id
stringlengths 27
33
| source
stringclasses 1
value | format
stringclasses 1
value | text
stringlengths 13
1.81M
|
|---|---|---|---|
warning/0004/hep-ph0004059.html
|
ar5iv
|
text
|
# 1 Introduction.
## 1 Introduction.
The scattering of electroweak probes off hadrons serves as a clean tool, free of complications from hadron-hadron reactions on both the theoretical and experimental sides, for the extraction of reliable information on the substructure of strongly interacting particles. Using the photon as a probe, (the absorptive part of) the forward virtual Compton (VC) process $`\gamma ^{}(q)N(P)\gamma ^{}(q)N(P)`$ allows to study the strong interaction dynamics and at the same time it has a simple QCD description in the hard regime, — when $`q^2m_{\mathrm{hadr}}^2`$. In more general settings it is instructive to address the non-forward scattering $`\gamma ^{()}(q_1)N(P_1)\gamma ^{()}(q_2)N(P_2)`$. The systematic approach to its calculation is established only for the deep inelastic domain, where the QCD factorization theorems separate short and long distance phenomena into a perturbative parton subprocess and soft functions which encode information about the strongly coupled regime. The latter, known as the skewed parton distributions (SPDs), being studied for some time , have attracted increased attention in light of the conceivable opportunity to learn more about the spin structure of the nucleon . They are also of interest in their own right being hybrids of parton densities/distribution amplitudes and form factors. They share properties of the former in different regions of phase space which have been studied perturbatively to a great extent at one and two-loop orders . Deeply virtual Compton scattering (DVCS), with $`q_2^2=0`$, proves to be an experimentally accessible reaction . In electroproduction processes of a real photon there is a strong contamination of DVCS by the Bethe-Heitler (BH) process. In view of the extreme interest to extract, or at least to constrain, SPDs it is timely to address the question of the best observables that allow to get rid of unwanted background. Fortunately, the interference of the two processes provides a rich source of information. It was suggested that diverse asymmetries can be used to disentangle the real and imaginary parts of DVCS and thus give access to SPDs. In the present contribution we consider a number of spin, azimuthal and charge asymmetries which share these properties and give predictions for the kinematics of the HERA and HERMES experiment. Spin asymmetries make it possible to extract the imaginary part of the DVCS amplitude and thus, due to the reality of SPDs, which holds owing to the spatial and time reversal invariance of strong interactions, give directly a measurement of the shape (at leading order in $`\alpha _s`$ in complete analogy to DIS) of SPDs on the diagonal $`t=\xi `$.
The paper is organized as follows. In section 2 we calculate the squared amplitude for DVCS, BH, and the interference term to leading twist-two accuracy. In section 3 we present simplified formulae for different cross sections and give a qualitative discussion of the feasibility to measure the DVCS leading twist-two amplitudes in different kinematical regions. After introducing models for SPDs, we give an estimate for different asymmetries for HERA and HERMES kinematics in 4. Finally, in section 5 we summarize.
## 2 Cross section.
To start let us discuss different contribution to the differential cross section of the electroproduction process $`e(k,\lambda )N(P_1,S_1)e(k^{},\lambda )N(P_2,S_2)\gamma (q_2,\mathrm{\Lambda })`$, given by the standard formula
$`d\sigma ={\displaystyle \frac{1}{4k.P_1}}|𝒯|^2(\lambda ,S_1)(2\pi )^4\delta ^4(k+P_1k^{}P_2q_2){\displaystyle \frac{d^3𝒌^{}}{2\omega ^{}(2\pi )^3}}{\displaystyle \frac{d^3𝑷_2}{2E_2(2\pi )^3}}{\displaystyle \frac{d^3𝒒_2}{2\nu (2\pi )^3}}.`$ (1)
The scattering amplitude squared $`|𝒯|^2`$ in the cross section, contains beside the VCS \[Fig. 1(a)\] and BH \[Fig. 1(b) and crossed contribution\] parts also the interference term:
$`|𝒯|^2(\lambda ,S_1)={\displaystyle \underset{\lambda ^{},S_2,\mathrm{\Lambda }}{}}\left\{|𝒯_{VCS}|^2+|𝒯_{BH}|^2+𝒯_{VCS}𝒯_{BH}^{}+𝒯_{VCS}^{}𝒯_{BH}\right\}.`$ (2)
The BH-amplitude is purely real and is given as a contraction of the leptonic tensor, at leading order in the fine structure constant $`\alpha `$,
$$L_{\mu \nu }=\overline{u}(k^{},\lambda ^{})[\gamma _\mu (/k/\mathrm{\Delta })^1\gamma _\nu +\gamma _\nu (/k^{}+/\mathrm{\Delta })^1\gamma _\mu ]u(k,\lambda ),$$
(3)
with the hadronic current
$`J_\nu =\overline{U}(P_2,S_2)\left\{F_1(\mathrm{\Delta }^2)\gamma _\nu +iF_2(\mathrm{\Delta }^2)\sigma _{\nu \tau }{\displaystyle \frac{\mathrm{\Delta }^\tau }{2M}}\right\}U(P_1,S_1),\text{where}\mathrm{\Delta }=P_2P_1=q_1q_2,`$ (4)
parametrized in terms of Dirac, $`F_1`$, and Pauli, $`F_2`$, form factors normalized according to $`F_1^p(0)=1`$, $`F_2^p(0)\kappa _p=1.79`$, and $`F_1^n(0)=0`$, $`F_2^n(0)\kappa _n=1.91`$, for proton and neutron, respectively. Thus, the BH amplitude is of the form
$`𝒯_{BH}={\displaystyle \frac{e^3}{\mathrm{\Delta }^2}}ϵ_\mu ^{}L^{\mu \nu }J_\nu .`$ (5)
The form factors are known fairly well from experimental measurements and can be parametrized by dipole formulae in the small $`\mathrm{\Delta }^2`$ region
$$G_E^p(\mathrm{\Delta }^2)=(1+\kappa _p)^1G_M^p(\mathrm{\Delta }^2)=\kappa _n^1G_M^n(\mathrm{\Delta }^2)=\left(1\frac{\mathrm{\Delta }^2}{m_V^2}\right)^2,G_E^n(\mathrm{\Delta }^2)=0,$$
(6)
where we have introduced the electric, $`G_E^i(\mathrm{\Delta }^2)=F_1^i(\mathrm{\Delta }^2)+\frac{\mathrm{\Delta }^2}{4M^2}F_2^i(\mathrm{\Delta }^2)`$, and magnetic, $`G_M^i(\mathrm{\Delta }^2)=F_1^i(\mathrm{\Delta }^2)+F_2^i(\mathrm{\Delta }^2)`$, form factor characterized by cutoff mass $`m_V=0.84\mathrm{GeV}`$, see e.g. . The hadronic tensor $`T_{VCS}`$ is:
$`𝒯_{VCS}={\displaystyle \frac{e^3}{q_1^2}}ϵ_\mu ^{}T^{\mu \nu }\overline{u}(k^{})\gamma _\nu u(k),\text{where}\{{\displaystyle \genfrac{}{}{0pt}{}{\text{ for }e^{}}{+\text{ for }e^+}}.`$ (7)
It is defined by the time ordered product of two electromagnetic currents
$`T_{\mu \nu }(q,P_1,P_2)=i{\displaystyle 𝑑xe^{ix.q}P_2,S_2|Tj_\mu (x/2)j_\nu (x/2)|P_1,S_1},`$ (8)
where $`q=(q_1+q_2)/2`$ (and the index $`\mu `$ refers to the outgoing real photon). It contains for a spin-1/2 target twelve<sup>1</sup><sup>1</sup>1$`12=\frac{1}{2}\times 3\text{ (virtual photon)}\times 2\text{ (final photon)}\times 2\text{ (initial nucleon)}\times 2\text{ (final nucleon)}`$. The reduction factor $`1/2`$ is a result of parity invariance. independent kinematical structures . In this paper we restrict ourselves to the twist-2 part of $`T_{\mu \nu }`$ that does not contain transversal photon spin flip contributions. Such contributions arise in the next-to-leading order of perturbation theory due to the gluon transversity and are seperately considered in . From the structure of the OPE we immediately learn that these contributions are contained in the following form factor decomposition<sup>2</sup><sup>2</sup>2We adopt throughout the conventions of , e.g. $`ϵ^{0123}=1`$:
$`T_{\mu \nu }(q,P,\mathrm{\Delta })`$ $`=`$ $`\stackrel{~}{g}_{\mu \nu }{\displaystyle \frac{q_\sigma V_1^\sigma }{P.q}}i\stackrel{~}{ϵ}_{\mu \nu q\sigma }{\displaystyle \frac{A_1^\sigma }{P.q}}+\mathrm{},`$ (9)
where $`P=P_1+P_2`$ and the gauge invariant tensors $`\stackrel{~}{t}_{\mu \nu }=𝒫_{\mu \rho }t_{\rho \sigma }𝒫_{\sigma \nu }`$ are constructed by means of the projection tensor $`𝒫_{\mu \nu }g_{\mu \nu }q_{1\mu }q_{2\nu }/q_1.q_2`$. The ellipsis indicate twist-three and higher contributions. The vectors $`V_{i\mu }`$ and axial-vectors $`A_{i\nu }`$ can be expressed by a form factor decomposition
$`V_{1\mu }=\overline{U}(P_2,S_2)\left(_1\gamma _\mu +_1{\displaystyle \frac{i\sigma _{\mu \nu }\mathrm{\Delta }^\nu }{2M}}\right)U(P_1,S_1)+\mathrm{},`$ (10)
$`A_{1\mu }=\overline{U}(P_2,S_2)\left(\stackrel{~}{}_1\gamma _\mu \gamma _5+\stackrel{~}{}_1{\displaystyle \frac{\mathrm{\Delta }_\mu \gamma _5}{2M}}\right)U(P_1,S_1)+\mathrm{},`$ (11)
where higher twist contributions are neglected. These form factors depend on the following variables
$`\xi ={\displaystyle \frac{Q^2}{P.q}},Q^2=q^2={\displaystyle \frac{1}{4}}(q_1+q_2)^2,\mathrm{\Delta }^2.`$
Note that in general (for off-shell final photons) a second scaling variable $`\eta =\frac{\mathrm{\Delta }.q}{P.q}`$ appears, which is related however to $`\xi `$, i.e. $`\eta =\xi \left(1\frac{\mathrm{\Delta }^2}{4Q^2}\right)\xi `$. The amplitudes are given as convolution in $`t`$, $`dt`$, of perturbatively calculable hard scattering parts with SPDs:
$`\left\{{\displaystyle \genfrac{}{}{0pt}{}{_1}{_1}}\right\}(\xi ,Q^2,\mathrm{\Delta }^2)`$ $`=`$ $`T_1(\xi ,Q^2,\mu ^2,t)\left\{{\displaystyle \genfrac{}{}{0pt}{}{H}{E}}\right\}(t,\xi ,\mathrm{\Delta }^2,\mu ^2),`$ (12)
$`\left\{{\displaystyle \genfrac{}{}{0pt}{}{\stackrel{~}{}_1}{\stackrel{~}{}_1}}\right\}(\xi ,Q^2,\mathrm{\Delta }^2)`$ $`=`$ $`\stackrel{~}{T}_1(\xi ,Q^2,\mu ^2,t)\left\{{\displaystyle \genfrac{}{}{0pt}{}{\stackrel{~}{H}}{\stackrel{~}{E}}}\right\}(t,\xi ,\mathrm{\Delta }^2,\mu ^2),`$ (13)
with summation over the different parton species implied and $`\mu ^2`$ being the factorization scale. The hard scattering amplitudes are available in next-to-leading order (NLO) approximation and they read in LO for a quark of charge $`Q_i`$
$$\xi T^{i(0)}(\xi ,t)=\frac{Q_i^2}{1t/\xi iϵ}(tt),$$
(14)
with $``$ ($`+`$) for parity even (odd) cases.
In the consequent presentation we give our results in the laboratory frame (see Fig. 2) in which we use the kinematical variables
$`k=(E,0,0,E),k^{}=(E^{},E^{}\mathrm{cos}\varphi _e\mathrm{sin}\theta _e,E^{}\mathrm{sin}\varphi _e\mathrm{sin}\theta _e,E^{}\mathrm{cos}\theta _e),`$ (15)
$`P_1=(M,0,0,0),P_2=(E_2,|𝑷_2|\mathrm{cos}\varphi _N\mathrm{sin}\theta _N,|𝑷_2|\mathrm{sin}\varphi _N\mathrm{sin}\theta _N,|𝑷_2|\mathrm{cos}\theta _N).`$
Furthermore, we introduce the azimuthal angle $`\varphi _r=\varphi _N\varphi _e`$ between the lepton and hadron scattering planes as well as $`\varphi _s=\varphi _N+\varphi _e`$. The spin vector of the spin-1/2 target for longitudinal and transverse polarization is given by
$`S=(0,0,0,\mathrm{\Lambda })\text{with}\mathrm{\Lambda }=\pm 1,S=(0,\mathrm{cos}\mathrm{\Phi },\mathrm{sin}\mathrm{\Phi },0),`$ (16)
respectively.
From the experimental point of view one works with the variables $`𝒬^2q_1^2`$ and $`xq_1^2/(2P_1.q_1)`$, which are related to our variables by
$`Q^2={\displaystyle \frac{1}{2}}q_1^2\left(1{\displaystyle \frac{\mathrm{\Delta }^2}{2q_1^2}}\right){\displaystyle \frac{1}{2}}𝒬^2,\xi ={\displaystyle \frac{x\left(1\frac{\mathrm{\Delta }^2}{2q_1^2}\right)}{2x\left(1+\frac{\mathrm{\Delta }^2}{q_1^2}\right)}}{\displaystyle \frac{x}{2x}}.`$ (17)
After performing the phase space integration we obtain for the differential cross section (1)
$`{\displaystyle \frac{d\sigma }{dxd𝒬^2d|\mathrm{\Delta }^2|d\varphi _r}}={\displaystyle \frac{y}{𝒬^2}}{\displaystyle \frac{d\sigma }{dxdyd|\mathrm{\Delta }^2|d\varphi _r}}={\displaystyle \frac{\alpha ^3xy^2}{8\pi 𝒬^4}}\left(1+{\displaystyle \frac{4M^2x^2}{𝒬^2}}\right)^{1/2}\left|{\displaystyle \frac{𝒯}{e^3}}\right|^2,`$ (18)
where we introduced as well the conventional variable $`y=P_1.q_1/P_1.k`$ \[$`y=1E^{}/E`$ in the frame (15)\].
Here we present simple analytical expressions for the amplitudes entering the cross section for positron beam. Changing to electrons will generate a minus sign for the interference terms. To deduce them we present at first the amplitudes squared in terms of the form factors. The result for the DVCS amplitude reads in leading twist-2 approximation (we set $`|e|=1`$)
$`|𝒯_{\mathrm{DVCS}}|^2`$ $`=`$ $`8{\displaystyle \frac{22y+y^2}{y^2}}{\displaystyle \frac{\xi ^2}{𝒬^6}}(q.V_1q.V_1^{}+q.A_1q.A_1^{})`$ (19)
$`+8{\displaystyle \frac{\lambda (2y)}{y}}{\displaystyle \frac{\xi ^2}{𝒬^6}}(q.A_1q.V_1^{}+q.V_1q.A_1^{}),`$
where $`\lambda =1`$ means that the spin of the lepton is parallel to the beam direction. It is obvious that the leading contribution scales as $`1/𝒬^2`$.
The exact squared amplitude for BH reads in terms of the electromagnetic currents:
$`|𝒯_{\mathrm{BH}}|^2`$ $`=`$ $`{\displaystyle \frac{8}{\mathrm{\Delta }^2}}{\displaystyle \frac{q.Jk.J^{}+k.Jq.J^{}q.Jq.J^{}2k.Jk.J^{}}{(2k.\mathrm{\Delta }\mathrm{\Delta }^2)(𝒬^2+2k.\mathrm{\Delta })}}`$
$`{\displaystyle \frac{4}{\mathrm{\Delta }^4}}\left(1+{\displaystyle \frac{𝒬^4+\mathrm{\Delta }^4}{2(2k.\mathrm{\Delta }\mathrm{\Delta }^2)(𝒬^2+2k.\mathrm{\Delta })}}\right)J.J^{}`$
$`+{\displaystyle \frac{4i\lambda [(𝒬^2+4k.\mathrm{\Delta }\mathrm{\Delta }^2)ϵ_{q\mathrm{\Delta }JJ^{}}+\mathrm{\Delta }^2ϵ_{k(2q+\mathrm{\Delta })JJ^{}}]}{\mathrm{\Delta }^4(2k.\mathrm{\Delta }\mathrm{\Delta }^2)(𝒬^2+2k.\mathrm{\Delta })}},`$
with the latter being defined in Eq. (4). Instead of the exact expression we may use an approximated one. To be consistent we have to expand the squared BH-amplitude, which starts with $`1/\mathrm{\Delta }^2`$, and the interference term up to the same order as the squared DVCS-amplitude, namely to $`1/𝒬^2`$ accuracy. However, one has to take into account that the lepton propagator in the $`u`$-channel of the BH amplitude gives a contribution which behaves as $`(kq_2)^2=\frac{(1y)}{y}𝒬^2\left(1+𝒪(1/𝒬)\right)`$. Therefore, the Taylor expansion is not legitimate for large $`y`$ and sets, otherwise, an upper limit for $`y`$, namely $`1yM^2/𝒬^2`$. To avoid this problem, we have to expand the propagator,
$`(kq_2)^2`$ $`=`$ $`{\displaystyle \frac{𝒬^2}{y}}\{1y+2\sqrt{{\displaystyle \frac{\mathrm{\Delta }^2}{𝒬^2}}}\sqrt{1{\displaystyle \frac{\mathrm{\Delta }_{\mathrm{min}}^2}{\mathrm{\Delta }^2}}}\sqrt{1y}\sqrt{1x}\mathrm{cos}(\varphi _r){\displaystyle \frac{\mathrm{\Delta }^2}{𝒬^2}}[{\displaystyle \frac{1y}{2}}`$
$`+(1x)(12{\displaystyle \frac{\mathrm{\Delta }_{\mathrm{min}}^2}{\mathrm{\Delta }^2}})(1y)(x+2(1x){\displaystyle \frac{\mathrm{\Delta }_{\mathrm{min}}^2}{\mathrm{\Delta }^2}})\mathrm{cos}(2\varphi _r)]+O\left({\displaystyle \frac{1}{𝒬^3}}\right)\},`$
and the remaining parts separately in Taylor series which results in a Pade-like approximation for the squared BH term. Here $`\mathrm{\Delta }_{\mathrm{min}}^2`$ is the minimal value of $`\mathrm{\Delta }^2`$ which is defined by the kinematical restriction $`\mathrm{\Delta }_{\mathrm{min}}^2=M^2x^2/(1x+xM^2/𝒬^2)`$. Taking only the first terms in the expansion, thus, neglecting all contributions formally suppressed by $`1/𝒬`$, we get
$`|𝒯_{\mathrm{BH}}|^2`$ $`=`$ $`{\displaystyle \frac{2(22y+y^2)}{(1y)}}\left({\displaystyle \frac{J.J^{}}{\mathrm{\Delta }^4}}+4{\displaystyle \frac{q.Jq.J^{}}{\mathrm{\Delta }^2𝒬^4}}\right)4{\displaystyle \frac{\lambda (2y)y^2}{(1y)}}{\displaystyle \frac{iϵ_{k\mathrm{\Delta }JJ^{}}}{\mathrm{\Delta }^4𝒬^2}},`$ (22)
which, however, is valid for $`y1`$.
For interference terms we adhere to the same approximation and give here only the leading contributions in $`1/𝒬`$, i.e.
$`𝒯_{\mathrm{BH}}𝒯_{\mathrm{DVCS}}^{}`$ $`=`$ $`4{\displaystyle \frac{22y+y^2}{1y}}{\displaystyle \frac{\xi }{\mathrm{\Delta }^2𝒬^4}}[(k^\sigma {\displaystyle \frac{1}{y}}q^\sigma )(J_\sigma +2\mathrm{\Delta }_\sigma {\displaystyle \frac{q.J}{𝒬^2}})q.V_1^{}2iϵ_{kq\mathrm{\Delta }J}{\displaystyle \frac{q.A_1^{}}{𝒬^2}}]`$ (23)
$`4{\displaystyle \frac{\lambda (2y)y}{1y}}{\displaystyle \frac{\xi }{\mathrm{\Delta }^2𝒬^4}}[(k^\sigma {\displaystyle \frac{1}{y}}q^\sigma )(J_\sigma +2\mathrm{\Delta }_\sigma {\displaystyle \frac{q.J}{𝒬^2}})q.A_1^{}2iϵ_{kq\mathrm{\Delta }J}{\displaystyle \frac{q.V_1^{}}{𝒬^2}}].`$
Note that for a consistent expansion up to order $`1/𝒬^2`$ also twist-three contributions having both kinematical and dynamical origins must be considered (see Ref. for the case of a scalar target). In general these contributions are poorly understood and have to be studied in more detail.
As a next step for the calculation of the cross section we have to sum over the spin of the outgoing proton and write the final answer in terms of SPDs and electromagnetic form factors. Here we give the results for polarized spin-1/2 target, where $`\mathrm{\Lambda }=1`$ means polarization along the lepton beam, averaged over the azimuthal angle $`\varphi _s`$. The DVCS amplitude squared $`|𝒯_{\mathrm{DVCS}}|^2=|𝒯_{\mathrm{DVCS}}|_{\mathrm{unp}}^2+|𝒯_{\mathrm{DVCS}}|_{\mathrm{pol}}^2`$, with $`\mathrm{pol}=\{\mathrm{LP},\mathrm{TP}\}`$, consists of the elements
$`|𝒯_{\mathrm{DVCS}}|_{\mathrm{unp}}^2`$ $`=`$ $`{\displaystyle \frac{2(22y+y^2)}{y^2(2x)^2𝒬^2}}[4(1x)(_1_1^{}+\stackrel{~}{}_1\stackrel{~}{}_1^{})x^2(_1_1^{}+_1_1^{}+\stackrel{~}{}_1\stackrel{~}{}_1^{}+\stackrel{~}{}_1\stackrel{~}{}_1^{})`$ (24)
$`(x^2+(2x)^2{\displaystyle \frac{\mathrm{\Delta }^2}{4M^2}})_1_1^{}x^2{\displaystyle \frac{\mathrm{\Delta }^2}{4M^2}}\stackrel{~}{}_1\stackrel{~}{}_1^{}],`$
$`|𝒯_{\mathrm{DVCS}}|_{\mathrm{LP}}^2`$ $`=`$ $`{\displaystyle \frac{2\lambda \mathrm{\Lambda }(2y)}{y(2x)^2𝒬^2}}[4(1x)(_1\stackrel{~}{}_1^{}+\stackrel{~}{}_1_1^{})x^2(_1\stackrel{~}{}_1^{}+\stackrel{~}{}_1_1^{}+\stackrel{~}{}_1_1^{}+_1\stackrel{~}{}_1^{})`$ (25)
$`x({\displaystyle \frac{1}{2}}x^2+(2x){\displaystyle \frac{\mathrm{\Delta }^2}{4M^2}})(_1\stackrel{~}{}_1^{}+\stackrel{~}{}_1_1^{})],`$
$`|𝒯_{\mathrm{DVCS}}|_{\mathrm{TP}}^2`$ $`=`$ $`{\displaystyle \frac{8\mathrm{cos}(\mathrm{\Phi }\varphi _r/2)(22y+y^2)\sqrt{1x}}{\pi y^2(2x)^2𝒬^2}}\sqrt{{\displaystyle \frac{\mathrm{\Delta }^2}{M^2}}}\sqrt{1{\displaystyle \frac{\mathrm{\Delta }_{\mathrm{min}}^2}{\mathrm{\Delta }^2}}}`$
$`\times \left\{(2x)\left[(\mathrm{Re}_1)(\mathrm{Im}_1)(\mathrm{Im}_1)(\mathrm{Re}_1)\right]x\left[(\mathrm{Re}\stackrel{~}{}_1)(\mathrm{Im}\stackrel{~}{}_1)(\mathrm{Im}\stackrel{~}{}_1)(\mathrm{Re}\stackrel{~}{}_1)\right]\right\}`$
$`+{\displaystyle \frac{2\lambda \mathrm{sin}(\mathrm{\Phi }\varphi _r/2)(2y)\sqrt{1x}}{\pi y(2x)^2𝒬^2}}\sqrt{{\displaystyle \frac{\mathrm{\Delta }^2}{M^2}}}\sqrt{1{\displaystyle \frac{\mathrm{\Delta }_{\mathrm{min}}^2}{\mathrm{\Delta }^2}}}[2x(_1\stackrel{~}{}_1^{}+_1^{}\stackrel{~}{}_1)`$
$`2(2x)(\stackrel{~}{}_1_1^{}+\stackrel{~}{}_1^{}_1)+x^2(_1\stackrel{~}{}_1^{}+_1^{}\stackrel{~}{}_1)].`$
For the BH term we use an analogous decomposition with
$`|𝒯_{\mathrm{BH}}|_{\mathrm{unp}}^2`$ $`=`$ $`{\displaystyle \frac{2(22y+y^2)}{(1y)\mathrm{\Delta }^2}}\left[4{\displaystyle \frac{1x}{x^2}}\left(1{\displaystyle \frac{\mathrm{\Delta }_{\mathrm{min}}^2}{\mathrm{\Delta }^2}}\right)F_1^2+2(F_1+F_2)^2+\left({\displaystyle \frac{\mathrm{\Delta }^2}{\mathrm{\Delta }_{\mathrm{min}}^2}}1\right)F_2^2\right],`$ (27)
$`|𝒯_{\mathrm{BH}}|_{\mathrm{LP}}^2`$ $`=`$ $`{\displaystyle \frac{4\lambda \mathrm{\Lambda }(2y)y}{(1y)\mathrm{\Delta }^2}}(F_1+F_2)\left[2{\displaystyle \frac{1x}{x}}\left(1{\displaystyle \frac{\mathrm{\Delta }_{\mathrm{min}}^2}{\mathrm{\Delta }^2}}\right)F_1+F_1+F_2\right],`$ (28)
$`|𝒯_{\mathrm{BH}}|_{\mathrm{TP}}^2`$ $`=`$ $`{\displaystyle \frac{8\lambda \mathrm{sin}(\mathrm{\Phi }\varphi _r/2)(2y)y\sqrt{1x}}{\pi (1y)xM(\mathrm{\Delta }^2)^{3/2}}}\sqrt{1{\displaystyle \frac{\mathrm{\Delta }_{\mathrm{min}}^2}{\mathrm{\Delta }^2}}}(F_1+F_2)\left(2xM^2F_1+\mathrm{\Delta }^2F_2\right).`$ (29)
For a polarized lepton beam the interference term is decomposed into a contribution for an unpolarized and an additional one for a polarized target, i.e. $`^2𝒯_{\mathrm{BH}}𝒯_{\mathrm{DVCS}}^{}+𝒯_{\mathrm{DVCS}}𝒯_{\mathrm{BH}}^{}=_{\mathrm{unp}}^2(\lambda )+_{\mathrm{pol}}^2(\lambda )`$, where
$`_{\mathrm{unp}}^2(\lambda )`$ $`=`$ $`{\displaystyle \frac{8(22y+y^2)\sqrt{1x}}{\sqrt{1y}yx\sqrt{\mathrm{\Delta }^2𝒬^2}}}\sqrt{1{\displaystyle \frac{\mathrm{\Delta }_{\mathrm{min}}^2}{\mathrm{\Delta }^2}}}\mathrm{cos}(\varphi _r)\mathrm{Re}\{F_1_1+{\displaystyle \frac{x}{2x}}(F_1+F_2)\stackrel{~}{}_1`$
$`{\displaystyle \frac{\mathrm{\Delta }^2}{4M^2}}F_2_1\}{\displaystyle \frac{8\lambda (2y)\sqrt{1x}}{\sqrt{1y}x\sqrt{\mathrm{\Delta }^2𝒬^2}}}\sqrt{1{\displaystyle \frac{\mathrm{\Delta }_{\mathrm{min}}^2}{\mathrm{\Delta }^2}}}\mathrm{sin}(\varphi _r)\mathrm{Im}\{F_1_1`$
$`+{\displaystyle \frac{x}{2x}}(F_1+F_2)\stackrel{~}{}_1{\displaystyle \frac{\mathrm{\Delta }^2}{4M^2}}F_2_1\},`$
$`_{\mathrm{LP}}^2(\lambda )`$ $`=`$ $`{\displaystyle \frac{8\mathrm{\Lambda }(22y+y^2)\sqrt{1x}}{\sqrt{1y}yx\sqrt{\mathrm{\Delta }^2𝒬^2}}}\sqrt{1{\displaystyle \frac{\mathrm{\Delta }_{\mathrm{min}}^2}{\mathrm{\Delta }^2}}}\mathrm{sin}(\varphi _r)\mathrm{Im}\{{\displaystyle \frac{x}{2x}}(F_1+F_2)(_1+{\displaystyle \frac{x}{2}}_1)`$ (31)
$`+F_1\stackrel{~}{}_1{\displaystyle \frac{x}{2x}}({\displaystyle \frac{x}{2}}F_1+{\displaystyle \frac{\mathrm{\Delta }^2}{4M^2}}F_2)\stackrel{~}{}_1\}+{\displaystyle \frac{8\lambda \mathrm{\Lambda }(2y)\sqrt{1x}}{\sqrt{1y}x\sqrt{\mathrm{\Delta }^2𝒬^2}}}\sqrt{1{\displaystyle \frac{\mathrm{\Delta }_{\mathrm{min}}^2}{\mathrm{\Delta }^2}}}\mathrm{cos}(\varphi _r)`$
$`\times \mathrm{Re}\left\{{\displaystyle \frac{x}{2x}}(F_1+F_2)\left(_1+{\displaystyle \frac{x}{2}}_1\right)+F_1\stackrel{~}{}_1{\displaystyle \frac{x}{2x}}\left({\displaystyle \frac{x}{2}}F_1+{\displaystyle \frac{\mathrm{\Delta }^2}{4M^2}}F_2\right)\stackrel{~}{}_1\right\},`$
$`_{\mathrm{TP}}^2(\lambda )`$ $`=`$ $`{\displaystyle \frac{8(22y+y^2)}{\sqrt{1y}yx(2x)\sqrt{𝒬^2M^2}}}[{\displaystyle \frac{\mathrm{cos}(\mathrm{\Phi }3\varphi _r/2)}{2\pi }}(1{\displaystyle \frac{\mathrm{\Delta }_{\mathrm{min}}^2}{\mathrm{\Delta }^2}})(1x)\mathrm{Im}\{2F_2(_1+\stackrel{~}{}_1)`$
$`[(2x)F_1xF_2]_1xF_1\stackrel{~}{}_1\}+{\displaystyle \frac{\mathrm{cos}(\mathrm{\Phi }+\varphi _r/2)}{2\pi }}({\displaystyle \frac{\mathrm{\Delta }_{\mathrm{min}}^2}{\mathrm{\Delta }^2}}(F_1+F_2)\mathrm{Im}\{4(1x)`$
$`\times (_1\stackrel{~}{}_1)x^2(_1\stackrel{~}{}_1)\}+(1{\displaystyle \frac{\mathrm{\Delta }_{\mathrm{min}}^2}{\mathrm{\Delta }^2}}\left)\mathrm{Im}\right\{2(1x)F_2(_1\stackrel{~}{}_1)[(2x)F_1`$
$`+xF_2]_1+x(F_1+xF_2)\stackrel{~}{}_1\})]+{\displaystyle \frac{8\lambda (2y)}{\sqrt{1y}(2x)x\sqrt{𝒬^2M^2}}}[{\displaystyle \frac{\mathrm{sin}(\mathrm{\Phi }3\varphi _r/2)}{2\pi }}`$
$`\times \left(1{\displaystyle \frac{\mathrm{\Delta }_{\mathrm{min}}^2}{\mathrm{\Delta }^2}}\right)(1x)\mathrm{Re}\left\{2F_2(_1+\stackrel{~}{}_1)[(2x)F_1xF_2]_1xF_1\stackrel{~}{}_1\right\}`$
$`{\displaystyle \frac{\mathrm{sin}(\mathrm{\Phi }+\varphi _r/2)}{2\pi }}({\displaystyle \frac{\mathrm{\Delta }_{\mathrm{min}}^2}{\mathrm{\Delta }^2}}(F_1+F_2)\mathrm{Re}\{4(1x)(_1\stackrel{~}{}_1)x^2(_1\stackrel{~}{}_1)\}`$
$`+(1{\displaystyle \frac{\mathrm{\Delta }_{\mathrm{min}}^2}{\mathrm{\Delta }^2}})\mathrm{Re}\{2(1x)F_2(_1\stackrel{~}{}_1)[(2x)F_1+xF_2]_1`$
$`+x(F_1+xF_2)\stackrel{~}{}_1\})].`$
## 3 Extraction of leading twist-two amplitudes.
A strong motivation for the measurement of DVCS arises from the fact that the second moment of SPDs in the parity even sector is related to form factors appearing in the decomposition of the symmetric QCD energy-momentum tensor $`\mathrm{\Theta }_{\mu \nu }`$. A gauge-invariant decomposition of the matrix element of the angular-momentum tensor, $`_{\mu \nu ,\sigma }=x_\mu \mathrm{\Theta }_{\sigma \nu }x_\nu \mathrm{\Theta }_{\sigma \mu }`$, provides, therefore, a separate estimation of the total angular-momentum fraction carried by quarks and gluons . To achieve this goal it is necessary to interpolate the corresponding moments of the SPDs to forward kinematics $`\mathrm{\Delta }=0`$. The new information that is required (and not available from DIS) is contained in the SPDs appearing in the spin-flip amplitude $`_1`$. Having the spin puzzle in mind, we give special attention to the problem of extracting $`_1`$ from measurements in different kinematical domains.
The simplified explicit expressions (24-2) for the amplitudes squared allow us to discuss the extraction of the leading twist-two amplitudes from future experimental data. Moreover, they allow us to give a qualitative discussion of the ratio for the different cross sections in more detail. Note, however, that the formulae, presented below, are only valid in a small kinematical window and in general not sufficient for numerical estimates. In the following we divide the kinematical region into $`\mathrm{\Delta }^2\mathrm{\Delta }_{\mathrm{min}}^2`$, $`|\mathrm{\Delta }_{\mathrm{min}}^2|<|\mathrm{\Delta }^2|<M^2`$, and $`|\mathrm{\Delta }_{\mathrm{min}}^2|<M^2|\mathrm{\Delta }^2|`$. Furthermore, we separately consider the small $`x`$ region.
### 3.1 Small $`x`$ region.
Let us start with the small $`x`$ region, which is for instance relevant for the HERA experiments . We assume that the small $`x`$ behavior of $`_1`$ and $`\stackrel{~}{}_1`$ is not one power less than that of $`_1`$ and $`\stackrel{~}{}_1`$, and, furthermore, that the SPDs in the small $`x`$ region are essentially determined by the usual parton densities at least in the DGLAP region. It is easy to establish that for small values of $`x`$ the ratio of the real to imaginary part of the unpolarized amplitude $`_1`$ is about 0.3 (by means of dispersion relation the same magnitude has been obtained in Ref. ). In the polarized case we find for instance for the Gehrman-Stirling parametrization that this ratio for $`\stackrel{~}{}_1`$ is of order $`11.7`$. However, the contribution of the latter is quite small as compared to $`_1`$.
Taking into account these numbers, we find that for longitudinally polarized target Eqs. (24,25) can be approximated by:
$`|𝒯_{\mathrm{DVCS}}|^2`$ $``$ $`{\displaystyle \frac{2(22y+y^2)}{y^2𝒬^2}}\left[(\mathrm{Im}_1)^2{\displaystyle \frac{\mathrm{\Delta }^2}{4M^2}}\left\{(\mathrm{Im}_1)^2+(\mathrm{Re}_1)^2\right\}\right]`$
$`{\displaystyle \frac{4\lambda \mathrm{\Lambda }(2y)}{y𝒬^2}}\left(\mathrm{Im}_1\mathrm{Im}\stackrel{~}{}_1+\mathrm{Re}_1\mathrm{Re}\stackrel{~}{}_1\right).`$
Contributions containing $`\stackrel{~}{}_1`$ are down by two powers of $`x`$ or they are proportional to $`x\frac{\mathrm{\Delta }^2}{4M^2}`$. The transversally polarized part offers an opportunity to measure $`_1`$:
$`|𝒯_{\mathrm{DVCS}}|_{\mathrm{TP}}^2`$ $``$ $`{\displaystyle \frac{4\mathrm{cos}(\mathrm{\Phi }\varphi _r/2)(22y+y^2)}{\pi y^2𝒬^2}}\sqrt{{\displaystyle \frac{\mathrm{\Delta }^2}{M^2}}}\sqrt{1{\displaystyle \frac{\mathrm{\Delta }_{\mathrm{min}}^2}{\mathrm{\Delta }^2}}}\left[(\mathrm{Re}_1)(\mathrm{Im}_1)(\mathrm{Im}_1)(\mathrm{Re}_1)\right]`$
$`{\displaystyle \frac{4\lambda \mathrm{sin}(\mathrm{\Phi }\varphi _r/2)(2y)}{\pi y𝒬^2}}\sqrt{{\displaystyle \frac{\mathrm{\Delta }^2}{M^2}}}\sqrt{1{\displaystyle \frac{\mathrm{\Delta }_{\mathrm{min}}^2}{\mathrm{\Delta }^2}}}\left[(\mathrm{Im}\stackrel{~}{}_1)(\mathrm{Im}_1)+(\mathrm{Re}\stackrel{~}{}_1)(\mathrm{Re}_1)\right].`$
The interference term starts with $`x^1`$, however, since the sea quark and gluonic contributions to the DVCS amplitudes are expected to grow with $`x^1`$, the DVCS cross section and the interference term have the same $`x`$ dependence. For the longitudinally polarized case it reads for small $`x`$:
$`^2`$ $`=`$ $`{\displaystyle \frac{8\mathrm{sin}(\varphi _r)\sqrt{1\frac{\mathrm{\Delta }_{\mathrm{min}}^2}{\mathrm{\Delta }^2}}}{\sqrt{1y}yx\sqrt{\mathrm{\Delta }^2𝒬^2}}}\mathrm{Im}\{\lambda (2y)y(F_1_1{\displaystyle \frac{\mathrm{\Delta }^2}{4M^2}}F_2_1)\mathrm{\Lambda }(22y+y^2)`$
$`\times ({\displaystyle \frac{x}{2}}(F_1+F_2)_1+F_1\stackrel{~}{}_1)\}{\displaystyle \frac{8\mathrm{cos}(\varphi _r)\sqrt{1\frac{\mathrm{\Delta }_{\mathrm{min}}^2}{\mathrm{\Delta }^2}}}{\sqrt{1y}yx\sqrt{\mathrm{\Delta }^2𝒬^2}}}\mathrm{Re}\{(22y+y^2)`$
$`\times (F_1_1{\displaystyle \frac{\mathrm{\Delta }^2}{4M^2}}F_2_1)\lambda \mathrm{\Lambda }(2y)y({\displaystyle \frac{x}{2}}(F_1+F_2)_1+F_1\stackrel{~}{}_1)\}.`$
The interference term might give in future an opportunity to access the imaginary and real part of the linear combination $`F_1_1\frac{\mathrm{\Delta }^2}{4M^2}F_2_1`$ due to the charge and single lepton spin asymmetries for $`\mathrm{\Delta }_{\mathrm{min}}^2<\mathrm{\Delta }^2`$. For a polarized proton beam one may extract in an analogous way $`\mathrm{Im}\left(x(F_1+F_2)_1/2+F_1\stackrel{~}{}_1\right)`$ and in combination with a polarized lepton beam one also gets the real part of this expression. Note that one has now the opportunity, at least in principle, to extract the DVCS cross section (3.1) for unpolarized or double spin flip experiments and thus separate $`_1`$ from the $`_1`$ contributions for the imaginary and real part.
The squared of the BH term will generally start with $`x^2`$ (however, the spin dependent part goes only like $`x^1`$) and has therefore a similar $`x`$ dependence as the other ones, i.e. $`x\mathrm{Im}_1/F_i`$ is of order one or so, and grows only slightly with increasing $`x`$. However, in comparison with the squared DVCS term it is multiplied by $`y^2𝒬^2/\mathrm{\Delta }^2`$. Thus, one expects that the BH background is not large in the small $`y`$ and $`\mathrm{\Delta }^2/𝒬^2`$ region. We conclude from our discussion that the measurement of the unpolarized and longitudinal polarized cross sections as well as single spin asymmetries it is feasible to disentangle the imaginary and real part of $`_1`$, $`\stackrel{~}{}_1`$, and $`_1`$ at small $`x`$ and $`\mathrm{\Delta }^2`$ of the order of one $`\mathrm{GeV}^2`$ or larger. For smaller values $`_1`$ is in general kinematically reduced. Since $`\stackrel{~}{_1}`$ is suppressed at least by one power in $`x`$, we conclude that this amplitude will not be accessible in the small $`x`$ region.
As mentioned, if $`\mathrm{\Delta }^2`$ starts to be smaller, the contributions proportional to $`_1`$ die out. In the case of its value at the kinematical boundary, i.e. $`\mathrm{\Delta }^2\mathrm{\Delta }_{\mathrm{min}}^2`$, we see that Eqs. (27-29) are quite simple, and, for instance, for a longitudinally polarized target we obtain
$`|𝒯_{\mathrm{BH}}|^2={\displaystyle \frac{4(22y+y^2)}{(1y)\mathrm{\Delta }_{\mathrm{min}}^2}}(F_1+F_2)^2+{\displaystyle \frac{4\lambda \mathrm{\Lambda }(2y)y}{(1y)\mathrm{\Delta }_{\mathrm{min}}^2}}(F_1+F_2)^2.`$ (36)
Moreover, in this case the interference term for unpolarized or longitudinally polarized target drops out, too. It is worth to have a closer look at the ratio of the DVCS and BH cross section. The unpolarized cross section is essentially governed by the imaginary part of unpolarized parton distributions:
$`{\displaystyle \frac{d\sigma _{\mathrm{DVCS}}(\lambda =0,\mathrm{\Lambda }=0)}{d\sigma _{\mathrm{BH}}(\lambda =0,\mathrm{\Lambda }=0)}}={\displaystyle \frac{(1y)x^2M^2}{2y^2𝒬^2}}{\displaystyle \frac{\mathrm{Im}_1^2}{(F_1+F_2)^2}}.`$ (37)
It is remarkable that the ratio of double spin flip (DF) cross sections is proportional to the product of unpolarized and polarized parton distributions:
$`{\displaystyle \frac{d\sigma _{\mathrm{DVCS}}(\lambda =1,\mathrm{\Lambda }=1)d\sigma _{\mathrm{DVCS}}(\lambda =1,\mathrm{\Lambda }=1)}{d\sigma _{\mathrm{BH}}(\lambda =1,\mathrm{\Lambda }=1)d\sigma _{\mathrm{BH}}(\lambda =1,\mathrm{\Lambda }=1)}}={\displaystyle \frac{(1y)x^2M^2}{y^2𝒬^2}}{\displaystyle \frac{\mathrm{Im}_1\mathrm{Im}\stackrel{~}{}_1+\mathrm{Re}_1\mathrm{Re}\stackrel{~}{}_1}{(F_1+F_2)^2}}.`$ (38)
For the sake of completeness we want also to mention that the interference term for a transversally polarized target does not vanish for $`\mathrm{\Delta }^2\mathrm{\Delta }_{\mathrm{min}}^2`$, while both the DVCS and BH cross section are reduced to the unpolarized ones. This interference term is proportional to $`_1\stackrel{~}{}_1`$:
$`^2(\lambda )`$ $`=`$ $`{\displaystyle \frac{16(22y+y^2)}{\sqrt{1y}yx\sqrt{𝒬^2M^2}}}{\displaystyle \frac{\mathrm{cos}(\mathrm{\Phi }+\varphi _r/2)}{2\pi }}(F_1+F_2)\mathrm{Im}\left\{_1\stackrel{~}{}_1\right\}`$
$`{\displaystyle \frac{16\lambda (2y)}{\sqrt{1y}x\sqrt{𝒬^2M^2}}}{\displaystyle \frac{\mathrm{sin}(\mathrm{\Phi }+\varphi _r/2)}{2\pi }}(F_1+F_2)\mathrm{Re}\left\{_1\stackrel{~}{}_1\right\}.`$
Note that at the kinematical boundary the spin flip contributions are generally suppressed by at least a factor $`x^2`$. Thus, in this kinematical range it is hopeless to extract $`_1`$ or $`\stackrel{~}{}_1`$ from any experiment.
### 3.2 Asymmetries for $`|\mathrm{\Delta }^2||\mathrm{\Delta }_{\mathrm{min}}^2|`$.
Away from the kinematical boundary $`|\mathrm{\Delta }^2||\mathrm{\Delta }_{\mathrm{min}}^2|`$ we can consider different cross sections to get separately information on the real and imaginary parts of the leading twist-two amplitudes, $`_1,\mathrm{},\stackrel{~}{}_1`$. Since the BH cross section depends only on the product of the lepton and target polarization, i.e. on $`\lambda \mathrm{\Lambda }`$ or $`\lambda f(\mathrm{\Phi })`$ with $`f(\mathrm{\Phi }+\pi )=f(\mathrm{\Phi })`$, we have a direct access to the interference term in polarized experiments to leading order in $`1/\sqrt{𝒬^2}`$. Moreover, the relative sign of the interference term is determined by the charge of the lepton beam. Thus, the following cross sections allow one to get access to the leading twist-two structure functions in the DVCS amplitude. In the approximation used above and neglecting terms $`𝒪\left(\mathrm{\Delta }_{\mathrm{min}}^2/\mathrm{\Delta }^2\right)`$, they read:
1. Polarized positron beam and unpolarized target:
$`\mathrm{\Delta }_{\mathrm{SL}}d\sigma `$ $``$ $`d\sigma ^{}d\sigma ^{}`$
$`=`$ $`{\displaystyle \frac{16(2y)\sqrt{1x}}{\sqrt{1y}x\sqrt{\mathrm{\Delta }^2𝒬^2}}}\mathrm{sin}(\varphi _r)\mathrm{Im}\left\{F_1_1+{\displaystyle \frac{x}{2x}}(F_1+F_2)\stackrel{~}{}_1{\displaystyle \frac{\mathrm{\Delta }^2}{4M^2}}F_2_1\right\}d.`$
2. Unpolarized positron beam and longitudinally polarized target:
$`\mathrm{\Delta }_{\mathrm{SLN}}d\sigma `$ $``$ $`d\sigma _{}d\sigma _{}={\displaystyle \frac{16(22y+y^2)\sqrt{1x}}{\sqrt{1y}yx\sqrt{\mathrm{\Delta }^2𝒬^2}}}\mathrm{sin}(\varphi _r)`$
$`\times \mathrm{Im}\left\{{\displaystyle \frac{x}{2x}}(F_1+F_2)\left(_1+{\displaystyle \frac{x}{2}}_1\right)+F_1\stackrel{~}{}_1+{\displaystyle \frac{x}{2x}}\left({\displaystyle \frac{x}{2}}F_1+{\displaystyle \frac{\mathrm{\Delta }^2}{4M^2}}F_2\right)\stackrel{~}{}_1\right\}d.`$
3. Unpolarized positron beam and transversally polarized target: $`(\mathrm{\Phi }=\{0,\pi \})`$:
$`\mathrm{\Delta }_{\mathrm{STN}}d\sigma `$ $``$ $`d\sigma _{}d\sigma _{}={\displaystyle \frac{16(22y+y^2)}{\sqrt{1y}yx(2x)\sqrt{𝒬^2M^2}}}`$
$`\times `$ $`[{\displaystyle \frac{\mathrm{cos}(3\varphi _r/2)}{2\pi }}(1x)\mathrm{Im}\{2F_2(_1+\stackrel{~}{}_1)[(2x)F_1xF_2]_1xF_1\stackrel{~}{}_1\}`$
$`+{\displaystyle \frac{\mathrm{cos}(\varphi _r/2)}{2\pi }}\mathrm{Im}\{2(1x)F_2(_1\stackrel{~}{}_1)[(2x)F_1+xF_2]_1`$
$`+x(F_1+xF_2)\stackrel{~}{}_1\}]d.`$
4. Charge asymmetry in unpolarized experiment:
$`\mathrm{\Delta }_\mathrm{C}^{\mathrm{unp}}d\sigma `$ $``$ $`d{}_{}{}^{+}\sigma _{}^{\mathrm{unp}}d{}_{}{}^{}\sigma _{}^{\mathrm{unp}}={\displaystyle \frac{16(22y+y^2)\sqrt{1x}}{\sqrt{1y}yx\sqrt{\mathrm{\Delta }^2𝒬^2}}}\mathrm{cos}(\varphi _r)`$
$`\times \mathrm{Re}\left\{F_1_1+{\displaystyle \frac{x}{2x}}(F_1+F_2)\stackrel{~}{}_1{\displaystyle \frac{\mathrm{\Delta }^2}{4M^2}}F_2_1\right\}d.`$
5. Charge asymmetry in double spin-flip experiments with longitudinally polarized target:
$`\mathrm{\Delta }_\mathrm{C}^{\mathrm{DFL}}d\sigma `$ $``$ $`d{}_{}{}^{+}\sigma _{}^{}d{}_{}{}^{}\sigma _{}^{}\mathrm{\Delta }_\mathrm{C}d\sigma ^{\mathrm{unp}}={\displaystyle \frac{16(2y)\sqrt{1x}}{\sqrt{1y}x\sqrt{\mathrm{\Delta }^2𝒬^2}}}\mathrm{cos}(\varphi _r)`$
$`\times \mathrm{Re}\left\{{\displaystyle \frac{x}{2x}}(F_1+F_2)\left(_1+{\displaystyle \frac{x}{2}}_1\right)+F_1\stackrel{~}{}_1{\displaystyle \frac{x}{2x}}\left({\displaystyle \frac{x}{2}}F_1+{\displaystyle \frac{\mathrm{\Delta }^2}{4M^2}}F_2\right)\stackrel{~}{}_1\right\}d.`$
6. Charge asymmetry in double spin-flip experiments with transversally polarized target:
$`\mathrm{\Delta }_\mathrm{C}^{\mathrm{DFT}}d\sigma `$ $``$ $`d{}_{}{}^{+}\sigma _{}^{}d{}_{}{}^{}\sigma _{}^{}\mathrm{\Delta }_\mathrm{C}d\sigma ^{\mathrm{unp}}={\displaystyle \frac{8(2y)}{\sqrt{1y}(2x)x\sqrt{𝒬^2M^2}}}`$
$`\times `$ $`[{\displaystyle \frac{\mathrm{sin}(3\varphi _r/2)}{2\pi }}2(1x)\mathrm{Re}\{2F_2(_1+\stackrel{~}{}_1)[(2x)F_1xF_2]_1xF_1\stackrel{~}{}_1\}`$
$`+{\displaystyle \frac{\mathrm{sin}(\varphi _r/2)}{\pi }}\mathrm{Re}\{2(1x)F_2(_1\stackrel{~}{}_1)[(2x)F_1+xF_2]_1`$
$`+x(F_1+xF_2)\stackrel{~}{}_1\}]d,`$
where $`d=\frac{\alpha ^3xy}{8\pi 𝒬^2}\left(1+\frac{4M^2x^2}{𝒬^2}\right)^{1/2}dxdyd|\mathrm{\Delta }^2|d\varphi _r`$. Note that for a consequent numerical treatment the $`\mathrm{\Delta }_{\mathrm{min}}^2/\mathrm{\Delta }^2`$ dependence cannot be droped and can be easily restored from Eqs. (22).
As we see from this list, the two single spin cross sections of longitudinally polarized beam or target give us information on the imaginary part of two linear combinations of the four leading twist-two amplitudes. The real part of these two quantities is accessible via charge asymmetry in unpolarized or longitudinally double spin flip experiments. Obviously, at low $`|\mathrm{\Delta }^2|`$, i.e. compared to $`4M^24\mathrm{GeV}^2`$, there is a suppression of contributions proportional to $`E`$ and $`\stackrel{~}{E}`$ SPDs, which are theoretically not well constraint: the contribution proportional to $`E`$ SPDs completely drops, while $`\stackrel{~}{E}`$ is suppressed by $`x^2`$. For the kinematics of present experiments $`\mathrm{\Delta }^21\mathrm{G}\mathrm{e}\mathrm{V}^2`$, and thus it is not a good approximation to drop afore mentioned terms, $`\mathrm{\Delta }^2/4M^2𝒪(1)`$. However, in certain asymmetries these contributions are accompanied by a factor $`x`$ and thus can be safely discarded at smaller values of $`x`$. Further constraints separately for the imaginary and real part arise from spin and charge asymmetries for a transversally polarized target. Fortunately, due to different angular dependences, we have indeed two further independent linear combinations of the DVCS amplitudes. But the kinematical pre-factors and an additional angular dependence of the denominator for large $`y`$, which we have dropped for simplicity, makes their practical use questionable.
Finally, the whole cross section for unpolarized and polarized $`\gamma `$ production tests the leading twist-two approximation used above. However, to be consistent one has then to expand the BH and interference terms up to order $`1/𝒬^2`$. As outlined in section 2 this is straightforward for the BH cross section, while in the interference term new twist-three contributions will enter. These additional terms will be worked out in a forthcoming paper. Here we should only mention that the different azimuthal angle dependence can be used to pick up different combinations of helicity amplitudes, entering at different twist levels. For instance, for small $`y`$ it is justified to use the derived formulae which imply that all single spin and charge asymmetry cross sections integrated over the azimuthal angle $`\varphi _r`$ vanish at twist-two level, for instance,
$$_0^{2\pi }𝑑\varphi _r\frac{\mathrm{\Delta }_{\mathrm{SL}}d\sigma }{d\varphi _r}=𝒪\left(1/𝒬^2\right)\text{ or }_0^{2\pi }𝑑\varphi _r\frac{\mathrm{\Delta }_\mathrm{C}^{\mathrm{unp}}d\sigma }{d\varphi _r}=𝒪\left(1/𝒬^2\right).$$
(46)
## 4 Numerical estimates.
To give phenomenological predictions for the asymmetries discussed above we have to specify models for the SPDs. Definitely, this is the main source of uncertainty for the numerical estimates we present in this section. (Of course, the primary goal of experiments is rather to constrain the skewed functions via the theoretical formulae.)
Let us discuss in turn spin non-flip and flip functions. In the former case SPDs have a definite limit for forward kinematics when they reduce to the familiar parton densities. For spin non-flip SPDs we choose an oversimplified factorized form of the $`\mathrm{\Delta }^2`$ and $`(t,\xi )`$ dependence of the SPDs:
$$H^i(t,\xi ,\mathrm{\Delta }^2,𝒬^2)=F_1^i(\mathrm{\Delta }^2)q^i(t,\xi ,𝒬^2),\stackrel{~}{H}^i(t,\xi ,\mathrm{\Delta }^2,𝒬^2)=G_1^i(\mathrm{\Delta }^2)\mathrm{\Delta }q^i(t,\xi ,𝒬^2),$$
(47)
where $`i`$ denotes the quark flavour. Here $`F_1^i(\mathrm{\Delta }^2)`$ and $`G^i(\mathrm{\Delta }^2)`$ are elastic parton form factors normalized to unity at the origin, $`F_1^i(0)=G^i(0)=1`$, and $`q(t,\xi ,𝒬^2)`$ as well as $`\mathrm{\Delta }q(t,\xi ,𝒬^2)`$ are the non-forward functions specified below. The support of these functions is $`[1,1]`$, where for $`t>0`$ we have the quark distribution and for $`t<0`$ the antiquark distribution, i.e. $`\overline{q}(t,\xi ,𝒬^2)=q(t,\xi ,𝒬^2)`$ and $`\mathrm{\Delta }\overline{q}(t,\xi ,𝒬^2)=\mathrm{\Delta }q(t,\xi ,𝒬^2)`$. The normalization of $`H^i(t,\xi ,\mathrm{\Delta }^2,𝒬^2)`$ and $`\stackrel{~}{H}^i(t,\xi ,\mathrm{\Delta }^2,𝒬^2)`$ at $`\mathrm{\Delta }^2=0`$, $`\xi =0`$ is determined by parton densities $`H^i(t,0,0)=q^i(t)`$, $`\stackrel{~}{H}^i(t,0,0)=\mathrm{\Delta }q^i(t)`$. Since the $`\xi `$ dependence drops out in the first moment we can constrain the $`\mathrm{\Delta }^2`$ dependence by sum rules, e.g.
$$_1^1𝑑tH^i(t,\xi ,\mathrm{\Delta }^2)=F_1^i(\mathrm{\Delta }^2),_1^1𝑑tE^i(t,\xi ,\mathrm{\Delta }^2)=F_2^i(\mathrm{\Delta }^2),$$
(48)
with the Dirac and Pauli form factor, respectively. For non-polarized SPDs the valence $`u`$ and $`d`$ quark form factors in the proton can be easily deduced from (6) via $`F_I^{(\genfrac{}{}{0pt}{}{p}{n})}=2\left(\genfrac{}{}{0pt}{}{Q_u}{Q_d}\right)F_I^u+\left(\genfrac{}{}{0pt}{}{Q_d}{Q_u}\right)F_I^d`$ which results in
$$2F_I^u(\mathrm{\Delta }^2)=2F_I^p(\mathrm{\Delta }^2)+F_I^n(\mathrm{\Delta }^2),F_I^d(\mathrm{\Delta }^2)=F_I^p(\mathrm{\Delta }^2)+2F_I^n(\mathrm{\Delta }^2),\text{for}I=1,2.$$
(49)
For $`s`$ (or in general, sea) quark contribution in the parity even sector the first moment vanishes and the sum rule (48) does not give any constraint. Nevertheless, the counting rule for elastic form factors tells us that for large $`\mathrm{\Delta }^2`$ we have a $`(\mathrm{\Delta }^2)^3`$ behaviour. This suggests the following dipole fit with the mass cutoff $`m_V`$ chosen as for valence quarks:
$$G_E^{\mathrm{sea}}(\mathrm{\Delta }^2)=\frac{1}{1+\kappa _{\mathrm{sea}}}G_M^{\mathrm{sea}}(\mathrm{\Delta }^2)=\left(1\frac{\mathrm{\Delta }^2}{m_V^2}\right)^3,$$
(50)
where $`\kappa _{\mathrm{sea}}`$ will be specified below.
So far the model has been governed by the known forward densities. Unfortunately, a similar reduction is absent for the helicity flip amplitudes. For $`E^i`$ we adopt nevertheless
$$E^i(t,\xi ,\mathrm{\Delta }^2)=r^i(t,\xi )F_2^i(\mathrm{\Delta }^2)\text{with}r^i(t,\xi )=q^i(t,\xi ).$$
(51)
The identification of $`r^i(t,\xi )`$ with $`q^i(t,\xi )`$ ensures the sum rule for valence quark contributions. Note that the parameter $`\kappa _{\mathrm{sea}}`$ normalizes the sea quark contribution, for instance, $`\kappa _{\mathrm{sea}}=0`$ provides $`E^{\mathrm{sea}}=0`$. In the axial vector channel the quark form factors can be read off from the iso-triplet axial form factor $`G_1^{(3)}(\mathrm{\Delta }^2)`$ of the $`\beta `$-decay and related by isotopic symmetry to the form factor in question. The decay constant $`g_A^{(3)}`$ is expressed by Goldberger-Treiman relation $`g_A^{(3)}\frac{1}{\sqrt{2}M}f_\pi g_{\pi NN}`$ in terms of the pion decay constant $`f_\pi `$, nucleon mass $`M`$ and $`\pi NN`$-coupling, $`g_{\pi NN}`$, and has the numerical value $`g_A^{(3)}=1.267`$ . If we assume the same $`\mathrm{\Delta }^2`$-dependence for the iso-singlet $`G_1^{(0)}(\mathrm{\Delta }^2)`$ with the same cutoff and a constant $`g_A^{(0)}`$ we get for quarks
$$G_1^i(\mathrm{\Delta }^2)=\left(1\frac{\mathrm{\Delta }^2}{m_A^2}\right)^2,\text{for}i=\{u_v,d_v\},\text{and}G_1^{\mathrm{sea}}(\mathrm{\Delta }^2)=\left(1\frac{\mathrm{\Delta }^2}{m_A^2}\right)^3$$
(52)
with the scale $`m_A=0.9\mathrm{GeV}`$ .
Finally for the polarized spin-flip amplitude it was observed that, similar to the $`\beta `$-decay effective pseudoscalar form factors , one can approximate $`\stackrel{~}{E}`$ at small $`\mathrm{\Delta }^2`$ by the pion pole so that one ends up with the model
$$\stackrel{~}{E}^u=\stackrel{~}{E}^d=\frac{1}{2\xi }\theta \left(\xi |t|\right)\varphi _\pi \left(t/\xi \right)g_\pi (\mathrm{\Delta }^2),\text{with}g_\pi (\mathrm{\Delta }^2)=\frac{4g_A^{(3)}M^2}{m_\pi ^2\mathrm{\Delta }^2}\text{and}\varphi _\pi (x)=\frac{4}{3}(1x^2).$$
(53)
It is reliable to assume that the SPDs in the DGLAP region can be modelled by the forward parton distributions measured in inclusive reactions. In the simplest case we assume that this is the only contribution. We refer to this model as the forward parton distribution (FPD) model which has no skewedeness dependence at the input scale $`𝒬_0`$:
$$(\mathrm{\Delta })q^i(t,\xi ,𝒬_0^2)=(\mathrm{\Delta })q^i(t,𝒬_0^2)\text{for}t>0.$$
(54)
The contributions for $`t<0`$ are easily restored by means of symmetry. Although the input distribution does not depend on $`\xi `$ a small evolution step does generate such a $`\xi `$-dependence . A further model is based on an proposal for the so-called double distribution (DD) , namely,
$$q(t,\xi ,Q^2)=_1^1𝑑x_{1+|x|}^{1|x|}𝑑y\delta (x+\xi yt)f(y,x,Q^2).$$
(55)
The functional dependence in the $`x`$-subspace is given by the shape of the forward parton density, $`f(x)`$, while the $`y/[1|x|]`$-dependence of the integrand has to be similar to that of the distribution amplitude. Thus, $`f(y,x)`$ is given by the product of a forward distribution $`f(z)`$ (more precisely $`q(z)`$ for quarks and $`zg(z)`$ for gluons) with a profile function $`\pi `$
$$f(y,x)=\pi (y,x)f(x),$$
(56)
where $`\pi `$ for quarks and gluons is given by
$$\pi ^Q(y,x)=\frac{3}{4}\frac{[1|x|]^2y^2}{[1|x|]^3},\pi ^G(y,x)=\frac{15}{16}\frac{\left\{[1|x|]^2y^2\right\}^2}{[1|x|]^5}.$$
(57)
Now we are in a position to give numerical estimates for the cross sections defined in the preceding sections.
### 4.1 HERA kinematics.
In the small $`x`$ kinematics, $`10^5x10^2`$, first estimates for the unpolarized azimuthal and single spin asymmetry has been given in the factorization and BFKL approach. In the following we evaluate numerically different asymmetries in order to demonstrate that $`_1`$, $`\stackrel{~}{}_1`$ and $`_1`$ are measurable in HERA experiments. In the following we deal with the FPD model, where we equate the SPDs with the parton densities taking the MRS A and the GS A parametrization at the input scale $`𝒬^2=4\text{ GeV}^2`$ and using $`\overline{u}=\overline{d}=\overline{s}/2`$ as well as $`\mathrm{\Delta }\overline{u}=\mathrm{\Delta }\overline{d}=\mathrm{\Delta }\overline{s}`$. For simplicity we do not discuss the $`𝒬^2`$ dependence of our predictions due to the perturbative evolution.
The unpolarized azimuthal asymmetry is defined by
$`A={\displaystyle \frac{_{\pi /2}^{\pi /2}𝑑\varphi _r\frac{d\sigma ^{\mathrm{unp}}}{d\varphi _r}_{\pi /2}^{3\pi /2}𝑑\varphi _r\frac{d\sigma ^{\mathrm{unp}}}{d\varphi _r}}{_0^{2\pi }𝑑\varphi _r\frac{d\sigma ^{\mathrm{unp}}}{d\varphi _r}}},`$ (58)
and its first signature has been seen by the ZEUS collaboration . In the approximation (27) the unpolarized squared BH term is $`\varphi _r`$ independent. However, it turned out that this approximation may provide misleading results due to the neglected azimuthal dependence of the BH process. In Fig. 3(a) it can be seen that the dropped terms cause a strong $`\varphi _r`$-dependence and that only for the Pade-type approximation introduced in section 2 \[see Eq. (2) for the expansion of the propagator\] there is a good agreement with the exact expression. Since the BH cross section is in general suppressed in the upper hemisphere, one may expect quite different predictions depending on the approximations involved. Indeed, for small values of $`x`$, where $`y=𝒬^2/xs`$ with the center of mass energy $`s=427.5820\text{ GeV}^2`$, we find a strong deviation, even for small $`\mathrm{\Delta }^2`$. Note that the interference term is only taken into account up to order $`1/\sqrt{\mathrm{\Delta }^2𝒬^2}`$ and at present it is not clear whether the $`1/𝒬^2`$ term is crucial or not.
To get information on $`\left\{F_1_1\frac{\mathrm{\Delta }^2}{4M^2}F_2_1\right\}`$ in a cleaner way one may use the azimuthal asymmetry of the charge asymmetry $`\mathrm{\Delta }_C^{\mathrm{unp}}d\sigma =d{}_{}{}^{}\sigma _{}^{\mathrm{unp}}d{}_{}{}^{+}\sigma _{}^{\mathrm{unp}}`$ defined in Eq. (4) (in the following we restore the $`\mathrm{\Delta }_{\mathrm{min}}^2/\mathrm{\Delta }^2`$ dependence in the interference amplitudes):
$`A_\mathrm{C}={\displaystyle \frac{_{\pi /2}^{\pi /2}𝑑\varphi _r\frac{\mathrm{\Delta }_C^{\mathrm{unp}}d\sigma }{d\varphi _r}_{\pi /2}^{3\pi /2}𝑑\varphi _r\frac{\mathrm{\Delta }_C^{\mathrm{unp}}d\sigma }{d\varphi _r}}{_0^{2\pi }𝑑\varphi _r\frac{d{}_{}{}^{}\sigma _{}^{\mathrm{unp}}+d{}_{}{}^{+}\sigma _{}^{\mathrm{unp}}}{d\varphi _r}}},`$ (59)
and the single (lepton) spin asymmetry with unpolarized target $`\mathrm{\Delta }_{\mathrm{SL}}=d\sigma ^{}d\sigma ^{}`$ defined in Eq. (1):
$`A_{\mathrm{SL}}={\displaystyle \frac{_0^\pi 𝑑\varphi _r\frac{\mathrm{\Delta }_{\mathrm{SL}}d\sigma }{d\varphi _r}_\pi ^{2\pi }𝑑\varphi _r\frac{\mathrm{\Delta }_{\mathrm{SL}}d\sigma }{d\varphi _r}}{_0^{2\pi }𝑑\varphi _r\frac{d\sigma ^{}+d\sigma ^{}}{d\varphi _r}}}.`$ (60)
The former (later) one is proportional to the real (imaginary) part of the linear combination that was mentioned. Both of them give sizeable effects. Although $`A_{\mathrm{SL}}`$ is proportional to the imaginary part which is growing as $`x^1`$ this asymmetry is suppressed by a kinematical factor $`y`$ as compared to the charge asymmetry. Therefore, we find in the considered kinematics a two times larger value of the charge asymmetry as for the single spin one. This should be reversed for larger values of $`y`$ (see also the discussion in Ref. ). In Fig. 4 we demonstrate the influence of the amplitude $`_1`$ which is quite sizeable for larger values of $`\mathrm{\Delta }^2`$, where $`\mathrm{\Delta }^2/𝒬^2`$ remains small.
To measure $`\stackrel{~}{}_1`$ one have to consider the proton single spin asymmetry $`\mathrm{\Delta }_{\mathrm{SLN}}=d\sigma _{}d\sigma _{}`$ given in Eq. (2):
$`A_{\mathrm{SLN}}={\displaystyle \frac{_0^\pi 𝑑\varphi _r\frac{\mathrm{\Delta }_{\mathrm{SLN}}d\sigma }{d\varphi _r}_\pi ^{2\pi }𝑑\varphi _r\frac{\mathrm{\Delta }_{\mathrm{SLN}}d\sigma }{d\varphi _r}}{_0^{2\pi }𝑑\varphi _r\frac{d\sigma _{}+d\sigma _{}}{d\varphi _r}}}`$ (61)
and the charge asymmetry in double spin-flip experiments, i.e. $`\mathrm{\Delta }_\mathrm{C}^{\mathrm{DFL}}d\sigma =d{}_{}{}^{}\sigma _{}^{}d{}_{}{}^{+}\sigma _{}^{}\mathrm{\Delta }_\mathrm{C}d\sigma ^{\mathrm{unp}}`$, defined in Eq. (5)
$`A_\mathrm{C}^{\mathrm{DFL}}={\displaystyle \frac{_0^\pi 𝑑\varphi _r\frac{\mathrm{\Delta }_\mathrm{C}^{\mathrm{DFL}}d\sigma }{d\varphi _r}_\pi ^{2\pi }𝑑\varphi _r\frac{\mathrm{\Delta }_\mathrm{C}^{\mathrm{DFL}}d\sigma }{d\varphi _r}}{_0^{2\pi }𝑑\varphi _r\frac{d{}_{}{}^{}\sigma _{}^{\mathrm{unp}}+d{}_{}{}^{+}\sigma _{}^{\mathrm{unp}}}{d\varphi _r}}}.`$ (62)
Unfortunately, for the considered $`𝒬^2`$ value we find for our model that this single spin asymmetry is compatible with zero. Although, we have an enhancement for small $`y`$ due do the factor $`1/y`$ in comparison to the lepton single spin asymmetry, this enhancement cannot compensate the weaker rise of $`\stackrel{~}{}_1`$ for small $`x`$. For larger value of $`𝒬`$, i.e. also larger $`x`$, and $`\mathrm{\Delta }^2`$ and small $`y`$ we found a few percent effect. Note also that the ratio of proton to lepton single spin asymmetry gives us the ratio $`(\mathrm{Im}\stackrel{~}{}_1)/(\mathrm{Im}_1)`$ times $`1/y`$, which is sizeable at not too small $`x`$. The azimuthal asymmetry in double spin flip experiments normalized to the unpolarized cross section is tiny. However, if we would compare it with the extracted double spin flip part, we would get quite sizeable effects which are of the order of 10% or even more. Furthermore, to explore $`_1`$ separately, one can make use of the discussion given in section 3.1. Especially, for small $`y`$ the BH cross section is suppressed by $`y^2`$ in comparison to the DVCS one and, thus, the former one drops out.
Let us note that the perturbative NLO corrections to the imaginary part for small $`x`$ are of order of 20% for each quark species as well as for the gluon contribution.
### 4.2 HERMES kinematics.
Now let us turn to the HERMES experiment with a $`E=27.5\mathrm{GeV}`$ positron beam scattered by a hydrogen target and give predictions for the asymmetries which can be accesed there. To give numerical predictions we take the model for SPDs deduced from the double distribution model and forward (un-) polarized parton densities from Ref. () with $`\kappa _{\mathrm{sea}}=2.`$ As a starting point we choose $`𝒬^2=4\mathrm{GeV}^2`$, $`x`$-range $`0.10.4`$, and the $`t`$-channel momentum transfer $`\mathrm{\Delta }^2=0.10.5\mathrm{GeV}^2`$. Since the parton densities are defined at rather low momentum scale $`𝒬_0^2=0.23\mathrm{GeV}^2`$ we evolve them up to the experimental scale using the formalism of . We concentrate mostly on the cross sections proportional to the imaginary part of the DVCS amplitude. Since $`\stackrel{~}{E}_1`$ is concentrated in the ER-BL region with zero at $`|t|=\xi `$, $`\mathrm{Im}\stackrel{~}{}_1`$ vanishes at all order.
Similarly to the previous section we calculate the charge asymmetry for unpolarized experiments given in Eq. (59). As we see form the Fig. 5(a) it reaches the level of $`515\%`$ for $`\mathrm{\Delta }^2=0.5\mathrm{GeV}^2`$. The single lepton spin asymmetry (60) is much larger and can be as big as $`2030\%`$ \[see Fig. 5(b)\] which gives promises to measure it experimentally. Note that in both cases the cross sections are very sensitive to the helicity flip distribution $`E`$ which is responsible for the orbital momentum of partons in the proton.
In view of limited space let us consider finally only the spin asymmetry for a longitudinally and transversally polarized proton (3). In the former case azimuthal averaging of Eq. (2) is the same as in Eq. (60) and leads to a sizeable asymmetry of order 20%. In the later case in order to extract the combination of SPDs multiplied by $`\mathrm{cos}(\varphi _r/2)`$ we define the following azimuthal asymmetry
$`A_{\mathrm{STN}}={\displaystyle \frac{_{\pi /3}^{2\pi /3}𝑑\varphi _r\frac{\mathrm{\Delta }_{\mathrm{STN}}d\sigma }{d\varphi _r}_{2\pi /3}^{5\pi /3}𝑑\varphi _r\frac{\mathrm{\Delta }_{\mathrm{STN}}d\sigma }{d\varphi _r}}{_0^{2\pi }𝑑\varphi _r\frac{d\sigma _{}+d\sigma _{}}{d\varphi _r}}}.`$ (63)
The numerical estimate presented in Fig. 5 (d) demonstrates that it has a sizable effect which in contrast to the other symmetries does not vanish at the kinematical boundery $`\mathrm{\Delta }^2=\mathrm{\Delta }_{\mathrm{min}}^2`$.
## 5 Conclusions.
In this paper we have given theoretical predictions for diverse asymmetries which can be measured in exclusive leptoproduction experiments of a real photon. Our estimates are rather encouraging, since they demonstrate a possibility to separate the contributions coming from different leading twist-two DVCS amplitudes by means of polarized lepton beam, and longitudinally and transversally polarized targets. This is the first step on the way to constrain the form of the SPDs from experimental data. The models used for numerical estimates lead to large charge and single (lepton) spin asymmetries of order $`20\%`$. However, since an asymmetry gives information on a linear combination of SPD one has to resort to other combinations of cross sections in order to disentangle a given distribution.
We have not discussed in the present paper the NLO correction to the DVCS amplitudes. However, as has been shown in Ref. the latter could be very sizable and therefore change the LO predictions significantly in the valence quark region. Yet another important issue is the study of kinematical and dynamical higher twist corrections. Both of these problems deserve a detailed investigation and will be considered elsewhere.
We would like to thank M. Amarian, M. Diehl, and A.V. Radyushkin for constuctive correspondence. This work was supported by DFG and BMBF. D.M. thanks G. Sterman for the hospitality extended to him at the C.N. Yang Institute for Theoretical Physics while this work was finished.
|
warning/0004/physics0004041.html
|
ar5iv
|
text
|
# Structures and Stabilities of CaO and MgO Clusters and Cluster Ions: An alternative interpretation of the experimental mass spectra.
## I Introduction
Small clusters are of great interest both to the physical and chemical communities because of their numerous potential applications (for example, in nanoelectronics or catalysis), and also because one can gain important insight into the evolution from atomic and molecular properties to bulk and surface properties. To have a knowledge of the structures adopted by the clusters is of paramount importance, as many interesting cluster properties are largely determined by them. From the theoretical side, finding the lowest energy structure for each cluster is a complicated matter, because the number of isomers increases exponentially with cluster size. Other reasons are that one has to treat bulklike and surfacelike ions on an equal footing, and that the number of ions to be explicitely considered in a cluster is larger than in a bulk or surface study, where symmetry restrictions impose a number of useful atomic equivalences. From the experimental side, the problem is so difficult that during the approximately 30 years of intensive cluster research, the main source of structural information has been theory. Very recently, experimental techniques like electron diffraction from trapped clusters or measurements of cluster mobilities have been succesfully applied to study the structures of covalent and ionic clusters. Photoelectron spectroscopy has also been applied to study isomerization transitions in small alkali halide clusters, and measurements of ionization potential to detect structural transitions in barium oxide clusters. At the moment, however, these techniques need parallel theoretical calculations to make a definite assignment of the observed diffraction pattern, mobility or ionization potential to a specific isomer geometry.
A large amount of theoretical work has been devoted to metallic, semiconductor and noble gas clusters. The work on ionic materials has been centered mostly in the family of alkali halides, and studies of metal oxide clusters have been comparatively scarce, despite their importance in many branches of surface physics, like heterogeneous catalysis or corrotion. Saunders reported mass spectra and collision induced fragmentation data for stoichiometric (MgO)$`{}_{}{}^{+}{}_{n}{}^{}`$ and (CaO)$`{}_{}{}^{+}{}_{n}{}^{}`$ cluster ions, Martin and Bergmann published mass spectra of (CaO)<sub>n</sub>Ca<sup>2+</sup> cluster ions, and Ziemann and Castleman performed experimental measurements of several singly- and doubly-ionized cluster ions of MgO and CaO by using laser-ionization time-of-flight mass spectrometry. Theoretical calculations have been performed at different levels of accuracy: simple ionic models based on phenomenological pair potentials were used by Ziemann and Castleman to explain the global trends found in their experiments; Wilson has studied neutral (MgO)<sub>n</sub> (n$``$30) clusters by using a compressible-ion model that includes coordination-dependent oxide polarizabilities; semiempirical tight-binding calculations for MgO clusters were reported by Moukouri and Noguera; finally, ab initio calculations on stoichiometric MgO clusters have been presented recently by Recio et al., Malliavin and Coudray, Li et al, and de la Puente et al, and calculations on stoichiometric (Li<sub>2</sub>O)<sub>n</sub> clusters have been reported by Finocchi and Noguera. Regarding the nonstoichiometric cluster ions, Aguado et al. have studied the structures and stabilities of (MgO)<sub>n</sub>Mg<sup>2+</sup>.
Trying to find an interpretation of the obtained mass spectra, Ziemann and Castleman performed some simple pair potential calculations of the structures of (MgO)<sub>n</sub>Mg<sup>2+</sup> cluster ions by using a rigid ion model. The conclusion of those calculations was that the magic numbers can be explained in terms of highly compact structures that can only be obtained for certain cluster sizes, an interpretation very similar to that found in the closely related case of alkali halides. In our previous work, we showed that the structures of (MgO)<sub>n</sub>Mg<sup>2+</sup> cluster ions were quite different from those of alkali halides. Specifically, the influence of the large and coordination dependent polarizabilities of oxide anions (not included in the rigid ion model) favors the formation of surface oxide sites, and thus structures with bulk oxide anions (coordination 6) are not energetically competitive until large values of the number of molecules n are attained. For example, a highly compact 3$`\times `$3$`\times `$3 cube (where the notation denotes the number of atoms along three perpendicular edges) is not the ground state of (MgO)<sub>13</sub>Mg<sup>2+</sup>. Nevertheless, the agreement between the magic numbers obtained through an examination of the stabilities of the clusters against the loss of an MgO molecule and the experimental ones is complete. It is just the interpretation of them in terms of structures that is different, that is, model-dependent. It is interesting to study a similar system like calcium oxide in order to assess whether those trends are a general feature of alkaline-earth oxide clusters or not. Moreover, the experiments of Saunders suggest interesting structural differences between both materials, as the main fragments observed after collisions with inert gas ions were (MgO)<sub>3</sub> in one case and (CaO)<sub>2</sub> in the other, and the mass spectra of Ziemann and Castleman show different stabilities in the small size regime (magic numbers at n=5,8,11 for (MgO)<sub>n</sub>Mg<sup>2+</sup> and at n=5,7,9,11 for (CaO)<sub>n</sub>Ca<sup>2+</sup>), providing further motivation for our study. From the theoretical point of view, Ca<sup>2+</sup> is larger than Mg<sup>2+</sup>, so we can expect ionic size packing effects to play an important role in determining structural differences. Besides, Ca<sup>2+</sup> has a polarizability approximately 6 times larger than Mg<sup>2+</sup>, and the polarizabilities of the oxide anions are also larger in CaO because the bonding is weaker than in MgO.
In this work we present the results of an extensive and systematic study of (CaO)<sub>n</sub>Ca<sup>2+</sup> cluster ions with n up to 29, and of (MgO)<sub>n</sub> and (CaO)<sub>n</sub> neutral stoichiometric clusters with n=3,6,9,12,15,18. The rest of the paper is organized as follows: in Section II we give a brief resume of the theoretical model employed, as full an exposition has been reported already in previous works. The results are presented in Section III, and the main conclusions to be extracted from our study in Section IV.
## II The aiPI model and polarization corrections
The theoretical foundation of the ab initio perturbed ion model lies in the theory of electronic separability, and its practical implementation in the Hartree-Fock (HF) version of the theory of electronic separability. Very briefly, the HF equations of the cluster are solved stepwise, by breaking the cluster wave function into local group functions (ionic in nature in our case). In each iteration, the total energy is minimized with respect to variations of the electron density localized in a given ion, with the electron densities of the other ions kept frozen. In the subsequent iterations each frozen ion assumes the role of nonfrozen ion. When the self-consistent process finishes, the outputs are the total cluster energy and a set of localized wave functions, one for each geometrically nonequivalent ion of the cluster. These localized cluster-consistent ionic wave functions are then used to estimate the intraatomic correlation energy correction through Clementi’s Coulomb-Hartree-Fock method. The large multi-zeta basis sets of Clementi and Roetti are used for the description of the ions. At this respect, our optimizations have been performed using basis sets (5s4p) for $`Mg^{2+}`$ and (5s5p) for $`O^2`$, respectively. Inclusion of diffuse basis functions has been checked and shown unnecessary. One important advantage coming from the localized nature of the model is the linear scaling of the computational effort with the number of atoms in the cluster. This has allowed us to study clusters with as many as 59 atoms at a reasonable computational cost.
In our previous work on alkaline-earth oxide clusters, we concluded that the aiPI model is equivalent to a first-principles version of the semiempirical breathing shell model. The binding energy of the cluster can be written as a sum of deformation and interaction terms
$$E_{bind}=\underset{R}{}E_{bind}^R=\underset{R}{}(E_{def}^R+\frac{1}{2}E_{int}^R).$$
(1)
where the sum runs over all ions in the cluster. The interaction energy term is of the form
$$E_{int}^R=\underset{SR}{}E_{int}^{RS}=\underset{SR}{}(E_{class}^{RS}+E_{nc}^{RS}+E_X^{RS}+E_{overlap}^{RS}),$$
(2)
where the different energy contributions are: the classical electrostatic interaction energy between point-like ions; the correction to this energy due to the finite extension of the ionic wave functions; the exchange interaction energy between the electrons of ion R and those of the other ions in the cluster; and the overlap repulsive energy contribution. The deformation energy term $`E_{def}^R`$ is the self-energy of the ion R. It is an intrinsically quantum-mechanical many-body term that accounts for the energy change associated to the compression of the ionic wave functions upon cluster formation, and incorporates the correlation contribution to the binding energy. As the model assumes, for computational simplicity, that the ion densities have spherical symmetry, the only relevant terms that are lacking from the ab initio description are the polarization terms. In a polarizable point-ion approximation, the polarization contribution to the deformation and interaction energies is
$`E_{int}^{RS,pol}={\displaystyle \frac{q_R(\stackrel{}{\mu }_S\stackrel{}{r}_{RS})}{r_{RS}^3}}{\displaystyle \frac{q_S(\stackrel{}{\mu }_R\stackrel{}{r}_{RS})}{r_{RS}^3}}`$ (3)
$`3{\displaystyle \frac{(\stackrel{}{\mu }_R\stackrel{}{r}_{RS})(\stackrel{}{\mu }_S\stackrel{}{r}_{RS})}{r_{RS}^5}}+{\displaystyle \frac{(\stackrel{}{\mu }_R\stackrel{}{\mu }_S)}{r_{RS}^3}},`$ (4)
$$E_{def}^{R,pol}=\frac{\mu _R^2}{2\alpha _R},$$
(5)
where $`\alpha _R`$ is the polarizability of the ion R, and $`\stackrel{}{\mu }_R`$ the dipole moment induced on ion R. The new terms added to the interaction energy are the monopole-dipole and dipole-dipole interaction energy terms. The term added to the deformation energy represents the energy cost of deforming the charge density of the ion to create the dipole moment. The point-ion approximation provides just the asymptotic part of the polarization interaction energy, that is, it is exact only for large ionic separations. As soon as the ions begin to overlap, there is an important short-range contribution to the induced dipole moments, which is of opposite sign to the asymptotic limit for anions and may in some specific cases reverse the sign of the asymptotic value. These effects can be easily acomodated in the formalism by substituting the asymptotic value of the induced dipole moments by the following expression:
$$\mu _\alpha ^{total.R}=\mu _\alpha ^{asymp,R}+\mu _\alpha ^{sr,R},$$
(6)
with
$$\mu _\alpha ^{asymp,R}=\alpha ^R\underset{SR}{}\frac{r_{RS,\alpha }}{r_{RS}^3}q^R,$$
(7)
$$\mu _\alpha ^{sr,R}=\alpha ^R\underset{SR}{}\frac{r_{RS,\alpha }}{r_{RS}^3}f(r_{RS}),$$
(8)
where $`sr`$ stands for “short-range” and $`\mu _\alpha ^{total.R}`$ is the $`\alpha `$ component of the dipole moment vector induced on ion R. The physics behind the short-range polarization correction has been explained in Ref. , and is associated to the finite extents of the electron densities of the anions and cations. Then, $`f`$ is a short-range function that switchs on as the cation-anion overlap becomes appreciable. Madden and coworkers have employed the Tang and Toennies dispersion damping function as a suitable form for $`f`$:
$$f(r_{RS})=c\underset{k=0}{\overset{k_{max}}{}}\frac{b^k}{k!}e^{br_{RS}}.$$
(9)
This is a smoothed step function passing from zero for large $`r`$ to $`c`$ for r=0. The range of $`r`$ values at which $`f`$ becomes significantly different from zero is primarily determined by the range parameter $`b`$.
We have included the polarization terms in the energy calculation with this parameterised method, that calculates the induced dipole moments from eq. (5) and the correction to the deformation and interaction energies from eqs. (3) and (4), respectively. The “enlarged” aiPI+polarization model thus obtained accounts for all the relevant physical interactions. The relaxation of the assumption of spherical symmetry being computationally expensive, the price to be paid is the inclusion in the model of a set of parameters, namely, the polarizabilities $`\alpha _R`$ and the range parameter $`b`$. Appropriate values for the other two constants $`c`$ and $`k_{max}`$ can be taken equal to the bulk values ($`c`$=-3 and $`k_{max}`$=4). Given the meaning of the range parameter $`b`$, inversely related to half the interionic distance between first neighbors, one might expect different values of $`b`$ for clusters as compared to bulk materials if the interionic distances are substantially different. As a matter of fact, the evolution of those distances with cluster size is not too complicated in the case of ionic clusters. Specifically, the average interionic distance $`d`$ initially increases quite abruptly with the number of molecules n, and then slowly approaches the bulk limit. As a consequence of this behaviour, we will see that the bulk value ($`b`$=0.75 a.u.) is appropriate for all (CaO)<sub>n</sub>Ca<sup>2+</sup> clusters with $`n4`$. Different values of $`b`$ are needed just for $`n<`$4 to avoid overpolarization problems. Regarding the polarizabilities, oxide anions have the interesting property of showing strongly coordination-dependent values. In fact, the O<sup>2-</sup> anion does not exist as a free ion, which is equivalent to an infinite polarizability; in the solid phase it is stabilized by the crystal environment and has a finite material-dependent polarizability. Wilson has interpolated between those two limits and gives values for the coordination-dependent values of $`\alpha (O^2)`$ in MgO. We have assumed that the ratio of the bulk oxide polarizabilities for MgO and CaO ($`\alpha ^{bulk}(O^2:CaO)`$/$`\alpha ^{bulk}(O^2:MgO)`$=1.469) is independent of the oxide coordination, and have deduced the $`\alpha `$ values for CaO from those of MgO. This procedure is justified because the cation size does not change appreciably with coordination number. For the calcium cation we take the bulk polarizability (3.193 a.u.)
We close this section with a consideration of several criticisms that could be raised against (and of the advantages of) the employed methodology. We have chosen a mixed ab initio/semiempirical energy model in order to obtain a good compromise between computational efficiency and accuracy. All the relevant energy terms excluding polarization are described with an ab initio methodology. To include polarization, we have used an accurate model, where the parameters have been fitted by a comparison to ab initio calculations. Special care has been devoted to the separation of all the independent physical factors that influence a given quantity, thus avoiding a mixing of different effects in a single parameter and enhancing the transferability of the model. The good parameterisation is reflected in the fact that parameters can be transfered between closely related systems (like, for example, different metal oxides) by simple scaling arguments involving ionic radii. Thus, we think that the reliability of our calculations is reduced just a little compared to full ab initio methodologies. To support this expectation, we made a comparison with DMOL calculations performed on neutral (CaO)<sub>n</sub> clusters by Malliavin and Coudray. All the interionic distances were in agreement to their calculations up to differences of 4 %. The energetic ordering of the isomers, as well as the specific energy differences, are reproduced with a maximum error of 5 %. We believe that this is a very reasonable agreement, even more if we realize that we are neglecting dispersion interactions, and polarization interactions beyond the dipolar terms. The solid MgO is excellently described with the aiPI model (at least in its static properties). The model is then expected to transfer properly between both limits. The larger computational simplicity has been exploited to study large cluster sizes (up to 59 ions) with full relaxations of the geometries. Moreover, for each cluster size, a large number of isomers (between 10 and 15) have been investigated. The generation of the initial cluster geometries was accomplished by using a pair potential, as we explained in our previous publication. The optimization of the geometries has been performed by using a downhill simplex algorithm.
## III Results and Discussion
### A Structural Trends in (CaO)<sub>n</sub>Ca<sup>2+</sup> Cluster Ions
In Fig.1 we present the optimized aiPI+polarization structures of the ground state (GS) and lowest lying isomers or (CaO)<sub>n</sub>Ca<sup>2+</sup> (n=4–29) cluster ions. Below each isomer we show the energy difference (in eV) with respect to the ground state. For n$`<`$4, the clusters are not detected in the experiments, probably because they undergo a Coulomb explosion driven by the excess charge and the small cluster size, but we are not interested here in this aspect of the experiments. From n=4 to n=10, there is a predominance of m$`\times `$2$`\times `$2 fragments (m=2–5), that is, the (CaO)<sub>2</sub> subunit appears as the basic building block. The total number of ions in these nonstoichiometric clusters is an odd number, and thus those structures are never perfectly compact. There is either an extra cation added to or a missing anion removed from the perfect structure. Less compact structures as for example planar fragments are not energetically competitive. The structures in this size range tend to be elongated as a direct consequence of the excess cluster charge. When n=9, a 3$`\times `$3$`\times `$2+1 fragment is more stable than that based on the (CaO)<sub>2</sub> building block, and n=10 is the largest cluster size for which a fragment of this kind is the ground state. Another thing to be pointed out is that in this size range, the extra cation present in the m$`\times `$2$`\times `$2+1 structures induces a larger cluster distortion than the missing anion in the m$`\times `$2$`\times `$2-1 structures. We will see that this feature has important implications in the stability of the clusters.
From n=11 to n=15, the dominant fragments are based on m$`\times `$3$`\times `$2 units. For n=16 and 17, the most stable isomers are m$`\times `$4$`\times `$2 fragments. None of these structures has still developed an anion with full bulk coordination. In particular, the 3$`\times `$3$`\times `$3 isomer for n=13, which is particularly stable in the case of nonstoichiometric alkali halide cluster ions, does not even appear in Fig. 1. The large coordination-dependent values of the polarizabilities of the oxide anions favors the formation of surface sites, and gives rise to somewhat less compact ground state structures, for which the increase in dipolar energy compensate for the decrease in Madelung energy. The 3$`\times `$3$`\times `$3 (CaO)<sub>13</sub>Ca<sup>2+</sup> is specially unfavored by the dipolar energy terms because it has a central oxide anion with bulk coordination (so with a comparatively low polarizability), and another 12 anions with coordination 4. On the contrary, the largest coordination in the GS structure is five, and some three-coordinated anions (in corner positions) also appear, inducing a large dipolar energy stabilization. For n=18 and 19 there is a glimpse of a transition to more compact cluster structures. The important feature of the GS structures of these two cluster sizes, compared to the 3$`\times `$3$`\times `$3 for n=13, is that now there are oxide anions in corner positions. These make a large contribution to the polarization energy term, that added to the increased Madelung energy of a compact fragment, gives a total GS energy more negative than that of m$`\times `$3$`\times `$2 or m$`\times `$4$`\times `$2 structures. Nevertheless, the energy differences between isomers are small, and from n=20 on, ground state isomers without bulk anions are again obtained (n=24 and 27 are the only relevant exceptions, because the ground states of n=26 and n=29 can be considered degenerate within the accuracy of our theoretical model).
A general feature of (CaO)<sub>n</sub>Ca<sup>2+</sup> cluster ions with n$``$8 is that a$`\times `$b$`\times `$c+1 fragments are specially stable compared to other isomers whenever they can be formed. In Table I we show all the fragments of that kind relevant to the cluster size range considered in this study. Each series has a typical periodicity that could in principle be reflected in different portions of the mass spectra, given the high stability of these fragments. Some sizes can be accomodated in several families, that is, the classification is highly redundant, but useful anyway to our purposes. If for a given cluster size, a cluster with that formula can be formed, it is always the ground state structure. If it is possible to build up two different isomers with that formula (n=12, 18, 24), the more compact structure is energetically favored. This rule works as long as we do not consider structures that are not energetically competitive anymore (the isomer based on the (CaO)<sub>2</sub> building block of (CaO)<sub>14</sub>Ca<sup>2+</sup> is an example), and can be helpful in guessing specially stable structures for clusters larger than those studied here. For nearly all those cluster sizes with no competitive a$`\times `$b$`\times `$c+1 structure, a$`\times `$b$`\times `$c-1 fragments are obtained as the ground state or specially stable isomers (examples are found for n=5, 7, 11, 19, 23 and 29). The special stability of a$`\times `$b$`\times `$c+1 structures is sometimes reflected in high stabilities for the corresponding a$`\times `$b$`\times `$c+3 structures, comparable indeed to the stabilities of a$`\times `$b$`\times `$c-1 fragments; this occurs for n=13, 17, 19 and 26. With the only exceptions of n=13,14,22,26 and 29, all (CaO)<sub>n</sub>Ca<sup>2+</sup> GS structures are explained in terms of those three kinds of fragments.
Comparing to the results of our previous paper on (MgO)<sub>n</sub>Mg<sup>2+</sup> cluster ions, we can see that from n=4 to n=20 the GS structures are basically the same in both systems (the only exceptions are n=7 and n=13). Interesting structural differences between both materials appear in the size range n$`>`$20. Specifically, the transition to bulklike structures, containing inner anions with bulk coordination, is slower in the case of (CaO)<sub>n</sub>Ca<sup>2+</sup>. An analysis of the several energy components shows that the net effect of polarization is more important in calcium oxide. Although the polarizability of Ca<sup>2+</sup> is larger than that of Mg<sup>2+</sup> and the coordination-dependent values of $`\alpha `$(O<sup>2-</sup>) are larger in CaO than in MgO, this is not a trivial conclusion, because the interatomic distances are also larger in CaO, and so the electric fields acting on each ion are correspondingly smaller. If we consider that a highly ionic material is that one for which the Madelung energy term is almost completely dominant in determining structural and several other properties, we would conclude that the ionic character of (CaO)<sub>n</sub>Ca<sup>2+</sup> is smaller that that found for (MgO)<sub>n</sub>Mg<sup>2+</sup>. Indeed, being the polarization contribution more important, the structures of calcium oxide clusters have a larger directionality degree, a feature that is usually associated to covalency (opposite to the natural tendency of purely ionic systems to form isotropic structures). However, we think that the term “covalency” should be employed just in those situations where charge transfer between different atomic centers is important. As Madden and coworkers have discussed, polarization terms in ionic systems are responsible for a lot of properties traditionally attributed to “covalency”. One point that deserves further investigation, however, is whether the directional properties induced by polarization effects (and reflected in a lower average coordination) can be responsible for a larger charge transfer between different centers. This would be reasonable because the saturation of the bonds is less complete.
### B Relative stabilities and connection to experimental mass spectra
In the experimental mass spectra, the populations observed for some cluster sizes are enhanced over those of the neighboring sizes. These “magic numbers” are a consequence of the evaporation events that occur in the cluster beam, mostly after ionization. A magic cluster of size n has an evaporation energy that is large compared to that of the neighboring sizes (n-1) and (n+1). Thus, on the average, clusters of size n undergo a smaller number of evaporation events and this leads to the maxima in the mass spectra. As our main concern in this section is to compare with the experimental mass spectra, we calculate the evaporation energy as a function of cluster size. To do this, we assume that the dominant evaporation channel is the loss of a neutral (CaO) molecule, something supported by the experiments of Ziemann and Castleman. In the size range n$`<`$11, some other chanels seem to be opened in the experiments, and indeed for n$`<`$4 Coulomb explosion is dominant, that is the reason why we do not consider clusters with n$`<`$4. With that assumption, the evaporation energy of (CaO)<sub>n</sub>Ca<sup>2+</sup> reads
$`E_{evap}(n)=E_{cluster}[(CaO)_{n1}Ca^{2+}]+E(CaO)`$ (10)
$`E_{cluster}[(CaO)_nCa^{2+}].`$ (11)
Maxima in the evaporation energy curve do not always coincide with maxima in the experimental mass spectra. There are two main processes that contribute to enhance the cluster population for size n: a)A small evaporation energy for size (n+1); b)A large evaporation energy for size n. Thus, a most convenient quantity to compare with experiment is the second energy difference
$$\mathrm{\Delta }_2(n)=E_{evap}(n+1)E_{evap}(n).$$
(12)
A negative value of $`\mathrm{\Delta }_2(n)`$ indicates that the n-population increases by evaporations from the (n+1)–clusters more rapidly than it decays by evaporation to the (n-1)–clusters. Specifically, the specially stable cluster sizes will be reflected as minima in the $`\mathrm{\Delta }_2(n)`$ curve.
Now, the evaporation energy E$`{}_{evap}{}^{}(n)`$ of eq. (9) can be calculated in two different ways. In the first one, energy differences are always taken between the ground state structures of sizes n and (n-1). This procedure, which we call (by obvious reasons) adiabatic evaporation, reflects the stability of the clusters in the limit of small energy barriers between isomers or alternatively of large experimental times of flight. The stabilities calculated in this way are shown in the upper part of figure 2. Magic numbers are found for n=5,8,12,15,18,20,24,27. The only a$`\times `$b$`\times `$c-1 structure that shows a special stability is that of n=5. The rest of magic clusters belong to the a$`\times `$b$`\times `$c+1 family of structures. If n is a magic size, and both (n+1) and (n-1) GS structures do not belong to the a$`\times `$b$`\times `$c+1 family, a deep minima is found in the $`\mathrm{\Delta }_2(n)`$ curve (this happens for n=12 and 18). For the rest of magic sizes, the (n+1) GS structure has also the formula a$`\times `$b$`\times `$c+1, and has a correspondingly high stability reflected in a negative value of $`\mathrm{\Delta }_2(n+1)`$. In these cases the stability of size n is just slightly enhanced over that of size (n+1). One can appreciate the increasing relevance of the Madelung term in determining the cluster stabilities: when n$`<`$20, the most stable a$`\times `$b$`\times `$c+1 structures are the less compact ones (n=8 and 15 more stable than n=9 and 16, respectively); if n$``$20, that trend is reversed (n=20, 24 and 27 more stable than n= 21, 25, and 28, respectively). The special relevance of a$`\times `$b$`\times `$c+1 structures in explaining the cluster stabilities does not conform to the initial experimental expectations of high stabilities for a$`\times `$b$`\times `$c-1 structures. Analysing the energy components, we find that the polarization contribution stabilizes the a$`\times `$b$`\times `$c+1 structure more than the corresponding a$`\times `$b$`\times `$c-1 structure for all values of a,b,c. For the smallest cluster sizes, however, the extra cation present in a$`\times `$b$`\times `$c+1 structures induces a large cluster distortion compared to that induced by the missing anion in a$`\times `$b$`\times `$c-1 structures, and the Madelung contribution favors these last structures in a larger amount, making them more stable for some sizes.
The second kind of calculation of E<sub>evap</sub>(n) proceeds as follows: we consider the optimized GS structure of (CaO)<sub>n</sub>Ca<sup>2+</sup> and identify the CaO molecule that contributes the least to the cluster binding energy. Then we remove that molecule and relax the resulting (CaO)<sub>n-1</sub>Ca<sup>2+</sup> fragment to the nearest local minimum. This process can be termed locally adiabatic because both fragments are allowed to relax to the local minimum energy configuration after the evaporation. For some cluster sizes, the fragment of size (n-1) left when a CaO molecule is removed from (CaO)<sub>n</sub>Ca<sup>2+</sup> does not lie on the catchment basin of the (CaO)<sub>n-1</sub>Ca<sup>2+</sup> GS isomer, so that the locally adiabatic evaporation energies are larger than the energy differences between adjacent ground states minus E(CaO) in those cases. The locally adiabatic evaporation energies are plotted as a function of n in the lower part of Fig. 2. These will reflect the cluster stabilities in the limit of large energy barriers between isomers or of short experimental times of flight. Magic numbers are obtained for n=5,7,9,11,13,16,19,22,25 and 27, in complete agreement with the experiments of Ziemann and Castleman.
The main message to be extracted from these considerations is that the magic numbers obtained in the experiments might be dominated by the effects of kinetic traps occuring in the course of the evaporation process. Since our calculations are static, we can not rigorously assert that this is the only possible explanation, but a plausibility argument based on a comparison to the closely related and more thoroughly studied case of alkali halides supports our expectations. The mobility experiments performed by the group of Jarrold show that the relaxation dynamics to the ground state structure for sodium chloride clusters involves drift times of almost one second. The importance of kinetic traps in explaining these interesting results is shown in the theoretical works of Doye and Wales. Specifically, these authors show that the potential energy landscape of alkali halide clusters, calculated by using a phenomenological pair potential to describe the interactions, is structured in several funnels, separated from each other by high free-energy barriers. When a cluster evaporates a molecule, it cools in the process, so trapping kinetic effects are expected whenever parent and product GS structures belong to different funnels. Given the close similarities between halide and oxide systems, one expects similar effects in the evaporation kinetics of alkaline-earth oxides to be relevant. The main structural differences are due to the effects of polarization, and these could also affect the mechanisms of structural transitions. In the case of alkali halides, Doye and Wales find that a highly cooperative process is energetically less impeded by energy barriers than sequential ionic diffusion, with interesting implications for the mechanical properties of these clusters. Perhaps the same is true for the clusters studied here, but one has to keep in mind that polarization tends to lower the barriers against diffusion, and those effects are more important for oxides. We think that further calculations of this kind for oxide clusters would be very interesting. Mobility experiments on (MgO)<sub>n</sub>Mg<sup>2+</sup> or (CaO)<sub>n</sub>Ca<sup>2+</sup> could conclusively confirm the structural trends found in the present work.
### C Neutral Stoichiometric (MgO)<sub>n</sub> and (CaO)<sub>n</sub> clusters
The experiments performed by Saunders show that both (MgO)$`{}_{}{}^{+}{}_{n}{}^{}`$ and (CaO)$`{}_{}{}^{+}{}_{n}{}^{}`$ stoichiometric cluster ions with a number of molecules n=6,9,12 and 15 are expecially abundant in the mass spectra. However, when these clusters are allowed to collide with inert gas ions, the fragmentation channels are different: (MgO)<sub>3</sub> fragments are predominantly observed in one case and (CaO)<sub>2</sub> fragments in the other. These results suggest that the basic cluster building blocks are different for the two materials, but not so different as to lead to different magic numbers.
We found a similar scenario in the case of alkali halide clusters. Specifically, a universal set of magic numbers n=4,6,9,12,… was found for the whole family of (AX)<sub>n</sub> clusters, with A=Li,Na,K,Rb and X=F,Cl,Br,I. However, the cluster structures were not found to be the same for all the different materials. When the cation size is much smaller than the anion size (all lithium halides and sodium iodide), ground state structures based on the stacking of hexagonal (AX)<sub>3</sub> rings are obtained. For the rest of materials, the ground state structures are mostly obtained by stacking of rectangular (or double-chain) (AX)<sub>3</sub> planar fragments. This is just a packing effect: when the ratio of cation to anion size is very small, anion-anion overlap repulsive interactions are large, forcing an opening of the (AX)<sub>3</sub> rectangular fragments into hexagons. The magic numbers are the same for both structural families because it is for those cluster sizes that specially compact structures can be formed. When we studied (AX)<sub>n</sub>A<sup>+</sup> alkali halide cluster ions, we found that the structures were much more similar irrespective of packing considerations. The ring structures are not competitive in this case because it is not possible to build up a perfect hexagonal fragment with an odd number of ions. On the contrary, perfect cubic structures can be formed (as for example the 3$`\times `$3$`\times `$3 structure for n=13).
From our study on doubly-charged clusters, we have not found important structural differences between (CaO)<sub>n</sub>Ca<sup>2+</sup> and (MgO)<sub>n</sub>Mg<sup>2+</sup>, at least in the small size regime. We have performed aiPI+polarization calculations on the structures of (MgO)<sub>n</sub> and (CaO)<sub>n</sub> with n=3,6,9,12,15 and 18. Specifically, we have considered just those structures based on staking of hexagonal and rectangular (AO)<sub>3</sub> units, and those based on stacking of (AO)<sub>2</sub> units, with A=Mg or Ca. We find that the ground state structures of (MgO)<sub>n</sub> clusters are based on (MgO)<sub>3</sub> units, while those of (CaO)<sub>n</sub> clusters have a rectangular (CaO)<sub>3</sub> building block, being this the same packing effect found in the case of alkali halides. The structure of (CaO)<sub>6</sub> could be alternatively viewed as the stacking of three (CaO)<sub>2</sub> units, but for n=9,12,15 and 18, the tubular shapes obtained by stacking (CaO)<sub>2</sub> units are not competitive anymore. Were all the ground state structures of (CaO)<sub>n</sub> clusters based on the (CaO)<sub>2</sub> building block, we would expect a periodicity of 2 in the magic numbers observed in the mass spectra. Saunders shows the collision induced fragmentation spectra of (CaO)<sub>n</sub>, with n=4,6,8, which are certainly based on stacking of (CaO)<sub>2</sub> units, but does not show those for (CaO)<sub>9</sub> or (CaO)<sub>12</sub>, for example. Our main conclusion is that the special stability of (CaO)<sub>n</sub> clusters is also explained in terms of (CaO)<sub>3</sub> units, but with rectangular instead of hexagonal shape. This explains the same periodicities observed in the magic numbers of both materials.
## IV Summary
The ab initio perturbed ion model, supplemented with a parameterised treatment of dipolar terms, has been employed in order to study the structural and energetic properties of (CaO)<sub>n</sub>Ca<sup>2+</sup> (n=1–29) cluster ions. Polarization effects favor the formation of surface sites, and reduce the stability of highly compact structures containing anions with bulk coordination. Thus, despite many similarities in the experimental mass spectra, the structures of alkaline-earth oxide and alkali halide cluster ions are shown to be different. Most of the lowest energy structures have the formula a$`\times `$b$`\times `$c+1. The structures of (CaO)<sub>n</sub>Ca<sup>2+</sup> and (MgO)<sub>n</sub>Mg<sup>2+</sup> cluster ions are very similar for n$`<`$20, irrespective of differences in cationic size and polarization. It is just for n$``$20 that structural differences emerge, showing a slower convergence to bulk properties for CaO compared to MgO. The analysis of the stabilities suggests that the experimental mass spectra could be dominated by the effects of kinetic traps. Specifically, if we consider locally adiabatic evaporation events, complete agreement is found with the experimental stabilities. The neutral stoichiometric (MgO)<sub>n</sub> and (CaO)<sub>n</sub> clusters (n=3,6,9,12,15,18) show structural differences similar to those observed in neutral stoichiometric alkali halide clusters: the basic building block is an (MgO)<sub>3</sub> hexagonal fragment in the case of MgO and a (CaO)<sub>3</sub> rectangular (or double-chain) fragment in the case of CaO. This is just a packing effect due to the larger overlap repulsion between anions when the cation size is very small. While the structures of (CaO)<sub>n</sub> clusters, with n=4,6,8 are certainly based on (CaO)<sub>2</sub> units, as suggested by collision-induced fragmentation experiments, the specially stable (CaO)<sub>n</sub> clusters are based on a (CaO)<sub>3</sub> unit. This explains the same periodicity of 3 observed in the experimental magic numbers of both (MgO)$`{}_{}{}^{+}{}_{3}{}^{}`$ and (CaO)$`{}_{}{}^{+}{}_{3}{}^{}`$ clusters.
Captions of Figures and Tables.
Figure 1. Lowest-energy structure and low-lying isomers of (CaO)<sub>n</sub>Ca<sup>2+</sup> cluster ions. Dark balls are Ca<sup>2+</sup> cations and light balls are O<sup>2-</sup> anions. The energy difference (in eV) with respect to the most stable structure is given below the corresponding isomers.
Figure 2. Adiabatic (a) and locally adiabatic (b) evaporation energies required to remove a neutral CaO molecule from (CaO)<sub>n</sub>Ca<sup>2+</sup> cluster ions as a function of n. The local minima in the evaporation energy curve are shown explicitely.
Table I Possible different a$`\times `$b$`\times `$c+1 structures, with their inherent periodicities. Those cluster sizes n that are actually observed as ground state structures of (CaO)<sub>n</sub>Ca<sup>2+</sup> clusters are written in boldface.
\[
|
warning/0004/hep-th0004144.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
There has been renewed interest in compactification through the non-factorizable Kaluza-Klein (KK) metric Ansatz with the warp factor, after Randall and Sundrum (RS) showed that such unconventional compactification of the extra spatial dimensions provides with a simple solution to the hierarchy problem of particle physics. A novel and surprising feature of the RS scenario is that even if the extra spatial dimension is infinite in size, the Newton’s $`1/r^2`$ law of four-dimensional gravity is recovered with negligible correction from the massive KK modes of graviton. The previous works on gravitational aspects of the RS scenario have attempted to reproduce physics of four-dimensional gravity up to small corrections beyond the current experimental precision in an effort to provide with an evidence that the RS scenario may be a true description of our nature.
Especially, in Refs. the geodesic equation for a massless test particle in the bulk of a gravitating configuration in brane world is put into the form of the geodesic motion of a massive test particle in the corresponding gravitating configuration in one lower dimensions by parameterizing the geodesic path with an affine parameter associate with the canonical metric in one lower dimensions. Such result appears to indicate that the bulk geodesic motion is observed by a four-dimensional observer to reproduce physics of our four-dimensional spacetime. However, as we will see through the careful analysis of the geodesic motion of a test particle in the bulk spacetime of brane worlds, in general laws of physics governing the motion of a particle in four-dimensional spacetime is observed to be violated. Namely, we find that the equation describing the trajectory of a particle as observed in one lower dimensions of the brane world scenario has an extra force term which is parallel to the velocity of the particle. According to the so-far known four-dimensional physics, only the component of the extra non-gravitational four-force $`F^\mu `$ which is perpendicular to the particle’s four-velocity $`\frac{dx^\mu }{d\tau }`$ can influence the particle’s motion. Due to such unusual property which cannot be explained by the physics of four-dimensional spacetime, such extra force was dubbed in the previous literature as the fifth force. \[Such extra abnormal force is observed also in Refs. by analyzing the geodesic motion of a test particle in the five-dimensional Kaluza-Klein theory.\] The so-called fifth force generically exists in the KK theories (with constant moduli scalar fields of the extra space) when the spacetime metric depends on the extra spatial coordinates and the velocity of the particle has nonzero components in both the extra spatial direction and the direction of our three dimensional space. Therefore, it is inevitable that if our four-dimensional world is described by the RS scenario then a four-dimensional observer following a test particle along its bulk geodesic path should observe the violation of four-dimensional law of particle mechanics, since the RS scenario allows dependence of the spacetime metric on the extra spatial coordinate and physical process in the RS scenario is generally higher-dimensional in nature.
The paper is organized as follows. In section 2, we survey relevant aspects of domain wall solutions in and geodesic motion of a test particle in the bulk of the RS scenario. In this section, we also discuss well-known facts of particle mechanics in curved spacetime for the purpose of understanding the physical implication of the extra force observed in one lower dimensions. Although some aspects of the fifth force were already studied in the previous literature, we feel that its relation to the four-dimensional particle mechanics has not been clearly presented. We hope that the present paper will clarify some of confusing issues in the fifth force. We study the bulk geodesic motion of a test particle moving in general gravitating configurations in brane worlds as observed in one lower dimensions for the case corresponding to the KK zero mode bulk graviton in section 3 and for the case of general bulk graviton including the massive KK modes in section 4. In these sections, we also identify the extra force on the particle which is observed in one lower dimensions as non-gravitational. Conclusions are given in section 5.
## 2 Preliminaries and General Setup
In this section, we prepare for the main topic of this paper by surveying aspects of domain wall solutions and dynamics of a particle in brane worlds and in general relativity.
Generally, the spacetime metric <sup>2</sup><sup>2</sup>2In this paper, we use the mostly positive convention $`(+\mathrm{}+)`$ for the metric signature. for a $`D`$-dimensional domain wall can be put into the following form:
$$G_{MN}dx^Mdx^N=𝒲(y)\eta _{\mu \nu }dx^\mu dx^\nu +dy^2,$$
(1)
where $`M,N=0,1,\mathrm{},D1`$, $`\mu ,\nu =0,1,\mathrm{},D2`$ and $`𝒲(y)`$ is the warp factor. In particular, for the domain wall solution to the field equations of the following bulk action:
$$S=\frac{1}{2\kappa _D^2}d^Dx\sqrt{G}\left[_G\frac{4}{D2}_M\varphi ^M\varphi +e^{2a\varphi }\mathrm{\Lambda }\right],$$
(2)
the warp factor is given by
$$𝒲(y)=\left(1\frac{(D2)a^2}{2}\sqrt{\frac{D2}{4(D1)a^2(D2)^2}\mathrm{\Lambda }}|y|\right)^{\frac{8}{(D2)^2a^2}},$$
(3)
for $`a0`$, and
$$𝒲(y)=\mathrm{exp}\left(2\sqrt{\frac{\mathrm{\Lambda }}{(D1)(D2)}}|y|\right),$$
(4)
for $`a=0`$. Here, we have imposed the invariance under the $`𝐙_2`$ transformation $`yy`$ and chosen the warp to decrease so that the bulk graviton can be localized.
It is the purpose of this paper to study the dynamics of a test particle in the bulk spacetime with the following metric:
$$G_{MN}dx^Mdx^N=𝒲g_{\mu \nu }dx^\mu dx^\nu +dy^2.$$
(5)
This metric generically describes any gravitational configuration in brane worlds. The geodesic motion of a test particle, i.e., the motion of a particle which is acted on by the gravitational force only, is described by the following geodesic equations:
$$\frac{d^2x^R}{d\lambda ^2}+\widehat{\mathrm{\Gamma }}_{MN}^R\frac{dx^M}{d\lambda }\frac{dx^N}{d\lambda }=0,$$
(6)
where $`\widehat{\mathrm{\Gamma }}_{MN}^R`$ is the Christoffel symbol (of the second kind) for the metric $`G_{MN}`$ and $`\lambda `$ is an affine parameter for the geodesic path $`x^M(\lambda )`$. In addition, the metric compatibility along the geodesic path requires that
$$ϵ_D=G_{MN}\frac{dx^M}{d\lambda }\frac{dx^N}{d\lambda }=𝒲g_{\mu \nu }\frac{dx^\mu }{d\lambda }\frac{dx^\nu }{d\lambda }+\left(\frac{dy}{d\lambda }\right)^2,$$
(7)
where $`ϵ_D=1,0`$ respectively for a massive test particle (i.e., a timelike geodesic) and a massless test particle (i.e., a null geodesic). For a timelike geodesic, $`ϵ_D`$ actually can take any positive values, but one can always apply the affine transformation $`\lambda a\lambda +b`$ ($`a,b𝐑`$), which leaves the geodesic equations (6) invariant, to bring $`ϵ_D=1`$. In this paper, we shall not consider the spacelike geodesics, i.e. the $`ϵ_D=1`$ case.
In this paper, we shall re-express the geodesic equations (6) for the bulk geodesic motion of a test particle in terms of quantities of the hypersurface spacetime of one lower dimensions, for the purpose of learning how the bulk geodesic motion of the test particle is observed in one lower dimensions. In the previous related works , the bulk geodesic motion is reparameterized by an affine parameter $`\stackrel{~}{\lambda }`$ associated with the four-dimensional metric given by $`g_{\mu \nu }`$, namely the one satisfying the following:
$$ϵ_{D1}=g_{\mu \nu }\frac{dx^\mu }{d\stackrel{~}{\lambda }}\frac{dx^\nu }{d\stackrel{~}{\lambda }},$$
(8)
where $`ϵ_{D1}=1,0`$ respectively for a timelike and a lightlike motion observed in one lower dimensions. On the other hand, as the bulk test particle follows its geodesic path, it generally passes through one hypersurface to another. We note that the induced metric on the hypersurface at a specific value of $`y`$ is given by $`\stackrel{~}{g}_{\mu \nu }𝒲(y)g_{\mu \nu }`$. Therefore, it appears that the natural choice for the affine parameter $`\stackrel{~}{\lambda }`$ (or the standard clock on the hypersurface that follows the particle in its motion) for the motion observed on the (comoving) hypersurface is the one satisfying the following
$$ϵ_{D1}=𝒲(y)g_{\mu \nu }\frac{dx^\mu }{d\stackrel{~}{\lambda }}\frac{dx^\nu }{d\stackrel{~}{\lambda }}.$$
(9)
However, one might argue that $`\stackrel{~}{\lambda }`$ satisfying Eq. (8) actually corresponds to the affine parameter for the motion as observed on the hypersurface $`y=y_0=\mathrm{constant}`$, say the TeV brane of our world, at fixed distance from the Planck brane at $`y=0`$. Namely, the affine parameter at $`y=y_0`$ is defined by
$$ϵ_{D1}=𝒲(y_0)g_{\mu \nu }(x^\rho ,y_0)\frac{dx^\mu }{d\stackrel{~}{\lambda }}\frac{dx^\nu }{d\stackrel{~}{\lambda }},$$
(10)
and by applying the affine transformation $`\stackrel{~}{\lambda }𝒲(y_0)^{1/2}\stackrel{~}{\lambda }`$, one can bring this to the form (8) in the case when $`g_{\mu \nu }`$ is independent of the extra spatial coordinate $`y`$. But we, as beings adapted to sense only the four- or lower-dimensional phenomena, are incapable of looking into the extra spatial direction to observe objects in different four-dimensional hypersurface. Any object on different four-dimensional hypersurface cannot be observed; it simply exists in different four-dimensional universe which cannot be observed. So, the geodesic equations conveniently put into a four-dimensional form in term of a new affine parameter in Refs. actually do not describe the trajectory of a test particle observed in one lower dimensions, as one might naively assume although Refs. do not explicitly state that the bulk geodesic motion can be observed by an observer on the TeV brane. However, in the following we shall also consider the case associated with a choice of the affine parameter defined by Eq. (8), as well as the one defined by Eq. (9), just for the purpose of studying the general bulk geodesic motion also within the framework of the previous works.
In obtaining the equations for the particle motion observed in one lower dimensions, we assume that the affine parameter $`\stackrel{~}{\lambda }`$ for the spacetime in one lower dimensions is a smooth function of the affine parameter $`\lambda `$ for the bulk geodesic motion: $`\stackrel{~}{\lambda }=f(\lambda )`$. First, for $`\stackrel{~}{\lambda }`$ defined in Eq. (8), which we refer to as ‘case 1’, the relation (7) for the bulk geodesic motion is rewritten in terms of the new parameter $`\stackrel{~}{\lambda }`$ as
$$g_{\mu \nu }\frac{dx^\mu }{d\stackrel{~}{\lambda }}\frac{dx^\nu }{d\stackrel{~}{\lambda }}=𝒲^1\left[ϵ_D\left(\frac{d\lambda }{d\stackrel{~}{\lambda }}\right)^2+\left(\frac{dy}{d\stackrel{~}{\lambda }}\right)^2\right].$$
(11)
Second, for $`\stackrel{~}{\lambda }`$ defined in Eq. (9), which we refer to as ‘case 2’, Eq. (7) is rewritten in terms of $`\stackrel{~}{\lambda }`$ as
$$𝒲g_{\mu \nu }\frac{dx^\mu }{d\stackrel{~}{\lambda }}\frac{dx^\nu }{d\stackrel{~}{\lambda }}=\left[ϵ_D\left(\frac{d\lambda }{d\stackrel{~}{\lambda }}\right)^2+\left(\frac{dy}{d\stackrel{~}{\lambda }}\right)^2\right].$$
(12)
The parameter $`\stackrel{~}{\lambda }`$ is an affine parameter for the motion observed in one lower dimensions, if Eqs. (11) and (12) respectively take the forms (8) and (9), i.e., the RHS’s are either 0 or $`1`$.
We now discuss the condition for $`\stackrel{~}{\lambda }`$ being an affine parameter. First, for a massless test particle ($`ϵ_D=0`$), the RHS’s of Eqs. (11) and (12) can be 0 or $`1`$, depending on the motion of a test particle along the $`y`$-direction. When the test particle is confined to move along the longitudinal directions of the domain wall <sup>3</sup><sup>3</sup>3From Eq. (29) with $`ϵ_D=0`$, one can see that such bulk geodesic motion is possible for a massless test particle. (i.e., $`\frac{dy}{d\lambda }=0`$ and therefore $`\frac{dy}{d\stackrel{~}{\lambda }}=\frac{d\lambda }{d\stackrel{~}{\lambda }}\frac{dy}{d\lambda }=0`$), the RHS’s of Eqs. (11) and (12) are zero, corresponding to the lightlike motion as observed in one lower dimensions. In this case, one can choose $`\stackrel{~}{\lambda }=\lambda `$ as an affine parameter for the motion observed in one lower dimensions, as can be seen from Eqs. (11) and (12). When the $`y`$-component of the velocity of the massless test particle is nonzero (i.e., $`\frac{dy}{d\lambda }0`$), its motion is observed in one lower dimensions as timelike if the following is satisfied:
$$\left(\frac{dy}{d\stackrel{~}{\lambda }}\right)^2=𝒲,$$
(13)
for the case 1, and
$$\left(\frac{dy}{d\stackrel{~}{\lambda }}\right)^2=1,$$
(14)
for the case 2, for which cases the RHS’s of Eqs. (11) and (12) are $`1`$. Second, for a massive test particle ($`ϵ_D=1`$), the parameter $`\stackrel{~}{\lambda }`$ is an affine parameter for the timelike motion as observed in one lower dimensions, if $`\stackrel{~}{\lambda }`$ is related to the bulk affine parameter $`\lambda `$ as
$$\left(\frac{d\lambda }{d\stackrel{~}{\lambda }}\right)^2=𝒲\left(\frac{dy}{d\stackrel{~}{\lambda }}\right)^2,$$
(15)
for the case 1, and
$$\left(\frac{d\lambda }{d\stackrel{~}{\lambda }}\right)^2=1\left(\frac{dy}{d\stackrel{~}{\lambda }}\right)^2,$$
(16)
for the case 2.
In the following sections, we will see that from the perspective of an observer in one lower dimensions a bulk test particle appears to be under the influence of an abnormal non-gravitational force. So, it would be useful to discuss some aspects of dynamics of particles in general relativity to better understand physical implication of such extra force. Although we assume the spacetime to be four-dimensional in the discussion in the following paragraphs, our discussion holds for arbitrary spacetime dimensions without modification of equations.
In terms of the relativistic four-vector notation, the following classical Newton’s second law of mechanics plus the work-energy relation (the law of conservation of energy):
$$𝐅=\frac{d𝐩}{dt},\frac{dT}{dt}=𝐅𝐯,$$
(17)
where $`𝐅`$ is the force on the particle, $`𝐩=m_0𝐯=m_0\frac{d𝐱}{dt}`$ is the momentum of the particle with the inertial mass $`m_0`$ and $`T`$ is the kinetic energy of the particle, is written as
$$\frac{dp_\mu }{d\tau }=F_\mu ,$$
(18)
where $`p^\mu =m_0\frac{dx^\mu }{d\tau }`$ is the contravariant components of the four-momentum of a particle with the rest mass $`m_0`$, $`\tau `$ is the proper time defined from $`d\tau ^2=\eta _{\mu \nu }dx^\mu dx^\nu `$, $`F_\mu =(𝐅\frac{d𝐱}{d\tau },𝐅\frac{dt}{d\tau })`$, and $`(x^\mu )=(t,𝐱)`$. In curved spacetime with metric $`g_{\mu \nu }`$, Eq. (18) is modified to:
$$\frac{Dp^\mu }{d\tau }\frac{dp^\mu }{d\tau }+\mathrm{\Gamma }_{\rho \sigma }^\mu \frac{dx^\rho }{d\tau }p^\sigma =F^\mu ,$$
(19)
in the contravariant notation, or
$$\frac{Dp_\mu }{d\tau }\frac{dp_\mu }{d\tau }\frac{1}{2}\frac{g_{\rho \sigma }}{x^\mu }\frac{dx^\rho }{d\tau }p^\sigma =F_\mu ,$$
(20)
in the covariant notation, where $`\mathrm{\Gamma }_{\rho \sigma }^\mu `$ is the Christoffel symbol of the second kind for the metric $`g_{\mu \nu }`$. Here, $`F_\mu `$ and $`p^\mu `$ are defined in the same way as in the flat spacetime case except that the proper time $`\tau `$ is now defined from $`d\tau ^2=g_{\mu \nu }dx^\mu dx^\nu `$.
First, we show that a purely mechanical non-gravitational four-force $`F^\mu `$ acts on a particle perpendicularly to its four-velocity $`\frac{dx^\mu }{d\tau }`$. By taking the covariant derivative of $`g_{\mu \nu }\frac{dx^\mu }{d\tau }\frac{dx^\nu }{d\tau }=1`$ along the particle trajectory $`x^\mu (\tau )`$, one can see that $`g_{\mu \nu }\frac{D}{d\tau }\left(\frac{dx^\mu }{d\tau }\right)\frac{dx^\nu }{d\tau }=0`$. This implies that $`g_{\mu \nu }F^\mu \frac{dx^\nu }{d\tau }=g_{\mu \nu }\frac{Dp^\mu }{d\tau }\frac{dx^\nu }{d\tau }=m_0g_{\mu \nu }\frac{D}{d\tau }\left(\frac{dx^\mu }{d\tau }\right)\frac{dx^\nu }{d\tau }=0`$.
Next, we consider the possibility that the four-force $`F^\mu `$ has nonzero component parallel to the four-velocity $`\frac{dx^\mu }{d\tau }`$. In deriving the relation $`g_{\mu \nu }F^\mu \frac{dx^\nu }{d\tau }=0`$ in the previous paragraph, we assumed that the proper mass $`m_0`$ of the particle is constant in time. Had we considered the possibility that $`m_0`$ changes with time, we would instead have obtained
$$g_{\mu \nu }F^\mu \frac{dx^\nu }{d\tau }=g_{\mu \nu }\frac{Dp^\mu }{d\tau }\frac{dx^\nu }{d\tau }=m_0g_{\mu \nu }\frac{D}{d\tau }\left(\frac{dx^\mu }{d\tau }\right)\frac{dx^\nu }{d\tau }+\frac{dm_0}{d\tau }g_{\mu \nu }\frac{dx^\mu }{d\tau }\frac{dx^\nu }{d\tau }=\frac{dm_0}{d\tau }.$$
(21)
Therefore, the existence of parallel component of the force $`F^\mu `$ implies non-conservation of the proper mass $`m_0`$ of a particle. Generally, the change in the proper mass of a particle occurs when there exist some non-mechanical external forces which cause such change. To take into account of the additional non-mechanical forces, one has to modify $`F_0`$ in the following way:
$$F_0=𝐅\frac{d𝐱}{d\tau }Q\frac{dt}{d\tau },$$
(22)
where the first term on the RHS is the mechanical work done by the mechanical force $`𝐅`$ per unit time and the second term is the heat or non-mechanical energy developed per unit time. The extra term is added to take into account of the contribution from the non-mechanical energy so that the energy can be conserved in the system under consideration. Meanwhile, the remaining components of $`F_\mu `$ take the same form as the above, i.e., (the curved space analog of) the Newton’s second law of mechanics $`𝐅=\frac{d𝐩}{dt}`$ continues to hold. So, from Eqs. (21) and (22), we obtain
$$\frac{dm_0}{d\tau }=Q\left(\frac{dt}{d\tau }\right)^2,$$
(23)
namely, the proper mass is converted into the non-mechanical energy and vice versa.
Finally, we discuss the motion of a particle under the influence of an extra non-gravitational force $`F^\mu `$ in curved spacetime. First, when $`F^\mu `$ acts on the particle perpendicularly to its four-velocity $`\frac{dx^\mu }{d\tau }`$, i.e., the proper mass $`m_0`$ of the particle is constant in time, from Eqs. (19) and (20) we obtain the following equations for the particle trajectory $`x^\mu (\tau )`$:
$`{\displaystyle \frac{d^2x^\mu }{d\tau ^2}}+\mathrm{\Gamma }_{\rho \sigma }^\mu {\displaystyle \frac{dx^\rho }{d\tau }}{\displaystyle \frac{dx^\sigma }{d\tau }}={\displaystyle \frac{F^\mu }{m_0}},`$ (24)
$`{\displaystyle \frac{d^2x_\mu }{d\tau ^2}}{\displaystyle \frac{1}{2}}{\displaystyle \frac{g_{\rho \sigma }}{x^\mu }}{\displaystyle \frac{dx^\rho }{d\tau }}{\displaystyle \frac{dx^\sigma }{d\tau }}={\displaystyle \frac{F_\mu }{m_0}}.`$ (25)
As expected, when the particle is free, i.e., when only gravitational force is acted on the particle ($`F^\mu =0`$), the particle will execute geodesic motion. Second, when the force $`F^\mu `$ has nonzero parallel component, i.e., when $`m_0`$ is not conserved, the equation for the particle trajectory $`x^\mu (\tau )`$ takes the following form:
$$\frac{d^2x^\mu }{d\tau ^2}+\mathrm{\Gamma }_{\rho \sigma }^\mu \frac{dx^\rho }{d\tau }\frac{dx^\sigma }{d\tau }=\frac{F^\mu }{m_0}+g_{\rho \sigma }\frac{F^\rho }{m_0}\frac{dx^\sigma }{d\tau }\frac{dx^\mu }{d\tau },$$
(26)
which can be obtained from Eqs. (19) and (21). We note that the RHS of Eq. (26) is perpendicular to the four-velocity $`\frac{dx^\mu }{d\tau }`$, implying that only the perpendicular component of $`F^\mu `$ influences the motion of the particle. In other words, if $`F^\mu `$ is parallel to $`\frac{dx^\mu }{d\tau }`$, then the particle will just follow the geodesic path (since the RHS of Eq. (26) vanishes for this case) while its mass changing with time according to the relation (23). This result also shows that according to the particle mechanics of four-dimensional general relativity the extra non-gravitational force term in the equation for the particle trajectory cannot have non-zero component parallel to the four-velocity of the particle.
## 3 Dynamics in the Kaluza-Klein Zero Mode Spacetime
In this section, we consider the case when $`g_{\mu \nu }`$ in the bulk metric (5) does not depend on the extra spatial coordinate $`y`$. In this case, the bulk test particle is regarded as being under the influence of the Kaluza-Klein zero mode of graviton, only.
The geodesic equations (6), with the bulk metric given by Eq. (5) with $`g_{\mu \nu }=g_{\mu \nu }(x^\rho )`$, take the following forms:
$$\frac{d^2x^\rho }{d\lambda ^2}+\mathrm{\Gamma }_{\mu \nu }^\rho \frac{dx^\mu }{d\lambda }\frac{dx^\nu }{d\lambda }+\frac{𝒲^{}}{𝒲}\frac{dx^\rho }{d\lambda }\frac{dy}{d\lambda }=0,$$
(27)
$$\frac{d^2y}{d\lambda ^2}\frac{1}{2}𝒲^{}g_{\mu \nu }\frac{dx^\mu }{d\lambda }\frac{dx^\nu }{d\lambda }=0,$$
(28)
where $`\mathrm{\Gamma }_{\mu \nu }^\rho `$ is the Christoffel symbol for the metric $`g_{\mu \nu }`$ and $`𝒲^{}d𝒲/dy`$. By using the relation (7), one can put the $`y`$-component geodesic equation (28) into the following form:
$$\frac{d^2y}{d\lambda ^2}+\frac{1}{2}\frac{𝒲^{}}{𝒲}\left[ϵ_D+\left(\frac{dy}{d\lambda }\right)^2\right]=0.$$
(29)
### 3.1 Case 1
In this subsection, we study the motion of the bulk test particle as observed in one lower dimensions with the spacetime metric $`g_{\mu \nu }=g_{\mu \nu }(x^\rho )`$.
First, we consider the geodesic motion of a massless particle in the bulk spacetime, i.e. the $`ϵ_D=0`$ case. When $`\frac{dy}{d\lambda }=0`$, by choosing $`\stackrel{~}{\lambda }=\lambda `$ as an affine parameter, one can put the $`x^\rho `$-component bulk geodesic equation (27) into the following form:
$$\frac{d^2x^\rho }{d\stackrel{~}{\lambda }^2}+\mathrm{\Gamma }_{\mu \nu }^\rho \frac{dx^\mu }{d\stackrel{~}{\lambda }}\frac{dx^\nu }{d\stackrel{~}{\lambda }}=0.$$
(30)
So, the test particle’s motion is observed in one lower dimensions with the metric $`g_{\mu \nu }`$ as the lightlike geodesic motion. Next, when the $`y`$-component of the velocity of the test particle is nonzero (i.e., $`\frac{dy}{d\lambda }0`$), from Eqs. (13) and (29) with $`ϵ_D=0`$, one can see that the parameters $`\lambda `$ and $`\stackrel{~}{\lambda }`$ are related as
$$\left(\frac{d\stackrel{~}{\lambda }}{d\lambda }\right)^1\frac{d}{d\stackrel{~}{\lambda }}\left(\frac{d\stackrel{~}{\lambda }}{d\lambda }\right)=𝒲^{\frac{1}{2}}𝒲^{},$$
(31)
which can be solved by
$$\frac{d\stackrel{~}{\lambda }}{d\lambda }=𝒲^1.$$
(32)
By using Eqs. (13) and (31), one can express the $`x^\rho `$-component bulk geodesic equation (27) in terms of the $`(D1)`$-dimensional affine parameter $`\stackrel{~}{\lambda }`$. The resulting equation also takes the form (30). To summarize, a massless test particle moving in the bulk spacetime with the metric (5) is observed in one lower dimensions with the metric $`g_{\mu \nu }(x^\rho )`$ to be ($`i`$) a free massless particle if the motion of the test particle is confined along the longitudinal directions of the domain wall and ($`ii`$) a free massive particle if the $`y`$-component of its velocity is nonzero. This is just a generalization of the result in Ref. to the case of an arbitrary warp factor $`𝒲`$ and an arbitrary gravitating configuration with the metric $`g_{\mu \nu }(x^\rho )`$ within the brane world. (See also Ref. for the generalization to the case of multi-codimensional brane world.)
Second, we consider the geodesic motion of a massive bulk test particle, i.e., the $`ϵ_D=1`$ case. By using the relation (15), one can re-express the $`y`$-component geodesic equation (29) (with $`ϵ_D=1`$) in terms of the new parameter $`\stackrel{~}{\lambda }`$ as follows:
$$\frac{d^2y}{d\stackrel{~}{\lambda }^2}\frac{𝒲^{}}{𝒲}\left(\frac{dy}{d\stackrel{~}{\lambda }}\right)^2+\frac{1}{2}𝒲^{}=0.$$
(33)
Making use of Eqs. (15) and (33), one can again put the $`x^\rho `$-component bulk geodesic equation (27) into the form (30). So, we see that a massive test particle moving in the bulk spacetime with the metric (5) is observed in one lower dimensions with the metric $`g_{\mu \nu }(x^\rho )`$ to be a free massive particle. This result extends the previous studies on the geodesic motion in brane worlds to include the case of a massive test particle. \[Note, even if some aspects of the geodesic motion of a massive test particle in brane worlds have been previously studied, it has never been shown that the geodesic motion of a massive test particle in the bulk spacetime is observed in one lower dimensions as the motion of a free massive particle.\]
Just by looking at Eq. (30), it appears from the lower-dimensional perspective that the particle is under the influence of the gravitational field $`g_{\mu \nu }(x^\rho )`$, only. However, this is not the case, as we explain in the following. A test particle of mass $`m_0`$ in $`D`$-dimensional bulk spacetime with the metric (5) appears in $`(D1)`$-dimensional embedded spacetime with the metric $`g_{\mu \nu }`$ to have mass given by :
$$\stackrel{~}{m}_0=m_0\frac{d\stackrel{~}{\lambda }}{d\lambda }=m_0\left[𝒲\left(\frac{dy}{d\stackrel{~}{\lambda }}\right)^2\right]^{\frac{1}{2}},$$
(34)
where we used Eq. (15). So, as the test particle executes its geodesic motion with the nonzero $`y`$-component velocity $`\frac{dy}{d\stackrel{~}{\lambda }}`$, its mass appears to change with the following rate from the $`(D1)`$-dimensional perspective:
$$\frac{d\stackrel{~}{m}_0}{d\stackrel{~}{\lambda }}=\stackrel{~}{m}_0\frac{𝒲^{}}{𝒲}\frac{dy}{d\stackrel{~}{\lambda }}.$$
(35)
From Eqs. (30) and (35), we obtain the following equation describing the conservation of energy and the Newton’s second law of mechanics in $`(D1)`$-dimensions:
$$\frac{Dp^\mu }{d\stackrel{~}{\lambda }}\frac{dp^\mu }{d\stackrel{~}{\lambda }}+\mathrm{\Gamma }_{\rho \sigma }^\mu \frac{dx^\rho }{d\stackrel{~}{\lambda }}p^\sigma =\stackrel{~}{m}_0\frac{𝒲^{}}{𝒲}\frac{dy}{d\stackrel{~}{\lambda }}\frac{dx^\mu }{d\stackrel{~}{\lambda }},$$
(36)
where $`p^\mu =\stackrel{~}{m}_0\frac{dx^\mu }{d\stackrel{~}{\lambda }}`$ is the $`(D1)`$-momentum of the particle. This implies that from the $`(D1)`$-dimensional perspective the particle is under the influence of the extra non-gravitational velocity dependent force $`F^\mu =\stackrel{~}{m}_0\frac{𝒲^{}}{𝒲}\frac{dy}{d\stackrel{~}{\lambda }}\frac{dx^\mu }{d\stackrel{~}{\lambda }}`$. This extra force does not influence the motion of the particle, because it acts parallelly to the particle’s $`(D1)`$-velocity $`\frac{dx^\mu }{d\stackrel{~}{\lambda }}`$ (Cf. Eq. (26)), but is responsible for the change of the inertial mass $`\stackrel{~}{m}_0`$ of the particle (from the $`(D1)`$-dimensional perspective). The intuitive reason is that the effect of the force $`𝐅`$ on the particle velocity $`𝐯=\frac{d𝐱}{dt}`$ is canceled by the effect of the inertial mass change with time on $`𝐯`$. Although the extra force is gravitational in nature from the $`D`$-dimensional perspective, a $`(D1)`$-dimensional observer will sense that some non-mechanical non-gravitational force causes the inertial mass $`\stackrel{~}{m}_0`$ of the particle to be converted into the heat energy and vice versa. Such heat energy generated per unit time is
$$Q\frac{dt}{d\stackrel{~}{\lambda }}=\stackrel{~}{m}_0\frac{𝒲^{}}{𝒲}\frac{dy}{dt},$$
(37)
according to Eq. (23).
### 3.2 Case 2
In this subsection, we study the motion of a bulk test particle as observed on the (comoving) hypersurface with the spacetime metric $`\stackrel{~}{g}_{\mu \nu }=𝒲(y)g_{\mu \nu }(x^\rho )`$.
First, we consider the bulk geodesic motion of a massless particle ($`ϵ_D=0`$). When the motion of the test particle is confined along the longitudinal directions of the domain wall (i.e., $`\frac{dy}{d\lambda }=0`$), one can choose $`\stackrel{~}{\lambda }=\lambda `$ as an affine parameter on the hypersurface $`y=\mathrm{constant}`$ to express the $`x^\rho `$-component bulk geodesic equations (27) in the following form:
$$\frac{d^2x^\rho }{d\stackrel{~}{\lambda }^2}+\stackrel{~}{\mathrm{\Gamma }}_{\mu \nu }^\rho \frac{dx^\mu }{d\stackrel{~}{\lambda }}\frac{dx^\nu }{d\stackrel{~}{\lambda }}=0,$$
(38)
where $`\stackrel{~}{\mathrm{\Gamma }}_{\mu \nu }^\rho `$ is the Christoffel symbol for the metric $`\stackrel{~}{g}_{\mu \nu }=𝒲g_{\mu \nu }`$ and we used the fact that $`\stackrel{~}{\mathrm{\Gamma }}_{\mu \nu }^\rho =\mathrm{\Gamma }_{\mu \nu }^\rho `$. So, its motion observed on the hypersurface $`y=\mathrm{constant}`$ is that of a massless free particle. Nontrivial result arises when the massless test particle has nonzero $`y`$-component for its velocity. From Eqs. (14) and (29) with $`ϵ_D=0`$, one can see that the parameters $`\lambda `$ and $`\stackrel{~}{\lambda }`$ are related as
$$\left(\frac{d\stackrel{~}{\lambda }}{d\lambda }\right)^1\frac{d}{d\stackrel{~}{\lambda }}\left(\frac{d\stackrel{~}{\lambda }}{d\lambda }\right)=\frac{1}{2}\frac{𝒲^{}}{𝒲},$$
(39)
which can be solved by
$$\frac{d\stackrel{~}{\lambda }}{d\lambda }=𝒲^{\frac{1}{2}}.$$
(40)
So, the bulk geodesic equation (27) for the $`x^\rho `$-component motion is rewritten in terms of the new parameter $`\stackrel{~}{\lambda }`$ as
$$\frac{d^2x^\rho }{d\stackrel{~}{\lambda }^2}+\stackrel{~}{\mathrm{\Gamma }}_{\mu \nu }^\rho \frac{dx^\mu }{d\stackrel{~}{\lambda }}\frac{dx^\nu }{d\stackrel{~}{\lambda }}=\frac{1}{2}\frac{𝒲^{}}{𝒲}\frac{dx^\rho }{d\stackrel{~}{\lambda }}.$$
(41)
The bulk geodesic motion of a free massless particle with nonzero $`y`$-component for its velocity is therefore observed on the hypersurface $`y=y(\stackrel{~}{\lambda })`$ with the metric $`\stackrel{~}{g}_{\mu \nu }`$ as the motion of a massive particle which is under the influence of the extra non-gravitational force as well as the gravitational field $`\stackrel{~}{g}_{\mu \nu }`$.
Second, we consider the massive bulk test particle ($`ϵ_D=1`$). By using Eq. (16), one can re-express the $`y`$-component geodesic equation (29) (with $`ϵ_D=1`$) in terms of the new parameter $`\stackrel{~}{\lambda }`$ as
$$\frac{d^2y}{d\stackrel{~}{\lambda }^2}+\frac{1}{2}\frac{𝒲^{}}{𝒲}\left[1\left(\frac{dy}{d\stackrel{~}{\lambda }}\right)^2\right]=0.$$
(42)
The $`x^\rho `$-component geodesic equations (27) can be put into the following simplified form in terms of the new parameter $`\stackrel{~}{\lambda }`$ by applying Eqs. (16) and (42):
$$\frac{d^2x^\rho }{d\stackrel{~}{\lambda }^2}+\stackrel{~}{\mathrm{\Gamma }}_{\mu \nu }^\rho \frac{dx^\mu }{d\stackrel{~}{\lambda }}\frac{dx^\nu }{d\stackrel{~}{\lambda }}=\frac{1}{2}\frac{𝒲^{}}{𝒲}\frac{dy}{d\stackrel{~}{\lambda }}\frac{dx^\rho }{d\stackrel{~}{\lambda }}.$$
(43)
So, the geodesic motion of a free massive particle in the bulk spacetime is observed on the hypersurface $`y=y(\stackrel{~}{\lambda })`$ with the metric $`\stackrel{~}{g}_{\mu \nu }`$ as the motion of a massive particle under the additional influence of an extra non-gravitational force. For both the massive and massless test particle cases, the extra force term exists in the equations (41) and (43) for the particle motion, if the $`x^\rho `$-component of its velocity is nonzero, and acts on the particle parallelly to its four velocity $`\frac{dx^\rho }{d\stackrel{~}{\lambda }}`$.
We have seen in the previous section that the Newton’s second law of mechanics and the conservation of energy (and their curved space generalization) in four-dimensional spacetime imply that only orthogonal (to the four-velocity $`\frac{dx^\mu }{d\tau }`$ of the particle) component of non-gravitational force $`F^\mu `$ influences the motion of the particle. However, as can be seen from the equations (41) and (43) for the trajectory $`x^\rho (\stackrel{~}{\lambda })`$ of the particle (as observed on the comoving hypersurface $`y=y(\stackrel{~}{\lambda })`$), the particle’s motion is additionally influenced by the extra force term (the RHS’s of Eqs. (41) and (43)) which is parallel to its velocity $`\frac{dx^\rho }{d\stackrel{~}{\lambda }}`$. Existence of such abnormal force term implies violation of four-dimensional physics and therefore can be an implication of existence of extra spatial dimensions (since such force term cannot be explained by the known four-dimensional physics). Or maybe it is due to the wrong choice of frame, since we have seen in the previous subsection that in the metric frame of the case 1 such abnormal force term does not exist. The author does not have yet clear understanding of which metric frame is the correct choice for the spacetime in one lower dimensions. However, as we will see in the following section, when $`g_{\mu \nu }`$ depends on the extra spatial coordinate $`y`$, the $`x^\rho `$-component equation for the particle trajectory (expressed in terms of $`\stackrel{~}{\lambda }`$) has the extra abnormal force term for both case 1 and case 2. So, it seems inevitable that the abnormal force term in general exists for natural choices of metric frame, i.e., those associated with $`g_{\mu \nu }`$ and $`\stackrel{~}{g}_{\mu \nu }=𝒲g_{\mu \nu }`$. Due to extraordinary property of such abnormal force term, the previous literature dubbed the extra force as the fifth force. Actually, it should not be regarded as the violation of the four-dimensional physics, since such contradiction arises because we attempt to interpret the phenomenon which is higher-dimensional in nature from the perspective of lower dimensional physics. We, as beings incapable of sensing higher-dimensional spacetime, is apt to regard the higher-dimensional physical process as violation of four-dimensional physics. Anyhow, in the following we reconstruct the equation for the energy conservation and the Newton’s second law of mechanics in the spacetime of one lower dimensions from the equations for the particle trajectory to see any physical implication of the extra force in one lower dimensions.
On the hypersurface $`y=y(\stackrel{~}{\lambda })`$ with the metric $`\stackrel{~}{g}_{\mu \nu }=𝒲g_{\mu \nu }`$, a bulk test particle with mass $`m_0`$ appears to have mass given by
$$\stackrel{~}{m}_0=m_0\left[1\left(\frac{dy}{d\stackrel{~}{\lambda }}\right)^2\right]^{\frac{1}{2}},$$
(44)
where we used Eq. (16). By applying Eq. (42), we obtain the following mass change rate with $`\stackrel{~}{\lambda }`$ as observed on the (comoving) hypersurface:
$$\frac{d\stackrel{~}{m}_0}{d\stackrel{~}{\lambda }}=\frac{\stackrel{~}{m}_0}{2}\frac{𝒲^{}}{𝒲}\frac{dy}{d\stackrel{~}{\lambda }}.$$
(45)
From Eqs. (43) and (45), we obtain the following equation describing the conservation of energy and the Newton’s second law on the hypersurface:
$$\frac{Dp^\mu }{d\stackrel{~}{\lambda }}\frac{dp^\mu }{d\stackrel{~}{\lambda }}+\stackrel{~}{\mathrm{\Gamma }}_{\rho \sigma }^\mu \frac{dx^\rho }{d\stackrel{~}{\lambda }}p^\sigma =\stackrel{~}{m}_0\frac{𝒲^{}}{𝒲}\frac{dy}{d\stackrel{~}{\lambda }}\frac{dx^\mu }{d\stackrel{~}{\lambda }},$$
(46)
which has the same form as the equation (36) for the case 1. So, in both cases the particle appears to be under the influence of the extra force of the same form $`F^\mu =\stackrel{~}{m}_0\frac{𝒲^{}}{𝒲}\frac{dy}{d\stackrel{~}{\lambda }}\frac{dx^\mu }{d\stackrel{~}{\lambda }}`$. However, the motions of the particle are different for the two cases, because the mass $`\stackrel{~}{m}_0`$ changes with different rates due to different choices of the $`(D1)`$-dimensional metric. For the case 2, the particle mechanics of one lower dimensions appears to be violated, whereas for the case 1 it is not.
## 4 Dynamics in General Spacetime
In this section, we consider the case when $`g_{\mu \nu }`$ in the bulk metric (5) depends on the extra spatial coordinate $`y`$. In this case, a bulk test particle is regarded as being under the influence of both the zero and the massive KK modes of graviton. We will see that the massive KK modes of graviton induce the perpendicular component of the extra force $`F^\mu `$ for both cases 1 and 2, and the parallel component of the force term even for the case 1.
The bulk geodesic equations (6), with the bulk metric given by Eq. (5) with $`g_{\mu \nu }=g_{\mu \nu }(x^\rho ,y)`$, take the following forms:
$$\frac{d^2x^\rho }{d\lambda ^2}+\mathrm{\Gamma }_{\mu \nu }^\rho \frac{dx^\mu }{d\lambda }\frac{dx^\nu }{d\lambda }+𝒲^1g^{\rho \sigma }_y(𝒲g_{\sigma \mu })\frac{dx^\mu }{d\lambda }\frac{dy}{d\lambda }=0,$$
(47)
$$\frac{d^2y}{d\lambda ^2}\frac{1}{2}_y(𝒲g_{\mu \nu })\frac{dx^\mu }{d\lambda }\frac{dx^\nu }{d\lambda }=0.$$
(48)
The consistency condition for the geodesic motion in the bulk expressed in terms of an affine parameter $`\lambda `$ \[a new parameter $`\stackrel{~}{\lambda }=f(\lambda )`$\] takes the same forms (7) \[(11) and (12)\] except that the metric $`g_{\mu \nu }`$ now depends on $`y`$.
### 4.1 Case 1
In this subsection, we study the geodesic motions of a bulk test particle as observed in one lower dimensions with the spacetime metric $`g_{\mu \nu }(x^\rho ,y)`$.
First, we consider a massless test particle ($`ϵ_D=0`$) in the bulk spacetime. For the trivial case of the geodesic motion with $`\frac{dy}{d\lambda }=0`$, the motion is observed in one lower dimensions to be that of massless free particle under the influence of gravitational field $`g_{\mu \nu }`$, only. When $`\frac{dy}{d\lambda }0`$, by using Eqs. (13) and (48), one obtains the following relation between the two parameters $`\lambda `$ and $`\stackrel{~}{\lambda }`$:
$$\left(\frac{d\stackrel{~}{\lambda }}{d\lambda }\right)^1\frac{d}{d\stackrel{~}{\lambda }}\left(\frac{d\stackrel{~}{\lambda }}{d\lambda }\right)=\frac{1}{2}𝒲^{\frac{1}{2}}𝒲^{}+\frac{1}{2}𝒲^{\frac{1}{2}}_y(𝒲g_{\mu \nu })\frac{dx^\mu }{d\stackrel{~}{\lambda }}\frac{dx^\nu }{d\stackrel{~}{\lambda }}.$$
(49)
The $`x^\rho `$-component geodesic equation (47) takes the following form after Eqs. (13) and (49) are applied:
$$\frac{d^2x^\rho }{d\stackrel{~}{\lambda }^2}+\mathrm{\Gamma }_{\mu \nu }^\rho \frac{dx^\mu }{d\stackrel{~}{\lambda }}\frac{dx^\nu }{d\stackrel{~}{\lambda }}=\frac{1}{2}𝒲^{\frac{1}{2}}𝒲^{}\frac{dx^\rho }{d\stackrel{~}{\lambda }}\left[𝒲^{\frac{1}{2}}g^{\rho \sigma }+\frac{1}{2}𝒲^{\frac{1}{2}}\frac{dx^\rho }{d\stackrel{~}{\lambda }}\frac{dx^\sigma }{d\stackrel{~}{\lambda }}\right]\frac{dx^\mu }{d\stackrel{~}{\lambda }}_y(𝒲g_{\rho \mu }).$$
(50)
The bulk geodesic motion of the massless particle with $`\frac{dy}{d\lambda }0`$ is therefore observed in one lower dimensions as the motion of a massive particle under the additional influence of the extra non-gravitational force, unlike the case of the KK zero mode metric $`g_{\mu \nu }`$ as discussed in the previous section. The extra force term on the RHS of Eq. (50) has both the parallel and the perpendicular components given by
$`f_{}^\rho `$ $`=`$ $`{\displaystyle \frac{1}{2}}\left[𝒲^{\frac{1}{2}}𝒲^{}+𝒲^{\frac{1}{2}}_y(𝒲g_{\mu \nu }){\displaystyle \frac{dx^\mu }{d\stackrel{~}{\lambda }}}{\displaystyle \frac{dx^\mu }{d\stackrel{~}{\lambda }}}\right]{\displaystyle \frac{dy}{d\stackrel{~}{\lambda }}}{\displaystyle \frac{dx^\rho }{d\stackrel{~}{\lambda }}},`$ (51)
$`f_{}^\rho `$ $`=`$ $`\left[𝒲^{\frac{1}{2}}g^{\rho \sigma }+𝒲^{\frac{1}{2}}{\displaystyle \frac{dx^\rho }{d\stackrel{~}{\lambda }}}{\displaystyle \frac{dx^\sigma }{d\stackrel{~}{\lambda }}}\right]{\displaystyle \frac{dx^\mu }{d\stackrel{~}{\lambda }}}_y(𝒲g_{\sigma \mu }).`$ (52)
These vanish when $`g_{\mu \nu }`$ is independent of the extra spatial coordinate $`y`$, implying that the extra force is due to the massive KK modes of graviton.
Second, for a massive bulk test particle ($`ϵ_D=1`$), by using (15) one can express the $`y`$-component bulk geodesic equation (48) in terms of $`\stackrel{~}{\lambda }`$ as
$$\frac{d^2y}{d\stackrel{~}{\lambda }^2}\frac{1}{2}\frac{𝒲^{}}{𝒲}\left(\frac{dy}{d\stackrel{~}{\lambda }}\right)^2\frac{1}{2}\left[𝒲\left(\frac{dy}{d\stackrel{~}{\lambda }}\right)^2\right]𝒲^1_y(𝒲g_{\mu \nu })\frac{dx^\mu }{d\stackrel{~}{\lambda }}\frac{dx^\nu }{d\stackrel{~}{\lambda }}=0.$$
(53)
The $`x^\rho `$-component bulk geodesic equation (47) can be expressed in terms of $`\stackrel{~}{\lambda }`$ as follows by applying Eqs. (15) and (53):
$$\frac{d^2x^\rho }{d\stackrel{~}{\lambda }^2}+\mathrm{\Gamma }_{\mu \nu }^\rho \frac{dx^\mu }{d\stackrel{~}{\lambda }}\frac{dx^\nu }{d\stackrel{~}{\lambda }}=\frac{1}{2}\frac{𝒲^{}}{𝒲}\frac{dy}{d\stackrel{~}{\lambda }}\frac{dx^\rho }{d\stackrel{~}{\lambda }}𝒲^1\left[g^{\rho \sigma }+\frac{1}{2}\frac{dx^\rho }{d\stackrel{~}{\lambda }}\frac{dx^\sigma }{d\stackrel{~}{\lambda }}\right]\frac{dy}{d\stackrel{~}{\lambda }}\frac{dx^\mu }{d\stackrel{~}{\lambda }}_y\left(𝒲g_{\sigma \mu }\right).$$
(54)
So, the bulk geodesic motion of a massive test particle is observed in one lower dimensions as the motion of a massive particle under the additional influence of the extra non-gravitational force, also unlike the case of $`y`$-independent $`g_{\mu \nu }`$. The extra force term on the RHS of Eq. (54) has both the parallel and the perpendicular components given by
$`f_{}^\rho `$ $`=`$ $`{\displaystyle \frac{1}{2}}\left[{\displaystyle \frac{𝒲^{}}{𝒲}}+𝒲^1_y(𝒲g_{\mu \nu }){\displaystyle \frac{dx^\mu }{d\stackrel{~}{\lambda }}}{\displaystyle \frac{dx^\mu }{d\stackrel{~}{\lambda }}}\right]{\displaystyle \frac{dy}{d\stackrel{~}{\lambda }}}{\displaystyle \frac{dx^\rho }{d\stackrel{~}{\lambda }}},`$ (55)
$`f_{}^\rho `$ $`=`$ $`𝒲^1\left[g^{\rho \sigma }+{\displaystyle \frac{dx^\rho }{d\stackrel{~}{\lambda }}}{\displaystyle \frac{dx^\sigma }{d\stackrel{~}{\lambda }}}\right]{\displaystyle \frac{dy}{d\stackrel{~}{\lambda }}}{\displaystyle \frac{dx^\mu }{d\stackrel{~}{\lambda }}}_y(𝒲g_{\sigma \mu }).`$ (56)
As expected, these vanish when $`g_{\mu \nu }`$ is independent of $`y`$.
We now obtain the expression for the extra $`(D1)`$-force $`F^\mu `$ acting on the particle from the $`(D1)`$-dimensional perspective. The bulk test particle of mass $`m_0`$ is observed to have mass $`\stackrel{~}{m}_0`$ given by Eq. (34) from the perspective of $`(D1)`$-dimensional spacetime with the metric $`g_{\mu \nu }`$. By using Eq. (53), one obtains the following rate of mass change with $`\stackrel{~}{\lambda }`$ as observed in the embedded $`(D1)`$-dimensional spacetime:
$$\frac{d\stackrel{~}{m}_0}{d\stackrel{~}{\lambda }}=\frac{\stackrel{~}{m}_0}{2}\left[𝒲^1𝒲^{}𝒲^1_y(𝒲g_{\mu \nu })\frac{dx^\mu }{d\stackrel{~}{\lambda }}\frac{dx^\nu }{d\stackrel{~}{\lambda }}\right]\frac{dy}{d\stackrel{~}{\lambda }}.$$
(57)
From Eqs. (54) and (57), we obtain the following equation describing the conservation of energy and the Newton’s second law of mechanics in $`(D1)`$-dimensions:
$$\frac{Dp^\mu }{d\stackrel{~}{\lambda }}\frac{dp^\mu }{d\stackrel{~}{\lambda }}+\mathrm{\Gamma }_{\rho \sigma }^\mu \frac{dx^\rho }{d\stackrel{~}{\lambda }}p^\sigma =\stackrel{~}{m}_0𝒲^1g^{\mu \nu }\frac{dx^\rho }{d\stackrel{~}{\lambda }}_y(𝒲g_{\nu \rho })\frac{dy}{d\stackrel{~}{\lambda }}.$$
(58)
The extra force $`F^\mu `$ on the RHS of Eq. (58) has both the parallel and orthogonal components given by
$`F_{}^\mu `$ $`=`$ $`\stackrel{~}{m}_0𝒲^1{\displaystyle \frac{dx^\rho }{d\stackrel{~}{\lambda }}}{\displaystyle \frac{dx^\sigma }{d\stackrel{~}{\lambda }}}_y(𝒲g_{\rho \sigma }){\displaystyle \frac{dy}{d\stackrel{~}{\lambda }}}{\displaystyle \frac{dx^\mu }{d\stackrel{~}{\lambda }}},`$ (59)
$`F_{}^\mu `$ $`=`$ $`\stackrel{~}{m}_0𝒲^1\left[g^{\mu \nu }+{\displaystyle \frac{dx^\mu }{d\stackrel{~}{\lambda }}}{\displaystyle \frac{dx^\nu }{d\stackrel{~}{\lambda }}}\right]{\displaystyle \frac{dy}{d\stackrel{~}{\lambda }}}{\displaystyle \frac{dx^\rho }{d\stackrel{~}{\lambda }}}_y(𝒲g_{\nu \rho }).`$ (60)
As expected, when $`g_{\mu \nu }`$ does not depend on $`y`$, $`F_{}^\mu `$ vanishes and $`F_{}^\mu `$ takes the form of the RHS of Eq. (36).
Even with the choice of the metric by $`g_{\mu \nu }`$ for the $`(D1)`$-dimensional spacetime, which it has been previously regarded as the natural canonical choice for the $`(D1)`$-dimensional spacetime embedded in brane worlds, the motion of the particle observed in the $`(D1)`$-dimensional spacetime appears to be under the additional influence of the abnormal force term, which cannot be explained by laws of physics in $`(D1)`$-dimensions, if the metric $`g_{\mu \nu }`$ depends on the extra spatial coordinate $`y`$. Namely, the equations (50) and (54) describing the particle trajectory $`x^\mu (\stackrel{~}{\lambda })`$ observed in one lower dimensions have nonzero parallel component force term $`f_{}^\rho `$ and the the mass change (57) with $`\stackrel{~}{\lambda }`$ is not in accordance with the conventional formula (21) for a given extra non-gravitational force $`F^\mu =\stackrel{~}{m}_0𝒲^1g^{\mu \nu }\frac{dx^\rho }{d\stackrel{~}{\lambda }}_y(𝒲g_{\nu \rho })\frac{dy}{d\stackrel{~}{\lambda }}`$. Also, with a choice of the metric frame $`\stackrel{~}{g}_{\mu \nu }`$, the same holds true as expected, as we will see in the following subsection. Note, the abnormal force term is not due to the wrong choice <sup>4</sup><sup>4</sup>4If one chooses a non-affine parameter to describe the motion of a particle, the abnormal force term also occurs in the equation for particle trajectory. To see this, we consider the following geodesic equation for a free particle, whose motion is under the influence of the gravitational force, only:
$$\frac{d^2x^\rho }{ds^2}+\mathrm{\Gamma }_{\mu \nu }^\rho \frac{dx^\mu }{ds}\frac{dx^\nu }{ds}=0,$$
(61) where $`s`$ is an affine parameter. If we take a new parameter $`\stackrel{~}{s}=f(s)`$ to parameterize the motion of the particle, then the above geodesic equations transform to the following form:
$$\frac{d^2x^\rho }{d\stackrel{~}{s}^2}+\mathrm{\Gamma }_{\mu \nu }^\rho \frac{dx^\mu }{d\stackrel{~}{s}}\frac{dx^\nu }{d\stackrel{~}{s}}=\frac{d^2\stackrel{~}{s}/ds^2}{(d\stackrel{~}{s}/ds)^2}\frac{dx^\rho }{d\stackrel{~}{s}}.$$
(62) So, through non-affine transformation one induces an extra velocity dependent fictitious force term $`\frac{d^2\stackrel{~}{s}/ds^2}{(d\stackrel{~}{s}/ds)^2}\frac{dx^\rho }{d\stackrel{~}{s}}`$ parallel to the four-velocity $`\frac{dx^\rho }{d\stackrel{~}{s}}`$ of the particle. Eq. (62) also shows that the geodesic equations (61) are invariant only under the affine transformations $`s\stackrel{~}{s}=as+b`$, $`a,b𝐑`$. of parameter $`\stackrel{~}{\lambda }`$ describing motion observed in $`(D1)`$-dimensions, since we have fixed (up to affine transformations) the parameter $`\stackrel{~}{\lambda }`$ through the $`(D1)`$-dimensional affine conditions (8) and (9). So, the massive KK modes of graviton not only give a small correction to Newton’s $`1/r^2`$ law of four-dimensional gravity but also causes violation of four-dimensional laws of physics, which can be an indication that our four-dimensional world is embedded in higher-dimensional spacetime.
If we take the viewpoint that laws of physics in $`(D1)`$-dimensional spacetime should not be violated, then we have to take the metric $`\overline{g}_{\mu \nu }`$ for which the equation for the particle trajectory does not have abnormal force term as the physical metric of the $`(D1)`$-dimensional spacetime. With a choice of such metric, according to Eq. (21), the $`(D1)`$-dimensional mass $`\overline{m}_0`$ should change with an affine parameter $`\overline{\lambda }`$ as
$$\frac{d\overline{m}_0}{d\overline{\lambda }}=\overline{g}_{\mu \nu }F^\mu \frac{dx^\nu }{d\overline{\lambda }},$$
(63)
where $`\overline{m}_0=m_0\frac{d\overline{\lambda }}{d\lambda }`$, the parameter $`\overline{\lambda }`$ is defined through the relation $`\overline{g}_{\mu \nu }\frac{dx^\mu }{d\overline{\lambda }}\frac{dx^\nu }{d\overline{\lambda }}=1`$, $`F^\mu `$ is the extra non-gravitational force observed in $`(D1)`$-dimensions, and of course the parameter $`\lambda `$ is defined through Eq. (7) with $`ϵ_D=1`$. In the case where $`g_{\mu \nu }`$ is independent of $`y`$, such physical metric is given by $`\overline{g}_{\mu \nu }=g_{\mu \nu }`$. For a general $`y`$-dependent $`g_{\mu \nu }`$, it seems not clear whether a simple and natural form of the physical metric $`\overline{g}_{\mu \nu }`$ that satisfies this equation exists.
### 4.2 Case 2
In this subsection, we study the geodesic motion of a bulk test particle as observed in the (comoving) hypersurface $`y=y(\stackrel{~}{\lambda })`$ with the metric $`\stackrel{~}{g}_{\mu \nu }=𝒲(y)g_{\mu \nu }(x^\rho ,y)`$.
First, we consider the case of a free massless bulk particle ($`ϵ_D=0`$). When $`\frac{dy}{d\lambda }=0`$, one can put the $`x^\rho `$-component geodesic equation (47) to the form (38). When $`\frac{dy}{d\lambda }0`$, by using Eqs. (14) and (48), one can see that the affine parameters $`\lambda `$ and $`\stackrel{~}{\lambda }`$ are related as
$$\left(\frac{d\stackrel{~}{\lambda }}{d\lambda }\right)^1\frac{d}{d\stackrel{~}{\lambda }}\left(\frac{d\stackrel{~}{\lambda }}{d\lambda }\right)=\frac{1}{2}_y\left(𝒲g_{\mu \nu }\right)\frac{dx^\mu }{d\stackrel{~}{\lambda }}\frac{dx^\nu }{d\stackrel{~}{\lambda }}.$$
(64)
Applying Eqs. (14) and (64), one can put the $`x^\rho `$-component bulk geodesic equation (47) into the following form in terms of the new parameter $`\stackrel{~}{\lambda }`$:
$$\frac{d^2x^\rho }{d\stackrel{~}{\lambda }^2}+\stackrel{~}{\mathrm{\Gamma }}_{\mu \nu }^\rho \frac{dx^\mu }{d\stackrel{~}{\lambda }}\frac{dx^\nu }{d\stackrel{~}{\lambda }}=\left[𝒲^1g^{\rho \sigma }+\frac{1}{2}\frac{dx^\rho }{d\stackrel{~}{\lambda }}\frac{dx^\sigma }{d\stackrel{~}{\lambda }}\right]\frac{dx^\mu }{d\stackrel{~}{\lambda }}_y\left(𝒲g_{\sigma \mu }\right),$$
(65)
where $`\stackrel{~}{\mathrm{\Gamma }}_{\mu \nu }^\rho `$ is the Christoffel symbol for the metric $`\stackrel{~}{g}_{\mu \nu }=𝒲(y)g_{\mu \nu }(x^\rho ,y)`$ and we used the fact that $`\stackrel{~}{\mathrm{\Gamma }}_{\mu \nu }^\rho =\mathrm{\Gamma }_{\mu \nu }^\rho `$. So, the geodesic motion of a massless free particle in the bulk spacetime with the metric (5) is observed on the (comoving) hypersurface $`y=y(\stackrel{~}{\lambda })`$ as the motion of a massive particle under the additional influence of the extra non-gravitational force. Unlike the case of the $`y`$-independent $`g_{\mu \nu }`$, the extra force term on the RHS of Eq. (65) is no longer parallel to the $`(D1)`$-velocity $`\frac{dx^\rho }{d\stackrel{~}{\lambda }}`$ of the test particle. The parallel and the perpendicular components of the extra force term are given by
$`f_{}^\rho `$ $`=`$ $`{\displaystyle \frac{1}{2}}_y\left(𝒲g_{\mu \nu }\right){\displaystyle \frac{dx^\mu }{d\stackrel{~}{\lambda }}}{\displaystyle \frac{dx^\nu }{d\stackrel{~}{\lambda }}}{\displaystyle \frac{dx^\rho }{d\stackrel{~}{\lambda }}},`$ (66)
$`f_{}^\rho `$ $`=`$ $`\left[{\displaystyle \frac{dx^\rho }{d\stackrel{~}{\lambda }}}{\displaystyle \frac{dx^\sigma }{d\stackrel{~}{\lambda }}}+𝒲^1g^{\rho \sigma }\right]{\displaystyle \frac{dx^\mu }{d\stackrel{~}{\lambda }}}_y\left(𝒲g_{\sigma \mu }\right).`$ (67)
As expected, the perpendicular component $`f_{}^\rho `$ vanishes and the parallel component $`f_{}^\rho `$ takes the form of the RHS of Eq. (41), when $`g_{\mu \nu }`$ is independent of $`y`$, implying that the perpendicular component is due to the massive KK modes of graviton.
Second, we consider the case of a massive bulk test particle ($`ϵ_D=1`$). By using Eq. (16), one can put the $`y`$-component geodesic equations (48) into the following form in terms of a new parameter $`\stackrel{~}{\lambda }`$:
$$\frac{d^2y}{d\stackrel{~}{\lambda }^2}\frac{1}{2}\left[1\left(\frac{dy}{d\stackrel{~}{\lambda }}\right)^2\right]_y\left(𝒲g_{\mu \nu }\right)\frac{dx^\mu }{d\stackrel{~}{\lambda }}\frac{dx^\nu }{d\stackrel{~}{\lambda }}=0.$$
(68)
The $`x^\rho `$-component bulk geodesic equation (47) takes the following form after Eqs. (16) and (68) are applied:
$$\frac{d^2x^\rho }{d\stackrel{~}{\lambda }^2}+\stackrel{~}{\mathrm{\Gamma }}_{\mu \nu }^\rho \frac{dx^\mu }{d\stackrel{~}{\lambda }}\frac{dx^\nu }{d\stackrel{~}{\lambda }}=\left[𝒲^1g^{\rho \sigma }+\frac{1}{2}\frac{dx^\rho }{d\stackrel{~}{\lambda }}\frac{dx^\sigma }{d\stackrel{~}{\lambda }}\right]\frac{dy}{d\stackrel{~}{\lambda }}\frac{dx^\mu }{d\stackrel{~}{\lambda }}_y\left(𝒲g_{\sigma \mu }\right).$$
(69)
So, the geodesic motion of a massive particle in the bulk spacetime with the metric (5) is observed on the (comoving) hypersurface $`y=y(\stackrel{~}{\lambda })`$ as the motion of a massive particle under the additional influence of the extra non-gravitational force. As in the massless test particle case, the extra force term on the RHS of Eq. (69) has both parallel and perpendicular components given by
$`f_{}^\rho `$ $`=`$ $`{\displaystyle \frac{1}{2}}_y\left(𝒲g_{\mu \nu }\right){\displaystyle \frac{dy}{d\stackrel{~}{\lambda }}}{\displaystyle \frac{dx^\mu }{d\stackrel{~}{\lambda }}}{\displaystyle \frac{dx^\nu }{d\stackrel{~}{\lambda }}}{\displaystyle \frac{dx^\rho }{d\stackrel{~}{\lambda }}},`$ (70)
$`f_{}^\rho `$ $`=`$ $`\left[{\displaystyle \frac{dx^\rho }{d\stackrel{~}{\lambda }}}{\displaystyle \frac{dx^\sigma }{d\stackrel{~}{\lambda }}}+𝒲^1g^{\rho \sigma }\right]{\displaystyle \frac{dy}{d\stackrel{~}{\lambda }}}{\displaystyle \frac{dx^\mu }{d\stackrel{~}{\lambda }}}_y\left(𝒲g_{\sigma \mu }\right).`$ (71)
As expected, when $`g_{\mu \nu }`$ is independent of $`y`$, $`f_{}^\rho `$ takes the form of the RHS of Eq. (43) and $`f_{}^\rho `$ vanishes.
We now obtain the expression for the extra non-gravitational $`(D1)`$-force $`F^\mu `$ observed on the (comoving) hypersurface $`y=y(\stackrel{~}{\lambda })`$. A bulk test particle with mass $`m_0`$ is measured on the hypersurface $`y=y(\stackrel{~}{\lambda })`$ to have mass $`\stackrel{~}{m}_0`$ given by Eq. (44). By using Eq. (68), we obtain the following mass change with $`\stackrel{~}{\lambda }`$ as observed on the hypersurface:
$$\frac{d\stackrel{~}{m}_0}{d\stackrel{~}{\lambda }}=\frac{\stackrel{~}{m}_0}{2}_y(𝒲g_{\mu \nu })\frac{dx^\mu }{d\stackrel{~}{\lambda }}\frac{dx^\nu }{d\stackrel{~}{\lambda }}\frac{dy}{d\stackrel{~}{\lambda }}.$$
(72)
So, from Eqs. (69) and (72), we obtain the equation $`\frac{Dp^\mu }{d\stackrel{~}{\lambda }}=F^\mu `$ which has the same form (58) as the case 1. However, since we have chosen different $`(D1)`$-dimensional metric, the expressions for the parallel and the perpendicular components of $`F^\mu `$ are instead given by
$`F_{}^\mu `$ $`=`$ $`\stackrel{~}{m}_0{\displaystyle \frac{dx^\rho }{d\stackrel{~}{\lambda }}}{\displaystyle \frac{dx^\sigma }{d\stackrel{~}{\lambda }}}_y(𝒲g_{\rho \sigma }){\displaystyle \frac{dy}{d\stackrel{~}{\lambda }}}{\displaystyle \frac{dx^\mu }{d\stackrel{~}{\lambda }}},`$ (73)
$`F_{}^\mu `$ $`=`$ $`\stackrel{~}{m}_0\left[𝒲^1g^{\mu \nu }+{\displaystyle \frac{dx^\mu }{d\stackrel{~}{\lambda }}}{\displaystyle \frac{dx^\nu }{d\stackrel{~}{\lambda }}}\right]{\displaystyle \frac{dy}{d\stackrel{~}{\lambda }}}{\displaystyle \frac{dx^\rho }{d\stackrel{~}{\lambda }}}_y(𝒲g_{\nu \rho }).`$ (74)
As expected, when $`g_{\mu \nu }`$ is independent of $`y`$, $`F_{}^\mu `$ vanishes and $`F_{}^\mu `$ takes the form of the RHS of Eq. (46).
## 5 Conclusions
In this paper, we carefully studied the geodesic motions of a test particle in the bulk spacetime of general gravitating configurations in the RS scenario as observed in the embedded spacetime of one lower dimensions. We presented the explicit equations describing such particle motion perceived by an observer in one lower dimensions and the explicit forms of the extra force on the particle measured in one lower dimensions. Such equations and extra forces are inconsistent with laws of particle mechanics in one lower dimensions. Such inconsistency does not mean the violation of physics in one lower dimensions, but results from our effort to interpret the physical process which is higher-dimensional in nature with physics of one lower dimensions. The RS model assumes that the extra spatial dimension is noncompact, and therefore generically physical phenomena in the RS model have to show higher-dimensional character, which is observed to be inconsistent with physics of our four-dimensional world. So, one can test the RS scenario by detecting inconsistency with the four-dimensional physics such as the one present in this paper. However, one has to note that since the current RS scenario assumes that the motion of ordinary matter in our (visible) universe is confined within the TeV brane, which is assumed to be at fixed distance from the Planck brane, due to yet unknown non-gravitational mechanism, the extra force discussed in this paper will not be detected by the lower-dimensional observers on the (visible) TeV brane of the current RS model. The extra force discussed in this paper can be measured only by an observer who follows a test particle with the nonzero velocity component along the extra spatial direction.
|
warning/0004/hep-ph0004080.html
|
ar5iv
|
text
|
# Present Status of the Long Range Component of the Nuclear Force11footnote 1NUP-A-2000-10
## 1 Introduction
In the Yukawa model of the nuclear force, the pi-meson plays the central roll, since the nuclear potentials are assumed to arise from the exchanges of a pion and a set of pions at least outside of the inner core region. Although the two-pion exchange potentials constructed directly from the meson theory are not successful in understanding the details of the data of the nucleon-nucleon scatterings, various phenomenological nuclear potentials have been proposed to reproduce the existing data. In the constructions of the nuclear potentials, two features are commonly shared by all the nuclear potentials, which are inherited from the meson theory of the nuclear force. One is the one-pion exchange (OPE) part of the potential and the other is that the range of the non-OPE part is shorter than $`1/(2\mu )`$. In terms of the singularities of the scattering amplitudes $`A(s,t)`$, such features correspond to the one-pion exchange pole at $`t=\mu ^2`$ and the cut of the two-pion exchange which starts at $`t=4\mu ^2`$. These analytic structures are common to all the amplitudes calculated from the various proposed nuclear potentials.
On the other hand, in the nineteen sixties our views on the nucleons changed from ‘elementary particles’ to composite particles, and moreover in the most important models of hadron, such as the QCD and the dyon model, the constructive forces of the composite particles are the Coulombic types. Therefore we cannot exclude the possibility for the induced long range force to appear between the composite particles. In particular, in the dyon model of hadron, the appearance of the strong van der Waals interaction between hadrons is a natural cosequence, because in such a model the hadrons are regarded as the ‘magnetic atoms’. In general the long range force gives rise to a singularity at $`t=0`$ in the scattering amplitude $`A(s,t)`$, and therefore the analytic structure is completely different from the case of the purely short range interactions.
Since there is no apriori reason to believe that the strong interaction is a synonym for the short range interaction, it is desirable to search for the possible long range components of the nuclear force, whenever new precise data in $`|t|4\mu ^2`$ come to be available. It is the difference of the analytic structure that allows us to do the search in the clear-cut way without being disturbed by the ambiguities such as different parametrization of the nuclear potentials in the inner region. Therefore, our aim in the present article is to search for the possible extra singularity of $`A(s,t)`$ at $`t=0`$, which is characteristic of the long range force, or to find the corresponding extra singularity in the partial wave amplitude. It is important that the search of the extra singularity can be carried out independently of the uncertainties of the spectra related to other singularities. The reason of such possibility comes from the fact that the location of the extra singularity is the end point of the physical region $`4\nu t0`$, where the experimental data are available. Therefore, in contrast to the search for the short range force, our search of the extra singularity at $`t=0`$ does not require any analytic continuation from the physical region. In our approach based on the analytic structure, the dispersion technique will turn out to be useful.
In the search of the long range force, the required accuracy of the input data depends heavily on the power $`\gamma `$ of the threshold behavior of the spectral function $`A_t(s,t)`$ at $`t=0`$, where for small $`\nu `$ and $`|t|`$ the power $`\gamma `$ is introduced by $`A_t(s,t)=\pi C^{}t^\gamma +\mathrm{}`$. Since $`\gamma `$ and another power $`\alpha `$, which appears in the asymptotic behavior of the long range potential as $`V(r)C/r^\alpha `$, are related by $`\alpha =2\gamma +3`$, our observation of the extra singularity in the amplitude becomes more and more difficult and requires the higher precision of the input data, as $`\alpha `$ increases. For example, the van der Waals potentials of the London type ($`\alpha =6`$) and the Casimir-Polder type ($`\alpha =7`$) imply $`\gamma =1.5`$ and $`\gamma =2.0`$ respectively, and therefore to observe these van der Waals forces is to recognize the singularities of the type $`C^{}(t)^{3/2}`$ or $`C^{}t^2\mathrm{log}(t)`$ on the smooth back ground function of $`A(s,t)`$. In order to observe the long range force, it is essential to obtain the high precision data at least in the region of small $`|t|`$, when the power $`\alpha `$ of the asymptotic potential is not small.
In the hadron physics, the accuracies of the phase shift data of the S-wave proton-proton scattering in the low energy region are exceptinally high, and so it is better to analyse the S-wave amplitude $`h_0(\nu )`$ instead of $`A(s,t)`$ in our search for the long range force. After the partial wave projection, the singularity of $`A(s,t)`$ at $`t=t_1`$ becomes the left hand singularity at $`\nu =t_1/4`$ in $`h_0(\nu )`$. Therefore the extra singularity at $`t=0`$ appears at $`\nu =0`$, whereas the OPE pole changes to the logarithmic cut starting at $`\nu =\mu ^2/4`$, and the two-pion exchange spectrum starts at $`\nu =\mu ^2`$ in $`h_0(\nu )`$. In addition to these singularities, $`h_0(\nu )`$ has the unitarity cut in $`\nu 0`$. Moreover since we are considering the proton-proton scattering, $`h_0(\nu )`$ has also the cuts of the Coulombic interaction in $`\mathrm{}<\nu 0`$ and of the vacuum polarization in $`\mathrm{}<\nu m_e^2`$ .
In section 2, we shall construct a function $`K_0(\nu )`$ whose non-OPE part is free from the singularities in $`|\nu |<\mu ^2`$, when the forces are the short range types of the meson theory plus the electromagnetic interaction. If we take advantage that the scattering length is known within $`0.05\%`$, we can even consider the once subtracted function $`K_0^{once}(\nu )=(K_0(\nu )K_0(0))/\nu `$. The merit to use the once subtracted function instead of $`K_0(\nu )`$ is that, since the power of the threshold behavior at $`\nu =0`$ changes from $`\gamma `$ to $`(\gamma 1)`$, the extra singularity becomes much easier to be obsreved, when it exists. In section 3, by using the phase shift data we shall evaluate numerically the once subtracted Kantor amplitude $`K_0^{once}(\nu )`$, and its non-OPE part $`\stackrel{~}{K}_0^{once}(\nu )`$ will be tabulated. The result of the evaluation is that $`\stackrel{~}{K}_0^{once}(\nu )`$ has a sharp cusp at $`\nu =0`$.
In sections 4 and 5, $`\stackrel{~}{K}_0^{once}(\nu )`$ is fitted by using the spectrum of the long range force and that of the short range force respectively. It turns out in section 4 that the chi-square minimum occurs at $`\gamma =1.481.62`$, which is close to the $`\gamma `$ of the van der Waals interaction of the London type. In section 5, it is shown that the conventional spectrum of the short range interaction cannot reproduce the cusp.
In section 6, by assuming that the well-known mechanism to produce the van der Waals force is working, the lower bound of the strength $`{}_{}{}^{}e_{}^{2}`$ of the underlying Coulombic force is estimated by using the inequality of the strength $`C_6`$ of the Van der Waals potential. The lower bound of $`{}_{}{}^{}e_{}^{2}`$ becomes 3.3 or 14 depending on the value of the radius of the composite particle, which comes from the measurement of the nucleon form factor or the distance where the deviation from the asymptotic form of the van der Waals potential becomes appreciable, respectively. In section 6, it is also pointed out that the Coulombic interaction between the magnetic monopoles is an imprtant candidate of the underlying super-strong Coulombic force, and the dyon model of hadron is explained briefly. Section 7 is used for remarks and comments, in which low energy experiments to confirm the long range force, by observing the characteristic destructive interference pattern of the Coulomb and the strong Van der Waals forces, are proposed .
## 2 Selection of a Regular Function
The low energy proton-proton scattering is prominent in its accuracy of the measurements in the hadron physics. Especially the S-wave amplitude $`h_0(\nu )`$ in the low energy region provides the ideal place to answer to the fundamental question whether the hadron-hadron interactions are short range except for the electromagnetic components. In the following investigations, we shall make use of the difference of the analytic structures of the amplitudes $`h_0(\nu )`$. In general, for the short range interaction which arises from the exchange of a state with mass $`\sqrt{t_1}`$, a singularity appears at $`\nu =t_1/4`$ in the partial wave amplitude $`h_0(\nu )`$. On the other hand, the long range interaction gives rise to a singularity at $`\nu =0`$. Since the location of the extra singularity due to the long range force is the end point of the physical region $`\nu 0`$, where the experimental data are available, we can expect to observe the singularity directly, if it exists, without making any analytic continuation from the physical region. The first thing we have to do is to construct a function which is regular at $`\nu =0`$, when all the interactions are short range. Next thing is to eliminate the known near-by singularities from the function. The function with such a wider domain of analyticity serves to expose the extra singularity at $`\nu =0`$, and makes it easier for us to observe the long range interaction, when it exists.
Although the main aim of this section is to construct such an analytic function for the proton-proton scattering where the effects of the vaccum polarization as well as the Coulomb interaction are not negligible, it is instructive to start by constructing such a function for the neutron-neutron scattering first. It is well-known that the S-wave amplitude $`h_0(\nu )`$ has the unitarity cut in $`\nu >0`$ with the spectral function Im$`h_0(\nu )`$, where $`h_0(\nu )`$ relates to the phase shift by
$$h_0(\nu )=\frac{\sqrt{m^2+\nu }}{\sqrt{\nu }}e^{i\delta _0(\nu )}\mathrm{sin}\delta _0(\nu ).$$
(1)
The most famous function, which is analytic at $`\nu =0`$ and therefore accepts the Taylor expansion of $`\nu `$ when all forces are short range, is the effective range function $`X_0(\nu )`$ of Bethe. As it is well-known, $`X_0(\nu )`$ is defined by
$$X_0(\nu )=\sqrt{\nu }\mathrm{cot}\delta _0(\nu ).$$
(2)
From Eqs.(1) and (2), the relation between $`h_0(\nu )`$ and $`X_0(\nu )`$ is
$$h_0(\nu )=\frac{\sqrt{m^2+\nu }}{X_0(\nu )i\sqrt{\nu }}.$$
(3)
Since the one-pion exchange (OPE) contribution is
$$h_0^{1\pi }(\nu )=\frac{1}{4}\frac{g^2}{4\pi }\frac{\mu ^2}{4\nu }\mathrm{log}(1+\frac{4\nu }{\mu ^2}),$$
(4)
the values of the coupling constant $`g^2/4\pi `$ and the neutral pion mass $`\mu `$ are sufficient to eliminate the OPE cut from $`h_0(\nu )`$. In fact $`\stackrel{~}{h}_0(\nu )h_0(\nu )h_0^{1\pi }(\nu )`$ does not have the OPE cut. However Eq.(3) indicates that in order to eliminate the OPE cut from $`X_0(\nu )`$ information on Re$`h_0(\nu )`$ in $`\nu \mu ^2/4`$ as well as $`g^2/4\pi `$ and $`\mu `$ are necessary. Therefore the effective range function $`X_0(\nu )`$ is not adequate for our purpose to construct a function with wider domain of analyticity.
Another possibility is the Kantor amplitude, which is defined by
$$K_0(\nu )=h_0(\nu )\frac{1}{\pi }_0^{\mathrm{}}𝑑\nu ^{}\frac{\mathrm{Im}h_0(\nu ^{})}{\nu ^{}\nu }.$$
(5)
The Kantor amplitude does not have the unitarity cut, and the values of $`K_0(\nu )`$ for real positive $`\nu `$ can be evaluated directly from the experimental data, although the integration must be regarded as the principal value integration of Cauchy. It is straightforward to eliminate the OPE cut from the Kantor amplitude, and which is achieved by introducing the non-OPE Kantor amplitude $`\stackrel{~}{K}_0(\nu )`$ by
$$\stackrel{~}{K}_0(\nu )=K_0(\nu )K_0^{1\pi }(\nu ),$$
(6)
where the OPE part of the Kantor amplitude is common to that of the partial wave amplitude, namely $`K_0^{1\pi }(\nu )=h_0^{1\pi }(\nu )`$. It is useful to rewrite the Kantor amplitude introduced in Eq.(5) into the form of the contour integration,
$$K_0(\nu )=\frac{1}{2\pi i}_C𝑑\nu ^{}\frac{h_0(\nu ^{})}{\nu ^{}\nu },$$
(7)
where the closed contour $`C`$ is shown in Figure 1. The merit to write $`K_0(\nu )`$ in the form of Eq.(7) is that because of Eq.(3), when the effective range function $`X_0(\nu )`$ is a polynomial of $`\nu `$ or more generally a meromorphic function of $`\sqrt{\nu }`$, then the only singularities of $`h_0(\nu )`$ on the first sheet of $`\nu `$ are poles.<sup>2</sup><sup>2</sup>2Strictly speaking $`\sqrt{m^2+\nu }`$ of the numerator of Eq.(3) causes a branch point at $`\nu =m^2`$. Since $`\nu =m^2`$ is the far-away singularity, it does not affect our search for the extra singularity at $`\nu =0`$. However if we want to be more precise, we may redefine the amplitude by multiplying $`m/\sqrt{m^2+\nu }`$ to $`h_0(\nu )`$, and make necessary changes in the following calculations. Therefore by shrinking the contour $`C`$ to a point, $`K_0(\nu )`$ becomes the sum of the contributions of the poles on the first sheet of $`\nu `$ or in the upper half $`\sqrt{\nu }`$-plane. In this way, we need not carry out the principal value integration of the rapidly changing function of Eq.(5). This technique to change the integration into the sum of the contributions of poles will be used in the actual calculation of $`K_0(\nu )`$ in the next section. We are now in the position to examine $`\stackrel{~}{K}_0(\nu )`$ in search for the extra singularity at $`\nu =0`$, if the precise data of the neutron-neutron scatterings were available.
Let us turn to the proton-proton scattering, where the vacuum polarization as well as the Coulombic interactions are important. The Kantor amplitude introduced in Eq.(5) does not satisfy our requirement of the analyticity. This is because the Coulombic cut in $`\mathrm{}<\nu 0`$ and the cut of the vaccum polarization in $`\mathrm{}<\nu m_e^2`$ still remain in the Kantor amplitude $`K_0(\nu )`$ of Eq.(5). The difficulties are by-passed if we remember the modified effective range function $`X_0(\nu )`$ of the proton-proton scattering, which is regular at $`\nu =0`$ and accepts the effective range expansion, when all the forces are short range except for the terms of the Coulomb and of the vacuum polarization. The modified effective range function $`X_0(\nu )`$ for the phase shift $`\delta _0^E(\nu )`$ is
$$X_0(\nu )=\frac{C_0^2\sqrt{\nu }}{1\varphi _0}\{(1+\chi _0)\mathrm{cot}\delta _0^E\mathrm{tan}\tau _0\}+me^2h(\eta )+me^2\mathrm{}_0(\eta ).$$
(8)
In Eq.(8), two well-known functions with the Coulombic order of magnitudes appear, they are expressed using a new variable $`\eta =me^2/(2\sqrt{\nu })`$ :
$$C_0^2=\frac{2\pi \eta }{e^{2\pi \eta }1}andh(\eta )=\eta ^2\underset{\mathrm{}=1}{\overset{\mathrm{}}{}}\frac{1}{\mathrm{}(\mathrm{}^2+\eta ^2)}\mathrm{log}\eta 0.57722\mathrm{}.$$
(9)
In Eq.(7) $`\tau _0`$ is the phase shift due to the vacuum polarization potential
$$V^{vac}(r)=\lambda \frac{e^2}{r}_{4m_e^2}^{\mathrm{}}𝑑t\frac{e^{r\sqrt{t}}}{2t}(1+\frac{2m_e^2}{t})\sqrt{1\frac{4m_e^2}{t}}\lambda \frac{e^2}{r}I(r),$$
(10)
where $`m_e`$ is the mass of the electron and $`\lambda =2e^2/3\pi =1.549\times 10^3`$. Functions $`\tau _0`$, $`\chi _0`$, $`\varphi _0`$ and $`\mathrm{}_0(\eta )`$ have the order of magnitudes of the vaccum polarization, and it is sufficient for our purpose to retain only the first term in $`\lambda `$. In such approximation, they are expressed by $`F_0(r)`$ and $`G_0(r)`$, which are the regular and the irregular Coulombic wave functions respectively, as
$$\tau _0=2\eta \lambda _0^{\mathrm{}}𝑑r\frac{F_0(r)^2I(r)}{r}$$
(11)
$$\chi _0=\varphi _0=2\eta \lambda _0^{\mathrm{}}𝑑r\frac{F_0(r)G_0(r)I(r)}{r}$$
(12)
$$\mathrm{}_0(\eta )=\lambda _0^{\mathrm{}}𝑑r\frac{I(r)}{r}[(CG_0(r))^2(CG_0(r))_{\nu =0}^2],$$
(13)
where $`I(r)`$ is defined in Eq.(10). Exact definitions of these functions are found in the paper by Heller.
By using the modified effective range function $`X_0(\nu )`$, we define the S-wave amplitude $`h_0(\nu )`$ of the p-p scattering by
$$h_0(\nu )=\frac{\sqrt{m^2+\nu }}{X_0(\nu )me^2h(\eta )i\sqrt{\nu }C_0^2}.$$
(14)
The relation between $`h_0(\nu )`$ and the phase shift $`\delta _0^E`$ is obtained if we substitute $`X_0(\nu )`$ of Eq.(8) into Eq.(14), and which reduces to the well-known form
$$h_0(\nu )=\frac{1}{C_0^2}\frac{\sqrt{m^2+\nu }}{\sqrt{\nu }}e^{i\delta _0^E(\nu )}\mathrm{sin}\delta _0^E(\nu ),$$
(15)
if the functions related to the vacuum polarization are neglected. The form of $`h_0(\nu )`$ of Eq.(15) is the same as that of the neutron-neutron scattering of Eq.(1) except for the factor $`C_0^2`$ given in Eq.(9), which is the penetration factor. If we compare the S-wave amplitude of the p-p scattering with that of the n-n scattering, which are given in Eq.(14) and Eq.(3) respectively, a combination of functions $`(me^2h(\eta )i\sqrt{\nu }C_0^2)`$ appears in place of $`i\sqrt{\nu }`$. In order to investigate the analytic structure of $`h_0(\nu )`$, it is convenient to rewrite the combination as
$$me^2h(\eta )i\sqrt{\nu }C_0^2=i\sqrt{\nu }+me^2\{\mathrm{log}(i\eta )\psi (1+i\eta )\}.$$
(16)
Since the digamma function $`\psi (z)`$ has poles at non-positive integers, the poles on the $`\eta `$-plane appear on the positive imaginary axis. In terms of $`\sqrt{\nu }`$, which is $`me^2/(2\eta )`$, the series of poles appear on the negative imaginary axisis and converge to $`\sqrt{\nu }=0`$. It is the smallness of the fine structure constant $`e^2`$ and therefore of the residues of such poles that zeros of the denominator of Eq.(14) occur at points very close to the locations of the poles of $`(me^2h(\eta )i\sqrt{\nu }C_0^2)`$. Therefore the partial wave amplitude $`h_0(\nu )`$ of the p-p scattering has a series of poles on the second sheet of $`\nu `$, namely on the lower half plane of $`\sqrt{\nu }`$, whereas on the first sheet of $`\nu `$ the analytic structure of $`h_0(\nu )`$ does not change compared to the case of the n-n scattering. This fact implies that the same definition of the Kantor amplitude $`K_0(\nu )`$ introduced for the neutron-neutron scattering, which is given in Eq.(5), is valid also for the proton-proton scattering, as long as we evaluate Im$`h_0(\nu ^{})`$ of Eq.(5) from Eqs.(14) and (16). Therefore the Kantor amplitude of the p-p scattering $`K_0(\nu )`$ constructed in this way is free from the singularities in the neighborhood of $`\nu =0`$, and so does not have the cut of the vacuum polarization as well as that of the Coulomb interaction.
## 3 Numerical Calculation of the Kantor <br>Amplitude
Since the scattering length of the S-wave amplitude of the p-p scattering is known with high precision, we shall analyse the once subtracted Kantor amplitude, which is
$$K_0^{once}(\nu )\frac{K_0(\nu )K_0(0)}{\nu }.$$
(17)
Merits to use the once subtracted Kantor amplitude rather than $`K_0(\nu )`$ are twofold. The first one is the higher convergence of the integration
$$K_0^{once}(\nu )=\frac{h_0(\nu )h_0(0)}{\nu }\frac{1}{\pi }_0^{\mathrm{}}𝑑\nu ^{}\frac{Imh_0(\nu ^{})}{\nu ^{}(\nu ^{}\nu )},$$
(18)
namely the uncertainty arising from the lack of the very accurate data in the higher energy region is largely suppressed because of the extra factor $`\nu ^{}`$ in the denominator of the integrand. The second one is the change of the power of the threshold behavior of the spectral function from $`C^{}(\nu ^{})^\gamma `$ to $`C^{}(\nu ^{})^{\gamma 1}`$, and which makes it easier for us to recognize the extra singularity at $`\nu =0`$ in $`K_0^{once}(\nu )`$. Since the power $`\alpha `$, which appears in the asymptotic behavior of the potential as $`V(r)C_\alpha /r^\alpha `$, and the power $`\gamma `$ of the spectral function Im$`h_0(\nu ^{})`$ are related by
$$\alpha =2\gamma +3,$$
(19)
$`\gamma `$ is equal to $`3/2`$ for the van der Waals potential of the London type, and which is expected to occur in the dyon model of hadron. Therefore in such a case the behavior of the once subtracted Kantor amplitude in the neighborhood of $`\nu =0`$ is that
$$K_0^{once}(\nu )=\mathrm{const}.+C^{}\sqrt{\nu }+\mathrm{}\mathrm{with}C^{}>0,$$
(20)
and we must observe the extra singularity as a cusp at $`\nu =0`$ unless the coefficient $`C^{}`$ is very small.
Let us represent the effective range function of the p-p scattering $`X_0(\nu )`$ by a meromorphic function of $`\nu `$:
$$\frac{X_0(\nu )c_0c_1\nu }{\nu }(1\frac{\nu }{\nu _P})=a_0+a_1\nu +\frac{b_0+b_1\nu }{d_0+d_1\nu +\nu ^2},$$
(21)
in which $`\nu _P`$ is the location of the pole of $`X_0(\nu )`$. $`c_0`$ and $`c_1`$ are the inverse of the scattering length and the effective range devided by 2 respectively. The r.h.s. of Eq.(21) involves six free parameters to be fitted. In the fitting, the Compton wave length of the neutral pion will be used as the unit of the length. In the search of the free parameters, $`c_0`$, $`c_1`$ and $`\nu _P`$ are fixed beforehand which are
$$c_0=0.18698,c_1=0.9541and\nu _P=6.39.$$
(22)
Accuracies of $`c_0`$ and $`c_1`$ are as high as 0.05% and 0.6% respectively. On the other hand, uncertainty of $`\nu _P`$ is around 3 per cent. However reasonable accuracy of $`\nu _P`$ is sufficient for our purpose to search for the extra singularity at $`\nu =0`$ .
The S-wave phase shifts of the p-p scattering by the Nijmegen group are used as the input to evaluate the effective range function $`X_0(\nu )`$. The once subtracted form of $`X_0(\nu )`$, in which the pole at $`\nu =\nu _P`$ is eliminated by multiplying a factor $`(1\nu /\nu _P)`$, is displayed in figure 2. The points in the figure come from the multi-energy phase shifts of Nijmegen-90 .
The box at the left lower corner in the figure is enlarged and displayed in figure 3, and in which the low energy data in the energy range $`0.1MeV.T_{lab}30MeV.`$ of Nijmegen-88 are shown with their error bars. The data in $`0.5MeV.T_{lab}100MeV.`$ are fitted by the meromorphic function of Eq.(21), and the six parameters are determined by the chi square search. The searched curve is shown in figures 2 and 3 as the full line, and whose parameters are tabulated in the column \[sc\] in Table 1. On the other hand the dash curve in the figures is a different fit by a meromorphic function, in which the six parameters are fixed by the six data points at $`T_{lab}=`$ 1, 5, 10, 25, 50, 100 MeV. of the Nijmegen-90. The parameters of the fixed curve are tabulated in the column \[fc\] in Table 1.
In the same table, there are also columns of the curves of \[$`l`$c\] and \[uc\], in which the parameters are determined by fitting to the lower fringe and to the upper fringe of the error bars in $`0.5MeV.T_{lab}100MeV.`$ respectively. Such parameters are necessary to estimate the errors of the Kantor amplitude.
It is remarkable that in figures 2 and 3, the curvature of the curve in the small $`T_{lab}`$ region increases very rapidly as $`T_{lab}`$ decreases to zero. If we remember that the starting points of the spectra of the one-pion exchange(OPE) and of the two-pion exchange(TPE) are $`T_{lab}=9.71MeV.`$ and $`38.83MeV.`$ repectively, it is not easy to understand the rapid change of the curvature in the low energy region such as $`0<T_{lab}<10MeV.`$ . On the other hand, if we accept the long range force, the appearance of such a cusp at $`T_{lab}=0`$ is what is expected. Since the separation of the OPE term from $`X_0(\nu )`$ is not simple, in the present article we shall examine $`K_0^{once}(\nu )`$ because, contrary to $`X_0(\nu )`$, the separation of the OPE term from the Kantor amplitude is straightforward.
In Table 2, the locations of the poles of $`h_0(\nu )`$, which is introduced in Eq.(14), on the first sheet of $`\nu `$ and their residues are tabulated. The columns correspond to those in Table 1, and each of them has two real poles $`\nu _1`$ and $`\nu _2`$, and a pair of complex poles $`(\nu _c,\nu _c^{})`$ respectively, on the first sheet of $`\nu `$.
In Table 3, $`\stackrel{~}{K}_0^{once}(\nu )`$, which are the once subtracted Kantor amplitude of Eq.(17) minus the one-pion exchange contribution, are tabulated for \[sc\] and \[fc\]. Because of Eq.(7),
$$\stackrel{~}{K}_0^{once}(\nu )=\frac{r_1}{\nu _1(\nu \nu _1)}+\frac{r_2}{\nu _2(\nu \nu _2)}+2Re\{\frac{r_c}{\nu _c(\nu \nu _c)}\}+\mathrm{\Delta }_0(OPEC)$$
(23)
, where OPEC is the one-pion exchange contribution. The correction term $`\mathrm{\Delta }_0`$ arises from the approximation to $`X_0(\nu )`$ by the meromorphic function fitted to the phase shift in the energy range of $`0.5MeV.<T_{lab}<100MeV.`$.
However the numerical value of $`\mathrm{\Delta }_0`$ is extremely small, namely less than $`2\times 10^3`$ or less than 10% of the error of $`\stackrel{~}{K}_0^{once}(\nu )`$ for $`0<\nu <3`$. The smallness of $`\mathrm{\Delta }_0`$ comes from the facts that firstly the meromorphic function reproduce $`X_0(\nu )`$ well in the wider domain of $`T_{lab}\stackrel{<}{}180MeV.`$, secondly in the neighborhood of $`T_{lab}=248MeV.`$, where the phase shift passes through zero, the spectral function Im $`h_0(\nu ^{})/\nu ^{}`$ is small and thirdly in higher $`\nu ^{}`$ region the spectral function is suppressed by the factor $`\nu ^{}`$, because we are computing the once sutracted Kantor amplitude. The last column of Table 3 is the error of $`\stackrel{~}{K}_0^{once}(\nu )`$, which is defined by
$$\mathrm{\Delta }\stackrel{~}{K}_0^{once}(\nu )=\frac{1}{2}\{|\stackrel{~}{K}_0^{[uc]once}(\nu )\stackrel{~}{K}_0^{[sc]once}(\nu )|+|\stackrel{~}{K}_0^{[lc]once}(\nu )\stackrel{~}{K}_0^{[sc]once}(\nu )|\}.$$
(24)
In figure 4, $`\stackrel{~}{K}_0^{once}(\nu )`$ is plotted against $`T_{lab}`$, and in which the diamonds are \[sc\] and the triangles are \[fc\] respectively. Figure 5 is the enlarged graph of the lower energy part of figure 4. The curves in the figures are that $`(L)_3`$ is the 3-parameter fit by the spectral function of the long range force, whereas $`(sa)_3`$ (dotted curve) and $`(sb)_3`$ (dash curve) are the fits by the short range potential with three free parameters. Details of the curves will be explained in sections 5. A characteristic feature of $`\stackrel{~}{K}_0^{once}(\nu )`$ is that it has a sharp cusp at $`\nu =0`$, contrary to what is expected in the meson theory of the nuclear force, where the location of the two-pion exchange spectrum is ‘far away’ compared to the width of the cusp. In the one-pion exchange term of Eq.(6), $`g^2/4\pi =14.4`$ is used as the $`\pi `$-$`N`$ coupling constant. However modifications to the different values of $`g^2/4\pi `$ are straightforward if we remember Eq.(4). It must be pointed out that as the coupling constant $`g^2/4\pi `$ decreases the cusp of $`\stackrel{~}{K}_0^{once}(\nu )`$ becomes more prominent.
In the next two sections, $`\stackrel{~}{K}_0^{once}(\nu )`$ of Table 3 will be analysed to understand the physical meanings of the cusp at $`\nu =0`$. In section 4, the cusp will be fitted by the spectrum of the long range force, in which the power $`\gamma `$ of the threshold behavior of the spectrum will be searched. On the other hand, in section 5 it will be examined whether the short range nuclear potential $`V(r)`$ can reproduce well the cusp of $`\stackrel{~}{K}_0^{once}(\nu )`$.
## 4 Fits by the Spectrum of the Long Range <br>Interactions
Figures 4 and 5 indicate that $`\stackrel{~}{K}_0^{once}(\nu )`$ has a cusp pointed upward at $`\nu =0`$. If there exists the spectrum of the long range force with the attractive sign, appearance of such a cusp in $`\stackrel{~}{K}_0^{once}(\nu )`$ is a natural consequence. In this section, we shall determine the power $`\gamma `$ and the strength of the threshold behavior of the spectral function at $`\nu =0`$. More conventional fit will be tried in the next section, in which the possibility to understand the cusp of $`\stackrel{~}{K}_0^{once}(\nu )`$ in terms of the ordinary short range potential of the nuclear force will be considered. The integral representation of the scattering amplitude $`\stackrel{~}{A}(s,t)`$ of the non-OPE part is
$$\stackrel{~}{A}(s,t)=\frac{1}{\pi }_{t_0}^{\mathrm{}}𝑑t^{}\frac{\stackrel{~}{A}_t(s,t^{})}{t^{}t}\pm \frac{1}{\pi }_{u_0}^{\mathrm{}}𝑑u^{}\frac{\stackrel{~}{A}_u(s,u^{})}{u^{}u},$$
(25)
where $`t_0`$ and $`u_0`$ are $`4\mu ^2`$ for the short range potential whereas zero for the long range interaction, respectively. The tilde such as $`\stackrel{~}{A}`$ means the function in which the OPE part is already subtracted. Since we are considering the scattering of the identical fermi particles, the signs between the integrations of Eq.(25) are plus for the spin-singlet and minus for the spin-triplet states respectively. Moreover the spectral functions of t and u channels are the same, namely $`\stackrel{~}{A}_t(s,)=\stackrel{~}{A}_u(s,)`$. The partial wave projection of Eq.(25) gives
$$\stackrel{~}{h}_0(\nu )=\frac{1}{\pi }_0^{\mathrm{}}𝑑t\stackrel{~}{A}_t(s,t)\frac{1}{2\nu }Q_0(1+\frac{t}{2\nu })+(tu).$$
(26)
Therefore the spectral function of the left hand cut of the S-wave amplitude $`\stackrel{~}{h}_0(\nu )`$ is
$$\mathrm{Im}\stackrel{~}{h}_0(\nu )=\frac{1}{4\nu }\{_{t_0}^{4\nu }𝑑t\stackrel{~}{A}_t(s,t)+_{u_0}^{4\nu }𝑑u\stackrel{~}{A}_u(s,u)\}for\nu <t_0/4.$$
(27)
Let us search the value of the power $`\gamma `$ of the spectral function of $`\stackrel{~}{A}_t(4m^2,t)`$ at $`t=0`$ or of $`\mathrm{Im}\stackrel{~}{h}_0(\nu )`$ at $`\nu =0`$. Because of Eq.(27), the powers $`\gamma `$’s of these functions are the same. Since $`\stackrel{~}{A}_t(4m^2,t)`$ directly relates to the potential by the Laplace transformation
$$rV(r)=\frac{1}{\pi m^2}_0^{\mathrm{}}𝑑t^{}A_t(4m^2,t^{})e^{r\sqrt{t^{}}},$$
(28)
we shall assume the functinal form of $`\stackrel{~}{A}_t(4m^2,t)`$ and search the parameters involved. In the search of the power $`\gamma `$, three-parameter form of the spectral function will be used, which is
$$\stackrel{~}{A}_t(4m^2,t)=\pi C^{}t^\gamma \mathrm{exp}[\beta t],$$
(29)
where the exponential factor is necessary to make the integration Eq.(26) convergent.<sup>3</sup><sup>3</sup>3Moreover the parameter $`\beta `$ plays the roll to incorporate the two-pion exchange spectrum in Eq.(29).
In the search, 68 points of $`\stackrel{~}{K}_0^{once}(\nu )`$ in $`0.6MeV.<T_{lab}<125MeV.`$ are fitted. The fitted points are chosen to be equispacing with respect to $`\sqrt{\nu }`$, or more precisely $`\sqrt{\nu _n}=0.025n`$ with $`5n72`$. The parameters $`(\gamma ,\beta ,C^{})`$ and $`\chi `$ of the best fits are
$`\gamma =1.543,\beta =0.06264,C^{}=0.1762and\chi =0.441`$ $`for`$ \[sc\] (30)
$`\gamma =1.569,\beta =0.06544,C^{}=0.1734and\chi =0.343`$ $`for`$ \[ fc \] (31)
respectively. The curves in figures 4 and 5 are the fits of \[sc\] (full line) and \[fc\] (dash line) respectively.
In figure 6, $`\chi (\gamma )`$ is plotted against $`\gamma `$, in which the full line is \[sc\] and the dash line is \[fc\] respectively. $`\chi (\gamma )`$ is obtained by making the chi-square search for fixed $`\gamma `$. From the curves, minimum points are determined, which are $`\gamma _{min}=1.543\pm 0.055`$ and $`1.569\pm 0.044`$ for \[sc\] and \[fc\] respectively. The range of the uncertainty of $`\gamma _{min}`$ is estimated by the condition that $`\chi (\gamma )1.25\chi _{min}`$, in which $`\chi _{min}`$ is the minimum value of $`\chi (\gamma )`$. It is remarkable that the value of the power $`\gamma `$ is close to what is expected in the case of the van der Waals potential of the London type namely to $`\gamma =1.50`$.
However before we discuss about the physical meaning of the value that $`\gamma 1.5`$, it is necessary to examine the dependency of $`\gamma _{min}`$ on the fitted range of energy. In figure 7, three curves of $`\chi (\gamma )`$ with different fitted energy domains are shown. They are $`(0.630)`$ MeV. and $`(0.6175)`$ MeV., in addition to the domain $`(0.6125)`$ MeV., which was shown in fig.6. The figure indicates that as the energy domain shrinks, $`\gamma _{min}`$ moves slowly to samaller value. For $`0.6MeV.<T_{lab}<30MeV.`$ the best fit is that $`\gamma _{min}=1.467\pm 0.111`$ and $`\chi _{min}=0.423`$ for \[sc\]. On the other hand, as the energy region expands, $`\gamma _{min}`$ increases and moreover the value of $`\chi _{min}`$ increases rapidly. Therefore the 3-parameter function of Eq.(29) is not suitable for the fit to so many data points. In fact, for the domain $`0.6MeV.T_{lab}175MeV.`$, the curve in the figure indicates that $`\gamma _{min}=1.695\pm \genfrac{}{}{0pt}{}{0.192}{0.150}`$ and $`\chi _{min}=1.410`$ for \[sc\]. Therefore even if we change the energy domain, the power $`\gamma `$ still remains in the neighborhood of 1.5 .
## 5 Fits by the Spectral Function of the Short Range Interactions
In the previous section, it was found that the cusp in $`\stackrel{~}{K}_0(\nu )`$ obtained in section 3 was reproduced well by the spectral function $`\stackrel{~}{A}_t(4m^2,t)`$ of the long range interaction given in Eqs.(29),(30) and (31). However it is important to ask whether the cusp can be understood also in the framework of the more conventional approach, namely in terms of the spectral function of the short range interaction expected in the meson theory. It is well known that, in the meson theory of the nuclear force, the spectral function of the two-pion exchange is
$$\stackrel{~}{A}_t(4m^2,t)=\{\begin{array}{cc}0\hfill & \mathrm{in}t4\mu ^2\hfill \\ \pi C_{2\pi }^{}(t4\mu ^2)^{1/2}+\mathrm{}\hfill & \mathrm{in}t>4\mu ^2\hfill \end{array}$$
(32)
, in the neighborhood of the two-pion threshold at $`t=4\mu ^2`$. The power 1/2 of Eq.(32) comes from the functional form of the phase volume of the two-particle state exchanged. The coefficient $`C_{2\pi }^{}`$ can be estimated from the phase shifts of the $`\pi `$-$`N`$ and the $`\pi `$-$`\pi `$ scatterings by using the generalized unitarity of the dispersion technique. From Eqs.(27) and (32), the spectral function of the left hand cut of $`\stackrel{~}{h_0}(\nu )`$, which is the same as that of $`\stackrel{~}{K_0}(\nu )`$, becomes
$$\mathrm{Im}\stackrel{~}{h}_0(\nu ^{})=\{\begin{array}{cc}0\hfill & \mathrm{in}\nu ^{}>1\hfill \\ \frac{8\pi }{3\nu ^{}}C_{2\pi }^{}(\nu ^{}1)^{3/2}+\mathrm{}\hfill & \mathrm{in}\nu ^{}1.\hfill \end{array}$$
(33)
Since the power of the spectral function at the threshold $`\nu =1`$ increases to 3/2, the distance of the average location of the spectrum from $`\nu =0`$ is much larger compared to that of the threshold point. With such a spectral function, it will be not easy to reproduce the cusp of $`\stackrel{~}{K}_0^{once}(\nu )`$ observed in $`\nu 1`$, where in our unit $`\nu =1`$ corresponds to $`T_{lab}=38.83MeV.`$ . On the other hand, the updated nuclear potentials are believed to fit well to the precise phase shift data..
In this section, we shall examine the updated nuclear potentials more closely, and in which we shall concentrate on the cusp at $`\nu =0`$. Among various potentials, we shall choose the regularized Reid potential updated by the Nijmegen group (Reid 93) as an example, in which 50 phenomenological parameters are searched to fit to 1787 p-p data and 2514 n-p data in the energy range of $`0<T_{lab}<350MeV.`$. It is remarkable that by using such potential they achieved excellent fits $`\chi ^2=1.03`$ per datum, which is close to the chi-square value $`0.99`$ of the energy dependent phase shift analysis. In the Reid93, the S-wave potential of the non-OPE part of the p-p scattering involves five parameters:
$$V_{pp}(^1S_0)=A_2Y(2)+A_3Y(3)+A_4Y(4)+A_5Y(5)+A_6Y(6),$$
(34)
where the regularized Yukawa functions of the range $`1/(p\mu )`$ are
$$Y(p)=p\frac{1}{r}[e^{p\overline{\mu }r}e^{\mathrm{\Lambda }r}(1+\frac{\mathrm{\Lambda }^2p^2\overline{\mu }^2}{2\mathrm{\Lambda }^2}\mathrm{\Lambda }r)]$$
(35)
and $`\overline{\mu }`$ is the average masses of the three types of pions and $`\mathrm{\Lambda }=8\overline{\mu }`$. Contrary to the original Reid potential, this potential has a characteristic feature that the $`Y(2)`$ term exists. Such term gives rise to a pole at $`t=4\mu ^2`$ in the scattering amplitude $`A(s,t)`$, whereas in the meson theory $`t=4\mu ^2`$ is merely the starting point of the continuous spectrum of the two-pion exchange. Therefore it is not desirable from the point of view of the meson theory to have the $`Y(2)`$ term in Eq.(34). On the other hand the inclusion of the $`Y(2)`$ term is necessary, in fitting the very precise low energy data of the p-p scattering, to reduce the the chi-square value. The contradictory situation already indicates the difficulty to reproduce the cusp from the far-away spectral function of the meson theory of the nuclear force.
In figure 5, results of the 3-parameter fits to the data points in $`0.6MeV.<T_{lab}<125MeV.`$ are compared. The curve $`(L)_3`$ is the fit by the spectrum of the long range force, while the curves of $`(sa)_3`$ and $`(sb)_3`$ are the fits by the short range potentials of the Reid types of Eq.(34). The $`(sa)_3`$ is the sum of $`Y(2)`$, $`Y(3)`$ and $`Y(4)`$, on the other hand $`(sb)_3`$ is the sum of $`Y(3)`$, $`Y(4)`$ and $`Y(5)`$, where $`Y(p)`$ is the regularized Yukawa potential of range $`1/p`$ defined in Eq.(35). The $`\chi `$-values per datum of $`(L)_3`$, $`(sa)_3`$ and $`(sb)_3`$ are 0.441, 1.82 and 3.11 respectively. It is evident that the short range potentials with three parameters cannot reproduce the cusp at $`\nu =0`$. It is interesting that the inclusion the $`Y(2)`$ term always reduce the $`\chi `$-value, this is because the lack of the spectrum of small $`t`$ is the reason of the deviation from the data points of $`\stackrel{~}{K}_0^{once}(\nu )`$.
In figure 8, the seven curves are the fits to $`\stackrel{~}{K}_0^{once}(\nu )`$ of \[sc\] in the off-cusp region $`T_{lab}=`$ $`(20125)MeV.`$. The fitted curves are then extrapolated to $`T_{lab}=0`$. The $`(L)_3`$ (full line) curve is the 3-parameter fit by the spectrum of the long range force of Eq.(29) as before, on the other hand $`(sa)_n`$ (dotted line) and $`(sb)_n`$ (dash line) are the n-paremeter fits by the Reid type potentials. The differences of $`(sa)_n`$ and $`(sb)_n`$ are that, although $`(sa)_n`$ involves $`Y(2)`$ term, in $`(sb)_n`$ the Yukawa term of the longest range is $`Y(3)`$. In general as the numbers of the parameters $`n`$ increase, the curves come closer to the data points. However the curves of Fig.8 indicate that the curves of the short range interaction $`(sb)_n`$ cannot reproduce the cusp of the data points at $`\nu =0`$.
In the meson theory, the nearest singularity of $`\stackrel{~}{K}_0^{once}(\nu )`$ occurs at $`\nu =1`$ or in terms of $`T_{lab}`$ at $`38.83MeV.`$, which is the threshold of the two-pion exchange spectrum. Since Eq.(33) indicates that the power of of the threshold behavior of the two-pion exchange spectrum at $`\nu =1`$ is 3/2, the distance to the average location of the spectrum from the origin is much larger than one. What is important is that the length of our extrapolation ($``$ 0.5) is short compared not only to the size of the fitted domain ($`3.0`$) but also to the distance to the left hand spectrum ($`>1`$). Therefore if the analytic structure expected in the meson theory is correct, the extrapolated curve must approximately follow the data points. However figure 8 indicates that this is not the case. On the other hand, the extrapolated curve of the long range force $`(L)_3`$ traces the data point nicely. Therefore we conclude that the analytic structure of the amplitude of the nucleon-nucleon scattering is different from what is expected in the meson theory of the nuclear force, rather the nuclear force involves the large component of the long range force.
## 6 Strong van der Waals Interaction as a <br>Candidate of the Nuclear Force
In section 2, we introduced the S-wave Kantor amplitude of the proton-proton scattering $`K_0(\nu )`$, which was free from the singularities of the vacuum polarization and the Coulombic interaction as well as the unitarity cut. By subtracting the one-pion exchange contribution from $`K_0(\nu )`$, the domain of analyticity expands further, and the nearest singularity of the non-OPE Kantor amplitude $`\stackrel{~}{K}_0(\nu )`$ occurs at $`\nu =1`$, and which is the branch point of the two-pion exchange spectrum. By using the phase shift data of the Nijmegen group , the once subtracted non-OPE Kantor amplitude $`\stackrel{~}{K}_0^{once}(\nu )`$ defined in Eqs.(6)and (18) was calculated in section 3. The appearance of the cusp at $`\nu =0`$ is rather surprising, because in the meson theory of the nuclear force, $`\stackrel{~}{K}_0^{once}(\nu )`$ is regular at least in $`|\nu |<1`$.
It turns out that the conventional fits of the spectra of the short range forces cannot reproduce the cusp properly even if the potentials with five parameters given in Eq.(34) are used in the fitting. On the other hand, if we use the spectra of the long range force, three parameters of Eq.(29) are sufficient to reproduce the cusp of $`\stackrel{~}{K}_0^{once}(\nu )`$. In summary, the short range condition, that Im$`h_0(\nu )0`$ in $`\nu >1`$, is too severe to reproduce the cusp in $`|\nu |1`$. Therefore it is natural to regard the cusp as an effect of the long range interaction between hadrons. The power $`\gamma `$ of the threshold behavior was searched in section 4, and it turned out that $`\gamma =1.543\pm 0.056`$ from the multi-energy data \[sc\] and $`\gamma =1.569\pm 0.046`$ determined from the fit \[fc\], respectively. Since the power $`\alpha `$ of the asymptotic behavior of the long range potential $`V(r)`$ is related to the power $`\gamma `$ by $`\alpha =2\gamma +3`$, the values of $`\gamma `$ imply $`\alpha =6.09\pm 0.11`$ and $`\alpha =6.17\pm 0.09`$ for \[sc\] and \[fc\] respectively.
It is remarkable that the values of $`\alpha `$ are close to that of the van der Waals potential of the London type ( $`\alpha =6`$ ). Although $`\alpha `$ can assume any value from $`1`$ (Coulomb) to infinity (Yukawa), Nature chooses a special value in the neighborhood of $`\alpha =6`$. Therefore it is difficult to believe the occurrence of the value $`\alpha =5.986.26`$ is merely accidental, rather the ordinary mechanism to produce the van der Waals interaction must be working. It is well-known in the atomic and molecular physics that, when the underlying Coulombic force exists, the dipole-dipole interaction occurs between the neutral composite particles, and the two-step process of such interactions causes the attractive potential. The form of the potential between particle 1 and particle 2 is
$$V(R)=\stackrel{}{}_n\frac{(H^{})_{0n}(H^{})_{n0}}{E_nE_0},$$
(36)
where $`H^{}`$ is the hamiltonian of the dipole-dipole interaction
$$H^{}=\frac{{}_{}{}^{}e_{}^{2}}{R^3}(x_1x_2+y_1y_22z_1z_2),$$
(37)
in which $`{}_{}{}^{}e_{}^{2}`$ is the strength of the underlying Coulombic interaction.
It is important to notice that each term in the summation of Eq.(36) is positive definite. Therefore if we replace all the denominators $`(E_nE_0)`$ in Eq.(36) by the first excitation energy $`(\mathrm{\Delta }E_1)`$, then by using the closure property the upper bound of the strength $`C_6`$ of the van der Waals potential, which is introduced by $`V(R)=C_6/R^6+\mathrm{}`$, is obtained. It is
$$C_6<\frac{R^6(H_{}^{}{}_{}{}^{2})_{00}}{(\mathrm{\Delta }E_1)}=\frac{2}{3}\frac{{}_{}{}^{}e_{}^{4}}{(\mathrm{\Delta }E_1)}\overline{r_1^2}\overline{r_2^2}.$$
(38)
On the other hand, if we retain only the terms of the first excited states in the summation, Eq.(36) gives the lower bound of $`C_6`$, namely
$$C_6>\frac{2}{3}\frac{{}_{}{}^{}e_{}^{4}}{(\mathrm{\Delta }E_1)}\stackrel{~}{R_1}^2\stackrel{~}{R_2}^2,$$
(39)
where $`\stackrel{~}{R_i}^2`$ relates to the dipole transition amplitude $`\stackrel{~}{z}`$ by
$$\stackrel{~}{R}^2=3|\stackrel{~}{z}|^2\mathrm{with}\stackrel{~}{z}=d^3r\psi _{1,1,0}^{}(r\mathrm{cos}\theta )\psi _{0,0,0}.$$
(40)
Since we know the value of $`C_6`$, Eqs.(38) and (39) will be used in turn to give the lower and the upper bounds of the strength of the underlying Coulombic force $`{}_{}{}^{}e_{}^{2}`$ respectively, and which are
$$\sqrt{\frac{3}{2}C_6(\mathrm{\Delta }E_1)}\frac{1}{\overline{r^2}}<^{}e^2<\sqrt{\frac{3}{2}C_6(\mathrm{\Delta }E_1)}\frac{1}{\stackrel{~}{R}^2}$$
(41)
The numerical values of $`C^{}`$ and $`\beta `$ of Eq.(29) for $`\gamma =1.5`$ are obtained by shrinking the energy range to be fitted. The results are
$`\gamma `$ $`=1.5`$ $`,\beta =0.0547andC^{}=0.174for[sc]`$ (42)
$`\gamma `$ $`=1.5`$ $`,\beta =0.0541andC^{}=0.172for[fc]`$ (43)
in the unit of the neutral pion mass, and the energy range of the fits are $`(0.634)`$ MeV. and $`(0.639)`$ MeV. for \[sc\] and \[ fc \] respectively. Since the coefficient $`C_\alpha `$ of the asymptotic potential is
$$C_\alpha =C^{}\frac{2\mathrm{\Gamma }(\alpha 1)}{m^2}\mathrm{with}\alpha =2\gamma +3,$$
(44)
$`C_6=0.9933C^{}`$ in our unit. If we use the mass difference of $`\mathrm{\Delta }(1232)`$ and $`N`$ as the first excitation energy, namely $`\mathrm{\Delta }E=2.18`$, then the lower bound of $`{}_{}{}^{}e_{}^{2}`$ becomes
$${}_{}{}^{}e_{}^{2}>\frac{\sqrt{3.27C^{}}}{\overline{r_1^2}}.$$
(45)
To go further we need the size of the composite particle, the radius of proton in particular. Although we do not have much information on the radius, we shall consider two cases. In the first case, $`\overline{r_1^2}`$ is $`(0.7\mathrm{fm}.)^2`$, which comes from the nucleon form factor. In another case $`\overline{r_1^2}`$ is set equal to $`\beta `$ of Eqs.(42) and (43), because at the range $`r=\sqrt{\beta }`$ the potential starts to deviate appreciablly from the asymptotic form of the van der Waals potential of $`V(R)C_6/R^6`$ and therefore $`\sqrt{\beta }`$ may be regarded as the size of the composite particle. The results are :
$`{}_{}{}^{}e_{}^{2}>3.27`$ $`(`$ $`\mathrm{from}\mathrm{the}\mathrm{form}\mathrm{factor})`$
$`{}_{}{}^{}e_{}^{2}>13.8`$ $`(`$ $`\mathrm{from}\overline{r_1^2}=\beta ).`$
As it is expected, the underlying Coulombic force is super-strong, because the induced van der Waals force has already the order of magnitude of the strong interaction.
The coupling constants $`{}_{}{}^{}e_{}^{2}`$ of the QED and the QCD are too small, they are $`(137.036)^1`$ and 0.32 respectively. On the other hand, the Coulombic force of the magnetic monopoles is an imortant candidate of the underlying Coulombic interaction. It is well-known from the charge quantization condition of Dirac that
$${}_{}{}^{}e_{}^{2}=\frac{137.036}{4}.$$
(46)
Therefore $`{}_{}{}^{}e_{}^{2}`$ of the magnetic charge is large enough to satisfies the condition of the lower bound, moreover it is not very large compared to the value of the lower bound. This situation is satisfactory, because the upper bound of Eq.(41) must have the value of the same order of magnitude as the lower bound.
Since the Coulombic interaction between the magnetic monopoles is a suitable candedite of the underlying dynamics of the nuclear force, it is worthwhile to repeat briefly the magnetic monopole model of hadron. Although it is overshadowed by the QCD, it still has virtue to be considered. The monopole model of hadron is essentially the quark model, in which the quarks bear the magnetic charges. Such a fundamental particle is often called dyon, since it is doublly charged, namely electrically and magnetically charged. Because of the superstrong Coulombic force between magnetic charges, the dyons form the bound states of magnetic charge zero, and such composite particles are identified with the hadrons. In a sense, hadrons are regarded as “ magnetic atoms ” in this model. Therefore it is not surprising to observe the strong van der Waals forces between hadrons. In this article we actually observed the strong van der Waals interaction in the proton-proton scattering. Since the van der Waals interaction is universal, we may expect to observe such a force also in other scattering processes. However because of the lower precision of the data, the elimination of the two-pion exchange cut is necessary at least in the threshold region of the spectrum. After carrying out such eliminations, we can actually observe the long range force in the P-wave amplitudes of the $`\pi `$-$`N`$ and the $`\pi `$-$`\pi `$ scatterings.
## 7 Remarks and Comments
In this paper we searched for the long range force in the S-wave amplitude of the proton-proton scattering. In order to see the extra singularity of the long range force in $`h_0(\nu )`$ at $`\nu =0`$ as clearly as possible, we constructed a regular function $`K_0(\nu )`$ from the modified effective range function $`X_0(\nu )`$ of the proton-proton scattering, where the function $`K_0(\nu )`$ does not have the normal singularities in the neighborhood of $`\nu =0`$. By separating the OPE part and then making the once subtracted function $`\stackrel{~}{K}_0^{once}(\nu )`$, it becomes much easier for us to observe the extra singularity at $`\nu =0`$ when it exists. It is impressive to observe that $`\stackrel{~}{K}_0^{once}(\nu )`$ has a sharp cusp at $`\nu =0`$, and the form of which is close to the square root type, namely to $`\stackrel{~}{K}_0^{once}(\nu )=(C_0^{\prime \prime }+C^{\prime \prime }\sqrt{\nu })`$ with positive $`C^{\prime \prime }`$. The $`\sqrt{\nu }`$ singularity is exactly what is expected in the van der Waals interaction of the London type. Moreover the observed sign of $`C^{\prime \prime }`$ indicates that the long range force is attractive.
If we state it more precisely, when the spectral function has the form of $`A_t(4m^2,t)=\pi C^{}t^\gamma \mathrm{exp}(\beta t)`$, the chi-square fit to the multi-energy phase shift data determines the three parameters involved. They are $`\gamma =1.54`$, $`\beta =0.063`$ and $`C^{}=0.175`$ in the unit of the neutral pion mass. It is interesting to see that the spectral function $`A_t(4m^2,t)`$ has a peak at $`\sqrt{t}=4.9`$ namely at 660MeV.. On the other hand, the location of the peak of the spectrum $`A_t(4m^2,t)/t`$ of the once subtracted amplitude is $`\sqrt{t}=2.93`$ or 395MeV.. When we do not need the very precise description of the low energy phase shifts, the spectral function can be replaced by a $`\delta `$-function located somewhere between 400 and 650MeV.. Such a $`\delta `$-function may correspond to the fictitious $`\sigma `$-meson of the one-boson exchange model of the nuclear potential .
Since the power $`\gamma `$ of the threshold behavior of the spectral function $`A_t(4m^2,t)`$ is close to that of the van der Waals force of the London type, we may assume that the ordinary mechanism, which produce the van der Waals force, is working. Then the strength $`{}_{}{}^{}e_{}^{2}`$ of the underlying Coulombic force can be estimated. This is possible because there are the well-known inequalities on the coefficient $`C_6`$ of the van der Waals potential, in particular the upper bound of which is obtained by using the strength $`{}_{}{}^{}e_{}^{2}`$ and the radius $`(\overline{r^2})^{1/2}`$ of the composite particle. Since we know the strength $`C_6`$ of the van der Waals potential from the fitting, the inequality gives instead the lower bound of $`{}_{}{}^{}e_{}^{2}`$, which depends on size of the composite state. For $`(\overline{r^2})^{1/2}=0.5`$, which is obtained by the measurement of the nucleon form factor, we obtain $`{}_{}{}^{}e_{}^{2}>3.3`$, whereas for $`(\overline{r^2})^{1/2}=\sqrt{\beta }=0.25`$, we obtain $`{}_{}{}^{}e_{}^{2}>14`$. In any case, the underlying Coulombic force is super-strong. Among the interactions of the Coulombic form, the force between the magnetic monopoles satisfies this inequalty, because $`{}_{}{}^{}e_{}^{2}=137.036/4`$ due to the charge quantization condition of Dirac. Therefore the dyon model of hadron, in which the constituent particle dyon bears the magnetic charge as well as the electric charge, must be an important candidate of the model of hadron.
Finally, in order to confirm the long range force in the proton-proton scattering, it is desirable to observe directly the difference of the interference patterns by measuring precisely the angular distributions of the cross section in the low energy experiments. Our proposal is to observe the interference between the repulsive Coulomb force and the attractive long range interaction obtained in the present article. Since the interference pattern of the conventional short range foece and the Coulomb force is known, the difference can be predicted by using the values of parameters $`\gamma `$, $`\beta `$ and $`C`$. It turns out that the most favorable energy is in $`T_{lab}=25MeV.40MeV.`$. Since the relative deviation, namely $`\mathrm{\Delta }(d\sigma /d\mathrm{\Omega })(d\sigma /d\mathrm{\Omega })^1`$, has a dip around $`\theta _{c.m.}=10^{}`$ with the depth $`0.3\%`$, there exist two challenging requirements in the measurements. They are the accuracy and the domain of $`\theta _{c.m.}`$. However since the shape of the interference pattern is important, there is no severe conditions on the normalization of the cross section. Details will be published in a separate paper.
|
warning/0004/hep-th0004139.html
|
ar5iv
|
text
|
# REGULATING THE 𝑃⁺=0 SINGULARITY
## 1 Introduction
I want to discuss the problems involved in regulating the $`p^+=0`$ singularity, a task which always must be done when using the light-cone representation (or light-cone gauge) to study quantum field theories. The $`p^+=0`$ point can have singularities which are associated with both the ultraviolet and the infrared structure of the theories. I shall give some examples of each type of singularity. For the case of ultraviolet singularities I shall briefly discuss the case of the one loop correction to the mass in Yukawa theory. For the case of infrared singularities I shall discuss the Schwinger model — the best understood case.
In each case we shall see that there are issues, unfamiliar from the equal-time representation, with how to regulate the theory and with how to find the operator mixing which results from the regulation. At the end I shall remark briefly on the import (in my view) of the examples to the problem of performing practical calculations for realistic theories in the light-cone representation.
## 2 Ultraviolet Singularities
The first example formed the starting point for the work that Stan Brodsky, John Hiller and I have been doing for some time . I want to use it to, among other things, warn you against an argument for the equivalence of the light-cone and equal-time representations, at least at the level of perturbation theory. The argument is sometimes given that if a finite (that is, regulated) Feynman integral is given by
$$d^4kf(k_\mu )$$
(1)
And if
$$𝑑k_+f(k_\mu )=g(k_{},k_{})$$
(2)
And if light-cone perturbation theory for the same quantity leads to the integral
$$𝑑k_{}d^2k_{}g(k_{},k_{})$$
(3)
Then clearly, at least perturbatively, the two formulations are the same.
Not so . Let us look at an example. For scalar Yukawa theory it is known that one Pauli-Villars field is sufficient to regulate the Fermion self-mass. The one loop correction to the mass is easily written down as
$$\stackrel{~}{u}(p)u(p)\delta m=\frac{i\alpha }{4\pi ^3}\stackrel{~}{u}(p)d^4k\frac{\gamma (pk)+m}{[(pk)^2m^2+iϵ][k^2\mu ^2]}u(p)[\mu \mu _1]$$
(4)
The integral can be evaluated in general but we need only the small-m limit, which is
$$\delta m=\frac{3}{2}\alpha m\mathrm{ln}\frac{\mu _1}{\mu }$$
(5)
We recall particularly, that the correction to the mass is zero if the bare mass is zero. That result is true to all orders since the zero-bare-mass Fermion is protected from gaining a mass by a discrete chiral symmetry. We can easily calculate that
$`(\stackrel{~}{u}(p)u(p))^1{\displaystyle \frac{i\alpha }{4\pi ^3}}\stackrel{~}{u}(p){\displaystyle }dk_+{\displaystyle \frac{\gamma (pk)+m}{[(pk)^2m^2+iϵ][k^2\mu ^2]}}u(p))[\mu \mu _1]=`$
$`{\displaystyle \frac{\alpha }{4m\pi ^2}}{\displaystyle \frac{p^2\stackrel{}{q}_{}^2+(2pq)^2m^2}{p^2\stackrel{}{q}_{}^2+q^2m^2+p(pq)\mu ^2}})[\mu \mu _1]`$ (6)
Furthermore, we find that light-cone perturbation theory gives
$$\frac{\alpha }{4m\pi ^2}_0^p\frac{dq}{p(pq)}d^2q_{}\frac{p^2\stackrel{}{q}_{}^2+(2pq)^2m^2}{p^2\stackrel{}{q}_{}^2+q^2m^2+p(pq)\mu ^2}[\mu \mu _1]$$
(7)
It is easy to check that the above integral is divergent. If we use two Pauli-Villars fields we find that the light-cone integral is now finite, but is not equal to the Feynman result and not proportional to m. We get
$$\delta m=\frac{\alpha }{8\pi ^2}\left[\underset{i=0}{\overset{2}{}}\mu _i^2\mathrm{ln}\mu _i^2m(\frac{11}{2}+\underset{i=0}{\overset{2}{}}\mathrm{ln}\mu _i^2)\right]$$
(8)
If we go to three Pauli-Villars fields the light-cone integral is both finite and proportional to m. Indeed, with three Pauli-Villars fields the renormalized light-cone series is equal to the renormalized Feynman series‘.
It is worthwhile understanding the source of the failure of the argument presented at the beginning of this section. The failure of the argument is due to the meaning of the word “convergent” when we say that the Feynman integral is “convergent”. The integral is conditionally convergent and thus any value ascribed to it is a prescription. The standard value presented above comes from the prescription: one Pauli-Villars field plus a Wick rotation. Since the integral is conditionally convergent it is guaranteed that there exists some way of expanding the domain of integration to cover the whole space such that any value will be obtained for the integral. The light-cone integral includes regions of the integration domain in an order not equivalent to a Wick rotation; the result might therefore be different and in this case it is.
If the “covariant regulator” is used, where the range of integration is limited by
$$\frac{\stackrel{}{k}^2+\mu ^2}{q}+\frac{\stackrel{}{k}^2+m^2}{pq}\frac{\mathrm{\Lambda }^2}{p}$$
(9)
the result for the Fermion self-mass is
$`{\displaystyle \frac{\alpha }{8\pi ^2}}[({\displaystyle \frac{\mathrm{\Lambda }^2}{2}}\mu ^2ln\mathrm{\Lambda }^2+\mu ^2ln\mu ^2{\displaystyle \frac{\mu ^4}{2\mathrm{\Lambda }^2}})`$
$`+m^2(4ln\mathrm{\Lambda }^24ln\mu ^2{\displaystyle \frac{11}{2}}+{\displaystyle \frac{\mu ^2}{\mathrm{\Lambda }^2}}ln\mathrm{\Lambda }^2{\displaystyle \frac{\mu ^2}{\mathrm{\Lambda }^2}}ln\mu ^2+{\displaystyle \frac{6\mu ^2}{\mathrm{\Lambda }^2}})]`$ (10)
The bad features of this result make it very difficult, perhaps impossible to devise an effective renormalization. These observations form the basis for the work of Brodsky, Hiller and McCartor .
## 3 Infrared Singularities
The example here is the Schwinger model. The Lagrangian for the Schwinger model in light-cone gauge is
$$=i\overline{\psi }\gamma ^\mu _\mu \psi \frac{1}{4}F^{\mu \nu }F_{\mu \nu }A^\mu J_\mu \lambda A^+$$
(11)
Notice that we have used a Lagrange multiplier, $`\lambda `$, to impliment the gauge choice. The degrees of freedom in $`\lambda `$ are essential to the solution and the reason has to do with regulating the $`p^+=0`$ singularity. The operator solution is
$$\mathrm{\Psi }_+=Z_+e^{\mathrm{\Lambda }_+^{()}}\sigma _+e^{\mathrm{\Lambda }_+^{(+)}}$$
(12)
$$\mathrm{\Lambda }_+=i2\sqrt{\pi }(\eta (x^+)+\stackrel{~}{\mathrm{\Sigma }}(x^+,x^{}))$$
(13)
$$Z_+^2=\frac{m^2e^\gamma }{8\pi \kappa }$$
(14)
$$\mathrm{\Psi }_{}=\psi _{}=Z_{}e^{\mathrm{\Lambda }_{}^{()}}\sigma _{}e^{\mathrm{\Lambda }_{}^{(+)}}$$
(15)
$$Z_{}^2=\frac{\kappa e^\gamma }{2\pi }$$
(16)
$$\mathrm{\Lambda }_{}=i2\sqrt{\pi }\varphi (x^+)$$
(17)
$$\lambda =m_+(\eta \varphi )$$
(18)
$$A_+=\frac{2}{m}_+(\eta +\stackrel{~}{\mathrm{\Sigma }})$$
(19)
In these relations
$$\stackrel{~}{\mathrm{\Sigma }}=MASSIVE(m=\frac{e}{\sqrt{\pi }})PSEUDOSCALAR$$
(20)
$$\varphi =CHIRALSCALAR;\eta =CHIRALGHOST$$
(21)
The chiral fields $`\varphi `$ and $`\eta `$ are functions of $`x^+`$; they have been omitted from many previous formulations. Together they make up $`\lambda `$. All the massless fields are regulated by the Kleiber method
$$\varphi ^{(+)}(x^+)=i(4\pi )^{\frac{1}{2}}_0^{\mathrm{}}𝑑k_+k_+^1d(k_+)\left(\mathrm{e}^{ik_+x^+}\theta (\kappa k_+)\right)$$
(22)
The $`x^+`$-dependent fields are necessary to regulate the theory. Let us understand that point. The singularity in the Fermi products must be
$$\mathrm{\Psi }_+^{}(x+ϵ)\mathrm{\Psi }_+(x)\frac{1}{2\pi ϵ^{}}$$
(23)
But for the part independent of the chiral fields we have
$$:e^{i2\sqrt{\pi }\stackrel{~}{\mathrm{\Sigma }}(x+ϵ)}::e^{i2\sqrt{\pi }\stackrel{~}{\mathrm{\Sigma }}(x)}:e^{2\gamma }\frac{4}{m^2}\frac{1}{ϵ^+ϵ^{}}$$
(24)
For the part depending on the chiral fields we have
$$e^{i2\sqrt{\pi }\eta ^{()}(x+ϵ)}\sigma _+^{}e^{i2\sqrt{\pi }\eta ^{(+)}(x+ϵ)}e^{i2\sqrt{\pi }\eta ^{()}(x)}\sigma _+e^{i2\sqrt{\pi }\eta ^{(+)}(x)}e^\gamma \kappa ϵ^+$$
(25)
Thus the chiral fields are essential for regulating the infrared singularity. The chiral fields, necessary for regulation the theory, have other important effects. When the completion necessary to form the full representation space is done, additional states, not in the representation space of free theory, are included. These are translationally invariant and can mix with the vacuum; indeed gauge invariance requires that they mix with the vacuum to form a vacuum of the $`\theta `$-state form. In such a state, $`\overline{\mathrm{\Psi }}\mathrm{\Psi }`$ has a nonzero expectation value.
The chiral fields have no effect on the spectrum as long as the bare mass is zero. But if the bare mass, $`\mu `$, is not zero, the interaction between the chiral fields and the physical fields leads to a new term in $`P^{}`$ which does act in the physical subspace. We find that in the physical subspace the operator is
$$\delta P^{}\mu \sigma _{}^{}\sigma _+Z_{}Z_+\text{:}e^{i\sqrt{\pi }\stackrel{~}{\mathrm{\Sigma }}(0,x^{})}\text{:}dx^{}+C.C.$$
(26)
Where the wavefunction renormalization constants have the small-$`\mu `$ expansions
$$Z_{}=Z_+(\mu )=\sqrt{\frac{\chi e^\gamma }{2\pi }}+𝒪(\mu )$$
(27)
$$Z_+=Z_+(\mu )=\sqrt{\frac{m^2e^\gamma }{8\pi \chi }}+𝒪(\mu )$$
(28)
and, more generally, are determined by the relations
$$\{\mathrm{\Psi }_+(x^+),\mathrm{\Psi }_+(x^++ϵ_+)\}=\delta (ϵ_+)$$
(29)
$$\{\mathrm{\Psi }_{}(x),\mathrm{\Psi }_{}(x+ϵ_{spacelike})\}=\delta (ϵ_{spacelike})$$
(30)
This operator provides a shift in the mass-squared of the physical Schwinger particle of $`2m\mu e^\gamma cos\theta `$, where $`\theta `$ is the vacuum angle. In this sense the Schwinger model provides a model for the $`\eta ^{}`$. It gets a mass from the anomaly which is independent of the bare mass, then has a term in the mass squared which is linear in the bare mass and linear in the chiral condensate.
Once we have identified the new term, we can drop the chiral fields and use DLCQ. If we stay in the continuum we can calculate the renormalization constants as above (at least in principle); in DLCQ we would have to fit them to data.
## 4 Remarks
Regulating the $`p^+=0`$ singularity induces additional operators into the theory. The form of the operators depends on the regulator used. It is not clear to me that all regulators will allow an effective renormalization; certainly some will make renormalization much more difficult than others. All of this is familiar from the equal-time representation.
What is new to the light-cone representation is the unfamiliar form the new operators take and, it seems, the greater difficulty in finding them. I am still hopeful that for the ultraviolet singularities, procedures such as using Pauli-Villars fields may provide a more or less automatic way of including the necessary operators. I do not think that that will prove to be true in the case of infrared singularities, particularly for gauge theories. I believe that operators similar to that of equation (26) will occur in four dimensions and will provide the mechanism for chiral symmetry breaking. I do not believe that the light-quark hadrons can be correctly treated without these operators. I do not yet know of an automatic procedure for finding them, but our knowlege of them is growing.
## Acknowledgments
This work was supported by the U.S. Department of Energy.
## References
|
warning/0004/nlin0004023.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction <br>
In a recent article we determined the Lie point symmetries of a class of equations that we called “generalized Toda field theories”. They all involved various types of exponentials and had the form
$$u_{n,xy}=F_n,F_n=\underset{m=nn_1}{\overset{n+n_2}{}}K_{nm}\mathrm{exp}\left(\underset{l=mn_3}{\overset{m+n_4}{}}H_{ml}u_l\right),$$
(1.1)
where $`K`$ and $`H`$ are some real constant matrices and $`n_1,\mathrm{},n_4`$ are some finite non-negative integers. The symmetries were obtained independently for three different ranges of the variable $`n`$. Namely, we considered the infinite case $`\mathrm{}<n<\mathrm{}`$, the semi-infinite one, $`1n<\mathrm{}`$, and also the finite case, $`1nN<\mathrm{}`$.
In the infinite and semi-infinite cases eq. (1.1) was treated as a differential-difference equation. If the range of $`n`$ is finite, eq. (1.1) is simply a system of $`N`$ differential equations for $`N`$ unknowns $`u_n`$.
A sizable literature exists on symmetries of difference equations -. Different approaches differ in their treatment of independent variables and also in the degree of generalization of the concept of “point symmetries” that is involved in passing from differential equations to difference ones.
In Ref. we adopted the “differential equation” approach proposed earlier . Essentially, eq. (1.1) was treated as a system of infinitely many differential equations for infinitely many fields $`u_n(x,y)`$ with $`\mathrm{}<n<\mathrm{}`$, or $`1n<\mathrm{}`$, respectively.
The purpose of this article is to investigate applications of the obtained Lie point symmetries. In particular we shall show how one can perform various types of symmetry reduction, using these symmetries. The reductions will be from infinite systems to semi-infinite, finite or periodic ones. We will also consider reductions of the number of independent variables, both continuous and discrete. Each time a reduction is performed the resulting equations will inherit some of the symmetries of the original system. We will show how to obtain the “inherited” symmetry group and will compare it with the entire Lie point symmetry group of the reduced system.
The symmetries are point ones, in the sense that we assume that the symmetry algebra, i.e. the Lie algebra of the symmetry group, is realized by vector fields of the form
$$\widehat{v}=\xi (x,y,\{u_k\})_x+\eta (x,y,\{u_k\})_y+\underset{j}{}\varphi _j(x,y,\{u_k\})_{u_j},$$
(1.2)
where $`\{u_k\}`$ denotes the set of all fields (with $`k`$ in an infinite, semi-infinite, or finite range, repectively). The summation in eq. (1.2) is also over the appropriate range of values of $`j`$.
Thus, if eq. (1.1) is considered as an equation on a lattice, the coefficients of the vector field $`\widehat{v}`$ can depend on the values of $`u_k`$ at all points of the lattice.
Whatever the range of the discrete variable $`n`$, the range of the interaction $`F_n`$ in eq. (1.1) is assumed to be finite, i.e. the integers $`n_1,\mathrm{},n_4`$ are finite. Obviously, this allows much more general interactions than nearest neighbour ones. Indeed eq. (1.1) is general enough to include all Toda field theories that, to our knowledge, occur in the literature, be they integrable or not.
Thus, if we put $`H_{n,n1}=H_{nn}=1`$ and $`K_{nn}=K_{nn+1}=1`$, and all the other components of $`H`$ and $`K`$ equal to zero, we obtain the “two-dimensional Toda lattice”
$$u_{n,xy}=e^{u_{n1}u_n}e^{u_nu_{n+1}},\mathrm{}<n<\mathrm{}$$
(1.3)
originally introduced by Mikhailov and Fordy and Gibbons and studied further in Ref. .
Other well known Toda systems (with nearest neighbour interactions) are obtained if we set $`H_{ml}=\delta _{ml}`$
$$u_{n,xy}=\underset{m=nn_1}{\overset{n+n_2}{}}K_{nm}\mathrm{exp}u_m,$$
(1.4)
or vice versa $`K_{nm}=\delta _{nm}`$
$$u_{n,xy}=\mathrm{exp}\underset{l=nn_3}{\overset{n+n_4}{}}H_{nl}u_l.$$
(1.5)
All of these systems have been studied from the point of view of their integrability and solutions -, usually for a finite number of fields and usually for $`K`$, or respectively $`H`$, the Cartan matrix of a simple, or affine, Lie algebra.
We shall call the theories (1.3), (1.4) and (1.5) type I, II and III, respectively. We shall use the $`n\mathrm{}`$ generalization of an $`\text{s}l(n+1,)`$ Cartan matrix, i.e. put $`K_{n1n}=K_{n+1n}=1`$, $`K_{nn}=2`$, $`K_{nm}=0`$ for $`mn,n\pm 1`$ in eq. (1.4) for $`\mathrm{}<n<\mathrm{}`$. The same will be chosen for $`H`$ in eq. (1.5).
We note that the equations (1.4) and (1.5) are equivalent if the range of $`n`$ is finite and the matrices $`H`$ and $`K`$ are invertible. Indeed, if $`u_n(x,y)`$ satisfies eq. (1.4), then $`w=K^1u`$ satisfies eq. (1.5) with $`H=K`$, and vice-versa. For $`\mathrm{}<n<\mathrm{}`$, or $`1n<\mathrm{}`$, this is no longer the case and their symmetry groups, in general, are different.
The original Toda lattice is obtained from eq. (1.3) by symmetry reduction, using translational invariance, i.e. looking for solutions invariant under translations generated by
$$P=_x_y,u_n(x,y)=u_n(t),t=x+y.$$
(1.6)
Since the matrices $`H`$ and $`K`$ are constant, the same reduction takes eq. (1.1) into the generalized Toda lattice
$$\ddot{u}_n=F_n,F_n=\underset{m=nn_1}{\overset{n+n_2}{}}K_{nm}\mathrm{exp}\left(\underset{l=mn_3}{\overset{m+n_4}{}}H_{ml}u_l\right).$$
(1.7)
In Section 2, we study the case where the range of $`n`$ in eq. (1.1) is infinite. We give the symmetry algebra $`L_f`$ of the general Toda field theory (1.1) as calculated in . The same results concerning the particular theories of types I, II and III are also given. Eq. (1.7) inherits a subgroup of the symmetry group of eq. (1.1). The Lie algebra $`L_0`$ of the “inherited symmetry group” will be the normalizer of the vector field $`P=_x_y`$ in $`L_f`$
$$\text{nor}_{L_f}P=\{xL_f|[x,P]=\lambda P\},\lambda .$$
(1.8)
Commuting $`P`$ with a general element of the algebra $`L_f`$, imposing the above normalizer condition and letting the resulting vector field act on functions of $`t`$ and $`u_n`$ we obtain the inherited symmetry algebra of the Toda lattice. We establish that the entire symmetry algebra of the generalized Toda lattice (1.7) coincides with the one inherited from $`L_f`$. From this general result, the symmetry algebras of the Toda lattices corresponding to the types I, II and III are obtained explicitely.
In Section 3, we restrict the range of the discrete variable to be $`1n<\mathrm{}`$. This means that starting from some value $`n_0`$ of $`n`$ the eq. (1.1) and (1.7) will be the same as in the infinite case but for $`1n<n_0`$ they will be modified. Their actual form will depend on the choice of the matrices $`K`$ and $`H`$. Our procedure will be the same in all cases. We start from the already established symmetry algebra in the case of infinitely many fields. Its prolongation will annihilate all the equations in the system that are not modified by the boundary conditions. We apply the prolonged vector field to the equations for $`1n<n_0`$ and require that these equations also be annihilated on the solution set. This will provide us with a subgroup of the original symmetry group which is the symmetry group of the semi-infinite Toda field theory. It must then be checked whether this “inherited” symmetry group is indeed the entire symmetry group of the corresponding semi-infinite system, or only a subgroup of it.
In Section 4, we further restrict the range of the variable $`n`$ and consider Toda field theories and Toda lattices with a finite number of fields, $`1nN`$. We shall proceed as in the semi-infinite case, that is start from the symmetries of Toda field theories, or Toda lattices, with $`\mathrm{}<n<\mathrm{}`$. We then impose that the general symmetry generator should also annihilate the modified equations at the beginning and end of the chain. This is equivalent to starting from a semi-infinite Toda theory and requiring that the symmetry generator should also annihilate the equation for $`n=N`$.
In Section 5, we view the relation between infinite and periodic (generalized) Toda systems as symmetry reduction. Indeed, let us consider the Toda field theory (1.1) for $`n`$ and its symmetry algebra. In order to be able to impose periodicity
$$u_{n+N}(x,y)=u_n(x,y),$$
(1.9)
the symmetry algebra of the generalized Toda field theories or of the generalized Toda lattices, should be enlarged by adding the operator
$$\widehat{N}=_n$$
(1.10)
to the corresponding symmetry algebra, whenever possible. This operator generates shifts (“translations” in $`n`$), though the corresponding translation group parameter must be by an integer, $`\stackrel{~}{n}=n+\lambda ,\lambda `$. Imposing periodicity, as in eq. (1.9), means that $`\widehat{N}`$ must be removed from the symmetry algebra. The symmetry algebra of the periodic system, inherited from the infinite one, will be the normalizer of $`\widehat{N}`$ in the original algebra. The operator $`\widehat{N}`$ of eq. (1.10) can also be used to reduce differential equations on lattices to differential-delay equations .
In section 6, we study a further application of the symmetry group. We look at symmetry reductions involving both discrete and continuous variables.
## 2 Symmetries of infinite generalized Toda field theories and lattices
### 2.1 General results
Let us now consider the differential-difference eq. (1.1) and for $`\mathrm{}<n<\mathrm{}`$ impose that the matrices $`H`$ and $`K`$ are band matrices with finite bands of constant width:
$$H_{nm}=H_{n,n+\sigma }=\{\begin{array}{c}h_\sigma (n)\sigma [p_1,p_2]\hfill \\ 0\sigma \overline{)}[p_1,p_2]\hfill \end{array},h_{p_1}(n)0,h_{p_2}(n)0,p_1p_2.$$
(2.1)
Similarly,
$$K_{nm}=K_{m+\sigma ,m}=\{\begin{array}{c}k_\sigma (m)\sigma [q_1,q_2]\hfill \\ 0\sigma \overline{)}[q_1,q_2]\hfill \end{array},k_{q_1}(m)0,k_{q_2}(m)0,q_1q_2.$$
(2.2)
We introduce the quantity $`\rho _n`$ which, as a function of $`n`$, is any particular solution of the inhomogeneous linear finite-difference equation
$$\underset{\sigma =p_1}{\overset{p_2}{}}h_\sigma (n)\rho _{\sigma +n}=1$$
(2.3)
and the quantities $`\psi _n^j`$, $`j=1,\mathrm{},p_2p_1`$ and $`\varphi _m^l`$, $`l=1,\mathrm{},q_2q_1`$, which are, respectively, linearly independent solutions of the homogeneous linear equations
$$\underset{\sigma =p_1}{\overset{p_2}{}}h_\sigma (n)\psi _{\sigma +n}=0,\underset{\sigma =q_1}{\overset{q_2}{}}k_\sigma \left(m\right)\varphi _{\sigma +m}=0.$$
(2.4)
Without proof we present the following result .
###### Theorem 1
The symmetry algebra of the infinite Toda field theory (1.1) with $`\mathrm{}<n<\mathrm{}`$ and $`H`$ and $`K`$ satisfying (2.1) and (2.2) has a basis consisting of the following vector fields
$$X=\xi (x)_x+\eta (y)_y(\xi _x+\eta _y)\underset{n=\mathrm{}}{\overset{\mathrm{}}{}}\rho _n_{u_n},$$
(2.5a)
$$V_j=(r_j(x)+s_j(y))\underset{n=\mathrm{}}{\overset{\mathrm{}}{}}\psi _n^j_{u_n},$$
(2.5b)
$$Z_{jl}=\left(\underset{m=\mathrm{}}{\overset{\mathrm{}}{}}\varphi _m^lu_m\right)\left(\underset{n=\mathrm{}}{\overset{\mathrm{}}{}}\psi _n^j_{u_n}\right),$$
(2.5c)
$$j=1,\mathrm{},p_2p_1,l=1,\mathrm{},q_2q_1.$$
The functions $`\xi (x),\eta (y),r_j\left(x\right)`$ and $`s_j\left(y\right)`$ are arbitrary and $`\rho _n`$, $`\psi _n^j`$ and $`\varphi _n^l`$ are defined in eq. (2.3) and (2.4), respectively.
The operator $`X`$ reflects the fact that the generalized Toda field theories with $`\mathrm{}<n<\mathrm{}`$ are always conformally invariant, be they integrable or not. If we have $`p_2p_11`$ the theory is invariant under gauge transformations. If we have $`q_2q_11`$ a further type of gauge invariance exists, represented by the operator $`Z_{jl}`$. If $`Z_{jl}`$ is absent the gauge group is abelian, otherwise it is non abelian. The commutation relations are given elsewhere .
The generalized Toda Lattice (1.7) is obtained from the generalized Toda field theory by symmetry reduction. Indeed, let us reduce by the translation generator $`P=_x_y`$ as in eq. (1.6). Eq. (1.1) reduces to eq. (1.7). We calculate the symmetry subgroup of the symmetry group of eq. (1.1) inherited by eq. (1.7) using the procedure explained in the Introduction. We obtain the following result.
###### Theorem 2
The basis for the inherited symmetry algebra for the infinite Toda lattice (1.7) with $`H`$ and $`K`$ satisfying (2.1) and (2.2) is given by
$$\begin{array}{cc}P=_t,& D=t_t2\underset{n=\mathrm{}}{\overset{\mathrm{}}{}}\rho _n_{u_n}\\ & \\ U_j=\underset{n=\mathrm{}}{\overset{\mathrm{}}{}}\psi _{n}^{}{}_{}{}^{j}_{u_n},& W_j=t\underset{n=\mathrm{}}{\overset{\mathrm{}}{}}\psi _{n}^{}{}_{}{}^{j}_{u_n}\end{array}$$
(2.6)
and $`Z_{jl}`$ as in eq. (2.5c). The inherited algebra coincides with the actual symmetry algebra of eq. (1.7).
Just as in the case of differential equations, there is no guarantee that the symmetry algebra inherited from the original equation is the entire symmetry algebra of the reduced equation. However, in this case, a direct calculation of the symmetry algebra of the generalized Toda lattice shows that its symmetry algebra is indeed given by eq. (2.6).
### 2.2 Special cases
A. The type I system
We have $`p_2p_1=q_2q_1=1`$, and obtain
$$\psi _m=\varphi _m=1,\rho _n=n$$
(2.7)
in eqs. (1) and (2.6). Thus, (2.5a) is present, as are (2.5b) and (2.5c). The labels $`j=l=1`$ can be dropped.
B. The type II system with $`K_{n1n}=K_{n+1n}=1`$, $`K_{nn}=2`$
We have $`p_2=p_1=0`$, $`q_2q_1=1`$ and hence
$$\psi _m=0,\varphi _m=an+b,\rho _n=1,$$
(2.8)
with $`a`$ and $`b`$ constant. Conformal invariance (2.5a) is present but there are no gauge transformations (2.5b), nor (2.5c).
C. The type III system with $`H_{n1n}=H_{n+1n}=1`$, $`H_{nn}=2`$
We have $`p_2p_1=2`$, $`q_2q_1=0`$ and
$$\psi _n^1=1,\psi _n^2=n,\varphi _m=0,\rho _n=\frac{1}{2}n^2.$$
(2.9)
We have conformal invariance (2.5a), two operators $`V_1`$ and $`V_2`$, no $`Z`$.
The symmetries of the corresponding generalized Toda lattices are obtained by putting the above values of $`\rho _n`$, $`\psi _m`$ and $`\varphi _m`$ into eq. (2.6).
## 3 Reduction to semi-infinite theories
We will now calculate the symmetry groups of semi-infinite theories inherited from infinite Toda systems using the procedure explained in the Introduction.
Rather than impose general and arbitrary boundary conditions on the matrices $`K`$ and $`H`$ of eq. (1.1) we shall consider several special cases suggested by Lie group theory, that already occured in the literature (usually for $`n`$ varying in a finite range).
### 3.1 The semi-infinite field theories of type I related to simple Lie algebras
Let us consider a semi-infinite generalization of the system (1.3). The field equation is given by
$$𝐔_{xy}=\mu ^2\underset{i=1}{\overset{N}{}}\frac{𝜶_i}{𝜶_i^2}\mathrm{exp}(𝜶_i𝐔),$$
(3.1)
where $`𝐔=(u_1,\mathrm{},u_N)`$ is an $`N`$-tuple of real fields and $`(𝜶_1,\mathrm{},𝜶_N)`$ denote the simple roots of some finite simple Lie algebra of rank $`N`$. If this Lie algebra is $`\text{s}l(N+1,)`$ one obtains the usual finite Toda field theory. Instead of this we shall consider the Toda field theory (1.3) and let the Lie algebra root system run through semi-infinite extensions of the classical Cartan Lie algebras $`A_N`$, $`B_N`$, $`C_N`$ and $`D_N`$ with $`N\mathrm{}`$ (in one direction).
For the system (1.3) the symmetry algebra (1) reduces to
$`X=\xi (x)_x+\eta (y)_y+(\xi _x+\eta _y){\displaystyle \underset{n=\mathrm{}}{\overset{\mathrm{}}{}}}n_{u_n},`$ (3.2a)
$`V=(\beta (x)+\gamma (y)){\displaystyle \underset{n=\mathrm{}}{\overset{\mathrm{}}{}}}_{u_n},`$ (3.2b)
$`Z=\left({\displaystyle \underset{m=\mathrm{}}{\overset{\mathrm{}}{}}}u_m\right)\left({\displaystyle \underset{n=\mathrm{}}{\overset{\mathrm{}}{}}}_{u_n}\right).`$ (3.2c)
For the ordinary Toda lattice theory obtained by replacing $`u_{n,xy}`$ by $`u_{n,tt}`$ in eq. (1.3), the symmetry algebra (2.6) reduces to
$`P=_t,D=t_t+2{\displaystyle \underset{n=\mathrm{}}{\overset{\mathrm{}}{}}}n_{u_n},`$ (3.3a)
$`U={\displaystyle \underset{n=\mathrm{}}{\overset{\mathrm{}}{}}}_{u_n},W=t{\displaystyle \underset{n=\mathrm{}}{\overset{\mathrm{}}{}}}_{u_n},`$ (3.3b)
and $`Z`$ as in (3.2c).
We now turn to the semi-infinite case. For each algebra we show the modified equations (for $`n<n_0`$). Those not shown coincide with eq. (1.3).
1. The Cartan $`A`$ series
We have
$$u_{1,xy}=e^{u_1u_2}.$$
(3.4)
Taking a general element of the algebra (3.1), applying its prolongation to eq. (3.4) and taking into account that
$$\left(\underset{i=1}{\overset{\mathrm{}}{}}u_i\right)_{xy}=0$$
(3.5)
we find that the entire symmetry algebra (3.1) leaves eq. (3.4) invariant. Hence the symmetry algebras in the semi-infinite and infinite cases coincide, though in the semi-infinite case all summations are for $`1n<\mathrm{}`$. Eq. (3.5) was used to show that the $`Z`$ symmetry also survives.
The same is true for the Toda lattice equations in this case. Namely, the symmetry groups are the same in the infinite and semi-infinite cases.
2. The Cartan $`B`$ series
In this case, the first equation is
$$u_{1,xy}=e^{u_1}e^{u_1u_2}.$$
(3.6)
Applying the same procedure, we find that conformal invariance remains and is realized as in eq. (3.2a). The gauge symmetries (3.2b) and (3.2c) do not survive.
For the Toda lattice, the only surviving symmetries are the translation $`P`$ and dilation $`D`$ as in eq. (3.3a).
3. The Cartan $`C`$ series
We have
$$u_{1,xy}=e^{u_1u_2}+2e^{2u_1}.$$
(3.7)
In this case the gauge symmetries combine together with the conformal ones. The presence of eq. (3.7) excludes invariance under the $`Z`$ transformation of eq. (3.2c).
The entire set of equations is invariant under the transformations generated by
$$X=\xi (x)_x+\eta (y)_y+(\xi _x+\eta _y)\underset{n=1}{\overset{\mathrm{}}{}}\left(n\frac{1}{2}\right)_{u_n},$$
(3.8)
The surviving symmetries of the corresponding Toda lattice are $`P`$ and $`DU`$ (see eq. (3.1)).
4. The Cartan $`D`$ series
In this case, the first two equations must be distinguished. They are
$`u_{1,xy}=e^{u_1u_2}e^{u_1u_2},`$ (3.9a)
$`u_{2,xy}=e^{u_1u_2}+e^{u_1u_2}e^{u_2u_3}.`$ (3.9b)
The surviving algebra is given by
$$X=\xi (x)_x+\eta (y)_y+(\xi _x+\eta _y)\underset{n=1}{\overset{\mathrm{}}{}}(n1)_{u_n}.$$
(3.10)
Reducing further to the corresponding Toda lattice, we find that it is invariant under translations and dilations generated by $`P`$ and $`D2U`$ with $`P`$, $`D`$ and $`U`$ as in eq. (3.1).
Without presenting the proof we state that the above symmetries, inherited from the infinite case, represent the entire symmetry algebra in the semi-infinite case.
### 3.2 The semi-infinite Toda field theories of type II related to simple Lie algebras
The infinite system is given by eq. (1.1) with $`H_{nm}=\delta _{nm}`$ and $`K`$ a Cartan matrix. In other words the infinite system is
$$u_{n,xy}=e^{u_{n1}}+2e^{u_n}e^{u_{n+1}}$$
(3.11)
and the entire symmetry algebra is generated by
$$X=\xi (x)_x+\eta (y)_y(\xi _x+\eta _y)\underset{n=\mathrm{}}{\overset{\mathrm{}}{}}_{u_n}.$$
(3.12)
The gauge transformations discussed in Section 2 are all absent since we have $`p_2=p_1`$. It was also shown in Section 2 that the existence of conformal invariance depends only on the matrix $`H`$ which in the field theory (1.4) is an (infinite) identity matrix. The $`A`$, $`B`$, $`C`$ and $`D`$ Cartan series have different matrices $`K`$ but this has no influence on conformal invariance. Hence the infinite, semi-infinite and finite theories all have the same symmetry algebra (3.12). Moreover the same symmetries will exist even if $`K`$ is not a Cartan matrix.
The reduction to the corresponding Toda lattice yields the symmetries
$$P=_t,D=t_t2\underset{n=1}{\overset{\mathrm{}}{}}_{u_n}$$
(3.13)
in agreement with eq. (2.6).
### 3.3 The semi-infinite field theories of type III related to simple Lie algebras
We are now restricting the matrix $`K`$ to satisfy $`K_{nm}=\delta _{nm}`$ and $`H`$ to be a Cartan matrix. Thus, in the infinite case, the equations we are studying are
$$u_{n,xy}=e^{u_{n1}+2u_nu_{n+1}}.$$
(3.14)
The symmetry algebra in this case is generated by
$`X=\xi (x)_x+\eta (y)_y+{\displaystyle \frac{1}{2}}(\xi _x+\eta _y){\displaystyle \underset{n=\mathrm{}}{\overset{\mathrm{}}{}}}n^2_{u_n},`$ (3.15a)
$`V={\displaystyle \underset{n=\mathrm{}}{\overset{\mathrm{}}{}}}\left(r_1(x)+s_1(y)+n(r_2(x)+s_2(y))\right)_{u_n},`$ (3.15b)
where $`\xi `$, $`\eta `$, $`r_1`$, $`r_2`$, $`s_1`$, and $`s_2`$ are arbitrary functions.
In the case of the corresponding infinite lattice the symmetry algebra is generated by
$`P=_t,`$ $`D=t_t+{\displaystyle \underset{n=\mathrm{}}{\overset{\mathrm{}}{}}}n^2_{u_n},`$ (3.16a)
$`U_1={\displaystyle \underset{n=\mathrm{}}{\overset{\mathrm{}}{}}}_{u_n},`$ $`W_1=t{\displaystyle \underset{n=\mathrm{}}{\overset{\mathrm{}}{}}}_{u_n}`$ (3.16b)
$`U_2={\displaystyle \underset{n=\mathrm{}}{\overset{\mathrm{}}{}}}n_{u_n},`$ $`W_2=t{\displaystyle \underset{n=\mathrm{}}{\overset{\mathrm{}}{}}}n_{u_n}.`$ (3.16c)
Now let us look at individual semi-infinite cases.
1. The Cartan $`A`$ series
The first equation is
$$u_{1,xy}=e^{2u_1u_2}.$$
(3.17)
The others are as in (3.14). Requiring that eq. (3.17) also be annihilited by the symmetry algebra, we obtain the constraint $`r_1+s_1=0`$. Thus the surviving symmetry algebra is the same as in eq. (3.3) but with $`r_1=s_1=0`$.
For the $`A`$ Toda lattice, only $`P`$, $`D`$, $`U_2`$ and $`W_2`$ in (3.3) survive.
2. The Cartan $`B`$ series
In this case the two first equations are modified. They are
$$\begin{array}{c}u_{1,xy}=e^{2u_1u_2},\\ u_{2,xy}=e^{2u_1+2u_2u_3}.\end{array}$$
(3.18)
In this case only conformal invariance survives the reduction and takes the form
$$X=\xi (x)_x+\eta (y)_y+\frac{1}{2}(\xi _x+\eta _y)\underset{n=1}{\overset{\mathrm{}}{}}n(n1)_{u_n}.$$
(3.19)
However a direct calculation of the symmetry algebra shows that there is a further gauge symmetry given by
$$V=(r(x)+s(y))\underset{n=1}{\overset{\mathrm{}}{}}a_n_{u_n},a_1=1,a_k=2\text{f}ork2.$$
(3.20)
Similarly for the Toda lattice in this case the only inherited symmetries from (3.3) are $`P`$ and $`DU_2`$. From a direct calculation we obtain an additional gauge symmetry $`W=(ct+d)_{n=1}^{\mathrm{}}a_n_{u_n}`$ with $`a_n`$ as in (3.20) and $`c`$ and $`d`$ constants.
3. The Cartan $`C`$ series
Only the first equation is modified and is
$$u_{1,xy}=e^{2u_12u_2}.$$
(3.21)
The only surviving symmetries from eq. (3.3) are given by
$$X=\xi (x)_x+\eta (y)_y+\frac{1}{2}(\xi _x+\eta _y)\underset{n=1}{\overset{\mathrm{}}{}}n(n2)_{u_n}$$
(3.22)
and gauge transformations generated by vector fields of the form (3.15b) with $`r_2=s_2=0`$.
Similarly for the Toda lattice in this case the symmetries inherited from (3.3) are
$$\begin{array}{cc}P=_t,& D=t_t+\underset{n=\mathrm{}}{\overset{\mathrm{}}{}}n(n2)_{u_n}\\ & \\ U_1=\underset{n=\mathrm{}}{\overset{\mathrm{}}{}}_{u_n},& W_1=t\underset{n=\mathrm{}}{\overset{\mathrm{}}{}}_{u_n}\end{array}$$
(3.23)
2. The Cartan $`D`$ series
In this case the two first equations are modified. They are
$$\begin{array}{c}u_{1,xy}=e^{2u_1u_3},\\ u_{2,xy}=e^{2u_2u_3}.\end{array}$$
(3.24)
In this case only conformal invariance survives the reduction and takes the form
$$X=\xi (x)_x+\eta (y)_y+\frac{1}{2}(\xi _x+\eta _y)\underset{n=1}{\overset{\mathrm{}}{}}(n^23n+2)_{u_n}.$$
(3.25)
However a direct calculation of the symmetry algebra shows that there is a further gauge symmetry given by
$$V=(r(x)+s(y))\underset{n=1}{\overset{\mathrm{}}{}}a_n_{u_n},a_1=a_2=1,a_k=2\text{f}ork3.$$
(3.26)
Similarly for the Toda lattice in this case the only inherited symmetries from (3.3) are $`P`$ and $`D3U_2+2U_1`$. From a direct calculation we obtain an additional gauge symmetry $`W=(ct+d)_{n=1}^{\mathrm{}}a_n_{u_n}`$ with $`a_n`$ as in (3.26) and $`c`$ and $`d`$ constants.
## 4 Reduction to finite Toda systems
We will now calculate subgroups of symmetry groups of infinite Toda systems inherited by finite ones. The reduction procedure is explained in the Introduction. It must then be verified that the inherited groups indeed correspond to the complete symmetry groups of the finite systems.
### 4.1 Finite Toda theories of type I
The symmetries of the semi-infinite Toda field theories and Toda lattices corresponding to $`A`$, $`B`$, $`C`$ and $`D`$ Cartan series were established in Section 3.1 above.
The corresponding finite systems are obtained by setting $`u_k=0`$, $`kN+1`$ and imposing the equation
$$u_{N,xy}=e^{u_{N1}u_N},\text{o}ru_{N,tt}=e^{u_{N1}u_N}$$
(4.1)
respectively. It is easy to check that eq. (4.1) is invariant under the entire algebra (3.1), or (3.1), respectively (with all summations in the range $`1nN`$).
We obtain the following result.
###### Theorem 3
The symmetries of the finite $`A_N`$, $`B_N`$, $`C_N`$ and $`D_N`$ Toda field theories of type (1.3) are all inherited from the symmetries (3.1) of the infinite theories and coincide with those of the corresponding semi-infinite ones. The same is true for the $`A_N`$, $`B_N`$, $`C_N`$ and $`D_N`$ Toda lattices.
### 4.2 Finite Toda theories of type II
The equations under consideration have the form (3.11). However we modify the matrix $`K`$ in eq. (1.4), the symmetries remain the same, namely conformal invariance (3.12) for Toda field theories and translations $`P=_t`$ and dilations $`D=t_t2_{n=1}^N_{u_n}`$ for Toda lattices. Thus, the symmetries are the same in the infinite case and in all semi-infinite and finite cases.
### 4.3 Finite Toda theories of type III
The finite case is obtained from the semi-infinite one of Section 3.3 by setting $`u_n=0`$, $`nN+1`$ and imposing
$$u_{N,xy}=e^{u_{N1}+2u_N},\text{o}ru_{N,tt}=e^{u_{N1}+2u_N},$$
(4.2)
respectively. The requirement that eq. (4.2) be invariant will restrict the symmetry algebras of Section 3.3.
Let us consider individual cases.
1. The Cartan series $`A_N`$
In the $`A_{1/2\mathrm{}}`$ case we have the prolonged symmetry vector
$$\text{pr}\widehat{v}=\xi (x)_x+\eta (y)_y+\left[\frac{1}{2}(\xi _x+\eta _y)n^2+(r(x)+s(y))n\right]_{u_n}(\xi _x+\eta _y)u_{n,xy}_{u_{n,xy}}.$$
This will annihilate the eq. (4.2) only if
$$r(x)+s(y)=\frac{N+1}{2}(\xi _x+\eta _y).$$
(4.3)
Thus, the inherited symmetry algebra of the finite $`A_N`$ Toda field theory consists of the conformal transformation generated by
$$X=\xi (x)_x+\eta (y)_y+\frac{1}{2}(\xi _x+\eta _y)\underset{n=1}{\overset{N}{}}n(nN1)_{u_n}.$$
(4.4)
Reducing to the $`A_N`$ Toda lattice we have
$$P=_t,D=t_t+\underset{n=1}{\overset{N}{}}n(nN1)_{u_n}$$
(4.5)
2. The Cartan series $`B_N`$
We require that the algebra (3.19), (3.20) should annihilate eq. (4.2) (on its solution set). This requires $`r+s=\frac{1}{4}N(N+1)(\xi _x+\eta _y)`$ and restricts the symmetry algebra to
$$X=\xi (x)_x+\eta (y)_y+\frac{1}{4}(\xi _x+\eta _y)\left[N(N+1)_{u_1}+2\underset{n=2}{\overset{N}{}}[n(n1)N(N+1)]_{u_n}\right].$$
(4.6)
For the Toda lattice this reduces further to
$$P=_t,D=t_t+\frac{1}{4}\left[N(N+1)_{u_1}+2\underset{n=2}{\overset{N}{}}[n(n1)N(N+1)]_{u_n}\right].$$
(4.7)
3. The Cartan series $`C_N`$
Starting from the symmetries (3.22) and the remaining gauge transformations, we find that the symmetries of the $`C_N`$ Toda field theories reduce to conformal transformations, realized as
$$X=\xi (x)_x+\eta (y)_y+\frac{1}{2}(\xi _x+\eta _y)\underset{n=1}{\overset{N}{}}[n(n2)N^2+1)]_{u_n}.$$
(4.8)
For the $`C_N`$ Toda lattice this reduces to
$$P=_t,D=t_t+\underset{n=1}{\overset{N}{}}[n(n2)N^2+1]_{u_n}.$$
(4.9)
4. The Cartan series $`D_N`$
Starting from the symmetries (3.25) and (3.26) we find that eq. (4.2) reduces the symmetries to
$`X=`$ $`\xi (x)_x+\eta (y)_y+{\displaystyle \frac{1}{4}}(\xi _x+\eta _y)[N(N1)(_{u_1}+_{u_2})`$
$`+2{\displaystyle \underset{n=3}{\overset{N}{}}}[(n2)(n1)N(N1)]_{u_n}]`$
and respectively
$`P=_t,`$ $`D=t_t+{\displaystyle \frac{1}{2}}[N(N1)(_{u_1}+_{u_2})`$
$`+2{\displaystyle \underset{n=3}{\overset{N}{}}}[(n2)(n1)N(N1)]_{u_n}].`$
The result is finally quite simple, namely:
###### Theorem 4
Toda field theories of the type (1.5) for a finite number $`N`$ of fields, based on the Cartan algebras $`A_N`$, $`B_N`$, $`C_N`$ and $`D_N`$ are invariant under conformal transformations only. They are realized by the vector fields (4.4), (4.6), (4.8) and (LABEL:4.10) for $`A_N`$, $`B_N`$, $`C_N`$ and $`D_N`$, respectively. The symmetries are inherited from the semi-infinite case (however, for the $`B`$ and $`D`$ series the symmetries in the semi-infinite are not all inherited from the infinite case). The finite Toda lattices of type (1.5) are invariant only under a translation, and the appropriate dilations.
## 5 Reduction to periodic Toda systems
We will now study reductions from infinite (generalized) Toda systems to periodic ones using the procedure explained in the Introduction. In order to be able to impose periodicity (1.9) and enlarge the symmetry algebra of the infinite system by adding the vector field (1.10), a further condition is necessary, namely that the recursion relations (2.3) and (2.4) should have constant coefficients:
$$h_\sigma (n)=h_\sigma (n+1),k_\sigma (m)=k_\sigma (m+1).$$
(5.1)
Let us look at individual cases.
### 5.1 Periodic Toda systems of type I
The inherited symmetry algebra of the usual periodic Toda field theory (1.3) is given by
$`L=x_xy_y,P_1=_x,P_2=_y,`$
$`V=[\beta (x)+\gamma (y)]{\displaystyle \underset{n=1}{\overset{N1}{}}}u_n,Z={\displaystyle \underset{m=1}{\overset{N1}{}}}u_m{\displaystyle \underset{n=1}{\overset{N1}{}}}_{u_n}.`$
We see that the infinite dimensional conformal algebra is reduced to the Poincaré one. For the Toda lattice the symmetry algebra (3.1) in the periodic case is reduced to $`\{P,U,W,Z\}`$.
The periodic Toda field theory (1.3) can be written as eq. (1.1) with
$$K=\left(\begin{array}{ccccc}\hfill 1& \hfill 1& \hfill 0& \hfill \mathrm{}& \hfill 0\\ \hfill 0& \hfill 1& \hfill 1& \hfill \mathrm{}& \hfill 0\\ \hfill \mathrm{}& & \hfill \mathrm{}& \hfill \mathrm{}& \\ \hfill 0& \hfill 0& \hfill \mathrm{}& \hfill 1& \hfill 1\\ \hfill 1& \hfill 0& \hfill \mathrm{}& \hfill 0& \hfill 1\end{array}\right),H=\left(\begin{array}{ccccc}\hfill 1& \hfill 0& \hfill 0& \hfill \mathrm{}& \hfill 1\\ \hfill 1& \hfill 1& \hfill 0& \hfill \mathrm{}& \hfill 0\\ \hfill 0& \hfill 1& \hfill 1& \hfill \mathrm{}& \hfill 0\\ \hfill \mathrm{}& & \hfill \mathrm{}& \hfill \mathrm{}& \\ \hfill 0& \hfill 0& \hfill \mathrm{}& \hfill 1& \hfill 1\end{array}\right).$$
(5.3)
The vector $`\overline{\mathrm{𝟏}}_N=(1,1,\mathrm{},1)`$ is not in the image of $`H`$, the kernel of $`H`$ and $`K^T`$ are one-dimensional. From Theorem 2 of Ref. we conclude that the periodic Toda field theories and lattices of type (1.3) have no further symmetries: all symmetries are inherited.
### 5.2 Periodic Toda systems of type II
All elements of the symmetry algebra (3.12) commute with $`\widehat{N}`$ of eq. (1.10). Hence, in this case the symmetry algebra is the same in the periodic case as in the infinite one (and also the semi-infinite and all finite ones). Thus, the corresponding Toda field theory is conformally invariant, the Toda lattice is invariant under translations $`P`$ and dilations $`D`$ as in eq. (3.13). No new symmetries, due to the reduction, arise.
### 5.3 Periodic Toda systems of type III
The inherited symmetries from eq. (3.3) in this case are Poincaré and gauge invariance:
$`L=x_xy_y,P_1=_x,P_2=_y,`$
$`V=[\beta (x)+\gamma (y)]{\displaystyle \underset{n=1}{\overset{N1}{}}}u_n.`$
For the corresponding periodic Toda lattice the only inherited symmetries are
$$P=_t,U=_{u_n},W=t_{u_n}.$$
(5.5)
These are the only symmetries of the system.
## 6 Symmetry reductions involving continuous and discrete variables
In this Section, we will extend the symmetry algebras of the infinite Toda lattices by the generator $`\widehat{N}`$ given in eq. (1.10). It generates transformations of the independent variable $`n`$ given by
$$n^{}=n+N.$$
(6.1)
The quantity $`N`$ is to be viewed as a discrete group parameter. We can then act with $`\widehat{N}`$ as if it were an element of the symmetry algebra and calculate group transformations and invariants in exactly the same manner as for differential equations.
### 6.1 The infinite Toda lattice of type I
The symmetry algebra in this case is given by eq. (3.1). All the possible reductions have been studied in . Let us look at the interesting case of the reduction by the generator
$$_n+a_t,$$
(6.2)
where $`a`$ is a constant . The invariants are $`\xi =ant`$ and $`u_n`$. We thus consider solutions of the infinite Toda lattice of the form
$$u_n(t)=F(\xi ),\xi =ant.$$
(6.3)
Substituing into (1.3) we find the following equation for $`F`$
$$F_{\xi \xi }=\text{e}^{F(\xi a)F(\xi )}\text{e}^{F(\xi )F(\xi +a)}.$$
(6.4)
A soliton solution of (6.4) is given by
$$F(\xi )=\text{ln}\left(\frac{1+\mathrm{exp}[2\xi \mathrm{sinh}\frac{\alpha }{2}\pm \alpha ]}{1+\mathrm{exp}[2\xi \mathrm{sinh}\frac{\alpha }{2}]}\right),a=\pm \frac{\alpha }{2\mathrm{sinh}\frac{\alpha }{2}}.$$
(6.5)
### 6.2 The infinite Toda lattice of type II
The symmetry algebra in this case is given by eq. (3.13). We first consider reductions by the generator (6.2). Invariant solutions then have the form given by eq. (6.3) and the function $`F`$ satisfies
$$F_{\xi \xi }=\text{e}^{F(\xi a)}+2\text{e}^{F(\xi )}\text{e}^{F(\xi +a)}.$$
(6.6)
We can also consider reductions by the operator $`\widehat{N}+aD`$ with $`D`$ given in eq. (3.13). Invariant solutions are then of the form
$$u_n(t)=F(\eta )2an,\eta =t\text{e}^{an}.$$
(6.7)
The function $`F`$ satisfies the equation
$$F_{\eta \eta }=\text{e}^{F(\text{e}^a\eta )}+2\text{e}^{F(\eta )}\text{e}^{F(\text{e}^a\eta )}.$$
(6.8)
### 6.3 The infinite Toda lattice of type III
The symmetry algebra in this case is given by eq. (3.3). We first consider reductions by generators of the form
$$\widehat{N}+aP+cU_1+dU_2+eW_1+fW_2,$$
(6.9)
where $`P`$, $`U_1`$, $`U_2`$, $`W_1`$ and $`W_2`$ are given in (3.3). Invariant solutions will have the form
$$u_n(t)=F(\xi )+(c\xi e)n+\frac{n^2}{2}(D+ae\xi f)+af\frac{n^3}{3},$$
(6.10)
where $`\xi =ant`$. The function $`F`$ then satisfies the equation
$$F_{\xi \xi }=\mathrm{exp}[F(\xi +a)+2F(\xi )F(\xi a)+aed+f\xi ].$$
(6.11)
In the case when $`d=ae`$ and $`f=0`$, we have a solution quadratic in $`\xi `$
$$F=\alpha \xi ^2+\beta \xi +\gamma ,$$
(6.12)
where $`\alpha `$ is determined in terms of $`a`$ by the equation
$$2\alpha =\text{e}^{2\alpha a^2}.$$
(6.13)
The constants $`\beta `$ and $`\gamma `$ are free.
We also consider reductions by generators of the form
$$\widehat{N}+bD+cU_1+dU_2+eW_1+fW_2$$
(6.14)
with $`b`$ nonzero. In this case invariant solutions have the form
$$u_n(t)=b\frac{n^3}{3}+d\frac{n^2}{2}+cn+\frac{\text{e}^{bn}}{b^2}\eta (be+bff)+F(\eta ),$$
(6.15)
where $`\eta =t\text{e}^{bn}`$. The function $`F`$ must then satisfy the equation
$$F_{\eta \eta }=\mathrm{exp}(dF(\text{e}^b\eta )+2F(\eta )F(\text{e}^b\eta )).$$
(6.16)
## 7 Conclusions
The starting point of this article are the symmetries (1) of a large class of classical field theories with exponential interactions described by eq. (1.1). The symmetries (1), established in our earlier article are present when the discrete variable $`n`$, labeling the fields, varies in an infinite range $`\mathrm{}<n<\mathrm{}`$. The “interaction matrices” $`K`$ and $`H`$ in eq. (1.1) are very general band matrices with bands of constant width (see eq. (2.1) and (2.2)). From eq. (2.5a) we see that the theory is always conformally invariant. In view of eq. (2.5b), the theory is gauge invariant, unless the matrix $`H`$ is diagonal. If $`K`$ is diagonal, the gauge group is abelian. If both $`H`$ and $`K`$ are non diagonal, the gauge group is nonabelian.
If the generalized Toda field theories (1.1) correspond to difference equations with constant coefficients, then the symmetry group includes a further transformation, namely translations of $`n`$
$$\stackrel{~}{n}=n+N,N,$$
(7.1)
formally generated by the operator (1.10). This occurs if the band matrices $`H`$ and $`K`$, specified in eq. (2.1) and (2.2), satisfy condition (5.1), that is, if the corresponding difference equations have constant ($`n`$-independent) coefficients.
The symmetry group of the infinite ($`\mathrm{}<n<\mathrm{}`$) generalized Toda field theory is applied to study different types of reductions of the system.
It should be emphasized that when we perform symmetry reduction, for differential equations, difference equations, or differential-difference ones, we obtain a symmetry algebra inherited from the original symmetry algebra. This may not be the entire symmetry algebra of the reduced system. A prime example for this is provided by the Laplace equation in three-dimensional euclidian space $`E_3`$. It is invariant under the conformal group O(4,1). If we reduce it by translational invariance to the Laplace equation in $`E_2`$, we obtain the inherited symmetry group O(3,1). However, the reduced equation is invariant under a much larger group, the infinite-dimensional conformal group $`E_2`$. The presence of this noninherited symmetry group is an indication of the fact that two-dimensional theories differ qualitatively from three-dimensional ones. We have seen above that no such radical difference between “inherited symmetries” and “all symmetries” occurs in reductions of Toda systems.
Among open problems, not addressed in this article, we mention two related ones. They are: under what conditions on the matrices $`H`$ and $`K`$ are the generalized Toda systems (1.1) and (1.7) integrable? When do these equations allow higher symmetries that depend on the derivatives of $`u_n`$, or that are nonlocal?
## Acknowledgments
The research reported in this article is part of a project supported by NATO grant CRG 960717. P.W. and S.L. acknowledges support from NSERC of Canada, and FCAR du Québec. They also thank the Dipartimento di Fisica, Università di Lecce, for its hospitality.
|
warning/0004/hep-th0004026.html
|
ar5iv
|
text
|
# Light-front vacuum and instantons in two dimensions
## I INTRODUCTION
Non-abelian quantum gauge theories are still far from being satisfactorily understood. Though some non-perturbative features are thought to be transparent, a consistent framework in the continuum is lacking.
Therefore one often resorts to the simplified context of two-dimensional theories where exact solutions can sometimes be available. In two dimensions the theory looks seemingly trivial when quantized in the light-cone gauge (LCG) $`A_{}\frac{A_0A_1}{\sqrt{2}}=0.`$ As a matter of fact, in the absence of dynamical fermions, no physical local degrees of freedom appear in the Lagrangian.
Still topological degrees of freedom occur if the theory is put on a (partially or totally) compact manifold, whereas the simpler behavior on the plane enforced by the LCG condition entails a severe worsening in its infrared structure. These features are related aspects of the same basic issue: even in two dimensions ($`D=2`$) the theory contains some non-trivial dynamics. We can say that, in LCG, dynamics gets hidden in the very singular nature of correlators at large distances (IR singularities).
In order to fully appreciate this point and the controversial aspects related to it, let us briefly review the ’t Hooft’s model for $`QCD_2`$ at large $`N`$, $`N`$ being the number of colours . In LCG no self-interaction occurs for the gauge fields; in the large-$`N`$ limit planar diagrams dominate, without quark loops. The $`q\overline{q}`$ interaction is mediated by the exchange
$$𝒟(x)=\frac{i}{2}|x^{}|\delta (x^+),$$
(1)
which looks instantaneous if $`x^+`$ is considered as a time variable. Eq.(1) is the Fourier transform of the quantity
$$\stackrel{~}{𝒟}(k)=\frac{1}{k_{}^2},$$
(2)
the singularity at $`k_{}=0`$ being interpreted as a Cauchy principal value. Such an expression in turn can be derived by quantizing the theory on the light front (at equal $`x^+`$), $`A_+`$ behaving as a constraint .
The full set of ladder diagrams can easily be summed, leading to a beautiful pattern of $`q\overline{q}`$-bound states with squared masses lying on rising Regge trajectories. This was the first evidence, to our knowledge, of a stringy nature of $`QCD`$ in its confining regime, reconciling dual models with a partonic field theory.
Still, if the theory within the same gauge choice is canonically quantized at equal times, a different expression is obtained for the exchange in eq.(1)
$$𝒟_c(x)=\frac{1}{2\pi }\frac{x^{}}{x^++iϵx^{}},$$
(3)
and its Fourier transform
$$\stackrel{~}{𝒟}_c(k)=\frac{1}{(k_{}+iϵk_+)^2},$$
(4)
can now be interpreted as a causal Feynman propagator .
This expression, first proposed by Wu , is nothing but the restriction at $`D=2`$ of the prescription for the LCG vector propagator in four dimensions suggested by Mandelstam and Leibbrandt (ML), and derived in ref. by equal-time canonical quantization of the theory.
In dimensions higher than two, where “physical” degrees of freedom are switched on (transverse “gluons”), this causal prescription is mandatory in order to get correct analyticity properties, which in turn are the basis of any consistent renormalization program .
When eq.(4) is used in summing the very same set of planar diagrams considered by ’t Hooft, no rising Regge trajectories are found in the spectrum of the $`q\overline{q}`$-system. The bound-state integral equation looks difficult to be solved; early approximate treatments as well as a more detailed recent study indicate the presence of a massless solution, with a fairly obscure interpretation, at least in this context. Confinement seems lost.
Then, how can it be that the causal way to treat the infrared (IR) singularities, which is mandatory in higher dimensions, leads to a disastrous result when adopted at $`D=2`$ ? In order to get an answer we address ourselves to the $`q\overline{q}`$-potential.
## II THE WILSON LOOP
A very convenient gauge invariant way of looking at the $`q\overline{q}`$-potential is to consider a rectangular Wilson loop, centered at the origin, with sides parallel to a spatial direction and to the time direction, of length $`2L`$ and $`2T`$ respectively
$$𝒲=\frac{1}{N}0|\mathrm{Tr}\left[𝒯𝒫\mathrm{exp}\left(ig_\gamma 𝑑x^\mu A_\mu (x)\right)\right]|0,$$
(5)
the symbols $`𝒯`$ and $`𝒫`$ denoting temporal ordering of operators and colour ordering.
It is well known that the Wilson loop we have hitherto introduced can be thought to describe the interaction of a couple of static $`q\overline{q}`$ at the distance $`2L`$ from each other. We can turn to the Euclidean formulation replacing $`T`$ with $`iT`$. If we denote by $`_0(L)`$ the ground state energy of the system, we get for large $`T`$
$$𝒲=\mathrm{exp}[(4m2_0)T]__0^{\mathrm{}}𝑑\rho (L,)\mathrm{exp}[2T(_0)].$$
(6)
Unitarity requires the spectral density $`\rho (L,)`$ to be a non-negative measure. Then $`𝒲`$ is positive and the coefficient of the exponential factor $`\mathrm{exp}[(4m2_0)T]`$ is a non-increasing function of $`T`$.
We can define the $`q\overline{q}`$-potential as
$$𝒱(L)=_0(L)2m.$$
If the theory confines, $`𝒱(L)`$ is an increasing function of the distance $`L`$; if at large distances the increase is linear in $`L`$, namely $`𝒱(L)2\sigma L,`$ we obtain an area-law behaviour for the leading exponent with a string tension $`\sigma `$.
For $`D>2`$ perturbation theory is unreliable in computing the true spectrum of the $`q\overline{q}`$-system. However, when combined with unitarity, it puts an intriguing constraint on the $`q\overline{q}`$-potential. To realize this point, let us consider the formal expansion
$$𝒱(L)=g^2𝒱_1(L)+g^4𝒱_2(L)+\mathrm{},$$
(7)
$`g`$ being the QCD coupling constant.
When inserted in the expression $`\mathrm{exp}[2𝒱(L)T]`$, it gives
$$\mathrm{exp}[2T𝒱]=12T\left[g^2𝒱_1+g^4𝒱_2+\mathrm{}\right]+2T^2\left[g^4𝒱_1^2+\mathrm{}\right]+\mathrm{}.$$
(8)
At $`𝒪(g^4)`$, the coefficient of the leading term at large $`T`$ should be half the square of the term at $`𝒪(g^2)`$. This constraint has often been used as a check of (perturbative) gauge invariance.
Therefore, if we denote by $`C_{F(A)}`$ the quadratic Casimir expression for the fundamental (adjoint) representation of $`SU(N)`$ and remember that $`𝒱_1`$ is proportional to $`C_F`$, at $`𝒪(g^4)`$ the term with the coefficient $`C_FC_A`$ should be subleading in the large-$`T`$ limit with respect to the Abelian-like term, which is proportional to $`C_F^2`$.
Such a calculation at $`𝒪(g^4)`$ for the loop $`𝒲`$ has been performed using Feynman gauge in , with the number of space-time dimensions larger than two ($`D>2`$). The result depends on the area $`𝒜=4LT`$ and on the dimensionless ratio $`\beta =\frac{L}{T}`$. The $`𝒪(g^2)`$-term is obviously proportional to $`C_F`$; at $`𝒪(g^4)`$ we find that the non-Abelian term is indeed subleading
$$T^2𝒱^{na}C_FC_A𝒜^2T^{42D}.$$
(9)
Therefore agreement with exponentiation holds and the validity of previous perturbative tests of gauge invariance in higher dimensions is vindicated.
The limit of our result when $`D2`$ is finite and depends only on $`𝒜`$, as expected on the basis of the invariance of the theory in two dimensions under area-preserving diffeomorphisms. However the non-Abelian term is no longer subleading in the limit $`T\mathrm{}`$, as it is clear from eq.(9); we get instead
$$2T^2𝒱^{na}=C_FC_A\frac{𝒜^2}{16\pi ^2}(1+\frac{\pi ^2}{3}).$$
(10)
We conclude that the limits $`T\mathrm{}`$ and $`D2`$ do not commute.
This result is confirmed by a calculation of $`𝒲`$ performed in LCG with the ML prescription for the vector propagator . At odds with Feynman gauge where the vector propagator is not a tempered distribution at $`D=2`$, in LCG the calculation can also be performed directly in two space-time dimensions. The result one obtains does not coincide with eq.(10). One gets instead
$$2T^2𝒱^{na}=C_FC_A\frac{𝒜^2}{48}.$$
(11)
The extra term in eq.(10) originates from the self-energy correction to the vector propagator. In spite of the fact that the triple vector vertex vanishes in two dimensions in LCG, the self-energy correction does not. We stress that this “anomaly-like” contribution is not a pathology of LCG, it is needed to comply with Feynman gauge.
Perturbation theory is discontinuous at $`D=2`$. We conclude that the perturbative result, no matter what gauge one adopts, conflicts with unitarity in two dimensions.
Taking advantage of the invariance under area-preserving diffeomorphisms in dimensions $`D=2`$, Staudacher and Krauth were able to generalize our $`𝒪(g^4)`$ result (eq.(11)) by fully resumming the perturbative series. In the Euclidean formulation, which is allowed as the causal propagator can be Wick-rotated, and with a particular choice of the contour (a circumference), they get
$$𝒲(𝒜)=\frac{1}{N}\mathrm{exp}\left[\frac{g^2𝒜}{4}\right]L_{N1}^{(1)}\left(\frac{g^2𝒜}{2}\right),$$
(12)
the function $`L_{N1}^{(1)}`$ being a Laguerre polynomial.
This result can be further generalized to a loop winding $`n`$-times around the countour
$$𝒲=\frac{1}{N}\mathrm{exp}\left[\frac{g^2𝒜n^2}{4}\right]L_{N1}^{(1)}\left(\frac{g^2𝒜n^2}{2}\right).$$
(13)
¿From eq.(12) one immediately realizes that, for even values of $`N`$, the result is no longer positive in the large-$`T`$ limit. Moreover in the ’t Hooft’s limit $`N\mathrm{}`$ with $`g^2N=2\widehat{g}^2`$ fixed, the string tension vanishes and eq.(12) becomes
$$𝒲\frac{1}{\sqrt{\widehat{g}^2𝒜}}J_1(2\sqrt{\widehat{g}^2𝒜}),$$
(14)
$`J_1`$ being the usual Bessel function. Confinement is lost.
This explains the failure of the Wu’s approach in getting a bound state spectrum lying on rising Regge trajectories in the large-$`N`$ limit.
However in LCG the theory can also be quantized on the light-front (at equal $`x^+`$); with such a choice, in pure YMT and just in two dimensions, no dynamical degrees of freedom occur as the non vanishing component of the vector field does not propagate, but rather gives rise to an instantaneous (in $`x^+`$) Coulomb-like interaction (see eq.(1)).
Only planar diagrams contribute to the Wilson loop $`𝒲`$ for any value of $`N`$, thanks to the “instantaneous” nature of such an exchange; the perturbative series can be easily resummed, leading to the result (for imaginary time)
$$𝒲(𝒜)=\mathrm{exp}\left[\frac{g^2N𝒜}{4}\right],$$
(15)
to be compared with eq.(12).
Not only is this result in complete agreement with the exponentiation required by unitarity; it also exhibits, in the ’t Hooft’s limit $`N\mathrm{}`$ with $`g^2N=2\widehat{g}^2`$ fixed, confinement with a finite string tension $`\sigma =\frac{\widehat{g}^2}{2}`$. This explains the success of ’t Hooft’s approach in computing the spectrum of the $`q\overline{q}`$ bound states. The deep reason of this good behaviour lies in the absence of ghosts in this formulation; however there is no smooth way of deriving it from any acceptable gauge choice in higher dimension $`(D>2)`$. Moreover the confinement exhibited at this stage is, in a sense, trivial, being shared by $`QED_2`$.
We end up with two basically different results for the same model and with the same gauge choice (LCG), according to the different ways in which IR singularities are regularized. Moreover we are confronted with the following paradox: the prescription which is mandatory in dimensions $`D>2`$ is the one which fails at $`D=2`$. What is the meaning (if any) of eq.(12)?
## III THE GEOMETRICAL APPROACH
In order to understand this point, it is worthwhile to study the problem on a compact two-dimensional manifold; possible IR singularities will be automatically regularized in a gauge invariant way. For simplicity, we choose the sphere $`S^2`$. We also consider the slightly simpler case of the group $`U(N)`$. On $`S^2`$ we envisage a smooth non self-intersecting closed contour and a loop winding around it a number $`n`$ of times. We call $`A`$ the total area of the sphere, which eventually will be sent to $`\mathrm{}`$, whereas $`𝒜`$ will be the area “inside” the loop we keep finite in this limit.
Our starting point is the well-known heat-kernel expressions of a non self-intersecting Wilson loop for a pure $`U(N)`$ YMT on a sphere with area $`A`$
$`𝒲_n(A,𝒜)={\displaystyle \frac{1}{𝒵(A)N}}{\displaystyle \underset{R,S}{}}d_Rd_S\mathrm{exp}\left[{\displaystyle \frac{g^2𝒜}{2}}C_2(R){\displaystyle \frac{g^2(A𝒜)}{2}}C_2(S)\right]`$ (16)
$`\times {\displaystyle }dU\mathrm{Tr}[U^n]\chi _R(U)\chi _S^{}(U),`$ (17)
$`d_{R(S)}`$ being the dimension of the irreducible representation $`R(S)`$ of $`U(N)`$; $`C_2(R)`$ ($`C_2(S)`$) is the quadratic Casimir expression, the integral in (16) is over the $`U(N)`$ group manifold while $`\chi _{R(S)}`$ is the character of the group element $`U`$ in the $`R(S)`$ representation. $`𝒵(A)`$ is the partition function of the theory, its explicit form being easily obtained from $`𝒲_0(A,𝒜)=1`$.
We write eq.(16) explicitly for $`N>1`$ and $`n>0`$ in the form
$`𝒲_n(A,𝒜)={\displaystyle \frac{1}{𝒵(A)}}{\displaystyle \underset{m_i=\mathrm{}}{\overset{+\mathrm{}}{}}}\mathrm{\Delta }(m_1,\mathrm{},m_N)\mathrm{\Delta }(m_1+n,m_2,\mathrm{},m_N)`$ (19)
$`\times \mathrm{exp}\left[{\displaystyle \frac{g^2A}{4}}{\displaystyle \underset{i=1}{\overset{N}{}}}(m_i)^2\right]\mathrm{exp}\left[{\displaystyle \frac{g^2n}{4}}(A𝒜)(n+2m_1)\right].`$
We have described the generic irreducible representation by means of the set of integers $`m_i=(m_1,\mathrm{},m_N)`$, related to the Young tableaux, in terms of which we get
$$C_2(R)=\frac{N}{24}(N^21)+\frac{1}{2}\underset{i=1}{\overset{N}{}}(m_i\frac{N1}{2})^2,d_R=\mathrm{\Delta }(m_1,\mathrm{},m_N).$$
(20)
$`\mathrm{\Delta }`$ is the Vandermonde determinant and the integration in eq.(16) has been performed explicitly, using the well-known formula for the characters in terms of the set $`m_i`$ and taking symmetry into account.
¿From eq.(19) it is possible to derive, for $`n=1`$ and in the large-$`A`$ decompactification limit, precisely the expression (15) we obtained by resumming the perturbative series in the ’t Hooft’s approach. This is a remarkable result as it has now been derived in a purely geometrical way without even fixing a gauge. Actually, in the decompactification limit $`A\mathrm{}`$ at fixed $`𝒜`$, from eq.(19) one gets the following expression for any value of $`n`$ and $`N`$
$`𝒲_n(𝒜;N)={\displaystyle \frac{1}{nN}}\mathrm{exp}\left[{\displaystyle \frac{g^2𝒜}{4}}n(N+n1)\right]`$ (21)
$`\times {\displaystyle \underset{k=0}{}}{\displaystyle \frac{(1)^k\mathrm{\Gamma }(N+nk)}{k!\mathrm{\Gamma }(Nk)\mathrm{\Gamma }(nk)}}\mathrm{exp}\left[{\displaystyle \frac{g^2𝒜}{2}}nk\right].`$ (22)
We notice that when $`n>1`$ the simple abelian-like exponentiation is lost. In other words the theory starts feeling its non-abelian nature as the appearance of different “string tensions” makes clear. The winding number $`n`$ probes its colour content. The related light-front vacuum, although simpler than the one in the equal-time quantization, cannot be considered trivial any longer.
Eq.(21) exhibits an interesting symmetry under the exchange of $`N`$ and $`n`$. More precisely, we have that
$$𝒲_n(𝒜;N)=𝒲_N(\stackrel{~}{𝒜};n),\stackrel{~}{𝒜}=\frac{n}{N}𝒜,$$
(23)
a relation that is far from being trivial, involving an unexpected interplay between the geometrical and the algebraic structure of the theory .
Looking at eq.(23), the abelian-like exponentiation for $`U(N)`$ when $`n=1`$ appears to be related to the $`U(1)`$ loop with $`N`$ windings, the “genuine” triviality of Maxwell theory providing the expected behaviour for the string tension. Moreover we notice the intriguing feature that the large-$`N`$ limit (with $`n`$ fixed) is equivalent to the limit in which an infinite number of windings is considered with vanishing rescaled loop area. Alternatively, this rescaling could be thought to affect the coupling constant $`g^2\frac{n}{N}g^2`$.
¿From eq.(21), in the limit $`N\mathrm{}`$, one can recover the Kazakov-Kostov result
$$𝒲_n(𝒜;\mathrm{})=\frac{1}{n}L_{n1}^{(1)}(\frac{\widehat{g}^2𝒜n}{2})\mathrm{exp}\left[\frac{\widehat{g}^2𝒜n}{4}\right].$$
(24)
Now, using eq.(23) we are able to perfom another limit, namely $`n\mathrm{}`$ with fixed $`n^2𝒜`$
$$\underset{n\mathrm{}}{lim}𝒲_n(𝒜;N)=\frac{1}{N}L_{N1}^{(1)}\left(g^2𝒜n^2/2\right)\mathrm{exp}\left[\frac{g^2𝒜n^2}{4}\right].$$
(25)
We remark that this large-$`n`$ result reproduces the resummation of the perturbative series (for any $`n`$) (eq.(13)) in the causal formulation of the theory.
We go back to the exact expression we have found on the sphere for the Wilson loop (eq.19). As first noted by Witten , it is possible to represent $`𝒲_n(A,𝒜)`$ (and consequently $`𝒵(A)`$) as a sum over instable instantons, where each instanton contribution is associated to a finite, but not trivial, perturbative expansion. The easiest way to see it, is to perform a Poisson resummation
$`{\displaystyle \underset{m_i=\mathrm{}}{\overset{+\mathrm{}}{}}}F(m_1,\mathrm{},m_N)={\displaystyle \underset{f_i=\mathrm{}}{\overset{+\mathrm{}}{}}}\stackrel{~}{F}(f_1,\mathrm{},f_N),`$ (26)
$`\stackrel{~}{F}(f_1,\mathrm{},f_N)`$ $`=`$ $`{\displaystyle _{\mathrm{}}^+\mathrm{}}𝑑z_1\mathrm{}𝑑z_NF(z_1,\mathrm{},z_N)\mathrm{exp}\left[2\pi i(z_1f_1+\mathrm{}+z_Nf_N)\right]`$ (27)
in eq.(19). One gets
$`𝒲_n(A,𝒜)={\displaystyle \frac{1}{𝒵(A)}}\mathrm{exp}\left[{\displaystyle \frac{g^2n^2(A2𝒜)^2}{16A}}\right]\times `$ (28)
$`{\displaystyle \underset{f_i=\mathrm{}}{\overset{+\mathrm{}}{}}}\mathrm{exp}\left[S_{inst}(f_i)\right]W(f_1,\mathrm{},f_N)\mathrm{exp}\left[2\pi inf_1{\displaystyle \frac{A𝒜}{A}}\right],`$ (29)
where
$$S_{inst}(f_i)=\frac{4\pi ^2}{g^2A}\underset{i=1}{\overset{N}{}}f_i^2,$$
(30)
and
$`W(f_1,\mathrm{},f_N)={\displaystyle _{\mathrm{}}^+\mathrm{}}dz_1\mathrm{}dz_N\mathrm{exp}[{\displaystyle \frac{1}{g^2A}}{\displaystyle \underset{i=1}{\overset{N}{}}}z_i^2]\mathrm{exp}\left({\displaystyle \frac{inz_1}{2}}\right)\times `$ (31)
$`\mathrm{\Delta }(z_12\pi \stackrel{~}{f}_1,\mathrm{},z_N2\pi f_N)\mathrm{\Delta }(z_1+2\pi \stackrel{~}{f}_1,\mathrm{},z_N+2\pi f_N),`$ (32)
with
$$\stackrel{~}{f}_1=f_1+\frac{ig^2n}{8\pi }(A2𝒜).$$
These formulae have a nice interpretation in terms of instantons. Indeed, on $`S^2`$, there are non trivial solutions of the Yang-Mills equation, labelled by the set of integers $`f_i=(f_1,\mathrm{},f_N)`$
$$𝒜_\mu (x)=\mathrm{Diag}(f_1𝒜_\mu ^0(x),f_2𝒜_\mu ^0(x),\mathrm{},f_N𝒜_\mu ^0(x))$$
(33)
where $`𝒜_\mu ^0(x)=𝒜_\mu ^0(\theta ,\varphi )`$ is the Dirac monopole potential,
$$𝒜_\theta ^0(\theta ,\varphi )=0,𝒜_\varphi ^0(\theta ,\varphi )=\frac{1\mathrm{cos}\theta }{2},$$
$`\theta `$ and $`\varphi `$ being spherical coordinates on $`S^2`$. The term $`\mathrm{exp}\left[2\pi inf_1\frac{A𝒜}{A}\right]`$ in eq.(28) corresponds to the classical contribution of such field configurations to the Wilson loop.
Only the zero instanton contribution should be obtainable by means of a genuine perturbative calculation. Therefore in the following we single out the zero-instanton contribution ($`f_q=0`$, $`q`$) to the Wilson loop in eq.(28), obviously normalized to the zero instanton partition function .
The equation, after a suitable rescaling, becomes
$$𝒲_n^{(0)}(A,𝒜)=\frac{1}{𝒵^{(0)}(A)}\mathrm{exp}\left[\frac{g^2n^2(A2𝒜)^2}{16A}\right]W_1(0,\mathrm{},0)$$
(34)
with
$`W_1(0,\mathrm{},0)={\displaystyle _{\mathrm{}}^+\mathrm{}}𝑑z_1\mathrm{}𝑑z_N\mathrm{exp}\left[{\displaystyle \frac{1}{2}}{\displaystyle \underset{i=1}{\overset{N}{}}}z_i^2\right]\mathrm{exp}\left({\displaystyle \frac{in\sqrt{g^2A}z_1}{2\sqrt{2}}}\right)`$ (36)
$`\times \mathrm{\Delta }(z_1{\displaystyle \frac{in}{4}}\sqrt{{\displaystyle \frac{2g^2}{A}}}(A2𝒜),\mathrm{},z_N)\mathrm{\Delta }(z_1+{\displaystyle \frac{in}{4}}\sqrt{{\displaystyle \frac{2g^2}{A}}}(A2𝒜),\mathrm{},z_N).`$
The two Vandermonde determinants can be expressed in terms of Hermite polynomials and then expanded in the usual way. The integrations over $`z_2,\mathrm{},z_N`$ can be performed, taking the orthogonality of the polynomials into account; we get
$`𝒲_n^{(0)}(A,𝒜)=\mathrm{exp}\left[{\displaystyle \frac{g^2n^2(A2𝒜)^2}{16A}}\right]{\displaystyle \underset{n=0}{\overset{N}{}}}{\displaystyle \frac{1}{n!}}{\displaystyle \underset{k=2}{\overset{N}{}}}(j_k1)!{\displaystyle \frac{\epsilon ^{j_1\mathrm{}j_N}\epsilon _{j_1\mathrm{}j_N}}{𝒵^{(0)}(A)}}`$ (37)
$`{\displaystyle _{\mathrm{}}^+\mathrm{}}𝑑z_1\mathrm{exp}\left[{\displaystyle \frac{1}{2}}z_1^2\right]\mathrm{exp}\left({\displaystyle \frac{in\sqrt{g^2A}z_1}{2\sqrt{2}}}\right)He_{j_11}(z_{1+})He_{j_11}(z_1),`$ (38)
where
$$z_{1\pm }=z_1\pm \frac{in}{4}\sqrt{\frac{2g^2}{A}}(A2𝒜).$$
(39)
Integration over $`z_1`$ finally gives
$$𝒲_n^{(0)}(A,𝒜)=\frac{1}{N}L_{N1}^{(1)}\left(\frac{g^2𝒜(A𝒜)n^2}{2A}\right)\mathrm{exp}\left[\frac{g^2𝒜(A𝒜)n^2}{4A}\right].$$
(40)
At this point we remark that, in the decompactification limit $`A\mathrm{}`$, $`𝒜`$ fixed, the quantity in the equation above exactly coincides, for any value of $`N`$, with eq.(13), which was derived following completely different considerations. We recall indeed that eq.(13) was obtained by a full resummation of the perturbative expansion of the Wilson loop in terms of causal Yang-Mills propagators in LCG. Its meaning is elucidated by noting that it just represents the zero-instanton contribution to the Wilson loop, a genuinely perturbative quantity .
In turn it also coincides with the expression of the exact result in the large-$`n`$ limit, keeping fixed the value of $`n^2𝒜`$ (eq.(25)). This feature can be understood if we remember that instantons have a finite size; therefore small loops are essentially blind to them .
If the perturbative result has been correctly interpreted, it should be related to the local behaviour of the theory: in particular it should be possible to derive it starting from any topology, when the decompactification limit is eventually performed. In the zero-instanton contribution to a homologically trivial Wilson loop on the torus $`T^2`$ has been computed: the Poisson resummation is harder there and a larger number of classical solutions complicates the geometrical structure. In spite of the different complexity, the zero-instanton contribution, in the decompactification limit, still coincides with the perturbative result, as expected.
## IV THE $`𝒌`$-SECTORS
It was first noticed by Witten that two-dimensional Yang-Mills theory and two-dimensional QCD with adjoint matter do possess $`k`$-sectors. We consider $`SU(N)`$ as the gauge group: since Yang-Mills fields transform in the adjoint representation, the true local symmetry is the quotient of $`SU(N)`$ by its center, $`Z_N`$. A standard result in homotopy theory tells us that the quotient is not simply connected, the first homotopy group being $`\mathrm{\Pi }_1(SU(N)/Z_N)=Z_N.`$ This result is of particular relevance for the vacuum structure of a two-dimensional gauge theory: in the case at hand we have exactly $`N`$ inequivalent quantizations, parametrized by a single integer $`k`$, taking the values $`k=0,1,..,N1`$. Concerning the pure $`SU(N)`$ Yang-Mills theory, the explicit solution when $`k`$-states are taken into account was presented in Ref. : their main result, the heat-kernel propagator on the cylinder, allows to compute partition functions and Wilson loops on any two-dimensional compact surface, therefore generalizing the well-known Migdal’s solution to $`k`$-sectors. Wilson loops, in this case, strongly depend on $`k`$: for a non self-intersecting loop we have, on the plane,
$`𝒲_k(A)={\displaystyle \frac{1}{N^21}}[1+{\displaystyle \frac{kN(N+2)(Nk)}{(k+1)(Nk+1)}}e^{\frac{g^2A}{2}(N+1)}`$ (41)
$`+`$ $`{\displaystyle \frac{(N+1)(Nk1)}{k+1}}e^{\frac{g^2A}{2}(Nk)}+{\displaystyle \frac{(N+1)(k1)}{Nk+1}}e^{\frac{g^2A}{2}k}].`$ (42)
This result can be obtained starting from the true $`SU(N)/Z_N`$ theory on the sphere , in the decompactification limit, or directly on the plane, using the procedure of , working with $`SU(N)`$ and simulating the $`k`$-sectors with a Wilson loop at infinity in the $`k`$fundamental representation. The very same result can be obtained through a perturbative resummation with ’t Hooft potential and the $`k`$-loop at infinity. On the other hand we expect that the truly perturbative physics ignore the existence of the $`k`$ parameter: by Poisson-resumming the result on the sphere for $`SU(N)/Z_N`$, we arrive to an instanton representation different from the $`SU(N)`$ case. Contribution from $`N`$ inequivalent classes of instantons ensues, with instanton numbers generalized to rational values by the effect of $`k`$. The zero-instanton limit does not depend on $`k`$ and still reproduces the WML computation (in the decompactification limit) without the loop at infinity
$`𝒲_k^{(0)}(A)={\displaystyle \frac{1}{N+1}}+{\displaystyle \frac{N}{𝒵(N+1)}}{\displaystyle _{\mathrm{}}^+\mathrm{}}dz_1\mathrm{}dz_N\mathrm{exp}[{\displaystyle \frac{1}{2}}{\displaystyle \underset{j=1}{\overset{N}{}}}z_j^2]\times `$ (43)
$`\mathrm{exp}\left[ig(z_1z_2)\sqrt{{\displaystyle \frac{A}{2}}}\right]\mathrm{\Delta }^2(z_1,\mathrm{},z_N),`$ (44)
where $`𝒵=𝒟F\mathrm{exp}(\frac{1}{2}\mathrm{Tr}F^2).`$ In the presence of a $`k`$-loop at infinity, the WML computation, although coinciding with the zero-instanton limit of the quantum average of two nested ($`k`$ and adjoint) loops, does depend on $`k`$. We conclude that $`\mathrm{𝑜𝑛𝑙𝑦}`$ for the complete theory on the plane (i.e. full-instanton resummed and then decompactified) the equivalence between $`k`$-sectors and theories with $`k`$-fundamental Wilson loops at infinity holds.
|
warning/0004/astro-ph0004158.html
|
ar5iv
|
text
|
# Vertical structure of the accreting two-temperature corona and the transition to an ADAF
## 1 Introduction
There are strong observational evidence for the existence of the hot plasma $`10^9`$K in the neighborhood of the accretion disc in active galactic nuclei (AGN) and galactic black holes (GBH). Soft disc photons are Compton upscattered by this optically thin plasma and form an approximately power law hard X-ray spectral component. The presence of the Compton reflection hump and the cold iron $`K_\alpha `$ line in the hard spectrum is a signature of strong illumination of cold disc matter by X-ray radiation (for a review see Mushotzky, Done & Pounds 1993, Tanaka & Lewin 1995, Madejski 1999).
The location of the hot plasma is still under debate, but one of the attractive models postulated by Liang & Price (1977) and Paczyński (1978) introduces hot dissipative corona above accretion disc following the example of the solar corona. via magnetohydrodynamic waves, jets or other nonthermal energy transport, and then thermalized in corona. But such ways of energy transport occurred to be inefficient, only $`10\%`$ of energy generated in the disc can be carried out towards corona. Within the frame of such geometry, observations seem to require that in some sources and/or some spectral states the innermost part of the disk is either disrupted and replaced by an optically thin flow or there is a strong coronal outflow (e.g. Esin, McClintock & Narayan 1997, Zdziarski, Lubiński & Smith 1999, Beloborodov 1999). Therefore, it is important to determine whether any of such effects are expected on the basis of the physical description of the accretion flow (Narayan 1997).
The disc/corona models were recently investigated following two basic lines.
The first group of models does not explain the mechanism which heats up the plasma in the disk/corona part of the flow. The fraction of energy generated in the corona $`f_{\mathrm{cor}}`$ is treated as a free parameter of the model (Haardt & Maraschi 1991, Svensson & Zdziarski 1994) or changes according to the artificial formula being inversely proportional to the distance form a central object (Esin et al. 1997). By changing $`f_{\mathrm{cor}}`$ and the geometrical shape of the hot region (continuous, patchy or clumpy corona), such models can be fitted to the observed X-ray spectra (for review see Pountanen 1999). The main conclusion from those models is that static corona agrees with observations only when it does not cover the whole disc. However, more accurate considerations of the reflection, taking into account the optical depth of the Compton heated zone (Różańska & Czerny 1996, Nayakshin, Kazanas & Kallman 1999) or an outflow (Beloborodov 1999) may influence this result to some extent.
The second group of models follow Shapiro, Lightman & Eardley (1976) solution when hot optically thin flow exists in the inner part of the cold disc. They assume both cold and hot flow to be driven via accretion and depending on the model they take into account mechanical energy transport in the radial direction: like advection (Ichimaru 1977; Narayan & Yi 1994, 1995a, 1995b; Abramowicz et al. 1995; Chen et al. 1995; for review see Narayan, Mahadevan & Quataert 1998 and Kato, Fukue & Mineshige 1998), turbulent diffusion (Honma 1996), hydrodynamical inflow/outflow (Narayan & Yi 1994, 1995a; Blandford & Begelman 1999) and/or in the vertical direction: thermal conduction, evaporation or coronal wind (Meyer & Meyer-Hofmeister 1994 hereafter MMH94, Meyer-Hofmeister & Meyer 1999 Liu et. al 1999, Turolla & Dullemond 2000).
In both groups of models the crucial thing is the transition between the hot and the cold phase. In general picture such a transition occurs in two directions: vertical transition when corona accretes above a cold disc, and radial transition when disk/corona system is replaced with a single optically thin flow close to the black hole (Fig. 1). Therefore, as was pointed by MMH94, in general picture both radial and vertical structure of the hot and cold phase should be considered.
In this paper we present the simplest disc/corona model which is the combination of two groups of models described above. The similar model was presented by Esin et al. (1997) with better description of ADAF region, and with non-local radiative interaction between the hot flow and cold disc. Nevertheless, they were not able to determine self consistently the transition between disc/corona flow, since they did not calculate its vertical structure. Also MMH94 stressed that thermal conduction and mass exchange between disc and corona in vertical direction are important do determine the transition. However, in their model (Meyer-Hofmeister & Meyer 1999 Liu et. al 1999) the radiative interaction between phases was neglected, an assumption of a single-temperature plasma was used, and the conductive flux on the bottom of the hot phase was artificially taken from the solar corona.
Our model takes into account radiative coupling between phases, thermal conduction and evaporation in vertical direction, and radial energy transport via advection. The mass exchange rate is calculated self-consistently from the physical conditions in the transition zone. Hot corona is treated as two temperature plasma and, similarly to the disc, it is powered by accretion (Witt, Czerny & Życki 1997; hereafter WCŻ97, Esin et al. 1997).
On each radius the vertical structure is calculated depending on comparison of accretion and vertical evaporation timescales. When accretion time scale is long enough, the mass exchange can be neglected and the ’static’ equilibrium is achieved. In such a case, for vertical transition we use the condition based on considerations of Krolik (1998), Dullemond (1999), and Różańska & Czerny (2000) which is relevant to the dissipative corona. When accretion time scale is short mass exchange cannot be neglected, and we determine the evaporation rate through the disc/corona transition.
Vertical properties are used to compute the radial trends solving continuity equation. The aim of this paper is to determine the radial dependence of the fraction of energy liberated in the corona for a given accretion rate, viscosity and a mass of the black hole and to check when the disk evaporation leads to a transition from disc/corona flow to the single optically thin flow. The radial transition occurs at the radius when all the mass starts to be carried by the corona.
Observational consequences of our model occur in the dependence of the transition radius on accretion rate, which can be seen in different shapes of the reflected component (Done & Życki 1999). Also the evaporation leads to the new stable branch on the accretion flow’s stability curve ( i.e. $`\dot{m}(\mathrm{\Sigma })`$), allowing for new evolution cycle to exist. If accretion rate is too low optically thick disk can evaporate completely in the innermost parts and the position of transition radius moves outward. This can explain the different luminosity states in AGNs and GBHs
In Sec 2 we present the model. Calculation are shown in Sec. 3, and we discuss our results in Sec. 4. Evaporation rate is determined in Appendix A.
## 2 Model
The basic problem in modeling the two-phase accretion flow onto a black hole is the determination of the relative importance of the hot optically thin flow and cold optically thick flow. It was shown by MMH94 in the case of dwarf novae that such estimates have to be based on the study of the vertical structure of such a flow since only in that case we can determine the mass exchange between the phases and the conditions of their coexistence. Therefore we follow this basic approach but we adjust the physical input to the conditions appropriate for the accretion close to the black hole.
The main difference from a dwarf novae is that in the case of inner regions of the accretion disc in AGNs and GBH, two-temperature plasma should be taken into account (Narayan & Yi 1995b). This is easy to see from the following simple argument. With the assumption of local viscous heating of ions, the total flux generated locally in the standard disk is:
$$F_{\mathrm{gen}}=\frac{3}{2}\alpha P\mathrm{\Omega }_\mathrm{K}H,$$
(1)
where $`\mathrm{\Omega }_\mathrm{K}`$ is the Keplerian angular frequency equal to $`(GM/r^3)^{1/2}`$, $`\alpha `$ is the viscosity, $`H`$ is the height scale of the flow and P is the pressure.
On the other hand, the total flux dissipated locally in a stationary Keplerian disk can be described as:
$$F_{\mathrm{gen}}=\frac{3GM\dot{M}}{8\pi r^3}(1f(r/R_{\mathrm{Schw}})),$$
(2)
where $`f(r/R_{\mathrm{Schw}})`$ represent the inner boundary condition, in our case taken from the Newtonian approximation, $`f(r/R_{\mathrm{Schw}})=(3R_{\mathrm{Schw}}/r)^{1/2}`$.
The optical depth of the flow can be described by $`\tau _{es}=\kappa _{es}\rho H`$, where $`\kappa _{es}`$ is the opacity coefficient for electron scattering and $`\rho `$ is the density. We use also following equation of state: $`P=\frac{k}{\mu m_\mathrm{H}}\rho T`$, with the value of molecular weight $`\mu =0.5`$ for cosmic chemical composition. Boltzmann constant is denoted as $`k`$ and the mass of hydrogen atom as $`m_\mathrm{H}`$.
Combining these equations, we can estimate the temperature of the optically thin $`\tau _{\mathrm{es}}1`$ flow as:
$$T=2.31\times 10^{13}\frac{\dot{m}}{\alpha }(R_{\mathrm{Schw}}/r)^{3/2}[1(3R_{\mathrm{Schw}}/r)^{1/2}]$$
(3)
We use accretion rate $`\dot{m}=\dot{M}/\dot{M}_{Edd}`$ in units of Eddington accretion rate described as:
$$\dot{M}_{Edd}=\frac{4\pi GM}{\eta \kappa _{es}c}$$
(4)
with accretion efficiency $`\eta =1/12`$. Thus the temperature does not depend on the mass of the black hole $`M`$, and can be computed for assumed value of Eddington accretion rate, radius and viscosity. Taking a realistic value $`\dot{m}/\alpha 1`$, and the distance appropriate for dwarf nowae which is $`10^4R_{\mathrm{Schw}}`$ we obtain the temperature equal to $`10^7`$ K, reasonable for the single temperature plasma. But, for $`r=10R_{\mathrm{Schw}}`$, the temperature attains the value of $`4\times 10^{11}`$ K, far too high for electrons, according to the observational data, and two-temperature plasma should be considered.
### 2.1 Basic assumptions and model parameters
In the present paper we consider the case of stationary accretion onto a non-rotating black hole. We assume that accreting material forms an optically thick disk underlying an optically thin hotter skin - a corona. The flow is described as a continuous medium so a possible clumpiness of either the disk or the corona are neglected. The transition of the flow in the vertical direction from a corona to a disk is gradual and corresponds to a change of the cooling mechanism with an increasing density and pressure towards the equatorial plane. Since the computations of the optically thick part of the flow required another approach than the optically thin parts we still introduce a formal division of the flow into a disk and the corona but the physical parameters at the bottom of the corona (pressure, density and temperature) are equal to the disk values so there is no discontinuity between the two media.
We assume that both the disk and the corona are powered directly by the release of the gravitational energy due to accretion. Viscous torque is parameterized as in the standard disk of Shakura & Sunyaev (1973). This approach is supported to some extent by the magnetohydrodynamical simulations of Hawley, Gammi & Balbus (1996). Such assumption was used e.g. by MMH94 and WCŻ97. Since the corona is accreting we assume the direct heating of ions and the subsequent energy transfer to electrons through the Coulomb coupling. Therefore, the corona is basically a two-temperature medium unless the Coulomb coupling is efficient enough to equalize the temperatures. This assumption is a simple, well determined starting point but in the future the ion-electron energy exchange should be considered more closely. Proton heated plasma, adopted in our paper, is hardly thermalised (Gruzinov & Quataert 1999). Spruit & Haardt (2000), motivated by large mean free path of ions considered the disk heating not as conduction due to electrons, but conduction due to the ions. Another point is that protons are heated efficiently only in the case of a weakly magnetized plasma; in a plasma with field in equipartition heat goes to electrons (Quataert & Gruzinov 1999). Also Ohmic heating stressed by Bisnovatyi-Kogan & Lovelace (1997) can efficiently heat up the electrons and advection may be not important. Diminishing the role of advection may have further consequences for the efficiency of mass exchange between the disk and the corona (Różańska & Czerny 2000).
We take into account the radial advection, Compton and bremsstrahlung cooling and the conduction due to electrons and the vertical mass exchange, which are important at the basis of the corona. Disk and corona are radiatively coupled, as in the classical paper of Haart & Maraschi (1991) although the cooling process is now treated locally. We assume that, the cold disc is the origin of the seed photons at a given radius. We neglect the contribution of the synchrotron radiation to the seed photons for Comptonization (Narayan & Yi 1994), as well as non-local photons which were included in the study of Esin et al. (1997). We also neglect the formation of wind at the top of corona discussed in some papers (e.g. Begelman, McKee & Shields 1983, Begelman & Blandford 1999).
In flows cooled by advection, the motion may be not Keplerian and proper calculation of equation of motion should be done to describe this effect. Nevertheless, the aim of this paper is to calculate properly the vertical structure while the radial one is computed less carefully. Therefore, together with hydrostatic equilibrium in vertical direction, we assume the Keplerian motion of the material.
The model parameters are: mass of the black hole, $`M`$, accretion rate, $`\dot{m}`$, and the viscosity parameter $`\alpha `$, assumed here for simplicity to be the same both in the disk and in the corona. All other quantities, including their radial dependences, are uniquely determined by those global parameters for a stationary flow. Corona model does not actually depend on $`M`$ but there is a weak dependence of the cold disk properties on the mass of the black hole itself so this parameter cannot be entirely neglected.
### 2.2 Vertical structure of the corona at a given radius
In the present study we simplify the problem of hydrostatic equilibrium. Instead of solving complex differential equations as in MMH94, we simply study algebraic dependences of various quantities on the pressure which is a monotonic function of the distance from equatorial plane. With such assumption the local viscous heating of ions $`Q_\mathrm{i}^+`$ can be described following the Eq. 1:
$$Q_\mathrm{i}^+=\frac{3}{2}\alpha P\mathrm{\Omega }_\mathrm{K},$$
(5)
A fraction of this energy is carried by advection. This fraction is proportional to the ratio of the ion temperature $`T_\mathrm{i}`$ to the virial temperature $`T_\mathrm{v}`$ with the coefficient $`\delta `$ determined by the radial derivatives of thermodynamical quantities (e.g. Ichimaru 1977, Muchotrzeb & Paczyński 1982, Abramowicz et al. 1995, Janiuk, Życki & Czerny 2000). At present we simply assume $`\delta =1`$ but in principle this value should be determined iteratively.
Remaining fraction of energy is transfered to electrons via Coulomb coupling
$$Q_\mathrm{i}^+\left(1\delta \frac{T_\mathrm{i}}{T_\mathrm{v}}\right)=Q_\mathrm{e}^+=\frac{3}{2}\frac{k}{m_\mathrm{H}}D\rho ^2(T_\mathrm{i}T_\mathrm{e})T_\mathrm{e}^{3/2}$$
(6)
where $`T_\mathrm{e}`$ is the electron temperature, and D is the Coulomb coupling constant equal $`4.88\times 10^{22}`$ (in s<sup>-1</sup>). At this exploratory stage, we neglect relativistic corrections to this equation given by Stepney & Guilbert (1983).
In the upper layers of the corona the electrons cool down radiatively, i.e.
$$Q_\mathrm{e}^+=\mathrm{\Lambda }(\rho ,T_\mathrm{e}).$$
(7)
The processes included are the Compton cooling due to the presence of soft photons $`F_{\mathrm{soft}}`$ from the disk and through the bremsstrahlung cooling
$$\mathrm{\Lambda }(\rho ,T_\mathrm{e})=F_{\mathrm{soft}}\kappa _{\mathrm{es}}\frac{4k}{m_\mathrm{e}c^2}\rho T_\mathrm{e}+B\rho ^2T_\mathrm{e}^{1/2}.$$
(8)
Here $`F_{\mathrm{soft}}`$ is the soft radiation flux coming from the disk due to the energy generation in the optically thick part of the flow as well as due to interception and reemission of the coronal emission $`F_{\mathrm{hard}}`$ (Haardt & Maraschi 1991)
$$F_{\mathrm{soft}}=F_{\mathrm{disk}}+0.5𝐅_{\mathrm{hard}},$$
(9)
$`m_\mathrm{e}`$ \- mass of the electron, $`c`$ \- light velocity, and $`B`$ \- bremsstrahlung cooling constant equal $`6.6\times 10^{20}`$ erg s<sup>-1</sup>cm<sup>-2</sup> g<sup>-2</sup>. We assume that half of the coronal flux $`𝐅_{\mathrm{hard}}`$, illuminates the disc and all of this is absorbed and subsequently reemited, so albedo is equal to zero.
In Eq. 8 we include only the Compton cooling and we neglect, for simplicity, the Compton heating term which may be important within the transition zone where the electron temperature drops below $`10^8`$ K.
Equations 6 and 8 determine the temperature profile in the upper part of the corona, or more precisely, the dependence of the ion and electron temperature on the gas pressure which we identify with the ion pressure for simplicity, $`P=k/m_\mathrm{H}\rho T_\mathrm{i}`$.
At some depth, however, a transition to a cool disk is expected. Within such a transition zone the essential role is played by conduction, neglected so far. Here we assume that the transition from hot to cold gas happens very sharply, under the constant pressure, $`P_{\mathrm{base}}`$, so the drop in the temperature is simply compensated by the increase in the density. Determination of this pressure is the key element of the model since the value of this pressure uniquely determines the fraction of the energy $`f_{\mathrm{cor}}`$ liberated in the corona in the following way.
Since the ion temperature is close to the virial temperature, the pressure scale high is roughly of the order of the radius, $`r`$, so instead of solving precisely the hydrostatic equilibrium we can estimate the total energy flux generated in the corona as
$$F_{\mathrm{cor}}=Q_\mathrm{i}^+(P_{\mathrm{base}})r,$$
(10)
and the flux emitted by corona as
$$F_{\mathrm{hard}}=Q_\mathrm{i}^+\left(1\delta \frac{T_\mathrm{i}}{T_\mathrm{v}}\right)r.$$
(11)
The total energy dissipated due to accretion both in the disk and the corona is given by formula:
$$F_{\mathrm{tot}}=F_{\mathrm{cor}}+F_{\mathrm{disk}}=F_{\mathrm{soft}}+0.5F_{\mathrm{hard}}\frac{1+\delta \frac{T_\mathrm{i}}{T_\mathrm{v}}}{1\delta \frac{T_\mathrm{i}}{T_\mathrm{v}}},$$
(12)
which relates to the accretion rate both within the disk and the corona due to the assumption of stationarity and Keplerian motion and $`F_{\mathrm{tot}}=F_{\mathrm{gen}}`$ given by Eq. 2.
In the accreting corona model, the fraction of energy dissipated in the corona corresponds to the same fraction of the mass accreting through the hot phase, i.e.
$$\frac{\dot{m}_{\mathrm{cor}}}{\dot{m}}=\frac{F_{\mathrm{cor}}}{F_{\mathrm{tot}}}$$
(13)
In this way the determination of the pressure at the basis of the corona allows to determine its structure at a given radius from the global model parameters, $`M`$, $`\dot{m}`$, and $`\alpha `$. Therefore the way how we determine this value is a crucial element of the model itself which removes arbitrariness from the disk/corona accretion flow present in most models apart from MMH94, Życki, Collin-Souffrin & Czerny (1995), WCŻ97 and the directly related papers.
The computations of the vertical structure of the optically thick part of the flow are not necessary for the computation of the corona. However, when the corona structure is determined the disk structure is calculated, with the boundary conditions imposed by the presence of corona, as described by Różańska et al. (1999). No iterations between the disk and the corona computations are needed.
The properties of the transition zone are calculated following the approach of Krolik (1998), Dullemond (1999), and Różańska & Czerny (2000). We consider two cases: a ’static’ balance solution without the mass exchange between the disk and the corona and the self-consistent case based on the continuity equation of the coronal flow.
#### 2.2.1 Equilibrium solutions without mass exchange
Interesting and important solutions are obtained under an assumption that the accretion flow is slow in comparison with the timescale of exchange of the mass between the disk and the corona. In that case the equilibrium solution is defined by the condition that there is no mass exchange between the disk and the corona (see Różańska & Czerny 2000).
To describe the equilibrium solution we integrate the energy balance equation as shown in Appendix A, with velocity in vertical direction $`v_\mathrm{z}`$ equal zero (see Eq. 34). It meant, that the heat flux which is transported via electrons from hot upper corona towards the cold matter is balanced by radiative cooling operating in the disc. Since the transition is described under constant pressure the appropriate criterion reduces to an integral (Różańska & Czerny 2000)
$$_{T_\mathrm{s}}^{T_\mathrm{e}(P)}(Q_\mathrm{e}^+\mathrm{\Lambda })\kappa _\mathrm{o}T_\mathrm{e}^{5/2}𝑑T_\mathrm{e}=0$$
(14)
where $`T_\mathrm{s}`$ is the cold disk temperature and $`T_\mathrm{e}(P)`$ is the electron temperature within the upper corona part determined at the pressure P and satisfying the heating/cooling balance formulated by Equation 7. Here, $`\kappa =\kappa _\mathrm{o}T^{5/2}`$ is thermal conductivity coefficient in $`\mathrm{ergs}\mathrm{cm}^1\mathrm{s}^1\mathrm{K}^1`$ and conductivity constant $`\kappa _\mathrm{o}`$ is assumed to be equal $`5.6\times 10^7`$. The ion temperature in this equation has to be determined from the Coulomb coupling equation (Eq. 6) and the density from the value of the (constant) pressure and the ion temperature.
Equation 14 determines the pressure of the transition zone, thus closing up the set of equations determining the disk/corona flow.
#### 2.2.2 Solutions with evaporation/condensation
The mass exchange between the disk and the corona cannot be neglected if the timescale of accretion is short so the ’static’ type of equilibrium cannot develop. It is also necessary if we want to describe the process of corona formation. This approach is more appropriate but more complex since it is essentially non-local and requires a global solution for all radii. Here we basically follow an approach of MMH94 by finding a family of numerical solutions at a single radii, determining their scaling properties with the radius and finally finding unique global solution at the basis of this scaling.
Therefore we allow the Equation 14 to be not satisfied which means that we have to solve the vertical energy transfer in the transition zone allowing for evaporation or condensation, depending on the choice of the pressure $`P_{\mathrm{base}}`$. If radiative processes are to low to balance the heat flux, the disc will evaporate as shown by Dullemond (1999). However, radiative cooling may overwhelm the heat flux if pressure at the base is high enough (higher than the pressure under which mass exchange equals to zero, as shown by McKee & Begelman in 1990), and in such situation matter will condense.
To determine the mass exchange rate we integrate the energy balance equation (32) in the way detaily presented in Appendix A. Combining equations (35) and (38) the evaporation rate $`\dot{m}_\mathrm{z}`$ across the disk/corona boundary is given by the formula.
$$\dot{m}_\mathrm{z}=\frac{_{T_\mathrm{s}}^{T_\mathrm{e}(P)}[Q_\mathrm{e}^+\mathrm{\Lambda }]\kappa _\mathrm{o}T_\mathrm{e}^{5/2}𝑑T_\mathrm{e}}{|_{T_\mathrm{s}}^{T_\mathrm{e}(P)}(Q_\mathrm{e}^+\mathrm{\Lambda })\kappa _\mathrm{o}T_\mathrm{e}^{5/2}𝑑T_\mathrm{e}|^{1/2}}\left(\frac{2}{5}\frac{m_\mathrm{H}}{kT_\mathrm{i}(P)}\right)$$
(15)
The sign of this quantity is either positive (disk material evaporates into the corona) or negative (coronal material condensates into the disk), depending on the net effect of the electron heating and cooling balance.
The radial dependence of $`P_{\mathrm{base}}`$ is subsequently determined from the continuity equation (see Section 2.3.2).
### 2.3 Radial component of the continuity equation and global solutions
#### 2.3.1 Solutions without mass exchange
In this case the equilibrium is built locally so the determination of the disk/corona structure at a given radius is completed if the local solution at this radius is found. The accretion rate within the corona is determined at each radius separately and the mass exchange between the disk and the corona account for its dependence on the disk radius, i.e. its evolution with the flow. All parameters, like accretion rate in the corona, soft X-ray flux from the disk, the ion and electron temperature profile, the density profile and the corona optical depth as well as the pressure at the basis of the corona where the temperature drops down to match the disk values are determined by the choice of the radius $`r`$ and the three global model parameters, $`M`$, $`\dot{m}`$ and $`\alpha `$.
#### 2.3.2 Solutions with evaporation/condensation
If the timescale of building the evaporation/condensation equilibrium is long in comparison with the viscous timescale in the corona we have to describe the build-up process of the corona, i.e. the accretion rate in the corona at any radius $`r`$ will reflect the amount of material which evaporated from the disk at all radii larger and equal to $`r`$.
The change of the accretion rate in the corona $`\dot{M}_{\mathrm{cor}}`$ is therefore due to the mass exchange with the disk
$$\frac{d\dot{M}_{\mathrm{cor}}}{dr}=2\pi r\dot{m}_\mathrm{z}.$$
(16)
which in dimensionless units reduces to
$$\frac{d\dot{m}_{\mathrm{cor}}}{dR}=\frac{2\eta \kappa _{\mathrm{es}}G}{c^3}MR\dot{m}_\mathrm{z}=27.9M_{\mathrm{BH8}}R\dot{m}_\mathrm{z}$$
(17)
where $`R=r/R_{\mathrm{Schw}}`$ and $`M_{\mathrm{BH8}}=M/10^8M_{}`$.
In order to solve this equation we follow the approach of MMH94. We find a family of solutions at a given radii for a set of values of the pressure $`P_{\mathrm{base}}`$. We express those solutions through the computed $`\dot{m}_{\mathrm{cor}}`$, thus obtaining a numerical relation $`\dot{m}_\mathrm{z}(\dot{m}_{\mathrm{cor}})`$. We study the scaling of this relation with radius, thus finding an analytical expression valid in the entire disk. This analytical expression is supplied to Equation 17 and the final analytical result for the dependence $`\dot{m}_{\mathrm{cor}}(r)`$ is found.
### 2.4 Cold disk structure
Determination of the corona properties, i.e. the fraction of energy dissipated in the corona and the pressure at the corona basis provides the boundary conditions necessary to calculate the vertical structure of the cold disk flow (e.g. Rȯżańska et al. 1999). Therefore, the description of the hot flow influences the properties of the cold part of the flow to their strong coupling through these boundary conditions.
In the present paper we concentrate mostly on the hot flow properties and the cold disk properties are only required for the qualitative discussion of the connection between the single phase hot flow and disk/corona flow (see Section 3.3) so we simplify the description of the cold flow. We apply the vertically averaged approach but we determine the dimensionless coefficients involved in this process on the basis of full computations of the disk vertical structure, as in Janiuk et al. (2000) (see also Muchotrzeb & Paczyński 1982, Abramowicz et al. 1988).
We assume the same value of the viscosity coefficient in the hot and in the cold flow although they may in principle be considerably different due to the different plasma properties. We include advection in our description of the flow, and the appropriate coefficients used are: $`B_1=B_2=B_3=B_4=1`$, and $`q_{\mathrm{adv}}=2`$. The disk is assumed to be optically thick so the current approach does not allow yet to follow the final transition from the disk/corona flow to a single phase optically thin accretion.
## 3 Results
### 3.1 Solutions without mass exchange
#### 3.1.1 Vertical structure of the flow at a given radius
We first discuss the vertical ion and electron temperature and density profiles at a representative radius $`r=10.8R_{\mathrm{Schw}}`$ for mass of the black hole equal $`10^8\mathrm{M}_{}`$ and the viscosity parameter $`\alpha =0.1`$. The virial temperature corresponding to this radius is equal $`10^{12}`$ K.
The set of algebraic equations (Eq. 2 and Eqs. 5 \- 12) describing the upper layers of the corona formally posses two solutions for a given pressure (corresponding to a certain distance $`z`$ from the equatorial plane): one solution is mostly advection-dominated, low density branch and the other one is high density, mostly radiatively cooled branch. This property, found also by Dullemond (1999), is a direct analogy of the vertically averaged optically thin solutions found by Abramowicz et al. (1995). Both solutions exists for low values of the pressure but they approach each other and finally merge at a certain value of the pressure, corresponding to an ion temperature $`T_\mathrm{i}=0.5T_\mathrm{v}`$. Dipper inside there are no solutions, as in the case of too high accretion rate in optically thin flows.
However, the situation of the disk/corona accretion is not the same as in the case of a single phase flow. Only one of the above solutions is acceptable. At the advection-dominated branch the ion temperature decreases with the increase towards the midplane and the density increase thus opening a possibility to match this solution to the cool disk part of the flow. The other branch shows opposite trends and therefore we reject it as unphysical in the case of two-phase accretion.
The fact, that the algebraic solution does not extends very deeply does not pose a problem since we have now to take into account the transition to the disk flow under no evaporation conditions. The value of the pressure which allows the coexistence of the disk and coronal flow under no evaporation/condensation condition given by Equation 14 is always smaller than the value of the pressure at which the two solutions merge.
The example of the change of the ion and electron temperature with a pressure increasing towards the equatorial plane is shown in Fig. 2 for the accretion rate $`\dot{m}=1.3\times 10^3`$ in units of Eddington accretion rate. The ion temperature is very close to the virial value, most of the heat generated in the corona is advected towards smaller radii. The electron temperature is always significantly lower than the ion temperature since the Coulomb coupling is nor very efficient but it slowly increases towards the equatorial plane (i.e. increasing $`P`$). For the assumed parameters the equilibrium pressure $`P_{\mathrm{base}}`$ is equal $`7.94\times 10^3`$ g cm<sup>-3</sup>.
The fraction of energy generated in the corona is equal 0.89, so the soft photon flux is predominantly due to the fraction of corona emission absorbed by the disk ($`F_{\mathrm{soft}}=5.2\times 10^{12}`$erg s<sup>-1</sup>cm<sup>-2</sup>). The optical depth of the corona is very low, $`6\times 10^3`$ and the density at the upper edge of the transition zone is equal $`5\times 10^{16}`$ g s<sup>-2</sup> cm<sup>-1</sup>.
The transition zone itself is also complex. We assumed that the transition happens at a constant pressure so its structure cannot be shown in Fig. 2. Instead, we draw the relation between the ion and electron temperature within this zone, as predicted, neglecting thermal conduction (only for radiative processes and advection), in Fig. 3.
In the upper part of the transition zone itself the ion temperature is still very close to the virial temperature, with advection as a dominant cooling mechanism for ions. However, deeper inside the ion temperature decreases and the density increases. Radiative cooling becomes more efficient. The advection-dominated flow is replaced by the radiatively cooled flow, basically of the type studied by Shapiro, Lightman & Eardley (1976). This branch was found to be unstable (Piran 1978, Narayan & Yi 1995b) and it does not provide us with a monotonic decrease of the electron temperature so we assume that actually a sharp transition takes place along a vertical line, as marked in Fig. 3. Our integral of the difference between the heating and cooling multiplied by the conduction flux (Equation 14) is calculated applying such a profile. Full solution of differential equations with the conduction due to electrons as well as ions would produce a smooth solution matching the upper and lower branch without the third, intermediate branch. We intend to study this problem in the future but for the determination of the basis solution properties the detailed knowledge of the transition zone is not necessary as the optical depth of this zone is generally very small.
The fraction of the energy generated in the corona for an accretion rate of $`\dot{m}=1.47\times 10^3`$ is high but generally this is not the case. In Fig. 4 we show the dependence of this value on the accretion rate at the radius 10 $`R_{\mathrm{Schw}}`$. For large accretion rates the corona is weak and its strength increases rapidly when the accretion rates drops down to $`\dot{m}=1.53\times 10^3`$. Below $`1.47\times 10^3`$ there are no solutions for disk/corona equilibrium without the mass exchange. Solutions with negative value of $`f_{\mathrm{disk}}`$ are unphysical. Those results are qualitatively very similar to the results obtained by WCŻ97 which were based on approximate description of the disk/corona equilibrium through the condition on the value of the ionization parameter $`\mathrm{\Xi }`$ (shown in Fig. 4 as a dashed line).
Since the corona is not isothermal there is no single value of the ion and electron temperature for a given radius. However, the representative values are those at the top of the transition zone, at $`P_{\mathrm{base}}`$. We plot them also as functions of the accretion rate (Figs. 5). We see that the trends are similar to those determined by WCŻ97 but there are quantitative differences. The ion temperature is always very close to the virial temperature and, at that radius, it never approaches the electron temperature. Finally, we plot the optical depth of the corona at the same radius (Fig. 6). The corona is much thinner than predicted by WCŻ97, its optical depth strongly depends on the accretion rate.
#### 3.1.2 Radial trends
All quantities depend not only on the accretion rate but also on the disk radius. Therefore we study now the radial dependences of the coronal parameters for a representative value of the accretion rate $`\dot{m}=0.05`$.
There is an essential, qualitative change with respect to the results obtained by WCŻ97.
In WCŻ97 the strength of the corona increases outwards: corona is weak close the black hole while all energy is generated within the corona at a certain critical distance $`r_{\mathrm{max}}`$. Beyond this radius the disk/corona configuration does not exists. According to our new description of the disk/corona transition under no evaporation condition the solutions at radii larger than a certain $`r_{\mathrm{max}}`$ also do not exists but at a radius only marginally smaller than $`r_{\mathrm{max}}`$ the fraction of energy generated in the corona is not 1, but about 0.5 (Fig. 7). In a very narrow radial region below $`r_{\mathrm{max}}`$ two solutions exist: one with weaker corona and the other one with strong corona, which merge at $`r_{\mathrm{max}}`$. The strong corona solution reaches the value $`f_{\mathrm{cor}}=1`$ at a radius not much lower than $`r_{\mathrm{max}}`$ and the solution does not continue down to smaller radii (mathematically, it exists as unphysical solution with $`f_{\mathrm{cor}}>1`$). Therefore this branch of solution does not offer a global solution for the corona. On the other hand the weak corona branch continues down to the marginally stable orbit and this branch represents the solution acceptable under adopted conditions (no evaporation/condensation). The nature of this multiple branch solutions is connected with the presence of advection, as studied for the case of WCŻ97 condition of disk/corona transition by Janiuk, Życki & Czerny (2000). The advection dominated branch tends to vanish if the solution is computed more accurately by determination of the coefficient $`\delta `$ iteratively from the radial derivatives of physical quantities.
Therefore, according to our model, the corona in equilibrium with the disk covers only a finite part of the disk, its strength increases outwards but it never dissipates more than about a half of the gravitational energy available at a given radius (Fig. 8). The transition from outer part without a coronal solution to the inner part with corona is sharp if studied within the frame of solutions with no disk/corona mass exchange and the true nature of this transition can be understood only within the frame of the solutions with evaporation (see Section 3.2).
The ion temperature at the top of the transition zone is always quite close to the local virial temperature so it clearly decreases with the radius (Fig. 9). On the other hand the electron temperature at $`P_{\mathrm{base}}`$ increases outwards on the basic branch, as in WCŻ97. Its value at the outer edge of the coronal solution is of the same order as predicted by WCŻ97 but closer to the central object is much cooler.
The optical depth of the corona is not constant, as in WCŻ97, but it strongly decreases inwards along the basic branch (Fig. 10). Therefore, the effect of Comptonization apart from the very outer edge of the coronal solutions is completely negligible as long as the inner disk is not evaporated (see Section 3.2). In this paper we neglect Compton heating, since the accretion is a main source of heating. However, Compton heating may increase the optical depth of the base of the corona and we plan to study this effect in a future work.
The radial trends change with the accretion rate is a way similar to that described by WCŻ97. The extension of the corona, $`r_{\mathrm{max}}`$, is large for large accretion rate and decreases with decreasing accretion rate. For accretion rates below $`\dot{m}=1.68\times 10^4`$ no equilibrium solutions are obtained. Again, the nature of this phenomenon can be understood by analyzing the solutions which allow for mass exchange in the disk/corona system (see Section 3.2).
### 3.2 Solutions with evaporation/condensation
Construction of those solutions is more complex as the solutions have global character, i.e. the results at a given radius are coupled to those at other radii through the continuity equation for the corona (Equation 17). Since we follow semi-analytical approach to solution of Eq. 17 we first present local numerical solutions which are parameterized by the coronal accretion rate, $`\dot{m}_{\mathrm{cor}}`$, in addition to global parameters $`M`$, $`\dot{m}`$ and $`\alpha `$, and we give the analytical formulae approximating the relation between $`\dot{m}_{\mathrm{cor}}`$ and $`\dot{m}_\mathrm{z}`$.
The vertical structure of the disk/corona system with evaporation is qualitatively similar to the case without mass exchange. However, the global trends are essentially different so in this section we concentrate on those aspects of the model.
In Fig. 11 we show the dependence of the evaporation rate of the cool disk on the accretion rate in the corona at $`10R_{\mathrm{Schw}}`$ for a small value of the total accretion rate ($`\dot{m}=5\times 10^4`$). From Fig. 4 we see that we should not expect an equilibrium solution without evaporation so it is not surprising that the evaporation rate $`\dot{m}_\mathrm{z}`$ is positive for all values of the accretion rate within the corona from 0 to the total accretion flux.
The relation is well approximated by an analytical formula
$$\dot{m}_\mathrm{z}=1.27\times 10^2\dot{m}_{\mathrm{cor}}^{5/3}R^{3/2}\alpha _{0.1}^{7/2}M_{\mathrm{BH8}}^1[\mathrm{gs}^1\mathrm{cm}^2].$$
(18)
Here, viscosity is described in dimensionless units: $`\alpha _{0.1}=\alpha /0.1`$. This formula neglects the effect of condensation and it is a good approximation of the solution only for very low accretion rates $`\dot{m}`$. Applicability of this asymptotic expression will be qualitatively discussed at the end of this Section.
Such an analytical expression can be now conveniently used to determine the global solution by substituting it to the continuity equation 17.
The radial dependence of the coronal accretion rate and evaporation rate are therefore:
$$\dot{m}_{\mathrm{cor}}=3.09\alpha _{0.1}^{}{}_{}{}^{21/4}R^{3/4}$$
(19)
$$\dot{m}_\mathrm{z}=8.23\times 10^2M_{\mathrm{BH8}}^1\alpha _{0.1}^{21/4}R^{11/4}.$$
(20)
These formulae are qualitatively similar to the formulae derived by MMH94 in the case of a single-temperature plasma and no condensation condition at the basis of the corona. The fraction of mass carried by the corona increases inwards. Finally, at a certain radius $`R_{\mathrm{evap}}`$ all the mass is in the coronal flow due to the complete disk evaporation: $`\dot{m}=\dot{m}_{\mathrm{cor}}`$ (Fig. 7 long dashed line). Its value
$$R_{\mathrm{evap}}=4.51\alpha _{0.1}^{}{}_{}{}^{7}\dot{m}^{4/3}$$
(21)
is surprisingly similar to the value determined by Liu et al. (1999) on the basis of Mayer & Mayer-Hoffmeister’s results. The only essential difference is the dependence on the viscosity present in our approach but absent in theirs. This is basically the result of our different description of the conduction flux close to the disk surface and the presence of the additional step in energy transfer in the form of ion-electron coupling which results in the electron temperature much lower than the virial temperature, depending also on the viscosity.
Assuming the viscosity parameter $`\alpha =0.1`$ we obtain practically the same results as Liu et al. (1999) but larger (smaller) viscosity in our case result in larger (smaller) purely ’coronal’ region. The dependence on the accretion rate is almost identical in both cases (see also Section 4.1.2).
We now proceed to the case of large accretion rate. From Fig. 4 we see that for large accretion rate $`\dot{m}=0.05`$ at 10 $`R_{\mathrm{Schw}}`$ there is an ’static’ equilibrium solution so generally we can expect either evaporation, or condensation, depending on the value of the accretion rate within the corona itself. Fig. 12 confirms those expectations. For extremely low values of coronal accretion rate $`\dot{m}_\mathrm{z}`$ is still well approximated by the analytical formula (Eq. 18) but a turnover appears at larger values and finally evaporation is replaced with condensation, as $`\dot{m}_\mathrm{z}`$ becomes negative.
Therefore we need to replace Equation 18 with more general formula able to represent both evaporation and condensation phenomenon.
$`\dot{m}_\mathrm{z}=1.27\times 10^2\dot{m}_{\mathrm{cor}}^{}{}_{}{}^{5/3}R^{3/2}\alpha _{0.1}^{}{}_{}{}^{7/2}M_{\mathrm{BH8}}^1\times [1`$ (22)
$`1\times 10^{42}\dot{m}^{10}\dot{m}_{\mathrm{cor}}^{}{}_{}{}^{5/2}R^{12.5}\alpha _{0.1}^{}{}_{}{}^{8.5}][\mathrm{gs}^1\mathrm{cm}^2].`$
This formula is oversimplified as it does not take into account the non-monotonic two-value dependence of $`f_{\mathrm{cor}}`$ on the radius for models in equilibrium, for which $`\dot{m}_\mathrm{z}=0`$ (see Fig. 8). However, it reproduces accurately enough the extension of the strong condensation zone. More accurate description would require numerical computations of the vertical structure of the corona together with integration of Eq. 17.
Previously found global solution based solely on evaporation applies everywhere in the disk down to the marginally stable orbit if the second term in Eq. 22 is smaller than 1, i.e. when the total accretion rate is below the limiting value $`\dot{m}=1.31\times 10^3\alpha _{0.1}^{17/25}`$ for $`R=3`$. In objects with such a low accretion rate the corona formation proceeds monotonically, corona carry an increasing fraction of mass as the flow approaches a black hole and finally, at radius given by Eq. 21 the disk disappears, as it is replaced by an optically thin flow of ADAF type.
In objects with significantly higher accretion rate the situation is more complex than in low accretion rate case. The equations 22 and 17 have to be now solved numerically. An example is shown in Fig. 13 (thick continuous line). At the outermost part of the disk the corona formation proceeds according to Equation 19, as before. Closer in the cooling becomes more efficient and the corona formation proceeds more slowly than predicted by Equation 19. When we approach the radius for which an equilibrium solution exists (i.e. possible balance between the disk and the corona without evaporation, see Sect. 3.1) the corona reaches its maximum strength. In the innermost part of the disk the corona strength decreases as the evaporation is replaced by condensation. The radial dependence of the fraction of energy generated in the corona in this region is qualitatively similar to the result given under assumption of no mass exchange between the disk and the corona. Trends are the same as those given by WCŻ97 although the actual radial dependence is much steeper.
A sequence of solutions showing the radial dependence of the fraction of the energy generated in the corona for several values of the accretion rate is shown in Fig. 14. For significantly higher accretion rates the maximum strength of corona never approaches the value of 1, so the two-phase flow extends to the marginally stable orbit. When the total accretion rate is intermediate the evaporation is efficient enough to cause a transition to an optically thin flow ($`f_{\mathrm{cor}}=1`$) at a radius approximately given by Eq. 21. However, in the innermost region the condensation is efficient if there is large soft photon flux available to cool electrons through Comptonization. We expect that it might lead to cool clump formation or even a reconstruction of the disk in this region. Such a secondary disk rebuilding is now present in our model for accretion rates higher than $`\dot{m}^{\mathrm{min}}`$. More accurate numerical results for an accretion rate which leads to an ADAF type flow in the innermost part give the value
$$\dot{m}^{\mathrm{min}}=1.9\times 10^3\alpha _{0.1}^{17/25}.$$
(23)
However, the current version of our model does not allow to describe yet the details of the transition from optically thick disk through optically thin disk to complete disappearance of the disk flow and eventual reverse of this process in the innermost part of the flow. Therefore, at present, we usually assume that once the transition occurs from the flow from disk/corona geometry to an optically thin flow at a certain radius it means that there is no disk flow closer to a black hole. However, we also discuss an opposite case in Section 3.3.
Further decrease of the accretion rate results in a smooth expansion of the ADAF zone. In all other models the first appearance of the single phase hot flow is expected at the marginally stable orbit (e.g. Esin et al. 1997). In our model the disk either extends to the marginally stable orbit or down to a radius larger than the minimum radius dependent on the viscosity, with no solutions for the transition radius ever expected between these two radii. This property of our solutions is caused by the domination of the condensation over evaporation in the innermost part of the disk where the pressure is large.
The dependence of the transition radius on the accretion rate is shown in Fig. 15. We see that for low accretion rates the position of the transition radius depends on the accretion rate and the relation is well described by Eq. 21, as expected. However, at a certain value of the accretion rate the condensation takes over and the coronal dissipation decreases in the innermost part of the disk enough to prevent the transition to an ADAF flow. This critical accretion rate $`\dot{m}_{\mathrm{evap}.\mathrm{branch}}`$ is well described by the two analytical conditions: Eq. 21 and $`\dot{m}_\mathrm{z}=0`$ (see Eq. 22) which leads to
$$\dot{m}_{\mathrm{evap}.\mathrm{branch}}=6.92\times 10^2\alpha _{0.1}^{3.3}.$$
(24)
This accretion marks the formation of the evaporation branch of the disk/corona solution which is essential for explanation of specific luminosity states of accreting black holes and time-dependent behaviour of X-ray sources (see also Sec. 3.3).
However, the essential point which directly results from radial computations is the understanding of the inner region without disk/corona solutions. What WCŻ97 interpreted as a ’bare’ disk is actually a ’bare corona’, or ADAF type of flow.
Determination of the coronal accretion rate as a function of the total accretion rate is a starting point of the discussion of the stability of solution and eventual time-dependent evolution.
### 3.3 Expected time-dependent behaviour
#### 3.3.1 Surface density of the accretion flow
The stability and time evolution of the accretion flow is most conveniently discussed on the plot showing the accretion rate, $`\dot{m}`$, and the surface density of the flow, $`\mathrm{\Sigma }`$, at a representative radius.
We calculate the fraction of the energy generated in the corona at by numerical integration of Eq. 17, assuming the analytical formula for the local evaporation/condensation rate given by Eq. 22. Knowing the corona properties we solve the cold disk structure at this radius, with the appropriate boundary conditions (see Sec. 2.4).
Generally, for $`\dot{m}>\dot{m}_{\mathrm{evap}.\mathrm{branch}}`$ the strength of corona is small and the solution is practically the same like in the thin disc without corona. However, when $`\dot{m}`$ approaches $`\dot{m}_{\mathrm{evap}.\mathrm{branch}}`$ this is no longer true and the solution starts to depart horizontally from standard $`\mathrm{log}\dot{m}`$$`\mathrm{log}\mathrm{\Sigma }`$ curve (Fig. 16). The plot depends on a distance from the black hole, i.e. whether it is higher or lower than the critical radius for which the curve in Fig. 14 has maximum at $`f_{\mathrm{cor}}=1`$. The critical radius can be determined analytically by combining Eq. 21 and Eq. 24 which gives
$$R_{\mathrm{evac}}=158.7\alpha _{0.1}^{2.6}.$$
(25)
For $`R>R_{\mathrm{evac}}`$, $`f_{\mathrm{cor}}`$ increases with decreasing $`\dot{m}`$, and finally $`f_{\mathrm{cor}}=1`$ for a certain value of accretion rate $`\dot{m}_{\mathrm{evap}}`$ smaller then $`m_{\mathrm{evap}.\mathrm{branch}}`$. As $`f_{\mathrm{cor}}1`$, the disc surface density drops with $`\dot{m}\dot{m}_{\mathrm{evap}}`$.
For $`R<R_{\mathrm{evac}}`$ the plot depends on the assumption of considering or rejecting the solutions representing the secondary disc condensation (see dashed line in Fig. 14 and Sec. 3.2). In Fig. 16 we show the relation between the accretion rate and the surface density for $`r=10R_{\mathrm{Schw}}`$ and $`\alpha =0.1`$.
If we assume that once the disk is evaporated it never forms again closer to the black hole, then the transition radius between the disk/corona and single-phase ADAF, suddenly drops to the marginally stable orbit. In such case, when decreasing accretion rate attains the value of $`m_{\mathrm{evap}.\mathrm{branch}}`$, the discontinuous jump in the corona strength occurs from $`f_{\mathrm{cor}}<1`$ to $`f_{\mathrm{cor}}=1`$. This sharp transition is marked as a horizontal line called evaporation branch. In time-dependent solution the disc structure may actually follow this branch if $`\dot{m}`$ decreases suddenly.
On the other hand, if we allow for existence of the reconstructed branch of the disk at the innermost part of the disk, whenever it is allowed, the solution enters the evaporation branch gradually, it is shallow and positioned at much lower accretion rate (long dashed line). However, such a solution has ’an ADAF hole’ for a certain range of radii. In further considerations we adopt the first solution as the correct one since the inner secondary disk formation is not consistently described within the frame of our model and may not represent a physically acceptable solution.
The mechanism of disk evaporation leads to departure from standard optically thick solution in the form of practically horizontal branch which develops at intermediate accretion rates. For $`\alpha =0.1`$ the position of this branch is approximately at the same value of the accretion rate as the maximum of the accretion rate at the optically thin branch, as described by Narayan & Yi (1995b), Abramowicz et al. (1995) and Chen et al. (1995), but this is not the case for other values of $`\alpha `$ since the optically thin branch was computed for a fixed assumed value of the Compton parameter characterizing the cooling of the optically thin flow. Correct solution should be obtained taking into account the actual soft X-ray flux coming from outer parts of the disk and should be a natural extension of the two-phase solution in the case of cold disk absence. At present we cannot extend our horizontal branch to the low surface density range and to reproduce the solution for a single phase medium since the present model is based on assumption that the underlying disk is optically thick and absorbs half of the X-ray radiation emitted by the corona at a given radius and reemits this radiation in the form of soft flux. The description of the transition from a two-phase flow to a single-phase optically thin flow would require relaxation of this assumption and a number of other modifications, like non-local description of the soft photon radiation field since the outermost part of the ADAF flow will be predominantly cooled by soft photons emitted by the outer parts (e.g. Esin et al. 1997) and/or soft photons originating from synchrotron radiation. We plan to study this problem in the future, in order to give a complete view of the transition between those two branches.
The position of evaporation branch in our model depends on the viscosity - lower viscosity parameter $`\alpha `$ corresponds to lower accretion rate at this branch. For $`\alpha =0.1`$ and large black hole mass it joins the (almost) standard Shakura-Sunyaev branch just below the turning point marking the gas pressure dominated branch. Therefore, the transition practically happens from unstable radiation pressure dominated branch to evaporation branch with a very narrow range of accretion rates in between. Adopting lower value of viscosity, however, broaden this range considerably. For low mass of the black hole the stable gas dominated branch exists even for $`\alpha =0.1`$ since the position of the transition from radiation pressure to gas pressure dominated solution in Shakura-Sunyaev disk depends on the mass of the c entral object.
As seen from Fig. 15, the position of the evaporation branch does not depend on radius in the innermost part of the disk. At $`R<R_{\mathrm{evac}}`$ the position of the evaporation branch is given by Eq. 24 while for $`R>R_{\mathrm{evac}}`$ this position is determined by Eq. 21. This property has important consequences for the limit cycle evolution expected for a range of accretion rates corresponding to the unstable radiation pressure dominated branch of thick disk solution.
#### 3.3.2 Stationary solutions and outbursts
This diagram (Fig. 16), although not complete, shows the importance of the evaporation branch for stationary solutions and time-dependent evolution of accretion disk. If the accretion rate is larger than the Eddington limit the model is located on the stable optically thick ADAF branch. Stationary solution well represents an accretion process at this parameter range. When the accretion rate is very low, below the evaporation branch, the flow is described by the optically thin stable ADAF solution in the innermost part, and a disk/corona flow in the outer part. The solution should be also well represented by a stationary model,unless the disk/corona interaction lead to some instabilities.
However, for intermediate accretion rates, the disk/corona flow, as the classical radiation pressure dominated model, may display a complex limit cycle, as already studied, without taking into account the evaporation branch, by Szuszkiewicz & Miller (1998), Honma, Matsumoto & Kato (1991) and Janiuk et al. (2000). The solution would oscillate between the upper branch during the eruption and and the evaporation branch between the outbursts. Actual position of the upper branch may be at lower accretion rates, if the strong outflow in addition to radial advection, is included (e.g. Nayakshin, Rappaport & Melia 1999, Janiuk et al. 2000). The timescale for such evolution is shorter than the viscous timescale of the standard Shakura-Sunyaev stationary disk at this radius because of the reduced surface density. The exact prediction of the global outburst involves the entire unstable part of the disk so, without actual computations, the exact timescale and the amplitude cannot be determined but order of magnitude timescales involved are seconds for galactic sources and months/years for AGN.
The character of this evolution would depend significantly on the radial extension of the instability zone. If this extension is below $`R_{\mathrm{evac}}`$ the position of the evaporation branch is the same for all radii and the entire disk will oscillate between the upper (advection or outflow) dominated branch and the evaporation branch so the optically thick disk will be present during the entire cycle. However, if the radial extension is larger than $`R_{\mathrm{evac}}`$ the evaporation branch at the outermost radius involved in the outburst is lower than evaporation branch inside. The accretion rate in low luminosity phase of the outburst is then too low for optically thick disk in the innermost part, the transition to ADAF takes place so the optically thick disk will be temporarily evacuated during such a cycle. It is interesting to note that probably both kinds of outbursts are seen in the case of microquasar GRS 1915+105, as two distinct kind of spectral behaviour and of the relation between the length of outburst and the frequency of QPO phenomenon are observed (Trudolyubov, Churazov & Gilfanov 1999). Also the observed X-ray delay in the outburst of soft X-ray transient GRO J1655-40 shows that most probably the system made the transition from an inner ADAF/outer disc/corona state to a pure disc/corona state Hameury et al. (1997). The same transition is seen in Cyg X-1 (Esin et al. 1997).
The expected amplitude of the outburst is considerably reduced by the presence of the evaporation branch - expected luminosity changes are by an order of magnitude while, if the gas pressure dominated branch is the lower branch of the limit cycle the outburst of up to four orders of magnitude would be expected, against any observational evidence for the timescales involved.
According to our model, there are strong systematic differences expected between the AGN and GBH since the range of the radiation pressure instability depends on the central mass while the position of the evaporation branch does not. If the evaporation branch is positioned significantly below the radiation pressure dominated branch we can also expect a stable accretion flow in the accretion rate range corresponding to the gas pressure dominated solution. It means that, independent on the viscosity, the range of accretion rates corresponding to stable, gas dominated solution is always broader for galactic black holes than for AGN. In particular. no such range at all is predicted in AGN for $`\alpha =0.1`$ since this solution is entirely replaced by evaporation branch and ADAF.
## 4 Discussion
### 4.1 Comparison to other models
#### 4.1.1 Underlying assumptions
Our approach to modeling the disk/corona structure follows the MMH94 attitude. We also model the energy dissipation within a corona as due to the accretion flow, we take the transition between the disk and the corona which allows to determine the fraction of energy generated in the corona as a function of radius as well as the transition radius to the optically thin flow as a function of the mass of the black hole, accretion rate and the viscosity parameter in the corona.
There are, however, major important differences between our model and MMH94: (i) our corona is basically a two-temperature medium with energy dissipation due to ion collisions and the Coulomb transfer of energy from ions to electrons, as in SLE and ADAF models (ii) apart from bremsstrahlung we allow also for Compton cooling (iii) we determine the conductive flux at the basis of the corona for the adopted heating/cooling mechanism instead of using the scaled prediction from solar corona. This last modification is responsible for the dependence of the accretion rate in the corona on the viscosity parameter in the case of our model and the lack of such dependence in MMH94 and subsequent papers (Liu, Meyer & Meyer-Hofmeister 1995, Liu et al. 1999, Meyer, Liu & Meyer-Hofmeister 2000). The predicted dependence of the coronal accretion rate on the radius is qualitatively similar in both papers in the case of low accretion rate but not in the case of large accretion rate when the condensation, neglected by MMH94, is important. Our approach to conduction is therefore the same as of Dullemond (1999) but he did not consider a two-temperature plasma and Compton cooling.
Two-temperature accreting corona was discussed by WCŻ97 but their condition of disk/corona transition based on analogy with Compton heating medium was not accurate numerically and missed an important insight on corona formation in the innermost and outermost parts of the flow.
#### 4.1.2 Transition radius to ADAF-type flow
The transition from an outer optically thick disk flow to an inner optically thin ADAF was predicted by a number of models. In some models the transition was based on ADAF principle (i.e. the transition happens whenever an ADAF solution is possible), but in other cases the necessity of the transition resulted from the model itself (MMH94 and applications, our model). It is therefore interesting to compare quantitatively those estimates as they may in principle be estimated observationally.
Condition based on the strong ADAF principle, i.e. on the absence of ADAF solutions:
$$r_{\mathrm{tr}}=1.9\times 10^4\dot{m}^2\alpha _{0.1}^4R_{\mathrm{Schw}}$$
(26)
(Abramowicz et al. 1995, Honma 1996, Kato & Nakamura 1998).
Approximation to the results of Esin et al. (1997) for $`\alpha =0.25`$ and $`\beta =0.5`$:
$$\begin{array}{cc}r_{\mathrm{tr}}=\times 10^4R_{\mathrm{Schw}}\hfill & \dot{m}<0.084\\ r_{\mathrm{tr}}=30R_{\mathrm{Schw}}\hfill & \dot{m}0.084\\ r_{\mathrm{tr}}=330R_{\mathrm{Schw}}\hfill & 0.084<\dot{m}<0.092\\ r_{\mathrm{tr}}=3R_{\mathrm{Schw}}\hfill & \dot{m}>0.092\end{array}$$
(27)
Liu et al. (1999) gives:
$$r_{\mathrm{tr}}=18.3\dot{m}^{1/1.17}R_{\mathrm{Schw}}$$
(28)
Our results:
$$\begin{array}{cc}r_{\mathrm{tr}}=4.51\dot{m}^{1.33}\alpha _{0.1}^7R_{\mathrm{Schw}}\hfill & \dot{m}<6.92\times 10^2\alpha _{0.1}^{3.3}\hfill \\ r_{\mathrm{tr}}=3R_{\mathrm{Schw}}\hfill & \dot{m}>6.92\times 10^2\alpha _{0.1}^{3.3}\hfill \end{array}$$
(29)
All those conditions are roughly in agreement with observational requirement that transition between the high and low state happens at luminosity $`10^{37}`$ erg s<sup>-1</sup> (Tanaka 1999). The accurate dependence of the transition radius on the accretion rate can in principle be determined from the shape of the reflected component (Done & Życki 1999) in X-ray novae but the errors are considerable and the presence of the hot Compton heated skin on the disk surface further complicates the interpretation of the data. However, high precision spectroscopy from new X-ray satellites should give stronger constraints on the available models.
### 4.2 Observational consequences
#### 4.2.1 Luminosity states
The transition from optically thick flow to ADAF, i.e. the accretion rate at the evaporation branch marks the transition between the hard state and soft state in accreting binaries, according to a number of suggestions (e.g. Narayan 1996, Esin et al. 1997, Zdziarski, Lubiński & Smith 1999). When the accretion rate is high enough and two-phase flow exists up to the marginally stable orbit then the soft flux is always dominant, cooling of corona is efficient and produces soft X-ray spectrum. Such a solution is stable in GBH for moderate accretion rates and may correspond to the soft (high) state. If the accretion rate is very high the disk is unstable and evolving between the evaporation branch and advection-dominated optically thick branch - such a parameter range may be appropriate for a Very High State, with its complex spectral behaviour (e.g. Takizawa et al. 1997, Revnivtsev et al. 2000). On the other hand, when the accretion rate is very low, the innermost part of the disk is an optically thin ADAF, dominating the spectrum, so such a case corresponds to the Low (or Hard) State.
In our model, the position of horizontal evaporation branch does not depends on the mass of the black hole (Fig. 17). Nevertheless the turning point between the unstable radiation pressure branch and the stable evaporation branch is different for AGNs and GBHs and the possibility of formation of the High State in AGN is strongly suppressed.
Since most of the energy generates at about 10 $`R_{\mathrm{Schw}}`$ we can roughly identify the accretion rate ranges for various luminosity states on the basis of Fig 17. Assuming $`\alpha =0.1`$ we obtain:
$$\begin{array}{ccc}State\hfill & GBH& AGN\\ VeryHighState\hfill & \dot{m}>0.1& \dot{m}>0.07\\ HighState\hfill & 0.07<\dot{m}<0.1& \\ LowState\hfill & \dot{m}<0.07& \dot{m}<0.07\end{array}$$
(30)
However, we stress that those limits depend significantly on the adopted value of the viscosity parameter $`\alpha `$.
#### 4.2.2 The nature of the Intermediate State in GBH
The transition of X-ray novae from soft state with the dominant disk emission to a hard state with the dominant thin plasma (e.g. ADAF) contribution proceeds through the Intermediate State. The analysis of the Nova Musca revealed that in the intermediate state the reflection component is seen with an amplitude of order of 1 while the kinematical determination of the inner disk radius from Doppler shifts of the reflected component show disk disruption (Życki, Done & Smith 1998). We can interpret such a state within the frame of our model as corresponding to an accretion rate for which a strong corona forms at intermediate radii. In this case the reflection is strongly biased towards larger radii instead of towards innermost disk radius. Also very high values of the ionization parameter may be consistent with the irradiating source being directly above an accretion disk. Our interpretation is therefore different from that of Esin et al. (1997) since in their approach the intermediate state corresponds to a disrupted disk with an innermost ADAF flow instead of the two-phase disk/corona flow.
#### 4.2.3 Variability
The lack of High State in AGN, as predicted by our model for the viscosity parameter $`\alpha =0.1`$ may offer an explanation why the observed variability properties of AGN and GBH are not the same. GBH, when in their High State, dominated by disk emission, are only weakly variable in X-rays in the timescales of about a second, with rms amplitude of order of a percent (e.g. Takizawa et al. 1997). Spectrally, such states are basically similar to quasar spectra. However, quasars are strongly variable, with rms amplitude in the optical band dominated by the disk emission is of order of 15 per cent (e.g. Giveon et al. 1999) in the timescales of years, so the simple scaling with the mass of the black hole does not seem to hold. It may mean that quasars are actually counterparts of the GBH in their Very High State. Since the spectra of GBH in Very High State are quite complex and not always dominated by the disk emission we can also expect such weak disk spectral states among the quasar/Seyfert class. Well studies source MCG-6-15-30 (Tanaka et al. 1995 and subsequent papers) may be an example, due to its exceptionally low mass and consequently high luminosity to the Eddington luminosity ratio (Nowak & Chiang 1999).
#### 4.2.4 Broad Line Region and LINERs
Nicastro (2000) proposed a very interesting model of the Broad Line Region based on the disk/corona model of WCŻ97. The basic idea is that the strong outflow develops at radii where the corona strength reaches its maximum and cool clouds forming in this disk wind are responsible for the emission of broad lines. Predictions based on our new model does not alter presented estimates significantly. The radius of the maximum corona strength in the new model can be determined from the condition $`\dot{m}_\mathrm{z}=0`$ (see Eq. 22), using the result with negligible evaporation to determine the local coronal accretion rate (Eq. 19) which gives the dependence
$$R_{\mathrm{max}}=1.02\times 10^3\dot{m}^{16/23}\alpha _{0.1}^{37/115},$$
(31)
smaller by a factor of few than the WCŻ97 formula (see Eq. 3 of Nicastro 2000). On the other hand the new corona efficiency is now roughly symmetric around this point so the weighted formula for the outflow used by Nicastro would not decrease this radius, as it did in the case of sharp edge corona of WCŻ97 so the net effect will be roughly the same as before.
It was suggested by Lasota et al. (1996), that LINERs may represent the systems with ADAFs. The transition from objects with broad emission lines to LINERs happens, according to Nicastro (2000), due to the switch off the outflow if the accretion rate is too low for the existence of the radiation pressure dominated zone. The most efficient part of the corona, however, is located in the gas pressure dominated region so this connection is not clear. On the other hand, for very small accretion rate, within the frame of our new model, there is a transition from secondary disk rebuilding phase to a definite ADAF flow in the innermost part (see Eq. 23). Such a change may be important from the point of view of the radiative cooling, the efficiency of advection and, consequently, the mass outflow. The mass loss through the wind as a function of radius is not calculated yet in the present model but this can be done in the future, following the method of e.g. Meyer & Hoffmeister (1994) and WCŻ97.
###### Acknowledgements.
We are grateful to Drs R. Narayan and A. Esin, for their helpful comments and suggestions on the original version of our paper. We thank A. Zdziarski for his computer program generating optically thin solutions used in Fig. 16 and to Piotr Życki for helpful discussions. Part of this work was supported by grant 2P03D01816 of the Polish State Committee for Scientific Research.
## Appendix A Evaporation/condensation rate at the basis of the corona
The electron temperature profile within the transition zone is determined by the effect of heating, radiative cooling, conduction and the vertical motion of the material (e.g. Begelman & McKee 1990, McKee & Begelman 1990, MMH94). The general form of this equation in a plane parallel geometry is the following
$$\frac{d}{dz}\left[\rho v_\mathrm{z}\frac{5}{2}\frac{kT_\mathrm{i}}{m_\mathrm{H}}\right]+\frac{dF_{\mathrm{cond}}}{dz}=Q_\mathrm{e}^+\mathrm{\Lambda },$$
(32)
where the conduction flux due to electrons is given by
$$F_{\mathrm{cond}}=\kappa _\mathrm{o}T_\mathrm{e}^{5/2}\frac{dT_\mathrm{e}}{dz}$$
(33)
Here we included the effect of radial advection into the cooling term and we neglected the conduction due to ions.
Following McKee & Begelman (1990) we multiply the Equation 32 by $`F_{\mathrm{cond}}`$ and integrate in the vertical direction across the transition zone:
$`{\displaystyle _{z_1}^{z_2}}{\displaystyle \frac{d}{dz}}\left[\rho v_\mathrm{z}{\displaystyle \frac{5}{2}}{\displaystyle \frac{kT_\mathrm{i}}{m_\mathrm{H}}}\right]F_{\mathrm{cond}}𝑑z+{\displaystyle \frac{1}{2}}\left[F_{\mathrm{cond}}^2(z_2)F_{\mathrm{cond}}^2(z_1)\right]=`$ (34)
$`={\displaystyle _{T_\mathrm{s}}^{T_\mathrm{e}(P)}}[Q_\mathrm{e}^+\mathrm{\Lambda }]\kappa _\mathrm{o}T_\mathrm{e}^{5/2}𝑑T_\mathrm{e}`$
The conduction flux at the upper and lower point of the transition zone vanishes which is used as a boundary condition in full numerical computations (e.g. Dullemond 1999) so the second term on the left hand side of the equation vanishes thus leaving a relation between the mass transfer (first term) and the integral on the right hand side.
We assume that the transition from hot coronal to cold disk material in the vertical direction takes place in a very narrow geometrical zone so any change of the pressure across this zone can be neglected. This allows to calculate the right hand side integral conveniently, using algebraic expressions for the density, ion temperature and the heating and cooling as functions of the electron temperature.
The left hand side can be approximately expressed by estimating first the Field length (Field 1965) measuring the thickness of the transition zone
$`\lambda _\mathrm{F}`$ $`=`$ $`\left({\displaystyle \frac{\kappa _\mathrm{o}T_\mathrm{e}^{7/2}(P)}{<|Q_\mathrm{e}^+\mathrm{\Lambda }>}}\right)^{1/2}`$ (35)
$``$ $`{\displaystyle \frac{\kappa _\mathrm{o}T_\mathrm{e}^{7/2}(P)}{|_{T_\mathrm{s}}^{T_\mathrm{e}(P)}[Q_\mathrm{e}^+\mathrm{\Lambda }]\kappa _\mathrm{o}T_\mathrm{e}^{5/2}𝑑T_\mathrm{e}|^{1/2}}}.`$
We now estimate the typical value of the conductive flux within the transition zone as
$$<F_{\mathrm{cond}}>=\kappa _\mathrm{o}T_\mathrm{e}(P)^{7/2}/\lambda _\mathrm{F}$$
(36)
which allows us to express the left hand side term as
$$_{z_1}^{z_2}\frac{d}{dz}\left[\rho v_\mathrm{z}\frac{5}{2}\frac{kT_\mathrm{i}}{m_\mathrm{H}}\right]F_{\mathrm{cond}}𝑑z<F_{\mathrm{cond}}>\dot{m}_\mathrm{z}\frac{5}{2}\frac{kT_\mathrm{i}(P)}{m_\mathrm{H}}.$$
(37)
Therefore the approximate formula for the condensation/evaporation rate at the disk/corona boundary reads
$$\dot{m}_\mathrm{z}=\frac{_{T_\mathrm{s}}^{T_\mathrm{e}(P)}[Q_\mathrm{e}^+\mathrm{\Lambda }]\kappa _\mathrm{o}T_\mathrm{e}^{5/2}𝑑T_\mathrm{e}}{\kappa _\mathrm{o}T_\mathrm{e}(P)^{7/2}}\lambda _\mathrm{F}\left(\frac{2}{5}\frac{m_\mathrm{H}}{kT_\mathrm{i}(P)}\right)$$
(38)
|
warning/0004/cond-mat0004427.html
|
ar5iv
|
text
|
# Electronic and Structural Properties of C36 Molecule
## I INTRODUCTION
Recently, a new member of fullerenes, C<sub>36</sub>, was synthesized by arc-discharge method and purified in bulk quantities. Up to now, it is the smallest fullerene ever discovered. The C<sub>36</sub> molecule is more curved than C<sub>60</sub> because of the adjacent pentagonal rings and the small number of carbon atoms. This feature suggests stronger electron-phonon interaction and possible higher superconducting transition temperatures in the alkali-doped C<sub>36</sub> solids than those of C<sub>60</sub>. The Solid-State Nuclear Magnetic Resonant experiment suggested that the most favorable configuration of C<sub>36</sub> molecule has $`D_{6h}`$ symmetry . This confirmed the results of the early ab initio pseudopotential density functional calculations, which indicated that the $`D_{6h}`$ structure is one of the most energetically favorable structures among several possibilities.
Looked in a direction perpendicular to the six-fold axis(Fig.1), the molecule is composed of six parallel layers. On each layer, six carbon atoms lie at the vertexes of a hexagon. The edges of the hexagons of the 1st, 2nd, 5th, 6th layers are parallel, while those of the two hexagons in the middle two layers are turned by an angle of $`30^{}`$. In addition to this special structure, the six-fold principal axis is distinct form the ordinary five-fold principal axis in the cases of C<sub>60</sub> and C<sub>70</sub> and little attention has been paid to it. In this paper, we discuss both the electronic and structural properties of C<sub>36</sub> molecule, emphasizing the interesting properties brought by this unusual $`D_{6h}`$ symmetry.
We employ a simple and elegant model , extended SSH model, which is successful in treating C<sub>60</sub> and C<sub>70</sub>. The BdeG formalism is performed to obtain the stable lattice configuration and the corresponding electronic states self-consistently. The electronic part of Hamiltonian without Coulomb interaction can be solved analytically by methods of group theory. We find that the electron densities of HOMO($`B_{1u}`$) and LUMO($`B_{2g}`$) are zero at the two middle layers. Furthermore, the HOMO and LUMO are mainly confined in the 2nd and 5th layers with a possibility of $`90\%`$. The gap between them is considerably small, compared to that of C<sub>60</sub> and C<sub>70</sub>, because the splitting of HOMO and LUMO is due to a long-distance hopping in our case. We find the large splitting of the triplet and singlet excitons due to more localized HOMO an LUMO in comparison with C<sub>60</sub> and C<sub>70</sub>. As a result, the triplet exciton’s energy is very small and possible experiments are suggested to test this phenomenum.
The geometrical figures of C<sub>36</sub> molecule and polarons are discussed and bond lengths and angles are calculated. The charge density on each layer are given. We find that the differences of charge densities among polarons and molecule are mainly in the 2nd and 5th layers.
The following sections are arranged as: in Sec.$`\mathrm{I}\mathrm{I}`$, the extended SSH model is introduced; in Sec.$`\mathrm{I}\mathrm{I}\mathrm{I}`$, we analytically solve the electronic part of the Hamiltonian without Coulomb interaction; in Sec.$`\mathrm{I}\mathrm{V}`$, the whole Hamiltonian is solved self-consistently; in Sec.$`\mathrm{V}`$, results and discussions are presented; and conclusions are made in the final section.
## II MODEL
The Hamiltonian of the extended SSH model to the C<sub>36</sub> molecule is written as:
$$H=H_0+H_{int}+H_{elas}$$
(1)
The first term $`H_0`$ of Eq.(1) is the hopping term of $`\pi `$-electrons.
$`H_0=`$ $``$ $`{\displaystyle \underset{i,j}{}}[t_0\alpha (l_{ij}l_0)](c_{i\sigma }^{}c_{j\sigma }+h.c.){\displaystyle \underset{i,j}{}^{}}t_1(c_{i\sigma }^{}c_{j\sigma }+h.c.)`$ (2)
$``$ $`{\displaystyle \underset{i,j}{}^{\prime \prime }}t_2(c_{i\sigma }^{}c_{j\sigma }+h.c.)`$ (3)
where the three terms describe hopping between the nearest neighbors, the next nearest neighbors and the third neighbors respectively, with the corresponding hopping integrals $`t_0,t_1`$ and $`t_2`$. The influence of electron-phonon coupling is only included in the first term of $`H_0`$, since those of the last two terms are much smaller than $`t_0`$. The third term is important here because it changes the accidental degeneracy of HOMO and LUMO states , which will be explained in the next section. The larger distance terms are neglected, as they are small and do not bring any new effects.
The second term of Eq.(1) is the screened Coulomb interaction expressed in the Hubbard model.
$$H_{int}=U\underset{i}{}n_in_i+V\underset{i,j,\sigma ,\sigma ^{}}{}n_{i\sigma }n_{j\sigma ^{}}$$
(4)
where $`U`$ is the strength of the on-site interaction, and $`V`$ is that between the nearest neighbors.
The third term of Eq.(1) is the elastic energy of the lattice. This term is composed of three parts,
$$H_{elas}=\frac{1}{2}K_1\underset{i,j}{}(l_{ij}l_0)^2+\frac{1}{2}K_2\underset{i}{}d\theta _{i5}^{}{}_{}{}^{2}+\frac{1}{2}K_3\underset{i}{}d\theta _{i6}^{}{}_{}{}^{2},$$
(5)
where the first part describes the spring energy of bond-length terms, and the next two terms describe the spring energy of the angular terms. $`K_i(i=13)`$ are the elastic constants for these different kinds of lattice vibrations. $`d\theta _{i5},d\theta _{i6}`$ are bond angle deviations from the original angle $`108^{},120^{}`$. e.g. $`d\theta _{i5}=\theta _{i5}108^{}`$. The first summation is taken up over all nearest pairs. The second is of all interior angles of pentagons. And the third is of all interior angles of hexagons.
Since there are no enough experiment data to determine the semiempirical parameters, we set the values in the scope of fullerenes such as C<sub>60</sub> and C<sub>70</sub> which can produce reasonable results. We adjust them to give the gap between HOMO and LUMO and the bond length consistent with those of the pseudopotential density functional approach. We take $`t_0=2.5`$ eV, $`\alpha =5.6`$ eV/Å, $`L_0=1.55`$ Å, $`K_1=47`$ eV/ Å<sup>2</sup>, $`K_2=8`$ eV/rad<sup>2</sup>, $`K_3=7`$ eV/rad<sup>2</sup>, $`t_1=0.168t_0`$. In fact, the particular values, except for the value of $`t_2`$, would not affect the physics too much. Based on the calculation of the third-nearest $`\pi `$-orbit integral, $`t_2`$ was estimated to be 0.111$`t_0`$.
## III ANALYTICAL RESULT OF THE Electronic HAMILTONIAN without Coulomb Interaction
Fully exploiting the high $`D_{6h}`$ symmetry, we solve $`H_0`$ algebraically as in the case of C<sub>60</sub>. For simplicity, we temporarily ignore $`t_1`$ and $`t_2`$ terms and will discuss their effects in detail later. The thirty-six $`\pi `$ orbits form a 36$`\times `$36 representation of the $`D_{6h}`$ group. It can be reduced into the sum of following irreducible representations:
$$\{3A_{1g}B_{1g}2B_{2g}3E_{1g}3E_{2g}\}\{2B_{1u}3A_{2u}B_{2u}3E_{1u}3E_{2u}\}$$
(6)
The even-order axis C<sub>6</sub> brings properties different from those odd-order axis $`C_5`$ characterized in C<sub>60</sub> and C<sub>70</sub>, in the way that it has 1-D representations of kind B.
We reduce this problem by $`D_{6h}`$’s subgroup $`C_{6h}`$, using the quantum number m(=0,$`\pm 1`$,$`\pm 2`$,3) which corresponds to the irreducible representation of $`C_{6h}`$, and $`P(=\pm 1)`$ to parity. The thirty-six $`\pi `$ orbits are recombined as:
$$|\mathrm{\Psi }_{mp}^{(l)}=\underset{i}{}\eta ^{mi}\{|l,i+P|7l,i\}.$$
(7)
where $`|l,i`$ represents the $`i`$th($`i=16`$) carbon atom’s $`\pi `$ orbit in the $`l`$th layer. Here $`l`$ is only ranged for 1 to 3, and $`\eta =e^{i\pi /3}`$
The energy eigenstate wavefunction $`\mathrm{\Phi }_{mp}^l`$ can be expanded with these base vectors.
$$\mathrm{\Phi }_{mp}^i=\underset{l=1,3}{}c_{mp,l}^i\mathrm{\Psi }_{mp}^l$$
(8)
Consequently, $`H_0`$ can also be reduced to 3$`\times `$3 matrices in the subspace which is spanned by the new bases:
$$H_{mp}=\left[\begin{array}{ccc}t_a(\eta ^m+\eta ^m)& t_b& 0\\ t_b& 0& t_c(1+\eta ^m)\\ 0& t_c(1+\eta ^m)& t_dP\eta ^{3m}\end{array}\right]$$
(9)
Because of the $`D_{6h}`$ symmetry, there are only four different kinds of bond lengths ($`L_aL_d`$), see Fig.2. The corresponding nearest hopping integrals are $`t_i=t_0\alpha (L_iL_0)(i=ad)`$. The energy eigenvalues $`E_{pm}^l`$ are determined by
$`\lambda ^3+A\lambda ^2+B\lambda +C=0`$ (10)
where
$`A=2t_a\mathrm{cos}{\displaystyle \frac{m\pi }{3}}+(1)^mptd,`$ (11)
$`B=2\mathrm{cos}{\displaystyle \frac{m\pi }{3}}[(1)^mpt_at_dt_c^2](2t_c^2+t_b^2),`$ (12)
$`C=4\mathrm{cos}{\displaystyle \frac{m\pi }{3}}(1+\mathrm{cos}{\displaystyle \frac{m\pi }{3}})t_at_c^2(1)^mpt_b^2t_d`$ (13)
Equation(9) can be solved analytically, so does the eigenvectors. Here we would not list the complicated analytical results. Instead, we present numerical results by using parameters sets,$`t_a=3.30`$eV, $`t_b=2.95`$eV, $`t_c=3.24`$eV, $`t_d=3.08`$eV, which are determined by the bond lengths $`L_i`$. The coefficients $`c_{mp,l}^i`$ of the energy eigenstate wavefunction $`\mathrm{\Phi }_{mp}^i`$ and their irreducible representations of $`D_{6h}`$ that they belong to are showed in Table 1.
In the C<sub>36</sub> molecule, because the hexagons in two middle layers turn an angle of $`30^{}`$ to the four hexagons at two ends, the wavefunctions of $`B_{2u}`$ and $`B_{1g}`$ energy eigenstates are entirely composed of the atom orbits on the two middle layers, while those of $`B_{1u}`$ or $`B_{2g}`$ energy eigenstates only have amplitude on the other four layers. There are six energy eigenstates of the B kind representations. The 13th($`B_{2u})`$, 22nd($`B_{1g})`$, the accidentally degenerate 35th($`B_{1u}`$) and 36th($`B_{2g}`$) levels lie either far below or above the fermi level, so they are of no interest. There are also another pair of accidentally degenerate 18th($`B_{1u}`$) and 19th($`B_{2g}`$) levels which are half filled. We also notice in Table 1. that the 18th($`B_{1u}`$) and 19th($`B_{2g}`$) electrons are distributed in the 2nd and 5th layers with $`90\%`$ possibility and $`10\%`$ in the 1st and 6th layers.
When $`t_1(16.8\%t_0)`$ and $`t_2(11.1\%t_0)`$ are considered, $`H_{mp}`$ have a different form. The long distance hopping($`t_2`$) coupling the 2nd and 5th layers results in $`B_{1u}`$ and $`B_{2g}`$’s splitting. This is the reason that we introduce much longer distance hopping term than usual. Because the charge densities of $`B_{1u}`$ and $`B_{2g}`$ are mainly distributed in the 2nd and 5th layers, the gap between HOMO($`B_{1u}`$) and LUMO($`B_{2g}`$) is approximately $`2t_2`$. When the electron-electron interaction is taken into account, the electronic Hamiltonian has to be solved in the HF mean field theory with self-consistent method. The energy levels of ground state configuration are shown in Fig.3. We can see that in Fig.3 and Table 1, the relative positions of the energy levels only change slightly, except the splitting of $`B_{1u}`$ and $`B_{2g}`$. In Fig.2, the separation between the 16th, 17th levels($`E_{1g}`$) and HOMO is about 0.2eV and that between the 20th, 21st levels ($`E_{2g}`$) and LUMO is 1.3eV.
## IV NUMERICAL RESULT OF THE MOLECULE
Under the adiabatic approximation, we apply the BdeG formalism to the Hamiltonian of the molecule to obtain the stable lattice configuration and the corresponding electronic structure. Molecular dynamical procedure is used to gradually approach the minimum of the potential surface from an initial lattice configuration. The following two equations are used,
$`F_{i\sigma }`$ $`=`$ $`{\displaystyle \frac{dV_{eff}}{dx_{i\sigma }}}(\sigma =13)`$ (14)
$`v_{i\sigma }`$ $`=`$ $`{\displaystyle \frac{dx_{i\sigma }}{dt}}`$ (15)
where $`V_{eff}`$ is the effective lattice potential that includes the elastic energy of the lattice and the electron-lattice interaction energy, both of which depends on the lattice coordinates. $`F_{i\sigma }`$ is the $`\sigma `$ component of the effective force acted on the $`ith`$ atom under this potential. $`x_{i\sigma }`$ is the $`\sigma `$ coordinate of the $`ith`$ atom.
In each step of the dynamical procedure, we solve the electrical part of the Hamiltonian under the lattice configuration given by the last step. The HF mean field theory is performed to decouple the electron-electron interaction. Then the electronic states and wavefunctions are obtained self-consistently.
$`H_{MF}^{el}`$ $`=`$ $`H_0+U{\displaystyle \underset{i}{}}\{n_in_i+n_in_in_in_i\}`$ (18)
$`+V{\displaystyle \underset{i,j,\sigma ,\sigma ^{}}{}}\{n_{i\sigma }n_{j\sigma ^{}}+n_{j\sigma ^{}}n_{i\sigma }n_{i\sigma }n_{j\sigma ^{}}\}`$
$`V{\displaystyle \underset{i,j,\sigma }{}}\{C_{i\sigma }^{}C_{j\sigma }C_{j\sigma }^{}C_{i\sigma }+C_{j\sigma }^{}C_{i\sigma }C_{i\sigma }^{}C_{j\sigma }C_{i\sigma }^{}C_{j\sigma }C_{j\sigma }^{}C_{i\sigma }\}`$
The ground state and the lowest triplet and singlet excitons are investigated. There are four lowest exciton configurations $`A,B,C`$ and $`D`$, see Fig. 4. Due to Coulomb interaction, $`C`$ and $`D`$ are mixed to give singlet and triplet,
$`C|H_{int}`$ $`|D={\displaystyle \underset{i}{}}U\psi _\alpha ^{}(i)\psi _\beta ^{}(i)\psi _\beta (i)\psi _\alpha (i)`$ (20)
$`V{\displaystyle \underset{ij}{}}\{\psi _\alpha ^{}(i)\psi _\beta ^{}(j)\psi _\beta (i)\psi _\alpha (j)+\psi _\alpha ^{}(j)\psi _\beta ^{}(i)\psi _\beta (j)\psi _\alpha (i)\}`$
where $`\alpha `$ and $`\beta `$ denote the 18th and 19th energy level respectively.
## V RESULTS AND DISCUSSION
The configurations and energy of ground and low excited states are given below,
$`E_1^{el}`$ $`=`$ $`64.25eV|\mathrm{\Phi }_1\text{(spin singlet ground state)},`$ (22)
$`E_2^{el}`$ $`=`$ $`64.00eV|\mathrm{\Phi }_2^{13}=|A,|B,{\displaystyle \frac{1}{\sqrt{2}}}(|C+|D)\text{(spin triplets)},`$ (23)
$`E_3^{el}`$ $`=`$ $`63.37eV,|\mathrm{\Phi }_3={\displaystyle \frac{1}{\sqrt{2}}}(|C|D)\text{(spin singlet)}.`$ (24)
We can see that the ground state $`\mathrm{\Psi }_1`$ is a spin singlet rather than triplet after the long distance hopping $`t_2`$ and electron-electron interaction are taken into account, which is an improvement of the prediction of simple Hückel theory, and there is no need to add the hybridizing of $`\sigma `$ and $`\pi `$ bonds to investigate the qualitative physics of C<sub>36</sub> as discussed in Ref..
The lowest excited states $`\mathrm{\Psi }_2^{13}`$ are spin triplet excitons, which are only about $`0.25`$eV above the ground states. The energy of singlet exciton $`\mathrm{\Psi }_3`$ is very high, and the splitting between the triplet and singlet excitons is about 0.63eV, which is much larger than 0.2eV in the cases of C<sub>60</sub> and C<sub>70</sub>. This is because the electrons of HOMO and LUMO are more localized, $`90\%`$ in the 2nd and 5th layers and $`10\%`$ in the 1st and 6th layers. Compared to the C<sub>36</sub> case, the electron densities of HOMO and LUMO in C<sub>60</sub> and C<sub>70</sub> are distributed more uniformly over all sites, so the splitting of the singlet and triplet exciton due to the Coulomb interaction is relatively small. The fact that the triplet exciton has low energy can be verified through experiment observation: When illuminated by external light, the C<sub>36</sub> can be excited to the state of singlet exciton and then may transit to triplet exciton through nonradiative decay. The transition rate to ground state is slow because of the different spin configuration and small energy splitting. We predict that the triplet excitons are metastable states, and the phosphorescence phenomena is possible to be observed. The triplet exciton is paramagnetic while the ground state is a diamagnetic singlet. So the magnetic susceptibility is increased upon illumination of the external light. Electron Spin Resonance(ESR) experiment can be performed to detect the triplet exciton states. When ESR is performed to a solution sample, the usual unimportant magnetic dipole interaction between electrons becomes crucial because its anisotropy can smear the resonance peak. However, this interaction is decayed rapidly with distance as $`R^3`$. The electrons in HOMO and LUMO are mainly uniformly distributed on 12 carbon atoms in two layers, the possibility of their short distance is considerably small compared to other small organic molecule, such as naphthalene. So the resonance peak is possible to be observed. The phone absorption and luminescence spectra experiments can be perform to test the lowest singlet exciton’s energy.
The shape of the C<sub>36</sub> molecule is an ellipsoid with high aspect ratios. Our calculation shows that the distance between the 1st and 6th layers is $`5.2A^{}`$, and that between the parallel vertical hexagon planes is $`4.2A^{}`$. The charge density, bond lengths and bond angles of C<sub>36</sub> molecule and ions are calculated and shown in Table 2, 3 and 4. We note that because of the $`D_{6h}`$ symmetry of the molecule, the 1st and 6th layers are regular hexagons, with each internal angle $`120^{}`$ and the shortest bonds ($`L_1`$) $`1.41A^{}`$. The six vertical hexagons around the equator are slightly deformed due to the drag of the 1st and 6th layers from each side. $`L_4`$ is $`2.8\%`$ longer than $`L_1`$ and $`2\%`$ than $`L_3`$. The $`L_2`$ bonds are relatively weaker than those in hexagons, because they couple the hexagons in the two ends with the hexagons around the equator. $`L_2`$ is $`4.4\%`$ longer than $`L_1`$. This result is consistent with that of LDA except that the $`L_4`$ is as long as $`L_3`$ in LDA, and we think that our result is more reasonable.
The bond lengths of $`L_3`$ and $`L_4`$ of ions differ from those of molecule very slightly. But the negative polarons’ $`L_1`$ is longer and positive polaron’ is shorter than the neutral molecule’s, while the $`L_2`$ changes in the opposite way.
The reason is that the wavefunctions of the electron in $`B_{1u}`$ and $`B_{2g}`$ has opposite sign between two nearest sites connected by $`L_1`$, same sign by $`L_2`$, and have zero amplitude on the 3rd, 4th layers. According to the contribution f these two states form hopping term, the sites connected by $`L_1`$ are repulsive, but those connected by $`L_2`$ are attractive. The angles in the pentagons and hexagons around the equator deviate from $`108^{}`$ and $`120^{}`$ slightly, which justifies the validity of $`H_{elas}`$. The bond angles in polarons differ from those of molecule very slightly, so we omit them in Table 4. Form C$`{}_{36}{}^{}{}_{}{}^{2+}`$ polaron to C$`{}_{36}{}^{}{}_{}{}^{2}`$, the adding electrons distribute mostly in the 2nd and 5th layers. This is because of the particular charge density distribution of the 18th and 19t levels.
## VI CONCLUSION
In this paper, we have carefully studied the effect of $`D_{6h}`$ symmetry on C<sub>36</sub>’s electronic properties under the extended SSH model. A small gap between HOMO($`B_{2u}`$) and LUMO($`B_{1g}`$) is obtained due to long distance hopping. The large splitting of the spin triplet and singlet lowest excitons, the differences of bond lengthes and electron density between molecule and polarons are discussed as results brought by the more localized HOMO and LUMO and their special symmetries. Possilbe experiments are suggested.
ACKNOWLEDGMENT
We thank Dr. Xi Dai for his helpful discussions.
|
warning/0004/cond-mat0004189.html
|
ar5iv
|
text
|
# Phase diagram of diluted magnetic semiconductor quantum wells
\[
## Abstract
The phase diagram of diluted magnetic semiconductor quantum wells is investigated. The interaction between the carriers in the hole gas can lead to first order ferromagnetic transitions, which remain abrupt in applied fields. These transitions can be induced by magnetic fields or, in double-layer systems by electric fields. We make a number of precise experimental predictions for observing these first order phase transitions.
\]
Semiconductor and ferromagnetic materials have complementary properties in information processing and storage technologies. Recent advances in the MBE growing techniques have made possible the fabrication of $`Mn`$-based diluted magnetic semiconductors (DMS) with a rather high ferromagnetic-paramagnetic critical temperature $`T_c`$. In semiconductors it is possible to modulate spatially the density of carriers by changing the doping profiles but in DMS it is also suitable to vary the magnetic order of the carriers by changing the magnetic ion densities. The combination of these two possibilities opens a rich field of applications for these materials.
The high critical temperature DMS’s have a high concentration, $`c`$, of $`Mn^{2+}`$ ions, randomly located. The itinerant carriers in the $`Ga_{1x}Mn_xAs`$ systems are holes and their density, $`c^{}`$, is much smaller than the magnetic ion density. In doped semiconductor the spin $`S`$=$`5/2`$ $`Mn^{2+}`$ ions feel a long range ferromagnetic interaction created by the coupling mediated by the itinerant spin polarized carriers. This interaction has some resemblance with the one observed in the pyrochlores, where a similar type of coupling between a narrow electronic band and magnetic ions is assumed to exist. In the latter case, the phase diagram is significantly different from that of a conventional itinerant ferromagnet, showing first order transitions and phase separation and/or the formation of localized textures near the transition temperature. These features lead to interesting transport properties, like colossal magnetoresistance. We expect these effects to be amplified in low density two-dimensional (2D) doped semiconductors, where the carrier-carrier interaction can play a significant role, and also favors ferromagnetism.
In this work, we investigate the nature of the phase diagram of diluted magnetic semiconductor quantum wells, with emphasis on the existence of abrupt transitions for experimentally accessible parameters. We analyze quantum wells made of $`GaMnAs`$ growth in the $`<001>`$ direction and with thickness $`w`$. The confinement of the carriers in the $`z`$-direction can be obtained by sandwiching the $`GaMnAs`$ system between non-magnetic $`GaAlAs`$ semiconductors.
The system is described by the following Hamiltonian,
$$=_M+_{holes}+J\underset{I}{}𝐒_I𝐬_i\delta \left(𝐫_i𝐑_I\right).$$
(1)
We now analyze the three terms contributing to $``$.
i)$`_M`$ is the Hamiltonian of the localized spins interacting with an external magnetic field $`B`$. Direct interactions between the magnetic moments of the $`Mn`$ ions are much smaller than the interaction with the carrier spins and therefore, unlike in the case of the pyrochlores, we neglect this coupling.
ii)$`_{holes}`$ is the part of the Hamiltonian which describe the itinerant holes. It is the sum of the kinetic energy of the holes and the hole-hole interaction energy. We treat the kinetic energy of the carriers in the framework of the envelope function approximation. The confinement of the carriers in the quantum well produces the quantization of their motion and the appearance of subbands which, for sufficiently narrow quantum wells, have a clear light-hole or heavy-hole character. We assume that the hole density is low enough so that only one of the subbands, heavy-like, is occupied. With this, the in plane motion of the holes can be approximated by a single parabolic band of effective mass $`m^{}`$. We describe the interaction between the electrical carriers with the local spin density approximation (LSDA). Except at very low densities, the LSDA gives an accurate description of the electron gas confined in quantum wells. Since the hole $`g`$-factor is much smaller than the $`Mn`$ $`g`$-factor we neglect in $`_{holes}`$ the coupling between the hole spins and the applied magnetic field.
iii)The last term is the antiferromagnetic exchange interaction between the spin of the $`Mn^{2+}`$ ions located at $`𝐑_I`$ and the spins , $`\stackrel{}{𝐬}_i`$, of the itinerant carriers. The interaction between ions mediated by the conduction holes is of long range. Thus we will assume that thermal distribution of the orientation of the $`Mn`$ spins is that induced by an effective field, due to the holes, which should be calculated selfconsistently.
In order to obtain the phase diagram of the DMS system we have to minimize the free energy $``$. The critical temperatures for the ferromagnetic to paramagnetic transition in DMS is typically smaller than 100$`K`$, and for these temperatures we can consider that the electron gas is degenerate. Hence, the only temperature dependence in $``$ is due to thermal fluctuations of the $`Mn`$ spins. Treating the holes in the mean field approximation the free energy per unit area takes the form:
$$=_{ions}+\frac{\mathrm{}^2}{m^{}}\frac{\pi }{2}n_{2D}^2(1+\xi ^2)+E_{xc}(n_{2D},\xi ).$$
(2)
Here $`\xi `$ is the carrier spin polarization, $`n_{2D}`$=$`c^{}w`$ is the two dimensional density of holes in the system, $`E_{xc}`$ is the exchange correlation energy for the holes and $`_{ions}`$ is the contribution of the magnetic ions to the free energy:
$$_{ions}=Tcw\mathrm{log}\frac{\mathrm{sinh}[\beta h(S+1/2)]}{\mathrm{sinh}(\beta h/2)}$$
(3)
being
$$h=\frac{J}{2}\frac{n_{2d}}{w}\xi +B$$
(4)
the effective magnetic field than the $`Mn`$ spins feel. In obtaining Eq. we have assumed that the hole wave function in the $`z`$-direction has the form $`w^{1/2}`$.
The phase diagram with parameters $`J=0.15`$eV nm<sup>3</sup>, ion concentration $`c=1`$nm<sup>-3</sup> and width of the well, $`d=10`$nm, is shown in fig.{1}. We include a single hole band of effective mass $`m_{}=0.11m_e`$, and a dielectric constant $`ϵ_0=12.2`$.
The calculations have been done by minimizing the value of the free energy, Eq. and using the expression given by Vosko et al. for the exchange correlation potential. In the high density region the line of continuous transitions agrees with that obtained from the divergences of the magnetic susceptibility. The main novelty of our calculation is the identification of a first order transition to a fully polarized state at intermediate densities. This transition takes place at higher temperatures than that at which the magnetic susceptibility of the system diverges. The existence of a first order transition between the paramagnetic and the ferromagnetic phases implies the existence of a discontinuity in the chemical potential at $`T_c`$. This discontinuity indicates the existence of a region in the phase diagram where phase separation can take place. At higher densities, when a continuous transition between $`\xi =0`$ and $`\xi 0`$ occurs, we also find a first order phase transition between the partially polarized system, $`\xi <1`$, and the fully polarized phase, $`\xi `$=1.
In order to understand better the phase diagram it is illustrative to derive the existence of these first order transitions analytically. For doing that we treat the hole-hole interaction in the Hartree-Fock (HF) approximation. We obtain that the HF results are in rather good agreement with the obtained using the LSDA. In the HF approach the hole energy per unit area can be written as,
$`E_{holes}={\displaystyle \frac{\mathrm{}^2}{m^{}}}{\displaystyle \frac{\pi }{2}}n_{2d}^2(1+\xi ^2)`$ (5)
$`{\displaystyle \frac{3}{8\pi }}{\displaystyle \frac{e^2}{ϵ_0}}\left({\displaystyle \frac{3\pi ^2}{w}}\right)^{1/3}n_{2d}^{4/3}\left[(1+\xi )^{4/3}+(1\xi )^{4/3}\right],`$ (6)
where the last term is the exchange energy of the holes. The free energy of the ions is given by Eq.. Near a paramagnetic-ferromagnetic transition, the hole spin polarization is small and we can expand the total free energy in terms of powers of $`\xi `$:
$`_{tot}`$ $``$ $`{\displaystyle \frac{\mathrm{}^2}{m^{}}}{\displaystyle \frac{\pi }{2}}n_{2d}^2C_1{\displaystyle \frac{3}{4}}{\displaystyle \frac{n_{2d}^{4/3}}{w^{1/3}}}Tcw\mathrm{log}(2S+1)`$ (7)
$`+`$ $`\xi ^2\left[{\displaystyle \frac{\mathrm{}^2}{m^{}}}{\displaystyle \frac{\pi }{2}}n_{2d}^2C_1{\displaystyle \frac{1}{6}}{\displaystyle \frac{n_{2d}^{4/3}}{w^{1/3}}}{\displaystyle \frac{1}{T}}{\displaystyle \frac{c}{w}}{\displaystyle \frac{S(S+1)}{24}}J^2n_{2d}^2\right]`$ (8)
$`+`$ $`\xi ^4[C_1{\displaystyle \frac{5}{324}}{\displaystyle \frac{n_{2d}^{4/3}}{w^{1/3}}}+{\displaystyle \frac{1}{T^3}}{\displaystyle \frac{c}{w^3}}{\displaystyle \frac{J^4n_{2d}^4}{16}}\times `$ (10)
$`({\displaystyle \frac{7}{360}}(S+{\displaystyle \frac{1}{2}})^4{\displaystyle \frac{S^2(S+1)^2}{72}})]+\mathrm{}`$
with
$$C_1=\left(\frac{3}{\pi }\right)^{1/3}\frac{e^2}{ϵ_0}.$$
We expect a paramagnetic-ferromagnetic transition when the quadratic term in $`\xi `$ is zero. This implies a critical temperature
$$T_C=\frac{1}{12w}\frac{cJ^2S(S+1)}{\pi \frac{\mathrm{}^2}{m^{}}\frac{1}{3\pi }\frac{e^2}{ϵ_0}\left(\frac{3\pi ^2}{w}\right)^{1/3}n_{2d}^{2/3}}.$$
(11)
At high densities, this approximation gives $`T_C8.7`$K, in good agreement with the LSDA calculation, and with the RKKY solution. Note that the only dependence of $`T_c`$ on $`n_{2d}`$ is through the hole-hole interaction. This is due to the fact that the two-dimensional density of states is energy independent. In three dimensional systems the kinetic energy scales as $`n^{5/3}`$ and $`T_c`$ is proportional to $`n^{1/3}`$ in agreement with the RKKY theory. The order of the transition can be inferred from the sign of the quartic term in Eq.. If the quartic term is positive the transition is of second order, while a negative quartic term implies the existence of a first order phase transition. Using the previous expression for $`T_C`$, a first order transition takes place if:
$`{\displaystyle \frac{e^2}{ϵ_0}}{\displaystyle \frac{1}{\pi }}\left({\displaystyle \frac{3\pi ^2}{w}}\right)^{1/3}{\displaystyle \frac{5}{2916}}{\displaystyle \frac{n_{2d}^{8/3}}{\left(\frac{7}{30}(S+\frac{1}{2})^4\frac{S^2(S+1)^2}{6}\right)}}\times `$ (12)
$`{\displaystyle \frac{c^2J^2S^3(S+1)^3}{\left(\pi \frac{\mathrm{}^2}{m^{}}\right)^3\left(1\frac{m^{}}{\mathrm{}^2}\frac{1}{3\pi ^2}\frac{e^2}{ϵ_0}\left(\frac{3\pi ^2}{w}\right)^{1/3}n_{2d}^{2/3}\right)^3}}>1`$ (13)
which can be solved to give $`n_{2d}^{first}2.1\times 10^{12}\mathrm{cm}^2`$. This estimate is also in good agreement with the LSDA results shown in fig.{1}.
Some comments about the LSDA exchange correlation potential are in order:
i)In the HF treatment, the existence of a negative quartic term in the expansion of the exchange energy in powers of $`\xi `$ is the source for the appearance of first order phase transitions. In the LSDA, the intermediate spin polarization correlation energy is obtained by assuming that it has the same polarization dependence as the exchange energy, so that the LSDA expression for $`E_{xc}`$ also leads to a negative quartic term when expanded in powers of $`\xi `$, and a first order transition is expected. On the other hand, correlation effects weaken the spin dependence of the interaction energy as compared with the results in HF, and we find that the LSDA gives a value for $`n_{2d}^{first}`$ slightly smaller than the observed in the HF approximation. Numerical evaluation of the partially spin polarized $`E_{xc}`$ also shows a negative quartic term in the expansion of $`E_{xc}`$ versus $`\xi `$. Hence, we believe that the existence of a first order phase transition is a robust result, independent of the model used for describing the interaction between carriers. Note, finally, that the existence of a first order transition implies the absence of long range critical fluctuations, lending further support to the adequacy of the methods used in this work.
ii)The interpolation formula used for describing the $`\xi `$ dependence of $`E_{xc}`$ is not analytic at $`\xi `$=1, Eq.. This implies that there is, for each pair of values $`n_{2d}`$ and $`T`$, a $`\xi `$ range near $`\xi `$=1 which cannot be reached by minimizing the total free energy. When $`^2/\xi ^2`$ is smaller than zero, the system prefers to be completely spin polarized. More accurate descriptions of the LSDA $`E_{xc}`$ results in an analytic behavior at $`\xi `$=1. Therefore we believe that, in the results shown in fig.{1}, the discontinuous transition which occurs at lower temperatures than the second order transition $`T_c`$, is, probably, a spurious result due to the use of a HF-like interpolation for $`E_{xc}`$.
We remark again here that the appearance of a first order transition when decreasing $`n_{2d}`$ is a real result, not related with the non analyticity of the LSDA expression for $`E_{xc}`$.
Now we analyze the effect in the phase diagram of an external magnetic field. For $`B`$0, the discontinuous transitions shown in fig.{1} are changed into transitions between phases with two different polarizations. General thermodynamic arguments show that this line of first order transitions should end in a critical point, in an analogous way to the liquid-vapor phase diagram. This critical point belongs to the Ising universality class. In the present calculations, the first order transition persists at all fields. This is due to the non analytic of the exchange energy at $`\xi `$=1. We expect that a correct description of $`E_{xc}`$ would induce the termination of the line of first order transitions at a critical value of the field. Finite temperature effects in the hole gas, not taken into account, also will weaken the $`\xi `$ dependence of the free energy, leading to the existence of a critical point.
The line of discontinuous transitions in an applied field is shown in fig.{2}.
The existence of discontinuous transitions leads to the possibility of phase separation. Using Maxwell construction, we find a region of phase separation near the line of first order transitions. In order to analyze its occurrence, we need to include electrostatic effects. The simplest situation where phase separation can be observed is in a bilayer. Let us imagine two wells with nominal hole density equal to $`n_0`$. At the temperatures and fields where the first order transition occurs, there are two phases with the same free energies, $`_1(n_0)=_2(n_0)`$. The chemical potential of these two phases differ at this point, and we define $`\mathrm{\Delta }\mu =\mu _1\mu _2`$. This difference in chemical potentials will induce a transfer of charge between the two layers, $`\delta n`$, until $`\mathrm{\Delta }\mu `$ is compensated by the induced electrostatic potential, $`V(e^2\delta nd)/ϵ_0`$, where $`d`$ is the distance between the layers. Thus, we obtain $`\delta n(\mathrm{\Delta }\mu ϵ_0)/(e^2d)`$. For reasonable values of $`d1050`$nm, we find that the charge transfer is small, $`\delta n/n_010^2`$. The change in the electrostatic barrier, $`V\mathrm{\Delta }\mu `$ is, however, comparable to the width of the band, and can change significantly the transport properties.
In a bilayer system, the first order transitions analyzed here can be induced by an applied electric field. The field induces a difference in the chemical potential of the two layers, which leads to a charge transfer. By suitably tuning the parameters, the density in one of the layers will reach the value at which the first order transition discussed above takes place. At this point, there will be an abrupt change in the charge transfer, which can be measured with standard capacitive techniques. The charge transfer as function of electric field, for reasonable parameters, is shown in fig.{3}.
In conclusion, we have analyzed the possible discontinuous transitions in two-dimensional diluted magnetic semiconductors. We find that the interaction between carriers can lead to first order transitions in two-dimensional quantum wells. At the transition, the holes become fully polarized. This transition can be induced by a change in the density of the hole gas, the temperature, a magnetic field, and, in multilayer systems, an applied electric field. At these transitions, the minority spin band becomes empty, and the density of states at the Fermi level is reduced by one half. In double layer systems, significant electrostatic barriers can appear near the transition. This change can alter significantly the transport properties. Thus, it can be important in the operation of devices made with these materials.
LB thanks A.H.MacDonald, C.Tejedor and A.Rubio for helpful discussions. We acknowledge financial support from grants PB96-0875 and PB96-0085 (MEC, Spain) and (07N/0045/98 and 07N/0027/99) (C. Madrid).
|
warning/0004/astro-ph0004077.html
|
ar5iv
|
text
|
# Interpretation of the UV spectrum of some stars with little reddening.
## 1 Introduction
This paper is the second of a serie dedicated to the interaction of starlight and interstellar grains in the UV and supported by the observations of the International Ultraviolet Explorer (IUE) satellite.
The properties of the interstellar grains can be studied either by their capacity to scatter starlight, by observations of reflection nebulae, or by their capacity to extinguish starlight, by observing a star through an interstellar cloud. The former method allows the evaluation of the phase function and of the albedo of interstellar grains. The latter gives more specific information on the proportion of starlight to be absorbed or scattered at each wavelength. If extinction is the only process involved, the curve which is obtained by dividing the spectrum of a star by the spectrum of an unreddened star of same spectral type shows the relative capacity, from one wavelength to another, of the grains to extinguish starlight in the direction of the star. This is an intrinsic property of the interstellar grains present in the direction of the observation.
The most salient feature of the UV spectrum of a star is the $`2200\mathrm{\AA }`$ bump which appears when there is interstellar matter between the star and the observer. From the standpoint of interstellar grain properties, and if no scattered starlight is introduced into the beam of the observation, this implies the existence of a particular class of grains, the bump carriers, which extinguish light at wavelengths close to $`\lambda _b2175\mathrm{\AA }`$. Those particles have never been formally identified to date. Efforts to comprehend the variations of the stars UV spectrum in different interstellar environments, all of which have assumed starlight extinction as the only process involved, did not really succeed in bringing a global understanding of the UV extinction curve (see Savage et al. savage85 (1985), Fitzpatrick & Massa fitzpatrick86 (1986) and fitzpatrick88 (1988)).
In a preceding paper (Zagury zagury1 (2000), paper I hereafter) the existence of the bump carriers was questioned as these carriers do not affect the UV spectrum of reflection nebulae. The UV spectrums of the bright nebulae presented in paper I were all interpreted as the result of starlight scattered by interstellar grains with identical albedo and phase function across the UV and in the different directions of space sampled by the nebulae. It was also noted that these properties of the grains may be identical in the optical spectral range. No bump or other particular feature is created at $`2200\mathrm{\AA }`$ in the spectrum of the light scattered by a nebula.
If the bump carriers are not present in nebulae, the presence of an additional component due to scattering becomes the most reasonable explanation of the bump (Bless & Savage bless72 (1972), and the appendix of this paper). To explain the variations of the surface brightness of a nebula in function of its distance $`\theta `$ from the illuminating star (paper I), the interstellar grains must have a strong forward scattering phase function. In paper I it was found that the maximum surface brightness a nebula can reach varies as a power law $`\theta ^\alpha `$ ($`\alpha <1`$) of $`\theta `$. A value $`\alpha =2`$ was found. Consequently, the starlight scattered into the beam of observation at close angular distance to the star may be a non-negligible proportion of the direct starlight. The scattered starlight will be more important in a wavelength range for which the scattering medium has an optical depth close to $`\tau _{max}12`$.
The present paper will further develop these ideas. I will study the UV spectrum of a selected sample of stars with a bump and little reddening. Contrary to previous studies which use the logarithm of the spectrums as a mean of obtaining the extinction curve, in this paper, the direct linear data will be used. This is necessary if the spectrums can be decomposed into two separate components, the direct starlight and the scattered starlight.
The stars to be studied in this paper have been selected because of the simple and straightforward interpretation which can be given of their UV spectrum. The specific aspect of the UV reduced spectrum of these stars, defined as the ratio of the star to an unreddened star of same spectral type spectrums (section 2), clearly separates two spectral regions. The long wavelength part of the spectrum is correlated with the reddening of the star, $`E(BV)`$, and, for some stars, it is fitted down to the bump spectral region by an exponential of $`1/\lambda `$. The exponential will be interpreted as the extinction of starlight $`e^{\tau _\lambda }`$, where $`\tau _\lambda `$ is a linear function of $`1/\lambda `$ (section 4). The linear in $`1/\lambda `$ dependence of $`\tau _\lambda `$ was established in the optical in the 1930’s (Greenstein & Henyey greenstein41 (1941) and references therein) and detailed in more recent studies by Rieke & Lebofsky (rieke85 (1985)) and Cardelli, Clayton and Mathis (cardelli89 (1989)). For the stars I have selected, this law extends to the UV.
At shorter wavelengths, $`\lambda <\lambda _b`$, a bump-like feature appears, superimposed to the exponential decrease. This feature will be analysed in section 6 and attributed to additional starlight scattered by interstellar dust at very small (compared to the beam of the observations) angle to the star. Hence, the $`2200\mathrm{\AA }`$ bump is no longer considered as a depression, but rather the point at which scattering becomes noticeable.
The consequences of this interpretation are discussed in the conclusion.
## 2 Data
The IUE experiment (Boggess et al. 1978a , 1978b ) and the process employed to obtain the final IUE spectrums have been described in paper I.
This paper will be concerned with seven stars, listed in table 1. Two of these stars, HD23480 (Merope) and HD200775 illuminate the Merope nebulae and NGC7023, which have been studied in paper I. For each star, the spectrum presented in the paper is an average of the best observations of the object. Some stars having been observed many times, only a few observations were necessary to ensure sufficient accuracy. At times I used high dispersion spectrums and decreased the resolution by a median filter.
The spectrum of the seven stars selected for this study are presented in figure 7, as they can be seen at IUE website.
The ratio of a reddened star spectrum to the spectrum of an unreddened star of same spectral type (reference star) will be called a ‘reduced spectrum’ of the reddened star. Each star has many reduced spectrums. All reduced spectrums of a star are proportional.
The reduced spectrum of the seven stars, scaled to a common value at $`3\mu \mathrm{m}^1`$, are presented in figure 2.
The ‘absolute reduced spectrum’ of a star is the reduced spectrum of the star obtained if the reference star is the star itself, corrected for reddening. It characterizes the interstellar medium between the star and the observer (section 3). The logarithm of the absolute spectrum of a star is proportional to the extinction curve, $`A_\lambda `$ versus $`1/\lambda `$, in the direction of the star.
The reference stars which have been used to establish the reduced spectrum of the stars are listed in table 2.
A star spectrum can be presented as a function of the wavelength $`\lambda `$ or as a function of $`\lambda ^1`$. The former manner keeps close to the data, whereas the latter is preferable since the optical depth varies as $`1/\lambda `$. The latter presentation, more useful when dealing with scattering, has been chosen for most plots.
## 3 The $`2200\mathrm{\AA }`$ bump and the extinction of starlight
When only extinction affects the transport of light between a star and the observer the spectrum of a star is the product of two terms: the unreddened flux of the star $`F_{,\lambda }^0`$ and the extinction $`e^{\tau _\lambda }`$, where $`\tau _\lambda =0.92A_\lambda `$ is the optical depth at wavelength $`\lambda `$ of the medium in front of the star. A reduced spectrum of the star, proportional to $`e^{\tau _\lambda }`$, depends only on the interstellar medium between the star and the observer. The absolute reduced spectrum of the star is $`e^{\tau _\lambda }`$.
The presence of the bump has complicated the interpretation of the UV spectrum of the stars but has not changed the ideas presented so far: the bump is usually considered as a particular feature in the $`\tau _\lambda `$ function.
The bump was proven to originate in the interstellar medium and related to the quantity of interstellar matter in front of the star (Savage savage75 (1975), see also Savage, Massa and Meade savage85 (1985)). Stars of same spectral type with a bump have significant differences in their U.V. spectrums while stars with no bump and close spectral types superimpose very well after multiplication by an appropriate factor (figure 1).
## 4 The exponential decrease
Figure 2 plots the reduced spectrums of the seven selected stars. The reduced spectrums are smoothed by a median filter and scaled to have the same value at $`3\mu \mathrm{m}^1`$. Stars are listed on the figure by order of increasing $`E(BV)`$, written after the star name.
Each spectrum consists of 2 parts, clearly separated at $`\lambda _b`$. The long wavelength part ($`3\mu \mathrm{m}<1/\lambda <1/\lambda _\mathrm{b}`$) varies steeply with increasing $`1/\lambda `$. This decrease will be called the ‘exponential decrease’.
There is a close correlation between the $`E(BV)`$ value of a star and the steepness of its’ exponential decrease: stars with larger $`E(BV)`$ have a more rapid decrease.
For all the stars of the sample, the exponential decrease is well fitted by an exponential, $`Ex=\beta e^{e/\lambda }`$, of $`1/\lambda `$. The exponent $`e`$ can be estimated by taking the logarithm of the reduced spectrums (figure 3, left). In some directions (HD147889, BD $`+\mathrm{62\hspace{0.17em}2154}`$), especially the directions of highest $`E(BV)`$, the exponential fit is best at long wavelengths, close to the optical wavelengths. In other directions, e.g. the directions of lowest $`E(BV)`$ (HD23480, HD37903, HD149757, HD62542, HD200775), the fit applies to the totality of the exponential decrease.
Figure 3, right, plots $`e`$ as a function of $`E(BV)`$. Within the error margin of $`E(BV)`$ (estimated to be $`0.1`$ mag) and on the determination of $`e`$ ($`5\%`$), we have: $`e2E(BV)`$. This relation is justified if the linear relation which holds between $`A_\lambda =1.08\tau _\lambda `$ and $`1/\lambda `$ in the visible (Cardelli, Clayton & Mathis cardelli89 (1989), Rieke & Lebofsky rieke85 (1985)) is extended to the UV. In this case:
$`A_\lambda `$ $`=`$ $`{\displaystyle \frac{E(BV)}{\frac{1}{\lambda _B}\frac{1}{\lambda _V}}}({\displaystyle \frac{1}{\lambda }}{\displaystyle \frac{1}{\lambda _V}})+A_V`$ (1)
$`=`$ $`2.2E(BV)({\displaystyle \frac{1\mu \mathrm{m}}{\lambda }}+0.46(R_V4))`$
$`\tau _\lambda `$ $`=`$ $`2E(BV)({\displaystyle \frac{1\mu \mathrm{m}}{\lambda }}+0.46(R_V4))`$ (2)
with $`R_V^1=(A_BA_V)/A_V=\tau _B/\tau _V1`$.
In the spectral range where equation 2 applies, extinction decreases as $`e^{\tau _\lambda }e^{2E(BV)/\lambda }`$. The $`Ex`$ functions which fit the near UV spectrum of the stars we are concerned with have an identical exponential dependence on $`1/\lambda `$ (exponent$`=2E(BV)/\lambda `$). The $`Ex`$ function prolong the linear optical extinction in the UV. Since there is no reason to suspect an abrupt change in the optical depth -more specifically in the constant term of equation 2\- when going from the optical to the UV, relation 2 must hold, in the directions sampled by the stars, from the near infrared to the near UV ($`\lambda >\lambda _b`$).
The left plots of figures 4, 5 and 6-top represent a reduced spectrum in the directions where the exponential decrease can be fitted by the $`Ex`$ function down to $`\lambda _b`$. The solid line is the exponential fit. The close fit the $`Ex`$ function provides to the exponential decrease indicates that, in these directions, the extinction curve is a linear function of $`1/\lambda `$ according to equation 1 from the optical to $`\lambda _b`$.
## 5 Absolute reduced spectrums
Knowledge of $`E(BV)`$ along with relation 2 permits an exact scaling of the reduced spectrums in the directions where the exponential decrease is well fitted by an exponential function of $`1/\lambda `$.
In these directions, the exponential $`Ex=\beta e^{2E(BV)/\lambda }`$ multiplied by a scaling factor $`\alpha _s`$ must equal the extinction of starlight, $`e^{\tau _\lambda }`$. $`\alpha _s`$ is unambiguously determined from equation 2 by:
$`\alpha _s`$ $`=`$ $`{\displaystyle \frac{1}{\beta }}e^{0.92E(BV)[R_V4]}`$ (3)
$`=`$ $`{\displaystyle \frac{1}{\beta }}e^{0.92A_V(1\frac{4}{R_V})}`$ (4)
$`\alpha _s`$ is the factor to be applied to the star reduced spectrum in order to obtain the absolute reduced spectrum and the extinction curve in the direction of the star.
In low density regions where $`R_V`$ is supposed to be $`3`$ (Cardelli et al. cardelli89 (1989)), we have:
$`\tau _\lambda `$ $`=`$ $`2E(BV)({\displaystyle \frac{1\mu \mathrm{m}}{\lambda }}0.46)`$ (5)
$`\alpha _s`$ $`=`$ $`{\displaystyle \frac{1}{\beta }}e^{0.92E(BV)}`$ (6)
With equation 6 the absolute reduced spectrum in the direction of a star can be deduced from a reduced spectrum and from the associated $`Ex`$ function. This property was used to calibrate the right axis of the left hand plots and the right hand plots of figures 4 to 6.
## 6 The FUV spectrum
### 6.1 Analysis
There is no evident correlation between $`E(BV)`$ in the direction of a star and the short wavelength part ($`1/\lambda _\mathrm{b}<1/\lambda <8\mu \mathrm{m}`$) of its reduced spectrum (figure 2). The bump-like feature at $`1/\lambda >1/\lambda _b`$ is added to the tail of the exponential decrease. When the latter is slow, indicating small $`E(BV)`$, the bump feature is tilted because of the underneath exponential decrease. HD23480, HD149757, HD62542, HD200775 are typical examples (see figure 2 and the left plots of figures 4 and 5).
In the directions of the stars selected in figure 4 to figure 6, the $`Ex`$ function fits the exponential decrease down to $`\lambda _b`$, showing no excess of extinction or other peculiarity at this wavelength. In all directions the reduced spectrum seems to catch up with the exponential decrease for large $`\tau _\lambda `$-values.
The reduced spectrum of the stars, right hand plots of figure 4 to figure 6, comprises the expected exponential extinction of starlight and the additional bump-like feature at short wavelengths. The bump feature can be isolated if, for each direction, $`e^{\tau _\lambda }`$ is substracted from the absolute reduced spectrum in the same direction. The resulting curves are plotted at the right hand of figure 4 to 6.
The short wavelength bump in the different directions are represented on the same plot, figure 6, down and left. With increasing $`E(BV)`$, the height of the bumps tends to decrease and the short wavelength decrease (high $`\tau _\lambda `$) is steeper.
### 6.2 Scattering
Scattering was ruled out as an explanation of the $`2200\mathrm{\AA }`$ bump for reasons which are summarized in the appendix and are questionable. The spectrum of nebulae (paper I) did not reveal the presence of the bump carriers in the interstellar medium. The exponential decrease of the UV spectrums can also logically be interpreted as the extinction of starlight by the same mean grain population responsible of the ‘normal’ extinction process of starlight. For the stars studied here there does not seem to be any peculiarity at $`\lambda _b`$, and there is no need of a particular class of grains which would extinguish starlight at this wavelength. The bump-like FUV feature is superimposed on the exponential decrease and appears as an additive feature to the extinction of the direct starlight $`e^{\tau _\lambda }`$. It must arise (Bless et Savage bless72 (1972) and the Appendix, this paper) as a result of the introduction of scattered starlight into the beam of observation.
If scattering is added to direct starlight, the light we receive from the direction of a star is the sum of the direct starlight, $`F_{,\lambda }^0e^{\tau _\lambda }`$, and of the scattering component $`F_{,\lambda }^0Sca_\lambda e^{\tau _\lambda ^{}}`$. $`F_{,\lambda }^0`$ is the stellar flux at $`\lambda `$ corrected for reddening, $`Sca_\lambda `$ is the proportion (relative $`F_{,\lambda }^0`$) of starlight at wavelength $`\lambda `$ scattered into the beam. $`\tau _\lambda ^{}<\tau _\lambda `$ is the optical depth which accounts for the extinction of starlight between the star and the scattering medium and for the extinction of the scattered light.
The $`Sca_\lambda `$ function depends on the structure of the medium sampled by the line of sight. The light scattered by a medium made of clumps with low density will not have the same spectrum as the light scattered by high density clumps. The small scale structure of the medium, that is the density distribution of the different regions which compose the scattering medium at scales probably smaller than the IUE beam (Falgarone et al falgarone (1998)), determines the shape of the $`Sca_\lambda `$ function.
With increasing $`E(BV)`$ the FUV bump has a smaller maximum and a steeper FUV decrease (left-bottom plot of figure 6). This can be attributed to the $`e^{\tau _\lambda ^{}}`$ term of the scattered emission which affects the shortest wavelengths.
In the right hand bottom plot of figure 6 the FUV bumps are multiplied by an appropriate function $`e^{\tau _\lambda ^{}}=e^{\gamma \tau _\lambda ^0}`$, $`\tau _\lambda ^0`$ given by equation 5 with $`E(BV)=1`$, and $`\gamma `$ an appropriate constant. $`\gamma `$ is adjusted for all the curves to have comparable maximums. All the curves of the figure have an identical growth between $`\lambda _b`$ and the maximum.
The curves for the direction of lowest reddening (HD23480, HD37903, HD149757) superimpose well, supporting the idea of scattering by a medium with similar characteristics in these directions: the $`Sca_\lambda `$ functions are identical and the scattered component of the absolute reduced spectrums differ by the extinction term $`e^{\tau _\lambda ^{}}`$ only.
The curves for the directions of HD200775 and HD62542 cannot be superimposed to the others, thus differing by the $`Sca_\lambda `$ function, which indicates an environment of different nature.
### 6.3 Implications
If scattering is responsible for the FUV bump of a star, the bottom plots of figure 6 show that it can represent up to $`15\%`$ of the direct starlight received on earth and corrected for extinction.
Most of the additional light has to be scattered within the $`\theta _0=1\mathrm{"}`$ angle of the small $`S`$ aperture of the IUE telescope since the differences between observations made with the $`S`$ and with the large $`L`$ apertures of the IUE telescop are explained by pointing problems (section A.1). Because the ratio of $`L`$ to $`S`$ observations is generally greater than $`1.2`$, the difference of the amount of scattered starlight between the two types of observations must be less than $`20\%`$.
Thus, if scattering is responsible for the FUV bump of a star, the scattered light is necessarily an appreciable proportion of direct starlight and scattering by interstellar dust must be strongly oriented in the forward direction.
### 6.4 Case of a $`\theta ^2`$ dependence of the maximum surface brightness of nebulae
Suppose the $`\theta ^2`$ dependence of the maximum surface brightness of a nebula (paper I) on angular distance $`\theta `$ to the illuminating star is verified and applies to very small angles, down to a lower limit $`\theta _{min}`$. According to the equation 6 of paper I, the amplitude of the FUV bump, $`15\%`$ of the unreddened flux of the star measured on earth, will be justified if:
$`\theta _{min}`$ $`<`$ $`\theta _0e^{\frac{0.07}{\pi c}}`$ (7)
$`\theta _{min}`$ $`<`$ $`\mathrm{6\hspace{0.17em}10}^4\mathrm{"}`$
The latter inequality assumes $`c\mathrm{3\hspace{0.17em}10}^3`$ and $`\theta _0=1\mathrm{"}`$. Note that this upper estimate of $`\theta _{min}`$ is extremely sensitive to the $`c`$ parameter. A value of $`c=\mathrm{15\hspace{0.17em}10}^3`$, within the possible range of values observed in paper I, gives $`\theta _{min}<0.2\mathrm{"}`$. $`c=10^3`$ gives $`\theta _{min}<\mathrm{2\hspace{0.17em}10}^{10}\mathrm{"}`$.
If $`r_a`$ is the ratio of the maximum light received from the direction of a star observed with apertures $`\theta _1`$ and $`\theta _0`$:
$`r_a`$ $`=`$ $`{\displaystyle \frac{\pi c(1+2\mathrm{ln}(\theta _1/\theta _{min}))}{\pi c(1+2\mathrm{ln}(\theta _0/\theta _{min}))}}`$ (8)
$``$ $`{\displaystyle \frac{\mathrm{ln}(\theta _1/\theta _{min})}{\mathrm{ln}(\theta _0/\theta _{min})}}`$
$``$ $`1{\displaystyle \frac{\mathrm{ln}\theta _1}{\mathrm{ln}\theta _{min}}}`$
with $`\theta _1`$ and $`\theta _{min}`$ in arcsecond.
If $`\theta _1\mathrm{\hspace{0.17em}10}\mathrm{"}`$, corresponding to the large aperture of the IUE telescope, and $`\theta _{min}`$ is of order $`10^\alpha \mathrm{"}`$, equation 8 has the simple form:
$$r_a=1+\frac{1}{\alpha }$$
(9)
Within this framework, conditions 7 and 8 refine the validity domain of the $`\theta ^2`$ law. Condition 7 must be satisfied to account for the amount of scattered light which is observed. The power received in the direction of a star will not differ from the small ($`\theta _01\mathrm{"}`$) to the large ($`\theta _110\mathrm{"}`$) aperture of the IUE telescope if $`\alpha 1`$ (equation 9). $`r_a`$ less than $`1.2`$ will be achieved for $`\alpha 5`$, i.e. $`\theta _{min}10^5\mathrm{"}`$.
In general, for observations made with an aperture $`\theta _110^{\alpha _m}\mathrm{"}`$, and provided that the $`\theta ^2`$ dependence of the surface brightness extends to $`10^{\alpha _m}\mathrm{"}`$, $`r_a`$ will be $`1+\alpha _m/\alpha `$. This result also depends upon the filling factor of the scattering medium.
## 7 Conclusion
The UV spectrum of selected stars with a bump and moderate reddening was decomposed into two parts. The first part is the expected extinction of starlight, $`e^{\tau _\lambda }`$, with $`\tau _\lambda `$ a linear function of $`1/\lambda `$. The second is the FUV bump, observed in figure 2, which is superimposed on the tail of the exponential decrease. This additional component was interpreted as starlight scattered by interstellar dust at very small angle to the star.
This decomposition is justified by the very close exponential fit which can be applied to the long wavelength exponential decrease of the spectrums (figure 4 to 6). The exponent of the exponential is $`e=2E(BV)/\lambda `$, as in the optical. The exponential fit extends to the UV the linear relation between $`A_\lambda `$ and $`1/\lambda `$ which applies in the optical. There is no excess of extinction at $`2200\mathrm{\AA }`$ in the spectrum of the stars which have been selected.
The additive nature of the FUV bump follows from the analysis of figure 2 carried out in sections 4 and 6. Its’ interpretation as scattered starlight is then the most credible one. This interpretation provides a simple explanation of the similarities and of the differences between the bumps in the different directions (section 6.2). If scattering is present in the spectrum of the stars, most of it must occur at less than $`1\mathrm{"}`$ to the star. It implies a strong forward scattering phase function of the interstellar grains, as found in paper I. An attempt to justify the amount of scattered light ($`15\%`$ of the star flux measured on earth and corrected for reddening) was proposed in section 6.4. It involves the power law found in paper I, which need to be confirmed, of the variation of a nebula maximum surface brightness with angular distance to the illuminating star.
This interpretation of the UV spectrum of the stars has two important consequences.
If the UV spectrum of some stars with a bump is explained without requiring an excess of extinction from the bump carriers at $`\lambda _b`$, all the stars’ UV spectrum must have a similar interpretation. There is no reason why scattering should affect the spectrum of some stars solely and/or why the bump carriers should be present in some nebulae only.
In the directions of the selected stars, the extinction curve, $`A_\lambda `$ as a function of $`1/\lambda `$, is a straight line from the near infrared to the FUV. The value of $`A_\lambda `$ at wavelength $`\lambda `$ is given by the equation 1. If, as it was suggested in paper I, grains have the same properties in all directions of space, this law must hold for all directions.
Why is scattering so important in the FUV? A first reason comes from the extinction of starlight which is increased when moving to the shortest wavelengths. Consequently, at short wavelengths, the scattered starlight will be a larger part of the total light received in the direction of a star. It is also plausible that the structure of the interstellar medium favors scattering at particular wavelengths. A medium constituted of small clumps of similar $`A_V`$ will scatter starlight in a particular wavelength range: high $`A_V`$ clumps will preferentially scatter toward the red while very low column density clumps will scatter in the UV. Although this aspect of scattering was not developped here, it may be an important step in the comprehension of the spectrum of the stars.
## Appendix A The possible explanations of the $`2200\mathrm{\AA }`$ bump
Bless & Savage (bless72 (1972)) review the possible causes of the bump: stars with a bump have a peculiar energy distribution; a large amount of starlight is scattered into the line of sight by interstellar dust; the extinction properties of the interstellar medium are particular in the UV.
Because of its relation to $`E(BV)`$, the bump does not originate in the stars’ atmosphere, nor does it come from a special energy distribution.
One explanation has been widely accepted and developped in all studies of the feature: the bump is due to a special class of interstellar grains which extinguish light at $`1175\mathrm{\AA }`$. According to studies of the bump in various environments (Savage savage75 (1975), Jenniskens & Greenberg jenniskens93 (1993), Nandy et al. nandy76 (1976)), those grains are well mixed (in all environments) to the large grains population responsible for the ‘normal’ extinction process of starlight.
The last possibility, that scattered light enters into the beam, was ruled out for reasons to be discussed in A.1.
### A.1 Arguments used against scattering
3 types of arguments were used to rule out the possibility of scattering as an explanation for the bump.
According to Bless & Savage (bless72 (1972)), Code has calculated that grains of albedo close to one are required to produce enough scattering to explain the UV extinction curve. However those calculations suppose isotropic scattering. The importance of forward scattering, and its’ consequences for our interpretation of the UV spectrum of nebulae has been emphasized in paper I. Introduction of forward scattering will invalidate Code’s calculations.
Snow & York (snow74 (1974)) have compared the spectrums of $`\sigma \mathrm{Sco}`$ from 2 different observations, each of which involves a camera of different aperture. The Wisconsin spectrometer aboard the Orbiting astronomical Observatory 2 (OAO-2) has a large aperture of $`8^{}\times 3^{}`$, while the Princeton OAO-3 has an entrance slit $`0.3^{\prime \prime }\times 39^{\prime \prime }`$, $`10^3`$ smaller than OAO-2. Thus, if the UV spectrum of stars with a bump was due to pollution by scattered nebular light, the authors expect to find differences between the 2 spectrums. No such difference is observed.
Snow & York’s work imposed a serious constraint on interstellar grains but did not demonstrate the abscence of scattering in the UV spectrum of stars. If the phase function is strongly forward scattering, that is if the assymetry parameter $`g_0`$ is close to 1, significant scattering will come only from directions very close to the star. Both observations used by Snow & York have relatively large apertures, and will receive nearly equal amounts of scattered light.
Snow & York’s experiment can be repeated with IUE data. In most of the stars observed with both apertures of the IUE camera there is a scaling factor of $`1.2`$ to $`2`$ between the $`L`$ and $`S`$ aperture spectrums, regardless of the star reddening. This difference also affects unreddened stars and is probably due to the difficulty of holding the star within the beam when using the small aperture. Hence, if UV spectrums are affected by scattering, it must occur at angular distances less than $`1.5\mathrm{"}`$ from the star.
Witt & Lillie (witt73 (1973)) study the diffuse Galactic light (DGL) spectrum from the OAO-2 satellite. The arguments of this paper rely on models of both the interstellar medium, assumed to be a plane parallel slab of uniform density, and of interstellar grains. The apparent disagreement between the model and the observations is attributed to changes of dust albedo with wavelength and to a pronounced minimun of the albedo at $`2200\mathrm{\AA }`$ ($`\lambda _b`$). No DGL spectra is presented in Witt & Lillie’s paper but their results imply a pronounced minimum of the DGL at $`\lambda _b`$.
IUE has observed over 400 off-positions, ‘IUE SKY’, free of luminous objects. Many of the observations have some cirrus on their line of sight, the $`100\mu `$m IRAS emission can range from 1 to a few $`10`$MJy/sr. The only region I have found with a reliable signal across the entire LWR camera wavelength range is spatially close to -and probably scatters the light of- the star cluster NGC1910. It has a level of $`\mathrm{8\hspace{0.17em}10}^{14}`$ $`\mathrm{erg}/\mathrm{cm}^2/\mathrm{s}/\mathrm{\AA }`$ and no bump. In all other observations the emission shortward of $`2600\mathrm{\AA }`$ has a very broad amplitude which can be attributed either to noise or to a very low level of signal. None of the spectrums, even when many of them are co-added, shows evidence for extinction at $`2200\mathrm{\AA }`$.
|
warning/0004/cond-mat0004412.html
|
ar5iv
|
text
|
# Collective Phenomena in Defect Crystals
## I Introduction
Interacting quantum impurities occur in various systems, such as spin glasses, magnetic systems with random couplings, and ionic crystals with substitutional defects. Spin glasses have been studied as a prominent example for systems far from thermal equilibrium. Though real spin glasses are of quantum nature, most theoretical approaches rely on classical models. Work on true quantum spin glasses has been mainly directed towards elucidating the nature of the phase transition to a frozen low-temperature state and at studying the influence of quantum fluctuations on the order parameter (for a recent review, see Ref. ). Likewise with quantum ferromagnets. Theoretical work has concentrated on the quantum phase transition that occurs at $`T=0`$. In disordered quantum ferromagnets, additional interest stems from the fact that rare fluctuations of the disorder configurations leading to Griffiths singularities have a much stronger effect in the presence of quantum fluctuations than in the corresponding classical systems.. Pertinent experiments have been performed on the diluted dipolar Ising magnet LiHo<sub>c</sub>Y<sub>1-c</sub>F<sub>4</sub>, which exhibits both, magnetic and spin glass order, depending on the concentration $`c`$ of Ho ions. Agreement between experiment and theory for this system is, however, still not in a satisfactory state.
Early work on ionic crystals with substitutional impurities, such as lithium or cyanide in various alkali halides (KCl:Li, KBr:CN,…), focused on the thermal properties; experimental findings at high concentration indicated the relevance of the dipolar interactions. More recently, sensitive echo measurements gave precise information on pairs of interacting impurities that occur at low doping. At impurity concentrations of a few hundred ppm, the dipolar couplings destroy the coherent motion of the impurities. Though a few questions would seem settled, the role of interaction, especially the dynamical aspects, is still controversial.
At still higher concentration, one obtains mixed crystals that show glassy behavior with respect to the rotational motion of the impurities, such as KBr<sub>(1-c)</sub>(CN)<sub>c</sub> or (MF<sub>2</sub>)<sub>(1-c)</sub>(LF<sub>3</sub>)<sub>c</sub>, with M=Ca,Sr,Ba. For this latter example of mixed fluorite crystals, the coupled density of states of the impurity degrees of freedom was deduced from Raman spectra, at concentrations $`c=0.050.45`$.
In the present paper, we study the consequences of interactions between substitutional defects for the low $`T`$ properties of defect crystals. We are interested in describing the effects of interactions on quantized defect motion, as well as in analyzing the crossover to glassy behavior at sufficiently high defect concentration. Interactions may be of an electric dipolar nature, or mediated by elastic strain fields.
We consider $`N`$ impurities that are randomly distributed on a lattice with $`N_0`$ sites. The defect concentration $`c=N/N_0`$ is usually much smaller than unity; the most interesting physics typically occurs in a range from $`c=10^5`$ to a few per cent.
The impurity Hamiltonian comprises a one-particle crystal field potential that is identical for each defect, and an interaction term that reflects the random configuration on the host lattice. The one-particle potential for a given impurity is written most conveniently in terms of local coordinates. For Li impurities in a KCl host, these coordinates could be chosen as $`𝒗_i=𝑹_i𝑹_i^0`$, where $`𝑹_i`$ is the impurity position at the lattice site $`𝑹_i^0`$. Tunneling arises from degenerate minima at off-center positions of the crystal field potential $`G(𝒗_i)`$. The local displacement $`|𝒗_i|`$ is significantly smaller than the lattice constant. For other systems, the $`𝒗_i`$ might describe orientational degrees of freedom of a defect molecule.
In a dipole approximation, the interaction of impurities $`i`$ and $`j`$ is linear in the local variables $`𝒗_i`$ and $`𝒗_j`$ and involves a coupling parameter $`J_{ij}`$. It may arise from electric dipole moments $`𝒑_i=q𝒗_i`$, or from elastic quadrupole moments; in both cases the interaction parameter $`J_{ij}`$ is proportional to the inverse cube of the impurity distance $`r_{ij}=|𝑹_i^0𝑹_j^0|`$, that is $`J_{ij}r_{ij}^3`$.
In our investigation of collective effects we shall, in what follows, resort to a simplified description in terms of an effective scalar mean-field analysis that consists of the following approximations. (i) The dependence of $`J_{ij}`$ on the relative angles of $`𝒗_i`$, $`𝒗_j`$, and $`(𝑹_i^0𝑹_j^0)`$, is simplified to two possible signs. (ii) The local coordinates $`𝒗_i`$ are replaced by the one-dimensional scalar variables $`v_i`$, the one-particle crystal field potential by $`G(v_i)`$, and the dipolar interaction by $`J_{ij}v_iv_j`$. In this scalar version, the impurity Hamiltonian takes the form
$$=\underset{i}{}\frac{p_i^2}{2m}+U_{\mathrm{int}}(\{v_i\})$$
(1)
with
$$U_{\mathrm{int}}(\{v_i\})=\frac{1}{2}\underset{ij}{}J_{ij}v_iv_j+\underset{i}{}G(v_i).$$
(2)
The distribution $`P_0`$ of dipolar couplings $`J_{ij}`$ can be evaluated for spacings $`r_{ij}`$ much larger than the lattice constant, i.e., for small impurity concentration $`c1`$. In a continuum approximation, the $`J_{ij}r_{ij}^3`$ scaling translates into
$$P_0(J_{ij})=\frac{1}{2(N_01)}\frac{J_{\mathrm{max}}}{J_{ij}^2},$$
(3)
for $`J_{\mathrm{max}}/N_0|J_{ij}|J_{\mathrm{max}}`$. The factor $`\frac{1}{2}`$ accounts for the two possible signs of $`J_{ij}`$. The upper bound $`J_{\mathrm{max}}`$ is determined by the interaction of closest neighbors,
$$J_{\mathrm{max}}=\frac{3q^2}{4\pi ϵϵ_0r_{\mathrm{min}}^3}.$$
(4)
where we have used the form of the dipole moment $`p_i=qv_i`$. (Note that $`r_{\mathrm{min}}`$ is of the order of the lattice spacing.) Since we have replaced the dipolar interaction in three dimensions by the simpler expression $`J_{ij}v_iv_j`$, the numerical constant in (4) is to some extent arbitrary. Using a different scheme we would obtain $`2/3`$ instead of the factor $`3/4\pi `$. Because of the $`J_{ij}r_{ij}^3`$ scaling, the lower bound $`J_{\mathrm{max}}/N_0`$ gives the coupling of impurities whose distance is of the order of the sample size. (iii) Lastly, we neglect the distance dependence of the couplings and approximate the random-site character of the defect problem by a mean field model with randomly distributed all-to-all connections. In a further simplifying step, the $`J_{ij}`$-distribution is taken to be of the Gaussian form
$$P(J_{ij})=\sqrt{\frac{N}{2\pi }}\frac{1}{J}\mathrm{exp}\left(\frac{NJ_{ij}^2}{2J^2}\right),$$
(5)
with zero mean $`\overline{J_{ij}}=0`$ and finite second moment,
$$\overline{J_{ij}^2}=J^2/N,$$
(6)
which is determined in such a way that it coincides with that of the actual distribution function $`P_0(J_{ij})`$,
$$𝑑J_{ij}J_{ij}^2P_0(J_{ij})=\frac{1}{N_0}J_{\mathrm{max}}^2.$$
(7)
When putting $`J^2/N=J_{\mathrm{max}}^2/N_0`$ and using $`c=N/N_0`$, we obtain the scaling of the parameter $`J`$ with defect concentration $`c`$,
$$J=\sqrt{c}J_{\mathrm{max}}.$$
(8)
The Gaussian distribution $`P`$ is roughly constant for $`|J_{ij}|`$ smaller than $`J/\sqrt{N}`$ and vanishes rapidly for higher values, whereas the more precise function $`P_0`$ shows a power law behavior. We repeat that, with (8), their second moments are identical.
Within our scalar approximation, we take crystal symmetries into account by demanding that the one-particle potential satisfies the condition $`G(v_i)=G(v_i)`$. More specifically we choose $`G(v_i)`$ to exhibit degenerate minima at finite displacement $`v_i=\pm a`$. As the simplest choice of an on-site potential with these properties we use
$$G(v)=g(v^2a^2)^2.$$
(9)
The coupling constant $`g`$ is chosen in such a way that the tunnel splitting $`\mathrm{\Delta }_0`$ for impurities moving in an isolated double well of the form (9) reproduces experimentally observed results for the lowest excitation energy of an isolated defect. E.g., in the case of <sup>7</sup>Li in a KCl host, one would require $`\mathrm{\Delta }_01.1`$ K, given a value of $`a0.7`$ Å.
Due to the simplifications introduced above, the interaction energy (2) of the defect Hamiltonian has been made to resemble that of the SK spin-glass model, albeit one with continuous degrees of freedom rather than Ising spins. Models of this type have recently been proposed as candidates for describing low temperature anomalies of glassy and amorphous systems. Indeed, it has been shown that a frustrated interaction of the form considered here is able to produce a potential energy landscape comprising an ensemble of single- and double well configurations – the latter with a broad distribution of asymmetries and, in the translationally invariant case, also barrier heights, even if one starts out with single-site potentials $`G(v)`$ which are a-priori of single well form.
Our model of interacting impurities can be analyzed exactly within mean–field theory, i.e. a self-consistent representation of the interaction energy as a sum of effective single-site potential energies
$$U_{\mathrm{int}}(\{v_i\})\underset{i}{}U_{\mathrm{eff}}(v_i)$$
(10)
becomes exact in the thermodynamic limit. The ensemble of effective single site potentials $`U_{\mathrm{eff}}(v_i)`$ represents the potential energy landscape of the system of interacting defects. It will be found to contain randomly varying parameters whose distribution can be computed. By quantizing the defect motion within the collectively determined ensemble of effective single site potentials $`U_{\mathrm{eff}}(v_i)`$, one finally obtains a semiclassical description of interaction effects on the behavior of quantum impurities.
We have organized the remainder of our material as follows. In Sec. II we analyze the potential energy landscape of the interacting system within mean-field theory. The analysis follows a proposal previously advocated for glasses. Sec. III describes the analytic solution of the mean-field theory in the weak coupling regime. Thermodynamic consequences of interaction effects are explored in Sec. IV, and simplifying features of two-state and WKB approximations are discussed in Sec. V. Sec. VI is devoted to dynamic effects, in particular to a computation of the distribution of relaxation rates, and to an analysis of the dynamic susceptibility. We discuss our results in some detail in Sec. VII, comparing them with those of complementary approaches, and with experiments, and close with a brief summary in Sec. VIII.
## II Mapping out the Potential Energy Surface
To map out the potential energy surface of the interacting system, one computes the configurational free energy
$`f_N(\beta )=(\beta N)^1\mathrm{ln}{\displaystyle \underset{i}{}dv_i\mathrm{exp}[\beta U_{\mathrm{pot}}(\{v_i\})]},`$ (11)
using replica theory to average over the ensemble of random $`J_{ij}`$ matrices, so as to get typical results. The $`T=0`$ limit is eventually taken to select one of the (possibly many) classical ground-state configurations of the interacting system.
As announced above, a mean-field decoupling produces a collection of independent single-site potentials $`U_{\mathrm{eff}}(v_i)`$ with random parameters (which are comparable to randomly varying local fields in the context of spin-glasses). Technically, this decoupling within replica theory can be seen as a method for the self-consistent determination of the distribution of these random parameters.
We shall not repeat here the details of such a calculation, as they follow standard lines of reasoning . One obtains $`f(\beta )=lim_{n0}f_n(\beta )`$ for the quenched free energy, with
$`nf_n(\beta )`$ $`=`$ $`{\displaystyle \frac{1}{4}}\beta J^2{\displaystyle \underset{a,b}{}}q_{a,b}^2`$ (13)
$`\beta ^1\mathrm{ln}{\displaystyle \underset{a}{}dv^a\mathrm{exp}\left[\beta U_{\mathrm{eff}}(\{v^a\})\right]}.`$
Here
$$U_{\mathrm{eff}}(\{v^a\})=\frac{1}{2}\beta J^2\underset{a,b}{}q_{ab}v^av^b+\underset{a}{}G(v^a)$$
(14)
is an effective replicated single-site potential, and the order parameters $`q_{ab}=N^1_iv_i^av_i^b`$ are determined as solutions of the fixed point equations
$$q_{ab}=v^av^b,a,b=1,\mathrm{},n,$$
(15)
where angular brackets denote a Gibbs average corresponding to the effective replica potential (14), and where it is understood that the limit $`n0`$ is eventually to be taken.
We are, in what follows, going to solve the self-consistency equations only within the so-called replica symmetric (RS) ansatz for order parameters
$$q_{aa}=\widehat{q},q_{ab}=q,ab.$$
(16)
The two order parameters of the RS ansatz must satisfy
$$\widehat{q}=v^2_z,q=v^2_z,$$
(17)
in which $`\mathrm{}_z`$ denotes an average over a zero-mean unit-variance Gaussian $`z`$ while $`\mathrm{}`$ without subscript is a Gibbs average generated by the effective replica-symmetric single–site potential
$`U_{\mathrm{RS}}(v)=J\sqrt{q}zv{\displaystyle \frac{1}{2}}J^2𝒞v^2+G(v).`$ (18)
with $`𝒞=\beta (\widehat{q}q)`$. The RS single site potential contains a Gaussian random variable $`z`$. A sum of single site potentials randomly drawn from the RS Gaussian ensemble (18) constitutes the mean-field representation (10) of the potential energy landscape in terms of an ensemble of effective single site potentials. Before analyzing this ensemble in greater detail, let us note its two most salient features.
First, due to a collective effect mediated by the interaction, there is a systematic deepening of the double-well crystal field potential experienced by the defects. Second, there is an induced distribution of asymmetries owing to the linear contribution to $`U_{\mathrm{RS}}(v)`$. Note that the deepening of the double well potential will in particular give rise to a renormalization of the tunnel splitting: $`\mathrm{\Delta }_0\stackrel{~}{\mathrm{\Delta }}_0<\mathrm{\Delta }_0`$, reducing the smallest excitation energy that occurs in the system of tunneling impurities. However, due to the spectrum of asymmetries there will also be a spread of excitation energies towards larger values, as we shall describe in greater detail below.
Either of the fixed point equations for $`\widehat{q}`$ or $`q`$ above may be replaced by one for $`𝒞=\beta (\widehat{q}q)`$
$`𝒞={\displaystyle \frac{1}{J\sqrt{q}}}{\displaystyle \frac{d}{dz}}v_z={\displaystyle \frac{1}{J\sqrt{q}}}zv_z,`$ (19)
which turns out to aquire a finite value in the $`\beta \mathrm{}`$–limit that is of interest to us here. The RS free energy is
$`f_{\mathrm{RS}}`$ $`(\beta )={\displaystyle \frac{1}{4}}J^2𝒞(\widehat{q}+q)`$ (21)
$`\beta ^1\mathrm{ln}{\displaystyle 𝑑v\mathrm{exp}\left[\beta U_{\mathrm{RS}}(v)\right]}_z.`$
As the $`\beta \mathrm{}`$–limit is taken, Gibbs averages generated by $`U_{\mathrm{RS}}(v)`$ are dominated by the value(s) of $`v`$ which minimize $`U_{\mathrm{RS}}(v)`$, which we denote by $`\widehat{v}=\widehat{v}(z)`$ (displaying its dependence on the value of the Gaussian $`z`$). An immediate consequence is that $`\widehat{q}=q`$ in this limit, provided that $`q0`$. The $`T=0`$ fixed point equations are then
$$\widehat{q}=q=\widehat{v}(z)^2_z,𝒞=\frac{1}{J\sqrt{q}}\frac{d}{dz}\widehat{v}(z)_z$$
(22)
with $`\widehat{v}=\widehat{v}(z)`$ minimizing $`U_{\mathrm{RS}}(v)`$, i.e., to be determined as the appropriate solution(s) of
$$G^{}(\widehat{v})=J\sqrt{q}z+J^2𝒞\widehat{v},$$
(23)
the prime denoting differentiation with respect to $`v`$. The solution $`\widehat{v}(z)`$ of (23) is a smooth function of $`z`$, except at $`z=0`$ where $`\widehat{v}(z)`$ has a jump-discontinuity, owing to the fact that among the solutions of (23) we have to choose the one corresponding to the absolute minimum of $`U_{\mathrm{RS}}`$. For $`z0`$ then, we obtain
$$\frac{d}{dz}\widehat{v}(z)=\frac{J\sqrt{q}}{G^{\prime \prime }(\widehat{v})J^2𝒞}$$
(24)
Taking the jump discontinuity of $`\widehat{v}(z)`$ , hence the $`\delta `$–function singularity of its $`z`$–derivative at $`z=0`$ into account, we obtain the following form of the fixed point equation for $`𝒞`$
$$𝒞=\frac{1}{G^{\prime \prime }(\widehat{v})J^2𝒞}_z+\frac{\mathrm{\Delta }\widehat{v}(0)}{J\sqrt{2\pi q}},$$
(25)
where $`\mathrm{\Delta }\widehat{v}(0)`$ denotes the size of the jump of $`\widehat{v}`$ at $`z=0`$. The $`T=0`$ limit of the free energy, i.e., the internal energy $`u`$ is
$$u=\frac{1}{4}J^2𝒞(\widehat{q}+q)+U_{\mathrm{RS}}(\widehat{v}(z))_z.$$
(26)
Owing to symmetry, the fixed point equations always admit of a $`q=0`$ solution. In a situation, where $`G(v)`$ is of double–well form (with minima at $`\pm a`$), however, only the $`q0`$ solution is thermodynamically acceptable as $`T0`$. Using the fixed point equations, we find that the internal energy is given by
$$u=J^2𝒞\widehat{q}+G(\widehat{v}(z))_z.$$
(27)
in this limit.
## III The Low–Concentration or Weak–Coupling Regime
The weak–coupling (low–concentration) limit is formally defined by the inequality $`Ja^2/ga^41`$. It states that typical interaction energies are much smaller than the bare classical barrier height of the on–site potential $`G(v)`$. In this limit, one has approximately $`\widehat{v}(z)\stackrel{~}{a}\mathrm{sgn}(z)`$ for moderate values of the Gaussian $`z`$, hence $`\widehat{q}=q\stackrel{~}{a}^2`$ and $`𝒞J^1\sqrt{2/\pi }`$. Here $`\pm \stackrel{~}{a}`$ denotes the coordinates of the minima of $`U_{\mathrm{RS}}(v)`$ at $`z=0`$, that is
$$\stackrel{~}{a}=a\left(1+\sqrt{\frac{2}{\pi }}\frac{J}{4ga^2}\right)^{1/2}$$
(28)
In the weak coupling limit, therefore, the effective single site potential experienced by the defects is of the form
$$U_{\mathrm{RS}}(v)=J\stackrel{~}{a}zv\frac{J}{\sqrt{2\pi }}v^2+G(v),$$
(29)
exhibiting both, the systematic deepening of the double well potential experienced by the defects, and the induced distribution of asymmetries due to the linear contribution to $`U_{\mathrm{RS}}(v)`$ mentioned above. Both scale linearly with $`J`$, hence with the square root of the defect concentration in the low concentration regime (see (8)).
As Figure 1 shows, the results of the low concentration approximation agree fairly well with those of a full numerical solution of the fixed point equations for all concentrations of interest.
## IV Thermodynamics at Low Temperatures
The contribution of the defects to the thermodynamics of the system at low temperatures is dominated by quantum effects. That is, one has to consider the quantized motion of the defects in the effective single site potentials given by (18) or by their weak-coupling approximations (29). The first task therefore consists in determining the energy levels $`E_n`$ and the corresponding eigenstates $`\psi _n`$ for the ensemble of effective single site Hamiltonians
$$_{\mathrm{eff}}=\frac{p^2}{2m}+U_{\mathrm{RS}}(v).$$
(30)
This done, one may proceed to compute densities of state. Normalized with respect to the total number $`N_0`$ of lattice sites, this gives
$$\rho _n(E)=c\delta (E\stackrel{~}{E}_n)_z,$$
(31)
where $`\stackrel{~}{E}_n=E_nE_0`$. The defect contribution to the specific heat in the same normalization gives
$$C=ck_\mathrm{B}\beta ^2_{\mathrm{eff}}^2_{\mathrm{eff}}^2_z,$$
(32)
in which now
$$\mathrm{}=\frac{\mathrm{Tr}(\mathrm{})\mathrm{exp}(\beta _{\mathrm{eff}})}{\mathrm{Tr}\mathrm{exp}(\beta _{\mathrm{eff}})}.$$
(33)
Another quantity of interest is the static (dielectric) susceptibility, originating from the interaction of the defect-dipoles with an external field $``$, $`_{\mathrm{ext}}=q_iv_i`$. The static Kubo-formula gives
$`\chi =c`$ $`\beta q^2v_{mm}^2v_{mm}^2`$ (35)
$`{\displaystyle \frac{1}{𝒵_{\mathrm{eff}}}}{\displaystyle \underset{mn}{}}{\displaystyle \frac{e^{\beta E_m}e^{\beta E_n}}{\beta E_m\beta E_n}}v_{mn}^2_z.`$
The $`v_{mn}=(\psi _n,v\psi _m)`$ denote matrix elements of the position operator between the various eigenstates of the effective Hamiltonian $`_{\mathrm{eff}}`$, and $`𝒵_{\mathrm{eff}}`$ is the canonical partition sum generated by it.
As shown in Figure 2, the lowest band of excitation energies, originating from tunneling-excitations in the ensemble of asymmetric double-well potentials extends to much lower energies than the other excitations. The latter are related to harmonic excitations about the two minima of $`G`$ with energies $`\mathrm{}\omega _0`$ only beyond 100 K. The pair of bands with peaks near 99 K and 125 K again corresponds to a pair of states mixed due to tunneling between the wells. The other two bands (with peaks near 192 K and 260 K correspond to states with energies above the barrier.
Tunneling excitations within the space spanned by the oscillator ground states in the two wells will dominate the low temperature physics. This is nicely seen in the specific heat data exhibited in Figure 4. Notice that the Schottky peak at low concentration is modified through interaction effects. As $`c`$ is increased beyond 1000 ppm, a range of temperatures develops where the specific heat starts to show a linear temperature dependence much like in glasses. This is due to the fact that the density of states corresponding to the (lowest band of) tunneling excitations shown in Figure 2 becomes nearly constant in the energy range $`\stackrel{~}{\mathrm{\Delta }}_0<E<\sqrt{c}J_{\mathrm{max}}a^2`$, which covers a large range of energies, as the concentration is sufficiently increased; see Figure 3 which also exhibits the renormalization of the tunneling matrix element towards lower energies due to interaction effects.
In the static susceptibility plotted in Figure 5, no significant contribution of higher order excitations is detectable (on the scale of the figure) at all, even up to temperatures as high as 60 K. The same quantity evaluated in a two-state approximation would be indistinguishable from what is shown here. Note that the concentration dependence at low temperatures is proportional to $`c`$ at low concentrations, but crosses over to a $`\sqrt{c}`$ behavior at larger concentrations. We shall return to this in greater detail later on.
## V Two-state approximation
At low temperatures, $`k_\mathrm{B}T\mathrm{}\omega _0`$, only the two lowest lying states are significantly populated. For a sufficiently high barrier they can be constructed in terms of a basis of pocket states $`|L`$ and $`|R`$ that are localized about the potential minima at $`\pm \stackrel{~}{a}`$ of the two wells.
The corresponding Hamilton matrix reads
$$=\frac{1}{2}\left(\begin{array}{cc}\mathrm{\Delta }& \stackrel{~}{\mathrm{\Delta }}_0\\ \stackrel{~}{\mathrm{\Delta }}_0& \mathrm{\Delta }\end{array}\right),$$
(36)
where the off-diagonal elements are given by the renormalized tunnel energy. The diagonal entries account for an asymmetry energy between the two wells, $`\mathrm{\Delta }=[U_{\mathrm{RS}}(\stackrel{~}{a})U_{\mathrm{RS}}(\stackrel{~}{a})]`$; with (29) we find
$$\mathrm{\Delta }=2J\stackrel{~}{a}^2z\overline{\mathrm{\Delta }}z.$$
(37)
The two-state Hamiltonian is easily diagonalized. Its eigenvalues are given by $`\pm \frac{1}{2}E`$ with
$$E=\sqrt{\stackrel{~}{\mathrm{\Delta }}_{0}^{}{}_{}{}^{2}+\mathrm{\Delta }^2};$$
(38)
the corresponding eigenstates $`|\pm `$ may be written in terms of a mixing angle $`\mathrm{tan}\varphi =\stackrel{~}{\mathrm{\Delta }}_0/\mathrm{\Delta }`$,
$`|+`$ $`=`$ $`\mathrm{cos}\varphi |L+\mathrm{sin}\varphi |R,`$ (39)
$`|`$ $`=`$ $`\mathrm{sin}\varphi |L\mathrm{cos}\varphi |R.`$ (40)
Note that the asymmetry energy $`\mathrm{\Delta }`$ is linear in the Gaussian variable $`z`$; its distribution law is thus given by
$$P_\mathrm{\Delta }(\mathrm{\Delta })=\frac{1}{\sqrt{2\pi }\overline{\mathrm{\Delta }}}\mathrm{exp}\left(\frac{\mathrm{\Delta }^2}{2\overline{\mathrm{\Delta }}^2}\right).$$
(41)
As a consequence, both the two-state energy splitting $`E`$ and the mixing angle $`\varphi `$ are spread over a range determined by this distribution function.
The resulting density of states for the lowest band of excitation energies ($`\rho (E)=\rho _1(E)`$ in the notation of the previous Section)
$$\rho (E)=c\delta \left(E\sqrt{\stackrel{~}{\mathrm{\Delta }}_{0}^{}{}_{}{}^{2}+\mathrm{\Delta }^2}\right)_z$$
(42)
is easily calculated. As is obvious from (38), we have $`\rho (E)=0`$ for $`E<\stackrel{~}{\mathrm{\Delta }}_0`$; at larger energies it is determined by the distribution of asymmetries, so
$$\rho (E)=c\frac{2E}{\sqrt{E^2\stackrel{~}{\mathrm{\Delta }}_{0}^{}{}_{}{}^{2}}}P_\mathrm{\Delta }\left(\sqrt{E^2\stackrel{~}{\mathrm{\Delta }}_{0}^{}{}_{}{}^{2}}\right)$$
(43)
for $`E\stackrel{~}{\mathrm{\Delta }}_0`$, and shows a square root singularity for $`E\stackrel{~}{\mathrm{\Delta }}_0`$; the pre-factor 2 is due the restriction $`E>0`$.
At very low concentration, the width of $`P_\mathrm{\Delta }(\mathrm{\Delta })`$ is much smaller than the tunnel energy $`\stackrel{~}{\mathrm{\Delta }}_0`$, and $`\rho (E)`$ is sharply peaked about $`\stackrel{~}{\mathrm{\Delta }}_0`$. As soon as the typical interaction energy $`\overline{\mathrm{\Delta }}2Ja^2`$ exceeds the tunnel energy $`\stackrel{~}{\mathrm{\Delta }}_0`$, $`\rho (E)`$ is well approximated by a Gaussian above $`\stackrel{~}{\mathrm{\Delta }}_0`$. It remains to compute the renormalized tunnel energy $`\stackrel{~}{\mathrm{\Delta }}_0`$.
With respect to the computation of $`\stackrel{~}{\mathrm{\Delta }}_0`$, we have looked at two variants. In the first, both $`\mathrm{\Delta }_0`$ and its renormalized value $`\stackrel{~}{\mathrm{\Delta }}_0`$ are computed exactly by solving the Schrödinger equation of the impurity moving in the appropriate crystal field potential, i.e., $`G(v)`$ for the former, and the renormalized potential $`\stackrel{~}{G}(v)=\frac{J}{\sqrt{2\pi }}v^2+G(v)`$ for the latter. As mentioned above, the first computation is actually used to fix the coupling $`g`$ in (9) so as to reproduce the experimentally observed value – e.g., $`\mathrm{\Delta }_01.1`$ K for <sup>7</sup>Li in KCl. With parameters appropriate for the KCl:<sup>7</sup>Li example, this requires a potential with bare barrier height $`ga^4174`$ K.
In the second variant, both $`\mathrm{\Delta }_0`$ and its renormalized value $`\stackrel{~}{\mathrm{\Delta }}_0`$ are computed within a WKB approximation. The bare tunneling matrix element is in this setting usually parameterized as
$$\mathrm{\Delta }_0=\mathrm{}\omega _0\mathrm{exp}(\lambda )$$
(44)
with
$$\lambda =d\sqrt{2mV_B/\mathrm{}^2},$$
(45)
and parameters derived from the bare on–site potential $`G`$ and the mass $`m`$ of the tunneling particle — the frequency $`\omega _0=\sqrt{8ga^2/m}`$ of harmonic oscillations about the minima of $`G`$, the distance $`d=2a`$ between the minima, and the height $`V_B=ga^4\mathrm{}\omega _0/2`$ of the classical barrier above ground state in the two wells. Once more, this first calculation would be used to fix the coupling constant $`g`$ of the bare potential (9). Within this approximation, the classical barrier required to reproduce the bare tunneling energy reported for KCl:<sup>7</sup>Li is $`ga^484`$ K, i.e. only approximately half the value obtained from the numerically exact analysis. The renormalized tunneling matrix element is computed in the same way, except that parameters are computed from the renormalized on–site potential. This gives
$`\stackrel{~}{\omega }_0`$ $`=`$ $`\omega _0\stackrel{~}{a}/a`$ (46)
$`\stackrel{~}{V}_B`$ $`=`$ $`V_B+{\displaystyle \frac{1}{2}}\left(\sqrt{{\displaystyle \frac{2}{\pi }}}Ja^2+\mathrm{}\omega _0\left(1{\displaystyle \frac{\stackrel{~}{a}}{a}}\right)\right)`$ (47)
$`\stackrel{~}{d}`$ $`=`$ $`2\stackrel{~}{a}`$ (48)
to lowest order in small parameters (with $`\stackrel{~}{a}/a1+J/(4\sqrt{2\pi }ga^2`$).
The renormalized tunnel energy $`\stackrel{~}{\mathrm{\Delta }}_0=\mathrm{}\stackrel{~}{\omega }_0\mathrm{e}^{\stackrel{~}{\lambda }}`$, with
$$\stackrel{~}{\lambda }=\stackrel{~}{d}\sqrt{2m\stackrel{~}{V}_B/\mathrm{}^2}$$
(49)
may thus be expressed as
$$\stackrel{~}{\mathrm{\Delta }}_0=\mathrm{\Delta }_0\sqrt{1+\epsilon }\mathrm{e}^\delta .$$
(50)
Here
$$\epsilon =\sqrt{\frac{2}{\pi }}\frac{Ja^2}{4ga^4}$$
(51)
and
$$\delta =\lambda \frac{\epsilon }{2}\left(1+2\frac{ga^4}{V_B}\frac{\mathrm{}\omega _0}{4V_B}\right)+𝒪(\epsilon ^2),$$
(52)
the latter quantity being evaluated in the weak coupling approximation, $`Ja^2ga^4`$ which is always appropriate for the case at hand. If, moreover, the barrier is assumed to be high, $`\mathrm{}\omega _0V_B`$, so that there are many oscillator levels between the potential minima at $`v\pm a`$ and the top of the barrier at $`v=0`$, the expression simplifies to
$$\delta \lambda \frac{3\epsilon }{2}=\lambda \frac{3}{8}\sqrt{\frac{2}{\pi }}\frac{Ja^2}{ga^4}\sqrt{c}3\sqrt{\frac{2}{\pi }}\frac{J_{\mathrm{max}}a^2}{\mathrm{}\omega _0},$$
(53)
in which corrections involving powers of $`Ja^2/ga^4`$ and $`\mathrm{}\omega _0/ga^4`$ have been neglected. There is no restriction on the ratio $`Ja^2/\mathrm{}\omega _0`$. It turns out, however, that in the KCl:Li case the ratio $`\mathrm{}\omega _0/ga^4`$ is roughly 1.2, thus not small, so that the corresponding simplifications are not available. Figure 6 compares the ratio $`\stackrel{~}{\mathrm{\Delta }}_0/\mathrm{\Delta }_0`$ evaluated numerically and via the WKB approximation (44), (45) and (50)-(52). The renormalization of the tunnel energy becomes noticeable when the concentration exceeds 100 ppm, a concentration at which the typical asymmetry $`\overline{\mathrm{\Delta }}`$ becomes comparable with the vibrational energy $`\mathrm{}\omega _0`$ in the two wells; the renormalization effect is, however, overestimated within the WKB approximation.
In the two-state approximation, specific heat and static susceptibility are given by
$$C=ck_\mathrm{B}dE\rho (E)\frac{(\beta E/2)^2}{\mathrm{cosh}^2(\beta E/2)}.$$
(54)
and
$`\chi =c\beta (q\stackrel{~}{a})^2`$ $`{\displaystyle \frac{(\mathrm{\Delta }/E)^2}{\mathrm{cosh}^2(\beta E/2)}}`$ (56)
$`+2k_\mathrm{B}T{\displaystyle \frac{\stackrel{~}{\mathrm{\Delta }}_{0}^{}{}_{}{}^{2}}{E^3}}\mathrm{tanh}(\beta E/2)_z,`$
respectively. The results are valid as long as temperature is much smaller than the librational energy $`\mathrm{}\omega _0`$. In (56) we have also introduced the usual approximate representation $`v=\stackrel{~}{a}\sigma _z`$ of the position operator in a basis of pocket states. This latter approximation is the largest source of errors in the expression (56) for the susceptibility in that it overestimates matrix elements by roughly 10%. It turns out tat this has a noticeable effect only on an overall pre-factor (at least for isolated or weakly coupled defects), but not on the temperature dependence.
## VI Dynamics
### A Phonon damping
The interaction of an impurity at site $`i`$ with elastic waves is described by the coupling potential
$$\gamma \epsilon v_i/a,$$
(57)
with the elastic deformation potential $`\gamma `$, the reduced coordinate $`v_i/a`$, and the phonon strain field
$$\epsilon =\underset{𝒒,s}{}\sqrt{\frac{\mathrm{}}{2V\varrho \omega _{𝒒s}}}qi\left(b_{𝒒s}b_{𝒒s}^{}\right),$$
(58)
where $`\varrho `$ is the mass density of the host crystal, and $`𝒒`$ labels the wave vector of three acoustic phonon branches $`s`$. There is ample evidence that the defect-phonon coupling is weak; therefore it may be treated in first Born approximation.
All dynamic information may be obtained from the two-time correlation function
$$G(t)=(1/2v^2)v(0)v(t)+v(t)v(0),$$
(59)
with the normalization condition $`G(0)=1`$. At low temperatures, we may use the two-state approximation. We shall here restrict ourselves to the simplified representation $`v=\stackrel{~}{a}\sigma _z`$ of the position operator in the basis of pocket states as introduced above; it is sufficiently precise to give qualitatively reliable results.
The theory of a weakly damped two-state system has been derived in many places; here we merely quote the result for the correlation function,
$`G(t)`$ $`=`$ $`{\displaystyle \frac{\stackrel{~}{\mathrm{\Delta }}_{0}^{}{}_{}{}^{2}}{E^2}}\mathrm{cos}(Et/\mathrm{})e^{\frac{1}{2}\mathrm{\Gamma }t}`$ (61)
$`+{\displaystyle \frac{\mathrm{\Delta }^2}{E^2}}\left((1Q)e^{\mathrm{\Gamma }t}+Q\right),`$
with $`Q=\mathrm{tanh}(\beta E/2)^2`$ and the one-phonon damping rate
$$\mathrm{\Gamma }=\frac{1}{2\pi }\frac{3\gamma ^2}{\mathrm{}^4\varrho v_s^5}\stackrel{~}{\mathrm{\Delta }}_{0}^{}{}_{}{}^{2}E\mathrm{coth}(\beta E/2).$$
(62)
Here $`v_s`$ denotes the sound velocity. $`G(t)`$ determines the linear response of a two-state system with dipole moment $`p=qa\sigma _z`$, tunnel energy $`\stackrel{~}{\mathrm{\Delta }}_0`$ and asymmetry $`\mathrm{\Delta }`$. There are several interesting issues arising from the above model.
First, the Gaussian distribution for the bias $`\mathrm{\Delta }`$ leads to a distribution of resonance energies $`E`$ and relaxation rates $`\mathrm{\Gamma }`$.
Second, because of the variation of the typical asymmetry $`\overline{\mathrm{\Delta }}`$ with concentration $`c`$, the relaxation behavior of the impurities changes significantly with increasing $`c`$. From (61) it is immediately clear that there is no significant relaxation contribution for $`\overline{\mathrm{\Delta }}\stackrel{~}{\mathrm{\Delta }}_0`$, i.e., at low concentration. In the opposite case $`\overline{\mathrm{\Delta }}\stackrel{~}{\mathrm{\Delta }}_0`$, the oscillatory (resonant) part of $`G(t)`$ is insignificant, and the relaxation term governs the impurity dynamics.
Third, the temperature factor of the relaxation contribution, $`1Q=\mathrm{cosh}(\beta E/2)^2`$, vanishes at very low temperatures $`k_\mathrm{B}TE`$, whereas it tends towards unity at higher $`T`$ and large concentration.
### B Rate distribution
Recent measurements of the dielectric constant of KCL:Li revealed relaxational motion of the lithium impurities over several decades in the kHz range. Such a broad relaxation spectrum is characteristic for disordered system in general. Therefore we discuss in some detail the distribution of relaxation rates
$$P_\mathrm{\Gamma }(\mathrm{\Gamma })=\frac{1}{N}\underset{i}{}\delta (\mathrm{\Gamma }\mathrm{\Gamma }_i).$$
(63)
In our two-state model there is only one variable, $`\mathrm{\Delta }=z\overline{\mathrm{\Delta }}`$, which is entirely determined by the Gaussian distribution of $`z`$ and the constant $`\overline{\mathrm{\Delta }}`$; cf. (38).
We give explicitly the rate distribution at zero temperature, because of its simplicity and since it shows the essential features. At $`T=0`$, Eq. (62) gives
$$\mathrm{\Gamma }=\frac{\stackrel{~}{\mathrm{\Gamma }}_0}{\stackrel{~}{\mathrm{\Delta }}_0}E$$
(64)
with
$$\stackrel{~}{\mathrm{\Gamma }}_0=\frac{\stackrel{~}{\mathrm{\Delta }}_{0}^{}{}_{}{}^{3}}{\mathrm{\Delta }_{0}^{}{}_{}{}^{3}}\mathrm{\Gamma }_0\mathrm{and}\mathrm{\Gamma }_0=\frac{1}{2\pi }\frac{3\gamma ^2}{\mathrm{}^4\varrho v_s^5}\mathrm{\Delta }_0^3.$$
(65)
Hence, except for scales, the rate distribution is at the presently chosen level of description basically equivalent to the density of states in the two-level approximation. Substituting the Gaussian variable $`z`$ appearing in the energy $`E`$, we find
$`P_\mathrm{\Gamma }(\mathrm{\Gamma })`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2\pi }}}{\displaystyle \frac{\stackrel{~}{\mathrm{\Delta }}_0}{\overline{\mathrm{\Delta }}\stackrel{~}{\mathrm{\Gamma }}_0}}{\displaystyle \frac{2\mathrm{\Gamma }}{\sqrt{\mathrm{\Gamma }^2\stackrel{~}{\mathrm{\Gamma }}_0^2}}}`$ (67)
$`\mathrm{exp}\left[{\displaystyle \frac{1}{2}}{\displaystyle \frac{\stackrel{~}{\mathrm{\Delta }}_0^2}{\overline{\mathrm{\Delta }}^2\stackrel{~}{\mathrm{\Gamma }}_0^2}}\left(\mathrm{\Gamma }^2\stackrel{~}{\mathrm{\Gamma }}_0^2\right)\right]`$
for $`\mathrm{\Gamma }\stackrel{~}{\mathrm{\Gamma }}_0`$.
Here, $`\stackrel{~}{\mathrm{\Gamma }}_0`$ is the minimum rate at finite doping. According to (65), it is reduced by a factor $`\stackrel{~}{\mathrm{\Delta }}_{0}^{}{}_{}{}^{3}/\mathrm{\Delta }_{0}^{}{}_{}{}^{3}`$ as compared to the rate in the dilute limit, $`\mathrm{\Gamma }_0`$. Like the density of states, $`P_\mathrm{\Gamma }(\mathrm{\Gamma })`$ contains a Gaussian factor and a square root singularity at the minimum value $`\stackrel{~}{\mathrm{\Gamma }}_0`$. The latter is of little relevance, and the rate distribution is governed the exponential in (67). We point out the most salient features of the rate distribution.
(i) For low impurity concentration, we have $`\stackrel{~}{\mathrm{\Delta }}_0\overline{\mathrm{\Delta }}`$, and $`P_\mathrm{\Gamma }(\mathrm{\Gamma })`$ is a sharply peaked function with the lower cut-off $`\stackrel{~}{\mathrm{\Gamma }}_0`$. In the dilute limit $`c0`$, the distribution tends towards a delta function at $`\mathrm{\Gamma }_0`$. However, already at concentrations as low as 1 ppm, the rate distribution has a visible tail towards higher rates due to the presence of asymmetries (see Figure 7).
(ii) In the opposite case of high doping, the maximum asymmetry exceeds by far the reduced tunnel energy, $`\stackrel{~}{\mathrm{\Delta }}_0\overline{\mathrm{\Delta }}`$. As a consequence, the Gaussian factor in (67) leads to a wide distribution that is almost constant between the lower bound $`\stackrel{~}{\mathrm{\Gamma }}_0`$ and the effective upper cut-off $`(\overline{\mathrm{\Delta }}/\stackrel{~}{\mathrm{\Delta }}_0)\stackrel{~}{\mathrm{\Gamma }}_0`$.
(iii) The lower bound of the distribution, $`\stackrel{~}{\mathrm{\Gamma }}_0`$, depends on the impurity concentration. In the dilute case, the tunnel energy is given by the bare value $`\mathrm{\Delta }_0`$. Yet at finite concentration, the renormalization of the tunnel energy reduces the rate by the factor $`\stackrel{~}{\mathrm{\Delta }}_{0}^{}{}_{}{}^{3}/\mathrm{\Delta }_{0}^{}{}_{}{}^{3}`$. At $`c=0.01`$, the minimum rate is by roughly one order of magnitude smaller than in the dilute case. (The WKB approximation predicts more than two orders of magnitude; see Figure 6). Thus interaction leads to both slow and very fast relaxation, as compared to the low-doping case.
Figure 7 shows the evolution from low to high doping. For concentrations up to 100 ppm, the curves have the same shape as those for the density of states (as predicted within the two-level approximation using the pocket state approximation for the position operator. The curve for 1000 ppm develops a different shape at high rates, because the pocket state representation of the position operator becomes less and less precise at large asymmetries (compare Figures 7 and 3).
### C Dynamic susceptibility
Experimental investigations of the two-state dynamics in terms of elastic or dielectric response function involve the dynamical susceptibility $`\chi (\omega )=\chi ^{}(\omega )+i\chi ^{\prime \prime }(\omega )`$, whose spectral function
$$\chi ^{\prime \prime }(\omega )=c\frac{2}{\mathrm{}}\mathrm{tanh}(\beta \mathrm{}\omega /2)G^{\prime \prime }(\omega )_z$$
(68)
is related to the average of the motional spectrum $`G^{\prime \prime }(\omega )`$, i.e. of the Fourier transform of (61). (The additional prefactor $`(q\stackrel{~}{a})^2`$ appearing in the susceptibility (56) and (35) is due to the fact that in Secs. III and IV we have considered an external field coupling to the dipole operator $`qv`$ rather than to a normalized position operator $`v/\sqrt{v^2}`$ whose correlator is considered in (59)).
Here we give explicitly the real part that describes the sound velocity or the reactive part of the dielectric function. For relevant frequencies $`\mathrm{}\omega \stackrel{~}{\mathrm{\Delta }}_0`$ (and relaxation rates at typical concentrations satisfying $`\mathrm{\Gamma }_i\stackrel{~}{\mathrm{\Delta }}_0`$) one has
$`\chi ^{}(\omega )`$ $``$ $`{\displaystyle \frac{2}{N_0}}{\displaystyle \underset{i}{}}{\displaystyle \frac{\stackrel{~}{\mathrm{\Delta }}_{0}^{}{}_{}{}^{2}}{E_i^3}}\mathrm{tanh}(\beta E_i/2)`$ (70)
$`+{\displaystyle \frac{\beta }{N_0}}{\displaystyle \underset{i}{}}{\displaystyle \frac{\mathrm{\Delta }_i^2}{E_i^2}}\mathrm{cosh}(\beta E_i/2)^2{\displaystyle \frac{\mathrm{\Gamma }_i^2}{\omega ^2+\mathrm{\Gamma }_i^2}}`$
within the two-state approximation. There are two contributions of different origin to the susceptibility. The first or “resonant” part is dominant in the dilute limit, where the typical asymmetry is significantly smaller than the tunnel energy, $`\overline{\mathrm{\Delta }}\stackrel{~}{\mathrm{\Delta }}_0`$. On the other hand, the second contribution prevails at strong doping, where $`\overline{\mathrm{\Delta }}\stackrel{~}{\mathrm{\Delta }}_0`$.
In Fig. 8 we plot $`\chi ^{}(\omega )`$ as a function of temperature for various impurity concentrations $`c`$. At low doping, the susceptibility is of the van Vleck type and hardly depends on the external frequency. The temperature variation is given by the occupation difference of the two levels, $`\mathrm{tanh}(\stackrel{~}{\mathrm{\Delta }}_0/kT)`$; accordingly $`\chi ^{}(\omega )`$ is constant for $`kT<\stackrel{~}{\mathrm{\Delta }}_0`$. At larger doping, pronounced relaxational contributions occur, with peak-positions strongly dependent on frequency.
The zero-temperature value of $`\chi ^{}`$ is a convenient measure of the relevance of interaction effects. The temperature factor in the relaxation contribution vanishes at $`T=0`$, and one has
$$\chi ^{}=c\frac{1}{\stackrel{~}{\mathrm{\Delta }}_0}(1+z^2\overline{\mathrm{\Delta }}^2/\stackrel{~}{\mathrm{\Delta }}_{0}^{}{}_{}{}^{2})^{3/2}_z.$$
(71)
The average over $`z`$ can be written in terms of the confluent hyper-geometric function $`U(a,b,x)`$,
$$\chi ^{}=c\frac{1}{\stackrel{~}{\mathrm{\Delta }}_0}\frac{1}{\sqrt{2\nu ^2}}U(\frac{1}{2},0,\frac{1}{2\nu ^2}),$$
(72)
where we have defined the dimensionless coupling parameter $`\nu =(\overline{\mathrm{\Delta }}/\stackrel{~}{\mathrm{\Delta }}_0)`$,
$$\nu =\sqrt{c}\mathrm{\hspace{0.17em}2}\frac{J_{\mathrm{max}}\stackrel{~}{a}^2}{\stackrel{~}{\mathrm{\Delta }}_0}\sqrt{c}\mathrm{\hspace{0.17em}2}\frac{J_{\mathrm{max}}a^2}{\stackrel{~}{\mathrm{\Delta }}_0}\sqrt{\frac{c}{c_0}}.$$
(73)
For weak doping, the asymmetry $`\overline{\mathrm{\Delta }}`$ is much smaller than the tunnel energy; with $`\nu 1`$ and $`U(1/2,0,1/2\nu ^2)\sqrt{2}\nu `$ we recover the obvious result $`\chi =c/\mathrm{\Delta }_0`$. In the opposite case of strong doping we have $`\nu 1`$ and $`U(1/2,0,1/\nu ^2)=u_{\mathrm{}}=1.128\mathrm{}`$ From (72) we thus obtain in the limiting cases
$$\chi ^{}=\{\begin{array}{cc}c/\stackrel{~}{\mathrm{\Delta }}_0\hfill & \text{for }cc_0\hfill \\ c/\overline{\mathrm{\Delta }}=u_{\mathrm{}}\sqrt{cc_0}/\stackrel{~}{\mathrm{\Delta }}_0\hfill & \text{for }cc_0\hfill \end{array}.$$
(74)
The cross-over occurs at a concentration $`c_0`$ where the asymmetry attains the value of the tunnel energy, $`\overline{\mathrm{\Delta }}\stackrel{~}{\mathrm{\Delta }}_0`$, or $`\nu 1`$. Eq. (72) yields
$$c_0=\left(\frac{\stackrel{~}{\mathrm{\Delta }}_0}{2J_{\mathrm{max}}a^2}\right)^2.$$
(75)
Because of the concentration dependence of $`\stackrel{~}{\mathrm{\Delta }}_0`$ this is an implicit equation for $`c_0`$. With parameters appropriate for <sup>7</sup>KCl:Li as used before, one finds $`c_0=𝒪(10^6)`$; compare Figure 9.
The relaxation contribution to (70) strongly depends on the frequency $`\omega `$ and the rates $`\mathrm{\Gamma }_i`$. Relaxation is most efficient where the rate of thermal two-level systems are close to the external frequency $`\omega `$; lowering the frequency shifts the peak to lower temperatures. The relaxation peak is significantly broadened by the rate distribution (67).
## VII Discussion
### A Cross-over to relaxation
Broad spectra of energies and relaxation rates are characteristic for any disordered system. In the case of quantum impurities with dipolar interactions the broadening is tuned by the parameter $`\nu `$, which measures the strength of the interaction in units of the tunnel energy and increases with the square root of the impurity concentration $`c`$; cf. (73).
At low doping the motional spectrum is peaked about the bare tunnel frequency $`\mathrm{\Delta }_0/\mathrm{}`$. The relaxation rates that are close to $`\mathrm{\Gamma }_0`$ as defined in (65) are of little importance, since the weight of the zero-frequency feature of the susceptibility is small. This changes as the impurity concentration reaches the value $`c_0`$, i.e. as $`\nu `$ tends towards unity. Then the average asymmetry energy is comparable to the tunnel energy; as a consequence, a strong relaxation peak emerges, and both the density of states $`\rho (E)`$ and the rate distribution $`P_\mathrm{\Gamma }(\mathrm{\Gamma })`$ broaden; cf. Figs. 3 and 7.
The static part of the susceptibility at $`T=0`$ provides the clearest signature of this cross-over to relaxation. In Fig. 9 we compare the result of the present work (72) with that obtained previously by one of us in a Mori projection scheme,
$$\chi ^{}=\chi _0\frac{(\sqrt{1+\mu ^2}\mu )^2}{\sqrt{1+\mu ^2}},$$
(76)
where the dimensionless coupling constant
$$\mu =c\mathrm{\hspace{0.17em}2}\frac{J_{\mathrm{max}}a^2}{\mathrm{\Delta }_0}$$
(77)
may be considered as the rescaled concentration. (The extra factor 2 in (77) as compared to the definition in Ref. is due to the fact that our convention for the coupling between impurities differs from that used in Ref. by a factor 1/2). The susceptibility in the absence of interaction, $`\chi _0=c/\mathrm{\Delta }_0`$, varies linearly with the concentration; in Fig. 9 it is indicated as a dotted line. Moreover we have indicated experimental data for various doped alkali halides.
At very low doping, both expressions (72) and (76) are linear in the concentration, as expected for non-interacting impurities, whereas at higher impurity density the dipolar interactions of the tunnel systems strongly reduce the susceptibility. We discuss the differences of the theoretical results obtained from the present mean-field model (72) and the Mori projection scheme (76).
(i) Both parameters $`\mu `$ and $`\nu `$ are proportional to the ratio of the maximum interaction $`J_{\mathrm{max}}a^2`$ and the tunnel energy. In the present approach we found a square root dependence $`\nu \sqrt{c}`$, whereas projection scheme yields a linear law $`\mu c`$. (Note that in the expression for $`\nu `$ we have neglected the weak $`c`$-dependence of $`\stackrel{~}{\mathrm{\Delta }}_0`$.)
(ii) As a consequence or these different powers, the cross-over to relaxation occurs at much higher concentration in (76), $`\mu 1`$ or $`c_0=\mathrm{\Delta }_0/J_{\mathrm{max}}a^2`$; the present mean-field theory yields, at $`\nu 1`$, the square of this quantity; cf. (75).
(iii) In the strong-doping limit, the susceptibility in (72) varies as $`\chi ^{}\sqrt{c}`$; it still increases with concentration, whereas (76) results in a decrease, $`\chi ^{}c^2`$.
These discrepancies are easily traced back to the basic assumptions of the models. The present mean-field approach relies on random couplings to all impurities. The characteristic interaction $`J`$ is given by the second moment of the Gaussian distribution and thus is proportional to the square root of the concentration; cf. (8).
In the previous Mori approach , the characteristic interaction is determined by a few nearest neighbors of a given impurity. (Because of the $`r^3`$-dependence on distance of the dipolar interaction and because of its random sign, neighbors at larger distances contribute little to field at a given site.) The $`r^3`$-law results in the linear variation of $`J`$ with concentration.
It should be noted that the square root law (8) can be modified by a change of the mean-field model in a manner that takes the dominant effect of the interactions of a defect with a few nearest neighbors into account. In a mean-field context this might be achieved by replacing the fully connected random interaction matrix by a sparse random matrix such as in the Viana-Bray model, in which typical interaction scales may be chosen to scale linearly with the concentration. In the details, however, such a diluted mean-field model is much more complicated, requiring the introduction of infinitely many order parameters.
### B Comparison with experiment
The most significant result of the present work consists in the cross-over at $`c=c_0`$ from coherent one-particle motion to relaxation driven by the dipolar interaction. (The pre-factor $`(\mathrm{\Delta }/E)^2`$ of the relaxation feature in the susceptibility (70) vanishes for zero interaction, whereas it tends towards unity in the strong-coupling limit.)
Such a cross-over and the emergence of a relaxation peak in the motional spectrum have been observed for various impurity systems such as KCl:Li and NaCl:OH. As discussed above and shown in Fig. 9, the concentration dependence of the susceptibility in the strong-doping limit obtained in the present work, $`\chi ^{}\sqrt{c}`$, differs from the previous result $`\chi ^{}c^2`$ (Ref. ). The data for the highest concentrations ($`\mu >0.1`$) clearly show the relevance of the dipolar interactions; the measured values are orders of magnitude smaller than expected for non-interacting impurities.
In spite of their marked differences of the theoretical laws (72) and (76), the available data do not permit us a definite statement on the power law $`\chi c^\alpha `$. Though (76) provides a better fit at intermediate concentrations, $`0.001<\mu <1`$, the data do not settle the power law at high concentrations. (Unfortunately, presently available crystals do not satisfy the strong doping criterion $`\mu 1`$.) Moreover, it is rather a difficult matter to properly separate the relaxation contribution for strongly doped crystals; thus the data points at the highest concentrations may well be too large. Further experiments at high densities would be most desirable, in view of the above discrepancy between $`\alpha =\frac{1}{2}`$ in the present approach and $`\alpha =2`$ from Ref. .
We now turn to the rate distribution plotted in Fig. 7, that shows a strong broadening with rising concentration. Indeed, the relaxation spectra observed at high concentrations become very broad and develop a significant low-frequency wing . On the other hand, specific heat measurements give clear evidence for a broadened density of states, in qualitative agreement with Fig. 3. (Cf. Ref. and original literature cited therein.)
We close the discussion of alkali halides with a remark on the temperature dependence of the relaxation amplitude in the susceptibility (70). Since our mean-field theory reduces to a set of effective two-state systems, the relaxation feature shows the well-known factor $`\mathrm{cosh}(\frac{1}{2}\beta E)^2`$ and thus vanishes exponentially in the limit $`T0`$, whereas the rate $`\mathrm{\Gamma }`$ tends towards a constant. This behavior is characteristic for a local degree of freedom. The Mori projection approach, on the other hand, results in a constant relaxation amplitude and a rate that decreases at low $`T`$, thus showing the generic behavior of collective relaxation. Again, the experimental situation would seem not entirely conclusive, though the data on KCl:Li of Ref. would suggest that both behaviors are present, i.e. the relaxation spectrum would comprise one-particle and collective contributions.
Finally we discuss recent Raman light scattering experiments on mixed fluorite crystals (MF<sub>2</sub>)<sub>(1-c)</sub>(LF<sub>3</sub>)<sub>c</sub>, with M=Ca,Sr,Ba. The reported data give strong evidence that the coupled density of states $`D(E)`$ of the two-level systems increases linearly with $`c`$ in the range $`c=0.050.5`$, i.e. $`D(E)=cD_0(E)`$; it would seem that the shape function $`D_0(E)`$ does not change in this range.
We consider very likely that the broad distribution of two-level energies observed in Ref. arises from elastic coupling of LF<sub>3</sub> tunneling states. In our model the TLS coupled density of states of is given by
$$D(E)=\frac{\stackrel{~}{\mathrm{\Delta }}_0^2}{E^2}\rho (E).$$
(78)
With the expression (43) for $`\rho (E)`$ one easily finds that $`D(E)`$ varies over a wide range as $`E^2`$ and increases with $`c`$. Though these dependencies on $`E`$ and $`c`$ qualitatively agree with the data of Ref. , this experiment certainly requires a more careful study of the high-doping case. Possible discrepancies with the present description for this particular case arise from the manner in which defect crystals can, or cannot become glassy at high doping, an issue to which we now turn.
### C Crossover to glassy behavior
The appearance in our theory of a broad spectrum of excitation energies and a correspondingly broad spectrum of relaxation rates due to interaction effects at high doping is reminiscent of the physics characteristic of structural glasses. In this sense, our mean-field approach appears to describe a crossover to glassy physics at high doping. Indeed, one of the original motivations for studying defect crystals has been that they constitute systems in which – unlike in glasses – the nature of the tunneling excitations is well understood, and which at the same time offer the possibility of approaching a glassy limit by increasing the defect concentration (for a recent review, see Ref. ). It should, however, be noted that the type of solely interaction-mediated glassiness described in the present paper is different from that believed to describe glasses proper in one essential aspect: while tunneling systems do occur with a broad distribution of asymmetries, there is – unlike in glasses – no corresponding randomness in barrier heights. This feature entails for instance that the typical glassy plateau of the internal friction as a function of temperature would be absent in systems to which the present theory applies. On the other hand, such plateaus are known develop in certain defect crystals such as (CaF<sub>2</sub>)<sub>(1-c)</sub>(LF<sub>3</sub>)<sub>c</sub>, but they require defect concentrations exceeding the 10 % range. There are basically two ways to create the required randomness also in the barrier heights. First, it can be solely interaction-mediated, if interactions are translationally invariant. This mechanism is, however, not available for a system of defects embedded in a crystalline host. The other possibility is to allow the crystal field potentials $`G(v_i)`$ to vary randomly from defect-site to defect-site. For highly doped defect crystals, this would seem like a realistic, and indeed expected feature. Unfortunately, though, there are no good theoretical models around to predict the associated kind of randomness and, in particular, its variation with defect concentration. In this sense, glassy defect crystals are apparently not much simpler systems than structural glasses proper.
### D Comparison with classical reorientation model
The dissipation rate (62) describes jumps between two quantum levels; it may be decomposed as $`\mathrm{\Gamma }=\mathrm{\Gamma }_{}+\mathrm{\Gamma }_{}`$, where $`\mathrm{\Gamma }_{}`$ accounts for thermally activated jumps from the ground state upwards and $`\mathrm{\Gamma }_{}`$ for the reverse process. Detailed balance requires $`\mathrm{\Gamma }_{}=e^{\beta E}\mathrm{\Gamma }_{}`$; thus we obtain with $`EkT`$ the thermally activated rate
$$\mathrm{\Gamma }_{}=\stackrel{~}{\mathrm{\Gamma }}_0e^{\beta E},$$
(79)
where the activation energy is, in the strong-coupling limit, given by the dipolar interaction, $`E=2J\stackrel{~}{a}^2`$. Such a rate, with an appropriate distribution of barrier heights, has been used in previous work on a classical reorientation model for quadrupolar glasses such as KBr<sub>(1-c)</sub>(CN)<sub>c</sub> ; in the classical picture the reorientation of the cyanide molecules (i.e. rotation by $`\pi `$) requires to overcome their quadrupole-quadrupole interaction that corresponds to our $`2J\stackrel{~}{a}^2`$. Thus the present quantum mechanical calculation yields in the classical limit the proper temperature dependence of the rates. (The present rate distribution is, however, much simpler than that derived in Ref. .)
### E Assumptions, Approximations and Potential Extensions
Here we briefly summarize once more the main assumptions and approximations made in this paper, trying to assess their quality, and we indicate possibilities for further improvements.
(i) The 3D local coordinates of the impurities have been replaced by a scalar variable, and the dipolar couplings by scalar couplings with a random sign. In the analysis of dynamical effects we have introduced a further simplification by truncating the Hilbert space of the single-particle effective Hamiltonian to the ground state doublet. Whereas the neglect of two additional spatial coordinates should be of little relevance, in that the simplified model still brings out the salient features of the problem of interacting tunneling impurities, the truncation of the one-particle Hilbert space certainly breaks down as soon as the thermal energy or the mean dipolar interaction exceed the vibrational energy within one well of the impurity crystal potential. Yet this is not a severe constraint for our model since the most interesting physics occurs at low temperatures and for dipolar couplings comparable to the tunnel energy.
(ii) Interactions between tunneling impurities are analyzed classically at the mean field level. This by itself is expected to give a reliable picture for much of the physics, as long as we are not interested in coherent quantum motion of several particles (see below) or critical behaviour associated with phase transitions.
(iii) In actually performing the mean-field analysis we have replaced the distribution (3) by the Gaussian (5) with the same variance. As discussed in Sect. VII-A above, this is certainly a most serious approximation, which is dictated solely by the demand for analytic tractability, and which can not be justified on physical grounds. There would be basically two possibilities to improve on this within random-bond modeling and replica techniques as used in the present paper. One is to consider sparse connectivity models, the other to follow a cavity approach proposed by Cizeau and Bouchaud. However, the former gives rise to an infinite set of order parameters and is quantitatively more or less untractable, the latter uses the two-state nature of Ising spins in ways so essential to the analysis that we have seen no way to apply it to continuous degrees of freedom as they occur in the present problem. Whether techniques invented for models with site-randomness will offer a way out of this dilemma, is currently under study.
Note that the uncertainties with respect to the concentration dependence alluded to above are quite general and arise in related approaches to, e.g., quantum spin sytems .
A possible and uncontrolled ad-hoc modification of the theory which would give a $`c^{1/3}`$ behavior of the $`T=0`$ susceptibility at large doping instead of the $`\sqrt{c}`$ behavior, and which would therefore fit the high doping behaviour of this quantity rather well, is briefly mentioned in our summary below.
(iv) The approximation of replica symmetry used in the quantitative analysis of collective behaviour is not expected to be a serious source of errors. We have checked elsewhere for related systems that replica symmetry breaking effects are small e.g. in the low-temperature specific heat, but also in distributions of parameters characterizing double well-potentials.
(vi) The mean-field model itself constitutes an approximation that is by no means innocuous for all questions one might wish to look at. By mapping the pair interaction on an effective one-particle potential, we discard all effects of coherent motion of the impurities. As a straightforward consequence, the relaxation amplitude of the susceptibility (60) vanishes in the limit of zero temperature.
(vii) Finally, we briefly mention minor approximations that are of little significance with respect to the basic features of our model. In order to obtain simple laws in terms of the concentrations we have truncated various powers series in Sect. VI, even if higher-order corrections are not always negligible.
## VIII Summary
We have studied interacting quantum impurities in terms of a mean-field model with a Gaussian distribution for the couplings. We briefly summarize our main results.
(i) The dipolar interaction leads to a reduction of the tunneling amplitude and to a wide distribution of the asymmetry energy. As a consequence, the density of states is smeared out to both to smaller and higher values as compared to the unperturbed tunnel energy $`\mathrm{\Delta }_0`$. Similarly, in the strong-coupling limit the relaxation spectrum covers several orders of magnitude. These features account for the collective nature of the underlying relaxation process; they are in qualitative agreement with experiments on lithium doped potassium chloride.
(ii) As a clear signature of interaction effects we consider how the zero-frequency susceptibility at low $`T`$ varies with the impurity concentration $`c`$. The present result, $`\chi \sqrt{c}`$, strongly deviates from the law $`\chi c^2`$ found previously in a different approach. Though more recent experiments on KCl:Li would seem to speak in favor of this latter result, the data presently available in the strong-doping regime $`\mu >1`$ do not unambiguously answer this question. Note, however, that the $`\sqrt{c}`$ scaling is clearly a consequence of the ‘naive’ mean-field assumption of identically distributed all-to-all interactions. It is likely to be modified in more realistic approaches. For instance, by introducing an ad-hoc $`c`$ scaling of our interaction parameters in an otherwise structurally unmodified mean-field approach, one would obtain a $`c^{1/3}`$ behavior of the $`T=0`$ susceptibility at large doping instead of the $`\sqrt{c}`$ behavior.
(iii) Since the above square root behavior results from a generic feature of the Gaussian mean-field model, an experimental test of this law would be interesting for a wider class of models. Mean-field models with random couplings similar to that considered in this paper are frequently studied in view of quantum spin glasses and other disorder quantum systems.
(iv) Our theory describes a crossover to glassy behavior in the (restricted) sense that broad and rather flat distributions of excitation energies and relaxation times develop at high doping. We have argued that true glassiness will develop in defect crystals only, when mechanisms are invoked which are beyond those originating from defect interactions. As a consequence, true glassy defect crystals are almost as difficult to grasp theoretically as truly amorphous systems.
###### Acknowledgements.
Very useful discussions with C. Enss, J. Classen, S. Hunklinger, S. Ludwig, P. Nalbach, R.O. Pohl, M. Thesen, and B. Thimmel are gratefully acknowledged. We dedicate this paper to Franz Wegner on the occasion of his 60-th birthday, thanking him for numerous discussions and inspiration throughout the years.
|
warning/0004/hep-th0004004.html
|
ar5iv
|
text
|
# Black Hole Evaporation and Large Extra Dimensions
## Abstract
We study the evaporation of black holes in space-times with extra dimensions of size $`L`$. We first obtain a potential which describes the expected behaviors of very large and very small black holes and then show that a (first order) phase transition, possibly signaled by an outburst of energy, occurs in the system when the horizon shrinks below $`L`$ from a larger value. This is related to both a change in the topology of the horizon and the restoring of translational symmetry along the extra dimensions.
thanks: Email: casadio@bo.infn.itthanks: Email: bharms@bama.ua.edu
Since Hawking’s semiclassical computation , one of the most elusive riddles of contemporary theoretical physics has been to understand black hole evaporation in a fully dynamical framework which accounts for the backreaction of the emitted particles on the space-time geometry. Intrinsically related to this (technically and conceptually) difficult issue is the role played by quantum gravity, since, by extrapolating from the semiclassical picture, one expects that black holes are capable of emitting particles up to any physical mass scale, including the Planck mass $`m_p=\mathrm{}_p^1`$. Therefore, black holes appear as the most natural window to look through for any theory involving quantum gravity.
In a series of papers the point of view of statistical mechanics was taken and the analysis based on the well known fact that the canonical ensemble cannot be consistently defined for a black hole and its Hawking radiation (conventionally viewed as point-like). Instead, the well posed microcanonical description led to the conclusion that black holes are (excitations of) extended objects ($`p`$-branes), a gas of which satisfies the bootstrap condition. This yielded the picture in which a black hole and the emitted particles are of the same nature and an improved law of black hole decay which is consistent with unitarity (energy conservation). The weakness of the statistical mechanical analysis is that it does not convey the geometry of the space-time where the evaporation takes place: although the luminosity of the black hole can be computed as a function of its mass, the backreaction on the metric remains an intractable problem .
This scenario has received support from investigations in fundamental string theory, where it is now accepted that extended D$`p`$-branes are a basic ingredient . States of such objects were constructed which represent black holes and corroborate the old idea that the area of the horizon is a measure of the quantum degeneracy of the black hole . However, this latter approach works mostly for very tiny black holes and suffers from the same shortcoming that the determination of the space-time geometry during the evaporation is missing.
It appears natural, in the framework of string theory, to consider the case when the black hole is embedded in a space-time of higher dimensionality. Indeed, the interest in models with extra spatial dimensions has been recently revived since they deliver a possible solution to the hierarchy problem without appealing to supersymmetry (see and references therein). One qualitatively views the four dimensional space-time as a D3-brane embedded in a bulk space-time of dimension $`4+d`$. Since matter is described by open strings with endpoints on the D$`p`$-branes, one expects that matter fields are confined to live on the D3-brane in the low energy limit (e.g., for energy smaller than the electroweak scale $`\mathrm{\Lambda }_{EW}`$), while gravity, being mediated by closed strings, can propagate also in the bulk. The $`d`$ extra spatial dimensions can be either compact or infinitely extended . Black holes in the former scenario were considered in and some light on the non-compact case was shed in (see also and References therein).
For compact extra dimensions of typical size $`L`$, states of the gravitational field living in the bulk have $`d`$ momentum components quantized in units of $`2\pi /L`$. For an observer living on the D3-brane, they then appear as particles with mass
$`m^{(\stackrel{}{n})}m_p{\displaystyle \frac{\mathrm{}_p}{L}}{\displaystyle \underset{i=1}{\overset{d}{}}}n_i,`$ (1)
where all $`n_i=0`$ for the (ground state) massless excitations corresponding to four dimensional gravitons, and states with some $`n_i>0`$ are known as Kaluza-Klein (KK) modes which induce short-range deviations from Newton’s law. The masses $`m^{(\stackrel{}{n})}`$ can be relatively small and the energy of matter confined on the D3-brane is prevented from leaking into the extra dimensions by the small coupling ($`m_p^1`$) between KK states of gravity and the matter energy-momentum tensor. This implies that processes involving particles within the standard model should result in energy loss into the extra dimensions and other phenomena potentially observable only above the TeV scale . However, this protection is not effective with Hawking radiation, since a black hole can evaporate into all existing particles whose masses are lower than its temperature, thus providing an independent way of testing the existence of extra dimensions.
The easiest way to estimate deviations from Newton’s law is to evaluate the potential generated by a point-like source of (bare) mass $`M`$ by means of Gauss’ law . Let us denote by $`r_b`$ the usual area coordinate on the four dimensional D3-brane. For distances $`r_bL`$ one then recovers the standard form
$`V_{(4)}=G_N{\displaystyle \frac{M}{r_b}},`$ (2)
where $`G_N=m_p^2`$ is Newton’s constant in four dimensions. For $`r_b<L`$ one has
$`V_{(4+d)}=G_{(4+d)}{\displaystyle \frac{M}{r_b^{1+d}}},`$ (3)
with $`G_{(4+d)}=M_{(4+d)}^{2d}=L^dG_N`$. This implies that the huge Planck mass $`m_p^2=M_{(4+d)}^{2+d}L^d`$ and, for sufficiently large $`L`$ and $`d`$, the bulk mass scale $`M_{(4+d)}`$ (eventually identified with the fundamental string scale) can be as small as $`1`$TeV. Since
$`L\left[1\mathrm{TeV}/M_{(4+d)}\right]^{1+\frac{2}{d}}\mathrm{\hspace{0.17em}10}^{\frac{31}{d}16}\mathrm{mm},`$ (4)
requiring that Newton’s law not be violated for distances larger than $`1`$mm restricts $`d2`$ .
A form for the potential which yields the expected behaviour at both small and large distance and ensures that a test particle on the D3-brane does not experience any discontinuity in the force when crossing $`r_bL`$ <sup>1</sup><sup>1</sup>1An important effect to be tested by the next generation of table-top experiments, see, e.g., . is given by
$`V=G_N{\displaystyle \frac{M}{r_b}}\left[1+{\displaystyle \underset{n=1}{\overset{d}{}}}C_n\left({\displaystyle \frac{L}{r_b}}\right)^n\right],`$ (5)
where $`C_n`$ are numerical coefficients (possibly functions of $`M`$, see below). The second (Yukawa) contribution in (5) can be related to the exchange of (massive) KK modes and, hence, encodes tidal effects from the bulk due to the presence of the mass $`M`$ on the D3-brane. We further note that the case $`d=2`$ can be used to make estimates for one non-compact extra dimension , provided one introduces an effective size $`L^2M_{(5)}^3/\mathrm{\Lambda }`$, with $`\mathrm{\Lambda }`$ the (negative) cosmological constant in the bulk AdS<sub>5</sub> .
The behavior of both very large ($`R_HL`$) and very small ($`R_HL`$) black holes is by now relatively well understood. In the former case, one can unwrap the compact extra dimensions and regard the real singularity as spread along a (black) $`d`$-brane of uniform density $`M/L^d`$, thus obtaining the Schwarzschild metric on the orthogonal D3-brane, in agreement with the weak field limit $`V_{(4)}`$, and an approximate “cylindrical” horizon topology $`S^2\times \text{I R}^d`$. In the latter case a solution is known , for one infinite extra dimension , which still has the form of a black string extending all the way through the bulk AdS<sub>5</sub>. However, this solution is unstable and believed to further collapse into one point-like singularity . This can be also argued from the observation that the Euclidean action of a black hole is proportional to its horizon area and is thus minimized by the spherical topology $`S^{2+d}`$. Hence, small black holes are expected to correspond to a generalization of the Schwarzschild metric to $`4+d`$ dimensions and should be colder and (possibly much) longer lived .
However, there is still a point to be clarified for small black holes, namely one should find an explicit matching between the spherical metric (for $`r_b<L`$) and the cylindrical metric (for $`r_b>L`$). This is not a trivial detail, since the ADM mass of a spherical $`4+d`$ dimensional black hole is zero as seen from the D3-brane because there is no $`1/r_b`$ term in the large $`r_b`$ expansion of the time-time component of the metric tensor . Thus, one concludes that the 4 dimensional ADM mass of a small black hole can be determined as a function of the $`4+d`$ dimensional mass parameter only after such a matching is provided explicitly. Even less is known about black holes of size $`R_HL`$, and a complete description is likely to be achieved only by solving the entire set of field equations for an evaporating black hole in $`4+d`$ dimensions.
Instead of tackling this intractable backreaction problem, we extrapolate from the weak field limit on the D3-brane given in (5) the (time and radial components of the) metric in $`4+d`$ dimensions as
$`g_{tt}`$ $``$ $`12V(r)`$
$`=`$ $`12G_N{\displaystyle \frac{M}{r}}\left[1+{\displaystyle \underset{n=1}{\overset{d}{}}}C_n\left({\displaystyle \frac{L}{r}}\right)^n\right]`$
$`g_{rr}`$ $``$ $`g_{tt}^1,`$
where $`r`$ now stands for the area coordinate in $`4+d`$ dimensions. This yields $`M`$ as the ADM mass of the black hole (see for a similar solution in the scenario of ) and the radius of the horizon is determined by
$`g_{tt}=0R_H=2G_NM\left[1+{\displaystyle \underset{n=1}{\overset{d}{}}}C_n\left({\displaystyle \frac{L}{R_H}}\right)^n\right].`$ (7)
The above ansatz does not provide an exact solution of vacuum Einstein equations, since some of the components of the corresponding Einstein tensor in $`4+d`$ dimensions $`G_{ij}=8\pi G_{(4+d)}T_{ij}0`$. However, it is possible to choose the coefficients $`C_n`$ (as functions of $`M`$) in such a way that the “effective matter contribution” $`T_{ij}`$ from the region outside the black hole horizon is small. In particular, one can require that the contribution to the ADM mass be negligible,
$`m{\displaystyle _{R_H}^{\mathrm{}}}d^{4+d}xT_t^t={\displaystyle \frac{1}{8\pi G_{(4+d)}}}{\displaystyle _{R_H}^{\mathrm{}}}d^{4+d}xG_t^t(\{C_n\})M.`$ (8)
In this sense one can render the above metric a good approximation to a true black hole in $`4+d`$ dimensions. One should also recall that the black hole must be a classical object, to wit its Compton wavelength $`\mathrm{}_M\mathrm{}_p(m_p/M)R_H`$. Further, once Hawking radiation is included, its backreaction on the metric at small $`r`$ is likely to be significant for $`R_HL`$, so that a true vacuum solution would not be practically much more useful.
When $`R_HL`$ one has $`rr_b`$, therefore, the metric (LABEL:met) is approximately cylindrically symmetric (along the extra dimensions). Further, on setting all $`C_n=0`$ yields $`m=0`$, and Eq. (7) coincides with the usual four dimensional Schwarzschild radius $`R_H2\mathrm{}_p(M/m_p)`$. Correspondingly one has the inverse Hawking temperature ($`\beta _H=T_H^1`$) and Euclidean action
$`\beta _H^>`$ $``$ $`8\pi \mathrm{}_p(M/m_p)`$ (9)
$`S_E^>`$ $``$ $`4\pi (M/m_p)^2A_{(4)}/4\mathrm{}_p^2,`$ (10)
where $`A_{(D)}`$ is the area of the horizon in $`D`$ space-time dimensions and the condition $`R_HL`$ translates into
$`Mm_p(L/\mathrm{}_p)M_c,`$ (11)
(e.g., $`M_c10^{27}`$g for $`L1`$mm). The fact that the extra dimensions do not play any significant role at this stage is further confirmed by $`T_H^>2\pi /Lm^{(1)}`$ (the mass of the lightest KK mode), therefore no KK particles can be produced.
For $`R_HL`$, one can again take advantage of the coefficients $`C_n`$ to lower $`m`$ as much as possible. For instance, for $`d=2`$ and $`M=10^{15}`$g ($`M_c`$) the values $`C_2=C_1=1`$ yield $`m10^3M`$ (more details will be given in ). Eq. (7) then leads to
$`R_H\left(2C_dL^dG_NM\right)^{\frac{1}{1+d}},`$ (12)
and the consistency conditions $`\mathrm{}_MR_HL`$ hold for
$`m_p(\mathrm{}_p/L)^{\frac{d}{2+d}}MM_c.`$ (13)
Since we have assumed that the spherical symmetry extends to $`4+d`$ dimensions, one obtains
$`\beta _H^<L(M/M_c)^{\frac{1}{1+d}}`$ (14)
$`S_E^<\left(L/\mathrm{}_p\right)^2\left(M/M_c\right)^{\frac{2+d}{1+d}}A_{(4+d)}/\mathrm{}_p^2L^d,`$ (15)
which reduce back to (9) and (10) if one pushes down $`L\mathrm{}_p`$ ($`M_cm_p`$). For $`MM_c`$, the temperature $`T_H^<`$ is sufficient to excite KK modes, although it is lower than that of a four dimensional black hole of equal mass. Correspondingly, the Euclidean action $`S_E^<(M)S_E^>(M)`$, yielding a smaller probability
$`P\mathrm{exp}\left(S_E\right).`$ (16)
for the Hawking particles “to come into existence” in the $`4+d`$ dimensional scenario.
The luminosity of a black hole (provided microcanonical corrections are negligible ) can be approximated by employing the canonical expression
$`F_{(D)}A_{(D)}{\displaystyle \underset{s}{}}{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{\mathrm{\Gamma }_sd\omega ^D}{e^{\beta _H\omega }1}}=A_{(D)}N_{(D)}T_H^D,`$ (17)
with $`\mathrm{\Gamma }`$ the grey-body factor and $`N`$ a coefficient which depends upon the number of available particle species $`s`$ with mass smaller than $`T_H`$. For small black holes $`D=4+d`$ and $`T_H=T_H^<`$, hence $`F_{(4+d)}(M/M_c)^{\frac{2}{1+d}}`$ is far less intense than $`F_{(4)}(M/M_c)^2`$ obtained from $`T_H^>`$ . It also follows from (17) that $`F_{(D)}`$ as a function of $`R_H`$ (not $`M`$) depends on $`D`$ only through $`N_{(D)}`$. Therefore, the energy emitted into KK modes must be a small fraction of the total luminosity,
$`{\displaystyle \frac{F_{KK}}{F_{(4+d)}}}{\displaystyle \frac{N_{KK}}{N_{(4)}+N_{KK}}}1,`$ (18)
because the number of degrees of freedom of KK gravitons ($`N_{KK}`$) is much smaller than the number of particles in the standard model ($`N_{(4)}`$), and energy conservation in $`4+d`$ dimensions requires that $`F_{(4+d)}=F_{(4)}+F_{KK}`$ (see also for similar arguments). However, one might question the efficiency of the confining mechanism for matter on the D3-brane at very high energy<sup>2</sup><sup>2</sup>2There is indeed evidence that only the zero modes of standard model fields can be confined, thus allowing matter to leak in the bulk (see, e.g., ). (say greater than $`\mathrm{\Lambda }_{EW}`$). One should then include bulk standard model fields among KK modes and consider $`N_{KK}`$ as a growing function of the temperature. If this is the case, the luminosity will greatly increase and the ratio $`F_{KK}/F_{(4+d)}`$ will eventually approach unity when $`T_H^<>\mathrm{\Lambda }_{EW}`$.
An inspection of the Klein-Gordon equation in $`4+d`$ dimensions supports the conclusion that the luminosity of small black holes is relatively fainter. In fact, there is a potential barrier in the equation governing the radial propagation outside the horizon of the form
$`W{\displaystyle \frac{d}{2}}\left({\displaystyle \frac{d}{2}}+1\right){\displaystyle \frac{1}{r^2}}\left[1\left({\displaystyle \frac{R_H}{r}}\right)^{1+d}\right]^2,`$ (19)
which all Hawking particles have to tunnel through in order to escape. One can roughly reproduce $`W`$ by assuming that all particles are emitted with an effective angular momentum $`l+d/3`$, which substantially lowers the grey-body factor .
Let us now consider the case when $`R_HL`$ (i.e., $`MM_c`$) and try to understand what happens as the horizon shrinks below the size of the extra dimensions. It is easy to solve (7) for the “toy model” $`d=1`$ and estimate the relation between the ADM mass and the radius of the horizon at all scales
$`R_H=G_NM\left(1+\sqrt{1+(2L/G_NM)}\right).`$ (20)
This leaves only the topology of the horizon to be specified for $`2R_HL`$. According to the area law $`S_EA_{(4+d)}/16\pi G_{(4+d)}`$, one has for the “cylinder” $`S^2\times \text{I R}`$ and for the three sphere $`S^3`$ respectively
$`S_E^cR_H^2/8\mathrm{}_p^2,S_E^sR_H^3/16\mathrm{}_p^2L,`$ (21)
and the ratio $`S_E^c/S_E^s4`$ for $`2R_H=L`$, so that the spherical topology is favored once the horizon has become small enough to “close up” along the extra dimension. The inverse temperature in the two cases is given by
$`\beta _H^c{\displaystyle \frac{R_H}{4\mathrm{}_p^2}}{\displaystyle \frac{R_H}{M}},\beta _H^s{\displaystyle \frac{3R_H^2}{16\mathrm{}_p^2L}}{\displaystyle \frac{R_H}{M}},`$ (22)
and is discontinuous as well, since $`\beta _H^c/\beta _H^s2/3`$ for $`2R_H=L`$. The physically interesting case $`d=2`$ is algebraically more involved but the qualitative picture is the same as for $`d=1`$ if the topology of the extra dimensions is $`\text{I R}^2`$ .
We can now get some further physical insight by appealing to statistical mechanics. Although the use of the canonical ensemble is known to be incorrect, we utilize it in the analysis below in order to conform to the standard description and introduce a partition function for the black hole and its Hawking radiation as the Laplace transform of the microcanonical density of states ,
$`Z_\beta {\displaystyle 𝑑Me^{\beta M}e^{S_E(M)}},`$ (23)
where we have estimated the internal degeneracy of a black hole as the inverse of the probability (16), in agreement with the area law and the bootstrap relation . Since the entropy $`S=\beta ^2(F/\beta )`$, where $`\beta F=\mathrm{ln}Z`$ is the (Helmholtz) free energy, it follows from (21) that there is a discontinuity in the first derivative of $`F`$ at
$`\beta _c\beta _H^s(M_c)L1/m^{(1)}.`$ (24)
This behavior is characteristic of a first order phase transition : In the cold phase, $`T_H<\beta _c^1`$, the system appears “condensed” into the low energy standard model fields living on the four dimensional D3-brane and translational invariance in the $`d`$ extra directions is therefore broken by the D3-brane itself. For $`T_H>\beta _c^1`$ translational invariance begins to restore, and the system starts spreading over all bulk space, with D3-brane vibrations acting as Nambu-Goldstone bosons which give mass to the KK modes<sup>3</sup><sup>3</sup>3One might argue that it is the cylindrical topology which is translationally symmetric along the extra dimensions, however for $`T_H<T_c`$ there are no physical particles in the bulk, and such a symmetry remains a purely geometrical concept (as well as the bulk itself) with no dynamical counterpart. .
Of course, in the statistical mechanical analysis the time is missing, and one can more realistically think that, during the evaporation of an initially large black hole, the horizon starts to bend in the extra dimensions when $`T_H^c`$ approaches $`T_H^s(M_c)`$ (from below). If the temperature remains constant (and approximately equal to the lower value $`T_H^s(M_c)T_c`$) during the transition, then a qualitative picture of the effect is given in Fig. 1. The change in the topology of the horizon could then be accompanied by a burst of energy to get rid quickly of the excess mass (equal to the width of the plateau in Fig. 1).
In the above approximation the specific heats in the cold and hot phases are respectively given by
$`C_V^>\left(M/m_p\right)^2,C_V^<LM\left(M/M_c\right)^{\frac{1}{1+d}},`$ (25)
Since $`C_V^<`$ is negative, one is eventually forced to employ the microcanonical description when the temperature becomes too hot<sup>4</sup><sup>4</sup>4In order to make sense of the partition function, one should introduce an ultra-violet cut off $`\mathrm{\Lambda }_{UV}`$ to render the (divergent) integral (23) finite. Then, the canonical description is a good approximation only for $`T\mathrm{\Lambda }_{UV}`$ , although the fact that $`|C_V^<|<|C_V^>|`$ suggests that the extra dimensions make the inconsistency of the canonical description milder .
This work was supported in part by the U.S. Department of Energy under Grant no. DE-FG02-96ER40967 and by NATO grant no. CRG.973052.
|
warning/0004/cond-mat0004069.html
|
ar5iv
|
text
|
# Finite-temperature resistive transition in the two-dimensional 𝑋𝑌 gauge glass model
## Abstract
We investigate numerically the resistive transition in the two-dimensional $`XY`$ gauge glass model. The resistively-shunted junction dynamics subject to the fluctuating twist boundary condition is used and the linear resistances in the absence of an external current at various system sizes are computed. Through the use of the standard finite-size scaling method, the finite temperature resistive transition is found at $`k_BT_c=0.22(2)`$ (in units of the Josephson coupling strength) with dynamic critical exponent $`z=2.0(1)`$ and the static exponent $`\nu =1.2(2)`$, in contrast to widely believed expectation of the zero-temperature transition. Comparisons with existing experiments and simulations are also made.
Since the prediction of the vortex glass (VG) phase in the high-$`T_c`$ materials, the properties and the existence of VG phase have drawn intensive interest for a decade. In experiments the VG phase has a significant practical importance since the high-$`T_c`$ materials can be truly superconducting in this phase where vortices are completely frozen at random positions, while the Abrikosov vortex lattice in pure type-II superconductors dissipate energy at any amount of external currents. To study VG from a theoretical point of view simplified discretized models such as the $`XY`$ gauge glass and the $`XY`$ spin glass have been widely used. For the gauge glass model, there is a growing consensus that the low critical dimension is less than 3, which means that there exists VG phase at finite temperatures in three dimensions (3D). This has been confirmed by the defect wall energy calculations at zero temperature as well as by finite-temperature simulations. Although the issue about the validity of using the gauge glass model to describe real bulk high-$`T_c`$ materials is not completely settled (e.g., the former is isotropic with a vanishing net external magnetic field, while the latter is anisotropic with nonzero net magnetic field), experiments on high-$`T_c`$ materials like YBCO and BiSCCO also have yielded results in accordance with the existence of a VG phase at finite temperatures.
In 2D, all existing defect wall energy calculations have unanimously revealed that the stiffness exponent has a negative value, which has been confirmed in Ref. through a simple argument but has been interpreted as an artifact of ubiquitous spin wave fluctuations. Finite-temperature simulations have so far yielded contradicting results: On the one hand, zero-temperature transition has been concluded from Monte-Carlo (MC) simulations in earlier studies and from the current-voltage ($`IV`$) characteristics in resistively-shunted junction (RSJ) simulations. On the other hand, there have been evidences of the finite-temperature transition from the $`IV`$ characteristics and from the divergence of the relaxation time scale in RSJ dynamic simulations, and from the very recent extensive finite-temperature MC simulations. Also, Langevin dynamic simulation of a vortex model with quenched impurities, which is closely related with the gauge glass model, also has found the finite-temperature transition in 2D. To make things more complicated, absence of finite-temperature VG phase has been reported in early experiments on very thin YBCO films, whereas a recent study on thin YBCO films, where the correlation length in $`c`$-direction exceeds the film thickness, as well as experiments on highly isotropic BiSCCO \[Bi(2:2:2:3)\] has obtained results which can be interpreted as indications of finite-$`T`$ VG phase in 2D. Consequently, we believe that the question about the existence of the finite-$`T`$ VG transition is not resolved yet. To our knowledge, a detailed numerical study of transport properties of 2D gauge glass model has not been performed: Existing RSJ studies did not consider the finite-size scaling in a proper way and in some of them the temperature range used did not cover the expected transition temperature ($`T_c0.22`$ in Refs. and , and $`T_c0.15`$ in Ref. ).
In this paper, we use RSJ dynamic simulation to investigate the resistive transition in the 2D $`XY`$ gauge glass model in the absence of external currents. Since the voltage is always nonzero in the presence of a finite current because of the nucleation process, vanishing of the linear resistance, which is defined in the limit of zero external current, is the appropriate definition of a superconductor. The fluctuating twist boundary condition (FTBC) for Langevin-type dynamic simulation made it possible to calculate the linear resistance with preserved periodicity of the phase variables (similar methods have also been used in Ref. ). We here use FTBC to calculate the linear resistances, and then, through the standard finite-size scaling analysis (see, e.g., Refs. and ), a resistive transition at a finite temperature is concluded.
The Hamiltonian of the 2D $`L\times L`$ $`XY`$ gauge glass model subject to FTBC is written as
$$H=J\underset{ij}{}\mathrm{cos}(\theta _i\theta _j𝐫_{ij}𝚫A_{ij}),$$
(1)
where $`J`$ is the Josephson coupling strength, the summation is over all nearest neighboring bonds, and $`\theta _i`$ (with periodicity $`\theta _i=\theta _{i+L\widehat{𝐱}}=\theta _{i+L\widehat{𝐲}}`$) is the phase of the superconducting order parameter at site $`i`$. The twist variable $`𝚫=(\mathrm{\Delta }_x,\mathrm{\Delta }_y)`$ measures the global twist of phase variables in each direction, and $`𝐫_{ij}(=\widehat{𝐱},\widehat{𝐲})`$ is the unit vector from site $`i`$ to the nearest-neighbor site $`j`$ (the lattice spacing is taken to be unity). The $`XY`$ gauge glass model is characterized by the quenched random variable $`A_{ij}`$ uniformly distributed in $`[\pi ,\pi )`$, whereas in the $`XY`$ spin glass model $`A_{ij}`$ can have binary values $`0`$ and $`\pi `$ with equal probability. Recently $`XY`$ spin glass model draws attention in relation to the $`\pi `$-junctions due to the $`d`$-wave symmetry of order parameters.
The equations of motion for RSJ dynamics subject to FTBC are determined from the local and the global current conservations and are written in dimensionless forms (see Refs. and for details): For phase variables
$$\dot{\theta }_i=\underset{j}{}G_{ij}\underset{k}{}^{^{}}[\mathrm{sin}(\theta _j\theta _k𝐫_{jk}𝚫A_{jk})+\eta _{jk}],$$
(2)
where $`G_{ij}`$ is the 2D square lattice Green’s function and the primed summation is over four nearest neighbors $`k`$ of site $`j`$, while for the twist variables
$$\dot{𝚫}=\frac{1}{L^2}\frac{H}{𝚫}+\eta _𝚫.$$
(3)
The thermal noise terms satisfy $`\eta _{ij}=\eta _𝚫=\eta _{ij}\eta _𝚫=\eta _{\mathrm{\Delta }_x}\eta _{\mathrm{\Delta }_y}=0`$ and
$`\eta _{ij}(t)\eta _{kl}(0)=2T\delta (t)(\delta _{ik}\delta _{jl}\delta _{il}\delta _{jk}),`$
$`\eta _{\mathrm{\Delta }_x}(t)\eta _{\mathrm{\Delta }_x}(0)=\eta _{\mathrm{\Delta }_y}(t)\eta _{\mathrm{\Delta }_y}(0)={\displaystyle \frac{2T}{L^2}}\delta (t),`$
where $`\mathrm{}`$ is the thermal average, and the time $`t`$ and the temperature $`T`$ have been normalized in units of $`\mathrm{}^2/4e^2R_0J`$ with the shunt resistance $`R_0`$, and $`J/k_B`$, respectively. The equations of motion were integrated numerically by using the efficient second-order Runge-Kutta-Helfand-Greenside algorithm with the discrete time step $`\mathrm{\Delta }t=0.05`$. For each disorder realization we neglected an initial stage of the time evolution up to $`t_0`$, and the measurements were made between $`t_0`$ and $`t_0+t_m`$. For $`L=4`$, 6, 8, and 10, disorder averages were performed over $`N_s=400`$, 200, 100, 50 samples, respectively. We verified that $`t_0=10^6`$ and $`t_m=5\times 10^6`$ (corresponding to $`10^8`$ time steps) were sufficiently large; it is more important to increase the number of sample averages to have more reliable results than increasing $`t_m`$ further. The highest temperature $`T=0.30`$ in the present work corresponds to the lowest temperature in Ref. , and this makes it necessary to use much larger value of $`t_m`$ than in Ref. \[by factor $`O(10^2)`$\].
The linear resistance in $`x`$ direction is given by the Nyquist formula:
$$R_x=\frac{1}{2T}\left[_{\mathrm{}}^{\mathrm{}}𝑑tV_x(t)V_x(0)\right]$$
(4)
with the disorder average $`[\mathrm{}]`$ and the voltage across the whole sample $`V_x=L\mathrm{\Delta }_x`$, which is then approximated as
$$R_x\frac{L^2}{2T}\frac{1}{\mathrm{\Theta }}\left[[\mathrm{\Delta }_x(\mathrm{\Theta })\mathrm{\Delta }_x(0)]^2\right],$$
(5)
for sufficiently large $`\mathrm{\Theta }`$. For convenience, we define
$$f(\mathrm{\Theta })\frac{L^2}{2T}\frac{\left[[\mathrm{\Delta }_x(\mathrm{\Theta })\mathrm{\Delta }_x(0)]^2\right]+\left[[\mathrm{\Delta }_y(\mathrm{\Theta })\mathrm{\Delta }_y(0)]^2\right]}{2}$$
(6)
and calculate the linear resistance $`R`$ from the least-square fit to the form $`f(\mathrm{\Theta })=R\mathrm{\Theta }`$.
We plot in Fig. 1 for $`L=6`$ at $`T=0.22`$ (a) the time evolution of $`L\mathrm{\Delta }_y`$ for a given disorder realization, and (b) $`f(\mathrm{\Theta })`$ obtained from the average over 200 samples. The full line in Fig. 1(b) obtained from the fit determines the value of $`R`$. Figure 2 shows $`R`$ obtained in this way as a function of $`L`$ at $`T=0.30`$, 0.25, 0.23, 0.22, 0.21, 0.20, and 0.15. For each $`L`$ and $`T`$, we divided the total number of samples, $`N_s`$, into five groups, each of which contains $`N_s/5`$ samples, and $`R`$ was calculated for each group, leading to the estimation of error bars (standard deviations) in Fig. 2. The linear resistance shows clear change of curvature around $`T0.22`$: In the high-temperature regime $`R`$ appears to saturate as $`L`$ is increased, while in the low-temperature regime $`R`$ drops down more rapidly as $`L`$ is increased. To see this behavior more clearly, we also display in the inset of Fig. 2 $`R(T=0.15)/R(T=0.30)`$ as a function of $`L`$ in log scales, which does not exhibit any sign of saturation in terms of $`L`$: On the contrary, the downward curvature implies that in the thermodynamic limit $`R(T=0.15)`$ vanishes. These observations lead to the conclusion that the 2D $`XY`$ gauge glass model has the finite-temperature resistive transition.
To obtain more precise value of $`T_c`$, we proceed to the finite-size scaling analysis of the linear resistance. In the dynamic scaling theory for the usual second-order phase transition in 2D, the linear resistance has the standard finite-size scaling form:
$$R(L,T)=L^z\rho \left((TT_c)L^{1/\nu }\right),$$
(7)
where $`\rho (x)`$ is the scaling function with the scaling variable $`x`$, $`z`$ is the dynamic critical exponent, and the static critical exponent $`\nu `$ is defined by the divergence of the correlation length, i.e., $`\xi |TT_c|^\nu `$. At $`T=T_c`$, $`RL^z`$ \[see Eq. (7)\] and from Fig. 2 we obtain $`z2.0`$. Although this value is much smaller than the values usually measured in experiments ($`z=49`$), it should be noted that the similar values have also been observed in various numerical simulations with RSJ dynamics and MC dynamics of 2D gauge glass model ($`z2.2`$ in Ref. and $`z2.4`$ in Ref. , respectively) as well as the Langevin-type relaxational dynamics of vortices in the presence of quenched impurities ($`z2.1`$, Ref. ). Furthermore, a recent analytic calculation based on the dynamic renormalization group method also has found $`z=2`$ for 2D gauge glass model with a purely relaxational equations of motion. In contrast, widely believed zero-temperature transition in 2D gauge glass model implies that the linear resistance should have Arrhenius form and thus corresponds to $`z=\mathrm{}`$. We also tried to fit our data in Fig. 2 to the Arrhenius form (see Fig. 3) and found that the thermal activation barrier, which is proportional to the slopes in Fig. 3, strongly depends on the system size. Furthermore, $`R`$ is found to deviate from the Arrhenius form in a systematic way at the lowest temperature at all system sizes.
Figure 4 shows the scaling plot Eq. (7) with $`z=2.0`$, $`T_c=0.22`$, and $`\nu =1.2`$. We tried to vary parameter values used in the scaling plot and concluded
$`T_c`$ $`=`$ $`0.22(2),`$ (8)
$`z`$ $`=`$ $`2.0(1),`$ (9)
$`\nu `$ $`=`$ $`1.2(2),`$ (10)
where numbers in parenthesis denote errors in the last digits, and $`T_c0.22`$ is in a good agreement with Refs. and . It is worth mentioning that the same method (calculation of the linear resistance with FTBC accompanied by the finite-size scaling analysis) has lead to a very precise determination of $`T_c`$ in 3D $`XY`$ model. In recent studies of 2D $`XY`$ gauge glass model based on the same RSJ dynamics, the finite-size scaling analysis has not been used and the temperature range did not cover the critical temperature ($`T_c0.22`$), leading to the different conclusion of the zero-temperature transition.
Recently, the possibility of the quasi-long-range glass order with the vanishing glass order parameter in 2D gauge glass model has been suggested in Ref. , where it has been argued that the correlation length $`\xi `$ diverges in the whole low-temperature phase. Although our scaling plot in Fig. 4 was obtained on the assumption of the usual second order transition, where $`\xi `$ is finite both below and above $`T_c`$, it is still plausible to have the quasi-long-range glass order: In this case, the quasi-criticality should be reflected in the temperature-dependent dynamic critical exponent below $`T_c`$ with $`R(L,T)L^{z(T)}`$, while the high-temperature phase still obeys the scaling form in Eq. (7). Our current results cannot rule out this possibility, and if we neglect the smallest size $`L=4`$ in Fig. 2, $`R`$ in the low-temperature phase indeed appears to exhibit the simple algebraic form $`RL^z`$ with increasing $`z`$ as $`T`$ is decreased, in accord with the idea of the quasi-long-range glass order.
In conclusion, we studied numerically the resistive transition in the 2D $`XY`$ gauge glass model by using the RSJ dynamic equations subject to the fluctuating twist boundary condition. The standard finite-size scaling analysis applied to the linear resistances lead to the strong evidence of the finite-temperature resistive transition at $`T_c=0.22(2)`$ with the dynamic critical exponent $`z=2.0(1)`$ and the static exponent $`\nu =1.2(2)`$. However, the nature of this finite-temperature resistive transition needs to be investigated more in detail.
###### Acknowledgements.
The author is grateful to Prof. Petter Minnhagen for useful discussions. This work was supported by the Swedish Natural Research Council through contract FU 04040-332.
|
warning/0004/hep-th0004056.html
|
ar5iv
|
text
|
# Living on the edge: cosmology on the boundary of anti-de Sitter space
## 1 Introduction
In this article we develop what we feel is a particularly simple and compelling cosmological model based on the semi-phenomenological Randall–Sundrum models for low-energy string theory . For some early tentative steps along these lines see the papers of Gogberashvili , plus more recent developments in and .<sup>1</sup><sup>1</sup>1 Note that many aspects of this recent work can be viewed as extending domain-wall physics in (3+1) dimensions to brane physics in (4+1) dimensions, and so owes much to early papers on domain-wall physics . In developing our cosmology, we wish to minimize the number of baroque features coming from the underlying string theory, and maximize the use of symmetry principles, in order to develop a picture that is as simple and attractive as possible, with good prospects for being observationally testable.
Perhaps the most compelling model along these lines can be built by considering a (4+1)-dimensional manifold with a single (3+1)-dimensional boundary. This boundary is taken to be a D-brane (a membrane on which the fundamental string fields satisfy Dirichlet type boundary conditions), and the D-brane is assigned an intrinsic energy density and pressure arising both from some underlying brane tension and from matter \[ordinary (3+1)-dimensional matter\] that is trapped on the D-brane by stringy effects.
Since this point has the capacity to cause serious confusion, let us try to make it a little more explicit:<sup>2</sup><sup>2</sup>2 We are trying to make this article comprehensible to string theorists, relativists, and astrophysicists. Accordingly some comments may be trivial to one of the three communities, but we would rather err on the side of clarity and simplicity than either impenetrable brevity or excessive technical detail. We are viewing ordinary matter as open-string excitations of the D-brane boundary. But since open strings by definition have their end-points on the D-brane, an open string of energy $`E`$ is strictly limited in how far it can stretch off the D-brane: Its maximum extension into any higher-dimensional bulk is simply $`L_{\mathrm{stringy}}<E/(2\alpha ^{})`$ where $`\alpha ^{}`$ is the fundamental open string tension.<sup>3</sup><sup>3</sup>3 This whole D-brane picture only makes sense for string excitations of low energy compared to the string scale: $`E<\sqrt{\mathrm{}c\alpha ^{}}`$. So the thickness of the cloud of excitations surrounding the D-brane is at most of order $`L_{\mathrm{stringy}}<\sqrt{\mathrm{}c/(2\alpha ^{})}`$. In contrast, gravitons are represented by closed string loops which are not trapped on the D-brane — gravitons (and non-perturbative gravity) can very easily penetrate finite distances into the higher-dimension bulk. Thus gravity is in our model fundamentally a (4+1)-dimensional effect and we will be using the (4+1)-dimensional Einstein equations to deduce the analog of the Friedmann equations of motion for the (3+1)-dimensional D-brane boundary.
Now while gravitons can easily penetrate into the bulk, one does not want them to be too effective at doing so. Once one turns away from the large-scale average properties of the cosmological FLRW geometry, to consider the gravitational field generated by astrophysical perturbations (planets, stars, galaxies) one does not want the virtual-graviton cloud surrounding these objects to be completely free to move into the (4+1)-dimensional bulk, since then one would see an inverse-cube law for gravity in lieu of the observed inverse-square law. This is where the Randall–Sundrum mechanism is critical — virtual gravitons generated by matter perturbations are (weakly) trapped near the D-brane, not by stringy effects, but rather by the bulk gravitational field and the tightly constrained location of the sources.<sup>4</sup><sup>4</sup>4 The distance scale on which gravitons are trapped is generically set by the Riemann curvature of the higher-dimensional bulk; in the Randall–Sundrum models the relevant parameter is $`L_{\mathrm{graviton}}=\sqrt{6/|\mathrm{\Lambda }_{4+1}|}`$, defined by the cosmological constant in the higher-dimensional bulk. We belabor this point because we have seen it generate considerable confusion within the relativity and astrophysics communities: Gravity is not used to trap matter on the D-brane and the Randall–Sundrum models have more in common with the field-theory-based trapping mechanisms of Akama and Shaposhnikov than they do with the gravity-based trapping mechanism of .
In the interests of simplicity and clarity the (4+1)-dimensional bulk will always be taken to be static and hyper-spherically symmetric, though we shall quickly specialize to Reissner–Nordström–de Sitter space, or even more particularly, to anti-de Sitter space. The boundary will always be taken to be hyper-spherically symmetric in the (4+1)-dimensional sense, with this hyper-spherical symmetry reducing to translation invariance when viewed from the (3+1)-dimensional point of view. In picking this particular starting point we have been guided by many recent publications; including the Randall–Sundrum scenarios (which will used to describe the physics near the brane), the single-brane models of Gogberashvili , various previous versions of Randall–Sundrum based cosmology , and by a desire to have a framework that is at least plausibly connectable to the complex of ideas going under the name of the adS/CFT correspondence .
We start the analysis by a discussion of what it means to apply the Einstein equations to a manifold with boundary, interpreting this process in terms of an extension of the Israel–Lanczos–Sen thin shell formalism . This permits us to write down an analog of the usual Friedmann equation of FLRW cosmology, and in the next section we discuss how to make this cosmologically viable. Going beyond the FLRW cosmological fluid approximation we verify that the essential portion of the Randall–Sundrum model (having to do with the weak trapping of perturbatively generated gravitons near the brane) continues to work in the present context. Finally we indicate some possible variants on the present model and describe areas where the present ideas may lead to observational tests.
## 2 D-brane surgery
### 2.1 Extrinsic and intrinsic geometries:
We start by considering a rather general static hyper-spherically symmetric geometry in (4+1) dimensions. (This is not the most general such metric, but quite sufficient for our purposes.)<sup>5</sup><sup>5</sup>5 Note that the technical computations closely parallel those for spherically symmetric (2+1)-dimensional domain walls symmetrically embedded in a spherically symmetric (3+1)-dimensional spacetime. See, for instance, references .
$$\mathrm{d}s_{4+1}^2=F(r)\mathrm{d}t^2+\frac{\mathrm{d}r^2}{F(r)}+r^2\mathrm{d}\mathrm{\Omega }_3^2.$$
(1)
$$\mathrm{d}\mathrm{\Omega }_3^2\mathrm{d}\chi ^2+\mathrm{sin}^2\chi (\mathrm{d}\theta ^2+\mathrm{sin}^2\theta \mathrm{d}\varphi ^2).$$
(2)
To build the class of (3+1)-dimensional geometries we are interested in, we start by simply truncating the (4+1)-dimensional geometry at some time-dependent radius $`a(t)`$, keeping only the interior portion and discarding the exterior. Kinematically, the surface of this truncated geometry (which we take to be the location of the D-brane) is automatically a (3+1)-dimensional closed ($`k=+1`$, positive spatial curvature) FLRW geometry with induced metric
$$\mathrm{d}s_{3+1}^2=\left[F(a(t))\frac{1}{F(a(t))}\left(\frac{\mathrm{d}a}{\mathrm{d}t}\right)^2\right]\mathrm{d}t^2+a(t)^2\mathrm{d}\mathrm{\Omega }_3^2.$$
(3)
Now consider radial motion of the D-brane; this is radial motion in the embedding (4+1)-dimensional hyperspace. We start the analysis by first parameterizing the motion in terms of proper time along a curve of fixed $`\chi `$, $`\theta `$, and $`\varphi `$ (these are comoving coordinates in the FLRW cosmology). That is: the D-brane sweeps out a world-volume
$$X^\mu (\tau ,\chi ,\theta ,\varphi )=(t(\tau ),a(\tau ),\chi ,\theta ,\varphi ).$$
(4)
The 5-velocity of the $`(\chi ,\theta ,\varphi )`$ element of the D-brane can then be defined as
$$V^\mu =(\frac{\mathrm{d}t}{\mathrm{d}\tau },\frac{\mathrm{d}a}{\mathrm{d}\tau },0,0,0).$$
(5)
Using the normalization condition and the assumed form of the metric, and defining $`\dot{a}=\mathrm{d}a/\mathrm{d}\tau `$,
$$V^\mu =(\frac{\sqrt{F(a)+\dot{a}^2}}{F(a)},\dot{a},0,0,0);V_\mu =(\sqrt{F(a)+\dot{a}^2},\frac{\dot{a}}{F(a)},0,0,0).$$
(6)
The unit normal vector to the hypersphere $`a(\tau )`$ is
$$n^\mu =(\frac{\dot{a}}{F(a)},\sqrt{F(a)+\dot{a}^2},0,0,0);n_\mu =(+\dot{a},\frac{\sqrt{F(a)+\dot{a}^2}}{F(a)},0,0,0).$$
(7)
\[We shall take the unit normal to be inward pointing, into the bulk of the (5+1) geometry.\] The extrinsic curvature can be written in terms of the normal derivative<sup>6</sup><sup>6</sup>6 Unfortunately sign conventions differ on this point. We follow .
$$K_{\mu \nu }=\frac{1}{2}\frac{g_{\mu \nu }}{\eta }=\frac{1}{2}n^\sigma \frac{g_{\mu \nu }}{x^\sigma }.$$
(8)
If we go to an orthonormal basis, the $`\widehat{\chi }\widehat{\chi }`$ component is easily evaluated
$$K_{\widehat{\chi }\widehat{\chi }}=K_{\widehat{\theta }\widehat{\theta }}=K_{\widehat{\varphi }\widehat{\varphi }}=\frac{1}{2}\sqrt{F(a)+\dot{a}^2}\frac{g_{\chi \chi }}{r}g^{\chi \chi }=\frac{\sqrt{F(a)+\dot{a}^2}}{a}$$
(9)
The $`\tau \tau `$ component is a little messier, but generalizing the calculation of (which amounts to calculating the five-acceleration of the brane, this is explained in more detail in ) quickly leads to
$$K_{\widehat{\tau }\widehat{\tau }}=+\frac{1}{2}\frac{1}{\sqrt{F(a)+\dot{a}^2}}\left(\frac{\mathrm{d}F(r)}{\mathrm{d}a}+2\ddot{a}\right)=+\frac{\mathrm{d}}{\mathrm{d}a}\left(\sqrt{F(a)+\dot{a}^2}\right).$$
(10)
In contrast to the extrinsic geometry, the intrinsic geometry of the D-brane is in these coordinates simply
$$\mathrm{d}s_{3+1}^2=\mathrm{d}\tau ^2+a(\tau )^2\mathrm{d}\mathrm{\Omega }_3^2.$$
(11)
### 2.2 The D-brane as boundary:
A perhaps unusual (and for us very useful) feature of D-brane physics is that the D-brane can be viewed as an actual physical boundary to spacetime, with the “other side” of the D-brane being empty (null and void).<sup>7</sup><sup>7</sup>7 A brief sketch of these ideas, from the (3+1)-dimensional point of view where one is dealing with holes in spacetime (voids), was presented in . Here we expand on these ideas in a more explicit manner. In general relativity, as it is normally formulated, the notion of an actual physical boundary to spacetime (that is, an accessible boundary reachable at finite distance) is complete anathema. The reason that spacetime boundaries are so thoroughly deprecated in general relativity is that they are artificial special places in the manifold where some sort of boundary condition has to be placed on the physics. Without such a postulated boundary condition all predictability is lost, and the theory is not physically acceptable. Since without some deeper underlying theory there is no physically justifiable reason for picking any one particular type of boundary condition (Dirichlet, Neumann, Robin, or something more complicated), the attitude in standard general relativity has been to simply exclude boundaries.
The key difference when a D-brane is used as a boundary is that now there is a specific and well-defined boundary condition for the physics: D-branes (remember that “D is for Dirichlet”) are defined as the loci on which the fundamental open strings end (and satisfy Dirichlet-type boundary conditions). D-branes are therefore capable of providing both a physical boundary for the spacetime and a plausible boundary condition for the physics residing in the spacetime.<sup>8</sup><sup>8</sup>8 A word of warning: D-branes by definition provide boundary conditions directly on the fundamental string states, and so, since all physics in string theory can be viewed in terms of some combination of string states, D-branes will in principle provide boundary conditions for all the physics. In practice the route from string state to low-energy effective field theory may be rather indirect, and elucidation of the proper boundary condition may be a little obscure; when in doubt use symmetry as much as possible, and be prepared to keep at least a few adjustable constants as part of the low-energy semi-phenomenological theory.
When it comes to specific calculations, this is however not be the best mental picture to have in mind — after all, how would you try to calculate the Riemann tensor for the edge of spacetime? And what would happen to the Einstein equations at the edge? There is a specific technical trick that clarifies the situation: Take the manifold with D-brane boundary and make a second copy (including a second D-brane boundary), then sew the two manifolds together along their respective D-brane boundaries, creating a single manifold without boundary that contains the doubled D-brane, and exhibits a $`Z_2`$ symmetry on reflection around the D-brane. Because this new manifold is a perfectly reasonable no-boundary manifold containing a (thin shell) D-brane, the gravitational field can be analyzed using a slight generalization of the usual Israel–Lanczos–Sen thin-shell formalism of general relativity . We now need to consider (3+1) shells propagating in (4+1) space, but this merely changes a few integer coefficients. The metric is continuous, the connection exhibits a step-function discontinuity, and the Riemann curvature a delta-function at the D-brane. The dynamics of the D-brane can then be investigated in this $`Z_2`$-doubled manifold, and once the dynamical equations and their solutions have been investigated the second surplus copy of spacetime can quietly be forgotten (effectively halving the strength of the delta-function contribution to the Riemann tensor). That is, as long as one is working in the $`Z_2`$-doubled manifold the discontinuity in the extrinsic curvature is twice the extrinsic curvature as seen from either side
$$\kappa _{\alpha \beta }(Z_2)=[K_{\alpha \beta }]=K_{\alpha \beta }^+K_{\alpha \beta }^{}=2K_{\alpha \beta }.$$
(12)
Consequently the Riemann tensor in the $`Z_2`$-doubled manifold is
$`R(Z_2)_{\alpha \beta \gamma \delta }`$ $`=`$ $`2\delta (\eta )\left[K_{\alpha \gamma }n_\beta n_\delta +K_{\beta \delta }n_\alpha n_\gamma K_{\alpha \delta }n_\beta n_\gamma K_{\beta \gamma }n_\alpha n_\delta \right]`$ (13)
$`+\mathrm{\Theta }(\eta )R_{\alpha \beta \gamma \delta }^++\mathrm{\Theta }(\eta )R_{\alpha \beta \gamma \delta }^{}.`$
We now define the Riemann tensor of the manifold with boundary by throwing away half of the $`Z_2`$-doubled manifold, and in view of the manifest symmetry of the situation, also throwing away half the delta-function contribution.<sup>9</sup><sup>9</sup>9 If for whatever reason one does not wish to work with the $`Z_2`$-doubled manifold, there is an alternative construction that leads to the same result that we present in Appendix A.
After doing all this, near the D-brane boundary the Riemann tensor takes the form
$$R_{\alpha \beta \gamma \delta }=\delta (\eta )\left[K_{\alpha \gamma }n_\beta n_\delta +K_{\beta \delta }n_\alpha n_\gamma K_{\alpha \delta }n_\beta n_\gamma K_{\beta \gamma }n_\alpha n_\delta \right]+R_{\alpha \beta \gamma \delta }^{\mathrm{bulk}}.$$
(14)
This is the relevant generalization of equation (14.23) of to a manifold with boundary; note that there is only one side to the boundary and that we explicitly use only the extrinsic curvature of that one side (which is half the extrinsic curvature discontinuity in the $`Z_2`$-doubled manifold). This particular formula is valid for any (\[n–1\]+1)-dimensional boundary to a (n+1)-dimensional bulk. It does assume that the normal $`n`$ is spacelike, though no symmetry assumptions are made. If we introduce the general projection tensor
$$h_{\mu \nu }=g_{\mu \nu }n_\mu n_\nu ,$$
(15)
then this projection tensor is the induced metric on the boundary and in the particular application we have in mind will be the physical spacetime metric of our universe. Performing the relevant contractions, and still working in an arbitrary number of bulk dimensions
$$R_{\mu \nu }=\delta (\eta )\left[K_{\mu \nu }+Kn_\mu n_\nu \right]+R_{\mu \nu }^{\mathrm{bulk}};$$
(16)
$$R=2K\delta (\eta )+R^{\mathrm{bulk}};$$
(17)
$$G_{\mu \nu }=\delta (\eta )\left[K_{\mu \nu }Kh_{\mu \nu }\right]+G_{\mu \nu }^{\mathrm{bulk}}.$$
(18)
These formulae generalize (14.25)–(14.27) of to a manifold with boundary. With hindsight this makes perfectly good sense since if we now integrate over the complete manifold (bulk plus boundary)
$`{\displaystyle \mathrm{d}^{n+1}x\sqrt{g_{n+1}}R}`$ $`=`$ $`{\displaystyle _{\mathrm{bulk}}}\mathrm{d}^{n+1}x\sqrt{g_{n+1}}R_{\mathrm{bulk}}`$ (19)
$``$ $`2{\displaystyle _{\mathrm{boundary}}}\mathrm{d}^{[n1]+1}x\sqrt{g_{[n1]+1}}K.`$
Which means that we have automatically recovered the Gibbons–Hawking surface term for the gravitational action, in addition to the Einstein–Hilbert bulk term.
We also take the total stress-energy tensor to be given by a combination of surface and bulk components
$$T_{\mu \nu }=\delta (\eta )T_{\mu \nu }^{\mathrm{surface}}+T_{\mu \nu }^{\mathrm{bulk}},$$
(20)
and normalize our (n+1)-dimensional bulk Newton constant $`G_{n+1}`$ by
$$G_{\mu \nu }=8\pi G_{n+1}T_{\mu \nu }.$$
(21)
Then in particular, picking off the surface contribution to both the Einstein tensor and the stress-energy
$$8\pi G_{n+1}T_{\mu \nu }^{\mathrm{surface}}=\left[K_{\mu \nu }Kh_{\mu \nu }\right]$$
(22)
Whether or not this surface stress tensor satisfies the energy conditions depends on the signs of the eigenvalues of the extrinsic curvature. By looking at the $`Z_2`$-doubled geometry it is a general result that a convex boundary (when viewed from the bulk) violates the null energy condition (NEC), while a concave boundary satisfies it. (This is intimately related to the fact that traversable wormholes violate the null energy condition, see .)
### 2.3 Cosmology
Now particularize to the (4+1)-dimensional version of the thin-shell formalism, and use the FLRW symmetries of the D-brane (some of the integer coefficients and exponents appearing below are dimension dependent):
$$8\pi G_{4+1}\rho _{3+1}=3\frac{\sqrt{F(a)+\dot{a}^2}}{a}.$$
(23)
(Note that the energy density is positive definite, in agreement with the fact that this boundary is concave when viewed from the bulk.)<sup>10</sup><sup>10</sup>10 Because of this feature the D-brane occurring here is guaranteed to have positive tension, and we do not need to worry (at least not at the cosmological level) about the possibility of energy-condition-violating negative tension D-branes, and the somewhat peculiar features \[traversable wormholes, etc.\] that negative tension D-branes can introduce into the low energy effective theory .
$$8\pi G_{4+1}p_{3+1}=\frac{1}{a^2}\frac{\mathrm{d}}{\mathrm{d}a}\left(a^2\sqrt{F(a)+\dot{a}^2}\right).$$
(24)
These equations can easily be seen to be compatible with the conservation of the stress energy localized on the D-brane<sup>11</sup><sup>11</sup>11 There is another potential source of confusion here: Since the (3+1)-dimensional D-brane is sweeping through the (4+1)-dimensional bulk, why is it that the D-brane does not exchange energy with the bulk? One might at first glance expect violations of (3+1)-dimensional stress-energy conservation due to (4+1)-dimensional matter entering or leaving the D-brane. In fact, in general this might happen, and it is potentially an interesting observational signal to look for — but in the present cosmological context this effect is zero: as the D-brane moves through (4+1)-space, it is the “flux” of (4+1)-dimensional matter onto the brane, defined by
$$J_\mu =n_\alpha T^{\alpha \beta }\left[g_{\beta \mu }n_\beta n_\mu \right],$$
that determines whether or not (4+1)-dimensional stress-energy conservation holds . In all of the bulk geometries considered in this article, this flux is identically zero (in fact the stress-energy tensor is diagonal).
$$\frac{\mathrm{d}}{\mathrm{d}\tau }(\rho _{3+1}a^3)+p_{3+1}\frac{\mathrm{d}}{\mathrm{d}\tau }(a^3)=0.$$
(25)
So as usual, two of these three equations are independent, and the third is redundant.
The conservation equation is identical to that for standard cosmology, while the D-brane version of the Friedmann equation, obtained by rearranging the equation for the surface energy density that was given above, is seen to be
$$\left(\frac{\dot{a}}{a}\right)^2=\frac{F(a)}{a^2}+\left(\frac{8\pi G_{4+1}\rho _{3+1}}{3}\right)^2.$$
(26)
In contrast the standard Friedmann equation (for a $`k=+1`$ closed FLRW universe) is
$$\left(\frac{\dot{a}}{a}\right)^2=\frac{1}{a^2}+\frac{\mathrm{\Lambda }}{3}+\frac{8\pi G_{3+1}\rho }{3}.$$
(27)
To get a brane cosmology that is not wildly in conflict with observation, we split the (3+1)-dimensional energy into a constant $`\rho _0`$ determined by the brane tension, plus ordinary matter $`\rho `$, with $`\rho \rho _0`$ to suppress the quadratic term in comparison to the linear . Then with $`\rho _{3+1}=\rho _0+\rho `$ we have
$$\left(\frac{\dot{a}}{a}\right)^2=\frac{F(a)}{a^2}+\left(\frac{8\pi G_{4+1}\rho _0}{3}\right)^2+\left(\frac{16\pi G_{4+1}\rho _0}{3}\right)\left(\frac{8\pi G_{4+1}}{3}\right)\left[\rho +\frac{1}{2}\frac{\rho ^2}{\rho _0}\right].$$
(28)
Picking out the term linear in $`\rho `$, this permits us to identify
$$G_{3+1}=G_{4+1}\left(\frac{16\pi G_{4+1}\rho _0}{3}\right);\text{that is}G_{4+1}=\sqrt{\frac{3G_{3+1}}{16\pi \rho _0}}.$$
(29)
Therefore
$$\left(\frac{\dot{a}}{a}\right)^2=\frac{F(a)}{a^2}+\left(\frac{8\pi G_{3+1}}{3}\right)\left[\frac{1}{2}\rho _0+\rho +\frac{1}{2}\frac{\rho ^2}{\rho _0}\right].$$
(30)
Since we want $`\rho _0\rho `$ to suppress the quadratic term, this leaves us with a large (3+1)-dimensional cosmological constant that we will need to eliminate by cancelling it (either fully or partially) with some term in $`F(a)`$ . This result is in its own way quite remarkable: up to this point no assumptions had been made about the size of the brane tension, or even whether or not the brane tension was zero. Nor had any assumption been made up to this point about the existence or otherwise of any cosmological constant in the (4+1)-dimensional bulk. It is observational cosmology that first forces us to take $`\rho _0`$ large \[electro-weak scale or higher to avoid major problems with nucleosynthesis\], and then forces us to deduce the presence of an almost perfectly countervailing cosmological constant in the bulk .
In the next section we shall make use of this still relatively general formalism by specializing $`F(r)`$ to the Reissner–Nordström–de Sitter form.
### 2.4 Reissner–Nordström–de Sitter surgery
For the (4+1)-dimensional Reissner–Nordström–de Sitter geometry
$$F(r)=1\frac{2M_{4+1}}{r^2}+\frac{Q_{4+1}^2}{r^4}\frac{\mathrm{\Lambda }_{4+1}r^2}{6}.$$
(31)
Here $`M_{4+1}`$ is a (4+1)-dimensional “mass” parameter, corresponding to the mass of the central object in (4+1)-space — it does not have a ready (3+1)-dimensional interpretation and is best carried along as an extra free parameter that from the 4-dimensional point of view can be adjusted to taste. Similarly, $`Q_{4+1}`$ corresponds to an “electric charge” in the (4+1)-dimensional sense. Our universe, the boundary D-brane, must then be viewed as carrying an equal but opposite charge to allow field lines to terminate. From the (3+1)-dimensional view this may be taken to be a second free parameter. The (4+1)-dimensional cosmological constant combines with the term coming from the D-brane tension to give an effective (3+1)-dimensional cosmological constant
$$\mathrm{\Lambda }=\frac{\mathrm{\Lambda }_{4+1}}{2}+4\pi G_{3+1}\rho _0.$$
(32)
In the original Randall–Sundrum models these two terms were fine-tuned by hand to obtain complete cancellation. In view of the recent observational evidence for a small cosmological constant in our observable universe we need merely assert that this effective cosmological constant is presently relatively small (Λ
<
8πG3+1ρcritical
<
Λ8𝜋subscript𝐺31subscript𝜌critical\Lambda\mathrel{\lower 2.5pt\vbox{\hbox{$<$}\hbox{$\sim$}}}8\pi G_{3+1}\;\rho_{\mathrm{critical}}; this is small by particle physics standards, but can be quite significant by cosmological standards).<sup>12</sup><sup>12</sup>12 A small effective cosmological constant would indeed imply deviations from the original Randall–Sundrum scenario, but on a distance scale determined by this effective cosmological constant (and observationally this distance scale would be of order Giga-parsecs or larger). So we are not too concerned about this issue in that the implications for particle phenomenology are negligible.
Since $`\rho _0`$ is guaranteed positive,<sup>13</sup><sup>13</sup>13 Tricky point: actually it is $`\rho _{3+1}`$ that is guaranteed to be positive, and this holds because the D-brane universe is taken to be convex as seen from the bulk. Then the same logic that leads to energy condition violations for traversable wormholes now applies in reverse, and the (3+1) null energy condition is generically satisfied in this type of cosmological model. (Violating the strong energy condition, which is relevant for cosmological inflation, is much easier .) this implies that $`\mathrm{\Lambda }_{4+1}`$ should be negative, and so if this model is correct we are living on the edge of a bulk anti-de Sitter space. The D-brane dynamical equation now reads
$$\left(\frac{\dot{a}}{a}\right)^2=\frac{1}{a^2}+\frac{2M_{4+1}}{a^4}\frac{Q_{4+1}^2}{a^6}+\frac{\mathrm{\Lambda }}{3}+\left(\frac{8\pi G_{3+1}}{3}\right)\left[\rho +\frac{1}{2}\frac{\rho ^2}{\rho _0}\right].$$
(33)
It is clear that by tuning these parameters appropriately one can recover standard cosmology to arbitrary accuracy. The $`M_{4+1}`$ parameter can be used to mimic an arbitrary quantity of what would usually be called “radiation” (relativistic fluid, $`\rho =3p`$), while the $`Q_{4+1}`$ parameter mimics “stiff” matter ($`\rho =p`$, though with an overall minus sign. An observational astrophysicist or cosmologist could now simply forget about the underlying string theory and D-brane physics, take this expression as the D-brane inspired generalization of the Friedmann equations, and treat $`M_{4+1}`$, $`Q_{4+1}`$, $`\mathrm{\Lambda }`$, and $`\rho _0`$ as parameters to be observationally determined.
Since we actually want to do more than just reproduce standard cosmology we should seek some additional constraints on these parameters — and this is where the phenomenon of weak localization of the graviton near the brane comes into play.
## 3 Weak localization of perturbative gravity
Suppose that the observable universe is large compared to the natural distance scale in the (4+1)-dimensional bulk, that is: $`a(\tau )\sqrt{6/|\mathrm{\Lambda }_{4+1}|}`$ (so that the universe has “grown up”), and both $`M_{4+1}`$ and $`Q_{4+1}`$ are sufficiently small to allow us to approximate
$$F(r)\frac{|\mathrm{\Lambda }_{4+1}|r^2}{6};\text{for}ra.$$
(34)
Then near the D-brane we can write
$$\mathrm{d}s_{4+1}^2\frac{|\mathrm{\Lambda }_{4+1}|r^2}{6}\mathrm{d}t^2+\frac{6}{|\mathrm{\Lambda }_{4+1}|r^2}\mathrm{d}r^2+r^2\mathrm{d}\mathrm{\Omega }_3^2.$$
(35)
In terms of the normal distance (proper distance) from the D-brane,
$$\eta \sqrt{\frac{6}{|\mathrm{\Lambda }_{4+1}|}}\mathrm{ln}(r/a),$$
(36)
this implies
$$\mathrm{d}s_{4+1}^2+\mathrm{d}\eta ^2+\mathrm{exp}\left(2\sqrt{\frac{|\mathrm{\Lambda }_{4+1}|}{6}}\eta \right)\left[\frac{|\mathrm{\Lambda }_{4+1}|a^2}{6}\mathrm{d}t^2+a^2\mathrm{d}\mathrm{\Omega }_3^2\right].$$
(37)
If we now re-label our time parameter in terms of proper time measured along the D-brane (that is, use the proper time of a cosmologically comoving observer),
$$\tau \sqrt{\frac{|\mathrm{\Lambda }_{4+1}|a^2}{6}}t,$$
(38)
and introduce quasi-Cartesian coordinates to the tangent space at any arbitrary point point of the D-brane then
$$\mathrm{d}s_{4+1}^2+\mathrm{d}\eta ^2+\mathrm{exp}\left(2\sqrt{\frac{|\mathrm{\Lambda }_{4+1}|}{6}}\eta \right)\left[\mathrm{d}\tau ^2+\mathrm{d}x^2+\mathrm{d}y^2+dz^2\right].$$
(39)
Thus in this approximation the near-brane metric is precisely of the Randall–Sundrum form and we know from their analysis that there is a graviton bound state attached to the brane with an exponential falloff controlled by the distance scale parameter<sup>14</sup><sup>14</sup>14 Of course this is little more than the statement that if we are interested in laboratory physics in the here and now, then a tangent space approximation to cosmology had better work: Minkowski space is an excellent approximation for physics here on Earth and so the D-brane must exhibit at least approximate Lorentz symmetry if it is to be acceptable as a model of empirical reality. Moving off the D-brane and into the bulk, the only essential item is that at large enough distances \[from the (4+1) “centre”\] we must have $`F(r)r^2`$. Thus as long as the (4+1) geometry is asymptotically anti-de Sitter space we will recover Randall–Sundrum phenomenology on small scales. (And eventually, on large enough distance scales, the simple Randall–Sundrum phenomenology will break down either because of cosmological expansion, or because of the small effective (3+1)-dimensional cosmological constant, or simply because of the positive spatial curvature \[$`k=+1`$ and $`a`$ is finite\].)
$$L_{\mathrm{graviton}}=\sqrt{\frac{6}{|\mathrm{\Lambda }_{4+1}|}}.$$
(40)
Now the experimental fact that we do not see short distance deviations from the inverse square law of gravity at least down to centimetre scales implies that $`L_{\mathrm{graviton}}`$ is certainly less than one centimetre (and many would argue that it is at most one millimetre). Numerous experiments designed to tighten this limit are currently planned and in progress. Within the approximation that the (3+1)-dimensional effective cosmological constant is negligible we get
$$G_{3+1}=G_{4+1}\frac{2}{L_{\mathrm{graviton}}}.$$
(41)
The importance of these results for cosmology is that, given the observed almost perfect cancellation of the net cosmological constant,
$$\rho _0\frac{3}{4\pi G_{3+1}L_{\mathrm{graviton}}^2}=\frac{3}{4\pi }\frac{L_{\mathrm{Planck}}^2}{L_{\mathrm{graviton}}^2}\rho _{\mathrm{Planck}}$$
(42)
While this number is certainly large on a usual astrophysics scale, and is rather large even compared with nuclear densities, it could still be much less than the Planck scale and yet be compatible with experiment. Indeed if $`L_{\mathrm{graviton}}`$ is as large as a centimetre then the quadratic terms in the density become important once temperatures reach the electroweak scale (about 100 GeV). The good news is that this implies the model is compatible with standard cosmology at least back to the electroweak scale; the better news is that there are possibilities of seeing deviations from the standard cosmology as we go further back. The larger $`L_{\mathrm{graviton}}`$ is, the better things are with regard to the hierarchy problem of particle physics and the lower the brane tension needs to be. On the other hand, the lower $`L_{\mathrm{graviton}}`$ is the better the brane is at trapping gravitational perturbations and the less risk there is of conflict with gravity-based experiment.
## 4 Discussion
The Randall–Sundrum scenarios , and earlier tentative steps along these lines , have engendered a tremendous amount of activity, both in terms of particle physics and in terms of cosmology . In this paper we have sketched what we feel is perhaps the simplest most symmetric cosmology that can be based on these ideas: We have reduced the number of D-branes to exactly one, and have only one bulk (4+1)-dimensional region. The D-brane (which our observable universe lives on) is here viewed as an actual physical boundary to the higher-dimensional spacetime, and we have demonstrated how to write down both curvature tensor and field equations for a manifold with boundary.
We have verified that standard $`k=+1`$ FLRW cosmology can very easily be reproduced, and that we do not have massive present day violations of observational constraints. If you absolutely insist on a spatially flat $`k=0`$ geometry (or even a spatially hyperbolic $`k=1`$ geometry) that can also be achieved along the lines of this article, but at some cost in elegance, and for very little real purpose. Remember that for $`a(\tau )`$ large enough a $`k=+1`$ spatial slice mimics $`k=0`$ to arbitrary accuracy. In Appendix B and Appendix C we sketch how one could nevertheless force spatially flat or spatially hyperbolic FLRW cosmologies into this framework.
As is by now not unexpected , likely places to look for observational signatures are in short-distance (centimetre) deviations from the gravitational inverse square law, and in very early universe cosmology (before densities drop to the electro-weak scale; this is the region where the quadratic density term in the generalized Friedmann equation might come into play).<sup>15</sup><sup>15</sup>15 In particular, for $`\rho \rho _0`$ even ordinary radiation ($`\rho a^4`$) acts as though it has a $`a^8`$ behaviour, and this is enough to drive an epoch of power-law inflation with $`a(t)t^{1/4}`$ .
Because we are viewing the D-brane as an actual boundary, the conjectured connections between the Randall–Sundrum models and Maldacena’s adS/CFT conjecture are perhaps more compelling — we no longer have to deal with a $`Z_2`$-doubled version of the adS/CFT conjecture, but can work directly on a boundary of the (asymptotic) anti-de Sitter space. As the universe evolves in time the D-brane boundary moves further out into the asymptotic anti-de Sitter region, and this hints at a possible connection between cosmological time, the holographic hypothesis, and renormalization group flow .
## Acknowledgments
The research of CB was supported by the Spanish Ministry of Education and Culture (MEC). MV was supported by the US Department of Energy.
## Appendix A: Alternative construction for the <br>Riemann tensor of a manifold with boundary.
Take your original manifold $``$, with boundary $``$, and join to the boundary a hyper-tube of topology $`=(\mathrm{},0)`$. Let the metric on this hyper-tube be specified in terms of the induced metric on the boundary and the flat 1-dimensional metric:
$$g()=d\eta ^2g().$$
(43)
Then by construction $`K_{\alpha \beta }^{}=0`$ and $`K_{\alpha \beta }^+=K_{\alpha \beta }`$, so that in this geometry
$$\kappa _{\alpha \beta }()=[K_{\alpha \beta }]=K_{\alpha \beta }^+K_{\alpha \beta }^{}=K_{\alpha \beta },$$
(44)
leading to the Riemann tensor
$`R()_{\alpha \beta \gamma \delta }`$ $`=`$ $`\delta (\eta )\left[K_{\alpha \gamma }n_\beta n_\delta +K_{\beta \delta }n_\alpha n_\gamma K_{\alpha \delta }n_\beta n_\gamma K_{\beta \gamma }n_\alpha n_\delta \right]`$ (45)
$`+\mathrm{\Theta }(\eta )R()_{\alpha \beta \gamma \delta }^{\mathrm{bulk}}+\mathrm{\Theta }(\eta )R()_{\alpha \beta \gamma \delta }^{\mathrm{bulk}}.`$
Now truncate the geometry by simply throwing away the hyper-tube $``$. The Riemann tensor in the remaining manifold $``$ is, as before
$$R()_{\alpha \beta \gamma \delta }=\delta (\eta )\left[K_{\alpha \gamma }n_\beta n_\delta +K_{\beta \delta }n_\alpha n_\gamma K_{\alpha \delta }n_\beta n_\gamma K_{\beta \gamma }n_\alpha n_\delta \right]+R()_{\alpha \beta \gamma \delta }^{\mathrm{bulk}}.$$
(46)
There is now no symmetry to suggest that one should perform any particular splitting of the delta-function contribution at the boundary, and in fact the observation that the extrinsic curvature is by construction zero on the hyper-tube side of the boundary is an indication that you should assign all the delta-function contribution to $``$, the resulting manifold with boundary. Either construction (hyper-tube addition or $`Z_2`$-doubling) leads to the same result for the Riemann tensor, but some may be happier with one construction over the other.
## Appendix B: Spatially flat FLRW cosmology.
By a little guess-work based on hyper-spherically symmetric Reissner–Nordström–de Sitter space one is led to consider the metric
$$\mathrm{d}s_{4+1}^2=F(r)\mathrm{d}t^2+\frac{\mathrm{d}r^2}{F(r)}+r^2\left[\mathrm{d}x^2+\mathrm{d}y^2+\mathrm{d}z^2\right].$$
(47)
with (note the absence of the leading $`1`$!)
$$F(r)=\frac{2M_{4+1}}{r^2}+\frac{Q_{4+1}^2}{r^4}\frac{\mathrm{\Lambda }_{4+1}r^2}{6}.$$
(48)
This metric still satisfies the (4+1)-dimensional Einstein–Maxwell equations, but with a hyper-planar symmetry instead of a hyper-spherical symmetry. You can now re-do the analysis of this note by placing a spatially flat D-brane boundary at $`r=a(t)`$ and will obtain very similar results to those of this article. The intrinsic geometry of the D-brane will now be
$$\mathrm{d}s_{3+1}^2=\mathrm{d}\tau ^2+a^2(\tau )\left[\mathrm{d}x^2+\mathrm{d}y^2+\mathrm{d}z^2\right].$$
(49)
It is not clear to us that the marginal change in the Friedmann equation is worth the loss of hyper-spherical symmetry. The point $`r=0`$ is still (for $`M_{4+1}0`$ or $`Q_{4+1}0`$) a curvature singularity of the (4+1)-dimensional bulk, but whether you really want to call it the “center” of the bulk (as opposed to say a “focal point”) is somewhat less than clear.
## Appendix C: Spatially hyperbolic FLRW cosmology.
Inspired by the previous guess-work one is led to consider the metric (note the presence of the $`\mathrm{sinh}`$ function)
$$\mathrm{d}s_{4+1}^2=F(r)\mathrm{d}t^2+\frac{\mathrm{d}r^2}{F(r)}+r^2\left[\mathrm{d}\chi ^2+\mathrm{sinh}^2\chi (\mathrm{d}\theta ^2+\mathrm{sin}^2\theta \mathrm{d}\varphi ^2)\right].$$
(50)
with (note the presence of the leading minus $`1`$!)
$$F(r)=1\frac{2M_{4+1}}{r^2}+\frac{Q_{4+1}^2}{r^4}\frac{\mathrm{\Lambda }_{4+1}r^2}{6}.$$
(51)
This metric also satisfies the (4+1)-dimensional Einstein–Maxwell equations, but with a hyperbolic symmetry instead of either hyper-spherical or hyper-planar symmetry. You can now re-do the analysis of this note by placing a spatially hyperbolic D-brane boundary at $`r=a(t)`$ and will again obtain very similar results to those of this article. The intrinsic geometry of the D-brane will now be
$$\mathrm{d}s_{3+1}^2=\mathrm{d}\tau ^2+a^2(\tau )\left[\mathrm{d}\chi ^2+\mathrm{sinh}^2\chi (\mathrm{d}\theta ^2+\mathrm{sin}^2\theta \mathrm{d}\varphi ^2)\right].$$
(52)
For all three cases, $`k=+1,0,1`$, the formal dynamical equation for the brane motion \[equation (30), valid for arbitrary $`F(r)`$\] is unchanged, while the explicit dynamical equation after Reissner–Nordström–de Sitter surgery (the generalized Friedman equation) becomes
$$\left(\frac{\dot{a}}{a}\right)^2=\frac{k}{a^2}+\frac{2M_{4+1}}{a^4}\frac{Q_{4+1}^2}{a^6}+\frac{\mathrm{\Lambda }}{3}+\left(\frac{8\pi G_{3+1}}{3}\right)\left[\rho +\frac{1}{2}\frac{\rho ^2}{\rho _0}\right].$$
(53)
|
warning/0004/astro-ph0004075.html
|
ar5iv
|
text
|
# The Cosmological Constant
## 1 Introduction
### 1.1 Truth and beauty
Science is rarely tidy. We ultimately seek a unified explanatory framework characterized by elegance and simplicity; along the way, however, our aesthetic impulses must occasionally be sacrificed to the desire to encompass the largest possible range of phenomena (i.e., to fit the data). It is often the case that an otherwise compelling theory, in order to be brought into agreement with observation, requires some apparently unnatural modification. Some such modifications may eventually be discarded as unnecessary once the phenomena are better understood; at other times, advances in our theoretical understanding will reveal that a certain theoretical compromise is only superficially distasteful, when in fact it arises as the consequence of a beautiful underlying structure.
General relativity is a paradigmatic example of a scientific theory of impressive power and simplicity. The cosmological constant, meanwhile, is a paradigmatic example of a modification, originally introduced to help fit the data, which appears at least on the surface to be superfluous and unattractive. Its original role, to allow static homogeneous solutions to Einstein’s equations in the presence of matter, turned out to be unnecessary when the expansion of the universe was discovered , and there have been a number of subsequent episodes in which a nonzero cosmological constant was put forward as an explanation for a set of observations and later withdrawn when the observational case evaporated. Meanwhile, particle theorists have realized that the cosmological constant can be interpreted as a measure of the energy density of the vacuum. This energy density is the sum of a number of apparently unrelated contributions, each of magnitude much larger than the upper limits on the cosmological constant today; the question of why the observed vacuum energy is so small in comparison to the scales of particle physics has become a celebrated puzzle, although it is usually thought to be easier to imagine an unknown mechanism which would set it precisely to zero than one which would suppress it by just the right amount to yield an observationally accessible cosmological constant.
This checkered history has led to a certain reluctance to consider further invocations of a nonzero cosmological constant; however, recent years have provided the best evidence yet that this elusive quantity does play an important dynamical role in the universe. This possibility, although still far from a certainty, makes it worthwhile to review the physics and astrophysics of the cosmological constant (and its modern equivalent, the energy of the vacuum).
There are a number of other reviews of various aspects of the cosmological constant; in the present article I will outline the most relevant issues, but not try to be completely comprehensive, focusing instead on providing a pedagogical introduction and explaining recent advances. For astrophysical aspects, I did not try to duplicate much of the material in Carroll, Press and Turner , which should be consulted for numerous useful formulae and a discussion of several kinds of observational tests not covered here. Some earlier discussions include , and subsequent reviews include . The classic discussion of the physics of the cosmological constant is by Weinberg , with more recent work discussed by . For introductions to cosmology, see .
### 1.2 Introducing the cosmological constant
Einstein’s original field equations are
$$R_{\mu \nu }\frac{1}{2}Rg_{\mu \nu }=8\pi GT_{\mu \nu }.$$
(1)
(I use conventions in which $`c=1`$, and will also set $`\mathrm{}=1`$ in most of the formulae to follow, but Newton’s constant will be kept explicit.) On very large scales the universe is spatially homogeneous and isotropic to an excellent approximation, which implies that its metric takes the Robertson-Walker form
$$\mathrm{d}s^2=\mathrm{d}t^2+a^2(t)R_0^2\left[\frac{\mathrm{d}r^2}{1kr^2}+r^2\mathrm{d}\mathrm{\Omega }^2\right],$$
(2)
where $`\mathrm{d}\mathrm{\Omega }^2=\mathrm{d}\theta ^2+\mathrm{sin}^2\theta \mathrm{d}\varphi ^2`$ is the metric on a two-sphere. The curvature parameter $`k`$ takes on values $`+1`$, $`0`$, or $`1`$ for positively curved, flat, and negatively curved spatial sections, respectively. The scale factor characterizes the relative size of the spatial sections as a function of time; we have written it in a normalized form $`a(t)=R(t)/R_0`$, where the subscript $`0`$ will always refer to a quantity evaluated at the present time. The redshift $`z`$ undergone by radiation from a comoving object as it travels to us today is related to the scale factor at which it was emitted by
$$a=\frac{1}{(1+z)}.$$
(3)
The energy-momentum sources may be modeled as a perfect fluid, specified by an energy density $`\rho `$ and isotropic pressure $`p`$ in its rest frame. The energy-momentum tensor of such a fluid is
$$T_{\mu \nu }=(\rho +p)U_\mu U_\nu +pg_{\mu \nu },$$
(4)
where $`U^\mu `$ is the fluid four-velocity. To obtain a Robertson-Walker solution to Einstein’s equations, the rest frame of the fluid must be that of a comoving observer in the metric (2); in that case, Einstein’s equations reduce to the two Friedmann equations
$$H^2\left(\frac{\dot{a}}{a}\right)^2=\frac{8\pi G}{3}\rho \frac{k}{a^2R_0^2},$$
(5)
where we have introduced the Hubble parameter $`H\dot{a}/a`$, and
$$\frac{\ddot{a}}{a}=\frac{4\pi G}{3}(\rho +3p).$$
(6)
Einstein was interested in finding static ($`\dot{a}=0`$) solutions, both due to his hope that general relativity would embody Mach’s principle that matter determines inertia, and simply to account for the astronomical data as they were understood at the time.<sup>1</sup><sup>1</sup>1This account gives short shrift to the details of what actually happened; for historical background see . A static universe with a positive energy density is compatible with (5) if the spatial curvature is positive ($`k=+1`$) and the density is appropriately tuned; however, (6) implies that $`\ddot{a}`$ will never vanish in such a spacetime if the pressure $`p`$ is also nonnegative (which is true for most forms of matter, and certainly for ordinary sources such as stars and gas). Einstein therefore proposed a modification of his equations, to
$$R_{\mu \nu }\frac{1}{2}Rg_{\mu \nu }+\mathrm{\Lambda }g_{\mu \nu }=8\pi GT_{\mu \nu },$$
(7)
where $`\mathrm{\Lambda }`$ is a new free parameter, the cosmological constant. Indeed, the left-hand side of (7) is the most general local, coordinate-invariant, divergenceless, symmetric, two-index tensor we can construct solely from the metric and its first and second derivatives. With this modification, the Friedmann equations become
$$H^2=\frac{8\pi G}{3}\rho +\frac{\mathrm{\Lambda }}{3}\frac{k}{a^2R_0^2}.$$
(8)
and
$$\frac{\ddot{a}}{a}=\frac{4\pi G}{3}(\rho +3p)+\frac{\mathrm{\Lambda }}{3}.$$
(9)
These equations admit a static solution with positive spatial curvature and all the parameters $`\rho `$, $`p`$, and $`\mathrm{\Lambda }`$ nonnegative. This solution is called the “Einstein static universe.”
The discovery by Hubble that the universe is expanding eliminated the empirical need for a static world model (although the Einstein static universe continues to thrive in the toolboxes of theorists, as a crucial step in the construction of conformal diagrams). It has also been criticized on the grounds that any small deviation from a perfect balance between the terms in (9) will rapidly grow into a runaway departure from the static solution.
Pandora’s box, however, is not so easily closed. The disappearance of the original motivation for introducing the cosmological constant did not change its status as a legitimate addition to the gravitational field equations, or as a parameter to be constrained by observation. The only way to purge $`\mathrm{\Lambda }`$ from cosmological discourse would be to measure all of the other terms in (8) to sufficient precision to be able to conclude that the $`\mathrm{\Lambda }/3`$ term is negligibly small in comparison, a feat which has to date been out of reach. As discussed below, there is better reason than ever before to believe that $`\mathrm{\Lambda }`$ is actually nonzero, and Einstein may not have blundered after all.
### 1.3 Vacuum energy
The cosmological constant $`\mathrm{\Lambda }`$ is a dimensionful parameter with units of (length)<sup>-2</sup>. From the point of view of classical general relativity, there is no preferred choice for what the length scale defined by $`\mathrm{\Lambda }`$ might be. Particle physics, however, brings a different perspective to the question. The cosmological constant turns out to be a measure of the energy density of the vacuum — the state of lowest energy — and although we cannot calculate the vacuum energy with any confidence, this identification allows us to consider the scales of various contributions to the cosmological constant .
Consider a single scalar field $`\varphi `$, with potential energy $`V(\varphi )`$. The action can be written
$$S=d^4x\sqrt{g}\left[\frac{1}{2}g^{\mu \nu }_\mu \varphi _\nu \varphi V(\varphi )\right]$$
(10)
(where $`g`$ is the determinant of the metric tensor $`g_{\mu \nu }`$), and the corresponding energy-momentum tensor is
$$T_{\mu \nu }=\frac{1}{2}_\mu \varphi _\nu \varphi +\frac{1}{2}(g^{\rho \sigma }_\rho \varphi _\sigma \varphi )g_{\mu \nu }V(\varphi )g_{\mu \nu }.$$
(11)
In this theory, the configuration with the lowest energy density (if it exists) will be one in which there is no contribution from kinetic or gradient energy, implying $`_\mu \varphi =0`$, for which $`T_{\mu \nu }=V(\varphi _0)g_{\mu \nu }`$, where $`\varphi _0`$ is the value of $`\varphi `$ which minimizes $`V(\varphi )`$. There is no reason in principle why $`V(\varphi _0)`$ should vanish. The vacuum energy-momentum tensor can thus be written
$$T_{\mu \nu }^{\mathrm{vac}}=\rho _{\mathrm{vac}}g_{\mu \nu },$$
(12)
with $`\rho _{\mathrm{vac}}`$ in this example given by $`V(\varphi _0)`$. (This form for the vacuum energy-momentum tensor can also be argued for on the more general grounds that it is the only Lorentz-invariant form for $`T_{\mu \nu }^{\mathrm{vac}}`$.) The vacuum can therefore be thought of as a perfect fluid as in (4), with
$$p_{\mathrm{vac}}=\rho _{\mathrm{vac}}.$$
(13)
The effect of an energy-momentum tensor of the form (12) is equivalent to that of a cosmological constant, as can be seen by moving the $`\mathrm{\Lambda }g_{\mu \nu }`$ term in (7) to the right-hand side and setting
$$\rho _{\mathrm{vac}}=\rho _\mathrm{\Lambda }\frac{\mathrm{\Lambda }}{8\pi G}.$$
(14)
This equivalence is the origin of the identification of the cosmological constant with the energy of the vacuum. In what follows, I will use the terms “vacuum energy” and “cosmological constant” essentially interchangeably.
It is not necessary to introduce scalar fields to obtain a nonzero vacuum energy. The action for general relativity in the presence of a “bare” cosmological constant $`\mathrm{\Lambda }_0`$ is
$$S=\frac{1}{16\pi G}d^4x\sqrt{g}(R2\mathrm{\Lambda }_0),$$
(15)
where $`R`$ is the Ricci scalar. Extremizing this action (augmented by suitable matter terms) leads to the equations (7). Thus, the cosmological constant can be thought of as simply a constant term in the Lagrange density of the theory. Indeed, (15) is the most general covariant action we can construct out of the metric and its first and second derivatives, and is therefore a natural starting point for a theory of gravity.
Classically, then, the effective cosmological constant is the sum of a bare term $`\mathrm{\Lambda }_0`$ and the potential energy $`V(\varphi )`$, where the latter may change with time as the universe passes through different phases. Quantum mechanics adds another contribution, from the zero-point energies associated with vacuum fluctuations. Consider a simple harmonic oscillator, i.e. a particle moving in a one-dimensional potential of the form $`V(x)=\frac{1}{2}\omega ^2x^2`$. Classically, the “vacuum” for this system is the state in which the particle is motionless and at the minimum of the potential ($`x=0`$), for which the energy in this case vanishes. Quantum-mechanically, however, the uncertainty principle forbids us from isolating the particle both in position and momentum, and we find that the lowest energy state has an energy $`E_0=\frac{1}{2}\mathrm{}\omega `$ (where I have temporarily re-introduced explicit factors of $`\mathrm{}`$ for clarity). Of course, in the absence of gravity either system actually has a vacuum energy which is completely arbitrary; we could add any constant to the potential (including, for example, $`\frac{1}{2}\mathrm{}\omega `$) without changing the theory. It is important, however, that the zero-point energy depends on the system, in this case on the frequency $`\omega `$.
A precisely analogous situation holds in field theory. A (free) quantum field can be thought of as a collection of an infinite number of harmonic oscillators in momentum space. Formally, the zero-point energy of such an infinite collection will be infinite. (See for further details.) If, however, we discard the very high-momentum modes on the grounds that we trust our theory only up to a certain ultraviolet momentum cutoff $`k_{\mathrm{max}}`$, we find that the resulting energy density is of the form
$$\rho _\mathrm{\Lambda }\mathrm{}k_{\mathrm{max}}^4.$$
(16)
This answer could have been guessed by dimensional analysis; the numerical constants which have been neglected will depend on the precise theory under consideration. Again, in the absence of gravity this energy has no effect, and is traditionally discarded (by a process known as “normal-ordering”). However, gravity does exist, and the actual value of the vacuum energy has important consequences. (And the vacuum fluctuations themselves are very real, as evidenced by the Casimir effect .)
The net cosmological constant, from this point of view, is the sum of a number of apparently disparate contributions, including potential energies from scalar fields and zero-point fluctuations of each field theory degree of freedom, as well as a bare cosmological constant $`\mathrm{\Lambda }_0`$. Unlike the last of these, in the first two cases we can at least make educated guesses at the magnitudes. In the Weinberg-Salam electroweak model, the phases of broken and unbroken symmetry are distinguished by a potential energy difference of approximately $`M_{\mathrm{EW}}200`$ GeV (where 1 GeV $`=1.6\times 10^3`$ erg); the universe is in the broken-symmetry phase during our current low-temperature epoch, and is believed to have been in the symmetric phase at sufficiently high temperatures in the early universe. The effective cosmological constant is therefore different in the two epochs; absent some form of prearrangement, we would naturally expect a contribution to the vacuum energy today of order
$$\rho _\mathrm{\Lambda }^{\mathrm{EW}}(200\mathrm{GeV})^43\times 10^{47}\mathrm{erg}/\mathrm{cm}^3.$$
(17)
Similar contributions can arise even without invoking “fundamental” scalar fields. In the strong interactions, chiral symmetry is believed to be broken by a nonzero expectation value of the quark bilinear $`\overline{q}q`$ (which is itself a scalar, although constructed from fermions). In this case the energy difference between the symmetric and broken phases is of order the QCD scale $`M_{\mathrm{QCD}}0.3`$ GeV, and we would expect a corresponding contribution to the vacuum energy of order
$$\rho _\mathrm{\Lambda }^{\mathrm{QCD}}(0.3\mathrm{GeV})^41.6\times 10^{36}\mathrm{erg}/\mathrm{cm}^3.$$
(18)
These contributions are joined by those from any number of unknown phase transitions in the early universe, such as a possible contribution from grand unification of order $`M_{\mathrm{GUT}}10^{16}`$ GeV. In the case of vacuum fluctuations, we should choose our cutoff at the energy past which we no longer trust our field theory. If we are confident that we can use ordinary quantum field theory all the way up to the Planck scale $`M_{\mathrm{Pl}}=(8\pi G)^{1/2}10^{18}`$ GeV, we expect a contribution of order
$$\rho _\mathrm{\Lambda }^{\mathrm{Pl}}(10^{18}\mathrm{GeV})^42\times 10^{110}\mathrm{erg}/\mathrm{cm}^3.$$
(19)
Field theory may fail earlier, although quantum gravity is the only reason we have to believe it will fail at any specific scale.
As we will discuss later, cosmological observations imply
$$|\rho _\mathrm{\Lambda }^{(\mathrm{obs})}|(10^{12}\mathrm{GeV})^42\times 10^{10}\mathrm{erg}/\mathrm{cm}^3,$$
(20)
much smaller than any of the individual effects listed above. The ratio of (19) to (20) is the origin of the famous discrepancy of 120 orders of magnitude between the theoretical and observational values of the cosmological constant. There is no obstacle to imagining that all of the large and apparently unrelated contributions listed add together, with different signs, to produce a net cosmological constant consistent with the limit (20), other than the fact that it seems ridiculous. We know of no special symmetry which could enforce a vanishing vacuum energy while remaining consistent with the known laws of physics; this conundrum is the “cosmological constant problem”. In section 4 we will discuss a number of issues related to this puzzle, which at this point remains one of the most significant unsolved problems in fundamental physics.
## 2 Cosmology with a cosmological constant
### 2.1 Cosmological parameters
From the Friedmann equation (5) (where henceforth we take the effects of a cosmological constant into account by including the vacuum energy density $`\rho _\mathrm{\Lambda }`$ into the total density $`\rho `$), for any value of the Hubble parameter $`H`$ there is a critical value of the energy density such that the spatial geometry is flat ($`k=0`$):
$$\rho _{\mathrm{crit}}\frac{3H^2}{8\pi G}.$$
(21)
It is often most convenient to measure the total energy density in terms of the critical density, by introducing the density parameter
$$\mathrm{\Omega }\frac{\rho }{\rho _{\mathrm{crit}}}=\left(\frac{8\pi G}{3H^2}\right)\rho .$$
(22)
One useful feature of this parameterization is a direct connection between the value of $`\mathrm{\Omega }`$ and the spatial geometry:
$$k=\mathrm{sgn}(\mathrm{\Omega }1).$$
(23)
\[Keep in mind that some references still use “$`\mathrm{\Omega }`$” to refer strictly to the density parameter in matter, even in the presence of a cosmological constant; with this definition (23) no longer holds.\]
In general, the energy density $`\rho `$ will include contributions from various distinct components. From the point of view of cosmology, the relevant feature of each component is how its energy density evolves as the universe expands. Fortunately, it is often (although not always) the case that individual components $`i`$ have very simple equations of state of the form
$$p_i=w_i\rho _i,$$
(24)
with $`w_i`$ a constant. Plugging this equation of state into the energy-momentum conservation equation $`_\mu T^{\mu \nu }=0`$, we find that the energy density has a power-law dependence on the scale factor,
$$\rho _ia^{n_i},$$
(25)
where the exponent is related to the equation of state parameter by
$$n_i=3(1+w_i).$$
(26)
The density parameter in each component is defined in the obvious way,
$$\mathrm{\Omega }_i\frac{\rho _i}{\rho _{\mathrm{crit}}}=\left(\frac{8\pi G}{3H^2}\right)\rho _i,$$
(27)
which has the useful property that
$$\frac{\mathrm{\Omega }_i}{\mathrm{\Omega }_j}a^{(n_in_j)}.$$
(28)
The simplest example of a component of this form is a set of massive particles with negligible relative velocities, known in cosmology as “dust” or simply “matter”. The energy density of such particles is given by their number density times their rest mass; as the universe expands, the number density is inversely proportional to the volume while the rest masses are constant, yielding $`\rho _\mathrm{M}a^3`$. For relativistic particles, known in cosmology as “radiation” (although any relativistic species counts, not only photons or even strictly massless particles), the energy density is the number density times the particle energy, and the latter is proportional to $`a^1`$ (redshifting as the universe expands); the radiation energy density therefore scales as $`\rho _\mathrm{R}a^4`$. Vacuum energy does not change as the universe expands, so $`\rho _\mathrm{\Lambda }a^0`$; from (26) this implies a negative pressure, or positive tension, when the vacuum energy is positive. Finally, for some purposes it is useful to pretend that the $`ka^2R_0^2`$ term in (5) represents an effective “energy density in curvature”, and define $`\rho _k(3k/8\pi GR_0^2)a^2`$. We can define a corresponding density parameter
$$\mathrm{\Omega }_k=1\mathrm{\Omega };$$
(29)
this relation is simply (5) divided by $`H^2`$. Note that the contribution from $`\mathrm{\Omega }_k`$ is (for obvious reasons) not included in the definition of $`\mathrm{\Omega }`$. The usefulness of $`\mathrm{\Omega }_k`$ is that it contributes to the expansion rate analogously to the honest density parameters $`\mathrm{\Omega }_i`$; we can write
$$H(a)=H_0\left(\underset{i(k)}{}\mathrm{\Omega }_{i0}a^{n_i}\right)^{1/2},$$
(30)
where the notation $`_{i(k)}`$ reflects the fact that the sum includes $`\mathrm{\Omega }_k`$ in addition to the various components of $`\mathrm{\Omega }=_i\mathrm{\Omega }_i`$. The most popular equations of state for cosmological energy sources can thus be summarized as follows:
| | $`w_i`$ | $`n_i`$ |
| --- | --- | --- |
| matter | 0 | 3 |
| radiation | $`1/3`$ | 4 |
| “curvature” | $`1/3`$ | 2 |
| vacuum | $`1`$ | 0 |
(31)
The ranges of values of the $`\mathrm{\Omega }_i`$’s which are allowed in principle (as opposed to constrained by observation) will depend on a complete theory of the matter fields, but lacking that we may still invoke energy conditions to get a handle on what constitutes sensible values. The most appropriate condition is the dominant energy condition (DEC), which states that $`T_{\mu \nu }l^\mu l^\nu 0`$, and $`T^\mu {}_{\nu }{}^{}l_{}^{\mu }`$ is non-spacelike, for any null vector $`l^\mu `$; this implies that energy does not flow faster than the speed of light . For a perfect-fluid energy-momentum tensor of the form (4), these two requirements imply that $`\rho +p0`$ and $`|\rho ||p|`$, respectively. Thus, either the density is positive and greater in magnitude than the pressure, or the density is negative and equal in magnitude to a compensating positive pressure; in terms of the equation-of-state parameter $`w`$, we have either positive $`\rho `$ and $`|w|1`$ or negative $`\rho `$ and $`w=1`$. That is, a negative energy density is allowed only if it is in the form of vacuum energy. (We have actually modified the conventional DEC somewhat, by using only null vectors $`l^\mu `$ rather than null or timelike vectors; the traditional condition would rule out a negative cosmological constant, which there is no physical reason to do.)
There are good reasons to believe that the energy density in radiation today is much less than that in matter. Photons, which are readily detectable, contribute $`\mathrm{\Omega }_\gamma 5\times 10^5`$, mostly in the $`2.73^{}`$K cosmic microwave background . If neutrinos are sufficiently low mass as to be relativistic today, conventional scenarios predict that they contribute approximately the same amount . In the absence of sources which are even more exotic, it is therefore useful to parameterize the universe today by the values of $`\mathrm{\Omega }_\mathrm{M}`$ and $`\mathrm{\Omega }_\mathrm{\Lambda }`$, with $`\mathrm{\Omega }_k=1\mathrm{\Omega }_\mathrm{M}\mathrm{\Omega }_\mathrm{\Lambda }`$, keeping the possibility of surprises always in mind.
One way to characterize a specific Friedmann-Robertson-Walker model is by the values of the Hubble parameter and the various energy densities $`\rho _i`$. (Of course, reconstructing the history of such a universe also requires an understanding of the microphysical processes which can exchange energy between the different states.) It may be difficult, however, to directly measure the different contributions to $`\rho `$, and it is therefore useful to consider extracting these quantities from the behavior of the scale factor as a function of time. A traditional measure of the evolution of the expansion rate is the deceleration parameter
$$\begin{array}{ccc}\hfill q& & \frac{\ddot{a}a}{\dot{a}^2}\hfill \\ & =& \underset{i}{}\frac{n_i2}{2}\mathrm{\Omega }_i\hfill \\ & =& \frac{1}{2}\mathrm{\Omega }_\mathrm{M}\mathrm{\Omega }_\mathrm{\Lambda },\hfill \end{array}$$
(32)
where in the last line we have assumed that the universe is dominated by matter and the cosmological constant. Under the assumption that $`\mathrm{\Omega }_\mathrm{\Lambda }=0`$, measuring $`q_0`$ provides a direct measurement of the current density parameter $`\mathrm{\Omega }_{\mathrm{M0}}`$; however, once $`\mathrm{\Omega }_\mathrm{\Lambda }`$ is admitted as a possibility there is no single parameter which characterizes various universes, and for most purposes it is more convenient to simply quote experimental results directly in terms of $`\mathrm{\Omega }_\mathrm{M}`$ and $`\mathrm{\Omega }_\mathrm{\Lambda }`$. \[Even this parameterization, of course, bears a certain theoretical bias which may not be justified; ultimately, the only unbiased method is to directly quote limits on $`a(t)`$.\]
Notice that positive-energy-density sources with $`n>2`$ cause the universe to decelerate while $`n<2`$ leads to acceleration; the more rapidly energy density redshifts away, the greater the tendency towards universal deceleration. An empty universe ($`\mathrm{\Omega }=0`$, $`\mathrm{\Omega }_k=1`$) expands linearly with time; sometimes called the “Milne universe”, such a spacetime is really flat Minkowski space in an unusual time-slicing.
### 2.2 Model universes and their fates
In the remainder of this section we will explore the behavior of universes dominated by matter and vacuum energy, $`\mathrm{\Omega }=\mathrm{\Omega }_\mathrm{M}+\mathrm{\Omega }_\mathrm{\Lambda }=1\mathrm{\Omega }_k`$. According to (32), a positive cosmological constant accelerates the universal expansion, while a negative cosmological constant and/or ordinary matter tend to decelerate it. The relative contributions of these components change with time; according to (28) we have
$$\mathrm{\Omega }_\mathrm{\Lambda }a^2\mathrm{\Omega }_ka^3\mathrm{\Omega }_\mathrm{M}.$$
(33)
For $`\mathrm{\Omega }_\mathrm{\Lambda }<0`$, the universe will always recollapse to a Big Crunch, either because there is a sufficiently high matter density or due to the eventual domination of the negative cosmological constant. For $`\mathrm{\Omega }_\mathrm{\Lambda }>0`$ the universe will expand forever unless there is sufficient matter to cause recollapse before $`\mathrm{\Omega }_\mathrm{\Lambda }`$ becomes dynamically important. For $`\mathrm{\Omega }_\mathrm{\Lambda }=0`$ we have the familiar situation in which $`\mathrm{\Omega }_\mathrm{M}1`$ universes expand forever and $`\mathrm{\Omega }_\mathrm{M}>1`$ universes recollapse; notice, however, that in the presence of a cosmological constant there is no necessary relationship between spatial curvature and the fate of the universe. (Furthermore, we cannot reliably determine that the universe will expand forever by any set of measurements of $`\mathrm{\Omega }_\mathrm{\Lambda }`$ and $`\mathrm{\Omega }_\mathrm{M}`$; even if we seem to live in a parameter space that predicts eternal expansion, there is always the possibility of a future phase transition which could change the equation of state of one or more of the components.)
Given $`\mathrm{\Omega }_\mathrm{M}`$, the value of $`\mathrm{\Omega }_\mathrm{\Lambda }`$ for which the universe will expand forever is given by
$$\mathrm{\Omega }_\mathrm{\Lambda }\{\begin{array}{cc}0\hfill & 0\mathrm{\Omega }_\mathrm{M}1\hfill \\ 4\mathrm{\Omega }_\mathrm{M}\mathrm{cos}^3\left[\frac{1}{3}\mathrm{cos}^1\left(\frac{1\mathrm{\Omega }_\mathrm{M}}{\mathrm{\Omega }_\mathrm{M}}\right)+\frac{4\pi }{3}\right]\hfill & \mathrm{\Omega }_\mathrm{M}>1.\hfill \end{array}$$
(34)
Conversely, if the cosmological constant is sufficiently large compared to the matter density, the universe has always been accelerating, and rather than a Big Bang its early history consisted of a period of gradually slowing contraction to a minimum radius before beginning its current expansion. The criterion for there to have been no singularity in the past is
$$\mathrm{\Omega }_\mathrm{\Lambda }4\mathrm{\Omega }_\mathrm{M}\mathrm{coss}^3\left[\frac{1}{3}\mathrm{coss}^1\left(\frac{1\mathrm{\Omega }_\mathrm{M}}{\mathrm{\Omega }_\mathrm{M}}\right)\right],$$
(35)
where “coss” represents $`\mathrm{cosh}`$ when $`\mathrm{\Omega }_\mathrm{M}<1/2`$, and $`\mathrm{cos}`$ when $`\mathrm{\Omega }_\mathrm{M}>1/2`$.
The dynamics of universes with $`\mathrm{\Omega }=\mathrm{\Omega }_\mathrm{M}+\mathrm{\Omega }_\mathrm{\Lambda }`$ are summarized in Figure (1), in which the arrows indicate the evolution of these parameters in an expanding universe. (In a contracting universe they would be reversed.)
This is not a true phase-space plot, despite the superficial similarities. One important difference is that a universe passing through one point can pass through the same point again but moving backwards along its trajectory, by first going to infinity and then turning around (recollapse).
Figure (1) includes three fixed points, at $`(\mathrm{\Omega }_\mathrm{M},\mathrm{\Omega }_\mathrm{\Lambda })`$ equal to $`(0,0)`$, $`(0,1)`$, and $`(1,0)`$. The attractor among these at $`(1,0)`$ is known as de Sitter space — a universe with no matter density, dominated by a cosmological constant, and with scale factor growing exponentially with time. The fact that this point is an attractor on the diagram is another way of understanding the cosmological constant problem. A universe with initial conditions located at a generic point on the diagram will, after several expansion times, flow to de Sitter space if it began above the recollapse line, and flow to infinity and back to recollapse if it began below that line. Since our universe has undergone a large number of $`e`$-folds of expansion since early times, it must have begun at a non-generic point in order not to have evolved either to de Sitter space or to a Big Crunch. The only other two fixed points on the diagram are the saddle point at $`(\mathrm{\Omega }_\mathrm{M},\mathrm{\Omega }_\mathrm{\Lambda })=(0,0)`$, corresponding to an empty universe, and the repulsive fixed point at $`(\mathrm{\Omega }_\mathrm{M},\mathrm{\Omega }_\mathrm{\Lambda })=(1,0)`$, known as the Einstein-de Sitter solution. Since our universe is not empty, the favored solution from this combination of theoretical and empirical arguments is the Einstein-de Sitter universe. The inflationary scenario provides a mechanism whereby the universe can be driven to the line $`\mathrm{\Omega }_\mathrm{M}+\mathrm{\Omega }_\mathrm{\Lambda }=1`$ (spatial flatness), so Einstein-de Sitter is a natural expectation if we imagine that some unknown mechanism sets $`\mathrm{\Lambda }=0`$. As discussed below, the observationally favored universe is located on this line but away from the fixed points, near $`(\mathrm{\Omega }_\mathrm{M},\mathrm{\Omega }_\mathrm{\Lambda })=(0.3,0.7)`$. It is fair to conclude that naturalness arguments have a somewhat spotty track record at predicting cosmological parameters.
### 2.3 Surveying the universe
The lookback time from the present day to an object at redshift $`z_{}`$ is given by
$$\begin{array}{ccc}\hfill t_0t_{}& =& _t_{}^{t_0}𝑑t\hfill \\ & =& _{1/(1+z_{})}^1\frac{da}{aH(a)},\hfill \end{array}$$
(36)
with $`H(a)`$ given by (30). The age of the universe is obtained by taking the $`z_{}\mathrm{}`$ ($`t_{}0`$) limit. For $`\mathrm{\Omega }=\mathrm{\Omega }_\mathrm{M}=1`$, this yields the familiar answer $`t_0=(2/3)H_0^1`$; the age decreases as $`\mathrm{\Omega }_\mathrm{M}`$ is increased, and increases as $`\mathrm{\Omega }_\mathrm{\Lambda }`$ is increased. Figure (2) shows the expansion history of the universe for different values of these parameters and $`H_0`$ fixed; it is clear how the acceleration caused by $`\mathrm{\Omega }_\mathrm{\Lambda }`$ leads to an older universe.
There are analytic approximation formulas which estimate (36) in various regimes , but generally the integral is straightforward to perform numerically.
In a generic curved spacetime, there is no preferred notion of the distance between two objects. Robertson-Walker spacetimes have preferred foliations, so it is possible to define sensible notions of the distance between comoving objects — those whose worldlines are normal to the preferred slices. Placing ourselves at $`r=0`$ in the coordinates defined by (2), the coordinate distance $`r`$ to another comoving object is independent of time. It can be converted to a physical distance at any specified time $`t_{}`$ by multiplying by the scale factor $`R_0a(t_{})`$, yielding a number which will of course change as the universe expands. However, intervals along spacelike slices are not accessible to observation, so it is typically more convenient to use distance measures which can be extracted from observable quantities. These include the luminosity distance,
$$d_L\sqrt{\frac{L}{4\pi F}},$$
(37)
where $`L`$ is the intrinsic luminosity and $`F`$ the measured flux; the proper-motion distance,
$$d_M\frac{u}{\dot{\theta }},$$
(38)
where $`u`$ is the transverse proper velocity and $`\dot{\theta }`$ the observed angular velocity; and the angular-diameter distance,
$$d_A\frac{D}{\theta },$$
(39)
where $`D`$ is the proper size of the object and $`\theta `$ its apparent angular size. All of these definitions reduce to the usual notion of distance in a Euclidean space. In a Robertson-Walker universe, the proper-motion distance turns out to equal the physical distance along a spacelike slice at $`t=t_0`$:
$$d_M=R_0r.$$
(40)
The three measures are related by
$$d_L=(1+z)d_M=(1+z)^2d_A,$$
(41)
so any one can be converted to any other for sources of known redshift.
The proper-motion distance between sources at redshift $`z_1`$ and $`z_2`$ can be computed by using $`ds^2=0`$ along a light ray, where $`ds^2`$ is given by (2). We have
$$\begin{array}{ccc}\hfill d_M(z_1,z_2)& =& R_0(r_2r_1)\hfill \\ & =& R_0\mathrm{sinn}\left[_{t_1}^{t_2}\frac{dt}{R_0a(t)}\right]\hfill \\ & =& \frac{1}{H_0\sqrt{|\mathrm{\Omega }_{k0}|}}\mathrm{sinn}\left[H_0\sqrt{|\mathrm{\Omega }_{k0}|}_{1/(1+z_1)}^{1/(1+z_2)}\frac{da}{a^2H(a)}\right],\hfill \end{array}$$
(42)
where we have used (5) to solve for $`R_0=1/(H_0\sqrt{|\mathrm{\Omega }_{k0}|})`$, $`H(a)`$ is again given by (30), and “sinn($`x`$)” denotes $`\mathrm{sinh}(x)`$ when $`\mathrm{\Omega }_{k0}<0`$, $`\mathrm{sin}(x)`$ when $`\mathrm{\Omega }_{k0}>0`$, and $`x`$ when $`\mathrm{\Omega }_{k0}=0`$. An analytic approximation formula can be found in . Note that, for large redshifts, the dependence of the various distance measures on $`z`$ is not necessarily monotonic.
The comoving volume element in a Robertson-Walker universe is given by
$$dV=\frac{R_0^3r^2}{\sqrt{1kr^2}}drd\mathrm{\Omega },$$
(43)
which can be integrated analytically to obtain the volume out to a distance $`d_M`$:
$$\begin{array}{ccc}\hfill V(d_M)& =& \frac{1}{2H_0^3\mathrm{\Omega }_{k0}}[H_0d_M\sqrt{1+H_0^2\mathrm{\Omega }_{k0}d_M^2}\hfill \\ & & \frac{1}{\sqrt{|\mathrm{\Omega }_{k0}|}}\mathrm{sinn}^1(H_0\sqrt{|\mathrm{\Omega }_{k0}|}d_M)],\hfill \end{array}$$
(44)
where “sinn” is defined as below (42).
### 2.4 Structure formation
The introduction of a cosmological constant changes the relationship between the matter density and expansion rate from what it would be in a matter-dominated universe, which in turn influences the growth of large-scale structure. The effect is similar to that of a nonzero spatial curvature, and complicated by hydrodynamic and nonlinear effects on small scales, but is potentially detectable through sufficiently careful observations.
The analysis of the evolution of structure is greatly abetted by the fact that perturbations start out very small (temperature anisotropies in the microwave background imply that the density perturbations were of order $`10^5`$ at recombination), and linearized theory is effective. In this regime, the fate of the fluctuations is in the hands of two competing effects: the tendency of self-gravity to make overdense regions collapse, and the tendency of test particles in the background expansion to move apart. Essentially, the effect of vacuum energy is to contribute to expansion but not to the self-gravity of overdensities, thereby acting to suppress the growth of perturbations .
For sub-Hubble-radius perturbations in a cold dark matter component, a Newtonian analysis suffices. (We may of course be interested in super-Hubble-radius modes, or the evolution of interacting or relativistic particles, but the simple Newtonian case serves to illustrate the relevant physical effect.) If the energy density in dynamical matter is dominated by CDM, the linearized Newtonian evolution equation is
$$\ddot{\delta }_\mathrm{M}+2\frac{\dot{a}}{a}\dot{\delta }_\mathrm{M}=4\pi G\rho _\mathrm{M}\delta _\mathrm{M}.$$
(45)
The second term represents an effective frictional force due to the expansion of the universe, characterized by a timescale $`(\dot{a}/a)^1=H^1`$, while the right hand side is a forcing term with characteristic timescale $`(4\pi G\rho _\mathrm{M})^{1/2}\mathrm{\Omega }_\mathrm{M}^{1/2}H^1`$. Thus, when $`\mathrm{\Omega }_\mathrm{M}1`$, these effects are in balance and CDM perturbations gradually grow; when $`\mathrm{\Omega }_\mathrm{M}`$ dips appreciably below unity (as when curvature or vacuum energy begin to dominate), the friction term becomes more important and perturbation growth effectively ends. In fact (45) can be directly solved to yield
$$\delta _\mathrm{M}(a)=\frac{5}{2}H_0^2\mathrm{\Omega }_{\mathrm{M0}}\frac{\dot{a}}{a}_0^aH^3(a^{})𝑑a^{},$$
(46)
where $`H(a)`$ is given by (30). There exist analytic approximations to this formula , as well as analytic expressions for flat universes . Note that this analysis is consistent only in the linear regime; once perturbations on a given scale become of order unity, they break away from the Hubble flow and begin to evolve as isolated systems.
## 3 Observational tests
It has been suspected for some time now that there are good reasons to think that a cosmology with an appreciable cosmological constant is the best fit to what we know about the universe . However, it is only very recently that the observational case has tightened up considerably, to the extent that, as the year 2000 dawns, more experts than not believe that there really is a positive vacuum energy exerting a measurable effect on the evolution of the universe. In this section I review the major approaches which have led to this shift.
### 3.1 Type Ia supernovae
The most direct and theory-independent way to measure the cosmological constant would be to actually determine the value of the scale factor as a function of time. Unfortunately, the appearance of $`\mathrm{\Omega }_k`$ in formulae such as (42) renders this difficult. Nevertheless, with sufficiently precise information about the dependence of a distance measure on redshift we can disentangle the effects of spatial curvature, matter, and vacuum energy, and methods along these lines have been popular ways to try to constrain the cosmological constant.
Astronomers measure distance in terms of the “distance modulus” $`mM`$, where $`m`$ is the apparent magnitude of the source and $`M`$ its absolute magnitude. The distance modulus is related to the luminosity distance via
$$mM=5\mathrm{log}_{10}[d_L(\mathrm{Mpc})]+25.$$
(47)
Of course, it is easy to measure the apparent magnitude, but notoriously difficult to infer the absolute magnitude of a distant object. Methods to estimate the relative absolute luminosities of various kinds of objects (such as galaxies with certain characteristics) have been pursued, but most have been plagued by unknown evolutionary effects or simply large random errors .
Recently, significant progress has been made by using Type Ia supernovae as “standardizable candles”. Supernovae are rare — perhaps a few per century in a Milky-Way-sized galaxy — but modern telescopes allow observers to probe very deeply into small regions of the sky, covering a very large number of galaxies in a single observing run. Supernovae are also bright, and Type Ia’s in particular all seem to be of nearly uniform intrinsic luminosity (absolute magnitude $`M19.5`$, typically comparable to the brightness of the entire host galaxy in which they appear) . They can therefore be detected at high redshifts ($`z1`$), allowing in principle a good handle on cosmological effects .
The fact that all SNe Ia are of similar intrinsic luminosities fits well with our understanding of these events as explosions which occur when a white dwarf, onto which mass is gradually accreting from a companion star, crosses the Chandrasekhar limit and explodes. (It should be noted that our understanding of supernova explosions is in a state of development, and theoretical models are not yet able to accurately reproduce all of the important features of the observed events. See for some recent work.) The Chandrasekhar limit is a nearly-universal quantity, so it is not a surprise that the resulting explosions are of nearly-constant luminosity. However, there is still a scatter of approximately $`40\%`$ in the peak brightness observed in nearby supernovae, which can presumably be traced to differences in the composition of the white dwarf atmospheres. Even if we could collect enough data that statistical errors could be reduced to a minimum, the existence of such an uncertainty would cast doubt on any attempts to study cosmology using SNe Ia as standard candles.
Fortunately, the observed differences in peak luminosities of SNe Ia are very closely correlated with observed differences in the shapes of their light curves: dimmer SNe decline more rapidly after maximum brightness, while brighter SNe decline more slowly . There is thus a one-parameter family of events, and measuring the behavior of the light curve along with the apparent luminosity allows us to largely correct for the intrinsic differences in brightness, reducing the scatter from $`40\%`$ to less than $`15\%`$ — sufficient precision to distinguish between cosmological models. (It seems likely that the single parameter can be traced to the amount of <sup>56</sup>Ni produced in the supernova explosion; more nickel implies both a higher peak luminosity and a higher temperature and thus opacity, leading to a slower decline. It would be an exaggeration, however, to claim that this behavior is well-understood theoretically.)
Following pioneering work reported in , two independent groups have undertaken searches for distant supernovae in order to measure cosmological parameters. Figure (3) shows the results for $`mM`$ vs. $`z`$ for the High-Z Supernova Team , and Figure (4) shows the equivalent results for the Supernova Cosmology Project . Under the assumption that the energy density of the universe is dominated by matter and vacuum components, these data can be converted into limits on $`\mathrm{\Omega }_\mathrm{M}`$ and $`\mathrm{\Omega }_\mathrm{\Lambda }`$, as shown in Figures (5) and (6).
It is clear that the confidence intervals in the $`\mathrm{\Omega }_\mathrm{M}`$-$`\mathrm{\Omega }_\mathrm{\Lambda }`$ plane are consistent for the two groups, with somewhat tighter constraints obtained by the Supernova Cosmology Project, who have more data points. The surprising result is that both teams favor a positive cosmological constant, and strongly rule out the traditional $`(\mathrm{\Omega }_\mathrm{M},\mathrm{\Omega }_\mathrm{\Lambda })=(1,0)`$ favorite universe. They are even inconsistent with an open universe with zero cosmological constant, given what we know about the matter density of the universe (see below).
Given the significance of these results, it is natural to ask what level of confidence we should have in them. There are a number of potential sources of systematic error which have been considered by the two teams; see the original papers for a thorough discussion. The two most worrisome possibilities are intrinsic differences between Type Ia supernovae at high and low redshifts , and possible extinction via intergalactic dust . (There is also the fact that intervening weak lensing can change the distance-magnitude relation, but this seems to be a small effect in realistic universes .) Both effects have been carefully considered, and are thought to be unimportant, although a better understanding will be necessary to draw firm conclusions. Here, I will briefly mention some of the relevant issues.
As thermonuclear explosions of white dwarfs, Type Ia supernovae can occur in a wide variety of environments. Consequently, a simple argument against evolution is that the high-redshift environments, while chronologically younger, should be a subset of all possible low-redshift environments, which include regions that are “young” in terms of chemical and stellar evolution. Nevertheless, even a small amount of evolution could ruin our ability to reliably constrain cosmological parameters . In their original papers , the supernova teams found impressive consistency in the spectral and photometric properties of Type Ia supernovae over a variety of redshifts and environments (e.g., in elliptical vs. spiral galaxies). More recently, however, Riess et al. have presented tentative evidence for a systematic difference in the properties of high- and low-redshift supernovae, claiming that the risetimes (from initial explosion to maximum brightness) were higher in the high-redshift events. Apart from the issue of whether the existing data support this finding, it is not immediately clear whether such a difference is relevant to the distance determinations: first, because the risetime is not used in determining the absolute luminosity at peak brightness, and second, because a process which only affects the very early stages of the light curve is most plausibly traced to differences in the outer layers of the progenitor, which may have a negligible affect on the total energy output. Nevertheless, any indication of evolution could bring into question the fundamental assumptions behind the entire program. It is therefore essential to improve the quality of both the data and the theories so that these issues may be decisively settled.
Other than evolution, obscuration by dust is the leading concern about the reliability of the supernova results. Ordinary astrophysical dust does not obscure equally at all wavelengths, but scatters blue light preferentially, leading to the well-known phenomenon of “reddening”. Spectral measurements by the two supernova teams reveal a negligible amount of reddening, implying that any hypothetical dust must be a novel “grey” variety. This possibility has been investigated by a number of authors . These studies have found that even grey dust is highly constrained by observations: first, it is likely to be intergalactic rather than within galaxies, or it would lead to additional dispersion in the magnitudes of the supernovae; and second, intergalactic dust would absorb ultraviolet/optical radiation and re-emit it at far infrared wavelengths, leading to stringent constraints from observations of the cosmological far-infrared background. Thus, while the possibility of obscuration has not been entirely eliminated, it requires a novel kind of dust which is already highly constrained (and may be convincingly ruled out by further observations).
According to the best of our current understanding, then, the supernova results indicating an accelerating universe seem likely to be trustworthy. Needless to say, however, the possibility of a heretofore neglected systematic effect looms menacingly over these studies. Future experiments, including a proposed satellite dedicated to supernova cosmology , will both help us improve our understanding of the physics of supernovae and allow a determination of the distance/redshift relation to sufficient precision to distinguish between the effects of a cosmological constant and those of more mundane astrophysical phenomena. In the meantime, it is important to obtain independent corroboration using other methods.
### 3.2 Cosmic microwave background
The discovery by the COBE satellite of temperature anisotropies in the cosmic microwave background inaugurated a new era in the determination of cosmological parameters. To characterize the temperature fluctuations on the sky, we may decompose them into spherical harmonics,
$$\frac{\mathrm{\Delta }T}{T}=\underset{lm}{}a_{lm}Y_{lm}(\theta ,\varphi ),$$
(48)
and express the amount of anisotropy at multipole moment $`l`$ via the power spectrum,
$$C_l=|a_{lm}|^2.$$
(49)
Higher multipoles correspond to smaller angular separations on the sky, $`\theta =180^{}/l`$. Within any given family of models, $`C_l`$ vs. $`l`$ will depend on the parameters specifying the particular cosmology. Although the case is far from closed, evidence has been mounting in favor of a specific class of models — those based on Gaussian, adiabatic, nearly scale-free perturbations in a universe composed of baryons, radiation, and cold dark matter. (The inflationary universe scenario typically predicts these kinds of perturbations.)
Although the dependence of the $`C_l`$’s on the parameters can be intricate, nature has chosen not to test the patience of cosmologists, as one of the easiest features to measure — the location in $`l`$ of the first “Doppler peak”, an increase in power due to acoustic oscillations — provides one of the most direct handles on the cosmic energy density, one of the most interesting parameters. The first peak (the one at lowest $`l`$) corresponds to the angular scale subtended by the Hubble radius $`H_{\mathrm{CMB}}^1`$ at the time when the CMB was formed (known variously as “decoupling” or “recombination” or “last scattering”) . The angular scale at which we observe this peak is tied to the geometry of the universe: in a negatively (positively) curved universe, photon paths diverge (converge), leading to a larger (smaller) apparent angular size as compared to a flat universe. Since the scale $`H_{\mathrm{CMB}}^1`$ is set mostly by microphysics, this geometrical effect is dominant, and we can relate the spatial curvature as characterized by $`\mathrm{\Omega }`$ to the observed peak in the CMB spectrum via
$$l_{\mathrm{peak}}220\mathrm{\Omega }^{1/2}.$$
(50)
More details about the spectrum (height of the peak, features of the secondary peaks) will depend on other cosmological quantities, such as the Hubble constant and the baryon density .
Figure 7 shows a summary of data as of 1998, with various experimental results consolidated into bins, along with two theoretical models. Since that time, the data have continued to accumulate (see for example ), and the near future should see a wealth of new results of ever-increasing precision. It is clear from the figure that there is good evidence for a peak at approximately $`l_{\mathrm{peak}}200`$, as predicted in a spatially-flat universe. This result can be made more quantitative by fitting the CMB data to models with different values of $`\mathrm{\Omega }_\mathrm{M}`$ and $`\mathrm{\Omega }_\mathrm{\Lambda }`$ , or by combining the CMB data with other sources, such as supernovae or large-scale structure . Figure 8 shows the constraints from the CMB in the $`\mathrm{\Omega }_\mathrm{M}`$-$`\mathrm{\Omega }_\mathrm{\Lambda }`$ plane, using data from the 1997 test flight of the BOOMERANG experiment . (Although the data used to make this plot are essentially independent of those shown in the previous figure, the constraints obtained are nearly the same.) It is clear that the CMB data provide constraints which are complementary to those obtained using supernovae; the two approaches yield confidence contours which are nearly orthogonal in the $`\mathrm{\Omega }_\mathrm{M}`$-$`\mathrm{\Omega }_\mathrm{\Lambda }`$ plane. The region of overlap is in the vicinity of $`(\mathrm{\Omega }_\mathrm{M},\mathrm{\Omega }_\mathrm{\Lambda })=(0.3,0.7)`$, which we will see below is also consistent with other determinations.
### 3.3 Matter density
Many cosmological tests, such as the two just discussed, will constrain some combination of $`\mathrm{\Omega }_\mathrm{M}`$ and $`\mathrm{\Omega }_\mathrm{\Lambda }`$. It is therefore useful to consider tests of $`\mathrm{\Omega }_\mathrm{M}`$ alone, even if our primary goal is to determine $`\mathrm{\Omega }_\mathrm{\Lambda }`$. (In truth, it is also hard to constrain $`\mathrm{\Omega }_\mathrm{M}`$ alone, as almost all methods actually constrain some combination of $`\mathrm{\Omega }_\mathrm{M}`$ and the Hubble constant $`h=H_0/(100`$ km/sec/Mpc); the HST Key Project on the extragalactic distance scale finds $`h=0.71\pm 0.06`$ , which is consistent with other methods , and what I will assume below.)
For years, determinations of $`\mathrm{\Omega }_\mathrm{M}`$ based on dynamics of galaxies and clusters have yielded values between approximately $`0.1`$ and $`0.4`$ — noticeably larger than the density parameter in baryons as inferred from primordial nucleosynthesis, $`\mathrm{\Omega }_\mathrm{B}=(0.019\pm 0.001)h^20.04`$ , but noticeably smaller than the critical density. The last several years have witnessed a number of new methods being brought to bear on the question; the quantitative results have remained unchanged, but our confidence in them has increased greatly.
A thorough discussion of determinations of $`\mathrm{\Omega }_\mathrm{M}`$ requires a review all its own, and good ones are available . Here I will just sketch some of the important methods.
The traditional method to estimate the mass density of the universe is to “weigh” a cluster of galaxies, divide by its luminosity, and extrapolate the result to the universe as a whole. Although clusters are not representative samples of the universe, they are sufficiently large that such a procedure has a chance of working. Studies applying the virial theorem to cluster dynamics have typically obtained values $`\mathrm{\Omega }_\mathrm{M}=0.2\pm 0.1`$ . Although it is possible that the global value of $`M/L`$ differs appreciably from its value in clusters, extrapolations from small scales do not seem to reach the critical density . New techniques to weigh the clusters, including gravitational lensing of background galaxies and temperature profiles of the X-ray gas , while not yet in perfect agreement with each other, reach essentially similar conclusions.
Rather than measuring the mass relative to the luminosity density, which may be different inside and outside clusters, we can also measure it with respect to the baryon density , which is very likely to have the same value in clusters as elsewhere in the universe, simply because there is no way to segregate the baryons from the dark matter on such large scales. Most of the baryonic mass is in the hot intracluster gas , and the fraction $`f_{\mathrm{gas}}`$ of total mass in this form can be measured either by direct observation of X-rays from the gas or by distortions of the microwave background by scattering off hot electrons (the Sunyaev-Zeldovich effect) , typically yielding $`0.1f_{\mathrm{gas}}0.2`$. Since primordial nucleosynthesis provides a determination of $`\mathrm{\Omega }_\mathrm{B}0.04`$, these measurements imply
$$\mathrm{\Omega }_\mathrm{M}=\mathrm{\Omega }_\mathrm{B}/f_{\mathrm{gas}}=0.3\pm 0.1,$$
(51)
consistent with the value determined from mass to light ratios.
Another handle on the density parameter in matter comes from properties of clusters at high redshift. The very existence of massive clusters has been used to argue in favor of $`\mathrm{\Omega }_\mathrm{M}0.2`$ , and the lack of appreciable evolution of clusters from high redshifts to the present provides additional evidence that $`\mathrm{\Omega }_\mathrm{M}<1.0`$.
The story of large-scale motions is more ambiguous. The peculiar velocities of galaxies are sensitive to the underlying mass density, and thus to $`\mathrm{\Omega }_\mathrm{M}`$, but also to the “bias” describing the relative amplitude of fluctuations in galaxies and mass . Difficulties both in measuring the flows and in disentangling the mass density from other effects make it difficult to draw conclusions at this point, and at present it is hard to say much more than $`0.2\mathrm{\Omega }_\mathrm{M}1.0`$.
Finally, the matter density parameter can be extracted from measurements of the power spectrum of density fluctuations (see for example ). As with the CMB, predicting the power spectrum requires both an assumption of the correct theory and a specification of a number of cosmological parameters. In simple models (e.g., with only cold dark matter and baryons, no massive neutrinos), the spectrum can be fit (once the amplitude is normalized) by a single “shape parameter”, which is found to be equal to $`\mathrm{\Gamma }=\mathrm{\Omega }_\mathrm{M}h`$. (For more complicated models see .) Observations then yield $`\mathrm{\Gamma }0.25`$, or $`\mathrm{\Omega }_\mathrm{M}0.36`$. For a more careful comparison between models and observations, see .
Thus, we have a remarkable convergence on values for the density parameter in matter:
$$0.1\mathrm{\Omega }_\mathrm{M}0.4.$$
(52)
Even without the supernova results, this determination in concert with the CMB measurements favoring a flat universe provide a strong case for a nonzero cosmological constant.
### 3.4 Gravitational lensing
The volume of space back to a specified redshift, given by (44), depends sensitively on $`\mathrm{\Omega }_\mathrm{\Lambda }`$. Consequently, counting the apparent density of observed objects, whose actual density per cubic Mpc is assumed to be known, provides a potential test for the cosmological constant . Like tests of distance vs. redshift, a significant problem for such methods is the luminosity evolution of whatever objects one might attempt to count. A modern attempt to circumvent this difficulty is to use the statistics of gravitational lensing of distant galaxies; the hope is that the number of condensed objects which can act as lenses is less sensitive to evolution than the number of visible objects.
In a spatially flat universe, the probability of a source at redshift $`z_s`$ being lensed, relative to the fiducial ($`\mathrm{\Omega }_\mathrm{M}=1`$, $`\mathrm{\Omega }_\mathrm{\Lambda }=0`$) case, is given by
$$P_{\mathrm{lens}}=\frac{15}{4}\left[1(1+z_s)^{1/2}\right]^3_1^{a_s}\frac{H_0}{H(a)}\left[\frac{d_A(0,a)d_A(a,a_s)}{d_A(0,a_s)}\right]𝑑a,$$
(53)
where $`a_s=1/(1+z_s)`$.
As shown in Figure (9), the probability rises dramatically as $`\mathrm{\Omega }_\mathrm{\Lambda }`$ is increased to unity as we keep $`\mathrm{\Omega }`$ fixed. Thus, the absence of a large number of such lenses would imply an upper limit on $`\mathrm{\Omega }_\mathrm{\Lambda }`$.
Analysis of lensing statistics is complicated by uncertainties in evolution, extinction, and biases in the lens discovery procedure. It has been argued that the existing data allow us to place an upper limit of $`\mathrm{\Omega }_\mathrm{\Lambda }<0.7`$ in a flat universe. However, other groups have claimed that the current data actually favor a nonzero cosmological constant. The near future will bring larger, more objective surveys, which should allow these ambiguities to be resolved. Other manifestations of lensing can also be used to constrain $`\mathrm{\Omega }_\mathrm{\Lambda }`$, including statistics of giant arcs , deep weak-lensing surveys , and lensing in the Hubble Deep Field .
### 3.5 Other tests
There is a tremendous variety of ways in which a nonzero cosmological constant can manifest itself in observable phenomena. Here is an incomplete list of additional possibilities; see also .
* Observations of numbers of objects vs. redshift are a potentially sensitive test of cosmological parameters if evolutionary effects can be brought under control. Although it is hard to account for the luminosity evolution of galaxies, it may be possible to indirectly count dark halos by taking into account the rotation speeds of visible galaxies, and upcoming redshift surveys could be used to constrain the volume/redshift relation .
* Alcock and Paczynski showed that the relationship between the apparent transverse and radial sizes of an object of cosmological size depends on the expansion history of the universe. Clusters of galaxies would be possible candidates for such a measurement, but they are insufficiently isotropic; alternatives, however, have been proposed, using for example the quasar correlation function as determined from redshift surveys , or the Lyman-$`\alpha `$ forest .
* In a related effect, the dynamics of large-scale structure can be affected by a nonzero cosmological constant; if a protocluster, for example, is anisotropic, it can begin to contract along a minor axis while the universe is matter-dominated and along its major axis while the universe is vacuum-dominated. Although small, such effects may be observable in individual clusters or in redshift surveys .
* A different version of the distance-redshift test uses extended lobes of radio galaxies as modified standard yardsticks. Current observations disfavor universes with $`\mathrm{\Omega }_\mathrm{M}`$ near unity (, and references therein).
* Inspiralling compact binaries at cosmological distances are potential sources of gravitational waves. It turns out that the redshift distribution of events is sensitive to the cosmological constant; although speculative, it has been proposed that advanced LIGO detectors could use this effect to provide measurements of $`\mathrm{\Omega }_\mathrm{\Lambda }`$ .
* Finally, consistency of the age of the universe and the ages of its oldest constituents is a classic test of the expansion history. If stars were sufficiently old and $`H_0`$ and $`\mathrm{\Omega }_\mathrm{M}`$ were sufficiently high, a positive $`\mathrm{\Omega }_\mathrm{\Lambda }`$ would be necessary to reconcile the two, and this situation has occasionally been thought to hold. Measurements of geometric parallax to nearby stars from the Hipparcos satellite have, at the least, called into question previous determinations of the ages of the oldest globular clusters, which are now thought to be perhaps 12 billion rather than 15 billion years old (see the discussion in ). It is therefore unclear whether the age issue forces a cosmological constant upon us, but by now it seems forced upon us for other reasons.
## 4 Physics issues
In Section (1.3) we discussed the large difference between the magnitude of the vacuum energy expected from zero-point fluctuations and scalar potentials, $`\rho _\mathrm{\Lambda }^{\mathrm{theor}}2\times 10^{110}\mathrm{erg}/\mathrm{cm}^3`$, and the value we apparently observe, $`\rho _\mathrm{\Lambda }^{(\mathrm{obs})}2\times 10^{10}\mathrm{erg}/\mathrm{cm}^3`$ (which may be thought of as an upper limit, if we wish to be careful). It is somewhat unfair to characterize this discrepancy as a factor of $`10^{120}`$, since energy density can be expressed as a mass scale to the fourth power. Writing $`\rho _\mathrm{\Lambda }=M_{\mathrm{vac}}^4`$, we find $`M_{\mathrm{vac}}^{(\mathrm{theory})}M_{\mathrm{Pl}}10^{18}`$ GeV and $`M_{\mathrm{vac}}^{(\mathrm{obs})}10^3`$ eV, so a more fair characterization of the problem would be
$$\frac{M_{\mathrm{vac}}^{(\mathrm{theory})}}{M_{\mathrm{vac}}^{(\mathrm{obs})}}10^{30}.$$
(54)
Of course, thirty orders of magnitude still constitutes a difference worthy of our attention.
Although the mechanism which suppresses the naive value of the vacuum energy is unknown, it seems easier to imagine a hypothetical scenario which makes it exactly zero than one which sets it to just the right value to be observable today. (Keeping in mind that it is the zero-temperature, late-time vacuum energy which we want to be small; it is expected to change at phase transitions, and a large value in the early universe is a necessary component of inflationary universe scenarios .) If the recent observations pointing toward a cosmological constant of astrophysically relevant magnitude are confirmed, we will be faced with the challenge of explaining not only why the vacuum energy is smaller than expected, but also why it has the specific nonzero value it does.
### 4.1 Supersymmetry
Although initially investigated for other reasons, supersymmetry (SUSY) turns out to have a significant impact on the cosmological constant problem, and may even be said to solve it halfway. SUSY is a spacetime symmetry relating fermions and bosons to each other. Just as ordinary symmetries are associated with conserved charges, supersymmetry is associated with “supercharges” $`Q_\alpha `$, where $`\alpha `$ is a spinor index (for introductions see ). As with ordinary symmetries, a theory may be supersymmetric even though a given state is not supersymmetric; a state which is annihilated by the supercharges, $`Q_\alpha |\psi =0`$, preserves supersymmetry, while states with $`Q_\alpha |\psi 0`$ are said to spontaneously break SUSY.
Let’s begin by considering “globally supersymmetric” theories, which are defined in flat spacetime (obviously an inadequate setting in which to discuss the cosmological constant, but we have to start somewhere). Unlike most non-gravitational field theories, in supersymmetry the total energy of a state has an absolute meaning; the Hamiltonian is related to the supercharges in a straightforward way:
$$H=\underset{\alpha }{}\{Q_\alpha ,Q_\alpha ^{}\},$$
(55)
where braces represent the anticommutator. Thus, in a completely supersymmetric state (in which $`Q_\alpha |\psi =0`$ for all $`\alpha `$), the energy vanishes automatically, $`\psi |H|\psi =0`$ . More concretely, in a given supersymmetric theory we can explicitly calculate the contributions to the energy from vacuum fluctuations and from the scalar potential $`V`$. In the case of vacuum fluctuations, contributions from bosons are exactly canceled by equal and opposite contributions from fermions when supersymmetry is unbroken. Meanwhile, the scalar-field potential in supersymmetric theories takes on a special form; scalar fields $`\varphi ^i`$ must be complex (to match the degrees of freedom of the fermions), and the potential is derived from a function called the superpotential $`W(\varphi ^i)`$ which is necessarily holomorphic (written in terms of $`\varphi ^i`$ and not its complex conjugate $`\overline{\varphi }^i`$). In the simple Wess-Zumino models of spin-0 and spin-1/2 fields, for example, the scalar potential is given by
$$V(\varphi ^i,\overline{\varphi }^j)=\underset{i}{}|_iW|^2,$$
(56)
where $`_iW=W/\varphi ^i`$. In such a theory, one can show that SUSY will be unbroken only for values of $`\varphi ^i`$ such that $`_iW=0`$, implying $`V(\varphi ^i,\overline{\varphi }^j)=0`$.
So the vacuum energy of a supersymmetric state in a globally supersymmetric theory will vanish. This represents rather less progress than it might appear at first sight, since: 1.) Supersymmetric states manifest a degeneracy in the mass spectrum of bosons and fermions, a feature not apparent in the observed world; and 2.) The above results imply that non-supersymmetric states have a positive-definite vacuum energy. Indeed, in a state where SUSY was broken at an energy scale $`M_{\mathrm{SUSY}}`$, we would expect a corresponding vacuum energy $`\rho _\mathrm{\Lambda }M_{\mathrm{SUSY}}^4`$. In the real world, the fact that accelerator experiments have not discovered superpartners for the known particles of the Standard Model implies that $`M_{\mathrm{SUSY}}`$ is of order $`10^3`$ GeV or higher. Thus, we are left with a discrepancy
$$\frac{M_{\mathrm{SUSY}}}{M_{\mathrm{vac}}}10^{15}.$$
(57)
Comparison of this discrepancy with the naive discrepancy (54) is the source of the claim that SUSY can solve the cosmological constant problem halfway (at least on a log scale).
As mentioned, however, this analysis is strictly valid only in flat space. In curved spacetime, the global transformations of ordinary supersymmetry are promoted to the position-dependent (gauge) transformations of supergravity. In this context the Hamiltonian and supersymmetry generators play different roles than in flat spacetime, but it is still possible to express the vacuum energy in terms of a scalar field potential $`V(\varphi ^i,\overline{\varphi }^j)`$. In supergravity $`V`$ depends not only on the superpotential $`W(\varphi ^i)`$, but also on a “Kähler potential” $`K(\varphi ^i,\overline{\varphi }^j)`$, and the Kähler metric $`K_{i\overline{ȷ}}`$ constructed from the Kähler potential by $`K_{i\overline{ȷ}}=^2K/\varphi ^i\overline{\varphi }^j`$. (The basic role of the Kähler metric is to define the kinetic term for the scalars, which takes the form $`g^{\mu \nu }K_{i\overline{ȷ}}_\mu \varphi ^i_\nu \overline{\varphi }^j`$.) The scalar potential is
$$V(\varphi ^i,\overline{\varphi }^j)=e^{K/M_{\mathrm{Pl}}^2}\left[K^{i\overline{ȷ}}(D_iW)(D_{\overline{ȷ}}\overline{W})3M_{\mathrm{Pl}}^2|W|^2\right],$$
(58)
where $`D_iW`$ is the Kähler derivative,
$$D_iW=_iW+M_{\mathrm{Pl}}^2(_iK)W.$$
(59)
Note that, if we take the canonical Kähler metric $`K_{i\overline{ȷ}}=\delta _{i\overline{ȷ}}`$, in the limit $`M_{\mathrm{Pl}}\mathrm{}`$ ($`G0`$) the first term in square brackets reduces to the flat-space result (56). But with gravity, in addition to the non-negative first term we find a second term providing a non-positive contribution. Supersymmetry is unbroken when $`D_iW=0`$; the effective cosmological constant is thus non-positive. We are therefore free to imagine a scenario in which supersymmetry is broken in exactly the right way, such that the two terms in parentheses cancel to fantastic accuracy, but only at the cost of an unexplained fine-tuning (see for example ). At the same time, supergravity is not by itself a renormalizable quantum theory, and therefore it may not be reasonable to hope that a solution can be found purely within this context.
### 4.2 String theory
Unlike supergravity, string theory appears to be a consistent and well-defined theory of quantum gravity, and therefore calculating the value of the cosmological constant should, at least in principle, be possible. On the other hand, the number of vacuum states seems to be quite large, and none of them (to the best of our current knowledge) features three large spatial dimensions, broken supersymmetry, and a small cosmological constant. At the same time, there are reasons to believe that any realistic vacuum of string theory must be strongly coupled ; therefore, our inability to find an appropriate solution may simply be due to the technical difficulty of the problem. (For general introductions to string theory, see ; for cosmological issues, see ).
String theory is naturally formulated in more than four spacetime dimensions. Studies of duality symmetries have revealed that what used to be thought of as five distinct ten-dimensional superstring theories — Type I, Types IIA and IIB, and heterotic theories based on gauge groups E(8)$`\times `$E(8) and SO(32) — are, along with eleven-dimensional supergravity, different low-energy weak-coupling limits of a single underlying theory, sometimes known as M-theory. In each of these six cases, the solution with the maximum number of uncompactified, flat spacetime dimensions is a stable vacuum preserving all of the supersymmetry. To bring the theory closer to the world we observe, the extra dimensions can be compactified on a manifold whose Ricci tensor vanishes. There are a large number of possible compactifications, many of which preserve some but not all of the original supersymmetry. If enough SUSY is preserved, the vacuum energy will remain zero; generically there will be a manifold of such states, known as the moduli space.
Of course, to describe our world we want to break all of the supersymmetry. Investigations in contexts where this can be done in a controlled way have found that the induced cosmological constant vanishes at the classical level, but a substantial vacuum energy is typically induced by quantum corrections . Moore has suggested that Atkin-Lehner symmetry, which relates strong and weak coupling on the string worldsheet, can enforce the vanishing of the one-loop quantum contribution in certain models (see also ); generically, however, there would still be an appreciable contribution at two loops.
Thus, the search is still on for a four-dimensional string theory vacuum with broken supersymmetry and vanishing (or very small) cosmological constant. (See for a general discussion of the vacuum problem in string theory.) The difficulty of achieving this in conventional models has inspired a number of more speculative proposals, which I briefly list here.
* In three spacetime dimensions supersymmetry can remain unbroken, maintaining a zero cosmological constant, in such a way as to break the mass degeneracy between bosons and fermions . This mechanism relies crucially on special properties of spacetime in (2+1) dimensions, but in string theory it sometimes happens that the strong-coupling limit of one theory is another theory in one higher dimension .
* More generally, it is now understood that (at least in some circumstances) string theory obeys the “holographic principle”, the idea that a theory with gravity in $`D`$ dimensions is equivalent to a theory without gravity in $`D1`$ dimensions . In a holographic theory, the number of degrees of freedom in a region grows as the area of its boundary, rather than as its volume. Therefore, the conventional computation of the cosmological constant due to vacuum fluctuations conceivably involves a vast overcounting of degrees of freedom. We might imagine that a more correct counting would yield a much smaller estimate of the vacuum energy , although no reliable calculation has been done as yet.
* The absence of manifest SUSY in our world leads us to ask whether the beneficial aspect of canceling contributions to the vacuum energy could be achieved even without a truly supersymmetric theory. Kachru, Kumar and Silverstein have constructed such a string theory, and argue that the perturbative contributions to the cosmological constant should vanish (although the actual calculations are somewhat delicate, and not everyone agrees ). If such a model could be made to work, it is possible that small non-perturbative effects could generate a cosmological constant of an astrophysically plausible magnitude .
* A novel approach to compactification starts by imagining that the fields of the Standard Model are confined to a (3+1)-dimensional manifold (or “brane”, in string theory parlance) embedded in a larger space. While gravity is harder to confine to a brane, phenomenologically acceptable scenarios can be constructed if either the extra dimensions are any size less than a millimeter , or if there is significant spacetime curvature in a non-compact extra dimension . Although these scenarios do not offer a simple solution to the cosmological constant problem, the relationship between the vacuum energy and the expansion rate can differ from our conventional expectation (see for example ), and one is free to imagine that further study may lead to a solution in this context (see for example ).
Of course, string theory might not be the correct description of nature, or its current formulation might not be directly relevant to the cosmological constant problem. For example, a solution may be provided by loop quantum gravity , or by a composite graviton . It is probably safe to believe that a significant advance in our understanding of fundamental physics will be required before we can demonstrate the existence of a vacuum state with the desired properties. (Not to mention the equally important question of why our world is based on such a state, rather than one of the highly supersymmetric states that appear to be perfectly good vacua of string theory.)
### 4.3 The anthropic principle
The anthropic principle is essentially the idea that some of the parameters characterizing the universe we observe may not be determined directly by the fundamental laws of physics, but also by the truism that intelligent observers will only ever experience conditions which allow for the existence of intelligent observers. Many professional cosmologists view this principle in much the same way as many traditional literary critics view deconstruction — as somehow simultaneously empty of content and capable of working great evil. Anthropic arguments are easy to misuse, and can be invoked as a way out of doing the hard work of understanding the real reasons behind why we observe the universe we do. Furthermore, a sense of disappointment would inevitably accompany the realization that there were limits to our ability to unambiguously and directly explain the observed universe from first principles. It is nevertheless possible that some features of our world have at best an anthropic explanation, and the value of the cosmological constant is perhaps the most likely candidate.
In order for the tautology that “observers will only observe conditions which allow for observers” to have any force, it is necessary for there to be alternative conditions — parts of the universe, either in space, time, or branches of the wavefunction — where things are different. In such a case, our local conditions arise as some combination of the relative abundance of different environments and the likelihood that such environments would give rise to intelligence. Clearly, the current state of the art doesn’t allow us to characterize the full set of conditions in the entire universe with any confidence, but modern theories of inflation and quantum cosmology do at least allow for the possibility of widely disparate parts of the universe in which the “constants of nature” take on very different values (for recent examples see ). We are therefore faced with the task of estimating quantitatively the likelihood of observing any specific value of $`\mathrm{\Lambda }`$ within such a scenario.
The most straightforward anthropic constraint on the vacuum energy is that it must not be so high that galaxies never form . From the discussion in Section (2.4), we know that overdense regions do not collapse once the cosmological constant begins to dominate the universe; if this happens before the epoch of galaxy formation, the universe will be devoid of galaxies, and thus of stars and planets, and thus (presumably) of intelligent life. The condition that $`\mathrm{\Omega }_\mathrm{\Lambda }(z_{\mathrm{gal}})\mathrm{\Omega }_\mathrm{M}(z_{\mathrm{gal}})`$ implies
$$\frac{\mathrm{\Omega }_{\mathrm{\Lambda }0}}{\mathrm{\Omega }_{\mathrm{M0}}}a_{\mathrm{gal}}^3=(1+z_{\mathrm{gal}})^3125,$$
(60)
where we have taken the redshift of formation of the first galaxies to be $`z_{\mathrm{gal}}4`$. Thus, the cosmological constant could be somewhat larger than observation allows and still be consistent with the existence of galaxies. (This estimate, like the ones below, holds parameters such as the amplitude of density fluctuations fixed while allowing $`\mathrm{\Omega }_\mathrm{\Lambda }`$ to vary; depending on one’s model of the universe of possibilities, it may be more defensible to vary a number of parameters at once. See for example .)
However, it is better to ask what is most likely value of $`\mathrm{\Omega }_\mathrm{\Lambda }`$, i.e. what is the value that would be experienced by the largest number of observers ? Since a universe with $`\mathrm{\Omega }_{\mathrm{\Lambda }0}/\mathrm{\Omega }_{\mathrm{M0}}1`$ will have many more galaxies than one with $`\mathrm{\Omega }_{\mathrm{\Lambda }0}/\mathrm{\Omega }_{\mathrm{M0}}100`$, it is quite conceivable that most observers will measure something close to the former value. The probability measure for observing a value of $`\rho _\mathrm{\Lambda }`$ can be decomposed as
$$d𝒫(\rho _\mathrm{\Lambda })=\nu (\rho _\mathrm{\Lambda })𝒫_{}(\rho _\mathrm{\Lambda })d\rho _\mathrm{\Lambda },$$
(61)
where $`𝒫_{}(\rho _\mathrm{\Lambda })d\rho _\mathrm{\Lambda }`$ is the a priori probability measure (whatever that might mean) for $`\rho _\mathrm{\Lambda }`$, and $`\nu (\rho _\mathrm{\Lambda })`$ is the average number of galaxies which form at the specified value of $`\rho _\mathrm{\Lambda }`$. Martel, Shapiro and Weinberg have presented a calculation of $`\nu (\rho _\mathrm{\Lambda })`$ using a spherical-collapse model. They argue that it is natural to take the a priori distribution to be a constant, since the allowed range of $`\rho _\mathrm{\Lambda }`$ is very far from what we would expect from particle-physics scales. Garriga and Vilenkin argue on the basis of quantum cosmology that there can be a significant departure from a constant a priori distribution. However, in either case the conclusion is that an observed $`\mathrm{\Omega }_{\mathrm{\Lambda }0}`$ of the same order of magnitude as $`\mathrm{\Omega }_{\mathrm{M0}}`$ is by no means extremely unlikely (which is probably the best one can hope to say given the uncertainties in the calculation).
Thus, if one is willing to make the leap of faith required to believe that the value of the cosmological constant is chosen from an ensemble of possibilities, it is possible to find an “explanation” for its current value (which, given its unnaturalness from a variety of perspectives, seems otherwise hard to understand). Perhaps the most significant weakness of this point of view is the assumption that there are a continuum of possibilities for the vacuum energy density. Such possibilities correspond to choices of vacuum states with arbitrarily similar energies. If these states were connected to each other, there would be local fluctuations which would appear to us as massless fields, which are not observed (see Section 4.5). If on the other hand the vacua are disconnected, it is hard to understand why all possible values of the vacuum energy are represented, rather than the differences in energies between different vacua being given by some characteristic particle-physics scale such as $`M_{\mathrm{Pl}}`$ or $`M_{\mathrm{SUSY}}`$. (For one scenario featuring discrete vacua with densely spaced energies, see .) It will therefore (again) require advances in our understanding of fundamental physics before an anthropic explanation for the current value of the cosmological constant can be accepted.
### 4.4 Miscellaneous adjustment mechanisms
The importance of the cosmological constant problem has engendered a wide variety of proposed solutions. This section will present only a brief outline of some of the possibilities, along with references to recent work; further discussion and references can be found in .
One approach which has received a great deal of attention is the famous suggestion by Coleman , that effects of virtual wormholes could set the cosmological constant to zero at low energies. The essential idea is that wormholes (thin tubes of spacetime connecting macroscopically large regions) can act to change the effective value of all the observed constants of nature. If we calculate the wave function of the universe by performing a Feynman path integral over all possible spacetime metrics with wormholes, the dominant contribution will be from those configurations whose effective values for the physical constants extremize the action. These turn out to be, under a certain set of assumed properties of Euclidean quantum gravity, configurations with zero cosmological constant at late times. Thus, quantum cosmology predicts that the constants we observe are overwhelmingly likely to take on values which imply a vanishing total vacuum energy. However, subsequent investigations have failed to inspire confidence that the desired properties of Euclidean quantum cosmology are likely to hold, although it is still something of an open question; see discussions in .
Another route one can take is to consider alterations of the classical theory of gravity. The simplest possibility is to consider adding a scalar field to the theory, with dynamics which cause the scalar to evolve to a value for which the net cosmological constant vanishes (see for example ). Weinberg, however, has pointed out on fairly general grounds that such attempts are unlikely to work ; in models proposed to date, either there is no solution for which the effective vacuum energy vanishes, or there is a solution but with other undesirable properties (such as making Newton’s constant $`G`$ also vanish). Rather than adding scalar fields, a related approach is to remove degrees of freedom by making the determinant of the metric, which multiplies $`\mathrm{\Lambda }_0`$ in the action (15), a non-dynamical quantity, or at least changing its dynamics in some way (see for recent examples). While this approach has not led to a believable solution to the cosmological constant problem, it does change the context in which it appears, and may induce different values for the effective vacuum energy in different branches of the wavefunction of the universe.
Along with global supersymmetry, there is one other symmetry which would work to prohibit a cosmological constant: conformal (or scale) invariance, under which the metric is multiplied by a spacetime-dependent function, $`g_{\mu \nu }e^{\lambda (x)}g_{\mu \nu }`$. Like supersymmetry, conformal invariance is not manifest in the Standard Model of particle physics. However, it has been proposed that quantum effects could restore conformal invariance on length scales comparable to the cosmological horizon size, working to cancel the cosmological constant (for some examples see ). At this point it remains unclear whether this suggestion is compatible with a more complete understanding of quantum gravity, or with standard cosmological observations.
A final mechanism to suppress the cosmological constant, related to the previous one, relies on quantum particle production in de Sitter space (analogous to Hawking radiation around black holes). The idea is that the effective energy-momentum tensor of such particles may act to cancel out the bare cosmological constant (for recent attempts see ). There is currently no consensus on whether such an effect is physically observable (see for example ).
If inventing a theory in which the vacuum energy vanishes is difficult, finding a model that predicts a vacuum energy which is small but not quite zero is all that much harder. Along these lines, there are various numerological games one can play. For example, the fact that supersymmetry solves the problem halfway could be suggestive; a theory in which the effective vacuum energy scale was given not by $`M_{\mathrm{SUSY}}10^3`$ GeV but by $`M_{\mathrm{SUSY}}^2/M_{\mathrm{Pl}}10^3`$ eV would seem to fit the observations very well. The challenging part of this program, of course, is to devise such a theory. Alternatively, one could imagine that we live in a “false vacuum” — that the absolute minimum of the vacuum energy is truly zero, but we live in a state which is only a local minimum of the energy. Scenarios along these lines have been explored ; the major hurdle to be overcome is explaining why the energy difference between the true and false vacua is so much smaller than one would expect.
### 4.5 Other sources of dark energy
Although a cosmological constant is an excellent fit to the current data, the observations can also be accommodated by any form of “dark energy” which does not cluster on small scales (so as to avoid being detected by measurements of $`\mathrm{\Omega }_\mathrm{M}`$) and redshifts away only very slowly as the universe expands \[to account for the accelerated expansion, as per equation (32)\]. This possibility has been extensively explored of late, and a number of candidates have been put forward.
One way to parameterize such a component $`X`$ is by an effective equation of state, $`p_X=w_X\rho _X`$. (A large number of phenomenological models of this type have been investigated, starting with the early work in ; see for many more references.) The relevant range for $`w_X`$ is between $`0`$ (ordinary matter) and $`1`$ (true cosmological constant); sources with $`w_X>0`$ redshift away more rapidly than ordinary matter (and therefore cause extra deceleration), while $`w_X<1`$ is unphysical by the criteria discussed in Section 2.1 (although see ). While not every source will obey an equation of state with $`w_X=`$ constant, it is often the case that a single effective $`w_X`$ characterizes the behavior for the redshift range over which the component can potentially be observed.
Current observations of supernovae, large-scale structure, gravitational lensing, and the CMB already provide interesting limits on $`w_X`$ , and future data will be able to do much better . Figure (10) shows an example, in this case limits from supernovae and large-scale structure on $`w_X`$ and $`\mathrm{\Omega }_\mathrm{M}`$ in a universe which is assumed to be flat and dominated by $`X`$ and ordinary matter. It is clear that the favored value for the equation-of-state parameter is near $`1`$, that of a true cosmological constant, although other values are not completely ruled out.
The simplest physical model for an appropriate dark energy component is a single slowly-rolling scalar field, sometimes referred to as “quintessence” . In an expanding universe, a spatially homogeneous scalar with potential $`V(\varphi )`$ and minimal coupling to gravity obeys
$$\ddot{\varphi }+3H\dot{\varphi }+V^{}(\varphi )=0,$$
(62)
where $`H`$ is the Hubble parameter, overdots indicate time derivatives, and primes indicate derivatives with respect to $`\varphi `$. This equation is similar to (45), with analogous solutions. The Hubble parameter acts as a friction term; for generic potentials, the field will be overdamped (and thus approximately constant) when $`H>\sqrt{V^{\prime \prime }(\varphi )}`$, and underdamped (and thus free to roll) when $`H<\sqrt{V^{\prime \prime }(\varphi )}`$. The energy density is $`\rho _\varphi =\frac{1}{2}\dot{\varphi }^2+V(\varphi )`$, and the pressure is $`p_\varphi =\frac{1}{2}\dot{\varphi }^2V(\varphi )`$, implying an equation of state parameter
$$w=\frac{p}{\rho }=\frac{\frac{1}{2}\dot{\varphi }^2V(\varphi )}{\frac{1}{2}\dot{\varphi }^2+V(\varphi )},$$
(63)
which will generally vary with time. Thus, when the field is slowly-varying and $`\dot{\varphi }^2<<V(\varphi )`$, we have $`w1`$, and the scalar field potential acts like a cosmological constant.
There are many reasons to consider dynamical dark energy as an alternative to a cosmological constant. First and foremost, it is a logical possibility which might be correct, and can be constrained by observation. Secondly, it is consistent with the hope that the ultimate vacuum energy might actually be zero, and that we simply haven’t relaxed all the way to the vacuum as yet. But most interestingly, one might wonder whether replacing a constant parameter $`\mathrm{\Lambda }`$ with a dynamical field could allow us to relieve some of the burden of fine-tuning that inevitably accompanies the cosmological constant. To date, investigations have focused on scaling or tracker models of quintessence, in which the scalar field energy density can parallel that of matter or radiation, at least for part of its history . (Of course, we do not want the dark energy density to redshift away as rapidly as that in matter during the current epoch, or the universe would not be accelerating.) Tracker models can be constructed in which the vacuum energy density at late times is robust, in the sense that it does not depend sensitively on the initial conditions for the field. However, the ultimate value $`\rho _{\mathrm{vac}}(10^3\mathrm{eV})^4`$ still depends sensitively on the parameters in the potential. Indeed, it is hard to imagine how this could help but be the case; unlike the case of the axion solution to the strong-CP problem, we have no symmetry to appeal to that would enforce a small vacuum energy, much less a particular small nonzero number.
Quintessence models also introduce new naturalness problems in addition to those of a cosmological constant. These can be traced to the fact that, in order for the field to be slowly-rolling today, we require $`\sqrt{V^{\prime \prime }(\varphi _0)}H_0`$; but this expression is the effective mass of fluctuations in $`\varphi `$, so we have
$$m_\varphi H_010^{33}\mathrm{eV}.$$
(64)
By particle-physics standards, this is an incredibly small number; masses of scalar fields tend to be large in the absence of a symmetry to protect them. Scalars of such a low mass give rise to long-range forces if they couple to ordinary matter; since $`\varphi `$ does couple to gravity, we expect at the very least to have non-renormalizable interactions suppressed by powers of the Planck scale. Such interactions are potentially observable, both via fifth-force experiments and searches for time-dependence of the constants of nature, and current limits imply that there must be suppression of the quintessence couplings by several orders of magnitude over what would be expected . The only known way to obtain such a suppression is through the imposition of an approximate global symmetry (which would also help explain the low mass of the field), of the type characteristic of pseudo-Goldstone boson models of quintessence, which have been actively explored . (Cosmological pseudo-Goldstone bosons are potentially detectable through their tendency to rotate polarized radiation from galaxies and the CMB .) See for a discussion of further fine-tuning problems in the context of supersymmetric models.
Nevertheless, these naturalness arguments are by no means airtight, and it is worth considering specific particle-physics models for the quintessence field. In addition to the pseudo-Goldstone boson models just mentioned, these include models based on supersymmetric gauge theories , supergravity , small extra dimensions , large extra dimensions , and non-minimal couplings to the curvature scalar . Finally, the possibility has been raised that the scalar field responsible for driving inflation may also serve as quintessence , although this proposal has been criticized for producing unwanted relics and isocurvature fluctuations .
There are other models of dark energy besides those based on nearly-massless scalar fields. One scenario is “solid” dark matter, typically based on networks of tangled cosmic strings or domain walls . Strings give an effective equation-of-state parameter $`w_{\mathrm{string}}=1/3`$, and walls have $`w_{\mathrm{wall}}=2/3`$, so walls are a better fit to the data at present. There is also the idea of dark matter particles whose masses increase as the universe expands, their energy thus redshifting away more slowly than that of ordinary matter (see also ). The cosmological consequences of this kind of scenario turn out to be difficult to analyze analytically, and work is still ongoing.
## 5 Conclusions: the preposterous universe
Observational evidence from a variety of sources currently points to a universe which is (at least approximately) spatially flat, with $`(\mathrm{\Omega }_\mathrm{M},\mathrm{\Omega }_\mathrm{\Lambda })(0.3,0.7)`$. The nucleosynthesis constraint implies that $`\mathrm{\Omega }_\mathrm{B}0.04`$, so the majority of the matter content must be in an unknown non-baryonic form.
Nobody would have guessed that we live in such a universe. Figure (11) is a plot of $`\mathrm{\Omega }_\mathrm{\Lambda }`$ as a function of the scale factor $`a`$ for this cosmology. At early times, the cosmological constant would have been negligible, while at later times the density of matter will be essentially zero and the universe will be empty. We happen to live in that brief era, cosmologically speaking, when both matter and vacuum are of comparable magnitude. Within the matter component, there are apparently contributions from baryons and from a non-baryonic source, both of which are also comparable (although at least their ratio is independent of time). This scenario staggers under the burden of its unnaturalness, but nevertheless crosses the finish line well ahead of any competitors by agreeing so well with the data.
Apart from confirming (or disproving) this picture, a major challenge to cosmologists and physicists in the years to come will be to understand whether these apparently distasteful aspects of our universe are simply surprising coincidences, or actually reflect a beautiful underlying structure we do not as yet comprehend. If we are fortunate, what appears unnatural at present will serve as a clue to a deeper understanding of fundamental physics.
## Acknowledgments
I wish to thank Greg Anderson, Tom Banks, Robert Caldwell, Gordon Chalmers, Michael Dine, George Field, Peter Garnavich, Jeff Harvey, Gordy Kane, Manoj Kaplinghat, Bob Kirshner, Lloyd Knox, Finn Larsen, Laura Mersini, Ue-Li Pen, Saul Perlmutter, Joe Polchinski, Ted Pyne, Brian Schmidt, and Michael Turner for numerous useful conversations, Patrick Brady, Deryn Fogg and Clifford Johnson for rhetorical encouragement, and Bill Press and Ed Turner for insinuating me into this formerly-disreputable subject.
|
warning/0004/gr-qc0004008.html
|
ar5iv
|
text
|
# Hadamard Regularization
## I Introduction
The Hadamard regularization , based on the concept of finite part (“partie finie”) of a singular function or a divergent integral, plays an important role in several branches of Mathematical Physics (see for reviews). Typically one deals with functions admitting some non-integrable singularities on a discrete set of isolated points located at finite distances from the origin. The regularization consists of assigning by definition a value for the function at the location of one of the singular points, and for the (generally divergent) integral of that function. The definition may not be fully deterministic, as the Hadamard partie finie depends in general on some arbitrary constants. The Hadamard regularization is one among several other possible regularizations .
A motivation for investigating the properties of a regularization comes from the physical problem of the gravitational interaction of compact bodies in general relativity. As it is hopeless to find a sufficiently general exact solution of this problem, we resort to successive post-Newtonian approximations (limit $`c+\mathrm{}`$). Within the post-Newtonian framework, it makes sense to model compact objects like black holes by point-like particles. This is possible at the price of introducing a regularization, in order to cure the divergencies due to the infinite self-field of the point-masses. However, general relativity is a non-linear theory and, if we want to go to high post-Newtonian approximations, involving high non-linear terms, the process of regularization must be carefully defined. In particular, it turns out that, from the third-post-Newtonian approximation (3PN or $`1/c^6`$), the problem becomes complicated enough that a rather sophisticated version of the Hadamard regularization, including a theory of generalized functions, is required. By contrast, a cruder form of the Hadamard regularization, using merely the concept of partie finie of singular functions , is sufficient to treat the problem up to the 2PN order. Furthermore, we know that the answer provided by the Hadamard regularization up to the 2PN order is correct, in the sense that the field of the two bodies matches the inner field generated by two black holes , and the result for the equations of motion can be recovered without need of any regularization from computations valid for extended non-singular objects . Conforted by these observations we systematically investigate in this paper the Hadamard regularization as well as a theory of associated generalized functions, in a form which can be directly applied to the study of the dynamics of two point-like particles at the 3PN order . (We therefore restrict our attention to two singular points; however most of the results of the paper can be generalized to any number of points.) Notice that this problem enjoys a direct relevance to the future gravitational-wave experiments LIGO and VIRGO, which should be able to detect the radiation from black-hole and/or neutron-star binaries which a precision compatible with the 3PN approximation .
Consider the class $``$ of functions on $`^3`$ that are smooth except at two isolated singularities 1 and 2, around which they admit some power-like singular expansions. The Hadamard partie finie $`(F)_1`$ of $`F`$ at the location of singularity 1, as reviewed in Section II, is defined by the average over spatial directions of the finite-part coefficient in the expansion of $`F`$ around 1. On the other hand, the Hadamard partie finie $`\mathrm{Pf}d^3𝐱F`$ of the divergent integral of $`F`$, we will review in Section III, is obtained from the removal to the integral of the divergent part arising when two regularizing volumes surrounding the singularities shrink to zero. Both concepts of partie finie are closely related. Notably, the partie-finie integral of a gradient is equal to the sum of the parties finies (in the former sense) of the surface integrals surrounding the singularities, in the limit of vanishing areas. In Section IV we investivage several alternative expressions of the Hadamard partie finie of integrals, some of them based on a finite part defined by means of an analytic continuation process (see for a relation between partie finie and analytic continuation). In our terminology, we adopt the name “partie finie” for the specific definitions due to Hadamard, and speak of a “finite part” when referring to other definitions, based for instance on analytic continuation. In Section V we focus to the case (important in applications) of the partie finie of a Poisson integral of $`F`$.
To any $`F`$, we associate in Section VI a generalized function, or partie-finie “pseudo-function” $`\mathrm{Pf}F`$, which is a linear form on $``$ defined for any $`G`$ by the duality bracket $`<\mathrm{Pf}F,G>=\mathrm{Pf}d^3𝐱FG`$. When restricted to the set $`𝒟`$ of smooth functions with compact support the pseudo-function $`\mathrm{Pf}F`$ is a distribution in the sense of Schwartz (see also for more details about generalized functions and distributions), i.e. a linear form which is continuous with respect to the Schwartz topology. \[However, we do not attempt here to introduce a topology on $``$; we simply define the set of algebraic and differential rules, needed in applications, that are satisfied by the pseudo-functions on $``$.\] The product of pseudo-functions coincides with the ordinary (“pointwise”) product used in Physics, namely $`\mathrm{Pf}F.\mathrm{Pf}G=\mathrm{Pf}(FG)`$. An important particular case is the pseudo-function $`\mathrm{Pf}\delta _1`$ obtained (in Section VI) from the pseudo-function associated with the Riesz delta-function , and that satisfies $`G`$, $`<\mathrm{Pf}\delta _1,G>=(G)_1`$. The “Dirac pseudo-function” $`\mathrm{Pf}\delta _1`$ plays in the present context the same role as plays the Dirac measure in distribution theory. We introduce also more complicated objects such as $`\mathrm{Pf}(F\delta _1)`$. In Sections VII and VIII we show how to construct a derivative operator on $``$, generalizing for this class of function the standard distributional derivative operator on $`𝒟`$ and satisfying basically the so-called rule of integration by parts, namely $`F,G`$, $`<_i(\mathrm{Pf}F),G>=<_i(\mathrm{Pf}G),F>`$. In addition we require that the derivative reduces to the “ordinary” derivative for functions that are bounded in a neighbourhood of the singular points, and that the rule of commutation of derivatives holds. We find that this derivative operator is uniquely defined modulo a dependence on an arbitrary numerical constant (see Theorem 4 in Section VIII). It represents a natural notion of derivative within the context of Hadamard regularization of the functions in $``$. However, it does not satisfy in general the Leibniz rule for the derivative of a product (in agreement with a theorem of Schwartz ). See Colombeau for a multiplication of distributions and associated distributional derivative satisfying the Leibniz rule. Further, we obtain the rules obeyed by the new derivative operator when acting on pseudo-functions such as $`\mathrm{Pf}(F\delta _1)`$ in Section VII, and we investigate the associated Laplacian operator in Section VIII. Finally, in Section IX, we consider the case of partial derivatives with respect to the singular points 1 and 2, as well as the time derivative when both singular points depend on time (i.e. represent the trajectories of real particules). Within this approach, the latter distributional derivative constitutes an important tool when studying the problem of the gravitational dynamics of point-particles at the 3PN order .
Notation. $``$, $``$, $``$ and $``$ are the usual sets of non-negative integers, integers, real numbers and complex numbers; $`^+`$ is the set of strictly positive real numbers $`s>0`$; $`^3`$ is the usual three-dimensional space endowed with the Euclidean norm $`|𝐱|=(x_1^2+x_2^2+x_3^2)^{1/2}`$; $`C^p(\mathrm{\Omega })`$ is the set of $`p`$-times continuously differentiable functions on the open set $`\mathrm{\Omega }`$ ($`p+\mathrm{}`$); $`L_{\mathrm{loc}}^1(\mathrm{\Omega })`$ is the set of locally integrable functions on $`\mathrm{\Omega }`$; the $`o`$ and $`O`$ symbols for remainders have their standard meaning; distances between the field point $`𝐱`$ and the source points $`𝐲_1`$ and $`𝐲_2`$ are denoted by $`r_1=|𝐱𝐲_1|`$ and $`r_2=|𝐱𝐲_2|`$; unit directions are $`𝐧_1=(𝐱𝐲_1)/r_1`$ and $`𝐧_2=(𝐱𝐲_2)/r_2`$; $`d\mathrm{\Omega }_1`$ and $`d\mathrm{\Omega }_2`$ are the solid angle elements associated with $`𝐧_1`$ and $`𝐧_2`$; $`r_{12}=|𝐲_1𝐲_2|`$; $`_1(s)`$ and $`_2(s)`$ denote the closed spherical balls of radius $`s`$ centered on $`𝐲_1`$ and $`𝐲_2`$; $`_i=/x^i`$, $`{}_{1}{}^{}_{i}^{}=/y_1^i`$, $`{}_{2}{}^{}_{i}^{}=/y_2^i`$; $`L=i_1i_2\mathrm{}i_l`$ is a multi-index with length $`l`$; $`n_1^L=n_1^{i_1}\mathrm{}n_1^{i_l}`$ and $`_L=_{i_1}\mathrm{}_{i_l}`$; the symmetric-trace-free (STF) projection is denoted by $`\widehat{n}_1^L=\mathrm{STF}(n_1^L)`$; $`(ij)=\frac{ij+ji}{2}`$ and $`[ij]=\frac{ijji}{2}`$; $`12`$ means the same expression but corresponding to the point 2; for clearer reading, we use left-side labels 1 and 2 when the quantity appears within the text, like for the partial derivatives $`{}_{1}{}^{}_{i}^{}`$ and $`{}_{2}{}^{}_{i}^{}`$ or the coefficients $`{}_{1}{}^{}f_{a}^{}`$ and $`{}_{2}{}^{}f_{b}^{}`$, and labels placed underneath the quantity when it appears in an equation; iff means if and only if.
## II Hadamard partie finie
### A A class of singular functions
All over this paper we consider the class of functions of a “field” point $`𝐱^3`$ that are singular at the location of two “source” points $`𝐲_1`$ and $`𝐲_2`$ around which they admit some singular expansions.
###### Definition 1
A real function $`F(𝐱)`$ on $`^3`$ is said to belong to the class of functions $``$ iff
(i) $`F`$ is smooth on $`^3`$ deprived from $`𝐲_1`$ and $`𝐲_2`$, i.e. $`FC^{\mathrm{}}(^3\{𝐲_1,𝐲_2\})`$.
(ii) There exists an ordered family of indices $`(a_i)_i`$ with $`a_i`$, and a family of coefficients $`\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}_{a_i}`$, such that
$$N,F(𝐱)=\underset{i=0}{\overset{i_N}{}}r_1^{a_i}\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{a_i}{}^{}(𝐧_1)+\stackrel{}{R}\genfrac{}{}{0pt}{}{}{1}{}_{N}{}^{}(𝐱).$$
(1)
Here $`r_1=|𝐱𝐲_1|`$ and $`𝐧_1=(𝐱𝐲_1)/r_1`$; $`i_N`$ satisfies $`a_0<a_1<\mathrm{}<a_{i_N}N<a_{i_N+1}`$; and the “remainder” is
$$\stackrel{}{R}\genfrac{}{}{0pt}{}{}{1}{}_{N}{}^{}(𝐱)=o(r_1^N)\text{when }r_10.$$
(2)
(iii) Idem with indices $`(b_i)_i`$, coefficients $`\stackrel{}{f}\genfrac{}{}{0pt}{}{}{2}_{b_i}`$, remainder $`{}_{2}{}^{}R_{N}^{}`$, $`r_1r_2`$ and $`𝐧_1𝐧_2`$.
In addition to Definition 1, we always assume that the functions $`F`$ decrease sufficiently fast at infinity (when $`|𝐱|\mathrm{}`$) so that all integrals we meet are convergent at infinity. Thus, when discussing the integral $`d^3𝐱F`$, we suppose implicitly that $`F=o(|𝐱|^3)`$ at infinity \[or sometimes $`F=O(|𝐱|^{3ϵ})`$ where $`ϵ>0`$\], so that the possible divergencies come only from the bounds at the singular points $`𝐲_{1,2}`$. Similarly, when considering the integral $`d^3𝐱FG`$, we suppose $`FG=o(|𝐱|^3)`$, but for instance we allow $`F`$ to blow up at infinity, say $`F=O(|𝐱|)`$, if we know that $`G`$ decreases rapidly, e.g. $`G=o(|𝐱|^4)`$; in the case of $`d^3𝐱_iF`$, we generally assume $`F=o(|𝐱|^2)`$. \[Clearly, from Definition 1 the ordinary product $`FG`$ of two functions of $``$ is again a function of $``$; and similarly the ordinary gradient $`_iF`$.\]
An important assumption in Definition 1 is that the powers of $`r_1`$ in the expansion of $`F`$ when $`r_10`$ (and similarly when $`r_20`$) are bounded from below, i.e. $`a_0a_i`$ where the most “divergent” power of $`r_1`$, which clearly depends on $`F`$, is $`a_0=a_0(F)`$. Thus the part of the expansion which diverges when $`r_10`$ is composed of a finite number of terms. Notice also that we have excluded in Definition 1 the possible appearance of logarithms of $`r_1`$ (or $`r_2`$) in the expansion of $`F`$. See Sellier for a more general study in the case where some arbitrary powers of logarithms are present. We will discuss the occurence of logarithms in Section V, when dealing with the Poisson integral of $`F`$. At last, we point out that the coefficients $`{}_{1}{}^{}f_{a}^{}`$ (and similarly $`{}_{2}{}^{}f_{b}^{}`$) do not depend only on $`𝐧_1`$, but also they do on the source points $`𝐲_1`$ and $`𝐲_2`$, so that in principle we should write $`{}_{1}{}^{}f_{a}^{}(𝐧_1;𝐲_1,𝐲_2)`$; however, for simplicity’s sake we omit writing the dependence on the source points. The coefficients could also depend on other variables such as the velocities $`𝐯_1`$ and $`𝐯_2`$ of the source points, but the velocities do not participate to the process of regularization and can be ignored for the moment (we will return to this question in Section IX when considering the time dependence of $`F`$).
Once the class $``$ has been defined, we shall often write in this paper the expansions of $`F`$ when $`r_{1,2}0`$ in the simplified forms
$`F(𝐱)`$ $`=`$ $`{\displaystyle \underset{a_0aN}{}}r_1^a\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{a}{}^{}(𝐧_1)+o(r_1^N)\text{when }r_10,`$ (4)
$`F(𝐱)`$ $`=`$ $`{\displaystyle \underset{b_0bN}{}}r_2^b\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{2}}{}_{b}{}^{}(𝐧_2)+o(r_2^N)\text{when }r_20,`$ (5)
by which we really mean the expansions in Definition 1, i.e. in particular where the indices $`a(a_i)_i`$ and $`b(b_i)_i`$, and are a priori real. However, most of the time (in applications), it is sufficient to assume that the powers of $`r_{1,2}`$ are relative integers $`a,b`$. We can then write the expansion $`r_10`$ in the form
$$F=\underset{k=0}{\overset{k_0}{}}\frac{1}{r_1^{1+k}}\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{1k}{}^{}+\underset{k=0}{\overset{N}{}}r_1^k\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{k}{}^{}+o(r_1^N),$$
(6)
where $`k_0=1a_0`$. In the following we shall sometimes derive the results in the simpler case where the powers $``$, being always undertood that the generalization to the case of real powers is straightforward. Finally, it is worth noting that the assumption (i) in Definition 1, that $`F`$ is $`C^{\mathrm{}}`$ outside $`\{𝐲_1,𝐲_2\}`$, can often be relaxed to allow some functions to have integrable singularities. An example is the function $`𝐱1/|𝐱𝐱^{}|`$ encountered in Section V, depending on a fixed “spectator” point $`𝐱^{}`$ distinct from $`𝐲_1`$ and $`𝐲_2`$. To treat such objects, we introduce a larger class of functions, $`_{\mathrm{loc}}`$.
###### Definition 2
$`F(𝐱)`$ is said to belong to the class of functions $`_{\mathrm{loc}}`$ iff
(i’) $`F`$ is locally integrable on $`^3`$ deprived from $`𝐲_1`$ and $`𝐲_2`$, i.e. $`FL_{\mathrm{loc}}^1(^3\{𝐲_1,𝐲_2\})`$.
(ii)-(iii) in Definition 1 hold.
For simplicity, in the following, we shall derive most of the results for functions belonging to the class $``$ (even if the generalization to $`_{\mathrm{loc}}`$ is trivial); $`_{\mathrm{loc}}`$ will be employed only occasionally.
### B Partie finie of a singular function
The first notion of Hadamard partie finie is that of a singular function at the very location of one of its singular points.
###### Definition 3
Given $`F`$ we define the Hadamard partie finie of $`F`$ at the point $`𝐲_1`$ to be
$$(F)_1=\frac{d\mathrm{\Omega }_1}{4\pi }\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{0}{}^{}(𝐧_1),$$
(7)
where $`d\mathrm{\Omega }_1=d\mathrm{\Omega }(𝐧_1)`$ denotes the solid angle element of origin $`𝐲_1`$ and direction $`𝐧_1`$.
In words, the partie finie of $`F`$ at point 1 is defined by the angular average, with respect to the unit direction $`𝐧_1`$, of the coefficient of the zeroth power of $`r_1`$ in the expansion of $`F`$ near 1 (and similarly for the point 2). There is a non zero partie finie only if the family of indices $`(a_i)_i`$ in Definition 1 contains the value 0, i.e. $`i_0`$ such that $`a_{i_0}=0`$. The latter definition applied to the product $`FG`$ of two functions in $``$ yields
$$(FG)_1=\underset{a_0(F)aa_0(G)}{}\frac{d\mathrm{\Omega }_1}{4\pi }\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{a}{}^{}\stackrel{}{g}\genfrac{}{}{0pt}{}{}{1}{}_{a}{}^{},$$
(8)
where $`{}_{1}{}^{}f_{a}^{}`$ and $`{}_{1}{}^{}g_{a}^{}`$ are the coefficients in the expansions of $`F`$ and $`G`$ when $`r_10`$ (the summation over $`a`$ is always finite). From (8) it is clear that the Hadamard partie finie is not “distributive” with respect to the multiplication, in the sense that
$$(FG)_1(F)_1(G)_1\text{in general}.$$
(9)
The partie finie picks up the angular average of $`{}_{1}{}^{}f_{0}^{}(𝐧_1)`$, namely the scalar or $`l=0`$ piece in the spherical-harmonics expansion ($`Y_{lm}`$), or, equivalently, in the expansion on the basis of symmetric and trace-free (STF) products of unit vectors $`𝐧_1=(n_1^i)`$. For any $`l`$, we denote by $`L=i_1i_2\mathrm{}i_l`$ a multi-index composed of $`l`$ indices, and similarly $`L1=i_1i_2\mathrm{}i_{l1}`$, $`P=j_1j_2\mathrm{}j_p`$. In general we do not need to specify the carrier index $`i`$ or $`j`$, so a tensor with $`l`$ upper indices is denoted $`T^L`$, and for instance the scalar formed by contraction with another tensor $`U^L`$ of the same type is written as $`S=T^LU^L=T^{i_1\mathrm{}i_l}U^{i_1\mathrm{}i_l}`$, where we omit writing the $`l`$ summations over the $`l`$ indices $`i_k=1,2,3`$. We denote a product of $`l`$ components of the unit vector $`n_1^i`$ by $`n_1^L=n_1^{i_1}\mathrm{}n_1^{i_l}`$, and the STF projection of that product by $`\widehat{n}_1^L\mathrm{STF}(n_1^L)`$: e.g., $`\widehat{n}_1^{ij}=n_1^in_1^j\frac{1}{3}\delta ^{ij}`$, $`\widehat{n}_1^{ijk}=n_1^in_1^jn_1^k\frac{1}{5}(n_1^i\delta ^{jk}+n_1^j\delta ^{ki}+n_1^k\delta ^{ij}`$). More generally, we denote by $`\widehat{T}^L`$ the STF projection of $`T^L`$; that is, $`\widehat{T}^L`$ is symmetric, and satisfies $`\delta _{i_{l1}i_l}\widehat{T}^{i_{l1}i_lL2}=0`$ (see and the appendix A of for a compendium of formulas using the STF formalism). The coefficients $`{}_{1}{}^{}f_{a}^{}`$ of the expansion of $`F`$ admit the STF decomposition
$$\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{a}{}^{}(𝐧_1)=\underset{l=0}{\overset{+\mathrm{}}{}}n_1^L\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{a}{}^{L},$$
(10)
where the $`{}_{1}{}^{}\widehat{f}_{a}^{L}`$’s are constant STF tensors, given by the inverse formula:
$$\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{a}{}^{L}=\frac{(2l+1)!!}{l!}\frac{d\mathrm{\Omega }_1}{4\pi }\widehat{n}_1^L\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{a}{}^{}(𝐧_1).$$
(11)
In STF notation, the Hadamard partie finie of $`F`$ at 1 reads simply
$$(F)_1=\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{0}{}^{},$$
(12)
where $`{}_{1}{}^{}\widehat{f}_{a}^{}`$ denotes the first term in the expansion (10).
###### Lemma 1
The partie finie at 1 of the gradient $`_iF`$ (as defined outside the singularities) of any function $`F`$ satisfies
$$(_iF)_1=3\left(\frac{n_1^i}{r_1}F\right)_1.$$
(13)
This Lemma is particularly useful as it permits replacing systematically the differential operator $`_i`$ by the algebraic one $`3\frac{n_1^i}{r_1}`$ when working under the partie-finie sign $`(..)_1`$.
Proof. The expansion when $`r_10`$ of the gradient is readily obtained from the expansion of $`F`$ itself as
$$_iF=\underset{a}{}r_1^{a1}[an_1^i\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{a}{}^{}+d_1^i\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{a}{}^{}],$$
(14)
(with over-simplified notation for the sum), where the operator $`d_1^i`$ is defined as $`r_1_i`$ when applied on a function of the sole unit vector $`𝐧_1`$. Hence, explicitly, $`d_1^i=(\delta ^{ij}n_1^{ij})\frac{}{n_1^j}`$. This operator is evidently transverse to $`𝐧_1`$ : $`n_1^id_1^i=0`$, and we get, from the decomposition (10),
$$d_1^i\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{a}{}^{}=\underset{l=0}{\overset{+\mathrm{}}{}}l(n_1^{L1}\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{a}{}^{iL1}n_1^{iL}\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{a}{}^{L}).$$
(15)
Thus, by averaging over angles,
$$\frac{d\mathrm{\Omega }_1}{4\pi }d_1^i\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{a}{}^{}=\frac{2}{3}\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{a}{}^{i}=2\frac{d\mathrm{\Omega }_1}{4\pi }n_1^i\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{a}{}^{}.$$
(16)
We readily deduce that the partie finie of the gradient (14) is given by
$$(_iF)_1=3\frac{d\mathrm{\Omega }_1}{4\pi }n_1^i\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{1}{}^{}=\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{1}{}^{i}\text{(}\text{QED}\text{).}$$
(17)
As an example of application of Lemma 1, we can write, using an operation by parts, $`(r_1^3_iF)_1=[_i(r_1^3F)_i(r_1^3)F]_1=[3n_1^ir_1^2F_i(r_1^3)F]_1`$, from which it follows that
$$(r_1^3_iF)_1=0.$$
(18)
Another consequence of Lemma 1, resulting from two operations by parts, is $`(r_1^2\mathrm{\Delta }F)_1=[3n_1^ir_1_iF_i(r_1^2)_iF]_1=(n_1^ir_1_iF)_1=[3F_i(n_1^ir_1)F]_1`$ (where the Laplacian $`\mathrm{\Delta }=_i_i`$), hence the identity
$$(r_1^2\mathrm{\Delta }F)_1=0.$$
(19)
By the same method we obtain also
$$(_{ij}F)_1=\left(\frac{15n_1^{ij}3\delta ^{ij}}{r_1^2}F\right)_1=2(\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{2}{}^{ij}+\delta ^{ij}\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{2}{}^{}),$$
(20)
the right-hand side of the last equality being expressed in terms of the STF tensors parametrizing (10). Tracing out the previous formula, we find
$$(\mathrm{\Delta }F)_1=\left(\frac{6}{r_1^2}F\right)_1=6\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{2}{}^{}.$$
(21)
Finally, let us quote the general formula for the partie finie of the $`l`$th derivative $`_LF=_{i_1}\mathrm{}_{i_l}F`$:
$$(_LF)_1=l!\underset{k=0}{\overset{\left[\frac{l}{2}\right]}{}}\delta ^{(2K}\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{l}{}^{L2K)}.$$
(22)
Here, $`\left[\frac{l}{2}\right]`$ denotes the integer part of $`\frac{l}{2}`$, $`\delta ^{2K}`$ is the product of Kronecker symbols $`\delta ^{i_1i_2}\delta ^{i_3i_4}\mathrm{}\delta ^{i_{2k1}i_{2k}}`$, and $`{}_{1}{}^{}\widehat{f}_{l}^{L2K}={}_{1}{}^{}\widehat{f}_{l}^{i_{2k+1}\mathrm{}i_l}`$; the parenthesis around the indices denote the symmetrization. One may define the “regular” part of the function $`F`$ near the singularity 1 as the formal Taylor expansion when $`r_10`$ obtained using (22). Thus,
$$F_1^{\mathrm{reg}}\underset{l=0}{\overset{+\mathrm{}}{}}\frac{1}{l!}r_1^ln_1^L(_LF)_1=\underset{l=0}{\overset{+\mathrm{}}{}}r_1^l\underset{k=0}{\overset{\left[\frac{l}{2}\right]}{}}n_1^{L2K}\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{l}{}^{L2K}.$$
(23)
## III Partie-finie integrals
### A The partie finie of a divergent integral
The second notion of Hadamard partie finie is that of the integral $`d^3𝐱F(𝐱)`$, where $`F`$. This integral is generally divergent because of the presence of the singular points $`𝐲_1`$ and $`𝐲_2`$ (recall that we always assume that the function decreases sufficiently rapidly at infinity so that we never have any divergency coming from the integration bound $`|𝐱|+\mathrm{}`$). Consider first the domain $`^3`$ deprived from two spherical balls $`_1(s)`$ and $`_2(s)`$ of radius $`s`$, centered on the two singularities $`𝐲_1`$, $`𝐲_2`$: $`_1(s)=\{𝐱;r_1s\}`$ and $`_2(s)=\{𝐱;r_2s\}`$. We assume that $`s`$ is small enough, i.e. $`s<\frac{r_{12}}{2}`$ where $`r_{12}=|𝐲_1𝐲_2|`$, so that the two balls do not intersect. For $`s>0`$ the integral over this domain, say $`I(s)=_{^3_1(s)_2(s)}d^3𝐱F`$, is well-defined and generally tends to infinity when $`s0`$. Thanks to the expansions (assumed in Definition 1) of $`F`$ near the singularities, we easily compute the part of $`I(s)`$ that blows up when $`s0`$; we find that this divergent part is given, near each singularity, by a finite sum of strictly negative powers of $`s`$ (a polynomial of $`1/s`$ in general) plus a term involving the logarithm of $`s`$. By subtracting from $`I(s)`$ the corresponding divergent part, we get a term that possesses a finite limit when $`s0`$; the Hadamard partie finie is defined as this limit. Associated with the logarithm of $`s`$, there arises an ambiguity which can be viewed as the freedom in the re-definition of the unit system we employ to measure the length $`s`$. In fact it is convenient to introduce two constant length scales $`s_1`$ and $`s_2`$, one per singularity, in order to a-dimensionalize the logarithms as $`\mathrm{ln}(\frac{s}{s_1})`$ and $`\mathrm{ln}(\frac{s}{s_2})`$.
###### Definition 4
For any $`F`$ integrable in a neighbourhood of $`|𝐱|=+\mathrm{}`$, we define the Hadamard partie finie of the divergent integral $`d^3𝐱F`$ as
$`\mathrm{Pf}_{s_1,s_2}{\displaystyle }d^3𝐱F=\underset{s0}{lim}\{{\displaystyle _{^3_1(s)_2(s)}}d^3𝐱F`$ $`+`$ $`{\displaystyle \underset{a+3<0}{}}{\displaystyle \frac{s^{a+3}}{a+3}}{\displaystyle }d\mathrm{\Omega }_1\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{a}{}^{}+\mathrm{ln}\left({\displaystyle \frac{s}{s_1}}\right){\displaystyle }d\mathrm{\Omega }_1\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}_3`$ (24)
$`+`$ $`12\},`$ (25)
where $`12`$ means the same previous two terms but concerning the singularity 2.
This notion of partie finie can be extended to functions which are locally integrable outside the singularities, i.e. $`F_{\mathrm{loc}}`$ (see Definition 2). In (24) the divergent terms are composed of a sum over $`a`$ such that $`a+3<0`$ as well as a logarithmic term, by which we really mean, using the more detailed notation of Definition 1,
$$\underset{i=0}{\overset{i_l1}{}}\frac{s^{a_i+3}}{a_i+3}d\mathrm{\Omega }_1\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{a_i}{}^{}+\delta _{3,a_{i_l}}\mathrm{ln}\left(\frac{s}{s_1}\right)d\mathrm{\Omega }_1\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{a_{i_l}}{}^{}+12,$$
where $`i_l`$ is such that $`a_0<a_1<\mathrm{}<a_{i_l1}<3a_{i_l}`$ (the sum is always finite); we have introduced a Kronecker symbol $`\delta _{3,a_{i_l}}`$ to recall that the logarithm is present only if the family of indices $`(a_i)_i`$ contains the integer $`3`$ (i.e. $`a_{i_l}=3`$). The divergent terms in (24) can also be expressed by means of the partie finie defined by (7). Indeed, they read
$$4\pi \left[\underset{a+3<0}{}\frac{s^{a+3}}{a+3}\left(\frac{F}{r_1^a}\right)_1+\mathrm{ln}\left(\frac{s}{s_1}\right)\left(r_1^3F\right)_1\right]+12$$
\[coming back to the less detailed notation of (24)\].
The partie-finie integral (24) depends intrinsically on the two arbitrary constants $`s_1`$ and $`s_2`$ introduced above. There is another way to interpret these constants besides the necessity to take into account the dimension of $`s`$, which is discussed by Sellier in . With this point of view we initially define the partie finie using two arbitrarily shaped volumes $`𝒱_1`$ and $`𝒱_2`$ instead of the two spherical balls $`_1`$ and $`_2`$. Consider for instance the two volumes $`𝒱_1=\{𝐱;r_1s\rho _1(𝐧_1)\}`$ and $`𝒱_2=\{𝐱;r_2s\rho _2(𝐧_2)\}`$, where $`s^+`$ measures the size of the volumes and the two functions $`\rho _1`$ and $`\rho _2`$ describe their shape (the balls $`_1`$ and $`_2`$ corresponding simply to $`\rho _1`$ and $`\rho _21`$). Here, we assume for simplicity that the volumes remain isometric to themselves when $`s`$ varies. Then, the partie finie is defined as the limit of the integral over $`^3𝒱_1𝒱_2`$ to which we subtract the corresponding divergent terms when $`s0`$, without adding any normalizing constant to the logarithms. In this way, we find that the alternative definition is equivalent to our definition (24) provided that $`s_1`$ and $`s_2`$ are related to the shapes of the regularizing volumes $`𝒱_1`$ and $`𝒱_2`$ through the formula
$$\mathrm{ln}s_1d\mathrm{\Omega }_1\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{3}{}^{}=d\mathrm{\Omega }_1\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{3}{}^{}\mathrm{ln}\rho _1,$$
(26)
(and similarly for $`s_2`$). The arbitrariness on the two original regularizing volumes is therefore encoded into the two (and only two) constants $`s_1`$ and $`s_2`$. A closely related way to interpret them is linked to the necessity to allow the change of the integration variable $`𝐱`$ in the integral $`d^3𝐱F`$. Such an operation modifies the size and shape of the regularizing volumes, thus the balls $`_1`$ and $`_2`$ are in general transformed into some new volumes $`𝒱_1`$ and $`𝒱_2`$; so, according to the previous argument, the freedom of choosing the integration variable reflects out in the freedom of choosing two arbitrary constants $`s_1`$ and $`s_2`$. (In this paper we shall assume that $`s_1`$ and $`s_2`$ are fixed once and for all.)
An alternative expression of the Hadamard partie finie is often useful because it does not involve the limit $`s0`$, but is written with the help of a finite parameter $`s^{}^+`$. Consider some $`s^{}`$ such that $`0<s<s^{}`$, and next, split the integral over $`^3_1(s)_2(s)`$ into the sum of the integral over $`^3_1(s^{})_2(s^{})`$ and the two integrals over the ring-shaped domains $`_1(s^{})_1(s)`$ and $`_1(s^{})_1(s)`$. If $`s<s^{}1`$ we can substitute respectively into the ring-shaped integrals the expansions of $`F`$ when $`r_10`$ and $`r_20`$ \[see (II A)\]. The terms that are divergent in $`s`$ cancel out, so we can apply the limit $`s0`$ (with fixed $`s^{}`$). This yields the following expression for the partie finie: $`N`$,
$`\mathrm{Pf}_{s_1,s_2}{\displaystyle d^3𝐱F}={\displaystyle _{^3_1(s^{})_2(s^{})}}d^3𝐱F`$ $`+`$ $`{\displaystyle \underset{\genfrac{}{}{0pt}{}{a+3N}{a+30}}{}}{\displaystyle \frac{s_{}^{}{}_{}{}^{a+3}}{a+3}}{\displaystyle }d\mathrm{\Omega }_1\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{a}{}^{}+\mathrm{ln}\left({\displaystyle \frac{s^{}}{s_1}}\right){\displaystyle }d\mathrm{\Omega }_1\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}_3`$ (27)
$`+`$ $`12+o(s_{}^{}{}_{}{}^{N+3}),`$ (28)
which is valid for an arbitrary fixed $`s^{}`$. Of course, up to any given finite order $`N`$ the second member of (27) depends on $`s^{}`$, but in the formal limit $`N+\mathrm{}`$, this dependence disappears and, in fine, the partie finie is independent of $`s^{}`$.
### B Partie-finie integral of a gradient
A fundamental feature of the Hadamard partie finie of a divergent integral is that the integral of a gradient $`_iF`$ is a priori not zero, since the surface integrals surrounding the two singularities become infinite when the surface areas shrink to zero, and may possess a finite part.
###### Theorem 1
For any $`F`$ the partie finie of the gradient of $`F`$ is given by
$$\mathrm{Pf}d^3𝐱_iF=4\pi (n_1^ir_1^2F)_1+12,$$
(29)
where the singular value at point 1 is defined by $`(\text{7})`$.
In the case of a regular function, the result is always zero from the simple fact that the surface areas tend to zero — cf the factor $`r_1^2`$ in the right side of (29). However, for $`F`$, the factor $`r_1^2`$ is in general compensated by a divergent term in the expansion of $`F`$, possibly producing a finite contribution.
Proof. We apply (24) to the case of the gradient $`_iF`$, using the expansion of $`_iF`$ when $`r_10`$ as given by (14). The expression of the divergent terms is simplified with the help of the identity (16), which shows notably that the logarithms and associated constants $`s_{1,2}`$ disappear. This leads to
$$\underset{s0}{lim}\{_{^3_1(s)_2(s)}d^3𝐱_iF+\underset{a+2<0}{}s^{a+2}d\mathrm{\Omega }_1n_1^i\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{a}{}^{}+12\}.$$
(30)
Next, the first term inside the braces is transformed via the Gauss theorem into two surface integrals at $`r_1=s`$ and $`r_2=s`$, where we can replace $`F`$ by the corresponding expansions around $`𝐲_1`$ and $`𝐲_2`$ respectively. We get
$`\underset{s0}{lim}\{{\displaystyle \underset{a}{}}s^{a+2}{\displaystyle }d\mathrm{\Omega }_1n_1^i\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{a}{}^{}+{\displaystyle \underset{a+2<0}{}}s^{a+2}{\displaystyle }d\mathrm{\Omega }_1n_1^i\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{a}{}^{}\}`$ $`=`$ $`{\displaystyle }d\mathrm{\Omega }_1n_1^i\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}_2`$
(and similarly when $`12`$); QED.
From Theorem 1 it results that the correct formula for “integrating by parts” under the sign $`\mathrm{Pf}`$ is
$$\mathrm{Pf}d^3𝐱F_iG=\mathrm{Pf}d^3𝐱G_iF4\pi (n_1^ir_1^2FG)_14\pi (n_2^ir_2^2FG)_2.$$
(31)
Note also that the partie-finie integrals of a double derivative as well as a Laplacian are given by
$`\mathrm{Pf}{\displaystyle d^3𝐱_{ij}F}`$ $`=`$ $`4\pi \left(r_1(\delta ^{ij}2n_1^{ij})F\right)_1+12,`$ (33)
$`\mathrm{Pf}{\displaystyle d^3𝐱\mathrm{\Delta }F}`$ $`=`$ $`4\pi (r_1F)_1+12.`$ (34)
### C Parties finies and the Riesz delta-function
The Riesz delta-function plays an important role in the context of Hadamard parties finies. It is defined for any $`\epsilon ^+`$ by $`{}_{\epsilon }{}^{}\delta (𝐱)=\frac{\epsilon (1\epsilon )}{4\pi }|𝐱|^{\epsilon 3}`$; when $`\epsilon 0`$, it tends, in the usual sense of distribution theory, towards the Dirac measure in three dimensions — i.e. $`lim_{\epsilon 0}{}_{\epsilon }{}^{}\delta =\delta `$, as can be seen from the easily checked property that $`\mathrm{\Delta }(|𝐱|^{\epsilon 1})=4\pi {}_{\epsilon }{}^{}\delta (𝐱)`$. The point for our purpose is that when defined with respect to one of the singularities, the Riesz delta-function belongs to $``$. Thus, let us set, $`\epsilon ^+`$,
$${}_{\epsilon }{}^{}\delta _{1}^{}(𝐱){}_{\epsilon }{}^{}\delta (𝐱𝐲_1)=\frac{\epsilon (1\epsilon )}{4\pi }r_1^{\epsilon 3}$$
(35)
(and idem for 2). Now we can apply to $`{}_{\epsilon }{}^{}\delta _{1}^{}(𝐱)`$ the previous definitions for parties finies. In particular, from Definition 3, we see that $`{}_{\epsilon }{}^{}\delta _{1}^{}`$ has no partie finie at 1 when $`\epsilon `$ is small enough: $`\left({}_{\epsilon }{}^{}\delta _{1}^{}\right)_1=0`$. From Definition 4:
###### Lemma 2
For any $`F`$, we have
$$\underset{\epsilon 0}{lim}\mathrm{Pf}d^3𝐱{}_{\epsilon }{}^{}\delta _{1}^{}F=(F)_1,$$
(36)
where the value of $`F`$ at point 1 is given by the prescription $`(\text{7})`$.
Proof. For $`\epsilon >0`$ we evaluate the finite part of the integral for the product $`{}_{\epsilon }{}^{}\delta _{1}^{}F`$ using the specific form (27) of the partie finie defined in terms of a given finite $`s^{}`$. The expansions of $`{}_{\epsilon }{}^{}\delta _{1}^{}F`$ when $`r_{1,2}`$ tend to zero are readily determined to be
$`{}_{\epsilon }{}^{}\delta _{1}^{}F`$ $`=`$ $`{\displaystyle \frac{\epsilon (1\epsilon )}{4\pi }}{\displaystyle \underset{a}{}}r_1^{a+\epsilon 3}\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{a}{}^{}(𝐧_1)\text{for }r_10,`$ (38)
$`{}_{\epsilon }{}^{}\delta _{1}^{}F`$ $`=`$ $`{\displaystyle \frac{\epsilon (1\epsilon )}{4\pi }}{\displaystyle \underset{l0}{}}{\displaystyle \frac{()^l}{l!}}\stackrel{}{}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{L}{}^{}r_{12}^{\epsilon 3}{\displaystyle \underset{b}{}}r_2^{b+l}n_2^L\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{2}}{}_{b}{}^{}(𝐧_2)\text{for }r_20.`$ (39)
In the second equation we used the Taylor expansion $`r_1^{\epsilon 3}=_{l0}\frac{()^l}{l!}r_2^ln_2^L{}_{1}{}^{}_{L}^{}r_{12}^{\epsilon 3}`$ when $`r_20`$, with notation $`n_2^L=n_2^{i_1}\mathrm{}n_2^{i_l}`$ and $`{}_{1}{}^{}_{L}^{}={}_{1}{}^{}_{i_1}^{}\mathrm{}{}_{1}{}^{}_{i_l}^{}`$. Hence, we can write the partie-finie integral in the form ($`N`$; with fixed $`s^{}`$ such that $`0<s^{}<1`$)
$`{\displaystyle _{^3_1(s^{})_2(s^{})}}d^3𝐱{}_{\epsilon }{}^{}\delta _{1}^{}F`$ $`+`$ $`{\displaystyle \frac{\epsilon (1\epsilon )}{4\pi }}{\displaystyle \underset{a+\epsilon N}{}}{\displaystyle \frac{s_{}^{}{}_{}{}^{a+\epsilon }}{a+\epsilon }}{\displaystyle }d\mathrm{\Omega }_1\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}_a`$
$`+`$ $`{\displaystyle \frac{\epsilon (1\epsilon )}{4\pi }}{\displaystyle \underset{l0}{}}{\displaystyle \frac{()^l}{l!}}\stackrel{}{}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{L}{}^{}r_{12}^{\epsilon 3}[{\displaystyle \underset{\genfrac{}{}{0pt}{}{b+l+3N}{\mathrm{and}0}}{}}{\displaystyle \frac{s_{}^{}{}_{}{}^{b+l+3}}{b+l+3}}{\displaystyle }d\mathrm{\Omega }_2n_2^L\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{2}}_b`$
$`+\mathrm{ln}\left({\displaystyle \frac{s^{}}{s_2}}\right){\displaystyle }d\mathrm{\Omega }_2n_2^L\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{2}}{}_{l3}{}^{}]+o(s_{}^{}{}_{}{}^{N}).`$
Here, we have discarded the term with $`\mathrm{ln}\left(\frac{s^{}}{s_1}\right)`$ by choosing $`\epsilon >0`$ to be so small that all denominators $`a+\epsilon `$ differ from zero. Since $`{}_{\epsilon }{}^{}\delta _{1}^{}`$ tends towards the Dirac measure when $`\epsilon 0`$, the integral over $`^3_1(s^{})_2(s^{})`$ goes to zero. Because of the factor $`\epsilon `$ present in the numerators, so do the other terms when $`\epsilon 0`$, except for those whose denominators involve a compensating $`\epsilon `$. Now, the only term having the required property corresponds to $`a=0`$ in the previous expression. Therefore, taking the limit $`\epsilon 0`$ (with fixed $`s^{}`$), we get
$$\underset{\epsilon 0}{lim}\mathrm{Pf}d^3𝐱{}_{\epsilon }{}^{}\delta _{1}^{}F=\frac{d\mathrm{\Omega }_1}{4\pi }\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{0}{}^{}(𝐧_1)+o(s_{}^{}{}_{}{}^{N}),$$
and this being true for any $`N`$, we conclude
$$\underset{\epsilon 0}{lim}\mathrm{Pf}d^3𝐱{}_{\epsilon }{}^{}\delta _{1}^{}F=\frac{d\mathrm{\Omega }_1}{4\pi }\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{0}{}^{}(𝐧_1)=(F)_1\text{(}\text{QED}\text{).}$$
As we can infer from Lemma 2, the Riesz delta-function $`{}_{\epsilon }{}^{}\delta _{1}^{}`$ should constitute in the limit $`\epsilon 0`$ an appropriate extension of the notion of Dirac distribution to the framework of parties finies of singular functions in $``$. The precise definition of a “partie-finie Dirac function” necessitates the introduction of the space of linear forms on $``$ and will be investigated in Section VI (see Definition 7).
## IV Alternative forms of the partie finie
### A Partie finie based on analytic continuation
Practically speaking, the Hadamard partie-finie integral in the form given by (24) is rather difficult to evaluate, because it involves an integration over the complicated volume $`^3_1(s)_2(s)`$. Fortunately, there exist several alternative expressions of the Hadamard partie finie, which are much better suited for practical computations. The first one is based on a double analytic continuation, with two complex parameters $`\alpha `$, $`\beta `$, of the integral
$$I_{\alpha ,\beta }=d^3𝐱\left(\frac{r_1}{s_1}\right)^\alpha \left(\frac{r_2}{s_2}\right)^\beta F,$$
(40)
where the constants $`s_1`$ and $`s_2`$ are the same as those introduced within the definition (24). The point for our purpose is that the integral (40) does range over the complete set $`^3`$. First of all, we propose to check that $`I_{\alpha ,\beta }`$ is defined by analytic continuation in a neighbourhood of the origin $`\alpha =0=\beta `$ in $`^2`$, except at the origin itself where it generically admits a simple pole in $`\alpha `$ or $`\beta `$ or both. We start by splitting $`I_{\alpha ,\beta }`$ into three contribution: $`{}_{1}{}^{}I_{\alpha ,\beta }^{}`$ extending over the ball $`_1(s)`$ of radius $`s`$ surrounding 1, $`{}_{2}{}^{}I_{\alpha ,\beta }^{}`$ extending over the ball $`_2(s)`$ surrounding 2, and $`{}_{3}{}^{}I_{\alpha ,\beta }^{}`$ extending over the rest $`^3_1(s)_2(s)`$. The integral $`{}_{1}{}^{}I_{\alpha ,\beta }^{}`$ is initially convergent for $`\mathrm{}(\alpha )>a_03`$ and any $`\beta `$, where $`a_0`$ is the most singular power of $`r_1`$ in the expansion of $`F`$ near $`𝐲_1`$; similarly, $`{}_{2}{}^{}I_{\alpha ,\beta }^{}`$ exists only if $`\mathrm{}(\beta )>b_03`$ and any $`\alpha `$ ($`b_0`$ is the analogous to $`a_0`$ that relates to $`𝐲_2`$), and $`{}_{3}{}^{}I_{\alpha ,\beta }^{}`$ exists if $`\mathrm{}(\alpha +\beta )<ϵ`$, where $`ϵ>0`$ is such that $`F=O(|𝐱|^{3ϵ})`$ when $`|𝐱|+\mathrm{}`$. As the third contribution $`{}_{3}{}^{}I_{\alpha ,\beta }^{}`$ is clearly defined in a neighbourhood of the origin, including the origin itself, we consider simply the part $`{}_{1}{}^{}I_{\alpha ,\beta }^{}`$ (the same reasoning applies to $`{}_{2}{}^{}I_{\alpha ,\beta }^{}`$). Within the integrand, we replace the product $`r_2^\beta F`$ by its expansion in the neighbourhood of $`𝐲_1`$ (using a Taylor expansion for $`r_2^\beta `$), and find that the dependence on $`\beta `$ occurs through some everywhere well-defined quantity, namely $`{}_{1}{}^{}_{L}^{}r_{12}^\beta `$. After performing the angular integration over $`d\mathrm{\Omega }_1`$, we obtain a remaining radial integral consisting of a sum of terms of the type $`_0^s𝑑r_1r_1^{\alpha +a+l+2}=s^{\alpha +a+l+3}/(\alpha +a+l+3)`$, that clearly admit a unique analytic continuation on $``$; hence our statement (a simple pole at the origin arises when $`a=l3`$).
###### Theorem 2
For any function $`F`$ that is summable at infinity, the Hadamard partie finie of the integral is given by
$$\mathrm{Pf}_{s_1,s_2}d^3𝐱F=\mathrm{FP}_{\genfrac{}{}{0pt}{}{\alpha 0}{\beta 0}}d^3𝐱\left(\frac{r_1}{s_1}\right)^\alpha \left(\frac{r_2}{s_2}\right)^\beta F=\mathrm{FP}_{\genfrac{}{}{0pt}{}{\beta 0}{\alpha 0}}d^3𝐱\left(\frac{r_1}{s_1}\right)^\alpha \left(\frac{r_2}{s_2}\right)^\beta F,$$
(41)
where $`\mathrm{FP}_{\genfrac{}{}{0pt}{}{\alpha 0}{\beta 0}}`$ means taking the finite parts in the Laurent expansions when $`\alpha 0`$ and $`\beta 0`$ successively.
The proof of Theorem 2 is relegated to Appendix A. Notice our convention regarding the notation: while “$`\mathrm{Pf}`$” always stands for the Partie finie of an integral in the specific sense of Hadamard , we refer to “$`\mathrm{FP}`$” as the Finite Part or zeroth-order coefficient in the Laurent expansion with respect to some complex parameter ($`\alpha `$, $`\beta `$, or $`B`$ as in the next subsection). We see from Theorem 2 that the partie finie $`\mathrm{Pf}`$ can be viewed as a finite part $`\mathrm{FP}`$ and vice versa. The link between analytic continuation and Hadamard partie finie is pointed out by Schwartz . More precisely, Theorem 2 says how to calculate the Hadamard partie finie; the procedure consists of: (i) performing the Laurent expansion of $`I_{\alpha ,\beta }`$ when $`\alpha 0`$ while $`\beta `$ remains a fixed (“spectator”) non-zero complex number, i.e.
$$I_{\alpha ,\beta }=\underset{p=p_{\mathrm{min}}}{\overset{+\mathrm{}}{}}\alpha ^pI_{(p),\beta },$$
where $`p`$ and where the coefficients $`I_{(p),\beta }`$ depend on $`\beta `$; (ii) achieving the Laurent expansion of the zeroth-$`\alpha `$-power coefficient $`I_{(0),\beta }`$ when $`\beta 0`$, i.e.
$$I_{(0),\beta }=\underset{q=q_{\mathrm{min}}}{\overset{+\mathrm{}}{}}\beta ^qI_{(0,q)},$$
to finally arrive at the zeroth-$`\beta `$-power coefficient $`I_{(0,0)}`$. Indeed, we find that the same result can be obtained by proceeding the other way around, first expanding around $`\beta =0`$ with a fixed $`\alpha `$, then expanding the coefficients $`I_{\alpha ,(0)}`$ near $`\alpha =0`$. Thus,
$$\stackrel{}{\mathrm{FP}}\genfrac{}{}{0pt}{}{}{\beta 0}\left\{\stackrel{}{\mathrm{FP}}\genfrac{}{}{0pt}{}{}{\alpha 0}I_{\alpha ,\beta }\right\}=I_{(0,0)}=\stackrel{}{\mathrm{FP}}\genfrac{}{}{0pt}{}{}{\alpha 0}\left\{\stackrel{}{\mathrm{FP}}\genfrac{}{}{0pt}{}{}{\beta 0}I_{\alpha ,\beta }\right\}.$$
(42)
We emphasize that the definition (24) of the partie finie yields unambiguously the result $`I_{(0,0)}`$, which corresponds to taking independently the two limits $`\alpha 0`$ and $`\beta 0`$ (the limiting process does not allow for instance to keep $`\alpha =\beta `$). The final value $`I_{(0,0)}`$ is the same as the one given by the regularization adopted by Jaranowski and Schäfer (see their appendix B.2).
In practice the expression (41) is used in connection with the Riesz formula , valid for any $`\gamma `$, $`\delta `$ except at some isolated poles,
$$d^3𝐱r_1^\gamma r_2^\delta =\pi ^{3/2}\frac{\mathrm{\Gamma }\left(\frac{\gamma +3}{2}\right)\mathrm{\Gamma }\left(\frac{\delta +3}{2}\right)\mathrm{\Gamma }\left(\frac{\gamma +\delta +3}{2}\right)}{\mathrm{\Gamma }\left(\frac{\gamma }{2}\right)\mathrm{\Gamma }\left(\frac{\delta }{2}\right)\mathrm{\Gamma }\left(\frac{\gamma +\delta +6}{2}\right)}r_{12}^{\gamma +\delta +3},$$
(43)
with $`r_{12}=|𝐲_1𝐲_2|`$; here, $`\mathrm{\Gamma }`$ denotes the Eulerian function. According to Theorem 2, the formula (43) permits computing the partie finie of any integral of a product between powers of $`r_1`$ and $`r_2`$. Consider the (not so trivial) case of the integral of $`r_1^3r_2^3`$, which is divergent at both points 1 and 2. From the Riesz formula, with $`\gamma =\alpha 3`$ and $`\delta =\beta 3`$, we have
$$I_{\alpha ,\beta }=\pi ^{3/2}\frac{\mathrm{\Gamma }\left(\frac{\alpha }{2}\right)\mathrm{\Gamma }\left(\frac{\beta }{2}\right)\mathrm{\Gamma }\left(\frac{\alpha +\beta 3}{2}\right)}{\mathrm{\Gamma }\left(\frac{\alpha 3}{2}\right)\mathrm{\Gamma }\left(\frac{\beta 3}{2}\right)\mathrm{\Gamma }\left(\frac{\alpha +\beta }{2}\right)}\frac{r_{12}^{\alpha +\beta 3}}{s_1^\alpha s_2^\beta }.$$
We compute the Laurent expansion when $`\alpha 0`$ with fixed $`\beta `$ and obtain a simple pole in $`\alpha `$ followed by a $`\beta `$-dependent finite part given by
$$I_{(0),\beta }=\pi ^{3/2}\frac{\mathrm{\Gamma }(1)}{\mathrm{\Gamma }\left(\frac{3}{2}\right)}\frac{r_{12}^{\beta 3}}{s_2^\beta }\left[\frac{2}{\beta }+\mathrm{\Psi }(1)\mathrm{\Psi }\left(1+\frac{\beta }{2}\right)+\mathrm{\Psi }\left(\frac{3}{2}\right)\mathrm{\Psi }\left(\frac{3}{2}\frac{\beta }{2}\right)+2\mathrm{ln}\left(\frac{r_{12}}{s_1}\right)\right],$$
with $`\mathrm{\Psi }(z)=\frac{d}{dz}\mathrm{ln}\mathrm{\Gamma }(z)`$. This finite part itself includes a simple pole in $`\beta `$, and then we obtain the corresponding finite part when $`\beta 0`$ as
$$I_{(0,0)}=\frac{\pi ^{3/2}}{r_{12}^3}\frac{\mathrm{\Gamma }(1)}{\mathrm{\Gamma }\left(\frac{3}{2}\right)}\left[2\mathrm{ln}\left(\frac{r_{12}}{s_1}\right)+2\mathrm{ln}\left(\frac{r_{12}}{s_2}\right)\right].$$
At last, Theorem 2 tells us that
$$\mathrm{Pf}_{s_1,s_2}\frac{d^3𝐱}{r_1^3r_2^3}=\frac{4\pi }{r_{12}^3}\left[\mathrm{ln}\left(\frac{r_{12}}{s_1}\right)+\mathrm{ln}\left(\frac{r_{12}}{s_2}\right)\right].$$
(44)
Some more complicated integrals will be obtained in the next subsection.
### B Partie finie based on angular integration
The idea is to compute the partie-finie integral by performing an angular integration, followed by the integration over some radial variable. In a first stage, consider an integral that diverges at the point 1, but converges at the point 2. According to (24), we need to compute it over the domain $`^3_1(s)`$; so it is natural to change the integration variable $`𝐱`$ to $`𝐫_1𝐱𝐲_1`$, carry on the angular integration over $`d\mathrm{\Omega }_1=d\mathrm{\Omega }(𝐧_1)`$, and then, the radial integration over $`r_1=|𝐫_1|`$ varying from $`s`$ to infinity, i.e.
$$_{^3_1(s)}d^3𝐱F=_{r_1>s}d^3𝐫_1F=_s^+\mathrm{}𝑑r_1r_1^2𝑑\mathrm{\Omega }_1F.$$
(45)
In the more general case where the integral is simultaneously divergent at the two points 1 and 2, this method stricto sensu is no longer valid since the radial integration in (45) becomes divergent when $`r_1=r_{12}`$. Yet, still it is advantageous to dispose of a mean to change the variable $`𝐱`$ into $`𝐫_1`$ in order to obtain a convenient radial integration (even at the price of breaking the symmetry between the points 1 and 2). We shall derive here two Propositions, based on this idea, whose implementation in practical computations constitutes a very efficient mean to determine the partie finie, without any a priori restriction on the form of integrand as in the application of the Riesz formula (43).
As a matter of fact, in the first proposition, the computation of a partie-finie integral with two singularities 1 and 2 boils down to the computation of a partie-finie integral with singularity 1 and a finite-part integral ($`\mathrm{FP}`$) whose singularity is located at infinity: $`r_1|𝐱𝐲_1|+\mathrm{}`$ (so to speak, the singularity 2 is “rejected” to infinity).
###### Proposition 1
For any function $`F`$ in the class $``$ we can write :
$$\mathrm{Pf}_{s_1,s_2}d^3𝐱F=\mathrm{Pf}_{s_1}\{\stackrel{}{\mathrm{FP}}\genfrac{}{}{0pt}{}{}{B0}d^3𝐫_1\left(\frac{r_1}{s_2}\right)^B[F\underset{b+30}{}r_2^b\stackrel{}{f}\genfrac{}{}{0pt}{}{}{2}{}_{b}{}^{}]\},$$
(46)
where the $`{}_{2}{}^{}f_{b}^{}`$’s denote the coefficients of the expansion of $`F`$ near $`r_2=0`$.
In words, in order to compute the partie finie one can (i) “regularize” $`F`$ around the point 2 by subtracting out from it the terms yielding a divergence at 2, i.e.
$$\stackrel{~}{F}_2F\underset{b+30}{}r_2^b\stackrel{}{f}\genfrac{}{}{0pt}{}{}{2}{}_{b}{}^{},$$
(47)
and (ii) compute the integral of the regularized $`\stackrel{~}{F}_2`$ using the partie finie around 1 and the finite part when $`B0`$ to deal with the divergency at infinity. Notice that the latter divergency has been introduced simply because of the term corresponding to $`b=3`$ in (47) if non-zero. By finite part when $`B0`$ we mean the zeroth-order coefficient in the Laurent expansion of the analytic continuation with respect to the parameter $`B`$. The analytic continuation is straightforwardly defined from the domain of the complex plane $`\mathrm{}(B)>0`$ in which the integral converges at infinity.
Proof. We consider two open domains $`𝒟_1`$ and $`𝒟_2`$ that are supposed to be disjoined, $`𝒟_1𝒟_2=\mathrm{}`$, complementary in $`^3`$, i.e. $`\overline{𝒟_1𝒟_2}=^3`$, and such that $`𝐲_1𝒟_1`$ and $`𝐲_2𝒟_2`$. From Definition 4, the partie-finie integral over $`𝒟_2`$ reads as (for small enough $`s`$)
$$\mathrm{Pf}_{𝒟_2}d^3𝐱F=\underset{s0}{lim}\{_{𝒟_2_2(s)}d^3𝐱F+\underset{b+3<0}{}\frac{s^{b+3}}{b+3}d\mathrm{\Omega }_2\stackrel{}{f}\genfrac{}{}{0pt}{}{}{2}{}_{b}{}^{}+\mathrm{ln}\left(\frac{s}{s_2}\right)d\mathrm{\Omega }_2\stackrel{}{f}\genfrac{}{}{0pt}{}{}{2}{}_{3}{}^{}\}.$$
Now, two short computations reveal that
$`{\displaystyle \underset{b+3<0}{}}{\displaystyle _{^3_2(s)}}d^3𝐱r_2^b\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{2}}_b`$ $`=`$ $`{\displaystyle \underset{b+3<0}{}}{\displaystyle \frac{s^{b+3}}{b+3}}{\displaystyle }d\mathrm{\Omega }_2\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{2}}{}_{b}{}^{},`$ (49)
$`\stackrel{}{\mathrm{FP}}{\displaystyle \genfrac{}{}{0pt}{}{}{B0}}{\displaystyle _{^3_2(s)}}d^3𝐱\left({\displaystyle \frac{|𝐱|}{s_2}}\right)^B{\displaystyle \frac{1}{r_2^3}}\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{2}}_3`$ $`=`$ $`\mathrm{ln}\left({\displaystyle \frac{s}{s_2}}\right){\displaystyle }d\mathrm{\Omega }_2\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{2}}{}_{3}{}^{}.`$ (50)
Furthermore, since the integral appearing in (49) is convergent at infinity, one can add without harm the same finite part operation when $`B0`$ as in (50). Thus, the integral over $`𝒟_2`$ may be re-written as
$`\underset{s0}{lim}\{{\displaystyle _{𝒟_2_2(s)}}d^3𝐱F{\displaystyle \underset{b+30}{}}\stackrel{}{\mathrm{FP}}{\displaystyle \genfrac{}{}{0pt}{}{}{B0}}{\displaystyle _{^3_2(s)}}d^3𝐱\left({\displaystyle \frac{|𝐱|}{s_2}}\right)^Br_2^b\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{2}}{}_{b}{}^{}\}`$
$`=\underset{s0}{lim}\{\stackrel{}{\mathrm{FP}}{\displaystyle \genfrac{}{}{0pt}{}{}{B0}}{\displaystyle _{𝒟_2_2(s)}}d^3𝐱\left({\displaystyle \frac{|𝐱|}{s_2}}\right)^B\stackrel{~}{F}_2{\displaystyle \underset{b+30}{}}\stackrel{}{\mathrm{FP}}{\displaystyle \genfrac{}{}{0pt}{}{}{B0}}{\displaystyle _{𝒟_1}}d^3𝐱\left({\displaystyle \frac{|𝐱|}{s_2}}\right)^Br_2^b\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{2}}{}_{b}{}^{}\}`$
$`=\underset{s0}{lim}\{\stackrel{}{\mathrm{FP}}{\displaystyle \genfrac{}{}{0pt}{}{}{B0}}{\displaystyle _{𝒟_2}}d^3𝐱\left({\displaystyle \frac{|𝐱|}{s_2}}\right)^B\stackrel{~}{F}_2{\displaystyle \underset{b+30}{}}\stackrel{}{\mathrm{FP}}{\displaystyle \genfrac{}{}{0pt}{}{}{B0}}{\displaystyle _{𝒟_1_1(s)}}d^3𝐱\left({\displaystyle \frac{|𝐱|}{s_2}}\right)^Br_2^b\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{2}}{}_{b}{}^{}\}.`$
We have used the facts that the integral of $`F`$ converges at infinity (first equality) and the integral of $`\stackrel{~}{F}_2`$ converges at the singularity 2 (second equality). Adding up the other contribution extending over $`𝒟_1`$, we readily obtain the complete partie finie as
$$\underset{s0}{lim}\{\stackrel{}{\mathrm{FP}}\genfrac{}{}{0pt}{}{}{B0}_{^3_1(s)}d^3𝐱\left(\frac{|𝐱|}{s_2}\right)^B\stackrel{~}{F}_2+\underset{a+3<0}{}\frac{s^{a+3}}{a+3}d\mathrm{\Omega }_1\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{a}{}^{}+\mathrm{ln}\left(\frac{s}{s_1}\right)d\mathrm{\Omega }_1\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{3}{}^{}\}.$$
Since the coefficients $`{}_{1}{}^{}f_{a}^{}`$, for $`a3`$, are those of the expansion when $`r_10`$ of $`F`$ as well as of $`\stackrel{~}{F}_2`$, we recognize in the expression above the partie finie (with respect to 1 only) of the integral of the regularized function $`\stackrel{~}{F}_2`$. Hence the intermediate expression
$$\mathrm{Pf}_{s_1,s_2}d^3𝐱F=\mathrm{Pf}_{s_1}\left\{\stackrel{}{\mathrm{FP}}\genfrac{}{}{0pt}{}{}{B0}d^3𝐱\left(\frac{|𝐱|}{s_2}\right)^B\stackrel{~}{F}_2\right\}.$$
(51)
To establish the proposition it remains to change of variable $`𝐱`$ into $`𝐫_1`$. At that point, we must be careful, because under this change of variable the regularization factor $`|𝐱|^B`$ changes itself in a complicated way. Fortunately, we can limit ourselves to the case where $`B`$ is infinitesimal, since we shall take the finite part afterwards, making $`B0`$. We substitute to $`|𝐱|^B`$ in the right side of (51) its equivalent expression in terms of $`𝐫_1`$ and where we expand when $`B0`$, i.e.
$$|𝐱|^B=r_1^Be^{B\mathrm{ln}\left(\frac{|𝐱|}{r_1}\right)}=r_1^B\left\{1+\frac{B}{2}\mathrm{ln}\left[1+2\frac{𝐧_1.𝐲_1}{r_1}+\frac{𝐲_1^2}{r_1^2}\right]+O(B^2)\right\},$$
(52)
where $`𝐧_1.𝐲_1`$ denotes the usual scalar product on $`^3`$ (and $`𝐲_1^2=𝐲_1.𝐲_1`$). Now, the dominant term in the latter expansion amounts simply to replacing $`|𝐱|^B`$ by $`r_1^B`$, which would yield precisely the result (46) we want to prove; but we have still to show that all the extra terms in the expansion (52), which carry at least a factor $`B`$ in front, do not contribute to the final result, i.e. that
$$\stackrel{}{\mathrm{FP}}\genfrac{}{}{0pt}{}{}{B0}\left[\frac{B}{2}^+\mathrm{}d^3𝐫_1\left(\frac{r_1}{s_2}\right)^B\mathrm{ln}\left[1+2\frac{𝐧_1.𝐲_1}{r_1}+\frac{𝐲_1^2}{r_1^2}\right]\stackrel{~}{F}_2+O(B^2)\right]=0.$$
(53)
Because of the factor $`B`$ in front, the only possible contribution to the finite part for $`B0`$ occurs when the integral develops a pole at $`B=0`$ due to the behaviour of the integrand at infinity ($`r_1+\mathrm{}`$). Hence, as indicated in (53), the value of the integral depends only on the bound at infinity \[this is also why we did not write a $`\mathrm{Pf}_{s_1}`$ symbol in (53): the partie finie deals with the bound $`r_1=0`$, which is irrelevant to this case\]. In order to evaluate the pole, we replace the integrand by its expansion when $`r_1+\mathrm{}`$. We know that $`F`$ behaves as $`o\left(\frac{1}{|𝐱|^3}\right)`$ at a maximum $`|𝐱|+\mathrm{}`$ to ensure the convergence of the integral of $`F`$ at infinity, swhen o we have $`F=o\left(\frac{1}{r_1^3}\right)`$ when $`r_1+\mathrm{}`$. Now, from the defining expression (47) of $`\stackrel{~}{F}_2`$, we obtain
$$\stackrel{~}{F}_2=\frac{1}{r_1^3}\stackrel{}{f}\genfrac{}{}{0pt}{}{}{2}{}_{3}{}^{}(𝐧_1)+o\left(\frac{1}{r_1^3}\right)\text{when }r_1+\mathrm{},$$
(54)
after making the replacements of $`r_2`$ and $`𝐧_2`$ by $`r_1`$ and $`𝐧_1`$ which are permitted because we are working at the dominant order when $`r_1+\mathrm{}`$. On the other hand, we have $`\mathrm{ln}\left[1+2\frac{𝐧_1.𝐲_1}{r_1}+\frac{𝐲_1^2}{r_1^2}\right]=2\frac{𝐧_1.𝐲_1}{r_1}+O\left(\frac{1}{r_1^2}\right)`$. So that the integral to be computed (as concerns the only relevant bound at infinity) reads as
$$^+\mathrm{}d^3𝐫_1r_1^B\mathrm{ln}[1+2\frac{𝐧_1.𝐲_1}{r_1}+\frac{𝐲_1^2}{r_1^2}]\stackrel{~}{F}_2=2^+\mathrm{}dr_1r_1^{B2}\{d\mathrm{\Omega }_1𝐧_1.𝐲_1\stackrel{}{f}\genfrac{}{}{0pt}{}{}{2}{}_{3}{}^{}(𝐧_1)+o\left(r_1^0\right)\}.$$
This integral cannot generate a pole at $`B=0`$ since such a pole could come only from a radial integral of the type $`^+\mathrm{}𝑑r_1r_1^{B1}`$ (after the angular integration has been performed). Repeating the same reasoning to any higher orders in $`B`$, we prove the equation (53) as well as Proposition 1.
In practice, Proposition 1 is used with the integration with respect to $`𝐧_1`$, followed by the integration over $`r_1`$ varying from 0 ($`\mathrm{Pf}_{s_1}`$ takes care of this bound) to infinity (where $`\mathrm{FP}_{B0}`$ does the work); Proposition 1 justifies this process even when the original integral is divergent at both singularities. The result of the angular integration depends on where the field point is located, either inside the ball $`_1(r_{12})`$ centered on $`𝐲_1`$ and of radius $`r_{12}`$ (the point 2 lies on the surface of this ball) or in the complementary domain $`^3_1(r_{12})`$. Therefore, a natural splitting of the integral (46) is
$$\mathrm{Pf}_{s_1,s_2}d^3𝐱F=\mathrm{Pf}_{s_1}_{_1(r_{12})}d^3𝐫_1\stackrel{~}{F}_2+\stackrel{}{\mathrm{FP}}\genfrac{}{}{0pt}{}{}{B0}_{^3_1(r_{12})}d^3𝐫_1\left(\frac{r_1}{s_2}\right)^B\stackrel{~}{F}_2,$$
(55)
taking into account the fact that the partie finie $`\mathrm{Pf}_{s_1}`$ applies only to the inner integral, over $`_1(r_{12})`$, and the finite part $`\mathrm{FP}_{B0}`$ only to the outer one, over $`^3_1(r_{12})`$. To be more specific, the angular integral of $`\stackrel{~}{F}_2`$ defines two angular-average functions $`\stackrel{~}{I}_2(r_1)`$ and $`\stackrel{~}{J}_2(r_1)`$ depending on whether $`𝐱`$ is in $`_1(r_{12})`$ or its complementary:
dΩ14πF~2={
~I2(r1)
when r1r12,
~J2(r1) when >r1r12.
\int{d\Omega_{1}\over 4\pi}~{}{\widetilde{F}}_{2}=\left\{\vbox{\vskip 1.20007pt\hbox{$~{}{\widetilde{I}}_{2}(r_{1})\qquad$
when $r_{1}\leq r_{12}~{},$}\vskip 6.0pt\hbox{
${\widetilde{J}}_{2}(r_{1})\qquad$ when $r_{1}>r_{12}~{}.$}\vskip 1.20007pt}\right. (56)
The functions $`\stackrel{~}{I}_2`$ and $`\stackrel{~}{J}_2`$ depend also explicitely on the source points $`𝐲_1`$ and $`𝐲_2`$. \[As an example, in the case $`\stackrel{~}{F}_2=r_2`$, we find $`\stackrel{~}{I}_2=r_{12}+\frac{r_1^2}{3r_{12}}`$ and $`\stackrel{~}{J}_2=r_1+\frac{r_{12}^2}{3r_1}`$.\] Now, knowing $`\stackrel{~}{I}_2`$ and $`\stackrel{~}{J}_2`$, we can achieve the radial integration according to the formula
$$\mathrm{Pf}_{s_1,s_2}d^3𝐱F=4\pi \mathrm{Pf}_{s_1}_0^{r_{12}}𝑑r_1r_1^2\stackrel{~}{I}_2+4\pi \stackrel{}{\mathrm{FP}}\genfrac{}{}{0pt}{}{}{B0}_{r_{12}}^+\mathrm{}𝑑r_1\left(\frac{r_1}{s_2}\right)^Br_1^2\stackrel{~}{J}_2.$$
(57)
The first term in (57) is quite simple to handle in practice, whereas the second one is more difficult because it requires a priori the knowledge of a closed-form expression for the integral of $`r_1^{B+2}\stackrel{~}{J}_2`$, valid for any $`B`$ such that $`\mathrm{}(B)>0`$. Obtaining this may not be feasible if $`F`$ is too complicated; in this event, we should use a different form of the integral at infinity. The second proposition, which provides the appropriate form, constitutes, perhaps, the most powerful way to compute the partie finie in rather complicated applications.
###### Proposition 2
The partie finie of the integral of $`F`$ (if convergent at infinity) reads as:
$`\mathrm{Pf}_{s_1,s_2}{\displaystyle d^3𝐱F}`$ $`=`$ $`4\pi \mathrm{Pf}_{s_1}{\displaystyle _0^{r_{12}}}𝑑r_1r_1^2\stackrel{~}{I}_2(r_1)+4\pi {\displaystyle _{r_{12}}^+\mathrm{}}𝑑r_1\left[r_1^2\stackrel{~}{J}_2(r_1)+{\displaystyle \frac{1}{r_1}}\left(r_2^3F\right)_2\right]`$ (58)
$`+`$ $`4\pi \left(r_2^3F\right)_2\mathrm{ln}\left({\displaystyle \frac{r_{12}}{s_2}}\right),`$ (59)
(and similarly by interchange of $`1`$ and $`2`$).
Proof. Consider the angular average of the expansion of $`\stackrel{~}{F}_2`$ when $`r_1+\mathrm{}`$ which has been determined in (54). We get
$$\stackrel{~}{J}_2\frac{d\mathrm{\Omega }_1}{4\pi }\stackrel{~}{F}_2=\frac{1}{r_1^3}\left(r_2^3F\right)_2+o\left(\frac{1}{r_1^3}\right),$$
(60)
where the coefficient of the dominant term is made of a Hadamard partie finie at point 2. Let us subtract and add to $`\stackrel{~}{J}_2`$ inside the second integral in (57) the previous dominant term at infinity. In this way, we may re-write it as the sum of a convergent integral at infinity on one hand, to which we can then remove the finite part prescription, and a simple extra integral on the other hand. Namely,
$`{\displaystyle _{r_{12}}^+\mathrm{}}𝑑r_1\left[r_1^2\stackrel{~}{J}_2+{\displaystyle \frac{1}{r_1}}\left(r_2^3F\right)_2\right]`$ $``$ $`\left(r_2^3F\right)_2\stackrel{}{\mathrm{FP}}{\displaystyle \genfrac{}{}{0pt}{}{}{B0}}{\displaystyle _{r_{12}}^+\mathrm{}}{\displaystyle \frac{dr_1}{r_1}}\left({\displaystyle \frac{r_1}{s_2}}\right)^B.`$
The extra integral is finally computed in a simple way as
$`\stackrel{}{\mathrm{FP}}{\displaystyle \genfrac{}{}{0pt}{}{}{B0}}{\displaystyle _{r_{12}}^+\mathrm{}}{\displaystyle \frac{dr_1}{r_1}}\left({\displaystyle \frac{r_1}{s_2}}\right)^B=\stackrel{}{\mathrm{FP}}{\displaystyle \genfrac{}{}{0pt}{}{}{B0}}[{\displaystyle \frac{1}{B}}\left({\displaystyle \frac{r_{12}}{s_2}}\right)^B]=\mathrm{ln}\left({\displaystyle \frac{r_{12}}{s_2}}\right),`$
where we used the properties of the analytic continuation. QED.
Thanks to Proposition 2 we are now able to compute many integrals which could not be deduced from the Riesz formula (43), unlike for (44). For instance,
$`\mathrm{Pf}_{s_1,s_2}{\displaystyle \frac{d^3𝐱}{r_1^3r_2^3(r_1+r_2)}}={\displaystyle \frac{4\pi }{r_{12}^4}}\left[\mathrm{ln}\left({\displaystyle \frac{r_{12}}{s_1}}\right)+\mathrm{ln}\left({\displaystyle \frac{r_{12}}{s_2}}\right){\displaystyle \frac{8}{3}}\mathrm{ln}2+{\displaystyle \frac{2}{3}}\right],`$ (62)
$`\mathrm{Pf}_{s_1,s_2}{\displaystyle \frac{d^3𝐱}{r_1^3r_2^3(r_1+r_2+r_{12})}}={\displaystyle \frac{2\pi }{r_{12}^4}}\left[\mathrm{ln}\left({\displaystyle \frac{r_{12}}{s_1}}\right)+\mathrm{ln}\left({\displaystyle \frac{r_{12}}{s_2}}\right)+{\displaystyle \frac{\pi ^2}{3}}4\right].`$ (63)
The result for the integral (63) is in agreement with the one that follows from a recent generalization of the Riesz formula to include arbitrary powers of $`r_1+r_2+r_{12}`$, which has been obtained by Jaranowski and Schäfer (see the appendix B.2 in ). In any case, the dependence of the partie-finie integral on the two constants $`s_1`$ and $`s_2`$ is given by
$`\mathrm{Pf}_{s_1,s_2}{\displaystyle d^3𝐱F}`$ $`=`$ $`4\pi \left(r_1^3F\right)_1\mathrm{ln}\left({\displaystyle \frac{r_{12}}{s_1}}\right)+4\pi \left(r_2^3F\right)_2\mathrm{ln}\left({\displaystyle \frac{r_{12}}{s_2}}\right)`$ (64)
$`+`$ $`\text{terms independent of }s_1\text{}s_2.`$ (65)
## V Partie finie of Poisson integrals
In this section we investigate the main properties of the partie finie of Poisson integrals of singular functions in the class $``$. We have in view the application to the post-Newtonian motion of particles in general relativity, since the post-Newtonian iteration proceeds typically through Poisson (or Poisson-type) integrals. Consider a fixed (“spectator”) point $`𝐱^{}^3`$ and, for each value of $`𝐱^{}`$, define the function $`S_𝐱^{}(𝐱)=F(𝐱)/|𝐱𝐱^{}|`$ where $`F`$. Clearly, for any given $`𝐱^{}`$, the function $`S_𝐱^{}`$ belongs to the class $`_{\mathrm{loc}}`$, introduced in Section II, Definition 2. In addition, when the spectator point $`𝐱^{}`$ coincides with the singular point $`𝐲_1`$ (and similarly for $`𝐲_2`$), we have $`S_{𝐲_1}`$. Since (as already mentionned) Definition 4 can be extended to functions in the class $`_{\mathrm{loc}}`$, we can consider the partie-finie integral
$$P(𝐱^{})=\frac{1}{4\pi }\mathrm{Pf}d^3𝐱S_𝐱^{}(𝐱)=\frac{1}{4\pi }\mathrm{Pf}\frac{d^3𝐱}{|𝐱𝐱^{}|}F(𝐱).$$
(66)
This is, indeed, what we shall call the “Poisson” integral of $`F`$. In particular, when the spectator point $`𝐱^{}`$ is equal to $`𝐲_1`$, we shall write
$$P(𝐲_1)=\frac{1}{4\pi }\mathrm{Pf}d^3𝐱S_{𝐲_1}(𝐱)=\frac{1}{4\pi }\mathrm{Pf}\frac{d^3𝐱}{r_1}F(𝐱).$$
(67)
The Poisson integral is not continuous at the singular point $`𝐲_1`$ because $`P(𝐱^{})`$, when initially defined for $`𝐱^{}𝐲_1`$, admits an expansion that is singular when $`𝐱^{}`$ tends to $`𝐲_1`$. In the present Section, our aim is to understand the limit relation of the integral $`P(𝐱^{})`$ when $`r_1^{}|𝐱^{}𝐲_1|0`$, and to connect it with the “regularized” integral $`P(𝐲_1)`$ given by (67). In particular, we shall show that the “partie finie” (in an extended Hadamard’s sense) of $`P(𝐱^{})`$ at $`𝐱^{}=𝐲_1`$ is related in a precise way to $`P(𝐲_1)`$. Let us make clear straight away that $`P(𝐱^{})`$, as a function of $`𝐱^{}`$ different from $`𝐲_1`$ (and $`𝐲_2`$), does not belong to the class $``$ as the Poisson integral typically generates logarithms in the expansion when $`r_1^{}0`$. In particular, the coefficient of zeroth power of $`r_1^{}`$ in the latter expansion contains a priori a $`\mathrm{ln}r_1^{}`$ term, and its partie finie in the sense of Definition 3 is in fact not finite at all, because of the presence of this formally infinite constant $`\mathrm{ln}r_1^{}=\mathrm{}`$. A possible way to deal with this problem, followed by Sellier in , is to exclude the $`\mathrm{ln}r_1^{}`$ (and any higher power of $`\mathrm{ln}r_1^{}`$) from the definition of the partie finie. On the other hand, in applications to the physical problem, the constant $`\mathrm{ln}r_1^{}`$ can be viewed as a “renormalization” constant, which is better to keep as it appears all the way through the calculation. Therefore, we simply include here the renormalization constant $`\mathrm{ln}r_1^{}`$ into the definition; but, for simplicity’s sake, we stick to the name of “partie finie” in this case (although the $`\mathrm{ln}r_1^{}`$ makes it formally infinite). Thus, for a function like $`P`$ admitting a logarithmic expansion:
$$N,P(𝐱^{})=\underset{\genfrac{}{}{0pt}{}{aN}{p=0,1}}{}r_{1}^{}{}_{}{}^{a}(\mathrm{ln}r_1^{})^p\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{a,p}{}^{}(𝐧_1^{})+o\left(r_{1}^{}{}_{}{}^{N}\right)\text{when }r_1^{}0,$$
(68)
we define the Hadamard partie finie of $`P`$ at 1 by
$$(P)_1=\frac{d\mathrm{\Omega }_1^{}}{4\pi }\left[\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{0,0}{}^{}(𝐧_1^{})+\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{0,1}{}^{}(𝐧_1^{})\mathrm{ln}r_1^{}\right].$$
(69)
###### Theorem 3
The Hadamard partie finie at 1 (in the previous sense) of the Poisson integral of any $`F`$ reads as
$$(P)_1=\frac{1}{4\pi }\mathrm{Pf}_{s_1,s_2}\frac{d^3𝐱}{r_1}F(𝐱)+\left[\mathrm{ln}\left(\frac{r_1^{}}{s_1}\right)1\right]\left(r_1^2F\right)_1,$$
(70)
with $`r_1^{}=|𝐱^{}𝐲_1|`$. Furthermore the constants $`s_1`$ cancel each other from the two terms in the right side of (70) (so the partie finie depends on the two constants $`\mathrm{ln}r_1^{}`$ and $`\mathrm{ln}s_2`$).
In words, the partie finie of the Poisson integral at 1 is equal to the regularized integral $`P(𝐲_1)`$, obtained from the replacement $`𝐱^{}𝐲_1`$ inside the integrand of $`P(𝐱^{})`$, augmented by a term associated with the presence of the (infinite) constant $`\mathrm{ln}r_1^{}`$.
Proof. The fact that the constants $`s_1`$ cancel out (so $`s_1`$ is “replaced” by $`r_1^{}`$) is a trivial consequence of the dependence of the partie finie on $`s_1`$ and $`s_2`$ determined in (64). For our proof, we need the explicit expressions of the objects $`P(𝐱^{})`$, when $`𝐱^{}`$ is different from $`𝐲_1`$ and $`𝐲_2`$, and $`P(𝐲_1)`$, following from Definition 4. For $`𝐱^{}𝐲_1`$ and $`r_1=|𝐱𝐲_1|0`$, we have the expansion
$$S_𝐱^{}(𝐱)=\underset{l0}{}\frac{()^l}{l!}_L^{}\left(\frac{1}{r_1^{}}\right)\underset{a}{}r_1^{a+l}n_1^L\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{a}{}^{}(𝐧_1)$$
(71)
(and idem $`12`$), where $`r_1^{}=|𝐱^{}𝐲_1|`$, $`_L^{}`$ being the multi-spatial derivative acting on $`𝐱^{}`$. From (24), we get the expression (for $`𝐱^{}𝐲_1`$ and $`𝐲_2`$)
$`P(𝐱^{})=`$ $``$ $`{\displaystyle \frac{1}{4\pi }}\underset{s0}{lim}\{{\displaystyle _{^3_1(s)_2(s)}}{\displaystyle \frac{d^3𝐱}{|𝐱𝐱^{}|}}F`$ (72)
$`+`$ $`{\displaystyle \underset{l0}{}}{\displaystyle \frac{()^l}{l!}}_L^{}\left({\displaystyle \frac{1}{r_1^{}}}\right)[{\displaystyle \underset{a+l+3<0}{}}{\displaystyle \frac{s^{a+l+3}}{a+l+3}}{\displaystyle }d\mathrm{\Omega }_1n_1^L\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{a}{}^{}+\mathrm{ln}\left({\displaystyle \frac{s}{s_1}}\right){\displaystyle }d\mathrm{\Omega }_1n_1^L\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{3l}{}^{}]`$ (73)
$`+`$ $`12\}.`$ (74)
Applying the recipe (69), we start by computing the angular integral over $`𝐧_1^{}=(𝐱^{}𝐲_1)/r_1^{}`$ (for a fixed $`r_1^{}`$) of $`P(𝐱^{})`$ in the form given by (72), and consider the limit $`r_1^{}0`$ afterwards. Since $`s`$ is fated to tend to zero first, one can choose $`s<r_1^{}`$, and as we are ultimately interested in the limit $`r_1^{}0`$, we also assume $`r_1^{}<r_{12}`$. To compute the angular average of the divergent terms in (72), we make use of the identities
$`{\displaystyle \frac{d\mathrm{\Omega }_1^{}}{4\pi }_L^{}\left(\frac{1}{r_1^{}}\right)}`$ $`=`$ $`{\displaystyle \frac{\delta _{0l}}{r_1^{}}},`$ (76)
$`{\displaystyle \frac{d\mathrm{\Omega }_1^{}}{4\pi }_L^{}\left(\frac{1}{r_2^{}}\right)}`$ $`=`$ $`_L\left({\displaystyle \frac{1}{r_{12}}}\right)`$ (77)
(where $`\delta _{0l}`$ denotes the Kronecker symbol). On the other hand, the relevant formula to treat the integral in the right side of (72) is
dΩ14π1|𝐱𝐱|={
1r1 (if <r1r1) ,
1r1 (if <r1r1) .
\int{d\Omega^{\prime}_{1}\over 4\pi}~{}{1\over|{\bf x}-{\bf x}^{\prime}|}=\left\{\vbox{\vskip 1.20007pt\hbox{$~{}{1\over r^{\prime}_{1}}\qquad$ (if $r_{1}<r^{\prime}_{1}$)~{},}
\vskip 6.0pt\hbox{
${1\over r_{1}}\qquad$ (if $r^{\prime}_{1}<r_{1}$)~{}.}\vskip 1.20007pt}\right. (78)
We split this integral into three other ones, the first of them extending over the “exterior” domain $`^3_1(r_1^{})_2(r_1^{})`$, and the two remaining ones over the ring-shaped regions $`_1(r_1^{})_1(s)`$ and $`_2(r_1^{})_2(s)`$. Hence:
$`{\displaystyle \frac{d\mathrm{\Omega }_1^{}}{4\pi }P(𝐱^{})}`$ $`=`$ $`{\displaystyle \frac{1}{4\pi }}\underset{s0}{lim}\{{\displaystyle _{^3_1(r_1^{})_2(r_1^{})}}{\displaystyle \frac{d^3𝐱}{r_1}}F`$ (79)
$`+`$ $`{\displaystyle \frac{1}{r_1^{}}}{\displaystyle _{_1(r_1^{})_1(s)}}d^3𝐱F+{\displaystyle _{_2(r_1^{})_2(s)}}{\displaystyle \frac{d^3𝐱}{r_1}}F`$ (80)
$`+`$ $`{\displaystyle \frac{1}{r_1^{}}}[{\displaystyle \underset{a+3<0}{}}{\displaystyle \frac{s^{a+3}}{a+3}}{\displaystyle }d\mathrm{\Omega }_1\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{a}{}^{}+\mathrm{ln}\left({\displaystyle \frac{s}{s_1}}\right){\displaystyle }d\mathrm{\Omega }_1\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{3}{}^{}]`$ (81)
$`+`$ $`{\displaystyle \underset{l0}{}}{\displaystyle \frac{()^l}{l!}}_L\left({\displaystyle \frac{1}{r_{12}}}\right)[{\displaystyle \underset{b+l+3<0}{}}{\displaystyle \frac{s^{b+l+3}}{b+l+3}}{\displaystyle }d\mathrm{\Omega }_2n_2^L\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{2}}{}_{b}{}^{}+\mathrm{ln}\left({\displaystyle \frac{s}{s_2}}\right){\displaystyle }d\mathrm{\Omega }_2n_2^L\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{2}}{}_{3l}{}^{}]\}.`$ (82)
Next, supposing that $`r_1^{}`$ is small enough, we may replace $`F`$ in the second and third terms by its own expansions around 1 and 2 respectively. We find that the divergent terms in $`s`$ cancel out, so we are allowed to apply the limit $`s0`$. This yields
$`{\displaystyle \frac{d\mathrm{\Omega }_1^{}}{4\pi }P(𝐱^{})}`$ $`=`$ $`{\displaystyle \frac{1}{4\pi }}\{{\displaystyle _{^3_1(r_1^{})_2(r_1^{})}}{\displaystyle \frac{d^3𝐱}{r_1}}F`$ (84)
$`+`$ $`{\displaystyle \frac{1}{r_1^{}}}[{\displaystyle \underset{a+3<0}{}}{\displaystyle \frac{r_{1}^{}{}_{}{}^{a+3}}{a+3}}{\displaystyle }d\mathrm{\Omega }_1\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{a}{}^{}+\mathrm{ln}\left({\displaystyle \frac{r_1^{}}{s_1}}\right){\displaystyle }d\mathrm{\Omega }_1\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{3}{}^{}+r_1^{}{\displaystyle }d\mathrm{\Omega }_1\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{2}{}^{}]`$ (85)
$`+`$ $`{\displaystyle \underset{l0}{}}{\displaystyle \frac{()^l}{l!}}_L\left({\displaystyle \frac{1}{r_{12}}}\right)[{\displaystyle \underset{b+l+3<0}{}}{\displaystyle \frac{r_{1}^{}{}_{}{}^{b+l+3}}{b+l+3}}{\displaystyle }d\mathrm{\Omega }_2n_2^L\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{2}}{}_{b}{}^{}+\mathrm{ln}\left({\displaystyle \frac{r_1^{}}{s_2}}\right){\displaystyle }d\mathrm{\Omega }_2n_2^L\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{2}}{}_{3l}{}^{}]`$ (86)
$`+`$ $`o(r_{1}^{}{}_{}{}^{0})\}`$ (87)
(the remainder dies out when $`r_1^{}0`$). Under the latter form we recognize most of the terms composing the integral $`P(𝐲_1)`$. Indeed, we have, respectively when $`r_10`$ and $`r_20`$,
$`S_{𝐲_1}(𝐱)`$ $`=`$ $`{\displaystyle \underset{a}{}}r_1^{a1}\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{a}{}^{}(𝐧_1),`$ (89)
$`S_{𝐲_1}(𝐱)`$ $`=`$ $`{\displaystyle \underset{l0}{}}{\displaystyle \frac{()^l}{l!}}_L\left({\displaystyle \frac{1}{r_{12}}}\right){\displaystyle \underset{b}{}}r_2^{b+l}n_2^L\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{2}}{}_{b}{}^{}(𝐧_2).`$ (90)
Now, using the form (27) of the partie finie with the change of notation $`s^{}=r_1^{}`$, we find
$`P(𝐲_1)=`$ $``$ $`{\displaystyle \frac{1}{4\pi }}\{{\displaystyle _{^3_1(r_1^{})_2(r_1^{})}}{\displaystyle \frac{d^3𝐱}{r_1}}F+{\displaystyle \underset{a+2<0}{}}{\displaystyle \frac{r_{1}^{}{}_{}{}^{a+2}}{a+2}}{\displaystyle }d\mathrm{\Omega }_1\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{a}{}^{}+\mathrm{ln}\left({\displaystyle \frac{r_1^{}}{s_1}}\right){\displaystyle }d\mathrm{\Omega }_1\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}_2`$ (91)
$`+`$ $`{\displaystyle \underset{l0}{}}{\displaystyle \frac{()^l}{l!}}_L\left({\displaystyle \frac{1}{r_{12}}}\right)[{\displaystyle \underset{b+l+3<0}{}}{\displaystyle \frac{r_{1}^{}{}_{}{}^{b+l+3}}{b+l+3}}{\displaystyle }d\mathrm{\Omega }_2n_2^L\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{2}}{}_{b}{}^{}+\mathrm{ln}\left({\displaystyle \frac{r_1^{}}{s_2}}\right){\displaystyle }d\mathrm{\Omega }_2n_2^L\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{2}}{}_{3l}{}^{}]`$ (92)
$`+`$ $`o(r_{1}^{}{}_{}{}^{0})\}.`$ (93)
We finally evaluate the difference between (84) and (91) and look for the partie finie in the sense of (69) (i.e. keeping the $`\mathrm{ln}r_1^{}`$ term). We obtain
$$(P)_1P(𝐲_1)=[\mathrm{ln}\left(\frac{r_1^{}}{s_1}\right)1]\frac{d\mathrm{\Omega }_1}{4\pi }\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{2}{}^{}\text{(}\text{QED}\text{).}$$
(94)
The same type of result can be proved for the partie finie of the “twice-iterated” Poisson integral defined by
$$Q(𝐱^{})=\frac{1}{4\pi }\mathrm{Pf}d^3𝐱|𝐱𝐱^{}|F(𝐱).$$
(95)
We find, analogously to (70), that
$$(Q)_1=\frac{1}{4\pi }\mathrm{Pf}d^3𝐱r_1F(𝐱)+\left[\mathrm{ln}\left(\frac{r_1^{}}{s_1}\right)+\frac{1}{2}\right]\left(r_1^4F\right)_1.$$
(96)
For the parties finies of the gradients of the Poisson and twice-iterated Poisson integrals, we get
$`(_iP)_1`$ $`=`$ $`{\displaystyle \frac{1}{4\pi }}\mathrm{Pf}{\displaystyle d^3𝐱\frac{n_1^i}{r_1^2}F(𝐱)}+\mathrm{ln}\left({\displaystyle \frac{r_1^{}}{s_1}}\right)\left(n_1^ir_1F\right)_1,`$ (98)
$`(_iQ)_1`$ $`=`$ $`{\displaystyle \frac{1}{4\pi }}\mathrm{Pf}{\displaystyle d^3𝐱n_1^iF(𝐱)}\left[\mathrm{ln}\left({\displaystyle \frac{r_1^{}}{s_1}}\right){\displaystyle \frac{1}{2}}\right]\left(n_1^ir_1^3F\right)_1.`$ (99)
Those results are proved in the same way as in Theorem 3 (with similar cancellations of the constants $`s_1`$).
## VI Partie finie pseudo-functions
### A A class of pseudo-functions
The concept of Hadamard partie finie of the divergent integral of functions $`F`$ yields a natural definition of a class of pseudo-functions $`\mathrm{Pf}F`$ (“partie finie” of $`F`$), namely linear forms on a subset of $``$, of the type $`G<\mathrm{Pf}F,G>`$, where the result of the action of $`\mathrm{Pf}F`$ on $`G`$ is denoted using a duality bracket $`<,>`$.
###### Definition 5
For any function $`F`$ we define the pseudo-function $`\mathrm{Pf}F`$ as the linear functional which associates to any $`G`$, such that $`FG=o(|𝐱|^3)`$ when $`|𝐱|+\mathrm{}`$, the partie-finie integral of the product $`FG`$, i.e.
$$<\mathrm{Pf}F,G>=\mathrm{Pf}d^3𝐱FG,$$
(100)
where the partie-finie integral is defined by $`(\text{24})`$.
As we can see, the pseudo-function $`\mathrm{Pf}F`$ is not a linear form on $``$ itself but on the subset of $``$ such that the integral converges at infinity. For simplicity’s sake we will always say that statements like (100) are valid $`G`$, without mentioning this restriction. Note also that the partie-finie integral depends on the two constants $`s_1`$, $`s_2^+`$, and so is the pseudo-function which should indeed be denoted $`\mathrm{Pf}_{s_1,s_2}F`$. In our simplified notation we omit indicating $`s_1`$ and $`s_2`$.
An evident property of the duality bracket is its “symmetry” by exchanging the roles of the two slots of the bracket, namely:
$$(F,G)^2,<\mathrm{Pf}F,G>=<\mathrm{Pf}G,F>.$$
(101)
Also evident are the properties:
$$<\mathrm{Pf}F,GH>=<\mathrm{Pf}G,FH>=<\mathrm{Pf}(FG),H>=<\mathrm{Pf}(FGH),1>.$$
In the following we generally do not distinguish between the two slots in $`<,>`$. Accordingly we define the object
$$<F,\mathrm{Pf}G><\mathrm{Pf}G,F>.$$
Even more, we allow for a bracket in which the two slots are filled with pseudo-functions. Thus, we write
$$<\mathrm{Pf}F,\mathrm{Pf}G><\mathrm{Pf}F,G>=<\mathrm{Pf}G,F>,$$
which constitutes merely the definition of the new object $`<\mathrm{Pf}F,\mathrm{Pf}G>`$.
We denote by $`^{}`$ the set of pseudo-functions $`\mathrm{Pf}F`$, when $`F`$ describes the class $``$, introduced by Definition 5 : $`^{}=\left\{\mathrm{Pf}F;F\right\}`$. Later we shall extend the definition of $`^{}`$ to include the “limits” of some pseudo-functions. Roughly, the set $`^{}`$ plays a role analogous to the set $`𝒟^{}`$ in distribution theory , which is dual to the class $`𝒟`$ of functions which are both $`C^{\mathrm{}}(^3)`$ (for what we are concerned about here) and zero outside a compact subset of $`^3`$. In distribution theory the set $`𝒟`$ is endowed with the Schwartz topology : a sequence $`(\phi _n)_n`$ of elements of $`𝒟`$ converges to zero if and only if (i) $`n_0`$ and a compact $`K`$ of $`^3`$ such that $`nn_0`$, $`\mathrm{supp}(\phi _n)K`$, and (ii) for any multi-index $`L=i_1i_2\mathrm{}i_l`$, $`_L\phi _n`$ converges uniformly to zero. $`𝒟^{}`$ is the set of linear forms on $`𝒟`$ that are continuous with respect to that topology. In this paper we shall not attempt to define a topology on the class $``$, and shall limit ourselves (having in view the physical application) to the definition of the algebraic and differential rules obeyed by the pseudo-functions of $`^{}`$. However we can state :
###### Lemma 3
The pseudo-functions of $`^{}`$, when restricted to the set $`𝒟`$ of $`C^{\mathrm{}}(^3)`$ functions with compact support, are distributions in the sense of Schwartz :
$$\mathrm{Pf}F_{|_𝒟}𝒟^{}.$$
(102)
Proof. All we need to check is that the pseudo-function $`\mathrm{Pf}F_{|_𝒟}`$ is continuous with respect to the Schwartz topology . Consider a sequence $`\phi _n𝒟`$ tending to zero in the sense recalled above. Applying the partie-finie integral in the form (27), we get ($`s^{}1`$ and $`N`$)
$`<\mathrm{Pf}F_{|_𝒟},\phi _n>`$ $`=`$ $`{\displaystyle _{^3_1(s^{})_2(s^{})}}d^3𝐱F\phi _n`$
$`+`$ $`{\displaystyle \underset{l0}{}}{\displaystyle \frac{1}{l!}}_L\phi _n(𝐲_1)[{\displaystyle \underset{\genfrac{}{}{0pt}{}{a+l+3N}{\mathrm{and}0}}{}}{\displaystyle \frac{s_{}^{}{}_{}{}^{a+l+3}}{a+l+3}}{\displaystyle }d\mathrm{\Omega }_1n_1^L\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{a}{}^{}+\mathrm{ln}\left({\displaystyle \frac{s^{}}{s_1}}\right){\displaystyle }d\mathrm{\Omega }_1n_1^L\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{l3}{}^{}]`$
$`+`$ $`12+o(s_{}^{}{}_{}{}^{N}).`$
Since $`\phi _n`$ and all its derivatives $`_L\phi _n`$ tend uniformly towards zero in a given compact $`K`$, clearly so does the sequence of real numbers $`<\mathrm{Pf}F_{|_𝒟},\phi _n>`$, which shows that $`\mathrm{Pf}F_{|_𝒟}`$ is indeed continuous (QED).
###### Definition 6
The product (“.”) of $`F`$ and of $`\mathrm{Pf}G^{}`$, and the product of two pseudo-functions $`\mathrm{Pf}F`$ and $`\mathrm{Pf}G`$, are defined as
$$F.\mathrm{Pf}G\mathrm{Pf}F.\mathrm{Pf}G\mathrm{Pf}(FG)^{}.$$
(103)
In particular $`F.\mathrm{Pf}G=G.\mathrm{Pf}F`$.
In the following, we will remove the dot indicating the product and write indifferently
$$F\mathrm{Pf}G=G\mathrm{Pf}F=\mathrm{Pf}(FG)=\mathrm{Pf}F\mathrm{Pf}G=FG\mathrm{Pf}1.$$
(104)
Notice that from the symmetry of the duality bracket we have, $`H`$,
$$<G\mathrm{Pf}F,H>=\mathrm{Pf}d^3𝐱FGH=<\mathrm{Pf}F,GH>.$$
(105)
Therefore, when applied to the restriction of pseudo-functions to $`𝒟`$, the product of Definition 6 agrees with the product of a distribution and a function $`\psi C^{\mathrm{}}(^3`$), i.e.
$$\phi 𝒟,<\psi \mathrm{Pf}F_{|_𝒟},\phi >=<\mathrm{Pf}F_{|_𝒟},\psi \phi >.$$
(106)
### B A Dirac delta-pseudo-function
Consider, for $`\epsilon ^+`$, the Riesz delta-function $`{}_{\epsilon }{}^{}\delta _{1}^{}`$ that we introduced in (35). Since $`{}_{\epsilon }{}^{}\delta _{1}^{}`$ we can associate to it the pseudo-function $`\mathrm{Pf}{}_{\epsilon }{}^{}\delta _{1}^{}`$. Now, Lemma 2 \[see (36)\] can be re-stated by means of the duality bracket as
$$\underset{\epsilon 0}{lim}<\mathrm{Pf}{}_{\epsilon }{}^{}\delta _{1}^{},F>=(F)_1.$$
(107)
This motivates the following definition.
###### Definition 7
We define the pseudo-function $`\mathrm{Pf}\delta _1`$ by
$$F,<\mathrm{Pf}\delta _1,F>=(F)_1.$$
(108)
We then extend the definition of the set $`^{}`$ to include this pseudo-function: $`\mathrm{Pf}\delta _1^{}`$.
Obviously $`\mathrm{Pf}\delta _1`$ can be viewed as the “limit” \[but we have not defined a topology on $``$\] of the pseudo-functions $`\mathrm{Pf}{}_{\epsilon }{}^{}\delta _{1}^{}`$ when $`\epsilon 0`$. The restriction of $`\mathrm{Pf}\delta _1`$ to $`𝒟`$ is identical to the usual Dirac measure,
$$\mathrm{Pf}\delta _{1}^{}{}_{|_𝒟}{}^{}=\delta _1\delta (𝐱𝐲_1),$$
(109)
so that the pseudo-function $`\mathrm{Pf}\delta _1`$ appears as a natural generalization of the Dirac measure in the context of Hadamard parties finies. In the following, we shall do as if $`\delta _1`$ would belong to the original class of functions $``$, writing for instance
$$<\mathrm{Pf}F,\delta _1><\mathrm{Pf}\delta _1,F>=(F)_1.$$
(110)
Of course, this equation constitutes in fact the definition of the bracket $`<\mathrm{Pf}F,\delta _1>`$.
###### Definition 8
For any $`F`$ the pseudo-function $`\mathrm{Pf}(F\delta _1)`$ is defined, consistently with the product $`(\text{103})`$, by
$$G,<\mathrm{Pf}(F\delta _1),G>=(FG)_1.$$
(111)
We include into $`^{}`$ all the pseudo-functions of this type: $`\mathrm{Pf}(F\delta _1)^{}`$ (that is, we consider $`_{\mathrm{new}}^{}=^{}+\delta _1+\delta _2`$; and we henceforth drop the “new”).
Notice that an immediate consequence of the “non-distributivity” of the Hadamard partie finie, namely $`(FG)_1(F)_1(G)_1`$, is the fact that
$$\mathrm{Pf}(F\delta _1)(F)_1\mathrm{Pf}\delta _1.$$
(112)
As an example, we have $`(r_1)_1=0`$; but $`\mathrm{Pf}(r_1\delta _1)`$ is not zero, since $`<\mathrm{Pf}(r_1\delta _1),1/r_1>=1`$ for instance. The pseudo-function $`\mathrm{Pf}(F\delta _1)`$ represents the product of a delta-function with a function that is singular on its own support, whereas this product is ill-defined in the standard distribution theory. However, this object, as seen as a distribution, i.e. when restricted to the class $`𝒟`$ of smooth functions with compact support, does exist in the standard theory. Using the Taylor expansion when $`r_10`$ of any $`\phi 𝒟`$, that is $`_{l0}\frac{1}{l!}r_1^ln_1^L_L\phi (𝐲_1)`$, we obtain
$$<\mathrm{Pf}(F\delta _1)_{|_𝒟},\phi >=(F\phi )_1=\underset{l0}{}\frac{1}{l!}_L\phi (𝐲_1)\frac{d\mathrm{\Omega }_1}{4\pi }n_1^L\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{l}{}^{},$$
(113)
where $`{}_{1}{}^{}f_{l}^{}`$ denotes the coefficient of $`1/r_1^l`$ in the expansion of $`F`$ when $`r_10`$. Notice that the sum in (113) is always finite because $`la_0`$, where $`a_0=a_0(F)`$ is the smallest exponent of $`r_1`$ in the expansion of $`F`$ (see Definition 1). From (113) we derive immediately the “intrinsic” form of the distribution $`\mathrm{Pf}(F\delta _1)_{|_𝒟}`$, that is
$$\mathrm{Pf}(F\delta _1)_{|_𝒟}=\underset{l0}{}\frac{()^l}{l!}_L\delta _1\frac{d\mathrm{\Omega }_1}{4\pi }n_1^L\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{l}{}^{}=\underset{l0}{}\frac{()^l}{l!}\left(r_1^ln_1^LF\right)_1_L\delta _1,$$
(114)
where $`_L\delta _1`$ denotes the $`l`$th partial derivative of the Dirac measure (and where the sums are finite). We have for instance
$$\mathrm{Pf}\left(\frac{\delta _1}{r_1^2}\right)_{|_𝒟}=\frac{1}{6}\mathrm{\Delta }\delta _1.$$
(115)
Note also that the distribution $`\mathrm{Pf}(F\delta _1)_{|_𝒟}`$ can be recovered, quite naturally, from the Laplacian (in the ordinary distributional sense) of the bracket corresponding to the “Poisson” integral of $`\mathrm{Pf}(F\delta _1)`$, i.e. formed by $`\mathrm{Pf}(F\delta _1)`$ acting on the function $`𝐱1/|𝐱𝐱^{}|`$. For any given $`𝐱^{}`$, this function belongs to $`_{\mathrm{loc}}`$ and we are still allowed to consider such a bracket (see also Section V). Thus we define
$$G(𝐱^{})=\frac{1}{4\pi }<\mathrm{Pf}(F\delta _1),\frac{1}{|𝐱𝐱^{}|}>=\frac{1}{4\pi }\left(\frac{F(𝐱)}{|𝐱𝐱^{}|}\right)_1.$$
(116)
For $`𝐱^{}`$ different from the singularity $`𝐲_1`$, we find, using the Taylor expansion of $`1/|𝐱𝐱^{}|`$ around $`𝐲_1`$,
$$G(𝐱^{})=\frac{1}{4\pi }\underset{l0}{}\frac{()^l}{l!}\left(r_1^ln_1^LF\right)_1_L^{}\left(\frac{1}{r_1^{}}\right).$$
(117)
Clearly the function $`G`$, if considered as a function of the variable $`𝐱^{}`$, belongs to $``$. Now, we see from (114) that the “ordinary” Laplacian of $`G(𝐱^{})`$ is precisely equal to $`\mathrm{Pf}(F\delta _1)_{|_𝒟}`$, namely
$$\mathrm{\Delta }^{}G_{}^{}{}_{|_𝒟}{}^{}=\underset{l0}{}\frac{()^l}{l!}\left(r_1^ln_1^LF\right)_1_L^{}\delta _1^{}=\mathrm{Pf}(F^{}\delta _1^{})_{|_𝒟}.$$
(118)
Let us point out that $`G`$ has no partie finie at the point 1: $`(G)_1=0`$; so, in order to compute its partie finie at 1, we are not allowed to replace formally $`𝐱^{}`$ by $`𝐲_1`$ inside the defining expression (116):
$$0=4\pi (G)_1<\mathrm{Pf}(F\delta _1),\frac{1}{r_1}>=\left(\frac{F}{r_1}\right)_1=\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{1}{}^{}.$$
(119)
\[The function $`G(𝐱^{})`$ is not continuous at 1, as we can easily see from its singular expansion (117).\]
Finally let us mention how to give a sense to a pseudo-function that would be associated with the square of the delta-function. $`\epsilon >0`$, we have $`{}_{\epsilon }{}^{}\delta _{1}^{2}`$, and hence, we can consider the partie-finie integral of $`{}_{\epsilon }{}^{}\delta _{1}^{2}F`$. In the limit $`\epsilon 0`$ we get
$$\underset{\epsilon 0}{lim}<\mathrm{Pf}{}_{\epsilon }{}^{}\delta _{1}^{2},F>=\underset{\epsilon 0}{lim}\mathrm{Pf}d^3𝐱{}_{\epsilon }{}^{}\delta _{1}^{2}F=0,$$
(120)
essentially because we have a square $`\epsilon ^2`$ in factor which kills any divergencies arising from the integral. Therefore $`\mathrm{Pf}\delta _1^2`$ is (defined to be) identically zero. More generally,
$$F,\mathrm{Pf}(F\delta _1^2)=0,$$
(121)
and we shall not hesitate to write such identities as
$$<\mathrm{Pf}\delta _1,F\delta _1>=<\delta _1,\mathrm{Pf}(F\delta _1)>=<\mathrm{Pf}(F\delta _1^2),1>=0.$$
(122)
Note also that
$$\mathrm{Pf}(F\delta _1\delta _2)=0.$$
(123)
## VII Derivative of pseudo-functions
### A A derivative operator on $``$
From now on we shall generally suppose, in order to simplify the presentation, that the powers of $`r_1`$ and $`r_2`$ in the expansions of $`F`$ around the two singularities are positive or negative integers ($``$). Our aim is to define an appropriate partial derivative operator acting on the pseudo-functions of the type $`\mathrm{Pf}F`$. First of all, we know (Lemma 3) that the restriction of $`\mathrm{Pf}F`$ to $`𝒟`$ is a distribution in the ordinary sense, so we already have at our disposal the derivative operator of distribution theory , which is uniquely determined — as well as any higher-order derivatives — by the requirement:
$$\phi 𝒟,<_i\left(\mathrm{Pf}F_{|_𝒟}\right),\phi >=<\mathrm{Pf}F_{|_𝒟},_i\phi >.$$
(124)
It is clear from viewing $`\mathrm{Pf}F_{|_𝒟}`$ as an integral operator acting on $`\phi `$, that (124) corresponds to a rule of “integration by part” in which the “all-integrated” (surface) term vanishes. In particular the “integral of a gradient” is zero. This motivates the following definition.
###### Definition 9
A partial derivative operator $`_i`$ acting on pseudo-functions of $`^{}`$ is said to satisfy the rule of integration by parts iff
$$F,G,<_i(\mathrm{Pf}F),G>=<_i(\mathrm{Pf}G),F>.$$
(125)
Notice the symmetry between the two slots of the duality bracket in (125). As an immediate consequence, for a derivative operator satisfying this rule, we have
$$F,<_i(\mathrm{Pf}F),F>=0.$$
(126)
Furthermore, if we assume $`_i(\mathrm{Pf1})=0`$ in addition to Definition 9, then
$$F,<_i(\mathrm{Pf}F),1>=0.$$
(127)
Of course, both (126) and (127) correspond to the intuitive idea that the integral of a gradient (in a “distributional-extended” sense) should be zero.
###### Proposition 3
The most general derivative operator on $`^{}`$ satisfying the rule of integration by parts $`(\text{125})`$ reads
$$_i(\mathrm{Pf}F)=\mathrm{Pf}(_iF)+\mathrm{D}_i[F]^{},$$
(128)
where $`\mathrm{Pf}(_iF)`$ represents the “ordinary” derivative, and where the “distributional” term $`\mathrm{D}_i[F]=\mathrm{H}_i[F]+\mathrm{D}_i^{\mathrm{part}}[F]`$ is the sum of the general solution of the homogeneous equation, i.e. a linear functional $`\mathrm{H}_i[F]`$ such that
$$F,G,<\mathrm{H}_i[F],G>+<\mathrm{H}_i[G],F>=0,$$
(129)
and of the particular solution defined by
$$\mathrm{D}_i^{\mathrm{part}}[F]=4\pi \mathrm{Pf}(n_1^i[\frac{1}{2}r_1\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{1}{}^{}+\underset{k0}{}\frac{1}{r_1^k}\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{2k}{}^{}]\delta _1+12).$$
(130)
When applied on any $`G`$, the particular solution reads
$$<\mathrm{D}_i^{\mathrm{part}}[F],G>=d\mathrm{\Omega }_1n_1^i[\frac{1}{2}\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{1}{}^{}\stackrel{}{g}\genfrac{}{}{0pt}{}{}{1}{}_{1}{}^{}+\underset{k0}{}\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{2k}{}^{}\stackrel{}{g}\genfrac{}{}{0pt}{}{}{1}{}_{k}{}^{}]+12.$$
(131)
Proof. We replace the form (128) of the derivative operator into the rule (125) and find
$$<\mathrm{D}_i[F],G>+<\mathrm{D}_i[G],F>=<\mathrm{Pf}(_iF),G><\mathrm{Pf}(_iG),F>.$$
The right-hand side can be readily re-written as the partie-finie integral of a gradient,
$$<\mathrm{D}_i[F],G>+<\mathrm{D}_i[G],F>=\mathrm{Pf}d^3𝐱_i(FG).$$
(132)
Now we know from (29) that the integral of a gradient is equal to the partie finie of the surface integrals around the singularities when the surface areas shrink to zero; thus
$$<\mathrm{D}_i[F],G>+<\mathrm{D}_i[G],F>=4\pi (n_1^ir_1^2FG)_1+12.$$
We replace into the right side $`F`$ and $`G`$ by their expansions around 1, and after an easy calculation we arrive at
$$<\mathrm{D}_i[F],G>+<\mathrm{D}_i[G],F>=d\mathrm{\Omega }_1n_1^i\left[\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{1}{}^{}\stackrel{}{g}\genfrac{}{}{0pt}{}{}{1}{}_{1}{}^{}+\underset{k0}{}(\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{2k}{}^{}\stackrel{}{g}\genfrac{}{}{0pt}{}{}{1}{}_{k}{}^{}+\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{k}{}^{}\stackrel{}{g}\genfrac{}{}{0pt}{}{}{1}{}_{2k}{}^{})\right]+12.$$
(133)
It is clear that the particular solution given by (130) or (131) solves the latter equation. As a consequence, the most general solution is simply obtained by adding the general solution of the homogeneous equation, i.e. (133) with zero in the right side, which is precisely a $`\mathrm{H}_i[F]`$ satisfying the “anti-symmetry” property $`<\mathrm{H}_i[F],G>+<\mathrm{H}_i[G],F>=0`$. QED.
As we see from Proposition 3, the rule of integration by parts does not permit, unlike in the case of distribution theory \[see (124)\], to fully specify the derivative operator. Obviously, we must supplement the rule by another statement indicating the cases for which the new derivative should reduce to the “ordinary” one, i.e. when we should have $`_i(\mathrm{Pf}F)=\mathrm{Pf}(_iF)`$. Clearly, we would like to recover the ordinary derivative in the cases where the function is “not too much singular”. In the following, we shall require essentially that our derivative reduces to the ordinary one when the function $`F`$ is bounded near the singularities \[in addition of belonging to $`C^{\mathrm{}}(^3\{𝐲_{1,2}\})`$\], in the sense that there exists a neighbourhood $`𝒩`$ containing the two singularities $`𝐲_1`$ and $`𝐲_2`$ and a constant $`M^+`$ such that $`𝐱𝒩|F(𝐱)|M`$. Let us refer to the coefficients of the negative powers of $`r_1`$ and $`r_2`$ in the expansions of $`F`$, i.e. the $`{}_{1}{}^{}f_{1k}^{}`$’s and $`{}_{2}{}^{}f_{1k}^{}`$’s where $`k`$, as the singular coefficients of $`F`$ (recall that we assumed that the powers of $`r_1`$ and $`r_2`$ are integers). Clearly, a function is bounded near the singularities if and only if all its singular coefficients vanish. This means that we shall require that the distributional term $`\mathrm{D}_i[F]`$, which is a linear functional of the coefficients in the expansions of $`F`$, should depend only on the singular coefficients $`{}_{1}{}^{}f_{1k}^{}`$ and $`{}_{2}{}^{}f_{1k}^{}`$ of $`F`$. This is already the case of our particular solution $`\mathrm{D}_i^{\mathrm{part}}[F]`$ in (130). We now look for the most general possible $`\mathrm{H}_i[F]`$ depending on the $`{}_{1}{}^{}f_{1k}^{}`$’s (and $`12`$).
All the singular coefficients admit some spherical-harmonics or equivalently STF expansions of the type (10)-(11), with STF-tensorial coefficients $`{}_{1}{}^{}\widehat{f}_{1k}^{L}`$ \[where $`L=i_1\mathrm{}i_l`$; see (10) for definition\], so we are led to requiring that $`\mathrm{H}_i[F]`$ be the most general (linear) functional of the STF tensors $`{}_{1}{}^{}\widehat{f}_{1k}^{L}`$ and $`12`$. Moreover, we demand that $`\mathrm{H}_i[F]`$, like $`\mathrm{D}_i^{\mathrm{part}}[F]`$, is proportional to the Dirac pseudo-function $`\mathrm{Pf}\delta _1`$ (as we shall see, the gradient of $`\mathrm{Pf}\delta _1`$ is itself proportional to $`\mathrm{Pf}\delta _1`$ so there is no loss of generality). Now, we have also to take into account the fact that the dimensionality of $`\mathrm{H}_i[F]`$ should be compatible with the one of $`\mathrm{Pf}(_iF)`$. Endowing $`^3`$ with a unit of length to measure the space coordinates, the Dirac pseudo-function $`\mathrm{Pf}\delta _1`$ takes the dimension of the inverse cube of a length, and $`\mathrm{H}_i[F]`$ the dimension of $`F`$ divided by this length (in physical applications, we do not want to introduce any special physical scale). We conclude that $`\mathrm{H}_i[F]`$ must be of the general form:
$$\mathrm{H}_i[F]=\underset{k0}{}\underset{l=0}{\overset{+\mathrm{}}{}}\mathrm{Pf}\left([\alpha _{k,l}\widehat{n}_1^{iL}\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{1k}{}^{L}+\beta _{k,l}n_1^L\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{1k}{}^{iL}]r_1^{1k}\delta _1\right)+12,$$
(134)
where the $`\alpha _{k,l}`$’s and $`\beta _{k,l}`$’s denote some purely constant numerical coefficients (and where, as usual, the sum over $`k`$ is finite). Applying this $`\mathrm{H}_i[F]`$ on any $`G`$ we readily obtain
$$<\mathrm{H}_i[F],G>=\underset{k0}{}\underset{l=0}{\overset{+\mathrm{}}{}}\frac{l!}{(2l+1)!!}[\frac{l+1}{2l+3}\alpha _{k,l}\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{1k}{}^{L}\stackrel{}{\widehat{g}}\genfrac{}{}{0pt}{}{}{1}{}_{1+k}{}^{iL}+\beta _{k,l}\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{1k}{}^{iL}\stackrel{}{\widehat{g}}\genfrac{}{}{0pt}{}{}{1}{}_{1+k}{}^{L}]+12.$$
(135)
At last we must impose the anti-symmetry condition (129). For any $`G`$ whose all singular coefficients vanish we have $`<\mathrm{H}_i[G],F>=0`$; then, the anti-symmetry condition tells us that (135) should be identically zero for any such $`G`$ and any $`F`$. Therefore, we must have $`\alpha _{k,l}=0`$ and $`\beta _{k,l}=0`$ whenever $`k1`$, so we are left with only the coefficients $`\alpha _{0,l}`$ and $`\beta _{0,l}`$, and the condition (129) now implies
$$0=\underset{l=0}{\overset{+\mathrm{}}{}}\frac{l!}{(2l+1)!!}(\frac{l+1}{2l+3}\alpha _{0,l}+\beta _{0,l})[\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{1}{}^{L}\stackrel{}{\widehat{g}}\genfrac{}{}{0pt}{}{}{1}{}_{1}{}^{iL}+\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{1}{}^{iL}\stackrel{}{\widehat{g}}\genfrac{}{}{0pt}{}{}{1}{}_{1}{}^{L}]+12,$$
which can clearly be satisfied only if (and only if) $`\frac{l+1}{2l+3}\alpha _{0,l}+\beta _{0,l}=0`$. Thus, posing $`\alpha _l\alpha _{0,l}`$, we have just proved:
###### Lemma 4
The most general $`\mathrm{H}_i[F]`$ that vanishes for any bounded function $`F`$ and possesses the correct dimension depends only on (the STF-harmonics of) the singular coefficients $`{}_{1}{}^{}f_{1}^{}`$ and $`{}_{2}{}^{}f_{1}^{}`$ and is given by
$$\mathrm{H}_i[F]=\underset{l=0}{\overset{+\mathrm{}}{}}\alpha _l\mathrm{Pf}\left([\widehat{n}_1^{iL}\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{1}{}^{L}\frac{l+1}{2l+3}n_1^L\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{1}{}^{iL}]r_1\delta _1\right)+12,$$
(136)
where the $`\alpha _l`$’s form a countable set of arbitrary numerical coefficients.
\[The angular dependence of the first term in (136) is expressed by means of the STF tensor $`\widehat{n}_1^{iL}`$.\] Equivalently we have
$$<\mathrm{H}_i[F],G>=\underset{l=0}{\overset{+\mathrm{}}{}}\alpha _l\frac{(l+1)!}{(2l+3)!!}[\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{1}{}^{L}\stackrel{}{\widehat{g}}\genfrac{}{}{0pt}{}{}{1}{}_{1}{}^{iL}\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{1}{}^{iL}\stackrel{}{\widehat{g}}\genfrac{}{}{0pt}{}{}{1}{}_{1}{}^{L}]+12.$$
(137)
This expression is anti-symmetric in the exchange $`FG`$ as required.
To sum up, we have obtained the most general derivative operator $`_i(\mathrm{Pf}F)=\mathrm{Pf}(_iF)+\mathrm{D}_i[F]`$ that satisfies the rule of integration by parts and depends only on the singular coefficients of $`F`$. The distributional term $`\mathrm{D}_i[F]`$ is the sum of a “particular” solution fully specified by (130) or (131), and of a “homogeneous” solution given by (136) or (137) in terms of an infinite set of arbitrary numerical coefficients $`\alpha _l`$ (and $`l`$). In Section VIII we shall see how one can reduce the arbitrariness of the definition of the derivative to only one single coefficient $`K`$.
### B Some properties of the derivative
At this stage, one can already investigate some properties of the distributional term $`\mathrm{D}_i[F]=\mathrm{D}_i^{\mathrm{part}}[F]+\mathrm{H}_i[F]`$, using the fact that the yet un-specified $`<\mathrm{H}_i[F],G>`$ depends only on $`{}_{1}{}^{}f_{1}^{}`$ and $`{}_{1}{}^{}g_{1}^{}`$ (and $`12`$). Let us first check that the derivative operator, when restricted to the smooth and compact-support functions of $`𝒟`$, reduces to the distributional derivative of distribution theory . This must actually be true since the fundamental property (124) of the distributional derivative is a particular case of our rule of integration by parts, and because the derivative of $`\phi 𝒟`$ reduces to the ordinary one. However, it is instructive to verify directly this fact using the expression (130). Applying $`\mathrm{D}_i[F]`$ on $`\phi 𝒟`$ and using the Taylor expansion of $`\phi `$ around 1: $`\phi =_{k0}\frac{1}{k!}r_1^kn_1^K(_K\phi )(𝐲_1)`$, we obtain
$`<\mathrm{D}_i[F],\phi >`$ $`=`$ $`{\displaystyle \underset{k0}{}}{\displaystyle \frac{1}{k!}}(_K\phi )(𝐲_1){\displaystyle }d\mathrm{\Omega }_1n_1^in_1^K\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{2k}{}^{}+12.`$
Hence the intrinsic expression of the distributional terms on $`𝒟`$,
$$\mathrm{D}_i[F]_{|_𝒟}=\underset{k0}{}\frac{()^k}{k!}_K\delta _1d\mathrm{\Omega }_1n_1^in_1^K\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{2k}{}^{}+12,$$
(138)
which agrees with the distributional part of the derivative of a function with tempered singularities in distribution theory. For example, we can write
$$\mathrm{D}_i\left[\frac{1}{r_1^3}\right]_{|_𝒟}=\frac{4\pi }{3}_i\delta _1.$$
(139)
However, when acting on functions of the full set $``$, the derivative generally leads to properties which have no equivalent in distributional theory. For instance, although the distributional derivative of $`1/r_1^2`$ reduces on $`𝒟`$ to the ordinary derivative, i.e. $`\mathrm{D}_i[\frac{1}{r_1^2}]_{|_𝒟}=0`$, on $``$ it does not:
$$_i\left(\mathrm{Pf}\frac{1}{r_1^2}\right)=\mathrm{Pf}\left(2\frac{n_1^i}{r_1^3}+4\pi n_1^i\delta _1\right).$$
(140)
For the distributional derivative of $`1/r_1^3`$ on $``$ we find
$$_i\left(\mathrm{Pf}\frac{1}{r_1^3}\right)=\mathrm{Pf}\left(3\frac{n_1^i}{r_1^4}+4\pi \frac{n_1^i}{r_1}\delta _1\right).$$
(141)
The expression of the distributional term is apparently different from the corresponding result (139) in distribution theory. However we shall see after learning how to differentiate the Dirac pseudo-function $`\mathrm{Pf}\delta _1`$ that the distributional term $`\mathrm{D}_i[\frac{1}{r_1^3}]`$ takes in fact the same form on $``$ as on $`𝒟`$ \[see (154) below\].
We come now to an important point. In this paper we have defined a “pointwise” product of pseudo-functions (see Definition 6), which reduces to the ordinary product in all the cases where the functions are regular enough. For instance, it coincides with the ordinary product for $`C^{\mathrm{}}`$ functions, or even continuous or locally integrable functions (adopting the class $`_{\mathrm{loc}}`$). Next, we introduced a derivative operator that acts merely as the ordinary derivative for a large class of not-too-singular functions (those which are bounded near the singularities, see Proposition 3). In particular, the derivative is equal to the ordinary one when the functions are $`C^1`$ at the location of the two singularities. However, we know from a theorem of Schwartz that it is impossible to define a multiplication for distributions having the previous properties and such that the distributional derivation satisfies the standard formula for the derivation of a product (Leibniz’s rule). In agreement with that theorem, we find that the derivative operator defined by (128)-(130) does not obey in general the Leibniz rule, whereas it does satisfy it by definition in an “integrated sense”, namely:
$$<_i[\mathrm{Pf}(FG)],1>=0=<_i(\mathrm{Pf}F)G+F_i(\mathrm{Pf}G),1>.$$
(142)
However it does not satisfy the Leibniz rule in a “local sense”, i.e. we have, generically for two functions $`F,G`$,
$$_i[\mathrm{Pf}(FG)]_i(\mathrm{Pf}F)GF_i(\mathrm{Pf}G)0.$$
(143)
This means that, a priori,
$$<_i[\mathrm{Pf}(FG)],H><_i(\mathrm{Pf}F),GH><_i(\mathrm{Pf}G),FH>0,$$
(144)
or, equivalently, since the Leibniz rule is satisfied by the ordinary derivative,
$$<\mathrm{D}_i[FG],H><\mathrm{D}_i[F],GH><\mathrm{D}_i[G],FH>0.$$
(145)
Actually, in accordance with the theorem in , (143) must be true even when the pseudo-function is regarded as a distribution on $`𝒟`$. To check this, let us compute the left side of (145) in the case where $`\mathrm{D}_i`$ is the particular solution $`\mathrm{D}_i^{\mathrm{part}}`$ defined by (130), and where $`H`$ is equal to some $`\phi 𝒟`$. We employ the Taylor expansion of $`\phi `$ around 1 and 2, and, strictly following the definition of the distributional term in (130), we arrive at
$`\left[\mathrm{D}_i^{\mathrm{part}}[FG]F\mathrm{D}_i^{\mathrm{part}}[G]G\mathrm{D}_i^{\mathrm{part}}[F]\right]_{|_𝒟}`$ (146)
$`={\displaystyle \underset{k1}{}}{\displaystyle \frac{()^k}{k!}}_K\delta _1{\displaystyle }d\mathrm{\Omega }_1n_1^in_1^K[{\displaystyle \frac{1}{2}}\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{1}{}^{}\stackrel{}{g}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{1k}{}^{}+{\displaystyle \frac{1}{2}}\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{1k}{}^{}\stackrel{}{g}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{1}{}^{}{\displaystyle \underset{j=0}{\overset{k}{}}}\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{1j}{}^{}\stackrel{}{g}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{jk1}{}^{}]+12.`$ (147)
The right side of (146) equals $`\frac{2\pi }{3}_i\delta _1`$ in the case where $`F=\frac{1}{r_1}`$ and $`G=\frac{1}{r_1^2}`$ for instance. It is not possible to add a homogeneous solution of the form (136) so as to get always zero. As the result (146) depends only on the singular coefficients of $`F`$ and $`G`$, we recover the Leibniz rule whenever $`F`$ or $`G`$ is bounded near the singularities. Besides, we can verify directly on (146) that the Leibniz rule is indeed true in an integrated sense, since the integral over $`^3`$ of (146) picks up only the term with $`k=0`$ which gives no contribution.
### C Derivative of the Dirac pseudo-function
In this subsection we compute the distributional term $`<\mathrm{D}_i[F],G>`$ given by the sum of (131) and (137) assuming that either $`F`$ or $`G`$ is equal to the Riesz delta-function $`{}_{\epsilon }{}^{}\delta _{1}^{}=\frac{\epsilon (\epsilon 1)}{4\pi }r_1^{\epsilon 3}`$ for some small $`\epsilon >0`$. (We come back for a moment to Definition 1 in which the powers of $`r_1`$ and $`r_2`$ in the expansions of $`F`$ or $`G`$ are real.) We notice first that the terms depending on the singular coefficients $`{}_{1}{}^{}f_{1}^{}`$ and $`{}_{1}{}^{}g_{1}^{}`$ are present only when the exponent $`1`$ belongs to both families of indices $`(a_i)_i`$ corresponding to $`F`$ and $`G`$ (remind Definition 1). This means that, choosing $`\epsilon `$ to be different from 2, these terms will not contribute to the present calculation, and in particular that the homogeneous part $`<\mathrm{H}_i[F],G>`$ will always give zero, provided that either $`F`$ or $`G`$ is equal to $`{}_{\epsilon }{}^{}\delta _{1}^{}`$. From the expression (131) we get
$`<\mathrm{D}_i[{}_{\epsilon }{}^{}\delta _{1}^{}],G>`$ $`=`$ $`\epsilon (1\epsilon ){\displaystyle }{\displaystyle \frac{d\mathrm{\Omega }_1}{4\pi }}n_1^i\stackrel{}{g}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{1\epsilon }{}^{},`$ (149)
$`<\mathrm{D}_i[F],{}_{\epsilon }{}^{}\delta _{1}^{}>`$ $`=`$ $`\epsilon (1\epsilon ){\displaystyle \underset{l0}{}}{\displaystyle \frac{()^l}{l!}}\stackrel{}{}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{L}{}^{}r_{12}^{\epsilon 3}{\displaystyle }{\displaystyle \frac{d\mathrm{\Omega }_2}{4\pi }}n_2^{iL}\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{2}}{}_{2l}{}^{}.`$ (150)
Furthermore, by choosing $`\epsilon `$ smaller than the spacing between some exponents $`a_i`$ of $`G`$ (specifically $`\epsilon <1a_{i_1}`$ with $`a_{i_1}`$ is such that $`a_{i_1}<1a_{i_1+1}`$) we can arrange for having $`{}_{1}{}^{}g_{1\epsilon }^{}=0`$ so that (149) becomes identically zero. Anyway, in the limit $`\epsilon 0`$ we come up formally with both relations $`<\mathrm{D}_i[\delta _1],G>=0`$ and $`<\mathrm{D}_i[F],\delta _1>=0`$. The former tells us that the distributional derivative of $`\mathrm{Pf}\delta _1`$ reduces to the ordinary one, i.e.
$$_i(\mathrm{Pf}\delta _1)=\mathrm{Pf}(_i\delta _1).$$
(151)
The latter \[that we already knew from (122)\] shows via the rule of integration by parts that the action of $`_i(\mathrm{Pf}\delta _1)`$ over any function $`F`$ is equal to minus the action of $`\mathrm{Pf}\delta _1`$ over the derivative $`_iF`$.
###### Definition 10
The derivative of the Dirac pseudo-function $`\mathrm{Pf}\delta _1`$ is defined by
$$F,<_i(\mathrm{Pf}\delta _1),F>=<\mathrm{Pf}\delta _1,_iF>(_iF)_1.$$
(152)
We can summarize the properties of the derivative of the Dirac pseudo-function by writing the successive identities
$$<_i(\mathrm{Pf}\delta _1),F>=<\mathrm{Pf}(_i\delta _1),F>=<\mathrm{Pf}\delta _1,_iF>=(_iF)_1,$$
as well as similar identities obtained by exchanging the roles of $`F`$ and $`\delta _1`$,
$$<_i(\mathrm{Pf}F),\delta _1>=<\mathrm{Pf}(_iF),\delta _1>=<\mathrm{Pf}F,_i\delta _1>=(_iF)_1.$$
###### Lemma 5
The intrinsic form of the derivative of the Dirac pseudo-function is
$$_i(\mathrm{Pf}\delta _1)=\mathrm{Pf}\left(3\frac{n_1^i}{r_1}\delta _1\right).$$
(153)
The proof is evident from using the identity (13). The form (153) \[with (151)\] is quite useful in practice; for instance, it permits us to re-write the derivative of the pseudo-function $`\mathrm{Pf}(\frac{1}{r_1^3})`$ as computed in (141) into the form
$$_i\left(\mathrm{Pf}\frac{1}{r_1^3}\right)=\mathrm{Pf}\left(3\frac{n_1^i}{r_1^4}\frac{4\pi }{3}_i\delta _1\right),$$
(154)
where the distributional term takes the same form as in the distribution theory \[compare with (139)\].
The preceding definition and lemma are easily extended to the case of the pseudo-functions $`\mathrm{Pf}(F\delta _1)`$. The derivative of these objects is defined by the mean of the relation
$$<_i[\mathrm{Pf}(F\delta _1)],G>=<\mathrm{Pf}(F\delta _1),_iG>=(F_iG)_1.$$
(155)
Then, from the identity $`(\text{13})`$, we readily get the intrinsic form
$$_i[\mathrm{Pf}(F\delta _1)]=\mathrm{Pf}\left[r_1^3_i\left(\frac{F}{r_1^3}\right)\delta _1\right].$$
(156)
Notice the interesting particular case
$$_i[\mathrm{Pf}(r_1^3\delta _1)]=0,$$
(157)
which is also an immediate consequence of (18). Finally, let us mention that the Leibniz rule happens to hold in the special case where one of the pseudo-functions is of the type $`\mathrm{Pf}(G\delta _1)`$, i.e.
$$_i[\mathrm{Pf}F.\mathrm{Pf}(G\delta _1)]=_i(\mathrm{Pf}F).\mathrm{Pf}(G\delta _1)+\mathrm{Pf}F._i[\mathrm{Pf}(G\delta _1)]$$
(158)
(the verification is straightforward).
## VIII Multiple derivatives
### A General construction
From Proposition 3 we can give a meaning to
$$<_i(\mathrm{Pf}F),G>=\mathrm{Pf}d^3𝐱_iFG+<\mathrm{D}_i[F],G>,$$
(159)
which will be also denoted $`<_i(\mathrm{Pf}F),\mathrm{Pf}G>`$. We now define the more complicated object $`<_i(\mathrm{Pf}F),_j(\mathrm{Pf}G)>`$. Since the distributional term $`\mathrm{D}_i[F]`$ has the form $`\mathrm{Pf}(H\delta _1)`$ plus $`12`$, and because (121)-(123) entail such identities as $`<\mathrm{Pf}(G\delta _1),\mathrm{Pf}(H\delta _1)>=0=<\mathrm{Pf}(G\delta _1),\mathrm{Pf}(H\delta _2)>`$, we deduce that the duality bracket applied on any two distributional terms is always zero:
$$F,G,<\mathrm{D}_i[F],\mathrm{D}_j[G]>=0.$$
(160)
When constructing the bracket $`<_i(\mathrm{Pf}F),_j(\mathrm{Pf}G)>`$ we shall meet a product of two distributional terms which gives zero by (160), and we shall be left only with the ordinary part as well as the two cross terms involving one distributional term. Therefore,
$`<_i(\mathrm{Pf}F),_j(\mathrm{Pf}G)>`$ $`=`$ $`\mathrm{Pf}{\displaystyle d^3𝐱_iF_jG}`$ (161)
$`+`$ $`<\mathrm{D}_i[F],_jG>+<\mathrm{D}_j[G],_iF>.`$ (162)
\[The ordinary part can equivalently be written as
$$\mathrm{Pf}d^3𝐱_iF_jG=<\mathrm{Pf}(_iF),\mathrm{Pf}(_jG)>=<\mathrm{Pf}(_iF),_jG>=<_iF,\mathrm{Pf}(_jG)>.]$$
We now intend to introduce the second-order derivative operator. The generalization to any $`l`$th-order derivative is straightforward and will be stated without proof. By extending the rule of integration by parts presented in Definition 9, we are led, quite naturally, to require that
$$F,G,<_{ij}(\mathrm{Pf}F),G>=<_j(\mathrm{Pf}F),_i(\mathrm{Pf}G)>,$$
(163)
where the object $`<_j(\mathrm{Pf}F),_i(\mathrm{Pf}G)>`$ has just been given in (161). For the moment, we are careful at distinguishing the order of the indices $`i`$ and $`j`$. Let us look for the expression of the distributional term $`\mathrm{D}_{ij}[F]`$ corresponding to the double derivative, viz
$$_{ij}(\mathrm{Pf}F)=\mathrm{Pf}(_{ij}F)+\mathrm{D}_{ij}[F],$$
(164)
in terms of the single-derivative term $`\mathrm{D}_i[F]`$. Inserting (164) into the required property (163) we arrive immediately at
$$<\mathrm{D}_{ij}[F],G>=\mathrm{Pf}d^3𝐱_i\left(_jFG\right)<\mathrm{D}_j[F],_iG><\mathrm{D}_i[G],_jF>.$$
Next recall the formula (132) which tells us that any partie-finie integral of a gradient is the sum of two distributional contributions. Using this property we obtain the simple result
$`<\mathrm{D}_{ij}[F],G>`$ $`=`$ $`<\mathrm{D}_i[_jF],G><\mathrm{D}_j[F],_iG>`$ (165)
$`=`$ $`<\mathrm{D}_i[_jF],G>+<_i\mathrm{D}_j[F],G>.`$ (166)
The formula (155) allowed us to obtain the second equality; so the intrinsic form of the second-order distributional term is obtained as
$$\mathrm{D}_{ij}[F]=\mathrm{D}_i[_jF]+_i\mathrm{D}_j[F].$$
(167)
This result is easily extendible to any multiple derivatives, demanding that, to any order $`l`$,
$$<_{i_1i_2\mathrm{}i_l}(\mathrm{Pf}F),G>=<_{i_2\mathrm{}i_l}(\mathrm{Pf}F),_{i_1}(\mathrm{Pf}G)>,$$
(168)
where the right side is obtained in a way similar to (161). We can even impose the more general rule of integration by parts, that for any $`k=1,\mathrm{},l`$,
$$<_{i_1i_2\mathrm{}i_l}(\mathrm{Pf}F),G>=()^k<_{i_{k+1}i_{k+2}\mathrm{}i_l}(\mathrm{Pf}F),_{i_ki_{k1}\mathrm{}i_1}(\mathrm{Pf}G)>.$$
(169)
Then the following is proved by induction over $`l`$.
###### Proposition 4
If a multi-derivative operator
$$_{i_1i_2\mathrm{}i_l}(\mathrm{Pf}F)=\mathrm{Pf}(_{i_1i_2\mathrm{}i_l}F)+\mathrm{D}_{i_1i_2\mathrm{}i_l}[F],$$
(170)
satisfies the rule of integration by parts $`(\text{168})`$ or $`(\text{169})`$, then the $`l`$-th order distributional term $`\mathrm{D}_{i_1i_2\mathrm{}i_l}[F]`$ is given in terms of the first-order $`\mathrm{D}_{i_k}[F]`$’s by
$$\mathrm{D}_{i_1i_2\mathrm{}i_l}[F]=\underset{k=1}{\overset{l}{}}_{i_1\mathrm{}i_{k1}}\mathrm{D}_{i_k}[_{i_{k+1}\mathrm{}i_l}F].$$
(171)
Recall that this result is valid for any distributional derivative of the form given by Proposition 3, i.e. $`\mathrm{D}_i[F]=\mathrm{D}_i^{\mathrm{part}}[F]+\mathrm{H}_i[F]`$. Therefore, the rule of integration by parts has permitted us to construct uniquely all higher-order derivatives from a given choice of first-order derivative $`\mathrm{D}_i[F]`$, i.e. from a given choice of “homogeneous” solution $`\mathrm{H}_i[F]`$. Notice that a priori this construction does not yield some commuting multi-derivatives (i.e. the Schwarz lemma is not valid in general), because evidently the right side of the formula (171) is not necessarily symmetric in all its indices. However, as a central result of this paper, we shall show now that it is possible to find an initial $`\mathrm{H}_i[F]`$ such that the derivatives do commute to any order.
###### Theorem 4
The most general derivative operator $`_i(\mathrm{Pf}F)=\mathrm{Pf}(_iF)+\mathrm{D}_i[F]`$ such that
(i) the distributional term $`\mathrm{D}_i[F]`$ depends only on the singular coefficients of $`F`$,
(ii) all multi-derivatives satisfy the rule of integration by parts,
(iii) all multi-derivatives commute (i.e. the $`\mathrm{D}_{i_1i_2\mathrm{}i_l}[F]`$’s are symmetric in $`i_1i_2\mathrm{}i_l`$),
is given by
$$\mathrm{D}_i[F]=4\pi \underset{l=0}{\overset{+\mathrm{}}{}}\mathrm{Pf}(C_l[n_1^{iL}\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{1}{}^{L}n_1^L\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{1}{}^{iL}]r_1\delta _1+\underset{k0}{}\frac{n_1^{iL}}{r_1^k}\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{2k}{}^{L}\delta _{1}^{})+12,$$
(172)
where the coefficients $`C_l=(l+1)\left[K+_{j=1}^l\frac{1}{j+1}\right]`$ depend on an arbitrary constant $`K`$.
(Actually the theorem states that the derivative operator depends a priori on two different constants $`K_1`$ and $`K_2`$ for each of the two singularities. In the following we shall assume for simplicity that the constants are the same, so that the way to differentiate does not distinguish between the different singularities.) Notice that $`\mathrm{D}_i[F]`$ differs from the particular solution $`\mathrm{D}_i^{\mathrm{part}}[F]`$ given by (130) only in the terms depending on the “least singular” coefficients $`{}_{1}{}^{}f_{1}^{}`$ and $`{}_{2}{}^{}f_{1}^{}`$.
Proof. According to the assumptions (i) and (ii) we already know (see Proposition 3 and Lemma 4) that the distributional term must be of the form $`\mathrm{D}_i[F]=\mathrm{D}_i^{\mathrm{part}}[F]+\mathrm{H}_i[F]`$, where the particular solution is given explicitly by (130), and where the homogeneous term takes the form (136) depending on a set of arbitrary coefficients $`\alpha _l`$. Furthermore, we know from Proposition 4 that all higher-order derivatives are generated from the first-order one in the way specified by (171). It only remains to show that the coefficients $`\alpha _l`$ can be computed in order that the assumption (iii) of commutation of derivatives be fulfilled, and that the derivative is given by (172).
What we want then is to impose the symmetry of $`\mathrm{D}_{ij}[F]`$ in $`ij`$. We compute the anti-symmetric projection $`[ij]\frac{ijji}{2}`$ of the second-order distributional term associated with the particular solution (130),
$$\mathrm{D}_{[ij]}^{\mathrm{part}}[F]=\mathrm{D}_{[i}^{\mathrm{part}}[_{j]}F]+_{[i}\mathrm{D}_{j]}^{\mathrm{part}}[F].$$
(173)
The first term is readily obtained from (14) which tells us that the $`a`$th coefficient in the $`r_1`$-expansion of the gradient is $`{}_{1}{}^{}f_{a}^{}(_jF)=(a+1)n_1^j{}_{1}{}^{}f_{a+1}^{}+d_1^j{}_{1}{}^{}f_{a+1}^{}`$. On the other hand, the second term in (173) comes directly from using the formula (156). It follows that the anti-symmetric projection depends only on the expansion coefficients $`{}_{1}{}^{}f_{0}^{}`$, $`{}_{1}{}^{}f_{1}^{}`$ and $`12`$ through the simple formula,
$$\mathrm{D}_{[ij]}^{\mathrm{part}}[F]=2\pi \mathrm{Pf}\left(n_1^{[i}[r_1d_1^{j]}\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{0}{}^{}+d_1^{j]}\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{1}{}^{}]\delta _1\right)+12,$$
(174)
or, using the relation (15) for the operator $`d_1^j`$,
$$\mathrm{D}_{[ij]}^{\mathrm{part}}[F]=2\pi \underset{l=0}{\overset{+\mathrm{}}{}}(l+1)\mathrm{Pf}\left(n_1^{L[i}[r_1\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{0}{}^{j]L}+\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{1}{}^{j]L}]\delta _1\right)+12.$$
(175)
Note that by applying this on any $`G`$, we get
$$<\mathrm{D}_{[ij]}^{\mathrm{part}}[F],G>=2\pi \underset{l=0}{\overset{+\mathrm{}}{}}\frac{(l+1)(l+1)!}{(2l+3)!!}(\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{0}{}^{L[i}\stackrel{}{\widehat{g}}\genfrac{}{}{0pt}{}{}{1}{}_{1}{}^{j]L}+\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{1}{}^{L[i}\stackrel{}{\widehat{g}}\genfrac{}{}{0pt}{}{}{1}{}_{0}{}^{j]L})+12.$$
Next, we add the homogeneous solution. By performing a computation similar as the previous one (but a bit more involved) we find, based on the expression (136),
$$\mathrm{H}_{[ij]}[F]=\underset{l=0}{\overset{+\mathrm{}}{}}\frac{l+1}{2l+3}[(l+2)\alpha _l(l+1)\alpha _{l+1}]\mathrm{Pf}\left(n_1^{[i}[r_1\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{0}{}^{j]L}+\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{1}{}^{j]L}]\delta _1\right)+12.$$
(176)
Remarkably, $`\mathrm{H}_{[ij]}[F]`$ takes exactly the same form as (175). Hence, we are able to determine a relation to be satisfied by the looked-for coefficients $`\alpha _l`$ for any $`l`$ in order that the non-commuting part (175) associated to the particular solution be cancelled out by that of the homogeneous one: $`l`$, $`(l+2)\alpha _l(l+1)\alpha _{l+1}=2\pi (2l+3)`$. Given any initial value for $`\alpha _0`$ the solution reads as
$$\alpha _l=(l+1)\left[\alpha _0+2\pi \underset{j=1}{\overset{l}{}}\left(\frac{1}{j}+\frac{1}{j+1}\right)\right]=2\pi +4\pi (l+1)\left[K+\underset{j=1}{\overset{l}{}}\frac{1}{j+1}\right],$$
(177)
in which we have introduced the new arbitrary constant $`K=\frac{\alpha _0}{4\pi }+\frac{1}{2}`$. Inserting (177) back into the expression for $`\mathrm{D}_i[F]`$ leads to the announced result (172). At last, we find that for any choice of the constant $`K`$ the second-derivative operator commutes, i.e.
$$\mathrm{D}_{[ij]}[F]=\mathrm{H}_{[ij]}[F]+\mathrm{D}_{[ij]}^{\mathrm{part}}[F]=0.$$
(178)
Let us verify from (178) that all higher-order multi-derivative operators commute as well, i.e. $`\mathrm{D}_{i_1i_2\mathrm{}i_l}[F]`$ given by the formula (171) is symmetric in all its indices. This is easily proved by induction over $`l`$. Suppose that to the $`(l1)`$th order $`\mathrm{D}_{i_1i_2\mathrm{}i_{l1}}[F]`$ is symmetric, and re-write the formula (171) into both forms
$`\mathrm{D}_{i_1i_2\mathrm{}i_l}[F]`$ $`=`$ $`\mathrm{D}_{i_1}[_{i_2\mathrm{}i_l}F]+_{i_1}\mathrm{D}_{i_2\mathrm{}i_l}[F]`$
$`=`$ $`\mathrm{D}_{i_1\mathrm{}i_{l1}}[_{i_l}F]+_{i_1\mathrm{}i_{l1}}\mathrm{D}_{i_l}[F].`$
Clearly, $`\mathrm{D}_{i_1\mathrm{}i_l}[F]`$ is symmetric with respect to both $`i_1\mathrm{}i_{l1}`$ and $`i_2\mathrm{}i_l`$, so must be symmetric in all its indices (the symmetry with respect to the first and last indices being a consequence of the other symmetries). QED.
We should mention that the dependence upon the arbitrary constant $`K`$ of the derivative operator defined by Theorem 4 is
$$\mathrm{D}_i[F]_{|_K}=4\pi K\underset{l=0}{\overset{+\mathrm{}}{}}(l+1)\mathrm{Pf}\left([n_1^{iL}\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{1}{}^{L}n_1^L\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{1}{}^{iL}]r_1\delta _1\right)+12,$$
(179)
which can also be cast into the more interesting form
$$\mathrm{D}_i[F]_{|_K}=4\pi K_i\left[\mathrm{Pf}(r_1^2\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{1}{}^{}\delta _{1}^{})\right]+12.$$
(180)
We see that the “ambiguity” linked with the constant $`K`$ when deriving the pseudo-function $`\mathrm{Pf}F`$ is related to an ambiguity resulting from the addition of the term $`4\pi K\mathrm{Pf}(r_1^2{}_{1}{}^{}f_{1}^{}\delta _1)+12`$ to $`\mathrm{Pf}F`$. In a sense, one can also view the constant $`K`$ as a measure of how much the distributional derivative of the pseudo-function $`\mathrm{Pf}(1/r_1)`$ differs from the ordinary one, i.e.
$$\mathrm{D}_i\left[\frac{1}{r_1}\right]=4\pi K\mathrm{Pf}\left(n_1^ir_1\delta _1\right).$$
(181)
Indeed, for functions which are more singular than a simple $`1/r_1`$, there is no dependence on the constant $`K`$; see e.g. (140)-(141).
### B The Laplacian operator
Let us compute the second-derivative of $`\mathrm{Pf}(1/r_1)`$ using the formula $`\mathrm{D}_{ij}[1/r_1]=\mathrm{D}_i[n_1^j/r_1^2]+_i\mathrm{D}_j[1/r_1]`$. The first term is obtained directly from the definition (172), and the second term is computed with the help of the formula (156) applied on (181). As a result, we get
$$\mathrm{D}_{ij}\left[\frac{1}{r_1}\right]=\frac{4\pi }{3}\mathrm{Pf}\left(\left[\delta ^{ij}+3(3K+1)\widehat{n}_1^{ij}\right]\delta _1\right),$$
(182)
where $`\widehat{n}_1^{ij}=n_1^in_1^j\frac{1}{3}\delta ^{ij}`$. Evidently (because of the trace-free $`\widehat{n}_1^{ij}`$), when we restrict ourselves to smooth functions of the set $`𝒟`$, we recover the usual formula of distributional theory,
$$\mathrm{D}_{ij}\left[\frac{1}{r_1}\right]_{|_𝒟}=\frac{4\pi }{3}\delta ^{ij}\delta _1.$$
(183)
Since the dependence over $`K`$ in (182) drops out when taking the trace over the indices $`ij`$, we have $`\mathrm{D}_{ii}[1/r_1]=4\pi \mathrm{Pf}\delta _1`$ (even on the set $``$). This means that the Laplacian of $`1/r_1`$ on $``$ takes the same form as the well-known formula of distribution theory:
$$\mathrm{\Delta }\left(\mathrm{Pf}\frac{1}{r_1}\right)=4\pi \mathrm{Pf}\delta _1.$$
(184)
We infer from the rule of integration by parts that
$$<\mathrm{\Delta }(\mathrm{Pf}F),\frac{1}{r_1}>=<\mathrm{\Delta }\left(\mathrm{Pf}\frac{1}{r_1}\right),F>=4\pi (F)_1,$$
(185)
which can be phrased by saying that the Poisson integral of the Laplacian of a singular function, as evaluated at a singular point, is equal to the partie finie of the function at that point. More generally, the Laplacian acting on any pseudo-function in $`^{}`$ is defined by
$$\mathrm{\Delta }(\mathrm{Pf}F)=\mathrm{Pf}(\mathrm{\Delta }F)+\mathrm{D}_{ii}[F],$$
(186)
where the distributional term is given by
$$\mathrm{D}_{ii}[F]=_i\mathrm{D}_i[F]+\mathrm{D}_i[_iF].$$
(187)
###### Proposition 5
Under the hypothesis of Theorem 4 the distributional term associated with the Laplacian operator reads
$$\mathrm{D}_{ii}[F]=4\pi \underset{l=0}{\overset{+\mathrm{}}{}}\mathrm{Pf}((l+1)C_{l1}n_1^L[\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{1}{}^{L}+r_1\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{0}{}^{L}]\delta _1\underset{k0}{}(2k+1)\frac{n_1^L}{r_1^k}\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{1k}{}^{L}\delta _{1}^{})+12.$$
(188)
The proof is straightforward and will not be detailed. Note that the dependence on $`K`$ occurs only for functions owing some non-zero coefficients $`{}_{1}{}^{}f_{1}^{}`$ or $`{}_{1}{}^{}f_{0}^{}`$, or $`12`$; for instance
$`\mathrm{D}_{ii}[n_1^j]`$ $`=`$ $`8\pi K\mathrm{Pf}\left(n_1^ir_1\delta _1\right),`$
$`\mathrm{D}_{ii}[n_1^j/r_1]`$ $`=`$ $`8\pi (K{\displaystyle \frac{1}{2}})\mathrm{Pf}\left(n_1^i\delta _1\right).`$
But, for more singular functions like $`1/r_1^3`$, we have
$$\mathrm{\Delta }\left(\mathrm{Pf}\frac{1}{r_1^3}\right)=\mathrm{Pf}\left(\frac{6}{r_1^5}\frac{20\pi }{r_1^2}\delta _1\right).$$
(189)
###### Lemma 6
The Laplacian of the pseudo-function $`\mathrm{Pf}(F\delta _1)`$ is given by
$$\mathrm{\Delta }\left[\mathrm{Pf}(F\delta _1)\right]=\mathrm{Pf}\left(r_1^3\mathrm{\Delta }\left[\frac{F}{r_1^3}\right]\delta _1\right).$$
(190)
The proof is similar to the one of the formula (156). Two immediate particular applications are
$`\mathrm{\Delta }\left(\mathrm{Pf}\delta _1\right)=\mathrm{Pf}\left({\displaystyle \frac{6}{r_1^2}}\delta _1\right),`$ (192)
$`\mathrm{\Delta }\left[\mathrm{Pf}(r_1^2\delta _1)\right]=0,`$ (193)
which can also be deduced respectively from (21) and (19). \[(192) is in agreement with (115).\] Let us add that
$$\mathrm{\Delta }\left[\mathrm{Pf}\left(\frac{r_1}{2}\delta _1\right)\right]=\mathrm{Pf}\left(\frac{\delta _1}{r_1}\right).$$
(194)
In practice, Lemma 6 may be used to determine some solutions of Poisson equations “in the sense of distributions” on $``$. For instance combining (192) with the formula (189), we can write
$$\mathrm{\Delta }\left[\mathrm{Pf}\left(\frac{1}{r_1^3}+\frac{10\pi }{3}\delta _1\right)\right]=\mathrm{Pf}\left(\frac{6}{r_1^5}\right),$$
(195)
which provides a solution of the Poisson equation with source $`\mathrm{Pf}(6/r_1^5)`$ in the sense of these distributions. Such a solution is by no means unique, since, from Lemma 6, one can add to it any “homogeneous” solution of the form $`\mathrm{Pf}(H^{\mathrm{hom}}\delta _1)`$ where $`H^{\mathrm{hom}}`$ is the product of $`r_1^3`$ with an arbitrary solution of the Laplace equation. Notice that (195) as it stands is well-defined in distribution theory and so takes the same form when restricted to $`𝒟`$ ($`\mathrm{\Delta }\delta _1`$ is meaningful on this set). However,
$$\mathrm{\Delta }\left[\mathrm{Pf}\left(\frac{1}{r_1^2}+6\pi r_1\delta _1\right)\right]=\mathrm{Pf}\left(\frac{2}{r_1^4}\right)$$
(196)
has no equivalent in distribution theory.
## IX Time derivative and partial derivatives
The functions $`F`$ depend on the field point $`𝐱`$ and on the two singular source points $`𝐲_1`$ and $`𝐲_2`$. We shall now consider the situation where the two source points represent the trajectories of actual particles, and therefore depend on time $`t`$. We assume that the two trajectories $`𝐲_1(t)`$ and $`𝐲_2(t)`$ are smooth, that is $`𝐲_1,𝐲_2C^{\mathrm{}}()`$. In general (e.g. in the application to the problem of motion of point-particles) the function $`F`$ will also depend on time through the two velocities $`𝐯_1(t)=d𝐲_1(t)/dt`$ and $`𝐯_2(t)=d𝐲_2(t)/dt`$. We suppose that $`F`$ is a smooth functional of $`𝐯_1`$ and $`𝐯_2`$. Therefore, in this section, $`F`$ is supposed to take the form
$$F=F(𝐱,t;𝐲_1(t),𝐲_2(t)).$$
(197)
We want to investigate the partial derivatives (in a distributional sense) of the pseudo-function $`\mathrm{Pf}F`$ with respect to the source points $`𝐲_1`$ and $`𝐲_2`$, as well as the derivative of $`\mathrm{Pf}F`$ with respect to time $`t`$. Obviously, the partial derivatives $`{}_{1}{}^{}_{i}^{}/𝐲_1`$ and $`12`$ are closely related to the time derivative $`_t/t`$ on account of the fact that
$$_tF=\dot{F}+v_1^i\stackrel{}{}\genfrac{}{}{0pt}{}{}{1}{}_{i}{}^{}F+v_2^i\stackrel{}{}\genfrac{}{}{0pt}{}{}{2}{}_{i}{}^{}F,$$
(198)
(in the ordinary sense), where $`\dot{F}`$ denotes the contribution of the time-derivative due to the dependence over the velocities, i.e. $`\dot{F}=a_1^iF/v_1^i+a_2^iF/v_2^i`$ ($`a_1^i`$ and $`a_2^i`$ denoting the two accelerations). In applications it is frequent that $`F`$ depends on the trajectories only through the two distances to the field point $`𝐫_1=𝐱𝐲_1`$ and $`𝐫_2=𝐱𝐲_2`$; in that case
$$_iF+\stackrel{}{}\genfrac{}{}{0pt}{}{}{1}{}_{i}{}^{}F+\stackrel{}{}\genfrac{}{}{0pt}{}{}{2}{}_{i}{}^{}F=0.$$
(199)
The general function (197) does not necessarily satisfy the latter identity. However, let us guess from (199) the result for the distributional terms $`{}_{1}{}^{}\mathrm{D}_{i}^{}[F]`$ (and $`12`$) associated with the partial derivative $`{}_{1}{}^{}_{i}^{}`$ acting on the pseudo-function $`\mathrm{Pf}F`$. Since we have supposed that the dependence of $`F`$ on the velocities is smooth, the distributional terms will depend only on that part of the function which becomes singular when $`r_10`$, and so, because as far as the singular part is concerned, the function behaves like (199), the distributional terms $`{}_{1}{}^{}\mathrm{D}_{i}^{}[F]`$ and $`{}_{2}{}^{}\mathrm{D}_{i}^{}[F]`$ should satisfy
$$\mathrm{D}_i[F]+\stackrel{}{\mathrm{D}}\genfrac{}{}{0pt}{}{}{1}{}_{i}{}^{}[F]+\stackrel{}{\mathrm{D}}\genfrac{}{}{0pt}{}{}{2}{}_{i}{}^{}[F]=0.$$
(200)
Now, from Theorem 4, we know that $`\mathrm{D}_i[F]`$ can be naturally split into two parts associated respectively with the singularities 1 and 2. Therefore, we expect that the correct distributional term $`{}_{1}{}^{}\mathrm{D}_{i}^{}[F]`$ is equal to minus that part of $`\mathrm{D}_i[F]`$ which corresponds to 1. Namely, using (172),
$$\stackrel{}{\mathrm{D}}\genfrac{}{}{0pt}{}{}{1}{}_{i}{}^{}[F]=4\pi \underset{l=0}{\overset{+\mathrm{}}{}}\mathrm{Pf}(C_l[n_1^{iL}\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{1}{}^{L}n_1^L\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{1}{}^{iL}]r_1\delta _1+\underset{k0}{}\frac{n_1^{iL}}{r_1^k}\stackrel{}{\widehat{f}}\genfrac{}{}{0pt}{}{}{1}{}_{2k}{}^{L}\delta _{1}^{})$$
(201)
(and idem for 2). This expectation is confirmed by the following definition and proposition.
###### Definition 11
The partial derivative $`\stackrel{}{}\genfrac{}{}{0pt}{}{}{1}_i`$ (and $`12`$) acting on pseudo-functions is said to satisfy the rule of integration by parts iff
$$F,G,<\stackrel{}{}\genfrac{}{}{0pt}{}{}{1}{}_{i}{}^{}(\mathrm{Pf}F),G>+<\stackrel{}{}\genfrac{}{}{0pt}{}{}{1}{}_{i}{}^{}(\mathrm{Pf}G),F>=\stackrel{}{}\genfrac{}{}{0pt}{}{}{1}{}_{i}{}^{}[<\mathrm{Pf}F,G>].$$
(202)
Similarly, the time derivative $`_t`$ is said to satisfy the rule of integration by parts iff
$$<_t(\mathrm{Pf}F),G>+<_t(\mathrm{Pf}G),F>=\frac{d}{dt}[<\mathrm{Pf}F,G>].$$
(203)
Notice that $`<\mathrm{Pf}F,G>=\mathrm{Pf}d^3𝐱FG`$ is a function of the source points $`𝐲_1(t)`$ and $`𝐲_2(t)`$, as well as $`t`$ independently if either $`F`$ or $`G`$ depends on the velocities. The time derivative in the right side of (203) means the total time derivative we get by taking into account both the variable $`t`$ occuring through $`𝐲_1(t)`$ and $`𝐲_1(t)`$, and the independent $`t`$ coming from the velocities. Let us now state a result analogous to Theorem 4, whose proof will not be given since it represents a simple adaptation of the one of that theorem.
###### Proposition 6
Under the hypothesis of Theorem 4 the partial derivative with respect to $`𝐲_1`$ (and idem with $`12`$) is determined as
$$\stackrel{}{}\genfrac{}{}{0pt}{}{}{1}{}_{i}{}^{}(\mathrm{Pf}F)=\mathrm{Pf}(\stackrel{}{}\genfrac{}{}{0pt}{}{}{1}{}_{i}{}^{}F)+\stackrel{}{\mathrm{D}}\genfrac{}{}{0pt}{}{}{1}{}_{i}{}^{}[F],$$
(204)
where $`{}_{1}{}^{}\mathrm{D}_{i}^{}[F]`$ is given by $`(\text{201})`$. And the time derivative is determined as
$$_t(\mathrm{Pf}F)=\mathrm{Pf}(_tF)+\mathrm{D}_t[F],$$
(205)
where $`\mathrm{D}_t[F]`$ is given by
$$\mathrm{D}_t[F]=v_1^i\stackrel{}{\mathrm{D}}\genfrac{}{}{0pt}{}{}{1}{}_{i}{}^{}[F]+v_2^i\stackrel{}{\mathrm{D}}\genfrac{}{}{0pt}{}{}{2}{}_{i}{}^{}[F].$$
(206)
Higher-order derivatives are constructed as in Section VIII. We find for instance
$$\stackrel{}{}\genfrac{}{}{0pt}{}{}{1}{}_{ij}{}^{}(\mathrm{Pf}F)=\mathrm{Pf}(\stackrel{}{}\genfrac{}{}{0pt}{}{}{1}{}_{ij}{}^{}F)+\stackrel{}{\mathrm{D}}\genfrac{}{}{0pt}{}{}{1}{}_{i}{}^{}[\stackrel{}{}\genfrac{}{}{0pt}{}{}{1}{}_{j}{}^{}F]+\stackrel{}{}\genfrac{}{}{0pt}{}{}{1}{}_{i}{}^{}\stackrel{}{\mathrm{D}}\genfrac{}{}{0pt}{}{}{1}{}_{j}{}^{}[F],$$
(207)
Idem for the second-order time derivative, which reads as
$$_t^2(\mathrm{Pf}F)=\mathrm{Pf}(_t^2F)+\mathrm{D}_t[_tF]+_t\mathrm{D}_t[F],$$
(208)
where $`_tF`$ is given by (198) and $`\mathrm{D}_t[F]`$ is defined in (206). Furthermore the mixing up of derivatives of different type is allowed, and proceeds in the expected way. E.g.
$$\stackrel{}{}\genfrac{}{}{0pt}{}{}{1}{}_{i}{}^{}\stackrel{}{}\genfrac{}{}{0pt}{}{}{2}{}_{j}{}^{}(\mathrm{Pf}F)=\mathrm{Pf}(\stackrel{}{}\genfrac{}{}{0pt}{}{}{1}{}_{i}{}^{}\stackrel{}{}\genfrac{}{}{0pt}{}{}{2}{}_{j}{}^{}F)+\stackrel{}{\mathrm{D}}\genfrac{}{}{0pt}{}{}{1}{}_{i}{}^{}[\stackrel{}{}\genfrac{}{}{0pt}{}{}{2}{}_{j}{}^{}F]+\stackrel{}{}\genfrac{}{}{0pt}{}{}{1}{}_{i}{}^{}\stackrel{}{\mathrm{D}}\genfrac{}{}{0pt}{}{}{2}{}_{j}{}^{}[F].$$
(209)
Another example is
$$_t_{ij}(\mathrm{Pf}F)=\mathrm{Pf}(_t_{ij}F)+\mathrm{D}_t[_{ij}F]+_t\mathrm{D}_i[_jF]+_t_i\mathrm{D}_j[F].$$
(210)
###### Acknowledgements.
The authors are grateful to Antoine Sellier for discussion and his interesting comments. This work was supported in part by the National Science Foundation under Grant no PHY-9900776.
## A Proof of Theorem 2
Basically the proof establishes the legitimacy of commuting some discrete series with integrals. Consider $`F`$. We start by evaluating the integrals
$`\mathrm{Pf}_{\genfrac{}{}{0pt}{}{\beta 0}{\alpha 0}}{\displaystyle _{_1(s)}}d^3𝐱{\displaystyle \underset{a3}{}}\left({\displaystyle \frac{r_1}{s_1}}\right)^\alpha \left({\displaystyle \frac{r_2}{s_2}}\right)^\beta r_1^a\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{a}{}^{}(𝐧_1)`$ (A2)
$`\mathrm{and}`$ $`\mathrm{Pf}_{\genfrac{}{}{0pt}{}{\alpha 0}{\beta 0}}{\displaystyle _{_1(s)}}d^3𝐱{\displaystyle \underset{a3}{}}\left({\displaystyle \frac{r_1}{s_1}}\right)^\alpha \left({\displaystyle \frac{r_2}{s_2}}\right)^\beta r_1^a\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{a}{}^{}(𝐧_1),`$ (A3)
where the $`{}_{1}{}^{}f_{a}^{}`$’s are the coefficients of the expansion of $`F`$ when $`r_10`$, and where $`_1(s)`$ is the ball centred on $`𝐲_1`$ and of radius $`s^+`$ (chosen to be $`s<r_{12}`$). From the definition of the class $``$ the sums over $`a`$ in (A) are finite. When the real part of $`\beta `$ is such that $`0\mathrm{}(\beta )1`$, the integrand of (A2) is majored by
$$\left(\frac{r_1}{s_1}\right)^{\mathrm{}(\alpha )}\left(\frac{r_2}{s_2}\right)^{\mathrm{}(\beta )}\underset{a3}{}r_1^a|\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{a}{}^{}(𝐧_1)|\left(\frac{r_1}{s_1}\right)^{\mathrm{}(\alpha )}\mathrm{max}(1,\frac{r_2}{s_2})\underset{a3}{}r_1^a|\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{a}{}^{}(𝐧_1)|,$$
which can be integrated on $`_1(s)`$. Thus the theorem of dominated convergence of an integral can be applied, with the result that
$`\mathrm{Pf}_{\genfrac{}{}{0pt}{}{\beta 0}{\alpha 0}}{\displaystyle _{_1(s)}}d^3𝐱{\displaystyle \underset{a3}{}}\left({\displaystyle \frac{r_1}{s_1}}\right)^\alpha \left({\displaystyle \frac{r_2}{s_2}}\right)^\beta r_1^a\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{a}{}^{}=\mathrm{Pf}_{\alpha 0}{\displaystyle _{_1(s)}}d^3𝐱{\displaystyle \underset{a3}{}}\left({\displaystyle \frac{r_1}{s_1}}\right)^\alpha r_1^a\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}_a`$
$`={\displaystyle \underset{a+3<0}{}}{\displaystyle \frac{s^{a+3}}{a+3}}{\displaystyle }d\mathrm{\Omega }_1\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{a}{}^{}+\mathrm{ln}\left({\displaystyle \frac{s}{s_1}}\right){\displaystyle }d\mathrm{\Omega }_1\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{3}{}^{}.`$
The second integral is more difficult to compute because the limit $`\alpha 0`$ does not commute with the integration sign. We must expand $`r_2^\beta `$ as a power series of $`r_1`$. $`r_1<r_{12}`$,
$$r_2^\beta =r_{12}^\beta (1+2𝐧_1.𝐧_{12}\frac{r_1}{r_{12}}+\frac{r_1^2}{r_{12}^2})^{\beta /2}=r_{12}^\beta \underset{l=0}{\overset{+\mathrm{}}{}}C_l^{\beta /2}(𝐧_1.𝐧_{12})\left(\frac{r_1}{r_{12}}\right)^l,$$
(A4)
where $`𝐧_{12}=(𝐲_1𝐲_2)/r_{12}`$, and where $`C_l^\lambda (t)`$ denotes the Gegenbauer polynomial, which is by definition the coefficient of $`x^l`$ in the power-series expansion of the function $`(12tx+x^2)^\lambda `$ when $`x0`$ (with $`\lambda `$, $`t`$). See e.g. Morse and Feshbach p. 602. When $`t`$ and is such that $`|t|1`$ (as is the case here since $`t=𝐧_1.𝐧_{12}`$), we can obtain a majoration of the Gegenbauer polynomial. From the formula (cf Gradshteyn and Ryzhik p. 1030)
$$C_l^\lambda (\mathrm{cos}\theta )=\underset{\genfrac{}{}{0pt}{}{k,h0}{k+h=l}}{}\frac{\mathrm{\Gamma }(\lambda +k)\mathrm{\Gamma }(\lambda +h)}{k!h![\mathrm{\Gamma }(\lambda )]^2}\mathrm{cos}[(kh)\theta ],$$
we find that $`l0`$, $`|C_l^\lambda (\mathrm{cos}\theta )|`$ is always less than
$$\underset{\genfrac{}{}{0pt}{}{k,h0}{k+h=l}}{}\frac{(|\lambda |+k1)(|\lambda |+k2)\mathrm{}|\lambda |(|\lambda |+h1)(|\lambda |+h2)\mathrm{}|\lambda |}{k!h!}=C_l^{|\lambda |}(1).$$
Therefore the series $`_l|C_l^{\beta /2}(𝐧_1.𝐧_{12}))(r_1/r_{12})^l|`$ is bounded by $`(12r_1/r_{12}+r_1^2/r_{12}^2)^{|\beta |/2}`$, and thus admits a limit. Thus (A4) converges absolutly and (when $`\mathrm{}(\alpha )`$ is large enough) the signs $``$ and $``$ can be interchanged:
$$_{_1(s)}d^3𝐱\underset{a+30}{}\left(\frac{r_1}{s_1}\right)^\alpha \left(\frac{r_2}{s_2}\right)^\beta r_1^a\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{a}{}^{}=\left(\frac{r_{12}}{s_2}\right)^\beta \underset{l=0}{\overset{+\mathrm{}}{}}\underset{a+30}{}_{_1(s)}d^3𝐱\frac{r_1^{\alpha +a+l}}{s_1^\alpha r_{12}^l}\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{a}{}^{}C_{l}^{\beta /2},$$
where $`C_l^{\beta /2}C_l^{\beta /2}(𝐧_1.𝐧_{12})`$. We obtain the two terms
$`\left({\displaystyle \frac{r_{12}}{s_2}}\right)^\beta {\displaystyle \underset{l=0}{\overset{+\mathrm{}}{}}}{\displaystyle \underset{\genfrac{}{}{0pt}{}{a+30}{a+l+30}}{}}{\displaystyle \frac{s^{\alpha +a+l+3}}{s_1^\alpha r_{12}^l(\alpha +a+l+3)}}{\displaystyle 𝑑\mathrm{\Omega }_1}\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{a}{}^{}C_{l}^{\beta /2}`$
$`+\left({\displaystyle \frac{r_{12}}{s_2}}\right)^\beta {\displaystyle \underset{\genfrac{}{}{0pt}{}{l=0}{\mathrm{finite}\mathrm{sum}}}{\overset{+\mathrm{}}{}}}{\displaystyle \frac{1}{\alpha }}\left({\displaystyle \frac{s}{s_1}}\right)^\alpha {\displaystyle \frac{1}{r_{12}^l}}{\displaystyle 𝑑\mathrm{\Omega }_1}\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{l3}{}^{}C_{l}^{\beta /2}.`$
The finite part when $`\alpha 0`$ of the second term reads simply
$$\left(\frac{r_{12}}{s_2}\right)^\beta \mathrm{ln}\left(\frac{s}{s_1}\right)\underset{l=0}{\overset{+\mathrm{}}{}}\frac{1}{r_{12}^l}𝑑\mathrm{\Omega }_1\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{l3}{}^{}C_{l}^{\beta /2}.$$
(A5)
On the other hand, in order to treat the first term, we must justify the commutation of the finite part with the infinite sum. Consider the series
$$\underset{l=0}{\overset{+\mathrm{}}{}}\frac{1}{\alpha +a+l+3}\left(\frac{s}{r_{12}}\right)^l𝑑\mathrm{\Omega }_1\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{a}{}^{}C_{l}^{\beta /2}.$$
For $`\alpha `$ in a disk of the complex plane centered on $`0`$ and of radius $`ϵ`$ (with $`0<ϵ<1`$), we can bound the generic term of that series (for large enough $`l`$) by
$$\frac{1}{|a+l+3|ϵ}\left(\frac{s}{r_{12}}\right)^l|d\mathrm{\Omega }_1\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{a}{}^{}C_{l}^{\beta /2}|,$$
which is independent of $`\alpha `$, and whose corresponding series in $`l`$ converges. Therefore we can apply the limit $`\alpha 0`$ through the summation over $`l`$ and deduce
$`\underset{\alpha 0}{lim}\left\{\left({\displaystyle \frac{r_{12}}{s_2}}\right)^\beta {\displaystyle \underset{l=0}{\overset{+\mathrm{}}{}}}{\displaystyle \underset{\genfrac{}{}{0pt}{}{a+30}{a+l+30}}{}}{\displaystyle \frac{s^{\alpha +a+l+3}}{s_1^\alpha r_{12}^l(\alpha +a+l+3)}}{\displaystyle 𝑑\mathrm{\Omega }_1}\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{a}{}^{}C_{l}^{\beta /2}\right\}`$ (A6)
$`=\left({\displaystyle \frac{r_{12}}{s_2}}\right)^\beta {\displaystyle \underset{a+30}{}}s^{a+3}{\displaystyle \underset{\genfrac{}{}{0pt}{}{l=0}{a+l+30}}{\overset{+\mathrm{}}{}}}{\displaystyle \frac{1}{a+l+3}}\left({\displaystyle \frac{s}{r_{12}}}\right)^l{\displaystyle 𝑑\mathrm{\Omega }_1}\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{a}{}^{}C_{l}^{\beta /2}.`$ (A7)
Next we apply the finite part $`\mathrm{Pf}_{\beta 0}`$ to the sum of (A5) and (A6), which involves finding the limit when $`\beta 0`$ of the series
$$\underset{\genfrac{}{}{0pt}{}{l=0}{a+l+30}}{\overset{+\mathrm{}}{}}\frac{1}{a+l+3}\left(\frac{s}{r_{12}}\right)^l𝑑\mathrm{\Omega }_1\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{a}{}^{}C_{l}^{\beta /2}.$$
(A8)
In any case the absolute value of the quantity under the sign $``$ is smaller than
$$\left(\frac{s}{r_{12}}\right)^lC_l^{|\beta |/2}(1)d\mathrm{\Omega }_1|\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{a}{}^{}|.$$
Furthermore we know that $`C_l^\lambda (1)=\mathrm{\Gamma }(2\lambda +l)/[l!\mathrm{\Gamma }(2\lambda )]`$. For $`l0`$, $`C_l^{|\lambda |}(1)=(2|\lambda |+l1)(2|\lambda |+l2)\mathrm{}(2|\lambda |)/l!`$ is manifestly an increasing function of $`|\lambda |`$, and, for $`l=0`$, $`C_0^{|\lambda |}(1)=1`$ is constant. From this we infer that $`l`$ and for $`\beta `$ in the disk centred on $`0`$ and of radius $`ϵ`$, $`C_l^{|\beta |/2}(1)C_l^{ϵ/2}(1)`$ holds, which leads to the $`\beta `$-independent bound
$$\left(\frac{s}{r_{12}}\right)^lC_l^{ϵ/2}(1)d\mathrm{\Omega }_1|\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{a}{}^{}|,$$
which is manifestly the general term of a convergent series. Therefore the series (A8) possesses a limit when $`\beta 0`$ which is simply obtained by setting $`\beta =0`$ under the sign $``$. Using $`C_l^0(\mathrm{cos}\theta )=\delta _{l0}`$ we find this limit to be $`0`$ if $`a=3`$ and
$$\frac{1}{a+3}d\mathrm{\Omega }_1\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}_a$$
(A9)
if $`a3`$. Gathering the results (A5)-(A6) and (A9), we arrive to
$`{\displaystyle \underset{a+3<0}{}}{\displaystyle \frac{s^{a+3}}{a+3}}{\displaystyle }d\mathrm{\Omega }_1\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{a}{}^{}+\mathrm{ln}\left({\displaystyle \frac{s}{s_1}}\right){\displaystyle }d\mathrm{\Omega }_1\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{3}{}^{}`$ (A10)
$`=\mathrm{FP}_{\genfrac{}{}{0pt}{}{\alpha 0}{\beta 0}}{\displaystyle _{_1(s)}}d^3𝐱{\displaystyle \underset{a+30}{}}\left({\displaystyle \frac{r_1}{s_1}}\right)^\alpha \left({\displaystyle \frac{r_2}{s_2}}\right)^\beta r_1^a\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}_a`$ (A11)
$`=\mathrm{FP}_{\genfrac{}{}{0pt}{}{\beta 0}{\alpha 0}}{\displaystyle _{_1(s)}}d^3𝐱{\displaystyle \underset{a+30}{}}\left({\displaystyle \frac{r_1}{s_1}}\right)^\alpha \left({\displaystyle \frac{r_2}{s_2}}\right)^\beta r_1^a\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{a}{}^{},`$ (A12)
from which we can now easily prove the equivalence with the Hadamard partie finie. Like in the proof of Proposition 1 we consider two open domains $`𝒟_1`$ and $`𝒟_2`$, disjoined and complementary in $`^3`$, and such that $`_1(s)𝒟_1`$ and $`_2(s)𝒟_2`$. We can write
$$_{𝒟_1}d^3𝐱\left(\frac{r_1}{s_1}\right)^\alpha \left(\frac{r_2}{s_2}\right)^\beta F=_{𝒟_1_1(s)}d^3𝐱\left(\frac{r_1}{s_1}\right)^\alpha \left(\frac{r_2}{s_2}\right)^\beta F+_{_1(s)}d^3𝐱\left(\frac{r_1}{s_1}\right)^\alpha \left(\frac{r_2}{s_2}\right)^\beta F,$$
(A13)
where each of the objects is defined by complex analytic continuation in a neighbourhood of $`\alpha =\beta =0`$ \[proof similar to the one after (40)\]. Like in (47) we associate to $`F`$ the function $`\stackrel{~}{F}_1`$ representing its “regularization” around the point 1,
$$\stackrel{~}{F}_1=F\underset{a+30}{}r_1^a\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{a}{}^{},$$
(A14)
and we re-write the right side of (A13) as
$$_{𝒟_1_1(s)}d^3𝐱\left(\frac{r_1}{s_1}\right)^\alpha \left(\frac{r_2}{s_2}\right)^\beta F+_{_1(s)}d^3𝐱\left(\frac{r_1}{s_1}\right)^\alpha \left(\frac{r_2}{s_2}\right)^\beta \stackrel{~}{F}_1+_{_1(s)}d^3𝐱\left(\frac{r_1}{s_1}\right)^\alpha \left(\frac{r_2}{s_2}\right)^\beta \underset{a+30}{}r_1^a\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{a}{}^{}.$$
Of these three terms, the first two are well-defined when $`\alpha `$ and $`\beta `$ tend to zero, hence their finite parts are simply obtained by posing $`\alpha =0=\beta `$. On the other hand we have proved previously that the finite parts $`\mathrm{Pf}_{\genfrac{}{}{0pt}{}{\alpha 0}{\beta 0}}`$ and $`\mathrm{Pf}_{\genfrac{}{}{0pt}{}{\beta 0}{\alpha 0}}`$ of the third term are equal and we have found their common value to be given by (A10). This shows immediately that
$`\mathrm{FP}_{\genfrac{}{}{0pt}{}{\alpha 0}{\beta 0}}{\displaystyle _{𝒟_1}}d^3𝐱\left({\displaystyle \frac{r_1}{s_1}}\right)^\alpha \left({\displaystyle \frac{r_2}{s_2}}\right)^\beta F`$ $`=`$ $`{\displaystyle _{𝒟_1_1(s)}}d^3𝐱F+{\displaystyle _{_1(s)}}d^3𝐱\stackrel{~}{F}_1`$ (A15)
$`+`$ $`{\displaystyle \underset{a+3<0}{}}{\displaystyle \frac{s^{a+3}}{a+3}}{\displaystyle }d\mathrm{\Omega }_1\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}{}_{a}{}^{}+\mathrm{ln}\left({\displaystyle \frac{s}{s_1}}\right){\displaystyle }d\mathrm{\Omega }_1\stackrel{}{f}{\displaystyle \genfrac{}{}{0pt}{}{}{1}}_3`$ (A16)
(and idem with $`\mathrm{Pf}_{\genfrac{}{}{0pt}{}{\beta 0}{\alpha 0}}`$). We recognize in the right side of (A15) the Hadamard partie finie of the integral. Indeed the second term clearly admits an expansion in positive powers of $`s`$,
$$N,_{_1(s)}d^3𝐱\stackrel{~}{F}_1=\underset{0<a+3N}{}\frac{s^{a+3}}{a+3}d\mathrm{\Omega }_1\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{a}{}^{}+o(s^N),$$
(A17)
so we recover exactly the partie-finie integral over $`𝒟_1`$ in the form given by (27). QED.
|
warning/0004/math0004049.html
|
ar5iv
|
text
|
# Spectral radii of bounded operators on topological vector spaces
## 0. Introduction
The spectral radius of a bounded linear operator $`T`$ on a Banach space is defined by the Gelfand formula $`r(T)=lim_n\sqrt[n]{T^n}`$. It is well known that $`r(T)`$ equals the actual radius of the spectrum $`\left|\sigma (T)\right|=sup\{|\lambda |:\lambda \sigma (T)\}`$. Further, it is known that the resolvent $`R_\lambda =(\lambda IT)^1`$ is given by the Neumann series $`_{i=0}^{\mathrm{}}\frac{T^i}{\lambda ^{i+1}}`$ whenever $`|\lambda |>r(T)`$. It is natural to ask if similar results are valid in a more general setting, e.g., for a bounded linear operator on an arbitrary topological vector space. The author arrived to these questions when generalizing some results on Invariant Subspace Problem from Banach lattices to ordered topological vector spaces. One major difficulty is that it is not clear which class of operators should be considered, because there are several non-equivalent ways of defining bounded operators on topological vector spaces. Another major difficulty is the lack of a readily available developed spectral theory. The spectral theory of operators on Banach spaces has been thoroughly studied for a long time, and is extensively used. Unfortunately, little has been known about spectral theory of bounded operators on general topological vector spaces, and many techniques used in Banach spaces cannot be applied for operators on topological vector spaces. In particular, the spectrum, the spectral radius, and the Neumann series are the tools which are widely used in the study of the Invariant Subspace Problem in Banach spaces, but which have not been sufficiently studied for general topological vector spaces. To overcome this obstacle we have developed a version of the spectral theory of bounded operators on general topological vector spaces and on locally convex spaces. Some results in this direction have also been obtained by B. Gramsch \[Gra66\], and by F. Garibay and R. Vera \[GBVM97, GBVM98, VM97\].
In particular, we consider the following classification of bounded operators on a topological vector space. We call a linear operator $`T`$
* nb-bounded if $`T`$ maps some neighborhood of zero into a bounded set,
* nn-bounded if there is a base of neighborhoods of zero such that $`T`$ maps every neighborhood in this base into a multiple of itself, and
* bb-bounded if $`T`$ maps bounded sets into bounded sets.
The classes of all linear operators, of all bb-bounded operators, of all continuous operators, of all nn-bounded operators, and of all nb-bounded operators form nested algebras. The spectrum of an operator $`T`$ in each of these algebras is defined as usual, i.e., the set of $`\lambda `$’s for which $`\lambda IT`$ is not invertible in this algebra. We show that the well known Gelfand formula for the spectral radius of an operator on a Banach space $`r(T)=\underset{n\mathrm{}}{lim}\sqrt[n]{T^n}`$ can be generalized to each of the five classes of operators on topological vector spaces, and then we use this formula to define the spectral radius of an operator in each of the classes. Then in Section 5 we show that if $`T`$ is a continuous operator on a sequentially complete locally convex space and $`|\lambda |`$ is greater than the spectral radius of $`T`$ in any of the five classes, then the Neumann series $`_{n=0}^{\mathrm{}}\frac{T^n}{\lambda ^{n+1}}`$ converges in the topology of the class, and $`\lambda `$ does not belong to the corresponding spectrum of $`T`$, i.e., the spectral radius is greater than or equal to the geometrical radius of the spectrum. In Sections 6 and 7 we show that the radii are equal for nb-bounded and compact operators.
This paper is based on a part of the author’s Ph.D. thesis \[Tro99\]. I would like to thank Yuri Abramovich for his encouragement and for numerous suggestions and improvements. Robert Kaufman read parts of the draft and asked interesting questions, which resulted in new developments. Thanks are also due to Heinrich Lotz, Michael Neumann, and Joseph Rosenblatt for their support and interest in my work.
## 1. Preliminaries and notation
The symbols $`X`$ and $`Y`$ always denote topological vector spaces. A neighborhood of a point $`xX`$ is any subset of $`X`$ containing an open set which contains $`x`$. Neighborhoods of zero will often be referred to as zero neighborhoods. Every zero neighborhood $`V`$ is absorbing, i.e., $`_{n=1}^{\mathrm{}}nV=X`$. In every topological vector space (over $``$ or $``$) there exists a base $`𝒩_0`$ of zero neighborhoods with the following properties:
1. Every $`V𝒩_0`$ is balanced, i.e., $`\lambda VV`$ whenever $`|\lambda |1`$;
2. For every $`V_1,V_2𝒩_0`$ there exists $`V𝒩_0`$ such that $`VV_1V_2`$;
3. For every $`V𝒩_0`$ there exists $`U𝒩_0`$ such that $`U+UV`$;
4. For every $`V𝒩_0`$ and every scalar $`\lambda `$ the set $`\lambda V`$ is in $`𝒩_0`$.
Whenever we mention a base zero neighborhood, we assume that the base satisfies these properties.
A topological vector space is called normed if the topology is given by a norm. In this case the collection of all balls centered at zero is a base of zero neighborhoods. A complete normed space is referred to as a Banach space. See \[DS58\] for a detailed study of normed and Banach spaces.
A subset $`A`$ of a topological vector space is called bounded if it is absorbed by every zero neighborhood, i.e., for every zero neighborhood $`V`$ one can find $`\alpha >0`$ such that $`A\alpha V`$. A set $`A`$ in a topological vector space is said to be pseudo-convex or semi-convex if $`A+A\alpha A`$ for some number $`\alpha `$ (for convex sets $`\alpha =2`$). If $`U`$ is a zero neighborhood, $`(x_\gamma )`$ is a net in $`X`$, and $`xX`$, we write $`x_\gamma \stackrel{𝑈}{}x`$ if for every $`\epsilon >0`$ one can find an index $`\gamma _0`$ such that $`x_\gamma x\epsilon U`$ whenever $`\gamma \gamma _0`$. It is easy to see that when $`U`$ is pseudo-convex, this convergence determines a topology on $`X`$, and the set of all scalar multiples of $`U`$ forms a base of the topology. We denote $`X`$ equipped with this topology by $`(X,U)`$. Clearly, $`(X,U)`$ is Hausdorff if and only if $`_{n=1}^{\mathrm{}}\frac{1}{n}U=\{0\}`$.
A topological vector space is said to be locally bounded if there exists a bounded zero neighborhood. Notice that if $`U`$ is a bounded zero neighborhood then it is pseudo-convex. Conversely, if $`U`$ is a pseudo-convex zero neighborhood, then $`(X,U)`$ is locally bounded. Recall that a quasinorm is a real-valued function on a vector space which satisfies all the axioms of norm except the triangle inequality, which is substituted by $`x+yk\left(x+y\right)`$ for some fixed positive constant $`k`$. It is known (see, e.g., \[Köt60\]) that a topological vector space is quasinormable if and only if it is locally bounded and Hausdorff. A complete quasinormed space is called quasi-Banach.
If the topology of a topological vector space $`X`$ is given by a seminorm p, we say that $`X=(X,p)`$ is a seminormed space. Clearly, in this case $`X=(X,U)`$ where the convex set $`U`$ is the unit ball of $`p`$ and, conversely, $`p`$ is the Minkowski functional of $`U`$. A Hausdorff topological vector space is called locally convex if there is a base of convex zero neighborhoods or, equivalently, if the topology is generated by a family of seminorms (the Minkowski functionals of the convex zero neighborhoods). When dealing with locally convex spaces we will always assume that the base zero neighborhoods are convex. Similarly, a Hausdorff topological vector space is said to be locally pseudo-convex if it has a base of pseudo-convex zero neighborhoods. A complete metrizable topological vector space is usually referred to as a Fréchet space.
Further details on topological vector spaces can be found in \[DS58, Köt60, RR64, Edw65, Sch71, KN76\]. For details on locally bounded and quasinormed topological vector spaces we refer the reader to \[Köt60, KPR84, Rol85\].
By an operator we always mean a linear operator between vector spaces. We will usually use the symbols $`S`$ and $`T`$ to denote operators. Recall that an operator $`T`$ between normed spaces is said to be bounded if its operator norm defined by $`T=sup\{Tx:x1\}`$ is finite. It is well known that an operator between normed spaces is bounded if and only if it is continuous. An operator between two vector spaces is said to be of finite rank if the range of $`T`$ is finite dimensional.
If $`𝒜`$ is a unital algebra and $`a𝒜`$, then the resolvent set of $`a`$ is the set $`\rho (a)`$ of all $`\lambda `$ such that $`e\lambda a`$ is invertible in $`𝒜`$. The resolvent set of an element $`a`$ in a non-unital algebra $`𝒜`$ is defined as the set of all $`\lambda `$ for which $`e\lambda a`$ is invertible in the unitalization $`𝒜_\times `$ of $`𝒜`$. The spectrum of an element of an algebra is defined via $`\sigma (a)=\rho (a)`$. It is well-known that whenever $`𝒜`$ is a unital Banach algebra then $`\sigma (a)`$ is compact and nonempty for every $`a𝒜`$. In this case the spectral radius $`r(a)`$ is defined via Gelfand formula: $`r(a)=\underset{n\mathrm{}}{lim}\sqrt[n]{a^n}`$. It is well known that $`r(a)=\left|\sigma (a)\right|`$, where $`\left|\sigma (a)\right|`$ is the geometrical radius of $`\sigma (a)`$, i.e., $`\left|\sigma (a)\right|=sup\{|\lambda |:\lambda \sigma (a)\}`$. An element $`aA`$ is said to be quasinilpotent if $`\sigma (a)=\{0\}`$.
If $`T`$ is a bounded operator on a Banach space $`X`$ then we will consider the spectrum $`\sigma (T)`$ and the resolvent set $`\rho (T)`$ in the sense of the Banach algebra of bounded operators on $`X`$. If $`\lambda \rho (T)`$ then the inverse $`(I\lambda T)^1`$ is called the resolvent operator and is denoted by $`R(T;\lambda )`$ or just $`R_\lambda `$. It is well known that if $`\lambda `$ satisfies $`|\lambda |>r(T)`$ then the Neumann series $`_{i=0}^{\mathrm{}}\frac{T^i}{\lambda ^{i+1}}`$ converges to $`R_\lambda `$ in operator norm. We say that $`T`$ is locally quasinilpotent at $`xX`$ if $`\underset{n\mathrm{}}{lim}\sqrt[n]{T^nx}=0`$.
## 2. Bounded operators
There are various definitions for a bounded linear operator between two topological vector spaces. To avoid confusion, we will, of course, give different names to different types of boundedness.
###### Definition 2.1.
Let $`X`$ and $`Y`$ be topological vector spaces. An operator $`T:XY`$ is said to be
1. bb-bounded if it maps every bounded set into a bounded set;
2. nb-bounded if it maps some neighborhood into a bounded set;
Further, if $`X=Y`$ we will say that $`T:XX`$ is nn-bounded if there exists a base $`𝒩_0`$ of zero neighborhoods such that for every $`U𝒩_0`$ there is a positive scalar $`\alpha `$ such that $`T(U)\alpha U`$.
###### Remark 2.2.
\[Edw65\] and \[KN76\] present (i) as the definition of a bounded operator on a topological vector space, while \[RR64\] and \[Sch71\] use (ii) for the same purpose. As we will see, these definitions are far from being equivalent.
###### Proposition 2.3.
Let $`X`$ and $`Y`$ be topological vector spaces. For an operator $`T:XY`$ consider the following statements:
1. $`T`$ is bb-bounded;
2. $`T`$ is continuous;
3. $`T`$ is nn-bounded;
4. $`T`$ is nb-bounded.
Then (iv) $``$ (ii) $``$ (i). Furthermore, if $`X=Y`$ then (iv) $``$ (iii) $``$ (ii) $``$ (i).
###### Proof.
The implications (iv) $``$ (ii) $``$ (i) are trivial. To show (iv) $``$ (iii) $``$ (ii) assume that $`X=Y`$ and fix a base $`𝒩_0`$ of zero neighborhoods. If $`T`$ is nb-bounded then $`T(U)`$ is bounded for some $`U𝒩_0`$. Note that $`\stackrel{~}{𝒩}_0=\{\lambda UV:\lambda >0,V𝒩_0\}`$ is another base of zero neighborhoods. For each $`W=\lambda UV`$ in $`\stackrel{~}{𝒩}_0`$ we have $`T(W)\lambda T(U)`$. But $`T(U)`$ is bounded and so $`T(W)\lambda T(U)\lambda \alpha W`$ for some positive $`\alpha `$, i.e., $`T`$ is nn-bounded.<sup>1</sup><sup>1</sup>1Note that if the topology is locally convex, then we can assume that $`U`$ is convex and $`𝒩_0`$ consists of convex neighborhoods. In this case $`\stackrel{~}{𝒩}_0`$ also consists of convex neighborhoods.
Finally, if $`T`$ is nn-bounded, then there is a base $`𝒩_0`$ such that for every zero neighborhood $`U𝒩_0`$ there is a positive scalar $`\alpha `$ such that $`T(U)\alpha U`$. Let $`V`$ be an arbitrary zero neighborhood. Then there exists $`U𝒩_0`$ such that $`UV`$, so that $`T(U)\alpha U\alpha V`$ for some $`\alpha >0`$. Taking $`W=\frac{1}{\alpha }U`$ we get $`T(W)V`$, hence $`T`$ is continuous. ∎
###### 2.4.
It can be easily verified that if $`T`$ is an operator on a locally bounded space then all the statements in Lemma 2.3 are equivalent. In general, however, these notions are not equivalent. Obviously, the identity operator $`I`$ is always nn-bounded, continuous, and bb-bounded, but $`I`$ is nb-bounded if and only if the space is locally bounded. Every bb-bounded operator between two locally convex spaces is continuous if and only if the domain space is bornological. (Recall that a locally convex space is bornological if every balanced convex set absorbing every bounded set is a zero neighborhood, for details see \[Sch71, RR64\].)
###### Example 2.5.
A continuous but not nn-bounded operator. Let $`T`$ be the left shift on the space of all real sequences $`^{}`$ with the topology of coordinate-wise convergence, i.e., $`T:(x_0,x_1,x_2,\mathrm{})(x_1,x_2,x_3,\mathrm{})`$. Clearly $`T`$ is continuous. We will show that $`T`$ is not nn-bounded. Assume that for every zero neighborhood $`U`$ in some base $`𝒩_0`$ there is a positive scalar $`\alpha `$ such that $`T(U)\alpha U`$. Since the set $`\{x=(x_k):|x_0|<1\}`$ is a zero neighborhood, there must be a base neighborhood $`U𝒩_0`$ such that $`U\{x:|x_0|<1\}`$. Since $`T(U)\alpha U`$ for some positive $`\alpha `$ then $`T^n(U)\alpha ^nU`$, so that if $`x=(x_k)U`$ then $`T^nx\alpha ^nU`$, so that $`|x_n|=\left|(T^nx)_0\right|<\alpha ^n`$. Hence $`U\{x:|x_n|<\alpha ^n\text{ for each }n>0\}`$. But this set is bounded, while the space is not locally bounded, a contradiction.
###### 2.6.
Algebraic properties of bounded operators. The sum of two bb-bounded operators is bb-bounded because the sum of two bounded sets in a topological vector space is bounded. Clearly the product of two bb-bounded operators is bb-bounded. It is well known that sums and products of continuous operators are continuous. Obviously, the product of two nn-bounded operators is nn-bounded, and it can be easily verified that the sum of two nn-bounded operators on a locally convex (or locally pseudo-convex) space is again nn-bounded. It is not difficult to see that the sum of two nb-bounded operator is nb-bounded. Indeed, suppose that $`T_1`$ and $`T_2`$ are two nb-bounded operators, then the sets $`T_1(U_1)`$ and $`T_2(U_2)`$ are bounded for some base zero neighborhoods $`U_1`$ and $`U_2`$. There exists another base zero neighborhood $`UU_1U_2`$, then the sets $`T_1(U)`$ and $`T_2(U)`$ are bounded, so that $`(T_1+T_2)(U)T_1(U)+T_2(U)`$ is bounded. Finally, it is not difficult to see that the product of two nb-bounded operators is again nb-bounded. In fact, it follows immediately from Proposition 2.3 and the following simple observation: if we multiply an nb-bounded operator by a bb-bounded operator on the left or by an nn-bounded operator on the right, the product is nb-bounded.
Thus, the class of all bb-bounded operators, the class of all continuous operators, and the class of all nb-bounded operators are subalgebras of the algebra of all linear operators. The class of nn-bounded operators is an algebra provided that the space is locally (pseudo-)convex.
Boundedness in terms of convergence
Suppose $`T:XY`$ is an operator between two topological vector spaces. It is well known that $`T`$ is continuous if and only if it maps convergent nets to convergent nets.
Notice that a subset of a topological vector space is unbounded if and only if it contains an unbounded sequence. Therefore, an operator is bb-bounded if and only if it maps bounded sequences (nets) to bounded sequences (respectively nets).
It is easy to see that $`T`$ is nn-bounded if and only if $`T`$ maps $`U`$-bounded ($`U`$-convergent to zero) sequences to $`U`$-bounded (respectively $`U`$-convergent to zero) sequences for every base zero neighborhood $`U`$ in some base of zero neighborhoods. We say that a net $`(x_\gamma )`$ is $`U`$-bounded if it is contained in $`\alpha U`$ for some $`\alpha >0`$, and $`x_\gamma \stackrel{𝑈}{}0`$ if for every $`\alpha >0`$ there exits $`\gamma _0`$ such that $`x_\gamma \alpha U`$ whenever $`\gamma >\gamma _0`$.
###### 2.7.
Suppose $`T`$ is nb-bounded, then $`T(U)`$ is bounded for some zero neighborhood $`U`$. Obviously $`x_\gamma \stackrel{𝑈}{}0`$ implies $`Tx_\gamma 0`$. The converse implication is also valid: if $`T`$ maps $`U`$-convergent sequences to convergent sequences, then $`T`$ has to be nb-bounded and the set $`T(U)`$ is bounded. Indeed, if $`T(U)`$ is unbounded, then there is a zero neighborhood $`V`$ in $`Y`$ such that $`V`$ does not absorb $`T(U)`$. Then for every $`n1`$ there exists $`y_nT(U)nV`$. Suppose $`y_n=Tx_n`$ for some $`x_nU`$, then $`\frac{x_n}{n}\stackrel{𝑈}{}0`$, but $`T(\frac{x_n}{n})=\frac{y_n}{n}V`$, so that $`T(\frac{x_n}{n})`$ does not converge to zero.
Normed, quasinormed, and seminormed spaces
Next, we discuss bounded operators in some particular topologies. Notice that every normed, seminormed, or quasinormed vector space is locally bounded. Therefore bb-boundedness, continuity, nn-boundedness and nb-boundedness coincide for operators on such spaces.
Locally convex topology
Similarly to the norm of an operator on a Banach space, we introduce the seminorm of an operator on a seminormed space.
###### Definition 2.8.
Let $`T`$ be an operator on a seminormed vector space $`(X,p)`$. As in the case with normed spaces, $`p`$ generates an operator seminorm $`p(T)`$ defined by
$$p(T)=\underset{p(x)0}{sup}\frac{p(Tx)}{p(x)}.$$
More generally, let $`S:XY`$ be a linear operator between two seminormed spaces $`(X,p)`$ and $`(Y,q)`$. Then we define a mixed operator seminorm associated with $`p`$ and $`q`$ via
$$𝔪_{pq}(S)=\underset{p(x)0}{sup}\frac{q(Sx)}{p(x)}.$$
The seminorm $`𝔪_{pq}(S)`$ is a measure of how far in the seminorm $`q`$ the points of the $`p`$-unit ball can go under $`S`$. Notice, that $`p(T)`$ and $`𝔪_{pq}(S)`$ may be infinite. Clearly, if $`T`$ is an operator on a seminormed space $`(X,p)`$, then $`𝔪_{pp}(T)=p(T)`$.
###### Lemma 2.9.
If $`S:XY`$ is an operator between two seminormed spaces $`(X,p)`$ and $`(Y,q)`$, then
1. $`𝔪_{pq}(S)=\underset{p(x)=1}{sup}q(Sx)=\underset{p(x)1}{sup}q(Sx)`$;
2. $`q(Sx)𝔪_{pq}(S)p(x)`$ whenever $`𝔪_{pq}(S)<\mathrm{}`$.
###### Proof.
The first equality in (i) follows immediately from the definition of $`p(T)`$. We obviously have
$$\underset{p(x)=1}{sup}q(Sx)\underset{p(x)1}{sup}q(Sx).$$
In order to prove the opposite inequality, notice that if $`0<p(x)1`$, then $`q(Sx)\frac{q(Sx)}{p(x)}𝔪_{pq}(S)`$. Thus, it is left to show that $`p(x)=0`$ implies $`q(Sx)𝔪_{pq}(S)`$. Pick any $`z`$ with $`p(z)>0`$, then
$$p(\frac{z}{n})=p(x+\frac{z}{n}x)p(x+\frac{z}{n})+p(x)=p(x+\frac{z}{n})p(x)+p(\frac{z}{n})=\frac{p(z)}{n},$$
so that $`p(x+\frac{z}{n})=p(\frac{z}{n})(0,1)`$ for $`n>p(z)`$. Further, since $`Sx+\frac{Sz}{n}`$ converges to $`Sx`$ we have
$$q(Sx)=\underset{n\mathrm{}}{lim}q(Sx+\frac{Sz}{n})\underset{n\mathrm{}}{lim}\frac{q\left(S(x+\frac{z}{n})\right)}{p(x+\frac{z}{n})}𝔪_{pq}(S).$$
Finally, (ii) follows directly from the definition if $`p(x)0`$. In the case when $`p(x)=0`$, again pick any $`z`$ with $`p(z)>0`$, then $`p(x+\frac{z}{n})0`$ and
$$q(Sx)=\underset{n\mathrm{}}{lim}q(Sx+\frac{Sz}{n})=\underset{n\mathrm{}}{lim}q\left(S(x+\frac{z}{n})\right)\underset{n\mathrm{}}{lim}𝔪_{pq}(S)p(x+\frac{z}{n})=0.$$
###### Corollary 2.10.
If $`T`$ is an operator on a seminormed space $`(X,p)`$, then
1. $`p(T)=\underset{p(x)=1}{sup}p(Tx)=\underset{p(x)1}{sup}p(Tx)`$;
2. $`p(Tx)p(T)p(x)`$ whenever $`p(T)<\mathrm{}`$.
The following propositions characterize continuity and boundedness of an operator on a locally convex space in terms of operator seminorms. We assume that $`X`$ and $`Y`$ are locally convex spaces with generating families of seminorms $`𝒫`$ and $`𝒬`$ respectively.
###### Proposition 2.11.
Let $`S`$ be an operator from $`X`$ to $`Y`$, then $`S`$ is continuous if and only if for every $`q𝒬`$ there exists $`p𝒫`$ such that $`𝔪_{pq}(S)`$ is finite.
###### Proposition 2.12.
An operator $`T`$ on $`X`$ is nn-bounded if and only if $`p(T)`$ is finite for every $`p𝒫`$, or, equivalently, if $`T`$ maps $`p`$-bounded sets to $`p`$-bounded sets for every $`p`$ in some generating family $`𝒫`$ of seminorms.
###### Proposition 2.13.
Let $`S:XY`$ be a linear operator, then the following are equivalent:
1. $`S`$ is nb-bounded;
2. $`S`$ maps $`p`$-bounded sets into bounded sets for some $`p𝒫`$;
3. There exists $`p𝒫`$ such that $`𝔪_{pq}(S)<\mathrm{}`$ for every $`q𝒬`$.
Since the balanced convex hall of a bounded set in a locally convex space is again bounded, we also have the following characterization.
###### Proposition 2.14.
An operator $`S:XY`$ is bb-bounded if and only if $`𝔪_{pq}(S)`$ whenever $`q𝒬`$ and $`p`$ is the Minkowski functional of a convex balanced bounded set.
Operator topologies
For each of the five classes of operators, we introduce an appropriate natural operator topology. The class of all linear operators between two topological vector spaces will be usually equipped with the strong operator topology. Recall, that a sequence $`(S_n)`$ of operators from $`X`$ to $`Y`$ is said to converge strongly or pointwise to a map $`S`$ if $`S_nxSx`$ for every $`xX`$. Clearly, $`S`$ will also be a linear operator.
The class of all bb-bounded operators will usually be equipped with the topology of uniform convergence on bounded sets. Recall, that a sequence $`(S_n)`$ of operators is said to converge to zero uniformly on $`A`$ if for each zero neighborhood $`V`$ in $`Y`$ there exists an index $`n_0`$ such that $`S_n(A)V`$ for all $`n>n_0`$. We say that $`(S_n)`$ converges to $`S`$ uniformly on bounded sets if $`(S_nS)`$ converges to zero uniformly on bounded sets. Recall also that a family $`𝒢`$ of operators is called uniformly bounded on a set $`AX`$ if the set $`_{S𝒢}S(A)`$ is bounded in $`Y`$.
###### Lemma 2.15.
If a sequence $`(S_n)`$ of bb-bounded operators converges uniformly on bounded sets to an operator $`S`$, then $`S`$ is also bb-bounded.
###### Proof.
Fix a bounded set $`A`$. Since $`SS_n`$ converges to zero uniformly on bounded sets then for every base zero neighborhood $`V`$ there exists an index $`n_0`$ such that $`(S_nS)(A)V`$ whenever $`nn_0`$. This yields $`S(A)S_n(A)+V\gamma V`$ since $`S_n(A)`$ is bounded. Thus, $`S(A)`$ is bounded for every bounded set $`A`$, so that $`S`$ is bb-bounded. ∎
The class of all continuous operators will be usually equipped with the topology of equicontinuous convergence. Recall, that a family $`𝒢`$ of operators from $`X`$ to $`Y`$ is called equicontinuous if for each zero neighborhood $`V`$ in $`Y`$ there is a zero neighborhood $`U`$ in $`X`$ such that $`S(U)V`$ for every $`S𝒢`$. We say that a sequence $`(S_n)`$ converges to zero equicontinuously if for each zero neighborhood $`V`$ in $`Y`$ there is a zero neighborhood $`U`$ in $`X`$ such that for every $`\epsilon >0`$ there exists an index $`n_0`$ such that $`S_n(U)\epsilon V`$ for all $`n>n_0`$.
###### Lemma 2.16.
If a sequence $`S_n`$ of continuous operators converges equicontinuously to $`S`$, then $`S`$ is also continuous.
###### Proof.
Fix a zero neighborhood $`V`$, there exist zero neighborhoods $`V_1`$ and $`U`$ and an index $`n_0`$ such that $`V_1+V_1V`$ and $`(S_nS)(U)V_1`$ whenever $`n>n_0`$. Fix $`n>n_0`$. The continuity of $`S_n`$ guarantees that there exists a zero neighborhood $`WU`$ such that $`S_n(W)V_1`$. Since $`(S_nS)(W)V_1`$, we get $`S(W)S_n(W)+V_1V_1+V_1V`$, which shows that $`S`$ is continuous. ∎
The class of all nn-bounded operators will be usually equipped with the topology of nn-convergence, defined as follows. We will call a collection $`𝒢`$ of operators uniformly nn-bounded if there exists a base $`𝒩_0`$ of zero neighborhoods such that for every $`U𝒩_0`$ there exists a positive real $`\beta `$ such that $`S(U)\beta U`$ for each $`S𝒢`$. We say that a sequence $`(S_n)`$ nn-converges to zero if there is a base $`𝒩_0`$ of zero neighborhoods such that for every $`U𝒩_0`$ and every $`\epsilon >0`$ we have $`S_n(U)\epsilon U`$ for all sufficiently large $`n`$.
###### Question.
Is the class of all nn-bounded operators closed relative to nn-convergence?
Finally, the class of all nb-bounded operators will be usually equipped with the topology of uniform convergence on a zero neighborhood.
###### Example 2.17.
The class of nb-bounded operators is not closed in the topology of uniform convergence on a zero neighborhood. Let $`X=^{}`$, the space of all real sequences with the topology of coordinate-wise convergence. Let $`T_n`$ be the projection on the first $`n`$ components. Clearly, every $`T_n`$ is nb-bounded because it maps the zero neighborhood $`U_n=\{(x_i)_{i=1}^{\mathrm{}}:|x_i|<1\text{ for }i=1,\mathrm{},n\}`$ to a bounded set. On the other hand, $`(T_n)`$ converges uniformly on $`X`$ to the identity operator, while the identity operator on $`X`$ is not nb-bounded.
## 3. Spectra of an operator
Recall that if $`T`$ is a continuous operator on a Banach space, then its resolvent set $`\rho (T)`$ is the set of all $`\lambda `$ such that the resolvent operator $`R_\lambda =(\lambda IT)^1`$ exists, while the spectrum of $`T`$ is defined by $`\sigma (T)=\rho (T)`$. The Open Mapping Theorem guarantees that if $`R_\lambda `$ exists then it is automatically continuous. Now, if $`T`$ is an operator on an arbitrary topological vector space and $`\lambda `$ then the algebraic inverse $`R_\lambda =(\lambda IT)^1`$ may exist but not be continuous, or may be continuous but not nb-bounded, etc. In order to treat all these cases properly we introduce the following definitions.
###### Definition 3.1.
Let $`T`$ be a linear operator on a topological vector space. We denote the set of all scalars $`\lambda `$ for which $`\lambda IT`$ is invertible in the algebra of linear operators by $`\rho ^l(T)`$. We say that $`\lambda \rho ^{bb}(T)`$ (respectively $`\rho ^c(T)`$ or $`\rho ^{nn}(T)`$) if the inverse of $`\lambda IT`$ is bb-bounded (respectively continuous or nn-bounded). Finally, we say that $`\lambda \rho ^{nb}(T)`$ if the inverse of $`\lambda IT`$ belongs to the unitalization of the algebra of nb-bounded operators, i.e., when $`(\lambda IT)^1=\alpha I+S`$ for a scalar $`\alpha `$ and an nb-bounded operator $`S`$.
The spectral sets $`\sigma ^l(T)`$, $`\sigma ^{bb}(T)`$, $`\sigma ^c(T)`$, $`\sigma ^{nn}(T)`$, and $`\sigma ^{nb}(T)`$ are defined to be the complements of the resolvent sets $`\rho ^l(T)`$, $`\rho ^{bb}(T)`$, $`\rho ^c(T)`$, $`\rho ^{nn}(T)`$, and $`\rho ^{nb}(T)`$ respectively.<sup>2</sup><sup>2</sup>2 We use superscripts in order to avoid confusion with $`\sigma _c(T)`$, which is commonly used for continuous spectrum. We will denote the (left and right) inverse of $`\lambda IT`$ whenever it exists by $`R_\lambda `$.
###### 3.2.
It follows immediately from Proposition 2.3 that $`\sigma ^l(T)\sigma ^{bb}(T)\sigma ^c(T)\sigma ^{nn}(T)\sigma ^{nb}(T)`$. It follows from the Open Mapping Theorem that for a continuous operator $`T`$ on a Banach space all the introduced spectra coincide with the usual spectrum $`\sigma (T)`$. Since the Open Mapping Theorem is still valid on Fréchet spaces, we have $`\sigma ^l(T)=\sigma ^{bb}(T)=\sigma ^c(T)`$ for a continuous operator $`T`$ on a Fréchet space.
###### 3.3.
If $`T`$ is an operator on a locally bounded space $`(X,U)`$, then by 2.4 bb-boundedness of $`T`$ is equivalent to nb-boundedness, so that $`\sigma ^{bb}(T)=\sigma ^c(T)=\sigma ^{nn}(T)=\sigma ^{nb}(T)`$. We will denote this set by $`\sigma _U(T)`$ to avoid ambiguity. Spectral theory of continuous operators on quasi-Banach spaces was developed in \[Gra66\].
###### 3.4.
There are several reasons why we define $`\sigma ^{nb}`$ in a slightly different fashion than the other spectra. Namely, for $`\lambda `$ to be in $`\rho ^{nb}(T)`$ we require $`(\lambda IT)^1`$ be not just nb-bounded, but be nb-bounded up to a multiple of the identity operator. On one hand, this is the standard way to define the spectrum of an element in a non-unital algebra, and we know that the algebra of nb-bounded operators is unital only when the space is locally bounded. On the other hand, if we defined $`\rho ^{nb}(T)`$ as the set of all $`\lambda `$ for which $`(\lambda IT)^1`$ is nb-bounded, then we wouldn’t have gotten any deep theory because $`(\lambda IT)^1`$ is almost never nb-bounded when the space is not locally bounded.
Indeed, suppose that $`X`$ is not locally bounded, $`T`$ is a bb-bounded operator on $`X`$, and $`\lambda `$. Then $`R_\lambda =(\lambda IT)^1`$ cannot be nb-bounded, because in this case $`I=(\lambda IT)R_\lambda `$ would be nb-bounded by 2.6 as a product of a bb-bounded and an nb-bounded operators. But we know that $`I`$ is not nb-bounded because $`X`$ is not locally bounded.
We will see in Proposition 6.3 that in a locally convex but non locally bounded space nb-bounded operators are never invertible, which implies that in such spaces $`(\lambda IT)^1`$ is not nb-bounded for any linear operator $`T`$.
###### 3.5.
Next, let $`T`$ be a (norm) continuous operator on a Banach space, $`\sigma (T)`$ the usual spectrum of $`T`$, and let $`\sigma ^l(T)`$, $`\sigma ^{bb}(T)`$, $`\sigma ^c(T)`$ be computed with respect to the weak topology. It is known that an operator on a Banach space is weak-to-weak continuous if and only if it is norm-to-norm continuous; therefore it follows that $`\sigma ^c(T)=\sigma (T)`$. Furthermore, $`\sigma ^l(T)`$ does not depend on the topology, so that it also coincides with $`\sigma (T)`$. Thus $`\sigma ^l(T)=\sigma ^{bb}(T)=\sigma ^c(T)=\sigma (T)`$.
## 4. Spectral radii of an operator
The spectral radius of a bounded linear operator $`T`$ on a Banach space is usually defined via the Gelfand formula $`r(T)=\underset{n\mathrm{}}{lim}\sqrt[n]{T^n}`$. The formula involves a norm and so makes no sense in a general topological vector space. Fortunately, this formula can be rewritten without using a norm, and then generalized to topological vector spaces. Similarly to the situation with spectra, this generalization can be done in several ways, so that we will obtain various types of spectral radii for an operator on a topological vector space. We will show later that, as with the Banach space case, there are some relations between the spectral radii, the radii of the spectra, and the convergence of the Neumann series of an operator on a locally convex topological vector space. The content of this section may look technical at the beginning, but later on the reader will see that all the facts lead to a simple and natural classification. We start with an almost obvious numerical lemma.
###### Lemma 4.1.
If $`(t_n)`$ is a sequence in $`^+\{\mathrm{}\}`$, then
$$\underset{n\mathrm{}}{lim\; sup}\sqrt[n]{t_n}=inf\{\nu >0:\underset{n\mathrm{}}{lim}\frac{t_n}{\nu ^n}=0\}=inf\{\nu >0:\underset{n\mathrm{}}{lim\; sup}\frac{t_n}{\nu ^n}<\mathrm{}\}.$$
###### Proof.
Suppose $`\underset{n\mathrm{}}{lim\; sup}\sqrt[n]{t_n}=r`$. If $`0<\nu <r`$, then $`\sqrt[n_k]{t_{n_k}}>\mu >\nu `$ for some $`\mu `$ and some subsequence $`(t_{n_k})`$, so that $`\frac{t_{n_k}}{\nu ^{n_k}}>\frac{\mu ^{n_k}}{\nu ^{n_k}}\mathrm{}`$ as $`k`$ goes to infinity. It follows that $`lim\; suplim_n\mathrm{}\frac{t_n}{\nu ^n}=\mathrm{}`$. On the other hand, if $`r`$ is finite and $`\nu >r`$ then $`\sqrt[n]{t_n}<\mu <\nu `$ for some $`\mu `$ and for all sufficiently large $`n`$. Then $`\underset{n\mathrm{}}{lim}\frac{t_n}{\nu ^n}\underset{n\mathrm{}}{lim}\frac{\mu ^n}{\nu ^n}=0`$. ∎
This lemma implies that the spectral radius $`r(T)`$ of a (norm) continuous operator $`T`$ on a Banach space equals the infimum of all positive real scalars $`\nu `$ such that the sequence $`\left(\frac{T^n}{\nu ^n}\right)`$ converges to zero (or just is bounded) in operator norm topology. This can be considered as an alternative definition of the spectral radius, and can be generalized to any topological vector space. Since for each of the five considered classes of operators on topological vector spaces we introduced appropriate concepts of convergent and bounded sequences, we arrive to the following definition.
###### Definition 4.2.
Given a linear operator $`T`$ on a topological vector space $`X`$, define the following numbers:
$`r_l(T)`$ $`=`$ $`inf\{\nu >0:\text{the sequence }\left(\frac{T^n}{\nu ^n}\right)\text{ converges strongly to zero}\};`$
$`r_{bb}(T)`$ $`=`$ $`inf\{\nu >0:\frac{T^n}{\nu ^n}0\text{ uniformly on every bounded set}\};`$
$`r_c(T)`$ $`=`$ $`inf\{\nu >0:\frac{T^n}{\nu ^n}0\text{ equicontinuously }\};`$
$`r_{nn}(T)`$ $`=`$ $`inf\{\nu >0:\left(\frac{T^n}{\nu ^n}\right)\text{ nn-converges to zero}\};`$
$`r_{nb}(T)`$ $`=`$ $`inf\{\nu >0:\frac{T^n}{\nu ^n}0\text{ uniformly on some 0-neighborhood}\}.`$
The following proposition explains the relations between the introduced radii.
###### Proposition 4.3.
If $`T`$ is a linear operator on a topological vector space $`X`$, then $`r_l(T)r_{bb}(T)r_c(T)r_{nn}(T)r_{nb}(T)`$.
###### Proof.
Let $`T`$ be a linear operator on a topological vector space $`X`$. Since every singleton is bounded then $`r_l(T)r_{bb}(T)`$. Next, assume $`\nu >r_c(T)`$, fix $`\mu `$ such that $`r_c(T)<\mu <\nu `$, then the sequence $`(\frac{T^n}{\mu ^n})`$ converges to zero equicontinuously. Take a bounded set $`A`$ and a zero neighborhood $`U`$. There exists a zero neighborhood $`V`$ and a positive integer $`N`$ such that $`\frac{T^n}{\mu ^n}(V)U`$ whenever $`nN`$. Also, $`A\alpha V`$ for some $`\alpha >0`$, so that $`\frac{T^n}{\nu ^n}(A)\frac{\mu ^n}{\nu ^n}\frac{T^n}{\mu ^n}(\alpha V)\frac{\mu ^n\alpha }{\nu ^n}UU`$ for all sufficiently large $`n`$. It follows that the sequence $`(\frac{T^n}{\nu ^n})`$ converges to zero uniformly on $`A`$ and, therefore, $`\nu r_{bb}(T)`$. Thus, $`r_{bb}(T)r_c(T)`$.
To prove the inequality $`r_c(T)r_{nn}(T)`$ we let $`\nu >r_{nn}(T)`$. Then for some base $`𝒩_0`$ of zero neighborhoods and for every $`V𝒩_0`$ and $`\epsilon >0`$ there exists a positive integer $`N`$ such that $`\frac{T^n}{\nu ^n}(V)\epsilon V`$ for every $`nN`$. Given a zero neighborhood $`U`$, we can find $`V𝒩_0`$ such that $`VU`$. Then $`\frac{T^n}{\nu ^n}(V)\epsilon V\epsilon U`$ for every $`nN`$, so that the sequence $`(\frac{T^n}{\nu ^n})`$ converges to zero equicontinuously, and, therefore, $`\nu r_c(T)`$.
Finally, we must show that $`r_{nn}(T)r_{nb}(T)`$. Suppose that $`\nu >r_{nb}(T)`$, we claim that $`\nu r_{nn}(T)`$. Take $`\mu `$ so that $`\nu >\mu >r_{nb}(T)`$. One can find a zero neighborhood $`U`$ such that for every zero neighborhood $`V`$ there is a positive integer $`N`$ such that $`\frac{T^n}{\mu ^n}(U)V`$ for every $`nN`$. Fix a base $`𝒩_0`$ of zero neighborhoods, and define a new base $`\stackrel{~}{𝒩}_0`$ of zero neighborhoods via $`\stackrel{~}{𝒩}_0=\{mUW:m,W𝒩_0\}`$. Let $`V\stackrel{~}{𝒩}_0`$ and $`\epsilon >0`$. Then $`V=mUW`$ for some positive integer $`m`$ and $`W𝒩_0`$. Then $`\frac{T^n}{\mu ^n}(V)m\frac{T^n}{\mu ^n}(U)mV`$ and for every sufficiently large $`n`$, so that $`\frac{T^n}{\nu ^n}(V)\frac{\mu ^n}{\nu ^n}mV\epsilon V`$, for each sufficiently large $`n`$, which implies $`\nu r_{nn}(T)`$. ∎
The following lemma shows that, similarly to the case of Banach spaces, one can use boundedness instead of convergence when defining the spectral radii of an operator on a topological vector space. This gives alternative ways of computing the radii, which are often more convenient.
###### Lemma 4.4.
Let $`T`$ be a linear operator on a topological vector space, then
1. $`r_l(T)=inf\{\nu >0:\left(\frac{T^nx}{\nu ^n}\right)\text{ is bounded for every }x\};`$
2. if $`T`$ is bb-bounded then $`r_{bb}(T)=inf\{\nu >0:\left(\frac{T^n}{\nu ^n}\right)\text{ is uniformly bounded on every bounded set}\};`$
3. if $`T`$ is continuous then $`r_c(T)=inf\{\nu >0:\left(\frac{T^n}{\nu ^n}\right)\text{ is equicontinuous}\};`$
4. if $`T`$ is nn-bounded then $`r_{nn}(T)=inf\{\nu >0:\left(\frac{T^n}{\nu ^n}\right)\text{ is uniformly nn-bounded}\};`$
5. if $`T`$ is nb-bounded then $`r_{nb}(T)=inf\{\nu >0:\left(\frac{T^n}{\nu ^n}\right)\text{ is uniformly bounded on some 0-neighborhood}\}.`$
Moreover, in each of these cases it suffices to consider any tail of the sequence $`\left(\frac{T^n}{\nu ^n}\right)`$ instead of the whole sequence.
###### Proof.
To prove (i) let
$$r_l^{}(T)=inf\{\nu >0:\left(\frac{T^n}{\nu ^n}x\right)\text{ is bounded for every }x\}.$$
Since every convergent sequence is bounded, we certainly have $`r_l(T)r_l^{}(T)`$. Conversely, suppose $`\nu >r_l^{}(T)`$, and take any positive scalar $`\mu `$ such that $`\nu >\mu >r_l^{}(T)`$. Then for every $`xX`$ the sequence $`\frac{T^n}{\mu ^n}x`$ is bounded, and it follows that the sequence $`\frac{T^nx}{\nu ^n}=\frac{\mu ^n}{\nu ^n}\frac{T^nx}{\mu ^n}`$ converges to zero, so that $`\nu r_l(T)`$ and, therefore $`r_l^{}(T)r_l(T)`$.
To prove (ii), suppose $`T`$ is bb-bounded, and let
$$r_{bb}^{}(T)=inf\{\nu >0:\left(\frac{T^n}{\nu ^n}\right)\text{ is uniformly bounded on every bounded set}\}.$$
We’ll show that $`r_{bb}^{}(T)=r_{bb}(T)`$. If $`\left(\frac{T^n}{\nu ^n}\right)`$ converges to zero uniformly on every bounded set, then for each bounded set $`A`$ and for each zero neighborhood $`U`$ there exists a positive integer $`N`$ such that $`\frac{T^n}{\nu ^n}(A)U`$ whenever $`nN`$. Also, since $`T`$ is bb-bounded, then for every $`n<N`$ we have $`\frac{T^n}{\nu ^n}(A)\alpha _nU`$ for some $`\alpha _n>0`$. Therefore, if $`\alpha =\mathrm{max}\{\alpha _1,\mathrm{},\alpha _{N1},1\}`$, then $`\frac{T^n}{\nu ^n}(A)\alpha U`$ for every $`n`$, so that the sequence $`\frac{T^n}{\nu ^n}`$ is uniformly bounded on $`A`$. Thus $`\nu r_{bb}^{}(T)`$, so that $`r_{bb}^{}(T)r_{bb}(T)`$.
Now suppose $`\nu >r_{bb}^{}(T)`$. There exists $`\mu `$ such that $`\nu >\mu >r_{bb}^{}(T)`$. The set $`_{n=1}^{\mathrm{}}\frac{T^n}{\mu ^n}(A)`$ is bounded for every bounded set $`A`$, so that for every zero neighborhood $`U`$ there exists a scalar $`\alpha `$ such that $`\frac{T^n}{\mu ^n}(A)\alpha U`$ for every $`n`$. Then $`\frac{T^n}{\nu ^n}(A)\frac{\mu ^n\alpha }{\nu ^n}UU`$ for all sufficiently large $`n`$. This means that the sequence $`\left(\frac{T^n}{\nu ^n}\right)`$ converges to zero uniformly on $`A`$, and it follows that $`\nu r_{bb}(T)`$.
Further, if $`T`$ is bb-bounded, then any finite initial segment $`(\frac{T^n}{\nu ^n})_{n=0}^N`$ is always uniformly bounded on every bounded set, so that a tail $`(\frac{T^n}{\nu ^n})_{n=N}^{\mathrm{}}`$ is uniformly bounded on every bounded set if and only if the whole sequence $`(\frac{T^n}{\nu ^n})_{n=0}^{\mathrm{}}`$ is uniformly bounded on every bounded set.
The statements (iii), (iv), and (v) can be proved in a similar way. ∎
###### 4.5.
Locally bounded spaces. If $`T`$ is a linear operator on a locally bounded topological vector space $`(X,U)`$, then it follows directly from Definition 4.2 that $`r_{bb}(T)=r_c(T)=r_{nn}(T)=r_{nb}(T)`$, because the corresponding convergences are equivalent. In this case we would denote each of these radii by $`r_U(T)`$.
Spectral radii via seminorms
The following proposition provides formulas for computing spectral radii of an operator on a locally convex space in terms of seminorms.
###### Proposition 4.6.
If $`T`$ is an operator on a locally convex space $`X`$ with a generating family of seminorms $`𝒫`$, then
1. $`r_l(T)=\underset{p𝒫,xX}{sup}\underset{n\mathrm{}}{lim\; sup}\sqrt[n]{p(T^nx)}`$;
2. $`r_{bb}(T)=\underset{p,q𝒫}{sup}\underset{n\mathrm{}}{lim\; sup}\sqrt[n]{𝔪_{pq}(T^n)}`$, where $``$ is the collection of Minkowski functionals of all balanced convex bounded sets in $`X`$;
3. $`r_c(T)=\underset{q𝒫}{sup}\underset{p𝒫}{inf}\underset{n\mathrm{}}{lim\; sup}\sqrt[n]{𝔪_{pq}(T^n)}`$;
4. $`r_{nn}(T)=\underset{𝒬}{inf}\underset{p𝒬}{sup}\underset{n\mathrm{}}{lim\; sup}\sqrt[n]{p(T^n)}`$, where the infimum is taken over all generating families of seminorms;
5. $`r_{nb}(T)=\underset{p𝒫}{inf}\underset{q𝒫}{sup}\underset{n\mathrm{}}{lim\; sup}\sqrt[n]{𝔪_{pq}(T^n)}`$;
###### Proof.
It follows from the definition of $`r_l(T)`$ and Lemma 4.1 that
$$\begin{array}{c}r_l(T)=inf\{\nu >0:\underset{n\mathrm{}}{lim}p\left(\frac{T^nx}{\nu ^n}\right)=0\text{ for every }xX,p𝒫\}\hfill \\ \hfill =\underset{xX,p𝒫}{sup}inf\{\nu >0:\underset{n\mathrm{}}{lim}\frac{p(T^nx)}{\nu ^n}=0\}=\underset{xX,p𝒫}{sup}\underset{n\mathrm{}}{lim\; sup}\sqrt[n]{p(T^nx)}.\end{array}$$
Similarly, since the balanced convex hull of a bounded set is bounded,
$$\begin{array}{c}r_{bb}(T)=inf\{\nu >0:\text{ bounded }AV𝒩_0NnN\frac{T^n}{\nu ^n}(A)V\}\hfill \\ \hfill =inf\{\nu >0:pq𝒫NnN𝔪_{pq}\left(\frac{T^n}{\nu ^n}\right)1\}\\ \hfill =\underset{p,q𝒫}{sup}inf\{\nu >0:\underset{n\mathrm{}}{lim}\frac{𝔪_{pq}(T^n)}{\nu ^n}1\}=\underset{p,q𝒫}{sup}\underset{n\mathrm{}}{lim\; sup}\sqrt[n]{𝔪_{pq}(T^n)}.\end{array}$$
Let $`U_p=\{xX:p(x)<1\}`$ for every $`p𝒫`$. Then, rephrasing the definition of $`r_c(T)`$ and applying Lemma 4.1, we have
$$\begin{array}{c}r_c(T)=inf\{\nu >0:q𝒫p𝒫\epsilon >0NnN\frac{T^n}{\nu ^n}(U_p)\epsilon U_q\}\hfill \\ \hfill =\underset{q𝒫}{sup}\underset{p𝒫}{inf}inf\{\nu >0:\epsilon >0NnN𝔪_{pq}\left(\frac{T^n}{\nu ^n}\right)<\epsilon \}\\ \hfill =\underset{q𝒫}{sup}\underset{p𝒫}{inf}inf\{\nu >0:\underset{n\mathrm{}}{lim}\frac{𝔪_{pq}(T^n)}{\nu ^n}=0\}=\underset{q𝒫}{sup}\underset{p𝒫}{inf}\underset{n\mathrm{}}{lim\; sup}\sqrt[n]{𝔪_{pq}(T^n)}.\end{array}$$
Similarly,
$$\begin{array}{c}r_{nn}(T)=inf\{\nu >0:𝒬p𝒬\epsilon >0Nn>N\frac{T^n}{\nu ^n}(U_p)\epsilon U_p\}\hfill \\ \hfill =\underset{𝒬}{inf}\underset{p𝒬}{sup}inf\{\nu >0:\underset{n\mathrm{}}{lim}\frac{p(T^n)}{\nu ^n}=0\}=\underset{𝒬}{inf}\underset{p𝒬}{sup}\underset{n\mathrm{}}{lim\; sup}\sqrt[n]{p(T^n)}.\end{array}$$
Finally,
$$\begin{array}{c}r_{nb}(T)=inf\{\nu >0:p𝒫q𝒫Nn>N\frac{T^n}{\nu ^n}(U_p)U_q\}\hfill \\ \hfill =\underset{p𝒫}{inf}\underset{q𝒫}{sup}inf\{\nu >0:\underset{n\mathrm{}}{lim\; sup}\frac{𝔪_{pq}(T^n)}{\nu ^n}1\}=\underset{p𝒫}{inf}\underset{q𝒫}{sup}\underset{n\mathrm{}}{lim\; sup}\sqrt[n]{𝔪_{pq}(T^n)}.\end{array}$$
Some special properties of $`r_c(T)`$
Continuity of an operator can be characterized in terms of neighborhoods (the preimage of every neighborhood contains a neighborhood) or, alternatively, in terms of convergence (every convergent net is mapped to a convergent net). Analogously, though defined in terms of neighborhoods, $`r_c(T)`$ can also be characterized in terms of convergent nets. This approach was used by F. Garibay and R. Vera in a series of papers \[GBVM97, GBVM98, VM97\]. Recall that a net $`(x_\alpha )`$ in a topological vector space is said to be ultimately bounded if every zero neighborhood absorbs some tail of the net, i.e., for every zero neighborhood $`V`$ one can find an index $`\alpha _0`$ and a positive real $`\delta >0`$ such that $`x_\alpha \delta V`$ whenever $`\alpha >\alpha _0`$. As far as we know, ultimately bounded sequences were first studied in \[DeV71\] for certain locally-convex topologies. The relationship between ultimately bounded nets and convergence of sequences of operators on locally convex spaces was studied in \[GBVM97, GBVM98, VM97\]. The following proposition (which is, in fact, a version of \[VM97, Corollary 2.14\]) shows how $`r_c(T)`$ can be characterized in terms of the action of powers of $`T`$ on ultimately bounded sequences. It also implies that $`r_c(T)`$ coincides with the number $`\gamma (T)`$ which was introduced in \[GBVM97, GBVM98, VM97\] for a continuous operator on locally convex spaces.
###### Proposition 4.7.
Let $`T`$ be a linear operator on a topological vector space $`X`$, then
$$\begin{array}{c}r_c(T)=inf\{\nu >0:\underset{n,\alpha }{lim}\frac{T^n}{\nu ^n}x_\alpha =0\text{ whenever }(x_\alpha )\text{ is ultimately bounded}\}\hfill \\ \hfill =inf\{\nu >0:\left(\frac{T^n}{\nu ^n}x_\alpha \right)_{n,\alpha }\text{ is ultimately bounded whenever }(x_\alpha )\text{ is ultimately bounded}\}.\end{array}$$
###### Proof.
To prove the first equality it suffices to show that $`r_c(T)<1`$ if and only if $`\underset{n,\alpha }{lim}T^nx_\alpha =0`$ whenever $`(x_\alpha )`$ is an ultimately bounded net. Suppose that $`r_c(T)<1`$, and let $`V`$ be a zero neighborhood. One can find a zero neighborhood $`U`$ such that for every $`\epsilon >0`$ there exists $`n_0`$ such that $`T^n(U)\epsilon V`$ for each $`n>n_0`$. Let $`(x_\alpha )`$ be an ultimately bounded net. There exists an index $`\alpha _0`$ and a number $`\delta >0`$ such that $`x_\alpha \delta U`$ whenever $`\alpha >\alpha _0`$. Then for $`\epsilon =\delta ^1`$ one can find $`n_0`$ such that $`T^n(U)\delta ^1V`$ for each $`n>n_0`$, so that $`T^nx_\alpha \delta T^n(U)V`$ whenever $`\alpha >\alpha _0`$ and $`n>n_0`$. This means that $`\underset{n,\alpha }{lim}T^nx_\alpha =0`$.
Conversely, suppose that $`\underset{n,\alpha }{lim}T^nx_\alpha =0`$ for each ultimately bounded net $`(x_\alpha )`$, and assume that $`T^n`$ does not converge equicontinuously to zero. Then there exists a zero neighborhood $`V`$ such that for every zero neighborhood $`U`$ one can find $`\epsilon _U`$ such that for every $`m`$ there exists $`n_{U,m}>m`$ with $`T^{n_{U,m}}(U)\epsilon _UV`$. Then there exists $`x_{U,m}U`$ such that
(1)
$$T^{n_{U,m}}x_{U,m}\epsilon _UV.$$
The collection of all zero neighborhood ordered by inclusion is a directed set, so that $`(x_{U,n})`$ is an ultimately bounded net. Indeed, if $`W`$ is a zero neighborhood then $`x_{U,n}W`$ for each zero neighborhood $`UW`$ and every $`n`$. But it follows from (1) that the net $`\left(T^nx_{U,m}\right)_{n,m,U}`$ does not converge to zero.
To prove the second equality, let
$`\gamma _1`$ $`=`$ $`inf\{\nu >0:\underset{n,\alpha }{lim}\frac{T^n}{\nu ^n}x_\alpha =0\text{ whenever }(x_\alpha )\text{ is ultimately bounded}\}\text{ and}`$
$`\gamma _2`$ $`=`$ $`inf\{\nu >0:\left(\frac{T^n}{\nu ^n}x_\alpha \right)_{n,\alpha }\text{ is ultimately bounded if }(x_\alpha )\text{ is ultimately bounded}\}.`$
Since every net which converges to zero is necessarily ultimately bounded, it follows that $`\gamma _1\gamma _2`$. Now let $`\nu >\gamma _2`$, and let $`(x_\alpha )`$ be an ultimately bounded sequence. Suppose that $`\gamma _2<\mu <\nu `$, then $`\left(\frac{T^n}{\mu ^n}x_\alpha \right)_{n,\alpha }`$ is ultimately bounded, that is, for each zero neighborhood $`V`$ there exists an indices $`\alpha _0`$ and $`n_0`$ and a positive $`\epsilon `$ such that $`\frac{T^n}{\mu ^n}x_\alpha \epsilon V`$ whenever $`\alpha >\alpha _0`$ and $`n>n_0`$. It follows that $`\frac{T^n}{\nu ^n}x_\alpha \frac{\mu ^n\epsilon }{\nu ^n}VV`$ for $`\alpha >\alpha _0`$ and all sufficiently large $`n`$. This implies that $`\underset{n,\alpha }{lim}\frac{T^n}{\nu ^n}x_\alpha =0`$ so that $`\nu \gamma _1`$. ∎
###### Question.
Are there similar ways for computing $`r_l(T)`$, $`r_{bb}(T)`$, $`r_{nn}(T)`$, and $`r_{nb}(T)`$ in terms of nets?
Proposition 4.7 enables us to prove some important properties of $`r_c`$. The following lemma is analogous to Lemma 3.13 of \[VM97\].
###### Lemma 4.8.
If $`S`$ and $`T`$ are two commuting linear operators on a topological vector space $`X`$ such that $`r_c(S)`$ and $`r_c(T)`$ are finite, then $`r_c(ST)r_c(S)r_c(T)`$.
###### Proof.
Suppose $`\mu >r_c(S)`$ and $`\nu >r_c(T)`$ and let $`(x_\alpha )`$ be an ultimately bounded net in $`X`$. Then the net $`(\frac{T^nx_\alpha }{\nu ^n})_{n,\alpha }`$ is ultimately bounded by Proposition 4.7. By applying Proposition 4.7 again we conclude that $`(\frac{S^mT^nx_\alpha }{\mu ^m\nu ^n})_{m,n,\alpha }`$ converges to zero. In particular, $`\underset{n,\alpha }{lim}\frac{(ST)^nx_\alpha }{(\mu \nu )^n}=\underset{n,\alpha }{lim}\frac{S^nT^nx_\alpha }{\mu ^n\nu ^n}=0`$, and applying Proposition 4.7 one more time we get $`\mu \nu >r_c(ST)`$. ∎
###### Theorem 4.9.
If $`S`$ and $`T`$ are two commuting continuous operators on a locally convex space $`X`$ then $`r_c(S+T)r_c(S)+r_c(T)`$.
###### Proof.
Assume without loss of generality that both $`r_c(S)`$ and $`r_c(T)`$ are finite. Suppose that $`\eta >r_c(S)+r_c(T)`$ and take $`\mu >r_c(S)`$ and $`\nu >r_c(T)`$ such that $`\eta >\mu +\nu `$. Let $`(x_\alpha )`$ be an ultimately bounded net in $`X`$. By Proposition 4.7 it suffices to show that $`\underset{n,\alpha }{lim}\frac{1}{\eta ^n}(S+T)^nx_\alpha =0`$. Notice that the net $`\left(\frac{T^n}{\nu ^n}x_\alpha \right)_{n,\alpha }`$ is ultimately bounded. This implies that the net $`\left(\frac{S^m}{\mu ^m}\frac{T^n}{\nu ^n}x_\alpha \right)_{m,n,\alpha }`$ converges to zero. Fix a seminorm $`p`$, then there exist indices $`n_0`$ and $`\alpha _0`$ such that $`p(S^mT^nx_\alpha )<\mu ^m\nu ^n`$ whenever $`m,nn_0`$ and $`\alpha \alpha _0`$. Also, notice that we can split $`\eta `$ into a product of two terms $`\eta =\eta _1\eta _2`$ such that $`\eta _1>1`$ while still $`\eta _2>\mu +\nu `$. Further, if $`n>2n_0`$ and $`\alpha \alpha _0`$ then we have
$$\begin{array}{c}p\left(\frac{1}{\eta ^n}(S+T)^nx_\alpha \right)\hfill \\ \hfill \frac{1}{\eta ^n}\underset{k=0}{\overset{n_0}{}}\left(\genfrac{}{}{0pt}{}{n}{k}\right)p\left(S^kT^{nk}x_\alpha \right)+\frac{1}{\eta ^n}\underset{k=n_0+1}{\overset{nn_0}{}}\left(\genfrac{}{}{0pt}{}{n}{k}\right)p\left(S^kT^{nk}x_\alpha \right)+\frac{1}{\eta ^n}\underset{k=nn_0+1}{\overset{n}{}}\left(\genfrac{}{}{0pt}{}{n}{k}\right)p\left(S^kT^{nk}x_\alpha \right).\end{array}$$
Since $`\left(\genfrac{}{}{0pt}{}{n}{k}\right)=\frac{(nk+1)\mathrm{}(n1)n}{12\mathrm{}(k1)k}n^k`$ and $`_{k=0}^n\left(\genfrac{}{}{0pt}{}{n}{k}\right)\mu ^k\nu ^{nk}=(\mu +\nu )^n`$, we have
$$\begin{array}{c}p\left(\frac{1}{\eta ^n}(S+T)^nx_\alpha \right)\hfill \\ \hfill \frac{n^{n_0}}{\eta ^n}\underset{k=0}{\overset{n_0}{}}p\left(S^kT^{nk}x_\alpha \right)+\frac{1}{\eta ^n}\underset{k=n_0+1}{\overset{nn_0}{}}\left(\genfrac{}{}{0pt}{}{n}{k}\right)\mu ^k\nu ^{nk}+\frac{n^{n_0}}{\eta ^n}\underset{k=nn_0+1}{\overset{n}{}}p\left(S^kT^{nk}x_\alpha \right)\\ \hfill \frac{n^{n_0}}{\eta _1^n}\frac{1}{\eta _2^n}\underset{k=0}{\overset{n_0}{}}\left(p\left(T^{nk}S^kx_\alpha \right)+p\left(S^{nk}T^kx_\alpha \right)\right)+\frac{(\mu +\nu )^n}{\eta ^n}.\end{array}$$
Notice that $`\underset{n\mathrm{}}{lim}\frac{(\mu +\nu )^n}{\eta ^n}=0`$ and that $`\underset{n\mathrm{}}{lim}\frac{n^{n_0}}{\eta _1^n}=0`$. Since $`T`$ is continuous, the net $`(T^kx_\alpha )_\alpha `$ is ultimately bounded for every fixed $`k`$, so that $`\underset{n,\alpha }{lim}\frac{1}{\eta _2^{nk}}S^{nk}T^kx_\alpha =0`$. It follows that for every $`k`$ between $`0`$ and $`n_0`$ the expression $`\frac{1}{\eta _2^n}p\left(S^{nk}T^kx_\alpha \right)`$ is uniformly bounded for all sufficiently large $`n`$ and $`\alpha `$. Similarly, for every $`k`$ between $`0`$ and $`n_0`$ the expression $`\frac{1}{\eta _2^n}p\left(T^{nk}S^kx_\alpha \right)`$ is uniformly bounded for all sufficiently large $`n`$ and $`\alpha `$. Therefore there exist indices $`n_1`$ and $`\alpha _1`$ such that the finite sum
$$\frac{1}{\eta _2^n}\underset{k=0}{\overset{n_0}{}}\left(p\left(T^{nk}S^kx_\alpha \right)+p\left(S^{nk}T^kx_\alpha \right)\right)$$
is uniformly bounded for all $`nn_1`$ and $`\alpha \alpha _1`$. It follows that $`\underset{n,\alpha }{lim}p\left(\frac{1}{\eta ^n}(S+T)^nx_\alpha \right)=0`$, so that $`\eta >r_c(S+T)`$. ∎
###### Corollary 4.10.
If $`T`$ is a continuous operator on a locally convex space with finite $`r_c(T)`$ then $`r_c\left(P(T)\right)`$ is finite for every polynomial $`P(z)`$.
###### Definition 4.11.
We say that a sequence $`(x_n)`$ in a topological vector space is fast null if $`\underset{n\mathrm{}}{lim}\alpha ^nx_n=0`$ for every positive real $`\alpha `$.
###### Lemma 4.12.
If $`T`$ is a linear operator on a topological vector space with $`r_c(T)<\mathrm{}`$ then $`(T^nx_n)`$ is fast null whenever $`(x_n)`$ is fast null.
###### Proof.
Suppose $`(x_n)`$ is a fast null sequence in a topological vector space and $`r_c(T)<\mathrm{}`$. Let $`\nu >r_c(T)`$, the sequence $`\nu ^n\alpha ^nx_n`$ converges to zero, hence is ultimately bounded, then by Proposition 4.7 we have
$$\underset{n\mathrm{}}{lim}\alpha ^nT^nx_n=\underset{n\mathrm{}}{lim}\frac{T^n}{\nu ^n}\nu ^n\alpha ^nx_n=0.$$
## 5. Spectra and spectral radii
It is well known that for a continuous operator $`T`$ on a Banach space its spectral radius $`r(T)`$ equals the geometrical radius of the spectrum $`\left|\sigma (T)\right|=sup\{|\lambda |:\lambda \sigma (T)\}`$. Further, whenever $`|\lambda |>r(T)`$, the resolvent operator $`R_\lambda =(\lambda IT)^1`$ is given by the Neumann series $`_{i=0}^{\mathrm{}}\frac{T^i}{\lambda ^{i+1}}`$. We are going to show in the next five theorems that the spectral radii that we have introduced are upper bounds for the actual radii of the correspondent spectra, and that when $`|\lambda |`$ is greater than or equal to any of these spectral radii, then the Neumann series converges in the correspondent operator topology to the resolvent operator.
In the following Theorems 5.15.5 we assume that $`T`$ is a linear operator on a sequentially complete locally convex space, $`\lambda `$ is a complex number, and $`R_\lambda `$ is the resolvent of $`T`$ at $`\lambda `$ in the sense of Definition 3.1.
###### Theorem 5.1.
If $`|\lambda |>r_l(T)`$ then the Neumann series converges pointwise to a linear operator $`R_\lambda ^0`$, and $`R_\lambda ^0(\lambda IT)=I`$. Moreover, if $`T`$ is continuous, then $`R_\lambda ^0=R_\lambda `$ and $`\left|\sigma ^l(T)\right|r_l(T)`$.
###### Proof.
For any $`\lambda `$ such that $`|\lambda |>r_l(T)`$ one can find $`z`$ such that $`0<|z|<1`$ and $`\lambda z>r_l(T)`$. Consider a point $`xX`$ and a base zero neighborhood $`U`$. Since by the definition of $`r_l(T)`$ the sequence $`\left(\frac{T^nx}{(\lambda z)^n}\right)`$ converges to zero, there exist a positive integer $`n_0`$, such that $`\frac{T^nx}{(\lambda z)^n}U`$ whenever $`nn_0`$. Therefore, $`\frac{T^nx}{\lambda ^n}z^nU|z|^nU`$ because $`U`$ is balanced. Thus, if $`nmn_0`$, then $`_{i=n}^m\frac{T^ix}{\lambda ^i}_{i=n}^m|z|^iU\left(_{i=n}^m|z|^i\right)U`$ because $`U`$ is convex. Since $`|z|<1`$, we have $`_{i=n}^m|z|^i<1`$ for sufficiently large $`m`$ and $`n`$, and so $`_{i=n}^m\frac{T^ix}{\lambda ^i}U`$ because $`U`$ is balanced. Therefore $`R_{\lambda ,n}x=\frac{1}{\lambda }_{i=0}^n\frac{T^ix}{\lambda ^i}`$ is a Cauchy sequence and hence it converges to some $`R_\lambda ^0x`$ because $`X`$ is sequentially complete.
Clearly, $`R_\lambda ^0`$ is a linear operator. Notice that $`R_{\lambda ,n}(\lambda xTx)=x\frac{T^{n+1}x}{\lambda ^{n+1}}`$ for every $`x`$. As $`n`$ goes to infinity, the left hand side of this identity converges to $`R_\lambda ^0(\lambda xTx)`$, while the right hand side converges to $`x`$. Thus it follows that $`R_\lambda ^0(\lambda IT)=I`$.
Finally, notice that $`R_{\lambda ,n}`$ commutes with $`T`$ for every $`n`$. Therefore, if $`T`$ is continuous, then
$$R_\lambda ^0Tx=\underset{n\mathrm{}}{lim}R_{\lambda ,n}Tx=\underset{n\mathrm{}}{lim}TR_{\lambda ,n}x=T(\underset{n\mathrm{}}{lim}R_{\lambda ,n}x)=TR_\lambda ^0x$$
for every $`x`$. This implies that $`(\lambda IT)R_\lambda ^0=R_\lambda ^0(\lambda IT)=I`$, so that $`R_\lambda ^0`$ is the (left and right) inverse of $`\lambda IT`$. This means that $`R_\lambda ^0=R_\lambda `$ and $`\lambda \rho ^l(T)`$. Thus, $`\left|\sigma ^l(T)\right|r_l(T)`$. ∎
###### Theorem 5.2.
If $`T`$ is bb-bounded and $`|\lambda |>r_{bb}(T)`$, then the Neumann series converges uniformly on bounded sets, and its sum $`R_\lambda ^0`$ is bb-bounded. Moreover, if $`T`$ is continuous, then $`R_\lambda ^0=R_\lambda `$ and $`\left|\sigma ^{bb}(T)\right|r_{bb}(T)`$.
###### Proof.
Suppose that $`|\lambda |>r_{bb}(T)`$, then the sum $`R_\lambda ^0`$ of the Neumann series exists by Theorem 5.1. As in the proof of Theorem 5.1 we denote the partial sums of the Neumann series by $`R_{\lambda ,n}`$. Fix $`z`$ such that $`0<|z|<1`$ and $`\lambda z>r_{bb}(T)`$, and consider a bounded set $`A`$ and a closed base zero neighborhood $`U`$. Since $`\frac{T^n}{(\lambda z)^n}`$ converges to zero uniformly on $`A`$, there exits $`n_0`$ such that $`\frac{T^n}{\lambda ^nz^n}(A)U`$ for all $`n>n_0`$. Also, since $`|z|<1`$, we can assume without loss of generality that $`_{i=n_0}^{\mathrm{}}|z|^i<|\lambda |`$. Then
$$\frac{1}{\lambda }\underset{i=n+1}{\overset{m}{}}\frac{T^ix}{\lambda ^i}\frac{1}{\lambda }\left(\underset{i=n+1}{\overset{m}{}}|z|^i\right)UU$$
whenever $`xA`$ and $`m>n>n_0`$. Since $`U`$ is closed, we have
$$R_\lambda ^0xR_{\lambda ,n}x=\underset{m\mathrm{}}{lim}\frac{1}{\lambda }\underset{i=n+1}{\overset{m}{}}\frac{T^ix}{\lambda ^i}U$$
for each $`xA`$ and $`n>n_0`$, so that $`(R_\lambda ^0R_{\lambda ,n})(A)U`$ whenever $`n>n_0`$. This shows that $`R_{\lambda ,n}`$ converges to $`R_\lambda ^0`$ uniformly on bounded sets. By Lemma 2.15 this implies that $`R_\lambda ^0`$ is bb-bounded.
Further, if $`T`$ is continuous, then by Theorem 5.1 we have $`R_\lambda =R_\lambda ^0`$, so that $`\lambda \rho ^{bb}(T)`$, whence it follows that $`\left|\sigma ^{bb}(T)\right|r_{bb}(T)`$. ∎
The next theorem is similar to Theorem 2.18 of \[VM97\].
###### Theorem 5.3.
If $`T`$ is a continuous and $`|\lambda |>r_c(T)`$, then the Neumann series converges equicontinuously to $`R_\lambda `$, and $`R_\lambda `$ is continuous. In particular, $`\left|\sigma ^c(T)\right|r_c(T)`$ holds.
###### Proof.
Let $`|\lambda |>r_c(T)`$. It follows from Theorem 5.1 that the Neumann series converges to $`R_\lambda `$. Again, we denote the partial sums of the Neumann series by $`R_{\lambda ,n}`$. Let $`z`$ be such that $`0<|z|<1`$ and $`\lambda z>r_c(T)`$. For a fixed closed zero neighborhood $`U`$ there exists a zero neighborhood $`V`$ such that $`\frac{T^n}{\lambda ^nz^n}(V)U`$ for every $`n0`$. Let $`\epsilon >0`$, then $`_{i=n_0}^{\mathrm{}}|z|^i<\epsilon |\lambda |`$ for some $`n_0`$. Then
$$\frac{1}{\lambda }\underset{i=n+1}{\overset{m}{}}\frac{T^ix}{\lambda ^i}\frac{1}{\lambda }\left(\underset{i=n+1}{\overset{m}{}}|z|^i\right)\epsilon UU$$
whenever $`xV`$ and $`m>n>n_0`$. Since $`U`$ is closed, we have
$$R_\lambda xR_{\lambda ,n}x=\underset{m\mathrm{}}{lim}\frac{1}{\lambda }\underset{i=n+1}{\overset{m}{}}\frac{T^ix}{\lambda ^i}\epsilon U$$
for each $`xV`$ and $`n>n_0`$, so that $`(R_\lambda R_{\lambda ,n})(V)\epsilon U`$ whenever $`n>n_0`$. This shows that $`R_{\lambda ,n}`$ converges to $`R_\lambda `$ equicontinuously, and Lemma 2.16 yields that $`R_\lambda `$ is continuous.
###### Theorem 5.4.
If $`T`$ is nn-bounded and $`|\lambda |>r_{nn}(T)`$, then the Neumann series nn-converges to $`R_\lambda `$ and $`R_\lambda `$ is nn-bounded. In particular, $`\left|\sigma ^{nn}(T)\right|r_{nn}(T)`$ holds.
###### Proof.
Let $`|\lambda |>r_{nn}(T)`$. By Theorem 5.1 the Neumann series $`_{i=0}^{\mathrm{}}\frac{T^i}{\lambda ^{i+1}}`$ converges to $`R_\lambda `$. Again, we denote the partial sums of the Neumann series by $`R_{\lambda ,n}`$. Fix some $`z`$ such that $`0<|z|<1`$ and $`\lambda z>r_{nn}(T)`$. There exists a base $`𝒩_0`$ of closed convex zero neighborhoods such that for every $`U𝒩_0`$ there is a scalar $`\beta >0`$ such that $`\frac{T^n}{(\lambda z)^n}(U)\beta U`$ for all $`n0`$. Fix $`U𝒩_0`$, then for each $`n0`$ we have $`\frac{T^n}{\lambda ^nz^n}(U)\beta U`$ for some $`\beta >0`$, so that $`\frac{T^nx}{\lambda ^n}|z|^n\beta U`$ whenever $`xU`$. It follows that
$$R_{\lambda ,n}x=\frac{1}{\lambda }\underset{i=0}{\overset{n}{}}\frac{T^ix}{\lambda ^i}\frac{\beta }{\lambda }\left(\underset{i=0}{\overset{n}{}}|z|^i\right)U.$$
Then $`R_\lambda x\frac{\beta }{\lambda (1|z|)}U`$, so that $`R_\lambda (U)\frac{\beta }{\lambda (1|z|)}U`$, which implies that $`R_\lambda `$ is nn-bounded, and, therefore, $`\left|\sigma ^{nn}(T)\right|r_{nn}(T)`$ holds.
Fix $`\epsilon >0`$. Then $`_{i=N}^{\mathrm{}}|z|^i<|\lambda |`$ for some $`N`$. Then for every $`U𝒩_0`$ we have
$$\frac{1}{\lambda }\underset{i=n+1}{\overset{m}{}}\frac{T^ix}{\lambda ^i}\frac{1}{\lambda }\left(\underset{i=n+1}{\overset{m}{}}|z|^i\right)\epsilon UU$$
whenever $`xU`$ and $`N<n<m`$. Since $`U`$ is closed, we have
$$R_\lambda xR_{\lambda ,n}x=\underset{m\mathrm{}}{lim}\frac{1}{\lambda }\underset{i=n+1}{\overset{m}{}}\frac{T^ix}{\lambda ^i}\epsilon U$$
for each $`xU`$ and $`n>N`$, so that $`(R_\lambda R_{\lambda ,n})(U)\epsilon U`$ whenever $`N<n`$. This shows that $`R_{\lambda ,n}`$ nn-converges to $`R_\lambda `$. ∎
###### Theorem 5.5.
If $`T`$ is nb-bounded and $`|\lambda |>r_{nb}(T)`$, then the Neumann series converges to $`R_\lambda `$ uniformly on a zero neighborhood. Further, $`\left|\sigma ^{nb}(T)\right|r_{nb}(T)`$ holds.
###### Proof.
Let $`|\lambda |>r_{nb}(T)`$. By Theorem 5.1 the Neumann series $`_{i=0}^{\mathrm{}}\frac{T^i}{\lambda ^{i+1}}`$ converges to $`R_\lambda `$. Since $`r_{bb}(T)r_{nb}(T)`$ then $`R_\lambda `$ is bb-bounded by Theorem 5.2. But then $`_{i=0}^{\mathrm{}}\frac{T^i}{\lambda ^{i+1}}=\frac{1}{\lambda }I+\frac{1}{\lambda }R_\lambda T`$. Notice that $`R_\lambda T`$ is nb-bounded as a product of a bb-bounded and an nb-bounded operators (see 2.6).
Suppose that $`|\lambda |>r_{nb}(T)`$. Fix $`z`$ such that $`0<|z|<1`$ and $`\lambda z>r_{nb}(T)`$, then the sequence $`\left(\frac{T^n}{\lambda ^nz^n}\right)`$ converges to zero uniformly on some base zero neighborhood $`U`$. We will show that the Neumann series converges uniformly on $`U`$. As in the proof of Theorem 5.1, we denote the partial sums of the Neumann series by $`R_{\lambda ,n}`$. Fix a closed base zero neighborhood $`V`$. Since $`\left(\frac{T^n}{\lambda ^nz^n}\right)`$ converges to zero uniformly on $`U`$, there exits $`n_0`$ such that $`\frac{T^n}{\lambda ^nz^n}(U)V`$ for all $`n>n_0`$. Also, since $`|z|<1`$, we can assume without loss of generality that $`_{i=n_0}^{\mathrm{}}|z|^i<|\lambda |`$. Then
$$\frac{1}{\lambda }\underset{i=n+1}{\overset{m}{}}\frac{T^ix}{\lambda ^i}\frac{1}{\lambda }\left(\underset{i=n+1}{\overset{m}{}}|z|^i\right)VV$$
whenever $`xA`$ and $`m>n>n_0`$. Since $`V`$ is closed, we have
$$R_\lambda xR_{\lambda ,n}x=\underset{m\mathrm{}}{lim}\frac{1}{\lambda }\underset{i=n+1}{\overset{m}{}}\frac{T^ix}{\lambda ^i}V$$
for each $`xU`$ and $`n>n_0`$, so that $`(R_\lambda R_{\lambda ,n})(U)V`$ whenever $`n>n_0`$. ∎
In rest of this section we present some remarks on Theorems 5.15.5. In particular, we discuss the conditions of sequential completeness and the local convexity and consider several examples and special cases.
###### 5.6.
It is easy to see that each spectral radius is exactly the radius of convergence of the Neumann series in the correspondent operator convergence. Indeed, in each of Theorems 5.15.5 the convergence of the Neumann series implies that the terms of the series tend to zero. It follows that $`|\lambda |`$ is greater than or equal to the corresponding spectral radius.
Clearly, if $`X`$ is a Banach space, then the norm topology on $`X`$ and the weak topology on $`X^{}`$ are sequentially complete. The weak topology of $`X`$ is sequentially complete if $`X`$ is reflexive. Also, it is known that the weak topologies of $`\mathrm{}_1`$ and of $`L_1[0,1]`$ are sequentially complete. Since all these topologies are locally convex, Theorems 5.15.5 are applicable to each of them.
###### 5.7.
Monotone convergence property. Notice that if $`T`$ is a positive operator on a locally convex-solid vector lattice (i.e., a locally convex space which is also a vector lattice such that $`|x||y|`$ implies $`p(x)p(y)`$ for every generating seminorm $`p`$) then we can substitute the sequential completeness condition in Theorems 5.15.5 by a weaker condition called sequential monotone completeness property: a locally convex-solid vector lattice is said to satisfy the sequential monotone completeness property if every monotone Cauchy sequence converges in the topology of $`X`$. For details, see \[AB78\]. Indeed, we used the sequential completeness at just one point — we used it in the proof of Theorem 5.1 to claim that since $`R_{\lambda ,n}x=\frac{1}{\lambda }_{i=0}^n\frac{T^ix}{\lambda ^i}`$ is a Cauchy sequence, then it converges to some $`R_\lambda x`$. But if $`T`$ is positive, then $`R_{\lambda ,n}x^+`$ and $`R_{\lambda ,n}x^{}`$ are increasing sequences, and the sequential monotone completeness property ensures the convergence.
###### 5.8.
Pointwise convergence. It can be easily verified that the space of continuous functions on $`[0,1]`$ with pointwise convergence topology is not sequentially complete, the sequence $`x_n(t)=t^n`$ is a counterexample. The same counterexample shows that this space does not have the monotone convergence property either.
Consider the sequence spaces $`\mathrm{}_p`$ for $`0<p\mathrm{}`$, $`c`$, $`c_0`$, and $`c_{00}`$ (the space of eventually vanishing sequences). None of these spaces is sequentially complete in the topology of coordinate-wise convergence: take the following sequence for a counterexample:
$$x_n(i)=\{\begin{array}{cc}i\hfill & \text{if }i<n;\hfill \\ 0\hfill & \text{otherwise}.\hfill \end{array}$$
The same example shows that these spaces do not have the monotone convergence property either. Therefore neither of Theorems 5.15.5 or 5.7 can be applied.
###### Example 5.9.
Theorems 5.15.5 fail without sequential completeness. Consider the space $`c_0`$ with the topology of coordinate-wise convergence. Let $`T`$ be the forward shift operator on $`c_0`$, that is, $`Te_k=e_{k+1}`$, where $`e_k`$ is the $`k`$-th unit vector of $`c_0`$. Let $`V`$ be any base zero neighborhood, we can assume without loss of generality that $`V=\{xc_0:|x_{i_1}|<1,\mathrm{},|x_{i_k}|<1\}`$ where $`i_1<i_2<\mathrm{}<i_k`$ are positive integers. If $`xU`$ then $`T^nx`$ has zero components $`1`$ through $`n`$, in particular for every positive $`\nu `$ we have $`\frac{T^nx}{\nu ^n}V`$ whenever $`n>i_k`$. Therefore $`\left(\frac{T^n}{\nu ^n}\right)`$ converges uniformly on $`c_0`$ for every $`\nu >0`$, so that $`r_{nb}(T)=0`$. It follows from Proposition 4.3 that $`r_l(T)=r_{bb}(T)=r_c(T)=r_{nn}(T)=0`$. On the other hand, $`_{n=1}^{\mathrm{}}T^ne_1`$ diverges in $`c_0`$. Since $`T`$ is obviously continuous, this shows that Theorems 5.15.5, do not hold in $`c_0`$. Thus, sequential completeness condition is essential in the theorems.
###### 5.10.
Banach spaces. If $`T`$ is a (norm) continuous operator on a Banach space, then it follows from 3.2 and 4.5 that $`\sigma ^l(T)=\sigma ^{bb}(T)=\sigma ^c(T)=\sigma ^{nn}(T)=\sigma ^{nb}(T)=\sigma (T)`$ and $`r_{bb}(T)=r_c(T)=r_{nn}(T)=r_{nb}(T)=r(T)`$, where $`\sigma (T)`$ and $`r(T)`$ are the usual spectrum and the spectral radius of $`T`$. Further, it follows from Lemma 4.3 that $`r_l(T)r(T)`$. On the other hand, since $`r(T)=\left|\sigma (T)\right|`$, then $`r(T)r_l(T)`$ by Theorem 5.1, so that $`r_l(T)=r(T)`$.
###### 5.11.
The following argument is a counterpart to 3.5. Let $`T`$ be a (norm) continuous operator on a Banach space $`X`$ and $`r(T)`$ the usual spectral radius of $`T`$, while $`r_l(T)`$ and $`r_{bb}(T)`$ be computed with respect to the weak topology of $`X`$. We claim that if the weak topology of $`X`$ is sequentially complete, then $`r_l(T)=r_{bb}(T)=r(T)`$. Indeed, $`r(T)r_l(T)`$ by 3.5 and Theorem 5.1 because $`\sigma (T)=\sigma ^l(T)`$. In view of Proposition 4.3 it suffices to show that $`r_{bb}(T)r(T)`$. Let $`\nu >r(T)`$, and let $`A`$ be a weakly bounded subset of $`X`$. Then $`A`$ is norm bounded, so that the sequence $`\frac{T^n}{\nu ^n}`$ converges to zero uniformly on $`A`$ in the norm topology. In particular, the set $`_{n=0}^{\mathrm{}}\frac{T^n}{\nu ^n}(A)`$ is norm bounded, hence weakly bounded, so that $`\nu >r_{bb}(T)`$.
Quasinilpotence
Recall that a norm continuous operator $`T`$ on a Banach space $`X`$ is said to be quasinilpotent if $`r(T)=0`$ or, equivalently, if $`\sigma (T)=\{0\}`$. Quasinilpotent operators on Banach spaces have some nice properties, therefore in the framework of topological vector spaces it is interesting to study operators having some of their spectra trivial or some of their spectral radii being zero. Notice, for example, that it follows from Proposition 4.6 that if $`T`$ is an operator on a locally convex topological vector space, then $`r_l(T)=0`$ if and only if $`\underset{n\mathrm{}}{lim}\sqrt[n]{p(T^nx)}=0`$ for every seminorm $`p`$ in a generating family of seminorms and for every $`xX`$. Further, if the space is in addition sequentially complete, then for such an operator we would have $`\sigma ^l(T)=\{0\}`$ by Theorem 5.1.
Recall also that a norm continuous operator $`T`$ on a Banach space $`X`$ is said to be locally quasinilpotent at a point $`xX`$ if $`\underset{n\mathrm{}}{lim}\sqrt[n]{T^nx}=0`$. Using Lemma 4.1, the concept of local quasinilpotence can be naturally generalized to topological vector spaces: an operator $`T`$ on a topological vector space $`X`$ is said to be locally quasinilpotent at a point $`xX`$ if $`\underset{n\mathrm{}}{lim}\frac{T^nx}{\nu ^n}=0`$ for every $`\nu >0`$. It follows immediately from the definition of $`r_l(T)`$ that $`r_l(T)=0`$ if and only if $`T`$ is locally quasinilpotent at every $`xX`$. It is known that a continuous operator on a Banach space is quasinilpotent if and only if it is locally quasinilpotent at every point. We see now that this is just a corollary of 5.10. The following example shows that a similar result for general topological vector spaces is not valid, that is, $`r_l(T)`$ may be equal to zero without the other radii be equal to zero.
###### Example 5.12.
A continuous operator with $`r_l(T)=0`$ but $`r_{bb}(T)=r_c(T)=r_{nn}(T)=r_{nb}(T)=\mathrm{}`$. Consider the space of all bounded real sequences $`\mathrm{}_{\mathrm{}}=\{x=(x_1,x_2,\mathrm{}):sup|x_k|<\mathrm{}\}`$ with the topology of coordinate-wise convergence. This topology can be generated by the family of coordinate seminorms $`\{p_m\}_{m=1}^{\mathrm{}}`$ where $`p_m(x)=|x_m|`$. Let $`e_k`$ denote the $`k`$-th unit vector in $`\mathrm{}_{\mathrm{}}`$.
Define an operator $`T:\mathrm{}_{\mathrm{}}\mathrm{}_{\mathrm{}}`$ via $`Te_k=\frac{(k1)^{k1}}{k^k}e_{k1}`$ if $`k>1`$, and $`Te_1=0`$. Then $`T^ne_k=\frac{(kn)^{kn}}{k^k}e_{kn}`$ if $`n<k`$ and zero otherwise. Clearly $`T`$ is continuous. In order to show that $`r_l(T)=0`$ fix a positive real number $`\nu `$ and $`x\mathrm{}_{\mathrm{}}`$, then
$$\left|\left(\frac{T^nx}{\nu ^n}\right)_m\right|=\left|\frac{m^m}{(m+n)^{m+n}\nu ^n}x_{n+m}\right|\underset{n}{sup}\frac{m^m}{(m+n)^{m+n}\nu ^n}\underset{n}{sup}|x_n|<\mathrm{}$$
It follows from Lemma 4.4(i) that $`r_l(T)=0`$.
Now we show that $`r_{bb}(T)=\mathrm{}`$ by presenting a bounded set $`A`$ in $`\mathrm{}_{\mathrm{}}`$ such that the sequence $`\left(\frac{T^n}{\nu ^n}\right)`$ in not uniformly bounded on $`A`$ for every positive $`\nu `$. Let
$$A=\{x\mathrm{}_{\mathrm{}}:x_n(2n)^{2n}\text{ for all }n0\}.$$
Then $`(2n)^{2n}e_nA`$ for each $`n>0`$ and $`\left(\frac{T^{n1}}{\nu ^{n1}}(2n)^{2n}e_n\right)_1=\frac{(2n)^{2n}}{n^n\nu ^n}`$ is unbounded. Then by Lemma 4.4(ii) we have $`r_{bb}(T)=\mathrm{}`$, and it follows from Proposition 4.3 that $`r_c(T)=r_{nn}(T)=r_{nb}(T)=\mathrm{}`$.
It is not difficult to show that $`\sigma ^l(T)=\{0\}`$, while $`\sigma ^c(T)=\sigma ^{nn}(T)=\sigma ^{nb}(T)=`$.
Non-locally convex spaces
We proved the key Theorems 5.15.5 for locally convex spaces, but they are still valid for locally pseudo-convex spaces. The local convexity of $`X`$ was used only once in the proof of Theorem 5.1, while Theorems 5.25.5 used Theorem 5.1. Hence it would suffice to modify the proof of Theorem 5.1 in such a way that it would work for locally pseudo-convex spaces instead of locally convex. Local convexity was used in the proof of Theorem 5.1 to show that if $`\frac{T^nx}{(\lambda z)^n}U`$ for all $`n>n_0`$ and some $`n_0`$, then there exists $`m_0`$ such that $`_{i=n}^m\frac{T^ix}{\lambda ^i}U`$ for all $`m,n>m_0`$. (Recall that $`T`$ is a linear operator, $`\lambda ,z`$ such that $`0<|z|<1`$ and $`\lambda z>r_l(T)`$, $`xX`$, and $`U`$ is a base zero neighborhood in $`X`$.) If $`X`$ is locally pseudo-convex, then we can assume that $`U+U\alpha U`$ for some $`\alpha >0`$, so that $`(X,U)`$ is a locally bounded space. Let $``$ be the Minkowski functional of $`U`$, then (see \[KPR84, pages 3 and 6\]) for any $`x_1,x_2,\mathrm{},x_n`$ in $`X`$ we have
$$x_1+\mathrm{}+x_k4^{\frac{1}{p}}\left(x_1^p+\mathrm{}+x_k^p\right)^{\frac{1}{p}},$$
where $`2^{\frac{1}{p}}=\alpha `$. Notice that $`\frac{T^nx}{\lambda ^n}|z|^n`$ for all $`n>n_0`$. Since $`|z|<1`$, then there exists $`m_0`$ such that $`_{i=n}^m|z|^{ip}<\frac{1}{4}`$ whenever $`n,m>m_0`$. But then
$$\underset{i=n}{\overset{m}{}}\frac{T^ix}{\lambda ^i}4^{\frac{1}{p}}\left(\underset{i=n}{\overset{m}{}}\frac{T^ix}{\lambda ^i}^p\right)^{\frac{1}{p}}4^{\frac{1}{p}}\left(\underset{i=n}{\overset{m}{}}|z|^{ip}\right)^{\frac{1}{p}}<1,$$
so that $`\underset{i=n}{\overset{m}{}}\frac{T^ix}{\lambda ^i}U`$.
The following example shows that Theorems 5.1, 5.2, and 5.3 fail if we assume no convexity conditions at all.
###### Example 5.13.
An operator on a complete non locally pseudo-convex space, whose spectral radii are 1, and whose Neumann series nevertheless diverges at $`\lambda =2`$. Let $`X`$ be the space of all measurable functions on $`[0,1]`$ with the topology of convergence in measure (which is not pseudo-convex). We identify the endpoints $`0`$ and $`1`$ and consider the interval as a circle. Fix an irrational $`\alpha `$ and define a linear operator $`T`$ on $`X`$ as the translation by $`\alpha `$, i.e., $`(Tf)(t)=f(t\alpha )`$. It is easy to see that $`\frac{T^nf}{\nu ^n}`$ converges in measure to zero for every $`fX`$ if and only if $`\nu >1`$. We conclude, therefore, that $`r_l(T)=1`$. Moreover, since the sets of the form $`W_{\epsilon ,\delta }=\{fX:\mu (f>\epsilon )<\delta \}`$ form a zero neighborhood base for the topology of convergence in measure, and $`T(W_{\epsilon ,\delta })W_{\epsilon ,\delta }`$, it follows that $`r_{nn}(T)1`$. Then by Proposition 4.3 we have $`r_l(T)=r_{bb}(T)=r_c(T)=r_{nn}(T)=1`$. Nevertheless, we are going to present a function $`hX`$ such that the Neumann series $`_{n=0}^{\mathrm{}}\frac{T^nh}{2^n}`$ does not converge in measure, which means that the conclusions of Theorems 5.15.5 do not hold for this space
For each $`n=1,2,3,\mathrm{}`$ one can find a positive integer $`M_n`$ such that the intervals $`[k\alpha ,k\alpha +\frac{1}{n}]`$ (mod 1) for $`k=1,\mathrm{},M_n`$ cover the circle. Let $`s_n=_{i=1}^nM_i`$, and let $`h`$ be the step function taking value $`2^{s_n}`$ on the interval $`(\frac{1}{n+1},\frac{1}{n}]`$. If $`s_{n1}<ks_n`$ for some positive integers $`n`$ and $`k`$, then on $`[0,\frac{1}{n}]`$ we have $`h2^{s_n}2^k`$, so that $`\frac{h}{2^k}1`$ on $`[0,\frac{1}{n}]`$, and it follows that $`\frac{T^kh}{2^k}1`$ on $`[k\alpha ,k\alpha +\frac{1}{n}]`$.
Now, given any positive integer $`N`$, we have $`Ns_{n1}`$ for some $`n`$. Then for each $`k=s_{n1}+1,\mathrm{},s_n`$ we have $`\frac{T^kh}{2^k}1`$ on the interval $`[k\alpha ,k\alpha +\frac{1}{n}]`$. It follows that
$$\underset{k=s_{n1}+1}{\overset{s_n}{}}\frac{T^kh}{2^k}1\text{ on the set }\underset{k=s_{n1}+1}{\overset{s_n}{}}[k\alpha ,k\alpha +\frac{1}{n}]=s_{n1}\alpha +\underset{k=1}{\overset{M_n}{}}[k\alpha ,k\alpha +\frac{1}{n}]=[0,1],$$
so that the series $`_{n=0}^{\mathrm{}}\frac{T^nh}{2^n}`$ does not converge in measure.
Other approaches to spectral theory
There are different approaches to spectral theory of operators on topological vector spaces, e.g., \[Wae54\] and \[All65\]. For example, Allan \[All65\] defines the spectrum of an element $`x`$ of locally-convex algebra $`B`$ as the set of all $`\lambda `$ such that $`\lambda ex`$ is not invertible or the inverse is not bounded, where $`yB`$ is said to be bounded if $`\{c^ny^n\}_{n=1}^{\mathrm{}}`$ is a bounded set for some real $`c>0`$. In our terms, this means that $`R_\lambda `$ has finite spectral radius. Allan’s spectrum is, therefore, bigger than ours. Allan defines the radius of boundedness of $`\beta (x)`$, which in our terms is exactly the spectral radius, and he shows that $`\beta (x)`$ is less than or equal to the geometrical radius of his spectrum. This result nicely complements our Theorems 5.15.5 where we showed that a spectral radius of an operator is greater that or equal to the geometrical radius of the corresponding spectrum.
For example, if $`X`$ is a locally-convex space then it can be easily verified that the collection of all continuous operators on $`X`$ equipped with the topology of uniform convergence on bounded sets is a locally convex algebra. For a base of convex neighborhoods of zero in this algebra one can take the sets $`𝒱_{A,U}`$ of all continuous $`T`$ such that $`T(A)U`$, where $`AX`$ is bounded and $`U`$ is a convex base zero neighborhood in $`X`$. Therefore, the result of Allan is applicable in the setup of Theorem 5.2.
## 6. nb-bounded operators
Since nb-boundedness is the strongest of the boundedness conditions we have introduced, it is natural to expect that stronger results can be obtained for nb-bounded operators.
###### 6.1.
The following argument is often useful when dealing with nb-bounded operators. Suppose that $`X`$ and $`Y`$ are topological vector spaces and $`T:XY`$ is nb-bounded, then $`T(U)`$ is a bounded set in $`Y`$ for some base zero neighborhood $`U`$. We claim that if $`Y`$ is Hausdorff, then $`_{n=1}^{\mathrm{}}\frac{1}{n}U\mathrm{Null}T`$. Indeed, it suffices to show that if $`x\frac{1}{n}U`$ for every $`n1`$ then $`Tx`$ belongs to every zero neighborhood $`V`$ of $`Y`$. But $`T(U)\alpha V`$ for some positive $`\alpha `$ (depending on $`V`$), and hence $`Tx\frac{1}{n}T(U)\frac{\alpha }{n}VV`$ whenever $`n\alpha `$.
It follows that if $`T`$ is one-to-one, then $`U`$ cannot contain any nontrivial linear subspaces. In particular, if $`U`$ is convex then the locally bounded space $`(X,U)`$ is Hausdorff, hence quasinormable. In this case $`T`$ is a continuous operator from $`(X,U)`$ to $`Y`$, and, moreover, if $`X=Y`$, then $`T`$ is continuous as an operator from $`(X,U)`$ to $`(X,U)`$.
In fact, many “classical” topological vector spaces have the property that every zero neighborhood contains a nontrivial linear subspace, e.g., topologies of pointwise or coordinate-wise convergence, weak topologies, etc.
###### Example 6.2.
A topological vector space in which no base zero neighborhood contains a nontrivial linear subspace. Let X be the space of all analytic functions on $``$ equipped with the topology of uniform convergence on compact subsets of $``$. The sets
$$U_{n,\epsilon }=\{fX:|f(z)|<\epsilon \text{ whenever }|z|n\}(n0\text{ and }\epsilon >0)$$
form a zero neighborhood base of this topology. Clearly, no $`U_{n,\epsilon }`$ contains a non-trivial linear subspace. Indeed, if there is a function $`f`$ in $`X`$ and a zero neighborhood $`U_{n,\epsilon }`$ such that $`\lambda fU_{n,\epsilon }`$ for every scalar $`\lambda `$, then $`f(z)=0`$ whenever $`|z|<n`$, and it follows that $`f`$ is identically zero on $``$. Note that this topology is generated by the countable sequence of seminorms $`f_n=\underset{|z|n}{sup}\left|f(z)\right|`$; clearly $`_n`$ is the Minkowski functional of $`U_{n,1}`$.
###### Proposition 6.3.
If $`X`$ is a complete locally convex space then $`X`$ is locally bounded if and only if $`X`$ admits an nb-bounded bijection.
###### Proof.
If $`X`$ is locally bounded then the identity map is an nb-bounded bijection. Suppose that $`T`$ is an nb-bounded bijection on $`X`$. Then there exists a closed base zero neighborhood $`U`$ in $`X`$ such that $`T(U)`$ is bounded. Let $`A=\overline{T(U)}`$, then $`A`$ is convex, bounded, balanced, and absorbing. It follows that the space $`(X,A)`$ is a locally convex and locally bounded, denote it by $`X_A`$. Notice also that the topology of $`X_A`$ is finer than the original topology on $`X`$ because $`A`$ is bounded. In particular, $`X_A`$ is Hausdorff.
We claim that $`X_A`$ is complete. Indeed, if $`(x_n)`$ is a Cauchy sequence in $`X_A`$, then it is also Cauchy in the original topology of $`X`$, which is complete, so that $`x_n`$ converges to some $`x`$. Fix $`\epsilon >0`$, then there exists $`n_0`$ such that $`x_nx_m\epsilon A`$ whenever $`n,mn_0`$. Let $`m\mathrm{}`$, since $`A`$ is closed we have $`x_nx\epsilon A`$, i.e., $`x_nx`$ in $`X_A`$. Thus, $`X_A`$ is complete, hence Banach.
Since $`A`$ is bounded, we can find $`m`$ such that $`AmU`$. Then $`T(A)T(mU)mA`$, so that $`T`$ is bounded in $`X_A`$. Then $`T^1`$ is also bounded in $`X_A`$ by the Banach Theorem, so that $`U=T^1\left(T(U)\right)T^1(A)nA`$ for some $`n>0`$, hence $`U`$ is bounded. ∎
###### Proposition 6.4.
Let $`T:XY`$ be an nb-bounded operator between Hausdorff topological vector spaces such that $`X`$ is not locally bounded. If
1. every zero neighborhood in $`X`$ contains a non-trivial linear subspace, or
2. both $`X`$ and $`Y`$ are Fréchet spaces,
then $`T`$ is not a bijection.
###### Proof.
If every zero neighborhood of $`X`$ contains a non-trivial linear subspace, then $`T`$ cannot be one-to-one by 6.1. Suppose now that $`X`$ and $`Y`$ are Fréchet and assume that $`T`$ is a bijection. Let $`S:YX`$ be the linear inverse of $`T`$. The Open Mapping Theorem implies that $`S`$ is continuous and hence bb-bounded. It follows that the identity operator of $`X`$ is nb-bounded being the composition of the nb-bounded operator $`T`$ and the bb-bounded operator $`S`$. But the identity operator is nb-bounded if and only if the space is locally bounded, a contradiction. ∎
Weak topologies
We are going to show that every operator which is nb-bounded relative to a weak topology has to be of finite rank. In order to prove this we need the following well-known lemma. For completeness we provide a simple proof of it.
###### Lemma 6.5.
Let $`T`$ be a linear operator on a vector space $`L`$, and let $`f_1,\mathrm{},f_n`$ be linear functionals on $`L`$ such that $`Tx=0`$ whenever $`f_i(x)=0`$ for every $`i=1,\mathrm{},n`$. Then $`T`$ is a finite rank operator of rank at most $`n`$.
###### Proof.
Define a linear map $`\pi `$ from $`L`$ to $`^n`$ via $`\pi (x)=(f_1(x),\mathrm{},f_n(x))`$. Then the dimension of the range $`\pi (L)`$ is at most $`n`$. Define also a linear map $`\phi `$ from $`\pi (L)`$ to $`L`$ via $`\phi (\pi (x))=Tx`$. It can be easily verified that $`\phi `$ is well-defined. Then the range of $`T`$ coincides with the range $`\phi (\pi (L))`$, which is of dimension at most $`n`$. ∎
###### Proposition 6.6.
Let $`X`$ be a locally convex space, and $`T`$ an operator on $`X`$ such that $`T`$ is nb-bounded with respect to the weak topology of $`X`$. Then $`T`$ is of finite rank.
###### Proof.
Suppose $`T`$ maps some weak base zero neighborhood $`U=\{xX:|f_i(x)|<1,i=1,\mathrm{},n\}`$ ($`f_1,\mathrm{},f_nX^{}`$), to a weakly bounded set. Since the weak topology is Hausdorff, it follows from 6.1 that $`_{n=1}^{\mathrm{}}\frac{1}{n}U\mathrm{ker}T`$. In particular, $`Tx=0`$ whenever $`f_i(x)=0`$ for every $`i=1,\mathrm{},n`$. Then Lemma 6.5 implies that $`T`$ is a finite rank operator. ∎
Spectra and spectral radii of nb-bounded operators
###### Proposition 6.7.
If $`T`$ is an nb-bounded operator on a topological vector space then $`\sigma ^{bb}(T)=\sigma ^c(T)=\sigma ^{nn}(T)=\sigma ^{nb}(T)`$.
###### Proof.
If $`X`$ is locally bounded then the result is trivial by 3.3. Suppose that $`X`$ is not locally bounded, then, in view of 3.2, it suffices to show that $`\rho ^{bb}(T)\rho ^{nb}(T)`$. Let $`\lambda \rho ^{bb}(T)`$, then $`R_\lambda `$ is bb-bounded. If $`\lambda 0`$, then it follows from $`R_\lambda (\lambda IT)=I`$ that $`R_\lambda =\frac{1}{\lambda }R_\lambda T+\frac{1}{\lambda }I`$. Thus, $`R_\lambda `$ is a sum of an nb-bounded operator and a multiple of the identity operator, which yields $`\lambda \rho ^{nb}(T)`$. To finish the proof, it suffices to show that $`\lambda =0`$ necessarily belongs to $`\sigma ^{bb}(T)`$ (and, therefore, to $`\sigma ^c(T)`$, $`\sigma ^{nn}(T)`$, and $`\sigma ^{nb}(T)`$). Indeed, if the resolvent $`R_\lambda =T^1`$ were bb-bounded, then $`I=T^1T`$ would be nb-bounded, which is impossible in a non-locally bounded space, a contradiction. ∎
###### Proposition 6.8.
If $`T`$ is an nb-bounded operator on a topological vector space, then $`r_{bb}(T)=r_c(T)=r_{nn}(T)=r_{nb}(T)`$.
###### Proof.
By Proposition 4.3 it suffices to show that $`r_{bb}(T)r_{nb}(T)`$. Since $`T`$ is nb-bounded, then $`T(U)`$ is a bounded set for some zero neighborhood $`U`$. Let $`\nu >r_{bb}(T)`$ and fix a zero neighborhood $`V`$. Then $`\nu V`$ is again a zero neighborhood. In particular, since the sequence $`\frac{T^n}{\nu ^n}`$ converges to zero uniformly on bounded sets, we have $`\frac{T^n}{\nu ^n}\left(T(U)\right)\nu V`$ for all sufficiently large $`n`$. Then $`\frac{T^{n+1}}{\nu ^{n+1}}(U)V`$, so that $`\frac{T^n}{\nu ^n}`$ converges to zero uniformly on $`U`$. Therefore $`\nu r_{nb}(T)`$, so that $`r_{bb}(T)r_{nb}(T)`$. ∎
###### 6.9.
In view of Propositions 6.7 and 6.8 we can write $`\sigma (T)`$ instead of $`\sigma ^{bb}(T)`$, $`\sigma ^c(T)`$, $`\sigma ^{nn}(T)`$, and $`\sigma ^{nb}(T)`$ and $`r(T)`$ instead of $`r_{bb}(T)`$, $`r_c(T)`$, $`r_{nn}(T)`$, and $`r_{nb}(T)`$.
We have established in Theorems 5.15.5 that under some conditions the spectral radii of a linear operator are upper bounds for the geometrical radii of the corresponding spectra. Of course we would like to know when the equalities hold. It is well known that the equality $`\left|\sigma (T)\right|=r(T)`$ holds for every continuous operator on a Banach space. Moreover, it was shown in \[Gra66\] that this equality also holds for every continuous operator on a quasi-Banach space (a complete quasinormed space). Further, by means of Proposition 4.7 the main result of \[GBVM98\] is equivalent to the following statement: $`r(T)=\left|\sigma (T)\right|`$ for every nb-bounded operator $`T`$ on a complete locally convex space. Here we present a direct proof of this. Our proof is a simplified version of the proof of \[GBVM98\].
###### Theorem 6.10.
If $`T`$ is an nb-bounded linear operator on a sequentially complete locally convex space, then $`\left|\sigma (T)\right|=r(T)`$.
###### Proof.
Suppose $`T(U)`$ is bounded for some base zero neighborhood $`U`$. It follows from Propositions 6.7, 6.8, and 4.3, and Theorem 5.5 that it suffices to show that $`\left|\sigma ^{nn}(T)\right|r_{nb}(T)`$. We are going to show that $`T`$ induces a continuous operator $`\stackrel{~}{T}`$ on some Banach space such that $`\sigma \left(\stackrel{~}{T}\right)\sigma ^{nn}(T)\{0\}`$ while $`r\left(\stackrel{~}{T}\right)r_{nb}(T)`$, and then appeal to the fact that the spectral radius of a continuous operator on a Banach space equals the geometrical radius of the spectrum.
Consider $`T`$ as a continuous operator on the locally bounded space $`X_U=(X,U)`$. Then $`\sigma _U(T)`$ is defined by 3.3 and $`r_U(T)`$ is defined by 4.5. We claim that $`r_U(T)r_{nb}(T)`$. To see this, suppose $`r_U(T)<\nu `$, then $`\frac{T^n}{\nu ^n}(U)U`$ for all sufficiently large $`n`$. Let $`V`$ be a base zero neighborhood, then $`T(U)\alpha V`$ for some $`\alpha >0`$, so that $`\frac{T^n}{\nu ^n}(U)=\frac{T}{\nu }\frac{T^{n1}}{\nu ^{n1}}(U)\frac{1}{\nu }T(U)\frac{\alpha }{\nu }V`$ for sufficiently large $`n`$. This implies that $`\nu r_{nb}(T)`$, and it follows that $`r_U(T)r_{nb}(T)`$.
On the other hand, we claim that $`\sigma _U(T)\sigma ^{nn}(T)`$. Suppose $`\lambda \rho ^{nn}(T)`$, then $`R_\lambda `$ is nn-bounded with respect to some base $`𝒩_0`$ of zero neighborhood. We can assume without loss of generality that $`U𝒩_0`$, so that $`R_\lambda (U)\beta U`$ for some $`\beta >0`$. It follows that $`\lambda \rho _U(T)`$.
Since $`U`$ is convex, the the space $`X_U`$ is, in fact, a seminormed space. We can assume without loss of generality that it is a normed space, because otherwise we can consider the quotient space $`X_U/(\mathrm{Null}T)`$ and the quotient operator $`\widehat{T}`$ on this quotient space instead of $`T`$. Indeed, since $`_{n=1}^{\mathrm{}}\frac{1}{n}U\mathrm{Null}T`$ by 6.1, we conclude that the quotient space $`X_U/(\mathrm{Null}T)`$ is Hausdorff. It follows then that $`X_U/(\mathrm{Null}T)`$ is a normed space, and $`\widehat{T}`$ is norm bounded. The spectrum $`\sigma _U(T)`$ becomes even smaller when we substitute $`T`$ with $`\widehat{T}`$. Indeed, suppose $`\lambda \rho _U(T)`$, then the resolvent $`R_\lambda `$ exists in $`X_U`$ and is continuous. If $`x\mathrm{ker}T`$, then $`x=R_\lambda (\lambda IT)x=\lambda R_\lambda x`$, so that $`R_\lambda `$ leaves $`\mathrm{ker}T`$ invariant, and, therefore, induces a quotient operator $`\widehat{R_\lambda }`$ on $`X_U/\mathrm{ker}T`$ via $`\widehat{R_\lambda }([x])=[R_\lambda x]`$. Clearly, $`\widehat{R_\lambda }`$ is continuous: if $`[x_n][x]`$ in $`X_U/\mathrm{ker}T`$ then $`x_nz_nx`$ in $`X_U`$ for some $`(z_n)_{n=1}^{\mathrm{}}`$ in $`\mathrm{ker}T`$, so that $`[R_\lambda x_n]=[R_\lambda (x_nz_n)][R_\lambda x]`$. On the other hand, $`r_U(\widehat{T})r_U(T)`$, because if $`\nu >r_U(\widehat{T})`$ then $`\frac{\widehat{T}^n}{\nu ^n}([U])[U]`$ for all sufficiently large $`n`$, then $`\frac{T^n}{\nu ^n}(U)U+\mathrm{ker}T`$, so that $`\frac{T^{n+1}}{\nu ^{n+1}}(U)\frac{1}{\nu }T(U)\frac{\alpha }{\nu }U`$ for some $`\alpha >0`$. It follows that $`\nu r_U(T)`$ and, therefore, $`r_U(\widehat{T})r_U(T)`$.
Finally, we consider the completion $`\stackrel{~}{X}_U`$ of $`X_U`$, and extend $`T`$ to a continuous linear operator $`\stackrel{~}{T}`$ on the completion. The spectrum of $`\stackrel{~}{T}`$ is smaller that the spectrum of $`T`$, because if $`\lambda \rho _U(T)`$ then the resolvent $`R_\lambda `$ can be extended by continuity to $`\stackrel{~}{R_\lambda }`$ on $`\stackrel{~}{X}`$, and $`\stackrel{~}{R_\lambda }`$ is a continuous inverse to $`\lambda I\stackrel{~}{T}`$, so that $`\lambda \rho (\stackrel{~}{T})`$. On the other hand, $`r(\stackrel{~}{T})r_U(T)`$ because if $`\nu >r(\stackrel{~}{T})`$ then $`\frac{\stackrel{~}{T}^n}{\nu ^n}(\stackrel{~}{U})\stackrel{~}{U}`$ for all sufficiently large $`n`$, which implies $`\frac{T^n}{\nu ^n}(U)U`$ since $`T`$ is a restriction of $`\stackrel{~}{T}`$ on $`X`$. ∎
## 7. Compact operators
As with bounded operators, there is more than one way to define compact operators on an arbitrary topological vector space. A subset of a topological vector space is called precompact if its closure is compact. Given a linear operator $`T`$ on a topological vector space, $`T`$ is called Montel if it maps every bounded set into a precompact set and compact if it maps some neighborhood into a precompact set. To be consistent, we should have probably called these operators “b-compact” and “n-compact” respectively, but the names “Montel” and “compact” are commonly accepted. Obviously, every compact operator is Montel and nb-bounded (hence continuous); every Montel operator is bb-bounded.
###### 7.1.
If $`T`$ is compact or Montel, then sequential completeness is not needed in Theorems 5.15.5. Indeed, we used sequential completeness just once, namely, in the proof of Theorem 5.1 to justify the convergence of the sequence $`R_{\lambda ,n}x=\frac{1}{\lambda }_{i=0}^n\frac{T^ix}{\lambda ^i}`$. But since the sequence $`(R_{\lambda ,n}x)_n`$ is Cauchy and, therefore, bounded, the sequence $`(TR_{\lambda ,n}x)_n`$ has a convergent subsequence whenever $`T`$ is compact or Montel. Furthermore, it follows from $`R_{\lambda ,n+1}x=\frac{1}{\lambda }(I+TR_{\lambda ,n})x`$ that $`(R_{\lambda ,n}x)_n`$ has a convergent subsequence hence converges.
Let $`K`$ be a compact operator on an arbitrary topological vector space, and let $`\sigma (K)`$ and $`r(K)`$ be as in 6.9. It was proved in \[Pec91\] that $`\sigma (K)=\{0\}`$ implies $`r_l(K)=0`$. In the following theorem we use the technique of \[Pec91\] to improve this result by showing that in general $`r(K)\left|\sigma (K)\right|`$.
###### Theorem 7.2.
If $`K`$ is a compact operator on a Hausdorff topological vector space $`X`$, then $`r(K)\left|\sigma (K)\right|`$.
###### Proof.
Assume that $`\left|\sigma (K)\right|<r(K)`$. Without loss of generality (by scaling $`K`$) we can assume that $`\left|\sigma (K)\right|<1<r(K)`$. Since $`K`$ is compact, there is a closed base zero neighborhood $`U`$ such that $`\overline{K(U)}`$ is compact. In particular $`\overline{K(U)}`$ is bounded, so that $`\overline{K(U)}\eta U`$ for some $`\eta >0`$. We can assume without loss of generality that $`\eta >1`$. We define the following subsets of $`U`$:
$$U_1=\overline{K(U)}U,U_{n+1}=K(U_n)U(n=1,2,\mathrm{}),\text{and}U_0=\underset{n=1}{\overset{\mathrm{}}{}}U_n.$$
Notice, that $`U_1`$ is compact because $`\overline{K(U)}`$ is compact and $`U`$ is closed. Also, if $`U_n`$ is compact, then $`K(U_n)`$ is compact as the image of a compact set under a continuous operator. Therefore, every $`U_n`$ for $`n1`$ is compact. Using induction, we can show that the sequence $`(U_n)`$ is decreasing. Indeed, $`U_1U`$ by definition, $`U_2=K(U_1)UK(U)UU_1`$, and if $`U_nU_{n1}`$, then $`U_{n+1}=K(U_n)UK(U_{n1})U=U_n`$. It follows also that $`U_0`$ is compact and contains zero.
Notice that $`K`$ maps every balanced set to a balanced set. Since $`U`$ is balanced, $`U_n`$ is balanced for each $`n0`$. If $`A`$ is a balanced subset of $`U`$, then obviously $`A(\eta A)U`$, and when we apply the same reasoning to $`\frac{1}{\eta }K(A)`$ instead of $`A`$ (which is also a balanced subset of $`U`$), we get $`\frac{1}{\eta }K(A)K(A)U`$. We use this to show by induction that $`\frac{1}{\eta ^n}K^n(U)U_n`$ for every $`n1`$. Indeed, for $`n=1`$ we have $`\frac{1}{\eta }K(U)K(U)UU_1`$. Suppose $`\frac{1}{\eta ^n}K^n(U)U_n`$ for some $`n1`$, then
$$\frac{1}{\eta ^{n+1}}K^{n+1}(U)\frac{1}{\eta }K(U_n)K(U_n)U=U_{n+1},$$
which proves the induction step.
Next, we claim that there exists an open zero neighborhood $`V`$ and an increasing sequence of positive integers $`(n_j)`$ such that $`U_{n_j}V`$ is nonempty for every $`j1`$. Assume for the sake of contradiction that for every open zero neighborhood $`V`$ we have $`U_nV`$ for all sufficiently large $`n`$. Since $`\frac{1}{2}U`$ contains an open zero neighborhood, then there exists a positive integer $`N`$ such that $`U_n\frac{1}{2}U`$ whenever $`nN`$. This implies that $`U_{N+m}=K^m(U_N)`$ for all $`m0`$. Indeed, this holds trivially for $`m=0`$. Suppose that $`U_{N+m}=K^m(U_N)`$ for some $`m0`$. Then $`U_{N+m+1}=K(U_{N+m})U=K^{m+1}(U_N)U`$, and this implies that $`U_{N+m+1}=K^{m+1}(U_N)`$ because $`U_{N+m+1}\frac{1}{2}U`$. Now take any open zero neighborhood $`V`$, then $`\frac{1}{\eta ^N}V`$ is again a zero neighborhood, and by assumption there exists a positive integer $`M`$ such that $`U_n\frac{1}{\eta ^N}V`$ whenever $`nM`$. Let $`n\mathrm{max}\{M,N\}`$, then
$$V\eta ^NU_n=\eta ^NK^{nN}(U_N)\eta ^NK^{nN}\left(\frac{1}{\eta ^N}K^N(U)\right)=K^n(U),$$
which contradicts the hypothesis $`r_{nb}(K)=r(K)>1`$.
It follows from $`U_{n_j}V\mathrm{}`$ for every $`j1`$ that $`U_nV\mathrm{}`$ for all sufficiently large $`n`$ because $`U_n`$ is a decreasing sequence. Since $`U_nV`$ is a decreasing sequence of nonempty compact sets, then $`U_0V=_{n=1}^{\mathrm{}}(U_nV)\mathrm{}`$, so that $`U_0\{0\}`$.
For every $`n1`$ we have $`U_0U_n`$, it follows that $`K(U_0)K(U_n)`$ and, therefore, $`K(U_0)_{n=1}^{\mathrm{}}K(U_n)`$. Actually, the reverse inclusion also holds. To see this, let $`y_{n=1}^{\mathrm{}}K(U_n)`$. Then $`y=Kx_n`$, where $`x_nU_nU_1`$. Since $`U_1`$ is compact, the sequence $`(x_n)`$ has a cluster point, i.e., $`x_{n_j}x`$ for some subsequence $`(x_{n_j})`$ and some $`x`$. Since $`K`$ is continuous we have $`y=Kx`$. On the other hand, since every $`U_{n_j}`$ is closed we have $`xU_{n_j}`$, so that $`x_{n=1}^{\mathrm{}}U_{n_j}=U_0`$. Thus $`K(U_0)=_{n=1}^{\mathrm{}}K(U_n)`$.
Next, we claim that $`U_0K(U_0)\eta U_0`$. Indeed,
$$U_0=\underset{n=2}{\overset{\mathrm{}}{}}U_n=\underset{n=2}{\overset{\mathrm{}}{}}\left[K(U_{n1})U\right]\underset{n=2}{\overset{\mathrm{}}{}}K(U_{n1})=K(U_0).$$
On the other hand, since $`U_n`$ are decreasing and $`\eta >1`$, we have $`K(U_n)K(U_{n1})\eta K(U_{n1})`$ and $`K(U_n)K(U)\eta U`$, so that $`K(U_n)\eta K(U_{n1})\eta U=\eta U_n`$, and this implies $`K(U_0)K(U_n)\eta U_n`$ for every $`n`$. Thus $`K(U_0)\eta U_0`$.
Since $`\overline{K(U)}`$ is compact, hence bounded, then $`\overline{K(U)}+\overline{K(U)}`$ is also bounded. Then there is a positive constant $`\gamma `$ such that $`\overline{K(U)}+\overline{K(U)}\gamma U`$. Without loss of generality we can assume $`\gamma 2`$. It follows that
$$U_1+U_1=\overline{K(U)}U+\overline{K(U)}U\overline{K(U)}+\overline{K(U)}\gamma U.$$
We use induction to show that $`U_n+U_n\gamma U_{n1}`$. Indeed, since $`AB+CD(A+C)(B+D)`$ for any four sets $`A`$, $`B`$, $`C`$, and $`D`$, then
$$\begin{array}{c}U_{n+1}+U_{n+1}=K(U_n)U_n+K(U_n)U_n\hfill \\ \hfill \left[K(U_n)+K(U_n)\right](U_n+U_n)K(U_n+U_n)(U_n+U_n)\\ \hfill K(\gamma U_{n1})\gamma U_{n1}=\gamma \left[K(U_{n1})U_{n1}\right]=\gamma U_n.\end{array}$$
Finally, $`U_0+U_0_{n=1}^{\mathrm{}}(U_n+U_n)_{n=1}^{\mathrm{}}\gamma U_n=\gamma U_0`$.
Next, consider the set $`F=_{n=1}^{\mathrm{}}nU_0`$. This set is closed under multiplication by a scalar, and $`U_0+U_0\gamma U_0`$ implies that $`F`$ is a linear subspace of $`X`$. We consider the locally bounded topological vector space $`(F,U_0)`$ with multiples of $`U_0`$ as the base of zero neighborhoods. Since $`U_0`$ is balanced by definition, this topology is linear, and it is Hausdorff because $`U_0`$ is compact. Also, it is finer than the topology on $`F`$ inherited from $`X`$ because $`U_0`$ is compact and, therefore, bounded in $`X`$.
We claim that $`(F,U_0)`$ is complete. Indeed, if $`(x_n)`$ is a Cauchy sequence in $`(F,U_0)`$ then there exists $`k>0`$ such that $`x_nkU_0`$ for each $`n>0`$. Since $`U_0`$ is compact, the sequence $`(x_n)`$ has a subsequence which converges to some $`xkU_0`$ in the topology of $`X`$. Moreover, $`\underset{n\mathrm{}}{lim}x_n=x`$ because the sequence $`(x_n)`$ is Cauchy in $`X`$. Fix $`\epsilon >0`$, then there exists $`n_0`$ such that $`x_nx_m\epsilon U_0`$ whenever $`n,mn_0`$. Let $`m\mathrm{}`$, since $`U_0`$ is is closed we have $`x_nx\epsilon U_0`$, i.e., $`x_nx`$ in $`(F,U_0)`$. Thus, $`(F,U_0)`$ is complete and, therefore, quasi-Banach.
It follows from $`U_0K(U_0)\eta U_0`$ that $`F`$ is invariant under $`K`$ and the restriction $`\stackrel{~}{K}=K|_F`$ is continuous. We claim that $`\sigma (\stackrel{~}{K})\sigma (K)\{0\}`$. Suppose that $`\lambda \rho (K)`$ and $`\lambda 0`$, then $`(\lambda IK)`$ is a homeomorphism, so that $`(\lambda IK)(U)`$ is a closed zero neighborhood, and $`\alpha U_1(\lambda IK)(U)`$ for some positive real $`\alpha `$ because $`U_1`$ is bounded. Further, $`\alpha K(U_1)K(\lambda IK)(U)(\lambda IK)K(U)`$. Therefore
$$\alpha U_2\alpha K(U_1)\alpha U_1(\lambda IK)K(U)(\lambda IK)(U),$$
and since $`\lambda IK`$ is one-to-one we get $`\alpha U_2(\lambda IK)(K(U)U)(\lambda IK)(U_1)`$. Similarly, we obtain $`\alpha U_{n+1}(\lambda IK)(U_n)`$ for all $`n1`$, and then $`\alpha U_0(\lambda IK)(U_0)`$. This implies that the restriction of $`\lambda IK`$ to $`F`$ is onto, invertible, and the inverse is continuous. Thus, $`\lambda \rho (\stackrel{~}{K})`$.
In particular this implies that $`\left|\sigma (\stackrel{~}{K})\right|\left|\sigma (K)\right|<1`$. On the other hand, it follows from $`U_0K(U_0)`$ that $`U_0\stackrel{~}{K}^n(U_0)`$ for all $`n0`$, so that $`\stackrel{~}{K}^n`$ does not converge to zero uniformly on $`U_0`$, whence $`r(\stackrel{~}{K})=r_{bb}(\stackrel{~}{K})1`$. This produces a contradiction because it was proved in \[Gra66\] that the spectral radius of a continuous operator on a quasi-Banach space equals the radius of the spectrum. ∎
###### Corollary 7.3.
If $`K`$ is a compact operator on a locally convex (or pseudo-convex) space, then $`r(K)=\left|\sigma (K)\right|`$.
## 8. Closed operators
In certain situation one has to deal with unbounded linear operators in Banach spaces. For example, the generator of a strongly continuous operator semigroup is generally a closed operator with dense domain (see e.g. \[HP57, DS58\]). Through this section $`T`$ will be a closed operator on a Banach space $`X`$ with domain $`𝒟(T)`$. As usually, we define $`𝒟(T^{n+1})=\{x𝒟(T^n):T^nx𝒟(T)\}`$ and $`D=_{n=0}^{\mathrm{}}𝒟(T^n)`$. In case when $`T`$ is the infinitesimal generator of an operator semigroup, $`D`$ is dense in the range of the semigroup, which is usually assumed to be all of $`X`$. The set $`D`$ with the locally-convex topology $`\tau `$ given by the sequence of norms $`x_n=_{k=0}^nT^kx`$ is a Fréchet space. Clearly, $`D`$ is invariant under $`T`$, and the restriction operator $`T_{|D}`$ is continuous because $`x_\alpha \stackrel{𝜏}{}0`$ in $`D`$ implies $`Tx_\alpha _nx_\alpha _{n+1}0`$ for each $`n`$.
We investigate the relation between the spectral properties of the original operator $`T`$ on $`X`$ and of the restriction $`T_{|D}`$ on $`D`$. A different approach to this question can be found in \[Wro99\]. Recall that $`\lambda \rho (T)`$ if $`R(\lambda ;T)=(\lambda IT)^1:X𝒟(T)`$ exists (it is automatically bounded by \[HP57, Theorem 2.16.3\]), and $`\sigma (T)=\rho (T)`$.
###### Lemma 8.1.
If $`\lambda \rho (T)`$ then $`R(\lambda ;T)`$ is a bijection of $`D`$ commuting with $`T`$ on $`𝒟(T)`$. Further, for each $`n0`$ there is a constant $`C_n`$ such that $`R(\lambda ;T)x_nC_nx_{n1}`$ for each $`xD`$.
###### Proof.
It can be easily verified that $`R(\lambda ;T)`$ is a bijection from $`𝒟(T^n)`$ onto $`𝒟(T^{n+1})`$ and, therefore, the restriction $`R(\lambda ;T)_{|D}`$ is a bijection. Notice that since $`R(\lambda ;T)Tx=x`$ for each $`x𝒟(T)`$ and $`TR(\lambda ;T)x=x`$ for each $`xX`$, then
$`TR(\lambda ;T)x`$ $`=`$ $`\lambda R(\lambda ;T)xx\text{ for each }x𝒟(T)\text{ and}`$
$`R(\lambda ;T)Tx`$ $`=`$ $`\lambda R(\lambda ;T)xx\text{ for each }xX,`$
so that $`T`$ and $`R(\lambda ;T)`$ commute on $`𝒟(T)`$. It also follows that for each $`xD`$ we have
$`T^2R(\lambda ;T)x`$ $`=`$ $`\lambda TR(\lambda ;T)xTx=\lambda ^2R(\lambda ;T)\lambda xTx,`$
$`T^3R(\lambda ;T)x`$ $`=`$ $`\lambda ^2TR(\lambda ;T)\lambda TxT^2x\lambda ^3R(\lambda ;T)\lambda ^2x\lambda TxT^2x,`$
$`\mathrm{}`$
$`T^kR(\lambda ;T)x`$ $`=`$ $`\lambda ^kR(\lambda ;T)x\lambda ^{k1}x\lambda ^{k2}Tx\mathrm{}\lambda T^{k2}xT^{k1}x.`$
It follows that
$$T^kR(\lambda ;T)x|\lambda |^kR(\lambda ;T)x+|\lambda |^{k1}x+|\lambda |^{k2}Tx+\mathrm{}+T^{k1}x$$
for each $`xD`$, so that
$$\begin{array}{c}R(\lambda ;T)x_n=\underset{k=0}{\overset{n}{}}T^kR(\lambda ;T)x\hfill \\ \hfill \mu _nR(\lambda ;T)x+\mu _{n1}x+\mu _{n2}Tx+\mathrm{}+\mu _0T^{n1}x,\end{array}$$
where $`\mu _k=1+|\lambda |+\mathrm{}+|\lambda |^k`$. Since $`R(\lambda ;T)xR(\lambda ;T)x`$ it follows that $`R(\lambda ;T)x_nC_n\left(x+Tx+\mathrm{}+T^{n1}x\right)`$ for some $`C_n`$, so that $`R(\lambda ;T)x_nC_nx_{n1}`$. ∎
###### Proposition 8.2.
The inclusion $`\rho (T)\rho ^{nn}(T_{|D})`$ holds. Moreover, if $`D`$ is dense in $`X`$ and $`T`$ is the smallest closed extension of $`T_{|D}`$, then $`\rho (T)=\rho ^{nn}(T_{|D})`$.
###### Proof.
Suppose that $`\lambda \rho (T)`$ and consider the resolvent operator $`R(\lambda ;T)`$ on $`X`$. Then $`R(\lambda ;T)x_nC_nx_{n1}C_nx_n`$ hence $`R(\lambda ;T)_{|D}`$ is nn-bounded and $`\lambda \rho ^{nn}(T_{|D})`$.
Suppose now that $`D`$ is dense in $`X`$, $`T`$ is the smallest closed extension of $`T_{|D}`$, and $`\lambda \rho ^{nn}(T_{|D})`$. Then there exists an nn-bounded operator $`R(\lambda ;T_{|D}):DD`$ such that $`R(\lambda ;T_{|D})(\lambda IT)x=(\lambda IT)R(\lambda ;T_{|D})x=x`$ for each $`xD`$. Then there is a constant $`C>0`$ such that $`R(\lambda ;T_{|D})x=R(\lambda ;T_{|D})x_0Cx_0=Cx`$ for each $`xD`$. It follows that $`R(\lambda ;T_{|D})`$ can be extended to a bounded operator $`R`$ on $`X`$. Fix $`xX`$ and pick $`(x_n)`$ in $`D`$ such that $`x_nx`$. Then $`R(\lambda ;T_{|D})x_nRx`$ and $`(\lambda IT)R(\lambda ;T_{|D})x_n=x_nx`$. Since $`\lambda IT`$ is closed we have $`(\lambda IT)Rx=x`$. It follows, in particular, that $`\lambda IT`$ is onto.
Since $`(\lambda IT)x\frac{1}{C}x`$ for each $`xD`$, it follows that for every nonzero $`yX`$ the pair $`(y,0)`$ doesn’t belong to the closure of the graph of $`\lambda IT_{|D}`$. But the closure of the graph of $`\lambda IT_{|D}`$ is the graph of $`\lambda IT`$ because $`\lambda IT`$ is the smallest closed extension of $`\lambda IT_{|D}`$. It follows that $`\lambda IT`$ is one-to-one, hence $`\lambda \rho (T)`$. ∎
Suppose now that $`S`$ is a bounded operator on $`X`$ such that $`D`$ is invariant under $`S`$ and $`STx=TSx`$ for each $`xD`$. Then
$$Sx_m=\underset{k=0}{\overset{m}{}}T^kSxS\underset{k=0}{\overset{m}{}}T^kx=Sx_m,$$
so that $`S_{|D}_nS`$. Moreover, if $`mk`$ then $`x_mx_k`$, so that the mixed seminorm $`𝔪_{km}(S_{|D})S`$. It also follows from Proposition 4.6 that $`r_{nn}(S_{|D})r(S)`$.
Further, we claim that if $`R=R(\lambda ;T)`$ for some $`\lambda \rho (T)`$, then $`r_{nb}(R_{|D})r(R)`$. Indeed, recursive application of Lemma 8.1 yields $`R^nx_kM_kR^{nk}x`$ for each $`xD`$ and $`kn`$, where $`M_k=\mathrm{\Pi }_{i=1}^kC_i`$. It follows that the mixed seminorm
$$\begin{array}{c}𝔪_{mk}(R_{|D}^n)=sup\{R^nx_k:xD,x_m1\}\hfill \\ \hfill sup\{M_kR^{nk}x:x1\}=M_kR^{nk}.\end{array}$$
Therefore $`lim_n\sqrt[n]{𝔪_{mk}(R_{|D}^n)}lim_n\sqrt[n]{R^n}=r(T)`$ for any $`m,k0`$. Now Proposition 4.6 yields $`r_{nb}(R_{|D})r(R)`$.
|
warning/0004/astro-ph0004206.html
|
ar5iv
|
text
|
# The Redshift Distribution of FIRST Radio Sources at 1 mJy
## 1 INTRODUCTION
An accurate definition of the redshift distribution of radio sources at faint flux densities has become particularly important in the last decade for both radio astronomy and cosmology. It is critical, for example, in testing radio-source unification (e.g. Jackson & Wall, 1998), and in large-scale structure studies (e.g. Loan et al. 1997, Magliocchetti et al. 1999) to permit the conversion of angular clustering estimates to the spatial clustering estimates required to evaluate structure formation models.
Classical large-scale-structure studies have used wide-area optical and IR surveys to measure the clustering of galaxies, but these are limited to low redshifts, with a peak redshift selection of $`0.1`$ (e.g. APM, Maddox et al., 1990; IRAS, Fisher et al., 1993). Deep small-area surveys such as the Hubble Deep Field North (Williams et al., 1996) or the CFRS survey (Lilly et al., 1995) have been used to measure clustering to much higher redshifts (see e.g. Le Fevre et al., 1995; Magliocchetti & Maddox, 1998), but they probe small volumes and can measure clustering only on small scales ($`1`$ Mpc). On the other hand radio objects are detected to high redshifts ($`z4`$) and sample a much larger volume for a given number of galaxies, so that they have the potential to provide information on the growth of structure on large physical scales. Recently, clustering statistics in deep radio surveys have been measured (Cress et al., 1997; Loan et al., 1997; Baleisis et al., 1997; Magliocchetti et al., 1998) and these analyses have shown radio sources to be reliable tracers of the mass distribution, even though they may be biased (Magliochetti et al., 1999).
However the relationship between angular measurements and the physically meaningful spatial quantities is highly uncertain without an accurate estimate of the redshift distribution of radio objects $`N(z)`$ (Magliocchetti et al., 1999). The current estimates of $`N(z)`$ for radio sources at mJy levels are largely based on predictions from the local radio luminosity functions and evolution (see e.g. Dunlop & Peacock, 1990), and are poorly defined.
Indeed the predictions require the knowledge of more than one luminosity function. Deep radio surveys (Condon & Mitchell, 1984; Windhorst et al., 1985; Fomalont et al., 1993) have shown a flattening of the differential source count $`\frac{dN}{dS}`$ below S $`10`$ mJy. This flattening is generally interpreted as due to the presence of a population of radio sources differing from the radio AGN which dominate at higher flux densities. Condon (1984) suggested a population of strongly-evolving normal spiral galaxies, while others (Windhorst et al., 1985; Danese et al., 1987) claimed the presence of an actively star-forming galaxy population. Observations supported this latter suggestion; the identifications of many of the faint sources are with galaxies with spectroscopic and photometric properties similar to ‘IRAS galaxies’ (Franceschini et al., 1988; Benn et al., 1993). The model predictions of $`N(z)`$ at faint flux densities from the luminosity functions of Dunlop & Peacock (1990) diverge greatly because of inadequate definition of the radio AGN luminosity functions; but in addition the contribution to $`N(z)`$ of the starburst population needs consideration.
In order to define $`N(z)`$ and the population mix at mJy levels, we have observed radio sources from the FIRST survey (Becker et al., 1995) using the WYFFOS multi-object spectrograph on the William Herschel Telescope in 8 1-degree diameter fields. We show how these observations constrain both the form of $`N(z)`$ and the population mix at S $`1`$ mJy for $`z0.3`$. Section 2 of the paper introduces this radio sample, while Sections 3 and 4 describe acquisition and reduction of the data. In Sections 5, 6 and 7 we present the radio, spectroscopic and photometric properties of the sample. Section 8 is devoted to the analysis of the observed $`N(z)`$ and comparison with models; the conclusions are in Section 9.
## 2 THE RADIO DATA
The FIRST (Faint Images of the Radio Sky at Twenty centimetres) survey (Becker, White and Helfand, 1995) began in the spring of 1993 and will eventually cover some 10,000 square degrees of the sky in the north Galactic cap and equatorial zones. The VLA is being used in B-configuration to take 3-min snapshots of 23.5-arcmin fields arranged on a hexagonal grid. The beam-size is 5.4 arcsec at 1.4 GHz, with an rms sensitivity of typically 0.14 mJy/beam. A map is produced for each field and sources are detected using an elliptical Gaussian fitting procedure (White et al., 1997); the $`5\sigma `$ source detection limit is $``$1 mJy. The surface density of objects in the catalogue is $`90`$ per square degree, though this is reduced to $`80`$ per square degree if we combine multi-component sources (Magliocchetti et al., 1998). The depth, uniformity and angular extent of the survey are excellent attributes for investigating the clustering properties of faint sources.
We used the 4 April 1998 version of the catalogue which contains approximately 437,000 sources and is derived from the 1993 through 1997 observations covering nearly 5000 square degrees of sky, including most of the area $`7^h20^m<\mathrm{RA}(2000)<17^h20^m`$, $`22.2^{}<\mathrm{Dec}<57.5^{}`$ and $`21^h20^m<\mathrm{RA}(2000)<3^h20^m`$, $`2.5^{}<\mathrm{Dec}<1.6^{}`$. This catalogue has been estimated to be 95 per cent complete at 2 mJy and 80 per cent complete at 1 mJy (Becker et al., 1995).
## 3 OBSERVATIONS
### 3.1 Field Selection
For the purpose of determining the redshift distribution, the precise position of the fields is not crucial, although several factors were taken into account when choosing which to observe. The fields must span a range of RA’s to ensure they could be observed at low zenith distance throughout the allocated nights. The fields are at as high galactic latitude as possible to reduce the probability of confused optical identifications and to ensure low galactic extinction (the allocated time meant the fields had to be on opposite sides of the galactic plane). We also cross-checked with fields which will be observed as part of the INT Wide-Field Camera survey, and chose our fields to overlap with these, so that deep optical images will be available in the near future.
These constraints lead to the choice of three main areas: $`16^h11^m,+54^{}22^{}`$ (which covers the ISO ELAIS field); $`22^h40^m,+0^{}0^{}`$, and $`02^h30^m,+0^{}0^{}`$. We also observed a field adjacent to the ELAIS field, and one on each side of the $`22^h40^m,+0^{}0^{}`$, and $`02^h30,+0^{}0^{}`$ fields. The areas actually observed are shown in Figure 1, and the field centres are listed in Table 1.
### 3.2 WYFFOS and AUTOFIB2
The 4.2-m William Herschel Telescope (WHT) has a prime-focus corrector with an atmospheric-dispersion compensator that provides a field of view of one degree in diameter. AUTOFIB2 is a robotic positioner, which places up to 120 optical fibre feeds in the focal plane, each fibre collecting light from a $`2.7^{\prime \prime }`$ diameter patch of sky. The fibres are fed to WYFFOS which is a multi-object spectrograph, at one of the WHT Nasmyth foci. The spectrograph can use a range of gratings offering 30 - 500 Å/mm dispersion. We chose to use the 600 lines/mm grating with the Tek1024<sup>2</sup> CCD, which gives a dispersion of $`3`$ Å per pixel, and a resolution of $`10`$Å. For a central wavelength at 6000Å this set up gives spectra from $`4400`$Å to $`7400`$Å.
### 3.3 Astrometry and Fibre Positioning
The fibres on WYFOSS cover 2.7 arcsecs on the sky, but to obtain the best throughput, they must be positioned to an accuracy of $`0.5^{\prime \prime }`$ or better. The astrometric reference frame of the FIRST survey is accurate to $`0.05^{\prime \prime }`$, and individual sources have 90% confidence error circles of radius $`<0.5^{\prime \prime }`$ at the 3 mJy level and $`1^{\prime \prime }`$ at the survey threshold. This means that, in principle, the radio source positions are accurate enough to position the fibres, even without optically identified counterparts. Using the radio positions means that our sample is not biased towards optically bright sources. Although our spectroscopy will clearly be limited by the optical brightness, it should be possible to obtain redshifts for any sources that have strong optical emission lines but very faint continuum. These sources would have been rejected from the sample if we required every target to have an optical counterpart. However there are a few complications in using the radio positions and these must be taken into account.
First, the telescope must be positioned and guided using fiducial stars. We selected stars with $`13<b_J<15`$ from APM scans of UKST plates for the equatorial fields, and of POSS2 plates for the ELAIS fields. These star positions are accurate to about $`0.1^{\prime \prime }`$, but have been measured in an optical reference frame, which may be offset from the radio reference frame by more than an arcsecond. Therefore we matched up the radio sources and optical images for each Schmidt plate.
Figure 2 shows a typical plot of the positional residuals between the optical and radio positions. There is a uniform background of points with residuals greater than $`1^{\prime \prime }2^{\prime \prime }`$, which come from random coincidences between optical and radio sources. The true optical identifications show up as the concentration of points near zero offset. The mean positional offset of the identifications is $`0.9^{\prime \prime }`$ in RA and $`0.5^{\prime \prime }`$ in Dec, corresponding to the offset between the optical and radio astrometric frames, and this offset was applied to each position in the ELAIS field so that the optical and radio astrometry coincide. Similar offsets were calculated for each of the other observed fields. The scatter in residuals about the mean offset is about $`0.5^{\prime \prime }`$ which is consistent with the quoted positional accuracy of the surveys.
The second complication is that the optical ID of an extended radiosource will not necessarily lie at the radio catalogue position. In the case of a classical double-lobed radiosource, the radio core and optical ID will lie between the lobes, which themselves may be catalogued as two separate radio sources. In the case of a core-jet radiosource, the radio core and optical ID will usually lie away from the centroid of the radio emission. To address this problem, we extracted $`4^{}\times 4^{}`$ radio maps from the FIRST database and optical images covering a similar area from the Digital Sky Survey (DSS). We visually inspected each of the radio maps, and overlaid the optical contours so that we could identify possible offsets between the radio and optical emission. Almost all of the radiosources were either unresolved or dominated by an unresolved component in the FIRST maps. Only in a few cases it was necessary to manually adjust the positions. A typical plot showing the FIRST radio map and overlaid optical contours is shown in Figure 3.
As well as the fiducial stars and target objects, it is necessary to observe some regions of blank sky, so that the sky spectrum during the observations can be estimated and subtracted from the target spectra. On average there are 70 radio sources per field (see Table 1), leaving 50 fibres which could be used for sky. Two sets of sky positions were generated for each field. The first set were the target positions in the focal plane when the telescope is offset by 5 arcminutes in Dec. Note that the spherical geometry of the sky means that new positions do not exactly correspond to a simple addition of 5 arcmins to each Dec. The low surface density of bright optical images meant that these positions were blank sky for all but a couple of cases. The second set were simply randomly chosen positions; again most of these are blank sky. Given the large number of sky fibres we could median them together to obtain a highly accurate estimate of the true sky spectrum, even if a few were contaminated by objects.
### 3.4 Field Configurations
Each fibre for AUTOFIB2 can be placed within a restricted area defined by a maximum radial extension and a maximum tilt angle from the radial direction. Also no fibre can cross another, and there must be a small buffer-zone around each fibre. These constraints limit the number of targets that can be allocated to a fibre within a particular field. The program CONFIGURE is part of the WYFFOS software package, and it attempts to find the most efficient set of target-to-fibre allocations. This configuration was then inspected and some changes were made manually to increase the number of fibres allocated to targets. A typical configuration is shown in Figure 4. The total number of targets and number of fibres allocated is shown in Table 1.
### 3.5 Observing Procedure
Once a field has been configured and acquired we took a standard sequence of calibration frames and observations of the target field. The sequence began with an arc lamp for wavelength calibration; then 1800 sec on the targets; then three 600 sec offset sky integrations; another 1800 sec on the targets; another three 600 sec skies; and finally 1800 sec on targets. Each of the offset skies was given a different offset to ensure that the sky spectra were not contaminated by random objects that happen to fall on the fibres.
Most of the targets are very faint, so accurate sky subtraction is crucial. The main aim of the observing sequence is to provide an accurate estimate of the sky spectrum during the observations, and also to calibrate the relative throughputs of each fibre for this particular configuration. Since the field is vignetted, the apparent transmission of a fibre will depend on its position in the field. Also the efficiency of light entering a fibre will depend on the precise angle that the fibre button settled when placed by AUTOFIB2, so there may be small variations even if the fibres are repositioned to the same configuration. This also means that flat-fields from internal lamps will not exactly reproduce the throughputs as seen on the sky. The offset skies are observed with exactly the same set-up as the target objects, and so provide very accurate relative throughput estimates for the object and sky fibres. The large number of sky fibres provides an accurate estimate of the actual sky spectrum during the object integrations.
In principle the $`5^{}`$ offset skies allow a more efficient observing mode that does not require separate offset sky integrations. If the telescope is nodded back and forth between the target positions, and the offset fibres, light from the targets is always being collected. Also the sky spectrum can be measured through the actual fibre used for each target. The disadvantage of this approach is that a sky fibre must be allocated to each object fibre, and this leads to quite severe configuration constraints. Typically less than 50% of the available targets could be allocated a pair of fibres, and so this turned out to be a rather inefficient option for our project. Nevertheless we did successfully apply the method for one field (F0234).
During daylight we took high signal-to-noise tungsten lamp flat-fields which were used to define the position of each fibre spectrum as described below.
## 4 Data Reduction
The data obtained were reduced using a package within IRAF written specifically to reduce WYFFOS data. The first steps in the data reduction are the standard bias level subtraction and trimming of the overscan regions. Then the tungsten flat-fields are used to locate and fit polynomials to the spectrum of each fibre. These polynomials allow 1-D spectra to be extracted from the 2-D images. The package uses information in the file headers to ensure the correct link between fibre number and spectrum number, so that each spectrum is identified with the correct input target.
Next, for each target field, the 2-D images from the individual integrations were averaged together, using the IRAF ccdreject option. Similarly the offset skies for each field were combined. This way of combining the images provides good signal-to-noise for pixels where all component images are good data, and effectively rejects pixels contaminated by cosmic rays or random objects that happened to fall in a sky fibre for one particular offset.
As a simple but effective method of sky subtraction, the 2-D averaged offset skies were rescaled to match the exposure time in the target frame and then simply subtracted from the 2-D target data. This technique was used because WYFFOS suffers from light scattered within the spectrograph. The broad scattered light features seen in Figure 5(a) have $`10\%`$ of the counts in the actual spectra, and so it is very important to subtract them accurately. The standard approach attempts to fit a smooth 2-D polynomial to the light in between the fibres, and subtracts this from the 2-D data frame. Then 1-D spectra are extracted from the 2-D data, and the sky-fibres are used to define the mean sky spectrum, which is rescaled according to a throughput map generated from offset skies, and subtracted from each object spectrum. We found that it was very difficult to fit the scattered light accurately enough to leave a clean 2-D image with reliable spectra. Since our target galaxies are very faint, the scattered light is almost entirely from the sky, even in the target fields. Therefore simply re-scaling the 2-D offset sky frames and subtracting from the 2-D target frame means the scattered light and the sky are both properly subtracted. This is very effective, as seen in Figure 5(b).
This sky-subtraction approach is susceptible to errors if the sky is rapidly varying, because the actual sky spectrum during the target integration is not used. Any such errors can be simply corrected by using the spectra from the sky fibres in sky-subtracted 2-D images to estimate the average residual sky which can then subtracted from the target spectra. As can be seen in the spectra, the residuals from sky-subtraction are very small, confirming that the approach works very well.
The wavelength calibration for each data-set was carried out using the standard IRAF tasks. This automatically accounts for the ’saw-tooth’ arrangement of fibres on the spectrograph slit.
## 5 Spectral Analysis and Classification
### 5.1 REDSHIFT DETERMINATION
Redshifts were determined in 46 objects, corresponding to $`13`$ per cent of the spectroscopic sample. Determinations were obtained in two separate ways: first by visual inspection of the spectra and attempting to identify individual features; second via cross-correlation with a range of templates.
The cross-correlation analysis uses the same algorithm as the 2dF redshift survey code (Seaborne, 2000) but has small changes to the filtering parameters, the wavelength coverage and allowed redshift range appropriate for ou WYFOSS data. The cross-correlation functions between the data and a set of template spectra are calculated and the highest peak is located. An index $`0q4`$ is automatically assigned according to the height of the peak and the noise and assesses the quality of the redshift measurement: 0 corresponding to no redshift estimate, and 4 a reliable redshift. Each redshift estimate was also checked by eye at this stage, and the quality flag was updated if necessary. We accepted estimates with $`q`$ as low as 3, so long as they agreed with the results from visual inspection. A few objects (all absorption systems) with $`q=2`$ were included in the final sample, when the cross-correlation estimate was in agreement with the results from visual inspection. Table 2 lists the measured redshifts for the observed targets.
### 5.2 SPECTRAL CLASSIFICATION
The optical counterparts of radio sources with a redshift determination were classified into 6 main groups according to their spectral features:
1. Early-type galaxies where spectra were dominated by continua much stronger than the intensity of any emission lines present. These objects can then be divided into two further categories:
* Galaxies with absorption lines only.
* Galaxies with absorption lines + weak but non-negligible OII. emission lines
2. Late-type galaxies showing strong emission lines characteristic of star-forming activity, together with a detectable continuum.
3. Starburst galaxies where the continuum is missing and the spectra only showed strong emission lines due to star-formation activity.
4. Seyfert1 galaxies showing strong broad emission lines. These objects could be quasars.
5. Seyfert2 galaxies showing narrow emission lines due to the presence of an active galactic nucleus. To have an objective classification of the narrow emission line objects and in particular to distinguish Seyfert2’s from galaxies heated by OB stars (classes 2-3), we used the diagnostic emission line ratios of Baldwin, Phillips & Terlevich, 1981 and Rola, Terlevich & Terlevich, 1997.
6. A final class consisted of those objects whose spectra show a detectable continuum but the complete absence of features; no redshift determinations for these were possible. These spectra are generally believed to be associated with early-type galaxies (see Gruppioni, Mignoli & Zamorani, 1998), given the lack of emission lines.
Also, four spectra appear to be stars at zero redshift; these are most likely to be random positional coincidences with galactic stars.
Table 2 lists the spectral classes. All spectra with significant signal are shown in Figure 6. The spectrum at the bottom of each panel show the sky spectrum, with strong emission lines marked. The galaxy spectra have not been flux calibrated, and the line on the lower right is an approximate atmospheric transmission curve, showing the position and shape of the atmospheric A and B absorption bands.
## 6 Photometry
### 6.1 The APM Survey
Some of the fields (F0226, F0230, F0234, F2236, F2240, F2244) targeted in our WHT observations lie in regions of the sky sampled by the APM scan of UKST plates (http://www.ast.cam.ac.uk/apmcat/). The APM data provided R and B magnitudes of the objects in the sample brighter than B $`22`$, R $`21`$, the limiting magnitudes of the APM survey.
To obtain these magnitude measurements we searched the UKST catalogue for optical counterparts lying within 2.7 arcseconds from the corresponding radio position. Note that the fibre position always coincides with the radio coordinates except in the case of double sources (see Section 3.3). The tolerance in positional matching allows for measurement errors and the fact that the centre of radio emission might be displaced from the centre of the optical emission, especially if the radio object is extended. If no optical object within this offset was found in the UKST catalogue we assumed the object to be fainter than the limiting magnitude.
A few objects in the F0230 and F2240 fields show a displacement greater than the value of 2.7<sup>′′</sup> (but within 3<sup>′′</sup> \- Table 2). These are the sources observed at the MDM Observatory (see Section 6.2).
Figure 10 plots the apparent magnitude R vs. the radio flux S$`_{1.4\mathrm{GHz}}`$ for all those sources with an optical counterpart in the APM catalogue. Filled dots represent objects with a redshift determination, while circles are for those objects with no $`z`$. The plot indicates that the percentage of successful redshift determinations is $`100\%`$ to R $`18.6`$ and then decreases with increasing apparent magnitude at a rate which varies from field to field. Figure 10 also shows that almost all the objects in our spectroscopic sample with a redshift determination are brighter than the APM limiting magnitude; only 3 sources with a measured redshift are not detected in the APM data.
### 6.2 Further observations
CCD observations at the MDM Observatory have yielded R magnitudes ($`21<\mathrm{R}<26`$) for objects in the F0230 and F2240 fields as shown in Figure 10 and Table 2 (D. Helfand, private communication).
### 6.3 Field-to-field scatter
The success rate of redshift determination varied strongly from field to field as shown in Table 1; $`N_{obs}`$ and $`N_z`$ are respectively the number of observed sources and the number of sources with redshift determinations. The striking differences in redshift success between fields are illustrated in Figure 10; the fields at 22<sup>h</sup> have far more redshift determinations than the fields at 02<sup>h</sup>, which average only a $`5`$ per cent success rate. This region also has an anomalously low number of optical identifications in the FIRST survey (D. Helfand private communication), even though the surface density of radio sources is close to the average.
Comparing the two fields for which we have photometric information down to R $`26`$, in the case of F2240, E(B-V)=0.067, while for F0230, E(B-V)=0.027 (D. Helfand, private communication). Thus it is not Galactic extinction that produces the very different identification rates. We are looking at either intergalactic extinction or evidence of large-scale structure.
If we assume that radio galaxies cluster as optical galaxies, we can make a crude estimate of the expected variance in the number of measured redshifts in each WHT field. As discussed later, the typical redshift of galaxies in our sample is $`z0.2`$, and so each 1 field samples a volume of roughly $`1.7\times 10^4h^3`$ Mpc<sup>3</sup> out to $`z=0.2`$. Ignoring the difference in shape, this volume is equivalent to a 26$`h^1`$Mpc cube, for which we would expect a fractional variance in the number of optical galaxies to be between 0.3 and 0.4 (Loveday et al., 1992). The observed field-to-field variance is 0.6, which may reflect the higher clustering amplitude of radio galaxies.
Whatever the true explanation, the observed variation is important in itself because of implications for the number of radio-sources with faint optical continuum but bright emission lines. We return to this issue in Section 8 when estimating $`N(z)`$.
## 7 OPTICAL AND RADIO PROPERTIES OF THE SAMPLE
Table 2 shows the properties of objects in our sample for which we have data (either spectroscopic or photometric) on the optical counterpart. The columns are as follows:
(1) Source number.
(2) Right ascension and declination of the targeted object. These coordinates are from those of the FIRST survey except in the case of 2240\_004 for which the fibre was placed at the centroid of a double source.
(3) Offset of the optical counterpart in the APM catalogue before correction for sytematic offsets.
(4-5) Apparent magnitudes R and B of the optical counterpart.
(6) Radio flux density at 1.4 GHz.
(7) Redshift.
(8) Spectral classification.
(9) Emission lines detected.
Plots for the subsample with redshift determinations are presented in Figure 11 for (1) R magnitude vs. radio flux, (2) R vs. redshift, (3) B magnitude vs. radio flux, and (4) B vs. redshift. The solid line in the R-$`z`$ plot of Figure 11 represents the relation
$`RM_R=5+5\mathrm{l}\mathrm{o}\mathrm{g}_{10}d_L(pc)255\mathrm{l}\mathrm{o}\mathrm{g}_{10}\mathrm{H}_0+`$
$`+5\mathrm{l}\mathrm{o}\mathrm{g}cz+1.086(1q_0)z`$ (1)
($`d_L`$ the luminosity distance), obtained by a minimum $`\chi ^2`$ fit to the data with the absolute magnitude $`M_R`$ as free parameter. This analysis, for $`H_0=`$ 50 Km sec<sup>-1</sup>Mpc<sup>-1</sup> and $`q_0=0.5`$, gives a value of $`M_R23.1`$ with little scatter ($`\mathrm{\Delta }M_R0.5`$ at the $`2\sigma `$ level - dashed lines in Figure 11). There are clearly two populations in the diagram, and to find the best fit for early types we excluded Seyfert-type galaxies and starbursts; the result from using all galaxies was virtually identical. The result is in very good agreement with previous estimates (see e.g. Rixon, Wall & Benn, 1991; Gruppioni, Mignoli & Zamorani, 1998; Georgakakis et al., 1999), showing that passive radio galaxies are reliable standard candles. The solid line in the B-$`z`$ plane of Figure 11 shows a $`\chi ^2`$ fit yielding $`M_B21.3`$, although the fit is not as good as that in the R-$`z`$ plane. The fit is improved if only early-type galaxies are included.
Figure 12 shows radio flux S<sub>1.4GHz</sub> vs. redshift for the objects with a redshift determination. Though the number of objects in each class is small, we can see that different classes of objects occupy different regions of the planes in Figures 11 and 12:
* Early-type galaxies are found at intermediate redshifts ($`0.1z0.6`$) and make up the entire spectroscopic sub-sample with $`0.2z0.5`$. Their radio fluxes lie in the range $`1\mathrm{S}_{1.4\mathrm{GHz}}10`$ mJy and optically they appear as relatively faint objects (R $`17`$, B $`18.5`$). Their absolute magnitude $`M_R23`$ is consistent with FRI-type radio galaxies (Ledlow & Owen, 1996).
* Late-type galaxies are found from low to intermediate redshifts ($`0.05z0.2`$) perhaps constituting an intermediate step between early-type and starburst galaxies. Their radio fluxes range from S$`{}_{1.4\mathrm{GHz}}{}^{}1.5`$ mJy to S$`{}_{1.4\mathrm{GHz}}{}^{}6`$ mJy, with a single exception of one radio-bright object (2236\_013, S$`{}_{1.4\mathrm{GHz}}{}^{}=116`$ mJy). The B and R magnitudes show intermediate values, $`15\mathrm{R}17`$ and $`15\mathrm{B}19`$.
* Starburst galaxies in the sample are very nearby ($`z0.05`$), optically-bright (R $`14`$, B $`14.5`$), and faint in radio emission (S$`{}_{1.4\mathrm{GHz}}{}^{}3`$ mJy). Two of them (0230\_071 and 0230\_061) are separated by $`1h^1`$Mpc and show very similar properties, suggesting that interaction may be responsible for the enhanced star-formation rate.
* Seyfert2 galaxies are found at intermediate redshifts ($`z0.2`$) and their radio flux densities are extremely low (S$`{}_{1.4\mathrm{GHz}}{}^{}1`$ mJy). Photometric data are available for only one. These objects lie close to the line delimiting Seyferts from HII galaxies in the log(OIII/$`H\beta `$)/log(OII/$`H\beta `$) plane (Rola, Terlevich & Terlevich, 1997).
* Seyfert1 galaxies / quasars exhibiting broad emission lines are only found at $`z0.9`$. These objects are very bright in the radio but quite faint in both the R and B bands. 2240\_004 is a double source showing characteristic radio lobes symmetric with respect to the optical centre (which does not show a radio core).
* Unclassified objects. These are probably early-type galaxies since they occupy the region in the R-S and B-S planes which is occupied by early-type objects. Also their colours are consistent with them being red/old objects; almost 50 per cent of objects this subsample have R magnitudes (with $`18.6\mathrm{R}20`$) in the APM catalogue but no B magnitudes, because they are below the APM survey limit. For these objects we used the R-$`z`$ relation of Figure 11 to estimate redshifts. The corresponding values are given in Table 2, distinguished by asterisks. (Note that we do not use these redshift estimates in the following analysis.)
## 8 THE REDSHIFT DISTRIBUTION $`N(z)`$
To construct the redshift distribution of radio sources $`N(z)`$ at the 1-mJy level we use the WYFFOS data described here (the WHT sample) and measurements from the Phoenix survey (Hopkins et al., 1998).
### 8.1 The WHT Sample
The number of objects in the WHT sample belonging to each spectroscopic class (Section 6) are summarized in Table 4.
The redshift distribution of these objects split by their spectral type is presented in Figure 13. Here AGN-powered sources (i.e. Seyfert1 /quasars and Seyfert2) are plotted together, the low-$`z`$ objects corresponding to Seyfert2’s. Our results are in good agreement with the predictions of Jackson & Wall (1998) for the relative contribution of different classes of objects at mJy levels (see Figure 17 of Jackson & Wall, 1998). At S$`{}_{1.4\mathrm{GHz}}{}^{}1`$ mJy their models predict (C.A. Jackson, private communication) 6.4 per cent of the whole population to be broad- and narrow-emission line objects (high excitation FRII’s, comparable with our class of Seyfert1+Seyfert2 galaxies), 48.1 per cent to be low-excitation FRI’s and FRII’s (early-type galaxy spectra), 28.7 per cent to be starforming galaxies (i.e. starbursting+late type) and a final 16.8 per cent to be BL Lacs (no features in their spectra - some of which may be amongst our unclassified objects).
Figure 14 shows the redshift distribution of all extragalactic objects of our sample with S$`{}_{1.4\mathrm{GHz}}{}^{}1`$ mJy and $`z0.6`$. These flux-density and redshift limits cut 8 objects from the spectroscopic sample (all 5 Seyfert1 galaxies, one early-type, one Seyfert2 and one starburst galaxy), so that Figure 14 contains a total of N$`{}_{\mathrm{TOT}}{}^{}=34`$ sources.
### 8.2 The Phoenix Sample
The Phoenix survey at 1.46 GHz (Hopkins et al., 1998) includes observations from two different fields. The Phoenix Deep Field (PDF) covers an area $`2^{}`$ in diameter centred at RA(2000)=$`01^h\mathrm{\hspace{0.33em}14}^m\mathrm{\hspace{0.33em}12}^s.16`$, Dec(2000)=$`45^{}\mathrm{\hspace{0.33em}44}^{}\mathrm{\hspace{0.33em}8}^{\prime \prime }.0`$, while the Phoenix Deep Field Sub-region (PDFS) covers an area of $`36^{}`$ in diameter centred at R(2000)=$`01^h\mathrm{\hspace{0.33em}11}^m\mathrm{\hspace{0.33em}13}^s.0`$, Dec(2000)=$`45^{}\mathrm{\hspace{0.33em}45}^{}\mathrm{\hspace{0.33em}0}^{\prime \prime }.0`$. The completeness of the radio catalogue is estimated to be $``$ 80 per cent down to 0.4 mJy for the PDF catalogue and $``$ 90 per cent down to 0.15 mJy for the PDFS catalogue. Optical and near-infrared data were obtained for many radio sources in the sample. A catalogue of 504 objects with redshift estimates and photometric information down to a limiting magnitude of R=22.5 has been produced. This represents $``$ 47 per cent of the whole radio sample. From this catalogue, the properties of the sub-mJy population have been analysed (Georgakakis et al., 1999).
To compare directly with our data, we have selected only objects with radio fluxes S$`{}_{1.4\mathrm{GHz}}{}^{}1`$ mJy and with a reliable redshift estimate, yielding a total of N$`{}_{TOT}{}^{}=52`$ sources. Figure 15 shows the redshift distribution of this sample for $`z0.8`$. The solid line is for all the objects, while the dashed line is the distribution obtained by imposing the further constraint R $`18.6`$, the limiting magnitude for the completeness of our sample.
The $`N(z)`$ distribution from the Phoenix sample is dominated by two peaks at $`z=0.15`$ and $`z=0.4`$, presumably due to two major galaxy concentrations. These features, together with the exceptional field-to-field variations which occur in the WHT sample show how seriously large-scale structures affect deep radio surveys. An $`N(z)`$ determination requires as many independent areas as possible; and to this end we have no hesitation in joining our sample with the Phoenix sample to make such a determination.
### 8.3 Incompleteness
We must consider the factors affecting the redshift completeness of the samples before comparing to model predictions.
For the Phoenix survey the completeness down to 0.4 mJy is $`80\%`$ in the shallower region, so we expect that it is essentially complete to $`1`$ mJy. Thus incompleteness in the spectroscopic sample comes only from the limiting magnitude R=22.5 which restricts identification and spectroscopy to 47 per cent of the sample.
The WHT sample has three types of incompleteness:
* Incompleteness in the radio catalogue. Becker, White & Helfand (1995) estimated the catalogue obtained from the FIRST survey to be 80 per cent complete down to a flux density of 1 mJy. The issue is how this incompleteness affects the population mix in that compact sources at 1 mJy will be present, while resolution will preferentially lose sources of extended structure. Paradoxically this is more of a problem for AGN and as these have been excised from our sample by virtue of their high redshifts, this issue may be of no significance. Starburst galaxies have relatively compact structures (Richards, 1999) so that the resolution effects of a 5 arcsec beam will not be serious.
* Incompleteness in the acquisition of the spectroscopic data. Not all the catalogued radio-source positions in the 8 fields observed had a fibre placed on them (Section 3 and Figures 1 and 4). This was due (1) the geometry of the WYFFOS spectrograph, and (2) the issue of multi-component objects (see Magliocchetti et al., 1998) affecting $`10`$ per cent of the whole sample. In these (recognised) cases the fibre was placed at the mid-point of the double/triple source.
* Incompleteness in redshift determination. Several poorly-defined factors are involved. For instance, as the sample was selected on the basis of radio flux alone, the process of redshift determination is biased against intrinsically dusty sources; but this same bias applies to the Phoenix sample. Intergalactic dust screening could also be involved; see Section 6.3 regarding field-to field variation.
Figure 16 compares the number of sources in the whole spectroscopic sample per radio flux interval with the relative distribution obtained for the sub-sample with measured redshifts (hereafter, the ‘$`z`$-sample’). The ratio $`N(S)(z)/N(S)13`$ per cent between the samples stays approximately constant with S, at least up to S $`15`$ mJy, at which point the number of objects in the samples becomes too small to make a meaningful comparison.
The constancy of the $`N(S)(z)/N(S)`$ ratio in Figure 12 implies that there is no radio flux-bias in the process of $`z`$ determination. Thus a reasonable supposition is that incompleteness in the radio survey (resolution in particular) does not seriously affect the population mix of the $`z`$-sample. We therefore suggest that apart from the known incompleteness due to fibre coverage, the optical flux is the prime cause of incompleteness. Though we have not explicitly imposed a magnitude limit, the redshifts require adequate signal-to-noise in the optical spectra. For non-emission line galaxies, this effectively means a completeness in the continuum magnitude limit, which, as seen in Figure 10, corresponds to R $`18.6`$ mag. Using the R-$`z`$ relationship of Figure 10 and equation (1), for a value of R $`18.6`$, the $`z`$-sample (which only includes objects with radio fluxes $`S_{1.4\mathrm{GHz}}1`$ mJy) is complete in redshift up to $`z0.3\pm 0.1`$. The uncertainty corresponds to the scatter in the R-$`z`$ relation. The procedure cannot be applied to Seyfert1 objects, which do not follow the $`Rz`$ relation. However our conclusions are not affected given that this class of objects is mainly found at redshifts $`z>0.3`$.
As a further test, Figure 15 shows the variation in the $`N(z)`$ for the Phoenix survey if we consider a magnitude cut of R=18.6 (dashed line). The number of sources per redshift is unaffected to $`z=0.25`$. The percentage then starts dropping as the R=18.6 magnitude limit comes into play. From a similar analysis using our R-$`z`$ relation, we find that the Phoenix survey sample is essentially complete to $`z0.9`$ for objects other than radio AGN.
### 8.4 Comparison with models and results
The $`N(z)`$ model predictions used here are from the study by Dunlop & Peacock (1990; hereafter DP) in which Maximum Entropy analysis was used to determine the coefficients of polynomial expansions representing the epoch-dependent radio luminosity functions of radio AGNs. The analysis incorporated the then-available identification and redshift data for complete samples from radio surveys at several frequencies. Indeed the 7 derivations of luminosity functions carried out by DP predict source counts and $`N(z)`$ distributions for frequencies 150 MHz to 5 GHz down to 100 mJy with considerable accuracy, the input data spanning approximately this parameter space. However, extrapolating down to mJy levels, the models show large variations in the predicted $`N(z)`$; see Figure 1 of Magliocchetti et al., 1999. Nevertheless, all predictions show a broad peak with median redshift of about 1. Some models also produce a ’spike’ at small redshifts indicating that at such low flux densities, the lowest-power tail of the local radio luminosity function begins to contribute substantial numbers of low-redshift sources. Note that the DP formulation did not encompass any evolving starburst-galaxy population explicitly; it was restricted to two radio AGN populations, the ’steep-spectrum’ extended-structure (FRI and FRII) objects and the ’flat spectrum’, compact objects (predominantly Seyfert1/quasars and BL Lacs). Hence, at some level, the low-$`z`$ spike must be considered as a fortunate artifact of the models.
To compare the predictions from DP models with the data we have to consider the incompleteness effects described above. With regard to normalization of surface density, the DP predictions give $`508N_{TOT}(\mathrm{S}1\mathrm{mJy})570`$ for an area of $`8\times (\pi \mathrm{\hspace{0.33em}0.5}^2)`$ square degrees. For our 8 chosen fields of the FIRST survey, we found N=525 (after correcting for the presence of multicomponent sources), in excellent agreement.
Figure 17 shows the distribution of objects in the WHT+Phoenix sample (76 objects up to $`z=0.52`$) as a function of redshift. The shaded area corresponds to the redshift range $`z0.3\pm 0.1`$ where the combined sample starts losing completeness due to magnitude incompleteness in the WHT sample. The smooth curves are the predictions from DP models. None of these provides a good fit. The percentage of objects at $`z0.1`$ is seriously overestimated in all the models, especially 6 and 7 that feature an unrealistic $`z0`$ spike, presumably an artifact caused by extrapolating too far. Though the uncertainties are large, only one model (model 3) roughly follows the steady rise in $`N(z)`$ to $`z0.2`$, while all the others either show a plateau or decrease for $`z0.1`$. Note that model 3 is also the closest to the correct low-$`z`$ normalisation.
Our data show that the population of starburst galaxies constitutes only a small fraction of the radio objects at the mJy level, contrary to early claims (e.g. Windhorst et al., 1985). If the majority of objects at radio fluxes 1 mJy $`S3`$ mJy were starburst galaxies we would have obtained redshifts for all of them. In fact such objects are on average the brightest in apparent magnitude and the closest in the $`z`$-sample (see also Benn et al., 1993; Gruppioni, Mignoli & Zamorani, 1998; Georgakakis et al., 1999). Allowing for the sensitivity limit of our redshift determinations, and the strong emission line spectrum for this type of objects we would expect $``$100 per cent redshift completeness up to $`z0.2`$. If indeed they constituted most of the sources at 1 mJy, we would have therefore found a ratio $`N(S)(z)/N(S)`$ to be $`1`$ at low radio fluxes. Figure 16 shows that this is not the case; all flux intervals have the similar probability ($`13`$ per cent) of redshift determination. Thus only $`5`$ per cent of the population at S$`_{1.4\mathrm{GHz}}`$ = 1 mJy can be starburst objects; the great majority of identifications remain AGN, with predominantly early-type galaxies as hosts, too faint in optical magnitude to be detected in our spectroscopic observations. Very similar conclusions are reached by Gruppioni, Mignoli & Zamorani, 1998 and Georgakakis et al., 1999 for their mJy/sub-mJy samples, in which the optical observations were pushed down to R=24 and R=22.5 magnitudes respectively.
We stress again that we did not impose a magnitude limit before our spectroscopic observations, so that any optically-faint star-forming galaxies would have been observed in our sample. The presence of emission lines would have allowed us to get redshifts for those sources much fainter than the APM survey magnitude limit. However, we found that that for R $`17`$ the only classes of objects detected were early-type and Seyfert1 galaxies. This makes the constraint on the percentage of star-bursting galaxies at mJy level yet more robust.
We note that our determination of $`N(z)`$ at low $`z`$ provides an additional datum on the shape of the complete $`N(z)`$ distribution. The overall normalisations between our sample and the DP predictions agree; yet the DP predictions significantly overestimate $`N(z)`$ at $`z<0.4`$. This disagreement at low redshifts may imply that we expect to find many more sources at $`z0.3`$, perhaps consistent with the DP model 3.
However there is a significant uncertainty caused by the large variation in fraction of optically identified sources in different 1 degree WHT fields (the fraction varies by a factor 4, see Table 1). The combined Phoenix and WHT samples are equivalent to roughly 10 times the area of one WHT field, so assuming Poisson statistics we expect roughly a factor 1.5 uncertainty in the overall surface density. Allowing for such variations, any of the DP models 1-4 too are reasonably consistent with the data.
## 9 CONCLUSIONS
We have carried out multi-object spectroscopy of an unbiased selection of FIRST radio sources (S $`0.8`$ mJy) by placing fibres at the positions of 365 sources ($``$ 69 per cent of the complete radio sample). The spectra obtained have enabled us to measure 46 redshifts, $`13`$ per cent of the targeted objects. APM data have provided morphology and photometric data for the corresponding optical identifications. The photometry shows that redshift measurements were obtained only for objects brighter than R $`20.5`$ mag; from the tight R-$`z`$ relation observed, the redshift sample is estimated to be $``$100 per cent complete to R=18.6 mag.
The objects in the spectroscopic sample with R $`20.5`$ are a mixture of early-type galaxies at relatively high redshifts, $`z0.2`$ ($`45`$ per cent of the sample), late-type galaxies at intermediate redshifts, $`0.02z0.2`$ ($`15`$ per cent), and very local starburst galaxies with $`z0.05`$ ($`6`$ per cent). We also found a number of Seyfert1/quasar type galaxies, all at $`z0.8`$ ($`9`$ per cent of the sample), two Seyfert2’s (4 per cent), and 4 stars. The number of objects with featureless spectra are most probably early-type galaxies, given the shape of the continuum, the lack of emission lines and the red colours. Using the R-$`z`$ relation derived for our sample, we conclude that they are likely to have $`z0.3`$. The redshift incompleteness does not depend on radio flux density; optical apparent magnitude is the only identifiable factor. Using again the R-$`z`$ relation determined for the sample, we estimate $``$ 100 per cent completeness for the spectroscopic sample up to $`z0.3\pm 0.1`$.
To define $`N(z)`$ at S$`_{1.4\mathrm{GHz}}`$ as well as possible, we have combined our sample with the Phoenix spectroscopic sample (Georgakakis et al., 1999), which we estimate to be complete (for non-AGN objects) to $`z=0.9`$. The combined distribution (Figure 17) shows the following:
* The redshift distribution rises up to $`z0.05`$ and is then approximately leveled to $`z=0.3`$.
* The total number of sources predicted by the luminosity-function models of Dunlop & Peacock (1990) agrees with that observed.
* None of the models provides a good fit to the shape of $`N(z)`$. The percentage of objects at $`z0.1`$ is seriously overestimated in almost all the models, especially for the pure-luminosity and luminosity/density evolution models (DP 6 and 7) that feature an unrealistic $`z0`$ spike.
* The normalization of the models appears to be too high to fit the observed $`N(z)`$ for $`z0.2`$. This disagreement may imply that the model shape is wrong, and there are more sources at $`z0.3`$ than indicated by the models. Alternatively, the discrepancy could be due to observing a low density by chance, given a large variance in galaxy density caused by strong clustering.
* The $`N(z)`$ at $`S_{1.4\mathrm{GHz}}=1`$ mJy is dominated by AGN, and starburst objects constitute less than 5 per cent of the total. This is a robust conclusion. More starburst galaxies would have substantially raised the proportion of objects with redshift determinations; and a significant intrusion of starburst galaxies at the lowest radio flux densities would have resulted in a higher success rate in redshift determination with decreasing flux density. The great majority of objects in the sample at this level are AGN associated with early-type galaxies whose optical continua and weak-to non-existent emission lines place them at or below the limit of our spectroscopic survey.
The accurate definition of the low$`z`$ end of the $`N(z)`$ relation has impact in four areas: (i) the population mix, which is critical for testing and refining dual-population unified models; (ii) the definition of the local luminosity functions, important for modelling both the form and cosmic evolution of the overall luminosity functions; (iii) the derivation of spatial measurements of the large-scale structure from angular measurements; and (iv) constraints which it enables to be placed on the global star-formation-rate up to $`z=0.3`$.
ACKNOWLEDGEMENTS
MM acknowledges support from the Isaac Newton Scholarship. We thank David Helfand for extremely helpful discussions. The WHT is operated on the island of La Palma by the Isaac Newton Group in the Spanish Observatorio del Roque de los Muchachos of the Instituto de Astrofisica de Canarias. GC acknowledges a PPARC Postdoctoral Research Fellowship.
|
warning/0004/physics0004077.html
|
ar5iv
|
text
|
# Design and Simulation of a High Frequency High Power Rf Extraction Device Using a Dielectric-Loaded Waveguide
## 1 Introduction
One trend in proposed future high energy linear collider designs is the use of high accelerating gradients in room temperature high frequency structures. This approach has the advantage that the average rf power for a given accelerating gradient scales as 1/(frequency)<sup>2</sup> R1 , while the breakdown field is empirically found to scale approximately as (frequency)<sup>1/3</sup> R2 . At the same time, the technology of conventional (klystron) rf sources is difficult to implement at frequencies higher than X-band (11.4 GHz) R1 . Two- beam acceleration R3 overcomes this difficulty by using a conventionally accelerated high current drive beam as the rf source for the high energy accelerator.
Two beam acceleration requires generation and propagation of a high current drive beam, and an efficient transfer structure to extract energy from the drive beam. We propose that a dielectric loaded waveguide represents an optimal power extraction/transfer structure. Dielectric structures have been investigated experimentally at some length R4 ; R5 . We summarize some of the important features of this class of devices compared to conventional copper rf structures:
* Simplicity of construction. As shown in figure 1, the structure is a simple ceramic tube which is inserted into a conducting copper sleeve.
* Higher order mode damping. Parasitic HEM modes generated by off-axis injection errors can be easily suppressed R6 by interrupting the outer conductor with axial cuts and surrounding the sleeve with an absorbing material. Note that this technique requires no modification to the dielectric tube and only modest machining of the exterior conducting sleeve.
* Recent advances in dielectric materials for microwave applications. Very low loss ($`tan\delta 10^4`$) and high Q materials are commercially available R7 .
Other issues which may represent possible limitations on the performance of dielectric structures are actively being investigated:
* Efficient rf coupling. The wakefield generated by the drive beam must be coupled out to a rectangular waveguide for transfer to the accelerating structure. This involves a mode conversion from TM<sub>01</sub> in the dielectric tube to TE in the waveguide. Significant progress on efficient coupling design has been made R8 .
* Charging. Beam halo or missteered beam impinging on the dielectric tube can cause large electrostatic charge buildup, resulting in deflection of the beam or volume breakdown of the structure. We have observed both these effects in our experiments using polymer based materials such as rexolite or nylon but have not observed this in any of the glass or ceramic materials we have tested. Changes in microwave dielectric properties with time due to radiation damage have also not been seen in our experiments.
* Rf breakdown. The work in this area has been almost entirely confined to copper structures R2 . We have currently generated rf surface fields in ceramic structures $`>`$ 10 MV/m at 15 GHz R5 (limited by the available drive beam properties); no breakdown was observed at this level. We expect that significantly higher fields can be developed at these frequencies before breakdown occurs but it is worth pointing out that these devices are interesting as power extraction structures even at 10 MeV/m scale gradients.
## 2 Rf Power Generation in a Dielectric Structure
We follow the approach and notation of reference R9 . Consider a dielectric structure (figure 1) with inner radius a and outer radius b. An electron bunch train passing through the device will generate a steady state RF power
$$P=\frac{l_s^2\omega }{4c}(\frac{q_b}{T_b})^2(\frac{R^{}}{Q})(\frac{1}{\beta _g}1)F^2(\sigma _z)$$
(1)
where $`l_s`$ is the length of the structure, $`q_b`$ is the charge/bunch, $`cT_b`$ is the bunch separation, $`\beta _g`$ is the group velocity and $`F(\sigma _z)`$ is a form factor proportional to the bunch spectrum (assumed Gaussian). $`R^{}/Q`$ is the normalized ratio of the shunt impedance of the structure to its quality factor and encapsulates all the details of the structure geometry.
To compute $`R^{}/Q`$ for a dielectric structure, we use
$$R^{}/Q=\frac{4E_{z0}}{q_b\omega }$$
(2)
where $`E_{z0}`$ is the longitudinal decelerating wakefield in the device. In the limit of an infinitely long structure, the analytic expression for the wakefield is given in R10 . This provides $`R^{}/Q`$ via equation 2, and we can proceed to compute and optimize the properties of our structures.
Since the formalism of reference R10 does not allow for the effects of structure boundaries, rf coupling etc., it only yields an approximate evaluation of the properties of these structures. We will also give more accurate results of numerical calculations which include finite structure length and rf extraction effects.
## 3 Design of the Dielectric Power Extraction/Transfer Structures
In this section we give a reference design for both 15 and 30 GHz structures. While the choice of device parameters are not identical to the CLIC final design, this work allows the study of the physics and engineering aspects of dielectric- based power extraction devices. We have attempted to develop power levels of 100 MW with large coupling to the beam and using a common, commercially available dielectric.
While 30 GHz is the design frequency of CLIC R9 , there are a number of advantages in developing a 15 GHz structure first. The lower frequency is compatible with both the microwave measurement equipment and the beam spectrum available at the AWA R5 . Also, there is some question whether the use of 30 GHz rf does in fact provide significant performance improvement over 10 GHz range devices R11 .
The choice of dielectric is Cordierite which has a relatively low permittivity (4.5) and low losses (Q=5000, $`tan\delta =0.0002`$). This material is commercially available R7 , and we have verified its dielectric properties in bench tests.
The choice of dielectric is not unique; for example, steatite, another commercially available material, has a permittivity of 5.8 and loss tangent of 0.0001 and would perform equally well. Theoretically one could also use a low coupling strength $`R^{}/Q`$ using a thin layer of high permittivity material to line the structure wg .
The inner radius a must be chosen to transport all the drive beam. The choice of dielectric material and the device frequency then fix the outer radius b. Using the methods outlined previously, we can calculate $`R^{}/Q`$ and power generated in these devices assuming CLIC test facility beam parameters.
Because of the simple geometry, the structures will be easy to machine. Conservatively assuming that the outer and inner radii of the structures can be machined to a tolerance of 25 $`\mu `$m, the worst case deviations from nominal frequency due to this tolerance are 3.4% for the 30 GHz device and 1.5% for the 15 GHz structure.
## 4 Numerical Calculations
We also used the finite difference time domain program ARRAKIS R12 to calculate the power generated in these structures while taking into account the finite structure length, group velocity of the wakefield, etc. In order to simulate the effects of power extraction from the structure, an absorbing boundary was placed at the downstream end of the dielectric tube in the computational geometry.
Figures 2a and 3a show the longitudinal wake potentials for the 15 and 30 GHz devices for a single drive bunch and for a train of 1 nC bunches spaced by 10 cm. Buildup of fields to equilibrium occurs relatively quickly– after four bunches (1.33 ns) for the 15 GHz device and three bunches (1.0 ns) for the 30 GHz structure due to its larger group velocity. The apparent “damping” of the wakefield after each bunch is in fact due to the large group velocity of the fields and the finite length of the devices. The 15 GHz wake shows some additional structure due to the excitation of the TM<sub>02</sub> mode at 42.8 GHz and the TM<sub>03</sub> at 73.4 GHz. The higher frequency modes result in some inefficiency but in general will not be transported out of the structure when the coupling is optimized for the fundamental frequency.
The bunch train wakes shown in figures 2b and 3b reflect finite structure length and group velocity effects by their not purely sinusoidal character. In order to compute the useful power generated at the fundamental, we filter the data by performing an FFT and zeroing all Fourier components outside a fixed window in $`\mathrm{\Delta }f/f_0`$. The inverse FFT then yields the amplitude W of the wake within that bandwidth, and the power generated is
$$P=W(MeV/m/nC)\times q_b(nC)\times \overline{I}(=q_b/T_b)(A)\times (l_s(m)).$$
(3)
Table 1 summarizes the parameters of these structures. The power generated is large, on the order of 100 MW for a 30 cm test structure with $`\mathrm{\Delta }f/f_0=4\%`$ and scaling quadratically with the length of the device. In the 30 GHz case the dielectric device performance is comparable to the CLIC designs R9 while being easier to manufacture and to incorporate parasitic mode damping.
## 5 Summary
We have presented designs for dielectric power extraction structures for two-beam accelerators which are able to produce power levels competitive with conventional copper structures while at the same time possessing the advantages of ease of construction and straightforward implementation of parasitic mode damping. We plan to build and test a 15 GHz device at our laboratory and if successful build a 30 GHz device for the CLIC test facility.
|
warning/0004/hep-th0004152.html
|
ar5iv
|
text
|
# Scale Relativity in Cantorian ℰ^(∞) Space and Average Dimensions of Our World
## 1 Introduction
Rephrasing I.Kant we can say that any physical theory should start from the first principles and assumptions based on intuition directly related to the existing physical phenomena, then proceed to the concepts, complete theory with a central idea , and crown it with the experimental verification and testable predictions. One of the best examples of this path is provided by the theory of relativity and quantum mechanics. Their further development spurred by the need to unite gravity and quantum mechanics resulted in the birth of the string theory.
During 30 years of its development the string theory demonstrated remarkable achievements in all but one respect: there is no direct way to verify its theoretical results. In addition, there is at least one ”thorny” question: why does the world we are living in appears to be $`3+1`$-dimensional? Up to now string theory could not produced a satisfactory answer. It seems that string theory ignored possible alternative basic assumptions which, as we will see later, could have produced a plausible answer: multi-fractality of spacetime, a possibility of spacetime dimensions ranging from negative dimensions to infinite ones,as in the $`^{\mathrm{}}`$ Cantorian theory of El Naschie and the resolution-dependent character of physical laws as described for example by Nottale’s scale relativity .
One of the basic problem facing future developments of a successful physical theory is a description of the quantum ”reality” without introducing by hand a priori existing background. Interestingly enough, as one of the first steps in formulating the theoretical foundations of physics Newton disposed of this problem rather easily by simply introducing (almost at the very beginning of ”Principia”) the absolute space and time as entities ”which do not bear any relation to anything external”. It is clear that without an attendant background ( either not changing or changing rather slowly as compared to developing phenomena) any description of a physical theory looks impossible.
To formalize dynamics of physical phenomena we have to answer the following question: the dynamics with respect to what? For example, the Lagrangian of the string theory refers to the background of proper time (which is rather a vague notion since we have to have a physical device attached to a string and measuring time intervals which in turn are not defined precisely) and coordinates along the string. The very spacetime exists only to ”the extent that it can be reconstructed from the 2-D field theory”.
n other words, there is no ” background spacetime ” $`abinitio`$. Quantum Spacetime is truly a ” process in the making ” which implies that gravity should be described by a nonlinear theory. We are dealing with a nonlinear complex dynamical system that is able of self-tuning in a fashion similar to biological systems in nature . To put it succinctly, we can say that the universe could be viewed as the ”organism”, or using a more fashionable wording, the ultimate quantum Machian computer. Here we refer to the Mach principle: a physical theory should contain only the relations between physical quantities. It is the relations (reflected in algebraic operations) among their representative elements ( monads) that govern a system’s evolution . Therefore algebraic-categorical relations are the only meaningful relations which are applicable for a description of such an ”organism” \- .
he problem of spacetime generation could be addressed with the help of what we call a new relativity -,, a theory which is based on a few postulates. One of these postulates is an old bootstrap idea by Chew: a physical system in a process of its development simultaneously generates its own background. In particular, geometry ( initially nonexistent) is produced by a recursive ( self-referential) process starting from a ”hyperpoint” endowed with infinite dimensions, or equivalently with an infinite amount of information ,-.
The latter implies a possibility of emergence of an ensemble of interacting universes having all possible dimensions ranging from negative values to infinity. The ensemble average yields the 3-dimensional world of our perception. Thus an assumption about existence of such an ensemble ( emerging as a result of a bootstrapping process) of the universes with all possible dimensions points in the direction where even string theory with its 10, 11, 26 dimensions does not go. Implications of infinite dimensions which in the final run can be viewed as ”words” constructed out of infinite information contained in a ”hyperpoint” provides us with a source of a bootstrapping mechanism of spacetime generation. Equally, the energy also emerges as a result of self-organization of the initial infinite amount of information (number of bits).
The inter-connectivity of energy, dimensions, information becomes especially visible if we consider the information contained in all possible $`p`$-loop histories. If we recall that $`0`$-loop corresponds to a point (0-D),$`1`$-loop corresponds to a line ($`1`$-D), etc. then a total number of all the $`p`$-loop histories in $`D`$-dimensions is $`2^D`$. As a result the respective Shannon information $`KLn(N)`$ is equivalent to the number of dimensions $`D`$ ( up to a non-essential constant factor). Information in turn creates energy and is created by energy. This leads to a ”triad” of equivalence generated by a recursive loop: dimensions $``$ information $``$ energy. The recursive loop could be viewed as some kind of a spiral of eternal creation and annihilation of energy, dimensions and by implication geometry.
Another basic postulate of a new relativity is Nottale’s scale relativity takes the Planck scale $`\mathrm{\Lambda }=1.62\times 10^{35}`$m as the ultimate ( and absolute) scale. This postulate allows one to unify the spaces of different dimensions similar to the unification of time and space in Einstein’s relativity ( on the basis of the absolute character of the speed of light). Subsequently, a new relativity theory does not need to deal with compactification and decompactification of superstring theory. The latter pose a serious problem since it leads to a huge number of possible phenomenological theories of the world. A new relativity theory adopting the infinite-dimensional universe and the ultimate scale $`\mathrm{\Lambda }`$ attainable in Nature suggests that topological field theories are the most natural candidates for a ”theory of the world”. Indirectly this is confirmed by the assumption that below Planck scale there is no such thing as a distance ( interval) meaning that the topology becomes a decisive factor there.
The third postulate of a new relativity is an extension of the ordinary spacetime to noncommutative C-spaces thus making a quantum mechanical loop equations fully covariant. This is done by extending the concepts of ordinary spacetime vectors and tensors to non-commutative Clifford manifolds where all p-branes are unified with the help of Clifford multivectors. There exists a one-to-one correspondence between a nested hierarchy of 0-loop,1-loop,…, p-loop histories in $`D`$ dimensions coded in terms of hypermatrices and single lines in Clifford manifolds. This is roughly equivalent to Penrose twistor program. The respective non-commutative geometry is taken to be the transfinite continuum of Cantorian fractal spacetime studied extensively by M.El Naschie and G.Ord .
Recently a new relativity theory claimed a few successes. One of those is especially interesting. A few years ago the leading figure in string theory E.Witten remarked, probably with some disappointment, that ” a proper theoretical framework for the extra term in the uncertainty” (stringy) ”relation has not emerged”. It has turned that a full-blown stringy uncertainty relation ( and its truncated version) could be straightforwardly derived within the framework of a new relativity with the help of its basic postulates . Also, in Appendix A we provide an elementary heuristic derivation of the stringy uncertainty relation based only on the postulate of the minimum attainable scale ($`\mathrm{\Lambda }`$) in Nature.
In addition, it was shown that a new relativity disposes of both EPR and black hole information loss paradoxes. More recently , it allowed one of us to derive in an elementary fashion the black hole area-entropy relation valid for any number of dimensions. In Appendix B we briefly review this derivation and demonstrate that Bekenstein-Hawking relation is a particular case of a more general relation. Furthermore, we argue why the holographic principle is a direct result of a new relativity.
A new relativity’s most surprising prediction of group velocity exceeding speed of light ( without resorting to tachyons) found its verification in recent experiments by Ranfagni and collaborators who discovered the evidence for ” faster than light ” wave propagation to distances up to one meter.
In section 2 we consider an ensemble of dimensions (including infinity) of quantum spacetime and derive its distribution function. The subsequent calculation of the respective average dimension associated with this distribution is given in section 3 where we obtain the value for this average which is very close to $`4`$ \+ $`\varphi ^3`$ ($`\varphi =(\sqrt{5}1)/2`$ is the golden mean) previously obtained on the basis of a Cantorian fractal spacetime model. A perceived 4-dimensional world then follows as a result of a coarse-grained long range averaging effect of the underlying Cantorian fractal geometry, as discussed by El Naschie . In addition we present a brief discussion of a plausible chiral symmetry breaking mechanism in Nature.
In section 4 we show why the ”cosmological constant problem ” is never an issue within the framework of a new relativity. This follows from the idea of the automatic self-organization and self-tuning of the universe in accordance with the renormalization group flow with respect to a local scaling microscopic arrow of time ( see Appendix C). We also propose a cosmological scenario as an alternative to the big bang, inflationary, brane-worlds and variable speed of light cosmologies. According to our model, the world began as a result of a non-equilibrium process of self-organized critical phenomena launched by vacuum fluctuations in Cantorian fractal spacetime. Its future evolution proceeded in such a way that that determined our existence in a metastable vacuum with a a subsequent transition to the renormalization group fixed point of a dimension $`D`$ = $`4`$ \+ $`\varphi ^3`$ .
The above ”fixed critical point” is not a true fixed point but rather a metastable one, which means that it does not represent a true vacuum. In section 5 we discuss how a new phase transition will eventually drive the universe to another (quasi-crystal) phase of a lower average dimension of $`<D>=\varphi ^3`$ representing the true vacuum. Using one of the basic postulate of a new relativity ( Nottale’s scale relativity) we arrive at two integral expressions which determine the upper limits imposed on a) the Hubble radius and b) the size of the quasi-crystal phase. These expressions show that both limits depend on the value $`\varphi `$ of the golden mean which is a recurrent feature of our analysis.
Finally, in the concluding section 6 we summarize our results and write down, for finite values of $`D`$, the unique quantum master action functional for the world in C-space, (outside spacetime). This action functional governs the full quantum dynamics for the creation of spacetime, gravity, and all the fundamental forces in nature. It appears that the quantum symmetry in such a world should be represented by a braided Hopf quantum Clifford algebra. The quantum field theory which follows from such an action is currently under investigation .
## 2 Distributions : Fermat’s Last Theorem and a Multidimensional World
To proceed with calculation of an average dimension of the observable world we make an assumption that there exists a certain ensemble of universes with dimensions ranging from some zero (reference) point( to be determined) to infinity. The above reference point is analogous for example to a zero point energy of a quantum oscillator. To simplify our analysis we choose a spherical symmetry and represent the introduced ensemble as a set of hyper-spheres whose radii are integer multiples of the fundamental quantum of length, the Planck scale $`\mathrm{\Lambda }`$. If we would have chosen , say hyper-cubes instead of hyper-spheres then it would have changed the respective geometry of the ensemble but not its topology. However the latter is more fundamental than the former and therefore justifies a choice of the spheres as the basic constituents of the ensemble.
In what follows we pattern our derivation after Planck’s approach to the black body radiation. The ensemble of the universes is in contact with an information bath prepared by some ”universal observer” existing (we use the word $`\mathrm{𝑒𝑥𝑖𝑠𝑡𝑖𝑛𝑔}`$ at the wont of a more appropriate word) $`\mathrm{𝐨𝐮𝐭𝐬𝐢𝐝𝐞}`$ of the ensemble. Such a frame of reference associated with the ” universal observer ” can be associated with a hyperpoint of infinite dimensions having an infinite amount of information ( see Appendix C) which could undergo a self-ordering ( bootstrapping) when perturbed ( even slightly).
The members of the ensemble ( hyper-spheres, or p-branes) can exchange energy ( and information) with the ”walls” of this bath at the Planck temperature $`T=T_P=1/\mathrm{\Lambda }`$ ( in units c=$`\mathrm{}`$=K =1) which is proportional to the number of hyper-spheres, that is to the information content of the bath. We assume ( in the spirit of things quantum) that the energy exchange occurs in integer ”chunks”. Equivalently, the hyper-spheres can absorb quanta ( from the bath) emit quanta ( to the bath), and exchange quanta between themselves. These quanta represent background independent geometric bits ( see Appendix B) or the true quanta of spacetime and thus replace gravitons of the linearized gravity which are not background independent. The geometric bits form their own background where they continue to evolve further. This is in agreement with one of the postulates of a new relativity ( described earlier), namely the Chew bootstrap idea. When the latter is applied to p-branes ( our hyper-spheres) it states that all the p-branes are made of each other.
ince geometric bits are quanta of fundamental excitations of spacetime we do not need to have $`\mathrm{𝑎𝑝𝑟𝑖𝑜𝑟𝑖}`$ an embedding target spacetime background where excitations of $`p`$-branes are to propagate. The above ensemble is self-supporting ( self-referential),i.e. it obeys the bootstrap conditions. This results in a background independent formulation of the emergence of spacetime in a sense that the target background spacetime is not fixed $`\mathrm{𝑎𝑝𝑟𝑖𝑜𝑟𝑖}`$. The ensemble itself creates ( self-reproduces) its own background in a process of the evolution. Quite recently Smolin and Kauffman pointed that self-organized critical phenomena might play a role in the description of quantum gravity. In particular such processes would have allowed the emerging universe to self-tune its fundamental constants, e.g., the cosmological constant.
Interestingly enough,there exists a connection between our ensemble of universes and p-adic topology. In p-adic topology all sets are simultaneously open and closed ( $`\mathrm{𝑐𝑙𝑜𝑝𝑒𝑛}`$), and every point of the set is its center. This means that there is no preferred point which could be considered as a reference point; the only meaningful statements which can be made is about relations between different points. If the introduced ensemble of universes is also simultaneously opened and closed then there is no preferred hyper-spheres in the ensemble: the only thing that counts is their relations which nicely fits into Mach’s principle.
Since there is no preferred ”point” in the ensemble we are left with only one choice for the reference frame, namely the thermal bath prepared by the ”universal observer” ( see above). Within this frame of reference we can define the ground ( reference) energy, an analog of the zero point energy of a quantum oscillator. To this end we assume that all the hyper-spheres in the ensemble have the same radius equal to the Planck length, $`R=\mathrm{\Lambda }`$, that is the minimum attainable length in nature. The respective Planck energy $`1/\mathrm{\Lambda }`$ could be taken as the reference energy.
The latter statement raises the following question. Do all the hyper-spheres of the Planck radius but of different dimensionality have the same value of reference energy? If we assume that the respective energies are different we arrive at a contradiction, since we have chosen the reference point to be the same for any dimension. A choice of different values of the reference energy ( according to the dimensionality) would violate the poly-dimensional covariance (one of the postulates of a new relativity) stating that all the dimensions must be treated on the same footing ( there is no preferred dimension). Hence all the unit hyper-spheres in any dimension have the same reference ( vacuum) energy $``$.
We postulate the invariance of the bulk energy density for hyper-spheres of arbitrary radii $`R_1,R_2,etc`$ in any dimension $`D`$, that is
$$\frac{E_D(R_1)}{C_DR_1^D}=\frac{E_D(R_2)}{C_DR_2^D}=\mathrm{}=\frac{}{C_D\mathrm{\Lambda }^D}.$$
$`(2.1)`$
where $`C_D=R^D\pi ^{D/2}/\mathrm{\Gamma }(\frac{D+2}{2})`$, $`E_D(R_j),j=1,2,\mathrm{}`$ is a radial excitation energy of a hyper-sphere of a radius $`R_j`$. The above postulate is based upon ”incompressiblity” (or volume-preserving diffeomorphism symmetry) that appeared within the context of $`p`$-branes on many occasions. The hyper-spheres are quantized in integers of Planck units. An increase in size of a $`D`$-dim hyper-sphere in a process of absorption of energy ”bits” occurs in such a way as to maintain the energy density equal to the vacuum energy density, $`/C_D\mathrm{\Lambda }^D`$
From eq.(2.1) follows that the energy $`E_D`$ is:
$$E_D=\frac{}{C_D\mathrm{\Lambda }^D}C_DR^D=(\frac{R}{\mathrm{\Lambda }})^D.$$
$`(2.2)`$
Thus we arrive at the following picture: the ground state of the initial ensemble comprises a condensate of hyper-spheres ( at Planck temperature $`T_P`$) of all possible dimensions, where each hyper-sphere has a unit radius (in Planck units) and a constant energy $`_{vac}`$. A pioneering concept of a vacuum state of quantum gravity as a condensate was originally introduced by G.Chapline . A condensation at high temperatures was studied by Rojas et al .
The condensation occurs in such a fashion that the hyper-spheres in the ensemble tends to distribute themselves so as to minimize the energy densities. For example, the spheres in the dimension range $`D=46`$ ( corresponding to the maximum volume of a unit sphere) will have a higher statistical weight than those in the extreme values $`D=2,\mathrm{}`$ of the dimensions corresponding to a zero volume of a unit hypersphere.
Using the principle of a minimum energy density we introduce the gamma distribution as the distribution function for the ensemble of hyper-spheres ( with the Planck radius normalized to one):
$$C_D=\frac{\sqrt{\pi }^D}{\mathrm{\Gamma }(\frac{D+2}{2})}.$$
$`(2.3)`$
Here the values of $`C_D`$ are the statistical weights of the distribution. To account for thermal effects due to a decrease of the average temperature , from the Planck temperature to about 3 Kelvins we include into distribution (2.3) the energy dependence by using the Bose-Einstein distribution based on the hyper-sphere energy $`E_D`$ and its temperature $`T`$. Therefore the distribution density $`\rho `$ will take the following form
$$\rho (D,E)=C_D/[e^{\frac{E_D}{KT}}1]=C_D/[e^{\frac{(R/\mathrm{\Lambda })^D}{KT}}1]$$
$`(2.4)`$
where we use eq.(2.2).
At the beginning of this section we postulated that the introduced ensemble contains hyper-spheres of infinite dimensions. Now we prove this statement ( at least for integer dimensions) with the help of the Fermat’s last theorem. Upon collision two hyper-spheres, $`A`$ and $`B`$ can produce a third hyper-sphere $`C`$ analogous either to a final product of a chemical reaction or to a 3-point vertex in string field theory. Let us assume for the moment that all the hyper-spheres have the same dimension $`D`$. The above interaction then conserves energy. Taking into account additivity of the energy we get
$$E_A+E_B=E_C(n_A)^D+(n_B)^D=(n_C)^D.$$
$`(2.5)`$
where, after a cancellation of a common factor $``$ we get the energies expressed as integers ( quanta) $`n_A,n_B,n_C`$.
According to the Fermat’s last theorem, eq.(2.5) has no solution in nonnegative integers for integer dimensions greater than $`2`$. Since we live in the universe of the average dimension $`D>2`$ the Fermat last theorem requires an energy balance different from (2.5) :
$$(n_A)^{D_A}+(n_B)^{D_B}=(n_C)^{D_C}.$$
$`(2.6)`$
where the dimensionalities $`D_A,D_B,D_C`$ cannot be all equal. Hence we have arrived at one of the most important results of our work. According to the Fermat last theorem the equilibrium ( or quasi-equilibrium ) state of a thermalization process can be attained only if the dimensions of the colliding hyper-spheres must change in the process. Since dimension is a topological invariant, such a collision represents a simple example of a topology changing process while geometry is restricted to spherical geometry.<sup>1</sup><sup>1</sup>1As an aside note we should mention that alternative geometries , for example hyperbolic geometries, such as the upper complex plane, de Sitter,and anti de Sitter spaces could be studied using a different framework. The problem with the respective topologies is that they comprised open and noncompact spaces. One could compactified them by attaching the projective boundaries at infinity. This would be topologically equivalent (in the anti de Sitter case) to spherical geometry $`S^n`$ thus reducing the problem to the previous case. We also would like to point out that a phase transition might not only change topologies but also transform one geometry into another.
Let us consider an example of a simple dimension-changing process: $`(2)^1+(5)^2=(3)^3`$, that is $`A(n_A=2;D=1)B(n_B=5;D=2)C(n_C=3;D=3).`$ Here 2 quanta (geometric bits) plus 25 quanta ( geometric bits) yield 27 quanta ( geometric bits). Since energy has been conserved in this process so does the information. The inverse process is also possible. State $`C`$ can decay into a sum $`A+B`$. This would look like some dimensional ” compactification ” process. A $`D=3`$ sphere of radius $`3`$ in Planck units has ”compactified ” into a sphere of $`D=1`$, of radius $`2`$ in Planck units, and another sphere of $`D=2`$, of radius $`5`$ in Planck units. Using the analogous arguments we can consider more complicated collisions , like $`A+BC+D`$.
Since the dimensions’ fluctuations are not necessarily small one has to abandon a perturbative expansion used in string theory. For example, the perturbative string theory could not even reproduce a dimension change from $`D=10`$ to $`D=11`$, let alone handle infinity of dimensions. On the other hand, by using all the dimensions ( including infinite ones) on equal footing and employing the Fermat last theorem we bypass the need to use perturbative methods and an unpleasant task of summation over all the topologies. The thermalization process of collisions (described by the Fermat theorem) automatically performs summation over different topologies. Thus, for example, $`S^2`$ does not have the same topology as $`S^4`$ simply because the respective dimensions are different, the fact taken into account by eq.(2.6).
## 3 Average Dimensions and Cantorian-Fractal Spacetime
### 3.1 Calculation of Average Dimensions
Using the distribution density $`\rho `$ for the ensemble of hyper-spheres we get the average dimension of the ensemble as follows :
$$<D>=\frac{_d^{\mathrm{}}𝑑D𝑑ED\rho (D,E)}{_d^{\mathrm{}}𝑑D𝑑E\rho (D,E)}.$$
$`(3.1)`$
The lower limit of integration $`d`$ must be found on the physical grounds. In a new relativity dimensions ( as everything else) are defined in relation to a reference point. From the dependence of either a volume or an area on dimensions follows that for any finite radius both the volume and the area tend to 0 at $`D_0=2`$. We take this value as the reference point. Another limiting value of $`D`$ resulting in zero volume and area is $`D\mathrm{}`$. These considerations determine the limits of integration in (3.1).
Since a hyper- sphere energy is given by (2.2) we immediately obtain
$$dE=D(\frac{R}{\mathrm{\Lambda }})^{D1}d(\frac{R}{\mathrm{\Lambda }}).$$
$`(3.2)`$
where $`\mathrm{\Lambda }`$ (in units of $`\mathrm{}=c=1`$) can be represented in terms of Newton’s gravitational constant $`G_D,G_{D1},G_{D2},\mathrm{}`$ in $`D,D1,D2,\mathrm{}`$ dimensions.
$$\mathrm{\Lambda }=G_D^{\frac{1}{D2}}=G_{D1}^{\frac{1}{D3}}=\mathrm{}$$
$`(3.3)`$
In $`D=4`$ the Planck scale is the familiar $`10^{33}cms`$. When $`D=2`$ the Einstein-Hilbert action is a topological invariant and (3.2) results in singularity. We can avoid this by choosing $`G_2=1`$. This means that we can also choose the value of the universal scale to be $`1`$ in the units of $`\mathrm{}=c=G_2=1`$.
Upon substitution of (2.4) in (3.1) we get
$$<D>=\frac{_d^{\mathrm{}}𝑑DD^2\sqrt{\pi }^D[\mathrm{\Gamma }(\frac{D+2}{2})]^1_1^{(R_H/\mathrm{\Lambda })}d(R/\mathrm{\Lambda })(R/\mathrm{\Lambda })^{D1}[e^{\frac{(R/\mathrm{\Lambda })^D}{KT}}1]^1}{_d^{\mathrm{}}𝑑DD\sqrt{\pi }^D[\mathrm{\Gamma }(\frac{D+2}{2})]^1_1^{(R_H/\mathrm{\Lambda })}d(R/\mathrm{\Lambda })(R/\mathrm{\Lambda })^{D1}[e^{\frac{(R/\mathrm{\Lambda })^D}{KT}}1]^1}.$$
$`(3.4)`$
Evaluating the integral over$`R/\mathrm{\Lambda }`$ we easily obtain :
$$<D>=\frac{_d^{\mathrm{}}𝑑DD\sqrt{\pi }^D[\mathrm{\Gamma }(\frac{D+2}{2})]^1F(a,b^D)}{_d^{\mathrm{}}𝑑D\sqrt{\pi }^D[\mathrm{\Gamma }(\frac{D+2}{2})]^1F(a,b^D)}.$$
$`(3.5)`$
where $`R_H`$ is the Hubble radius whose value we do not fix a priori and thus leave for the time being arbitrary.
$$a=\frac{}{KT},b=\frac{R_H}{\mathrm{\Lambda }},F(a,b^D)=a(b^D1)[\frac{e^a1}{e^{ab^D}1}].$$
$`(3.6)`$
The average dimension given by (3.5) cannot be expressed in closed form. Therefore we will evaluate it numerically. Still we can extract some information directly from (3.5) before resorting to numerical integration. In fact, we can distinguish $`3`$ easily identifiable regions in terms of the parameters $`a`$ and $`b`$:
(i) $`a=1`$. This means that the thermal factor $`F(a,b^D)0`$ and thus does not influence the resulting value of $`<D>`$ leaving it as if it were no thermal factor.
(ii)$`a>>1,b>>1`$. In this case the thermal factor $`F(a,b^D)`$ once again tends to zero and will be factored out of the expression for $`<D>`$. The latter remains the same as without the thermal factor, with the only difference that the final temperature $`T`$ will be much lower than the Planck temperature $`T_P`$ corresponding to the vacuum energy $``$.
(iii) $`a>1`$, $`b>1`$. In this case we cannot factor out the thermal factor $`F(a,b^D)`$, and as a result it contributes significantly to the value of $`<D>`$. Numerical simulations show that in this case the fluctuations in $`R_H/\mathrm{\Lambda }`$ and $`T`$ will initially lead to an increase of $`<D>`$ and then ,after reaching a maximum value, to its monotonic decrease approaching asymptotically the metastable fixed point $`<D>=4+\varphi ^3=4.236\mathrm{}`$ of the renormalization group. This point corresponds to our world which emerged from chaos ( a multitude of dimensions) as soon as the ensemble of the hyper-spheres began evolving (generating the respective average temperature and dimensions) towards the metastable fixed point of the renormalization group
As a next step we will provide the numerical value of $`<D>`$ and demonstrate an importance ( and ubiquitousness) of the golden mean $`\varphi `$ in evaluations of the average dimensions of the world. In particular, it will help us to understand why we live in four dimensions, why their signature is $`3+1`$, and why there might exist a deep reason of chiral symmetry breaking in nature.
### 3.2 The Cantorian number $`<D>=4+\varphi ^3=4.236067977\mathrm{}`$ as the Exact Average Limiting Dimension
We have already mentioned that a new relativity requires that dimensions should be calculated with respect to the reference point ( zero point dimension)$`D_0`$ , that is any dimension $`D`$ is in fact $`D^{}=DD_0`$. In the previous section we have already chosen this reference point $`D_0=2`$ Introducing $`<D^{}>`$ into eq.(3.5) we get
$$<D^{}(t)>=\frac{<D^{}F(a(t),b(t)^D^{})>}{<F(a(t),b(t)^D^{})>}<D^{}>_{lower}.$$
$`(3.7)`$
Here $`<D^{}>_{lower}`$ is the average dimension without quantum dissipative effects (responsible for the initial chaotic phase). The latter will increase the average dimension as compared to $`<D^{}>_{lower}`$ and prevent the expanding hyper-spheres from cooling down below the Planck temperature $`T_P`$. As soon as the average dimension reaches a peak value, the continuing expansion of the hyper-spheres will gain the ”upper hand” over the re-heating due to quantum-dissipative effects. As a result the hyper-spheres will begin to cool, and the average dimension will begin to decrease towards its initial value. This will signal an onset of the ” ordered ” phase emerging from the initial chaotic phase (quantum dissipation ). It is possible to demonstrate that the average dimension cannot fall below the metastable fixed point $`4+\varphi ^3`$ ( metastable vacuum) of renormalization group until the system will reach the phase transition point and move towards the quasi-crystal phase ( the true vacuum ).
Evaluating expression (3.5) for cases (i) and (ii)we easily find that
$$<D^{}>=6.3876=<DD_o><D>=4.38764+\varphi ^2=4.3819\mathrm{}..$$
$`(3.8)`$
This value represents the upper value of $`<D>`$. The respective lower value can be evaluated as follows. Assuming that the dimensions are integers we replace in (3.5) the integrals by the sums and obtain
$$<D_n>=4.3267$$
$`(3.9)`$
which is smaller ( as expected ) than the value given by (3.8) , that is
$$4.38764+\varphi ^2=4.3819\mathrm{}$$
$`(3.10)`$
The values of $`<D>`$ based on a discrete (3.9) and integral(3.8) averages are in a very good agreement with El Naschie results obtained with the help of the transfinite $`_{\mathrm{}}`$Cantorian fractal spacetime models in ,. These models give the following exact value of the average dimension :
$$<D>=<dim_F^{(\mathrm{})}>=\frac{1+\varphi }{1\varphi }=\frac{1}{\varphi ^3}=4+\varphi ^3=4.236\mathrm{}$$
$`(3.11)`$
where $`\varphi =\frac{\sqrt{5}1}{2}=0.618\mathrm{}<ln2/ln3`$ and $`ln2/ln3`$ is the dimension of the middle segment Cantor set .
The golden mean $`\varphi `$ provides us with the basic ”unit”, or the elementary dimension building block from which all the sets of the transfinite Cantorian fractal spacetime models are constructed. These spaces are the densest spaces obtained from infinite intersections of infinite unions. The value of $`d_c^{(0)}=\varphi =0.618033984\mathrm{}.`$ is the fractal dimension of a physical structure living in a $`0`$ topological dimension. This can be interpreted as packing, compressing ( information, for example ) the fractal dust points of dimensionality equal to $`\varphi `$ into a ” point ”. We put quotation marks around ” point ” to emphasize that since there exists the minimum attainable (non-zero) scale in nature ( Planck scale) there are no points in nature. A geometrical point is in fact smeared out into a ” fuzzy ” ball/sets of all possible topological dimensions, from $`\mathrm{}`$ to $`\mathrm{}`$. For this reason the naive concept of a fixed and well defined dimension cannot be used anymore.
The set of $`\mathrm{}`$ dimensions $`^{(\mathrm{})}`$ is called the void virtual set and the set of $`\mathrm{}`$ dimensions $`^{(\mathrm{})}`$ is called the universal set. The void set has in fact the fractal dimension $`0`$, i.e the void set comprises one single fractal dust point. The universal set,on the contrary, has an uncountable infinity of fractal dust points which means that its fractal Cantorian dimension is $`\mathrm{}`$. Thus the respective dimensions are dual to each other : $`d_c^{(\mathrm{})}=\mathrm{}`$, $`d_c^{(\mathrm{})}=0=1/d_c`$.
The virtual sets of negative topological dimensions can be endowed with a physical meaning if we would view them as carriers of negative information entropy, or anti-entropy . Such an interpretation is very close to Dirac’s sea of negative energy where we substitute negative entropy for the negative energy. The true vacuum of the theory requires the sea of negative topological dimensions located below the topological value of a zero dimension $`D_0=2`$ to be filled up. This is necessary for keeping the quantum universe from ”cascading” down and disappearing in a void set $`^{\mathrm{}}`$.
Thus we arrive at the following result: the average dimensions obtained with the help of the Cantorian fractal spacetime model, discrete sums average, and the integral average respectively satisfy the following relation :
$$4+\varphi ^3=4.236\mathrm{}<4.3267(discrete)<4.3876(integral)4+\varphi ^2.$$
$`(3.12)`$
The difference between various averages is surprisingly small, taking into account that our model uses smooth spheres and that $`4`$+$`\varphi ^3`$ is the exact result obtained by El Nashie on the basis of the densest packing allowed and which generalizes the Maudlin-Williams golden mean theorem. The spaces are filled-in totally with ” fractal dust points ”. Another values for the average dimension $`<D>=2/ln\varphi =4.156\mathrm{}`$ (based on the knot theory and Jones polynomial ,) and $`<D>=4.074.09`$ (based on the Leech lattice packing) are only an approximation to the exact value of $`<D>`$=$`4`$+$`\varphi ^3`$. This is due to the fact that these models used packing different form the densest packing.
We believe that connection of the exact average dimension of the world to the golden mean is not a simple numerical coincidence. In fact, this value $`<dim^{\mathrm{}}>=d_c^{(4)}`$ = $`4`$+$`\varphi ^3`$ indicates the onset of quasi-ergodicity ,, that is the topological dimension is of the same order as the Fractal dimension . $`D_T>D_F`$ represents a stable region and $`D_T<D_F`$ represent ergodic case. We are living in a metastable quasi-ergodic state or ” vacuum ”, $`<D>`$ =$`4`$+$`\varphi ^3`$. On the other hand, the true vacuum, as was argued earlier (in connection with $`^{(\mathrm{})}`$ and $`^{\mathrm{}}`$) corresponds to the dual inverse fractal dimension of $`d_c^{(4)}`$ equal to $`d_c^{(2)}=1/(4+\varphi ^3)=\varphi ^3`$; or $`2+\varphi ^3`$ relative to the zero point $`D_0=2`$.
Still the major question remains unanswered: why do we live in a $`3+1`$ dimensions? To answer we have to consider the long range coarse-grain averaging process of a fractal geometry underlying the perceived spacetime of our existence.
Such an averaging process results in a breakdown of poly-dimensional isotropy. This means that prior to the break-down there are (on average) four orthogonal dimensions which separately can deviate from their respective values by the same amount $`(ϵ/4)=(\varphi ^3/4)`$ . As the result the emerging average dimension is
$$4(1+\frac{\varphi ^3}{4})=4+\varphi ^3=4.236\mathrm{}=<dim_F^{(\mathrm{})}>=dim^{(4)}=d_c^{(4)}.$$
$`(3.13)`$
After the breakdown of isotropy $`3+ϵ=3(1+ϵ/3)`$ spatial dimensions plus $`1+ϵ`$ temporal dimensions emerge. Here the fluctuations of the temporal dimension is 3 times greater then the fluctuation $`ϵ/3`$ of each of the spatial dimensions.The long-range coarse-grain averaging process could be viewed as a process of ”projecting ” El Naschie’s effective average $`4`$+$`\varphi ^3`$-dimensions onto the four-dimensional $`\mathrm{𝑠𝑚𝑜𝑜𝑡ℎ}`$ outer surface, which we perceive as our reality. As a result of the projection the initially orthogonal spatial and temporal dimensions will become entangled. A good analogy could be a two-dimensional view of a three dimensional knot.
This could explain a perception of the three-spatial dimensions, at any given fixed moment of clock time. One can slice ( at least locally) the four-dimensional spacetime into $`3`$-dimensional sheets at any given fixed value of the clock time resulting in a possibility of measurements in extended space at a fixed moment of time . Inversely, it is difficult ( if not impossible) for a macroscopic observer to perceive an extended interval of time, within an extended region of $`3`$-space. Paraphrasing Wheeler <sup>2</sup><sup>2</sup>2”Time exists to prevent all things from happening at once”, we can say, ” space exists to prevent everything from happening one point ”.
On the other hand, in our model of infinite dimensions at the scales approaching Planck scale particles can move forward and backward in time simply because they have access to more dimensions than we do. In other words, particles are fully ”aware” of the fractality of spacetime. For example, they do not ”see” the two slits in the two-slit experiment as two separate points since from their perspective spacetime is a discrete, fractal-like Cantor set. When one constructs the Cantor set by removing the middle segment $`adinfinitum`$ , one is literally ” removing ” the space between the fractal dust points thus eliminating a separation between the slits. Since at this scale there is no separation between the fractal dust points, the Cantor set has zero measure = zero length. It is our mind which fills the void creating the illusion that there is a separation between the points of a fractal dust. This leads to the apparent ” non-locality ” and other paradoxes of quantum mechanics. Nature does not abhor the vacuum , it is our mind which does so by filling in the voids. In fact, nature is not only quantum and discrete, it is fractal.
Thus the entanglement and long range averaging of the perturbed spatial and temporal dimensions yield the observed $`4`$-dimensional world. This means that within the framework of Cantorian fractal spacetime we have the following entangled dimensions yielding the perceived $`4`$ dimensions
$$(3+ϵ)(1+ϵ)=44+ϵ=\frac{1}{ϵ}ϵ=\varphi ^3.$$
$`(3.14)`$
As a result of this process the golden mean appears once again as an essential parameter of reality. On a surface this looks like a forced fit to the preconceived notion. And really, why does not $`ϵ=\varphi ^3`$ appear as a result of some other splitting, say $`(5+ϵ_1)(1+ϵ_2)=6`$? . It is obvious that there are infinite number of solutions $`ϵ_1,ϵ_2`$ satisfying suac splitting. However, to require that dimensional perturbation $`ϵ_1=ϵ_2=\varphi ^3`$ must be the only solution indicating the emergence of $`3`$-dimensional entangled imposes a very stringent constraint. We must mention that the condition $`ϵ_1=ϵ_2=\varphi ^3`$ satisfies the symmetric split $`(2+ϵ)(2+ϵ)=5`$ between $`2`$ spatial, $`2`$ temporal, and the universal fluctuation $`ϵ`$. Is all this just a numerical coincidence or design ?
We can explain this situation from the lattice point of view. If we consider a quasi-ergodic self-similar fractal grid/lattice, or topologically equivalent fractal sphere of dimensionality $`4`$+$`\varphi ^3`$ the coarse-grain averaging process will break down the ”translational ” symmetry from
$$𝒯[4+\varphi ^3]𝒯[4]=T[(3+ϵ)(1+ϵ)]whereϵ=\varphi ^3.$$
$`(3.15)`$
It transpires that our perception ( which by the very definition amounts to averaging) of $`3`$ dimensions is in fact the long range coarse-grain effect of co-existence of $`3+\varphi ^3`$ (spatial ) and $`1+\varphi ^3`$ (temporal ) dimensions. The set $`^{(4)}`$ whose dimension is $`4`$+$`\varphi ^3`$ is ”packed” inside a four-dimensional sphere. This reminds a picture of the grains of sand on a beach. The sand ”looks” two-dimensional to us when in fact it is $`3`$-dimensional. Due to a coarse-graining effect we as macroscopic observers can only perceive the ”skin ”, or a $`3`$-dimensional outer surface (where space and time co-exist) of the ”onion”-like world. In other words, we perceive only projections onto the ” four-dimensional external surface ” of the average $`4`$+$`\varphi ^3`$ dimensions of a truly infinite-dimensional world lying figuratively speaking underneath our ”feet” and above ”our heads”. The four-dimensional ”skin” encloses the exact average value of $`<D>`$=$`4`$+$`\varphi ^3`$ where averaging is performed over all possible topological dimensions (ranging from $`\mathrm{}`$ to $`\mathrm{}`$) of the sets $`^{(i)}`$ of Cantorian fractal spacetime. A similar mechanism may be responsible for the chiral-symmetry breakdown in Nature without supersymmetry.
If there will be enough energy to move ”inside” the ”skin” into the transfinite continuum of Cantorian spacetime we will be able to discover the true fractal average dimension of spacetime, at each layer(ladder) of the renormalization group flow. From the renormalization group treatment of quantum field theory one can infer that the universe, upon reaching the metastable fixed point $`4+\varphi ^3`$ of the renormalization group will signal the beginning of a dimensional phase transition, from $`4`$+$`\varphi ^3`$ to $`\varphi ^3`$. At this fixed point a conformal invariance exists which means that the world will become self-similar at every scale.
From this perspective it seems that theories of gravity and all other fundamental forces in nature are effective theories since they emerge from a deeper theory , namely a new relativity in Cantorian fractal spacetime. In fact, vacuum fluctuations in Cantorian fractal spacetime generate the long range Einstein’s gravity theory and all other fundamental forces ”residing” inside spacetime. It has turned out that a new relativity allows one to write down the quantum master action for the ”master” field residing in C-space, ( outside spacetime) whose vacuum fluctuations generate classical spacetime, gravity and all other fundamental forces. We believe, this is a reason why attempts to quantize Einstein’s gravity have been futile so far. It looks as if Cantorian fractal geometry is the bridge connecting both worlds, quantum and classical. Therefore noncommutative geometry together with the renormalization group approach are mathematical tools allowing us to probe this Cantorian fractal world .
Thus if general relativity required four-dimensional Riemannian geometry for its formulation a new relativity requires the transfinite continuum of the noncommutative von Neumann’s Cantorian fractal spacetime $`^{(\mathrm{})}`$ -. What we perceive as a smooth four-dimensional topological space ( a sphere, for example ) is an illusion. There are no ” points ” in this New Relativity, due to the fact that the Planck scale is the minimum distance in Nature. As we zoom in using a ”microscope” of the renormalization group, we uncover that each point is also a four-sphere, and that each point within that sphere is also four-dimensional, and so on and so forth…$`adinfinitum`$.
## 4 A Solution to the Cosmological Constant Problem
Here we will present a solution to the cosmological ” constant” problem which parallels Nottale’s derivation based on his scale relativity as well as El Naschie’s $`^{\mathrm{}}`$ Cantorian spacetime theory. Depending either on the values of the ”constants” $`a,b`$ of the previous section or initial conditions, one will have several different cosmological scenarios. We write ”constants” because in reality they have an explicit scaling ” temporal ” dependence consistent with the renormalization group. The parameters $`a,b`$ will change with scaling time, $`t`$. We set the initial scaling time to be $`t_0=1`$ , since $`ln(t_0)=0`$ and logarithmic dependencies play an important part in scale relativity.
Using (3.6) and the scale relativity we get
$$a(t)=(\frac{E_0}{KT})(t)=(\frac{}{KT_P})^{(t/t_o)^\alpha }.\frac{log[E_o/K]}{log[/KT_P]}=(t/t_o)^\alpha .$$
$$b(t)=(\frac{R}{\mathrm{\Lambda }})(t)=(\frac{}{\mathrm{\Lambda }})^{(t/t_o)^\beta }.\frac{log[R/\mathrm{\Lambda }]}{log[/\mathrm{\Lambda }]}=(t/t_o)^\beta $$
$`(4.1))`$
Since we introduce the time dependence of the parameters $`a`$ and $`b`$ we have two different values of the vacuum energy: $`E_0`$ denotes the vacuum energy at a given time $`t`$ and $``$ denotes the initial vacuum energy. The average value of the dimension is respectively:
$$\frac{<D(t)>}{<D(t_o)>}=(\frac{t}{t_o})^\delta .\frac{ln<D(t)>}{lnt}=\delta .$$
$`(4.2)`$
where $`\alpha `$ and $`\beta `$ are the constants entering Nottale’s scale relativity , and $`\delta `$ is the universal scaling exponent. Equation (4.2) mean that the average dimension changes monotonically with scaling time. On the other hand, eq.(4.1) postulates a power-scaling law for the ratios of the fundamental constants in nature. According to Nottale’s scale relativity the scaling coefficient themselves are resolution-dependent. However for the simplicity sake we assume that they are universal constants.
In Appendix B it is shown that the slope $`dS/dA`$ of entropy vs. area diverges between the values $`D=4`$ and $`D=5`$ . This represents an onset of a dimensional phase transition, that is the universe has reached $`<D>=4+\varphi ^3`$ and begins its transition to the quasi-crystal phase characterized by $`<D>=\varphi ^3`$ (or to $`2+\varphi ^3`$ if we take $`2`$ as zero reference point). To study this process in more details one needs to consider Nottale’s scale relativistic resolution-dependent coefficients $`\alpha `$ and $`\beta `$. At the critical point these three coefficients are connected by a certain relation which is typical for critical phenomena.
For example if initially, at the Planck era the vacuum energy $``$ was of the same order of magnitude as the thermal Planck energy $`KT_P`$ then generally speaking its value will subsequently change (with the renormalization group flow) to a much smaller value $`E_o`$ of today. However $``$ still will be large in comparison with today’s background thermal energy of $`2.763`$ Kelvins. We would like to reiterate that in a new relativity it makes sense to speak only about relations between quantities ( and by implication about the respective ratios). In this sense a concept of a very small is intrinsically connected with a concept of a very large.
The quantities $`,`$ are respectively the vacuum energies and sizes of the original baby universes at the ” initial” scaling time of $`t_0=1`$, at the beginning of the ”count down”. This means that the scaling ”clock ” starts ticking at the moment when the radius of the hyper-sphere is $``$ and the value of the vacuum energy is $``$. After that moment infinitesimal perturbations of both, the radius and energy will begin to grow. This results in the following physical picture.
Initially the vacuum fluctuations about the perfect balance conditions $`(R/\mathrm{\Lambda })`$ = $`E/KT_P`$= $`1`$ were infinitesimally small. However, because the system has an infinite degrees of freedom these fluctuations could not have been damped, and this would have lead to a disruption of equilibrium. Inversely, had the number of degrees of freedom be finite the infinitesimal perturbations would have been damped and equilibrium would have been restored.
Thus, as a result of infinite number of degrees of freedom the universe was driven out of balance, out of the metastable state of average dimension $`4`$+$`\varphi ^3`$, to states of higher and higher average dimensions, until it reached a maximum average dimension which depends on initial perturbations and the non-linear dynamics of the system.After reaching the maximum average dimension , the universe began its ”descent” from this maximum to the initial value of $`D`$ = $`4`$ \+ $`\varphi ^3`$.
Since the initial perturbations drove the average dimension to a value higher than $`4`$ \+ $`\varphi ^3`$ one might be inclined to think that the average temperature will also surpass the Planck temperature $`T_P`$.However this is a wrong conclusion since according to a new relativity the effective values of the Boltzamann constant $`K_{eff}(E)`$ and Planck constant $`\mathrm{}_{eff}(E)`$ are energy dependent ( see e.g, Appendix A). Both constants increase in such a way as to keep the temperature below $`T_P`$. For example, a very energetic photon of frequency $`\omega `$ being emitted by the walls of the reservoir cannot exceed the upper temperature bound of $`(\mathrm{}_{eff}\omega /K_{eff})=T_P`$.
As we discussed earlier, during the initial chaotic phase the quantum-dissipative effects re-heat the expanding universe in such a way as to keep $`T_P`$ constant. After the universe had reached its maximum average dimension, it began to ”roll down”, in both dimensions and temperature: the ordered-phase began. The renormalization group local scaling arrow of time also appeared at the moment when the universe started its descent back to the metastable point. Prior to that there was no arrow of time since when the universe was in perfect balance nothing changed, nothing ”happened ”. Spacetime as we know it did not exist and emerged only afterwards. The universe began due to a non-equilibrium process of self-organization as was argued long ago by Prigogine.
As the Universe moves from higher average dimensions and higher vacuum energies , to lower dimensions and lower vacuum energies, information is lost. Since the total entropy of the universe plus reservoir cannot decrease, this means that entropy flowed out of the universe into the reservoir which comprises the infinite family of hyper-spheres of all possible dimensions and radii. The quanta of spacetime create their own background in which all the $`p`$-branes live in ; this background is self-referential and self-supporting.
Keeping in line with the second law of thermodynamics, the information entropy which flows out of a universe into the self-referential thermal background is recycled over and over again due to the bootstrap principle , i.e., the quanta/hyper-spheres are made of each other. This recycling process of information-energy-dimensions bootstrapping, the creation of the fundamental particles in the universe and its life forms occurs in a hierarchical multi-fractal fashion. Accordingly self-organization proceeds in discrete jumps from a smaller scale to a larger scale, to a larger scale, etc. Therefore we subscribe to Penrose’s view that consciousness is a non-algorithmic process which is compatible with our idea that an uncountable infinity of dimensions implies existence of an uncountable infinity of information leading to the emergence of conscious life.
As the emerging local scaling time begins its ”count” the values of the ”universal constants ” begin change with ”time”. This process is consistent with a modern theory of variable speed of light cosmologies which gains a certain support. Choosing a simple situation characterized by $`\alpha =\beta `$ and $`/KT_P=/\mathrm{\Lambda }`$ we get from (5.1)
$$\frac{(E_o/KT)}{(R/\mathrm{\Lambda })}(t)=[\frac{/KT_P}{/\mathrm{\Lambda }}]^{(t/t_o)^\alpha }=1.$$
$`(4.3)`$
From Eq.(4.3) follows that in this simple case for all values of renormalization group scaling time the following is true
$$(\frac{R}{\mathrm{\Lambda }})(t)=\frac{E_o}{KT}(t).$$
$`(4.4)`$
Taking, for example, $`R_{Hubble}=10^{60}\mathrm{\Lambda }`$ we find from (4.4) that according to this simple scenario the present-day value of the vacuum energy should be
$$E_o(today)=(kT_H)(\frac{R_H}{\mathrm{\Lambda }})(today)=10^{60}(kT_H).$$
$`(4.5)`$
This result means that the present vacuum energy would be huge as compared to the thermal energy $`T_H`$. This might serve as a very straightforward explanation of the fact that today’s universe is expanding much more rapidly than was predicted by the existing theories. The explanation follows as a natural result of a new relativity theory and Cantorian fractal spacetime which in turn might be viewed as an indirect proof that the respective world view may be correct, or at least points in the right direction.
Let us look at the famous experiments $`COBE`$. The $`T_{COBE}`$ value of $`2.763`$ Kelvins was obtained by ”looking” into the past of the universe. The photons we detect today were emitted in the past and were redshifted due to the expansion of the universe. Since the speed of light in a new relativity changes with the renormalization group flow, calculation of the exact values of the redshifts is going to be a rather difficult task. The correct way to do this would be to carry out integration along the renormalization group trajectory backwards in scaling time (cf. use of convective derivatives in fluid mechanics) to find out the true frequency of the photons upon emission.
In the widely accepted theories ”past” refers to the Big Bang. On the other hand, there was no Big Bang in our cosmological scenario. Our model implies an ever expanding Universe where expansion begins from the moment when the infinitesimal perturbations drove the Universe out of balance. By expansion we mean a synchronous change of universe’s radius with the renormalization group flow representing the true arrow of time.<sup>3</sup><sup>3</sup>3Here we must add a word of caution concerning the use of the true ” local scaling ” renormalization group flow arrow of time. It should not be confused with the conventional coordinate time. H. Kitada using Godel’s incompleteness theorem has proven that a local time exists even in the absence of global time in traditional quantum cosmology.
Since the average dimension of the universe changes in the process it leads to a change of the respective volume even for a fixed radius. However to take the volume of the hyper-spheres as an indicator of temporal evolution is not a good idea. The radius is the more appropriate indicator of change. Quite analogously to the radius being an indicator of the renormalization group flow, the COBE data’s temperature $`T_{COBE}`$ may also serve as an appropriate ” thermometer ” of the evolving ensemble of an infinite number of bubbles/universes. For this reason, in this model, one could set $`T_H=T_{COBE}`$ and conclude that within the model the vacuum energy today would be of the order $`10^{60}KT_{COBE}`$. Another way to look at the COBE data is to look for deviations from the $`2.76`$ Kelvin due to a greater redshifts caused by a greater expansion of a cooling universe. Recent experimental work by De Bernardis et al and the recent BOOMERANG data indicates that this is the case .
The renormalization group scaling time changes rather slowly as compared to the coordinate time of our clocks since the former is logarithmic and the latter is roughly linear. In turn the fundamental constants of nature change with respect to the renormalization group scaling time, and for this reason they change very slowly as compared to our daily experience. For example, the speed of light at the time of relativity formulation was essentially the same as today. However, this was not the case when the universe began its evolution at the Planck era. At that period the fundamental constants changed more rapidly. We are living now in a metastable phase( a slow, ”predictable”) and for this reason life was possible at this stage of evolution.
It would be hard( if not impossible) for life forms to emerge during a phase where the fundamental constants changed rapidly relative to the renormalization group scaling time flow. Such a rate of change corresponds to the situation where the information-entropy is no longer a linear function of the area which in turn is a result of taking into account the infinitely-dimensional picture of the universe ( Appendix B). As for ourselves, we are living in a linear world where information-entropy transfer per unit area is constant. This is another reason why $`D=4,5`$ are the optimum dimensions conducive for the appearance of humans, which is consistent with the average dimension $`D=4`$.
Let us look at the cosmological constant $`\lambda `$ in today’s $`4`$-dimensional world. It has units of a $`(length)^2`$ because it must have the dimensions of curvature appearing in Einstein’s equations. The units of $`\lambda `$ are also the same as the units of a string tension, namely energy per unit length. Let us now compare today’s $`\lambda `$ with the cosmological ”constant” at the moment when the scaling renormalization group time was launched (at the Planck era) by simply evaluating the ratios of energy/length. One can argue that at the very beginning the large scale structure of spacetime ( as we know it ) did not even exist. Spacetime was barely born at that moment . Therefore Einstein’s equations could not be applied and this period is a truly quantum gravity period. Still we can compute the relevant numbers, at least crudely.
Setting the vacuum energies to be of the same orders of magnitude as their Compton energies , $`E_{today}\frac{1}{R_H}`$ and $`\frac{1}{\mathrm{\Lambda }}`$ and using eqs.(4.3-4.5) we arrive at the following ratios of the cosmological ”constants” :
$$\frac{E_{min}/}{/\mathrm{\Lambda }}=(\frac{\mathrm{\Lambda }}{})^2=\frac{T_o}{T_P}.$$
$`(4.6)`$
Let us assume for the sake of the argument that one could take the upper Nottale scale to be $``$ $`10^{61}\mathrm{\Lambda }`$ $`>`$ $`10^{60}\mathrm{\Lambda }`$. Using this assumption in (4.6) we obtain the following ratio of the cosmological ” constants ” :
$$(\frac{\mathrm{\Lambda }}{})^2=10^{122}=\frac{T_o}{T_P}.$$
$`(4.7)`$
This result implies that the cosmological ” constant ” (when the world was of the Planck radius - a mini-black hole ) was 122 orders of magnitude larger than the cosmological ” constant ” of the universe having the maximum radius of $``$. Once again we wish to emphasize that we are referring to the radius of the universe since in general dimensions change and therefore it is wrong to compare volumes at different epochs.
At another limit the universe would move to an extremely cold world of minimum $`T_o`$ which is close to zero but not zero. In fact if the minimum $`T=0`$ this would contradict the duality of nature requiring that maximum $`T_P`$ must be dual to a minimum $`0`$ temperature since in this case $`1/0=\mathrm{}`$. The state corresponding to $`T_o`$ is the noncommutative quasi-crystal phase of the coldest world whose dimension $`d_c^{(2)}=\varphi ^3=0.236..`$. This means that as the universe reaches the renormalization group metastable point of $`D=4.236\mathrm{}`$ it would begin to move very slowly towards another phase transition and not less slowly proceed to a very cold world whose lowest temperature $`T_o`$ will be 122 orders of magnitude smaller than the Planck temperature $`T_P`$ that is $`T_o10^{32}.10^{122}=10^{90}`$ Kelvin.
At a first glance it seems that the assumption about inverse proportionality of vacuum energy and the scale (that is $`E1/R`$) contradicts eq.(4.4) which states that the energy is proportional to the scale. However after performing the duality transformation $`T_oT_P,\mathrm{\Lambda }`$ in eq.(4.4) and leaving $`_{min}`$ unchanged we obtain
$$\frac{_{min}}{KT_P}=\frac{\mathrm{\Lambda }}{}.$$
$`(4.8)`$
It is immediately seen that eq.(4.8) is in full agreement with the statement that $`_{min}1/`$. Moreover, combining eq.(4.4) and (4.8) we arrive at eq.(4.7). Interestingly enough, if we apply the duality transformation to a bubble of Planck radius, and Planck temperature $`T_P`$, we find that the Planck scale is ” self dual ” in the sense that the Compton wavelength associated with the Planck’s mass $`M_P`$ is in excellent agreement with the value of its Schwarzchild radius , $`RM_P`$. But this is precisely how one defines the Planck scale from the very beginning.
As we have already indicated, the universe will keep expanding until it will asymptotically approach the upper Nottale scale $``$, reaching the final quasi-crystal phase. The scale relativistic corrections prohibit scales exceeding $``$. This restriction serves as the natural infrared regulator which is analogous to viewing the Planck scale as the natural ultraviolet regulator. The relation between the two represents the duality of UV/IR. On the other hand, the ultimate scale $``$ corresponds to the lowest temperature $`T_o`$ dual to the maximum attainable temperature $`T_P`$ ( a postulate of thermal relativity). This means that the universe’s evolution from the minimum Planck scale $`L`$ to the Nottale’s maximum scale $``$ is accompanied by a change of the temperature from the maximum temperature $`T_p`$ to the minimum (dual) temperature $`T_o`$.
Between these 2 limits the universe can ”hover” in the metastable state ( $`<D>`$ =$`4`$ \+ $`\varphi ^3`$ = $`4.236\mathrm{}`$) long enough to favor an emergence of a conscious life. This will continue until the whole ensemble of infinite quanta (bubbles,universes) begins its collective slow evolution towards the lower ( true vacuum ) stable state of average dimension $`\varphi ^3=0.236\mathrm{}`$ (or $`2+\varphi ^3`$ if considered relative to the $`D_o=2`$) of extremely cold temperatures transforming itself into a quasi-crystal state. The phase transition to the lower dimensions could be viewed as the ”big crunch” which transforms the universe into a ”pancake”. The latter is consistent with the observational data .
The disruption of the metastable state is caused by vacuum fluctuations. These fluctuations can be arbitrarily large due to the fact that the ensemble of universes comprises an infinity of quanta (bubbles of different dimensions and radii). An analogous situation arises in the field theory where quantum fluctuations can become large since the respective systems have an infinite number of degrees of freedom. <sup>4</sup><sup>4</sup>4This picture is not so unusual in physical applications if we recall that for example Van der Waals forces arise as a result of molecular fluctuations. The idea of an analogous origin of gravity was expressed independently by Feynman and by Vigier and Petroni . Our view is that classical spacetime and gravity emerged from vacuum fluctuations of the Cantorian fractal geometry ( dimensions, for example ). Within the framework of a new relativity and Cantorian fractal spacetime, one now needs to define the ” field ” (which we call Cantorian ) whose fluctuations cause the respective fluctuations of dimensions. Quite recently Sidharth has also pointed out the importance of fluctuations in Cantorian fractal spacetime. It is these fluctuations that generate classical spacetime, gravity, and all fundamental forces of nature. The quasi-crystal will eventually reach an enormous size while evolving towards the regime of the minimal energy density configuration. In fact this follows directly form eq.(4.7) which tells us that the final value of the cosmological ” constant ” will be dramatically smaller than it was at the birth of the universe. This regime ( if it will be reached in a final time interval) will indicate the ” end ” of spacetime, matter, energy,…and life.
Still we should not despair since it is quite possible that the cold quasi-crystallized ensemble will ” collide ”( in C-space, outside spacetime ) with another ensemble of much higher vacuum energy, higher temperatures ( we are assuming a ” multi-verse”), and the world will again begin to reheat and climb up the dimensional ladder, evolving towards the metastable state of $`D=4+\varphi ^3`$. Vacuum fluctuations will trigger another life cycle of the emerging universe ( in general different from the one that descended into quasi-crystal state), and the process will repeat itself time and time again. A more rigorous picture of this cyclical universes scenario requires the construction of the quantum field theory in noncommutative C-spaces .
We can speculate that this ” cosmic dance ” has been going on forever, and it will continue on forever within a cyclical scaling time associated with the renormalization group flow. The cyclical renormalization group scaling time is the true universal arrow of time. Its existence is predicated on the existence of negative dimensions ( a new relativity sea of negative dimensions). On the other hand, the time measured by our clocks is just one of the coordinates which is interchangeable with any other coordinate , the fact following from the diffeomorphic invariance of general relativity.
Our arguments showing that there is ” no cosmological constant problem ” in a new relativity agree with Nottale’s arguments who explained in simple terms a huge discrepancy between the cosmological ” constant ” measured at the cosmological scales, with the cosmological ” constant ” measured at Planck scales. It was necessary to formulate scale relativity to point out an existence of a dependence of measurements’ scales on a frame of reference. In a new relativity theory, there is no such thing as the cosmological ” constant problem ” because all the constants in nature are subject to renormalization group flow with scaling time. It is precisely because of a process of self-organization and self-tuning that all the constants in nature properly adjust themselves to the renormalization group scaling flow , or arrow of time. Such arrow of time itself materialized only because the world emerged from a strongly non-equilibrium state.
## 5 Exact Evaluation of the Nottale’s Upper Scale In Nature
Here we determine the exact value of the minimum non-zero temperature of the quasi-crystal phase. We will see that is directly related to the maximum upper scale in nature which ( as we already indicated) is dual to the Planck minimum scale. Nottale already gave estimates of what this scale should be. However at that time he was not aware of the power of the duality principle provided by the string theory and therefore had no way of knowing how to calculate the maximum upper scale from the basic principles. On the other hand, we will utilize duality and provide such an evaluation.
This duality between the large and the small is at the heart of the UV/IR entanglement of quantum field theory in noncommutative spaces . If the upper scale were infinity, and the lowest scale were zero, such an entanglement wouldn’t be possible. The spacetime coordinates of a noncommutative space do not commute because the Planck scale is not zero,i.e $`[X^i,X^j]\mathrm{\Lambda }^2`$. This property represents one of natural consequences of a new relativity : ordinary ” point ” coordinates $`x^\mu `$ are to be replaced by Clifford-algebra valued X multi-vectors ( matrices ). It is clear that in general the latter do not commute.
As we have already argued, the Planck scale is the ultimate UV regulator and the upper scale is the ultimate IR regulator. If in preparation of the ensemble of hyper-spheres we take into account Nottale’s scale relativistic corrections, the fractalization of these hyper-spheres would be taken into account from the very beginning. The respective volumes will be not only radius and dimension dependent but also resolution dependent. We will show that by including Nottale’s scale relativistic corrections (that is resolution-dependence of physical quantities) the system’s average dimension will change from the ” metastable ”quasi-ergodic value of $`<D>`$ = $`4`$ \+ $`\varphi ^3`$ to the true vacuum value of $`<D>=\varphi ^3`$.
In Section 3 we found that the average dimensions calculated with the use of either the discrete sum or the integral eq.(3.12) is very close to the exact average dimension $`d_c^{(4)}`$=$`4`$ \+ $`\varphi ^3`$ found on the basis of fractal dimensions and thus signaling the presence of a quasi-ergodic, metastable state. For this reason the world remains for a very long period (conducive to an emergence and existence of life forms) in a state whose average topological dimension is indeed very close to the value of $`D=4`$ associated with a smooth spheres ( manifolds).
To precisely calculate the average dimensions we should include the scale relativistic corrections . In essence these corrections amount to a computation of the averages by packing the ensemble thermal reservoir (”box”) with fractal spheres instead of smooth ones. The averaging requires integration with respect to the radii $`R`$ from $`R_H`$ to $``$ in the quasi-crystal phase and from $`\mathrm{\Lambda }`$ to $`R_H`$ in the metastable phase. The former case allows us to find the maximum upper length implicitly:
$$I=\frac{_0^{\mathrm{}}𝑑DD^2\sqrt{\pi }^D[\mathrm{\Gamma }(\frac{D+2}{2})]^1_{(R_H/\mathrm{\Lambda })}^{(/\mathrm{\Lambda })}d(\frac{R}{\mathrm{\Lambda }})(\frac{R}{\mathrm{\Lambda }})^{D1}[e^{\frac{E_{min}(R/\mathrm{\Lambda })^D}{KT_o}}1]^1J}{_0^{\mathrm{}}𝑑DD\sqrt{\pi }^D[\mathrm{\Gamma }(\frac{D+2}{2})]^1_{R_H/\mathrm{\Lambda }}^{(/\mathrm{\Lambda })}d(\frac{R}{\mathrm{\Lambda }})(\frac{R}{\mathrm{\Lambda }})^{D1}[e^{\frac{E_{min}(R/\mathrm{\Lambda })^D}{KT_o}}1]^1J}.$$
$`(5.1)`$
where $`J=_1^{R/\mathrm{\Lambda }}d(\frac{\xi }{\mathrm{\Lambda }})(\frac{\xi }{\mathrm{\Lambda }})^{D(\gamma 1)}`$, $`\gamma (\xi )=(1\beta ^2)^{1/2}`$ and $`\beta ^2=ln^2(\xi /R)/ln^2(R/\mathrm{\Lambda })`$ are Nottale’s scale-relativistic parameters, and $`\xi `$ is the resolution $`\mathrm{\Lambda }\xi R`$. At the ultimate resolution allowed in nature $`\xi =\mathrm{\Lambda }`$, and the volume<sup>5</sup><sup>5</sup>5$`D[\gamma (\xi )1]=D_FD_T`$ is the difference between the fractal (Hausdorf) and topological dimensions of a fractal hyper-sphere $`V=V_0\xi ^{D[\gamma (\xi )1]}`$ goes to 0 meaning that the infinite-dimensional limit is reached where spacetime evaporates into a sea of fractal dust, the hyper-point. This would require an infinite amount of energy.
Generally speaking, the average dimension given by eq.(5.1) depends on two parameters $`a=E_{min}/KT_o`$ and $`b=/\mathrm{\Lambda }`$, whose values should be chosen in such a way as to fit the experimental observations. Thus effectively our theory is a 2-parameter theory which is a great advantage as compared with other theories involving much more phenomenological parameters.
For example one can choose $`a=b`$ as a test condition. This yields an integral equation that will define the maximum upper scale $``$ exactly when we require the average dimension ( with respect to the zero reference point $`D_o=2`$ ) to be :
$$<D^{}>=2+\varphi ^3=2.236\mathrm{}.=I_1(E_{min}/KT_o=/\mathrm{\Lambda };/\mathrm{\Lambda };R_H/\mathrm{\Lambda }).$$
$`(5.2)`$
where $`I_1`$ is the value of I for $`a=b`$. Result (5.2) is tantamount to choosing an average value of $`<D>=\varphi ^3`$. This value reflects a decrease (on average) by one of each of the three spatial and one temporal dimensions ( plus fluctuations ) of the metastable vacuum whose overall average dimension is $`<D>`$ = $`4`$ \+ $`\varphi ^3`$.
In particular, scale relativistic corrections can shift the value $`<D^{}>=6.38\mathrm{}.`$ to the exact value of $`6`$+$`\varphi ^3`$ = $`6.236`$ by carefully selecting the value of the Hubble radius $`R_H`$ at the end of the metastable phase before it begins its evolution towards the true vacuum , the quasi-crystal phase. The respective temperature $`T_H`$ ( which is less than the $`T_{COBE}`$ ) is found from the following expression:
$$\frac{E_{vacuum}}{KT_H}=\frac{R_H}{\mathrm{\Lambda }}.$$
$`(5.3)`$
This temperature will influence the second integrals in (5.1),namely the integrals involving the Bose-Einstein distributions. Including the scale relativistic corrections, and integrating between the limits $`\mathrm{\Lambda }`$ and $`R_H`$ we arrive at one more integral equation:
$$6ϵ=I_2(E_{vac}/KT_H=R_H/\mathrm{\Lambda };R_H/\mathrm{\Lambda })=6+\varphi ^3=6.236\mathrm{}.$$
$`(5.4)`$
Equation (5.4) will give the exact ratios of $`R_H/\mathrm{\Lambda }=E_{vac}/KT_H`$ to fit the metastable fixed renormalization group point( relative to the zero point $`D_o=2`$) of $`6+\varphi ^3`$ . Here $`I_2`$ is the value of $`I`$ for $`E_{vac}/KT_H=R_H/\mathrm{\Lambda }`$.
It is seen that without using fractal spheres and the respective Nottale’s scale relativistic corrections it is impossible to fit the metastable state’s value $`6+\varphi ^3`$ . The latter is calculated relative to $`D_o=2`$ which implies that $`<D>=4+\varphi ^3`$. Such value precisely corresponds to the fractal dimensions of a set structure living ”underneath” a four dimensional smooth sphere. $`d_c^{(4)}`$ =$`4`$ \+ $`\varphi ^3`$ is exactly the Cantorian fractal dimension of the set $`^{(4)}`$.
Thus we have demonstrated an existence of a very close connection of the Cantorian fractal spacetime model and Nottale’s scale relativity. Moreover, using duality arguments of the string theory we were able to exactly define the upper Nottale’s scale $``$ via the integral equation that requires fitting $`\varphi ^3`$ as the true dimension of the true vacuum of our theory.
It is tempting to speculate that because there are $`3`$ different scales, $`\mathrm{\Lambda },R_H,`$ they are not necessarily independent. As a test function connecting these scales we can use the mean geometric formula:
$$R_H^2=\times \mathrm{\Lambda }(R_H/\mathrm{\Lambda })=(/R_H)(/\mathrm{\Lambda })=(R_H/\mathrm{\Lambda })^2..$$
$`(5.5)`$
The final verdict on the validity of (5.5) can be pronounced only if we will be able to calculate the integrals of (5.1).
## 6 Conclusion
It is our thesis that the evolutionary process of the universe is closely connected to the dynamics of self-organized criticality, complex systems, self- referential noise and quantum dissipative processes as fundamental aspects of reality. The universe as it evolves simultaneously self-tunes. Roughly speaking, this process could be described by what Finkelstein has called a variable quantum law, or $`q`$-process.
Clearly the naive Lagrangian formalism used in ordinary spaces will not work here. The master action functional
$$S\{\mathrm{\Psi }[X(\mathrm{\Sigma })]\}=[\frac{1}{2}(\frac{\delta \mathrm{\Psi }}{\delta X}\frac{\delta \mathrm{\Psi }}{\delta X}+^2\mathrm{\Psi }\mathrm{\Psi })+\frac{g_3}{3!}\mathrm{\Psi }\mathrm{\Psi }\mathrm{\Psi }+\frac{g_4}{4!}\mathrm{\Psi }\mathrm{\Psi }\mathrm{\Psi }\mathrm{\Psi }]DX(\mathrm{\Sigma }).$$
$`(6.1)`$
of quantum field theory in C-space ”lives” outside spacetime and as such requires a more general formalism. The latter is provided by the renormalization group approach which is an essential ingredient of the propagation of strings in curved spacetime backgrounds. The curved spacetime itself is the solution of the coupled Einstein-Yang-Mills equations which can be described as the vanishing of the beta functions associated with the world sheet couplings. The $`scaling`$ dynamics encodes the $`motion`$ dynamics of the strings.
The quantum group symmetry of such noncommutative quantum field theory is described by Braided Hopf Quantum Clifford algebras. The respective vertices are described as follows. The 2-point vertex corresponds to the pairing of the quantum algebra. The 3-point vertex is given by product and coproduct of the quantum algebra, that is annihilation of two C-lines and generation of the third line, or creation of two C-lines out of one C-line.
The Cantorian field $`\mathrm{\Psi }[X(\mathrm{\Sigma })]`$ is a hyper-complex number of Clifford algebra-valued object. In particular it could be quaternionic or octonionic-valued. The C-lines in C-space are nothing more than the Clifford-algebraic extension of Penrose’s twistors into a complex field. Action (6.1) is unique in a sense that the braided Hopf quantum Clifford algebra fixes the types of terms allowed by the action. Quantum fluctuations of the field $`\mathrm{\Psi }[X(\mathrm{\Sigma })]`$ of the Cantorian fractal Spacetime were responsible for the creation of the present quantum universe.
In $`D=\mathrm{}`$ limit it may be possible to construct a unique topological action for the world. The large $`D=N`$ limit was discussed briefly in with respect to the relations between conformally invariant $`\sigma `$ models on anti de Sitter spaces, $`AdS_{2N}`$, and Zaikov’s Chern-Simons $`p^{}`$-brane field theories residing on the projective boundaries of anti de Sitter spaces, $`S_{2N1}`$ spheres. When $`D=2N`$ and $`2N=p^{}`$ tend to $`\mathrm{}`$ then there is no distinction between $`D`$ and $`p^{}`$. As a result, Zaikov’s Chern-Simons $`p^{}`$-brane quantum field theory is the natural candidate for the topological field theory for the world. In this case ($`2N=D\mathrm{}`$) the Chern-Simons $`p^{}`$-brane becomes the infinite-dimensional spacetime filling $`p`$-brane .
Zaikov’s Chern-Simons classical $`pbrane`$ field theory admits $`W_1\mathrm{}`$ algebras as an algebra of constraints. Its respective connection to Vasiliev’s higher spin conformal field theories , $`W`$-geometry , and $`W_{\mathrm{}}`$ algebras based on Moyal-Fedosov quantization was given in . In the limit $`D\mathrm{}`$ one encounters a transition to the transfinite continuum of Cantorian fractal spacetime. This would require a study of infinite-dimensional loop spaces and associated loop algebras.
Finally, we believe that the evolving universe constructs its own Hilbert space. Hence, nonlinear complex dynamical systems has to be an essential part of reality. String theory, noncommutative geometry, quantum groups, Hopf algebras and the new scale relativity, for example, have already shown that at the fundamental level coordinates do not commute, and the Heisenberg uncertainty relations are to be modified to account for the Planck scale to be the minimum attainable length in nature. As we try to compress the strings (membranes, $`p`$-branes…) to scales smaller than the Planck scale the strings (membranes, $`p`$-branes…) begin to grow in size signaling that there is a nontrivial ultra violet/infrared entanglement. Let us not forget that the Planck scale should not to be confused with the string scale $`l_s`$ since for example the $`D`$-branes can probe distances smaller than $`l_s`$. Non-Archimedean geometry and $`p`$-Adic numbers are the natural geometry and numbers required by the New Relativity at Planck scales consistent with the Cantorian-Fractal nature of spacetime .
Self-tuning of the universe in process of its evolution implies that values of the fundamental ”constants”, including the cosmological constant are also adjusted accordingly. In particular the classical $`spacetime`$ of our perception is constantly evolving entity which does not exist $`abinitio`$. We can say that it is a $`process`$ in the making . Thus we can abandon the idea of a single universe inflated to the size observed today. Instead we can introduce a true q-process (self-referential) with a hierarchical family of universes (reminiscent of a matryoshka doll) where each representative member of the family has an average dimension (as seen today) of approximately $`4`$ \+ $`\varphi ^3`$ = $`4.236067977\mathrm{}`$. Therefore within this model we view our world only as a representative of an infinite ensemble of universes instead of a ”given ” and fixed universe inflated to the sizes of today’s universe. This ensemble approach allowed us to calculate universe’s average dimension by using the ensemble averaging. If we look at this differently, we in fact performed the Feynman path integration over the infinite possible scenarios/histories of the world.
A world as we see it is a ”perceptual averaging projection” of a perpetually changing processes underlying the visible world of our senses. The world tomorrow will not the same as the world today. Quoting Wheeler, it is possible to say that we ourselves are true observers and participants in this averaging process which we perceive as reality. This represents (in different terms) the Everret-De Witt-Wheeler many worlds interpretation of quantum mechanics. We perform billions of measurements every day using one of the most sophisticated measuring devises : our brains . Every time an observation is made, a branching of possible scenarios will occur. For this reason we wholeheartedly subscribe to Penrose’s view that the physics of the human brain ought to be included into future physical theories .
Acknowledgements
We thank D. Chakalov, S.Paul King , D. Finkelstein, E. Spallucci, E. Guendelmann , G. Chapline G. Bekkum and R. Guevara for many discussions. One of us (CC) wishes to thank A. Cabo, H. Perez, C. Trallero, D. Villarroel, M. Chaichan. To C. Handy. M. Handy, D. Bessis, A. Boedo, M. Bowers, A. Bowers our deepest gratitude for their assistance and encouragement.
## 7 Appendix A
Using the notion of a quantum ”path” as a fractal curve , with poles at the time interval $`\mathrm{\Delta }t=0`$ and shifting this value to the minimum attainable time interval , Planck’s time $`\tau _P`$
$$\mathrm{\Delta }t=\tau _P$$
$`(A1)`$
we get the following (cf.)
$$E=\frac{m}{2}(\frac{\mathrm{\Delta }x}{\mathrm{\Delta }t})^2=\frac{a}{\mathrm{\Delta }t\tau _p}.$$
$`(A2)`$
where a is a constant to be determined. We rewrite (A2)
$$\frac{m}{2}(\frac{\mathrm{\Delta }x}{\mathrm{\Delta }t})^2=\frac{m}{2}(\frac{\mathrm{\Delta }x}{\mathrm{\Delta }t})^2\frac{\tau _p}{\mathrm{\Delta }t}+\frac{a}{\mathrm{\Delta }t}.$$
$`(A3)`$
If we divide both sides of (A3) by $`m\mathrm{\Delta }x/2(\mathrm{\Delta }t)^2`$ and introduce the average momentum $`<p>`$= $`m(\mathrm{\Delta }x/\mathrm{\Delta }t)`$ then (A3) yields a relation between the spatial resolution $`\mathrm{\Delta }x`$ and the average momentum:
$$\mathrm{\Delta }x=\tau _P\frac{<p>}{m}+\frac{2a}{<p>}.$$
$`(A4)`$
Since by operational definition the resolution of a physical device $`\mathrm{\Delta }x`$ must be less ( or equal) than a statistical mean square deviation $`\mathrm{\Delta }x_s`$we get from (A4)
$$\mathrm{\Delta }x_s\mathrm{\Delta }x=\tau _P\frac{<p>}{m}+\frac{2a}{<p>}.$$
$`(A5)`$
The minimum of the r.h.s. of (A5) is reached at
$$\mathrm{\Delta }x_{min}=2\sqrt{2\tau _P\frac{a}{m}}$$
$`(A6)`$
which in particular means $`\mathrm{\Delta }x_s\mathrm{\Delta }x_{min}.`$ Since the average momentum $`<p>`$ is of the order of the mean square deviation,that is $`<p>^2(\mathrm{\Delta }p_s)^2`$ we get from (A6) the following
$$\mathrm{\Delta }x_s\mathrm{\Delta }p_s\frac{\tau _P}{m}(\mathrm{\Delta }p_s)^2+2a.$$
$`(A7)`$
If we set $`\tau _p`$ =$`0`$ we recover from (A7) the conventional uncertainty relation of quantum mechanics which means that $`2a=\mathrm{}`$. By setting the min resolution $`\mathrm{\Delta }x_{min}`$ = $`\mathrm{\Lambda }`$ we obtain from (A6) that
$$\frac{\tau _P}{m}=\frac{\mathrm{\Lambda }^2}{4\mathrm{}}.$$
$`(A8)`$
Upon inserting this value and $`2a=\mathrm{}`$ back into (A7) we arrive at the following
$$\mathrm{\Delta }x_s\mathrm{\Delta }p_s\mathrm{}[1+\frac{1}{22!}(\frac{\mathrm{\Lambda }}{\mathrm{}})^2\mathrm{\Delta }p_{s}^{}{}_{}{}^{2}].$$
$`(A9)`$
which is a truncated stringy uncertainty relation of string theory. The right hand side of (A9) (with $`<\delta p_s><p>^2=\mathrm{}^2k^2`$) represents an ”effective” Planck constant
$$h_{eff}=\mathrm{}[1+\frac{1}{22!}\mathrm{\Lambda }^2k^2].$$
$`(A10)`$
## 8 Appendix B
An Elementary Derivation of the Area-Entropy Relation In Any Dimension
In this appendix we review the derivation of the area-entropy relation and explain why the stringy holographic principle is a direct result of a new relativity. Let us consider an infinite-dimensional quantum spacetime quantized in discrete geometric bits of point, 1-loop, 2-loop ,…, $`p`$-loop histories. The latter play the same role as photons in quantum electrodynamics with a difference that $`p`$-loop histories ( quanta of spacetime) form a self-referential ( bootstrapping) medium of a dynamical spacetime in the making. Thus counting the number of these quanta we will get the information content of the attendant spacetime.
It has turned out that the appropriate counting mechanism is provided by Clifford algebra. In fact, a Clifford algebra in $`D`$ dimensions or degree $`D`$ has $`2^D`$ independent ” components ” that represent the total number of the point , holographic area, holographic volume, holographic four volume,etc. coordinates associated with the hierarchy of point, 1-loop, 2-loop, … histories, or excitations of $`D`$-dimensional spacetime - . This is simultaneously the total number of ”geometric quanta” $`N`$, meaning that $`N`$=$`2^D`$. The respective Shannon information entropy is
$$S=Kln2^D=KDln2.$$
$`(B1)`$
Therefore
$$D=\frac{S}{Kln2}s.$$
$`(B2)`$
On the other hand,the $`(D1)`$-dim hyper-area enclosing a $`D`$\- dimensional( bulk) hyper-volume of a $`D`$-dimensional sphere of a unit Planck radius is :
$$A_{D1}=D\frac{\sqrt{\pi }^D}{\mathrm{\Gamma }(\frac{D+2}{2})}.$$
$`(B3)`$
which means that $`D`$ and hence $`s`$ ( eq.B1) is an implicit function of the hyper-area $`A`$:
$$s=D(A)$$
$`(B4)`$
Since this function cannot be expressed analytically we restrict ourselves to its graphical representation shown in Fig. 1. It is easily seen that the graph has a region of a linear dependence $`s`$ vs. $`A`$ in a rather narrow range of dimensions, namely from $`D=2`$ to $`D=5`$.
The slope of the linear region is $`1/6.9`$ which is less than the Bekenstein upper bound. For the dimensions outside this range the dependence clearly deviates from linearity. If we consider a sphere with a non-unit radius then a respective graph of $`s`$ vs. $`A`$ will look similar to Fig.1, with the only difference that for the same $`s`$ the values of the respective areas $`A`$ will increase with the radius increase, as in Fig 2.
Let us apply this result to macroscopic black holes.We can consider a macroscopic black-hole as being built from many ”bits” or mini black holes of unit Planck radius. The state of a macroscopic black-hole is evaluated by simply ”counting ” the number of micro-states accessible to the macroscopic system.
Since Bekenstein and Hawking assumed that there exists a linear relation between area and entropy (valid only for a certain range of dimensions, as was shown above) , the entropy could be found by simply adding elementary portions (or bits) of areas. In other words, a linear superposition of the fundamental mini black holes states was possible due to the linearity of the area-entropy relation . A more general way of combining bits of information-entropy is as follows.
Suppose we wish to calculate the entropy of a cubic object. Since a cube is topologically equivalent to a sphere, we simply deform the cube into a sphere, simultaneously preserving the cube’s area. This leads to an increase of the sphere’s volume as compared to the original cube. As a next step we evaluate the surface area of the sphere suitably normalized to Planck units. If now we divide this area by the area of a unit sphere (in Planck units) then the result would give us the information entropy ( divided by $`KLn2`$) of the original cube.
However there are two obvious objections to this procedure:
1.Is it possible to pack all unit spheres into the big sphere without leaving any voids ?
2. What if the information-entropy, given by the ratio of areas (in units of $`kln2`$) is not an integer ?
To answer the first objection we perform an inverse deformation of each unit sphere into a fundamental small cube of the same area, then we deform the large sphere into the original cube of the same area. Now we can pack all the elementary cubes into a large cube without leaving voids.
The second objection is a little bit more tricky. In this case a small region of the cube remains unpacked. To resolve this problem we use properties of fractal geometry: the large cube is to be filled with cubes of fractal dimensions. Since fractals have a property of space filling the cube will be filled without any voids left.
Now we make a plausible assumption that only one bit of information can be ascribed to a minimum attainable scale, the Planck scale $`\mathrm{\Lambda }.`$ This assumption is supported by the holographic origins of chaos in $`M`$ theory . An onset of chaos is signaled by impossibility to pack the energy levels ( spectra ) into small regions of very high energy ( spectral ) density distributions. The repulsive ( holographic ) feature of these spectral lines is the indication of quantum chaos.
This means that for every two adjacent cubes/cells packed side by side, information will be ”pushed” out to the respective surfaces. As a result their common boundary surface would contain two bits of information per unit area (in Planck units) , which contradicts our initial assumption that one cannot have more than one bit per unit area ( an ” exclusion principle ”). Applying this argument to the overall packing-process we arrive at a situation analogous to the Gauss’s law in electrostatics: the information can reside only on the boundary surface of a cube, and hence on the surface of the original sphere. For the importance of the Cantorian fractal spacetime model for the theory of superconductivity, and its potential high temperature applications, see .
In general, the graphs $`s=D(A)`$ (figs. $`1`$ and $`2`$) exhibit a variable slope indicating that information density changes with the area. The slope reaches infinity in two cases: 1) at a maximum attainable area and 2)in the limit of $`D\mathrm{}`$ when the area goes to $`0`$. The latter signals an onset of a phase transition under which spacetime ”evaporates” leaving behind a fractal Cantor dust. Since the area tends to $`0`$ the information density blows up.
In the former case ( for simplicity consider a unit sphere) the slope becomes infinite between $`D=4`$ and $`D=5`$. This is the metastable renormalization group fixed point of $`D`$= $`4`$+$`ϵ`$ = $`4`$ \+ $`\varphi ^3`$. Not surprisingly, this is another physical reason why we live in $`D=4`$. Can we actually reach $`D=\mathrm{}`$ ? In other words, how much energy this process would require? In we have discussed why dimensions, energy, information are indistinguishable at the ultimate scale provided by nature, the Planck scale $`\mathrm{\Lambda }`$. Once Nottale’s scale relativistic corrections to resolutions of physical devices are taken into account, it turns out that to reach $`D=\mathrm{}`$ (which is tantamount to reaching the Planck scale ) one needs an infinite amount of energy.
## 9 Appendix C
Renormalization group flow and a local scaling arrow of time.
Here we would like to point out a connection of our work to the renormalization group techniques of quantum field theory within the context of fractals. The emergence of a microscopic Liouville arrow of time within the context of $`D`$-branes and the breakdown of Lorentz invariance at very short scales has been analyzed by Ellis et al . In a sense, a picture of an infinite unfolding of a hierarchy of infinite-dimensional worlds represents a visualization of actions of the renormalization group. In simple terms we can recreate this picture as follows. Let us focus our eyes at a small region of a monitor which we will conditionally call a ”point”. Certainly one can even imagine to be ”inside” this point. A physical realization of this immersion would require a lot of energy which will increase with a decrease of the point’s size. As more energy is supplied to get inside the ”point ” (corresponding to a greater zoom) an observer will notice unfolding of additional dimensions. This resembles playing with a toy matryoshka doll. Since energy generates information ( and generated by information) an increase of energy in this process will generate additional information and as a result will reverse a local microscopic scaling arrow of time.
We begin to move into a different hierarchical family of the Cantorian matryoshka-like spacetimes whose local average dimension will be greater than our initial $`D=4`$. The opening worlds would be the members of the family which belongs to our ”past” with respect to the local scaling renormalization group arrow of time. However to reverse the renormalization group arrow of time of the whole (global) universe is a totally different enterprise because energy involved in achieving such a feat would be enormous, of the order of the total mass-energy of the whole universe.
Kreimer has provided an ample proof that the renormalization and combinatorial techniques of quantum field theory contains a Hopf algebra associated with the Feynamn’s diagrams. A multi fractal, multi-vector , and multi-scaling basis of a new relativity allows one to utilize a rich and diverse mathematical apparatus of quantum groups, noncommutative geometry, Feynman graphs, Hopf algebras, etc. as an algebraic-combinatorial and self-recursive way of coding the infinite unfolding of the transfinite Cantorian multi fractal hierarchical matryoshka doll-like worlds. Interestingly enough, the Euler gamma function ( which is self-recursive and which plays an important role in our derivation)is an essential tool in the renormalization process of quantum field theory.
To make our statement about reversing a scaling arrow of time more precise we invoke Zamolodchikov’s central charge theorem which originally was formulated for flows in conformal field theories in $`D=2`$ but can be easily generalized to $`D=2n`$. The theorem states that the central charge $`c[g^i,\mu ]`$ monotonically decreases along its renormalization group trajectory moving towards the infrared region
$$\frac{g^i}{ln\mu }\frac{c[g^i,\mu ]}{g^i}=\frac{c[g^i,\mu ]}{ln\mu }0.$$
$`(C1)`$
Here $`g^i`$ are the world sheet couplings and $`\mu `$ is the scale,or the subtraction point.
On the other hand, in string theory and conformal field theory central charges are directly related to dimensions. For example, critical strings are anomaly free in $`D=26`$ to cancel the $`c=26`$ charge associated with the ghosts; critical superstrings are anomaly free in $`D=10`$, etc. This fact indicates that our assertion about ” reversal” of the scaling renormalization group flow on its way to regions of higher central charges and hence higher dimensions is compatible with Zamolodchikov’s theorem. Moreover, there exists a connection between conformal field theory , the monster group, and sphere packing which could be found in ref. . Roughly speaking , the monster group associated with the Golay error correcting code for a self-recursive system encodes the unfolding process into higher dimensions. And conversely, upon reversal of the scaling arrow of time, the monster group encodes condensation of dimensions from the higher( on the average) initial values to the smaller ( on the average ) values. This could be viewed as coding a monotonic (quasi) crystallization process. Similar ideas about a code underlying spacetime being a q-network of q-processes was expressed by Finkelstein long time ago .
|
warning/0004/gr-qc0004068.html
|
ar5iv
|
text
|
# Black Holes with Zero Mass
## I Introduction
Global monopoles are topological defects that arise in certain theories where a global symmetry is spontaneously broken. The simplest and most studied example is the $`O(3)`$ model, in which one finds that the static, spherically symmetric solutions, have in general an energy density that decreases for large distances as $`1/r^2`$. This would lead in a Newtonian analysis to a divergent expression for the total mass. When we turn to the general relativistic analysis, this problem translates in the fact that the resulting spacetime is not asymptotically flat, and thus the standard ADM mass is not well defined. The effect of the $`1/r^2`$ behavior of the density on the spacetime is that the latter develops a deficit angle at large distances The fact that, at small distances, the behavior deviates from that, results in the appearance of a small phenomenological ”core mass” which turns out to be negative in all cases considered . Moreover, an analysis of the behavior of geodesics in the large distance regime does indeed support such interpretation of this core mass because its effect turns out to be repulsive. This “core mass” is then evidently not the standard ADM mass. The question of what exactly one is talking about when referring to this core mass has been resolved in through the application of the standard type of Hamiltonian analysis to the class of spacetimes that are Asymptotically-flat-but-for-a-deficit-angle (AFDA$`\alpha `$) .
For these (AFDA$`\alpha `$) spacetimes one can also define future and past conformal null infinity, and thus the notion of a black hole and of its horizon. In fact solutions corresponding to global monopoles with such interior horizons have been found in , .
In this paper we study the behavior of the ADM mass of these AFDA$`\alpha `$ spacetimes as a function of the horizon area, concentrating in particular on its sign, which we find changes in the regime where one would interpret as going from a situation that would be naturally described as a “black hole inside a monopole core” to that which would be naturally described as a “black hole with a global monopole inside”.
We shall adhere to the following conventions on index notation in this paper: Greek indices ($`\alpha `$, $`\beta `$, $`\mu `$, $`\nu `$,…) range from $`0`$ to $`3`$, and denote tensors on (four-dimensional) spacetime. Latin indices, alphabetically located after the letter i (i,j,k,…) range from $`1`$ to $`3`$, and denote tensors on a spatial hipersurface $`\mathrm{\Sigma }`$; whereas Latin indices, from the beginning of the alphabet (a,b,c,d,…) range from $`1`$ to $`3`$, and denote indices in the internal space of the scalar fields. The metric for the internal space is just the flat Euclidean metric $`\delta _{ab}`$. The signature of the spacetime metric g is $`(,+,+,+)`$. Geometrized units, for which $`G_N=c=1`$ are used in this paper.
## II The Global Monopole Spacetime
The theory of a scalar field with spontaneously broken internal $`O(3)`$ symmetry, minimally coupled to gravitation, is described by the action:
$$S=\sqrt{(g)}[(1/16\pi )R(1/2)(^\mu \varphi _a)(_\mu \varphi ^a)V(\varphi )]d^4x.$$
(1)
where R is the scalar curvature of the spacetime metric, $`\varphi _a`$ is a triplet of scalar fields, and $`V(\varphi )`$, is potential depending only on the magnitude $`\varphi =(\varphi _a\varphi ^a)^{1/2}`$, which we will take to be the “Mexican Hat” $`V(\varphi )=(\lambda /4)(\varphi ^2\eta ^2)^2`$.
The gravitational field equations following from the Lagrangian (1) can be written as
$$R^{\mu \nu }\frac{1}{2}g^{\mu \nu }R=8\pi T^{\mu \nu }$$
(2)
where
$`T_{\mathrm{sf}}^{\mu \nu }`$ $`=`$ $`^\mu \varphi ^a^\nu \varphi _ag^{\mu \nu }\left[{\displaystyle \frac{1}{2}}(\varphi ^a)(\varphi _a)+V(\varphi )\right].`$ (3)
The equation of motion for the scalar fields become
$`\mathrm{}\varphi ^a={\displaystyle \frac{V(\varphi )}{\varphi _a}}.`$ (4)
We are interested in spacetimes with topology $`\mathrm{\Sigma }\times R`$, where $`\mathrm{\Sigma }`$ has the topology of $`(R^3B)C`$, with $`B`$ a 3-ball, and $`C`$ a compact manifold with $`S^2`$ boundary.
We will focus on the sector corresponding to the asymptotic behavior characteristic of the Hedgehog ansatz:
$$\varphi ^a\eta x^a/r.$$
(5)
where the $`x^a`$ are asymptotic Cartesian coordinates. Within this sector, there is a static, spherically symmetric solution with metric given by:
$$ds^2=B(r)dt^2+S(r)dr^2+r^2(d\theta ^2+\mathrm{sin}^2(\theta )d\phi ^2),$$
(6)
and scalar field
$$\varphi ^a=\eta f(r)x^a/r$$
(7)
and with the following asymptotic behavior of,
$$BS^11\alpha 2M/r+O(1/r^2),f1+O(1/r^2)$$
(8)
where $`\alpha =8\pi \eta ^2`$. Redefining the $`r`$ and $`t`$ coordinates as $`r(1\alpha )^{1/2}r`$ and $`t(1\alpha )^{1/2}t`$, respectively, and defining $`\stackrel{~}{M}=M(1\alpha )^{3/2}`$, we obtain the asymptotic form for the metric:
$$ds^2=(12\stackrel{~}{M}/r)dt^2+(12\stackrel{~}{M}/r)^1dr^2+(1\alpha )r^2(d\theta ^2+\mathrm{sin}^2(\theta )d\phi ^2).$$
(9)
As we mentioned it is natural to associate the parameter $`\stackrel{~}{M}`$ with the mass of the configuration because it can be seen that the proper acceleration of the ($`\theta ,\phi ,r`$) = constant world lines is $`a=\stackrel{~}{M}/(r(r2\stackrel{~}{M}))`$. However, as we explained before it is not the standard ADM mass. This is also evident from the fact that in the specific solution $`\stackrel{~}{M}`$ turns out to be negative , while the matter certainly satisfy the dominant energy condition, under which the ADM mass of a regular solution would be positive , .
These issues are clarified by the introduction of concept of asymptotically-flat-but-for-a-deficit-angle spacetimes (A.F.D.A $`\alpha `$) and the standard asymptotically-flat-but-for-a-deficit-angle spacetime (S.A.F.D.A $`\alpha `$). The ADM mass of any spatial hipersurface of the former being defined in terms of its spatial metric and the metric of a particular slice of the S.A.F.D.A $`\alpha `$ spacetime (see for details).
In fact we take as the S.A.F.D.A $`\alpha `$ spacetime the metric,
$$ds^2=dt^2+dr^2+(1\alpha )r^2(d\theta ^2+\mathrm{sin}^2\theta d\phi ^2).$$
(10)
The natural generalization of the ADM mass for the A.F.D.A $`\alpha `$ spacetimes is
$$16\pi (1\alpha )M_{ADM\alpha }=_\mathrm{\Sigma }𝑑S_i(h^{0ij}h^{0kl}h^{0ik}h^{0jl})D_k^0(h_{jl}).$$
(11)
where $`h_{ij}`$ is the spatial metric of the slice of the spacetime for which the ADM$`\alpha `$ mass will be evaluated, $`h_{ij}^0`$ is the spatial metric of a static slice of the S.A.F.D.A $`\alpha `$ spacetime, and $`D_j^0`$ the covariant derivative associated with the latter (in fact, it looks just like the usual ADM formula, but with the quantities associated with the flat metric replaced by the S.A.F.D.A.$`\alpha `$ metric), and just like this, it is the numerical value of the true Hamiltonian ( a true generator of “time translations”); so, it is natural to interpret this as the mass (or energy) of the A.F.D.A.$`\alpha `$ spacetimes.
Let us write a static spherically symmetric metric in the form
$$ds^2=\left(1\frac{2m}{r}\right)e^{2\delta }dt^2+\left(1\frac{2m}{r}\right)^1dr^2+(1\alpha )r^2(d\theta ^2+\mathrm{sin}^2\theta d\phi ^2).$$
(12)
where the functions $`m(r)`$ and $`\delta (r)`$ depend only on the radial coordinate $`r`$.
We also introduce the following dimensionless quantities
$`\stackrel{~}{r}`$ $`:=`$ $`r\eta \lambda ^{1/2},`$ (13)
$`\stackrel{~}{m}`$ $`:=`$ $`m\eta \lambda ^{1/2},`$ (14)
$`\stackrel{~}{E}`$ $`:=`$ $`E{\displaystyle \frac{1}{\eta ^2\lambda }},`$ (15)
$`\alpha `$ $`:=`$ $`8\pi \eta ^2,`$ (16)
Introducing the metric (12) and the ansatz (7) in the expressions (2) and (4), we obtain the final form of the equations of motion :
$`_{\stackrel{~}{r}}\stackrel{~}{m}`$ $`=`$ $`4\pi \stackrel{~}{r}^2\stackrel{~}{E}{\displaystyle \frac{\alpha }{2(1\alpha )}},`$ (17)
$`_{\stackrel{~}{r}}\delta `$ $`=`$ $`{\displaystyle \frac{\alpha \stackrel{~}{r}}{2}}(_{\stackrel{~}{r}}f)^2,`$ (18)
$`_{\stackrel{~}{r}\stackrel{~}{r}}f`$ $`=`$ $`\left[{\displaystyle \frac{2}{\stackrel{~}{r}}}_{\stackrel{~}{r}}\delta 2A^2\left(4\pi \stackrel{~}{r}\stackrel{~}{E}{\displaystyle \frac{\alpha }{2(1\alpha )\stackrel{~}{r}}}{\displaystyle \frac{\stackrel{~}{m}}{\stackrel{~}{r}^2}}\right)\right](_{\stackrel{~}{r}}f)`$ (20)
$`+A^2\left[f(f^21)+{\displaystyle \frac{2f}{(1\alpha )\stackrel{~}{r}^2}}\right],`$
where
$`A^2=\left(1{\displaystyle \frac{2\stackrel{~}{m}}{\stackrel{~}{r}}}\right)^1,`$ (21)
and
$`\stackrel{~}{E}`$ $`=`$ $`{\displaystyle \frac{\alpha }{8\pi }}\left[{\displaystyle \frac{(_{\stackrel{~}{r}}f)^2}{2A^2}}+{\displaystyle \frac{f^2}{(1\alpha )\stackrel{~}{r}^2}}+{\displaystyle \frac{(f^21)^2}{4}}\right].`$ (22)
In this paper, we will make use of the fact that when the metric is written as (12), the formula (11) for the ADM$`\alpha `$ mass of the spacetime can be expressed as:
$$M_{\mathrm{ADM}\alpha }=M=\mathrm{lim}_r\mathrm{}m(r).$$
(23)
Regular solutions require the following boundary conditions at the origin $`f(0)=0`$, $`\stackrel{~}{m}(0)=0`$, $`\stackrel{~}{m}(0)_{,\stackrel{~}{r}}=\alpha /(2(1\alpha ))`$ and using a standard shooting method to compute $`f(0)_{,\stackrel{~}{r}}`$. The solutions are found by the standard one parameter shooting method . This is possible because the equation for $`\delta `$ decouples, and can thus be solved independently, after the rest of the functions are solved, and because, once the function $`f`$ approaches $`1`$ (sufficiently fast) at $`\mathrm{}`$, eq.(17) ensures that $`\stackrel{~}{m}`$ converges to a finite value.
A very useful tool in the analysis of this kind of configurations is provided by the consideration of the limit in which $`\lambda \mathrm{}`$ which can formally be taken by replacing the potential term by the constraint $`\varphi ^2=\eta ^2`$.
In fact in this case the field equations become;
$$R^{\mu \nu }\frac{1}{2}g^{\mu \nu }R=8\pi T_{\mathrm{const}}^{\mu \nu }$$
(24)
where
$`T_{\mathrm{const}}^{\mu \nu }`$ $`=`$ $`^\mu \varphi ^a^\nu \varphi _ag^{\mu \nu }\left[{\displaystyle \frac{1}{2}}(\varphi ^a)(\varphi _a)\right].`$ (25)
while the equation of motion for the scalar fields become
$`\mathrm{}\varphi ^a={\displaystyle \frac{1}{\eta ^2}}\varphi ^b(\mathrm{}\varphi _b)\varphi ^a.`$ (26)
Using the metric (12) we can verify that the anzats $`\varphi _a=\eta x_a/r`$, satisfies identically eq. (26) and the eqs. (24) reduce to
$`_{\stackrel{~}{r}}\stackrel{~}{m}`$ $`=`$ $`0,`$ (27)
$`_{\stackrel{~}{r}}\delta `$ $`=`$ $`0,`$ (28)
and
$`\stackrel{~}{E}`$ $`=`$ $`{\displaystyle \frac{1}{8\pi }}\left({\displaystyle \frac{\alpha }{(1\alpha )\stackrel{~}{r}^2}}\right).`$ (29)
Which results in the solution given by the metric
$$ds^2=\left(1\frac{2M}{r}\right)dt^2+\left(1\frac{2M}{r}\right)^1dr^2+(1\alpha )(r^2d\theta ^2+r^2\mathrm{sin}^2\theta d\phi ^2).$$
(30)
which for the choice $`M=0`$, is in fact what is taken as the S.A.F.D.A $`\alpha `$ spacetime (10).
## III Solutions with Black Hole Horizons
We note that eq. (30) with $`M>0`$ corresponds to the case of a solution with a regular event horizon. In this case, as in the standard asymptotically flat case, the mass is in fact positive and given by $`M=\left(\frac{A_H}{16\pi (1\alpha )}\right)^{1/2}`$ where $`A_H`$ is the area of the horizon.
The issue is then, whether in the general situation, i.e. without taking the limit $`\lambda \mathrm{}`$, we will find also positive masses, or whether the presence of the monopole will in some instances dominate and make the mass negative?. Intuition of course suggests that the latter will be the case, and that the two regimes (positive and negative mass) are possible, with the interplay between the two scales, the monopole core scale given by $`r_c(\eta \lambda ^{1/2})^1`$, and the black hole radius given by $`r_H`$, defining which one prevails. In the case $`r_H>r_c`$, which we will consider as a “monopole within a black hole‘”, one expects the black hole features to dominate, and that therefore the $`M_{\mathrm{ADM}\alpha }`$ will be positive. In the case $`r_H<r_c`$, which we will consider as a “black hole inside a monopole”, one expects the monopole features to dominate, and thus that the $`M_{\mathrm{ADM}\alpha }`$ will be negative. Furthermore there should exist a specific regime where $`r_Hr_c`$ and for which the two tendencies will exactly compensate each other so that the mass should be zero.
We will investigate these issues numerically and will find that the above picture is in fact confirmed by the results.
Configurations corresponding to static spherically symmetric regular black hole horizons with area $`A=4\pi r_H^2=\frac{4\pi \stackrel{~}{r}_H^2}{\lambda \eta ^2}`$ are those that satisfy the standard boundary conditions at infinity, i.e., $`\mathrm{lim}_r\mathrm{}f(r)=1`$, $`\mathrm{lim}_r\mathrm{}m(r)=M`$, where $`M`$ is a constant, and that at $`\stackrel{~}{r}=\stackrel{~}{r}_H`$ satisfy:
$`2\stackrel{~}{m}(\stackrel{~}{r}_H)=\stackrel{~}{r}_H.`$ (31)
We ensure that the time-translational Killing field of the metric is normalized to unit at infinity. This is done by fixing the constant that appears in the integration of the equation for $`\delta `$ in such a way that $`\mathrm{lim}_r\mathrm{}\delta (r)=0`$.
The value of the monopole field at $`\stackrel{~}{r}=\stackrel{~}{r}_H`$, $`f(\stackrel{~}{r}_H)=f_H`$, is taken as the shooting parameter, and is thus fixed by the boundary conditions at spatial infinity. The derivatives at $`r=r_H`$ of the functions $`f(r)`$ and $`m(r)`$ are given by:
$`\stackrel{~}{m}(\stackrel{~}{r}_H)_{,\stackrel{~}{r}}={\displaystyle \frac{\alpha (f_H^21)}{2(1\alpha )}}\left[1+{\displaystyle \frac{(1\alpha )\stackrel{~}{r}_H^2(f_H^21)}{4}}\right],`$ (32)
$`f(\stackrel{~}{r}_H)_{,\stackrel{~}{r}}={\displaystyle \frac{f_H(2+\stackrel{~}{r}_H^2(1\alpha )(f_H^21))}{\stackrel{~}{r}_H(1\alpha )(12m(\stackrel{~}{r}_H)_{,\stackrel{~}{r}})}},`$ (33)
In the same way as in the regular case the system is analyzed as a standard one dimensional shooting problem.
In the $`\lambda \mathrm{}`$ limit we can see from (30) that the ADM mass is always positive, and this can be easily understood if we note that this limit can be thought to represent the extreme case of a monopole inside a black hole .
The Fig.1 shows the behavior of the ADM$`\alpha `$ mass as a function of the horizon radius. Note that as expected there exists a radius where the mass ADM$`\alpha `$ vanishes. These are therefore zero mass black holes.
We must emphasize that in this class of spacetimes the mass is not positive definite and therefore there is nothing really paradoxical about the fact that there are black holes that have zero mass. However we must also point out that the mass in this case is not just a definition as it really reflects the effects on the test particles at large distances from the ”body” which in the cases treated here are the monopole core and/or the black hole horizon. In this case a zero value for the mass means that at large distances, the proper acceleration of the static bodies (i.e., those following integral curves of the static Killing field) falls off faster than $`1/r^2`$. Moreover, as pointed out in these black holes satisfy the standard laws of black hole dynamics and are nondegenerate as can be seen from the evaluation of the surface gravity $`\kappa `$. This is obtained from the expression $`t^\mu _\mu t^\nu =\kappa t^\nu `$, i.e., the surface gravity is defined in terms of the acceleration at the horizon of the time-translational Killing field $`t^\mu `$ which is unit at infinity. In general for a spherically symmetric system the surface gravity is ,
$$\kappa =(\eta \lambda ^{1/2})\frac{1}{2\stackrel{~}{r}_\mathrm{H}}e^{\delta (\stackrel{~}{r}_\mathrm{H})}\left[12\stackrel{~}{m}(\stackrel{~}{r}_H)_{,\stackrel{~}{r}}\right]$$
(34)
here the derivative of the metric function evaluated at the horizon can be evaluated from formula (32). The surface gravity is positive definite as it is shown in Figure 2.
Previous works on this system (e.g.) were unable to study the main point of this work because they lacked an appropriate definition of the mass for the class of A.F.D.A.$`\alpha `$ spacetimes.
These type of models with unusual asymptotic are a very interesting ground to investigate the robustness of the standard results of the physics of black holes and to understand which of those results are specific to the asymptotic flatness assumption. This point might seem to be of purely academic interest, but we must remember that our universe is not asymptotically flat, and that the latter is just an approximation that is introduced in order to simplify the treatment of regions of spacetime that might be regarded as “isolated” from the rest of the universe, thus the issue of the degree to which the standard results are independent of the precise form of the asymptotics is indeed of practical importance because it will tell us which of them might have to be taken as really pertinent to our universe. In this work we have learned for example that the mass of a black hole need not be positive and might indeed be negative or even zero. Here it is worth recalling that, as a result of the fact that the spacetimes in question are not asymptotically flat, the notion of ADM mass used throughout this work is different from the usual one, and thus, in particular, the standard theorems concerning the positivity of the standard ADM mass do not apply (for details see). Other results that we feel should be studied in this context include, for example, the black hole uniqueness theorems, the black hole entropy results arising from the various proposals for a quantum theory of gravity, etc. Some of these issues are currently under investigation and will be the subject of forthcoming articles.
## Acknowledgments
This work was in part supported by DGAPA-UNAM grant No IN121298 and by CONACyT grant 32272-E. U.N. is supported by a CONACyT Postdoctoral Fellowship Grant 990490.
## References
|
warning/0004/hep-th0004159.html
|
ar5iv
|
text
|
# Open Superstring and Noncommutative Geometry
## I Introduction
The noncommutative geometry has been considered for some time in connection with various physics subjects, which include the lowest Landau level physics in condensed matter physics, the quantum plane in mathematical physics, and the geometrical interpretation of Yang-Mills-Higgs action and so forth. Recent motivation to study the noncommutative geometry mainly comes from the string theory. If the matrix model of M-theory is compactified on tori in the presence of an appropriate background field, the noncommutative supersymmetric Yang-Mills theory arises . It implies that the $`D`$-brane world volume theory is noncommutative. This point has been discussed later by a more direct approach. If we quantize the open string in the background of a $`D`$-brane with the NS $`B`$-field, the ends points of the string, attached on the $`D`$-brane is shown to be noncommutative . The effective action for the $`D`$-brane in the low energy regime is induced by the open string on it, thus it should be noncommutative. In their recent paper Seiberg and Witten extensively discussed the various aspects of the noncommutative geometry in the context of the string theory such as the equivalence between ordinary gauge fields and the noncommutative gauge fields, Morita equivalence, and its implications in M-theory.
The noncommutativity in the open string theory can be most easily seen in the framework of canonical quantization. The approach based on the canonical quantization was adopted in the earlier works on the open string in the $`D`$-brane background and further elaborated recently in refs. . In we show that the set of constraints generated by the mixed boundary condition can be explicitly solved. Solving the constraint conditions we find the open string is governed by a free string Hamiltonian defined on the target space with the effective metric $`G`$. Resorting to the canonical analysis we also evaluate the Polyakov path integral for the open string and obtain the noncommutative Dirac-Born-Infeld action as the low energy effective action for the $`D`$-brane, which reduces to the noncommutative Yang-Mills theory in the zero slope limit. If the constraint conditions are imposed, the Wilson loop operator with the ordinary gauge field becomes the Wilson loop operator with a noncommutative gauge field. Hence the Seiberg-Witten map, which connects the ordinary gauge field to the noncommutative one, can be understood in the framework of the canonical quantization .
The appearance of the noncommutative geometry in the bosonic string theory is now well understood. However, the noncommutative geometry in the framework of the superstring theory needs further study. Since the $`D`$-brane is a BPS object, many interesting features of the $`D`$-brane are associated with the supersymmetry. Therefore, it is important to understand how the noncommutativity interplay with the supersymmetry. In the present paper we attempt to extend the previous canonical analysis to the supersymmetric theory. To this end we take the Neveu-Schwarz-Ramond (NSR) superstring in the background of $`D`$-brane with NS $`B`$-field.
## II The NSR Superstring in the background of $`D`$-brane
We begin with the world sheet action for the free NSR superstring in superspace. In the presence of the $`D`$-p-brane the action for the superstring is given as
$`S`$ $`=`$ $`{\displaystyle \frac{1}{4\pi \alpha ^{}}}{\displaystyle _M}d^2\xi d^2\theta g_{\mu \nu }D_+Z^\mu D_{}Z^\nu `$ (1)
$`Z^\mu (\xi ,\theta _\pm )`$ $`=`$ $`X^\mu +\theta _+\psi _{}^\mu +\theta _{}\psi _+^\mu +\theta _+\theta _{}F^\mu `$ (2)
$`D_+`$ $`=`$ $`{\displaystyle \frac{}{\theta _{}}}i\theta _{}_+,D_{}={\displaystyle \frac{}{\theta _+}}+i\theta _+_{}`$ (3)
where we choose the world sheet metric as $`(,+)`$ and $`F^\mu `$ is the auxiliary field. $`Z^a`$, $`a=p+1,\mathrm{},9`$ of the transverse directions are subject to the Dirichlet boundary condition, $`_\tau Z^a=0`$. If the $`D`$ brane carries the NS $`B`$-field, the action for the longitudinal coordinates $`Z^i`$, $`i=0,\mathrm{}p`$, is replaced with the following one
$`S`$ $`=`$ $`{\displaystyle \frac{1}{4\pi \alpha ^{}}}{\displaystyle _M}d^2\xi d^2\theta (g+2\pi \alpha ^{}B)_{ij}D_+Z^iD_{}Z^j.`$ (4)
Since the action for the transverse coordinates is rather trivial, we will be concerned with the action for the longitudinal coordinates only hereafter. For the background with constant $`g`$ and $`B`$, the action is greatly simplified to
$`S`$ $`=`$ $`{\displaystyle \frac{1}{4\pi \alpha ^{}}}{\displaystyle _M}d^2\xi \left(E_{ij}_+X^i_{}X^ji\psi _+^iE_{ij}_{}\psi _+^ji\psi _{}^iE_{ij}^T_+\psi _{}^j\right),`$ (5)
$`E_{ij}`$ $`=`$ $`(g+2\pi \alpha ^{}B)_{ij}.`$ (6)
The boundary conditions are given as
$`g_{ij}_\sigma X^i2\pi \alpha ^{}B_{ij}_\tau X^i`$ $`=`$ $`0,`$ (8)
$`E_{ij}\psi _+^jE_{ij}^T\psi _{}^j`$ $`=`$ $`0,`$ (9)
for $`\sigma =0`$, $`\pi `$.
For canonical quantization of the bosonic part we refer to ref. : The boundary condition Eq.(8) generates a set of second class constraints, which can be explicitly solved. After solving the constraints, the bosonic part of the Hamiltonian and the string coordinate variables are obtained as
$`H_B`$ $`=`$ $`(2\pi \alpha ^{}){\displaystyle \frac{1}{2}}p_i(G^1)^{ij}p_j+(2\pi \alpha ^{}){\displaystyle \underset{n=1}{}}\left\{{\displaystyle \frac{1}{2}}K_{in}(G^1)^{ij}K_{jn}+{\displaystyle \frac{1}{(2\pi \alpha ^{})^2}}{\displaystyle \frac{n^2}{2}}Y_n^iG_{ij}Y_n^j\right\},`$ (11)
$`X^i(\sigma )`$ $`=`$ $`x^i+\theta _{NC}^{ij}p_j\left(\sigma {\displaystyle \frac{\pi }{2}}\right)+\sqrt{2}{\displaystyle \underset{n=1}{}}\left(Y_n^i\mathrm{cos}n\sigma +{\displaystyle \frac{1}{n}}\theta _{NC}^{ij}K_{jn}\mathrm{sin}n\sigma \right)`$ (12)
where $`Y_n^i`$ and $`K_n^i`$ satisfy the usual commutation relation
$`[Y_n^i,Y_m^j]=0,[Y_n^i,K_{jm}]=i\delta ^i{}_{j}{}^{}\delta _{nm}^{},[K_{in},K_{jm}]=0`$ (13)
and
$`\theta _{NC}^{ij}`$ $`=`$ $`(2\pi \alpha ^{})^2\left({\displaystyle \frac{1}{g+2\pi \alpha ^{}B}}B{\displaystyle \frac{1}{g2\pi \alpha ^{}B}}\right)^{ij}`$ (15)
$`(G^1)^{ij}`$ $`=`$ $`\left({\displaystyle \frac{1}{g+2\pi \alpha ^{}B}}g{\displaystyle \frac{1}{g2\pi \alpha ^{}B}}\right)^{ij}.`$ (16)
Now let us turn to the ferminonic part. The NSR string has two sectors, depending on the boundary conditions for the fermionic variables: The Ramond sector with periodic boundary condition
$`\psi ^{i+}(\sigma )={\displaystyle \underset{n}{}}\psi _n^{i+}e^{in\sigma },\psi ^i(\sigma )={\displaystyle \underset{n}{}}\psi _n^ie^{in\sigma },`$ (17)
and the Neveu-Schwarz sector with anti-periodic boundary condition
$`\psi ^{i+}(\sigma )={\displaystyle \underset{n}{}}\psi _{n+1/2}^{i+}e^{i(n+1/2)\sigma },\psi ^i(\sigma )={\displaystyle \underset{n}{}}\psi _{n+1/2}^ie^{i(n+1/2)\sigma }`$ (18)
where $`n𝐙`$. Since the constraint Eq.(9) is linear in the fermionic variables, it is compatible with these boundary conditions.
We discuss the Ramond sector first. The canonical analysis of the Neveu-Schwarz sector is not much different from that of the Ramond sector. In the Ramond sector the fermionic part of the action in the normal modes is written as
$`L_F`$ $`=`$ $`i{\displaystyle \underset{n}{}}\left(\psi _n^{i+}g_{ij}\dot{\psi }_n^{j+}+\psi _n^ig_{ij}\dot{\psi }_n^j\right)H,`$ (20)
$`H_F`$ $`=`$ $`{\displaystyle \underset{n}{}}n\left(\psi _n^{i+}E_{ij}\psi _n^{j+}+\psi _n^iE_{ij}^T\psi _n^j\right).`$ (21)
We can get the Poisson bracket from the this canonical form as
$`\{\psi _n^{i+},\psi _m^{j+}\}_{PB}=i(g^1)^{ij}\delta (n+m),\{\psi _n^i,\psi _m^j\}_{PB}=i(g^1)^{ij}\delta (n+m).`$ (22)
The boundary conditions accordingly read as
$`\phi _{0i}`$ $`=`$ $`E_{ij}{\displaystyle \underset{n}{}}\psi _n^{j+}E_{ij}^T{\displaystyle \underset{n}{}}\psi _n^j=0,`$ (24)
$`\overline{\phi }_{0i}`$ $`=`$ $`E_{ij}{\displaystyle \underset{n}{}}(1)^n\psi _n^{j+}E_{ij}^T{\displaystyle \underset{n}{}}(1)^n\psi _n^j=0.`$ (25)
We may choose the boundary condition Eq.(24) as a primary constraint. Then the Dirac procedure requires to introduce the following secondary constraint in order to be consistent
$`[H,\phi _i]_{PB}=i{\displaystyle \underset{n}{}}n\left(E_{ij}\psi _n^{j+}E_{ij}^T\psi _n^j\right)=0.`$ (26)
It yields the secondary constraint as
$`\phi _{1i}={\displaystyle \underset{n}{}}n\left(E_{ij}\psi _n^{j+}E_{ij}^T\psi _n^j\right)=0.`$ (27)
Then the Dirac procedure further requires
$`[H,\phi _{1i}]_{PB}=0`$ (28)
It leads us to another secondary constraint
$`\phi _{2i}={\displaystyle \underset{n}{}}n^2\left(E_{ij}\psi _n^{j+}E_{ij}^T\psi _n^j\right)=0.`$ (29)
We may repeat this procedure until no new secondary constraints are generated. By repetition we get
$`\phi _{mi}={\displaystyle \underset{n}{}}n^m\left(E_{ij}\psi _n^{j+}E_{ij}^T\psi _n^j\right)=0,m=1,2,\mathrm{}.`$ (30)
Since the obtained constraints are of second class, we should introduce the Dirac bracket. However, it may not be convenient to construct the Dirac bracket with this set of constraints, $`\{\phi _{mi}=0,m=1,2,3,\mathrm{}\}`$. As in the case of the bosonic string theory, the following observation turns out to be very useful: We can easily disentangle the set of constraints and find that they are equivalent to
$`\left\{E_{ij}\psi _n^{j+}E_{ij}^T\psi _n^j=0,n𝐙\right\}.`$ (31)
(We may also take the boundary condition, Eq.(25), imposed on the other end of the open superstring. But it generates the same set of the constraints, Eq.(31); it is redundant.)
The fermionic degrees of freedom are halved by the set of conditions Eq.(31). They reduce to
$`\left\{\psi _n^{i+}\psi _n^i=0,n𝐙\right\}`$ (32)
when the NS $`B`$-field is absent. In the case of the free superstring theory we get rid of $`\psi _n^i`$ in favor of $`\psi _n^{i+}`$ and choose $`\{\psi _n^{i+}\}`$ as a proper basis for the fermionic degrees of freedom. Suppose that we choose $`\{\psi _n^{i+}\}`$ as the basis for the fermionic degrees of freedom in the present case. If we make use of the constraints, $`\psi _n^i=((E^T)^1E)^i{}_{j}{}^{}\psi _{n}^{j+}`$, we find that the fermionic part of the Lagrangian is written as
$`L_F`$ $`=`$ $`i{\displaystyle \underset{n}{}}\left(\psi _n^{i+}g_{ij}\dot{\psi }_n^{j+}\right)H_F,`$ (33)
$`H_F`$ $`=`$ $`{\displaystyle \underset{n}{}}n\left(\psi _n^{i+}g_{ij}\psi _n^{j+}\right)`$ (34)
where we make use of
$`E^TE^1g(E^1)^TE=E^TG^1E=g.`$ (35)
Here $`\psi ^{i+}`$ is scaled as $`\psi ^{i+}\psi ^{i+}/\sqrt{2}`$. At first appearance the Lagrangian looks same as that of the free superstring and the NS $`B`$-field does not affect the fermionic part. However, this conclusion is misleading. If the constraint conditions are imposed the bosonic part respects the effective metric $`G`$ instead of $`g`$. If the fermionic part still respects the metric $`g`$ after imposing the constraints, the super-Virasoro algebra would not be consistent. We will discuss this point in the next section in some detail.
## III Supersymmetry and Noncommutative Geometry
Let us recall the bosonic part of the Hamiltonian for the open string in the background of NS $`B`$-field. If the background NS $`B`$-field is constant, the Hamiltonian is given as
$`H^B`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left(L_0^B+\overline{L}_0^B\right),`$ (37)
$`L_0^B`$ $`=`$ $`(2\pi \alpha ^{})p_{Li}(g^1)^{ij}p_{Lj}+(2\pi \alpha ^{}){\displaystyle \underset{n=1}{}}nA_{in}(g^1)^{ij}A_{jn}^{},`$ (38)
$`\overline{L}_0^B`$ $`=`$ $`(2\pi \alpha ^{}){\displaystyle \frac{1}{2}}p_{Ri}(g^1)^{ij}p_{Rj}+(2\pi \alpha ^{}){\displaystyle \underset{n=1}{}}n\overline{A}_{in}(g^1)^{ij}\overline{A}_{jn}^{},`$ (39)
where
$`p_{Li}`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2}}}\left(p_i{\displaystyle \frac{1}{2\pi \alpha ^{}}}E_{ij}^Ta^j\right),`$ (41)
$`p_{Ri}`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2}}}\left(p_i+{\displaystyle \frac{1}{2\pi \alpha ^{}}}E_{ij}a^j\right).`$ (42)
The left and right movers are defined as
$`A_{in}^{}`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2n}}}\left(P_{in}+{\displaystyle \frac{in}{2\pi \alpha ^{}}}E_{ij}^TX_n^j\right),`$ (44)
$`A_{in}`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2n}}}\left(P_{in}{\displaystyle \frac{in}{2\pi \alpha ^{}}}E_{ij}^TX_n^j\right),`$ (45)
$`\overline{A}_{in}^{}`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2n}}}\left(P_{in}+{\displaystyle \frac{in}{2\pi \alpha ^{}}}E_{ij}X_n^j\right),`$ (46)
$`\overline{A}_{in}`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2n}}}\left(P_{in}{\displaystyle \frac{in}{2\pi \alpha ^{}}}E_{ij}X_n^j\right),`$ (47)
and their commutation relations are
$`[A_{in},A_{jm}^{}]=(2\pi \alpha ^{})^1g_{ij}\delta _{nm},[\overline{A}_{in},\overline{A}_{jm}^{}]=(2\pi \alpha ^{})^1g_{ij}\delta _{nm}.`$ (48)
Using the constraint conditions
$`a^i=\theta ^{ij}p_j,\overline{Y}_n^i={\displaystyle \frac{1}{\sqrt{2}}}(X_n^iX_n^i)={\displaystyle \frac{1}{n}}\theta ^{ij}K_{jn},\overline{K}_{in}={\displaystyle \frac{1}{\sqrt{2}}}(P_n^iP_n^i)=0,`$ (49)
we may remove $`a^i`$, $`\overline{Y}_n^i`$, $`\overline{K}_{in}`$ in favor of $`(x^i,p_i)`$, $`(Y_n^i,K_{in})`$, which are canonical variables for the open string. As a result we find that the bosonic part of the Hamiltonian is just the same as the free Hamiltonian for the open string on the target space with metric $`G`$
$`H^B`$ $`=`$ $`(2\pi \alpha ^{}){\displaystyle \frac{1}{2}}\left(p^2+{\displaystyle \underset{n=1}{}}nA_{in}(E^{})\left(G^1\right)^{ij}A_{jn}^{}(E^{})\right),`$ (51)
$`A_{in}(E^{})`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2n}}}\left(K_{in}{\displaystyle \frac{in}{2\pi \alpha ^{}}}G_{ij}Y_n^j\right),`$ (52)
$`A_{in}^{}(E^{})`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2n}}}\left(K_{in}+{\displaystyle \frac{in}{2\pi \alpha ^{}}}GijY_n^j\right),`$ (53)
where the left movers and the right movers satisfy
$`[A_{in}(E^{}),A_{jm}^{}(E^{})]=(2\pi \alpha ^{})^1G_{ij}\delta (n+m).`$ (54)
It is interesting to note that the left and right movers $`A^{}(E)`$, $`\overline{A}^{}(E)`$ are related to the left and right movers, $`A^{}(E^{})`$, $`\overline{A}^{}(E^{})`$ by a T-dual transformation
$`T=\left(\begin{array}{cc}I& 0\\ (2\pi \alpha ^{})^1\theta & I\end{array}\right).`$ (57)
We may have geometric interpretation of this T-dual transformation as discussed in ref. .
Now let us turn to the fermionic constraint $`F_0`$, which forms the super-algebra with the Hamiltonian $`H`$,
$`\{F_0,F_0\}+\{\overline{F}_0,\overline{F}_0\}=2\left(L_0+L_0^F\right)+2\left(\overline{L}_0+\overline{L}_0^F\right)=2\left(H^B+H^F\right).`$ (58)
In the presence of the NS $`B`$-field the fermionic constraint $`F_0`$ is given as
$`(2\pi \alpha ^{})^{\frac{1}{2}}F_0`$ $`=`$ $`p_{Li}\psi _0^{i+}+{\displaystyle \underset{n=1}{}}\sqrt{n}\left(A_{in}\psi _n^{i+}+A_{in}^{}\psi _n^{i+}\right),`$ (60)
$`(2\pi \alpha ^{})^{\frac{1}{2}}\overline{F}_0`$ $`=`$ $`p_{Ri}\psi _0^i+{\displaystyle \underset{n=1}{}}\sqrt{n}\left(\overline{A}_{in}\psi _n^i+\overline{A}_{in}^{}\psi _n^i\right).`$ (61)
From the canonical analysis of the bosonic part we expect that the fermionic constraint is rewritten as
$`(2\pi \alpha ^{})^{\frac{1}{2}}\left(F_0+\overline{F}_0\right)=p_i\widehat{\psi }^i+{\displaystyle \underset{n=1}{}}\sqrt{n}\left(A_{in}^{}(E^{})\widehat{\psi }_n^i+A_{in}(E^{})\widehat{\psi }_n^i\right)`$ (62)
where the fermion operator $`\widehat{\psi }_n^i`$ satisfies
$`\{\widehat{\psi }_n^i,\widehat{\psi }_m^j\}_{PB}=(G^1)^{ij}\delta (n+m).`$ (63)
It follows then that the fermion operator $`\widehat{\psi }_n^i`$ may be defined as
$`\widehat{\psi }_n^i={\displaystyle \frac{1}{\sqrt{2}}}\left(\psi _n^{i+}+\psi _n^i\right)=\sqrt{2}(G^1E)^i{}_{j}{}^{}\psi _{}^{j+}=\sqrt{2}(G^1E^T)^i{}_{j}{}^{}\psi _{}^{j}.`$ (64)
Rewriting the fermionic part of the Lagrangian Eq.(II) as
$`L_F`$ $`=`$ $`i{\displaystyle \underset{n}{}}\left(\widehat{\psi }_n^iG_{ij}_\tau \widehat{\psi }_n^j\right)H_F,`$ (66)
$`H_F`$ $`=`$ $`{\displaystyle \underset{n}{}}n\left(\widehat{\psi }_n^iG_{ij}\widehat{\psi }_n^j\right),`$ (67)
we confirm that the fermionic part also respects the effective open string metric $`G`$. It is consistent with the commutation relation among $`\{\widehat{\psi }_n^i\}`$ and the supersymmetry.
Being equipped with the canonical analysis of the fermionic part, we discuss the noncommutativity in the superspace
$`Z^i(\sigma )=X^i(\sigma )+\theta _+{\displaystyle \underset{n}{}}\psi _n^ie^{in\sigma }+\theta _{}{\displaystyle \underset{n}{}}\psi _n^{i+}e^{in\sigma }+\theta _+\theta _{}{\displaystyle \underset{n}{}}F^ie^{in\sigma }.`$ (68)
We may rewrite the fermionic part of $`Z^i`$ in terms of $`\widehat{\psi }_n^i`$ as
$`Z_F^i`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2}}}{\displaystyle \underset{n}{}}\left\{\left(\theta \mathrm{cos}n\sigma i\overline{\theta }\mathrm{sin}n\sigma \right)\widehat{\psi }_n^i+i{\displaystyle \frac{\left(\theta _{NC}G\right)^i_j}{2\pi \alpha ^{}}}\left(\theta \mathrm{sin}n\sigma +i\overline{\theta }\mathrm{cos}n\sigma \right)\widehat{\psi }_n^j\right\}`$ (69)
where
$`\theta ={\displaystyle \frac{1}{\sqrt{2}}}(\theta _++\theta _{}),\overline{\theta }={\displaystyle \frac{1}{\sqrt{2}}}(\theta _+\theta _{}),\overline{\theta }\theta =\theta _+\theta _{}.`$ (70)
Note that the end points of the string are no longer holomorphic in $`\theta `$
$`Z_F^i(0)`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2}}}{\displaystyle \underset{n}{}}\left(\theta \widehat{\psi }_n^i\overline{\theta }{\displaystyle \frac{\left(\theta _{NC}G\right)^i_j}{2\pi \alpha ^{}}}\widehat{\psi }_n^j\right),`$ (72)
$`Z_F^i(\pi )`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2}}}{\displaystyle \underset{n}{}}(1)^n\left(\theta \widehat{\psi }_n^i\overline{\theta }{\displaystyle \frac{\left(\theta _{NC}G\right)^i_j}{2\pi \alpha ^{}}}\widehat{\psi }_n^j\right).`$ (73)
It may have some consequences in construction of the vertex operators. The noncommutativity can be easily seen if we evaluate the commutator between $`Z_F^i`$
$`[Z^i(\sigma ),Z^j(\sigma ^{})]=[X^i(\sigma ),X^j(\sigma ^{})]+[Z_F^i(\sigma ),Z_F^j(\sigma ^{})].`$ (74)
The noncommutativity in the bosonic sector is discussed in details in the previous works . The commutation relation between $`Z_F^i(\sigma )`$ is found to be
$`[Z_F^i(\sigma ),Z_F^j(\sigma ^{})]`$ $`=`$ $`\{\begin{array}{cc}\hfill \frac{i}{2\pi \alpha ^{}}(\theta \overline{\theta })\theta _{NC}^{ij}:& \sigma =\sigma ^{}=0,\hfill \\ \hfill \frac{i}{2\pi \alpha ^{}}(\overline{\theta }\theta )\theta _{NC}^{ij}:& \sigma =\sigma ^{}=\pi \hfill \\ \hfill 0:& \mathrm{otherwise}\hfill \end{array}`$ (78)
where we use $`\zeta (0)=_n1=1/2`$.
The canonical analysis of the Neveu-Schwarz Sector is obtained as we replace the integer modes of the fermion variables by the half-integer modes, $`\psi _n\psi _{n+1/2}`$.
## IV Concluding Remarks
A few remarks are in order to conclude the paper. The first one is a brief summary of the present work. We perform the canonical quantization of the NSR open superstring attached on the $`D`$-brane with a NS $`B`$-field. The open superstring has fermionic boundary conditions to be imposed on the ends of the string in addition to the bosonic ones. Taking the fermionic boundary conditions as primary ones, we obtain a set of infinite secondary constraints, which turn out easy to solve. Choosing an appropriate basis for the fermion variables, we find that the fermion variables also respect the effective metric $`G`$ and the fermionic part of the Hamiltonian is just same as that of the free open string Hamiltonian defined on the target space with the effective metric $`G`$. Thus, the canonical analysis of the bosonic part is extended to the fermionic part explicitly. The interesting point we should note is that the superspace coordinate $`Z^i`$ is no longer holomorphic in $`\theta `$ at the end points of the string. We expect that this has some important consequences when we construct the vertex operators representing physical states.
The present canonical analysis suggests a number of interesting directions to explore along the line of this work. We may construct the vertex operator for emission of a scalar to study the recent issues associated with the noncommutative field theories in the context of superstring theory. The supersymmetric Dirac-Born-Infeld action may be derived if we construct the vertex operator for emission of a massless vector and evaluate the Polyakov string path integral over a disk on the D-brane word sheet. It would also bring us to the Seiberg-Witten map in the context of the supersymmetric noncommutative field theory. The T-duality also deserves further study and the Morita equivalence may be extended to the supersymmetric theory. After all these related subjects may be discussed in a single framework of the second quantized open superstring theory, if properly constructed. The canonical analysis presented here would be certainly useful to develop the second quantized open superstring theory on the noncommutative geometry.
## Acknowledgement
This work was supported in part by KOSEF (995-0200-005-2). Part of the work was done during the author’s visit to KIAS.
|
warning/0004/gr-qc0004079.html
|
ar5iv
|
text
|
# Large-distance behaviour of the graviton two-point function in de Sitter spacetime
## 1 Introduction
De Sitter spacetime is a maximally symmetric solution of the vacuum Einstein equations with positive cosmological constant,
$$R_{ab}\frac{1}{2}g_{ab}R+\mathrm{\Lambda }g_{ab}=0.$$
(1)
(See Ref. for a detailed description of de Sitter spacetime.) Physics in this spacetime has been studied extensively due to its relevance to inflationary cosmologies . The graviton two-point function has been of particular interest in this context. Ford and Parker analysed linearised gravity in spatially-flat de Sitter spacetime and found that the mode functions in a physical gauge are similar to those of minimally-coupled massless scalar field . Since the latter theory exhibits infrared (IR) divergences similar to those of massless scalar field theory in two-dimensional Minkowski spacetime , one may suspect that there would be IR divergences in linearised gravity in de Sitter spacetime as well. However, it was shown that there are no physical IR divergences in the graviton two-point function in de Sitter spacetime which was obtained by analytic continuation from that on the 4-sphere . It was also found that the behaviour of mode functions responsible for the apparent IR divergences in spatially-flat de Sitter spacetime is a gauge artifact , and it was shown explicitly that the IR divergences in the graviton two-point function in a physical gauge can be gauged away . Thus, it has been established that there are no physical IR divergences in linearised gravity in de Sitter spacetime. However, there is another apparent problem which is closely related: the graviton two-point function grows logarithmically with distance. If this behaviour were physical, then it would have a significant effect on physics of inflationary cosmologies.
The aim of this paper is to show that this logarithmic growth is also a gauge artifact. Specifically, we show that the large-distance logarithmic behaviour of the physical graviton two-point function computed by Allen can be gauged away by demonstrating that the same behaviour arises in the two-point function of pure-gauge form. The rest of the paper is organised as follows. In section 2 we present the graviton two-point function in a physical gauge and show that its logarithmically growing part can be gauged away. In section 3 we show that this logarithmic behaviour can be reproduced by a pure-gauge field obeying a relativistic field equation. We summarise our results and make some remarks in section 4. In Appendix A, we list some integrals used in this work. Appendix B contains details of the calculations in section 3. We adopt the metric signature $`(+++)`$ and set $`\mathrm{}=c=16\pi G=1`$ throughout this paper.
## 2 The physical graviton two-point function
We work with the metric which covers half of de Sitter spacetime:
$$ds^2=\frac{1}{H^2\lambda ^2}(d\lambda ^2+dx^2+dy^2+dz^2),$$
(2)
where $`H^2=\mathrm{\Lambda }/3`$, i.e. $`g_{ab}=(H\lambda )^2\mathrm{diag}(1,1,1,1)`$. In the expanding half of de Sitter spacetime, the parameter $`\lambda `$ takes positive values and decreases from $`\mathrm{}`$ to $`0`$ towards the future. By letting $`g_{ab}=g_{ab}^{(0)}+h_{ab}`$, where $`g_{ab}^{(0)}`$ is the de Sitter metric (2), and linearising the Hilbert-Einstein Lagrangian density, we have for linearised gravity
$``$ $`=`$ $`\sqrt{g^{(0)}}[{\displaystyle \frac{1}{2}}_ah^{ac}^bh_{bc}{\displaystyle \frac{1}{4}}_ah_{bc}^ah^{bc}+{\displaystyle \frac{1}{4}}(^ah2^bh_b^a)_ah`$ (3)
$`{\displaystyle \frac{1}{2}}H^2(h_{ab}h^{ab}+{\displaystyle \frac{1}{2}}h^2)],`$
where the covariant derivative $`_a`$ is compatible with the background metric $`g_{ab}^{(0)}`$. Indices are raised and lowered by $`g_{ab}^{(0)}`$ and $`h`$ is the trace of $`h_{ab}`$. (We will denote $`g_{ab}^{(0)}`$ by $`g_{ab}`$ from now on.) The Lagrangian density (3) yields the following Euler-Lagrange equation:
$`{\displaystyle \frac{1}{2}}\left(\mathrm{}h_{ab}_a_ch_b^c_b_ch_a^c+_a_bh\right)`$
$`+{\displaystyle \frac{1}{2}}g_{ab}(_c_dh^{cd}\mathrm{}h)H^2\left(h_{ab}+{\displaystyle \frac{1}{2}}g_{ab}h\right)=0,`$ (4)
where $`\mathrm{}=_a^a`$. Equation (4) is invariant under the gauge transformations
$$h_{ab}h_{ab}+_a\mathrm{\Lambda }_b+_b\mathrm{\Lambda }_a.$$
(5)
We will first fix the gauge completely at the classical level.
By imposing the gauge condition,
$$_bh^{ab}\frac{1}{2}^ah=0,$$
(6)
we find from (4)
$$\frac{1}{2}\mathrm{}h_{ab}\frac{1}{4}g_{ab}\mathrm{}hH^2\left(h_{ab}+\frac{1}{2}g_{ab}h\right)=0.$$
(7)
The trace $`h`$ can be gauged away as follows. By taking the trace of (7) we have
$$(\mathrm{}+6H^2)h=0.$$
(8)
Using this equation, we can replace $`h_{ab}`$ by a traceless field satisfying (7),
$$\stackrel{~}{h}_{ab}=h_{ab}+\frac{1}{6H^2}_a_bh,$$
(9)
which is gauge-equivalent to the original field $`h_{ab}`$. Thus, the trace $`h`$ can be gauged away.
The field $`\stackrel{~}{h}_{ab}`$, which we will denote by $`h_{ab}`$ from now on, is transverse-traceless, (i.e. it satisfies $`^bh_{ab}=h_c^c=0`$) and obeys the following equation:
$$(\mathrm{}2H^2)h_{ab}=0.$$
(10)
The general solution of this equation can be found, e.g. in . Now, this equation allows solutions which are pure gauge: the field
$$h_{ab}^{(\xi )}=_a\xi _b+_b\xi _a$$
(11)
is transverse-traceless and satisfies (10) if $`^c\xi _c=0`$ and
$$(\mathrm{}+3H^2)\xi _a=0.$$
(12)
This gauge freedom allows us to fix the gauge further and in effect we can impose the following gauge conditions on the field $`h_{ab}`$:
$`\mathrm{traceless}:`$ $`h_{}^{c}{}_{c}{}^{}=0;`$
$`\mathrm{transverse}:`$ $`_bh^{ab}=0;`$
$`\mathrm{synchronous}:`$ $`t^ah_{ab}=0,`$
where $`t^a`$ is the future-pointing unit vector parallel to $`(/\lambda )^a`$.
Allen considered quantisation of linearised gravity in this gauge, which we call the physical gauge, and computed the symmetrised two-point function. Here, we present essentially the same results for the unsymmetrised two-point function, $`G_{aba^{}b^{}}(x,x^{})=0|h_{ab}(x)h_{a^{}b^{}}(x^{})|0`$, where the state $`|0`$ is the so-called Euclidean vacuum . The unprimed indices refer to the spacetime point $`x`$, whereas the primed indices refer to the spacetime point $`x^{}`$.
Following Allen, we define the projection operator $`P_{ab}=g_{ab}+t_at_b`$ at point $`x`$, which projects tensors onto the flat spatial section of constant $`\lambda `$. The tensor $`P_{ab}`$ is the metric tensor on this spatial section. In our coordinate system it has the form
$$P_{ab}=(H\lambda )^2\mathrm{diag}(0,1,1,1).$$
(13)
We define $`P_{a^{}b^{}}`$ to be the same projection operator at point $`x^{}`$. We define a bi-covector $`P_{ab^{}}`$ by
$$P_{ab^{}}=(H^2\lambda \lambda ^{})^1\mathrm{diag}(0,1,1,1).$$
(14)
Next we define the comoving spatial separation $`r`$ of two points $`x=(\lambda ,x_1,x_2,x_3)`$ and $`x^{}=(\lambda ^{},x_1^{},x_2^{},x_3^{})`$ as
$$r(x,x^{})=\sqrt{(x_1x_1^{})^2+(x_2x_2^{})^2+(x_3x_3^{})^2}.$$
(15)
For any given two points $`x`$ and $`x^{}`$ the vectors $`V^a`$ and $`V^a^{}`$ are defined in components as
$$V^a=\frac{H\lambda }{r}(0,𝐱𝐱^{}),V^a^{}=\frac{H\lambda ^{}}{r}(0,𝐱^{}𝐱).$$
(16)
The vector $`V^a`$ and $`V^a^{}`$ are the unit vectors at points $`x`$ and $`x^{}`$, respectively, which is parallel to the projection of the tangent vector to the geodesic joining two points $`x`$ and $`x^{}`$ onto the constant $`\lambda `$ hypersurface.
The field $`h_{ab}(\lambda ,𝐱)`$ has the following mode expansion :
$$h_{ab}(\lambda ,𝐱)=d^3𝐤\underset{s=1}{\overset{2}{}}[b^{(s)}(𝐤)\frac{H}{4\sqrt{2}\pi }\lambda ^{3/2}H_{3/2}^{(1)}(k\lambda )\widehat{H}_{ab}^{(𝐤,s)}e^{i𝐤𝐱}+\mathrm{h}.\mathrm{c}.],$$
(17)
where the symmetric traceless tensors $`\widehat{H}_{ab}^{(𝐤,s)}`$ satisfy $`\widehat{H}_{ab}^{(𝐤,s)}\widehat{H}^{(𝐤,s^{})ab}=\delta ^{ss^{}}`$ and $`k^a\widehat{H}_{ab}^{(𝐤,s)}=t^a\widehat{H}_{ab}^{(𝐤,s)}=0`$. We have defined $`k=𝐤`$. Here, $`d^3𝐤=dk_1dk_2dk_3`$ and $`𝐤𝐱=k_1x_1+k_2x_2+k_3x_3`$. The Hankel function $`H_{3/2}^{(1)}(z)`$ is given by
$$H_{3/2}^{(1)}(z)=\sqrt{\frac{2}{\pi }}\left(\frac{i}{z^{3/2}}\frac{1}{z^{1/2}}\right)e^{iz}.$$
(18)
The operators $`b^{(s)}(𝐤)`$ and $`b^{(s)}(𝐤)^{}`$ satisfy the commutation relations
$$[b^{(s)}(𝐤),b^{(s^{})}(𝐤^{})^{}]=\delta ^{ss^{}}\delta ^3(𝐤𝐤^{}),$$
(19)
with all other commutators vanishing. The Euclidean vacuum $`|0`$ is defined by $`b^{(s)}(𝐤)|0=0`$ for all $`𝐤`$ and $`s`$. By using the mode expansion (17) and remembering that the two-point function is a maximally symmetric bi-tensor in the spatial sections, we find
$$G_{aba^{}b^{}}(x,x^{})=f_1(\lambda ,\lambda ^{},r)\theta _{aba^{}b^{}}^{(1)}+f_2(\lambda ,\lambda ^{},r)\theta _{}^{(2)}{}_{aba^{}b^{}}{}^{}+f_3(\lambda ,\lambda ^{},r)\theta _{}^{(3)}{}_{aba^{}b^{}}{}^{}$$
(20)
where the bi-tensors $`\theta _{}^{(i)}{}_{aba^{}b^{}}{}^{}`$ are given by
$`\theta _{}^{(1)}{}_{aba^{}b^{}}{}^{}`$ $`=`$ $`(V_aV_b{\displaystyle \frac{1}{3}}P_{ab})(V_a^{}V_b^{}{\displaystyle \frac{1}{3}}P_{a^{}b^{}})`$ (21)
$`\theta _{}^{(2)}{}_{aba^{}b^{}}{}^{}`$ $`=`$ $`P_{aa^{}}P_{bb^{}}+P_{ba^{}}P_{ab^{}}{\displaystyle \frac{2}{3}}P_{ab}P_{a^{}b^{}}`$ (22)
$`\theta _{}^{(3)}{}_{aba^{}b^{}}{}^{}`$ $`=`$ $`4V_aV_bV_a^{}V_b^{}+P_{aa^{}}V_bV_b^{}+P_{ba^{}}V_aV_b^{}+P_{ab^{}}V_bV_a^{}+P_{bb^{}}V_aV_a^{}.`$ (23)
The functions $`f_1`$, $`f_2`$ and $`f_3`$ are given by
$`f_1`$ $`=`$ $`{\displaystyle \frac{H^2}{8\pi ^2}}\left[{\displaystyle \frac{3}{4}}V^2(V^23)\psi _2{\displaystyle \frac{1}{5}}\left(15V^440V^212\right)\right]`$ (24)
$`+{\displaystyle \frac{H^2\lambda \lambda ^{}}{8\pi ^2r^2}}\left[{\displaystyle \frac{3}{4}}(5V^29)\psi _2+{\displaystyle \frac{15V^437V^2+16}{1V^2}}\right],`$
$`f_2`$ $`=`$ $`{\displaystyle \frac{H^2}{8\pi ^2}}\left[{\displaystyle \frac{1}{5}}\psi _1+{\displaystyle \frac{1}{20}}V^2(V^2+5)\psi _2{\displaystyle \frac{1}{75}}(15V^4+80V^232)\right]`$ (25)
$`+{\displaystyle \frac{H^2\lambda \lambda ^{}}{8\pi ^2r^2}}\left[{\displaystyle \frac{1}{4}}(V^2+3)\psi _2+{\displaystyle \frac{3V^4+7V^24}{3(1V^2)}}\right],`$
$`f_3`$ $`=`$ $`{\displaystyle \frac{H^2}{8\pi ^2}}\left[{\displaystyle \frac{1}{4}}V^2(V^2+1)\psi _2{\displaystyle \frac{1}{15}}(15V^4+20V^2+8)\right]`$ (26)
$`+{\displaystyle \frac{H^2\lambda \lambda ^{}}{8\pi ^2r^2}}\left[{\displaystyle \frac{1}{4}}(5V^2+3)\psi _2+{\displaystyle \frac{15V^4V^28}{3(1V^2)}}\right],`$
where
$$V=\frac{\lambda \lambda ^{}+iϵ}{r}$$
(27)
and
$`\psi _1`$ $`=`$ $`\mathrm{log}[\alpha ^4r^4(1V^2)^2]+4\gamma ,`$ (28)
$`\psi _2`$ $`=`$ $`V\mathrm{log}\left({\displaystyle \frac{V+1}{V1}}\right)^2.`$ (29)
Here, $`\gamma `$ is Euler’s constant and $`\alpha (>0)`$ is an infrared cut-off. Since these results are essentially the same as those in Ref. — recall that we have set $`16\pi G=1`$ — we omit the details of the calculations. The method is similar to that used in the next section.
In the large-$`r`$ limit these functions become
$`f_1`$ $`=`$ $`{\displaystyle \frac{3H^2}{10\pi ^2}}+O(r^2),`$ (30)
$`f_2`$ $`=`$ $`{\displaystyle \frac{H^2}{40\pi ^2}}\left({\displaystyle \frac{32}{15}}2\mathrm{log}\alpha ^2r^24\gamma \right)+O(r^2),`$ (31)
$`f_3`$ $`=`$ $`{\displaystyle \frac{H^2}{15\pi ^2}}+O(r^2).`$ (32)
Notice that the two-point function $`G_{aba^{}b^{}}`$ is IR divergent and grows logarithmically with distance due to the behaviour of the function $`f_2`$. However, this behaviour can be a gauge artifact because linearised gravity has gauge invariance (5). The two-point function $`G_{aba^{}b^{}}`$ is physically equivalent to $`G_{aba^{}b^{}}^{\mathrm{mod}}`$ if
$$G_{aba^{}b^{}}(x,x^{})=G_{aba^{}b^{}}^{\mathrm{mod}}(x,x^{})+_a_a^{}T_{bb^{}}+_b_a^{}T_{ab^{}}+_a_b^{}T_{ba^{}}+_b_b^{}T_{aa^{}}.$$
(33)
If there is a field $`T_{aa^{}}`$ such that $`_a_a^{}T_{bb^{}}+_b_a^{}T_{ab^{}}+_a_b^{}T_{ba^{}}+_b_b^{}T_{aa^{}}`$ contains a term proportional to $`(H^2/20\pi ^2)\mathrm{log}\alpha ^2r^2\times \theta _{aba^{}b^{}}^{(2)}`$, then the modified two-point function $`G_{aba^{}b^{}}^{\mathrm{mod}}`$ will be IR finite and has no logarithmic growth with distance $`r`$. Allen found that the IR divergence could be gauged away in a similar manner. However, it is not difficult to show that the logarithmic growth with distance $`r`$ can be gauged away together with the IR divergence. Indeed we find
$`\mathrm{log}\alpha ^2r^2\times \theta _{aba^{}b^{}}^{(2)}`$ $`=`$ $`{\displaystyle \frac{1}{6}}\left(_a_a^{}K_{bb^{}}+_a_b^{}K_{ba^{}}+_b_a^{}K_{ab^{}}+_b_b^{}K_{aa^{}}\right)`$ (34)
$`+{\displaystyle \frac{1}{3}}(P_{aa^{}}V_bV_b^{}+P_{a^{}b}V_aV_b^{}+P_{a^{}b}V_a^{}V_b+P_{bb^{}}V_aV_a^{}`$
$`+4P_{ab}V_a^{}V_b^{}+4P_{a^{}b^{}}V_aV_b4P_{aa^{}}P_{bb^{}}4P_{ab^{}}P_{a^{}b}`$
$`8V_aV_bV_a^{}V_b^{})`$
with
$$K_{aa^{}}=\frac{r^2}{H^2\lambda \lambda ^{}}(V_aV_a^{}+2P_{aa^{}})\mathrm{log}\alpha ^2r^2.$$
(35)
Therefore we have
$`G_{aba^{}b^{}}(x,x^{})`$ $`=`$ $`G_{aba^{}b^{}}^{\mathrm{mod}}(x,x^{})`$ (36)
$`+{\displaystyle \frac{H^2}{120\pi ^2}}\left(_a_a^{}K_{bb^{}}+_b_a^{}K_{ab^{}}+_a_b^{}K_{ba^{}}+_b_b^{}K_{aa^{}}\right),`$
where $`G_{aba^{}b^{}}^{\mathrm{mod}}(x,x^{})`$ does not grow logarithmically as the function of the distance between the two points $`x`$ and $`x^{}`$ and is IR finite.
This proves that the $`\mathrm{log}r`$ behaviour of the two-point function $`G_{aba^{}b^{}}`$ is a gauge artifact. However, it may be desirable to use the two-point function $`0|_{(a}\xi _{b)}(x)_{(a^{}}\xi _{b^{})}(x^{})|0`$ of a vector field $`\xi _a`$ for gauging away the $`\mathrm{log}\alpha ^2r^2`$ term in order to have a better understanding of the logarithmic growth. In the next section we show how this can be done.
## 3 Pure-gauge two-point function
Recall that the tensor $`h_{ab}^{(\xi )}=_a\xi _b+_b\xi _a`$ given by (11) satisfies $`^bh_{ab}^{(\xi )}=h_a^{(\xi )a}=0`$ and $`(\mathrm{}2H^2)h_{ab}^{(\xi )}=0`$. These are the equations satisfied by $`h_{ab}`$ in the physical gauge. Hence, it is natural to expect that the two-point function of $`h_{ab}^{(\xi )}`$ has a structure similar to $`G_{aba^{}b^{}}`$. We will find that this is indeed the case and that the $`\mathrm{log}\alpha ^2r^2`$ term in the two-point function $`G_{aba^{}b^{}}`$ can be gauged away, in the manner described at the end of the previous section, using the two-point function of $`h_{ab}^{(\xi )}`$ with an additional condition $`t^a\xi _a=0`$.
First we note that the transverse solutions to equation (12) satisfying the condition $`t^a\xi _a=0`$ are
$$\xi _a^{(s)}(𝐤,𝐱,\lambda )=\frac{H}{4\sqrt{2}\pi }\lambda ^{3/2}H_{5/2}^{(1)}(k\lambda )e^{i𝐤𝐱}\widehat{h}_a^{(𝐤,s)},$$
(37)
where the polarisation vectors $`\widehat{h}_a^{(𝐤,s)}`$, $`s=1,2`$, are orthogonal to $`k^a`$ and $`t^a`$ and satisfy
$$P^{ab}h_a^{(𝐤,s)}h_b^{(𝐤,s^{})}=\delta ^{ss^{}}.$$
(38)
The Hankel function $`H_{5/2}^{(1)}(z)`$ is given by
$$H_{5/2}^{(1)}(z)=\sqrt{\frac{2}{\pi }}\left(\frac{3i}{z^{5/2}}\frac{3}{z^{3/2}}+\frac{i}{z^{1/2}}\right)e^{iz}.$$
(39)
We expand the field $`\xi _a(x)`$ as
$$\xi _a(𝐱,\lambda )=\underset{s=1}{\overset{2}{}}d^3𝐤\left[c^{(s)}(𝐤)\xi _a^{(s)}(𝐤,𝐱,\lambda )+c^{(s)}(𝐤)^{}\overline{\xi _a^{(s)}(𝐤,𝐱,\lambda )}\right].$$
(40)
We then impose the commutation relations
$$[c^{(s)}(𝐤),c^{(s^{})}(𝐤^{})^{}]=\delta ^{ss^{}}\delta ^3(𝐤𝐤^{}),$$
(41)
with all other commutators being zero. (Since the field $`\xi _a`$ is a fictitious field introduced to rewrite the two-point function $`G_{aba^{}b^{}}`$, there is no need to derive these commutation relations.) Requiring that $`a^{(s)}(𝐤)|0=0`$ for all $`s`$ and $`𝐤`$, we have the two-point function of $`\xi _a`$,
$`M_{aa^{}}(x,x^{})`$ $`=`$ $`0|\xi _a(x)\xi _a^{}(x^{})|o`$ (42)
$`=`$ $`{\displaystyle d^3𝐤\underset{s=1}{\overset{2}{}}\xi _a^{(s)}(𝐤,𝐱,\lambda )\overline{\xi _a^{}^{(s)}(𝐤,𝐱^{},\lambda ^{})}}.`$
By using (37) and (39), we find
$`{\displaystyle \underset{s=1}{\overset{2}{}}}\xi _a^{(s)}(𝐤,𝐱,\lambda )\overline{\xi _a^{}^{(s)}(𝐤,𝐱^{},\lambda ^{})}`$
$`={\displaystyle \frac{H^2E(k,\lambda ,\lambda ^{})}{16\pi ^3\lambda \lambda ^{}}}e^{ik(\lambda \lambda ^{})}e^{i𝐤(𝐱𝐱^{})}{\displaystyle \underset{s=1}{\overset{2}{}}}\widehat{h}_a^{(s)}(𝐤)\widehat{h}_a^{}^{(s)}(𝐤),`$ (43)
where
$`E(k,\lambda ,\lambda ^{})`$ $`=`$ $`9k^59i(\lambda \lambda ^{})k^4+[9\lambda \lambda ^{}3(\lambda ^2+\lambda _{}^{}{}_{}{}^{2})]k^3`$ (44)
$`3i\lambda \lambda ^{}(\lambda \lambda ^{})k^2+\lambda ^2\lambda _{}^{}{}_{}{}^{2}k^1.`$
Using the properties of $`\widehat{h}_a^{(s)}(𝐤)`$, we get
$$\underset{s=1}{\overset{2}{}}\widehat{h}_a^{(s)}(𝐤)\widehat{h}_a^{}^{(s)}(𝐤)=\frac{1}{H^2\lambda \lambda ^{}}\left(\stackrel{~}{\delta }_{aa^{}}\frac{\stackrel{~}{k}_a\stackrel{~}{k}_a^{}}{k^2}\right),$$
(45)
where $`\stackrel{~}{\delta }_{aa^{}}=\lambda \lambda ^{}P_{aa^{}}`$. (In components, $`\stackrel{~}{\delta }_{11}=\stackrel{~}{\delta }_{22}=\stackrel{~}{\delta }_{33}=1`$ and all other components vanish.) We have also defined the spatial part of $`k_a`$ as $`\stackrel{~}{k}_a=P_a^bk_b`$. It is also convenient to define the spacelike projections of the partial derivatives as $`P_a^c_c=\stackrel{~}{}_a`$ and $`P_a^{}^c^{}_c^{}=\stackrel{~}{}_a^{}`$. Combining equations (42)–(45), we obtain
$`M_{aa^{}}(x,x^{})`$ $`=`$ $`{\displaystyle \frac{1}{16\pi ^3\lambda ^2\lambda ^2}}{\displaystyle d^3𝐤E(k,\lambda ,\lambda ^{})e^{ik[(\lambda \lambda ^{})+iϵ]}e^{i𝐤(𝐱𝐱^{})}\left(\stackrel{~}{\delta }_{aa^{}}\frac{\stackrel{~}{k}_a\stackrel{~}{k}_a^{}}{k^2}\right)}`$ (46)
$`=`$ $`{\displaystyle \frac{1}{16\pi ^3\lambda ^2\lambda ^2}}{\displaystyle 𝑑kk^2E(k,\lambda ,\lambda ^{})e^{ik\mathrm{\Delta }\lambda }}`$
$`\times {\displaystyle }d\mathrm{\Omega }_𝐤e^{i𝐤(𝐱𝐱^{})}(\stackrel{~}{\delta }_{aa^{}}{\displaystyle \frac{\stackrel{~}{k}_a\stackrel{~}{k}_a^{}}{k^2}}),`$
where the convergence factor $`e^{kϵ}`$ has been introduced to make the integral converge for large wave numbers $`k`$. We have also defined $`\mathrm{\Delta }\lambda =\lambda \lambda ^{}+iϵ`$, and $`\mathrm{\Omega }_𝐤`$ is the solid angle in the $`𝐤`$ space. Noting that $`\stackrel{~}{}_ae^{i𝐤(𝐱𝐱^{})}=i\stackrel{~}{k}_ae^{i𝐤(𝐱𝐱^{})}`$ and
$$𝑑\mathrm{\Omega }_𝐤e^{i𝐤(𝐱𝐱^{})}=\frac{4\pi \mathrm{sin}kr}{kr},$$
(47)
we obtain
$`M_{aa^{}}(x,x^{})`$ $`=`$ $`{\displaystyle \frac{1}{4\pi ^2\lambda ^2\lambda ^2}}{\displaystyle 𝑑kk^2E(k,\lambda ,\lambda ^{})e^{ik\mathrm{\Delta }\lambda }}`$ (48)
$`\times \left(\stackrel{~}{\delta }_{aa^{}}{\displaystyle \frac{\stackrel{~}{}_a\stackrel{~}{}_a^{}}{k^2}}\right){\displaystyle \frac{\mathrm{sin}kr}{kr}}.`$
We will show that the pure-gauge two-point function,
$$T_{aba^{}b^{}}40|_{(a}\xi _{b)}_{(a^{}}\xi _{b^{})}|0,$$
(49)
has the same $`\mathrm{log}\alpha ^2r^2`$ term as $`G_{aba^{}b^{}}`$. We first note that
$$_a\xi _b+_b\xi _a=_\xi g_{ab}=\xi ^c_cg_{ab}+g_{ac}_b\xi ^c+g_{bc}_a\xi ^c,$$
(50)
where $`_\xi `$ is the Lie derivative with respect to the vector $`\xi ^a`$. Since the vector field we consider (i.e., $`\xi ^a`$) satisfies $`t^a\xi _a=0`$, we have $`\xi ^c_cg_{ab}=0`$. \[Recall that the metric $`g_{ab}`$ depends only on time $`\lambda `$ and that $`t^a`$ is proportional to $`(/\lambda )^a`$.\] Hence we find
$`T_{aba^{}b^{}}`$ $``$ $`40|_{(a}\xi _{b)}(x)_{(a^{}}\xi _{b^{})}(x^{})|0`$ (51)
$`=`$ $`_a_a^{}M_{bb^{}}+_a_b^{}M_{ba^{}}+_b_a^{}M_{ab^{}}+_b_b^{}M_{aa^{}}`$
$`=`$ $`g_{ac}g_{a^{}c^{}}_b_b^{}0|\xi ^c\xi ^c^{}|0+g_{ac}g_{b^{}c^{}}_b_a^{}0|\xi ^c\xi ^c^{}|0`$
$`+g_{bc}g_{a^{}c^{}}_a_b^{}0|\xi ^c\xi ^c^{}|0+g_{bc}g_{b^{}c^{}}_a_a^{}0|\xi ^c\xi ^c^{}|0.`$
Define $`T_{aba^{}b^{}}^\mathrm{A}`$ to be the spacelike projection of $`T_{aba^{}b^{}}`$,
$$T_{aba^{}b^{}}^\mathrm{A}=P_a^cP_b^dP_a^{}^c^{}P_b^{}^d^{}T_{cdc^{}d^{}}.$$
(52)
Then, remembering that $`t_a\xi ^a=0`$, we find
$$T_{aba^{}b^{}}^\mathrm{A}=\stackrel{~}{}_b\stackrel{~}{}_b^{}M_{aa^{}}(x,x^{})+\stackrel{~}{}_b\stackrel{~}{}_a^{}M_{ab^{}}(x,x^{})+\stackrel{~}{}_a\stackrel{~}{}_b^{}M_{ba^{}}(x,x^{})+\stackrel{~}{}_a\stackrel{~}{}_a^{}M_{bb^{}}(x,x^{}),$$
(53)
where $`M_{aa^{}}`$ is given by (48). The components with one timelike index can be represented by
$`T_{aa^{}b^{}}^\mathrm{B}`$ $``$ $`P_a^ct^dP_a^{}^c^{}P_b^{}^d^{}T_{cdc^{}d^{}}`$ (54)
$`=`$ $`{\displaystyle \frac{H}{\lambda }}\stackrel{~}{}_b^{}{\displaystyle \frac{}{\lambda }}[\lambda ^2M_{aa^{}}(x,x^{})]{\displaystyle \frac{H}{\lambda }}\stackrel{~}{}_a^{}{\displaystyle \frac{}{\lambda }}[\lambda ^2M_{ab^{}}(x,x^{})].`$
(The minus sign arises due to the fact that the parameter $`\lambda `$ decreases towards the future.) We define $`T_{aba^{}}^\mathrm{B}`$ as the tensor obtained by interchanging the primed and unprimed indices in $`T_{aa^{}b^{}}^\mathrm{B}`$. The components with two timelike indices can be represented by
$$T_{ab^{}}^\mathrm{C}P_a^ct^dt^c^{}P_b^{}^d^{}T_{cdc^{}d^{}}=\frac{H^2}{\lambda \lambda ^{}}\frac{^2}{\lambda \lambda ^{}}[\lambda ^2\lambda ^2M_{ab^{}}(x,x^{})].$$
(55)
The pure-gauge two-point function $`T_{aba^{}b^{}}`$ is given by
$`T_{aba^{}b^{}}`$ $`=`$ $`T_{aba^{}b^{}}^\mathrm{A}+t_bT_{aa^{}b^{}}^\mathrm{B}+t_aT_{ba^{}b^{}}^\mathrm{B}+t_b^{}T_{aba^{}}^\mathrm{B}+t_a^{}T_{abb^{}}^\mathrm{B}`$ (56)
$`+t_at_a^{}T_{bb^{}}^\mathrm{C}+t_bt_a^{}T_{ab^{}}^\mathrm{C}+t_bt_a^{}T_{ab^{}}^\mathrm{C}+t_bt_b^{}T_{aa^{}}^\mathrm{C}.`$
We will find that $`T_{aba^{}b^{}}^\mathrm{A}`$ contains a term proportional to $`\mathrm{log}\alpha ^2r^2\times \theta _{aba^{}b^{}}^{(2)}`$, whereas the other two tensors $`T_{aa^{}b^{}}^\mathrm{B}`$ and $`T_{ab^{}}^\mathrm{C}`$ contain no terms which grow with the distance $`r`$.
By using (48) and (53) we have
$`T_{aba^{}b^{}}^\mathrm{A}`$ $`=`$ $`{\displaystyle \frac{1}{4\pi ^2\lambda ^2\lambda ^2}}{\displaystyle 𝑑kk^2E(k,\lambda ,\lambda ^{})e^{ik\mathrm{\Delta }\lambda }}`$ (57)
$`\times (\stackrel{~}{}_b\stackrel{~}{}_b^{}\stackrel{~}{\delta }_{aa^{}}+\stackrel{~}{}_b\stackrel{~}{}_a^{}\stackrel{~}{\delta }_{ab^{}}+\stackrel{~}{}_a\stackrel{~}{}_b^{}\stackrel{~}{\delta }_{ba^{}}+\stackrel{~}{}_a\stackrel{~}{}_a^{}\stackrel{~}{\delta }_{bb^{}}{\displaystyle \frac{4}{k^2}}\stackrel{~}{}_a\stackrel{~}{}_a^{}\stackrel{~}{}_b\stackrel{~}{}_b^{})`$
$`\times {\displaystyle \frac{\mathrm{sin}kr}{kr}}.`$
This tensor can be evaluated by using the integration formulas in Appendix A. The result is
$$T_{aba^{}b^{}}^A=T^{(1)}\times \theta _{aba^{}b^{}}^{(1)}+T^{(2)}\times \theta _{aba^{}b^{}}^{(2)}+T^{(3)}\times \theta _{aba^{}b^{}}^{(3)},$$
(58)
where the bi-tensors $`\theta _{aba^{}b^{}}^{(i)}`$ are defined in section 2, while the $`T^{(i)}`$ are given as follows:
$`T^{(1)}`$ $`=`$ $`{\displaystyle \frac{H^4}{4\pi ^2}}\{{\displaystyle \frac{36}{5}}+3V^2(4+3V^2(4\psi ))`$ (59)
$`+3{\displaystyle \frac{\lambda \lambda ^{}}{r^2}}\left[{\displaystyle \frac{12V^2}{1V^2}}+8+15V^2(4\psi )\right]`$
$`+3{\displaystyle \frac{\lambda ^2\lambda _{}^{}{}_{}{}^{2}}{r^4}}(15(4\psi )+{\displaystyle \frac{4(7+5V^2)}{(1V^2)^2}})\},`$
$`T^{(2)}`$ $`=`$ $`{\displaystyle \frac{H^4}{4\pi ^2}}\{{\displaystyle \frac{18}{5}}\gamma {\displaystyle \frac{108}{25}}+{\displaystyle \frac{1}{5}}V^2[4+3V^2(4\psi )]`$ (60)
$`+{\displaystyle \frac{\lambda \lambda ^{}}{r^2}}\left[2+3V^2(4\psi ){\displaystyle \frac{6V^2}{1V^2}}\right]`$
$`+{\displaystyle \frac{\lambda ^2\lambda _{}^{}{}_{}{}^{2}}{r^4}}[{\displaystyle \frac{4(V^22)}{(1V^2)^2}}+3(4\psi )]+{\displaystyle \frac{9}{5}}\mathrm{log}\alpha ^2r^2(1V^2)\},`$
$`T^{(3)}`$ $`=`$ $`{\displaystyle \frac{H^4}{4\pi ^2}}\{{\displaystyle \frac{3}{5}}+V^2(4+3V^2(4\psi )){\displaystyle \frac{3V^2(1+V^2(V^22))}{(1V^2)^3}}`$ (61)
$`+{\displaystyle \frac{\lambda \lambda ^{}}{r^2}}\left[1+15V^2(4\psi ){\displaystyle \frac{3[1+V^2(11+V^2(7V^217))]}{(1V^2)^3}}\right]`$
$`+{\displaystyle \frac{\lambda ^2\lambda _{}^{}{}_{}{}^{2}}{r^4}}[15(4\psi ){\displaystyle \frac{4[9+V^2(5V^212)]}{(1V^2)^3}}]\}.`$
To compute $`T_{aa^{}b^{}}^\mathrm{B}`$ we use (48) and (54) and find
$`T_{aa^{}b^{}}^\mathrm{B}`$ $`=`$ $`{\displaystyle \frac{H}{4\pi ^2\lambda \lambda ^2}}{\displaystyle 𝑑kk^2\frac{}{\lambda }\left[E(k,\lambda ,\lambda ^{})e^{ik\mathrm{\Delta }\lambda }\right]}`$ (62)
$`\times \left(\stackrel{~}{\delta }_{aa^{}}\stackrel{~}{}_b^{}+\stackrel{~}{\delta }_{ab^{}}\stackrel{~}{}_a^{}2{\displaystyle \frac{\stackrel{~}{}_b^{}\stackrel{~}{}_a\stackrel{~}{}_a^{}}{k^2}}\right){\displaystyle \frac{\mathrm{sin}kr}{kr}}.`$
Then, using the integral formulas in Appendix A, we have
$$T_{aa^{}b^{}}^\mathrm{B}=\stackrel{~}{T}^{(1)}\beta _{aa^{}b^{}}^{(1)}+\stackrel{~}{T}^{(2)}\beta _{aa^{}b^{}}^{(2)}$$
(63)
with
$`\beta _{aa^{}b^{}}^{(1)}=P_{aa^{}}V_b^{}+P_{ab^{}}V_a^{}+2V_aV_a^{}V_b^{},`$ (64)
$`\beta _{aa^{}b^{}}^{(2)}=P_{a^{}b^{}}V_aP_{aa^{}}V_b^{}P_{ab^{}}V_a^{}5V_aV_a^{}V_b^{},`$ (65)
and
$`\stackrel{~}{T}^{(1)}`$ $`=`$ $`{\displaystyle \frac{H^4}{4\pi ^2}}\{{\displaystyle \frac{3\lambda }{r}}{\displaystyle \frac{1}{1V^2}}+{\displaystyle \frac{1}{r^3}}{\displaystyle \frac{2\lambda \lambda ^{}(3\lambda \lambda ^{})}{(1V^2)^2}}`$ (66)
$`+{\displaystyle \frac{1}{r^4}}{\displaystyle \frac{8\lambda ^2\lambda _{}^{}{}_{}{}^{2}V}{(1V^2)^3}}\},`$
$`\stackrel{~}{T}^{(2)}`$ $`=`$ $`{\displaystyle \frac{H^4}{4\pi ^2}}\{{\displaystyle \frac{2\lambda }{r}}+{\displaystyle \frac{1}{r^3}}[{\displaystyle \frac{2\lambda \lambda ^{}(3\lambda +\lambda ^{})}{1V^2}}+6\lambda ^2(\lambda +\lambda ^{})]`$ (67)
$`+{\displaystyle \frac{1}{r^4}}[{\displaystyle \frac{2\lambda ^2\lambda _{}^{}{}_{}{}^{2}V(3V^25)}{(1V^2)^2}}{\displaystyle \frac{3}{2}}\lambda ^4\psi _2]\}.`$
We have defined $`\psi _2=V^1\psi `$.
Next we examine the tensor $`T_{ab^{}}^\mathrm{C}`$ by using (55) and (48). We first find
$`T_{ab^{}}^C`$ $`=`$ $`{\displaystyle \frac{H^2}{4\pi ^2\lambda \lambda ^{}}}{\displaystyle 𝑑kk^2\frac{^2}{\lambda \lambda ^{}}E(k,\lambda ,\lambda ^{})e^{ik\mathrm{\Delta }\lambda }}`$ (68)
$`\times \left(\stackrel{~}{\delta }_{ab^{}}{\displaystyle \frac{\stackrel{~}{}_a\stackrel{~}{}_b^{}}{k^2}}\right){\displaystyle \frac{\mathrm{sin}kr}{kr}}.`$
By integrating over $`k`$, we have
$$T_{ab^{}}^C=S^{(1)}P_{ab^{}}+S^{(2)}V_aV_a^{},$$
(69)
where the $`S^{(1)}`$ and $`S^{(2)}`$ are given by
$`S^{(1)}`$ $`=`$ $`{\displaystyle \frac{\lambda \lambda ^{}H^4}{2\pi ^2}}[{\displaystyle \frac{1}{r^2}}{\displaystyle \frac{V}{(1V^2)^2}}`$ (70)
$`+{\displaystyle \frac{1}{r^4}}({\displaystyle \frac{2\lambda \lambda ^{}V^2}{(1V^2)^2}}+{\displaystyle \frac{4\lambda \lambda ^{}V^2}{(1V^2)^3}})],`$
$`S^{(2)}`$ $`=`$ $`{\displaystyle \frac{\lambda \lambda ^{}H^4}{2\pi ^2}}\left[{\displaystyle \frac{1}{r^2}}{\displaystyle \frac{1}{(1V^2)^2}}+{\displaystyle \frac{1}{r^4}}{\displaystyle \frac{4\lambda \lambda ^{}}{(1V^2)^3}}\right].`$ (71)
Details of the calculations presented here can be found in Appendix B.
## 4 Conclusion
It can be seen from the results of the previous section that $`T_{aa^{}b^{}}^B`$ and $`T_{ab^{}}^C`$ are of order $`r^1`$ and of order $`r^2`$, respectively, and that they are both IR finite. Hence, we have from (56)
$`T_{aba^{}b^{}}`$ $`=`$ $`T_{aba^{}b^{}}^\mathrm{A}+O(r^1)`$ (72)
$`=`$ $`{\displaystyle \frac{H^4}{4\pi ^2}}\left({\displaystyle \frac{18}{5}}\gamma {\displaystyle \frac{108}{25}}+{\displaystyle \frac{9}{5}}\mathrm{log}\alpha ^2r^2\right)\times \theta _{aba^{}b^{}}^{(2)}`$
$`{\displaystyle \frac{36}{5}}\times \theta _{aba^{}b^{}}^{(1)}+{\displaystyle \frac{3}{5}}\times \theta _{aba^{}b^{}}^{(3)}+O(r^1).`$
By comparing this result and the physical two-point function $`G_{aba^{}b^{}}`$ given by (20) with (30)–(32), we find
$$G_{aba^{}b^{}}(x,x^{})=G_{aba^{}b^{}}^{\mathrm{mod}}(x,x^{})+\frac{4}{9H^2}0|_{(a}\xi _{b)}(x)_{(a^{}}\xi _{b^{})}(x^{})|0,$$
(73)
where the two-point function $`G_{aba^{}b^{}}^{\mathrm{mod}}(x,x^{})`$, which is related to $`G_{aba^{}b^{}}(x,x^{})`$ by a gauge transformation, is IR finite and does not grow logarithmically with $`r`$. Thus, we have shown that the logarithmic behaviour in $`G_{aba^{}b^{}}(x,x^{})`$ can be gauged away by a two-point function of a pure-gauge field.
Recently, Hawking, Hertog and Turok have found that the physical graviton two-point function is well-behaved for large distances in open de Sitter spacetime. Now, gauge-invariant correlation functions must be the same in the de Sitter invariant vacuum, whether we use the spatially-flat or open coordinate system. Hence, their result implies that there cannot be any physical effects due to the logarithmic term in the two-point function (20). This is consistent with our result. Finally, it will be interesting to investigate the implication of our result for two-point functions in covariant gauges (see, e.g. Ref. ).
Acknowledgement
We thank Bernard Kay for useful discussions. Some of the calculations in this paper were performed using Maple V Release 5.1.
## Appendix A. Some useful integrals
We first obtain
$$_\alpha ^{\mathrm{}}\frac{dk}{k}e^{ik(x+iϵ)}=_{\alpha (x+iϵ)}^{\mathrm{}}\frac{d\kappa }{\kappa }\mathrm{cos}\kappa +i_{\alpha x}^{\mathrm{}}\frac{d\kappa }{\kappa }\mathrm{sin}\kappa =\gamma \mathrm{log}\alpha (x+iϵ)+\frac{\pi i}{2},$$
(A1)
where the terms of order $`\alpha `$ or higher are neglected. By differentiating this formula with respect to $`x`$ we find
$$_0^{\mathrm{}}𝑑ke^{ik(x+iϵ)}=\frac{i}{x+iϵ}.$$
(A2)
$$_0^{\mathrm{}}𝑑kk^2e^{ik(x+iϵ)}=\frac{2i}{(x+iϵ)^3},$$
(A3)
$$_0^{\mathrm{}}𝑑kke^{ik(x+iϵ)}=\frac{1}{(x+iϵ)^2},$$
(A4)
We define $`K=\frac{\pi i}{2}\gamma `$ and use integration by parts to find the following formulas:
$$_\alpha ^{\mathrm{}}\frac{dk}{k^2}e^{ikx}=\frac{1}{\alpha }+ix[K+1\mathrm{log}\alpha x],$$
(A5)
$$_\alpha ^{\mathrm{}}\frac{dk}{k^3}e^{ikx}=\frac{1}{2\alpha ^2}+\frac{ix}{\alpha }\frac{x^2}{2}\left[K+\frac{3}{2}\mathrm{log}\alpha x\right],$$
(A6)
$$_\alpha ^{\mathrm{}}\frac{dk}{k^4}e^{ikx}=\frac{1}{3\alpha ^3}+\frac{ix}{2\alpha ^2}\frac{x^2}{2\alpha }\frac{ix^3}{6}\left[K+\frac{11}{6}\mathrm{log}\alpha x\right],$$
(A7)
$$_\alpha ^{\mathrm{}}\frac{dk}{k^5}e^{ikx}=\frac{1}{4\alpha ^4}+\frac{ix}{3\alpha ^3}\frac{x^2}{4\alpha ^2}\frac{ix^3}{6\alpha }+\frac{x^4}{24}\left[K+\frac{25}{12}\mathrm{log}\alpha x\right],$$
(A8)
$$_\alpha ^{\mathrm{}}\frac{dk}{k^6}e^{ikx}=\frac{1}{5\alpha ^5}+\frac{ix}{4\alpha ^4}\frac{x^2}{6\alpha ^3}\frac{ix^3}{12\alpha ^2}+\frac{x^4}{24\alpha }\frac{ix^5}{120}\left[K+\frac{137}{60}\mathrm{log}\alpha x\right].$$
(A9)
## Appendix B. Details of the calculation of the pure-gauge two-point function
B1. Spatial components
To calculate $`T_{aba^{}b^{}}^\mathrm{A}`$ we start from the quantity
$`A_{aba^{}b^{}}`$ $`=`$ $`{\displaystyle \frac{4\pi }{(\lambda \lambda ^{})^2}}(\stackrel{~}{\delta }_{aa^{}}\stackrel{~}{}_b\stackrel{~}{}_b^{}+\stackrel{~}{\delta }_{ba^{}}\stackrel{~}{}_a\stackrel{~}{}_b^{}+\stackrel{~}{\delta }_{ab^{}}\stackrel{~}{}_b\stackrel{~}{}_a^{}+\stackrel{~}{\delta }_{bb^{}}\stackrel{~}{}_a\stackrel{~}{}_a^{}`$ (B1)
$`{\displaystyle \frac{4}{k^2}}\stackrel{~}{}_a\stackrel{~}{}_b\stackrel{~}{}_a^{}\stackrel{~}{}_b^{}){\displaystyle \frac{\mathrm{sin}kr}{kr}}`$
in equation (57). By using
$$\stackrel{~}{}_af(r)=\frac{f}{r}\frac{r_a}{r},\stackrel{~}{}_ar_a^{}=\stackrel{~}{\delta }_{aa^{}},$$
(B2)
and using symmetries of $`A_{aba^{}b^{}}`$ we find
$`A_{aba^{}b^{}}`$ $`=`$ $`{\displaystyle \frac{1}{(\lambda \lambda ^{})^2}}[A^{(1)}{\displaystyle \frac{r_ar_br_a^{}r_b^{}}{r^4}}`$ (B3)
$`+A^{(2)}{\displaystyle \frac{\stackrel{~}{\delta }_{aa^{}}r_br_b^{}+\stackrel{~}{\delta }_{ba^{}}r_ar_b^{}+\stackrel{~}{\delta }_{ab^{}}r_br_a^{}+\stackrel{~}{\delta }_{bb^{}}r_ar_a^{}}{r^2}}`$
$`+A^{(3)}{\displaystyle \frac{\stackrel{~}{\delta }_{a^{}b^{}}r_ar_b+\stackrel{~}{\delta }_{ab}r_a^{}r_b^{}}{r^2}}+A^{(4)}(\stackrel{~}{\delta }_{a^{}b}\stackrel{~}{\delta }_{ab^{}}+\stackrel{~}{\delta }_{bb^{}}\stackrel{~}{\delta }_{aa^{}})`$
$`+A^{(5)}\stackrel{~}{\delta }_{ab}\stackrel{~}{\delta }_{a^{}b^{}}],`$
where
$`A^{(1)}`$ $`=`$ $`4\pi \left({\displaystyle \frac{4k\mathrm{sin}kr}{r}}{\displaystyle \frac{40\mathrm{cos}kr}{r^2}}+{\displaystyle \frac{180\mathrm{sin}kr}{kr^3}}+{\displaystyle \frac{420\mathrm{cos}kr}{k^2r^4}}{\displaystyle \frac{420\mathrm{sin}kr}{k^3r^5}}\right),`$ (B4)
$`A^{(2)}`$ $`=`$ $`4\pi \left({\displaystyle \frac{k\mathrm{sin}kr}{r}}{\displaystyle \frac{7\mathrm{cos}kr}{r^2}}+{\displaystyle \frac{27\mathrm{sin}kr}{kr^3}}+{\displaystyle \frac{60\mathrm{cos}kr}{k^2r^4}}{\displaystyle \frac{60\mathrm{sin}kr}{k^3r^5}}\right),`$ (B5)
$`A^{(3)}`$ $`=`$ $`4\pi \left({\displaystyle \frac{4\mathrm{cos}kr}{r^2}}{\displaystyle \frac{24\mathrm{sin}kr}{kr^3}}{\displaystyle \frac{60\mathrm{cos}kr}{k^2r^4}}+{\displaystyle \frac{60\mathrm{sin}kr}{k^3r^5}}\right),`$ (B6)
$`A^{(4)}`$ $`=`$ $`4\pi \left({\displaystyle \frac{2\mathrm{cos}kr}{r^2}}+{\displaystyle \frac{6\mathrm{sin}kr}{kr^3}}+{\displaystyle \frac{12\mathrm{cos}kr}{k^2r^4}}{\displaystyle \frac{12\mathrm{sin}kr}{k^3r^5}}\right),`$ (B7)
$`A^{(5)}`$ $`=`$ $`4\pi \left({\displaystyle \frac{4\mathrm{sin}kr}{kr^3}}+{\displaystyle \frac{12\mathrm{cos}kr}{k^2r^4}}{\displaystyle \frac{12\mathrm{sin}kr}{k^3r^5}}\right).`$ (B8)
Using the definitions for $`P_{ab}`$ and $`V_a`$, we can express $`A_{aba^{}b^{}}`$ in terms of $`P_{ab}`$ and $`V_a`$ as
$`H^4A_{aba^{}b^{}}`$ $`=`$ $`A^{(1)}\times V_aV_bV_a^{}V_b^{}+A^{(2)}\times \left(P_{aa^{}}V_bV_b^{}+P_{ba^{}}V_aV_b^{}+P_{ab}V_bV_a^{}+P_{bb^{}}V_aV_a^{}\right)`$ (B9)
$`+A^{(3)}\times \left(P_{a^{}b^{}}V_aV_b+P_{ab}V_a^{}V_b^{}\right)+A^{(4)}\times \left(P_{a^{}b}P_{ab^{}}+P_{bb^{}}P_{aa^{}}\right)`$
$`+A^{(5)}\times P_{ab}P_{a^{}b^{}}.`$
The identity $`\stackrel{~}{\delta }^{ab}A_{aba^{}b^{}}=0`$ implies that $`A^{(1)}4A^{(2)}+3A^{(3)}=A^{(3)}+2A^{(4)}+3A^{(5)}=0`$. These equations are indeed satisfied. They allow us to eliminate $`A^{(1)}`$ and $`A^{(5)}`$. Thus, we obtain
$$A_{aba^{}b^{}}=3A^{(3)}\times \theta _{aba^{}b^{}}^{(1)}+A^{(4)}\times \theta _{aba^{}b^{}}^{(2)}+A^{(2)}\times \theta _{aba^{}b^{}}^{(3)},$$
(B10)
where the $`\theta ^{(i)}`$ are defined in section 2. By substituting this in equation (57) and noting that
$`{\displaystyle \frac{\mathrm{cos}kr}{k^n}}e^{ik\mathrm{\Delta }\lambda }={\displaystyle \frac{1}{2}}\left[{\displaystyle \frac{e^{ik(r+\mathrm{\Delta }\lambda )}}{k^n}}+{\displaystyle \frac{e^{ik(r+\mathrm{\Delta }\lambda )}}{k^n}}\right],`$ (B11)
$`{\displaystyle \frac{\mathrm{sin}kr}{k^n}}e^{ik\mathrm{\Delta }\lambda }={\displaystyle \frac{1}{2i}}\left[{\displaystyle \frac{e^{ik(r+\mathrm{\Delta }\lambda )}}{k^n}}{\displaystyle \frac{e^{ik(r+\mathrm{\Delta }\lambda )}}{k^n}}\right],`$ (B12)
we can calculate $`T_{aba^{}b^{}}^\mathrm{A}`$, using the integrals in Appendix A. Thus we obtain
$$T_{aba^{}b^{}}^\mathrm{A}=T^{(1)}\times \theta _{aba^{}b^{}}^{(1)}+T^{(2)}\times \theta _{aba^{}b^{}}^{(2)}+T^{(3)}\times \theta _{aba^{}b^{}}^{(3)}$$
(B13)
where
$`T^{(1)}`$ $`=`$ $`{\displaystyle \frac{H^4}{4\pi ^2}}\{{\displaystyle \frac{36\lambda \lambda ^{}(\mathrm{\Delta }\lambda )^2}{r^2(r\mathrm{\Delta }\lambda )(r+\mathrm{\Delta }\lambda )}}{\displaystyle \frac{60\lambda ^2\lambda _{}^{}{}_{}{}^{2}(\mathrm{\Delta }\lambda )^2}{r^2(r\mathrm{\Delta }\lambda )^2(r+\mathrm{\Delta }\lambda )^2}}`$ (B14)
$`{\displaystyle \frac{84\lambda ^2\lambda _{}^{}{}_{}{}^{2}}{(r\mathrm{\Delta }\lambda )^2(r+\mathrm{\Delta }\lambda )^2}}+{\displaystyle \frac{36}{5}}+{\displaystyle \frac{36h(\lambda ,\lambda ^{})}{r^4}}+{\displaystyle \frac{12(\lambda ^2+\lambda _{}^{}{}_{}{}^{2})}{r^2}}`$
$`{\displaystyle \frac{18(\lambda ^5\lambda _{}^{}{}_{}{}^{5})}{r^5}}\mathrm{log}{\displaystyle \frac{r+\mathrm{\Delta }\lambda }{r+\mathrm{\Delta }\lambda }}\},`$
$`T^{(2)}`$ $`=`$ $`{\displaystyle \frac{H^4}{4\pi ^2}}\{{\displaystyle \frac{18}{5}}\gamma {\displaystyle \frac{108}{25}}{\displaystyle \frac{2\lambda \lambda ^{}(3(\mathrm{\Delta }\lambda )^2+2\lambda \lambda ^{})}{r^2(r\mathrm{\Delta }\lambda )(r+\mathrm{\Delta }\lambda )}}{\displaystyle \frac{4\lambda ^2\lambda _{}^{}{}_{}{}^{2}}{(r\mathrm{\Delta }\lambda )^2(r+\mathrm{\Delta }\lambda )^2}}`$ (B15)
$`+{\displaystyle \frac{4}{5}}{\displaystyle \frac{(\mathrm{\Delta }\lambda )^2}{r^2}}+{\displaystyle \frac{8\lambda \lambda ^{}}{r^2}}+{\displaystyle \frac{12}{5}}{\displaystyle \frac{h(\lambda ,\lambda ^{})}{r^4}}`$
$`+{\displaystyle \frac{9}{5}}\mathrm{log}\alpha ^2(r^2(\mathrm{\Delta }\lambda )^2){\displaystyle \frac{6}{5}}{\displaystyle \frac{\lambda ^5\lambda _{}^{}{}_{}{}^{5}}{r^5}}\mathrm{log}{\displaystyle \frac{r+\mathrm{\Delta }\lambda }{r+\mathrm{\Delta }\lambda }}\},`$
$`T^{(3)}`$ $`=`$ $`{\displaystyle \frac{H^4}{4\pi ^2}}\{[3((\mathrm{\Delta }\lambda )^2\lambda \lambda ^{})r^6+(33\lambda \lambda ^{}(\mathrm{\Delta }\lambda )^26(\mathrm{\Delta }\lambda )^4+36\lambda ^2\lambda _{}^{}{}_{}{}^{2})r^4`$ (B16)
$`+(51\lambda \lambda ^{}(\mathrm{\Delta }\lambda )^448\lambda ^2\lambda _{}^{}{}_{}{}^{2}+3(\mathrm{\Delta }\lambda )^6)r^2`$
$`+20\lambda ^2\lambda _{}^{}{}_{}{}^{2}(\mathrm{\Delta }\lambda )^4+21\lambda \lambda ^{}(\mathrm{\Delta }\lambda )^6]/r^2(r+\mathrm{\Delta }\lambda )^3(r\mathrm{\Delta }\lambda )^3{\displaystyle \frac{3}{5}}`$
$`+{\displaystyle \frac{\lambda \lambda ^{}+4(\mathrm{\Delta }\lambda )^2}{r^2}}+{\displaystyle \frac{12h(\lambda ,\lambda ^{})}{r^4}}{\displaystyle \frac{6(\lambda ^5\lambda _{}^{}{}_{}{}^{5})}{r^5}}\mathrm{log}{\displaystyle \frac{r+\mathrm{\Delta }\lambda }{r+\mathrm{\Delta }\lambda }}\},`$
with
$$h(\lambda ,\lambda ^{})=\lambda ^4+\lambda ^3\lambda ^{}+\lambda ^2\lambda _{}^{}{}_{}{}^{2}+\lambda \lambda _{}^{}{}_{}{}^{3}+\lambda _{}^{}{}_{}{}^{4}.$$
(B17)
The expression (61) follows by reexpressing these in terms of $`V`$ and $`\psi `$.
B2. Components with one time index
We first find a simpler form for the expression (62). We note that
$`{\displaystyle \frac{4\pi }{H^3\lambda \lambda ^2}}\left(\stackrel{~}{\delta }_{aa^{}}\stackrel{~}{}_b^{}{\displaystyle \frac{1}{k^2}}\stackrel{~}{}_b^{}\stackrel{~}{}_a\stackrel{~}{}_a^{}\right){\displaystyle \frac{\mathrm{sin}kr}{kr}}`$
$`=4\pi \left({\displaystyle \frac{\mathrm{cos}kr}{r}}{\displaystyle \frac{\mathrm{sin}kr}{kr^2}}\right)\times P_{aa^{}}V_b^{}`$
$`+4\pi \left({\displaystyle \frac{\mathrm{sin}kr}{kr^2}}+{\displaystyle \frac{3\mathrm{cos}kr}{k^2r^3}}{\displaystyle \frac{3\mathrm{sin}kr}{k^3r^4}}\right)\times \left(P_{aa^{}}V_b^{}+P_{a^{}b^{}}V_aP_{ab^{}}V_a^{}\right)`$
$`+4\pi \left({\displaystyle \frac{\mathrm{cos}kr}{r}}{\displaystyle \frac{6\mathrm{sin}kr}{kr^2}}{\displaystyle \frac{15\mathrm{cos}kr}{k^2r^3}}+{\displaystyle \frac{15\mathrm{sin}kr}{k^3r^4}}\right)V_aV_a^{}V_b^{}.`$ (B18)
We also find
$$k^2\frac{}{\lambda }\left[E(k,\lambda ,\lambda ^{})e^{ik(\lambda \lambda ^{})}\right]=[3\lambda k^13i\lambda \mathrm{\Delta }\lambda +\lambda \lambda ^{}(\lambda ^{}+3\lambda )k+i\lambda ^2\lambda _{}^{}{}_{}{}^{2}k^2]e^{ik\mathrm{\Delta }\lambda }.$$
(B19)
Therefore we have
$`{\displaystyle \frac{H}{\lambda }}{\displaystyle \frac{}{\lambda }}\stackrel{~}{}_b^{}[\lambda ^2M_{aa^{}}(x,x^{})]`$
$`={\displaystyle \frac{H^4}{16\pi ^3}}{\displaystyle 𝑑k\left\{3\lambda k^13i\lambda \mathrm{\Delta }\lambda +\lambda \lambda ^{}(\lambda ^{}+3\lambda )k+i\lambda ^2\lambda _{}^{}{}_{}{}^{2}k^2e^{ik\mathrm{\Delta }\lambda }\right\}}`$
$`\times \{4\pi [{\displaystyle \frac{\mathrm{cos}kr}{r}}{\displaystyle \frac{\mathrm{sin}kr}{kr^2}}]\times P_{aa^{}}V_b^{}`$
$`+4\pi \left[{\displaystyle \frac{\mathrm{sin}kr}{kr^2}}+{\displaystyle \frac{3\mathrm{cos}kr}{k^2r^3}}{\displaystyle \frac{3\mathrm{sin}kr}{k^3r^4}}\right]\times \left(P_{aa^{}}V_b^{}+P_{a^{}b^{}}V_aP_{ab^{}}V_a^{}\right)`$
$`+4\pi [{\displaystyle \frac{\mathrm{cos}kr}{r}}{\displaystyle \frac{6\mathrm{sin}kr}{kr^2}}{\displaystyle \frac{15\mathrm{cos}kr}{k^2r^3}}+{\displaystyle \frac{15\mathrm{sin}kr}{k^3r^4}}]V_aV_a^{}V_b^{}\}.`$ (B20)
We obtain the analogous expression for $`(H/\lambda )(/\lambda )\stackrel{~}{}_a^{}[\lambda ^2M_{ab^{}}(x,x^{})]`$ by interchanging $`a^{}`$ with $`b^{}`$. Thus, we have
$`T_{aa^{}b^{}}^\mathrm{B}`$ $`=`$ $`{\displaystyle \frac{H}{\lambda }}{\displaystyle \frac{}{\lambda }}\stackrel{~}{}_b^{}[\lambda ^2M_{aa^{}}(x,x^{})]{\displaystyle \frac{H}{\lambda }}{\displaystyle \frac{}{\lambda }}\stackrel{~}{}_a^{}[\lambda ^2M_{ab^{}}(x,x^{})]`$ (B21)
$`=`$ $`{\displaystyle \frac{H^4}{16\pi ^3}}{\displaystyle 𝑑k\left[3\lambda k^13i\lambda \mathrm{\Delta }\lambda +\lambda \lambda ^{}(\lambda ^{}+3\lambda )k+i\lambda ^2\lambda _{}^{}{}_{}{}^{2}k^2\right]}`$
$`\times e^{ik\mathrm{\Delta }\lambda }\times D_{aa^{}b^{}},`$
where $`D_{aa^{}b^{}}`$ is given by
$$D_{aa^{}b^{}}=D^{(1)}\beta _{aa^{}b^{}}^{(1)}+D^{(2)}\beta _{aa^{}b^{}}^{(2)}$$
(B22)
with
$`D^{(1)}`$ $`=`$ $`4\pi \left({\displaystyle \frac{\mathrm{cos}kr}{r}}{\displaystyle \frac{\mathrm{sin}kr}{kr^2}}\right)`$
$`D^{(2)}`$ $`=`$ $`8\pi \left({\displaystyle \frac{\mathrm{sin}kr}{kr^2}}+{\displaystyle \frac{3\mathrm{cos}kr}{k^2r^3}}{\displaystyle \frac{3\mathrm{sin}kr}{k^3r^4}}\right)`$ (B23)
and
$`\beta _{aa^{}b^{}}^{(1)}=P_{aa^{}}V_b^{}+P_{ab^{}}V_a^{}+2V_aV_a^{}V_b^{},`$ (B24)
$`\beta _{aa^{}b^{}}^{(2)}=P_{a^{}b^{}}V_aP_{aa^{}}V_b^{}P_{ab^{}}V_a^{}5V_aV_a^{}V_b^{}.`$ (B25)
Again, by performing the $`k`$ integration using the formulas in Appendix A we find the result in (63).
B3. Components with two time indices
We begin by noting that
$`{\displaystyle \frac{4\pi }{H^2\lambda \lambda ^{}}}\left(\stackrel{~}{\delta }_{ab^{}}{\displaystyle \frac{\stackrel{~}{}_a\stackrel{~}{}_b^{}}{k^2}}\right){\displaystyle \frac{\mathrm{sin}kr}{kr}}`$
$`=4\pi \left({\displaystyle \frac{\mathrm{sin}kr}{kr}}+{\displaystyle \frac{\mathrm{cos}kr}{k^2r^2}}{\displaystyle \frac{\mathrm{sin}kr}{k^3r^3}}\right)P_{aa^{}}`$
$`+4\pi \left({\displaystyle \frac{\mathrm{sin}kr}{kr}}+{\displaystyle \frac{3\mathrm{cos}kr}{k^2r^2}}{\displaystyle \frac{3\mathrm{sin}kr}{k^3r^3}}\right)V_aV_a^{}`$ (B26)
and
$$k^2\frac{^2}{\lambda \lambda ^{}}\left[E(k,\lambda ,\lambda ^{})e^{ik\mathrm{\Delta }\lambda }\right]=\left(\lambda \lambda ^{}ki\lambda \lambda ^{}\mathrm{\Delta }\lambda k^2+\lambda ^2\lambda _{}^{}{}_{}{}^{2}k^3\right)e^{ik\mathrm{\Delta }\lambda }.$$
(B27)
Therefore we find
$`T_{ab^{}}^\mathrm{C}`$ $`=`$ $`{\displaystyle \frac{H^2}{\lambda \lambda ^{}}}{\displaystyle \frac{^2}{\lambda \lambda ^{}}}[(\lambda \lambda ^{})^2M_{ab^{}}(x,x^{})]`$ (B28)
$`=`$ $`{\displaystyle \frac{H^4}{16\pi ^3}}{\displaystyle 𝑑k\left(\lambda \lambda ^{}ki\lambda \lambda ^{}\mathrm{\Delta }\lambda k^2+\lambda ^2\lambda _{}^{}{}_{}{}^{2}k^3\right)e^{ik\mathrm{\Delta }\lambda }\times E_{ab^{}}},`$
where $`E_{ab^{}}`$ is given by
$$E_{ab^{}}=E^{(1)}P_{ab^{}}+E^{(2)}V_aV_b^{}$$
(B29)
with
$`E^{(1)}`$ $`=`$ $`4\pi \left({\displaystyle \frac{\mathrm{sin}kr}{kr}}+{\displaystyle \frac{\mathrm{cos}kr}{k^2r^2}}{\displaystyle \frac{\mathrm{sin}kr}{k^3r^3}}\right),`$ (B30)
$`E^{(2)}`$ $`=`$ $`4\pi \left({\displaystyle \frac{\mathrm{sin}kr}{kr}}+{\displaystyle \frac{3\mathrm{cos}kr}{k^2r^2}}{\displaystyle \frac{3\mathrm{sin}kr}{k^3r^3}}\right).`$ (B31)
We integrate over $`k`$ using the formulas in Appendix A and obtain
$$T_{ab^{}}=S^{(1)}P_{ab^{}}+S^{(2)}V_aV_b^{},$$
(B32)
with
$`S^{(1)}`$ $`=`$ $`{\displaystyle \frac{H^4\lambda \lambda ^{}}{2\pi ^2}}\left[{\displaystyle \frac{\lambda ^2+\lambda _{}^{}{}_{}{}^{2}}{(r\mathrm{\Delta }\lambda )^2(r+\mathrm{\Delta }\lambda )^2}}+{\displaystyle \frac{4\lambda \lambda ^{}(\mathrm{\Delta }\lambda )^2}{(r+\mathrm{\Delta }\lambda )^3(r\mathrm{\Delta }\lambda )^3}}\right],`$ (B33)
$`S^{(2)}`$ $`=`$ $`{\displaystyle \frac{H^4\lambda \lambda ^{}r^2}{2\pi ^2}}\left[{\displaystyle \frac{1}{(r\mathrm{\Delta }\lambda )^2(r+\mathrm{\Delta }\lambda )^2}}+{\displaystyle \frac{4\lambda \lambda ^{}}{(r+\mathrm{\Delta }\lambda )^3(r\mathrm{\Delta }\lambda )^3}}\right].`$ (B34)
By expressing these in terms of $`V`$ and powers of $`\lambda \lambda ^{}`$ we find (71).
|
warning/0004/astro-ph0004083.html
|
ar5iv
|
text
|
# Radio supernovae, supernova remnants and H ii regions in NGC 2146 observed with MERLIN and the VLA
## 1 Introduction
NGC 2146 is a starburst galaxy located at a distance of 14.5 Mpc (1″$``$ 70 pc). Its infrared luminosity measured with IRAS at 60 $`\mu \mathrm{m}`$ and 100 $`\mu \mathrm{m}`$ is $`6.6\times 10^{10}`$ $`\mathrm{L}_{}`$. This value and the large 25 $`\mu \mathrm{m}`$ to 60 $`\mu \mathrm{m}`$ flux ratio place the object in the lower part of the IR luminosity/spectral index plane populated by superluminous galaxies without active nuclei (Hutchings et al. hutch90 (1990)). NGC 2146 has a central molecular ring (Jackson & Ho jack88 (1988), Young et al. young88 (1988)), and an outflow of hot gas along the minor axis driven by supernova explosions and stellar winds in the starburst region (Armus et al. armus95 (1995), Della Ceca et al. della99 (1999)). These characteristics reveal a strong similarity to the prototype starburst galaxy M 82 (at a distance of 3.2 Mpc and with an infrared luminosity of $`2.4\times 10^{10}`$ $`\mathrm{L}_{}`$), although NGC 2146 does not have a companion that may have triggered the starburst (Fisher $`\&`$ Tully fisher76 (1976)). It has been suggested that the starburst in NGC 2146 is the result of a far evolved merger (Condon et al. condon82 (1982); Young et al. young88 (1988), Hutchings et al. hutch90 (1990)), but a fully convincing kinematic and material trace of the merger has not yet been found.
The nuclear region is partly obscured by a strongly absorbing dust lane (Benvenuti et al. benve75 (1975)). The high extinction (A<sub>v</sub> $`>`$ 5 mag; Young et al. young88bis (1988); Hutchings et al. hutch90 (1990); Smith et al. smith95 (1995)) makes the optical observation of the inner star-forming regions impossible.
Strong non–thermal radio emission from the centre of NGC 2146 has been detected by Kronberg & Biermann (kron81 (1981); hereafter KB), by Condon et al. (condon82 (1982)), and by Lisenfeld et al. (lisenfeld96 (1996), lisenfeld97 (1997)). About 20 point sources were detected at 5 GHz in the central 800 pc of NGC 2146 by Glendenning & Kronberg (glen86 (1986)) using the VLA, and these were interpreted as radio supernovae (RSN) or supernova remnants (SNRs). In this paper we report the detailed spatial distribution of these point sources, together with their number and fluxes derived from radio continuum observations at 5 GHz obtained with MERLIN and the VLA. From additional 1.6 GHz MERLIN observations we obtained the spectral index of 9 of these sources, which allow to identify their nature. Our results open the possibility of a direct comparison with similar observations of M 82 (Kronberg et al. kron85 (1985); Muxlow et al. mux94 (1994)) and other strong and nearby starburst galaxies, such as NGC 1808, NGC 4736 and NGC 5253 (Saikia et al. saikia90 (1990); Duric & Dittmar duric88 (1988); Turner et al. turner98 (1998)).
## 2 The observations and image processing
NGC 2146 was observed for 23 h with MERLIN (6 antennas) in November 1996 (one day) and February 1997 (two days). The observing frequency was 4.994 GHz ($`\lambda `$ = 6 cm), with a bandwidth of 15 MHz in both circular polarizations; the data were taken in spectral-line mode (15 $`\times `$ 1–MHz channels). OQ208 (2.39 Jy) and 0602+780 (0.126 Jy) were used as flux and phase calibration sources, respectively.
NGC 2146 was observed in the L-Band with MERLIN (6 antennas, including the Lovell telescope) in May 1997, for a period of 15 h. The observing frequency was 1.658 GHz ($`\lambda `$ = 18 cm), with a bandwidth of 15 MHz in both circular polarizations. The passband and relative gains of the antennas were determined using the point source calibrator 0552+398, with a flux density of 2.31 Jy; the compact source 0602+780 (0.109 Jy) was used to determine the telescope phases.
Moreover, we used observations of NGC 2146 taken in September 1985 with the VLA <sup>1</sup><sup>1</sup>1The National Radio Astronomy Observatory is a facility of the National Science Foundation operated under cooperative agreement by Associated Universities, Inc. (27 antennas) in the A, B and C array configurations; the observing frequency was 4.885 GHz, with an integration time of 18 h.
The 5 GHz data from MERLIN and the VLA A-array were combined to increase the brightness sensitivity and $`uv`$ coverage, in order to improve the information about the weaker sources already marginally detected in the 5 GHz MERLIN observation. To combine the two data sets it was necessary to shift the MERLIN phase reference position slightly. A re-weighting was also necessary to compensate for different origins of the data sets. Several images were produced using the AIPS task IMAGR, and deconvolved with the CLEAN algorithm (Högbom hoegbom74 (1974)). The details of these maps such as restoring beam, the applied weighting, and the rms noise, are summarized in Table 1. The rms noise in source-free areas of the images is consistent with the expected thermal noise levels.
Since the Wardle–telescope was not present in the observations, the length of the shortest baseline of the MERLIN array at 1.6 GHz is 9.2 km (Darnhall–Lovell telescopes). At this frequency the interferometer is then not sensitive to structures larger than $`\vartheta _{\mathrm{max}}^{1.6\mathrm{GHz}}=4\mathrm{}`$ (280 pc). At 5 GHz the shortest baseline of the MERLIN array is 9.5 km (Darnhall–Mark2 telescopes), so that $`\vartheta _{\mathrm{max}}^{5\mathrm{G}\mathrm{H}\mathrm{z}}`$ = $`1.3\mathrm{}`$ (90 pc). The combination of the 5 GHz MERLIN data with the VLA data gives a shortest baseline of 0.68 km, so that $`\vartheta _{\mathrm{max}}^{5\mathrm{G}\mathrm{H}\mathrm{z}}`$ = $`18\mathrm{}`$ (1250 pc). The short-spacing effect does not influence the results at either frequency since the size of the sources we are searching for is smaller than $`1\mathrm{}`$ (100 pc). Furthermore, most of the sources are either unresolved or have extensions of a few beam widths and hence the flux measurement is not influenced by the lack of short spacings.
## 3 The results
Figure 1 shows the uniformly weighted ABC-array VLA map (A) of NGC 2146. Due to the higher resolution (0$`\stackrel{}{.}`$37 $`\times `$ 0$`\stackrel{}{.}`$32 instead of 1$`\stackrel{}{.}`$2 $`\times `$ 1$`\stackrel{}{.}`$2) the comparison of this map with that of KB reveals a larger number of compact radio sources ($``$ 18 instead of 5). Any feature above 0.1 mJy (3$`\sigma `$ level) in the 5 GHz MERLIN + VLA map (C) is taken to be a source. Our map lacks the long ’tongue’ of emission to the SE, seen in the map of KB. The global features found by KB, such as the S-shape of the central emission region (by KB assumed to be a ’triple’ source, but now resolved into a larger number of sources), its position angle ($``$ 105<sup>°</sup>), and the main ’bar’ of radio emission ($``$ 128<sup>°</sup>), are consistent with those in our map. Convolution of our VLA map to the 1$`\stackrel{}{.}`$2 resolution of the KB map shows that the peak brightness is of the order of 6 mJy/beam as measured by KB, and that the SE tongue of emission is still present, though less extended than in the earlier maps. From the map shown in Fig. 1 it is evident that the point sources present in the SE part of the galaxy are more numerous than in the NW part (13 compared to 4, assuming that the centre of symmetry of NGC 2146 is near the strong central source 37.6+24.2). The extended emission of the SE region is also more pronounced, giving the impression of an asymmetry in NGC 2146, in favour of a stronger starburst activity along the SE arm. $`{}_{}{}^{12}\mathrm{CO}`$ interferometer observations (Young et al. 1988) show a higher concentration of molecular gas in this part of the galactic disk.
Table 2 contains the positions and the fluxes of the eighteen compact sources derived from our observation, listed by coordinate name with increasing R.A., following the convention of Kronberg & Wilkinson (kron75 (1975)). The reference position is at $`\alpha =06^\mathrm{h}18^\mathrm{m}00\stackrel{s}{.}0`$ and $`\delta =78\mathrm{°}21\mathrm{}00\stackrel{}{.}0`$ (J2000). The coordinates of the emission peaks were determined from a Gaussian fit of each component using the AIPS task JMFIT (which also gives the absolute positional uncertainties) in the uniformly weighted 5 GHz MERLIN + VLA map (C). From this map we also derived the peak and integrated flux densities at 5 GHz (S<sub>5</sub>). The same quantities were derived at 1.6 GHz (S<sub>1.6</sub>) from the naturally weighted MERLIN map (D). The numbers in parentheses are the errors of the derived quantities. Only nine sources are detected at 1.6 GHz, due to a lack of sensitivity in the MERLIN observations, and possibly also to the steep positive spectral indices which may place the remaining sources below the 3 $`\sigma `$ level. The spectral indices ($`\alpha _{1.6}^5`$) were obtained from the integrated flux densities at both frequencies, where we adopt the convention $`\mathrm{S}_\nu \nu ^\alpha `$. The flux densities have been computed using the AIPS task JMFIT, when the sources were either unresolved or shell-like, and using the task BLSUM for sources with extended structures impossible to fit by a single Gaussian. The integrated flux densities were calculated up to a 3$`\sigma `$ level to avoid confusion with the extended background emission.
In Table 3 we report the deduced quantities for the eighteen sources: the total emitted power at 5 GHz (L<sub>5</sub>) and 1.6 GHz (L<sub>1.6</sub>), for a distance of 14.5 Mpc, based on the peak flux values of Table 2, and the brightness temperature calculated from the integrated flux intensities at 5 GHz (Table 2). The diameter (D) was measured as described in Appendix 1 of Burns et al. (burns79 (1979)), where $`\mathrm{R}_{\mathrm{resolved}}=2\sqrt{\left(\frac{\mathrm{R}_{\mathrm{map}}}{2}\right)^2(0.85\mathrm{HWHM}_{\mathrm{beam}})^2}`$,
$`\mathrm{R}_{\mathrm{unresolved}}=0.85\mathrm{HWHM}_{\mathrm{beam}}`$, $`\mathrm{R}_{\mathrm{map}}`$ the radius taken from the map, and $`\mathrm{HWHM}_{\mathrm{beam}}`$ the half width at half maximum of the Gaussian beam. The diameter for the unresolved sources is 8.6 pc (Table 3) and their brightness temperatures are obviously lower limits because their true dimensions are not known.
### 3.1 Sources with negative spectral indices
#### 3.1.1 The “central source” 37.6+24.2
37.6+24.2 has a large negative spectral index of $`\alpha _{1.6}^5`$ = –1.1, consistent with optically thin synchrotron emission. The source has a position offset of $`1\stackrel{}{.}15`$ north-east of the dynamic center of NGC 2146 (R.A. = 06<sup>h</sup>18<sup>m</sup>37$`\stackrel{s}{.}`$4 $`\pm `$ 0$`\stackrel{s}{.}`$4, Dec. = 78°21′23$`\stackrel{}{.}`$2 $`\pm `$ 2″) deduced from the $`{}_{}{}^{12}\mathrm{CO}(10)`$ rotation curve measured with the Plateau de Bure interferometer, adopting a systemic velocity of 850 $`\mathrm{km}\mathrm{s}^1`$. The source is not resolved in the 0$`\stackrel{}{.}`$15 resolution 5-GHz map (C), and its brightness temperature is too high to be due to free-free emission from ionized gas. The strength of the source allowed us to map it with a resolution of 0$`\stackrel{}{.}`$04 using only the MERLIN observations at 5 GHz (B). Even in this case the source is not resolved, indicating that it is a very strong and compact object (less than 3 pc diameter). This source was considered earlier to be the true nucleus of NGC 2146 (KB). However, the properties of the source are consistent with those found for RSN and SNRs, and because of its steep spectrum and small size, we believe that this source is a RSN (Weiler et al. weiler86 (1986)).
#### 3.1.2 Source 38.9+22.5
38.9+22.5 has also a large negative spectral index $`\alpha _{1.6}^5`$ = -1.1. The object is resolved both at 5 GHz and 1.6 GHz, and has the structure of a partially filled shell of $``$ 30 pc diameter (Fig. 2). As a consequence of its morphology, the size obtained from a Gaussian fit corresponds to its largest angular extent, which may explain the relatively low brightness temperature at 5 GHz. Similar to the previous source we believe that also in this case we are dealing with either a SNR or a RSN. While the steep spectrum indicates it to be a RSN, its large size and the shell-like morphology agree better with that of a SNR.
#### 3.1.3 Source 41.4+15.0
This source has a negative spectral index of $`\alpha _{1.6}^5`$ = - 0.6, again consistent with synchrotron emission. It is not resolved and the derived brightness temperature of 1200 K is a lower limit. The spectral index suggests that this source is a SNR.
### 3.2 Sources with positive spectral indices
There are 6 sources which have a positive spectral index ($`\alpha _{1.6}^5>0`$). Four of them (36.6+27.5, 39.0+17.9, 40.2+18.1 and 37.7+23.7) are not resolved, so that their brightness temperatures are only lower limits. The other two sources (37.9+23.7 and 39.5+17.7) are resolved into amorphous structures. The brightness temperatures, though relatively high, are consistent with free-free emission.
## 4 Discussion
In the following Sects. (4.1 and 4.2) we try to give an explanation of the physical nature of the thermal and non-thermal sources detected in our observations.
### 4.1 Thermal sources (positive spectral index)
A positive spectral index with a slope higher than 0.5 is expected to arise from free-free emission either in the turnover region, where $`\tau 1`$, or in the optically thick region, where $`\tau >1`$. The free-free optical depth in the radio domain is
$$\tau =8.2\times 10^2\left(\frac{\nu }{\mathrm{GHz}}\right)^{2.1}\left(\frac{\mathrm{EM}}{\mathrm{pc}\mathrm{cm}^6}\right)\left(\frac{\mathrm{T}_\mathrm{e}}{\mathrm{K}}\right)^{1.35}$$
(1)
(Mezger & Henderson mezger67 (1967), Carlstrom & Kronberg carlstrom91 (1991), Wills et al. wills98 (1998)), where $`\nu `$ is the observing frequency, $`\mathrm{T}_\mathrm{e}`$ the electron temperature, and EM the emission measure, defined as
$$\mathrm{EM}=_0^\mathrm{L}\left(\frac{\mathrm{n}_\mathrm{e}}{\mathrm{cm}^3}\right)^2\left(\frac{d\mathrm{s}}{\mathrm{pc}}\right),$$
(2)
where $`\mathrm{n}_\mathrm{e}`$ is the electron density and L the pathlength. Using in Eq.(1) the electron temperature T<sub>e</sub> = 8000 K derived from radio recombination line observations of NGC 2146 (Zhao et al. zhao96 (1996)), we obtain for $`\tau `$ = 1 the emission measure $`\mathrm{EM}_{\tau =1}^{1.6\mathrm{GHz}}6\times 10^6`$ $`\mathrm{cm}^6\mathrm{pc}`$ at 1.6 GHz and $`\mathrm{EM}_{\tau =1}^{5\mathrm{GHz}}7\times 10^7`$ $`\mathrm{cm}^6\mathrm{pc}`$ at 5 GHz. For sources with a very steep positive spectral index these values is only a lower limit since for these sources $`\tau >1`$. If the thermal gas is optically thin ($`\tau 0.1`$) and has an electron temperature of 8000 K, the applied source detection limit at 5 GHz of 0.1 mJy corresponds for an unresolved source to an EM of $`4\times 10^6`$. Using the value EM($`\tau `$=1), and for the pathlength L the diameter D of the sources (Table 3), we derive the electron densities given in Table 4.
If we assume that the free–free emission is thermal
bremsstrahlung from H ii regions photoionized by hot, luminous stars, it is possible to estimate the Lyman continuum photon rate $`\mathrm{N}_{\mathrm{Ly}}[\mathrm{s}^1]`$ from the measured radio flux, and from $`\mathrm{N}_{\mathrm{Ly}}`$ it is possible to estimate the number of exciting stars. For optically thin emission, Hjellming & Newell (hjellming83 (1983)) give
$`\mathrm{N}_{\mathrm{Ly}}^{\mathrm{thin}}`$ $`=`$ $`3.1\times 10^{43}\left({\displaystyle \frac{\mathrm{T}_\mathrm{e}}{4000\mathrm{K}}}\right)^{0.16}\left({\displaystyle \frac{\nu }{4.9\mathrm{GHz}}}\right)^{0.11}`$ (3)
$`\times \left({\displaystyle \frac{\mathrm{d}}{180\mathrm{p}\mathrm{c}}}\right)^2\left({\displaystyle \frac{\mathrm{S}_\nu }{7\mathrm{m}\mathrm{J}\mathrm{y}}}\right),`$
where d is the distance of the galaxy; for optically thick emission, Turner et al. (turner98 (1998)) give
$$\mathrm{N}_{\mathrm{Ly}}^{\mathrm{thick}}=4\pi \left(\frac{\mathrm{R}}{\mathrm{cm}}\right)^3\left(\frac{\mathrm{n}_\mathrm{e}}{\mathrm{cm}^3}\right)^2\frac{\alpha _\mathrm{B}}{3},$$
(4)
where $`\alpha _\mathrm{B}`$ is the coefficient for recombinations to all but the ground state of hydrogen (=$`3.2\times 10^{13}`$; tabulated for different electron temperatures and densities in Hummer & Storey hummer87 (1987)). Having thus computed the Lyman continuum photon rate $`\mathrm{N}_{\mathrm{Ly}}`$, we use the compilation of stellar properties by Panagia (panagia73 (1973)) to derive the number of stars embedded in the individual objects observed by us. The number of equivalent O6 stars<sup>2</sup><sup>2</sup>2This is an arbitrary, though often used, unit. Using instead the also common unit of O7V or O5.5 stars will change the number of equivalent stars in Table 4 by $`\pm `$ 50 %, respectively. is given in Table 4.
Are there objects in NGC 2146 which have emission measures of the order of EM $`7\times 10^7`$ $`\mathrm{cm}^6\mathrm{pc}`$, and if so, can there be stellar clusters of 1000 to 2000 O6 stars associated with them?
’Classical’ Galactic H ii regions have EMs smaller than $`10^6`$ $`\mathrm{cm}^6\mathrm{pc}`$. Applying the definition of the emission measure, Carlstrom & Kronberg (carlstrom91 (1991)) derived a hypothetical upper limit of $`4.5\times 10^7`$ $`\mathrm{cm}^6\mathrm{pc}`$ for the EMs of the ionized gas in M 82, assuming that the gas is uniformly distributed throughout a 500 pc disk, with an average density less than 300 $`\mathrm{cm}^3`$ (Duffy et al. duffy87 (1987)). From radio and millimeter observations (Seaquist et al. seaquist85 (1985); Carlstrom & Kronberg carlstrom91 (1991); Wills et al. wills98 (1998)), values of up to $`5\times 10^6`$ $`\mathrm{cm}^6\mathrm{pc}`$ have been derived for M 82.
The EMs required to explain the observations of NGC 2146 are higher and suggest that the sources we are dealing with must have outstanding properties. Large EMs of up to $`10^9`$ $`\mathrm{cm}^6\mathrm{pc}`$ have been found in our galaxy in compact and ultracompact H ii regions (see Churchwell churchwell90 (1990)) with typical sizes of 0.01 pc. High EMs ($`10^7`$-$`10^9`$ $`\mathrm{cm}^6\mathrm{pc}`$) have been recently also found in two galaxies, the starburst dwarf galaxy NGC 5253 (Turner et al. turner98 (1998)) and the blue compact galaxy Henize 2-10 (Kobulnicky & Johnson kobul00 (2000)), in less compact though ultra-dense H ii regions with size between 0.1 and 8 pc. The parameters of our unresolved thermal sources (turnover frequency, size, EM), together with their estimated equivalent number of O6 stars, are consistent with these sources. As far as the resolved thermal sources (37.9+23.7; 39.5+17.7) are concerned, the equivalent number of O6 stars seems clearly unrealistic, especially for 37.9+23.7, which is surely seen in the optically thick regime <sup>3</sup><sup>3</sup>3The sources with less steep spectra belong to the turnover region, so the equivalent number of O6 stars should lie between the values computed for the optically thick and optically thin case (which is a much smaller number).. The star cluster associated with 37.9+23.7 is not seen at optical wavelengths, nor is any other super star cluster.
We thus propose that the objects with positive spectral index are in reality a mixture of thermal and non-thermal sources, too compact to be fully resolved in our map. However, at the single frequency of 5 GHz it is difficult to separate between thermal and non-thermal emission. Furthermore free-free absorption by ionized gas could play a role in obscuring non-thermal sources, such as SNRs and/or RSN, embedded in or located behind optically thick H ii regions (with different sizes and densities). Of course such an explanation presents difficulties in being physically realistic. When we assume that the turnover frequency is equal to or larger than 5 GHz, the EM of the thermal sources should be larger than $`10^8`$ $`\mathrm{cm}^6\mathrm{pc}`$, implying H ii regions that surround the non-thermal objects, with very small sizes (0.01-0.1 pc), but containing hundreds of massive stars. If the thermal sources are instead located in front of, and are not associated with the SNRs or RSN, then the number of stars should be even larger. However, the statistical probability of a line-of-sight coincidence of SNR/RSN and non-associated compact H ii regions should be very small for the six observed sources.
Fortunately, the discovery of very dense thermal sources by Kobulnicky & Johnson (kobul00 (2000)), with sizes up to 8 pc and containing $``$ 750 O7V equivalent stars, makes the possibility less extreme of actually dealing with thermal sources of parsec scales with a very high stellar density. In addition to this, we cannot exclude that the free-free absorption is affecting only the 1.6 GHz observations. In this case the required EM would be $`10^7`$ $`\mathrm{cm}^6\mathrm{pc}`$, and as a consequence the sizes of the H ii regions could be larger, and the stellar density smaller. Better constraints on the exact turnover frequency, the mixture of thermal and non-thermal emission, the Lyman continuum rate from the free-free emission and the equivalent number of O6 stars present both, in the resolved and unresoved sources, can be obtained from high-resolution observations at still higher frequencies (for instance 8 and 15 GHz). At these frequencies the free-free radio emission will originate either from optically thin H ii regions (in this case the spectral index will change to a slope of $``$ -0.1), or from non-thermal sources no longer affected by free-free absorption (in this case the slope will be typical for synchrotron emission).
The presence of a number of compact and ultra-compact H ii regions in NGC 2146 has also been claimed by Puxley et al. (puxley91 (1991)) and Zhao et al. (zhao96 (1996)) using observations of H53$`\alpha `$ (43 GHz) and H92$`\alpha `$ (8 GHz) recombination lines. To explain both, the strength of the 8 GHz continuum and the recombination lines, they require a two–component model of H ii regions with high electron densities ($`10^5`$ cm<sup>-3</sup>) and sizes of 0.2 pc, and lower electron density H ii regions ($`10^310^4`$ cm<sup>-3</sup>) with parsec-scale sizes.
The reason why we do not see such ultra–compact/ultra–dense sources in the starburst galaxy M 82, which in many respects is similar to NGC 2146, is unknown. A sequence for the starburst development has been proposed for some starburst galaxies, including M 82, by Rieke et al. (rieke88 (1988)) and this picture may also apply to NGC 2146. According to this sequence, NGC 5253 for instance, which has a flat radio spectrum and similar compact sources as NGC 2146, is classified to be in a very young phase of a starburst, while M 82 is showing an older phase in which the massive stars have already evolved off the main sequence, with many SNRs clearly visible. NGC 2146 is probably in between these two phases because it still has optically thick H ii regions like those in NGC 5253, and a smaller number of SNRs than M 82, but nevertheless, more similar to M 82, exhibits a global radio spectral index with a non-thermal slope of -0.74.
### 4.2 Non-Thermal sources
#### 4.2.1 Luminosities and energetics
Earlier radio frequency observations of NGC 2146 (McCutcheon mccut73 (1973); KB) show that the galaxy is filled with non-thermal synchrotron emission of spectral index $`\alpha 0.74`$ <sup>4</sup><sup>4</sup>4This value has been confirmed using the total flux density of 475 $`\pm `$ 39 mJy obtained from new single-dish 5 GHz observations of NGC 2146, obtained with the 100m Effelsberg radio telescope of the MPIfR, and the total flux at 20 cm of 1228 mJy taken from the Green Bank 1.4 GHz Northern Sky Survey (White & Becker white92 (1992)).. This situation resembles that of M 82, where the spectral index is $`\alpha 0.6`$ (White & Becker white92 (1992); Kühr et al. kuehr81 (1981)). Following current theories, this synchrotron emission is due to SN, RSN, and SNRs which preferentially occur in the starburst region, as supported by the observed X-ray emission and outflows of hot gas along the minor axis. Using minimum-energy arguments (Miley miley80 (1980)), Muxlow et al. (mux94 (1994)) have computed the number of SNRs ($``$ 2000) necessary to explain the observed flux density at 5 GHz of the extended radio component in the central part of M 82 and consequently the time ($`4\times 10^4`$ yr) necessary to produce such a flux density, assuming a supernova rate of 0.05 $`\mathrm{yr}^1`$. Following this computation, we use the relation given by Longair (longair94 (1994)) to calculate the minimum energy requirements for synchrotron radiation:
$`\mathrm{W}_{\mathrm{min}}`$ $``$ $`8\times 10^6(1+\beta )^{4/7}\left({\displaystyle \frac{\mathrm{V}}{\mathrm{cm}^3}}\right)^{3/7}\left({\displaystyle \frac{\nu }{\mathrm{Hz}}}\right)^{2/7}`$ (6)
$`\times \left({\displaystyle \frac{\mathrm{L}_\nu }{\mathrm{erg}\mathrm{s}^1\mathrm{Hz}^1}}\right)^{4/7}[\mathrm{erg}]`$
where $`\beta `$ is the energy ratio of relativistic protons to electrons, taken to be 1, V is the volume of the considered region, $`\nu `$ is the frequency and $`\mathrm{L}_\nu `$ the luminosity. The 5-GHz flux of the central region of NGC 2146 (assumed to be a disk with a radius of 750 pc and thickness of 100 pc) is 0.25 Jy, which corresponds to a minimum energy of $`5\times 10^{53}\mathrm{erg}`$. To produce this energy, $`10^4`$ evolved SNRs similar to those in the LMC ($`5\times 10^{49}\mathrm{erg}`$ each) are required. A supernova rate of 0.15 $`\mathrm{yr}^1`$ (see Sect. 4.2.2) could produce the extended non-thermal radio emission over a time of $`6\times 10^4`$ yr. The individual events are probably too old and faint still to be seen individually, but their number and/or strength must be higher than in M 82, making the small number of SNRs detected surprising. The strength of the overall magnetic field is not taken into account, though it must play an important role in increasing/decreasing the synchrotron emissivity and the global spectral index of a galaxy.
We also computed the minimum energies and the magnetic field strengths of the three SNR candidates presented in the previous section. The equation for the magnetic field (Longair longair94 (1994)) is
$$\mathrm{B}_{\mathrm{min}}=9.3\times 10^3\left(\frac{\eta \mathrm{L}_\nu }{\mathrm{V}}\right)^{2/7}\nu ^{1/7}[\mathrm{Gauss}]$$
(7)
with the same units as in Eq. 6. The results are compiled in Table 5. Compared with the values obtained by Duric et al. (duric88 (1988)) for non-thermal compact sources in NGC 4736, the agreement of minimum energy requirements and the higher magnetic field strength again favour an identification with fairly young SNRs. The total emitted power at 5 GHz (Table 3) is typically more than ten times higher than those found by Duric et al. (duric88 (1988)) and than that of Cas A. Instead it falls in the upper end of the range of the M 82 sources identified as young SNRs or RSN.
#### 4.2.2 The number of SNRs and RSN in NGC 2146.
The star formation rate (SFR) of NGC 2146, calculated using the extinction-corrected $`\mathrm{H}\alpha `$ flux, is $`6\mathrm{M}_{}\mathrm{yr}^1`$and is three times higher than that of M 82 (Young et al. young88bis (1988)). The same result is obtained from the FIR flux (Moshir et al. moshir90 (1990)), using the relation derived by Thronson & Telesco (thronson86 (1986)) and Thronson et al. (thronson91 (1991)). The values for the SFRs in NGC 2146 and M 82 then are 15 $`\mathrm{M}_{}\mathrm{yr}^1`$ and 5 $`\mathrm{M}_{}\mathrm{yr}^1`$ respectively. Van Buren & Greenhouse (vanburen94 (1994)) use the FIR flux to calculate the supernova rate in M 82 to be 0.06 $`\mathrm{yr}^1`$. Applying this method to NGC 2146 we obtain a supernova rate of 0.15 $`\mathrm{yr}^1`$, which is again higher than the one in M 82. From these results, and despite the fact that the volume of the starburst region in NGC 2146 is larger than that in M 82, we can infer that the bursts in both galaxies have different strengths. As a consequence, we should have a higher number of SNRs and RSN in NGC 2146 than in M 82.
However, when we place the $`40`$ SNRs and RSN detected in M 82 at 5 GHz (Kronberg et al. kron85 (1985); Muxlow et al. mux94 (1994)) at the distance of NGC 2146, given the sensitivity ($`0.1`$ mJy at $`3\sigma `$ level) of our MERLIN + VLA map (C) we expect to find at least 8 sources . The number of sources detected in our observations is only one third of that expected. For 9 sources, however, we do not have spectral indices, and their continuum spectra may be distorted by free-free or synchrotron self-absorption as explained above. A larger number of SNRs than those detected in NGC 2146 would also be expected if, as argued by Armus et al. (armus95 (1995)), SNRs are the possible cause of the hard X-ray component in NGC 2146 We may therefore speculate that the starburst in NGC 2146 is young, or repetitive, so that we are presently observing a strong and young burst still lacking a large number of SNRs and RSN, but instead emitting strong optically thick thermal emission from ionized gas regions, possibly associated with super starclusters. A first, also strong, burst of star formation may have occurred in the past, of which we now observe the intense extended radio component, and which could account for the overall steep non-thermal radio spectrum of the galaxy.
A realistic hypothesis is also that some SNRs and RSN are hidden because of strong free-free absorption, viz. in the ’mixed’ regions mentioned in the previous subsection.<sup>5</sup><sup>5</sup>5For the unresolved sources synchrotron-self absorption that may lead to spectral indices up to 2.5 cannot a priori be excluded. Free-free absorption has in fact been also invoked for several compact sources in M 82 (Wills et al. wills98 (1998)), even though with turnovers at much lower frequencies, with the only exception of 44.01+59.6 that has a positive spectral index at frequencies higher than 1 GHz.
## 5 Conclusions
The combination of VLA and MERLIN 5 GHz observations of the starburst galaxy NGC 2146 allowed us to resolve seven of the eighteen hitherto unresolved sources detected with the VLA. We have measured the spectral indices between 1.6 and 5 GHz for nine sources. Three sources (37.6+24.2, 38.9+22.5, 41.4+15.0) seem to be good candidates for SNRs or RSN, with total emitted powers at 5 GHz comparable to the strongest SNRs found in M 82. Four other sources (36.6+27.5, 39.0+17.9, 40.2+18.1 and 37.7+23.7) have instead been identified with compact or ultra-dense H ii regions that have large EMs and a high-frequency turnover. The emitted thermal emission indicates a large equivalent number of O6 stars, exceeding the number of those found in similar objects in other galaxies. The large number of O6 stars may, however, be consistent with stars in super starclusters. The nature of the last two sources (37.9+23.7 and 39.5+17.7) for which we have spectral index information are still a matter of discussion. It seems realistic that they emit a mixture of thermal and non-thermal radiation, and may manifest the most powerful H ii regions ever observed. However, the burst of star formation in NGC 2146 has an extraordinary power, stronger than that in M 82. Despite this and the fact that the overall radio spectral index of the galaxy reveals a non-thermal origin, the number of discovered SNRs is much smaller than that in M 82. For the explanation of this fact we propose two scenarios:
* NGC 2146 is experiencing a burst of star formation stronger than that in M 82, but which we are observing in a younger phase. This phase still lacks a large number of SNRs, but instead gives rise to strong thermal emission from optically thick dense ionized gas regions, associated with massive or super starclusters. The origin of the non-thermal extended component must be a trace of past SN explosions, individually too old and faint to be seen, and belonging to a previous burst in the galaxy.
* Strong free-free absorption by very dense H ii regions and, for the most compact sources, possibly also
synchrotron-self absorption, may conceal a certain number of SNRs or RSN, rendering the identification of their nature difficult if only based on measurements of the spectral indices around the turnover frequencies. In this case the global number of SNRs or RSN can be higher, and this would corroborate the similarity between NGC 2146 and M 82. This interpretation raises, however, the question why we do not observe similar compact or ultra-dense H ii regions in M 82 as well.
Observations at 15 GHz will be able to better clarify the nature of the sources, and shed light on the unusual phase of the star burst we are observing in NGC 2146.
###### Acknowledgements.
We would like to acknoledge Dr. Peter Thomasson for his helpful support and Dr. Peter Biermann for useful discussions. We are grateful to the referee Dr. Jean Turner for constructive comments and valuable suggestions. This work was supported by the DFG through the Graduiertenkolleg ’The Magellanic System, Galaxy Interaction and the Evolution of Dwarf Galaxies’. This research has made use of the NASA/IPAC Extragalactic Database (NED) which is operated by the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration.
|
warning/0004/cond-mat0004053.html
|
ar5iv
|
text
|
# On the Bragg, Leibfried, and Modified Leibfried Numbers
## Abstract
The Bragg, Leibfried, and modified Leibfried numbers are defined in the context of a theory of dislocation-mediated melting, and their values are determined from the properties of the dislocation ensemble at the melting temperature. The approximate numerical coincidence of the Bragg and modified Leibfried numbers is explained. The parameter $`K`$ in the definition of the modified Leibfried number is shown to be the natural logarithm of the effective coordination number. Our analysis reveals that the Bragg number can be considered an elemental constant, in contrast to the Leibfried and modified Leibfried numbers.
Key words: melting, dislocation, Bragg, Leibfried, modified Leibfried
PACS: 61.72.Bb, 61.72.Lk, 64.70.Dv, 64.90.+b
In Leibfried’s study of melting he observed that for a number of metals (specifically Al, Ag, Au, Cu, Pd and Pb)
$$LRT_m/GV0.042,$$
(1)
where $`R`$ is the gas constant, $`T_m`$ is the melting temperature, $`G`$ is the ambient shear modulus, and $`V`$ is the molar volume. The factor $`RT_m`$ is an approximation to the heat of fusion, $`L_m,`$ which follows from Richard’s rule
$$L_m/T_m=\mathrm{}S_m2.0\mathrm{e}.\mathrm{u}.R$$
(2)
(1 e.u. $`=1`$ cal/mol K), since the numerical value of $`R`$ is 1.987 e.u. Bragg , on the other hand, noted that
$$BL_m/GV0.034$$
(3)
for a few metals (specifically Al, Ag, Au, Co, Cu, Fe, Ni and Pb).
Although $`L,`$ the Leibfried number, and $`B,`$ the Bragg number, seem to be intrinsically related, through Richard’s rule, there is a relatively large discrepancy between their values. The importance of both numbers in the calculation of the size factors caused Gschneidner to reanalyze them in his review . First, he noticed that Richard’s rule may not be a valid approximation. In fact, Stull and Sinke had earlier used two different values for estimating heats of fusion, 2.3 e.u. for elements which crystallize as face-centered cubic (fcc) or hexagonal close-packed (hcp) structures, and 1.9 e.u. for elements which have body-centered cubic (bcc) structure. Second, Gschneidner introduced the modified Leibfried number, $`L^{},`$ which differs from the Leibfried number $`L`$ in that the term $`RT_m`$ is replaced by the term $`KT_m.`$ The value of $`K`$ depends on the crystal structure of the element just below its melting point: 2.29 e.u. for fcc or hcp metals, 1.76 e.u. for bcc metals, etc. Finally, he concluded that the modified Leibfried and Bragg numbers do agree within uncertainties.
In this Technical Note we offer an explanation for the numerical values of both $`L^{}`$ and $`B`$ in a theory of dislocation-mediated melting. We obtain the numerical values of $`B,`$ $`L`$ and $`L^{}`$ from the properties of the dislocation ensemble at the melting point. We also explain the approximate numerical coincidence of the values of $`L^{}`$ and $`B.`$
In our previous study of melting as a dislocation-mediated phase transition on a lattice , we obtained two relations:
$$k_BT_m=\frac{\lambda \kappa Gv_{WS}}{8\pi \mathrm{ln}z^{}}\mathrm{ln}\left(\frac{\alpha ^2}{4b^2\rho (T_m)}\right),$$
(4)
$$L_m=\frac{1}{\lambda }b^2\rho (T_m)RT_m\mathrm{ln}z^{},$$
(5)
where $`\rho (T_m)`$ is the critical dislocation density at melt, $`v_{WS}`$ is the Wigner-Seitz volume, $`1/\kappa =(1\nu /2)\pm \nu /25/6,`$ $`\nu `$ being the Poisson ratio, $`b`$ is the shortest perfect-dislocation Burgers vector, $`\lambda b^3/v_{WS}4/3,`$ and $`\alpha =2.9`$ accounts for nonlinear effects in a dislocation core. Also, $`z^{},`$ the effective coordination number for a dislocation as a random walk on the lattice, satisfies $`z^{}z,`$ where $`z`$ is the coordination number of the lattice. These relations hold to $`20`$% accuracy for those elements of the Periodic Table (more than half) for which sufficient data exist.
It then follows that the Bragg number is
$$B=\frac{\kappa }{8\pi }b^2\rho (T_m)\mathrm{ln}\left(\frac{\alpha ^2}{4b^2\rho (T_m)}\right).$$
(6)
We have found that the critical dislocation density at melt is an elemental constant with an approximate numerical value of $`2/3b^2`$ for three-quarters of the Periodic Table. It corresponds to the situation where, on average, half of the atoms are within a dislocation core at melt. Hence, the Bragg number must be an elemental constant, and its approximate numerical value is, in view of Eq. (6) with the numerical values of the parameters quoted above,
$$B\frac{0.92}{8\pi }0.037,$$
(7)
in agreement with Bragg’s original estimate. We discuss this point in more detail below.
Similarly, the Leibfried number is
$$L=\frac{\lambda \kappa }{8\pi \mathrm{ln}z^{}}\mathrm{ln}\left(\frac{\alpha ^2}{4b^2\rho (T_m)}\right).$$
(8)
It depends, through $`\mathrm{ln}z^{},`$ on the crystal structure of the solid phase of the element from which melting occurs, in agreement with Gschneidner’s observation. It is, however, possible to define modified Leibfried numbers that approximately coincide numerically with the Bragg number :
$$L^{}\frac{K}{R}L,$$
(9)
where $`K`$ may be defined in either of two ways:
$$K=\frac{2b^2\rho (T_m)\mathrm{ln}z^{}}{\lambda }\mathrm{e}.\mathrm{u}.$$
(10)
or
$$K=\mathrm{ln}z^{}\mathrm{e}.\mathrm{u}.$$
(11)
In the first case,
$$L^{}=\frac{\kappa }{4\pi }b^2\rho (T_m)\mathrm{ln}\left(\frac{\alpha ^2}{4b^2\rho (T_m)}\right)\frac{\mathrm{e}.\mathrm{u}.}{R}$$
(12)
will approximately coincide numerically with $`B`$ given in Eq. (6), since the numerical value of $`R`$ in units of e.u. is very close to 2. In the second case,
$$L^{}=\frac{\lambda \kappa }{8\pi }\mathrm{ln}\left(\frac{\alpha ^2}{4b^2\rho (T_m)}\right)\frac{\mathrm{e}.\mathrm{u}.}{R}$$
(13)
will again approximately coincide numerically with $`B`$ given in (6), since $`R2`$ e.u., $`b^2\rho (T_m)2/3`$ and $`\lambda 4/3.`$
In the first case, the numerical values of $`K`$ in e.u., according to Eq. (10), are 2.26 for fcc and hcp metals (for which $`\lambda =\sqrt{2}`$ and $`z=12)`$ and 1.99 for bcc metals (for which $`\lambda 1.3`$ and $`z=8),`$ where we have taken $`b^2\rho (T_m)=2/3`$ and $`z^{}=z1`$ . These values of $`K`$ are consistent with those presented by Gschneidner, 2.29 and 1.76, respectively. In the second case, the corresponding values for $`z^{}=z1`$ are 2.40 for fcc and hcp metals and 1.95 for bcc metals, and for $`z^{}=z2`$ these values are 2.30 for fcc and hcp metals and 1.79 for bcc metals. It is therefore seen that in the second case the choice $`z^{}=z2`$ leads to better agreement between the values of $`K`$ calculated from Eq. (11) and those given by Gschneidner. This is not sufficient, however, to conclude that the choice $`z^{}=z2`$ is better than the choice $`z^{}=z1`$ used in , since Gschneidner did not specify the criteria for his choice of the values of $`K.`$ Equation (5) with $`z^{}=z1,`$ for example, explains the Stull-Sinke rule, since it gives $`L_m/T_m=2.25`$ e.u. for fcc and hcp metals, and 1.98 e.u. for bcc metals, in good agreement with the values 2.3 e.u. and 1.9 e.u. used by Stull and Sinke. In any case, the uncertainty in the value of $`\mathrm{ln}z^{}`$ associated with the choice between $`\mathrm{ln}(z1)`$ and $`\mathrm{ln}(z2)`$ does not exceed 8%, which is well within the $`20`$% uncertainty of Eqs. (4),(5).
In ref. , for three-quarters of the Periodic Table, we established that
$$b^2\rho (T_m)=0.64\pm 0.14.$$
(14)
It then follows from Eqs. (3)-(5) with $`z^{}=z1`$ and the numerical values of the parameters quoted above, that
$$B=0.0369\pm 0.0093,$$
(15)
where most of the error comes from uncertainty in the value of $`\kappa .`$ Hence, uncertainty in the numerical value of $`B`$ is about 25%, which is sufficiently small that $`B`$ can be considered an elemental constant.
As follows from Eqs. (1), (4) with $`z^{}=z1,`$ and Eq. (14),
$$L=0.0308\pm 0.0113,$$
(16)
where we have used $`z=8`$ and 12 to fix the error bar. Also, from using this value of $`L`$, the modified Leibfried number defined by Eqs. (9) and (11) with $`z^{}=z1`$ is
$$L^{}=0.0337\pm 0.0119,$$
(17)
where, again, the error reflects the two values of $`z.`$ These values of $`L`$ and $`L^{}`$ are in agreement with the values obtained in ref. for almost all the elements of the Periodic Table: $`L=0.0305\pm 0.0135,`$ $`L^{}=0.0334\pm 0.0145.`$ Although the values of $`B,`$ $`L,`$ and $`L^{}`$ do agree within errors, uncertainties associated with both $`L`$ and $`L^{}`$ are about 35% which is so large that they cannot be considered elemental constants.
To conclude, we have defined and evaluated the Bragg, Leibfried and modified Leibfried numbers in the theory of dislocation-mediated melting. We have explained the approximate numerical coincidence of the Bragg and modified Leibfried numbers. We have shown that the Bragg number can be considered an elemental constant, in contrast to the Leibfried and modified Leibfried numbers. In our analysis, the quantity $`\mathrm{ln}z^{}`$ is identified with Gschneidner’s constant $`K.`$ The factor $`\mathrm{ln}z^{}`$ can only arise naturally in a theory of line-like defects.
|
warning/0004/math0004108.html
|
ar5iv
|
text
|
# New Numerical Algorithm for Modeling of Boson-Fermion Stars in Dilatonic Gravity This work was supported by Bulgarian National Scientific Fund, Contr. NoNo F610/99, MM602/96 and by Sofia University Research Fund, Contr. No. 245/99.
## 1 Introduction
In this paper we present a numerical method for solving the equations of the general scalar-tensor theories of gravity including a dilaton potential term for the general case of mixed boson-fermion star. This method is an impruvment of the method concerning the similar problem proposed in our recent work . The original domain is splitted to two domains: inner inside the star and external outside the star. The solutions in these regions are obtained separately and after that they are matched. Before we go to the substantial part of this paper we will describe briefly the physical problem.
## 2 Main model
Boson stars are gravitationally bound macroscopic quantum states made up of scalar bosons . They differ from the usual fermionic stars in that they are only prevented from collapsing gravitationally by the Heisenberg uncertainty principle. For self-interacting boson field the mass of the boson star, even for small values of the coupling constant turns out to be of the order of Chandrasekhar’s mass when the boson mass is similar to a proton mass. Thus, the boson stars arise as possible candidates for non-baryonic dark matter in the universe and consequently as a possible solution of the one of outstanding problems in today’s astrophysics - the problem of nonluminous matter in the universe. Most of the stars are of primordial origin being formed from an original gas of fermions and bosons in the early universe. That is why it should be expected that most stars are a mixture of both, fermions and bosons in different proportions.
Boson-fermion stars are also good model to learn more about the nature of strong gravitational fields not only in general relativity but also in the other theories of gravity.
The most natural and promising generalizations of general relativity are the scalar-tensor theories of gravity . In these theories the gravity is mediated not only by a tensor field (the metric of space-time) but also by a scalar field (the dilaton). These dilatonic theories of gravity contain arbitrary functions of the scalar field that determine the gravitational “constant” as a dynamical variable and the strength of the coupling between the scalar field and matter. It should be stressed that specific scalar-tensor theories of gravity arise naturally as low energy limit of string theory which is the most promising modern model of unification of all fundamental physical interactions.
Boson stars in scalar-tensor theories of gravity with massless dilaton have been widely investigated recently both numerically and analytically . Mixed boson-fermion stars in scalar tensor theories of gravity however have not been investigated so far in contrast to general relativistic case where boson-fermion stars have been investigated .
In present paper we consider boson-fermion stars in the most general scalar-tensor theory of gravity with massive dilaton.
In the Einstein frame the field equations in presence of fermion and boson matter are:
$`G_m^n=\kappa _{}(\stackrel{B}{T_m^n}+\stackrel{F}{T_m^n})+2_m\phi ^n\phi `$ $`^l\phi _l\phi \delta _m^n+{\displaystyle \frac{1}{2}}U(\phi )\delta _m^n,`$
$`_m^m\phi +{\displaystyle \frac{1}{4}}U^{}(\phi )`$ $`={\displaystyle \frac{\kappa _{}}{2}}\alpha (\phi )(\stackrel{B}{T}+\stackrel{F}{T}),`$
$`_m^m\mathrm{\Psi }+2\alpha (\phi )^l\phi _l\mathrm{\Psi }`$ $`=2A^2(\phi ){\displaystyle \frac{\stackrel{~}{W}(\mathrm{\Psi }^+\mathrm{\Psi })}{\mathrm{\Psi }^+}},`$
$`_m^m\mathrm{\Psi }^++2\alpha (\phi )^l\phi _l\mathrm{\Psi }^+`$ $`=2A^2(\phi ){\displaystyle \frac{\stackrel{~}{W}(\mathrm{\Psi }^+\mathrm{\Psi })}{\mathrm{\Psi }^+}}`$
where $`_m`$ is the Levi-Civita connection with respect to the metric $`g_{mn},(m=0,\mathrm{},3;`$ $`n=0,\mathrm{},3)`$. The constant $`\kappa _{}`$ is given by $`\kappa _{}=8\pi G_{}`$ where $`G_{}`$ is the bare Newtonian gravitational constant. The physical gravitational “constant” is $`G_{}A^2(\phi )`$ where $`A(\phi )`$ is a function of the dilaton field $`\phi `$ depending on the concrete scalar-tensor theory of gravity. $`\stackrel{~}{W}(\mathrm{\Psi }^+\mathrm{\Psi })`$ is the potential of boson field. The dilaton potential $`U(\phi )`$ can be written in the form $`U(\phi )=m_D^2V(\phi )`$ where $`m_D`$ is the dilaton mass and $`V(\phi )`$ is dimensionless function of $`\phi `$.
The function $`\alpha (\phi )=\frac{d}{d\phi }\left[\mathrm{ln}A(\phi )\right]`$ determines the strength of the coupling between the dilaton field $`\phi `$ and the matter. The functions $`\stackrel{B}{T}`$ and $`\stackrel{F}{T}`$ are correspondingly the trace of the energy-momentum tensor of the fermionic matter<sup>1</sup><sup>1</sup>1In the present article we consider fermionic matter only in macroscopic approximation, i.e., after averaging quantum fluctuations of the corresponding fermion fields. Thus we actually consider standard classical relativistic matter. $`\stackrel{F}{T_m^n}`$ and bosonic matter $`\stackrel{B}{T_m^n}`$. Their explicit forms are
$`\stackrel{B}{T_m^n}=`$ $`{\displaystyle \frac{1}{2}}A^2(\phi )\left(_m\mathrm{\Psi }^+^n\mathrm{\Psi }+_m\mathrm{\Psi }^n\mathrm{\Psi }^+\right)`$ (2)
$`{\displaystyle \frac{1}{2}}A^2(\phi )\left[_l\mathrm{\Psi }^+^l\mathrm{\Psi }2A^2(\phi )\stackrel{~}{W}(\mathrm{\Psi }^+\mathrm{\Psi })\right]\delta _m^n,`$
$`\stackrel{F}{T_m^n}=`$ $`\left(\epsilon +p\right)u_mu^np\delta _m^n.`$ (3)
Here $`\mathrm{\Psi }`$ is a complex scalar field describing the bosonic matter while $`\mathrm{\Psi }^+`$ is its complex conjugated function. The energy density and the pressure of the fermionic fluid in the Einstein frame are $`\epsilon =A^4(\phi )\stackrel{~}{\epsilon }`$ and $`p=A^4(\phi )\stackrel{~}{p}`$ where $`\stackrel{~}{\epsilon }`$ and $`\stackrel{~}{p}`$ are the physical energy density and pressure. Instead to give the equation of state of the fermionic matter in the form $`\stackrel{~}{p}=\stackrel{~}{p}(\stackrel{~}{\epsilon })`$ it is more convenient to write it in a parametric form
$$\stackrel{~}{\epsilon }=\stackrel{~}{\epsilon }_0g(\mu )\stackrel{~}{p}=\stackrel{~}{\epsilon }_0f(\mu )$$
(4)
where $`\stackrel{~}{\epsilon }_0`$ is a properly chosen dimensional constant and $`\mu `$ is dimensionless Fermi momentum.
The physical four-velocity of the fluid is denoted by $`u_\mu `$. The potential for the boson field has the form
$$\stackrel{~}{W}(\mathrm{\Psi }^+\mathrm{\Psi })=\frac{m_D^2}{2}\mathrm{\Psi }^+\mathrm{\Psi }\frac{1}{4}\stackrel{~}{\mathrm{\Lambda }}(\mathrm{\Psi }^+\mathrm{\Psi })^2.$$
Field equations together with the Bianchi identities lead to the local conservation law of the energy-momentum of matter
$$_n\stackrel{F}{T_m^n}=\alpha (\phi )\stackrel{F}{T}_m\phi .$$
(5)
We will consider a static and spherically symmetric mixed boson-fermion star in asymptotic flat space-time. This means that the metric $`g_{mn}`$ has the form
$$ds^2=\mathrm{exp}\left[\nu ()\right]dt^2\mathrm{exp}\left[\lambda ()\right]d^2^2\left(d\theta ^2+\mathrm{sin}^2\theta d\psi ^2\right)$$
(6)
where $`,\theta ,\psi `$ are usual spherical coordinates. The field configuration is static when the boson field $`\mathrm{\Psi }`$ satisfies the relationship
$$\mathrm{\Psi }=\stackrel{~}{\sigma }()\mathrm{exp}(i\omega t).$$
Here $`\omega `$ is a real number and $`\stackrel{~}{\sigma }()`$ is a real function. Taking into account the assumption that have been made the system of the field equation is reduced to a system of ordinary differential equations. Before we explicitly write the system we are going to introduce a rescaled (dimensionless) radial coordinate by $`r=m_B`$ where $`m_B`$ is the mass of the bosons. From now on, a prime will denote a differentiation with respect to the dimensionless radial coordinate $`r`$. After introducing the dimensionless quantities
$$\mathrm{\Omega }=\frac{\omega }{m_B},\sigma =\sqrt{\kappa }_{}\stackrel{~}{\sigma },\mathrm{\Lambda }=\frac{\stackrel{~}{\mathrm{\Lambda }}}{\kappa _{}m_{}^{2}{}_{B}{}^{}},\gamma =\frac{m_D}{m_B}.$$
and defining the dimensionless energy-momentum tensors as $`T_m^n=\frac{\kappa _{}}{m_B^2}T_m^n`$ the components of the dimensionless energy-momentum tensor of the fermionic and bosonic matter become correspondingly
$`\stackrel{F}{T_0^0}=bA^4(\phi )g(\mu ),\stackrel{F}{T_1^1}=bA^4(\phi )f(\mu ),`$
$`\stackrel{B}{T_0^0}`$ $`={\displaystyle \frac{1}{2}}A^2(\phi )\left[\mathrm{\Omega }^2\sigma ^2(r)\mathrm{exp}(\nu (r))+\left({\displaystyle \frac{\sigma }{dr}}\right)^2\mathrm{exp}(\lambda (r))\right]A^4(\phi )W(\sigma ^2),`$
$`\stackrel{B}{T_1^1}`$ $`={\displaystyle \frac{1}{2}}A^2(\phi )\left[\mathrm{\Omega }^2\sigma ^2(r)\mathrm{exp}(\nu (r))+\left({\displaystyle \frac{\sigma }{r}}\right)^2\mathrm{exp}(\lambda (r))\right]A^4(\phi )W(\sigma ^2).`$
Here the parameter $`b=\frac{\kappa _{}\stackrel{~}{\epsilon }_0}{m_B^2}`$ describes the relation between the Compton length of dilaton and the usual radius of neutron star in general relativity.
The functions $`\stackrel{B}{T}`$ and $`\stackrel{F}{T}`$ describing the trace of energy-momentum tensor have a form:
$$\stackrel{B}{T}=A^2(\phi )[\mathrm{\Omega }^2\sigma ^2(r)\mathrm{exp}(\nu (r))\left(\frac{\sigma }{dr}\right)^2\mathrm{exp}(\lambda (r))]4A^4(\phi )W(\sigma ^2),$$
$$\stackrel{F}{T}=bA^4(\phi )[g(\mu )3f(\mu )].$$
For the independent (dimensionless) radial coordinate $`r`$ we have $`r[0,R_s][R_s,\mathrm{})`$ where $`0<R_s<\mathrm{}`$ is the unknown radius of the fermionic part of the mixed boson-fermion star.
With all definitions we have given the main system of differential equations governing the structure of a static and spherically symmetric boson-fermion star can be written in the following form:
1. In the interior of the fermionic part of the star (the functions in this domain are subscribed by $`i`$)
$`{\displaystyle \frac{d\lambda }{dr}}`$ $`=F_{1,i}{\displaystyle \frac{1\mathrm{exp}(\lambda )}{r}}+r\{\mathrm{exp}(\lambda )[\stackrel{F}{T_0^0}+\stackrel{B}{T_0^0}+{\displaystyle \frac{1}{2}}\gamma ^2V(\phi )]+\left({\displaystyle \frac{d\phi }{dr}}\right)^2\},`$
$`{\displaystyle \frac{d\nu }{dr}}`$ $`=F_{2,i}{\displaystyle \frac{1\mathrm{exp}(\lambda )}{r}}r\{\mathrm{exp}(\lambda )[\stackrel{F}{T_1^1}+\stackrel{B}{T_1^1}+{\displaystyle \frac{1}{2}}\gamma ^2V(\phi )]\left({\displaystyle \frac{d\phi }{dr}}\right)^2\},`$
$`{\displaystyle \frac{d^2\phi }{dr^2}}`$ $`=F_{3,i}{\displaystyle \frac{2}{r}}{\displaystyle \frac{d\phi }{dr}}+{\displaystyle \frac{1}{2}}\left(F_{1,i}F_{2,i}\right){\displaystyle \frac{d\phi }{dr}}`$ (7)
$`+{\displaystyle \frac{1}{2}}\mathrm{exp}(\lambda )[\alpha (\phi )(\stackrel{F}{T}+\stackrel{B}{T})+{\displaystyle \frac{1}{2}}\gamma ^2V^{^{}}(\phi )],`$
$`{\displaystyle \frac{d^2\sigma }{dr^2}}`$ $`=F_{4,i}{\displaystyle \frac{2}{r}}{\displaystyle \frac{d\sigma }{dr}}+\left[{\displaystyle \frac{1}{2}}\left(F_{1,i}F_{2,i}\right)2\alpha (\phi ){\displaystyle \frac{d\phi }{dr}}\right]{\displaystyle \frac{d\sigma }{dr}}`$
$`\sigma \mathrm{exp}(\lambda )\left[\mathrm{\Omega }^2\mathrm{exp}(\nu )+2A^2(\phi )W^{^{}}(\sigma ^2)\right],`$
$`{\displaystyle \frac{d\mu }{dr}}`$ $`=F_{5,i}{\displaystyle \frac{g(\mu )+f(\mu )}{f^{^{}}(\mu )}}\left[{\displaystyle \frac{1}{2}}F_2+\alpha (\phi ){\displaystyle \frac{d\phi }{dr}}\right].`$
Here $`\lambda (r),\nu (r),\phi (r),\sigma (r)`$ and $`\mu (r)`$ are unknown functions of $`r`$ and $`\mathrm{\Omega }`$ is a unknown parameter. Having in mind the physical assumptions we have to solve the equations (7) under following boundary conditions:
$$\lambda (0)=\frac{d\phi }{dr}(0)=\frac{d\sigma }{dr}(0)=0,\sigma (0)=\sigma _c,\mu (0)=\mu _c,$$
(8)
$$\mu (R_s)=0$$
(9)
where $`\sigma _c`$ and $`\mu _c`$ are the values of density of the bosonic and fermionic matter, respectively at the star’s centre. The first three conditions in (8) guarantee the nonsingularity of the metrics and the functions $`\lambda (r),`$ $`\phi (r),`$ $`\sigma (r)`$ at the star’s centre.
2. In the external domain (subscribed by $`e`$) there is not fermionic matter, i.e., one can suppose formally that the function $`\mu (r)0`$ if $`xR_s`$. The fermionic part of the energy-momentum tensor vanishes identically also and thus the differential equations with respect to the rest four unknown functions $`\lambda (r),`$ $`\nu (r),`$ $`\phi (r)`$ and $`\sigma (r)`$ are:
$`{\displaystyle \frac{d\lambda }{dr}}`$ $`=F_{1,e}{\displaystyle \frac{1\mathrm{exp}(\lambda )}{r}}+r\{\mathrm{exp}(\lambda )[\stackrel{B}{T_0^0}+{\displaystyle \frac{1}{2}}\gamma ^2V(\phi )]+\left({\displaystyle \frac{d\phi }{dr}}\right)^2\},`$
$`{\displaystyle \frac{d\nu }{dr}}`$ $`=F_{2,e}{\displaystyle \frac{1\mathrm{exp}(\lambda )}{r}}r\{\mathrm{exp}(\lambda )[\stackrel{B}{T_1^1}+{\displaystyle \frac{1}{2}}\gamma ^2V(\phi )]\left({\displaystyle \frac{d\phi }{dr}}\right)^2\},`$
$`{\displaystyle \frac{d^2\phi }{dr^2}}`$ $`=F_{3,e}{\displaystyle \frac{2}{r}}{\displaystyle \frac{d\phi }{dr}}+{\displaystyle \frac{1}{2}}\left(F_{1,e}F_{2,e}\right){\displaystyle \frac{d\phi }{dr}}`$ (10)
$`+{\displaystyle \frac{1}{2}}\mathrm{exp}(\lambda )\left[\alpha (\phi )\stackrel{B}{T}+{\displaystyle \frac{1}{2}}\gamma ^2V^{^{}}(\phi )\right],`$
$`{\displaystyle \frac{d^2\sigma }{dr^2}}`$ $`=F_{4,e}{\displaystyle \frac{2}{r}}{\displaystyle \frac{d\sigma }{dr}}+\left[{\displaystyle \frac{1}{2}}\left(F_{1,e}F_{2,e}\right)2\alpha (\phi ){\displaystyle \frac{d\phi }{dr}}\right]{\displaystyle \frac{d\sigma }{dr}}`$
$`\sigma \mathrm{exp}(\lambda )\left[\mathrm{\Omega }^2\mathrm{exp}(\nu )+2A^2(\phi )W^{^{}}(\sigma ^2)\right].`$
As it is required by the asymptotic flatness of space-time the boundary conditions at the infinity are
$$\nu (\mathrm{})=0,\phi (\mathrm{})=0,\sigma (\mathrm{})=0$$
(11)
where we denote $`()(\mathrm{})=lim_r\mathrm{}()(r)`$.
We seek for a solution $`[\lambda (r),\nu (r),\phi (r),\sigma (r),\mu (r),R_s,\mathrm{\Omega }]`$ subjected to the nonlinear ODEs (7) and (10), satisfying the boundary conditions (8), (9) and (11). At that we assume the function $`\mu (r)`$ is continuous in the interval $`[0,R_s],`$ whilst the functions $`\lambda (r),`$ $`\nu (r)`$ are continuous and the functions $`\phi (r),`$ $`\sigma (r)`$ are smooth in the whole interval $`[0,\mathrm{})`$, including the unknown inner boundary $`r=R_s`$.
The so posed BVP is a two-parametric eigenvalue problem with respect to the quantities $`R_s`$ and $`\mathrm{\Omega }`$.
Let us emphasize that a number of methods for solving the free-boundary problems are considered in detail in .
Here we aim at applying the new solving method to the above formulated problem. This method differs from that one proposed in and for the governing field equations written in the forms (7) and (10) it possesses certain advantage.
## 3 Method of solution
At first we scale the variable $`r`$ using the Landau transformation and in this way we obtain a fixed computational domain. Namely
$$x=\frac{r}{R_s},x[0,1][1,\mathrm{}).$$
For our further considerations it is convenient to present the systems (7) and (10) in following equivalent form as systems of first order ODEs:
$`𝐲_i^{}+R_s𝐅_i(R_sx,𝐲_i,\mathrm{\Omega })`$ $`=`$ $`0,`$ (12)
$`𝐲_e^{}+R_s𝐅_e(R_sx,𝐲_e,\mathrm{\Omega })`$ $`=`$ $`0`$ (13)
with respect to the unknown vector functions
$$𝐲_i(x)(\lambda (x),\nu (x),\phi (x),\xi (x),\sigma (x),\eta (x),\mu (x))^T,$$
$$𝐲_e(x)(\lambda (x),\nu (x),\phi (x),\xi (x),\sigma (x),\eta (x))^T$$
and right hand sides $`𝐅_i(F_1,F_2,\xi ,F_3,\eta ,F_4,F_5)^T`$, $`𝐅_e(F_1,F_2,\xi ,F_3,\eta ,F_4)^T`$ where $`(.)^{}`$ stands for differentiation towards the new variable $`x`$.
For given values of the parameters $`R_s`$ and $`\mathrm{\Omega }`$ the independent solving of the inner system (12) requires seven boundary conditions. At the same time we have at disposal only six conditions of the kind (8) and (9). In order to complete the problem we set additionally one more parametric condition (the value of someone from among the functions $`\lambda (x),\nu (x),\phi (x),\xi (x),\sigma (x)`$ or $`\eta (x)`$) at the point $`x=1)`$. Let us set for example
$$\phi _i(1)=\phi _s$$
(14)
where $`\phi _s`$ is a parameter. Then the boundary conditions (8), (9) and (14) of the inner BVP can be presented in the form:
$$B_{0,i}𝐲_i(0)D_{0,i}=0,B_{1,i}𝐲_i(1)D_{1,i}(\phi _s)=0.$$
(15)
Here the matrices $`B_{0,i}=diag(1,0,0,1,1,1,1)`$, $`D_{0,i}=diag(0,0,0,0,\sigma _c,0,\mu _c)`$, $`B_{1,i}=diag(0,0,1,0,0,0,1)`$, $`D_{1,i}=diag(0,0,\phi _s,0,0,0,0)`$.
Obviously the solution in the inner domain $`x[0,1]`$ depends not only on the variable $`x,`$ but it is a function of the three parameters $`R_s,\mathrm{\Omega },\phi _s`$ as well, i.e., $`𝐲_i=𝐲_i(x,\mathrm{\Omega },R_s,\phi _s).`$
In the external domain $`x1`$ the vector of solutions
$$𝐲_e(x)(\lambda (x),\nu (x),\phi (x),\xi (x),\sigma (x),\eta (x))^T$$
is 6D. Thereupon six boundary conditions are indispensable to solving of the equation (13). At the same time the three boundary conditions (11) are only known. Let us consider that the solution $`𝐲_i(x)`$ in the inner domain $`x[0,1]`$ is knowledged. Then we postulate the rest three deficient conditions to be the continuity conditions at the point $`x=1.`$The first of them is similar to the condition (14) and the else two we assign to arbitrary two functions from among $`\lambda (x),`$ $`\nu (x),\xi (x),`$ $`\sigma (x)`$ and $`\eta (x)`$, for example
$$\lambda _e(1)=\lambda _i(1),\phi _e(1)=\phi _s,\sigma _e(1)=\sigma _i(1).$$
It is convenient to present the boundary conditions in the external domain in matrix form again:
$$B_{1,e}𝐲_e(1)D_{1,e}(\phi _s)=0,B_{\mathrm{},e}𝐲_e(\mathrm{})=0$$
(16)
where the matrices $`B_{1,e}=diag(1,0,1,0,1,0)`$, $`D_{1,e}=diag(\lambda _i(1),0,\phi _s,0,\sigma _i(1),0)`$, $`B_{\mathrm{},e}=diag(0,1,1,0,1,0)`$.
Let the solutions $`𝐲_i=𝐲_i(x,\mathrm{\Omega },R_s,\phi _s)`$ and $`𝐲_e=𝐲_e(x,\mathrm{\Omega },R_s,\phi _s)`$ be supposed known. Generally speaking for given arbitrary values of the parameters $`R_s,\mathrm{\Omega }`$ and $`\phi _s`$ the continuity conditions with respect to the functions $`\nu (x),\xi (x)`$ and $`\eta (x)`$ at the point $`x=1`$ are not satisfied. We choose the parameters $`R_s,\mathrm{\Omega }`$ and $`\phi _s`$ in such manner that the continuity conditions for the functions $`\nu (x)`$, $`\xi (x)`$ and $`\eta (x)`$ to be held, i.e.,
$`\nu _e(1,R_s,\mathrm{\Omega },\phi _s)\nu _i(1,R_s,\mathrm{\Omega },\phi _s)`$ $`=0,`$
$`\xi _e(1,R_s,\mathrm{\Omega },\phi _s)\xi _i(1,R_s,\mathrm{\Omega },\phi _s)`$ $`=0,`$ (17)
$`\eta _e(1,R_s,\mathrm{\Omega },\phi _s)\eta _i(1,R_s,\mathrm{\Omega },\phi _s)`$ $`=0.`$
These conditions should be interpreted as three nonlinear algebraic equations in regard to the unknown quantities $`R_s,\mathrm{\Omega }`$ and $`\phi _s`$. The usual way for solving of the above mentioned kind of equations (17) is by means of various iteration methods, for example Newton’s methods. The traditional technology similarly to the methods like shutting , requires separate treatment of the BVPs and the algebraic continuity equations and brings itself to additional linear ODEs for elements of the corresponding to (17) Jacobi matrix. These elements are functions of the variable $`x`$ and they have to be known actually only at the point $`x=1`$. The solving of the both nonlinear BVPs (12), (15) and (13), (16) along with the attached linear equations is another hard enough task.
At the present work using the CANM , we propose a common treatment of both, differential and algebraic problems.
We suppose that the nonlinear spectral problem (12), (15), (13), (16) and (17) has a “well separated” exact solution. Let the functions $`𝐲_{i,0}(x),𝐲_{e,0}(x)`$ and the parameters $`R_{s,0},\mathrm{\Omega }_0,\phi _{s,0}`$ are initial approximations to this solution. The CANM leads to the following iteration process:
$`𝐲_{i,k+1}(x)`$ $`=𝐲_{i,k}(x)+\tau _k𝐳_{i,k}(x),`$ (18)
$`𝐲_{e,k+1}(x)`$ $`=𝐲_{e,k}(x)+\tau _k𝐳_{e,k}(x),`$ (19)
$`R_{s,k+1}`$ $`=R_{s,k}+\tau _k\rho _k,`$ (20)
$`\mathrm{\Omega }_{k+1}`$ $`=\mathrm{\Omega }_k+\tau _k\omega _k,`$ (21)
$`\phi _{s,k+1}`$ $`=\phi _{s,k}+\tau _k\varphi _k.`$ (22)
Here $`\tau _k(0,1]`$ is a parameter which can rule the convergence of iteration process. The increments $`𝐳_{i,k}(x),`$ $`𝐳_{e,k}(x),\rho _k,\omega _k`$ and $`\varphi _k`$, $`k=0,1,2,\mathrm{}`$ satisfy the linear ODEs (for sake of simplicity henceforth we will omit the number of iterations $`k`$):
$`𝐳_i^{}+R_s{\displaystyle \frac{𝐅_i}{𝐲_i}}𝐳_i+\left(R_s{\displaystyle \frac{𝐅_i}{R_s}}+𝐅_i\right)\rho +R_s{\displaystyle \frac{𝐅_i}{\mathrm{\Omega }}}\omega `$ $`=𝐲_i^{}R_s𝐅_i,`$ (23)
$`𝐳_e^{}+R_s{\displaystyle \frac{𝐅_e}{𝐲_e}}𝐳_e+\left(R_s{\displaystyle \frac{𝐅_e}{R_s}}+𝐅_e\right)\rho +R_s{\displaystyle \frac{𝐅_e}{\mathrm{\Omega }}}\omega `$ $`=𝐲_e^{}R_s𝐅_e.`$ (24)
All the coefficients and right hand sides as well in the above two equations are known functions of the arguments $`x`$, $`R_s`$, $`\mathrm{\Omega }`$ by means of the solution from the previous iteration. We seek for the unknowns $`𝐳_i(x)`$ of equation (23) and $`𝐳_e(x)`$ of equation (24) as a linear combinations with coefficients $`\rho ,\omega `$ and $`\varphi `$:
$`𝐳_i(x)`$ $`=`$ $`𝐬_i(x)+\rho 𝐮_i(x)+\omega 𝐯_i(x)+\varphi 𝐰_i(x),`$ (25)
$`𝐳_e(x)`$ $`=`$ $`𝐬_e(x)+\rho 𝐮_e(x)+\omega 𝐯_e(x)+\varphi 𝐰_e(x).`$ (26)
Here $`𝐬_i(x),𝐮_i(x)`$, $`𝐯_i(x)`$, $`𝐰_i(x)`$, $`𝐬_i(x)`$ and $`𝐬_e(x),𝐮_e(x)`$, $`𝐯_e(x)`$, $`𝐰_e(x)`$ are new unknown functions, which are defined in either, internal or external domains. Substituting for the decomposition (25) into equation (23) after reduction we obtain:
$`𝐬_i^{}+Q_i(x)𝐬_i`$ $`=𝐲_i^{}R_s𝐅_i,`$
$`𝐮_i^{}+Q_i(x)𝐮_i`$ $`=\left(𝐅_i+R_s{\displaystyle \frac{𝐅_i}{R_s}}\right),`$
$`𝐯_i^{}+Q_i(x)𝐯_i`$ $`=R_s{\displaystyle \frac{𝐅_i}{\mathrm{\Omega }}},`$
$`𝐰_i^{}+Q_i(x)𝐰_i`$ $`=0`$
where $`Q_i\left(x\right)R_s\frac{𝐅_i(R_sx,𝐲_i,\mathrm{\Omega })}{𝐲_i}`$ stands for a square matrix ($`7\times 7)`$, which consists of the Frechet derivatives of operator $`𝐅_i`$ at the point $`\{𝐲_i(x),R_s,\mathrm{\Omega }\}`$.
Similarly applying the CANM to the boundary conditions (15) and taking into account the dependence of matrix $`D_{1,i}`$ on the parameter $`\phi _s`$ yields:
$$B_{0,i}𝐳_i(0)=D_{0,i}B_{0,i}𝐲_i(0),B_{1,i}𝐳_i(1)=D_{1,i}B_{1,i}𝐲_i(1)D_{1,i}^{}\varphi .$$
By means of the decomposition (25) we obtain the following eight boundary conditions (four left + four right) for the equations (LABEL:lini):
$`B_{0,i}𝐬_i(0)=D_{0,i}B_{0,i}𝐲_i(0),B_{1,i}𝐬_i(1)=D_{1,i}B_{1,i}𝐲_i(1),`$
$`B_{0,i}𝐮_i(0)=0,B_{1,i}𝐮_i(1)=0,`$
(28)
$`B_{0,i}𝐯_i(0)=0,B_{1,i}𝐯_i(1)=0,`$
$`B_{0,i}𝐰_i(0)=0,B_{1,i}𝐰_i(1)=D_{1,i}^{}(\phi _s).`$
Let us now substitute for decomposition (26) into the linear equations for external domain (24). As result we obtain the following four vector equations with regard to the unknown functions $`𝐬_e(x),𝐮_e(x)`$, $`𝐯_e(x)`$ and $`𝐰_e(x)`$ with eight boundary conditions (four left + four right):
$`𝐬_e^{}+Q_e(x)𝐬_e`$ $`=𝐲_e^{}R_s𝐅_e,`$
$`𝐮_e^{}+Q_e(x)𝐮_e`$ $`=\left(𝐅_e+R_s{\displaystyle \frac{𝐅_e}{R_s}}\right),`$
$`𝐯_e^{}+Q_e(x)𝐯_e`$ $`=R_s{\displaystyle \frac{𝐅_e}{\mathrm{\Omega }}},`$
$`𝐰_e^{}+Q_e(x)𝐰_e`$ $`=0.`$
Here $`Q_e\left(x\right)R_s\frac{𝐅_e[R_sx,𝐲_e(x),\mathrm{\Omega }]}{𝐲_e}`$ is a square matrix $`(6\times 6)`$ whose elements are Frechet’s derivatives of the operator $`𝐅_e`$ at the point $`\{𝐲_e(x),R_s,\mathrm{\Omega }\}`$.
The corresponding linear BC are obtained in the same way as (28) and they become:
$`B_{1,e}𝐬_e(1)`$ $`=D_{1,e}B_{1,e}𝐲_e(1),B_{\mathrm{},e}𝐬_e(\mathrm{})=B_{\mathrm{},e}𝐲_e(\mathrm{}),`$
$`B_{1,e}𝐮_e(1)`$ $`=0,B_{\mathrm{},e}𝐮_e(\mathrm{})=0,`$
$`B_{1,e}𝐯_e(1)`$ $`=0,B_{\mathrm{},e}𝐯_e(\mathrm{})=0,`$
$`B_{1,e}𝐰_e(1)`$ $`=D_{1,e}^{}(\phi _s),B_{\mathrm{},e}𝐰_e(\mathrm{})=0.`$
In the end to compute the increments $`\{\rho ,\omega ,\varphi \}`$ of parameters $`R_s,\mathrm{\Omega }`$ and $`\phi _s`$ we use the three conditions (17).
Let the solutions of linear BVP (LABEL:lini), (28) and (LABEL:line), (LABEL:lbce) at the $`k`$th iteration stage are assumed to be known. For sake of the simplicity we introduce the vector $`\stackrel{~}{𝐲}(x)(\nu (x),\xi (x),\eta (x))^T`$. For two arbitrary functions $`h_i(x)`$ and $`h_e(x)`$, defined in left and right vicinity of the point $`x=1`$, we set $`\mathrm{\Delta }hh_e(1)h_i(1)`$. Then applying the CANM to the equations (17) and having in mind the decompositions (25),(26), we attain the vector equation
$$\mathrm{\Delta }\stackrel{~}{𝐮}\rho +\mathrm{\Delta }\stackrel{~}{𝐯}\omega +\mathrm{\Delta }\stackrel{~}{𝐰}\varphi =\left(\mathrm{\Delta }\stackrel{~}{𝐲}+\mathrm{\Delta }\stackrel{~}{𝐬}\right),$$
(31)
which represents an algebraic system consisting of three linear scalar equations with respect to the three unknowns $`\rho `$, $`\omega `$ and $`\varphi `$.
The general sequence of the algorithm can be recapitulated in the following way. Let us assume that the functions $`𝐲_{i,k}(x)`$, $`𝐲_{e,k}(x)`$, and parameters $`R_{s,k}`$, $`\mathrm{\Omega }_k`$, $`\phi _{s,k}`$ are given for $`k0`$. We solve the linear BVPs (LABEL:lini), (28) and thus we compute the functions $`𝐬_{i,k}(x),𝐮_{i,k}(x)`$, $`𝐯_{i,k}(x)`$, $`𝐰_{i,k}(x)`$ in the inner domain $`x[0,1]`$. Then we solve the linear BVPs (LABEL:line), (LABEL:lbce) in the external domain $`x[1,\mathrm{}]`$ and compute the functions $`𝐬_{e,k}(x),𝐮_{e,k}(x)`$, $`𝐯_{e,k}(x)`$ and $`𝐰_{e,k}(x)`$. Next, to obtain the increments $`\rho _k`$, $`\omega _k`$ and $`\varphi _k`$ we solve the linear algebraic system (31). So using the decompositions (25), (26) and then the formulae (18) - (22) we calculate the functions $`𝐲_{i,k+1}(x)`$, $`𝐲_{e,k+1}(x)`$, the radius of the star $`R_{s,k+1}`$, the quantity $`\mathrm{\Omega }_{k+1}`$ and the parameter boundary condition $`\phi _{s,k+1}`$ as well at the new iteration stage $`k+1`$.
At the every iteration $`k`$ an optimal time step $`\tau _{opt}`$ is determinated in accordance with the Ermakov&Kalitkin formula
$$\tau _{opt}\frac{\delta (0)}{\delta (0)+\delta (1)}$$
where the residual $`\delta (\tau )`$ is calculated as follows
$$\delta (\tau _k)=\mathrm{max}[\delta _f,(R_{s,k}+\tau _k\rho _k)^2,(\mathrm{\Omega }_k+\tau _k\omega _k)^2,(\phi _{s,k}+\tau _k\varphi _k)^2]$$
and $`\delta _f`$ is the Euclidean residual of right hand side of the first equations in the systems (LABEL:lini), (28) and (LABEL:line), (LABEL:lbce).
The criterion for termination of the iterations is $`\delta (\tau _{opt})<\epsilon `$ where $`\epsilon 10^8÷10^{12}`$ for some $`k`$.
Taking into account the smoothness of sought solutions we solve the linear BVPs (LABEL:lini), (28) and (LABEL:line), (LABEL:lbce) employing Hermitean splines and spline collocation scheme of fourth order of approximation . At that we utilize essentially the important feature that everyone of the above mentioned two groups vector BVPs (inner and external) have one and the same left hand sides.
It is worth to note that the algebraic systems of linear equations and the system (31) as well become ill-posed in the vicinity of the “exact” solution, i.e., for sufficiently small residuals $`\delta `$. That is why for small $`\delta `$, for example if $`\delta <10^3`$ (then $`\tau _{opt}1`$ usually), it is expedient to use the Newton-Kantorovich method when the respective matrices are fixed for some $`\delta 10^3`$.
## 4 Some numerical results
For a purpose of illustrating we will consider and discuss some results obtained from numerical experiments. A detailed description and analysing of results from physical point of view will be object of another our paper.
In present article we consider concrete scalar-tensor model with functions (see Section 2)
$`A(\phi )=\mathrm{exp}({\displaystyle \frac{\phi }{\sqrt{3}}}),V(\phi )={\displaystyle \frac{3}{2}}(1A^2(\phi ))^2,`$
$`f(\mu )`$ $`={\displaystyle \frac{1}{8}}\left[(2\mu 3)\sqrt{\mu +\mu ^2}+3\mathrm{ln}\left(\sqrt{\mu }+\sqrt{1+\mu }\right)\right],`$
$`g(\mu )`$ $`={\displaystyle \frac{1}{8}}\left[(6\mu +3)\sqrt{\mu +\mu ^2}3\mathrm{ln}\left(\sqrt{\mu }+\sqrt{1+\mu }\right)\right],`$
$$W(\sigma ^2)=\frac{1}{2}\left(\sigma ^2+\frac{1}{2}\mathrm{\Lambda }\sigma ^4\right).$$
The quantities $`b,\mathrm{\Lambda }`$ are given parameters. For completeness we note that in the concrete case the functions $`f(\mu )`$ and $`g(\mu )`$ represent in parametric form the equation of state of noninteracting neutron gas while the function $`W(\sigma ^2)`$ describes the boson field with quartic self-interaction.
The calculated functions $`\sigma (x)`$, $`\phi (x)`$, $`\nu (x)`$ and $`\mu (x)`$ are plotted correspondingly in Fig. 1, 2, 3 and 4 for the values of the parameters $`\gamma =0.1`$, $`\mathrm{\Lambda }=10`$ and $`b=1.`$ The behaviour of the mentioned functions is typical for wider range of the parameters not only for those values presented in the figures. The function $`\sigma (x)`$ decreases rapidly from its central value $`\sigma _c=0.4`$ (in the case under consideration) to zero, at that when dimensionless coordinate $`x>6`$ the function does not exceed $`10^4`$. Similarly the function $`\nu (x)`$ has most large derivative for $`x(0,9)`$ after that it approaches slowly zero at infinity like $`\frac{1}{x}`$. For example when $`x9`$ the derivative $`\nu ^{}(x)10^2`$, while for $`x>27`$ $`\nu ^{}(x)<10^4`$, i.e., the asymptotical behaviour of calculated grid function and its derivative agrees very well with the theoretical prediction (see ). The function $`\phi (x)`$ increases rapidly for $`x<4`$, besides that it trends asymptotically to zero. Obviously the quantitative behaviour of $`\phi (x)`$ for central value $`\sigma _c=0.4`$ is determinated by the dominance of the term $`\stackrel{B}{T}`$ over the term $`\stackrel{F}{T}`$ (see ). At last the function $`\mu (x)`$ is nontrivial in the inner domain $`x[0,1]`$, i.e. inside the star. Here it varies monotonously and continuously from its central value (in the case under consideration) $`\mu _c=1.2`$ till zero at $`x=1`$ corresponding to the radius of the star.
From physical point of view it is important to know the mass of the boson-fermion star and the total number of particles (bosons and fermions) making up the star.
The dimensionless star mass can be calculated via the formula
$$M=_0^{\mathrm{}}r^2(\stackrel{B}{T_0^0}+\stackrel{F}{T_0^0}+\mathrm{exp}(\lambda )\left(\frac{d\phi }{dr}\right)^2+\frac{\gamma ^2}{2}V(\phi ))dr.$$
The dimensionless rest mass of the bosons (total number of bosons times the boson mass) is given by
$$M_{RB}=\mathrm{\Omega }_0^{\mathrm{}}r^2A^2(\phi )\mathrm{exp}\left(\frac{\lambda \nu }{2}\right)\sigma ^2𝑑r.$$
The dimensionless rest mass of the fermions is correspondingly
$$M_{RF}=b_0^{\mathrm{}}r^2A^3(\phi )\mathrm{exp}\left(\frac{\lambda }{2}\right)n(\mu )𝑑r$$
where $`n(\mu )`$ is the density of the fermions. In the case we consider we have $`n(\mu )=\mu ^{\frac{3}{2}}(x)`$.
The dependencies of the star mass $`M`$ (solid line) and the rest mass of fermions $`M_{RF}`$ (dash line) on the central value $`\mu _c`$ of the function $`\mu (x)`$ are shown in configuration diagram on Fig. 5 for $`\lambda =0`$, $`\gamma =0.1`$, $`b=1`$ and $`\sigma _c=0.002`$. It should be pointed that for so small central value $`\sigma _c`$ we have in practice pure fermionic star. On the figure it is seen that from small values of $`\mu _c`$ to values near beyond the peak the rest mass is greater than the total mass of the star which means that the star is potentially stable.
On Fig. 6 the binding energy of the star $`E_b=MM_{RB}M_{RF}`$ is drawn against the rest mass of fermions $`M_{RF}`$ for $`\lambda =0`$, $`\gamma =0.1`$, $`b=1`$ and $`\sigma _c=0.002`$. Fig. 6 is actually a bifurcation diagram. With increasing the central value of function $`\mu (x)`$ one meets a cusp. The appearance of a cusp shows that the stability of the star changes - one perturbation mode develops instability. Beyond the cusp the star is unstable and may collapse eventually forming black hole. The corresponding physical results for pure boson stars are considered in our recent paper .
## Acknowledgments
We are grateful to Prof. I.V. Puzynin (JINR, Dubna, Russia) for helpful discussion.
|
warning/0004/hep-th0004103.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
Recently, there has been renewed interest in the idea that our world may be a domain wall in a higher dimensional spacetime . This scenario, often known as the “Brane World” scenario , is motivated by the existence of solitonic objects in string theory, such as D-branes, in which gauge and matter fields are localized on lower-dimensional subspaces. Indeed, if the Standard Model fields are localized on a brane, whereas gravity propagates in a higher dimensional spacetime, a number of phenomenological issues have to be revisited. A rather interesting 5-dimensional example was suggested in where gravity seems to be localized on a 4-dimensional “brane”, even though the 5-th dimension has an infinite extent. A key ingredient is their setup is that the ambient spacetime is warped; the localization of gravity is due to an exponentially damped warp factor. Several attempts have been made to find domain wall solutions in 5D gauged supergravity that may perhaps localize gravity . A warped metric can also be used to generate the hierarchy between the electroweak and the four-dimensional Planck scale . The possible embedding of such a scenario in string theory was discussed in .
In fact, warped compactifications are a rather logical generalization of compactifications of the product type $`_4\times 𝒦`$, and were considered some time ago in the context of supergravity and heterotic string theory . In perturbative string theory or supergravity there are no branes and in these scenarios there are therefore no gauge and matter fields which are a priori localized in some lower dimensional subspace. In strongly coupled heterotic string theory, on the other hand, there are two “end of the universe”- branes at the ends of the non-perturbative 11-th direction on which gauge and matter fields are localized . In general, the four-form field strength $`G`$ is non-zero, and as a result, spacetime is warped. However, the long wavelength expansion in eleventh dimensional supergravity is valid provided that $`R_{11}R_{CY}/\kappa ^{2/3}`$ where $`\kappa `$ is the eleventh dimensional gravitational coupling, and $`R_{CY}`$ is the size of the Calabi-Yau on which the heterotic string theory is compactified. One cannot make the extra dimension transverse to the branes too large, or else the effective field theory breaks down.
F theory provides another powerful tool in studying non-perturbative string vacua (see,e.g., , for four-dimensional F theory compactifications). In compactifying M and F theory to lower dimensions, anomaly cancellation (cancellation of tadpoles) often requires one to introduce branes (or alternatively, to turn on some background flux ). All $`p`$-brane supergravity solutions are examples of warped metrics due to the backreaction of the branes. The recent developments give us motivation to better understand these warped compactifications in a variety of circumstances. And while the consistency requirements for M theory compactifications on Calabi-Yau four-folds (and their lifting to F theory) have been studied in some detail more recently , the form of the background geometry has not been determined explicitly in the presence of non-zero background flux (see, however ). A purpose of the present paper is to take some modest steps in this direction.
In the first part of this paper, we will determine the warp factor for some M/F theory compactifications with background flux and analyze its behavior. To solve the equation for the warp factor explicitly, we consider orbifold limits of Calabi-Yau spaces. In the F theory limit, the dilaton will also be taken to be constant over the base (the $`7`$-brane charges are locally cancelled). It is, however, likely that the generic features we find here will also hold for general Calabi-Yau manifolds. We find that the background flux contributes to the warp factor in the same manner and with the same scale as $`p`$-branes (both are $`\alpha ^{}`$ suppressed as they are due to backreaction). The resulting metrics are thus generalizations of the $`p`$-brane ones with a contribution to the harmonic function from the background flux. In particular the warp factor always contains a constant part. This is to some extent expected as the contribution of the background flux to the energy momentum tensor scales inversely with the compactification volume. In the large volume limit, the contribution is negligible, and we should recover the regular $`p`$-brane metric. By way of comparison, the contribution of a cosmological constant (as in ) does not scale with the compactification volume. In the proposal of , however, one restricts attention to the near-horizon region where the constant part of the warp factor is unimportant and the effective geometry changes. The compactification data as well as the undecoupled gravitional interactions are encoded in a hypothetical Planck brane at the edge of the AdS throat.
In the second part of this paper we will address some aspects of the computation of (the closed string part of) the low-energy spectrum for compactifications with background flux. Generic string compactifications suffer from an abundance of massless fields. The presence of background flux, however, will lift a number of these . A priori, the number of massless fields are determined by the moduli of the Calabi-Yau four-fold, subject to a set of topological consistency conditions on the non-zero background flux. On the other hand, a Kaluza-Klein reduction of an arbitrary warped metric generates extra mass terms due to the warp factors. Generically the massless spectrum is no longer determined purely from the dimensions of spaces of harmonic forms on the internal space. However, the powers of the warp factors that appear in this type of warped compactifications, which are determined by supersymmetry, balance each other out, so one may indeed use topological information to determine the massless spectrum. Moreover the non-trivial parts of the warp factor are generated solely by the gravitational backreaction, and in the weak coupling limit space-time reduces back to a product form. Using this information we recover explicitly the superpotential for the complex structure moduli conjectured in .
The organization of the paper is as follows. In Section 2, we review the results of M and F theory compactifications on Calabi-Yau 4-folds in the presence of background fluxes. In Section 3, we determine the warp factor for orbifold examples, in which the anomaly is cancelled by a combination of background flux (from the untwisted and twisted sectors) and $`2/3`$-branes. In Section 4, we comment on the shape of the graviton wavefunction in this type of warped compactification. For completeness, we include in the appendix a parallel discussion on the graviton wavefunction for Heterotic-M theory on a Calabi-Yau 3-fold. In Section 5, we present a general discussion on determining the number of complex and Kahler moduli in vacua with background flux. For illustative purpose, we consider in Section 6 compactifications of M theory with background flux which satisfy the field equations for an orbifold limit of $`K3\times K3`$. The conditions on the background flux are greatly simplified if we choose it to be a product of $`(1,1)`$ forms of each $`K3`$. For this example we will determine the spectrum using the explicit topological data and we will discuss how to solve more general cases. We end with some comments in Section 7.
## 2 M and F theory Vacua with Background Fluxes
Warped compactifications of perturbative string theories, or rather supergravities, preserving minimal supersymmetry in four dimensions were considered in . The supersymmetry requirements are highly restrictive; it was found that for Type II theories there are no non-trivial solutions if the four non-compact dimensions are Minkowski . For type I/heterotic theory there is a solution if the internal space has nonzero torsion: i.e. the vacuum expectation value of the three-form NS-NS field strength $`H^{NS}`$ is non-zero. In this case the warp factor equals the dilaton, whose profile is determined by the vacuum expectation value $`H^{NS}`$.
It has since been discovered that the M theory effective action has at the first subleading order in the derivative expansion a topological term
$$CX_8(R),$$
(1)
where
$$X_8=\frac{1}{84!}\left(\text{tr}R^4\frac{1}{4}(\text{tr}R^2)^2\right).$$
(2)
The inclusion of such higher derivative terms in the low energy supergravities allows for new solutions preserving minimal supersymmetry, i.e. four supercharges, which are of the warped kind . (see also ). Let us give a brief review.
On compactifications down to three dimensions, the eight-form $`X_8(R)`$ can take on a background expectation value. This will act as a source term for the three-form field $`C`$. To maintain a solution to the field equations one needs to introduce M2-brane sources or non-trivial $`G`$-flux; $`G`$ is the four-form which locally equals $`dC`$. Either the M2-branes or the non-trivial $`G`$-flux or both induce in turn a warping of the metric
$$ds^2=e^{\varphi (y)}\eta _{\mu \nu }dx^\mu dx^\nu +e^{\frac{1}{2}\varphi (y)}g_{a\overline{b}}dy^ad\overline{y}^{\overline{b}}.$$
(3)
The requirement that minimal supersymmetry (four supercharges, $`𝒩=2`$ in $`d=3`$) is preserved determines the relative weights of the warp-factors, as is known from $`p`$-brane solutions . In addition, supersymmetry demands that $`g_{a\overline{b}}`$ is the metric of a Calabi-Yau four-fold. The four-form $`G`$ must furthermore obey
$$G_{abcd}=0=G_{abc\overline{d}},g^{c\overline{d}}G_{a\overline{b}c\overline{d}}=0,G_{\mu \nu \rho a}=ϵ_{\mu \nu \rho }_ae^{\frac{3}{2}\varphi }.$$
(4)
The first two conditions mean that $`G`$ is a (2,2) form on $`CY_4`$ which is self-dual or, equivalently, primitive with respect to the Kahler form $`J`$ of the $`CY_4`$
$$G=GJG=0.$$
(5)
Dirac quantization requires $`G`$ to be an element of integer cohomology. The last equation of (4) says that the only nonvanishing part of the three-form $`C`$ is
$$C_{\mu \nu \rho }=ϵ_{\mu \nu \rho }e^{\frac{3}{2}\varphi }.$$
(6)
By definition the antisymmetric tensor is taken with respect to the unwarped metric.<sup>1</sup><sup>1</sup>1The three-form thus equals the induced volume on the three-dimensional space $`C_{\mu \nu \rho }`$ $`=`$ $`\sqrt{det|g_{MN}_\mu z^M_\nu z^N|}ϵ_{\mu \nu \rho }^{LeviCivita}`$ as is again familiar from $`p`$-branes. For the Dirac-delta function in curved space, we use the convention that $`𝑑x\sqrt{g(x)}\delta (xx_i)=1`$.
The field equation for $`C`$ determines the warp factor
$$\mathrm{}_{CY_4}(e^{3\varphi /2})=_{CY_4}(X_8\frac{1}{2}GG)\underset{j=1}{\overset{n}{}}\delta ^8(yy_j).$$
(7)
The factor $`GG`$ has its origins in the well known $`CGG`$ supergravity interaction and we have introduced $`n`$ M2-branes at arbitrary points. This number $`n`$ is determined by the consistency requirement - the absence of tadpoles or anomalies - that the integral over the right hand side of (7) vanish,
$$_{CY_4}X_8=\frac{\chi _{CY_4}}{24}=n+\frac{1}{2}_{CY_4}GG.$$
(8)
#### Lift to F theory
Such compactifications can be lifted to F theory if the $`CY_4`$ is elliptically fibered. Four-dimensional Lorentz-invariance requires that the four-form $`G`$ has one leg on the toroidal fiber and three on the base $``$ . We will limit our attention to the case where the $`G`$-flux is not localized in the fiber of the elliptic fibration, but is constant. In that case the M theory solution can straightforwardly be lifted to F theory with
$`ds^2`$ $`=`$ $`e^{\frac{3}{4}\varphi }dx^2+e^{\frac{3}{4}\varphi }g_{a\overline{b}}^{}dy^ad\overline{y}^{\overline{b}},`$
$`H^{NS}`$ $`=`$ $`\omega _{}\omega ,`$
$`H^{RR}`$ $`=`$ $`\omega \tau _{}\omega \overline{\tau },`$
$`D_{\lambda \mu \nu \rho }^+`$ $`=`$ $`ϵ_{\lambda \mu \nu \rho }e^{\frac{3}{2}\varphi },`$ (9)
where $`\omega ϵH^{(1,2)}()`$ with $`(\tau \overline{\tau })_{}\omega \omega =_{CY_4}GGϵ^+`$. The metric and four-form potential are again that of the D3-brane solution. Formally the warp factor is still given by (7), but one has to be careful with the dependence on the internal directions. We will solve for the warp factor for a four-fold of the form $`K3\times T^4/_2`$ where the fiber belongs to $`T^4/_2`$. In this case, the warp factor can equivalently be determined from the IIB field equation for $`D^+`$,
$$ddD^+=\frac{1}{16}\underset{i=1}{\overset{4}{}}\text{tr}(RR)\delta ^2(z^1z_i^1)H^{NS}H^{RR}_{}\underset{j=1}{\overset{n}{}}\delta ^2(z^1z_j^1)\delta ^4(\omega \omega _j).$$
(10)
Here, $`i`$ labels the fixed points of $`T^2/_2`$ and $`j`$ labels the $`n`$ D3-branes. Tadpole cancellation requires that this number equals
$$_{}H^{NS}H^{RR}+n=\frac{\chi _{K3\times K3}}{24}=24.$$
(11)
Substituting the background expectation value for $`D^+`$ we find the F theory analogue of (7)
$$\mathrm{}_{}(e^{3\varphi /2})=_{}(\frac{1}{16}\underset{i=1}{\overset{4}{}}\text{tr}(RR)\delta ^2(z^1z_i^1)H^{NS}H^{RR})\underset{j=1}{\overset{n}{}}\delta ^2(z^1z_j^1)\delta ^4(\omega \omega _j),$$
(12)
which determines the shape of the warp factor. In the orientifold limit, the first term of the expression contains the contribution of the orientifold planes .
We will solve the warp-factor equations (7) and (12) for an orbifold limit of the Calabi-Yau. Specifically we will concentrate on the case $`K3\times K3`$ $`=T^4/_2\times T^4/_2`$. This allows us to take the F theory limit in the end, for which we will then attempt to compute the massless four-dimensional spectrum.
Let us note from the outset that the warp factor, as a solution to (7),(12) is determined only up to an integration constant whose value is fixed by the boundary conditions of the warp factor. If the fluxes are sufficiently localized, the metric (3) should be approximately flat away from such special points and the constant can be fixed to $`1`$. We will discuss the effect of a constant flux in the next section.
#### The Dilaton
In the previously known warped compactifications of (heterotic) string theory, the warp factor turned out to be equivalent to the dilaton . Let us briefly recall why this is not the case for the Type IIB dilaton when we lift the above M theory vacua to F theory . Weyl rescaling between the Einstein and the string frame can therefore alter the form of the warp factors if a non-constant dilaton profile is consistent with the field equations.
By compactifying M theory in the eleventh and ninth-direction to IIA on $`S_1`$ and T-dualizing on the latter one finds that the IIB dilaton is given by a ratio of the compactification radii
$$\mathrm{exp}(\phi _{IIB})=\frac{R_{11}}{R_9}.$$
(13)
Since in the background solution for an ellipitically fibered $`CY_4`$,
$$ds^s=e^\varphi dx^2+e^{\varphi /2}(g_{a\overline{b}}dy^ad\overline{y}^{\overline{b}}+R_9^2(y)dw_1^2+R_{11}^2(y)dw_2^2),$$
(14)
the warp factor is common to both $`R_{11}`$ and $`R_9`$, the IIB dilaton is independent hereof. In general, though, the dilaton will be a function of the internal dimensions $`y_a`$. In F theory, the $`7`$-branes are sources of the complex field $`\tau =a+ie^\phi `$ where $`a`$ is the RR axion. For example, near the location of a $`7`$-brane at $`z=z_i`$ ($`z`$ is the complex coordinate transverse to the $`7`$-branes),
$$\tau \frac{1}{2\pi i}\text{log}(zz_i).$$
(15)
In the Einstein frame the metric is then warped to
$$g_{z\overline{z}}=\tau _2\eta ^2\overline{\eta }^2\underset{i}{}(zz_i)^{1/12}\underset{i}{}(\overline{z}\overline{z}_i)^{1/12}.$$
(16)
In F theory, the D7-branes are not mutually local, but in the orientifold limit, which we will consider, the charges of the D-branes are locally cancelled against the orientifold planes (which for non-zero $`g_s`$, are bound states of some $`(p,q)`$ $`7`$-branes). In that case the complex field $`\tau `$ becomes a constant, and the metric (16) reduces to a flat one (except for conical singularities at the locations of the orientifold planes). In the general case where the $`7`$-brane charges are not locally cancelled, the calculation of the effective four dimensional Planck scale would involve an integral over a rather non-trivial function of the internal manifold (due to (15), (16)). Another complication is that the metric for the 3-7 brane system where the 7-brane charges are not locally cancelled has not yet been solved; for progress in this direction see .
## 3 The Warp Factor
#### M theory
The warp factor is to be determined from the equation of motion (7) for the gauge field $`C_{\mu \nu \rho }`$. To find its explicit form, we have to invert the Laplacian on the compact internal space. This can be done with the help of Green’s functions. On a compact space, or equivalently when the Laplacian has non-trivial zero modes, the inversion is only defined on the subset of functions orthogonal to these zero modes, i.e. on those functions not belonging to the kernel of the Laplacian (see appendix for a brief review). For our purposes, it is relevant to know that on a compact space the scalar Green’s function obeys,
$$\mathrm{}G(x,x_i)=\delta ^d(xx_i)\frac{1}{\text{Vol}}$$
(17)
where $`\delta ^d(xx_i)`$ is the $`d`$-dimensional Dirac delta-function, and “Vol” is the volume of the compact space.
On an orbifold, internal fluxes fall into two categories: those localized at the fixed points and those wrapped over periods of the underlying torus. The equation (7) determining the warp factor thus has two different kinds of contributions: the part of the flux term $`GG`$ which consists of a constant flux corresponding to “untwisted” cycles and that part which is built from “twisted” cycles, proportional to Dirac-delta-functions at the fixed points. The constant fluxes are naturally expressed as an integer divided by the volume (after the Hodge dual has been taken). In our case, the orbifold limit of $`K3\times K3`$, we take the G-flux to be constant on one of the K3’s, so that we may take the F theory limit later. Then the flux term on the r.h.s. of (7) can be written as
$$(GG)=\frac{r_1r_2}{V_1V_2}+\frac{r_1}{V_1}\underset{x_p}{}m_p\delta ^4(xx_p),$$
(18)
where $`r_i`$, $`V_i`$ is respectively the total untwisted flux and volume on each $`K3`$ and $`m_p`$ is the total twisted flux located at each fixed point $`x_p`$ of the second $`K3`$.
In the orbifold limit the curvature is also localized at the fixed points. On $`K3\times K3`$ the total contribution to the warp factor from the curvature is
$$X_8=\frac{\chi _{K3\times K3}}{24}=24.$$
(19)
The $`(T^4/_2)^2`$ limit has $`16^2`$ fixed points. Each contributes of course equally to the curvature and the first term in (7) may thus be written as
$$X_8=\frac{24}{16^2}\underset{x_p,z_p}{}\delta ^4(xx_p)\delta ^4(zz_p),$$
(20)
where $`z_p`$ are the fixed points of the first $`K3`$.
Combining this information the equation (7) determining the warp factor reduces to
$`\mathrm{}(e^{3\varphi /2})`$ $`=`$ $`{\displaystyle \frac{3}{32}}{\displaystyle \underset{x_p,z_p}{}}\delta ^4(xx_p)\delta ^4(zz_p){\displaystyle \frac{r_1r_2}{2V_1V_2}}{\displaystyle \frac{r_1}{2V_1}}{\displaystyle \underset{x_p}{}}m_p\delta ^4(xx_p)`$ (21)
$`{\displaystyle \underset{i=1}{\overset{n}{}}}\delta ^4(xx_i)\delta ^4(zz_i)`$
$`=`$ $`{\displaystyle \frac{3}{32}}{\displaystyle \underset{x_p,z_p}{}}\left(\delta ^4(xx_p)\delta ^4(zz_p){\displaystyle \frac{1}{V_1V_2}}\right){\displaystyle \underset{i=1}{\overset{n}{}}}\left(\delta ^4(xx_i)\delta ^4(zz_i){\displaystyle \frac{1}{V_1V_2}}\right)`$
$`{\displaystyle \frac{r_1}{2V_1}}{\displaystyle \underset{x_p}{}}m_p\left(\delta ^4(xx_p){\displaystyle \frac{1}{V_2}}\right).`$
Here we have made use of the fact that tadpole cancellation implies that $`48=r_1r_2+r_1m_p+2n`$.
The warp factor is now easily expressed in terms of Green’s functions
$`e^{3\varphi /2}=c_0+{\displaystyle \frac{3}{32}}{\displaystyle \underset{x_p,z_p}{}}G^{(8)}(x,(x_p,z_p)){\displaystyle \underset{x_i,z_i}{}}G^{(8)}(x,(x_i,z_i)){\displaystyle \frac{r_1}{2V_1}}{\displaystyle \underset{x_p}{}}m_pG^{(4)}(x,x_p).`$ (22)
Note that the completely constant flux term, proportional to $`r_2`$, does not contribute to the warp factor in any obvious way. The Green’s functions $`G^{(8)}`$ on $`K3\times K3`$ and $`G^{(4)}`$ on a single $`K3`$ can be constructed in the orbifold limit by the method of images. For example, $`G^{(4)}`$ is given by
$$G^{(4)}(x,x_i)=\underset{\stackrel{}{p}}{}\frac{e^{i\stackrel{}{p}(\stackrel{}{x}\stackrel{}{x_i})}+e^{i\stackrel{}{p}(\stackrel{}{x}+\stackrel{}{x_i})}}{\stackrel{}{p}^2}.$$
(23)
The momenta $`\stackrel{}{p}`$ are quantized in units of the inverse radii of the $`T^4`$. The second term is due to the $`_2`$ image. In the above sum, the zero momentum mode ($`p_1=p_2=p_3=p_4=0`$) is excluded.
The integration constant $`c_0`$ is determined by the boundary conditions. We already explained why, if all fluxes are localized ($`r_2=0`$ in our example above) this constant is fixed to unity. In that case far away from the points where flux (energy density) is localized, spacetime should be approximately flat. In the case of constant G-flux over the whole internal manifold, we can determine $`c_0`$ from the curvature in the internal directions by looking at the Einstein equation,
$$R_{MN}=\frac{1}{2}\left(G_MG_N\frac{3}{D2}g_{MN}GG\right),$$
(24)
where
$$GG=\frac{G_{MNRS}G^{MNRS}}{4!};G_MG_N=\frac{G_{MABC}G_N^{ABC}}{3!}.$$
(25)
We have ignored contributions to the stress-tensor from the curvature and the M2-branes as they are localized. Substituting the background expectation values (and noting that $`G`$ is self-dual in the internal dimensions)
$$C_{\mu \nu \rho }=ϵ_{\mu \nu \rho }e^{3\varphi /2},G_{a\overline{b}c\overline{d}}\stackrel{~}{G}_{a\overline{b}c\overline{d}}0\stackrel{~}{G}\stackrel{~}{G}=r/V_{CY},$$
(26)
one finds for the Ricci curvature in the internal directions<sup>2</sup><sup>2</sup>2To find this answer one needs to use that the field equations require that $`\stackrel{~}{G}_m\stackrel{~}{G}_n={\displaystyle \frac{g_{mn}}{2}}\stackrel{~}{G}\stackrel{~}{G}.`$ This is the combined constraint of primitivity and $`GϵH^{(2,2)}`$ expressed in components.,
$$R_{mn}=\frac{1}{2}\left(_m\mathrm{ln}e^{3\varphi /2}_n\mathrm{ln}e^{3\varphi /2}\frac{g_{mn}}{3}(\mathrm{ln}e^{3\varphi /2})^2+g_{mn}\frac{e^{3\varphi /2}}{6}\stackrel{~}{G}\stackrel{~}{G}\right),$$
(27)
with $`g_{mn}`$ the metric on the $`CY_4`$. In the decompactification limit the explicit $`G`$-flux term vanishes as it scales inversely with the volume. The warp factor does not depend explicitly on the constant flux and we should therefore recover the regular $`p`$-brane metric for which $`c_0`$=1. (Strictly speaking we have only shown that $`lim_{V_{CY}\mathrm{}}c_0(V)=1`$). The above argument is rather general, and is not restricted to orbifold cases where the curvatures are localized. This is again because $`X_8`$ and $`GG`$ are quantized (to a finite number), and so their contribution to the internal curvature is small when the size of the Calabi-Yau is large. In other words, both $`X_8`$ and $`GG`$, unlike the cosmological constant, are not extensive quantities. By way of comparison, a non-zero $`c_0`$ means that the decompactification limit of the present scenario has significantly different properties than those studied in . For instance, one would not obtain a finite Planck scale in this limit.
An issue regarding the solution (22) is that due to the opposite sign of the Green’s functions corresponding to the curvature induced charge, the metric has a naked singularity at the fixed points of the orbifold: the conical orbifold singularity. This is, however, an artifact of the long-range supergravity approximation. The solution may roughly only be trusted as long as the distance to any special point is larger than the Planck length or string scale. The naked singularity is expected to be cured by stringy effects. Examples of this have recently been discussed in . Since these effects will only modify the metric close to the singularity, they are unlikely to change the arguments above.
#### F theory
For F theory the solution is similar. We denote $`V_6`$ as the volume of the 6 compactified dimensions and $`V_2`$ the volume of the two dimensions transverse to the $`7`$-branes. The twisted fluxes are proportional to $`\delta ^4(xx_p)`$ since we consider only fluxes that are not localized around a singular fiber of the elliptic fibration: constant on one of the K3’s. The flux factor can thus be written as
$$(H^{NS}H^{RR})=\frac{r}{V_6}+\frac{1}{V_2}\underset{x_p}{}m_p\delta ^4(xx_p).$$
(28)
The curvature contribution to the equation for the warp factor is due to the (unit charge) D3-branes or O3-planes (charge $`\mu _O`$). In the orbifold limit the curvature is completely localized at the fixed points and thus gives rise to a six-dimensional delta-function. This corresponds to the fact that we are considering limit that the D7-brane charges are locally cancelled and we are left with only O3-plane charge at the fixed points. Explicitly
$`\mathrm{}e^{\frac{3}{2}\varphi }`$ $`=`$ $`{\displaystyle \underset{x_O}{}}\mu _O\delta ^6(xx_O)(H^{NS}H^{RR}){\displaystyle \underset{i=1}{\overset{n}{}}}\delta ^6(xx_i).`$ (29)
Substituting (28) we can rewrite this as
$`\mathrm{}e^{\frac{3}{2}\varphi }`$ $`=`$ $`{\displaystyle \underset{x_O}{}}\mu _O\delta ^4(xx_O)\delta ^2(zz_O){\displaystyle \frac{r}{V_6}}{\displaystyle \frac{1}{V_2}}{\displaystyle \underset{x_p}{}}m_p\delta ^4(xx_p){\displaystyle \underset{i=1}{\overset{n}{}}}\delta ^6(xx_i)`$ (30)
$`=`$ $`{\displaystyle \underset{x_O}{}}\mu _O\left(\delta ^4(xx_O)\delta ^2(zz_O){\displaystyle \frac{1}{V_6}}\right){\displaystyle \frac{1}{V_2}}{\displaystyle \underset{x_p}{}}m_p\left(\delta ^4(xx_p){\displaystyle \frac{1}{V_4}}\right)`$
$`{\displaystyle \underset{i=1}{\overset{n}{}}}\left(\delta ^6(xx_i){\displaystyle \frac{1}{V_6}}\right),`$
where we have made use of the fact that anomaly cancellation implies $`r=n+_{x_p}(m_p+\mu _O)`$. In terms of Green’s functions the warp factor is:
$$e^{\frac{3}{2}\varphi }=c_0\frac{1}{V_2}\underset{x_p}{}m_pG^4(x,x_p)+\underset{x_O}{}\mu _OG^6(x,x_O)\underset{i=1}{\overset{n}{}}G^6(x,x_i).$$
(31)
For the more general case where the $`7`$-brane charges are not locally cancelled, equation (10) (which is written in the string frame), should receive contributions from O3- and O7-planes and D3- and D7-branes plus an appropriate contribution from the dilaton which is no longer constant. The net effect is that, in the Einstein frame, the warp factor is given by an equation of the form
$$_i\left(\sqrt{g}g^{ij}_je^{3\varphi /2}\right)=\underset{x_O}{}\mu _O\delta ^6(xx_O)H^{NS}H^{RR}\underset{i=1}{\overset{n}{}}\delta ^6(xx_i),$$
(32)
where the metric $`g_{ij}`$ takes into account the backreaction of the $`7`$-branes according to (16). The exact solution to this equation is not known, although it can be solved approximately .
## 4 The Shape of Gravity
With the warped metric (2) in F theory, the tree-level four-dimensional Planck scale is given in terms of the 10-dimensional Planck scale by
$$M_4^2=M_{10}^8_{}e^{2\phi }e^{3\varphi /2}\sqrt{g^{}},$$
(33)
whereas the tree-level gauge coupling on a three-brane at $`y=y_i`$ has no warp factor contribution and is given as usual by
$$\frac{1}{g_{YM}^2(y_i)}=e^{2\phi (y_i)}.$$
(34)
As the non-trivial terms in the warp factor are solely due to backreaction, the usual relation between the four-dimensional Newton constant and the three-brane gauge coupling still holds at leading order.
At subleading order, the warp factor might come into play. However, an explicit calculation shows that this is not the case. First, let us consider the case where the dilaton is constant. Note that the power of the warp factor that appears in (33) is exactly equal to the one in the field equation (12). Since the Green’s function may be written as
$$G(y,y_i)=\underset{\lambda 0}{}\frac{\overline{f}_\lambda (y)f_\lambda (y_i)}{\lambda ^2},$$
(35)
where $`f_\lambda `$ are orthonormal eigenfunctions of the Laplacian with eigenvalues $`\lambda `$, the four-dimensional Planck mass can be expressed as
$$M_4^2=M_{10}^8e^{2\phi }\left(d^6y\sqrt{g^{}}\underset{n0}{}\frac{1}{\lambda ^2}d^6y^{}\sqrt{g^{}}\rho (y^{})f_\lambda (y^{})d^6y\sqrt{g^{}}\overline{f}_\lambda (y)\right).$$
(36)
The quantity $`\rho `$ equals the charge density on the r.h.s. of (12),
$$\rho (y^{})=_{}(\frac{1}{16}\underset{i=1}{\overset{4}{}}\text{tr}(RR)\delta ^2(y^{}y_i)H^{NS}H^{RR})\underset{j=1}{\overset{n}{}}\delta ^6(y^{}y_j).$$
(37)
Because the zero mode is explicitly excluded in the Green’s function, the second term in (36) vanishes. This bears out our expectations. The warp factor does not change the relation between the four and ten-dimensional Planck scale.<sup>3</sup><sup>3</sup>3 If the dilaton is not constant (taking into account the $`7`$-branes), the net effect is the modification of the field equation to (32) where the metric is expressed in the Einstein frame. The above argument based on the properties of the Green’s function should also hold in this case.
In , an exponentially decaying wavefunction of the graviton was used to generate the hierarchy between the electroweak scale and the four-dimensional Planck scale. The exponential decay is due to the fact that the wavefunction of the massless graviton is dressed by some powers of the warp factor. By way of comparison, let us also deduce the shape of the graviton wavefunction in the present setup by linearizing fluctuations about the background warped metric in Section 3. We will see that although the massless spectrum is indeed unaffected, the masses of the KK modes will get dressed by powers of the warp factor. A few comments are in order. If the compactification is of the order of string scale, we expect, a priori, that the wavefunctions of the KK modes are no longer given by linearizing Einstein gravity since stringy effects may become important. However, in compactifying Heterotic-M theory on a Calabi-Yau three-fold (the G flux in this case is generically non-zero), there is a regime in which the theory is effectively five-dimensional, the wavefunctions of some of the lower-lying KK modes (in the fifth dimension) can still be obtained by linearizing Einstein gravity. The analysis for Heterotic-M theory on $`CY_3`$ is very similar, so for completeness, we have included a derivation of the KK spectrum in the appendix.
For a metric of the form
$$ds^2=e^{2A(y)}\eta _{\mu \nu }dx^\mu dx^\nu +e^{2B(y)}\overline{g}_{ab}dy^ady^b,$$
(38)
the gravitational fluctuations are given by $`\eta _{\mu \nu }\eta _{\mu \nu }+h_{\mu \nu }`$ where $`h_{\mu \nu }`$ is small compared with $`\eta _{\mu \nu }`$.
From Einstein’s equations, if we choose the gauge $`^\mu h_{\mu \nu }=0`$, the linear fluctutations can be shown to satisfy the covariant wave equation (see the discussion in the next section or e.g. ):
$$\frac{1}{\sqrt{g}}_M\left(\sqrt{g}g^{MN}_Nh_{\mu \nu }\right)=0,$$
(39)
where the indices $`M,N=0,1,\mathrm{},9`$ are raised and lowered with the warped background metric (in Einstein frame). For the warped metric in M and F theory that we considered (i.e., $`A(y)`$ is proportional to $`B(y)`$), this reduces to
$$\left(e^{2A(y)}\mathrm{}_{Mink}+\mathrm{}_{\overline{g}}\right)h_{\mu \nu }=0.$$
(40)
Expanding $`h_{\mu \nu }(x,y)=\psi (y)\widehat{h}_{\mu \nu }(x)`$, with $`\mathrm{}_x\widehat{h}_{\mu \nu }(x)=m^2\widehat{h}_{\mu \nu }(x)`$, we have
$$\mathrm{}_{\overline{g}}\psi (y)=m^2e^{2A(y)}\psi (y).$$
(41)
Hence the masses of the KK modes depend on the warp factor, which in turn depends on the locations of the branes.
For the massless graviton (i.e., $`m^2=0`$), (41) always admit the solution $`\psi (y)=`$ constant. The wavefunction for the graviton, however, should be properly normalized:
$$Sd^{10}x\sqrt{g}_\lambda h_{\mu \nu }^\lambda h_{\mu \nu }+\mathrm{}=𝑑ye^{2A(y)}\sqrt{g(y)}\psi ^2(y)d^4x_{\widehat{\lambda }}\widehat{h}_{\mu \nu }(x)^{\widehat{\lambda }}\widehat{h}_{\mu \nu }(x)+\mathrm{}$$
(42)
where hatted indices are with respect to the unwarped metric. Therefore, the properly normalized wavefunction is
$$\mathrm{\Psi }(y)=\left[e^{2A(y)}\sqrt{g(y)}\right]^{1/2}\psi (y)=e^\phi e^{3\varphi /4}\left[detg_{ab}^{}\right]^{1/4},$$
(43)
which is the square root of the integrand in (33), as expected. For the orbifold examples that we discuss in Section 3, the dilaton $`\phi `$ and $`\left[detg_{ab}^{}\right]^{1/4}`$ are constant, with $`e^{3\varphi /2}`$ given in terms of the Green’s functions, hence
$$\mathrm{\Psi }(y)\left[c_0\frac{1}{V_2}\underset{x_p}{}m_pG^4(x,x_p)+\underset{x_O}{}\mu _OG^6(x,x_O)\underset{i=1}{\overset{n}{}}G^6(x,x_i)\right]^{1/2}.$$
(44)
An alternative view of these warped compactifications was suggested in . In extending the AdS/CFT correspondence to the full string theory (i.e., without taking the scaling limits as in ), the closed string degrees of freedom including gravity are no longer decoupled from the effective theory on the world-volume of the branes. To account for the closed string degrees of freedom, one introduces into the AdS supergravity a hypothetical Planck brane with dynamical degrees of freedom representing these closed string modes. This hypothetical Planck brane is placed at the edge of the AdS throat created by the branes and effectively cuts off the radial AdS coordinate. In the AdS/CFT correspondence, distances from the brane correspond to energy scales in the worldvolume theory. Therefore, the Planck brane serves as an UV cutoff and quantum gravity effects become important as we get closer to the Planck brane.
In the case that the spacetime transverse to the branes is compactified, the information about the compactification geometry would then be encoded in vacuum expectation values of the excitations of the Planck brane. In principle one could derive this set-up by integrating out coordinate shells of constant warp factor (momentum shells on the world volume) extending beyond the throat of the AdS near horizon region.
## 5 The Low Energy Spectrum
The low energy spectrum of warped compactifications is naively different from that of product space compactifications where one may use topological information of the internal manifold to determine the massless spectrum. For example the equation of motion of a scalar field,
$$\frac{1}{\sqrt{g}}_M\sqrt{g}g^{MN}_N\varphi (z)=0,$$
(45)
when reduced on a generic warped metric of the form
$$g_{MN}(z)=\left(\begin{array}{cc}e^{2A(y)}\stackrel{~}{g}_{\mu \nu }(x)& 0\\ 0& \overline{g}_{ab}(y)\end{array}\right)$$
(46)
yields
$$(e^{2A}\mathrm{}_{\stackrel{~}{g}}+\mathrm{}_{\overline{g}}+d\overline{g}^{ab}_aA_b)\varphi (z)=0.$$
(47)
Here $`d`$ denotes the dimension of the internal manifold; $`D`$ is the dimension of the ambient space. Redefining $`\varphi (z)=e^{dA/2}\stackrel{~}{\varphi }(z)`$ one finds that the low energy modes descending from this field are determined by
$$\left(e^{2A}\mathrm{}_{\stackrel{~}{g}}+e^{\frac{dA}{2}}\left[\mathrm{}_{\overline{g}}\frac{d}{2}\mathrm{}_{\overline{g}}A\frac{d^2}{4}(A)^2\right]\right)\stackrel{~}{\varphi }(z)=0.$$
(48)
The last two terms act as additional mass terms for the dimensionally reduced field. The massless modes are those in which the eigenvalue of the internal Laplacian cancels the terms descending from the warp factor. Naively one thus loses the power of topological arguments to determine the massless spectrum.
The particular warped solutions known in string theory/supergravity, the $`p`$-brane metrics, belong to a special class, however. The internal manifold itself is also multiplied by a warp factor which precisely compensates for the warp factor of the external space. For example, in the scalar field above, the effect of an extra warp factor on the internal space,
$$g_{MN}=\left(\begin{array}{cc}e^{2A(y)}\stackrel{~}{g}_{\mu \nu }(x)& 0\\ 0& e^{2bA}\overline{g}_{ab}(y)\end{array}\right),$$
(49)
changes the field equation for the low-energy modes $`\varphi `$ to
$$(e^{2A}\mathrm{}_{\stackrel{~}{g}}+\mathrm{}_{\overline{g}}+(db(Dd2))\overline{g}^{ab}_aA_b)\varphi (z)=0.$$
(50)
This internal warp factor can cancel the additional mass terms in (48) if $`b`$ is chosen appropriately: $`b=d/(Dd2)`$. This is in fact exactly the combination one finds for $`p`$-brane metrics in supergravity. Here, the physical reason is that the warp factor is solely due to the backreaction of the vacuum configuration. Indeed one can explicitly see from the solution that the non-constant terms in the warp factor are suppressed by powers of the gravitational coupling constant. In the limit where this vanishes the warp factor is trivial and the space is of the product form $`M_4\times CY_4`$. This is another argument why also in the case with background fluxes, the integration constant equals unity. In fact, the “balancing” of the warp factors is essentially the reason why when we first quantize open strings with Dirichlet-boundary conditions, the gravitational backreaction of the D-branes does not change the corresponding massless closed string spectrum.
The argument that in “balanced” warp metrics the warp factor is to a large extent inconsequential as regards to the massless part of the low-energy spectrum holds irrespective of whether $`G`$-flux is present or not. The introduction of the latter does complicate the determination of the massless spectrum as in the presence of non-trivial background expectation values of matter fields the mass matrix of low energy modes is generically off-diagonal in terms of the original fields. Fortunately the fact that the allowed background fluxes are subject to topological conditions - $`G`$ should be primitive and a (2,2) form - in addition to the topological interpretation of the moduli of the Calabi-Yau four-fold ought to allow one to also use topological methods to determine the massless spectrum with $`G`$-flux . In general these low-energy modes could be a linear combination of the original fields, but their number can be determined by looking at the topological constraints.
To be specific, those Kahler moduli which spoil the primitivity condition $`JG=0`$ are lifted as well as those complex structure moduli which fail to keep $`G`$ a (2,2) class. In addition those Wilson lines which are not orthogonal to $`G`$: $`CG0`$ are lifted as well . In principle one has hereby computed the massless spectrum. In practice, the first and the last constraint are readily solved but the counting of those complex structure deformations which keep $`G`$ a (2,2) class is more involved. Consider a deformation of the complex structure. This is given by a coordinate transformation which is not holomorphic
$$z^iy^i(z,\overline{z}).$$
(51)
Infinitesimally the mixed and the pure deformations of the metric under an arbitrary coordinate transformation are
$`\delta g_{a\overline{b}}`$ $`=`$ $`g_{a\overline{c}}\overline{}_{\overline{b}}\overline{y}^{\overline{c}}+g_{\overline{b}c}_ay^c+y^c_cg_{a\overline{b}}+\overline{y}^{\overline{c}}\overline{}_{\overline{c}}g_{a\overline{b}},`$
$`\delta g_{ab}`$ $`=`$ $`g_{a\overline{c}}_b\overline{y}^{\overline{c}}+g_{b\overline{c}}_a\overline{y}^{\overline{c}}g_{\overline{c}(a}\overline{\chi }_{b)}^{\overline{c}}.`$ (52)
This is just the well known fact that non-holomorphic coordinate changes correspond to pure type metric deformations. By contracting $`g^{\overline{c}a}\delta g_{ab}h_b^{\overline{c}}`$ with the constant anti-holomorphic $`(0,n)`$ form one recovers the $`(1,n1)`$ forms that are in one-to-one correspondence with complex structure deformations. Under such a transformation a (2,2) form transforms infinitesimally as
$$GG_{a\overline{b}c\overline{d}}dz^ad\overline{z}^{\overline{b}}dz^cd\overline{z}^{\overline{d}}G+K^{(3,1)}+K^{(1,3)}$$
(53)
with
$$K^{(1,3)}=G_{a\overline{b}c\overline{d}}(\chi _{\overline{e}}^a\delta _e^c\chi _{\overline{e}}^c\delta _e^a)dz^ed\overline{z}^{\overline{b}}d\overline{z}^{\overline{d}}d\overline{z}^{\overline{e}}.$$
(54)
For those complex structure deformations which keep $`G`$ a (2,2) class $`K^{(1,3)}`$ must vanish. This constraint can be expressed as follows. The complex structure deformations $`\chi _{\overline{b}}^a`$ are representatives of the cohomology $`H^{(0,1)}(T)`$. Define the form $`\stackrel{~}{G}ϵH^{(0,2)}(^2T)`$
$$\stackrel{~}{G}_{\overline{c}\overline{d}}^{ab}=G_{\overline{c}\overline{d}ef}\mathrm{\Omega }^{efab}.$$
(55)
The constraint that $`K^{(3,1)}`$ must vanish can then be expressed by requiring that the triple intersection numbers
$$\mathrm{\Omega }_{abcd}\left(\chi _{(0,1)}^{[a}\stackrel{~}{G}_{(0,2)}^{bc}\stackrel{~}{\alpha }_{I(0,1)}^{d]}\right)\mathrm{\Omega }_{(4,0)}=0$$
(56)
vanish for all basis elements $`\stackrel{~}{\alpha }_I`$ of $`H^{(0,1)}(T)`$. The latter are constructed from basis elements $`\alpha _I`$ of $`H^{(3,1)}(CY_4)`$ as in (55). Equation (56) thus represents that the natural inner product of $`K^{(1,3)}`$ with an arbitrary element $`\alpha _I`$ vanish. There is however no general analysis of this condition if the (2,2) form $`G`$ is also required to be integral. One therefore has to solve (56) on a case by case basis. Below we will do so for the simple example where the $`CY_4`$= $`K3\times K3`$.
The deformations that do not respect the constraints should give rise to mass terms when we perform a KK reduction. By our previous reasoning we can to first order ignore the effects of the warp factor. Then, we are on a product space and the KK reduction is straightforward, with the caveat that the field equations are now not satisfied. One finds that the kinetic term for $`C`$ is responsible for mass terms for the pure and mixed type metric fluctuations (numerical coefficients are suppressed)
$$m^2h^2h^{ab}G_{a\overline{b}c\overline{d}}G_{b\overline{c}}^{c\overline{d}}h^{\overline{b}\overline{c}}+h^{a\overline{b}}G_{ab\overline{e}\overline{d}}G_{\overline{b}\overline{c}}^{\overline{g}\overline{e}}h^{b\overline{c}}+G_{a\overline{b}c\overline{d}}h^{\overline{d}\overline{k}}h_{\overline{k}}^eG_e^{a\overline{b}c}+G_{a\overline{b}c\overline{d}}h^{\overline{d}k}h_k^eG_e^{a\overline{b}c}.$$
(57)
In the first and the third term one recognizes the “square” of $`K^{(3,1)}`$ in (54). For massless deformations it must vanish.
One may compare this with the conjectures made in . There it was argued that the complex structure deformations have a superpotential
$$W_C=\mathrm{\Omega }G,$$
(58)
where $`\mathrm{\Omega }`$ is the holomorphic (4,0) form. For the calculation of the mass terms involving the complex structure deformations, the relevant parts of the first and second deformation of the four-form $`\mathrm{\Omega }`$ are
$`\left(\delta \mathrm{\Omega }\right)_{abc\overline{n}}`$ $``$ $`\mathrm{\Omega }_{abcd}g^{d\overline{m}}h_{\overline{m}\overline{n}}+\mathrm{},`$
$`\left(\delta ^2\mathrm{\Omega }\right)_{ab\overline{n}\overline{f}}`$ $``$ $`\mathrm{\Omega }_{abcd}g^{c\overline{m}}h_{\overline{m}\overline{n}}g^{d\overline{e}}h_{\overline{e}\overline{f}}+\mathrm{}`$ (59)
The non-trivial quadratic fluctuations in the superpotential are thus given by
$$\delta ^2W=\delta ^2\mathrm{\Omega }G\sqrt{g}h_{\overline{m}}^nh_{\overline{c}}^d\mathrm{\Omega }_{abdn}G^{ab\overline{c}\overline{m}},$$
(60)
where we have used the fact that the background value of $`G^{(2,2)}`$ is selfdual. Reducing to $`N=0`$ components and noting that the $`W^2|_{\theta =0}`$ term does not contribute, we find the mass term for the pure deformations
$$m^2h_{zz}^2h_{\overline{c}}^d\mathrm{\Omega }_{abdn}G^{ab\overline{c}\overline{m}}G^{np\overline{q}\overline{r}}\overline{\mathrm{\Omega }}_{\overline{r}\overline{q}\overline{s}\overline{m}}h_p^{\overline{s}}.$$
(61)
Using the relation
$$\mathrm{\Omega }_{abcd}\overline{\mathrm{\Omega }}_{\overline{a}\overline{b}\overline{c}\overline{d}}ϵ_{abcd\overline{a}\overline{b}\overline{c}\overline{d}},$$
(62)
the mass term reduces to
$$m^2h_{zz}^2h_{\overline{s}}^dG_{\overline{q}\overline{r}nd}G^{np\overline{q}\overline{r}}h_p^{\overline{s}}+h_{\overline{q}}^dG_{\overline{s}\overline{r}nd}G^{np\overline{q}\overline{r}}h_p^{\overline{s}},$$
(63)
which is of the same form as the first and third mass terms from the Lagrangian in (57).
For the Kahler deformations the authors of conjectured the superpotential
$$W_K=𝒦𝒦G,$$
(64)
where $`K`$ is the complexified Kahler form $`𝒦=J+iB`$. The reduction to $`N=0`$ components in this case is more involved. This will be reported elsewhere.
## 6 An Example: $`K3\times K3`$
In this section, we will compute the spectrum for the simple example where the $`CY_4`$ equals $`K3\times K3`$. We will first determine the number of complex and Kahler moduli by explicitly constructing the complete parameter space of solutions to the $`G`$-flux constraints on the orbifold limit of $`K3\times K3`$. This allows us to simply count the number of moduli by hand. At the end we will compare it to the topological determination of allowed deformations outlined above.
#### Vanishing Fluxes
Before proceeding let us briefly recall for comparison the spectrum of M theory on a $`K3\times K3`$ without G-flux. In this case there are $`H^{(3,1)}+H^{(2,1)}`$ $`N=1,d=4`$ chiral multiplets, corresponding to deformations of the complex structure and Wilson lines of the three-form respectively, and $`H^{(1,1)}`$ $`N=1,d=4`$ vector fields, corresponding to Kahler deformations. For $`CY_4`$ = $`K3\times K3`$ the Hodge diamond equals
$`\begin{array}{ccccccccc}& & & & h^{0,0}& & & & \\ & & & h^{1,0}& & h^{0,1}& & & \\ & & h^{2,0}& & h^{1,1}& & h^{0,2}& & \\ & h^{3,0}& & h^{2,1}& & h^{1,2}& & h^{0,3}& \\ h^{4,0}& & h^{3,1}& & h^{2,2}& & h^{1,3}& & h^{0,4}\end{array}=\begin{array}{ccccccccc}& & & & 1& & & & \\ & & & 0& & 0& & & \\ & & 2& & 40& & 2& & \\ & 0& & 0& & 0& & 0& \\ 1& & 40& & 404& & 40& & 1\end{array}`$ (65)
The $`K3\times K3`$ compactifaction is non-minimal in that it preserves four more supercharges than the minimal four required by $`N=1,d=4`$ supersymmetry. Indeed we see that the spectrum is given by 40 $`N=2,d=4`$ vector multiplets.
Taking one of $`K3`$’s to be elliptically fibered we can lift this M theory compactification to F theory on $`K3\times K3`$, which is related to a number of other compactifications by a chain of dualities . Since F theory on $`K3`$ is heterotic on $`T^2`$, compactifying both sides on another K3 gives heterotic on $`K3\times T^2`$. On the other hand, heterotic on $`K3`$ is F theory on a Calabi-Yau threefold $`CY_3`$. As a result, F theory on $`K3\times K3`$ is dual to F theory on $`CY_3\times T^2`$.
The orientifold dual of the above F theory compactification can be found via Sen’s map . We take the orbifold limit $`K3=T^4/_2`$ for each $`K3`$. Let $`(z_1,z_2,z_3,z_4)`$ be the complex coordinates of $`T^8`$, with $`z_4`$ being the fiber coordinate. The orientifold dual is given by Type IIB on $`T^6/\{R_1\times (\mathrm{\Omega }(1)^{F_L}R_2)\}`$ where $`\mathrm{\Omega }`$ is the worldsheet parity operation, $`R_1`$ and $`R_2`$ act as follows:
$`R_1z_{1,2}`$ $`=`$ $`z_{1,2},R_1z_3=z_3,`$ (66)
$`R_2z_{1,2}`$ $`=`$ $`z_{1,2},R_2z_3=z_3.`$ (67)
The resulting orientifold dual is therefore the Gimon-Polchinski model further compactified on a $`T^2`$ and then T-dualized in the $`T^2`$ directions, so that instead of $`5`$-branes and $`9`$-branes, we have $`3`$-branes and $`7`$-branes. At a special point of the moduli space where $`\tau `$ is constant, i.e., four $`D7`$-branes are placed on top on each $`O7`$-plane, the gauge group from the $`7`$-branes is $`U(4)^4`$. In addition, there are two $`𝒩=2`$ hypermultiplets in the $`\mathrm{𝟔}`$ representation of each $`U(4)`$ from the $`77`$ open strings. The $`3`$-branes give rise to additional gauge symmetries. In the case of maximal gauge symmetries, i.e. the $`3`$-branes are sitting on top of each other, the gauge group from the $`3`$-branes is $`U(16)`$. The $`33`$ open strings also give rise to two $`𝒩=2`$ hypermultiplets in the $`\mathrm{𝟏𝟐𝟎}`$ representations of $`U(16)`$. Finally, there is a hypermultiplet in the bi-fundamental representation $`(\mathrm{𝟏𝟔},\mathrm{𝟒})`$ of $`U(16)\times U(4)`$ if the $`3`$-branes sit on one of the four groups of $`7`$-branes. The orientifold model at a generic point of the moduli space can be obtained from the above by Higgsing.
### 6.1 G-flux conditions
Now we turn on background fluxes. The field-equations demand that we seek an integer (2,2) form on $`K3\times K3`$ = $`T^4/_2\times T^4/_2`$ which is primitive, i.e.
$$JG=0.$$
(68)
We will choose $`G`$ of the form
$$G=\omega _1\omega _2,$$
(69)
where $`\omega _iϵH^{(1,1)}(T^4/_2)H^{(2)}(T^4/_2,)`$. This guarantees that $`GϵH^{(2,2)}(T^4/_2,)`$ though it is not necessary. This simplifies the solutions to the quantization conditions (as compared to the general case ). We could have chosen $`\omega _iϵH^{(2,0)}(T^4/_2,)`$ or $`H^{(0,2)}(T^4/_2,)`$ but these forms correspond to the class of the fiber and its Hodge dual and are no longer normalizable when we shrink the volume of the fiber in the F theory limit .
The complex structure is inherited from the tori and the condition (68) requires that each $`\omega _i`$ is primitive with respect to $`J_i`$ . Thus our task is reduced to finding primitive (1,1) forms on $`T^4/_2`$. The cohomology $`H^{(1,1)}(K3=T^4/_2)`$ has dimension 20, but these can be subdivided in four “untwisted” (1,1) forms inherited from the $`T^4`$ and sixteen “twisted” ones. As any integer form is an integer combination of the basis forms we can consider the two situations separately.
#### Constant fluxes
Consider untwisted forms first. These are inherited from the $`T^4`$ and are constant on the orbifold. The intersection matrix of two-forms on K3$`=T^4/_2`$ is block-diagonal with in the upper left-hand corner minus the Cartan-matrix of $`E_8\times E_8`$ and in the lower right-hand corner three times the Pauli matrix $`\sigma _1`$. The latter blocks equal the intersection-matrix of two-forms on $`T^4`$ and thus correspond to the untwisted forms. Hence we only have to check that the periods of the untwisted forms are integer over the untwisted cycles. This means that we have reduced our problem to finding the set of integer (1,1) forms $`D`$ on $`T^4`$.
For each $`T^2`$ (with volume normalized to 1), we make the standard choice for the periods
$$_{\gamma _x^j}𝑑z^i=\delta _j^i_{\gamma _y^j}𝑑z^i=\tau _i\delta _j^i\text{(no sum on }i\text{)},$$
(70)
where $`\gamma _x^i`$ and $`\gamma _y^i`$ are the $`x`$ and $`y`$ cycles; $`i,j=1,2`$. The (1,1) forms $`\stackrel{~}{\gamma }_x^i`$ and $`\stackrel{~}{\gamma }_y^i`$ dual to these cycles are
$$\stackrel{~}{\gamma }_x^i=dy^i\stackrel{~}{\gamma }_y^i=dx^i,$$
(71)
with the volume element
$$_{T_i}𝑑x^idy^i=1\text{(no sum on }i\text{)}.$$
(72)
The Kahler form is given by
$$J_i=dz^id\overline{z}^i,$$
(73)
and equals $`(\overline{\tau }_i\tau _i)`$ times the volume form $`dx^idy^i`$.
Take the Kahler form $`J`$ for the $`T^4`$ to be the sum of $`J_1`$ and $`J_2`$. The general form of (1,1) forms obeying the primitivity condition is then
$$JD=0D=Adz_1d\overline{z}_2+\overline{A}d\overline{z}_1dz_2+iB(dz_1d\overline{z}_1dz_2d\overline{z}_2).$$
(74)
where $`A`$ is complex and $`B`$ is real. The requirement that $`DϵH^{(1,1)}(T^4,)H^{(1,1)}(T^4/_2,)`$ demands that
$`A+\overline{A}`$ $`=`$ $`n,`$
$`\overline{\tau }_2A+\tau _2\overline{A}`$ $`=`$ $`m,`$ (75)
$`\tau _1A+\overline{\tau }_1\overline{A}`$ $`=`$ $`p,`$
$`\tau _1\overline{\tau }_2A+\overline{\tau }_1\tau _2\overline{A}`$ $`=`$ $`q,`$
where $`n,m,p,qϵ`$. In addition
$`iB(\overline{\tau }_1\tau _1)`$ $`=`$ $`v,`$
$`iB(\overline{\tau }_2\tau _2)`$ $`=`$ $`w,`$ (76)
where $`v,wϵ`$. These are six equations with in principle three unknowns: Re$`(A)`$, Im$`(A)`$, $`B`$; the system is overconstrained and has no solutions.
However, if one relaxes the requirement that all the $`\tau _i`$ are free parameters, namely one allows the complex structure $`\tau _2`$ of one torus to be determined in terms of the other, and in addition requires that the ratio of the imaginary parts of $`\tau _1`$ and $`\tau _2`$ be a rational number, then there is a solution. This poses three real constraints on the moduli of the $`T^4`$. The first constraint can be seen by rewriting the integrality conditions as
$`\overline{A}`$ $`=`$ $`nA,`$
$`A`$ $`=`$ $`{\displaystyle \frac{pn\overline{\tau }_1}{(\tau _1\overline{\tau }_1)}}={\displaystyle \frac{mn\tau _2}{(\overline{\tau }_2\tau _2)}},`$ (77)
$`{\displaystyle \frac{\tau _1\overline{\tau }_1}{\tau _2\overline{\tau }_2}}`$ $`=`$ $`{\displaystyle \frac{n\overline{\tau }_1p}{mn\tau _2}},`$
$`\tau _1\overline{\tau }_2A+\overline{\tau }_1\tau _2\overline{A}`$ $`=`$ $`q.`$
Note, however, that we have imposed both the primitivity and integrality conditions in deriving these constraints. Moreover, in (74) we are only considering (1,1) classes. Hence, the three constraints determine the loci in the combined space of complex and Kahler structures on K3 where one may find primitive integral forms which are purely (1,1). The number of constraints therefore correspond to the total number of moduli, complex structure plus Kahler, which are lifted. Complex structure deformations must form complex pairs. The total number of moduli which become massive is therefore 1 (complex valued) complex structure deformation and 1 real Kahler deformation. For instance, one can see this explicitly by generalizing the choice of the Kahler class to $`J_1+aJ_2`$ and then noting that the analogue of eq.(76) fixes the value of $`a`$ in terms of the complex structures $`\tau _1`$ and $`\tau _2`$.
In the end we are interested in the $`H^{(2,2)}(T^4,)`$ flux
$$_{T^4}DD=2(|A|^2B^2)(\tau _1\overline{\tau }_1)(\tau _2\overline{\tau }_2).$$
(78)
which should be an integer. This is proportional to the number of M2/D3-branes we will have to introduce. Substituting the second equation of (77) in (78) we find that
$$_{T^4}DD=2vw2(pn\overline{\tau _1})(mn\overline{\tau _2}).$$
(79)
As the last factor should be a negative semi-definite integer ($`\tau _iϵH^+`$ in (78)) let us make the Ansatz that
$`mn\overline{\tau _2}={\displaystyle \frac{r}{(pn\overline{\tau _1})}}`$ $``$ $`\overline{\tau }_2=\left({\displaystyle \frac{m}{n}}{\displaystyle \frac{r}{n(pn\overline{\tau _1})}}\right)`$ (80)
is indeed a solution to (77). Here $`rϵ^+/2`$. Substituting this relation into the four equations (77) we find that our Ansatz is a solution with $`r=qn+mp`$. Note that $`mpqn`$. Hence the flux equals
$$_{T^4}DD=2vw2(pn\overline{\tau _1})(mn\overline{\tau _2})=2(qnmpvw).$$
(81)
Note that if $`D`$ is also an element of $`H^{(1,0)}(T^2,)\times H^{(0,1)}(T^2,)H^{(1,1)}(T^4,)`$, i.e. if $`n=n_1n_2`$; $`m=n_1m_2`$; $`p=m_1n_2`$ and $`q=m_1m_2`$ for integer $`n_1,n_2,m_1,m_2`$ then the flux $`_{T^4}DD=2(m_1m_2n_1n_2n_1m_2m_1n_2)`$ vanishes.
#### Twisted fluxes
We will restrict ourselves to twisted forms whose dual cycles are localized at the fixed points. These are two-speres $`S^2CP_1`$ shrunk to a point. Denoting them as $`B^i`$ with $`i=1,\mathrm{},16`$ running over the fixed points, the general form of a twisted flux $`D`$ is
$$D=c_iB^i.$$
(82)
The $`B^i`$ are a linear combination of the generating forms $`V^n`$ of $`H^{(1,1)}(T^4/_2,)`$: $`B^i=a_n^iV^n`$ with $`a_n^i`$ integer. As we are limiting our attention to twisted fluxes we may restrict the $`V^n`$ to those generating minus the $`E_8\times E_8`$ Cartan matrix $`I^{nm}`$. This also means that $`D`$ is automatically primitive as the Kahler form $`J`$ consists purely of untwisted forms.
The condition that also $`DϵH^{(1,1)}(T^4/_2,)`$ means that
$$_{V^n}D=c_ia_m^iI^{mn}=p^n,$$
(83)
with $`p^nϵ`$ for all $`n`$. The cycles $`B^i`$ have an intersection matrix
$$B^iB^j=2\delta ^{ij}=a_n^iI^{nm}a_m^j.$$
(84)
This means that the coefficients $`c_i`$ are all multiples of $`1/2`$,
$$c^i=\frac{p^na_n^i}{2}.$$
(85)
The four-form flux of interest to us $`_{T^4/_2}DD`$ equals
$$_{T^4/_2}DD=2\underset{i}{}c^ic^i=p^nI_{nm}^1p^m,$$
(86)
as one can show that
$$\underset{i}{}a_n^ia_m^i=2I_{nm}^1.$$
(87)
The inverse of the Cartan matrix has integer entries because the $`E_8`$ lattice is even and self-dual, and the diagonal entries are even. The r.h.s. of (86) is therefore an integer as expected. Note again that $`DD`$ is negative semidefinite just as in the case of constant fluxes.
Note that the $`B_i`$ or more precisely arbitrary half-integer combinations thereof do not generate $`H^{(1,1)}(T^4/_2,)`$ , but a larger group. For integer cohomology the $`c^i`$ are subject to the additional constraint
$$\underset{i}{}a_m^ic^i=p^nI_{nm}^1.$$
(88)
We are therefore not missing any integer forms by limiting our attention to fluxes of the form (82).
In this case those Kahler deformations $`\delta J=\beta _nV^n`$ where $`V^nϵH^{(1,1)}(T^4/_2,)`$ are twisted fluxes that do not preserve the primitivity condition are frozen. Those deformations such that
$$\delta JD=\beta _nI^{nm}c_ia_m^i=0\beta _n$$
(89)
do survive. The integrated primitivity condition is sufficient to guarantee the local one .
In fact we can determine the contribution to $`DD`$ of each fixed point separately. As the intersection matrix of the cycles $`B^i`$ is proportional to the unit matrix, the contribution of the ‘$`i`$’th cycle is just given by the ‘$`i`$’th term in equation (86) . Since the orientifold planes are also localized at the fixed points twisted fluxes change the effective O-plane charge.
### 6.2 The Spectrum
In the previous section, we explicitly constructed the parameter space of solutions for pure $`(1,1)`$ forms on $`K3`$ that are primitive and integral. For constant flux this showed that the dimension of the space of complex structures and the dimension of the space of Kahler forms are each reduced by one. In this section we will compare these results with a topological deformation analysis.
Suppose one is given a complex structure and Kahler form on $`K3`$ for which it is possible to find a primitive integral (1,1) flux $`\omega _i`$. The $`\omega _iH^{(1,1)}(K3)H^2(K3,)`$ define a single direction in $`H^{(1,1)}(K3)H^2(K3,)`$. The Kahler deformations which are preserved are those which are orthogonal to $`\omega _i`$ in the sense that
$$\delta J\omega _i=0.$$
(90)
Thus there is always exactly one Kahler deformation in each K3 that is lifted.
As for the complex structure deformations, on a $`K3`$ they will cause a (1,1) form to become a mixture of a $`(1,1)`$ and a $`(0,2)`$+$`(2,0)`$ form; the latter being complex conjugates. Since $`H^{(0,2)}`$ is 1 dimensional, a 19 dimensional subspace of (1,1) forms is preserved. These allowed deformations will not spoil integrality, but could spoil primitivity. One can, however, always find a compensating Kahler deformation which restores primitivity. This is illustrated in the constant flux example by our earlier explicit calculation.
Since on $`K3\times K3`$, the (2,2) classes come from $`(1,1)\times (1,1)`$ and $`(2,0)\times (0,2)`$, we find that there is a $`38=19+19`$ dimensional subspace of preserved complex structure deformations on $`K3\times K3`$.
An alternative way to see this embodies the symmetry between complex structure deformations and Kahler deformations. Consider in analogy with the superpotential (58) of the intersection
$$\mathrm{\Omega }_i\omega _i$$
(91)
for each $`K3`$. $`\mathrm{\Omega }_i`$ is now the holomorphic (2,0) form Under a complex structure deformation this potential changes to
$$\delta \mathrm{\Omega }_i\omega _i.$$
(92)
We see that that deformation which is parallel to the $`\omega _i`$ flux is lifted.
These constraints thus change the number of chiral and vector multiplets by one in each K3. As a result, a total of two $`N=2`$, $`d=4`$ vector multiplets are lifted due to the G-flux. The spectrum (in addition to that from the branes) thus consists of 38 $`N=2`$, $`d=4`$ $`U(1)`$ vector multiplets, coupled to gravity.
## 7 Discussion
In this paper, we have taken some modest steps towards understanding warped compactifications in M and F theory on Calabi-Yau four-folds in the presence of non-trivial background flux. The introduction of background flux can have interesting consequences for low energy physics and this subject is certainly worthy of further study. Detailed investigations require the explicit form of the warp factor, and to facilitate its determination, we considered orbifold limits of these compactifications. The contribution of the background flux is a backreaction effect similar to explicit $`p`$-brane sources and because the energy density associated with the G-flux is inversely proportional to the volume, the leading term in the warp factor is always a constant. This contrasts sharply to what has been found in a number of other warped scenarios. In the orbifold limit the background fluxes are either constant or localized. The constant fluxes allow the warp factor to be consistently solved in terms of Green’s functions on the internal space. The twisted fluxes act as sources for the Green’s functions.
The extensive nature of the background flux suggests that its introduction will not have drastic consequences; indeed we find that the usual relation between the four- and ten-dimensional Planck scales is recovered. This, again contrasts with the scenarios involving a cosmological constant , in which gravity is localized. The shape of the graviton wavefunction is sharply peaked at the location of the branes but this is just the source divergence of a Green’s function. In a similar analysis for Heterotic-M theory on a Calabi-Yau three-fold, discussed in an appendix, we also derived the wavefunctions of the KK modes In our M/F theory setup, the higher KK modes can also be obtained by linearizing gravity if the compactification size is large. Interestingly, aside from the compactification size, at subleading order the masses of the KK modes appear to depend on the location of the branes.
In the view of , in which the renormalization flow of the effective world volume theory is correlated with a translation in one of the real internal directions, the compactification geometry is encoded by a hypothetical Planck brane at the throat of the near-horizon AdS created by the branes. The shape of the warp factor should determine the characteristics of this Planck brane.
Finally we discussed the low energy spectrum of these warped compactifications. Supersymmetry or the argument that the warp factor is solely due to backreaction guarantees that topological arguments may still be used to determine the massless spectrum. The presence of background fluxes can lift a number of moduli, but the rules to determine them are of (pseudo) topological nature as well. For $`K3\times K3`$ we found that two $`N=2`$ vector multiplets were lifted. In addition to the complex and Kahler moduli that arise in these compactifications, the low energy degrees of freedom include gauge and matter fields on the branes. For example, in the $`K3\times K3`$ case, there are generically $`3`$ and $`7`$ branes on which gauge and matter fields are localized. We recalled in Section 6 the spectrum from the branes when the background flux is zero. The determination of the spectrum from the brane sector of F theory compactifications in the presence of background three-form NS-NS and R-R fluxes is an interesting direction to pursue in the future.
Acknowledgments
We would like to thank Jan de Boer, Claude LeBrun, Lennaert Huiszoon, Zurab Kakushadze, David Morrison, Horatiu Nastase, Peter van Nieuwenhuizen and Savdeep Sethi for discussions and correspondence. The research of G.S. is partially supported by the NSF grant PHY-97-22101, and that of B.R.G. is partially supported by the DOE grant DE-FG02-92ER40699B. K.S. acknowledges the hospitality of the C.N. Yang ITP and the Spinoza Institute.
## Appendix
### A. Green’s Functions on a compact space
On a compact space, or equivalently when the Laplacian has non-trivial zero modes, the inversion of the Laplacian is only defined on the subset of functions orthogonal to these zero-modes. One can see this explicitly by constructing the Green’s function in terms of the complete set of eigenfunctions of the Laplacian. These eigenfunctions must obey the same boundary conditions as the function on which the Laplacian acts. Recalling that the Laplacian has non-positive definite eigenvalues we can denote an orthonormal set by $`\varphi _m(x)`$ with
$$\mathrm{}\varphi _m(x)=m^2\varphi _m(x).$$
(93)
The Green’s function is then
$$G(x,x^{})=\underset{m0}{}\frac{\overline{\varphi }_m(x)\varphi _m(x^{})}{m^2}.$$
(94)
All the zero modes of the Laplacian must be omitted in the sum on the r.h.s. This is a generic feature of all Green’s functions. When one writes
$$\mathrm{}G(x,x^{})=11=\text{}\delta ^D(xx^{})\text{},$$
(95)
it is implicitly understood that the $`\delta `$-function or identity-operator is only defined in the space of functions orthogonal to the zero-modes of the Laplacian. This is not necessarily the same as the Dirac-delta function. On a circle of radius $`R`$ for instance, there is exactly one non-trivial zero-mode, the constant function. The Green’s function equals
$$G(x,x^{})=\frac{1}{2\pi R}\underset{n0}{}\frac{e^{\frac{in(xx^{})}{R}}}{n^2/R^2}$$
(96)
and therefore obeys<sup>4</sup><sup>4</sup>4Recall that we use the convention that the Dirac-delta function in curved space obeys $`𝑑x\sqrt{g(x)}f(x)\delta (xx_i)=f(x_i)`$.
$$\mathrm{}G(x,x^{})=_x^2G(x,x^{})=\frac{1}{2\pi R}\underset{n0}{}e^{\frac{in(xx^{})}{R}}=\delta (xx^{})\frac{1}{2\pi R},$$
(97)
rather than
$$\mathrm{}G(x,x^{})=\delta (xx^{}).$$
(98)
The former combination $`\delta (xx^{})\frac{1}{2\pi R}`$ is obviously the delta-function in the space of functions orthogonal to the zero-mode. In the infinite volume limit we are left with the conventional Dirac-delta function.
The above is a reflection of the fact that for differential forms, the Green’s function can be used as a projector onto the space of non-harmonic forms, see e.g. . An arbitrary $`p`$-form $`\omega `$ can be decomposed into an harmonic $`p`$-form $`\gamma `$ and a non-harmonic part as
$$\omega =\gamma +\mathrm{}G\omega .$$
(99)
The combination $`\mathrm{}G`$ is equivalent to the projector
$$\mathrm{}G=11\underset{n}{}\frac{\gamma _n}{\left(\gamma _n\gamma _n\right)}\gamma _n.$$
(100)
Here the $`\gamma _n`$ form a basis of harmonic $`p`$-forms. For scalars on a compact manifold there is only one harmonic form, the constant, and for 0-forms the above combination reduces to
$$\mathrm{}G(x,x^{})=\delta ^D(xx^{})\frac{1}{\text{Vol}}.$$
(101)
For one dimension this equals the r.h.s. of (97)
Particularly in the case of tori, one occasionally imposes the boundary conditions signifying the compactness of the space by introducing image charges on the covering space. The torus is considered as the quotient $`^d/\mathrm{\Lambda }_d`$ of $`^d`$ by the lattice $`\mathrm{\Lambda }_d`$. The Green’s function on $`^d/\mathrm{\Lambda }_d`$ is then a sum of regular $`^d`$ Green’s functions obeying
$$\mathrm{}G_^d(x,x^{})=\delta _{Dirac}^D(xx^{})$$
(102)
such that $`G_{^d/\mathrm{\Lambda }_d}`$ and its derivative is periodic, i.e.
$$G_{^d/\mathrm{\Lambda }_d}(x,x^{})=\underset{\stackrel{}{n}}{}G_^d(x,x^{}+\stackrel{}{n}\stackrel{}{e}).$$
(103)
Here $`\stackrel{}{e}`$ are the lattice vectors generating $`\mathrm{\Lambda }_d`$.
The Laplacian acting on $`G_{^d/\mathrm{\Lambda }_d}(x,x^{})`$ now gives
$$\mathrm{}G_{^d/\mathrm{\Lambda }_d}(x,x^{})=\underset{\stackrel{}{n}}{}\delta _{Dirac}^D(xx^{}+\stackrel{}{n}\stackrel{}{e}).$$
(104)
The only relevant part of the r.h.s, however, is the term with $`n=0`$ as both $`x,x^{}`$ belong to a single fundamental region $`x^iϵ[0,e^i)`$. The delta functions for other values of $`\stackrel{}{n}`$ never contribute.
If one chooses to solve for the warp factor using such a Green’s function, one would ignore the contribution from any constant terms on the r.h.s. This is evident from the preceding discussion. However, a consistency condition is required, namely, the total integral on the r.h.s. of the warp factor equation of motion must vanish. This just says that the total flux on the compact space is zero or that the anomaly cancels.
### B. Heterotic-M theory on $`CY_3`$
The effective five-dimensional theory by compactifying Heterotic-M theory on a Calabi-Yau 3-fold was derived in . Here, we will follow their discussion and notation. We will consider the simplest case, in which there are no 5-branes, and the low energy degrees of freedom include only gravity and the hypermultiplet containing the breathing mode $`V`$ of the Calabi-Yau (the “universal” solution in ). With the standard embedding of the spin connection in one of the $`E_8`$’s, the effective action becomes
$`2\kappa _5^2S_5`$ $`=`$ $`{\displaystyle _{M_5}}\sqrt{g}\left[R+{\displaystyle \frac{1}{2}}V^2g^{55}V^2+{\displaystyle \frac{1}{3}}V^2\alpha ^2\right]`$ (105)
$`+2\sqrt{2}{\displaystyle _{M_4^{(1)}}}\sqrt{g}V^1\alpha 2\sqrt{2}{\displaystyle _{M_4^{(2)}}}\sqrt{g}V^1\alpha ,`$
where $`\kappa _5`$ is the five-dimensional gravitational coupling. The constant $`\alpha `$ is given by
$$\alpha =\frac{1}{8\sqrt{2}\pi v}\left(\frac{\kappa }{4\pi }\right)^{2/3}_{CY_3}\omega \text{tr}RR,v=_{CY_3}\sqrt{g_{CY_3}},$$
(106)
where $`\omega `$ is the Kahler form on the $`CY_3`$. The metric is warped to
$$ds^2=a^2(y)\eta _{\mu \nu }dx^\mu dx^\nu +b^2(y)dy^2.$$
(107)
The solution to the field equations is:
$`a`$ $`=`$ $`a_0H^{1/2},`$
$`b`$ $`=`$ $`b_0H^2,H={\displaystyle \frac{\sqrt{2}}{3}}\alpha |y|+c_0c|y|+c_0,`$ (108)
$`V`$ $`=`$ $`b_0H^3,`$
where $`a_0`$, $`b_0`$ and $`c_0`$ are integration constants.
The Einstein equation linearized in the fluctuations give:
$$\frac{1}{2}h_{\mu \nu }^{\prime \prime }\frac{1}{2}\left(4\frac{a^{}}{a}\frac{b^{}}{b}\right)h_{\mu \nu }^{}\left(3cb_0H3\frac{c}{H}\right)\left(\delta (y)\delta (y\pi \rho )\right)h_{\mu \nu }=\frac{1}{2}(\frac{b_0}{a_0})^2H^3h_{\mu \nu ,\lambda \lambda }.$$
(109)
Let us first consider the equation in the bulk. It is easy to see that the above equation can be written as the covariant wave equation (39). As in Section 4, we expand $`h_{\mu \nu }(x,y)=\widehat{h}_{\mu \nu }(x)\psi (y)`$ with $`\mathrm{}_x\widehat{h}_{\mu \nu }(x)=m^2\widehat{h}_{\mu \nu }(x)`$. The properly normalized wavefunction is
$$\mathrm{\Psi }(y)=[e^{2A(y)}\sqrt{g(y)}]^{1/2}\psi (y)=ab^{1/2}\psi (y).$$
(110)
For massless graviton,
$$\mathrm{\Psi }(y)ab^{1/2}H^{3/2}.$$
(111)
Therefore, the wavefunction $`\mathrm{\Psi }(y)`$ vanishes at the singularity where $`H(y)`$ is zero.
The compactification scale of the M theory direction is usually taken to be slightly larger than that of the Calabi-Yau (from gauge and gravitational unification ). Therefore, there is a regime in which the theory is five-dimensional, and the wavefunction of some of the low-lying massive KK gravitons can still be described by the above wave equation
$$\frac{1}{2}\psi ^{^{\prime \prime }}(y)\left(3cb_0H3\frac{c}{H}\right)\left(\delta (y)\delta (y\pi \rho )\right)\psi (y)=\frac{1}{2}m^2(\frac{b_0}{a_0})^2H^3\psi (y).$$
(112)
Let us rewrite the above equation such that $`m^2`$ becomes the eigenvalue of a Schrodinger equation. Define a new variable $`u=(2/5)c^1(b_0/a_0)H^{5/2}`$ and hence $`du=(b_0/a_0)H^{3/2}dy`$. In terms of this variable, the linearized equation becomes
$$\frac{1}{2}\ddot{\psi }\frac{3c}{4}(\frac{a_0}{b_0})H^{5/2}\dot{\psi }3a_0c\left(\frac{1}{H^{1/2}}\frac{1}{b_0H^{5/2}}\right)\left(\delta (uu_1)\delta (uu_2)\right)\psi =\frac{1}{2}m^2\psi ,$$
(113)
where the dot denotes the derivative with respect to $`u`$. The locations $`u_1`$ and $`u_2`$ are defined by $`u=u_i`$ when $`y=0`$ and $`\pi \rho `$ respectively.
Finally, to eliminate the first derivative term in the above equation. Define $`\psi =H^s\widehat{\psi }`$. It is easy to see that first order derivatives of $`\widehat{\psi }`$ do not appear if $`s=3/4`$. The function $`\widehat{\psi }`$ satisfies the following equation
$$\frac{1}{2}\frac{d^2}{du^2}\widehat{\psi }(u)+𝒱(u)\widehat{\psi }(u)=\frac{1}{2}m^2\widehat{\psi }(u),$$
(114)
where
$`𝒱(u)`$ $`=`$ $`{\displaystyle \frac{21}{32}}\left({\displaystyle \frac{a_0}{b_0}}\right)^2{\displaystyle \frac{c^2}{H^5}}\left({\displaystyle \frac{a_0}{b_0}}\right){\displaystyle \frac{c}{H^{5/2}}}\left(3b_0H^2{\displaystyle \frac{15}{4}}\right)\left(\delta (uu_1)\delta (uu_2)\right)`$
$`=`$ $`{\displaystyle \frac{21}{200u^2}}{\displaystyle \frac{2}{5u}}\left[3b_0\left({\displaystyle \frac{5c}{2}}{\displaystyle \frac{a_0}{b_0}}u\right)^{4/5}{\displaystyle \frac{15}{4}}\right]\left(\delta (uu_1)\delta (uu_2)\right).`$
The potential is not of the “volcano” type . In particular, we have seen that gravity is not localized.
We have seen that the wavefunction of the massless graviton can be deduced without having to solve (114) since it is clear from (41) that $`\psi =`$ constant is a solution. The massive modes can be found by solving (41), or equivalently (114). Let us focus on the latter, as it allows us to compare our solution with that of . Let $`\widehat{\psi }=u^{1/2}f`$, the Schrodinger equation can be written as
$$v^2\frac{^2f}{v^2}+v\frac{f}{v}+(v^2\frac{1}{25})f(v)=0,$$
(115)
where $`v=m(u+u_0)`$. The solution is simply
$$f=k_1J_{1/5}\left(m(u+u_0)\right)+k_2J_{1/5}\left(m(u+u_0)\right),$$
(116)
where $`k_1`$, $`k_2`$ are constants which can be determined by the matching conditions at the delta function sources. The boundary conditions at the two delta sources also imply that the masses $`m`$ of the KK modes are quantized in units of $`1/\rho `$.
|
warning/0004/nlin0004040.html
|
ar5iv
|
text
|
# References
Loop-Algebra and Virasoro Symmetries of Integrable Hierarchies of KP Type
H. Aratyn<sup>1</sup>, J.F. Gomes<sup>2</sup>, E. Nissimov<sup>3</sup> and S. Pacheva<sup>3</sup>
## Abstract
We propose a systematic treatment of symmetries of KP integrable systems, including constrained (reduced) KP models $`\mathrm{𝑐𝐾𝑃}_{R,M}`$ (generalized AKNS hierarchies), and their multi-component (matrix) generalizations. Any $`\mathrm{𝑐𝐾𝑃}_{R,M}`$ integrable hierarchy is shown to possess $`\left(\widehat{U}(1)\widehat{SL}(M)\right)_+\left(\widehat{SL}(M+R)\right)_{}`$ loop-algebra (additional) symmetry. Also we provide a systematic construction of the full algebra of Virasoro additional symmetries in the case of constrained KP models which requires a nontrivial modification of the known Orlov-Schulman construction for the general unconstrained KP hierarchy. Multi-component KP hierarchies are identified as ordinary (scalar) one-component KP hierarchies supplemented with the Cartan subalgebra of the additional symmetry algebra, which provides the basis of a new method for construction of soliton-like solutions. Davey-Stewartson and $`N`$-wave resonant systems arise as symmetry flows of ordinary $`\mathrm{𝑐𝐾𝑃}_{R,M}`$ hierarchies.
1. Introduction
The Kadomtsev-Petviashvili (KP) hierarchy of integrable soliton evolution equations, together with its reductions and multi-component (matrix) generalizations, describe a variety of physically important nonlinear phenomena (for a review, see e.g. ). Constrained (reduced) KP models are intimately connected with the matrix models in non-perturbative string theory of elementary particles at ultra-high energies ( and references therein). They provide an unified description of a number of basic soliton equations such as Korteveg-de-Vries, nonlinear Schrödinger (AKNS hierarchy in general), Yajima-Oikawa, coupled Boussinesq-type equations etc.. Recently it was found that dispersionless KP models play a fundamental role in the description of interface dynamics (Laplacian growth problem). Multi-component (matrix) KP hierarchies, in turn, contain such physically interesting systems as 2-dimensional Toda lattice, Davey-Stewartson, $`N`$-wave resonant system etc.. Recently it has been shown that multi-component KP tau-functions provide solutions to the basic Witten-Dijkgraaf-Verlinde-Verlinde equations in topological field theory.
In the present paper we propose a systematic approach, within Sato pseudo-differential operator framework, for treating symmetries of KP integrable systems, including constrained KP models $`\mathrm{𝑐𝐾𝑃}_{R,M}`$ (generalized AKNS hierarchies – see Eq.(20) below), and their multi-component generalizations. Any $`\mathrm{𝑐𝐾𝑃}_{R,M}`$ hierarchy is shown to possess $`\left(\widehat{U}(1)\widehat{SL}(M)\right)_+\left(\widehat{SL}(M+R)\right)_{}`$ loop-algebra (additional) symmetry generated by squared eigenfunction potentials. The latter subscripts $`(\pm )`$ indicate taking the positive/negative-grade part of the corresponding loop algebra. The symmetry flows generating the above two mutually commuting loop-algebra factors will be called ”positive”/”negative” for brevity.
Furthermore, we provide a systematic construction of the full algebra of Virasoro additional symmetries in the case of constrained KP models which requires a nontrivial modification of the known Orlov-Schulman Virasoro construction for the general unconstrained KP hierarchy.
Multi-component (matrix) KP hierarchies are identified as ordinary (scalar) one-component KP hierarchies supplemented with a special set of commuting additional symmetries, namely, the Cartan subalgebra of the underlying loop algebra. This identification leads to new systematic methods of constructing soliton-like solutions of multi-component KP hierachies by employing the well-established techniques of Darboux-Bäcklund transformations in ordinary one-component KP hierarchies. In particular, Davey-Stewartson and $`N`$-wave resonant systems arise as symmetry flows of ordinary $`\mathrm{𝑐𝐾𝑃}_{R,M}`$ hierarchies.
2. Sato Formalism for Additional Symmetries of Integrable Hierarchies
The general one-component (scalar) KP hierarchy is given by a pseudo-differential<sup>1</sup><sup>1</sup>1Department of Physics, University of Illinois at Chicago, 845 W. Taylor St., Chicago, IL 60607-7059, U.S.A.; e-mail: aratyn@uic.edu<sup>1</sup><sup>1</sup>footnotetext: In what follows the operator $`D`$ is such that $`[D,f]=f=f/x`$ and the generalized Leibniz rule holds: $`D^nf=_{j=0}^{\mathrm{}}\left(\genfrac{}{}{0pt}{}{n}{j}\right)(^jf)D^{nj}`$ with $`n\text{}`$. In order to avoid confusion we shall employ the following notations: for any (pseudo-)differential operator $`A=_ka_kD^k`$ and a function $`f`$, the symbol $`A(f)`$ will indicate application (action) of $`A`$ on $`f`$, whereas the symbol $`Af`$ will denote simply operator product of $`A`$ with the zero-order (multiplication) operator $`f`$. Projections $`(\pm )`$ are defined as: $`A_+=_{k0}a_kD^k`$ and $`A_{}=_{k1}a_kD^k`$. Finally, $`\mathrm{Res}Aa_1`$. Lax operator $``$ obeying Sato evolution equations (also known as isospectral flow equations; for a systematic exposition, see ) :
$$=D+\underset{k=1}{\overset{\mathrm{}}{}}u_kD^k,\frac{}{t_n}=[\left(^n\right)_+,]$$
(1)
with Sato dressing operator $`W`$ :
$$=WDW^1,\frac{}{t_n}W=\left(WD^nW^1\right)_{}W,W=\underset{k=0}{\overset{\mathrm{}}{}}\frac{p_k([])\tau (t)}{\tau (t)}D^k$$
(2)
and (adjoint) Baker-Akhiezer (BA) wave functions $`\psi _{BA}^{()}(t,\lambda )`$ :
$$^{()}\psi _{BA}^{()}=\lambda \psi _{BA}^{()},\frac{}{t_n}\psi _{BA}^{()}=\pm \left(_{}^{()}{}_{}{}^{n}\right)_+(\psi _{BA}^{()})$$
(3)
$$\psi _{BA}^{()}(t,\lambda )=W^{(1)}\left(e^{\pm \xi (t,\lambda )}\right)=\frac{\tau (t[\lambda ^1])}{\tau (t)}e^{\pm \xi (t,\lambda )},\xi (t,\lambda )\underset{\mathrm{}=1}{\overset{\mathrm{}}{}}t_{\mathrm{}}\lambda ^{\mathrm{}}$$
(4)
where the tau-function $`\tau (t)`$ satisfies the relation:
$$_x\frac{}{t_n}\mathrm{ln}\tau =\mathrm{Res}^n$$
(5)
Here and below we employ the following short-hand notations: $`(t)(t_1x,t_2,\mathrm{})`$ for the set of isospectral time-evolution parameters; $`[](\frac{}{t_1},\frac{1}{2}\frac{}{t_2},\frac{1}{3}\frac{}{t_3},\mathrm{})`$ and $`[\lambda ^1](\lambda ^1,\frac{1}{2}\lambda ^2,\frac{1}{3}\lambda ^3,\mathrm{})`$; $`p_k(.)`$ indicate the well-known Schur polynomials.
There exist few other objects in Sato formalism for integrable hierarchies which play fundamental role in our construction. (Adjoint) eigenfunctions $`\mathrm{\Phi }(t)`$ ($`\mathrm{\Psi }(t)`$, respectively) are those functions of KP “times” $`(t)`$ satisfying:
$$\frac{}{t_l}\mathrm{\Phi }=(^l)_+(\mathrm{\Phi }),\frac{}{t_l}\mathrm{\Psi }=(^l)_+^{}(\mathrm{\Psi })$$
(6)
According to second Eq.(3), (adjoint) BA functions are special cases of (adjoint) eigenfunctions, which in addition satisfy spectral equations (first Eq.(3)).
It has been shown in ref. that any (adjoint) eigenfunction possesses a spectral representation of the form<sup>2</sup><sup>2</sup>2Instituto de Física Teórica – IFT/UNESP, Rua Pamplona 145, 01405-900, São Paulo - SP, Brazil; e-mail: jfg@ift.unesp.br<sup>2</sup><sup>2</sup>footnotetext: Integrals over spectral parameters are understood as: $`𝑑\lambda _0\frac{d\lambda }{2i\pi }=\mathrm{Res}_{\lambda =0}`$. :
$$\mathrm{\Phi }(t)=d\lambda \phi (\lambda )\psi _{BA}(t,\lambda ),\mathrm{\Psi }(t)=d\lambda \psi (\lambda )\psi _{BA}^{}(t,\lambda )$$
(7)
with appropriate spectral densities $`\phi (\lambda )`$ and $`\psi (\lambda )`$ which are formal Laurent series in $`\lambda `$. Clearly, any KP hierarchy possesses an infinite set of independent (adjoint) eigenfunctions in one-to-one correspondence with the space of all independent formal Laurent series in $`\lambda `$.
The next important object is the so called squared eigenfunction potential (SEP) – a function $`S(\mathrm{\Phi }(t),\mathrm{\Psi }(t))`$ associated with an arbitrary pair of (adjoint) eigenfunctions $`\mathrm{\Phi }(t),\mathrm{\Psi }(t)`$ which possesses the following characteristics:
$$\frac{}{t_n}S(\mathrm{\Phi }(t),\mathrm{\Psi }(t))=\mathrm{Res}\left(D^1\mathrm{\Psi }(^n)_+\mathrm{\Phi }D^1\right)$$
(8)
In particular, for $`n=1`$ Eq.(8) implies $`_xS(\mathrm{\Phi }(t),\mathrm{\Psi }(t))=\mathrm{\Phi }(t)\mathrm{\Psi }(t)`$ (recall $`_x/t_1`$). Eq.(8) determines $`S(\mathrm{\Phi }(t),\mathrm{\Psi }(t))^1\left(\mathrm{\Phi }(t)\mathrm{\Psi }(t)\right)`$ up to a shift by a trivial constant which is uniquely fixed by the fact that any SEP obeys the following double-spectral representation :
$`^1\left(\mathrm{\Phi }(t)\mathrm{\Psi }(t)\right)={\displaystyle 𝑑\lambda 𝑑\mu \psi (\lambda )\phi (\mu )\frac{1}{\lambda }\psi _{BA}^{}(t,\lambda )\psi _{BA}(t+[\lambda ^1],\mu )}`$
$`={\displaystyle 𝑑\lambda 𝑑\mu \frac{\psi (\lambda )\phi (\mu )}{\lambda \mu }e^{\xi (t,\mu )\xi (t,\lambda )}\frac{\tau (t+[\lambda ^1][\mu ^1])}{\tau (t)}}`$ (9)
with $`\phi (\lambda ),\psi (\lambda )`$ being the respective spectral densities in (7). It is in this well-defined sense that inverse space derivatives $`^1`$ will appear throughout our construction below.
A flow on the space of Sato pseudo-differential Lax operators $``$ or, equivalently, on the space of Sato dressing operators $`W`$ is given by:
$$\delta _\alpha =[_\alpha ,],\delta _\alpha W=_\alpha W$$
(10)
where $`_\alpha `$ is a purely pseudo-differential operator. A flow $`\delta _\alpha `$ (10) is a symmetry if and only if it commutes with the isospectral flows $`\frac{}{t_l}`$ :
$$[\delta _\alpha ,\frac{}{t_l}]=0\frac{}{t_l}_\alpha =[(^l)_+,_\alpha ]_{}$$
(11)
The general form of $`_\alpha `$ obeying (11) is provided by ref. :
$$_\alpha =𝑑\lambda 𝑑\mu \rho _\alpha (\lambda ,\mu )\psi _{BA}(t,\mu )D^1\psi _{BA}^{}(t,\lambda )=\underset{I,J𝒜}{}c_{IJ}^{(\alpha )}\mathrm{\Phi }_JD^1\mathrm{\Psi }_I$$
(12)
where $`\rho _\alpha (\lambda ,\mu )`$ is arbitrary (in the case of the general unconstrained KP hierarchy) double Laurent series in $`\lambda `$ and $`\mu `$. In the second equality above the sums run in general over an infinite set $`𝒜`$ of indices, and $`\{\mathrm{\Phi }_I,\mathrm{\Psi }_I\}_{I𝒜}`$ are (adjoint) eigenfunctions of the Lax operator $``$ (6) for $`\mathrm{\Phi }=\mathrm{\Phi }_I`$ and $`\mathrm{\Psi }=\mathrm{\Psi }_I`$. The second equality in (12) arises from the general representation of the “bispectral” density:
$$\rho _\alpha (\lambda ,\mu )=\underset{I,J𝒜}{}c_{IJ}^{(\alpha )}\phi _J(\mu )\psi _I(\lambda )$$
(13)
in terms of basis of Laurent series $`\left\{\phi _I(\mu )\right\}`$ and $`\left\{\psi _I(\lambda )\right\}`$, together with spectral representation theorem for (adjoint) eigenfunctions of Sato Lax operators (cf. Eqs.(7)).
Our main objective below will be to construct explicitly symmetry-flow generating operators (12), such that the corresponding flows both preserve the constrained form of Lax operators definining constrained KP hierarchies, as well as they yield a closed algebra of symmetries. At this point let us recall that the following special form of $`\rho _\alpha (\lambda ,\mu )=\lambda ^k\left(/\lambda \right)^l\delta (\lambda \mu )`$ in (12) yields the well-known Orlov-Schulman $`W_{1+\mathrm{}}`$ additional symmetries in the case of the general unconstrained KP hierarchy. On the other hand, these standard Orlov-Schulman symmetry flows fail to produce symmetries in the case of constrained KP hierarchies since they do not preserve the constrained form of the pertinent Lax operators. The solution to this problem is provided by a non-trivial modification of the former flows (see ref. and Section 6 below).
On any (adjoint) eigenfunction the action of the flow $`\delta _\alpha `$ (10) takes the form:
$$\delta _\alpha \mathrm{\Phi }=_\alpha (\mathrm{\Phi })+^{(\alpha )},\delta _\alpha \mathrm{\Psi }=_\alpha ^{}(\mathrm{\Psi })+𝒢^{(\alpha )}$$
(14)
where $`^{(\alpha )}`$ and $`𝒢^{(\alpha )}`$ are other (adjoint) eigenfunctions. Eqs.(14) follow from the second Eq.(10) taking again into account the spectral representation theorem . Note that the emergence of the additional (adjoint) eigenfunctions terms on the r.h.s. of (14) is due to the fact that the spectral densities of $`\mathrm{\Phi }`$ and $`\mathrm{\Psi }`$ in (7) may in general vary under the action of $`\delta _\alpha `$. Moreover, as it will be seen in Section 4 below, in the case of constrained KP hierarchies the presence of the additional terms on the r.h.s. of Eqs.(14) is mandatory for consistency of the flow action (10) with the constrained form of the pertinent Lax operator, which accordingly uniquely fixes the form of $`^{(\alpha )}`$ and $`𝒢^{(\alpha )}`$.
Making use of the well-known pseudo-differential operator identities (cf. e.g. the appendix in first ref.) :
$`M_1M_2=M_1(f_2)D^1g_2+f_1D^1M_2^{}(g_1)`$ (15)
$`M_{1,2}f_{1,2}D^1g_{1,2},M_1(f_2)=f_1^1\left(g_1f_2\right)etc.`$
one easily finds that the flows $`\delta _a`$ (10) span (an infinite-dimensional, in general) closed algebra:
$$[\delta _\alpha ,\delta _\beta ]=[\delta _\alpha _\beta \delta _\beta _\alpha [_\alpha ,_\beta ],]$$
(16)
where:
$`\delta _\alpha _\beta \delta _\beta _\alpha [_\alpha ,_\beta ]_{[\alpha ,\beta ]}=`$
$`{\displaystyle \underset{I,J𝒜}{}}\left(c^{(\beta )}a^{(\alpha )}c^{(\alpha )}a^{(\beta )}+b^{(\alpha )}c^{(\beta )}b^{(\beta )}c^{(\alpha )}\right)_{IJ}\mathrm{\Phi }_JD^1\mathrm{\Psi }_I`$ (17)
Here the matrices $`a_{IJ}^{(\alpha )}`$ and $`b_{IJ}^{(\alpha )}`$ appear in the inhomogeneous terms in the $`\delta _\alpha `$-flow equations for $`\mathrm{\Phi }_I`$ and $`\mathrm{\Psi }_I`$, respectively, according to Eqs.(14) :
$$\delta _\alpha \mathrm{\Phi }_I=_\alpha (\mathrm{\Phi }_I)+\underset{J}{}a_{IJ}^{(\alpha )}\mathrm{\Phi }_J,\delta _\alpha \mathrm{\Psi }_I=_\alpha ^{}(\mathrm{\Psi }_I)+\underset{J}{}b_{JI}^{(\alpha )}\mathrm{\Psi }_J$$
(18)
and similarly for $`a_{IJ}^{(\beta )}`$ and $`b_{IJ}^{(\beta )}`$. In the case of general KP hierarchy the specific form of $`a_{IJ}^{(\alpha )}`$ and $`b_{IJ}^{(\alpha )}`$ for any flow (with label $`\alpha `$) is arbitrary a priori, and it is subject to the only condition of fulfillment of Jacobi identities for the flow commutator (17). However, for constrained KP hierarchies the form of $`a_{IJ}^{(\alpha )}`$ and $`b_{IJ}^{(\alpha )}`$ is determined uniquely from the consistency (Eq.(23) below) of the flow action with the constrained form of the pertinent Lax operator, see Section 4 below.
Finally, starting from relation (5) and using (11) we find for the transformation of the tau-function under the action of $`\delta _\alpha `$-flow (10) :
$$\delta _\alpha \mathrm{ln}\tau =^1\left(\mathrm{Res}_\alpha \right)=\underset{I,J}{}c_{IJ}^{(\alpha )}^1\left(\mathrm{\Phi }_J\mathrm{\Psi }_I\right)$$
(19)
3. Constrained KP Hierarchies. Inverse Powers of Lax Operators
So far we have considered the general case of unconstrained KP hierarchy. Now we are interested in symmetries for constrained KP hierarchies $`\mathrm{𝖼𝖪𝖯}_{R,M}`$ with Lax operators (cf. and references therein) :
$$_{R,M}=D^R+\underset{i=0}{\overset{R2}{}}v_iD^i+\underset{j=1}{\overset{M}{}}\mathrm{\Phi }_jD^1\mathrm{\Psi }_j=L_{M+R}L_M^1$$
(20)
where $`\{\mathrm{\Phi }_i,\mathrm{\Psi }_i\}_{i=1}^M`$ is a set of (adjoint) eigenfunctions of $``$.
The second representation of $`_{R,M}`$<sup>3</sup><sup>3</sup>3Institute of Nuclear Research and Nuclear Energy, Boul. Tsarigradsko Chausee 72, BG-1784 Sofia, Bulgaria; e-mail: nissimov@inrne.bas.bg , svetlana@inrne.bas.bg<sup>3</sup><sup>3</sup>footnotetext: Henceforth we shall employ the short-hand notation $``$ for $`_{R,M}`$ (20) whenever this will not lead to a confusion. is in terms of a ratio of two monic purely differential operators $`L_{M+R}`$ and $`L_M`$ of orders $`M+R`$ and $`M`$, respectively (see and references therein). For $`_{R,M}`$ the Sato evolution (isospectral flow) Eqs.(1), the equations for (adjoint) BA (3) and (adjoint) eigenfunctions (6) acquire the form:
$$\frac{}{t_n}=[\left(^{\frac{n}{R}}\right)_+,],^{()}\psi _{BA}^{()}=\lambda ^R\psi _{BA}^{()},\frac{}{t_n}\psi _{BA}^{()}=\pm (^{()})_+^{\frac{n}{R}}(\psi _{BA}^{()})$$
(21)
$$\frac{}{t_n}\mathrm{\Phi }=(^{\frac{n}{R}})_+(\mathrm{\Phi }),\frac{}{t_n}\mathrm{\Psi }=(^{\frac{n}{R}})_+^{}(\mathrm{\Psi })$$
(22)
In the case of constrained hierarchies (20), we have the following additional condition on the symmetry generating operator $`_\alpha `$ since the flow (10) must preserve the constrained form (20) of the pertinent Lax operator (cf.(14) and (12)) :
$$\underset{i=1}{\overset{M}{}}\left[_i^{(\alpha )}D^1\mathrm{\Psi }_i+\mathrm{\Phi }_iD^1𝒢_i^{(\alpha )}\right]=\underset{I,J\{\alpha \}}{}c_{IJ}\left[\mathrm{\Phi }_J^{(\alpha )}D^1^{}(\mathrm{\Psi }_I^{(\alpha )})(\mathrm{\Phi }_J^{(\alpha )})D^1\mathrm{\Psi }_I^{(\alpha )}\right]$$
(23)
where (cf. (14)) :
$$\delta _\alpha \mathrm{\Phi }_i=_\alpha (\mathrm{\Phi }_i)+_i^{(\alpha )},\delta _\alpha \mathrm{\Psi }_i=_\alpha ^{}(\mathrm{\Psi }_i)+𝒢_i^{(\alpha )}$$
(24)
Eq.(23) uniquely fixes the form of the additional terms $`_i^{(\alpha )}`$ and $`𝒢_i^{(\alpha )}`$ in (24).
In what follows we will also need the $`\delta _\alpha `$-flow equations on inverse powers of the Lax operator $`=L_{M+R}L_M^1`$ (20). First, let us recall that the inverses of the underlying purely differential operators are given by:
$$L_M^1=\underset{i=1}{\overset{M}{}}\phi _iD^1\psi _i,L_{M+R}^1=\underset{a=1}{\overset{M+R}{}}\overline{\phi }_aD^1\overline{\psi }_a$$
(25)
where the functions $`\left\{\phi _i\right\}_{i=1}^M`$ and $`\left\{\psi _i\right\}_{i=1}^M`$ span $`Ker(L_M)`$ and $`Ker(L_M^{})`$, respectively, whereas $`\left\{\overline{\phi }_a\right\}_{a=1}^{M+R}`$ and $`\left\{\overline{\psi }_a\right\}_{a=1}^{M+R}`$ span $`Ker(L_{M+R})`$ and $`Ker(L_{M+R}^{})`$, respectively. Therefore we have:
$$=()_++\underset{i=1}{\overset{M}{}}L_{M+R}(\phi _i)D^1\psi _i,i.e.\mathrm{\Phi }_i=L_{M+R}(\phi _i),\mathrm{\Psi }_i=\psi _i$$
(26)
$$^1=\underset{a=1}{\overset{M+R}{}}L_M(\overline{\phi }_a)D^1\overline{\psi }_a$$
(27)
$$^N=\underset{a=1}{\overset{M+R}{}}\underset{s=0}{\overset{N1}{}}^{(N1)+s}\left(L_M(\overline{\phi }_a)\right)D^1\left(^s\right)^{}(\overline{\psi }_a)$$
(28)
Compare the last formula (28) with the formula for the negative pseudo-differential part of a positive power of $``$ (20):
$$\left(^N\right)_{}=\underset{i=1}{\overset{M}{}}\underset{s=0}{\overset{N1}{}}^{N1s}(\mathrm{\Phi }_i)D^1\left(^s\right)^{}(\mathrm{\Psi }_i)$$
(29)
Let us also note that the following simple consequences from the definitions of the corresponding objects will play essential role for the consistency of the constructions involving inverse powers of $``$ :
$$\left(L_M(\overline{\phi }_a)\right)=0,^{}(\overline{\psi }_a)=0,^1(\mathrm{\Phi }_i)=0,\left(^1\right)^{}(\mathrm{\Psi }_i)=0$$
(30)
Applying the flow Eq.(10) to $`^1`$ (27) $`\delta _\alpha ^1=[_\alpha ,^1]`$ and taking into account the explicit form of $`_\alpha `$ (second equality (12)) we obtain:
$$\delta _\alpha \left(L_M(\overline{\phi }_a)\right)=_\alpha \left(L_M(\overline{\phi }_a)\right)+\overline{}_a^{(\alpha )},\delta _\alpha \overline{\psi }_a=_\alpha ^{}(\overline{\psi }_a)+\overline{𝒢}_a^{(\alpha )}$$
(31)
with consistency condition for the “shift” functions $`\overline{}_a^{(\alpha )}`$ and $`\overline{𝒢}_a^{(\alpha )}`$ (the analog of Eq.(24)) :
$$\underset{a=1}{\overset{M+R}{}}\left[\overline{}_a^{(\alpha )}D^1\overline{\psi }_a+L_M(\overline{\phi }_a)D^1\overline{𝒢}_a^{(\alpha )}\right]=\underset{I,J𝒜}{}c_{IJ}^{(\alpha )}\left[\mathrm{\Phi }_JD^1\left(^1\right)^{}(\mathrm{\Psi }_I)^1(\mathrm{\Phi }_J)D^1\mathrm{\Psi }_I\right]$$
(32)
Also, from the isospectral flows equations applied on $`^1`$, i.e., $`\frac{}{t_n}^1=[_+^{\frac{n}{R}},^1]`$, we find, taking into account (30), that $`L_M(\overline{\phi }_a)`$ and $`\overline{\psi }_a`$ are (adjoint) eigenfunctions of $``$ (cf. (22)) :
$$\frac{}{t_n}L_M(\overline{\phi }_a)=(^{\frac{n}{R}})_+\left(L_M(\overline{\phi }_a)\right),\frac{}{t_n}\overline{\psi }_a=(^{\frac{n}{R}})_+^{}(\overline{\psi }_a)$$
(33)
4. Loop-Algebra Symmetries of KP Hierarchies
Let us consider the following system of $`M`$ infinite sets of (adjoint) eigenfunctions of $`_{R,M}`$ (20) :
$$\mathrm{\Phi }_i^{(n)}^{n1}(\mathrm{\Phi }_i),\mathrm{\Psi }_i^{(n)}\left(^{}\right)^{n1}(\mathrm{\Psi }_i),n=1,2,\mathrm{};i=1,\mathrm{},M$$
(34)
which are expressed in terms of the $`M`$ pairs of (adjoint) eigenfunctions entering the pseudo-differential part of $`_{R,M}`$ (20). Using (34) we can build the following infinite set of symmetry flows (cf. (10) and (12)) :
$$\delta _A^{(n)}=[_A^{(n)},],_A^{(n)}\underset{i,j=1}{\overset{M}{}}A_{ij}^{(n)}\underset{s=1}{\overset{n}{}}\mathrm{\Phi }_j^{(n+1s)}D^1\mathrm{\Psi }_i^{(s)}$$
(35)
where $`A^{(n)}`$ is an arbitrary constant $`M\times M`$ matrix, i.e., $`A^{(n)}Mat(M)`$. Consistency of the flow action (35) with the constrained form (20) of $`_{R,M}`$ (cf. (23)) implies the following flow action on the involved (adjoint) eigenfunctions:
$`\delta _A^{(n)}\mathrm{\Phi }_i^{(m)}=_A^{(n)}(\mathrm{\Phi }_i^{(m)}){\displaystyle \underset{j=1}{\overset{M}{}}}A_{ij}^{(n)}\mathrm{\Phi }_j^{(n+m)}`$
$`\delta _A^{(n)}\mathrm{\Psi }_i^{(m)}=\left(_A^{(n)}\right)^{}(\mathrm{\Psi }_i^{(m)})+{\displaystyle \underset{j=1}{\overset{M}{}}}A_{ji}^{(n)}\mathrm{\Psi }_j^{(n+m)}`$ (36)
The specific form of the inhomogeneous terms on the r.h.s. of Eqs.(36) is the main ingredient of our construction. It is precisely these inhomogeneous terms which yield non-trivial loop-algebra additional symmetries.
Using the pseudo-differential operator identities (15) and taking into account (36) we can show that (cf. Eq.(17)):
$$\delta _A^{(n)}_B^{(m)}\delta _B^{(m)}_A^{(n)}[_A^{(n)},_B^{(m)}]=_{[A,B]}^{(n+m)}$$
(37)
Eq.(37) implies that the symmetry flows (35)–(36) span the following infinite-dimensional algebra:
$$[\delta _A^{(n)},\delta _B^{(m)}]=\delta _{[A,B]}^{(n+m)};A^{(n)},B^{(m)}Mat(M),n,m=1,2,\mathrm{}$$
(38)
isomorphic to $`\left(\widehat{U}(1)\times \widehat{SL}(M)\right)_+`$ where the subscript $`(+)`$ indicates taking the positive-grade subalgebra of the corresponding loop-algebra. We observe, that in the case of $`\mathrm{𝖼𝖪𝖯}_{R,M}`$ models we have $`_{A=\text{1}\text{l}}^{(n)}=\left(_{R,M}^n\right)_{}`$ (insert (34) into first relation (35) for $`A^{(n)}=\text{1}\text{l}`$ and compare with (29)). Therefore, the flows $`\delta _{A=\text{1}\text{l}}^{(n)}`$ for $`\mathrm{𝖼𝖪𝖯}_{R,M}`$ models coincide upto a sign with the ordinary isospectral flows modulo $`R`$: $`\delta _{A=\text{1}\text{l}}^{(n)}=\frac{}{t_{nR}}`$ (cf. Eqs.(21)). Thereby the flows $`\delta _A^{(n)}`$ (35) will be called ”positive” for brevity.
Now we consider another infinite set of (adjoint) eigenfunctions of $`_{R,M}`$ expressed in terms of the (adjoint) eigenfunctions entering the inverse power of $`_{R,M}`$ (27) :
$`\mathrm{\Phi }_a^{(m)}^{(m1)}\left(L_M(\overline{\phi }_a)\right),\mathrm{\Psi }_a^{(m)}\left(^{(m1)}\right)^{}(\overline{\psi }_a)`$ (39)
$`m=1,2,\mathrm{},a=1,\mathrm{},M+R`$ (40)
Using (39) we obtain the following set of ”negative” symmetry flows which parallels completely the set of ”positive” flows (35) :
$$\delta _𝒜^{(n)}=[_𝒜^{(n)},],_𝒜^{(n)}\underset{a,b=1}{\overset{M+R}{}}𝒜_{ab}^{(n)}\underset{s=1}{\overset{n}{}}\mathrm{\Phi }_b^{(n1+s)}D^1\mathrm{\Psi }_a^{(s)}$$
(41)
where $`𝒜_{ab}^{(n)}`$ is an arbitrary constant $`(M+R)\times (M+R)`$ matrix, i.e., $`𝒜^{(n)}Mat(M+R)`$. In fact, since according to (28) we have $`_{𝒜=\text{1}\text{l}}^{(n)}=^n`$, the flows $`\delta _{𝒜=\text{1}\text{l}}^{(n)}`$ vanish identically, i.e., $`\delta _{𝒜=\text{1}\text{l}}^{(n)}=0`$, therefore, we restrict $`𝒜^{(n)}SL(M+R)`$.
Consistency of the flow action (41) with the constrained form (20) of $`_{R,M}`$ (cf. (23)) and with the constrained form (27) of the inverse $`^1`$ implies the following $`\delta _𝒜^{(n)}`$-flow action on the involved (adjoint) eigenfunctions (using short-hand notations (34) and (39)) :
$$\delta _𝒜^{(n)}\mathrm{\Phi }_i^{(m)}=_𝒜^{(n)}(\mathrm{\Phi }_i^{(m)}),\delta _𝒜^{(n)}\mathrm{\Psi }_i^{(m)}=\left(_𝒜^{(n)}\right)^{}(\mathrm{\Psi }_i^{(m)})$$
(42)
$`\delta _𝒜^{(n)}\mathrm{\Phi }_a^{(m)}=_𝒜^{(n)}(\mathrm{\Phi }_a^{(m)}){\displaystyle \underset{b=1}{\overset{M+R}{}}}𝒜_{ab}^{(n)}\mathrm{\Phi }_b^{(nm)}`$
$`\delta _𝒜^{(n)}\mathrm{\Psi }_a^{(m)}=\left(_𝒜^{(n)}\right)^{}(\mathrm{\Psi }_a^{(m)})+{\displaystyle \underset{b=1}{\overset{M+R}{}}}𝒜_{ba}^{(n)}\mathrm{\Psi }_b^{(nm)}`$ (43)
Similarly, consistency of ”positive” $`\delta _A^{(n)}`$-flow action (35) with the constrained form (27) of the inverse Lax operator implies:
$$\delta _A^{(n)}\mathrm{\Phi }_a^{(m)}=_A^{(n)}(\mathrm{\Phi }_a^{(m)}),\delta _A^{(n)}\mathrm{\Psi }_a^{(m)}=\left(_A^{(n)}\right)^{}(\mathrm{\Psi }_a^{(m)})$$
(44)
Using again the pseudo-differential operator identities (15) we find from (42)–(44) (cf. Eq.(37)) :
$$\delta _A^{(n)}_{}^{(m)}\delta _{}^{(m)}_A^{(n)}[_A^{(n)},_{}^{(m)}]=0$$
(45)
$$\delta _𝒜^{(n)}_{}^{(m)}\delta _{}^{(m)}_𝒜^{(n)}[_𝒜^{(n)},_{}^{(m)}]=_{[𝒜,]}^{(nm)}$$
(46)
Eqs.(45)–(46) imply that the ”negative” symmetry flows (41)–(43) commute with the ”positive” flows (35)–(36):
$$[\delta _A^{(n)},\delta _{}^{(m)}]=0$$
(47)
and that they themselves span the following infinite-dimensional algebra:
$$[\delta _𝒜^{(n)},\delta _{}^{(m)}]=\delta _{[𝒜,]}^{(nm)};𝒜^{(n)},^{(m)}SL(M+R),n,m=1,2,\mathrm{}$$
(48)
which is isomorphic to $`\left(\widehat{SL}(M+R)\right)_{}`$ (the subscript $`()`$ indicates taking the negative-grade subalgebra of the corresponding loop-algebra).
Therefore,we conclude that the full loop algebra of (additional) symmetries of $`\mathrm{𝑐𝐾𝑃}_{R,M}`$ hierarchies (20) is the direct sum:
$$\left(\widehat{U}(1)\widehat{SL}(M)\right)_+\left(\widehat{SL}(M+R)\right)_{}$$
(49)
The construction above can be straightforwardly extended to the case of the general unconstrained KP hierarchy defined by (1). All relations (35)–(38) and (41)–(48) remain intact where now:
$$\{\mathrm{\Phi }_i^{(n)},\mathrm{\Psi }_i^{(n)}\}_{i=1,\mathrm{},M}^{n=1,2,\mathrm{}},\{\mathrm{\Phi }_a^{(n)},\mathrm{\Psi }_a^{(n)}\}_{a=1,\mathrm{},M+R}^{n=1,2,\mathrm{}}$$
(50)
form an infinite system of independent (adjoint) eigenfunctions of the general Lax operator (1) with $`M`$, $`M+R`$ being arbitrary positive integers.
5. Multi-Component KP Hierarchies from One-Component Ones
Let us now consider the following subset of ”positive” flows $`\delta _{E_k}^{(n)}`$ (35) for the general KP hierarchy (1) corresponding to:
$$E_k=diag(0,\mathrm{}0,1,0,\mathrm{},0),i.e._{E_k}^{(n)}=\underset{s=1}{\overset{n}{}}\mathrm{\Phi }_k^{(n+1s)}D^1\mathrm{\Psi }_k^{(s)}$$
(51)
Due to Eq.(38) the flows $`\delta _{E_k}^{(n)}`$ span an infinite-dimensional Abelian algebra and, by construction, they commute with the original isospectral flows $`\frac{}{t_n}`$ as well. Comparison with our construction in refs. allows us to identify the set of isospectral flows plus the set of $`\delta _{E_k}^{(n)}`$-flows (51) :
$$\frac{}{t_n}/\stackrel{(1)}{t_n},\delta _{E_k}^{(n)}/\stackrel{(k+1)}{t_n},k=1,\mathrm{},M$$
(52)
with the set of isospectral flows $`\left\{\stackrel{(\mathrm{})}{t_n}\right\}_{n=1,2,\mathrm{}}^{\mathrm{}=1,\mathrm{},M+1}`$ of the (unconstrained) $`M+1`$-component matrix KP hierarchy. The latter is defined in terms of the $`M+1\times M+1`$ matrix Hirota bilinear identities (see refs.) :
$`{\displaystyle \underset{k=1}{\overset{M+1}{}}}\epsilon _{ik}\epsilon _{jk}{\displaystyle }d\lambda \lambda ^{\delta _{ik}+\delta _{jk}2}e^{\xi (\stackrel{\left(k\right)}{t}\stackrel{\left(k\right)}{t^{}},\lambda )}\tau _{ik}(\mathrm{},\stackrel{(k)}{t}[\lambda ^1],\mathrm{})\tau _{kj}(\mathrm{},\stackrel{(k)}{t^{}}+[\lambda ^1],\mathrm{})=0`$ (53)
which are obeyed by a set of $`M(M+1)+1`$ tau-functions $`\left\{\tau _{ij}\right\}`$ expressed in terms of the single tau-function $`\tau `$ and the “positive” symmetry flow generating (adjoint) eigenfunctions (50) in the original one-component (scalar) KP hierarchy (1)–(5) as follows:
$$\tau _{11}=\tau _{ii}=\tau ,\tau _{1i}=\tau \mathrm{\Phi }_{i1}^{(1)},\tau _{i1}=\tau \mathrm{\Psi }_{i1}^{(1)}$$
$$\tau _{ij}=\epsilon _{ij}\tau ^1\left(\mathrm{\Phi }_{j1}^{(1)}\mathrm{\Psi }_{i1}^{(1)}\right),ij,i,j=2,\mathrm{}M+1$$
(54)
Here $`\epsilon _{ij}=1`$ for $`ij`$ and $`\epsilon _{ij}=1`$ for $`i>j`$, and $`\delta _{ij}`$ are the usual Kronecker symbols.
The above construction of multi-component (matrix) KP hierarchies out of ordinary one-component ones can be straighforwardly carried over to the case of constrained KP models (20) using the identification (34) for the symmetry-generating (adjoint) eigenfunctions. In this case, however, there is a linear dependence among the flows (52) $`_{k=1}^M\delta _{E_k}^{(n)}=\frac{}{t_n}`$, therefore, the associated constrained multi-component KP hierarhy is now $`M\times M`$ matrix hierarchy.
Similarly, we can start with the subset of “negative” symmetry flows $`\delta _{E_k}^{(n)}`$ (41) for $`\mathrm{𝑐𝐾𝑃}_{R,M}`$ hierarchy<sup>4</sup><sup>4</sup>4The flow $`\delta _{E_k}^{(n)}`$ for $`k=1`$ is excluded since $`_{k=1}^{M+R}\delta _{E_k}^{(n)}=\delta _{𝒜=\text{1}\text{l}}^{(n)}`$ which vanishes identially as explained in the previous section. :
$$\delta _{E_k}^{(n)}/\stackrel{(k)}{t}{}_{n}{}^{},_{E_k}^{(n)}=\underset{s=1}{\overset{n}{}}\mathrm{\Phi }_k^{(n1+s)}D^1\mathrm{\Psi }_k^{(s)},k=2,\mathrm{},M+R,n=1,2,\mathrm{}$$
(55)
which also span an infinite-dimensional Abelian algebra of flows commuting with the isospectral flows. Then, following the steps of our construction in ref. we arrive at $`(M+R)`$-component constrained KP hierarchy given in terms of $`(M+R)(M+R1)+1`$ tau-functions $`\left\{\stackrel{~}{\tau }_{ab}\right\}`$ obeying the corresponding $`(M+R)\times (M+R)`$ matrix Hirota bilinear identities (cf. (53)). The latter tau-functions are expressed in terms of the original single tau-function $`\tau `$ and the “negative” flow generating (adjoint) eigenfunctions (39) in the original ordinary $`\mathrm{𝑐𝐾𝑃}_{R,M}`$ hierarchy as follows:
$$\stackrel{~}{\tau }_{11}=\stackrel{~}{\tau }_{aa}=\tau ,\stackrel{~}{\tau }_{1a}=\tau L_M(\overline{\phi }_a),\stackrel{~}{\tau }_{a1}=\tau \overline{\psi }_a$$
$$\stackrel{~}{\tau }_{ab}=\epsilon _{ab}\tau ^1\left(L_M(\overline{\phi }_b)\overline{\psi }_a\right),ab,a,b=2,\mathrm{}M+R$$
(56)
Let us recall that multi-component (matrix) KP hierarchies (53) contain various physically interesting nonlinear systems such as Davey-Stewartson and $`N`$-wave systems, which now can be written entirely in terms of objects belonging to ordinary one-component (constrained) KP hierarchy. For instance, the $`N`$-wave resonant system ($`N=M(M+1)/2`$) is given by:
$$_cf_{ij}=f_{ik}f_{kj},ijk,i,j,k=1,\mathrm{},M+1$$
(57)
$$_k/\stackrel{(k)}{t}{}_{1}{}^{},f_{1i}\mathrm{\Phi }_{i1}^{(1)},f_{i1}\mathrm{\Psi }_{i1}^{(1)}$$
$$f_{ij}\epsilon _{ij}^1\left(\mathrm{\Phi }_{j1}^{(1)}\mathrm{\Psi }_{i1}^{(1)}\right),ij,i,j=2,\mathrm{}M+1$$
(58)
As a further example, we will demonstrate that the well-known Davey-Stewartson system arises as particular subset of symmetry flow equations obeyed by any pair of adjoint eigenfunctions $`(\mathrm{\Phi }_i,\mathrm{\Psi }_i)`$ ($`i`$=fixed) or $`(L_M(\overline{\phi }_a),\overline{\psi }_a)`$ ($`a`$=fixed). In fact, for $`(\mathrm{\Phi }_i,\mathrm{\Psi }_i)`$ ($`i`$=fixed) this has already been done in last ref.. Here for simplicity we take $`\mathrm{𝑐𝐾𝑃}_{1,M}`$ hierarchy (Eq.(20) with $`R=1`$; the general case is straightforward generalization of the formulas below) and consider a pair of “negative” symmetry flow generating (adjoint) eigenfunctions, e.g., $`\left(\varphi L_M(\overline{\phi }_a),\psi \overline{\psi }_a\right)`$ ($`a`$=fixed), which obeys the following subset of flow equations – w.r.t. $`/t_2`$, $`\overline{}/\stackrel{(a)}{t}_1`$ and $`/\overline{t}_2/\stackrel{(a)}{t}_2`$ (cf. Eqs.(43)) :
$$\frac{}{t_2}\varphi =(^2+2\underset{i=1}{\overset{M}{}}\mathrm{\Phi }_i\mathrm{\Psi }_i)\varphi ,\frac{}{t_2}\psi =(^2+2\underset{i=1}{\overset{M}{}}\mathrm{\Phi }_i\mathrm{\Psi }_i)\psi $$
(59)
$$\overline{}\varphi =_1^{(1)}(\varphi )^1(\varphi ),\overline{}\psi =\left(_1^{(1)}\right)^{}(\psi )+\left(^1\right)^{}(\psi )$$
(60)
$$/\overline{t}_2\varphi =_1^{(2)}(\varphi )^2(\varphi ),/\overline{t}_2\psi =\left(_1^{(2)}\right)^{}(\psi )+\left(^2\right)^{}(\psi )$$
(61)
where:
$$_1^{(1)}\varphi D^1\psi ,_1^{(2)}^1(\varphi )D^1\psi +\varphi D^1\left(^1\right)^{}(\psi )$$
(62)
Using (60) we can rewrite Eqs.(61) as purely differential equation w.r.t. $`\overline{}`$ :
$$\frac{}{\overline{t}_2}\varphi =[\overline{}^2+2\overline{}\left(^1\left(\varphi \psi \right)\right)]\varphi ,\frac{}{\overline{t}_2}\psi =[\overline{}^22\overline{}\left(^1\left(\varphi \psi \right)\right)]\psi $$
(63)
Now, introducing new time variable $`T=t_2\overline{t}_2`$ and the short-hand notation $`Q_{i=1}^M\mathrm{\Phi }_i\mathrm{\Psi }_i2(\varphi \psi )2\overline{}\left(^1(\varphi \psi )\right)`$, and subtracting Eqs.(63) from Eqs.(59), we arrive at the following system of $`(2+1)`$-dimensional nonlinear evolution equations:
$$\frac{}{T}\varphi =\left[\frac{1}{2}(^2+\overline{}^2)+Q+2\varphi \psi \right]\varphi $$
(64)
$$\frac{}{T}\psi =\left[\frac{1}{2}(^2+\overline{}^2)+Q+2\varphi \psi \right]\psi $$
(65)
$$\overline{}Q+(+\overline{})^2\left(\varphi \psi \right)=0$$
(66)
which is precisely the standard Davey-Stewartson system for the “negative” (adjoint) eigenfunction pair $`\left(\varphi L_M(\overline{\phi }_a),\psi \overline{\psi }_a\right)`$ ($`a`$=fixed).
The construction in this Section allows us to employ the well-known Darboux-Bäcklund techniques from ordinary one-component (scalar) KP hierarchies (full or constrained) in order to obtain new soliton-like solutions of multi-component (matrix) KP hierarchies (see last ref.).
6. The Full Virasoro Algebra of Additional Symmetries
In refs. we have constructed an essential modification to the original Orlov-Schulman additional Virasoro symmetry flows needed in the case of $`\mathrm{𝖼𝖪𝖯}_{R,M}`$ reduced KP models (20) for $`n0`$, i.e., for the Borel subalgebra (henceforth $`_{R,M}`$) :
$$\delta _n^V=[\left(M^n\right)_{}+𝒳_n,]$$
(67)
or, equivalently:
$$\delta _n^V=[\left(M^n\right)_++𝒳_n,]+^n$$
(68)
where $`\delta _n^VL_{n1}`$ (in terms of standard Virasoro notations). Here:
$$[,M]=\text{1}\text{l},M=\underset{k1}{}kt_k^{k1}+\underset{j1}{}(jp_j([])\mathrm{ln}\tau )^{j1}$$
(69)
$$𝒳_n\underset{i=1}{\overset{M}{}}\underset{j=0}{\overset{n2}{}}\left(j\frac{1}{2}(n2)\right)^{n2j}(\mathrm{\Phi }_i)D^1\left(^j\right)^{}(\mathrm{\Psi }_i)$$
(70)
The presence of the additional terms $`𝒳_n`$ in (67) is very crucial to ensure that the flows $`\delta _n^V`$ preserve the constrained form of the pertinent pseudo-differential Lax operator (20). The ordinary Orlov-Schulman flows $`\delta _n^{OS}=[\left(M^n\right)_{},]`$ do not define symmetries for constrained $`\mathrm{𝖼𝖪𝖯}_{R,M}`$ hierarchies.
The action of $`\delta _n^V`$-flows on the pertinent (adjoint) eigenfunctions reads accordingly (for $`n0`$) :
$$\delta _n^V\mathrm{\Phi }_i=\left[\left(M^n\right)_++𝒳_n\right](\mathrm{\Phi }_i)+\frac{n}{2}^{n1}(\mathrm{\Phi }_i)$$
(71)
$$\delta _n^V\mathrm{\Psi }_i=\left[\left(M^n\right)_+^{}+𝒳_n^{}\right](\mathrm{\Psi }_i)+\frac{n}{2}\left(^{n1}\right)^{}(\mathrm{\Psi }_i)$$
(72)
Similarly, for the (adjoint) eigenfunctions entering the inverse Lax powers we find from $`\delta _n^V^1=[\left(M^n\right)_{}+𝒳_n,^1]`$ and Eqs.(30) (for $`n0`$) :
$$\delta _n^VL_M(\overline{\phi }_a)=[\left(M^n\right)_++𝒳_n]\left(L_M(\overline{\phi }_a)\right),\delta _n^V\overline{\psi }_a=[\left(M^n\right)_+^{}+𝒳_n^{}](\overline{\psi }_a)$$
(73)
Here we want to extend the above construction to cover the case of the full Virasoro algebra of additional symmetries. For the negative flows we must therefore find the appropriate additional terms $`𝒳_{(n)}`$ :
$$\delta _n^V=[\left(M^n\right)_{}+𝒳_{(n)},]$$
(74)
or, equivalently:
$$\delta _n^V=[\left(M^n\right)_++𝒳_{(n)},]+^n$$
(75)
so that the consistency condition (23) is satisfied. Using again the pseudo-differential operator identities (15) and taking into account the relevant formulas for negative Lax powers (28) we obtain the following explicit expressions for $`𝒳_{(n)}`$ :
$$𝒳_{(n)}=\underset{a=1}{\overset{M+R}{}}\underset{j=0}{\overset{n}{}}\left(\frac{n}{2}j\right)^{(nj)}\left(L_M(\overline{\phi }_a)\right)D^1\left(^j\right)^{}(\overline{\psi }_a)$$
(76)
The consistency of the negative flow definitions (75) with $`𝒳_{(n)}`$ as in Eq.(76) crucially depends on the relations (30).
The flows $`\delta _n^V`$ act on the constituent (adjoint) eigenfunctions of $``$ as:
$$\delta _n^V\mathrm{\Phi }_i=[\left(M^n\right)_++𝒳_{(n)}](\mathrm{\Phi }_i),\delta _n^V\mathrm{\Psi }_i=[\left(M^n\right)_+^{}+𝒳_{(n)}^{}](\mathrm{\Psi }_i)$$
(77)
and similarly on the (adjoint) eigenfunctions $`L_M(\overline{\phi }_a)`$, $`\overline{\psi }_a`$ entering the inverse powers of $``$ :
$$\delta _n^V\left(L_M(\overline{\phi }_a)\right)=\left[\left(M^n\right)_++𝒳_{(n)}\right]\left(L_M(\overline{\phi }_a)\right)\left(\frac{n}{2}+1\right)^{(n+1)}\left(L_M(\overline{\phi }_a)\right)$$
(78)
$$\delta _n^V\overline{\psi }_a=\left[\left(M^n\right)_+^{}+𝒳_{(n)}^{}\right](\overline{\psi }_a)\left(\frac{n}{2}+1\right)\left(^{(n+1)}\right)^{}(\overline{\psi }_a)$$
(79)
Let us now consider the commutator of the Virasoro flows $`\delta _n^VL_{n1}`$ and $`\delta _m^VL_{m1}`$ acting on $``$ (cf. Eq.(16)) where $`(n,m)`$ are arbitrary non-negative or negative indices:
$$[\delta _n^V,\delta _m^V]=\delta _n^V\left(\left(M^m\right)_{}+𝒳_m\right)\delta _m^V\left(\left(M^n\right)_{}+𝒳_n\right)[\left(M^n\right)_{}+𝒳_n,\left(M^m\right)_{}+𝒳_m]$$
(80)
Using the identity:
$`\delta _n^V\left(M^m\right)_{}\delta _m^V\left(M^n\right)_{}=(nm)\left(M^{n+m1}\right)_{}`$
$`[\left(M^n\right)_{},\left(M^m\right)_{}]+[𝒳_n,M^m]_{}[𝒳_m,M^n]_{}`$ (81)
the r.h.s. of Eq.(80) can be rewritten in the form:
$$(nm)\left(M^{n+m1}\right)_{}+\delta _n^V𝒳_m[\left(M^n\right)_+,𝒳_m]_{}\delta _m^V𝒳_n+[\left(M^m\right)_+,𝒳_n]_{}[𝒳_n,𝒳_m]$$
(82)
Now, employing the pseudo-differential identities (15) it is easy to show, taking into account (71)–(73) and (77)–(79), that the sum of all terms in (82) involving $`𝒳_{n,m}`$ yield:
$$\delta _n^V𝒳_m[\left(M^n\right)_+,𝒳_m]_{}\delta _m^V𝒳_n+[\left(M^m\right)_+,𝒳_n]_{}[𝒳_n,𝒳_m]=(nm)𝒳_{n+m1}$$
(83)
Thus, we have verified the closure of the full Virasoro algebra of additional symmetries without central extension:
$$[\delta _n^V,\delta _m^V]=(nm)\delta _{n+m1}^V$$
(84)
Outlook In a subsequent paper we will generalize the present construction of additional symmetries to the case of supersymmetric integrable hierarchies. We will continue the derivation and study of properties of new soliton-like solutions of multi-component KP hierarchies obtained via standard Darboux-Bäcklund methods for ordinary one-component KP models which has been already initiated in the last ref.. In a forthcoming more detailed paper we will systematically study the construction of additional (loop-algebra and Virasoro) symmetries within a generalized Drinfeld-Sokolov formalism both in ordinary and supersymmetric integrable systems of KP type. Also we will relate the algebraic dressing method to Sato pseudo-differential operator approach.
Acknowledgements H.A. is partially supported by NSF (PHY-9820663) and J.F.G. is partially supported by CNPq and Fapesp (Brazil). Three of us (H.A., E.N. and S.P.) gratefully acknowledge support from NSF grant INT-9724747. E.N. and S.P. are partially supported by Bulgarian NSF grant F-904/99.
|
warning/0004/physics0004052.html
|
ar5iv
|
text
|
# I Introduction
## I Introduction
A problem of long-standing interest is that of setting confidence intervals for an unknown non-negative signal $`\mu `$ in the presence of a known mean background $`b`$ when the measurement $`n`$ is Poisson distributed as $`p(n;\mu +b)`$. When $`n<b`$, the usual estimate for $`\mu `$, i.e. $`nb`$, is negative, leading in most constructions to small upper limits that imply unrealistically high confidence in small values of $`\mu `$. In a recent paper, Roe and Woodroofe propose a construction that produces more believable intervals and contains the unifying feature that one need not decide beforehand whether to set a confidence interval or an upper confidence bound. However, since the Roe-Woodroofe confidence belt (of confidence level $`\alpha `$) is not constructed from an unconditional probability density and does not have coverage in the usual sense (i.e. unconditional coverage), one cannot state that the unconditional probability of the interval enclosing the true value is at least $`\alpha `$. Our comment is that a straightforward modification of the Roe-Woodroofe confidence belt gives it coverage, making the construction effectively an ordering principle applied to the Poisson pdf, albeit reached by circuitous means.
## II Roe-Woodroofe Confidence Intervals
Roe and Woodroofe are motivated by the observation that the measurement $`n=0`$ implies that zero signal (as well as zero background) is seen; thus, the resulting estimate for $`\mu `$ is zero, independent of $`b`$. They argue therefore that the confidence interval for $`\mu `$ for n=0 must be independent of $`b`$. Extending the argument, they note that for any observation $`n`$, one has observed a result $`n`$ from the Poisson pdf $`p(n;\mu +b)`$ and a background of at most $`n`$. They formulate a method of obtaining confidence intervals based on the conditional probability to observe $`n`$ given a background $`n`$ and obtain the desired result for $`n=0`$ and approximately the classical confidence intervals for $`n>b`$. While they identify their method as an ordering principle, it is not one in the same sense as Ref.s and which explicitly choose a confidence belt of probability $`\alpha `$ using the Poisson pdf $`p(n;\mu +b)`$ and the Likelihood Ratio Construction and invert it to find confidence intervals. The latter methods do not obtain intervals that are independent of $`b`$ for $`n=0`$ and yield confidence intervals which are unphysically small for $`n<b`$.
Although the Roe-Woodroofe construction does not have coverage in the usual sense , it can be easily modified to obtain coverage, by retaining the left-hand boundary of the confidence belt and adjusting the right-hand boundary so that for all $`\mu `$ the horizontal intervals contain probability $`\alpha `$. In Fig. 1 we show the Roe-Woodroofe 90% intervals for $`b=3`$ along with one-sided and central confidence belts <sup>*</sup><sup>*</sup>*We show the confidence belt consisting of central intervals \[$`n_1(\mu _0)`$, $`n_2(\mu _0)`$\] containing at least 90% of the probability for unknown Poisson mean $`\mu _0`$ in the absence of any known background (dotted) and the 90% one-sided belt consisting of intervals \[0, $`n_{os}(\mu _0)`$\](dashed). There is some arbitrariness in the choice of a central interval for a discrete variate. We choose the smallest interval such that there is $``$ 90% of the probability in the center and $``$5%, but as close as possible to 5%, on the right. The alternative of requiring $``$5%, but as close as possible to 5%, on the left gives slightly less symmetrical intervals. For the latter choice the 90% Poisson upper limit for $`n=0`$ is $`\mu _0=3.0`$ compared to $`\mu _0=2.62`$ for our choice. For $`\mu _0<2.62`$, according to this prescription, one cannot construct an interval containing probability $`>90\%`$ that does not include $`n=0`$ and we adopt 90% one-sided intervals. for the Poisson distribution without background. We note that the Roe-Woodroofe horizontal intervals do not coincide with the one-sided intervals shown for $`\mu <2.44`$. Therefore for some values of $`\mu `$ in this range, the confidence belt does not satisfy the coverage requirement that $``$90% of the probability is contained. Because coverage cannot be exact when the variable is discrete, the error for the example given here is not of great numerical significance. The minimum coverage of $`0.87`$ is obtained at $`\mu 0.4`$. Undercoverage is more severe for greater $`b`$; for $`b=10.0`$, the minimum coverage is $`0.78`$. However, it is desirable to have coverage, which we obtain as shown in Fig. 2 where we have changed the right side of the confidence belt so that the horizontal intervals contain probability $``$90%. We note that the confidence intervals for small $`n`$, i.e. $`n<b`$, are unchanged. Intervals for both constructions are given in Table I.
It would be nice to devise an ordering principle that can be directly applied to the Poisson pdf $`p(n;\mu +b)`$ to obtain the confidence belt shown in Fig. 2, if only because the construction we have used here is aesthetically unpleasing. This method, which consists of first determining vertical intervals per Ref. , and then fixing them, leaves something to be desired. However, in the end the method of construction does not really matter. What results here is an ordering procedure that yields a confidence belt with coverage and produces physically sensible intervals.
B. Roe has noted that our modification is equally applicable to a construction due to R. Cousins, in which the Roe-Woodroofe method of conditioning is applied to the Gaussian-with-boundary problem. Here, for example, an interval of confidence level $`\alpha `$ is sought for an unknown non-negative signal $`\mu `$ and the measurements x are normally distributed as N(x;$`\mu `$). As for the Roe-Woodroofe construction referred to above, the Cousins construction produces physically sensible confidence intervals for all $`x`$ including $`x<0`$. However this construction significantly undercovers for $`\mu <0.5`$ and significantly overcovers for $`\mu 1`$. In order to produce exact coverage using the Cousins construction, we retain the left hand (upper) curve of the confidence belt $`x_l(\mu )`$ and recalculate the right hand (lower) curve $`x_r(\mu )`$ so that the horizontal intervals contain probability $`\alpha `$ using:
$$2\alpha =erf(\frac{\mu x_l}{\sqrt{2}})+erf(\frac{x_r\mu }{\sqrt{2}}).$$
(1)
## III Conclusion
For the case of Poisson distributed measurements $`n`$ with a non-negative signal mean $`\mu `$ and known mean background $`b`$, the Roe-Woodroofe construction produces well-behaved confidence intervals, particularly for $`n<b`$ where other constructions yield unphysically small intervals. Since the construction is not based on integrating probabilities that arise from an unconditional pdf, it does not produce a confidence belt with coverage in the usual frequentist sense. We suggest a modification that provides coverage while preserving the desirable features of the construction. While the changes introduced by this modification are relatively small for the example given here (they are larger for greater b), nevertheless the procedure corrects a formal defect in the original construction. A similar modification provides coverage for a construction recently discussed by R. Cousins for the Gaussian-with-boundary problem.
|
warning/0004/hep-th0004169.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
Since its discovery, turbulence has attracted an enormous effort from scientists to study it . In recent years, there has been an increasing interest in studying turbulent hydrodynamics because lots of phenomena in Nature are turbulent, like in the flow of blood in biomechanics, meteorology and ocean engineering, astrophysics and formation of galaxies in the Universe. Beyond these phenomena, turbulence has became a very fruitful field of research for theoreticians, who have studied the analogies between turbulence and field theory, critical phenomena and condensed matter physics leading to renewed optimism to solve the problem of turbulence described by the Navier-Stokes equations
$$\frac{\stackrel{}{u}}{t}=\stackrel{}{w}\times \stackrel{}{u}\left(\frac{p}{\rho }+\frac{u^2}{2}\right)+\nu ^2\stackrel{}{u},$$
(1)
where $`\stackrel{}{u}(\stackrel{}{x},t)`$ is the velocity field, $`\stackrel{}{w}(\stackrel{}{x},t)`$ the vorticity field, $`p(\stackrel{}{x},t)`$ is the pressure, $`\rho `$ is the density and $`\nu `$ the kinematic viscosity.
Among turbulent phenomena, the hydrodynamic turbulence is one in which physicists have enormous interest because of the universal characteristics stressed by an incompressible fluid at high Reynolds numbers in the fully-developed turbulent regime. The regime related to the limit of high Reynolds number, $`R\mathrm{}`$, with $`R(LV)/\nu `$, which measures the competition between convective and diffusive processes in an incompressible fluid described by the Navier-Stokes equations. Above, $`L`$ is the integral length-scale of the largest eddies that should appear and $`V`$, is a characteristic large-scale velocity.
An important feature in turbulence is the presence of vorticities , but another flow quantity, the Lamb vector, can be used to describe turbulence, together with vorticities in the so called metafluid dynamics. In metafluid dynamics, the study of average quantities of an incompressible fluid in the fully developed regime is proposed, given a system where the average fields show up in a continuum inter-relation and respond as waves to quantities named turbulent sources. To do so, the Lamb vector $`\stackrel{}{l}`$ and the vorticity become the kernel of the turbulent dynamics, being related by $`\stackrel{}{l}(\stackrel{}{x},t)=\stackrel{}{w}\times \stackrel{}{u}`$. Defining Bernoulli energy function $`\varphi `$, $`\varphi (\stackrel{}{x},t)=\frac{p}{\rho }+\frac{u^2}{2}`$, one can introduce a theoretical concept, the turbulent charge density $`n`$, $`n(\stackrel{}{x},t)=^2\varphi `$. In the inertial range, the Bernoulli energy function is conserved, the energy generated at larger scales is transferred to smaller wave numbers across the region and energy pumping and dissipation are not relevant allowing us to make a very close analogy between turbulent hydrodynamics and electromagnetism . Later on, we show there is no loss of generality in treating the turbulent flow in inertial range from the point of view of our proposal.
This analogy between hydrodynamic turbulence and the electromagnetism in inertial range makes possible to study metafluid dynamics. So as in the electromagnetism , where we write the Lagrangian density as $`=\frac{1}{2}(\stackrel{}{E}^2\stackrel{}{B}^2)`$ ($`\stackrel{}{E}`$ is the electric field and $`\stackrel{}{B}`$ is the magnetic field), one can write down the Lagrangian density for the theory of turbulence as $`=\frac{1}{2}(\stackrel{}{l}^2\stackrel{}{\omega }^2)`$ or $`=\frac{1}{4}F_{\mu \nu }F^{\mu \nu }J_\mu V^\mu `$, with $`F^{\mu \nu }=^\mu V^\nu ^\nu V^\mu ,V^\mu =(\varphi ,\stackrel{}{u}),\text{ and }J^\mu =(n,\stackrel{}{J})`$, $`\stackrel{}{J}=\stackrel{}{u}n+\times (\stackrel{}{u}.\stackrel{}{\omega })\stackrel{}{u}+\stackrel{}{\omega }\times (\varphi +\stackrel{}{u}^2)+2(\stackrel{}{l}.)\stackrel{}{u}`$ is the turbulent current density. Writting $`\stackrel{}{J}=\stackrel{}{J}_{tr}+\stackrel{}{J}_l`$, where $`\stackrel{}{J}_{tr}`$ is the turbulent transverse current density and $`\stackrel{}{J}_l=\dot{\varphi }`$ is the turbulent longitudinal current density, the Lagrangian density becomes $`=\frac{1}{4}F_{\mu \nu }F^{\mu \nu }J_\mu V^\mu +\stackrel{}{u}.\dot{\varphi }`$, where now $`J^\mu =(n,J_{tr}^i)`$. We note that $`.\stackrel{}{J}_{tr}=0`$.
Despite of all resemblance between hydrodynamic turbulence and the dynamics of electromagnetic fields, there is a conceptual difference in the identification of the physical entities. In the classical electromagnetism, the physical fields are the electric and magnetic fields and the potentials are just mathematical artifices while in the metafluid dynamics the potentials play physical role.
The analogy between turbulent hydrodynamics and electromagnetism suggests we can study the turbulence as a constrained system. The theory of constrained systems plays an important role in the study of physical systems. It includes almost all known fundamental theories. Constrained systems were first studied systematically by Dirac .
Constrained systems are characterised in phase space by the presence of constraints, which are functions of the coordinates and momenta. These constraints can be classified into primary, secondary, etc., or first and second class . First class constraints imply the presence of gauge invariance in the theory, since they generate gauge transformations. Beside that, they exhibit the structure of the corresponding gauge group. On the other hand, second class constraints do not have these properties.
There exist several methods to treat constrained systems based on the classification given above. Most of these deal with first class constraints, for the second class ones there is the Dirac bracket method. There is also a more recent method by Faddeev and Jackiw , which does not follow the above classification. In this method, the Dirac formalism can be avoided and basic bracket relations can be obtained without using the usual Dirac brackets.
We will consider the Dirac and Faddeev-Jackiw formalisms to treat turbulence as constrained system. Those procedures have been used with great success in Quantum Field Theory for quantizing some models.
## 2 Dirac formalism
Under the point of view of a geometric interpretation, Arnold showed, using Lie algebra, that Euler flow can be described in the Hamiltonian formalism in any dimension. This has a lot of interesting consequencies for fluid mechanics and has been studied , but it is not quite obvious that this method could be used for viscous fluid. That is where the metafluid dynamics comes, showing a way to find a Hamiltonian formalism even for turbulent flow with viscosity.
Using again the analogy between electromagnetism and turbulence, one obtains the conjugate momentum for $``$ as $`\pi ^\mu =\frac{}{\dot{V}_\mu }=F^{\mu 0}`$ that immediately leads us to a primary constraint $`\pi ^00`$. So, the primary Hamiltonian becomes $`H_p=d\stackrel{}{x}\left(\pi ^\mu \dot{V}_\mu +\lambda \pi ^0\right)`$ $`=\mathrm{d}\stackrel{}{x}(\frac{1}{2}\stackrel{}{\pi }^2+\frac{1}{2}(\times \stackrel{}{u})^2+2\stackrel{}{\pi }.\varphi `$ $`+\pi ^0\dot{\varphi }+\frac{1}{2}(\varphi )^2+J^0\varphi \stackrel{}{J}_{tr}.\stackrel{}{u}+\lambda \pi ^0)`$.
In order to apply the Dirac Hamiltonian method , we need to look for secondary constraints. Imposing that primary constraint must be conserved in time, we get $`\dot{\pi }^0=\{\pi ^0,H_p\}0`$. This consistency condition over the primary constraint leads to the secondary constraint $`.\stackrel{}{\pi }J^00`$, that is exactly one of the metafluid dynamics equations. If one goes on and imposes the consistency condition over this secondary constraint, we observe that no new constraints are obtained via this iterative procedure.
The constraints $`\pi ^00`$ and $`.\stackrel{}{\pi }J^00`$ are first class constraints. It means this new theory of turbulence is gauge-invariant. A result expected due to the analogy between electromagnetism and turbulence. This analogy allows us to apply to this new theory of turbulence all the machinery associated with gauge theories. Since we have two constraints of first class, we have two degrees of freedom to field $`V^\mu `$. These two degrees of freedom can be associated to the vorticity. This means we must explore the gauge invariance of the theory to choose two gauges. Following the analogy between electromagnetism and turbulence we find the gauges $`.\stackrel{}{u}=0,\text{(Coulomb or transverse gauge) and }V^0=\alpha \text{ (constant)}\text{(Temporal gauge)}`$, where the first gauge comes from condition of incompressibility of fluid and the second, from the fact that Bernoulli energy function is constant.
So, the theory has the constraints $`T_1=V^0\alpha 0,T_2=\pi ^00,T_3=.\stackrel{}{u}0,\text{ and }T_4=.\stackrel{}{\pi }J^00`$,which are second class ones. To implement the Dirac brackets we need to compute the matrix elements of their Poisson brackets, which reads $`C_{12}=\delta (\stackrel{}{x}\stackrel{}{y})=C_{21},C_{34}=^2\delta (\stackrel{}{x}\stackrel{}{y})=C_{43}`$, and all other elements are zero. Using the Dirac brackets $`\{A,B\}^{}=\{A,B\}\{A,T_\alpha \}C_{\alpha \beta }^1\{T_\beta ,B\}`$, we get
$`\{u_i(\stackrel{}{x}),u_j(\stackrel{}{y})\}^{}`$ $`=`$ $`0=\{\pi _i(\stackrel{}{x}),\pi _j(\stackrel{}{y})\}^{},`$
$`\{u_i(\stackrel{}{x}),\pi _j(\stackrel{}{y})\}^{}`$ $`=`$ $`\left(\delta _{ij}{\displaystyle \frac{_i^x_j^x}{^2}}\right)\delta (\stackrel{}{x}\stackrel{}{y}).`$ (2)
Inspite of these Dirac brackets have been computed in the inertial range, they are valid for all fully-developed turbulence, as we show bellow.
## 3 Faddeev-Jackiw formalism
With the purpose to show that the presence of viscous term in the Navier-Stokes equation does not change the Dirac brackets, let’s analyze the metafluid dynamics using a second treatment for this constrained theory, the Faddeev-Jackiw procedure, also known as sympletic formalism.
Faddeev and Jackiw showed how to implement the constraints directly into the canonical part of the first order Lagrangian. Once this method is applied to first order Lagrangians, for implementing the Faddeev-Jackiw procedure we write the Lagrangian density of turbulence making explicit the time-derivative of the fields $`V_\mu `$ :
$``$ $`=`$ $`{\displaystyle \frac{1}{2}}\dot{\stackrel{}{u}}+\dot{\stackrel{}{u}}.\varphi +{\displaystyle \frac{1}{2}}(\varphi )^2{\displaystyle \frac{1}{2}}(\times \stackrel{}{u})^2\nu \dot{\stackrel{}{u}}.^2\stackrel{}{u}`$ (3)
$``$ $`\nu (^2\stackrel{}{u}).(\varphi )+{\displaystyle \frac{1}{2}}\nu ^2(^2\stackrel{}{u})^2+\stackrel{}{u}.\stackrel{}{J}\nu \stackrel{}{u}.n\varphi n,`$
and for transforming it from second to first order, we consider the momentum as the auxiliary field, extending the configuration space. In that way, we get the equation of motion for $`\stackrel{}{\pi }`$ : $`\stackrel{}{\pi }=\varphi \dot{\stackrel{}{u}}+\nu ^2\stackrel{}{u}`$. Replacing back this result in the Lagrangian we return to the original quadratic term. Using that $`\stackrel{}{J}=\stackrel{}{J}_{tr}+\stackrel{}{J}_l`$, the Lagrangian density in eq.(3) becomes
$``$ $`=`$ $`{\displaystyle \frac{1}{2}}\stackrel{}{\pi }^2\stackrel{}{\pi }.\varphi \stackrel{}{\pi }.\dot{\stackrel{}{u}}+\nu \stackrel{}{\pi }.^2\stackrel{}{u}+{\displaystyle \frac{1}{2}}(\times \stackrel{}{u})^2`$ (4)
$``$ $`\nu \stackrel{}{u}.n+\stackrel{}{u}.\stackrel{}{J}_{tr}+\stackrel{}{u}.\dot{\varphi }n\varphi `$
In this case, the first order Lagrangian reads $`^{(0)}=\stackrel{}{\pi }.\dot{\stackrel{}{u}}\dot{\stackrel{}{u}}.\varphi U^{(0)}`$, where the potential density is $`U^{(0)}=\frac{1}{2}\stackrel{}{\pi }^2+\stackrel{}{\pi }.\varphi \frac{1}{2}(\times \stackrel{}{u})^2\nu \stackrel{}{\pi }.^2\stackrel{}{u}\stackrel{}{u}.\stackrel{}{J}_{tr}+\varphi n+\nu \stackrel{}{u}.n`$.
The initial set of sympletic variables $`\xi _i^{(0)}=\{u_i,\pi _i,\varphi \}`$ allows us to identify the quantities $`a_i^{(0)\stackrel{}{u}}=\pi _i_i\varphi ,a_i^{(0)\stackrel{}{\pi }}=0,\text{ and }a_i^{(0)\varphi }=0`$. These lead us to the following matrix $`f^{(0)}`$,
$$f^{(0)}=\left(\begin{array}{ccc}0& _j^y& 0\\ _i^y& 0& \delta _{ij}\\ 0& \delta _{ij}& 0\end{array}\right)\delta (\stackrel{}{x}\stackrel{}{y}),$$
(5)
which obviously is singular. Thus, the system has constraints in the Faddeev-Jackiw formalism. The components of the zero-mode $`\stackrel{~}{v}^{(0)}=(v^\varphi ,\mathrm{\hspace{0.17em}0},v^\stackrel{}{\pi })`$ satisfy $`v_i^\stackrel{}{\pi }_iv^\varphi =0`$.
The primary constraint is obtained from $`\mathrm{d}\stackrel{}{x}v^\varphi (\stackrel{}{x})[.\stackrel{}{\pi }(\stackrel{}{x})+n(\stackrel{}{x})]=0`$. Since $`v^\varphi (\stackrel{}{x})`$ is an arbitrary function and the fluid is taken incompressible, we obtain the constraint $`.\stackrel{}{\pi }(\stackrel{}{x})n(\stackrel{}{x})=0`$.
Introducing this constraint back into the Lagrangian by means of a Lagrange multiplier $`\lambda `$ one obtains $`^{(1)}=\stackrel{}{\pi }.\stackrel{}{u}\varphi .\dot{\stackrel{}{u}}+\dot{\lambda }(.\stackrel{}{\pi }n)U^{(1)}`$ where $`U^{(1)}=\frac{1}{2}\stackrel{}{\pi }^2\frac{1}{2}(\times \stackrel{}{u})^2\nu \stackrel{}{\pi }.^2\stackrel{}{u}\stackrel{}{u}.\stackrel{}{J}_{tr}+\nu \stackrel{}{u}.n`$.
Considering now that the new set of sympletic variables is given in the following order, $`\xi =(u_i,\varphi ,\pi _i,\lambda )`$, we have $`a_i^{(1)\stackrel{}{u}}=\pi _i_i\varphi ,a_i^{(1)\varphi }=0,\text{ and }a^{(1)\lambda }=.\stackrel{}{\pi }n`$. From there we obtain the sympletic matrix $`f^{(1)}`$,
$$f^{(1)}=\left(\begin{array}{cccc}0& _i^y& \delta _{ij}& 0\\ _j^y& 0& 0& 0\\ \delta _{ij}& 0& 0& _i^y\\ 0& 0& _j^x& 0\end{array}\right)\delta (\stackrel{}{x}\stackrel{}{y}).$$
(6)
that is a nonsingular matrix. Then we can identify it as the sympletic tensor of the constrained theory. The inverse of $`f^{(1)}`$ gives us the Dirac brackets of the physical fields and can be obtained in a straightforward calculation. The result is
$$(f^{(1)})^1=\left(\begin{array}{cccc}0& \frac{_i^x}{^2}& \delta _{ij}\frac{_i^x_j^x}{^2}& 0\\ \frac{_j^x}{^2}& 0& 0& \frac{1}{^2}\\ \delta _{ij}+\frac{_i^x_j^x}{^2}& 0& 0& \frac{_i^x}{^2}\\ 0& \frac{1}{^2}& \frac{_j^x}{^2}& 0\end{array}\right)\delta (\stackrel{}{x}\stackrel{}{y})$$
(7)
From the matrix $`(f^{(1)})^1`$ we identify
$$\{u_i(\stackrel{}{x}),\dot{u}_j(\stackrel{}{y})\}^{}=\left(\delta _{ij}+\frac{_i^x_j^x}{^2}\right)\delta (\stackrel{}{x}\stackrel{}{y})$$
(8)
where the auxiliary fields were eliminated and we returned to the original variables. These brackets are nothing more than brackets of the turbulence gauge field in the gauge $`.\stackrel{}{u}=0`$, obtained in eq. (2), showing us that the Dirac brackets are the same for the Navier-Stokes equation with or without the viscous term.
The straightforward step would be the quantization of this constrained theory, that can be obtained by the well known canonical quantization rule $`\{,\}^{}i[,]`$. Doing so, we get the commutators
$`[u_i(\stackrel{}{x}),u_j(\stackrel{}{y})]`$ $`=`$ $`0=[\pi _i(\stackrel{}{x}),\pi _j(\stackrel{}{y})],`$
$`[u_i(\stackrel{}{x}),\pi _j(\stackrel{}{y})]`$ $`=`$ $`i\left(\delta _{ij}{\displaystyle \frac{_i^x_j^x}{^2}}\right)\delta (\stackrel{}{x}\stackrel{}{y}).`$ (9)
Once we have the canonical quantization rule, we can apply standard Quantum Field Theory technics to find the generating functional, consequently correlation functions and all physical quantities one wish.
## 4 Conclusion
In conclusion, we have proposed to study incompressible turbulent hydrodynamics in the metafluid dynamics formalism applying tools commonly used in Quantum Field Theory for constrained systems. This study is possible because of the close analogy between incompressible turbulent hydrodynamics and the electromagnetism. Exploiting this analogy we computed the Dirac brackets using the Dirac and Fadeev-Jackiw formalisms. We showed that Dirac brackets found in the Dirac formalism in the inertial regime ($`\nu =0`$) is valid for all regime of fully-developed turbulence, as we found in the Fadeev-Jackiw formalism. We believe this approach of turbulence as a gauge theory can lead to results that tend to renew the optimism to solve the difficult problem of turbulence.
## 5 Acknowledgements
This work was partially supported by CNPq and FAPEMIG. One of us (A.C.R.M) thanks to CNPq for the financial support and (W.O and F.I.T.) would like to thank FAPEMIG and CNPq for partial support.
|
warning/0004/cond-mat0004483.html
|
ar5iv
|
text
|
# Entropy-vanishing transition and glassy dynamics in frustrated spins
## a Model
One of the simplest non-randomly frustrated spin models is the triangular-lattice Ising antiferromagnet (TIAFM). The TIAFM has an exponentially large number of ground states and has a zero-temperature critical point. The model studied in this letter is the compressible TIAFM (CTIAFM) in which the coupling of the spins to the elastic strain fields removes the exponential degeneracy of the ground-state. We solve the CTIAFM exactly within the ground-state ensemble of the TIAFM and show that there is an entropy-vanishing transition which involves extended structures. We then present results of simulations which indicate that the entropy-vanishing transition leads to glassy dynamics.
The Hamiltonian of the CTIAFM is:
$`H`$ $`=`$ $`J{\displaystyle \underset{<ij>}{}}S_iS_jϵJ{\displaystyle \underset{\alpha }{}}e_\alpha {\displaystyle \underset{<ij>_\alpha }{}}S_iS_j`$ (1)
$`+`$ $`N{\displaystyle \frac{E}{2}}{\displaystyle \underset{\alpha }{}}e_\alpha ^2.`$ (2)
Here $`J`$, the strength of the anti-ferromagnetic coupling, is modulated by the presence of the second term which defines a coupling between the spins and the homogeneous strain fields $`e_\alpha ,\alpha =1,2,3`$, along the three nearest-neighbor directions on the triangular lattice. The last term stabilizes the unstrained lattice. The total number of spins in the system is given by $`N`$. The ground-state of the CTIAFM is a three-fold degenerate striped phase where up and down spins alternate between rows and there is a shear distortion characterized by $`e_1=e`$ and $`e_2=e_3=e`$ if the rows are along the direction of $`e_1`$. Within the manifold of the TIAFM ground-states, the competition between energy gained from the lattice distortions and the extensive entropy of the TIAFM ground-states leads to an entropy-vanishing transition. The order parameter associated with this entropy-vanishing transition is the density of extended string-like structures which characterize the TIAFM ground states.
The string picture of the TIAFM ground states derives from a well-known mapping of these states to dimer coverings. In a ground state, there is one unsatisfied bond per triangular plaquette and the dimers are the filled-in bonds of the dual honeycomb lattice that cross the unsatisfied bonds of the triangular lattice, as shown in Fig.(1). Superposing a dimer configuration on a “standard” dimer configuration where all the dimers are vertical leads to a string configuration (cf Fig.(1)). Assuming spin-flip dynamics for the moment, the only spins that can be changed while the system remains in the ground-state manifold, are the ones which have a coordination of 3-3 (3 satisfied and 3 unsatisfied bonds). These are the fast spins in the system, and, as shown in Figure (1), are located at isolated kinks on the strings. The strings, therefore, play the role of dynamical heterogeneities in this lattice model.
## b Exact results
We can solve the CTIAFM exactly within the restricted spin ensemble of the ground states of the TIAFM. All the states in this ensemble can be classified according to the string density $`p=N_s/L`$ where $`L`$ is the linear dimension of the sample and $`N_s`$ is the number of strings. The number of spin states ($`\mathrm{\Omega }(p)`$) belonging to a particular string-density sector $`p`$ has been shown to be exponentially large: $`\mathrm{\Omega }(p)=\mathrm{exp}(N\gamma (p))`$, with $`N=L\times L`$ being the total number of spins. As shown in Fig. (2), the entropy $`\gamma (p)`$ has a peak at $`p=2/3`$. In the CTIAFM (Eq. 2), the strain $`e_\alpha `$ couples to $`S_iS_j_\alpha `$. This average counts the number of strings along the $`\alpha `$ direction, i.e. the number of strings obtained by taking an overlap with the standard dimer state with all the dimers perpendicular to the $`\alpha `$ direction. Under periodic boundary conditions, only one of the three string densities is independent.
The strain field appears in the Hamiltonian as a purely Gaussian variable and can be integrated out, and the CTIAFM energy per spin, in the restricted ensemble, can be written in terms of the one, independent string density $`p`$:
$$E(p)=(\mu /2)((12p)^2+2(1p)^2)).$$
(3)
Here $`\mu =ϵ^2J^2/E`$. The energy function $`E(p)`$ distinguishes different string sectors and is minimized by $`p=0`$. The entropy of the ground states, $`\gamma (p)`$, on the other hand favors the $`p=2/3`$ sector and the competition between energy and entropy leads to the possibility of a phase transition. In the thermodynamic limit, the partition function $`Z=_p\mathrm{exp}^{Nf(p)}`$, is dominated by the string density which minimizes $`f(p)=\beta E(p)\gamma (p)`$, where $`\beta `$ is the inverse temperature. The exact free energy corresponds to this minimum value of $`f(p)`$ and the only relevant coupling constant in the problem is $`\beta \mu `$. For small values of the coupling constant, the function $`f(p)`$ shown in Fig (2(b)), has only one minimum at $`p=2/3`$. As the coupling constant is increased, this minimum stays pinned at $`2/3`$ and a second minimum starts developing at $`p=0`$. The $`p=0`$ state stays metastable until at $`\mu /T_1=(3/4)\gamma (2/3)0.24`$ there is a first-order transition from the $`p=2/3`$ state to the $`p=0`$ state. The $`p=2/3`$ state loses its metastability at a larger coupling given by $`\mu /T^{}=\sqrt{3}\pi /12`$. At $`T^{}`$, the entropy vanishes and the order parameter shows a discontinuous change from $`p=2/3`$ to $`p=0`$. This transition is reminiscent of the transitions observed in p-spin spin glasses.
These exact results show that in the CTIAFM, there is an entropy vanishing transition which defines the limit of stability of the homogeneous, liquid-like $`p=2/3`$ state. In the Adam-Gibbs scenario, such a transition underlies the glass transition. The exact results are valid for the CTIAFM acting within the ground-state ensemble of the TIAFM. Defects, which correspond to triangles with all three bonds unsatisfied, can take the system out of the ground-state manifold. For low defect densities, however, it is possible that the entropy-vanishing transition survives in some form and leads to slow glassy dynamics. We have investigated this scenario by performing Monte Carlo simulations of the CTIAFM.
## c Simulations
The parameters of the model were chosen to be $`J=1`$, $`ϵ=0.6`$ and $`E=2`$. These values yield a small value of $`\mu `$ ($`=0.18`$) and ensures that at the transition, $`T^{}`$ ($`=0.397`$), the defect density is low. The average defect number-density was measured to be $`0.04\%`$ at $`T=0.45`$. We used Monte Carlo simulations to study the dynamics of the supercooled state following instantaneous quenches to temperatures below $`T_1=0.75`$. Spin-exchange kinetics was extended to include moves which attempted changes of the strain fields $`e_\alpha `$. Details of the simulation algorithm have been published earlier. System sizes ranged from 48x48 up to 120x120. Unless otherwise stated, the results presented in this letter were obtained from 96x96 systems.
As the glass transition is approached, global quantities such as the energy per spin should exhibit anomalously slow relaxation processes. Fig (3) shows the energy autocorrelation function $`C_E(t,t_0)=<E(t_0)E(t+t_0)>`$. For quench temperatures between $`T=0.6`$ and T=0.47, $`C_E(t,t_0)`$ is independent of the time origin $`t_0`$ and has a stretched exponential form; $`\mathrm{exp}^{(t/\tau _E)^\beta }`$. The stretching-exponent $`\beta `$ decreases from 0.45 to 0.35 over this temperature range and the time scale $`\tau _E`$ increases rapidly as is evident from the top panel of Fig (3). Below $`T0.47`$, the energy autocorrelation function depends on the time origin $`t_0`$, indicating that the equilibration times have become longer than our observation times. To illustrate the dependence on the waiting time $`t_0`$, we have shown the autocorrelation function averaged over three different ranges of $`t_0`$ at $`T=0.45`$. The system is seen to relax more slowly for longer waiting times $`t_0`$. This behavior of the energy autocorrelation function is similar to that of supercooled liquids and the temperature $`T=0.47`$ is analogous to the laboratory glass transition temperature at which the equilibration time becomes longer than the observation time. In the CTIAFM, the proximity of this transition to $`T^{}`$ suggests that the glassy dynamics is related to the entropy vanishing transition.
The microscopic picture of the entropy-vanishing transition is one where the string density vanishes. An analysis of the string-density relaxation can, therefore provide some insight into the nature of the dynamics at the glass transition. We find that the string-density autocorrelation functions are well described by exponential relaxations, in contrast to the energy autocorrelation functions. Fig. (4) shows the results of our simulations for the relaxation times and fits to a power-law and a Vogel-Fulcher form. Both fits yield a time-scale divergence at a temperature $`TT^{}`$. The $`\tau _E`$ obtained from the stretched exponential fits to $`C_E(t,t_0)`$ has a temperature dependence which tracks that of the string relaxation time. This observation suggests that the slow, non-exponential relaxations are a consequence of the freezing of the string-density relaxation which, in turn, is related to the entropy vanishing transition. The static susceptibility associated with the string density changes only by a factor $`2`$ over the temperature range in which the time scales change by a factor $`40`$, indicating that the entropy-vanishing transition suffers from anomalously strong critical slowing down.
In order to investigate the nature of the string-relaxations further, we measured the distribution, $`P(\mathrm{\Delta }p)`$, of $`\mathrm{\Delta }p=p(t+t_0)p(t_0)`$, the deviation of the string density in the time interval $`t`$ (Fig. (5)). The distributions were generated by choosing different time origins $`t_0`$. The most striking feature of the distributions, observed at temperatures close to $`T=T^{}`$, is its non-Gaussian nature at intermediate times. At $`T=0.55`$, the non-Gaussian feature is most pronounced at $`t=4000`$. Beyond this time the distribution relaxes towards a Gaussian and for $`t8000`$, the distribution is time independent. At $`T=0.47`$, a time-independent behavior is not observed for times as long as 30,000 and the distributions are non-Gaussian at all intermediate times. In usual critical-point dynamics, one would expect to find a distribution of $`\mathrm{\Delta }p`$ which takes longer to reach its time-independent form as the critical point is approached and to find non-Gaussian behavior (within the limits of finite-size cutoffs) in the stationary distribution. In contrast, we observe the most pronounced non-Gaussian features at intermediate times. Drawing an analogy with critical phenomena, this observation leads us to speculate that there exists a time-dependent length scale, $`\xi (\tau )`$ has a peak at a time $`\tau _0(T)`$. As $`TT^{}`$, the time scale $`\tau _0`$ and the height of the peak, $`\xi (\tau _0)`$ appear to diverge. The exact nature of the divergence is difficult to extract from the current data. These apparent divergences indicate that the thermodynamic transition present in the zero-defect sector has been replaced by a dynamical transition.
## d Connection to real glasses
In conclusion, our study of the CTIAFM provides strong indication that the slow, glassy dynamics in this model is associated with an entropy-vanishing transition involving extended, string-like structures. These structures are a manifestation of the frustration embodied in the nearest-neighbor, anti-ferromagnetic interactions, and the entropy-vanishing transition is a consequence of coupling to another degree of freedom, the lattice strain, which tends to remove the frustration in the system. The strings are naturally occurring dynamical heterogeneities since they demarcate regions of fast spin flips. If the relation between frustration and dynamical heterogeneities is a generic feature of glass formers, then our observations would suggest that the clusters of mobile particles observed in Lennard-Jones simulations and in colloidal systems should consist of the most frustrated particles in the system. In a Lennard-Jones mixture, these should be the particles with the least number of unlike bonds (if unlike bonds are energetically preferred) and in a colloidal system, these should be the particles which have coordinations that are furthest from being icosahedral. An experimental verification of this correlation between geometry and mobility would be a direct test of the connection between the glass-transition and an entropy-vanishing transition involving extended structures forced in by frustration.
The work of BC was supported in part by NSF grant number DMR-9815986 and the work of HY was supported by DOE grant DE-FG02-ER45495. We would like to thank R. K. P. Zia, W. Klein, H. Gould, S. R. Nagel and J. Kondev for many helpful discussions.
|
warning/0004/gr-qc0004047.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
There have been many topics issued on the quantum mechanics of the neutrino oscillation in the flat spacetime and in the curved spacetime , most of them centered on the understanding of the quantum coherence condition between the various mass-eigenstates, as well as on the calculations of the phase factor in the spacetime. The mass neutrino oscillation problem is connected basically to the phase difference between different mass-eigenstate, a property intimately related to the basic principles of quantum mechanics. In the standard treatment of the neutrino oscillation, the condition of same-energy (same-momentum) with different momenta (energies) of the mass-eigenstates is introduced , which arises the same practically applied massive neutrino oscillation phase, however it is indicated that the unambiguous theoretical description of the massive neutrino oscillation phase will be involved in the wave packet formalism and not in the plane wave approximation . Further, it is shown that the same energy or same momentum description is just the arbitrariness of the choice of the Lorentz frame in which it is valid, nothing to do with the physical argument . The change of flavors is described by the oscillation length, and it is usually claimed that both conditions (same-energy and same-momentum) present practically the same neutrino oscillation results. Sometimes, a source-dependent condition is added , which implies in giving up either the same-energy or the same-momentum prescription. This calculation of the neutrino oscillation phase, however, yields the same result of the standard treatment. Furthermore, the velocity difference of the various mass-eigenstates results in a spacetime separation for neutrinos of different flavor , which is another source of confusion on the interference condition for the different neutrino. The same velocity description has been paid much attention , but this prescription is pointed out to be forbidden kinematically . Actually, considerable confusion has arisen in the description, interpretation and understanding of the neutrino oscillation, a problem which involves the fundamental principles of both quantum mechanics and special relativity, such as the uncertainty principle, the superposition principle, and simultaneity problems because the mass neutrinos are high energy quantum objects. Moreover it is often noted the factor of 2 of the neutrino phase calculations in the flat spacetime and in the curved spacetime , which is believed to be the consideration of the space phase or the time phase , as well as the arrival time difference of the two mass neutrinos .
In the Minkowski diagram of flat spacetime, we discuss how the same-energy, the same-momentum of the neutrino phase present the same practical result, and the relation between the arrival time difference and the double counting of the phase, and further, we extend our discussions to the curved spacetime. We also calculate the phase factor in the case of the same velocity description although this process is not physically realized. Here we stress that all our calculations are based on the plane wave treatment of the neutrino, otherwise the wave packet should be considered . We set $`\mathrm{}=c=1`$ throughout this article.
## 2 The Standard Treatment of the Neutrino Oscillation
In the standard treatment of the neutrino oscillation, if two generations are taken into account, the neutrino flavor-state is written as
$$|\nu _\alpha (x,t)=\underset{j}{}U_{\alpha j}\mathrm{exp}\left[i\mathrm{\Phi }_j\right]|\nu _j,$$
(1)
where
$$(U)=U_{\alpha j}=\left(\begin{array}{cc}\mathrm{cos}\theta & \mathrm{sin}\theta \\ \mathrm{sin}\theta & \mathrm{cos}\theta \end{array}\right),$$
(2)
with $`\theta `$ the mixing angle. In vector form,
$$\left(\begin{array}{c}|\nu _e(x,t)\\ |\nu _\mu (x,t)\end{array}\right)=\mathrm{exp}(i𝚽)𝐔\left(\begin{array}{c}|\nu _\mathrm{𝟏}\\ |\nu _\mathrm{𝟐}\end{array}\right),$$
(3)
where
$$\left(\mathrm{exp}(i𝚽)\right)=\left(\begin{array}{cc}\mathrm{exp}(i\mathrm{\Phi }_1)& 0\\ 0& \mathrm{exp}(i\mathrm{\Phi }_2)\end{array}\right),$$
(4)
with $`\mathrm{\Phi }_j`$ the eigenvalue of the phase operator
$$\mathrm{\Phi }_j=\left(E_jdtP_jdx\right).$$
(5)
Here, flavor (mass) indices are expressed by Greek (Latin) letters, and $`\nu _e`$ and $`\nu _\mu `$ are represented respectively by $`\nu _1`$ and $`\nu _2`$. The matrix elements $`U_{\alpha j}`$ comprise the transformation between the flavor and mass basis. Now, we suppose a pure flavor-state electron neutrino $`|\nu _e`$ is at the initial source position $`𝐀`$ ($`x=0`$ and $`t=0`$). The mass eigenstates are taken to be the eigenstates of energy (momentum). The momentum (energy) then satisfies the mass shell relation
$$𝐏=\sqrt{𝐄^2𝐦^2}𝐄\frac{𝐦^2}{2𝐄},$$
(6)
where $`𝐦`$ is the rest-mass operator corresponding to the mass eigenvalue $`m_j`$ of the eigenstate $`|\nu _j`$ .
The oscillation probability $`𝒫`$ from flavor $`|\nu _e`$ at the source position A, to flavor $`|\nu _\mu `$ at the detector position $`𝐁`$ ($`x=L`$) is given by ,
$$𝒫(\nu _e\nu _\mu )=|\nu _\mu |\nu _e(x,t)|^2=\mathrm{sin}^22\theta \mathrm{cos}\frac{\mathrm{\Delta }\mathrm{\Phi }}{2},$$
(7)
where
$$\mathrm{\Delta }\mathrm{\Phi }=\mathrm{\Phi }_2\mathrm{\Phi }_1$$
(8)
is the phase difference between the different mass states. The oscillation length is defined by taking $`\mathrm{\Delta }\mathrm{\Phi }=2\pi `$. From the standard treatment of the neutrino oscillation, the common momentum (energy), and the approximation condition (6) lead to a phase difference $`\mathrm{\Delta }\mathrm{\Phi }`$ and an oscillation length $`L_{osc}^S`$ given respectively by
$$\mathrm{\Delta }\mathrm{\Phi }=\frac{\mathrm{\Delta }m^2}{2E}L,$$
(9)
and
$$L_{osc}^S=2\pi \frac{2E}{\mathrm{\Delta }m^2},$$
(10)
where $`\mathrm{\Delta }m^2=m_2^2m_1^2`$, with $`m_2>m_1`$. To compute the oscillation probability, the following three assumptions are often applied : ($`i`$) The mass eigenstates are taken to be the energy (momentum) eigenstates, with a common energy (momentum); ($`ii`$) up to $`𝒪`$($`m/E`$), we have the approximation $`PE>>m`$; ($`iii`$) a massless trajectory is assumed, which means that the neutrino travels along the null trajectory defined by $`dx=dt`$. With these assumptions, the flavor state is simplified as
$$|\nu _\alpha (x,t)=\underset{j}{}U_{\alpha j}\mathrm{exp}\left[i\left(\frac{m_j^2}{2E}\right)x\right]|\nu _j.$$
(11)
This state is then used to compute the oscillation amplitude.
## 3 Reexamination of the Standard Treatments
In this section, we discuss the standard procedures for the calculation of the phase factor along both the particle world-line and the null trajectory in Minkowski spacetime . First, let us recall that, when the null condition is applied, the standard treatment yields
$$\mathrm{\Phi }^S=(EdtPdx)(EP)𝑑x=\frac{m^2}{2E}L.$$
(12)
If the null condition $`dx=dt`$ is not used, we find
$$\mathrm{\Phi }=(EdtPdx)=\frac{(E^2P^2)}{P}𝑑x=\frac{m^2}{P}L2\mathrm{\Phi }^S.$$
(13)
The phase difference is then
$$\mathrm{\Delta }\mathrm{\Phi }=\left(\frac{m_2^2}{P_2}\frac{m_1^2}{P_1}\right)L2\mathrm{\Delta }\mathrm{\Phi }^S.$$
(14)
In order to illustrate the problem explicitly, the original definition of the phase factor will be obtained by means of the interval in the Minkowski spacetime. From Fig.1, we see that the interval of the world line, when the neutrino propagates from $`𝐀`$ to $`𝐁`$ in spacetime, is
$$ds^2(AD)=\overline{BD}^2\overline{BN}^2=(\overline{BD}+\overline{BN})\overline{ND}=\frac{L}{v}^2L^2=\frac{L^2}{v^2\gamma ^2},$$
(15)
and consequently
$$\mathrm{\Phi }=mds(AD)=\frac{mL}{v\gamma }=\left(\frac{m^2}{E}L\right)\left(\frac{1}{v}\right)2\mathrm{\Phi }^S.$$
(16)
The standard treatment gives
$$mds(AD)=\left(\frac{m\overline{BD}}{ds}\right)\overline{BD}\left(\frac{m\overline{BN}}{ds}\right)\overline{BN}=(EP)\overline{BN}=\frac{m^2}{2E}L=\mathrm{\Phi }^S.$$
(17)
Although the difference $`\overline{ND}=\overline{BD}\overline{BN}`$ between the particle world time and the null time is small, the phase is not solely related to the distance in space, but also to the interval in the spacetime, and the phase is sensitive to the null condition when the velocity of the neutrino approaches the speed of light. If we neglect this small difference in the time interval, a factor of 2 will appear in the phase factor! To illustrate this conclusion, we inspect in the next subsections the standard treatments in more detail.
### 3.1 Reexamination of the Same-Energy Prescription
If two neutrinos share the same energy ($`E_1=E_2=E`$, $`E^2P_j^2=m_j^2`$), but present different momenta, the contribution to the phase difference will come from the integration of the momentum in space because the same-energy condition makes the integration of the energy in time to vanish. This is the main point of the same-energy condition in the standard treatment. Now, we will explore this point in more detail by using the Minkowski diagram of Fig.2.
For convenience, we will keep using the convention $`m_2>m_1`$, which leads to $`\gamma _1>\gamma _2`$, $`P_1>P_2`$ and $`v_1>v_2`$. In the Minkowski diagram of Fig.2, the faster the frame the closer it is of the null line . This property will be helpful for our analysis.
According to the standard treatment in the case of same-energy ($`E_1=E_2`$), we have
$`\mathrm{\Delta }\mathrm{\Phi }^S`$ $`=`$ $`{\displaystyle _A^B}(E_2dtP_2dx){\displaystyle _A^B}(E_1dtP_1dx)`$ (18)
$`=`$ $`{\displaystyle _A^B}(P_2P_1)𝑑x={\displaystyle \frac{\mathrm{\Delta }m^2}{2E}}L.`$
This computation seems not to use the null condition $`dx`$ = $`dt`$, and it does not use the fact that, because the velocities are not the same, the neutrinos with velocities $`v_1`$ and $`v_2`$ follow different world-lines. However, the null condition is in fact used when we replace both $`\overline{BD_1}`$ and $`\overline{BD_2}`$ by $`\overline{BN}`$ (see Fig.2). It is thus important to remark that it is the null condition, not the same energy condition, that accounts for the cancellation of the time phase.
What does it happen if the real world-lines (geodesic) of the neutrinos are taken into account? In order to answer this question, it is important to notice that the massive neutrinos $`\nu _1`$ and $`\nu _2`$ describe two different world lines from the source A to the detector B, given respectively by $`AD_1`$ and $`AD_2`$ (see Fig.2). Taking this into account, we obtain:
$`\mathrm{\Delta }\mathrm{\Phi }`$ $`=`$ $`{\displaystyle _A^{D_2}}(E_2dtP_2dx){\displaystyle _A^{D_1}}(E_1dtP_1dx)`$ (19)
$`=`$ $`{\displaystyle _{D_1}^{D_2}}E𝑑t{\displaystyle _A^B}(P_2P_1)𝑑x`$
$`=`$ $`{\displaystyle \frac{\mathrm{\Delta }m^2E}{2P_1P_2}}L+{\displaystyle \frac{\mathrm{\Delta }m^2}{2E}}L`$
$``$ $`{\displaystyle \frac{\mathrm{\Delta }m^2}{2E}}L+{\displaystyle \frac{\mathrm{\Delta }m^2}{2E}}L=2\mathrm{\Delta }\mathrm{\Phi }^S.`$
If the arrival time-difference of the neutrinos is considered, therefore, the time-phase is not canceled. As it is as large as the space-phase, the real phase results twice the value yielded by the standard treatment. A similar conclusion has also been obtained in Ref. . The computation of (18) and (19) indicates that it is the null condition, not the same energy condition, the responsible for the duplication of the standard phase difference.
### 3.2 Reexamination of the Same-Momentum Prescription
Following a procedure similar to that used in the case of the same-energy prescription, we examine now the same-momentum prescription ($`P_1=P_2=P`$, $`E_j^2P^2=m_j^2`$). We still suppose $`m_2>m_1`$, which leads to $`\gamma _1>\gamma _2`$, $`E_1<E_2`$ and $`v_1>v_2`$. These relations allow us to use the same Minkowski diagram of Fig.2. We find in this case
$`\mathrm{\Delta }\mathrm{\Phi }`$ $`=`$ $`{\displaystyle _A^{D_2}}(E_2dtP_2dx){\displaystyle _A^{D_1}}(E_1dtP_1dx)`$ (20)
$`=`$ $`{\displaystyle _B^{D_2}}E_2𝑑t{\displaystyle _B^{D_1}}E_1𝑑t`$
$`=`$ $`{\displaystyle \frac{\mathrm{\Delta }m^2}{P}}L2\mathrm{\Phi }^S.`$
It is the same result of the case of the same-energy prescription. However, a small difference exists. In order to show the influence of the same-momentum and of the null-condition on the calculation of the phase difference, we examine the phase computation of the same-momentum prescription of the standard treatment, but using the null condition. This amounts to replace $`BD_2`$ and $`BD_1`$ by the null time corresponding to $`BN`$. We find
$`\mathrm{\Delta }\mathrm{\Phi }^S`$ $`=`$ $`{\displaystyle _A^B}(E_2dtP_2dx){\displaystyle _A^B}(E_1dtP_1dx)`$ (21)
$`=`$ $`{\displaystyle _B^N}E_2𝑑t{\displaystyle _B^N}E_1𝑑t`$
$`=`$ $`{\displaystyle _B^N}(E_2E_1)𝑑x={\displaystyle \frac{\mathrm{\Delta }m^2}{2P}}L=\mathrm{\Phi }^S.`$
We see in this way that the same-momentum condition cancels the space phase in both (20) and (21). However, the time-phase in (20) is twice the value found in (21).
A problem then arises: Why do both the same-energy and the same-momentum prescriptions yield the same practical result? A possible answer is that the time-phase and the space-phase might be equivalent when the null condition is used, as already concluded in Ref. . Somehow, the null condition implies in neglecting the arrival time-difference of the two neutrinos. If this is correct, we should get $`\mathrm{\Phi }^S`$ when computing the phase factor under the assumption of the same-velocity prescription, as in this case no arrival time-difference exists.
### 3.3 On the Same-Velocity Prescription
Although the same velocity descrition is forbidden by the kinematical consideration , here, we still explore in the Minkowski diagram how it arises the standard oscillation phase.
Instead of supposing $`E_1=E_2`$ or $`P_1=P_2`$, let us suppose the same-velocity prescription for the neutrinos motion : $`v_1`$ = $`v_2`$ ($`\gamma _1`$ = $`\gamma _2`$). We have in this case
$$\frac{m_2}{m_1}=\frac{E_2}{E_1}=\frac{P_2}{P_1},$$
(22)
and
$$\frac{P_1}{E_1}=\frac{P_2}{E_2}.$$
(23)
The phase difference can be computed along the world line shown in Fig.1. The result is
$`\mathrm{\Delta }\mathrm{\Phi }`$ $`=`$ $`{\displaystyle _A^D}(E_2dtP_2dx){\displaystyle _A^D}(E_1dtP_1dx)`$ (24)
$`=`$ $`{\displaystyle _B^D}{\displaystyle \frac{(P_2P_1)}{v_o}}\left({\displaystyle \frac{dx}{v_o}}\right){\displaystyle _A^B}(P_2P_1)𝑑x`$
$`=`$ $`{\displaystyle \frac{(P_2P_1)}{\gamma ^2v_o^2}}L={\displaystyle \frac{(P_2^2P_1^2)}{\gamma ^2v_o^2(P_2+P_1)}}L`$
$`=`$ $`{\displaystyle \frac{\mathrm{\Delta }m^2}{(P_2+P_1)}}L\mathrm{\Delta }\mathrm{\Phi }^S.`$
Like the same-energy and the same-momentum prescriptions, the same-velocity prescription gives exactly the value of the standard treatment. We conclude in this way that the confusion on the understanding of the neutrino phase factor has its origin in the arrival-time difference. The null condition of the standard treatment includes the information of the same speed of the two neutrinos.
## 4 geodesic phase and null phase in the curved spacetime
As discussed in the flat spacetime, calculating the phase along the geodesic will produce a factor of 2, now we can also obtain this conclusion in the general curved spacetime, not only in the Schwarzschild spacetime . The velocity of an extremely relativistic neutrino is nearly the speed of light in the curved spacetime. Although seemingly irrelevant to think about the difference between the geodesic and the null, this tiny deviation becomes important for the understanding of the neutrino oscillation. Motivated by this argument, we will compare the neutrino phase when calculated along the geodesic and along the null-line. With this, we will be able to verify the factor of 2 when the null is replaced by the geodesic. This study can be shown to remain valid in the case of the flat spacetime.
Let $`n^\mu `$ and $`\stackrel{}{n}^\mu `$ be the tangent vectors to the geodesic and to the null-line, respectively, their difference $`ϵ^\mu `$ being a small quantity for the case of an extremely relativistic neutrino. Here, we suppose that the two neutrinos, the massless and massive, start their journey at the same initial spacetime position A, and their space routes are almost the same. But, there will be an arrival time-difference at the detector position B. This means that their 4-dimensional spacetime trajectories are not the same, and consequently the tangent vectors will present a small difference. Thus, we have
$$n^\nu =\stackrel{}{n}{}_{}{}^{\nu }+ϵ^\nu ,\text{or}P^\nu =\stackrel{}{P}{}_{}{}^{\nu }+mϵ^\nu ,$$
(25)
where $`P^\nu =mn^\nu `$ ($`\stackrel{}{P}{}_{}{}^{\nu }=m\stackrel{}{n}^\nu `$) is the 4-momentum along the geodesic (null-line) with
$$\stackrel{}{n}{}_{}{}^{\nu }=\frac{dx^\mu }{d\lambda }=\frac{d\stackrel{}{x}^\nu }{ds}$$
(26)
and
$$n^\mu =\frac{dx^\mu }{ds}.$$
(27)
In these expressions, $`\lambda `$ and $`s`$ are respectively affine parameters along the null and the geodesic lines. These two tangent vectors satisfy the mass shell relations of the geodesic and the null-line:
$$g_{\mu \nu }n^\mu n^\nu =1$$
(28)
and
$$g_{\mu \nu }\stackrel{}{n}{}_{}{}^{\mu }\stackrel{}{n}{}_{}{}^{\nu }=0.$$
(29)
Now, substituting (25) into (28), we obtain
$$g_{\mu \nu }(\stackrel{}{n}{}_{}{}^{\mu }+ϵ^\mu )(\stackrel{}{n}{}_{}{}^{\nu }+ϵ^\nu )=1,$$
(30)
or, by using (29),
$$2g_{\mu \nu }\stackrel{}{n}{}_{}{}^{\mu }ϵ_{}^{\nu }+𝒪(ϵ^2)=1.$$
(31)
We can estimate the order of $`\{n^\mu \}`$ and $`\{\stackrel{}{n}{}_{}{}^{\mu }\}`$ by noting that $`nϵ1/2`$, which implies that $`ϵn^1\frac{m}{E}`$, where $`E=P^oP^i`$ $`(i=1,2,3)`$ for a relativistic neutrino.
The neutrino phase induced by the null condition, as in the standard treatment, comes from the 4-momentum $`P^\nu `$ defined along the geodesic line, and the tangent vector $`\{\stackrel{}{n}{}_{}{}^{\mu }\}`$ to the null-line . We notice that, if the 4-momentum $`\stackrel{}{P}^\nu `$ defined along the null-line was instead used to compute the null phase, we would obtain zero because of the null condition. Therefore, the phase along the geodesic line (geodesic phase) and the phase along the null-line (null phase) can be written respectively as
$$\mathrm{\Phi }(\mathrm{geod})=m𝑑s=g_{\mu \nu }P^\mu n^\nu 𝑑s,$$
(32)
and
$$\mathrm{\Phi }(\mathrm{null})=g_{\mu \nu }P^\mu \stackrel{}{n}{}_{}{}^{\nu }ds.$$
(33)
Therefore, the difference between the geodesic phase and the null phase, by using Eq.(31), is
$`\mathrm{\Phi }(\mathrm{geod})`$ $``$ $`\mathrm{\Phi }(\mathrm{null})={\displaystyle }g_{\mu \nu }P^\mu (n^\nu \stackrel{}{n}{}_{}{}^{\nu })ds`$
$`=`$ $`{\displaystyle g_{\mu \nu }P^\mu ϵ^\nu 𝑑s}={\displaystyle g_{\mu \nu }}\stackrel{}{P}{}_{}{}^{\mu }ϵ_{}^{\nu }ds+𝒪(ϵ^2)`$
$`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle m𝑑s}+𝒪(ϵ^2)={\displaystyle \frac{1}{2}}\mathrm{\Phi }(\mathrm{geod})+𝒪(ϵ^2),`$
that is
$$\mathrm{\Phi }(\mathrm{geod})=2\mathrm{\Phi }(\mathrm{null})+𝒪(ϵ^2).$$
(34)
This conclusion, valid for a general curved spacetime, is similar to that found in in a Schwarzschild spacetime. Concerning the Schwarzschild spacetime, Bhattacharya et al have the following argument for the factor of 2. As the neutrino energy is fixed, but the masses are different, if an interference is to be observed at the same final spacetime point B$`(r_B,t_B)`$, the relevant components of the wave function could not both have started at the same initial spacetime point A$`(r_A,t_A)`$ in the semiclassical approximation. Instead, the lighter mass (hence faster moving) component must either have started at the same time from a spatial location $`r<r_A`$, or (what is equivalent) started from the same location $`r_A`$ at a later time $`t_A+\mathrm{\Delta }t`$. Hence, there is already an initial phase difference between the two mass components due to this time gap, even before the transport from $`r_A`$ to $`r_B`$ which leads to the phase $`\mathrm{\Phi }(\mathrm{null})`$, i.e., the additional initial phase difference may be taken into account .
## 5 discussions and conclusions
As mentioned above, the same-energy and the same-momentum prescriptions in calculating the mass neutrino interference phase will obtain the same result if the arrival time difference is taken into account, which will be, mathematically, equivalent to the phase of a treatment by using the same-velocity condition. However, we remark that generally neither energy nor momentum are equal in the factual physical process , the standard treatment of the same-energy and the same-momentum prescriptions of massive neutrino phases is just a mathematical simplification in some sense when applying the plane wave approximation. Then our analysis of the massive neutrino standard phases in the Minkowski diagram indicates that the same phase factor will be obtained if the two massive neutrinos follow the null line or possess the same velocity. If the phase calculated along the particle world line, the realtive phase will produce a factor of 2 , the reason is that we despise the arrival time difference of the two mass neutrinos, which results in a double counting effect. This conclusion is correct in both flat spacetime and the curved spacetime. Further, for the better understanding of the mass neutrino interference, we should consider the neutrino mixing state as a wave-packet , or a relativistic quantum ball. According to this scheme, two physical aspects must be considered. On one hand, there is the classical trajectory of the ball which is defined by giving the initial conditions in the classical sense. This motion is relativistic, leading thus to the relativity of the simultaneity as the velocities of the different mass neutrinos are not the same. The classical trajectory, therefore, is defined in terms of macroscopic quantities of the packet, the quantum average of quantities, in accordance with the correspondence principle of quantum mechanics. In the classical sense, the average mass and velocity of the packet ball define its orbit. The distance from the source to the detector is usually measured accurately, which is a classical measurement with a macroscopic precision. On the other hand, there is the quantum dynamics of the packet, which is a two-state system like a neutral kaon or a B meson. This fuzzy dynamics takes place in the microscopic scale $`d`$, the characteristic dimension of the ball, which is small compared to the dimension $`L`$ of the macroscopic trajectory. This quantum dynamics, that is, the internal evolution of the packet, is well described and understood in the particle physics context.
As a final comment, we would like to argue that, if the individual mass-eigenstate is supposed to have a well defined classical velocity, its orbit would be well defined at any spacetime position. This, however, is not in accordance with the quantum interference description, which consequently would never occur. This means that the quantum fuzzy is necessary for the wave-packet, a well defined orbit being valid only for the classical (macroscopic) quantum average. According to this point of view, two basic points concerns the description of the neutrino oscillation. For the kinematics, we use the plane-wave description with a Lorentz invariant phase. For the dynamics, we use the wave-packet and fuzzy quantum path. We use, therefore, two concepts for the neutrinos, particle and wave, classical and quantum. The neutrino is described in terms of a spinor wave-packet, and the spacetime translation induces the spinor to precess in the state vector-space. In other words, we take the mixing state as a classical object propagating along the classical world-line with a well defined velocity — the group velocity. The individual mass-eigenstates propagate with a phase velocity, and only the relative phase velocity will produce the realistic interference phenomenon.
## Acknowledgments
Thanks are due to the scientific visiting support by ICTP, Trieste, Italy. A.B. is supported by NRF of South Africa. Discussions with A. Smirnov are highly appreciated. The authors are grateful to the modification suggestions from the anonymous referee.
|
warning/0004/astro-ph0004078.html
|
ar5iv
|
text
|
# The Origin of Star Formation Gradients in Rich Galaxy Clusters
## 1. Introduction
Extensive observational work has established that the star formation properties of galaxies in rich clusters differ significantly from those of field galaxies (e.g., Osterbrock (1960); Dressler et al. (1985); Balogh et al. (1997); Koopmann and Kenney (1998)). In the field, galaxies form stars at rates several times higher than systems of similar luminosity in the cores of clusters. This is partly a result of the well-known morphology-density relation, since ellipticals and S0 galaxies are more abundant in clusters (Dressler (1980); Whitmore et al. (1993)), but there is evidence that even later type galaxies in clusters form stars at lower rates then in the field. For example, Balogh et al (Balogh et al. (1999)) report that field galaxies of given bulge-to-disk ratio and luminosity have, on average, much larger \[OII\] equivalent widths than their counterparts in rich clusters, suggesting that the cluster environment somehow curbs the star formation rates of all galaxies, regardless of morphology.
Various physical mechanisms have been proposed to explain this and other systematic differences between the field and cluster galaxy populations. Ram pressure stripping by the intracluster medium, for example, has been suggested as a means of removing the gaseous component of disk galaxies and of dramatically altering their morphology and subsequent star forming history (e.g., Gott and Gunn (1972); Fujita and Nagashima (1999); Abadi et al. (1999)). Tides, either by the main cluster potential (e.g. Byrd and Valtonen (1990); Fujita (1998)) or by fly-by encounters with other cluster galaxies (“galaxy harassment”, Moore et al. (1996)), have also been proposed to explain the observations.
Although the above processes are all plausible, the actual changes in star formation rate induced by them remain a matter of debate. For example, some authors have suggested that the cluster environment triggers intense bursts of star formation that rapidly consume the gas of an infalling cluster galaxy (Dressler and Gunn (1983); Barger et al. (1996); Poggianti et al. (1999)), while others have favored a scenario where star formation is truncated in a galaxy almost immediately after it is accreted into the cluster (Abraham et al. (1996); Newberry et al. (1990); Morris et al. (1998); Jones et al. (2000)).
The virtues of these scenarios can in principle be tested observationally, since they are expected to produce a population of galaxies with unusually strong Balmer absorption lines in which star formation has been recently terminated (e.g., Dressler and Gunn (1983); Couch and Sharples (1987)). Recently, Balogh et al. (Balogh et al. (1999)) have applied this idea to galaxies in the CNOC1 survey of X-ray luminous clusters (Yee et al. (1996)). From the paucity of galaxies with strong H$`\delta `$ absorption in their spectra they conclude that the decline of star formation in these clusters may actually be a fairly gradual process. This is in contradiction to other authors’ conclusions, which are derived from datasets with relatively high numbers of H$`\delta `$-strong objects (e.g. Barger et al. (1996); Poggianti et al. (1999)); whether this discrepancy is due to selection effects or to real differences in the clusters studied has not yet been fully elucidated.
A natural timescale for a gentler reduction in star formation rate may be gleaned from the long–known observation that, given their present disk gas content, normal field spirals currently form stars at rates that cannot be sustained over a Hubble time. At face value, it appears that most spirals would use up their disk gas supply in a few Gyrs (e.g., Gallagher et al. (1989)), though this timescale may be considerably extended for some choices of the IMF and if the effects of non-instantaneous mass recycling is taken into account (Kennicutt et al. (1994)). Star formation lifetimes can be significantly longer if galaxies continuously accrete fresh star formation fuel from their surroundings (e.g., Blitz et al. (1999)). If, as proposed by Larson, Tinsley & Caldwell (Larson et al. (1980)), this extended reservoir is stripped off a galaxy when it first enters the cluster, its star formation rate may decay significantly within a few Gyr, leading to large differences in the cluster and field populations. In clusters that are built hierarchically (as in the “bottom-up” scenario favored by cold dark matter cosmogonies) this mechanism would also establish a radial gradient in the star formation properties of cluster galaxies, reflecting the relation between the clustercentric radius of a galaxy and the time of its accretion into the cluster. This is an important ingredient in the success of semianalytic models, which have been shown to match global cluster properties such as the morphological composition and the blue galaxy fraction (Baugh et al. (1996); Kauffmann and Charlot (1998)).
We explore here a simple model based on this interpretation where the mass accretion history of a cluster obtained from numerical simulations of a universe dominated by cold dark matter is coupled with a simple star formation prescription for galaxies following accretion. Once calibrated to reproduce the properties of local field galaxies, the model has no free parameters and its results can be compared directly with observations. We shall focus our analysis on a quantitative discussion of the clustercentric radial gradients in star formation properties and galaxy colors expected in this model.
In §2 we describe the observational dataset we use, selected from the CNOC1 survey. The numerical simulations used to derive mass accretion histories of different clusters are described in §3. Our model prescriptions are presented in §4 and our results in §5. We discuss the implications of our results in §6, and list our conclusions in §7.
## 2. Observational Dataset
We use in this paper the CNOC1 cluster redshift survey dataset (Yee et al. (1996)), which consists of CFHT spectra for $`2000`$ galaxies in 15 X–ray luminous clusters at $`0.19<z<0.55`$. The observational selection effects of this survey are well understood and are discussed in detail in Yee, Ellingson & Carlberg (Yee et al. (1996)) and Balogh et al. (Balogh et al. (1999)). For the present analysis, we weigh the raw data by three factors: one to account for the primary selection effect due to source magnitude, and two secondary corrections which depend on the galaxy color and position on the CCD. One of the main advantages of this survey, especially for the purposes of the present study, is that the dataset includes spectra of foreground and background field galaxies projected onto each cluster. Since both field and cluster galaxies are selected using the same criteria, relative differences between the two samples are rather insensitive to uncertainties in the procedure used to correct for selection effects.
Since our analysis concentrates on radial gradients within clusters, we only consider those clusters with well defined centers in position and velocity space and thus exclude from the sample the bimodal clusters MS0906+11 and MS1358+62 (Carlberg et al. (1996)). We also restrict the redshift range of the galaxy sample to $`0.19<z<0.45`$, in order to facilitate comparisons with simulations analyzed at a single epoch and to minimize effects due to global changes in the galaxy population as a function of redshift. This effectively removes two more clusters from the CNOC1 sample, MS0016+16 and MS0451-03. The final sample has twelve clusters in total, which are scaled and co-added together to construct a “fiducial cluster” sample where effects due to substructure and asphericity of individual clusters are minimized (Carlberg et al. (1997)). The full procedure is described in detail in Balogh et al (Balogh et al. (1997, 1999)).
In brief, we use the mass models of Carlberg, Yee & Ellingson (Carlberg et al. (1997)) to divide the sample into a cluster and field sample; galaxies are deemed to be cluster members if they are within 3$`\sigma `$ of the (radially dependent) cluster velocity dispersion, and field members if they are beyond 6$`\sigma `$. Cluster galaxy positions are all measured relative to the brightest cluster galaxy (BCG), and normalized to $`R_{200}`$, the radius at which the mean inner density is 200 times the critical density. The BCGs themselves are omitted from the final sample, as they are likely to have a unique formation history which may differ from the general cluster population; we briefly compare their properties with those of the full sample in §5.1. Finally, we impose an absolute magnitude limit on the sample, considering only galaxies brighter than $`M_r=18.5+5\mathrm{log}h`$ at $`z=0`$ (Gunn-$`r`$, $`q_0=0.1`$); when appropriately weighted, this sample is statistically complete. Because each individual cluster is at a different redshift a small evolutionary correction is applied to this cutoff assuming that luminosity increases in direct proportion to $`(1+z)`$ (e.g., Lin et al. (1999)). At $`z0.3`$, the luminosity cutoff we adopt is therefore $`M_r=18.8+5\mathrm{log}h`$. This correction has little effect on our results because of the narrow redshift range under consideration. For the sample considered here, the luminosity function of the cluster population is similar to that of the field galaxy population.
Individual star formation rates (SFRs) for galaxies in the sample are computed from the rest frame equivalent width of the \[OII\]$`\lambda `$3727 emission line and the rest frame B–band luminosity relative to solar ($`L_B/L_{B,}`$) using the relation,
$$\dot{M_{}}=3.4\times 10^{12}\left(\frac{L_B}{L_{B,}}\right)W_{}(\mathrm{OII})\mathrm{E}(\mathrm{H}\alpha )\mathrm{M}_{}\text{ }\mathrm{yr}^1.$$
(1)
Here E(H$`\alpha `$) is the extinction at H$`\alpha `$ which, following Kennicutt (Kennicutt (1992)), we take to be one magnitude. The coefficient in Equation 1 has been chosen so that, on average, the SFRs of field galaxies are consistent with their colors and luminosities, based on models of their star formation history that are discussed in detail in §5.2. This coefficient is consistent with that empirically determined by Barbaro & Poggianti (Barbaro and Poggianti (1997)), and about 30% larger than that measured by Kennicutt (Kennicutt (1992)) — well within the uncertainty of its determination (see Kennicutt Kennicutt (1998)). Because we are concerned with relative differences, neither this normalization nor the extinction correction has a significant effect on our conclusions, unless these quantities vary dramatically from the cluster to the field.
Uncertainties in the equivalent widths have been assessed assuming Poisson statistics, and have been internally calibrated to account for additional systematic errors, as described in Balogh et al. (Balogh et al. (1999)). We exclude from the sample all galaxies with very large $`W_{}(`$OII$`)`$ uncertainties, $`\mathrm{\Delta }`$ $`W_{}(`$OII$`)`$$`>15`$Å ($`6\%`$ of the sample), and all galaxies for which \[OII\]$`\lambda `$3727 lies outside the observed spectral range. The final sample with measured SFRs consists of 556 cluster galaxies and 339 field galaxies.
The $`W_{}(`$OII$`)`$ measurements from which we derive SFRs are computed by adding up the observed flux (accounting for partial pixels) above the continuum level in the wavelength range $`3713<\lambda /`$Å$`<3741`$. The continuum level is estimated by fitting a straight line to the flux in the ranges $`3653<\lambda /`$Å$`<3713`$ and $`3741<\lambda /`$Å$`<3801`$. For weak or absent \[OII\] features, the $`W_{}(`$OII$`)`$ index will be sensitive to features in the continuum in these two regions. A crude estimate of uncertainties introduced in the $`W_{}(`$OII$`)`$ measurements by these features may be obtained by using eq. 1 to compute SFRs of galaxies expected to have little or no ongoing star formation; these are red cluster galaxies with large 4000Å breaks ($`(gr)_{}>0.35`$, $`D_{4000}>1.8`$) found within $`0<R/R_{200}<0.3`$. The 3$`\sigma `$–clipped mean SFR of this population is -0.057 $`h^2M_{}\text{yr}^1`$, and the standard deviation is $`0.156h^2M_{}\text{yr}^1`$. This small systematic offset and uncertainty are taken into account in our modeling, as described in §4.2.
## 3. Numerical Simulations
Cluster mass accretion rates are computed directly from N–body simulations of the formation of six massive clusters ($`0.7<M/(10^{15}M_{})<2.3`$) in a COBE-normalized, $`\mathrm{\Lambda }=0.7`$, $`\mathrm{\Omega }_0=0.3`$ cosmology. The simulations are similar to those described in detail by Eke, Navarro, & Frenk (Eke et al. (1998)). We assume that “light traces mass” in the clusters and identify statistically each dark matter particle with a “galaxy”. The virial radius, $`R_{\mathrm{vir}}`$, of a cluster is computed using the overdensity prescription described in Eke, Cole & Frenk (Eke et al. (1996)), which differs slightly from $`R_{200}`$. At $`z=0.3`$, the mean redshift of our cluster sample, $`R_{\mathrm{vir}}1.2R_{200}1.4`$-$`2.4`$ Mpc. Each cluster has about $`9,000`$ particles within $`2R_{\mathrm{vir}}`$ at $`z=0.3`$. The observations and model parameters are all presented in terms of the simulation cosmology; we use $`H_0=70`$ km s<sup>-1</sup> Mpc<sup>-1</sup>, which gives a present age of the universe of 13.5 Gyr. At $`z=0.3`$ the universe is $`11`$ Gyr old.
## 4. The Model
As discussed in §1, our modeling assumes that cluster galaxies differ from their field counterparts in their star forming properties because they are stripped of their surrounding gas reservoirs as they are accreted into the cluster. Star formation in cluster galaxies thus declines after accretion as more and more of the galaxy’s remaining gas content gets turned into stars.
The model involves the following steps:
(i) All particles within $`2R_{200}`$ from the center of each simulated cluster at $`z0.3`$ are selected.
(ii) Each particle is traced back in time to find out when it was first accreted into the cluster. Two definitions of “accretion time”, $`t_{\mathrm{acc}}`$, are used. The first ($`t_{\mathrm{acc}}=t_{\mathrm{cluster}}`$) is defined to be the time when the particle (“galaxy”) first finds itself within $`R_{\mathrm{vir}}`$ from the center of the current most massive progenitor of the cluster, and the second ($`t_{\mathrm{acc}}=t_{\mathrm{group}}`$) is the time when a galaxy is first accreted into any large clump, not necessarily the most massive one. Details of this procedure are given in §4.1.
(iii) A SFR is chosen for each galaxy at $`t=t_{\mathrm{acc}}`$. For simplicity, these are taken at random (using appropriate weights as discussed in §2) from the distribution of SFRs computed for $`z0.3`$ field galaxies in our sample. This assumes implicitly that field SFRs do not evolve with time, arguably the simplest possible model. We explore in §5.2 the consequences of relaxing this assumption.
(iv) We use a simple gas consumption model to estimate SFRs following accretion and to compute the SFR of each cluster galaxy at $`z0.3`$. Details are discussed in §4.2.
(v) Each simulated cluster is projected onto three orthogonal planes. Field contamination in the observational sample is accounted for as discussed in §4.3.
This procedure uniquely defines a final ($`z=0.3`$) SFR for each galaxy in the simulated clusters; the mean SFR can then be computed as a function of clustercentric distance, and compared with the field galaxy observations. We discuss now the various steps of the model in some detail.
### 4.1. Accretion Times
We have explored two ways of assigning “accretion times”, $`t_{\mathrm{acc}}`$, to galaxies in the simulated clusters. One choice defines $`t_{\mathrm{acc}}=t_{\mathrm{cluster}}`$ as the time when a particle is first found within the virial radius of the most massive progenitor present at that time. Following Eke, Cole & Frenk (Eke et al. (1996)), the virial radius, $`R_{\mathrm{vir}}`$, is defined as the radius where the mean inner density of the cluster is $`\overline{\rho }(R_{\mathrm{vir}})=\mathrm{\Delta }_c(z)\rho _c(z)`$, where $`\rho _c(z)`$ is the critical density at the redshift $`z`$, and $`\mathrm{\Delta }_c(\mathrm{\Omega }(z))`$ is the “critical” overdensity for spherical collapse (which takes the familiar value of 178 for $`\mathrm{\Omega }=1`$ models).
The above definition of $`t_{\mathrm{cluster}}`$ may not be completely appropriate, since in a hierarchically clustering universe particles may be first accreted into another “protocluster” before being accreted into the main progenitor of the final cluster. Therefore, we consider as an alternative definition of $`t_{\mathrm{acc}}`$ the time when a particle first finds itself associated with a clump with circular velocity exceeding $`V_c=500`$ km s<sup>-1</sup>. (This circular velocity corresponds to a virial temperature of $`0.8`$ keV). Particles are associated with clumps via a friends-of-friends algorithm, with an evolving linking length parameter taken to be $`10\%`$ of the mean inter-particle separation at each time. Accretion times defined this way are labeled $`t_{\mathrm{group}}`$. We shall see in §5.1 that our results are rather insensitive to the particular choice of accretion time definition.
The “resolution” of $`t_{\mathrm{acc}}`$ is limited by the number of times particle positions are output by the simulations. In this case, we have outputs every 1.34 Gyr; we therefore add a uniform random number between 0 and 1.34 Gyr to each particle’s $`t_{\mathrm{acc}}`$ to smooth out this resolution. We do not expect our results to be sensitive to this resolution, since we are combining the results of six clusters, which will smooth out the effect of discrete merger events.
### 4.2. Star Formation Prescription
For normal, field spiral galaxies, the SFR per unit area averaged over the disk, $`\mathrm{\Sigma }_{\mathrm{SFR}}`$, depends on the disk gas content in a manner well approximated by a Schmidt law (Schmidt (1959)). From Kennicutt (Kennicutt (1998)), this relation is given by
$$\mathrm{\Sigma }_{\mathrm{SFR}}=(2.5\pm 0.7)\times 10^4\left(\frac{\mathrm{\Sigma }_{\mathrm{gas}}}{M_{}\text{pc}^2}\right)^NM_{}\text{yr}^1\text{kpc}^2,$$
(2)
where $`\mathrm{\Sigma }_{\mathrm{gas}}`$ is the average gas surface density, and the exponent $`N=1.4\pm 0.15`$. Converting surface densities to integrated values using the source diameters tabulated by Kennicutt, galaxies that evolve according to this relation and accrete no extra gas would form stars at a steadily decreasing rate given by,
$$\dot{M_{}}(t^{})=\dot{M_{}}(0)\left(1+0.33\frac{t^{}}{t_e}\right)^{3.5}M_{}\text{yr}^1,$$
(3)
where $`\dot{M_{}}(0)`$ is the initial SFR and $`t_e1.48(\dot{M_{}}(0)/M_{}\mathrm{yr}^1)^{0.29}`$ Gyr is the characteristic gas consumption timescale. As discussed in §4, we adopt $`\dot{M_{}}(0)`$ values taken at random from the measured field SFRs and set $`t^{}=tt_{\mathrm{acc}}`$. With these assumptions it is possible to compute SFRs of cluster galaxies at $`z0.3`$, the mean redshift of the observational sample. In the above calculation, we have neglected the effects of gas recycling. If we assume instantaneous recycling, where a fraction $`R`$ of every solar mass of stars formed is returned as gas, the timescale $`t_e`$ is increased by a factor $`1/(1R)`$. Assuming $`R=0.33`$ (Kennicutt et al. (1994))<sup>1</sup><sup>1</sup>1This large factor of $`R`$ approximates the effects of a non-instantaneous recycling calculation, assuming a Scalo IMF, and is fairly good for slowly evolving disks; for rapidly evolving disks, $`R`$ can be as high as 0.75., we obtain $`t_e2.2(\dot{M_{}}(0)/M_{}\mathrm{yr}^1)^{0.29}`$ Gyr. This increase does not have a significant effect on the results we present, so we neglect the effects of recycling throughout this work.
Finally, we correct the values obtained from eq. 3 for the systematic offset and uncertainty in the SFR measurements discussed in §2. This is done by adding a random number, drawn from a Gaussian distribution with a mean of $`0.12M_{}`$ yr<sup>-1</sup> and a variance of $`0.32M_{}`$ yr<sup>-1</sup>, to the model SFR. These parameters were chosen from the SFR distribution of red, central cluster galaxies (with $`h=0.7`$), as described in §2.
### 4.3. Field Contamination
Observational cluster datasets are contaminated by field galaxies, projected onto the cluster, that have Hubble-flow redshifts similar to the peculiar velocities induced by the cluster potential. Our model takes this into account following the procedure of Carlberg et al. (Carlberg et al. (1997)). In brief, the velocity differences between cluster and individual galaxies in each dataset are scaled to the velocity dispersion of each cluster and their clustercentric projected radii to $`R_{200}`$. These normalized velocities, $`V_{\mathrm{norm}}=\mathrm{\Delta }v/\sigma `$, and radii, $`R_{\mathrm{proj}}/R_{200}`$, are then combined into a single dataset. The density of galaxies per unit $`V_{\mathrm{norm}}`$ at $`5<V_{\mathrm{norm}}<25`$ is computed in radial bins to account for the radially varying $`\sigma `$. Under the assumption that this density remains constant within each bin, it is used to calculate the expected number of interloper galaxies, i.e., those within 3$`\sigma `$ in each radial bin. This provides a direct estimate of the fraction of galaxies deemed cluster members that are actually field galaxies projected onto the cluster in “redshift space”. This fraction is typically $`1\%`$ in the central cluster regions, but may be as large as $`12\%`$ at $`R_{200}`$. Our modeling accounts for this effect by including an appropriate number of “field” galaxies (randomly selected from the field sample) in all computations.
## 5. Results
### 5.1. Non-evolving field SFR model
Figure 1 shows the mean SFR per galaxy as a function of projected distance from the cluster center, normalized to $`R_{200}`$. The solid squares correspond to the CNOC1 data for the complete sample of galaxies brighter than $`M_r=19.6`$ (at $`z=0.3,h=0.7`$), averaged over radial bins and with 1-$`\sigma `$ jackknife error bars. The mean SFR per galaxy increases systematically from the center of the cluster outwards, from almost zero near the center to about $`0.7M_{}`$ yr<sup>-1</sup> in the outskirts of the cluster. We note that the absolute SFR values quoted here are sensitive to a number of sample selection parameters; in particular to the luminosity cutoff and the uncertain coefficient in eq. 1, so care must be exercised when comparing these results to other work. On the other hand, relative differences between cluster and field populations are robust, since the two samples have similar luminosity functions, and are drawn from the same survey with identical selection criteria.
As shown in Figure 1, the average SFR per galaxy in the field is significantly higher than in clusters. Remarkably, even at radii as far from the center of the cluster as $`2R_{200}`$, cluster star formation rates remain depressed by almost a factor of two relative to the field.<sup>2</sup><sup>2</sup>2Because the luminosity functions of the field and cluster population considered in this sample are quite similar, the same result is obtained for the SFR per unit luminosity as for the SFR per galaxy.. We note here that the BCGs, which are omitted from the data sample, have unusually high SFRs; the mean SFR of the 11 BCGs which satisfy our selection criteria<sup>3</sup><sup>3</sup>3No spectrum is available for the BCG in MS 0451.5+0250. is 14 $`M_{}`$ yr<sup>-1</sup>, much greater than the mean field value. Only four of these eleven galaxies have no significant \[OII\] emission. The low redshift BCGs are discussed in more detail in Davidge & Grinder (Davidge and Grinder (1995)); we do not consider these galaxies further in this study.
Figure 1 also shows, with open symbols, the results of the modeling procedure outlined in §4. Open squares and triangles correspond to the two different accretion time definitions discussed in §4.1. Error bars represent the 1-$`\sigma `$ variance of the 18 numerical realizations (6 simulated clusters and 3 orthogonal projections per cluster). The agreement between the model and the observations is remarkable, especially considering that the modeling involves no free parameters. The model reproduces the observed SFR gradient and even the observed depression, relative to the field, of SFRs outside $`R_{200}`$.
The latter result is somewhat surprising, since in spherical accretion models particles outside $`R_{200}`$ are infalling into the cluster for the first time (Bertschinger (1985); White et al. (1993)) and therefore their SFRs have yet to feel the effects of the cluster environment. The main reason for our result is that a substantial fraction ($`54\pm 20\%`$ for the six clusters we studied) of particles between $`1`$ and $`2R_{200}`$ have actually been inside the virial radius of the main progenitor at some earlier time. These are often particles that populated the outskirts of recently accreted clumps and that, although still bound to the system, have been scattered to large apocenter orbits during the merger process.
Interestingly, assuming that the onset of the SFR decline occurs when a galaxy is accreted into any clump with circular velocity exceeding $`500`$ km s<sup>-1</sup> rather than the cluster’s main progenitor has only a small effect on this result within $`R_{200}`$ (witness the good agreement between open triangles and squares in Figure 1). Note that the mean SFR beyond $`R_{200}`$ is further suppressed under this assumption, resulting in even better agreement with observations. This is because some particles beyond $`R_{200}`$ are found within fairly massive groups, although they may never have been within the virial radius of the main cluster.
An important feature of this model is that many cluster galaxies at $`z=0.3`$ have substantial SFRs. In particular, near $`R_{200}`$, $`20\%`$ of cluster galaxies have SFRs in excess of 1$`M_{}\text{yr}^1`$, and this declines to about $`10\%`$ at $`R=0.5R_{200}`$; fractions which are consistent with the CNOC1 data. These large SFRs are not the result of cluster-induced starbursts, but correspond to recently accreted field galaxies in which star formation has not yet been completely quenched. In a recent study, Balogh & Morris (Balogh and Morris (2000)) have measured H$`\alpha `$ equivalent widths for galaxies in Abell 2390, and they failed to find a substantial population with large H$`\alpha `$ fluxes that were undetected in \[OII\]. Thus, there does not appear to be a population of dust-obscured starburst galaxies in this cluster which were missed in the CNOC1 survey. Our modeling indicates that cluster-induced starbursts are not necessary to generate the levels of star formation seen in these clusters.
We conclude from this comparison that a model based on the simple assumption that the SFR of a galaxy decreases on gas consumption timescales is able to reproduce the data extremely well, lending support to the underlying hypothesis that continuous accretion of external gas is responsible for maintaining the SFRs of normal spirals over a Hubble time.
### 5.2. Evolving field SFR model
There are two weaknesses in the model explored in the previous sections. One is the assumption that the SFRs of field galaxies, which are used to assign “initial” SFRs to cluster galaxies at accretion time (see eq. 3), do not evolve with time. Given the mounting evidence that the average SFR per unit volume evolves strongly with lookback time (Lilly et al. (1995); Madau et al. (1996)) it is necessary to explore the consequences of relaxing the non-evolving SFR assumption on our results. The second is the luminosity evolution of cluster galaxies, which must also be modeled in order to account for galaxies that may fade beyond the observational magnitude limit when their SFR declines.
A proper treatment of these effects, which must take into account the merger and star formation history of galaxies, is of great interest, but well beyond the scope of this study. In order to at least explore the effects that this more complete treatment will have, we use simple $`\tau `$models to model the SFR evolution of field galaxies in our sample, i.e., by assuming that,
$$\dot{M_{}}(t)=\dot{M_{}}(t_0)\mathrm{e}^{(tt_0)/\tau },$$
(4)
where $`\dot{M_{}}(t_0)`$ corresponds to the SFR of galaxies in the field at $`t=t_011`$ Gyr, the age of the universe at $`z0.3`$. Once $`\tau `$ is determined for each galaxy it is possible to construct, at arbitrary times, an “evolving” field SFR distribution which matches the observations at $`z0.3`$. For a given IMF, and constant reddening, $`\tau `$ is uniquely determined by the observed $`(gr)_{}`$ color <sup>4</sup><sup>4</sup>4 This color is available for all galaxies in the CNOC1 dataset. We have made a small adjustment to the zero point of the observed $`(gr)_{}`$ colors, reducing them by $`0.05`$ so that the reddest observed galaxies have the colors predicted by a model in which all stars formed at $`t=0`$ in an instantaneous burst. of the galaxy, as shown in Figure 2. We have used the PEGASE (Fioc and Rocca-Volmerange (1997)) spectrophotometric models with a Salpeter IMF (with lower and upper limit given by $`0.1<M/M_{}<120`$) to determine $`\tau `$ for each galaxy in our field sample. The bluest galaxies ($`(gr)_{}<0.04`$) have $`\tau <0`$, corresponding to a SFR that increases with time, while moderately blue galaxies ($`(gr)_{}0`$) have $`\tau `$ approaching (positive or negative) infinity, corresponding to a constant SFR. The reddest field galaxies, those with $`\tau <300`$ Myr, are assumed to have formed in a single burst at high redshift, and are modeled as such in the cluster. The net result is a rather extreme model where the global star formation rate in the field population increases steadily out to $`z10`$. We explore below as well the consequences of restricting this increase at large $`z`$, and conclude that our results are quite insensitive to the precise nature of the redshift evolution in the field.
The normalization constant of eq. 4, $`\dot{M_{}}(t_0)`$, may in principle be determined by the observed field galaxy SFR at $`t_0`$, but we found that, because of the large uncertainties in individual SFR determinations, this procedure leads to large discrepancies in the total luminosity of galaxies in the sample once the SFRs are integrated over time. Because total luminosities are considerably better determined than SFRs in our dataset, and are less sensitive to burst–like events that eq. 4 cannot reproduce, we compute $`\dot{M_{}}(t_0)`$ for each field galaxy by requiring the model to reproduce its observed total Gunn-$`r`$ luminosity at $`z=0.3`$. This choice implies that, on a galaxy by galaxy basis, the model $`z=0.3`$ SFRs no longer correspond to those derived from the $`W_{}(`$OII$`)`$ measurements. The observed and model field SFR distributions are compared in the bottom right panel of Figure 3; where the coefficient in eq. 1 ($`3.4\times 10^{12}`$) has been chosen so that both distributions have identical averages. Although the difference is statistically significant, the two distributions are qualitatively fairly similar, indicating that, on average, the simple $`\tau `$-model provides a reasonable description of the $`z=0.3`$ field. This consistency allows us to compare model and observed SFRs, even though the former is based on galaxy color, while the latter is determined from nebular emission.
A further advantage of modeling the SFR history of the field population is that it allows us to model the color evolution of cluster galaxies. Color ($`(gr)_{}`$) distributions of model galaxies (which include the addition of a Gaussian random number to account for the $`0.02`$ magnitude error typical of the CNOC1 data) are compared with the observational data in the left panels of Figure 3. The model shown adopts $`t_{\mathrm{acc}}=t_{\mathrm{group}}`$, and excludes galaxies fainter than $`M_r=19.6`$ at $`z=0.3`$. This model is very successful at producing a cluster core dominated by red galaxies from an initial field galaxy sample that has a much broader and bluer color distribution. In the outskirts of the cluster, both the observed and model color distributions show a significant increase in the population of blue galaxies. The qualitative agreement between model and observed SFR and color distributions throughout the cluster is quite remarkable for such a simple prescription.
The right panels of Figure 3 compare the observed SFR distributions with this same model. In agreement with observations, model distributions show a strong reduction in the fraction of galaxies forming stars at rates higher than about $`1M_{}`$ yr<sup>-1</sup> near the cluster center. The agreement between model and observations is quite good, although model galaxies form stars at slightly higher rates in both bins.
The differences between model and observed mean cluster SFRs are clearly apparent in Figure 4, which shows the clustercentric gradient predicted by three variants of the evolving field-SFR model. The open circles correspond to the fiducial model as described above, and are generally larger than the observed cluster mean. The open triangles show how this result changes if we neglect galaxy fading; i.e., when we include all cluster galaxies in the average. This recovers the good match to the data, and shows that it is primarily the luminosity fading of galaxies that affects our result, and not the nature of the redshift evolution itself. To demonstrate this explicitly, we present a model (open squares) in which the field galaxy SFRs are evolved as in the fiducial model but are held constant for $`z>1.5`$. The results obtained with this model do not differ significantly from those of the fiducial model, lending support to our conclusion that our results depend only mildly on the SFR redshift evolution.
We wish to stress that the purpose of the above exercise is to assess the sensitivity of our modeling to various assumptions about the redshift dependence of the SFR of the field galaxy population rather than to build a realistic model of field galaxy evolution. We conclude that the luminosity evolution of the field galaxies has a small, but non-negligible effect on our results. Although it appears that qualitatively our conclusions are safe, it is clear that definitive answers will have to wait for a more realistic modeling of the field galaxy star formation evolution, such as the one implemented in semianalytic models of galaxy formation (e.g. Baugh et al. (1996), Diaferio et al. in preparation).
## 6. Discussion
We present models of the clustercentric dependence of star formation and colors of galaxies in rich clusters, under the following assumptions: (i) the cluster galaxy population is built by the ongoing accretion of field galaxies, (ii) SFRs in field galaxies are sustained by regular accretion of gas from their surroundings, and (iii) reservoirs of fresh star formation fuel are lost as galaxies plunge into the cluster potential.
Within this context, our model provides support for a gradual decline (over a timescale of a few Gyrs) in the star formation rates of cluster galaxies after accretion. Actually, results similar to those presented in the previous section may be obtained if the SFR in all cluster galaxies is assumed to decay exponentially after accretion with fixed timescales $`1t_e3`$ Gyrs. Decline timescales longer than $`3`$ Gyr lead to unacceptably large star formation rates in model clusters at $`z0.3`$. On the other hand, sharp truncation of star formation ($`t_e1`$ Gyr) would result in too little star formation within clusters and, furthermore, would lead to an abundant population of galaxies with strong Balmer lines but no nebular emission lines (K+A galaxies), and these appear to be rare in the very luminous X-ray clusters we study here (Balogh et al. (1999)). Larger fractions of K+A galaxies have been reported in other cluster datasets, in particular by the MORPHS collaboration (Dressler et al. (1999); Poggianti et al. (1999)), but it is still unclear whether this apparent disagreement is a result of the procedure used to select the spectroscopic sample, the effects of dust obscuration, or perhaps a genuine effect of the dependence of SFRs on other cluster properties such as X-ray luminosity or temperature (Balogh et al. (1999)). Further analysis is required in order to assess whether the simple model we propose here is applicable to clusters other than the relatively massive, X-ray luminous systems targeted by the CNOC1 survey.
With this caveat, the success of our model strongly suggests that (i) gradients in galaxy properties arise from gradients in accretion times and (ii) that the cessation of star formation need not take place abruptly to explain the observed SFRs of cluster galaxies. On the other hand, our understanding of the star formation history of cluster galaxies is bound to remain incomplete until the physical mechanism responsible for the decline of star formation in cluster galaxies is fully elucidated. The loss of external gas reservoirs advocated here is attractive, but conclusive observational evidence that such reservoirs exist in isolated field galaxies has been slow to emerge (Benson et al. (1999), but see Blitz et al. (1999)). Other processes that may reduce star formation rates, such as tides, harassment, and ram pressure tripping, operate on similar timescales once a galaxy has been accreted into the cluster and it is therefore unlikely that analysis of the kind we present here will be able to distinguish clearly amongst them.
Another major question that our models do not address is the origin of gradients in galaxy morphology that parallel the color and star formation gradients we focus on (e.g., Dressler (1980); Dressler et al. (1997); Koopmann and Kenney (1998)). The over-representation of ellipticals near the center of X-ray luminous clusters may reflect higher merger rates between nearly equal mass systems in systems that collapse early to form cluster cores, but this is an issue that remains unexplored in our model. The construction of large spectroscopical datasets that probe the dependence of galaxy SFRs, colors and morphologies on cluster properties such as concentration, richness, velocity dispersion and X-ray properties, coupled with numerical and/or semianalytical models that treat self-consistently the accretion history and dynamics (and hydrodynamics) of galaxy and cluster formation, are probably the most promising way to make substantial progress in the subject.
## 7. Conclusions
We present a simple model to account for the systematic differences in the star formation properties of galaxies in CNOC1 clusters and the field. The model assumes that the cluster is built through the accretion of field galaxies whose star formation rates gradually decline after entering the cluster as a result of the removal, through tides or ram pressure, of the gaseous envelopes needed to supply normal spirals with the fuel needed to sustain their present star formation rates over a Hubble time. Once calibrated to reproduce observations of nearby spirals the model has no free parameters.
Using cluster mass accretion rates determined from N–body simulations of cluster formation in a $`\mathrm{\Lambda }`$CDM universe, our model is able to reproduce qualitative and quantitative differences in the mean star formation rates and colors between clusters and the field. The model demonstrates that the origin of radial gradients in these properties is the natural consequence of the strong correlation between radius and accretion times which results from the hierarchical assembly of the cluster. Interestingly, the model also explains why star formation in the outskirts of clusters is found to be almost a factor of two below the field average as far out as twice the virial radius of the cluster. This is a result of cluster members being pushed onto highly eccentric, loosely bound orbits during major merger events. Our results are robust to evolution in the star formation properties of field galaxies, but somewhat more sensitive to the manner in which fading of cluster galaxies is modeled. Realistic modeling of the field population star formation histories is required to fully understand the consequences of this fading.
We conclude that the stripping of extended gaseous reservoirs by the cluster environment and the gradual decline that follows gas consumption is likely to be the main mechanism that differentiates the star formation properties of cluster and field galaxies.
We have made extensive use of the CNOC1 dataset of intermediate–redshift clusters. We are grateful to all the consortium members and to the CFHT staff for their contributions to this project. We thank the referee for a careful reading of the manuscript, and for useful suggestions which improved this paper. In Victoria, MLB was supported by a Natural Sciences and Engineering Research Council of Canada (NSERC) research grant to C. J. Pritchet and by an NSERC postgraduate scholarship. In Durham, MLB is supported by a PPARC rolling grant for extragalactic astronomy and cosmology.
|
warning/0004/astro-ph0004281.html
|
ar5iv
|
text
|
# Oscillations and convective motion in stars with URCA shells
## Oscillations and convective motion in stars with <br>URCA shells
G.S. Bisnovatyi-Kogan<sup>1</sup>
<sup>1</sup> Space Research Institute, Moscow
## Abstract
It is shown that in presence of URCA shell pulsational energy losses due to neutrino emission and nonequilibrium beta heating are much less than energy losses by excitation of short-wavelength acoustic waves. Convective motion in presence of URCA shell is considered, and equations generalizing the mean free path model of convection are derived.
## 2 Introduction
It was suggested in that in the convective region cooling of matter may be enhanced in presence URCA shells appearing when matter contains an isotope with a threshold Fermi energy for an electron capture, coresponding to a density less then the central one. Presence of such isotope leads to existence of a jump in the composition at a density, corresponding to a threshold energy. During convective motion the matter in eddies around this density crosses periodically the boundary. That implies continious beta capture and beta decay in the matter of these eddies.
Because of heating of a degenerate matter due to nonequilibrium beta processes , with account of convective URCA shell different conclusions had been done with respect to stabilizing or destabilizing the carbon burning in the convective degenerate core . Here we calculate damping of stellar oscillations in presence of URCA shell (see ), analyze physical processes in the convective URCA shells and formulate approximate quantitative approach to the solution of this problem.
## 3 Linear oscillations of a slab with a phase transition
Consider the plane-parallel layer in the constant gravitational field with an acceleration $`g`$, with a phase transition at the pressure $`P_{}`$, and polytropic equation of state $`P=K\rho ^\gamma `$, with $`K=K_1`$ at $`P<P_{}`$ and $`K=K_2`$ at $`P>P_{}`$, $`K_1>K_2`$. In static equilibrium $`P=gM\left(1\frac{m}{M}\right),\rho =\left(\frac{gM}{K}\right)^{1/\gamma }\left(1\frac{m}{M}\right)^{1/\gamma }.`$ Here the pressure is continuous, but the density $`\rho `$ has a jump at $`P_{}`$ due to the jump of the constant $`K`$, $`M`$ is the mass of one cm<sup>2</sup> of the slab, $`m`$ is the mass of one cm<sup>2</sup> of the slab under the layer with a Lagrangian coordinate $`x`$, which is related to the density as $`\rho =\left(\frac{\gamma 1}{\gamma K}\right)^{\frac{1}{\gamma 1}}(Cgx)^{\frac{1}{\gamma 1}}`$. In presence of a phase transition we have $`\rho _0=\left(\frac{gM}{K_2}\right)^{1/\gamma },P_0=gM,gx_{}=C_2\frac{\gamma }{\gamma 1}\frac{P_{}}{\rho _2},x_0=\frac{C_1}{g},C_2=\frac{\gamma }{\gamma 1}(gM)^{\frac{\gamma 1}{\gamma }}K_2^{1/\gamma },C_1=C_2+\frac{\gamma }{\gamma 1}\frac{P_{}}{\rho _2}(\lambda 1).`$ Here $`\rho _0`$ and $`P_0`$ are the density and the pressure at the bottom of the slab, $`x_0`$ and $`x_{}`$ are total thickness and thickness of the inner denser phase layer of the slab, $`\lambda =\rho _2/\rho _1`$. The phase transition in the slab happens only if its specific mass $`M>P_{}/g=M_{}`$.
Linear oscillations of the slab are reduced to Bessel equation for perturbations $`\stackrel{~}{P},\stackrel{~}{v}=\stackrel{~}{P}_a,\stackrel{~}{v}_a\mathrm{exp}(i\omega t)`$ with solutions
$$\stackrel{~}{P}_a=A\sqrt{z}J_{\frac{\gamma }{\gamma 1}}(\eta )+B\sqrt{z}Y_{\frac{\gamma }{\gamma 1}}(\eta ),$$
(1)
$$\stackrel{~}{v}_a=\frac{i}{\sqrt{\gamma P_0\rho _0}}z^{\frac{1}{2\gamma }}\left[AJ_{\frac{1}{\gamma 1}}(\eta )+BY_{\frac{1}{\gamma 1}}(\eta )\right],$$
(2)
where for two phases $`\eta _2=\frac{2\gamma }{\gamma 1}\frac{M\omega }{\sqrt{\gamma \rho _0P_0}}z^{\frac{\gamma 1}{2\gamma }},\eta _1=\eta _2\sqrt{\lambda },z_{}=1\frac{m}{M}`$. The frequency of oscillations $`\omega `$ and relations between constants $`A_1,B_1,A_2,B_2`$ are obtained from boundary conditions and from relations on the phase jump . The dispersion equation are obtained analytically for a frozen
$$J_{\frac{\gamma }{\gamma 1}}\left(\mathrm{\Omega }\sqrt{\lambda }z_{}^{\frac{\gamma 1}{2\gamma }}\right)\left[J_{\frac{1}{\gamma 1}}(\mathrm{\Omega })Y_{\frac{1}{\gamma 1}}\left(\mathrm{\Omega }z_{}^{\frac{\gamma 1}{2\gamma }}\right)J_{\frac{1}{\gamma 1}}\left(\mathrm{\Omega }z_{}^{\frac{\gamma 1}{2\gamma }}\right)Y_{\frac{1}{\gamma 1}}(\mathrm{\Omega })\right]$$
(3)
$$\sqrt{\lambda }J_{\frac{1}{\gamma 1}}\left(\mathrm{\Omega }\sqrt{\lambda }z_{}^{\frac{\gamma 1}{2\gamma }}\right)\left[J_{\frac{1}{\gamma 1}}(\mathrm{\Omega })Y_{\frac{\gamma }{\gamma 1}}\left(\mathrm{\Omega }z_{}^{\frac{\gamma 1}{2\gamma }}\right)J_{\frac{\gamma }{\gamma 1}}\left(\mathrm{\Omega }z_{}^{\frac{\gamma 1}{2\gamma }}\right)Y_{\frac{1}{\gamma 1}}(\mathrm{\Omega })\right]=0$$
and equilibrium phase transition
$$J_{\frac{\gamma }{\gamma 1}}\left(\mathrm{\Omega }\sqrt{\lambda }z_{}^{\frac{\gamma 1}{2\gamma }}\right)\left[J_{\frac{1}{\gamma 1}}(\mathrm{\Omega })Y_{\frac{1}{\gamma 1}}\left(\mathrm{\Omega }z_{}^{\frac{\gamma 1}{2\gamma }}\right)J_{\frac{1}{\gamma 1}}\left(\mathrm{\Omega }z_{}^{\frac{\gamma 1}{2\gamma }}\right)Y_{\frac{1}{\gamma 1}}(\mathrm{\Omega })\right]$$
$$\left[\sqrt{\lambda }J_{\frac{1}{\gamma 1}}\left(\mathrm{\Omega }\sqrt{\lambda }z_{}^{\frac{\gamma 1}{2\gamma }}\right)(\lambda 1)\frac{\gamma 1}{2}\mathrm{\Omega }z_{}^{\frac{\gamma 1}{2\gamma }}J_{\frac{\gamma }{\gamma 1}}\left(\mathrm{\Omega }\sqrt{\lambda }z_{}^{\frac{\gamma 1}{2\gamma }}\right)\right]$$
(4)
$$\times [J_{\frac{1}{\gamma 1}}(\mathrm{\Omega })Y_{\frac{\gamma }{\gamma 1}}\left(\mathrm{\Omega }z_{}^{\frac{\gamma 1}{2\gamma }}\right)J_{\frac{\gamma }{\gamma 1}}\left(\mathrm{\Omega }z_{}^{\frac{\gamma 1}{2\gamma }}\right)Y_{\frac{1}{\gamma 1}}(\mathrm{\Omega })]=0.$$
In the limiting case $`m_{}=0`$, when the boundary between phases is on the inner boundary of the layer, and $`z_{}=1`$ the dispersion equation is reduced to $`J_{\frac{1}{\gamma 1}}(\mathrm{\Omega }\sqrt{\lambda })=0`$ for a frozen, and
$$\sqrt{\lambda }J_{\frac{1}{\gamma 1}}(\mathrm{\Omega }\sqrt{\lambda })(\lambda 1)\frac{\gamma 1}{2}\mathrm{\Omega }J_{\frac{\gamma }{\gamma 1}}(\mathrm{\Omega }\sqrt{\lambda })=0$$
(5)
for an equilibrium phase transitions. At $`m_{}M,z_{}0`$ when the level between phases is moving to the outer boundary, we have the dispesion equation $`J_{\frac{1}{\gamma 1}}(\mathrm{\Omega })=0`$ in both cases. Here
$$\mathrm{\Omega }=\frac{2\gamma }{\gamma 1}\frac{M\omega }{\sqrt{\gamma \rho _0P_0}},$$
(6)
To investigate the dependence of oscillattion modes with an ideal phase transition on $`\gamma `$ it is convenient to introduce $`\stackrel{~}{\mathrm{\Omega }}=\frac{\gamma 1}{2\sqrt{\gamma }}\mathrm{\Omega }`$, with the first root $`\stackrel{~}{\mathrm{\Omega }}^2\frac{1}{\lambda 1}`$ at $`\gamma \mathrm{}`$. All other roots tend to infinity at $`\gamma \mathrm{}`$. First three roots of equations (3), (4) are presented in Figs.1,2<sup>1</sup><sup>1</sup>1Numerical solution of dispersion equations had been done by O.V.Shorokhov..
Caption to Figures 1,2,3
Frequencies $`\mathrm{\Omega }`$ of the slab oscillations as functions of $`z_{}`$ of the basic mode (Fig.1), modes with one (Fig.2) and two (Fig.3) nodes. Upper and lower curves correspond to frozen and equilibrium cases, relatively. Frequencies of these two cases coincide, when one of the node coincides with the phase transition.
## 4 Damping of oscillations due to URCA shell in a highly degenerate matter
Consider ultrarelativistic degenerate electron gas a good approximation in most URCA shells. It corresponds to the polytrope with $`\gamma =4/3`$, and constant $`K=\frac{c\mathrm{}}{12\pi ^2}\left(\frac{3\pi ^2}{\mu _Zm_p}\right)^{4/3}`$, where $`\mu _Z=\left(x_Z\frac{Z}{A}+x_{Z1}\frac{Z1}{A}\right)^1`$ is the average number of nucleons on one electron. Here a two component mixture is considered consisting of elements with an atomic weight $`A`$ and atomic numbers $`Z`$ and $`Z1`$, with a beta transitions between them, $`x_Z`$ and $`x_{Z1}`$ are mass concentrations of these elements, $`x_Z+x_{Z1}=1`$.
Let $`u_{fe}`$ and $`\delta `$ be Fermi energy plus rest mass energy of the electrons, and threshold energy for a beta capture, in units of $`m_ec^2`$; $`g_z`$ and $`g_{Z1}`$ be statistical weights of the elements $`(A,Z)`$ and $`(A,Z1)`$; $`Ft_{1/2}`$ be a nondimensional value measured in the beta-decay experiments, or estimated theoretically. For small difference $`|\delta u_{fe}|1`$ we have simple expressions for an entropy increase during beta decay and capture in a fully degenerate matter
$$\rho T\frac{S}{t}=\mathrm{\Phi }(\delta u_{fe})^4n_{Z1},\rho T\frac{S}{t}==\mathrm{\Phi }(u_{fe}\delta )^4\frac{g_{Z1}}{g_Z}n_Z,\mathrm{\Phi }=m_ec^2\mathrm{ln}2\frac{(\delta ^21)^{1/2}\delta }{12(Ft_{1/2})_{Z1}}.$$
(7)
Here the rate of the entropy increase is equal to 1/3 of the energy loss rate by neutrino emission. During linear oscillations the beta reactions take place only in a thin layer of matter, crossing in its motion the boundary $`x=x_{}`$, $`m=m_{}`$, $`P=P_{}`$. In this layer the pressure may be represented by an expansion
$$P=P_{}g(mm_{})+\stackrel{~}{P},P^{1/4}=P_{}^{1/4}+\frac{1}{4}P_{}^{3/4}[\stackrel{~}{P}g(mm_{})],$$
(8)
Equations describing change of concentrations during oscillations averaged over the layer $`\mathrm{\Delta }m=\stackrel{~}{P}_a/g`$, are written as
$$\frac{d\overline{x}_Z}{dt}=R\frac{P_{}^{9/4}}{256}\stackrel{~}{P}_a^3\mathrm{cos}^4\omega t\frac{g_{Z1}}{g_Z},P_{eq}<P_{},P>P_{};$$
(9)
$$\frac{d\overline{x}_{Z1}}{dt}=R\frac{P_{}^{9/4}}{256}\stackrel{~}{P}_a^3\mathrm{cos}^4\omega t,P_{eq}>P_{},P<P_{}.$$
Here $`R=\mathrm{ln}2\frac{(\delta ^21)^{1/2}\delta }{3(Ft_{1/2})_{Z1}}\left(\frac{12\pi ^2\mathrm{}^3}{m_e^4c^5}\right)^{3/4}.`$ To derive an equation describing decrease of the pulsational amplitude, we should take into account the change of pressure due to change of the electron concentration. Let $`\stackrel{~}{p}`$ be a perturbation of the pressure determined by the beta reactions $`P=g(Mm)+\stackrel{~}{P}+\stackrel{~}{p}`$, with $`\stackrel{~}{p}=\frac{4}{3}P\frac{\stackrel{~}{x}_Z}{Z},P_{eq}<P_{},P>P_{};\stackrel{~}{p}=\frac{4}{3}P\frac{\stackrel{~}{x}_{Z1}}{Z1},P_{eq}>P_{},P<P_{}.`$
In presence of damping we represent the velocity perturbation in a form $`\stackrel{~}{v}=\stackrel{~}{v}_a(t)\mathrm{sin}\omega t`$, and define $`V(t)=\stackrel{~}{v}_{a,out}(m_{},t)`$. so that $`\stackrel{~}{v}=V\mathrm{sin}\omega t`$. The amplitude of the perturbed outside pressure my be written as a function of $`V`$ as $`\stackrel{~}{P}_a=z_{}^{\frac{7}{8}}\frac{\sqrt{\gamma P_0\rho _0}}{\sqrt{\lambda }}\frac{J_4(\eta _1)}{J_3(\eta _1)}V,\gamma =\frac{4}{3}`$. The equation of motion gives the relation describing damping of oscillations in the layer $`\mathrm{\Delta }m(t)=\frac{\stackrel{~}{P}_a\mathrm{cos}\omega t}{g}`$, after subtraction of the proper oscillations of the slab at the frozen composition and averaging over the motion of the whole slab
$$\mathrm{sin}\omega t\frac{dV}{dt}=\frac{\mathrm{\Delta }\stackrel{~}{p}}{\mathrm{\Delta }m(t)}\frac{\mathrm{\Delta }m}{M}\frac{g\stackrel{~}{p}(m_{})}{\stackrel{~}{P}_a\mathrm{cos}\omega t}\frac{\mathrm{\Delta }m}{M},$$
(10)
where $`\mathrm{\Delta }m=\frac{\stackrel{~}{P}_a}{g}`$. Taking into account, $`M=gP_0`$, we get
$$\mathrm{sin}\omega t\frac{dV}{dt}=\frac{g\stackrel{~}{p}(m_{})}{P_0\mathrm{cos}\omega t}.$$
(11)
Averaging over the oscillation period we obtain equation for decreasing of the amplitude of oscillations in the form
$$\frac{dV}{dt}=DV^3,D=\frac{RgP_{}^{5/4}}{768\omega P_0Z}\left(1+\frac{g_{Z1}}{g_Z}\right)(\gamma \lambda P_0\rho _0)^{3/2}z_{}^{\frac{21}{8}}\frac{J_4^3(\eta _1)}{J_3^3(\eta _1)}.$$
(12)
## 5 Energy balance and damping of oscillations
Averaging equations (7) over the time and space we get an expression for heating rate of the oscillating slab due to nonequilibrium beta processes in the layer around the URCA shell
$$\dot{Q}_\nu =R_1\frac{P_{}^3}{g\mathrm{\hspace{0.17em}32}\times 75\pi }\stackrel{~}{P}_a^5\left(1+\frac{g_{Z1}}{g_Z}\right),R_1=\mathrm{ln}2\frac{(\delta ^21)^{1/2}\delta }{12(Ft_{1/2})_{Z1}Am_p}\frac{12\pi ^2\mathrm{}^3}{m_e^3c^3}.$$
(13)
The rate of the neutrino energy losses $`L_\nu `$ (ergs/s/cm<sup>2</sup>) is obtained by averaging over a time of losses in the whole oscillating layer. We get similar to (13)
$$L_\nu =3R_1\frac{P_{}^3}{g\mathrm{\hspace{0.17em}32}\times 75\pi }\stackrel{~}{P}_a^5\left(1+\frac{g_{Z1}}{g_Z}\right).$$
(14)
Similar dependence of neutrino energy losses during oscillations because of URCA shell had been obtained by Tsuruta and Cameron (1970), who also took (14) for the rate of loss of kinetic energy of oscillations. In the approximation of strong degeneracy the matter will be heated during oscillations with the rate (13), and neutrino luminosity is determined by (14). The source of both kind of energy fluxes is the pulsation energy of the slab, giving the rate of pulsation energy losses directly connected with beta reactions as
$$\dot{E}_{pul}^{(\beta )}=(\dot{Q}_\nu +L_\nu )=\frac{R_1P_{}^3}{600\pi g}\stackrel{~}{P}_a^5\left(1+\frac{g_{Z1}}{g_Z}\right).$$
(15)
Defining $`E_{\mathrm{pul}}\frac{1}{2}MV^2`$, and using approximately $`V=\frac{\stackrel{~}{P}_a}{\sqrt{\gamma P_0\rho _0}}`$, for URCA shell in the middle of the slab, we get from (12) the equation for decreasing of pulsational energy, connected with hydrodynamic processes, in the form
$$\dot{E}_{\mathrm{pul}}^{(\mathrm{dyn})}=\frac{RP_{}^{5/4}}{768Z\omega \sqrt{\gamma P_0\rho _0}}\stackrel{~}{P}_a^4(1+\frac{g_{Z1}}{g_Z}).$$
(16)
A ratio of damping rates of oscillation on $`i`$-th mode is $`\frac{\dot{E}_{\mathrm{pul}}^{(\beta )}}{\dot{E}_{\mathrm{pul}}^{(\mathrm{dyn})}}\frac{\stackrel{~}{P}_a}{P_0}\frac{\omega _i}{\omega _1}.`$ It follows that the main source of damping of oscillations in presence of URCA shell is connected not with the neutrino emission / nonequilibrium heating, but with a dynamical action of the nonequilibrium layer of the slab, where beta reactions take place. This action leads to an exitation of short-wavelength acoustic waves with length $`l\frac{\mathrm{\Delta }m}{\rho _0}x_0`$. When the wavelength of the excited eigen-mode approaches the thickness of the nonequilibrium layer $`\mathrm{\Delta }x`$, formed by oscillations, both mechanisms of damping become comparable. For $`l_i\mathrm{\Delta }x=\frac{\stackrel{~}{P}_a}{g\rho _{}}`$ with account of relations $`\omega _1\frac{1}{x_0}\sqrt{\frac{P_0}{\rho _0}}`$, $`x_0\frac{P_0}{g\rho _0}`$, $`\omega _i\frac{1}{l_i}\sqrt{\frac{P_0}{\rho _0}}`$, and $`\rho _{}\rho _0`$, we get $`\dot{E}_{\mathrm{pul}}^{(\beta )}/\dot{E}_{\mathrm{pul}}^{(\mathrm{dyn})}1`$. Importance of the dynamical damping of stellar oscillations in presence of URCA shell is connected with non-linearity of weak interaction rates, and deviations from eigen-oscillations under an action of weak interactions in this layer, leading to excitation of acoustic waves. The dissipation of pressure oscillations, connected with excitation of sound waves is inherent to any kind of dissipation, when the main term is nonlinear. The linear mechanism, connected with a conventional bulk viscosity, does not change the form of the eigenfunction of oscillations, and so no additional waves are excited.
## 6 URCA shell in a convective motion
It was concluded in that nonequilibrium heating is balanced by the change in convective flow, leading to the net cooling due to convective URCA shell. Nine years later same authors changed their mind, concluding that ”convective URCA process can reduce the rate of heating by nuclear reactions but cannot result in a net decrease in entropy, and hence in temperature, for a constant or increasing density.” This conclusion, as well as opposite one, made by using thermodynamic relations only, seems to be not convincing. Following this line, let us present two plausible scenarios, leading to two opposite conclusions.
A. Due to action of a nonlinear bulk viscosity, the convection is damping in the vicinity of the URCA shell, decreasing the convective heat flux from the central part of the star. In the general heat balance of the star it means that cooling become less effective, and nuclear reactions become thermally unstable and lead to a nuclear explosion earlier, than without a presence of the URCA shell. Nonequibrium heating give additional heating, supporting the earlier nuclear explosion.
B. Due to action of a nonlinear bulk viscosity, the convection is damping in the vicinity of the URCA shell, decreasing the convective heat flux from the central part of the star. Due to local decrease of the heat flux from the core the average temperature gradient increases, leading finally to increase of the convective flux soon after entering an URCA shell into a convective zone. If the increase of the convective flux prevails the nonequilibrium heating in the URCA shell, the general heat balance would be shifted to a larger temperature with more effective cooling, and the boundary of the thermal explosion would be postponed in time, if not eliminated.
I cannot choose between these two scenario without construction of the numerical model taking into account all processes mentioned above. In such a highly nonlinear system, as a star with nuclear reactions, neutrino losses, degeneracy, convection, and many feedback influences it seems to be impossible to make a conclusion about the direction of process under the action of addtional URCA shell, basing only on thermodynamical ground.
Convective modes belong to $`g`$-mode family, in which the local pressure perturbations are small, and could be neglected when the convective velocity is much less than the velocity of the sound. In this situation the sound wave dissipation of the convective modes imposed by URCA shell is negligible. The equations of stellar evolution in presence of URCA shell should take into account the following physical processes
1. loss of enegry due to neutrino emission in the URCA shell
2. heating of the matter in the convective region around URCA shell due to nonequilibrium beta processes.
3. Decrease of the convective velocity in the layer around the URCA shell due to energy dissipation connected with the nonequilibrium beta processes. Kinetic energy of the convection is the source of energy for neutrino losses and nonequilibrium heating of the matter. In the condition of static equilibrium only energy and heat transfer equations should be modified. In the energy equation
$$T\frac{dS}{dt}=\frac{dE}{dt}\frac{P}{\rho ^2}\frac{d\rho }{dt}=ϵ_nϵ_\nu +ϵ_\nu ^{CU}\frac{1}{4\pi \rho r^2}\frac{dL_r}{dr},$$
(17)
in addition to other neutrino cooling processes $`ϵ_\nu `$, the new term $`ϵ_\nu ^{CU}`$ is connected with heating due to nonequilibrium beta processes around the URCA shell. having in mind strong degeneracy of electrons in this region. Neutrino emission in nonequilibrium URCA processes is accompanied by heating at high degeneracy, because the positive term $`\mu _idn`$ exceed the energy carried away by neutrino (Bisnovatyi-Kogan and Seidov, 1970). Convective motion, consisting of convective vortexes around an URCA shell is a source of additional neutrino energy losses, and of heating of the matter. This dissipation of convective energy may be described in the same way, as corresponding dissipation and heating during stellar pulsations. Therefore we use for description of these processes the formulae from the previous sections. If we accept that the pressure difference in the convective vortex is about one half of the local pressure, we use for the amplitude of pressure pulsations in (13) and (16) $`\stackrel{~}{P}_a=\frac{\alpha _p}{4}P_{}`$. Taking also approximately $`P_0=P_{},\rho _0=\rho _{},`$ and $`u_{fe}=\delta `$ we get
$$\dot{Q}_\nu =\stackrel{~}{R}\frac{m_ec^2}{Am_p}\frac{\delta ^8\alpha _p^5}{g\mathrm{\hspace{0.17em}2}^{10}\times 75\pi },\stackrel{~}{R}=\mathrm{ln}2\frac{(\delta ^21)^{1/2}\delta }{12(Ft_{1/2})_{Z1}}\frac{m_e^4c^5}{12\pi ^2\mathrm{}^3}\left(1+\frac{g_{Z1}}{g_Z}\right).$$
(18)
Equation (18) is related to energy losses averaged over the whole slab. In the convective motion the losses are localized in the layer around the URCA shell radius $`r_{}`$
$$r_{}+l_{\mathrm{conv}}<r<r_{}l_{\mathrm{conv}},l_{\mathrm{conv}}=\alpha _p\frac{P}{P},$$
(19)
here $`l_{\mathrm{conv}}`$ is taken from the mean free path model. The local rate is obtained from (18), if we take into account that the whole heating is concentrated inside the layer (19). We get than
$$ϵ_\nu ^{CU}=\frac{\dot{Q}_\nu }{2\rho l_{\mathrm{conv}}}.$$
(20)
Convective velocity suffers from additional damping due to URCA shell, because kinetic energy of the convection is a source of both nonequilibrium heating and of additional neutrino losses. The only relation of the convetional mixing length model of the convetion (see e.g. ) should be modified, determining the convective velocity with an additional damping
$$\frac{1}{2}\rho v_{\mathrm{conv}}^2=\frac{1}{8}(\mathrm{\Delta }T)l^2\left(\frac{\rho }{T}\right)|_Pg\frac{\dot{E}_{\mathrm{conv}}^{(\beta )}}{v_{\mathrm{conv}}},$$
(21)
where $`\dot{E}_{\mathrm{conv}}^{(\beta )}=4\dot{Q}_\nu `$ is found from (18). These relations may be applied for description of URCA shell convection only for sufficiently strong convective motion, when the first term in (21) exceeds considerably the second one. The equation (21) has roots only when
$$\dot{E}_{\mathrm{conv}}^{(\beta )}<\frac{1}{8\sqrt{\rho }}\left[\frac{1}{3}(\mathrm{\Delta }T)l^2\left(\frac{\rho }{T}\right)|_Pg\right]^{3/2}.$$
(22)
Violation of this inequality may result in an abrupt termination of the convection in the layer (19) around the URCA shell. When analysing the URCA shell convection in star, it would be premature to predict the results of evolutionary calculations with account of convective URCA shell before such calculations are done. Two possibilities may be expected . One is connected with obtaining of a definite result which has a little sensitivity to the input parameters of the problem, such as $`\alpha _p`$, $`Ft_{1/2}`$, accepted rates of nuclear reactions, neutrino losses etc. Another possibility could be a great sensitivity of the result to the same input parameters. If the second possibility would be realized we could still remain in situation of ambiguity, because the set of the input parameters for presupernovae model cannot be established with a sufficient precision.
## Acknowledgements
Author is grateful to R.Canal, S.I.Blinnikov, J.Isern, R.Mochkovich for useful discussions; to W.Hillebrandt and E.Müller for the invitations and posssibility to participate in workshops ”Nuclear Astrophysics’, to O.V.Shorokhov for help. This work was partly supported by Russian Basic Research Foundation grant No. 99-02-18180 and grant of a Ministry of Science and Technology 1.2.6.5.
|
warning/0004/astro-ph0004001.html
|
ar5iv
|
text
|
# Photometric Light Curves and Polarization of Close-in Extrasolar Giant Planets
## 1 Introduction
The discovery of the planet 51 Peg b in 1995 (Mayor & Queloz 1995), only 0.051 AU from its parent star, heralded an unexpected new class of planets. Due to gravitational selection effects, several more Jupiter-mass close-in extrasolar giant planets (CEGPs) have been discovered since that time (Butler et al. 1997, 1998; Mayor et al. 1999; Mazeh et al. 2000). To date there are 5 extrasolar giant planets $`<`$ 0.05 AU from their parent stars, and an additional 9 $`<`$ 0.23 AU (see Schneider 2000). Relevant data about the close-in planet-star systems (orbital distance $`<`$ 0.05 AU) are listed in Table 1. Ongoing radial velocity searches will certainly uncover more CEGPs in the near future. The CEGPs are being bombarded by radiation from their parent stars, and could be very bright in the optical. At best the CEGPs could be four to five orders of magnitude fainter than their primary star, compared to Jupiter which is 10 orders of magnitude fainter than the Sun.
The recent transit detection of HD 209458 b by Charbonneau et al. (2000) and Henry et al. (2000b) confirms that the CEGPs are gas giants, gives the planet radius, and fixes the orbital inclination, which removes the $`\mathrm{sin}i`$ ambiguity in mass and provides the average planet density. HD 209458 has $`R_{}=1.2\pm 0.1R_{}`$ and $`R_P=1.40\pm 0.17R_J`$ (Mazeh et al. 2000), where $`R_{}`$ is the stellar radius and $`R_P`$ is the planet radius. Transits are definitely ruled out for $`\tau `$ Boo b, 51 Peg b, $`\upsilon `$ And b, HD 187123, and $`\rho ^1`$ Cnc b, whether they are assumed to be gas giants with radius 1.2 $`R_J`$, or smaller rocky planets with radius $``$0.4 $`R_J`$ (Henry et al. 2000a, 1997; Baliunas et al. 1997; G. Henry, private communication). Transits are also ruled out for HD 75289 (M. Mayor, private communication). For a transit to be observable, a CEGP must be aligned with the star as seen from Earth with an inclination $`i>\theta _T`$, where $`\theta _T=\mathrm{cos}^1((R_{}+R_P)/D)`$. For random orientations, the probability for $`i`$ to be between $`90^{}`$ and $`j^{}`$ is $`P(j)=\mathrm{cos}(j)`$. With $`\theta _T83^{}`$, the CEGPs have transit probabilities of ten percent. By the same $`\theta _T`$ criterion, the non-detection of transits puts limits on the orbital inclinations to approximately 83 (for $`R_{}=1.16`$ $`R_{}`$, $`R_P=1.2`$ $`R_J`$, and $`D=0.051`$ AU). Several groups (e.g. STARE (PI T. Brown), Vulcan Camera Project (PI W. Borucki), WASP (PI S. Howell)) are monitoring thousands of stars without known planets, searching with high precision photometry for periodic fluctuations indicative of a planetary transit. Follow-up observations by radial velocity techniques (or astrometry in the future) will be needed to fix the orbital radius in order to determine the planet mass. Edge-on CEGP systems are the most promising for reflected light signals.
Several observational approaches to detecting and characterizing CEGP atmospheres have been developed. These include spectral separation, transmission spectra observations during transit, infrared observations, and optical photometric light curve observations.
Charbonneau et al. (1999) and Cameron et al. (1999) have developed a direct detection technique: a spectral separation technique to search for the reflected spectrum in the combined star-planet light. Both groups have observed the $`\tau `$ Boo system. $`\tau `$ Boo A is one of the brightest (fourth magnitude), hottest (F7V) parent stars, and $`\tau `$ Boo b has one of the smallest semi-major axes; these three properties make $`\tau `$ Boo a promising candidate for this technique. From a non-detection, Charbonneau et al. (1999) have put upper limits on the planet-star flux ratio ranging from $`5\times 10^5`$ for $`\mathrm{sin}i1`$ to $`1\times 10^4`$ for $`\mathrm{sin}i0.5`$. The result is within the strict assumptions that the light curve is fairly isotropic and that the reflected spectrum is an exact copy of the stellar spectrum from 4668 to 4987 Å. Their upper limit on the geometric albedo is 0.3 for $`\mathrm{sin}i1`$. The same technique for the $`\tau `$ Boo system has been used by Cameron et al. (1999) who claim a possible detection at an inclination of 29, and give a planet-star flux ratio of $`1.9\times 10^4`$ at $`i=90^{}`$. Given $`R_P`$, the albedo derived from this type of observation can provide a weak constraint on theoretical models.
A second approach is to observe transmission spectra during a planet transit. The stellar flux will pass through the optically thin part of the planet atmosphere. Theoretical predictions show the planetary absorption features will be at the $`10^4`$ to $`10^3`$ level (Seager & Sasselov 2000). Successful observations will constrain the cloud depth and may give important spectral diagnostics such as the presence of CH<sub>4</sub> which is a good temperature indicator for the upper atmosphere layers.
A third technique under development is the use of the Keck infrared interferometer in the differential phase mode to directly detect and spectroscopically characterize the CEGPs. The technique is based on the difference between the very smooth infrared stellar spectrum and the strong water absorption bands and possibly methane bands in the CEGP’s infrared spectrum. See Akeson & Swain (1999) for more details.
In this paper we present theoretical photometric light curves and polarization curves of the CEGP systems. As the planet orbits the star, the planet changes phase as seen from Earth. The planet and star are too close together for their light to be separated, but this small separation means the stellar flux hitting the planet is large, and the reflected light variation in the combined light of the system from the planet’s different phases may be detectable. We focus on the optical where there is a clear signature of reflected light: the planet’s dark side has no reflection or emission in optical light. In contrast, there is no large light variation in the infrared where the CEGPs are bright on both the day and night side from reemission of absorbed light. Scattered light is minimal in the infrared, and is difficult to disentangle from the emitted light. Unlike transits, which can only be seen for inclinations $`>\theta _T`$, the reflected light curves of lower inclinations are theoretically visible and may be detectable. This work is motivated by upcoming microsatellites MOST ($``$2002) (Matthews 1997), COROT ($``$2003) (Baglin 1998, 2000), and MONS ($``$2003) (Christensen-Dalsgaard 2000). Initially intended for asteroseismology, these satellites have capabilities to detect $`\mu `$mag variability. MOST will observe known stars in a broad visual waveband, one at a time for a period of roughly one month, including one with a known CEGP in its first year. COROT’s exoplanet approach will use two CCDs in two colors to observe several fields of $``$6000 stars for a few months each. Because of this wide-field approach the stars with known CEGPs will not be observed by COROT. While COROT’s main exoplanet focus is on transits, probability estimates suggest several CEGP light curves from reflection should be detected. Precision of ground-based photometry on the CEGP parent stars is currently at 100 $`\mu `$mag, and could reach 50 $`\mu `$mag in the near future with dedicated automatic photometric telescopes (Henry et al. 2000a). We also present polarization signatures although they are well under the current limits of detectability which is a few $`\times 10^4`$ in fractional polarization of the system (e.g. Huovelin et al. 1989).
This paper, to our knowledge, is the first to describe photometric light curves and polarization of CEGP systems: gas giants in close orbits around Sun-like stars. Although our own Solar System planets have been well studied in reflected and polarized light, the CEGPs have effective temperatures an order of magnitude higher, so completely different cloud species and atmospheric parameters are expected. If observable, the light curves would roughly constrain the type and size distribution of condensates in the planetary atmosphere. In §2 we present definitions and analytical estimates of reflected light from the CEGPs, in §3 a description of our model, and in §4 results and discussion.
## 2 Analytical Estimate of the Light Curves and Polarization
An analytical estimate of the amount of reflected light and polarization of an EGP system is useful for both comparison with simulations and for upper limit predictions. A good idealized case for such estimates is provided by modeling a planet as, for example, a Lambert sphere. The Lambert sphere derives from the law of diffuse reflection proposed by Lambert, postulating a reflecting surface with a reflection coefficient that is constant for all angles of incidence. The reflection coefficient is simply the ratio of the amount of light diffusely reflected in all directions by an element of the surface to the incident amount of light which falls on this element. The general conditions postulated by Lambert, for example angle independence, are satisfied strictly only for an absolute blackbody and an ideal reflecting surface (often called “absolutely white”, “ideally matted”, etc.). Thus derives Lambert’s definition of albedo, with its inherent ambiguities as discussed at the end of the 19$`th`$ century by Seeliger, and the ultimate decision by Russell (1916) to endorse Bond’s (1861) definition of albedo for use in the Solar System.
The arguments offered by Russell (1916) in favor of the Bond albedo (over other albedo definitions in use at the time) are still relevant for Solar System objects, but not necessarily for EGPs. One important point in favor of the Bond albedo was that it is derivable from observations. The Bond albedo is defined as the ratio of the total amount of reflected light to the total amount of incident plane-parallel light integrated over all angles. Note that at the time of Russell — before multi-wavelength observations of the Solar System planets — the Bond albedo, $`A`$, was not defined as an integrated quantity over all wavelengths as it is today. However, the discussion below is still valid with either definition. The Bond albedo $`A`$ can be separated into two quantities,
$$A=pq,$$
(1)
where $`p`$ is the geometric albedo and $`q`$ is the phase integral. The geometric albedo is defined as the planet flux divided by the reflected flux from a perfectly diffusing disk of the same radius. The phase integral $`q`$ is defined as
$$q=_0^\pi \varphi (\alpha )\mathrm{sin}\alpha d\alpha ,$$
(2)
where $`\varphi (\alpha )`$ is the phase function, or the brightness variation of the planet at different phases. The phase angle, $`\alpha `$, is the angle between the star and Earth as seen from the planet; $`\alpha =0`$ corresponds to “opposition” when the planet is maximally illuminated as seen from Earth. $`p`$ is measureable for all Solar System planets because it is a geometric and photometric quantity. $`q`$ is measureable from Earth for Mercury, Venus, Mars, and the Moon; for the outer planets whose phase angle variation is only up to several degrees from Earth satellite mission observations were necessary. Thus the benefit of the Bond albedo: it is a physically meaningful quantity but it can be determined empirically for the Solar System planets.
In contrast, the Bond albedo cannot be determined from observations for extrasolar planets. Because the EGP systems are so distant from Earth, only the CEGPs have prospects for measurement of $`p`$ in the foreseeable future, and even then only the most reflective CEGPs will be bright enough, and only $``$10% of those will have orbital inclinations near $`90^{}`$. Charbonneau et al. (1999) and Cameron et al. (1999) have developed a spectral separation detection technique which can put upper limits on — and in the best case measure — $`p`$ in a narrow wavelength region. For CEGPs that are reflective, and at $`i90^{}`$, $`q`$ should be measureable with the upcoming satellite missions. However if a given CEGP system is at $`i<90^{}`$, the full range of phases will not be visible (i.e. $`\alpha `$ will not be fully probed), and $`p`$ and $`q`$ will not be measureable. As discussed in §4.7.2, EGPs beyond $`D=0.1`$ AU will not be detectable in optically reflected light even with the upcoming satellite missions. The EGPs certainly have promise for detection in the infrared where they emit most of their energy. However, most of this energy is reprocessed absorbed energy; it is not possible to measure the Bond albedo with infrared observations. To summarize, the Bond albedo came into standard usage because it was a measureable quantity. This is not possible for almost all of the EGPs because of the distance of the systems and random orbital inclinations.
The goal of this paper is a presentation of light curves at all viewing angles and at different inclinations, instead of a Bond albedo. We begin with the analytical estimate, where in the idealized case we are simply interested in the ratio, $`ϵ`$, of the observed flux at Earth from the EGP at full phase ($`\alpha =0`$) to that of the star: $`ϵ=p(R_P/D)^2`$. Here $`D`$ is the star-planet distance, and $`p`$ and $`R_P`$ are as previously defined. For a Lambert sphere the single scattering albedo $`\stackrel{~}{\omega }=1`$; no photons are absorbed, and so $`A=1`$. For a Lambert sphere, all incoming photons are singly, isotropically scattered, and $`p=2/3`$. The light variation of the Lambert sphere is only due to phase effects $``$ the phase function, $`\varphi `$, is simply (Russell 1916):
$$\varphi (\alpha )=\frac{\mathrm{sin}(\alpha )+(\pi \alpha )\mathrm{cos}(\alpha )}{\pi },$$
(3)
and the phase-dependent flux ratio is $`ϵ\varphi (\alpha )`$. Note that the phase angle, $`\alpha `$, is a function of the orbital phase and inclination. We convert this flux ratio to variation in micromagnitudes by
$$\mathrm{\Delta }m=2.5\mathrm{log}_{10}(1+ϵ\varphi (\alpha ))10^6.$$
(4)
Figure 1a shows the photometric light variation at $`i=90^{}`$ for a Lambert sphere of $`R=1.2R_J`$ at various $`D`$ corresponding to known CEGP systems. The large variation of 110 and 140 $`\mu `$mag for planets with $`D`$ corresponding to $`\tau `$ Boo b and HD 187123 respectively is above current ground-based limits (Henry et al. 2000a). The Lambert sphere light curve is unrealistic, and the point of this paper is to show that the situation is far more complex and almost always conducive to a smaller light variation for a few reasons. Firstly, the single scattering albedo, $`\stackrel{~}{\omega }`$, is generally different from one. When optical photons are absorbed by condensates or gas in the CEGP atmospheres, they are re-emitted in the infrared or contribute to the thermal pool. Secondly, multiple scattering gives more of a chance for absorption over single scattering for the same single scattering albedo. Each time a photon scatters, its next encounter has a scattering probability of $`\stackrel{~}{\omega }`$, but when a photon is absorbed it cannot contribute to the scattered light. This effect depends on density, i.e. on the mean free path of the photon. Thirdly, when particles are large compared to the wavelength of light, the particle scatters preferentially in the forward direction. In this case, photons that enter the atmosphere are likely to be multiply scattered down into the atmosphere and eventually absorbed, rather than to be backscattered and escape the planetary atmosphere.
Figure 1b shows the fractional polarization of the total light of a Lambert sphere at $`i=90^{}`$ at various $`D`$ corresponding to known CEGP systems, and assuming $`R_P=1.2R_J`$. We assume Rayleigh scattering linear polarization of unpolarized incident light $`Pol_{Ray}=\mathrm{sin}^2\theta _S/(1+\mathrm{cos}^2\theta _S)`$ (Chandrasekhar 1960). Here $`\theta _S`$ is the scattering angle: $`\theta _S=0^{}`$ is the forward direction and $`\theta _S=180^{}`$ the backward direction. $`Pol_{Ray}`$ peaks at a scattering angle of $`90^{}`$. For single scattering and $`\stackrel{~}{\omega }=1`$, all scattered light is polarized. The polarization signature plotted in Figure 1b does not peak at $`\alpha =90^{}`$ because the polarization is modulated by the reflected light curve; the scattered light is maximally polarized at $`\alpha =90^{}`$, but the amount of scattered light peaks at $`\alpha =0^{}`$. Plotted is $`Pol=(S_{}S_{})/(S+F)=ϵ\varphi (\alpha )Pol_{Ray}`$, where $`S`$ is the total scattered light, $`F`$ is the unpolarized stellar flux, $`SF`$, and $`S_{}`$ and $`S_{}`$ are the perpendicular and parallel components of the scattered light respectively. In reality polarization is much lower than this best case estimate, for a few reasons. Firstly, the amount of scattered light is expected to be lower as described above. Secondly, for the case of multiple scattering not all of the scattered light is polarized – multiple scatterings mean the photon loses some of its polarization signature. Thirdly, when the particle is large compared to the wavelength of light, different light paths through the particle and interference effects cause the polarized light to be lower and to have more than one peak, and so a strong single peaked signal such as shown in Figure 1b may not be reached.
As shown in Figures 1a and 1b, the light curves and polarization are very sensitive to $`D`$, since $`ϵ1/D^2`$. For example, the Lambert sphere at $`D=0.042`$ AU (corresponding to HD 187123) has a light curve and polarization curve with amplitude twice as high as a Lambert sphere at $`D=0.059`$ AU (corresponding to $`\upsilon `$ And b). The light curve and polarization curve estimates are also sensitive to $`R_P`$. Both can be scaled — the light curve approximately and the polarization curve exactly — for different $`R_P`$ by the factor $`(R_P/1.2R_J)^2`$. For HD 209458 b with $`R_P=1.40\pm 0.17R_J`$ (Mazeh et al. 2000), this factor is 1.36.
## 3 Model Atmosphere
The model atmosphere code consistently solves for the planetary emergent flux and temperature-pressure structure by simultaneously solving hydrostatic equilibrium, radiative and convective equilibrium, and chemical equilibrium in a plane-parallel atmosphere, with upper boundary condition equal to the incoming radiation. The code is described in Seager (1999) and is improved over our code described in Seager & Sasselov (1998) in two major ways. One is a Gibbs free energy minimization code to calculate equilibrium abundances of solids and gases, the second is condensate opacities for 3 solid species. So while in Seager & Sasselov (1998) we considered neither the depletion of TiO nor accurate formation of MgSiO<sub>3</sub>, in the new models we do.
We compute the photometric light curves and polarization curves with a 3D Monte Carlo code (Whitney, Wolff, & Clancy 1999), using the atmospheric profiles (opacities and densities as a function of radial depth) generated by the Seager & Sasselov code. While in principle one could compute light curves from the model atmosphere code described above, the Monte Carlo scheme can treat a much more sophisticated scattering method, with anisotropic scattering and polarization, than our model atmosphere program. Both are described in this section.
### 3.1 Chemical Equilibrum
The equation of state is calculated using a Gibbs free energy minimization method, originally developed by White, Johnson, & Danzig (1958), and followed up by a number of papers in the 1960s and 1970s including a particularly useful one by Eriksson (1971). A detailed description of this method can be found in those papers; more recent treatments relevant for astrophysics are described in Sharp & Huebner (1990), Petaev & Wood (1998), and Burrows & Sharp (1999). A more complete chemical equlibrium calculation applied to Gliese 229 B is described in Fegley and Lodders (1996).
We have selected the most important species from Burrows and Sharp (1999) for brown dwarfs and from Allard & Hauschildt (1995) for cool stars. We used fits to the Gibbs free energy from Sharp & Huebner (1990), from Falkesgaard (private communication), or fitted from the JANAF tables (Chase 1998) following the normalization procedure described in Sharp & Huebner (1990). We include ions using a charge conservation in place of the usual mass balance constraint (equation (10) in Sharp & Huebner 1990). In the Gibbs method we include 27 elements, with 90 gaseous species and 4 solid species: H, He, O, C, Ne, N, Mg, Si, Fe, S, Ar, Al, Ca, Na, Ni, Cr, P, Mn, Cl, K, Ti, Co, F, V, Li, Rb, Cs, CO, H<sub>2</sub>, OH, SH, N<sub>2</sub>, O<sub>2</sub>, SiO, TiO, SiS, H<sub>2</sub>O, C<sub>2</sub>, CH, CN, CS, SiC, NH, SiH, NO, SN, SiN, SO, S<sub>2</sub>, C<sub>2</sub>H, HCN, C<sub>2</sub>H<sub>2</sub>, CH<sub>4</sub>, AlH, AlOH, Al<sub>2</sub>O, CaOH, MgH, MgOH, VO, VO<sub>2</sub>, CO<sub>2</sub>, TiO<sub>2</sub>, Si<sub>2</sub>C, SiO<sub>2</sub>, FeO, FeS, NH<sub>2</sub>, NH<sub>3</sub>, CH<sub>2</sub>, CH<sub>3</sub>, H<sub>2</sub>S, KOH, NaOH, NaCl, NaF, KCl, KF, LiCl, LiF, CsCl, CsF, H<sup>+</sup>, H<sup>-</sup>, H$`{}_{}{}^{}{}_{2}{}^{}`$, H$`{}_{}{}^{+}{}_{2}{}^{}`$, Na<sup>-</sup>, K<sup>-</sup>, Li<sup>-</sup>, Cs<sup>-</sup>, Fe (solid), CaTiO<sub>3</sub>, Al<sub>2</sub>O<sub>3</sub>, MgSiO<sub>3</sub>.
While a full treatment of all naturally occuring elements and $``$2500 compounds (e.g. Fegley & Lodders 1996) is possible, our abridged choice is certainly adequate for a first prediction of CEGP photometric light curves. There may be non-equilibrium chemistry involved (e.g. photochemistry on our own Solar System planets) that is not addressed by even a complete thermodynamical equilibrium calculation. In general, as the temperature decreases from the inner atmosphere to the outer atmosphere, the metal gases are depleted into solids which are efficient absorbers or reflectors. The three condensate opacities we chose (solid Fe, MgSiO<sub>3</sub>, and Al<sub>2</sub>O<sub>3</sub>) have very different optical constants (see §3.3 and §4.2), and are among the dominant solids expected at the relevant temperatures and pressures.
### 3.2 Radiative Transfer
The flux from the parent star travels through the planetary atmosphere, interacting with absorbers and scatterers in a frequency-dependent manner. In a condensate-free CEGP atmosphere (Seager & Sasselov 1998), blue light will Rayleigh scatter deep in the atmosphere where the density of scatterers is highest, while infrared light will be absorbed high in the atmosphere due to strong absorbers such as TiO and H<sub>2</sub>O. Similar results were found for dusty models by Marley et al. (1999), although they considered an isolated planet of the equilibrium effective temperature. In other words, they assumed that the absorbed stellar flux can be accounted for as thermalized intrinsic flux for the calculation of the atmospheric structure. For differences in the self-consistent treatment of irradiation and an isolated planet at the same $`T_{\mathrm{eff}}`$, see Seager & Sasselov (1998) and Seager (1999).
The equilibrium effective temperature is defined by $`T_{\mathrm{eq}}=T_{}(R_{}/2D)^{1/2}[f(1A)]^{1/4}`$. Here the subscript * refers to the parent star, $`D`$ is the star-planet distance, $`A`$ is the Bond albedo, $`f`$=1 if the heat is evenly distributed, and $`f`$=2 if only the heated side reradiates the energy. Physically, $`T_{\mathrm{eq}}`$ is the effective temperature attained by an isothermal planet (after it has reached complete equilibrium with its star).
Our approach is to use $`F_{}=\sigma T_{}^4R_{}^2/4D^2`$ as the incoming flux and assume that $`f`$=1, as it will for planets with a thick atmosphere, due to rapid zonal and meridional circulation patterns (Guillot et al. 1996). The factor of 4 is due to the assumption that the absorbed incoming radiation is efficiently distributed to all parts of the planet: radiation incoming to a cross section of $`\pi R_P^2`$ is reemitted into $`4\pi R_P^2`$. With this approach we need not use $`T_{\mathrm{eq}}`$, nor the Bond albedo used in the $`T_{\mathrm{eq}}`$ definition. Heating of the planet happens in a frequency- and depth-dependent manner, and the heating, as well as the planet’s $`T_{\mathrm{eff}}`$, comes out of the model atmosphere solution. We treat the incoming flux as plane-parallel, which is accurate for isotropic scattering (see §4.3.2).
An additional but small contribution to the $`T_{\mathrm{eff}}`$ is the internal planetary heat. Because of the strong irradiation, the internal temperature of the planet is greater than the internal temperature of an isolated planet (Guillot et al. 1996). The planet possesses an intrinsic luminosity because it leaks some of the heat acquired during formation by loss of gravitational energy. Thus, the atmosphere’s inner boundary condition is age- and mass-dependent, and needs a self-consistent atmospheric and evolutionary calculation with accurate irradiation and spectral modeling. In any case the reflected spectra are not affected by the lower boundary condition; any good guess is too cool to produce light in the optical, which is entirely reflected light.
We solve the radiative transfer equation using the Feautrier method with 100 angular points and 3500 wavelength points. We include isotropic scattering except for Rayleigh scattering which can be added to the Feautrier method via a modified source function (Chandrasekhar 1960).
### 3.3 Opacities
We get the optical constants of MgSiO<sub>3</sub> (enstatite) from Dorschner et al. (1995), of Al<sub>2</sub>O<sub>3</sub> (corundum) from Koike et al. (1995) and Begemann et al. (1997), and of Fe (iron) from Ordal (1985) and Johnson & Christy (1973). In all cases the optical constants were extrapolated below 0.2 $`\mu `$m. The condensate opacities (absorption and scattering) were computed using Mie theory for spherical particles with a version of the code from Bohren & Huffman (1983). The condensate opacities dominate over gaseous Rayleigh scattering.
For H<sub>2</sub>O, which is the dominant infrared absorber, we use the straight means opacities (Ludwig 1971). TiO is only present very deep in the atmosphere for the hottest models; we use straight means opacities from Collins & Fay (1974). Even if the features do not appear in the atmosphere, the opacity contributes to the temperature-pressure structure. We also include H<sub>2</sub>-H<sub>2</sub> and H<sub>2</sub>-He collision-induced opacities from Borysow et al. (1997), and Rayleigh scattering by H<sub>2</sub> and He from Mathisen (1984). CH<sub>4</sub> opacities are taken from the GEISA database (Husson et al. 1994) which is incomplete for the high temperatures of the CEGPs. Unfortunately the only existing optical CH<sub>4</sub> opacities are coefficients derived from Jupiter (Karkoschka 1994), and we do not include them.
The alkali metals, noteably Na I and K I, are very important opacity sources in brown dwarf spectra (Tsuji et al. 1999; Burrows, Marley, & Sharp 2000). The alkali metals’ (Na, K, Li, Cs) oscillator strengths and energy levels were taken from Radzig & Smirnov (1985). We only include the low-lying resonance lines which may have large absorption troughs in the optical. We compute line broadening using a Voigt profile with H<sub>2</sub> and He broadening, and Doppler broadening.
Many better line lists exist, and other opacities that are present in L dwarf spectra may also appear in the CEGPs, but they are not necessary for a first approximation of light curves. We plan to include them in future work. While the opacities are necessary for a self-consistent solution of the atmosphere profile, the reflected spectra are not sensitive to small details in the infrared spectra.
### 3.4 Condensates
The atmospheric structure and emergent spectra of our “dusty” models are highly dependent on condensates, as first noted for brown dwarf models in Lunine et al. (1989), and subsequently by other modelers (e.g. Tsuji et al. 1996). The CEGPs have an extra sensitivity to condensates because the strong irradiation will heat up the upper atmosphere according to the condensate amount and absorptivity (see Figure 4). Also, because the incoming radiation is strongly peaked in the optical, in contrast to isolated brown dwarfs which have little optical emission, the condensates will cause strong reflection or absorption in the optical.
We consider four different sizes of condensates, based on the Solar System planets. The cases are intended to explore the expected size range, in part because the cloud theories are limited and may be in enough error that such assumptions are just as good. The particle sizes considered are mean radius $`\overline{r}=`$ 0.01 $`\mu `$m, 0.1 $`\mu `$m, 1 $`\mu `$m, and 10 $`\mu m`$. All have gaussian size distributions with a standard deviation of 0.1 times mean particle radius. This choice is narrow enough to attribute specific effects to a given particle size, but wide enough to prevent interference effects. Cloud particles in Venus have $`r=0.851.15\mu `$m, and a haze layer above that has particles with $`r=0.2\mu `$m (Knibbe et al. 1997). The clouds on Jupiter range from an upper haze layer with $`r=0.5\mu `$m, and lower cloud decks with $`r=0.75\mu `$m, and $`r=0.4550`$ $`\mu `$m (Taylor & Irwin 1999). We assume that particles are distributed homogenously horizontally and vertically from the cloud base. The limitations of this assumption are discussed in §4.5
For the light curve calculation with the Monte Carlo scattering code we compute the scattering matrix elements (Van de Hulst 1957) from Mie theory, which describe the anisotropic scattering phase function and the polarization.
### 3.5 Monte Carlo method for scattering
We use the atmosphere structure generated in the Seager & Sasselov code and then compute the light curves and polarization curves using a Monte Carlo scattering code. In principle it is possible to compute the light curves with a model atmosphere code, but it is much more accurate to use the Monte Carlo code since it can deal with anisotropic scattering, the spherical geometry of the planet, and can easily compute all viewing angles — inclinations and phases — from one run.
The basic principle of the Monte Carlo scattering method is that photon paths and interactions are simulated by sampling randomly from the various probability distribution functions that determine the interaction lengths, scattering angles, and absorption rates. Incoming photons at a given frequency travel into the atmosphere (to a location sampled from a probability distribution function), and scatter using random numbers to sample from probabilistic interaction laws. At each scatter, the photon’s polarization and direction changes according to the phase function. Photons are followed until they are absorbed (they can no longer contribute to the reflected light), or until they exit the sphere. On exit, the photons are binned into direction and location; the result is flux and polarization as a function of phase and inclination.
The code we use was adapted from several previous codes described in the literature (Whitney 1991; Whitney & Hartmann 1992, 1993; Code & Whitney 1995; Whitney et al. 1999). Improvements from previous versions include exact sampling of the scattering phase function for any grain composition, arbitrary atmospheric density profiles, and inclusion of arbitrary opacity sources. Phase functions of MgSiO<sub>3</sub>, Al<sub>2</sub>O<sub>3</sub>, and Fe are computed using Mie theory. Additional opacities include Rayleigh scattering by H<sub>2</sub> and He, and absorption by H<sub>2</sub>O, TiO, H<sub>2</sub>–H<sub>2</sub>, H<sub>2</sub>–He, and H<sup>-</sup>. Once absorbed, the photons are considered destroyed — they contribute to the thermal pool and no longer can contribute to scattered light. The Monte Carlo code uses the atmospheric structure (density profiles) and opacities computed from the detailed plane-parallel radiative and convective equilibrium code of Seager & Sasselov, and computes scattering from a spherical planet with such an atmospheric structure. (As long as the scale-height of the atmosphere is small the plane-parallel approximation is sufficient to determine atmospheric structure). At visual wavelengths, the contribution of thermal emission from the planet is essentially zero and the reflected light can be treated as a scattering problem, where the incident radiation comes from the nearby star, and absorbed flux is ignored. Because condensate scattering is coherent we follow only one wavelength at a time. Because the CEGPs are so close to their parent stars that plane-parallel irradiation may not be accurate, we use the correct angular distribution of $`\mathrm{tan}^1(R_{}/D)`$ (see §4.3.2).
The Monte Carlo scattering method is preferable over “traditional” radiative transfer techniques because it can treat complex geometries, and its probabilistic nature gives all viewing angles at once. In the traditional plane-parallel method one can only solve along the line of sight, and must use the same angles for the incoming radiation as for the outgoing radiation (i.e. the emergent spectra). For the plane-parallel atmosphere models, one model must be computed for each phase angle and for each inclination. Our particular model atmosphere only considers isotropic or Rayleigh scattering, but realistically anisotropic scattering is important (see §4.3.2). Polarization is complicated and unnecessary when solving the model atmosphere for hydrostatic equilibrium, and for radiative and convective equilibrium in the traditional method.
Because of the need to use very large numbers of photons ($`10^7`$ to $`5\times 10^8`$) in order to fully sample the probability distribution function space, the Monte Carlo method cannot solve the model atmosphere problem (it is slow for optically thick regimes), although progress is being made in this direction for radiative equilibrium but with only a few line opacities (Bjorkman & Wood 2000).
## 4 Results and Discussion
As an example of a CEGP we use 51 Peg b (Mayor & Queloz 1995). There are many uncertainties about the CEGPs, including mass, radius, gravity, composition, T<sub>eff</sub>, etc. We have chosen only one example out of a large range of parameter space: $`M=0.47M_\mathrm{J}`$, log $`g`$ (cgs) = 3.2, metallicity that of the parent star, and $`R_P=1.2R_J`$. (Note that the CEGPs’ radii depend on mass, heavy element enrichment and parent star heating. Evolutionary models show $`R_P`$ for a given CEGP could be larger than the known radius of HD 209458b ($`R_P=1.4R_J`$) or as small as 0.9$`R_J`$ (Guillot 1999)). The incident flux of 51 Peg A (G2V, T$`{}_{\mathrm{eff}}{}^{}=5750`$, metallicity \[Fe/H\] = +0.21, and log $`g`$ (cgs) = 4.4 (Gonzalez 1998)) was calculated from the model grids of Kurucz (1992).
### 4.1 Atmospheric Structure and Emergent Spectra
We leave the detailed discussion of spectra and irradiative effects on temperature and pressure profiles for a separate paper. However, because the condensate assumptions affect the temperature-pressure profile and hence emergent spectra, we discuss the general properties here. These models supersede those in Seager & Sasselov (1998) since grain formation and grain opacities are considered, most noteably, TiO condenses out of the upper atmosphere.
#### 4.1.1 Theoretical Spectra: Main Characteristics
Figure 2 shows a model of 51 Peg b with a homogeneous cloud of MgSiO<sub>3</sub> particles with $`\overline{r}=0.01`$ $`\mu `$m, with $`T_{\mathrm{eff}}=1170`$ K. The effective temperature $`T_{\mathrm{eff}}`$ refers to the thermal emission only. The most noticeable feature is the large optical flux, many orders of magnitude greater than a blackbody (dotted line) of the same $`T_{\mathrm{eff}}`$. The CEGPs have negligible optical emission of their own, although the hottest ones at $`T_{\mathrm{eff}}=1600`$ K may have some emission $`>`$ 7000 Å, due to high absorption in the infrared which forces flux blueward. The CEGP in this model has 2–3 orders of magnitude more reflected flux than an isolated planet of the same $`T_{\mathrm{eff}}`$ has emitted flux. The CEGPs are at very different effective temperatures from their parent stars of $``$6000 K, so they have almost no molecular or atomic spectral features in common. Thus the spectral features in the blue and UV, blueward of $``$5200 Å, are largely spectral copies of the stellar spectra. Spectral features may also be reflected at longer wavelengths where no absorbers are present, for example H$`\alpha `$ at 6565 Å.
The reflected optical component of the spectrum in this model comes from reflection from a homogeneous cloud of solid grains of MgSiO<sub>3</sub> with particles with $`\overline{r}=0.01`$ $`\mu `$m. Rayleigh scattering from H<sub>2</sub> and He is negligible compared to the highly efficient scattering condensate, but plays a role deep in the atmosphere. The visual geometric albedo in this model is 0.18. The reflected stellar features between 4000 and 5200 Å follow the slope of the scattering coefficient of MgSiO<sub>3</sub>. In our models the condensate absorption and scattering features, such as the well known 10 $`\mu `$m feature in comet reflectance spectra, do not emerge in the CEGP spectra, since they occur where thermal emission of the planet is strongest.
The absorption line at 7670 Å is the K I $`4^2p`$$`4^2s`$ resonance doublet. Its broad wings extend for several hundred Å and are responsible for the slope redward to 1 $`\mu `$m. This effect is the cause of the large optical continuum depression in T dwarf spectra. (Tsuji et al. 1999; Burrows et al. 1999). This extreme broadening of the K I resonance doublet is also seen in cool L dwarf spectra (e.g. Tinney et al. 1999). Such broad atomic absorption of this kind — wings of thousands of Å — is not seen in any stellar atmosphere and indeed came as quite a surprise in the L dwarf observations. The cause is twofold: 1) strong pressure broadening of a fairly abundant species; and 2) there are no other strong absorbers in that wavelength region. The extreme broadening is not as surprising if we consider, for example, that if the Sun had no other absorbers than Lyman $`\alpha `$ (at 1215 Å) the wings would be visible out to the infrared (R. Kurucz, private communication). Other alkali metal lines are visible in the sample spectrum shown in Figure 2: Na I resonance doublet at 5893.6 Å ($`3^2p`$$`3^2s`$), Cs I at 8945.9 Å ($`6^2p_{1/2}`$$`6^2s`$) and 8523.5 Å ($`6^2p_{3/2}`$$`6^2s`$), Li I at 6709.7 Å ($`4^2p`$$`4^2s`$) . With very low ionization potentials — between 3.89 and 5.39 eV — the alkali metals are in neutral atomic form for much of the temperature-pressure regime in the CEGP atmosphere, although they do coexist with the gaseous metal chlorides and fluorides in the very upper atmospheres. The alkali metals are ionized in stars, and form alkali metal chloride solids in cooler planets such as Jupiter. Rb atomic lines should also be present but are not included in our model atmosphere. In principle Rayleigh scattering from Na I and K I could contribute a small amount to the scattering (Dalgarno 1968), but would only become important in condensate-free atmospheres.
The water bands are the most prominent absorption features in the infrared, with broad absorption troughs at 1.15, 1.4, 1.9, and 2.7 $`\mu `$m. Because the depth of the troughs is related to the temperature gradient, the spectral shape is expected to change depending on the amount of upper atmosphere heating, and to be different for irradiated planet atmospheres compared to isolated planet atmospheres. The infrared flux is thermal emission from absorbed and reradiated heat. Condensate absorption and scattering affects this wavelength region as well, as described in the next subsection.
The absorption trough at 3.3 $`\mu `$m is CH<sub>4</sub>, and more minor methane features are apparent at 1.6 and 2.3 $`\mu `$m. The methane lines are strong for this particular model. However as described in the next subsection (4.1.2) the presence of methane at all is very sensitive to the amount of heating in the upper atmosphere.
#### 4.1.2 Effects of Condensates on Spectra
The CEGP spectra are extremely dependent on the type, size, and amount of condensates in the planetary atmosphere, and Figure 2 represents only one specific model. Indeed it is impossible to predict the spectra, albedo, or light curve without referring to a specific condensate mix and size distribution. Figure 3 compares different low-resolution spectra at $`\alpha =0`$ of a subset of condensate cases considered in this paper. The curves are spectra from the following condensate assumptions: the solid line is a model with MgSiO<sub>3</sub> particles of mean radius $`\overline{r}=0.01\mu `$m (shown in Figure 2), the dotted line is a model with a MgSiO<sub>3</sub>-Fe-Al<sub>2</sub>O<sub>3</sub> mix of particles of $`\overline{r}=0.01\mu `$m, the dashed line is a model with a MgSiO<sub>3</sub>-Fe-Al<sub>2</sub>O<sub>3</sub> mix of particles of $`\overline{r}=0.1\mu `$m, and the dot-dashed line is a model with a MgSiO<sub>3</sub>-Fe-Al<sub>2</sub>O<sub>3</sub> mix of particles of $`\overline{r}=10\mu `$m.
The dotted curve is the most absorptive case which resembles a blackbody of $`T_{\mathrm{eff}}`$ close to $`T_{\mathrm{eq}}`$. Reflected features (not visible) appear at a very low magnitude. The most noticeable feature in the dashed curve is the broad dip between approximately 3000–10,000 Å. The cause is Rayleigh scattering from the condensates, which in this wavelength region have $`\overline{r}\lambda `$. The slope is $`\lambda ^4`$, but displaced compared to gaseous Rayleigh scattering since the Rayleigh scattering criterion is valid in a region of longer wavelength. The reflected spectral features are still visible on this Rayleigh scattering slope, in between the planetary atomic absorption lines. The K I and Na I lines are visible, but the extreme broadening shown in the solid line is not present because scattering and absorption high in the atmosphere means the deep atmosphere where the pressure broadening occurs is not sampled.
The most noticeable difference in the dot-dashed curve compared to the other spectra in Figure 3 is the presence of weak TiO features in the optical. Condensates with $`\overline{r}=10`$ $`\mu `$m have more scattering relative to absorption, and less absorption overall, right across the wavelength range. In this case, incoming radiation penetrates deep into the atmosphere, heating the atmosphere over a large depth to temperatures where enough TiO is present to produce weak absorption features.
CH<sub>4</sub> is an excellent temperature diagnostic for the CEGPs’ upper atmospheres. The strong CH<sub>4</sub> 3.3 $`\mu `$m band, and weaker CH<sub>4</sub> features at 1.6 and 2.3 $`\mu `$m are present only in the coolest, least absorptive model (solid line). The H<sub>2</sub>O bands also differ among the different models. They are much shallower for the atmospheres with absorptive condensates, due to absorption of incoming light by the condensates.
#### 4.1.3 Temperature-Pressure Profiles
The temperature-pressure profiles (which are the basis for the emergent spectra) also vary depending on the type and size distribution of condensates in the planet atmosphere. Figure 4 shows the temperature-pressure profiles of the four models shown in Figure 3, together with the equilibrium condensation curves. In contrast to the T dwarfs which do not need clouds to be modeled, the irradiated CEGPs have heated upper atmospheres that bring the temperature closer to the equilibrium condensation curves so are more likely to have clouds near the top of the atmosphere. In Figure 4 the model with particle mix with $`\overline{r}=0.1\mu `$m (dashed line) is highly absorbing and results in a temperature inversion in the upper atmosphere layers. The model with cloud with particles with $`\overline{r}=10\mu `$m is much less absorbing and results in a cooler temperature in the upper atmosphere layers. A highly reflective model (solid line) shows that much less heating occurs when molecules such as H<sub>2</sub>O are the primary absorbers. With the clouds at low pressure, at $`10^3`$$`10^4`$ dyne cm<sup>-2</sup>, the equilibrium condensation curves for MgSiO<sub>3</sub> and Fe are close together so a cloud mix of both particles could exist. Even if the uppermost cloud dominates the reflected light curve and spectra, the heating from lower cloud layers such as Al<sub>2</sub>O<sub>3</sub> are important and do alter the temperature-pressure profile. The $`T_{\mathrm{eff}}`$s of these models range from 1170 K to 1270 K.
Another interesting consequence of the irradiative heating, evident from Figure 4, is the proximity of the temperature-pressure profiles to the CO/CH<sub>4</sub> equilibrium curve. As noted in Goukenleuque et al. (1999) CO is expected to dominate over CH<sub>4</sub>, but CH<sub>4</sub> is abundant enough to produce absorption bands. We have found that the strength of the CH<sub>4</sub> features is sensitive to the upper atmosphere temperature which is in turn dependent on the amount of irradiative heating. Thus the CH<sub>4</sub> bands are a good temperature diagnostic. (The CH<sub>4</sub> bands are also useful, but less sensitive, as a pressure diagnostic (Seager 1999)).
Alternate approaches to modeling the temperature-pressure profiles have been taken. Marley et al. (1999) consider the temperature-pressure profile of an isolated object of the same effective temperature. Sudarsky et al. (1999) use ad hoc modified isolated temperature-pressure profiles (based on the temperature-pressure profiles in Seager & Sasselov 1998), to simulate heating, instead of computing irradiative heating. As a result, the temperature gradient is much steeper because most of the heat comes from the bottom of the atmosphere. These models have clouds at the 1 bar level, near the bottom of the atmosphere. One consequence of this assumption is the strength of the K I and Na I absorption. Because of the clear atmosphere down to the 1 bar level, the K I and Na I resonance lines are extremely pressure broadened and absorb essentially all incoming optical radiation redward of 500 nm. This is in contrast to the spectra shown in Figure 3 where the K I and Na I resonance lines are relatively narrow — the deep pressure zones where the broad line wings are formed are not sampled.
Although our approach is to compute the temperature-pressure profiles and reflected light in a consistent manner, we emphasize that in general there are many uncertainties in current CEGP models including photochemistry, cloud assumptions, and heat redistribution by winds. More specific to our models is the internal heat assumptions. For numerical reasons we must assume a lower boundary condition to our atmosphere in the form of a net flux coming from the planet interior. The assumption we have made is a net flux of approximately 1/10th of the absorbed flux. This may be too high (T. Guillot, private communication), and using a much lower value would produce a more isothermal atmosphere at the highest pressures in our models. More work is needed to understand the 3D heating redistribution in CEGPs. Importantly, the lower boundary condition has little effect on the upper atmospheric temperature and the reflected light curves.
### 4.2 Condensates and the Scattering Asymmetry Parameter
The shapes of the CEGP reflected light curves depend on the absorptivity and directional scattering probability of the condensates. Figure 5 shows the scattering asymmetry parameter and the single scattering albedo at 5500 Å for the three condensates considered as a function of particle size. The scattering asymmetry parameter $`g`$ is defined by
$$g=<\mathrm{cos}\theta _S>=_{4\pi }\mathrm{cos}\theta _SP^{11}\frac{d\mathrm{\Omega }}{4\pi },$$
(5)
where $`\theta _S`$ is the scattering angle and $`P^{11}`$ is the phase function (see e.g. Van de Hulst 1957). The scattering phase function is the directional scattering probability of condensates (see Figures 9 and 13), and should not be confused with the planetary phase function introduced in §2 The scattering asymmetry parameter $`g`$ varies from -1 to 1 and is 0 for isotropic scattering. The higher $`g`$ is, the more forward throwing the particle. The curves for $`g`$ and $`\stackrel{~}{\omega }`$ in Figure 5 can predict, or help interpret, the light curves. Small particles compared to wavelength scatter as Rayleigh scattering; $`g=0`$ in this case, where the forward and backward scattering average out. This is seen for particles with $`r=0.01`$ $`\mu `$m (along the $`y`$ axis). In addition, Al<sub>2</sub>O<sub>3</sub> and Fe have $`\stackrel{~}{w}=0`$ for $`r=0.01`$ $`\mu `$m, so for these small particles absorption dominates over all scattering and the light curves will show little variation. For particles with $`r=0.1`$ $`\mu `$m more scattering will occur than for $`r=0.01`$ $`\mu `$m; $`g=0.2`$ and $`\stackrel{~}{w}`$ is high. In this case, scattering is not too forward throwing, and the probability of scattering over absorption is high. For particles with $`r=10\mu `$m, both $`g`$ and $`\stackrel{~}{w}`$ are high. High $`g`$ means the particle will scatter light preferentially in the forward direction. Coupled with high $`\stackrel{~}{w}`$, the photons will multiply scatter forward into the planet resulting in little reflected light. These effects will be partially borne out in the light curves shown in the next section.
The single scattering albedos and the asymmetry parameter curves are generally similar for large $`r`$ for the three particles considered because the parameters — indeed light scattering in general — are determined largely by the particle size compared to the wavelength of light. The curves are different from each other because of the different nature of the particles, specifically the real and complex indicies of refraction. At 5500 Å, Fe has a very high complex index of refraction, while that of MgSiO<sub>3</sub> is essentially zero. The variations for a given curve, notably at $`r=0.5`$ $`\mu `$m, are interference effects between diffracted light rays and rays that refract twice through the particle. This effect gets damped out for absorptive particles (e.g. Fe). For a concise discussion of asymmetry parameters and phase functions see Hansen & Travis (1974).
Although we have chosen the dominant solids expected at the relevant temperatures and pressures from equilibrium calculations, it is certainly possible that nature has provided CEGPs with a different condensate size distribution, and different condensate particles and shapes than the ones used here. This will be investigated in future work.
### 4.3 Visual Photometric Light Curves and Polarization
In the next few subsections we present the photometric light curves and fractional polarization curves. The results from our Monte Carlo scattering code give a planet-star flux ratio, and we use equation (4) to convert to variation in $`\mathrm{\Delta }m`$, but where $`ϵ\varphi (\mathrm{\Theta })`$ is the planet-star flux ratio ($`\mathrm{\Theta }`$ is defined below). The results from the Monte Carlo scattering code also give the percent polarization, $`Pol`$, and we convert this to fractional polarization of the system as described in §2, by $`P_{\mathrm{frac}}=ϵ\varphi (\mathrm{\Theta })Pol`$, with the flux ratio in place of $`ϵ\varphi (\mathrm{\Theta })`$.
Figures 6a, 7a, 8a, and 11a show the 5500 Å light curves with orbital angle $`\mathrm{\Theta }`$ for the four cases of particles with mean radius $`\overline{r}=`$ 0.01, 0.1, 1, and 10 $`\mu `$m. Here we use orbital angle instead of orbital phase used for radial velocity measurements because an angular variable is more convenient for the analysis. We define $`\mathrm{\Theta }`$ as the angle in the orbital plane of the planet and star. With this definition, an orbital angle of 0 occurs when the planet is farthest from Earth, and an orbital angle of 180 occurs when the planet is between Earth and the star. In addition, for $`i=90^{}`$ orbital angle and phase angle are equivalent, and orbital angle 0 corresponds to orbital phase of 3/4 (used in radial velocity measurements). In each figure the first curve is for inclination $`i=90^{}`$, which for all CEGPs except HD 209458 b, has already been excluded by transit non-detections. The other curves are for $`i=82^{}`$, $`i=66^{}`$, $`i=48^{}`$, and $`i=21^{}`$, none of which can be excluded by transit non-detections. The $`y`$ axis scale differs among the different figures. Transits occur only for $`i\theta _T`$ ($`\theta _T=83.3^{}`$ for 51 Peg b with $`R_{}=1.16`$ $`R_{}`$, $`R_P=1.2`$ $`R_J`$, and $`D=0.051`$ AU), and within $`\mathrm{\Theta }=180^{}\pm (90\theta _T)`$. They are barely visible on these figures, since transits darken rather than brighten the light of the system. In addition, the transit light curves are on the order of millimagnitudes, $``$2 orders of magnitude greater than the reflected light effect. Nevertheless the start of the drop in the light curve for $`i=90^{}`$ is shown at $`\mathrm{\Theta }=173.7^{}`$ and $`\mathrm{\Theta }=186.3^{}`$ at first and fourth contact respectively. Similarly, for $`i=90^{}`$ — and only for $`i>\theta _T`$ — the reflected planetary light is not visible as the planet goes behind the star at $`\mathrm{\Theta }=360^{}(90^{}\theta _T)`$, and reemerges at $`\mathrm{\Theta }=(90^{}\theta _T)`$; that area is shaded in on the figures.
Figures 6b, 7b, 8b, and 11b show the fractional polarization with orbital angle for the four cases of mean particle radius 0.01, 0.1, 1, and 10 $`\mu `$m. Inactive solar-type stars are very weakly polarized, on the order of a few $`\times `$ $`10^2`$ percent, so we treat the incoming light as unpolarized. We have plotted the fractional linear polarization of the system as described in §2, so that the polarization is modulated by the amount of scattered light, which peaks at an orbital angle of zero. Circular polarization is a secondary effect, and smaller than the errors in our scattering simulation.
#### 4.3.1 Particles with $`\overline{r}=0.01\mu `$m
Figure 6a shows the light curve for particles with $`\overline{r}=0.01\mu `$m, the case of very small particle size compared to wavelength. The amount of scattered light is tiny due to the high absorptivity of Fe and Al<sub>2</sub>O<sub>3</sub>, as described in §4.2 That small particles obey Rayleigh scattering can be seen from the smooth, Rayleigh-like shape of the light curve. Rayleigh scattering produces more backscattering and results in a slightly different light curve from isotropic scattering. The Rayleigh scattering phase function $`P_{Ray}=3/4(1+\mathrm{cos}^2\theta _S)`$ (where $`\theta _S`$ is the scattering angle), whereas for purely isotropic scattering $`P_{iso}=1`$. At $`\mathrm{\Theta }=0^{}`$ and $`i=90^{}`$, $`P_{Ray}`$ = 1.5, compared to $`P_{iso}=1`$, and at $`\mathrm{\Theta }=90^{}`$ and $`i=90^{}`$, $`P_{Ray}`$ = 0.75 which is smaller than $`P_{iso}=1`$.
Rayleigh scattered light is maximally polarized at a scattering angle of 90. Figure 6b shows the fractional polarization of the CEGP system (described in §2), i.e. the ratio of polarized light to total white light of the star + planet. Because absorption is so high for this case, photons that exit the sphere have singly scattered, and the scattered light is 100% polarized. Even so, the fractional polarization is tiny because the amount of scattered light is very small compared to the unpolarized stellar flux. Figure 6b also shows that different inclinations have the same polarization peaks, but with smaller amplitudes. Fractional polarization at $`i=21^{}`$ is noisier than at other inclinations, since less radiation scatters into these phase angles, as indicated by the light curve in Figure 6a.
We also ran simulations with MgSiO<sub>3</sub> particles as the only condensate present in the planetary atmosphere (not shown in the figures), to investigate the lightcurves without the highly absorbing Fe and Al<sub>2</sub>O<sub>3</sub> condensates (see Table 2). For $`\overline{r}=0.01`$ $`\mu `$m at $`i=90^{}`$, the light curve peaks at 25 $`\mu `$mag, which corresponds to a geometric albedo $`p=0.18`$. Although $`\stackrel{~}{\omega }`$ is relatively high for this case, multiple scattering makes the resulting geometric albedo much smaller than for single scattering because it gives more chance for absorption. For the case where only MgSiO<sub>3</sub> with particles of $`\overline{r}=0.01`$ $`\mu `$m is present, the polarization fraction peaks at $`5.5\times 10^6`$, which is almost 2 orders of magnitude higher than the MgSiO<sub>3</sub>-Fe-Al<sub>2</sub>O<sub>3</sub> mix. However, it is still below current detectability limits.
#### 4.3.2 Particles with $`\overline{r}=0.1\mu `$m
The light curves for particles with $`\overline{r}=0.1\mu `$m, shown in Figure 7a, are similar to those for $`\overline{r}=0.01\mu `$m, but have a much larger amplitude. This can also be seen from Figure 5 which shows that for visual wavelengths the single scattering albedos are higher than those for $`\overline{r}=0.01\mu `$m, and the scattering asymmetry parameter is only 0.2, which is reasonably isotropic.
Polarization, shown in Figure 7b, is also similar to the $`\overline{r}=0.01\mu `$m case, with a much larger amplitude. Because the particles are still somewhat small compared to the wavelength of light, the scattering is largely Rayleigh scattering and the peaks are similar to those in Figure 6b. However, not shown is that the scattered light is 55% polarized.
For this case of particles with $`\overline{r}=0.1\mu `$m, $`\stackrel{~}{\omega }`$ of Fe dominates. As seen from Figure 5, if $`\stackrel{~}{\omega }`$ from MgSiO<sub>3</sub> or Al<sub>2</sub>O<sub>3</sub> dominates instead, the light curve will have a higher amplitude. For example, for a model with pure MgSiO<sub>3</sub> clouds, the $`i=90^{}`$ light curve peaks at 95 $`\mu `$mag, which corresponds to a geometric albedo $`p=0.69`$, and the fractional polarization peaks at $`8.6\times 10^5`$. The $`i=21^{}`$ light curve peaks at 42 $`\mu `$mag. This case has the highest reflectivity of all of our models.
#### 4.3.3 Particles with $`\overline{r}=1\mu `$m
For particles with $`\overline{r}=1\mu `$m, which are larger than visual wavelengths, the light curve shown in Figure 8a shows effects both from forward throwing and from a narrowly peaked backscattering function. Figure 9a shows the phase function for the three different condensates, plotted with scattering angle $`\theta _S`$. Although the phase function represents single scattering, it can be used to interpret the light curve which arises from multiple scattering. The narrow backscattering (at $`\theta _S=180^{}`$) is responsible for the narrow peak in the light curve: there is a high probability of backscattering but only for a narrow angular range. The high probability for forward throwing means that photons are likely to be forward scattered into the atmosphere where they will be absorbed; this is the cause for the otherwise reduced light curve (in the “wings”) compared to the Rayleigh-shaped light curves in Figures 6a and 7a.
We also plot the phase function as a polar diagram in Figure 9b, where the light is incoming from the left, and the condensate particle is marked at the origin. Figures 9a and 9b also show that the three condensates have different amounts of backscattering, indeed slightly different phase functions overall. In Figure 10a we plot the three light curves from each of the condensates, considering that each condensate is the only one present in the atmosphere. This shows that the $`\stackrel{~}{\omega }`$ of MgSiO<sub>3</sub> dominates, as compared to particles with $`\overline{r}=0.01\mu `$m where Fe dominates, and also that each condensate has unique properties. Both Fe and Al<sub>2</sub>O<sub>3</sub> have very forward throwing phase functions without a strong backward peak; the result is that incoming light is forward scattered into the atmosphere where it is not likely to contribute to reflected light. As a result, their light curve amplitudes are much smaller than MgSiO<sub>3</sub>’s; Al<sub>2</sub>O<sub>3</sub> alone would not even be detectable by the planned microsatellites. The same comparison for particles with $`\overline{r}=0.01\mu `$m and $`\overline{r}=0.1\mu `$m, where the phase functions and light curves are more isotropic, reveals that the main effect of each condensate is mostly a change in magnitude. This is because the particles are smaller than the wavelength of light, and to first order light is Rayleigh scattered.
The phase function of MgSiO<sub>3</sub>, the solid curve in Figure 9a, is typical of those for spheres of size parameter $`x`$ 1, with complex index of refraction $`n_i0`$ (see Hansen & Travis 1974), where $`x=2\pi r/\lambda `$. The forward throwing is caused by diffraction of light rays around the particle and depends on the geometrical cross-section of the particle, and so would also occur for non-spherical particles. The backward peak, known as the “glory”, is specific to spherical particles and is related to interfering surface waves on the particle sphere. For absorbing particles (with high $`n_i`$, such as Fe), this effect is damped out, as seen from the Fe phase function in Figure 9a, which shows no rise towards $`\theta _S=180^{}`$. For randomly oriented axi-symmetric spheroids, the backward peak would not be as severe (Mishchenko et al. 1997). The phase function of randomly oriented axi-symmetric spheroids depends on the distribution of both particle axis sizes and particle orientation. To estimate the light curve from a reduced backscattering peak, we assume all of the condensates scatter like Fe, and in another case all like Al<sub>2</sub>O<sub>3</sub>. In other words, we use the same total opacity of the MgSiO<sub>3</sub>-Fe-Al<sub>2</sub>O<sub>3</sub> mix. The resulting light curves are shown in Figure 10b. Although their peaks are much lower than the strong backscattering case, with the exception of Al<sub>2</sub>O<sub>3</sub> this variation is still detectable by the upcoming microsatellite missions.
Because the CEGPs are very close to their parent stars, light rays hitting the planet may not be well approximated as plane-parallel. We use the correct angular distribution of $`\mathrm{tan}^1(R_{}/D)`$ (6 for 51 Peg b). The main effect from using this angular distribution compared to plane-parallel rays is that the backscattering peak is reduced by 12%, because the backscattering phase function peaks sharply at $`180^{}`$. A more minor improvement is that the lower inclination light curves are a few percent lower. For isotropic or Rayleigh scattering we find no difference in the light curve from using either plane-parallel rays or the correct angular distribution of incoming radiation. Because $`\mathrm{tan}^1(R_{}/D)`$ is such a small angle, isotropic irradiation, used in atmosphere codes that treat feedback from the star’s own corona, is not accurate.
The polarized light curve shown in Figure 8b is very different from the Rayleigh scattering polarization curves for $`\overline{r}=0.01\mu `$m and $`\overline{r}=0.1\mu `$m. The polarization is more complex than Rayleigh scattering as the light rays reflect from and refract through the particles, and the scattered light rays interfere. In addition, the polarization from each condensate is different, since polarization depends in part on the index of refraction, which is very different for each of the three condensates in this study. The peak of the polarization has a similar peak to the light curve, since fractional polarization is plotted which follows the scattered light. Polarization of the scattered light alone shows a smaller central peak and additional smaller peaks at 10 and at 70 for $`i=90^{}`$.
#### 4.3.4 Particles with $`\overline{r}=10\mu `$m
Light curves from particles with $`\overline{r}=10\mu `$m, shown in Figure 11a, are similar to the light curves from $`\overline{r}=1\mu `$m particles, but with a more pronounced effect from strong forward throwing and an even more narrowly peaked backscattering probability. Outside of the backscattering peak, the light curve is much smaller due to the forward throwing effects discussed above. For $`i=90^{}`$ and $`\mathrm{\Theta }=160^{}`$, there is a rise in the light curve just before the transit. This is from light that enters the atmosphere near the limb, and scatters through the top of the atmosphere due to the high forward throwing nature of the large particles. For particles with $`\overline{r}=10\mu `$m, the polarization is very similar to the $`\overline{r}=1\mu `$m case, but with a greater peak from the greater amount of backscattering, and an otherwise lower amplitude from the higher forward throwing.
### 4.4 U, B, V, R Photometric Light Curves and Polarization
In this section we compare the light curves and polarization at the U, B, V, and R effective wavelengths (U = 3650 Å, B = 4400 Å, V = 5500 Å, R = 7000 Å). The incoming stellar flux has many spectral features over the wavelength range of a band, but the same features are reflected by the planet (Figure 3); the light curve depends on the flux ratio and the features cancel out. However, for a second order calculation absorption by alkali metal line wings (in the planetary atmosphere) or the opacity variation from condensates within a color band may play a role.
In general the light curves are a function of the opacity and of the phase function. The opacity effects include density effects and the single scattering albedo. Photons travel into the atmosphere and encounter condensates (or atoms or molecules) which will scatter or absorb them. If the photons scatter, they will scatter according to the condensate phase function. A phase function that preferentially scatters photons into the forward direction will generate a very different light curve than a Rayleigh phase function, as shown in §4.3 The phase function for given optical constants depends on the size parameter $`x=2\pi r/\lambda `$, and the condensate index of refraction.
Figure 12a shows the light curves of the 51 Peg b system for particles with $`\overline{r}=0.1\mu `$m. The main difference between the colors is caused by the different size parameters: different wavelengths of light for a fixed particle size. For example, $`x=0.90`$ at R but $`x=1.72`$ at U. The effects of this are seen in the light curves: R has a Rayleigh-scattering-like light curve compared to U. The phase functions for the 4 colors for Fe are shown in Figure 13. As mentioned in §4.3.2, Fe opacity dominates this case of particles with $`\overline{r}=0.1\mu `$m. In fact, the U light curve is close to the shape of the Fe-only light curve for $`\overline{r}=1\mu `$m particles at V (dashed line in Figure (10a)). Because of the $`x`$ dependency of the phase functions, the different colors for a fixed size go through the same shapes as shown for the V light curves which describe fixed wavelength for a varying particle size. However, there are differences due to opacity variation with color.
Figure 12b shows the polarization fraction for $`\overline{r}=0.1\mu `$m. Effects from both phase function and opacity contribute. The polarization peak of each color is at a different angle, due to the different indicies of refraction of the particles at different colors. The difference in the polarization peaks is much greater than the difference in the light curve peaks between the colors. The polarization curves reflect the higher asymmetry in the scattering in U and the greater absorption at this wavelength.
### 4.5 Cloud Layers
The light curves and polarization curves presented in this paper have been computed under the assumption that the clouds are not in layers, but that above a given equilibrium condensation curve all of the gas condenses into solids and the particles are suspended uniformly in the atmosphere.
There are two consequences of finite cloud layers. The first difference is that stratification will separate different condensates into different layers, where the light curve and flux signature will come largely from the cloud closest to the top of the atmosphere. In contrast to our models of vertically homogenous clouds, a cloud confined to one pressure scale height with the same solid mass fraction would have a higher optical depth. With a high optical depth the reflected light cloud signature is even more likely to resemble the uppermost cloud only than the situation of a homogenous mix.
The second difference is that the gaseous scatterers and absorbers above the cloud layer could play a role. The dominant gaseous scatterer is Rayleigh scattering by H<sub>2</sub>, and the dominant absorbers are alkali metals, particularly K I (7670 Å) and Na I (5894 Å). In our approximate models with cloud layers, the alkali metals are strong, but narrow. Because the irradiated temperature-pressure profiles cross the condensation boundaries at relatively low pressure, the strong pressure broadening of the alkali metals does not occur in our models.
With the choice of a finite cloud layer, an additional free parameter becomes the location of the cloud base. For example, in the case considered here the temperature-pressure curve could cross the Fe and MgSiO<sub>3</sub> condensation curves at low $`P`$, where they overlap. In this case a Fe-MgSiO<sub>3</sub> mix will prevail (see Figure 4). In contrast, as discussed in §4.1.3, if the cloud is low in the atmosphere (around 1 bar e.g. Sudarsky et al. 1999) the incoming radiation will be absorbed by broad lines of Na I and K I before reaching the scattering cloud, causing zero optical albedo redward of 500 nm.
We have rudimentarily explored finite cloud layers where all clouds are 1 pressure scale height above their base at the equilibrium condensation curve. The main difference in reflected light curves from our vertically homogenous assumption is an increase in magnitude. The reason is that MgSiO<sub>3</sub>, which has the coolest condensation curve (shown in Figure 4), would be the top layer; MgSiO<sub>3</sub> is both the most reflective (see $`\stackrel{~}{\omega }`$ in Figure 5) and has the largest backscattering probability of the three condensates considered here. The light curve shape changes little with the cloud layer models. In the case of small particles compared to wavelength ($`\overline{r}=0.01\mu `$m for 5500 Å), the shape of the phase function is Rayleigh-like for all three condensates. In the case for large particles compared to wavelength ($`\overline{r}=10\mu `$m for 5500 Å), $`\stackrel{~}{\omega }`$ of MgSiO<sub>3</sub> already dominates in the three condensate mix for large particles. However, we caution that much more work needs to be done both in cloud models and particle size distribution. We emphasize the difficulty in predicting the reflected light curves due to the large parameter space of irradiative heating, cloud models, and particle type and size distribution. Thus these exploratory models should be considered as a useful interpretative tool rather than a predictive tool.
### 4.6 Comparison with Observations
#### 4.6.1 Spectral Separation
Cameron et al. (1999) and Charbonneau et al. (1999) have given the first observational results for a CEGP atmosphere, $`\tau `$ Boo b. The results are marginally in conflict (see below); the first group claims a probable detection with a flux-ratio of $`ϵ=1.9\times 10^4`$ and the second group a null result with upper limit of $`ϵ=5\times 10^5`$. Although in this paper we are modeling 51 Peg b, we can assume to first order that $`\tau `$ Boo b has a similar atmosphere. However, $`\tau `$ Boo b is hotter than 51 Peg b and may be heated above the condensation boundary of some grain species.
To first order such observations are extremely useful in addressing whether or not there are reflective clouds near the top of the atmosphere. For example, if the probable Cameron et al. result is confirmed, $`\tau `$ Boo b must have a very reflective cloud of particles fairly high in the planet atmosphere. The reason for this is the orbital inclination was measured to be 29, meaning only small phase angles are observed during the planet’s orbit and the planet must be very bright to be detectable at all. If the particles are spherical, they must also be smaller than the wavelength of light, because the strongly peaked wavefunctions of highly reflecting spherical particles (e.g. particles with $`\overline{r}=10\mu `$m with light curve shown in Figure 11) generate a small light curve amplitude at low inclinations. Cameron et al. also find a wavelength dependence of the albedo. If confirmed this will reveal molecular absorbers in the atmosphere, but not cloud characteristics (which are grey at optical wavelengths).
The Charbonneau et. al result is inclination dependent, and provides useful constraints if the system is at high orbital inclination. Their upper limit of $`p=0.3`$ at high inclinations excludes atmospheres with extremely reflective clouds (such as pure MgSiO<sub>3</sub> clouds of small particles.) Bright models with light curve shapes different from isotropic, e.g. those shown in Figures 5 and 6 (which have $`p>0.3`$ for high $`i`$) are not excluded, because of the phase function assumption (see below). Their upper limit is based on the assumption that the reflected planetary light is an exact copy of the stellar light for 4668 to 4987 Å. For lower inclinations ($`i30^{}`$), Charbonneau et al. give an upper limit of $`1\times 10^4`$. This does not provide a useful model constraint because at low inclinations only small phases of the planet are visible, and only a few extreme cases of highly reflective clouds can be excluded.
Beyond characterizing the very general cloud reflectance property, the spectral separation observations cannot constrain the particle type or size distribution. One reason is that the observations measure a combination of the geometric albedo and planet area: $`p(R_P/D)^2`$. The planet radius is not known, except for a transiting planet. Cameron et al. assume a Jupiter-like albedo of $`p=0.55`$ and derive $`R_P=1.8R_J`$; Charbonneau et al. assume $`R_P=1.2R_J`$ (based on evolutionary models from Guillot et al. 1996) and derive an upper limit for $`p`$.
A second difficulty in using the spectral separation results to constrain cloud details is the phase function assumption that goes into the results. The results given by both groups are expressed as a flux ratio or geometric albedo at opposition. Neither group observed near opposition (“full phase” for $`i=90^{}`$), but instead extrapolated their results based on an assumed phase function. Cameron et al. used an empirically determined polynomial approximation to the phase function of Venus (Hilton 1992), which also resembles that of Jupiter (Hovenier & Hage 1989). Charbonneau et al. used the Lambert sphere phase function, which derives from isotropic scattering and is approximately valid for Uranus and Venus. The differences in $`p`$ from using these different phase function assumptions is 20% and this may be one contribution to the conflicting observational results. For more details see Cameron et al. (2000) and Charbonneau & Noyes (2000). The inherent uncertainty in the albedo measurement from both the phase function and the radius/albedo degeneracy prevents any serious constraint on atmosphere models. Nevertheless, for comparison Table 2 shows the geometric albedos of our 4 models at 4800 Å, roughly the center of both groups’ wavelength range.
Results from this paper show that the light curve observations needed to constrain atmosphere models are those at different colors in narrow wavelength bands. The specific anisotropic scattering properties for a given particle size distribution and grain indicies of refraction will determine the light curve. The indicies of refraction are both wavelength- and particle type-dependent, which is why the color dependence is important. Opacity effects in a narrow wavelength range are less important because the condensates are generally grey in the optical.
#### 4.6.2 Ground-based Photometric Light Curves
Observations by G. Henry (private communication) with ground-based automatic photometric telescopes can currently reach a precision of near 100 $`\mu `$mag and could be as precise as 50 $`\mu `$mag with a dedicated automatic photometric telescope. The precision is attainable with observations over many orbital periods because the phase effect is strictly repeating. With this limit, reflected light detections of high-orbital inclination systems should be possible, or at least useful constraints on the models. Furthermore, these photometric limits should allow confirmation of the Cameron et al. result. Their result of $`ϵ=1.9\times 10^4`$, which using the polynomial approximation to Venus (Hilton 1992) for $`i=29^{}`$ translates roughly into a flux-ratio of $`9\times 10^5`$ and an amplitude of $`7\times 10^5`$ which corresponds to $`80\mu `$mag.
### 4.7 Other EGPs
#### 4.7.1 Close-in EGPs
Because $`ϵ(R_P/D)^2`$, the flux variation of other CEGPs can be estimated using the data in Table 1, and multiplying the light curves by the $`(D_{51Peg}/D_{EGP})^2`$ ratio. However $`\mathrm{\Delta }m`$ is a flux ratio and is not affected by the magnitude of the parent star, although magnitude is observationally important. This estimate also assumes that the radii of the planets are the same (cf. Guillot et al. 1996). This estimate, shown for a Lambert sphere in Figure 1, shows a variation of a factor of 2 between $`D=0.042`$ AU and $`D=0.059`$ AU.
There are additional differences among the different CEGPs that affect scattering. One is from density. For example, 51 Peg b has almost an order of magnitude lower minimum mass than $`\tau `$ Boo b for the same planetary radius. A less dense atmosphere has a longer photon mean free path, which has two effects. One is more backscattered light. The second, which is minor, is more unscattered radiation passing through the limb, and less forward-scattered radiation traveling through the upper atmosphere. Coincidentally, some of the enhanced scattering effects gained from 51 Peg b’s lower surface gravity atmosphere compared to $`\tau `$ Boo b are lost with the larger distance from the parent star. With a lower surface gravity, 51 Peg b’s atmosphere is less dense (has a lower $`P_g`$) for the same Rosseland mean optical depth. Figure 14 shows the light curves for both $`\tau `$ Boo b and 51 Peg b, for the three condensate mix of particles with $`\overline{r}=10`$ $`\mu `$m, at $`i=82^{}`$, which is not excluded by a transit non-detection. Also plotted in Figure 14 is 51 Peg b’s light curve at $`D_{\tau Boo}`$. Because of its lower surface gravity atmosphere, 51 Peg b shows effects from more scattering; a higher backscattering peak ($`335^{}<\theta _S<25^{}`$), and more light scattered through the upper atmosphere ($`\theta _S>170^{}`$). The difference at $`25^{}<\theta _S<335^{}`$ is also due to the different atmosphere densities. Figure 14 shows that the observations would not be able to constrain the density; away from $`\mathrm{\Theta }=0^{}`$ the differences are very small, and near $`\mathrm{\Theta }=0^{}`$ the amplitude difference is degenerate with change in density, $`\stackrel{~}{\omega }`$, $`R_p`$, etc.
A planet with a larger radius than we have assumed, such as $`R_P=1.40\pm 0.17R_J`$ derived from the transit of HD 209458 b (Mazeh et al. 2000) would have a larger reflected light signal. The density effects described above would be a secondary effect.
Discrete cloud layers (not modeled here but discussed in §4.5) could have an effect on the CEGPs. The bases of different cloud types may be at different depths, as on Jupiter, because thermodynamical equilibrium calculations predict condensation curves with different temperature and pressure dependencies for each species (see Figure 4). CEGPs that have a hotter parent star or a smaller $`D`$ will have different temperature-pressure structures than the cooler ones; they may be different enough that different cloud layers are more or less visible. For example, we find that for particles with $`\overline{r}<1\mu `$m, $`\tau `$ Boo b’s atmosphere may be heated enough to have a temperature above that of MgSiO<sub>3</sub> condensation.
#### 4.7.2 EGPs Beyond 0.05 AU
Because we are investigating planets that could be detectable in reflected light in the near future, we focus on the CEGPs. However, for a very rough estimate of EGPs with $`D>0.05`$ AU, the light and polarization curves in this paper can be scaled by $`(D_{51Peg}/D_{EGP})^2`$ (since $`ϵ1/D^2`$). This rough estimate ignores the cloud particles, which would be different than those in the CEGP atmospheres. An EGP such as $`\rho ^1`$ Cnc, at 0.12 AU from its parent star, is 2.35 times as far as 51 Peg b from its parent star. The maximum amplitude light curve in this paper (60 $`\mu `$mag at $`i=90^{}`$) would be only be 11 $`\mu `$mag at the distance of $`\rho ^1`$ Cnc. This variation is barely detectable by the upcoming microsatellite missions. An EGP such as $`\rho `$ CrB at $`D=0.23`$ AU is 4.5 times farther from its parent star than 51 Peg b. The maximum amplitude light curve at this distance is only 3 $`\mu `$mag. Such an EGP will not be photometrically detectable in the foreseeable future. Thus the CEGPs have the brightest prospects for detection.
## 5 Summary and Conclusion
We have presented photometric light curves and fractional polarization curves for 51 Peg b for 4 mean particle sizes, and discussed the differences with color and with other CEGPs. The light curves are very sensitive to condensate type and size distribution, hence observations will be able to distinguish between extreme scenarios. However, more detailed information such as the exact size distribution and particle type will be more difficult to extract.
The temperature-pressure profiles are also extremely dependent on the condensate assumptions. We have briefly discussed these, along with the emergent spectra. In contrast to T dwarfs which have no observeable clouds, the CEGPs should have clouds closer to the top of their atmospheres because irradiation heats the upper atmosphere to temperatures closer to the equilibrium condensation curve (i.e. the cloud base). The condensates that may not contribute to reflected light because they are sequestered below the top cloud layer will still affect the temperature-pressure profile by heating the lower atmosphere. Thus observations of the light curves which should constrain the general cloud properties will help distinguish between atmosphere models.
The light curves may be very different from sine curves; their shape depends not just on the particle size and type, but also on $`\stackrel{~}{\omega }`$ and the atmospheric density. Inclinations other than the narrow angular range possible for transits are theoretically detectable for the CEGPs. Many cases of the light curves are detectable by upcoming space missions, and some of the largest amplitude cases (e.g. from pure MgSiO<sub>3</sub> particles with $`\overline{r}=0.1\mu `$m) might be detectable from the ground in the near future. See Table 2, and equation (4) using $`ϵ=p(R_P/D)^2`$. Results from this paper show that the light curve observations that would best constrain atmosphere models are those at different colors in narrow wavelength bands.
Geometric albedos at 5500 Å in this study for a cloud mix of particles of MgSiO<sub>3</sub>, Fe, and Al<sub>2</sub>O<sub>3</sub> range from $`p=0.44`$ for particles with $`\overline{r}=10\mu `$m that have a strong backscattering peak, to $`p=0.0013`$ for particles with $`\overline{r}=0.01\mu `$m which include highly absorbing Fe. Clouds of pure MgSiO<sub>3</sub> are much more reflective for all particle sizes (see Table 2).
The polarization of the CEGP systems is not detectable with current techniques. Rayleigh scattering polarization peaks at orbital angle 90, but with modulation of the reflected light the fractional polarization has an asymmetric peak at around 70. Polarization from particles that are large compared to the wavelength will have more than one peak due to interference effects from different light ray paths through the particle.
Other CEGP systems in our study give similar light curves to 51 Peg b, but effects from distance from the parent star and density are important. Planets farther than 0.1 AU from their parent stars are too faint in reflected light to be detected photometrically in the foreseeable future.
We emphasize that there are many unknowns in the model atmospheres, and much room for improvement — mainly more realistic cloud modeling, heat redistribution by winds, photochemistry, non-spherical particles, and other types of condensates. These ingredients will result in different light curves than those shown in this paper. That the predictions are so varied means observations should be able to identify the gross cloud characteristics. In this sense the theory should be seen as an interpretive rather than a predictive tool. Observations by the upcoming satellites MOST, COROT, and MONS and ground-based work will help constrain the CEGP atmosphere models and at best will reveal the nature of their atmospheres directly.
We thank Kenny Wood for illuminating discussions on the Monte Carlo scattering method, Dave Charbonneau for many useful discussions, Jens Falkesgaard for many of the Gibbs Free energy fits. We also thank Mark Marley, Bob Noyes, Mike Wolff, Greg Henry, and Adam Burrows for useful discussions. We thank the referee Tristan Guillot for helpful comments that improved the paper. SS is supported by NSF grant PHY-9513835, BAW acknowledges support by NAG5-8587, and DDS acknowledges support from the Alfred P. Sloan Foundation.
|
warning/0004/astro-ph0004030.html
|
ar5iv
|
text
|
# The Evolution of 3He, 4He and D in the Galaxy
## 1. Introduction
As discussed by Tosi (this volume), chemical evolution models are useful both to derive the primordial abundances of D, <sup>3</sup>He and <sup>4</sup>He and to give informations on stellar nucleosynthesis. In this work we show the predictions of the two-infall model (CMG) for the chemical evolution of the above elements in the solar vicinity and for their distribution along the galactic disk. We adopt a new version of the two-infall model which includes the contribution by novae enrichment and the new proposed mechanism of extra-mixing in low mass stars (Charbonnel, Sackman, this meeting). The model was calibrated to the solar galactocentric distance of 8 kpc (we were still adopting 10 kpc in CMG to better compare our predictions with the ones of Matteucci and François 1989). This model assumes two main infall episodes for the formation of the halo (and part of the thick disk) and thin disk, respectively. The timescale for the formation of the thin disk is much longer than that of the halo, implying that the infalling gas forming the thin disk comes not only from the halo but mainly from the intergalactic medium. The timescale for the formation of the thin disk is assumed to be a function of the galactocentric distance, leading to an inside-out picture for the Galaxy disk buildup. The two-infall model differs from other models in the literature mainly in two aspects: i) it considers an almost independent evolution between the halo and thin disk components (see also Pagel & Tautvaisiene 1995) and ii) it assumes a threshold in the star formation process (Kennicutt 1989). The last point has important consequences for the predicted abundance gradients (Chiappini, Matteucci & Romano 2000 - CMR).
## 2. Results
### 2.1. The solar Vicinity
Our present model differs from those of CMG in i) the adopted yields for the low and intermediate mass range stars which are now taken from van den Hoek & Groenewegen (1997) instead of Renzini & Voli (1981); ii) the fact that now we are including the explosive nucleosynthesis from nova outbursts (see Romano et al. 1999) and iii) the adopted solar galactocentric distance of 8 kpc. Moreover, in the present model we adopt a primordial helium-4 abundance of 0.241 (by mass) instead of 0.23 as recently suggested by Viegas et al. (1999).
Our model is in good agreement with what is called the minimum set of observational constraints, among which the most important is the G-dwarf metallicity distribution (see Tosi 2000 for a recent review; see Figure 1 and Tables 1 and 2).
The primordial abundances by mass of D and <sup>3</sup>He were taken to be 4.4 $`\times `$ 10<sup>-5</sup> and 2.0 $`\times `$ 10<sup>-5</sup> respectively. While the D primordial value is an upper limit (as can be seen in Figure 2a) the <sup>3</sup>He is a lower limit (see Figure 2b). As can be seen in Figure 2a, the observations of the local interstellar medium (ISM) and the solar system represent tight constraints to the deuterium primordial abundance. In fact, models that can reproduce the bulk of the observational data predict only a modest D destruction (in our case a factor $``$ 1.6; see Tosi et al. 1998).
With respect to our old predictions for the solar vicinity (CMG), we now have better agreement between the predicted solar abundances of <sup>3</sup>He, <sup>12</sup>C and <sup>4</sup>He (Table 2). The improvement in the predicted <sup>12</sup>C is due to the fact that we adopt the new nucleosynthetic yields from van den Hoek & Groenewegen (1997) for low and intermediate mass stars. Moreover, the adoption of a higher Y primordial abundance (where Y is the <sup>4</sup>He abundance by mass) leads to a better agreement with the solar value. The improvement in the predicted <sup>3</sup>He is due to the fact that we are taking into account the recently proposed extra-mixing mechanism in low mass stars ($`M<2.5M_{}`$, eg. Charbonnel & do Nascimento 1998). In Figure 2b we show models with different assumptions with respect to the extra-mixing mechanism. The model without extra-mixing clearly overproduces <sup>3</sup>He with respect to both solar and ISM observed abundances (even with the lower limit primordial abundance of <sup>3</sup>He adopted in the present model). The two other curves show our predictions for models which assume that 93% (solid line) and 99% (dashed-line) of stars with M$`<`$ 2.5 M completely destroy their <sup>3</sup>He, respectively. A model assuming that, when a star suffers extra-mixing (93%) a fraction of $``$ 1/100 of its <sup>3</sup>He is preserved, gives essentially the same result as the solid line model shown in Figure 2b. The predicted Y vs Z and Y vs O/H are shown in Figures 3a and 3b respectively.
### 2.2. The Galactic Disk
In this section we show our results for the predicted abundance distributions of D (Figure 4a), <sup>3</sup>He (Figure 4b) and <sup>4</sup>He (Figure 5). The predicted gradient for D is positive and steep. This is due to the faster evolution of the inner disk regions as compared with the outer parts (which are still in the process of formation thus having an almost primordial composition). In fact, of the various elements, D is probably the most sensitive to radial variations in the timescale of disk formation. A hope for the future is to have some D abundance measurements in regions outside the solar vicinity. This would certainly represent a very important constraint to the disk-formation mechanism !
In Figure 4b the predicted <sup>3</sup>He abundance gradient is shown. It can be seen that the assumption that 99% of the low mass stars suffer extra-mixing leads to a flat gradient (dashed line). The solid line (93% of low mass stars destroy their <sup>3</sup>He) is in marginal agreement with the data, but it is still acceptable. Again, in this case we see that this gradient is sensitive to the adopted timescales of disk formation. In fact, in the inner regions (older in the inside-out scenario) the contribution of low mass stars for the <sup>3</sup>He enrichment of the ISM has been more important than in the outer regions, and a negative gradient is predicted. More data on <sup>3</sup>He abundance at different galactocentric distances are welcome and would be very important to better constrain our models and the low-mass stellar nucleosynthesis as well. Finally, in Figure 5 the <sup>4</sup>He gradient is shown. The predicted gradient is $``$ $``$0.003 dex/kpc over the 4-14kpc galactocentric range. This value is in agreement with the results presented by Maciel (this meeting) based on disk planetary nebulae.
#### Acknowledgments.
C. C. thanks C. Charbonnel, M. Tosi, W. Maciel and B. Pagel for fruitful discussions during the meeting. The authors thank D. Romano who introduced in our code the new yields of van den Hoek & Groenewegen (1997). C. C. acknowledges financial support by CNPq and FAPERJ (Brazil).
## References
Anders, E. & Grevesse, N. 1989, Geochim. Cosmochim. Acta 53, 197
Charbonnel, C., do Nascimento, J.D. Jr. 1998, A&A 336, 915
Charbonnel, C. 2000 (this volume)
Chiappini, C., Matteucci, F., Gratton, R. 1997, ApJ 477, 765
Chiappini, C., Matteucci, F., Romano, D. 2000 (in preparation)
Geiss, J. & Gloeckler, G. 1998, in Primordial Nuclei and their Galactic evolution, N. Prantzos, M. Tosi & R. von Steiger eds., Space Sci. Rev. 84, 239
Kennicutt, R. C. Jr. 1989, ApJ 344, 685
Linsky, J. L. 1998, in Primordial Nuclei and their Galactic evolution, N. Prantzos, M. Tosi & R. von Steiger eds., Space Sci. Rev. 84, 239
Maciel, W. J. 2000 (this volume)
Matteucci, F. & François, P. 1989, MNRAS 239, 885
Pagel, B.E.J. & Tautvaisiene, G. 1995, MNRAS 276, 505
Renzini, A. & Voli, M. 1981, A&A 94, 175
Rocha-Pinto, H.J. & Maciel, W. J. 1996, MNRAS 279, 447
Romano, D., Matteucci, F., Molaro, P., Bonifacio, P. 1999, A&A 352, 117
Sackman, J. 2000 (this volume)
Tosi, M. 2000 (this volume)
Tosi, M. 2000, in The chemical evolution of the Milky Way: stars versus clusters, F. Giovanelli and F. Matteucci eds., Kluwer (in press)
Tosi, M., Steigman, G., Matteucci, F., Chiappini, C. 1998, ApJ 498, 226
van den Hoek, L. B. and Groenewegen, M. A. T. 1997, A&AS 123, 305
Viegas, M. S., Gruenwald, R. & Steigman, G. 1999 (astroph/9909213)
|
warning/0004/math-ph0004001.html
|
ar5iv
|
text
|
# Existence and properties of 𝑝-tupling fixed points
## 1 Introduction
Two problems have a strong resemblance, and have both found their origin in the theory of period doubling for maps of the interval \[F1, F2, CT\]. The first is to prove the existence and properties of solutions of the $`(p+1)`$-Cvitanović-Feigenbaum functional equation, i.e. fixed points of the $`(p+1)`$-tupling operator $`_{p+1}`$:
$$g(x)=(_{p+1}g)(x)=\frac{1}{\lambda }g^{p+1}(\lambda x),g(0)=1.$$
(1.1)
Here $`g`$ is required to be an even, $`𝒞^1`$ map of $`[1,1]`$ into itself, strictly decreasing on $`[0,1]`$ and $`\lambda =g^{p+1}(0)`$ is required to be in $`(0,1)`$. More precisely, the restrictions $`g_+`$ and $`g_{}`$ to $`[0,1]`$ and $`[1,0]`$, respectively, must satisfy
$$g_+=\frac{1}{\lambda }g_{}^{p1}g_+^2(\lambda ),g(0)=1.$$
(1.2)
Denoting $`u`$ the inverse function of $`g_+`$, and $`\stackrel{ˇ}{u}(z)=u(z)`$, this can be reexpressed as
$$u=\frac{1}{\lambda }u\stackrel{ˇ}{u}^p\lambda .$$
(1.3)
We shall also require $`g_+`$ to have the form
$$g_+(x)=f(x^r)x[0,1],$$
(1.4)
where $`r>1`$ is a real number, and $`f`$ is real-analytic on $`[0,1]`$, with $`f^{}(x)<0`$ on this closed interval. The class of those $`g`$ having the property (1.4) for a fixed $`r`$ is left invariant by $`_{p+1}`$.
The second problem is to prove the existence and properties of solutions of the system
$$\begin{array}{ccc}\eta & =& \frac{1}{\lambda }\eta ^p\xi (\lambda ),\\ \xi & =& \frac{1}{\lambda }\eta (\lambda ),\xi (0)=1,\end{array}\lambda =\eta (0)(0,1).$$
(1.5)
Here $`\xi `$ is a real $`𝒞^1`$, strictly increasing function defined on a certain interval $`[L,0]`$ of the negative real axis ($`L>1`$) and satisfies $`\xi (x)>x`$ on this interval. Again $`\xi `$ is required to be of the form
$$\xi (x)=f(|x|^r)x[L,0],$$
(1.6)
where $`r>1`$ is a real number, and $`f`$ is real analytic without critical points on $`[0,L^r]`$. Let $`u`$ be the inverse function of $`\xi `$, and $`\stackrel{ˇ}{u}(z)=u(z)`$. Then (1.5) implies
$$u=\frac{1}{\lambda ^2}u\lambda \stackrel{ˇ}{u}^p\lambda .$$
(1.7)
The system (1.5) is part of the theory, initiated in \[FKS\] and \[ORSS\], of critical circle mappings whose rotation number has the continued fraction expansion $`[p,p,\mathrm{},p,\mathrm{}]`$. It is natural to attempt a unified treatment of the two functional equations (1.3) and (1.7) by introducing an interpolating parameter $`\nu [1,2]`$ and considering the functional equation
$$u=\frac{1}{\lambda ^\nu }u\lambda ^{\nu 1}\stackrel{ˇ}{u}^p\lambda .$$
(1.8)
As a device for avoiding repetitions, this works rather well for $`p=1`$, (\[EE, E2, E3\]). It is much less effective, as we shall see, for $`p>1`$. It is also of some interest to consider the case when $`\nu <1`$. The history of this subject is long, even if restricted to rigorous results (see e.g. \[L1, L2\]), and the literature has experienced a veritable explosion in recent times. For the case of interval maps, the paper of M. Lyubich \[Ly\] (a kind of culminating point) contains a historical note and references to which I refer the reader. For the case of circle maps, the reader is referred to the paper of M. Yampolsky \[Y\] and to references therein. However the literature has tended to concentrate on the case of integer $`r`$, with notable exceptions such as \[CEL, JR, M, MO\]. Another, most important exception is the whole theory of “real a priori bounds” (see \[dMvS, S1, S2, Sw, dFdM\] and other references given in \[Ly, Y\]). In this paper, we look for solutions of the functional equations (1.8), for arbitrary real $`r`$, which are subjected to some additional constraints (see Section 3). All the available theoretical and numerical evidence indicates that, for each $`\nu [1,2]`$, each $`r>1`$, and each $`p1`$, there is one and only one solution obeying all the constraints. This suggests that the solution (and in particular $`\lambda `$) must depend analytically on the parameters $`\nu `$, $`r`$, and $`p`$. For $`p=1`$, it has been proved in \[E3\] that solutions exist for all $`\nu [1,2]`$ and all $`r>1`$, and the proof extends without any change to the case $`\nu (0,1]`$ provided $`r\nu >1`$. In the case $`\nu =1`$, the existence of solutions for all $`p`$ and all $`r>1`$ has been proved by M. Martens \[M\], whose results go much farther since they include all possible periodic points and kneading sequences. In this paper, the existence of solutions will be proved, by another method, in the case $`1\nu 2`$, for all $`r>1`$ and all (integer) $`p1`$. It will be seen that in the case $`0<\nu 1`$, the condition $`r\nu 1(p1)(1\nu )>0`$ is necessary and sufficient for the existence of solutions. This work had remained unfinished for a long time<sup>1</sup><sup>1</sup>1The case $`\nu =1`$, all $`r`$ and $`p`$, was presented at the Meeting on new developments in Mathematical Physics and Neuroscience (Hunziker-Hepp Fest), ETH, Zurich, 21-23/9/1995. when I belatedly became aware of the paper of B. Mestel and A. Osbaldestin \[MO\], devoted to the proof of the existence of a period 2 point of the doubling operator for non-even maps (with arbitrary $`r>1`$). One of the ideas in that paper allowed me to finish the proof of existence in the case $`\nu >1`$ (see Subsection 6.2).
Section 2 collects some notations and well-known or straightforward facts (see \[D, V\]). Sections 3-6 contain the proofs of existence. In Section 7 some properties of the solutions are derived (univalence, boundedness). In Section 8 it is shown that for $`\nu (0,1]`$, when $`r`$ tends to infinity the solutions behave similarly to those of the case $`p=1`$ (see \[EW, E1, EE\]).
Acknowledgement. I wish to thank Oscar Lanford and Michael Yampolsky for many helpful discussions. I am also indebted to O. Lanford for his kind permission to include the contents of Subsection 7.4.
## 2 Notations and preliminaries
1. We denote $`𝐂_+=𝐂_{}=\{z𝐂:\mathrm{Im}z>0\}`$. A function $`f`$ is a Herglotz or Pick function \[D\] (and $`f`$ is an anti-Herglotz function) if it is holomorphic in $`𝐂_+𝐂_{}`$, $`f(z^{})=f(z)^{}`$, and $`f`$ maps $`𝐂_+`$ (resp. $`𝐂_{}`$) into its closure, $`f(𝐂_\pm )\overline{𝐂_\pm }`$. If $`f`$ is also holomorphic on a real non-empty open segment $`(a,b)`$, then, for each $`x(a,b)`$, and each $`N𝐍`$ the $`N\times N`$ matrix $`M`$ with components $`M_{jk}=D^{j+k+1}f(x)/(j+k+1)!`$, ($`0j,k<N`$), is positive. This follows immediately from the Herglotz integral representation theorem (\[D\], pp. 20 ff.). The case $`N=2`$ shows that if $`f`$ is not a constant, then $`f^{}(x)>0`$ $`x(a,b)`$, and $`f`$ has non-negative Schwarzian derivative $`Sf=(f^{\prime \prime }/f^{})^{}(f^{\prime \prime }/f^{})^2/2`$ in $`(a,b)`$. Denote $`v=f^{\prime \prime }/f^{}`$ and suppose $`a<x<y<b`$. If $`v`$ does not vanish in $`[x,y]`$,
$$\frac{1}{v(x)}\frac{1}{v(y)}\frac{yx}{2}.$$
(2.9)
If $`v(x)>0`$ then $`v(y)>0`$ since $`v`$ is increasing, hence $`v(x)2/(yx)`$, which also holds if $`v(x)0`$. Similarly $`v(y)2/(yx)`$. Letting $`y`$ tend to $`b`$ or $`x`$ tend to $`a`$, we find:
$$\frac{2}{za}\frac{f^{\prime \prime }(z)}{f^{}(z)}\frac{2}{bz}z(a,b).$$
(2.10)
If $`f((a,b))`$ has a finite upper bound, then $`f|(a,b)`$ extends continuously to $`(a,b]`$ with $`f(b)=supf((a,b))`$, and similarly if there is a finite lower bound. If $`f`$ maps $`[a,b]`$ into $`(a,b)`$, it has a fixed point in $`(a,b)`$ which (by Schwarz’s lemma) is unique and attractive; in this case every subinterval of $`(a,b)`$ which contains the fixed point is mapped into itself by $`f`$. If $`F`$ is an increasing function with non-negative Schwarzian on $`(0,+\mathrm{})`$ (in particular if $`F`$ is a Herglotz function holomorphic in $`𝐂_+𝐂_{}(0,+\mathrm{})`$), then its restriction to $`(0,+\mathrm{})`$ is concave as a special case of (2.10). The following corollary will be needed later:
###### Lemma 1
Let $`A<a<b<B`$ be real numbers, and $`f`$ be a Herglotz function holomorphic in $`𝐂_+𝐂_{}(A,B)`$. Then, for each $`z(a,b)`$,
$$f(z)\frac{(Bb)(za)f(b)+(Ba)(bz)f(a)}{(ba)(Bz)}.$$
(2.11)
Proof. We define $`F(z)=f(Bz^1)`$, i.e. $`f(z)=F(1/(Bz))`$. Then $`F`$ is a Herglotz function holomorphic in $`𝐂_+𝐂_{}(1/(BA),+\mathrm{})`$. Setting now $`a^{}=1/(Ba)`$, $`b^{}=1/(Bb)`$ and $`z^{}=1/(Bz)`$, with $`z(a,b)`$, the concavity of $`F`$ implies
$$F(z^{})\frac{z^{}a^{}}{b^{}a^{}}F(b^{})+\frac{b^{}z^{}}{b^{}a^{}}F(a^{}),$$
(2.12)
which translates back into (2.11).
2. Let $`A`$, $`B`$, $`A^{}`$, $`B^{}`$ be strictly positive real numbers. Then the homographic function
$$zm(z;A,B,A^{},B^{})=\frac{z\left(\frac{1}{A}+\frac{1}{B}\right)}{z\left(\frac{1}{AB^{}}\frac{1}{BA^{}}\right)+\frac{1}{A^{}}+\frac{1}{B^{}}}$$
(2.13)
is a bijection of $`𝐂_+𝐂_{}[A,B]`$ onto $`𝐂_+𝐂_{}[A^{},B^{}]`$, and fixes $`0`$. Its derivative at $`0`$ is
$$m^{}(0;A,B,A^{},B^{})=\frac{A^{}B^{}(A+B)}{AB(A^{}+B^{})}.$$
(2.14)
Hence if $`f`$ is a holomorphic map of $`𝐂_+𝐂_{}(A,B)`$ into $`𝐂_+𝐂_{}(A^{},B^{})`$ which fixes $`0`$, it follows from Schwarz’s lemma applied to $`m^1f`$, that $`|f^{}(0)|m^{}(0;A,B,A^{},B^{})`$.
3. Let $`b,s`$ be real numbers, with $`0<b<1`$, and $`s>1`$. Then the homographic function
$$z\chi _{b,s}(z)=1+b^{s1}\frac{b(1+b)(z1)}{1+bb^2(z1)}$$
(2.15)
is holomorphic in $`\mathrm{\Omega }(1/b,1/b^2)`$, Herglotzian, and
$$\chi _{b,s}(1/b)>0,\chi _{b,s}(1/b^2)<1/b^2,\chi _{b,s}(1)=1,\chi _{b,s}^{}(1)=b^s.$$
(2.16)
4. In the sequel, if $`s`$ and $`t>s`$ are real numbers, we shall denote $`\mathrm{\Omega }(s,t)`$ the domain
$$\mathrm{\Omega }(s,t)=𝐂_+𝐂_{}(s,t).$$
(2.17)
If $`u_{}`$ and $`u_+`$ are two real numbers in $`(0,1)`$, we denote $`𝐄_0(u_{},u_+)`$ the space of functions $`\psi `$, holomorphic and anti-Herglotzian in $`\mathrm{\Omega }(1/u_{},1/u_+)`$, and such that $`\psi (0)=1`$, $`\psi (1)=0`$. Such a function has an integral representation
$$\mathrm{log}\psi (z)=_{𝐑(1/u_{},1)}\sigma (t)\left[\frac{1}{t}\frac{1}{tz}\right]𝑑t,z\mathrm{\Omega }(1/u_{},1).$$
(2.18)
Here $`\sigma `$ is an $`L^{\mathrm{}}`$ function with $`0\sigma 1`$ and $`\sigma (t)=1`$ for all $`t[1,1/u_+]`$. It follows that $`\psi `$ satisfies the following inequalities:
$$\begin{array}{ccc}\frac{\psi (z)(1u_+z)}{1z}& 1\frac{\psi (z)(1+u_{}z)}{1z}& \mathrm{for}\mathrm{all}z(0,1/u_+)\{1\},\\ & & \mathrm{reversed}\mathrm{for}z(1/u_{},0),\end{array}$$
(2.19)
$$\begin{array}{ccc}\frac{1u_+}{(1z)(1u_+z)}& \frac{\psi ^{}(z)}{\psi (z)}& \frac{1+u_{}}{(1z)(1+u_{}z)}\hfill \\ & & \mathrm{for}\mathrm{all}z(1/u_{},1/u_+)\{1\}.\hfill \end{array}$$
(2.20)
Suppose now that $`\psi 𝐄_0(u_{},u_+)`$ has a finite upper bound $`M`$ on $`(u_{},u_+)`$. (By (2.19), $`M`$ must satisfy $`M\psi (1/u_{})(1+u_{})/(u_++u_{})`$ .) Then
$$_{𝐑(1/u_{},1)}\frac{\sigma (t)dt}{t(1+u_{}t)}\mathrm{log}M,$$
(2.21)
Therefore, if $`1/u_{}<z<0`$,
$`{\displaystyle \frac{\psi ^{}(z)}{\psi (z)}}`$ $`=`$ $`{\displaystyle _{𝐑(1/u_{},1)}}{\displaystyle \frac{\sigma (t)}{t(1+u_{}t)}}{\displaystyle \frac{t(1+u_{}t)}{(tz)^2}}𝑑t`$ (2.22)
$``$ $`(\mathrm{log}M)\underset{t(1/u_{},1)}{\mathrm{max}}{\displaystyle \frac{t(1+u_{}t)}{(tz)^2}}{\displaystyle \frac{\mathrm{log}M}{(4z)(1+u_{}z)}}.`$
## 3 More precise statement of the problem
We begin with a few heuristic considerations. It is easy to see that if $`u`$ is a solution of (1.8), it will analytically extend to the real interval $`(1/\lambda ,1)`$. Moreover, since the function $`f`$ is analytic without critical point in an open real interval containing 0, its inverse function, denoted $`U`$, will be analytic, with strictly negative derivative, in an open interval containing 1. The functions $`u`$ and $`U`$ are related by $`U(z)=u(z)^r`$. Thus
$$U(z)=\frac{1}{\lambda ^{r\nu }}U(\lambda ^{\nu 1}\stackrel{ˇ}{u}^p(\lambda z))$$
(3.23)
should hold wherever both sides are analytic. The main condition which we impose on the solutions we seek is that $`u`$ and $`U`$ be anti-Herglotz functions. It is in fact sufficient to impose this condition on $`u`$. Indeed denote
$$\phi =\lambda ^{\nu 1}\stackrel{ˇ}{u}^p\lambda .$$
(3.24)
Then $`\phi `$ is Herglotzian and the equation (3.23), rewritten as
$$U(z)=\frac{1}{\lambda ^{r\nu }}U(\phi (z))$$
(3.25)
shows first that $`\phi (1)`$ must be a zero of $`U`$, i.e. that $`\phi (1)=1`$, and then that $`U`$ is a linearizer of $`\phi `$ at 1, the multiplier $`\phi ^{}(1)`$ being equal to $`\lambda ^{r\nu }<1`$. Therefore $`U`$ is also anti-Herglotzian, and is holomorphic in the basin of attraction of 1 for $`\phi `$. The reasons for imposing the Herglotz condition have been given e.g. in \[EE, E2\]. It is more convenient to work with $`\psi =U/U(0)`$ rather than with $`U`$, and we denote $`z_1/\lambda ^{\nu 1}`$ the quantity $`u(0)`$. This implies $`z_1=\lambda ^{\nu 1}\stackrel{ˇ}{u}(0)\phi (0)<1`$. We thus adopt the following definition:
Given two real numbers $`\nu (0,2]`$, $`r>1`$, and an integer $`p1`$, a solution associated with these values consists of two functions $`\psi `$ and $`u`$ and two real numbers $`\lambda (0,1)`$ and $`z_1(0,1)`$ with the following properties:
(1) $`\psi `$ is an anti-Herglotzian function holomorphic in $`\mathrm{\Omega }(1/\lambda ,1/a)`$ for some $`a(0,1)`$ with $`\psi (1)=0`$ and $`\psi (0)=1`$.
(2) $`u`$ is holomorphic and anti-Herglotzian in $`\mathrm{\Omega }(1/\lambda ,1)`$, and is given there by
$$u(z)=\frac{z_1}{\lambda ^{\nu 1}}\psi (z)^{1/r}.$$
(3.26)
(3) The identity
$$\psi (z)=\frac{1}{\lambda ^{r\nu }}\psi (\lambda ^{\nu 1}\stackrel{ˇ}{u}^p(\lambda z))$$
(3.27)
holds for all $`z\mathrm{\Omega }(1/\lambda ,1/a)`$, where again $`\stackrel{ˇ}{u}(z)=u(z)`$.
The following theorem will be proved.
###### Theorem 1
(i) For any integer $`p1`$ and any real $`\nu (0,1]`$, there exist solutions if and only if $`r`$ satisfies
$$r\nu 1(p1)(1\nu )>0.$$
(3.28)
(ii) For any integer $`p1`$ and any real $`\nu [1,2]`$, there exist solutions for all real $`r>1`$.
The necessity of the condition (3.28) will be shown in the next section, but it is easy to see that the conditions we have imposed require $`r\nu >1`$. It suffices to consider the case $`0<\nu 1`$. In this case, we must have
$$\psi (1/\lambda )=\frac{1}{\lambda ^{r\nu }}\psi (\phi (1/\lambda ))\frac{1}{\lambda ^{r\nu }}u(1/\lambda )\frac{z_1}{\lambda ^{2\nu 1}}<\frac{1}{\lambda },$$
(3.29)
from which it follows that $`\phi =\lambda ^{\nu 1}\stackrel{ˇ}{u}^p\lambda `$ is analytic in $`\mathrm{\Omega }(1/\lambda ,1/\lambda ^2)`$ and that:
$$\phi (1/\lambda )0,\phi (1)=1,\phi (1/\lambda ^2)\frac{z_1}{\lambda ^\nu }.$$
(3.30)
We can apply Remark 2 of Section 2, i.e. Schwarz’s lemma as in (2.14) to bound $`\phi ^{}(1)`$, with $`A=1+1/\lambda `$, $`B=1/\lambda ^21`$, $`A^{}=1`$, $`B^{}=\lambda ^\nu 1`$ and find
$$\phi ^{}(1)=\lambda ^{r\nu }\frac{\lambda (1\lambda ^\nu )}{(1\lambda ^2)}\frac{\lambda }{1+\lambda }.$$
(3.31)
This implies $`\lambda ^{r\nu 1}<1`$, i.e. $`r\nu >1`$.
In the case $`\nu >1`$, it is well-known (see \[JR\]), and easy to verify, that for $`r=1`$, $`p1`$, there is a solution such that $`\psi (z)=1z`$, all functions $`u`$, $`\phi `$, etc. are affine, $`z_1=\lambda ^{\nu 1}`$, and $`\lambda `$ is the unique solution in $`(0,1)`$ of
$$\lambda ^\nu +p\lambda ^{\nu 1}1=0.$$
(3.32)
Moreover it is proved in \[JR\] that (in the case $`p=1`$, $`\nu =2`$) there exist solutions for all sufficiently small $`r1>0`$.
The proof of Theorem 1 will occupy Sections 4-6. Many repetitions occur in these sections, since variations of the same method apply to several cases. But avoiding the repetitions would produce more obscurity than brevity.
## 4 Existence for $`r>1`$, $`p1`$ and $`\nu (0,1]`$
In this section, $`r>1`$ and $`\nu (0,1]`$ are fixed real numbers such that $`r\nu >1`$, and $`p1`$ is a fixed integer. The real number $`b(0,1)`$ is also fixed, but its value will be chosen later (as a function of $`r`$). For any two $`s,t(0,1)`$, we denote
$$h_{s,t}(z)=\frac{z(s+1)}{z(st)+1+t}\{\begin{array}{ccc}0& & 0\\ 1& & 1\\ 1/s& & 1/t\end{array}$$
(4.33)
Obviously $`h_{s,t}=h_{t,s}^1`$.
We denote $`𝒬_0(b,r\nu )`$ the space of all functions $`\mathrm{\Phi }`$ with the following properties:
(Q1) $`\mathrm{\Phi }`$ is a Pick function holomorphic in the domain:
$$\mathrm{\Omega }(1/b,1/b^2)=𝐂_+𝐂_{}(\frac{1}{b},\frac{1}{b^2}),$$
(4.34)
and maps this domain into itself.
(Q2) $`\mathrm{\Phi }(z)0`$ for all $`z(1/b,1/b^2)`$,
(Q3) $`\mathrm{\Phi }(1)=1`$, and $`0<\mathrm{\Phi }^{}(1)b^{r\nu }`$.
We regard $`𝒬_0(b,r\nu )`$ as a subset of the real Fréchet space of the self-conjugated functions holomorphic in $`\mathrm{\Omega }(1/b,1/b^2)`$, equipped with the topology of uniform convergence on compact subsets. We shall define a continuous operator $`B(b,r,p,\nu )`$ on this space by describing its action on an arbitrary $`\mathrm{\Phi }_0𝒬_0(b,r\nu )`$.
Given $`\mathrm{\Phi }_0𝒬_0(b,r\nu )`$, we denote $`\lambda =\mathrm{\Phi }_0^{}(1)^{1/r\nu }`$. By (Q3), $`0<\lambda b`$. We define a function $`\phi _0`$ by
$$\phi _0=h_{b,\lambda }\mathrm{\Phi }_0h_{b,\lambda }^1.$$
(4.35)
If $`\lambda =b`$, $`h_{b,\lambda }`$ is the identity. Otherwise, since $`\lambda <b`$, its pole is below $`1/b`$ and $`h_{b,\lambda }`$ maps $`\mathrm{\Omega }(1/b,1/b^2)`$ onto $`\mathrm{\Omega }(1/\lambda ,1/a_0(\lambda ))`$ where
$$\frac{1}{a_0(\lambda )}=h_{b,\lambda }\left(\frac{1}{b^2}\right),$$
$$b^2a_0(\lambda )=b\lambda (1b)<b.$$
(4.36)
The function $`\phi _0`$ possesses the following properties:
(Q$`{}_{}{}^{}1`$) $`\phi _0`$ is a Pick function holomorphic in $`\mathrm{\Omega }(1/\lambda ,1/a_0(\lambda ))`$, and maps this domain into itself.
(Q$`{}_{}{}^{}2`$) $`\phi _0(z)0`$ for all $`z(1/\lambda ,1/a_0(\lambda ))`$.
(Q$`{}_{}{}^{}3`$) $`\phi _0(1)=1`$, and $`\phi _0^{}(1)=\lambda ^{r\nu }`$.
We denote $`\psi `$ the linearizer of $`\phi _0`$ normalized by the condition $`\psi (0)=1`$, i.e. the unique function, holomorphic in $`\mathrm{\Omega }(1/\lambda ,1/a_0(\lambda ))`$, such that
$$\psi (z)=\frac{1}{\lambda ^{r\nu }}\psi (\phi _0(z))z\mathrm{\Omega }(1/\lambda ,1/a_0(\lambda )),\psi (1)=0,\psi (0)=1.$$
(4.37)
This is an anti-Herglotz function given, as it is well known (\[Mi, V\]), by
$$\psi (z)=h(z)/h(0),h(z)=\underset{n\mathrm{}}{lim}\frac{1}{\lambda ^{nr\nu }}(\phi _0^n(z)1).$$
(4.38)
The limit converges uniformly on compact subsets of $`\mathrm{\Omega }(1/\lambda ,1/a_0(\lambda ))`$, which is a basin of attraction of 1 for $`\phi _0`$. It is easy to check that $`\psi `$ depends continuously on $`\phi _0`$. On $`[1/\lambda ,1/a_0(\lambda ))`$, $`\psi `$ is strictly decreasing and, because $`0\phi _0(1/\lambda ))<1`$,
$$\psi (1/\lambda )=\frac{1}{\lambda ^{r\nu }}\psi (\phi _0(1/\lambda ))\frac{1}{\lambda ^{r\nu }}.$$
(4.39)
$`\psi `$ satisfies the inequalities (2.19) and (2.20), with $`u_{}=\lambda `$ and $`u_+=a_0(\lambda )`$. We also define
$$v(z)=(\psi (z))^{1/r}z𝐂_+𝐂_{}(1,1/\lambda ).$$
(4.40)
$`v`$ is a Pick function which extends to a strictly increasing continuous function on $`[1,1/\lambda ]`$. It satisfies:
$$v(1)=0,v(0)=1,v(1/\lambda )\lambda ^\nu .$$
(4.41)
We now show that there is a unique $`z_1(0,1)`$ such that
$$\left(z_1\lambda ^{1\nu }v\right)^p(\lambda )=\lambda ^{1\nu }.$$
(4.42)
As a consequence of the inequality in (4.41), the function $`(s\lambda ^{1\nu }v)^k`$ is defined on $`[1,1/\lambda ]`$ for every $`s[0,1]`$ and every integer $`k0`$. The functions $`sx_k(s)=(s\lambda ^{1\nu }v)^k(\lambda )`$ are thus defined, continuous, and strictly increasing in $`s`$ for $`k>0`$ and $`s[0,1]`$. For $`s_{}=\lambda ^\nu /v(\lambda )`$, $`x_k(s_{})=\lambda `$ for all $`k`$. For $`s>s_{}`$, $`x_1(s)>x_1(s_{})=\lambda =x_0(s)`$, hence $`x_0(s)<x_1(s)<\mathrm{}x_{p+1}(s)`$. Since $`x_p(1)>x_1(1)=\lambda ^{1\nu }v(\lambda )>\lambda ^{1\nu }`$, there exists a unique $`z_1(s_{},1)`$ such that $`x_p(z_1)=\lambda ^{1\nu }`$. In the case $`p=1`$, (4.42) reduces to $`z_1=1/v(\lambda )`$.
The derivative $`x_p^{}(z_1)`$ is strictly positive. Therefore, if $`v`$ is allowed to change slightly, $`z_1`$ can be computed by a Cauchy integral along a small circle which remains fixed. Thus $`z_1`$ depends continuously (in fact analytically) on $`v`$, hence on $`\phi _0`$.
We denote $`\zeta _j=(z_1\lambda ^{1\nu }v)^j(\lambda )`$, $`(0jp+1)`$. By the preceding argument,
$$\lambda =\zeta _0<\zeta _1<\mathrm{}<\zeta _p=\lambda ^{1\nu }<\zeta _{p+1}=z_1\lambda ^{1\nu }v(\lambda ^{1\nu }),$$
(4.43)
Since $`v(\lambda ^{1\nu })\lambda ^\nu `$,
$$z_1>\lambda ^\nu ,z_1\lambda ^{1\nu }>\lambda .$$
(4.44)
Note also that, since $`v(\lambda )>v(0)=1`$,
$$\zeta _1>z_1\lambda ^{1\nu }.$$
(4.45)
The last inequality in (4.43), the upper bound on $`\psi (\lambda ^{1\nu })`$ from (2.19), and $`\nu 1`$ give
$$z_1\left(\frac{1\lambda ^{2\nu }}{1+\lambda ^{1\nu }}\right)^{\frac{1}{r}}>\frac{1\lambda }{2}$$
(4.46)
We can now define a new function $`\phi `$ by:
$$\phi (z)=\lambda ^{\nu 1}\left(z_1\lambda ^{1\nu }v\right)^p(\lambda z),z\mathrm{\Omega }(1/\lambda ,1/\lambda ^2).$$
(4.47)
In this domain, $`\phi `$ is a Pick function, which extends continuously to the ends of its real interval of definition, and
$$\phi (1/\lambda )=0\mathrm{if}p=1,$$
(4.48)
$$\phi (1/\lambda )=\lambda ^{\nu 1}(z_1\lambda ^{1\nu }v)^{p1}(0)z_1>\lambda ^\nu \mathrm{if}p2,$$
(4.49)
$$\phi (1)=1,\phi (1/\lambda ^2)=\lambda ^{\nu 1}(z_1\lambda ^{1\nu }v)^p(1/\lambda )z_1/\lambda ^\nu .$$
(4.50)
The domain $`\mathrm{\Omega }(1/\lambda ,1/\lambda ^2)`$ is thus a basin of attraction of the fixed point 1 of $`\phi `$. This domain contains $`\mathrm{\Omega }(1/\lambda ,1/a_0(\lambda ))`$ since $`a_0(\lambda )\lambda ^2`$ (see (4.36)).
We now use Schwarz’s lemma, as mentioned in Section 2, to obtain an upper bound for $`\phi ^{}(1)`$. If $`p2`$,
$$\phi ^{}(1)\frac{A^{}B^{}(A+B)}{AB(A^{}+B^{})}\mathrm{with}$$
$$A=1+\frac{1}{\lambda },B=\frac{1}{\lambda ^2}1,A^{}=1z_1,B^{}=\frac{z_1}{\lambda ^\nu }1.$$
This gives
$$\phi ^{}(1)\frac{\lambda (1z_1)(z_1\lambda ^\nu )}{z_1(1\lambda ^\nu )(1\lambda ^2)}.$$
(4.51)
When $`z_1(\lambda ^\nu ,1)`$ this expression is maximum at $`z_1=\lambda ^{\nu /2}`$, so that
$$\phi ^{}(1)\frac{\lambda }{Z_1(\lambda )}=\frac{\lambda }{(1+\lambda )(1+\sqrt{\lambda })^2}<\frac{1}{8}\mathrm{if}p2.$$
(4.52)
Therefore, if $`p2`$ and we choose $`bb_0(r\nu )=(1/8)^{1/r\nu }`$, then $`\phi ^{}(1)<b^{r\nu }`$. For a slightly better choice of $`b`$, we note that $`\lambda \lambda /Z_1(\lambda )`$ is increasing on $`(0,1)`$, so that $`\lambda b\phi ^{}(1)b/Z_1(b)`$. This will be less than $`b^{r\nu }`$ if $`bb_1(r\nu )`$, where $`sb_1(s)`$ is the solution of $`b_1^s=b_1/Z_1(b_1)`$, i.e. the inverse function (defined on $`(1,\mathrm{})`$) of
$$b1+\frac{\mathrm{log}Z_1(b)}{\mathrm{log}(1/b)}=1+\frac{\mathrm{log}((1+b)(1+\sqrt{b})^2)}{\mathrm{log}(1/b)}.$$
(4.53)
This last function is strictly increasing on $`(0,1)`$, and tends to 1 as $`b`$ tends to 0 and to $`+\mathrm{}`$ as $`b`$ tends to 1. Obviously $`b_1(s)b_0(s)`$. A useful inequality (proved in Appendix 1) is:
$$\frac{\mathrm{log}((1+b)(1+\sqrt{b})^2)}{\mathrm{log}(1/b)}>\frac{2b}{1b}b(0,1)$$
(4.54)
i.e.
$$b=b_1(s),s>1s>\frac{1+b}{1b}.$$
(4.55)
If $`p=1`$,
$$\phi ^{}(1)\frac{\lambda (1\lambda ^\nu )}{(1\lambda ^2)}\frac{\lambda }{(1+\lambda )}\frac{1}{2},(p=1).$$
(4.56)
Thus, if $`b`$ is chosen at least equal to $`b_2(r\nu )=(1/2)^{1/r\nu }`$ or to $`b_3(r\nu )`$, where $`b_3`$ is the inverse function of $`b1+\mathrm{log}(1+b)/\mathrm{log}(1/b)`$, it follows from $`\lambda b`$ that $`\phi ^{}(1)b^{r\nu }`$.
We now define the action of the operator $`B(b,r,p,\nu )`$ on $`\mathrm{\Phi }_0`$ by
$$B(b,r,p,\nu )\mathrm{\Phi }_0=\mathrm{\Phi }=h_{b,\lambda }^1\phi h_{b,\lambda }.$$
(4.57)
This definition implies that if $`\mathrm{\Phi }_0`$ is a fixed point of $`B(b,r,p,\nu )`$, i.e. $`\mathrm{\Phi }=\mathrm{\Phi }_0`$, then the functions $`\phi _0`$ and $`\phi `$ constructed above coincide, and $`\lambda `$, $`z_1`$, $`\psi `$, and $`u(z)=z_1\lambda ^{1\nu }\psi (z)^{1/r}`$ provide a solution to the problem set in Section 3. Conversely, given a solution to the problem, the function $`\mathrm{\Phi }`$ given by Eq. (4.57) (with $`\lambda =\mathrm{\Phi }^{}(1)^{1/r\nu }`$) is a fixed point of $`B(b,r,p,\nu )`$ for any $`b[\lambda ,1)`$.
The preceding estimates show that if $`p2`$ and $`bb_0(r\nu )`$ or $`bb_1(r\nu )`$, or if $`p=1`$ and $`bb_2(r\nu )`$ or $`bb_3(r\nu )`$, then
$$B(b,r,p,\nu )𝒬_0(b,r\nu )𝒬_0(b,r\nu ).$$
(4.58)
The same estimates show that, for any solution of our problem, the inequalities $`\lambda <b_j(r\nu )`$ must hold ($`j=0,1`$ for $`p2`$, $`j=2,3`$ for $`p=1`$).
The set $`𝒬_0(b,r\nu )`$ is not compact: we have to guard against $`\lambda `$ tending to zero, i.e. to find a reproducing lower bound for $`\lambda `$. This will be feasible only under certain restrictions on $`r`$, $`\nu `$, and $`p`$. We first show that such restrictions are unavoidable. $`\phi ^{}(1)`$ is given by
$`\phi ^{}(1)=\lambda ^\nu {\displaystyle \underset{j=0}{\overset{p1}{}}}z_1\lambda ^{1\nu }v^{}(\zeta _j)`$ $`=`$ $`\lambda ^\nu {\displaystyle \underset{j=0}{\overset{p1}{}}}{\displaystyle \frac{\zeta _{j+1}v^{}(\zeta _j)}{v(\zeta _j)}}`$
$`={\displaystyle \underset{j=0}{\overset{p1}{}}}{\displaystyle \frac{\zeta _jv^{}(\zeta _j)}{v(\zeta _j)}}`$ $`=`$ $`{\displaystyle \underset{j=0}{\overset{p1}{}}}{\displaystyle \frac{\zeta _j\psi ^{}(\zeta _j)}{r\psi (\zeta _j)}}.`$ (4.59)
Here we have used $`\zeta _p/\zeta _0=\lambda ^\nu `$. The upper bound in (2.20) (with $`u_{}=\lambda `$) give
$`\phi ^{}(1){\displaystyle \underset{j=0}{\overset{p1}{}}}{\displaystyle \frac{\zeta _j(1+\lambda )}{r(1+\zeta _j)(1\lambda \zeta _j)}}`$ $``$ $`{\displaystyle \frac{\lambda }{r(1\lambda ^2)}}\left({\displaystyle \frac{\lambda ^{1\nu }(1+\lambda )}{r(1+\lambda ^{1\nu })(1\lambda ^{2\nu })}}\right)^{p1}`$ (4.60)
$``$ $`\lambda ^{1+(p1)(1\nu )}(r(1\lambda ))^p.`$
If we suppose $`p2`$ and $`\lambda b_1(r\nu )`$, then $`r(1\lambda )1+\lambda >1`$ by the inequality (4.55). Therefore a fixed point can exist only if
$$r\nu 1(p1)(1\nu )>0.$$
(4.61)
Using (4.59) and the lower bound in (2.20), with $`u_+=a_0(\lambda )<b`$ gives
$$\phi ^{}(1)\underset{j=0}{\overset{p1}{}}\frac{\zeta _j(1b)}{r(1+\zeta _j)(1+b\zeta _j)},$$
(4.62)
and using $`1\zeta _jz_1\lambda ^{1\nu }`$ (for $`j>0`$), and $`z_1(1b)/2`$ (see (4.46)), we find
$$\phi ^{}(1)c^p\lambda ^{1+(p1)(1\nu )},c=\frac{(1b)^2}{4r(1+b)}.$$
(4.63)
Assume now that $`\lambda \lambda _0>0`$. Then a sufficient condition for $`\phi ^{}(1)\lambda _0^{r\nu }`$ to hold is that
$$\lambda _0^{r\nu 1(p1)(1\nu )}c^p.$$
(4.64)
If the condition (4.61) holds, we can take
$$\lambda _0=\lambda _0(p,r,\nu )=\left(\frac{(1b)^2}{4r(1+b)}\right)^{\frac{p}{r\nu 1(p1)(1\nu )}}(0,1).$$
(4.65)
Assume that the inequality (4.61) holds. Let, for definiteness, $`b(r\nu )=b_1(r\nu )`$ if $`p2`$, and $`b(r\nu )=b_3(r\nu )`$ if $`p=1`$. We observe that
$$𝒬_1(p,r,\nu )=𝒬_0(b(r\nu ),r\nu )\{\mathrm{\Phi }:\mathrm{\Phi }^{}(1)\lambda _0(p,r,\nu )^{r\nu }\}$$
(4.66)
is not empty. Indeed the function $`\chi _{b,s}`$ (see (2.16)) with $`b=b(r\nu )`$ and $`s=r\nu `$ belongs to $`𝒬_0(b(r\nu ),r\nu )`$ and $`\chi _{b,s}^{}=b^{r\nu }`$. Therefore $`\mathrm{\Phi }=B(b(r\nu ),r,p,\nu )\chi _{b,s}`$ also belongs to $`𝒬_0(b(r\nu ),r\nu )`$, and the preceding estimates show that $`b^{r\nu }\mathrm{\Phi }^{}(1)b^{1+(p1)(1\nu )}c^p`$ with $`c`$ as in (4.63), and $`b=b(r\nu )`$. Hence $`b(r\nu )\lambda _0(p,r,\nu )`$, in particular $`\chi _{b,s}𝒬_1(p,r,\nu )`$. (This is not really essential since we could have redefined $`\lambda _0(p,r,\nu )`$ to be less than $`b(r\nu )`$.) The continuous map $`B(b(r\nu ),r,p,\nu )`$ maps the compact convex non-empty set $`𝒬_1(p,r,\nu )`$ into itself. Therefore it has a fixed point there by the Schauder-Tikhonov theorem. As noted before, if $`\mathrm{\Phi }_0=\mathrm{\Phi }`$ is such a fixed point, the functions $`\phi _0`$ and $`\phi `$ constructed as above coincide, and $`\psi `$ and $`u(z)=z_1\lambda ^{1\nu }\psi (z)^{1/r}`$ provide a solution to our problem. Note that here again any solution must satisfy $`\lambda \lambda _0(p,r,\nu )`$, since it must satisfy $`\phi ^{}(1)=\lambda ^{r\nu }c^p\lambda ^{1+(p1)(1\nu )}`$, with $`c`$ as in (4.63). Thus any solution is associated to a fixed point of $`B(b(r\nu ),r,p,\nu )`$ in $`𝒬_1(p,r,\nu )`$.
## 5 Case $`p=1`$ and $`\nu [1,2]`$
This case has been dealt with in \[E3\]. It will be shown in this section that the method of the preceding section also applies to this case with minor modifications. Let $`r>1`$ and $`\nu [1,2]`$ be fixed reals. We define the space $`𝒬(b,r\nu )`$ and the operator $`B(b,r,1,\nu )`$ in the same way as in the preceding section. In particular, starting from $`\mathrm{\Phi }_0𝒬(b,r\nu )`$, the functions $`\phi _0`$, $`\psi `$ and $`v`$ are defined by the same formulae and have the same properties, in particular
$$v(1)=0,v(0)=1,v(1/\lambda )\lambda ^\nu ,$$
(5.67)
but we note that now $`\lambda ^\nu 1/\lambda `$. We define
$$z_1=1/v(\lambda )(0,1).$$
(5.68)
It follows from (5.67) that
$$z_1>\lambda ^\nu $$
(5.69)
and from the upper bound (2.19) on $`\psi (\lambda )`$ that
$$z_1(1\lambda )^{1/r}>1\lambda .$$
(5.70)
The function
$$z\phi (z)=z_1v(\lambda z)$$
(5.71)
is again Herglotzian, holomorphic in $`\mathrm{\Omega }(1/\lambda ,1/\lambda ^2)`$, continuous on $`[1/\lambda ,1/\lambda ^2]`$ with
$$\phi (1/\lambda )=0,\phi (1)=1,\phi (1/\lambda ^2)=z_1v(1/\lambda )z_1/\lambda ^\nu <1/\lambda ^2.$$
(5.72)
Schwarz’s lemma can be again applied as in the preceding section, but now with
$$A=1+\frac{1}{\lambda },B=B^{}=\frac{1}{\lambda ^2}1,A^{}=1.$$
(5.73)
This gives
$$\phi ^{}(1)\lambda ,$$
(5.74)
which is not sufficient for our purposes. We therefore use the bound (2.22) with $`u_{}=\lambda `$ and $`M=1/\lambda ^{r\nu }`$, to get
$$\frac{zv^{}(z)}{v(z)}=\frac{z\psi ^{}(z)}{r\psi (z)}\frac{\nu \mathrm{log}(1/\lambda )}{4(1\lambda z)}z(0,1/\lambda ),$$
(5.75)
hence
$$\phi ^{}(1)=\frac{\lambda v^{}(\lambda )}{v(\lambda )}\frac{\mathrm{log}(1/\lambda )}{2(1\lambda ^2)}.$$
(5.76)
The r.h.s. of this inequality is a decreasing function of $`\lambda `$, tending to $`+\mathrm{}`$ when $`\lambda 0`$, and to $`1/4`$ when $`\lambda 1`$. For any choice of $`b(0,1)`$, if $`\lambda <b^{r\nu }`$, then $`\phi ^{}(1)<b^{r\nu }`$ by (5.74). If $`\lambda b^{r\nu }`$, then, by (5.76), a sufficient condition for $`\phi ^{}(1)<b^{r\nu }`$ is that $`b^{r\nu }>\mu `$, where $`\mu `$ is the unique zero, in $`(0,1)`$, of the increasing function
$$xx\frac{\mathrm{log}(1/x)}{2(1x^2)}.$$
(5.77)
One finds $`\mu <0.479`$, and we choose, from now on, $`b=b_5(r\nu )=(0.479)^{1/r\nu }`$. Note that for any solution, $`\phi ^{}(1)=\lambda ^{r\nu }`$ must satisfy (5.76), and since the rhs of this inequality is decreasing, the function defined in (5.77) must be negative at $`x=\lambda ^{r\nu }`$, i.e. $`\lambda b_5(r\nu )`$.
The lower bound in (2.20), with $`u_+=a_0(\lambda )<b`$, gives
$$\phi ^{}(1)=\frac{\lambda \psi ^{}(\lambda )}{r\psi (\lambda )}\frac{\lambda (1a_0(\lambda ))}{r(1+\lambda )(1+\lambda a_0(\lambda ))}\frac{\lambda (1b)}{r(1+b)(1+b^2)}.$$
(5.78)
If $`\lambda \lambda _0>0`$ then $`\phi ^{}(1)\lambda _0^{r\nu }`$ provided $`\lambda _0\lambda _0(r,\nu )`$ with
$$\lambda _0(r,\nu )=\left(\frac{(1b)}{r(1+b)(1+b^2)}\right)^{1/(r\nu 1)},b=b_5(r\nu ).$$
(5.79)
Therefore the operator $`B(b_5(r\nu ),r,1,\nu )`$ preserves the compact convex set
$$𝒬_0(b_5(r\nu ),r\nu )\{\mathrm{\Phi }:\mathrm{\Phi }^{}(1)\lambda _0(r,\nu )\}.$$
(5.80)
Again any solution must satisfy $`\lambda \lambda _0(r,\nu )`$.
## 6 Case $`p2`$ and $`\nu [1,2]`$
In this section, $`r>1`$ and $`\nu [1,2]`$ are fixed real numbers, and $`p2`$ is a fixed integer. The real number $`b[1/2,1)`$ is also fixed, but its value will be chosen later (as a function of $`r`$). We shall need the function $`a:[0,1][0,1]`$ given by
$$a(t)=\mathrm{min}\{\frac{2t}{1t},\frac{1+t}{2}\}=\{\begin{array}{ccc}\frac{2t}{1t}& \mathrm{if}& 0t\sqrt{5}2,\\ \frac{1+t}{2}& \mathrm{if}& \sqrt{5}2t1.\end{array}$$
(6.81)
$`ta(t)`$ is continuous and strictly increasing in $`[0,1]`$. (Note that $`\sqrt{5}20.236<1/4`$.)
We denote $`\stackrel{~}{𝒬}_0(b,\nu r)`$ the space of all functions $`\mathrm{\Phi }`$ with the following properties:
(Q̃1) $`\mathrm{\Phi }`$ is a Pick function holomorphic in the domain:
$$\mathrm{\Omega }(1/b,1/a(b))=𝐂_+𝐂_{}(\frac{1}{b},\frac{1}{a(b)}),a(b)=\frac{1+b}{2},$$
(6.82)
and maps this domain into itself.
(Q̃2) $`\mathrm{\Phi }(z)0`$ for all $`z(1/b,1/a(b))`$,
(Q̃3) $`\mathrm{\Phi }(1)=1`$, and $`0<\mathrm{\Phi }^{}(1)b^{\nu r}`$.
$`\stackrel{~}{𝒬}_0(b,\nu r)`$ is a convex subset of the real Fréchet space of all self-conjugated functions holomorphic in $`\mathrm{\Omega }(1/b,1/a(b))`$. It is not empty since it contains the function $`\chi _{b,\nu r}`$ (see Section 2).
### 6.1 The operator $`B(b,r,p,\nu )`$
We shall define a continuous operator $`B(b,r,p,\nu )`$ on the space $`\stackrel{~}{𝒬}_0(b,\nu r)`$ by describing its action on an arbitrary element $`\mathrm{\Phi }_0`$.
Given $`\mathrm{\Phi }_0\stackrel{~}{𝒬}_0(b,\nu r)`$, we denote $`\lambda =\mathrm{\Phi }_0^{}(1)^{1/r\nu }`$. Note that $`\lambda b`$. We define a function $`\phi _0`$ by
$$\phi _0=h_{b,\lambda }\mathrm{\Phi }_0h_{b,\lambda }^1.$$
(6.83)
Here $`h_{b,\lambda }`$ is the homographic function defined in Section 4 (see (4.33)). It maps the domain $`\mathrm{\Omega }(1/b,1/a(b))`$ onto $`\mathrm{\Omega }(1/\lambda ,1/a_1(\lambda ))`$, where $`1/a_1(\lambda )=h_{b,\lambda }(1/a(b))`$, i.e.
$$a_1(\lambda )=\frac{1+\lambda }{2}+\frac{b\lambda }{1+b}a(b)a(\lambda ),a_1(\lambda )a_1(0)=\frac{1+3b}{2(1+b)}.$$
(6.84)
The function $`\phi _0`$ possesses the following properties:
(Q̃$`{}_{}{}^{}1`$) $`\phi _0`$ is a Pick function holomorphic in $`\mathrm{\Omega }(1/\lambda ,1/a_1(\lambda ))`$, and maps this domain into itself.
(Q̃$`{}_{}{}^{}2`$) $`\phi _0(z)0`$ for all $`z(1/\lambda ,1/a_1(\lambda ))`$.
(Q̃$`{}_{}{}^{}3`$) $`\phi _0(1)=1`$, and $`\phi _0^{}(1)=\lambda ^{\nu r}`$.
As in previous sections we denote $`\psi `$ the linearizer of $`\phi _0`$, normalized by the condition $`\psi (0)=1`$, i.e.
$$\psi (z)=\frac{1}{\lambda ^{\nu r}}\psi (\phi _0(z))z\mathrm{\Omega }(1/\lambda ,1/a_1(\lambda )),\psi (1)=0,\psi (0)=1.$$
(6.85)
$`\psi `$ is anti-Herglotz, holomorphic in $`\mathrm{\Omega }(1/\lambda ,1/a_1(\lambda ))`$, and satisfies the inequalities (2.19) and (2.20), with $`u_{}=\lambda `$ and $`u_+=a_1(\lambda )`$. Also,
$$\psi (1/\lambda )=\frac{1}{\lambda ^{\nu r}}\psi (\phi _0(1/\lambda ))\frac{1}{\lambda ^{\nu r}}.$$
(6.86)
We again define $`v(z)=(\psi (z))^{1/r}`$ fo all $`z\mathrm{\Omega }(1,1/\lambda )`$. This is a Pick function which extends to a strictly increasing continuous function on $`[1,1/\lambda ]`$. It satisfies:
$$v(1)=0,v(0)=1,v(1/\lambda )\lambda ^\nu .$$
(6.87)
We now show that there is a unique $`z_1(0,1)`$ such that
$$\left(\frac{z_1}{\lambda ^{\nu 1}}v\right)^p(\lambda )=\frac{1}{\lambda ^{\nu 1}}.$$
(6.88)
For real $`s0`$ let $`x_0(s)=\lambda `$, $`x_1(s)=s\lambda ^{1\nu }v(\lambda )`$. The function $`sx_1(s)`$ is strictly increasing on $`𝐑_+`$ and takes the values $`\lambda `$ at $`s_{}=\lambda ^\nu /v(\lambda )`$ and $`1/\lambda `$ at $`s_1=\lambda ^{\nu 2}/v(\lambda )`$. By induction we can construct a strictly decreasing infinite sequence $`s_1>\mathrm{}>s_j>\mathrm{}>s_{}`$ such that, for $`j2`$, $`sx_j(s)=(s\lambda ^{1\nu }v)^j(\lambda )`$ is continuous and strictly increasing on $`[s_{},s_{j1}]`$, $`x_0(s)<\mathrm{}<x_j(s)`$ in $`(s_{},s_{j1}]`$, $`x_j(s_{})=\lambda `$, and $`x_j(s_j)=1/\lambda `$. Indeed it follows that $`x_{j+1}(s)=s\lambda ^{1\nu }v(x_j(s))`$ is defined, continuous, and strictly increasing on $`[s_{},s_j]`$ and $`x_{j+1}(s)>s\lambda ^{1\nu }v(x_{j1}(s))=x_j(s)`$ for all $`s(s_{},s_j]`$. Since $`x_{j+1}(s_j)>x_j(s_j)=1/\lambda `$ and $`x_{j+1}(s_{})=\lambda `$, $`s_{j+1}`$ exists in $`(s_{},s_j)`$. In particular $`x_p(s_{p1})>1/\lambda `$. Therefore there is a unique $`z_1(s_{},s_{p1})`$ such that $`x_p(z_1)=\lambda ^{1\nu }`$. It must satisfy $`z_1<1`$ since $`z_1v(x_{p1}(z_1))=1`$, and $`v(x_{p1}(z_1))>1`$. Note also that for $`s(s_{},s_{p1})`$, there exists a unique $`x_1(s)<x_0(s)=\lambda `$ such that $`s\lambda ^{1\nu }v(x_1(s))=\lambda `$. The function $`zs_{}\lambda ^{1\nu }v(z)`$ maps $`\mathrm{\Omega }(1,1/\lambda )`$ into $`\mathrm{\Omega }(0,\lambda ^{1\nu }/v(\lambda ))`$, so that it has a unique and attractive fixed point at $`\lambda `$ by Schwarz’s lemma. Hence $`s_{}\lambda ^{1\nu }v(x)x`$ for all $`x[1,\lambda ]`$. When $`s>s_{}`$, $`s\lambda ^{1\nu }v(x)>x`$ for all $`x[1,\lambda ]`$. Since this includes $`[x_1(s),x_0(s)]`$, it follows that $`s\lambda ^{1\nu }v(x)>x`$ for all $`x[1,x_p(s)]`$, for all $`s(s_{},z_1]`$. The function $`x_p`$ is analytic, with a strictly positive derivative, on $`(s_{},s_{p1})`$. Therefore $`z_1`$ depends continuously on $`v`$, hence on $`\phi _0`$.
We denote $`\zeta _j=(z_1\lambda ^{1\nu }v)^j(\lambda )`$, $`(0jp+1)`$ :
$$\lambda =\zeta _0<\zeta _1<\mathrm{}\zeta _p=\frac{1}{\lambda ^{\nu 1}}<\zeta _{p+1}=\frac{z_1}{\lambda ^{\nu 1}}v(1/\lambda ^{\nu 1}),$$
(6.89)
Since $`v(1/\lambda ^{\nu 1})\lambda ^\nu `$,
$$z_1>\lambda ^\nu ,\frac{z_1}{\lambda ^{\nu 1}}>\lambda ,$$
(6.90)
and since $`v(\lambda )>v(0)=1`$,
$$\zeta _1>\frac{z_1}{\lambda ^{\nu 1}}.$$
(6.91)
We have seen above that
$$\frac{z_1}{\lambda ^{\nu 1}}v(x)>xx[1,1/\lambda ^{\nu 1}].$$
(6.92)
Applying this to $`x=1`$ gives $`z_1/\lambda ^{\nu 1}1/v(1)`$, and, using (2.19),
$$\frac{z_1}{\lambda ^{\nu 1}}\left(\frac{1\lambda }{2}\right)^{\frac{1}{r}}>\frac{1\lambda }{2}.$$
(6.93)
The function $`\phi `$ is defined by:
$$\phi (z)=\lambda ^{\nu 1}\left(\frac{z_1}{\lambda ^{\nu 1}}v\right)^p(\lambda z),z𝐂_+𝐂_{}(1/\lambda ,\zeta _1/\lambda ).$$
(6.94)
In this domain, $`\phi `$ is a Pick function, which extends continuously to the ends of its real interval of definition, and
$`\phi (1/\lambda )`$ $`=`$ $`\lambda ^{\nu 1}\left({\displaystyle \frac{z_1}{\lambda ^{\nu 1}}}v\right)^{p1}(0)z_1\lambda ^\nu \lambda ^2,`$
$`\phi (1)`$ $`=`$ $`1,\phi (\zeta _1/\lambda )=z_1v(1/\lambda ^{\nu 1}){\displaystyle \frac{z_1}{\lambda ^\nu }}<\zeta _1/\lambda .`$ (6.95)
(Note that the first inequality in (6.95) has used $`p2`$.) The domain $`\mathrm{\Omega }(1/\lambda ,\zeta _1/\lambda )`$ is a basin of attraction of the fixed point 1 of $`\phi `$, hence $`\phi ^{}(1)<1`$ by Schwarz’s lemma. For a better upper bound on this derivative, we shall need a better lower bound for $`\zeta _1/\lambda `$. This is provided by
###### Lemma 2
The inequality
$$\frac{z_1}{\lambda ^{\nu 1}}v(z)\frac{z(12\lambda )+\lambda }{1\lambda z}$$
(6.96)
holds for all $`z[0,1]`$.
Proof. This is simply the result of applying Lemma 1 of Section 2, with $`f=z_1\lambda ^{1\nu }v`$, and $`a=0`$, $`b=1`$, $`B=1/\lambda `$. This function satisfies $`f(0)=z_1\lambda ^{1\nu }\lambda `$ by (6.90), and $`f(1)1`$ by (6.92).
For $`z=\zeta _0=\lambda `$, this gives $`\zeta _12\lambda /(1+\lambda )`$. Since we also have the lower bounds (6.91) and (6.93),
$$\frac{\zeta _1}{\lambda }\mathrm{max}\{\frac{2}{1+\lambda },\frac{1\lambda }{2\lambda }\}=\frac{1}{a(\lambda )}.$$
(6.97)
This is the reason for our original definition of the function $`a`$ in (6.81). We conclude that the domain $`\mathrm{\Omega }(1/\lambda ,\zeta _1/\lambda )`$ where $`\phi `$ is holomorphic, and which it maps into itself, certainly contains the domain of analyticity $`\mathrm{\Omega }(1/\lambda ,1/a_1(\lambda ))`$ of $`\phi _0`$ in view of (6.84). We now use Schwarz’s lemma, as mentioned in Section 2, to obtain an upper bound for $`\phi ^{}(1)`$ :
$`\phi ^{}(1){\displaystyle \frac{A^{}B^{}(A+B)}{AB(A^{}+B^{})}}`$ $`\mathrm{with}`$
$`A=1+{\displaystyle \frac{1}{\lambda }}`$ , $`B=B^{}={\displaystyle \frac{1}{a(\lambda )}}1,A^{}=1\lambda ^2.`$ (6.98)
This gives
$`\phi ^{}(1)`$ $``$ $`{\displaystyle \frac{(1\lambda ^2)(a(\lambda )+\lambda )}{(1+\lambda )(1a(\lambda )\lambda ^2)}}`$ (6.99)
$``$ $`Z(\lambda )={\displaystyle \frac{1+3\lambda }{2+2\lambda +\lambda ^2}}`$
$``$ $`Z_{\mathrm{max}}=9/(4+2\sqrt{13})<0.803.`$
Therefore choosing $`bb_6(r\nu )=m_0^{1/r\nu }`$, $`m_0=0.803`$ ensures that $`\phi ^{}(1)<b^{r\nu }`$.
We define the operator $`B(b,r,p,\nu )`$ by
$$\mathrm{\Phi }=B(b,r,p,\nu )\mathrm{\Phi }_0=h_{b,\lambda }^1\phi h_{b,\lambda }.$$
(6.100)
It then follows from the preceding estimates that, if $`bb_6(r\nu )`$,
$$B(b,r,p,\nu )\stackrel{~}{𝒬}_0(b,\nu r)\stackrel{~}{𝒬}_0(b,\nu r).$$
(6.101)
In the remainder of this section, it will always be understood that $`b=b_6(r\nu )`$.
#### 6.1.1 Lower bound for $`\phi ^{}(1)`$.
We use
$$\phi ^{}(1)=\underset{j=0}{\overset{p1}{}}\frac{\zeta _j\psi ^{}(\zeta _j)}{r\psi (\zeta _j)}.$$
(6.102)
The lower bound in (2.20) gives, for $`\zeta [0,1/\lambda )`$,
$$\frac{\zeta \psi ^{}(\zeta )}{\psi (\zeta )}\frac{\zeta (1c)}{(1+\zeta )(1+c\zeta )},$$
(6.103)
Here $`c=a_1(\lambda )`$, where $`a_1(\lambda )`$ is given by (6.84) and satisfies
$$a(\lambda )a(b)a_1(\lambda )<a_1(0)=\frac{1+3b}{2(1+b)}.$$
(6.104)
However it will be convenient to suppose only, at first, that (6.103) holds for a certain $`c`$ satisfying $`a(\lambda )c<1`$. For $`j>0`$, $`\zeta _j\zeta _1\lambda /a(\lambda )`$ hence $`\zeta _j[\lambda /c,1/\lambda ]`$. When $`\zeta `$ varies in this interval, the second expression in (6.103) is minimum at $`\zeta =1/\lambda `$. Therefore
$$\phi ^{}(1)\lambda ^p\left(\frac{1c}{r(1+\lambda )(1+c\lambda )}\right)\left(\frac{1c}{r(1+\lambda )(\lambda +c)}\right)^{p1}.$$
(6.105)
It is easy to verify that the rhs of this inequality is decreasing in $`c`$ and increasing in $`\lambda `$ provided $`c\lambda ^2`$ (note that $`a(\lambda )>\lambda `$). Setting now $`c=a_1(\lambda )`$, and using the inequalities (6.104) and $`\lambda b`$, this gives
$$\phi ^{}(1)\lambda ^p\left(\frac{1b}{16r}\right)^p.$$
(6.106)
Supposing $`\lambda \lambda _0>0`$, the last inequality will imply $`\phi ^{}(1)\lambda _0^{r\nu }`$ if $`\lambda _0`$ satisfies
$$\lambda _0^{r\nu p}\left(\frac{1b}{16r}\right)^p,$$
(6.107)
and this is possible only if $`r\nu p>0`$. In this case we can choose
$$\lambda _0=\lambda _0(r,\nu )=\left(\frac{1b}{16r}\right)^{\frac{p}{r\nu p}},$$
(6.108)
and obtain the existence of a fixed point in the same way as in the preceding sections. Recall that in these formulae, $`b`$ stands for $`b_6(r\nu )=m_0^{1/r\nu }`$. It is easy to verify that $`\lambda _0(r,\nu )1`$ when $`r\mathrm{}`$.
The condition $`r\nu p>0`$ is just a limitation of the present method. The inadequacy of the estimate (6.106) is due to the fact that $`a_1(\lambda )`$ does not tend to 0 as $`\lambda `$ tends to 0. By contrast, in the case of fixed points, the lower bound on $`\lambda `$ can be improved. Indeed, since $`\psi `$ and $`\phi `$ are holomorphic in $`\mathrm{\Omega }(1/\lambda ,1/a(\lambda ))`$, the bound (6.103) and consequently (6.105) hold with $`c`$ replaced by $`a(\lambda )`$ (instead of $`a_1(\lambda )`$). Assume $`\lambda 1/7`$. We can then set $`c=2\lambda /(1\lambda )`$ in (6.105) and obtain
$$\phi ^{}(1)\lambda \left(\frac{13\lambda }{r(1+\lambda )(1\lambda +2\lambda ^2)}\right)\left(\frac{13\lambda }{r(1+\lambda )(3\lambda )}\right)^{p1}>\lambda (6r)^p.$$
(6.109)
Therefore the lower bound
$$\lambda =\phi ^{}(1)^{1/r\nu }\mathrm{min}\{1/7,(6r)^{p/(r\nu 1)}\}$$
(6.110)
holds for all fixed points.
This fact suggests the use of another operator instead of $`B(b,r,p,\nu )`$, and this will be done in the next subsection.
### 6.2 The operator $`N(b,r,p,\nu ,\lambda _1)`$
In this subsection, we define a new operator $`N(b,r,p,\nu ,\lambda _1)`$ on the space $`\stackrel{~}{𝒬}_0(b,\nu r)`$. This construction closely follows an idea of Mestel and Osbaldestin \[MO\]. It consists in replacing the operator $`B(b,r,p,\nu )`$ (which is analytic on $`\stackrel{~}{𝒬}_0(b,\nu r)`$) by a “truncated version” $`N(b,r,p,\nu ,\lambda _1)`$ which is only continuous, but maps $`\stackrel{~}{𝒬}_0(b,\nu r)`$ into a compact subset. This operator depends on an additional real parameter $`\lambda _1(0,1/2)`$. It will be shown later that for small values of this parameter, any fixed point of $`N(b,r,p,\nu ,\lambda _1)`$ is a fixed point of $`B(b,r,p,\nu )`$.
The notations are the same as in the preceding subsection unless explicitly mentioned. In particular $`\nu [1,2]`$ and $`r>1`$ are fixed and $`b`$ will stand for $`b_6(r\nu )=m_0^{1/r\nu }`$, $`m_0=0.803`$ . We denote $`\tau _1=\lambda _1^r`$.
We define $`N(b,r,p,\nu ,\lambda _1)`$ by its action on an arbitrary element $`\mathrm{\Phi }_0`$ of $`\stackrel{~}{𝒬}_0(b,\nu r)`$. Let $`\sigma ^\nu =\mathrm{\Phi }_0^{}(1)`$. Recall that, by the definition of $`\stackrel{~}{𝒬}_0(b,\nu r)`$, $`\sigma ^\nu b^{r\nu }`$.
If $`\sigma ^\nu \tau _1^\nu `$, we define
$$N(b,r,p,\nu ,\lambda _1)\mathrm{\Phi }_0=B(b,r,p,\nu )\mathrm{\Phi }_0,(\sigma ^\nu \tau _1^\nu ).$$
(6.111)
If $`\sigma ^\nu <\tau _1^\nu `$, we define $`\lambda =\lambda _1`$ (so that $`\lambda b`$), and define $`\phi _0`$, as before, by
$$\phi _0=h_{b,\lambda }\mathrm{\Phi }_0h_{b,\lambda }^1.$$
(6.112)
The function $`\phi _0`$ is holomorphic and Herglotzian in the domain $`\mathrm{\Omega }(1/\lambda ,1/a_1(\lambda ))`$, which it maps into itself. Here $`a_1(\lambda )=a_1(\lambda _1)`$ is given by (6.84). $`\phi _0`$ possesses the same properties as in Subsection 6.1, except for
$$\phi _0^{}(1)=\sigma ^\nu .$$
(6.113)
The linearizer $`\psi _1`$ is the unique function holomorphic in $`\mathrm{\Omega }(1/\lambda ,1/a_1(\lambda ))`$ such that
$$\psi _1(z)=\frac{1}{\sigma ^\nu }\psi _1(\phi _0(z))z\mathrm{\Omega }(1/\lambda ,1/a_1(\lambda )),\psi _1(0)=1,\psi _1(1)=0.$$
(6.114)
It is anti-Herglotzian and satisfies
$$\psi _1(1/\lambda )=\frac{1}{\sigma ^\nu }\psi _1(\phi _0(1/\lambda ))\frac{1}{\sigma ^\nu }.$$
(6.115)
In the preceding subsection, much depended on the bound $`\psi (1/\lambda )\lambda ^{r\nu }`$. To restore an analogous situation we define a new function $`\psi `$ as
$$\psi =\theta _{\sigma ^\nu ,\tau _1^\nu }\psi _1,$$
(6.116)
where $`\theta _{\sigma ^\nu ,\tau _1^\nu }`$ denotes the homographic function which fixes 0 and 1, and sends $`\sigma ^\nu `$ to $`\tau _1^\nu `$:
$$\theta _{\sigma ^\nu ,\tau _1^\nu }(z)=\frac{z(1\sigma ^\nu )}{z(\tau _1^\nu \sigma ^\nu )+1\tau _1^\nu }.$$
(6.117)
This function is Herglotzian and has a pole at a negative value temporarily denoted $`k`$. As a consequence $`\psi `$ is holomorphic and anti-Herglotzian in $`\mathrm{\Omega }(1/\lambda ,1/a_2)`$, where $`1/a_2=\psi _1^1(k)`$ if $`k\psi _1((1,1/a_1(\lambda )))`$, and $`1/a_2=1/a_1(\lambda )`$ otherwise. For $`z>1`$, $`\psi _1(z)<0`$ and (using the inequalities (2.19)),
$$\psi _1(z)\psi _2(z)=\frac{1z}{1a_1(\lambda )z}$$
(6.118)
If $`y=\psi _1^1(k)<1/a_1`$, we have, since $`\psi _2`$ is decreasing,
$$k=\psi _1(y)\psi _2(y),\psi _2^1(k)y.$$
(6.119)
Thus $`\psi `$ is holomorphic in $`\mathrm{\Omega }(1/\lambda ,1/\mathrm{})`$, where $`1/\mathrm{}=\psi _2^1(k)=\psi _2(k)`$. This gives:
$$\mathrm{}=\frac{\tau _1^\nu \sigma ^\nu +(1\tau _1^\nu )a_1(\lambda )}{1\sigma ^\nu },$$
(6.120)
$$a(\lambda )a_1(\lambda )\mathrm{}<a_3(\lambda )=\tau _1^\nu +(1\tau _1^\nu )a_1(\lambda ).$$
(6.121)
The function $`\psi `$ has been defined so as to satisfy $`\psi (1/\lambda )\lambda ^{r\nu }`$. We now proceed to define $`v`$, $`z_1`$, $`\phi `$ etc. exactly as in the preceding subsection and obtain the same inequalities with the single exception that, in the lower bound (6.105), $`c`$ must be replaced by $`a_3(\lambda )`$. Since $`\lambda =\lambda _1`$, we find
$$\phi ^{}(1)l(\lambda _1)=\lambda _1^p\left(\frac{1a_3(\lambda _1)}{r(1+\lambda _1)(1+a_3(\lambda _1)\lambda _1)}\right)\left(\frac{1a_3(\lambda _1)}{r(1+\lambda _1)(\lambda _1+a_3(\lambda _1))}\right)^{p1}.$$
(6.122)
Recall that $`\phi `$ is holomorphic in $`\mathrm{\Omega }(1/\lambda ,1/a(\lambda ))`$ and maps this domain into itself, with $`a(\lambda )`$ given by (6.81). The bound (6.122) also holds in the cases when $`\lambda >\lambda _1`$ since then $`a_1(\lambda )<a_3(\lambda _1)`$.
Finally we define
$$N(b,r,p,\nu ,\lambda _1)\mathrm{\Phi }_0=h_{b,\lambda }^1\phi h_{b,\lambda }.$$
(6.123)
The operator $`N(b,r,p,\nu ,\lambda _1)`$ maps the domain $`\stackrel{~}{𝒬}_0(b,\nu r)`$ into $`\stackrel{~}{𝒬}_0(b,\nu r)\{\mathrm{\Phi }:\mathrm{\Phi }^{}(1)l(\lambda _1)\}`$, which is compact and convex, hence it has fixed points there.
Our task is now to prove that if $`\lambda _1`$ has been chosen sufficiently small, any fixed point of $`N(b,r,p,\nu ,\lambda _1)`$ is actually a fixed point of $`B(b,r,p,\nu )`$. We assume, from now on, that $`\lambda _11/8`$. Let $`\mathrm{\Phi }_0`$ be a fixed point of $`N(b,r,p,\nu ,\lambda _1)`$. If $`\sigma ^\nu =\mathrm{\Phi }_0^{}(1)\tau _1^\nu `$, there is nothing to prove. Otherwise, we have $`\lambda =\lambda _1`$ and $`\phi _0=\phi `$, so that $`\phi _0`$ and $`\psi _1`$ are now holomorphic in $`\mathrm{\Omega }(1/\lambda _1,1/a(\lambda _1))`$. Thus $`\psi `$ is now holomorphic in $`\mathrm{\Omega }(1/\lambda _1,1/a_4(\lambda _1))`$, with
$$a(\lambda _1)<a_4(\lambda _1)=\frac{\tau _1^\nu \sigma ^\nu +(1\tau _1^\nu )a(\lambda _1)}{1\sigma ^\nu }<\tau _1^\nu +(1\tau _1^\nu )a(\lambda _1).$$
(6.124)
Recalling that $`\lambda _11/8`$, we find
$$a_4(\lambda _1)\lambda _1^{r\nu }+2\lambda _1\frac{1\lambda _1^{r\nu }}{1\lambda _1}\frac{3\lambda _1}{1\lambda _1}$$
(6.125)
Inserting this in the lower bound obtained by setting $`\lambda =\lambda _1`$ and $`c=a_4(\lambda _1)`$ in (6.105) gives
$$\phi ^{}(1)\lambda _1\left(\frac{14\lambda _1}{r(1+\lambda _1)(1\lambda _1+3\lambda _1^2)}\right)\left(\frac{14\lambda _1}{r(1+\lambda _1)(4\lambda _1)}\right)^{p1},$$
(6.126)
and, using $`\lambda _11/8`$,
$$\phi ^{}(1)\lambda _1(9r)^p,$$
(6.127)
and since $`\lambda _1(\phi ^{}(1))^{1/r\nu }`$,
$$\phi ^{}(1)(9r)^{pr\nu /(r\nu 1)}.$$
(6.128)
If we assume that $`\lambda _1`$ has been chosen so that
$$\lambda _1<(9r)^{p/(r\nu 1)},$$
(6.129)
the inequality (6.128) contradicts our hypothesis that $`\mathrm{\Phi }_0^{}(1)<\lambda _1^{r\nu }`$. Therefore $`\mathrm{\Phi }_0`$ is a fixed point of $`B(b,r,p,\nu )`$.
## 7 Properties of solutions
This section is devoted to some properties of the solutions, i.e. of functions $`\psi `$ and $`u`$, and numbers $`\nu (0,1]`$, $`p2`$, $`r>1`$, ($`r\nu >1+(p1)(1\nu )`$ if $`\nu <1`$), $`\lambda `$, $`z_1`$, satisfying the requirements of Section 3. These properties are extensions of those established for $`p=1`$ in \[EE, EL, E2\]. We do not consider the case $`p=1`$.
We denote $`\phi =\phi _0`$, $`v`$, $`\zeta _0,\mathrm{},\zeta _{p+1}`$, the objects constructed from $`\psi `$ as in the definition of $`B(b,r,p,\nu )`$. We also denote $`\tau =\lambda ^r`$, and
$`u(z)=\stackrel{ˇ}{u}(z)`$ $`=`$ $`{\displaystyle \frac{z_1}{\lambda ^{\nu 1}}}v(z)`$
$`=`$ $`{\displaystyle \frac{z_1}{\lambda ^{\nu 1}}}\psi (z)^{1/r}=U(z)^{1/r},z\mathrm{\Omega }(1/\lambda ,1),`$
$$U(z)=\left(\frac{z_1}{\lambda ^{\nu 1}}\right)^r\psi (z),z\mathrm{\Omega }(1/\lambda ,\zeta _1/\lambda ).$$
(7.130)
Recall that it has been shown in Section 4 that
$$\frac{1}{\tau ^\nu }\frac{1}{\lambda }(1+\lambda )(1+\sqrt{\lambda })^2>8,\lambda b_1(r\nu ),\mathrm{if}0<\nu 1,p2,$$
(7.131)
and
$$r\nu \frac{1+\lambda }{1\lambda },\lambda ^{r\nu 1(p1)(1\nu )}(1+\lambda )^p\mathrm{if}0<\nu 1,p2.$$
(7.132)
Moreover (4.65) and $`r\nu (1+b)/(1b)`$ give
$$\lambda (4r^3\nu ^2)^{p/(r\nu 1(p1)(1\nu ))}\mathrm{if}0<\nu 1,p2.$$
(7.133)
For $`1<\nu 2`$, it was shown in Section 6 that
$$\lambda b_6(r\nu )=m_0^{1/r\nu },m_0=0.803,\frac{\zeta _1}{\lambda }\frac{1}{a(\lambda )},(1<\nu 2,p2),$$
(7.134)
where $`a(\lambda )`$ is defined in (6.81), and that
$$\lambda \mathrm{min}\{1/7,(6r)^{p/(r\nu 1)}\},(1<\nu 2,p2),$$
(7.135)
$$\lambda \left(\frac{1m_0^{1/r\nu }}{16r}\right)^{p/(r\nu p)},(1<\nu 2,2p<r\nu ).$$
(7.136)
The function $`u`$ has an angular derivative at infinity equal to zero (i.e. $`u(z)/z`$ tends to 0 as $`z\mathrm{}`$ in non real directions) because $`u(z)=U(z)^{1/r}`$, $`U`$ is anti-Herglotzian, and $`r>1`$. Similarly $`v`$ and $`\phi `$ have zero angular derivative at infinity.
### 7.1 Analyticity
The function $`\phi `$ is holomorphic in $`\mathrm{\Omega }(1/\lambda ,\xi _{\mathrm{max}})`$, where $`\xi _{\mathrm{max}}=\lambda ^2`$ if $`\stackrel{ˇ}{u}(1/\lambda )1/\lambda `$ (as is the case for $`\nu 1`$, since $`\lambda \stackrel{ˇ}{u}(1/\lambda )z_1\lambda ^{22\nu }<1`$). In this case,
$$\phi (\xi _{\mathrm{max}})=\phi (\lambda ^2)=z_1v(\stackrel{ˇ}{u}^{p1}(\lambda ^1))<\frac{z_1}{\lambda ^\nu }<\lambda ^2.$$
(7.137)
If $`\stackrel{ˇ}{u}(1/\lambda )>1/\lambda `$, we denote $`\xi _p=1/\lambda ^2`$ and $`\lambda \xi _{p1}=\stackrel{ˇ}{u}^1(1/\lambda )`$. We construct by a descending induction the strictly increasing sequence $`\xi _1,\mathrm{},\xi _p`$ satisfying $`\stackrel{ˇ}{u}^j(\lambda \xi _{pj})=\lambda \xi _p=1/\lambda `$. Supposing $`\xi _{pj}<\mathrm{}\xi _p`$ already constructed for a certain $`j<p1`$, we have $`\stackrel{ˇ}{u}^{j+1}(\lambda \xi _{pj})=\stackrel{ˇ}{u}(1/\lambda )>1/\lambda `$, while $`\stackrel{ˇ}{u}^{j+1}(\lambda )=\zeta _{j+1}<1/\lambda `$. Hence $`\lambda \xi _{pj1}=\stackrel{ˇ}{u}^{(j+1)}(1/\lambda )`$ exists in $`(1,\xi _{pj})`$. We set $`\xi _{\mathrm{max}}=\xi _1`$ so that $`\stackrel{ˇ}{u}^{p1}(\lambda \xi _{\mathrm{max}})=1/\lambda `$. Recalling that $`\stackrel{ˇ}{u}^{p1}(\zeta _1)=\lambda ^{1\nu }`$, we find:
$$\frac{\zeta _1}{\lambda }\xi _{\mathrm{max}}=\xi _1<\xi _2<\mathrm{}<\xi _p=\lambda ^2.$$
(7.138)
The first inequality here is replaced by the equality $`\xi _{\mathrm{max}}=\zeta _1/\lambda `$ when $`\nu =2`$ (and, of course, $`p>1`$). More generally $`\stackrel{ˇ}{u}^{pj}(\zeta _j)=\lambda ^{1\nu }`$ implies $`\zeta _j\lambda \xi _j`$ for all $`j[1,p1]`$, equality holding when $`\nu =2`$. Note (see (6.89) and (6.91)) that $`z_1/\lambda ^{\nu 1}<\zeta _1<1/\lambda ^{\nu 1}`$, and
$$\phi (\xi _{\mathrm{max}})=z_1v(\lambda ^1)\frac{z_1}{\lambda ^\nu }<\zeta _1/\lambda .$$
(7.139)
In both cases the whole domain $`\mathrm{\Omega }(1/\lambda ,\xi _{\mathrm{max}})`$ is a basin of attraction of the fixed point 1 of $`\phi `$, hence the domain of $`\psi `$ is also $`\mathrm{\Omega }(1/\lambda ,\xi _{\mathrm{max}})`$, and
$`\psi (z)`$ $`=`$ $`{\displaystyle \frac{1}{\lambda ^{\nu r}}}\psi (\phi (z)),`$
$`\phi (z)`$ $`=`$ $`\lambda ^{\nu 1}\stackrel{ˇ}{u}^p(\lambda z),`$ (7.140)
hold for all $`z\mathrm{\Omega }(1/\lambda ,\xi _{\mathrm{max}})`$. Also
$$u(z)=\frac{1}{\lambda ^\nu }u(\lambda ^{\nu 1}\stackrel{ˇ}{u}^p(\lambda z)),z\mathrm{\Omega }(1/\lambda ,1).$$
(7.141)
### 7.2 Univalence for $`p2`$
We prove in this subsection that $`\psi `$ and $`\phi `$ are univalent in $`\mathrm{\Omega }(1/\lambda ,\xi _{\mathrm{max}})`$. We temporarily denote
$`\varphi _j(z)`$ $`=`$ $`\stackrel{ˇ}{u}^j(\lambda z),\mathrm{for}0jp1,`$
$`\varphi _p(z)`$ $`=`$ $`\phi (z).`$ (7.142)
Let $`c`$ be fixed with $`1<c<\mathrm{min}\{1/\lambda ,\zeta _1/\lambda \}`$. We first verify that each $`\varphi _j`$, $`0jp`$, maps the interval $`(1/\lambda ,c)`$ into an open interval $`X_j`$ with closure contained in $`(1/\lambda ,c)`$. This is clear in the case $`j=p`$, since $`\varphi _p=\phi `$. For $`j=0`$, $`\varphi _0(1/\lambda )=1`$, and $`\varphi _0(c)=\lambda c>1`$. If $`1jp1`$, $`\varphi _j`$ is decreasing, $`\varphi _j(c)<\varphi _j(1/\lambda )0`$ and $`\varphi _j(c)>\stackrel{ˇ}{u}^j(\zeta _1)=\zeta _{j+1}1/\lambda `$. Let $`X`$ be the convex hull of $`X_0\mathrm{}X_p`$. This is an open interval with closure contained in $`(1/\lambda ,c)`$, such that, for all $`j=0,\mathrm{},p`$, $`\varphi _j((1/\lambda ,c))X`$.
Suppose that $`w^{}`$ and $`w^{\prime \prime }`$ are distinct points in $`\mathrm{\Omega }(1/\lambda ,\xi _{\mathrm{max}})`$ such that $`\psi (w^{})=\psi (w^{\prime \prime })`$. This implies that $`w^{}`$ and $`w^{\prime \prime }`$ are not real, and have imaginary parts of the same sign. We inductively construct a sequence of triples $`\{w_n^{},w_n^{\prime \prime },j_n\}_{0n<\mathrm{}}`$, where $`w_0^{}=w^{}`$, $`w_0^{\prime \prime }=w^{\prime \prime }`$, and, for all $`n0`$, $`w_n^{}w_n^{\prime \prime }`$ are non-real, $`\psi (w_n^{})=\psi (w_n^{\prime \prime })`$, and $`0j_np`$ is such that $`w_{n+1}^{}=\varphi _{j_n}(w_n^{})`$, $`w_{n+1}^{\prime \prime }=\varphi _{j_n}(w_n^{\prime \prime })`$. Assuming that $`w_n^{}`$ and $`w_n^{\prime \prime }`$ have already been constructed, it follows from (7.140) and the definition (7.142) of the functions $`\varphi _j`$ that there is a unique $`j_n`$ in $`[0,p]`$ such that $`\varphi _{j_n}(w_n^{})\varphi _{j_n}(w_n^{\prime \prime })`$ and either $`j_n=p`$ or $`\varphi _{j_n+1}(w_n^{})=\varphi _{j_n+1}(w_n^{\prime \prime })`$. This implies that $`\psi (\varphi _{j_n}(w_n^{}))=\psi (\varphi _{j_n}(w_n^{\prime \prime }))`$, and we take $`w_{n+1}^{}=\varphi _{j_n}(w_n^{})`$, $`w_{n+1}^{\prime \prime }=\varphi _{j_n}(w_n^{\prime \prime })`$. It is easy to see (as e.g. in \[E2\]) that, as $`n`$ tends to infinity, the Poincaré distances, relative to $`𝐂_+𝐂_{}X`$, of $`w_n^{}`$ and $`w_n^{\prime \prime }`$ to the segment $`X`$ tend to 0. Therefore as $`n`$ becomes sufficiently large, the points $`w_n^{}`$ and $`w_n^{\prime \prime }`$ enter a complex neighborhood of the real segment $`X`$ so thin that $`\psi `$ is injective there, producing a contradiction. Thus $`\psi `$, and therefore also $`u`$ and $`\phi `$ are univalent in their respective domains.
### 7.3 Boundary values of $`u`$
We show in this subsection that the restriction of $`u`$ to the upper half-plane $`𝐂_+`$ extends to a continuous bounded injective function on the closed upper half-plane $`\overline{𝐂_+}`$. The same, of course, holds in the lower half-plane, since $`u(z)=u^{}(z^{})`$.
We rewrite (7.141) as
$$u(z)=F(u(\lambda z)),$$
(7.143)
$$F(z)=\frac{1}{\lambda ^\nu }u(\lambda ^{\nu 1}\stackrel{ˇ}{u}^{p1}(z)).$$
(7.144)
The function $`F`$ is anti-Herglotzian, holomorphic and univalent in $`\mathrm{\Omega }(1,\zeta _1)`$ and vanishes at $`\zeta _1=u(\lambda )`$. It has a fixed point at $`u(0)=z_1\lambda ^{1\nu }`$ with $`F^{}(z_1\lambda ^{1\nu })=1/\lambda `$, and (since it is strictly decreasing) no other fixed point in the real interval $`[1,\zeta _1]`$. The equation (7.143) can be rewritten as $`F=u(\lambda ^1)u^1`$ on the intersection of the domain of $`F`$ with the range of $`u`$. This range includes the real segment $`(0,u(\lambda ^1))`$ and hence $`(0,\zeta _1]`$, since $`u(\lambda ^1)>u(\lambda )=\zeta _1`$. Any periodic orbit of $`F`$ in $`[1,\zeta _1]`$ must be contained in $`(0,\zeta _1]`$, so that $`\{u(0)\}`$ is the only such orbit. Therefore the Herglotz function $`F^2`$ also has $`u(0)`$ as its unique real fixed point, with $`F_{}^{2}{}_{}{}^{}(u(0))=\lambda ^2`$. Both $`F`$ and $`F^2`$ have zero angular derivative at $`\mathrm{}`$ since $`zF(z)^r`$ is anti-Herglotzian and $`zF^2(z)^r`$ is Herglotzian. As in \[EL, E2\], the theory of iterations of maps of $`𝐂_+`$ into itself (Wolff-Denjoy-Valiron Theorem, see \[V, Mi\]) shows that, uniformly on every compact subset of $`𝐂_+`$, $`F^{2n}`$ tends to a finite constant $`c`$ as $`n\mathrm{}`$. In particular for all $`z`$ in a fixed compact subset of $`𝐂_{}`$, $`u(\lambda ^{2n}z)=F^{2n}(u(z))`$ tends uniformly to $`c`$, i.e. $`c=u(i\mathrm{})`$.
For $`n=0,1,2,\mathrm{}`$, we denote $`I_n`$ the closed real interval
$$I_n=(\lambda )^n[1,\lambda ^2].$$
(7.145)
We first consider the case when $`\xi _{\mathrm{max}}=\lambda ^2`$ (in particular the case $`0<\nu 1)`$, for which the argument of the case $`p=1`$ \[EL, E2\] can be repeated almost verbatim. Let $`z`$ follow $`I_0i0`$. Then $`u(z)`$ follows the segment
$$\tau _0=e^{i\pi /r}[0,|U(\lambda ^2)|^{1/r}].$$
(7.146)
If $`z`$ crosses the interior of $`I_0`$ into $`𝐂_+`$, $`u`$ gets continued by $`v_0e^{2\pi i/r}u`$. The image of $`𝐂_+`$ given by $`v_0`$ is contained in $`\{z:\pi /r<\mathrm{arg}z<2\pi /r\}`$. It is contained in $`𝐂_+`$ if and only if $`r2`$ (in particular if $`r`$ is an integer). Let $`𝒱_0`$ denote the open set $`\{z𝐂_+:v_0(z)𝐂_+\}`$, and, for $`n1`$, $`𝒱_n=\lambda ^1𝒱_{n1}^{}`$ (so that $`𝒱_n=𝐂_+`$ when $`r2`$). If $`z`$ follows $`I_1i0`$, then $`\lambda z`$ follows $`I_0+i0`$ and, by (7.143), $`u(z)`$ follows the analytic arc $`\tau _1=F(\tau _0^{})`$ which lies entirely in $`𝐂_+`$ except for its starting point, $`u(\lambda ^1)=F(0)`$. An easy induction shows that when $`z`$ follows $`I_ni0`$, ($`n1`$), $`u(z)`$ follows an analytic arc $`\tau _n`$ lying entirely in $`𝐂_+`$ for $`n>1`$, and $`u`$ can be continued across the interior of $`I_n`$ into $`𝐂_+`$ by a function $`v_n`$ holomorphic in $`𝒱_n`$, with $`v_n(𝒱_n)𝐂_+`$ and
$`\tau _n`$ $`=`$ $`F(\tau _{n1}^{}),`$
$`v_n(z)`$ $`=`$ $`F(v_{n1}(\lambda z^{})^{}).`$ (7.147)
The starting point of $`\tau _{n+2}`$ is the end of $`\tau _n`$, and $`\tau _{n+2}=F^2(\tau _n)`$. Hence the arcs $`\tau _n`$ tend to the point $`c`$. Thus $`u|𝐂_+`$ extends to a continuous bounded function on $`\overline{𝐂_+}`$. This function is injective. Indeed at each step of its inductive construction, a new extension is obtained by composing copies of the previously constructed extension and scalars.
We now consider the case when $`\xi _{\mathrm{max}}<\lambda ^2`$ which occurs if $`\stackrel{ˇ}{u}(\lambda ^1)>\lambda ^1`$. (In particular for $`\nu =2`$, $`\zeta _p=\lambda ^1<\zeta _{p+1}=\stackrel{ˇ}{u}(\lambda ^1)`$.) Recall that we denote $`\xi _j`$, for $`1jp`$, the unique number in $`[\zeta _1/\lambda ,\lambda ^2]`$ such that
$$\stackrel{ˇ}{u}^{pj}(\lambda \xi _j)=\lambda ^1,$$
(7.148)
and that
$$\xi _{\mathrm{max}}=\xi _1<\mathrm{}<\xi _p=\lambda ^2,$$
$$\zeta _j\lambda \xi _jj[1,p].$$
(7.149)
Suppose $`z`$ follows $`[1,\xi _{\mathrm{max}}]i0`$. Then $`u(z)=e^{\pm i\pi /r}|U(z)|^{1/r}`$ follows the segment $`e^{\pm i\pi /r}[0,|U(\xi _{\mathrm{max}})|^{1/r}]`$. Hence if $`z`$ follows $`(1/\lambda )[1,\xi _{\mathrm{max}}]i0`$, $`u(z)`$ is given by (7.143), and follows an analytic arc entirely contained in $`𝐂_+`$ except for its starting point $`\stackrel{ˇ}{u}(1/\lambda )`$. Thus $`zu(zi0)`$ now has a continuous, non-real extension to $`[\xi _{\mathrm{max}}/\lambda ,1/\lambda )`$. The extension thus obtained of $`u|𝐂_{}`$ to $`𝐂_{}[\xi _1/\lambda ,\xi _1]`$ is also obviously injective, as well as the conjugate extension of $`u|𝐂_+`$. Recall also that
$$1/\lambda <\stackrel{ˇ}{u}(1/\lambda )z_1\lambda ^{12\nu }<\zeta _1/\lambda ^\nu \xi _{\mathrm{max}}/\lambda .$$
(7.150)
Hence
$$\stackrel{ˇ}{u}(1/\lambda )(1/\lambda ,\xi _{\mathrm{max}}/\lambda ),$$
(7.151)
and
$$1\lambda ^{\nu 2}<\lambda ^{\nu 1}\stackrel{ˇ}{u}(1/\lambda )<\zeta _1/\lambda .$$
(7.152)
This shows that $`\lambda ^{\nu 1}\stackrel{ˇ}{u}(1/\lambda )`$ is in the domain of analyticity of $`U`$ and $`U`$ is negative there.
We assume inductively that there exists, for a certain $`j[1,p1]`$, a continuous injective extension, temporarily denoted $`u_{(j)}`$, of $`u|𝐂_{}`$ to $`𝐂_{}[\xi _j/\lambda ,\xi _j]`$, such that $`u_{(j)}(zi0)𝐂_+`$ if $`z[\xi _j/\lambda ,1/\lambda )`$ or if $`z(1,\xi _j]`$. This implies of course a symmetrical situation for $`u|𝐂_+`$. By abuse of notation we also denote $`u_{(j)}(z+i0)=u_{(j)}(z^{}i0)^{}`$. We also assume that (7.141) holds with $`u`$ replaced by $`u_{(j)}`$ in the domain of the latter. In order to prove the same for $`j+1`$, we denote
$$u_{(j+1/2)}(zi0)=\lambda ^\nu u_{(j)}(\lambda ^{\nu 1}\stackrel{ˇ}{u}_{(j)}^j(\stackrel{ˇ}{u}_{(j)}^{pj}(\lambda zi0))).$$
(7.153)
Note that the rhs of the above equation is equal to $`u_{(j)}(zi0)`$ in the domain of $`u_{(j)}`$ by the induction hypothesis. Thus $`u_{(j+1/2)}`$ is a new extension of $`u|𝐂_{}`$, which is injective wherever it is defined. If $`z`$ increases along $`(\xi _j,\xi _{j+1}]`$, $`\stackrel{ˇ}{u}_{(j)}^{pj}(\lambda zi0)`$ moves along $`(1/\lambda ,\stackrel{ˇ}{u}(1/\lambda )]i0`$. This, by the induction hypothesis and (7.151), is within the domain of the already constructed $`\stackrel{ˇ}{u}_{(j)}`$, and $`\stackrel{ˇ}{u}_{(j)}^j(\stackrel{ˇ}{u}_{(j)}^{pj}(\lambda zi0))`$ moves along an arc entirely contained in $`𝐂_{}`$, and $`u_{(j+1/2)}(zi0)`$ moves along an arc entirely contained in $`𝐂_+`$. A little more detail is needed, since $`\stackrel{ˇ}{u}(1/\lambda )`$ is real, when $`z`$ moves in a small real interval containing $`\xi _j`$ so that $`\stackrel{ˇ}{u}^{pj}(\lambda zi0)`$ moves along a small real interval containing $`1/\lambda `$. If $`j=1`$, then $`u_{(1)}`$ is continuous and non-real at $`\lambda ^{\nu 1}\stackrel{ˇ}{u}(1/\lambda )\pm i0`$ since, as noted above, this is a point of analyticity of $`U`$. If $`j>1`$, $`\stackrel{ˇ}{u}_{(j)}^2(1/\lambda \pm i0)`$ is defined and non-real by (7.151). Denote now $`u_{(j+1)}(zi0)=\lambda ^\nu u_{(j+1/2)}(\lambda ^{\nu 1}\stackrel{ˇ}{u}_{(j+1/2)}^p(\lambda zi0))`$. This is a continuous injective extension of $`u|𝐂_{}`$ to $`𝐂_{}[\xi _{j+1}/\lambda ,\xi _{j+1}]`$ since the arc $`u_{(j+1/2)}([\xi _j,\xi _{j+1}]i0)`$ is entirely contained in $`𝐂_\pm `$. The construction makes it obvious that (7.141) holds with $`u`$ replaced by $`u_{(j+1)}`$ in the domain of the latter.
We conclude that $`u|𝐂_{}`$ has a continuous injective extension to $`𝐂_{}[\lambda ^3,\lambda ^2]`$ which takes real values only on $`[1/\lambda ,1]`$. It maps $`I_0i0`$ onto a union $`\tau _0`$ of $`p`$ consecutive arcs contained in $`𝐂_+`$ except for the point $`u(1)=0`$ : $`\tau _0=\tau _{00}\mathrm{}\tau _{0(p1)}`$, with $`\tau _{00}=u([1,\xi _1]i0)`$ and $`\tau _{0j}=u([\xi _j,\xi _{j+1}]i0)`$ for $`1j<p`$. It maps $`I_1i0`$ onto another finite union $`\tau _1=\tau _{10}\mathrm{}\tau _{1(p1)}`$, with $`\tau _{1j}=F(\tau _{0j}^{})`$, contained in $`𝐂_+`$ except for the point $`u(1/\lambda )`$. As in the previous case, $`u|𝐂_{}`$ extends to a continuous injective function on $`\overline{𝐂_{}}`$. The images $`\tau _n=u(I_ni0)`$ all lie in $`𝐂_+`$ for $`n>1`$. The sequence of the $`\tau _n=F(\tau _{n1}^{})=F^2(\tau _{n2})`$ tends to the point $`c`$.
Let $`I_{00}=[1,\xi _1]`$, $`I_{0j}=[\xi _j,\xi _{j+1}]`$ for $`1j<p`$, and $`I_{nj}=(\lambda )^nI_{0j}`$ for $`n𝐍`$, $`0j<p`$. If $`z`$ crosses $`(1,\xi _{\mathrm{max}})`$ from $`𝐂_{}`$ into $`𝐂_+`$, $`u(z)`$ gets continued by $`v_{00}(z)=e^{2i\pi /r}u(z)`$. If $`z`$ crosses $`(\xi _{\mathrm{max}}/\lambda ,1/\lambda )`$ from $`𝐂_{}`$ into $`𝐂_+`$, $`u(z)`$ gets continued by $`v_{10}(z)=F(v_{00}(\lambda z^{})^{})`$, holomorphic in $`𝒱_{10}=𝒱_1`$. If $`z`$ crosses $`(\xi _j,\xi _{j+1})`$ from $`𝐂_{}`$ into $`𝐂_+`$ (with $`1j<p`$), $`u(z)`$ gets continued by $`v_{0j}(z)=\lambda ^\nu u(\lambda ^{\nu 1}\stackrel{ˇ}{u}^{j1}(v_{10}(\stackrel{ˇ}{u}^{pj}(\lambda z^{}))^{}))`$. If $`z`$ crosses $`(\xi _{j+1}/\lambda ,\xi _j/\lambda )`$ from $`𝐂_{}`$ into $`𝐂_+`$, $`u(z)`$ gets continued by $`v_{1j}(z)=F(v_{0j}(\lambda z^{})^{})`$. If $`z`$ crosses the interior of $`I_{nj}`$ from $`𝐂_{}`$ into $`𝐂_+`$, $`u(z)`$ gets continued by $`v_{nj}(z)=F(v_{(n1)j}(\lambda z^{})^{})`$. If $`r2`$, all the functions $`v_{nj}`$ are holomorphic in $`𝐂_+`$ and map it into itself.
Note that, in all cases, the extension of $`u`$ to $`\overline{𝐂_{}}`$ (resp. $`\overline{𝐂_+}`$) takes real values only on $`[\lambda ^1,1]`$. The function $`F|𝐂_+`$ (resp. $`F|𝐂_{}`$) also has a continuous injective extension to $`\overline{𝐂_+}`$ (resp. $`\overline{𝐂_{}}`$) which takes real values only on the real segment $`[1,\zeta _1]`$. The point $`c`$ cannot be real. Indeed if we suppose it is and let $`w_0𝐂_+`$, $`w_n=F^{2n}(w_0)`$ for $`n𝐍`$, the sequences $`\{w_n\}`$ and $`\{F^2(w_n)\}`$ both tend to $`c`$, so that, by the continuity of the extensions of $`F`$ to $`\overline{𝐂_\pm }`$, $`F^2(c+i0)=c`$. Since this is real, $`F(c+i0)`$, hence also $`c`$, must belong to $`[1,\zeta _1]`$, and $`c`$ is a fixed point of $`F^2`$, i.e. $`c`$ coincides with $`z_1\lambda ^{1\nu }`$. But the latter is repulsive, contradicting the attractive property of $`c`$. Hence $`c`$ is in $`𝐂_+`$ and is a fixed point of $`F^2`$. It is attractive and unique by Schwarz’s lemma applied to $`F^2|𝐂_+`$. Therefore the compact sets $`\tau _n`$ converge geometrically to $`c`$. It follows that the functions $`u`$, $`\phi `$, $`\psi `$ are all bounded.
### 7.4 Commutativity for $`\nu =2`$
The following is a transcription into the notations of this paper of (a special case of) a result due to O. Lanford. This will prove that the properties of $`u`$ and $`\psi `$ recalled at the beginning of this section suffice to imply, in the case $`\nu =2`$, a form of commutativity for the functions $`\xi `$ and $`\eta `$ given by
$$\xi =(u)^1,\eta =\lambda \xi (1/\lambda ),\eta ^1=\lambda \stackrel{ˇ}{u}(1/\lambda ).$$
(7.154)
Recall that the functional equations (7.140), (7.141), and $`\nu =2`$, imply that $`\xi `$ and $`\eta `$ satisfy the system (1.5). With the notations of the beginning of this section, we have
###### Lemma 3
(Lanford) For every solution with $`\nu =2`$,
$$\psi (\lambda u(z/\lambda ))=\lambda ^r\psi (u(z)/\lambda )\mathrm{for}\mathrm{all}z\mathrm{\Omega }(\lambda ,1).$$
(7.155)
Proof. The domains of the two anti-Herglotzian functions
$`F_1`$ $`=`$ $`\psi \lambda u(1/\lambda ),`$
$`F_2`$ $`=`$ $`\lambda ^r\psi (1/\lambda )u,`$ (7.156)
are equal to $`\mathrm{\Omega }(\lambda ,1)`$. Indeed the function $`z\lambda u(z/\lambda )`$ has this domain and maps $`\lambda `$ to 0, and 1 to $`z_1v(1/\lambda )z_1/\lambda ^2<\zeta _1/\lambda `$, hence it maps $`\mathrm{\Omega }(\lambda ,1)`$ into the domain of $`\psi `$. The function $`(1/\lambda )u`$ is holomorphic in $`\mathrm{\Omega }(1/\lambda ,1)`$. It maps 1 to 0 and $`\lambda `$ to $`\zeta _1/\lambda `$, hence it also maps $`\mathrm{\Omega }(\lambda ,1)`$ into the domain of $`\psi `$ (and $`F_2`$ has a branch point at $`\lambda `$ since $`\psi `$ has one at $`\zeta _1/\lambda `$). We now substitute for $`u`$, in the equation for $`F_1`$, the r.h.s. of (7.141), and substitute for $`\psi `$, in the equation for $`F_2`$, the r.h.s. of the first equation in (7.140). This gives
$$F_1=\frac{1}{\lambda ^r}F_2G_0,F_2=\frac{1}{\lambda ^r}F_1G_0,$$
(7.157)
$$G_0(z)=\lambda \stackrel{ˇ}{u}^p(z)=\phi (z/\lambda ).$$
(7.158)
Since the functional equations (7.141) and (7.140) hold with domains, so does the system (7.157). In fact the anti-Herglotzian function $`G_0`$ maps the domain $`\mathrm{\Omega }(\lambda ,1)`$ into itself, since $`G_0`$ is holomorphic in $`\mathrm{\Omega }(\zeta _1,1)`$ and satisfies:
$$G_0(1)=\phi (1/\lambda )0,G_0(0)=\phi (0)<1,G_0(\lambda )=\phi (1)=1.$$
(7.159)
Since $`G_0`$ is strictly decreasing on $`[\lambda ,1]`$, it has there a unique fixed point $`\overline{x}(0,1)`$ which, by Schwarz’s lemma, is attractive and has $`\mathrm{\Omega }(\lambda ,1)`$ as a basin of attraction. Let $`\kappa =G_0^{}(\overline{x})(0,1)`$, and let $`h`$ be the linearizer of $`G_0`$ at $`\overline{x}`$, normalized by $`h^{}(\overline{x})=1`$. This is a function holomorphic in $`\mathrm{\Omega }(\lambda ,1)`$, and satisfying, in this domain, $`h=(1/\kappa )hG_0`$ (in particular $`h(\overline{x})=0`$). The point $`\overline{x}`$ is also the unique fixed point of the function $`G_0^2`$ in $`\mathrm{\Omega }(\lambda ,1)`$ and its normalized linearizer is also $`h`$ . On the other hand, because $`G_0`$ maps $`\mathrm{\Omega }(\lambda ,1)`$ into itself, the equation obtained by substituting the second equation in (7.157) into the first,
$$F_1=\frac{1}{\lambda ^{2r}}F_1G_0^2$$
(7.160)
holds in $`\mathrm{\Omega }(\lambda ,1)`$. Therefore $`F_1=c_1h`$ with $`c_1=F_1^{}(\overline{x})`$, and $`\kappa =\lambda ^r`$. The second equation in (7.157) now reads $`F_2=(c_1/\kappa )hG_0=c_1h`$. Therefore $`F_1`$ and $`F_2`$ coincide, which is the assertion of the lemma.
In particular, for $`z=\lambda `$, this gives $`1=\psi (0)=\lambda ^r\psi (\zeta _1/\lambda )`$, i.e. $`\psi (\zeta _1/\lambda )=1/\lambda ^r`$. Both sides of (7.155) must vanish at $`\overline{x}`$, hence $`\stackrel{ˇ}{u}(\overline{x}/\lambda )=1/\lambda `$, $`u(\overline{x})=\lambda `$ so that $`\overline{x}=\lambda \zeta _{p1}=\zeta _1`$. Since $`F_2(1)=\lambda ^r`$, the common range of $`F_1(z)`$ and $`F_2(z)`$ as $`z`$ varies in $`[\lambda ,1]`$ is $`[\lambda ^r,1]`$. The identity (7.155) continues to hold if $`\psi `$ is replaced with $`U=(z_1/\lambda )^r\psi `$ on both sides. In order to translate this identity in terms of $`\xi `$ and $`\eta `$, we denote
$$q(z)=|z|^r\mathrm{sign}(z)z𝐑,\widehat{u}(z)=q^1U(z)z[1/\lambda ,\zeta _1/\lambda ].$$
(7.161)
The function $`\widehat{u}`$ is strictly decreasing, with range containing $`[z_1/\lambda ^2,1/\lambda ]`$, coincides with $`u`$ on $`[1/\lambda ,1]`$, and satisfies
$$\begin{array}{ccc}& \widehat{u}=(1/\lambda ^2)\widehat{u}\lambda \stackrel{ˇ}{u}^p\lambda & \mathrm{on}[1/\lambda ,\zeta _1/\lambda ],\\ & \widehat{u}\lambda u(1/\lambda )=\lambda \widehat{u}(1/\lambda )u& \mathrm{on}[\lambda ,1].\end{array}$$
(7.162)
Let
$$\widehat{\xi }=(\widehat{u})^1,\widehat{\eta }=\lambda \widehat{\xi }(1/\lambda ).$$
(7.163)
Then $`\widehat{\xi }`$ is an extension of $`\xi `$ to an interval containing $`(1/\lambda ,z_1/\lambda ^2)`$, $`\widehat{\eta }`$ is an extension of $`\eta `$, and
$$\begin{array}{ccc}& \widehat{\xi }=(1/\lambda ^2)\eta ^p\widehat{\xi }\lambda ^2,\widehat{\xi }=(1/\lambda )\widehat{\eta }(\lambda ),& \mathrm{on}(1/\lambda ,z_1/\lambda ^2),\\ & \eta \widehat{\xi }=\xi \widehat{\eta }& \mathrm{on}(z_1/\lambda ,z_1).\end{array}$$
(7.164)
## 8 Behavior of fixed points as $`r\mathrm{}`$, $`0<\nu 1`$
In this section we consider only the cases when $`0<\nu 1`$ and $`p2`$ (and, of course, $`r\nu 1(p1)(1\nu )>0`$). In the case $`p=1`$, the behavior of solutions as $`r\mathrm{}`$ was first elucidated by Eckmann and Wittwer in \[EW\], and also studied in \[E1\] (for $`\nu =1`$) and \[EE\] (for $`1\nu 2`$), and the method of \[E1, EE\] extends trivially to $`0<\nu 1`$. The case $`p2`$ requires some additional work.
### 8.1 The functions $`V`$ and $`W`$
The functional equation implies
$$\psi (z)=V(\psi (\lambda z))=W(\psi (\lambda ^2z)),z𝐂_+𝐂_{}(1/\lambda ,1/\lambda ^2),$$
(8.165)
where
$`V(\zeta )`$ $`=`$ $`f(\zeta ^{1/r}),`$
$`f(z)`$ $`=`$ $`{\displaystyle \frac{1}{\lambda ^{r\nu }}}\psi (\lambda ^{\nu 1}\stackrel{ˇ}{u}^{p1}(z_1\lambda ^{1\nu }z)),`$
$`W`$ $`=`$ $`VV.`$ (8.166)
Recall that
$$\stackrel{ˇ}{u}(z)=z_1\lambda ^{1\nu }\psi (z)^{1/r}.$$
(8.167)
The function $`f`$ is anti-Herglotzian and holomorphic in $`\mathrm{\Omega }(1/z_1\lambda ^{1\nu },1/z_1\lambda ^{2\nu })`$. We denote $`\zeta _{\mathrm{max}}=(1/z_1\lambda ^{2\nu })^r`$. These functions satisfy
$$V(1)=f(1)=1,V^{}(1)=\frac{1}{\lambda },f^{}(1)=\frac{r}{\lambda }.$$
(8.168)
Since $`\psi (1)=0`$, $`V`$ vanishes at $`\alpha =\psi (\lambda )`$, and $`f`$ vanishes at $`v(\lambda )=(z_1\lambda ^{1\nu })^1\zeta _1`$. We also define
$$\widehat{V}(\zeta )=1V(1\zeta ),\widehat{W}=\widehat{V}\widehat{V}.$$
(8.169)
Since the functional equations (8.165) hold for all $`z`$ in the domain of $`\psi `$, the real ranges of $`V`$ and $`W`$ contain that of $`\psi `$. The following estimates follow \[EE\] and \[E1\]. In the domain of $`V`$,
$$\frac{V^{\prime \prime }(\zeta )}{V^{}(\zeta )}=\frac{1}{r\zeta }\left(r1\frac{zf^{\prime \prime }(z)}{f^{}(z)}\right),z=\zeta ^{1/r}.$$
(8.170)
For real $`\zeta (0,\zeta _{\mathrm{max}})`$,
$`{\displaystyle \frac{V^{\prime \prime }(\zeta )}{V^{}(\zeta )}}`$ $``$ $`{\displaystyle \frac{1}{r\zeta }}\left(r1{\displaystyle \frac{2z}{1/\lambda ^{2\nu }z_1z}}\right)`$ (8.171)
$`=`$ $`{\displaystyle \frac{1}{r\zeta }}\left(r{\displaystyle \frac{1+\lambda ^{2\nu }z_1z}{1\lambda ^{2\nu }z_1z}}\right).`$
Recalling the bound $`r\nu (1+\lambda )/(1\lambda )`$, we find that
$$\frac{V^{\prime \prime }(\zeta )}{V^{}(\zeta )}\frac{1\nu }{\zeta }\mathrm{for}0<\zeta (z_1\lambda ^{1\nu })^r$$
(8.172)
This is in particular satisfied if $`\zeta =\alpha =((z_1\lambda ^{1\nu })^1\zeta _1)^r`$, since $`\zeta _1\lambda ^{1\nu }1`$. Integrating the inequality (8.172) from 1 to $`\zeta >1`$, using $`V^{}(1)=1/\lambda `$ and $`V(1)=1`$, gives
$$V(\zeta )>1\frac{1}{\lambda \nu }(\zeta ^\nu 1)\alpha >(1+\lambda \nu )^{1/\nu }(1+\lambda ).$$
(8.173)
It follows similarly from (8.170) that
$$\frac{V^{\prime \prime }(\zeta )}{V^{}(\zeta )}\frac{1}{r\zeta }(r1+\frac{2z}{z+\lambda ^{\nu 1}/z_1})=\frac{1}{\zeta }(1\frac{1z_1\lambda ^{1\nu }z}{r(1+z_1\lambda ^{1\nu }z)}),$$
(8.174)
so that
$$\frac{V^{\prime \prime }(\zeta )}{V^{}(\zeta )}\frac{1}{\zeta }\zeta (0,\alpha ).$$
(8.175)
Integrating this from 1 to $`\zeta >1`$ gives
$$V^{}(\zeta )\frac{1}{\lambda \zeta }\zeta (1,\alpha ),$$
(8.176)
$$V(\zeta )1\frac{1}{\lambda }\mathrm{log}\zeta \zeta (1,\alpha )\alpha e^\lambda .$$
(8.177)
Since $`V=fq^1`$ where $`q^1(\zeta )=\zeta ^{1/r}`$, the Schwarzian derivative $`SV`$ of $`V`$ satisfies, for real $`\zeta `$ in the domain of $`V`$,
$$SV(\zeta )Sq^1(\zeta )=\frac{1r^2}{2\zeta ^2}.$$
(8.178)
The function $`W`$ is Herglotzian and holomorphic in $`\mathrm{\Omega }(0,\alpha )`$, where $`\alpha =\psi (\lambda )=V^1(0)`$ (since $`V(0)\lambda ^{r\nu }<\zeta _{\mathrm{max}}`$). It has a repelling fixed point at 1 with multiplier $`\lambda ^2`$. $`\widehat{W}`$ is Herglotzian and holomorphic in $`\mathrm{\Omega }(1\alpha ,1)`$ and has a fixed point at 0. By (8.178),
$$SW(\zeta )\frac{(1r^2)}{2}\left(\frac{V^{}(\zeta )^2}{V(\zeta )^2}+\frac{1}{\zeta ^2}\right).$$
(8.179)
### Lower bound for $`\widehat{W}`$ in $`[0,1]`$.
For $`0<\zeta <1`$, the convexity of $`V`$ implies:
$$V^{}(\zeta )\frac{V(\zeta )1}{1\zeta },$$
(8.180)
hence
$$\frac{V(\zeta )}{V^{}(\zeta )}1\zeta \frac{1}{V^{}(\zeta )}1\zeta +\lambda ,$$
(8.181)
It follows that
$$2SW(\zeta )(1r^2)\left[\frac{1}{(1\zeta +\lambda )^2}+\frac{1}{\zeta ^2}\right],$$
(8.182)
and hence
$$2S\widehat{W}(\zeta )(1r^2)\left[\frac{1}{(\zeta +\lambda )^2}+\frac{1}{(1\zeta )^2}\right].$$
(8.183)
In (0, 1), the r.h.s. has a minimum at $`\zeta =(1\lambda )/2`$, and, using the bound on $`r(1+\lambda )/(1\lambda )`$, we get
$$\frac{d}{d\zeta }\frac{\widehat{W}^{\prime \prime }(\zeta )}{\widehat{W}^{}(\zeta )}S\widehat{W}(\zeta )s(\lambda )\frac{16\lambda }{(1+\lambda )^4}.$$
(8.184)
By (8.175) and
$$\frac{\widehat{W}^{\prime \prime }(\zeta )}{\widehat{W}^{}(\zeta )}=\frac{\widehat{V}^{\prime \prime }(\widehat{V}(\zeta ))}{\widehat{V}^{}(\widehat{V}(\zeta ))}\widehat{V}^{}(\zeta )+\frac{\widehat{V}^{\prime \prime }(\zeta )}{\widehat{V}^{}(\zeta )},$$
(8.185)
it follows that
$$\frac{\widehat{W}^{\prime \prime }(0)}{\widehat{W}^{}(0)}=\left(\frac{1}{\lambda }1\right)\frac{\widehat{V}^{\prime \prime }(0)}{\widehat{V}^{}(0)}=\left(\frac{1}{\lambda }1\right)\frac{V^{\prime \prime }(1)}{V^{}(1)}\left(\frac{1}{\lambda }1\right).$$
(8.186)
Hence,
$$\frac{\widehat{W}^{\prime \prime }(\zeta )}{\widehat{W}^{}(\zeta )}\frac{\widehat{W}^{\prime \prime }(0)}{\widehat{W}^{}(0)}+s(\lambda )\zeta \left(\frac{1}{\lambda }1\right)+s(\lambda )\zeta ,$$
(8.187)
$`\mathrm{log}\widehat{W}^{}(\zeta )`$ $``$ $`2\mathrm{log}(1/\lambda )\left({\displaystyle \frac{1}{\lambda }}1\right)\zeta +s(\lambda )\zeta ^2/2`$ (8.188)
$``$ $`\left[2\mathrm{log}(1/\lambda )\left({\displaystyle \frac{1}{\lambda }}1\right)\right]+s(\lambda )\zeta ^2/2.`$
As a function of $`\lambda `$ in $`(0,1)`$, the first bracket in the last expression has a unique maximum at $`1/2`$ and vanishes at 1. Since it is positive at $`e^1`$, it is non negative in $`[e^1,1]`$. Hence, for $`\lambda e^1`$ and $`0\zeta <1`$,
$$\widehat{W}^{}(\zeta )1+s(\lambda )\zeta ^2/2,$$
(8.189)
$$\widehat{W}(\zeta )\zeta \left(1+\frac{s(\lambda )}{6}\zeta ^2\right),$$
(8.190)
and we note that, for $`\lambda 1/4`$, $`s(\lambda )1`$.
On the other hand $`\widehat{W}`$ is Pick with 0 angular derivative at infinity in $`𝐂_+𝐂_{}(1\alpha ,1)`$, and vanishes at 0. Hence there is a positive measure $`\rho `$ with support in $`𝐑(1\alpha ,1)`$ such that
$$\widehat{W}(\zeta )=_{𝐑(1\alpha ,1)}\left(\frac{1}{t\zeta }\frac{1}{t}\right)𝑑\rho (t),_{𝐑(1\alpha ,1)}\frac{d\rho (t)}{t^2}=\frac{1}{\lambda ^2}.$$
(8.191)
Hence, for $`0\zeta <1`$,
$`\widehat{W}(\zeta ){\displaystyle \frac{\zeta }{\lambda ^2}}\underset{t(1\alpha ,1)}{inf}{\displaystyle \frac{t}{t\zeta }}`$ $`=`$ $`{\displaystyle \frac{\zeta (\alpha 1)}{\lambda ^2(\alpha 1+\zeta )}}`$ (8.192)
$``$ $`{\displaystyle \frac{\zeta }{\lambda (1+\lambda )}}.`$
Here we have used the lower bound (4.46) for $`\alpha `$. For $`\lambda 1/2`$, this implies $`\widehat{W}(\zeta )4\zeta /3\zeta (1+\zeta ^2/6)`$, so that, for all $`\lambda `$ and all $`\zeta (0,1)`$,
$$\widehat{W}(\zeta )\zeta (1+c^{}\zeta ^2),c^{}=1/6.$$
(8.193)
###### Remark 8.1
Let $`\zeta `$, $`y`$, $`a^{}`$, and $`m`$ be strictly positive real numbers such that
$$0<\zeta (1+a^{}\zeta ^2)ym.$$
(8.194)
Then
$$\zeta y(1ay^2),a=\frac{a^{}}{1+3a^{}m^2}.$$
(8.195)
Indeed, note first that $`am^21/3<1`$. Moreover $`\zeta z`$ for any $`z`$ such that $`a^{}z^3+zy0`$, and inserting $`z=y(1ay^2)`$ in this expression gives
$$y^3[a^{}(1ay^2)^3a]y^3[a^{}(13am^2)a]=0.$$
This remark (with $`m=1`$) and the lower bound (8.193) imply that $`\widehat{W}^1`$ is defined on $`[0,1)`$, and that, for all $`y[0,1)`$,
$$\widehat{W}^1(y)y(1cy^2),c=\frac{c^{}}{1+3c^{}}=1/9.$$
(8.196)
### Lower bound for $`W`$ in $`[1,\alpha ]`$.
For $`1\zeta \alpha `$, the inequalities (8.179) and (8.176), together with $`0V(\zeta )1`$, give
$$SW(\zeta )\frac{1}{2\zeta ^2}(1r^2)(\lambda ^2+1)\frac{1}{\zeta ^2}\frac{2}{(1+\lambda )^2}(\lambda ^1+\lambda ).$$
(8.197)
The last inequality follows from the lower bound on $`r`$ already used above. The last expression is decreasing in $`\lambda `$, so that, finally,
$$SW(\zeta )\frac{1}{\zeta ^2}\zeta (1,\alpha ).$$
(8.198)
Since $`W^{\prime \prime }(1)/W^{}(1)=\widehat{W}^{\prime \prime }(0)/\widehat{W}^{}(0)0`$ (see (8.186)), for $`1\zeta \alpha `$,
$$\frac{W^{\prime \prime }(\zeta )}{W^{}(\zeta )}_1^\zeta t^2𝑑t=(\zeta 1)/\zeta (\zeta 1)/e,$$
(8.199)
by using the upper bound $`\alpha e`$, and hence
$`W^{}(\zeta )`$ $``$ $`\lambda ^2(1+(\zeta 1)^2/2e),`$
$`W(\zeta )1`$ $``$ $`(\zeta 1)(1+k^{}(\zeta 1)^2),k^{}=1/6e\zeta (1,\alpha ).`$ (8.200)
The function $`\underset{¯}{W}(\zeta )=W(\zeta +1)1`$ is thus defined on $`[0,\alpha 1]`$, where it satisfies
$$\underset{¯}{W}(\zeta )\zeta (1+k^{}\zeta ^2).$$
(8.201)
We note that $`W(\alpha )=W(\psi (\lambda ))=\psi (\lambda ^1)`$, hence the range of $`W|(1,\alpha )`$ contains in particular $`\psi (1)`$. We wish to apply Remark 8.1 to the inverse function $`\underset{¯}{W}^1`$ restricted to $`[0,\psi (1)1]`$, and we first obtain an upper bound for $`\psi (1)`$. We use the representation (2.18) :
$$\begin{array}{cccc}\mathrm{log}\psi (1)\mathrm{log}\psi (\lambda )& =& _{𝐑(\lambda ^1,1)}\sigma (t)\left(\frac{1}{t+\lambda }\frac{1}{t+1}\right)𝑑t& \\ & & _{𝐑(\lambda ^1,1)}\left(\frac{1}{t+\lambda }\frac{1}{t+1}\right)𝑑t& =\mathrm{log}2\end{array}$$
(8.202)
which yields (using (8.177))
$$\psi (1)2\psi (\lambda )2e^\lambda <2e.$$
(8.203)
Thus (8.201) and Remark 8.1, with $`m=2e1`$, show that
$$\underset{¯}{W}^1(y)y(1ky^2),k=1/(6e+3(2e1)^2),y[0,\psi (1)1].$$
(8.204)
Note that we have obtained the following bounds:
$$1+\lambda \psi (\lambda )e^\lambda ,\psi (1)2\psi (\lambda )2e^\lambda .$$
(8.205)
This provides upper and lower bounds for $`y_0=z_1^r`$. Indeed from $`\zeta _1=z_1\lambda ^{1\nu }v(\lambda )\lambda ^{1\nu }=\zeta _p`$, and $`z_1\lambda ^{1\nu }v(1)z_1\lambda ^{1\nu }v(\zeta _p)\zeta _p`$, it follows $`z_11/v(\lambda )`$ and $`z_11/v(1)`$, hence
$$(2e)^1y_0(1+\lambda )^1.$$
(8.206)
### 8.2 The functions $`H_\pm `$
We define
$$H_\pm (w)=\psi (\pm e^{\beta w}),\beta =\mathrm{log}(1/\lambda ),\widehat{H}_\pm =1H_\pm .$$
(8.207)
$`H_+`$ is holomorphic in rhe cut strip
$$\mathrm{\Delta }_+(\lambda )=\{w𝐂:|\mathrm{Im}w|<\pi /\beta \}(2+𝐑_+).$$
(8.208)
It maps points in $`𝐂_\pm `$ into $`𝐂_{}`$. It is decreasing on the reals, tends to 1 at $`\mathrm{}`$, and vanishes at 0. $`H_{}`$ is holomorphic in the cut strip
$$\mathrm{\Delta }_{}(\lambda )=\{w𝐂:|\mathrm{Im}w|<\pi /\beta \}(1+𝐑_+),$$
(8.209)
maps points in $`𝐂_\pm `$ into $`𝐂_\pm `$, is increasing on the reals and tends to 1 at $`\mathrm{}`$. They satisfy
$`H_\pm (w)`$ $`=`$ $`V(H_{}(w1))=W(H_\pm (w2)),`$
$`\widehat{H}_\pm (w)`$ $`=`$ $`\widehat{V}(\widehat{H}_{}(w1))=\widehat{W}(\widehat{H}_\pm (w2)).`$ (8.210)
Moreover
$$\frac{H_\pm ^{\prime \prime }(w)}{H_\pm ^{}(w)}=\frac{\widehat{H}_\pm ^{\prime \prime }(w)}{\widehat{H}_\pm ^{}(w)}=\beta \left(1+\frac{z\psi ^{\prime \prime }(z)}{\psi ^{}(z)}\right),z=\pm e^{\beta w}.$$
(8.211)
Since (for $`0<\nu 1`$) $`\psi `$ is anti-Herglotzian in $`𝐂_+𝐂_{}(1/\lambda ,1/\lambda ^2)`$, the inequalities (2.10) imply, for $`0<z=e^{\beta w}<1/\lambda `$, i.e. for all $`w(\mathrm{},1)`$,
$$\frac{H_+^{\prime \prime }(w)}{H_+^{}(w)}\beta \left(1\frac{2\lambda z}{1+\lambda z}\right)0.$$
(8.212)
For $`0<z=e^{\beta w}<1/\lambda `$, i.e. again for all $`w(\mathrm{},1)`$, we find similarly that
$$\frac{H_{}^{\prime \prime }(w)}{H_{}^{}(w)}\beta \left(1+\frac{2\lambda ^2z}{1\lambda ^2z}\right)0.$$
(8.213)
In other words, $`H_{}`$ and $`\widehat{H}_+`$ are increasing and convex, $`H_+`$ is decreasing and concave. In particular, for $`w<0`$, using (8.196),
$`2\widehat{H}_+^{}(w)`$ $``$ $`\widehat{H}_+(w)\widehat{H}_+(w2)`$ (8.214)
$`=`$ $`\widehat{H}_+(w)\widehat{W}^1(\widehat{H}_+(w))c\widehat{H}_+(w)^3.`$
Integrating this with the initial condition $`\widehat{H}_+(0)=1`$ gives
$`\widehat{H}_+(w)`$ $``$ $`(1cw)^{1/2},`$
$`H_+(w)`$ $``$ $`1(1cw)^{1/2}w𝐑_{}(c=1/9).`$ (8.215)
Similarly, defining $`\underset{¯}{H}_{}(w)=H_{}(w)1`$, recalling that $`H_{}(0)=\psi (1)`$, $`H_{}(1)=\psi (\lambda )`$, we obtain, using (8.204),
$$\underset{¯}{H}_{}^{}(w)k\underset{¯}{H}_{}(w)^3/2,$$
$$H_{}^{}(w)k(H_{}(w)1)^3/2w𝐑_{}(k=1/(6e+3(2e1)^2)).$$
(8.216)
We will need a lower bound for $`H_{}^{}(w)/H_{}(w)`$ in the interval $`w[1,0]`$. This is provided by the lower bound $`H_{}(1)=\psi (\lambda )1+\lambda `$, and by (8.216) :
$`{\displaystyle \frac{H_{}^{}(w)}{H_{}(w)}}`$ $``$ $`{\displaystyle \frac{k(H_{}(w)1)^3}{2H_{}(w)}}`$ (8.217)
$``$ $`{\displaystyle \frac{k(H_{}(1)1)^3}{2H_{}(1)}}{\displaystyle \frac{k\lambda ^3}{2(1+\lambda )}}w[1,0].`$
### 8.3 Lower bound on $`\tau `$
Recall that the function $`\phi `$ satisfies
$`\phi (z)`$ $`=`$ $`\lambda ^{\nu 1}\stackrel{ˇ}{u}^p(\lambda z),z𝐂_+𝐂_{}(1/\lambda ,1/\lambda ^2),\phi (1)=1,`$
$`\phi ^{}(1)`$ $`=`$ $`\tau ^\nu =\lambda ^\nu {\displaystyle \underset{j=0}{\overset{p1}{}}}\stackrel{ˇ}{u}^{}(\zeta _j),\lambda \zeta _j=\stackrel{ˇ}{u}^j(\lambda )\lambda ^{1\nu }.`$ (8.218)
Let $`T(w)=e^{\beta w}`$, $`\beta =\mathrm{log}(1/\lambda )`$. Then the function
$$X=T^1\phi T$$
(8.219)
is given by
$`X(w)`$ $`=`$ $`\nu +1+Y^p(w1)w(\mathrm{},2),`$
$`Y(w)`$ $`=`$ $`T^1\stackrel{ˇ}{u}T(w)`$ (8.220)
$`=`$ $`{\displaystyle \frac{\mathrm{log}y_0}{\mathrm{log}(1/\tau )}}+\nu 1+{\displaystyle \frac{1}{\mathrm{log}(1/\tau )}}\mathrm{log}H_{}(w)w(\mathrm{},1).`$
It satisfies $`X(0)=0`$ and
$$X^{}(0)=\tau ^\nu =\underset{j=0}{\overset{p1}{}}Y^{}(w_j)=\underset{j=0}{\overset{p1}{}}\frac{1}{\mathrm{log}(1/\tau )}\frac{H_{}^{}(w_j)}{H_{}(w_j)},$$
(8.221)
where
$$1w_j=\frac{\mathrm{log}\zeta _j}{\mathrm{log}(1/\lambda )}\nu 1.$$
(8.222)
Hence by (8.217),
$$\tau ^{\nu /p3/r}\mathrm{log}(1/\tau )\frac{k}{4},k=1/(6e+3(2e1)^2).$$
(8.223)
When $`r>3p/\nu `$, this provides a lower bound for $`\tau `$. We may e.g. rewrite (8.223) as
$$y\mathrm{log}(1/y)(\nu /p3/r)k/4,y=\tau ^{\nu /p3/r}.$$
(8.224)
### 8.4 Limiting fixed points
The preceding subsections have shown that, for any solution, the associated functions have the following properties:
(1) The function $`V`$ is holomorphic and anti-Herglotzian in $`𝐂_+𝐂_{}(0,\zeta _{\mathrm{max}})`$, where $`\zeta _{\mathrm{max}}\tau ^{\nu 2}8^{(2\nu )/\nu }`$. It satisfies $`V(1)=1`$ and $`V^{}(1)=1/\lambda `$.
(2) The function $`W=VV`$ is holomorphic and Herglotzian in $`𝐂_+𝐂_{}(0,\alpha )`$, where $`(1+\lambda )\alpha =V^1(0)e^\lambda `$.
(3) The function $`H_+`$ is holomorphic in the cut strip $`\mathrm{\Delta }_+(\lambda )`$ (see (8.208)), maps points in $`𝐂_\pm `$ into $`𝐂_{}`$, vanishes at 0, and satisfies the bound (8.215).
(4) The function $`H_{}`$ is holomorphic in the cut strip $`\mathrm{\Delta }_{}(\lambda )`$ (see (8.209)), maps points in $`𝐂_\pm `$ into $`𝐂_\pm `$, and satisfies $`H_{}(1)=\alpha `$ and the bounds (8.216) and (8.217).
(5) $`\tau =\lambda ^r`$ is bounded above by $`\tau 8^{1/\nu }`$. For sufficiently large $`r`$, its is bounded below by (8.223), and for all $`r`$ by $`\lambda _0(p,r,\nu )^r`$ (see (4.65)).
(6) $`y_0=z_1^r`$ satisfies (8.206).
As a consequence every infinite sequence of solutions, with fixed $`\nu `$ and $`p`$, such that $`r\mathrm{}`$, contains an infinite subsequence such that $`\tau `$ and $`y_0`$ have limits in $`(0,1)`$, and that the functions $`V`$, $`W`$, $`H_\pm `$ tend, uniformly over compact sets, to non-constant functions, holomorphic in cut planes. Meanwhile, $`\lambda `$ and $`z_1>\lambda ^\nu `$ tend to 1 (see (7.133)), $`\psi `$ and $`u`$ tend to 1, uniformly over compact subsets of $`𝐂_+𝐂_{}(1,1)`$ (see (8.215)). However the functions
$$S_\pm (\zeta )=U(\pm \zeta ^{1/r})=y_0\tau ^{1\nu }H_\pm \left(\frac{\mathrm{log}\zeta }{\mathrm{log}(1/\tau )}\right)$$
(8.225)
have non-trivial limits and obey the functional equation:
$$S_\pm (\zeta )=\frac{1}{\tau ^\nu }S_+(\tau ^{\nu 1}S_{}^{p1}S_{}(\tau \zeta )).$$
(8.226)
## Appendix A Appendix 1. Proof of the inequality (4.54)
This inequality is equivalent to
$$(1x^2)\mathrm{log}((1+x^2)(1+x)^2)+4x^2\mathrm{log}(x)>0x(0,1),$$
(A.227)
or to
$$f_1(x)4xf_2(x)>0x(0,1),$$
(A.228)
where
$$f_1(x)=\mathrm{log}((1+x^2)(1+x)^2),f_2(x)=\frac{x\mathrm{log}(1/x)}{1x^2}.$$
(A.229)
The derivative $`f_2^{}(x)`$ has the sign of
$$2\mathrm{log}(x)2\left(\frac{1x^2}{1+x^2}\right)=\mathrm{log}(t)\frac{4}{1+t}+2,t=x^2.$$
(A.230)
The last expression vanishes at 1 and has negative derivative in $`t`$ on $`(0,1)`$. Hence $`f_2`$ is increasing on $`(0,1)`$. It tends to $`1/2`$ as $`x`$ tends to 1, so that $`f_2<1/2`$ on $`(0,1)`$. It now suffices to prove that $`f_1(x)2x>0`$ for all $`x(0,1)`$. This quantity vanishes for $`x=0`$, and
$$f_1^{}(x)2=\frac{2x^2(1x)}{(1+x^2)(1+x)}>0x(0,1).$$
(A.231)
|
warning/0004/cond-mat0004267.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
The possibility of realizing Bose–Einstein condensation in trapped dilute gases (see reviews ) demonstrated that a macroscopic number of bosons could be produced in a single quantum state of trapped atoms. The occupation of a single quantum state by a large number of bosons is the matter–wave analog of the storage of photons in a single mode of a laser cavity. The device that could emit coherent beams of Bose atoms, similarly to the emission of photon rays by light lasers, has been called atom laser \[3–10\]. Output couplers that are used for extracting condensed atoms from a trap are based on transferring atoms from a magnetically trapped state to another internal state that is not trapped \[11–14\]. The transferring field can be either a weak radiofrequency field including spin flips between trapped and untrapped states \[11–14\] or a field of a laser beam stimulating Raman transitions between magnetic sublevels . In the first case, atoms in the output state simply fall down from the trap under the action of gravity. In the second case, when the atoms are transferred to a magnetic untrapped state, they are also given a momentum kick from the photon recoil, so that they exit the trap in a well–defined beam whose direction and velocity are governed by the irradiating Raman laser.
Since the very first condition on any laser is that its output be highly directional , it is important to develop mechanisms permitting to create well–collimated beams of atoms coupled out of a trap. A novel mechanism of creating highly directional beams if atoms leaving a trap has recently been suggested and different regimes of operations have been studied \[17–20\].
This mechanism is based on transferring trapped atoms to a nonadiabatic initial state, which can be done by a short triggering pulse. There exist such initial spin states for which the motion of atoms becomes semiconfined for the same configuration of the trap magnetic field. Then atoms quit the trap flying out predominantly in one direction and forming a well–collimated beam. In this report, we give a brief outline of the semiconfining regime of motion.
## 2 Solution of Differential Equations
The main equations of motion for an atom of mass $`m`$ and magnetic moment $`\mu _0`$ are usually written in the semiclassical approximation for the quantum–mechanical average of the real–space coordinate, $`\stackrel{}{R}=\{R_\alpha \}`$, and for the average $`\stackrel{}{S}=\{S_\alpha \}`$ of the spin operator, with $`\alpha =x,y,z`$. The first equation writes
$$\frac{d^2R_\alpha }{dt^2}=\frac{\mu _0}{m}\stackrel{}{S}\frac{\stackrel{}{B}}{R_\alpha },$$
(1)
where $`\stackrel{}{B}`$ is a magnetic field and, for a while, we omit the gravitational force and a collision term whose role will be discussed later on. The equation for the average spin is
$$\frac{d\stackrel{}{S}}{dt}=\frac{\mu _0}{\mathrm{}}\stackrel{}{S}\times \stackrel{}{B}.$$
(2)
The total magnetic field $`\stackrel{}{B}=\underset{1}{\overset{}{B}}+\underset{2}{\overset{}{B}}`$ consists of two terms. The first is the quadrupole field
$$\underset{1}{\overset{}{B}}=B_1^{}(R_x\underset{x}{\overset{}{e}}+R_y\underset{y}{\overset{}{e}}+\lambda R_z\underset{z}{\overset{}{e}}),$$
(3)
which we write in a slightly more general form than in Ref. , including the anisotropy parameter $`\lambda `$. The latter was taken in Ref. in the standard form as $`\lambda =2`$. But, if several magnetic coils are involved in the formation of the field (3), the anisotropy parameter $`\lambda `$ can be varied. The possibility of such a variation will be important in what follows. The second part of the magnetic field $`\stackrel{}{B}`$ is a transverse field
$$\underset{2}{\overset{}{B}}=B_2\stackrel{}{h}(t),\stackrel{}{h}(t)=h_x\underset{x}{\overset{}{e}}+h_y\underset{y}{\overset{}{e}},$$
(4)
in which $`h_\alpha =h_\alpha (t)`$ and $`|\stackrel{}{h}(t)|=1`$. In Ref. the rotating transverse field, with
$$h_x=\mathrm{cos}\omega t,h_y=\mathrm{sin}\omega t,$$
(5)
was considered, although, as will be shown here, this is not the sole possibility. Introducing the dimensionless space variable $`\stackrel{}{r}\stackrel{}{R}/R_0=\{x,y,z\}`$, measured in units of the length $`R_0B_2/B_1^{}`$, and defining the characteristic frequencies
$$\omega _1\sqrt{\frac{\mu _0B_1^{}}{mR_0}}^{}\omega _2\frac{\mu _0B_2}{\mathrm{}},$$
(6)
we may rewrite Eq. (1) as
$$\frac{d^2\stackrel{}{r}}{dt^2}=\omega _1^2(S_x\underset{x}{\overset{}{e}}+S_y\underset{y}{\overset{}{e}}+\lambda S_z\underset{z}{\overset{}{e}})$$
(7)
and the spin equation (2) as
$$\frac{d\stackrel{}{S}}{dt}=\omega _2\widehat{A}\stackrel{}{S},$$
(8)
where the matrix $`\widehat{A}=[A_{\alpha \beta }]`$ is antisymmetric, $`A_{\alpha \beta }=A_{\beta \alpha },A_{\alpha \alpha }=0`$, and
$$A_{12}=\lambda z,A_{23}=x+h_x,A_{31}=y+h_y.$$
(9)
Keeping in mind the following discussion on the generality of the approach, we do not require here that $`h_x`$ and $`h_y`$ in Eq. (9) be necessarily of the rotating form (5). What is required is the existence of small parameters, one of which is
$$\frac{\omega _1}{\omega _2}1$$
(10)
and another one is
$$\frac{\omega }{\omega _2}1,\omega \underset{t}{\mathrm{max}}\left|\frac{d\stackrel{}{h}}{dt}\right|.$$
(11)
In the particular case of the rotating field (5), the value of $`\omega `$ in Eq. (11) coincides with the rotation frequency.
It is worth noting that the inequalities (10) and (11) are assumed from the very beginning. Numerical estimates of Ref. give $`\omega _1/\omega _210^5`$ and $`\omega /\omega _210^3`$. The existence of the small parameters (10) and (11) allows us to apply for solving the system of equations (7) and (8) averaging methods \[21–23\] as developed for multifrequency systems . A generalization of averaging methods, called the scale separation approach , has been developed for treating statistical systems that can be described by stochastic differential equations. However, while equations (7) and (8) contain no stochastic terms, the mathematical foundation of solving them is directly based on the averaging methods \[21–25\].
According to the small parameters (10) and (11), the functional variables $`\stackrel{}{r}`$ and $`\stackrel{}{h}`$ are to be treated as slow compared to the fast variable $`\stackrel{}{S}`$. This means that $`\stackrel{}{r}`$ and $`\stackrel{}{h}`$ change very little, being almost constant, during the oscillation time of $`\stackrel{}{S}`$. Therefore, $`\stackrel{}{r}`$ and $`\stackrel{}{h}`$ can be regarded as quasi–invariants with respect to $`\stackrel{}{S}`$. Such quasi–invariants in the theory of classical Hamiltonian systems are called adiabatic invariants , whose existence should not be confused with the adiabatic approximation. Following the averaging methods \[21–25\], we can solve equation (8) for the fast spin variable with the slow variables $`\stackrel{}{r}`$ and $`\stackrel{}{h}`$ treated as quasi–invariants. The difference between the latter is that $`\stackrel{}{r}`$, as a function of time, is defined implicitly by Eq. (7), while $`\stackrel{}{h}`$ is an explicitly given function. Of course, this difference is merely technical but not principal, since both $`\stackrel{}{r}`$ and $`\stackrel{}{h}`$, according to inequalities (10) and (11), are slow variables.
The matrix $`\widehat{A}`$ defined in Eq. (9) depends on time only through $`\stackrel{}{r}`$ and $`\stackrel{}{h}`$. This permits us, introducing the small parameter
$$\epsilon sup\{\frac{\omega _1}{\omega _2},\frac{\omega }{\omega _2}\}1,$$
(12)
to write the evolution equation
$$\frac{d\widehat{A}}{dt}=\epsilon \widehat{C},$$
(13)
in which $`\widehat{C}`$ is a matrix with bounded elements for any $`\epsilon <1`$ and all $`t0`$. Consequently, the matrix $`\widehat{A}`$ has also to be treated as a slow matrix variable. This allows us to look for a particular solution of the spin equation (8) in the form
$$\underset{i}{\overset{}{S}}(t)=\underset{i}{\overset{}{b}}(t)\mathrm{exp}\left\{\phi _i(t)\right\},$$
(14)
in which $`\underset{i}{\overset{}{b}}`$, with $`i=1,2,3`$, are the eigenvectors of the matrix $`\widehat{A}`$, given by the eigenproblem $`\widehat{A}\underset{i}{\overset{}{b}}=\alpha _i\underset{i}{\overset{}{b}}`$. The nice antisymmetric structure of $`\widehat{A}`$ makes it possible to easily organize that the set $`\{\underset{i}{\overset{}{b}}\}`$ would form an orthonormal basis. Substituting expression (14) into Eq. (8), we find
$$\phi _i(t)=_0^t(\omega _2\alpha _i\underset{i}{\overset{}{\stackrel{}{b}}}\dot{\stackrel{}{b}}_i)dt,$$
(15)
as in Ref. . As is clear, the form (14) is not, strictly speaking, an exact solution of Eq. (8), but this is what in the theory of differential equations called an asymptotically exact solution with respect to the small parameter $`\epsilon `$, in the sense that this solution becomes exact when $`\epsilon 0`$. The latter is evident from Eq. (13), since, when $`\epsilon 0`$, the matrix $`\widehat{A}`$ becomes a matrix with constant coefficients.
In the guiding center approach , which is a variant of the averaging methods \[21–25\], the solution of type (14) is called the guiding center because it describes the leading behaviour of an exact solution. Corrections to the guiding center can be obtained by presenting the exact solution as an expansion around this guiding center. Such expansions, because the guiding center itself is generally a function of the small parameter, are called the generalized asymptotic expansions. These have been introduced and studied by Lindstedt and Poincaré and are actively used in averaging methods \[24–26\]. Corrections to the guiding center describe small fastly oscillating ripples of higher harmonics and with the amplitudes of the orders of increasing power of the small parameter. As is well known \[29–31\], the construction of generalized asymptotic series is not unique, being based on an incomplete expansion in powers of the small parameter. For instance, one can separately expand amplitudes and phases, as in the Lindstedt–Poincaré technique \[28–31\]. In such incomplete expansions, one may keep different number of terms in the amplitudes and phases, which is dictated by the properties of the particular equation. For example, this can be necessitated by cancelling secular terms or by considering more accurately the phase that, entering in an exponential, can influence the solution stronger than corrections to the amplitude. In our case, the nonuniqueness of defining the guiding center (14) could result in the possibility of writing, instead of the phase (15), the phase
$$\phi _i(t)=\omega _2_0^t\alpha _i(t)𝑑t.$$
(16)
This is admissible, since the second term in the integrand of Eq. (15) is small as compared to the first one. The sole doubt could be whether it is possible to omit the second term in Eq. (15) if one of the eigenvalues $`\alpha _i`$ is zero. In our case it is really so for $`\alpha _3=0`$. However, again owing to the nice antisymmetry property of $`\widehat{A}`$, we can show that $`\underset{1}{\overset{}{\stackrel{}{b}}}=\underset{2}{\overset{}{b}}`$ and $`\underset{3}{\overset{}{\stackrel{}{b}}}=\underset{3}{\overset{}{b}}`$. Then, because of the normalization condition $`|\underset{3}{\overset{}{b}}|^2=\underset{3}{\overset{2}{\stackrel{}{b}}}=1`$, we have $`2\underset{3}{\overset{}{\stackrel{}{b}}}\dot{\stackrel{}{b}}_3=d|\underset{3}{\overset{}{b}}|^2/dt=0`$. Thus, $`\phi _3(t)=0`$ exactly. Which of the expressions, (15) or (16), is to be used makes no difference for what follows, since at the next step of the method \[24–26\] one has to substitute the solution for the fast variable into the right–hand side of the equation for the slow variable and to average over time this right–hand side. In the course of this averaging, all fastly oscillating terms disappear, irrespectively to the fine structure of their expressions.
When speaking about asymptotically exact solutions, one should not confuse this with the asymptotic behaviour with respect to small time $`t0`$. The solutions we are talking about are asymptotic with respect to small parameters, such as (10), (11), or (12). These parameters are always small for any $`t0`$. There is also the following invariance property of Eq. (8) if $`\widehat{A}`$ is an antisymmetric matrix. Let $`[S_{ij}(t)]`$ be the fundamental matrix of solutions to Eq. (8). Then , one has
$$\mathrm{det}[S_{ij}(t)]=\mathrm{det}[S_{ij}(t_0)]\mathrm{exp}\left\{\omega _2_{t_0}^t\mathrm{Tr}\widehat{A}(t^{})𝑑t^{}\right\}$$
for any $`t_0`$. For an antisymmetric matrix, $`\mathrm{Tr}\widehat{A}=0`$. Hence, $`\mathrm{det}[S_{ij}(t)]=const`$. Thus, if $`[S_{ij}(t_0)]`$ is the fundamental matrix at some $`t=t_0`$, then the solutions $`\underset{i}{\overset{}{S}}(t)`$ are linearly independent for any $`t0`$. The total solution of Eq. (8) is the linear combination
$$\stackrel{}{S}(t)=\underset{i=1}{\overset{3}{}}a_i\underset{i}{\overset{}{S}}(t)$$
(17)
of linear independent particular solutions, with the coefficients $`a_i=\underset{0}{\overset{}{S}}\underset{i}{\overset{}{b}}(0)`$ defined by initial conditions. In our case, the asymptotically exact particular solutions are given by Eq. (14).
Let us emphasize that all the consideration is based on and justified by the averaging methods \[21–26\] requiring the existence of small parameters that permit one to classify functional variables onto fast and slow. Without such small parameters we would have quite a different story.
After the solution (17) for the fast variable is found, then following the multifrequency averaging methods \[24–26\], one has to substitute it into the equation (7) for the slow variable and to average over time this right–hand side, which gives
$$\frac{d^2\stackrel{}{r}}{dt^2}=\stackrel{}{F}\omega _1^2<S_x\underset{x}{\overset{}{e}}+S_y\underset{y}{\overset{}{e}}+\lambda S_z\underset{z}{\overset{}{e}}>.$$
(18)
Till now, the concrete form of the transverse field (4) has been of no importance, provided that condition (11) holds true. To explicitly accomplish the time averaging in Eq. (18), we need to specify the field $`\stackrel{}{h}`$. With the rotating field (5), we find
$$\stackrel{}{F}=\frac{1}{2}f_1\omega _1^2[(1+x)S_x^0+yS_y^0+\lambda zS_z^0](x\underset{x}{\overset{}{e}}+y\underset{y}{\overset{}{e}}+2\lambda ^2z\underset{z}{\overset{}{e}}),$$
where
$$f_1\left[(1+2x+x^2+y^2+\lambda ^2z^2)(1+x^2+y^2+\lambda ^2z^2)\right]^{1/2}.$$
As is shown in paper , the semiconfining regime of motion can be realized taking the initial spin polarization given by the initial conditions
$$S_x^0=S_y^0=0,S_z^0S0.$$
(19)
Such initial conditions can be prepared in several ways. The first possibility is to confine atoms in a trap, in which all atoms are polarized having their spins in the $`z`$ direction, which e.g. can be done by means of the trap of Ref. , being a quadrupole trap with a bias field along the $`z`$ axis. Then the longitudinal bias field is quickly switched off and, at the same time, a transverse field, as in Ref. , is switched on. Thus, we get the desired initial conditions. Another possibility is to prepare spin polarized atoms in one trap and quickly load them into another trap with the required field configuration. The possibility of realizing two ways of transferring atoms from one trap to another, by means of sudden transfer as opposed to adiabatic transfer, is discussed in Ref. . The third possibility could be by acting on the trapped atoms with a short pulse of strong magnetic field, polarizing atomic spins in the necessary way.
With the initial conditions (19), equation (18) reduces to the system
$$\frac{d^2x}{dt^2}=\frac{\lambda }{2}S\omega _1^2f_1zx,\frac{d^2z}{dt^2}=\lambda ^3S\omega _1^2f_1z^2,$$
(20)
the equation for the $`y`$ component having the same form as for the $`x`$ component. The system of equations (20) can be solved analytically only for the case $`|\stackrel{}{r}|1`$ when $`f_11`$, as is considered in Ref. . This consideration proves the appearance of the semiconfining regime of motion and the existence of collimation with the aspect ratio
$$\frac{x(t)}{z(t)}|tt_0|^{3/2},$$
as $`tt_0`$, where $`t_0`$ is defined by the initial conditions for the space variable. When $`|\stackrel{}{r}|`$ becomes not small, one has to resort to numerical solution which has been done in Refs. confirming the existence of the semiconfining regime and the formation of well-collimated atomic beams.
Let us stress again that asymptotic solutions obtained by the averaging methods are asymptotic with respect to small parameters and have nothing to do with short–time approximation. As follows from the theory of the averaging methods \[21–25\], the latter provide accurate solutions to differential equations for long time intervals.
## 3 Applicability to Different Fields
The semiconfining regime of motion can be realized for a wide class of magnetic fields. Not only the anisotropy parameter $`\lambda `$ in the quadrupole field (3) can be varied, but also different transverse fields (4) can be employed. In the previous section, the particular form of a transverse field has not been specified till formula (18), showing by this that the same consideration is valid for any transverse field satisfying condition (11). Now we shall show that the choice of the rotating field (5) is not compulsory and other transverse fields can be taken provided condition (11) holds true.
Let us consider the simple case of a constant transverse field
$$h_x=const,h_y=const,$$
(21)
where, similarly to Eq. (4), it is assumed that $`|\stackrel{}{h}|=1`$. Then the small parameter (11) is identically zero.
With the transverse field (21), the right–hand side of Eq. (18) becomes
$$\stackrel{}{F}=f_2\omega _1^2[(x+h_x)S_x^0+(y+h_y)S_y^0+\lambda zS_z^0][(x+h_x)\underset{x}{\overset{}{e}}+(y+h_y)\underset{y}{\overset{}{e}}+\lambda ^2z\underset{z}{\overset{}{e}}],$$
(22)
where
$$f_2\left[(x+h_x)^2+(y+h_y)^2+\lambda ^2z^2\right]^1.$$
Accepting the initial conditions (19), we come to the equations
$$\frac{d^2x}{dt^2}=\lambda S\omega _1^2f_2z(x+h_x),\frac{d^2z}{dt^2}=\lambda ^3S\omega _1^2f_2z^2.$$
(23)
One may notice that these equations are invariant with respect to the change $`\lambda \lambda `$, $`SS`$ or to the change $`SS`$, $`zz`$. Therefore, it would be sufficient to consider a case of one fixed sign for $`\lambda S`$, say, $`\lambda S>0`$. Passing to another sign of $`\lambda S<0`$ is equivalent to the inversion $`zz`$.
Equations (23) are very similar to Eqs. (20). In the same way as in Ref. , we can demonstrate that the solution to Eqs. (23) corresponds to the semiconfining regime of motion. For $`|\stackrel{}{r}|1`$, when $`f_21`$, we have
$$\frac{d^2x}{dt^2}=\lambda S\omega _1^2h_xz,\frac{d^2z}{dt^2}=\lambda ^3S\omega _1^2z^2.$$
(24)
The second equation here is the same as in Eq. (20) if $`f_11`$. The analysis shows that the motion along the $`z`$ axis is semiconfined. The collimation of the forming directed beam is characterized by the aspect ratio
$$\frac{x(t)}{z(t)}|tt_0|^2\mathrm{ln}|tt_0|,$$
(25)
as $`tt_0`$. Comparing this with the aspect ratio for the rotating field, we take into account that for $`\tau |tt_0|1`$ one has $`\tau ^{1/2}|\mathrm{ln}\tau |1`$. Therefore, the constant transverse field provides even better collimation than the rotating field.
## 4 Influence of Gravitational Force
Atoms having mass are certainly subject to the action of the gravitational force. For the adiabatic motion of atoms confined in a trap, this force is not as significant since it can always be easily compensated by an additional gradient magnetic field. But what is the role of this force in a nonadiabatic motion? The principal answer to this question is, actually, almost evident: If the gravitational force can be compensated by additional gradient fields, it will lead to no principal changes in the motion of atoms inside a trap. However, to be precise, let us turn to mathematics.
Recall first of all that the coordinate axes everywhere in our formulas have been related to the magnetic field configuration, so that the axis $`z`$ is an axis of the device, but not necessary the vertical axis related to gravity. As far as the device can be oriented arbitrarily, the gravitational force can also be directed along different axes. For instance, assume that this force is $`mg\underset{x}{\overset{}{e}}`$, where $`g10^3`$ cm s<sup>-2</sup> is the standard gravitational acceleration. Adding this force to the right–hand side of Eq. (18) results, instead of Eqs. (20), in the equations
$$\frac{d^2x}{dt^2}=\omega _1^2\left(\frac{\lambda }{2}Sf_1zx\alpha \right),\frac{d^2z}{dt^2}=\lambda ^3S\omega _1^2f_1z^2,$$
(26)
in which the notation
$$\alpha \frac{g}{R_0\omega _1^2}$$
(27)
is used. The occurrence of an additional term in the first of equations (26) will lead to the gravitational drift along the $`x`$ axis. To compensate the gravitational force, an auxiliary gradient magnetic field is to be superimposed. This is equivalent to the increase of the coefficient near $`\underset{x}{\overset{}{e}}`$ in the quadrupole field (3), which yields the scaling $`x\lambda _xx`$ in Eq. (26). One may also increase the anisotropy coefficient $`\lambda `$. The final aim is to make the ratio $`\alpha /\lambda \lambda _x`$ small, which thus reduces the influence of the gravity force. This ratio is actually small already for existing traps, without involving additional compensating fields. For instance, taking the values $`\omega _1(10^210^3)`$s<sup>-1</sup>, $`R_0(0.10.5)`$cm, $`|\lambda |=2`$, and $`\lambda _x=1`$, typical of many modern traps, we have $`\alpha /\lambda \lambda _x`$ of the order $`10^110^3`$. Increasing the parameters $`\lambda `$ and $`\lambda _x`$, by switching on additional gradient fields, the ratio $`\alpha /\lambda \lambda _x`$ can be made arbitrarily small, thus making the influence of the gravitational force negligible.
Moreover, we have an additional possibility of changing the orientation of the device with respect to the gravitational force. For instance, the latter can be directed along the $`z`$ axis, so that this force is $`mg\underset{z}{\overset{}{e}}`$. Then, instead of Eqs. (26), we get
$$\frac{d^2x}{dt^2}=\frac{\lambda }{2}S\omega _1^2f_1xz,\frac{d^2z}{dt^2}=\omega _1^2(\lambda ^3Sf_1z^2\alpha ).$$
(28)
From here it is seen that the influence of the gravitational force can be reduced by increasing the anisotropy parameter $`\lambda `$. To demonstrate this more accurately, let us consider the same case as earlier, when $`|\stackrel{}{r}|1`$. Then, integrating once the second equation in the system (28), we have
$$\left(\frac{dz}{dt}\right)^2=\frac{2}{3}\lambda ^3S\omega _1^2\left[z^3z_0^3\frac{3\alpha }{\lambda ^3S}(zz_0)+\zeta \right],$$
where
$$\zeta \frac{3\dot{z}_0^2}{2\lambda ^3S\omega _1^2}.$$
The latter equation, by means of the substitution
$$z(t)=\frac{6}{\lambda ^3S}𝒫(\tau \tau _0),\tau \omega _1t,$$
(29)
can be transformed to the Weierstrass equation
$$\left(\frac{d𝒫}{d\tau }\right)^2=4𝒫^3g_2𝒫g_3,$$
with the Weierstrass invariants
$$g_2=\frac{\alpha }{3}\lambda ^3S,g_3=\frac{(\lambda ^3S)^3}{54}\left(z_0^3\frac{3\alpha }{\lambda ^3S}z_0\zeta \right).$$
From the properties of the Weierstrass function $`𝒫(\tau )`$ it follows that the semiconfining regime of motion is realized when $`g_2^3<27g_3^2`$, which gives
$$\frac{4\alpha ^3}{\lambda ^9S^3}<\left(z_0^3\frac{3\alpha }{\lambda ^3S}z_0\zeta \right)^2.$$
(30)
As is evident form this inequality, the terms related to the gravity can be strongly reduced by increasing the anisotropy parameter $`\lambda `$, that is by increasing the field gradient in the $`z`$ direction. Even for the typical value $`|\lambda |=2`$, without involving additional compensating fields, the left–hand side of the inequality (30) is of the order $`10^3\alpha ^3`$. For $`\alpha 10^3`$ this yields $`10^{12}`$, which is very small. Increasing the parameter $`\lambda `$ can practically completely eliminate the influence of gravity. Note also that for the case $`\lambda S<0`$, the inequality (30) is always valid since its left–hand side is negative while the right–hand side is nonnegative. The latter case shows that the gravitational force even may help to realize semiconfinement, which happens when the direction of this force coincides with that of the atomic escape. In this way, it is not difficult to prepare such magnetic fields and to choose the orientation of the device so that the gravitational force would not essentially disturb the semiconfining regime of motion.
## 5 Role of Random Collisions
In order to investigate the role of atomic collisions, we have to compliment the right–hand side of the evolution equation (18) by an additional term describing these collisions, so that we can write
$$\frac{d^2\stackrel{}{r}}{dt^2}=\stackrel{}{F}+\gamma \stackrel{}{\xi },$$
(31)
where $`\stackrel{}{F}`$ is the same expression as in Eq. (18), $`\gamma `$ is a collision rate, and $`\stackrel{}{\xi }`$ is a vector whose properties are defined by the details of the pair collisions. Two principally different situations can exist. One is when the effective collision force $`\gamma \stackrel{}{\xi }`$ is larger than or comparable with $`\stackrel{}{F}`$. Then it is clear that the atomic motion is essentially governed by the collision force whose particular properties become important. However, this case is of no interest for us since, as is evident, in the presence of strong and frequent collisions no directed semiconfined motion can occur. Any organized directed motion will be strongly suppressed by disorganizing random collisions. The second situation is when $`\gamma \stackrel{}{\xi }`$ is much weaker than $`\stackrel{}{F}`$. And solely this case is of interest since only then an organized unidirectional motion can arise. But when the force $`\gamma \stackrel{}{\xi }`$ representing random pair collisions is just a weak perturbation, then the details of this force are not of great importance and it can be modelled by the stochastic white–noise variable. It is this approach that was employed in Ref. , where some additional simplifications for $`\stackrel{}{\xi }`$ were assumed. These simplifications are not principal, and the consideration can be easily generalized to the case of an anisotropic random vector $`\stackrel{}{\xi }=\{\xi _\mu \}`$, where $`\mu =x,y,z`$. In the general anisotropic case, the stochastic properties of the set $`\{\xi _\mu \}`$ of white–noise random variables is characterized by the stochastic averages
$$\xi _\mu (t)=0(\mu =x,y,z),\xi _\mu (t)\xi _\nu (t^{})=2D_\mu \delta _{\mu \nu }\delta (tt^{}),$$
(32)
in which $`D_\mu `$ is a diffusion rate in the $`\mu `$ direction.
Adding the anisotropic random force to Eq. (20), for $`|\stackrel{}{r}|1`$, we have
$$\frac{d^2x}{dt^2}=\frac{\lambda }{2}S\omega _1^2zx+\gamma \xi _x,\frac{d^2z}{dt^2}=\lambda ^3S\omega _1^2z^2+\gamma \xi _z.$$
(33)
As far as only the case of small disturbance of the semiconfining regime of motion is of our concern, we can solve Eqs. (33) by perturbation theory writing the solutions as
$$x=x_1+x_2,z=z_1+z_2,$$
(34)
where $`x_1`$ and $`z_1`$ are the solutions of the unperturbed equations (20) and $`x_2`$ and $`z_2`$ are the solutions to the equations
$$\frac{d^2x_2}{dt^2}=\frac{\lambda }{2}S\omega _1^2(z_1x_2+x_1z_2)+\gamma \xi _x,\frac{d^2z_2}{dt^2}=2\lambda ^3S\omega _1^2z_1z_2+\gamma \xi _z.$$
(35)
From here, similarly to the way of Ref. , we get
$$x_2(t)=_0^tG_x(tt^{})\left[\gamma \xi _x(t^{})+\frac{\lambda }{2}S\omega _1^2x_1z_2(t^{})\right]𝑑t^{},$$
$$z_2(t)=_0^tG_z(tt^{})\gamma \xi _z(t^{})𝑑t^{},$$
(36)
where
$$G_x(t)=\frac{\mathrm{sinh}(\beta t)}{\beta },G_z(t)=\frac{\mathrm{sinh}(\beta _zt)}{\beta _z},\beta =\left(\frac{\lambda }{2}S\omega _1^2z_1\right)^{1/2},\beta _z=2|\lambda |\beta .$$
Because of Eqs. (32), we have $`x_2=z_2=0`$. And for the mean square deviations, we obtain
$$x^2=\frac{\gamma ^2D_xt}{\beta ^2}[\frac{\mathrm{sinh}(2\beta t)}{2\beta t}1]+\frac{\beta ^4x_1^2\gamma ^2D_zt}{\beta _z^2(\beta _z^2\beta ^2)^2z_1^2}\times $$
$$\times \left\{\frac{\mathrm{sinh}(\beta _zt)}{\beta _zt}\left[\mathrm{cosh}(\beta _zt)\mathrm{cosh}(\beta t)\right]\frac{\beta _z}{\beta }\mathrm{sinh}(\beta _zt)\mathrm{sinh}(\beta t)+\mathrm{cosh}(\beta _zt)\mathrm{cosh}(\beta t)1\right\},$$
(37)
$$z^2=\frac{\gamma ^2D_zt}{\beta _z^2}[\frac{\mathrm{sinh}(2\beta _zt)}{2\beta _zt}1].$$
(38)
These formulas generalize the result of Ref. . However the main aim in considering the role of random collisions is not just to derive formulas as above but rather to find conditions under which these collisions would not essentially disturb the semiconfined motion. Such a condition can be written as
$$\frac{\gamma ^2D}{\omega _1^3}1,Dsup\{D_x,D_y,D_z\}.$$
(39)
For estimates, we can take the collision rate as $`\gamma \mathrm{}\rho a_s/m`$, where $`\rho `$ is the density of atoms and $`a_s`$, scattering length, and for the diffusion rate we may write $`Dk_BT/\mathrm{}`$, where $`T`$ is temperature. Then condition (39) yields
$$(\rho a_s^3)^2k_BT\left(\frac{\mathrm{}^2}{ma_s^2}\right)^2(\mathrm{}\omega _1)^3.$$
(40)
This shows that random atomic collisions will not disturb much the organized semiconfined motion if density and temperature are small enough to satisfy condition (40).
Acknoledgements
We are grateful for discussions to V.S. Bagnato. Financial support from the São Paulo State Research Foundation is appreciated.
|
warning/0004/hep-th0004162.html
|
ar5iv
|
text
|
# I INTRODUCTION
## I INTRODUCTION
It is by now well-known that when particles of conventional statistics are coupled to pure Chern-Simons (CS) gauge field, this field creates an Aharonov-Bohm (AB) like interaction which converts the particles to charge-flux tube composites . Somewhat later, it was shown in the context of a Galilean field theory of scalar fields minimally coupled to a pure CS field that, one can approach the problem of calculating an arbitrary scattering process by restricting consideration to an N-body sector, allowing one to derive a Schroedinger equation for N-body problem with each pair interacting as zero radius flux tubes. This has led to the claim that the field theory, sector by sector, is formally equivalent to a conventional Schroedinger equation . Specifically in two particle sector of this equivalent field theory, one gets a Schroedinger equation similar to the AB equation .
These developments brought back the long-standing issue of failure of the quantum mechanical perturbation theory for the AB scattering amplitude . The failure of the Born approximation, for instance, is known to be due to the fact that the lowest partial wave amplitude satisfies an integral equation whose interaction term is quadratic in terms of the flux parameter. As the exact lowest partial wave contribution to the scattering amplitude is known to be linear in this parameter, then, it is absent in the first order Born approximation.
There have been several attempts to solve this problem for the spinless case in the context of direct AB scattering , anyon physics , and scalar Galilean CS gauge field theory . For instance, in Ref.7, it was shown, through a perturbative calculation of the two-particle scattering amplitude, up to one-loop order, that this amplitude is non-renormalizable, unless a contact interaction is introduced, which however for a given strength of the interaction (critical value corresponding to the self-dual limit) reduces to the same order term of the series expansion of the exact quantum mechanical amplitude. The same procedure is generalized to the non-Abelian case with similar conclusions in the second work of Ref. 7. One should note that before the introduction of the contact interaction, the failure of the naive perturbation expansion of the Galilean CS field theory (namely the first Born term for s-wave is wrong, while the second Born term is infinite) is very similar to that of the Born series in direct AB scattering. As the exact AB amplitude posseses scale invariance, it is natural that the agreement is obtained only after the introduction of contact interaction whose coupling strenght has the critical value for which scale invariance is restored. This scale invariance at the critical coupling is explicitely checked up to three loops in Ref.8, and up to all orders in Ref.9.
The Born approximation problem for the spinless case was adressed from a more general point of view in Ref. 10 and Ref.11 questioning whether the exact (non-perturbative) quantum mechanical AB amplitude can be reproduced order-by-order perturbatively in the framework of scalar Galilean CS field theory. Ref. 10 concludes that the full agreement is obtained if the renormalized strength of the contact interaction is chosen to be related to the self-adjoint extension parameter, for fixed renormalization scale. However, the conclusion of more recent work is not in full agreement with that of the Ref.10. They show that the full agreement can be obtained only in some special regimes. Thus, we see that the general problem in the context of Galilean scalar field theory is not satisfactorily settled yet.
In Ref.12, it was shown that if one starts from the relativistic scalar gauge field theory of the CS interaction, one finds a renormalizable one-loop scattering amplitude which remains so in the non-relativistic limit as well, thus reproducing the correct series expansion of the exact quantum mechanical expression without the need to introduce a contact interaction term. It is not clear yet whether the issue raised in Ref.10 and Ref.11 would be relevant for the relativistic field theories. Obviously there are some fundamental differences between the non-relativistic and the relativistic cases. For instance, in the non-relativistic case, the necessity of a cut-off is not a relic of some unknown ultraviolet physics, but rather an artefact of the perturbative methods used. This is in contrast with the conventional wisdom on renormalization, whose natural habitat is the relativistic field theories.
AB scattering of spin-$`\frac{1}{2}`$ particles from an infinitely long solenoid was considered in the context of Dirac equation formalism in Ref. 13, and using covariant perturbation theory in Ref. 14. In these works, it was shown that that Born approximation works, that is, it agrees with the corresponding term in the series expansion of the exact amplitude. The agreement obtained in the framework of Dirac equation is not surprising at all. Because the Dirac Hamiltonian is linear in momenta, and the corresponding integral equation determining the lowest partial wave amplitude involves a term linear in flux parameter.
The spin-$`\frac{1}{2}`$ AB problem was also considered in the framework of equivalent Galilean CS gauge field theory in Ref. 15 and Ref. 16 from different perspectives. In these works the consistency of the perturbative treatment was checked up to one-loop order. As the exact amplitude is proportional to $`\mathrm{sin}\pi \alpha `$ (with $`\alpha =\frac{e\varphi }{2\pi }`$, and $`\varphi `$ is the magnetic flux carried by the solenoid), the series expansion of this term contains terms of order O($`\alpha `$), O($`\alpha ^3`$),…; that is O($`\alpha ^2`$) is missing. Thus a complete check of the consistency of perturbative approach, not only should get agreement on O($`\alpha `$) terms, but also should show the vanishing of O($`\alpha ^2`$) terms (1-loop terms in the language of the field theory).
In Ref. 15, it was shown that the two-particle sector of the Galilean field theory again leads to an AB-like equation. Then, the two particle scattering amplitude is computed up to 1-loop order. The tree-contribution (O($`\alpha `$)) agrees with the corresponding term in the series expansion of the exact amplitude; the 1-loop contribution (O($`\alpha ^2`$)) is finite and vanishes. This completes the check of consistency of the Born approximation to lowest order, in the (sector by sector) equivalent field theory framework.
Encouraged by the results of Ref. 15 and Ref. 16 in the Galilean field theory framework, it is aimed in this work to carry out the second order analysis in direct version of the problem, namely the relativistic scattering of spin-$`\frac{1}{2}`$ particles from an infinitely long solenoid, and check the consistency of the Born approximation fully, by demonstrating that O($`\alpha ^2`$) contribution to the scattering amplitude in the framework of covariant perturbation theory vanishes. We will show that this is indeed what happens.
We would like to note that the subtleties pointed out in Ref.10 and Ref.11 for the spinless case are natually overcome in the relativistic case considered in this work. This does not create any difficulty in establishing parallelism with the results obtained in Ref.15 in the context of Galilean CS field theory. Because it was already shown in Ref.15 that, in contrast with the crucial role played by the contact interaction in the scalar case, the contribution of the Pauli term formally corresponding to the contact interaction (produced in the non-relativistic limit of the fermionic CS gauge field theory with given coupling strength) to 1-loop diagrams are finite and null, thanks to the statistics.
This paper is organized as follows: In Sect $`2`$, we briefly review the results of Ref. 14 for the general discussion of the Helicity conservation. In Sect $`3`$, we review the covariant perturbation theory approach to lowest order for the problem under consideration. In Sect $`4`$, the O($`\alpha ^2`$) contribution to the scattering amplitude is computed; and it is shown that this contribution vanishes. Sect $`5`$ is devoted to the discussion of the results.
## II HELICITY CONSERVATION AND THE EXACT SCATTERING AMPLITUDE
The basic starting point of Ref. 14 is the well-known observation that the helicity of a spin-$`\frac{1}{2}`$ particle is unchanged by a time-independent magnetic field .
Defining the Helicity eigenstates in the initial and final states as $`|\pm _{i,f}`$ , the first observation is that $`|\pm _i|\pm _f`$ transitions proceed with unit probability in the Helicity space. Denoting the scattering matrix by S this reads as
$`|_f\pm |S|\pm _i|^2=1`$
$`|_f\pm |S|_i|^2=0`$ (1)
Thus the differential cross section for $`|\pm _i|\pm _f`$ per unit length is determined by the phase space only, and thus equal to the unpolarized cross section:
We next consider the scattering from an initial state polarized along the direction of an arbitrary unit vector $`\widehat{n}`$ to a final state moving along $`\theta `$, in which the beam is polarized again in the same $`\widehat{n}`$ direction. Denoting the spherical angles of $`\widehat{n}`$ with respect to the initial beam axis (chosen as x-axis) by ($`\theta ^{},\phi ^{}`$ ) these states are given as
$`|i(\stackrel{}{p}_i,\widehat{n})`$ $`=`$ $`\mathrm{cos}{\displaystyle \frac{\theta ^{}}{2}}e^{\frac{i\phi ^{}}{2}}|+_i+\mathrm{sin}{\displaystyle \frac{\theta ^{}}{2}}e^{\frac{i\phi ^{}}{2}}|_i`$
$`|f(\stackrel{}{p}_f,\widehat{n})`$ $`=`$ $`(\mathrm{cos}{\displaystyle \frac{\theta }{2}}\mathrm{cos}{\displaystyle \frac{\theta ^{}}{2}}e^{\frac{i\phi ^{}}{2}}+\mathrm{sin}{\displaystyle \frac{\theta }{2}}\mathrm{sin}{\displaystyle \frac{\theta ^{}}{2}}e^{\frac{i\phi ^{}}{2}}|+_f`$ (2)
$`+(\mathrm{cos}{\displaystyle \frac{\theta }{2}}\mathrm{sin}{\displaystyle \frac{\theta ^{}}{2}}e^{\frac{i\phi ^{}}{2}}\mathrm{sin}{\displaystyle \frac{\theta }{2}}\mathrm{cos}{\displaystyle \frac{\theta ^{}}{2}}e^{\frac{i\phi ^{}}{2}}|_f`$
Using equations (2), one readily gets
$`f|S|i=\mathrm{cos}{\displaystyle \frac{\theta }{2}}i\mathrm{sin}{\displaystyle \frac{\theta }{2}}\mathrm{sin}\theta ^{}\mathrm{sin}\phi ^{}`$ (3)
Thus the polarized cross section per unit length of the solenoid is obtained as
$`{\displaystyle \frac{d\sigma }{d\theta }}=[1(\widehat{n}\times \widehat{z})^2\mathrm{sin}^2{\displaystyle \frac{\theta }{2}}]({\displaystyle \frac{d\sigma }{d\theta }})_{unpol}`$ (4)
Here $`\widehat{z}`$ is the unit vector in the direction of the solenoid. Thus the cross section differs from the unpolarized case (or the spinless case) when the spin of the particle has components in the scattering plane(chosen here as x-y plane).
## III COVARIANT PERTURBATION THEORY; FIRST ORDER (BORN APPROXIMATION)
The purpose of this section is to show that Born approximation reproduces the correct result, that is it agrees with the corresponding terms in the series expansion of the exact amplitude.
The S-matrix element for a spin-$`\frac{1}{2}`$ particle scattering from an external electromagnetic field to lowest order is given by:
$`S_{fi}^{(1)}={\displaystyle d^4z\overline{\psi }_f(z)(ie\gamma _\mu A^\mu (z))\psi _i(z)}`$ (5)
where in the Bjorken-Drell convention
$`\psi _i(z)`$ $`=`$ $`\sqrt{{\displaystyle \frac{m}{E_iV}}}u(p_i,s_i)e^{ip_{i\mu }z^\mu }`$
$`\psi _f(z)`$ $`=`$ $`\sqrt{{\displaystyle \frac{m}{E_fV}}}u(p_f,s_f)e^{ip_{f\mu }z^\mu }`$ (6)
The vector potential of the solenoid, taken along the 3rd axis, in the Coulomb gauge $`\stackrel{}{}\stackrel{}{A}=0`$ is given as
$`A_1(z)`$ $`=`$ $`{\displaystyle \frac{\varphi }{2\pi }}{\displaystyle \frac{z_2}{z_1^2+z_2^2}}`$
$`A_2(z)`$ $`=`$ $`{\displaystyle \frac{\varphi }{2\pi }}{\displaystyle \frac{z_1}{z_1^2+z_2^2}}`$
$`A_3(z)`$ $`=`$ $`A_0(z)=0`$ (7)
where $`\varphi `$ is the magnetic flux carried by the solenoid. Denoting $`\stackrel{}{q}=\stackrel{}{p_f}\stackrel{}{p_i}`$, and carrying out the z- integrals, we find
$`S_{fi}^{(1)}={\displaystyle \frac{4\pi ^2}{V}}(me\varphi )\delta (E_fE_i)\delta (p_{f3}p_{i3}){\displaystyle \frac{\overline{u}(f)(\gamma ^2q_1\gamma ^1q_2)u(i)}{\sqrt{E_fE_i}(q_1^2+q_2^2)}}`$ (8)
As the initial beam is in the 1st direction ($`p_{i3}=0`$), denoting $`t=\overline{u}(f)(\gamma ^2q_1\gamma ^1q_2)u(i)`$, the differential cross section per unit solenoid length, to this order, is given as
$`({\displaystyle \frac{d\sigma }{d\theta }})^{Born}={\displaystyle \frac{m^2e^2\varphi ^2}{2\pi |\stackrel{}{p_i}|(q_1^2+q_2^2)^2}}|t|^2`$ (9)
with $`|\stackrel{}{p}_i|=|\stackrel{}{p}_f|=k`$ and $`E_i=E_f`$, as imposed by the $`\delta `$-functions. We can proceed in two ways: a)We can sum over final polarizations, and average over the initial ones to get the unpolarized cross section by direct use of Dirac matrix algebra
$`({\displaystyle \frac{d\sigma }{d\theta }})^{Born}={\displaystyle \frac{e^2\varphi ^2}{8\pi k\mathrm{sin}^2\frac{\theta }{2}}}`$ (10)
where $`\stackrel{}{p_i}=k\widehat{x}`$. b)We can compute the polarized amplitude, using the explicit expressions of the Dirac spinors for the polarized initial and final electrons.
$`u(i)`$ $`=`$ $`\mathrm{cos}{\displaystyle \frac{\theta ^{}}{2}}e^{\frac{i\phi ^{}}{2}}u_+(i)+\mathrm{sin}{\displaystyle \frac{\theta ^{}}{2}}e^{\frac{i\phi ^{}}{2}}u_{}(i)`$
$`u(f)`$ $`=`$ $`(\mathrm{cos}{\displaystyle \frac{\theta }{2}}\mathrm{cos}{\displaystyle \frac{\theta ^{}}{2}}e^{\frac{i\phi ^{}}{2}}+\mathrm{sin}{\displaystyle \frac{\theta }{2}}\mathrm{sin}{\displaystyle \frac{\theta ^{}}{2}}e^{\frac{i\phi ^{}}{2}})u_+(f)`$ (11)
$`+(\mathrm{cos}{\displaystyle \frac{\theta }{2}}\mathrm{sin}{\displaystyle \frac{\theta ^{}}{2}}e^{\frac{i\phi ^{}}{2}}\mathrm{sin}{\displaystyle \frac{\theta }{2}}\mathrm{cos}{\displaystyle \frac{\theta ^{}}{2}}e^{\frac{i\phi ^{}}{2}})u_{}(f)`$
where
$`u_+(i)`$ $`=`$ $`N_i\left(\begin{array}{c}1\\ 0\\ \mu _i\\ 0\end{array}\right),u_{}(i)=N_i\left(\begin{array}{c}0\\ 1\\ 0\\ \mu _i\end{array}\right)`$ (20)
$`N_i`$ $`=`$ $`\sqrt{{\displaystyle \frac{E_i+m}{2m}}},\mu _i={\displaystyle \frac{|\stackrel{}{p}_i|}{E_i+m}}`$
$`u_+(f)`$ $`=`$ $`N_f\left(\begin{array}{c}cos\frac{\theta }{2}\\ sin\frac{\theta }{2}\\ \mu _fcos\frac{\theta }{2}\\ \mu _fsin\frac{\theta }{2}\end{array}\right),u_{}(f)=N_f\left(\begin{array}{c}sin\frac{\theta }{2}\\ cos\frac{\theta }{2}\\ \mu _fsin\frac{\theta }{2}\\ \mu _fcos\frac{\theta }{2}\end{array}\right)`$ (29)
$`N_f`$ $`=`$ $`\sqrt{{\displaystyle \frac{E_f+m}{2m}}},\mu _f={\displaystyle \frac{|\stackrel{}{p_f}|}{E_f+m}}`$ (30)
Using (11), (30), we can compute t, and find
$`t={\displaystyle \frac{2k^2}{m}}\mathrm{sin}{\displaystyle \frac{\theta }{2}}[\mathrm{cos}{\displaystyle \frac{\theta }{2}}i\mathrm{sin}{\displaystyle \frac{\theta }{2}}\mathrm{sin}\theta ^{}\mathrm{sin}\phi ^{}]`$ (31)
Substituting (31) in (9), we get
$`({\displaystyle \frac{d\sigma }{d\theta }})_{pol}^{Born}=({\displaystyle \frac{d\sigma }{d\theta }})_{unpol}^{Born}[1(\widehat{n}\times \widehat{z})^2\mathrm{sin}^2{\displaystyle \frac{\theta }{2}}]`$ (32)
Thus, Born approximation indeed works in the polarized case. The scattering amplitude (and thus the cross section) is effected by the same expression in the Born approximation as in case of the exact amplitude. However this does not constitute a complete check of the consistency of the Born approximation in the relativistic spin-$`\frac{1}{2}`$ AB effect yet. As the exact amplitude is proportional to $`\mathrm{sin}\pi \alpha `$, a full consistency would require that the $`O(\alpha ^2)`$ contribution to the scattering amplitude should vanish; and this is what we will check next.
## IV COVARIANT PERTURBATION THEORY- SECOND ORDER
The S-matrix in the second order is given as
$`S_{fi}^{(2)}={\displaystyle d^4xd^4y\overline{\psi _f}(x)(ie\gamma ^\mu A_\mu (x))iS_F(xy)(ie\gamma ^\nu A_\nu (y))\psi _i(y)}`$ (33)
where $`S_F(xy)=\frac{d^4p}{(2\pi )^4}e^{ip(xy)}\frac{\gamma ^\mu p_\mu +m}{p^2m^2+i\epsilon }.`$ Carrying out the spatial integrals we get
$`S_{fi}^{(2)}`$ $`=`$ $`{\displaystyle \frac{i}{V}}(e^2\varphi ^2){\displaystyle \frac{m}{\sqrt{E_iE_f}}}\delta (E_fE_i)\delta (p_{f3}p_{i3})I`$
$`I`$ $`=`$ $`{\displaystyle d^2p_{}\frac{N}{(\stackrel{}{p_f_{}}^2\stackrel{}{p_{}}^2)(\stackrel{}{p_fp})_{}^2(\stackrel{}{p_ip})_{}^2}}`$ (34)
where N is obtained as
$`N`$ $`=`$ $`(p_ip)_2(p_fp)_1\overline{u}_f\gamma ^1P\gamma ^3u_i+(p_ip)_1(p_fp)_2\overline{u}_f\gamma ^3P\gamma ^1u_i`$ (35)
$`(p_ip)_1(p_fp)_1\overline{u}_f\gamma ^1P\gamma ^1u_i(p_ip)_2(p_fp)_2\overline{u}_f\gamma ^3P\gamma ^3u_i`$
with
$`P=\gamma ^0E_f\gamma ^3p_1\gamma ^1p_2+m`$ (36)
Denoting the polar angle in the $`p_{}`$ plane by $`\phi `$, and making use of the energy conservation mandated by $`\delta (E_fE_i),\stackrel{}{p_i}^2=\stackrel{}{p_f}^2k^2`$ with $`\stackrel{}{p}_i=k\widehat{x}`$, then N can be written as
$`N`$ $`=`$ $`\alpha +\beta \mathrm{cos}\phi +\gamma \mathrm{sin}\phi `$
$`\alpha `$ $`=`$ $`k^3\{{\displaystyle \frac{E_i}{k}}\{A\mathrm{sin}\theta B\mathrm{cos}\theta \}u^2\{{\displaystyle \frac{E_i}{k}}B+D\mathrm{sin}\theta +C(1+\mathrm{cos}\theta )\}`$
$`+{\displaystyle \frac{m}{k}}\{A^{}\mathrm{sin}\theta +B^{}\mathrm{cos}\theta +B^{}u^2\}\}`$
$`\beta `$ $`=`$ $`k^3\{Cu^3+u\{D\mathrm{sin}\theta +C\mathrm{cos}\theta +{\displaystyle \frac{E_i}{k}}\{(1+\mathrm{cos}\theta )BA\mathrm{sin}\theta )\}\}`$
$`{\displaystyle \frac{mu}{k}}\{A^{}\mathrm{sin}\theta +B^{}(1+\mathrm{cos}\theta )\}\}`$
$`\gamma `$ $`=`$ $`k^3\{Du^3+u\{C\mathrm{sin}\theta D\mathrm{cos}\theta {\displaystyle \frac{E_i}{k}}\{A(1\mathrm{cos}\theta )B\mathrm{sin}\theta \}\}`$ (37)
$`{\displaystyle \frac{mu}{k}}\{A^{}(1\mathrm{cos}\theta )+B^{}\mathrm{sin}\theta \}\}`$
with $`u\frac{p}{k}`$ and
$`A`$ $`=`$ $`i\overline{u}_f\gamma ^0\mathrm{\Sigma }_2u_i,A^{}=i\overline{u}_f\mathrm{\Sigma }_2u_i,\mathrm{with}\Sigma _2=\left(\begin{array}{cc}\sigma _2& 0\\ 0& \sigma _2\end{array}\right)`$ (40)
$`B`$ $`=`$ $`\overline{u}_f\gamma ^0u_i,B^{}=\overline{u}_fu_i`$
$`C`$ $`=`$ $`\overline{u}_f\gamma ^3u_i,D=\overline{u}_f\gamma ^1u_i`$ (41)
The $`\phi `$ integration can be carried out using the complex integration techniques. That is we define $`z=e^{i\phi }`$, and the $`\phi `$ integration is converted into a contour integration over the unit circle $`|z|=1`$. Thus
$`I={\displaystyle \frac{e^{i\theta }}{2ik}}{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{du}{u(1u^2)}}{\displaystyle _{|z|=1}}𝑑z{\displaystyle \frac{(z,\overline{z})}{(z^2+12az)(z^2+e^{2i\theta }2aze^{i\theta })}}`$ (42)
with $`a=\frac{u^2+1}{2u}`$ and
$`(z,\overline{z})=c_0+c_1z+c_2z^2`$ (43)
where
$`c_0`$ $`=`$ $`\{C+iD\}u^3+u\{(CiD)e^{i\theta }+{\displaystyle \frac{E_i}{k}}(BiA+(B+iA)e^{i\theta })`$
$`{\displaystyle \frac{m}{k}}(B^{}+iA^{}+(B^{}iA^{})e^{i\theta })\}`$
$`c_1`$ $`=`$ $`{\displaystyle \frac{2E_i}{k}}\{A\mathrm{sin}\theta B\mathrm{cos}\theta \}2u^2\{{\displaystyle \frac{E_i}{k}}B{\displaystyle \frac{m}{k}}B^{}+D\mathrm{sin}\theta +C(1+\mathrm{cos}\theta )\}`$
$`+2{\displaystyle \frac{m}{k}}\{A^{}\mathrm{sin}\theta +B^{}\mathrm{cos}\theta \}`$
$`c_2`$ $`=`$ $`\{CiD\}u^3+u\{(C+iD)e^{i\theta }+{\displaystyle \frac{E_i}{k}}(B+iA+(BiA)e^{i\theta })`$ (44)
$`{\displaystyle \frac{m}{k}}(B^{}iA^{}+(B^{}+iA^{})e^{i\theta })\}`$
The z-integral now can be carried out, using the Cauchy theorem, and we get
$`J`$ $`=`$ $`2\pi ie^{i\theta }{\displaystyle \frac{2u^2}{k}}`$ (45)
$`\times {\displaystyle \frac{\{(E_iBmB^{})u^2+E_i(A\mathrm{sin}\theta B\mathrm{cos}\theta )+m(A^{}\mathrm{sin}\theta +B^{}\mathrm{cos}\theta )\}}{(u^2e^{i\theta })(u^2e^{i\theta })}}\epsilon (u1)`$
Substituting (45) in (42) , we get
$`I`$ $`=`$ $`{\displaystyle \frac{2\pi }{k^2}}{\displaystyle _0^{\mathrm{}}}u𝑑u\epsilon (u1)`$ (46)
$`\times {\displaystyle \frac{\{(E_iBmB^{})u^2+E_i(A\mathrm{sin}\theta B\mathrm{cos}\theta )+m(A^{}\mathrm{sin}\theta +B^{}\mathrm{cos}\theta )\}}{(u^21)(u^2e^{i\theta })(u^2e^{i\theta })}}`$
Changing variables, $`u^2=v`$, (46) could be rewritten as
$`I`$ $`=`$ $`{\displaystyle \frac{\pi }{k^2}}{\displaystyle _0^{\mathrm{}}}𝑑v\epsilon (v1)`$ (47)
$`\times \{{\displaystyle \frac{E_iBmB^{}}{(ve^{i\theta })(ve^{i\theta })}}+{\displaystyle \frac{(E_iA+mA^{})\mathrm{sin}\theta +(E_iBmB^{})(1\mathrm{cos}\theta )}{(v1)(ve^{i\theta })(ve^{i\theta })}}\}`$
The first integral in (47), can easily be shown to vanish, with the help of a variable change $`v=\frac{1}{w}`$ in the $`(1,\mathrm{})`$ interval. Thus, we finally end up with
$`I={\displaystyle \frac{\pi T}{k^2}}{\displaystyle _0^{\mathrm{}}}𝑑v{\displaystyle \frac{\epsilon (v1)}{(v1)(ve^{i\theta })(ve^{i\theta })}}`$ (48)
where
$`T`$ $`=`$ $`(E_iA+mA^{})\mathrm{sin}\theta +(E_iBmB^{})(1\mathrm{cos}\theta )`$ (49)
$`=`$ $`\overline{u}_f(E_i\gamma ^0m)(1\mathrm{cos}\theta +i\mathrm{sin}\theta \mathrm{\Sigma }_2)u_i`$
Using the definition in (41), the profactor T can be shown to vanish.
## V CONCLUSIONS AND DISCUSSION
In Ref. 14 it was claimed that the Born approximation for relativistic spin-$`\frac{1}{2}`$ AB scattering works, by demonstrating that this amplitude agrees with the corresponding terms in the series expansion of the exact amplitude. As the exact amplitude is proportional to $`\mathrm{sin}\pi \alpha `$, the demonstration of the full consistency of the Born approximation however requires a further step, namely the vanishing of the $`O(\alpha ^2)`$ contributions. This was already done in the context of the Galilean invariant field theory whose 2-particle sector is known to be equivalent to the AB Schroedinger equation. Encouraged by the success of these works, we have addressed the same issue directly, namely by considering the $`O(\alpha ^2)`$ contribution for the relativistic scattering of spin-$`\frac{1}{2}`$ particles from an infinitely long solenoid in the context of covariant perturbation theory, and shown that it indeed vanishes, thus completing the consistency check of the Born approximation for the relativistic spin-$`\frac{1}{2}`$ problem.
|
warning/0004/cond-mat0004353.html
|
ar5iv
|
text
|
# Peculiarities of anharmonic lattice dynamics and thermodynamics of alkaline-earth metals
## I Introduction
The investigation of anharmonic effects (AE) in lattice dynamics is a classical problem of solid-state physics. It is important, particularly, because of the role these effects can play in phenomena associated with structural phase transitions and melting in crystals (see, e.g. ). At the same time, obtaining any information about the magnitude and scale of AEs from experiment and theory is a difficult problem. The experimental study of such ”basic” AEs as the frequency shift and damping of phonons is very difficult and leads to a large uncertainty in the results (see, e.g., the data presented in for bcc and fcc metals, respectively). Up to now first-principles microscopic calculations of AEs have been performed for one point of the Brillouin zone (N) in the bcc phase of Zr and four points (N,P,$`\omega `$,G) in Mo . Detailed information about AEs in the entire Brillouin zone and their temperature dependence has been obtained in on the basis of pseudopotential theory for the bcc phases of alkali and alkaline-earth metals. For these metals the most striking manifestations of AEs are due to the ”soft-mode behavior”(the anomalous temperature dependence of the phonon frequencies) of the $`\mathrm{\Sigma }_4`$ branch. It is of interest to calculate AEs for the ”general position”, i.e. for crystals which do not possess soft vibrational modes. Such crystals include most metals with close-packed structures, for example, fcc. In a recent paper the results of calculations performed for the anharmonic effects in lattice dynamics of Ir having the fcc structure have been presented. In order to understand the specific features of AEs in lattice dynamics, related to structural phase transitions, it is interesting to investigate them in several phases for polymorphic metals. A classical example of structural phase transitions in metals is the fcc-bcc temperature transition in Ca and Sr . Therefore studies of AE features in these metals seem to be important. In the present paper the AE features in Ca and Sr lattice dynamics and thermodynamics have been investigated basing on microscopic calculations of the heat capacity, thermal expansion, temperature frequency shifts, damping of phonons and Gruneisen parameters.
## II Approximations and computation procedure
As noted in Introduction, at present, ab initio calculations of phonon frequency with regard to AE have been only done for some highly symmetric points of the Brillouin zone in a few metals . Consistent ab initio calculations of such quantities as phonon damping, as well as thermal expansion coefficients and other integral anharmonic effects in metals are now extremely difficult. Therefore in works reported so far particular models of interatomic interactions were used. For example, for alkali metals and Ir the pair interaction approximation basing on pseudopotential theory was used, which gives a reliable description of a variety of lattice properties of these metals. Construction of a similar model for alkaline-earth metals appears to be difficult because of the specific features of electron structure. The fcc phase of Ca and Sr has many van Hove singularities in the density of electron states near the Fermi level $`E_F`$ and, therefore, for the adequate description of lattice properties allowance for related contributions to the total energy is important . At the same time, the Fermi surface for these metals is close to that in the approximation of almost free electrons, and parameter $`|V_𝒈|/E_F`$ (where g is the reciprocal lattice vector, $`|V_𝒈|`$ is the Fourier component of pseudopotential) is small (less than $`0.1`$) . As shown in , under these conditions the proximity of the Fermi surface to the Brillouin zone boundary essentially contributes to the elastic moduli and to frequencies of long-wave phonons with the wave vector
$$qq_c=g_1\sqrt{\frac{|V_𝒈||E_cE_F|}{E_F}}$$
(1)
where $`E_c`$ is the van-Hove singularity nearest to $`E_F`$, $`g_1`$ is the minimum vector of reciprocal lattice while the corresponding anomalous contributions to frequencies of phonons with $`q>q_c`$ are much weaker. Therefore it may be expected that for the description of thermodynamics of fcc phase of Ca and Sr for $`T\mathrm{\Theta }_Dq_c/g_10.1\mathrm{\Theta }_D`$ (where $`\mathrm{\Theta }_D`$ is the Debye temperature) the ordinary approximation of pairwise interactions corresponding to the second order to perturbation theory for $`|V_𝒈|/E_F`$ would be applicable. This model was successfully used in for the description of the phonon spectra of Ca and Sr in the high temperature bcc phase.
In the present paper the phonon spectra, thermodynamic and anharmonic properties were calculated using the model corresponding to the second order of perturbation theory for pseudopotential (see, e.g. ). For the latter the Animalou-Heine expression
$$V(q)=\frac{4\pi Ze^2}{q^2}\left[\mathrm{cos}qr_0+U\left(\frac{\mathrm{sin}qr_0}{qr_0}\mathrm{cos}qr_0\right)\right]\mathrm{exp}\left[0.03\left(\frac{q}{2k_{F_0}}\right)^4\right]$$
(2)
was used, where $`Z=2`$ is the ion charge, $`e`$ is the electron charge, $`r_0`$, U are the pseudopotential parameters listed in table I, $`k_{F_0}`$ is the Fermi momentum at zero pressure. For the screening the Geldart-Taylor approximation with the Ceperley-Alder expression for correlation energy was used (see for details). To eliminate the influence of fcc band structure peculiarities (the closeness of the Fermi surface to the bounadaries of the Brillouin zone), parameters $`r_0`$ and $`U`$ were fitted to the properties of high-temperature bcc phase: $`\mathrm{\Omega }=\mathrm{\Omega }_{exp}(T=T_s)`$ and $`\mathrm{\Theta }_D=\mathrm{\Theta }_D^{exp}(T=T_s)`$ for Sr and $`\mathrm{\Omega }=\mathrm{\Omega }_{exp}(T=T_s)`$ and $`C^{}=C_{exp}^{}`$ for Ca. Here $`T_s`$ is the temperature of bcc-fcc transition (see table I), index ”exp” indicates the related experimental value, $`C^{}=1/2(C_{11}C_{12})`$ is one of the shear moduli.
For the AE calculations the standard anharmonic perturbation theory was used . With an accuracy up to terms of $`\varkappa ^2`$ order, where $`\varkappa =(m/M)^{1/4}`$ is the adiabatic parameter, $`m`$, $`M`$ are the masses of electron and ion, respectively, the Hamiltonian of phonon subsystem can be written as:
$$H=H_0+H_{qh}+H_3+H_4$$
(3)
where
$$H_0=\underset{\lambda }{}\omega _\lambda b_\lambda ^+b_\lambda $$
(4)
is the Hamiltonian of harmonic approximation, $`\lambda 𝒒\xi `$, q is the wave vector of phonon, $`\xi `$ is the number of phonon branch, $`b_\lambda ^+`$, $`b_\lambda `$ are the phonon operators of creation and annihilation.
$$H_{qh}=\underset{\lambda i}{}\left(\frac{\omega _\lambda }{u_i}\right)_{u_i=0}u_ib_\lambda ^+b_\lambda $$
(5)
is the quasiharmonic Hamiltonian, $`u_i`$ are the deformation parameters (in cubic crystals it is sufficient to allow for dilatation $`u_1`$, $`du_1=d\mathrm{ln}\mathrm{\Omega }`$, where $`\mathrm{\Omega }`$ is the lattice volume)
$`H_3`$ $`=`$ $`{\displaystyle \underset{\lambda _1\lambda _2\lambda _3}{}}V^{(3)}(\lambda _1,\lambda _2,\lambda _3)Q_{\lambda _1}Q_{\lambda _2}Q_{\lambda _3}`$ (6)
$`=`$ $`{\displaystyle \underset{\lambda _1\lambda _2\lambda _3}{}}\mathrm{\Phi }^{(3)}(\lambda _1,\lambda _2,\lambda _3)A_{\lambda _1}A_{\lambda _2}A_{\lambda _3}`$ (7)
$`H_4`$ $`=`$ $`{\displaystyle \underset{\lambda _1\lambda _2\lambda _3\lambda _4}{}}V^{(4)}(\lambda _1,\lambda _2,\lambda _3,\lambda _4)Q_{\lambda _1}Q_{\lambda _2}Q_{\lambda _3}Q_{\lambda _4}`$ (8)
$`=`$ $`{\displaystyle \underset{\lambda _1\lambda _2\lambda _3\lambda _4}{}}\mathrm{\Phi }^{(4)}(\lambda _1,\lambda _2,\lambda _3,\lambda _4)A_{\lambda _1}A_{\lambda _2}A_{\lambda _3}A_{\lambda _4}`$ (9)
are the Hamiltonians of three- and four- phonon processes, respectively,
$$Q_\lambda =\frac{1}{\sqrt{2M\omega _\lambda }}A_\lambda ,A_\lambda =b_\lambda +b_\lambda ^+$$
(10)
is the operator of phonon coordinate, $`\lambda 𝐪,\xi `$. Here and below we put the Planck constant as $`\mathrm{}=1`$. Expressions for the amplitudes of three- and four-phonon processes $`\mathrm{\Phi }^{(3)}`$, $`\mathrm{\Phi }^{(4)}`$, in terms of the pseudopotential model used are shown in . With an accuracy up to terms of $`\varkappa ^2`$ order we have the following expressions for anharmonic shifts of phonon frequencies and damping (see Appendix):
$$\mathrm{\Delta }_\lambda =\mathrm{\Delta }_\lambda ^{(qh)}+\mathrm{\Delta }_\lambda ^{(3)}+\mathrm{\Delta }_\lambda ^{(4)}$$
(11)
$$\mathrm{\Delta }_\lambda ^{(qh)}=\gamma _\lambda \frac{\mathrm{\Delta }\mathrm{\Omega }}{\mathrm{\Omega }}$$
(12)
$`\mathrm{\Delta }_\lambda ^{(3)}`$ $`=`$ $`18{\displaystyle \underset{\lambda _2\lambda _3}{}}|\mathrm{\Phi }^{(3)}(\lambda ,\lambda _2,\lambda _3)|^2\{(1+N_{\lambda _2}+N_{\lambda _3})[{\displaystyle \frac{1}{\omega _\lambda +\omega _{\lambda _2}+\omega _{\lambda _3}}}`$ (14)
$`+{\displaystyle \frac{𝒫}{\omega _{\lambda _2}+\omega _{\lambda _3}\omega _\lambda }}]+(N_{\lambda _2}N_{\lambda _3})[{\displaystyle \frac{𝒫}{\omega _{\lambda _3}\omega _{\lambda _2}\omega _\lambda }}{\displaystyle \frac{𝒫}{\omega _{\lambda _2}\omega _{\lambda _3}\omega _\lambda }}]\}`$
$$\mathrm{\Delta }_\lambda ^{(4)}=12\underset{\lambda ^{}}{}\mathrm{\Phi }^{(4)}(\lambda ,\lambda ,\lambda ^{},\lambda ^{})(1+2N_\lambda ^{})$$
(15)
$`\mathrm{\Gamma }_\lambda `$ $`=`$ $`18\pi {\displaystyle \underset{\lambda _2\lambda _3}{}}|\mathrm{\Phi }^{(3)}(\lambda ,\lambda _2,\lambda _3)|^2\{(1+N_{\lambda _2}+N_{\lambda _3})\delta (\omega _{\lambda _2}+\omega _{\lambda _3}\omega _\lambda )`$ (17)
$`+(N_{\lambda _2}N_{\lambda _3})[\delta (\omega _{\lambda _3}\omega _{\lambda _2}\omega _\lambda )\delta (\omega _{\lambda _2}\omega _{\lambda _3}\omega _\lambda )]\}`$
here
$$\gamma _\lambda =\frac{\mathrm{ln}\omega _\lambda }{\mathrm{ln}\mathrm{\Omega }}$$
(18)
are the Gruneisen parameters, $`\mathrm{\Delta }\mathrm{\Omega }`$ is the change in the crystal volume due to the thermal expansion
$$N_\lambda =\frac{1}{\mathrm{exp}(\omega _\lambda /T)1}$$
(19)
is the Planck distribution function, $`𝒫`$ is the symbol of principle value. In the particular calculations the approximation
$$𝒫\frac{1}{x}\frac{x}{x^2+\epsilon ^2}$$
(20)
$$\delta (x)\frac{1}{\pi }\frac{\epsilon }{x^2+\epsilon ^2}$$
(21)
was used for different small positive $`\epsilon `$ with subsequent extrapolation $`\epsilon 0`$. The details of the calculations, in particular, the summing up over the Brillouin zone is discussed in detail in .
## III Calculations of phonon spectra in harmonic approximation
The results of calculations of phonon spectra and Gruneisen parameters (18) for fcc and bcc phases of Ca and Sr are shown in Figs. 1-5. As noted above, the theoretical model used is unapplicable for the description of phonon spectra in the fcc phase of these metals in the nearest vicinity of point $`\mathrm{\Gamma }`$ because of unaccounted effects of Fermi surface proximity to the Brillouin zone boundaries . It follows from Fig. 1 that $`q_c`$c from Eq. (1) is of the order of $`0.1g_1`$. Fig. 2 shows that the phonon spectrum in the bcc phase of Ca and Sr is qualitatively similar to phonon spectra in the bcc phase of alkali metals , exhibiting the same characteristic features: presence of ”soft” $`\mathrm{\Sigma }_4`$ branch and prominent minimum for branch F<sub>1</sub>. In the fcc phase of these metals there are no soft mode anomalies. Fig. 3 displays such phonon spectra of bcc Sr for $`T=T_s`$ with allowance for anharmonic frequency shifts and corresponding experimental data in accordance with . It is seen that there is a reasonable agreement between the calculated phonon spectra and experimental data. Unfortunately, any direct experimental data for $`\omega (𝐪)`$ in the fcc phase of Ca and Sr are unavailable. The comparison of theory with the experiment for this phase will be done in the next Section basing on the data on temperature dependence of the lattice heat capacity. As seen from Figs. 4, 5 the Gruneisen model $`\gamma _\lambda =`$const is absolutely unapplicable, even qualitatively, for the description of bcc phases in alkali metals. In the fcc phase the q-dependence is much weaker (provided that the vicinity of point $`\mathrm{\Gamma }`$, where the model itself becomes unapplicable, is not considered). This seems to be because of lack of ”soft modes” in the phonon spectra of fcc phase.
## IV Anharmonic effects in lattice dynamics
The calculation results for the anharmonic effects in the lattice dynamics for Ca and Sr are given in Figs. 6-15. The temperature dependencies of the frequency shift and phonon damping are shown for Ca as an example (Figs. 6, 7, 12, 13); for Sr the anharmonic effects are somewhat stronger than for Ca, nevertheless all dependencies are similar.
Figs. 6, 7 displays temperature dependencies of phonon frequency shifts:
$`\mathrm{\Delta }\omega (T)`$ $`=`$ $`\omega (T)\omega (T=0)`$ (22)
$`=`$ $`\mathrm{\Delta }_\lambda ^{(qh)}+\mathrm{\Delta }_\lambda ^{(3)}+\mathrm{\Delta }_\lambda ^{(4)}`$ (23)
Here $`\mathrm{\Delta }_\lambda ^{(qh)}`$, $`\mathrm{\Delta }_\lambda ^{(3)}`$, $`\mathrm{\Delta }_\lambda ^{(4)}`$ are values determined by Eqs. (12-17) respectively without the contribution of zero point oscillations, since such contributions enter both $`\omega (T)`$ and $`\omega (T=0)`$ and are compensated in the difference (23). Note also that since fitting of the pseudopotential papameters was done at the volume $`\mathrm{\Omega }=\mathrm{\Omega }_{exp}(T=T_s)`$ (see Section II), $`\mathrm{\Delta }\mathrm{\Omega }`$ entering right hand side of Eq. (12) is determined by the following relation: $`\mathrm{\Delta }\mathrm{\Omega }=\mathrm{\Omega }(T)\mathrm{\Omega }(T_s)`$
It is seen that both in the bcc and in fcc phases $`\mathrm{\Delta }\omega (T)`$ early (at $`T0.1÷0.2T_{pl}`$) comes to high-temperature behavior $`\mathrm{\Delta }\omega (T)T`$. In the bcc phases of Ca and Sr the typical soft-mode behavior $`d\omega /dT>0`$ for $`\mathrm{\Sigma }_4`$ branch takes place (see the data for $`N_4`$ in Fig. 7). As pointed out in the anharmonic effects in the bcc phases of alkaline-earth and alkaline metals are qualitatively similar in the whole, but, however, the scale of these effects in alkaline-earth metals is essentially larger. In the fcc phase of Ca and Sr there are no ”soft modes” and $`d\omega /dT<0`$ for all vibrational modes.
Figs. 8, 9 shows the q-dependence of the relative frequency shift in Ca and Sr. The comparison of the results for the fcc phase (Fig. 8) with those obtained in for Ir indicates that the positions of frequency shift minima and maxima in the Brillouin zone depends essentially on the ion charge; in Ir at $`Z=4.5`$ maxima of $`\delta `$ are reached at symmetric points X and L, while in alkaline-earth metals they are significantly shifted. Moreover, in contrast with the temperature dependencies $`\delta (T)`$, there is a noticeable difference in the behavior of $`\delta (𝒒)`$ in Ca and Sr (cf., for example, the dependence in $`\mathrm{\Gamma }L`$ direction in Fig. 8).
For the bcc phases of Ca and, particularly, Sr, $`\delta (𝐪)`$ has a prominent maximum in the narrow region of $`𝐪`$-space near point N for the soft branch $`\mathrm{\Sigma }_4`$ for which $`\delta >0`$ (see Fig. 9).
On example of Ca, Figs. 10, 11 show the contributions of three- and four- phonon processes to the frequency shift. In both phases these contributions have opposite signs in the larger part of the Brillouin zone ($`\mathrm{\Delta }_3<0,\mathrm{\Delta }_4>0`$) The mentioned above ”soft mode” behavior of the branch $`\mathrm{\Sigma }_4`$ in the bcc phase is due to a sharp increase in $`\mathrm{\Delta }_4`$ as in alkali metals and in $`BaTiO_3`$-type ferroelectrics .
In Figs. 12, 13 the temperature dependence of phonon damping in the symmetric points of Brillouin zone is shown for Ca as an example. First of all it should be pointed out that, in accordance with Eq. (17) the damping does not vanish even at $`T=0`$ . It is due to three-phonon processes (phonon decay into two phonons) and has a smallness of $`\varkappa ^2`$ order. It is seen from Figs. 12, 13 that for Ca it is of the order of $`0.2÷0.5`$% of phonon frequency. As in the case of anharmonic frequency shift, it is seen that transition to asymptotic $`\mathrm{\Gamma }T`$ is reached very early (at $`T0.1T_{pl}`$).
Figs. 14, 15 show the q-dependence of relative damping of phonons $`\eta `$ in Ca and Sr. In the fcc phase $`\eta `$ has essentially nonmonotonous dependence on the wave vector, reaches values of about 25% in Sr and does not exceed 12% in Ca. In the bcc phase of Sr in the vicinity of point N $`\eta `$ reaches about 1 for the soft mode $`\mathrm{\Sigma }_4`$, which formally indicates inapplicability of the anharmonic perturbation theory in this narrow region. At the same time in the larger part of the Brillouin zone the anharmonic perturbation theory seems to be applicable. In view of this a problem arises on accurate separation of the contribution of the vicinity of point N to integral anharmonic characteristics such as free energy. This question will be considered elsewhere .
## V Analysis of experimental data on heat capacity and thermal expansion
### A Heat capacity
The lattice heat capacity is determined by the phonon state density $`g(\omega )`$. Some characteristics of the latter, which can be used for comparing our description of phonon spectra with the experiment can be found from the analysis of temperature dependence of heat capacity at the constant pressure $`C_p(T)`$.
Here we carried out the separation of the lattice components from the heat capacity measured in the experiment and determined some average frequencies (momenta) of phonon spectrum, without using any model.
In our analysis we used the following set of data: for Sr – direct experimental data from , for Ca – the data from handbooks; in the region 2-50 K – the data from handbook which in the region 2-10 K is based on , and in the region 50-600 – the data taken from handbook .
In the analysis we neglected the temperature dependence of the electron heat capacity coefficient and assumed that the anharmonism is weak. The assumption on smallness of anharmonic contributions to the thermodynamical properties of metals is justified as it was checked by a direct calculation and found to be true even for alkali metals having soft phonon modes . Under these assumptions the heat capacity measured experimentally at a constant pressure, $`C_p`$ is described by the relations:
$$C_p=C_{ph}+C_a$$
(24)
$$C_a=\left\{\gamma +(A\gamma )(C_{ph}/3R)^2\right\}T$$
(25)
where
$`C_{ph}`$ is the phonon component of the heat capacity in the harmonic approximation,
$`C_a`$ is the sum of contributions to the heat capacity, having the linear temperature dependence and coming from the anharmonic effects, thermal expansion of the lattice and conduction electrons;
$`\gamma `$ is the coefficient of electron heat capacity at low temperatures,
$`A`$ is the coefficient of linear in temperature term at high temperatures,
$`R`$ is the gas constant.
The interpolation formula (25) for $`C_a`$ gives the corresponding linear temperature asymptotics both at low- and high-temperatures and ensures a smooth transition between the low and high temperature asymptotics in accordance with the law similar to Nernst-Linderman formula .
In the high-temperature region the phonon component of heat capacity was described by the expression proposed by Naumov
$$C_{ph}=3R\left\{1\frac{1}{12}\left(\frac{\mathrm{\Omega }_2}{T}\right)^2+\frac{1}{240}\left(\frac{\mathrm{\Omega }_4}{T}\right)^4+\phi \left(\frac{\mathrm{\Omega }_{}}{T}\right)\right\}$$
(26)
$$\phi (z)=\frac{z^2exp(z)}{\left(1exp(z)\right)^2}\left(1\frac{1}{12}z^2+\frac{1}{240}z^4\right)$$
(27)
Here the asymptotic expansion of phonon heat capacity over small parameter $`z=\mathrm{\Omega }/T`$ is used:
$$C_{ph}=3R\left(1\underset{n=2}{\overset{\mathrm{}}{}}\frac{(n1)B_n}{n!}\left(\frac{\mathrm{\Omega }_n}{T}\right)^n\right)$$
(28)
where $`B_n`$ are the Bernulli numbers ($`B_2=1/6`$, $`B_4=1/30`$, $`B_66=1/42`$, $`B_8=1/30`$, $`B_{10}=5/66`$ etc. For the odd n, beginning from n = 3, all $`B_n`$ = 0).
In (26) the second and fourth order correction for $`\mathrm{\Omega }/T`$ are separately written, and function $`\phi (\frac{\mathrm{\Omega }_{}}{T})`$ allows for all higher corrections in the ” Einstein” approximation, i.e. assuming $`\mathrm{\Omega }_n=\mathrm{\Omega }_{}`$ for all $`n6`$. The values $`\mathrm{\Omega }_n`$ characterize momenta (mean frequencies) of phonon spectrum according to the relation:
$$(\mathrm{\Omega }_n)^n=<\omega ^n>=_0^{\mathrm{}}g(\omega )\omega ^n𝑑\omega /_0^{\mathrm{}}g(\omega )𝑑\omega $$
(29)
The value $`\gamma `$ entering $`C_a`$ (25) was determined in the standard manner: by fitting of heat capacity in the low temperature region (particularly, 2-5 K for Ca and 5-15 K for Sr) by the relation $`C_p=\gamma T+\beta T^3`$. The values $`\gamma `$, $`\beta `$ as well as the limiting low-temperature value of the Debye temperature $`\mathrm{\Theta }_{LT}`$ related to $`\beta `$ by relation $`\beta =12\pi ^4R/(5\mathrm{\Theta }_{LT}^3)`$ are listed in Table II. The estimations of $`\mathrm{\Theta }_{LT}`$ we obtained from the set of data used reasonably agree with those available in the literature (see ).
The value $`A`$ together with $`\mathrm{\Omega }_2`$ , $`\mathrm{\Omega }_4`$ and $`\mathrm{\Omega }_{}`$ were determined using the least square method by the fitting of heat capacity by relations (26, 27) in the temperature region 40-600 K for Ca and 32-350 K for Sr. Within this temperature range these relations described the experimental results with least square deviation of the order of 1% for Ca and 0.3% for Sr. The parameters $`A`$, $`\mathrm{\Omega }_2`$, $`\mathrm{\Omega }_4`$ and $`\mathrm{\Omega }_{}`$ determined by the least square method are shown in Table II. The table also gives the limiting high-temperature value of Debye temperature $`\mathrm{\Theta }_{HT}=\mathrm{\Theta }_D(T)`$ where $`\mathrm{\Theta }_{LT}T<T_m`$ ($`T_m`$ is the melting temperature) related to the second momentum of the phonon spectrum by: $`\mathrm{\Omega }_2=\mathrm{\Theta }_{HT}\sqrt{3/5}`$ .
It should be pointed out that the lower value of the least square deviation for Sr is due to that over the whole temperature range the experimental data from the same source were used. On the other hand, its higher value for Ca seems to be due to non-smooth joining of the data at the transition from one source to another one.
The analysis performed permitted the contribution of $`C_{ph}`$ in the harmonic approximation to be separated from the total heat capacity and the electron and anharmonic contributions to be left out. Some phonon spectrum momenta are expressed directly via phonon heat capacity integrals .
$$<\omega >=2_0^{\mathrm{}}\left(1\frac{C_{ph}}{3R}\right)𝑑T$$
(30)
$$<\omega ^1>=\frac{3}{\pi ^2}_0^{\mathrm{}}\frac{C_{ph}}{3R}T^2𝑑T$$
(31)
$$<\omega ^2>=0.138651_0^{\mathrm{}}\frac{C_{ph}}{3R}T^3𝑑T$$
(32)
$$<\omega ^1log\omega >=\frac{3}{\pi ^2}_0^{\mathrm{}}\frac{C_{ph}}{3R}log\left(\frac{T}{0.70702}\right)T^2𝑑T$$
(33)
We calculated these momenta with the integration over the experimental points in the temperature region 10-300 K and outside this region – over low temperature asymptotic $`C_{ph}=\beta T^3`$ and high temperature asymptotic in the Debye spectrum model as in . The mean frequencies corresponding to these momenta are listed in Table II. The meanings of $`\mathrm{\Omega }_2`$, $`\mathrm{\Omega }_1`$, $`\mathrm{\Omega }_1`$ corresponding to that of $`\mathrm{\Omega }_n`$ in relation 29 for n = -2, - 1 and 1 while the meaning of $`\mathrm{\Omega }_{log}`$ is determined by relation:
$`\mathrm{log}(\mathrm{\Omega }_{log})`$ $`=`$ $`{\displaystyle \frac{<\omega ^1\mathrm{log}\omega >}{<\omega ^1>}}`$ (34)
$`=`$ $`{\displaystyle \underset{0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{g(\omega )\mathrm{log}\omega }{\omega }}𝑑\omega /{\displaystyle \underset{0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{g(\omega )}{\omega }}𝑑\omega `$ (35)
The contribution of $`C_{ph}`$ in the harmonic approximation separated from the total heat capacity is shown in Figs. 16, 17.
### B Thermal expansion
The coefficient of thermal expansion $`\alpha (T)`$ contains two contributions: the electron one $`\alpha _e`$ and the lattice one $`\alpha _l`$. In the low temperature region the electron contribution is linearly dependent on temperature, while the lattice contribution is proportional to $`T^3`$. This makes it possible to separate these contributions fitting the experimental data by the dependence:
$$\alpha (T)=DT+ET^3$$
(36)
In the paper the electron contribution to thermal expansion of Ca and Sr is determined. For the determination of the lattice contribution to thermal expansion we used the following set of experimental data: in the low temperature region — the data from and in the high temperature region — the data from for Ca and for Sr. The phonon contribution to thermal expansion was determined as the difference of the experimental data and the electron contribution. It was assumed that the electron contribution is linear by temperature over the whole temperature region, its value being determined by low temperature asymptotic in . The lattice contribution to thermal expansion separated from the experimental data is shown in Figs. 18, 19.
The lattice contributions to the heat capacity and the coefficient of thermal expansion, separated from the experimental data were used for the determination of the Gruneisen lattice parameter by the formula:
$$G=\frac{3\alpha BV_m}{C_{ph}}$$
(37)
where $`B`$ is the bulk modulus, $`\alpha `$ is the temperature coefficient of linear expansion, $`V_m`$ is the molar volume. The temperature dependence of the Gruneisen parameter $`\gamma (T)`$ is shown in Figs. 18, 19.
## VI Calculations of heat capacity and thermal expansion
Figs. 16, 17 give the computation results of lattice heat capacity of Ca and Sr in the harmonic approximation and the effective Debye temperature. It is seen that, with the exception of the region of lowest temperatures where the phonon spectrum model itself becomes inapplicable (see Section II), the theory is in a perfect agreement with the experiment. It is especially well seen in Table II, where the computation data are compared with the momenta of phonon density of states, found from the experimental data.
Figs. 18, 19 displays the computation results for the temperature dependence of the volume $`\mathrm{\Delta }\mathrm{\Omega }(T)`$ and of the Gruneisen parameter $`\gamma (T)`$ for Ca and Sr.It is related to the Gruneisen parameters by the following expression:
$$\gamma (T)=\frac{{\displaystyle \underset{\lambda }{}}\gamma _\lambda N_\lambda (1+N_\lambda )\omega _\lambda ^2}{{\displaystyle \underset{\lambda }{}}N_\lambda (1+N_\lambda )\omega _\lambda ^2}$$
(38)
One can see that the temperature dependence of the macroscopic Gruneisen parameter is a rather smooth. At the same time, our calculations show a strong temperature dependence of microscopic Gruneisen parameters for the soft branch in BCC phases of Ca and Sr (see Table III). It would be very interesting to check this prediction experimentally.
## VII Conclusions
In the present paper a simple, and at the same time, adequate model of the interatomic interactions is proposed, which describes with a reasonably high accuracy the lattice properties of fcc and bcc phases of Ca and Sr. As a result, the dependence of AE in the dynamics of lattice on its geometry could be investigated. The comparison of the results obtained for the fcc phase of Ca and Sr with earlier results for Ir demonstrates an essential dependence of AE on the character of itneratomic interactions for the given crystal structure. As the question on the AE scale, particularly, for metals with polymorphic transformation (in this case bcc-fcc) is one of the key problems in the lattice dynamics, performance of further experimental investigations is desirable.
##
Here we shall give a short derivation of equations for the temperature shift of frequencies and phonon damping (11-17). In contrast to the basing work and subsequent publications, where the diagram method for Matsubara Green functions was used, we employ the method of two time Green functions . We shall begin with the chain of equations of motion for delayed two time Green functions
$$\omega A|B_\omega =[A,B]+[A,H]|B$$
(39)
where
$$A|B_\omega =i\underset{0}{\overset{\mathrm{}}{}}dt\mathrm{exp}\left[i(\omega +i\epsilon )t\right][A(t),B(0)]|_{\epsilon +0}$$
(40)
The bracketed expression indicates the commutator of operators, and that within the French quats — the averaging over the Gibbs ensemble. For the Hamiltonian (4-9) we have the accurate equation:
$`(\omega \omega _\lambda )b_\lambda |b_\lambda ^+_\omega `$ $`=`$ $`1+3{\displaystyle \underset{\lambda _2\lambda _3}{}}\mathrm{\Phi }^{(3)}(\lambda ,\lambda _2,\lambda _3)A_{\lambda _2}A_{\lambda _3}|b_\lambda ^+_\omega `$ (42)
$`+4{\displaystyle \underset{\lambda _2\lambda _3\lambda _4}{}}\mathrm{\Phi }^{(4)}(\lambda ,\lambda _2,\lambda _3,\lambda _4)A_{\lambda _2}A_{\lambda _3}A_{\lambda _4}|b_\lambda ^+_\omega `$
With an accuracy up to $`\varkappa ^2`$ the Green function describing the contribution of four-phonon processes to (42) can be decoupled
$`A_{\lambda _2}A_{\lambda _3}A_{\lambda _4}|b_\lambda ^+_\omega `$ $`=`$ $`A_{\lambda _2}A_{\lambda _3}A_{\lambda _4}|b_\lambda ^+_\omega +A_{\lambda _2}A_{\lambda _4}A_{\lambda _3}|b_\lambda ^+_\omega `$ (44)
$`+A_{\lambda _3}A_{\lambda _4}A_{\lambda _2}|b_\lambda ^+_\omega `$
$`=`$ $`\delta _{\lambda _2\lambda _3}(1+2N_{\lambda _2})\delta _{\lambda _4\lambda }+\delta _{\lambda _2\lambda _4}(1+2N_{\lambda _1})\delta _{\lambda _3\lambda }`$ (46)
$`+\delta _{\lambda _2\lambda _4}(1+2N_{\lambda _3})\delta _{\lambda _2\lambda }+b_\lambda |b_\lambda ^+_\omega `$
This is immediately followed by the result (15) for the contribution of four-phonon processes to the frequency shift. For the determination of the contribution of three-phonon processes we must write the chain of equations for the Green-functions entering the following relation
$`A_{\lambda _2}A_{\lambda _3}|b_\lambda ^+_\omega `$ $`=`$ $`b_{\lambda _2}b_{\lambda _3}|b_\lambda ^+_\omega +b_{\lambda _2}b_{\lambda _3}|b_\lambda ^+_\omega `$ (48)
$`+b_{\lambda _2}b_{\lambda _3}|b_\lambda ^+_\omega +b_{\lambda _2}b_{\lambda _3}|b_\lambda ^+_\omega `$
and make there decoupling. Upon separation of the real and imaginary parts of Green function $`b_\lambda |b_\lambda ^+_\omega `$ we obtain immediately the result (11-17). The quasiharmonic frequency shift proves to be then as expressed through the amplitudes of three-phonon processes:
$$\mathrm{\Delta }_\lambda ^{(qh)}=36\underset{\lambda _2\lambda _3}{}\frac{1+2N_{\lambda _3}}{\omega _{\lambda _2}}\mathrm{\Phi }(\lambda ,\lambda ,\lambda _2)\mathrm{\Phi }(\lambda _3,\lambda _3,\lambda _2)$$
(49)
As far as we know this expression is new, although the relation itself of thermal expansion to three-phonon anharmonism was qualitatively discussed as early as in the classical book by Paierls .
|
warning/0004/astro-ph0004389.html
|
ar5iv
|
text
|
# Delayed Recombination
## 1. Introduction
The measurements of the anisotropies of the cosmic microwave background (the CMB) offer extraordinarily powerful tests of the relativistic Friedmann-Lemaître cosmological model and the nature of the early stages of cosmic structure formation (e.g. Jungman et al. 1996). Indeed, the recent detection of the first peak in the angular power spectrum of the CMB temperature indicates space is close to flat (Miller et al. 1999; Melchiorri et al. 1999; de Bernardis et al. 2000). The interpretation is quite indirect, however so we must seek diagnostics for possible complications.
There are relatively few physical effects that can increase the angular scale of the first peak. Because of this fact, the observed large angular scale of the peak is believed to strongly disfavour open universes. Of the fundamental cosmological parameters, only a Hubble constant well in excess of observations can substantially increase the scale of the peak in an open or flat universe (Hu & Sugiyama 1995).
One possibility is that some process at redshift $`z1000`$ delayed recombination of the primeval plasma. This would increase the sound horizon at last scattering and decrease the angular size distance to last scattering, moving the first peak of the CMB temperature fluctuation spectrum to smaller angular wavenumber (Hu & White 1996; Weller et al. 1999), and biasing the measure of space curvature to a too large (more positive) apparent value. Another consequence of delayed recombination is that it would suppress the secondary peaks due to an increase in the time available for acoustic oscillations to dissipate (Silk 1968; Hu & White 1996).
Because the first peak is not observed to have suffered substantial dissipation, recombination in the delayed model must be rapid compared with the cosmological expansion rate. Thus if recombination were delayed by ionizing radiation from decaying dark matter (e.g. Sarkar & Cooper 1983; Sciama el al 1991; Ellis et al 1992) or evaporating primeval black holes (Nasel’skij & Polonarëv 1987), or by thermal energy input from cosmic string wakes (Weller et al. 1999), the source would have to terminate quite abruptly and well before $`z100`$ when the universe starts to become optically thin even if the baryons are fully ionized.
Here we consider a picture that more naturally allows rapid recombination: sources of radiation at $`z1000`$ that produce many more photons in the resonance Ly $`\alpha `$ line than ionizing photons, in the manner of a quasar. The Ly $`\alpha `$ photons would increase the population in the principal quantum number $`n=2`$ levels of atomic hydrogen, increasing the rate of photoionization from $`n=2`$ by the CMB. Since the rate of thermal photoionization from $`n=2`$ varies rapidly with redshift at $`z1000`$, the delayed recombination is rapid, and the residual ionization can be small enough that the optical depth for Thomson scattering after recombination is well below unity. Thus the height of the first peak in the spectrum of CMB temperature fluctuations is little affected by the delayed recombination. The shift in the angular wavenumber at the first peak is modest also, even if early sources produce many Ly $`\alpha `$ photons per baryon, but the shift can be considerably larger than the projected precision of the measurements. Thus it is fortunate that we seem to have an unambiguous diagnostic for delayed recombination in the suppression of the secondary peaks.
After this work was substantially complete we learned that the recent measurements by the BOOMERanG experiment in fact favor a substantial suppression of the second peak (de Bernardis et al. 2000). We emphasize that there are many other ways to account for this effect, and that there is no evidence for the early source of Ly $`\alpha `$ radiation assumed in our picture. Within the conventional adiabatic CDM model the recombination history is well understood (Seager et al. 2000 and earlier references therein), so there is good reason to expect the residual fluctuations in the CMB may be related to the cosmology in a simple and computable way. But it is good science to bear in mind the possibility that Nature is more complicated than our ideas, a rule that has particular force in cosmology because of the limited empirical basis.
## 2. The Model
Since the early source of radiation is purely conjectural, a simple model is appropriate. We assume the rate of production of Ly $`\alpha `$ resonance photons per unit volume (in excess of those produced by the primeval plasma) is
$$dn_\alpha /dt=ϵ_\alpha n_\mathrm{H}H(t),$$
(1)
where $`n_\mathrm{H}`$ is the number density of hydrogen nuclei, $`H(t)=\dot{a}/a`$ is the expansion rate, and $`ϵ_\alpha `$ is a free parameter. The computation of the ionization history by Seager et al. (2000) is readily adjusted to take account of this new source term if it is approximately homogeneous. The results in Figure 1 assume the Hubble parameter is $`h=0.7`$ in units of $`100`$ km s<sup>-1</sup> Mpc<sup>-1</sup>, the baryon density parameter is $`\mathrm{\Omega }_\mathrm{b}h^2=0.02`$ (Tytler et al. 2000) and the total matter density parameter is $`\mathrm{\Omega }_\mathrm{m}=0.25`$. The ionization history is quite similar at $`\mathrm{\Omega }_\mathrm{m}=0.4`$.
The optical depth for Thomson scattering subsequent to redshift $`z`$ is
$$\tau =63x\left(\frac{1Y_p}{0.76}\right)\left(\frac{\mathrm{\Omega }_\mathrm{b}h^2}{0.02}\right)\left(\frac{\mathrm{\Omega }_\mathrm{m}h^2}{0.122}\right)^{1/2}\left(\frac{z}{10^3}\right)^{3/2},$$
(2)
if the fractional ionization is constant at $`x`$ at redshift less than $`z`$. Figure 1 shows that the delayed recombination through the redshift $`z_{}`$ where $`\tau `$ passes through unity is about as fast relative to the expansion rate as in the standard recombination model. Furthermore the decrease in $`z_{}`$ itself is approximately
$$\frac{z_{}(ϵ_\alpha )}{z_{}(0)}=(1+3ϵ_\alpha )^{0.042}.$$
(3)
Sources of Ly $`\alpha `$ photons may also produce ionizing radiation. Since the rate of recombination of fully ionized baryons at the cosmic mean density is faster than the rate of expansion the ionization, $`x`$ may be approximated by the equilibrium equation
$$\alpha n_\mathrm{H}x^2ϵ_\mathrm{i}H.$$
(4)
The definition of the parameter $`ϵ_\mathrm{i}`$ follows equation (1), and $`\alpha `$ is the recombination coefficient for principal quantum numbers $`n2`$. At this ionization the optical depth for Thomson scattering is
$$\tau ϵ_\mathrm{i}^{1/2}\left(\frac{1Y_p}{0.76}\right)^{1/2}\left(\frac{\mathrm{\Omega }_\mathrm{b}h^2}{0.02}\right)^{1/2}\left(\frac{\mathrm{\Omega }_\mathrm{m}h^2}{0.122}\right)^{1/4}\left(\frac{z}{10^3}\right)^{1.2}.$$
(5)
Figure 2 shows in more detail the effect of the production of ionizing photons on the ionization history. If $`ϵ_\mathrm{i}1`$ the slow decrease in the optical depth through $`\tau 1`$ would remove the first peak in the anisotropy power spectrum, an effect the measurements indicate is unacceptable.
Sources of Ly $`\alpha `$ photons may also add energy to the CMB by inverse Compton scattering, but if the energy added is comparable to the energy in Ly $`\alpha `$ photons the effect on the $`y`$-parameter is small.
## 3. Temperature Anisotropy Power Spectrum
In the adiabatic CDM model the ionization history and the cosmological parameters fix the CMB anisotropy power spectrum; the results in Figure 2 for the cosmologically flat model with the parameters in Figure 1, and in Figure 3 for an open model with the cosmological parameters $`\mathrm{\Omega }_\mathrm{m}=0.6`$, $`\mathrm{\Omega }_\mathrm{\Lambda }=0`$, $`h=0.7`$, $`\mathrm{\Omega }_\mathrm{b}h^2=0.02`$ are computed using a code based on White & Scott (1996).
The shift in the angular wavenumber $`l_1`$ at the peak is noteworthy because $`l_1`$ is used to infer the space curvature. A general expression for its value in adiabatic models is (Hu & Sugiyama 1995, with $`\mathrm{}_10.73\mathrm{}_p`$)
$`\mathrm{}_1`$ $``$ $`{\displaystyle \frac{125}{\sqrt{\mathrm{\Omega }_\mathrm{m}+\mathrm{\Omega }_\mathrm{\Lambda }}}}[1+\mathrm{ln}(1\mathrm{\Omega }_\mathrm{\Lambda })^{0.085}]\left({\displaystyle \frac{z_{}}{10^3}}\right)^{1/2},`$ (6)
$`\times \left({\displaystyle \frac{1}{\sqrt{R_{}}}}\mathrm{ln}{\displaystyle \frac{\sqrt{1+R_{}}+\sqrt{R_{}+r_{}R_{}}}{1+\sqrt{r_{}R_{}}}}\right)^1,`$
where
$`r_{}`$ $`=`$ $`0.042m^1(z_{}/10^3),`$
$`R_{}`$ $`=`$ $`30b(z_{}/10^3)^1,`$
$`z_{}(0)`$ $``$ $`1008(1+0.00124b^{0.74})(1+c_1m^{c_2}),`$
$`c_1`$ $`=`$ $`0.0783b^{0.24}(1+39.5b^{0.76})^1,`$
$`c_2`$ $`=`$ $`0.56(1+21.1b^{1.8})^1,`$ (7)
with $`b\mathrm{\Omega }_\mathrm{b}h^2`$ and $`m\mathrm{\Omega }_\mathrm{m}h^2`$. Although in isocurvature models $`\mathrm{}_1`$ is larger by 50% or more, it scales with cosmological parameters in the same way.
The leading order dependence of the peak scale is then
$$\mathrm{}_1(\mathrm{\Omega }_\mathrm{m}+\mathrm{\Omega }_\mathrm{\Lambda })^{1/2}z_{}^{1/2},$$
(8)
such a $`10\%`$ change in $`z_{}`$ shifts $`\mathrm{}_1`$ by 5% and can compensate a 10% change in the spatial curvature $`\mathrm{\Omega }_\mathrm{m}+\mathrm{\Omega }_\mathrm{\Lambda }`$. In our model for delayed recombination with $`z_{}`$ given by eqn. (3), $`ϵ_\alpha =1000`$ can make an open universe with $`\mathrm{\Omega }_\mathrm{m}=0.6`$ appear flat (see Figure 4).
Even in the context of cosmologically flat models, one might want to consider the possibility of delayed recombination. The observed peak at $`\mathrm{}_1=197\pm 12`$ (95% CL, de Bernardis et al. 2000) favors either a slightly closed low density universe or one of high density ($`\mathrm{\Omega }_\mathrm{m}h^2`$). While flat models are certainly still consistent with the data, the expected increase in precision of the measurements could force one into accepting either a closed geometry, large Hubble constant or delayed recombination.
The amplitudes of the secondary peaks break this approximate degeneracy of delayed recombination with space curvature or the model for structure formation. While the promptness of the delayed recombination nearly preserves the first peak in the spectrum the secondary peaks are strongly suppressed because of the extra time the photons are allowed to diffuse (Silk 1968). As discussed in Hu & White (1996), the ratio of the first peak location to the location of the damping tail is a sensitive test of delayed recombination. An equally important but more subtle effect is that a delay in recombination raises the baryon-photon momentum density ratio $`R_{}`$ due to the redshifting of the photons. This effect suppresses the second peak and raises the third. Though an increase in the baryon density parameter also has this effect, it simultaneously reduces rather than increases the damping and can in principle be distinguished through the higher peaks.
Under standard recombination, the the ratio of power at the second versus first peak scales as
$$A_2(0)\frac{C_\mathrm{}_2}{C_\mathrm{}_1}0.7\left[1+\left(\frac{\mathrm{\Omega }_bh^2}{0.016}\right)^4\right]^{1/4}2.4^{n1}$$
(9)
In our model, $`A_2(ϵ_\alpha )A_2(0)z_{}(ϵ_\alpha )/z_{}(0)`$ such that delayed recombination is twice as effective in suppressing power at the second peak as it is at shifting the first peak. Observations currently favor $`A_21/3`$ (see Fig. 4) compared with $`0.5`$ in our fiducial $`\mathrm{\Lambda }`$CDM model with standard recombination. Of course the tilt $`n`$ and the baryon density can also lower this ratio.
Because of its cumulative effect over the secondary peaks, $`ϵ_\alpha `$ has statistically significant effects as long as it is greater than
$$ϵ_{\mathrm{min}}=\left[\underset{\mathrm{}=2}{\overset{\mathrm{}_{\mathrm{max}}}{}}(\mathrm{}+1/2)\left(\frac{\mathrm{ln}C_{\mathrm{}}}{ϵ}\right)^2\right]^{1/2},$$
(10)
where $`\mathrm{}_{\mathrm{max}}`$ is the largest $`\mathrm{}`$ for which the measurements are cosmic variance limited. For $`\mathrm{}_{\mathrm{max}}=500`$, $`ϵ_{\mathrm{min}}=0.006`$; for $`\mathrm{}_{\mathrm{max}}=1000`$, $`ϵ_{\mathrm{min}}=0.002`$.
## 4. Discussion
Our model for delayed recombination with $`ϵ_\alpha =1`$ yields a 5% shift of the first peak and a reduction of the secondary peak by 10%, an observationally interesting effect. The sources must have $`ϵ_\mathrm{i}1`$ (see Figure 2). This condition requires that the Ly $`\alpha `$ sources (perhaps hot stars or quasars) be surrounded by envelopes of neutral primeval material which strongly absorbs the ionizing radiation.
Our picture does not follow from the conventional CDM model for structure formation. Apparent problems with excess small-scale clustering in this model (Moore et al. 1999; Klypin et al. 1999) have motivated discussions of modifications (Spergel & Steinhardt 1999; Kamionkowski & Liddle 1999; Peebles 2000; Hu, Barkana, & Gruzinov 2000). These modifications would tend to delay the appearance of the first generation of gravitationally bound systems, however, which is in the opposite direction to what is postulated here. These lines of thought do not rule out early sources of Ly $`\alpha `$ photons, of course, but they do suggest one might best look for a source outside the adiabatic CDM model. Perhaps cosmic strings produced wakes that were subdominant to the primeval CDM density fluctuations in determining the mass fluctuation power spectrum but did produce occasional non-gaussian density fluctuations (Contaldi et al. 1999) large enough to have collapsed to stars or active black holes that could have produced Ly $`\alpha `$ photons.
To summarize, we have argued that the picture of delayed recombination caused by early sources of Ly $`\alpha `$ radiation, modeled after quasar spectra, but with suppressed X-ray emission, has the virtue that it naturally preserves the observed first peak in the CMB temperature angular power spectrum while shifting the peak to larger scales. The picture is ad hoc but important as a conceivable complication in the application of an exceedingly powerful cosmological test. The most direct diagnostic for delayed recombination seems to be the suppression of the secondary peaks in the spectrum. The amplitude of the third peak, which may be measured by the BOOMERanG experiment or the MAP satellite<sup>1</sup><sup>1</sup>1http://map.gsfc.nasa.gov, is crucial for distinguishing the effect of delayed recombination from that of adjustments of the values of space curvature, the baryon density, the Hubble constant or the shape of the spectrum of the primeval density fluctuations. If the measured secondary peaks agreed with the standard model for recombination with astronomically acceptable cosmological parameters that fit the primary peak, it would convincingly rule out our model for delayed recombination.
We have benefitted from discussions with Martin Rees. P.J.E.P. is supported in part by the NSF. S.S. is supported by NSF grant PHY-9513835. W.H. is supported by the Keck Foundation and a Sloan Fellowship.
|
warning/0004/hep-th0004082.html
|
ar5iv
|
text
|
# Vilkovisky-DeWitt Effective Action for Einstein Gravity on Kaluza-Klein Spacetimes 𝑀⁴×𝑆^𝑁
## I Introduction
Appelquist and Chodos AC (1) were the first to use the effective potential formalism to study the problem of spontaneous compactification in Kaluza-Klein theories. The hope was that quantum effects could explain the smallness of the extra dimensions. It was soon realized that results obtained using the standard effective action theory were dependent on which quantum gauge fixing condition was used RDS (2, 3). This non-uniqueness was overcome BO (4, 5) by the introduction of a new effective action formulated by Vilkovisky GV (6) and modified by DeWitt BD1 (7). It is now known as the Vilkovisky-DeWitt (VD) effective action and has the merit of being gauge choice independent. Progress in compactification, however, immediately slowed to a snail’s pace BLO (8, 9, 10, 11, 12, 13, 14, 15), because the VD effective action for gravity involves evaluating determinants of complicated non-local operators (even at one-loop). In CK1 (16), we were able to make progress with the six-dimensional case of a general background spacetime. We evaluated the divergent part of the VD effective action by extending the four-dimensional methods of Barvinsky and Vilkovisky BV (17). Due to the complexity of this calculation, we concluded that it is next to impossible to push this method to higher dimensional general spacetimes.
In this paper we therefore restrict ourselves to specific even-dimensional Kaluza-Klein backgrounds of the form $`M^4\times S^N`$. In the next section we briefly review the VD effective action formalism. We then extend the method of Barvinsky and Vilkovisky to these higher dimensional cases and expand the effective action in terms of functional traces of various operators. In this way we can extract the divergent part of the effective action by only considering a finite number of terms.
In Section III, the formalism is applied to Einstein gravity. The eigenvalues of the operators RO (18, 19) mentioned above are evaluated for gravity fields on $`M^4\times S^N`$ backgrounds. Using these eigenvalues, we then extract the divergent parts of the VD effective action. Because the internal geometry is assumed static the effective action only gives the effective potential. In Section IV two applications are made of these results. First, gauge-independent trace anomalies for gravitons are explicitly given for $`N`$=2, 4, and 6. Second, if the divergent part of the effective potential dominates the dynamics of the internal spheres, self-consistent stable configurations are shown not to exist. Conclusions are given in Section V, and formulae for the divergent parts of functional traces, relevant to these calculations, are listed in the Appendix.
## II Vilkovisky-DeWitt Effective Action
In this section we briefly review the formalism of the Vilkovisky-DeWitt effective action. We follow closely the method of Barvinsky and Vilkovisky BV (17), as well as their notation.
Consider first a general gauge theory with the action $`S^G[\mathrm{\Phi }]`$, where $`\mathrm{\Phi }^i`$ is the set of fields with $`i`$ in the condensed notation representing both the spacetime and gauge indices. Let $`Q_\alpha ^i`$ be the generators of gauge transformations,
$$\delta \mathrm{\Phi }^i=Q_\alpha ^iϵ^\alpha ,$$
(1)
where $`ϵ^\alpha `$ is the gauge parameter. Since the action $`S^G`$ is gauge invariant,
$$Q_\alpha ^i\frac{\delta S^G}{\delta \mathrm{\Phi }^i}=0.$$
(2)
Up to one-loop, the gauge fixing action in the background field gauge is given by
$$S^{GF}=\frac{1}{2}\chi ^\alpha c_{\alpha \beta }\chi ^\beta ,$$
(3)
where $`\chi ^\alpha `$ is a linear gauge condition, and $`c^{\alpha \beta }`$ is a local, invertible matrix. Both $`\chi ^\alpha `$ and $`c_{\alpha \beta }`$ may depend on the background field. The corresponding ghost operator is $`Q_\alpha ^i(\delta \chi ^\beta /\delta \mathrm{\Phi }^i)`$, and the one-loop effective action can be written as
$$iW=\frac{1}{2}\mathrm{Trln}F_{ij}+\mathrm{Trln}\left(Q_\alpha ^i\frac{\delta \chi ^\beta }{\delta \mathrm{\Phi }^i}\right),$$
(4)
where
$$F_{ij}=\frac{\delta ^2S^G}{\delta \mathrm{\Phi }^i\delta \mathrm{\Phi }^j}\frac{\delta \chi ^\alpha }{\delta \mathrm{\Phi }^i}c_{\alpha \beta }\frac{\delta \chi ^\beta }{\delta \mathrm{\Phi }^j}.$$
(5)
However, the one-loop effective action $`W`$ is gauge dependent in general, that is, it depends on the choice of the gauge fixing action $`S^{GF}`$ when the background field is not a solution of the classical equation of motion,
$$_i\frac{\delta S^G}{\delta \mathrm{\Phi }^i}=0.$$
(6)
This poses a problem using the effective action formalism in off-shell calculations, for example, in calculating the trace anomalies for gravitons CK3 (20), and in studying the spontaneous compactification of Kaluza-Klein spaces HKLT (5). This gauge fixing problem can be resolved by using the VD effective action, since this effective action is manifestly independent to the choice of gauge conditions.
At the one-loop level, the VD effective can be obtained simply by replacing the functional derivative in the ordinary effective action by a covariant functional derivative
$`{\displaystyle \frac{\delta ^2S^G}{\delta \mathrm{\Phi }^i\delta \mathrm{\Phi }^j}}`$ $``$ $`{\displaystyle \frac{D}{\delta \mathrm{\Phi }^i}}\left({\displaystyle \frac{\delta S^G}{\delta \mathrm{\Phi }^j}}\right)`$ (7)
$`=`$ $`{\displaystyle \frac{\delta ^2S^G}{\delta \mathrm{\Phi }^i\delta \mathrm{\Phi }^j}}\mathrm{\Gamma }_{ij}^k{\displaystyle \frac{\delta S^G}{\delta \mathrm{\Phi }^k}},`$
where the connection consists of two parts,
$$\mathrm{\Gamma }_{ij}^k=\left\{\begin{array}{c}k\\ ij\end{array}\right\}+T_{ij}^k.$$
(8)
$`\left\{\begin{array}{c}k\\ ij\end{array}\right\}`$ is the local Christoffel symbol constructed in the usual manner from a configuration space metric $`\gamma _{ij}`$,
$$\left\{\begin{array}{c}k\\ ij\end{array}\right\}=\frac{1}{2}\gamma ^{kl}(\gamma _{li,j}+\gamma _{lj,i}\gamma _{ij,l}),$$
(9)
where the derivative in $`\gamma _{li,j}=\delta \gamma _{li}/\delta \mathrm{\Phi }^j`$ represents the ordinary functional derivative. The configuration space metric is the new ingredient in the VD theory. A prescription for defining it has been given by Vilkovisky GV (6). The non-local part $`T_{ij}^k`$ of the connection comes from the gauge constraints,
$$T_{ij}^k=2Q_{\alpha ;(i}^k\gamma _{j)l}N^{\alpha \beta }Q_\beta ^l+\gamma _{(il}N^{\alpha \mu }Q_\mu ^lQ_\alpha ^mQ_{\beta ;m}^k\gamma _{j)n}N^{\beta \nu }Q_\nu ^n,$$
(10)
where the derivative in $`Q_{\beta ;m}^k`$ is the covariant functional derivative defined with the Christoffel symbol $`\left\{\begin{array}{c}k\\ ij\end{array}\right\}`$, and $`N^{\alpha \beta }`$ is the inverse of $`N_{\alpha \beta }^1`$,
$$N_{\alpha \beta }^1N^{\beta \gamma }=\delta _\alpha ^\gamma ,$$
(11)
with
$$N_{\alpha \beta }^1=\gamma _{ij}Q_\alpha ^iQ_\beta ^j.$$
(12)
Here we have used the convention of symmetrization such that $`A_{(i}B_{j)}=\frac{1}{2}(A_iB_j+A_jB_i)`$. A detailed derivation of $`T_{ij}^k`$ can be found, for example, in GK (21). Therefore, the one-loop VD effective action can be written as
$$iW_{unique}=\frac{1}{2}\mathrm{Trln}_{ij}+\mathrm{Trln}\left(Q_\alpha ^i\frac{\delta \chi ^\beta }{\delta \mathrm{\Phi }^i}\right),$$
(13)
where
$`_{ij}`$ $`=`$ $`{\displaystyle \frac{D}{\delta \mathrm{\Phi }^i}}\left({\displaystyle \frac{\delta S^G}{\delta \mathrm{\Phi }^j}}\right){\displaystyle \frac{\delta \chi ^\alpha }{\delta \mathrm{\Phi }^i}}c_{\alpha \beta }{\displaystyle \frac{\delta \chi ^\beta }{\delta \mathrm{\Phi }^j}}`$ (14)
$`=`$ $`S_{,ij}^G{\displaystyle \frac{\delta \chi ^\alpha }{\delta \mathrm{\Phi }^i}}c_{\alpha \beta }{\displaystyle \frac{\delta \chi ^\beta }{\delta \mathrm{\Phi }^j}}\mathrm{\Gamma }_{ij}^k_k.`$
To calculate the divergent part of $`iW_{unique}`$, we separate the local and the non-local parts of $``$. First we define the Green’s function $`𝒢`$ such that
$$\left\{S_{,ij}^G\left\{\begin{array}{c}k\\ ij\end{array}\right\}_k\frac{\delta \chi ^\alpha }{\delta \mathrm{\Phi }^i}c_{\alpha \beta }\frac{\delta \chi ^\beta }{\delta \mathrm{\Phi }^j}\right\}𝒢^{jl}=\delta _i^l.$$
(15)
Therefore, using $`𝒢^1`$ to represent the above operator in the parenthesis,
$`_{ij}`$ $`=`$ $`𝒢_{ij}^1T_{ij}^k_k`$ (16)
$`=`$ $`𝒢_{il}^1(\delta _j^l+𝒢^{lm}T_{mj}^k_k).`$
The contribution of $``$ to the VD effective action can then be written as
$`{\displaystyle \frac{1}{2}}\mathrm{Trln}`$ $`=`$ $`{\displaystyle \frac{1}{2}}\mathrm{Trln}𝒢^1{\displaystyle \frac{1}{2}}\mathrm{Trln}(1+𝒢T)`$ (17)
$`=`$ $`{\displaystyle \frac{1}{2}}\mathrm{Trln}𝒢^1{\displaystyle \frac{1}{2}}\mathrm{Trln}M+{\displaystyle \frac{1}{4}}\mathrm{Trln}M^2+\mathrm{},`$
where $`M_j^l=𝒢^{lm}T_{mj}^k_k`$. The various traces can be evaluated with the help of the following identities. From Eq. (2),
$`{\displaystyle \frac{\delta }{\delta \mathrm{\Phi }^j}}\left(Q_\alpha ^i{\displaystyle \frac{\delta S^G}{\delta \mathrm{\Phi }^i}}\right)=0`$ (18)
$``$ $`\left({\displaystyle \frac{\delta Q_\alpha ^i}{\delta \mathrm{\Phi }^j}}\right)_i+Q_\alpha ^iS_{,ij}^G=0.`$
From the definition of the operator $`𝒢^1`$ in Eq. (15),
$$S_{,ij}^G=𝒢_{ij}^1+\left\{\begin{array}{c}k\\ ij\end{array}\right\}_k+\frac{\delta \chi ^\alpha }{\delta \mathrm{\Phi }^i}c_{\alpha \beta }\frac{\delta \chi ^\beta }{\delta \mathrm{\Phi }^j}.$$
(19)
If we choose the DeWitt background gauge BD2 (22),
$$\frac{\delta \chi ^\alpha }{\delta \mathrm{\Phi }^i}=c_{}^{1}{}_{}{}^{\alpha \beta }Q_\beta ^j\gamma _{ji},$$
(20)
we can obtain minimal operators CK2 (19) with the appropriate choice of $`c_{\alpha \beta }`$. In the DeWitt gauge,
$$\frac{\delta \chi ^\alpha }{\delta \mathrm{\Phi }^i}c_{\alpha \beta }\frac{\delta \chi ^\beta }{\delta \mathrm{\Phi }^j}=\gamma _{ik}Q_\alpha ^kc_{}^{1}{}_{}{}^{\alpha \beta }Q_\beta ^l\gamma _{lj},$$
(21)
and the ghost operator
$$Q_\alpha ^i\frac{\delta \chi ^\beta }{\delta \mathrm{\Phi }^i}=N_{\alpha \mu }^1c_{}^{1}{}_{}{}^{\mu \beta }.$$
(22)
Therefore, Eq. (18) becomes
$`\left({\displaystyle \frac{\delta Q_\alpha ^i}{\delta \mathrm{\Phi }^j}}\right)_i+Q_\alpha ^i\left(𝒢_{ij}^1+\left\{\begin{array}{c}k\\ ij\end{array}\right\}_k+\gamma _{ik}Q_\mu ^kc_{}^{1}{}_{}{}^{\mu \nu }Q_\nu ^l\gamma _{lj}\right)=0`$ (25)
$``$ $`N_{\alpha \beta }c_{}^{1}{}_{}{}^{\beta \mu }Q_\mu ^i\gamma _{ij}=Q_\alpha ^i𝒢_{ij}^1Q_{\alpha ;j}^i_i.`$ (26)
Multipling both sides by $`𝒢^{jk}N^{\alpha \nu }`$, we have
$$Q_\nu ^i\gamma _{ij}𝒢^{jk}=Q_\alpha ^kN^{\alpha \beta }c_{\beta \nu }Q_{\alpha ;j}^i_i𝒢^{jk}N^{\alpha \beta }c_{\beta \nu }.$$
(27)
This is the needed basic identity. If we multiply both sides by $`N^{\mu \gamma }Q_{\gamma ;k}^l_l`$,
$`Q_\nu ^i\gamma _{ij}𝒢^{jk}Q_{\gamma ;k}^l_lN^{\mu \gamma }`$ $`=`$ $`N^{\mu \gamma }Q_{\gamma ;k}^l_lQ_\alpha ^kN^{\alpha \beta }c_{\beta \nu }N^{\mu \gamma }Q_{\gamma ;k}^i_i𝒢^{kl}Q_{\alpha ;l}^j_jN^{\alpha \beta }c_{\beta \nu }`$ (28)
$`=`$ $`U_{1}^{}{}_{\nu }{}^{\mu }U_{2}^{}{}_{\nu }{}^{\mu },`$
where
$`U_{1}^{}{}_{\nu }{}^{\mu }`$ $``$ $`N^{\mu \gamma }Q_\gamma ^kQ_{\alpha ;k}^l_lN^{\alpha \beta }c_{\beta \nu },`$ (29)
$`U_{2}^{}{}_{\nu }{}^{\mu }`$ $``$ $`N^{\mu \gamma }Q_{\gamma ;k}^i_i𝒢^{kl}Q_{\alpha ;l}^j_jN^{\alpha \beta }c_{\beta \nu }.`$ (30)
We have also used the fact that
$`Q_\alpha ^iQ_{\gamma ;i}^j_j`$ $`=`$ $`Q_\alpha ^i[(Q_\gamma ^j_j)_{;i}Q_\gamma ^j_{j;i}]`$ (33)
$`=`$ $`Q_\alpha ^iQ_\gamma ^j\left[S_{,ij}^G\left\{\begin{array}{c}k\\ ij\end{array}\right\}_k\right]`$
$`=`$ $`Q_\gamma ^iQ_{\alpha ,i}^j_j,`$ (34)
is symmetric with respect to the indices $`\alpha `$ and $`\gamma `$. We can obtain yet another identity by multipling the basic identity, Eq. (27), with $`\gamma _{kl}Q_\gamma ^l`$,
$$Q_\nu ^i\gamma _{ij}𝒢^{jk}\gamma _{kl}Q_\gamma ^l=c_{\nu \gamma }Q_{\alpha ;j}^i_i𝒢^{jk}N^{\alpha \beta }c_{\beta \nu }\gamma _{kl}Q_\gamma ^l.$$
(35)
Applying again the basic identity, Eq. (35) becomes
$`Q_\nu ^i\gamma _{ik}𝒢^{kl}\gamma _{lj}Q_\gamma ^j`$ $`=`$ $`c_{\nu \gamma }Q_{\alpha ;j}^i_iN^{\alpha \beta }c_{\beta \nu }(Q_\mu ^jN^{\mu \rho }c_{\rho \gamma }Q_{\mu ;k}^l_l𝒢^{jk}N^{\mu \rho }c_{\rho \gamma })`$ (36)
$`=`$ $`c_{\nu \beta }(\delta _\gamma ^\beta U_{1}^{}{}_{\gamma }{}^{\beta }+U_{2}^{}{}_{\gamma }{}^{\beta }).`$
With these identities we are in a position to evaluate the various traces in Eq. (17). First,
$`\mathrm{Tr}M`$ $`=`$ $`𝒢^{ij}T_{ij}^k_k`$ (37)
$`=`$ $`2(U_{1}^{}{}_{\alpha }{}^{\alpha }U_{2}^{}{}_{\alpha }{}^{\alpha })+U_{1}^{}{}_{\beta }{}^{\alpha }(\delta _\alpha ^\beta U_{1}^{}{}_{\alpha }{}^{\beta }+U_{2}^{}{}_{\alpha }{}^{\beta })`$
$`=`$ $`\mathrm{Tr}U_1+2\mathrm{T}\mathrm{r}U_2\mathrm{Tr}U_1^2+\mathrm{Tr}U_1U_2,`$
where we have made use of both the identities in Eqs. (28) and (36). Similarly, we can evaluate the traces of higher powers of $`M`$.
$`\mathrm{Tr}M^2`$ $`=`$ $`\mathrm{Tr}U_1^2+2\mathrm{T}\mathrm{r}U_2+2\mathrm{T}\mathrm{r}U_1^32\mathrm{T}\mathrm{r}U_1U_2+\mathrm{Tr}U_1^46\mathrm{T}\mathrm{r}U_1^2U_2+4\mathrm{T}\mathrm{r}U_2^2`$ (38)
$`2\mathrm{T}\mathrm{r}U_1^3U_2+4\mathrm{T}\mathrm{r}U_1U_2^2+O(^6),`$
$`\mathrm{Tr}M^3`$ $`=`$ $`2\mathrm{T}\mathrm{r}U_1^33\mathrm{T}\mathrm{r}U_1U_26\mathrm{T}\mathrm{r}U_1^2U_2+6\mathrm{T}\mathrm{r}U_2^2`$ (39)
$`3\mathrm{T}\mathrm{r}U_1^5+9\mathrm{T}\mathrm{r}U_1^3U_23\mathrm{T}\mathrm{r}U_1U_2^2+O(^6),`$
$`\mathrm{Tr}M^4`$ $`=`$ $`\mathrm{Tr}U_1^4+2\mathrm{T}\mathrm{r}U_2^24\mathrm{T}\mathrm{r}U_1^5+16\mathrm{T}\mathrm{r}U_1^3U_212\mathrm{T}\mathrm{r}U_1U_2^2+O(^6),`$ (40)
$`\mathrm{Tr}M^5`$ $`=`$ $`\mathrm{Tr}U_1^5+5\mathrm{T}\mathrm{r}U_1^3U_25\mathrm{T}\mathrm{r}U_1U_2^2+O(^6).`$ (41)
Note that we have expanded the various expressions only up to the fifth power of the first functional derivative $`_i`$ of the action. To calculate the divergent part of the VD effective action, one needs only terms up to some power of $`_i`$, depending on the dimensionality of the spacetime that one is considering.
Finally, the VD effective action in Eq. (13) can be expanded in terms of the various traces of the operators $`U_1`$ and $`U_2`$,
$`iW_{unique}`$ $`=`$ $`{\displaystyle \frac{1}{2}}\mathrm{Trln}+\mathrm{Trln}N^1`$ (42)
$`=`$ $`{\displaystyle \frac{1}{2}}\mathrm{Trln}𝒢^1+\mathrm{Trln}N^1`$
$`+{\displaystyle \frac{1}{2}}\mathrm{Tr}U_1+{\displaystyle \frac{1}{4}}\mathrm{Tr}U_1^2{\displaystyle \frac{1}{2}}\mathrm{Tr}U_2+{\displaystyle \frac{1}{6}}\mathrm{Tr}U_1^3{\displaystyle \frac{1}{2}}\mathrm{Tr}U_1U_2+{\displaystyle \frac{1}{8}}\mathrm{Tr}U_1^4`$
$`{\displaystyle \frac{1}{2}}\mathrm{Tr}U_1^2U_2+{\displaystyle \frac{1}{4}}\mathrm{Tr}U_2^2+{\displaystyle \frac{1}{10}}\mathrm{Tr}U_1^5{\displaystyle \frac{1}{2}}\mathrm{Tr}U_1^3U_2+{\displaystyle \frac{1}{2}}\mathrm{Tr}U_1U_2^2+O(^6).`$
Using this expression we evaluate the divergent part of the VD effective action for Einstein gravity on Kaluza-Klein spaces $`M^4\times S^N`$ in the following section.
## III Einstein gravity on Kaluza-Klein spaces $`M^4\times S^N`$
Here we shall apply the formalism set up in the last section to the case of Einstein gravity. Because such calculations are very tedious to carry out for general spacetimes (see CK1 (16)), we restrict ourselves to Kaluza-Klein backgrounds of the form $`M^4\times S^N`$.
### III.1 Einstein gravity
We start with the n-dimensional Einstein-Hilbert action,
$$S^G=d^nx\sqrt{g}(R2\mathrm{\Lambda }),$$
(43)
where $`\mathrm{\Lambda }`$ is the cosmological constant. The first functional derivative
$$^{\mu \nu x}=\frac{\delta S^G}{\delta g_{\mu \nu }(x)}=(R^{\mu \nu }\frac{1}{2}g^{\mu \nu }(R2\mathrm{\Lambda })).$$
(44)
The gauge symmetry is the general coordinate invariance,
$$\delta g_{\mu \nu }(x)=_\mu ϵ_\nu +_\nu ϵ_\mu =Q_{\mu \nu x,\alpha y}ϵ^\alpha (y),$$
(45)
where $`ϵ_\mu (x)`$ is the gauge transformation parameter, $`_\mu `$ is the ordinary covariant derivative, and
$$Q_{\mu \nu x,\alpha y}=(g_{\mu \alpha }_\nu +g_{\nu \alpha }_\mu )\delta ^n(xy).$$
(46)
Because of this gauge symmetry, the second functional derivative of the action, $`S_{,ij}^G`$, is a singular operator.
$`{\displaystyle \frac{\delta S^G}{\delta g_{\mu \nu }(x)\delta g_{\alpha \beta }(y)}}`$ (47)
$`=`$ $`C^{\mu \nu ,\rho \sigma }[\delta _{\rho \sigma }^{\alpha \beta }\mathrm{}\delta _{(\rho }^{(\alpha }_{\sigma )}^{\beta )}\delta _{(\rho }^{(\alpha }^{\beta )}_{\sigma )}+g^{\alpha \beta }_{(\rho }_{\sigma )}+2R_{\rho \sigma }^{(\alpha \beta )}+2\delta _{(\rho }^{(\alpha }R_{\sigma )}^{\beta )}`$
$`g^{\alpha \beta }R_{\rho \sigma }{\displaystyle \frac{2}{n2}}g_{\rho \sigma }R^{\alpha \beta }+{\displaystyle \frac{1}{n2}}g_{\rho \sigma }g^{\alpha \beta }R(R2\mathrm{\Lambda })\delta _{\rho \sigma }^{\alpha \beta }]\delta ^n(xy)/\sqrt{g},`$
where
$`\delta _{\rho \sigma }^{\alpha \beta }`$ $`=`$ $`\delta _{(\rho }^\alpha \delta _{\sigma )}^\beta ,`$ (48)
$`C^{\mu \nu ,\alpha \beta }`$ $`=`$ $`{\displaystyle \frac{1}{4}}(g^{\mu \alpha }g^{\nu \beta }+g^{\mu \beta }g^{\nu \alpha }g^{\mu \nu }g^{\alpha \beta }).`$ (49)
To invert (47) we need to pick a gauge, and again as in the last section, we choose the DeWitt gauge. The DeWitt gauge requires a metric for the configuration space of $`g_{\mu \nu }(x)`$. The one given by Vilkovisky is GV (6)
$$\gamma ^{\mu \nu x,\alpha \beta y}=C^{\mu \nu ,\alpha \beta }\sqrt{g}\delta ^n(xy),$$
(50)
in which $`C^{\mu \nu ,\alpha \beta }`$ is exactly the same as the factor in front of the operator $`S_{,ij}^G`$. With this metric $`\gamma _{ij}`$ and the choice of $`c_{\mu \nu }=\delta _{\mu \nu }`$, the DeWitt gauge becomes the harmonic gauge and the corresponding graviton operator in Eq. (4) is minimal CK2 (19),
$`F_{\mu \nu }^{\alpha \beta }`$ $`=`$ $`\delta _{\mu \nu }^{\alpha \beta }\mathrm{}+2R_{\mu \nu }^{(\alpha \beta )}+2\delta _{(\mu }^{(\alpha }R_{\nu )}^{\beta )}g^{\alpha \beta }R_{\mu \nu }`$ (51)
$`{\displaystyle \frac{2}{n2}}g_{\mu \nu }R^{\alpha \beta }+{\displaystyle \frac{1}{n2}}g_{\mu \nu }g^{\alpha \beta }R(R2\mathrm{\Lambda })\delta _{\mu \nu }^{\alpha \beta }.`$
Note that for simplicity we have left out the spacetime coordinate indices, and we shall do so in the subsequent discussion. The ghost operator in Eq. (22) is now simply,
$$N_{\mu \nu }=(g_{\mu \nu }\mathrm{}+R_{\mu \nu }).$$
(52)
To find the expression for the local operator $`𝒢^1`$ in the VD effective action, we need the Christoffel symbol $`\left\{\begin{array}{c}k\\ ij\end{array}\right\}`$. From the metric $`\gamma ^{\mu \nu ,\alpha \beta }`$ above,
$$\left\{\begin{array}{c}g_{\rho \sigma }\\ g_{\mu \nu }g_{\alpha \beta }\end{array}\right\}=\frac{1}{4}g^{\mu \nu }\delta _{\rho \sigma }^{\alpha \beta }+\frac{1}{4}g^{\alpha \beta }\delta _{\rho \sigma }^{\mu \nu }\frac{1}{2}\delta _{\rho \sigma }^{\mu (\alpha }g^{\beta )\nu }\frac{1}{2}\delta _{\rho \sigma }^{\nu (\alpha }g^{\beta )\mu }+\frac{1}{n2}g_{\rho \sigma }C^{\mu \nu ,\alpha \beta }.$$
(53)
The operator $`𝒢^1`$ has the form
$$𝒢_{}^{1}{}_{\mu \nu }{}^{\alpha \beta }=(\delta _{\mu \nu }^{\alpha \beta }\mathrm{}+P_{\mu \nu }^{\alpha \beta }),$$
(54)
where
$`P_{\mu \nu }^{\alpha \beta }`$ $`=`$ $`2R_{\mu \nu }^{(\alpha \beta )}{\displaystyle \frac{1}{2}}g^{\alpha \beta }R_{\mu \nu }{\displaystyle \frac{1}{n2}}g_{\mu \nu }R^{\alpha \beta }`$ (55)
$`+{\displaystyle \frac{1}{n2}}(\delta _{\mu \nu }^{\alpha \beta }+{\displaystyle \frac{1}{2}}g_{\mu \nu }g^{\alpha \beta })R{\displaystyle \frac{n}{2(n2)}}(R2\mathrm{\Lambda })\delta _{\mu \nu }^{\alpha \beta }.`$
Note that we have left out an overall factor $`C^{\mu \nu ,\alpha \beta }`$ which is irrelevant in this particular calculation.
Next we would like to find the expressions for the operators $`U_{1}^{}{}_{\nu }{}^{\mu }`$ and $`U_{2}^{}{}_{\nu }{}^{\mu }`$. To do so we need to use the following equations.
$`(Q_{\alpha \beta ,\mu })^{;\rho \sigma }^{\alpha \beta }`$ (56)
$`=`$ $`[R^{\nu (\rho }\delta _\mu ^{\sigma )}_\nu +R_\mu ^{(\rho }^{\sigma )}{\displaystyle \frac{1}{2}}g^{\rho \sigma }R_\mu ^\nu _\nu +{\displaystyle \frac{1}{2}}(R{\displaystyle \frac{2n}{n2}}\mathrm{\Lambda })\delta _\mu ^{(\rho }^{\sigma )}`$
$`+{\displaystyle \frac{1}{2}}(R^{\rho \sigma }{\displaystyle \frac{1}{2}}g^{\rho \sigma }R)_\mu +{\displaystyle \frac{n}{2(n2)}}\mathrm{\Lambda }g^{\rho \sigma }_\mu +(_\mu R^{\rho \sigma }){\displaystyle \frac{1}{2}}g^{\rho \sigma }(_\mu R)],`$
and
$`Q_{\rho \sigma ,\mu }(Q_{\alpha \beta ,\nu })^{;\rho \sigma }^{\alpha \beta }`$ (57)
$`=`$ $`g_{\mu \nu }R^{\alpha \beta }_\alpha _\beta R_{\mu \nu }\mathrm{}{\displaystyle \frac{1}{2}}\left(R{\displaystyle \frac{2n}{n2}}\mathrm{\Lambda }\right)(g_{\mu \nu }\mathrm{}+R_{\mu \nu })`$
$`R_\mu ^\alpha R_{\alpha \nu }+R^{\alpha \beta }R_{\mu \alpha \nu \beta }(^\alpha R_{\mu \nu })_\alpha +(_\mu R_\nu ^\alpha )_\alpha (_\nu R_\mu ^\alpha )_\alpha .`$
Now, the operator $`U_{1}^{}{}_{\nu }{}^{\mu }`$ and $`U_{2}^{}{}_{\nu }{}^{\mu }`$ are given by
$`U_{1}^{}{}_{\nu }{}^{\mu }`$ $`=`$ $`N_\alpha ^\mu [\delta _\beta ^\alpha R^{\sigma \lambda }_\sigma _\lambda R_\beta ^\alpha \mathrm{}{\displaystyle \frac{1}{2}}(R{\displaystyle \frac{2n}{n2}}\mathrm{\Lambda })(\delta _\beta ^\alpha \mathrm{}+R_\beta ^\alpha )`$ (58)
$`R_\lambda ^\alpha R_\beta ^\lambda +R^{\sigma \lambda }R_{\sigma \beta \lambda }^\alpha (^\lambda R_\beta ^\alpha )_\lambda +(^\alpha R_\beta ^\lambda )_\lambda (_\beta R^{\alpha \lambda })_\lambda ]N^\beta _\nu `$
$`U_{2}^{}{}_{\nu }{}^{\mu }`$ $`=`$ $`N_\alpha ^\nu \left\{{\displaystyle \frac{1}{2}}\left[(^\alpha R^{\sigma \lambda })(^\sigma R^{\lambda \alpha })(^\lambda R^{\sigma \alpha })\right]+\left[D^{\sigma \lambda ,\alpha \mu }\left(R{\displaystyle \frac{2n}{n2}}\mathrm{\Lambda }\right)C^{\sigma \lambda ,\alpha \mu }\right]_\mu \right\}`$ (59)
$`𝒢_{\sigma \lambda ,\omega ϵ}\left\{(_\beta R^{ϵ\omega }){\displaystyle \frac{1}{2}}g^{ϵ\omega }(_\beta R)\left[D_\beta ^{\omega ϵ,\gamma }\left(R{\displaystyle \frac{2n}{n2}}\mathrm{\Lambda }\right)C_\beta ^{\omega ϵ,\gamma }\right]_\gamma \right\}N_\mu ^\beta ,`$
where $`N_\alpha ^\mu `$ is the ghost Green’s function
$$(\delta _\alpha ^\mu \mathrm{}+R_\alpha ^\mu )N_\nu ^\alpha =\delta _\nu ^\mu ,$$
(60)
and
$$D^{\sigma \lambda ,\alpha \mu }\frac{1}{2}(g^{\sigma \lambda }R^{\alpha \mu }g^{\alpha \mu }R^{\sigma \lambda }+g^{\alpha \sigma }R^{\lambda \mu }g^{\lambda \mu }R^{\alpha \sigma }+g^{\alpha \lambda }R^{\sigma \mu }g^{\sigma \mu }R^{\alpha \lambda }.$$
(61)
We have written down the expressions for the operators $`𝒢^1`$, $`N^1`$, $`U_1`$, and $`U_2`$, which are present in the VD effective action. In principle, the divergent part of the VD effective action can now be evaluated by working out the traces of different combinations of these operators. However, the algebra gets exceedingly tedious as one goes to higher dimensions. In CK1 (16), we have considered the case of $`n=6`$, but going to larger values of $`n`$ seems to be impossible. In the following analysis, we therefore restrict ourselves to the specific cases of Kaluza-Klein spaces of the form $`M^4\times S^N`$.
### III.2 Eigenvalues of various operators on $`M^4\times S^N`$
Here we work out the eigenvalues of the operators $`N^1`$, $`𝒢^1`$, $`U_1`$, and $`U_2`$. To distinguish between external spacetime and internal space indices, we use Greek letters for external spacetime $`M^4`$ and Latin letters for internal space $`S^N`$. Since the traces of these operators are divergent in general, a regularization scheme is needed. We adopt the method of dimensional regularization by taking the external spacetime dimension to be a general $`d`$ and we take $`d4ϵ`$ at the end to extract the divergent parts. Hence in this subsection we first assume the spacetime dimension to be $`M^d\times S^N`$.
First we consider the ghost operator $`N^1`$:
$`N_{}^{1}{}_{\nu }{}^{\mu }`$ $`=`$ $`\delta _\nu ^\mu (\mathrm{}_d+\mathrm{}_N),`$ (62)
$`N_{}^{1}{}_{b}{}^{a}`$ $`=`$ $`\delta _b^a[(\mathrm{}_d+\mathrm{}_N)+R_b^a]`$ (63)
$`=`$ $`\delta _b^a\left[(\mathrm{}_d+\mathrm{}_N)+{\displaystyle \frac{1}{r^2}}(N1)\right],`$
where $`\mathrm{}_d`$ is the d’Alembertian on $`M^d`$, while $`\mathrm{}_N`$, $`r`$, and $`R_b^a`$ are the Laplacian, the radius, and the Ricci tensor on $`S^N`$, respectively. The off-diagonal terms $`N_{}^{1}{}_{a}{}^{\mu }`$ and $`N_{}^{1}{}_{\mu }{}^{a}`$ vanish.
The eigenvector $`\eta ^\nu `$ of $`N_{}^{1}{}_{\nu }{}^{\mu }`$ is a vector with $`d`$ components on $`M^d`$ and a scalar on $`S^N`$. Consequently, RO (18),
$$N_{}^{1}{}_{\nu }{}^{\mu }\eta ^\nu =(k^2\mathrm{\Lambda }_l)\eta ^\mu ,$$
(64)
where $`k^\mu `$ is the momentum and
$$\mathrm{\Lambda }_l=\frac{l(l+N1)}{r^2},$$
(65)
for $`l=0,1,2,\mathrm{}`$ is the scalar Laplacian eigenvalue on $`S^N`$ with degeneracy
$$D_l^{(s)}(N)=\frac{(2l+N1)(l+N2)!}{l!(N1)!}.$$
(66)
The eigenvectors of $`N_{}^{1}{}_{b}{}^{a}`$ are $`\eta _{}^{b}{}_{}{}^{\mathrm{T}}`$ and $`\eta _{}^{b}{}_{}{}^{\mathrm{L}}`$. $`\eta _{}^{b}{}_{}{}^{\mathrm{T}}`$ is the transverse part of $`\eta ^b`$, which is a scalar on $`M^d`$ and a vector on $`S^N`$, with
$$(_N)_b\eta _{}^{b}{}_{}{}^{\mathrm{T}}=0.$$
(67)
The eigenvalue of $`\eta _{}^{b}{}_{}{}^{\mathrm{T}}`$ is
$$N_{}^{1}{}_{b}{}^{a}\eta _{}^{b}{}_{}{}^{\mathrm{T}}=\left[k^2\mathrm{\Lambda }_l\frac{N}{r^2}\right]\eta _{}^{a}{}_{}{}^{\mathrm{T}},$$
(68)
for $`l=1,2,\mathrm{}`$ with degeneracy
$$D_l^{(v)}(N)=\frac{l(l+N1)(2l+N1)(l+N3)!}{(N2)!(l+1)!}.$$
(69)
And the eigenvalue of $`\eta _{}^{b}{}_{}{}^{\mathrm{L}}`$, the longitudinal part of $`\eta ^b`$, is
$$N_{}^{1}{}_{b}{}^{a}\eta _{}^{b}{}_{}{}^{\mathrm{L}}=\left(k^2\mathrm{\Lambda }_l\frac{2}{r^2}(N1)\right)\eta _{}^{a}{}_{}{}^{\mathrm{L}},$$
(70)
for $`l=1,2,\mathrm{}`$, with degeneracy $`D_l^{(s)}(N)`$.
For the operator $`𝒢^1`$, the eigenfunctions are symmetric tensors. First, $`\eta _{\alpha \beta }`$, which is a symmetric tensor on $`M^d`$ and a scalar on $`S^N`$, can be decomposed RO (18) into the transverse-traceless (TT) part $`\eta _{\alpha \beta }^{\mathrm{TT}}`$, the longitudinal-transverse-traceless (LTT) part $`\eta _{\alpha \beta }^{\mathrm{LTT}}`$, the longitudinal-longitudinal-traceless (LLT) part $`\eta _{\alpha \beta }^{\mathrm{LLT}}`$, and the trace (Tr) part $`\eta _{\alpha \beta }^{\mathrm{Tr}}`$. $`\eta _{\alpha \beta }^{\mathrm{TT}}`$ has $`(d2)(d+1)/2`$ components on $`M^d`$,
$$𝒢_{}^{1}{}_{\mu \nu }{}^{\alpha \beta }\eta _{\alpha \beta }^{\mathrm{TT}}=\left(k^2\mathrm{\Lambda }_l+\frac{1}{2r^2}N(N1)\frac{N+d}{N+d2}\mathrm{\Lambda }\right)\eta _{\mu \nu }^{\mathrm{TT}}.$$
(71)
$`\eta _{\alpha \beta }^{\mathrm{LTT}}`$ has $`(d1)`$ components on $`M^d`$,
$$𝒢_{}^{1}{}_{\mu \nu }{}^{\alpha \beta }\eta _{\alpha \beta }^{\mathrm{LTT}}=\left(k^2\mathrm{\Lambda }_l+\frac{1}{2r^2}(N1)\frac{N+d}{N+d2}\mathrm{\Lambda }\right)\eta _{\mu \nu }^{\mathrm{LTT}}.$$
(72)
$`\eta _{\alpha \beta }^{\mathrm{LLT}}`$ has 1 component on $`M^d`$,
$$𝒢_{}^{1}{}_{\mu \nu }{}^{\alpha \beta }\eta _{\alpha \beta }^{\mathrm{LLT}}=\left(k^2\mathrm{\Lambda }_l+\frac{1}{2r^2}(N1)\frac{N+d}{N+d2}\mathrm{\Lambda }\right)\eta _{\mu \nu }^{\mathrm{LLT}}.$$
(73)
$`\eta _{\alpha \beta }^{\mathrm{Tr}}`$ also has only 1 component on $`M^d`$,
$$𝒢_{}^{1}{}_{\mu \nu }{}^{\alpha \beta }\eta _{\alpha \beta }^{\mathrm{Tr}}=\left(k^2\mathrm{\Lambda }_l+\frac{1}{2r^2}\frac{N(N1)(N2)}{N+d2}\frac{N+d}{N+d2}\mathrm{\Lambda }\right)\eta _{\mu \nu }^{\mathrm{Tr}}.$$
(74)
They all have degeneracy $`D_l^{(s)}(N)`$ with $`l=0,1,2,\mathrm{}`$.
Similarly, $`\eta _{ab}`$ is a scalar on $`M^d`$ and a symmetric tensor on $`S^N`$.
$$𝒢_{}^{1}{}_{ab}{}^{cd}\eta _{cd}^{\mathrm{TT}}=\left(k^2\mathrm{\Lambda }_l+\frac{1}{2r^2}N(N1)\frac{N+d}{N+d2}\mathrm{\Lambda }\right)\eta _{ab}^{\mathrm{TT}},$$
(75)
for $`l=2,3,\mathrm{}`$, with degeneracy
$$D_l^{(t)}(N)=\frac{(N+1)(N2)(l+N)(l1)(2l+N1)(l+N3)!}{2(N1)!(l+1)!}.$$
(76)
$$𝒢_{}^{1}{}_{ab}{}^{cd}\eta _{cd}^{\mathrm{LTT}}=\left(k^2\mathrm{\Lambda }_l+\frac{1}{2r^2}N(N3)\frac{N+d}{N+d2}\mathrm{\Lambda }\right)\eta _{ab}^{\mathrm{LTT}},$$
(77)
for $`l=2,3,\mathrm{}`$, with degeneracy $`D_l^{(v)}(N)`$.
$$𝒢_{}^{1}{}_{ab}{}^{cd}\eta _{cd}^{\mathrm{LLT}}=\left(k^2\mathrm{\Lambda }_l+\frac{1}{2r^2}(N1)(N4)\frac{N+d}{N+d2}\mathrm{\Lambda }\right)\eta _{ab}^{\mathrm{LLT}},$$
(78)
for $`l=2,3,\mathrm{}`$, with degeneracy $`D_l^{(s)}(N)`$.
$`𝒢_{}^{1}{}_{ab}{}^{cd}\eta _{cd}^{\mathrm{Tr}}`$ $`=`$ $`(k^2\mathrm{\Lambda }_l+{\displaystyle \frac{1}{2r^2}}{\displaystyle \frac{N^3+N^2(2d7)2N(3d7)+4(d2)}{N+d2}}`$ (79)
$`{\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda })\eta ^{\mathrm{Tr}}_{ab},`$
for $`l=0,1,2,\mathrm{}`$, with degeneracy $`D_l^{(s)}(N)`$. We also have the off-diagonal terms,
$`𝒢_{}^{1}{}_{\mu \nu }{}^{ab}\eta _{ab}^{\mathrm{Tr}}`$ $`=`$ $`{\displaystyle \frac{(N1)(N2)d}{2r^2(N+d2)}}\eta _{\mu \nu }^{\mathrm{Tr}},`$ (80)
$`𝒢_{}^{1}{}_{ab}{}^{\mu \nu }\eta _{\mu \nu }^{\mathrm{Tr}}`$ $`=`$ $`{\displaystyle \frac{N(N1)(d2)}{2r^2(N+d2)}}\eta _{ab}^{\mathrm{Tr}}.`$ (81)
The eigenfunctions $`\eta _{\mu \nu }^{\mathrm{Tr}}`$ and $`\eta _{ab}^{\mathrm{Tr}}`$ are seen to be coupled together, and $`𝒢^1`$ forms a $`2\times 2`$ matrix $`𝒢_{\mathrm{Tr}}^1`$ in the subspace of these eigenfunctions. The determinant of this matrix is
$`\mathrm{det}𝒢_{\mathrm{Tr}}^1`$ $`=`$ $`\left[k^2\mathrm{\Lambda }_l+{\displaystyle \frac{1}{2r^2}}{\displaystyle \frac{N(N1)(N2)}{N+d2}}{\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda }\right]`$ (82)
$`\left[k^2\mathrm{\Lambda }_l+{\displaystyle \frac{1}{2r^2}}{\displaystyle \frac{N^3+N^2(2d7)2N(3d7)+4(d2)}{N+d2}}{\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda }\right]`$
$`+{\displaystyle \frac{N(N1)^2(N2)d(d2)}{4r^4(N+d2)^2}}.`$
There are also eigenfunctions $`\eta _{\nu b}^{\mathrm{T},\mathrm{T}}`$, $`\eta _{\nu b}^{\mathrm{T},\mathrm{L}}`$, $`\eta _{\nu b}^{\mathrm{L},\mathrm{T}}`$, and $`\eta _{\nu b}^{\mathrm{L},\mathrm{L}}`$. $`\eta _{\nu b}^{\mathrm{T},\mathrm{T}}`$ is a transverse vector on $`M^d`$ as well as on $`S^N`$, and so on. $`\eta _{\nu b}^{\mathrm{T},\mathrm{T}}`$ has $`(d1)`$ components on $`M^d`$,
$$𝒢_{}^{1}{}_{\mu a}{}^{\nu b}\eta _{\nu b}^{\mathrm{T},\mathrm{T}}=\frac{1}{2}\left(k^2\mathrm{\Lambda }_l+\frac{1}{2r^2}(N+1)(N2)\frac{N+d}{N+d2}\mathrm{\Lambda }\right)\eta _{\mu a}^{\mathrm{T},\mathrm{T}},$$
(83)
for $`l=1,2,\mathrm{}`$, with $`D_l^{(v)}(N)`$. $`\eta _{\nu b}^{\mathrm{T},\mathrm{L}}`$ has $`(d1)`$ components on $`M^d`$,
$$𝒢_{}^{1}{}_{\mu a}{}^{\nu b}\eta _{\nu b}^{\mathrm{T},\mathrm{L}}=\frac{1}{2}\left(k^2\mathrm{\Lambda }_l+\frac{1}{2r^2}(N1)(N2)\frac{N+d}{N+d2}\mathrm{\Lambda }\right)\eta _{\mu a}^{\mathrm{T},\mathrm{L}},$$
(84)
for $`l=1,2,\mathrm{}`$, with $`D_l^{(s)}(N)`$. $`\eta _{\nu b}^{\mathrm{L},\mathrm{T}}`$ has 1 components on $`M^d`$,
$$𝒢_{}^{1}{}_{\mu a}{}^{\nu b}\eta _{\nu b}^{\mathrm{L},\mathrm{T}}=\frac{1}{2}\left(k^2\mathrm{\Lambda }_l+\frac{1}{2r^2}(N+1)(N2)\frac{N+d}{N+d2}\mathrm{\Lambda }\right)\eta _{\mu a}^{\mathrm{L},\mathrm{T}},$$
(85)
for $`l=1,2,\mathrm{}`$, with $`D_l^{(v)}(N)`$. $`\eta _{\nu b}^{\mathrm{L},\mathrm{L}}`$ has 1 components on $`M^d`$,
$$𝒢_{}^{1}{}_{\mu a}{}^{\nu b}\eta _{\nu b}^{\mathrm{L},\mathrm{L}}=\frac{1}{2}\left(k^2\mathrm{\Lambda }_l+\frac{1}{2r^2}(N1)(N2)\frac{N+d}{N+d2}\mathrm{\Lambda }\right)\eta _{\mu a}^{\mathrm{L},\mathrm{L}},$$
(86)
for $`l=1,2,\mathrm{}`$, with $`D_l^{(s)}(N)`$.
For $`U_{1}^{}{}_{\nu }{}^{\mu }`$, the eigenfunction $`\eta ^\nu `$ is a vector on $`M^d`$ and a scalar on $`S^N`$. $`\eta ^\nu `$ has $`d`$ components on $`M^d`$,
$`U_{1}^{}{}_{\nu }{}^{\mu }\eta ^\nu `$ $`=`$ $`[({\displaystyle \frac{1}{2r^2}}N(N1){\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda })(k^2)`$ (87)
$`({\displaystyle \frac{1}{2r^2}}(N1)(N2){\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda })(\mathrm{\Lambda }_l)\left]\right({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l}})^2\eta ^\mu ,`$
for $`l=0,1,2,\mathrm{}`$, with degeneracy $`D_l^{(s)}(N)`$.
For $`U_{1}^{}{}_{b}{}^{a}`$, the eigenfunctions are $`\eta _{}^{b}{}_{}{}^{\mathrm{T}}`$ and $`\eta _{}^{b}{}_{}{}^{\mathrm{L}}`$.
$`U_{1}^{}{}_{b}{}^{a}\eta _{}^{b}{}_{}{}^{\mathrm{T}}`$ $`=`$ $`[({\displaystyle \frac{1}{2r^2}}(N+2)(N1){\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda })(k^2)`$ (88)
$`({\displaystyle \frac{1}{2r^2}}N(N1){\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda })(\mathrm{\Lambda }_l+{\displaystyle \frac{N}{r^2}})\left]\right({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l\frac{N}{r^2}}})^2\eta ^a^\mathrm{T},`$
for $`l=1,2,\mathrm{}`$, with degeneracy $`D_l^{(v)}(N)`$. And
$`U_{1}^{}{}_{b}{}^{a}\eta _{}^{b}{}_{}{}^{\mathrm{L}}`$ $`=`$ $`[({\displaystyle \frac{1}{2r^2}}(N+2)(N1){\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda })(k^2)`$ (89)
$`({\displaystyle \frac{1}{2r^2}}N(N1){\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda })(\mathrm{\Lambda }_l+{\displaystyle \frac{2}{r^2}}(N1))]`$
$`\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l\frac{2}{r^2}(N1)}}\right)^2\eta _{}^{a}{}_{}{}^{\mathrm{L}},`$
for $`l=1,2,\mathrm{}`$, with degeneracy $`D_l^{(s)}(N)`$.
The eigenvalues of $`U_2`$ are more complicated. After lengthy calculations, we obtained the following results. For the eigenfunction $`\eta _{}^{\nu }{}_{}{}^{\mathrm{T}}`$, which has $`(d1)`$ components on $`M^d`$,
$`U_{2}^{}{}_{\nu }{}^{\mu }\eta _{}^{\nu }{}_{}{}^{\mathrm{T}}`$ (90)
$`=`$ $`[({\displaystyle \frac{1}{2r^2}}N(N1){\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda })^2(k^2)\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l+\frac{1}{2r^2}N(N1)\frac{N+d}{N+d2}\mathrm{\Lambda }}}\right)`$
$`\left({\displaystyle \frac{1}{2r^2}}(N1)(N2){\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda }\right)^2(\mathrm{\Lambda }_l)`$
$`\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l+\frac{1}{2r^2}(N1)(N2)\frac{N+d}{N+d2}\mathrm{\Lambda }}}\right)\left]\right({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l}})^2\eta ^\mu ^\mathrm{T},`$
for $`l=0,1,2,\mathrm{}`$, with degeneracy $`D_l^{(s)}(N)`$. For eigenfunction $`\eta _{}^{b}{}_{}{}^{\mathrm{T}}`$, which is a scalar on $`M^d`$,
$`U_{2}^{}{}_{b}{}^{a}\eta _{}^{b}{}_{}{}^{\mathrm{T}}`$ (91)
$`=`$ $`[({\displaystyle \frac{1}{2r^2}}(N+2)(N1){\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda })^2(k^2)\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l+\frac{1}{2r^2}(N+1)(N2)\frac{N+d}{N+d2}\mathrm{\Lambda }}}\right)`$
$`\left({\displaystyle \frac{1}{2r^2}}N(N1){\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda }\right)^2\left(\mathrm{\Lambda }_l+{\displaystyle \frac{N}{r^2}}\right)`$
$`\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l+\frac{1}{2r^2}N(N3)\frac{N+d}{N+d2}\mathrm{\Lambda }}}\right)\left]\right({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l\frac{N}{r^2}}})^2\eta ^a^\mathrm{T},`$
for $`l=1,2,\mathrm{}`$, with degeneracy $`D_l^{(v)}(N)`$.
For the eigenfunctions $`\eta _{}^{\nu }{}_{}{}^{\mathrm{L}}`$ and $`\eta _{}^{b}{}_{}{}^{\mathrm{L}}`$, they are coupled together. For $`l=0`$,
$`U_{2}^{}{}_{\nu }{}^{\mu }\eta _{}^{\nu }{}_{}{}^{\mathrm{L}}`$
$`=`$ $`[2\left({\displaystyle \frac{d1}{d}}\right)({\displaystyle \frac{1}{2r^2}}N(N1){\displaystyle \frac{N+d}{N+d2}})^2\left({\displaystyle \frac{1}{k^2}}\right)\left({\displaystyle \frac{1}{k^2+\frac{1}{2r^2}N(N1)\frac{N+d}{N+d2}\mathrm{\Lambda }}}\right)`$
$`\left({\displaystyle \frac{d2}{d}}\right)\left({\displaystyle \frac{1}{2r^2}}N(N1){\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda }\right)\left({\displaystyle \frac{1}{2r^2}}{\displaystyle \frac{N(N1)(N2)}{N+d2}}{\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda }\right)`$
$`\left(k^2+{\displaystyle \frac{1}{2r^2}}{\displaystyle \frac{N^3+N^2(2d7)2N(3d7)+4(d2)}{N+d2}}{\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda }\right)\left({\displaystyle \frac{1}{k^2}}\right){\displaystyle \frac{1}{\mathrm{det}𝒢_{l=0}^1}}`$
$`{\displaystyle \frac{N(N1)^2(N2)(d2)^2}{4r^4(N+d2)^2}}\left({\displaystyle \frac{1}{2r^2}}N(N1){\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda }\right)\left({\displaystyle \frac{1}{k^2}}\right){\displaystyle \frac{1}{(\mathrm{det}𝒢_{\mathrm{Tr}}^1)_{l=0}}}`$
$`{\displaystyle \frac{N(N1)(d2)}{2r^2(N+d2)}}({\displaystyle \frac{1}{2r^2}}(N1)(N2){\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda }){\displaystyle \frac{1}{(\mathrm{det}𝒢_{\mathrm{Tr}}^1)_{l=0}}}]\eta ^\mu ^\mathrm{L},`$
where $`(\mathrm{det}𝒢_{\mathrm{Tr}}^1)_{l=0}`$ is the expression $`\mathrm{det}𝒢_{\mathrm{Tr}}^1`$ in Eq. (82) evaluated at $`l=0`$. The other eigenfunctions do not contribute in this case. For $`l=1,2,\mathrm{}`$,
$`U_{2}^{}{}_{\nu }{}^{\mu }\eta _{}^{\nu }{}_{}{}^{\mathrm{L}}`$
$`=`$ $`[2\left({\displaystyle \frac{d1}{d}}\right)({\displaystyle \frac{1}{2r^2}}N(N1){\displaystyle \frac{N+d}{N+d2}})^2(k^2)`$
$`\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l}}\right)^2\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l+\frac{1}{2r^2}N(N1)\frac{N+d}{N+d2}\mathrm{\Lambda }}}\right)`$
$`\left({\displaystyle \frac{1}{2r^2}}(N1)(N2){\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda }\right)^2(\mathrm{\Lambda }_l)`$
$`\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l}}\right)^2\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l+\frac{1}{2r^2}(N1)(N2)\frac{N+d}{N+d2}\mathrm{\Lambda }}}\right)`$
$`\left({\displaystyle \frac{d2}{d}}\right)\left({\displaystyle \frac{1}{2r^2}}N(N1){\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda }\right)\left({\displaystyle \frac{1}{2r^2}}{\displaystyle \frac{N(N1)(N2)}{N+d2}}{\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda }\right)(k^2)`$
$`\left(k^2\mathrm{\Lambda }_l+{\displaystyle \frac{1}{2r^2}}{\displaystyle \frac{N^3+N^2(2d7)2N(3d7)+4(d2)}{N+d2}}{\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda }\right)`$
$`\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l}}\right)^2{\displaystyle \frac{1}{\mathrm{det}𝒢_{\mathrm{Tr}}^1}}`$
$`{\displaystyle \frac{N(N1)^2(N2)(d2)^2}{4r^4(N+d2)^2}}\left({\displaystyle \frac{1}{2r^2}}N(N1){\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda }\right)(k^2)\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l}}\right)^2{\displaystyle \frac{1}{\mathrm{det}𝒢_{\mathrm{Tr}}^1}}`$
$`{\displaystyle \frac{N(N1)(d2)}{2r^2(N+d2)}}({\displaystyle \frac{1}{2r^2}}(N1)(N2){\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda })(k^2)\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l}}\right){\displaystyle \frac{1}{\mathrm{det}𝒢_{\mathrm{Tr}}^1}}]\eta ^\mu ^\mathrm{L},`$
$`U_{2}^{}{}_{\nu }{}^{a}\eta _{}^{\nu }{}_{}{}^{\mathrm{L}}`$
$`=`$ $`[({\displaystyle \frac{1}{2r^2}}(N+2)(N1){\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda })({\displaystyle \frac{1}{2r^2}}(N1)(N2){\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda })(k^2)`$
$`\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l}}\right)\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l\frac{2}{r^2}(N1)}}\right)\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l+\frac{1}{2r^2}(N1)(N2)\frac{N+d}{N+d2}\mathrm{\Lambda }}}\right)`$
$`\left({\displaystyle \frac{1}{2r^2}}(N+2)(N1){\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda }\right)\left({\displaystyle \frac{1}{2r^2}}{\displaystyle \frac{N(N1)(N2)}{N+d2}}{\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda }\right)(k^2)`$
$`\left(k^2\mathrm{\Lambda }_l+{\displaystyle \frac{1}{2r^2}}{\displaystyle \frac{N^3+N^2(2d7)2N(3d7)+4(d2)}{N+d2}}{\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda }\right)`$
$`\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l}}\right)\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l\frac{2}{r^2}(N1)}}\right){\displaystyle \frac{1}{\mathrm{det}𝒢_{\mathrm{Tr}}^1}}`$
$`{\displaystyle \frac{N(N1)^2(N2)d(d2)}{4r^4(N+d2)^2}}\left({\displaystyle \frac{1}{2r^2}}(N+2)(N1){\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda }\right)(k^2)`$
$`\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l}}\right)\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l\frac{2}{r^2}(N1)}}\right){\displaystyle \frac{1}{\mathrm{det}𝒢_{\mathrm{Tr}}^1}}`$
$`{\displaystyle \frac{(N1)(N2)(d2)}{2r^2(N+d2)}}\left({\displaystyle \frac{1}{2r^2}}N(N1){\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda }\right)(k^2)`$
$`\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l\frac{2}{r^2}(N1)}}\right){\displaystyle \frac{1}{\mathrm{det}𝒢_{\mathrm{Tr}}^1}}]\eta ^a^\mathrm{L},`$
$`U_{2}^{}{}_{b}{}^{\mu }\eta _{}^{b}{}_{}{}^{\mathrm{L}}`$
$`=`$ $`[({\displaystyle \frac{1}{2r^2}}(N1)(N2){\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda })({\displaystyle \frac{1}{2r^2}}(N+2)(N1){\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda })(\mathrm{\Lambda }_l)`$
$`\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l}}\right)\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l\frac{2}{r^2}(N1)}}\right)\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l+\frac{1}{2r^2}(N1)(N2)\frac{N+d}{N+d2}\mathrm{\Lambda }}}\right)`$
$`+\left({\displaystyle \frac{1}{2r^2}}(N1)(N2){\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda }\right)`$
$`\left({\displaystyle \frac{1}{2r^2}}{\displaystyle \frac{N(N1)(N+2d2)}{N+d2}}{\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda }\right)(\mathrm{\Lambda }_l)`$
$`\left(k^2\mathrm{\Lambda }_l+{\displaystyle \frac{1}{2r^2}}{\displaystyle \frac{N(N1)(N2)}{N+d2}}{\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda }\right)\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l}}\right)`$
$`\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l\frac{2}{r^2}(N1)}}\right){\displaystyle \frac{1}{\mathrm{det}𝒢_{\mathrm{Tr}}^1}}`$
$`+{\displaystyle \frac{N(N1)^2(N2)d(d2)}{4r^4(N+d2)^2}}\left({\displaystyle \frac{1}{2r^2}}(N1)(N2){\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda }\right)(\mathrm{\Lambda }_l)`$
$`\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l}}\right)\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l\frac{2}{r^2}(N1)}}\right){\displaystyle \frac{1}{\mathrm{det}𝒢_{\mathrm{Tr}}^1}}`$
$`{\displaystyle \frac{(N1)(N2)(d2)}{2r^2(N+d2)}}\left({\displaystyle \frac{1}{2r^2}}N(N1){\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda }\right)(\mathrm{\Lambda }_l)`$
$`\left(k^2\mathrm{\Lambda }_l{\displaystyle \frac{2}{r^2}}{\displaystyle \frac{N^2+N(d3)(d2)}{N+d2}}\right)\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l}}\right)`$
$`\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l\frac{2}{r^2}(N1)}}\right){\displaystyle \frac{1}{\mathrm{det}𝒢_{\mathrm{Tr}}^1}}]\eta ^\mu ^\mathrm{L},`$
$`U_{2}^{}{}_{b}{}^{a}\eta _{}^{b}{}_{}{}^{\mathrm{L}}`$
$`=`$ $`[2\left({\displaystyle \frac{N1}{N}}\right)({\displaystyle \frac{1}{2r^2}}N(N1){\displaystyle \frac{N+d}{N+d2}})^2(\mathrm{\Lambda }_l+{\displaystyle \frac{N}{r^2}})`$
$`\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l\frac{2}{r^2}(N1)}}\right)^2\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l+\frac{1}{2r^2}(N1)(N4)\frac{N+d}{N+d2}\mathrm{\Lambda }}}\right)`$
$`+({\displaystyle \frac{1}{2r^2}}(N+2)(N1){\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda })^2(k^2)`$
$`\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l\frac{2}{r^2}(N1)}}\right)^2\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l+\frac{1}{2r^2}(N1)(N2)\frac{N+d}{N+d2}\mathrm{\Lambda }}}\right)`$
$`+\left({\displaystyle \frac{N2}{N}}\right)\left({\displaystyle \frac{1}{2r^2}}N(N1){\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda }\right)`$
$`\left({\displaystyle \frac{1}{2r^2}}{\displaystyle \frac{N(N1)(N+2d2)}{N+d2}}{\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda }\right)(\mathrm{\Lambda }_l)`$
$`\left(k^2\mathrm{\Lambda }_l+{\displaystyle \frac{1}{2r^2}}{\displaystyle \frac{N(N1)(N2)}{N+d2}}{\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda }\right)\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l\frac{2}{r^2}(N1)}}\right)^2{\displaystyle \frac{1}{\mathrm{det}𝒢_{\mathrm{Tr}}^1}}`$
$`+{\displaystyle \frac{(N1)^2(N2)^2d(d2)}{4r^4(N+d2)^2}}\left({\displaystyle \frac{1}{2r^2}}N(N1){\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda }\right)(\mathrm{\Lambda }_l)`$
$`\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l\frac{2}{r^2}(N1)}}\right)^2{\displaystyle \frac{1}{\mathrm{det}𝒢_{\mathrm{Tr}}^1}}`$
$`{\displaystyle \frac{(N1)(N2)d}{2r^2(N+d2)}}\left({\displaystyle \frac{1}{2r^2}}(N+2)(N1){\displaystyle \frac{N+d}{N+d2}}\mathrm{\Lambda }\right)(\mathrm{\Lambda }_l)`$
$`(k^2\mathrm{\Lambda }_l{\displaystyle \frac{2}{r^2}}{\displaystyle \frac{N^2+N(d3)(d2)}{N+d2}})\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l\frac{2}{r^2}(N1)}}\right)^2{\displaystyle \frac{1}{\mathrm{det}𝒢_{\mathrm{Tr}}^1}}]\eta ^b^\mathrm{L}.`$
Both $`\eta _{}^{\nu }{}_{}{}^{\mathrm{L}}`$ and $`\eta _{}^{b}{}_{}{}^{\mathrm{L}}`$ have degeneracy $`D_l^{(s)}(N)`$.
### III.3 VD effective actions for gravitons on $`M^4\times S^N`$
With the eigenvalues of the various operators in the last subsection and the divergent parts of their traces listed in the Appendix, we can evaluate the divergent parts of the VD effective action on $`M^4\times S^N`$.
Let us start with Trln$`N^1`$, the ghost contribution to the VD effection action. From Eqs. (64), (68), and (70), we obtain on $`M^4\times S^N`$,
$`\mathrm{Trln}N^1|^{div}`$ $`=`$ $`d{\displaystyle \underset{k}{}}{\displaystyle \underset{l=0}{\overset{\mathrm{}}{}}}D_l^{(s)}(N)\mathrm{ln}(k^2\mathrm{\Lambda }_l)+{\displaystyle \underset{k}{}}{\displaystyle \underset{l=1}{\overset{\mathrm{}}{}}}D_l^{(v)}(N)\mathrm{ln}\left(k^2\mathrm{\Lambda }_l{\displaystyle \frac{N}{r^2}}\right)`$ (97)
$`+{\displaystyle \underset{k}{}}{\displaystyle \underset{l=1}{\overset{\mathrm{}}{}}}D_l^{(s)}(N)\mathrm{ln}(k^2\mathrm{\Lambda }_l{\displaystyle \frac{2}{r^2}}(N1))|^{div}.`$
If we start the summations over $`l`$ from $`l=0`$,
$`\mathrm{Trln}N^1|^{div}`$ $`=`$ $`d{\displaystyle \underset{k}{}}{\displaystyle \underset{l=0}{\overset{\mathrm{}}{}}}D_l^{(s)}(N)\mathrm{ln}(k^2\mathrm{\Lambda }_l)+{\displaystyle \underset{k}{}}{\displaystyle \underset{l=0}{\overset{\mathrm{}}{}}}D_l^{(v)}(N)\mathrm{ln}\left(k^2\mathrm{\Lambda }_l{\displaystyle \frac{N}{r^2}}\right)`$ (98)
$`+{\displaystyle \underset{k}{}}{\displaystyle \underset{l=0}{\overset{\mathrm{}}{}}}D_l^{(s)}(N)\mathrm{ln}\left(k^2\mathrm{\Lambda }_l{\displaystyle \frac{2}{r^2}}(N1)\right)`$
$`\delta _{N2}{\displaystyle \underset{k}{}}\mathrm{ln}(k^2{\displaystyle \frac{N}{r^2}}){\displaystyle \underset{k}{}}\mathrm{ln}(k^2{\displaystyle \frac{2}{r^2}}(N1))|^{div}`$
$`=`$ $`d{\displaystyle \underset{k}{}}{\displaystyle \underset{l=0}{\overset{\mathrm{}}{}}}D_l^{(s)}(N)\mathrm{ln}(k^2\mathrm{\Lambda }_l)`$
$`+{\displaystyle \underset{k}{}}{\displaystyle \underset{l=0}{\overset{\mathrm{}}{}}}D_l^{(v)}(N)\mathrm{ln}(k^2\mathrm{\Lambda }_l){\displaystyle \underset{k}{}}{\displaystyle \underset{l=0}{\overset{\mathrm{}}{}}}D_l^{(v)}(N){\displaystyle \underset{q=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{q}}\left({\displaystyle \frac{N}{r^2}}\right)^q\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l}}\right)^q`$
$`+{\displaystyle \underset{k}{}}{\displaystyle \underset{l=0}{\overset{\mathrm{}}{}}}D_l^{(s)}(N)\mathrm{ln}(k^2\mathrm{\Lambda }_l){\displaystyle \underset{k}{}}{\displaystyle \underset{l=0}{\overset{\mathrm{}}{}}}D_l^{(s)}(N){\displaystyle \underset{q=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{q}}\left({\displaystyle \frac{2(N1)}{r^2}}\right)^q\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l}}\right)^q`$
$`\delta _{N2}{\displaystyle \underset{k}{}}\mathrm{ln}k^2+\delta _{N2}{\displaystyle \underset{k}{}}{\displaystyle \underset{q=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{q}}\left({\displaystyle \frac{N}{r^2}}\right)^q\left({\displaystyle \frac{1}{k^2}}\right)^q`$
$`{\displaystyle \underset{k}{}}\mathrm{ln}k^2+{\displaystyle \underset{k}{}}{\displaystyle \underset{q=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{q}}\left({\displaystyle \frac{2(N1)}{r^2}}\right)^q\left({\displaystyle \frac{1}{k^2}}\right)^q|^{div},`$
where we have Taylor-expanded the logarithmic function. Using the results in the Appendix,
$`\mathrm{Trln}N^1|^{div}`$
$`=`$ $`{\displaystyle \frac{iV_4}{(4\pi r^2)^2ϵ}}[F^{(v)}(N)+5F^{(s)}(N){\displaystyle \underset{q=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{q}}\left({\displaystyle \frac{N}{r^2}}\right)^qG^{(v)}(0,q,N)`$
$`{\displaystyle \underset{q=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{q}}\left({\displaystyle \frac{2(N1)}{r^2}}\right)^qG^{(s)}(0,q,N)+\delta _{N2}\left({\displaystyle \frac{1}{2}}\right)\left({\displaystyle \frac{N}{r^2}}\right)^2(2r^4)+\left({\displaystyle \frac{1}{2}}\right)\left({\displaystyle \frac{2(N1)}{r^2}}\right)^2(2r^4)].`$
Therefore, for $`N=2`$,
$$\mathrm{Trln}N^1|^{div}=\frac{iV_4}{(4\pi r^2)^2ϵ}[\frac{24}{35}].$$
(100)
For $`N=4`$,
$$\mathrm{Trln}N^1|^{div}=\frac{iV_4}{(4\pi r^2)^2ϵ}[\frac{4640}{189}].$$
(101)
For $`N=6`$,
$$\mathrm{Trln}N^1|^{div}=\frac{iV_4}{(4\pi r^2)^2ϵ}[\frac{1232816}{10395}].$$
(102)
Similarly for Trln$`𝒢^1`$, we have for $`N=2`$,
$$\mathrm{Trln}𝒢^1|^{div}=\frac{iV_4}{(4\pi r^2)^2ϵ}[\frac{106}{15}\frac{257}{10}(\mathrm{\Lambda }r^2)+36(\mathrm{\Lambda }r^2)^2\frac{189}{8}(\mathrm{\Lambda }r^2)^3].$$
(103)
For $`N=4`$,
$$\mathrm{Trln}𝒢^1|^{div}=\frac{iV_4}{(4\pi r^2)^2ϵ}[\frac{7574}{45}+\frac{67552}{315}(\mathrm{\Lambda }r^2)\frac{13504}{135}(\mathrm{\Lambda }r^2)^2+\frac{1664}{81}(\mathrm{\Lambda }r^2)^3\frac{128}{81}(\mathrm{\Lambda }r^2)^4].$$
(104)
For $`N=6`$,
$`\mathrm{Trln}𝒢^1|^{div}={\displaystyle \frac{iV_4}{(4\pi r^2)^2ϵ}}`$ $`[{\displaystyle \frac{5833069}{3360}}{\displaystyle \frac{26312047}{24192}}(\mathrm{\Lambda }r^2)+{\displaystyle \frac{6498535}{24192}}(\mathrm{\Lambda }r^2)^2{\displaystyle \frac{8375}{256}}(\mathrm{\Lambda }r^2)^3`$ (105)
$`+{\displaystyle \frac{18125}{9216}}(\mathrm{\Lambda }r^2)^4{\displaystyle \frac{6875}{147456}}(\mathrm{\Lambda }r^2)^5].`$
Next, for the operator $`U_1`$, one can evaluate the trace of some general power of the operator. The results are the following. For $`N=2`$,
$`\mathrm{Tr}U_1|^{div}`$ $`=`$ $`{\displaystyle \frac{iV_4}{(4\pi r^2)^2ϵ}}\left[{\displaystyle \frac{148}{15}}{\displaystyle \frac{26}{5}}(\mathrm{\Lambda }r^2)\right],`$ (106)
$`\mathrm{Tr}U_1^2|^{div}`$ $`=`$ $`{\displaystyle \frac{iV_4}{(4\pi r^2)^2ϵ}}\left[2440(\mathrm{\Lambda }r^2)+18(\mathrm{\Lambda }r^2)^2\right],`$ (107)
$`\mathrm{Tr}U_1^3|^{div}`$ $`=`$ $`{\displaystyle \frac{iV_4}{(4\pi r^2)^2ϵ}}\left[{\displaystyle \frac{57}{5}}{\displaystyle \frac{69}{2}}(\mathrm{\Lambda }r^2)+{\displaystyle \frac{81}{2}}(\mathrm{\Lambda }r^2)^2{\displaystyle \frac{81}{4}}(\mathrm{\Lambda }r^2)^3\right].`$ (108)
For $`N=4`$,
$`\mathrm{Tr}U_1|^{div}`$ $`=`$ $`{\displaystyle \frac{iV_4}{(4\pi r^2)^2ϵ}}\left[{\displaystyle \frac{13432}{45}}{\displaystyle \frac{95504}{2835}}(\mathrm{\Lambda }r^2)\right],`$ (109)
$`\mathrm{Tr}U_1^2|^{div}`$ $`=`$ $`{\displaystyle \frac{iV_4}{(4\pi r^2)^2ϵ}}\left[{\displaystyle \frac{6908}{5}}{\displaystyle \frac{3808}{9}}(\mathrm{\Lambda }r^2)+{\displaystyle \frac{13312}{405}}(\mathrm{\Lambda }r^2)^2\right],`$ (110)
$`\mathrm{Tr}U_1^3|^{div}`$ $`=`$ $`{\displaystyle \frac{iV_4}{(4\pi r^2)^2ϵ}}\left[{\displaystyle \frac{9612}{5}}996(\mathrm{\Lambda }r^2)+{\displaystyle \frac{1600}{9}}(\mathrm{\Lambda }r^2)^2{\displaystyle \frac{896}{81}}(\mathrm{\Lambda }r^2)^3\right],`$ (111)
$`\mathrm{Tr}U_1^4|^{div}`$ $`=`$ $`{\displaystyle \frac{iV_4}{(4\pi r^2)^2ϵ}}[{\displaystyle \frac{29412}{35}}{\displaystyle \frac{3136}{5}}(\mathrm{\Lambda }r^2)+{\displaystyle \frac{2752}{15}}(\mathrm{\Lambda }r^2)^2`$ (112)
$`{\displaystyle \frac{2048}{81}}(\mathrm{\Lambda }r^2)^3+{\displaystyle \frac{1024}{729}}(\mathrm{\Lambda }r^2)^4].`$
For $`N=6`$,
$`\mathrm{Tr}U_1|^{div}`$ $`=`$ $`{\displaystyle \frac{iV_4}{(4\pi r^2)^2ϵ}}\left[{\displaystyle \frac{1670863}{945}}{\displaystyle \frac{67301}{756}}(\mathrm{\Lambda }r^2)\right],`$ (113)
$`\mathrm{Tr}U_1^2|^{div}`$ $`=`$ $`{\displaystyle \frac{iV_4}{(4\pi r^2)^2ϵ}}\left[{\displaystyle \frac{803450}{63}}{\displaystyle \frac{101855}{63}}(\mathrm{\Lambda }r^2)+{\displaystyle \frac{77815}{1512}}(\mathrm{\Lambda }r^2)^2\right],`$ (114)
$`\mathrm{Tr}U_1^3|^{div}`$ $`=`$ $`{\displaystyle \frac{iV_4}{(4\pi r^2)^2ϵ}}\left[{\displaystyle \frac{193775}{6}}{\displaystyle \frac{53575}{8}}(\mathrm{\Lambda }r^2)+{\displaystyle \frac{22475}{48}}(\mathrm{\Lambda }r^2)^2{\displaystyle \frac{2125}{192}}(\mathrm{\Lambda }r^2)^3\right],`$ (115)
$`\mathrm{Tr}U_1^4|^{div}`$ $`=`$ $`{\displaystyle \frac{iV_4}{(4\pi r^2)^2ϵ}}[{\displaystyle \frac{2197000}{63}}{\displaystyle \frac{91375}{9}}(\mathrm{\Lambda }r^2)+1125(\mathrm{\Lambda }r^2)^2`$ (116)
$`{\displaystyle \frac{8125}{144}}(\mathrm{\Lambda }r^2)^3+{\displaystyle \frac{625}{576}}(\mathrm{\Lambda }r^2)^4],`$
$`\mathrm{Tr}U_1^5|^{div}`$ $`=`$ $`{\displaystyle \frac{iV_4}{(4\pi r^2)^2ϵ}}[{\displaystyle \frac{41524375}{3024}}{\displaystyle \frac{996875}{192}}(\mathrm{\Lambda }r^2)+{\displaystyle \frac{1611875}{2016}}(\mathrm{\Lambda }r^2)^2{\displaystyle \frac{18125}{288}}(\mathrm{\Lambda }r^2)^3`$ (117)
$`+{\displaystyle \frac{15625}{6144}}(\mathrm{\Lambda }r^2)^4{\displaystyle \frac{3125}{73728}}(\mathrm{\Lambda }r^2)^5].`$
For traces involving the operator $`U_2`$, the calculation is more complicated. Following the same procedure as above, we obtain, for $`N=2`$,
$`\mathrm{Tr}U_2|^{div}`$ $`=`$ $`{\displaystyle \frac{iV_4}{(4\pi r^2)^2ϵ}}\left[10+{\displaystyle \frac{1}{2}}(\mathrm{\Lambda }r^2){\displaystyle \frac{45}{2}}(\mathrm{\Lambda }r^2)^2+{\displaystyle \frac{81}{4}}(\mathrm{\Lambda }r^2)^3\right],`$ (118)
$`\mathrm{Tr}U_1U_2|^{div}`$ $`=`$ $`{\displaystyle \frac{iV_4}{(4\pi r^2)^2ϵ}}\left[{\displaystyle \frac{151}{12}}{\displaystyle \frac{73}{2}}(\mathrm{\Lambda }r^2)+{\displaystyle \frac{81}{2}}(\mathrm{\Lambda }r^2)^2{\displaystyle \frac{81}{4}}(\mathrm{\Lambda }r^2)^3\right].`$ (119)
For $`N=4`$,
$`\mathrm{Tr}U_2|^{div}`$ $`=`$ $`{\displaystyle \frac{iV_4}{(4\pi r^2)^2ϵ}}[{\displaystyle \frac{1724}{5}}{\displaystyle \frac{328}{3}}(\mathrm{\Lambda }r^2)+{\displaystyle \frac{20432}{405}}(\mathrm{\Lambda }r^2)^2`$ (120)
$`{\displaystyle \frac{128}{9}}(\mathrm{\Lambda }r^2)^3+{\displaystyle \frac{1024}{729}}(\mathrm{\Lambda }r^2)^4],`$
$`\mathrm{Tr}U_1U_2|^{div}`$ $`=`$ $`{\displaystyle \frac{iV_4}{(4\pi r^2)^2ϵ}}[{\displaystyle \frac{5211}{5}}{\displaystyle \frac{4976}{15}}(\mathrm{\Lambda }r^2){\displaystyle \frac{176}{15}}(\mathrm{\Lambda }r^2)^2`$ (121)
$`+{\displaystyle \frac{128}{9}}(\mathrm{\Lambda }r^2)^3{\displaystyle \frac{1024}{729}}(\mathrm{\Lambda }r^2)^4],`$
$`\mathrm{Tr}U_1^2U_2|^{div}`$ $`=`$ $`{\displaystyle \frac{iV_4}{(4\pi r^2)^2ϵ}}[{\displaystyle \frac{4419}{5}}{\displaystyle \frac{1936}{3}}(\mathrm{\Lambda }r^2)+{\displaystyle \frac{5008}{27}}(\mathrm{\Lambda }r^2)^2`$ (122)
$`{\displaystyle \frac{2048}{81}}(\mathrm{\Lambda }r^2)^3+{\displaystyle \frac{1024}{729}}(\mathrm{\Lambda }r^2)^4],`$
$`\mathrm{Tr}U_2^2|^{div}`$ $`=`$ $`{\displaystyle \frac{iV_4}{(4\pi r^2)^2ϵ}}[{\displaystyle \frac{4657}{5}}{\displaystyle \frac{9952}{15}}(\mathrm{\Lambda }r^2)+{\displaystyle \frac{25312}{135}}(\mathrm{\Lambda }r^2)^2`$ (123)
$`{\displaystyle \frac{2048}{81}}(\mathrm{\Lambda }r^2)^3+{\displaystyle \frac{1024}{729}}(\mathrm{\Lambda }r^2)^4].`$
For $`N=6`$,
$`\mathrm{Tr}U_2|^{div}`$ $`=`$ $`{\displaystyle \frac{iV_4}{(4\pi r^2)^2ϵ}}[{\displaystyle \frac{2356505}{2016}}+{\displaystyle \frac{10560265}{32256}}(\mathrm{\Lambda }r^2){\displaystyle \frac{12249815}{96768}}(\mathrm{\Lambda }r^2)^2`$ (124)
$`+{\displaystyle \frac{717125}{36864}}(\mathrm{\Lambda }r^2)^3{\displaystyle \frac{26875}{18432}}(\mathrm{\Lambda }r^2)^4+{\displaystyle \frac{3125}{73728}}(\mathrm{\Lambda }r^2)^5],`$
$`\mathrm{Tr}U_1U_2|^{div}`$ $`=`$ $`{\displaystyle \frac{iV_4}{(4\pi r^2)^2ϵ}}[{\displaystyle \frac{26262425}{2304}}{\displaystyle \frac{2887775}{1536}}(\mathrm{\Lambda }r^2)+{\displaystyle \frac{3084025}{18432}}(\mathrm{\Lambda }r^2)^2`$ (125)
$`{\displaystyle \frac{230125}{12288}}(\mathrm{\Lambda }r^2)^3+{\displaystyle \frac{26875}{18432}}(\mathrm{\Lambda }r^2)^4{\displaystyle \frac{3125}{73728}}(\mathrm{\Lambda }r^2)^5],`$
$`\mathrm{Tr}U_1^2U_2|^{div}`$ $`=`$ $`{\displaystyle \frac{iV_4}{(4\pi r^2)^2ϵ}}[{\displaystyle \frac{3026375}{144}}{\displaystyle \frac{39073375}{8064}}(\mathrm{\Lambda }r^2)+{\displaystyle \frac{9993125}{32256}}(\mathrm{\Lambda }r^2)^2`$ (126)
$`+{\displaystyle \frac{43625}{6144}}(\mathrm{\Lambda }r^2)^3{\displaystyle \frac{26875}{18432}}(\mathrm{\Lambda }r^2)^4+{\displaystyle \frac{3125}{73728}}(\mathrm{\Lambda }r^2)^5],`$
$`\mathrm{Tr}U_1^3U_2|^{div}`$ $`=`$ $`{\displaystyle \frac{iV_4}{(4\pi r^2)^2ϵ}}[{\displaystyle \frac{5418875}{384}}{\displaystyle \frac{3549125}{672}}(\mathrm{\Lambda }r^2)+{\displaystyle \frac{7434625}{9216}}(\mathrm{\Lambda }r^2)^2`$ (127)
$`{\displaystyle \frac{387875}{6144}}(\mathrm{\Lambda }r^2)^3+{\displaystyle \frac{15625}{6144}}(\mathrm{\Lambda }r^2)^4{\displaystyle \frac{3125}{73728}}(\mathrm{\Lambda }r^2)^5],`$
$`\mathrm{Tr}U_2^2|^{div}`$ $`=`$ $`{\displaystyle \frac{iV_4}{(4\pi r^2)^2ϵ}}[{\displaystyle \frac{609125}{96}}+{\displaystyle \frac{998125}{1536}}(\mathrm{\Lambda }r^2){\displaystyle \frac{2394875}{4608}}(\mathrm{\Lambda }r^2)^2`$ (128)
$`+{\displaystyle \frac{163625}{2304}}(\mathrm{\Lambda }r^2)^3{\displaystyle \frac{36875}{9216}}(\mathrm{\Lambda }r^2)^4+{\displaystyle \frac{3125}{36864}}(\mathrm{\Lambda }r^2)^5],`$
$`\mathrm{Tr}U_1U_2^2|^{div}`$ $`=`$ $`{\displaystyle \frac{iV_4}{(4\pi r^2)^2ϵ}}[{\displaystyle \frac{117011375}{8064}}{\displaystyle \frac{38508875}{7168}}(\mathrm{\Lambda }r^2)+{\displaystyle \frac{26252375}{32256}}(\mathrm{\Lambda }r^2)^2`$ (129)
$`{\displaystyle \frac{583625}{9216}}(\mathrm{\Lambda }r^2)^3+{\displaystyle \frac{15625}{6144}}(\mathrm{\Lambda }r^2)^4{\displaystyle \frac{3125}{73728}}(\mathrm{\Lambda }r^2)^5].`$
Putting all these together, we can obtain the divergent parts of the VD effective actions for various $`M^4\times S^N`$ Kaluza-Klein spaces. For $`M^4\times S^2`$, we have
$`iW_{unique}^{div}`$ $`=`$ $`\mathrm{Trln}N^1{\displaystyle \frac{1}{2}}\mathrm{Trln}𝒢^1+{\displaystyle \frac{1}{2}}\mathrm{Tr}U_1+{\displaystyle \frac{1}{4}}\mathrm{Tr}U_1^2{\displaystyle \frac{1}{2}}\mathrm{Tr}U_2+{\displaystyle \frac{1}{6}}\mathrm{Tr}U_1^3{\displaystyle \frac{1}{2}}\mathrm{Tr}U_1U_2|^{div}`$ (130)
$`=`$ $`{\displaystyle \frac{iV_4}{(4\pi r^2)^2ϵ}}\left[{\displaystyle \frac{2249}{840}}+{\displaystyle \frac{25}{2}}(\mathrm{\Lambda }r^2){\displaystyle \frac{63}{4}}(\mathrm{\Lambda }r^2)^2+{\displaystyle \frac{135}{16}}(\mathrm{\Lambda }r^2)^3\right].`$
This result is consistent with that in ref.CK1 (16) where the divergent part of the VD effective action was calculated in a general six-dimensional background spacetime. For $`M^4\times S^4`$,
$`iW_{unique}^{div}`$ $`=`$ $`\mathrm{Trln}N^1{\displaystyle \frac{1}{2}}\mathrm{Trln}𝒢^1+{\displaystyle \frac{1}{2}}\mathrm{Tr}U_1+{\displaystyle \frac{1}{4}}\mathrm{Tr}U_1^2{\displaystyle \frac{1}{2}}\mathrm{Tr}U_2+{\displaystyle \frac{1}{6}}\mathrm{Tr}U_1^3{\displaystyle \frac{1}{2}}\mathrm{Tr}U_1U_2`$ (131)
$`+{\displaystyle \frac{1}{8}}\mathrm{Tr}U_1^4{\displaystyle \frac{1}{2}}\mathrm{Tr}U_1^2U_2+{\displaystyle \frac{1}{4}}\mathrm{Tr}U_2^2|^{div}`$
$`=`$ $`{\displaystyle \frac{iV_4}{(4\pi r^2)^2ϵ}}[{\displaystyle \frac{41657}{540}}{\displaystyle \frac{7850}{81}}(\mathrm{\Lambda }r^2)+{\displaystyle \frac{6152}{135}}(\mathrm{\Lambda }r^2)^2`$
$`{\displaystyle \frac{2176}{243}}(\mathrm{\Lambda }r^2)^3+{\displaystyle \frac{448}{729}}(\mathrm{\Lambda }r^2)^4].`$
Finally for $`M^4\times S^6`$ we obtain,
$`iW_{unique}^{div}`$ $`=`$ $`\mathrm{Trln}N^1{\displaystyle \frac{1}{2}}\mathrm{Trln}𝒢^1+{\displaystyle \frac{1}{2}}\mathrm{Tr}U_1+{\displaystyle \frac{1}{4}}\mathrm{Tr}U_1^2{\displaystyle \frac{1}{2}}\mathrm{Tr}U_2+{\displaystyle \frac{1}{6}}\mathrm{Tr}U_1^3{\displaystyle \frac{1}{2}}\mathrm{Tr}U_1U_2`$ (132)
$`+{\displaystyle \frac{1}{8}}\mathrm{Tr}U_1^4{\displaystyle \frac{1}{2}}\mathrm{Tr}U_1^2U_2+{\displaystyle \frac{1}{4}}\mathrm{Tr}U_2^2+{\displaystyle \frac{1}{10}}\mathrm{Tr}U_1^5{\displaystyle \frac{1}{2}}\mathrm{Tr}U_1^3U_2+{\displaystyle \frac{1}{2}}\mathrm{Tr}U_1U_2^2|^{div}`$
$`=`$ $`{\displaystyle \frac{iV_4}{(4\pi r^2)^2ϵ}}[{\displaystyle \frac{476483023}{591360}}+{\displaystyle \frac{10896475}{21504}}(\mathrm{\Lambda }r^2){\displaystyle \frac{32112235}{258048}}(\mathrm{\Lambda }r^2)^2`$
$`+{\displaystyle \frac{549625}{36864}}(\mathrm{\Lambda }r^2)^3{\displaystyle \frac{10625}{12288}}(\mathrm{\Lambda }r^2)^4+{\displaystyle \frac{625}{32768}}(\mathrm{\Lambda }r^2)^5].`$
## IV Applications
In this section we make two applications of the above effective actions. We give trace anomalies and we search for self-consistent configurations of the internal Kaluza-Klein spheres.
Some explanation of a trace anomaly is in order for gravity. Einstein gravity is not Weyl invariant and the trace part we calculate is NOT a combination of the anomalous and the “normal” contributions. What we calculate below is the “anomalous” part of the trace, as discussed in previous works. In “Trace anomaly for gravitons” RC (27) Critchley argues that the “anomalous” part of the trace is given by $`a_2`$, while the first terms of his (4) gives the “normal” contribution, irrespective to the value of $`\zeta `$. For Einstein gravity, this is equivalent to his (48), which is the same as our calculation. It is in this sense that we calculate the “anomalous” part of the trace anomaly. This is not the same as the usual trace anomaly in which the original classical action is conformal, while the quantum action is not. For nonconformal theories, Duff MJD (28) in his (29) gave a definition for the anomaly, which we suspect is the same as what we have calculated.
This problem is also discussed in “Quantum fields in curved space” BiD (29). On page 179, the authors mention that for fields which are not conformally invariant, there will be extra non-anomalous (normal) terms, which are really the same as the first terms of (4) in Critchley’s paper.
The paper “Non-conformal renormalized stress tensors in Robertson-Walker space-times”, BuD (30) by Bunch and Davies, gave a rather detailed calculation for the stress tensor for a massless, but minimally coupled scalar (therefore a non-conformally invariant field) in a conformal Robertson-Walker spacetime. Their (3.16) gives the total trace of the stress tensor of the quantum theory and (3.18) gives the “normal” part of the trace. Subtracting (3.18) from (3.16) (using Duff’s definition) gives (3.19), the “anomalous” trace part which should be equivalent to what we have calculated because (3.19) is proportional to $`a_2`$.
In the spirit of the above we can easily determine the gauge-independent VD trace anomaly for gravitons from the divergent part of the effective action of the corresponding Kaluza-Klein space. The appropriate expression is:
$`T_\mu ^\mu _{ren}`$ $`=`$ $`{\displaystyle \frac{ϵW_{unique}^{div}}{V_4V_{S^N}}},`$
$`\mathrm{where}V_{S^N}`$ $`=`$ $`{\displaystyle \frac{2\pi ^{(N+1)/2}r^N}{\mathrm{\Gamma }\left(\frac{N+1}{2}\right)}},`$ (133)
is the volume of the sphere $`S^N`$. Thus, for $`M^4\times S^2`$,
$$T_\mu ^\mu _{ren}=\frac{1}{(4\pi )^3}\left[\frac{135}{16}\mathrm{\Lambda }^3\frac{63}{4}\left(\frac{\mathrm{\Lambda }^2}{r^2}\right)+\frac{25}{2}\left(\frac{\mathrm{\Lambda }}{r^4}\right)\frac{2249}{840}\left(\frac{1}{r^6}\right)\right],$$
(134)
for $`M^4\times S^4`$,
$$T_\mu ^\mu _{ren}=\frac{1}{(4\pi )^4}\left[\frac{896}{243}\mathrm{\Lambda }^4\frac{4352}{81}\left(\frac{\mathrm{\Lambda }^3}{r^2}\right)+\frac{12304}{45}\left(\frac{\mathrm{\Lambda }^2}{r^4}\right)\frac{15700}{27}\left(\frac{\mathrm{\Lambda }}{r^6}\right)+\frac{41657}{90}\left(\frac{1}{r^8}\right)\right],$$
(135)
and for $`M^4\times S^6`$,
$`T_\mu ^\mu _{ren}`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^5}}[{\displaystyle \frac{9375}{8192}}\mathrm{\Lambda }^5{\displaystyle \frac{53125}{1024}}\left({\displaystyle \frac{\mathrm{\Lambda }^4}{r^2}}\right)+{\displaystyle \frac{336875}{384}}\left({\displaystyle \frac{\mathrm{\Lambda }^3}{r^4}}\right){\displaystyle \frac{160561175}{21504}}\left({\displaystyle \frac{\mathrm{\Lambda }^2}{r^6}}\right)`$ (136)
$`+{\displaystyle \frac{54482375}{1792}}\left({\displaystyle \frac{\mathrm{\Lambda }}{r^8}}\right){\displaystyle \frac{476483023}{9856}}\left({\displaystyle \frac{1}{r^{10}}}\right)].`$
The pure $`\mathrm{\Lambda }`$ terms in all three cases can be compared with CK3 (20) and the N = 2 case agrees with CK1 (16).
Our second use of the VD effective actions computed above is to produce stable configurations of the internal spheres. Because quantum fluctuations of the gravity field itself (about the Kaluza-Klein background) are generating the corrections to the effective potential, any stable configuration should be called a self-consistent dimensionally reduced configuration. For a stable configuration to be of interest it should have a positive renormalized Newton’s constant $`G_0`$. Most efforts to find such configurations have used the naive effective action (see RDS (2, 25, 26)) and have failed. Stable configurations were found but only with negative gravity constants. Those efforts that attempted to correctly use the VD effective potential have not gotten past the simplest cases: $`M^4\times S^1`$ or $`M^4\times S^2,`$ and $`M^4\times S^1\times S^1`$. For these cases no acceptable configurations were found either BO (4, 5, 8, 9, 10, 11, 15) .
Because of our lack of knowledge of the finite part of the effective potential we will assume that the divergent part ($`1/ϵ`$) dominates at one-loop. We can then easily seek stable configurations for the internal geometry, i.e., configurations where the classical gravity forces balance the quantum gravity (Casimir) pressures. We follow CW (24) and seek configurations satisfying
$$V(r)=\frac{V}{r}=0,\frac{^2V}{r^2}>0,$$
where the potential $`V(r)`$ is the negative of the 1-loop corrected effective action:
$$V(r)\frac{1}{16\pi G}\left(\frac{N(N1)}{r^2}2\mathrm{\Lambda }\right)+\frac{W_{unique}^{div}}{V_4}.$$
(137)
The bare value of Newton’s 4-dimensional gravity parameter G depends inversely on the sphere’s volume $`V_{S^N}`$ \[see (133)\], i.e., $`G\times V_{S^N}`$ is the initial gravity constant in 4+N dimensions. Values for $`W_{unique}^{div}`$ can be found in (130), (131), and (132).
There are only two static configurations for even dimensions $`6`$ (see columns 2 and 3 of Table I), one stable and one not.
Columns 4 and 5 relate the bare parameters to their renormalized values. Because Einstein gravity is not renormalizable we are rather unconstrained in our use of the effective actions (potentials). We have done the 1-loop renormalization by rewriting (137) as
$`V(r)`$ $`=`$ $`{\displaystyle \frac{1}{16\pi G}}\left({\displaystyle \frac{N(N1)}{r^2}}2\mathrm{\Lambda }\right)`$ (138)
$`+`$ $`{\displaystyle \frac{1}{(4\pi )^2ϵr^4}}\left\{c_0+c_1(\mathrm{\Lambda }r^2)+c_2(\mathrm{\Lambda }r^2)^2+\mathrm{}+c_{N/2+2}(\mathrm{\Lambda }r^2)^{N/2+2}\right\},`$
and collecting the $`r^N`$ and $`r^{N2}`$ terms to get two equations:
$$\frac{1}{16\pi G}\left(2\mathrm{\Lambda }\right)+\frac{1}{(4\pi )^2ϵ}\left\{c_{N/2+2}\mathrm{\Lambda }^{N/2+2}r^N\right\}=\frac{1}{16\pi G_0}\left(2\mathrm{\Lambda }_0\right),$$
(139)
$$\frac{1}{16\pi G}\left(\frac{N(N1)}{r^2}\right)+\frac{1}{(4\pi )^2ϵ}\left\{c_{N/2+1}\mathrm{\Lambda }^{N/2+1}r^{N2}\right\}=\frac{1}{16\pi G_0}\left(\frac{N(N1)}{r^2}\right).$$
(140)
The constants $`c_{N/2+1}`$ and $`c_{N/2+2}`$ can be read from (130),(131), and (132). Solving these two equations gives:
$$\frac{G}{G_0}=1+c_{N/2+1}\left(\mathrm{\Lambda }r^2\right)^{N/2+1}(G/\pi ϵr^2)/N(N1),$$
(141)
$$\frac{\mathrm{\Lambda }}{\mathrm{\Lambda }_0}=\frac{1+c_{N/2+1}\left(\mathrm{\Lambda }r^2\right)^{N/2+1}(G/\pi ϵr^2)/N(N1)}{1c_{N/2+2}\left(\mathrm{\Lambda }r^2\right)^{N/2+1}(G/\pi ϵr^2)/2.}$$
(142)
Both stationary configurations in the Table have negative values for the external dimensional regularization parameter $`ϵ=4d`$ and positive values for the renormalized Newton’s and cosmological constants G<sub>0</sub> and $`\mathrm{\Lambda }_0`$. The $`N=2`$ configuration is stable but requires a negative bare gravity constant G. The $`N=6`$ configuration has a positive unrenormalized gravity constant G, but is unstable.
Two input parameters, e.g., $`G_0`$ and $`\mathrm{\Lambda }_0`$, are required to evaluate all other parameters (see columns 6-8) including the renormalized $`O(N+1)`$ coupling constant $`g^2=(N+1)8\pi G_0/r^2`$, see CW (24). As expected, this theory cannot apply to the current phase of the universe where $`\mathrm{\Lambda }_010^{52}m^2`$ and $`G_0=2.6\times 10^{70}m^2`$, but perhaps to one of the pre-inflation phases where $`G_0\mathrm{\Lambda }_010^2`$.
## V Conclusions
We have demonstrated how to evaluate the divergent part of the VD effective action for Einstein gravity on even-dimensional Kaluza-Klein spacetimes of the form $`M^4\times S^N`$. First the effective action is expanded as a series of functional traces of various operators including $`N^1`$, $`𝒢^1`$, $`U_1`$, and $`U_2`$ and some of their products. Then the eigenvalues of these operators are obtained by decomposing the corresponding eigenfunctions, vectors or symmetric tensors, into their irreducible parts. Using these eigenvalues, the divergent parts of the traces can be obtained, thus giving the divergent parts of the VD effective action. The formulae used to extract these divergent parts are tabulated in the Appendix for $`N`$=2, 4, and 6.
Although the above procedure becomes more and more tedious as one goes to higher dimensions, there is no conceptual difficulty in doing so. One can therefore extend this method to even dimensions with $`N8`$, as well as to other more general coset spaces (provided eigenvalues for the corresponding Laplacians are known).
From the divergent parts of the VD effective action, we have obtained the gauge-independent trace anomaly for gravitons on $`M^4\times S^N`$. The trace anomaly for gravitons derived from the usual effective action depends on the choice of gauge condition in the off-shell case; however, the VD formalism provides an alternative definition of a unique trace anomaly, even when off-shell CK3 (20). Our final application was an attempt to find self-consistent dimensionally reduced Kaluza-Klein spaces. Only one was found (N=2) and it required that we start with a negative bare Newton’s constant. It also possessed much too small of a gauge coupling constant to represent the current stage of the universe.
###### Acknowledgements.
H. T. Cho is supported by the National Science Council of the Republic of China under contract number NSC 87-2112-M-032-001.
##
In this appendix we evaluate the divergent part of the traces of various operators. We use the dimensional regularization scheme in which the dimension of the external spacetime is taken to be
$$d4ϵ.$$
(143)
Then we extract the part which is proportional to $`1/ϵ`$ as $`ϵ0`$.
The traces that we shall consider are
$`{\displaystyle \underset{k}{}}\mathrm{ln}k^2,`$
$`{\displaystyle \underset{k}{}}\left({\displaystyle \frac{1}{k^2}}\right)^p,`$
$`{\displaystyle \underset{k}{}}{\displaystyle \underset{l=0}{\overset{\mathrm{}}{}}}D_l^{(s,v,t)}(N)\mathrm{ln}(k^2\mathrm{\Lambda }_l(N)){\displaystyle \frac{iV_4}{(4\pi r^2)^2ϵ}}F^{(s,v,t)}(N),`$
$`{\displaystyle \underset{k}{}}{\displaystyle \underset{l=0}{\overset{\mathrm{}}{}}}D^{(s,v,t)}(N)(k^2)^p\left({\displaystyle \frac{1}{k^2\mathrm{\Lambda }_l(N)}}\right)^q{\displaystyle \frac{iV_4}{(4\pi r^2)^2ϵ}}G^{(s,v,t)}(p,q,N),`$ (144)
for $`p0`$, $`q1`$.
First, using the proper time method, the divergent part of $`_k\mathrm{ln}k^2`$ can be written as
$`{\displaystyle \underset{k}{}}\mathrm{ln}k^2|^{div}`$ $`=`$ $`{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{d\tau }{\tau }}{\displaystyle \underset{k}{}}e^{\tau k^2}e^{\tau m^2}|_{m0}^{div}`$ (145)
$`=`$ $`{\displaystyle \frac{iV_d}{(4\pi )^{d/2}}}{\displaystyle _0^{\mathrm{}}}𝑑\tau \tau ^{d/21}e^{\tau m^2}|_{m0}^{div}`$
$`=`$ $`{\displaystyle \frac{iV_d}{(4\pi )^2}}{\displaystyle \frac{1}{ϵ}}(m^4)|_{m0}`$
$`=`$ $`0.`$
Similarly,
$$\underset{k}{}\left(\frac{1}{k^2}\right)^p|^{div}=\frac{iV_4}{(4\pi )^2}\frac{1}{ϵ}(2)\delta _{p2},$$
(146)
that is, the trace has a divergent part only when $`p=2`$.
Next, we consider $`F^{(s,v,t)}(N)`$. For $`N=2`$, $`D_l^{(t)}=0`$ for $`l1`$ and $`D_{l=0}^{(t)}=3`$. While
$$D_l^{(v)}=D_l^{(s)}=2l+1,$$
(147)
and
$$\mathrm{\Lambda }_l(2)=\frac{l(l+1)}{r^2},$$
(148)
for $`l0`$. Therefore,
$`{\displaystyle \underset{k}{}}{\displaystyle \underset{l=0}{\overset{\mathrm{}}{}}}D_l^{(t)}(2)\mathrm{ln}(k^2\mathrm{\Lambda }_l(2))|^{div}`$ $`=`$ $`{\displaystyle \underset{k}{}}(3)\mathrm{ln}k^2|^{div}`$ (149)
$`=`$ $`0,`$
and
$`{\displaystyle \underset{k}{}}{\displaystyle \underset{l=0}{\overset{\mathrm{}}{}}}D_l^{(v)}(2)\mathrm{ln}(k^2\mathrm{\Lambda }_l(2))|^{div}`$ (150)
$`=`$ $`{\displaystyle \underset{k}{}}{\displaystyle \underset{l=0}{\overset{\mathrm{}}{}}}(2l+1)\mathrm{ln}(k^2+{\displaystyle \frac{l(l+1)}{r^2}})|^{div}`$
$`=`$ $`{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{d\tau }{\tau }}\left({\displaystyle \underset{k}{}}e^{\tau k^2}\right)\left({\displaystyle \underset{l=0}{\overset{\mathrm{}}{}}}(2l+1)e^{\tau l(l+1)/r^2}\right)e^{\tau m^2}|_{m0}^{div}`$
$`=`$ $`{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{d\tau }{\tau }}{\displaystyle \frac{iV_d}{(4\pi )^{d/2}}}{\displaystyle \frac{e^{\tau m^2}}{\tau ^{d/2}}}\left({\displaystyle \frac{r^2}{\tau }}+{\displaystyle \frac{1}{3}}+{\displaystyle \frac{\tau }{15r^2}}+{\displaystyle \frac{4\tau ^2}{315r^4}}+\mathrm{}\right)|_{m0}^{div}`$
$`=`$ $`{\displaystyle \frac{iV_4}{(4\pi r^2)^2}}{\displaystyle \frac{1}{ϵ}}\left({\displaystyle \frac{8}{315}}\right).`$
Note that we have used an asymptotic expansion for the summation over $`l`$ above for small values of $`\tau `$ BKM (23). Now we have
$`F^{(t)}(2)`$ $`=`$ $`0,`$ (151)
$`F^{(v)}(2)`$ $`=`$ $`F^{(s)}(2)={\displaystyle \frac{8}{315}}.`$ (152)
For $`N=4`$,
$`D_l^{(t)}`$ $`=`$ $`{\displaystyle \frac{5}{6}}(2l+3)(l+4)(l1),`$ (153)
$`D_l^{(v)}`$ $`=`$ $`{\displaystyle \frac{1}{2}}(2l+3)(l+3)l,`$ (154)
$`D_l^{(s)}`$ $`=`$ $`{\displaystyle \frac{1}{6}}(2l+3)(l+2)(l+1),`$ (155)
$`\mathrm{\Lambda }_l`$ $`=`$ $`{\displaystyle \frac{l(l+3)}{r^2}},`$ (156)
for $`l0`$, and the asymptotic expansions,
$`{\displaystyle \underset{l=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{5}{6}}(2l+3)(l+4)(l1)e^{\tau l(l+3)/r^2}`$
$`=`$ $`{\displaystyle \frac{5r^4}{6\tau ^2}}{\displaystyle \frac{10r^2}{3\tau }}{\displaystyle \frac{91}{18}}{\displaystyle \frac{508\tau }{189r^2}}{\displaystyle \frac{127\tau ^2}{378r^4}}+{\displaystyle \frac{806\tau ^3}{2079r^6}}+{\displaystyle \frac{21311\tau ^4}{81081r^8}}+{\displaystyle \frac{3416\tau ^5}{57915r^{10}}}+\mathrm{},`$
$`{\displaystyle \underset{l=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{2}}(2l+3)(l+3)le^{\tau l(l+3)/r^2}`$
$`=`$ $`{\displaystyle \frac{r^4}{2\tau ^2}}{\displaystyle \frac{11}{30}}{\displaystyle \frac{46\tau }{315r^2}}+{\displaystyle \frac{19\tau ^2}{210r^4}}+{\displaystyle \frac{388\tau ^3}{3465r^6}}+{\displaystyle \frac{6179\tau ^4}{135135r^8}}{\displaystyle \frac{268\tau ^5}{225225r^{10}}}+\mathrm{},`$
$`{\displaystyle \underset{l=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{6}}(2l+3)(l+2)(l+1)e^{\tau l(l+3)/r^2}`$
$`=`$ $`{\displaystyle \frac{r^4}{6\tau ^2}}+{\displaystyle \frac{r^2}{3\tau }}+{\displaystyle \frac{29}{90}}+{\displaystyle \frac{37\tau }{189r^2}}+{\displaystyle \frac{149\tau ^2}{1890r^4}}+{\displaystyle \frac{179\tau ^3}{10395r^6}}{\displaystyle \frac{1387\tau ^4}{405405r^8}}{\displaystyle \frac{13162\tau ^5}{2027025r^{10}}}+\mathrm{}.`$ (159)
We obtain
$`F^{(t)}(4)`$ $`=`$ $`{\displaystyle \frac{127}{189}},`$ (160)
$`F^{(v)}(4)`$ $`=`$ $`{\displaystyle \frac{19}{105}},`$ (161)
$`F^{(s)}(4)`$ $`=`$ $`{\displaystyle \frac{149}{945}}.`$ (162)
For $`N=6`$,
$`D_l^{(t)}`$ $`=`$ $`{\displaystyle \frac{7}{60}}(2l+5)(l+6)(l+3)(l+2)(l1),`$ (163)
$`D_l^{(v)}`$ $`=`$ $`{\displaystyle \frac{1}{24}}(2l+5)(l+5)(l+3)(l+2)l,`$ (164)
$`D_l^{(s)}`$ $`=`$ $`{\displaystyle \frac{1}{120}}(2l+5)(l+4)(l+3)(l+2)(l+1),`$ (165)
$`\mathrm{\Lambda }_l`$ $`=`$ $`{\displaystyle \frac{l(l+5)}{r^2}},`$ (166)
for $`l0`$, and the asymptotic expansions,
$`{\displaystyle \underset{l=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{7}{60}}(2l+5)(l+6)(l+3)(l+2)(l1)e^{\tau l(l+5)/r^2}`$ (169)
$`=`$ $`{\displaystyle \frac{7r^6}{30\tau ^3}}{\displaystyle \frac{21r^2}{5\tau }}{\displaystyle \frac{262}{27}}{\displaystyle \frac{4133\tau }{450r^2}}{\displaystyle \frac{248\tau ^2}{99r^4}}+{\displaystyle \frac{958729\tau ^3}{289575r^6}}+{\displaystyle \frac{14624\tau ^4}{3861r^8}}`$
$`+{\displaystyle \frac{11267779\tau ^5}{16409250r^{10}}}{\displaystyle \frac{357108736\tau ^6}{168358905r^{12}}}+\mathrm{},`$
$`{\displaystyle \underset{l=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{24}}(2l+5)(l+5)(l+3)(l+2)le^{\tau l(l+5)/r^2}`$
$`=`$ $`{\displaystyle \frac{r^6}{12\tau ^3}}+{\displaystyle \frac{r^4}{4\tau ^2}}{\displaystyle \frac{1823}{3780}}{\displaystyle \frac{487\tau }{1260r^2}}+{\displaystyle \frac{3671\tau ^2}{13860r^4}}+{\displaystyle \frac{264611\tau ^3}{405405r^6}}+{\displaystyle \frac{1603\tau ^4}{4290r^8}}`$
$`{\displaystyle \frac{9333977\tau ^5}{45945900r^{10}}}{\displaystyle \frac{13067106599\tau ^6}{23570246700r^{12}}}+\mathrm{},`$
$`{\displaystyle \underset{l=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{120}}(2l+5)(l+4)(l+3)(l+2)(l+1)e^{\tau l(l+5)/r^2}`$
$`=`$ $`{\displaystyle \frac{r^6}{60\tau ^3}}+{\displaystyle \frac{r^4}{12\tau ^2}}+{\displaystyle \frac{r^2}{5\tau }}+{\displaystyle \frac{1139}{3780}}+{\displaystyle \frac{833\tau }{2700r^2}}+{\displaystyle \frac{137\tau ^2}{660r^4}}+{\displaystyle \frac{121442\tau ^3}{2027025r^6}}{\displaystyle \frac{45251\tau ^4}{810810r^8}}`$
$`{\displaystyle \frac{23068481\tau ^5}{229729500r^{10}}}{\displaystyle \frac{1974977293\tau ^6}{23570246700r^{12}}}+\mathrm{}.`$
We obtain
$`F^{(t)}(6)`$ $`=`$ $`{\displaystyle \frac{496}{99}},`$ (170)
$`F^{(v)}(6)`$ $`=`$ $`{\displaystyle \frac{3671}{6930}},`$ (171)
$`F^{(s)}(6)`$ $`=`$ $`{\displaystyle \frac{137}{330}}.`$ (172)
For $`G^{(s,v,t)}(p,q,N)`$, we again use the asymptotic expansions for various dimensions and the same procedure to extract the divergent parts. The results are listed out in the following tables.
| $`G^{(t)}(p,q,2)`$ | $`q=1`$ | 2 | 3 | 4 | 5 | 6 |
| --- | --- | --- | --- | --- | --- | --- |
| $`p=0`$ | 0 | $`6r^4`$ | 0 | 0 | 0 | 0 |
| 1 | 0 | 0 | $`6r^4`$ | 0 | 0 | 0 |
| 2 | 0 | 0 | 0 | $`6r^4`$ | 0 | 0 |
| 3 | 0 | 0 | 0 | 0 | $`6r^4`$ | 0 |
| $`G^{(v)}(p,q,2)`$ | $`q=1`$ | 2 | 3 | 4 | 5 | 6 |
| --- | --- | --- | --- | --- | --- | --- |
| $`p=0`$ | $`\frac{2r^2}{15}`$ | $`\frac{2r^4}{3}`$ | $`r^6`$ | 0 | 0 | 0 |
| 1 | $`\frac{16}{315}`$ | $`\frac{4r^2}{15}`$ | $`\frac{2r^4}{3}`$ | $`\frac{2r^6}{3}`$ | 0 | 0 |
| 2 | $`\frac{4}{105r^2}`$ | $`\frac{16}{105}`$ | $`\frac{2r^2}{5}`$ | $`\frac{2r^4}{3}`$ | $`\frac{r^6}{2}`$ | 0 |
| 3 | $`\frac{64}{1155r^4}`$ | $`\frac{16}{105r^2}`$ | $`\frac{32}{105}`$ | $`\frac{8r^2}{15}`$ | $`\frac{2r^4}{3}`$ | $`\frac{2r^6}{5}`$ |
| $`G^{(s)}(p,q,2)`$ | $`q=1`$ | 2 | 3 | 4 | 5 | 6 |
| --- | --- | --- | --- | --- | --- | --- |
| $`p=0`$ | $`\frac{2r^2}{15}`$ | $`\frac{2r^4}{3}`$ | $`r^6`$ | 0 | 0 | 0 |
| 1 | $`\frac{16}{315}`$ | $`\frac{4r^2}{15}`$ | $`\frac{2r^4}{3}`$ | $`\frac{2r^6}{3}`$ | 0 | 0 |
| 2 | $`\frac{4}{105r^2}`$ | $`\frac{16}{105}`$ | $`\frac{2r^2}{5}`$ | $`\frac{2r^4}{3}`$ | $`\frac{r^6}{2}`$ | 0 |
| 3 | $`\frac{64}{1155r^4}`$ | $`\frac{16}{105r^2}`$ | $`\frac{32}{105}`$ | $`\frac{8r^2}{15}`$ | $`\frac{2r^4}{3}`$ | $`\frac{2r^6}{5}`$ |
| $`G^{(t)}(p,q,4)`$ | $`q=1`$ | 2 | 3 | 4 | 5 | 6 | 7 | 8 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| $`p=0`$ | $`\frac{1016r^2}{189}`$ | $`\frac{91r^4}{9}`$ | $`\frac{10r^6}{3}`$ | $`\frac{5r^8}{18}`$ | 0 | 0 | 0 | 0 |
| 1 | $`\frac{254}{189}`$ | $`\frac{2032r^2}{189}`$ | $`\frac{91r^4}{9}`$ | $`\frac{20r^6}{9}`$ | $`\frac{5r^8}{36}`$ | 0 | 0 | 0 |
| 2 | $`\frac{3224}{693r^2}`$ | $`\frac{254}{63}`$ | $`\frac{1016r^2}{63}`$ | $`\frac{91r^4}{9}`$ | $`\frac{5r^6}{3}`$ | $`\frac{r^8}{12}`$ | 0 | 0 |
| 3 | $`\frac{340976}{27027r^4}`$ | $`\frac{12896}{693r^2}`$ | $`\frac{508}{63}`$ | $`\frac{4064r^2}{189}`$ | $`\frac{91r^4}{9}`$ | $`\frac{4r^6}{3}`$ | $`\frac{r^8}{18}`$ | 0 |
| 4 | $`\frac{54656}{3861r^6}`$ | $`\frac{1704880}{27027r^4}`$ | $`\frac{32240}{693r^2}`$ | $`\frac{2540}{189}`$ | $`\frac{5080r^2}{189}`$ | $`\frac{91r^4}{9}`$ | $`\frac{10r^6}{9}`$ | $`\frac{5r^8}{126}`$ |
| $`G^{(v)}(p,q,4)`$ | $`q=1`$ | 2 | 3 | 4 | 5 | 6 | 7 | 8 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| $`p=0`$ | $`\frac{92r^2}{315}`$ | $`\frac{11r^4}{15}`$ | 0 | $`\frac{r^8}{6}`$ | 0 | 0 | 0 | 0 |
| 1 | $`\frac{38}{105}`$ | $`\frac{184r^2}{315}`$ | $`\frac{11r^4}{15}`$ | 0 | $`\frac{r^8}{12}`$ | 0 | 0 | 0 |
| 2 | $`\frac{1552}{1155r^2}`$ | $`\frac{38}{35}`$ | $`\frac{92r^2}{105}`$ | $`\frac{11r^4}{15}`$ | 0 | $`\frac{r^8}{20}`$ | 0 | 0 |
| 3 | $`\frac{98864}{45045r^4}`$ | $`\frac{6208}{1155r^2}`$ | $`\frac{76}{35}`$ | $`\frac{368r^2}{315}`$ | $`\frac{11r^4}{15}`$ | 0 | $`\frac{r^8}{30}`$ | 0 |
| 4 | $`\frac{4288}{15015r^6}`$ | $`\frac{98864}{9009r^4}`$ | $`\frac{3104}{231r^2}`$ | $`\frac{76}{21}`$ | $`\frac{92r^2}{63}`$ | $`\frac{11r^4}{15}`$ | 0 | $`\frac{r^8}{42}`$ |
| $`G^{(s)}(p,q,4)`$ | $`q=1`$ | 2 | 3 | 4 | 5 | 6 | 7 | 8 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| $`p=0`$ | $`\frac{74r^2}{189}`$ | $`\frac{29r^4}{45}`$ | $`\frac{r^6}{3}`$ | $`\frac{r^8}{18}`$ | 0 | 0 | 0 | 0 |
| 1 | $`\frac{298}{945}`$ | $`\frac{148r^2}{189}`$ | $`\frac{29r^4}{45}`$ | $`\frac{2r^6}{9}`$ | $`\frac{r^8}{36}`$ | 0 | 0 | 0 |
| 2 | $`\frac{716}{3465r^2}`$ | $`\frac{298}{315}`$ | $`\frac{74r^2}{63}`$ | $`\frac{29r^4}{45}`$ | $`\frac{r^6}{6}`$ | $`\frac{r^8}{60}`$ | 0 | 0 |
| 3 | $`\frac{22192}{135135r^4}`$ | $`\frac{2864}{3465r^2}`$ | $`\frac{596}{315}`$ | $`\frac{296r^2}{189}`$ | $`\frac{29r^4}{45}`$ | $`\frac{2r^6}{15}`$ | $`\frac{r^8}{90}`$ | 0 |
| 4 | $`\frac{210592}{135135r^6}`$ | $`\frac{22192}{27027r^4}`$ | $`\frac{1432}{693r^2}`$ | $`\frac{596}{189}`$ | $`\frac{370r^2}{189}`$ | $`\frac{29r^4}{45}`$ | $`\frac{r^6}{9}`$ | $`\frac{r^8}{126}`$ |
| $`G^{(t)}(p,q,6)`$ | p=1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| $`q=0`$ | $`\frac{4133r^2}{225}`$ | $`\frac{524r^4}{27}`$ | $`\frac{21r^6}{5}`$ | 0 | $`\frac{7r^{10}}{360}`$ | 0 | 0 | 0 | 0 | 0 |
| 1 | $`\frac{992}{99}`$ | $`\frac{8266r^2}{225}`$ | $`\frac{524r^4}{27}`$ | $`\frac{14r^6}{5}`$ | 0 | $`\frac{7r^{10}}{900}`$ | 0 | 0 | 0 | 0 |
| 2 | $`\frac{3834916}{96525r^2}`$ | $`\frac{992}{33}`$ | $`\frac{4133r^2}{75}`$ | $`\frac{524r^4}{27}`$ | $`\frac{21r^6}{10}`$ | 0 | $`\frac{7r^{10}}{1800}`$ | 0 | 0 | 0 |
| 3 | $`\frac{233984}{1287r^4}`$ | $`\frac{15339664}{96525r^2}`$ | $`\frac{1984}{33}`$ | $`\frac{16532r^2}{225}`$ | $`\frac{524r^4}{27}`$ | $`\frac{42r^6}{25}`$ | 0 | $`\frac{r^{10}}{450}`$ | 0 | 0 |
| 4 | $`\frac{90142232}{546975r^6}`$ | $`\frac{1169920}{1287r^4}`$ | $`\frac{7669832}{19305r^2}`$ | $`\frac{9920}{99}`$ | $`\frac{4133r^2}{45}`$ | $`\frac{524r^4}{27}`$ | $`\frac{7r^6}{5}`$ | 0 | $`\frac{r^{10}}{720}`$ | 0 |
| 5 | $`\frac{11427479552}{3741309r^8}`$ | $`\frac{180284464}{182325r^6}`$ | $`\frac{1169920}{429r^4}`$ | $`\frac{15339664}{19305r^2}`$ | $`\frac{4960}{33}`$ | $`\frac{8266r^2}{75}`$ | $`\frac{524r^4}{27}`$ | $`\frac{6r^6}{5}`$ | 0 | $`\frac{r^{10}}{1080}`$ |
| $`G^{(v)}(p,q,6)`$ | p=1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| $`q=0`$ | $`\frac{487r^2}{630}`$ | $`\frac{1823r^4}{1890}`$ | 0 | $`\frac{r^8}{12}`$ | $`\frac{r^{10}}{144}`$ | 0 | 0 | 0 | 0 | 0 |
| 1 | $`\frac{3671}{3465}`$ | $`\frac{487r^2}{315}`$ | $`\frac{1823r^4}{1890}`$ | 0 | $`\frac{r^8}{24}`$ | $`\frac{r^{10}}{360}`$ | 0 | 0 | 0 | 0 |
| 2 | $`\frac{1058444}{135135r^2}`$ | $`\frac{3671}{1155}`$ | $`\frac{487r^2}{210}`$ | $`\frac{1823r^4}{1890}`$ | 0 | $`\frac{r^8}{40}`$ | $`\frac{r^{10}}{720}`$ | 0 | 0 | 0 |
| 3 | $`\frac{12824}{715r^4}`$ | $`\frac{4233776}{135135r^2}`$ | $`\frac{7342}{1155}`$ | $`\frac{974r^2}{315}`$ | $`\frac{1823r^4}{1890}`$ | 0 | $`\frac{r^8}{60}`$ | $`\frac{r^{10}}{1260}`$ | 0 | 0 |
| 4 | $`\frac{37335908}{765765r^6}`$ | $`\frac{12824}{143r^4}`$ | $`\frac{2116888}{27027r^2}`$ | $`\frac{7342}{693}`$ | $`\frac{487r^2}{126}`$ | $`\frac{1823r^4}{1890}`$ | 0 | $`\frac{r^8}{84}`$ | $`\frac{r^{10}}{2016}`$ | 0 |
| 5 | $`\frac{104536852792}{130945815r^8}`$ | $`\frac{74671816}{255255r^6}`$ | $`\frac{38472}{143r^4}`$ | $`\frac{4233776}{27027r^2}`$ | $`\frac{3671}{231}`$ | $`\frac{487r^2}{105}`$ | $`\frac{1823r^4}{1890}`$ | 0 | $`\frac{r^8}{112}`$ | $`\frac{r^{10}}{3024}`$ |
| $`G^{(s)}(p,q,6)`$ | p=1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| $`q=0`$ | $`\frac{833r^2}{1350}`$ | $`\frac{1139r^4}{1890}`$ | $`\frac{r^6}{5}`$ | $`\frac{r^8}{36}`$ | $`\frac{r^{10}}{720}`$ | 0 | 0 | 0 | 0 | 0 |
| 1 | $`\frac{137}{165}`$ | $`\frac{833r^2}{675}`$ | $`\frac{1139r^4}{1890}`$ | $`\frac{2r^6}{15}`$ | $`\frac{r^8}{72}`$ | $`\frac{r^{10}}{1800}`$ | 0 | 0 | 0 | 0 |
| 2 | $`\frac{485768}{675675r^2}`$ | $`\frac{137}{55}`$ | $`\frac{833r^2}{450}`$ | $`\frac{1139r^4}{1890}`$ | $`\frac{r^6}{10}`$ | $`\frac{r^8}{120}`$ | $`\frac{r^{10}}{3600}`$ | 0 | 0 | 0 |
| 3 | $`\frac{362008}{135135r^4}`$ | $`\frac{1943072}{675675r^2}`$ | $`\frac{274}{55}`$ | $`\frac{1666r^2}{675}`$ | $`\frac{1139r^4}{1890}`$ | $`\frac{2r^6}{25}`$ | $`\frac{r^8}{180}`$ | $`\frac{r^{10}}{6300}`$ | 0 | 0 |
| 4 | $`\frac{92273924}{3828825r^6}`$ | $`\frac{362008}{27027r^4}`$ | $`\frac{971536}{135135r^2}`$ | $`\frac{274}{33}`$ | $`\frac{833r^2}{270}`$ | $`\frac{1139r^4}{1890}`$ | $`\frac{r^6}{15}`$ | $`\frac{r^8}{252}`$ | $`\frac{r^{10}}{10080}`$ | 0 |
| 5 | $`\frac{15799818344}{130945815r^8}`$ | $`\frac{184547848}{1276275r^6}`$ | $`\frac{362008}{9009r^4}`$ | $`\frac{1943072}{135135r^2}`$ | $`\frac{137}{11}`$ | $`\frac{833r^2}{225}`$ | $`\frac{1139r^4}{1890}`$ | $`\frac{2r^6}{35}`$ | $`\frac{r^8}{336}`$ | $`\frac{r^{10}}{15120}`$ |
|
warning/0004/cond-mat0004115.html
|
ar5iv
|
text
|
# Mean Field critical behaviour for a Fully Frustrated Blume-Emery-Griffiths Model
## 1 Introduction
In the last two decades the physics of complex systems, ranging from dilute magnets to structural glasses has been captured by models which couple Ising variables with lattice gas or Potts variables -, i.e. models with this type of Hamiltonian:
$$\beta =\underset{ij}{}J\epsilon _{ij}S_iS_jn_in_j+\underset{ij}{}Kn_in_j+\mu \underset{i}{}n_i,$$
(1)
where $`\epsilon _{ij}=\pm 1`$ are quenched variables associated to pairs of nearest neighbour sites, $`J>0`$ is the interaction between the Ising spin variables$`(S=\pm 1),`$ $`K`$ is the interaction between the particles, $`n_i=0,1`$ are the lattice gas variables, $`\mu `$ is the chemical potential. The spins can interact each other ferromagnetically ($`\epsilon _{ij}=1`$) or antiferromagnetically ($`\epsilon _{ij}=1`$).
For $`\epsilon _{ij}=1`$ everywhere, this model concides with the original Blume-Emery-Griffiths model (BEG) - with an extra degeneracy 2 at each empty site. $`J\epsilon _{ij}`$ is the bilinear interaction, $`K`$ the quadrupolar interaction, and $`\mu `$ the crystal field. In the last few years the disordered BEG model has been studied for random values of the $`\epsilon _{ij}=\pm 1`$ . Recently the Degenerate BEG (DBEG) has been found suitable to describe the martensitic trasformation.
It may be useful to write the Hamiltonian of Eq. (1) in the following way:
$$\beta =\underset{ij}{}[J(\epsilon _{ij}S_iS_j1)n_in_j+\eta Jn_in_j]+\mu \underset{i}{}n_i,$$
(2)
where $`\eta =K/J+1.`$ For $`\eta =0`$ and $`J=\mathrm{},`$ this model has been extensively studied in the last few years to study glassy systems and granular materials in the disordered case (i.e. when the $`\epsilon _{ij}`$ variables are randomly distributed on the lattice ) , -. This model can be considered as a model of particles with an internal degree of freedom ($`S=\pm 1)`$ that interact with an effective coupling $`J(\epsilon _{ij}S_iS_j1)`$ which is zero for spin configurations that satisfy the interactions (i.e. $`\epsilon _{ij}S_iS_j=1)`$ and gives an infinite repulsion, for those that do not satisfy the interaction (i.e. $`\epsilon _{ij}S_iS_j1)`$. So these last configurations are forbidden for $`J=\mathrm{}`$.
Here we analyze the model for $`J`$ finite and $`\eta 0.`$ For $`\eta 0`$ there is an extra interaction between a pair of n.n. particles,while finite values of $`J`$ correspond to softening the hard core potential between the spin variables.
In particular we present a mean field analysis of the Hamiltonian of Eq. (1) in the fully frustrated (FF) case on the square lattice. In this case the $`\epsilon _{ij}`$ variables are choosen in such a way that every plaquette (i.e. elementary cell of the lattice) is frustrated. In other terms every plaquette has an odd number of $`\epsilon _{ij}=1`$, so that the four spins of the plaquette cannot completely satisfy the interactions. In Fig. 1 we show the Villain scheme for the 2D FF model, highlighting the differences between the A and B sublattices. For this FF lattice we have recently made a mean field analysis of the Frustrated Percolation problem -.
In Sec. 3 and Sec. 4 we write down the equations for site magnetizations $`(m_A`$ and $`m_B)`$ and site densities ($`D_A`$ and $`D_B)`$ and these enable us to find the critical lines for the order-disorder transitions in our model for the FF case.
For $`K/J`$ $`>1`$ (i.e. $`\eta >0)`$ there is a tricritical point which separates the critical line in two branches, respectively characterized by first-order and second-order transitions. On the other hand for $`K/J=1(`$i.e. $`\eta =0)`$ the transitions are second-order for any $`\mu .`$
Finally we compare the FF behaviour with that of the original Ferromagnetic BEG with and without degeneracy.
## 2 Mean field analysis
We will study the model defined by the Hamiltonian (1) by evaluating its free energy in a mean field approximation. For convenience we will set $`\kappa =K/J`$
At each site $`i`$ of the lattice we have to consider the variables $`S_i=\pm 1`$ and $`n_i=0,1`$. For notation purposes it is useful to introduce a new 4-state variable $`\nu _i`$ such that $`\left\{\nu _i\right\}=\left\{n_i\right\}\left\{S_i\right\}=\{1,1,0,0\}\{1,2,3,4\}`$. We can express the old variables in terms of this new variable by means of the relations: $`n_iS_i=\delta _{\nu _i,1}\delta _{\nu _i,2}`$ and $`n_i=\delta _{\nu _i,1}+\delta _{\nu _i,2}`$.
Moreover, using the index $`r`$ to denote one of the four states of $`\nu _i`$, we can define $`p_r^i=\delta _{\nu _i,r}`$, i.e. the probability that the site $`i`$ will be found in the state $`\nu _i=r`$. Here the angular brackets represent, as usual, the average done with the Hamiltonian of Eq. (1).
To obtain the free energy we evaluate first the internal energy of the system, which is the expectation value of our Hamiltonian:
$`\beta 𝒰`$ $``$ $`\beta =\beta {\displaystyle \underset{ij}{}}_{ij}\beta {\displaystyle \underset{i}{}}_i=`$
$`=`$ $`J{\displaystyle \underset{ij}{}}\epsilon _{ij}(\delta _{\nu _i,1}\delta _{\nu _i,2})(\delta _{\nu _j,1}\delta _{\nu _j,2})+`$
$`+\kappa {\displaystyle \underset{ij}{}}(\delta _{\nu _i,1}+\delta _{\nu _i,2})(\delta _{\nu _j,1}+\delta _{\nu _j,2})+\mu {\displaystyle \underset{i}{}}\delta _{\nu _i,1}+\delta _{\nu _i,2}.`$
In the MF context we neglect the fluctuations and can simply put
$$\delta _{\nu _i,r}\delta _{\nu _i,s}=\delta _{\nu _i,r}\delta _{\nu _i,s},$$
(4)
so relation (4) implies
$`\beta _{ij}`$ $`=`$ $`J\left[\epsilon _{ij}\left(p_1^ip_2^i\right)\left(p_1^jp_2^j\right)+\kappa \left(p_1^i+p_2^i\right)\left(p_1^j+p_2^j\right)\right],`$ (5a)
$`\beta _i`$ $`=`$ $`\mu \left(p_1^i+p_2^i\right).`$ (5b)
The order parameters we will use in the following are the site magnetization $`m_i`$ and the lattice gas particle density $`D_i`$ expressed by
$`m_i`$ $`=`$ $`S_in_i=\delta _{\nu _i,1}\delta _{\nu _i,2}=p_1^ip_2^i,`$ (6a)
$`D_i`$ $`=`$ $`n_i=\delta _{\nu _i,1}+\delta _{\nu _i,2}=p_1^i+p_2^i,`$ (6b)
from which we have:
$$p_1^i=\frac{1}{2}(D_i+m_i)p_2^i=\frac{1}{2}(D_im_i).$$
(7a)
These relations and the equivalence condition $`p_3^i=p_4^i`$, together with the normalization $`_{r=1}^4p_r^i=1`$, imply:
$$p_3^i=p_4^i=\frac{1}{2}(1D_i).$$
(7b)
Moreover we invoke the typical translation invariance requirement of the MF approximation, taken separately on the two sublattices. Then we look for a solution in which all the sites of sublattice A (B) have the same probabilities, i.e. $`p_r^i=p_r^AiA`$ and $`p_r^i=p_r^BiB`$. This solution is one of the many occurring in the degenerate ground state.
Using the translation invariance we can write
$$\beta _{AB}=J\left[m_Am_B+\kappa D_AD_B\right]$$
for the expectation value $`\beta _{ij}`$ of the partial Hamiltonian relative to any AB ferromagnetic bond, i.e. any ferromagnetic bond $`ij`$ such that $`iA`$ and $`jB`$. A similar relation holds for all the partial Hamiltonians relative to any AA ferromagnetic bond. On the other hand, the expectation value of the partial Hamiltonian relative to any BB antiferromagnetic bond ($`\epsilon _{ij}=1)`$ is given by
$$\beta _{BB}=J\left[m_B^2+\kappa D_B^2\right].$$
Therefore, for $`N`$ sites, since the number of A sites and the number of B sites are both $`N/2`$, the internal energy is
$`{\displaystyle \frac{\beta 𝒰}{N}}`$ $`=`$ $`{\displaystyle \frac{1}{N}}{\displaystyle \underset{i=1}{\overset{N}{}}}\left[{\displaystyle \frac{1}{2}}{\displaystyle \underset{j:ij}{}}\beta _{ij}+\mu D_i\right]`$
$`=`$ $`{\displaystyle \frac{1}{N}}{\displaystyle \underset{iA}{}}{\displaystyle \frac{1}{2}}\left\{{\displaystyle \frac{z}{2}}\beta _{AA}+{\displaystyle \frac{z}{2}}\beta _{AB}+\mu D_A\right\}`$
$`+{\displaystyle \frac{1}{N}}{\displaystyle \underset{iB}{}}{\displaystyle \frac{1}{2}}\left\{{\displaystyle \frac{z}{2}}\beta _{BA}+{\displaystyle \frac{z}{2}}\beta _{BB}+\mu D_B\right\}`$
$`=`$ $`{\displaystyle \frac{Jz}{8}}\left[m_A^2+2m_Am_Bm_B^2+\kappa \left(D_A+D_B\right)^2\right]+{\displaystyle \frac{1}{2}}\mu \left(D_A+D_B\right).`$
(8)
For the evaluation of the MF entropic term we use the factorization property of the probability distribution $`𝒫(\nu _1,\mathrm{},\nu _N)`$ and therefore get $`𝒮k_{\left\{\nu \right\}}𝒫\mathrm{ln}𝒫=k_{i=1}^N_{r=1}^4p_r^i\mathrm{ln}p_r^i`$.
Using the translation invariance, this can be written in the form
$$\frac{𝒮}{kN}=\frac{1}{2}\underset{r=1}{\overset{4}{}}(p_r^A\mathrm{ln}p_r^A+p_r^B\mathrm{ln}p_r^B).$$
(9)
Using Eqs. (2) and (9) we can finally write the MF free energy per site of the lattice:
$$\beta f\frac{\beta }{N}\frac{\beta 𝒰}{N}\frac{𝒮}{kN},$$
(10)
where the probabilities $`p_r^A`$ and $`p_r^B`$ have to be expressed in terms of the local order parameters $`m_A,m_B,D_A`$ and $`D_B`$ through Eq (7).
## 3 Equations for the site Magnetizations and Densities
The knowledge of the free energy allows us to write down easily the MF equations that must be satisfied by the order parameters $`m_A`$, $`m_B`$, $`D_A`$ and $`D_B`$.
From the stationary relations $`f/m_A=0`$ and $`f/m_B=0`$ it follows that
$$m_A=D_A\mathrm{tanh}\left(\frac{\lambda }{2}(m_A+m_B)\right),m_B=D_B\mathrm{tanh}\left(\frac{\lambda }{2}(m_Am_B)\right).$$
(11)
Here $`\lambda =4J=4J_o/kT=T_c/T`$ where $`T_c`$ $`4J_0/k`$ is the mean field critical temperature of the isotropic Ising model recovered by the isotropic version of the Hamiltonian (1) in the $`\mu \mathrm{}`$ limit.
Morover from the stationary relations $`f/D_A=0`$ and $`f/D_B=0`$ we deduce that
$`e^{\kappa \lambda (D_A+D_B)+2\mu }`$ $`=`$ $`{\displaystyle \frac{D_A^2m_A^2}{(1D_A)^2}},`$
$`e^{\kappa \lambda (D_A+D_B)+2\mu }`$ $`=`$ $`{\displaystyle \frac{D_B^2m_B^2}{(1D_B)^2}}.`$
These relations give in implicit form $`D_A`$ and $`D_B`$ for every $`m_A`$ and $`m_B.`$
Now, replacing Eqs. (11) into Eqs. (3) we get stationarity in the four order parameters $`m_A`$, $`m_B`$, $`D_A`$ and $`D_B`$. After straightforward calculations we find:
$`D_A`$ $`=`$ $`{\displaystyle \frac{\mathrm{cosh}\left[(\lambda /2)(m_A+m_B)\right]}{e^{(\kappa \lambda /2)(D_A+D_B)\mu }+\mathrm{cosh}\left[(\lambda /2)(m_A+m_B)\right]}},`$ (13a)
$`m_A`$ $`=`$ $`{\displaystyle \frac{\mathrm{sinh}\left[(\lambda /2)(m_A+m_B)\right]}{e^{(\kappa \lambda /2)(D_A+D_B)\mu }+\mathrm{cosh}\left[(\lambda /2)(m_A+m_B)\right]}},`$ (13b)
$`D_B`$ $`=`$ $`{\displaystyle \frac{\mathrm{cosh}\left[(\lambda /2)(m_Am_B)\right]}{e^{(\kappa \lambda /2)(D_A+D_B)\mu }+\mathrm{cosh}\left[(\lambda /2)(m_Am_B)\right]}},`$ (14a)
$`m_B`$ $`=`$ $`{\displaystyle \frac{\mathrm{sinh}\left[(\lambda /2)(m_Am_B)\right]}{e^{(\kappa \lambda /2)(D_A+D_B)\mu }+\mathrm{cosh}\left[(\lambda /2)(m_Am_B)\right]}}.`$ (14b)
These equations can be studied numerically for different values of $`\kappa `$ in order to find the fixed points for every $`\lambda `$ and $`\mu `$. This analysis, together with the values of the free energy (10) for each fixed point, has enabled us to find for every $`\mu `$ the critical value $`\lambda _c`$ where the order parameters $`m_A`$ , $`m_B`$ , $`D_A`$ and $`D_B`$ undergo a first-order or second-order transition.
## 4 Critical lines and Results
We have done our analysis for a number of values of the $`\kappa `$ parameter, but report here, for convenience, only the most interesting cases in the range$`\kappa 1`$ (i.e. $`\eta 0)`$. Note that the antiquadrupolar phase that generally appears in the BEG model for $`\kappa <0`$ does not appear here because our sublattice partition is intrinsically different from the usual BEG sublattice partition. The critical behaviours are reported in Fig. 2–6 respectively for $`\eta =1.16,`$ $`1,`$ $`.84,`$ $`.5,`$ $`0.`$ To appreciate the differences between the FF model and the Ferromagnetic model (i.e. $`\epsilon _{ij}=1`$ for all bonds), each figure contains the (a)-section in which we report the behaviour of the Degenerate FF BEG model and the (b)-section relative to behaviour of Degenerate Ferromagnetic BEG model. In the (a)-section for each $`T/T_c`$ we give the field $`\mu /\lambda `$ were the transition from the high-field disordered phase ($`m_A=m_B=0`$ and $`D_A=D_B`$) to the low-field ordered phase ($`m_A>m_B0`$ and $`D_A>D_B`$) takes place. Bold (dotted) lines represents second-order (first-order) transitions. Dashed lines represent the spinodals, i.e. the boundaries of areas of metastability that surround any first-order transition line. Below the first-order transition line. the metastable phase is the disordered phase, above this line the metastable phase is the ordered phase . In the (b)-section for each $`T/T_c`$ we give the field $`\mu /\lambda `$ were the transition from the high-field disordered phase ($`m=0`$ and $`D1/2`$) to the low-field ordered phase ($`m>0`$ and $`D1/2)`$ takes place. As for the (a)-section, bold (dotted) lines represents second-order (first-order) transitions; dashed lines represent the spinodals. Fig. 2-6 is relative to decreasing values of the extra-interaction $`\eta =\kappa +1.`$ The overall feature is that decreasing $`\eta `$ we obtain a smaller ordered region. This is expected if we look at the Hamiltonian (2) since $`\eta `$ is the extra interaction among the particles. In the insert of Fig. 3b and Fig. 4b we report also the behaviour of the original BEG.
For the Ferromagnetic Degenerate BEG we find that the degeneracy reduces the area of the ordered region and increases the area of the region of first-order transitions, in agreement with recent results .
On the other hand it is known that the frustration has the conflicting effect of reducing this region both for the original BEG with random bonds and for the DBEG with random field . Here we find that the frustration reduces the ordered region and moves the tricritical point toward low temperatures, i.e. the frustration in the Fully-Frustrated model (in spite of the small degeneracy present) reduces the first order region.
These results may be useful to study the effects of the softening of the hard core potential and the effect of the attraction between particles for systems described by Hamiltonian (2) such as for example glasses and granular material.
Acknowledgements
We gratefully aknowledge A. Coniglio and A. De Candia for useful discussions.
Figure captions
$`2d`$ FF model on the square lattice. Straight (wavy) lines represent ferromagnetic (antiferromagnetic) interactions. $`z=4`$ ferromagnetic interactions start from each site of the sublattice $`A`$ (open circles); $`z/2`$ ferromagnetic interactions and $`z/2`$ antiferromagnetic interactions start from each site of the sublattice $`B`$ (closed circles).
(a) Critical lines for the FF lattice for $`\kappa =+.16(`$i.e. $`\eta =1.16)`$. Bold (dotted) lines represents second-order (first-order) transitions. Dashed lines represent the spinodals. (b) Corresponding critical lines for the ferromagnetic model (i.e. $`\epsilon _{ij}=1`$ for all bonds).
(a) Critical lines for the FF lattice for $`\kappa =0`$ ($`\eta =1)`$. The tricritical point is located at $`T/T_c0,233`$ and $`\mu /\lambda =(1/\lambda )\mathrm{ln}\left(1+\lambda /\sqrt{2}\right).166`$. (b) Corresponding critical lines for the ferromagnetic model. The insert reports the critical lines for the original BEG .
(a) Critical lines for the FF lattice for $`\kappa =.16(\eta =.84)`$., (b) Corresponding critical lines for the ferromagnetic model Ferromagnetic. In the insert we report the corresponding critical lines for the original BEG .
(a) Critical lines for the FF lattice for $`\kappa =.5(\eta =+.5)`$, (b) Corresponding critical lines for the ferromagnetic model. Observe that both in the ferromagnetic and fully-frustrated case the first-order transition line continues in the ordered phase, below the tricrical point,similarly to the corresponding behaviour of the original BEG .
(a) Critical lines for the FF lattice for $`\kappa =1(\eta =0)`$, (b) Corresponding critical lines for the ferromagnetic model. Observe that the first-order transition line now disappears, differently from what happens in the spin glass case .
|
warning/0004/hep-th0004014.html
|
ar5iv
|
text
|
# References
UNIL-IPT-00-08
hep-th/0004014
April 2000
Localizing Gravity on a String-Like Defect
in Six Dimensions
Tony Gherghetta<sup>1</sup><sup>1</sup>1Email: tony.gherghetta@ipt.unil.ch and Mikhail Shaposhnikov<sup>2</sup><sup>2</sup>2Email: mikhail.shaposhnikov@ipt.unil.ch
Institute of Theoretical Physics
University of Lausanne
CH-1015 Lausanne, Switzerland
## Abstract
We present a metric solution in six dimensions where gravity is localized on a four-dimensional singular string-like defect. The corrections to four-dimensional gravity from the bulk continuum modes are suppressed by $`𝒪(1/r^3)`$. No tuning of the bulk cosmological constant to the brane tension is required in order to cancel the four-dimensional cosmological constant.
It is an old idea that spacetime may have more than four dimensions, with extra coordinates being unobservable at available energies. A first possibility arises in Kaluza-Klein type theories (see e.g and references therein), where the D-dimensional metric has the form
$$ds^2=g_{\mu \nu }(x^\mu )dx^\mu dx^\nu \gamma _{ab}(x^a)dx^adx^b.$$
(1)
Here $`g_{\mu \nu }`$ is the metric of our four-dimensional world, while $`\gamma _{ab}`$ is the metric associated with $`D4`$ (small, with a size $`M^1`$) compact extra dimensions. The compactness of extra dimensions makes them unobservable at energies $`E<M`$, and manifests itself in the existence of an infinite tower of states with four-dimensional masses $`M`$.
In fact, the Kaluza-Klein metric is not the most general metric consistent with Poincare invariance in four dimensions. Its generalization was proposed in , and is given by
$$ds^2=\sigma (x^a)g_{\mu \nu }(x^\mu )dx^\mu dx^\nu \gamma _{ab}(x^a)dx^adx^b.$$
(2)
where $`\sigma (x^a)`$ is a conformal factor depending on the extra coordinates only. A number of specific solutions of the Einstein equations in six-dimensional (6d) spacetime with a positive 6d cosmological constant were found in , leading to non-compact extra dimensions while still leaving them unobservable at low energies.
Yet another idea leading to non-compact extra dimensions was suggested in . In this case the four dimensions of our world were identified with the internal space of topological defects residing in a higher-dimensional spacetime (e.g. a domain wall in 5d, string in 6d, monopole in 7d, instanton in 8d, etc). In these type of backgrounds, as a rule, there are fermionic and scalar zero modes, that can be associated with the four-dimensional particles that we observe. At that time it was not clear how to localize the gauge fields and gravity on topological defects in order to make the whole construction realistic.
The solitons of string theory - D-branes - provide a natural framework for the localization of gauge and matter fields on the worldvolume of the branes . In field theory language the branes can be associated with topological defects. Moreover, in Ref. it was discovered that gravity could be localized on the 3-brane domain wall in 5d spacetime. A normalizable graviton zero mode residing on the brane correctly reproduces 4d gravity, while the continuum spectrum of 5d gravitons living in the bulk, gives only a small correction $`𝒪(1/r^2)`$ to Newton’s law at large distances . The metric of the corresponding 5d spacetime has the general structure of eq. (2).
The aim of the present paper is to see what happens with gravity around a 3-brane of a specific structure (local string defect in field theory language) in 6d spacetime with a negative cosmological constant. In fact, a regular solution of the Einstein equations in this case for an empty space follows immediately from , but does not give any possibility of compactification. However, the existence of a brane with positive tension changes the situation and we find a solution which is very similar to that of Ref. . In contrast to 5d case, there is no fine-tuning of the cosmological constant in the bulk to the tension of the brane (the origin of this difference is that the 1d space in the domain wall scenario is flat, while the 2d space around the string defect can be curved). Similarly to the solution in Ref. , there is a normalizible graviton zero mode attached to the string-like defect, and the contribution of bulk gravitons is suppressed, leading to $`𝒪(1/r^3)`$ violations of Newton’s law. A hierarchy between the four-dimensional Planck scale and the Planck scale in 6d can be obtained, leading to a solution of the gauge hierarchy problem similar to that of ref. .
Other solutions obtained with two transverse dimensions include a generalization of the original 5d domain wall setup to the case of parallel brane sources , and the case of global string defects . Furthermore, a class of radially symmetric solutions was considered in .
Let us begin with the details of our solution. In 6d the Einstein equations with a bulk cosmological constant $`\mathrm{\Lambda }`$ and stress-energy tensor $`T_{AB}`$ are
$$R_{AB}\frac{1}{2}g_{AB}R=\frac{1}{M_6^4}\left(\mathrm{\Lambda }g_{AB}+T_{AB}\right),$$
(3)
where $`M_6`$ is the six-dimensional reduced Planck scale. We will assume that there exists a solution that respects 4d Poincare invarance. A six-dimensional metric satisfying this ansatz is
$$ds^2=\sigma (\rho )g_{\mu \nu }dx^\mu dx^\nu d\rho ^2\gamma (\rho )d\theta ^2,$$
(4)
where the metric signature is $`(+,,,)`$. For the two extra spatial dimensions we have introduced polar coordinates $`(\rho ,\theta )`$, where $`0\rho <\mathrm{}`$ and $`0\theta <2\pi `$. With our metric ansatz (4), the general expression for the four-dimensional reduced Planck scale, $`M_P`$ expressed in terms of $`M_6`$ is
$$M_P^2=2\pi M_6^4_0^{\mathrm{}}𝑑\rho \sigma \sqrt{\gamma }.$$
(5)
The nonzero components of the stress-energy tensor $`T_B^A`$ are assumed to be
$$T_\nu ^\mu =\delta _\nu ^\mu f_0(\rho ),T_\rho ^\rho =f_\rho (\rho ),\mathrm{and}T_\theta ^\theta =f_\theta (\rho ),$$
(6)
where we have introduced three source functions $`f_0,f_\rho `$ and $`f_\theta `$, which depend only on the radial coordinate $`\rho `$. Using the cylindrically symmetric metric ansatz (4) and the stress energy tensor (6), the Einstein equations become
$`{\displaystyle \frac{3}{2}}{\displaystyle \frac{\sigma ^{\prime \prime }}{\sigma }}+{\displaystyle \frac{3}{4}}{\displaystyle \frac{\sigma ^{}}{\sigma }}{\displaystyle \frac{\gamma ^{}}{\gamma }}{\displaystyle \frac{1}{4}}{\displaystyle \frac{\gamma ^2}{\gamma ^2}}+{\displaystyle \frac{1}{2}}{\displaystyle \frac{\gamma ^{\prime \prime }}{\gamma }}`$ $`=`$ $`{\displaystyle \frac{1}{M_6^4}}(\mathrm{\Lambda }+f_0(\rho ))+{\displaystyle \frac{1}{M_P^2}}{\displaystyle \frac{\mathrm{\Lambda }_{phys}}{\sigma }},`$
$`{\displaystyle \frac{3}{2}}{\displaystyle \frac{\sigma ^2}{\sigma ^2}}+{\displaystyle \frac{\sigma ^{}}{\sigma }}{\displaystyle \frac{\gamma ^{}}{\gamma }}`$ $`=`$ $`{\displaystyle \frac{1}{M_6^4}}(\mathrm{\Lambda }+f_\rho (\rho ))+{\displaystyle \frac{1}{M_P^2}}{\displaystyle \frac{2\mathrm{\Lambda }_{phys}}{\sigma }},`$
$`2{\displaystyle \frac{\sigma ^{\prime \prime }}{\sigma }}+{\displaystyle \frac{1}{2}}{\displaystyle \frac{\sigma ^2}{\sigma ^2}}`$ $`=`$ $`{\displaystyle \frac{1}{M_6^4}}(\mathrm{\Lambda }+f_\theta (\rho ))+{\displaystyle \frac{1}{M_P^2}}{\displaystyle \frac{2\mathrm{\Lambda }_{phys}}{\sigma }},`$ (7)
where the denotes differentiation $`d/d\rho `$. The constant $`\mathrm{\Lambda }_{phys}`$ represents the physical four-dimensional cosmological constant, where
$$R_{\mu \nu }^{(4)}\frac{1}{2}g_{\mu \nu }R^{(4)}=\frac{1}{M_P^2}\mathrm{\Lambda }_{phys}g_{\mu \nu }.$$
(8)
By eliminating two of the equations in (S0.Ex1), the sources can be related by the equation
$$f_\rho ^{}=2\frac{\sigma ^{}}{\sigma }(f_0f_\rho )+\frac{1}{2}\frac{\gamma ^{}}{\gamma }(f_\theta f_\rho ).$$
(9)
In the absence of source terms, the discussion of solutions to this coupled system of differential equations for arbitrary values of $`\mathrm{\Lambda }_{phys}`$ and $`\mathrm{\Lambda }>0`$ can be found in . However, the case $`\mathrm{\Lambda }<0`$ was not considered there because the vacuum solutions lead to noncompact transverse spaces, and therefore using (5), one cannot obtain a finite value of the Planck scale. Here we propose adding singular source terms in order to obtain a transverse space with finite volume (which leads to a finite four-dimensional Planck scale). Thus, the system of equations (S0.Ex1) and (9) describes the generalization of the setup considered in , to the case where source terms are included. Similar equations of motion in the global string context, were also considered in Refs. .
Specifically, we will assume that there is a 3-brane at the origin $`\rho =0`$ which is a four-dimensional local string-like topological defect in the six-dimensional spacetime, and has a nonzero stress-energy tensor $`T_B^A`$ parametrized by (6). For example, one may think of the Nielsen-Olesen string solution in the 6d Abelian Higgs model. The source functions describe a continuous matter distribution within the core of radius $`ϵ`$ and vanish for $`\rho >ϵ`$. At the origin we will require that our solution satisfies the boundary conditions
$$\sigma ^{}|_{\rho =0}=0,(\sqrt{\gamma })^{}|_{\rho =0}=1\mathrm{and}\gamma |_{\rho =0}=0.$$
(10)
We have set $`\sigma (0)=A`$, where $`A`$ is a constant. Following , we can integrate over the disk of small radius $`ϵ`$ containing the 3-brane, and define various components of the string tension per unit length as
$$\mu _i=_0^ϵ𝑑\rho \sigma ^2\sqrt{\gamma }f_i(\rho ).$$
(11)
where $`i=0,\rho ,\theta `$. Using the system of equations (S0.Ex1) we obtain the following boundary conditions
$$\sigma \sigma ^{}\sqrt{\gamma }|_0^ϵ=\frac{1}{2M_6^4}(\mu _\rho +\mu _\theta ),$$
(12)
and
$$\sigma ^2(\sqrt{\gamma })^{}|_0^ϵ=\frac{1}{M_6^4}(\mu _0+\frac{1}{4}\mu _\rho \frac{3}{4}\mu _\theta ),$$
(13)
where it is understood that the limit $`ϵ0`$ is taken. By analogy with string defects in four dimensions, $`\mu _\rho +\mu _\theta `$ can be referred to as the Tolman mass (per unit length) . Its nonzero value in four dimensions gives rise to the Melvin branch for local string defects . Similarly, the analogous equation of (13) in four dimensions is related to the string angular deficit . Thus, with these general conditions, any metric solution to the Einstein equations with sources will lead to nontrivial relationships between the components of the string tension per unit length.
Let us now restrict to the case where the four-dimensional cosmological constant $`\mathrm{\Lambda }_{phys}=0`$, and look for a solution outside the core $`(\rho >ϵ)`$ of the form
$$\sigma (\rho )=e^{c\rho }.$$
(14)
Note that we have chosen the arbitrary integration constant, which corresponds to an overall rescaling of the coordinates $`x^\mu `$, such that $`\underset{ϵ0}{lim}\sigma (ϵ)=1`$. Then, a solution to the coupled set of equations (S0.Ex1) can be found with $`\gamma (\rho )=R_0^2\sigma (\rho )`$ and
$$c=\sqrt{\frac{2}{5}\frac{(\mathrm{\Lambda })}{M_6^4}},$$
(15)
where $`R_0`$ is an arbitrary length scale that can be fixed from eqs. (12) and (13). Clearly, the negative exponential solution (14) requires that $`\mathrm{\Lambda }<0`$. If we now demand that the solution (15) is consistent with the boundary conditions (12) and (13), the components of the string tension per unit length must satisfy
$$\mu _0=\mu _\theta +A^2M_6^4,$$
(16)
where $`\mu _\rho `$ remains undetermined. In fact choosing $`\mu _\rho =0`$ gives
$$\mu _\theta =2R_0M_6^4c.$$
(17)
Thus, as long as sources are introduced at the origin $`\rho =0`$ satisfying (16), we obtain a flat Poincare invariant solution in four dimensions. Since the solution is already valid for $`\mathrm{\Lambda }_{phys}=0`$, there is no need to tune the brane tension to the bulk cosmological constant, $`\mathrm{\Lambda }`$ as in the case . However, there is still a tuning in order to satisfy (16).
Having found a solution with a finite volume transverse space the four-dimensional reduced Planck scale now becomes
$$M_P^2=2\pi R_0M_6^4_0^{\mathrm{}}𝑑\rho \sigma ^{3/2}=\frac{5\pi }{3}\frac{\mu _\theta }{\mathrm{\Lambda }}M_6^4,$$
(18)
where we have used the relation (17). The inequality $`M_6M_P`$ is possible by adjusting the string tension or the bulk cosmological constant, and thus could lead to a solution of the gauge hierarchy problem along the lines of .
In order to see that gravity is only localized on the 3-brane, let us now consider the equations of motion for the linearized metric fluctuations. We will only concentrate on the spin-2 modes and neglect the scalar modes, which needs to be taken into account together with the bending of the brane . The vector modes are massive as follows from a simple modification of the results in Ref. . For a fluctuation of the form $`h_{\mu \nu }(x,z)=\mathrm{\Phi }(z)h_{\mu \nu }(x)`$ where $`z=(\rho ,\theta )`$ and $`^2h_{\mu \nu }(x)=m_0^2h_{\mu \nu }(x)`$ we can separate the variables by defining $`\mathrm{\Phi }(z)=_{lm}\varphi _m(\rho )e^{il\theta }`$. The radial modes satisfy the equation
$$\frac{1}{\sigma \sqrt{\gamma }}_\rho \left[\sigma ^2\sqrt{\gamma }_\rho \varphi _m\right]=m^2\varphi _m,$$
(19)
where $`m_0^2=m^2+l^2/R_0^2`$ contains the contributions from the orbital angular momentum $`l`$. The differential operator (19) is self-adjoint provided that we impose the boundary conditions
$$\varphi _m^{}(0)=\varphi _m^{}(\mathrm{})=0,$$
(20)
where the modes $`\varphi _m`$ satisfy the orthonormal condition
$$_0^{\mathrm{}}𝑑\rho \sigma \sqrt{\gamma }\varphi _m^{}\varphi _n=\delta _{mn}.$$
(21)
Using the specific solution (15), the differential operator (19) becomes
$$\varphi _m^{\prime \prime }\frac{5}{2}c\varphi _m^{}+m^2e^{c\rho }\varphi _m=0.$$
(22)
This equation is the same as that obtained for the 5d domain wall solution , except that the coefficient of the the first-derivative term is $`2`$ instead of $`5/2`$. This difference is due to the extra spatial coordinate in the transverse space. When $`m=0`$ we clearly see that $`\varphi _0(\rho )`$ = constant is a solution. Since the modes satisfy the orthonormal condition
$$R_0_0^{\mathrm{}}𝑑\rho e^{\frac{3}{2}c\rho }\varphi _m^{}\varphi _n=\delta _{mn},$$
(23)
a wavefunction in flat space can be defined as
$$\psi _m=e^{\frac{3}{4}c\rho }\varphi _m.$$
(24)
Thus the zero-mode wavefunction becomes
$$\psi _0(\rho )=\sqrt{\frac{3c}{2R_0}}e^{\frac{3}{4}c\rho },$$
(25)
which shows that the zero-mode tensor fluctuation is localized near the origin $`\rho =0`$ and is normalizable.
The contribution from the nonzero modes will modify Newton’s law on the 3-brane. In order to calculate this contribution we need to obtain the wavefunction for the nonzero modes at the origin. The nonzero mass eigenvalues can be obtained by imposing the boundary conditions (20) on the solutions of the differential equation (22). The solutions of (22) are
$$\varphi _m(\rho )=e^{\frac{5}{4}c\rho }\left[C_1J_{5/2}(\frac{2m}{c}e^{\frac{c}{2}\rho })+C_2Y_{5/2}(\frac{2m}{c}e^{\frac{c}{2}\rho })\right],$$
(26)
where $`C_1,C_2`$ are constants and $`J_{5/2},Y_{5/2}`$ are Bessel functions which can be expressed in terms of elementary functions. In the limit that $`\rho \mathrm{}`$, the solutions for nonzero $`m`$ grow exponentially. One way to regulate this behaviour is to introduce a finite radial distance cutoff $`\rho _{\mathrm{max}}`$. Then imposing the boundary conditions (20) at $`\rho =\rho _{\mathrm{max}}`$ (instead of $`\rho =\mathrm{}`$) will lead to a discrete mass spectrum, where for sufficiently large integer $`n`$ we obtain
$$m_nc(n\frac{1}{2})\frac{\pi }{2}e^{\frac{c}{2}\rho _{\mathrm{max}}}.$$
(27)
With this discrete mass spectrum we find that in the limit of vanishing mass $`m_n`$,
$$\varphi _{m_n}^2(0)=\frac{4}{cR_0}m_n^2e^{\frac{c}{2}\rho _{\mathrm{max}}}.$$
(28)
On the 3-brane the gravitational potential between two point masses $`m_1`$ and $`m_2`$, will receive a contribution from the discrete nonzero modes given by
$$\mathrm{\Delta }V(r)G_N\frac{m_1m_2}{r}\underset{n}{}e^{m_nr}\frac{8}{3c^2}m_n^2e^{\frac{c}{2}\rho _{\mathrm{max}}},$$
(29)
where $`G_N`$ is Newton’s constant. In the limit that $`\rho _{\mathrm{max}}\mathrm{}`$, the spectrum becomes continuous and the discrete sum is converted into an integral. Thus the contribution to the gravitational potential becomes
$`\mathrm{\Delta }V(r)`$ $``$ $`{\displaystyle \frac{16G_N}{3\pi c^3}}{\displaystyle \frac{m_1m_2}{r}}{\displaystyle _0^{\mathrm{}}}𝑑mm^2e^{mr}`$ (30)
$`=`$ $`{\displaystyle \frac{32G_N}{3\pi c^3}}{\displaystyle \frac{m_1m_2}{r^4}}.`$ (31)
Thus we see that the correction to Newton’s law from the bulk continuum states grows like $`1/r^3`$. This correction is more suppressed than in 5d, because now the gravitational field of the bulk continuum modes spreads out in one extra dimension and so their effect on the 3-brane is weaker.
Some remarks are now in order:
(i) If different components of the brane tension do not satisfy equation (16), a more general solution to the system of equations (S0.Ex1) can be found along the lines of ref. . Using the parametrisation $`\sigma =z^{4/5}`$, and $`\gamma =\alpha ^2(z^{})^2z^{6/5}`$ (with $`\alpha =4R_0/(5c)`$) the general solution can be written as
$$z(\rho )=\mathrm{exp}(\frac{5}{4}c\rho )+2\beta \mathrm{sinh}(\frac{5}{4}c\rho ),$$
(32)
where $`\beta =0`$ corresponds to the case (14). The general condition for the brane tension components now becomes
$$\mu _0\mu _\theta =\beta (\beta +1)(\frac{3}{2}\mu _\theta \frac{5}{2}\mu _\rho 4\mu _0)+(1+2\beta )^2A^2M_6^4.$$
(33)
The choice of $`\beta <0`$ does not lead to any compactification because $`\sigma `$ diverges at large $`\rho `$. However, $`\beta >0`$ leads to non-compact spaces defined for a finite interval $`0<\rho <\frac{2}{5c}\mathrm{log}(\frac{1+\beta }{\beta })`$ of the type discussed in that may be used as a description of four-dimensional space.
(ii) The metric solution that we have found can also be written in the form
$$ds^2=z^2g_{\mu \nu }dx^\mu dx^\nu z^2R_0^2d\theta ^2\frac{4}{c^2z^2}dz^2.$$
(34)
where $`z=\mathrm{exp}(\frac{c}{2}\rho )`$. In this way we see that the origin $`\rho =0`$ is now mapped to $`z=1`$. The singular source is spread around the circumference of a disk of radius $`R_0`$. This suggests that the 3-brane at the origin $`\rho =0`$ can be interpreted as a wrapped 4-brane where all angular points, $`\theta `$ are identified. In other words, denoting the wrapped 4-brane by $`_4`$, then the 3-brane corresponds to $`_4/S^1`$. While we have given the explicit solution in six dimensions, our solution can be generalized and presumably similar solutions exist at the core of topological defects in higher dimensions where for $`n2`$ transverse dimensions the 3-brane can be identified with $`_{n+2}/S^{n1}`$, where $`_{n+2}`$ has $`n1`$ coordinates spherically wrapped. Again, the corrections to 4d gravity on the 3-brane are expected to be small since the bulk continuum modes live in a higher-dimensional space and by Gauss’s law the effects on the 3-brane are suppressed.
(iii) It is also interesting to study whether our solution (or its generalization in higher dimensions) can be realized in an effective supergravity theory. This would be one step towards embedding the scenario in string theory.
Acknowledgments: We wish to thank S. Dubovsky and P. Tinyakov for helpful discussions. This work was supported by the FNRS, contract no. 21-55560.98.
|
warning/0004/cond-mat0004407.html
|
ar5iv
|
text
|
# Random networks created by biological evolution
## I Introduction
Irregular networks or random graphs composed of units of various kind are very frequent both in nature and society (which is, however, nothing but a special segment of nature). Examples range from vulcanized polymers, silica glasses, force chains in granular materials , mesoscopic quantum wires to food webs , herding effects in economics , world-wide-web links , “small-world” networks of personal contacts between humans and scientific collaboration networks .
Modeling of such networks is not quite easy and analytical results are relatively rare (examples, without any pretence of completeness, can be found in ). Numerical simulations are still one of the principal tools. However, even in the case when the properties of a given class of random networks are relatively well established, either analytically or numerically, as is the case of small-world networks, the serious question remains, why do these networks occur in nature. In other words, what are the dynamical processes, which generate these networks.
Indeed, one can study, for example, various networks of mutual dependence of species in a model of co-evolution , but it is difficult to infer from these studies only, which networks are closer to the reality that the others. In the context of biological evolution models, there was recently a few attempts to let the networks evolve freely, in order to check, which types of topologies might correspond to “attractors” of the process of natural evolution .
The model introduced by us in a preceding Letter is based on extremal dynamics and basically follows the Bak-Sneppen model of biological evolution . Extremal dynamics (ED) models are used in wide area of problems, ranging from growth in disordered medium , dislocation movement , friction to biological evolution . Among them, the Bak-Sneppen (BS) model plays the role of a testing ground for various analytical as well as numerical approaches (see for example ).
The idea of ED is the following. The dynamical system in question is composed of a large number of simple units, connected in a network. Each site of the network hosts one unit. The state of each unit is described by a single dynamical variable $`b`$, called barrier. In each step, the unit with minimum $`b`$ is mutated by updating the barrier. The effect of the mutation on the environment is taken into account by changing $`b`$ also at all sites connected to the minimum site by a network link. Because a perturbation can propagate through the links, we should expect that the topology of the network can affect substantially the ED evolution.
General feature of ED models is the avalanche dynamics. The forward $`\lambda `$-avalanches are defined as follows . For fixed $`\lambda `$ we define active sites as those having barrier $`b<\lambda `$. Appearance of one active site can lead to avalanche-like proliferation of active sites in successive time steps. The avalanche stops, when all active sites disappear again. Generically, there is a value of $`\lambda `$, for which the probability distribution of avalanche sizes obeys a power law without any parameter tuning, so that the ED models are classified as a subgroup of self-organized critical models . (This, of course, can hold only for networks of unlimited size.) The set of exponents describing the critical behavior determines the dynamical universality class the model belongs to.
It was found that the universality class depends on the topology of the network. Usually, regular hyper-cubic networks or Cayley trees are investigated. For random neighbor networks, mean-field solution was found to be exact . Also the tree models were found to belong to the mean-field universality class. A one-dimensional model in which the links were wired randomly with probability decaying as a power $`\mu `$ of the distance was introduced . It was found that the values of critical exponents depend continuously on $`\mu `$. The BS model on a small-world network was also studied .
Recently, BS model on random networks, produced by bond percolation on fully connected lattice, was studied . Two universality classes were found. Above the percolation threshold, the system belongs to the mean-field universality class, while exactly at the percolation threshold, the avalanche exponent is different. A dynamics changing the topology in order to drive the network to critical connectivity was suggested.
There are also several recent results for random networks produced by different kind of dynamics than ED, especially for the threshold networks and Boolean networks .
The geometry of the world-wide web was intensively studied very recently. It was found experimentally that the network exhibits scale-free characteristics, measured by the power-law distribution of connectivities of the sites . Similar power-law behavior was observed also in the actor collaboration graph and in the power grids . A model was suggested to explain this behavior, whose two main ingredients are continual growth and preferential attachment of new links, where sites with higher connectivity having higher probability to receive additional links. The latter feature resembles the behavior of additive-multiplicative random processes, which are well known to produce power-law distributions .
The model introduced in is exactly soluble . Variants including aging of sites , decaying and rewiring links were also studied. The preferential attachment rule, which apparently requires unrealistic knowledge of connectivities of the whole network before a single new link is established, was justified in a very recent work , where higher probability of attachment at highly connected sites results from local search by walking on the network.
In the preceding Letter we concentrated on the self-organized critical behavior and extinction dynamics of a model in which the network changes dynamically by adding and removing sites. It was shown that the extinction exponent is larger than the upper bound for the BS model (given by the mean-field value) and is closer to the experimentally found value than any previous version of the BS model. In the present work we introduce in Sec. II a generalized version of the model defined in and further investigate the self-organized critical behavior in Sec. III. However, our main concern will be about the geometric properties of the network, produced during the dynamics. These results are presented in Sec. IV. Section V makes conclusions from the results obtained.
## II Evolution model on evolving network
We consider a system composed of varying number $`n_\mathrm{u}`$ of units connected in a network, subject to extremal dynamics. Each unit bears a dynamical variable $`b`$. In the context of biological evolution these units are species and $`b`$ represent the barrier against mutations. For the main novelty of our model consists in adding (speciation) and removing (extiction) units, let us first define the rules for extinction and speciation. The rules determining which of the existing units will undergo speciation or extinction will be specified afterwards.
(i) If a unit is chosen for extinction, it is completely removed from the network without any substitution and all links it has, are broken.
(ii) If a unit is chosen for speciation, it acts as a “mother” giving birth to a new, “daughter” unit. A new unit is added into the system and the links are established between the new unit and the neighbors of the “mother” unit: each link of the “mother” unit is inherited with the probability $`p`$ by the “daughter” unit. This rule reflects the fact that the new unit is to a certain extent a copy of the original, so the relations to the environment will be initially similar to the ones the old unit has. Moreover, if a unit which speciates has only one neighbor, a link between “mother” and “daughter” is also established.
The extremal dynamics rule for our model is the following.
(iii) In each step, the unit with minimum $`b`$ is found and mutated. The barrier of the mutated unit is replaced by a new random value $`b^{}`$ taken from uniform distribution on the interval $`(0,1)`$. Also the barriers of all its neighbors are replaced by new random numbers from the same distribution.
The rules determining whether a unit is chosen for extiction or speciation are the following.
(iv) If the newly assigned barrier of the mutated unit $`b^{}`$ is larger than new barriers of all its neighbors, the unit is chosen for speciation. If $`b^{}`$ is lower than barriers of all neighbors, the unit is chosen for extinction. In other cases neither extinction nor speciation occcurs. As a boundary condition, we use the following exception: if the network consists of a single isolated unit only, it is always chosen for speciation.
(v) If a unit is chosen for extinction, all its neighbors which are not connected to any other unit also chosen for extinction. We call this kind of extinctions singular extinctions.
The rule (iv) is motivated by the following considerations. We assume, that well-adapted units proliferate more rapidly and chance for speciation is bigger. However, if the local biodiversity, measured by connectivity of the unit, is bigger, there are fewer empty ecological niches and the probability of speciation is lower. On the other hand, poorly adapted units are more vulnerable to extinction, but at the same time larger biodiversity (larger connectivity) may favor the survival. Our rule corresponds well to these assumptions: speciation occurs preferably at units with high barrier and surrounded by fewer neighbors, extinction is more frequent at units with lower barriers and lower connectivity. Moreover, we suppose that a unit completely isolated from the rest of the ecosystem has very low chance to survive. This leads to the rule (v).
From the rule (iv) alone follows equal probability of adding and removing a unit, because the new random barriers $`b`$ are taken from uniform distribution. At the same time the rule (v) enhances the probability of the removal. Thus, the probability of speciation is slightly lower than the probability of extinction.
The degree of disequilibrium between the two depends on the topology of the network at the moment and can be quantified by the frequency of singular extinctions. The number of units $`n_\mathrm{u}`$ perform a biased random walk with reflecting boundary at $`n_\mathrm{u}=1`$. The bias towards small values is not constant, though, but fluctuates as well.
The above rules are illustrated by the examples shown in Fig. 1. The networks in (a) show the effect of the speciation: a new site is created and some to the links to the mother’s neighbors are established. In (b) the extinction is shown. One of the units is removed also due to a singular extinction (rule (v)). In (c) we illustrate the possibility that in the extinction event the network can be split into several disconnected clusters.
## III Self-organized critical behavior
### A Crossover scaling
The model investigated in the preceding Letter corresponds to the value $`p=1`$. We found that in this case the model is self-organized critical. We defined newly the mass extinctions, as number of units removed during an avalanche. The distribution of mass extinctions obeys a power law with the exponent $`\tau _{\mathrm{ext}}=2.32\pm 0.05`$. In this section we present improved analysis of the data for the self-organized critical behavior.
We measured the distribution of forward $`\lambda `$-avalanches and we observed, contrary to the BS model that two power-law regimes with two different exponents occur. The crossover value $`s_{\mathrm{cross}}`$ which separates the two regimes depends on $`\lambda `$. We observed that the distributions for different $`\lambda `$ collapse onto single curve, if plotted against the rescaled avalanche size $`s/s_{\mathrm{cross}}`$, i. e.
$$P_{\mathrm{fwd}}^>(s)f_{\mathrm{cross}}=g(s/s_{\mathrm{cross}})$$
(1)
where $`g(x)x^{\tau +1}`$ for $`x1`$ and $`g(x)x^{\tau ^{}+1}`$ for $`x1`$. The data are plotted in Fig. 2. For the values of the exponents, we found $`\tau =1.98\pm 0.04`$ and $`\tau ^{}=1.65\pm 0.05`$.
We investigated the dependence of the scaling parameters $`s_{\mathrm{cross}}`$ and $`f_{\mathrm{cross}}`$ on $`\lambda `$ and we found that both of them behave as a power law with approximately equal exponent, $`s_{\mathrm{cross}}f_{\mathrm{cross}}\lambda ^\sigma ^{}`$ with $`\sigma ^{}3.5`$ (see inset in the Fig. 2). The role of critical $`\lambda `$ at which the distribution of forward avalanches follows a power law is assumed by the value $`\lambda =0`$. This result is easy to understand. In fact, in models with fixed (or at least bounded) connectivity $`c`$, the critical $`\lambda `$ is roughly $`1/c`$. As will be shown in the next section, in our case the size of the system and average connectivity grows without limits, and thus the critical $`\lambda `$ tends to zero. Note that it is difficult to see this result without resort to the data collapse (1). Indeed, for any finite time of the simulation, the connectivity and the system size reaches only a limited value and the critical $`\lambda `$ seen in the distribution of forward avalanches has apparently non-zero value.
### B Comparison with the Bak-Sneppen model
If we compare the above findings with the BS model, we can deduce that in our model, with $`p=1`$, the exponent $`\tau `$ corresponds to the usual forward-avalanche exponent, while the exponent $`\tau ^{}`$ is new. The above described scaling (1) breaks down for $`p<1`$ because the connectivity and the system size are limited (cf. next section).
The main difference from the usual BS model is the existence of the second power-law regime, for $`ss_{\mathrm{cross}}`$. It can be particularly well observed for values of $`\lambda `$ close to $`1`$, where the crossover avalanche size $`s_{\mathrm{cross}}`$ is small. We have seen that such avalanches start and end mostly when number of units is close to its minimum value equal to 1. Between these events the evolution of the number of units is essentially a random walk, because singular extinctions are rare . This fact can explain, why the exponent $`\tau ^{}`$ is not too far from the value $`3/2`$ corresponding to the distribution of first returns to the origin for the random walk. The difference is probably due to the presence of singular extinctions.
We measured also the distribution of barriers $`P(b)`$ and the distribution of barriers on the extremal site $`P_{\mathrm{min}}(b_{\mathrm{min}})`$. In Fig. 3 we can compare the results for $`p=1`$ and $`p=0.95`$. The sharp step observed in BS model is absent here, because the connectivity is not uniform. (For comparison, we measured also the barrier distribution in the model of Ref. , where the network is static, but the connectivity is not uniform. Also in that case the step was absent and the distribution was qualitatively very similar to the one shown in Fig. 3.) The large noise level for $`b`$ close to 1 is due to the fact that units with larger $`b`$ undergo mutations rarely.
## IV Network geometry
In this section we analyze the geometrical properties of the network and their dependence on the parameter $`p`$.
### A Size of the network
The first important feature of the networks created by the dynamics of the model is their size, or the number of units within the network. This is a strongly fluctuating quantity, but on average it grows initially and after some time it saturates and keeps fluctuating around some average value, which depend on $`p`$. Fig. 4 shows the probability distribution of number of units $`n_\mathrm{u}`$ for several values of $`p`$. The average number of units $`n_\mathrm{u}`$ was computed from these distributions and its dependence on $`p`$ is shown in the inset of Fig. 4. We can see that the average network size diverges for $`p1`$ as a power law, $`n_u(1p)^{\alpha _n}`$ with the exponent $`\alpha _n0.8`$.
We can see from Fig. 5 that the distribution of number of units has an exponential tail. This corresponds to the fact that the time evolution of the network size is a random walk with reflecting boundary at $`n_\mathrm{u}=1`$, with a bias to lower values, caused by the singular extinctions (for the analysis of biased random walks repelled from zero see e. g. ). From the decrease of average size with decreasing $`p`$ we deduce that the bias due to the singular extinctions has larger effect for smaller $`p`$, i. e. if the new unit created in a speciation event has fewer links to the neighbors.
### B Connectivity
In Fig. 6 we show the probability distribution of the connectivity of network sites $`P_{\mathrm{all}}(c)`$ and distribution of connectivity of the extremal unit $`P_{\mathrm{extremal}}(c)`$. We can observe the tendency that the extremal unit has larger connectivity than average. This is in accord with the findings of Ref. obtained on static networks. It can be also easily understood intuitively. Indeed, in a mutation event the barriers of neighbors of the mutated unit are changed. So, the neighbors have enhanced probability to be extremal in the next time step. Therefore, the sites with higher number of neighbors have larger probability that a mutation occurs in their neighborhood and that they are then mutated in the subsequent step.
The average connectivity $`c`$ computed from the distributions $`P_{\mathrm{all}}(c)`$ is shown in the inset of Fig. 6. We can observe that analogically to the system size also the average connectivity diverges for $`p1`$ as a power law, but the value of the exponent is slightly different. We find $`c(1p)^{\alpha _c}`$ with the exponent $`\alpha _c0.75`$. From the data available we were not able do decide, whether the exponents $`\alpha _n`$ and $`\alpha _c`$ are equal within the statistical noise.
In Fig. 5 we can see that also the distribution of connectivity has an exponential tail, similarly to the distribution of network size. We measured also the joint probability density $`P(n_\mathrm{u},c)`$ for the number of units and the connectivity. The result is shown as a contour plot in Fig. 7. We can see that also for large networks (large $`n_\mathrm{u}`$) the most probable connectivity is small and nearly independent on $`n_\mathrm{u}`$. This means that the overall look of the network created by the dynamics of our model is that there are a few sites with large connectivity, surrounded by many sites with low connectivity.
An interesting observation can be drawn from the results shown in Fig. 8. It depicts the joint probability density as a function of connectivity at fixed system sizes. We can see that for smaller system sizes, closer to the average number of units, the distribution is exponential, while if we increase the system size a power-law dependence develops. For example for the system size fixed at $`n_\mathrm{u}=170`$ we observe a power-law behavior $`P(n_\mathrm{u},c)c^\eta `$ nearly up to the geometric cutoff, $`c<n_\mathrm{u}`$. The value of the exponent was about $`\eta 2.3`$.
This finding may be in accord with the power-law distribution in growing networks . Indeed, in our model the power-law behavior applies only for networks significantly larger than the average size. Such networks are created during time-to-time fluctuations leading to temporary expansion of the network. So, the power-law is the trace of expansion periods in the network evolution, corresponding to continuous growth in the model of . The preferential attachment, which is the second key ingredient in has also an analog in our model; highly connected units are more likely to be mutated, as was already mentioned in the discussion of Fig. 6. However, here the preference of highly connected sites is a dynamical phenomenon, resulting from the extremal dynamics rules of our model.
### C Clusters
As noted already in the Sec. II, the network can be split into several disconnected clusters. The clusters cannot merge, but they may vanish due to extinctions. We observed qualitatively that after initial growth the number of clusters exhibits stationary fluctuations around an average value, which increases when $`p`$ approaches 1. We measured both the distribution of the number of clusters and the distribution of their sizes. In Fig. 9 we show the distribution of the number of clusters. The most probable situation is that there is only a single cluster. However, there is a broad tail, which means that even large number of clusters can be sometimes created. The tail has a power-law part with an exponential cutoff. The value of the exponent in the power-law regime $`P(n_c)n_c^\rho `$ was about $`\rho 1.2`$. We have observed that the width of the power-law regime is larger for larger $`p`$. This leads us to the conjecture that in the limit $`p1`$ the number of clusters is power-law distributed.
On the other hand, the distribution of cluster sizes shown in Fig. 10 has maximum at very small values. This is due to two effects. First, already the distribution of network size has maximum at small sizes, and second, if the network is split into many clusters, they have small size and remain unchanged for long time. The reason why small clusters change very rarely (and therefore can neither grow nor disappear) can be also seen from Fig. 10, where the distribution of sizes of the clusters containing the extremal site is shown. The latter distribution is significantly different from the size distribution for all clusters and shows that the extremal site belongs mostly to large clusters. In fact, we measured also the fraction indicating how often the extremal unit is in the largest cluster, if there are more than one cluster. For the same run from which the data shown in Fig. 10 were collected, we found that this fraction is 0.97, i. e. very close to 1. A similar “screening effect” was reported also in the Cayley tree models : the small isolated portions of the network are very stable and nearly untouched by the evolution.
### D Mean distance
An important feature of a random network is also the mean distance $`\overline{d}`$ between two sites, measured as minimum number of links, which should be passed in order to get from one site to the other. In $`D`$-dimensional lattices, the mean distance depends on the number of sites $`N`$ as $`\overline{d}N^{1/D}`$, while in completely random networks the dependence is $`\overline{d}\mathrm{log}N`$. In the small-world networks, the crossover from the former to the latter behavior is observed .
The dependence of the average distance within a cluster on the size of the cluster in our model is shown in Fig. 11. We can observe global tendency to decrease $`\overline{d}`$ when increasing $`p`$. This result is natural, because larger $`p`$ means more links from a randomly chosen site and thus shorter distance to other sites. The functional form of the size dependence is not completely clear. However, for larger cluster sizes, greater than about 25, the dependence seems to be faster than logarithmic, as can be seen from the inset in Fig. 11. So, the networks created in our model seem to be qualitatively different from the random networks studied previously, as far as we know.
## V Conclusions
We studied an extremal dynamics model motivated by biological evolution on dynamically evolving random network. The properties of the model can be tuned by the parameter $`p`$, the probability that a link is inherited in the process of speciation. For $`p=1`$ the model is self-organized critical and the average system size and connectivity grows without limits. Contrary to the usual BS model, we find two power-law regimes with different exponents in the statistics of forward $`\lambda `$-avalanches. The crossover avalanche size depends on $`\lambda `$ and diverges for $`\lambda 0`$ as a power law. The reason why the critical $`\lambda `$ is zero in this model is connected with the fact that time-averaged connectivity diverges for $`p=1`$.
We investigated the geometrical properties of the random networks for different values of $`p`$. The average network size and average connectivity diverge as a power of $`1p`$. The probability distribution of system sizes has an exponential tail, which suggests that the dynamics of the system size is essentially a biased random walk with a reflecting boundary, The value of the bias grows with decreasing $`p`$. The joint distribution of size and connectivity shows that even for large network sizes the most probable connectivity is low. Hence, there are few highly-connected sites linked to the majority of sites with small connectivity. Moreover, the situations where the system size is far above its mean value are characterized by power-law distribution of connectivity, like in the models of growing networks with preferential attachment.
The network can consist of several mutually disconnected clusters. Even though the most probable situation contains only a single cluster, the distribution of cluster numbers has a broad tail, which shows a power-law regime with exponential cutoff. We observed the “screening effect”, characterized by very small probability that the extremal site is found in any other cluster than the largest one. So, there is a central large cluster, where nearly everything happens, surrounded by some small peripheral clusters, frozen for the major part of the evolution time.
We measured also the mean distance measured along the links within one cluster. The distance grows very slowly with the cluster size; however, the increase seems to be faster than logarithmic.
Summarizing, we demonstrated that the extremal dynamics, widely used in previous studies on macroevolution in fixed-size systems is useful in creating random networks of variable size. It would be of interest to compare the properties of the networks created in our model with food webs and other networks found in the nature. For example the studies of food webs in isolated ecologies give for network sizes about 30 average connectivities in the range from 2.2 to 9, which is not in contradiction with the findings of our model. However, more precise comparisons are necessary for any reliable conclusions about real ecosystems.
## Acknowledgments
We wish to thank K. Sneppen, A. Markoš and A. Pȩkalski for useful discussions.
|
warning/0004/hep-ph0004084.html
|
ar5iv
|
text
|
# Diffraction at the LHC – antishadow scattering?
## Introduction
Soft hadronic interactions observe a time oscillating pattern in the interest from a high-energy physics community. The peaks of the interest coincide as usual with the beginning of the new machine operation. Nowadays RHIC is preparing for operation and LHC would start to provide first results in the not too distant future. Under these circumstances the interest to experimental and theoretical studies in this field is increasing.
There are many open problems in hadron physics at large distances and their importance has not been overshadowed by the exciting expectations of the new particles discoveries in the newly opening energy range of the LHC.
The most global characteristic of the hadronic collision is the total cross–section and the most important problem here is the nature of the total–cross section rising energy dependence. There are various approaches which provide total cross-section rising with energy but the underlying microscopic mechanism leading to this increase remains obscure. However, the growing understanding how QCD works at large distances could finally lead to a final explanation of this longstanding problem .
In this connection the TOTEM experiment approved recently at the LHC could be more valuable than just a tool for checking numerous model predictions and background and luminosity estimates. It could have a definite discovery potential and our main goal in this note is to discuss one of the such aspects related to the possible observation of the antishadow scattering mode at the LHC.
## 1 When will asymptotics be seen?
The answer on the above question currently is model dependent. There are many model parameterizations for the total cross-sections using $`\mathrm{ln}^2s`$ dependence for $`\sigma _{tot}(s)`$. This implies the saturation of the Froissart–Martin bound and what is unnatural the presence of the asymptotical contributions already at the very moderate energies. On the other side the power-like parameterizations of $`\sigma _{tot}(s)`$ neglect the Froissart–Martin bound considering it as a matter of the distant unknown asymptopia.
It seems that the both approaches are limited and their limitations reflect the real energy range available for the analysis of the experimental data. For example, it is not clear whether the power–like parameterizations respect unitarity limit for the partial–wave amplitudes $`|f_l(s)|1`$. We are keeping in mind here only the accelerator data (cosmic ray data will be briefly commented below).
Unitarity is an important principle and the unitarization procedure of some input power-like “amplitude” leads to the complicated energy dependence of $`\sigma _{tot}(s)`$ which can be approximated by the various functions depending on the particular energy range under consideration. Moreover, unitarity implies the appearance of the new scattering mode – antishadow (see and references therein). Here we provide numerical estimates at LHC energies based on the $`U`$-matrix unitarization method and the particular model for $`U`$-matrix and argue that antishadow mode could be revealed already at the LHC energy $`\sqrt{s}=14`$ TeV.
## 2 Antishadow scattering at LHC
In the impact parameter representation the unitarity equation written for the elastic scattering amplitude $`f(s,b)`$ at high energies has the form
$$Imf(s,b)=|f(s,b)|^2+\eta (s,b)$$
(1)
where the inelastic overlap function $`\eta (s,b)`$ is the sum of all inelastic channel contributions. It can be expressed as a sum of $`n`$–particle production cross–sections at the given impact parameter
$$\eta (s,b)=\underset{n}{}\sigma _n(s,b).$$
(2)
Unitarity equation has the two solutions for the case of pure imaginary amplitude:
$$f(s,b)=\frac{i}{2}[1\pm \sqrt{14\eta (s,b)}].$$
(3)
Eikonal unitarization with pure imaginary eikonal corresponds to the choice of the particular solution with sign minus.
In the $`U`$–matrix approach the form of the elastic scattering amplitude in the impact parameter representation is the following:
$$f(s,b)=\frac{U(s,b)}{1iU(s,b)}.$$
(4)
$`U(s,b)`$ is the generalized reaction matrix, which is considered as an input dynamical quantity similar to eikonal function.
Inelastic overlap function is connected with $`U(s,b)`$ by the relation
$$\eta (s,b)=\frac{ImU(s,b)}{|1iU(s,b)|^2}.$$
(5)
Construction of particular models in the framework of the $`U`$–matrix approach proceeds the standard steps, i.e. the basic dynamics as well as the notions on hadron structure are used to obtain a particular form for the $`U`$–matrix.
However, the two unitarization schemes ($`U`$–matrix and eikonal) lead to different predictions for the inelastic cross–sections and for the ratio of elastic to total cross-section. This ratio in the $`U`$–matrix unitarization scheme reaches its maximal possible value at $`s\mathrm{}`$, i.e.
$$\frac{\sigma _{el}(s)}{\sigma _{tot}(s)}1,$$
(6)
which reflects in fact that the bound for the partial–wave amplitude in the $`U`$–matrix approach is $`|f(s,b)|1`$ while the bound for the case of imaginary eikonal is (black disk limit): $`|f(s,b)|1/2`$.
When the amplitude exceeds the black disk limit (in central collisions at high energies) then the scattering at such impact parameters turns out to be of an antishadow nature. In this antishadow scattering mode the elastic amplitude increases with decrease of the inelastic channels contribution.
The shadow scattering mode is considered usually as the only possible one. But the two solutions of the unitarity equation have an equal meaning and the antishadow scattering mode could also appear in the central collisions first as the energy becomes higher. The both scattering modes are realized in a natural way under the $`U`$–matrix unitarization despite the two modes are described by the two different solutions of unitarity.
Appearance of the antishadow scattering mode is consistent with the basic idea that the particle production is the driving force for elastic scattering. Indeed, the imaginary part of the generalized reaction matrix is the sum of inelastic channel contributions:
$$ImU(s,b)=\underset{n}{}\overline{U}_n(s,b),$$
(7)
where $`n`$ runs over all inelastic states and
$$\overline{U}_n(s,b)=d\mathrm{\Gamma }_n|U_n(s,b,\{\xi _n\}|^2$$
(8)
and $`d\mathrm{\Gamma }_n`$ is the $`n`$–particle element of the phase space volume. The functions $`U_n(s,b,\{\xi _n\})`$ are determined by the dynamics of $`2n`$ processes. Thus, the quantity $`ImU(s,b)`$ itself is a shadow of the inelastic processes. However, unitarity leads to self–damping of the inelastic channels and increase of the function $`ImU(s,b)`$ results in decrease of the inelastic overlap function $`\eta (s,b)`$ in accord with Eq. (5) when $`ImU(s,b)`$ exceeds unity.
Let us consider the transition to the antishadow scattering mode . With conventional parameterizations of the $`U`$–matrix the inelastic overlap function increases with energies at modest values of $`s`$. It reaches its maximum value $`\eta (s,b=0)=1/4`$ at some energy $`s=s_0`$ and beyond this energy the antishadow scattering mode appears at small values of $`b`$. The region of energies and impact parameters corresponding to the antishadow scattering mode is determined by the conditions $`Imf(s,b)>1/2`$ and $`\eta (s,b)<1/4`$. The quantitative analysis of the experimental data gives the threshold value: $`\sqrt{s_0}2`$ TeV.
Thus, the function $`\eta (s,b)`$ becomes peripheral when energy is increasing. At such energies the inelastic overlap function reaches its maximum value at $`b=R(s)`$ where $`R(s)`$ is the interaction radius. So, beyond the transition threshold there are two regions in impact parameter space: the central region of antishadow scattering at $`b<R(s)`$ and the peripheral region of shadow scattering at $`b>R(s)`$.
The region of the LHC energies is the one where antishadow scattering mode is to be presented. It will be demonstrated in the next section that this mode can be revealed directly measuring $`\sigma _{el}(s)`$ and $`\sigma _{tot}(s)`$ and not only through the analysis of impact parameter distributions.
## 3 Estimates and transition to asymptotics
We use chiral quark model for the $`U`$–matrix . The function $`U(s,b)`$ is chosen in the model as a product of the averaged quark amplitudes
$$U(s,b)=\underset{Q=1}{\overset{N}{}}f_Q(s,b)$$
(9)
in accordance with assumed quasi-independent nature of the valence quark scattering in some effective field. The essential point here is the rise with energy of the number of the scatterers like $`\sqrt{s}`$ (cf. ). The $`b`$–dependence of the function $`f_Q`$ is related to the quark formfactor $`F_Q(q)`$ and has a simple form $`f_Q(b)\mathrm{exp}(m_Qb/\xi )`$, i.e. the valence quarks in the model have a complicated structure with quark matter distribution approximated by the function $`f_Q(b)`$.
The generalized reaction matrix (in a pure imaginary case) gets the following form
$$U(s,b)=ig\left[1+\alpha \frac{\sqrt{s}}{m_Q}\right]^N\mathrm{exp}(Mb/\xi ),$$
(10)
where $`M=_{Q=1}^Nm_Q`$. Here $`m_Q`$ is the mass of constituent quark, which is taken to be $`0.35`$ $`GeV`$, $`N`$ is the total number of valence quarks in the colliding hadrons, i.e. $`N=6`$ for $`pp`$–scattering. The values for the other parameters were obtained in : $`g=0.24`$, $`\xi =2.5`$, $`\alpha =0.5610^4`$. These parameters were adjusted to the experimental data on the total cross–sections in the range up to the Tevatron energy. With such small number of free parameters the model is in a rather good agreement with the data .
For the LHC energy $`\sqrt{s}=14`$ $`TeV`$ the model gives
$$\sigma _{tot}230\text{mb}$$
(11)
and
$$\sigma _{el}/\sigma _{tot}0.67.$$
(12)
Thus, the antishadow scattering mode could be discovered at the LHC by measuring $`\sigma _{el}/\sigma _{tot}`$ ratio which is greater than the black disc value $`1/2`$.
However, the LHC energy is not in the asymptotic region yet; the total, elastic and inelastic cross-sections behave like
$$\sigma _{tot,el}\mathrm{ln}^2\left[g\left(1+\alpha \frac{\sqrt{s}}{m_Q}\right)^N\right],$$
(13)
$$\sigma _{inel}\mathrm{ln}\left[g\left(1+\alpha \frac{\sqrt{s}}{m_Q}\right)^N\right].$$
(14)
True asymptotical regime
$$\sigma _{tot,el}\mathrm{ln}^2s,\sigma _{inel}\mathrm{ln}s$$
(15)
is expected at $`\sqrt{s}>100`$ $`TeV`$.
Another predictions of the chiral quark model is decreasing energy dependence of the the cross-section of the inelastic diffraction at $`s>s_0`$. Decrease of diffractive production cross–section at high energies ($`s>s_0`$) is due to the fact that $`\eta (s,b)`$ becomes peripheral at $`s>s_0`$ and the whole picture corresponds to the antishadow scattering at $`b<R(s)`$ and to the shadow scattering at $`b>R(s)`$ where $`R(s)`$ is the interaction radius:
$$\frac{d\sigma _{diff}}{dM_X^2}\frac{8\pi g^{}\xi ^2}{M_X^2}\eta (s,0).$$
(16)
The parameter $`g^{}<1`$ is the probability of the excitation of a constituent quark during interaction. Diffractive production cross–section has familiar $`1/M_X^2`$ dependence which is related in this model to the geometrical size of excited constituent quark.
At the LHC energy $`\sqrt{s}=14`$ $`TeV`$ the value of the single diffractive inelastic cross-sections is limited by the value
$$\sigma _{diff}(s)2.4\text{mb}.$$
(17)
The above predicted values for the global characteristics of $`pp`$ – interactions at the LHC differ from the most common predictions of the other models. First of all total cross–section is predicted to be twice as much of the standard predictions in the range 95-120 mb and it also overshoots the existing cosmic ray data. However, extracting proton–proton cross sections from cosmic ray experiments is model dependent and far from straightforward (see, e.g. and references therein). Those experiments measure the attenuation lengths of showers initiated by the cosmic particles in the atmosphere and are sensitive to the model dependent parameter called inelasticity. So the disagreement of the particular model with the cosmic ray measurements means that the data should be recalculated in the framework of this model and in addition assumptions on the energy dependence of inelasticity should be involved also.
## 4 Discussions and conclusion
The main goal of this note is to point out that the antishadow scattering mode at the LHC can be detected measuring elastic to total cross section ratio which is predicted to be greater than the black disc limit $`1/2`$. The considered model estimates also the total cross section values significantly higher than the values conventional parameterizations provide.
The studies of soft interactions at the LHC energies can lead to the discoveries of fundamental importance. The genesis of hadron scattering with rising energy can be described as transition from the grey to black disc and eventually to black ring with the antishadow scattering mode in the center. It is worth noting that the appearance of the antishadow scattering mode at the LHC implies a somewhat unusual scattering picture. At high energies the proton should be represented as a very loosely bounded composite system and it appears that this system has a high probability to reinstate itself only in the central collisions where all of its parts participate in the coherent interactions. Therefore the central collisions are mostly responsible for elastic processes while the peripheral ones where only few parts of weekly bounded protons are involved result mainly in the production of secondary particles. This leads to the peripheral impact parameter profile of the inelastic overlap function. The above picture would imply interesting consequences for the multiplicities in hadronic collisions, i.e. up to the threshold energy $`s_0`$ the picture will correspond to the fragmentation concept which supposes larger multiplicity for the higher value of momentum transfer. The increase of the mean multiplicity in hadron interactions with $`t`$ is in agreement with the hadronic experimental data. However, when the energy becomes greater than $`s_0`$ and antishadow mode develops, momentum transfer dependence of multiplicity would change. Loosely speaking the picture described above correspond to the scattering of extended objects at lower energies and transition to the scattering of weakly bounded systems at higher energies. This picture has an illustrative value and is in general compliance with asymptotic freedom of QCD and parton picture.
Finally, we would like to note that the numerical predictions depend on the particular choice of the model for the $`U`$-matrix, but appearance of the antishadow scattering mode is an inherent feature of the considered approach.
## Acknowledgements
Authors are grateful to W. Kienzle, A. Krisch, W. Lorenzon and V. Roinishvili for the interesting discussions. This work was supported in part by the RFBR Grant No. 99-02-17995.
|
warning/0004/hep-th0004073.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
The advent of D-branes and orientifolds in string theory gave new tools to study gauge theories. The vacua of gauge theories with classical groups, compactified on tori with commuting holonomies can be straightforwardly described by configurations of D-branes and orientifold planes (see and references quoted there). For theories with unitary or symplectic groups, holonomies are specified in the fundamental representation, for theories with orthogonal groups the holonomies are in the vector representation. All these representations still have a non-trivial centre.
When fields are invariant under the centre, typical for conventional open string perturbation theory, one can allow for holonomies that commute up to an element of the centre, as was first considered for gauge theory by ’t Hooft in terms of so-called twisted boundary conditions. But it was only relatively recent that this extra freedom was first considered in the context of string theory. In Witten studied the case of $`SO(4N)`$ on 2– and 3–tori, which can be described by a configuration of D-branes on an orientifold.
In the appendix of the same paper, Witten used a construction involving D-branes on an orientifold to show that for orthogonal groups the moduli space of flat connections on the 3–torus with periodic boundary conditions has an extra component not considered before, and that this seems to solve an old problem concerning the computation of the Witten index . Motivated by this, various authors have subsequently shown the necessary existence of extra vacuum components for exceptional groups, so as to solve the Witten index problem for these groups, for which no D-brane construction is available. The authors of also included the case of general boundary conditions, likewise demonstrating a richer structure than considered earlier in the computation of the Witten index with twisted boundary conditions . It has remained a challenge to translate all these results into configurations of D-branes and orientifold planes, which allowed Witten to make his discovery for the orthogonal groups . Here we close the circle by addressing this translation for all classical groups, with arbitrary boundary conditions.
The fact that in standard open string perturbation theory, all representations are conjugate to the adjoint (and therefore invariant under the centre) is believed to be false outside perturbation theory . It is argued that the full gauge group is actually $`Spin(32)/_2`$, and that also spinorial representations occur. But even configurations that would be consistent for $`Spin(32)/_2`$-gauge theory, can be shown to be inconsistent for the string theory by more subtle arguments . However, in this paper we wish to elucidate the underlying gauge group structure, which we stress is interesting in its own right. It is also an essential step towards a cleaner, and more complete derivation of the string consistency conditions governing orientifolds. We should note however, that there will be important modifications to our results on allowed gauge groups and representations in orientifold compactifications in $`D<10`$ upon imposition of the string consistency conditions.
The main part of this paper is devoted to a discussion of compactification with orthogonal and symplectic groups on 2– and 3–tori, with twisted boundary conditions (as previously analysed in ), but we will also discuss $`U(n)`$ theory with various boundary conditions. After one T-duality, these theories correspond to configurations of branes and orientifold fixed planes on the Möbius strip, the Klein bottle, and tori that are not rectangular. We derive how to T-dualise the Möbius strip and Klein bottle in the direction orthogonal to the first T-duality. This leads to orientifolds with fixed planes of different type, much like in . With these methods every possible flat connection for symplectic gauge groups or orthogonal gauge groups allows a translation in terms of a configuration of D-branes on an orientifold. Holonomies and possible enhanced symmetry groups are easily read off from the configuration.
We will open with a discussion of a $`U(n)`$-theory on a 2–torus with special boundary conditions: Along one of the directions of the 2–torus there is a holonomy that is not an element of $`U(n)`$.
## 2 The T-dual of a cross-cap
We start by considering $`U(n)`$ theory on a circle. $`U(n)`$ can be embedded in an open string theory by attaching Chan-Paton charges on the ends of oriented strings. Compactifying this string theory on a circle and applying a T-duality transformation, we obtain a configuration of $`n`$ D-branes that are transverse to a dual circle, each intersecting the dual circle in one point. The location of the D-branes is controlled by the holonomy $`\mathrm{\Omega }_1`$ along the circle in the original theory. We are interested in configurations with discrete symmetries. The discrete symmetries of the circle are the shift symmetries, shifting the circle by an angle $`2\pi q`$ with $`q`$ a rational number, and the order $`2`$ reflection symmetry. Only for specific choices of the original holonomy will the D-brane configuration respect one or some of these symmetries. In this section, we are interested in the reflection symmetry.
The reflection on the circle has two fixed points, which will be taken to be at $`X=0`$ and $`X=\pi R`$ ($`X`$ being the coordinate along the circle, and $`2\pi R`$ its circumference). This can always be arranged: in the original theory we had a holonomy in $`U(n)`$, which is locally equivalent to $`U(1)\times SU(n)`$. The holonomy for the $`U(1)`$-factor can be chosen arbitrarily, since it does not couple to anything. In the dual theory this corresponds to an overall translation, which we use to set the coordinates of the fixed points to the above values.
Now compactify in addition on another circle of radius $`R^{}`$ with holonomy $`\mathrm{\Omega }_2`$ along this circle. The standard formalism assumes holonomies that can be diagonalised within the group. In case we have the above $`_2`$ symmetry, we may consider a holonomy that includes the $`_2`$ reflection. Gluing the circle to a reflected circle upon going around the second cycle, one does not obtain a 2–torus, but a Klein bottle. Instead of a non-trivial line bundle over a circle, we will represent the Klein bottle here as a cylinder of length $`\pi R`$, circumference $`4\pi R^{}`$, bounded by two cross-caps at the end, the cross-cap being an identification over half the period of the circle.
The D-branes are wrapped around the cylinder, parallel to the cross-caps. There are two possibilities, controlled by the holonomy $`\mathrm{\Omega }_1`$ in the original theory. D-branes in the bulk (away from the cross-caps) represent branes that were reflected into their image. In this representation, D-brane and image are represented as one brane (which is in this sense a brane pair). In the original theory there can also be D-branes at the fixed point(s) of the $`_2`$-reflection. In this representation of the Klein-bottle, they are located at the cross-cap. Under a smooth deformation of the original holonomy, only even numbers of D-branes can move away from the cross-cap. Hence for $`U(n)`$ with $`n`$ odd there is at least one brane fixed under the $`_2`$ reflection and therefore stuck to a cross-cap. For $`n`$ even there are two possibilities: the number of branes at each cross-cap is either even or odd. In the latter situation there is at least one brane at each cross-cap.
The above is reminiscent of the situation for orientifold planes. For orientifolds, a brane and an image brane on the double cover are mapped to a brane-pair in the orientifold. There is also the possibility of single branes being stuck at an orientifold plane (in the case of $`O^{}`$ planes. By $`O^{}`$ we denote the orientifold plane that gives orthogonal gauge symmetry, and $`O^+`$ is an orientifold plane that gives symplectic gauge symmetry).
The above configuration can be interpreted in terms of the original gauge theory. $`U(n)`$ is locally $`U(1)\times SU(n)`$, and the $`U(1)`$ background is fixed. For $`n>2`$, $`SU(n)`$ possesses an outer automorphism, which, in a suitable representation, corresponds to complex conjugation. We will be working in the fundamental representation, and denote complex conjugation as $`C`$ with action
$$C:UU^{}USU(n)$$
(1)
One can extend this action to $`U(n)`$ as $`C`$ also has a simple action on $`U(1)`$, and now one may also extend to $`n2`$. One normally considers holonomies taking values in the gauge group, which corresponds to combining a translation in space with the action of an inner automorphism (i.e. a conjugation) on the group. One may also consider a holonomy that corresponds to an outer automorphism, and this is precisely what we are doing in the above. The outer automorphism $`C`$ can be combined with an inner one, say conjugation with a group element $`A`$. To avoid ambiguities we require that $`AC=CA`$, which is true if $`A`$ is real, that is $`AO(N)`$. The holonomy $`\mathrm{\Omega }_2`$ combines the action of $`C`$ with conjugation with $`A`$, and we denote it as $`\mathrm{\Omega }_2=AC`$, with $`A`$ in the fundamental representation of $`U(n)`$, and $`C`$ the operator that implements complex conjugation. The holonomy $`\mathrm{\Omega }_1`$ is an “ordinary” holonomy, and we write $`\mathrm{\Omega }_1=B`$, with $`B`$ an element of $`U(n)`$ in the fundamental representation. $`\mathrm{\Omega }_1`$ should commute with $`\mathrm{\Omega }_2`$, which is solved by taking $`B`$ commuting with $`A`$ and $`BO(N)`$ . Continuous variation of the $`U(1)`$-background is incompatible with complex conjugation; in the D-brane picture this corresponds to the fact that a global translation on the D-branes is incompatible with the reflection for generic cases.
By conjugation with $`O(n)`$ matrices, we may transform $`A`$ and $`B`$ to a block diagonal form with $`2\times 2`$ blocks of the form
$$\left(\begin{array}{cc}\hfill \mathrm{cos}\varphi & \hfill \mathrm{sin}\varphi \\ \hfill \mathrm{sin}\varphi & \hfill \mathrm{cos}\varphi \end{array}\right)$$
(2)
on the diagonal, and some $`1`$’s and $`1`$’s as remaining diagonal entries. In the following, we will take $`A`$ and $`B`$ to be of this standard form.
We wish to T-dualise the cylinder with the two cross-caps in the direction of the circle. Ignoring the cross-caps one would roughly expect this to lead to a dual theory on a cylinder. The inclusion of the cross-caps can be analysed by examining the symmetries of the original theory.
$`A`$ and $`B`$ are elements of the vector representation of $`O(n)`$, and in particular their eigenvalues occur in pairs: If $`\mathrm{exp}\mathrm{i}\varphi `$ is an eigenvalue, then so is $`\mathrm{exp}\mathrm{i}\varphi `$. The ordering is unimportant as there are symmetries that allow the exchange of $`\mathrm{exp}\mathrm{i}\varphi `$ and $`\mathrm{exp}\mathrm{i}\varphi `$, for every $`\varphi `$ separately. If $`A`$ and $`B`$ where holonomies for an $`O(n)`$-theory in an orientifold description this symmetry would be simply the orientifold projection itself. This suggests that also for this $`U(n)`$-theory the dual should be some orientifold.
The radius of the dual theory is expected to be $`1/(2R^{})`$, half the “normal” radius. The coordinates of the D-branes in this theory reflect the eigenvalues of the holonomies in the original theory. Naively mapping these onto the dual circle suggests a circle of radius $`1/R^{}`$, which seems to lead to a contradiction. The resolution to this paradox lies in the presence of the operator $`C`$. If $`\mathrm{\Omega }_i`$ are the holonomies for a certain theory, then the holonomies $`\mathrm{\Omega }_i^{}=g\mathrm{\Omega }_ig^1`$ with $`g`$ some element of $`U(n)`$ represent the same theory. Consider the set of diagonal matrices with entries $`\pm 1`$, $`\pm \mathrm{i}`$ on the diagonal that commute with the $`A`$ and $`B`$. Taking $`g`$ to be a specific element from this set has the effect
$$g:(B,AC)(gBg^1,gACg^1)=(B,Ag^2C).$$
This leaves $`\mathrm{\Omega }_1`$ and $`\mathrm{\Omega }_2`$ in standard form, but with $`A`$ replaced by $`Ag^2`$. Hence in this construction, $`A`$ and $`Ag^2`$ have to be identified. $`g^2`$ is an element of $`O(n)`$ that commutes with $`A`$. By a suitable choice of $`g^2`$, any eigenvalue $`\mathrm{exp}\mathrm{i}\varphi `$ of $`A`$ can be mapped to $`\mathrm{exp}\mathrm{i}\varphi =\mathrm{exp}\mathrm{i}(\varphi +\pi )`$. Therefore the periods of the circle and the eigenvalues of the holonomies match, and the dual theory is indeed an orientifold. Now we examine the orientifold planes.
To find maximal symmetry groups we set $`B`$ to either $`\mathrm{𝟏}`$ or $`\mathrm{𝟏}`$. If we set $`A=\mathrm{𝟏}`$, then the surviving symmetry group is the subgroup of $`U(n)`$ that is invariant under $`C`$, which is $`O(n)`$. For another maximal symmetry group, assume $`n`$ to be even for a moment and take $`A`$ to be of block diagonal form with $`2\times 2`$ blocks
$$\left(\begin{array}{cc}\hfill 0& \hfill 1\\ \hfill 1& \hfill 0\end{array}\right),$$
(3)
on the diagonal, and call this matrix $`J`$. The unbroken symmetry group is then the subgroup of $`U(n)`$ of matrices $`U`$ that commute with $`JC`$. $`C`$ transforms $`UU^{}`$, but as $`U`$ is unitary, $`U^{}=(U^1)^T`$, where $`T`$ is for transposed. We may then rewrite the invariance condition to
$$U^TJU=J$$
(4)
which, together with the unitarity condition defines the symplectic group. It is obvious how to generalise to arbitrary holonomies, and odd $`n`$: a holonomy with $`k`$ blocks $`\text{diag}(1,1)`$ and $`k^{}`$ blocks (3) gives rise to $`O(k)\times Sp(k^{})`$-symmetry, completed with some $`U(m)`$-factors, whenever $`m`$ eigenvalues not equal to $`\pm 1`$, $`\pm \mathrm{i}`$ coincide.
The above $`U(n)`$ theory on a Klein-bottle is thus T-dual to an orientifold $`T^2/_2`$, where two of the four orientifold planes are of $`O^{}`$-type and two are of $`O^+`$-type. The holonomy $`B=\pm \mathrm{𝟏}`$ distinguishes two parallel configurations of one $`O^+`$ and one $`O^{}`$-plane, whereas in the theory on the Klein bottle it distinguished the two parallel cross-caps. As a rule of thumb one may therefore state that the dual of the cross-cap is a configuration of one $`O^+`$ and one $`O^{}`$-plane. This fits with the usual charge assignments: opposite charges for the $`O^+`$ and $`O^{}`$ plane versus no charge for the cross-cap. The original theory may have had isolated D-branes at the cross-caps. In the dual theory the isolated branes should be located at the $`O^{}`$-planes, since the $`O^+`$ planes cannot support isolated branes. Examining the holonomies that will lead to such a situation indeed shows this to be the case.
These ideas are independent of whether D-branes are static in the background, or used as “probes”. The above orientifold background is identical to a IIB-orientifold encountered in , but consistency requires absence of D-branes. This suggests to regard this model as a “$`U(0)`$-theory” with a holonomy that includes complex conjugation. Its duality to IIA on a Klein-bottle is obvious from the above. Considering various limits one may also reach other theories discussed in and .
We interpreted the Klein-bottle theory as created by combining a translation with an outer automorphism (complex conjugation). Outer automorphisms can always be divided even if not combined with a translation. Dividing a $`U(n)`$ group by its outer automorphism will give a symplectic or orthogonal theory, where the ambiguity comes from the fact that an outer automorphism may be combined with an inner automorphism to give another outer automorphism. In our case one may consider, instead of $`C`$, an operator $`AC`$ with $`A`$ an element of $`U(n)`$. One should require $`A`$ to commute with $`C`$ and therefore $`AO(n)`$. Consistency also requires that $`AC`$ acting on the group squares to the identity. The group action on itself is always in the adjoint representation, and hence we have the possibilities $`A^{}A=(A^1)^TA=\pm \mathrm{𝟏}`$, so $`A`$ is either symmetric or antisymmetric. One may now copy a standard textbook derivation to show that this leads to either symplectic or orthogonal groups. The reasoning is parallel to that for orientifolds, so one may interpret the introduction of an orientifold plane as quotienting the gauge group by an outer automorphism. With this point made, which is not stressed in the literature, we may also say that in the above a translation is combined with an orientifold action, as the theory in the last chapter of was originally motivated.
## 3 Twisted boundary conditions on the 2–torus
### 3.1 Twist in unitary groups
The $`U(n)`$-gauge group allows a second form of twist. The circle also has discrete shift symmetries by angles $`2\pi qR`$, with $`q`$ a rational number, which can be chosen on the interval $`[0,1)`$. Choose a configuration of the $`n`$ D-branes that respects one or some of these shifts. In that case the number $`q`$ is a multiple of $`1/n`$. Now compactify on a second circle with a holonomy that includes the shift over $`2\pi qR`$. This results in a theory on the 2–torus, not with $`n`$ D-branes, but with $`k=\text{gcd}(qn,n)`$ branes, wrapped $`n/k`$ times around the torus. This theory is naturally interpreted as a $`U(n)`$-theory with twisted boundary conditions . We will not have much new to say on this theory, but mention it for completeness, and to point out some effects that are encountered in other theories as well.
Let $`X_1`$ and $`X_2`$ be the coordinates transverse resp. parallel to the branes. Then this 2–torus is $`^2`$ with coordinates $`(X_1,X_2)`$, quotiented by a lattice generated by the vectors
$$e_1=2\pi (qR_1,R_2)e_2=2\pi (R_1,0)$$
(5)
Now transform to an $`SL(2,)`$-equivalent form. Let $`n^{}=n/k`$. Then $`n^{}`$ and $`qn^{}`$ are integer, and $`\text{gcd}(qn^{},n^{})=1`$. Hence the equation
$$n^{}a+qn^{}b=1,a,b$$
has a solution, which can be found using Euclid’s algorithm. The solution is not unique as $`aa+mqn^{}`$, $`bbmn^{}`$ with integer $`m`$ gives another solution. Use this arbitrariness to select a $`b`$ such that $`0b<n^{}`$. Then change the fundamental domain of the torus by using the $`SL(2,)`$ transformation
$$\left(\begin{array}{c}x^{}\\ y^{}\end{array}\right)=\left(\begin{array}{cc}n^{}& b\\ qn^{}& a\end{array}\right)\left(\begin{array}{c}x\\ y\end{array}\right),$$
(6)
where $`(x,y)`$ are coordinates for the lattice vectors $`xe_1+ye_2`$. Under this transformation the basis vectors transform as
$$2\pi (qR_1,R_2)2\pi (0,n^{}R_2)2\pi (R_1,0)2\pi (R_1/n^{},bR_2)$$
(7)
On this fundamental domain only $`k`$ D-branes (which are in a sense configurations of $`n^{}`$-tuples of branes) are visible. This is analogous to the two different representations of the Klein bottle in the previous section. The set-up is the one considered in , which is argued to lead to a Yang-Mills theory on a non-commutative torus in a suitable limit.
We may T-dualise our original theory back to an open string theory (with Neumann boundary conditions) on a torus, using the standard methods . The resulting theory has a non-zero $`B`$-field (with $`B=q`$ in appropriate units) in the background, as our original theory does not live on a square torus.
Combining the discrete shift over $`2\pi qR`$, with the $`Z_2`$ reflection does not lead to anything new. The resulting transformation is of the form
$$XX+2\pi qR$$
which is just a $`_2`$-reflection, but with other fixed points. This is related to the $`U(n)`$ theory of the previous section by a trivial translation.
### 3.2 Twist in symplectic groups on the 2–torus
Symplectic groups can be realised in string theories by combining the Chan-Paton construction with the gauging of world sheet parity . This gives a theory of unoriented strings. To complete the description of the theory one has to prescribe how world sheet parity acts on the Chan-Paton matrices. If the reflection of the world sheet is combined with the action of an anti-symmetric matrix on the Chan-Paton indices, the resulting theory will have symplectic gauge symmetry.
Compactifying this theory on a circle of radius $`R_1`$ and T-dualising leads to an oriented string theory, living on an interval $`I=S^1/_2`$ of size $`(2R_1)^1`$, bounded by two $`O^+`$-planes. For an $`Sp(k)`$-theory there will be $`k`$ D-brane pairs distributed along the interval. The two $`O^+`$-planes do not allow any freely acting shift. We will instead assume that the D-branes are distributed in a configuration that is invariant under the reflection that exchanges the two $`O^+`$-planes. For odd $`k`$ one brane-pair is fixed in the middle of the interval. Now compactify on another circle of radius $`R_2`$ with a holonomy that implements the $`_2`$-reflection. The resulting compactification manifold is a Möbius strip with an $`O^+`$-plane as edge. For $`k`$ even half of the D-branes are exchanged with the other half on going around the circle. For $`k`$ odd half of $`(k1)/2`$ pairs are exchanged with another $`(k1)/2`$ and one brane-pair is fixed by the $`_2`$ reflection. Another representation of the Möbius strip, is a cylinder of diameter $`2R_2`$ and length $`(4R_1)^1`$. One end of the cylinder ends in a cross-cap, the other end is formed by a single $`O^+`$-plane. On the cylinder there are $`k/2`$ D-branes pairs for $`k`$ even, and $`(k1)/2`$ for $`k`$ odd in which case there is a brane-pair stuck at the cross-cap.
This theory is a $`U(2k)`$-theory as described in section 2 with an extra orientifold plane inserted. The mirror symmetry of the orientifold plane turns the Klein bottle into a Möbius strip. Take the circle that is dual to the circle of radius $`R_1`$ and choose coordinates as follows: we will take the orientifold planes at $`X=0`$ and $`X=\pi /R_1`$, and the fixed points of the $`_2`$-reflection at $`X=\pi /2R_1`$ and $`X=3\pi /2R_1`$. The description from the $`U(n)`$ theory has to be slightly modified, as the fixed points of the $`_2`$ are no longer located at $`X=0`$ and $`X=\pi R`$, as before. The action of the $`_2`$-reflection can be interpreted in the original theory as accomplished by the operator $`(\mathrm{𝟏})C`$, which is complex conjugation combined with multiplying by $`(\mathrm{𝟏})`$. In symplectic theories one projects onto states invariant under $`JC`$, with $`J`$ the matrix composed of $`2\times 2`$-blocks of the form (3), and the invariance condition is (4). In the orientifold projected theory, the operator $`(\mathrm{𝟏})C`$ is identified with $`(\mathrm{𝟏})\text{Ad}J`$, which has as action “conjugate with $`J`$ and multiply with $`\mathrm{𝟏}`$”. Multiplying by $`\mathrm{𝟏}`$ is not an outer automorphism of $`Sp(k)`$ (in fact, the symplectic groups do not posses any outer automorphism at all), and it can be realised by conjugation, as we will show later.
With the appropriate symmetries realised, we can pass from the $`U(n)`$-theory to the symplectic theory as follows. We argued that the $`U(n)`$-theory had as its holonomies $`(\mathrm{\Omega }_1,\mathrm{\Omega }_2)=(B,AC)`$. Replace the operator $`C`$ by $`(\mathrm{𝟏})C`$, and then perform the orientifold projection. The resulting holonomies are then $`(\mathrm{\Omega }_1,\mathrm{\Omega }_2)=(B,A(\mathrm{𝟏})\text{Ad}J)`$. $`\mathrm{\Omega }_1`$ and $`\mathrm{\Omega }_2`$ do not commute, but anticommute. Their eigenvalues can be read of from $`B`$ and $`A`$, but we have to find a way to implement the action of $`\mathrm{𝟏}`$.
Anticommutativity of the holonomies is allowed in symplectic theories, provided all representations of $`Sp(k)`$ have trivial centre (this is the case for all representations one encounters in $`Sp(k)`$ string perturbation theory. These are the adjoint, which is the symmetric two-tensor; a $`k(2k1)1`$ dimensional representation which is the antisymmetric tensor with an extra singlet removed; and the singlet). This theory may also be analysed by the methods of . Here we will reproduce the results of such an analysis by a different method.
The T-dual theory to the Möbius strip is an orientifold $`T^2/_2`$, with the size of the $`T^2`$ being $`(2R_1)^1\times (2R_2)^1`$ (one fourth of the usual size, compare with ). At the four fixed points we find orientifold fixed planes. The original $`O^+`$-plane splits into two $`O^+`$-planes intersecting the torus at a point. The cross-cap will dualise into one $`O^+`$-plane and one $`O^{}`$-plane, so we have a total of 3 $`O^+`$-planes and 1 $`O^{}`$-plane. On the dual we have $`k/2`$ D-brane pairs at arbitrary positions if $`k`$ is even. If $`k`$ is odd, there are $`(k1)/2`$ D-branes whose positions can be chosen freely. The remaining brane pair was stuck at the cross-cap, so in the dual picture there is an isolated brane at an orientifold plane, which should be the $`O^{}`$.
The corresponding holonomies can be read of as follows. A brane pair in the bulk has two coordinates, and each corresponds to four eigenvalues $`\lambda _i,\lambda _i,\lambda _i^1,\lambda _i^1`$ with $`\lambda _i=\mathrm{exp}(2\pi \mathrm{i}X_i/R_i)`$, $`X_i`$ and $`R_i`$ being the coordinate and the radius of the corresponding dimension (the $`O^{}`$ plane is located at $`(X_1/R_1,X_2/R_2)=(1/4,1/4)`$, the remaining $`O^+`$ at $`(0,0)`$, $`(1/4,0)`$, $`(0,1/4)`$). Corresponding to these eigenvalues we have $`2\times 2`$ blocks on the diagonals of the holonomies of the form
$$\left(\begin{array}{cc}\hfill \lambda _1& \hfill 0\\ \hfill 0& \hfill \lambda _1\end{array}\right)\left(\begin{array}{cc}\hfill 0& \hfill \lambda _2\\ \hfill \lambda _2& \hfill 0\end{array}\right)$$
(8)
where the left block appears in one of the holonomies and the other, resulting from multiplying a diagonal block with a block of the form (3) in the other holonomy. There is a second set of blocks with $`(\lambda _1,\lambda _2)`$ replaced by $`(\lambda _1^1,\lambda _2^1)`$. For a single brane located at the $`O^{}`$ plane we get blocks with $`(\lambda _1,\lambda _2)=(\mathrm{i},\mathrm{i})`$. One easily verifies that this prescription leads to anticommuting elements in the fundamental representation of the symplectic group.
On the orientifold $`T^2/_2`$ one should introduce a $`B`$-field which is half-integer valued. For orthogonal groups this is well known, and it is usually deduced from a path-integral argument . It may also be deduced from duality. The Möbius strip we used may be described as the torus $`T^2`$ quotiented by the lattice generated by $`2\pi (0,2R_2)`$ and $`2\pi (R_1/2,R_2)`$, quotiented by an orientifold action that takes $`(X_1,X_2)(X_1,X_2)`$. Omitting the orientifold for a moment, we see that the torus is skew, implying a half-integer value for the $`B`$-field in its dual . The same reasoning applies to a Möbius strip, where the edge is formed by an $`O^{}`$ instead of $`O^+`$ -plane. This corresponds to an orthogonal theory without vector structure, as described in , and reproduced by our analysis later.
The resulting orientifolds describe the moduli space of compactifications of $`Sp(k)`$ theories with twisted boundary conditions. As a check consider the cases $`k=1`$ and $`k=2`$, since as $`Sp(1)/_2=SU(2)/_2=SO(3)`$, and $`Sp(2)/_2=SO(5)`$ these results should be reproduced by other orientifolds. $`Sp(1)`$ with twist corresponds to an orientifold with 3 $`O^+`$-planes, and a single D-brane stuck to the $`O^{}`$ plane. The resulting configuration allows no continuous gauge freedom, in accordance with the standard description of $`SU(2)`$ with twist. The single D-brane at the $`O^{}`$ fixed plane gives $`O(1)=_2`$ residual symmetry; this should be interpreted as the symmetry of the centre of $`SU(2)`$ which is the only symmetry of $`SU(2)`$ that survives the twist.
For $`k=2`$ the dual description consists of a single D-brane-pair on the orientifold with 3 $`O^+`$ and 1 $`O^{}`$-planes. The rank of the unbroken group is 1, and generically it is $`U(1)`$. At the $`O^{}`$-plane this is enhanced to $`O(2)`$, while at any of the three $`O^+`$-planes it is enhanced to $`Sp(1)=SU(2)`$. We will see in the next section that this nicely agrees with the orientifold description of the $`O(5)`$ orientifold corresponding to the $`_2`$-twisted case.
For higher $`k`$ the analysis is similar. For $`k`$ even the generic unbroken group is $`U(1)^{k/2}`$, which can be enhanced to $`U(k/2)`$ at a generic position at the orientifold, to $`Sp(k/2)`$ at one of the three $`O^+`$ planes or $`O(k)`$ at the $`O^{}`$ point. For $`k`$ odd this analysis can be copied while replacing $`k`$ by $`k1`$, with the exception that at the $`O^{}`$ $`O(k)`$ symmetry is possible because of the brane already present there.
### 3.3 Twist in orthogonal groups on the 2–torus
Twist in the orthogonal groups gives a more involved situation and we can distinguish several possibilities. Every orthogonal group has a two-fold cover, so the resulting $`Spin`$-group has at least a $`_2`$ centre. Compactification on a two torus with twist in this $`_2`$ will lead to absence of “spin-structure”: fields in the spin representation are not allowed since the holonomies will not commute in this representation.
For $`SO(N)`$-theories with $`N`$-odd this is all one can do apart from compactification with periodic boundary conditions. For $`N`$ even, $`SO(N)`$ already has a non-trivial centre and the above mentioned $`_2`$ is just a subgroup of the whole centre. For $`N`$ divisible by $`4`$, the centre of $`Spin(N)`$ is $`_2\times _2`$. $`_2\times _2`$ allows three $`_2`$ subgroups (basically each of the $`_2`$ factors, and a diagonal embedding). Of these two are related by the outer automorphism of the $`Spin(N)`$-groups with $`N`$ even. Hence there are two options for twisting by a $`_2`$: The already above mentioned $`_2`$ leading to compactification without spin structure, and a second one, named “compactification without vector structure”. The latter is named so because in this compactification the vector representation is not an allowed one.
For $`Spin(N)`$ with $`N`$ even but not divisible by $`4`$, the centre is $`_4`$. The previously mentioned $`_2`$ is generated by the order $`2`$ element in $`_4`$. It is also possible to twist by an element of $`_4`$ generating the whole centre. We will call this compactification without vector structure, since in this case the vector representation is not an allowed one either. Note however that this twist forbids any representation with a non-trivial centre, so the spin representation should be absent as well. The only representations allowed in this case are conjugate to the adjoint. The results from this section may also be derived with the methods of . We will follow a different route.
#### 3.3.1 No spin structure
Absence of spin-structure does not forbid the vector representation, so one can use an ordinary orientifold $`T^2/_2`$ with 4 $`O^{}`$-planes. The topological non-triviality has to be treated with the technique of Stiefel-Whitney classes, following the appendix of . Absence of spin-structure implies that the second Stiefel-Whitney class is non-vanishing. We will keep on demanding that the first Stiefel-Whitney class vanishes (which implies $`SO(N)`$-symmetry, not only $`O(N)`$), and hence the total Stiefel-Whitney class should be $`w=1+\omega _2`$, with $`\omega _2`$ the 2-form on the 2-torus. For $`SO(N)`$ with $`N`$ odd, this is accomplished by placing 3 branes at the three non-trivial orientifold fixed points, placing $`(N3)/2`$ pairs at arbitrary points. For $`SO(N)`$ with $`N`$ even one distributes 4 branes over all orientifold fixed points and has $`(N/22)`$ pairs at arbitrary locations. These are the only solutions.
It is instructive to compare the cases of $`SO(3)`$ and $`SO(5)`$ without spin structure to the analysis for $`Sp(1)`$ and $`Sp(2)`$ with twist, as $`Spin(3)=Sp(1)`$ and $`Spin(5)=Sp(2)`$. For $`SO(3)`$, 3 isolated D-branes are located at three orientifold planes, and there is no continuous gauge freedom at all, just like in the $`Sp(1)`$ case. There are discrete symmetries $`O(1)^3=_2^3`$, but these just correspond to the $`8`$ diagonal $`O(3)`$-matrices. Of these, 4 are not elements of $`SO(3)`$, and of the remaining 4, 3 lift to elements that anticommute with the holonomies in $`Spin(3)=Sp(1)`$. Only the identity remains, which lifts to the two centre elements of $`Sp(1)`$, which is how the $`_2`$ discrete symmetry there is recovered.
For $`SO(5)`$, we have 3 orientifold planes occupied by one brane each, and a pair of branes at an arbitrary point. Generically the unbroken symmetry is $`U(1)`$, which can be enhanced to $`O(3)`$ at any of the three points where an orientifold plane with brane is sitting. At the remaining orientifold plane $`U(1)`$ is enhanced to $`O(2)`$. Stressing again that $`SO(3)`$ is the double cover of $`Sp(1)`$, we see that this is exactly the same as the $`Sp(2)`$ orientifold with twist.
Further easy examples are $`SO(4)`$ and $`SO(6)`$ without spin structure. For $`SO(4)`$ there is no residual gauge symmetry. The $`_2`$ associated with vector structure acts as twist in both $`SU(2)`$-factors of $`Spin(4)`$, eliminating all gauge freedom. $`SO(6)`$ without spin structure is equivalent to $`Spin(6)=SU(4)`$ with $`_2`$-twist. From the above description this gives a rank 1 subgroup, which can be enhanced to $`O(3)`$ at 4 orientifold-planes. This coincides with the $`SU(4)`$-description, where $`SU(2)`$ is a maximal symmetry group.
#### 3.3.2 No vector structure
The case of absence of vector structure was already analysed by Witten for $`O(4N)`$. Here we present an analysis from a different point of view, which also nicely extends to the case of $`O(4N+2)`$.
Compactifying a string theory with orthogonal gauge symmetry $`SO(2k)`$ on a circle, and T-dualising along this circle, gives a theory on the interval $`I=S^1/_2`$. The interval is bounded by two $`O^{}`$-planes, and on the interval we have $`k`$ pairs of D-branes. We will assume that the $`O^{}`$ planes do not contain any isolated D-branes, since if both of them would be occupied we do not have $`SO(2k)`$ but $`O(2k)`$-symmetry, and if only one of them would be occupied this would indicate $`O(n)`$-symmetry with $`n`$ odd (actually, for $`n`$ odd the following construction is impossible, which is a reflection of the fact that $`SO(n)`$ with $`n`$ odd allows only one kind of twist).
Again the only possible discrete symmetry is $`_2`$ reflection symmetry, and we will henceforth assume that this is realised. Compactifying on an extra circle, with a holonomy implementing this reflection leads again to a theory on the Möbius strip, this time with $`O^{}`$planes on the boundary. We again go to the representation in which the Möbius strip is a cylinder, bounded on one end by an $`O^{}`$ plane, and on the other end by the cross-cap. Notice that for $`k`$ odd, there is a pair of D-branes fixed by the reflection and, on the cylinder it has to be located at the cross-cap. Half the number of the remaining D-branes are visible in this representation. Since we are restricting to $`SO(n)`$-configurations, we can repeat the whole discussion presented for symplectic groups, with the difference that the orientifold planes we insert here will not give symplectic but orthogonal symmetry. From the geometric picture one may again deduce anticommutativity of the holonomies $`\mathrm{\Omega }_1`$ and $`\mathrm{\Omega }_2`$.
We may now T-dualise as before, and obtain an orientifold with two $`O^{}`$-planes from the original $`O^{}`$-plane, and an $`O^+`$ and $`O^{}`$-plane from the cross-cap. This explains the relation between the IIA-theory on a Möbius strip, that is discussed in , and the IIB orientifold in . Both feature in a network of theories describing duals of the CHL-string . The strong coupling limit of IIA theory on the Möbius strip gives M theory on a Möbius strip, which has another weak coupling description as a heterotic $`E_8\times E_8`$ string, with a holonomy interchanging the two $`E_8`$-factors . Via heterotic duality, this corresponds to a compactification of the $`Spin(32)/_2`$-string without vector structure as described in . In the latter paper, the strong coupling description of the heterotic $`Spin(32)/_2`$-string without vector structure is derived to be the IIB-orientifold with a single $`O^+`$ and 3 $`O^{}`$-planes. The $`_2`$ projection causing the reduction of rank of the gauge group is geometrical in the IIA and M theory descriptions (compare with similar constructions in ). The T-duality derived here closes the circle, and relates the mechanism for rank reduction in the IIB-theory to the more transparant one of the IIA-theory.
We may represent the parts of the holonomies corresponding to branes in the bulk in the same way as in the symplectic case. These can be conjugated to matrices in the real vector representation of $`O(n)`$. Readers who prefer the real representation of $`O(n)`$ should substitute for each brane in the bulk the following $`4\times 4`$-blocks
$$\left(\begin{array}{cccc}\mathrm{cos}\varphi _1& 0& \mathrm{sin}\varphi _1& 0\\ 0& \mathrm{cos}\varphi _1& 0& \mathrm{sin}\varphi _1\\ \mathrm{sin}\varphi _1& 0& \mathrm{cos}\varphi _1& 0\\ 0& \mathrm{sin}\varphi _1& 0& \mathrm{cos}\varphi _1\end{array}\right)\left(\begin{array}{cccc}0& \mathrm{cos}\varphi _2& 0& \mathrm{sin}\varphi _2\\ \mathrm{cos}\varphi _2& 0& \mathrm{sin}\varphi _2& 0\\ 0& \mathrm{sin}\varphi _2& 0& \mathrm{cos}\varphi _2\\ \mathrm{sin}\varphi _2& 0& \mathrm{cos}\varphi _2& 0\end{array}\right)$$
(9)
as can be derived straightforwardly. The parameters $`\varphi _i=X_i/2\pi R_i`$ follow from the coordinates of the D-branes on the torus.
For $`k`$ odd, there was an odd number of D-brane pairs at the cross-cap, and hence the $`O^{}`$-plane coming from the cross-cap has to contain an odd number of branes. The other two orientifold points also contain isolated branes. This is possible because the even number of branes that should be on the edge of the Möbius strip (to ensure $`SO(2k)`$-symmetry), may translate into odd numbers of branes at each of the corresponding $`O^{}`$-planes in the dual theory.
Consider a single brane stuck to the $`O^{}`$-plane that came from dualising the cross-cap. As can be seen from the geometric picture, this corresponds to a block $`\text{diag}(\mathrm{i},\mathrm{i})`$ in the holonomy $`\mathrm{\Omega }_1`$, or equivalently, a block
$$\left(\begin{array}{cc}\hfill 0& \hfill 1\\ \hfill 1& \hfill 0\end{array}\right)$$
in the real representation of $`O(n)`$. Demanding anticommutativity with a $`2\times 2`$ block in the holonomy $`\mathrm{\Omega }_2`$ leads to the unique solution $`\text{diag}(1,1)`$ (in the real representation, up to conjugation with an element of $`SO(2)`$). One may also consider a single brane stuck to one of the other $`O^{}`$ planes. This defines a block $`\text{diag}(1,1)`$ in the holonomy $`\mathrm{\Omega }_1`$. Demanding anticommutativity with a second $`2\times 2`$ block leads to two inequivalent possibilities, being
$$\left(\begin{array}{cc}\hfill 0& \hfill 1\\ \hfill 1& \hfill 0\end{array}\right)\left(\begin{array}{cc}\hfill 0& \hfill 1\\ \hfill 1& \hfill 0\end{array}\right)$$
These two possibilities correspond to the two $`O^{}`$-planes that came from dualising the original $`O^{}`$-plane. The point is now that occupying 1 or 2 of the $`O^{}`$ planes by an odd number of branes corresponds to holonomies in $`O(n)`$, but occupying all three at once with an odd number of single branes does give holonomies in $`SO(n)`$. This also gives the interpretation for the orientifold with 3 $`O^{}`$ planes and one $`O^+`$-plane where not all of the $`O^{}`$ planes are occupied; these represent $`O(k)`$ configurations that cannot be represented in $`SO(k)`$. We see that if we demand $`SO(n)`$-symmetry, and occupy one $`O^{}`$-plane with an odd number of branes, we have to occupy all $`O^{}`$ planes by an odd number of branes.
On the resulting dual orientifold we have $`k/2`$ pairs of D-branes if $`k`$ is even, or $`(k3)/2`$ if $`k`$ is odd. For $`k`$ even we obtain back the description of , with the possibility of $`O(k)`$ symmetry at 3 planes, and $`Sp(k/2)`$-symmetry at 1 plane. For $`k`$ odd we have the possibility of $`O(k2)`$ at three planes, and $`Sp((k3)/2)`$ at one plane.
It is again instructive to look at a few examples. $`k=1`$ is impossible in $`SO(2)`$. $`k=2`$ corresponds to $`SO(4)`$-theory without vector structure. Since $`Spin(4)`$ is $`SU(2)\times SU(2)`$, this corresponds to twist in one of the $`SU(2)`$ factors, and arbitrary holonomies in the other $`SU(2)`$-factor. In the orientifold description, we have the possibility of enhanced symmetries $`O(2)`$ and $`Sp(1)`$. $`Sp(1)=SU(2)`$ obviously corresponds to the unbroken second factor. The $`O(2)`$’s correspond to situations in which the holonomies in the second factor are $`(1,\mathrm{i}\sigma ^3)`$, $`(\mathrm{i}\sigma ^3,1)`$, $`(\mathrm{i}\sigma ^3,\mathrm{i}\sigma ^3)`$. The extra parity transformation is due to the fact that $`\mathrm{i}\sigma ^3`$ anticommutes with the elements $`\mathrm{i}\sigma ^{1,2}`$, which lifts to the double cover as commutation symmetry.
$`k=3`$ gives $`SO(6)`$ without vector structure. This corresponds to $`Spin(6)=SU(4)`$ with $`_4`$-twist. The absence of remaining gauge freedom is completely in agreement with the $`SU(4)`$ description.
$`k=4`$ gives $`SO(8)`$ without vector structure, which due to triality should be equivalent to $`SO(8)`$ without spin structure. We certainly do find $`Sp(2)=Spin(5)`$ symmetry in both descriptions. $`O(4)`$ is less visible in the above description of $`SO(8)`$ without spin structure, but this is due to the fact that the parity in $`O(4)`$ is actually not a real symmetry (compare to the $`O(2)`$ symmetry found for $`k=2`$), and $`Spin(4)=SU(2)\times SU(2)`$ can also be obtained outside orientifold fixed planes. The asymmetry in the two description is then due to the fact that they represent different projections from the same moduli space.
## 4 Compactifications on a 3–torus
### 4.1 Commuting triples for orthogonal and symplectic groups
For compactifications on a 3–torus with periodic boundary conditions, the vacua are classified by its three holonomies, which should be 3 commuting elements in the gauge group.
Finding such a triple for a symplectic theory amounts to placing a number of D-brane pairs on an orientifold $`T^3/_2`$ where the fixed points are all $`O^+`$-planes. A similar thing can be done for orthogonal theories, with the difference that on the fixed points of the $`_2`$ orientifold action one introduces $`O^{}`$-planes, which can also support single D-branes. Not all such configurations correspond to flat connections with periodic boundary conditions. Alternative boundary conditions can be described by this orientifold, as long as the holonomies commute in the vector representation of $`O(N)`$. For a periodic connection one should however demand that the holonomies commute in $`Spin(N)`$, which is a more severe restriction. To solve which configurations of $`O(N)`$ holonomies lift tusefulo $`Spin(N)`$ holonomies, one may calculate the Stiefel-Whitney class for the configurations . Only if the Stiefel-Whitney class is trivial the configuration can be lifted to $`Spin(N)`$.
Solving for these requirements Witten found new vacua for orthogonal gauge theory with periodic boundary conditions on a 3–torus whose existence was explained from the group theory point of view in . A configuration with 7 D-branes on the 7 fixed points excluding the origin of an orientifold, and $`N`$ pairs at arbitrary positions parametrises a periodic connection for $`SO(2N+7)`$-theory. Similarly, a configuration with 8 D-branes distributed over all 8 fixed points, and $`N`$ pairs at arbitrary positions parametrises a periodic connection of $`SO(2N+8)`$-theory. Besides these there are always the flat periodic connections that are smoothly connected to the configuration where all three holonomies are equal to the identity. For $`SO(N)`$ with $`N`$ even, this gives a configuration of only D-brane pairs on the orientifold. For $`N`$ odd there is a single isolated D-brane at the origin of the orientifold. Forgetting about the D-brane pairs for a moment, we see that for each theory there are two solutions, one in which a number $`k`$ of orientifold planes is occupied by single branes ($`k`$ being $`0`$ or $`1`$) and one where $`(8k)`$ orientifold planes are occupied by single branes. This is a useful observation for later.
### 4.2 $`c`$-triples for symplectic groups
For symplectic groups on the 3–torus one may also choose non-periodic boundary conditions. This can be done in various ways, but by using $`SL(3,)`$ transformations on the torus, all possibilities are isomorphic to one standard form. We can choose the standard form to have twist between the holonomies in the $`1`$ and $`2`$ direction, and the third holonomy commuting with the former two. Following , we call a triple of such holonomies a $`c`$-triple, where $`c`$ denotes that the three holonomies only commute up to a (non-trivial) centre element of the gauge group.
From our analysis for twist in symplectic gauge theories on the 2–torus, one easily deduces that the corresponding orientifold description has 6 $`O^+`$-planes and 2 $`O^{}`$planes. The two planes with 3 $`O^+`$ and 1 $`O^{}`$, are distinguished by the eigenvalue $`\pm 1`$ in the third holonomy. Eigenvalues for the third holonomy can be read of in the usual way, with the remark that their multiplicities should be doubled. A configuration for the 2–torus may therefore be imported in either of these planes, corresponding to choosing the third holonomy in $`Sp(k)`$ to be $`\pm \mathrm{𝟏}`$, which are the two elements of the centre of $`Sp(k)`$. One quickly deduces that there are always 2 disconnected possibilities for placing the D-branes in this orientifold background.
First suppose $`Sp(k)`$ symmetry with $`k`$ even. For the description with twist on a 3–torus, this should give $`k/2`$ pairs of D-branes in the above orientifold background. There are two possibilities to distribute the D-branes. First, one can have $`k/2`$ pairs at arbitrary locations on the orientifold. But one can also split one pair, put one D-brane on one $`O^{}`$-plane and the other on the other $`O^{}`$-plane, and have the remaining $`(k/21)`$-pairs at arbitrary locations. Both possibilities are legitimate, since $`Svectorstructurep(k)`$ is simply connected. The conclusion is thus, that $`Sp(k)`$-theory on a 3–torus with twisted boundary conditions has a moduli space of 2 components, one with a rank $`k/2`$ unbroken gauge group, and one with a rank $`k/21`$ unbroken gauge group. We can now perform a Witten index count for this theory, as also performed in : The two components will contribute $`k/2+1`$ and $`k/2`$ to the index giving the total value $`k+1`$ in agreement with both the periodic boundary conditions case, and the infinite volume case .
For $`k`$ is odd the procedure should also be clear. One can place $`(k1)/2`$ pairs of D-branes at arbitrary points in the orientifold background. The single D-brane that is left can go on either of the two $`O^{}`$planes. These are inequivalent possibilities, and hence also in this case the moduli space consists of 2 components. Each of these components contributes $`(k+1)/2`$ to a Witten index calculation, giving also the correct result $`k+1`$ .
Again we check the $`k=1`$ and $`k=2`$ cases. According to the above, $`Sp(1)`$-theory with twist on a 3–torus gives a moduli space of 2 components. On each component the gauge group is completely broken. This is as it should be as $`Sp(1)=SU(2)`$ which, when compactified on a 3–torus with twist has a moduli space that looks like this. We again have the remaining $`O(1)=_2`$ symmetry corresponding to the centre of $`SU(2)`$, which commutes with everything.
Perhaps more interesting is the $`Sp(2)`$-case, where we have one component for which the gauge group is completely broken, and another where a rank 1 group survives. The rank one gauge group is generically $`U(1)`$, but can be enhanced to $`O(2)`$ at two planes or $`Sp(1)`$ at six other planes. This coincides with the description we will find for $`SO(5)=Sp(2)/_2`$, without spin structure.
### 4.3 $`c`$-triples for orthogonal groups
For orthogonal groups on a 3–torus there are more possibilities for the boundary conditions. Like in the case for the 2–torus, we can have absence of either spin- or . By $`SL(3,)`$-transformations on the torus, we can again arrange that the holonomies for the 1 and 2-direction are the ones that do not commute (in the Spin-cover of the group), while the third holonomy does commute with the other two.
In the case of $`SO(4n)`$ there is however a new possibility. For $`Spin(4n)`$ the centre of the gauge group is not cyclic but a product of cyclic groups, being $`_2\times _2`$. Call the generator of the first $`_2`$ $`z_s`$ (with $`s`$ for spin), and the generator of the second $`_2`$ $`z_c`$ ($`c`$ being the standard notation for the second spin-representation). Also define $`z_v=z_sz_c`$. This notation is motivated by the fact that identifying $`z_v\mathrm{𝟏}`$ gives the vector representation.
We can now also impose the following twist conditions on the holonomies:
$$\mathrm{\Omega }_1\mathrm{\Omega }_2=z_s\mathrm{\Omega }_2\mathrm{\Omega }_1\mathrm{\Omega }_2\mathrm{\Omega }_3=z_c\mathrm{\Omega }_3\mathrm{\Omega }_2\mathrm{\Omega }_3\mathrm{\Omega }_1=z_v\mathrm{\Omega }_1\mathrm{\Omega }_3$$
(10)
This can be thought of as a standard form. $`SL(3,)`$-transformations result in an isomorphic moduli-space. We will call this case “spin nor vector”-structure, and treat it separately.
#### 4.3.1 No spin structure
This is the easiest case, provided we use some previously obtained knowledge. From our description of orthogonal theories on a 2–torus without spin structure, a particular case for the 3–torus can be obtained as follows.
For $`SO(k)`$ with $`k`$ odd, place 3 single D-branes at three $`O^{}`$-planes within one plane within the orientifold $`T^3/_2`$ leaving the fixed plane at the origin empty, and place the others in pairs at arbitrary points at the orientifold. For $`k`$ even one should place 4 single D-branes at 4 orientifold fixed planes within one plane of $`T^3/_2`$.
For $`k4`$ there is always a second possibility. Remember from section 4.1 and that a configuration of 8 D-branes distributed at all orientifold planes has a trivial Stiefel-Whitney class. We may “add” this orientifold configuration to another as follows. Take a specific configuration of D-branes at the orientifold. This has a certain Stiefel-Whitney class, which can be thought of as providing a topological classification for the configuration. Now adding 8 more D-branes at the orientifold fixed points will not affect the Stiefel-Whitney class. This is so because the Stiefel-Whitney class of the 8 D-branes is trivial, and the Stiefel-Whitney class of the ”new” configuration may be simply obtained by multiplying the class of the ”old” configuration with that of the added configuration (it is important to realise that Stiefel-Whitney classes are $`_2`$ valued, and that $`1=1\text{mod}2`$, so there is no ordering ambiguity). One may also add or delete any pair of D-branes without affecting the class, also because of its $`_2`$ nature.
We thus obtain the following possibilities: For $`SO(k)`$ with $`k`$ odd, we had 3 single D-branes at three $`O^{}`$-planes. Adding the 8 D-branes and reducing modulo 2, we obtain a configuration of 5 D-branes with the same topological classification as the previous one. The 5 D-branes are precisely at the orientifold planes that were not occupied previously, and in a sense one could speak of a $`_2`$-complement. One can add pairs of D-branes to again obtain an $`SO(k)`$ configuration.
For $`k`$ even one had 4 single D-branes at 4 $`O^{}`$-planes. Taking the $`_2`$-complement, we get an inequivalent configuration with 4 single D-branes at the other 4 $`O^{}`$-planes, with the same topological classification. Of course, afterwards we must add pairs of D-branes to acquire $`SO(k)`$.
We will now discuss several cases. $`k=3`$ corresponds to $`SO(3)=SU(2)/Z_2`$ with twist on the 3–torus. The $`SU(2)`$ description has two components. The $`SO(3)`$-description has also two components, but these cannot be distinguished by their holonomies. In a particular representation, the $`SU(2)`$ holonomies read
$$\mathrm{\Omega }_1=\mathrm{i}\sigma _3;\mathrm{\Omega }_2=\mathrm{i}\sigma _1;\mathrm{\Omega }_3=\pm 1,$$
(11)
but $`\pm 1`$ in $`SU(2)`$ are both projected to the same element of $`SO(3)`$ being the identity.
$`k=4`$ gives $`SO(4)`$ which gives two orientifolds, but some thought will reveal that also in this case there are twice as many components in moduli space. Using that $`Spin(4)=SU(2)\times SU(2)`$, the no-spin-structure condition amounts to twisting both $`SU(2)`$ factors simultaneously. For the third holonomy one has then 4 possibilities, being any combination of plus or minus the identity in each $`SU(2)`$-factor. These 4 possibilities project to only two sets of holonomies in $`SO(4)`$, and hence two orientifold descriptions.
$`k=5`$ gives us $`SO(5)`$ which is interesting because we should be able to reproduce the $`Sp(2)`$-results here. $`SO(5)`$ without spin-structure gives two orientifolds. On one we have 5 fixed D-branes and hence no residual gauge symmetry. On the other we have 3 fixed D-branes and a pair wandering freely. Possible enhanced gauge symmetries are $`O(3)`$ at three points, and $`O(2)`$ at five points. However, all but one of the “parity” symmetries (corresponding to elements with $`det=1`$) in these $`O(n)`$ groups are “fake” in the sense that they correspond to elements that anticommute in $`Spin(5)`$. The remaining $`O(2)`$ corresponds to the $`O(2)`$’s we encountered in the $`Sp(2)`$ case, and the $`SO(3)`$’s map to the $`Sp(1)`$-unbroken subgroups in $`Sp(2)`$. That the multiplicities of these enhanced symmetry groups are only half of those encountered in the $`Sp(2)`$ description reflects the fact that $`SO(5)`$ is a double cover of $`Sp(2)`$, which also translates to the fact that the moduli space of $`Sp(2)`$-triples is a double cover of the space of $`SO(5)`$-triples. The moduli space for the gauge theory is the moduli space of $`Sp(2)`$-triples, as every set of $`SO(5)`$ holonomies has two inequivalent realisations in terms of gauge fields. Note however that here the number of components in moduli space agree; the two $`SO(5)`$-components are double covers of two $`Sp(2)`$ components, not of 4 $`Sp(2)`$ components
$`k=6`$ gives $`SO(6)`$ whose spin cover is $`SU(4)`$. Here there are two equal dimension components in moduli space, both with a rank $`1`$ gauge group which can be enhanced to $`SO(3)`$ at 4 points.
It is easy to perform the Witten index count for $`k5`$ . In these cases we have always two components of the moduli space and two corresponding orientifold representations. For $`k`$ even, both components contribute $`k/21`$, for a total of $`k2`$. For $`k`$ odd, one component contributes $`(k1)/2`$ whereas the second contributes $`(k3)/2`$ for a total of $`k2`$. Of course these answers are as they should be.
#### 4.3.2 No vector structure
From the analysis for the 2–torus we deduce that $`O(2k)`$ without vector structure on a 3–torus corresponds to an orientifold background with 6 $`O^{}`$-planes and 2 $`O^+`$-planes. Again eigenvalues for the third holonomy can be read off in the usual way, except that their multiplicities should be doubled. One obvious solution to the boundary conditions is to import the solution for the 2–torus here.
For $`k`$ even we have seen that a particular solution is given by placing all D-brane pairs at arbitrary points. For the second solution we take as before the $`_2`$-complement. We have not defined how the operation of ”$`_2`$-complement” acts on the $`O^+`$-planes but this is not hard to guess. Since $`O^+`$ planes cannot support isolated D-branes, they should remain empty. Hence the second solution has all $`O^{}`$ planes occupied by one D-brane, and $`k3`$ pairs at arbitrary points. A way to see this is as follows. The smallest group for which the configuration with six isolated branes exists is $`SO(12)`$. One can take the $`SO(6)`$ holonomies $`\mathrm{\Omega }_1`$ and $`\mathrm{\Omega }_2`$ that gave “no vector structure” on the 2–torus (these are unique up to gauge transformations), to construct the $`SO(12)`$-holonomies $`\mathrm{\Omega }_1\mathrm{\Omega }_1`$, $`\mathrm{\Omega }_2\mathrm{\Omega }_2`$ and $`\mathrm{𝟏}\mathrm{𝟏}`$, where $`\mathrm{𝟏}`$ stands for the identity in $`SO(6)`$. That these $`SO(12)`$-matrices satisfy the required boundary conditions is obvious, and liftings to $`Spin(12)`$ can be constructed from the liftings of the $`SO(6)`$-holonomies to $`Spin(6)=SU(4)`$. That $`SO(12)`$ is the smallest group allowing these extra solutions can also be deduced with the techniques from .
For $`k`$ is odd we have 3 $`O^{}`$-planes within one plane occupied by D-branes. A priori one has two possible planes, and actually both give distinct solutions. Note that also in these cases the solutions are each others $`_2`$-complement. For $`k`$ odd all components of the moduli space are isomorphic, as also follows form the analysis of .
$`SO(4)`$ without vector structure on a 3-torus gives only one solution, since there are simply not enough D-branes to realise the second one. This gives a rank 1 unbroken gauge group which can be enhanced to $`Sp(1)`$. A naive calculation of the Witten index would give half of the right answer, but as before, the moduli space consists of 2 components that cannot be distinguished by their holonomies in $`SO(4)`$. We therefore have to multiply the naive value for the index by two, again obtaining the right answer.
$`SO(6)`$ without vector structure gives two solutions, but from $`SU(4)`$-analysis one expects four. Again this is due to the fact that inequivalent solutions exist that cannot be distinguished by their holonomies in $`SO(6)`$. Notice that the gauge group is completely broken, which is as it should be.
A Witten index calculation is straightforward for these theories. For $`SO(4N+2)`$ there are always 4 components (projected to two orientifolds) that are all isomorphic. Each orientifold has $`2N+1`$ branes on it, of which $`3`$ are stuck. The rest should organise in $`N1`$ pairs, giving a rank $`N1`$ gauge group and contribution $`N`$ to the Witten index. With 4 components one obtains the total value $`4N`$, which is indeed the dual Coxeter number for these theories.
For $`SO(4N)`$ one has 2 components (1 orientifold), where no branes are stuck and $`N`$ pairs move freely, giving a contribution of $`2N+2`$ to the index. For $`N3`$ this is not sufficient, but for these cases there exist 2 more components (1 orientifold) having 6 stuck branes, and $`N3`$ pairs at arbitrary points. The total adds up to $`2(N+1)+2(N2)=4N2`$, the right answer.
#### 4.3.3 Spin nor vector structure
For $`SO(4k)`$ there is the possibility of holonomies satisfying equation (10). In some sense this should encompass both the case of no spin- as well as no vector-structure. The orientifold background is as in the case without vector structure, $`T^3/_2`$ with 2 $`O^+`$-planes and 6 $`O^{}`$-planes. On top of this $`2k`$ branes should be distributed.
A clue on the D-brane configuration can be found from T-dualising in the direction of the line connecting the two $`O^+`$ planes. This gives a theory on the product of a circle and an orientifold $`T^2/_2`$ with 3 $`O^{}`$’s and 1 $`O^+`$ plane. This orientifold corresponded to an orthogonal theory without vector structure on the 2–torus. In our case, the gauge group is $`SO(4k)`$, and we know that if there are branes at an orientifold plane, their number should be even. Translating back to the orientifold $`T^3/_2`$, the pair of orientifold planes corresponding to one $`O^{}`$-plane in $`T^2/_2`$ will be occupied either both by an even number of branes, or both by an odd number of branes. This leaves $`8`$ possibilities, 2 of which can be quickly discarded as they correspond to a $`SO(4k)`$-theory without vector, but with spin structure. The remaining possibilities have thus either $`2`$ or $`4`$ branes stuck at $`O^{}`$ planes, and hence $`2k2`$, resp. $`2k4`$ D-branes in the bulk
T-dualising in another direction, along a line connecting an $`O^+`$ and an $`O^{}`$-plane will lead to an orientifold of the form $`((T^2/_2)\times S^1)/_2`$. One can represent this by an orientifold $`T^2/_2`$ with 4 $`O^{}`$ planes, quotiented by a $`_2`$-reflection in one of the points halfway on the line between 2 $`O^{}`$-planes (the multiple possibilities are related by $`SL(2,)`$-transformations). This reflection has a second fixed point. Over the orientifold $`(T^2/_2)/_2`$ one erects a circle everywhere, except at the two fixed points of the $`_2`$-reflection where it is replaced by a cross-cap. On the orientifold $`T^2/_2`$ we should have absence of spin structure, meaning that all $`O^{}`$-planes are occupied by an odd number of branes (remember that $`4k`$ is even). This translates to occupancy of both $`O^{}`$-planes in the quotient $`(T^2/_2)/_2`$. We now have two possibilities; either there are an odd number of branes at both cross-caps, or there are an even number. For the orientifold $`T^3/_2`$, these two possibilities translate into the situations with two $`O^{}`$-planes occupied (cross-caps occupied by even number of branes), and four $`O^{}`$-planes occupied (cross-caps occupied by odd number of branes). Thus both possibilities can be realised.
The 3 possibilities of occupying 2 $`O^{}`$-planes are related by $`SL(2,)`$, as are the 3 possibilities of occupying 4 $`O^{}`$ planes. Only one of each set of possibilities solves the boundary conditions (10) (as these are not $`SL(2,)`$ invariant), and actually, the two possibilities are each others $`_2`$ complement, as before.
We therefore have two orientifolds representing $`SO(4k)`$-theory on a 3–torus with spin- nor vector structure. Each orientifold background is of the form $`T^3/_2`$, with 6 $`O^{}`$ planes, and 2 $`O^+`$-planes. One orientifold has 2 $`O^{}`$ planes occupied, and the other has the remaining 4 $`O^{}`$-planes occupied.
$`SO(4)`$ with spin- nor vector structure has only 2 branes on the dual orientifold, so only one out of the two possibilities mentioned can be realised. This corresponds to 4 components on the moduli space, all consisting of a single point. This can be seen as follows. $`Spin(4)=SU(2)\times SU(2)`$, and therefore we may write $`Spin(4)`$ holonomies as $`SU(2)`$-pairs. The holonomies obeying the boundary conditions are (up to conjugation)
$$\mathrm{\Omega }_1=(\mathrm{i}\sigma _3,\mathrm{i}\sigma _3)\mathrm{\Omega }_2=(\pm \mathrm{i}\sigma _1,\pm \mathrm{i}\sigma _3)\mathrm{\Omega }_3=(\mathrm{i}\sigma _1,\mathrm{i}\sigma _1)$$
(12)
There are four inequivalent possible choices for the signs in $`\mathrm{\Omega }_2`$ (one cannot change a sign in $`\mathrm{\Omega }_2`$ by conjugation without changing some sign in the other holonomies).
For $`SO(4k)`$ $`(k>1)`$ and larger groups each of the two orientifold descriptions represents two components. Each of the two orientifold descriptions again represents 2 components. For $`SO(4k)`$, the two components contribute $`k`$, resp. $`k1`$ to the Witten index. Taking into account the correct multiplicities, gives the correct answer $`4k2`$ for the Witten index .
## 5 Conclusions
We have shown how to construct orientifold configurations describing compactifications of gauge theories with classical groups on a 2– or 3–torus. Conversely these results show how various orientifolds should be interpreted as gauge theories. The methods can be extended to higher dimensional tori.
Notice that for two dimensional orientifolds $`T^2/_2`$ any configuration of $`O^+`$ and $`O^{}`$-planes is possible. For 3–dimensional orientifolds $`T^3/_2`$ we only find even numbers of $`O^+`$ (and $`O^{}`$) planes. There is a simple argument why odd numbers of $`O^+`$ planes are not allowed .
Some of these configurations are of immediate interest for consistent string-theories. We already mentioned that the $`U(n)`$-theory on a 2–torus, with holonomy with outer automorphism has precisely the same background as a certain IIB orientifold, though consistency requires the absence of D-branes, and hence the absence of gauge symmetry (so $`n=0`$ in a sense). For the type I string theory on a 3–torus and its duals, which are argued to have $`Spin(32)/_2`$ as its gauge group, it seems that there are four configurations of interest, two describing periodic boundary conditions, and two describing absence of vector structure. In a subsequent publication it will be argued however that two of the four configurations actually suffer a relatively subtle inconsistency, invalidating the naive application of group theory methods in string theory.
Acknowledgements: We would like to thank Pierre van Baal and Jan de Boer for useful discussions and reading a draft version of this article.
|
warning/0004/physics0004057.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
A fundamental problem in formalizing our intuitive ideas about information is to provide a quantitative notion of “meaningful” or “relevant” information. These issues were intentionally left out of information theory in its original formulation by Shannon, who focused attention on the problem of transmitting information rather than judging its value to the recipient. Correspondingly, information theory has often been viewed as being strictly a theory of communication, and this view has become so accepted that many people consider statistical and information theoretic principles as almost irrelevant for the question of meaning. In contrast, we argue here that information theory, in particular lossy source compression, provides a natural quantitative approach to the question of “relevant information.” Specifically, we formulate a variational principle for the extraction or efficient representation of relevant information. In related work we argue that this single information theoretic principle contains as special cases a wide variety of problems, including prediction, filtering, and learning in its various forms.
The problem of extracting a relevant summary of data, a compressed description that captures only the relevant or meaningful information, is not well posed without a suitable definition of relevance. A typical example is that of speech compression. One can consider lossless compression, but in any compression beyond the entropy of speech some components of the signal cannot be reconstructed. On the other hand, a transcript of the spoken words has much lower entropy (by orders of magnitude) than the acoustic waveform, which means that it is possible to compress (much) further without losing any information about the words and their meaning.
The standard analysis of lossy source compression is “rate distortion theory,” which characterizes the tradeoff between the rate, or signal representation size, and the average distortion of the reconstructed signal. Rate distortion theory determines the level of inevitable expected distortion, $`D`$, given the desired information rate, $`R`$, in terms of the rate distortion function $`R(D)`$. The main problem with rate distortion theory is in the need to specify the distortion function first, which in turn determines the relevant features of the signal. Those features, however, are often not explicitly known and an arbitrary choice of the distortion function is in fact an arbitrary feature selection.
In the speech example, we have at best very partial knowledge of what precise components of the signal are perceived by our (neural) speech recognition system. Those relevant components depend not only on the complex structure of the auditory nervous system, but also on the sounds and utterances to which we are exposed during our early life. It therefore is virtually impossible to come up with the “correct” distortion function for acoustic signals. The same type of difficulty exists in many similar problems, such as natural language processing, bioinformatics (for example, what features of protein sequences determine their structure) or neural coding (what information is encoded by spike trains and how). This is the fundamental problem of feature selection in pattern recognition. Rate distortion theory does not provide a full answer to this problem since the choice of the distortion function, which determines the relevant features, is not part of the theory. It is, however, a step in the right direction.
A possible solution comes from the fact that in many interesting cases we have access to an additional variable that determines what is relevant. In the speech case it might be the transcription of the signal, if we are interested in the speech recognition problem, or it might be the speaker’s identity if speaker identification is our goal. For natural language processing, it might be the part of speech labels for words in grammar checking, but the dictionary senses of ambiguous words in information retrieval. Similarly, for the protein folding problem we have a joint database of sequences and three dimensional structures, and for neural coding a simultaneous recording of sensory stimuli and neural responses defines implicitly the relevant variables in each domain. All of these problems have the same formal underlying structure: extract the information from one variable that is relevant for the prediction of another one. The choice of additional variable determines the relevant components or features of the signal in each case.
In this short paper we formalize this intuitive idea using an information theoretic approach which extends elements of rate distortion theory. We derive self consistent equations and an iterative algorithm for finding representations of the signal that capture its relevant structure, and prove convergence of this algorithm.
## 2 Relevant quantization
Let $`X`$ denote the signal (message) space with a fixed probability measure $`p(x)`$, and let $`\stackrel{~}{X}`$ denote its quantized codebook or compressed representation. For ease of exposition we assume here that both of these sets are finite, that is, a continuous space should first be quantized.
For each value $`xX`$ we seek a possibly stochastic mapping to a representative, or codeword in a codebook, $`\stackrel{~}{x}\stackrel{~}{X}`$, characterized by a conditional p.d.f. $`p(\stackrel{~}{x}|x)`$. The mapping $`p(\stackrel{~}{x}|x)`$ induces a soft partitioning of $`X`$ in which each block is associated with one of the codebook elements $`\stackrel{~}{x}\stackrel{~}{X}`$, with probability given by
$$p(\stackrel{~}{x})=\underset{x}{}p(x)p(\stackrel{~}{x}|x).$$
(1)
The average volume of the elements of $`X`$ that are mapped to the same codeword is $`2^{H(X|\stackrel{~}{X})}`$, where
$$H(X|\stackrel{~}{X})=\underset{xX}{}p(x)\underset{\stackrel{~}{x}\stackrel{~}{X}}{}p(\stackrel{~}{x}|x)\mathrm{log}p(\stackrel{~}{x}|x)$$
(2)
is the conditional entropy of $`X`$ given $`\stackrel{~}{X}`$.
What determines the quality of a quantization? The first factor is of course the rate, or the average number of bits per message needed to specify an element in the codebook without confusion. This number per element of $`X`$ is bounded from below by the mutual information
$$I(X;\stackrel{~}{X})=\underset{xX}{}\underset{\stackrel{~}{x}\stackrel{~}{X}}{}p(x,\stackrel{~}{x})\mathrm{log}\left[\frac{p(\stackrel{~}{x}|x)}{p(\stackrel{~}{x})}\right],$$
(3)
since the average cardinality of the partitioning of $`X`$ is given by the ratio of the volume of $`X`$ to that of the mean partition, $`2^{H(X)}/2^{H(X|\stackrel{~}{X})}=2^{I(X;\stackrel{~}{X})}`$, via the standard asymptotic arguments. Notice that this quantity is different from the entropy of the codebook, $`H(\stackrel{~}{X})`$, and this entropy normally is not what we want to minimize.
However, information rate alone is not enough to characterize good quantization since the rate can always be reduced by throwing away details of the original signal $`x`$. We need therefore some additional constraints.
### 2.1 Relevance through distortion: <br>Rate distortion theory
In rate distortion theory such a constraint is provided through a distortion function, $`d:X\times \stackrel{~}{X}^+`$, which is presumed to be small for good representations. Thus the distortion function specifies implicitly what are the most relevant aspects of values in $`X`$.
The partitioning of $`X`$ induced by the mapping $`p(\stackrel{~}{x}|x)`$ has an expected distortion
$$d(x,\stackrel{~}{x})_{p(x,\stackrel{~}{x})}=\underset{xX}{}\underset{\stackrel{~}{x}\stackrel{~}{X}}{}p(x,\stackrel{~}{x})d(x,\stackrel{~}{x}).$$
(4)
There is a monotonic tradeoff between the rate of the quantization and the expected distortion: the larger the rate, the smaller is the achievable distortion.
The celebrated rate distortion theorem of Shannon and Kolmogorov (see, for example Ref. ) characterizes this tradeoff through the rate distortion function, $`R(D)`$, defined as the minimal achievable rate under a given constraint on the expected distortion:
$$R(D)\underset{\{p(\stackrel{~}{x}|x):d(x,\stackrel{~}{x})D\}}{\mathrm{min}}I(X;\stackrel{~}{X}).$$
(5)
Finding the rate distortion function is a variational problem that can be solved by introducing a Lagrange multiplier, $`\beta `$, for the constrained expected distortion. One then needs to minimize the functional
$$[p(\stackrel{~}{x}|x)]=I(X;\stackrel{~}{X})+\beta d(x,\stackrel{~}{x})_{p(x,\stackrel{~}{x})}$$
(6)
over all normalized distributions $`p(\stackrel{~}{x}|x)`$. This variational formulation has the following well known consequences:
###### Theorem 1
The solution of the variational problem,
$$\frac{\delta }{\delta p(\stackrel{~}{x}|x)}=0,$$
(7)
for normalized distributions $`p(\stackrel{~}{x}|x)`$, is given by the exponential form
$$p(\stackrel{~}{x}|x)=\frac{p(\stackrel{~}{x})}{Z(x,\beta )}\mathrm{exp}\left[\beta d(x,\stackrel{~}{x})\right],$$
(8)
where $`Z(x,\beta )`$ is a normalization (partition) function. Moreover, the Lagrange multiplier $`\beta `$, determined by the value of the expected distortion, $`D`$, is positive and satisfies
$$\frac{\delta R}{\delta D}=\beta .$$
(9)
Proof. Taking the derivative w.r.t. $`p(\stackrel{~}{x}|x)`$, for given $`x`$ and $`\stackrel{~}{x}`$, one obtains
$`{\displaystyle \frac{\delta }{\delta p(\stackrel{~}{x}|x)}}`$ $`=`$ $`p(x)[\mathrm{log}{\displaystyle \frac{p(\stackrel{~}{x}|x)}{p(\stackrel{~}{x})}}+1`$ (10)
$`{\displaystyle \frac{1}{p(\stackrel{~}{x})}}{\displaystyle \underset{x^{}}{}}p(x^{})p(\stackrel{~}{x}|x^{})+\beta d(x,\stackrel{~}{x})+{\displaystyle \frac{\lambda (x)}{p(x)}}],`$
since the marginal distribution satisfies $`p(\stackrel{~}{x})=_x^{}p(x^{})p(\stackrel{~}{x}|x^{})`$. $`\lambda (x)`$ are the normalization Lagrange multipliers for each $`x`$. Setting the derivatives to zero and writing $`\mathrm{log}Z(x,\beta )=\lambda (x)/p(x)`$, we obtain Eq. (8). When varying the normalized $`p(\stackrel{~}{x}|x)`$, the variations $`\delta I(X;\stackrel{~}{X})`$ and $`\delta d(x,\stackrel{~}{x})_{p(x,\stackrel{~}{x})}`$ are linked through
$$\delta =\delta I(X;\stackrel{~}{X})+\beta \delta d(x,\stackrel{~}{x})_{p(x,\stackrel{~}{x})}=0,$$
(11)
from which Eq. (9) follows. The positivity of $`\beta `$ is then a consequence of the concavity of the rate distortion function (see, for example, Chapter 13 of Ref. ).
### 2.2 The Blahut–Arimoto algorithm
An important practical consequence of the above variational formulation is that it provides a converging iterative algorithm for self consistent determination of the distributions $`p(\stackrel{~}{x}|x)`$ and $`p(\stackrel{~}{x})`$.
Equations (8) and (1) must be satisfied simultaneously for consistent probability assignment. A natural approach to solve these equations is to alternately iterate between them until reaching convergence. The following lemma, due to Csiszár and Tusnády , assures global convergence in this case.
###### Lemma 2
Let $`p(x,y)=p(x)p(y|x)`$ be a given joint distribution. Then the distribution $`q(y)`$ that minimizes the relative entropy or Kullback–Leibler divergence, $`D_{KL}[p(x,y)|p(x)q(y)]`$, is the marginal distribution
$$p(y)=\underset{x}{}p(x)p(y|x).$$
Namely,
$$I(X;Y)=D_{KL}[p(x,y)|p(x)p(y)]=\underset{q(y)}{\mathrm{min}}D_{KL}[p(x,y)|p(x)q(y)].$$
Equivalently, the distribution $`q(y)`$ which minimizes the expected relative entropy,
$$\underset{x}{}p(x)D_{KL}[p(y|x)|q(y)],$$
is also the marginal distribution $`p(y)=_xp(x)p(y|x)`$.
The proof follows directly from the non–negativity of the relative entropy.
This lemma guarantees the marginal condition Eq. (1) through the same variational principle that leads to Eq. (8):
###### Theorem 3
Equations (1) and (8) are satisfied simultaneously at the minimum of the functional,
$$=\mathrm{log}Z(x,\beta )_{p(x)}=I(X;\stackrel{~}{X})+\beta d(x,\stackrel{~}{x})_{p(x,\stackrel{~}{x})},$$
(12)
where the minimization is done independently over the convex sets of the normalized distributions, $`\{p(\stackrel{~}{x})\}`$ and $`\{p(\stackrel{~}{x}|x)\}`$,
$$\underset{p(\stackrel{~}{x})}{\mathrm{min}}\underset{p(\stackrel{~}{x}|x)}{\mathrm{min}}[p(\stackrel{~}{x});p(\stackrel{~}{x}|x)].$$
These independent conditions correspond precisely to alternating iterations of Eq. (1) and Eq. (8). Denoting by $`t`$ the iteration step,
$$\{\begin{array}{c}p_{t+1}(\stackrel{~}{x})=_xp(x)p_t(\stackrel{~}{x}|x)\hfill \\ p_t(\stackrel{~}{x}|x)=\frac{p_t(\stackrel{~}{x})}{Z_t(x,\beta )}\mathrm{exp}(\beta d(x,\stackrel{~}{x}))\hfill \end{array}$$
(13)
where the normalization function $`Z_t(x,\beta )`$ is evaluated for every $`t`$ in Eq. (13). Furthermore, these iterations converge to a unique minimum of $``$ in the convex sets of the two distributions.
For the proof, see references . This alternating iteration is the well known Blauht-Arimoto (BA) algorithm for calculation of the rate distortion function.
It is important to notice that the BA algorithm deals only with the optimal partitioning of the set $`X`$ given the set of representatives $`\stackrel{~}{X}`$, and not with an optimal choice of the representation $`\stackrel{~}{X}`$. In practice, for finite data, it is also important to find the optimal representatives which minimize the expected distortion, given the partitioning. This joint optimization is similar to the EM procedure in statistical estimation and does not in general have a unique solution.
## 3 Relevance through another variable: <br>The Information Bottleneck
Since the “right” distortion measure is rarely available, the problem of relevant quantization has to be addressed directly, by preserving the relevant information about another variable. The relevance variable, denoted here by $`Y`$, must not be independent from the original signal $`X`$, namely they have positive mutual information $`I(X;Y)`$. It is assumed here that we have access to the joint distribution $`p(x,y)`$, which is part of the setup of the problem, similarly to $`p(x)`$ in the rate distortion case.<sup>1</sup><sup>1</sup>1The problem of actually obtaining a good enough sample of this distribution is an interesting issue in learning theory, but is beyond the scope of this paper. For a start on this problem see Ref. .
### 3.1 A new variational principle
As before, we would like our relevant quantization $`\stackrel{~}{X}`$ to compress $`X`$ as much as possible. In contrast to the rate distortion problem, however, we now want this quantization to capture as much of the information about $`Y`$ as possible. The amount of information about $`Y`$ in $`\stackrel{~}{X}`$ is given by
$$I(\stackrel{~}{X};Y)=\underset{y}{}\underset{\stackrel{~}{x}}{}p(y,\stackrel{~}{x})\mathrm{log}\frac{p(y,\stackrel{~}{x})}{p(y)p(\stackrel{~}{x})}I(X;Y).$$
(14)
Obviously lossy compression cannot convey more information than the original data. As with rate and distortion, there is a tradeoff between compressing the representation and preserving meaningful information, and there is no single right solution for the tradeoff. The assignment we are looking for is the one that keeps a fixed amount of meaningful information about the relevant signal $`Y`$ while minimizing the number of bits from the original signal $`X`$ (maximizing the compression). <sup>2</sup><sup>2</sup>2It is completely equivalent to maximize the meaningful information for a fixed compression of the original variable. In effect we pass the information that $`X`$ provides about $`Y`$ through a “bottleneck” formed by the compact summaries in $`\stackrel{~}{X}`$.
We can find the optimal assignment by minimizing the functional
$$[p(\stackrel{~}{x}|x)]=I(\stackrel{~}{X};X)\beta I(\stackrel{~}{X};Y),$$
(15)
where $`\beta `$ is the Lagrange multiplier attached to the constrained meaningful information, while maintaining the normalization of the mapping $`p(\stackrel{~}{x}|x)`$ for every $`x`$. At $`\beta =0`$ our quantization is the most sketchy possible—everything is assigned to a single point—while as $`\beta \mathrm{}`$ we are pushed toward arbitrarily detailed quantization. By varying the (only) parameter $`\beta `$ one can explore the tradeoff between the preserved meaningful information and compression at various resolutions. As we show elsewhere , for interesting special cases (where there exist sufficient statistics) it is possible to preserve almost all the meaningful information at finite $`\beta `$ with a significant compression of the original data.
### 3.2 Self-consistent equations
Unlike the case of rate distortion theory, here the constraint on the meaningful information is nonlinear in the desired mapping $`p(\stackrel{~}{x}|x)`$ and this is a much harder variational problem. Perhaps surprisingly, this general problem of extracting the meaningful information—minimizing the functional $`[p(\stackrel{~}{x}|x)]`$ in Eq. (15)—can be given an exact formal solution.
###### Theorem 4
The optimal assignment, that minimizes Eq. (15), satisfies the equation
$$p(\stackrel{~}{x}|x)=\frac{p(\stackrel{~}{x})}{Z(x,\beta )}\mathrm{exp}\left[\beta \underset{y}{}p(y|x)\mathrm{log}\frac{p(y|x)}{p(y|\stackrel{~}{x})}\right],$$
(16)
where the distribution $`p(y|\stackrel{~}{x})`$ in the exponent is given via Bayes’ rule and the Markov chain condition $`\stackrel{~}{X}XY`$, as,
$$p(y|\stackrel{~}{x})=\frac{1}{p(\stackrel{~}{x})}\underset{x}{}p(y|x)p(\stackrel{~}{x}|x)p(x).$$
(17)
This solution has a number of interesting features, but we must emphasize that it is a formal solution since $`p(y|\stackrel{~}{x})`$ in the exponential is defined implicitly in terms of the assignment mapping $`p(\stackrel{~}{x}|x)`$. Just as before, the marginal distribution $`p(\stackrel{~}{x})`$ must satisfy the marginal condition Eq. (1) for consistency.
Proof. First we note that the conditional distribution of $`y`$ on $`\stackrel{~}{x}`$
$$p(y|\stackrel{~}{x})=\underset{xX}{}p(y|x)p(x|\stackrel{~}{x}),$$
(18)
follows from the Markov chain condition $`YX\stackrel{~}{X}`$.<sup>3</sup><sup>3</sup>3It is important to notice that this not a modeling assumption and the quantization $`\stackrel{~}{X}`$ is not used as a hidden variable in a model of the data. In the latter, the Markov condition would have been different: $`Y\stackrel{~}{X}X`$. The only variational variables in this scheme are the conditional distributions, $`p(\stackrel{~}{x}|x)`$, since other unknown distributions are determined from it through Bayes’ rule and consistency. Thus we have
$$p(\stackrel{~}{x})=\underset{x}{}p(\stackrel{~}{x}|x)p(x),$$
(19)
and
$$p(\stackrel{~}{x}|y)=\underset{x}{}p(\stackrel{~}{x}|x)p(x|y).$$
(20)
The above equations imply the following derivatives w.r.t. $`p(\stackrel{~}{x}|x)`$,
$$\frac{\delta p(\stackrel{~}{x})}{\delta p(\stackrel{~}{x}|x)}=p(x)$$
(21)
and
$$\frac{\delta p(\stackrel{~}{x}|y)}{\delta p(\stackrel{~}{x}|x)}=p(x|y).$$
(22)
Introducing Lagrange multipliers, $`\beta `$ for the information constraint and $`\lambda (x)`$ for the normalization of the conditional distributions $`p(\stackrel{~}{x}|x)`$ at each $`x`$, the Lagrangian, Eq. (15), becomes
$``$ $`=`$ $`I(X,\stackrel{~}{X})\beta I(\stackrel{~}{X},Y){\displaystyle \underset{x,\stackrel{~}{x}}{}}\lambda (x)p(\stackrel{~}{x}|x)`$ (24)
$`=`$ $`{\displaystyle \underset{x,\stackrel{~}{x}}{}}p(\stackrel{~}{x}|x)p(x)\mathrm{log}\left[{\displaystyle \frac{p(\stackrel{~}{x}|x)}{p(\stackrel{~}{x})}}\right]\beta {\displaystyle \underset{\stackrel{~}{x},y}{}}p(\stackrel{~}{x},y)\mathrm{log}\left[{\displaystyle \frac{p(\stackrel{~}{x}|y)}{p(\stackrel{~}{x})}}\right]`$
$`{\displaystyle \underset{x,\stackrel{~}{x}}{}}\lambda (x)p(\stackrel{~}{x}|x).`$
Taking derivatives with respect to $`p(\stackrel{~}{x}|x)`$ for given $`x`$ and $`\stackrel{~}{x}`$, one obtains
$`{\displaystyle \frac{\delta }{\delta p(\stackrel{~}{x}|x)}}`$ $`=`$ $`p(x)\left[1+\mathrm{log}p(\stackrel{~}{x}|x)\right]{\displaystyle \frac{\delta p(\stackrel{~}{x})}{\delta p(\stackrel{~}{x}|x)}}\left[1+\mathrm{log}p(\stackrel{~}{x})\right]`$ (25)
$`\beta {\displaystyle \underset{y}{}}{\displaystyle \frac{\delta p(\stackrel{~}{x}|y)}{\delta p(\stackrel{~}{x}|x)}}p(y)\left[1+\mathrm{log}p(\stackrel{~}{x}|y)\right]`$
$`\beta {\displaystyle \frac{\delta p(\stackrel{~}{x})}{\delta p(\stackrel{~}{x}|x)}}\left[1+\mathrm{log}p(\stackrel{~}{x})\right]\lambda (x).`$
Substituting the derivatives from Eq’s. (21) and (22) and rearranging,
$$\frac{\delta }{\delta p(\stackrel{~}{x}|x)}=p(x)\left\{\mathrm{log}\left[\frac{p(\stackrel{~}{x}|x)}{p(\stackrel{~}{x})}\right]\beta \underset{y}{}p(y|x)\mathrm{log}\left[\frac{p(y|\stackrel{~}{x})}{p(y)}\right]\frac{\lambda (x)}{p(x)}\right\}.$$
Notice that $`_yp(y|x)\mathrm{log}\frac{p(y|x)}{p(y)}=I(x,Y)`$ is a function of $`x`$ only (independent of $`\stackrel{~}{x}`$) and thus can be absorbed into the multiplier $`\lambda (x)`$. Introducing
$$\stackrel{~}{\lambda }(x)=\frac{\lambda (x)}{p(x)}\beta \underset{y}{}p(y|x)\mathrm{log}\left[\frac{p(y|x)}{p(y)}\right],$$
we finally obtain the variational condition:
$`{\displaystyle \frac{\delta }{\delta p(\stackrel{~}{x}|x)}}=p(x)\left[\mathrm{log}{\displaystyle \frac{p(\stackrel{~}{x}|x)}{p(\stackrel{~}{x})}}+\beta {\displaystyle \underset{y}{}}p(y|x)\mathrm{log}{\displaystyle \frac{p(y|x)}{p(y|\stackrel{~}{x})}}\stackrel{~}{\lambda }(x)\right]=0,`$ (26)
which is equivalent to equation (16) for $`p(\stackrel{~}{x}|x)`$,
$`p(\stackrel{~}{x}|x)={\displaystyle \frac{p(\stackrel{~}{x})}{Z(x,\beta )}}\mathrm{exp}(\beta D_{KL}[p(y|x)|p(y|\stackrel{~}{x})]),`$ (27)
with
$$Z(x,\beta )=\mathrm{exp}[\beta \stackrel{~}{\lambda }(x)]=\underset{\stackrel{~}{x}}{}p(\stackrel{~}{x})\mathrm{exp}(\beta D_{KL}[p(y|x)|p(y|\stackrel{~}{x})]),$$
the normalization (partition) function. Comments:
1. The Kullback–Leibler divergence, $`D_{KL}[p(y|x)|p(y|\stackrel{~}{x})]`$, emerged as the relevant “effective distortion measure” from our variational principle but is not assumed otherwise anywhere! It is therefore natural to consider it as the “correct” distortion $`d(x,\stackrel{~}{x})=D_{KL}[p(y|x)|p(y|\stackrel{~}{x})]`$ for quantization in the information bottleneck setting.
2. Equation (27), together with equations (18) and (19), determine self consistently the desired conditional distributions $`p(\stackrel{~}{x}|x)`$ and $`p(\stackrel{~}{x})`$. The crucial quantization is here performed through the conditional distributions $`p(y|\stackrel{~}{x})`$, and the self consistent equations include also the optimization over the representatives, in contrast to rate distortion theory, where the selection of representatives is a separate problem.
### 3.3 The information bottleneck iterative algorithm
As for the BA algorithm, the self consistent equations (16) and (17) suggest a natural method for finding the unknown distributions, at every value of $`\beta `$. Indeed, these equations can be turned into converging, alternating iterations among the three convex distribution sets, $`\{p(\stackrel{~}{x}|x)\}`$, $`\{p(\stackrel{~}{x})\}`$, and $`\{p(y|\stackrel{~}{x})\}`$, as stated in the following theorem.
###### Theorem 5
The self consistent equations (18), (19), and (27), are satisfied simultaneously at the minima of the functional,
$`[p(\stackrel{~}{x}|x);p(\stackrel{~}{x});p(y|\stackrel{~}{x})]`$ $`=`$ $`\mathrm{log}Z(x,\beta )_{p(x)}`$ (28)
$`=`$ $`I(X;\stackrel{~}{X})+\beta D_{KL}[p(y|x)|p(y|\stackrel{~}{x})]_{p(x,\stackrel{~}{x})},`$ (29)
where the minimization is done independently over the convex sets of the normalized distributions, $`\{p(\stackrel{~}{x})\}`$ and $`\{p(\stackrel{~}{x}|x)\}`$ and $`\{p(y|\stackrel{~}{x})\}`$. Namely,
$$\underset{p(y|\stackrel{~}{x})}{\mathrm{min}}\underset{p(\stackrel{~}{x})}{\mathrm{min}}\underset{p(\stackrel{~}{x}|x)}{\mathrm{min}}[p(\stackrel{~}{x}|x);p(\stackrel{~}{x});p(y|\stackrel{~}{x})].$$
This minimization is performed by the converging alternating iterations. Denoting by $`t`$ the iteration step,
$$\{\begin{array}{c}p_t(\stackrel{~}{x}|x)=\frac{p_t(\stackrel{~}{x})}{Z_t(x,\beta )}\mathrm{exp}(\beta d(x,\stackrel{~}{x}))\hfill \\ p_{t+1}(\stackrel{~}{x})=_xp(x)p_t(\stackrel{~}{x}|x)\hfill \\ p_{t+1}(y|\stackrel{~}{x})=_yp(y|x)p_t(x|\stackrel{~}{x})\hfill \end{array}$$
(30)
and the normalization (partition function) $`Z_t(\beta ,\stackrel{~}{x})`$ is evaluated for every $`t`$ in Eq. (30).
Proof. For lack of space we can only outline the proof. First we show that the equations indeed are satisfied at the minima of the functional $``$ (known for physicists as the “free energy”). This follows from lemma (2) when applied to $`I(X;\stackrel{~}{X})`$ with the convex sets of $`p(\stackrel{~}{x})`$ and $`p(\stackrel{~}{x}|x)`$, as for the BA algorithm. Then the second part of the lemma is applied to $`D_{KL}[p(y|x)|p(y|\stackrel{~}{x})]_{p(x,\stackrel{~}{x})}`$ which is an expected relative entropy. Equation (27) minimizes the expected relative entropy w.r.t. to variations in the convex set of the normalized $`\{p(y|\stackrel{~}{x})\}`$. Denoting by $`d(x,\stackrel{~}{x})=D_{KL}[p(y|x)|p(y|\stackrel{~}{x})]`$ and by $`\lambda (\stackrel{~}{x})`$ the normalization Lagrange multipliers, we obtain
$`\delta d(x,\stackrel{~}{x})`$ $`=`$ $`\delta \left({\displaystyle \underset{y}{}}p(y|x)\mathrm{log}p(y|\stackrel{~}{x})+\lambda (\stackrel{~}{x})({\displaystyle \underset{y}{}}p(y|\stackrel{~}{x})1)\right)`$ (31)
$`=`$ $`{\displaystyle \underset{y}{}}\left({\displaystyle \frac{p(y|x)}{p(y|\stackrel{~}{x})}}+\lambda (\stackrel{~}{x})\right)\delta p(y|\stackrel{~}{x}).`$ (32)
The expected relative entropy becomes,
$$\underset{x}{}\underset{y}{}\left(\frac{p(y|x)p(x|\stackrel{~}{x})}{p(y|\stackrel{~}{x})}+\lambda (\stackrel{~}{x})\right)\delta p(y|\stackrel{~}{x})=0,$$
(33)
which gives Eq. (27), since $`\delta p(y|\stackrel{~}{x})`$ are independent for each $`\stackrel{~}{x}`$. Equation (27) also have the interpretation of a weighted average of the data conditional distributions that contribute to the representative $`\stackrel{~}{x}`$.
To prove the convergence of the iterations it is enough to verify that each of the iteration steps minimizes the same functional, independently, and that this functional is bounded from below as a sum of two non–negative terms. The only point to notice is that when $`p(y|\stackrel{~}{x})`$ is fixed we are back to the rate distortion case with fixed distortion matrix $`d(x,\stackrel{~}{x})`$. The argument in for the BA algorithm applies here as well. On the other hand we have just shown that the third equation minimizes the expected relative entropy without affecting the mutual information $`I(X;\stackrel{~}{X})`$. This proves the convergence of the alternating iterations. However, the situation here is similar to the EM algorithm and the functional $`[p(\stackrel{~}{x}|x);p(\stackrel{~}{x});p(y|\stackrel{~}{x})]`$ is convex in each of the distribution independently but not in the product space of these distributions. Thus our convergence proof does not imply uniqueness of the solution.
### 3.4 The structure of the solutions
The formal solution of the self consistent equations, described above, still requires a specification of the structure and cardinality of $`\stackrel{~}{X}`$, as in rate distortion theory. For every value of the Lagrange multiplier $`\beta `$ there are corresponding values of the mutual information $`I_XI(X,\stackrel{~}{X})`$, and $`I_YI(\stackrel{~}{X},Y)`$ for every choice of the cardinality of $`\stackrel{~}{X}`$. The variational principle implies that
$$\frac{\delta I(\stackrel{~}{X},Y)}{\delta I(X,\stackrel{~}{X})}=\beta ^1>0,$$
(34)
which suggests a deterministic annealing approach. By increasing the value of $`\beta `$ one can move along convex curves in the “information plane” $`(I_X,I_Y)`$. These curves, analogous to the rate distortion curves, exists for every choice of the cardinality of $`\stackrel{~}{X}`$. The solutions of the self consistent equations thus correspond to a family of such annealing curves, all starting from the (trivial) point $`(0,0)`$ in the information plane with infinite slope and parameterized by $`\beta `$. Interestingly, every two curves in this family separate (bifurcate) at some finite (critical) $`\beta `$ through a second order phase transition. These transitions form a hierarchy of relevant quantizations for different cardinalities of $`\stackrel{~}{X}`$, as described in .
### Further work
The most fascinating aspect of the information bottleneck principle is that it provides a unified framework for different information processing problems, including prediction, filtering and learning . There are already several successful applications of this method to various “real” problems, such as semantic clustering of English words , document classification , neural coding, and spectral analysis.
#### Acknowledgements
Helpful discussions and insights on rate distortion theory with Joachim Buhmann and Shai Fine are greatly appreciated. Our collaboration was facilitated in part by a grant from the US–Israel Binational Science Foundation (BSF).
|
warning/0004/math0004022.html
|
ar5iv
|
text
|
# Local formula for the Index of a Fourier Integral Operator
## 1. Introduction
Let $`X`$ and $`Y`$ be two smooth closed connected Riemannian manifolds of the same dimension such that there exists a contact diffeomorphism $`\varphi :S^{}XS^{}Y`$ between the two unit cotangent bundles which induces a homogeneous symplectomorphism, still denoted by $`\varphi `$, from $`T^{}XX`$ onto $`T^{}YY`$.
We first recall the definition of the index of $`\varphi `$ when dim$`X3`$ following . We will denote by $`\mathrm{\Omega }_{\frac{1}{2}}`$ the half-density bundle over $`X`$ or $`Y`$. Let $`C_\varphi `$ be the graph of $`\varphi ^1`$ in $`(T^{}YY)\times (T^{}XX)`$ and $`L_{C_\varphi }`$ be the associated Maslov bundle. Let $`A:L^2(X,\mathrm{\Omega }_{\frac{1}{2}})L^2(Y,\mathrm{\Omega }_{\frac{1}{2}})`$ be an elliptic Fourier Integral Operator of order zero whose canonical relation is $`C_\varphi `$ and whose principal symbol is an invertible section of the bundle $`\mathrm{\Omega }_{\frac{1}{2}}L_{C_\varphi }C_\varphi `$ (see , ) for details). Suppose that $`B:L^2(Y,\mathrm{\Omega }_{\frac{1}{2}})L^2(X,\mathrm{\Omega }_{\frac{1}{2}})`$ is an elliptic Fourier Integral Operator of order zero whose canonical relation is $`C_{\varphi ^1}`$. Then $`BA:L^2(X,\mathrm{\Omega }_{\frac{1}{2}})L^2(X,\mathrm{\Omega }_{\frac{1}{2}})`$ is an elliptic scalar pseudo-differential operator of order zero. Since dim $`X3`$ there exists a smooth non vanishing function $`xXa(x)^{}`$ such that the principal symbol of $`BA`$ is homotopic to $`(x,\xi )T^{}Xa(x)^{}`$. In particular the index of $`BA`$ is zero. Thus $`\mathrm{Ind}B=\mathrm{Ind}A`$ for any Fourier Integral Operators $`A`$ and $`B`$ as above, and, as the corollary of this fact, $`\mathrm{Ind}A`$ does not depend on the choice of $`A`$. Since it only depends on the transformation $`\varphi `$, it is called the index of $`\varphi `$ and denoted by $`\mathrm{Ind}\varphi `$. A. Weinstein has proved (see ) that the integer $`\mathrm{Ind}\varphi `$ naturally appears if one wants to compare the spectrum $`(\lambda _k(X))_k`$ of the Laplace Beltrami operator $`\mathrm{\Delta }_X`$ of $`X`$ with the one of $`\mathrm{\Delta }_Y`$; for instance if $`T^{}XX`$ is simply connected then the sequence $`(\lambda _k(X)\lambda _{k\mathrm{Ind}\varphi }(Y))_k`$ is bounded.
The goal of this paper is to provide a geometric formula for the index an elliptic Fourier Integral Operator $`\mathrm{\Phi }`$ of order zero whose canonical relation is $`C_\varphi `$ ( we do not assume dim$`X3`$).
Let us first fix some notation. Given a smooth manifold $`X`$, we will use $`T^{}X`$ to denote the cotangent bundle of $`X`$ and $`\overline{B}^{}X`$ to denote the projective compactification of $`T^{}X`$. We will use $`M`$ to denote the the smooth manifold obtained by glueing at infinity $`\overline{B}^{}X`$ and $`\overline{B}^{}Y`$ with the help of the map
$$\varphi ^{}:(x,\xi )\varphi (x,\xi ).$$
Let $`S^0(T^{}X)`$ and $`S^0(T^{}Y)`$ denote the algebras of asymptotic symbols of pseudodifferential operators of order at most zero on $`X`$ and $`Y`$. Given an element $`aS^0(T^{}X)`$, we denote by $`a_{\mathrm{}}`$ the symbol $`a`$ scaled by $`\mathrm{}`$ in the cotangent direction and by Op($`a`$) the pseudodifferential operator associated to $`a`$. Given a pseudodifferential operator $`A`$, we denote by $`\sigma (A)`$ its full symbol (for the precise definition see the next section).
The general strategy is as follows. We interpret conjugation by $`\mathrm{\Phi }`$ as an isomorphism of the algebras of pseudodifferential operators on $`X`$ and $`Y`$. Translated into terms of formal deformations of the cotangent bundles, this allowes us to construct a formal deformation $`𝔸^{\mathrm{}}(M)`$ of $`C^{\mathrm{}}(M)`$ which on $`T^{}X`$ and $`T^{}Y`$ represents the calculus of differential operators, while on the common cosphere at infinity represents the calculus of pseudodifferential operators. While the symplectic structures on $`T^{}X`$ and $`T^{}Y`$ do not glue together (so there is in general no almost complex structure on $`B^{}X_\varphi ^{^{}}B^{}Y`$), there is a (noncanonical) symplectic Lie algebroid structure $`(,[,],\omega )`$ over $`M`$ and $`𝔸^{\mathrm{}}(M)`$ is a deformation associated to it in the sense of . The usual traces on the algebras of smoothing operators on $`X`$ and $`Y`$ give rise to a trace $`\tau _{\text{can}}`$ on $`𝔸^{\mathrm{}}(M)`$ such that $`\mathrm{ind}\mathrm{\Phi }=\tau _{\text{can}}(1)`$. An application of the general algebraic index theorem from gives the local formula for the index.
The content of the paper is given below.
1.
In the first section we recall the relation between the calculus of $`smoothing`$ operators on $`X`$ and a formal deformation of $`T^{}X`$ which is basically given by full symbol of a pseudodifferential operator.
2.
$`\overline{B}^{}X`$ carries a structure of symplectic Lie algebroid $`(_X,[,],\omega )`$ described in Section 2. The symbolic calculus of pseudodifferential operators gives rise to a formal deformation of the sphere at infinity of $`\overline{B}^{}X`$ which, together with the formal deformation of $`T^{}X`$ given above, gives rise to a formal deformation $`𝔸^{\mathrm{}}(X)`$ of $`\overline{B}^{}X`$ associated to $`(_X,[,],\omega )`$.
3.
Let us fix an almost unitary elliptic Fourier Integral Operator $`\mathrm{\Phi }`$ whose canonical relation is given by the graph of $`\varphi ^1`$. In Section 4 we show how to glue together the deformations $`𝔸^{\mathrm{}}(X)`$ and $`𝔸^{\mathrm{}}(Y)`$ into a formal deformation $`𝔸^{\mathrm{}}(M)`$ of $`M`$ associated to a symplectic Lie algebroid structure $`(,[,],\omega )`$ on $`M`$. The construction is based on the following strengthening of the Egorov theorem (see Theorem 1).
1. The map which to any $`aS^0(T^{}X)`$ associates the asymptotic expansion at $`\mathrm{}=0`$ of $`(\sigma (\mathrm{\Phi }\mathrm{Op}(a_{\mathrm{}})\mathrm{\Phi }^{}))_\mathrm{}^1`$ induces an algebra isomorphism
$$\stackrel{~}{\mathrm{\Phi }}:S^0(T^{}X)S^0(T^{}Y)$$
2. For each $`k^{}`$, there exists an $`_X`$differential operator $`D_k`$ on $`\overline{B}^{}X`$ such that, for any $`aS^0(T^{}X)`$, the following identity holds:
$$\stackrel{~}{\mathrm{\Phi }}(a)=(a+\underset{k1}{}\mathrm{}^kD_k(a))\varphi ^1.$$
Egorov theorem corresponds to the leading term in the above expansion.
The real symplectic vector bundle $``$ is isomorphic to $`TM`$ (as a vector bundle over $`M`$) and hence $`TM`$ is the realification of a complex vector bundle on $`M`$ which will be denoted by $`_{}`$.
4.
In Section 5 we identify the space traces on $`𝔸^{\mathrm{}}(M)`$ and relate it to the traces on the algebras of smoothing and of pseudodifferential operators.
5.
In Section 6 we identify the index of the Fourier Integral Operator with the trace of 1 in the formal deformation.
The local index formula for $`\mathrm{Ind}\mathrm{\Phi }`$ follows from the algebraic index theorem of , the class $`\theta _0`$ being the coefficient of $`\mathrm{}^0`$ in the characteristic class $`\theta `$ ( , ) of the deformation.
The main result can be formulated as follows.
1. Let $`\mathrm{\Phi }`$ be a Fourier Integral Operator and $`𝔸^{\mathrm{}}(M)`$ the formal deformation of $`M`$ associated to it as in Definition 2. Then
$$\mathrm{ind}\mathrm{\Phi }=_M\mathrm{e}^{\theta _0}\widehat{A}(M),$$
where $`\theta _0`$ denotes the characteristic class of the deformation of the Lie Algebroid ($`,[,],\omega `$) given by $`𝔸^{\mathrm{}}(M)`$.
2. Let $`_X`$ be a connection $`_X`$ on the tangent bundle $`T(\overline{B}^{}X)`$ and $`\widehat{A}(T^{}X)`$ an associated representative form of the $`\widehat{A}`$class of $`_X`$. The symplectomorphism $`\varphi `$ induces a connection $`\varphi _{}(_X)`$ on the tangent space of $`\overline{B}^{}YB^{}(Y)`$. Let $`_Y`$ denote its extension to a connection of $`T(Y)`$ and $`\widehat{A}(T^{}Y)`$ an associated representative differential form of the $`\widehat{A}`$-class of $`_Y`$. Then
$$\mathrm{ind}\mathrm{\Phi }=_{B^{}(X)}\mathrm{e}^{\theta _0}\widehat{A}(T^{}X)_{B^{}(Y)}\mathrm{e}^{\theta _0}\widehat{A}(T^{}Y)$$
6.
The computation of the characteristic class of the deformation is given in the last section, where we simultaneously construct a deformation of $`M`$ and the Fourier Integral Operator whose index is given by the trace of the 1 in the deformed algebra. As the starting point we give a somewhat nonstandard definition of the characteristic class of a formal deformation which is more amenable to computations in the case of deformations associated to (twisted) differential or pseudodifferential operators.
As a corollary we get the following result.
1. There exists an almost unitary Fourier integral operator $`\mathrm{\Phi }_0`$ whose canonical relation is $`C_\varphi `$ and such that:
$$\mathrm{ind}\mathrm{\Phi }_0=_M\widehat{A}(M)e^{\frac{1}{2}c_1(_{})}$$
2. If the dimension of $`M`$ is at least three, then
$$\mathrm{ind}(\varphi )=_M\widehat{A}(M)e^{\frac{1}{2}c_1(_{})}$$
(compare with , ).
In the case when $`X=Y`$ a straightforward Meyer-Vietoris type argument with the mapping torus of $`\varphi `$ shows that our results recover those of Epstein and Melrose.
7.
###### Remark 1.
The methods of this paper extend in a fairly straightforward manner to the case of a Fourier Integral Operator $`\mathrm{\Phi }`$ between $`L^2`$ sections of vector bundles $`E`$ and $`F`$ of the same dimension on $`X`$ and $`Y`$. In the case when both $`X`$ and $`Y`$ posses a metalinear structure, the corresponding index formula is given by the expression
$$\mathrm{ind}\mathrm{\Phi }=_{B^{}(X)}\text{ch}()\widehat{A}(T^{}X)_{B^{}(Y)}\text{ch}()\widehat{A}(T^{}Y)$$
Here $``$ is the vector bundle over $`M`$ obtained by glueing together pull backs (by the canonical projections $`\pi _X^{}`$ and $`\pi _Y^{}`$) to the cotangent bundles of $`X`$ (resp. $`Y`$) of the bundles $`\mathrm{\Lambda }^{\frac{n}{2}}(X)E`$ and $`\mathrm{\Lambda }^{\frac{n}{2}}(Y)F`$ with the help of the symbol of $`\mathrm{\Phi }`$.
Note that existence of an isomorphism of $`\pi _X^{}(\mathrm{\Lambda }^{\frac{n}{2}}(X)E)`$ and $`\pi _Y^{}(\mathrm{\Lambda }^{\frac{n}{2}}(Y)F)`$ over $`\varphi ^{^{}}`$ is equivalent to existence of an elliptic Fourier Integral Operator from $`L^2(X,E)`$ to $`L^2(Y,F)`$.
###### Remark 2.
It is easy to see that our local formula implies the following fact:
If $`\varphi `$ extends as a symplectomorphism $`:T^{}XT^{}Y`$ up to the zero section, then $`\mathrm{Ind}\mathrm{\Phi }=\mathrm{\hspace{0.17em}0}`$.
## 2. Symbolic calculus for $`\mathrm{\Psi }`$DO’s and formal deformations
### 2.1. Deformation of $`T^{}X`$
We will recall the pertinent facts from .
Let $`\chi `$ be a smooth, non-negative function on $`X\times X`$ satisfying the following conditions:
(1) $`\chi (x,y)=\chi (y,x)`$;
(2) $`\chi 1`$ on an open set containing the diagonal in $`X\times X`$.
(3) For each $`xX`$, the set $`D_x=\{yX/(x,y)\mathrm{supp}\chi \}`$ is geodesically convex.
We denote by $`\mathrm{Exp}_x^1`$ the unique smooth inverse to the exponential map:
$$\mathrm{Exp}_x:T_xXX$$
defined on $`D_x`$ and such that $`\mathrm{Exp}_x^1(x)=0`$.
Given $`xX,yD_x`$, let $`z`$ denote the midpoint of the unique geodesic joining $`x`$ and $`y`$ within $`D_x`$, and let $`vT_zX`$ be given by
(1)
$$v/2=\mathrm{Exp}_x^1(y)=\mathrm{Exp}_z^1(x)$$
Now, denote by $`S^m(T^{}X)`$ the space of classical symbols of order $`m`$ on $`X`$, i. e.. smooth functions $`\theta `$ on $`T^{}X`$ satisfying estimates of the form:
$$\underset{(x,\xi )}{sup}|_x^\alpha _\xi ^\beta \theta (x,\xi )|C_{\alpha ,\beta }(1+|\xi |^2)^{\frac{m|\beta |}{2}}$$
$`S^m(T^{}X)`$ is given the topology of (Frechet) topological vector space by the ”best” $`C_{\alpha ,\beta }`$.
We will denote by $`S^+\mathrm{}(T^{}X)=_mS^m(T^{}X)`$ the set of all classical symbols on $`T^{}X`$.
With the above notation (1), the map:
$$\mathrm{Op}:S^m(T^{}X)\mathrm{End}(C^{\mathrm{}}(X))$$
given by
$$\mathrm{Op}(\theta )(u)(x)=_{T_z^{}X}𝑑\xi 𝑑y\chi (x,y)\mathrm{e}^{i\xi .v}\theta (z,\xi )u(y)$$
defines a pseudo-differential operator. Conversely, if $`P`$ is a pseudo-differential operator on $`X`$ we define its complete symbol to be:
(2)
$$\sigma (P)(z,\xi )=P_y(\chi (x,y)\mathrm{e}^{i\xi .v})|_{x=y=z}$$
where $`z`$ is the midpoint of the geodesic joining $`x`$ and $`y`$ and $`v`$ satisfies (1). We observe that $`P\mathrm{Op}(\sigma (P))`$ is a smoothing operator whose Schwartz kernel vanishes to infinite order on the diagonal. Now, for a given $`\theta C^{\mathrm{}}(T^{}X)`$, we set: $`\theta _{\mathrm{}}(x,\xi )=\theta (x,\mathrm{}\xi )`$.
Following (), we endow the algebra
$$𝔸^{\mathrm{}}(T^{}X)=C^{\mathrm{}}(T^{}X)_{}[[\mathrm{}]]$$
with a star product $`_X`$ by defining, for any symbols $`\theta ^1,\theta ^2,S^m(T^{}X)`$, $`\theta ^1_X\theta ^2`$ to be the asymptotic expansion at $`\mathrm{}=0`$ of:
(3)
$$\sigma (\mathrm{Op}(\theta _{\mathrm{}}^1)\mathrm{Op}(\theta _{\mathrm{}}^2))_\mathrm{}^1$$
One sees immediately that $`_X`$ extends to $`𝔸^{\mathrm{}}(T^{}X)`$
Recall that there exists, unique up to normalization, a canonical trace on $`(𝔸^{\mathrm{}}(T^{}X),_X)`$, $`\mathrm{Tr}_{\mathrm{can}}^X`$, given by:
$$aS^{\mathrm{}}(T^{}X),\mathrm{Tr}_{\mathrm{can}}^X(a)=\mathrm{Tr}(\mathrm{Op}(a_{\mathrm{}}))=\frac{1}{n!\mathrm{}^n}_{T^{}X}a(\omega ^X)^n[\mathrm{}^1,\mathrm{}]]$$
( Proposition 2.5 (3) of ).
### 2.2. Lie algebroid structure and deformation quantization the projective completion $`\overline{B}^{}X`$
For any $`xX`$, we set $`B_x^{}X=\frac{_+T_x^{}X}{_+^{}}\{(0,0)\}`$, and embed $`T_x^{}X`$ in $`\overline{B}_{}^{}{}_{x}{}^{}X`$ by sending $`\xi `$ to the class of $`1\xi `$. We view $`B_x^{}X`$ as a compactification of $`T_x^{}X`$. Then we consider the fiber bundle $`\overline{B}^{}X`$ over $`X`$ defined by $`\overline{B}^{}X=_{xX}B_x^{}X`$. Therefore $`\overline{B}^{}X`$ is a compactification of $`T^{}X`$ and a smooth compact manifold with boundary: $`\overline{B}^{}X=\overline{B}^{}XT^{}X`$. Similarly one defines the bundle $`\overline{B}^{}Y`$ over $`Y`$. We observe that the map from $`S^{}X`$ into $`\overline{B}^{}X`$ given by $`\xi 0\xi `$ defines an isomorphism between $`S^{}X`$ and $`\overline{B}^{}XT^{}X`$. For any $`\xi =(x,\xi _x)T_x^{}X`$ we will define $`\xi `$ to be $`(x,\xi _x)T_x^{}X`$. Clearly, $`\varphi `$ induces a natural smooth isomorphism of manifolds with boundary:
$$\varphi ^{}:\overline{B}^{}XX\overline{B}^{}YY$$
defined by
$$\varphi ^{}(\lambda \xi )=(\lambda \varphi (\xi ))\mathrm{if}\xi T^{}XX,\varphi ^{}(\lambda 0)=(\lambda 0).$$
By glueing $`\overline{B}^{}X`$ and $`\overline{B}^{}Y`$ along the boundary $`\overline{B}^{}XT^{}X`$ with the help of $`\varphi ^{}`$, we define the following smooth compact manifold $`M`$:
(4)
$$M=\overline{B}^{}X_\varphi ^{}\overline{B}^{}Y$$
Let $`\mathrm{\Pi }_X:\overline{B}^{}XX`$ be the projection map. We denote by $`\mathrm{\Xi }^X`$ the set of smooth vectors fields of $`\overline{B}^{}X`$ which are tangent to all the submanifolds $`\mathrm{\Pi }_X^1(x)(\overline{B}^{}XT^{}X)`$, $`xX`$. Let $`(x,\xi )=(x_1,\mathrm{},x_n;\xi _1,\mathrm{},\xi _n)`$ be a local chart of $`T^{}X`$ and $`(\rho ,\theta )`$ be the polar coordinates: $`\rho =\xi ,\theta =\frac{\xi }{\xi }`$, where $`||||`$ denotes the Euclidean norm of $`T^{}X`$. Then a local chart of $`\overline{B}^{}X`$ near $`\overline{B}^{}XT^{}X`$ is given by
(5)
$$(x_1,\mathrm{},x_n;t=\frac{1}{\rho },\theta =(\theta _1,\mathrm{},\theta _{n1}))t0,\theta S^{n1}$$
In this local chart, $`\mathrm{\Xi }^X`$ is generated by the vector fields $`t\frac{}{x_j}`$, $`t\frac{}{t}`$, $`\frac{}{\theta _l}`$, where $`1jn`$, $`1ln1`$. We will use several times the following obvious Lemma
###### Lemma 1.
The vector fields $`\frac{}{\xi _j}(1jn)`$ belong to the $`C^{\mathrm{}}(\overline{B}^{}X)`$module $`t\mathrm{\Xi }^X`$ generated by $`t^2\frac{}{t}`$, $`t\frac{}{\theta _l}(1ln1)`$.
Moreover we observe that the set of classical symbols of order zero on $`T^{}X`$ is nothing else but $`C^{\mathrm{}}(\overline{B}^{}X)`$.
Before we continue, let us recall the definition of a symplectic Lie algebroid ( see for instance , ).
###### Definition 1.
1) A symplectic Lie algebroid on $`M`$ is a quadruple $`(,\rho ,[,],\omega )`$ on $`M`$, where $``$ is a smooth vector bundle on $`M`$, $`[,]`$ is a Lie algebra structure on the sheaf of sections of $``$, $`\rho `$ is a smooth map of vector bundles:
$$\rho :TM$$
such that the induced map:
$$\mathrm{\Gamma }(\rho ):C^+\mathrm{}(M,)C^+\mathrm{}(M,TM)$$
is a Lie algebra homomorphism and, for any sections $`\sigma `$ and $`\tau `$ of $``$ and any smooth function $`f`$ on $`M`$, the following identity holds:
$$[\sigma ,f\tau ]=\rho (\sigma )(f).\tau +f[\sigma ,\tau ]$$
Lastly, $`\omega `$ is a closed $``$two form on $`M`$ such that the associated linear map:
$$C^+\mathrm{}(M,)\times C^+\mathrm{}(M,)(U,V)\omega (U,V)C^+\mathrm{}(M)$$
defines a symplectic structure on $``$.
2) The ring of $``$differential operators is by definition the ring generated by smooth functions on $`M`$ and smooth sections of $``$.
3) We denote by $`{}_{}{}^{}\mathrm{\Omega }_{}^{.}=C^+\mathrm{}(M,\mathrm{\Lambda }^.^{})`$ the set of smooth sections on $`M`$ of the bundle of alternating multilinear forms on $``$.
We leave to the reader the easy proof of the following:
###### Proposition 1.
For any $`p\overline{B}^{}X`$, we set
$$_p^X=\frac{\mathrm{\Xi }^X}{I_p\mathrm{\Xi }^X}$$
where $`I_p`$ is the set of smooth real-valued functions on $`\overline{B}^{}X`$ which vanish at $`p`$. Then:
1) $`(_p^X)_{p\overline{B}^{}X}`$ form a smooth vector bundle, denoted $`^X`$, over $`\overline{B}^{}X`$ such that the set of smooth sections over $`\overline{B}^{}X`$ of $`^X`$ is the same as $`\mathrm{\Xi }^X`$. If $`U,V\mathrm{\Xi }^X`$ then the Lie bracket $`[U,V]`$ also belongs to $`\mathrm{\Xi }^X`$.
2) The fundamental two-form $`\omega ^X(=_{j=1}^nd\xi _jdx_j)`$ of $`T^{}X`$ induces a smooth form, still denoted $`\omega ^X`$, in $`C^{\mathrm{}}(\overline{B}^{}X;{}_{}{}^{}{}_{}{}^{X}\mathrm{\Omega }_{}^{2})`$. Moreover, $`(^X,[,],\omega ^X)`$ defines a symplectic Lie algebroid over $`\overline{B}^{}X`$.
###### Proposition 2.
The star product $`_X`$ on $`T^{}X`$ extends to a star product, still denoted $`_X`$, on $`\overline{B}^{}X`$ such that for any $`f,gC^{\mathrm{}}(\overline{B}^{}X)`$ we have:
$$f_Xg=fg+\underset{n1}{}\mathrm{}^nA^{(n)}(f,g)$$
where the $`A^{(n)}`$ are $`^X`$bidifferential operators.
###### Proof.
Let $`(x,\xi )=(x_1,\mathrm{},x_n;\xi _1,\mathrm{},\xi _n)`$ be a local chart of $`T^{}X`$. Then for any $`f,gC^{\mathrm{}}(\overline{B}^{}X)`$ and $`(x,\xi )`$ in the domain of this local chart we have:
$$f_Xg(x,\xi )=\underset{\alpha ,\beta ^n,|\beta ||\alpha |}{}\frac{\mathrm{}^{|\alpha |}}{\alpha !}c_{\alpha ,\beta }(x)D_\xi ^\alpha f(x,\xi )\frac{^\beta }{^\beta x}g(x,\xi )$$
then, using the local coordinates (5) and Lemma 1, one gets easily all the results of the proposition. ∎
Proposition 1 allows to formulate the following definition.
###### Definition 2.
1) A smooth real-vector bundle $`^Y`$ over $`\overline{B}^{}Y`$ is defined by setting $`_{|T^{}Y}^Y=T(T^{}Y)`$ and $`_{|\overline{B}^{}YY}^Y=\varphi _{}(_{|\overline{B}^{}XX}^X)`$.
2) By glueing $`^X`$ and $`^Y`$ along $`\overline{B}^{}XT^{}X`$ with the help of $`\varphi ^{}`$, one defines a smooth vector bundle $``$ over $`M`$ which is isomorphic to $`TM`$. A smooth exact differential form $`\omega C^{\mathrm{}}(M;{}_{}{}^{}\mathrm{\Omega }_{}^{2})`$ is defined by setting $`\omega _{|\overline{B}^{}X}=\omega ^X`$, $`\omega _{|\overline{B}^{}YY}=(\varphi ^1)^{}(\omega _{|\overline{B}^{}XX}^X)`$ and $`\omega _{|T^{}Y}=\omega ^Y`$ (where $`\omega ^Y`$ is the canonical two form of $`T^{}Y`$).
3) The natural injection:
$$C^+\mathrm{}(M,)C^+\mathrm{}(M,TM)$$
is induced by a bundle map $`\rho :TM`$ as in Proposition 1 and $`(,\rho ,[,],\omega )`$ defines a symplectic Lie algebroid which will be denoted $`(,[,],\omega )`$ in the sequel.
## 3. Regularized Index formula for a Fourier Integral Operator
Let $`C_\varphi `$ be the graph of $`\varphi ^1`$ in $`(T^{}YY)\times (T^{}XX)`$ and $`L_{C_\varphi }`$ be the associated Maslov bundle over $`C_\varphi `$. We fix $`\mathrm{\Phi }:L^2(X,\mathrm{\Omega }_{\frac{1}{2}})L^2(Y,\mathrm{\Omega }_{\frac{1}{2}})`$ an elliptic Fourier integral operator of order zero whose canonical relation is $`C_\varphi `$ and whose principal symbol $`a`$ is a unitary section of the bundle $`\mathrm{\Omega }_{\frac{1}{2}}L_{C_\varphi }C_\varphi `$: this means that $`a`$ is homogeneous of degree zero (i.e. constant on each ray) and that $`a\overline{a}1`$: see . We can, and will, assume in the sequel that $`\mathrm{\Phi }\mathrm{\Phi }^{}\mathrm{Id}`$ and $`\mathrm{\Phi }^{}\mathrm{\Phi }\mathrm{Id}`$ are smoothing. As observed in $`\mathrm{\Phi }`$ is Fredholm, with index defined by $`\mathrm{ind}\mathrm{\Phi }=\mathrm{dim}\mathrm{ker}\mathrm{\Phi }\mathrm{dim}\mathrm{coker}\mathrm{\Phi }`$. In order to give a formula ”via regularization” for $`\mathrm{ind}\mathrm{\Phi }`$ we introduce the following algebra $`𝒜`$ which will have a ”regularized” trace.
$$𝒜=\{(A,B)\mathrm{\Psi }^0(X,\mathrm{\Omega }_{\frac{1}{2}})\times \mathrm{\Psi }^0(Y,\mathrm{\Omega }_{\frac{1}{2}})|A\mathrm{\Phi }^{}B\mathrm{\Phi }\mathrm{is}\mathrm{smoothing}\}$$
We leave to the reader the easy proof of the following:
###### Proposition 3.
1) The map $`\tau :𝒜`$ given by:
$$(A,B)𝒜,\tau (A,B)=\mathrm{Tr}(A\mathrm{\Phi }^{}B\mathrm{\Phi })\mathrm{Tr}(B(\mathrm{Id}\mathrm{\Phi }\mathrm{\Phi }^{}))$$
is a trace
2) $`\mathrm{ind}\mathrm{\Phi }=\tau (\mathrm{Id},\mathrm{Id})`$.
###### Remark 3.
$`\tau (A,B)`$ is a ”regularization” of $`\mathrm{Tr}A\mathrm{Tr}B`$.
## 4. Algebraization of a Fourier Integral Operator
We are going to use the following (deformed quantized algebra), where the manifold $`Z`$ is equal to $`X`$ or $`Y`$:
$$𝔹^{\mathrm{}}(\overline{B}^{}Z)=\frac{C^{\mathrm{}}(\overline{B}^{}Z)}{C_0^{\mathrm{}}(\overline{B}^{}Z)}_{}[[\mathrm{}]]$$
where $`C_0^{\mathrm{}}(\overline{B}^{}Z)`$ denotes the set of smooth functions which vanish to infinite order at $`\overline{B}^{}ZT^{}Z`$. We observe that $`_Z`$ induces a star-product, still denoted $`_Z`$, on $`𝔹^{\mathrm{}}(\overline{B}^{}Z)`$.
###### Theorem 1.
1) The map which to any $`aS^0(T^{}X)`$ associates the asymptotic expansion at $`\mathrm{}=0`$ of $`(\sigma (\mathrm{\Phi }\mathrm{Op}(a_{\mathrm{}})\mathrm{\Phi }^{}))_\mathrm{}^1`$ induces an algebra isomorphism $`\stackrel{~}{\mathrm{\Phi }}`$ from $`(𝔹^{\mathrm{}}(\overline{B}^{}X),_X)`$ onto $`(𝔹^{\mathrm{}}(\overline{B}^{}Y),_Y)`$.
2) For each $`k^{}`$, there exists an $``$differential operator $`D_k`$ on $`\overline{B}^{}X`$ such that for any $`aC^+\mathrm{}(\overline{B}^{}X)`$ which is identically zero in a neighborhood of the zero section, we have the following identity :
$$\stackrel{~}{\mathrm{\Phi }}(a)=(a+\underset{k1}{}\mathrm{}^kD_k(a))\varphi ^1$$
in the vector space $`𝔹^{\mathrm{}}(\overline{B}^{}Y)`$
Before proving this theorem we state the next proposition which is an easy consequence of Proposition 2 and Theorem 1
###### Proposition 4.
The star product $`_Y`$ on $`T^{}Y`$ extends to a star product, still denoted $`_Y`$, on $`\overline{B}^{}Y`$ such that for any $`f,gC^{\mathrm{}}(\overline{B}^{}Y)`$ we have:
$$f_Yg=fg+\underset{n1}{}\mathrm{}^nB^{(n)}(f,g)$$
where the $`B^{(n)}`$ are $`{}_{}{}^{}{}_{}{}^{Y}`$bidifferential operators.
###### Proof.
Let us first assume part 2). Then, using the results of Section 2.1 and the fact that $`\mathrm{\Phi }\mathrm{\Phi }^{}\mathrm{Id}`$ and $`\mathrm{\Phi }^{}\mathrm{\Phi }\mathrm{Id}`$ are smoothing, one proves easily that $`\stackrel{~}{\mathrm{\Phi }}`$ is an isomorphism whose inverse is given by:
$$bS^0(T^{}Y)(\sigma (\mathrm{\Phi }^{}\mathrm{Op}(b_{\mathrm{}})\mathrm{\Phi }))_\mathrm{}^1$$
Now let us prove part 2). Following page 26, we recall that the Schwartz kernel of $`\mathrm{\Phi }`$ is the finite sum of a smooth function and of oscillatory integrals (supported in small coordinates charts) of the following type:
(6)
$$K(y,x)=_^ne^{i(\phi (y,\eta )x\eta )}b(y,\eta )𝑑\eta $$
where $`b(y,\eta )S^0(T^{}Y)`$ vanishes for $`\eta 1`$, $`\phi (y,\eta )`$ is an homogeneous phase function parametrizing locally the graph $`C_\varphi `$ of $`\varphi ^1`$ which satisfies $`\mathrm{det}\frac{^2\phi }{y\eta }0`$ so that locally we have:
$$\{(y,\phi _y^{}(y,\eta );\phi _\eta ^{}(y,\eta ),\eta )\}=C_\varphi $$
and $`\varphi ^1(y,\phi _y^{}(y,\eta ))=(\phi _\eta ^{}(y,\eta ),\eta )`$. Notice moreover that $`(y,\eta )(y,\phi _y^{}(y,\eta ))`$ and $`(y,\eta )(\phi _\eta ^{}(y,\eta ),\eta )`$ are local diffeomorphisms.
With these notations, the Schwartz kernel of $`\mathrm{\Phi }^{}`$ is the finite sum of a smooth function and of oscillatory integrals (supported in small coordinates charts) of the following type:
(7)
$$K^{}(x,y)=_^ne^{i(\phi (y,\eta )x\eta )}b_1(y,\eta )𝑑\eta $$
Let $`aC^+\mathrm{}(\overline{B}^{}X)`$ which is identically zero in a neighborhood of the zero section, in order to analyze $`\mathrm{\Phi }\mathrm{Op}(a_{\mathrm{}})`$ it is enough to study the operator $`K\mathrm{Op}(a_{\mathrm{}})`$ where $`K`$ denotes the operator whose Schwartz kernel is given by (6).
The Schwartz kernel of $`K\mathrm{Op}(a_{\mathrm{}})`$ is given by:
$$T(y,z)=_^n_^n_^ne^{i(\phi (y,\eta )x\eta )}b(y,\eta )a(x,\mathrm{}\xi )e^{i(xz)\xi }𝑑\xi 𝑑x𝑑\eta $$
In this integral we replace $`a(x,\mathrm{}\xi )`$ by its Taylor expansion:
$$\underset{\alpha ^n}{}\frac{1}{\alpha !}_x^\alpha a(z,\mathrm{}\xi )(xz)^\alpha $$
Using the following two identities
$$(xz)^\alpha e^{i(xz)\xi }=D_\xi ^\alpha (e^{i(xz)\xi })$$
$$_^ne^{ix(\xi \eta )}𝑑x=(2\pi )^n\delta _{\xi =\eta }$$
and integrating by parts we see that $`T(y,z)`$ is the sum of a smooth function and of
$`H(y,z)=`$
$$e^{i(\phi (y,\eta )x\eta )}b(y,\eta )\underset{\alpha ^n}{}\frac{1}{\alpha !}(\mathrm{})^{|\alpha |}_x^\alpha D_\xi ^\alpha a(z,\mathrm{}\xi )e^{i(xz)\xi }d\xi dxd\eta $$
$$=(2\pi )^n_^ne^{i(\phi (y,\eta )z\eta )}b(y,\eta )\underset{\alpha ^n}{}\frac{1}{\alpha !}(\mathrm{})^{|\alpha |}_x^\alpha D_\xi ^\alpha a(z,\mathrm{}\eta )d\eta $$
Now for $`\alpha ^n`$ we set
$$c_\alpha (z;\mathrm{}\eta )=_x^\alpha D_\xi ^\alpha a(z,\mathrm{}\eta )$$
and we consider
$$H_\alpha (y,z)=_^ne^{i(\phi (y,\eta )z\eta )}b(y,\eta )c_\alpha (z,\mathrm{}\eta )𝑑\eta $$
If we replace $`c_\alpha (z,\mathrm{}\eta )`$ by its Taylor expansion
$$\underset{\beta ^n}{}\frac{1}{\beta !}_z^\beta c_\alpha (\phi _\eta ^{};\mathrm{}\eta )(z\phi _\eta ^{})^\beta $$
then, using integrations by parts as above, it follows easily that $`H_\alpha (y,z)`$ is the sum of a smooth function and of
$$_^ne^{i(\phi (y,\eta )z\eta )}\underset{\beta ^n}{}\frac{1}{\beta !}D_\eta ^\beta \left(b(y,\eta )_z^\beta c_\alpha (\phi _\eta ^{};\mathrm{}\eta )\right)d\eta $$
We observe that if we apply the Leibniz rule for the term $`D_\eta ^\beta (\mathrm{}.)`$ in the previous integral then the following differential operators will appear
(8)
$$D_\eta ^{\beta \gamma }b(y,\eta )D_\eta ^{\gamma \gamma ^{}}(\phi _\eta ^{})D_\eta ^\gamma ^{}_z^\beta $$
It is clear from Lemma 1 that, expressed in the coordinates $`(\phi _\eta ^{}(y,\eta ),\eta )`$, these differential operators (8) are $``$differential operators. Therefore we have just proved that $`T(y,z)`$ is the sum of a smooth function and of:
$$_^ne^{i(\phi (y,\eta )z\eta )}\underset{k}{}\mathrm{}^kP_k(a)(\phi _\eta ^{}(y,\eta ),\mathrm{}\eta )d\eta $$
where the $`P_k`$ are $``$differential operators.
Now we recall that the Schwartz kernel of $`\mathrm{\Phi }^{}`$ is the finite sum of a smooth function and of terms of the type (7). So in order to analyze $`\mathrm{\Phi }\mathrm{Op}(a_{\mathrm{}})\mathrm{\Phi }^{}`$ it is enough to study the operator $`K\mathrm{Op}(a_{\mathrm{}})K^{}`$ whose Schwartz kernel is the finite sum of a smooth function and of integrals of the type:
$$e^{i(\phi (y,\eta )x\eta )}e^{i(\phi (y^{},\eta ^{})x\eta ^{})}P_k(a)(\phi _\eta ^{}(y,\eta ),\mathrm{}\eta )b_1(y^{},\eta ^{})𝑑x𝑑\eta ^{}𝑑\eta $$
(9)
$$=(2\pi )^n_^ne^{i(\phi (y,\eta )\phi (y^{},\eta )}P_k(a)(\phi _\eta ^{}(y,\eta ),\mathrm{}\eta )b_1(y^{},\eta )𝑑\eta $$
Moreover we can write $`\phi (y,\eta )\phi (y^{},\eta )=(yy^{}).\widehat{\eta }(y,y^{},\eta )`$ where $`\widehat{\eta }(y,y,\eta )=\phi _y^{}(y,\eta )`$ and we can assume (at the expense of shrinking the local coordinates charts) that $`\eta \widehat{\eta }(y,y^{},\eta )`$ is a local diffeomorphism whose inverse is denoted $`\widehat{\eta }\eta (y,y^{},\widehat{\eta })`$. With these notations, we set:
$$A_k(y,y^{},\mathrm{},\widehat{\eta })=P_k(a)(\phi _\eta ^{}(y,\eta ),\mathrm{}\eta )b_1(y^{},\eta )$$
Then a change of variable formula allows us to see that the oscillatory integral (9) is equal to
$$(2\pi )^n_^ne^{i(yy^{})\widehat{\eta }}A_k(y,y^{},\mathrm{},\widehat{\eta })|\frac{D\eta }{D\widehat{\eta }}|𝑑\widehat{\eta }$$
We observe that, expressed in the coordinates $`(\phi _\eta ^{}(y,\eta ),\eta )`$, the vector fields $`_\eta (\phi _\eta ^{})_y`$ are $``$differential operators. Therefore one proves easily the assertion of Part 2) of the Theorem by replacing $`A_k(y,y^{},\mathrm{},\widehat{\eta })`$ by its Taylor expansion
$$\underset{\beta ^n}{}\frac{1}{\beta !}_y^{}^\beta A_k(y,y,\mathrm{},\widehat{\eta })_{|y=y^{}}(y^{}y)^\beta $$
and using, as before, integration by parts.
## 5. The formal deformation and traces on $`\overline{B}^{}X_\varphi \overline{B}^{}Y`$ and regularized traces on $`\mathrm{\Psi }`$DO’s
Recall first that $`C^+\mathrm{}(M)`$ is exactly the set of functions $`(f,g)C^+\mathrm{}(\overline{B}^{}X)\times C^+\mathrm{}(\overline{B}^{}Y)`$ such that $`fg\varphi ^{}`$ vanish of infinite order at the boundary of $`\overline{B}^{}X`$.
We are going to use the $``$-products denoted $`_X`$, $`_Y`$ on $`\overline{B}^{}X`$ and $`\overline{B}^{}Y`$ defined in Propositions 2 and 4. We set $`𝔸^{\mathrm{}}(\overline{B}^{}X)=C^+\mathrm{}(\overline{B}^{}X)_{}[[\mathrm{}]]`$ and $`𝔸^{\mathrm{}}(\overline{B}^{}Y)=C^+\mathrm{}(\overline{B}^{}Y)_{}[[\mathrm{}]]`$ .
Let $`𝔸^{\mathrm{}}(M)`$ be the vector space given by
$$\{(a,b)𝔸^{\mathrm{}}(\overline{B}^{}X)\times 𝔸^{\mathrm{}}(\overline{B}^{}Y)\stackrel{~}{\mathrm{\Phi }}(\overline{a})=\overline{b}\}$$
where $`\overline{a}`$ (resp. $`\overline{b}`$) denotes an element of $`𝔹^{\mathrm{}}(\overline{B}^{}X)`$ (resp. $`𝔹^{\mathrm{}}(\overline{B}^{}Y)`$) induced by $`a`$ (resp. $`b`$). Theorem 1 shows that $`𝔸^{\mathrm{}}(M)`$ is an algebra with respect to the diagonal product: $`(_X,_Y)`$. In particular, pairs of the form $`(\sigma (\mathrm{\Phi }^{}\mathrm{\Phi }),\sigma (\mathrm{\Phi }\mathrm{\Phi }^{}))`$ belong to $`𝔸^{\mathrm{}}(M)`$.
In the statement of the next proposition we will use the notations of Theorem 1.
###### Proposition 5.
1) Let $`\chi C^+\mathrm{}(T^{}X,[0,1])`$ be such that $`\chi (x,\xi )=0`$ for $`\xi 1/2`$ and $`\chi (x,\xi )=1`$ for $`\xi 1`$. For any $`fC^+\mathrm{}(\overline{B}^{}X)`$ we have :
$$\stackrel{~}{\mathrm{\Phi }}(\chi f)(\chi f)\varphi ^1\mathrm{}𝔹^{\mathrm{}}(\overline{B}^{}Y)$$
2) For each $`bC^+\mathrm{}(\overline{B}^{}Y)`$ one defines $`b^{}C^+\mathrm{}(\overline{B}^{}Y)`$ by setting $`b^{}(\eta )=b(\eta )`$ for any $`\eta \overline{B}^{}Y`$. The following formula:
$$(a,b)𝔸^{\mathrm{}}(M),𝒰(a,b)=(a+\underset{k1}{}\mathrm{}^kD_k(a),b^{})C^+\mathrm{}(M)_{}[[\mathrm{}]]$$
defines a $`[[\mathrm{}]]`$linear isomorphism $`𝒰`$ from $`𝔸^{\mathrm{}}(M)`$ to $`C^+\mathrm{}(M)_{}[[\mathrm{}]]`$.
3) The product $`𝒰(_X,_Y)`$ defines an $``$deformation of $`M`$ (or a star product) associated to the symplectic Lie algebroid $`(,[,],\omega )`$ ( see section 3.3).
###### Proof.
Parts 1) and 2) are left to the reader. Part 3) is an easy consequence of part 2) and of Theorem 1 2). ∎
###### Definition 3.
$`𝔸^{\mathrm{}}(M)`$ denotes the formal deformation of $`M`$ associated to the symplectic Lie algebroid $`(,[,],\omega )`$ constructed in the proposition 5.
The linear functional
$$\tau _{\mathrm{can}}:𝔸^{\mathrm{}}(M)[\mathrm{}^1,\mathrm{}]]$$
is given by
(10)
$$(a,b)𝔸^{\mathrm{}}(M),\tau _{\mathrm{can}}(a,b)=\left\{\begin{array}{c}\text{ asymptotic expansion at }\mathrm{}=0\text{ of }\\ \mathrm{}\tau (\mathrm{Op}(a_{\mathrm{}}),\mathrm{Op}(b_{\mathrm{}}))\end{array}\right\}$$
where $`\tau `$ is the trace defined in Proposition 3. It follows immediately from the definition that $`\tau _{\mathrm{can}}`$ is a trace.
Computation of $`\tau `$.
Since the space of traces on $`𝔸^{\mathrm{}}(M)`$ may be very big we introduce the following algebra:
$$𝔻^{\mathrm{}}(M)=𝔸^{\mathrm{}}(M)\left[(\chi \xi ,(\chi \xi +\underset{k1}{}\mathrm{}^kD_k(\chi \xi ))\varphi ^1)\right]$$
Another way of describing $`𝔻^{\mathrm{}}(M)`$ is given by glueing from
$$\stackrel{~}{\varphi }:𝔸^{\mathrm{}}(\overline{B}^{}X)[\chi \xi ]𝔸^{\mathrm{}}(\overline{B}^{}Y)[\chi \xi ]$$
where $`\chi `$ is as in previous Proposition. It is easily seen that $`\tau _{\mathrm{can}}`$ defines, by the same formula as (10), a trace on $`𝔻^{\mathrm{}}(M)`$.
Next Proposition describes the space of traces on $`𝔻^{\mathrm{}}(M)`$.
###### Proposition 6.
The space of traces with values in $`[\mathrm{}^1,\mathrm{}]]`$ on the algebra $`𝔻^{\mathrm{}}(M)`$ is two dimensional over $`[\mathrm{}^1,\mathrm{}]]`$. A basis is given by $`(\tau _{\mathrm{can}},\tau _1)`$ where for any $`(a,b)𝔻^{\mathrm{}}(M)`$ $`\tau _1(a,b)=\mathrm{Res}_W(a)`$. Here $`\mathrm{Res}_W`$ denotes Wodzicki’s noncommutative residue.
###### Proof.
For $`Z=X`$ or $`Y`$ we set:
$$C_0^{\mathrm{}}[\overline{B}^{}Z)_{}[[\mathrm{}]]=𝔸_0^{\mathrm{}}(T^{}Z)$$
where $`C_0^{\mathrm{}}[\overline{B}^{}Z)`$ denotes the set of smooth functions which vanish of infinite order at $`\overline{B}^{}ZT^{}Z`$. Then we have the following exact sequence:
$$0𝔸_0^{\mathrm{}}(T^{}X)𝔸_0^{\mathrm{}}(T^{}Y)𝔻^{\mathrm{}}(M)𝒯(M)0$$
of $`[[\mathrm{}]]`$algebras. Here $`𝒯(M)`$ denotes the induced formal $``$-deformation of the sphere at infinity. A direct construction of this deformation may be described as follows. Let $`𝒫^i`$ denote the space of pseudodifferential operators on, say, $`X`$ of order $`i`$ modulo the smoothing operators. Then the space of doubly infinite sequences
$$\{P_i\}_i,P_i𝒫^i,\text{ there exists }i_0,P_i𝒫^{i_0}\text{ for }i\text{ large}$$
is a flat module over $`[[\mathrm{}]]`$, where the multiplication by $`\mathrm{}`$ acts as the right translation. If we endow it with the product
$$\{P_i\}_i\{Q_i\}_i=\{\underset{i+j=n}{}P_iQ_j\}_n$$
it is easily seen to be isomorphic to $`𝒯(M)`$. Any trace $`\tau `$ on $`𝒯(M)`$ is given by a sequence of $``$-linear, $`[[\mathrm{}]]`$-valued functionals $`\tau _n`$ on $`𝒫^n`$ such that
$$\tau (\{P_i\})=\tau _i(P_i).$$
The $`\mathrm{}`$-linearity of $`\tau `$ implies that $`\tau _{i+1}=\mathrm{}\tau _i`$ and the trace condition on $`\tau `$ implies that each $`\tau _i`$ is a trace on the algebra of pseudodifferential operators modulo the smoothing operators. Recall that, on this latter algebra, the Wodzicki residue $`res`$ is the unique trace up to multiplicative constant. Thus $`\tau `$ is, up to multiplicative constant, uniquely determined by $`\tau _n=res`$, and hence the space of traces on $`𝒯(M)`$ is one-dimensional.
We recall that $`𝔸_0^{\mathrm{}}(T^{}X)`$ is H-unital (in the sense of Wodzicki, see ), so we have the following long exact sequence in cyclic cohomology:
$$0HC^0(𝒯(M))HC^0(𝔻^{\mathrm{}}(M))HC^0(𝔸_0^{\mathrm{}}(T^{}X)𝔸_0^{\mathrm{}}(T^{}Y))$$
$$HC^1(𝒯(M))\mathrm{}$$
From Section 2.1 we recall that the space of $`[[\mathrm{}]]`$linear traces (with values in $`[\mathrm{}^1,\mathrm{}]]`$) on $`𝔸_0^{\mathrm{}}(T^{}X)`$ is one dimensional and generated by $`\mathrm{Tr}_{\mathrm{can}}^X`$. By above, $`HC^0(𝒯(M))`$ is one-dimensional. The connecting map:
$$\delta :HC^0(𝔸_0^{\mathrm{}}(T^{}X)𝔸_0^{\mathrm{}}(T^{}Y))HC^1(𝒯(M))$$
is given by taking a trace on $`𝔸_0^{\mathrm{}}(T^{}X)𝔸_0^{\mathrm{}}(T^{}Y)`$, extending it to a linear functional on $`𝔻^{\mathrm{}}(M)`$ and taking its Hochschild boundary. In particular, it is not zero (this is equivalent to existence of a pseudodifferential operator with nonzero index!). This implies that $`HC^0(𝔻^{\mathrm{}}(M))`$ is either one or two dimensional. Since, with the notations of the Proposition, $`\tau _{\mathrm{can}},\tau _1`$ are two linearly independent elements of the vector space of traces on $`𝔻^{\mathrm{}}(M)`$, the rest of the statement of above Proposition follows. ∎
## 6. The algebraic index theorem for the Lie algebroid $``$
The following Theorem is proved in and is an extension to the symplectic Lie algebroid $`(,[,],\omega )`$ of the Riemann Roch theorem (on symplectic manifolds) for periodic cochains of , .
###### Theorem 2.
The following diagram is commutative:
where $`\sigma `$ is the specialization map at $`\mathrm{}=0`$, $`\mu `$ is the Hochschild-Kostant-Rosenberg map, $`\mu ^{\mathrm{}}`$ is the trace density map defined in and $`\theta =\frac{1}{\sqrt{1}\mathrm{}}\omega +_{k0}\mathrm{}^k\theta _k{}_{}{}^{}H_{}^{2}(M,[[\mathrm{}]])`$ is the characteristic class of the deformation of the symplectic Lie algebroid $`(,[,],\omega )`$ ().
The natural injection $`𝔸^{\mathrm{}}(M)𝔻^{\mathrm{}}(M)`$ induces a natural map:
$$CC_{}^{\mathrm{per}}(𝔸^{\mathrm{}}(M))CC_{}^{\mathrm{per}}(𝔻^{\mathrm{}}(M))$$
Since the traces $`\tau _{\mathrm{can}}`$ and $`\tau _1`$ of Proposition 6, and the trace density map $`\mu ^{\mathrm{}}`$ extend to $`CC_{}^{\mathrm{per}}(𝔻^{\mathrm{}}(M))`$, they can be identified using the following result.
###### Proposition 7.
1) The $``$vector space $`{}_{}{}^{}H_{}^{2n}(M,)`$ is two-dimensional. The vector space of $``$linear forms on $`{}_{}{}^{}\mathrm{\Omega }_{}^{2n}(M)`$ which vanish on the range of the $``$exterior derivative $`{}_{}{}^{}d`$ admits a unique linear basis $`({}_{}{}^{\mathrm{reg}},_1)`$, characterized by the following properties
For any $`(\alpha ,\beta ){}_{}{}^{}\mathrm{\Omega }_{}^{2n}(M)`$ such that $`\alpha `$ (resp. $`\beta `$) is zero in a neighborhood of the boundary of $`\overline{B}^{}X`$ (resp. $`\overline{B}^{}Y`$)
$${}_{}{}^{\mathrm{reg}}(\alpha ,\beta )=_{T^{}X}\alpha _{T^{}Y}\beta ,_1(\alpha ,\beta )=0.$$
Moreover $`_1\mu ^{\mathrm{}}=\mathrm{Res}_W`$ .
2) There exists a constant $`C`$ such that
$$\tau _{\mathrm{can}}={}_{}{}^{\mathrm{reg}}\mu ^{\mathrm{}}+C_1\mu ^{\mathrm{}}$$
Moreover $`_1\mu ^{\mathrm{}}(1,1)=0`$ and for any $`(a,b)𝔻^{\mathrm{}}(M)`$ such that $`a`$ is zero in a neighborhood of the boundary of $`\overline{B}^{}X`$, $`_1\mu ^{\mathrm{}}(a,b)=0`$.
###### Proof.
1) A standard Mayer-Vietoris sequence argument shows that $`{}_{}{}^{}H_{}^{2n}(M,)`$ is indeed two dimensional. The fact that $`({}_{}{}^{\mathrm{reg}},_1)`$ defines a basis is left to the reader.
2) This is an easy consequence from part 1) and of the properties (see , ) of the trace density map $`\mu ^{\mathrm{}}`$. ∎
## 7. Local formula for the index of a Fourier Integral Operator
###### Theorem 3.
Let $`\mathrm{\Phi }`$ be a Fourier Integral Operator and $`𝔸^{\mathrm{}}(M)`$ the formal deformation of $`M`$ associated to it as in Definition 2. Then
$$\mathrm{ind}\mathrm{\Phi }=_M\mathrm{e}^{\theta _0}\widehat{A}(M),$$
where $`\theta _0`$ denotes the characteristic class of the deformation of the Lie Algebroid ($`,[,],\omega `$) given by $`𝔸^{\mathrm{}}(M)`$.
Let $`_X`$ be a connection $`_X`$ on the tangent bundle $`T(\overline{B}^{}X)`$ and $`\widehat{A}(T^{}X)`$ an associated representative form of the $`\widehat{A}`$class of $`_X`$. The symplectomorphism $`\varphi `$ induces a connection $`\varphi _{}(_X)`$ on the tangent space of $`\overline{B}^{}YB^{}(Y)`$. Let $`_Y`$ denote its extension to a connection of $`T(\overline{B}^{}Y)`$ and $`\widehat{A}(T^{}Y)`$ an associated representative differential form of the $`\widehat{A}`$-class of $`_Y`$. Then
$$\mathrm{ind}\mathrm{\Phi }=_{B^{}(X)}\mathrm{e}^{\theta _0}\widehat{A}(T^{}X)_{B^{}(Y)}\mathrm{e}^{\theta _0}\widehat{A}(T^{}Y)$$
###### Proof.
One obtain this formula by first applying Proposition 3, Theorem 2 and Proposition 7 and then by letting $`\mathrm{}0^+`$. As we will see below, the involved characteristic classes of vector bundles on $`M`$ are in fact standard de Rham cohomology classes and hence the regularized integral coincides with the orientation class of $`M`$. ∎
The previous formula shows that if $`\varphi `$ extends as a symplectomorphism $`T^{}XT^{}Y`$ up to the zero section then $`\mathrm{ind}\mathrm{\Phi }=\mathrm{\hspace{0.17em}0}`$.
For a deformation associated with a Fourier Integral Operator (as in Proposition 5) the characteristic class $`\theta `$ of Theorem 2 is in fact of the form
$$\theta =\frac{1}{\sqrt{1}\mathrm{}}\omega +\theta _0$$
where $`\theta _0H^2(M,)`$ is a closed differential form (not only an $``$differential form). In order to do this and to identify the relevant characteristic class we will give below a slightly nonstandard description of a formal deformation.
### 7.1. General construction of the characteristic class of a formal deformation
Let us start with some notation.
Let $`𝔸^{\mathrm{}}`$ denote the Weyl algebra of the symplectic vector space $`^{2n}`$ with the standard symplectic structure, i.e. the algebra generated by the vectors $`\widehat{x}_l,\widehat{\xi }_l`$ ($`1ln`$) satisfying the relations $`[\widehat{\xi }_k,\widehat{x}_l]=\sqrt{1}\mathrm{}\delta _{k,l}`$. The algebra $`𝔸^{\mathrm{}}`$ is completed in the topology associated to the ideal generated by $`\{\widehat{x}_l,\widehat{\xi }_l,\mathrm{};1ln\}`$ and has the grading induced by
$$deg\widehat{x}_l=deg\widehat{\xi }_l=1,deg\mathrm{}=2.$$
The corresponding Lie algebra $`\frac{1}{\mathrm{}}𝔸^{\mathrm{}}`$ will be denoted by $`\stackrel{~}{𝔤}`$. We set:
$$𝔤=\text{Der }(𝔸^{\mathrm{}})=\stackrel{~}{𝔤}/\text{center}$$
and
$$G=\text{Aut}(𝔸^{\mathrm{}})=\mathrm{exp}(𝔤_0)$$
We set
$$\stackrel{~}{G}=\{g\frac{1}{\mathrm{}}𝔸^{\mathrm{}}g𝔰p(2n,)mod𝔤_1\}$$
and will endow it with the group structure coming from the exponential map. Note that $`\stackrel{~}{G}`$ is an extension of $`G`$ associated to the (Lie algebra) central extension $`\stackrel{~}{𝔤}`$ of $`𝔤`$.
We endow the bundle $`^{2n}\times 𝔸^{\mathrm{}}`$ with the obvious fiber-wise action of $`\stackrel{~}{G}`$ and with the $`\stackrel{~}{g}`$-valued (Fedosov) connection
$$\stackrel{~}{}^0=\underset{l=1}{\overset{n}{}}(d\xi _l(_{\xi _l}\frac{1}{\sqrt{1}\mathrm{}}\widehat{x}_l)+dx_l(_{x_l}+\frac{1}{\sqrt{1}\mathrm{}}\widehat{\xi }_l)).$$
Let us recall (see Section 2.2) that a local chart of $`\overline{B}^{}^n`$ near $`\overline{B}^{}^nT^{}^n`$ is given by:
(11)
$$(x_1,\mathrm{},x_n;t=\frac{1}{\xi },\theta =(\theta _1,\mathrm{}\theta _{n1}))t0,\theta S^{n1}$$
By using the local coordinates (11) one checks easily that $`\stackrel{~}{}^0`$ extends as an $`^^n`$connection, still denoted $`\stackrel{~}{}^0`$, of $`\overline{B}^{}^n\times 𝔸^{\mathrm{}}`$.
The description given below of a formal deformation of a symplectic Lie algebroid structure on $`M`$ is just the representation of the Fedosov construction in terms of the bundle of jets on $`M`$ with the fiber-wise product structure induced by the \*-product (which is isomorphic to Weyl bundle) .
Local description of the characteristic class $`\theta `$ of a formal deformation.
The deformation is described by a local (Darboux) cover $`\{U_i\}_{iI}`$ of $`(M,\omega )`$, a collection of functions $`\{g_{i,j}:U_iU_j\stackrel{~}{G}\}`$ and a collection of $`\stackrel{~}{𝔤}`$-valued $``$connections $`\stackrel{~}{}_i`$ on $`U_i\times 𝔸^{\mathrm{}}`$ which, when expressed in terms of local Darboux coordinates $`(x_1,\mathrm{},x_n,\xi _1,\mathrm{},\xi _n)`$ (resp (11)) if $`U_i`$ does not meet (resp meets) the boundary at infinity , are equal to $`\stackrel{~}{}^0`$ modulo $`\stackrel{~}{𝔤}_1`$ and so that the three following conditions hold.
1) The cocycle condition:
$$g_{i,j}g_{j,i}=1\mathrm{and}g_{i,j}g_{j,k}=g_{i,k}\mathrm{on}U_iU_jU_k$$
In particular $`\{g_{i,j}:U_iU_j\stackrel{~}{G}\}`$ define a smooth bundle $`𝒲`$ of algebras over $`M`$ with fiber isomorphic to $`𝔸^{\mathrm{}}`$ and the structure group $`\stackrel{~}{G}`$.
2) The local connections $`\stackrel{~}{}_i`$ define a $`\stackrel{~}{𝔤}`$-valued connection $`\stackrel{~}{}`$ on the bundle $`𝒲`$, i.e.:
$$g_{i,j}\stackrel{~}{}_j=\stackrel{~}{}_ig_{i,j}$$
3) The induced $`𝔤`$-valued connection $``$ on the bundle $`𝒲`$ is flat, i. e. $`\theta =\stackrel{~}{}^2`$ is a globally defined differential form on $`M`$ with values in the center of $`\stackrel{~}{𝔤}`$, necessarily of the form
$$\frac{1}{\sqrt{1}\mathrm{}}\omega +\theta _0\text{ where }\theta _0\mathrm{\Omega }^2(M,[[\mathrm{}]]).$$
The algebra of $``$-flat sections of $`𝒲`$ is a formal deformation of $`(M,\omega )`$ whose characteristic class is $`\theta `$.
### 7.2. Local canonical liftings
We endow $`^{2n}`$ with its canonical symplectic structure $`\omega =_{l=1}^nd\xi _ldx_l`$. Given any smooth, $`[[\mathrm{}]]`$-valued function $`H`$ on $`^{2n}`$, we set
$$H_0=H(x,\xi )|_{\mathrm{}=0},H_1=\underset{l=1}{\overset{n}{}}(\widehat{x}_l_{x_l}H_0+\widehat{\xi }_l_{\xi _l}H_0)$$
and
$$\stackrel{~}{H}=\frac{\widehat{x}^\alpha \widehat{\xi }^\beta }{\alpha !\beta !}_x^\alpha _\xi ^\beta H.$$
We will associate to $`H`$ the following $`\stackrel{~}{𝔤}`$-lift of the Lie derivative $`_{\{H,\}}`$:
$$𝒟_H=_{\{H,\}}+\frac{1}{\mathrm{}}(\stackrel{~}{H}H_0H_1+\frac{1}{2}\mathrm{}\underset{l=1}{\overset{n}{}}_{x_l,\xi _l}^2H_0).$$
We can think of it as an element of the Lie algebra of the semidirect product of $`C^{\mathrm{}}(^{2n},\stackrel{~}{G})`$ by the pseudogroup of local diffeomorphisms of $`^{2n}`$. The $`𝒟_H`$’s form a Lie algebra, in fact
$$[𝒟_H,𝒟_K]=𝒟_{\frac{1}{\mathrm{}}(HKKH)},$$
and they satisfy
$$[𝒟_H,\stackrel{~}{}^0]=\frac{1}{2}d(\underset{l=1}{\overset{n}{}}_{x_l,\xi _l}^2H_0).$$
We will also have an occasion to use
(12)
$$𝒟_H^0=_{\{H,\}}+\frac{1}{\mathrm{}}(\stackrel{~}{H}H_0H_1),$$
which commutes with $`\stackrel{~}{}^0`$.
### 7.3. The cotangent bundle case
The deformation of $`T^{}X`$ associated to the sheaf of differential operators on $`X`$ can be now described as follows.
Locally on a coordinate domain $`UX`$ we use coordinates on $`U`$ to give an explicit symplectomorphism
$$T^{}U^{2n}$$
and use Weyl deformation of $`^{2n}`$ to construct the deformation of $`T^{}U`$. This amounts to the choice of a ($`\stackrel{~}{𝔤}`$-valued) connection given in our local coordinates $`(x_1,\mathrm{},x_n)`$ on $`U`$ and the induced local coordinates $`(x_i,\xi _i)_{i=1,\mathrm{}n}`$ on $`T^{}UU\times ^n`$ by
$$\stackrel{~}{}^0=d+\frac{1}{\mathrm{}}\underset{l=1}{\overset{n}{}}(dx_l\widehat{\xi }_ld\xi _l\widehat{x}_l).$$
The infinitesimal change of coordinates on $`U`$ is given by a vector field of the form $`_{l=1}^nX_l_{x_l}`$ and the associated infinitesimal symplectomorphism of $`T^{}U`$ is given by the Hamiltonian vector field $`\{_{l=1}^nX_l\xi _l,\}`$.
It is immediate to see that the map
$$\underset{l}{}X_l_{x_l}𝒟_{_lX_l\xi _l}$$
is the Lie algebra homomorphism.
The associated local diffeomorphisms (coordinate changes) $`\mathrm{exp}_lX_l_{x_l}`$ lift to a local isomorphisms of the bundle $`T^{}U\times 𝔸^{\mathrm{}}`$ given by $`\mathrm{exp}𝒟_{_lX_l\xi _l}`$.
Given a local coordinate cover $`\{U_i\}_{iI}`$ of $`X`$ it is now immediate to construct the associated $`\stackrel{~}{G}`$-valued cocycle $`\{g_{ij}\}`$ glueing the bundles together. Note that, since $`𝒟`$’s do not commute with the connection $`\stackrel{~}{}^0`$, the corresponding collection of connections
$$\stackrel{~}{}_i=\stackrel{~}{}^0\text{in i’th coordinate system on }T^{}U_i\text{ }$$
do not glue together. But it is not difficult to check that
$$g_{ij}\stackrel{~}{}_ig_{ji}=\frac{1}{2}d\mathrm{log}detDg_{ij},$$
where $`Dg_{ij}`$ is the induced action of $`g_{ij}`$ on the tangent bundle. By trivializing the cocycle $`\frac{1}{2}d\mathrm{log}detDg_{ij}`$ in $`\stackrel{ˇ}{C}^1(X,\mathrm{\Omega }^1(T^{}X))`$ we get a globally defined connection $`\stackrel{~}{}`$ and it is immediate that the characteristic class of the associated deformation is $`\frac{1}{2}\pi ^{}c_1(T_{}X)`$, where $`\pi :T^{}XX`$ is the canonical projection. It is also immediate that the deformation constructed in this way coincides with the one associated to the calculus of differential operators on $`X`$, while its jet at $`\xi =\mathrm{}`$ gives the deformation associated to the calculus of pseudodifferential operators on $`X`$, the characteristic class being given by the jets at $`\xi =\mathrm{}`$ of $`\frac{1}{2}c_1(_{})`$ (recall that the real symplectic vector bundle $``$ is the realification of a complex vector bundle $`_{}`$).
### 7.4. The Lie algebroid case
Recall now that the Lie algebroid (on $`M`$) $`(,[,],\omega )`$ is given by glueing (at infinity) the two cotangent bundles $`(T^{}X,\omega ^X)`$ and $`(T^{}Y,\omega ^Y)`$ by the symplectomorphism $`\varphi ^{}`$. To construct the deformation in this case, we will use the following data, whose existence follows immediately from the compactness of the co-sphere bundles of $`X`$ and $`Y`$.
1. A local coordinate cover $`\{U_i\}_{iI}`$ of $`X`$ and an open relatively compact neighborhood $`U_X`$ of the zero section in $`T^{}X`$;
2. A local coordinate cover $`\{V_i\}_{iI}`$ of $`Y`$ and an open relatively compact neighborhood $`U_Y`$ of the zero section in $`T^{}Y`$;
3. For each $`iI`$ a one-homogeneous real-valued function $`H_i`$ on $`T^{}XXT^{}YY`$ such that the restriction $`\varphi _i`$ of the symplectomorphism $`\varphi `$ to $`T^{}U_iU_X`$ is given by integrating the (time dependent) hamiltonian flow $`_{H_i}`$.
Using the above data, we can construct cocycles
$$T^{}U_iT^{}U_jpg_{ij}(p)C^{\mathrm{}}(T^{}U_iT^{}U_j,\stackrel{~}{G})$$
and
$$T^{}V_iT^{}V_jph_{ij}(p)C^{\mathrm{}}(T^{}V_iT^{}V_j,\stackrel{~}{G})$$
intertwining the flat connections $`\stackrel{~}{}_i^0`$ up to the term $`\frac{1}{2}d\mathrm{log}detDg_{ij}`$ ($`\frac{1}{2}d\mathrm{log}detDh_{ij}`$ respectively) as in the cotangent bundle case. We can also construct, using the notation of (12), a lifting of $`\varphi _i`$
$$\mathrm{exp}𝒟_{H_i}^0=\mathrm{\Psi }_i:\mathrm{\Gamma }(T^{}U_iU_X,𝔸^{\mathrm{}})\mathrm{\Gamma }(T^{}V_iU_Y,𝔸^{\mathrm{}}).$$
¿From now on we view $`\mathrm{\Phi }_i`$ as local isomorphisms of jets at infinity of the $`\stackrel{~}{G}`$-bundles on compactified cotangent bundles of $`X`$ and $`Y`$ constructed from the cocycles $`g_{i,j}`$ and $`h_{i,j}`$. While both $`g_{i,j}`$’s and $`h_{i,j}`$’s do satisfy the cocycle conditions on $`T^{}(X)`$ ($`T^{}(Y)`$ respectively), however
$$\lambda _{ij}=\mathrm{\Psi }_j^1h_{ji}\mathrm{\Psi }_ig_{ij}1.$$
and hence we do not yet have the data necessary to construct the bundle $`𝒲`$ over $`M`$.
The following facts are easy corollaries of the construction:
1. $`\lambda _{ij}=1\text{ mod }\stackrel{~}{G}_1`$;
2. $`\lambda _{ij}\stackrel{~}{}_j^0\lambda _{j,i}=\stackrel{~}{}_i^0\frac{1}{2}d\mathrm{log}detDg_{ij}\frac{1}{2}d\mathrm{log}detDh_{ij}`$;
3. $`\lambda _{ij}`$ form a two-cocycle with values in $`\stackrel{~}{G}`$.
To begin with, note that both $`\frac{1}{2}d\mathrm{log}detDg_{ij}`$ and $`\frac{1}{2}d\mathrm{log}detDh_{ij}`$ as cohomology classes on $`T^{}XX`$ and $`T^{}YY`$ represent (under our symplectomorphism) the same cohomology class, to wit half of the first Chern class of the tangent bundle with the complex structure induced by the symplectic form. Since these vanish, we can find a zero-$`\stackrel{ˇ}{C}`$ech cochain $`\tau _i`$ of the sheaf of functions with values in $`\{0\}`$ and such that $`\tau _i\lambda _{ij}\tau _j^1`$ intertwines the (local) flat connections $`\stackrel{~}{}_i^0`$ and $`\stackrel{~}{}_j^0`$.
In particular, $`\tau _i\lambda _{ij}\tau _j^1`$ are given by exponentials of jets of $`\stackrel{~}{}_i^0`$-flat sections of the bundle $`T^{}U_i\times 𝔸^{\mathrm{}}`$ and, using a partition of unity, they can be written in the form $`\tau _i\lambda _{ij}\tau _j^1=\lambda _i\lambda _j^1`$, where $`\lambda _i`$ is a jet of a flat section of the Weyl bundle supported on $`T^{}U_iU_X`$. We now define an operator $`\mathrm{\Psi }`$, acting on the set of sections of the Weyl bundle $`𝒲`$, by setting for each $`iI`$:
$$\mathrm{\Psi }_{|T^{}U_iU_X}=\mathrm{\Psi }_i\tau _i_{\lambda _i}.$$
Here $`_{\lambda _i}`$ stands for the operator of multiplication with the flat section $`\lambda _i`$. It is easy to see that $`\mathrm{\Psi }`$ descends to an isomorphism of jets at the sphere at infinity of the deformations of cotangent bundles constructed above so that
$$aC^{\mathrm{}}(\overline{B}^{}X),\mathrm{\Psi }(a)=(a+\underset{k1}{}\mathrm{}^k\widehat{D}_k(a))\varphi ^1$$
holds in $`𝔹^{\mathrm{}}(\overline{B}^{}Y)`$, where the $`\widehat{D}_k`$ are $``$differential operators. Hence, as in Proposition 5, $`\mathrm{\Psi }`$ induces a deformation of the Lie algebroid $`(,[,],\omega )`$.
### 7.5. The characteristic class of $`𝔸^{\mathrm{}}(M)`$
The characteristic class of the deformation constructed above can now be easily obtained as follows. The collection $`g_{ij},h_{ij}`$, and the jet at infinity of $`\mathrm{\Psi }_i_{\lambda _i}`$ give a cocycle with values in $`\stackrel{~}{G}`$, and it commutes with local flat connections up to the Cech cocycle given by the collection of differential forms
(13)
$$\frac{1}{2}d\mathrm{log}detDg_{ij},d\mathrm{log}\tau _i,\frac{1}{2}d\mathrm{log}detDh_{ij}.$$
As in the case of cotangent bundle, we can correct local connections by a scalar term. The characteristic class $`\theta _0`$ of the deformation is given by (13) as a cochain in $`\stackrel{ˇ}{C}^1(M,\mathrm{\Omega }^1(T^{}M))`$ Moreover, in the case that both $`X`$ and $`Y`$ admit metalinear structures, the collection $`\{\tau _i\}_{iI}`$ can be thought of as glueing of the pulled back)of the half-top form bundles of $`X`$ and $`Y`$ along the graph of the symplectomorphism into a line bundle $``$ over $`M`$ and, in this case,
$$\theta _0=c_1()$$
### 7.6. The Fourier Integral Operator
To get the Fourier integral operator we will work locally. We will dispense with the half-density bundles (trivial in any case) for the sake of simplicity of notation. We will begin by constructing, for each i, an operator on $`L^2(^n)`$ as follows. Choosing local coordinates on $`U_i`$ and $`V_i`$, we can assume that $`H_i`$ (introduced at the end of section 7.4) is actually a smooth function on $`T^{}^{2n}`$ which is 1-homogeneous in the cotangent direction. The solution of the following differential equation
$$\frac{d}{dt}T_i(t)=\text{Op}(\sqrt{1}H_i)T_i(t),T_i(0)=1$$
is a smooth family of bounded operators. Using the fact that $`\{\tau _i\}_{iI}`$ is a $`\stackrel{ˇ}{C}`$ech zero- cochain of the sheaf of functions with values in the unit circle and proceeding as in , one checks that $`T_i(1)`$ satisfies
$$\text{Ad }(T_i(1)\text{Op}(\lambda _i))\text{Op}(f_{\mathrm{}})\text{Op}(\mathrm{\Psi }(f)_{\mathrm{}})$$
mod $`\mathrm{}^+\mathrm{}`$ as $`\mathrm{}0`$ whenever supp$`fT^{}U_iX`$ ( recall that $`\lambda _i`$ is introduced in Section 7.4). In other words, the deformation constructed above is associated (in the sense of Proposition 5) to the almost unitary Fourier Integral Operator $`\mathrm{\Phi }=_{iI}T_i(1)\text{Op}(\lambda _i)`$ whose canonical relation is $`C_\varphi `$. Moreover, the index of this operator $`\mathrm{\Phi }_0`$ is given by
$$_M\widehat{A}(M)e^{\theta _0},$$
###### Remark 4.
The result above depends on the choice of the $`\tau _i`$’s which in turn determine the homotopy class of the symbol of the Fourier Integral Operator. Moreover, since the characteristic classes involved are given by differential forms associated to connections on a vector bundle over $`M`$ and $`\mathrm{\Omega }{}_{}{}^{}\mathrm{\Omega }`$, the $``$-classes involved in the index formulas are in fact identical with corresponding standard characteristic classes.
Let us recall that the real vector bundle $`TM`$ is given by realification of a complex vector bundle $`_{}`$ on $`M`$ (the almost complex structure coming from the symplectic vector bundle structure on $``$). Moreover, it is easy to see, that there exists a choice of the $`\tau _i`$’s such that the associated characteristic class of the deformation coincides with $`\frac{1}{2}c_1(_{})`$. This gives the following result (compare with and ):
###### Theorem 4.
There exists an almost unitary Fourier integral operator $`\mathrm{\Phi }_0`$ (as in Section 3) whose canonical relation is $`C_\varphi `$ and such that:
$$\mathrm{ind}\mathrm{\Phi }_0=_M\widehat{A}(M)e^{\frac{1}{2}c_1(_{})}$$
|
warning/0004/hep-ph0004222.html
|
ar5iv
|
text
|
# Untitled Document
IFT/00–12
Two sterile neutrinos as a consequence
of matter structure<sup>*</sup><sup>*</sup>* Work supported in part by the Polish KBN–Grant 2 P03B 052 16 (1999–2000).
Wojciech Królikowski
Institute of Theoretical Physics, Warsaw University
Hoża 69, PL–00–681 Warszawa, Poland
Abstract
An algebraic argument based on a series of generalized Dirac equations, truncated by an ”intrinsic Pauli principle”, shows that there should exist two sterile neutrinos as well as three families of leptons and quarks. Then, the influence of these additional neutrinos on neutrino oscillations is studied. As an example, a specific model of effective five–neutrino texture is proposed, where only the nearest neighbours in the sequence of five neutrinos ordered as $`\nu _s,\nu _s^{},\nu _e,\nu _\mu ,\nu _\tau `$ are coupled through the $`5\times 5`$ mass matrix. Its diagonal elements are taken as negligible in comparison with its nonzero off–diagonal entries.
PACS numbers: 12.15.Ff , 14.60.Pq , 12.15.Hh .
April 2000
1. Introduction
The existence problem of light sterile neutrinos , free of Standard Model gauge charges, is connected phenomenologically with the LSND effect for accelerator $`\nu _\mu `$’s . If confirmed, it would avail (jointly with the observed deficits of solar $`\nu _e`$’s and atmospheric $`\nu _\mu `$’s ) the existence of the third mass–square difference in neutrino oscillations, invoking necessarily at least one kind of light sterile neutrinos, in addition to the familiar three active neutrinos $`\nu _e,\nu _\mu ,\nu _\tau `$. From the theoretical viewpoint, however, light sterile neutrinos might exist even in the case, when the LSND effect was not confirmed . At any rate, there might be either three sorts of light Majorana sterile neutrinos
$$\nu _\alpha ^{(s)}=\nu _{\alpha R}+(\nu _{\alpha R})^c(\alpha =e,\mu ,\tau )$$
(1)
being structurally righthanded counterparts of familiar light Majorana active neutrinos
$$\nu _\alpha ^{(a)}=\nu _{\alpha L}+(\nu _{\alpha L})^c(\alpha =e,\mu ,\tau ),$$
(2)
or some quite new Dirac or Majorana light sterile neutrinos.
The first kind appears in the case of pseudo–Dirac option for neutrinos $`\nu _e,\nu _\mu ,\nu _\tau `$ that contrasts with the popular seesaw option involving heavy sterile neutrinos of the form (1). If the LSND effect did not exist, the seesaw option, operating effectively at low energies with three active neutrinos (2) only, would be phenomenologically most economical, beside the simple option of Dirac neutrinos $`\nu _\alpha =\nu _{\alpha L}+\nu _{\alpha R}`$. At the same time, the seesaw option would be favourable from the standpoint of GUT idea (say, in the SO(10) version), where a large mass scale of sterile neutrinos of the form (1) could be understood, at least qualitatively. On the other hand, however, such an option would not meet the needs of astrophysics for light sterile neutrinos useful in tentatively explaining heavy–element nucleosynthesis .
The sterile neutrinos of the second kind are suggested to exist in the framework of a theoretical scheme based on the series $`N=1,2,3,\mathrm{}`$ of generalized Dirac equations
$$\left\{\mathrm{\Gamma }^{(N)}\left[pgA(x)\right]M^{(N)}\right\}\psi ^{(N)}(x)=0,$$
(3)
where for any $`N`$ the Dirac algebra
$$\{\mathrm{\Gamma }_\mu ^{(N)},\mathrm{\Gamma }_\nu ^{(N)}\}=2g_{\mu \nu }$$
(4)
is constructed by means of a Clifford algebra,
$$\mathrm{\Gamma }_\mu ^{(N)}\underset{i=1}{\overset{N}{}}\gamma _{i\mu }^{(N)},\{\gamma _{i\mu }^{(N)},\gamma _{j\nu }^{(N)}\}=2\delta _{ij}g_{\mu \nu }$$
(5)
with $`i,j=1,2,\mathrm{},N`$ and $`\mu ,\nu =0,1,2,3`$. Here, the term $`g\mathrm{\Gamma }^{(N)}A(x)`$ symbolizes the Standard Model gauge coupling, involving $`\mathrm{\Gamma }_5^{(N)}i\mathrm{\Gamma }_0^{(N)}\mathrm{\Gamma }_1^{(N)}\mathrm{\Gamma }_2^{(N)}\mathrm{\Gamma }_3^{(N)}`$ as well as the color, weak–isospin and hypercharge matrices (this coupling is absent for sterile particles such as sterile neutrinos). The mass $`M^{(N)}`$ is independent of $`\mathrm{\Gamma }_\mu ^{(N)}`$. In general, the mass $`M^{(N)}`$ should be replaced by a mass matrix of elements $`M^{(N,N^{})}`$ which would couple $`\psi ^{(N)}(x)`$ with all appropriate $`\psi ^{(N^{})}(x)`$, and it might be natural to assume for $`NN^{}`$ that $`[\gamma _{i\mu }^{(N)},\gamma _{j\nu }^{(N^{})}]=0`$ i.e., $`[\mathrm{\Gamma }_\mu ^{(N)},\mathrm{\Gamma }_\nu ^{(N^{})}]=0`$.
The Dirac–type equation (3) for any $`N`$ implies that
$$\psi ^{(N)}(x)=\left(\psi _{\alpha _1\alpha _2\mathrm{}\alpha _N}^{(N)}(x)\right),$$
(6)
where each $`\alpha _i=1,2,3,4`$ is the Dirac bispinor index defined in its chiral representation in which the matrices
$$\gamma _{j5}^{(N)}i\gamma _{j0}^{(N)}\gamma _{j1}^{(N)}\gamma _{j2}^{(N)}\gamma _{j3}^{(N)},\sigma _{j3}^{(N)}\frac{i}{2}[\gamma _{j1}^{(N)},\gamma _{j2}^{(N)}]$$
(7)
are diagonal (note that all matrices (7), both with equal and different $`j`$’s, commute simultaneously). The wave function or field $`\psi ^{(N)}(x)`$ for any $`N`$ carries also the Standard Model (composite) label, suppressed in our notation. The mass $`M^{(N)}`$ gets also such a label. The Standard Model coupling of physical Higgs bosons should be eventually added to Eq. (3) for any $`N`$.
For $`N=1`$ Eq. (3) is, of course, the usual Dirac equation, for $`N=2`$ it is known as the Dirac form of the Kähler equation , while for $`N3`$ Eqs. (3) give us new Dirac–type equations . All of them describe some spin–halfinteger or spin–integer particles for $`N`$ odd and $`N`$ even, respectively. The nature of these particles is the main subject of the present paper (cf. also Ref. ).
The Dirac–type matrices $`\mathrm{\Gamma }_\mu ^{(N)}`$ for any $`N`$ can be embedded into the new Clifford algebra
$$\{\mathrm{\Gamma }_{i\mu }^{(N)},\mathrm{\Gamma }_{j\nu }^{(N)}\}=2\delta _{ij}g_{\mu \nu }$$
(8)
\[isomorphic with the Clifford algebra introduced for $`\gamma _{i\mu }^{(N)}`$ in Eq. (5)\], if $`\mathrm{\Gamma }_{i\mu }^{(N)}`$ are defined by the properly normalized Jacobi linear combinations of $`\gamma _{i\mu }^{(N)}`$. In fact, they are given as
$$\mathrm{\Gamma }_{1\mu }^{(N)}\mathrm{\Gamma }_\mu ^{(N)}\frac{1}{\sqrt{N}}\underset{i=1}{\overset{N}{}}\gamma _{i\mu }^{(N)},\mathrm{\Gamma }_{i\mu }^{(N)}\frac{1}{\sqrt{i(i1)}}\left[\gamma _{1\mu }^{(N)}+\mathrm{}+\gamma _{(i1)\mu }^{(N)}(i1)\gamma _{i\mu }^{(N)}\right]$$
(9)
for $`i=1`$ and $`i=2,\mathrm{},N`$, respectively. So, $`\mathrm{\Gamma }_1^{(N)}`$ and $`\mathrm{\Gamma }_2^{(N)},\mathrm{},\mathrm{\Gamma }_N^{(N)}`$ represent respectively the ”centre–of–mass” and ”relative” Dirac–type matrices. Note that the Dirac–type equation (3) for any $`N`$ does not involve the ”relative” Dirac–type matrices $`\mathrm{\Gamma }_2^{(N)},\mathrm{},\mathrm{\Gamma }_N^{(N)}`$, solely including the ”centre–of–mass” Dirac–type matrix $`\mathrm{\Gamma }_1^{(N)}\mathrm{\Gamma }^{(N)}`$. Since $`\mathrm{\Gamma }_i^{(N)}=_{j=1}^NO_{ij}\gamma _j^{(N)}`$, where the $`N\times N`$ matrix $`O=\left(O_{ij}\right)`$ is orthogonal ($`O^T=O^1`$), we obtain for the total spin tensor the formula
$$\underset{i=1}{\overset{N}{}}\sigma _{i\mu \nu }^{(N)}=\underset{i=1}{\overset{N}{}}\mathrm{\Sigma }_{i\mu \nu }^{(N)},$$
(10)
where
$$\sigma _{j\mu \nu }^{(N)}\frac{i}{2}[\gamma _{j\mu }^{(N)},\gamma _{j\nu }^{(N)}],\mathrm{\Sigma }_{j\mu \nu }^{(N)}\frac{i}{2}[\mathrm{\Gamma }_{j\mu }^{(N)},\mathrm{\Gamma }_{j\nu }^{(N)}].$$
(11)
Of course, the spin tensor (10) is the generator of Lorentz transformations for $`\psi ^{(N)}(x)`$.
It is convenient for any $`N`$ to pass from the chiral representations for individual $`\gamma _i^{(N)}`$’s to the chiral representations for Jacobi $`\mathrm{\Gamma }_i^{(N)}`$’s in which the matrices
$$\mathrm{\Gamma }_{j5}^{(N)}i\mathrm{\Gamma }_{j0}^{(N)}\mathrm{\Gamma }_{j1}^{(N)}\mathrm{\Gamma }_{j2}^{(N)}\mathrm{\Gamma }_{j3}^{(N)},\mathrm{\Sigma }_{j3}^{(N)}\frac{i}{2}[\mathrm{\Gamma }_{j1}^{(N)},\mathrm{\Gamma }_{j2}^{(N)}]$$
(12)
are diagonal (they all, both with equal and different $`j`$’s, commute simultaneously). Note that $`\mathrm{\Gamma }_{15}^{(N)}\mathrm{\Gamma }_5^{(N)}`$ is the Dirac–type chiral matrix as it is involved in the Standard Model gauge coupling in the Dirac–type equation (3).
Using the new Jacobi chiral representations, the ”centre–of–mass” Dirac-type matrices $`\mathrm{\Gamma }_{1\mu }^{(N)}\mathrm{\Gamma }_\mu ^{(N)}`$ and $`\mathrm{\Gamma }_{15}^{(N)}\mathrm{\Gamma }_5^{(N)}`$ can be taken in the reduced forms
$$\mathrm{\Gamma }_\mu ^{(N)}=\gamma _\mu \underset{N1\mathrm{times}}{\underset{}{\mathrm{𝟏}\mathrm{}\mathrm{𝟏}}},\mathrm{\Gamma }_5^{(N)}=\gamma _5\underset{N1\mathrm{times}}{\underset{}{\mathrm{𝟏}\mathrm{}\mathrm{𝟏}}},$$
(13)
where $`\gamma _\mu `$, $`\gamma _5i\gamma _0\gamma _1\gamma _2\gamma _3`$ and 1 are the usual $`4\times 4`$ Dirac matrices. For instance, the Jacobi $`\mathrm{\Gamma }_{i\mu }^{(N)}`$’s and $`\mathrm{\Gamma }_{i5}^{(N)}`$’s for $`N=3`$ can be chosen as
$$\begin{array}{ccccccc}\mathrm{\Gamma }_{1\mu }^{(3)}\hfill & =\hfill & \gamma _\mu \mathrm{𝟏}\mathrm{𝟏}\hfill & ,\hfill & \mathrm{\Gamma }_{15}^{(3)}\hfill & =\hfill & \gamma _5\mathrm{𝟏}\mathrm{𝟏},\hfill \\ & & & & & & \\ \mathrm{\Gamma }_{2\mu }^{(3)}\hfill & =\hfill & \gamma _5i\gamma _5\gamma _\mu \mathrm{𝟏}\hfill & ,\hfill & \mathrm{\Gamma }_{25}^{(3)}\hfill & =\hfill & \mathrm{𝟏}\gamma _5\mathrm{𝟏},\hfill \\ & & & & & & \\ \mathrm{\Gamma }_{3\mu }^{(3)}\hfill & =\hfill & \gamma _5\gamma _5\gamma _\mu \hfill & ,\hfill & \mathrm{\Gamma }_{35}^{(3)}\hfill & =\hfill & \mathrm{𝟏}\mathrm{𝟏}\gamma _5.\hfill \end{array}$$
(14)
Then, the Dirac–type equation (3) for any $`N`$ can be rewritten in the reduced form
$$\left\{\gamma \left[pgA(x)\right]M^{(N)}\right\}_{\alpha _1\beta _1}\psi _{\beta _1\alpha _2\mathrm{}\alpha _N}^{(N)}(x)=0,$$
(15)
where $`\alpha _1`$ and $`\alpha _2,\mathrm{},\alpha _N`$ are the ”centre–of–mass” and ”relative” Dirac bispinor indices, respectively (here, $`(\gamma p)_{\alpha _1\beta _1}=\gamma _{\alpha _1\beta _1}p`$ and $`\left(M^{(N)}\right)_{\alpha _1\beta _1}=\delta _{\alpha _1\beta _1}M^{(N)}`$, but the chiral coupling $`g\gamma A(x)`$ involves within $`A(x)`$ also the matrix $`\gamma _5`$ ). Note that in the Dirac–type equation (15) for any $`N>1`$ the ”relative” indices $`\alpha _2,\mathrm{},\alpha _N`$ are free, but still are subjects of Lorentz transformations (for $`\alpha _2`$ this was known already in the case of Dirac form of Kähler equation corresponding to our $`N=2`$).
Since in Eq. (15) the Standard Model gauge fields interact only with the ”centre–of–mass” index $`\alpha _1`$, this is distinguished from the physically unobserved ”relative” indices $`\alpha _2,\mathrm{},\alpha _N`$. Thus, it was natural for us to conjecture some time ago that the ”relative” bispinor indices $`\alpha _2,\mathrm{},\alpha _N`$ are all undistinguishable physical objects obeying Fermi statistics along with the Pauli principle requiring in turn the full antisymmetry of wave function $`\psi _{\alpha _1\alpha _2,\mathrm{},\alpha _N}(x)`$ with respect to $`\alpha _2,\mathrm{},\alpha _N`$ . Hence, only five values of $`N`$ satisfying the condition $`N14`$ are allowed, namely $`N=1,3,5`$ for $`N`$ odd and $`N=2,4`$ for $`N`$ even. Then, from the postulate of relativity and the probabilistic interpretation of $`\psi ^{(N)}(x)`$ we were able to infer that three $`N`$ odd and two $`N`$ even correspond to states with total spin 1/2 and total spin 0, respectively .
Thus, the Dirac–type equation (3), jointly with the ”intrinsic Pauli principle”, if considered on a fundamental level, justifies the existence in Nature of three and only three families of spin–1/2 fundamental fermions (i.e., leptons and quarks) coupled to the Standard Model gauge bosons. In addition, there should exist two and only two families of spin–0 fundamental bosons also coupled to the Standard Model gauge bosons.
For sterile particles, Eq. (15) with any $`N`$ goes over into the free Dirac–type equation
$$\left(\gamma _{\alpha _1\beta _1}p\delta _{\alpha _1\beta _1}M^{(N)}\right)\psi _{\beta _1\alpha _2\mathrm{}\alpha _N}^{(N)}(x)=0$$
(16)
(as far as only Standard Model gauge interactions are considered). Here, no Dirac bispinor index $`\alpha _i`$ is distinguished by the Standard Model gauge coupling which is absent in this case. The ”centre–of mass” index $`\alpha _1`$ is not distinguished also by its coupling to the particle’s four–momentum, since Eq. (16) is physically equivalent to the free Klein–Gordon equation
$$\left(p^2M^{(N)\mathrm{\hspace{0.17em}2}}\right)\psi _{\alpha _1\alpha _2\mathrm{}\alpha _N}^{(N)}(x)=0.$$
(17)
Thus, in this case the intrinsic Pauli principle requires that $`N4`$, leading to $`N=1,3`$ for $`N`$ odd and $`N=2,4`$ for $`N`$ even. Similarly as before, they correspond to states with total spin 1/2 and total spin 0, respectively .
Therefore, there should exist two and only two spin–1/2 sterile fundamental fermions (i.e., two sterile neutrinos $`\nu _s`$ and $`\nu _s^{}`$) and, in addition, two and only two spin–0 sterile fundamental bosons.
The wave functions or fields of active fermions (leptons and quarks) of three families and sterile neutrinos of two generations can be presented in terms of $`\psi _{\alpha _1\alpha _2\mathrm{}\alpha _N}^{(N)}(x)`$ as follows
$`\psi _{\alpha _1}^{(f)}(x)`$ $`=`$ $`\psi _{\alpha _1}^{(1)}(x),`$
$`\psi _{\alpha _1}^{(f^{})}(x)`$ $`=`$ $`{\displaystyle \frac{1}{4}}\left(C^1\gamma _5\right)_{\alpha _2\alpha _3}\psi _{\alpha _1\alpha _2\alpha _3}^{(3)}(x)=\psi _{\alpha _112}^{(3)}(x)=\psi _{\alpha _134}^{(3)}(x),`$
$`\psi _{\alpha _1}^{(f^{\prime \prime })}(x)`$ $`=`$ $`{\displaystyle \frac{1}{24}}\epsilon _{\alpha _2\alpha _3\alpha _4\alpha _5}\psi _{\alpha _1\alpha _2\alpha _3\alpha _4\alpha _5}^{(5)}(x)=\psi _{\alpha _11234}^{(5)}(x)`$ (18)
and
$`\psi _{\alpha _2}^{(\nu _s)}(x)`$ $`=`$ $`\psi _{\alpha _2}^{(1)}(x),`$
$`\psi _{\alpha _2}^{(\nu _s^{})}(x)`$ $`=`$ $`{\displaystyle \frac{1}{6}}\left(C^1\gamma _5\right)_{\alpha _2\alpha _3}\epsilon _{\alpha _3\alpha _4\alpha _5\alpha _6}\psi _{\alpha _4\alpha _5\alpha _6}^{(3)}(x)=\{\begin{array}{ccc}\hfill \psi _{134}^{(3)}(x)& \mathrm{for}\hfill & \alpha _2=1\hfill \\ \hfill \psi _{234}^{(3)}(x)& \mathrm{for}\hfill & \alpha _2=2\hfill \\ \hfill \psi _{312}^{(3)}(x)& \mathrm{for}\hfill & \alpha _2=3\hfill \\ \hfill \psi _{412}^{(3)}(x)& \mathrm{for}\hfill & \alpha _2=4\hfill \end{array},`$ (23)
respectively, where $`\psi _{\alpha _1\alpha _2\mathrm{}\alpha _N}^{(N)}(x)`$ for active fermions \[Eq. (18)\] carries also the Standard Model (composite) label, suppressed in our notation, and $`C`$ denotes the usual $`4\times 4`$ charge–conjugation matrix. We can see that due to the full antisymmetry in $`\alpha _i`$ indices for $`i2`$ these wave functions or fields appear (up to the sign) with the multiplicities 1, 4, 24 and 1, 6 , respectively. Thus, for active fermions and sterile neutrinos there is given the weighting matrix
$$\rho ^{(a)\mathrm{\hspace{0.17em}1}/2}=\frac{1}{\sqrt{29}}\left(\begin{array}{ccc}1& 0& 0\\ 0& \sqrt{4}& 0\\ 0& 0& \sqrt{24}\end{array}\right)$$
(24)
and
$$\rho ^{(s)\mathrm{\hspace{0.17em}1}/2}=\frac{1}{\sqrt{7}}\left(\begin{array}{cc}1& 0\\ 0& \sqrt{6}\end{array}\right),$$
(25)
respectively. For all neutrinos (i.e., $`\nu _e,\nu _\mu ,\nu _\tau `$ and $`\nu _s,\nu _s^{}`$) described jointly, the overall weighting matrix takes the form
$$\rho ^{(a+s)\mathrm{\hspace{0.17em}1}/2}=\frac{1}{\sqrt{36}}\left(\begin{array}{ccccc}1& 0& 0& 0& 0\\ 0& \sqrt{6}& 0& 0& 0\\ 0& 0& 1& 0& 0\\ 0& 0& 0& \sqrt{4}& 0\\ 0& 0& 0& 0& \sqrt{24}\end{array}\right),$$
(26)
if we use the ordering $`s,s^{},e,\mu ,\tau `$. Of course, for all these matrices Tr $`\rho =1`$.
Concluding this Introduction, we would like to say that in our approach to families of fundamental particles Dirac bispinor indices (”algebraic partons”) play the role of building blocks of composite states identified as fundamental particles. Any fundamental particle, active with respect to the Standard Model gauge interactions, contains one ”active algebraic parton” (coupled to the Standard Model gauge bosons) and a number $`N1`$ of ”sterile algebraic partons” (decoupled from these bosons). Due to the intrinsic Pauli principle obeyed by ”sterile algebraic partons”, the number $`N`$ of all ”algebraic partons” within a fundamental particle is restricted by the condition $`N14`$, so that only $`N=1,2,3,4,5`$ are allowed. It turns out that states with $`N=1,3,5`$ carry total spin 1/2 and are identified with three families of leptons and quarks, while states with $`N=2,4`$ get total spin 0 and so far are not identified. Any fundamental particle, sterile with respect to the Standard Model gauge interactions, contains only a number $`N4`$ of ”sterile algebraic partons”, thus only $`N=1,2,3,4`$ are allowed. States with $`N=1,3`$ correspond to total spin 1/2 and have to be identified as two sterile neutrinos, while states with $`N=2,4`$ have total spin 0 and are still to be identified.
Our algebraic construction may be interpreted either as ingeneously algebraic (much like the famous Dirac’s algebraic discovery of spin 1/2) or as the summit of an iceberg of really composite states of $`N`$ spatial partons with spin 1/2 whose Dirac bispinor indices manifest themselves as our ”algebraic partons”. In the former algebraic option, we avoid automatically the irksome existence problem of new interactions necessary to bind spatial partons within leptons and quarks of the second and third families. For the latter spatial option see some remarks in the second Ref. .
2. A model of five–neutrino texture
In this Section we construct the five–neutrino mass matrix $`M=\left(M_{\alpha \beta }\right)`$ under the conjecture that in ordering $`\alpha ,\beta =s,s^{},e,\mu ,\tau `$ of neutrino sequence $`\nu _\alpha =\nu _s,\nu _s^{},\nu _e,\nu _\mu ,\nu _\tau `$ only the nearest neighbours are coupled through the matrix $`M`$. In terms of our presentation of three families of active fermions, Eqs. (18), and two sterile neutrinos, Eqs. (19), such an ordering of five–neutrino sequence tells us that the chain
$$\nu _s\nu _s^{}\nu _e\nu _\mu \nu _\tau $$
of neutrino transitions corresponds to the chain
$$\alpha _2\alpha _2\alpha _3\alpha _4\alpha _1\alpha _1\alpha _3\alpha _4\alpha _1\alpha _2\alpha _3\alpha _4\alpha _5$$
of consecutive acts of creation or annihilation of index pairs $`\alpha _i\alpha _j`$ with $`i,j2`$ (pairs of ”sterile algebraic partons”), allowed by the intrinsic Pauli principle valid for $`\alpha _i`$ with $`i2`$. In one of these four acts, $`\alpha _1`$ must additionally be interchanged with $`\alpha _2`$, what should diminish the rate of $`\alpha _1\alpha _2\alpha _3\alpha _4`$ versus $`\alpha _1\alpha _1\alpha _3\alpha _4`$ (i.e., the magnitude of $`M_{s^{}e}`$ versus $`M_{e\mu }`$). One may also argue that the rate of $`\alpha _2\alpha _2\alpha _3\alpha _4`$ should be still more diminished versus $`\alpha _1\alpha _1\alpha _3\alpha _4`$,(i.e., $`M_{ss^{}}`$ versus $`M_{e\mu }`$), as being caused by two such additional interchanges of $`\alpha _1`$ with $`\alpha _2`$. The allowed alternative chain
$$\alpha _1\alpha _2\alpha _3\alpha _4\alpha _2\alpha _1\alpha _3\alpha _4\alpha _1\alpha _2\alpha _3\alpha _4\alpha _5$$
corresponding to
$$\nu _e\nu _s^{}\nu _s\nu _\mu \nu _\tau $$
does not contain the natural link $`\alpha _1\alpha _1\alpha _2\alpha _3`$ related to $`\nu _e\nu _\mu `$.
Thus, under the extra assumption that $`M_{ss}=0,M_{s^{}s^{}}=0`$ and $`M_{ee}=0`$, we can write
$$M=\left(\begin{array}{ccccc}0& M_{ss^{}}& 0& 0& 0\\ M_{s^{}s}& 0& M_{s^{}e}& 0& 0\\ 0& M_{es^{}}& 0& M_{e\mu }& 0\\ 0& 0& M_{\mu e}& M_{\mu \mu }& M_{\mu \tau }\\ 0& 0& 0& M_{\tau \mu }& M_{\tau \tau }\end{array}\right),$$
(27)
where $`M_{\beta \alpha }=M_{\alpha \beta }^{}`$ due to the hermicity of $`M`$. When the CP violation may be ignored in neutrino oscillations, we put $`M_{\alpha \beta }^{}=M_{\alpha \beta }`$ (and $`M_{\alpha \beta }>0`$ ).
Operating with the mass matrix (23), we will make the tentative assumption that, in comparison with its nonzero off–diagonal entries, its nonzero diagonal elements are small enough to be treated as a perturbation of the former. Such a property of $`M`$ may be related to a tiny neutrino mass scale involved in its diagonal elements. Then, in the zero perturbative order (where $`M_{\mu \mu }`$, $`M_{\tau \tau }`$ are put zero), the matrix (23) can be diagonalized exactly, giving the following zero–order neutrino masses $`\stackrel{o}{m}_i`$ numerated by $`i=4,5,1,2,3`$:
$$\stackrel{o}{m}_4=0,\stackrel{o}{m}_{5,1}=(A\sqrt{B^2})^{1/2},\stackrel{o}{m}_{2,3}=(A+\sqrt{B^2})^{1/2},$$
(28)
where
$`2A`$ $`=`$ $`|M_{e\mu }|^2+|M_{\mu \tau }|^2+|M_{ss^{}}|^2+|M_{s^{}e}|^2,`$
$`4B^2`$ $`=`$ $`\left(|M_{e\mu }|^2+|M_{\mu \tau }|^2|M_{ss^{}}|^2|M_{s^{}e}|^2\right)^2+4|M_{\mu \tau }|^2|M_{s^{}e}|^2.`$ (29)
Next, in the first perturbative order with respect to the ratios
$$\xi M_{\tau \tau }/|M_{e\mu }|,\chi M_{\mu \mu }/|M_{e\mu }|$$
(30)
we obtain $`m_i=\stackrel{o}{m}_i+\delta m_i`$, where
$$\delta m_i=(C_i/D_i)|M_{e\mu }|$$
(31)
with
$`C_i`$ $`=`$ $`(\xi +\chi )\stackrel{o}{m}_i^4\left[\xi \left(|M_{e\mu }|^2+|M_{ss^{}}|^2+|M_{s^{}e}|^2\right)+\chi \left(|M_{ss^{}}|^2+|M_{s^{}e}|^2\right)\right]\stackrel{o}{m}_i^2`$
$`+\xi |M_{e\mu }|^2|M_{ss^{}}|^2,`$
$`D_i`$ $`=`$ $`5\stackrel{o}{m}_i^43\left(|M_{e\mu }|^2+|M_{\mu \tau }|^2+|M_{ss^{}}|^2+|M_{s^{}e}|^2\right)\stackrel{o}{m}_i^2+|M_{\mu \tau }|^2|M_{s^{}e}|^2.`$ (32)
Note that the minus sign possible at $`m_5`$ and certain at $`m_2`$ is irrelevant since $`\nu _5`$ and $`\nu _2`$ are relativistic particles for which only $`m_5^2`$ and $`m_2^2`$ have physical meaning.
If our argument outlined in the first paragraph of this Section works, the mass matrix elements $`M_{ss^{}}`$ and $`M_{s^{}e}`$ (which couple the sterile neutrinos $`\nu _s,\nu _s^{}`$ among themselves and $`\nu _s^{}`$ with the active $`\nu _e`$, respectively) should be smaller than the elements $`M_{e\mu }`$ and $`M_{\mu \tau }`$ (coupling the active neutrinos $`\nu _e,\nu _\mu ,\nu _\tau `$), and also the element $`M_{ss^{}}`$ should be smaller than $`M_{s^{}e}`$: $`|M_{ss^{}}|<|M_{s^{}e}|<|M_{e\mu }|`$. Assuming tentatively $`|M_{ss^{}}||M_{e\mu }|`$ and $`|M_{s^{}e}||M_{e\mu }|`$, it can be seen that in the lowest approximation in the ratios
$$\lambda |M_{s^{}e}|/|M_{e\mu }|,\kappa |M_{ss^{}}|/|M_{e\mu }|$$
(33)
the formulae (24) and (27) give
$$\stackrel{o}{m}_4=0,\stackrel{o}{m}_{5,1}=(\lambda ^2c^2+\kappa ^2)^{1/2}|M_{e\mu }|,\stackrel{o}{m}_{2,3}=\frac{1}{s}|M_{e\mu }|$$
(34)
and
$$\delta m_4=\xi \frac{\kappa ^2s^2}{\lambda ^2c^2+\kappa ^2}|M_{e\mu }|,\delta m_{5,1}=\frac{1}{2}\xi \frac{\lambda ^2c^2s^2}{\lambda ^2c^2+\kappa ^2}|M_{e\mu }|,\delta m_{2,3}=\frac{1}{2}(\xi c^2+\chi )|M_{e\mu }|,$$
(35)
respectively, where the abbreviations
$$s\frac{|M_{e\mu }|}{\left(|M_{e\mu }|^2+|M_{\mu \tau }|^2\right)^{1/2}},c\frac{|M_{\mu \tau }|}{\left(|M_{e\mu }|^2+|M_{\mu \tau }|^2\right)^{1/2}}=(1s^2)^{1/2}$$
(36)
are used. Note that $`_i\delta m_i=M_{\mu \mu }+M_{\tau \tau }`$, as it should be because of $`_i\stackrel{o}{m}_i=0`$ and $`M_{ss}=M_{s^{}s^{}}=M_{ee}=0`$. For the masses $`m_i=\stackrel{o}{m}_i+\delta m_i`$, the formulae (30) and (31) show that $`m_5^2\stackrel{<}{}m_1^2m_2^2\stackrel{<}{}m_3^2`$ and $`m_4^2m_2^2`$.
Now, we can calculate the unitary matrix $`U=\left(U_{\alpha i}\right)`$ diagonalizing the mass matrix $`M=\left(M_{\alpha \beta }\right)`$ given in Eq. (23): $`U^{}MU=\mathrm{diag}(m_4,m_5,m_1,m_2,m_3)`$. In the zero perturbative order with respect to $`\xi ,\chi `$ and in the lowest approximation in $`\lambda ,\kappa `$, the result is
$$U=\left(\begin{array}{ccccc}f& \frac{\kappa }{\lambda c\sqrt{2}}f& \frac{\kappa }{\lambda c\sqrt{2}}f& 0& 0\\ 0& \frac{1}{\sqrt{2}}& \frac{1}{\sqrt{2}}& \frac{\lambda s^2}{\sqrt{2}}& \frac{\lambda s^2}{\sqrt{2}}\\ \frac{\kappa }{\lambda }f& \frac{c}{\sqrt{2}}f& \frac{c}{\sqrt{2}}f& \frac{s}{\sqrt{2}}& \frac{s}{\sqrt{2}}\\ 0& \frac{\lambda s^2}{\sqrt{2}}& \frac{\lambda ^2}{\sqrt{2}}& \frac{1}{\sqrt{2}}& \frac{1}{\sqrt{2}}\\ \frac{\kappa s}{\lambda c}f& \frac{s}{\sqrt{2}}f& \frac{s}{\sqrt{2}}f& \frac{c}{\sqrt{2}}& \frac{c}{\sqrt{2}}\end{array}\right),$$
(37)
where
$$f=\left(1+\frac{\kappa ^2}{\lambda ^2c^2}\right)^{1/2},\frac{\kappa }{\lambda c}f=\kappa \left(\lambda ^2c^2+\kappa ^2\right)^{1/2}$$
(38)
assuming that $`\lambda c0`$. In Eq. (33), a possible small effect of CP violation in neutrino oscillations is ignored by taking $`M_{\alpha \beta }=|M_{\alpha \beta }|`$.
If not only $`M_{ss^{}}M_{e\mu }`$ and $`M_{s^{}e}M_{e\mu }`$ but tentatively also $`M_{ss^{}}M_{s^{}e}`$, then beside $`\kappa 1`$ and $`\lambda 1`$ also $`\kappa \lambda `$, and so, we can put $`\kappa =0`$ and $`f=1`$. As is seen from Eq. (33), in this case the sterile neutrino $`\nu _s`$ is decoupled from $`\nu _s^{},\nu _e,\nu _\mu ,\nu _\tau `$ and, therefore, our five–neutrino texture is effectively reduced to a four–neutrino texture, where the masses $`m_i=\stackrel{o}{m}_i+\delta m_i`$ given in Eqs. (30) and (31) become
$$m_4=0,m_{5,1}=\left(\lambda c+\frac{1}{2}\xi s^2\right)M_{e\mu },m_{2,3}=\left[\frac{1}{s}+\frac{1}{2}\left(\xi c^2+\chi \right)\right]M_{e\mu }.$$
(39)
Here, $`m_4^2\stackrel{<}{}m_5^2\stackrel{<}{}m_1^2m_2^2\stackrel{<}{}m_3^2`$.
When the effect of mixing charged leptons $`e^{},\mu ^{},\tau ^{}`$ does not appear or may be ignored in the original lagrangian, then $`V=U^{}`$ is the five–neutrino extension of the lepton counterpart of the familiar Cabibbo–Kobayashi–Maskawa mixing matrix for quarks. In such a situation, the flavor neutrinos $`\nu _\alpha `$ and their states $`|\nu _\alpha `$ can be expressed as
$$\nu _\alpha =\underset{i}{}U_{\alpha i}\nu _i,|\nu _\alpha =\nu _\alpha ^{}|0=\underset{i}{}U_{\alpha i}^{}|\nu _i,$$
(40)
where $`\nu _i`$ and $`|\nu _i`$ are massive neutrinos and their states, numerated by $`i=4,5,1,2,3`$. Then, in the case of $`\kappa ^2\lambda ^2c^2`$, for instance, we obtain from Eqs. (33) the following simple mixing of massive neutrinos:
$`\nu _s`$ $`=`$ $`\nu _4{\displaystyle \frac{\kappa }{\lambda c}}{\displaystyle \frac{\nu _5\nu _1}{\sqrt{2}}},`$
$`\nu _s^{}`$ $`=`$ $`{\displaystyle \frac{\nu _5+\nu _1}{\sqrt{2}}}+\lambda s^2{\displaystyle \frac{\nu _2+\nu _3}{\sqrt{2}}},`$
$`\nu _e`$ $`=`$ $`c\left({\displaystyle \frac{\nu _5\nu _1}{\sqrt{2}}}+{\displaystyle \frac{\kappa }{\lambda c}}\nu _4\right)s{\displaystyle \frac{\nu _2\nu _3}{\sqrt{2}}},`$
$`\nu _\mu `$ $`=`$ $`{\displaystyle \frac{\nu _2+\nu _3}{\sqrt{2}}}\lambda s^2{\displaystyle \frac{\nu _5+\nu _1}{\sqrt{2}}},`$
$`\nu _\tau `$ $`=`$ $`s\left({\displaystyle \frac{\nu _5\nu _1}{\sqrt{2}}}+{\displaystyle \frac{\kappa }{\lambda c}}\nu _4\right)c{\displaystyle \frac{\nu _2\nu _3}{\sqrt{2}}}.`$ (41)
Here, we have $`f=1`$ up to $`O(\kappa ^2/\lambda ^2c^2)`$. Obviously, the assumption $`\kappa ^2\lambda ^2c^2`$ leading to $`\kappa ^2=0`$ is weaker than $`\kappa \lambda c`$ implying $`\kappa =0`$: in the former case the sterile neutrino $`\nu _s`$ is still coupled to $`\nu _s^{},\nu _e,\nu _\mu ,\nu _\tau `$, although by a small coefficient $`=O(\kappa /\lambda c)`$.
3. Five–neutrino oscillations
Finally, in this Section we can evaluate five–neutrino oscillation probabilities making use of the formulae
$$P(\nu _\alpha \nu _\beta )=|\nu _\beta |e^{iPL}|\nu _\alpha |^2=\delta _{\alpha \beta }4\underset{j>i}{}U_{\beta j}^{}U_{\alpha j}U_{\beta i}U_{\alpha i}^{}\mathrm{sin}^2x_{ji},$$
(42)
valid when the CP violation may be ignored (then $`U_{\alpha i}^{}=U_{\alpha i}`$). Here,
$$x_{ji}=1.27\frac{\mathrm{\Delta }m_{ji}^2L}{E},\mathrm{\Delta }m_{ji}^2=m_j^2m_i^2$$
(43)
with $`\mathrm{\Delta }m_{ji}^2`$, $`L`$ and $`E`$ expressed in eV<sup>2</sup>, km and GeV, respectively, while $`p_i=\sqrt{E^2m_i^2}Em_i^2/2E`$ are eigenvalues of neutrino momentum operator $`P`$ and $`L`$ denotes the experimental baseline.
In the case of our mixing matrix (33), valid in the zero perturbative order with respect to $`\xi ,\chi `$ and in the lowest approximation in $`\lambda ,\kappa `$, the formulae (38) in the case of $`\kappa ^2\lambda ^2c^2`$ give
$`P(\nu _e\nu _e)`$ $`=`$ $`1c^4\mathrm{sin}^2x_{15}(sc)^2\left(\mathrm{sin}^2x_{21}+\mathrm{sin}^2x_{31}+\mathrm{sin}^2x_{25}+\mathrm{sin}^2x_{35}\right)s^4\mathrm{sin}^2x_{32}`$
$``$ $`1c^4\mathrm{sin}^2x_{15}(2sc)^2\mathrm{sin}^2x_{21}s^4\mathrm{sin}^2x_{32},`$
$`P(\nu _\mu \nu _\mu )`$ $`=`$ $`1\mathrm{sin}^2x_{32},`$
$`P(\nu _\mu \nu _e)`$ $`=`$ $`s^2\mathrm{sin}^2x_{32}\lambda cs^3\left(\mathrm{sin}^2x_{21}\mathrm{sin}^2x_{31}\mathrm{sin}^2x_{25}+\mathrm{sin}^2x_{35}\right)`$ (44)
$``$ $`s^2\mathrm{sin}^2x_{32}.`$
Here, due to Eqs. (38) and (34) we have $`\mathrm{\Delta }m_{21}^2\mathrm{\Delta }m_{31}^2\mathrm{\Delta }m_{25}^2\mathrm{\Delta }m_{35}^2`$ and
$$\mathrm{\Delta }m_{15}^2=2\xi \lambda s^2cM_{e\mu }^2,\mathrm{\Delta }m_{32}^2=2\frac{\xi c^2+\chi }{s}M_{e\mu }^2,\mathrm{\Delta }m_{21}^2=\frac{1}{s^2}M_{e\mu }^2.$$
(45)
When $`1.27\mathrm{\Delta }m_{32}^2L_{\mathrm{atm}}/E_{\mathrm{atm}}=O(1)`$ for atmospheric $`\nu _\mu `$’s and thus $`\mathrm{\Delta }m_{32}^2=\mathrm{\Delta }m_{\mathrm{atm}}^23.5\times 10^3\mathrm{eV}^2`$ , the second formula (40) is able to describe atmospheric neutrino oscillations (dominated in our case by the mode $`\nu _\mu \nu _\tau `$) with maximal amplitude 1. Thus, the second equation (40) leads to the estimation
$$2\frac{\xi c^2+\chi }{s}M_{e\mu }^23.5\times 10^3\mathrm{eV}^2.$$
(46)
Hence, $`\xi +\chi 0`$ with $`c1`$ for $`M_{e\mu }`$ fixed. Also $`\xi `$ and $`\chi 0`$, since $`\xi `$ and $`\chi 0`$.
On the other hand, when $`\mathrm{\Delta }m_{15}^2L_{\mathrm{sol}}/E_{\mathrm{sol}}=O(1)`$ for solar $`\nu _e`$’s and so, $`\mathrm{\Delta }m_{15}^2=\mathrm{\Delta }m_{\mathrm{sol}}^2(6.5\times 10^{11}`$ or $`4.4\times 10^{10})\mathrm{eV}^2`$ (when considering the ”small” or ”large” vacuum solution), then the first formula (40) has a chance to describe solar neutrino oscillations (dominated now by the mode $`\nu _e\nu _s^{}`$) with the large amplitude $`c^4=\mathrm{sin}^22\theta _{\mathrm{sol}}0.72`$ or 0.90, respectively. In fact, due to $`\mathrm{\Delta }m_{15}^2\mathrm{\Delta }m_{32}^2\mathrm{\Delta }m_{21}^2`$ the first formula (40) becomes
$$P(\nu _e\nu _e)1c^4\mathrm{sin}^2x_{15}\left(2s^2c^2+s^4/2\right),$$
(47)
where the disturbing last term, $`2s^2c^2+s^4/20.27`$ or 0.099, may be too large, but it tends quickly to zero with $`c^41`$. Thus, from the first equation (41) the estimate
$$2\xi \lambda s^2cM_{e\mu }^2\left(6.5\times 10^{11}\mathrm{or}\mathrm{\hspace{0.33em}\hspace{0.17em}4.4}\times 10^{10}\right)\mathrm{eV}^2$$
(48)
is suggested.
Since $`c^20.85`$ or 0.95 and $`c0.92`$ or 0.97, Eqs. (42) and (44) in the case of $`\xi \chi `$ (i.e., $`M_{\tau \tau }M_{\mu \mu }`$ ) give the estimation
$$\xi M_{e\mu }^2\left(8.0\mathrm{or}\mathrm{\hspace{0.33em}\hspace{0.17em}4.2}\right)\times 10^4\mathrm{eV}^2,\lambda 2.9\times 10^7\mathrm{or}\mathrm{\hspace{0.33em}\hspace{0.17em}1.1}\times 10^5.$$
(49)
Such a tiny value $`\lambda `$ shows that $`M_{s^{}e}`$ (coupling $`\nu _s^{}`$ with $`\nu _e`$) is really very small versus $`M_{e\mu }:M_{s^{}e}=\lambda M_{e\mu }`$. In this case, we get from Eqs. (35)
$`m_{5,1}:m_{2,3}\lambda sc{\displaystyle \frac{1}{2}}\xi s^3(1.0\times 10^7`$ $`\mathrm{or}`$ $`2.3\times 10^6)2.4\times 10^6(\mathrm{eV}/M_{e\mu })^2,`$
$`m_{2,3}={\displaystyle \frac{1}{s}}M_{e\mu }`$ $``$ $`(2.6\mathrm{or}\mathrm{\hspace{0.33em}4.4})M_{e\mu }.`$ (50)
Thus, in order to obtain, for instance, $`|m_2|m_3(1\mathrm{to}\mathrm{\hspace{0.33em}10})`$ eV one should take $`M_{e\mu }(0.38\mathrm{to}\mathrm{\hspace{0.33em}3.8})`$ or $`(0.23\mathrm{to}\mathrm{\hspace{0.33em}2.3})`$ eV. Then, from the first Eq. (44) we infer that
$$\xi 5.6\times \left(10^3\mathrm{to}\mathrm{\hspace{0.33em}10}^5\right)\mathrm{or}\mathrm{\hspace{0.33em}7.9}\times \left(10^3\mathrm{to}\mathrm{\hspace{0.33em}10}^5\right)$$
(51)
for $`c^40.72`$ or 0.90, respectively. However, when $`M_{e\mu }`$ is kept fixed, $`\xi `$ tends quickly to zero with $`c^41`$.
In the case of Chooz experiment searching for oscillations of reactor $`\overline{\nu }_e`$’s , where it happens that $`1.27\mathrm{\Delta }m_{\mathrm{atm}}^2L_{\mathrm{Chooz}}/E_{\mathrm{Chooz}}=O(1)`$, the first formula (40) with $`\mathrm{\Delta }m_{32}^2=\mathrm{\Delta }m_{\mathrm{atm}}^2`$ and $`\mathrm{\Delta }m_{15}^2=\mathrm{\Delta }m_{\mathrm{sol}}^2`$ becomes
$$P(\overline{\nu }_e\overline{\nu }_e)1s^4\mathrm{sin}^2x_{32}2s^2c^21\left(2s^2c^2+s^4\right),$$
(52)
since $`\mathrm{\Delta }m_{15}^2\mathrm{\Delta }m_{32}^2\mathrm{\Delta }m_{21}^2`$. This is consistent with the negative result $`P(\overline{\nu }_e\overline{\nu }_e)=1`$ of Chooz experiment up to 28% or 10%, but this deviation from 1 tends quickly to zero with $`c^41`$. Note that $`U_{e3}=s/\sqrt{2}0.28`$ or 0.16, respectively.
The third formula (40) may imply the existence of $`\nu _\mu \nu _e`$ oscillations with the amplitude equal to $`s^20.15`$ or 0.05 and the mass–square scale given by $`\mathrm{\Delta }m_{32}^2=\mathrm{\Delta }m_{\mathrm{atm}}^23.5\times 10^3\mathrm{eV}^2`$, while the estimate from LSND experiment is, say, $`\mathrm{sin}^22\theta _{\mathrm{LSND}}0.02`$ and $`\mathrm{\Delta }m_{\mathrm{LSND}}^20.5\mathrm{eV}^2`$. Thus, our four–neutrino texture, if fitted to atmospheric and solar results, cannot explain the LSND observation. In order to include the LSND effect, one might depart from our conjecture on nearest–neighbour coupling in the four– or five–neutrino mass matrix.
4. Final remarks: a specific proposal for mass matrix elements
In a specific model of three–neutrino texture discussed by the author previously (e.g. and the first Ref. ), the following nonzero mass matrix elements were proposed:
$$M_{\mu \mu }=\frac{480}{9}\frac{\mu }{29},M_{\tau \tau }=\frac{24624}{25}\frac{\mu }{29},M_{e\mu }=\frac{2g}{29},M_{\mu \tau }=\frac{8\sqrt{3}g}{29},$$
(53)
where $`\mu `$ and $`g`$ stood for two small mass scales. Then,
$$M_{\tau \tau }=16.848M_{\mu \mu },M_{\mu \tau }=\sqrt{48}M_{e\mu }$$
(54)
and from Eqs. (26)
$$\xi =299.52\mu /g,\chi =17.778\mu /g=\xi /16.848.$$
(55)
Thus, $`\xi 1`$ if and only if $`g299.52\mu `$, the latter inequality implying $`g\mu `$ certainly. In the zero perturbative order with respect to $`\xi `$ or $`\mu /g`$ we put $`\mu =0`$.
When accepting the values (49) for $`M_{e\mu }`$ and $`M_{\mu \tau }`$, we obtain from Eqs. (32)
$$s=1/7=0.14286,c=\sqrt{48}/7=0.98974.$$
(56)
So, the estimation (42) provided by atmospheric neutrino experiments implies
$$\xi M_{e\mu }^22.4\times 10^4\mathrm{eV}^2.$$
(57)
Then, due to Eq. (43) and the first Eq. (41), the mass–square difference and oscillation amplitude for solar neutrinos should be
$$\mathrm{\Delta }m_{\mathrm{sol}}^2=2\xi \lambda s^2cM_{e\mu }9.7\times 10^6\lambda \mathrm{eV}^2,\mathrm{sin}^2\theta _{\mathrm{sol}}=c^4=0.95960,$$
(58)
respectively, while the disturbing last term would become smaller than before, giving now $`2s^2c^2+s^4/20.040`$.
The values (49) proposed for elements of the three–neutrino mass matrix $`M^{(a)}=\left(M_{\alpha \beta }\right)`$ $`(\alpha ,\beta =e,\mu ,\tau )`$ can be exactly deduced from the simple ansatz ( and the first Ref. ):
$$M^{(a)}=\rho ^{(a)1/2}\left[\mu (N^2N^2)+g(a+a^{})\right]\rho ^{(a)1/2}.$$
(59)
Here,
$$N=1+2a^{}a=\left(\begin{array}{ccc}1& 0& 0\\ 0& 3& 0\\ 0& 0& 5\end{array}\right)$$
(60)
is the matrix of number of all Dirac bispinor indices $`\alpha _i`$ (all ”algebraic partons”) used in Eqs. (18) to present active neutrinos $`\nu _e,\nu _\mu ,\nu _\tau `$, while
$$a=\left(\begin{array}{ccc}0& 1& 0\\ 0& 0& \sqrt{2}\\ 0& 0& 0\end{array}\right),a^{}=\left(\begin{array}{ccc}0& 0& 0\\ 1& 0& 0\\ 0& \sqrt{2}& 0\end{array}\right)$$
(61)
are (truncated) annihilation and creation matrices of index pairs $`\alpha _i\alpha _j`$ with $`i,j2`$ (pairs of ”sterile algebraic partons”) included in Eqs. (18) for active neutrinos. The latter matrices satisfy, jointly with the matrix of number of such index pairs,
$$n=a^{}a=\left(\begin{array}{ccc}0& 0& 0\\ 0& 1& 0\\ 0& 0& 2\end{array}\right),$$
(62)
the familiar commutation relations
$$[a,n]=a,[a^{},n]=a^{},$$
(63)
and, in addition, the truncation relations $`a^3=0`$ and $`a^3=0`$ consistent with the intrinsic Pauli principle for Dirac bispinor indices $`\alpha _i`$ with $`i2`$ (obviously, neither boson nor fermion canonical commutation relations, $`[a,a^{}]_{}=1`$, are satisfied here). Finally, $`\rho ^{(a)1/2}`$ stands in Eq. (55) for the active–neutrino weighting matrix (20).
In the mass matrix (55), the first term containing $`\mu N^2`$ may be intuitively interpreted as an interaction of all $`N`$ ”algebraic partons” treated on equal footing, while the second involving $`\mu N^2`$, as a subtraction term caused by the fact that there is one ”active algebraic parton” distinguished (by its external coupling) among all $`N`$ ”algebraic partons” of which $`N1`$, as ”sterile” are undistinguishable. This distinguished ”algebraic parton” appears, therefore, with the probability $`[N!/(N1)!]^1=N^1`$ that, when squared, leads to an additional interaction involving $`\mu N^2`$. The latter interaction should be subtracted from the former in order to obtain for $`N=1`$ the matrix element $`M_{ee}`$ assumed to be zero. The third term in the mass matrix (55) containing $`g(a+a^{})`$ annihilates and creates pairs of ”sterile algebraic partons” and so, is responsible in a natural way for mixing of three active neutrinos.
Of course, the three–neutrino matrix $`M^{(a)}=\left(M_{\alpha \beta }\right)`$ $`(\alpha ,\beta =e,\mu ,\tau )`$ considered in this Section is a submatrix of our five–neutrino mass matrix $`M=\left(M_{\alpha \beta }\right)`$ $`(\alpha ,\beta =s,s^{},e,\mu ,\tau )`$, viz.
$$M=(\begin{array}{ccccc}0& M_{ss^{}}& 0& 0& 0\\ M_{s^{}s}& 0& 0& 0& 0\\ 0& 0& 0& 0& 0\\ 0& 0& 0& 0& 0\\ 0& 0& 0& 0& 0\end{array})+(\begin{array}{ccccc}0& 0& 0& 0& 0\\ 0& 0& M_{s^{}e}& 0& 0\\ 0& M_{es^{}}& 0& 0& 0\\ 0& 0& 0& 0& 0\\ 0& 0& 0& 0& 0\end{array})+(\begin{array}{ccccc}0& 0& 0& 0& 0\\ 0& 0& 0& 0& 0\\ 0& 0& 0& M_{e\mu }& 0\\ 0& 0& M_{\mu e}& M_{\mu \mu }& M_{\mu \tau }\\ 0& 0& 0& M_{\tau \mu }& M_{\tau \tau }\end{array}),$$
(64)
where $`M_{ss^{}}=\kappa M_{e\mu }`$ and $`M_{s^{}e}=\lambda M_{e\mu }`$. Thus, the $`2\times 2`$ matrix involved in the middle $`5\times 5`$ matrix plays the role of coupling between the sterile $`2\times 2`$ matrix $`M^{(s)}`$ and active $`3\times 3`$ matrix $`M^{(a)}`$. If $`\lambda `$ were zero, both sterile neutrinos $`\nu _s`$ and $`\nu _s^{}`$ would be decoupled from the three active. When $`\kappa `$ is put zero, $`\nu _s`$ becomes decoupled from $`\nu _s^{}`$ as well as from $`\nu _e,\nu _\mu ,\nu _\tau `$ ($`M^{(s)}`$ is then a zero matrix).
Originally, the ansatz (55) was introduced for mass matrix $`M^{(e)}=\left(M_{\alpha \beta }^{(e)}\right)`$ $`(\alpha ,\beta =e,\mu ,\tau )`$ of charged leptons $`e^{},\mu ^{},\tau ^{}`$. In this case, in order to get a small but nonzero value of $`M_{ee}^{(e)}`$, the quantity $`\mu (1\epsilon )N^2`$ with a small $`\epsilon `$ (rather than the quantity $`\mu N^2`$) was used in the second term of $`M^{(e)}`$.Then, the nonzero mass matrix elements were
$`M_{ee}^{(e)}=\epsilon {\displaystyle \frac{\mu }{29}},M_{\mu \mu }^{(e)}`$ $`=`$ $`{\displaystyle \frac{4(80+\epsilon )}{25}}{\displaystyle \frac{\mu }{29}},M_{\tau \tau }^{(e)}={\displaystyle \frac{24(624+\epsilon )}{25}}{\displaystyle \frac{\mu }{29}},`$
$`M_{e\mu }^{(e)}`$ $`=`$ $`{\displaystyle \frac{2g}{29}},M_{\mu \tau }^{(e)}={\displaystyle \frac{8\sqrt{3}g}{29}}.`$ (65)
Making the conjecture that for charged leptons diagonal elements of $`M^{(e)}`$ dominate over its off–diagonal entries (i.e., $`\xi 1`$ what is certainly true for $`g\mu `$), we calculated the masses $`m_e,m_\mu ,m_\tau `$ as eigenvalues of $`M^{(e)}`$ in the lowest (quadratic) perturbative order with respect to $`1/\xi `$ or $`g/\mu `$. Then, we expressed $`m_\tau `$, $`\mu `$ and $`\epsilon `$ in terms of $`m_e,m_\mu `$ and $`(g/\mu )^2`$, obtaining
$`m_\tau `$ $`=`$ $`\left[1776.80+10.12112(g/\mu )^2\right]\mathrm{MeV},`$
$`\mu `$ $`=`$ $`\left\{85.9924+O\left[(g/\mu )^2\right]\right\}\mathrm{MeV},`$
$`\epsilon `$ $`=`$ $`0.172329+O\left[(g/\mu )^2\right],`$ (66)
where the experimental values of $`M_e`$ and $`M_\mu `$ were taken as the only input. Comparing this prediction for $`m_\tau `$ with the experimental value $`m_\tau ^{\mathrm{exp}}=1777.05_{0.26}^{+0.29}`$ MeV , we got
$$(g/\mu )^2=0.024_{0.025}^{+0.028}$$
(67)
for charged leptons. In such a way, we achieved in the case of charged leptons a really good agreement of our ansatz for $`M^{(e)}`$ with the experimental mass spectrum (even in the zero perturbative order).
This result has motivated the application of our ansatz for $`M^{(e)}`$ also to the case of active neutrinos $`\nu _e,\nu _\mu ,\nu _\tau `$ (corresponding to $`e^{},\mu ^{},\tau ^{}`$). In their case, however, the inverse conjecture $`\xi 1`$ or $`g\mu `$ seems natural in view of experimentally suggested large neutrino mixing that is in contradistinction to small mixing of charged leptons \[cf. Eq. (63)\]. In terms of three active neutrinos alone this ansatz leads to maximal amplitude for atmospheric $`\nu _\mu \nu _\mu `$ oscillations, but it requires introducing at least one sterile neutrino to explain solar $`\nu _e\nu _e`$ oscillations ( and the present paper). Even in this case, the LSND effect does not appear, however. Thus, if this effect was confirmed in a clear manner, the conjecture on nearest–neighbour coupling in the four– or five–neutrino mass matrix and, in particular, our ansatz (55) would not be correct for neutrinos.
In this case, a different neutrino texture, also including one or more sterile neutrinos, would be needed. If, on the contrary, the LSND effect was not seen, our four– or five–neutrino texture might be realized in Nature. However, much more economical would be then a three-neutrino texture involving (as e.g. in Ref. ) active neutrinos $`\nu _e,\nu _\mu ,\nu _\tau `$ with the mass hierarchy $`m_1^2\stackrel{<}{}m_2^2m_3^2`$ (in place of $`m_1^2m_2^2\stackrel{<}{}m_3^2`$ valid in the present paper). They ought to be coupled in a different way than in the present paper in order to explain both the atmospheric and solar neutrino results. The argument for our texture would be the absence of LSND effect and, at the same time, the experimental existence of one or two sterile neutrinos (for a possible astrophysical aspect of sterile neutrinos cf. e.g. Ref. ).
References
1. Cf. e.g. C. Giunti, hep–ph/990336; C. Giunti, M.C. Gonzalez–Garcia and Peña–Garay, hep–ph/0001101; and references therein.
2. C. Athanassopoulos et al. (LSND Collaboration), Phys. Rev. Lett. 75, 2650 (1995); Phys. Rev. C 54, 2685 (1996); Phys. Rev. Lett. 77, 3082 (1998); 81, 1774 (1998).
3. Cf. e.g. J.N. Bahcall, P.I. Krastev and A.Y. Smirnov, Phys. Rev. D 58, 096016 (1998); hep–ph/9905220v2; Phys. Lett. B 477, 401 (2000); hep–ph/0002293.
4. Y. Fukuda et al. (Super–Kamiokande Collaboration), Phys. Rev. Lett. 81, 1562 (1998) \[Errata 81, 4279 (1998)\]; 82, 1810 (1999); 82, 2430 (1999).
5. W. Królikowski, hep–ph/0001023.
6. Cf. e.g. W. Królikowski, Nuovo Cim. A 122, 893 (1999); Acta Phys. Pol. B 30, 2631 (1999); B 31, 663 (2000); and references therein.
7. M. Gell–Mann, P. Ramond and R. Slansky, in Supergravity , ed. P. van Nieuwenhuizen and D.Z. Freedman, North–Holland, Amsterdam 1979; T. Yanagida, in Proc. of the Workshop on the Unified Theory of the Baryon Number in the Universe, ed. O. Sawada and S. Sugamoto, KEK report No. 79–18, Tsukuba (Japan) 1979; see also R. Mohapatra and G. Senjanovic, Phys. Rev. Lett. 44, 912 (1980).
8. D.O. Caldwell, hep–ph/0003250.
9. W. Królikowski, Acta Phys. Pol. B 30, 227 (1999); in Proc. of the 12th Max Born Symposium, Wrocław (Poland) 1998, eds. A. Borowiec et al., Springer 2000, also hep–ph/9808307.
10. W. Królikowski, Acta Phys. Pol. B 21, 871 (1990); Phys. Rev. D 45, 3222 (1992); in Proc. of the 2nd Max Born Symposium, Wrocław (Poland) 1992, eds. Z. Oziewicz et al., Kluwer 1993; Acta Phys. Pol. B 24, 1149 (1993).
11. T. Banks, Y. Dothan and D. Horn, Phys. Lett. B 117, 413 (1982).
12. E. Kähler, Rendiconti di Matematica 21, 425 (1962); see also D. Ivanenko and L. Landau, Z. Phys. 48, 341 (1928).
13. M. Appolonio et al. (Chooz Collaboration), Phys. Lett. B 420, 397 (1998)
14. Particle Data Group, Review of Particle Physics, Eur. Phys. J. C 3, 1 (1998).
15. P.H. Chankowski, W. Królikowski and S. Pokorski, Phys. Lett. B 473, 109 (2000).
|
warning/0004/gr-qc0004030.html
|
ar5iv
|
text
|
# A Note on Brane Inflation
## I Introduction
Recently, an alternative mechanism of solving the hierarchy problem has been proposed . The novel proposal exploits the fact that the fundamental Planck scale $`M`$ is a TeV scale but in $`4+d`$ dimensions, where $`d`$ is the number of extra dimensions which are to be compactified. The Planck scale in the observable world is generated because of the presence of large extra dimensions: $`M_{\mathrm{Pl}}^2=M^{2+d}V_d`$. Here $`V_d`$ is the volume of the extra compactified dimensions. The proposal also demands that the standard model fields are bound to live in the observable world and gravity is the only force which mediates through the bulk and the observable world. The introduction of very large size of the compactified dimensions brings a completely new set of exciting problems ranging from phenomenology of accelerator physics to cosmology .
However, Randall and Sundrum introduced another twist in the physics beyond $`4`$ dimensions to account for the hierarchy between the Planck scale and the electroweak scale. They demonstrate that in a background of special non-factorizable geometry an exponential warp factor appears for the Poincaré invariant 3+1 dimensions. The model consists of two $`3`$ branes situated at the fixed positions along the 5th dimension compactified on a $`S^1/Z_2`$ orbifold symmetry. The space-time in the bulk is 5 dimensional anti de-Sitter space (AdS). The five dimensional Einstein’s equations permit a solution which preserves 4 dimensional Poincaré invariance on the brane, with the metric taking the form
$`ds^2=e^{2k|z|}g_{\mu \nu }dx^\mu dx^\nu +dz^2.`$ (1)
Here $`z`$ is the extra spatial dimension and $`\mu ,\nu =0,..,3`$. The constant $`k`$ is determined by the bulk cosmological constant $`\mathrm{\Lambda }=3M_5^3k^2`$ where $`M_5`$ is the five dimensional Planck scale (see the 5 dimensional action defined later). The 4 dimensional Poincaré invariance also requires a fine-tuning of the brane tensions. Namely, the positive tension $`\sigma `$ of the brane at $`z=0`$ is related to $`\mathrm{\Lambda }`$ and $`M_5`$, while the brane at $`z=z_c`$ has equal and opposite tension $`\sigma `$. Gravity is localized due to the exponentially decaying warp factor in the 5th dimension. The hierarchy between the Planck scale and the electroweak scale is explained by the suppression factor $`e^{kz_c}`$ on the negative tension brane, where the standard model (SM) particles are assumed to be.
The excitations around the background metric includes a massless 4 dimensional graviton zero modes and a set of massive Kaluza-Klein modes which have masses proportional to $`1/z_c`$. In addition, there is a massless four-dimensional scalar associated with the relative motions of the two branes, i.e., the radion field . The radion field would become massive once the separation of the two branes are stabilized due to certain mechanism . It is interesting to note that the reduced action of the massless graviton zero mode and the radion field in the effective $`4`$ dimensions on either of the branes, is not exactly the same as general relativity (GR). Instead, the actions on the positive and the negative branes mimic that of a deviant theory of gravity, popularly known as scalar tensor theory, with the radion field taking the role of a Brans-Dicke scalar with non-trivial Brans-Dicke parameter . In this theory, Einstein’s gravity is recovered in a limiting case.
The purpose of this paper is to investigate the possibility of inflationary scenarios on the branes. We assume that the inflation is induced by a scalar field confined on the brane which has a slow-roll potential. In our discussion, we also assume that the inflation on the brane takes place before the stabilization of the extra dimensions , hence during the inflationary era, the radion field is massless and has no additional potential. We find that it is possible to have an inflationary era induced by a slow-rolling inflaton on the positive tension brane. In particular, during this era the separation between the two branes increases. Namely, the bulk gravity reacts to inflation on the Planck brane and pushes the other brane away and enlarges the size of the AdS space. We find that the change in the size of the extra dimension can be sufficiently small such that the perturbative description of our frame work is still valid towards the end of inflation. However, on the negative tension brane, we fail to find a consistent inflationary solution within our assumptions of massless radion and slow-rolling inflaton.
In Section II, we write down the effective actions on the branes. We present the equations of motions in the next section and the inflationary solutions on both the positive and the negative tension branes. In Section IV, we discuss our results.
## II Effective actions
The full action consists of $`5`$ dimensional Einstein’s gravity with a negative cosmological constant in the bulk. The two branes are located on the orbifold $`S^1/Z_2`$ along the extra dimension, with the positive tension brane at $`z=0`$ and the negative tension brane at $`z=z_c`$.
$`S=2{\displaystyle d^4x_0^{z_c}𝑑z\sqrt{g_5}\left[\frac{M_5^3}{2}R_52\mathrm{\Lambda }\right]}`$ (2)
$`\sigma _+{\displaystyle d^4x\sqrt{g_+}}\sigma _{}{\displaystyle d^4x\sqrt{g_{}}},`$ (3)
where $`M`$ is the $`5`$ dimensional Planck mass, $`\sigma _\pm `$ and $`g_\pm `$ are the brane tensions and the corresponding induced metric on the respective branes. The effective 4 dimensional Planck scale: $`M_{Pl}=M_5^3(1e^{2kz_c})/(4k)`$ . The linearized gravity around its background including the massless degree of freedom, i.e., the massless 4 dimensional graviton and radion field, can be parametrized by the following metric solution ,
$`ds^2`$ $`=`$ $`e^{2kh(x,z)}\overline{g}_{\mu \nu }dx^\mu dx^\nu +h_{,z}^2dz^2,`$ (4)
$`h(x,z)`$ $`=`$ $`z+f(x)e^{2kz}.`$ (5)
where $`0<z<z_c`$, and $`h_{,z}`$ denotes the derivative with respect to $`z`$. The radion field $`f(x)`$, which is a function of the brane coordinates only, was also introduced in . For the classical solution, the induced metric on the positive tension brane at $`z=0`$ takes the form,
$`g_{+\mu \nu }=e^{2kf}\overline{g}_{\mu \nu },`$ (6)
and on the negative tension brane at $`z=z_c`$:
$`g_{\mu \nu }=\alpha ^1e^{2\alpha kf}\overline{g}_{\mu \nu },`$ (7)
where $`\alpha e^{2kz_c}`$ is a constant. With this metric ansatz one can integrate out the fifth dimension to achieve the effective four dimensional actions on the two branes .
$`S_\pm ={\displaystyle }d^4x\sqrt{g_\pm }{\displaystyle \frac{1}{16\pi }}[\mathrm{\Phi }^\pm R{\displaystyle \frac{\omega _\pm (\mathrm{\Phi }^\pm )}{\mathrm{\Phi }^\pm }}(\mathrm{\Phi }^\pm )^2],`$ (8)
where, $`\pm `$ denotes the positive and negative tension branes respectively. The fields $`\mathrm{\Phi }^\pm `$ and the parameter $`\omega _\pm `$ (dimensionful) are defined as ,
$`\mathrm{\Phi }^\pm ={\displaystyle \frac{2}{kG_5}}e^{k[(\alpha 1)f+z_c]}\mathrm{sinh}(k[(\alpha 1)f+z_c]),`$ (9)
$`\omega _\pm (f)=\pm 3e^{\pm k[(\alpha 1)f+z_c]}\mathrm{sinh}(k[(\alpha 1)f+z_c])`$ (10)
$`\left(1{\displaystyle \frac{e^{\pm k[(\alpha 1)f]}\mathrm{sinh}[k(\alpha 1)f]}{\mathrm{sinh}^2(kz_c)}}\right),`$ (11)
where $`G_5`$ is defined as $`G_5(8\pi M_5^3)^1`$. The effective action (8) is the correct description of the dynamics of the massless fields as long as the perturbative limit $`e^{2k\alpha f}\alpha `$ is satisfied <sup>*</sup><sup>*</sup>*Here we want to emphasis that in the strict limit $`e^{2k\alpha f}\alpha `$, the effective action on the brane (8) is equivalent to the original action in five dimensions. It is possible that effective brane action deduced from dimensional reduction could differ in higher orders from that of the original 5-dimensional action .. It is pointed out in that this action is similar to that of the scalar tensor theories , with $`\mathrm{\Phi }^+`$ and $`\mathrm{\Phi }^{}`$ are the scalar Brans-Dicke fields, whose dynamical evolution determines the strength of gravity, namely, the Newton’s constant $`G_N1/\mathrm{\Phi }_\pm `$. The strength of the coupling between gravity and the scalar field $`\mathrm{\Phi }^\pm `$ is determined by $`\omega _\pm `$. When $`\omega _\pm \mathrm{}`$, the scalar tensor theory approaches Einstein’s general theory of relativity provided $`\omega _\pm ^{}/\omega _\pm ^30`$ . The present observations such as light bending, perihelion precession and radar echo delay phenomena in the solar system suggests $`\omega _\pm >3000`$ . It is obvious that in the perturbative limit $`\omega _+`$ on the positive tension brane $`\omega _+3\alpha e^{2k\alpha f}/2`$ can easily satisfy the experimental constraint, while on the negative tension brane $`\omega _{}3/2`$ does not satisfy this bound . However, we take the view that the situation could be altered if additional massive degrees of freedom in the bulk associated with stabilization are included. In the following section we study the dynamics of both branes in the presence of a scalar field inflaton with a slow-roll potential on the brane. We are mainly interested in the inflationary aspects of the solutions. In the mean time, it is interesting to see if brane inflation could lead to a more physical $`\omega _{}`$.
## III Inflationary Solutions
To induce inflation on the brane, we add the action of a scalar field on the brane. We conjecture that adding an additional scalar field on both the branes will not affect the background classical solution of the gravity and the dynamics can be safely described with the usual perturbative theory . At the end we shall see that inflation on the positive tension brane indeed satisfies our assumption, while inconsistency appears on the negative tension brane. The action of a scalar field $`\chi `$ on the brane takes the form,
$$S_{m,\pm }=d^4x\sqrt{g_\pm }[g_{\pm \mu \nu }^\mu \chi ^\nu \chi V(\chi )],$$
(12)
where $`g_{\pm \mu \nu }`$ are the induced metric on the branes as defined previously.
To discuss inflation on the brane, we choose $`\overline{g}_{\mu \nu }`$ to be the Friedmann Robertson Walker (FRW) metric, $`\overline{g}_{\mu \nu }=diag\{1,A(t),A(t),A(t)\}`$, where $`A(t)`$ is the scale factor. In order to derive the equations of motions for the fields, one can perform a coordinate transformation on the brane, $`d\tau =\sqrt{|g_{\pm 00}|}dt`$, such that the metric on the branes, Eqs.(67) can be brought to the ordinary FRW metric by defining the scale factors $`a_+(\tau )e^{kf(t)}A(t)`$ and $`a_{}(\tau )\alpha ^{1/2}e^{\alpha kf(t)}A(t)`$. We have explicitly assumed that the radion field $`f`$ has solely the time dependence. Since we are interested in studying the inflationary solution, we can appropriately assume that the late time dynamics of $`f`$ does not depend on the spatial coordinates of the branes. Under these assumptions the equations for the scalar fields are
$`H^2+H{\displaystyle \frac{\dot{\mathrm{\Phi }}^\pm }{\mathrm{\Phi }^\pm }}{\displaystyle \frac{\omega }{6}}{\displaystyle \frac{\dot{\mathrm{\Phi }}^{\pm 2}}{\mathrm{\Phi }^{\pm 2}}}`$ (13)
$`=`$ $`{\displaystyle \frac{8\pi }{3\mathrm{\Phi }^\pm }}({\displaystyle \frac{1}{2}}\dot{\chi }^2+V(\chi )),`$ (14)
$`\ddot{\mathrm{\Phi }}^\pm +\dot{\mathrm{\Phi }}^\pm \left[3H+{\displaystyle \frac{\dot{\omega }}{2\omega +3}}\right]`$ (15)
$`=`$ $`8\pi {\displaystyle \frac{4V(\chi )\dot{\chi }^2}{2\omega +3}},`$ (16)
$`\ddot{\chi }+3H\dot{\chi }+V^{}(\chi )`$ $`=`$ $`0,`$ (17)
where the Hubble parameter $`H\dot{a}_\pm /a_\pm `$, the overdot denotes $`d/d\tau `$ and the prime denotes $`d/d\chi `$. An inflationary epoch, in which the scale factors $`a_\pm `$ are accelerating, requires the scalar field $`\chi `$ to evolve slowly compared to the expansion of the Universe. Thus, the following conditions of slow-rolling are required:
$$\ddot{\chi }H\dot{\chi },$$
(18)
$$\frac{1}{2}\dot{\chi }^2V(\chi ),$$
(19)
$$\ddot{\mathrm{\Phi }}^\pm H\dot{\mathrm{\Phi }}^\pm H^2\mathrm{\Phi }^\pm .$$
(20)
Under the slow-roll conditions, the Eqs.(1317) can be simplified. In the following subsections, we discuss the possible solutions on the positive and the negative tension branes respectively. For simplicity, we assume that $`V(\chi )=V_0`$ during the epoch when the slow-roll conditions Eqs.(1820) are satisfied. There are other forms of $`V(\chi )`$ that one can take to satisfy the slow-roll conditions , which we will not consider in this paper.
### A On the positive tension brane
On the positive tension brane, when $`e^{2k\alpha f}\alpha `$, $`\mathrm{\Phi }^+G_P(1e^{(2k\alpha f)}/\alpha )`$, and the Brans-Dicke parameter $`\omega _+`$ can be written as a function of $`\mathrm{\Phi }^+`$,
$`\omega _+(\mathrm{\Phi }^+){\displaystyle \frac{3}{2}}{\displaystyle \frac{\mathrm{\Phi }^+}{G_\mathrm{P}\mathrm{\Phi }^+}},`$ (21)
where $`G_P1/(kG_5)`$. In this limit, the value of $`\mathrm{\Phi }^+`$ approaches $`G_P`$. And $`\omega _+\frac{3}{2}\alpha 1`$. Following the slow-roll approximations Eqs.(1820), Eqs.(1315) can be expressed as:
$`3H^2(G_\mathrm{P}\mathrm{\Phi }^+)8\pi V_0(\mathrm{\Phi }^+)^1(G_\mathrm{P}\mathrm{\Phi }^+)`$ (22)
$`+{\displaystyle \frac{3}{4}}(\dot{\mathrm{\Phi }}^+)^2(\mathrm{\Phi }^+)^1,`$ (23)
$`3H\dot{\mathrm{\Phi }}^+(G_\mathrm{P}\mathrm{\Phi }^+)+{\displaystyle \frac{1}{2}}\dot{\mathrm{\Phi }}^{+2}(\mathrm{\Phi }^+)^1G_\mathrm{P}`$ (24)
$`{\displaystyle \frac{32}{3}}\pi V_0{\displaystyle \frac{(G_\mathrm{P}\mathrm{\Phi }^+)^2}{\mathrm{\Phi }^+}}.`$ (25)
By manipulating Eqs.(2224) we get:
$$\frac{(\dot{\mathrm{\Phi }}^+)^2}{(\mathrm{\Phi }^+)}\frac{G_\mathrm{P}}{G_\mathrm{P}\mathrm{\Phi }^+}6H\left[\frac{4}{3}H(G_\mathrm{P}\mathrm{\Phi }^+)\dot{\mathrm{\Phi }}^+\right],$$
(26)
where the slow-roll condition Eq.(20) has been used to neglect the term $`\frac{3}{4}(\dot{\mathrm{\Phi }}^+)^2(\mathrm{\Phi }^+)^1`$. The above equation can be solved approximately by:
$$\frac{\dot{\mathrm{\Phi }}^+}{G_\mathrm{P}\mathrm{\Phi }^+}\gamma H,$$
(27)
where $`\gamma =\sqrt{17}3`$, we have used Eq.(20) and the approximation that in the limit $`e^{2\alpha kf}\alpha `$, $`\mathrm{\Phi }^+G_\mathrm{P}`$. With Eq.(27) and the slow-roll condition, Eq.(22) can be reduced to
$$H\sqrt{\frac{8\pi V_0}{3\mathrm{\Phi }^+}}.$$
(28)
One can then solve for $`\mathrm{\Phi }^+`$ from Eq.(27),
$$\begin{array}{c}(\tau \tau _0)\sqrt{\frac{2\pi \gamma ^2}{3}\pi V_0}=\\ \left[\sqrt{\mathrm{\Phi }^+}+\frac{\sqrt{G_\mathrm{P}}}{2}\mathrm{ln}\frac{\sqrt{G_\mathrm{P}}+\sqrt{\mathrm{\Phi }^+}}{\sqrt{G_\mathrm{P}}\sqrt{\mathrm{\Phi }^+}}\right]_{\mathrm{\Phi }_0^+}^{\mathrm{\Phi }^+},\end{array}$$
(29)
where $`\mathrm{\Phi }_0^+`$ is the initial value of $`\mathrm{\Phi }^+`$. We take as an initial condition that at the beginning of the inflation $`f(\tau =\tau _0)=0`$, hence $`\mathrm{\Phi }_0^+=G_P(11/\alpha )`$. Solving Eq.(27) for the scale factor, we get:
$$\frac{a(\tau )}{a_0}\left[\frac{G_\mathrm{P}\mathrm{\Phi }_0}{G_P\mathrm{\Phi }(t)}\right]^{1/\gamma }.$$
(30)
Substituting Eq.(30) in Eq.(29) and assuming $`a(\tau )/a_01`$ at $`\tau \tau _0`$, we get the final expression
$$\frac{a(\tau )}{a_0}\mathrm{exp}\left(\sqrt{\frac{8\pi V_0}{3G_\mathrm{P}}}(\tau \tau _0)\right).$$
(31)
The above equation confirms the exponential growth in the scale factor during the inflationary era. The evolution of the Brans-Dicke field can be obtained by substituting Eq.(31) into Eq.(29).
$$\frac{\mathrm{\Phi }^+(\tau )}{G_P}1\frac{1}{\alpha }\mathrm{exp}\left(2\left[\sqrt{\frac{2\pi \gamma ^2V_0}{3G_P}(\tau \tau _0)}\right]\right),$$
(32)
where we have used the initial condition $`f(\tau =\tau _0)=0`$ and $`\mathrm{\Phi }_0^+=G_P(11/\alpha )`$. Eq.(32) shows that the effective Brans-Dicke field grows with time during the inflationary era and approaches its asymptotic value, $`(kG_5)^1`$ at $`\tau \mathrm{}`$. We can also estimate the final value of $`f(\tau _f)`$ in terms of the number of e-foldings $`N`$ of the inflation, by using the expression for $`\mathrm{\Phi }_+`$. From Eq.(31) and (32), we get:
$$f(\tau _f)=\frac{\gamma N}{2k\alpha },$$
(33)
where,
$`N\mathrm{ln}\left({\displaystyle \frac{a(\tau _f)}{a(\tau _0)}}\right).`$ (34)
If we assume that $`N60`$ e-foldings (to solve the flatness problem in the usual hot big bang Universe), it is not hard to realize that for sufficiently large $`\alpha `$, the condition $`e^{2k\alpha f(\tau _f)}e^{\gamma N}\alpha `$ could be satisfied such that during the complete process of inflation, the perturbative description of radion field remains valid. However, note that the size of $`\alpha `$ that is required here is larger than the required size of $`\alpha `$ to explain the hierarchy problem, i.e., $`e^{kz_c}10^{15}`$ is needed to produce TeV scale masses from the fundamental scale of $`M_{Pl}10^{19}`$ GeV . Since, we have assumed that the stabilization of the brane separation and the inflation are two independent processes taking place at different times, it is possible that the stabilization mechanism happens after the end of inflation and reduces the separation of the two branes. However, if $`z_c`$ at the onset of inflation does not satisfy: $`e^{\gamma N}\alpha `$, inflation would eventually destabilize the configuration of the two branes and the perturbative approach would break down. We also make no comments on how the inflation is stopped, however, we believe that the end of slow-roll conditions eventually leads to graceful exit of inflation. Though it would be interesting to investigate the possibility of the stabilization mechanism as a way to stop inflation and generate reheating, it is beyond the scope of this work. Similar expression can be derived for $`\mathrm{\Phi }^+(\tau _f)`$ in terms of the number of e-foldings $`N`$,
$`{\displaystyle \frac{\mathrm{\Phi }^+(\tau _f)}{G_P}}=1{\displaystyle \frac{e^{\gamma N}}{\alpha }}.`$ (35)
In the perturbative limit, $`f`$, the radion field determines the perturbation of the separation between the two branes. The result in Eq.(33) shows, that by assuming $`f=0`$ as an initial condition for inflation to occur, $`f(\tau )`$ increases during inflation. The increased value depends on the number of e-foldings that can be achieved during inflation. Therefore, inflation on the Planck brane results in increasing the separation between the two branes, i.e, the size of the AdS space. On the other hand Eq.(35) suggests that the increase in the radion field $`f(\tau )`$ also leads to gradual increase in $`\mathrm{\Phi }^+`$ and at the end of inflation it’s value asymptotically approaches $`G_\mathrm{P}1/(kG_5)`$.
### B On the negative tension brane
In this section we study the feasibility of an inflationary solution on the negative tension brane. The action on the negative tension brane is given in Eq.(8) and Eq.(12) with negative superscript. Again in the perturbative limit, $`\mathrm{\Phi }^{}`$ in Eq.(9), reduces to a simple form
$`\mathrm{\Phi }^{}{\displaystyle \frac{1}{kG_5}}\alpha e^{2k(\alpha 1)f},`$ (36)
and the parameter:
$`\omega _{}{\displaystyle \frac{3}{2}}(1+{\displaystyle \frac{2}{\alpha }}{\displaystyle \frac{3}{kG_5\mathrm{\Phi }^{}}}).`$ (37)
Within the perturbative limit $`e^{2k\alpha f}\alpha `$, $`2\omega _{}+3`$ is greater than zero, provided
$`f(\tau )<{\displaystyle \frac{\mathrm{ln}(3/2)}{2k\alpha }}.`$ (38)
In the perturbative limit, we also have:
$`kG_5\mathrm{\Phi }^{}\alpha 1,`$ (39)
for $`f0`$. Note, that the situation is exactly opposite to that of the positive tension brane, where $`\mathrm{\Phi }^+`$ is small and $`\omega _+1`$.
With these assumptions and with the help of slow-roll conditions Eqs.(1820), Eq.(13) can be simplified to yield,
$$3H^2\frac{8\pi V_0}{\mathrm{\Phi }^{}}.$$
(40)
Here we have used Eq.(37), and we have also assumed the initial condition $`f(\tau =\tau _0)0`$. On the other hand Eq.(15), around $`f0`$, reduces to:
$$\frac{\dot{\mathrm{\Phi }}^{}}{\mathrm{\Phi }^{}}\frac{16\pi }{27}\frac{kG_5V_0}{H(1G_\varphi \mathrm{\Phi }^{})},$$
(41)
where $`G_\varphi \frac{2}{3}\frac{kG_5}{\alpha }`$, such that around $`f0`$, $`G_\varphi \mathrm{\Phi }^{}2/3`$. Hence, the slow-roll condition: $`\dot{\mathrm{\Phi }}^{}/\mathrm{\Phi }^{}H`$ requires $`kG_5V_0/HH`$. However, it is easy to show that combined with the result of Eq.(40) this leads to an inconsistent result $`kG_5\mathrm{\Phi }^{}1`$, compared to Eq.(39). It leads to the conclusion that within the perturbative limit the slow-roll inflation on the negative tension brane breaks down from the very beginning.
The result can be understood as follows. On the negative tension brane, the effective physical scale is reduced by a factor of $`\alpha e^{2kz_c}`$ compared to the fundamental scale $`M_{Pl}`$, the same mechanism used by Randall-Sundrum to explain the hierarchy between the Planck scale and the electroweak scale . Hence, inflation is driven by the flat potential $`V_0/\alpha `$ of $`\chi `$ Eq.(40). On the other hand, due to the fact that $`\omega _{}`$ is small on the negative brane , in fact $`2\omega _{}+31/\alpha `$, the coupling between $`\mathrm{\Phi }^{}`$ and the inflaton $`\chi `$ is boosted by a factor of $`\alpha `$ (see the term on the right-hand of Eq.(15)). This creates an effective potential for $`\mathrm{\Phi }^{}`$ with a steep slope, hence the evolution of $`\mathrm{\Phi }^{}`$ is no more subdominant compared to the Hubble expansion rate which is set here by the reduced strength of $`V_0`$. And eventually the slow-roll assumptions break down. On the other hand, one may also conclude that the inflation quickly pushes the two branes out of their equilibrium positions, hence, destabilizes the configuration.
Note that the inconsistency is related to the fact that the value of $`\omega _{}`$ in the perturbative limit is non-physical, i.e., its value does not satisfy the present experimental bound $`\omega >3000`$ as we discussed earlier. One may consider a situation where the stabilization process is inter-winded with the inflation. By including new contributions from massive particles in the bulk, the effective steep potential of $`\mathrm{\Phi }^{}`$ could be balanced to provide a possible solution to the problem. (It was suggested by some of the studies on cosmology in Randall-Sundrum scenario .) Inflation in such a case would be driven by the massive Kaluza-Klein modes of the bulk field, which could possibly assist inflation on the brane, work in this direction is in progress.
## IV Discussion
We have shown that it is possible to discuss the dynamics of inflation with additional matter Lagrangian on the positive tension brane. Assuming inflation takes place before the stabilization of the brane separation, we find inflation on the positive tension brane displaces the radion field, but can stay within the perturbative limits with large brane separations. Due to the expansion of the positive tension brane, the separation between the two branes increases slightly and this could be the initial condition for a subsequent process in which the brane separation is stabilized and the radion field becomes massive. However, on the negative tension brane, the slow-roll inflation destabilizes the brane separation at the very beginning, and it is not possible to reach the general relativity limit for a given separation between the two branes.
## Acknowledgments
J.W. would like to thank M. Cvetič and D. Chung for helpful discussions. A.M. is supported by the Inlaks foundation and an ORS award. J.W. is supported by U.S. Department of Energy Grant No. DE-AC02-76CH03000.
|
warning/0004/math-ph0004022.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
For a long time, the diffuse part in diffraction spectra played only a minor role in crystallography. This was mainly due to experimental restrictions but also to a lack of theoretical studies. In this letter, we investigate some simple models with disorder, both deterministic and random, where the diffuse background of the diffraction must not be neglected in the structural analysis.
We first start with an introduction to the language of mathematical diffraction theory, which is necessary to get a full understanding of the spectrum. The following two examples have exactly the same diffraction pattern albeit their disorder is of completely different type. We then discuss models with singular continuous diffraction, a type of spectrum that is not present in classical crystallography but may appear in random structures with long range order. This is followed by the Ising lattice gas where the Bragg spectrum alone does not reflect the correct symmetry. Last, we compare square ice with an alternative model that can only be distinguished by the diffuse scattering.
## 2 Diffraction theory
Let us recall some notions of diffraction theory , adapted to our purposes using the measure theoretic approach . On the assumption of kinematical diffraction in the Fraunhofer picture, the diffracted intensity per atom $`\widehat{\gamma }_\omega `$ is given by the Fourier transform of the autocorrelation $`\gamma _\omega `$. Let the unit point measure (Dirac distribution) $`\delta _x`$ idealize an atom at position $`x`$ and $`\mathrm{\Lambda }`$ denote the point set of all possible atomic positions. Then the atomic structure is described by the so-called weighted Dirac comb
$$\omega _\mathrm{\Lambda }=\underset{x\mathrm{\Lambda }}{}w(x)\delta _x,$$
(1)
where $`w(x)\{0,1\}`$ indicates whether the point $`x`$ is actually occupied or not. More generally, $`w(x)`$ can be a bounded complex weight function. The autocorrelation (Patterson function) $`\gamma _\omega `$, assuming it is well defined, is given by
$$\gamma _\omega =\underset{z\mathrm{\Delta }}{}\nu (z)\delta _z,$$
(2)
with $`\mathrm{\Delta }=\mathrm{\Lambda }\mathrm{\Lambda }`$ being the set of difference vectors which we assume discrete and closed. The autocorrelation coefficient $`\nu (z)`$ can be calculated according to
$$\nu (z)=\underset{R\mathrm{}}{lim}\frac{1}{\mathrm{vol}(B_R)}\underset{\stackrel{y\mathrm{\Lambda }_R}{zy\mathrm{\Lambda }}}{}w(y)\overline{w(zy)},$$
(3)
where $`B_R`$ is the ball of radius $`R`$ around the origin, $`\mathrm{\Lambda }_R=\mathrm{\Lambda }B_R`$ and $`\overline{}`$ complex conjugation. If $`w(x)\{0,1\}`$, $`\nu (z)`$ is the frequency of having two scatterers at distance $`z`$.
The Fourier transform of a function $`\varphi `$ is
$$\widehat{\varphi }(k)=_^ne^{2\pi ikx}\varphi (x)𝑑x,$$
(4)
and we use the standard theory of tempered distributions from here, see for our conventions.
The diffraction image, $`\widehat{\gamma }_\omega `$, is a positive measure that tells us how much intensity is scattered into a given volume element. So, by the general decomposition theorem \[16, Ch. I.4\], it admits a unique decomposition with respect to Lebesgue’s measure into three parts, $`\widehat{\gamma }_\omega =(\widehat{\gamma }_\omega )_{pp}+(\widehat{\gamma }_\omega )_{sc}+(\widehat{\gamma }_\omega )_{ac}`$. The pure point part $`(\widehat{\gamma }_\omega )_{pp}`$ consists of the Bragg peaks, the absolutely continuous $`(\widehat{\gamma }_\omega )_{ac}`$ corresponds to the usual diffuse background that can often be described as a continuous function, otherwise as an $`L^1`$-function. The singular continuous component $`(\widehat{\gamma }_\omega )_{sc}`$ lies somewhere in between, i.e. it is neither a smooth function nor “as singular” as a Bragg peak, for a precise definition see .
## 3 Random versus deterministic disorder
In the following, we present two one-dimensional examples with completely different type of (dis)order but displaying the same diffraction spectrum, i.e. they are homometric.
The first model is a Bernoulli system on $``$. The structure is given by the stochastic Dirac comb $`\omega ^B=_m\eta ^B(m)\delta _m`$ where $`\eta ^B(m)`$ is a family of random variables that take the values $`h_1`$ and $`h_2`$ with probabilities $`p_1`$ and $`p_2`$. For such a lattice gas, one can prove that its diffraction measure almost surely consists of a uniform pure point and a constant absolutely continuous part,
$$\widehat{\gamma }_\omega ^B=|𝒉|^2\underset{m}{}\delta _m+\text{Var}(𝒉),$$
(5)
with mean $`𝒉=p_1h_1+p_2h_2`$ and variance $`\text{Var}(𝒉)=|𝒉|^2|𝒉|^2`$. For $`h_1=1`$, $`h_2=0`$ and $`p_1=p_2=1/2`$, we have $`|𝒉|^2=\text{Var}(𝒉)=1/4`$.
Let us compare this with the Rudin-Shapiro sequence, where the atoms are distributed deterministically. We construct $`\omega ^{RS}=_m\eta ^{RS}(m)\delta _m`$ according to the rule
$$\eta ^{RS}(0)=1,\eta ^{RS}(1)=0,$$
$$\eta ^{RS}(4m)=\eta ^{RS}(4m+1)=\eta ^{RS}(m),$$
(6)
$$\eta ^{RS}(4m+l)=\frac{1}{2}\left(1(1)^{m+l+\eta ^{RS}(m)}\right),l=2,3$$
Alternatively, this sequence may be defined via a substitution rule on a four letter alphabet . For the initiate, we add that we use the square of the traditional substitution to get the fixed point of the bi-infinite sequence. The diffraction spectrum can be calculated rigorously \[15, pp. 165–168\]. With our choice of the scattering strenghts, this results in
$$\widehat{\gamma }_\omega ^{RS}=\frac{1}{4}+\frac{1}{4}\underset{m}{}\delta _m.$$
(7)
Amazingly, though the structure is completely deterministic, its two-point correlations are destroyed systematically so that only a constant diffuse background remains in $`\widehat{\gamma }_\omega `$.
Therefore, with $`h_1=1`$ and $`h_2=0`$, we have $`\widehat{\gamma }_\omega ^B=\widehat{\gamma }_\omega ^{RS}`$, a distinction of the original structure on the basis of the diffraction spectrum is impossible. Even for finite systems, the slight difference in the fluctuations of the background is almost undetectable. Since the Fourier transform is unique, this means that the Rudin-Shapiro sequence is homometric to the above Bernoulli chain (a statement which is true with probability one). The remarkable feature of this example is the inclusion of a diffuse background, see for other cases of homometry.
Note that this example explores the full entropy range: the Bernoulli case has entropy $`\mathrm{log}2`$, the maximal value for a binary system, while Rudin-Shapiro has entropy 0. It is clear that one can find other examples with the same diffraction image but entropy between these extremal values. So, a resolution of the corresponding reconstruction problem, unless extra information is available, needs an optimization approach, e.g. by choosing the structure which maximizes the configuration entropy.
## 4 Singular continuous spectra
There is no simple characterization of the singular continuous part of a diffraction spectrum, since this term covers the whole range (in the sense of tempered distributions) between a continuous function, resp. $`L^1`$-function, and a Dirac distribution. As indication one may use the scaling behaviour of the intensities with the system size $`N`$: $`\widehat{\gamma }_{pp}N`$, $`\widehat{\gamma }_{ac}const.`$ and $`\widehat{\gamma }_{sc}N^\alpha `$, $`0<\alpha <1`$, but this recipe is not correct in general . A classical example where the scaling argument yields the right answer is the Thue-Morse sequence . Singular continuous spectra usually appear in structures where the constructive interference that is responsible for the Bragg peaks is impaired by some randomness. Nonetheless, some long range order is still strong enough to prevent the peaks from becoming completely diffuse. Note that singular “peaks” can never be isolated – a property that further complicates their analysis.
As an illustration in one dimension, let us consider a “tiling” of the line without gaps and overlaps which consists of two prototiles (intervals) of length 1 and $`\tau =(1+\sqrt{5})/2`$. The unit scatterers are located on the left endpoints of the intervals. By placing the intervals randomly with arbitrary (but fixed) probabilities, the resulting spectrum is absolutely continuous except for the trivial Bragg peak at $`k=0`$, see \[3, Thm. 2\]. On the other hand, arranging these intervals as a Fibonacci sequence leads to the well-known quasicrystal with purely discrete diffraction spectrum, cf .
With the same tiles, one can now also construct a “structure intermediate between quasiperiodic and random”. This is achieved using the so-called circle sequence . The positions $`x_n`$ of the atoms are given by
$$x_nx_{n1}=1+\xi \mathrm{𝟏}_{[0,\beta )}(n\alpha )$$
(8)
where $`0<\beta <1`$, $`\mathrm{𝟏}_{[0,\beta )}`$ is the characteristic function of the interval $`[0,\beta )`$ and $`x_0`$, $`\xi `$ are parameters; assume $`x_0=0`$ for simplicity. In order to have the same tile lengths and density $`d=1/(1+\beta \xi )`$ as in the Fibonacci case, we choose $`\xi =\tau 1`$ and $`\beta =2\tau `$. As was shown by Hof , the circle sequence has purely singular continuous diffraction spectrum apart from the trivial Bragg peak for every irrational $`\xi `$, every $`\beta `$ and generic (“most”) $`\alpha `$. Nevertheless, the assessment in each explicit case is open.
In two dimensions, a straightforward generalization of the above mentioned Thue-Morse chain is given by the next example. It is constructed by using a two-dimensional substitution rule on a two-letter alphabet:
$$\rho :a\begin{array}{cc}b& a\\ a& b\end{array};b\begin{array}{cc}a& b\\ b& a\end{array},$$
(9)
or, using two elementary cells with different atoms,
The resulting structure is a simple Cartesian product of Thue-Morse chains and can be treated using the method indicated in \[3, Sec. 4\]. The diffracted intensity is a product measure $`\widehat{\gamma }_\omega ^{2\text{D}}=\widehat{\gamma }_{\omega ^{(1)}}^{\text{TM}}\widehat{\gamma }_{\omega ^{(2)}}^{\text{TM}}`$ of the corresponding one-dimensional intensities $`\widehat{\gamma }_{\omega ^{(i)}}^{\text{TM}}`$ (compare ). We assign scattering strenghts of $`+1`$ (full circles) and $`1`$ (open circles) to the atoms. Taking values $`1`$ and $`0`$ would just lead to an additional, trivial Bragg part on $`^2`$. The remaining diffracted intensity is a product of singular continuous measures and thus itself purely singular continuous (Fig. 2). We have to emphasize that this singular continuous behaviour is by no means recognizable from the shape of the peaks and the decrease of their wings. The most likely fit is given by Gaussians, but this is also the case for true Bragg peaks.
The explicit scaling of the peaks with the system size can be obtained by applying the formulae of \[6, Sec. 4.3\] to our product structure.
## 5 The Ising lattice gas and its symmetry
It’s worth taking a glance at the well-known 2D Ising model where, in contrast to the previous examples, we have some interaction between the atoms. In the lattice gas formulation, we place atoms and holes (scattering strengths 1 and 0) on the square lattice subjected to coupling constants $`K_i=J_i/k_BT>0`$ in $`x`$\- and $`y`$-direction. There is a phase transition at $`(\mathrm{sinh}(2K_1)\mathrm{sinh}(2K_2))^1=1`$. One can show that the pure point part of the diffraction spectrum is fourfold symmetric for all $`K_i`$ and has the form
$$(\widehat{\gamma }_\omega )_{pp}=\{\begin{array}{cc}\frac{1}{4}_{𝒌^2}\delta _𝒌,\hfill & TT_c\hfill \\ \rho ^2_{𝒌^2}\delta _𝒌,\hfill & T<T_c,\hfill \end{array}$$
(10)
where $`\rho `$ is the ensemble average of the number of scatterers per unit volume, $`1/2\rho 1`$.
The correct (two-fold) symmetry is visible only from the diffuse part of the spectrum, the Bragg part alone is misleading. Using the asymptotic behaviour of the correlation functions, the background can be shown to be an absolutely continuous measure , which can be represented by a continuous function. The background concentrates around the Bragg peaks, as expected for an attractive interaction.
## 6 The ice model
Square ice is another example for which the relevant structural information of the diffraction image is only contained in the diffuse background. In this model, oxygen atoms are sitting on the vertices of a square lattice, see Fig. 4. The hydrogens are placed on the edges with a distance of $`1/3`$ to the next O according to the following ice (or Bernal-Fowler) rules ( and references therein):
1. There are two hydrogens adjacent to each oxygen.
2. There is only one hydrogen per bond.
By using the equivalence to a special case of the six-vertex model, Lieb solved the ice model and calculated its residual entropy to be $`\frac{3}{2}\mathrm{log}\frac{4}{3}`$. Although calorimetric measurements on three-dimensional ice confirmed the predicted value of the entropy very well, it was almost impossible to detect the hydrogen disorder in X-ray scattering experiments. Neutron studies are more promising, compare where a highly disordered hydrogen structure was shown in deuterated cubic ice Ic.
Apart from the weak scattering power of the hydrogens, another more fundamental problem arises. Consider the case of a Bernoulli system where we ignore the first ice rule, so that the hydrogens are distributed with equal probability independently of the others.
By assigning scattering strengths $`h_\text{O}`$ and $`h_\text{H}`$ to the atoms, a straightforward calculation yields the autocorrelation of the Bernoulli model:
$$\begin{array}{c}\gamma _\omega =\underset{𝒛^2}{}(h_\text{O}^2+h_\text{H}^2+\frac{h_\text{H}^2}{2}\underset{x,y=\pm \frac{1}{3}}{}\delta _{(x,y)}\hfill \\ \hfill +(h_\text{O}h_\text{H}+\frac{h_\text{H}^2}{4})\underset{x=\pm \frac{1}{3}}{}(\delta _{(x,0)}+\delta _{(0,x)}))\delta _𝒛\\ \hfill +h_\text{H}^2\left(\delta _0\frac{1}{4}\underset{x=\pm \frac{1}{3}}{}\left(\delta _{(x,0)}+\delta _{(0,x)}\right)\right).\end{array}$$
(11)
Thus, the Bragg part is given by $`(𝒌=(k_1,k_2))`$
$$\left(\widehat{\gamma }_\omega \right)_{pp}=\underset{𝒌^2}{}\left(h_\text{O}+h_\text{H}\left(\mathrm{cos}\frac{2\pi k_1}{3}+\mathrm{cos}\frac{2\pi k_2}{3}\right)\right)^2\delta _𝒌,$$
(12)
and the absolutely continuous background can be represented by the continuous function
$$g(𝒌)=h_\text{H}^2\left(\mathrm{sin}^2\frac{\pi k_1}{3}+\mathrm{sin}^2\frac{\pi k_2}{3}\right).$$
(13)
But due to the same averaged occupation probabilities of the hydrogens in the ice and the Bernoulli model, their pure point parts are indistinguishable. In other words, Eq. (12) is valid for the case of square ice, too. The background is different, of course, but very weak, in particular because, in reality, $`h_\text{O}h_\text{H}`$.
Fig. 5 displays the diffraction image of square ice. To focus on the contribution of the hydrogens, we set $`h_\text{O}=0`$. Positive $`h_\text{O}`$ would result in an increase of intensity of the Bragg peaks, except for the case of $`h_\text{O}=h_\text{H}`$ where we have extinction at $`(k_1,k_2)=(1,1)`$, (1,2), (2,1) and (2,2), according to Eq. (12). The continuous background, however, depends only on $`h_\text{H}`$, as in the Bernoulli case.
We have no explicit formula for the background in the ice model, since the correlation functions are not known, but we can calculate the diffraction image numerically (FFT) from a pattern of a Monte Carlo simulation, see \[13, Ch. 7\] for the method. The decay of the correlations suggests that the background is an absolutely continuous measure and there is no singular continuous component present. Note that the interpretation of the diffuse part is more involved here – it needs to include multi-particle effective interactions.
|
warning/0004/hep-ph0004221.html
|
ar5iv
|
text
|
# On the derivation of the neutrino oscillation length formula.
## 1 Three derivations
Let us represent the neutrino born at space-time point $`(0,0)`$ in some charged current reaction involving charged lepton $`l`$ by
$`|0,0>=|\nu _l>={\displaystyle \underset{h}{}}U_{lh}|h>`$ (1)
where $`|h>`$ is a mass eigenstate with eigenvalue $`m_h`$ and definite energy and momentum.
The fate of this neutrino is governed by the space-time translation operator
$`𝒰=e^{i(Ht\stackrel{}{P}\stackrel{}{r})}`$ and the problem amounts to correctly evaluate its action on (1):
$`𝒰|0,0>=|x,t>={\displaystyle \underset{h}{}}U_{lh}e^{i(E_htp_hx)}|h>`$ (2)
where we have assumed that the propagation is along the $`x`$ axis. The precise values of $`E_h`$ and $`p_h`$ in this formula depend on the production kinematics. In the case of a $`\pi _{l2}`$ decay for example, they are fixed by the masses in the $`\pi `$ rest-frame and from there in the lab, once the $`\pi `$’s decay angle and velocity are given. Projection of $`|x,t>`$ onto $`|\nu _l^{}>`$ yields then
$`A(ll^{})(x,t)={\displaystyle \underset{h}{}}U_{l^{}h}^{}U_{lh}e^{i(E_htp_hx)}`$ (3)
for the amplitude to detect a neutrino of flavor $`l^{}`$ at point $`x`$ and time $`t`$ relative to the production point at $`(0,0)`$, granted that the neutrino interacts.
Henceforth, we shall assume a two flavor-two mass world, which simplifies the matter greatly.
The all-important object is then the one phase difference which appears upon squaring (3):
$`P(ll^{}l)(x,t)=sin^2(2\theta )sin^2({\displaystyle \frac{\delta \varphi }{2}})`$ (4)
where we have reverted to the simplified notation in use in the two flavor world and abbreviated: $`\delta \varphi =\delta Et\delta px`$
In order to make contact with experiments which register the coordinates but not the time and because the object of study should be more properly described by some more or less localized wave function, a connection between $`x`$ and $`t`$ must be made at some point to describe what shall be considered as the motion of the center of the wave packet. Also, the phase difference should be expressed in terms of the quantities which are really at stake, viz. the masses and a single kinematical value representing the average energy or momentum of the beam. To this effect, various supplementary hypotheses are added to the basic ingredient represented by formula (2)
None of these is necessary in the case at hand as we shall see now, provided the first problem is properly treated.
### 1.1 Equal energies
It is assumed here that the two massive components have the same energy and different momenta; then $`\delta \varphi =\delta px`$ and the oscillation pattern is described by:
$$O(x)=sin^2(\frac{\delta px}{2})$$
Invoking a relativistic situation, people usually expand $`pE\frac{m^2}{2E}`$ hence $`\delta \varphi \frac{\delta m^2}{2E}x`$ and <sup>1</sup><sup>1</sup>1$`\delta `$ is a true, signed, difference, and $`\mathrm{\Delta }`$ is an absolute value
$`O(x)sin^2({\displaystyle \frac{\mathrm{\Delta }m^2}{2E}}x)`$ (5)
However, this is unnecessary since the exact relation:
$`\delta p^2=\delta E^2\delta m^2`$ (6)
yields in this case:
$$\delta p=\frac{\delta m^2}{\mathrm{\Sigma }p}=\frac{\delta m^2}{2\overline{p}}$$
with an obvious definition for $`\overline{p}`$.
Consequently:
$`O(x)=sin^2({\displaystyle \frac{\mathrm{\Delta }m^2}{4\overline{p}}}x)`$ $`\mathrm{and}`$ $`L_{osc}={\displaystyle \frac{4\pi \overline{p}}{\mathrm{\Delta }m^2}}`$ (7)
Simple as is it, the meaning of this procedure is completely clear only in the case of two masses, since in the more general situation we could not introduce the third momentum into the definition of $`\overline{p}`$.
### 1.2 Equal momenta
Here, $`\delta \varphi =\delta Et`$ and using again a first order expansion, this becomes: $`\delta \varphi \frac{\delta m^2}{2p}t`$ after which the further ’relativistic approximations’ $`tx`$ and $`pE`$ allow to find consistency with the approximate result (5).
This again is unnecessary because (6) yields here $`\delta E=\frac{\delta m^2}{\mathrm{\Sigma }E}`$ and, upon defining the velocity $`v`$ of the center of the would-be wave packet, one finds:
$$\delta \varphi =\frac{\delta m^2}{\mathrm{\Sigma }E}\frac{x}{v}=\frac{\delta m^2}{2p}x$$
provided
$$v=\frac{2p}{E_1+E_2}$$
which shall be justified presently, but is seen to agree with the arithmetic mean up to and including first degree terms in the small quantity $`\frac{\mathrm{\Delta }E}{\mathrm{\Sigma }E}`$
Hence
$`O(x)=sin^2({\displaystyle \frac{\mathrm{\Delta }m^2}{4p}}x)`$ $`\mathrm{and}`$ $`L_{osc}={\displaystyle \frac{4\pi p}{\mathrm{\Delta }m^2}}`$ (8)
exactly as in (1.1), granted the definition used for $`v`$, and with an analogous restriction since only two energies must be considered in defining $`v`$.
### 1.3 Equal velocities
In this case, $`\delta \varphi `$ does not reduce to a single term and it is very important not to approximate $`t`$ by $`x`$, for in so doing one would arrive at:
$$\delta \varphi =(\delta E\delta p)x=\delta me^\eta x$$
upon introducing $`\eta =\mathrm{tanh}^1(v)`$
The oscillating pattern would be described by:
$`sin^2({\displaystyle \frac{\mathrm{\Delta }m}{2}}e^\eta x)`$ $`\mathrm{and}\mathrm{the}\mathrm{oscillation}\mathrm{length}:`$ $`L_{osc}^{}={\displaystyle \frac{2\pi e^\eta }{\mathrm{\Delta }m}}`$ (9)
However, since
$$e^\eta =\frac{E_1+p_1}{m_1}=\frac{E_2+p_2}{m_2}\frac{4p}{m_1+m_2}$$
this yields finally
$`L_{osc.}^{}={\displaystyle \frac{8p\pi }{\mathrm{\Delta }m^2}}`$ (10)
viz. twice the usual value.
Confronted with this result, people have been tempted to think that the standard formula is either false or does not apply in the case at hand.
A more carefull treatment of the motion of the would-be wave packet shows that this is not correct. Indeed, if the hypothesis of equal velocities has any meaning, then the center of the wave packet moves with that velocity, not with velocity 1. Therefore, defining its position by $`x=vt`$ yields:
$$\delta \varphi =\delta E(tvx)=\delta m\gamma (1/vv)x=\frac{\delta m}{v\gamma }x$$
<sup>2</sup><sup>2</sup>2$`\gamma =1/\sqrt{1v^2}`$ as usual
Now
$$\frac{1}{v\gamma }=\frac{m_1}{p_1}=\frac{m_2}{p_2}=\frac{\mathrm{\Sigma }m}{2\overline{p}}$$
Hence $`\delta \varphi =\frac{\delta m^2}{2\overline{p}}x`$ and the correct formula found in (1.1) results.
The same restriction as before applies, since the third momentum cannot enter the definition of $`\overline{p}`$.
Clearly, replacing $`1/vv`$ by $`1v`$ (equivalent to $`tx`$) cannot be harmless; in (1.2), $`v`$ is only an overall factor, and replacing it by $`1`$ induces a relative error on the phase shift which goes to $`0`$ with $`1v`$. Not so in the present case where the relative error is $`1/(1+v)`$ \- hence the factor $`2`$ found above. Stated differently, $`v`$ and $`E`$ were treated separetly in (1.2) but here the connection between $`m`$, $`v`$ and $`E`$ (or $`p`$) must be used.
## 2 ..and a fourth one.
First observe that all three derivations above use exact relativistic kinematics but that none uses any sort of ’ultra-relativistic approximation’, especially not the ubiquitous but very unreasonable ’$`xt`$’. A moment of reflexion reveals their common feature: in all three cases, the center of the would-be wave packet is endowed with the average momentum and energy of the components: $`p^c=\overline{p}`$, $`E^c=\overline{E}`$, and it is assumed to have velocity $`v=\frac{p^c}{E^c}`$. This is the real justification behind the definition of $`v`$ in (1.2).
One is thus led to think that this is all that is needed to yield the well-known $`L_{osc}`$; indeed, baring any ad hoc hypothesis on the production kinematics:
$`v={\displaystyle \frac{p_1+p_2}{E_1+E_2}}`$ $`\delta \varphi =\delta px\delta Et`$ $`=(\delta p\delta E{\displaystyle \frac{\mathrm{\Sigma }E}{\mathrm{\Sigma }p}})x={\displaystyle \frac{(\delta p^2\delta E^2)x}{\mathrm{\Sigma }p}}`$
$`={\displaystyle \frac{\delta m^2x}{2\overline{p}}}`$
which proves our point.
## 3 Lessons
The usefulness of all this is of course lessened by the well known shortcomings of the use of plane waves for the purpose of describing neutrino oscillations (see e.g. for a list of these) ; the necessity of using wave packets (see e.g.) or field theory (,) has been the subject of a long and still ongoing debate and many sophisticated treatments have appeared over the years. However, these more elaborate methods together with the inclusion of the neutrino production and/or detection processes in the description of the phenomenon (,) all result in formulae which are subtended by the basic oscillation pattern described by (7), provided there exists a middle-zone where coherence is not lost but finite source length and momentum spread effects are negligible. In all cases, the same (vacuum) oscillation length obtains when and where resolution or decoherence do not blur the oscillations.
The above demonstration sheds some light on this robustness of the classical formula, by showing that none of the extra hypotheses usually made is necessary, at least in the two flavor world where the oscillation length has its clearest meaning. Buried at the heart of the more sophisticated treatments is always some definition of the $`xt`$ relationship which avoids the hand-waving $`xt`$.
It is also seen that the relevant variable is the momentum, not the energy, when the distinction applies; provided one uses this variable, the standard result seems to follow also in the non relativistic regime. This might be usefull if slow, Karmen-anomaly-like objects are confirmed; consideration of non-relativistic effects in oscillations due to such states have already appeared in the litterature (see )
A note of caution is, however, in order: it is not entirely clear that an oscillation probability -and therefore, an oscillation length- has a meaning in itself in the non-relativistic regime; the wave packet treatment (except in its most primitive form) and the field theoretical approach both include the production and/or detection processes in an overall probability calculation which generally factorizes in the relativistic regime, where all masses are small with respect to the kinetic energy scale. It should be determined under what conditions this also applies in the non relativistic regime. Since oscillations can only occur in the case of nearly degenerate mass states, the production phase-space mass dependence should not be of concern, but the detection reaction is a potential source of problem in that, e.g. the $`\nu _\tau `$ and possibly the $`\nu _\mu `$ component can be inhibited for lack of energy; one can really appreciate at this point how much ill-defined are the so called ’weak eigenstates’ (besides the fact that they are not eigenstates of any operator which distinguishes them from the mass eigenstates!)
Moreover, one must expect a larger yield of ’wrong’ helicity when $`\gamma 1`$ and therefore an additional entanglement of ’flavor’ with the other variables which might further preclude the definition of an ’oscillation probability’ disconnected from the rest of the process .
## 4 Summary and conclusion
Simple calculations show that none of the hypotheses usually employed in deriving the neutrino oscillation length in the plane wave formalism is necessary and that the only requirement is a proper treatment of the motion of the would-be wave packet, at least in a two-mass world. This should apply to real life whenever the number of active mass states is reduced to two, in particular when production phase-space is restricted or in case of degeneracy.
Although a plane wave treatment of neutrino oscillations is, admittedly, an over-simplification, we believe that what has been done here has the merit of giving some clues in answering questions as to what are the conditions that should be met to allow observation of oscillations or that should be hypothetized in a sensible theoretical treatment. In particular, it casts some shadow on the relevance of the equal energies versus equal momenta arguments that have appeared in the litterature, given that the one important ingredient seems to be a suitably defined velocity for the center of the wave packet representing the object under study.
|
warning/0004/cond-mat0004074.html
|
ar5iv
|
text
|
# Macroscopic Quantum Oscillations of Antiferromagnetic Nanoclusters in a Swept Magnetic Field
## Abstract
New macroscopic quantum interference effects in the behavior of the antiferromagnetic nanoclusters under action of a swept magnetic field are predicted and theoretically investigated, namely : oscillations of the magnetic susceptibility of nanocluster similar to the electron Bloch oscillations in crystal and the corresponding Stark ladder-like resonances in ac-magnetic field. These effects are related to that of the macroscopic quantum coherence phenomena. The analogy of the antiferromagnetic nanoclusters quantum behavior and Josephson junction with a small capacity was revealed as well.
PACS: 03.65.-w, 36.40.-c, 73.23.-b, 75.50.Xx
The quantum dynamics of magnetic nanoclusters with a large spin is drawing increasing attention, both from experiment and theory. Fundamental scientific questions at stake are the macroscopic quantum phenomena including the macroscopic tunneling and interference of wave functions and the transition from quantum to classical behavior – a process known as decoherence. From this point of view metallorganic molecules with large number of d-ions (Fe, Mn, Co etc) such as Mn<sub>12</sub>, Fe<sub>8</sub>, Fe<sub>10</sub> are of particular interest. The presence of strong enough antiferromagnetic interaction between d-ions is a characteristic feature of these objects. Some of these clusters have the magnetic structure of the ferrimagnetic type (Mn<sub>12</sub>, Fe<sub>8</sub>) with large spin moment in the ground state ($`S=10`$). As of now the quantum mechanical properties of these nanoclusters are investigated in great detail. Recently such new phenomena and specific features as molecular bistability , macroscopic quantum tunneling of magnetization, quantum hysteresis , the appearance of Berri–phase and quantum peculiarity of magnetic susceptibility behavior , have been discovered and predicted. The quantum properties of nanoclusters with magnetic structure of antiferromagnet type (Fe<sub>10</sub>, Fe<sub>6</sub>) are less well understood . The aim of the present paper is the investigation of macroscopic quantum dynamics phenomena of the antiferromagnetic clusters in a swept magnetic field. The reason to use the swept field is that it produces a torque on the spin system of the antiferromagnetic molecules accelerating the spin precession and displays a new macroscopic quantum interference effects in their behavior.
We consider the ring-like molecular magnets (Fe<sub>10</sub>, Fe<sub>6</sub>, Fe<sub>18</sub>) composed of $`N=2k`$ $`d`$–ions spaced on a circle lying in the $`xy`$-plane; there is an antiferromagnetic exchange interaction between them. The Hamiltonian of this system can be represented as
$$=J\left(\underset{i=1}{\overset{N1}{}}\stackrel{}{S}_i\stackrel{}{S}_{i+1}+\stackrel{}{S}_1\stackrel{}{S}_N\right)+\underset{i=1}{\overset{N}{}}H_{cr}\left(S_i\right)+g\mu _B\underset{i=1}{\overset{N}{}}\stackrel{}{S}_i\stackrel{}{B},$$
(1)
where $`J>0`$ is the exchange interaction constant, $`H_{cr}(S_i)`$ is the crystal field acting on i-th spin ($`H_{cr}=K_1S_z^2+K_2(S_x^2S_y^2)`$, with $`z`$-axis directed perpendicular to a molecule plane); for Fe<sup>3+</sup> $`S=5/2`$. In accordance with experimental results it will be assumed that $`JK`$ (the typical values are $`J`$ 10 cm<sup>-1</sup>, $`K`$1 cm<sup>-1</sup>). The anisotropy in the plane($`K_2`$) can be formed artificially, e.g. by means of external electric or magnetic fields, pressure, or using anisotropic substrate. Due to $`C_n`$-symmetry there is also ”$`n`$-gonal anisotropy in the plane”($`n=6(10)`$ for Fe<sub>6(10)</sub>). Following analysis can be easily reformulated to take into account this anisotropy. Let suppose that the magnetic field is $`\stackrel{}{B}=(0,\mathrm{\hspace{0.17em}0},B_z)`$, where $`B_z`$ depends on time generally. We decompose as usually for antiferromagnets the local spins into the two subsystems (“sublattices”) and it is reasonable to assume the coordinates of two magnetic sublattices, describing the macroscopic quantum behavior of the antiferromagnetic molecule as degrees of freedom. In order to describe such system the coherent quantum states will be used
$$|\theta _1,\phi _1;\theta _2,\phi _2,$$
(2)
where $`\theta _i,\phi _i`$ are the polar and azimuthal angles of the spin of the magnetic sublattices. (These angles are measured from the $`z`$ and $`x`$ axes respectively). The Lagrangian of the system can be represented as:
$`L`$ $`=`$ $`{\displaystyle \frac{M}{\gamma }}{\displaystyle \underset{i=1}{\overset{2}{}}}\mathrm{cos}\theta _i\dot{\phi }_i+B_zM{\displaystyle \underset{i=1}{\overset{2}{}}}\mathrm{cos}\theta _i{\displaystyle \frac{1}{2}}K_u{\displaystyle \underset{i=1}{\overset{2}{}}}\mathrm{sin}^2\theta _i{\displaystyle \frac{1}{2}}K_{}{\displaystyle \underset{i=1}{\overset{2}{}}}\mathrm{sin}^2\theta _i\mathrm{sin}^2\phi _iNJS^2[\mathrm{cos}(\theta _1+\theta _2)`$ (4)
$`+\mathrm{sin}\theta _1\mathrm{sin}\theta _2(1+\mathrm{cos}(\phi _1\phi _2))],`$
where $`M=1/2g\mu _BSN,\gamma =g\mu B/\mathrm{},K_u=NK_1S^2,K_{}=1/2NK_2S^2`$.
This Lagrangian can be derived by means of the standard technique of coherent quantum states (see for example ). All terms in (4) are of apparent physical meaning. The first term is called Wess-Zumino term (kinetic energy), the second term is Zeeman energy, the third and fourth term are the magnetic anisotropy and the fifth is the exchange interaction energy.
Furthermore it is convenient to use the new variables:
$$\theta =\frac{\theta _1+\theta _2}{2},\phi =\frac{\phi _1+\phi _2\pi }{2},$$
(5)
$$ϵ_1=\frac{\theta _1\theta _2}{2},ϵ_2=\frac{\phi _1\phi _2+\pi }{2},$$
(6)
Then the partition function of the system can be represented as the functional integral in the Eucledian space ($`\tau =it`$).
$$Z=Dϵ_1Dϵ_2D\theta D\phi e^{_0^\mathrm{}\beta 𝑑\tau L(\theta ,\phi ,ϵ_1,ϵ_2)}.$$
(7)
Upon integrating (7) over $`ϵ_1`$ and $`ϵ_2`$ one can obtain the following effective Lagrangian depending on the variable $`\phi (\tau )`$ only:
$$L=\frac{\chi _{}}{2}\left(\frac{\dot{\phi }}{\gamma }B_z\right)^2+K_{}\mathrm{cos}^2\phi .$$
(8)
The total magnetic moment of the antiferromagnetic molecule equals to:
$`M_z=\{\begin{array}{cc}\chi _{}\left(B_z\frac{\dot{\phi }}{\gamma }\right),& B_zB_c,\\ 2M,& B_zB_c,\end{array}`$ (11)
where
$$\chi _{}=\frac{Ng^2\mu _B^2}{4J},B_c=\frac{4JS}{g\mu _B}.$$
(12)
The following classical motion equation for variable $`\phi `$ can be put in correspondence to the Lagrangian (8):
$$\frac{\chi _{}}{\gamma ^2}\ddot{\phi }+\frac{\alpha M}{\gamma }\dot{\phi }+K_{}\mathrm{sin}2\phi \frac{\chi _{}\dot{B}_z}{2\gamma }=0.$$
(13)
Inserted here is the dissipative term $`\alpha M\dot{\phi }/\gamma `$ related to the attenuation occurence in Landau–Lifshits equations, where $`\alpha `$ is the dimensionless Hilbert constant.
For description of the thermodynamical and nonequilibrium properties of the molecule at finite temperature one can utilize the partition function as follows (see for example )
$$Z=𝑑\phi _0_{\phi (0)}^{\phi (\mathrm{}\beta )}D\phi (t)e^{\frac{S_{eff}\left[\phi (\tau )\right]}{\mathrm{}}},$$
(14)
where the effective action $`S_{eff}`$ equals to
$`S_{eff}\left[\phi (\tau )\right]`$ $`=`$ $`{\displaystyle _0^\mathrm{}\beta }d\tau L(\phi ,\dot{\phi },\tau )+{\displaystyle \frac{1}{2}}{\displaystyle _0^\mathrm{}\beta }d\tau {\displaystyle _0^\mathrm{}\beta }d\tau ^{}\alpha (\tau \tau )[\phi (\tau )\phi (\tau ^{})]^2,`$ (15)
$$\alpha (\tau )=\frac{\alpha M}{2\pi \gamma }\frac{\left(\frac{\pi }{\mathrm{}\beta }\right)^2}{\mathrm{sin}^2\left(\frac{\pi \tau }{\mathrm{}\beta }\right)}.$$
(16)
Henceforward we shall restrict our consideration to the only quantum properties (at $`T=`$ 0 K). The Hamiltonian of the system for the collective variable $`\phi `$ can be obtained from the Lagrangian (8) by the following procedure. The momentum $`p`$ corresponding to the independent coordinate $`\phi `$ equals to
$$p=\frac{L}{\dot{\phi }}=\frac{\chi _{}}{\gamma }\left(\frac{\dot{\phi }}{\gamma }B_z\right).$$
(17)
By further substituting this expression to the Hamilton function $`=p\dot{\phi }L`$ one can obtain
$$=\frac{1}{2\chi _{}}\left(\gamma p+\chi _{}B_z\right)^2\frac{\chi _{}B_z^2}{2}+U(\phi ).$$
(18)
By performing a standard quantization technique which consists in definition of the operators $`\widehat{p}`$ and $`\widehat{\phi }`$ by means of the commutation rule $`[\widehat{p},\widehat{\phi }]=i\mathrm{}`$ we obtain $`\widehat{p}=\frac{\mathrm{}}{i}\frac{}{\phi }`$. The magnetic momentum operator is determined as $`\widehat{M}_z=\gamma \widehat{p}`$. Substituting this expression into the Hamilton function (18) yields
$$=\frac{1}{2\chi _{}}\left(M_z\chi _{}B_z\right)^2\frac{\chi _{}B_z^2}{2}+U(\phi ).$$
(19)
The magnetic moment of the antiferromagnetic molecule and the angular variable $`\phi `$ are the conjugate variables: $`[\widehat{M}_z,\widehat{\phi }]=ig\mu _B`$.
Let’s consider first the case $`B_z=0`$. It is important that the equation (19) is isomorphous to the equation of particle in periodic potential, therefore one can use the results known for this model. The eigenstates of the Hamiltonian (19) are the Bloch functions
$$\mathrm{\Psi }_{sn}(\phi +\pi )=e^{i\pi m}\mathrm{\Psi }_{sm}(\phi ),$$
(20)
where $`m`$ is an arbitrary real number. By analogy with the term -“charge states” - for similar states in the theory of Josephson effect it is possible to call the states (20) as the “continuous spin states”. The Schroedinger equation for the Hamiltonian (19) is reduced to the Mathieu equation from the theory of which it follows that the energy spectrum of the Hamiltonian (19) has a band structure, i.e. the eigenvalues of (18) $`E_s(m)`$ are the functions defined in the appropriate Brillouin zones. At $`U(\phi )0`$ the band structure corresponds to the approximation of free electrons. In this case the energy spectrum is determined approximately by parabolic function
$$E_n(m)\frac{m^2\left(g\mu _B^2\right)}{2\chi _{}}$$
(21)
with the forbidden bands on the boundaries of the Brillouin zones: $`m_B=s\left(s=1,2,\mathrm{}S_{max}\right)`$, where $`S_{max}`$ is the maximal spin of the molecule. Near the Brillouin zone boundary the wave function can be represented as follows
$$\mathrm{\Psi }(\phi )=u(\phi )e^{iEt},$$
(22)
where $`u(\phi )=A_1e^{im\phi }+A_2e^{i(m+2)\phi }`$. The Schroedinger equation for the Hamiltonian (18) can be written as
$$u^{\prime \prime }+\left(\mu ^22b^2\mathrm{cos}2\stackrel{~}{\phi }\right)u=0,$$
(23)
where
$$\mu ^2=\frac{2\chi _{}E}{\left(g\mu _B\right)^2},b^2=\frac{\chi _{}K_{}}{2\left(g\mu _B\right)^2}.$$
(24)
Used here is a new variable $`\stackrel{~}{\phi }=\phi +\pi /2`$. The sign “$``$” will be further omitted. Substituting (22) to (19) yields
$$\mu ^2=\left(\frac{m^2+(m+2)^2}{2}\pm \sqrt{\frac{\left(m^2(m+2)^2\right)^2}{4}+b^4}\right).$$
(25)
In particular from the expression (25) it is clear that at $`m_B=1`$ the forbidden band width equals
$$\mathrm{\Delta }E_g=\frac{\left(g\mu _B\right)^2}{\chi _{}}b^2.$$
(26)
Equations (21),(23),(25) determine with sufficient accuracy the energy spectrum in the limit of first two Brillouin zones. The forbidden band between $`s1`$ and $`s`$ bands equals
$$\mathrm{\Delta }E_s\frac{g^2\mu _B^2}{2\chi _{}}\left(2b^2\right)^ss^{1s}.$$
(27)
If $`B_z0`$ in (19) it is possible to consider $`\chi _{}B_z`$ as a classical gauge field. The wave functions of the Hamiltonian (19) can be obtained from equation (20) by means of apparent gauge transformation. The gauge transformation can be presented as follows:
$$\mathrm{\Psi }_1(\phi )\mathrm{\Psi }(\phi )e^{\theta (\phi ,t)},$$
(28)
Then
$$i\mathrm{}\dot{\mathrm{\Psi }}_1=\left[\frac{1}{2\chi _{}}\left(\widehat{M}_zg\mu _B\frac{\theta }{\phi }\chi _{}B_z\right)^2u(\phi )\frac{\chi _{}B_z^2}{2}+\mathrm{}\dot{\theta }\right]\mathrm{\Psi }_1.$$
(29)
Assuming $`\theta (\phi ,t)=\frac{\chi _{}B_z\phi }{g\mu _B}`$, we have (the label “1” of the function $`\mathrm{\Psi }`$ will be further omitted)
$$i\mathrm{}\dot{\mathrm{\Psi }}=\left[\frac{M_z^2}{2\chi _{}}\frac{K_{}}{2}\mathrm{cos}2\phi \frac{\mathrm{}\chi _{}\dot{B}\phi }{g\mu _B}\frac{\chi _{}B_z^2}{2}\right]\mathrm{\Psi }.$$
(30)
The last term $`\frac{\mathrm{}\chi _{}\dot{B}\phi }{g\mu _B}`$ in equation (30) is playing the same role as energy – $`eFx`$ ($`F`$ characterizes the electrical field and $`x`$ is the electron coordinate) in the well-known problem of the dynamics of Bloch electron in electrical field.
Let consider the behavior of the molecule magnetization process under the action of time-dependent magnetic field $`B_z(t)`$ varying adiabatically slowly i.e.
$$\left|\frac{\chi _{}\dot{B}_z}{g\mu _B}\right|\frac{K_{}}{\mathrm{}}.$$
(31)
To describe the dynamics of the antiferromagnetic vector of the molecule (or “staggered magnetization”) let’s consider a wave packet of the Bloch functions (22). Let $`\overline{m}`$ and $`\overline{\phi }`$ be mean values of the generalized coordinate and momentum of the packet and the values $`\mathrm{\Delta }m`$ and $`\mathrm{\Delta }\phi `$ ($`\mathrm{\Delta }m\mathrm{\Delta }\phi 1`$) determine the packet width. Under influence of $`\dot{B}_z`$ which induces the rotational torque accelerating spin precession the wave packet moves to the boundary of Brillouin zone where Bragg reflection takes place. Here the wave packet group velocity changes its sign to an opposite after that a new process of the wave packet acceleration starts.
In mathematical terms this process can be described by the following equations:
$$\dot{\overline{m}}=\frac{\chi _{}}{g\mu _B}\dot{B},$$
(32)
$$\dot{\overline{\phi }}=\frac{1}{\mathrm{}}\frac{E_s(\overline{m})}{\overline{m}},$$
(33)
where $`E_s=E_s(\overline{m})`$ is energy spectrum of (19) for the s-band. Equations (32,33) must be completed by appropriate initial conditions.
In the adiabatic case the antiferromagnetic molecule remains in the state with the definite $`s`$ and all observed values as, for example, the magnetization (11), accounting (32,33) are the oscillating functions of time with the frequency
$$f_{Bloch}=\frac{\chi _{}\dot{B}_z}{g\mu _B}=\frac{g\mu _B\dot{B}_zN}{4J}.$$
(34)
These oscillations are identical to the oscillations that electrons are subjected in crystal or superlattice under electrical field (Bloch oscillations). This macroscopic quantum effect is essentially the macroscopic quantum coherence effect induced by time-increasing (decreasing) magnetic field.
If the external field can be represented by the following sum
$$B=B_0+at+b\mathrm{sin}2\pi ft,$$
(35)
then it is possible to expect the resonance origin at
$$f=f_{Bloch}$$
(36)
and on the frequences $`f=rf_{Bloch}`$ as well, where $`r`$ is a rational number (Stark ladder-like resonances). The constant shifting field $`B_0`$ plays here role of a chemical potential and can be used to provide optimal conditions for observing the considered macroscopic quantum effects.
Let’s present some numerical estimations. For Fe<sub>6</sub> and Fe<sub>10</sub> $`J=`$20.9 cm<sup>-1</sup> and $`J=9.6`$ cm<sup>-1</sup>, accordingly . The magnetic anisotropy is of the ”easy plane” type. The constants of anisotropy are $`K=`$ -0.3 cm<sup>-1</sup>/ion (Fe<sub>6</sub>) and $`K=`$ -0.1 cm<sup>-1</sup>/ion (Fe<sub>10</sub>). The estimation under the formulas (31-34) gives for Fe<sub>10</sub>: $`B_c=`$ 97 T; $`\omega _E=\mathrm{\hspace{0.17em}1.7}10^{13}`$ rad/s, $`f_{Bl}\mathrm{\hspace{0.17em}10}^5`$ Hz at $`\dot{B}_z=10^6`$ T/s.
Let’s examine briefly a role of dissipation. From equation (13) it follows, that the relaxation time of $`\phi `$ angular perturbations equals to $`\tau =(\alpha \omega _E)^1`$. For observation of the Bloch oscillations it is necessary that $`f_{Bloch}>\tau ^1`$ that restricts the value $`\dot{B}_z`$ from below. For the relaxation time estimation we shall take an advantage of the experimental data of paper for the antiferromagnet MnF<sub>2</sub> in which the exchange interaction constant, field $`B_c`$ and consequently $`\omega _E`$ are close to that for the antiferromagnetic molecules in the question. In the the linewidth of antiferromagnetic resonance $`\mathrm{\Delta }H`$ was measured which is linked to a Hilbert constant $`\alpha `$ and relaxation time by the relation $`\gamma \mathrm{\Delta }H=\alpha \omega _E=1/\tau `$, i.e. $`\tau =(\gamma \mathrm{\Delta }H)^1`$. In agreement to the temperature dependence $`\mathrm{\Delta }H(T)`$ at 5 K–40 K is an average between $`T^3`$ and $`T^4`$ with the tendency to $`T^4`$ (and even $`T^5`$) when approaching to $`T=`$ 5 K. At $`T=`$ 5 K $`\mathrm{\Delta }H`$ 0.1 Oe, therefore $`\tau \mathrm{\hspace{0.17em}0.5}10^6`$ s. It is possible to expect $`\tau >`$ 10<sup>-3</sup> s at 0.5 K.
On the other hand the condition (31) restricts from above the value $`\dot{B}_z`$ . Thus, the following inequality should be fulfilled:
$$\frac{4J}{Ng\mu _B\tau }\dot{B}_z\frac{\gamma K_{}}{\chi _{}}.$$
(37)
Accepting $`K_210^4J`$ we shall obtain following boundary values for an area of acceptable $`\dot{B}_z`$ values: $`10^4`$ T/s and $`10^{10}`$ T/s.
In conclusion, it was shown that the external time dependent magnetic field induces new macroscopic quantum effects in the behavior of the antiferromagnetic nanoclusters: oscillations of the magnetization similar to the electron Bloch oscillations in crystal and the corresponding resonances in ac-magnetic field. These effects are related to that of the macroscopic quantum coherence. The analogy of antiferromagnetic nanoclusters quantum behavior and Josephson junction was revealed as well.
Author thanks B. Barbara for discussions. This work was partially supported by RFBR (N 99-02-17830) and INTAS (N 97-705)
|
warning/0004/cond-mat0004478.html
|
ar5iv
|
text
|
# Correlation functions in the factorization approach of nonextensive quantum statistics
## I Introduction
The formulation of nonextensive statistical mechanics has raised numerous questions regarding its relevance to systems with long range interactions, long range microscopic memory, or multifractal properties . This formalism can also be understood as a generalization of Boltzmann-Gibbs statistics, opening the possibility to get a theoretical insight on the thermodynamics of systems whose behavior depart from Boltzmann-Gibbs statistical mechanics. The generalized entropy
$$S_q=\frac{k}{q1}\left(1\underset{R}{}p_R^q\right),$$
(1)
is a function of the probability $`p_R`$ for the ensemble to be in the state $`R`$ and a real parameter $`q`$. Equation (1) becomes the Shannon entropy as $`q1`$. In addition, Tsallis’ entropy shares all the properties of the Shannon entropy except that of additivity. Considering two independent systems $`\mathrm{\Sigma }`$ and $`\mathrm{\Sigma }^{}`$, the entropy $`S_q`$ satisfies the pseudoadditivity property
$$\frac{S_q^{\mathrm{\Sigma }\mathrm{\Sigma }^{}}}{k}=\frac{S_q^\mathrm{\Sigma }}{k}+\frac{S_q^\mathrm{\Sigma }^{}}{k}+(1q)\frac{S_q^\mathrm{\Sigma }}{k}\frac{S_q^\mathrm{\Sigma }^{}}{k}.$$
(2)
The probability distribution that results from extremizing the entropy with the constraints
$$\underset{R}{}p_R=1,$$
(3)
and
$$E=\underset{R}{}p_R^qE_R,$$
(4)
is given by the equation
$$p_R=\frac{\left[1+\beta (q1)(E_R\mu N)\right]^{1/(1q)}}{Z_q},$$
(5)
with the partition function
$$Z_q=\underset{R}{}\left[1+\beta (q1)(E_R\mu N)\right]^{1/(1q)},$$
(6)
and the total energy $`E_R=_jn_j\epsilon _j`$. Due to the mathematical complexity of the partition function in Equation (6), it has been necessary to study the validity of certain approximation procedures. Thus, the study of the consequences of nonextensivity has been mainly focused on either assuming a value of $`q1`$ or by approximating the partition function in Eq. (6) by a factorized partition function
$$Z=\underset{j=0}{}\underset{n_j=0}{}\left[1+\beta (q1)n_j(\epsilon _j\mu )\right]^{1/(q1)}.$$
(7)
The factorization approach has been shown to be a good approximation to Tsallis partition function outside certain narrow temperature interval that shifts to higher values of $`T`$ when the number of energy levels increases. An application to the Ising model and black body radiation can be found in Refs. and respectively. In this approximation the average number of particles with energy $`\epsilon `$ is given by the function
$$n=\frac{1}{\left[1+\beta (q1)(\epsilon \mu )\right]^{1/(q1)}+a},$$
(8)
where $`a=0,1,+1`$ for Maxwell-Boltzmann, Bose-Einstein and Fermi-Dirac cases respectively. It is important to remark that the function $`n`$ has also been obtained by extremizing the entropy related to the generalized dimensions of a fractal set. It has been also pointed out that nonextensive thermodynamics could also be understood in terms of $`q`$-deformations and possibly with the theory of quantum groups. Along this line of work, in Ref. we made a study of the basic thermodynamics that result from the particle distribution functions in Eq. (8) for classical and quantum gases. In that article we shown that the high temperature behavior is consistent with the thermodynamical limit provided that the internal energy is calculated according to the equation
$$U=\frac{4\pi V}{h^3}_0^{\mathrm{}}\frac{p^2}{2m}n(p)^qp^2𝑑p.$$
(9)
By introducing the usual creation and annihilation operator formalism we found that the boson hamiltonian that leads to the particle distribution, for $`a=1`$, in Eq.(8) is written as
$$\widehat{K}=\underset{j=0}{}(\epsilon _j\mu )\overline{\varphi }_j\varphi _j,$$
(10)
where the operator $`\varphi _j`$ and its adjoint $`\widehat{\varphi }_j`$ satisfy a deformed boson algebra. However, a comparison of the heat capacity and entropy functions, in Ref. , for systems with a particle distribution function as given by Eq. (8) with those for Bose and Fermi gases described by quantum group invariant hamiltonians shows that nonextensivity is unrelated to quantum group invariance.
In this paper we calculate the correlation function for boson systems with a particle distribution function as given by Eq. (8). Our main motivation in studying this system is twofold. First, it is of theoretical interest to study a thermodynamics system which obeys a statistical mechanics that generalizes Boltzmann-Gibbs statistics. Second, a calculation of the correlation functions will give us an insight on the long-range behavior dependence of these thermodynamical systems on the parameter $`q`$. Our calculations will show in a concrete fashion the relation between $`q`$ and long-range behavior, and will indicate whether the thermodynamics resulting from Eq. (8), proposed in Ref., has the long range behavior expected to be present in nonextensive thermodynamics. In Sec. II we calculate the correlation function for $`q1`$, and discuss the results for some particular values of this parameter. We specialize our discussion to the behavior near the critical temperature $`T_c`$ and at large values of $`T`$. In Sec. III we summarize our results.
## II Correlation functions
According to our previous work , the parameter $`q`$ cannot be any real number but its values are restricted to those such that $`1/(q1)`$ is an integer. In addition, the thermodynamic functions are well defined only in the interval $`1q1.5`$. The correlation for a Bose gas with particle distribution according to Eq.(8) is given by
$$G(𝐑)=\frac{1}{V}\underset{𝐤}{}\frac{e^{i𝐤𝐑}}{\left[1+\beta (q1)(\epsilon \mu )\right]^{1/(q1)}1},$$
(11)
As usual, this summation can be approximated by an integration over $`𝐤`$, leading after integration over the angles for $`d=3`$ to the equation
$$G(𝐑)=\frac{N_0}{V}+\frac{2}{\pi ^{1/4}R^{1/2}(\lambda \sqrt{q1})^{5/2}}_0^{\mathrm{}}x^{3/2}\underset{j=1}{\overset{\mathrm{}}{}}\left[1+\alpha (q1)+x^2\right]^{j/(q1)}J_{1/2}\left(2\pi ^{1/2}Rx/\lambda \sqrt{q1}\right)dx,$$
(12)
where the new variable $`x^2=\beta (q1)k^2\mathrm{}^2/2m`$ and, as usual, $`\alpha =\beta \mu `$ and $`\lambda `$ is the thermal wavelength. The integral in Eq. (12) is tabulated
$$_0^{\mathrm{}}\frac{J_\nu (bx)}{(x^2+c^2)^{\mu +1}}x^{\nu +1}𝑑x=\frac{c^{\nu \mu }b^\mu }{2^\mu \mathrm{\Gamma }(\mu +1)}K_{\nu \mu }(cb),$$
(13)
where $`1<Re\nu <Re(2\mu +3/2)`$, $`c>0`$ and $`b>0`$. It is simple to check that all these inequalities are satisfied for all the allowed values of $`q`$ in the interval $`1q1.5`$. Defining a length parameter $`\chi <\lambda `$
$$\chi =\frac{\lambda }{2\sqrt{\pi }}\sqrt{\frac{q1}{1+\alpha (q1)}},$$
(14)
and by use of the integral representation
$$K_\nu (zx)=\frac{\mathrm{\Gamma }(\nu +1/2)}{x^\nu \mathrm{\Gamma }(1/2)}(2z)^\nu _O^{\mathrm{}}\frac{\mathrm{cos}(xt)}{(t^2+z^2)^{\nu +1/2}}𝑑t,$$
(15)
the correlation function reduces to the expression
$$G(𝐑)=\frac{1}{\pi R\lambda ^2(q1)}_0^{\mathrm{}}\mathrm{cos}(t)\underset{n=1}{}\left[\frac{(R/\chi )^2}{(1+\alpha (q1))(t^2+(R/\chi )^2)}\right]^{n/(q1)1}dt.$$
(16)
Once we perform the summation we get a set of integrals in terms of elementary functions. These integrals are all tabulated and the correlation function for some values of $`q`$ follows
$$G(𝐑)=\{\begin{array}{ccc}(1/R\lambda ^2)\left[e^{R/\xi }e^{\sqrt{(R/\xi )^2+32\pi (R/\lambda )^2}}2e^A^{}\mathrm{sin}(B^{})\right]\hfill & \text{for }q=5/4\hfill & \\ & & \\ (1/R\lambda ^2)\left[e^{R/\xi }e^A\mathrm{cos}(B)\sqrt{3}e^A\mathrm{sin}(B)\right]\hfill & \text{for }q=4/3\hfill & \\ & & \\ (1/R\lambda ^2)\left[e^{R/\xi }e^{\sqrt{(R/\xi )^2+16\pi (R/\lambda )^2}}\right]\hfill & \text{for }q=3/2\hfill & \end{array},$$
(17)
where $`\xi =\lambda /2\sqrt{\pi \alpha }`$ is the correlation length and the exponents
$`A^2`$ $`=`$ $`12\pi (R/\lambda )^2\left[(1/2)\sqrt{(3/2+\alpha /3)^2+(3/4)}+(3/2+\alpha /3)\right]`$ (18)
$`B^2`$ $`=`$ $`12\pi (R/\lambda )^2\left[(1/2)\sqrt{(3/2+\alpha /3)^2+(3/4)}(3/2+\alpha /3)\right]`$ (19)
$`A^2`$ $`=`$ $`16\pi (R/\lambda )^2\left[(1/2)\sqrt{(1+\alpha /4)^2+1}+(1+\alpha /4)\right]`$ (20)
$`B^2`$ $`=`$ $`16\pi (R/\lambda )^2\left[(1/2)\sqrt{(1+\alpha /4)^2+1}(1+\alpha /4)\right]`$ (21)
From these equations it is clear that the correlation functions for $`q1`$ have a leading term
$$G(𝐑)(1/R\lambda ^2)e^{R/\xi },$$
(22)
minus some smaller additional terms which are absent in the Bose-Einstein case.
### A Correlation length
In order to compare these correlation functions with the standard correlation function $`G_{BE}(𝐑)`$ we first look at the case of high temperature $`T>>T_c`$. Although Eq. (22) has the same functional relation than
$$G_{BE}(𝐑)(1/R\lambda ^2)e^{R/\xi _{BE}},$$
(23)
the parameter $`\alpha `$ is the function
$$\alpha =\frac{1}{q1}\left[1+\left(\frac{2}{\sqrt{\pi }n(q1)^{3/2}\lambda ^3}S_2(q)\right)^{2(q1)/(53q)}\right],$$
(24)
with
$$S_2(q)=\frac{1}{3/21/(q1)}+\underset{m=1}{\overset{\mathrm{}}{}}\frac{(1)^m}{m!}(1/2)\mathrm{}(3/2m)\frac{1}{3/2m1/(q1)}.$$
(25)
In terms of the critical temperature $`T_c`$ the correlation length $`\xi `$ for high temperatures is given by the expression
$$\frac{1}{\xi }=2\sqrt{\frac{\pi }{q1}}\left(\frac{n}{G_{3/2}(1,q)}\right)^{1/3}\left(\frac{T}{T_c}\right)^{1/2}\left[1+\left(\frac{2}{\sqrt{\pi }G_{3/2}(1,q)(q1)^{3/2}}\left(\frac{T}{T_c}\right)^{3/2}S_2(q)\right)^{2(q1)/(53q)}\right]^{1/2},$$
(26)
where the function $`G_{3/2}(z=1,q)`$ was calculated in Ref. . In particular $`G_{3/2}(1,1)=2.612`$ and $`G_{3/2}(1,q)`$ increases with the value of $`q`$. A simple inspection shows that the correlation length $`\xi `$ is smaller than the correlation length for the standard Bose-Einstein case $`\xi ^{BE}`$
$$\frac{1}{\xi ^{BE}}=2\sqrt{\pi }\left(\frac{n}{2.612}\right)^{1/3}\left(\frac{T}{T_c^{BE}}\right)^{1/2}\mathrm{ln}^{1/2}\left[2.612(T/T_c^{BE})^{3/2}\right],$$
(27)
where $`T_c<T_c^{BE}`$. Figure 1 shows a comparison between these two functions for several values of the parameter $`q`$. Clearly, the correlation function decreases more rapidly as $`q`$ increases from the standard value $`q=1`$. At the critical temperature, $`\xi \mathrm{}`$ and the correlations become
$$G_c(𝐑)\frac{1}{R\lambda _c^2},$$
(28)
which, due to the inequality $`\lambda _c>\lambda _c^{BE}`$, is smaller than $`G_c^{BE}`$. Figure 2 shows $`\alpha `$ for $`TT_{critical}`$ and some values of $`q`$ in comparison to the $`q=1`$ case. Since, at $`T>T_c`$, $`\alpha `$ is smaller for $`q=1`$ we find that the correlation function for all temperatures is larger for the $`q=1`$ case.
### B Critical behavior
In order to study the behavior at the critical temperature we need to expand the function
$$G_{3/2}(z,q)=\frac{2}{\sqrt{\pi }(q1)^{3/2}}_0^{\mathrm{}}\frac{y^{1/2}}{\left[1+y(q1)\mathrm{ln}z\right]^{1/(q1)}1}𝑑y,$$
(29)
in powers of $`\alpha `$. It is clear from the previous discussion that $`G_{3/2}(z,q)`$ becomes the well known Bose-Einstein function $`g_{3/2}(z)`$ in the limit $`q1`$. For these purposes we apply the same method used in Ref., which gives a power series of $`\alpha `$ by performing first a Mellin transformation of the function and then applying the corresponding inverse transform. The Mellin transform of $`G_{3/2}(z,q)`$ is given by the equation
$`F_{3/2}(s)`$ $`=`$ $`{\displaystyle _0^{\mathrm{}}}G_{3/2}(z,q)\alpha ^{s1}𝑑\alpha ,`$ (30)
$`=`$ $`{\displaystyle \frac{\mathrm{\Gamma }(s)}{(q1)^{3/2+s}}}{\displaystyle \underset{m=1}{}}{\displaystyle \frac{\mathrm{\Gamma }\left(m/(q1)(3/2)s\right)}{\mathrm{\Gamma }\left(m/(q1)\right)}},`$ (31)
and the inverse transformation
$$G_{3/2}(z,q)=\frac{1}{2\pi i}_{ci\mathrm{}}^{c+i\mathrm{}}\frac{1}{(q1)^{3/2+s}}\mathrm{\Gamma }(s)\underset{m=1}{}\frac{\mathrm{\Gamma }\left(m/(q1)(3/2)s\right)}{\mathrm{\Gamma }\left(m/(q1)\right)}\alpha ^sds,$$
(32)
where the contour of integration closes on the left half plane. With use of the approximation
$$\frac{\mathrm{\Gamma }\left(m/(q1)(3/2)s\right)}{\mathrm{\Gamma }\left(m/(q1)\right)}=\frac{1}{(m/(q1))^{3/2+s}}\left[1+\frac{b}{m/(q1)}+\frac{c}{(m/(q1))^2}+\mathrm{}\right],$$
(33)
with the constants
$`b`$ $`=`$ $`{\displaystyle \frac{1}{2}}(3/2+s)(3/2+s+1),`$ (34)
$`c`$ $`=`$ $`{\displaystyle \frac{1}{12}}{\displaystyle \frac{\mathrm{\Gamma }(3/2s+1)}{\mathrm{\Gamma }(3)\mathrm{\Gamma }(3/2s1)}}(3(5/2+s)^2+s+1/2),`$ (35)
the function $`G_{3/2}(z,q)`$ is written
$`G_{3/2}(z,q)`$ $`=`$ $`{\displaystyle \frac{1}{2\pi i}}{\displaystyle _{ci\mathrm{}}^{c+i\mathrm{}}}{\displaystyle \frac{\mathrm{\Gamma }(s)\alpha ^s}{(q1)^{3/2+s}}}[(q1)^{3/2+s}\zeta (3/2+s)+b(q1)^{5/2+s}\zeta (5/2+s)`$ (36)
$`+`$ $`c(q1)^{7/2+s}\zeta (7/2+s)+\mathrm{}]ds.`$ (37)
The function $`\mathrm{\Gamma }(s)`$ has simple poles at $`s=n`$ with residues $`(1)^n/n!`$ and the zeta function $`\zeta (w)`$ has a simple pole at $`w=1`$ with residue equal to $`+1`$. With use of the residue theorem the function $`G_{3/2}(z,q)`$ is expressed as a power series of $`\alpha `$ as follows
$$G_{3/2}(z,q)=\frac{1}{(q1)^{3/2}}\underset{m=1}{}\underset{n=0}{}\frac{(\alpha (q1))^n}{n!}\frac{\mathrm{\Gamma }(m/(q1)3/2+n)}{\mathrm{\Gamma }(m/(q1))}+\mathrm{\Gamma }(1/2)\alpha ^{1/2}\frac{1}{6}(q1)^2\mathrm{\Gamma }(5/2)\alpha ^{5/2}+O(\alpha ^{7/2}).$$
(38)
A simple check shows that the series with integer powers of $`\alpha `$ reduces, as $`q1`$, to the result in Ref. $`_{n=0}(\alpha )^n\zeta (3/2n)/n!`$. At lowest order in $`\alpha `$ we obtain
$$G_{3/2}(z,q)\frac{1}{(q1)^{3/2}}\underset{m=1}{}\frac{\mathrm{\Gamma }(m/(q1)3/2)}{\mathrm{\Gamma }(m/(q1))}+\mathrm{\Gamma }(1/2)\alpha ^{1/2}.$$
(39)
The series in equation (39) approximates at lowest order to the function $`\zeta (3/2)`$ and the last term is identical to the standard, $`q=1`$, result. Therefore, since for $`\alpha 0`$ we have $`G_{3/2}(z,q)\alpha ^{1/2}`$, defining $`t=(TT_c)/T_c`$ we find that $`\alpha t^2`$. In addition, the correlation length $`\xi \lambda t^1`$, as the standard case. Since, near the critical temperature, the $`t`$ dependence of the functions $`G_\nu (z,q)`$ is the same as the case of the Bose-Einstein functions $`g_\nu (z)`$ and the thermodynamic functions of the two systems have the same functional form in terms of $`G_\nu `$ and $`g_\nu `$, the remaining critical exponents are also independent of $`q`$.
## III Conclusions
In this paper our main concern has been to study the long range behavior predicted by thermodynamical systems described by a factorized partition function . As pointed out in the Introduction, this factorized partition function has been shown to approximate well the partition function of nonextensive statistics. A calculation of the correlation functions indicate that a Bose gas for the $`q1`$ case is less correlated than for $`q=1`$. In particular, we showed that the correlation length for $`q=1`$ is larger than for $`q1`$ at all temperatures , and the critical exponents are independent of $`q`$. Certainly, our calculations can draw conclusions on the long range behavior of a thermodynamical system in the factorization approach only. On the other hand, correlations in nonextensive thermodynamics should be studied with use of the Tsallis’ partition function. The fact that a Bose-Einstein gas obeying the studied factorized partition function is less correlated than for the Boltzmann-Gibbs case implies that the long range behavior expected in nonextensive thermodynamics is lost when the Tsallis’ partition function is replaced by a factorized one. Thus, our results point out that the errors introduced by forcing factorization, which could be not significant for the evaluation of thermodynamic functions, are large enough that the long range behavior of Tsallis thermostatistics cannot be studied within this factorization approach.
FIG. 1. The function $`f=2\pi ^{1/2}n^{1/3}\xi `$, where $`n`$ is the average number of particles per volume and $`\xi `$ is the correlation length at high temperatures for the cases of $`q=1,6/5,3/2`$ as a function of $`T/T_{critical}`$. $`T_{critical}`$ refers to the critical temperature for each value of $`q`$. This graph shows that the correlation length decreases as the value of $`q`$ increases.
FIG. 2. The behavior of $`\alpha =\beta \mu `$ as a function of the temperature for the cases $`q=1,6/5,3/2`$, indicating that for low temperatures a Bose-Einstein gas is more strongly correlated for $`q=1`$.
|
warning/0004/quant-ph0004087.html
|
ar5iv
|
text
|
# Generalised coherent states for 𝑆𝑈(𝑛) systems
## I Introduction
Coherent states were originally constructed and developed for the Heisenberg–Weyl group to investigate quantized electromagnetic radiation . These coherent states were generated by the action of the Heisenberg–Weyl group operators on the vacuum state which led to group theoretic generalizations by Peleromov and Gilmore . These two mathematical frameworks differ in some points, such as the representations of groups and the reference states; these differences are summarized in . Coherent states for $`SU(2)`$ have also been called atomic coherent states , and have been found useful for treating atom systems, and also for investigations of quantum optical models such as nonlinear rotators . $`SU(2)`$ coherent states have been successfully applied to the analysis of the classical limit of quantum systems, and more recently, to the investigations of nonlinear quantum systems and quantum entanglement .
In spite of these many successful $`SU(2)`$ coherent state applications, not much work has been done towards generalising the analysis to other $`SU(n)`$ groups, although $`SU(3)`$ symmetries were employed to treat a schematic nuclear shell model . More recently, this lack has been addressed for $`SU(3)`$ systems with the explicit construction of the $`SU(3)`$ coherent states , the calculation of Clebsch–Gordon coefficients , and the investigation of Wigner functions . Further, geometrical phases for $`SU(3)`$ systems have been discussed in . These developments for $`SU(3)`$ are technologically useful and allow treatment of more complex quantum systems such as coupled Bose–Einstein condensates .
In this paper, we construct a set of explicit coherent states for $`SU(n)`$, and apply group theoretic techniques to facilitate the investigation of nonlinear quantum systems and quantum entanglement. In order to construct explicit coherent states, we need to specify the group representation and the reference states. For the chosen group representation, it is necessary to show a useful decomposition and a parameterization giving usable expressions for the coherent states. Formal approaches to the definition of coherent states are often not readily applicable. For instance, while the Baker–Campbell–Hausdorff relation derived for $`SU(n)`$ can be used to define coherent states, this approach does not yield explicit formulae and parameterisations.
In this paper, we employ the decomposition for $`SU(n)`$ in and exploit its symmetric parameterisation. A set of coherent states of $`SU(n)`$ is called an orbit, and is produced by the action of group elements on a reference state which here is chosen to be the highest weight state. For instance, for $`SU(2)`$ the highest weight state for a spin 1/2 system is spin-up, and the orbit is the surface of a 3-sphere. For general $`n`$, this orbit corresponds to a $`(2n1)`$–sphere, which is isomorphic to the coset space $`SU(n)/SU(n1)`$. The geometrical properties of this coset space generalise the $`SU(3)`$ properties described in . The coset space considered here, $`SU(n)/SU(n1)`$, differs slightly from the coset space normally considered for coherent states, $`SU(n)/U(n1)`$, by including an arbitrary phase. The coset space $`SU(n)/SU(n1)`$ enables us to provide a more general method to construct coherent states. Developing the representations and decompositions of higher rank groups becomes rapidly messy, however the decomposition in leads to a systematic procedure for the derivation of the coherent states on the coset space $`SU(n)/SU(n1)`$ without additional complexity. Thence we can easily extract an arbitrary phase carrying no physical significance for application to physical systems.
In Section II, we obtain an iterative equation in $`SU(n)`$ coherent states for the simplest irreducible unitary representation of $`SU(n)`$. We also show the geometrical structure of the coset space $`SU(n)/SU(n1)`$, and provide the metric and measure on the space. In Section III, our analysis is generalised to the case of finding coherent states of irreducible unitary representations for arbitrarily large dimension, and parametric representations are derived. We also show some properties of the coherent states. Finally, we summarize our results in Section IV.
## II Decomposition and coset spaces for fundamental representations of $`SU(n)`$
In order to construct the $`SU(n)`$ coherent states for the fundamental $`n\times n`$ matrix representation, we first specify the reference state $`|\varphi _0`$ as $`(1,0,\mathrm{},0)^T`$, where $`T`$ denotes transpose. This state is a highest weight state, in the sense that it is annihilated by each of the $`SU(n)`$ raising operators. The raising (lowering) operators $`J_j^h`$ are equivalent to elementary matrices $`e_j^h,h<j`$ ($`h>j`$) in the $`n\times n`$ matrix representation. Appendix A shows the commutation relations of these matrices. In this section we review the construction of $`SU(2)`$ coherent states, which provides the origin of the recursive relation of the $`SU(n)`$ coherent states. We then derive the displacement operators for $`SU(3)`$ and $`SU(4)`$, employing the $`n\times n`$ matrix representation of . Finally our results are extended to the $`SU(n)`$ case.
### A Review of $`SU(2)`$
Elements $`gSU(2)`$ in the fundamental $`2\times 2`$ matrix representation of $`SU(2)`$ may be parameterized as
$$g(\theta ,\phi _1,\phi _2)=\left(\begin{array}{cc}e^{i\phi _1}\mathrm{cos}\theta & e^{i\phi _2}\mathrm{sin}\theta \\ e^{i\phi _2}\mathrm{sin}\theta & e^{i\phi _1}\mathrm{cos}\theta \end{array}\right),$$
(1)
where angles are real and lie on $`0\theta \pi /2`$ and $`0\phi _1`$, $`\phi _22\pi `$ respectively. The standard approach to obtain a set of coherent states corresponding to $`SU(2)/SU(1)`$ begins with the decomposition of this matrix as
$$g(\theta ^{},\phi _1,\phi _2)=\left(\begin{array}{cc}\mathrm{cos}\frac{\theta ^{}}{2}& e^{i(\phi _2\phi _1)}\mathrm{sin}\frac{\theta ^{}}{2}\\ e^{i(\phi _2\phi _1)}\mathrm{sin}\frac{\theta ^{}}{2}& \mathrm{cos}\frac{\theta ^{}}{2}\end{array}\right)\left(\begin{array}{cc}e^{i\phi _1}& 0\\ 0& e^{i\phi _1}\end{array}\right),$$
(2)
where the new variable $`\theta ^{}`$ is introduced as $`\theta ^{}=2\theta `$. The action of the right matrix on the highest weight vector $`|\varphi _0=\left(\begin{array}{c}1\\ 0\end{array}\right)`$ changes only a phase factor and does not otherwise change the highest weight vector. Absolute phases $`\phi _1`$ and $`\phi _2`$ by themselves do not carry physical significance, and only their difference is physically important. This allows the removal of an arbitrary phase usually done by setting $`e^{i\phi _1}=1`$. Now the action of the left matrix on the highest weight state gives us coherent states $`|𝐧_2^{}=g|\varphi _0=\left(\begin{array}{c}\mathrm{cos}\frac{\theta ^{}}{2}\\ e^{i\phi _2}\mathrm{sin}\frac{\theta ^{}}{2}\end{array}\right),`$ which corresponds to the 2–sphere, i.e. the surface of three–dimensional ball, with the unit vector $`(\mathrm{cos}\theta ^{},e^{i\phi _2}\mathrm{sin}\theta ^{})`$, or $`(\mathrm{cos}\theta ^{},\mathrm{sin}\theta ^{}\mathrm{cos}\phi _2,\mathrm{sin}\theta ^{}\mathrm{cos}\phi _2)`$ in real coordinates, and the measure $`d\mu _2^{}=\mathrm{sin}\theta ^{}d\theta ^{}d\phi _2`$ . However this method is not very convenient to construct coherent states for general $`SU(n)`$. The decomposition of $`SU(n)`$ equivalent to (2) is not trivial especially for larger $`n`$, and is dependent on the choice of which arbitrary phase is extracted. In this paper, to avoid using the equivalent decomposition to (2), we begin more generally with the parameterisation (1), derive coherent states and then easily remove an arbitrary phase from our $`SU(n)`$ coherent states.
We now apply $`g(\theta ,\phi _1,\phi _2)`$ of (1) to the highest weight state $`|\varphi _0`$. The action yields the $`SU(2)`$ coherent states
$$|𝐧_2=g|\varphi _0=\left(\begin{array}{c}e^{i\phi _1}\mathrm{cos}\theta \\ e^{i\phi _2}\mathrm{sin}\theta \end{array}\right),$$
(3)
which correspond to points on a 3–sphere with unit vector $`(e^{i\phi _1}\mathrm{cos}\theta ,e^{i\phi _2}\mathrm{sin}\theta )`$, from which we derive the expression for the metric on the sphere
$$|ds_2|^2=d\theta ^2+\mathrm{cos}^2(\theta )d\phi _1^2+\mathrm{sin}^2(\theta )d\phi _2^2,$$
(4)
and the measure associated with this metric as
$$d\mu _2=\mathrm{cos}\theta \mathrm{sin}\theta d\theta d\phi _1d\phi _2.$$
(5)
From this set of coherent states, we now give the procedure to obtain the $`SU(2)`$ coherent states $`|𝐧_2^{}`$. This is done by setting $`e^{i\phi _1}=1`$ and introducing the variable $`\theta ^{}`$ so the coherent states now correspond to 2–sphere. This shows that one can readily remove an arbitrary phase from our $`SU(n)`$ coherent states without changing the decomposition of the group representation.
For the convenience of the later use, we here introduce $`\lambda `$–matrices and their parameterisation of $`SU(2)`$. $`g`$ may also be parameterised with using $`\lambda `$–matrices (i.e. Pauli matrices) $`\lambda _1=\sigma _1=\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right)`$, $`\lambda _2=\sigma _2=\left(\begin{array}{cc}0& i\\ i& 0\end{array}\right)`$, and $`\lambda _3=\sigma _3=\left(\begin{array}{cc}1& 0\\ 0& 1\end{array}\right)`$, as
$$g=e^{i\alpha \lambda _3}e^{i\beta \lambda _2}e^{i\gamma \lambda _3},$$
(6)
where $`\phi _1=\alpha +\gamma `$, $`\phi _2=\alpha +\gamma `$, and $`\theta =\beta `$, viz $`\alpha =\frac{1}{2}(\phi _1\phi _2)`$, $`\beta =\theta `$, and $`\gamma =\frac{1}{2}(\phi _1+\phi _2)`$.
### B Structure of $`SU(n)`$ for arbitrary $`n`$
We here employ the symmetric parameterisation for the $`SU(n)`$ matrices provided in to obtain an iterative equation for the $`SU(n)`$ coherent states. This matrix representation efficiently yields the orbit of the highest weight state $`(1,0,\mathrm{},0)^T`$, because of its symmetric decomposition. The parameterisation influences the structure of the iterative equation, which we demonstrate by example for small $`n`$. We derive firstly the $`SU(3)`$ coherent states (which may be compared with the simplest case of ), and secondly the $`SU(4)`$ coherent states. For each example, the expression for the coherent states shows their geometrical structure and determines the metric and measure of the coset space isomorphic to the coherent states. These examples are then generalised to $`SU(n)`$ by determining the iterative equation for the $`SU(n)`$ coherent states. Lastly, we give the measure of the coset space $`SU(n)/SU(n1)`$.
An arbitrary element $`gSU(n)`$ in the $`n\times n`$ matrix representation may be parameterised as
$`g`$ $`=`$ $`\left(\begin{array}{cccc}1& 0& \mathrm{}& 0\\ 0& & & \\ \mathrm{}& & X_{n1}& \\ 0& & & \end{array}\right)\left(\begin{array}{cccc}e^{i\phi }\mathrm{cos}\theta & & \mathrm{sin}\theta & \\ \mathrm{sin}\theta & & e^{i\phi }\mathrm{cos}\theta & 0\\ & & & \\ & 0& & I_{n2}\end{array}\right)\left(\begin{array}{cccc}1& 0& \mathrm{}& 0\\ 0& & & \\ \mathrm{}& & Y_{n1}& \\ 0& & & \end{array}\right)`$ (19)
$`=`$ $`L_{n1}M(\theta ,\phi )R_{n1}`$ (20)
where $`X_{n1},Y_{n1}`$ are the appropriate $`(n1)\times (n1)`$ matrices representing elements of $`SU(n1)`$, and $`I_k`$ is the $`k\times k`$ identity matrix, and we have defined three matrices $`L_{n1}`$, $`M`$, $`R_{n1}`$ for convenience. (See Appendix B for a justification of this parameterisation.)
#### 1 Structure of $`SU(3)`$
For $`SU(3)`$, since the matrices $`X_{n1},Y_{n1}`$ may be parameterised as (1), (20) gives
$`g(\phi _1,\xi _1,\phi _2,\phi ,\theta ,\phi _3,\xi _2,\phi _4)`$ (21)
$`=`$ $`\left(\begin{array}{ccc}1& 0& 0\\ 0& e^{i\phi _1}\mathrm{cos}\xi _1& e^{i\phi _2}\mathrm{sin}\xi _1\\ 0& e^{i\phi _2}\mathrm{sin}\xi _1& e^{i\phi _1}\mathrm{cos}\xi _1\end{array}\right)\left(\begin{array}{ccc}e^{i\phi }\mathrm{cos}\theta & \mathrm{sin}\theta & 0\\ \mathrm{sin}\theta & e^{i\phi }\mathrm{cos}\theta & 0\\ 0& 0& 1\end{array}\right)`$ (32)
$`\left(\begin{array}{ccc}1& 0& 0\\ 0& e^{i\phi _3}\mathrm{cos}\xi _2& e^{i\phi _4}\mathrm{sin}\xi _2\\ 0& e^{i\phi _4}\mathrm{sin}\xi _2& e^{i\phi _3}\mathrm{cos}\xi _2\end{array}\right)`$
$`=`$ $`L_2M(\theta ,\phi )R_2.`$ (33)
We take the highest weight state $`(1,0,0)^T`$ as the reference state and now obtain the expression of the orbit. Noting that the right matrix $`R_2`$ does not change the reference state, the displacement operator for the $`SU(3)`$ coherent states is the product of the left and middle matrices, $`L_2M(\theta ,\phi )`$. The left matrix $`L_2`$ corresponds to $`SU(2)`$, hence the orbit of the reference state is the coset space $`SU(3)/SU(2)`$. The first column of the middle matrix $`M(\theta ,\phi )`$ and the first and second columns of the left matrix $`L_2`$ in Eq. (33) can change the reference state, giving
$`|𝐧_3g|\varphi _0`$ $`=`$ $`L_2\left(\begin{array}{c}e^{i\phi }\mathrm{cos}\theta \\ \mathrm{sin}\theta \\ 0\end{array}\right)`$ (37)
$`=`$ $`\left(\begin{array}{ccc}1& 0& 0\\ 0& 0& 0\\ 0& 0& 0\end{array}\right)\left(\begin{array}{c}e^{i\phi }\mathrm{cos}\theta \\ 0\\ 0\end{array}\right)+\left(\begin{array}{ccc}1& 0& 0\\ 0& e^{i\phi _1}\mathrm{cos}\xi _1& e^{i\phi _2}\mathrm{sin}\xi _1\\ 0& e^{i\phi _2}\mathrm{sin}\xi _1& e^{i\phi _1}\mathrm{cos}\xi _1\end{array}\right)\left(\begin{array}{c}0\\ \mathrm{sin}\theta \\ 0\end{array}\right)`$ (50)
$`=`$ $`\left(\begin{array}{c}e^{i\phi }\mathrm{cos}\theta \\ 0\\ 0\end{array}\right)+\mathrm{sin}\theta \left(\begin{array}{c}0\\ \begin{array}{cc}& \\ |𝐧_2& \end{array}\end{array}\right).`$ (58)
The state $`|𝐧_2`$ is the $`SU(2)`$ coherent state given in (3), hence the coset space is isomorphic to a 5–sphere which has unit normal
$$𝐧_3=(e^{i\phi }\mathrm{cos}\theta ,e^{i\phi _1}\mathrm{sin}\theta \mathrm{cos}\xi _1,e^{i\phi _2}\mathrm{sin}\theta \mathrm{sin}\xi _1),$$
(59)
metric
$$|ds_3|^2=d\theta ^2+\mathrm{cos}^2\theta d\phi ^2+\mathrm{sin}^2\theta (d\xi _1^2+\mathrm{cos}^2\xi _1d\phi _1^2+\mathrm{sin}^2\xi _1d\phi _2^2),$$
(60)
and measure
$$d\mu _3=\mathrm{cos}\theta \mathrm{sin}^3\theta \mathrm{cos}\xi _1\mathrm{sin}\xi _1d\theta d\xi _1d\phi d\phi _1d\phi _2.$$
(61)
We note here that an arbitrary phase in these coherent states for $`SU(3)`$ can be easily removed as discussed in Subsection (II A), and this process is also applicable to general $`SU(n)`$ cases.
$`SU(3)`$ may also be decomposed using $`\lambda `$–matrices , which yields a slightly different parameterisation from (33). The $`SU(2)`$ $`\lambda `$–matrix decomposition for each matrix in (33) gives the $`\lambda `$–matrix expression for the $`SU(3)`$ coherent states. The middle matrix of (33) may be constructed using $`\lambda _2`$ and $`\lambda _3`$ as
$$M(\theta ,\phi )=e^{i\phi /2\lambda _3}e^{i\theta \lambda _2}e^{i\phi /2\lambda _3}.$$
(62)
The left matrix of (33) may be expressed by $`\lambda _7`$, $`\lambda _8`$, and $`\lambda _3`$. For convenience, we define a matrix $`\lambda _8^{}`$ as $`(\sqrt{3}\lambda _8\lambda _3)/2`$, which describes the $`SU(2)`$ diagonal generator $`\sigma _3`$ in the bottom right corner of the $`SU(3)`$ matrix $`\left(\begin{array}{cc}1& 0\\ 0& \sigma _3\end{array}\right)`$. The left matrix is
$`L_2`$ $`=`$ $`e^{i\alpha \lambda _8^{}}e^{i\beta \lambda _7}e^{i\gamma \lambda _8^{}}`$ (63)
$`=`$ $`e^{i\sqrt{3}\alpha /2\lambda _8}e^{i\alpha /2\lambda _3}e^{i\beta \lambda _7}e^{i\sqrt{3}\gamma /2\lambda _8}e^{i\gamma /2\lambda _3},`$ (64)
where $`\phi _1=\alpha +\gamma `$, $`\phi _2=\alpha +\gamma `$, and $`\xi _1=\beta `$. These two expressions (62) and (64) give
$`L_2M(\theta ,\phi )`$ $`=`$ $`e^{i\alpha \lambda _8^{}}e^{i\beta \lambda _7}e^{i\gamma \lambda _8^{}}e^{i\phi /2\lambda _3}e^{i\theta \lambda _2}e^{i\phi /2\lambda _3}`$ (65)
$`=`$ $`e^{i\sqrt{3}\alpha /2\lambda _8}e^{i\alpha /2\lambda _3}e^{i\beta \lambda _7}e^{i\sqrt{3}\gamma /2\lambda _8}e^{i\gamma /2\lambda _3}e^{i\phi /2\lambda _3}e^{i\theta \lambda _2}e^{i\phi /2\lambda _3}`$ (66)
The coherent states $`|𝐧_3`$ in this representation are thus
$`|𝐧_3`$ $`=e^{i\alpha \lambda _8^{}}e^{i\beta \lambda _7}e^{i\gamma \lambda _8^{}}e^{i\phi /2\lambda _3}e^{i\theta \lambda _2}e^{i\phi /2\lambda _3}|\varphi _0`$ (68)
$`=e^{i\sqrt{3}\alpha /2\lambda _8}e^{i\alpha /2\lambda _3}e^{i\beta \lambda _7}e^{i\sqrt{3}\gamma /2\lambda _8}e^{i\gamma /2\lambda _3}e^{i\phi /2\lambda _3}e^{i\theta \lambda _2}e^{i\phi /2\lambda _3}|\varphi _0.`$
#### 2 $`SU(4)`$ and $`SU(n)`$ for arbitrary $`n`$
Next we obtain the $`SU(4)`$ coherent states by applying the above procedure to the $`SU(3)`$ coherent states. This process shows the iterative structure of the $`SU(n)`$ coherent states, which allows us to define generalised coherent states for arbitrary $`n`$. An arbitrary element $`gSU(4)`$ can be factored by (20) as
$`g`$ $`=`$ $`L_3M(\theta ,\phi )R_3`$ (69)
$`=`$ $`\left(\begin{array}{cccc}1& 0& \mathrm{}& 0\\ 0& & & \\ \mathrm{}& & X_3& \\ 0& & & \end{array}\right)\left(\begin{array}{cccc}e^{i\phi }\mathrm{cos}\theta & & \mathrm{sin}\theta & 0\\ \mathrm{sin}\theta & & e^{i\phi }\mathrm{cos}\theta & 0\\ & & & \\ 0& 0& & I_2\end{array}\right)\left(\begin{array}{cccc}1& 0& \mathrm{}& 0\\ 0& & & \\ \mathrm{}& & Y_3& \\ 0& & & \end{array}\right)`$ (82)
$`=`$ $`\left(\begin{array}{cc}\begin{array}{cc}1& 0\\ 0& 1\end{array}& \begin{array}{cc}0& 0\\ 0& 0\end{array}\\ \begin{array}{cc}0& 0\\ 0& 0\end{array}& \begin{array}{c}X_2\end{array}\end{array}\right)\left(\begin{array}{cccc}1& 0& 0& 0\\ 0& e^{i\phi _1}\mathrm{cos}\xi _1& \mathrm{sin}\xi _1& 0\\ 0& \mathrm{sin}\xi _1& e^{i\phi _1}\mathrm{cos}\xi _1& 0\\ 0& 0& 0& 1\end{array}\right)\left(\begin{array}{cc}\begin{array}{cc}1& 0\\ 0& 1\end{array}& \begin{array}{cc}0& 0\\ 0& 0\end{array}\\ \begin{array}{cc}0& 0\\ 0& 0\end{array}& \begin{array}{c}Y_2\end{array}\end{array}\right)M(\theta ,\phi )R_3,`$ (105)
where (20) has been iteratively applied twice. Here $`X_3`$, $`Y_3`$ are $`SU(3)`$ matrices and $`X_2`$ may be parameterised as
$$X_2=\left(\begin{array}{cc}e^{i\phi _2}\mathrm{cos}\xi _2& e^{i\phi _3}\mathrm{sin}\xi _2\\ e^{i\phi _3}\mathrm{sin}\xi _2& e^{i\phi _2}\mathrm{cos}\xi _2\end{array}\right).$$
(106)
Taking the highest weight state $`|\varphi _0=(1,0,0,0)^T`$ and evaluating $`g|\varphi _0`$ as before, we observe that only two columns, the first column of $`X_3`$ and the first column of the matrix $`M(\theta ,\phi )`$, are important. The $`SU(4)`$ coherent states are
$`|𝐧_4=g|\varphi _0`$ $`=`$ $`e^{i\phi }\mathrm{cos}\theta \left(\begin{array}{c}1\\ 0\\ 0\\ 0\end{array}\right)+\mathrm{sin}\theta \left(\begin{array}{cc}\begin{array}{cc}1& 0\\ 0& 1\end{array}& \begin{array}{cc}0& 0\\ 0& 0\end{array}\\ \begin{array}{cc}0& 0\\ 0& 0\end{array}& \begin{array}{c}X_2\end{array}\end{array}\right)\left(\begin{array}{cccc}1& 0& 0& 0\\ 0& e^{i\phi _1}\mathrm{cos}\xi _1& \mathrm{sin}\xi _1& 0\\ 0& \mathrm{sin}\xi _1& e^{i\phi _1}\mathrm{cos}\xi _1& 0\\ 0& 0& 0& 1\end{array}\right)\left(\begin{array}{c}0\\ 1\\ 0\\ 0\end{array}\right)`$ (128)
$`=`$ $`\left(\begin{array}{c}e^{i\phi }\mathrm{cos}\theta \\ 0\\ 0\\ 0\end{array}\right)+\mathrm{sin}\theta \left(\begin{array}{c}0\\ \begin{array}{c}\\ |𝐧_3\end{array}\end{array}\right).`$ (138)
We note that the matrix including $`Y_2`$ in (105) commutes with the matrix to its right, that is $`[I_2Y_2,M(\theta ,\phi )]=0`$, and does not change the state $`|\varphi _0`$. The expression of the metric for this coset space is
$$|ds_4|^2=d\theta ^2+\mathrm{cos}^2(\theta )d\phi ^2+\mathrm{sin}^2(\theta )\{d\xi _1^2+\mathrm{cos}^2(\xi _1)d\phi _1^2+\mathrm{sin}^2(\xi _1)(d\xi _2^2+\mathrm{cos}^2(\xi _2)d\phi _2^2+\mathrm{sin}(\xi _2)d\phi _3)\},$$
(139)
and the measure is
$$d\mu _4=\mathrm{cos}(\theta )\mathrm{sin}^5(\theta )\mathrm{cos}(\xi _1)\mathrm{sin}^3(\xi _1)\mathrm{cos}(\xi _2)\mathrm{sin}(\xi _2)d\theta d\xi _1d\xi _2d\phi d\phi _1d\phi _2d\phi _3.$$
(140)
We note that the total volume is $`(2\pi )^4/(642)`$.
This establishes that the $`SU(n)`$ coherent states $`|𝐧_n`$ in this representation may be obtained from the iterative relation
$`|𝐧_n=g|\varphi _0`$ $`=`$ $`\left(\begin{array}{c}e^{i\phi }\mathrm{cos}\theta \\ 0\\ \mathrm{}\\ 0\end{array}\right)+\left(\begin{array}{cccc}1& 0& \mathrm{}& 0\\ 0& & & \\ \mathrm{}& & X_{n1}& \\ 0& & & \end{array}\right)\left(\begin{array}{c}0\\ \mathrm{sin}\theta \\ 0\\ \mathrm{}\\ 0\end{array}\right)`$ (154)
$`=`$ $`\left(\begin{array}{c}e^{i\phi }\mathrm{cos}\theta \\ 0\\ \mathrm{}\\ 0\end{array}\right)+\mathrm{sin}\theta \left(\begin{array}{c}0\\ \begin{array}{c}\\ |𝐧_{n1}\end{array}\end{array}\right),`$ (164)
where $`X_{n1}`$ are $`SU(n1)`$ matrices, and $`|𝐧_{n1}`$ is an $`SU(n1)`$ coherent state. Since $`|𝐧_n`$ is the unit vector of the $`(2n1)`$–sphere, the measure on the hypersphere is
$$d\mu _n=\mathrm{cos}\theta \mathrm{sin}^{2n3}(\theta )\mathrm{cos}(\xi _1)\mathrm{sin}^{2n5}(\xi _1)\mathrm{}\mathrm{cos}(\xi _{n2})\mathrm{sin}(\xi _{n2})d\theta d\xi _1\mathrm{}d\xi _{n2}d\phi d\phi _1\mathrm{}d\phi _{n1}.$$
(165)
## III Arbitrary $`SU(n)`$ representations
We extend the results of the previous section to irreducible unitary representations of arbitrarily large dimension for $`SU(n)`$. We define infinitesimal operators and the basis of the group representation. Using the decomposition of $`SU(n)`$, we derive an iterative equation for the $`SU(n)`$ coherent states and further obtain its recurrence equation. For the purpose of applications, some properties of the $`SU(n)`$ coherent states are given.
### A Infinitesimal operators
We denote $`T_n^N`$ as a representation of $`SU(n)`$ where the size number $`N`$ determines the dimension of the representation. A set of simultaneous normalised eigenstates of the Cartan operators $`J_h^h`$ $`(1<hn1)`$ is employed as an appropriate basis to describe the set of coherent states. This basis will be denoted as $`|m_1,\mathrm{},m_n`$ where the $`m_j`$ satisfy $`N=_{j=1}^nm_j`$. The basis elements are also simultaneous eigenstates of the size operator $`\widehat{N}`$ such that $`\widehat{N}|m_1,\mathrm{},m_n=N|m_1,\mathrm{},m_n`$. For $`SU(2)`$, these are equivalent to the angular momentum eigenstates. The $`n^2n`$ raising operators $`J_j^h`$, $`1h<jn`$, and the same number of lowering operators $`J_j^h`$, $`1j<hn`$, of $`SU(n)`$ satisfy the relations
$`[`$raising operators, $`(h<j)]`$
$$J_j^h|m_1,\mathrm{},m_h,\mathrm{},m_j,\mathrm{},m_n=\sqrt{(m_h+1)m_j}|m_1,\mathrm{},m_h+1,\mathrm{},m_j1,\mathrm{},m_n,$$
(166)
$`[`$lowering operators, $`(h>j)]`$
$$J_j^h|m_1,\mathrm{},m_j,\mathrm{},m_h,\mathrm{},m_n=\sqrt{m_h(m_j+1)}|m_1,\mathrm{},m_h1,\mathrm{},m_j+1,\mathrm{},m_n.$$
(167)
For Cartan operators $`J_h^h`$, $`1hn1`$ we have
$$J_h^h|m_1,\mathrm{},m_h,\mathrm{},m_j,\mathrm{},m_n=\sqrt{\frac{2}{h(h+1)}}\left(\underset{k=1}{\overset{h}{}}m_khm_{h+1}\right)|m_1,\mathrm{},m_h,\mathrm{},m_j,\mathrm{},m_n.$$
(168)
### B $`SU(n)`$ coherent states
It is appropriate to choose $`|\varphi _0=|N,0,\mathrm{},0`$ as the highest weight state, since the action of any raising operator (166) on this state gives zero. The parameterization (20) shows that the representation $`T_n^N(g)`$ may be decomposed as
$$T_n^N(g)=T^N(L_{n1})T^N(M)T^N(R_{n1}).$$
(169)
The action of the representation $`T^N(R_{n1})`$ does not change the highest weight state as we have seen in the examples in the previous section, hence the coherent state is determined as
$$|𝐧_n^N=T^N(L_{n1})T^N(M(\theta ,\phi ))|N,0,\mathrm{},0.$$
(170)
The right element $`T^N(M(\theta ,\phi ))`$ in (170) acts as an $`SU(2)`$ operator on the subspace $`|m_1,m_2`$, a cross section obtained by taking the first two elements of $`|m_1,m_2,\mathrm{},m_n`$. It is well-known that an arbitrary $`gSU(2)`$ may be decomposed as
$`g`$ $`=`$ $`e^{\zeta ^{}J_2^1}e^{\nu J_1^1}e^{\zeta J_1^2}e^{i\phi _1J_1^1}`$ (171)
$`=`$ $`e^{\zeta J_1^2}e^{\nu J_1^1}e^{\zeta ^{}J_2^1}e^{i\phi _1J_1^1}.`$ (172)
The parameters in the above expressions correspond to the angle parameters in (1) as $`\zeta =e^{i(\phi _2\phi _1)}\mathrm{tan}\theta `$, and $`\nu =\mathrm{ln}\mathrm{cos}\theta `$. Setting $`\phi _1=\phi `$ and $`\phi _2=0`$, $`T^N(M(\theta ,\phi ))`$ is decomposed as (172), and acts on the subspace $`|N,0`$ of $`|N,0,\mathrm{},0`$ as
$$T^N(M(\theta ,\phi ))|N,0,\mathrm{},0=\underset{j=0}{\overset{N}{}}e^{i\phi (Nj)}\mathrm{sin}^j(\theta )\mathrm{cos}^{Nj}(\theta )\left(\begin{array}{c}N\\ j\end{array}\right)^{1/2}|Nj,j,0,\mathrm{},0$$
(173)
The left element of the decomposition, $`T^N(L_{n1})`$, does not change the first element of the state $`|Nj,j,0,\mathrm{},0`$, and acts on the subspace $`|j,0,\mathrm{},0`$ in the state $`|Nj|j,0,\mathrm{},0`$. This element acts as an $`SU(n1)`$ operator on the subspace $`|j,0,\mathrm{},0`$, which generates the $`SU(n1)`$ coherent states, giving
$$|𝐧_n^N=\underset{j=0}{\overset{N}{}}e^{i\phi (Nj)}\mathrm{sin}^j(\theta )\mathrm{cos}^{Nj}(\theta )\left(\begin{array}{c}N\\ j\end{array}\right)^{1/2}|Nj|𝐧_{n1}^j.$$
(174)
In order to obtain a more convenient expression of the $`SU(n)`$ coherent states, we derive a recurrence relation from the above iterative equation (174). The last decomposition in (172) gives the SU(2) coherent states
$`|𝐧_2^N`$ $`=`$ $`{\displaystyle \underset{j=0}{\overset{N}{}}}e^{ij\phi _2}e^{i(Nj)\phi _1}\mathrm{sin}^j(\theta )\mathrm{cos}^{Nj}(\theta )\left(\begin{array}{c}N\\ j\end{array}\right)^{1/2}|Nj,j`$ (177)
$`=`$ $`{\displaystyle \underset{j=0}{\overset{N}{}}}\eta _j^N(\phi _1,\phi _2,\theta )|Nj,j,`$ (178)
where we define
$$\eta _j^N(\phi _1,\phi _2,\theta )e^{ij\phi _2}e^{i(Nj)\phi _1}\mathrm{sin}^j(\theta )\mathrm{cos}^{(Nj)}(\theta )\left(\begin{array}{c}N\\ j\end{array}\right)^{1/2}.$$
(179)
The $`SU(3)`$ coherent states are constructed using the $`SU(2)`$ coherent states, and the relations (174) and (179) give
$`|𝐧_3^N`$ $`=`$ $`{\displaystyle \underset{j_1=0}{\overset{N}{}}}e^{i\phi (Nj_1)}\mathrm{sin}^{j_1}(\theta )\mathrm{cos}^{(Nj_1)}(\theta )\left(\begin{array}{c}N\\ j_1\end{array}\right)^{1/2}|N,j_1|𝐧_2^{j_1}`$ (182)
$`=`$ $`{\displaystyle \underset{j_1=0}{\overset{N}{}}}\eta _{j_1}^N(\phi ,0,\theta ){\displaystyle \underset{j_2=0}{\overset{j_1}{}}}\eta _{j_2}^{j_1}(\phi _1,\phi _2,\xi _1)|Nj_1,j_1j_2,j_2,`$ (183)
in agreement with the $`SU(3)`$ coherent states developed in .
Recursively, the $`SU(n)`$ coherent states may be expressed by the function $`\eta _k^l(\alpha ,\beta ,\gamma )`$ of (179),
$`|𝐧_n^N`$ $`=`$ $`{\displaystyle \underset{j_1=0}{\overset{N}{}}}\eta _{j_1}^N(\phi ,0,\theta ){\displaystyle \underset{j_2=0}{\overset{j_1}{}}}\eta _{j_2}^{j_1}(\phi _1,0,\xi _1)\mathrm{}{\displaystyle \underset{j_{n2}=0}{\overset{j_{n3}}{}}}\eta _{j_{n2}}^{j_{n3}}(\phi _{n3},0,\xi _{n3})`$ (185)
$`{\displaystyle \underset{j_{n1}=0}{\overset{j_{n2}}{}}}\eta _{j_{n1}}^{j_{n2}}(\phi _{n2},\phi _{n1},\xi _{n2})|Nj_1,j_1j_2,\mathrm{},j_{n1}.`$
### C Properties of the $`SU(n)`$ coherent states
For the purpose of applications, here we describe some fundamental properties of the $`SU(n)`$ coherent states.
(1) Stereographic coordinates
The decomposition (172) implies that the $`SU(n)`$ coherent states may be represented in the complex numbers $`\zeta _k`$ such that $`\zeta _k=e^{i(\phi _{k+1}\phi _k)}\mathrm{tan}(\xi _k)`$. Routine change of variables gives the $`SU(n)`$ coherent states in this stereographic coordinates as
$`|𝐧_n^N=e^{i\phi N}\left({\displaystyle \frac{1}{1+|\zeta |^2}}\right)^N`$ $`{\displaystyle \underset{j_1=0}{\overset{N}{}}}`$ $`(\zeta )^{j_1}\left(\begin{array}{c}N\\ j_1\end{array}\right)^{1/2}\left({\displaystyle \frac{1}{1+|\zeta _1|^2}}\right)^{j_1}`$ (188)
$`{\displaystyle \underset{j_2=0}{\overset{j_1}{}}}`$ $`(\zeta _1)^{j_2}\left(\begin{array}{c}j_1\\ j_2\end{array}\right)^{1/2}\left({\displaystyle \frac{1}{1+|\zeta _2|^2}}\right)^{j_2}`$ (191)
$`\mathrm{}`$ (192)
$`{\displaystyle \underset{j_{n1}=0}{\overset{j_{n2}}{}}}`$ $`(\zeta _{n2})^{j_{n1}}\left(\begin{array}{c}j_{n2}\\ j_{n1}\end{array}\right)^{1/2}|Nj_1,\mathrm{},j_{n2}j_{n1},j_{n1}`$ (195)
(2) Resolution of unity
The set of coherent states provides a resolution of unity in the coset space as
$$\frac{(N+n1)!}{2\pi ^nN!}𝑑\mu _n|𝐧_n^N𝐧_n^N|=\widehat{I}.$$
(196)
The matrix $`|𝐧_n^N𝐧_n^N|`$ may be expanded in terms of matrices $`|Nj_1^{},\mathrm{},j_{n1}^{}Nj_1,\mathrm{},j_{n1}|`$ by using the expansion (185). The integrals with respect to $`\xi _k`$ are carried out by change of integral variables $`x\mathrm{cos}^2(\xi _k)`$ allowing $`_0^{\pi /2}𝑑\xi _k\mathrm{cos}^{(2(mn)+1)}(\xi _k)\mathrm{sin}^{(2n+1)}(\xi _k)=\frac{1}{2}\frac{n!(mn)!}{(m+1)!}`$, while the integrals in terms of $`\phi _j`$ produce delta functions. The result of the all integrals cancels with the normalization factor, and gives us $`_{j_1=0}^N\mathrm{}_{j_{n1}=0}^{j_{n2}}|Nj_1,\mathrm{},j_{n1}Nj_1,\mathrm{},j_{n1}|=\widehat{I}`$.
(3) Overlap of two coherent states
The overlap of two coherent states may be calculated from (174), as
$$𝐧_{}^{}{}_{n}{}^{N}|𝐧_n^N=\left(e^{i(\phi _{n1}\phi _{n1}^{})}\underset{k=0}{\overset{n2}{}}\mathrm{sin}\xi _k\mathrm{sin}\xi _k^{}+\underset{m=0}{\overset{n2}{}}\left[e^{i(\phi _m\phi _m^{})}\mathrm{cos}\xi _m\mathrm{cos}\xi _m^{}\underset{k=0}{\overset{m1}{}}\mathrm{sin}\xi _k\mathrm{sin}\xi _k^{}\right]\right)^N,$$
(197)
where we have changed the notation of angles, replacing $`\theta `$ with $`\xi _0`$, and $`\phi `$ with $`\phi _0`$, and where we have defined $`_{k=0}^1\mathrm{sin}\xi _k\mathrm{sin}\xi _k^{}=1`$.
## IV Summary
In conclusion, we have described $`SU(n)`$ coherent states for irreducible unitary representations for arbitrarily large dimension and some examples for small $`n`$ demonstrated. The geometric structure of the $`SU(n)`$ coherent states has been represented using spherical coordinates. We also gave expressions for the resolution of unity, and the non-orthogonality of the coherent states. It was shown the $`SU(n)`$ coherent states may be recursively derived from $`SU(2)`$ coherent states.
###### Acknowledgements.
The author would like to thank David De Wit for useful discussions and comments about group representations and Michael J. Gagen for useful suggestions. The author acknowledges the financial support of the Australian International Education Foundation (AIEF).
## A $`\lambda `$–matrices
In general $`SU(n)`$ generators can be represented by $`n^2n`$ off–diagonal matrices and $`n1`$ diagonal matrices. For example, $`SU(4)`$ has fifteen generators which can be constructed using twelve off–diagonal matrices and three diagonal matrices . We take $`\{e_j^h\}`$ as a basis for the group $`SU(n)`$, where $`e_j^h`$ are elementary matrices. We also define $`e_j^h,(h<j)`$ as raising operators, and $`e_j^h,(h>j)`$ as lowering operators respectively. Non-diagonal elements of this basis are
$$\{\beta _j^h=i(e_j^he_h^j),\mathrm{\Theta }_j^h=e_j^h+e_h^j,1h<jn\}.$$
(A1)
The commutation relations of these non–diagonal elements are
$$[\beta _j^h,\mathrm{\Theta }_e^k]=i\delta _j^k\mathrm{\Theta }_e^h+i\delta _e^h\mathrm{\Theta }_j^k+i\delta _h^k\mathrm{\Theta }_e^ji\delta _e^j\mathrm{\Theta }_h^k.$$
(A2)
The diagonal elements $`\{\eta _m^m|1mn1\}`$ are
$$\eta _m^m=\sqrt{\frac{2}{m(m+1)}}\left(\underset{j=1}{\overset{m}{}}e_j^jme_{m+1}^{m+1}\right).$$
(A3)
For instance in $`SU(4)`$ the fifteen $`\lambda `$–matrices are numbered as
$`\{\begin{array}{cc}\lambda _1=\mathrm{\Theta }_2^1& \lambda _2=\beta _2^1\\ \lambda _3=\eta _1^1,& \end{array}`$ (A6)
$`\{\begin{array}{cc}\lambda _4=\mathrm{\Theta }_3^1& \lambda _5=\beta _3^1\\ \lambda _6=\mathrm{\Theta }_3^2& \lambda _7=\beta _3^2\\ \lambda _8=\eta _2^2,& \end{array}`$ (A10)
$`\{\begin{array}{cc}\lambda _9=\mathrm{\Theta }_4^1& \lambda _{10}=\beta _4^1\\ \lambda _{11}=\mathrm{\Theta }_4^2& \lambda _{12}=\beta _4^2\\ \lambda _{13}=\mathrm{\Theta }_4^3& \lambda _{14}=\beta _4^3\\ \lambda _{15}=\eta _3^3.& \end{array}`$ (A15)
These $`\lambda `$–matrices are the generators of the representation $`T_4^1`$.
These $`SU(4)`$ generators allow another expression for the coherent states. Defining a matrix $`\lambda _{15}^{}`$ as $`(\sqrt{6}\lambda _{15}\sqrt{3}\lambda _8)/3`$, the decomposition of $`SU(4)`$ using $`\lambda `$–matrices gives expressions for the coherent states
$`|𝐧_4`$ $`=`$ $`e^{i\alpha \lambda _{15}^{}}e^{i\beta \lambda _{14}}e^{i\gamma \lambda _{15}^{}}e^{i\phi _1/2\lambda _8^{}}e^{i\xi _1\lambda _7}e^{i\phi _1/2\lambda _8^{}}e^{i\phi /2\lambda _3}e^{i\theta \lambda _2}e^{i\phi /2\lambda _3}|\varphi _0`$ (A16)
$`=`$ $`e^{i\sqrt{6}\alpha /3\lambda _{15}}e^{i\sqrt{3}\alpha /3\lambda _8}e^{i\beta \lambda _{14}}e^{i\sqrt{6}\gamma /3\lambda _{15}}e^{i\sqrt{3}\gamma /3\lambda _8}e^{i\sqrt{3}\phi _1/4\lambda _8}e^{i\phi _1/4\lambda _3}e^{i\xi _1\lambda _7}e^{i\sqrt{3}\phi _1/4\lambda _8}e^{i\phi _1/4\lambda _3}`$ (A18)
$`e^{i\phi /2\lambda _3}e^{i\theta \lambda _2}e^{i\phi /2\lambda _3}|\varphi _0,`$
where $`\phi _2=\alpha +\gamma `$, $`\phi _3=\alpha +\gamma `$, and $`\xi _2=\beta `$. These expressions have been obtained directly from (66) and (105), and show the displacement operator for the $`SU(4)`$ coherent states.
## B The symmetric parameterization for $`SU(n)`$
We here show a brief proof of the parameterization (20). A proof of this parameterization for $`n=3`$ was given in . Showing any element of $`gSU(n)`$ can be transformed into $`R_{n1}^{}`$, we give the parameterization (20) as an inverse equation in terms of the element $`g`$. We first review the proof for $`n=3`$, and prove (20) for arbitrary $`n`$ inductively.
(i) For n=3, an arbitrary element $`g=\left(\begin{array}{ccc}x_{11}& x_{12}& x_{13}\\ x_{21}& x_{22}& x_{23}\\ x_{31}& x_{32}& x_{33}\end{array}\right)`$ can be transformed as
$$\left(\begin{array}{ccc}1& 0& 0\\ 0& a^{}& b^{}\\ 0& b& a\end{array}\right)\left(\begin{array}{ccc}x_{11}& x_{12}& x_{13}\\ x_{21}& x_{22}& x_{23}\\ x_{31}& x_{32}& x_{33}\end{array}\right)=\left(\begin{array}{ccc}x_{11}& x_{12}& x_{13}\\ r_2^3& \frac{_{k=2}^3x_{k1}^{}x_{k2}}{r_2^3}& \frac{_{k=2}^3x_{k1}^{}x_{k3}}{r_2^3}\\ 0& & \end{array}\right),$$
(B1)
where $`a=x_{21}/r_2^3`$, $`b=x_{31}/r_2^3`$ and $`r_p^q=\sqrt{_{k=p}^q|x_{k1}|^2}`$. Applying a matrix $`\left(\begin{array}{ccc}x_{11}^{}& \sqrt{1|x_{11}|^2}& 0\\ \sqrt{1|x_{11}|^2}& x_{11}& 0\\ 0& 0& 0\end{array}\right)`$ from the left on the above matrix (B1) gives $`\left(\begin{array}{ccc}1& 0& 0\\ 0& & \\ 0& & \end{array}\right)`$, where we used the constraints on $`g`$, which are $`_{j=1}^3x_{jk}^{}x_{jl}=\delta _{kl}`$. With suitably chosen parameters, the inversion of the above relation gives the devised SU(3) parameterization (33). Now we extend this procedure to the general result, and prove it inductively.
(ii) We assume the result in the case (i), that is, for $`n=m`$ an arbitrary element $`gSU(m)`$ can be parametrized as (20), and for any $`g`$ a matrix $`X_{m1}SU(m1)`$ exists such that
$$\left(\begin{array}{cccc}1& 0& \mathrm{}& 0\\ 0& & & \\ \mathrm{}& & X_{m1}^{}& \\ 0& & & \end{array}\right)\left(\begin{array}{cccc}x_{11}& x_{12}& \mathrm{}& x_{1m}\\ x_{21}& x_{22}& \mathrm{}& x_{2m}\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\\ x_{m1}& x_{m2}& \mathrm{}& x_{mm}\end{array}\right)=\left(\begin{array}{cccc}x_{11}& x_{12}& \mathrm{}& x_{1m}\\ r_2^m& \frac{_{k=2}^mx_{k1}^{}x_{k2}}{r_2^m}& \mathrm{}& \frac{_{k=2}^mx_{k1}^{}x_{km}}{r_2^m}\\ 0& & & \\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\\ 0& & & \end{array}\right).$$
(B2)
(iii) For n=m+1, using (B2), an arbitrary matrix $`gSU(m+1)`$ can be transformed as
$`\left(\begin{array}{cccc}I_2& 0& \mathrm{}& 0\\ 0& & & \\ \mathrm{}& & Y_{m1}^{}& \\ 0& & & \end{array}\right)\left(\begin{array}{cccc}1& 0& 0& \\ 0& \frac{x_{21}^{}}{r_2^{m+1}}& \frac{r_3^{m+1}}{r_2^{m+1}}& 0\\ 0& \frac{r_3^{m+1}}{r_2^{m+1}}& \frac{x_{21}}{r_2^{m+1}}& \\ & 0& & I_{m2}\end{array}\right)\left(\begin{array}{cccc}I_2& 0& \mathrm{}& 0\\ 0& & & \\ \mathrm{}& & X_{m1}^{}& \\ 0& & & \end{array}\right)`$ (B15)
$`\times \left(\begin{array}{cccc}x_{11}& x_{12}& \mathrm{}& x_{1m+1}\\ x_{21}& x_{22}& \mathrm{}& x_{2m+1}\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\\ x_{m+11}& x_{m+12}& \mathrm{}& x_{m+1m+1}\end{array}\right)=\left(\begin{array}{cccc}x_{11}& x_{12}& \mathrm{}& x_{m+1}\\ r_2^{m+1}& \frac{_{k=2}^{m+1}x_{k1}^{}x_{k2}}{r_2^{m+1}}& \mathrm{}& \frac{_{k=2}^{m+1}x_{k1}^{}x_{km+1}}{r_2^{m+1}}\\ 0& & \mathrm{}& \\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\\ 0& & \mathrm{}& \end{array}\right),`$ (B25)
where $`I_k`$ are $`k\times k`$ identity matrices. Using the constraints for $`SU(m+1)`$ matrices, $`_{j=1}^{m+1}x_{jk}^{}x_{jl}=\delta _{kl}`$, the matrix on the right hand side can be transformed to contain an $`SU(m)`$ matrix as
$`\left(\begin{array}{ccc}x_{11}^{}& \sqrt{1|x_{11}|^2}& 0\\ \sqrt{1|x_{11}|^2}& x_{11}& 0\\ 0& & I_{m1}\end{array}\right)`$ $`\left(\begin{array}{cccc}x_{11}& x_{12}& \mathrm{}& x_{m+1}\\ r_2^{m+1}& \frac{_{k=2}^{m+1}x_{k1}^{}x_{k2}}{r_2^{m+1}}& \mathrm{}& \frac{_{k=2}^{m+1}x_{k1}^{}x_{km+1}}{r_2^{m+1}}\\ 0& & \mathrm{}& \\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\\ 0& & \mathrm{}& \end{array}\right)`$ (B34)
$`=\left(\begin{array}{cccc}1& 0& \mathrm{}& 0\\ 0& & & \\ \mathrm{}& & Y_m& \\ 0& & & \end{array}\right)`$ (B39)
Since the second matrix can be parametrized equivalently to the $`n=m`$ case, the product of the three matrices constructs an $`SU(n)`$ matrix. The inversion of this relation gives
$`\left(\begin{array}{cccc}x_{11}& x_{12}& \mathrm{}& x_{1m+1}\\ x_{21}& x_{22}& \mathrm{}& x_{2m+1}\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\\ x_{m+\mathrm{1\hspace{0.17em}1}}& x_{m+\mathrm{1\hspace{0.17em}2}}& \mathrm{}& x_{m+1m+1}\end{array}\right)`$ (B57)
$`=\left(\begin{array}{cccc}1& 0& \mathrm{}& 0\\ 0& & & \\ \mathrm{}& & X_m& \\ 0& & & \end{array}\right)\left(\begin{array}{cccc}x_{11}& \sqrt{1|x_{11}|^2}& & \\ \sqrt{1|x_{11}|^2}& x_{11}^{}& & 0\\ & & & \\ & 0& & I_{m1}\end{array}\right)\left(\begin{array}{cccc}1& 0& \mathrm{}& 0\\ 0& & & \\ \mathrm{}& & Y_m& \\ 0& & & \end{array}\right).`$
The elements, $`x_{11}`$ and $`\sqrt{1|x_{11}|^2}`$, can be parametrized as
$$x_{11}=e^{i\phi }\mathrm{cos}\theta ,\sqrt{1|x_{11}|^2}=\mathrm{sin}\theta .$$
(B58)
|
warning/0004/cond-mat0004299.html
|
ar5iv
|
text
|
# Untitled Document
Momentum and doping dependence of the electron spectrum and ($`\pi `$,0) feature in copper oxide materials
Feng Yuan<sup>a</sup>, Xianglin Ke<sup>a</sup>, and Shiping Feng<sup>a,b</sup>
<sup>a</sup>Department of Physics, Beijing Normal University, Beijing 100875, China
<sup>b</sup>National Laboratory of Superconductivity, Academia Sinica, Beijing 100080, China
The momentum and doping dependence of the electron spectrum of copper oxide materials in the underdoped regime is studied within the $`t`$-$`J`$ model. It is shown that the pseudogap opens near ($`\pi `$,0) point in the Brillouin zone, and the electron dispersion exhibits the flat band around ($`\pi `$,0), which are consistent with the experiments.
The experimental measurements from the photoemission spectroscopy show that electron spectrum in copper oxide materials is strongly momentum and doping dependent, and has an anomalous form as a function of energy $`\omega `$ for $`𝐤`$ in the vicinity of ($`\pi `$,0) point in the Brillouin zone, which leads to the flat band near momentum ($`\pi `$,0) with anomalously small changes of electron energy as a function of momentum. This flat band around ($`\pi `$,0) point has a particular importance in the mechanism of the normal-state pseudogap formation . Although the exact origin of the flat band still is controversial, a strongly correlated many-body like approach may be appropriate to describe the electronic structure of copper oxide materials. In this paper, we study this issue within the $`t`$-$`J`$ model ,
$`H=t{\displaystyle \underset{i\widehat{\eta }\sigma }{}}C_{i\sigma }^{}C_{i+\widehat{\eta }\sigma }+J{\displaystyle \underset{i\widehat{\eta }}{}}𝐒_i𝐒_{i+\widehat{\eta }},`$ (1)
acting on the space with no doubly occupied sites, where $`\widehat{\eta }=\pm \widehat{x},\pm \widehat{y}`$, $`C_{i\sigma }^{}`$ ($`C_{i\sigma }`$) are the electron creation (annihilation) operators, and $`𝐒_i`$ is spin operator. The strong electron correlation in the $`t`$-$`J`$ model manifests itself by the electron single occupancy on-site local constraint . To incorporate this local constraint, the fermion-spin theory based on the charge-spin separation, $`C_i=h_i^{}S_i^{}`$, $`C_i=h_i^{}S_i^+`$, has been proposed , where the spinless fermion operator $`h_i`$ keeps track of the charge (holon), while the pseudospin operator $`S_i`$ keeps track of the spin (spinon). In this paper, we hope to discuss the electronic structure of copper oxide materials, and therefore it needs to calculate the electron Green’s function $`G(ij,tt^{})=C_{i\sigma }(t);C_{j\sigma }^{}(t^{})`$. According to the fermion-spin theory, the electron Green’s function is a convolution of the spinon Green’s function $`D(ij,tt^{})=S_i^+(t);S_j^{}(t^{})`$ and holon Green’s function $`g(ij,tt^{})=h_i(t);h_j^{}(t^{})`$, and can be formally expressed in terms of the spectral representation as,
$`G(𝐤,\omega )={\displaystyle \frac{1}{N}}{\displaystyle \underset{q}{}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{d\omega ^{}}{2\pi }}{\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{d\omega ^{\prime \prime }}{2\pi }}A_h(𝐪,\omega ^{})\times `$
$`A_s(𝐪+𝐤,\omega ^{\prime \prime }){\displaystyle \frac{n_F(\omega ^{})+n_B(\omega ^{\prime \prime })}{\omega +\omega ^{}\omega ^{\prime \prime }}},(2)`$ (2)
where the holon spectral function $`A_h(𝐪,\omega )=2\mathrm{I}\mathrm{m}g(𝐪,\omega )`$, spinon spectral function $`A_s(𝐤,\omega )=`$ $`2\mathrm{I}\mathrm{m}D(𝐤,\omega )`$, and $`n_B(\omega )`$ and $`n_F(\omega )`$ are the boson and fermion distribution functions for spinons and holons, respectively. In this paper, we limit the spinon part to the first-order (mean-field level) since some physical properties can be well described at this level . On the other hand, it has been shown that there is a connection between the anomalously temperature dependence of the resistivity and the flat band around ($`\pi `$,0) point and normal-state pseudogap . Therefore we treat the holon part by using the loop expansion to the second-order correction as in the discussion of the charge dynamics . Within the fermion-spin theory, the mean-field spinon and full holon Green’s functions $`D^{(0)}(𝐤,\omega )`$ and $`g(𝐤,\omega )`$ have been evaluated in Ref. and Ref. , respectively. Substituting these Green’s functions into Eq. (2), we therefore can obtain the electron spectral function $`A(𝐤,\omega )=2\mathrm{I}\mathrm{m}G(𝐤,\omega )`$.
The numerical results of the electron spectral function at ($`\pi `$,0) point in the doping $`\delta =0.06`$ for the parameter $`t/J=2.5`$ in the zero temperature are plotted in Fig. 1 (solid line). For comparison, the corresponding mean-field results (dashed line) are also plotted in Fig. 1. These results indicate that at the mean-field level, the electron spectrum at ($`\pi `$,0) point consists of two main parts, which comes from noninteracting particles. After including the fluctuation, the mean-field part is renormalized and the spectral weight has been spread to lower energies, in particular, the sharp mean-field peak at ($`\pi `$,0) point near the chemical potential $`\mu `$ has been split into two peaks. Moreover, the low energy peaks are well defined at all momenta, and the positions of the dominant peaks in $`A(𝐤,\omega )`$ as a function of momentum in the doping $`\delta =0.06`$ for the parameter $`t/J=2.5`$ are shown in Fig. 2 (solid line). In comparison with corresponding mean-field results (dashed line) in Fig. 2, it is shown that the mean-field electron dispersion $`E_k^{(0)}`$ in the vicinity of ($`\pi `$,0) point has been split into two branches $`E_k^{()}`$ and $`E_k^{(+)}`$, and a pseudogap opens. The branch $`E_k^{()}`$ has a very weak dispersion around ($`\pi `$,0) point, and then the flat regime appears, while the Fermi energy is only slightly above this flat regime. Although the nature of the pseudogap is different in different theories, the present results show that the pseudogap near ($`\pi `$,0) point is closely related to the spinon fluctuation, since the full holon Green’s function (then the electron Green’s function) is obtained by considering the second-order correction due to the spinon pair bubble, where the single-particle hopping is strongly renormalized by the short-range antiferromagnetic order resulting in a bandwidth also of order of (a few) $`J`$, this renormalization is then responsible for the anomalous dispersion around ($`\pi `$,0) point and normal-state pseudogap.
Acknowledgements
The authors would like to thank Professor C.D. Gong, Professor H.Q. Lin, and Professor Z.X. Zhao for helpful discussions. This work was supported by the National Natural Science Foundation of China under Grant No. 19774014.
1. Z.X. Shen et al., Phys. Rep. 253 (1995) 1.
2. P.W. Anderson, Science 235 (1987) 1196.
3. S. Feng et al., Phys. Rev. B49 (1994) 2368.
4. S. Feng et al., Phys. Lett. A232 (1997) 293.
5. S. Feng et al., Phys. Rev. B55 (1997) 642.
Figure Captions
Figure 1. $`A(𝐤,\omega )`$ at ($`\pi `$,0) point at the doping $`\delta =0.06`$ for $`t/J=2.5`$. The dashed line is the result at the mean-field level.
Figure 2. Position of the dominant peaks in $`A(𝐤,\omega )`$ as a function of momentum at the doping $`\delta =0.06`$ for $`t/J=2.5`$. The dashed line is the result at the mean-field level.
|
warning/0004/hep-th0004133.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
There has been renewed interest in Kaluza–Klein theories over the past two years, mainly due to the realization that localization of matter and localization of gravity may drastically change the commonly assumed properties of such models. These theories clearly open new approaches to the cosmological constant problem<sup>*</sup><sup>*</sup>*The cosmological constant problem concerns the brane cosmological constant that governs the expansion of our universe and it has to be distinguished to the vacuum energy density, or tension, on the brane: in particular it is possible, for a non-vanishing tension, to find solutions to the Einstein equations with a flat Minkowski metric on the brane in which case the brane cosmological constant is zero., since using extra dimensions the no-go theorem of Weinberg for adjustment mechanisms may be circumvented. A particularly simple scenario has been recently suggested in to at least improve on the cosmological constant problem (see also for an earlier mechanism involving extra dimensions), by proposing a mechanism for the cancellation of all order Standard Model (SM) loop contributions and the leading (tree-level) gravity contribution to the effective 4D cosmological constant,
$$\underset{\text{tree-level}}{\underset{}{𝒪(M^4)}}+\underset{\text{SM loops}}{\underset{}{𝒪(T_{br})}}+\underset{\text{higher order}}{\underset{}{𝒪(T_{br}^2M^4)+\mathrm{}}}$$
(1.1)
where $`M`$ is the fundamental scale of gravity, the Planck scale in the bulk for instance. The mechanism is based on a single 3-brane (to which the SM fields are localized) embedded into a 4+1 dimensional space-time. The essential new ingredient is a bulk scalar field $`\varphi `$, which is coupled to the brane tension $`T_{br}`$ (assumed to be the full loop-corrected SM vacuum energy density). The authors of then showed that one can find static solutions to the classical equations of motion for a vanishing bulk potential of the scalar field, but for arbitrary values of the brane tension $`T_{br}`$. However, all the solutions found in which localize gravity involve a naked space-time singularity, which has been interpreted as the boundary of the extra dimension (see also for a discussion on singularity in a brane-world context). Since the bulk is effectively compactified, we have to worry whether the size of the extra dimension is compatible with our phenomenological knowledge of gravity which has been tested up to millimeter distances. The proper distance from the brane to the singularity is found to be $`y_c\kappa _5^2T_{br}^1`$ where $`M_5=\kappa _5^{2/3}`$ is the 5D Planck scale, which is related to the 4D Planck scale, $`M_4=\kappa _4^1`$, by $`T_{br}=\kappa _4^2/\kappa _5^4`$. If the tension, associated to the vacuum energy of the SM fields, is of the order of the electroweak scale, $`T_{br}\text{TeV}^4`$, we naturally obtainThe fact that the electroweak scale on the brane is five orders of magnitude below the fundamental scale, $`M_5`$, is the gauge hierarchy problem in this model and a mechanism, such as low energy supersymmetry for instance, is needed to cancel the quadratic divergences which should bring $`T_{br}`$ near $`M_5^4`$.
$$M_510^8\text{GeV }\text{and }y_c1\text{mm}$$
(1.2)
which is phenomenologically safe. Finally, while the SM contribution to the 4D cosmological constant would be of the order of $`10^{64}M_4^4`$, the self-tuning mechanism cancels this term. However, the next possible term is of the order of $`10^{84}M_4^4`$, which is still forty orders of magnitude too large. It is worth noticing that the self-tuning mechanism eliminates twenty orders of magnitude and thus should be considered on equal footing to the Randall–Sundrum solution to the hierarchy problem.
The necessity of singularities in the bulk is here only dictated by the phenomenological requirement of localized gravity. However it complements the singularity theorem of Hawking and Penrose under which generic initial conditions lead to singular solutions of Einstein’s equations .
Several works have studied the self-tuning mechanism. In this paper, we examine the general properties of the solutions to the coupled 5D scalar-gravity system in detail. First, we investigate the solutions in the presence of a general bulk potential for the scalar field. The motivation to include a scalar bulk potential are twofold: (i) to give a mass to the scalar field and thus evade the cosmological problems associated to a massless scalar field coupled to gravity and/or those associated to an unstabilized extra-dimension; (ii) to overcome the singularity problem by considering a more general bulk geometry. We explain that the self-tuning behavior is expected to occur for a generic bulk potential. However, in the generic solutions (except for a vanishing bulk potential) self-tuning does have the more restricted interpretation that there is at least a finite region for the brane tension for which static solutions can be found (but not necessarily for all values of $`T_{br}`$). More precisely, Standard Model corrections are expected to occur at the weak scale, whereas self-tuning works up to the larger 5D Planck scale.
After presenting our general results and methods on the solutions in the presence of a generic bulk potential we ask the following question: are there bulk potentials such that the self-tuning solutions avoid naked space-time singularities for a range of values of the brane tension in such a way, that gravity is localized to the brane (so as to reproduce 4D gravity for the observer on the brane). We show that for the 5D scalar-gravity system the only possibility for such solutions is when space-time asymptotes to five dimensional anti-de Sitter space (AdS<sub>5</sub>) far from the brane (therefore producing solutions of the sort considered in ). After a careful analysis we show that such solutions are always fine-tuned; that is, they occur only for particular isolated values of the brane tension $`T_{br}`$. We show that the number of allowed brane tensions is related to the number of negative stationary points of the bulk potential, which appears as a maximal index. In the case of an oscillatory bulk potential, we thus obtain a quantization of the brane tension as in a brane world construction from supergravity .
The paper is organized as follows: in Section 2 we present our method for searching for solutions of the coupled scalar-gravity system using the superpotential formalism of . The advantage of this method is that it results in first order ordinary differential equations, and all the degrees of freedom are transparent without having to make a particular ansatz for the solution. In Section 3 we present a class of exactly solvable systems, given by an exponential bulk potential, which includes all the models presented in . We analyze these models in detail and find that all of these solutions necessarily involve a naked singularity or lead to an infinite Planck scale on the brane. We also find that, by relaxing an ansatz for solutions made in , the exponential bulk potential is in fact self-tuning. In Section 4 we compare a perturbative and a numerical method to solve the equations for the most general bulk potentials, and present an example for both methods using the exactly solvable cases of Section 3. In Section 5 we present a no-go theorem for self-tuning branes that would localize gravity without singularities. In Section 6 we comment on the effect of resolution of the singularities on self-tuning. We conclude in Section 7.
## 2 Self-tuning and scalar fields
In this section we demonstrate that the 4D cosmological constant of branes coupled to scalar fields is generically self-tuned to zero, as in . We begin with the action for a brane coupled to gravity and a real scalar field (such dilatonic domain walls have been extensively studied from a supersymmetric point of view , a new ingredient here is the localization of SM fields)<sup>*</sup><sup>*</sup>*Our conventions correspond to a mostly positive Lorentzian signature $`(+\mathrm{}+)`$ and the definition of the curvature in terms of the metric is such that a Euclidean sphere has positive curvature. Bulk indices will be denoted by capital Latin indices and brane indices by Greek indices.,
$$S=d^Dx\frac{1}{2\kappa _D^2}\sqrt{|G|}\left[R\frac{D2}{D1}\left(_M\varphi ^M\varphi +V[\varphi ]\right)\right]d^{D1}x\sqrt{|g|}\frac{D2}{(D1)\kappa _D^2}f[\varphi ]T,$$
(2.1)
where $`G_{MN}`$ is the $`D`$-dimensional metric and $`g_{\mu \nu }`$ is the pullback of the metric onto the flat domain wall at $`x^{D1}y=0`$. For now we will not be concerned with the origin of the coupling $`f[\varphi ]`$ to the brane, and allow it to be arbitrary; $`T`$ is the non-canonically normalized brane tension and includes standard model vacuum contributions. We look for static solutions with the metric ansatzWe will not look for non-flat solution on the brane such as de Sitter or anti–de Sitter 4D configurations. When they exist simultaneously with flat Minkowskian solutions, it is a dynamical question to know which solution is preferred by stability. A nice feature of the case with vanishing bulk potential is that the flat solutions are the only 4D maximally symmetric solutions .,
$$ds^2=e^{2A(y)/(D1)}dx_{D1}^2+dy^2.$$
(2.2)
The unconventional normalization of the action (2.1) and the warp factor in (2.2) will be convenient in what follows.
The equations of motion which follow from the action (2.1) with the ansatz (2.2) are ,
$`A^{\prime \prime }(y)=\varphi ^{}(y)^2+f[\varphi (y)]T\delta (y),`$ (2.3)
$`A^{}(y)^2=\varphi ^{}(y)^2V[\varphi (y)],`$ (2.4)
$`\varphi ^{\prime \prime }(y)A^{}(y)\varphi ^{}(y)=\frac{1}{2}{\displaystyle \frac{V[\varphi ]}{\varphi }}+{\displaystyle \frac{f[\varphi ]}{\varphi }}T\delta (y),`$ (2.5)
where the primes denote derivatives with respect to $`y`$. If $`\varphi (y)`$ is monotonic in the bulk then the bulk equations of motion can be written in a first order form introducing an auxiliary field $`W[\varphi ]`$,
$`V[\varphi ]`$ $`=`$ $`\left({\displaystyle \frac{W[\varphi ]}{\varphi }}\right)^2W[\varphi ]^2,`$ (2.6)
$`\varphi ^{}(y)`$ $`=`$ $`{\displaystyle \frac{W[\varphi (y)]}{\varphi (y)}},`$ (2.7)
$`A^{}(y)`$ $`=`$ $`W[\varphi (y)].`$ (2.8)
Because of the relation (2.6), $`W[\varphi ]`$ will be called superpotential even though no supersymmetry is involved. We will use the first order formalism in order to construct solutions in the bulk on both sides of the brane: $`W_\pm `$, $`\varphi _\pm `$ and $`A_\pm `$ on the right ($`+`$) and left ($``$) hand side. But the equations of motion (2.6)-(2.8) must then be supplemented by boundary conditions at the brane. The boundary conditions due to the delta-function terms in (2.3) and (2.5) can be written,
$`\mathrm{\Delta }\varphi ^{}=\mathrm{\Delta }{\displaystyle \frac{W[\varphi (y)]}{\varphi (y)}}=T{\displaystyle \frac{f[\varphi (y)]}{\varphi (y)}}|_{y=0}`$
$`\mathrm{\Delta }A^{}=\mathrm{\Delta }W[\varphi (y)]=Tf[\varphi (y)]|_{y=0},`$ (2.9)
where $`\mathrm{\Delta }F`$ indicates the jump of a discontinuous function $`F`$ at $`y=0`$, $`F(0^+)F(0^{})`$. In addition, $`\varphi (y)`$ and $`A(y)`$ must be continuous across the boundary. Hence, there are four boundary conditions at the brane.
A count of free parameters in the solutions to the equations of motion immediately demonstrates that given $`f[\varphi (y)]`$, the tension $`T`$ is generically not fine tuned. If we do not impose orbifold boundary conditions in addition to those above, then there are naively six free parameters: one from the solution on each side of the brane of each of equations (2.6)-(2.8). However, one overall constant shift in $`A(y)`$ is not relevant because $`A(y)`$ enters into the equations of motion only through its derivatives. That leaves five free parameters and four boundary conditions. There is generically a (finite) line of solutions for a region of scalar-brane couplings $`f[\varphi ]T`$.
Once again, this is what we mean by self-tuning: Given an arbitrary scalar-brane coupling $`f[\varphi ]`$ (possibly satisfying some constraints), there is a range of “brane tensions” $`T`$ such that static solutions exist. In the generic case, as argued above, there is in fact a continuous set of static solutions for a given boundary condition specified by $`f[\varphi ]`$ and $`T`$. Furthermore, if $`f[\varphi ]=f^{}[\varphi ]`$ then the range of $`T`$ often extends to infinity. The reason is that if $`W[\varphi ]V[\varphi ]`$ asymptotically when $`W`$ is large, then $`W^{}W`$ there, and $`T`$ can be chosen arbitrarily large such that $`(fT,f^{}T)(2W,2W^{})`$.
Orbifold boundary conditions are more constraining. The additional constraints from the orbifold condition are $`A(y)=A(y)`$ and $`\varphi (y)=\varphi (y),`$ which implies that $`W_+[\varphi ]=W_{}[\varphi ]`$, where $`W_\pm [\varphi ]`$ are the solutions for $`W`$ on the two sides of the brane. However, continuity of $`A(y)`$ and $`\varphi (y)`$ is then guaranteed. Hence, in this case there are two free parameters (there would be three but a constant shift in $`A`$ is not relevant) and two boundary conditions, and there is generically a solution for a region of couplings $`f[\varphi ]T`$. Thus, there is generically self-tuning in this case, as well.
In the absence of orbifold type boundary conditions, if there are $`N`$ branes then there are $`3N+31=3N+2`$ free parameters and $`4N`$ constraints, leaving $`2N`$ free parameters in the solution for a given set of boundary conditions. Hence, there can be up to two branes without fine-tuning. With orbifold boundary conditions, where image branes are included in $`N`$ and there is assumed to be a brane at the orbifold fixed point (which contributes 1 to $`N`$), there are $`3(N1)/2+31=(3N+1)/2`$ free parameters and $`4(N1)/2+2=2N`$ constraints, leaving $`(1N)/2`$ free parameters in the solution. Hence, only if there is a single brane at the orbifold fixed point will self-tuning occur. If the space is compactified on a circle with orbifold boundary conditions, then a parameter count for the case of branes at the two orbifold fixed points demonstrates that a fine-tuning is necessary in this case, as well . Namely, there are two free parameters in the solution but four boundary conditions. In what follows we will concentrate on the case of a single brane coupled to a scalar field.
## 3 Integrable bulk potentials
In this section we study some special cases which were also partially discussed in . Our discussion of the exact solutions with a vanishing bulk potential is similar, but we will not restrict ourselves to the ansatz $`A^{}[\varphi ]\varphi ^{}`$ made in . In agreement with the parameter count in the previous section we will find that there is no fine-tuning in these theories, although some of the exact solutions exhibit non-generic behavior.
Let us first find the exactly solvable models. The challenge is finding a class of solutions to the nonlinear equation (2.6) for a given $`V[\varphi ]`$. If (in a region where<sup>*</sup><sup>*</sup>*The case of a positive bulk potential can be studied in a very similar way up to some changes of sign in the equations (3.1)–(3.2). $`V<0`$) we write $`W[\varphi ]`$ and $`W^{}[\varphi ]`$ asFrom now, we will denote by a prime a derivative of $`V`$, $`f`$, $`W`$ or $`\omega `$ with respect to $`\varphi `$; or a derivative of $`\varphi `$ or $`A`$ with respect to $`y`$.,
$`W`$ $`=`$ $`\frac{1}{2}\sqrt{V[\varphi ]}\left(w[\varphi ]+{\displaystyle \frac{1}{w[\varphi ]}}\right),`$ (3.1)
$`W^{}`$ $`=`$ $`\frac{1}{2}\sqrt{V[\varphi ]}\left(w[\varphi ]{\displaystyle \frac{1}{w[\varphi ]}}\right),`$ (3.2)
then (2.6) is immediately satisfied for any prepotential $`w[\varphi ]`$. The consistency of (3.1) and (3.2) then translates into a differential equation for $`w[\varphi ]`$:
$$\omega ^{}=\omega \frac{V^{}}{2V}\omega \frac{\omega ^2+1}{\omega ^21}$$
(3.3)
Exact solutions for $`w[\varphi ]`$ can be found when (3.3) is separable, i.e. when $`V^{}[\varphi ]/V[\varphi ]`$ is a constant.
We distinguish three cases:
* a vanishing bulk potential: $`V[\varphi ]=0.`$
* a negative bulk cosmological constant: $`V[\varphi ]=\mathrm{\Lambda }<0.`$
* an exponential bulk potential: $`V^{}[\varphi ]/V[\varphi ]=\mathrm{const}.0.`$
### 3.1 Vanishing bulk potential
This case has been extensively studied by refs. but it is a worthwhile exercise to repeat the discussion in terms of a superpotential. The equation for the superpotential can be solved without introducing a prepotential. Indeed, eq. (2.6) becomes simply
$$W^{}[\varphi ]^2=W[\varphi ]^2,$$
(3.4)
with two branches of solutions,
$$W[\varphi ]=ce^{ϵ\varphi }.$$
(3.5)
where $`c`$ is a constant of integration and $`ϵ`$ is a sign, both of them can take different values on the two sides of the brane (the constants of integration relative to the right (left) hand side of the brane will be denoted with a $`+`$ ($``$) subscript). With this form of the superpotential, the eqs. (2.7)-(2.8) can be easily solved as
$`\varphi (y)=ϵ\mathrm{ln}(dcy)`$ (3.6)
$`A(y)=\mathrm{ln}(dcy)+e`$ (3.7)
where $`d`$ and $`e`$ are some constants of integration that can also differ on the sides of the branes. Moreover, to make sense eq. (3.6) need: $`d_+>0`$ and $`d_{}>0`$. Thus the continuity requires
$`\varphi (0)\varphi _0=ϵ_+\mathrm{ln}d_+=ϵ_{}\mathrm{ln}d_{}`$ (3.8)
$`A(0)A_0=e_+\mathrm{ln}d_+=e_{}\mathrm{ln}d_{}`$ (3.9)
while the jump equations are
$`{\displaystyle \frac{ϵ_+c_+}{d_+}}{\displaystyle \frac{ϵ_{}c_{}}{d_{}}}=f^{}[\varphi _0]T`$ (3.10)
$`{\displaystyle \frac{c_+}{d_+}}{\displaystyle \frac{c_{}}{d_{}}}=f[\varphi _0]T`$ (3.11)
From the expression of the warp factor, we conclude that the Planck scale on the brane, $`\kappa _{D1}^2=\kappa _D^2𝑑ye^{(D3)A/(D1)}`$, is finite iff singularities are encountered on both sides of the brane i.e. $`c_+>0`$ and $`c_{}<0`$.
* if $`ϵ_+ϵ_{}=1`$: the consistency of the jump equations requires that $`f^{}[\varphi _0]=ϵf[\varphi _0]`$ and then it is possible to find solutions with or without singularities for any value of the brane tension but the singular solutions correspond to $`f[\varphi _0]T>0`$.
* if $`ϵ_+ϵ_{}=1`$: the solutions with singularities exist only if $`f[\varphi _0]T>0`$ and $`1<f^{}[\varphi _0]/f[\varphi _0]<1`$ but do not require any fine-tuning.
If we choose $`f[\varphi ]=Ce^{ϵ\varphi }`$ for the case $`ϵ=ϵ_+=ϵ_{}`$ then it is clear that the boundary conditions can be satisfied for any $`T`$. There are several important comments to make about this case, which is quite non-generic. First of all, the fact that a specific form had to be chosen for $`f[\varphi ]`$ is a result of two non-generic features of the model: First, there is a symmetry $`\varphi \varphi +\text{const.}`$. As a result of this symmetry, one of the free parameters, namely $`\varphi _0`$, does not appear on the left hand side of the boundary conditions for the derivatives $`\varphi ^{}`$ and $`A^{}`$. In addition, it turned out that the left hand sides of the two boundary conditions had the same form up to a sign. As a result, only $`f^{}[\varphi ]/f[\varphi ]`$ is relevant, and given one solution $`(f[\varphi ]T,f^{}[\varphi ]T)`$, there is an infinite set of solutions with the same $`f[\varphi ]`$ and arbitrary $`T`$. The fact that $`T`$ is completely arbitrary is most likely unique to the case of vanishing bulk potential. But the fact that given $`f,f^{}𝒪(M_5)`$, self-tuning occurs for $`T`$ to within $`𝒪(M_5)𝒪(M_{EW})`$ (where $`𝒪(M_{EW})`$ is the expected Standard Model contribution to the tension) is generic, and is what we mean by self-tuning.
Even given the constraint on $`f[\varphi ]`$ from the shift symmetry in $`\varphi `$ in this case, as explained in the required exponential form of $`f[\varphi ]`$ might be natural from a stringy perspective where $`\varphi `$ is interpreted as the dilaton.
Furthermore, as pointed out in , the solutions with singularities on either side of the brane must be chosen in order to have a finite gravitational coupling (assuming that the spacetime can be cutoff at the singularity in a consistent way). This feature will turn out to be generic except in theories which admit solitonic solutions for fine-tuned boundary conditions i.e. for special discrete values of the brane tension.
### 3.2 Bulk cosmological constant
Even though the case of a (negative) bulk cosmological constant can be studied without introducing a prepotential, we will present our method on this rather simple example before proceeding to the somewhat more complicated case of an exponential bulk potential in the next subsection. We will denote by $`V[\varphi ]=\mathrm{\Lambda }<0`$ the cosmological constant. The equations of motion are:
$`{\displaystyle \frac{d\omega }{d\varphi }}`$ $`=`$ $`\omega ,`$ (3.12)
$`{\displaystyle \frac{d\varphi }{dy}}`$ $`=`$ $`\frac{1}{2}\sqrt{\mathrm{\Lambda }}\omega (1\omega ^2),`$ (3.13)
$`{\displaystyle \frac{dA}{dy}}`$ $`=`$ $`\frac{1}{2}\sqrt{\mathrm{\Lambda }}\omega (1+\omega ^2).`$ (3.14)
The differential equation for the prepotential can be easily integrated,
$$\omega =ce^\varphi ,$$
(3.15)
where $`c`$ is a constant of integration. Plugging this expression for the prepotential into the remaining equations of motion, we obtain,
$`\varphi (y)`$ $`=`$ $`\mathrm{ln}(c\vartheta (y)),`$ (3.16)
$`A(y)`$ $`=`$ $`\mathrm{ln}|\vartheta (y)|+\mathrm{ln}|1\vartheta ^2(y)|+a,`$ (3.17)
where $`a`$ is a constant of integration and another constant of integration, $`y_c`$, also appears in the expression of the function $`\vartheta (y)`$ defined, on the two different branches of solutions, by
$$\vartheta (y)=\mathrm{tanh}\frac{\sqrt{\mathrm{\Lambda }}}{2}(yy_c)\text{or }\vartheta (y)=\mathrm{coth}\frac{\sqrt{\mathrm{\Lambda }}}{2}(yy_c).$$
(3.18)
The nature of the solution depends on the sign of $`y_c`$ on each side of the brane. On the right (left) hand side, a positive (negative) value of $`y_c`$ will correspond to a solution involving a singularity at a finite proper distance from the brane, $`y=y_c`$. Near this singularity, the warp factor, $`\mathrm{exp}(2A/(D1))`$, goes to zero so the singularity appears as an horizon in the bulk. Conversely, if $`y_c^+`$, the value of $`y_c`$ on the right hand side of the brane, is negative (respectively, $`y_c^{}>0`$), then the transverse dimension will be infinitely large; however, near infinity, the warp factor blows up and the Planck scale on the brane diverges, ruining any phenomenological relevance.
The values of the constants of integration are constrained by the continuity and jump equations (with $`\vartheta _\pm =\mathrm{tanh}\sqrt{\mathrm{\Lambda }}y_c^\pm /2`$),
$`\varphi _0\mathrm{ln}(c_+\vartheta _+)=\mathrm{ln}(c_{}\vartheta _{}),`$ (3.19)
$`A_0\mathrm{ln}|\vartheta _+|+\mathrm{ln}|1\vartheta _+^2|+a_+=\mathrm{ln}|\vartheta _{}|+\mathrm{ln}|1\vartheta _{}^2|+a_{},`$ (3.20)
$`\frac{1}{2}\sqrt{\mathrm{\Lambda }}\left(\frac{1+\vartheta _+^2}{\vartheta _+}\frac{1+\vartheta _{}^2}{\vartheta _{}}\right)=f[\varphi _0]T,`$ (3.21)
$`\frac{1}{2}\sqrt{\mathrm{\Lambda }}\left(\frac{1\vartheta _+^2}{\vartheta _+}\frac{1\vartheta _{}^2}{\vartheta _{}}\right)=f^{}[\varphi _0]T.`$ (3.22)
A solution to these equations can be found, provided that $`f^{}[\varphi _0]\pm f[\varphi _0]`$, for any value of the brane tension such that
$$T^2>\frac{4\mathrm{\Lambda }}{f[\varphi _0]^2f^{}[\varphi _0]^2}.$$
(3.23)
Moreover, singularities will exist on both sides of the brane iff
$$f[\varphi _0]T>0\text{and }1<\frac{f^{}[\varphi _0]}{f[\varphi _0]}<1,$$
(3.24)
just as in the case of a vanishing bulk potential.
It is interesting to look at the $`_2`$ symmetric solution for which there are the additional constraints:
$$y_c^+=y_c^{},c_+=c_{}\text{and }a_+=a_{}.$$
(3.25)
The continuity conditions are automatically satisfied but the consistency of the jump equations requires a fine-tuning,
$$f[\varphi _0]^2f^{}[\varphi _0]^2=\frac{4\mathrm{\Lambda }}{T^2}.$$
(3.26)
As already discussed in the case with a vanishing bulk potential, this fine-tuning is a consequence of the translational symmetry, $`\varphi \varphi +\text{const.}`$, in the theory. As before, because of this shift symmetry we lose the appearance of a free parameter to adjust in the jump equations, which leads to a more restricted set of boundary conditions than for a generic bulk potential $`V[\varphi ]`$. The fine-tuning (3.26) is precisely the one appearing in the Randall–Sundrum model when the scalar coupling to the brane is a constant:
$$\mathrm{\Lambda }_{bk}=\frac{D1}{8(D2)}\kappa _D^2T_{br}^2,$$
(3.27)
where $`\mathrm{\Lambda }_{bk}`$ and $`T_{br}`$ are the canonically normalized quantities ,
$$\mathrm{\Lambda }_{bk}=\frac{D2}{2(D1)\kappa _D^2}\mathrm{\Lambda }\text{and }T_{br}=\frac{D2}{(D1)\kappa _D^2}f[\varphi _0]T.$$
(3.28)
The solution constructed by Randall and Sundrum that localizes gravity with an infinitely large extra dimension corresponds to a limit of the singular solution where the singularities are pushed to infinity i.e. $`\vartheta _\pm =\pm 1`$. The jump equations then require
$$f^{}[\varphi _0]=0\text{and }f[\varphi _0]T=2\sqrt{\mathrm{\Lambda }}$$
(3.29)
i.e. the coupling between the brane and the scalar field vanishes and the canonically normalized brane tension, $`T_{br}`$, is fine-tuned to (3.27). In this limit, the expressions for the scalar field and the warp factor simply become
$$\varphi =\varphi _0\text{and }A=\sqrt{\mathrm{\Lambda }}|y|.$$
(3.30)
### 3.3 Exponential bulk potential
The last case that can be solved analytically with our method involves an exponential potential for the scalar field in the bulk. We will concentrate on negative potential while the case of positive potential requires some minimal changes. So the bulk potential will be parametrized by two real numbers $`a`$ and $`b`$:
$$V[\varphi ]=a^2e^{2b\varphi }.$$
(3.31)
On each side of the brane, the equations of motion are simply:
$`{\displaystyle \frac{d\omega }{d\varphi }}`$ $`=`$ $`\omega b\omega {\displaystyle \frac{\omega ^2+1}{\omega ^21}},`$ (3.32)
$`{\displaystyle \frac{d\varphi }{dy}}`$ $`=`$ $`\frac{1}{2}ae^{b\varphi }\omega (1\omega ^2),`$ (3.33)
$`{\displaystyle \frac{dA}{dy}}`$ $`=`$ $`\frac{1}{2}ae^{b\varphi }\omega (1+\omega ^2).`$ (3.34)
The sign of $`a`$ is not fixed and can be chosen independently on the two sides of the brane, as we will discuss. The first differential equation can be easily solved to express $`\varphi `$ in terms of $`\omega `$:
$$e^{b\varphi }=e^{bc}\left|\omega \right|^{b/(1+b)}\left|1+b(1b)\omega ^2\right|^{b^2/(b^21)},$$
(3.35)
where $`c`$ is a constant of integration. This last result can be used to obtain a parametric representation of $`A`$ and $`y`$ as functions of $`\omega `$:
$`y(\omega )=\frac{2}{a}e^{bc}{\displaystyle _{\omega _0}^\omega }𝑑\stackrel{~}{\omega }{\displaystyle \frac{\left|\stackrel{~}{\omega }\right|^{b/(1+b)}\left|1+b(1b)\stackrel{~}{\omega }^2\right|^{b^2/(b^21)}}{\left(1+b(1b)\stackrel{~}{\omega }^2\right)}},`$ (3.36)
$`A(\omega )=A_0{\displaystyle _{\omega _0}^\omega }𝑑\stackrel{~}{\omega }{\displaystyle \frac{(1+\stackrel{~}{\omega }^2)}{\stackrel{~}{\omega }\left(1+b(1b)\stackrel{~}{\omega }^2\right)}}.`$ (3.37)
On the two sides of the brane, the parameter $`a`$ can differ by a sign while the initial bound of integration, $`\omega _0`$, and the constant of integration, $`c`$, can take any values compatible with the continuity conditions. A $`_2`$ symmetric solution will correspond to two different choices of sign for $`a`$ but the same values for $`\omega _0`$ and $`c`$.
Different kinds of solutions can be obtained depending on the value of $`b`$ and of the range of integration for the variable $`\stackrel{~}{\omega }`$:
* $`b<1`$ : in that case, $`dy/d\omega `$ has an integrable singularity at $`\omega =+\mathrm{}`$ but a non-integrable singularity at $`\omega =0`$ while the singularities of $`dA/d\omega `$ are both non-integrable. So we can find solution with or without bulk singularity at finite proper distance.
+ Solutions without singularity will be given by (3.36)-(3.37) where on the right (left) hand side of the brane, $`a`$ has to be chosen positive (negative) and $`\omega _0`$ is a negative initial bound of integration. The parameter $`\omega `$ will range from $`\omega _0`$ to $`0^{}`$. It is easy to find the asymptotic behavior of this solution for large $`|y|`$, i.e. $`\omega `$ near $`0^{}`$:
$$A\underset{|y|\mathrm{}}{}\mathrm{ln}|y|,$$
(3.38)
from where it becomes evident that the singularity free solutions do not localize gravity because the Planck scale on the brane diverges.
+ Solutions with singularity will still be given by eqs. (3.36)-(3.37) with a positive (negative) parameter $`a`$ on the right (left) hand side of the brane. And the range of integration goes from a positive initial value, $`\omega _0`$, to $`+\mathrm{}`$. In that case, $`y`$ reaches a finite value $`y_c`$ while the warp factor goes like:
$$A\underset{yy_c}{}\mathrm{ln}|yy_c|,$$
(3.39)
which indicates that the singularity is a horizon where the metric on the brane vanishes. The behavior of the warp factor near the singularity insures that the Planck scale on the brane, $`\kappa _{D1}^2=\kappa _D^2𝑑ye^{(D3)A/(D1)}`$, is finite.
* $`1<b<1`$ : in that case the singularities of $`dy/d\omega `$ at $`\omega =\pm \mathrm{}`$, $`\omega =\pm \sqrt{(1+b)/(1b)}`$ and $`\omega =0`$ are integrable while those appearing in $`dA/d\omega `$ are non-integrable. All the solution are singular with a horizon or a curvature singularity.
* $`1<b`$ : this case is quite similar to the first case because the singularity of $`dy/d\omega `$ at $`\omega =0`$ is integrable but the singularity at $`\omega =+\mathrm{}`$ is non-integrable while the two singularities of $`dA/d\omega `$ are both non-integrable. So we can construct solutions with or without singularity.
+ Solutions without singularity will be given by (3.36)-(3.37) where on the right (left) hand side of the brane, $`a`$ has to be chosen negative (positive) and $`\omega _0`$ is a positive initial bound of integration. The parameter $`\omega `$ will range from $`\omega _0`$ to $`+\mathrm{}`$. It is easy to find the asymptotic behavior of this solution for large $`|y|`$, i.e. $`\omega `$ near $`+\mathrm{}`$:
$$A\underset{|y|\mathrm{}}{}\mathrm{ln}|y|,$$
(3.40)
from where, once again, it becomes evident that the singularity free solution does not localize gravity.
+ Solutions with singularity will still be given by eqs. (3.36)-(3.37) with a negative (positive) parameter $`a`$ on the right (left) hand side of the brane. And the range of integration goes from a negative initial value, $`\omega _0`$, to $`0^{}`$. In that case, $`y`$ reaches a finite value $`y_c`$ while the warp factor goes like:
$$A\underset{yy_c}{}\mathrm{ln}|yy_c|,$$
(3.41)
which indicates that the singularity is a horizon where the metric on the brane vanishes.
* $`b=\pm 1`$ : in these two cases, we can construct singularity free solutions with a blowing up warp factor as well as singular solutions with a horizon at a finite proper distance.
Figure 1 illustrates the different types of solutions.
The main result of the analysis of these integrable bulk potentials is that we find two kinds of solutions: (i) solutions with a horizon in the bulk at a finite proper distance from the brane and with a finite lower dimensional Planck scale; (ii) solutions without singularity at a finite proper distance but associated to a bulk geometry that decouples gravity on the brane. Both kinds of solutions do not require any fine-tuning and can be constructed for a range of brane tension, $`T`$. However it seems impossible to find a singularity free solution that localizes gravity on the brane, unless the brane tension is fine-tuned as in the Randall–Sundrum model. This point surely deserves further scrutiny which the next sections will be devoted to.
## 4 Perturbative and numerical methods
In this section we give two methods for finding approximate solutions to the coupled equations of motion in the bulk that satisfy the boundary conditions at the brane. First, we present a perturbative method, which we then show to break down for the interesting case of perturbing around a solution with localized gravity. Then we give a systematic numerical method for solving the equations.
### 4.1 A perturbative method
As emphasized in the previous section, the key to finding the self-tuned solutions is to utilize the fact that one can transform the superpotential without changing the bulk potential for the scalar field. Thus, one has an additional degree of freedom if one picks a different superpotential function on the two sides of the brane, $`W_+[\varphi ]`$ and $`W_{}[\varphi ]`$. Let us now assume that we have found a static solution to the equations of motion $`\varphi _0(y)`$ and $`A_0(y)`$, and assume that we have chosen one of the integration constants such that $`A_0(0)=0`$. This is a solution for a particular value $`T`$ of the brane tension. In order to find the solution obtained by perturbing the brane tension as $`TT+\delta T`$, we first need to find how one can change the superpotential $`W`$ around $`W_0[\varphi ]`$ so as to leave the bulk potential unchanged:
$$V[\varphi ]=W_0^{}[\varphi ]^2W_0[\varphi ]^2,$$
(4.1)
which should be invariant under $`WW+\delta W`$. Linearizing (4.1) around $`W_0`$ we obtain a differential equation for $`\delta W`$ of the form
$$\frac{\delta W^{}}{\delta W}=\frac{W_0}{W_0^{}},$$
(4.2)
which is solved by
$$\delta W[\varphi ]=C\mathrm{exp}\frac{W_0[\varphi ]}{W_0^{}[\varphi ]}𝑑\varphi ,$$
(4.3)
where the arbitrary constant $`C`$ yields the extra degree of freedom needed to find a solution for any value of $`T`$. The superpotential variation can be expressed in terms of the unperturbed background solution using the equations of motion $`d\varphi _0/W_0^{}[\varphi _0]=dy`$ and $`W_0[\varphi _0]=dA_0/dy`$, we obtain
$$\delta W[\varphi (y)]=C\mathrm{exp}𝑑y\frac{dA_0}{dy}=Ce^{A_0(y)}.$$
(4.4)
From (4.3) we can see that the change in the derivative of the superpotential is given by
$$\delta W^{}[\varphi (y)]=C\frac{A_0^{}(y)}{\varphi _0^{}(y)}e^{A_0(y)}.$$
(4.5)
In order to satisfy the jump equations (2.9) for the perturbed tension $`T+\delta T`$, we need to choose $`C`$ differently on the two sides of the brane. The values of $`C_\pm `$ are then given by
$$C_\pm =\frac{f^{}[\varphi _0]f[\varphi _0]\frac{A_0^{}(0)}{\varphi _0^{}(0)}}{\mathrm{\Delta }\frac{A_0^{}}{\varphi _0^{}}}\delta T,$$
(4.6)
where $`\mathrm{\Delta }A_0^{}/\varphi _0^{}`$ denotes the jump in $`A_0^{}/\varphi _0^{}`$ at $`y=0`$. Once $`C_\pm `$ is determined from (4.6), we can simply integrate the equations
$$\delta \varphi ^{}(y)=\delta W^{}[\varphi _0(y)],\delta A^{}(y)=\delta W[\varphi _0(y)]$$
(4.7)
to obtain the perturbed solutions $`\varphi _0(y)+\delta \varphi (y)`$ and $`A_0(y)+\delta A(y)`$. This method always results in a perturbed solution. However, in the most interesting case, when the unperturbed solution asymptotes to AdS space thereby localizing gravity to the brane (that is for $`A(y)(D1)|y|/R_{AdS}`$ for large values of $`y`$), one can easily see that the perturbative method presented here always breaks down. This can be seen by inspecting (4.4), which shows that in this case $`\delta W[\varphi (y)]e^{(D1)|y|/R_{AdS}}`$. Therefore the perturbed values of $`\delta \varphi `$ and $`\delta A`$ grow exponentially, and thus the linearized approximation breaks down. We will examine the case of localized gravity in detail in Section 5. But first we give a numerical method that can be used for any choice of the bulk potential.
### 4.2 A numerical method
Next we present a method for solving the system (2.6)-(2.8) numerically for any brane tension and potential in the bulk. Thus the input functions are the bulk potential $`V[\varphi ]`$, the brane tension $`T`$, the coupling of the scalar to the brane determined by the function $`f[\varphi ]`$, and in addition we can pick the value of the scalar field at the brane $`\varphi (0)\varphi _0`$ arbitrarily. In order to find a numerical solution to these equations, one has to first make sure that the boundary conditions that one imposes do satisfy the jump equations (2.9). Our strategy is the following: we first determine the superpotential functions $`W_+[\varphi ]`$ and $`W_{}[\varphi ]`$ to the left and the right of the brane numerically such that the boundary conditions arising from the coupling to the brane are satisfied. This can be done by noting that once $`\varphi _0`$ is fixed, the jump equations are just given by
$$W_+W_{}=f_0T,W_+^{}W_{}^{}=f_0^{}T,$$
(4.8)
where $`W_\pm `$ refers to the values of the superpotential functions to the right and left of the brane at $`\varphi _0`$, $`f_0=f[\varphi _0]`$, etc. In addition, the superpotential functions must be such that they reproduce the correct value of the bulk potential at the brane:
$$W_{+}^{}{}_{}{}^{2}W_+^2=V_0,W_{}^{}{}_{}{}^{2}W_{}^2=V_0,$$
(4.9)
where $`V_0=V[\varphi _0]`$. Eqs. (4.8) and (4.9) together are enough to determine the values of both $`W_\pm `$ and $`W_\pm ^{}`$ at the branes. They are given by the expressions:
$`W_\pm =\pm {\displaystyle \frac{1}{2}}f_0T+{\displaystyle \frac{1}{2}}f_0^{}\sqrt{T^2+{\displaystyle \frac{4V_0}{f_{0}^{}{}_{}{}^{2}f_0^2}}}`$
$`W_\pm ^{}=\pm {\displaystyle \frac{1}{2}}f_0^{}T+{\displaystyle \frac{1}{2}}f_0\sqrt{T^2+{\displaystyle \frac{4V_0}{f_{0}^{}{}_{}{}^{2}f_0^2}}}.`$ (4.10)
Due to the quadratic nature of equations (4.8) and (4.9) there is a second solution, where the signs in front of the square roots in (4.2) are both simultaneously flipped. Once the value of $`W_\pm `$ and $`W_\pm ^{}`$ are fixed, one can numerically integrate the equation<sup>*</sup><sup>*</sup>*We will see in Section 5 that, unless the brane tension is fine-tuned, the superpotential, $`W[\varphi ]`$, will be a monotonic function of $`\varphi `$ and thus $`W_\pm ^{}[\varphi ]`$ will keep the sign of $`W_\pm ^{}`$.
$$W_\pm ^{}[\varphi ]=\mathrm{sgn}(W_\pm ^{})\sqrt{V[\varphi ]+W_\pm [\varphi ]^2}$$
(4.11)
to obtain the superpotential functions to the left and the right of the brane that satisfy all boundary conditions. Once $`W_\pm [\varphi ]`$ are numerically known, we can simply integrate the equations
$$\varphi ^{}(y)=W^{}[\varphi (y)],A^{}(y)=W[\varphi (y)]$$
(4.12)
to the left and the right of the brane to obtain the numerical solutions for $`\varphi (y)`$ and $`A(y)`$. We will show an example for this below for the case when the exact solution is known, and compare the two results.
### 4.3 An example for the perturbative method
In this Section we test how well the perturbative method described in 4.1 works. We will compare the analytic solution of the model with a vanishing bulk potential to the perturbed solution around a different analytic solution. The main conclusions are as expected: the perturbative method works well far from the singularities. However it gets worse as we approach the singularity itself, and does not capture the essential feature of the self-tuning solutions: whereas in the self-tuning mechanism, for a fixed value of the scalar field on the brane, the place of the singularity adjusts itself with respect to the value of the brane tension, here the singularity of the perturbed solution remains at the same place where the singularity of the unperturbed solution was. Therefore, we conclude that this method is not very efficient in capturing the basic properties of the self-tuning solutions.
The example we consider is the vanishing bulk potential discussed in 3.1, with the choice $`ϵ_+=1,ϵ_{}=1`$, and we choose $`f[\varphi ]=e^{\varphi /2}`$, $`\varphi _0=1`$, and for the unperturbed solution we choose $`T=1`$, while we pick $`\delta T=0.1`$ for the perturbed solution. The analytic solution is given by (3.6)–(3.7), with
$$d_\pm =e^{\varphi _0},c_+=\frac{3}{4}Te^{\varphi _0/2},c_{}=\frac{1}{4}Te^{3\varphi _0/2},e_\pm =\varphi _0,$$
(4.13)
where $`A`$ has been normalized to zero on the brane. The perturbed solution from (4.7) is given by
$$\delta \varphi _+(y)=\frac{\delta T}{T}\mathrm{ln}\left(\frac{d_+c_+y}{d_+}\right),\delta \varphi _{}(y)=\frac{\delta T}{T}\mathrm{ln}\left(\frac{d_{}c_{}y}{d_{}}\right).$$
(4.14)
where $`\delta \varphi _\pm `$ has been normalized to zero on the brane. The perturbed solution obtained for $`T=1,\delta T=0.1`$ compared to the exact solution for $`T=1.1`$ can be seen in Fig. 2. As mentioned above, the perturbative solution nicely follows the exact solution away from the singularity, but deviates from it close to the singularity, in particular the place of the singularity is incorrectly predicted to coincide with the singularity of the unperturbed solution.
### 4.4 An example for the numerical method
We have shown above, that the perturbative method in general does not do a good job in finding the solutions, since it becomes unreliable close to the singularities. However, the numerical solution should not have these problems. Indeed, we analyze the same example as above (the case with vanishing bulk potential) using the numerical method, and find that the exact and numerical curves are virtually indistinguishable. Therefore, we suggest that in order to analyze potentials for which no exact solutions can be found, one should use the numerical method rather than the method based on perturbations.
We are looking for a numerical solution to the case analyzed perturbatively above, that is vanishing bulk potential, $`f[\varphi ]=e^{\varphi /2}`$, $`\varphi _0=1`$, $`T=1`$ and $`ϵ_+=1,ϵ_{}=1`$. From Eqs. (3.4) and (3.8) we find the starting values of the superpotential to the left and the right of the brane:
$$W_+=\frac{3}{4}e^{\varphi _0/2}T,W_{}=\frac{1}{4}e^{\varphi _0/2}T.$$
(4.15)
Numerically integrating the equation
$$W_\pm ^{}[\varphi ]=\pm W[\varphi ]$$
(4.16)
with the boundary conditions $`W_\pm [\varphi _0]=W_\pm `$ one obtains the numerical values for $`W_\pm (\varphi )`$. Finally, the values for $`\varphi (y)`$ can be obtained by numerically inverting the integral
$$y=_{\varphi _0}^\varphi \frac{d\varphi }{W^{}(\varphi )}$$
(4.17)
to the left and right of the brane. The numerical solution obtained this way overlayed on the exact solution of Section 2 can be seen in Fig. 3. One can see that the two curves are virtually indistinguishable, suggesting that the numerical method works very well around the singularities, and should be the preferred method of looking for solution in the absence of exact solutions.
## 5 Localized gravity without singularities
### 5.1 A no-go theorem
We would like to reexamine in this section the count of free parameters versus fine-tuning parameters needed to preserve a static Poincaré invariance on the brane, with the restriction that the solution corresponds to an infinitely large extra-dimension (without singularities) in the bulk and localizes gravity on the brane.
Consider the general D-dimensional background preserving $`\text{Poincaré}_{D1}`$
$$ds^2=e^{2A(y)/(D1)}dx_{D1}^2+dy^2.$$
(5.1)
The graviton zero-mode is localized on the brane<sup>*</sup><sup>*</sup>*We do not consider the recently proposed possibility that gravity might be quasi-localized to the brane, since in those models the $`A^{\prime \prime }>0`$ condition is not satisfied, therefore it is not possible to generate those backgrounds from a single scalar field. precisely when the effective Planck scale is finite on the brane . In terms of the warp factor, this condition is:
$$\frac{1}{\kappa _{D1}^2}=\frac{1}{\kappa _D^2}𝑑ye^{(D3)A(y)/(D1)}<\mathrm{}.$$
(5.2)
It is convenient to introduce a transverse coordinate $`z`$ for which the bulk metric is conformally flat:
$$ds^2=\mathrm{\Omega }^2(z)\left(dx_{D1}^2+dz^2\right),$$
(5.3)
where the conformal factor, $`\mathrm{\Omega }`$, is related to the warp factor, $`e^{2A\left(y\left(z\right)\right)/(D1)}`$, by the two identities:
$$\mathrm{\Omega }(z)=e^{A(y)/(D1)}\text{and }\mathrm{\Omega }^2(z)dz^2=dy^2.$$
(5.4)
Then the condition (5.2) is equivalent to having a massless normalizable bound state, which is interpreted as the graviton on the brane:
$$\psi _0\mathrm{\Omega }^{(D2)/2}\text{with }𝑑z|\psi _0|^2<\mathrm{}.$$
(5.5)
Let us assume that the behavior of $`\psi _0`$ at infinity is a power law:
$$\psi _0\underset{z\mathrm{}}{}z^\alpha .$$
(5.6)
The localization of gravity (5.5) then requires: $`\alpha >1/2`$.
In our study, the value of the parameter $`\alpha `$ is constrained by the fact that the background is created by a scalar field coupled to gravity. From the equations of motion, we easily deduce that $`A`$ has to satisfy: $`d^2A/dy^20`$, which translates, in the $`z`$ coordinate, in a lower bound on the value of $`\alpha `$:
$$\alpha \frac{D2}{2}.$$
(5.7)
Furthermore it is worth noticing that an upper bound on $`\alpha `$ comes by the requirement of a geometry without singularity at a finite proper distanceFor a power law conformal factor, the curvature always vanishes at infinity. However quadratic invariants such as $`R_{M_1M_2M_3M_4}R^{M_1M_2M_3M_4}`$ will be singular at infinity as soon as $`\alpha >(D2)/2`$.. Indeed the proper distance from the brane to infinity is given: $`l_{\mathrm{}}=𝑑z\mathrm{\Omega }`$ that diverges iff:
$$\alpha \frac{D2}{2}.$$
(5.8)
So the only background for the scalar field coupled to gravity that localizes gravity without singularity (with an infinitely large extra dimension) is asymptotic, at infinity, to the horizon of an anti-de Sitter space, as in the RS model, and corresponds to $`\alpha =(D2)/2`$. In that case, the warp factor is exponentially decreasing with the proper distance to the brane:
$$A\underset{|y|\mathrm{}}{}(D1)|y|/R_{AdS}.$$
(5.9)
The aim of this section is to show that such a background necessarily requires a fine-tuning between the brane and the bulk. This is not to say that there is no self-tuning in these models, only that the nonsingular solutions require fine-tuning.
The previous asymptotic behavior has a nice interpretation in terms of the superpotential, $`W[\varphi ]`$. According to the equations of motion, the fact that $`A`$ is asymptotically linear means that $`\varphi `$ becomes constant and we will denote by $`\varphi _c^{}`$ and $`\varphi _c^+`$ the asymptotic values of $`\varphi `$ at $`y=\mathrm{}`$ and $`y=+\mathrm{}`$ respectively.
The equation
$$\frac{\varphi }{y}=\frac{dW}{d\varphi },$$
(5.10)
is similar to an RGE with $`W^{}`$ playing the role of the $`\beta `$-function. In order for $`\varphi `$ to approach a constant (fixed point) at infinity, the $`\beta `$-function $`dW/d\varphi `$ must have zeroes. In other words, in order to extend the range of the transverse coordinate from $`y=\mathrm{}`$ to $`y=+\mathrm{}`$, the values $`\varphi _c^{}`$ and $`\varphi _c^+`$ at infinity must be some roots of $`dW/d\varphi `$:
$$\frac{dW}{d\varphi }|_{\varphi _c^{}}=0\text{and }\frac{dW}{d\varphi }|_{\varphi _c^+}=0.$$
(5.11)
Furthermore, at infinity, $`A`$ has to be linearly increasing in $`|y|`$; otherwise the conformal infinity of AdS would be reached without localized gravity . Given that
$$\frac{A}{y}=W[\varphi ],$$
(5.12)
we conclude that $`\varphi _c^{}`$ and $`\varphi _c^+`$ must satisfy,
$$W[\varphi _c^{}]<0\text{and }W[\varphi _c^+]>0.$$
(5.13)
Finally, $`\varphi _c^{}`$ and $`\varphi _c^+`$ must be dynamically reached at $`y=\mathrm{}`$ and $`y=+\mathrm{}`$, which according to (5.10) is possible iff:
$$\frac{d^2W}{d\varphi ^2}|_{\varphi _c^{}}>0\text{and }\frac{d^2W}{d\varphi ^2}|_{\varphi _c^+}<0,$$
(5.14)
or at least a similar condition for the first non-vanishing higher order derivatives at $`\varphi _c^{}`$ and $`\varphi _c^+`$.
Pictorially, the previous conditions are summarized in figure 4.
Finally, the last equation of motion that relates the superpotential, $`W`$, to the scalar potential in the bulk, $`V`$, partially fixes the possible values of $`\varphi _c^{}`$ and $`\varphi _c^+`$. Indeed this differential equation evaluated at infinity gives:
$$V[\varphi _c^{}]=W[\varphi _c^{}]^2<0\text{and }V[\varphi _c^+]=W[\varphi _c^+]^2<0.$$
(5.15)
while a differentiation with respect to $`\varphi `$ gives:
$$\frac{dV}{d\varphi }=2\frac{dW}{d\varphi }\left(\frac{d^2W}{d\varphi ^2}W\right);\text{ thus }\frac{dV}{d\varphi }_{|\varphi _c^{}}=0\text{and }\frac{dV}{d\varphi }_{|\varphi _c^+}=0.$$
(5.16)
More information on $`W`$ can be obtained by considering higher order derivatives of the differential equation between $`W`$ and $`V`$. Indeed it is easy to prove by induction the following relation:
$$V^{(n)}=\underset{k=1}{\overset{n}{}}2\left(\genfrac{}{}{0pt}{}{n1}{k1}\right)W^{(k)}\left(W^{(nk+2)}W^{(nk)}\right)$$
(5.17)
where $`V^{(n)}`$ denotes the $`n^{th}`$ order derivative of $`V`$ and similarly for $`W^{(n)}`$; in addition, we will denote by $`W_{}^{(n)}{}_{c}{}^{\pm }`$ the values of $`W^{(n)}`$ at $`\varphi =\varphi _c^\pm `$. At the second order, by evaluating (5.17) at $`\varphi =\varphi _c^\pm `$, we obtain a quadratic equation for $`W_{}^{(2)}{}_{c}{}^{\pm }`$:
$$2W_{}^{(2)}{}_{c}{}^{\pm }{}_{}{}^{2}2W_c^\pm W_{}^{(2)}{}_{c}{}^{\pm }V_{}^{(2)}{}_{c}{}^{\pm }=0.$$
(5.18)
The superpotential will be real-valued provided that:
$$V_{}^{(2)}{}_{c}{}^{\pm }>V_c^\pm /2,$$
(5.19)
and then there are four different branches at each asymptotic point:
$`W_c^\pm =ϵ_1\sqrt{V_c^\pm },`$ (5.20)
$`W_{}^{(2)}{}_{c}{}^{\pm }=ϵ_1{\displaystyle \frac{\sqrt{V_c^\pm }}{2}}+ϵ_2{\displaystyle \frac{\sqrt{2V_{}^{(2)}{}_{c}{}^{\pm }V_c^\pm }}{2}},`$ (5.21)
with $`ϵ_1=\pm ϵ_2=\pm 1`$.
However the compatibility between the gravity localization requirement (5.13) and the dynamics of the differential equation (5.14) can be fulfilled iff
$$V_{}^{(2)}{}_{c}{}^{\pm }0,$$
(5.22)
and only one out of the four branches is retained:
$`\varphi _c^{}:`$ $`W_c^{}=\sqrt{V_c^{}}W_{}^{(2)}{}_{c}{}^{}={\displaystyle \frac{\sqrt{V_c^{}}}{2}}+{\displaystyle \frac{\sqrt{2V_{}^{(2)}{}_{c}{}^{}V_c^{}}}{2}},`$ (5.23)
$`\varphi _c^+:`$ $`W_c^+=\sqrt{V_c^+}W_{}^{(2)}{}_{c}{}^{+}={\displaystyle \frac{\sqrt{V_c^+}}{2}}{\displaystyle \frac{\sqrt{2V_{}^{(2)}{}_{c}{}^{+}V_c^+}}{2}}.`$ (5.24)
At higher order, the relation (5.17) becomes linear in $`W_{}^{(n)}{}_{c}{}^{\pm }`$ and allows to compute $`W_{}^{(n)}{}_{c}{}^{\pm }`$ recursively in terms of lower derivatives:
$$W_{}^{(n)}{}_{c}{}^{\pm }=\frac{V_{}^{\left(n\right)}{}_{c}{}^{\pm }{\displaystyle \underset{k=3}{\overset{n1}{}}}2\left({\scriptscriptstyle \genfrac{}{}{0pt}{}{n1}{k1}}\right)W_{}^{\left(k\right)}{}_{c}{}^{\pm }\left(W_{}^{\left(nk+2\right)}{}_{c}{}^{\pm }W_{}^{\left(nk\right)}{}_{c}{}^{\pm }\right)+2\left(n1\right)W_{}^{\left(2\right)}{}_{c}{}^{\pm }W_{}^{\left(n2\right)}{}_{c}{}^{\pm }}{2\left(nW_{}^{\left(2\right)}{}_{c}{}^{\pm }W_c^\pm \right)}.$$
(5.25)
Of course, for this expression to make sense for any integer $`n>2`$, it is important that no solution to the equation $`nW_{}^{(2)}{}_{c}{}^{\pm }W_c^\pm =0`$ can be found. However, the roots of this equation would satisfy:
$$\frac{V_{}^{(2)}{}_{c}{}^{\pm }}{V_c^\pm }=\frac{2}{n}\left(1\frac{1}{n}\right),$$
(5.26)
which is incompatible with the physical requirements (5.15) and (5.22) of localized gravity.
At this stage, it is worth noticing that the uniqueness of a superpotential $`W`$ reaching a given asymptotic point follows from our requirements of a localized gravity without singularity. Otherwise, a scalar field $`V`$ in the bulk could be constructed such that the equation $`nW_{}^{(2)}{}_{c}{}^{\pm }W_c^\pm =0`$ has some solution in which case $`W_{}^{(n)}{}_{c}{}^{\pm }`$ would not be determined, leading to a continuum of solutions.
Finally, the whole expression of $`W`$ reaching the asymptotic points $`\varphi _c^\pm `$ can be uniquely reconstructed from its derivatives through a Taylor expansion:
$`y<0:`$ $`W_{}[\varphi ]={\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{n!}}W_{}^{(n)}{}_{c}{}^{}(\varphi \varphi _c^{})^n,`$ (5.27)
$`y>0:`$ $`W_+[\varphi ]={\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{1}{n!}}W_{}^{(n)}{}_{c}{}^{+}(\varphi \varphi _c^+)^n.`$ (5.28)
From the expression of the superpotential, we can solve the equations of motion on the two sides of the brane. And the jump conditions are given by:
$$\frac{W_+[\varphi _0]W_{}[\varphi _0]}{f[\varphi _0]}=\frac{W_+^{}[\varphi _0]W_{}^{}[\varphi _0]}{f^{}[\varphi _0]}=T.$$
(5.29)
The first equation will fix the value, $`\varphi _0`$, of the scalar field on the brane while the second one fine-tunes the value of the brane tension. Generically, we will obtain only discrete values of $`\varphi _0`$. Indeed, if there exist a continuum interval of solutions for $`\varphi _0`$, then the jump equation becomes a differential equation that can be integrated on this interval and we obtain that $`f[\varphi ]`$ has to be proportional to $`W_+[\varphi ]W_{}[\varphi ]`$, but in that case there is only one possible value for the brane tension Whereas the non-canonically normalized brane tension, $`T`$, is fine-tuned, the physical brane tension, $`T_{br}f[\varphi _0]T`$, is not fine-tuned since the value of $`\varphi _0`$ can vary continuously. Whether this is a solution to the cosmological constant problem or not, beside the fine-tuning requires on $`f[\varphi ]`$, depends whether SM loops will modify $`T`$ or $`T_{br}`$. This issue deserves further analysis in future work. In all the other cases, given two asymptotic points, $`\varphi _c^\pm `$, we will obtain discrete solutions for $`\varphi _0`$ and $`T`$. Moreover, using different critical asymptotic points, we usually find different values of the brane tension and the number of values of $`T`$ that allows an infinitely large extra dimension with localized gravity is related to the number of critical points of the bulk potential $`V`$ satisfying (5.15), (5.16) and (5.22):
$$n_T(3n_C2)n_S,$$
(5.30)
where $`n_T`$ stands for the number of values of $`T`$ such that gravity is localized without a singularity, while $`n_C`$ is the number of critical points of $`V`$ as defined above. The multiplicity factor comes from the fact that the scalar field can asymptote either the same critical point or two adjacent ones on the two sides of the brane. The upper bound has been semi-quantitatively corrected by taking into account the average number, $`n_S`$, of solutions to the jump equation for $`\varphi _0`$. It may also happen that some values of $`T`$ are degenerate. We will explicitly describe an example in the next subsection.
It is worth noticing that in the case of an oscillatory bulk potential, we will obtain an infinite number of discrete values for the brane tension that is as quantized.
To complete the proof of the no-go theorem presented above, we need to show that there is no loophole in the above argument due to the fact that we have used the superpotential formalism. The subtlety that one might worry about is that in the proof above we have implicitly assumed that $`\varphi `$ is monotonic, by writing the second order equations (2.3)-(2.4) in terms of first order equations involving $`W`$. In particular, if $`\varphi `$ is not monotonic (that is if $`\varphi ^{}=0`$ at a finite value of $`y`$) one does not have a globally defined superpotential function $`W(\varphi )`$, but instead one must define separate superpotential functions $`W_i`$ for the regions between $`y_i`$ and $`y_{i+1}`$, where $`\varphi ^{}(y_i)=\varphi ^{}(y_{i+1})=0`$. For these superpotentials that are not globally defined it is then possible to have $`W^{}(\varphi _{})=0`$ without satisfying $`V^{}(\varphi _{})=0`$. However, it is impossible to continue the solution beyond $`\varphi _{}`$. This by itself however may not be a problem, as long as at $`\varphi _{}`$ one is smoothly switching over to another branch of $`W`$. Of course this switch-over can only happen at a point where $`\varphi ^{}=0`$, since otherwise $`\varphi `$ is monotonic, and one can solve the equations in terms of the superpotential, which is well-defined around $`\varphi _{}`$. Next we show that such possibilities do not get around the no-go theorem presented above. The reason is that in order to have localized gravity with an infinitely large extra dimension, we need $`\varphi ^{}0,\varphi ^{\prime \prime }0`$ for $`|y|\mathrm{}`$, and therefore we find from the second order equation (2.5) that $`V^{}0`$. Thus the critical points at infinity must belong to an “ordinary branch” described above, where $`W`$ can be continued at both sides of $`\varphi _c`$. However, once we are on an “ordinary branch” which can be globally defined, all the critical points will actually happen at $`V^{}=0`$. Since the only possibility for switching over to another branch is at $`W^{}=0`$, and at those points $`V^{}=0`$, one can never switch off the ordinary branch, and therefore one can not circumvent the no-go theorem by gluing non-monotonic $`\varphi `$’s together.
### 5.2 A numerical example
In this section we demonstrate the ideas of the previous sections in a numerical example. The bulk potential in this example (Fig. 5) is
$$V(\varphi )=\varphi ^6+11\varphi ^47\varphi ^2+1,$$
(5.31)
which is generated by the superpotential $`W[\varphi ]=\varphi ^3\varphi `$.
As in some of our previous illustrative examples the potential (5.31) is unbounded from below, and so the theory might be unstable to quantum fluctuations. However, we will only be concerned with static solutions and for our purposes any instabilities will not be relevant.
As emphasized in Figure 5, the bulk potential (5.31) has two negative stationary points at $`\varphi =\pm 1/\sqrt{3}`$. According to our no-go theorem, there are isolated superpotentials solving the equations of motion with critical points at $`\pm 1/\sqrt{3}`$. Here these superpotentials are very simple because from (5.25) only a finite number of derivatives are non-vanishing thus the superpotentials are polynomial and take the form
$`W[\varphi ]=\varphi ^3\varphi `$ $`\text{reaches }\varphi _c=\pm 1/\sqrt{3}\text{on the negative (positive) branch},`$ (5.32)
$`W[\varphi ]=\varphi ^3+\varphi `$ $`\text{reaches }\varphi _c=\pm 1/\sqrt{3}\text{on the positive (negative) branch}.`$ (5.33)
Depending on which critical point we want to asymptote at infinity, we can construct four types of solutions that localizes gravity
* for $`(\varphi _c^+=1/\sqrt{3},\varphi _c^{}=1/\sqrt{3})`$ : the solution is
$`\varphi (y)=\frac{1}{\sqrt{3}}\mathrm{tanh}(\sqrt{3}|yy_c^\pm |)`$ (5.34)
$`A(y)=\frac{1}{18}\mathrm{tanh}^2(\sqrt{3}(yy_c^\pm ))+\frac{2}{9}\mathrm{ln}\mathrm{cosh}(\sqrt{3}(yy_c^\pm ))+a_\pm `$ (5.35)
where $`y_c^\pm `$ and $`a_\pm `$ are four constants of integration to be determined by the continuity and jump conditions. The continuity conditions imply that $`y_c^+=y_c^{}`$ and $`a_+=a_{}`$. The jump equations will depend on the precise form of $`f[\varphi ]`$. Except in the degenerate case where $`f[\varphi ]W_+[\varphi ]W_{}[\varphi ]`$ that has been discussed in a footnote on page 5.1, we will generically obtain a finite number of discrete values for $`\varphi _0`$. For instance in the case of an exponential coupling, $`f[\varphi ]=ae^{b\varphi }`$, the values of $`\varphi _0`$ will satisfy
$$\frac{3\varphi _0^21}{\varphi _0(\varphi _0^21)}=b,$$
(5.36)
which admits three solutions whatever the value of $`b`$ is. And thus there exist three values for the brane tension that will lead to a localized gravity.
* for $`(\varphi _c^+=1/\sqrt{3},\varphi _c^{}=1/\sqrt{3})`$ or $`(\varphi _c^+=1/\sqrt{3},\varphi _c^{}=1/\sqrt{3})`$ : in those cases, there is no discontinuity in the superpotential and the brane tension has to vanish.
* for $`(\varphi _c^+=1/\sqrt{3},\varphi _c^{}=1/\sqrt{3})`$ : this case is analogous to the first one and, for an exponential coupling, three values for the brane tension are possible and they are just the opposite of the ones obtained in the first case.
Besides the previous fine-tuned solutions with an infinitely large extra dimension and localized gravity, all other solutions will either decouple gravity on the brane or involve a horizon at a finite distance. We would like now to examine numerically the self-tuning of these singular solutions. To do that we scan for solutions to (2.6) by fixing the value of $`W[\varphi ]`$ at $`\varphi =0`$. The solutions are plotted in Fig. 6. Note that in agreement with the no-go theorem of the previous section there is a unique solution in this branch of solutions with stationary points at the positions of the minima of $`V[\varphi ]`$. As a result of the uniqueness of the solitonic solution, there are no solutions (in this branch) with $`0<|W[0]|<W[\varphi _{}].385`$, where $`\varphi _{}`$ is the value of the field at the position of the local minimum of $`V[\varphi ]`$, in this case $`\varphi _{}.577`$. Also in agreement with the results of previous sections, the solitonic solution is the unique regular solution for $`\varphi (y)`$ and $`A(y)`$, which in this case approximates the Randall-Sundrum solution for large $`|y|`$.
In order to determine the range of boundary conditions at the brane which could be satisfied, we note that except for the solitonic solution, all other solutions span an infinite range in $`\varphi `$. As noted in Section 4.2 the boundary conditions can be expressed in terms of $`W[\varphi ]`$ and $`W^{}[\varphi ]`$ at the boundary. Hence, if there is a space-filling range of solutions $`(W[\varphi ],W^{}[\varphi ])`$ then no fine-tuning is necessary in order to have a static solution with arbitrary boundary conditions in that range. In other words, a given $`(2W[\varphi _0],2W^{}[\varphi _0])`$ can be equated with a certain (orbifold) boundary condition $`(f[\varphi _0]T,f^{}[\varphi _0]T)`$. Then, given one such $`(f[\varphi _0]T,f^{}[\varphi _0]T)`$ there is a continuous set of solutions around that point. That means that for a given $`(f[\varphi _0],f^{}[\varphi _0])`$ in a certain range of $`f^{}/f`$ there is a range of tensions $`T`$ for which there is a solution; hence the theory is self-tuning. Fig. 7 is a parametric plot of $`W[\varphi ]`$ versus $`W^{}[\varphi ]`$ for several solutions of $`W[\varphi ]`$. It is intended to illustrate the fact that there is a space-filling region for $`W>.385`$. Given any boundary condition parametrized by $`f[\varphi ]`$ and $`T`$, $`T`$ can be rescaled by an amount given by the intersection of the set of solutions $`(2W,2W^{})`$ with a line through the origin and $`(fT,f^{}T)`$. As is generically expected, one can see from Figure 7 that if $`f,f^{}`$ are $`𝒪(1)=𝒪(M_5)`$, then $`T`$ can be rescaled by $`𝒪(M_5)`$ without eliminating a solution. Hence we have demonstrated the self-tuning of this model. However, the caveat is that because of the isolated solitonic solution, there is a fine-tuned region near $`W,W^{}𝒪(1)`$. If $`f`$ is $`𝒪(1)`$ at some matching point $`\varphi _0`$, and $`T`$ is $`𝒪(M_{EW}/M_5)1`$ then a fine tuning is reintroduced because of the uniqueness of the solitonic solution. There is still a large region of parameter space where the theory is self-tuning, but whether that region is natural or not requires exploration. This phenomenon is a result of the existence of solitonic solutions, and is nongeneric. Furthermore, if we do not require orbifold boundary conditions, then once again self-tuning is natural. Note also that asymptotically the parametric plots of $`W`$ vs. $`W^{}`$ include the line $`W=W^{}`$. This is generic in any region where the solutions satisfy $`W[\varphi ]V[\varphi ]`$. This also implies that for any solution $`ff^{}`$, there is a very large range for $`T`$ for which there are solutions. This behavior for large values of $`W`$ is common, and extends the range of $`T`$ over which self-tuning occurs possibly to $`\pm \mathrm{}`$ if $`f[\varphi ]=f^{}[\varphi ]`$, which may be natural from a stringy perspective .
## 6 Resolution of singularities and fine-tuning
In this section we reconsider the resolution of singularities<sup>*</sup><sup>*</sup>* As they stand in our solutions, Einstein’s equations are not satisfed at the singularity but require a singular stress-energy tensor located at the singularity that may correspond to a brane: the introduction of this brane, for instance, is what we mean by resolution of the singularity. as proposed in . As we have seen, the case of zero bulk potential, which is the case studied in , is quite non-generic. There is a shift symmetry in the scalar field which makes the boundary conditions more constraining, and two of the boundary conditions turn out to have the same form. Hence, we study the generic case, and propose a new resolution of the singularity which may restore self-tuning.
The idea of is to add a brane at each of the singularities such that the equations of motion, or boundary conditions, are satisfied there, as well. As pointed out in See also the first reference in for a discussion on vanishing 4D effective vacuum energy. the singularities contribute to the effective 4D energy density when integrated over the extra dimension, and the contribution of the singularities is essential for vanishing of the 4D cosmological constant.
However, addition of branes at the singularities adds new boundary conditions. In order to answer the question of whether or not the theory is self-tuning we must better understand the continuation past the singularities. In it is proposed that the spacetime is either periodically continued or cut-off at the singularity. In the case that the spacetime is cut-off at the singularities on each side, there are generically two new boundary conditions at each singularity, but no additional free parameters. Hence, without orbifold boundary conditions there are $`3+31=5`$ free parameters and $`4+2+2=8`$ boundary conditions, and the system is overconstrained. If one imposes orbifold boundary conditions, then there are only half as many additional boundary conditions ($`2+2=4`$ boundary conditions in all), but only $`31=2`$ free parameters. In either case the system is overconstrained. Hence, a fine-tuning is required. Although in the absence of a bulk potential the situation is modified because of the non-generic features mentioned above, in that case a fine-tuning at each of the singularities is required, as well.
Although the details of any continuation of the spacetime beyond the singularity will depend on quantum gravitational dynamics, we propose another scenario which does slightly better than the cut-off scenario, although fine-tuning will still be required. The singularities are generically horizons (c.f. Sections 3.3) because the warp factor $`e^{2A(y)/(D1)}`$ vanishes there. Hence, light does not cross the horizon and we can imagine a scenario in which there are bulk fields living beyond the singularity, out of causal contact with our universe. The Penrose diagram for this scenario is illustrated in Figure 8. However, the bulk fields across the horizon would provide an additional three free parameters which one might hope would help avoid fine-tuning. More precisely, following the counting in Section 2, if there are orbifold boundary conditions, then there are $`3+31=5`$ free parameters from the bulk fields on both sides of the singularity, but $`2+4=6`$ boundary conditions. Hence, although the system is less constrained than the case in which the spacetime is cut-off at the singularity, one fine-tuning is still required.
One might suspect that additional fields would add additional degrees of freedom which could help in self-tuning. For example, the addition of a scalar field with second derivatives in its equation of motion would contribute two additional free parameters on each side of a brane, but only two boundary conditions (continuity and change in the derivative at the brane). Hence in a system of branes there are net, at least naively, two additional free parameters from the extra scalar field, which could be used to restore self-tuning at the singularities or perhaps even produce nonsingular solutions. However, a more detailed analysis is required in this case.
## 7 Conclusions
We have studied brane worlds coupled to a scalar field and have found that self-tuning is a generic feature of these models. In these models the dynamics of the scalar field provides additional degrees of freedom, which generically alleviates the need for fine-tuning of static solutions. We have reexamined the exactly solvable models, two of which were studied previously , and have found that those case are more constrained than the generic case because of a shift symmetry in the scalar field in these models. Still, these theories are self-tuning, including the case of an exponential potential. Whereas in a fine-tuning was necessary in this case due to a particular ansatz for the scalar and graviton fields, we showed that the more general solution is not fine-tuned, in agreement with a counting of free parameters in these models. We demonstrated that singularities in the self-tuned solutions are generic if gravity is to be localized, and we presented a no-go theorem to this effect. We provided perturbative and numerical techniques in order to calculate the self-tuned solutions, and we illustrated the major points of the paper via several numerical examples. Finally, we pointed out that the fine-tuning that is required in order to resolve the singularities in the spirit of for the case of zero bulk potential is generic.
## Acknowledgments
We are extremely grateful to Chris Kolda for collaboration at early stages of this project. We thank Nima Arkani-Hamed, Tanmoy Bhattacharya, Pierre Binétruy, Martin Schmaltz and Anupam Singh for useful discussions. C.C. thanks the Theory Group at Berkeley for hospitality while this work was initiated. C.C. is a J. Robert Oppenheimer fellow at the Los Alamos National Laboratory. C.C., J.E. and T.H. are supported by the US Department of energy under contract W-7405-ENG-36. C.G. is supported in part by the US Department of energy under Contract DE-AC03-76SF00098 and in part by the National Science Foundation under grant PHY-95-14797.
|
warning/0004/math0004084.html
|
ar5iv
|
text
|
# Groupoids and the integration of Lie algebroids
## Introduction
Differentiable groupoids appear in geometry in various instances, for example in the theory of connections and parallel transport on fiber bundles, or in the theory of pseudogroups of transformations. The concept of groupoid is a generalization of the concept of group, the main difference being that not any two elements of a groupoid are composable. Intuitively, it is convenient to think of a groupoid as a set of arrows between various points, called units, two arrows being composable if, and only if, their ends match. (See Section 1 for precise definitions.)
One of the main features of differential groupoids is that they are geometric objects that interpolate between differentiable manifolds and Lie groups: differentiable manifolds have many units and few arrows; whereas Lie groups have many arrows and few units–actually only one. This “interpolation” property is valid also at the level of algebras: to a compact smooth manifold $`M`$ one associates the commutative algebra $`𝒞^{\mathrm{}}(M)`$ of its differentiable functions; whereas to a Lie group $`G`$ one associates the convolution algebra $`𝒞_\text{c}^{\mathrm{}}(G)`$ of compactly supported smooth functions on the group, which is usually highly non-commutative. In this way, differentiable groupoids provide a link between geometry and harmonic analysis.
The algebras $`𝒞^{\mathrm{}}(M)`$ and $`𝒞_\text{c}^{\mathrm{}}(G)`$ are particular cases of the convolution algebra of a differential groupoid, and this feature makes groupoids a favorite toy model in non-commutative geometry. From this point of view, the results of this paper are a first step towards a generalization of the results of , from étale groupoids to general differential groupoids.
Recall that a Lie algebroid is a vector bundle $`AM`$ on a differentiable manifold $`M`$ together with a Lie algebra structure on the space $`\mathrm{\Gamma }(A)`$ of its smooth sections and a Lie algebra morphism
$$\mathrm{\Gamma }(A)\mathrm{\Gamma }(TM),$$
defined by a structural vector bundle morphism $`q:ATM`$, called “the anchor map.” The Lie algebroid $`A`$ is called regular \[resp. transitive\] if, and only if, the anchor map $`q:ATM`$ has locally constant rank \[resp. it is onto\]. From a classical, differential geometric point of view, to a differential groupoid there is associated a “Lie algebroid,” which is some sort of “infinitesimal form” of the groupoid, generalizing both the Lie algebra of a group and the tangent space to a differentiable manifold.
In it was proved that any Lie algebroid such that $`q=0`$ is the Lie algebroid of a differentiable groupoid, in other words, it is integrable. This result can be regarded as a generalization of “Lie’s third theorem,” which states that every finite dimensional Lie algebra is the Lie algebra of a Lie group. This is relevant because, unlike (finite dimensional) Lie algebras, Lie algebroids do not always correspond to differential groupoids, that is, they are not always integrable. Actually, for a transitive groupoid, one can define an obstruction to integrability, see . Nevertheless, it is still interesting to construct differentiable groupoids that integrate specific Lie algebroids. Some examples can be found in . In that paper, to a differentiable groupoid $`𝒢`$ there was associated an algebra of pseudodifferential operators $`\mathrm{\Psi }^{}(𝒢)`$, such that, for $`𝒢^{(0)}`$ compact, all first order differential operators in $`\mathrm{\Psi }^{}(𝒢)`$ are linear combinations of sections of $`A(𝒢)`$ and operators of multiplication by functions. The integration of Lie algebroids is thus a first step toward constructing algebras of pseudodifferential operators, and this explains why we are interested in the problem of integrating Lie algebroids. See for more on the question of constructing pseudodifferential operators.
In this paper, we approach the problem of integrating Lie algebroids from an abstract point of view, looking for some general methods to integrate Lie algebroids. Since arbitrary Lie algebroids $`\pi :AM`$ behave rather wildly, we make two assumptions. First, we assume that the manifold $`M`$ has a stratification
$$M=S$$
into disjoint strata, each of which is invariant with respect to the diffeomorphisms generated by the sections of $`A`$ and, second, we assume that the restrictions
$$A_S=A|_S,$$
which are Lie algebroids precisely because the strata are invariant, are regular. (If $`M`$ satisfies the first assumption, we say that it has an “$`A`$-invariant regular stratification.”)
For the class of groupoids with $`A`$-invariant regular stratifications, one can approach the “integration problem” in two steps. First, a necessary condition for the integrability of $`A`$ is the integrability of each of the restrictions $`A_S`$, and hence the first step will be to integrate each of these restrictions (see also below). Assume then that we can find a differentiable groupoid $`𝒢_S`$ that integrates $`A_S`$, for each $`S`$. The second step is to “glue” the resulting groupoids $`𝒢_S`$. Surprisingly enough, this naïve approach actually works in most cases. It works for example if we choose the integrating groupoids $`𝒢_S`$ to be maximal in a suitable sense ($`d`$-simply connected), and this is our main general result on the integration of differential groupoids. As in the paper of Douady and Lazard , we obtain in general non-Hausdorff groupoids.
Since transitive Lie algebroids are a particular case of regular Lie algebroids, the first part of the problem–that is, integrating regular Lie algebroids–is similar to the problem of integrating transitive Lie algebroids, and probably can be handled similarly. In particular, it is clear that not all regular Lie algebroids are integrable. It is not our purpose in this paper to study the integration of general regular algebroids, but we do show how to integrate particular classes of regular algebroids. For example, we show that a regular algebroid $`A`$ is integrable if $`\mathrm{ker}(q)`$, the kernel of $`q`$, consists of semisimple Lie algebras, or if $`A`$ is a semi-direct product.
As for the second part of the problem, it turns out that there exists at most one way to glue the groupoids $`𝒢_S`$ that integrate $`A_S=A|_S`$, assuming that they exist. The problem is that the resulting glued space (a groupoid) is not always a smooth manifold, so this procedure does not lead directly to a differentiable groupoid. However, we show that this procedure does lead to a differentiable groupoid that integrates $`A`$, provided that all groupoid $`𝒢_S`$ are $`d`$-simply connected. We do not assume that the strata are regular here. We thus obtain the following result.
###### Theorem 1.
A Lie algebroid $`\pi :AM`$ on a manifold $`M`$ with an $`A`$-invariant stratification is integrable if, and only if, it is integrable along each stratum.
These results provide us with an explicit way of integrating many Lie algebroids. As an application of the theorem, we prove the integrability of certain Lie algebroids on foliated manifolds with corners. A foliated manifold with corners is a manifold with corners, each of whose open faces is a foliated manifold, the foliations being required to satisfy certain compatibility relations. This result generalizes a construction due to Winkelnkemper . Previously, Melrose and Mazzeo–Melrose have shown how to integrate certain particular algebroids. However, their framework was different from ours.
The results of this paper, together with the results of , can be used to construct a natural algebra of pseudodifferential operators on a complex algebraic variety endowed with a “$`𝒞^{\mathrm{}}`$–resolution of singularities,” thus making a substantial step towards a solution of the problem stated in . Then the methods of can presumably be applied to study the resulting algebras of operators. Algebras of pseudodifferential operators on groupoids are also a natural framework to study adiabatic limits . The problem of associating an algebra of pseudodifferential operators to a groupoid was first formally stated in a paper by Weinstein, . However, before that, in , Alain Connes has constructed algebras of pseudodifferential operators on foliations, which in our setting corresponds to the case of a regular groupoid with discrete holonomy (see also ). His methods have played a role in inspiring the constructions of .
I would like to thank Alan Weinstein for several useful comments on an earlier version of this paper. Also, I would like to thank an anonymous referee for very carefully reading this paper and for several useful suggestions.
## 1. Basic concepts
We begin this section by fixing notation and recalling some of the basic concepts used in this paper.
In the following, we shall use the framework of . In particular, a manifold is a smooth manifold, possibly with corners. By definition, every point $`m`$ in a manifold with corners $`M`$ has a coordinate neighborhood diffeomorphic to $`[0,1)^k\times ^{nk}`$, such that the transition functions are smooth (including on the boundary). Let $`k(m)`$ be the least $`k`$ such that $`m`$ is in a set diffeomorphic to $`[0,1)^k\times ^{nk}`$, and let $`_k(M)`$ be the set of points $`m`$ for which $`k(m)=k`$. A component of $`_k(M)`$ is called an open face of codimension $`k`$. A face of $`M`$ is the closure of an open face (of $`M`$). A hyperface of $`M`$ is a face $`H`$ of codimension one.
It is customary to assume that any hyperface $`H`$ of a manifold with corners $`M`$ is of the form $`H=\{x_H=0\}`$, where $`x_H`$ is a smooth positive function on $`M`$ such that $`dx_H0`$ on $`H`$. If this is the case, $`x_H`$ is called a defining function of $`H`$. We shall also assume in this paper that each hyperface of $`M`$ has a defining functions, although, most of our results are true even without this assumption. An interior point of $`M`$ is a point of $`M`$ that belongs to no hyperface.
A submersion $`f:MN`$ between two manifolds with corners is a differentiable map with surjective differential at each point, such that a non-zero tangent vector to $`M`$ points inward if, and only if, its image in $`TN`$ is non-zero and points inward. By definition, if $`f:MN`$ is a submersion, then the fibers $`f^1(y)`$ are smooth manifolds without corners. A submanifold with corners $`NM`$ is a closed submanifold $`N`$ of $`M`$ such that each face of $`N`$ is locally the transverse intersection of $`N`$ with a face of $`M`$.
We now define groupoids. Recall that a small category is a category whose class of morphisms is a set. By definition, a groupoid is a small category $`𝒢`$ in which every morphism is invertible. General results on groupoids can be found in .
We now fix some notation and make the definition of a groupoid more explicit. The set of objects (or units) of $`𝒢`$ is denoted by $`𝒢^{(0)}.`$ The set of morphisms (or arrows) of $`𝒢`$ is denoted by $`𝒢^{(1)}=\mathrm{Mor}(𝒢).`$ We shall sometimes write $`𝒢`$ instead of $`𝒢^{(1)}`$, by abuse of notation. For example, when we consider a space of functions on $`𝒢`$, we actually mean a space of functions on $`𝒢^{(1)}`$. We will denote by $`d(g)`$ \[respectively $`r(g)`$\] the domain \[respectively, the range\] of the morphism $`g:d(g)r(g).`$ We thus obtain functions
(1)
$$d,r:𝒢^{(1)}𝒢^{(0)}$$
that will play an important role. The multiplication $`\mu :(g,h)\mu (g,h)=gh`$ is defined on the set $`𝒢^{(2)}`$ of composable pairs of arrows:
(2)
$$\mu :𝒢^{(2)}=𝒢^{(1)}\times _M𝒢^{(1)}:=\{(g,h):d(g)=r(h)\}𝒢^{(1)}.$$
The inversion operation is a bijection $`\iota (g)=g^1`$ of $`𝒢^{(1)}`$. Denoting by $`u(x)`$ the identity morphism of the object $`x𝒢^{(0)}`$, we obtain an inclusion $`u:𝒢^{(0)}𝒢^{(1)}`$. We see that a groupoid $`𝒢`$ is completely determined by the spaces $`𝒢^{(0)}`$ and $`𝒢^{(1)}`$ and by the structural morphisms $`d,r,\mu ,u`$, and $`\iota `$. We sometimes write
$$𝒢=(𝒢^{(0)},𝒢^{(1)},d,r,\mu ,u,\iota ).$$
The structural maps satisfy the following properties:
(i) $`r(gh)=r(g)`$, $`d(gh)=d(h)`$ for any pair $`(g,h)𝒢^{(2)}`$, and the partially defined multiplication $`\mu `$ is associative.
(ii) $`d(u(x))=r(u(x))=x`$, $`x𝒢^{(0)}`$, $`u(r(g))g=g`$, and $`gu(d(g))=g`$, $`g𝒢^{(1)}`$, and $`u:𝒢^{(0)}𝒢^{(1)}`$ is injective.
(iii) $`r(g^1)=d(g)`$, $`d(g^1)=r(g)`$, $`gg^1=u(r(g))`$, and $`g^1g=u(d(g))`$.
By definition, a differentiable groupoid is a groupoid such that $`𝒢^{(0)}`$ and $`𝒢^{(1)}`$ are smooth manifolds with corners, $`𝒢^{(0)}`$ is smooth, all structural morphisms are differentiable, and $`d`$, the domain map, is a submersion.
We observe that $`\iota `$ is a diffeomorphism and hence $`d`$ is a submersion if, and only if, $`r=d\iota `$ is a submersion. Also, it follows from the definition of submersions of manifolds with corners that each fiber $`𝒢_x=d^1(x)𝒢^{(1)}`$ is a smooth manifold without corners whose dimension $`n`$ is constant on each connected component of $`𝒢^{(0)}`$. The étale groupoids considered in are extreme examples of differentiable groupoids (corresponding to $`dim𝒢_x=0`$). Note that we allow $`𝒢^{(1)}`$ to be non-Hausdorff. If we want to make more precise the space of units $`𝒢^{(0)}`$ of $`𝒢`$, we say that “$`𝒢`$ is a differentiable groupoid on $`𝒢^{(0)}`$.”
We now recall the definition of a Lie algebroid . See also .
###### Definition 1.
A Lie algebroid over a manifold $`M`$ is a vector bundle $`A`$ over $`M`$ together with a Lie algebra structure on the space $`\mathrm{\Gamma }(A)`$ of smooth sections of $`A`$, and a bundle map $`q:ATM`$, extended to a map between sections of these bundles, such that:
(i) $`q([X,Y])=[q(X),q(Y)]`$, and
(ii) $`[X,fY]=f[X,Y]+(q(X)f)Y`$,
for all smooth sections $`X`$ and $`Y`$ of $`A`$ and all smooth function $`f`$ on $`M`$.
The morphism $`q`$ is called the anchor map. Note that we allow the base $`M`$ in the definition above to be a manifold with corners. This is necessary in the most interesting examples related to singular spaces. We did not include in the definition the condition that $`q(\mathrm{\Gamma }(A))`$ consist of vector fields tangent to each face of $`M`$, but it will be satisfied in all the cases we consider.
To a differential groupoid $`𝒢`$ there is naturally associated a Lie algebroid $`A(𝒢)`$ as follows . Consider first the vertical tangent bundle of $`𝒢`$ along the fibers of the domain map $`d`$:
$$T_d𝒢=\mathrm{ker}(d_{})=\underset{x𝒢^{(0)}}{}T𝒢_xT𝒢^{(1)}.$$
By definition, $`A(𝒢)`$ is the restriction $`T_d𝒢|_{𝒢^{(0)}}`$ of $`T_d𝒢`$ to the set of units of $`𝒢`$. The space of $`d`$–vertical vector fields invariant with respect to right translations is closed with respect to the Lie bracket and identifies canonically with $`\mathrm{\Gamma }(A)`$. Thus we obtain a Lie algebra structure on $`\mathrm{\Gamma }(A)`$. The action on functions on the base is obtained by lifting a function on $`𝒢^{(0)}`$ to a function on $`𝒢^{(1)}`$ via $`r`$. Consequently, the anchor map $`q:A(𝒢)T𝒢^{(0)}`$ is obtained by restricting the differential of $`r`$ to $`A(𝒢)`$.
If $`A`$ is a Lie algebroid and $`𝒢`$ is a smooth groupoid such that $`A(𝒢)A`$, then we say that $`𝒢`$ integrates $`A`$. Not every Lie algebroid is integrable (see for an example). Nevertheless, it is important to provide examples of general methods to integrate Lie algebroids.
A differentiable groupoid is called $`d`$-connected \[respectively, $`d`$-simply connected\] if, and only if, each set $`𝒢_x`$ is path connected \[respectively, path connected and simply connected\]. As for Lie groups, all $`d`$-simply connected differential groupoids with isomorphic Lie algebroids are isomorphic.
If $`𝒢`$ is a differentiable groupoid, then there exists a $`d`$-simply connected groupoid $`𝒫𝒢`$, uniquely determined up to isomorphism, with the same Lie algebroid as $`𝒢`$; it is called the path groupoid associated to $`𝒢`$. As a set, $`𝒫𝒢`$ consists of fixed end-point homotopy classes of paths $`\gamma :[0,1]𝒢`$ such that $`d(\gamma (t))`$ is constant and $`\gamma (0)`$ is a unit. This result is due to Moerdijck in general (see also for a particular case).
## 2. A glueying theorem
Let $`M`$ be a manifold with corners. In this paper, by stratified manifold we mean a smooth manifold $`M`$, possibly with corners, together with a disjoint union decomposition $`M=S`$ of $`M`$ by a locally finite family of submanifolds $`S`$ without corners, called open strata, such that the closure (in $`M`$) of each stratum $`S`$ is a submanifold with corners and each $`S`$ is contained in a unique open face of $`M`$. (This notion is slightly stronger than that of a stratified space which is also a manifold with corners. Most of our results however hold in this greater generality.)
Consider a differentiable Lie algebroid $`A`$ with anchor map $`q:ATM`$ on a manifold $`M`$. A stratification $`M=S`$ of $`M`$ is called $`A`$-invariant if, and only if, for each point $`xS`$, the range of $`A_xT_xM`$ (from the fiber of $`A`$ at the point $`x`$ to the tangent space to $`M`$ at $`x`$) is contained in $`T_xS`$ (i.e., $`q(A_x)T_xS`$). The condition that
(3)
$$q(A_x)T_xS,$$
for all $`xS`$, is equivalent to the condition that each local diffeomorphism of the form $`\mathrm{exp}(q(X))`$, for some smooth section $`X`$ of $`A`$, preserve the strata of $`M`$, or, to the condition that the restriction $`A_S`$ of $`A`$ to $`S`$ be a Lie algebroid on $`S`$, for each $`S`$.
Define, for any subset $`SM`$ of the set of units of $`𝒢`$, the groupoid
$$𝒢_S:=d^1(S)r^1(S),$$
called the reduction of $`𝒢`$ to $`S`$. If $`M=S`$ is an $`A`$-invariant stratification of $`M`$ and, moreover, $`𝒢`$ is a differentiable groupoid on $`M`$ that integrates $`A`$, then $`d^1(S)=r^1(S)`$ and $`𝒢_S=d^1(S)`$ satisfies $`A_SA(𝒢_S)`$. This shows that, in order to integrate $`A`$, we need to integrate each restriction $`A_S`$. The main point of this section is that this is also enough (Theorem 3).
Recall from that an admissible section of a differential groupoid $`𝒢`$ on $`M`$ is a differentiable map
$$\sigma :M𝒢$$
such that $`d(\sigma (x))=x`$ and the map $`Mxr(\sigma (x))M`$ is a diffeomorphism. Then $`\sigma `$ also defines a diffeomorphism
(4)
$$𝒢g\sigma g:=\sigma (r(g))g𝒢.$$
The main example of an admissible section is
$$\sigma (x)=\mathrm{exp}(X_m)\mathrm{}\mathrm{exp}(X_1)x,$$
for suitable, smooth sections $`X_1,\mathrm{},X_m`$ of $`A`$. We will discuss this type of admissible sections in more detail below.
A differentiable family of admissible sections is a family $`\sigma _s:M𝒢`$, $`s[0,1]`$, of maps such that each $`\sigma _s`$, $`s[0,1]`$, is an admissible section and the induced map $`[0,1]\times M(s,x)\sigma _s(x)𝒢`$ is differentiable.
###### Lemma 1.
Let $`𝒢`$ be a differentiable groupoid on $`M`$ and let $`\sigma _s:M𝒢`$ be a differentiable family of local admissible sections. Then there exists a section $`X`$ of $`A`$ such that
(5)
$$_sf(\sigma _sg)|_{s=0}=Xf(\sigma _0g),g𝒢.$$
###### Proof.
Since the map $`[0,1]s\sigma _sg:=\sigma _s(r(g))g`$ is differentiable for all $`g`$, there exists a vector field $`X`$ on $`𝒢`$ satisfying (5). We need to check that $`X`$ is $`d`$-vertical and right invariant. We have that
$$d(\sigma _sg)=d(g),$$
which proves that $`X`$ is $`d`$-vertical, by definition. Also,
$$\sigma _s(gh)=\sigma _s(r(gh))gh=\sigma _s(r(g))gh=(\sigma _sg)h,$$
which proves that $`X`$ is also right invariant. ∎
In the following, the section $`X\mathrm{\Gamma }(A(𝒢))`$, defined in the above lemma, will be denoted $`_s\sigma _s|_{s=0}`$. We define in the same way $`_s\sigma _s`$ for all values of $`s`$.
We shall repeatedly use the exponential map, and hence we shall to consider diffeomorphisms obtained by integrating vector fields. Recall that a smooth vector field $`X`$ on a manifold $`M`$ is called complete if, and only if, there exists a differentiable map $`\varphi :\times MM`$ such that
$$X(\varphi (t,m))=_t(\varphi (t,m)),$$
for all $`(t,m)\times M`$. We then define
$$\mathrm{exp}(X)m:=\varphi (1,m).$$
Note that a complete vector field on a manifold with corners $`M`$ is necessarily tangent to each face of $`M`$.
As for manifolds without corners, it follows from the definition and basic results on ordinary differential equations that, if $`X`$ is complete, then $`tX`$ is also complete, for all $`t`$, and
$$\mathrm{exp}((t+s)X)=\mathrm{exp}(tX)\mathrm{exp}(sX),$$
for all $`s`$ and $`t`$. Consequently, $`\mathrm{exp}(X)`$ is a diffeomorphism.
###### Lemma 2.
Let $`𝒢`$ be a differentiable groupoid on $`M`$ with Lie algebroid $`A=A(𝒢)`$. If $`X\mathrm{\Gamma }(A)`$ is a section such that $`q(X)`$ is complete, then $`X`$, regarded as a $`d`$-vertical vector field on $`𝒢`$, is also complete.
###### Proof.
For manifolds without corners, this is a result from Kumpera and Spencer \[7, Appendix\]. For manifolds with corners the proof is the same. ∎
###### Proposition 1.
Let $`\pi :AM`$ be a Lie algebroid with anchor map $`q:ATM`$, and let $`X\mathrm{\Gamma }(A)`$ be a section such that $`q(X)`$ is complete. Then there exists a uniquely determined isomorphism $`E_X:AA`$ of Lie algebroids satisfying (i) and (ii):
(i) $`\pi E_X=\mathrm{exp}(q(X))\pi `$;
(ii) $`_tE_{tX}(Y)|_{t=0}=[X,Y]`$.
(iii) If, moreover, $`𝒢`$ is a differentiable groupoid integrating $`A`$ and $`X,Y\mathrm{\Gamma }(A)`$ are both complete, then $`E_X`$ also satisfies
$$\mathrm{exp}(X)\mathrm{exp}(Y)=\mathrm{exp}(E_X(Y))\mathrm{exp}(X),$$
as admissible sections.
###### Proof.
If $`A`$ is integrable, then (i) is Proposition 4.1.(v), (ii) is Proposition Proposition 4.11.(iii), and (iii) is Proposition 4.11.(ii) of . The proof of (i) and (ii) in the general case are the same. ∎
A family $`Y_1,\mathrm{},Y_n`$ of smooth sections of $`A`$ is a local basis of $`A`$ at $`yM`$ if $`Y_1(y),\mathrm{},Y_n(y)`$ is a basis of $`A_y`$. If $`t=(t_1,\mathrm{},t_n)^n`$ and $`Y_1,\mathrm{},Y_n`$ are sections of $`A`$, then we denote
$$\mathrm{Exp}(t,Y):=\mathrm{exp}(t_1Y_1)\mathrm{exp}(t_2Y_2)\mathrm{}\mathrm{exp}(t_nY_n).$$
Also, let
$$B_ϵ=\{t=(t_1,\mathrm{},t_n)^n,\underset{i=1}{\overset{n}{}}|t_i|<ϵ\}.$$
Recall that a differentiable local groupoid $`𝒰`$ on $`M`$ satisfies all the axioms of a differential groupoid, except that the multiplication $`gg^{}`$ is not defined for all pairs $`(g,g^{})`$ such that $`d(g)=r(g^{})`$, but only in a neighborhood $`𝒰_2`$ of the diagonal in the set $`\{(g,g^{}),d(g)=r(g^{})\}`$. See for details.
###### Lemma 3.
Let $`𝒢`$ be a local differential groupoid on $`M`$, and let $`\sigma :M𝒢`$ be an admissible section with $`r(\sigma (x_0))=y_0`$. Also, let $`Y_1,\mathrm{},Y_n`$ be a local base of $`A`$ at $`y_0`$. Then, for some small $`ϵ>0`$ and a relatively compact open neighborhood $`U`$ of $`x_0`$ in $`M`$, the map
(6)
$$\psi _Y^\sigma :^n\times M(t,x)\mathrm{Exp}(t,Y)\sigma (x)𝒢$$
is a diffeomorphism from $`B_ϵ\times U`$ to a neighborhood of $`\sigma (x_0)`$ in $`𝒢`$.
If $`\sigma =\mathrm{exp}(X_m)\mathrm{}\mathrm{exp}(X_1)`$, for some integrable $`X_1,\mathrm{},X_m\mathrm{\Gamma }(A)`$, then we denote $`\psi _Y^\sigma =\psi _Y^X`$.
###### Proof.
If $`𝒢`$ is a differentiable groupoid (not just a local one), this is Proposition 4.12 in . For local groupoids the proof is the same. ∎
Let $`𝒢`$ be a differentiable groupoid, and let $`X_1,\mathrm{},X_m`$ be smooth complete sections of $`A(𝒢)`$. Also, let $`Y_i`$ be smooth, complete sections of $`A(𝒢)`$ that form a local basis at some point $`x_0M`$, as above. For simplicity, we shall sometimes assume that the $`Y_i`$ are compactly supported, which is a stronger assumption than completeness. Since the admissible section
$$\sigma (x)=\mathrm{exp}(X_m)\mathrm{}\mathrm{exp}(X_1)x$$
defines a diffeomorphism $`\sigma :𝒢𝒢`$, equation (4), we obtain that the map
(7)
$$\varphi _Y^X:^n\times M(t,x)\mathrm{exp}(X_m)\mathrm{}\mathrm{exp}(X_1)\mathrm{Exp}(t,Y)x𝒢$$
is also a diffeomorphism from a set of the form $`B_ϵ\times U`$ to its image, for some small $`ϵ>0`$ and some small open subset $`UM`$. The maps $`\varphi _Y^X`$ are slightly more convenient to work with, in what follows, than the maps $`\psi `$ of the previous lemma.
Throughout the rest of this section, $`M`$ will be a smooth manifold and $`AM`$ will be a Lie algebroid. Moreover, $`M=S`$ is an $`A`$–invariant stratification of $`M`$, $`𝒢_S`$ is a differentiable groupoid integrating $`A_S`$, and $`𝒢=𝒢_S`$ (disjoint union).
On $`𝒢=𝒢_S`$ we can then define uniquely a natural groupoid structure on $`𝒢`$ such that the structural morphisms of each $`𝒢_S`$ are obtained from those of $`𝒢`$ by restriction. In particular, if two arrows $`g𝒢_S`$ and $`g^{}𝒢_S^{}`$ are composable, then $`S=S^{}`$, also, each $`𝒢_S`$ is a subgroupoid of $`𝒢`$.
A crucial observation is that we need not use the full differentiable structure on $`𝒢`$ to define the maps $`\psi _Y^X`$ of Lemma 3, or the maps $`\varphi _Y^X`$ of equation (7), for that matter. To define $`\varphi _Y^X`$, it is enough to use only the smooth structure on each $`𝒢_S`$. This will allow to extend the definition of $`\varphi _Y^X`$ to our case, when the groupoid $`𝒢`$ is obtained by glueying differential groupoids. This can be done as follows.
Let $`M=S`$ be an $`A`$-invariant stratification of $`M`$. Also, let for each $`S`$ $`𝒢_S`$ be a differentiable groupoid integrating $`A_S=A|_S`$ and $`𝒢=𝒢_S`$, as above. If $`X_i`$ and $`Y_j`$ are sections of $`A`$ on $`M=𝒢^{(0)}`$ such that the vector fields $`q(X_i)`$ and $`q(Y_j)`$ are integrable, then the restriction of these vector fields to each strata is again integrable, and hence the restriction of $`X_i`$ and $`Y_j`$ to each strata are integrable as vertical vector fields on $`𝒢_S`$. Then the map $`\varphi _Y^X`$ is defined on each $`^n\times S`$ (with values in $`𝒢_S`$), by glueying these maps, we obtain the desired definition of $`\varphi _Y^X:^n\times M𝒢`$, for $`𝒢=𝒢_S`$.
Since the maps $`\varphi _Y^X`$ and $`\psi _Y^X`$ play an important role in what follows, we now spell out their properties in more detail.
###### Lemma 4.
(i) The maps $`\varphi _Y^X`$ and $`\psi _Y^X`$ are related by
$$\varphi _Y^X=\psi _Y^{}^X,$$
where $`Y_j^{}=E_{X_m}E_{X_{m1}}\mathrm{}E_{X_1}(Y_j)`$.
(ii) Let $`Y_j^{\prime \prime }=Y_{n+1j}`$ and $`X_i^{\prime \prime }=X_{m+1i}`$. Then
$$\varphi _Y^X(x)^1=\psi _{Y^{\prime \prime }}^{X^{\prime \prime }}(\alpha (x)),$$
where $`\alpha `$ is the diffeomorphism $`\mathrm{exp}(q(X_m))\mathrm{}\mathrm{exp}(q(X_1))`$.
###### Proof.
Because each $`𝒢_S`$ is a differential groupoid, (i) follows from Proposition 1 on each $`𝒢_S`$. From this we obtain the desired relation everywhere on $`𝒢`$.
(ii) follows from the relation
$$(\mathrm{exp}(X)x)^1=\mathrm{exp}(X)(\mathrm{exp}(q(X))x),$$
valid on each $`𝒢_S`$, and hence everywhere on $`𝒢`$. ∎
We let as above $`X_1,\mathrm{},X_m,Y_1,\mathrm{},Y_n`$ be integrable sections of $`A`$. We shall also use the following lemma.
###### Lemma 5.
(i) Fix $`t^n`$ and $`x_0M`$. Then we can find $`ϵ>0`$, a neighborhood $`U`$ of $`x`$ in $`M`$, and a differentiable map
$$\tau :B_ϵ\times U^n\times M,\tau (0,x_0)=(0,x_0),$$
which is a diffeomorphism onto its image, such that
$$\varphi _Y^X(t+s,x)=\varphi _Y^X^{}(\tau (s,x)),$$
on $`B_ϵ\times U`$, where $`X_j^{}=X_j`$, for $`j=1,\mathrm{},m`$, and $`X_{j+m}^{}=t_jY_j`$, for $`j=1,\mathrm{},n`$.
(ii) For all $`Y_1,\mathrm{}Y_n`$, $`Y_1^{},\mathrm{},Y_n^{}`$, and $`x_0M`$, there exist $`\delta >ϵ>0`$, an open neighborhood $`U`$ of $`x_0`$, and a differentiable map
$$m_1:B_ϵ\times B_ϵ\times UB_\delta \times M,m_1(0,0,x)=(0,x),$$
such that
$$\mathrm{Exp}(t,Y)x(\mathrm{Exp}(t^{},Y^{})x)^1=\mathrm{Exp}(m(t,t^{},x)).$$
###### Proof.
We begin by writing $`\mathrm{exp}((t_i+s_i)X_i)=\mathrm{exp}(t_iX_i)\mathrm{exp}(s_iX_i)`$. Then, using Proposition 1, we can find sections $`Y_i^s`$ of $`A`$, $`s`$, $`Y_i^0=Y_i`$, depending smoothly on $`s`$, such that
$$\varphi _Y^X(t+s,x)=\varphi _Y^X^{}(\mathrm{Exp}(s,Y^s)x).$$
To complete (i), we now use a local groupoid $`𝒰`$ integrating $`A`$. The existence of such a $`𝒰`$ is ensured by . By replacing $`𝒰`$ with an open neighborhood of $`M`$, if necessary, we may identify $`𝒰`$ with a subset of $`𝒢`$, using the exponential map. Then, for small $`ϵ`$ and a relatively compact neighborhood $`U`$ of $`x_0`$, the maps
$$B_ϵ\times U(x,s)f_1(x,s):=\mathrm{Exp}(s,Y^s)x$$
and
$$B_ϵ\times U(x,s)f_2(x,s):=\mathrm{Exp}(s,Y)x$$
are diffeomorphisms onto neighborhoods of $`x_0`$ in $`𝒰`$, by Lemma 3. The desired map $`\tau `$ is obtained from $`f_2^1f_1`$.
The proof of (ii) is similar, using the differentiability of the multiplication in a local groupoid. ∎
The above result suggest to introduce the following family of maps.
The family $`\mathrm{\Phi }`$: Let $`f:V_0V`$ be a diffeomorphism (i.e., coordinate chart) from an open subset of $`^l`$ to an open, relatively compact subset of $`𝒢^{(0)}=M`$. Because the partition $`M=S`$ is locally finite, we can find, using Lemma 3, an $`ϵ>0`$ and sections $`X_i,Y_j\mathrm{\Gamma }(A)`$, for which $`\varphi _Y^X`$ defines a diffeomorphism from $`B_ϵ\times (VS)𝒢_S`$, for all $`S`$ such that the intersection $`SV`$ is not empty. The family $`\mathrm{\Phi }`$ then consists of all maps of the form
(8)
$$\phi (t,y)=\varphi _Y^X(t,f(y)):B_ϵ\times V_0𝒢.$$
Recall that a differentiable atlas on a set $`M_0`$ is a family of injective maps
$$\phi :V_\phi M_0,$$
defined on an open subset of $`^l`$, for some fixed $`l`$, such that $`\phi (V_\phi )`$ is a covering of $`M_0`$, $`\phi ^1(\phi _1(V_{\phi _1})`$ is an open subset $`V`$ of $`V_\phi `$, and the map $`\phi _1^1\phi `$ is differentiable on $`V`$.
We are ready now to prove the following theorem.
###### Theorem 2.
Let $`M=S`$ be an $`A`$-invariant stratification of $`M`$ and, for each $`S`$, let $`𝒢_S`$ be a $`d`$-connected differentiable groupoid on $`M`$ integrating $`A_S=A|_S`$. Then $`𝒢`$ has a differentiable structure making it a differentiable groupoid with $`A(𝒢)A`$ if, and only if, the family $`\mathrm{\Phi }`$, consisting of the maps in equation (8), is a differentiable atlas.
###### Proof.
Suppose first that the family $`\mathrm{\Phi }`$ is a differentiable atlas. We begin by showing that the groupoid structure on $`𝒢`$, induced from the groupoids $`𝒢_S`$, is compatible with the differentiable structure defined by $`\mathrm{\Phi }`$. That is, we need to check that the structural morphisms are differentiable. We now check this.
First, the domain map is differentiable and submersive because $`d(\varphi _Y^X(t,x))=x`$, for all $`X`$, $`Y`$, and $`xM`$. Next, it is enough to check that the map $`(g^{},g)g^{}g^1`$, defined on
$$\{(g^{},g),d(g^{})=d(g)\},$$
is differentiable. Let $`(g^{},g)`$ be such that $`d(g)=d(g^{})=x_0`$, and let $`\varphi _Y^X:^n\times M𝒢`$ and $`\varphi _Y^{}^X^{}:^n\times M𝒢`$ be two maps such that $`\varphi _Y^X(t,x_0)=g`$ and $`\varphi _Y^{}^X^{}(t^{},x_0)=g^{}`$, and such that their restrictions to some small set of the form $`B_ϵ\times U`$, $`t,t^{}B_ϵ`$, is in the family $`\mathrm{\Phi }`$. We need to show that the induced map
(9)
$$\mu _1:B_ϵ\times B_ϵ\times U(t,t^{},x)\varphi _Y^{}^X^{}(t^{},x)(\varphi _Y^X(t,x))^1𝒢,$$
is differentiable. It is enough to prove this in a small neighborhood of $`x_0`$. Because $`\mathrm{\Phi }`$ forms an atlas, using Lemma 5(i), we see that we may assume $`t=t^{}=0`$, eventually by changing $`X,X^{},Y,`$ or $`Y^{}`$. The differentiability of $`\mu _1`$ then follows by combining Lemma 5(ii), and Lemma 4(i). This is enough to conclude that $`𝒢`$ is a differentiable groupoid whenever $`\mathrm{\Phi }`$ is an atlas.
Let $`A(𝒢)`$ be the Lie algebroid of $`𝒢`$ corresponding to the differentiable structure defined by $`\mathrm{\Phi }`$. We need to check that $`A(𝒢)A`$. We have that $`A(𝒢)|_SA_S`$, by construction, and hence we can identify as a set $`A(𝒢)`$ with the disjoint union of the restrictions $`A_S`$. The two differentiable structures on $`A_S`$ (the first induced from $`A`$ and the second induced from $`A(𝒢)`$) are the same because the differential of the map $`\varphi _Y^\sigma `$ of Lemma 3, $`\sigma =id`$, canonically identifies $`A|_U`$ and $`A(𝒢)|_U`$, if $`U`$ is as in that lemma. The Lie algebra structures on $`\mathrm{\Gamma }(A)`$ and $`\mathrm{\Gamma }(A(𝒢))`$ also coincide because the two possible brackets of two vector fields coincide on each strata $`S`$, and hence they coincide everywhere. We have thus proved that, if $`\mathrm{\Phi }`$ forms an atlas, then $`𝒢`$ is a differentiable groupoid with $`A(𝒢)A`$.
Conversely, suppose now that $`𝒢`$ is endowed with a differentiable structure, and let $`g𝒢_S`$. Since $`𝒢_S`$ is $`d`$-connected, we can choose vector fields $`X_1,\mathrm{},X_m`$ such that
$$g=\mathrm{exp}(X_m)\mathrm{}\mathrm{exp}(X_1)x_0,$$
for some $`x_0S`$. If $`Y_1,\mathrm{},Y_n`$ are chosen to form a basis of $`A_{x_0}`$, as in Lemma 3, then the map $`\varphi _Y^X`$ must be a diffeomorphism of a set of the form $`B_ϵ\times U`$ onto an open neighborhood of $`g`$, for some open subset $`UM`$. We obtain that if $`𝒢=𝒢_S`$ is a differential groupoid such that $`A(𝒢)A`$, then the family $`\mathrm{\Phi }`$ is an atlas. ∎
In particular, we immediately obtain from the above theorem and its proof that the differentiable structure on $`M`$ and the groupoid structure on $`𝒢`$ uniquely determine the differentiable structure on $`𝒢`$ satisfying $`A(𝒢)A`$. Indeed, the differentiable structure on $`𝒢`$ is determined by the family $`\mathrm{\Phi }`$.
We now turn to the main theorem of this paper, Theorem 3, which shows that the assumptions of Theorem 2 are satisfied provided that the groupoids $`𝒢_S`$ are $`d`$-simply connected. We shall need an extension of the concept of differentiable family of sections to our case, $`𝒢=𝒢_S`$, when $`𝒢_S`$ are smooth groupoids but $`𝒢`$ is not endowed with any differentiable structure.
A differentiable family of admissible sections is a family $`\sigma _s:M𝒢`$, $`s[0,1]`$, of maps such that each $`\sigma _s`$, $`s[0,1]`$, is an admissible section, the induced map $`[0,1]\times S(s,x)\sigma _s(x)𝒢_S`$ is differentiable for each $`S`$, and the sections $`_s(\sigma _s|_S)`$ of $`A_S`$ can be glued to a smooth section of $`A`$ on $`[0,1]\times M`$. This definition of a differentiable family of smooth section is thus very similar to the corresponding definition in the case when $`𝒢`$ has a smooth structure, except that we replace the condition on the smoothness of the map $`[0,1]\times M𝒢`$ by the existence and smoothness of the derivatives $`_s\sigma _s`$ on $`[0,1]\times M`$.
The following lemma is an important technical part of the proof of Theorem 3. It achieves a continuous and smooth deformation of a local admissible section of $`𝒢`$ to a local identity section. We denote by $`\mathrm{\Gamma }_c(E)`$ the space of compactly supported, smooth sections of a vector bundle $`E`$.
###### Lemma 6.
Let $`𝒢=𝒢_S`$ be a union of $`d`$-simply connected differentiable groupoids with $`A(𝒢_S)=A_S`$, as in the statement of Theorem 2, and let $`x_0M`$. Suppose that $`X_1,\mathrm{},X_m\mathrm{\Gamma }_c(A)`$ satisfy
$$\mathrm{exp}(X_m)\mathrm{}\mathrm{exp}(X_1)x_0=x_0.$$
Then we can find a differentiable family of admissible sections $`\sigma _s:M𝒢`$ such that:
(i) $`\sigma _1(x)=\mathrm{exp}(X_m)\mathrm{}\mathrm{exp}(X_1)x`$, on $`M`$;
(ii) $`\sigma _0(x)=x`$, for all $`xM`$;
and, most importantly,
(iii) $`\sigma _s(x_0)=x_0,`$ for all $`s[0,1]`$.
This lemma remains true if we replace the condition that $`X_i`$ be compactly supported by the condition that they be integrable.
###### Proof.
Consider the curve
$$\varphi (t)=\mathrm{exp}(t^{}X_k)\mathrm{exp}(X_{k1})\mathrm{}\mathrm{exp}(X_1)x_0,$$
where $`t^{}=(mtk+1)`$ and $`k`$ is chosen such that $`0t^{}1`$. By assumption, $`\varphi `$ is a closed curve on $`𝒢_{x_0}`$, and hence, by the assumption that $`𝒢`$ is $`d`$–simply connected, we can continuously deform this curve to the constant curve $`x_0`$, within $`𝒢_{x_0}`$, through closed curves based at $`x_0`$. More precisely, we can find
$$\eta :[0,1]\times [0,1]𝒢_{x_0}$$
such that $`\eta (t,1)=\varphi (t)`$ and $`\eta (t,0)=\eta (0,s)=\eta (1,s)=x_0`$, for all $`t,s[0,1]`$.
By an approximation argument, we can assume that $`\eta (\frac{k}{m},s)`$ depends smoothly on $`s`$, for each integer $`k`$. Moreover, after replacing $`m`$ by a large multiple $`lm`$ and each $`X_k`$, $`k=1,\mathrm{},m`$, by $`l^1X_k`$ repeated $`l`$ times, we can assume that there exist compactly supported sections $`X_k^s\mathrm{\Gamma }(A)`$, depending smoothly on $`s`$, such that $`X_k^1=X_k`$, $`X_k^0=0`$, and
$$\eta (\frac{k}{m},s)=\mathrm{exp}(X_k^s)\eta (\frac{k1}{m},s).$$
The desired deformation is obtained by letting
$$\sigma _s(x)=\mathrm{exp}(X_m^s)\mathrm{}\mathrm{exp}(X_1^s)x.$$
To see that $`\sigma _s`$ is a smooth family of admissible sections, we use Proposition 1 for each $`𝒢_S`$ and the fact that each $`\mathrm{exp}(X_k^s)`$ is a smooth admissible section. ∎
We continue to assume that $`𝒢=𝒢_S`$ is as in the statement of Theorem 3.
###### Proposition 2.
Let $`\sigma _s`$, $`s[0,1]`$, be a differentiable family of local admissible sections. Assume that $`\sigma _s(x_0)=x_0`$, for all $`s[0,1]`$, and that $`\sigma _0(x)=x`$, for all $`xM`$. Let $`Y_1,\mathrm{},Y_n`$ be a local basis of $`A`$ at $`x_0`$. Then there exist $`ϵ>\delta >0`$, a neighborhood $`U`$ of $`x_0`$ in $`M`$, and a differentiable map
$$\tau :[0,1]\times B_\delta \times UB_ϵ$$
such that: (i) $`\tau (s,0,x_0)=0`$, (ii) for each fixed $`s[0,1]`$ and $`xU`$, the map $`B_\delta t\tau (s,t,x)B_ϵ`$ is a diffeomorphism onto its image, and
$$(iii)\sigma _s\mathrm{Exp}(t,Y)x=\mathrm{Exp}(\tau (s,t,x),Y)x,$$
for all $`s,t[0,1]`$, and $`xU`$.
###### Proof.
Let $`𝒰`$ be a local groupoid integrating $`A`$. Choose $`ϵ>0`$ and a neighborhood $`U_1`$ of $`x_0`$ small enough such that
$$\eta :B_{2ϵ}\times U_1𝒰,\eta (t,x)=\mathrm{Exp}(t,Y)x,$$
is a diffeomorphism onto an open neighborhood $`V`$ of $`x_0`$ in $`𝒰`$. Let $`U_0U_1`$ be a compact neighborhood of $`x_0`$. Because $`K:=\mathrm{exp}(\overline{B}_ϵ\times U_0)`$ is a compact subset of the open set $`V`$ and $`\sigma `$ is continuous, $`\sigma _s(x_0)=x_0`$, we can find a neighborhood $`U`$ of $`x_0`$ such that $`\sigma _sk`$ is defined in $`𝒰`$ whenever $`kK`$, and $`d(k)U`$, and such that
$$\sigma _s(U_1)KV.$$
Then we can define
$$\tau (s,t,x)=\eta ^1(\sigma _s\mathrm{Exp}(t,Y)x).$$
Because $`\sigma _s\mathrm{Exp}(t,Y)xV𝒰`$ and $`\eta `$ is a diffeomorphism, we obtain that $`\tau `$ is smooth also. ∎
###### Theorem 3.
Let $`A`$ be a Lie algebroid on a manifold with corners $`M`$. Suppose that $`M`$ has an $`A`$-invariant stratification $`M=S`$ such that, for each stratum $`S`$, the restriction $`A_S`$ is integrable, then $`A`$ is integrable.
More precisely, let $`𝒢_S`$ be $`d`$-simply connected differential groupoids such that $`A(𝒢_S)A_S`$. Then the disjoint union $`𝒢=𝒢_S`$ is naturally a differentiable groupoid such that $`A(𝒢)A`$.
###### Proof.
By Theorem 2, we see that we it is enough to show that the family $`\mathrm{\Phi }`$, defined using the maps $`\varphi _Y^X`$ of equation (7), is an atlas. Let $`\varphi _Y^X`$ and $`\varphi _Y^{}^X^{}`$ be two such maps, defined on $`B_ϵ\times U`$ and, respectively, on $`B_ϵ^{}\times U^{}`$, such that their images intersect. Let $`g𝒢`$ be an element of this intersection. As in the proof of Theorem 2, we can arrange that $`\varphi _Y^X(0,x_0)=\varphi _Y^{}^X^{}(0,x_0)=g`$. Then
$$\varphi _Y^X(0,x_0)=\mathrm{exp}(X_m)\mathrm{}\mathrm{exp}(X_1)x_0=\mathrm{exp}(X_m^{}^{})\mathrm{}\mathrm{exp}(X_1^{})x_0=\varphi _Y^{}^X^{}(0,x_0),$$
and hence
$$x_0=\mathrm{exp}(X_1^{})\mathrm{}\mathrm{exp}(X_m^{}^{})\mathrm{exp}(X_m)\mathrm{}\mathrm{exp}(X_1)x_0,$$
so we may also assume that $`X_i^{}=0`$ (and hence $`g=x_0`$). By replacing $`V`$ and $`V^{}`$ with some smaller, relatively compact neighborhoods, if necessary, we may further assume that the sections $`X_i`$ are compactly supported. Let
$$\sigma (x)=\mathrm{exp}(X_m)\mathrm{}\mathrm{exp}(X_1)x.$$
Since each $`𝒢_S`$ is $`d`$-simply connected, by Lemma 6 we can find a smooth family $`\sigma _s:M𝒢`$, $`s[0,1]`$, such that $`\sigma _0`$ is the identity, $`\sigma _1=\sigma `$, the restriction to each groupoid $`𝒢_S`$ is differentiable. The smoothness of the family $`\sigma _s`$ in this case is reduces to showing that $`_s\sigma _s`$ is a smooth section of $`A`$ over $`M`$. Let $`\tau `$ be the map defined in Proposition 2 using the admissible sections $`\sigma _s`$, and let $`\tau _1(t,x)=\tau (1,t,x)`$. Then
$$\varphi _Y^X(t,x)=\varphi _Y^{}^X^{}(\tau _1(t,x),x),$$
in a small neighborhood of $`x_0`$. Since $`\tau _1`$ is a local diffeomorphism for each fixed $`x`$, this proves the theorem. ∎
## 3. Applications to foliations
Recall that a Lie algebroid $`\pi :AM`$, with anchor map $`q:ATM`$, is called regular if, and only if, the range of $`q`$ has locally constant rank. Then the sections of $`q(A)`$ are the sections of the tangent bundle to a foliation $`_A`$ whose tangent bundle, a sub-bundle of $`TM`$, is denoted $`T_A`$. If, moreover, there exists a morphism $`\rho :\mathrm{\Gamma }(T_A)\mathrm{\Gamma }(A)`$ such that $`q\rho =id`$, then we may assume that $`A`$ is a semi-direct product.
Also, if $`A`$ is regular, the kernel of $`q`$ is a bundle of Lie algebras. See also . Recall that a result of Douady and Lazard from states that every bundle of Lie algebras is integrable. In particular the kernel $`\mathrm{ker}(q)`$ is integrable. (I am greatful to Alan Weinstein for pointing out this reference to me.)
We say that $`A`$ is (isomorphic to) the semi-direct product $`\mathrm{ker}(q)q(A)`$ if there exists a morphism of Lie algebras $`\rho :\mathrm{\Gamma }(q(A))\mathrm{\Gamma }(A)`$ that is a right inverse to $`q`$.
###### Proposition 3.
Let $`A`$ be a regular algebroid with anchor map $`q:ATM`$. Assume that $`A`$ is the semi-direct product $`\mathrm{ker}(q)q(A)`$. Then $`A`$ is integrable.
Before proceeding to the proof, we first introduce some terminology and make some comments. If $`EM`$ is a vector bundle on a foliated manifold $`M`$, then a leafwise connection on $`E`$ is a linear map
$$:\mathrm{\Gamma }(E)\mathrm{\Gamma }(ET^{}),$$
satisfying the Leibnitz identity. Here $``$ is the foliation of $`M`$, $`T`$ is the tangent bundle to this foliation, and $`T^{}`$ is the dual of this bundle. The equivalent definitions of a connection in terms of parallel transport or equivariant splittings of tangent spaces of principal bundles extend to the “leafwise” setting also.
The structure of regular, semi-direct product Lie algebroids is as follows. First, there exists a leafwise flat connection $``$, $`_Z(X)=[\rho (Z),X],`$ on $`\mathrm{ker}(q)`$, which preserves its Lie bundle structure, that is
$$_Z_Z^{}_Z^{}_Z=_{[Z,Z^{}]},\text{ and }$$
$$_Z([X,Y])=[_Z(X),Y]+[X,_Z(Y)],$$
for all $`Z,Z^{}\mathrm{\Gamma }(T_A)`$ and $`X,Y\mathrm{\Gamma }(\mathrm{ker}(q))`$. Then
$$AT_A\mathrm{ker}(q)$$
as vector bundles, with anchor map given by the projection onto the first component, and with Lie bracket on $`\mathrm{\Gamma }(A)`$ defined by
$$[(Z,X),(Z^{},X^{})]=([Z,Z^{}],_Z(X^{})_Z^{}(X)+[X,X^{}]),$$
for all $`Z,Z^{}\mathrm{\Gamma }(T_A)`$ and $`X,X^{}\mathrm{\Gamma }(\mathrm{ker}(q))`$.
With these comments, we are now ready to prove Proposition 3.
###### Proof.
Recall that $`𝒫`$, the path groupoid of the foliation $``$, consists of fixed end point homotopy classes of paths $`\gamma `$ that are fully contained in a single leaf, with respect to homotopies within that leaf. As explained above, the morphism $`\rho `$ defines a leafwise flat connection on $`\mathrm{ker}(q)`$ that preserves the Lie bracket. Let $`𝒦`$ be a $`d`$-simply connected groupoid that integrates $`\mathrm{ker}(q)`$.
We define then a groupoid $`𝒢`$ that integrates $`A`$ as follows. As a smooth manifold,
$$𝒢=\{(g,\gamma )𝒦\times 𝒫,d(g)=\gamma (1)\}.$$
To define the multiplication, observe first that the leafwise flat connection on $`\mathrm{ker}(q)`$ defines a parallel transport map
$$\rho (\gamma ):\mathrm{ker}(q)_{\gamma (0)}\mathrm{ker}(q)_{\gamma (1)},$$
which is a Lie algebra isomorphism for any path $`\gamma `$ fully contained in a leaf. Since $`𝒦_x`$ is simply connected for each $`x`$, we obtain by exponentiation a group morphism
$$\rho (\gamma ):𝒦_{\gamma (0)}𝒦_{\gamma (1)}.$$
That is, the leafwise flat connection on $`\mathrm{ker}(q)`$ lifts to a leafwise flat connection on $`𝒦`$ that preserves the Lie group structure on the fibers.
We are ready now to define the groupoid structure on $`𝒢`$. Note first that $`d(g)=r(g)`$ for $`g𝒦`$. Then $`d(g,\gamma )=\gamma (0)`$, $`r(g,\gamma )=r(g)`$, and the product on $`𝒢`$ is given by the formula
$$(g,\gamma )(g^{},\gamma ^{})=(g\rho (\gamma )(g^{}),\gamma \gamma ^{}),$$
where the composition of paths is given by concatenation. The flatness of $``$ gives that $`\rho (\gamma \gamma ^{})=\rho (\gamma )\rho (\gamma ^{})`$, which guarantees the associativity of the product. ∎
In certain cases we get integrability without assuming that $`A`$ is a semi-direct product.
###### Proposition 4.
Let $`A`$ be a regular Lie algebroid with anchor map $`q:ATM`$ such that $`\mathrm{ker}(q)`$ is a bundle of semisimple Lie algebras. Then $`A`$ is integrable.
###### Proof.
We may assume that $`M`$ is connected. Since semisimple Lie algebras are rigid, all fibers of $`\mathrm{ker}(q)`$ will be isomorphic Lie algebras. Fix one of these algebras and denote it by $`𝔤`$. Also, let $`G_0`$ be the group of automorphisms of $`𝔤`$. Then, if we define
$$P=_x\mathrm{Iso}(𝔤,\mathrm{ker}(q)_x),$$
(the fibers are the sets of Lie algebra isomorphisms $`𝔤\mathrm{ker}(q)_x`$), we obtain a $`G_0`$-principal bundle on $`M`$, which acquires by pull-back a foliation $``$ of the same codimension as $`_A`$. The path groupoid $`𝒫`$ of the foliation $``$ has an induced free action of $`G_0`$, and we define the groupoid $`𝒢`$ by $`𝒢=𝒫/G_0`$. The composition of two paths in $`𝒢`$ is obtained by choosing composable liftings in $`𝒫`$. Because the Lie algebra of $`G_0`$ is $`𝔤`$, we obtain that $`𝒢`$ integrates $`A`$. ∎
With the above results in mind, we now define $`A`$-invariant regular stratifications.
###### Definition 2.
Let $`\pi :AM`$ be a Lie algebroid with anchor map $`q:ATM`$ on the manifold with corners $`M`$. An invariant stratification $`M=S`$ is called regular if, and only if, the restriction $`A_S:=A|_S`$ is regular for each $`S`$.
Although almost all interesting Lie algebroids have invariant regular stratifications, this is not true in general. Consider, for example, a closed subset $`B^n`$ with empty interior, which is not a manifold. Let $`\varphi 0`$ be a smooth function that vanishes exactly on $`B`$, and let $`𝒱(^n)`$ be the Lie algebra of vector fields on $`^n`$. Since $`\varphi 𝒱(^n)`$ is a free $`𝒞^{\mathrm{}}(^n)`$ module, using the Serre-Swan theorem, we can define a vector bundle $`A`$ such that $`\mathrm{\Gamma }(A)=\varphi 𝒱(^n)`$. Moreover,
$$[\varphi X,\varphi Y]=\varphi \left(X(\varphi )YY(\varphi )X+\varphi [X,Y]\right),$$
for all vector fields $`X`$ and $`Y`$ on $`^n`$, which shows that $`A`$ is a Lie algebroid. It is not difficult to see that $`A`$ has no invariant regular stratification.
Although not all Lie algebroids have regular stratifications, this notion is useful because it is easier to integrate regular algebroids than general Lie algebroids, and we know that in order to integrate a Lie algebroid, it is enough to integrate it over each strata (Theorem 3).
###### Theorem 4.
Let $`A`$ be a Lie algebroid with anchor map $`q:ATM`$ on a manifold $`M`$ with a regular $`A`$-invariant stratification $`M=S`$. Assume, for each $`S`$, that either $`A_S`$ is the semi-direct product $`\mathrm{ker}(q)_Sq(A_S)`$, or that $`\mathrm{ker}(q)|_S`$ is a bundle of semisimple Lie algebras. Then $`A`$ is integrable.
###### Proof.
Using Proposition 3 or Proposition 4, we see that each of the algebroids $`A_S`$, obtained by restricting $`A`$ to the stratum $`S`$, is integrable. By Theorem 3, the Lie algebroid $`A`$ is then integrable. ∎
Let us consider now in greater detail a class of examples that is useful for the construction of pseudodifferential operators on complex algebraic varieties. It is possible to associate several “natural” algebroids to a complex algebraic variety, so the following constructions lead to a family of algebras of pseudodifferential operators associated to a complex algebraic variety. The details of this construction will be presented in a future paper. We do not know at this point which of the many algebras that one obtains is the “right” algebra to associate to a complex algebraic manifold, but we hope to address this question in a future paper.
Let $`M`$ be a manifold with corners, such that each hyperface (i.e., face of maximal dimension) $`HM`$ is given by $`H=\{x_H=0\}`$ for some function $`x_H0`$ with $`dx_H0`$ on $`H`$, (i.e., $`x_H`$ is a defining function of $`H`$.) If $`FM`$ is an arbitrary face of $`M`$ of codimension $`k`$, then $`F`$ is an open component of the intersection of the hyperfaces containing it. The set $`x_1,\mathrm{},x_k`$ of defining functions of these hypersurfaces are called the defining functions of $`F`$; thus $`F`$ is a connected component of $`\{x_1=x_2=\mathrm{}=x_k=0\}`$.
We first introduce the class of Lie algebroids we are interested in on a manifold with corners $`M`$. We call these algebroids “quasi-homogeneous,” and we now proceed to construct them. First we need some related definitions that will make it easier to describe our settings.
###### Definition 3.
A Lie flag on a manifold $`M`$ is an increasing finite sequence of sub-bundles
$$E_0E_1\mathrm{}E_lE_{\mathrm{}}:=TM$$
such that $`[\mathrm{\Gamma }(E_i),\mathrm{\Gamma }(E_j)]\mathrm{\Gamma }(E_{i+j})`$.
It follows from the definition that, if $`E_0\mathrm{}E_lTM`$ is a Lie flag on $`M`$, then the bundles $`E_0`$ and $`E_l`$ are integrable. We do not assume the above inclusions to be strict.
We now describe the type of behavior we want for the vector fields that are sections of a quasi-homogeneous Lie algebroid. Let $`p_i`$ denote the projection onto the $`i`$th component of a product.
###### Definition 4.
Let $`H=\{x_H=0\}M`$ be a hyperface, and let
$$\varphi _H=(\pi _H,x_H):V_HH\times [0,ϵ)$$
be a diffeomorphism defined in a neighborhood $`V_H`$ of $`H`$ in $`M`$. Also, let
$$E_0^H\mathrm{}E_{l_H}^HTH$$
be a Lie flag on $`H`$ and $`d_H\{\mathrm{}\}`$. Then a vector field $`X`$ on $`V_H`$ is called $`(E_i,\varphi _H,d_H)`$adapted if, and only if,
(10)
$$X\underset{j=0}{\overset{l_H}{}}x_H^j\mathrm{\Gamma }(\pi _H^{}(E_j^H)|_{V_H})+𝒞^{\mathrm{}}(V_H)x_H^{d_H+1}_{x_H},$$
(we agree that $`x_H^{\mathrm{}}=0`$).
Finally, the sections of our quasi-homogeneous Lie algebroids will consist of vector fields that are “adapted” to each hyperface, but the data at each hyperface must be compatible.
###### Definition 5.
A boundary Lie datum $`𝒟=(E_i^H,\varphi _H,d_H)`$ on a manifold with corners $`M`$, where $`H`$ ranges through the set of hyperfaces of $`M`$, consists of:
(i) Lie flags $`E_0^H\mathrm{}E_{l_H}^HTH`$ such that all intersections $`E_{i_1}^{H_1}\mathrm{}E_{i_t}^{H_t}`$, $`i_j\{0,1,\mathrm{},l_{H_j},\mathrm{}\}`$, as well as all finite sums of such intersections, have constant rank on the set where they are defined (and hence they are vector bundles), and form a distributive lattice.
(ii) Diffeomorphisms
$$\varphi _H=(\pi _H,x_H):V_HH\times [0,ϵ)$$
such that $`\pi _H\pi _H^{}=\pi _H^{}\pi _H`$ and $`x_H\pi _H^{}=x_H`$ on $`V_HV_H^{}`$, for all hyperfaces $`H`$ and $`H^{}`$.
(iii) Degrees $`d_H\{\mathrm{}\}`$.
If $`𝒟=(E_i^H,\varphi _H,d_H)`$ is a boundary Lie datum on $`M`$, then the diffeomorphism $`\varphi _H`$ defines a complement $`NH`$ to $`TH`$ in $`TM|_H`$.
###### Proposition 5.
Let $`𝒟=(E_i^H,\varphi _H,d_H)`$ be a boundary Lie datum on the manifold with corners $`M`$. Suppose that $``$ is a foliation of $`M`$ such that
$$T|_H=\{\begin{array}{cc}E_{l_H}^H+NH,\hfill & \text{ if }d_H<\mathrm{},\hfill \\ E_{l_H}^H,\hfill & \text{ if }d_H=\mathrm{}.\hfill \end{array}$$
If we define
$$𝒜_𝒟:=\{X\mathrm{\Gamma }(T),X\text{ is }(E_i^H,\varphi _H,d_H)\text{–adapted, for each hyperface }H\}$$
and $`[x_j_{x_j},𝒜_𝒟]𝒜_𝒟`$, then $`𝒜_𝒟`$ is a Lie algebra and a projective $`𝒞^{\mathrm{}}(M)`$–module.
###### Proof.
The set of points satisfying (10) at a face $`H`$ is closed under the Lie bracket, by definition of a Lie flag. Since $`\mathrm{\Gamma }(T)`$ is also closed under the Lie bracket, it follows that $`𝒜_𝒟`$ is a Lie algebra.
Fix a point $`x_0`$ in the interior of a face $`F`$ of codimension $`k`$ with defining functions $`x_1,\mathrm{},x_k`$. Let $`H_1,\mathrm{},H_k`$ be the corresponding hyperfaces, $`H_j=\{x_j=0\}`$, and let $`d_i=d_{H_i}`$. Also let
$$\pi =\pi _{H_1}\mathrm{}\pi _{H_k}:V:=V_{H_j}F,$$
where the order of composition is not important because $`\pi _H\pi _H^{}=\pi _H^{}\pi _H`$, by the definition of boundary Lie datum.
Let $`\nu =(\nu _1,\mathrm{},\nu _k)`$, $`0\nu _il_{H_i}`$, be a multi-index and
$$E_\nu =\pi ^{}(E_{\nu _1}^{H_1}\mathrm{}E_{\nu _k}^{H_k}),$$
which, we recall, is a vector bundle on $`H_1\mathrm{}H_k`$, again by the definition of a boundary Lie datum. Also, let
$$\nu ^{(i)}=(\nu _1,\mathrm{},\nu _{i1},\nu _i1,\nu _{i+1},\mathrm{},\nu _k)$$
and choose, arbitrarily, a complement $`Y_\nu `$ to $`E_{\nu ^{(i)}}`$ in $`E_\nu `$.
Let $`𝒵`$ be the set of vector fields $`\{x_i^{d_i+1}_{x_i},d_i<\mathrm{}\}`$ defined using the diffeomorphism $`(\pi ,x_1,\mathrm{},x_k):V=V_{H_i}F\times [0,ϵ)^k,`$ for some small $`ϵ`$. Then the restriction of $`𝒜_𝒟`$ to $`V`$ is
(11)
$$\begin{array}{c}𝒞^{\mathrm{}}(V)𝒜_𝒟|_V=\left(\underset{\nu =(\nu _1,\mathrm{},\nu _k)}{}x_1^{\nu _1}\mathrm{}x_k^{\nu _k}\mathrm{\Gamma }(\pi ^{}(Y_\nu ))\right)\left(\underset{Z_i𝒵}{}𝒞^{\mathrm{}}(V)Z_i\right)\hfill \\ \hfill \mathrm{\Gamma }(_\nu Y_\nu ^𝒵).\end{array}$$
Since this is a $`𝒞^{\mathrm{}}(V)`$-projective module–the module of sections of a bundle isomorphic to the direct sum of $`_\nu Y_\nu `$ and the trivial bundle generated by the set $`x_i^{d_i+1}_{x_i}`$ ($`d_i<\mathrm{}`$), we obtain that $`𝒜_𝒟`$ is a projective $`𝒞^{\mathrm{}}(M)`$–module, as desired. ∎
Let us examine now a particular case of this construction. Assume that in the definition above $`=M`$, that is, that there exists a single leaf, and that in the boundary Lie data $`d_H=0`$ and $`E_1^H=TH`$. The only choice then is that of the integrable bundles $`E_0^H`$, because the choice of the diffeomorphisms $`\varphi _H`$ is not important. The conditions that these bundles have to satisfy are that, for each face $`F`$ (contained in $`k`$ distinct hyperplanes $`H`$, where $`k`$ is the codimension of $`F`$), there exist $`2k`$ commuting surjective projections
$$p_{FH}:TM|_FE_0^H|_F,\text{ and }q_{FH}:TM|_FTH|_F.$$
We denote by $`A_𝒟`$ the vector bundle on $`M`$ with sections $`𝒜_𝒟`$, defined by the above proposition. Because $`𝒜_𝒟`$ is a Lie algebra, $`A_𝒟`$ is a Lie algebroid. A Lie algebroid $`A`$ is called quasi-homogeneous if it is isomorphic to one of the form $`A_𝒟`$ obtained as above.
###### Proposition 6.
Let $`ATM`$ be a quasi-homogeneous algebroid. Then the set of open faces of $`M`$ defines an $`A`$-invariant regular stratification of $`M`$. Moreover, if $`S=\mathrm{Int}(F)`$ is an open face of $`M`$, then the Lie algebroid $`q_S:A_S:=A|_STS`$ is integrable.
###### Proof.
Let $``$ and $`(E_i^H,\varphi _H,d_H)`$ be the foliation and, respectively, the boundary Lie data defining $`A`$. Observe that the sections of $`A`$ are vector fields that are tangent to all faces of $`M`$, and hence each open face of $`M`$ is $`A`$-invariant. This means that the stratification of $`M`$ by open faces is an $`A`$-invariant stratification of $`M`$.
Fix a face $`F`$ of codimension $`k`$ with defining functions $`x_1,\mathrm{},x_k`$, such that $`H_j=\{x_j=0\}`$. Let $`S:=\mathrm{Int}(F)`$ be the interior of $`F`$, and denote $`A_S:=A|_S`$.
For $`\nu _0=(0,\mathrm{},0)`$, the intersection
$$E_{\nu _0}:=E_0^{H_1}\mathrm{}E_0^{H_k}TS$$
is an integrable sub-bundle of $`TS`$ such that $`q_S:A_STS`$ maps $`\mathrm{\Gamma }(A_S)`$ surjectively onto $`\mathrm{\Gamma }(E_{\nu _0})`$. Moreover, the map $`\pi =\pi _{H_1}\mathrm{}\pi _{H_k}`$ defines a splitting of $`A_S`$, that is, a Lie algebra morphism $`\mathrm{\Gamma }(E_{\nu _0})\mathrm{\Gamma }(A_S)`$. (This splitting is implicit in the description of $`𝒜_𝒟`$ given in (11).) Consequently, $`A_S`$ is integrable, by Proposition 3. ∎
###### Theorem 5.
Every quasi-homogeneous algebroid is integrable.
###### Proof.
This follows from Proposition 6 and Theorem 4. ∎
Because the construction of the differentiable groupoid integrating a regular, semi-direct product Lie algebroid depends on the construction of a differentiable groupoid $`𝒦`$ integrating $`\mathrm{ker}(q_S)`$, it is useful to have an explicit construction of $`𝒦`$ in the above proposition. To this end, we use the notation in the proof of Proposition 5. Let $`Z_0Z:=\{x_i^{d_i+1}_{x_i}\}`$ be the set of normal vectors (with respect to the decomposition induced by $`(\pi ,x_1,\mathrm{},x_k)`$) such that $`d_i=d_{H_i}=0`$. Also, let $`A_0`$ be the sub-bundle of $`A|_S`$ generated by $`Z_0`$ and $`A_1`$ be the vector bundle generated by the complement of $`Z_0`$ in $`Z`$ and by $`E_\nu `$, for $`\nu (0,\mathrm{},0)`$. We then obtain the split exact sequence
$$0A_1\mathrm{ker}(q_S)A_00.$$
Now, each of the bundles $`A_0`$ and $`A_1`$ is a Lie algebroid with vanishing anchor map. In fact, each of $`A_0`$ and $`A_1`$ are bundles of commutative Lie algebras, and hence are integrable: to integrate them, we just consider each $`A_0`$ and $`A_1`$ as a bundle of commutative Lie groups. Moreover, $`A_1`$ is an ideal (in Lie algebra sense) of $`\mathrm{ker}(q_S)`$, and the sub-bundle $`A_1`$ acts by derivations on $`A_0`$ (the weights are exactly given by the exponents $`\nu =(\nu _i)`$). This action by derivations of each of the fibers $`(A_1)_x`$ on $`(A_0)_x`$ exponentiates to an action of the Lie group with Lie algebra $`(A_1)_x`$ on the Lie group with Lie algebra $`(A_0)_x`$. Denote the resulting semidirect product by $`𝒦_x`$. In this way we obtain on $`𝒦=𝒦_x`$ a Lie algebra bundle structure, which is isomorphic to $`\mathrm{ker}(q_S)`$, as a fiber bundle, via the exponential map.
If $`A`$ is an integrable Lie algebroid, then we denote by $`𝒢_A`$ a $`d`$-simply-connected differential groupoid that integrates $`A`$, which is unique up to isomorphism. If $`A`$ is a quasi-homogeneous algebroid and $`𝒢`$ is any differential groupoid on $`M`$ with Lie algebroid $`A`$, then we call $`𝒢`$ a quasi-homogeneous groupoid.
We now make some elementary remarks on quasi-homogeneous algebroids and groupoids. If $`AM`$ is a quasi-homogeneous algebroid and $`FM`$ is a face of $`M`$, then $`A_F`$ is not a quasi-homogeneous algebroid unless $`F=M`$. Let $`N_FTM`$ be the normal bundle to $`F`$, with the trivialization given by the boundary Lie data used in the definition of $`A`$. Then $`A_F`$ is the semidirect product of $`A_FTF`$ and $`N`$. The same will be true of the integrating groupoids. Thus, assume that $`F`$ has codimension $`k`$. Then there exists an action of $`^k`$ on $`𝒢_{A_FTF}`$, which fixes $`F`$, the set of units of $`𝒢_{A_FTF}`$, such that
(12)
$$𝒢_{A_F}𝒢_{A_FTF}^k.$$
This relates the differential groupoids $`𝒢_{A_F}`$, associated to the faces of $`M`$, to the quasi-homogeneous differential groupoids $`𝒢_{A_FTF}`$.
Let us now take a closer look at the simplest example of a quasi-homogeneous, non-regular algebroid, the algebroid $`𝒱_b(M)`$ of all vector fields tangent to the boundary $`M\mathrm{}`$ of a manifold with boundary $`M`$. The two strata of $`M`$ are $`M`$ and $`\mathrm{Int}(M)=MM`$. If $`X`$ is a topological space, we denote by $`𝒫_X`$ its path groupoid. Then the $`d`$-simply-connected differential groupoids $`𝒢_1`$ and $`𝒢_2`$ that integrate $`𝒱_b(M)|_M=T(M)N_M`$ and $`𝒱_b(M)|_{\mathrm{Int}(M)}`$ are, up to isomorphism,
$$𝒢_1𝒫_M\times ,\text{and}𝒢_2𝒫_{\mathrm{Int}(M)}.$$
By Theorem 3, or directly, the two groupoids $`𝒢_1`$ and $`𝒢_2`$ above can be smoothly glued to form a differentiable groupoid
$$𝒢_{𝒱_b(M)}=𝒢_1𝒢_2$$
that integrates $`𝒱_b(M)`$. This groupoid will be Hausdorff if, and only if, the morphism $`\pi _1(M)\pi _1(M)`$ is injective. The domain map
$$d:𝒢_{𝒱_b(M)}M$$
is a fibration (that is, it has the homotopy lifting property) if, and only if, $`\pi _1(M)\pi _1(M)`$ is surjective.
In general, the groupoid $`𝒢_{𝒱_b(M)}`$ is much larger than the “stretched product” $`M_b^2`$ considered by Melrose , which also gives a groupoid that integrates $`𝒱_b(M)`$ after we remove its off-diagonal faces. We get the same groupoid only if both $`\pi _1(M)`$ and $`\pi _1(M)`$ are trivial. Unlike our $`d`$-simply connected groupoid $`𝒢_{𝒱_b(M)}`$, the groupoid obtained from $`M_b^2`$ is always Hausdorff and the domain map $`d`$ is a fibration.
Let $`r2`$ be an integer. Then the same discussion applies to $`𝒱_{b,k}(M)`$, the Lie algebra of vector fields that at the boundary are of the form $`x^r_x+_j_{y_j}`$, for a suitable coordinate systems $`(x,y_1,\mathrm{},y_{n1})`$ in a neighborhood of a point of the boundary $`M=\{x=0\}`$ and a suitable choice of a complement to $`T(M)`$ in $`TM|_M`$.
As a final remark, we now use the algebroid $`𝒱_b(M)`$ to show that some conditions on the groupoids $`𝒢_S`$ are necessary in Theorem 3. Otherwise the glued groupoid might not be a smooth manifold. Indeed, consider the same groupoid $`𝒢_2𝒫_{\mathrm{Int}(M)}`$ for the big-open strata $`\mathrm{Int}(M)`$, but a smaller one for the boundary:
$$𝒢_1(M\times M)\times .$$
Then $`𝒢=𝒢_1𝒢_2`$ is not a smooth manifold if $`\pi _1(M)`$ is non-trivial.
|
warning/0004/cond-mat0004150.html
|
ar5iv
|
text
|
# Critical behaviour of the 1𝑑 annihilation fission process 2𝐴→"Ø", 2𝐴→3𝐴
\[
## Abstract
Numerical simulations and cluster mean-field approximations with coherent anomaly extrapolation show that the critical line of the $`1d`$ annihilation fission process is separated into two regions. In both the small and high diffusion cases the critical behavior is different from the well known universality classes of non-equilibrium phase transitions to absorbing states. The high diffusion region seems to be well described by the cyclically coupled directed percolation and annihilating random walk. Spreading exponents show non-universal behavior.
\]
Non-equilibrium phase transition may take place even in one dimensional systems. However the ordered phase lacks fluctuations that could destroy the state. If the system has evolved into that state it will be trapped there. We call this transition to an absorbing state. First order transition among these is very rare, there have been only a few system found that exhibit a discontinuous jump from the active to the absorbing phase .
Continuous phase transitions have been found to belong to a few universality classes, the most robust of them is the directed percolation (DP) class. According to the hypothesis of all continuous phase transitions to a single absorbing state in homogeneous systems with short ranged interactions belong to this class provided there is no additional symmetry and quenched randomness present. The most prominent system exhibiting phase transition of this class is the branching and annihilating random walk with one offspring (BARW). Furthermore systems with infinitely many absorbing states were also found to belong to this class .
An important exception from the DP class is the so-called parity-conserving (PC) class where particles follow a branching and annihilating random walk with two offsprings (BARW2) $`A3A`$, $`2A\text{Ø}`$ that conserves the parity of their number . Particles following BARW2 may also appear as kinks between ordered domains in systems exhibiting two absorbing states . However it was realized that the BARW2 dynamics alone is not sufficient condition for a transition in this class but an exact $`Z_2`$ symmetry between the two absorbing states is necessary, too .
Recently a study on the annihilation fission (AF) process $`2A\text{Ø}`$, $`2A3A`$ suggested that neither the BARW2 dynamics nor the $`Z_2`$ symmetry but simply the occurrence of two absorbing states can result in a PC class transition. This model was introduced first by in the context of a renormalization group analysis of the corresponding bosonic field theory. The main idea was to interpolate between systems with real and imaginary noise. These studies predicted a non-DP class transition, but they could not tell to which universality class this transition really belongs.
A fermionic version of the AF process introduced in is controlled by two parameters, namely the probability of pair annihilation $`p`$ and the probability of particle diffusion $`d`$. The dynamical rules are
$$\begin{array}{cc}\hfill AA\text{Ø},\text{Ø}AAAAA\text{with rate}& (1p)(1d)/2\hfill \\ \hfill AA\text{ØØ}\text{with rate}& p(1d)\hfill \\ \hfill A\text{Ø}\text{Ø}A\text{with rate}& d.\hfill \end{array}$$
(1)
Carlon et al. suggested that the phase diagram can be separated into two regions. For low values of $`d`$ (less than approximately 0.3), they found a continuous phase transition belonging to the PC universality class. For large values of $`d`$, however, a first order transition was reported. This claim is based on mean-field, pair mean-field and density matrix renormalization group method (DMRG) calculations. An even more recent preliminary numerical study found non-PC class like critical exponents and posed the question if there is a new type of non-equilibrium phase transition occurring here.
In this paper I report mored detailed numerical results: Simulations and generalized mean-field (GMF) approximations with coherent anomaly (CAM) extrapolations that give numerical evidence of a rich phase diagram with two different kinds of new universality classes in the low $`d`$ and high $`d`$ regions.
For simulations a parallel update version of the AF process was used since the model can effectively be mapped on a massively parallel processor ring . To avoid collisions by simultaneous updates the annihilation and exchange steps are done on two sub-lattices, while creation is performed on three sub-lattices (with a rate double as in eq.1). The critical point has been determined for $`d=0.05,0.1,0.2,0.5,0.9`$ by following the decay of the particles from a uniform, random initial state ($`\rho (t)t^\delta `$). The local slopes curve of the density
$$\delta _{eff}(t)=\frac{\mathrm{ln}\left[\rho (t)/\rho (t/m)\right]}{\mathrm{ln}(m)}$$
(2)
(where we use $`m=8`$ usually) at the critical point goes to exponent $`\delta `$ by a straight line, while in sub(super)-critical cases it veers down(up) respectively. As Fig. 1 shows at $`d=0.2`$ and $`p=0.24802`$ this quantity scales without any relevant corrections (i.e. the straight line is horizontal) and in the $`1/t0`$ limit goes to $`\delta =0.268(2)`$.
Very similar results has been obtained for $`d=0.05`$ and $`d=0.1`$ showing that the exponent $`\delta `$ is close to the corresponding value of the PC class ($`0.285(10)`$ ), but appears to be significantly smaller. This $`\delta `$ is in agreement within error margin with the value ($`0.272(18)`$) one can obtain from DMGR calculations using scaling laws $`\delta =\beta /\nu _{||}=\beta /\nu _{}/z`$.
For stronger diffusion rates $`d=0.5`$ and $`d=0.9`$ one can observe slower density decays with exponent $`\delta =0.215(10)`$ (Fig. 2). Note that I checked, that changing the abscissa on the local slopes graphs from $`1/t`$ to $`\mathrm{ln}(t)/t`$ does not eliminate the differences between small and large $`d`$ behaviors. Also by assuming strong correction to scaling I could not get coherent estimates for $`\beta `$ and $`\delta `$ that would be in agreement with DP or PC classes. This value of $`\delta `$ is in agreement with that of the coupled DP+annihilation random walk suggested by . This model was proposed to explain the space-time behavior of the AF with the help of a multi-component model, where $`A`$ particles (corresponding to pairs in AF) perform BARW process: $`A2A`$, $`A\text{Ø}`$ and occasionally create $`B`$ particles (corresponding to lonely particles in AF) who follow random walk + annihilation : $`2BA`$ (ARW). The space-time evolution picture of this model looks similar to that of AF (and very different from that of BARW2) but the numerical simulations predicted somewhat different exponents as those of the AF . By approaching $`d=1`$ corrections to scaling are getting stronger and this might leaded to the conclusion of that for $`d>0.3`$ the transition is of first order. The present study can not confirm those predictions based on extrapolations to DMRG results of small system sizes ($`L<100`$). The first order transition for $`d>0.3`$ appears to be unlikely also because the mean-field approximation of the model exhibits continuous phase transition (with $`\beta _{MF}=1`$) and that corresponds to the strong diffusion limit without fluctuations.
The differences from PC class and from first order transition can be seen even more clearly by direct measurements of the order parameter $`\beta `$ exponent ($`\rho (\mathrm{})ϵ^\beta `$). By looking at the effective exponent defined as
$$\beta _{eff}(ϵ_i)=\frac{\mathrm{ln}\rho (ϵ_i)\mathrm{ln}\rho (ϵ_{i1})}{\mathrm{ln}ϵ_i\mathrm{ln}ϵ_{i1}},$$
(3)
on Fig. 3 one can observe two regions again: for $`d<0.3`$ the $`\beta _{eff}`$ tends to $`0.58(2)`$ as $`ϵ=(p_cp)0`$, while for $`d>0.3`$ it goes to $`0.4(2)`$.
For small diffusion rates the value $`0.58`$ is very different from that of the PC class ($`0.94(10)`$) ) and from that of DP class ($`0.2765(1))`$, but is in agreement with preliminary simulation results of and . For strong diffusion it coincides fairly well with the value $`0.38(6)`$ that was found for the coupled DP+ARW model .
The results for the order parameter exponent has been verified by generalized mean-field (GMF) approximations with coherent anomaly extrapolation (CAM) . This method has been proven to give precise estimates for the DP and PC classes. The details of the method are described in and a forthcoming paper will discuss the results and show estimates for other exponents as well.
For technical reasons a variant of the AF model was investigated where the creation is symmetric: $`A\text{Ø}AAAA`$. This does not change the essence since the $`AA\text{Ø}/\text{Ø}AAAAA`$ process can be decomposed into a $`A\text{Ø}\text{Ø}A`$ exchange and a symmetric creation. One can easily check that the mean field approximation and result of it is the same
$$\rho (\mathrm{})=(13p)/(1p)$$
(4)
Higher order approximations for $`N=2,3,\mathrm{..7}`$ converge to the simulation results of this variant (Fig. 4) and the phase diagram qualitatively agrees with that of ref. . The leading order singularity at critical point remains mean-field type i.e. $`\beta _{GMF}=1`$. Furthermore dynamical MC simulation of this variant gives the same critical decay behavior as the original AF model. According to CAM the amplitudes of these singularities $`a(N)`$ scale in such a way that
$$a(N)(p_c(N)p_c)^{\beta \beta _{MF}}$$
(5)
the exponent of true singular behavior can be estimated.
The CAM extrapolation has been applied for the highest levels of approximations ($`N=4,5,6,7`$) and resulted in $`\beta _{CAM}`$ values is agreement with those of simulations (see summarizing Table I).
Note, that the CAM extrapolation does not work for $`d>0.5`$ and $`d<0.1`$, because corrections to scaling are getting strong, and it is very difficult to find the steady state solutions of the GMF equations for $`N>7`$.
Time dependent simulations from a single active seed for these models have been proven to be a very efficient method since the slowly spreading clusters do not exceed the allowed system sizes hence do not feel finite size effects. Very precise estimates for the cluster ”mass” exponent $`N(t)t^\eta `$ and the cluster survival probability exponent $`P(t)t^\delta `$ have been obtained in DP and PC classes (see references in ). However example in the case of systems with infinitely many absorbing states these exponents seem to depend on initial conditions . A rigorous explanation is still missing. The simulation results for the cluster mass (Fig. 5) show strong dependence of $`\eta `$ on $`d`$. For $`d=0.5`$ and $`d=0.9`$, where $`\beta `$ and $`\delta `$ are roughly the same it changes from $`0.23(2)`$ to $`0.48(1)`$.
The survival probability exponent $`\delta `$ is roughly zero, expressing the fact that a single particle survives with finite probability (see table I).
The continuously changing exponent $`\eta `$ is in agreement with the preliminary observations of , however $`\beta `$ and the $`\delta `$ seem to be constant in the region: $`0.05d0.2`$ and $`0.5d0.9`$, suggesting two different universality classes.
The non-universal behavior of the cluster exponent may not be so surprising since it was also observed in the $`d=0`$ limit of the AF in the pair contact process .
In conclusion numerical simulations and analytical GMF+CAM calculations suggest two new universality classes in the AF model along the phase transition line. For small diffusion single particles cannot escape clusters where pair annihilation and creation dominates. For larger diffusions these single particles will wander among pair creation-annihilation clusters and the coupled BARW + ARW description of will be valid. These new universal behaviors may sound uneasy following two decades when the DP hypothesis has been found to describe almost every continuous transition to absorbing state, but do not contradict to it at all. In the AF model the absorbing state is not singlet, there are two (non symmetric) absorbing states where the system can evolve with equal probabilities. One of them is not completely frozen, but a single particle diffuses in it. My results do not support the conclusions of for a PC class transition without BARW2 process and $`Z_2`$ symmetry nor the first order transition for large $`d`$, but the exponents do not contradict to those of within numerical precision. It is likely that the two exponent ratios determined by are close to those of PC class accidentally. The cluster exponent $`\eta `$ behaves non-universally like in the pair contact process, the $`d=0`$ limit of the AF. Deeper understanding of this model that would reveal hidden symmetries responsible for the strange critical behavior of AF would be highly desirable.
Acknowledgements:
The author would like to thank E. Carlon, P. Grassberger, M. Henkel, H. Hinrichsen and J. F. F. Mendes for helpful discussions. Support from Hungarian research fund OTKA (Nos. T-25286 and T-23552) and from Bólyai (No. BO/00142/99) is acknowledged. The simulations were performed on Aspex’s System-V parallel processing system (www.aspex.co.uk).
|
warning/0004/quant-ph0004002.html
|
ar5iv
|
text
|
# Non-relativistic quantum electrodynamics for strong laser-atom interaction
## I Introduction
The availability of powerful sources of laser light has permitted in recent years the study of light-matter interaction in regimes where known approximations fail. Indeed, a number of new phenomena has appeared as high-order harmonics of the laser frequency, ionization with a number of photons well above the requested threshold and possibly, there are also strong theoretical indications of existence of stabilization, meaning by this that the ionization rate goes to zero by increasing the intensity of the laser field . There has been a lot of theoretical work following the appearance of those effects and, as a general approach, people aims to develop models that account for the physics of strong-laser atom interaction trying to support them both by numerical and experimental work. So, a very succesful model for harmonic generation has been firstly put forward in . This is the recollision model where is assumed that the outer electron goes into continuum due to the laser field and here is accelerated by the field, this step being described classically, then it recombines with the core generating the harmonics. A quantum improved version, understanding the emitted radiation as bremsstrahlung radiation due to the electron approaching the core has been given in . This improvement has the advantage that an explanation of the appearance solely of odd harmonics of the laser frequency is given. Another quantum account of the recollision model has been given in . To obtain these quantum models a number of reasonable assumptions have been made as the disregarding of intermediate states, the absence of resonances and so on, that need to be justified. These quantum versions of the recollision model prove to be a satisfactory explanation in all the experiments carried out so far with harmonics. Then, any theory starting from quantum electrodynamics has to cope with such a succesful understanding of the situation at hand.
Since the initial work by Kulander et al. a lot of numerical work is currently performed to understand strong laser-atom interaction. Anyhow, although this work is highly interesting our aim in this paper is quite different. That is, we are trying to rederive the recollision model from purely quantum arguments without any other approximation than the schemes of current experiments in high-intensity laser field require. Our conclusions concern the justification needed to understand the theoretical models discussed in Ref.. In this light, we take for granted that numerical computations support the recollision model as the experiments do.
After the pioneering work in , a lot of studies have also been carried out assuming a simple two-level model. So, one of the open problems in this field of research is if such a simple model really accounts for the physics of harmonic generation. In this light, the general structure of the spectrum has been obtained by Floquet theory and a parallel with the recollision model has been given in for the so called cut-off law, that is the rule to obtain the region of the spectrum where the intensity of the harmonics goes rapidly down. It should be said that such a model cannot account for ionization and so appears somewhat rough at best.
In the light of the above approaches, in this paper we will try to change the point of view beginning directly from quantum electrodynamics, taking as a starting point our work in Ref. that here is deepened and improved. Indeed, modelling is an important way of understanding but, if one is able to solve the equations of the full theory without resorting to numerical methods, again a deep way to understand physics is also given. So, our main aim is to study in depth what is the physics of strong laser-atom interaction by directly solving the Schrödinger equation. The theory we obtain is anyway enough general to be possibly applicable to other physical situations where strong electromagnetic fields play a role as e.g for Rydberg atoms.
It is interesting to note that, in view of using the dual Dyson series presented in and discussed here by using the Birkhoff theorem , the Pauli-Fierz transformation and its classical counterpart, the Kramers-Hennerberger transformation, turn out to play a dominant role. Indeed, when unitary transformations are applied, using probability amplitudes permits to extract physics without worrying about. This is a standard approach e.g. in quantum field theory for the interaction picture. Things do not change for the dual Dyson series through the Pauli-Fierz transformation.
Some approximations to start with are needed, but we take them directly from the experiments to tailor the Hamiltonian of quantum electrodynamics to the kind of problems we want to discuss. So, we take the electromagnetic field to be second quantized and the atom field in the non-relativistic approximation. The quadratic term of the vector potential in the Hamiltonian is considered just to see, through a single mode approximation, that taking it into account would shift the frequency of the laser in the harmonics. So, the condition to neglect it is also given. Finally, as the wavelengths of interest are much larger than the atomic radius, the dipole or long wavelength approximation is also taken. Then, a Pauli-Fierz transformation is applied to study strong laser-atom interaction. We will show that only if a large number of photons is present and the amplitude of the quiver motion of the free-electron is much larger than the atomic radius, a condition normally met in this kind of experiments and a key approximation that we will assume in this paper, than just odd harmonics are generated, assuming the atomic potential to be spherically symmetric. Indeed, for a large number of photons the Pauli-Fierz transformation reduces to the Kramers-Henneberger transformation as should be.
An interesting result obtained in this way is that, the atom appears to be a kicked quantum system, being this result rigorously derived from the Hamiltonian of quantum electrodynamics. The kicking hypothesis has been assumed for Rydberg atoms in a microwave field firstly in Ref. in order to explain the experimental results about ionization. Indeed, kicking can mean localization as a counterpart of classical chaos . This is one of the main results of this paper pointing toward the merging of such different fields of research.
A kicked Hamiltonian can in principle be solved exactly. Anyhow, we want to study an atom in a strong laser field doing perturbation theory. So, we will obtain a perturbation series having the product $`\omega \gamma `$ between the Keldysh parameter $`\gamma =\sqrt{\frac{I_B}{2U_p}}`$, being $`I_B`$ the ionization energy and $`U_p=\frac{e^2E^2}{4m\omega ^2}`$ the ponderomotive energy for an electric field $`E`$, and the laser frequency $`\omega `$ smaller than the distances, properly shifted by the laser field, between the energy levels. This accomplish the task to obtain a perturbation theory dual to the standard time-dependent theory applied to an atom in a weak electromagnetic field. Using this approach, we are able to show that the rate of above threshold ionization determines the duration of the harmonics in the spectrum and then, also the form of the spectrum itself by Fourier transform. Beside, if a rough model for the outer electron in the field of the rest of the atom is taken having a Coulomb form with a $`Z_{eff}`$ for the atomic number to give the correct ionization energy, it can be seen that the duration of harmonics agrees fairly well with experimental results as given e.g. in . This simply means to scale all the formulas for hydrogen-like atoms by ionization potential, ponderomotive energy and laser frequency.
The applicability of perturbation theory is indeed possible as, through a modified Rayleigh-Schrödinger perturbation theory one can show that the energy levels of the atom are shifted in such a way to change very little the wave function giving a prove of stabilization in the limit of laser frequency increasingly large. This is what we call “rigidity” of the wave function, very similar to the behavior of a superconductor due to the presence of the binding energy of the Cooper pair. This should give a hint, in the framework of quantum chaos, to study the change of statistics of such energy levels, possibly by some numerical work.
Let us finally point out that a number of methods have been devised starting from Volkov states, that are the solution of the Schrödinger equation for a free particle in a plane wave. The prototype of these approaches is given in Ref.. Anyhow, our point of view assumes different asymptotic states that are naturally derived from dual Dyson series as an entangled state between the radiation and the atomic state: This is just the leading order approximation. Then, e.g. no hypothesys a priori on the atomic potential being short ranged is needed. On the other side, no general proof is known of Volkov states being the proper asymptotic states for perturbation theory in a strong electromagnetic field. Beside, it appears very difficult to understand why just odd harmonics are experimentally observed as this approach keeps both odd and even harmonics . Anyhow, an improved version has been recently given in Ref. by the Volkov states in a second quantized field giving just odd harmonics. This approach satisfy the principle of duality in perturbation theory but what is changed is the initial state used to obtain the perturbation series and is generally presented as a non-perturbative method through the results for the computation of the amplitudes given in Ref..
The paper is so structured. In sec.II we give a brief presentation of duality for the Schrödinger equation and discuss the Birkhoff theorem using the dual Dyson series. In sec.III we give the Hamiltonian formulation of quantum electrodynamics to start with and show why the quadratic term is negligible. Then we perform a Pauli-Fierz transformation to obtain the dual Hamiltonian and show how it reduces to the Kramers-Henneberger form. A discussion of the two-level approximation with respect to the full theory is also given. In sec.IV we apply the approximation of large amplitude of the quiver motion in the laser field with the respect to the atomic radius to show how kicking arises and why the atomic wave function turns out to be rigid in the sense given above. In sec.V perturbation theory is done in the tunnelling regime deriving the rate of ionization and consequently the spectrum of the harmonics. Finally, in sec.VI the conclusions are given.
## II Duality and Birkhoff Theorem
The duality principle in perturbation theory, when applied to the Schrödinger equation written as (here and in the following $`\mathrm{}=c=1`$)
$$(H_0+H_1(t))|\psi =i\frac{|\psi }{t}$$
(1)
states that, by choosing $`H_0`$ as unperturbed Hamiltonian, the perturbation series has a development parameter exactly the inverse of the series obtained by taking $`H_1`$ as unperturbed Hamiltonian. Formally, this latter case means that we are multiplying $`H_1`$ by an ordering parameter $`\lambda `$ going to infinity. Duality principle is enough, as shown in Ref., to prove that the dual to the Dyson series is just the adiabatic approximation and its higher order corrections. Indeed, from the theory of singular perturbation we can rederive a similar result for the equations of probability amplitudes by the Birkhoff theorem , where “singular” means that the small parameter multiplies the higher derivative in the equation. In this way, a dual Dyson series can be obtained also for the probability amplitudes. In fact, the dual perturbation series is obtained by solving at the leading order the equation
$$\lambda H_1(t)|\psi ^{(0)}=i\frac{|\psi ^{(0)}}{t}$$
(2)
taking $`\lambda \mathrm{}`$. But this can be stated as a singular perturbation problem. Indeed, using the eigenstates of the unperturbed part $`H_0|n>=E_n|n>`$ and putting $`|\psi (t)>=_ne^{iE_nt}a_n(t)|n>`$, one has for eq.(1) the system of linear differential equations for the amplitudes
$$iϵ\frac{da_m(t)}{dt}=\underset{n}{}e^{i(E_nE_m)t}m|H_1(t)|na_n(t)$$
(3)
being now $`ϵ=\frac{1}{\lambda }`$. If the Hamiltonian $`H_0`$ has a finite set of $`N`$ eigenstates and eigenvalues and considering the amplitudes $`a_m(t)`$ as elements of a vector $`𝐚(t)`$, we can apply a result obtained by Birkhoff on 1908 in singular perturbation theory , that states that the equation (3) has a fundamental set of solutions $`𝐛_j(t)`$ with $`1jN`$ if the eigenvalues $`\lambda _j(t)`$ are distinct, given by
$$b_{kj}(t)=\mathrm{exp}\left[i_{t_0}^t\lambda _j(t^{})𝑑t^{}\right]\mathrm{exp}[i\gamma _j(t)]u_{kj}(t)+O(ϵ)$$
(4)
$`k`$-th element of the vector $`𝐛_j(t)`$, $`\gamma _j(t)=_{t_0}^t𝐮_j^{}(t^{})i\frac{d}{dt^{}}𝐮_j(t^{})`$ the geometrical part of the phase, $`𝐮_j(t)`$ the eigenstates of the matrix having elements $`A_{mn}=e^{i(E_nE_m)t}m|H_1(t)|n`$. The condition of no crossing of eigenvalues $`\lambda _j(t)`$ is also required. The final leading order approximation is then written as
$$𝐚(t)=\underset{j=1}{\overset{N}{}}\alpha _j𝐛_j(t)$$
(5)
with the coefficient $`\alpha _j`$ given by the initial conditions $`𝐚(0)`$. Again, we realizes that formally this is the adiabatic approximation applied to the operator $`e^{iH_0t}H_1(t)e^{iH_0t}`$. Then, we are arrived at similar conclusions as in Ref. for dressed states where, instead, duality principle was used, in the limit $`\lambda \mathrm{}`$. But, the use of duality principle permits to get rid of the limitation on the spectrum of $`H_0`$ being bounded to $`N`$ eigenstates and eigenvalues and then, it appears as a more general tool to treat also this kind of problems. So, we relax the condition on the spectrum of $`H_0`$ required by the Birkhoff theorem, as duality permits to give an alternative way to derive it.
Indeed, one can build the dual Dyson series by also doing an unitary transformation that removes the perturbation $`H_1(t)`$ in eq.(1). This defines a dual interaction picture. In fact, one can see that the unitary transformation operator must be a solution of
$$\lambda H_1(t)U_F(t)=i\frac{U_F(t)}{t}$$
(6)
and so $`U_F(t)`$ turns out to be the evolution operator of the adiabatic approximation in the limit $`\lambda \mathrm{}`$, proving the equivalence. But, this equation can also be solved exactly in some cases, without resorting to the adiabatic approximation giving the series
$$|\psi (t)=U_F(t)𝒯\mathrm{exp}\left[i_{t_0}^t𝑑t^{}U_F^{}(t^{})H_0U_F(t^{})\right]|\psi (t_0)$$
(7)
with $`𝒯`$ the time-ordering operator.
## III Non-relativistic Formulation of Quantum Electrodynamics
### A Full Hamiltonian and the Quadratic Term
The starting assumptions about the formulation of quantum electrodynamics that we need are the following. Firstly, we takes a non-relativistic approximation and neglect spin effects. Secondly, we cut-off the wavelengths limiting the analysis to the long ones that is, we take the dipole approximation: currently, this is in agreement with the experimental results for harmonic generation. Finally, we do second quantization on the electromagnetic field as we want to understand the properties of the scattered light in experiments with strong fields. So, we write the Hamiltonian in the Coulomb gauge (here and in the following $`\mathrm{}=c=1`$)
$$H=\underset{\lambda ,𝐤}{}\omega _𝐤a_\lambda ^{}(𝐤)a_\lambda (𝐤)+\frac{𝐩^2}{2m}+V(𝐱)+\frac{e}{m}\mathrm{𝐀𝐩}+\frac{e^2}{2m}𝐀^2$$
(8)
with $`\lambda =1,2`$ meaning a sum over polarizations, $`V(𝐱)`$ the atomic potential, $`a_\lambda ^{}(𝐤),a_\lambda (𝐤)`$ the creation and annihilation operators for the mode $`𝐤`$ with polarization $`\lambda `$. So, we have
$$𝐀=\underset{\lambda ,𝐤}{}g_𝐤(\mathit{ϵ}_\lambda ^{}a_\lambda ^{}(𝐤)+\mathit{ϵ}_\lambda a_\lambda (𝐤))$$
(9)
being $`\mathit{ϵ}_\lambda `$ the complex polarization vector, $`g_𝐤=\sqrt{\frac{2\pi }{V\omega _𝐤}}`$ and a normalization in a box of volume $`V`$ is assumed everywhere.
To see how much relevant is the quadratic term, we refer to the experimental result that just harmonics of the laser frequency are observed. Then, specializing the above Hamiltonian to a single mode , we derive the Heisenberg motion equation for the creation and annihilation operators. So, we take
$$H=\omega a^{}a+\frac{𝐩^2}{2m}+V(𝐱)+\frac{e}{m}\mathrm{𝐀𝐩}+\frac{e^2}{2m}𝐀^2.$$
(10)
Assuming a linear polarization, that is
$$𝐀=\sqrt{\frac{2\pi }{V\omega }}\mathit{ϵ}(a^{}+a)$$
(11)
we can use the same argument given in Ref. that the laser frequency should be shifted by squeezing. Indeed, one has the Heisenberg equations for $`a`$ and $`a^{}`$
$`{\displaystyle \frac{da}{dt}}`$ $`=`$ $`i\omega ai{\displaystyle \frac{e^2g^2}{m}}(a+a^{})i{\displaystyle \frac{eg}{m}}\mathit{ϵ}𝐩`$ (12)
$`{\displaystyle \frac{da^{}}{dt}}`$ $`=`$ $`i\omega a+i{\displaystyle \frac{e^2g^2}{m}}(a+a^{})+i{\displaystyle \frac{eg}{m}}\mathit{ϵ}𝐩`$ (13)
and without going into details of computation, by duality, as a first approximation, one neglects the atomic Hamiltonian and then, managing $`𝐩`$ as a c-number, the time dependence of the creation and annihilation operators is harmonic with frequency $`\mathrm{\Omega }=\sqrt{\omega ^2+\frac{4\pi e^2}{mV}}`$ and not just $`\omega `$ as should be required by the experimental results. So, squeezing terms can be neglected otherwise a shifted frequency of the harmonics would be observed . This effect should be more pronounced as greater is the order of the harmonic. But, considering this shift with current frequencies and gas densities really negligible, the quadratic term of the field in the Hamiltonian (8) can be systematically neglected giving the final Hamiltonian
$$H=\underset{\lambda ,𝐤}{}\omega _𝐤a_\lambda ^{}(𝐤)a_\lambda (𝐤)+\frac{𝐩^2}{2m}+V(𝐱)+\frac{e}{m}\underset{\lambda ,𝐤}{}g_𝐤(\mathit{ϵ}_\lambda ^{}a_\lambda ^{}(𝐤)+\mathit{ϵ}_\lambda a_\lambda (𝐤))𝐩.$$
(14)
This will be the starting point for our further analysis.
### B Pauli-Fierz Transformation and Kramers-Henneberger Hamiltonian
We now try to approach the Hamiltonian (14) using duality principle, that is, we try to compute the dual Dyson series. So, one can remove the perturbation by a Pauli-Fierz transformation given by
$$U_{PF}=\mathrm{exp}\left[\underset{𝐤,\lambda }{}\left(\beta _\lambda ^{}(𝐤)a_\lambda (𝐤)\beta _\lambda (𝐤)a_\lambda ^{}(𝐤)\right)\right]$$
(15)
where one has
$$\beta _\lambda (𝐤)=\frac{e}{m\omega _𝐤}g_𝐤\mathit{ϵ}^{}𝐩$$
(16)
and obtains the Hamiltonian
$$H_{PF}=\underset{𝐤,\lambda }{}\omega _𝐤a_\lambda ^{}(𝐤)a_\lambda (𝐤)+\frac{𝐩^2}{2m^{}}+V\left[𝐱i\underset{𝐤,\lambda }{}\left(\beta _\lambda ^{}(𝐤)a_\lambda (𝐤)\beta _\lambda (𝐤)a_\lambda ^{}(𝐤)\right)\right]$$
(17)
being $`m^{}`$ the renormalized mass due to the field. Here we try another way, to agree with duality in perturbation theory. That is, firstly we consider the Hamiltonian (14) in the interaction picture giving
$$H_I=\frac{𝐩^2}{2m}+V(𝐱)+\frac{e}{m}\underset{\lambda ,𝐤}{}g_𝐤\left[\mathit{ϵ}_\lambda ^{}a_\lambda ^{}(𝐤)e^{i\omega _𝐤t}+\mathit{ϵ}_\lambda a_\lambda (𝐤)e^{i\omega _𝐤t}\right]𝐩.$$
(18)
Then, we solve the leading order equation
$$\frac{e}{m}\underset{\lambda ,𝐤}{}g_𝐤\left[\mathit{ϵ}_\lambda ^{}a_\lambda ^{}(𝐤)e^{i\omega _𝐤t}+\mathit{ϵ}_\lambda a_\lambda (𝐤)e^{i\omega _𝐤t}\right]𝐩U_F(t)=i\frac{U_F(t)}{t}.$$
(19)
The solution is standard and can be written as
$$U_F(t)=\mathrm{exp}\left[\underset{𝐤,\lambda }{}\gamma _{𝐤,\lambda }(t)(\mathit{ϵ}_\lambda ^{}𝐩)a_\lambda ^{}(𝐤)\right]\mathrm{exp}\left[\underset{𝐤,\lambda }{}\gamma _{𝐤,\lambda }^{}(t)(\mathit{ϵ}_\lambda 𝐩)a_\lambda (𝐤)\right]\mathrm{exp}\left[\underset{𝐤,\lambda }{}\alpha _{𝐤,\lambda }(t)(\mathit{ϵ}_\lambda ^{}𝐩)(\mathit{ϵ}_\lambda 𝐩)\right]$$
(20)
with
$$\gamma _{𝐤,\lambda }=\frac{e}{m}\sqrt{\frac{2\pi }{\omega _𝐤V}}\frac{e^{i\omega _𝐤t}1}{\omega _𝐤}$$
(21)
and
$$\alpha _{𝐤,\lambda }=\frac{1}{2}|\gamma _{𝐤,\lambda }(t)|^2.$$
(22)
So, we have found in this way a time-dependent version of the above Pauli-Fierz transformation. But now, by eq.(7) we have a dual Dyson series for quantum electrodynamics. From this, one has e.g. that the leading order wave-function in the interaction picture, if we assume the atom in the ground state, is given by
$$|\psi ^{(0)}(t)_I=U_F(t)|1s|\alpha $$
(23)
being $`|1s`$ the atomic state and $`|\alpha `$ the initial state of the laser field described by a coherent state. As it should be expected this is an entangled state between the atomic state and the field state . It is easy to realize, using the momentum representation for the atomic state, that the scattered light is indeed coherent and preserves the property of the initial state for the field.
On this basis, we expect the light of the harmonics to be coherent as well . In fact, in order to evaluate higher orders in the dual Dyson series we rewrite the Pauli-Fierz transformation as
$$U_F(t)=\mathrm{exp}\left[\underset{𝐤,\lambda }{}\left(\gamma _{𝐤,\lambda }^{}(t)(\mathit{ϵ}_\lambda 𝐩)a_\lambda (𝐤)\gamma _{𝐤,\lambda }(t)(\mathit{ϵ}_\lambda ^{}𝐩)a_\lambda ^{}(𝐤)\right)\right],$$
(24)
obtaining the transformed Hamiltonian for higher order computations
$$H_F=U_F^{}(t)\left(\frac{𝐩^2}{2m}+V(𝐱)\right)U_F(t)=\frac{𝐩^2}{2m}+V\left[𝐱+𝐗(t)\right]$$
(25)
being
$$𝐗(t)=i\underset{𝐤,\lambda }{}\left(\gamma _{𝐤,\lambda }^{}(t)\mathit{ϵ}_\lambda a_\lambda (𝐤)\gamma _{𝐤,\lambda }(t)\mathit{ϵ}_\lambda ^{}a_\lambda ^{}(𝐤)\right).$$
(26)
Now, we specialize the above construction assuming that initially the atom is exposed to an intense laser field. It is well-known that the creation and annihilation operators can be expressed through the number operator $`n_{𝐤,\lambda }=a_{𝐤,\lambda }^{}a_{𝐤,\lambda }`$ and a phase operator by the Susskind and Glogower or the Pegg and Barnett construction . But, the distribution probability of the phase can be made to coincide in such a way to give for a coherent state a definite phase having quantum fluctuactions that go to zero for a very large mean number of photon, that is the case of the intensity of the laser field of interest. Then, one can keep just one mode and operate the substitution
$$a_{𝐤,\lambda }\sqrt{n_{𝐤,\lambda }}e^{i\varphi _{𝐤,\lambda }}.$$
(27)
The Hamiltonian $`H_F`$ is then reduced to the Kramers-Henneberger form for a classical field that is,
$$H_{HK}=\frac{𝐩^2}{2m}+V\left[𝐱\lambda _L\mathit{ϵ}(\mathrm{sin}(\omega t\varphi )+\mathrm{sin}(\varphi ))\right]$$
(28)
where a linear polarization has been assumed, all subscripts have been dropped and $`\lambda _L`$, the free-electron maximum excursion in a monochromatic field, can be rewritten as $`\lambda _L=\frac{eE}{m\omega ^2}`$, being $`E`$ the intensity of the laser field. The above Hamiltonian can be obtained by an unitary transformation, the Kramers-Henneberger transformation, for an atom in a monochromatic classical field. Then, we simply neglect quantum fluctuaction beyond the leading order due to the characteristic of the electromagnetic field. The relevance of this limit for a coherent state is that it makes consistent the arguments that follow about the properties of the harmonics. We just note that the advantage of such a derivation is that it gives us the coherence property of the scattered laser light in atomic scattering experiments, and permits to understand the way from a quantum electrodynamics formulation to a classical Hamiltonian that fully accounts for the situation at hand.
As a general approach, in agreement with the duality principle in perturbation theory , we see that to study strong laser-atom interaction we apply unitary transformations to remove the big part of the Hamiltonian that in this case is due to the field. We now apply the same idea also to the two-level model going to a dual interaction picture.
### C Two-level Approximation
The two-level model generally adopted in the study of harmonic generation can be cast in the form (e.g. )
$$H=\frac{\mathrm{\Delta }}{2}\sigma _3+\mathrm{\Omega }\sigma _1\mathrm{cos}(\omega t)$$
(29)
being $`\mathrm{\Delta }`$ the distance between the two-level, $`\mathrm{\Omega }`$ the strength of the field, $`\omega `$ the frequency of the laser field and $`\sigma _1`$, $`\sigma _3`$ the Pauli matrices. The first point to note is that, in order to be in agreement with the recollision model, one of the two levels should be in the continuum or, at best, $`\frac{\mathrm{\Delta }}{2}`$ is the ionization energy $`I_B`$ . In this way, one can rederive the well-known cut-off law for the frequency of the maximum harmonic $`I_B+3U_p`$, being $`U_p=\frac{e^2E^2}{4m\omega ^2}`$ the ponderomotive energy that in this model should be proportional to $`\mathrm{\Omega }`$ as pointed out in Ref.. But, in this model the continuum is missing and then, ionization cannot be described. This means to lose any connection between ionization and harmonic generation that, as we will see, plays a role in the duration of the harmonics itself. Finally, let us apply a Pauli-Fierz transformation (or Kramers-Henneberger transformation) by removing the oscillating term as
$$U_F(t)=\mathrm{exp}\left[i\sigma _1\frac{\mathrm{\Omega }}{\omega }\mathrm{sin}(\omega t)\right]$$
(30)
to obtain the transformed Hamiltonian
$$H_F=\frac{\mathrm{\Delta }}{2}J_0\left(\frac{2\mathrm{\Omega }}{\omega }\right)+\mathrm{\Delta }\sigma _3\underset{n=1}{\overset{\mathrm{}}{}}J_{2n}\left(\frac{2\mathrm{\Omega }}{\omega }\right)\mathrm{cos}(2n\omega t)+\mathrm{\Delta }\sigma _2\underset{n=0}{\overset{\mathrm{}}{}}J_{2n+1}\left(\frac{2\mathrm{\Omega }}{\omega }\right)\mathrm{sin}((2n+1)\omega t).$$
(31)
From this Hamiltonian is quite easy to realize that both odd and even harmonics play the same role and the only way to make ones or the other appear is to choose proper initial conditions. To show that things are really in this way, we turn back to Hamiltonian (29) and apply the Birkhoff theorem to rederive the spectrum already obtained in Ref., given in a general form by the Floquet theory in Ref..
So, we have to solve the equation
$$\left[\frac{\mathrm{\Delta }}{2}\sigma _3+\mathrm{\Omega }\sigma _1\mathrm{cos}(\omega t)\right]|\psi (t)=i\frac{d|\psi (t)}{dt}$$
(32)
that in the interaction picture becomes the system of equations
$`i{\displaystyle \frac{d}{dt}}\left(\begin{array}{c}a_1(t)\\ a_2(t)\end{array}\right)=\left(\begin{array}{cccc}0& \mathrm{\Omega }e^{i\mathrm{\Delta }t}\mathrm{cos}(\omega t)\hfill & & \\ \mathrm{\Omega }e^{i\mathrm{\Delta }t}\mathrm{cos}(\omega t)& 0\hfill & & \end{array}\right)\left(\begin{array}{c}a_1(t)\\ a_2(t)\end{array}\right)`$ (39)
to which the Birkhoff theorem can be applied. It should be pointed out that this approximation is consistent with $`\mathrm{\Delta }\mathrm{\Omega }`$ and we take $`\omega \mathrm{\Omega },\mathrm{\Delta }`$, in agreement with the identifications of Ref. and in the way the model should compare with experiments. Otherwise, the Birkhoff theorem should be applied differently. Under these conditions, we show that even harmonics can be present depending on the initial conditions. These harmonics are known as hyper-Raman lines in the current literature .
Firstly, we note that eq.(39) is the same as eq.(32) when use has been made of the eigenstates of $`\sigma _3`$. Then, we can simply use the results of Ref. about the dressed states of Hamiltonian (29) and obtain the Birkhoff basis as
$$𝐛_1(t)=e^{i\frac{\mathrm{\Delta }}{2}t}e^{i\frac{\mathrm{\Omega }}{\omega }\mathrm{sin}(\omega t)}\frac{1}{\sqrt{2}}\left(\begin{array}{c}1\\ e^{i\mathrm{\Delta }t}\end{array}\right)$$
(40)
and
$$𝐛_2(t)=e^{i\frac{\mathrm{\Delta }}{2}t}e^{i\frac{\mathrm{\Omega }}{\omega }\mathrm{sin}(\omega t)}\frac{1}{\sqrt{2}}\left(\begin{array}{c}e^{i\mathrm{\Delta }t}\\ 1\end{array}\right).$$
(41)
Then, we conclude that the Birkhoff theorem gives the same results obtained by other approaches in Ref.. The spectrum is then given by
$`\psi (t)|\sigma _1|\psi (t)`$ $``$ $`a_2(0)a_1^{}(0)e^{i\mathrm{\Delta }_Rt}+a_2^{}(0)a_1(0)e^{i\mathrm{\Delta }_Rt}`$ (42)
$`+`$ $`(|a_1(0)|^2|a_2(0)|^2)\mathrm{\Delta }{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}J_{2n+1}\left({\displaystyle \frac{2\mathrm{\Omega }}{\omega }}\right){\displaystyle \frac{\mathrm{cos}((2n+1)\omega t)1}{(n+\frac{1}{2})\omega }}`$ (43)
$`+`$ $`i(a_2^{}(0)a_1(0)e^{i\mathrm{\Delta }_Rt}a_2(0)a_1^{}(0)e^{i\mathrm{\Delta }_Rt})\mathrm{\Delta }{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}J_{2n}\left({\displaystyle \frac{2\mathrm{\Omega }}{\omega }}\right){\displaystyle \frac{\mathrm{sin}(2n\omega t)}{n\omega }}`$ (44)
with $`\mathrm{\Delta }_R=\mathrm{\Delta }J_0\left(\frac{2\mathrm{\Omega }}{\omega }\right)`$ the renormalized level separation. This form of the spectrum is in agreement with the one derived by the Floquet theory as given in . In the range of validity of this approximation as stated above, this proves the assertion that the even harmonics should be considered on the same ground as the odd ones for the two-level model. As we are going to show, this makes the applicability of this approximation to harmonic generation somewhat unappropriate as the appearance of just odd harmonics in the experiments has a deep physical meaning that here is overlooked. In fact, the two-level model relies just on the initial conditions to select the properties of the spectrum.
## IV Properties of Atoms in Intense Laser Fields
### A Odd Harmonics and Kicking for an Electron in a Coulomb Field
We now turn our attention to the classical limit of the Pauli-Fierz Hamiltonian that is, the Kramers-Henneberger Hamiltonian that here we rewrite assuming the phase $`\varphi =0`$ obtaining
$$H_{HK}=\frac{𝐩^2}{2m}+V\left[𝐱\lambda _L\mathit{ϵ}\mathrm{sin}(\omega t)\right].$$
(45)
The potential has the Fourier series
$$V\left[𝐱\lambda _L\mathit{ϵ}\mathrm{sin}(\omega t)\right]=\underset{k=0}{\overset{\mathrm{}}{}}v_k(𝐱)i^k\left[e^{ik\omega t}+(1)^ne^{ik\omega t}\right]$$
(46)
where
$$v_k(𝐱)=_1^1\frac{dx^{}}{\pi }V(𝐱\mathit{ϵ}\lambda _Lx^{})\frac{T_k(x^{})}{\sqrt{1x^2}}$$
(47)
and being $`T_k(x)`$ the $`k`$-th Chebyshev polynomial of first kind. In this section we study the terms having $`k>1`$, deserving the analisys of the static contribution with $`k=0`$, i.e. the dressed potential, for the next section.
We now take the unperturbed potential $`V(𝐱)`$ to be spherical symmetric. It easily realized that one has different parity for even components being $`v_{2n}(𝐱)=v_{2n}(𝐱)`$, and odd components being $`v_{2n+1}(𝐱)=v_{2n+1}(𝐱)`$. But, the unperturbed part of the Hamiltonian is invariant by parity transformation while, the odd time-dependent part breaks this simmetry. So, by perturbation theory one has at first order $`\psi (t)|x|\psi (t)=\psi ^{(0)}(t)|x|\psi ^{(1)}(t)+c.c.`$ being $`|\psi ^{(0)}`$ the initial atomic state such that $`\psi ^{(0)}(t)|x|\psi ^{(0)}(t)=0`$, and the only way not to obtain a null spectrum through perturbation theory is from the odd components. This result turns out to be in agreement with the general one obtained through Floquet states in Ref..
To apply the above scheme to the Coulomb potential one has to consider the integrals
$$v_k(𝐱)=\frac{Ze^2}{\lambda _L}_1^1\frac{dx^{}}{\pi }\frac{1}{\sqrt{\left(\frac{𝐱}{\lambda _L}x^{}\right)^2+\left(\frac{𝐲}{\lambda _L}\right)^2+\left(\frac{𝐳}{\lambda _L}\right)^2}}\frac{T_k(x^{})}{\sqrt{1x^2}}$$
(48)
but we have to study them around the origin of coordinates where the Coulomb potential turns out to be singular. Indeed, at small distances the theory given above has its shortcomings due to the dipole approximation and the neglecting of the relativity. Then, some regularization is needed for the above integrals. So, let us study the integrals (48) at the point $`y=z=a_0`$, introducing in this way the cut-off $`\eta =\frac{a_0}{\lambda _L}1`$, and we are reduced to the analysis of the leading term
$$v_k(𝐱)=\frac{Ze^2}{\lambda _L}_1^1\frac{dx^{}}{\pi }\frac{1}{\sqrt{\left(\frac{𝐱}{\lambda _L}x^{}\right)^2+2\eta ^2}}\frac{T_k(x^{})}{\sqrt{1x^2}}$$
(49)
very near the origin. These integrals turn out to be zero for odd Chebyshev polynomials when one takes $`x=0`$ and, for even Chebyshev polynomials one gets an extremum such that the integral is bounded. The extremum depends on $`\eta `$ but increases slowly. Then, we can remove the even terms, after we take $`v_{2n}`$ at their values at the origin, by an unitary transformation. The same cannot be done for odd terms as they go to zero at the origin. These integrals can be studied by a Taylor series at $`x=0`$ giving for the dipolar term
$$v_k^{dip}(𝐱)=\frac{Ze^2}{\lambda _L}_1^1\frac{dx^{}}{\pi }\frac{1}{(x^2+2\eta ^2)^{\frac{3}{2}}}\frac{T_k(x^{})x^{}}{\sqrt{1x^2}}\frac{x}{\lambda _L}$$
(50)
and the integral gives the cut-off dependent approximation
$$_1^1\frac{dx^{}}{\pi }\frac{1}{(x^2+2\eta ^2)^{\frac{3}{2}}}\frac{T_k(x^{})x^{}}{\sqrt{1x^2}}\frac{2}{\pi }\mathrm{log}(\eta )(2n+1)(1)^n.$$
(51)
that has a logarithmic divergence for the cut-off $`\eta `$ going to zero. If we intrdouce the ratio $`\frac{a_B}{\lambda _L}1`$ that defines the region where we want to study the physics of the model, we can reabsorb the divergence $`\mathrm{log}(\eta )`$ as a renormalization into it and introduce the physical ratio $`ϵ=\frac{2}{\pi }\frac{a_B}{\lambda _L}`$. This means that we have two bare constants $`ϵ_0`$ and $`\eta `$ and this goes to zero through the relation $`\eta =e^{\frac{ϵ}{ϵ_0}}`$ with $`ϵ`$ the physical ratio. Then, the limit $`ϵ_00`$ implies $`\eta 0`$. Through this redefinition, the dipolar terms are
$$v_k^{dip}(𝐱)=ϵ\frac{Ze^2}{\lambda _L}(1)^n(2n+1)\frac{x}{a_B}$$
(52)
In this way, we can now prove that the electron in a Coulomb field and an intense laser field, such to have the parameter $`ϵ1`$, undergoes kicking turning the problem into one of quantum chaos .
The question of a kicked hydrogen atom has been put forward to explain experiments on ionization of Rydberg atoms in an intense microwave field through the mechanism of dynamical localization . Pioneering work has also been done on strong fields and atoms, using kicked models e.g. in Ref.. But, a rigorous proof of this behavior derived directly from quantum electrodynamics has not been given yet. Here, we accomplish this task by completing the analysis of the time-dependent part of the Kramers-Henneberger Fourier components. To complete this derivation, we point out that a periodic distribution as $`\delta _T(t)=_{n=\mathrm{}}^+\mathrm{}\delta (tnT)`$ that has period $`T`$, can be defined through the coefficients of its Fourier series. So, e.g. $`\delta _T(t)=\frac{1}{T}_{n=\mathrm{}}^+\mathrm{}e^{i2n\pi \frac{t}{T}}`$ has all constant coefficients given by $`\frac{1}{T}`$. To have convergence of the Fourier series in the sense of distributions, the coefficients $`f_n`$ must satisfy the criterium of being slowly varying , i.e. $`|f_n|M|n|^k`$, being $`M`$ and $`k`$ two constants.
Then, by using eqs.(52) and one gets the dipolar potential as
$$\stackrel{~}{V}_{KH}=i\frac{Ze^2}{\lambda _L}\frac{x}{a_B}ϵ\underset{k=\mathrm{}}{\overset{+\mathrm{}}{}}(2k+1)e^{i(2k+1)\omega t}=2ϵ\frac{Ze^2}{\lambda _L}\frac{x}{a_B}\underset{k=0}{\overset{+\mathrm{}}{}}(2k+1)\mathrm{sin}((2k+1)\omega t).$$
(53)
The Fourier series appearing here has convergence just in the sense of distributions. We now give an explicit form for the dipolar term. Indeed, one has
$$\stackrel{~}{\delta }_{\frac{T}{2}}(t)=\underset{k=\mathrm{}}{\overset{+\mathrm{}}{}}(1)^k\delta \left(tk\frac{T}{2}\right)=\frac{2}{T}\underset{k=\mathrm{}}{\overset{+\mathrm{}}{}}e^{i(2k+1)\omega t}$$
(54)
so that, the dipolar term is defined through the derivative of a periodic distribution of period $`\frac{T}{2}`$. One can write at last
$$\stackrel{~}{V}_{KH}=ϵ\frac{\pi }{\omega ^2}\frac{Ze^2}{\lambda _L}\frac{x}{a_B}\frac{d}{dt}\stackrel{~}{\delta }_{\frac{T}{2}}(t)+\mathrm{}$$
(55)
and we draw the conclusion that the motion of an atom in an intense laser field undergoes kicking with period $`\frac{T}{2}`$ so that, localization, a typical effect of quantum chaos, can happen.
Now, we show that the Rydberg atoms used in ionization experiments with microwave are indeed in a regime where the ratio between the atomic dimensions and the amplitude of the quiver motion of the free-electron $`\lambda _L`$ in the laser field is largely smaller than unity, and the above theory applies. As a starting point we take Ref. for a typical experiment. The less favourable case is for an intensity of the microwave field of 2.5 V/cm, at 12.4 GHz and the Rydberg atom has $`n_0`$=98. This gives for the ratio $`\frac{n_0^2a_B}{\lambda _L}=\frac{n_0^2\omega }{\sqrt{8I_BU_p}}`$ the value 0.0027, largely lesser than unity. Instead, the most favourable case is given by the intensity of the electric field of 21 V/cm, at 18 Ghz and $`n_0`$=64 giving a ratio of 0.00029 improving the situation of about a magnitude order. Then, we can conclude that Rydberg atoms are kicked by the microwave field and localization can happen, in agreement with all the current researches. This result comes directly from quantum electrodynamics and so, it is fully justified.
For the case of the harmonic generation the situation is still better because the Keldysh parameter $`\gamma `$ given by $`\sqrt{\frac{I_B}{2U_p}}`$ can be lesser than one permitting a straightforward application of the perturbation theory. A typical example of an experiment in this regime is given in Ref.. This possibility can be fully exploited if a set of unperturbed states can be found to do perturbation theory. This is indeed the case as we will see in the next section.
### B Rigidity of the Atomic Wave Function
In order to do perturbation theory, one generally needs a full set of orthonormal functions to start with, representing the unperturbed system. But, here one has to diagonalize the Hamiltonian
$$H_0=\frac{𝐩^2}{2m}+_1^1\frac{dx^{}}{\pi }V(𝐱\mathit{ϵ}\lambda _Lx^{})\frac{1}{\sqrt{1x^2}}$$
(56)
that is generally an impossible task analitically.
What we want to do is to analyze this problem in the limit where the amplitude of the quiver motion of the free-electron in the laser field is much larger than the atomic radius, further specializing the above problem to the Coulomb potential.
To evaluate the degree of deformation due to the laser field in this approximation we apply a modified Rayleigh-Schrödinger approximation as obtainable from the time-dependent perturbation series. This can be accomplished easily if one makes the multipolar expansion of the dressed Coulomb potential in eq.(56) as
$$V_{KH}=\frac{Ze^2}{r}\left[1+\underset{n=1}{\overset{+\mathrm{}}{}}A_n\left(\frac{\lambda _L}{r}\right)^{2n}P_{2n}\left(\frac{x}{r}\right)\right]$$
(57)
being $`A_n=_1^1𝑑xx^{2n}/(\pi \sqrt{1x^2})`$ and $`P_n`$ the $`n`$-th Legendre polynomial. So, we approach the static part of the Kramers-Henneberg Hamiltonian differently from the time-dependent part. The reason to do that is that we expect very large shifts of the energy levels and that just very few terms of the multipolar series (57) really contributes to the matrix elements in the Rayleigh-Schrödinger series.
The way we compute the Rayleigh-Schrödinger corrections is taken from the time-dependent perturbation theory. Indeed, let us suppose that one can neglect the time dependent part of the Kramers-Henneberger Hamiltonian, so that one is left with the time-independent problem
$$H_A=\frac{𝐩^2}{2m}\frac{Ze^2}{r}+\delta V_{KH}(𝐱)$$
(58)
where we have put
$$\delta V_{KH}=\frac{Ze^2}{r}\underset{n=1}{\overset{+\mathrm{}}{}}A_n\left(\frac{\lambda _L}{r}\right)^{2n}P_{2n}\left(\frac{x}{r}\right)$$
(59)
and we want to solve the problem
$$H_A|\psi (t)=i\frac{|\psi (t)}{t}.$$
(60)
This is the way stabilization is studied in the Kramers-Henneberger frame when the limit $`\omega \mathrm{}`$ is taken. Here we are able to prove that stabilization indeed exists in the limit of large ratio of free-electron quiver motion and Bohr radius due to the rigidity of the atomic wave function.
We take as unperturbed states the ones of the Coulomb problem by setting
$$|\psi (t)=\underset{n}{}a_n(t)e^{iE_nt}|n$$
(61)
being
$$\left(\frac{𝐩^2}{2m}\frac{Ze^2}{r}\right)|n=E_n|n.$$
(62)
One gets for the amplitudes
$$ia_m(t)=m|\delta V_{KH}(𝐱)|ma_m(t)+\underset{nm}{}e^{i(E_nE_m)t}m|\delta V_{KH}(𝐱)|na_n(t)$$
(63)
and introducing $`b_m(t)=e^{i\delta E_mt}a_m(t)`$, $`\delta E_m=m|\delta V_{KH}(𝐱)|m`$ and $`\stackrel{~}{E}_m=E_m+\delta E_m`$, we arrive finally at the equations
$$ib_m(t)=\underset{nm}{}e^{i(\stackrel{~}{E}_n\stackrel{~}{E}_m)t}m|\delta V_{KH}(𝐱)|nb_n(t).$$
(64)
Then, if the shifts $`\delta E_m`$ are really large in the limit of large ratio between the quiver amplitude of the motion of the free electron and the Bohr radius, the Rayleigh-Schrödinger corrections to the initial wave-function are really small and this state is “rigid” with respect to the perturbation introduced by the laser field. Indeed, one has
$$b_m(t)=b_m(0)+\underset{nm}{}e^{i(\stackrel{~}{E}_n\stackrel{~}{E}_m)t}\frac{m|\delta V_{KH}(𝐱)|n}{\stackrel{~}{E}_n\stackrel{~}{E}_m}b_n(0)+\mathrm{}$$
(65)
where an adiabatic switching of the perturbation has been introduced. We see that the first order correction gives a modification of the Rayleigh-Schrödinger perturbation theory as in place of the unperturbed energy levels $`E_m`$ there are the modified energy levels $`\stackrel{~}{E}_m`$. This appears also as an improvement with respect to the Brillouin-Wigner perturbation series. Indeed, one has for the wave function
$$|\psi (t)=\underset{n}{}e^{i\stackrel{~}{E}_nt}a_n(0)|n+\underset{n}{}e^{i\stackrel{~}{E}_nt}\underset{kn}{}e^{i(\stackrel{~}{E}_k\stackrel{~}{E}_n)t}\frac{n|\delta V_{KH}(𝐱)|k}{\stackrel{~}{E}_k\stackrel{~}{E}_n}a_k(0)|k+\mathrm{}.$$
(66)
In this way is possible to show that the first-order correction is really small.
For the level shift one gets for the first few levels (we have put $`a_B`$ for the Bohr radius, $`l`$ for the orbital angular moment and $`l_z`$ for the third component of the orbital angular moment):
n=1
$`\delta E_{001}`$ $`=`$ $`l=0,l_z=0,n=1|\delta V_{KH}(𝐱)|l=0,l_z=0,n=1=0`$ (67)
n=2
$`\delta E_{102}`$ $`=`$ $`l=1,l_z=0,n=2|\delta V_{KH}(𝐱)|l=1,l_z=0,n=2={\displaystyle \frac{1}{240}}{\displaystyle \frac{Ze^2}{a_B}}\left({\displaystyle \frac{\lambda _L}{a_B}}\right)^2`$ (68)
$`\delta E_{112}`$ $`=`$ $`\delta E_{112}=l=1,l_z=1,n=2|\delta V_{KH}(𝐱)|l=1,l_z=1,n=2={\displaystyle \frac{1}{480}}{\displaystyle \frac{Ze^2}{a_B}}\left({\displaystyle \frac{\lambda _L}{a_B}}\right)^2`$ (69)
$`\delta E_{002}`$ $`=`$ $`l=0,l_z=0,n=2|\delta V_{KH}(𝐱)|l=0,l_z=0,n=2=0`$ (70)
n=3
$`\delta E_{223}`$ $`=`$ $`l=2,l_z=2,n=3|\delta V_{KH}(𝐱)|l=2,l_z=2,n=3={\displaystyle \frac{Ze^2}{a_B}}\left({\displaystyle \frac{1}{5760}}\left({\displaystyle \frac{\lambda _L}{a_B}}\right)^2+{\displaystyle \frac{1}{816480}}\left({\displaystyle \frac{\lambda _L}{a_B}}\right)^4\right)`$ (71)
$`\delta E_{213}`$ $`=`$ $`\delta E_{213}=l=2,l_z=1,n=3|\delta V_{KH}(𝐱)|l=2,l_z=1,n=3={\displaystyle \frac{Ze^2}{a_B}}\left({\displaystyle \frac{1}{11340}}\left({\displaystyle \frac{\lambda _L}{a_B}}\right)^2+{\displaystyle \frac{1}{204120}}\left({\displaystyle \frac{\lambda _L}{a_B}}\right)^4\right)`$ (72)
$`\delta E_{203}`$ $`=`$ $`l=2,l_z=0,n=3|\delta V_{KH}(𝐱)|l=2,l_z=0,n=3={\displaystyle \frac{Ze^2}{a_B}}\left({\displaystyle \frac{1}{5670}}\left({\displaystyle \frac{\lambda _L}{a_B}}\right)^2+{\displaystyle \frac{1}{136080}}\left({\displaystyle \frac{\lambda _L}{a_B}}\right)^4\right)`$ (73)
$`\delta E_{103}`$ $`=`$ $`l=1,l_z=0,n=3|\delta V_{KH}(𝐱)|l=1,l_z=0,n=3={\displaystyle \frac{1}{810}}{\displaystyle \frac{Ze^2}{a_B}}\left({\displaystyle \frac{\lambda _L}{a_B}}\right)^2`$ (74)
$`\delta E_{113}`$ $`=`$ $`\delta E_{113}=l=1,l_z=1,n=3|\delta V_{KH}(𝐱)|l=1,l_z=1,n=3={\displaystyle \frac{1}{1620}}{\displaystyle \frac{Ze^2}{a_B}}\left({\displaystyle \frac{\lambda _L}{a_B}}\right)^2`$ (75)
$`\delta E_{003}`$ $`=`$ $`l=0,l_z=0,n=3|\delta V_{KH}(𝐱)|l=0,l_z=0,n=3=0`$ (76)
n=4
$`\delta E_{334}`$ $`=`$ $`l=3,l_z=3,n=4|\delta V_{KH}(𝐱)|l=3,l_z=3,n=4`$ (77)
$`=`$ $`{\displaystyle \frac{Ze^2}{a_B}}\left({\displaystyle \frac{1}{32256}}\left({\displaystyle \frac{\lambda _L}{a_B}}\right)^2+{\displaystyle \frac{3}{50462720}}\left({\displaystyle \frac{\lambda _L}{a_B}}\right)^4+{\displaystyle \frac{25}{113359454208}}\left({\displaystyle \frac{\lambda _L}{a_B}}\right)^6\right)`$ (78)
$`\delta E_{324}`$ $`=`$ $`\delta E_{324}=l=3,l_z=2,n=4|\delta V_{KH}(𝐱)|l=3,l_z=2,n=4`$ (79)
$`=`$ $`{\displaystyle \frac{Ze^2}{a_B}}\left({\displaystyle \frac{1}{7208960}}\left({\displaystyle \frac{\lambda _L}{a_B}}\right)^4+{\displaystyle \frac{25}{18893242368}}\left({\displaystyle \frac{\lambda _L}{a_B}}\right)^6\right)`$ (80)
$`\delta E_{314}`$ $`=`$ $`\delta E_{314}=l=3,l_z=1,n=4|\delta V_{KH}(𝐱)|l=3,l_z=1,n=4`$ (81)
$`=`$ $`{\displaystyle \frac{Ze^2}{a_B}}\left({\displaystyle \frac{1}{53760}}\left({\displaystyle \frac{\lambda _L}{a_B}}\right)^2+{\displaystyle \frac{1}{50462720}}\left({\displaystyle \frac{\lambda _L}{a_B}}\right)^4+{\displaystyle \frac{125}{37786484736}}\left({\displaystyle \frac{\lambda _L}{a_B}}\right)^6\right)`$ (82)
$`\delta E_{304}`$ $`=`$ $`l=3,l_z=0,n=4|\delta V_{KH}(𝐱)|l=3,l_z=0,n=4`$ (83)
$`=`$ $`{\displaystyle \frac{Ze^2}{a_B}}\left({\displaystyle \frac{1}{40320}}\left({\displaystyle \frac{\lambda _L}{a_B}}\right)^2+{\displaystyle \frac{3}{25231360}}\left({\displaystyle \frac{\lambda _L}{a_B}}\right)^4{\displaystyle \frac{125}{28339863552}}\left({\displaystyle \frac{\lambda _L}{a_B}}\right)^6\right)`$ (84)
$`\delta E_{224}`$ $`=`$ $`\delta E_{224}=l=2,l_z=2,n=4|\delta V_{KH}(𝐱)|l=2,l_z=2,n=4`$ (85)
$`=`$ $`{\displaystyle \frac{Ze^2}{a_B}}\left({\displaystyle \frac{1}{13440}}\left({\displaystyle \frac{\lambda _L}{a_B}}\right)^2+{\displaystyle \frac{3}{4587520}}\left({\displaystyle \frac{\lambda _L}{a_B}}\right)^4\right)`$ (86)
$`\delta E_{214}`$ $`=`$ $`\delta E_{214}=l=2,l_z=1,n=4|\delta V_{KH}(𝐱)|l=2,l_z=1,n=4`$ (87)
$`=`$ $`{\displaystyle \frac{Ze^2}{a_B}}\left({\displaystyle \frac{1}{26880}}\left({\displaystyle \frac{\lambda _L}{a_B}}\right)^2+{\displaystyle \frac{3}{1146880}}\left({\displaystyle \frac{\lambda _L}{a_B}}\right)^4\right)`$ (88)
$`\delta E_{204}`$ $`=`$ $`l=2,l_z=0,n=4|\delta V_{KH}(𝐱)|l=2,l_z=0,n=4`$ (89)
$`=`$ $`{\displaystyle \frac{Ze^2}{a_B}}\left({\displaystyle \frac{1}{13440}}\left({\displaystyle \frac{\lambda _L}{a_B}}\right)^2+{\displaystyle \frac{9}{2293760}}\left({\displaystyle \frac{\lambda _L}{a_B}}\right)^4\right)`$ (90)
$`\delta E_{114}`$ $`=`$ $`\delta E_{114}=l=1,l_z=1,n=4|\delta V_{KH}(𝐱)|l=1,l_z=1,n=4={\displaystyle \frac{1}{3840}}{\displaystyle \frac{Ze^2}{a_B}}\left({\displaystyle \frac{\lambda _L}{a_B}}\right)^2`$ (91)
$`\delta E_{104}`$ $`=`$ $`l=1,l_z=0,n=4|\delta V_{KH}(𝐱)|l=1,l_z=0,n=2={\displaystyle \frac{1}{1920}}{\displaystyle \frac{Ze^2}{a_B}}\left({\displaystyle \frac{\lambda _L}{a_B}}\right)^2`$ (92)
$`\delta E_{004}`$ $`=`$ $`l=0,l_z=0,n=4|\delta V_{KH}(𝐱)|l=0,l_z=0,n=4=0`$ (93)
It easily realized that just very few terms of the multipolar series of $`\delta V_{KH}`$ really contribute to level shifts making the argument working. Beside, as it should be expected the s-states have no shift.
Indeed, for the matrix elements, assuming as initial state the ground state of the atom one obtains
n=2
$`l=1,l_z=1,n=2|\delta V_{KH}(𝐱)|l=0,l_z=0,n=1`$ $`=`$ $`0`$ (94)
$`l=1,l_z=0,n=2|\delta V_{KH}(𝐱)|l=0,l_z=0,n=1`$ $`=`$ $`0`$ (95)
$`l=0,l_z=0,n=2|\delta V_{KH}(𝐱)|l=0,l_z=0,n=1`$ $`=`$ $`0`$ (96)
n=3
$`l=2,l_z=2,n=3|\delta V_{KH}(𝐱)|l=0,l_z=0,n=1`$ $`=`$ $`{\displaystyle \frac{Ze^2}{a_B}}{\displaystyle \frac{1}{720}}\left({\displaystyle \frac{\lambda _L}{a_B}}\right)^2`$ (97)
$`l=2,l_z=1,n=3|\delta V_{KH}(𝐱)|l=0,l_z=0,n=1`$ $`=`$ $`0`$ (98)
$`l=2,l_z=0,n=3|\delta V_{KH}(𝐱)|l=0,l_z=0,n=1`$ $`=`$ $`{\displaystyle \frac{Ze^2}{a_B}}{\displaystyle \frac{\sqrt{150}}{10800}}\left({\displaystyle \frac{\lambda _L}{a_B}}\right)^2`$ (99)
$`l=1,l_z=1,n=3|\delta V_{KH}(𝐱)|l=0,l_z=0,n=1`$ $`=`$ $`0`$ (100)
$`l=1,l_z=0,n=3|\delta V_{KH}(𝐱)|l=0,l_z=0,n=1`$ $`=`$ $`0`$ (101)
$`l=0,l_z=0,n=3|\delta V_{KH}(𝐱)|l=0,l_z=0,n=1`$ $`=`$ $`0`$ (102)
n=4
$`l=3,l_z=3,n=4|\delta V_{KH}(𝐱)|l=0,l_z=0,n=1`$ $`=`$ $`0`$ (103)
$`l=3,l_z=2,n=4|\delta V_{KH}(𝐱)|l=0,l_z=0,n=1`$ $`=`$ $`0`$ (104)
$`l=3,l_z=1,n=4|\delta V_{KH}(𝐱)|l=0,l_z=0,n=1`$ $`=`$ $`0`$ (105)
$`l=3,l_z=0,n=4|\delta V_{KH}(𝐱)|l=0,l_z=0,n=1`$ $`=`$ $`0`$ (106)
$`l=2,l_z=2,n=4|\delta V_{KH}(𝐱)|l=0,l_z=0,n=1`$ $`=`$ $`{\displaystyle \frac{Ze^2}{a_B}}{\displaystyle \frac{13\sqrt{150}}{150000}}\left({\displaystyle \frac{\lambda _L}{a_B}}\right)^2`$ (107)
$`l=2,l_z=1,n=4|\delta V_{KH}(𝐱)|l=0,l_z=0,n=1`$ $`=`$ $`0`$ (108)
$`l=2,l_z=0,n=4|\delta V_{KH}(𝐱)|l=0,l_z=0,n=1`$ $`=`$ $`{\displaystyle \frac{Ze^2}{a_B}}{\displaystyle \frac{13}{15000}}\left({\displaystyle \frac{\lambda _L}{a_B}}\right)^2`$ (109)
$`l=1,l_z=1,n=4|\delta V_{KH}(𝐱)|l=0,l_z=0,n=1`$ $`=`$ $`0`$ (110)
$`l=1,l_z=0,n=4|\delta V_{KH}(𝐱)|l=0,l_z=0,n=1`$ $`=`$ $`0`$ (111)
$`l=0,l_z=0,n=4|\delta V_{KH}(𝐱)|l=0,l_z=0,n=1`$ $`=`$ $`0.`$ (112)
Again we see that just very few terms of $`\delta V_{KH}`$ really contribute to the matrix elements. Beside, this contribution is smaller if not zero with respect to the level shifts. Then, these results strongly support the statement that the first-order correction into the series (66) is really small in the limit $`\frac{\lambda _L}{a_B}1`$ as it should be. This also supports stabilization when the limit of laser frequency going to infinity is taken, with the ponderomotive energy being constant as in this limit, the time-dependent part of the Kramers-Henneberger Hamiltonian can be neglected .
Finally, we conclude that a fairly good approximation for the unperturbed Hamiltonian for our aims is
$$H_A\underset{n}{}\stackrel{~}{E}_n|nn|$$
(113)
where just diagonal terms are kept and the off-diagonal terms
$$H_A^{}=\underset{nm}{}|mn|m|\delta V_{KH}|n$$
(114)
are neglected. We take the Hamiltonian (113) for time-dependent computations in perturbation theory.
## V Perturbation Theory for Atoms in Intense Laser Fields
The experiments carried out to produce harmonics have a small laser frequency with respect to ionization and ponderomotive energy, then also the time dependent components of the Kramers-Henneberger Hamiltonian need to be considered.
Our aim is to make a perturbaion analysis of the Hamiltonian
$$H=\underset{n}{}\stackrel{~}{E}_n|nn|+2ϵ\omega \gamma \frac{x}{a_B}\underset{k=0}{\overset{+\mathrm{}}{}}(2k+1)\mathrm{sin}((2k+1)\omega t)$$
(115)
being
$$\gamma =\frac{Ze^2}{\lambda _L\omega }=\sqrt{\frac{I_B}{2U_p}}$$
(116)
the Keldysh parameter. So, the perturbation theory is applicable in this case just when $`\gamma 1`$ that defines the so-called tunnelling regime for the electron in the laser field. But, the situation here is more favourable as we have just to require that the distance between energy levels is much larger than $`\omega \gamma `$ and we are able to account for a larger number of experiments than the tunnelling regime would permit.
We write the equations for the probability amplitudes as
$$i\dot{a}_m(t)=iϵ\omega \gamma \underset{n}{}e^{i(\stackrel{~}{E}_n\stackrel{~}{E}_m)t}a_n(t)m\left|\frac{x}{a_B}\right|n\underset{k=0}{\overset{+\mathrm{}}{}}\left[(2k+1)\left(e^{i(2k+1)\omega t}e^{i(2k+1)\omega t}\right)\right].$$
(117)
Then, a generic term, out of resonance, will be written as
$`a_m(t)`$ $`=`$ $`a_m(0)+`$ (120)
$`iϵ\omega \gamma {\displaystyle \underset{n}{}}a_n(0)m\left|{\displaystyle \frac{x}{a_B}}\right|n{\displaystyle \underset{k=0}{\overset{+\mathrm{}}{}}}[(2k+1)({\displaystyle \frac{e^{i((2k+1)\omega \stackrel{~}{E}_n+\stackrel{~}{E}_m)t}1}{(2k+1)\omega \stackrel{~}{E}_n+\stackrel{~}{E}_m}}+`$
$`{\displaystyle \frac{e^{i((2k+1)\omega +\stackrel{~}{E}_n\stackrel{~}{E}_m)t}1}{(2k+1)\omega +\stackrel{~}{E}_n\stackrel{~}{E}_m}})]+\mathrm{}.`$
This result shows that when there are two resonant states $`m`$ and $`n`$ in the laser field, then we have Rabi flopping with Rabi frequency given by
$$\frac{\mathrm{\Omega }_R}{2}=ϵ\gamma (2k+1)\omega \left|m\left|\frac{x}{a_B}\right|n\right|$$
(121)
determined by the $`(2k+1)`$-th harmonic of the laser frequency. This implies a significant modification of the spectrum of the harmonics with respect to the observed patterns. Otherwise, out of resonance amplitudes are really small due to the large shifts in the limit of large $`\frac{\lambda _L}{a_B}`$ ratio. So, the interesting case is the resonance with the set of continuous states of the atom representing the situation of the recollision model. In this case, instead of flopping, we can have decay, that means ionization, and a probability for the electron to turn back to the core emitting radiation.
Then, we can apply the Wigner-Weisskopf argument to eq.(120). The first approximation is to reduce the system to the two levels really resonating and to assume all the other amplitudes to be small in the sense given above. This means to approximate eq.(120) by
$`\dot{a}_0(t)`$ $``$ $`ϵ\omega \gamma {\displaystyle \underset{𝐩}{}}e^{i(E_𝐩E_0)t}0\left|{\displaystyle \frac{x}{a_B}}\right|𝐩a_𝐩(t){\displaystyle \underset{k=0}{\overset{+\mathrm{}}{}}}(2k+1)e^{i(2k+1)\omega t}`$ (122)
$`\dot{a}_𝐩(t)`$ $``$ $`ϵ\omega \gamma e^{i(E_0E_𝐩)t}𝐩\left|{\displaystyle \frac{x}{a_B}}\right|0a_0(t){\displaystyle \underset{k=0}{\overset{+\mathrm{}}{}}}(2k+1)e^{i(2k+1)\omega t}`$ (123)
where we have labelled the eigenstates of continuum for the atom by the momentum $`𝐩`$ and $`E_0=I_B`$. We seek a solution of this equations by setting
$$a_0(t)=e^{\frac{\mathrm{\Gamma }^{}}{2}t}.$$
(124)
By substituting eq.(124) into eq.(123) one arrives at the condition
$$\frac{\mathrm{\Gamma }^{}}{2}=iϵ^2\gamma ^2\omega ^2\underset{𝐩}{}\left|0\left|\frac{x}{a_B}\right|𝐩\right|^2\underset{k=0}{\overset{+\mathrm{}}{}}(2k+1)^2\frac{1e^{i[E_𝐩E_0(2k+1)\omega ]t+i\frac{\mathrm{\Gamma }^{}}{2}t}}{(2k+1)\omega E_𝐩+E_0i\frac{\mathrm{\Gamma }^{}}{2}}$$
(125)
For small $`\mathrm{\Gamma }^{}`$ this indeed reduces to a time-independent expression
$$\frac{\mathrm{\Gamma }^{}}{2}=iϵ^2\gamma ^2\omega ^2\underset{𝐩}{}\left|0\left|\frac{x}{a_B}\right|𝐩\right|^2\underset{k=0}{\overset{+\mathrm{}}{}}(2k+1)^2\left[𝒫\frac{1}{(2k+1)\omega E_𝐩+E_0}i\pi \delta (E_𝐩E_0(2k+1)\omega )\right]$$
(126)
with $`𝒫`$ meaning the principal value. So, finally, one gets the a.c. Stark shift of the ground state of the atom given by
$$\frac{\delta \omega }{2}=ϵ^2\gamma ^2\omega ^2\underset{𝐩}{}\left|0\left|\frac{x}{a_B}\right|𝐩\right|^2\underset{k=0}{\overset{+\mathrm{}}{}}(2k+1)^2𝒫\frac{1}{(2k+1)\omega E_𝐩+E_0}$$
(127)
and the decay rate for the ionization of the atom
$$\mathrm{\Gamma }=2\pi ϵ^2\gamma ^2\omega ^2\underset{𝐩}{}\left|0\left|\frac{x}{a_B}\right|𝐩\right|^2\underset{k=0}{\overset{+\mathrm{}}{}}(2k+1)^2\delta (E_𝐩E_0(2k+1)\omega ).$$
(128)
Then, for the continuum one has
$$a_𝐩(t)iϵ\gamma \omega \underset{k=0}{\overset{+\mathrm{}}{}}𝐩\left|\frac{x}{a_B}\right|0(2k+1)\frac{e^{i[E_𝐩E_0(2k+1)\omega +i\frac{\mathrm{\Gamma }^{}}{2}]t}1}{E_𝐩E_0(2k+1)\omega +i\frac{\mathrm{\Gamma }^{}}{2}}$$
(129)
so that, taking
$$|\psi (t)=a_0(t)e^{iI_Bt}|0>+\underset{𝐩}{}e^{iE_𝐩t}a_𝐩(t)|𝐩$$
(130)
one has for the harmonic spectrum
$$\psi (t)|x|\psi (t)=\underset{𝐩}{}\left(e^{i(E_𝐩+I_B)t}a_0(t)a_𝐩^{}(t)𝐩\left|x\right|0+e^{i(E_𝐩+I_B)t}a_𝐩(t)a_0^{}(t)0\left|x\right|𝐩\right)$$
(131)
where use has been made of the fact that $`0|x|0=𝐩|x|𝐩=0`$. Assuming the laser light coming from the far past, eq.(129) can rewritten as
$$a_𝐩(t)iϵ\gamma \omega \underset{k=0}{\overset{+\mathrm{}}{}}𝐩\left|\frac{x}{a_B}\right|0(2k+1)\frac{e^{i[E_𝐩E_0(2k+1)\omega +i\frac{\mathrm{\Gamma }^{}}{2}]|t|}}{E_𝐩E_0(2k+1)\omega +i\frac{\mathrm{\Gamma }^{}}{2}}$$
(132)
and we are left with the resonant part of the spectrum given by
$`\psi (t)|x|\psi (t)_R`$ $``$ $`2\pi ϵ\gamma \omega {\displaystyle \underset{𝐩}{}}|0\left|{\displaystyle \frac{x}{a_B}}\right|𝐩|^2{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}(2n+1){\displaystyle \frac{\frac{\mathrm{\Gamma }}{2\pi }}{\left[E_𝐩E_0(2k+1)\omega \frac{\delta \omega }{2}\right]^2+\frac{\mathrm{\Gamma }^2}{4}}}\times `$ (134)
$`e^{\mathrm{\Gamma }|t|}\mathrm{cos}\left[(2n+1)\omega t+{\displaystyle \frac{\delta \omega }{2}}t\right].`$
This form of the spectrum gives two main results: Firstly, one has all the harmonics shifted by the same quantity $`\frac{\delta \omega }{2}`$ and secondly each harmonic has a Lorentzian form that, for very small $`\mathrm{\Gamma }`$ can be reduced to a Dirac $`\delta `$ distribution. In this case the integration in $`𝐩`$ can be done taking a plane wave as a final state and neglecting the shift of the spectrum, one gets an improved version of the result of Ref., that is
$$\psi (t)|x|\psi (t)_R\frac{2}{\pi }\frac{2^{\frac{17}{2}}}{3^{\frac{9}{2}}}\frac{Ze^2\omega }{U_p^2}\gamma ^5\underset{n=n_0}{\overset{+\mathrm{}}{}}\frac{x_n^{\frac{3}{2}}}{\left(x_n+\frac{2\gamma ^2}{3}\right)^5}\mathrm{cos}((2n+1)\omega t)e^{\mathrm{\Gamma }|t|}.$$
(135)
being $`x_n=\frac{(2n+1)\omega I_B}{3U_p}`$ and $`n_0`$ is a lower cut-off given by the integer value such that $`(2n_0+1)\omega I_B0`$ firstly foreseen in Ref.. Being scaled in this way through $`U_p`$, $`I_B`$ and $`\omega `$, a simple model of a single electron in an atom with atomic number $`Z_{eff}`$ can be adopted for complex atoms in a strong laser field and the above result used also in this case. As shown in Ref., the order of magnitude is correct. The same is true for $`\mathrm{\Gamma }`$ given by
$$\mathrm{\Gamma }=\frac{256}{3\pi ^2}\frac{\omega ^2}{U_p}\gamma ^2\underset{n=n_0}{\overset{+\mathrm{}}{}}\left[\frac{I_B}{(2n+1)\omega }\right]^{\frac{5}{2}}\left[1\frac{I_B}{(2n+1)\omega }\right]^{\frac{3}{2}}$$
(136)
that for the experiment described in Ref. yields for helium .012 eV and for neon .01 eV improving the results given in Ref.. This in turn means that the mean lifetime of the harmonics ranges from 52 to 68 fs and, by e-folding, after a time of few 10<sup>2</sup> fs the harmonics disappear in agreement with the experimental results. The result is independent from the order of the harmonics. We expect that the value of $`\mathrm{\Gamma }`$ should be in better agreement with the experimental results when $`U_p`$ increases or being the same, when the Keldysh parameter $`\gamma `$ decreases marking the so called tunnelling regime.
## VI Conclusions
We have given a full theory of harmonic generation starting from non-relativistic quantum electrodynamics in the dipole approximation. A relation with quantum chaos is obtained through a kicked model that originates from the model at small distances where a regularization procedure is needed. Perturbation theory can be done in the tunnelling regime, where the Keldysh parameter is small, as the atomic wavefunction turns out to be rigid due to the way the energy levels are shifted by the laser field. In this way a deep connection between ionization rate and harmonic spectrum is given. Finally, a fully understanding of the recollision model is obtained and this was our main aim.
|
warning/0004/hep-ph0004002.html
|
ar5iv
|
text
|
# Spectators effect in inclusive beauty decays*footnote **footnote *To appear in the Proceedings of the Sixth Workshop on High Energy Physics Phenomenology held at the Institute of Mathematical Sciences, Chennai, India during Jan. 3-15, 2000.
## Abstract
I review the role of the spectator quarks effect in the inclusive beauty decays. The evaluation of the expectation values of four-quark operators between hadronic states and its consequences are discussed.
preprint: UOM/NPh/HQP/00-2
Inclusive decays of heavy hadrons are described by the heavy quark expansion (HQE), an expansion in the inverse powers of the heavy quark mass ($`m`$) based on the operator product expansion (OPE) in QCD and the heavy quark effective theory (HQET) assuming quark-hadron duality . The leading order hadronic decay rate, proportional to $`m^5`$, is that of the free heavy quark. Corrections appear at $`O(1/m^2)`$ and beyond. They are due to the heavy quark motion inside the hadron and the chromomagnetic interaction at $`O(1/m^2)`$ and the spectator quarks processes at $`O(1/m^3)`$. The decay rate at order two in $`1/m`$ splits up into the mesonic one on the one hand and the baryonic on the other. This is because of the vanishing chromomagnetic interaction in the baryons with an exception of $`\mathrm{\Omega }_Q`$. Among the predictions of the HQE for the inclusive properties which are confronted by the experimental values like the lifetime of $`\mathrm{\Lambda }_b`$, semileptonic branching ratio of $`B`$ and the charm counting in the final state , we address the ratio $`\tau (\mathrm{\Lambda }_b)/\tau (B^0)`$ which is 0.9 by theory but 0.79 from experiment), the specators effect in charmless semileptonic decay of $`\mathrm{\Lambda }_b`$ on $`Br(bX_ul\nu _l)`$ and the validity of the assumption of quark-hadron duality.
In view of the discrepancy of the theoretical prediction with the experimental one for $`\tau (\mathrm{\Lambda }_b)`$, it is necessary to accomodate the contribution coming from the third order term in the HQE:
$$C(\mu )<H|(\overline{b}\mathrm{\Gamma }q)(\overline{q}\mathrm{\Gamma }b)|H>$$
(1)
where the Wilson coefficient, $`C(\mu )`$, describes the spectator quarks processes: in the decay $`Q(q)Q^{}q_1q_2(q)`$, if either $`q_1`$ or $`q_2`$ is the same as $`q`$, then both of them interfere destructively; if $`q_1`$ or $`q_2`$ is the antiquark of $`q`$, then they weakly annihilate; and the other one is the $`W`$-scattering: $`Qq_{1(2)}WQ^{}q_{2(1)}`$. These processes are found to enhance the decay rate of $`\mathrm{\Lambda }_Q`$. On the other hand, the central issue in the systematic incorporation of the spectators effect is the evaluation of the expectation values of the four-quark operators ($`EV_{FQO}`$). Traditionally, for mesons, the $`EV_{FQO}`$ is obtained, with the vacuum saturation approximation, in terms of the leptonic decay constant of the hadron, $`f_H`$; on the other hand, for baryons, the valence quark model is employed. This procedure and other methods found that the FQO do not account for the discrepancy. However, we have shown in our recent works that the FQO accounts for the difference in the lifetimes of $`\mathrm{\Lambda }_b`$ and B.
The $`EV_{FQO}`$ between hadronic states is related to the form factor characterising the light quark scattering off the heavy quark inside the hadron :
$$\frac{1}{2M_H}<H|(\overline{b}\mathrm{\Gamma }q)(\overline{q}\mathrm{\Gamma }b)|H>=|\mathrm{\Psi }(0)|^2=\frac{d^3}{(2\pi )^3}F(q^2)$$
(2)
In , representing the form factor by $`e^{q^2/4\beta ^2}`$, the wave function density is obtained as
$$|\mathrm{\Psi }(0)|^2=\frac{\beta ^3}{4\pi ^{3/2}}$$
(3)
where $`\beta `$ is determined by solving the Schrödinger equation in Variational procedure for the wave function $`\frac{\beta ^{3/2}}{\pi ^{3/4}}e^{\beta ^2r^2/2}`$ with the potential $`V(r)_{meson}=a/r+br+c`$ and $`V(r)_{baryon}=a/r+br+\beta r^2+c`$. In this description, the baryon is considered as a two body system of a heavy quark-diquark. The $`\beta `$’s for the hadrons are: $`\beta _B^{}=0.4,\beta _{B^s}=0.44`$ and $`\beta _{\mathrm{\Lambda }_b}`$ = 0.72, all in $`GeV`$ units. Using these values for the wave function density, the ratio of lifetimes of $`\mathrm{\Lambda }_b`$ and $`B`$ is found to be 0.79.
If one assumes that the HQE is an asymptotic expansion, then the expansion for the decay rate can safely be considered as converging at $`O(1/m^3)`$. Recently, Voloshin has analysed the relations between the inclusive decay rates of the charmed and beauty baryons triplet ($`\mathrm{\Lambda }_Q,\mathrm{\Xi }_Q`$) Voloshin did not consider that the expansion converges at the third order which is consequencial.. The relations depend only on the HQE and on the flavour symmetry under $`SU(3)_f`$. In this procedure, the EV<sub>FQO</sub> between baryon states is obtained using the differences in the total decay rates. In , strongly assuming that the HQE converges at $`O(1/m^3)`$, we extended the Voloshin analysis to $`SU(3)_f`$ triplet of the $`B`$ mesons, $`B^{}`$, $`B^0`$ and $`B_s^0`$. Their total decay rate splits up due to their light quark flavour dependence at the third order in the HQE. The differences in the decay rates of the triplet, are related to the third order terms in $`1/m`$ by
$`d\mathrm{\Gamma }_{B^0B^{}}`$ $`=`$ $`\mathrm{\Gamma }_0^{}(1x)^2\left\{Z_1{\displaystyle \frac{1}{3}}(c_0+6)+(c_0+2)\right\}O_6_{B^0B^{}}`$ (4)
$`d\mathrm{\Gamma }_{B_s^0B^{}}`$ $`=`$ $`\mathrm{\Gamma }_0^{}(1x)^2\left\{Z_2{\displaystyle \frac{1}{3}}(c_0+6)+(c_0+2)\right\}O_6_{B_s^0B^{}}`$ (5)
$`d\mathrm{\Gamma }_{B_s^0B^0}`$ $`=`$ $`\mathrm{\Gamma }_0^{}(1x)^2\left\{(Z_1Z_2){\displaystyle \frac{1}{3}}(c_0+6)\right\}O_6_{B_s^0B^0}`$ (6)
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
$$d\mathrm{\Gamma }_{\mathrm{\Lambda }_b\mathrm{\Xi }^0}=\frac{3}{8}\mathrm{\Gamma }_0^{}(c_{00}2)O_6_{\mathrm{\Lambda }_b\mathrm{\Xi }^0}$$
(7)
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 EV<sub>FQO</sub> are obtained for $`B`$ meson, as an average, from Eqs. (4-6): $`O_6_B=8.08\times 10^3GeV^3`$. The EV<sub>FQO</sub> for the baryon $`O_6_{\mathrm{\Lambda }_b\mathrm{\Xi }^0}=3.072\times 10^2GeV^3`$, 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 EV<sub>FQO</sub> for baryon is about 3.8 times larger than that of B. For these values $`\tau (\mathrm{\Lambda }_b)/\tau (B)=0.78`$. 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 $`\tau (\mathrm{\Lambda }_b)`$ = 1.20 $`ps`$.
We now turn up to the spectator quarks effect in $`\mathrm{\Lambda }_b\mathrm{\Lambda }_ul\nu _l`$ , in view of the ALEPH measurement of $`Br(bX_ul\nu _l)`$. When $`b`$ decays into $`ul\nu _l`$, the final state $`u`$ quark constructively interferes with the $`u`$ quark in the initial state. This effect increases the decay rate leading to the ratio, using the $`EV_{FQO}`$ for baryons obtained above,
$$\frac{\mathrm{\Gamma }(\mathrm{\Lambda }_bX_ul\nu _l)}{\mathrm{\Gamma }(bX_ul\nu _l)}=1.34$$
(8)
The $`b`$-baryon contributes about 10% to $`Br(bX_ul\nu _l)`$. The above estimate will have effect on the branching ratio considerably if there is no compensation from elsewhere. The esitmate above will increase if the spectators effect from $`\mathrm{\Xi }^0`$. It seems that any compensation is absent to offset the above estimate plus that one from $`\mathrm{\Xi }^0`$. This will, though modestly, effect the value of the CKM matrix element $`|V_{ub}|`$.
Concerning quark-hadron duality in the heavy hadron decays, we make inference that follows the results obtained above. The agreement found between theory and experiment for $`\tau (\mathrm{\Lambda }_b)`$, besides consistency in the $`B`$ mesons case, clearly signals that quark-hadron duality holds good in the HQE. In the previous case, Eqs. (2-3), by the choice of the form factor representation, we obtained $`\tau (\mathrm{\Lambda }_b)/\tau (B)`$, whereas in the latter it is the very assumption that the HQE converges at $`O(1/m^3)`$ which leads to the prediction for the ratio. The validity of the assumption that we made needs to be verified . In the recent lattice study, the authors stated that the role of the FQO is significant to explain $`\tau (\mathrm{\Lambda }_b)`$. We hope that their claim will throw light.
In conclusion, we make a note of warning. The evaluation of the $`EV_{FQO}`$ is model dependent one in the first case and is subject to the validity of the assumption on the convergence of the HQE in the latter one. The intriguing point is that $`|\mathrm{\Psi }(0)|^2`$ for meson is smaller that the esitmate in terms of the leptonic decay constant. The ratio $`|\mathrm{\Psi }(0)|_{\mathrm{\Lambda }_b}^2/|\mathrm{\Psi }(0)|_B^2`$ is larger than expected.
###### Acknowledgements.
The author is grateful to Prof. H. Yamamoto, Prof. Rahul Sinha, Dr Anjan Giri, Dr Rukmani Mahanta and Mr K. R. S. Balaji for useful discussions. I acknowledge the encouragement being shown by Prof. P. R. Subramanian and Dr. D. Caleb Chanthi Raj.
|
warning/0004/hep-ph0004041.html
|
ar5iv
|
text
|
# 𝐵_𝑐 Meson Production in Nuclear Collisions at RHIC
## I Introduction
The recent observation of candidate $`B_c`$-meson events by the CDF collaboration yields measurements for the ground state mass and lifetime which are consistent with expectations from nonrelativistic potential models . This system of states is expected to have properties intermediate between the $`J/\psi `$ and $`\mathrm{{\rm Y}}`$ systems (except for its longer lifetime due to the absence of quark annihilation processes). Thus it is natural to investigate the fate of such states when produced in a deconfined environment, where suppression relative to “expected” yields may serve as a signature for the existence of deconfinement.
We have previously reported some preliminary work toward this end . Our calculations of the expected yield of $`B_c`$ at RHIC utilize estimates of the coherent production of both a $`b\overline{b}`$ and $`c\overline{c}`$ pair in the same initial hard perturbative QCD process, followed by hadronization into the $`B_c`$ states. This process is of order $`\alpha _s^4`$, and falls quite rapidly with decreasing energy. The resulting prediction at RHIC energies for the bound-state fraction relative to initial b-quark production is in the range $`10^4\mathrm{\hspace{0.17em}10}^5`$ . Combining this result with the expected yield of b-quarks, we find that even at design luminosity there will be at most a handful of $`B_c`$-mesons produced at RHIC which decay into observable final states (all scalar, vector, excited, particle/antiparticle $`\overline{b}c`$ bound states will be called $`B_c`$-mesons or simply $`B_c`$, except when otherwise noted). Thus a scenario of production via calculable hard QCD interactions followed by suppression in a deconfined medium will be irrelevant for RHIC parameters.
What we now present is a new production mechanism which itself depends on the existence of a deconfined state. The new mechanism becomes relevant for the first time at RHIC energies, since typical central collision events will have multiple pairs of initially-produced charm quarks. Estimates using perturbative QCD to calculate the initial production of heavy quarks predict approximately 10 $`c\overline{c}`$ pairs in each central event in Au-Au collisions at $`\sqrt{s}`$ = 200 GeV . Due to the larger mass, only about 1 in 20 central events will produce a $`b\overline{b}`$ pair.
Consider the events in which there is a single $`b\overline{b}`$ pair, along with the expected 10 $`c\overline{c}`$ pairs. If a region of deconfined matter is subsequently produced in the space-time region encompassing the heavy quarks, a b-quark will be able to “find” any of the initially-produced charm quarks with which to produce the final $`B_c`$ bound state. More generally, our study shows that the formation of quarkonium states within a deconfined space-time domain resulting from the interaction of mobile heavy quarks offers a very interesting signature for deconfinement.
In Sec. II we study quantitatively the probability that in the deconfined quark-gluon plasma (QGP) phase the mobile $`c,\overline{c}`$-quarks can seek, find and bind to a $`\overline{b},b`$ quark produced in the same event. We also consider the dynamical evolution of these heavy quark bound states in the deconfined phase. In Sec. II A we establish the kinetic model for the evolution in time of the $`\overline{b}c,b\overline{c}`$ bound state population. The required cross sections are presented in Sec. II B, and the formation and dissociation reaction rates obtained.
To calculate the final $`B_c`$ yield, one needs an estimate of the charm quark density during the deconfined phase. We therefore consider in Sec. III A a generic model for the expansion and cooling of the QGP. This also enables the comparison of direct (initial) and microscopic thermal charm production in Sec. III B. We confirm that at RHIC energy charm is produced primarily in the initial parton interactions, with only a minor contribution arising from the thermal plasma processes. We also show that the charm density at hadronization remains significantly in excess of that which would be present if full chemical equilibrium were reached.
We then present in Sec. IV a study of the influence of the model parameters on the pattern of $`B_c`$ production at RHIC. We present in Sec. IV B the fractional yield per initial bottom quark pair in QGP. We find that the final yield is most sensitive to the initial charm density, and there is very little dependence on the initial temperature of the dense deconfined state. In contrast, we show in Sec. IV D that the relative yield of the first radially excited state to the ground state, $`B_c(2S)/B_c(1S)`$ is significantly more dependent on the initial temperature in the plasma phase.
## II Chemical kinetics of quarkonium abundance
In our scenario, the $`b\overline{c}`$ and $`\overline{b}c`$ states formed in the QGP will be subject to collisions with gluons, which will dissociate them into their constituent quarks. This mechanism can be thought of as the dynamic counterpart of the plasma screening scenario, in which the color-confinement force is screened away in the hot dense plasma . We do not distinguish between the vacuum and plasma values for the mass and binding energy of the 1S ground states. Both are significantly larger than typical temperatures expected at RHIC. Estimates for their behavior as a function of screening mass have been made , and indicate that such an approximation is reasonable. Thus we no longer distinguish between the plasma bound states and the physical $`B_c`$ mesons which can be observed after hadronization.
The primary formation mechanism is just the inverse of the breakup reaction, in which unbound heavy quarks are captured in the bound states, emitting a color octet gluon. The competition between the rates of these reactions integrated over the lifetime of the QGP then determines the final $`B_c`$ populations. Note that in this scenario it is impossible to separate the formation process from the breakup (suppression) process. Both processes occur simultaneously, in contrast to the situation in which the formation only occurs at the initial times before the QGP is present. (Of course, one can include this case also in our scenario by adjusting the initial conditions. However, for the case of $`B_c`$ initial production at RHIC, we have already shown that this possibility is negligible.) A number of other reactions involving b-quarks are possible in the QGP, but the rates are much smaller than those above. For example, formation of bound states with light quarks, $`B_s,B_u,B_d`$, are not possible at the high initial temperatures expected at RHIC since their lower binding energies prevents them from existing in a hot QGP, or equivalently they are ionized on very short time scales. They will be formed predominantly at hadronization, but this process is too late to affect the final $`B_c`$ population. We also neglect the decrease in b-quark population due to $`b\overline{b}`$ annihilation into light quarks or formation of $`\mathrm{{\rm Y}}`$ states. Both of these processes depend quadratically on the b-quark densities, and proceed too slowly or too rarely, respectively, to be significant. We also neglect the breakup cross section due to collisions with light quarks, since their population in the QGP is expected to be very much suppressed relative to gluons as compared to chemical equilibrium values during the early times when the breakup reaction rate has its most significant effect .
### A Chemical rate equations
The abundance of bottom and charm quarks and their bound states is thus governed by very simple master (population) equations involving only two reactions. Specifically, the formation reaction F
$$b+\overline{c}B_c+g,$$
and the dissociation reaction D
$$B_c+gb+\overline{c},$$
and similar equations for the conjugate states determine the time evolution of the number of bound states in the deconfined region, $`N_{B_c}`$.
$$\frac{dN_{B_c}}{d\tau }=\lambda _\mathrm{F}N_b\rho _{\overline{c}}\lambda _\mathrm{D}N_{B_c}\rho _g.$$
(1)
Since the total number of b-quarks does not change under the assumptions above, the rate of change in the number of unbound $`b`$ quarks $`N_b`$ is just the negative of that for the number of bound state $`B_c`$ mesons :
$$\frac{dN_b}{d\tau }=\lambda _\mathrm{D}N_{B_c}\rho _g\lambda _\mathrm{F}N_b\rho _{\overline{c}}.$$
(2)
In Eqs. (1,2), $`\tau `$ is the proper time in a small volume cell, $`\rho _i`$ denotes the number density $`[L^3]`$ of species i and the reactivity $`\lambda `$ $`\left[L^3/\mathrm{time}\right]`$ is the momentum distribution averaged reaction rate:
$$\lambda \sigma v_{\mathrm{rel}}=\frac{d^3p_1d^3p_2f_1(p_1)f_2(p_2)\sigma (\sqrt{s})v_{\mathrm{rel}}}{d^3p_1d^3p_2f_1(p_1)f_2(p_2)},$$
(3)
Note that even though $`\lambda `$ and $`\rho `$ are not Lorentz invariant, their product, the rate of particle production, is invariant. Thus, we evaluate the rates $`\rho _i\lambda _{\mathrm{D}/\mathrm{F}}`$ in the volume cell frame, where the (local) densities are given by external inputs.
We study here the deconfined period during which parton distributions have kinetically (but not necessarily chemically) equilibrated. We thus consider an isotropic medium of mobile quarks and gluons with the momentum distribution function $`f_i`$ for particle i given by the thermal equilibrium distribution $`f=(e^{E/T}\pm 1)^1`$. $`\sigma `$ is the spin- and color-averaged total cross section for the relevant reaction which depends only on $`s=(p_1+p_2)^2`$, and $`v_{\mathrm{rel}}`$ is the relative speed between the two reacting particles. Except where otherwise indicated $`\rho _g`$ is assumed to be in chemical equilibrium. $`\rho _{\overline{c}}`$ will be determined by its own kinetic equation described in Sec. III B. The spatial distribution of charm and bottom quarks will be taken to be uniform, an approximation which indicates the accuracy to which we will pursue our following calculations.
### B In-plasma cross section and reaction rates
We now need to estimate the cross sections involved in the formation and breakup reactions. For these purposes, we first utilize a derivation based on the interaction of a gluon field with the color dipole moment of a nonrelativistic heavy quarkonium state. It is implemented via an operator product expansion technique , and has been applied to the $`J/\psi `$ breakup rates in a QGP . We generalize this result to any heavy quarkonium 1S state with arbitrary flavor content, so that the spin and color averaged dissociation cross section is written
$$\sigma _\mathrm{D}=\frac{2\pi }{3}\left(\frac{32}{3}\right)^2\left(\frac{2\mu }{ϵ}\right)^{1/2}\frac{k_0^{7/2}}{4\mu ^2}\frac{(kk_0)^{3/2}}{k^5}.$$
(4)
Here $`k`$ is the gluon momentum (in the quarkonium-rest frame), $`k_0`$ is the minimum value required to impart the binding energy $`ϵ`$ to the bound state, and $`\mu `$ is the reduced mass. This form is valid if the quarkonium system has a spatial size small compared with the inverse of $`\mathrm{\Lambda }_{QCD}`$, and its bound state spectrum is close to that in a nonrelativistic Coulomb potential with $`ϵ`$ large compared with $`\mathrm{\Lambda }_{QCD}`$. These conditions are marginally satisfied for the $`J/\psi `$, and should be somewhat better for the $`B_c`$ kinematics. The form above has also been altered to account for recoil of the finite mass system, since in the original form the values of $`ϵ`$ and $`k_0`$ were identical.
We use values $`m_b`$ = 5.8 GeV and $`m_c`$ = 1.3 GeV which are consistent with typical potential model fits to the spectra. The magnitude of the cross section is controlled by the geometrical factor $`(4\mu ^2)^1`$, which for these quark masses is consistent with the size of the bound state wave function in the same potential models. The rate of increase just above threshold is due to phase space and the p-wave color dipole interaction, and it reaches a maximum value for $`B_c`$ dissociation of about $`1.5`$ mb when $`k=\frac{10}{7}k_0`$. For our model calculations we use a central value of 6.3 GeV for the $`B_c`$ mass, and a binding energy of 0.84 GeV which follows from the hadronic open flavor B and D meson masses. We retain these values for our kinetic calculations, as an approximation for physical values in the QGP. One might expect that both would decrease somewhat in a QGP. Some potential model calculations utilizing screening indicate the magnitude of this effect is expected to be small for the $`B_c`$ at RHIC conditions .
Our dominant formation cross section $`\sigma _\mathrm{F}`$ can then be directly obtained from $`\sigma _\mathrm{D}`$ by utilizing the detailed balance relation. This is written in the zero-momentum (ZM) frame of two-body interactions as
$$(\sigma 𝐩^2g)_\mathrm{D}=(\sigma 𝐩^2g)_\mathrm{F},$$
(5)
where $`𝐩_{\mathrm{D}/\mathrm{F}}`$ is the 3-momentum of the initial state particles for these reactions in their respective ZM frames, and $`g_{\mathrm{D}/\mathrm{F}}`$ is the statistical degeneracy in the two channels. In the present case of an unpolarized QGP, $`g_\mathrm{D}=((3+1)1)(28)`$, counting spin and color multiplicities of $`B_c`$ color-singlet and gluon, and $`g_\mathrm{F}=(23)(23)`$, counting spin and color multiplicities of the initial state b- and c-quarks. In this formalism we have included both the pseudoscalar and vector 1S $`B_c`$ states with the same cross sections, since in the nonrelativistic approximation one would expect the same spatial wave function and no spin dependence. These two states are also included in the no-deconfinement scenario initial production estimates , so that a direct comparison can be made. Using these formation - dissociation cross sections we calculate the reactivities as defined by Eq. (3). As shown in Fig. 1, formation and dissociation tend to balance at very high temperature, where the endotherm nature of the dissociation is negligible. At temperatures in the range 150–300 MeV the formation reactivity dominates by about a factor 4.
### C $`B_c`$ dissociation time scales
From Eq. (1), one sees that if the system is at a constant temperature (or equivalently the reaction rates are sufficiently fast), the final ratio of bound state to free quark populations is just given by the ratio $`(\lambda _\mathrm{F}\rho _{\overline{c}})/(\lambda _\mathrm{D}\rho _g)`$. The relevant time scales are set by the magnitudes of either factor in this ratio. We have calculated the dissociation rates of several quarkonium states under breakup by gluons with full chemical equilibrium density, and the results are shown in Fig. 2.
As expected, these rates rise quite sharply with temperature. The $`B_c`$ curve fits quite nicely between those for the $`J/\psi `$ and $`\mathrm{{\rm Y}}`$, also as expected. We also show the same calculation for the $`B_s`$ state, although the approximations made for this cross section have a very marginal validity in view of such a large state with small binding energy. For the $`B_c`$, one sees that in the range of initial temperatures expected at RHIC (roughly 300 to 500 MeV), these dissociation rates imply time scales of order 1–10 fm/c. Since the total QGP lifetimes are also in this region, it is evident that the equilibrium solutions will not in general be reached at each temperature, and one must solve the rate equations numerically to obtain the final populations.
## III Evolution of charm density in QGP
### A Evolution of temperature
In order to obtain the chemical evolution of quark bound state abundances in QGP we need to establish a relation between plasma temperature and proper time. We assume that the expansion of the QGP follows an isentropic path . We utilize a generic scenario for the proper-time dependence of the volume involving both longitudinal and transverse expansion, and examine the sensitivity of our final results to variations in the parameters involved. As a specific example we consider an adiabatically expanding homogeneous QGP domain having the shape of an ellipsoid of revolution about the longitudinal axis with semi-major and semi-minor axes parameterized with an initial length $`l_i`$/2 = $`\tau _0`$ and an initial transverse radius $`r_i`$. These are fixed at the time of QGP equilibration at an initial temperature $`T_0`$, when our formation and breakup reactions are assumed to start. We will explore the range $`0.5<l_i<2`$ fm (equivalent to thermalization times between 0.25 and 1.0 fm), and take $`r_i`$ = 5 fm for Au–Au collisions at RHIC, corresponding to 15–20% most central interactions. The longitudinal growth occupies the region between the (almost unstopped) receding nuclei. For transverse growth we allow a radial expansion at a speed $`v_r`$. The speed of sound in an ideal relativistic gas $`v_r0.58`$ c is taken as a nominal value, but the effects of significant variations in parameter space will be considered. Thus the volume evolves as a function of proper time $`\tau `$ according to:
$$V(\tau )=\frac{4\pi }{3}(r_i+v_r\tau )^2(\frac{l_i}{2}+\tau ).$$
(6)
Note that we have rescaled the initial proper time to zero.
This simple approach produces temperature vs. time profiles which appear to be very similar to those arising in more complex studies. These are shown in Fig. 3 for a homogeneous bulk QGP state in a range of initial temperatures $`0.3T_00.6`$ GeV. Intersection with the horizontal line at freeze-out temperature $`T_f=0.15`$ GeV indicates the QGP lifespan between thermalization and hadronization.
### B Thermal charm production in QGP
From fits to experimental nucleon-nucleon reaction data it is estimated that at RHIC there will be an average number $`N_{c_0}`$ = 10 of directly produced $`c\overline{c}`$ pairs per central collision , and this is the standard value for which our calculations will be carried out. However, additional charmed quarks could be produced in the QGP by collisions of gluons:
$$g+gc+\overline{c},$$
and of light quarks:
$$q+\overline{q}c+\overline{c}.$$
The local evolution of charm density $`\rho _c`$ can be described in a fashion quite similar to the model developed for strangeness production, see e.g. . Allowing for charm production and volume dilution under adiabatic expansion, the local charm density obeys:
$$\frac{d\rho _c}{dT}=\frac{A_c[T]}{\dot{T}}\left(1\left(\frac{\rho _c[T]}{\rho _c^{\mathrm{}}[T]}\right)^2\right)+3\rho _c\frac{1}{T},$$
(7)
where $`\dot{T}\mathrm{d}T/\mathrm{d}\tau `$ . The rate of charm production is
$$A_c[T]=\frac{1}{2}\rho _g^\mathrm{}^2\lambda ^{ggc\overline{c}}+\rho _q^\mathrm{}^2\lambda ^{q\overline{q}c\overline{c}},$$
(8)
where $`\rho ^{\mathrm{}}`$ denotes the density in thermal and chemical equilibrium. $`A_c[T]`$ is shown in Fig. 4, obtained with running QCD parameters, $`\alpha _s(M_Z)=0.118,m_c(1\mathrm{G}\mathrm{e}\mathrm{V})=1.5`$ GeV .
Integrating Eq. 7, we obtain the total number of charmed quark pairs in plasma at freeze-out, with initial temperature $`T_0`$ being a parameter, along with the initial longitudinal size $`l_i`$ . Fig. 5 shows our results for three different initial longitudinal sizes $`l_i=2`$ fm (dotted), $`l_i=1`$ fm (dashed), and $`l_i=0.5`$ fm (solid): the three thick up-curving lines include both the 10 directly produced charm pairs and the QGP charm production. The larger the initial volume and the longer the lifespan at high temperature, the greater is the QGP contribution to charm yield.
For the range of initial temperatures expected at RHIC, ones sees that there is virtually no additional charm produced in the QGP, and also that charm annihilation processes are too slow to significantly reduce the initially-produced number of charm quark pairs. The thin lines show for comparison the number of charm quarks which would follow from a chemical equilibrium density at the hadron freeze-out temperature $`T_f=0.15`$ GeV for $`m_c=1.5`$ GeV. (This value depends on the initial temperature $`T_0`$ through its effect on the freezeout volume.) The chemical equilibrium value is always significantly smaller than the corresponding direct + thermally produced abundance, i.e. direct initial charm production at RHIC is predicted in general to significantly oversaturate the statistical phase space at freeze-out.
## IV Discussion of Results
### A $`B_c`$ survival in plasma
To see what temperature range contributes to breakup, Fig. 6 shows the survival probability of one $`B_c`$ meson being placed in the QGP at different initial temperatures. At low initial temperature ($`T350`$ MeV) the state will most likely remain bound, while already at $`T500`$ MeV the plasma will most likely dissolve the $`B_c`$. One can infer from this result that the final state population is mainly dominated by the formation process as the plasma cools between these temperatures. This of course is not in exact correspondence with the actual physical process, which involves the formation of the bound state at any time after QGP forms at the initial temperature, where the expansion has already produced a partial decrease in the gluon density.
Another way to interpret the results shown in Fig. 6 is that the effective collision-mediated color screening is most effective in dissolving $`B_c`$ bound state when the average gluon energy $`\overline{E}_g3T`$ exceeds the binding energy. Using this argument we can also easily see that the flavor transfer reaction $`\overline{b}s+c\overline{b}c+s`$ is unlikely, since the $`\overline{b}s`$ bound states have much smaller binding energies and are color screened already at all temperatures above the hadronization temperature.
### B Fractional $`B_c`$ yields
We numerically integrate the evolution of the $`B_c`$ abundance as function of the temperature $`T`$, using our expansion model to obtain the temperature as a function of time. The initial conditions are $`N_{B_c}`$ = 0, and $`N_b`$ = 1. We use $`N_{c_0}`$ = 10 and the volume expansion model for the charm quark density, and a full thermal gluon density. The final $`B_c`$ population per initial $`b\overline{b}`$ pair is shown in Fig. 7. The factor of 2 in the axis label indicates that on average an equal number of particle and antiparticle bound states will be produced. Thus the numbers can be directly compared with the bound state ratio $`10^4\mathrm{\hspace{0.17em}10}^5`$ expected if no deconfinement occurs.
The nominal initial volume and expansion parameters are used. We explore a range $`0.2T_00.6`$ GeV of initial temperature in steps of 0.1 GeV. We see that the $`B_c`$ abundance grows approximately linearly during the entire expansion as temperature decreases. This verifies that the formation and dissociation reactions do not come to equilibrium during the QGP phase, as previously anticipated in . This linearity is not of any inherent physical origin, as can be seen by changing variables in Eq. 1 from $`\tau `$ to $`T`$ and noting that $`N_{B_c}N_b`$. The linearity in $`T`$ is seen to follow from the numerical values of the product $`\dot{T}[\tau ]V[\tau ]\lambda _\mathrm{F}`$ remaining almost constant for $`T_00.6`$ GeV . The important conclusion we draw from the results shown in Fig. 7 is that we can expect a rather $`T_0`$ independent fractional $`B_c`$-yield, which for the main benchmark of our assumptions is at 5%.
Shown by dashed lines in Fig. 7 is the scenario where gluon density is $`1/2`$ its thermal equilibrium value $`\rho _g^{\mathrm{}}(T)`$. This corresponds to an effective reduction in the available degrees of freedom for gluons in a QGP for temperatures that are not yet ‘asymptotically’ large . One sees the expected effect of reduced gluon density leading to reduced dissociation, and hence increased final bound state populations. However, the increase is substantially less than linear, indicating that the formation term in the rate equation is dominant.
In order to better understand this result, and also to illustrate the dependence on initial QGP volume and transverse velocity, we show in Fig. 8 the freeze-out fractional $`B_c`$-yield per one bottom quark pair, as a function of $`T_0`$, while also varying the other parameters.
As $`T_0`$ increases, the yield initially increases. However, as the initial temperature further increases, the dissociation reaction becomes more prominent and adversely impacts the yield, an effect which is not fully compensated by the additional QGP-produced charm available at these temperatures. This is seen comparing the thick up-curving lines with thin lines which do not include QGP-produced charm. However, this competition between formation and dissociation does lead to a very broad yield maximum in the vicinity of the expected range of initial temperatures at RHIC. This effect has been noted for the nominal expansion and initial volume parameters in Fig. 7. Here we see this effect persists for a variety of these parameters.
In order to establish a lower limit for the final $`B_c`$ fractional abundance, we show by the dashed and dotted curves in Fig. 8 the effects of increasing the expansion rate, thus decreasing the lifetime of the QGP. We see that this effect is less than that generated by a change in initial volume, which controls the formation rate through the change in charm quark density. The different initial volumes are controlled by varying $`l_i`$. The set of 6 lines in the middle corresponds to $`l_i=1`$ fm, ($`V_0=52`$ fm<sup>3</sup>), our standard scenario, while the upper set of lines corresponds to the high density case $`l_i=0.5`$ fm, ($`V_0=26`$ fm<sup>3</sup>), and the lower set of lines represents the low density case, $`l_i=2`$ fm, ($`V_0=105`$ fm<sup>3</sup>).
In all of these calculations we have used an initial charm quark number $`N_{c_0}`$ = 10, and for the range of temperatures expected at RHIC, this number will not change appreciably during the QGP lifetime. Eq. 1 then predicts that $`N_{B_c}`$ will be exactly linear with the initial number of charm quarks, and our numerical results verify this. The calculated final $`B_c`$ yields utilizing the average initial $`N_{c_0}`$ thus will correctly include the fluctuations expected according to a binomial (or Poisson) distribution. This linear property is not evident in the results shown in Fig. 8 in terms of the initial charm density, since there one is changing the density by changing the volume, and this also effects the lifetime of the QGP and through that the final $`B_c`$ abundance.
Shown in Fig. 9 is the fractional $`B_c`$-yield dependence on initial volume measured through variation of $`l_i`$ for a range of initial temperatures. One sees that the temperature variation is less important than the initial volume effect, at least in the range of these parameters which we consider. For a fixed $`T_0`$, a change in the initial volume not only causes a inverse variation in the initial charm density $`\rho _cl_i^1`$, but also influences the plasma life time $`\tau _L`$. Empirically it was found that $`\tau _Ll_i^{1/2}`$ is a good fit. These two factors combine two give the roughly $`l_i^{1/2}`$ dependence of the final fractional $`B_c`$ meson yield shown in Fig. 9.
### C Sensitivity to Breakup and Formation Cross Sections
The breakup cross section magnitude and shape is an essential part of our dynamical calculation, since in addition to providing a time scale, it provides through detailed balance the relative magnitude of the formation cross section. The form in Eq. 4 has been used to estimate the breakup rate of $`J/\psi `$ due to final state collisions with hadrons, via convolution with the gluon structure function of the hadron . Recently, there have been several additional attempts to model the hadron-quarkonium cross section , which has lead to results which typically are much larger and have a more rapid rise above threshold. Here we investigate the sensitivity of our prediction $`B_c`$ yields to such variations in the fundamental cross sections. We show in Fig. 10 the change in the predicted $`B_c`$ yields at RHIC which follow if our breakup cross section in Eq. 4 is increased by a factor of 2. Also shown are corresponding results if the cross section is assumed to immediately rise to its maximum value just above threshold, with the same overall magnitudes as before. All other model parameters are kept at the nominal values. One sees that in all cases the final $`B_c`$ yields are increased. One can understand this behavior as a combination of two effects. First, the detailed balance relation provides the relative magnitudes of formation and breakup rates. Second, an increase in the magnitude of the cross section just decreases the corresponding time scales, allowing the favored formation reaction to proceed further toward completion. Thus the cross section we have utilized provides a conservative lower bound for our $`B_c`$ production estimates at RHIC via this new mechanism.
### D Relative excited state $`B_c`$ yield
We have also calculated the ratio of 2S to 1S-state $`B_c`$ yields within our model scenario. This is prompted by the observation that the corresponding ratio in the charmonium system at SPS may serve as a thermometer of the QGP phase .
As a first estimate of this ratio we use as a 2S dissociation cross section Eq. 4 with $`ϵ_{2s}=0.25`$ GeV. This of course is only a rough guess, since one should change the parameters to those corresponding to a 2S state, but this state has binding and energy levels which are very marginal in terms of the constraints used for the validity of the cross section formula. The individual population equations can be solved independently, since both final bound state fractions are small enough that the source of b-quarks is not significantly decreased.
Fig. 11 shows the ratio of yields $`B_c`$(2S)/$`B_c`$(1S) at hadronization as a function of the initial temperature for the three different initial volumes we consider: $`l_i=2`$ fm (dotted line), 1 fm (dashed line), and 0.5 fm (solid line). It appears that this ratio is somewhat more sensitive to the initial temperature than the individual yields, varying by at least a factor of two within the expected range for RHIC.
We also find that this yield ratio is insensitive to initial charm abundance and production, as charm density enters linearly in both the $`B_c`$(2S) and $`B_c`$(1S) population equations. Also, the initial volume as shown in Fig. 11 does not alter the abundance ratio significantly. Thus this yield ratio may allow one to draw conclusions about initial temperatures present in the QGP phase at RHIC, independent of the other parameters. To estimate the systematic uncertainty that would arise in such a procedure, we consider the effects of changing the cross sections (formation and by detailed balance also breakup) in the following cases: (a) increasing the (2S) cross sections by a factor of two, (b) increasing both 2S and 1s cross sections by factor of two, and (c) decreasing the $`B_c`$(1S) and $`B_c`$(2S) binding energies by 100 MeV. In all three cases the overall shape of the results as shown in Fig. 11 remain nearly unchanged. However, we note an overall increase in the relative yield by a factor of two in case (a), practically no change in the result in case (b) and a decrease by a factor of 1/3 in case (c). The ratio $`B_c`$(2S)/$`B_c`$(1S) is thus primarily sensitive to initial temperature, but to be able to draw firm conclusions we need accurate relative 2S to 1S cross sections, and a good understanding of in-plasma $`B_c`$ binding energy.
## V Summary and Conclusion
We have shown that this new mechanism of quarkonium production in a deconfined medium predicts the minimum final state abundance of $`B_c`$ mesons at RHIC to be of the order of 5% per initial $`b\overline{b}`$ pair. This result is relatively insensitive to the initial temperature and volume of the QGP, as well as to changes in the transverse expansion dynamics. There is a linear dependence on the charm quark abundance in the initial state, which in fact is the primary controlling factor in the final state $`B_c`$ yield.
Fractional $`B_c`$ yields at the level of $`5\times 10^2`$ significantly exceed expectations based on initial coherent one step production in individual nucleon-nucleon interactions, where a relative yield in range $`10^4\mathrm{\hspace{0.17em}10}^5`$ is expected. Such a small yield would not be observable at RHIC. Even the 500-1000 times greater multistep yield we obtain may pose considerable experimental challenges. However, should such an experiment succeed to even roughly confirm these predictions for the fractional $`B_c`$ yield, we would have a convincing evidence for the mobility of charmed quarks over an extended space-time region in the dense phase.
While in principle one could argue that incoherent $`B_c`$ formation could also occur in the hadronic phase in collision of D-mesons with B-mesons, such a process requires localization in phase space of both these hadrons, which is highly unlikely. A calculation has been performed for the analogous case in the $`J/\psi `$ system, using D-meson interactions . It was found that this mechanism is negligible even at LHC energies, except for possibly some observable effects in the $`\psi ^{}`$ yield.
We emphasize here that an essential element of this calculation relies on the assumption that colored heavy quarks will be subject to large energy loss processes in a plasma . This is necessary if these heavy quarks are to exist in a common region of the phase space volume. Without this stopping, heavy quarks are highly unlikely to remain for the required period of time inside the thermal deconfined phase. The details of this scenario cannot be fully justified at present, and considerable effort has to be vested to better understand the mechanisms which determine the initial phase space distribution of heavy quarks formed in nuclear collisions.
Another issue is the final state reduction of $`B_c`$ mesons due to collisions with comoving hadrons. However, we would expect the smaller size of these bound states relative to those extensively studied in the $`J/\psi `$ system will produce a much less significant effect in the $`B_c`$ system.
Our work relies heavily on the presence of mobile charmed quarks: investigation of other bound state formation, such as $`J/\psi `$, is currently underway. Initial results indicate that while a significant fraction of charm will be in fact bound (at the 10% level) this is too little to significantly alter the results we have presented for $`B_c`$ yields. In fact the overall initial charm production uncertainty is at least as large as this effect. One such source is the possible shadowing of gluons in nuclei, which could reduce the primary yield by amounts in the tens of percents .
If the experimental techniques enable observation of these predicted $`B_c`$ yields with significant efficiency, one has the possibility to embark upon a general study of both the vacuum and plasma properties of the $`B_c`$ system. Even should this prove not to be possible in the RHIC energy range, it will most certainly be easier at the Large Hadron Collider (LHC), where the collision energy is 30 times greater. Several bottom quark pairs will be produced in each central nuclear collision, along with literally hundreds of charm quark pairs, leading to expectations involving copious production of $`B_c`$ mesons. A study of relevant parameters for this production mechanism at LHC is underway.
### Acknowledgements
This work was supported in part by a grant from the U.S. Department of Energy, DE-FG03-95ER40937 .
|
warning/0004/hep-th0004143.html
|
ar5iv
|
text
|
# Remarks on Dirac-like Monopoles, Maxwell and Maxwell-Chern-Simons Electrodynamics in D=(2+1)
## Introduction
Field-theoretic models defined in a (2+1)-dimensional space-time have been studied for nearly two decades. Actually, lower-dimensional models have provided many interesting results which do not take place in the (3+1)D world, e.g., Schwinger’ mechanism in QED<sub>2</sub> and fractional statistics in three dimensions . Consequently, lower-dimensional theories cannot be considered as mere lower limits of four-dimensional ones; they have rather revealed characteristics that are intrinsic to its dimensionality.
On the other hand, some (2+1)D theories, whenever supplemented by a Chern-Simons’ term, turn out to exhibit a new interesting physical content, as for example, Maxwell and Einstein-Hilbert actions . Furthermore, it has been claimed that such models (mainly those in the context of MCS) have relevance for a deeper understanding of some Condensed Matter phenomena, like the Quantum Hall Effect (QHE) and High-Tc Superconductivity (see also, Ref. ).
Although Maxwell and Maxwell-Chern-Simons (mainly the latter, in both Abelian and non-Abelian frameworks) have attracted a great deal of efforts, it is curious that one has not provided an “electrodynamical body” (Liénard-Wiechert-type potentials, Larmor-like formula and so forth) for such (say, Abelian) theories which would be similar to the one we have for (3+1)D Maxwell<sup>1</sup><sup>1</sup>1Although in a different approach, a classical analysis of the non-Abelian case (SU(2), more precisely) was performed by D’Hoker and Vinet.. Thus, we shall try to draw the attention to the fact that the “lack” of a complete “electrodynamical body” is related to some serious difficulties, for instance, in calculating $`A_\mu `$ (and $`F_{\mu \nu }`$) for a single accelerated point-like charge. In view of that, a Larmor-like expression relating energy-flux (radiation) and the acceleration of the sources is still missing.
We start the present work by studying the Maxwell (massless) case. Some results are discussed and a number of difficulties are pointed out. Following, we add a Chern-Simons term to the former model and some consequences of such a procedure are worked out. Going on, we analyse the issue concerning the introduction of a Dirac-like monopole within both models and some properties of its field. Some effects of its potential on an usual electric charge are discussed in both classical and quantum (non-relativistic) frameworks. We close this paper by pointing out some Conclusions and Prospects.
i) Classical Maxwell Electrodynamics in D=(2+1)
Let us consider the D=(2+1) Maxwell Electrodynamics (MED<sub>3</sub>) Lagrangian: <sup>2</sup><sup>2</sup>2Our conventions read: $`diag(\eta _{\mu \nu })=(+,,)`$, greek letters running 0,1,2; the 2-D spatial coordinates are labeled by latin letters running 1,2; and $`ϵ_{012}=ϵ^{012}=ϵ_{12}=ϵ^{12}=+1`$.
$`_{MED}={\displaystyle \frac{1}{4}}F_{\mu \nu }F^{\mu \nu }+j_\mu A^\mu .`$ (1)
The invariance of the action under local Abelian gauge transformations, $`A_\mu (x)A_\mu (x)_\mu \mathrm{\Lambda }(x)`$, is ensured by the conservation of the 3-current, say, $`_\mu j^\mu =0`$. Moreover with the usual definition of the field strength, $`F_{\mu \nu }=_\mu A_\nu _\nu A_\mu `$, we get $`F_{\mu \nu }=(F_{0i}=+(\stackrel{}{E})_i;F_{12}=B)`$. Next, the field-strength clearly satisfies $`_\mu F^{\mu \nu }=j^\nu `$ and $`_\mu \stackrel{~}{F}^\mu =0`$, whence there follow:
$$B=_t\stackrel{}{E}^{}+\stackrel{}{j}^{},\stackrel{}{E}=\rho \text{and}\stackrel{}{E}^{}=_tB,$$
where we have defined $`\stackrel{~}{F}^\mu =\frac{1}{2}ϵ^{\mu \nu \kappa }F_{\nu \kappa }=(+B;\stackrel{}{E}^{})`$, with the components of a dual-vector given by $`(\stackrel{}{U}^{})_i=ϵ_{ij}U_j`$.
The dynamical equation for the more basic quantity, $`A_\mu `$, reads (in the gauge $`_\mu A^\mu =0`$): $`^2A_\mu (x)=j_\mu (x)`$. The solutions to this wave-equation may be readily obtained by means of the well-known Green’s function method (or by applying the Hadamard’s Descent Method, see Ref. for further details). Such a function, $`G^{2+1}(xy)`$, may be explicitly worked out and reads (the advanced function is easily got by introducing a $`\mathrm{\Theta }(\tau )`$; $`\mathrm{\Theta }`$ is the usual step-function):
$`G_{ret}^{2+1}(xy)={\displaystyle \frac{\mathrm{\Theta }(\tau )}{2\pi }}{\displaystyle _0^{\mathrm{}}}J_0(kr)\mathrm{sin}(k\tau )𝑑k={\displaystyle \frac{\mathrm{\Theta }(\tau )}{2\pi }}{\displaystyle \frac{\mathrm{\Theta }(\tau ^2r^2)}{\sqrt{\tau ^2r^2}}},`$ (2)
where $`\tau =x^0y^0`$ and $`r=|\stackrel{}{x}\stackrel{}{y}|`$. The integral above may be found, for example, in Ref. (on page 731 and eq. 6.671-7). It is worth to notice that $`G^{2+1}`$ presents a quite different behaviour respect to its (3+1)D-counterpart, $`G^{3+1}`$: the support of $`G^{2+1}`$ lies no longer on the surface of the light-cone, where $`(xy)^2=0`$, as is the case for $`G^{3+1}(xy)=\delta [(x^0y^0)^2|\stackrel{}{x}\stackrel{}{y}|^2]/2\pi `$. Indeed, it rather spreads throughout the whole internal region of the light-cone, where $`(xy)^2>0`$ (blowing up as $`(xy)^20_+`$ and vanishing for space-like intervals, $`(xy)^2<0`$). Thus, the Huyghens principle is satisfied by $`G^{3+1}`$ and violated by $`G^{2+1}`$.
As we shall see, this will lead to profound modifications in planar electrodynamics with respect to the (3+1)D Maxwell theory. For example, by virtue of the failure of Huyghens principle, electromagnetic signals reverberate in (2+1) dimensions, and a Larmor-like formula for the radiated power appears to be a highly non-trivial task.
Next, by taking a single point-like charge, $`j^\mu (y)=q_{\mathrm{}}^+\mathrm{}\dot{z}^\mu (s)\delta ^{2+1}(yz(s))𝑑s`$, we get the general form for its potential (we have omitted the homogeneous part of the potential):
$`A_{ret}^\mu (x)=+{\displaystyle \frac{q}{2\pi }}{\displaystyle _{\mathrm{}}^+\mathrm{}}\mathrm{\Theta }(x^0z^0(s)){\displaystyle \frac{\mathrm{\Theta }[(xz(s))^2]}{\sqrt{(xz(s))^2}}}\dot{z}^\mu (s)𝑑s,`$ (3)
with $`(xz)^2=[(x^0z^0)^2|\stackrel{}{x}\stackrel{}{z}|^2]`$. The expression for the field-strength is also obtained in the usual way, and reads (with $`P=(xz)^\alpha \dot{z}_\alpha `$ and $`Q=(xz)^\alpha \ddot{z}_\alpha `$):
$`F_{\mu \nu }(x)={\displaystyle \frac{q}{2\pi }}{\displaystyle _{\mathrm{}}^+\mathrm{}}{\displaystyle \frac{\mathrm{\Theta }(x^0z^0)\mathrm{\Theta }((xz)^2)}{P^2\sqrt{(xz)^2}}}[\ddot{z}_\nu (xz)_\mu P+\dot{z}_\nu (xz)_\mu (1Q)\mu \nu ]ds.`$ (4)
Here, it is worthy noticing that, in general, we do not get to solve the expressions above. Actually, we have tried to solve elementary accelerated motions, say parabolic and hyperbolic ones. Unfortunately, we have found serious difficulties in performing some integrals that are highly non-trivial and plagued with serious divergences that have to be suitable handled<sup>3</sup><sup>3</sup>3It was already pointed out in the literature that (2+1)D Electrodynamics indeed imposes additional troubles in calculating some quantities; for example, in Ref., the author discusses some difficulties brought about by the logarithmic behaviour of the potential.. In (3+1)D, the scenario is quite different, because we have a $`\delta ^{3+1}((xz)^2)`$ (instead of $`\mathrm{\Theta }((xz)^2)/\sqrt{(xz)^2}`$) which, in turn, implies in a straightforward factorisation of the integral in $`s`$-variable, by picking up only those points for which $`(xz)^2=0`$.
Hence, we conclude that the “lack” of closed analytic expressions for $`A_\mu `$ (and $`F_{\mu \nu }`$) in the case of an arbitrary motion (Liénard-Wiechert-type expressions) is deeply related to the failure of the Huyghens’ principle, since the solutions to the $`^2`$-operator in (2+1)D, $`G^{2+1}`$, do not satisfy such a principle (indeed, the same happens for any $`G^{n+1}`$, $`n`$ even. See, for example, Ref.).
On the other hand, even the static case (the constant motion may be easily got by a Lorentz’ boost) reveals some of the new characteristics of the model. Thus, by taking $`z^\mu =(s,\stackrel{}{0})\dot{z}^\mu =(1,\stackrel{}{0}),`$ we get:
$`A^\mu (x)=`$ $`\{\begin{array}{c}A^0(\stackrel{}{r},t)=\frac{q}{2\pi }\mathrm{ln}|\stackrel{}{r}|+\frac{q}{2\pi }lim_{\tau +\mathrm{}}\left(\mathrm{ln}|\tau +\sqrt{\tau ^2r^2}|\right)\hfill \\ \stackrel{}{A}(\stackrel{}{r},t)=0\hfill \end{array}`$ (7)
$`F_{\mu \nu }(x)=`$ $`\{\begin{array}{c}F_{0i}(\stackrel{}{r},t)=+\frac{q}{2\pi }\frac{r^i}{r^2}\frac{q}{2\pi }r^ilim_{\tau r^+}\left(\frac{\tau }{r^2\sqrt{\tau ^2r^2}}\right)\hfill \\ F_{ij}(\stackrel{}{r},t)=0\hfill \end{array}.`$ (10)
Here, we notice that, besides the well-known $`\mathrm{ln}|\stackrel{}{x}|`$-behaviour of the potential in planar Electrodynamics, there is an extra term which explicitly diverges. Such a term clearly represents the asymptotic value of the potential as $`|\stackrel{}{x}|+\mathrm{}`$ and is directly related to the infrared divergence of the theory. Indeed, by calculating $`A_0(x)`$ by means of $`\stackrel{~}{A}_0(k)`$ (its Fourier transform), we may clearly see that such a term arises when the mass term is set to zero, as below:
$`A^0(\stackrel{}{r},t)=+{\displaystyle \frac{q}{2\pi }}{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{J_0(pr)}{p}}𝑑p=\underset{\mu 0_+}{lim}K_0(\mu r)\mathrm{ln}|\stackrel{}{r}|\underset{\mu 0_+}{lim}\mathrm{ln}(\mu /2).`$
Whence, we see that as $`m0_+`$, the last term above blows up. On the other hand, the explicitly divergent term appearing in the $`F_{\mu \nu }`$ above may be removed by a suitable subtraction procedure, which is possible because such a quantity vanishes asymptotically. \[Among others, such subtleties shall be more explicitly discussed in Ref.\].
It is interesting to pay attention to the appearance of such an infrared divergence at the classical level; indeed, infrared problems in (2+1)D are much more severe than in 4 dimensions. For example, the non-Abelian case, even in the presence of massive matter, makes sense only for very special gauge choices .
Still concerning the general $`F_{\mu \nu }`$-form, eq. (4), there remains an interesting issue to be pointed out. By taking into account the terms proportional to the acceleration, $`\ddot{z}(s)`$, which are those that effectively contribute to the energy-flux and so, to a Larmor-like formula, we notice that such terms are proportional to $`𝑑s/\sqrt{(xz)^2}`$, and might surprisingly lead us to the result that radiation in (2+1)D no longer falls off with $`r^1`$. Indeed it may increase proportionally to $`\mathrm{ln}|\stackrel{}{r}|`$, as long as $`z(s)`$ depends on $`s^2`$, which is the case for constant accelerated motions.
Next, let us point out a rather peculiar characteristic of the model as long as the propagation of electromagnetic signals is concerned. Let us start by considering the charge configuration: $`\rho (\stackrel{}{y},t^{})=q\delta ^2(\stackrel{}{y})\delta (t^{})`$. Its potential reads:
$`\mathrm{\Phi }_{pulse}(\stackrel{}{x},t)={\displaystyle \frac{q}{2\pi }}{\displaystyle \frac{\mathrm{\Theta }(t|\stackrel{}{x}|)}{\sqrt{t^2|\stackrel{}{x}|^2}}},`$ (11)
in contrast with its (3+1)D-counterpart $`\mathrm{\Phi }_{pulse}(\stackrel{}{x},t)=q\delta (t|\stackrel{}{x}|)/4\pi |\stackrel{}{x}|`$. Clearly, although such a signal has been sharply sent (at $`t=0`$ it was just at $`|\stackrel{}{x}|=0`$) it cannot later be recorded as a sharp one: the pulse develops a “tail” (its spreading in time) and so it reverberates. Therefore, we now need a very long time to record a sharp signal sent at an earlier time. Next, we obtain the superposed case, which is got from $`\rho (\stackrel{}{y},t^{})=q\delta ^2(\stackrel{}{y})\mathrm{\Theta }(t^{}),`$ and reads:
$`\mathrm{\Phi }_{sup}(\stackrel{}{x},t)=+{\displaystyle \frac{q}{2\pi }}\mathrm{ln}\left({\displaystyle \frac{t+\sqrt{t^2|\stackrel{}{x}|^2}}{|\stackrel{}{x}|}}\right)\mathrm{\Theta }(t|\stackrel{}{x}|),`$ (12)
whence we see that these signals superpose in a logarithmic way, differently from the (3+1) dimensions, where such a superposition takes place linearly, $`\mathrm{\Phi }_{sup}(\stackrel{}{x},t)=q\mathrm{\Theta }(t|\stackrel{}{x}|)/4\pi |\stackrel{}{x}|`$.
The logarithmic superposition leads us to an interesting point if we compare with previous results when $`t`$ is equal or slightly greater than $`|\stackrel{}{x}|`$: while the single pulse’ potential, eq. (11), appears to be very strong, the contrary happens to the superposed case, which is very weak there. However, as time goes by, things straighten up: while single pulses fall off, their superposition appears to broaden the potential. \[The expressions for the electric field are also easily obtained and exhibit similar phenomenon concerning reverberation, while the superposition is “better-behaved” than the $`\mathrm{\Phi }`$-potential\]. Moreover, notice that as (and only as) $`t\mathrm{}`$, we recover the static potential, eq. (7).
Thus, the results discussed above bring an additional complication to the (classical, at least) electrodynamics of a system of interacting charges, since even single pulses emitted by an electric charge will demand a very long time to be completely ‘felt’ by another one. In other words, even the static feature of the potentials and fields will be no longer determined only by the static configuration of the charges. It rather demands a very long time to actually happen, since at finite times the electromagnetic quantities are time-dependent.
Indeed, in (2+1)D, we may regard the classical propagation of a signal as if the wave-front travels with velocity $`c`$, and decreasing in a such a way that the back point of the signal has null-velocity (this is exactly what eq. (11) says).
Actually, similar conclusions concerning the reverberation of signals were already discussed by other authors . For instance, Courant and Hilbert in their classical book analyse such a propagation and, by virtue of the failure of the Huyghens principle, they conclude that D’Alembertian’ waves (in general), even if sharply produced, cannot be later recorded with the same sharpness.
Furthermore, we would like here to raise a question in view of what we have understood about the spreading that unavoidably affects the classical propagation of sharp signals in (2+1)D. By facing an electromagnetic signal rather as a wave, reverberation affects its propagation and we can no longer speak of sharp pulses; on the other hand, if we are to give the electromagnetic signal the status of a particle, we wonder whether the concept of photon as a localised energy packet should not be reassessed in the framework of planar Electromagnetism.<sup>4</sup><sup>4</sup>4An analogous question is pertinent in the MCS-case (next section). There, however, by virtue of the mass gap, reverberation is more expected to happen, since massive (Klein-Gordon or Proca-like) fields exhibit such a phenomenum even in (3+1) dimensions . (See also, in which is studied a modification of the standard electromagnetism, by the inclusion of a Lorentz- and Parity-violating Chern-Simons-like term in (3+1) dimensions).
ii) Maxwell-Chern-Simons model
Let us write the Lagrangian for the Maxwell-Chern-Simons Electrodynamics (MCS):
$`_{MCS}={\displaystyle \frac{1}{4}}F_{\mu \nu }F^{\mu \nu }+{\displaystyle \frac{m}{2}}ϵ^{\mu \nu \kappa }A_\mu _\nu A_\kappa +j_\mu A^\mu ,`$ (13)
where $`\frac{m}{2}ϵ^{\mu \nu \kappa }A_\mu _\nu A_\kappa =\frac{m}{2}A_\mu \stackrel{~}{F}^\mu `$ is the (Abelian) Chern-Simons term, which provides a mass for the boson, $`A_\mu `$, without breaking the original local gauge symmetry of the action , $`S_{MCS}=d^{2+1}x_{MCS}(x)`$. Moreover, the mass parameter, $`m`$, may be taken to be positive or negative. Depending on the choice of its signal, the ‘massive photon’ will carry polarisation equal to +1 ($`m>0`$) or -1 ($`m<0`$).<sup>5</sup><sup>5</sup>5 Talking about spin in (2+1) dimensions, we should be careful, since its meaning is rather different from its (3+1)D-counterpart. In fact, for a massive particle, its “spin” in (2+1)D has some similarities with the helicity of its massless correspondent in (3+1)D: only the positive, +1, or negative, -1, polarisations may take place, while no component of zero-polarisation appears. See Refs. . Notice, however, that in both cases, massless or massive, the “photon” carries only one physical degree-of-freedom, which highlights its ‘scalar nature’. Actually, since its mass is given by means of a topological mass term, we do not expect to have any additional degree-of-freedom.
In a similar way to the massless case, $`A_\mu `$-potential can be worked out and reads as below:
$`A^\mu (x)={\displaystyle d^{2+1}y\left[G^{mass}(xy)\eta ^{\mu \nu }+\frac{m}{m^2}\left(G^{mass}(xy)G^{2+1}(xy)\right)ϵ^{\mu \nu \kappa }_\kappa \right]j_\nu (y)},`$ (14)
where the massive Green’ function is given by:
$$G_{\stackrel{ret}{adv}}^{mass}(xy)=\frac{1}{2\pi }\frac{\mathrm{\Theta }[t^2r^2]\mathrm{cos}\left(m\sqrt{t^2r^2}\right)}{\sqrt{t^2r^2}}\mathrm{\Theta }[\pm t].$$
with $`t=x^0y^0`$ and $`r=|\stackrel{}{x}\stackrel{}{y}|`$. We clearly see that, as $`m0`$, then $`G^{mass}G^{2+1}`$. Similarly to its massless counterpart, $`G^{mass}`$ does not satisfy the Huyghens’ principle: again, the support spreads throughout the whole region $`(xy)^20`$.
Next, the general expression for $`A_\mu `$, as produced by a single point-like charge, takes the form:
$`A^\mu (x)=`$ $`+{\displaystyle \frac{q}{2\pi }}{\displaystyle _{\mathrm{}}^+\mathrm{}}ds\mathrm{\Theta }(x^0z^0(s))\mathrm{\Theta }[(xz)^2]\{{\displaystyle \frac{\mathrm{cos}(m\sqrt{(xz)^2})}{\sqrt{(xz)^2}}}\dot{z}^\mu +`$ (15)
$`+{\displaystyle \frac{m}{m^2}}ϵ^{\mu \nu \kappa }[\dot{z}_\nu (xz)_\kappa ({\displaystyle \frac{m\mathrm{sin}(m\sqrt{(xz)^2})}{(\sqrt{(xz)^2})^2}}+{\displaystyle \frac{\mathrm{cos}(m\sqrt{(xz)^2})1}{(\sqrt{(xz)^2})^3}})+`$
$`+\ddot{z}_\nu \dot{z}_\kappa \left({\displaystyle \frac{\mathrm{cos}(m\sqrt{(xz)^2})1}{\sqrt{(xz)^2}}}\right)]\},`$
from which we may notice the difficulties which arise in trying to solve it for arbitrary motions of the charge (indeed, the general solution to such an expression deeply depends on the massless one). There is also a new sort of term, not present in the massless case, which is explicitly acceleration-dependent (a radiation-like term, the last one in the eq. above). Such a term, in turn, will lead to another one that explicitly depends on $`d^3z/ds^3`$ in the expression for $`F_{\mu \nu }`$: a back-reaction-like term. By virtue of its length, we shall not give the explicit form for this field here. We refer the reader to Ref., where a detailed derivation of the results above will be presented. We only anticipate that the possibility that the radiation increases like a $`\mathrm{ln}|\stackrel{}{r}|`$ also takes place here.
Even though a general solution for $`A_\mu `$ (and $`F_{\mu \nu }`$) for arbitrary motions appears to be far off our possibilities, it is instructive to work out static quantities which already exhibit some of the new properties brought about by the Chern-Simons term. They read as follows:
$`A^\mu (x)=\{\begin{array}{c}\mathrm{\Phi }(\stackrel{}{x})=+\frac{q}{2\pi }K_0(m|\stackrel{}{x}|)\hfill \\ A^i(\stackrel{}{x})=\frac{q}{2\pi }\frac{m}{m^2}\frac{ϵ^{ij}x^j}{|\stackrel{}{x}|}\left(\frac{1}{|\stackrel{}{x}|}mK_1(m|\stackrel{}{x}|)\right)\hfill \end{array},`$ (18)
$`F_{\mu \nu }(x)=\{\begin{array}{c}E^i(\stackrel{}{x})=\frac{q}{2\pi }\frac{mx^i}{|\stackrel{}{x}|}K_1(m|\stackrel{}{x}|)\hfill \\ B(\stackrel{}{x})=+\frac{q}{2\pi }mK_0(m|\stackrel{}{x}|)=m\mathrm{\Phi }(\stackrel{}{x})\hfill \end{array}.`$ (21)
Now, we see that $`A_\mu `$ acquires a better asymptotic behaviour: $`A_\mu 0`$ as $`|\stackrel{}{x}|\mathrm{}`$ (at large distances, $`K_0`$ and $`K_1`$ roughly behave as $`e^{|m\stackrel{}{x}|}/\sqrt{|m\stackrel{}{x}|}`$). Indeed, even the long-range sector of $`\stackrel{}{A}`$ now decreases as $`|\stackrel{}{x}|^1`$. Such a sector is related to the well-known non-dynamical massless pole and also to the possibility of topological objects such as vortex-like magnetic field. In addition, due to the Chern-Simons term, the charge now produces a non-vanishing static magnetic field. Nevertheless, this does not lead to radiation at all. Indeed, it is easily to show that $`\stackrel{}{S}^{}=(\stackrel{}{E}^{}B)=0`$, with $`\stackrel{}{S}^{}`$ being the Poynting vector.
We should now comment on the short-distance behaviour of these quantities. By recalling that, for $`|z|1`$, the modified Bessel functions behave as $`K_0(z)\mathrm{ln}(z/2)`$ and $`K_1(z)z^1`$, we see that, near the charge, $`\mathrm{\Phi }`$ and $`B`$ diverge as $`\mathrm{ln}|m\stackrel{}{x}|`$ while $`\stackrel{}{E}`$ blows up as $`|\stackrel{}{x}|^1`$. The vector potential, on the other hand, exhibits a very peculiar behaviour: it vanishes as $`|\stackrel{}{x}|0`$! Such a result is actually in accordance with eq. (14): the $`A^i`$ components should vanish as $`\sqrt{t^2|\stackrel{}{x}|^2}0`$.
Moreover, the fact that, as $`|\stackrel{}{x}|0`$, $`B\mathrm{ln}|m\stackrel{}{x}|`$ implies that a charge within the Chern-Simons framework is a richer object than within a pure Maxwell context: along with its massive electric field, it also produces a flux “tube” of magnetic field, of width $`m^1`$ and strength $`q/m`$ (what demands $`m`$ to be suficiently large). It is precisely in this non-vanishing character of $`B`$ that there lies the possibility of the fractional statistics exhibited by such ‘charges’ .
Furthermore, it is easy to conclude, using eq. (14) for example, that upon $`mm`$, $`A^0=\mathrm{\Phi }`$ and $`\stackrel{}{E}`$ remains unchanged while $`\stackrel{}{A}`$ and $`B`$ changes their signals.
Next, we shall treat the propagation of signals in the Maxwell-Chern-Simons framework. We shall start by obtaining and analysing the single pulse case, which is produced by $`\rho (\stackrel{}{y},t^{})=q\delta ^2(\stackrel{}{y})\delta (t^{})`$. The quantities read (we have omitted $`\mathrm{\Theta }(t|\stackrel{}{x}|)`$ in all expressions below):
$`\mathrm{\Phi }_{pulse}(\stackrel{}{x},t)=+{\displaystyle \frac{q}{2\pi }}{\displaystyle \frac{\mathrm{cos}(m\sqrt{t^2|\stackrel{}{x}|^2})}{\sqrt{t^2|\stackrel{}{x}|^2}}},`$ (22)
$`A_{pulse}^i(\stackrel{}{x},t)={\displaystyle \frac{q}{2\pi }}{\displaystyle \frac{m}{m^2}}ϵ^{ij}_j\left({\displaystyle \frac{\mathrm{cos}(m\sqrt{t^2|\stackrel{}{x}|^2})1}{\sqrt{t^2|\stackrel{}{x}|^2}}}\right),`$
for the potentials, while the fields are:
$`E_{pulse}^i(\stackrel{}{x},t)=+{\displaystyle \frac{q}{2\pi }}^i\left({\displaystyle \frac{\mathrm{cos}(m\sqrt{t^2|\stackrel{}{x}|^2})}{\sqrt{t^2|\stackrel{}{x}|^2}}}\right)+{\displaystyle \frac{q}{2\pi }}{\displaystyle \frac{m}{m^2}}ϵ^{ij}_t_j\left({\displaystyle \frac{\mathrm{cos}(m\sqrt{t^2|\stackrel{}{x}|^2})1}{\sqrt{t^2|\stackrel{}{x}|^2}}}\right)`$
$`B_{pulse}(\stackrel{}{x},t)={\displaystyle \frac{q}{2\pi }}{\displaystyle \frac{m}{m^2}}_x^2\left({\displaystyle \frac{\mathrm{cos}(m\sqrt{t^2|\stackrel{}{x}|^2})1}{\sqrt{t^2|\stackrel{}{x}|^2}}}\right).`$
The reverberation of the pulse is evident: it is very strong when $`t`$ is equal or slightly greater than $`|\stackrel{}{x}|`$ and decreases as time goes by, vanishing as $`t\mathrm{}`$. The superposed case is obtained by integrating expressions above from $`|\stackrel{}{x}|`$ to $`t`$. For example, the scalar potential superposes as:
$`\mathrm{\Phi }_{sup}(\stackrel{}{x},t)={\displaystyle _{|\stackrel{}{x}|}^t}\mathrm{\Phi }(\stackrel{}{x},\tau )𝑑\tau =+{\displaystyle \frac{q}{2\pi }}{\displaystyle _{|\stackrel{}{x}|}^t}{\displaystyle \frac{\mathrm{cos}(m\sqrt{\tau ^2|\stackrel{}{x}|^2})}{\sqrt{\tau ^2|\stackrel{}{x}|^2}}}𝑑\tau ,`$ (23)
Here, a new result takes place in the MCS framework: we cannot exactly evaluate how electromagnetic signals superpose for arbitrary cases (say, finite times), since the integral above is not available, in closed form, unless $`t\mathrm{}`$ (the other electromagnetic quantities also depend on the same integral). At this limit, we get (see, for example, Ref. , page 419, eq. 3.754-2):
$$\underset{t\mathrm{}}{lim}_{|\stackrel{}{x}|}^t\frac{\mathrm{cos}(m\sqrt{\tau ^2|\stackrel{}{x}|^2})}{\sqrt{\tau ^2|\stackrel{}{x}|^2}}𝑑\tau =K_0(m|\stackrel{}{x}|),$$
which, in turn, leads us to the static potential, eq. (18), as $`t\mathrm{}`$. A similar scenario holds for the other quantities. Thus, we see that, in the case of the $`\stackrel{}{E}`$-field, only its longitudinal component survives asymptotically.
iii)Dirac-like monopole and its tangential electric field
Now, let us draw the attention to the introduction of a Dirac-like object into the previously studied models and to discuss some characteristics and consequences of the fields produced by this sort of monopole.
As it is well-known, such an (point-like) object shows up by breaking the Bianchi’ identity:<sup>6</sup><sup>6</sup>6In the Maxwell-Chern-Simons case, the naïve breaking of such an identity yields the breaking of gauge invariance. Thus, one should take into account that the monopole induces an extra electric current in order to balance $`_\mu j^\mu =0`$, and so restores gauge invariance (see Ref. for details. See also Ref. for an alternative approach to a similar problem in (3+1) dimensions). $`_\mu \stackrel{~}{F}^\mu =g`$, which in terms of the potentials gets the form:
$`{\displaystyle _t}𝑑t{\displaystyle _{xy}}d^2x\left(ϵ_{ij}[_i,_t]A_j(\stackrel{}{x},t)[_x,_y]\mathrm{\Phi }(\stackrel{}{x},t)\right)=g;`$ (24)
in the static limit, it reduces to:
$`[_x,_y]\mathrm{\Phi }(\stackrel{}{x})=g\delta ^2(\stackrel{}{x}).`$ (25)
Now, the above equation may be satisfied only if $`\mathrm{\Phi }`$ carries a ‘singular structure’. Indeed, by recalling that
$$[_x,_y]\mathrm{arctan}\left(\frac{y}{x}\right)=_x\left(\frac{x}{x^2+y^2}\right)+_y\left(\frac{y}{x^2+y^2}\right)$$
exactly coincides with
$$^2\mathrm{ln}\sqrt{x^2+y^2}=+2\pi \delta (x)\delta (y),$$
we identically solve eq. (25) by taking (as usual $`r=\sqrt{x^2+y^2}`$ and $`\phi =\mathrm{arctan}(y/x)`$)
$`\mathrm{\Phi }(\stackrel{}{x})={\displaystyle \frac{g}{2\pi }}\mathrm{arctan}\left({\displaystyle \frac{y}{x}}\right)\mathrm{\Phi }(r,\phi )={\displaystyle \frac{g}{2\pi }}\phi ,`$ (26)
The appearance of the angle-function above suggests us the need for a single-valuedness requirement: $`\mathrm{\Phi }(\phi )=\mathrm{\Phi }(\phi +2\pi n)`$. Its remarkable angular (instead of being radial) dependence leads to a very interesting (static) electric field ($`\stackrel{}{}=(\mathrm{\Phi }+_t\stackrel{}{A})`$, as usual):
$`\stackrel{}{}(x,y)=+{\displaystyle \frac{g}{2\pi }}{\displaystyle \frac{x\widehat{y}y\widehat{x}}{x^2+y^2}}\stackrel{}{}(r,\phi )=+{\displaystyle \frac{g}{2\pi }}{\displaystyle \frac{\widehat{e}_\phi }{r}}.`$ (27)
Whence, we clearly see the announced property of the $`g`$-monopole: it yields a (static) tangential electric field<sup>7</sup><sup>7</sup>7Strictly speaking, such a field does not produce a genuine Newton’s force on another charge (usual or peculiar one), since the force between them does not lie on the line that links both particles, as may be readily seen.. \[As far as we have seen, such a peculiarity takes place only in (2+1)D Electrodynamics. Furthermore, we do expect that such a property survives at time-dependent regimes\]. Moreover, it is worth noticing that a point-like magnetic vortex is characterised by a vector potential identical in structure to the tangential electric field above. Thus, we may identify a “duality” between both objects: the vortex is obtained from the monopole (more precisely, from its “string” -see below) by taking the electric field and the charge of the first to be respectively the vector potential and the magnetic flux associated to the latter.
On the other hand, it is a well-known fact that in (2+1)D the ‘worldline’ of a monopole is reduced to a point in (2+1) dimensional space-time (see, for example, Ref.; see also Ref. ). Therefore, the singular point above cannot be identified with the monopole itself. Actually, the modified Bianchi equation, $`_\mu \stackrel{~}{F}^\mu =g\delta ^2(\stackrel{}{x})`$, have to be rather viewed as an equation for the “string” of to the monopole. What happens is that, at static limit the “string” (indeed, reduced to a spatial point in the (2+1)D-case) appears to be localised at the origin.
Although such a localisation seems to state us that $`g`$ should be rather faced as a peculiar electric charge, we stress that this is not so. Indeed, what occurs is that, at static limit, the vanishing of radiation, $`\stackrel{}{S}^{}d^2x=(\stackrel{}{}^{})d^2x=0`$, demands that the monopole’ magnetic field must also vanish. \[Notice that such a requiriment, $`=0`$, is intimately related to the tangential feature of $`\stackrel{}{}`$, once that $`\stackrel{}{}^{}`$ becomes radial, and so $`\stackrel{}{}^{}0`$\]. Hence, what we may state is that such an object yields only non-vashing (tangential) electric field at the static limit.
Next, we analyse the (classical) dynamics of a usual electric charge, $`q`$, with mass $`m`$, moving under the action of such a tangential field. Its equations of motion are easily obtained and read as follows:
$`{\displaystyle \frac{2\pi m}{gq}}\ddot{x}={\displaystyle \frac{y}{x^2+y^2}}\text{and}{\displaystyle \frac{2\pi m}{gq}}\ddot{y}=+{\displaystyle \frac{x}{x^2+y^2}},`$ (28)
or in $`(r,\phi )`$-coordinates:
$`{\displaystyle \frac{2\pi m}{gq}}(\ddot{r}r\dot{\phi }^2)=0\text{and}{\displaystyle \frac{2\pi m}{gq}}{\displaystyle \frac{d}{dt}}(r^2\dot{\phi })=1.`$ (29)
Now, due to the angle-dependent feature of the potential, we notice that the particle’ ‘angular momentum’ is clearly not conserved. As far as we have seen, such a non-conservation imposes an intricate coupling between the coordinates, what implies in serious difficults towards analytical resolution of the differential equations. A typical plot of the motion ($`xy`$-coordinates) of the charged particle is shown in Fig. 1. By virtue of the tangentially repulsive nature of the electric field, the particle is quickly drifted away, despite the signals of the charges.
A further system which deserves more attention is that in which we also have the presence of an external (constant, for concreteness) magnetic field. A realistic planar system may be obtained at very low temperatures (around or less than 1K) and suficiently strong magnetic field (at least 10T) perpendicular to a very thin plate <sup>8</sup><sup>8</sup>8Such systems may be realised, for instance, in the interface between two semi-conductors. Furthermore, since the motion of the charges (electrons, for concreteness) takes place as if the third dimension (perpendicular to the plane of motion) were frosen, the generally employed 2D (spatial) treatment is justified, and has been shown to gives us a very good explanation of the physical phenomena which occur whithin such systems, e.g., the Quantum Hall Effect.. Such a perpendicular field is got by taking a vector potential entirely confined to the 2D-spatial plane, for example, $`\stackrel{}{A}=\stackrel{}{A_1}=B_0x\widehat{j},\stackrel{}{A}=\stackrel{}{A_2}=B_0y\widehat{i}`$ (Landau gauges) or still $`\stackrel{}{A}=\frac{1}{2}\stackrel{}{A_1}+\stackrel{}{A_2}`$ (symmetric gauge).
Now, our present system is composed by the electric charge subject to the external magnetic and to the tangential electric field as well. Again, the classical eqs. of motion are easy to be obtained and read (eqs. of motion in $`r,\phi `$ imediately follow):
$`{\displaystyle \frac{m}{q}}\ddot{x}={\displaystyle \frac{q}{2\pi }}{\displaystyle \frac{y}{x^2+y^2}}+B_0\dot{y}\text{and}{\displaystyle \frac{m}{q}}\ddot{y}=+{\displaystyle \frac{g}{2\pi }}{\displaystyle \frac{x}{x^2+y^2}}B_0\dot{x},`$ (30)
Or, by defining complex dynamical variables as $`\eta =x+iy`$ and $`\eta ^{}=xiy`$, we get:
$$2m(\ddot{\eta }\eta ^{}+\eta \ddot{\eta }^{})+iqB_0(\dot{\eta }\eta ^{}\eta \dot{\eta }^{})=0\text{and}4\pi m(\ddot{\eta }\dot{\eta }^{}+\dot{\eta }\ddot{\eta }^{})+iqg\frac{(\dot{\eta }\eta ^{}\eta \dot{\eta }^{})}{\eta \eta ^{}}=0.$$
Despite their symmetric appearance, the resolution of the eqs. above is not too easy. Indeed, we expect that they may be even more difficult to be solved than those in the absence of magnetic field (previous case).
On the other hand, numerical resolution shows us that the magnetic field tends to compensate the repulsive effect of the electric one so that the (classical) motion of the particle appears to drift in a slower way, describing an almost regular spiral-like pattern (see Fig. 2). Notice also that the distance between two neighbour arms of such a pattern decreases as the radial distance increases: the particle asymptotically ‘approaches’ to perform a closed trajectory (in the next section, we shall see that, the quantum dynamics of the charged particle asymptoticaly, $`r\mathrm{}`$, reduces to that of one central harmonic oscillator).
There is, however, at least one important information which may be analytically obtained: in both cases, $`B_0=0`$ and $`B_00`$, the velocity of the charged particle is bounded by the angle, as below:
$`(\stackrel{}{v})^2(t)={\displaystyle \frac{qg}{m\pi }}\phi (t)+(\stackrel{}{v}_0)^2.`$ (31)
It is worthy noticing that the number of windings of the charge around the origin must be taken into account, i.e., the kinectic energy is determined by the total angle descrided by the charge. \[As a sort of quantum counterpart, we shall see that as $`r\mathrm{}`$ the (angular) energy eigenvalues have to be shifted as $`\phi \phi +2\pi `$ (see next section for details)\].
iv) Preliminary analysis of the quantum charge-monopole system
Next, we shall present a preliminary (non-relativistic) quantum analysis of the system above: one electric charge, $`q`$, moving under the action of the monopole scalar potential and of an external constant magnetic field, $`B_0`$. The Hamiltonian (the pure $`gq`$-system is readily got by setting $`\stackrel{}{A}=0`$),
$$H=\frac{1}{2m}(\stackrel{}{p}q\stackrel{}{A})^2+qV$$
for this system is obtained by taking $`\stackrel{}{A}`$ in a particular gauge (Landau or symmetric), as well as $`V(x,y)=\frac{g}{2\pi }\mathrm{arctan}(y/x)=\frac{g}{2\pi }\mathrm{arg}(\stackrel{}{r})`$. \[Notice that the potential remains invariant under general scale transformation, say: $`xf(x,y)x\text{and}yf(x,y)y,`$ but, the same symmetry is not present in the full Hamiltonian, even for $`f(x,y)=a=\text{constant}`$\].
For the analysis to be presented here, concerning the non-conservation of the angular-momentum and some of its consequences, as well as asymptotic bahaviours of the present system, it will be more convenient to write the Hamiltonian above in polar coordinates, $`r,\phi `$, and $`\stackrel{}{A}`$ in the symmetric gauge, like below:
$`H={\displaystyle \frac{1}{2m}}[p{}_{r}{}^{}{}_{}{}^{2}+{\displaystyle \frac{p_r}{r}}+(qB_0)^2r^2]+{\displaystyle \frac{1}{2m}}{\displaystyle \frac{p_{}^{2}{}_{\phi }{}^{}}{r^2}}+{\displaystyle \frac{qB_0}{2m}}p_\phi {\displaystyle \frac{gq}{2\pi }}\phi ,`$ (32)
with $`r`$ and $`\phi `$ defined as before and $`\stackrel{}{p}=p_r\widehat{e}_r+\frac{p_\phi }{r}\widehat{e}_\phi `$, whence there follows that $`p_ri\mathrm{}\frac{}{r}`$ and $`p_\phi i\mathrm{}\frac{}{\phi }`$.
Now, we notice the first remarkable feature of this Hamiltonian: $`H`$ is explicitly angle-dependent and so non-invariant under rotations; conversely, the angular momentum operator, $`J=p_\phi =i\mathrm{}\frac{}{\phi }`$, is not conserved, $`[J,H]=+i\mathrm{}gq/2\pi 0`$.
Although other angle-dependent Hamiltonians have been studied and shown to be relevant in Physics (see for example ), a remarkable difference between them and the one presented here is that the latter is not separable. Indeed, as far as we have seen, the system appears to present an intricate coupling between its degrees-of-freedom, despite of the coordinates chosen. \[Perhaps, some non-standard tranformation could lead us to such a separation, but could also lead us, on the other hand, to results which were of hard physical interpretation. Such an issue remains to be investigated\].
It is clear, from the Hamiltonian (32) and also from the fundamental commutation relations, $`[r,\phi ]=[p_r,p_\phi ]=0`$ and $`[r,p_r]=[\phi ,p_\phi ]=+i\mathrm{}`$, that the non-separability arises from the non-conservation of the angular momentum, $`[J,H]0`$. Indeed, as it may be easily checked, such an angular sector would be separable if it had the general form $`\frac{1}{r^2}(J^2+aJ+b\phi )`$. So, it is the lack of a $`1/r^2`$-factor in $`J`$ and in $`\phi `$-terms what prevents us from a split of variables.
On the other hand, by facing $`H`$ as being non-separable, the analytical resolution of the eigenvalue problem, $`H|\psi >=E|\psi >`$, appears to be of very hard achievement. \[Actually, the presence of the terms proportional to $`\phi `$ and $`r`$ -or powers of $`r`$\- in $`H`$ prevents us from solving this eigenvalue problem by means of, for example, hypergeometric functions (see, for example Ref. )\].
Therefore, a numerical resolution appears to be a more suitable (and direct) attempt towards solving the problem (results will be communicated as soon as they were obtained). Here, however, we shall deal with some analytical results at asymptotic limits, even though some of them appear to be quite qualitative. We shall mainly discuss the limits $`r0`$ and $`r\mathrm{}`$:
i) $`r0`$: we have seen that near the origin (where the “string” is localised), the charged particle experiences a very strong tangentially repulsive electric field (see previous section for details). Since as $`r0`$ this field blows up, it is expected that $`q`$ can never reach the origin, say, its wave-function must vanish there: $`|\psi (r=0,\phi )>0`$. Such a requirement may be viewed as the counterpart of the Dirac-veto in (3+1)D: a single charge moving under the action of the magnetic monopole field could not cross the string of its associated vector potential. \[In addition, such a requirement will impose (see $`r\mathrm{}`$-limit below) severe restrictions on the asymptotic wave solutions\].
Thus, what remains to be determined is how quickly $`|\psi >`$ vanishes as $`r0`$. Nevertheless, contrary to the $`r\mathrm{}`$-limit, in which the Hamiltonian gets separable (see below), here the variables are not naïvely separated. This arises because $`p_\phi ^2/r^2`$ is one of the leading terms, similarly to the original problem, described by the Hamiltonian (32). \[In this sense numerical techniques could help us in order to get some information about the $`gq`$-system as $`r0`$, say, the form of the wave-functions and eigenvalues\].
ii)$`r\mathrm{}`$: supposing that the canonical momenta remain finite in this limit, we get:
$`H(r,\phi )_r\mathrm{}{\displaystyle \frac{1}{2m}}(p_r^2+q^2B_0^2r^2)+{\displaystyle \frac{qB_0}{2m}}p_\phi {\displaystyle \frac{gq}{2\pi }}\phi ,`$ (33)
in which the variables appear explicitly split, say, $`H_r\mathrm{}=H_r\mathrm{}^r+H_r\mathrm{}^\phi `$. Thus, at this limit, we have that (the limit $`r\mathrm{}`$ is implicit hereafter)
$`(H|\psi (r,\phi )>)=(E_n|\psi (r,\phi )>)(H^r+H^\phi )|R(r)\mathrm{\Phi }(\phi ))>=((E^r+E^\phi )|R\mathrm{\Phi }>),`$ (34)
which leads us to:
$`H^rR_k(r)=E_k^rR_k(r)\text{and}H^\phi \mathrm{\Phi }_l(\phi )=E_l^\phi \mathrm{\Phi }_l(\phi ).`$ (35)
Therefore, as $`r\mathrm{}`$, we get the following set of differential eqs.:
$`\mathrm{}^2{\displaystyle \frac{d^2}{dr^2}}R+(2mE^rq^2B_0^2r^2)R=0,`$ (36)
$`i\mathrm{}{\displaystyle \frac{d}{d\phi }}\mathrm{\Phi }+(ϵ^\phi +\beta \phi )\mathrm{\Phi }=0,`$ (37)
with $`\beta =+mg/\pi B_0`$ and $`ϵ^\phi =2mE^\phi /qB_0`$.
We notice that, at this limit, the radial part of the Hamiltonian reduces to that of one central harmonic oscillator, whose solutions may be written in terms of Hermite polynomials, $`H_n`$:
$$R_k(u)=R_0e^{u^2/2}H_k(u),$$
with $`u=qB_0`$. This implies the well-known eigenvalues $`E_k^r=\mathrm{}\omega _c(k+1/2)`$, where $`\omega _c=qB_0/m`$ is the cyclotron frequency. Now, by virtue of the requirement $`|\psi (r=0,\phi )>0`$, only the $`k=\text{odd}`$ solutions will survive. This implies in a non-vanishing value for the lowest (renormalised) energy level, $`E_k^1=\mathrm{}\omega _c`$.
On the other hand, the angular sector appears to be quite unusual. Indeed, by solving the differential equation in $`\phi `$, we readily obtain
$`\mathrm{\Phi }_l(\phi )=\mathrm{\Phi }_0exp\left[{\displaystyle \frac{i}{\mathrm{}}}\left({\displaystyle \frac{\beta \phi }{2}}+ϵ_l^\phi \right)\phi \right].`$ (38)
It is worth noticing the new $`\phi ^2`$-like phase factor, along with the usual linear one. As a first remark, we should stress that it cannot be removed by any suitable gauge tranformation; indeed, it must rather be faced as a consequence of the $`\phi `$-like scalar potential. Although quite unusual, it leads us to new and interesting results. First, notice that $`\mathrm{\Phi }(\phi )`$ has periodicity $`2\pi (\beta \pi +ϵ_l^\phi )`$. Thus, the requirement that $`\mathrm{\Phi }`$ be single-valued, i.e., continuous, is equivalent to set
$`2\pi (\beta \pi +ϵ_l^\phi )=2\pi l\mathrm{}E_l^\phi =E_0^\phi +{\displaystyle \frac{\omega _c}{2}}l\mathrm{}.`$ (39)
Now, if we identify the parameter $`l`$ as the number of windings $`q`$ gives around $`g`$ (for example, in the counter-clock-wise sense), then $`l`$ shall be taken as a non-negative integer (indeed, the negative values would be associated to the clock-wise sense). Therefore, the eigenvalues associated to the angular variable feel whether it is running between $`0`$ and $`2\pi `$, $`2\pi `$ and $`4\pi `$, and so forth. In other words, whenever $`\phi `$ is shifted, say, by $`2\pi `$, its associated eigenvalues respond to this change by shifting up their values.
However, before completing a winding around $`g`$, $`q`$-particle would have vanishing energy, since $`l=0`$. It is precisely here that $`E_0^\phi =gq/2`$ enters: since $`E_0^\phi `$ is a classical value, the lowest (angular) energy level of $`q`$-charge is non-vanishing. In other words, since the angular potential is acting on $`q`$, it is expected that, by conservation of energy, this charge has non-vanishing (also angular) kinetic energy, as long as it has started its motion. This is the reason why there appears an $`E_0^\phi `$ with an intrinsically classical nature.
The results above may be viewed as a quantum analogue of eq. (31). For example, the fact that $`\stackrel{}{v}^2`$ be given by the full angle (including many windings) is now represented by parameter $`l`$; and the initial kinetic energy, $`\stackrel{}{v}_0^2`$ (which may be classically set to zero) ‘survives’ at quantum level, but acquiring a intrinsic non-vanishing value.
Moreover, we could be tempted to naively apply $`J`$-operator on $`|\psi _{kl}>`$ above, to get
$$J|\psi _{kl}>=(\beta \phi +ϵ_l^\phi )|\psi _{kl}>=(\beta (\phi +\pi )+l)|\psi _{kl}>,$$
and hence, to guess that $`|\psi _{kl}>`$ carry continuous angular momentum. However, this is not a legitimate procedure, because $`|\psi _{kl}>`$ are not eigenvectors of $`J`$ (recall that $`[J,H]0`$). Actually, as far as we have seen, the only two quantities which may be simoutaneously diagonalised in $`|\psi _{kl}>`$-basis are $`H_r`$ and $`H_\phi `$ (the components of the asymptotic Hamiltonian, eq. (33)).
Clearly, the results and remarks above are strictly valid only at the asymptotic limits specified previously. Whether similar scenario does happen at arbitrary distances (as the classical result (31) does), remains to be studied and will be strongly dependent on the separation of variables in the original Hamiltonian, eq. (32).
A naive analysis of the limits discussed above would lead us to conclude that, since the charged particle is repelled from the origin by the $`\phi `$-potential and since as $`r\mathrm{}`$ its dynamics reduces to that of one central harmonic oscillator (whose wave-functions fall off exponentially), it is expected that the system yields physical bound states. Therefore, even though the pure $`gq`$-system does not admit bound states (once that the confining $`r^2`$-type potential is absent, for this case, in eq.(33), we get indeed a radially free particle), when it is supplemented by a suitable external magnetic field, the possibility for such states may be raised.
Nevertheless, when electrons are moving on the plane subject only to a perpendicular magnetic field, then the choice of Landau gauge immediately reduces the quantum problem to that of one harmonic oscillator in one dimension, and a free particle motion in the other direction. In this case, we cannot have bound states.<sup>9</sup><sup>9</sup>9Nevertheless, whenever two species of fermions are combined into a unique four-component spinor, the presence of a constant magnetic field induces flavor symmetry breakdown and fermion condensates appear. Such condensates are, however, quite sensible to thermal effects and disappear at finite temperature . However, when the system is supplemented by an extra, say, scalar potential (as in the present case), it is also well-known that bound states show up, even in the case of repulsive potential. Here, we have just raised such a question, and a precise answer demands further investigation.
v) Conclusion and Prospects
We have shown that classical (2+1)D Maxwell and Maxwell-Chern-Simons Electrodynamics present some interesting novelties as compared to Maxwell theory in (3+1)D, namely, the reverberation of signals and the far-from-trivial question of a Larmor-like formula. As we have seen, such phenomena are intimately related to the failure of the Huyghens’ principle. Namely, the latter is very difficult to be obtained even for constant accelerated motions (parabolic and hyperbolic ones). The integrals involved are highly non-trivial and appear to diverge, so demanding some suitable regularisation scheme. On the other hand, we hope that some hints about such a Larmor’ formula could be obtained with the help of numerical calculations. Next, as a natural extension of our present results, we shall pursue an investigation of the canonical quantisation of the electromagnetic radiation for the models contemplated here .
Concerning the Dirac-like monopole, it also presents some new properties whenever compared to its (3+1)D-counterpart; for instance, its static tangential electric field. Furthermore, acting on a single charged particle, it leads us to interesting classical and quantum results. For example, the $`gq`$-system (with $`B_0`$) has been shown to give rise, at least asymptoticaly and at non-relativistic regimes, to a central harmonic oscillator, with an interesting angular sector which contributes to the energy-eigenvalues.
As future prospects, solutions to the Hamiltonian of eq. (32) in its general form shall be the object of a further investigation . It woul be also of relevance to compute possible effects of this peculiar potential on spin particles, for instance, planar Dirac fermions. Moreover, by virtue of its peculiar scalar potential (and unusual consequences), such a monopole could be relevant to Condensed Matter problems. For instance, by looking at this object as a sort of impurity (scatter) within a sample, could its presence modify the Hall conductivity? And eventually, how would such a modification actually look like?
Acknowledgements
The authors are grateful to Prof. S.A. Dias, Prof. B. Schroer, F. Araruna, H. Belich, J.L. Boldo, R. Casana, G. Cuba Castillo, O. Del Cima, R. Klippert, L. Moraes, A. Nogueira, R. Paunov and R. Rodrigues for useful discussions. They also express their gratitude to Prof. M. Henneaux for a careful reading of an earlier version of this manuscript and for having drawn their attention to the work of Reference. Prof. Ashok Das and Prof. M. Plyushchay are deeply ackowledged for a number of very pertinent comments and for having drawn our attention to some relevant references. Finally, the authors would also like to thank Prof. R. Jackiw for having pointed out the work of Reference. CNPq-Brasil is also acknowledged for the financial support.
|
warning/0004/cond-mat0004007.html
|
ar5iv
|
text
|
# A Pseudogap in the Single-Particle Density of States of a Tomonaga-Luttinger Liquid
The Tomonaga-Luttinger liquid theory predicts a power-law frequency dependence of the single-particle density of states (DOS). In an apparent contradiction with this prediction many quasi-one-dimensional materials (see, for example, ) exhibit a pseudo-gap type structure of DOS.
In this brief report we argue that it is perfectly possible to describe the pseudo-gap behavior in the Tomonaga-Luttinger framework provided one treats carefully the short-time part of the electron Green’s function.
Let us recall how DOS is calculated. The standard non-perturbative technique used to study low-dimensional systems - the bosonization approach yields thermodynamic Green’s functions in the Matsubara time representation $`G(\tau )`$. To extract DOS one has to Fourier transform this function to get $`G(i\omega _n)`$ and then do the analytic continuation $`i\omega _n\omega +i0`$.
As we have said, in the limit $`T0`$, the single-electron Green’s function in the Tomonaga-Luttinger theory has a power-law asymptotics:
$$G(\tau )=\frac{A}{\tau }\frac{1}{|\tau |^\theta }$$
(1)
where $`A`$ is a non-universal number and the exponent $`\theta `$ is determined by the interactions. For the standard spin-1/2 SU(2)-invariant Tomonaga-Luttinger liquid the exponent $`\theta `$ is related to the Luttinger parameter $`K_\mathrm{c}`$ :
$$\theta =\frac{1}{4}(K_\mathrm{c}^1+K_\mathrm{c}2)$$
(2)
Assuming that $`G(\tau )=G(\tau )`$ we can write down the Fourier transform as
$$G(i\omega _n)=2i_0^{\mathrm{}}𝑑\tau \mathrm{sin}(\omega _n\tau )G(\tau )$$
(3)
At small frequencies this integral is dominated by the asymptotics (1) when $`\theta <1`$. Then we have
$$G(i\omega _n)2iA_0^{\mathrm{}}𝑑\tau \frac{\mathrm{sin}(\omega _n\tau )}{\tau ^{1+\theta }}(\omega _n)^\theta $$
(4)
The analytic continuation is quite straightforward in this case and one gets
$`\mathrm{}mG^{(R)}(\omega )=`$ (5)
$`=`$ $`A(\omega )^\theta {\displaystyle \frac{\mathrm{sin}[\pi (1\theta /2)]\mathrm{sin}[\pi (\theta 1)/2]\mathrm{\Gamma }(2\theta )}{\theta (\theta 1)}}.`$ (6)
When $`\theta >1`$ the integral (3) converges at large Matsubara times. However, at $`\theta <2`$ its second derivative with respect to $`\omega _n`$ is still determined by the asymptotics at large $`\tau `$ such that we get
$$G(i\omega _n)=i\omega _ng(0)2(\omega _n)^\theta \frac{\mathrm{sin}[\pi (1\theta /2)]\mathrm{\Gamma }(2\theta )}{\theta (\theta 1)}\mathrm{}$$
(7)
After the analytic continuation only the second term contributes to the imaginary part and we find that Eq. (6) is still valid. Thus the value $`\theta =1`$ corresponding either to $`K_\mathrm{c}=32\sqrt{2}0.17`$ or $`K_\mathrm{c}=3+2\sqrt{2}5.83`$ marks a crossover into a pseudogap phase where $`\rho (\omega )/\omega `$ vanishes at $`\omega 0`$.
Small values of $`K_\mathrm{c}`$ can be achieved in systems with strong retardation effects such as systems with electron-phonon interactions or Kondo lattices . In all these systems small values of $`K_\mathrm{c}`$ are achieved asymptotically at small frequencies. Therefore it is interesting to learn what is the area of validity of the universal power law behavior (6). Below we calculate the DOS for a model with electron-phonon interactions.
The influence of phonons on the electron subsystem in quasi-one-dimensional metals have been studied by different authors; mostly for the case of noninteracting electrons (see , and references therein). The combined effects of the Coulomb and electron- phonon interactions was studied in , using the renormalization group approach. In these effects have been studied in the framework of Tomonaga-Luttinger theory.
Here we briefly repeat the derivation given in . The lattice effects can be included in the Hubbard Hamiltonian by making the hopping integral $`t`$ dependent on the intersite distance:
$`t_{\mathrm{ij}}t+{\displaystyle \frac{1}{2a}}\kappa (u_\mathrm{i}u_\mathrm{j}).`$
(It has been proposed for the first time by Su, Schriefer and Heeger to describe the essential physics of conducting polymers ). Then the Hamiltonian takes the form:
$`H`$ $`=t{\displaystyle \underset{\mathrm{j},\sigma }{}}\left(c_{\mathrm{j}+1,\sigma }^+c_{\mathrm{j},\sigma }+c_{\mathrm{j},\sigma }^+c_{\mathrm{j}+1,\sigma }\right)+U{\displaystyle \underset{\mathrm{j}}{}}n_\mathrm{j}n_\mathrm{j}`$ (8)
$``$ $`{\displaystyle \frac{1}{2a}}\kappa {\displaystyle \underset{\mathrm{j},\sigma }{}}\left(u_\mathrm{j}u_{\mathrm{j}+1}\right)\left(c_{\mathrm{j}+1,\sigma }^+c_{\mathrm{j},\sigma }+c_{\mathrm{j},\sigma }^+c_{\mathrm{j}+1,\sigma }\right)+H_{\mathrm{ph}},`$ (9)
where $`u_\mathrm{j}`$ is dimensionless and $`\kappa `$ has dimensions of energy. The $`c_{j,\sigma }`$ operators are the usual creation and annihilation operators for the electrons with spin $`\sigma `$ in the Wannier orbitals at site $`j`$ and $`n_{\mathrm{j},\sigma }`$ is the number of electrons. U is the repulsion of two electrons on the same site.
In the case of an incommensurate band filling ($`4k_\mathrm{F}2\pi /a`$) in the continuous approximation the electron-phonon part of this Hamiltonian generates a coupling between the lattice deformations and the $`2k_\mathrm{F}`$ and $`4k_\mathrm{F}`$ components of the charge density. The electron-phonon interaction contributes to an effectively retarded interaction between the electronic densities:
$$S_{\mathrm{int}}=d\tau d\tau {}_{}{}^{}\mathrm{d}x\underset{l=1,2}{}\rho (2lk_\mathrm{F},x)D_l(\tau \tau ^{})\rho (2lk_\mathrm{F},x),$$
(10)
where $`D_l(\tau )`$ is the phonons Green’s function at $`q=2lk_\mathrm{F}`$ . Thus the main contribution to the interaction comes from phonons with large frequency. This interaction effects both the spin and the charge sector. The phonons give a positive contribution to the current-current coupling constant in the spin sector (i.e. $`g_\mathrm{s}^{(0)}g_\mathrm{s}`$). In this sector the electron-phonon interaction competes with repulsive forces responsible for $`g_0`$. We assume that the renormalized coupling constant is still repulsive such that there is no spin gap.
In the charge sector the phonons influence the dynamics as well as the scaling dimensions, renormalizing the charge velocity $`v_\mathrm{c}`$ and $`K_\mathrm{c}`$. In the limit of small frequencies $`|\omega |\omega _\mathrm{l}`$ their renormalized values are:
$$K_\mathrm{c}=\left(\frac{m}{m^{}}\right)^{1/2}K_\mathrm{c}^0,\stackrel{~}{v}_\mathrm{c}=\left(\frac{m}{m^{}}\right)^{1/2}v_\mathrm{c}^0.$$
(11)
where $`m^{}`$ is interpreted as renormalized electrons mass with $`m`$ being the bare mass.
Since we are interested not only in the long time asymptotics of the Green’s function, but in its intermediate time behavior, we shall model these behavior adopting the modified Gaussian model with the time-dependent Luttinger parameter:
$$S=\frac{1}{2}\underset{\omega ,q}{}\mathrm{\Phi }_\mathrm{c}(\omega ,q)\left[\frac{1}{v_\mathrm{c}}\omega ^2f(i\omega )+v_\mathrm{c}q^2\right]\mathrm{\Phi }_\mathrm{c}(\omega ,q),$$
(12)
where the function $`f(\omega )`$ takes values between $`f(0)=m^{}/mK^2`$ and $`f(\mathrm{})=1`$. In this brief report we suggest a semi-phenomenological form for the function $`f(\omega )`$:
$$f(\omega )=1+\frac{\omega _0^2}{\omega ^2+\omega _0^2}(K^21),$$
(13)
The above model yields the following Matsubara time single-electron Green’s function
$`G(\tau )`$ $`=`$ $`{\displaystyle \frac{1}{\tau }}\mathrm{exp}\{{\displaystyle _0^{ϵ_\mathrm{F}/\omega _0}}{\displaystyle \frac{\mathrm{d}x}{x}}[\left({\displaystyle \frac{x^2+1}{x^2+K^2}}\right)^{1/4}`$ (14)
$``$ $`\left({\displaystyle \frac{x^2+K^2}{x^2+1}}\right)^{1/4}]^2\mathrm{sin}^2(\omega _0\tau x/2)\}.`$ (15)
where $`ϵ_\mathrm{F}`$ is the ultraviolet cut-off (recall that we consider the Green’s function at coinciding spatial points). Pictures of the single particle density of states are represented on Figure 1. They are obtained by analytic continuation of $`G(i\omega _\mathrm{n})`$ from the imaginary axis to just above the real axis.
The figures clearly show the crossover from the Luttinger-liquid type behavior with singular $`d\rho /d\omega `$ at $`K>0.17`$ to the pseudogap behavior at $`K<0.17`$. Another remarkable feature of these DOS is the peak at $`\omega K^1\omega _0/2`$ (see Fig. 2). Since the DOS is invariant under $`KK^1`$ this empirical formula can be generalized as
$$\omega _{\mathrm{peak}}=\frac{1}{2}(K^1+K)\omega _0.$$
(16)
It is interesting that the peak always occurs at frequencies larger than the characteristic phonon frequency $`\omega _0`$. At small (large) values of $`K`$ this discrepancy can be quite substantial. For example, a peak in DOS has been observed in K<sub>0.3</sub>MoO<sub>3</sub> at $`\omega 300`$ meV. The behavior of DOS at smaller frequencies is almost linear in $`\omega `$ which suggests $`K0.15`$. Then Eq. (16) gives a reasonable estimate for the phonon frequency: $`\omega _090`$ meV.
E.P. is grateful to Steve Allen and particularly to Dave Allen for helpful discussions.
|
warning/0004/cond-mat0004451.html
|
ar5iv
|
text
|
# On Fermi systems with strong forward scattering in 𝑑 spatial dimensions
## I Introduction
The question with regard to the existence of non-Fermi liquid (NFL) metallic states in spatial dimensions $`d`$ greater than one, specifically in $`d=2`$, has been subject of intensive investigations in the course of the past several years, following the observation that the normal-state properties of doped cuprate superconducting compounds markedly differ from those of conventional metals which are accurately described within the framework of the Fermi-liquid (FL) theory (Anderson 1988,1989,1990a,b,1997, Varma, Littlewood, Schmitt-Rink, Abrahams and Ruckenstein 1989, 1990, Littlewood and Varma 1991). Since according to a celebrated theorem due to Luttinger (1961), validity of the many-body perturbation theory to all orders would imply that metallic states of fermions in spatial dimensions greater than one would be FLs, <sup>*</sup><sup>*</sup>* Strictly speaking, this theorem has direct bearing on cases where fermions interact through a short-range two-body potential. failure of the FL picture in the case of the doped cuprates has been considered to signal breakdown of the many-body perturbation theory as applied to these materials. Indeed, in the course of the recent years the question with regard to breakdown of the many-body perturbation theory has been a central aspect in debates concerning the above-mentioned failure of the FL phenomenology in describing the normal-state properties of the doped cuprate compounds (Anderson 1990b, 1993, Engelbrecht and Randeria 1990, Stamp 1993) (see also Engelbrecht and Randeria (1991), Anderson (1991)). Consequently, in this period a number of non-perturbative methods have been put forward and applied to models of interacting fermions with the aim of determining whether or not NFL states are viable in $`d>1`$. Elsewhere (Farid 1999a,b) we have presented accounts of these developments and therefore refrain from repetition here. For our present considerations it is relevant to mention that the existence of NFL metallic states in realistic microscopic models of interacting fermions in $`d>1`$ has remained elusive to this date.
In view of the above-indicated apparent discrepancy that prevails between theoretical predictions and experimental observations, we have undertaken to subject a number of aspects that are crucial to establishing feasibility or otherwise of NFL metallic states in $`d>1`$ to an independent investigation. The present work is the third report of our investigations on this subject. Our first work (Farid 1999a) concerned a critical analysis of the above-indicated Luttinger (1961) theorem. We have demonstrated (Farid 1999a) that Luttinger’s (1961) proof of this theorem involves an implicit assumption which is strictly specific to FL metallic states, implying that the Luttinger (1961) theorem far from excluding the possibility of the existence of NFL metallic states in $`d>1`$ (for cases where the many-body perturbation theory is valid to all orders), in fact amounts to a statement of consistency of the FL states with the principles of the theory of interacting fermions as expressed in terms of the many-body perturbation theory.
Our second work (Farid 1999b) concerned an in-depth analysis of the non-perturbative bosonisation scheme due to Haldane (1992) for metallic states in $`d1`$ as applied to a model of interacting fermions in $`d=2`$, originally put forward by Houghton and Marston (1993) and subsequently extensively studied by Houghton, Kwon and Marston (1994), Houghton, Kwon, Marston and Shankar (1994) and Kwon, Houghton and Marston (1995). We have demonstrated failure of the Haldane (1992) bosonisation scheme in $`d>1`$ and established that this scheme invariably predicts the metallic state of the model under consideration to be an unconventional FL, irrespective of the nature (and strength, provided it be non-vanishing) of the fermion-fermion interaction function, that is whether this be short-range, long-range Coulomb or super-long range. The latter result is in particular at variance with an earlier finding by Bares and Wen (1993), It is also at variance with the findings by Houghton, Kwon, Marston and Shankar (1994) and Kwon, Houghton and Marston (1995). We have however shown (Farid 1999b) that this discrepancy is due to use of certain invalid approximations by the latter authors. indicating the metallic state of the system for super-long-range interactions to be a NFL. We have traced back the shortcoming of the Haldane (1992) bosonisation scheme in $`d>1`$ to the neglect of some terms in the commutation relation concerning some current operators; in particular, in the case of metallic systems with a sufficiently small (or vanishing) value for the jump $`Z_{𝐤_F}`$ in the momentum distribution function at the Fermi momentum, the neglected terms are more relevant than the sole term that is taken into account, namely the quantum anomaly. For completeness, we mention that by solely taking into account the quantum anomaly, the mentioned current operators form a Kac-Moody algebra (see, e.g., Goddard and Olive 1986). We have also shown that under all conditions the Fermi energy of the interacting system within the Haldane (1992) bosonisation scheme coincides with that of the non-interacting system. This unambiguously implies that even for weakly-interacting metallic states, for which $`Z_{𝐤_F}`$ is close to unity, neglect of the indicated terms in the mentioned commutation relation for current operators is not justified.
The exactly solvable one-dimensional Luttinger model (Luttinger 1963, Mattis and Lieb 1965) has been dealt with by three distinct methods. One is based on the technique of bosonisation, of which the Haldane (1992) scheme is a generalisation to higher dimensions. <sup>§</sup><sup>§</sup>§ The bosonisation method for calculation of correlation functions is due to Luther (1979) which derives from works by Luther and Peschel (1974), Mattis and Lieb (1965), Tomonaga (1950) and Bloch (1933, 1934). Parallel developments involving similar views as in the indicated works are due to Jordan (1935, 1936a,b), Krönig (1935a,b,c), Skyrme (1958, 1959, 1961a,b) and Coleman (1975). For reviews see Stone (1994) and Kopietz (1997). The second, due to Everts and Schulz (1974), is based on the solution of the equation of motion for the Green function which derives from an exact decoupling of the hierarchy of the Green functions. The third approach, due to Dzyaloshinskiǐ and Larkin (DL) (1974), For a detailed exposition of this approach see (Bohr 1981). exploits the conservation laws and the associated Ward identities. This scheme has been generalised to $`d>1`$ and reported in a series of articles by Di Castro and Metzner (1991), Metzner and Di Castro (1993), Castellani, Di Castro and Metzner (1994), Castellani and Di Castro (1994) and extensively reviewed by Metzner, Castellani and Di Castro (MCDC) (1998). It is this scheme that we subject to a critical analysis in the present work. In our considerations that follow, we shall almost exclusively refer to the last review article without thereby implying that the cited subject matters would have their origin in this work.
Contrary to case of $`d=1`$, associated with linear energy dispersions for the non-interacting left- and right-movers (that is the case of the one-dimensional Luttinger (1963) model), in cases of $`d>1`$ the generalised DL approach is not exact; MCDC (1998) assert, however, that for Fermi systems the generalised approach is asymptotically exact in the limit of strong forward scattering, that is in the limit of zero momentum transfer $`𝐪`$ for fermions in a narrow band of width $`\lambda `$ ($`\mathrm{\Lambda }`$ in the authors’ notation) encompassing the Fermi surface. In this work we demonstrate this not to be the case. We further support this finding by referring to the similarity between the expressions for the single-particle Green function within the framework of the generalised DL (MCDC 1998) scheme and the Haldane (1992) bosonisation approach and the fact that, as we have indicated above, the latter is not adequate for $`d>1`$ (Farid 1999b). We also establish that the rigorous formalism for systems of interacting fermions, even in the limit of strong forward scattering, is not amenable to treatment by the Wilson-Fisher (1972) (Wilson 1973) technique of dimensional continuation.
The organisation of this work is as follows. In § II we present an outline of the generalised DL formalism (MCDC 1998) which will facilitate our discussions in this paper. Anticipating the results that we present in § III, here we indicate that the conventional process of dimensional continuation is not applicable to a general formalism concerning systems of interacting fermions, not even in the limit where solely forward scattering processes are taken into account. In § III we analyse two crucial steps that are involved in the derivation of the generalised DL formalism and explicitly demonstrate their inadequacies in $`d>1`$; in § III.A we deal with a relationship between the charge and the current part of a vertex function that has been held to be asymptotically correct in the limit of strong forward-scattering and show that its validity in $`d>1`$ is strictly limited to the trivial case of non-interacting fermions; in § III.B we show the invalidity of a second relationship between two irreducible vertex functions, also held to be asymptotically correct in $`d>1`$ in the limit of strong forward-scattering. In § IV we explicitly demonstrate how the singular behaviour of the momentum distribution function, in particular that pertaining to the ground-state of the system, at the Fermi momentum, renders the mathematical manipulations that are central to the derivation of the generalised DL method in $`d>1`$ (MCDC 1998) inapplicable. Here we further show how similar manipulations, combined with a complete disregard of the errors associated with these, in $`d=1`$ (specifically in dealing with the one-dimensional Luttinger model) give rise to results which coincide with the exact results, thus misleadingly suggesting their general applicability even in $`d>1`$. We finally present a summary and concluding remarks of this work in § V. We devote an Appendix which follows the main text to a brief exposition of the Wilson-Fisher (Wilson and Fisher 1972, Wilson 1973) technique of dimensional continuation, which concerns continuation of the number of the spatial dimensions from integer to real, as well as complex, values. From the details in this Appendix it becomes apparent how application of this technique is hampered in the context of the many-body theory of interacting fermions. Here we propose an alternative scheme of dimensional continuation which, although not unique, is not restrictive with regard to the functional dependence of the integrands on the variable of integration.
## II Outline of the generalised Dzyaloshinskiǐ-Larkin scheme
Here we present a brief outline of the generalised DL scheme. For a detailed exposition of this scheme readers are referred to the article by MCDC (1998). In this Section we adopt the notational conventions of the latter work. For completeness, below $`p`$ and $`q`$ denote $`d+1`$-vectors $`(ip_0,𝐩)`$ and $`(iq_0,𝐪)`$ respectively, with $`p_0`$ ($`q_0`$) energy (throughout we assume $`\mathrm{}=1`$) and $`𝐩`$ ($`𝐪`$) momentum in $`d`$-dimensional momentum space (here $`d`$ is integer; we shall be explicit when $`d`$ is non-integer); $`(ip_0,𝐩)_\nu `$ represents $`ip_0`$ when $`\nu =0`$ and $`p_\nu `$ when $`\nu =1,2,\mathrm{},d`$; $`_kf(k)`$, with $`f(k)F(ik_0;𝐤)`$, is the short for $`(2\pi )^1dk_0(2\pi )^d\mathrm{d}^dkF(ik_0;𝐤)`$. For finite temperatures $`T`$, $`(2\pi )^1dk_0(\mathrm{})`$ symbolises a Matsubara sum, $`(2\pi )^1dk_0(\mathrm{})T_{k_0}(\mathrm{})`$, where the summation runs over the discrete fermion Matsubara energies $`2\pi i(k_0+1/2)T`$, with $`k_0=0,\pm 1,\mathrm{}`$. Below, unless specifically indicated otherwise, we assume the summation convention $`A_\nu B^\nu _{\nu =0}^dA_\nu B^\nu `$.
For a system whose Hamiltonian $`\widehat{H}`$ is of the form
$$\widehat{H}\widehat{H}_0+\widehat{H}_I:=\underset{𝐤\sigma }{}\epsilon _𝐤^0\widehat{a}_{𝐤,\sigma }^{}\widehat{a}_{𝐤,\sigma }+\frac{1}{2V}\underset{𝐤,𝐤^{},𝐪}{}\underset{\sigma \sigma ^{}}{}g_{𝐤𝐤^{}}^{\sigma \sigma ^{}}(𝐪)\widehat{a}_{𝐤𝐪/2,\sigma }^{}\widehat{a}_{𝐤+𝐪/2,\sigma }\widehat{a}_{𝐤^{}+𝐪/2,\sigma ^{}}^{}\widehat{a}_{𝐤^{}𝐪/2,\sigma ^{}},$$
(1)
with $`\widehat{a}_{𝐤,\sigma }^{}`$, $`\widehat{a}_{𝐤,\sigma }`$ the canonical creation and annihilation operators for fermions with spin $`\sigma `$, respectively, $`\epsilon _𝐤^0`$ the energy dispersion of the non-interacting fermions, We assume $`\epsilon _𝐤^0𝐤^2/2`$ (in units where $`m_f`$, the fermion mass, is equal to unity). For some relevant remarks see Footnote \*‡III A further on. $`V`$ volume of the system and $`g_{𝐤𝐤^{}}^{\sigma \sigma ^{}}(𝐪)`$ the fermion-fermion interaction function, MCDC (1998) attempt to transform the expression for the self-energy $`\mathrm{\Sigma }_\sigma (p)`$,
$$\mathrm{\Sigma }_\sigma (p)=\underset{\sigma ^{}}{}_p^{}D_{𝐩𝐩^{}}^{\sigma \sigma ^{}}(pp^{})G_\sigma (p^{})\mathrm{\Lambda }_{\sigma ^{}\sigma }^0([p+p^{}]/2;p^{}p),$$
(2)
into a closed form. Here $`D_{𝐩𝐩^{}}^{\sigma \sigma ^{}}(pp^{})`$ is the screened interaction function (see Eq. (17) below), $`G_\sigma (k)G_\sigma (ik_0;𝐤)`$ the (thermal) single-particle Green function,
$$\mathrm{\Lambda }_{\sigma ^{}\sigma }^0(p;q):=\widehat{n}_\sigma ^{}(q)\widehat{a}_{pq/2,\sigma }\widehat{a}_{p+q/2,\sigma }^{}_{\mathrm{tr}}^{\mathrm{irr}}$$
(3)
the irreducible density (or charge) vertex, in which <sup>\**</sup><sup>\**</sup>\** Note that $`_\sigma \widehat{n}_\sigma (q)\widehat{j}_0^0(q)\widehat{j}^0(q)`$, where $`\widehat{j}_0^0(q)`$ and $`\widehat{j}^0(q)`$ stand for the density (or charge) parts of $`\widehat{j}_0^\nu (q)`$ and $`\widehat{j}^\nu (q)`$, respectively (see Eqs. (18) and (47) below).
$$\widehat{n}_\sigma (𝐪):=\underset{𝐤}{}\widehat{a}_{𝐤𝐪/2,\sigma }^{}\widehat{a}_{𝐤+𝐪/2,\sigma }.$$
(4)
In Eq. (3) “tr” indicates that the external Green functions are not taken into account (i.e. they are truncated), which amounts to dividing the expression associated with the (thermal) average $`\mathrm{}^{\mathrm{irr}}`$, without subscript “tr”, by $`G_\sigma (p+q/2)G_\sigma (pq/2)`$; the superscript “irr” indicates that in the diagrammatic expansion of the associated correlation function only one-interaction-irreducible diagrams are taken into account, i.e. those which do not fall into two disjoint parts upon removing an interaction line in them. Through reliance upon the generalised closed-loop theorem (CLT) (Kopietz, Hermisson and Schönhammer 1995), <sup>††</sup><sup>††</sup>†† The CLT (or, loop-cancellation theorem) for the one-dimensional Luttinger (1963) model is due to Dzyaloshinskiǐ and Larkin (1974). For an extensive exposition of this theorem see Bohr (1981). MCDC (1998) relate $`\mathrm{\Lambda }_{\sigma ^{}\sigma }^\nu (p;q)`$ with the irreducible vertex part associated with the non-interacting density-current $`\widehat{j}_0^\nu (q)`$, that is
$$\mathrm{\Lambda }_\sigma ^\nu (p;q):=\widehat{j}_0^\nu (q)\widehat{a}_{pq/2,\sigma }\widehat{a}_{p+q/2,\sigma }^{}_{\mathrm{tr}}^{\mathrm{irr}},$$
(5)
as follows:
$$\mathrm{\Lambda }_{\sigma ^{}\sigma }^0(p;q)=\delta _{\sigma ^{},\sigma }\mathrm{\Lambda }_\sigma ^0(p;q),𝚲_{\sigma ^{}\sigma }(p;q)=\delta _{\sigma ^{},\sigma }\mathrm{\Lambda }_\sigma ^0(p;q)𝐯_𝐩^0.$$
(6)
Here the $`d`$-dimensional vector $`𝚲_{\sigma ^{}\sigma }(p;q)`$ corresponds to the $`\nu =1,2,\mathrm{},d`$ components of $`\mathrm{\Lambda }_{\sigma ^{}\sigma }^\nu (p;q)`$. The expressions in Eq. (6) can be written as
$$\mathrm{\Lambda }_{\sigma ^{}\sigma }^0(p;q)=\delta _{\sigma ^{},\sigma }\frac{(iq_0,𝐪)_\nu \mathrm{\Lambda }_\sigma ^\nu (p;q)}{iq_0𝐪𝐯_𝐩^0}.$$
(7)
The density-current vertex function $`\mathrm{\Lambda }_\sigma ^\nu (p;q)`$ is related to the vertex part
$$\mathrm{\Gamma }_\sigma ^\nu (p;q):=\widehat{j}^\nu (q)\widehat{a}_{pq/2,\sigma }\widehat{a}_{p+q/2,\sigma }^{}_{\mathrm{tr}}$$
(8)
and the irreducible vertex part corresponding to the interacting density-current $`\widehat{j}^\nu (q)`$, that is
$$\mathrm{\Lambda }_\sigma ^{}_{}{}^{}\nu (p;q):=\widehat{j}^\nu (q)\widehat{a}_{pq/2,\sigma }\widehat{a}_{p+q/2,\sigma }^{}_{\mathrm{tr}}^{\mathrm{irr}},$$
(9)
through the Dyson equation
$$\mathrm{\Gamma }_\sigma ^\nu (p;q)=\mathrm{\Lambda }_\sigma ^{}_{}{}^{}\nu (p;q)+\frac{1}{V}\underset{𝐤^{}\sigma ^{}}{}\underset{𝐤^{\prime \prime }\sigma ^{\prime \prime }}{}J^{\nu ,𝐤^{}\sigma ^{}}(q)g_{𝐤^{}𝐤^{\prime \prime }}^{\sigma ^{}\sigma ^{\prime \prime }}(q)\mathrm{\Lambda }_\sigma ^{𝐤^{\prime \prime }\sigma ^{\prime \prime }}(p;q),$$
(10)
where
$$J^{\nu ,𝐤\sigma }(q):=\frac{\mathrm{d}k_0}{2\pi }\widehat{j}^\nu (q)\widehat{a}_{k+q/2,\sigma }^{}\widehat{a}_{kq/2,\sigma }\frac{\mathrm{d}k_0}{2\pi }\mathrm{\Gamma }_\sigma ^\nu (k;q)G_\sigma (k+q/2)G_\sigma (kq/2),$$
(11)
$$\mathrm{\Lambda }_\sigma ^{𝐤\sigma ^{}}(p;q):=\frac{\mathrm{d}k_0}{2\pi }\widehat{a}_{kq/2,\sigma ^{}}^{}\widehat{a}_{k+q/2,\sigma ^{}}\widehat{a}_{pq/2,\sigma }\widehat{a}_{p+q/2,\sigma }^{}_{\mathrm{tr}}^{\mathrm{irr}}.$$
(12)
Owing to the exact Ward identity
$$(iq_0,𝐪)_\nu \mathrm{\Gamma }_\sigma ^\nu (k;q)=G_\sigma ^1(k+q/2)G_\sigma ^1(kq/2),$$
(13)
from Eq. (11) it readily follows that
$$(iq_0,𝐪)_\nu J^{\nu ,𝐤\sigma }(q)=𝗇_\sigma (𝐤𝐪/2)𝗇_\sigma (𝐤+𝐪/2),$$
(14)
where
$$𝗇_\sigma (𝐤\pm 𝐪/2):=\widehat{a}_{𝐤\pm 𝐪/2,\sigma }^{}\widehat{a}_{𝐤\pm 𝐪/2,\sigma }\frac{\mathrm{d}k_0}{2\pi }G_\sigma (k\pm q/2)$$
(15)
stands for the momentum distribution function corresponding to the spin-$`\sigma `$ fermions. Through relying upon (see however §§ III and IV) $`𝗇_\sigma (𝐤\pm 𝐪/2)=𝗇_\sigma (𝐤)\pm \frac{1}{2}𝐪\mathbf{}_𝐤𝗇_\sigma (𝐤)+𝒪(𝐪^2)`$, for small $`𝐪`$, MCDC (1998) deduce from Eq. (14) (c.f. Eq. (75) below)
$$(iq_0,𝐪)_\nu J^{\nu ,𝐤\sigma }(q)=𝐪\mathbf{}_𝐤𝗇_\sigma (𝐤)+𝒪(𝐪^3).$$
(16)
The screened interaction $`D_{𝐤𝐤^{}}^{\sigma \sigma ^{}}(q)`$ in Eq. (2) is that within the random-phase approximation (RPA),
$$D_{𝐤𝐤^{}}^{\sigma \sigma ^{}}(q)=g_{𝐤𝐤^{}}^{\sigma \sigma ^{}}(𝐪)+\underset{\sigma ^{\prime \prime }}{}_{k^{\prime \prime }}g_{𝐤𝐤^{\prime \prime }}^{\sigma \sigma ^{\prime \prime }}(𝐪)G(k^{\prime \prime }q/2)G(k^{\prime \prime }+q/2)D_{𝐤^{\prime \prime }𝐤^{}}^{\sigma ^{\prime \prime }\sigma ^{}}(q).$$
(17)
In practice, the interacting ‘bubble’ $`G(k^{\prime \prime }q/2)G(k^{\prime \prime }+q/2)`$ is replaced by the non-interacting one (MCDC 1998), i.e. $`G_0(k^{\prime \prime }q/2)G_0(k^{\prime \prime }+q/2)`$; this on account of the generalised CLT, under the assumption that $`𝐤^{\prime \prime }`$ is confined to a small neighbourhood of $`𝐤_F`$. For a comment on the generalised CLT in $`d>1`$ see (Farid 1999b).
From the equation of continuity $`[\widehat{n}(𝐪),\widehat{H}]_{}=𝐪\widehat{𝐣}(𝐪)`$, where
$$\widehat{n}(𝐪):=\underset{\sigma }{}\widehat{n}_\sigma (𝐪)$$
(18)
(see Eq. (4) above), one can obtain the conserved current $`\widehat{𝐣}(𝐪)`$ for the system governed by the Hamiltonian $`\widehat{H}`$ in Eq. (1). The conserved current $`\widehat{𝐣}_0(𝐪)`$ corresponding to $`\widehat{H}_0`$ in Eq. (1) is obtained through application of the continuity equation $`[\widehat{n}(𝐪),\widehat{H}_0]_{}=𝐪\widehat{𝐣}_0(𝐪)`$, <sup>‡‡</sup><sup>‡‡</sup>‡‡ Explicit calculation yields $`[\widehat{n}(𝐪),\widehat{H}_0]_{}=_{𝐤\sigma }\left(\epsilon _{𝐤+𝐪/2}^0\epsilon _{𝐤𝐪/2}^0\right)\widehat{a}_{𝐤𝐪/2,\sigma }^{}\widehat{a}_{𝐤+𝐪/2,\sigma }`$. With $`\epsilon _𝐤^0=𝐤^2/2`$ (see Footnote II above and Footnote \*‡III A below), one has $`\epsilon _{𝐤+𝐪/2}^0\epsilon _{𝐤𝐪/2}^0=𝐪𝐤𝐪𝐯_𝐤^0`$.
$$\widehat{𝐣}_0(𝐪)=\underset{𝐤\sigma }{}𝐯_𝐤^0\widehat{a}_{𝐤𝐪/2,\sigma }^{}\widehat{a}_{𝐤+𝐪/2,\sigma },$$
(19)
where $`𝐯_𝐤^0:=\mathbf{}_𝐤\epsilon _𝐤^0𝐤`$. The commutator $`[\widehat{n}(𝐪),\widehat{H}_I]_{}`$ gives rise to ‘non-diagonal’ quartic contributions (involving four different momenta) to the current operator $`\widehat{𝐣}_I(𝐪):=\widehat{𝐣}(𝐪)\widehat{𝐣}_0(𝐪)`$, which are irrelevant in the low-energy limit (from the renormalisation-group perspective); taking the ‘diagonal’ contributions, corresponding to the momentum transfer $`𝐪^{}`$ equal to either $`𝐪`$ or $`𝐪`$, the following (approximate) expression is obtained (MCDC 1998)
$`[\widehat{n}(𝐪),\widehat{H}_I]_{}`$ $`=`$ $`{\displaystyle \frac{1}{2V}}{\displaystyle \underset{𝐤\sigma }{}}{\displaystyle \underset{𝐤^{}\sigma ^{}}{}}g_{𝐤𝐤^{}}^{\sigma \sigma ^{}}(𝐪)\left[\widehat{𝗇}_\sigma (𝐤𝐪/2)\widehat{𝗇}_\sigma (𝐤+𝐪/2)\right]\widehat{a}_{𝐤^{}𝐪/2,\sigma ^{}}^{}\widehat{a}_{𝐤^{}+𝐪/2,\sigma ^{}}`$ (20)
$`+`$ $`{\displaystyle \frac{1}{2V}}{\displaystyle \underset{𝐤\sigma }{}}{\displaystyle \underset{𝐤^{}\sigma ^{}}{}}g_{𝐤𝐤^{}}^{\sigma \sigma ^{}}(𝐪)\widehat{a}_{𝐤𝐪/2,\sigma }^{}\widehat{a}_{𝐤+𝐪/2,\sigma }\left[\widehat{𝗇}_\sigma ^{}(𝐤^{}𝐪/2)\widehat{𝗇}_\sigma ^{}(𝐤^{}+𝐪/2)\right].`$ (21)
MCDC (1998) cast the right-hand side (RHS) of this expression into the form $`𝐪\widehat{𝐣}_I(𝐪)`$, with
$$\widehat{𝐣}_I(𝐪)=\underset{𝐤\sigma }{}\widehat{𝐯}_{𝐤\sigma }^I(𝐪)\widehat{a}_{𝐤𝐪/2,\sigma }^{}\widehat{a}_{𝐤+𝐪/2,\sigma },$$
(22)
where
$$\widehat{𝐯}_{𝐤\sigma }^I(𝐪):=\frac{1}{V}\underset{𝐤^{}\sigma ^{}}{}g_{𝐤𝐤^{}}^{\sigma \sigma ^{}}(𝐪)\mathbf{}_𝐤^{}\widehat{𝗇}_\sigma ^{}(𝐤^{}).$$
(23)
In obtaining the result in Eq. (22), MCDC (1998) employ $`\widehat{𝗇}_\sigma (𝐤𝐪/2)\widehat{𝗇}_\sigma (𝐤+𝐪/2)=𝐪[\mathbf{}_𝐤\widehat{𝗇}_\sigma (𝐤)]+𝒪(𝐪^3)`$ (see §§ III.B and IV however; also compare with Eq. (75) below), and the same for $`\widehat{𝗇}_\sigma ^{}(𝐤^{}𝐪/2)\widehat{𝗇}_\sigma ^{}(𝐤^{}+𝐪/2)`$. On account of the fact that for the ground and low-lying excited states of the system $`[\mathbf{}_𝐤\widehat{𝗇}_\sigma (𝐤)]`$ is most significant for $`𝐤=𝐤_F`$ (the same concerning $`[\mathbf{}_𝐤^{}\widehat{𝗇}_\sigma ^{}(𝐤^{})]`$ for $`𝐤^{}=𝐤_F^{}`$), MCDC (1998) make use of the following relationship concerning the radial part of the $`𝐤^{}`$ integration (that is that in the direction normal to the Fermi surface) in the expression for $`\widehat{𝐯}_{𝐤\sigma }^I(𝐪)`$ in Eq. (23), <sup>\**</sup><sup>\**</sup>\** In the thermodynamic limit one has $`V^1_𝐤^{}(\mathrm{})(2\pi )^d\mathrm{d}^d𝐤^{}(\mathrm{})`$. For $`d`$-dimensional integrations see Appendix.
$`{\displaystyle dk_r^{}k_{r}^{}{}_{}{}^{d1}g_{𝐤𝐤^{}}^{\sigma \sigma ^{}}(𝐪)\mathbf{}_𝐤^{}\widehat{𝗇}_\sigma ^{}(𝐤^{})}`$ $``$ $`𝐤_F^{}^{d1}g_{𝐤𝐤_F^{}}^{\sigma \sigma ^{}}(𝐪){\displaystyle dk_r^{}\mathbf{}_𝐤^{}\widehat{𝗇}_\sigma ^{}(𝐤^{})}`$ (24)
$`=`$ $`𝐤_F^{}^{d1}g_{𝐤𝐤_F^{}}^{\sigma \sigma ^{}}(𝐪)𝒏_{𝐤_F^{}},`$ (25)
where $`𝒏_{𝐤_F^{}}`$ denotes the outward unit vector normal to the Fermi surface at $`𝐤_F^{}`$. Here $`d`$ is not necessarily integer (see Appendix). In arriving at the last expression on the RHS of Eq. (24), MCDC (1998) have made use of the fact that for the ground and low-lying excited states of the system, $`\widehat{𝗇}_\sigma ^{}(𝐤^{})`$ yields unity for $`𝐤^{}`$ deep inside and zero for $`𝐤^{}`$ far outside the Fermi sea (that is for $`𝐤^{}=𝐤_i^{}`$ and $`𝐤^{}=𝐤_o^{}`$, respectively). MCDC (1998) observe that the RHS of Eq. (24) (which is a $`c`$-number) would remain intact if $`\widehat{𝗇}_\sigma ^{}(𝐤^{})`$ on the left-hand side (LHS) were the expectation value $`\widehat{𝗇}_\sigma ^{}(𝐤^{})𝗇_\sigma ^{}(𝐤^{})`$. <sup>\*†</sup><sup>\*†</sup>\*† MCDC (1998) state namely “The $`k_r^{}`$ integral over the operator expression thus yields a $`c`$-number, and the same $`c`$-number is obtained when replacing $`\widehat{𝗇}_\sigma ^{}(𝐤^{})`$ by its expectation value.” Thus MCDC (1998) employ (c.f. Eq. (23) above)
$$\widehat{𝐯}_{𝐤\sigma }^I(𝐪)𝐯_{𝐤\sigma }^I(𝐪):=\frac{1}{V}\underset{𝐤^{}\sigma ^{}}{}g_{𝐤𝐤^{}}^{\sigma \sigma ^{}}(𝐪)\mathbf{}_𝐤^{}𝗇_\sigma ^{}(𝐤^{}),$$
(26)
and consequently (c.f. Eq. (22) above)
$$\widehat{𝐣}_I(𝐪)\widehat{𝐣}_I(𝐪):=\underset{𝐤\sigma }{}𝐯_{𝐤\sigma }^I(𝐪)\widehat{a}_{𝐤𝐪/2,\sigma }^{}\widehat{a}_{𝐤+𝐪/2,\sigma }.$$
(27)
It is this $`\widehat{𝐣}_I(𝐪)`$ that MCDC (1998) employ in the construction of the generalised DL scheme. In § III.B we extensively discuss the shortcomings of the foundation of this expression for $`\widehat{𝐣}_I(𝐪)`$. Here we only briefly mention that, contrary to the statement by MCDC (1998), the relationship between the second and the third expression in Eq. (24) is defective: momenta being confined to a narrow band of width $`\lambda `$ circumscribing the Fermi surface, we have $`𝐤_o𝐤_i=\lambda `$. With $`0<\lambda 𝐤_F`$, unless the ground and low-lying excited states of the system be uncorrelated (or at most weakly correlated), it is not permitted to replace $`\widehat{𝗇}_\sigma ^{}(𝐤_i)`$ by $`1`$ and $`\widehat{𝗇}_\sigma ^{}(𝐤_o)`$ by $`0`$. To clarify this point, let us take the expectation value of the expressions in Eq. (24) with respect to the $`N`$-particle ground state of $`\widehat{H}`$. With $`0<\lambda 𝐤_F^{}`$, it is seen that $`1`$ on the RHS of Eq. (24) has to be replaced by $`𝗇_\sigma ^{}(𝐤_o)𝗇_\sigma ^{}(𝐤_i)Z_{𝐤_F^{}\sigma ^{}}`$, where $`Z_{𝐤_F^{}\sigma ^{}}`$ stands for the amount of discontinuity in $`𝗇_\sigma ^{}(𝐤^{})`$ at $`𝐤^{}=𝐤_F^{}`$ which is less than unity for interacting systems and is equal to zero for such NFL states as the marginal (Varma, et al., 1989, 1990, Littlewood and Varma 1991) and Luttinger liquids (Haldane 1980, 1981, Anderson 1997). This undue replacement of $`Z_{𝐤_F^{}\sigma ^{}}`$ by unity, is exactly the step undertaken in the construction of the Haldane (1992) bosonisation scheme in $`d>1`$; the consequences of this replacement would not be as severe were it not that in cases where $`Z_{𝐤_F^{}\sigma ^{}}`$ is small, or vanishing, the other contributions, which are neglected within the bosonisation scheme, were not so significantly more dominant than that of the quantum anomaly that in an enhanced way (through replacing $`Z_{𝐤_F^{}\sigma ^{}}`$ by unity) is taken into account (Farid 1999b). We observe that, unless the system under consideration be non-interacting, the constraint on $`\lambda `$, namely $`\lambda 𝐤_F`$, implies that in contradiction to the suggestion invoked by Eq. (24), $`dk_{}^{}{}_{r}{}^{}k_{}^{}{}_{r}{}^{d1}g_{𝐤𝐤^{}}^{\sigma \sigma ^{}}(𝐪)\mathbf{}_𝐤^{}\widehat{𝗇}_\sigma ^{}(𝐤^{})`$ is not even approximately a $`c`$-number. Be it as it may, as we shall demonstrate in § III.B, the expression in Eq. (22) with $`\widehat{𝐯}_{𝐤\sigma }^I(𝐪)`$ as defined in Eq. (23) is not the appropriate expression for $`\widehat{𝐣}_I(𝐪)`$ and, moreover, the appropriate expression (see Eqs. (57) and (55) below) even for large $`\lambda `$ cannot be replaced by a $`c`$-number.
Through applying $`(iq_0,𝐪)_\nu `$ to both sides of Eq. (10) and making use of Eqs. (14), (22) and (23), MCDC (1998) obtain the following crucial relationship
$$(iq_0,𝐪)_\nu \mathrm{\Gamma }_\sigma ^\nu (p;q)=(iq_0,𝐪)_\nu \mathrm{\Lambda }_\sigma ^\nu (p;q).$$
(28)
In arriving at this result, use has been made of the relations $`(2\pi )^1dk_0^{}\widehat{a}_{k^{}q/2,\sigma ^{}}^{}\widehat{a}_{k^{}+q/2,\sigma ^{}}=\widehat{a}_{𝐤^{}𝐪/2,\sigma ^{}}^{}\widehat{a}_{𝐤^{}+𝐪/2,\sigma ^{}}`$ and
$$𝚲_\sigma ^{}(p;q)+\widehat{𝐣}_I(q)\widehat{a}_{pq/2,\sigma }\widehat{a}_{p+q/2,\sigma }^{}_{\mathrm{tr}}^{\mathrm{irr}}\widehat{𝐣}_0(q)\widehat{a}_{pq/2,\sigma }\widehat{a}_{p+q/2,\sigma }^{}_{\mathrm{tr}}^{\mathrm{irr}}=:𝚲_\sigma (p;q).$$
(29)
From the exact Ward identity in Eq. (13) one finally obtains
$$(iq_0,𝐪)_\nu \mathrm{\Lambda }_\sigma ^\nu (p;q)=G_\sigma ^1(p+q/2)G_\sigma ^1(pq/2)$$
(30)
which in combination with Eq. (7) yields (MCDC 1998)
$$\mathrm{\Lambda }_{\sigma ^{},\sigma }^0(p;q)=\delta _{\sigma ^{},\sigma }\frac{G_\sigma ^1(p+q/2)G_\sigma ^1(pq/2)}{iq_0𝐪𝐯_𝐩^0}.$$
(31)
Eq. (2) in conjunction with Eqs. (17), (31) and the Dyson equation leads to a closed set of equations for the single-particle Green function $`G_\sigma (p)`$. One has
$$G_\sigma (p)=G_{0;\sigma }(p)\left\{1_p^{}\frac{D_{𝐩𝐩^{}}^{\sigma \sigma }(pp^{})}{i(pp_0)𝐯_{[𝐩+𝐩^{}]/2}^0(𝐩𝐩^{})}G_\sigma (p^{})+X_\sigma (p)G_\sigma (p)\right\},$$
(32)
where $`G_{0;\sigma }(p)`$ denotes the non-interacting counterpart of $`G_\sigma (p)`$ and
$$X_\sigma (p):=_p^{}\frac{D_{𝐩𝐩^{}}^{\sigma \sigma }(pp^{})}{i(pp_0)𝐯_{[𝐩+𝐩^{}]/2}^0(𝐩𝐩^{})}.$$
(33)
In practice (MCDC 1998) $`D_{𝐩𝐩^{}}^{\sigma \sigma }(pp^{})`$ (see Eq. (17)) is approximated by $`D_{𝐩𝐩}^{\sigma \sigma }(pp^{})=:D_𝐩^{\sigma \sigma }(pp^{})`$.
For the case of an isotropic system, through replacing $`𝐯_𝐤^0`$ by $`v_F^0`$ and employing for small $`𝐪:=𝐩𝐩^{}`$ (c.f. the denominators of the expressions on the RHSs of Eqs. (32) and (33))
$$𝐯_{[𝐩+𝐩^{}]/2}^0(𝐩𝐩^{})=v_F^0(p_rp_r^{})+𝒪(𝐪^2),$$
(34)
where $`p_r:=𝐩𝐤_F`$, $`p_r^{}:=𝐩^{}𝐤_F`$, and for small $`q=(iq_0,𝐪)`$,
$$D_𝐩^{\sigma \sigma }(q)=D(q_r/q_0,q_t/q_0),$$
(35)
where $`q_r:=𝐪𝐩/𝐩`$ and $`q_t:=(𝐪^2q_r^2)^{1/2}0`$, the radial and tangential components of $`𝐪`$, respectively, MCDC (1998) render the expression for $`G_\sigma (p)`$ in Eq. (32) suitable for treatment by the conventional method of dimensional continuation (see Appendix — we note in passing that here $`q_t`$ is the component of $`𝐪`$ inside the tangent space specified in Appendix). In this way MCDC (1998) establish that the cross-over dimension from Luttinger- to FL state is $`d=1`$; that is, the metallic state of the system under consideration is a FL (though not necessarily a conventional one) in $`d>1`$. This finding is in agreement with the $`ϵ`$-expansion result by Ueda and Rice (1984) around $`d=1`$; according to these authors, in the $`d=1+ϵ`$ fermion system the only weak-coupling fixed point is the FL one. As we shall demonstrate in § III.A, the expression in Eq. (31) whose denominator on the RHS is the origin of the denominators on the RHSs of the expressions in Eqs. (32) and (33) (which is further simplified in Eq. (34)), is not valid except in the strictly non-interacting case. In § III.A we further indicate that for NFLs, $`\mathrm{\Sigma }_\sigma (𝐩\pm 𝐪/2,\omega \pm \omega _0/2)`$ does not allow for an effective disentangling of $`𝐪`$ from $`𝐩`$ (as well as $`\omega _0`$ from $`\omega `$) and consequently for these systems the appropriate equation for $`G_\sigma (p)`$ does not accommodate terms involving contributions whose form would resemble the denominator of the integrands on the RHSs of Eqs. (32) and (33) which thus would be amenable to such reduction as in Eq. (34). Note that the expression in Eq. (34) is specific in that it does not involve a component of $`𝐩^{}`$ in the tangent subspace $`S_{}`$ (see Appendix). Thus we conclude that the general formalism for systems of interacting fermions even in the limit of strong forward scattering does not lend itself for treatment by the conventional method of dimensional continuation.
In the following Section we subject the generalised DL formalism to a critical analysis and demonstrate its failure in $`d>1`$. We also indicate the underlying reasons for the exactness of this formalism in $`d=1`$ when applied to the one-dimensional Luttinger model (1963).
## III Critical analyses
In this Section we present our analyses of some crucial elements that are involved in the derivation of the generalised DL method described in § II.
### A On the relationship between $`\mathrm{\Lambda }_\sigma ^0(p;q)`$ and $`𝚲_\sigma (p;q)`$
Here we restrict our considerations to the case of zero temperature ($`T=0`$) and deal with the analytic continuations of functions of imaginary energy along the real energy axis. We denote the analytic continuations of such functions as $`G_\sigma (p)`$ and $`\mathrm{\Sigma }_\sigma (p)`$ by $`G_\sigma (𝐩,\omega )`$ and $`\mathrm{\Sigma }_\sigma (𝐩,\omega )`$ respectively.
From the Dyson equation for the Green function it follows that <sup>\*‡</sup><sup>\*‡</sup>\*‡ In expressions akin to that in Eq. (36) (such as that presented in Eq. (46) below), MCDC (1998) add $`𝒪(𝐪^2)`$ to the RHSs in order to account for the fact that in general the non-interacting energy dispersions can be more complicated than linear or quadratic functions of momentum. We do not follow this practice for the reason that the one-particle kinetic-energy operator $`\widehat{H}_0`$ and the velocity operator $`\widehat{𝐯}`$ are related to one another through the momentum operator $`\widehat{𝐩}`$; we have $`\widehat{H}_0:=\widehat{𝐩}^2/2`$ and $`\widehat{𝐯}:=\widehat{𝐩}`$ (we use $`m_f=1`$ throughout), so that in the momentum representation must hold (assuming a uniform state) $`\epsilon _𝐩^0𝐩^2/2`$ and $`𝐯_𝐩^0𝐩`$ for all $`𝐩`$ (from this perspective $`\epsilon _{𝐩+𝐪/2}^0\epsilon _{𝐩𝐪/2}^0=𝐪𝐯_𝐩^0`$, whence absence of $`𝒪(𝐪^2)`$ in Eq. (46)). A fully consistent generalisation of the formalism considered in the present work, that is one capable of dealing with general non-interacting energy dispersions, must therefore involve some reformulation of the problem at hand.
$`G_\sigma ^1(𝐩+𝐪/2,\omega +\omega _0/2)G_\sigma ^1(𝐩𝐪/2,\omega \omega _0/2)=\omega _0𝐪𝐯_{𝐩\sigma }^0`$ (36)
$`\left[\mathrm{\Sigma }_\sigma (𝐩+𝐪/2,\omega +\omega _0/2)\mathrm{\Sigma }_\sigma (𝐩𝐪/2,\omega \omega _0/2)\right].`$ (37)
Assuming continuous differentiability of $`\mathrm{\Sigma }_\sigma (𝐩^{},\omega ^{})`$ with respect to $`𝐩^{}`$ and $`\omega ^{}`$ in neighbourhoods of $`𝐩^{}=𝐩`$ and $`\omega ^{}=\omega `$ respectively, for sufficiently small $`𝐪`$ and $`|\omega _0|`$ we have
$`G_\sigma ^1(𝐩+𝐪/2,\omega +\omega _0/2)`$ $``$ $`G_\sigma ^1(𝐩𝐪/2,\omega \omega _0/2)=\omega _0\left[1\mathrm{\Sigma }_\sigma (𝐩,\omega )/\omega \right]`$ (39)
$`𝐪\left[𝐯_{𝐩\sigma }^0+\mathbf{}_𝐩\mathrm{\Sigma }_\sigma (𝐩,\omega )\right]+𝒪(𝐪^\gamma ,|\omega _0|^\gamma ^{}),`$
where $`\gamma ,\gamma ^{}>1`$. From Eqs. (30) and (39), up to corrections of the form $`𝒪(𝐪^\gamma ,|\omega _0|^\gamma ^{})`$, we deduce
$$\mathrm{\Lambda }_\sigma ^0(𝐩,\omega )=1\frac{\mathrm{\Sigma }_\sigma (𝐩,\omega )}{\omega },𝚲_\sigma (𝐩,\omega )=𝐯_{𝐩\sigma }^0+\mathbf{}_𝐩\mathrm{\Sigma }_\sigma (𝐩,\omega ).$$
(40)
Following the expressions in Eq. (6) relating $`\mathrm{\Lambda }_\sigma ^0`$ to $`𝚲_\sigma `$, the results in Eq. (40) imply
$$\mathbf{}_𝐩\mathrm{\Sigma }_\sigma (𝐩,\omega )=\frac{\mathrm{\Sigma }_\sigma (𝐩,\omega )}{\omega }𝐯_{𝐩\sigma }^0,$$
(41)
which must hold exactly for $`𝐩=𝐩_F`$ and $`\omega =\omega _F`$ if the results in Eqs. (7) and (31) are to be asymptotically valid for $`𝐪0`$ and $`|\omega _0|0`$. We note that a $`\mathrm{\Sigma }_\sigma (𝐩,\omega )`$ whose associated $`\mathrm{\Sigma }_\sigma (𝐩_F,\omega )`$ and $`\mathrm{\Sigma }_\sigma (𝐩,\omega _F)`$ are continuously differentiable functions of $`\omega `$ and $`𝐩`$ in neighbourhoods of $`\omega =\omega _F`$ and $`𝐩=𝐩_F`$, respectively, by definition corresponds to a FL (Farid 1999a). In other words, if valid, Eq. (41) must hold for FLs when $`\omega =\omega _F`$ as $`𝐩𝐩_F`$ and $`𝐩=𝐩_F`$ as $`\omega \omega _F`$.
Let us now consider the following asymptotic expressions which apply to conventional FLs in $`d=2`$ for $`𝐩`$ close to the Fermi surface and $`\omega \omega _F`$ (Hodges, Smith and Wilkins 1971, Bloom 1975) See also Fujimoto (1990) and Fukuyama, Narikio and Hasegawa (1991). (in what follows we assume the system under consideration to be in the paramagnetic phase and thus suppress the spin indices):
$`\mathrm{Re}\mathrm{\Sigma }(𝐩,\omega )`$ $``$ $`\mathrm{\Sigma }(𝐩,\omega _F)+\beta _𝐩(\omega \omega _F),`$ (42)
$`\mathrm{Im}\mathrm{\Sigma }(𝐩,\omega )`$ $``$ $`\alpha _𝐩\mathrm{sgn}(\omega \omega _F)(\omega \omega _F)^2\mathrm{ln}|\omega \omega _F|,`$ (43)
where $`\beta _𝐩`$ and $`\alpha _𝐩`$ are real valued and for interacting FLs $`\beta _𝐩<0`$ (strictly negative; see text following Eq. (48) below) and $`\alpha _𝐩0`$; for conventional FLs in $`d=2`$, $`\mathrm{\Sigma }(𝐩,\omega _F)0`$ and $`\alpha _𝐩>0`$. Substitution of the RHSs of Eqs. (42) and (43) in Eq. (41) results in
$`\mathbf{}_𝐩\mathrm{\Sigma }(𝐩,\omega _F)+\left[\mathbf{}_𝐩\beta _𝐩\right](\omega \omega _F)`$ $``$ $`\beta _𝐩𝐯_𝐩^0,`$ (44)
$`\left[\mathbf{}_𝐩\alpha _𝐩\right](\omega \omega _F)\mathrm{ln}|\omega \omega _F|`$ $``$ $`\left(2\mathrm{ln}|\omega \omega _F|+1\right)\alpha _𝐩𝐯_𝐩^0.`$ (45)
We now consider the consequences of these results. As will become evident, Eqs. (44) and (45) hold only for strictly non-interacting FLs (in this case they hold identically), thus implying incorrectness of Eq. (41) and by extension that of the expressions in Eq. (6) for interacting systems. For $`\omega =\omega _F`$, Eq. (44) leads to $`\mathbf{}_𝐩\mathrm{\Sigma }(𝐩,\omega _F)\beta _𝐩𝐯_𝐩^0`$ (for $`𝐩`$ in the vicinity of the Fermi momentum), which, according to Eq. (44) for $`\omega \omega _F`$, implies $`\mathbf{}_𝐩\beta _𝐩\mathrm{𝟎}\beta _𝐩0`$ (see further on). This result in turn necessitates $`\mathbf{}_𝐩\mathrm{\Sigma }(𝐩,\omega _F)\mathrm{𝟎}`$; further, for $`\omega \omega _F`$ the LHS of Eq. (45) vanishes while, unless $`\alpha _𝐩0`$, the RHS diverges. It follows that unless the system of fermions (in $`d=2`$) be non-interacting, Eq. (41) and, in consequence, the relations in Eq. (6) are incorrect (see text following Eq. (49) below). For non-interacting systems we have namely (see Footnote \*‡III A above)
$$G_0^1(𝐩+𝐪/2,\omega +\omega _0/2)G_0^1(𝐩𝐪/2,\omega \omega _0/2)\omega _0𝐪𝐯_𝐩^0,$$
(46)
from which it is readily deduced that, with $`(1,𝐯_𝐩^0)^\nu `$ the non-interacting counterpart of the irreducible vertex part $`\mathrm{\Lambda }_\sigma ^\nu (p;q)`$, the relations in Eq. (6) are identically valid in the strictly non-interacting case. This completes the proof of our statement with regard to the condition for validity of the expressions in Eq. (6). We emphasise that the case of $`d=1`$ is very special when $`𝐯_𝐩^0`$ is a constant, say $`𝐯^0`$ (which points to a definite direction), as is the case for the one-dimensional Luttinger (1963) model. Here the relations in Eq. (6) hold identically, for whatever $`\mathrm{\Lambda }_\sigma ^0(p;q)`$ may be, from the definition for $`\widehat{j}_0^\nu (𝐪)`$, namely (see Eqs. (18), (4) and (19) above)
$$\widehat{j}_0^\nu (𝐪):=(\widehat{n}(𝐪),\widehat{𝐣}_0(𝐪))^\nu \underset{𝐤\sigma }{}(1,𝐯_𝐤^0)^\nu \widehat{a}_{𝐤𝐪/2\sigma }^{}\widehat{a}_{𝐤+𝐪/2\sigma },$$
(47)
it follows that $`𝚲_\sigma (p;q)`$ is $`𝐯^0`$ times $`\mathrm{\Lambda }_\sigma ^0(p;q)`$ (see Eq. (6)).
We conclude that the relations in Eq. (6) are not asymptotically correct in the forward-scattering limit when $`d>1`$, unless the system under consideration be non-interacting, in which case they are identically valid in any spatial dimension.
Above we have asserted that $`\mathbf{}_𝐩\beta _𝐩\mathrm{𝟎}\beta _𝐩0`$. We now demonstrate the validity of this equivalence. For FLs, it can be shown that (Farid 1999a)
$$\beta _𝐩=\frac{1}{\pi }_0^{\mathrm{}}\frac{\mathrm{d}\omega ^{}}{\omega _{}^{}{}_{}{}^{2}}\left[\mathrm{Im}\mathrm{\Sigma }(𝐩,\omega ^{}+\omega _F)\mathrm{Im}\mathrm{\Sigma }(𝐩,\omega ^{}+\omega _F)\right].$$
(48)
We note in passing that since for interacting FLs the integrand of this expression is not identically vanishing and, moreover, owing to the assumed stability of the system $`\mathrm{Im}\mathrm{\Sigma }(𝐩,\omega ^{}+\omega _F)`$ and $`\mathrm{Im}\mathrm{\Sigma }(𝐩,\omega ^{}+\omega _F)`$ are never positive for $`\omega ^{}0`$, it follows that $`\beta _𝐩<0`$ (strictly negative). The expression in Eq. (48) can be transformed into For FLs $`_\mu ^i\mathrm{}dz(z\mu )^2\mathrm{Im}[\mathrm{\Sigma }(𝐩,z)\mathrm{\Sigma }(𝐩,2\mu z)]`$ exists, however $`_\mu ^i\mathrm{}dz(z\mu )^2[\mathrm{\Sigma }(𝐩,z)\mathrm{\Sigma }(𝐩,2\mu z)]`$ does not. Addition of the infinitesimal $`0^+`$ to $`\mu `$ in the lower boundary of the latter integral (see Eq. (49)) formally renders this integral existent without affecting the value of $`\beta _𝐩`$.
$$\beta _𝐩=\frac{1}{\pi }\mathrm{Im}_{\mu +0^+}^i\mathrm{}\frac{\mathrm{d}z}{(z\mu )^2}\left[\mathrm{\Sigma }(𝐩,z)\mathrm{\Sigma }(𝐩,2\mu z)\right],$$
(49)
where $`\mu (\mu _N,\mu _{N+1})`$ with $`\mu _N:=E_{N,0}E_{N1,0}\omega _F`$ and $`\mu _{N+1}:=E_{N+1,0}E_{N,0}`$. Here $`E_{M,0}`$ stands for the total energy of the interacting $`M`$-particle ground state and $`N`$ denotes the actual number of the particles in the system. The $`z`$-integration in Eq. (49) is carried out along an arbitrary contour in the upper half of the complex $`z`$ plane, connecting $`z=\mu +0^+`$ with the point of infinity. The arbitrariness (under the mentioned constraints) of the contour of the $`z`$-integration has a far-reaching consequence: assuming differentiability of $`\beta _𝐩`$ with respect to $`𝐩`$ (in some neighbourhood of $`𝐩_F`$), it follows from $`\mathbf{}_𝐩\beta _𝐩\mathrm{𝟎}`$ that $`\mathbf{}_𝐩\left[\mathrm{\Sigma }(𝐩,z)\mathrm{\Sigma }(𝐩,2\mu z)\right]\mathrm{𝟎}`$ for all $`z`$ in the upper half of the complex $`z`$ plane; by the Riemann-Schwarz reflection property with respect to the real axis, this identity holds also for all points in the lower half of the complex $`z`$ plane. <sup>\*∥</sup><sup>\*∥</sup>\*∥ For $`\mathrm{Im}(z)0`$, $`\mathrm{\Sigma }(𝐩,z)`$ satisfies the Riemann-Schwarz reflection property $`\mathrm{\Sigma }(𝐩,z^{})\mathrm{\Sigma }^{}(𝐩,z)`$ (see e.g., DuBois 1959, Luttinger 1961, Farid 1999a). This implies that $`\mathrm{\Sigma }(𝐩,z)\mathrm{\Sigma }(𝐩,2\mu z)`$ over the entire complex plane. In view of the expressions in Eqs. (42) and (43), this result is seen to apply only if $`\mathrm{\Sigma }(𝐩,z)`$ is independent of $`z`$, a property which is specific to non-interacting systems. A corollary to this result is that $`\beta _𝐩0`$ (that is zero rather than any other constant that $`\mathbf{}_𝐩\beta _𝐩\mathrm{𝟎}`$ would suggest in the first instance; see the text following Eq. (45) above). <sup>\***</sup><sup>\***</sup>\*** Let $`\mathrm{Re}(z)=:\omega >\mu `$ and $`\mathrm{Im}(z)=:\eta `$, $`\eta 0`$. For such a $`z`$ we have $`\mathrm{Im}\mathrm{\Sigma }(𝐩,z)0`$ while $`\mathrm{Im}\mathrm{\Sigma }(𝐩,2\mu z)0`$ (validity of both of these relations is necessary for the stability of the system). Satisfaction of $`\mathrm{\Sigma }(𝐩,z)\mathrm{\Sigma }(𝐩,2\mu z)`$, for all $`z`$, in this particular case would imply $`\mathrm{Im}\mathrm{\Sigma }(𝐩,\omega +i\eta )0`$ for all $`\omega >\mu `$; it similarly would imply $`\mathrm{Im}\mathrm{\Sigma }(𝐩,\omega i\eta )0`$ for all $`\omega <\mu `$. By the Kramers-Krönig relation it follows that $`\mathrm{\Sigma }(𝐩,z)`$ is independent of $`z`$. Note that in obtaining this result we have not relied upon the specific form of the expressions in Eqs. (42) and (43).
Our above analysis has explicitly concerned FLs (through our explicit use of the relations in Eqs. (42) and (43), but more importantly Eq. (48)), however, as can be easily verified, for NFLs the RHS of Eq. (36) does not allow for any effective separation of $`𝐪`$ and $`\omega _0`$ from $`𝐩`$ and $`\omega `$ respectively, so that in $`d>1`$ the relations in Eq. (6) lack even the most fundamental basis for validity as regards NFLs.
### B On the relationship between $`\mathrm{\Lambda }_\sigma ^\nu (p;q)`$ and $`\mathrm{\Gamma }_\sigma ^\nu (p;q)`$
Evidently for obtaining the $`\widehat{𝐣}_I(𝐪)`$ as presented in Eq. (22) in terms of $`\widehat{𝐯}_{𝐤\sigma }^I(𝐪)`$ in Eq. (23), the functions in the square brackets on the RHS of Eq. (20) have been expanded to linear order in $`𝐪`$. Such an expansion is in general invalid for reasons that we discuss in some detail in § IV. A second step involved in the derivation of $`\widehat{𝐣}_I(𝐪)`$ in Eq. (27) is that of replacing $`\mathbf{}_𝐤\widehat{𝗇}_\sigma (𝐤)`$ by the expectation value $`\mathbf{}_𝐤\widehat{𝗇}_\sigma (𝐤)`$. The justification that has been given by MCDC (1998) for this substitution is untenable (in particular in $`d>1`$) (see also text following Eq. (27) above). For obtaining a generally valid expression for $`\widehat{𝐣}_I(𝐪)`$, the RHS of Eq. (20) must be first written in terms of four separate contributions, each involving only one of $`\widehat{𝗇}_\sigma (𝐤𝐪/2)`$, $`\widehat{𝗇}_\sigma (𝐤+𝐪/2)`$, $`\widehat{𝗇}_\sigma ^{}(𝐤^{}𝐪/2)`$ and $`\widehat{𝗇}_\sigma ^{}(𝐤^{}+𝐪/2)`$. Subsequently, in each of these contributions, by appropriate transformations of the summation variables $`𝐤`$ and $`𝐤^{}`$, the vector $`𝐪`$ in the arguments of these functions must be transferred into that of the interaction function $`g_{𝐤𝐤^{}}^{\sigma \sigma ^{}}(𝐪)`$, which we assume to be a continuously differentiable function of its indices $`𝐤`$ and $`𝐤^{}`$ (c.f. statement preceding Eq. (7) in Castellani, Di Castro and Metzner 1994). In this way one obtains (for small $`𝐪`$)
$`\widehat{𝐣}_I(𝐪)`$ $`=`$ $`{\displaystyle \frac{1}{2V}}{\displaystyle \underset{𝐤\sigma }{}}{\displaystyle \underset{𝐤^{}\sigma ^{}}{}}\{\left[\mathbf{}_𝐤g_{𝐤𝐤^{}}^{\sigma \sigma ^{}}(𝐪)\right]\widehat{𝗇}_\sigma (𝐤)\widehat{a}_{𝐤^{}𝐪/2,\sigma ^{}}^{}\widehat{a}_{𝐤^{}+𝐪/2,\sigma ^{}}`$ (51)
$`+\left[\mathbf{}_𝐤^{}g_{𝐤𝐤^{}}^{\sigma \sigma ^{}}(𝐪)\right]\widehat{a}_{𝐤𝐪/2,\sigma }^{}\widehat{a}_{𝐤+𝐪/2,\sigma }\widehat{𝗇}_\sigma ^{}(𝐤^{})\}.`$
Making use of the commutation relation
$$[\widehat{a}_{𝐤𝐪/2,\sigma }^{}\widehat{a}_{𝐤+𝐪/2,\sigma },\widehat{𝗇}_\sigma ^{}(𝐤^{})]_{}=\delta _{\sigma ,\sigma ^{}}\left\{\delta _{𝐤+𝐪/2,𝐤^{}}\delta _{𝐤𝐪/2,𝐤^{}}\right\}\widehat{a}_{𝐤𝐪/2,\sigma }^{}\widehat{a}_{𝐤+𝐪/2,\sigma },$$
(52)
which is obtained through a repeated application of the canonical anti-commutation relations for $`\widehat{a}_{𝐤,\sigma }^{}`$ and $`\widehat{a}_{𝐤,\sigma }`$, the expression in Eq. (51) can be written as
$`\widehat{𝐣}_I(𝐪)`$ $`=`$ $`{\displaystyle \underset{𝐤\sigma }{}}\widehat{𝐯}_{𝐤\sigma }^I(𝐪)\widehat{a}_{𝐤𝐪/2,\sigma }^{}\widehat{a}_{𝐤+𝐪/2,\sigma }`$ (54)
$`+{\displaystyle \frac{1}{2V}}{\displaystyle \underset{𝐤\sigma }{}}\left\{\left[\mathbf{}_𝐤^{}g_{𝐤𝐤^{}}^{\sigma \sigma }(𝐪)|_{𝐤^{}=𝐤+𝐪/2}\right]\left[\mathbf{}_𝐤^{}g_{𝐤𝐤^{}}^{\sigma \sigma }(𝐪)|_{𝐤^{}=𝐤𝐪/2}\right]\right\}\widehat{a}_{𝐤𝐪/2,\sigma }^{}\widehat{a}_{𝐤+𝐪/2,\sigma },`$
where (c.f. Eq. (23) above)
$$\widehat{𝐯}_{𝐤\sigma }^I(𝐪):=\frac{1}{V}\underset{𝐤^{}\sigma ^{}}{}\left[\mathbf{}_𝐤^{}\overline{g}_{𝐤𝐤^{}}^{\sigma \sigma ^{}}(𝐪)\right]\widehat{𝗇}_\sigma ^{}(𝐤^{}),$$
(55)
in which
$$\overline{g}_{𝐤𝐤^{}}^{\sigma \sigma ^{}}(𝐪):=\frac{1}{2}\left(g_{𝐤𝐤^{}}^{\sigma \sigma ^{}}(𝐪)+g_{𝐤^{}𝐤}^{\sigma ^{}\sigma }(𝐪)\right).$$
(56)
By assuming $`\mathbf{}_𝐤^{}g_{𝐤𝐤^{}}^{\sigma ,\sigma }(𝐪)`$ to be a continuously-differentiable function of $`𝐤^{}`$, the last term on the RHS of Eq. (54) is seen to be proportional to $`𝐪`$, so that in the limit of $`𝐪\mathrm{𝟎}`$ it can be neglected in comparison with the first term. For small $`𝐪`$ we can thus write
$$\widehat{𝐣}_I(𝐪)=\underset{𝐤\sigma }{}\widehat{𝐯}_{𝐤\sigma }^I(𝐪)\widehat{a}_{𝐤𝐪/2,\sigma }^{}\widehat{a}_{𝐤+𝐪/2,\sigma },$$
(57)
where $`\widehat{𝐯}_{𝐤\sigma }^I(𝐪)`$ is defined in Eq. (55). Note in passing, that for a constant interaction (i.e. a $`𝐤,𝐤^{}`$-independent $`g_{𝐤𝐤^{}}^{\sigma \sigma ^{}}(𝐪)`$), $`\widehat{𝐯}_{𝐤\sigma }^I(𝐪)\widehat{\mathrm{𝟎}}`$, so that Eq. (57) correctly yields $`\widehat{𝐣}_I(𝐪)\widehat{\mathrm{𝟎}}`$, or $`\widehat{𝐣}(𝐪)\widehat{𝐣}_0(𝐪)`$.
The correlated velocity operator in Eq. (55) is distinct from that in Eq. (23). To bring out the fundamental difference between the two, let us follow MCDC (1998) and replace the operator $`\widehat{𝗇}_\sigma ^{}(𝐤^{})`$ on the RHS of Eq. (55) by its expectation value $`𝗇_\sigma ^{}(𝐤^{})`$ and compare the resulting vector $`𝐯_{𝐤\sigma }^I(𝐪)`$ with that in Eq. (26). Since by assumption $`\mathbf{}_𝐤^{}\overline{g}_{𝐤𝐤^{}}^{\sigma \sigma ^{}}(𝐪)`$ is a smooth function of $`𝐤^{}`$ (see text preceding Eq. (51) above), we can write (recall that $`𝐤_o^{}𝐤_i^{}=\lambda 𝐤_F^{}`$)
$$dk_r^{}k_{r}^{}{}_{}{}^{d1}\left[\mathbf{}_𝐤^{}\overline{g}_{𝐤𝐤^{}}^{\sigma \sigma ^{}}(𝐪)\right]𝗇_\sigma ^{}(𝐤^{})\left[\mathbf{}_𝐤^{}\overline{g}_{𝐤𝐤^{}}^{\sigma \sigma ^{}}(𝐪)|_{𝐤^{}=𝐤_F^{}}\right]dk_r^{}k_{r}^{}{}_{}{}^{d1}𝗇_\sigma ^{}(𝐤^{}).$$
(58)
For the non-interacting case, where $`𝗇_\sigma ^{}(𝐤^{})=\mathrm{\Theta }(𝐤_F^{}𝐤^{})`$ (recall that distances are measured along $`𝒏_{𝐤_F^{}}`$), the integral on the RHS of Eq. (58) is equal to $`\left(𝐤_F^{}^d𝐤_i^{}^d\right)/d`$, so that the RHS of Eq. (58) becomes equal to $`\left[\mathbf{}_𝐤^{}\overline{g}_{𝐤𝐤^{}}^{\sigma \sigma ^{}}(𝐪)|_{𝐤^{}=𝐤_F^{}}\right]\left(𝐤_F^{}^d𝐤_i^{}^d\right)/d`$, to be compared with the RHS of Eq. (24), which is equally applicable to the non-interacting ground states, namely $`g_{𝐤𝐤_F^{}}^{\sigma \sigma ^{}}(𝐪)𝐤_F^{}^d𝒏_{𝐤_F^{}}`$. The two results are manifestly different. This demonstrates that Eq. (26), and therefore Eq. (27), is fundamentally incorrect. This aspect is additional to the shortcoming of the result in Eq. (24), which as we have discussed in the text following Eq. (27), shows up when the ground and low-lying excited states of the system are (strongly) correlated. Thus we have unequivocally established that not only the change in Eq. (27), corresponding to replacing $`\widehat{𝐯}_{𝐤\sigma }^I(𝐪)`$ as defined in Eq. (23) by $`𝐯_{𝐤\sigma }^I(𝐪)`$ as defined in Eq. (26), lacks justification (unless the system under consideration be non-interacting), but also that the expression for $`\widehat{𝐯}_{𝐤\sigma }^I(𝐪)`$ as presented in Eq. (23) (to be compared with the $`\widehat{𝐯}_{𝐤\sigma }^I(𝐪)`$ in Eq. (55)) is invalid, this on account of the fact that it relies on the Taylor expansion of an operator (that is $`\widehat{𝗇}_\sigma (𝐤\pm 𝐪/2)`$) whose, in particular, expectation values with respect to ground states are in general non-differentiable (see § IV).
A major consequence of use of Eq. (57), or Eq. (22), rather than one in which the operator $`\widehat{𝐯}_{𝐤\sigma }^I(𝐪)`$ is replaced by the $`c`$-number $`𝐯_{𝐤\sigma }^I(𝐪)`$ (effected through substituting $`𝗇_\sigma ^{}(𝐤^{})`$ for $`\widehat{𝗇}_\sigma ^{}(𝐤^{})`$ in Eq. (55), or Eq. (23)), is violation of Eq. (28) which is the crucial link between the exact Ward identity in Eq. (13) and the expression in Eq. (30). To make this statement explicit, we first apply $`(iq_o,𝐪)_\nu `$ to both sides of Eq. (10) and subsequently substitute the expression on the RHS of Eq. (14) for $`(iq_0,𝐪)_\nu J^{\nu ,𝐤^{}\sigma ^{}}(q)`$. Note that in doing so, we depart from the approach by MCDC (1998) who employ the expression on the RHS of Eq. (16). Next, we write the expression for $`(iq_0,𝐪)_\nu J^{\nu ,𝐤^{}\sigma ^{}}(q)`$ in the contribution corresponding to $`(iq_0,𝐪)_\nu \mathrm{\Gamma }_\sigma ^\nu (p;q)`$ as two separate terms and in each of these by transformations of the dummy variables of summation, remove $`𝐪/2`$ from arguments of $`𝗇_\sigma ^{}(𝐤^{}𝐪/2)`$ and $`𝗇_\sigma ^{}(𝐤^{}+𝐪/2)`$. This procedure is exactly the same as that which we employed for deducing $`\widehat{𝐣}_I(𝐪)`$ from the equation of continuity associated with $`\widehat{H}_I`$ (see Eq. (51) above and text following it). Thus we obtain
$`{\displaystyle \frac{1}{V}}{\displaystyle \underset{𝐤^{}\sigma ^{}}{}}{\displaystyle \underset{𝐤^{\prime \prime }\sigma ^{\prime \prime }}{}}(iq_0,𝐪)_\nu J^{\nu ,𝐤^{}\sigma ^{}}(q)g_{𝐤^{}𝐤^{\prime \prime }}^{\sigma ^{}\sigma ^{\prime \prime }}(q)\mathrm{\Lambda }_\sigma ^{𝐤^{\prime \prime }\sigma ^{\prime \prime }}(p;q)`$ (59)
$`=𝐪{\displaystyle \underset{𝐤^{\prime \prime }\sigma ^{\prime \prime }}{}}\left\{{\displaystyle \frac{1}{V}}{\displaystyle \underset{𝐤^{}\sigma ^{}}{}}𝗇_\sigma ^{}(𝐤^{})\left[\mathbf{}_𝐤^{}g_{𝐤^{}𝐤^{\prime \prime }}^{\sigma ^{}\sigma ^{\prime \prime }}(𝐪)\right]\right\}\mathrm{\Lambda }_\sigma ^{𝐤^{\prime \prime }\sigma ^{\prime \prime }}(p;q).`$ (60)
Neglecting the possible difference between $`g_{𝐤𝐤^{}}^{\sigma \sigma ^{}}(𝐪)`$ and $`\overline{g}_{𝐤𝐤^{}}^{\sigma \sigma ^{}}(𝐪)`$ (see Eq. (56) above), it is seen that the term enclosed by curly braces in Eq. (59) is equal to the expectation value of $`\widehat{𝐯}_{𝐤\sigma }^I(𝐪)`$ as presented in Eq. (55). As we have extensively discussed above, there exists no justification, whatever, for $`𝐯_{𝐤\sigma }^I(𝐪)`$ to be a good representation of $`\widehat{𝐯}_{𝐤\sigma }^I(𝐪)`$. If, as in the treatment by MCDC (1998), $`\widehat{𝐯}_{𝐤\sigma }^I(𝐪)`$ could be replaced by its expectation value, then making use of the defining expression for $`\mathrm{\Lambda }_\sigma ^{𝐤^{\prime \prime }\sigma ^{\prime \prime }}(p;q)`$ in Eq. (12) and the expression for $`\widehat{𝐣}_I(𝐪)`$ in Eq. (57) wherein $`\widehat{𝐯}_{𝐤\sigma }^I(𝐪)`$ is replaced by $`𝐯_{𝐤\sigma }^I(𝐪)`$, the RHS of Eq. (59) would reduce to $`𝐪𝚲_\sigma (p;q)`$. Adding this result to $`(iq_0,𝐪)_\nu 𝚲_{\sigma }^{}{}_{}{}^{\nu }(p;q)`$ (see Eq. (10), to whose both sides $`(iq_0,𝐪)_\nu `$ is applied), one would arrive at Eq. (28). We emphasise that in arriving at Eq. (28) it is the substitution of $`\widehat{𝐯}_{𝐤\sigma }^I(𝐪)`$ by $`𝐯_{𝐤\sigma }^I(𝐪)`$ that plays the vital role and not whether $`𝐯_{𝐤\sigma }^I(𝐪)`$ is the expectation value of $`\widehat{𝐯}_{𝐤\sigma }^I(𝐪)`$ as defined in Eq. (23) or Eq. (55). However, it should be realised that contrary to the case of the $`\widehat{𝐯}_{𝐤\sigma }^I(𝐪)`$ in Eq. (23) whose replacement by its expectation value can be justified for non-interacting systems (and by extension, weakly interacting systems) (see § II, text following Eq. (27)), replacement of the $`\widehat{𝐯}_{𝐤\sigma }^I(𝐪)`$ in Eq. (55), which we have demonstrated to be the correct velocity operator, by its expectation value is not justified, not even for cases where the ground and low-lying excited states of the system are non-interacting (or weakly interacting). Of course, in the non-interacting case $`g_{𝐤𝐤^{}}^{\sigma \sigma ^{}}(𝐪)`$ is identically vanishing so that the problem associated with the inadmissibility of replacing $`\widehat{𝐯}_{𝐤\sigma }^I(𝐪)`$ by $`𝐯_{𝐤\sigma }^I(𝐪)`$ does not arise in this limit (see Eq. (55) above).
Above we have demonstrated that in $`d>1`$, the trivial relationship between the vertex parts $`\mathrm{\Lambda }_\sigma ^\nu (p;q)`$ and $`\mathrm{\Gamma }_\sigma ^\nu (p;q)`$ as given in Eq. (28) is invalid and consequently no direct link exists between the exact Ward identity in Eq. (13) and the expression in Eq. (30). Further, our considerations in § III.A made evident that also Eq. (7) has no rightful place in the theory of interacting fermions in $`d>1`$.
## IV Mathematical transgressions and “blessings” of incorrect error estimation
In this Section we illustrate how in $`d=1`$, specifically in dealing with the one-dimensional Luttinger (1963) model at $`T=0`$, a combination of transgressions against mathematical rules and uncritical neglect of truncation errors can give rise to outcomes which coincide with the exact results, thus creating the false impression that this approach were justified. The particular example that we consider in this Section makes explicit that specifically in $`d>1`$ the same procedure yields manifestly incorrect results. Our illustrative example involves the ground-state momentum distribution function $`𝗇(𝐤)`$ which is singular at $`𝐤=𝐤_F`$ both in the case of FLs and NFLs.
We first explicitly deal with the one-dimensional Luttinger model for spin-less fermions (Luttinger 1963, Mattis and Lieb 1965), characterised by the anomalous dimension $`\alpha `$, $`\alpha (0,1)`$. We note in passing that for the one-dimensional Hubbard model $`0<\alpha <1/8`$, with $`\alpha =1/8`$ corresponding to the limit of infinite on-site repulsion (Voit 1993). By choosing the positive direction to be from left to right, below we consider the one-dimensional vectors $`𝐤`$ and $`𝐪`$ as scalars and $`k_F|𝐤_F|`$. Here the ground-state momentum-distribution function $`𝗇_r(𝐤)`$, with $`r=\pm `$ the branch index ($`r=+`$ indicates the branch of the right movers and $`r=`$ that of the left movers), is continuous but not differentiable at $`𝐤=rk_F`$. For the density-density correlation function $`\mathrm{\Pi }_{rr}(q)`$ pertaining to the interacting system we have (DL 1974; see, e.g., Metzner and Di Castro 1993, Voit 1994) <sup>\*††</sup><sup>\*††</sup>\*†† The boundaries of the integral $`g_r(𝐪)`$ are determined by the observation that $`𝐤`$ in $`𝗇_r(𝐤)`$ is restricted to the range $`\lambda /2+rk_F𝐤\lambda /2+rk_F`$.
$$\mathrm{\Pi }_{rr}(q)=\frac{1}{2\pi }\frac{g_r(𝐪)}{iq_0𝐯_r^0𝐪},g_r(𝐪):=_{\lambda /2+rk_F+𝐪/2}^{\lambda /2+rk_F𝐪/2}d𝐤\left[𝗇_r(𝐤𝐪/2)𝗇_r(𝐤+𝐪/2)\right],$$
(61)
where $`𝐯_r^0`$ stands for the constant velocity of the non-interacting fermions; it points to the positive direction for $`r=+`$ and the negative direction for $`r=`$. Further, $`\lambda `$ is a cut-off which is finite but can be arbitrarily large; for our following considerations it is required that $`\lambda `$ be sufficiently large so that $`𝗇_r(𝐤)`$ can be replaced by zero (unity) for $`r=+`$ ($`r=`$) when $`𝐤\lambda /2`$. The freedom to chose $`\lambda `$ arbitrarily large is a specific aspect of the one-dimensional Luttinger model. We note that although introduction of such a cut-off is necessary for a correct treatment of the Luttinger model (Mattis and Lieb 1965), imposition of a ‘sharp’ cut-off on $`𝗇_r(𝐤)`$, implying $`𝗇_\pm (\lambda /20^+)=0`$ and $`𝗇_\pm (\lambda /2\pm 0^+)=1`$, gives rise to an incorrect asymptotic result for the long-distance behaviour of in particular the single-particle Green functions $`G_r`$ pertaining to the model under consideration; <sup>\*‡‡</sup><sup>\*‡‡</sup>\*‡‡ By imposing a ‘sharp’ cut-off on $`𝗇_r(𝐤)`$, the long-distance behaviour of $`G_r`$ coincides with that specific to FLs (Farid 1999b); such a behaviour is incorrect as long as $`\alpha 0`$. for this long-distance asymptotic behaviour to be described correctly, it is necessary that $`𝗇_+(𝐤)`$ and $`𝗇_{}(𝐤)`$ are reduced to zero for $`𝐤\lambda /2`$, which can be achieved through imposing a ‘soft’ cut-off on these functions which becomes operative for $`|𝐤|`$ larger than $`\lambda /2`$ (Farid 1999b).
Through decomposing the integral in Eq. (61) into two integrals involving separately $`𝗇_r(𝐤𝐪/2)`$ and $`𝗇_r(𝐤+𝐪/2)`$ and subsequently transforming the integration variable $`𝐤`$, with $`𝐤𝐤+𝐪/2`$ in the first and $`𝐤𝐤𝐪/2`$ in the second integral, cancellation of a common contribution to the resulting integrals gives rise to $`g_r(𝐪)=_{\lambda /2+rk_F}^{\lambda /2+rk_F+𝐪}d𝐤𝗇_r(𝐤)_{\lambda /2+rk_F𝐪}^{\lambda /2+rk_F}d𝐤𝗇_r(𝐤)`$. By assuming $`\lambda `$ to be sufficiently large (see above), $`𝗇_r(𝐤)`$ in the first integral can be replaced by unity (zero) and in the second integral by zero (unity) for $`r=+`$ ($`r=`$), upon which one readily obtains
$$g_r(𝐪)=r𝐪.$$
(62)
From this and Eq. (61) it follows that
$$\mathrm{\Pi }_{rr}(q)=\frac{1}{2\pi }\frac{r𝐪}{iq_0𝐯_r^0𝐪}\mathrm{\Pi }_{rr}^0(q),$$
(63)
where $`\mathrm{\Pi }_{rr}^0(q)`$ stands for the density-density correlation function pertaining to the non-interacting system. This result is a direct manifestation of the CLT (DL 1974, Bohr 1981) which applies to the one-dimensional Luttinger model. Note that validity of Eq. (62) does not depend on any other aspect of $`𝗇_r(𝐤)`$ than it be zero (unity) for $`r=+`$ ($`r=`$) when $`𝐤`$ is sufficiently far to the right of $`rk_F`$ and unity (zero) for $`r=+`$ ($`r=`$) when $`𝐤`$ is sufficiently far to the left of $`rk_F`$. Thus Eq. (62) holds also for the case where $`𝗇_r(𝐤)`$ in Eq. (61) is replaced by its non-interacting counter-part $`\mathrm{\Theta }(k_Fr𝐤)`$; using this step function, the validity of this statement is trivially verified by explicit calculation.
Let us now assume $`|𝐪|`$ to be small. Using (inappropriately) $`𝗇_r(𝐤𝐪/2)𝗇_r(𝐤+𝐪/2)=𝐪𝗇_r(𝐤)/𝐤+𝒪(|𝐪|^3)`$ for all $`𝐤`$ (c.f. Eq. (16) above and Eq. (75) below), through integration-by-parts and making use of $`𝗇_r(\lambda /2+rk_F𝐪/2)=0(1)`$ and $`𝗇_r(\lambda /2+rk_F+𝐪/2)=1(0)`$ for $`r=+`$ ($`r=`$), up to $`𝒪(|𝐪|^3)`$ we evidently obtain the (exact) result in Eq. (62), in spite of the fact that $`𝗇_r(𝐤)`$ is not a differentiable function of $`𝐤`$ in a neighbourhood of $`𝐤=rk_F`$.
We shall now demonstrate that $`𝒪(|𝐪|^3)`$ in $`g_r(𝐪)=r𝐪+𝒪(|𝐪|^3)`$ is incorrect. The remainder in fact is of the form $`𝒪(|𝐪|^{1+\alpha })`$. Further, as the non-analytic form of this remainder suggests, the second- and higher-order expansions of $`𝗇_r(𝐤𝐪/2)𝗇_r(𝐤+𝐪/2)`$ (in powers of $`𝐪`$) result in divergent integrals, reflecting the fact that $`1<1+\alpha <2`$.
Below we consider $`_{\lambda _0/2+rk_F+𝐪/2}^{\lambda _0/2+rk_F𝐪/2}d𝐤\left[𝗇_r(𝐤𝐪/2)𝗇_r(𝐤+𝐪/2)\right]_{\lambda _0/2+𝐪/2}^{\lambda _0/2𝐪/2}d𝐤\left[𝗇_r(𝐤+rk_F𝐪/2)𝗇_r(𝐤+rk_F+𝐪/2)\right]`$ in terms of the following three constituent parts
$`_1`$ $`:=`$ $`{\displaystyle _{\lambda _0/2+𝐪/2}^{𝐪/2}}d𝐤\left[𝗇_r(𝐤+rk_F𝐪/2)𝗇_r(𝐤+rk_F+𝐪/2)\right],`$ (64)
$`_2`$ $`:=`$ $`{\displaystyle _{𝐪/2}^{𝐪/2}}d𝐤\left[𝗇_r(𝐤+rk_F𝐪/2)𝗇_r(𝐤+rk_F+𝐪/2)\right],`$ (65)
$`_3`$ $`:=`$ $`{\displaystyle _{𝐪/2}^{\lambda _0/2𝐪/2}}d𝐤\left[𝗇_r(𝐤+rk_F𝐪/2)𝗇_r(𝐤+rk_F+𝐪/2)\right],`$ (66)
where $`\lambda _0`$ is a constant which we chose to satisfy $`|𝐪|<\lambda _0\lambda `$. Our choice of the inequality $`\lambda _0\lambda `$ is motivated by the fact that it justifies our use of the following two-term asymptotic expression for $`𝗇_r(𝐤)`$ pertaining to the one-dimensional Luttinger model for spin-less fermions (Voit 1994)
$$𝗇_r(𝐤+rk_F)\frac{1}{2}rC\mathrm{sgn}(𝐤)|𝐤|^\alpha ,$$
(67)
where $`C`$ is a positive constant. Making use of the expression in Eq. (67), through exact evaluation of the integrals one obtains (below we assume $`𝐪>0`$):
$`_1=_3`$ $``$ $`\{\begin{array}{cc}\frac{rC}{\alpha +1}\left((\lambda _0/2)^{\alpha +1}(\lambda _0/2𝐪)^{\alpha +1}𝐪^{\alpha +1}\right),\hfill & \text{(No expansion)}\hfill \\ rC\left((\lambda _0/2𝐪/2)^\alpha (𝐪/2)^\alpha \right)𝐪,\hfill & \text{(Expansion)}\hfill \end{array}`$ (70)
$`_2`$ $``$ $`\{\begin{array}{cc}\frac{2rC}{\alpha +1}𝐪^{\alpha +1},\hfill & \text{(No expansion)}\hfill \\ \frac{2^{1\alpha }(\alpha +2)rC}{\alpha +1}𝐪^{\alpha +1}.\hfill & \text{(Expansion)}\hfill \end{array}`$ (73)
The dissimilarity of the results based upon the first-order expansion of $`𝗇_r(𝐤+rk_F𝐪/2)𝗇_r(𝐤+rk_F+𝐪/2)`$, indicated by “Expansion”, from the exact results, indicated by “No expansion”, confirms our above statement concerning the invalidity of the expansion procedure. Note for instance that $`_2`$ as determined through expansion coincides with the exact result only when $`\alpha =2`$. It is interesting also to note that contrary to the expression $`𝗇_r(𝐤𝐪/2)𝗇_r(𝐤+𝐪/2)=𝐪𝗇_r(𝐤)/𝐤+𝒪(|𝐪|^3)`$ which implies absence of a zeroth-order term in $`𝐪`$, from Eq. (67) one in fact has (c.f. Eq. (16) above)
$`𝗇_r(𝐤𝐪/2)𝗇_r(𝐤+𝐪/2)`$ $``$ $`2rC\mathrm{sgn}(𝐪)|𝐤rk_F|^\alpha +\alpha rC|𝐤rk_F|^{\alpha 1}𝐪,`$ (75)
$`\text{when}𝐤[rk_F𝐪/2,rk_F+𝐪/2],`$
which contains a zeroth-order term in $`𝐪`$ as long as $`𝐤\overline{)}=rk_F`$. Without taking into account the zeroth-order term in Eq. (75), we would have $`_22^{1\alpha }rC𝐪^{\alpha +1}`$ (Expansion) rather than the ‘(Expansion)’ result for $`_2`$ in Eq. (73); the two results coincide only for $`\alpha =1`$.
Adding the three contributions $`_j`$, $`j=1,2,3`$, as presented in Eqs. (70) and (73), we obtain
$`{\displaystyle _{\lambda _0/2+rk_F+𝐪/2}^{\lambda _0/2+rk_F𝐪/2}}d𝐤\left[𝗇_r(𝐤𝐪/2)𝗇_r(𝐤+𝐪/2)\right]`$ (76)
$`\{\begin{array}{cc}\frac{2rC}{\alpha +1}\left((\lambda _0/2)^{\alpha +1}(\lambda _0/2𝐪)^{\alpha +1}\right),\hfill & \text{(No expansion)}\hfill \\ \frac{2^{1\alpha }(\alpha +2)rC}{\alpha +1}𝐪^{\alpha +1}+2rC\left((\lambda _0/2𝐪/2)^\alpha (𝐪/2)^\alpha \right).\hfill & \text{(Expansion)}\hfill \end{array}`$ (79)
Assuming $`0<𝐪\lambda _0`$, and neglecting $`𝐪^{\alpha +1}`$, both of the results in Eq. (76) reduce to $`2^{1\alpha }rC\lambda _0^\alpha 𝐪`$. This analysis shows that the error in $`g_r(𝐪)`$ as obtained through the first-order expansion of the integrand of $`g_r(𝐪)`$ is of order $`|𝐪|^{\alpha +1}`$ and not of order $`|𝐪|^3`$. As we have indicated above, in practice $`\alpha `$ is small (recall that for the one-dimensional Hubbard model, $`0<\alpha <1/8`$), so that the range of the $`𝐪`$ values for which $`|𝐪|`$ is dominant with respect to $`|𝐪|^{\alpha +1}`$ is substantially reduced as compared with the range for which $`|𝐪|`$ is dominant with respect to $`|𝐪|^3`$. Note in passing that $`|𝐪|^{\alpha +1}=|𝐪|^3`$ provided $`\alpha =2`$ which coincides with the condition required for the equality of the $`_2`$’s as evaluated without relying upon expansion and through (correct) expansion of the integrand of $`_2`$.
In $`d>1`$, contrary to $`d=1`$ (in particular when dealing with the one-dimensional Luttinger model), the radial cut-off parameter $`\lambda `$ cannot be arbitrarily large. Consequently, the condition $`|𝐪|\lambda _0`$ is extremely restrictive, owing to $`\lambda _0\lambda 𝐤_F`$, and therefore should be relaxed. Correspondingly, evaluation of integrals subsequent to an expansion of the pertinent integrands yields results which even in the leading order deviate from the exact results (see Eqs. (70), (73) and (76) above).
Finally, the momentum distribution function in Eq. (67) is peculiar in that $`𝗇_r(𝐤)`$ does not allow for a Taylor expansion in a neighbourhood of $`𝐤=rk_F`$. In the case of the conventional FLs, although $`𝗇(𝐤)`$ is discontinuous at $`𝐤=𝐤_F`$, nevertheless it possesses regular (Taylor) expansions to the left and to the right of $`𝐤=𝐤_F`$. <sup>†\*</sup><sup>†\*</sup>†\* Within the framework of the RPA, $`𝗇(𝐤)`$ pertaining to the uniform system of electrons (in $`d=3`$) has been shown to possess finite left and right derivatives at $`𝐤=𝐤_F`$ (Daniel and Vosko 1960). On the other hand, for a uniform system of electrons interacting through a hard-core two-body potential, Belyakov (1961) (see also Sartor and Mahaux 1980) has obtained an $`𝗇(𝐤)`$ which has logarithmically divergent derivatives to the left and right of $`𝐤=𝐤_F`$. For a discussion of the behaviour of $`𝗇(𝐤)`$ in the close vicinity of $`𝐤=𝐤_F`$ see (Farid 1999a). Consequently, in cases where one has to do with conventional FLs, provided one takes appropriate care of the discontinuity at $`𝐤=𝐤_F`$ of the associated $`𝗇(𝐤)`$, the procedure based upon expansion of $`𝗇(𝐤𝐪/2)𝗇(𝐤+𝐪/2)`$ yields correct results to the expected order. One observes that mathematical procedures that are in general inapplicable, may be applied, with due care, when dealing with conventional FLs (see Farid 1999a). The same procedure fails however, in the case of unconventional FLs or NFLs.
## V Summary and concluding remarks
In this work we have presented a brief outline of the steps that underlie a generalised method which has been modelled on a formalism due to DL (1974) which exactly solves the one-dimensional Luttinger (1963) model. Key elements in both the original DL (1974) method and its generalisation (MCDC 1998) are the conservation laws and the associated Ward identities. These identities give rise to some auxiliary relationships which are exact in the one-dimensional Luttinger model, rendering this model exactly solvable, whereas though not exact, these auxiliary relationships have been asserted to be asymptotically exact for Fermi systems in $`d>1`$ in the limit where the scattering of the electrons in a thin layer circumscribing the Fermi surface is strictly in the forward-scattering channel (corresponding to scattering processes with vanishingly-small momentum transfers, that is $`𝐪\mathrm{𝟎}`$). We have presented an analysis concerning a crucial relationship between the irreducible density and current vertex parts, $`\mathrm{\Lambda }_\sigma ^0(p;q)`$ and $`𝚲_\sigma (p;q)`$ respectively, which is exactly valid in the case of the one-dimensional Luttinger model and asserted (MCDC 1998) to be asymptotically exact in the forward-scattering limit when $`d>1`$. We have explicitly demonstrated failure of this relationship for the conventional FL systems in $`d=2`$. We have further demonstrated that this relationship is exact, in any arbitrary spatial dimension, only in the trivial case of systems of strictly non-interacting fermions. <sup>††</sup><sup>††</sup>†† Recall that metallic states of non-interacting fermions are ideal FLs. From the same analysis it follows that for NFLs in $`d>1`$ there does not exist even the remotest ground on the basis of which the presumed asymptotic validity of the mentioned relationship can be justified: here, the interlocking of energies and momenta in the arguments of the corresponding self-energies, which by the very fact of being by definition non-differentiable in neighbourhoods of the Fermi energy and/or the Fermi momentum (otherwise the associated metallic states would not be NFLs), stands in the way of obtaining a relationship similar in appearance to the indicated supposed asymptotic relationship between the irreducible vertex parts.
We have further explicitly demonstrated that the continuity equation only in $`d=1`$ gives rise to a direct relationship between the vertex parts $`\mathrm{\Lambda }_\sigma ^\nu (p;q)`$ and $`\mathrm{\Gamma }_\sigma ^\nu (p;q)`$, $`\nu =0,1,\mathrm{},d`$, independent of the analytic properties of the momentum distribution function $`𝗇_\sigma (𝐤)`$. In $`d>1`$, continuity as well as continuous differentiability of this function in the above-indicated narrow band of the momentum space circumscribing the Fermi surface must be assumed before a similar direct relationship between the two vertex parts, as in $`d=1`$, can be deduced. As is well known, $`𝗇_\sigma (𝐤)`$ is discontinuous at $`𝐤=𝐤_F`$ both for FLs and certain class of NFLs (for the latter see Farid 1999a). For the case of the one-dimensional Luttinger model, though continuous, $`𝗇_\sigma (𝐤)`$ is not differentiable at $`𝐤=𝐤_F`$. Therefore, by assuming a similar behaviour for $`𝐧_\sigma (𝐤)`$ in $`d>1`$ as in $`d=1`$, we have established that the indicated simplified relationships between $`\mathrm{\Lambda }_\sigma ^\nu (p;q)`$ and $`\mathrm{\Gamma }_\sigma ^\nu (p;q)`$, $`\nu =0,1,\mathrm{},d`$, do not hold in $`d>1`$.
Each of the above-mentioned two findings removes ground for validity in $`d>1`$ of the heretofore supposed asymptotic Ward identities in the forward-scattering limit. This failure concerns both FLs (excluding strictly non-interacting metallic systems) and NFLs.
In an earlier work (Farid 1999b) we have demonstrated failure of the Haldane (1992) bosonisation scheme in $`d>1`$ both on general grounds and on grounds of an explicit calculation on a model of interacting fermions in $`d=2`$. Within the framework of the Haldane bosonisation scheme, the explicit expression for the single-particle Green function $`G`$ takes a simple form in the space-time representation. It is interesting that through replacing a screening function that features in this expression by its random-phase approximation (that is RPA) counterpart, one obtains an expression which coincides with that obtained within the generalised DL framework; compare Eq. (5.65) in (MCDC 1998) with for instance Eq. (24) in (Kwon, Houghton and Marston 1995). The use of the RPA within the framework of the generalised DL method is considered to be justified through reliance upon the so-called generalised CLT (Kopietz, Hermisson and Schönhammer 1996) which implies that while the RPA yields the exact response function in the case of the one-dimensional Luttinger model (DL 1974) (see Eq. (63) above), it yields an asymptotically exact response function in $`d>1`$ in the forward-scattering limit. In (Farid 1999b) we have commented on the generalised CLT, however independent of the true status of this theorem in $`d>1`$, and whether one has to do with FL or NFL systems, from the point of view of our considerations in the present work the coincidence of the two Green functions is of crucial importance as it indirectly establishes that indeed the generalised DL (MCDC 1998) formalism is of no validity for systems of interacting fermions in $`d>1`$.
To summarise the main conclusions arrived at in this and our earlier works (Farid 1999a,b), we can state that none of the presumed non-perturbative methods designed thus far for investigating the existence or otherwise of NFL metallic states in $`d>1`$ has proved capable of dealing with systems of interacting fermion in $`d>1`$, not even in the weakly-interacting limit. Renewed attempts concerning design and application of non-perturbative methods are therefore called for in order to settle the long-standing problem with regard to the existence or otherwise of NFL metallic states in spatial dimensions greater than one. It is therefore befitting to close this work by mentioning that Laughlin’s (1998) recent statement, that “… and all attempts to account for the existence of a Luttinger-liquid at zero temperature in spatial dimension greater than one have failed.”, must needs be qualified and in due course perhaps revoked altogether.
$`\mathrm{}`$
## Acknowledgements
It is a pleasure for me to thank Professor Peter B. Littlewood and members of the Theory of Condensed Matter Group for their kind hospitality at Cavendish Laboratory where this work was completed. I record my indebtedness to Girton College, Cambridge, for invaluable support. I extend my thanks to Dr Peter Kopietz for kindly providing me with references (Kopietz 1997) and (Bohr 1981). I dedicate this work to the memory of Lady Bertha Jeffreys (1903 - 1999).
## A On dimensional continuation and its application to interacting Fermi systems
In this Appendix we first present a brief account of the technique of dimensional continuation (Wilson 1973, appendix). <sup>†‡</sup><sup>†‡</sup>†‡ For a comprehensive review of this topic see Collins (1984, Chapter 4). Subsequently, we point out that the formalism of interacting fermions even in the limit of strong forward scattering cannot be subjected to the conventional process of dimensional continuation. We consequently propose an unconventional approach to this process which involves one continuous free parameter and therefore defines an infinite family of continuations that correctly interpolate between $`d=1`$ and $`d=2`$; they also correctly interpolate between $`d=2`$ and $`d=3`$ for the class of cylindrically symmetric integrands.
Consider
$$𝒥_d(𝐪):=_S\mathrm{d}^d𝐤f_𝐪(𝐤),$$
(A1)
where $`d`$ is some real or complex number. Let $`\{𝐞_1,\mathrm{},𝐞_J\}`$ be an orthonormal basis spanning the so-called parallel space (to whose complimentary space we refer as tangent space), large enough to accommodate the vector $`𝐪`$ (in general, all external momenta $`\{𝐪_j\}`$ upon which the integrand may depend). We have
$$SS_{}S_{}.$$
(A2)
With
$$𝐤=𝐤_{}+𝐤_{}\underset{j=1}{\overset{J}{}}k_{}^j𝐞_j+𝐤_{},$$
(A3)
one can therefore write
$$𝒥_d(𝐪)=_S_{}dk_{}^1\mathrm{}dk_{}^J_S_{}\mathrm{d}^{dJ}𝐤_{}f_𝐪(𝐤).$$
(A4)
By assuming independence of $`f_𝐪(𝐤)`$ upon direction of $`𝐤_{}`$, i.e. by assuming
$$f_𝐪(𝐤)\stackrel{~}{f}_𝐪(𝐤𝐤,𝐤𝐪),$$
(A5)
one has
$$_S_{}\mathrm{d}^{dJ}𝐤_{}f_𝐪(𝐤)=K_{dJ}_0^{\mathrm{}}\mathrm{d}𝐤_{}𝐤_{}^{dJ1}\stackrel{~}{f}_𝐪(𝐤𝐤,𝐤𝐪),$$
(A6)
where
$$K_\nu :=\frac{2\pi ^{\nu /2}}{\mathrm{\Gamma }(\nu /2)},$$
(A7)
with $`\mathrm{\Gamma }(z)`$ the gamma function. The value of $`K_\nu `$ in Eq. (A7) is equal to the area of the hyper-sphere of unit radius in $`\nu `$ dimensions. The $`J+1`$ integrals over $`k_{}^1,\mathrm{},k_{}^J`$ and $`𝐤_{}`$ are conventional and are evaluated in the usual way. It should be evident that the key element for reducing the abstract integral in Eq. (A1) into a conventional multi-dimensional integral is the assumption specified by the expression in Eq. (A5). Below we explicitly consider two cases which clarify the role played by the dimension of the parallel space, $`J`$.
Case I: Let $`J=1`$. Let further $`\theta `$ denote the planar angle between $`𝐤`$ and $`𝐪`$. With
$$𝐤_{}=𝐤\mathrm{sin}\theta ,k_{}^1=𝐤\mathrm{cos}\theta ,\theta [0,\pi ],$$
(A8)
one has $`\mathrm{d}k_{}^1\mathrm{d}𝐤_{}=\mathrm{d}\theta \mathrm{d}𝐤𝐤`$, $`𝐤_{}^{d2}=𝐤^{d2}\left(\mathrm{sin}\theta \right)^{d2}`$ and thus (Wilson 1973)
$`𝒥_d(𝐪)`$ $`=`$ $`{\displaystyle _{\mathrm{}}^{\mathrm{}}}dk_{}^1K_{d1}{\displaystyle _0^{\mathrm{}}}\mathrm{d}𝐤_{}𝐤_{}^{d2}\stackrel{~}{f}_𝐪(𝐤𝐤,𝐤𝐪)`$ (A9)
$``$ $`K_{d1}{\displaystyle _0^\pi }d\theta {\displaystyle _0^{\mathrm{}}}\mathrm{d}𝐤𝐤^{d1}\left(\mathrm{sin}\theta \right)^{d2}\stackrel{~}{f}_𝐪(𝐤^2,𝐤𝐪\mathrm{cos}\theta ).`$ (A10)
Since $`K_{d1}\left(\mathrm{sin}\theta \right)^{d2}\delta (\theta )+\delta (\theta \pi )`$ for $`d1`$, it follows that in the limit $`d1`$ the expression on the RHS of Eq. (A9) recovers the pertinent expression for $`d=1`$, i.e.
$`𝒥_d(𝐪)|_{d1}`$ $`=`$ $`{\displaystyle _0^{\mathrm{}}}\mathrm{d}𝐤\stackrel{~}{f}_𝐪(𝐤^2,𝐤𝐪)+{\displaystyle _0^{\mathrm{}}}\mathrm{d}𝐤\stackrel{~}{f}_𝐪(𝐤^2,𝐤𝐪)`$ (A11)
$``$ $`{\displaystyle _{\mathrm{}}^{\mathrm{}}}dk\stackrel{~}{f}_𝐪(k^2,k𝐪).`$ (A12)
For $`d=2`$, for which $`K_{d1}=2`$ holds, we have
$`𝒥_d(𝐪)|_{d=2}`$ $`=`$ $`2{\displaystyle _0^\pi }d\theta {\displaystyle _0^{\mathrm{}}}\mathrm{d}𝐤𝐤\stackrel{~}{f}_𝐪(𝐤^2,𝐤𝐪\mathrm{cos}\theta )`$ (A13)
$``$ $`{\displaystyle _0^{2\pi }}d\phi {\displaystyle _0^{\mathrm{}}}\mathrm{d}𝐤𝐤\stackrel{~}{f}_𝐪(𝐤^2,𝐤𝐪\mathrm{cos}\phi ).`$ (A14)
The latter result, whose validity can be readily verified, <sup>†§</sup><sup>†§</sup>†§ To this end employ the variable transformation $`\phi u`$ defined through $`u=\mathrm{cos}\phi `$, implying $`\mathrm{d}\phi =\mathrm{d}u/(1u^2)^{1/2}`$ for $`\phi [0,\pi ]`$ and $`\phi [\pi ,2\pi ]`$ respectively, and a subsequent reverse transformation $`u\phi `$. makes explicit that for $`d=2`$ one indeed recovers the conventional integral over $`𝐤`$ of $`\stackrel{~}{f}_𝐪(𝐤^2,𝐤𝐪)`$ in terms of the cylindrical-polar coordinates $`(𝐤,\phi )`$, with $`𝐪`$ the polar axis. It should be noted however, that not all two-dimensional integrals can be viewed as the $`d2`$ limit of the expression on the RHS of Eq. (A9).
Case II: Let $`J=2`$, so that $`\{𝐞_1,𝐞_2\}`$ spans the parallel space. With $`\phi `$ denoting the planar angle between $`𝐤_{}`$ and $`𝐞_1`$ (which without loss of generality we assume to be along $`𝐪`$) and $`\theta `$ that between $`𝐤`$ and $`𝐤_{}`$ (i.e. $`𝐤_{}`$ is taken as the polar axis), one has
$$𝐤_{}=𝐤\mathrm{cos}\theta ,k_{}^1=𝐤\mathrm{sin}\theta \mathrm{cos}\phi ,k_{}^2=𝐤\mathrm{sin}\theta \mathrm{sin}\phi ,\theta [0,\pi /2],\phi [0,2\pi ],$$
(A15)
so that $`\mathrm{d}k_{}^1\mathrm{d}k_{}^2\mathrm{d}𝐤_{}=\mathrm{d}\phi \mathrm{d}\theta \mathrm{d}𝐤𝐤^2\mathrm{sin}\theta `$, $`𝐤_{}^{d3}=𝐤^{d3}\left(\mathrm{cos}\theta \right)^{d3}`$ and thus
$`𝒥_d(𝐪)`$ $`=`$ $`{\displaystyle _{\mathrm{}}^{\mathrm{}}}dk_{}^1dk_{}^2K_{d2}{\displaystyle _0^{\mathrm{}}}\mathrm{d}𝐤_{}𝐤_{}^{d3}\stackrel{~}{f}_𝐪(𝐤𝐤,𝐤𝐪)`$ (A16)
$`=`$ $`K_{d2}{\displaystyle _0^{2\pi }}d\phi {\displaystyle _0^{\pi /2}}d\theta \left(\mathrm{cos}\theta \right)^{d3}\mathrm{sin}\theta `$ (A18)
$`\times {\displaystyle _0^{\mathrm{}}}\mathrm{d}𝐤𝐤^{d1}\stackrel{~}{f}_𝐪(𝐤^2,𝐤𝐪\mathrm{sin}\theta \mathrm{cos}\phi ).`$
Since $`K_{d2}\left(\mathrm{cos}\theta \right)^{d3}\delta (\theta \pi /2)`$ for $`d2`$, it follows that
$$𝒥_d(𝐪)|_{d2}=_0^{2\pi }d\phi _0^{\mathrm{}}\mathrm{d}𝐤𝐤\stackrel{~}{f}_𝐪(𝐤^2,𝐤𝐪\mathrm{cos}\phi ),$$
(A19)
which is indeed the expected expression for a two-dimensional integral in the cylindrical-polar coordinates $`(𝐤,\phi )`$. However, again as in the case considered above (concerning $`J=1`$), not all two-dimensional integrals over arbitrary functions of $`𝐤`$ can be obtained as the $`d2`$ limit of the $`d`$-dimensional integral in Eq. (A16).
The above considerations make explicit that whereas by an appropriate choice of the dimension of the parallel space one can analytically continue $`d`$ and obtain the standard integrals for integer dimensions, nonetheless the applicability of this procedure in an essential way depends on the requirement that $`f_𝐪(𝐤)`$ in so far as its dependence upon $`𝐤`$ is concerned, be a function of at most $`𝐤𝐤`$ and $`𝐤𝐪`$. In cases where the dependence of $`f_𝐪(𝐤)`$ upon $`𝐤`$ is more general, the conventional process of dimensional continuation may be possible in the limit of small $`𝐪`$. In this limit, provided $`f_𝐪(𝐤)`$ be continuously differentiable in a neighbourhood of $`𝐪=\mathrm{𝟎}`$, it may be possible (that is, under the condition that $`f_𝐪(𝐤)`$ to linear order in $`𝐪`$ can be described by a function of $`𝐤𝐤`$ and $`𝐤𝐪`$) to carry through the process of dimensional continuation, accurate to linear order in $`𝐪`$. <sup>†¶</sup><sup>†¶</sup>†¶ We note that here we only deal with scalar functions ($`f_𝐪(𝐤)`$) which are functions of scalar products, however, the technique of dimensional continuation is also applicable to tensor functions. See Wilson (1973, appendix) and Collins (1984, Chapter 4). This is precisely the case in the generalised DL formalism by MCDC (1998) (see the last part of § II).
It is readily verified that the following expression, whose origin does not lie in the conventional formalism for dimensional continuation (Wilson 1973), is a smooth analytic continuation for $`d(1,2]`$:
$`𝒥_d(𝐪)`$ $`:=`$ $`{\displaystyle \frac{K_{d1}}{\vartheta (d)}}\{{\displaystyle _0^\pi }\mathrm{d}\theta \left(\mathrm{sin}\theta \right)^{d2}{\displaystyle _0^{\mathrm{}}}\mathrm{d}𝐤𝐤^{d1}\overline{f}_𝐪(𝐤,\theta )`$ (A21)
$`+{\displaystyle _\pi ^{\vartheta (d)\pi }}\mathrm{d}\theta \left(\mathrm{sin}(\theta \pi )\right)^{d2}{\displaystyle _0^{\mathrm{}}}\mathrm{d}𝐤𝐤^{d1}\overline{f}_𝐪(𝐤,\theta )\},`$
where
$$\overline{f}_𝐪(𝐤,\theta )f_𝐪(𝐤),$$
(A22)
with $`(𝐤,\theta )`$ the cylindrical-polar coordinates of $`𝐤`$, and (see Figure)
$$\vartheta (d):=1+(3d)\mathrm{exp}\left(a\left[11/(d1)^{1/a}\right]\right),\text{with}a>0.$$
(A23)
We note that the integral in Eq. (A21) together with $`\vartheta (d)`$ in Eq. (A23) is also correct for $`d=3`$ over the space of functions with cylindrical symmetry; therefore it is also a smooth continuation for $`d[2,3]`$, however over a restricted set of functions. In this respect Eq. (A21) is of limited validity for $`d>2`$, in the same way that Eq. (A9) is of limited validity for $`d>1`$.
The choice for $`\vartheta (d)`$ in Eq. (A23) is not unique. However, in choosing any $`\vartheta (d)`$ interpolating between $`\vartheta (1)=1`$ and $`\vartheta (2)=2`$ (and possibly $`\vartheta (3)=1`$, etc.) care must be exercised that in the limit $`d1`$ the singularities of $`K_{d1}\left(\mathrm{sin}\theta \right)^{d2}`$ and $`K_{d1}\left(\mathrm{sin}(\theta \pi )\right)^{d2}`$ for respectively $`\theta \pi `$ and $`\theta \pi `$ do not yield repeated contributions. This is prevented by the requirement $`\left(\vartheta (d)1\right)^{d1}0`$ for $`d1`$; in consequence of this, for $`d1`$ the reduction in the width of the interval $`[\pi ,\vartheta (d)\pi ]`$ renders the growth in the integrand, due to the singularity of $`K_{d1}\left(\mathrm{sin}(\theta \pi )\right)^{d2}`$ at $`d=1`$, ineffective. It can be easily verified that $`\vartheta (d)`$ in Eq. (A23) indeed satisfies this requirement. The freedom in choice of the continuous positive parameter $`a`$ in Eq. (A23) implies that the dimensional continuation scheme as prescribed by Eqs. (A21) and (A23) represents an infinite family of dimensional continuations. However, in order to recover the expression in Eq. (A9), applicable to functions $`\{f_𝐪(𝐤)\}`$ whose dependence upon $`𝐤`$ is through $`𝐤𝐤`$ and $`𝐤𝐪`$, it is required that $`0<a1`$ (see Figure); in the limit $`a0`$ the expressions in Eqs. (A9) and (A21) coincide. $`\mathrm{}`$
|
warning/0004/nucl-th0004054.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
What happens to QCD at large density is a fascinating topic which is generating an intense activity in nuclear and particle communities. It seems that at an asymptotic density, the likely scenario is that diquarks condense into the matter as a consequence of which color superconductivity (CSC) may take place. For a recent review with references, see . But it is not at all clear whether the density involved for CSC is relevant to the dense matter that can be accessed by experiments and that we are interested in, namely neutron stars and heavy ion collisions that are currently studied or planned for the future. The question we would like to address here is: What happens to nuclear matter as density increases beyond the normal matter density toward the critical density for chiral phase transition? This question is not easy to answer since the coupling involved is not weak enough as what might happen at asymptotic density at which QCD weak-coupling property can be exploited but it is definitely more relevant to the physics we want to study in conjunction with experiments. We expect a priori that a variety of different phenomena will intervene in the regime of density that we are interested in and these may not render themselves to simple analyses. Some of them were discussed from a weak-coupling QCD point of view in . Here we address them from a different perspective.
The issue in question was studied by Langfeld and Rho using a semi-realistic field theory model of the NJL type. Some of the results found there were novel and suggestive but certain restrictive features of the NJL-type models (i.e., no confinement) rendered them problematic. See for discussions on related matters. Here we would like to analyze the problem using a simplified solvable quantum mechanical model and a two-dimensional field theory model. Our study here is exploratory and we cannot make quantitative statements (for instance, our analysis is “blind” to color and flavor contents) as to what really must happen but our hope is to gain some qualitative understanding of what might be taking place generically at a density greater than that of normal nuclear matter but much less than the asymptotic density relevant to the color superconductivity.
In , Langfeld and Rho suggested that when the isoscalar vector channel (called the $`\omega `$ channel) becomes sufficiently attractive as expected in the presence of collective phonon modes of the pertinent quantum numbers as discussed in , then the baryon density can have a significant jump at some critical chemical potential in conjunction with generating low-mass excitations with the quantum numbers of the $`\omega `$ meson. Since this can be expressed as a rapid increase – but not a discontinuous jump – in the ground state expectation value of the time component of the $`\omega `$ field, the process was called “induced symmetry breaking” (ISB) associated with the broken isoscalar vector symmetry characterized by the “vaccum” expectation value of the time component of the vector current. We should stress that in medium, $`\omega _0`$ is of course never zero, being proportional to density. Therefore $`\omega _0`$ cannot, strictly speaking, be used as a signal for a phase change. However it can have an anomalous increase at certain chemical potential reminiscent of an order parameter that can be associated with a phase transition. A simple condensed matter model that illustrates this phenomenon is discussed by Langfeld . We shall use somewhat abusively this notion in this paper with the caveat in mind.
A similar behavior of the baryon density vs. the chemical potential seen in the ISB has been also observed by Berges et. al in a renormalization-group flow analysis of a linear sigma model in which it was found that quark number density shows non-analytic behavior at $`\mu =1.025M_q`$, with $`M_q316.2MeV`$ an effective constituent quark mass, where the vacuum expectation value of the $`\sigma `$-field, chiral condensate, vanishes. As discussed in ref., this is easy to understand in terms of change in the number distribution at that $`\mu `$. The change in the number distribution can give a jump in the number density through $`\rho 𝑑k[\mathrm{exp}\beta (ϵ(k)\mu )+1]^1`$. That there is a rapid increase in density at the transition point is intrinsic in the chiral phase transition. The point of ref., however, was that the ISB is concurrent with chiral symmetry restoration and that the presence of the ISB could postpone color superconductivity until the $`\omega `$-channel becomes ineffective at some high density due to weakened QCD coupling. At what density this can happen, we cannot say but the density regime involved could be relevant to the interior of compact stars.
In this paper, we wish to address the problem in two aspects. First we take a quantum mechanical model studied by Ilieva and Thirring (IT) and study, exploiting its solvability, whether and how the ISB manisfests itself in the model. As stressed by IT, this analysis does not depend upon the dimension of the space considered. We shall discuss in section 2 the ISB phenomenon in the IT model and find that the model can have a phase in which both the ISB and fermion pair (Cooper) condensate co-exist and find that the ground state energy of the mixed phase is indeed lower than the normal state without such condensates. This model does not however render itself to a discussion of what happens with chiral (fermion-antifermion) condensates. To investigate whether this kind of mixed phase is possible and also how the chiral condensate figures in that phase, we resort to a soluble field theory model in two space-time dimensions studied by Chodos et al and examine the effective potentials to see whether both condensates can coexist in the global minimum. The two-dimensional model is subject to the Mermin-Wagner-Coleman no-go theorem for spontaneous symmetry breaking but we follow ref. in considering this model in the large $`N`$ sense a la Witten . It would be possible to formulate the problem in effective two dimensions dimensionally reduced from four dimensions, thereby avoiding the Mermin-Wagner-Coleman theorem. We find in section 3 that although gap equations exist in which both condensates are non-zero, the global minimum of the effective potential always occurs for the case when one or the other condensate vanishes as long as $`\overline{\psi }\psi `$ and $`\psi ^{}\psi `$ are concerned. On the other hand, we do find a stable mixed phase consisting of $`\psi \psi `$ and $`\psi ^{}\psi `$ at some high density. Section 4 contains concluding remarks. In Appendix we clarify the notion of grand canonical ensemble used in this paper and in .
## 2 Ilieva-Thirring Model and the ISB
Ilieva and Thirring studied the competition between $`\psi ^{}\psi `$ and $`\psi \psi `$ in a solvable quantum mechanical model (that we shall refer to as IT). Here we revisit their discussion in light of our objective as defined in the previous section. Since we shall use the notation and the results of IT, we first summarize them and then extract the information relevant to us.
It was argued in that two Hamiltonians are equivalent if they lead to the same time evolution of the local observables. This means that the effective Hamiltonian $`H_B`$
$`H_B={\displaystyle 𝑑p\{a^{}(p)a(p)[\omega (p)+\mathrm{\Delta }_M(p)]+\frac{1}{2}\mathrm{\Delta }_B(p)[a^{}(p)a^{}(p)+a(p)a(p)]\}}`$ (1)
is equivalent to $`H=H_{kin}+V_B+V_M`$,
$`H`$ $`=`$ $`H_{kin}+V_B+V_M`$
$`V_B`$ $`=`$ $`\kappa ^{3/2}{\displaystyle 𝑑p𝑑p^{}𝑑q𝑑q^{}a^{}(q)a^{}(q^{})a(p)a(p^{})v_B(q,q^{},p,p^{})e^{\kappa (p+p^{})^2\kappa (q+q^{})^2}}`$
$`V_M`$ $`=`$ $`\kappa ^{3/2}{\displaystyle 𝑑p𝑑p^{}𝑑q𝑑q^{}a^{}(q)a^{}(q^{})a(p)a(p^{})v_M(q,q^{},p,p^{})e^{\kappa (qp)^2\kappa (q^{}p^{})^2}}`$
where $`V_B`$ and $`V_M`$ support respectively the pairing gap $`\mathrm{\Delta }_B`$ and the “mean-field gap” $`\mathrm{\Delta }_M`$ for $`\kappa \mathrm{}`$, provided the gap equations
$`[a(k),V_B+V_M]=[a(k),H_BH_{kin}]`$ (2)
are satisfied. To solve the gap equations, IT choose the following potential describing an interaction concentrated about the Fermi surface
$`v_{B,M}(\stackrel{}{k},\stackrel{}{p})=\lambda _{B,M}S(\stackrel{}{k})S(\stackrel{}{p})`$
$`\mathrm{with}S(\stackrel{}{k})`$ $`=`$ $`{\displaystyle \frac{1}{2ϵ}}[\mathrm{\Theta }(|\stackrel{}{k}|\sqrt{\mu }+ϵ)\mathrm{\Theta }(|\stackrel{}{k}|\sqrt{\mu }ϵ)]`$ (3)
where $`\mathrm{\Theta }(x)`$ is the Heaviside function. With this potential and the additional assumption $`\omega (p)=p^2=\mu `$ (with $`2m=1`$), the gap equations read
$`{\displaystyle \frac{1}{2}}\mathrm{\Delta }_M(\mu )`$ $`=`$ $`\lambda _M\{{\displaystyle \frac{c^2(\mu )}{1+e^{\beta (\overline{W}\mu )}}}+{\displaystyle \frac{s^2(\mu )}{1+e^{\beta (\overline{W}\mu )}}}\}`$ (4)
$`\overline{W}`$ $`=`$ $`\lambda _B\mathrm{tanh}{\displaystyle \frac{\beta (\overline{W}\mu )}{2}}\mathrm{or}\mathrm{\Delta }_B=0`$ (6)
$`\overline{W}=\sqrt{[\mu +\mathrm{\Delta }_M]^2+\mathrm{\Delta }_B^2}`$
(where the pairing gap equation can have two solutions (6), the former non-trivial and the latter trivial) with the subsidiary conditions
$`c^2(\mu )+s^2(\mu )`$ $`=`$ $`1,`$
$`c^2(\mu )s^2(\mu )`$ $`=`$ $`(\omega (\mu )+\mathrm{\Delta }_M)/\overline{W},`$
$`2c(\mu )s(\mu )`$ $`=`$ $`\mathrm{\Delta }_B(\mu )/\overline{W}`$ (7)
where $`c`$ and $`s`$ are real coefficients. It follows from (4) that $`\mathrm{sgn}\mathrm{\Delta }_M=\mathrm{sgn}\lambda _M`$. It has been assumed that $`\overline{W}>0`$. From (1), we have the following thermal expectation value of the number density operator $`a^{}a`$,
$`a^{}(p)a(p^{})=\delta (pp^{})\{{\displaystyle \frac{c^2(p)}{1+e^{\beta (\overline{W}\mu )}}}+{\displaystyle \frac{s^2(p)}{1+e^{\beta (\overline{W}\mu )}}}\}.`$ (8)
### 2.1 Ground-State Energy
Let us now reformulate the IT model by performing Bogoliubov transformation explicitly and evaluate the ground-state energy which is related to the “effective energy” from which we can derive the same quantities. We write the reduced Hamitonian in the form
$`H={\displaystyle \underset{𝐤\sigma }{}}ϵ_𝐤a_{𝐤\sigma }^{}a_{𝐤\sigma }+{\displaystyle \underset{\mathrm{𝐤𝐥}}{}}V_{\mathrm{𝐤𝐥}}^Ba_𝐤^{}a_𝐤^{}a_𝐥a_𝐥+{\displaystyle \underset{\mathrm{𝐤𝐥}}{}}V_{\mathrm{𝐤𝐥}}^Ma_𝐤^{}a_𝐥^{}a_𝐤a_𝐥.`$ (9)
Introduce “order parameters” $`d_𝐤`$ and $`m_𝐤`$ by writing
$`a_𝐤a_𝐤`$ $`=`$ $`d_𝐤+(a_𝐤a_𝐤d_𝐤)`$
$`a_𝐤^{}a_𝐤`$ $`=`$ $`m_𝐤+(a_𝐤^{}a_𝐤m_𝐤)`$ (10)
and define
$`\mathrm{\Delta }_𝐤`$ $``$ $`{\displaystyle \underset{𝐥}{}}V_{\mathrm{𝐤𝐥}}^Bd_𝐥={\displaystyle \underset{𝐥}{}}V_{\mathrm{𝐤𝐥}}^Ba_𝐥a_𝐥`$
$`\overline{\mathrm{\Delta }}_𝐤`$ $``$ $`{\displaystyle \underset{𝐥}{}}V_{\mathrm{𝐤𝐥}}^Mm_𝐥={\displaystyle \underset{𝐥}{}}V_{\mathrm{𝐤𝐥}}^Ma_𝐥^{}a_𝐥.`$ (11)
In terms of $`\mathrm{\Delta }_𝐤`$ and $`\overline{\mathrm{\Delta }}_𝐤`$, the reduced Hamiltonian (9) becomes
$`H`$ $`=`$ $`{\displaystyle \underset{𝐤\sigma }{}}ϵ_𝐤a_{𝐤\sigma }^{}a_{𝐤\sigma }{\displaystyle \underset{𝐤}{}}(\mathrm{\Delta }_𝐤a_𝐤^{}a_𝐤^{}+\mathrm{\Delta }_𝐤^{}a_𝐤a_𝐤\mathrm{\Delta }_𝐤d_𝐤^{})`$ (12)
$`+{\displaystyle \underset{𝐤}{}}(2\overline{\mathrm{\Delta }}_𝐤a_𝐤^{}a_𝐤\overline{\mathrm{\Delta }}_𝐤m_𝐤).`$
To diagonalize the Hamiltonian, we make use of the Bogoliubov-Valatin canonical transformation with real coefficients
$`a_𝐤=c_𝐤b_𝐤+s_𝐤b_𝐤^{}`$
$`a_𝐤=c_𝐤b_𝐤s_𝐤b_𝐤^{}.`$ (13)
Substituting these new operators into (12), we obtain
$`H`$ $`=`$ $`{\displaystyle \underset{𝐤}{}}ϵ_𝐤[(c_𝐤^2s_𝐤^2)(b_𝐤^{}b_𝐤+b_𝐤^{}b_𝐤)+2c_𝐤s_𝐤b_𝐤^{}b_𝐤^{}+2c_𝐤s_𝐤b_𝐤b_𝐤+2s_𝐤^2]`$ (14)
$`+{\displaystyle \underset{𝐤}{}}[c_𝐤s_𝐤(\mathrm{\Delta }_𝐤+\mathrm{\Delta }_𝐤^{})(b_𝐤^{}b_𝐤+b_𝐤^{}b_𝐤1)(\mathrm{\Delta }_𝐤c_𝐤^2\mathrm{\Delta }_𝐤^{}s_𝐤^2)b_𝐤^{}b_𝐤^{}`$
$`+(\mathrm{\Delta }_𝐤s_𝐤^2\mathrm{\Delta }_𝐤^{}c_𝐤^2)b_𝐤b_𝐤+\mathrm{\Delta }_𝐤d_𝐤^{}]+{\displaystyle }_𝐤[2\overline{\mathrm{\Delta }}_𝐤(c_𝐤^2b_𝐤^{}b_𝐤+c_𝐤s_𝐤b_𝐤^{}b_𝐤^{}`$
$`+c_𝐤s_𝐤b_𝐤b_𝐤+s^2b_𝐤b_𝐤^{})\overline{\mathrm{\Delta }}_𝐤m_𝐤].`$
The diagonalization is effected by demanding that the coefficients of $`b_𝐤^{}b_𝐤^{}`$ and $`b_𝐤b_𝐤`$ vanish
$`2c_𝐤s_𝐤ϵ_𝐤+\mathrm{\Delta }_𝐤^{}s_𝐤^2\mathrm{\Delta }_𝐤c_𝐤^2+2\overline{\mathrm{\Delta }}_𝐤c_𝐤s_𝐤=0.`$ (15)
Multiplying $`\mathrm{\Delta }_𝐤^{}/c_𝐤^2`$ and defining $`x=\frac{s_𝐤}{c_𝐤}\mathrm{\Delta }_𝐤^{}`$, we get two solutions
$`x=(ϵ_𝐤+\overline{\mathrm{\Delta }}_𝐤)\pm \sqrt{(ϵ_𝐤+\overline{\mathrm{\Delta }}_𝐤)^2+\mathrm{\Delta }_𝐤^2}.`$ (16)
We take (+) sign here to get a stable minimum energy solution. Then assuming $`\mathrm{\Delta }_𝐤`$ to be a real quantity, we have two equations for $`c_𝐤`$ and $`s_𝐤`$:
$`{\displaystyle \frac{s_𝐤}{c_𝐤}}`$ $`=`$ $`{\displaystyle \frac{E_𝐤(ϵ_𝐤+\overline{\mathrm{\Delta }}_𝐤)}{\mathrm{\Delta }_𝐤}}`$
$`c_𝐤^2+s_𝐤^2`$ $`=`$ $`1`$ (17)
where $`E_𝐤\sqrt{(ϵ_𝐤+\overline{\mathrm{\Delta }}_𝐤)^2+\mathrm{\Delta }_𝐤^2}`$. Solving the equations, we have
$`c_𝐤^2={\displaystyle \frac{1}{2}}(1+{\displaystyle \frac{ϵ_𝐤+\overline{\mathrm{\Delta }}_𝐤}{E_𝐤}})`$
$`s_𝐤^2={\displaystyle \frac{1}{2}}(1{\displaystyle \frac{ϵ_𝐤+\overline{\mathrm{\Delta }}_𝐤}{E_𝐤}}).`$ (18)
To obtain the (coupled) gap equations for $`\mathrm{\Delta }_𝐤`$ and $`\overline{\mathrm{\Delta }}_𝐤`$, we rewrite $`\mathrm{\Delta }_𝐤`$ and $`\overline{\mathrm{\Delta }}_𝐤`$ in terms of the new operators defined in (13), i.e.,
$`\mathrm{\Delta }_𝐤`$ $`=`$ $`{\displaystyle \underset{𝐥}{}}V_{\mathrm{𝐤𝐥}}^Bc_𝐥s_𝐥1b_𝐥^{}b_𝐥b_𝐥^{}b_𝐥`$
$`=`$ $`{\displaystyle \underset{𝐥}{}}V_{\mathrm{𝐤𝐥}}^B{\displaystyle \frac{\mathrm{\Delta }_𝐥}{2E_𝐥}}\mathrm{tanh}{\displaystyle \frac{\beta (E_𝐥\mu )}{2}}`$
$`\overline{\mathrm{\Delta }}_𝐤`$ $`=`$ $`{\displaystyle \underset{𝐥}{}}V_{\mathrm{𝐤𝐥}}^Ms_𝐥^2+c_𝐥^2b_𝐥^{}b_𝐥s_𝐥^2b_𝐥^{}b_𝐥`$ (19)
$`=`$ $`{\displaystyle \underset{𝐥}{}}V_{\mathrm{𝐤𝐥}}^M[{\displaystyle \frac{c_𝐥^2}{e^{\beta (E_𝐥\mu )}+1}}+{\displaystyle \frac{s_𝐥^2}{e^{\beta (E_𝐥\mu )}+1}}]`$
where we have used $`b_{\pm 𝐥}^{}b_{\pm 𝐥}=1/(e^{\beta (E_𝐥\mu )}+1)`$. To compare (19) with the gap equations of IT model, we take $`V_{\mathrm{𝐤𝐥}}^B=2\lambda _B\delta _{|𝐤|,\sqrt{\mu }}\delta _{|𝐥|,\sqrt{\mu }}`$, $`V^M=2\lambda _M\delta _{|𝐤|,\sqrt{\mu }}\delta _{|𝐥|,\sqrt{\mu }}`$ with $`\delta _{a,b}`$ being Kronecker delta, $`\mathrm{\Delta }_𝐤=\mathrm{\Delta }_B(\mu )`$ and $`\overline{\mathrm{\Delta }}_𝐤=\mathrm{\Delta }_M(\mu )`$ and get
$`E(\mu )`$ $`=`$ $`\lambda _B\mathrm{tanh}{\displaystyle \frac{\beta (E(\mu )\mu )}{2}}or\mathrm{\Delta }_B(\mu )=0`$
$`\mathrm{\Delta }_M(\mu )`$ $`=`$ $`2\lambda _M[{\displaystyle \frac{c^2(\mu )}{e^{\beta (E(\mu )\mu )}+1}}+{\displaystyle \frac{s^2(\mu )}{e^{\beta (E(\mu )\mu )}+1}}].`$ (20)
We can easily read off the ground state energy $`U`$ from (14),
$`U`$ $`=`$ $`{\displaystyle \underset{𝐤}{}}[2ϵ_𝐤s_𝐤^22\mathrm{\Delta }_𝐤c_𝐤s_𝐤+\mathrm{\Delta }_𝐤d_𝐤^{}\overline{\mathrm{\Delta }}_𝐤m_𝐤+2\overline{\mathrm{\Delta }}_𝐤s_𝐤^2]`$ (21)
$`=`$ $`\mu E(\mu )+\mathrm{\Delta }_M(\mu )+{\displaystyle \frac{\mathrm{\Delta }_B^2(\mu )}{2\lambda _B}}{\displaystyle \frac{\mathrm{\Delta }_M^2(\mu )}{2\lambda _M}}`$
where the summation has been performed with $`\delta _{|𝐤|,\sqrt{\mu }}`$ in the spirit of the IT model<sup>1</sup><sup>1</sup>1In the usual BCS theory, the summation is replaced by the integration, $`_k_0^{\mathrm{}\omega _c}`$ with $`\mathrm{}\omega _c`$, a typical phonon energy. But since we are considering the very special case of an interaction concentrated about the Fermi surface, we impose the Kronecker delta in the summation.. We are now in a position to analyze various cases of interest.
### 2.2 $`\mathrm{\Delta }_B=0`$
This case corresponds to the simplest solution of gap equations for all values of $`\lambda _M,\lambda _B,\mu `$.
$``$ Case $`I`$-i : $`\mu +\mathrm{\Delta }_M>0`$
In this case,
$`E\mu =\mathrm{\Delta }_M`$ $`=`$ $`{\displaystyle \frac{2\lambda _M}{1+e^{\beta \mathrm{\Delta }_M}}}`$
$`\mathrm{\Delta }_M`$ $``$ $`\{\begin{array}{cc}2\lambda _M& \mathrm{for}\lambda _M<0\\ 0& \mathrm{for}\lambda _M>0\end{array},`$ (24)
as $`\beta \mathrm{}`$. Since the phase with $`\mathrm{\Delta }_M=2\lambda _M`$ has a non-zero positive ground state energy $`U=2\lambda _M`$ for $`\lambda _M<0`$, it does not interest us. We will not consider it anymore. Then, (8) becomes (after integrating over $`p^{}`$)
$`a^{}(p)a(p)`$ $`=`$ $`{\displaystyle \frac{1}{1+e^{\beta (ϵ\mu )}}}\mathrm{for}\mathrm{\Delta }_M=0`$ (25)
where $`ϵ=p^2`$. Note that this phase is nothing but the normal state ( $`\mathrm{\Delta }_B=0\mathrm{and}\mathrm{\Delta }_M=0`$).
$``$ Case $`I`$-ii : $`\mu +\mathrm{\Delta }_M<0`$
To satisfy the condition $`\mu +\mathrm{\Delta }_M<0`$, $`\mathrm{\Delta }_M`$ should be negative and therefore $`\lambda _M<0`$. Noting that $`(ϵ_𝐤+\overline{\mathrm{\Delta }}_𝐤)/E_𝐤=1`$ when we take $`\mathrm{\Delta }_𝐤(\mathrm{\Delta }_B)=0`$ and $`ϵ_𝐤=\mu `$ in (18), we have $`c^2(\mu )=0`$ and $`s^2(\mu )=1`$. Therefore we have
$`E+\mu `$ $`=`$ $`\mathrm{\Delta }_M`$
$`\mathrm{\Delta }_M`$ $`=`$ $`{\displaystyle \frac{2\lambda _M}{1+e^{\beta (\mathrm{\Delta }_M+2\mu )}}}`$ (28)
$``$ $`\{\begin{array}{cc}2\lambda _M& \mathrm{for}\mathrm{\Delta }_M+2\mu <0\\ 0& \mathrm{for}\mathrm{\Delta }_M+2\mu >0\end{array}`$
as $`\beta \mathrm{}`$. Noting that the second solution $`\mathrm{\Delta }_M=0`$ cannot satisfy the condition $`\mu +\mathrm{\Delta }_M<0`$, we have at a given energy $`ϵ(p)=p^2`$
$`a^{}(p)a(p)`$ $`=`$ $`{\displaystyle \frac{1}{1+e^{\beta (\mu |ϵ+\mathrm{\Delta }_M|)}}},ϵ+\mathrm{\Delta }_M<0,`$
$`or`$ $`=`$ $`{\displaystyle \frac{1}{1+e^{\beta (ϵ+\mathrm{\Delta }_M\mu )}}},ϵ+\mathrm{\Delta }_M>0`$ (29)
for $`\mathrm{\Delta }_M=2\lambda _M`$. The corresponding ground state energy $`U=2(\mu +\lambda _M)`$ is negative as long as $`\mu <|\lambda _M|`$ and therefore this phase is energetically favorable compared with the normal state of matter. Since the $`\mathrm{\Delta }_M`$ in (29) corresponds to $`V_0`$ in eq.(2) of ref., one may be tempted to conclude that it is a signal of an ISB. But this can not be a candidate for an ISB because $`\mathrm{\Delta }_M`$ in the IT model cannot be associated with spontaneous breaking of a symmetry and furthermore $`\mathrm{\Delta }_M`$ is a constant independent of the chemical potential.
### 2.3 $`\mathrm{\Delta }_B0`$, $`\lambda _B<0`$
In this case, we cannot expect the Cooper pairs to condensate if $`\lambda _M=0`$ as one can see clearly from the gap equation (6). If we take $`\lambda _M=0`$ and therefore $`\mathrm{\Delta }_M=0`$, the gap equation becomes
$`\sqrt{\mu ^2+\mathrm{\Delta }_B^2}=\lambda _B\mathrm{for}\mathrm{\Delta }_B0\mathrm{and}\beta \mathrm{}`$ (30)
which is contradictory with the conditions we are starting with.
It follows that the parameters under consideration must respect the following conditions :
$`\lambda _B<0,E<\mu ,\mathrm{\Delta }_M<0,\lambda _M<0.`$ (31)
Since we are interested in the physics of cold dense matter, we solve the gap equation taking $`\beta \mathrm{}`$ and get
$`E`$ $`=`$ $`\lambda _B,\mathrm{\Delta }_M=\lambda _M{\displaystyle \frac{\lambda _B\mu }{\lambda _B+\lambda _M}}`$
$`\mathrm{\Delta }_B`$ $`=`$ $`\pm {\displaystyle \frac{\lambda _B}{\lambda _B+\lambda _M}}\sqrt{(\lambda _B\mu )(\lambda _B+2\lambda _M+\mu )}.`$ (32)
Note that we have solutions only when $`\mu >E=\lambda _B`$. To investigate the competition between the mixed phase given by (32) and the $`\mathrm{\Delta }_M0`$ phase in $`I`$-ii, we take the following values of coupling constants with arbitrary dimension
$`S_1`$ $`:`$ $`\lambda _B=5,\lambda _M=5`$
$`S_2`$ $`:`$ $`\lambda _B=5,\lambda _M=10`$
$`S_3`$ $`:`$ $`\lambda _B=5,\lambda _M=2`$ (33)
and find the following critical chemical potential $`\mu _c`$
$`S_1`$ $`:`$ $`\mu _c=5`$
$`S_2`$ $`:`$ $`\mu _c=7.6`$
$`S_3`$ $`:`$ $`\mu _c^1=2,\mu _c^2=5.`$ (34)
In the case of $`S_1`$ and $`S_2`$, our system sits in the phase with $`\mathrm{\Delta }_B=0`$ and $`\mathrm{\Delta }_M0`$, $`I`$-ii, in the regime $`\mu <\mu _c`$ and rolls down to $`\mathrm{\Delta }_B0`$ and $`\mathrm{\Delta }_M0`$ above $`\mu _c`$. The case $`S_3`$ is quite puzzling in the sense that in the region $`\mu <\mu _c^1`$ our system sits in the phase with $`\mathrm{\Delta }_B=0`$ and $`\mathrm{\Delta }_M0`$, for $`\mu _c^1<\mu <\mu _c^2`$, we have the normal phase ($`\mathrm{\Delta }_B=0`$ and $`\mathrm{\Delta }_M=0`$) and in the region $`\mu >\mu _c^2`$, our system sits in the $`\mathrm{\Delta }_B0`$ and $`\mathrm{\Delta }_M0`$ phase. But in all three cases, we recover the normal phase at a high chemical potential, $`S_1:\mu 8.5,S_2:\mu 12.3,S_3:\mu 9.7`$. The phase diagram with the parameter set $`S_1`$ is shown in fig. 1(a). To discuss a possible realization of ISB at the critical potential $`\mu _c`$, we note that in the mixed phase $`U(1)`$ symmetry of the IT model is spontaneously broken to $`Z_2`$ by the superconducting gap $`\mathrm{\Delta }_B`$. Thus one may expect an ISB phenomenon to take place at the critical chemical potential but one should note that chemical potential does not break $`U(1)`$ symmetry explicitly at the level of an action. One of the essential points of ISB is that a small explicit breaking of a symmetry by chemical potential or external magnetic field at the level of Lagranigan leads to drastic changes in the vacuum structure of the theory and thereby gives a jump in number density or magnetization . In addition, the mixed phase in the IT model does not possess the exclusive feature that the presence of the ISB expel color superconductivity found in ref. . Thus we are led to suggest that the ISB phenomena may not be a relevant notion in the IT model <sup>2</sup><sup>2</sup>2We are grateful to K. Langfeld for emphasizing this point to us..
### 2.4 $`\mathrm{\Delta }_B0`$, $`\lambda _B>0`$
In this case, the solutions for $`\mathrm{\Delta }_B`$ and $`\mathrm{\Delta }_M`$ are the same as those in (32) but $`E=\lambda _B`$ with a restriction $`\lambda _B+\mu +2\lambda _M>0`$ in the limit $`\beta \mathrm{}`$. We find from the ground state energy given by the parameter choice (33) that as long as $`\mu <\mu _c=\lambda _B`$, we have a mixed phase, $`\mathrm{\Delta }_B0`$ and $`\mathrm{\Delta }_M0`$, and that in the regime $`\mu >\mu _c`$, our system sits in the normal phase. The resulting phase diagram is depicted in fig. 1 (b).
## 3 Field Theory Model For Competitions Between $`\overline{\psi }\psi `$, $`\psi \psi `$ And $`\psi ^{}\psi `$
In this section, we shall study a field-theory model by generalizing the model considered recently by Chodos et al to one that contains a number density field corresponding to $`\psi ^{}\psi `$. The (1+1)-dimensional toy model we shall consider is defined by the Lagrangian
$`L`$ $`=`$ $`i\overline{\psi }^{(i)}\overline{)}\psi ^{(i)}\mu \psi ^{}\psi +{\displaystyle \frac{1}{2}}g^2[\overline{\psi }^{(i)}\psi ^{(i)}][\overline{\psi }^{(j)}\psi ^{(j)}]`$ (35)
$`+2G^2[\overline{\psi }^{(i)}\gamma _5\psi ^{(j)}][\overline{\psi }^{(i)}\gamma _5\psi ^{(j)}]2F^2[\overline{\psi }^{(i)}\gamma ^\mu \psi ^{(i)}][\overline{\psi }^{(j)}\gamma _\mu \psi ^{(j)}]`$
where $`i`$ runs from 1 to $`N`$. The last term is the term we have added to the model of Chodos et al. We shall do a large $`N`$ approximation. We do a Hubbard-Stratanovich transformation using the auxiliary fields, $`\varphi `$, $`B`$ and $`V_\mu `$ by adding the term
$`\mathrm{\Delta }L`$ $`=`$ $`{\displaystyle \frac{1}{2g^2}}[\varphi +g^2\overline{\psi }^{(i)}\psi ^{(i)}]^2{\displaystyle \frac{1}{G^2}}(B^{}G^2ϵ_{\alpha \beta }\psi _\alpha ^{(i)}\psi _\beta ^{(i)})(B+G^2ϵ_{\gamma \delta }\psi _\gamma ^{(i)}\psi _\delta ^{(i)})`$ (36)
$`+{\displaystyle \frac{2}{F^2}}[V_\mu +F^2\overline{\psi }^{(i)}\gamma _\mu \psi ^{(i)}]^2.`$
The resulting Lagrangian $`L^{}=L+\mathrm{\Delta }L`$ is
$`L^{}`$ $`=`$ $`\overline{\psi }^{(i)}(i\overline{)}\varphi \mu \gamma _0+4\overline{)}V)\psi ^{(i)}{\displaystyle \frac{\varphi ^2}{2g^2}}{\displaystyle \frac{B^{}B}{G^2}}+{\displaystyle \frac{2}{F^2}}V_\mu V^\mu `$ (37)
$`+Bϵ_{\alpha \beta }\psi _\alpha ^{(i)}\psi _\beta ^{(i)}B^{}ϵ_{\alpha \beta }\psi _\alpha ^{(i)}\psi _\beta ^{(i)}.`$
We can now integrate out $`\psi `$ and $`\psi ^{}`$ to obtain the effective action (modulo a constant)
$`\mathrm{\Gamma }_{eff}(\varphi ,B,B^{},V_0)`$ $`=`$ $`{\displaystyle d^2x(\frac{\varphi ^2}{2g^2}\frac{B^{}B}{G^2}+\frac{2}{F^2}V_0^2)}{\displaystyle \frac{i}{2}}\mathrm{Tr}\mathrm{ln}(\mathrm{A}^\mathrm{T}\mathrm{A})`$ (38)
$`{\displaystyle \frac{i}{2}}\mathrm{Tr}\mathrm{ln}[1+\mathrm{M}^2(\mathrm{A}^\mathrm{T})^1\sigma _2\mathrm{A}^1\sigma _2]`$
where we have assumed that $`B`$, $`B^{}`$, $`\varphi `$ and $`V_0`$ are constant since we are interested in an effective potential with translation invariance and have defined
$`M^2`$ $`=`$ $`4B^{}B`$
$`A`$ $`=`$ $`i_0+i\sigma _3_3\stackrel{~}{\mu }\varphi \sigma _1`$
$`A^T`$ $`=`$ $`i_0i\sigma _3_3\stackrel{~}{\mu }\varphi \sigma _1`$
with $`\stackrel{~}{\mu }=\mu 4V_0`$.
Factoring out $`N`$ coming from the flavor trace, we write $`g^2N=\lambda `$, $`G^2N=\frac{\kappa }{4}`$ and $`F^2N=2\alpha `$ and define $`V_{eff}`$ as $`\mathrm{\Gamma }_{eff}=N(d^2x)V_{eff}`$:
$`V_{eff}={\displaystyle \frac{\varphi ^2}{2\lambda }}+{\displaystyle \frac{M^2}{\kappa }}{\displaystyle \frac{V_0^2}{\alpha }}+V_{eff}^{(1)}`$ (39)
where $`V_{eff}^{(1)}=\frac{i}{2}[\mathrm{Tr}\mathrm{ln}(\mathrm{A}^\mathrm{T}\mathrm{A})+\mathrm{Tr}\mathrm{ln}(1+\mathrm{M}^2(\mathrm{A}^\mathrm{T})^1\sigma _2\mathrm{A}^1\sigma _2)]`$. Then the gap equations are obtained by
$`{\displaystyle \frac{V_{eff}}{\varphi ^2}}={\displaystyle \frac{V_{eff}}{M^2}}={\displaystyle \frac{V_{eff}}{V_0^2}}=0`$ (40)
from which we get
$`{\displaystyle \frac{1}{2\lambda }}`$ $`=`$ $`{\displaystyle \frac{V_{eff}^{(1)}}{\varphi ^2}}=i{\displaystyle \frac{d^2k}{(2\pi )^2}\frac{[k_0^2k_1^2+\stackrel{~}{\mu }^2+M^2\varphi ^2]}{D}}`$
$`{\displaystyle \frac{1}{\kappa }}`$ $`=`$ $`{\displaystyle \frac{V_{eff}^{(1)}}{M^2}}=i{\displaystyle \frac{d^2k}{(2\pi )^2}\frac{[k_0^2k_1^2\stackrel{~}{\mu }^2M^2+\varphi ^2]}{D}}`$
$`{\displaystyle \frac{1}{\alpha }}`$ $`=`$ $`{\displaystyle \frac{V_{eff}^{(1)}}{V_0^2}}=i{\displaystyle \frac{d^2k}{(2\pi )^2}\frac{4\stackrel{~}{\mu }[k_0^2+k_1^2\stackrel{~}{\mu }^2M^2+\varphi ^2]}{V_0D}}`$ (41)
where $`D=[k_0^2k_1^2\stackrel{~}{\mu }^2M^2+\varphi ^2]^24[\varphi ^2k_0^2+\mu ^2k_1^2\varphi ^2k_1^2]`$ with $`k_0=k_0+iϵ\mathrm{sgn}k_0`$. After the $`k_0`$ integral, we have
$`{\displaystyle \frac{1}{2\lambda }}`$ $`=`$ $`{\displaystyle \frac{1}{8\pi }}{\displaystyle _\mathrm{\Lambda }^\mathrm{\Lambda }}𝑑k_1[{\displaystyle \frac{1}{k_+}}+{\displaystyle \frac{1}{k_{}}}+{\displaystyle \frac{M^2+\stackrel{~}{\mu }^2}{\sqrt{M^2\varphi ^2+\stackrel{~}{\mu }^2(k_1^2+\varphi ^2)}}}({\displaystyle \frac{1}{k_+}}{\displaystyle \frac{1}{k_{}}})]`$
$`{\displaystyle \frac{1}{\kappa }}`$ $`=`$ $`{\displaystyle \frac{1}{8\pi }}{\displaystyle _\mathrm{\Lambda }^\mathrm{\Lambda }}𝑑k_1[{\displaystyle \frac{1}{k_+}}+{\displaystyle \frac{1}{k_{}}}+{\displaystyle \frac{\varphi ^2}{\sqrt{M^2\varphi ^2+\stackrel{~}{\mu }^2(k_1^2+\varphi ^2)}}}({\displaystyle \frac{1}{k_+}}{\displaystyle \frac{1}{k_{}}})]`$
$`{\displaystyle \frac{1}{\alpha }}`$ $`=`$ $`{\displaystyle \frac{\stackrel{~}{\mu }}{V_0}}{\displaystyle \frac{1}{2\pi }}{\displaystyle _\mathrm{\Lambda }^\mathrm{\Lambda }}𝑑k_1[{\displaystyle \frac{1}{k_+}}+{\displaystyle \frac{1}{k_{}}}+{\displaystyle \frac{\varphi ^2+k_1^2}{\sqrt{M^2\varphi ^2+\stackrel{~}{\mu }^2(k_1^2+\varphi ^2)}}}({\displaystyle \frac{1}{k_+}}{\displaystyle \frac{1}{k_{}}})]`$
where $`\mathrm{\Lambda }`$ is a cutoff to regularize logarithmic divergences
and $`k_\pm =\sqrt{M^2+\varphi ^2+\stackrel{~}{\mu }^2+k_1^2\pm 2\sqrt{M^2\varphi ^2+\stackrel{~}{\mu }^2(k_1^2+\varphi ^2)}}`$.
By integrating with respect to $`\varphi ^2`$, $`M^2`$ and $`V_0^2`$, we have the unrenormalized $`V_{eff}^{(1)}`$,
$`V_{eff}^{(1)}={\displaystyle \frac{1}{2\pi }}{\displaystyle _0^\mathrm{\Lambda }}𝑑k_1[k_+k_{}+c(k_1)]+C(\mu )`$ (42)
where $`c(k_1)`$ and $`C(\mu )`$, the value of which are determined later, are constants of integration. In the unrenormalized effective potential in free space, there can be quadratic and logarithmic divergences. Since free-space renormalization will eliminate all deivergences even in medium, we will cancel out the quadractic diverence by choosing a suitable value of $`c(k_1)`$ and define renormalized coupling constants to eliminate logarithmic divergences. To be explicit, we take $`\mu =0`$. Then the $`V_{eff}^{(1)}`$ is given by
$`V_{eff}^{(1)}`$ $`=`$ $`{\displaystyle \frac{1}{2\pi }}{\displaystyle _0^\mathrm{\Lambda }}dk_1[\sqrt{M^2+\varphi ^2+16V_0^2+k_1^2+2\sqrt{M^2\varphi ^2+16V_0^2(k_1^2+\varphi ^2)}}`$ (43)
$`+\sqrt{M^2+\varphi ^2+16V_0^2+k_1^22\sqrt{M^2\varphi ^2+16V_0^2(k_1^2+\varphi ^2)}}2k_1]`$
where we have chosen $`c(k_1)=2k_1`$ to cancel the quadratic divergence, see (44) and $`C(\mu =0)=0`$, see (51).
### 3.1 $`\varphi =0,\mu =0`$
Since it is not so easy to integrate (43) explicitly to separate the infinities from finite quantities, let us first confine ourselves to the case of $`\varphi =0`$ <sup>3</sup><sup>3</sup>3In Ref. , it is shown that although a solution to the gap equations exists in which both condensates, $`\overline{\psi }\psi `$ and $`\psi \psi `$, are non-vanishing, the global minimum of the effective potential always occurs for the case when one or the other condensate vanishes in free space ($`\mu =0`$) except for one very special case which we will not consider here. If we take naively this result, setting $`\varphi =0`$ is not so bad an approximation. . The $`V_{eff}^{(1)}`$ is given by
$`V_{eff}^{(1)}(V_0,M)`$ $`=`$ $`{\displaystyle \frac{1}{2\pi }}{\displaystyle _0^\mathrm{\Lambda }}𝑑k_1[\sqrt{M^2+(4V_0+k_1)^2}+\sqrt{M^2+(4V_0k_1)^2}2k_1]`$ (44)
$`=`$ $`{\displaystyle \frac{1}{4\pi }}[(\mathrm{\Lambda }+4V_0)\sqrt{(\mathrm{\Lambda }+4V_0)^2+M^2}+(\mathrm{\Lambda }4V_0)\sqrt{(\mathrm{\Lambda }4V_0)^2+M^2}`$
$`+M^2\mathrm{ln}{\displaystyle \frac{\mathrm{\Lambda }+4V_0+\sqrt{(\mathrm{\Lambda }+4V_0)^2+M^2}}{4V_0+\sqrt{16V_0^2+M^2}}}`$
$`+M^2\mathrm{ln}{\displaystyle \frac{\mathrm{\Lambda }4V_0+\sqrt{(\mathrm{\Lambda }4V_0)^2+M^2}}{4V_0+\sqrt{16V_0^2+M^2}}}2\mathrm{\Lambda }^2].`$
The unrenormalized effective potential is given by
$`V_{eff}(V_0,M)`$ $`=`$ $`M^2({\displaystyle \frac{1}{\kappa }}{\displaystyle \frac{1}{4\pi }})V_0^2({\displaystyle \frac{1}{\alpha }}+{\displaystyle \frac{8}{\pi }})`$ (45)
$`{\displaystyle \frac{1}{4\pi }}[M^2\mathrm{ln}{\displaystyle \frac{\mathrm{\Lambda }+4V_0+\sqrt{(\mathrm{\Lambda }+4V_0)^2+M^2}}{4V_0+\sqrt{16V_0^2+M^2}}}`$
$`+M^2\mathrm{ln}{\displaystyle \frac{\mathrm{\Lambda }4V_0+\sqrt{(\mathrm{\Lambda }4V_0)^2+M^2}}{4V_0+\sqrt{16V_0^2+M^2}}}].`$
We define the renormalized couplings $`\kappa _R`$ and $`\alpha _R`$ as
$`{\displaystyle \frac{^2V_{eff}}{BB^{}}}|_{M=M_0,V_0=v_0}={\displaystyle \frac{4}{\kappa _R}},{\displaystyle \frac{^2V_{eff}}{V_0^2}}|_{M=M_0,V_0=v_0}={\displaystyle \frac{2}{\alpha _R}}`$ (46)
and get
$`{\displaystyle \frac{1}{\alpha _R}}`$ $`=`$ $`{\displaystyle \frac{1}{\alpha }}+{\displaystyle \frac{8}{\pi }}`$
$`{\displaystyle \frac{1}{\kappa _R}}`$ $`=`$ $`{\displaystyle \frac{1}{\kappa }}{\displaystyle \frac{1}{4\pi }}[\mathrm{ln}{\displaystyle \frac{\mathrm{\Lambda }+4v_0+\sqrt{(\mathrm{\Lambda }+4v_0)^2+M_0^2}}{4v_0+\sqrt{16v_0^2+M_0^2}}}`$ (47)
$`+\mathrm{ln}{\displaystyle \frac{\mathrm{\Lambda }4v_0+\sqrt{(\mathrm{\Lambda }4v_0)^2+M_0^2}}{4v_0+\sqrt{16v_0^2+M_0^2}}}1].`$
The renormalized effective potential is now found to be
$`V_{eff}^R=M^2({\displaystyle \frac{1}{\kappa _R}}{\displaystyle \frac{1}{2\pi }})V_0^2{\displaystyle \frac{1}{\alpha _R}}+{\displaystyle \frac{M^2}{4\pi }}\mathrm{ln}{\displaystyle \frac{M^2}{M_0^2}}.`$ (48)
Note that $`V_0`$ and $`M`$ are not coupled to each other. Solving the gap equations for $`V_0`$ and $`M`$, we obtain
$`V_0`$ $`=`$ $`0`$
$`M^2`$ $`=`$ $`M_0^2e^{1\frac{4\pi }{\kappa _R}}\mathrm{or}0.`$ (49)
At the solutions, the effective potential becomes
$`V_{eff}^R`$ $`=`$ $`0\mathrm{for}V_0=0,M^2=0`$
$`V_{eff}^R`$ $`=`$ $`{\displaystyle \frac{M_0^2}{4\pi }}e^{1\frac{4\pi }{\kappa _R}}\mathrm{for}V_0=0,M^20.`$ (50)
So the phase with $`V_0=0,M^20`$ is energetically favored and is a global minimum when $`\alpha _R<0`$.
### 3.2 $`\varphi =0`$$`\mu 0`$
In this case, the renormalized effective potential is given (up to a constant) by
$`V_{eff}^R=M^2({\displaystyle \frac{1}{\kappa _R}}{\displaystyle \frac{1}{2\pi }})V_0^2{\displaystyle \frac{1}{\alpha _R}}{\displaystyle \frac{1}{2\pi }}\mu ^2+{\displaystyle \frac{4}{\pi }}\mu V_0+{\displaystyle \frac{M^2}{4\pi }}\mathrm{ln}{\displaystyle \frac{M^2}{M_0^2}}+C(\mu )`$ (51)
where $`C(\mu )`$ will be fixed by the condition that $`V_{eff}^R(M=0,V_0=0)=0`$. We will make use of the condition $`V_{eff}^R(M=0,V_0=0)=0`$ throughout this paper. Then we have the renormalized effective potential,
$`V_{eff}^R=M^2({\displaystyle \frac{1}{\kappa _R}}{\displaystyle \frac{1}{2\pi }})V_0^2{\displaystyle \frac{1}{\alpha _R}}+{\displaystyle \frac{4}{\pi }}\mu V_0+{\displaystyle \frac{M^2}{4\pi }}\mathrm{ln}{\displaystyle \frac{M^2}{M_0^2}}.`$ (52)
As far as $`M^2`$ is concerned, our effective potential is the same as that in Ref. . Note that since $`V_0`$ and $`M`$ do not couple, it is easy to find solutions of the gap equations. We find
$`V_0`$ $`=`$ $`2{\displaystyle \frac{\alpha _R}{\pi }}\mu ,M^2=0\mathrm{or}`$ (53)
$`V_0`$ $`=`$ $`2{\displaystyle \frac{\alpha _R}{\pi }}\mu ,M^2=M_0^2e^{1\frac{4\pi }{\kappa _R}}.`$ (54)
At the solutions, the effective potential becomes
$`V_{eff}^R`$ $`=`$ $`{\displaystyle \frac{4}{\pi ^2}}\alpha _R\mu ^2\mathrm{for}V_0=\alpha _R{\displaystyle \frac{2}{\pi }}\mu ,M^2=0`$
$`V_{eff}^R`$ $`=`$ $`{\displaystyle \frac{4}{\pi ^2}}\alpha _R\mu ^2{\displaystyle \frac{M_0^2}{4\pi }}e^{1\frac{4\pi }{\kappa _R}}\mathrm{for}V_0=\alpha _R{\displaystyle \frac{2}{\pi }}\mu ,M^20.`$ (55)
From (55), we can see that the phase with $`M0`$ is energetically favored regardless of the behavior of $`V_0`$. Thus we expect there could exist a phase where superconductivity and mean fields coexist in the case of $`M_0^20`$ at finite density. Since we are expecting superconductivity at high density, it is plausible that we have $`M_0^20`$ with $`M_0`$ being small at some relevant density.
### 3.3 $`M=0,\mu =0`$
In this case, $`V_{eff}^{(1)}`$ takes the form
$`V_{eff}^{(1)}={\displaystyle \frac{1}{2\pi }}{\displaystyle _0^\mathrm{\Lambda }}𝑑k_1[\sqrt{\varphi ^2+k_1^2}+4V_0+\sqrt{\varphi ^2+k_1^2}4V_02k_1].`$ (56)
We can consider two cases:
$``$ $`|\varphi |>4|V_0|`$:
In this case, the effective potential $`V_{eff}^{(1)}`$ is independent of $`V_0`$. Using the exactly same method adopted in the previous section, we have
$`{\displaystyle \frac{1}{\lambda }}`$ $`=`$ $`{\displaystyle \frac{1}{\lambda _R}}+{\displaystyle \frac{1}{\pi }}\mathrm{ln}{\displaystyle \frac{\mathrm{\Lambda }+\sqrt{\mathrm{\Lambda }^2+\varphi ^2}}{\varphi _0}}{\displaystyle \frac{1}{\pi }}`$ (57)
$`{\displaystyle \frac{1}{\alpha }}`$ $`=`$ $`{\displaystyle \frac{1}{\alpha _R}}`$ (58)
$`V_{eff}^R`$ $`=`$ $`\varphi ^2({\displaystyle \frac{1}{2\lambda _R}}{\displaystyle \frac{3}{4\pi }}){\displaystyle \frac{1}{\alpha _R}}V_0^2+{\displaystyle \frac{\varphi ^2}{4\pi }}\mathrm{ln}{\displaystyle \frac{\varphi ^2}{\varphi _0^2}}`$ (59)
where $`\lambda _R`$ is defined by $`\frac{^2V_{eff}}{\varphi ^2}_{M=M_0,\varphi =\varphi _0}=\frac{1}{\lambda _R}`$. Since $`\alpha `$ and $`\alpha _R`$ differ by a constant, we will use, hereafter, $`\alpha `$ instead of $`\alpha _R`$ just for simplicity. The solutions of the gap equations are given by
$`V_0`$ $`=`$ $`0,\varphi =0`$ (60)
$`\mathrm{or}V_0`$ $`=`$ $`0,\varphi =\varphi _0e^{1\frac{\pi }{\lambda _R}}.`$ (61)
It can be easily seen that the phase with $`V_0=0`$ and $`\varphi 0`$, at which $`V_{eff}^R=\frac{1}{4\pi }\varphi _0^2e^{2\frac{2\pi }{\lambda _R}}`$, is favored. This comes as no surprise.
$``$ $`|\varphi |<4|V_0|`$:
We find
$`{\displaystyle \frac{1}{\alpha }}`$ $`=`$ $`{\displaystyle \frac{1}{\alpha _R}}{\displaystyle \frac{2}{\pi }}{\displaystyle \frac{16|v_0|}{\sqrt{16v_0^2\varphi _0^2}}}`$
$`{\displaystyle \frac{1}{\lambda }}`$ $`=`$ $`{\displaystyle \frac{1}{\lambda _R}}+{\displaystyle \frac{1}{\pi }}\mathrm{ln}{\displaystyle \frac{\mathrm{\Lambda }+\sqrt{\mathrm{\Lambda }^2+\varphi _0^2}}{4|v_0|+\sqrt{16v_0^2\varphi _0^2}}}+{\displaystyle \frac{1}{\pi }}{\displaystyle \frac{\varphi _0^2}{16v_0^2\varphi _0^2+4|v_0|\sqrt{16v_0^2\varphi _0^2}}}`$
$`V_{eff}^R`$ $`=`$ $`\varphi ^2({\displaystyle \frac{1}{2\lambda _R}}{\displaystyle \frac{1}{4\pi }}){\displaystyle \frac{1}{\alpha }}V_0^2`$ (62)
$`+{\displaystyle \frac{1}{2\pi }}\varphi ^2\mathrm{ln}{\displaystyle \frac{4|V_0|+\sqrt{16V_0^2\varphi ^2}}{4|v_0|+\sqrt{16v_0^2\varphi _0^2}}}`$
$`+{\displaystyle \frac{1}{2\pi }}\varphi ^2{\displaystyle \frac{\varphi _0^2}{16v_0^2\varphi _0^2+4|v_0|\sqrt{16v_0^2\varphi _0^2}}}{\displaystyle \frac{2}{\pi }}|V_0|\sqrt{16V_0^2\varphi ^2}.`$
Solving the gap equations, we find that the only possible solutions are $`V_0=0`$ and $`\varphi =0`$ where $`V_{eff}^R=0`$. So at zero chemical potential, the system is again characterized by $`V_0=0`$ and $`\varphi 0`$.
### 3.4 $`M=0,\mu 0`$
$``$ $`|\varphi |>4|V_0|`$:
For $`\mu <\varphi `$, the effective potential is equal to (59). In this case, our system is characterized by
$`V_0=0,\varphi =\varphi _0e^{1\frac{\pi }{\lambda _R}}.`$ (63)
For $`\mu >\varphi `$, we have
$`V_{eff}^R`$ $`=`$ $`\varphi ^2({\displaystyle \frac{1}{2\lambda _R}}{\displaystyle \frac{3}{4\pi }}){\displaystyle \frac{1}{\alpha }}V_0^2{\displaystyle \frac{1}{2\pi }}\stackrel{~}{\mu }\sqrt{\stackrel{~}{\mu }^2\varphi ^2}`$ (64)
$`{\displaystyle \frac{1}{2\pi }}\varphi ^2\mathrm{ln}{\displaystyle \frac{\varphi _0}{\stackrel{~}{\mu }+\sqrt{\stackrel{~}{\mu }^2\varphi ^2}}}+{\displaystyle \frac{1}{2\pi }}\mu ^2.`$
(Recall that $`\stackrel{~}{\mu }=\mu 4V_0`$.) The possible solutions of the gap equations are calculated to be
$`V_0`$ $`=`$ $`{\displaystyle \frac{1}{\frac{\pi }{2\alpha }4}}(\mu \varphi _0e^{1\frac{\pi }{\lambda _R}})`$
$`\varphi ^2`$ $`=`$ $`\stackrel{~}{\mu }^2{\displaystyle \frac{\pi ^2}{4\alpha ^2}}V_0^2.`$ (65)
At the solutions of the gap equations, we have
$`V_{eff}^R={\displaystyle \frac{1}{4\pi }}\mu ^2+({\displaystyle \frac{2}{\pi }}{\displaystyle \frac{1}{4\alpha _R}})\mu V_0{\displaystyle \frac{1}{4\pi }}(16{\displaystyle \frac{\pi ^2}{4\alpha _R^2}})V_0^2.`$ (66)
$``$ $`|\varphi |<4|V_0|`$:
$`V_{eff}^R`$ $`=`$ $`\varphi ^2({\displaystyle \frac{1}{2\lambda _R}}{\displaystyle \frac{1}{4\pi }}){\displaystyle \frac{1}{\alpha }}V_0^2`$ (67)
$`+{\displaystyle \frac{1}{2\pi }}\varphi ^2\mathrm{ln}{\displaystyle \frac{\stackrel{~}{\mu }+\sqrt{\stackrel{~}{\mu }^2\varphi ^2}}{4v_0+\sqrt{16v_0^2\varphi _0^2}}}+{\displaystyle \frac{1}{2\pi }}\mu ^2`$
$`+{\displaystyle \frac{1}{2\pi }}\varphi ^2{\displaystyle \frac{\varphi _0^2}{16v_0^2\varphi _0^2+4|v_0|\sqrt{16v_0^2\varphi _0^2}}}{\displaystyle \frac{1}{2\pi }}\stackrel{~}{\mu }\sqrt{\stackrel{~}{\mu }^2\varphi ^2}.`$
First, we consider the case of
$`\varphi =0\mathrm{and}V_0={\displaystyle \frac{1}{\frac{\pi }{2\alpha }+4}}\mu `$ (68)
which is one of the solutions of the gap equations. In this case, the effective potential becomes
$`V_{eff}^R={\displaystyle \frac{1}{2\pi }}(16+{\displaystyle \frac{2\pi }{\alpha }})V_0^2+{\displaystyle \frac{4}{\pi }}V_0\mu .`$ (69)
For $`\stackrel{~}{\mu }>0`$, we have the following solution for the gap equation
$`V_0`$ $`=`$ $`{\displaystyle \frac{1}{\frac{\pi }{2\alpha }4}}[\mu C_1e^{(\frac{\pi }{\lambda _R}+C_2)}]`$
$`\varphi ^2`$ $`=`$ $`\stackrel{~}{\mu }^2{\displaystyle \frac{\pi ^2}{4\alpha ^2}}V_0^2.`$ (70)
In the case of $`\stackrel{~}{\mu }<0`$, we have
$`V_0`$ $`=`$ $`{\displaystyle \frac{1}{\frac{\pi }{2\alpha }4}}[\mu +C_1e^{(\frac{\pi }{\lambda _R}+C_2)}]`$
$`\varphi ^2`$ $`=`$ $`\stackrel{~}{\mu }^2{\displaystyle \frac{\pi ^2}{4\alpha ^2}}V_0^2.`$ (71)
To see which phase is energetically favorable, we take $`\varphi _0=m_F`$ and $`M_00`$ <sup>4</sup><sup>4</sup>4Since it is not plausible to have superconductivity in free space, $`M_0`$ should be zero or very small, if any. We shall assume, however, that we can have a small but non-zero value of $`M_0`$ at finite density.
In free space ($`\mu =0`$), we have two phases:
$`V_{eff}^R`$ $`=`$ $`{\displaystyle \frac{1}{4\pi }}m_F^2e^{2\frac{2\pi }{\lambda _R}}\mathrm{for}V_0=0\mathrm{and}\varphi 0`$
$`V_{eff}^R`$ $`=`$ $`{\displaystyle \frac{M_0^2}{4\pi }}e^{1\frac{4\pi }{\kappa _R}}\mathrm{for}V_0=0\mathrm{and}M0.`$ (72)
Since $`M_0`$ will be zero or small, our system sits in the phase with $`V_0=0`$ and $`\varphi 0`$. In free space, therefore, we have chiral symmetry breaking but no superconductivity.
At finite density, the situation becomes more complicated. We have six sets of solutions of the gap equation: (54), (63), (65), (68), (70) and (71). Now let us investigate the competition between the phases characterized by (54), (63) and (68). The effective potentials given at their ground-state positions are
$`I:V_{eff}^R`$ $`=`$ $`{\displaystyle \frac{4}{\pi ^2}}{\displaystyle \frac{1}{\frac{1}{\alpha }+\frac{8}{\pi }}}\mu ^2{\displaystyle \frac{M_0^2}{4\pi ^2}}e^{1\frac{4\pi }{\kappa _R}}(\text{54})`$
$`II:V_{eff}^R`$ $`=`$ $`{\displaystyle \frac{1}{4\pi }}m_F^2e^{2\frac{2\pi }{\lambda _R}}(\text{63})`$
$`III:V_{eff}^R`$ $`=`$ $`{\displaystyle \frac{4}{\pi ^2}}{\displaystyle \frac{1}{\frac{1}{\alpha }+\frac{8}{\pi }}}\mu ^2(\text{68}).`$ (73)
Comparing $`I`$ and $`III`$, we find that our system will be in the mixed phase ($`I`$) characterized by $`\psi ^{}\psi 0`$ and $`\psi \psi 0`$ as long as $`M_0^20`$. From the condition that the phase $`I`$ is the global minimum of the potential (52), $`\frac{1}{\alpha }+\frac{8}{\pi }<0`$, we find $`0.125\pi <\alpha <0`$. It is expected that the phase $`II`$, which is valid when $`\mu <|\varphi |`$, becomes the absolute minimum of the potential at low density. To see the competition between $`I`$ and $`II`$ explicitly, we neglect a term with $`M_0^2`$ in the phase I and take $`\lambda _R=\pi `$ and $`\alpha =0.1\pi `$. Then we find $`\frac{\mu _c}{m_F}0.4`$. In the regime $`\mu <\mu _c`$, we have the chiral symmetry breaking phase ($`II`$) with almost zero fermion number density <sup>5</sup><sup>5</sup>5If we consider density dependence of $`m_F`$ and $`\lambda _R`$, the fermion number density may not be zero but is still expected to be small. defined by $`\rho _F=\frac{V_R^{eff}}{\stackrel{~}{\mu }}`$. In the case of $`\mu >\mu _c`$ the system sits in the mixed phase ($`I`$) with fermion number density
$`\rho _F`$ $`=`$ $`{\displaystyle \frac{8\alpha _R}{\pi ^28\pi \alpha _R}}\mu \mathrm{with}{\displaystyle \frac{1}{\alpha _R}}={\displaystyle \frac{1}{\alpha }}+{\displaystyle \frac{8}{\pi }}`$ (74)
$``$ $`0.25\mu \mathrm{for}\alpha =0.1\pi .`$
The fermion number density is depicted in Fig. 2. We should, however, consider competitions with other possible phases. To investigate numerically the phases characterized by (65), (70) and (71), we take the following parameters
$`\lambda _R`$ $`=`$ $`\pi `$
$`\alpha `$ $`=`$ $`\pi ,0.1\pi `$
$`{\displaystyle \frac{v_0}{\varphi _0}}`$ $`=`$ $`10,1.0,0.3.`$ (75)
With this parameter choice, we plot the renormalized effective potentials near the minimum-energy positions (i.e., solutions of the gap equations), (65), (70) and (71). We find that the phases characterized by (65), (70) and (71) correspond to the maximum or saddle point of the renormalized effective potential or does not satisfy some constraints, for example $`4|V_0|>|\varphi |`$, near the minimum-energy positions and therefore those phases are unstable or an unphysical “vacuum.”
## 4 Conclusion
We studied the competitions between induced symmetry breaking (ISB), Cooper pairing condensate and chiral condensate at finite baryon density.
By reformulating the IT model using the Bogoliubov-Valatin transformation, we show that the mixed phase, in which both $`\mathrm{\Delta }_M`$ and $`\mathrm{\Delta }_B`$ are non-zero, is energetically favored. By symmetry we argue that the ISB phenomena may not be a relevant notion in the IT model.
We calculated the effective potentials $`V(\varphi =0,V_0,M)`$ and $`V(M=0,V_0,\varphi )`$ which are functions of the fermion-antifermion (or chiral), ISB (or mean-field) and Cooper-pairing order parameters $`\varphi `$, $`V_0`$ and $`M`$, respectively. At zero chemical potential, we find that the global minimum of the effective potential is given by $`V_0=0`$ and $`M0`$ or $`V_0=0`$ and $`\varphi 0`$. Since $`M_0`$ should be zero or very small, it is reasonable to conclude that the system sits in the chiral symmetry breaking phase.
At finite density, taking $`\lambda _R=\pi `$ and $`\alpha =0.1\pi `$, we find that in the regime $`\mu <\mu _c`$, we have the chiral symmetry breaking phase(II) with almost zero fermion number density and in the case of $`\mu >\mu _c`$ the system sits in the mixed phase with nonzero “vacuum” (or rather ground-state) expectation values of $`\psi ^{}\psi `$ and $`\psi \psi `$, i.e, $`V_00`$ and $`M0`$. We observe an ISB-like behavior in the fermion number density as in fig. 2 but we fail to observe the exclusive competition obtained in . As we can see in (54) and also in ref., $`M`$ ($`\psi \psi `$) is independent of the chemical potential. We are unable to say whether or not these are an artifact of the simplified models not present in effective theories of QCD. A similar (uncertain) situation applies also to analysis made in QCD at weak coupling.
Our analysis depends on the value of the renomalization point which is arbitrary in general. The possible resolution to this arbitrariness is to do a renormalization group analysis at finite density discussed in . Such a procedure will replace the renormalization scale (up to a constant) by the chemical potential. We leave this exercise to a future publication.
## Acknowledgment
Part of this work was done at Korea Institute of Advanced Studies, Seoul, Korea and at the Theory Group, State University of New York, Stony Brook, N.Y. The hospitality of the two institutes is highly appreciated. We thank Kurt Langfeld for helpful comments. YK is supported in part by the Korea Ministry of Education (BSRI 99-2441) and KOSEF (Grant No. 985-0200-001-2) and he thanks Prof. Hyun Kyu Lee for his support.
## Appendix
In this appendix, we compare the grand canonical ensemble used in this work and in ref. (case B) with that widely used in BCS theory, for example see (case A). Here ($`a,a^{}`$) stand for the annihilation and creation operators for the “bare” fermions and ($`b,b^{}`$) for the quasiparticles as in section 2.
$`\mathrm{\Delta }_B0,\mathrm{\Delta }_M=0`$
Here we shall take the mean-field gap to be zero, i.e., $`\mathrm{\Delta }_M=0`$ for simplicity. Since the BCS ground state does not preserve the particle number of the ground state, a condition is imposed on the average;
$`\mathrm{\Psi }_{BCS}|N_{op}|\mathrm{\Psi }_{BCS}=\overline{N}`$ (76)
where $`N_{op}=_ka_k^{}a_k`$, the number operator. As described in ref. in detail, this relaxation of the condition of $`N`$ in the BCS and Bogoliubov theory must be distinguished from the use of grand canonical ensemble in statistical mechanics. In grand canonical ensemble of statistical mechanics, we deal with an ensemble of systems, with a distribution of particle numbers $`N`$ with the systems with N particles being weighted by a factor $`z^N`$ where $`z`$ called fugacity is defined by $`z=e^{\beta \mu }`$. However, each separate system within an assembly has its own definite number of particles $`N`$.
In the case of A, we add a term $`\mu _aa^{}a`$ to the Hamiltonian to specify the mean value of particle number. Normally in this case, we adjust the chemical potential $`\mu `$ to satisfy $`N_a=\overline{N}_a`$. Here the chemical potential $`\mu _a`$ is a variable conjugate to $`N_a=_ka_k^{}a_k`$. After specifying the mean number of $`N_a`$, we diagonalize the Hamiltonian to get $`H(b^{}b)`$ describing quasi-particles. We can specify a constant number (in the sense of the average given above) to a system characterized by this diagonalized Hamiltonian $`H(b^{}b)`$. Therefore this system becomes effectively canonical ensemble of the statistical system with definite number but we are essentially using the grand canonical ensemble. Then the probability of quasi-particle thermal excitation is given at a given energy $`ϵ`$ by $`b^{}b=(1+e^{\beta \sqrt{(ϵ\mu _a)^2+\mathrm{\Delta }_B^2}})^1`$.
In the case of B, we do not constrain the mean number in terms of $`N_a=_ka_k^{}a_k`$. Since our diagonalized Hamiltonian $`H(b^{}b)`$ has the same structure with that of non-interacting harmonic oscillator and preserves the particle number, we can define a simultaneous eigenstate of quasi-particle number and energy operators <sup>6</sup><sup>6</sup>6This is not true in the case of BCS and Bogoliubov theory.. Note that the quasiparticle operator $`b`$ kills, by definition, the BCS ground state $`b|\mathrm{\Psi }_{BCS}=0`$ and the lowest quasiparticle excitation state is given by $`b^{}|\mathrm{\Psi }_{BCS}`$. Following an elementary procedure in statistical mechanics, we have $`b^{}b=(e^{\beta (\sqrt{ϵ^2+\mathrm{\Delta }_B^2}\mu _b)}+1)^1`$ with $`\mu _b`$ being the “chemical potential” for quasiparticles. This formalism B is essential if we want to consider both $`\lambda _B>0`$ and $`\lambda _B<0`$ on the same footing. If we use the formalism A, the gap equation becomes
$`\sqrt{(ϵ\mu _a)^2+\mathrm{\Delta }_B^2}=\lambda _B\mathrm{tanh}{\displaystyle \frac{\beta \sqrt{(ϵ\mu _a)^2+\mathrm{\Delta }_B^2}}{2}}.`$ (77)
Taking $`\beta \mathrm{}`$, we have
$`\sqrt{(ϵ\mu _a)^2+\mathrm{\Delta }_B^2}=\lambda _B`$ (78)
showing that $`\lambda _B<0`$ does not enter into the consideration. Note that $`\lambda _B>0`$ corresponds to the usual BCS attraction.
Since the chemical potential $`\mu _b`$ essentially controls the number of quasiparticle excitations while the chemical potential $`\mu _a`$ fixes the mean number of the ground state, one might say that $`\mu _b`$ has nothing to do with $`\mu _a`$. But one should keep it in mind that $`b^{}`$ is a linear combination of the $`a`$ and $`a^{}`$ original fermion operators. Thus $`b^{}`$ can create and annihilate fermions ($`a,a^{}`$) from the ground state. This implies that controlling the number of quasi-particles must have something to do with that of the “bare” fermion $`(a^{},a)`$. Whether we fix the number of ground state($`A`$) or quasi-particle excitation($`B`$), they should give the same answer for $`a^{}a`$. Note that $`(a^{},a)`$ and $`(b^{},b)`$ are connected by the canonical transformation:
$`a^{}(p)a(p)`$ $`=`$ $`\{{\displaystyle \frac{c^2(p)}{1+e^{\beta (E\mu _b)}}}+{\displaystyle \frac{s^2(p)}{1+e^{\beta (E\mu _b)}}}\}\mathrm{with}E=\sqrt{ϵ^2+\mathrm{\Delta }_B^2}`$ (79)
$`=`$ $`\{{\displaystyle \frac{c^2(p)}{1+e^{\beta E}}}+{\displaystyle \frac{s^2(p)}{1+e^{\beta E}}}\}\mathrm{with}E=\sqrt{(ϵ\mu _a)^2+\mathrm{\Delta }_B^2}`$
where $`ϵ=p^2`$.
$`\mathrm{\Delta }_B=0,\mathrm{\Delta }_M0`$
In this case the probability of thermal excitation of “bare” fermion is given by (for $`ϵ+\mathrm{\Delta }_M>\mu _a`$)
$`a^{}(p)a(p)`$ $`=`$ $`{\displaystyle \frac{1}{1+e^{\beta (ϵ+\mathrm{\Delta }_M\mu _b)}}}\mathrm{with}B`$ (80)
$`=`$ $`{\displaystyle \frac{1}{1+e^{\beta (ϵ+\mathrm{\Delta }_M\mu _a)}}}\mathrm{with}A`$
$`=`$ $`b^{}(p)b(p)`$
showing $`\mu _a=\mu _b`$. Note that in the case $`\mathrm{\Delta }_B=0,\mathrm{\Delta }_M0`$, the KMS-state is defined for the operator $`(a,a^{})`$.
|
warning/0004/hep-th0004145.html
|
ar5iv
|
text
|
# References
Hodge decomposition theorem
for Abelian two-form gauge theory
E. Harikumar, R. P. Malik<sup>∗∗,†</sup> and M. Sivakumar <sup>∗,‡</sup>
School of Physics, University of Hyderabad,
Hyderabad-500 046, India
<sup>∗∗</sup> S. N. Bose National Centre for Basic Sciences,
Block-JD, Sector-III, Salt Lake, Calcutta-700 098, India
Abstract. We show that the BRST/anti-BRST invariant $`3+1`$ dimensional 2-form gauge theory has further nilpotent symmetries (dual BRST /anti-dual BRST) that leave the gauge fixing term invariant. The generator for the dual BRST symmetry is analogous to the co-exterior derivative of differential geometry. There exists a bosonic symmetry which keeps the ghost terms invariant and it turns out to be the analogue of the Laplacian operator. The Hodge duality operation is shown to correspond to a discrete symmetry in the theory. The generators of all these continuous symmetries are shown to obey the algebra of the de Rham cohomology operators of differential geometry. We derive the extended BRST algebra constituted by six conserved charges and discuss the Hodge decomposition theorem in the quantum Hilbert space of states.
footnotetext: E-mail: malik@boson.bose.res.infootnotetext: E-mail: mssp@uohyd.ernet.in
1 Introduction
For the covariant canonical quantization of gauge theories, one of the most elegant methods is the Becchi-Rouet-Stora-Tyutin (BRST) formalism where (quantum) gauge invariance and unitarity are respected together at any arbitrary order of perturbation theory. The first-class constraints of the original gauge theories are found to be encoded in the subsidiary condition ($`Q_B|phys>=0`$) when one requires that the physical subspace (of the total Hilbert space of states) contains only those states that are annihilated by the nilpotent ($`Q_B^2=0`$) and conserved ($`\dot{Q}_B=0`$) BRST charge $`Q_B`$. In fact, the condition $`Q_B|phys>=0`$ implies that the operator form of the first-class constraints annihilate the physical states. This requirement is essential for the consistent quantization of any theory endowed with the first-class constraints (Dirac’s prescription) . The nilpotency of the BRST charge ($`Q_B^2=0`$) and physicality criteria ($`Q_B|phys>=0`$) are the two key requirements for the discussion of cohomological aspects of BRST formalism \[5-8\] and its connection with the de Rham cohomology operator $`d`$ (exterior derivative; $`d^2=0`$) of differential geometry defined on a compact manifold. For instance, two physical states are said to belong to the same cohomology class w.r.t. $`Q_B`$ if they differ by a BRST exact state as two closed forms belong to the same cohomology class w.r.t. operator $`d`$ if they differ by an exact form. There are two other de Rham cohomology operators that are essential for the definition of the Hodge decomposition theorem which states that, on a compact manifold, any arbitrary $`n`$-form $`f_n(n=0,1,2,3\mathrm{}..)`$ can be written as a unique sum of a harmonic form $`h_n(\mathrm{\Delta }h_n=0,dh_n=0,\delta h_n=0)`$, an exact form ($`de_{n1}`$) and a co-exact form ($`\delta c_{n+1}`$):
$$\begin{array}{ccc}f_n=h_n+de_{n1}+\delta c_{n+1}\hfill & & \end{array}$$
$`(1.1)`$
where $`\delta =\pm d,(\delta ^2=0)`$ is the dual exterior derivative, $`\mathrm{\Delta }=(d+\delta )^2=d\delta +\delta d`$ is the Laplacian operator and $``$ is the so-called Hodge duality operation \[9-12\].
It is a well known fact that the cohomological operator $`d`$ of differential geometry finds its analogue in the local, conserved and nilpotent BRST charge $`Q_B`$ . It is, therefore, interesting to enquire if analogous local conserved charges (and corresponding local symmetry transformations for a given Lagrangian density) exist for the analogues of the other cohomological operators, viz; $`\delta `$ and $`\mathrm{\Delta }`$. Some interesting attempts \[13-16\] have been made to express $`\delta `$ and $`\mathrm{\Delta }`$ in the language of symmetry properties of a given Lagrangian density for the 1-form interacting gauge theory in any arbitrary spacetime dimension. The symmetry transformations, however, turn out to be non-local and non-covariant. In the covariant formulation , the nilpotency of transformations are dependent on the specific choice of parameters of the theory. Recently, it has been shown that the two-dimensional (2D) free Abelian as well as non-Abelian gauge theories (without any interaction with matter fields) provide a topological <sup>*</sup><sup>*</sup>* A theory with a flat spacetime metric and without any propagating degrees of freedom. field theoretical model for the Hodge theory where symmetry transformations corresponding to the de Rham cohomology operators ($`d,\delta ,\mathrm{\Delta }`$) are nilpotent (for $`d`$ and $`\delta `$), local, covariant and continuous \[18-20\]. The analogue of these local symmetries have also been shown to exist for the 2D topological fields (i.e., 2D Abelian gauge fields) coupled to matter (Dirac) fields in two-dimensions of spacetime .
In our present paper, we show the existence of symmetries corresponding to the de Rham cohomology operators for a field theoretical model in the physical four ($`3+1)`$ dimensional spacetime We follow the notations in which the flat Minkowski metric is $`\eta _{\mu \nu }=`$ diag $`(+,,,)`$ and Levi-Civita totally antisymmetric tensor $`\epsilon _{0123}=+1=\epsilon ^{0123},\epsilon _{0ijk}=\epsilon _{ijk}=\epsilon ^{ijk},\epsilon ^{\mu \nu \lambda \xi }\epsilon _{\mu \nu \lambda \xi }=4!,\epsilon ^{\mu \nu \lambda \xi }\epsilon _{\mu \nu \lambda \rho }=3!\delta _\rho ^\xi ,`$ etc. Here Greek indices: $`\mu ,\nu ,\lambda \mathrm{}\mathrm{}.`$ = 0, 1, 2, 3 and Latin indices: i , j, k …….= 1, 2, 3.. The search for such symmetries in the Abelian and non-Abelian 1-form gauge theories, even though quite illuminating, has not been fully successful and satisfactory, as stated earlier. Thus, the central theme of our present work is to show that the free Abelian antisymmetric (2-form) gauge theory in 4D provides a prototype example for Hodge theory where the de Rham cohomology operators correspond to the local and conserved charges. These charges turn out to be the generators of specific local, covariant and continuous symmtery transformations for the BRST invariant Lagrangian density of this theory.
The 2-form massless gauge theory is interesting by itself as it is a dual description for the massless scalar theory. It also has interesting constraint structure: stage-one reducibility and corresponding ghost for the ghost feature. In addition, the 2-form potential also appears naturally in supergravity and superstring theories including the recent developments in non-commutative geometry . Its different forms have appeared in other contexts of theoretical physics, e.g., QCD, cosmic strings and vortices, black holes, etc. \[23-26\]. In fact, this theory, coupled to a 1-form Abelian gauge field via a ‘topological’ $`BF`$ term, has rich mathematical structure and has been studied from various points view, viz., duality consideration , Dirac bracket analyses , BFT Hamiltonian formulation , BRST quantization , etc.
We shall consider the BRST invariant version (see, e.g., Section 2 (below)) of the free 4D Kalb-Ramond Lagrangian density
$$\begin{array}{ccc}=\frac{1}{12}H^{\mu \nu \lambda }H_{\mu \nu \lambda }\hfill & & \end{array}$$
$`(1.2)`$
where $`H_{\mu \nu \lambda }=_\mu B_{\nu \lambda }+_\nu B_{\lambda \mu }+_\lambda B_{\mu \nu }`$ is the totally antisymmetric curvature tensor constructed from the antisymmetric gauge field $`B_{\mu \nu }`$ The gauge field $`B_{\mu \nu }`$ is defined through 2-form: $`B=\frac{1}{2}B_{\mu \nu }dx^\mu dx^\nu `$ and the curvature tensor $`H_{\mu \nu \lambda }`$ is defined through 3-form as: $`H=dB`$. It can be readily seen that the gauge-fixing term $`_\mu B^{\mu \nu }`$ can be defined through one-form by the application of $`\delta `$ as: $`_\mu B^{\mu \nu }dx_\nu =\delta B`$ where $`\delta =d`$ is the dual exterior derivative of $`d`$. It is clear that the gauge-fixing term is the ‘Hodge dual’ of curvature term. and show that: (i) in addition to the usual BRST charge ($`Q_B`$), there exists a local, conserved and nilpotent dual(co)-BRST charge ($`Q_D`$) under which the gauge-fixing term of this theory remains invariant. This fact should be contrasted with the usual BRST transformations, under which, it is the kinetic energy term (more precisely curvature tensor $`H_{\mu \nu \lambda }`$ itself) that remains invariant. (ii) The anticommutator of BRST- and dual BRST transformations leads to a symmetry transformation that is generated by a local and conserved bosonic charge ($`W`$). This is analogous to the Laplacian operator in differential geometry where it is given by the anticommutator of $`d`$ and $`\delta `$. (iii) The conserved charges (e.g., $`Q_B,Q_D,W`$) can be exploited together for the discussion of the Hodge decomposition theorem in the quantum Hilbert space of states and for the anlaysis of the constraint structure on the physical (harmonic) states of the theory. (iv) A discrete transformation symmetry of the Lagrangian density relates $`Q_B`$ and $`Q_D`$ like a dual symmetry: $`Q_BQ_D,Q_DQ_B`$ and $`WW`$. This relationship maintains the anticommutator between $`Q_B`$ and $`Q_D`$ and the underlying discrete symmetry turns out to be a realization of the Hodge $``$ operation of differential geometry for this gauge theory. To the best of our knowledge, this is the first example of a field theoretical model for the Hodge theory in four ($`3+1`$) dimensional spacetime where the conserved charges corresponding to the de Rham cohomology operators generate the local, continuous and covariant transformations for the fields. The existence of new symmetries (corresponding to $`\delta `$ and $`\mathrm{\Delta }`$) and their generalizations might turn out to be useful in the proof of renormalizability of an interacting theory where the gauge fields are coupled to matter fields. Thus, our present work is the first step towards our main goal of having a complete understanding of the interacting theory.
The outline of our paper is as follows. In section 2, we recapitulate the essentials of the BRST formalism for the 2-form gauge theory and set up the notations for our further discussion. This is followed by the discussion and derivation of the dual BRST symmetry in section 3. We derive the symmetry generated by the Casimir operator in section 4 and obtain the corresponding conserved charge. Section 5 is devoted to the derivation of the extended BRST algebra and a brief discussion is provided for its possible connection to the de Rham cohomology operators of the differential geometry. In section 6, we discuss the Hodge decomposition theorem in the quantum Hilbert space of states and analyze the structure of constraints on the physical states of the theory. Finally, in section 7, we make some concluding remarks and point out some directions that can be pursued in the future.
2 Preliminary: BRST symmetry
We begin with the BRST invariant Lagrangian density
$$\begin{array}{ccc}_B\hfill & =& \frac{1}{12}H^{\mu \nu \lambda }H_{\mu \nu \lambda }\frac{1}{2}B^\mu B_\mu +B_\nu \left(_\mu B^{\mu \nu }^\nu \varphi _1\right)_\mu \overline{\beta }^\mu \beta \hfill \\ & +& \left(_\mu \overline{C}_\nu _\nu \overline{C}_\mu \right)^\mu C^\nu +\rho \left(_\mu C^\mu +\lambda \right)+\left(_\mu \overline{C}^\mu +\rho \right)\lambda \hfill \end{array}$$
$`(2.1)`$
where $`B_\mu ,\varphi _1,\lambda `$ and $`\rho `$ are the auxiliary fields <sup>§</sup><sup>§</sup>§ By integrating out the auxiliary fields, we will obtain the Lagrangian density which respects the on-shell nilpotent BRST symmetry. introduced to have the off-shell nilpotent BRST invariance. The following BRST transformations
$$\begin{array}{ccc}\hfill & & \delta _BB_{\mu \nu }=\eta \left(_\mu C_\nu _\nu C_\mu \right)\delta _BC_\mu =\eta _\mu \beta \delta _B\overline{C}_\mu =\eta B_\mu \hfill \\ & & \delta _B\varphi _1=\eta \lambda \delta _B\overline{\beta }=\eta \rho \delta _B(B_\mu ,\rho ,\lambda ,\beta )=0\hfill \end{array}$$
$`(2.2)`$
leave the Lagrangian density invariant up to a total derivative term.
The continuous symmetry transformations (2.2) lead to the following nilpotent ($`Q_B^2=0`$) and conserved ($`\dot{Q}_B=0`$) BRST charge due to Noether theorem
$$\begin{array}{ccc}Q_B=d^3x\left[H^{0ij}_iC_j+B_0\lambda \rho \dot{\beta }+(^0C^i^iC^0)B_i(^0\overline{C}^i^i\overline{C}^0)_i\beta \right].\hfill & & \end{array}$$
$`(2.3)`$
This charge turns out to be the generator for the transformations (2.2) if we exploit the following general relationship
$$\begin{array}{ccc}\delta _B\mathrm{\Phi }=i\eta [\mathrm{\Phi },Q_B]_\pm \hfill & & \end{array}$$
$`(2.4)`$
where $`[,]_\pm `$ stands for the (anti)commutator for the generic field $`\mathrm{\Phi }`$ being (fermionic)bosonic in nature. For the verification of (2.4), one has to use the canonical (anti)commutators for the Lagrangian density (2.1) as given below (with $`\mathrm{}=c=1`$)
$$\begin{array}{ccc}& & [B_{0i}(x,t),B^j(y,t)]=i\delta _i^j\delta (xy)[\beta (x,t),\dot{\overline{\beta }}(y,t)]=i\delta (xy)\hfill \\ & & [B_{ij}(x,t),H^{0kl}(y,t)]=i(\delta _i^k\delta _j^l\delta _i^l\delta _j^k)\delta (xy)\hfill \\ & & [\varphi _1(x,t),B_0(y,t)]=i\delta (xy)[\overline{\beta }(x,t),\dot{\beta }(y,t)]=i\delta (xy)\hfill \\ & & \{C_0(x,t),\rho (y,t)\}=i\delta (xy)\{\overline{C}_0(x,t),\lambda (y,t)\}=i\delta (xy)\hfill \\ & & \{C_i(x,t),\mathrm{\Pi }_C^j(y,t)\}=i\delta _i^j\delta (xy)\{\overline{C}_i(x,t),\mathrm{\Pi }_{\overline{C}}^j(y,t)\}=i\delta _i^j\delta (xy)\hfill \end{array}$$
$`(2.5)`$
where $`\delta (xy)`$ is the Dirac- delta function in 3D of space (i.e., $`\delta ^{(3)}(𝐱𝐲)`$) and the expression for the canonical momenta are:
$$\begin{array}{ccc}\mathrm{\Pi }_{(C)}^{(i)}=(^0\overline{C}^i^i\overline{C}^0)\mathrm{\Pi }_{(\overline{C})}^{(i)}=(^0C^i^iC^0).\hfill & & \end{array}$$
$`(2.6)`$
All the rest of the (anti)commutators are zero.
It can be readily seen that the ghost part of the Lagrangian density has the following discrete symmetry invariance
$$\begin{array}{ccc}& & \beta i\overline{\beta }C_\mu \pm i\overline{C}_\mu \rho \pm i\lambda \hfill \\ & & \overline{\beta }\pm i\beta \overline{C}_\mu \pm iC_\mu \lambda \pm i\rho .\hfill \end{array}$$
$`(2.7)`$
As a result of this symmetry, one can define an anti-BRST charge $`Q_{AB}`$ from (2.3) and one can obtain anti-BRST symmetry from (2.2) by exploiting (2.7). Furthermore, the total Lagrangian density (2.1) remains invariant under the following transformations:
$$\begin{array}{ccc}& & B_{\mu \nu }B_{\mu \nu }\varphi _1\varphi _1B_\mu B_\mu \hfill \\ & & \beta e^{2\mathrm{\Sigma }}\beta C_\mu e^\mathrm{\Sigma }C_\mu \lambda e^\mathrm{\Sigma }\lambda \hfill \\ & & \overline{\beta }e^{2\mathrm{\Sigma }}\overline{\beta }\overline{C}_\mu e^\mathrm{\Sigma }\overline{C}_\mu \rho e^\mathrm{\Sigma }\rho \hfill \end{array}$$
$`(2.8)`$
where $`\mathrm{\Sigma }`$ is a global (spacetime independent) scale transformation parameter. This continuous symmetry leads to the derivation of a conserved ghost charge ($`Q_g`$) as
$$\begin{array}{ccc}Q_g=d^3x\left[C_i\mathrm{\Pi }_{(C)}^{(i)}+C_0\mathrm{\Pi }_{(C)}^{(0)}+2\beta \mathrm{\Pi }_\beta 2\overline{\beta }\mathrm{\Pi }_{\overline{\beta }}\overline{C}^0\mathrm{\Pi }_{(\overline{C})}^{(0)}\overline{C}_i\mathrm{\Pi }_{(\overline{C})}^{(i)}\right]\hfill & & \end{array}$$
$`(2.9)`$
where $`\mathrm{\Pi }`$s are the canonical momenta w.r.t. ghost fields Besides (2.6), the other canonical momenta are $`\mathrm{\Pi }_\beta =\dot{\overline{\beta }},\mathrm{\Pi }_{\overline{\beta }}=\dot{\beta },\mathrm{\Pi }_{(C)}^{(0)}=\rho ,\mathrm{\Pi }_{(\overline{C})}^{(0)}=\lambda .`$. It can be readily seen, by exploiting the canonical (anti)commutators of (2.5), that
$$\begin{array}{ccc}& & Q_B^2=\frac{1}{2}\{Q_B,Q_B\}=0Q_{AB}^2=\frac{1}{2}\{Q_{AB},Q_{AB}\}=0\hfill \\ & & \{Q_B,Q_{AB}\}=0i[Q_g,Q_B]=+Q_Bi[Q_g,Q_{AB}]=Q_{AB}.\hfill \end{array}$$
$`(2.10)`$
Thus, we note that $`Q_B`$ and $`Q_{AB}`$ are the nilpotent operators of order 2 (i.e., $`Q_B^2=Q_{AB}^2=0`$) and the ghost number for them is $`+1`$ and $`1`$ respectively. This ghost number will also have relevance with some aspects of differential geometry (see, e.g., Section 5). Though the conserved and nilpotent charge $`Q_B`$ is the analogue of the exterior derivative $`d`$ , the conserved and nilpotent charge $`Q_{AB}`$ is not the analogue of the co-exterior derivative $`\delta `$. This is due to the fact that the anticommutator between $`d`$ and $`\delta `$ is not equal to zero (i.e., $`\{d,\delta \}0`$) whereas $`Q_B`$ and $`Q_{AB}`$ anticommute ($`\{Q_B,Q_{AB}\}=0`$) with each-other. Furhermore, there is no analogue of the Laplacian operator $`\mathrm{\Delta }`$ in (2.10). This fact can be succinctly expressed as
$$\begin{array}{ccc}& & Q_B^2=0d^2=0Q_{AB}^2=0\delta ^2=0\hfill \\ & & \{Q_B,Q_{AB}\}=0\{d,\delta \}=\mathrm{\Delta }0.\hfill \end{array}$$
$`(2.11)`$
Recently, it has been pointed out that the cohomologically higher-order BRST- and anti-BRST operators do not anticommute and their anticommutator leads to the definition of a higher-order Laplacian operator for the compact non-Abelian Lie algebras . This argument does not apply here in our discussion of the Abelian 2-form gauge theory because here the Lie algebra is a trivial (Abelian) algebra. Furthermore, we do not consider here the higher-order cohomology discussed in Ref. .
3 Dual BRST symmetry
In this Section, we discuss the ‘dual’ BRST symmetry which leaves the gauge-fixing term of the Lagrangian density invariant. This nilpotent symmetry should be contrasted with the BRST symmetry (and also anti-BRST symmetry) where it is the curvature term $`H=dB`$, that remains invariant. Just as one linearizes the gauge fixing term by introducing an auxiliary field $`B_\mu `$ and a scalar field $`\varphi _1`$ in the case of BRST invaraint Lagrangian density (2.1), one can linearize the the kinetic energy term by incorporating another auxiliary field $`_\mu `$ and a different scalar field $`\varphi _2`$ to obtain the off-shell nilpotent dual BRST invariance of the same Lagrangian density By integrating out the linearizing field $`_\mu `$ and the scalar field $`\varphi _2`$, we get back the BRST invariant Lagrangian density (2.1).. Such a BRST- and dual BRST invariant Lagrangian density, incorporating the above linearizations, is
$$\begin{array}{ccc}_D\hfill & =& \frac{1}{2}^\mu _\mu \frac{1}{3!}\epsilon _{\mu \nu \lambda \zeta }^\mu H^{\nu \lambda \zeta }+^\mu _\mu \varphi _2\frac{1}{2}B^\mu B_\mu +B_\nu \left(_\mu B^{\mu \nu }^\nu \varphi _1\right)\hfill \\ & & _\mu \overline{\beta }^\mu \beta +\left(_\mu \overline{C}_\nu _\nu \overline{C}_\mu \right)^\mu C^\nu +\rho \left(_\mu C^\mu +\lambda \right)+\left(_\mu \overline{C}^\mu +\rho \right)\lambda .\hfill \end{array}$$
$`(3.1)`$
Under the following off-shell nilpotent ($`\delta _D^2=0`$) dual BRST symmetry transformations:
$$\begin{array}{ccc}& & \delta _DB_{\mu \nu }=\eta \epsilon _{\mu \nu \lambda \zeta }^\lambda \overline{C}^\zeta \delta _D\overline{C}_\mu =\eta _\mu \overline{\beta }\delta _DC_\mu =\eta _\mu \hfill \\ & & \delta _D\beta =\eta \lambda \delta _D\varphi _2=\eta \rho \delta _D(\overline{\beta },\lambda ,\rho ,\varphi _1,B_\mu ,_\mu )=0\hfill \end{array}$$
$`(3.2)`$
the Lagrangian density (3.1) transforms as:
$$\begin{array}{ccc}\delta _D_D=\eta _\mu \left[\rho ^\mu +\lambda ^\mu \overline{\beta }+(^\mu \overline{C}^\nu ^\nu \overline{C}^\mu )_\nu \right].\hfill & & \end{array}$$
$`(3.3)`$
Thus, the above Lagrangian density (3.1) remains invariant under the dual BRST transformations (3.2) and the BRST transformations (2.2) (together with $`\delta _B(_\mu ,\varphi _2)=0`$). It is appropriate to call the symmetry transformations (3.2) as the ‘dual’ BRST transformations because it is the gauge-fixing term (i.e., $`\delta B=_\mu B^{\mu \nu }dx_\nu `$: the Hodge dual of the curvature $`dB=H`$) of the theory that remains invariant and the kinetic energy term (which remains invariant under BRST- and anti-BRST symmetries) transforms under it to compensate for the transformation of the ghost terms. The Noether conserved current, derived from the above symmetry transformations, is:
$$\begin{array}{ccc}J_D^\alpha =\epsilon ^{\alpha \beta \rho \sigma }B_\beta _\rho \overline{C}_\sigma ^\alpha \rho \lambda ^\alpha \overline{\beta }(^\alpha C^\lambda ^\lambda C^\alpha )_\lambda \overline{\beta }(^\alpha \overline{C}^\lambda ^\lambda \overline{C}^\alpha )_\lambda \hfill & & \end{array}$$
$`(3.4)`$
which ultimately leads to the derivation of a conserved ($`\dot{Q}_D=0`$) and nilpotent ($`Q_D^2=0`$) dual BRST charge ($`Q_D=d^3xJ_D^0`$) as:
$$\begin{array}{ccc}Q_D=d^3x\left[\epsilon ^{0ijk}(B_i)_j\overline{C}_k_0\rho \lambda \dot{\overline{\beta }}(^0C^i^iC^0)_i\overline{\beta }(^0\overline{C}^i^i\overline{C}^0)_i\right].\hfill & & \end{array}$$
$`(3.5)`$
To prove the conservation law for the Noether current in (3.4), one has to use some of the following equations of motion derived from the Lagrangian density (3.1)
$$\begin{array}{ccc}& & B=0=0\mathrm{}\varphi _1=\mathrm{}\varphi _2=0B_\mu =^\rho B_{\rho \mu }_\mu \varphi _1\hfill \\ & & _\mu =\frac{1}{3!}\epsilon _{\mu \nu \lambda \xi }H^{\nu \lambda \xi }_\mu \varphi _2\mathrm{}\rho =\mathrm{}\lambda =\mathrm{}\beta =\mathrm{}\overline{\beta }=0\hfill \\ & & \mathrm{}C^\mu ^\mu (C)+^\mu \lambda =0_\mu C^\mu +2\lambda =0,\hfill \\ & & \mathrm{}\overline{C}^\mu ^\mu (\overline{C})+^\mu \rho =0_\mu \overline{C}^\mu +2\rho =0\hfill \\ & & \mathrm{}B_\mu _\mu (B)=0\mathrm{}B_\mu =0\hfill \\ & & \mathrm{}_\mu _\mu ()=0\mathrm{}_\mu =0\hfill \\ & & \epsilon _{\mu \nu \lambda \xi }^\lambda ^\xi +(_\mu B_\nu _\nu B_\mu )=0.\hfill \end{array}$$
$`(3.6)`$
As the ghost part of the Lagrangian density (3.1) remains invariant under (2.7), it is very interesting to note that the bosonic part of this Lagrangian density remains invariant under the following discrete symmetry transformations
$$\begin{array}{ccc}& & _\mu iB_\mu \varphi _2i\varphi _1\varphi _1\pm i\varphi _2\hfill \\ & & B_\mu \pm i_\mu B_{\mu \nu }\frac{i}{2}\epsilon _{\mu \nu \lambda \xi }B^{\lambda \xi }.\hfill \end{array}$$
$`(3.7)`$
It is straightforward to check that that the total Lagrangian density (3.1) remains invariant under the combination of discrete symmetry transformations (2.7) and (3.7). We note here that the analogue of Hodge $``$ operation of differential geometry turns out to be the combined symmetries (2.7) and (3.7). This assertion can be verified by the validity of the following relation
$$\begin{array}{ccc}\delta _D\left(\mathrm{\Phi }\right)=\pm \delta _B\left(\mathrm{\Phi }\right)\hfill & & \end{array}$$
$`(3.8)`$
where $`(+)`$ stands for the generic field $`\mathrm{\Phi }`$ being (bosonic) fermionic in nature, $`\delta _D`$ and $`\delta _B`$ are the nilpotent transformations (2.2) and (3.2) and $``$ operation is the discrete transformations (2.7) and (3.7). Thus, we note that the dual BRST and BRST variations (on a field) are related to each-other in the same way as the action of an exterior derivative $`d`$ and co-exterior derivative $`\delta =\pm d`$ on a given differential form. This symmetry is also reflected in the expressions for BRST- and dual BRST charges. In fact, it can be readily seen that under the transformations (2.7) and (3.7), one obtains the following changes for these conserved and nilpotent charges:
$$\begin{array}{ccc}Q_BQ_DQ_DQ_B.\hfill & & \end{array}$$
$`(3.9)`$
In the language of symmetry transformations, this fact can be translated into: $`\delta _B(\mathrm{\Phi })\delta _D(\mathrm{\Phi }),\delta _D(\mathrm{\Phi })\delta _B(\mathrm{\Phi })`$ under (2.7) and (3.7). Here $`\mathrm{\Phi }`$ is the generic field representing bosonic as well as fermionic variables of the theory. It is interesting to note the similarity between relations (3.9) and the usual electro-magnetic duality present in the case of Maxwell equations (for $`U(1)`$ gauge theory) where $`𝐄𝐁,𝐁𝐄`$ under global duality transformations (see, e.g., Ref. ).
The existence of discrete symmetry for the ghost action, allows one to define an anti-dual BRST charge $`Q_{AD}`$ from the expression for $`Q_D`$ in (3.5). The off-shell nilpotent transformations generated by $`Q_{AD}`$ can be also derived from (3.2) by exploiting (2.7). Now, it is evident that the total Lagrangian density (3.1) respects four nilpotent symmetries which are generated by (anti) BRST- and (anti) dual BRST charges. The exact expressions for these charges for the Lagrangian density (3.1) are
$$\begin{array}{ccc}Q_B=d^3x\left[\epsilon ^{0ijk}_i_jC_k+B_0\lambda \rho \dot{\beta }+(^0C^i^iC^0)B_i(^0\overline{C}^i^i\overline{C}^0)_i\beta \right]\hfill & & \end{array}$$
$`(3.10)`$
$$\begin{array}{ccc}Q_D=d^3x\left[\epsilon ^{0ijk}B_i_j\overline{C}_k_0\rho \lambda \dot{\overline{\beta }}(^0C^i^iC^0)_i\overline{\beta }(^0\overline{C}^i^i\overline{C}^0)_i\right]\hfill & & \end{array}$$
$`(3.11)`$
$$\begin{array}{ccc}Q_{AB}=id^3x\left[\epsilon ^{0ijk}_i_j\overline{C}_k+B_0\rho +i\lambda \dot{\overline{\beta }}+(^0\overline{C}^i^i\overline{C}^0)B_i+i(^0C^i^iC^0)_i\overline{\beta }\right]\hfill & & \end{array}$$
$`(3.12)`$
$$\begin{array}{ccc}Q_{AD}=id^3x\left[\epsilon ^{0ijk}B_i_jC_k_0\lambda i\rho \dot{\beta }(^0C^i^iC^0)_ii(^0\overline{C}^i^i\overline{C}^0)_i\beta \right].\hfill & & \end{array}$$
$`(3.13)`$
4 Bosonic symmetry
It is evident that the total Lagrangian density $`_D`$ in (3.1) is endowed with four nilpotent symmetry transformations that are generated by the conserved and nilpotent charges (3.10–3.13). It is logical to expect that the anticommutator of the pair of these symmetries would also be the symmetry for (3.1). Since four anticommutators ($`\{Q_B,Q_{AB}\}=0,\{Q_D,Q_{AD}\}=0,\{Q_B,Q_{AD}\}=0,\{Q_D,Q_{AB}\}=0`$) are zero, the other two anticommutators ($`\{Q_B,Q_D\},\{Q_{AB},Q_{AD}\}`$) would lead to the definition of a bosonic operator $`W`$ which will generate a symmetry transformation $`\delta _W`$ for (3.1). The following transformations generated by the operator $`W`$ (with $`\kappa =i\eta \eta ^{}`$)
$$\begin{array}{ccc}& & \delta _WB_{\mu \nu }=i\kappa (_\mu _\nu _\nu _\mu +\epsilon _{\mu \nu \lambda \xi }^\lambda B^\xi )\delta _W\varphi _1=0\delta _W\varphi _2=0,\delta _WB_\mu =0\hfill \\ & & \delta _WC_\mu =i\kappa _\mu \lambda \delta _W\overline{C}_\mu =i\kappa _\mu \rho \delta _W\rho =0,\delta _W\lambda =0\delta _W_\mu =0\hfill \\ & & \delta _W(C)=i\kappa \mathrm{}\lambda \delta _W(\overline{C})=i\kappa \mathrm{}\rho \delta _W(^\rho B_{\rho \mu })=i\kappa (\mathrm{}_\mu _\mu ())\hfill \\ & & \delta _W\left(\frac{1}{3!}\epsilon _{\mu \nu \lambda \xi }H^{\nu \lambda \xi }\right)\delta _W\left(\frac{1}{2}\epsilon _{\mu \nu \lambda \xi }^\nu B^{\lambda \xi }\right)=i\kappa (\mathrm{}B_\mu _\mu (B))\delta _W\beta =0\delta _W\overline{\beta }=0\hfill \end{array}$$
$`(4.1)`$
turn out to be the symmetry transformations for $`_D;`$
$$\begin{array}{ccc}\delta _W_D\hfill & =& i\kappa _\alpha [X^\alpha ]\hfill \\ X^\alpha \hfill & =& \rho ^\alpha \lambda ^\alpha \rho \lambda +B^\rho ^\alpha _\rho \hfill \\ & & ^\rho ^\alpha B_\rho +^\alpha (B)B^\alpha ().\hfill \end{array}$$
$`(4.2)`$
Here $`\eta `$ and $`\eta ^{}`$ (in the definition of $`\kappa `$) are the fermionic spacetime indpendent parameters in the transformations corresponding to $`\delta _B`$ and $`\delta _D`$ of eqns. (2.2) and (3.2). The Noether conserved current corresponding to the transformations (4.1), is
$$\begin{array}{ccc}J_W^\alpha \hfill & =& i\epsilon ^{\alpha \beta \rho \sigma }\left(_\beta _\rho _\sigma +B_\beta _\rho B_\sigma \right)+i_\rho \left(^\rho B^\alpha ^\alpha B^\rho \right)\hfill \\ & +& i(^\alpha C^\lambda ^\lambda C^\alpha )_\lambda \rho +i(^\alpha \overline{C}^\mu ^\mu \overline{C}^\alpha )_\mu \lambda \hfill \end{array}$$
$`(4.3)`$
which finally leads to the derivation of a local conserved charge ($`W=d^3xJ_W^0`$) as
$$\begin{array}{ccc}W=id^3x\left[\epsilon ^{0ijk}\left(B_i_jB_k+_i_j_k\right)+(^0C^i^iC^0)_i\rho +(^0\overline{C}^i^i\overline{C}^0)_i\lambda \right].\hfill & & \end{array}$$
$`(4.4)`$
This charge can be directly computed from the anticommutators of $`\{Q_B,Q_D\}`$ or $`\{Q_{AB},Q_{AD}\}`$ by exploiting the analogue of canonical (anti)commutators in (2.5) for the Lagrangian density (3.1). In fact, all the (anti)commuators of (2.5) remain intact except the fact that now the canonical momenta w.r.t. $`B_{kl}`$ becomes $`\epsilon ^{0klm}_m`$ instead of $`H^{0kl}`$. Thus, one has to replace now $`H^{0kl}`$ by $`\epsilon ^{0klm}_m`$ in one of the commutators of eqn. (2.5).
There are simpler ways to compute this generator $`W`$ for the bosonic symmetry transformations in (4.1). Since the conserved and nilpotent charges in (3.10–3.13) are the generators of the nilpotent transformations, it can be readily seen that the following equations
$$\begin{array}{ccc}\delta _BQ_D\hfill & =& i\eta \{Q_D,Q_B\}=i\eta W\hfill \\ \delta _DQ_B\hfill & =& i\eta \{Q_B,Q_D\}=i\eta W\hfill \\ \delta _{AB}Q_{AD}\hfill & =& i\eta \{Q_{AD},Q_{AB}\}=i\eta W\hfill \\ \delta _{AD}Q_{AB}\hfill & =& i\eta \{Q_{AB},Q_{AD}\}=i\eta W\hfill \end{array}$$
$`(4.5)`$
can be exploited to derive $`W`$ from the expressions of charges in (3.10–3.13) and the transformations (2.2) and (3.2). It will be noticed that here $`\delta _{AB}`$ and $`\delta _{AD}`$ correspond to anti-BRST- and anti-dual BRST transformations that can be easily derived from eqns. (2.2) and (3.2). It is straightforward to check that $`\delta _DQ_B=i\eta W`$ leads to:
$$\begin{array}{ccc}W=id^3x\left[\epsilon ^{0ijk}_i_j_k+\rho \dot{\lambda }+(^0^i^i^0)B_i+(^0\overline{C}^i^i\overline{C}^0)_i\lambda \right].\hfill & & \end{array}$$
$`(4.6)`$
We can also obtain an expression for $`W`$ from the expression for $`Q_D`$ by applying the transformations $`\delta _B`$ (i.e., $`\delta _BQ_D=i\eta W`$) as given below
$$\begin{array}{ccc}W=id^3x\left[\epsilon ^{0ijk}B_i_jB_k+\lambda \dot{\rho }(^0B^i^iB^0)_i+(^0C^i^iC^0)_i\rho \right].\hfill & & \end{array}$$
$`(4.7)`$
It is obvious that the expressions (4.6) and (4.7) bear a different outlook than the expression derived in (4.4). All these expressions for $`W`$ are, however, identical if we exploit the appropriate equations of motion. Similar expressions emerge from the calculations of other expressions in (4.5). The most concise form of $`W`$ that can be derived from (4.5), is
$$\begin{array}{ccc}W=id^3x\left[\epsilon ^{0ijk}\left(B_i_jB_k+_i_j_k\right)+\lambda \dot{\rho }+\rho \dot{\lambda }\right].\hfill & & \end{array}$$
$`(4.8)`$
It will be noticed that we have exploited here only the off-shell nilpotent symmetries (and conserved charges) for the derivation of $`W`$.
One important point to be noticed here is the fact that the operator $`W`$ does not go to zero if we exploit the equations of motion. This feature is completely different from the discussion of the free 2D (non)Abelian gauge theories in Refs. \[18-20\] where it has been argued that the topological nature of these theories is encoded in the vanishing of the operator $`W`$ when equations of motion are used and all the fields are assumed to fall off rapidly at infinity.
5 Extended BRST algebra
In this Section, we concentrate on the derivation of an extended BRST algebra (which is found to be constituted by six conserved charges) and provide a possible connection of this algebra with the algebra of the de Rham cohomology operators of differential geometry. In the normal BRST algebra, there are three conserved charges (viz., $`Q_g,Q_B,Q_{AB}`$) of equations (2.9), (3.10) and (3.12). The existence of new symmetries, however, provide three more conserved charges (viz., $`Q_D,Q_{AD}`$ and $`W`$) which are given by equations (3.11), (3.13) and (4.8). If one exploits the canonical (anti)commutators of equation (2.5) for the Lagrangian density (3.1), one can show that all the six conserved charges obey the following extended BRST algebra
$$\begin{array}{ccc}& & [W,Q_k]=0k=g,B,AB,D,AD\hfill \\ & & Q_B^2=0Q_D^2=0Q_{AB}^2=0Q_{AD}^2=0\hfill \\ & & i[Q_g,Q_B]=+Q_Bi[Q_g,Q_D]=Q_Di[Q_g,Q_{AB}]=Q_{AB}\hfill \\ & & i[Q_g,Q_{AD}]=+Q_{AD}\{Q_B,Q_{AB}\}=0\{Q_D,Q_{AD}\}=0\hfill \\ & & \{Q_B,Q_{AD}\}=0\{Q_D,Q_{AB}\}=0\{Q_B,Q_D\}=\{Q_{AB},Q_{AD}\}=W.\hfill \end{array}$$
$`(5.1)`$
A few comments are in order. First of all, it is trivial to see that the operator $`W`$ is the Casimir operator for the whole extended BRST algebra. Secondly, there are four nilpotent (of order two) charges in the extended BRST algebra. Thirdly, two anticommutators (viz. $`\{Q_B,Q_D\},\{Q_{AB},Q_{AD}\}`$) lead to the definition of of the Casimir operator $`W`$. And, lastly, the ghost number for charges $`Q_B`$ and $`Q_{AD}`$ is $`+1`$ and that of $`Q_D`$ and $`Q_{AB}`$ is $`1`$. There are simpler ways to check the validity of the above statements. For instance, exploiting the symmetry transformations of eqns. (2.2), (3.2), (4.1) and an infinitesimal version of (2.8), it can be easily seen that
$$\begin{array}{ccc}& & \delta _BW=0\delta _DW=0\delta _gW=0\hfill \\ & & \delta _{AB}W=0\delta _{AD}W=0\delta _WW=0\hfill \end{array}$$
$`(5.2)`$
where the expression for the $`W`$ operator can be taken to be its most concise form of eqn. (4.8). Similarly other expressions for the (anti)commutators in (5.1) can be checked by merely using the symmetry transformation properties and the expressions for the conserved charges.
Next we present arguments to bring out the analogy between symmetry generators of this field theoretical model and the de Rham cohomology operators. It is a well known fact that the de Rham cohomology operators ($`d,\delta ,\mathrm{\Delta }`$) obey the following algebra
$$\begin{array}{ccc}& & d^2=0\delta ^2=0\mathrm{\Delta }=(d+\delta )^2=d\delta +\delta d\hfill \\ & & [\mathrm{\Delta },d]=0[\mathrm{\Delta },\delta ]=0\mathrm{\Delta }=\{d,\delta \}.\hfill \end{array}$$
$`(5.3)`$
Furthermore, a differential form of degree $`n`$ ($`f_n`$) becomes a differential form of degree $`n+1`$ ($`f_{n+1}`$) due to the application of operator $`d`$ (i.e., $`df_nf_{n+1}`$). In contrast, the operator $`\delta `$ reduces the degree of a form by one (i.e., $`\delta f_nf_{n1}`$) on which it acts and the Laplacian operator $`\mathrm{\Delta }`$ does not change the degree of the form (i.e., $`\mathrm{\Delta }f_nf_n`$). Now we observe that the ghost number of the state is parallel to the degree of the differential form and $`Q_B,Q_D`$ and $`W`$ play respectively the role of $`d,\delta `$ and $`\mathrm{\Delta }`$ in differential geometry. Exploiting the algebra (5.1), it can be readily seen that a state $`|\psi >_n`$ with ghost number $`n`$ (i.e., $`iQ_g|\psi >_n=n|\psi >_n`$) in the quantum Hilbert space will imply that the ghost number for the states $`Q_B|\psi >_n,Q_D|\psi >_n,W|\psi >_n`$ is $`(n+1),(n1),n`$ respectively. This fact can be succinctly expressed as
$$\begin{array}{ccc}iQ_gQ_B|\psi >_n\hfill & =& (n+1)Q_B|\psi >_n\hfill \\ iQ_gQ_D|\psi >_n\hfill & =& (n1)Q_D|\psi >_n\hfill \\ iQ_gW|\psi >_n\hfill & =& nW|\psi >_n\hfill \\ iQ_gQ_{AB}|\psi >_n\hfill & =& (n1)Q_{AB}|\psi >_n\hfill \\ iQ_gQ_{AD}|\psi >_n\hfill & =& (n+1)Q_{AD}|\psi >_n.\hfill \end{array}$$
$`(5.4)`$
Thus, now one can draw a parallel between the differential geometry (and the corresponding de Rham cohomology operators) defined on a compact manifold and the quantum states, conserved charges, etc., defined in the quantum Hilbert space of states. For instance, the differential forms are just like quantum states; a closed form ($`df=0`$) is just like a BRST closed (physical) state ($`Q_B|\psi >=0`$); a compact manifold is just like the quantum Hilbert space of states; degree of a form is analogous to the ghost number and the de Rham cohomolgy operators ($`d,\delta ,\mathrm{\Delta }`$) have their counterpart as conserved charges ($`Q_B,Q_D,W`$) and ($`Q_{AD},Q_{AB},W`$), etc. It is a very special feature of the BRST formalism that each of the de Rham cohomology operators $`d,\delta `$ can be identified with two symmetry generators. This, in turn, implies that irrespective of the nature (i.e., real or complex) of the compact manifold, its counterpart—- the quantum Hilbert space of states—- is always complex so that $`d`$ and $`\delta `$ have two representations and the analogue of the Laplacian operator (i.e., $`W`$) can also be expressed in two different ways (i.e., $`\{Q_B,Q_D\}=\{Q_{AB},Q_{AD}\}=W`$). However, if we retrace back, the full strength of the BRST cohomology and Hodge decomposition theorem implies that the compact manifold has to be a complex manifold so that one can achieve a complete analogy with BRST formalism. In other words, it should be possible to define ($`d,\overline{d}`$), ($`\delta ,\overline{\delta }`$) and ($`\mathrm{\Delta },\overline{\mathrm{\Delta }}`$) on the compact manifold so that cohomology operators $`\mathrm{\Delta }=\overline{\mathrm{\Delta }}`$ and $`\mathrm{\Delta }=d\delta +\delta d\overline{d}\overline{\delta }+\overline{\delta }\overline{d}`$ can be constructed on this manifold.
6 Constraint analysis
In this Section, we first discuss the Hodge decomposition theorem for a given state $`|\psi >_n`$ (with ghost number $`n`$) in the quantum Hilbert space of states. This is, then, followed by the discussion of constraints on the physical (harmonic) states by the imposition of the physicality criteria with conserved and nilpotent charges (i.e., $`Q_B|phys>=0,Q_D|phys>=0`$) which define the physical subspace of states in the total quantum Hilbert space of states. It is obvious from the algebra (5.1) and the ghost number analysis in eqn. (5.4) that, now any arbitrary state $`|\psi >_n`$ in the quantum Hilbert space of states can be written as
$$\begin{array}{ccc}|\psi >_n=|\omega >_n+Q_B|\theta >_{n1}+Q_D|\chi >_{n+1}\hfill & & \end{array}$$
$`(6.1)`$
where $`|\omega >_n`$ is the harmonic state (i.e., $`W|\omega >_n=0,Q_B|\omega >_n=0,Q_D|\omega >_n=0`$), $`Q_B|\theta >_{n1}`$ is a BRST exact state and $`Q_D|\chi >_{n+1}`$ is a co-BRST exact state. This equation is just the analogue of the Hodge decomposition theorem (1.1) written for a differential form in terms of the de Rham cohomology operators ($`d,\delta ,\mathrm{\Delta }`$) defined on a compact manifold. It is a special feature of the BRST formalism (and the corresponding extended BRST algebra (5.1)) that eqn. (1.1) can also be expressed in terms of the conserved and nilpotent charges $`Q_{AB}`$ and $`Q_{AD}`$ as
$$\begin{array}{ccc}|\psi >_n=|\omega >_n+Q_{AD}|\theta >_{n1}+Q_{AB}|\chi >_{n+1}\hfill & & \end{array}$$
$`(6.2)`$
where $`Q_{AB}`$ and $`Q_{AD}`$ are the anti-BRST and anti-dual BRST charges.
Unlike the uniqueness of the Hodge decomposition in the mathematical aspects of the de Rham cohomology, the uniqueness of the corresponding decomposition of the quantum states (cf. eqns. (6.1, 6.2)) is not obvious in the quantum Hilbert space of states.
It is a noteworthy point that the combined discrete transformations (2.7) and (3.7), turn out to be the symmetry of the Lagrangian density of the theory under discussion. It is obvious from our earlier arguments that this symmetry corresponds to the Hodge duality operation (i.e., $``$ operation) in differential geometry. Thus, we have a theory which is duality invariant due to the presence of the discrete symmetries (2.7) and (3.7). As a result, the vacuum- and physical states of the quantum theory should be also duality (i.e., BRST and dual BRST) invariant in the quantum Hilbert space of states. This feature, in fact, has been exploited in Refs. \[18-20\] to establish the topological nature of the 2D free (non)Abelian gauge theory. In the BRST formalism, physical states are those states that are annihilated by $`Q_B`$ (i.e. $`Q_B|phys>=0`$). Due to the presence of the discrete symmetry, it is obvious that $`Q_B`$ goes to $`Q_D`$ (cf. (3.9)) and hence the latter also annihilate the physical states (i.e., $`Q_B|phys>=0Q_D|phys>=0`$). These two together imply that the Casimir operator $`W`$ also annhilates the physical states. It is, therefore, clear that the physical states are the harmonic states. Of course, the vacuum state will be annihilated by all these charges, as they are the generators of the unbroken symmetry transformations<sup>∗∗</sup>. Thus, these states satisfy
$$\begin{array}{ccc}& & W|vac>=0Q_B|vac>=0Q_D|vac>=0\hfill \\ & & Q_{AB}|vac>=0Q_{AD}|vac>=0\hfill \\ & & W|phys>=0Q_B|phys>=0Q_D|phys>=0\hfill \\ & & Q_{AB}|phys>=0Q_{AD}|phys>=0.\hfill \end{array}$$
$`(6.3)`$
The conditions $`iQ_g|phys>=0`$ and $`iQ_g|vac>=0`$ imply that the ghost number of the physical- and vacuum states is zero. No other constraint emerge on physical states due to the existence of ghost charge $`Q_g`$. It will be noted that the conditions ($`Q_B|phys>=0,Q_{AB}|phys>=0`$) lead to the one and the same constraints on the physical state. Thus, we can choose one of them for the constraint analysis. Similar argument holds for the conditions $`Q_D|phys>=0`$ and $`Q_{AD}|phys>=0`$ and one can choose only one of these charges for the discussion of constraints. Thus, we see that the vacuum, as well as the physical (harmonic) states, of the theory respect three basic symmetries (cf. (6.3)) and the ghost number for them is zero. It will be noticed that these conclusions are arrived at by the symmetry considerations alone.
Before we concentrate on the constraint analysis of the Lagrangian density in (3.1), we shall dwell a bit on the nature of constraints for the original Lagrangian density $``$ of eqn. (1.2). It is evident that the canonical momenta w.r.t. the antisymmetric field $`B_{\mu \nu }`$ is:
<sup>∗∗</sup> If the discrete transformations (2.7) and (3.7) (which relate $`Q_B`$and $`Q_D`$) are not the symmetry of the Lagrangian density, the physical (harmonic) states can be assumed to be annihilated independently by the BRST and the dual BRST charges.
In what follows, we shall concentrate on the set of operators $`Q_B,Q_D,W`$ for the discussion of the Hodge decomposition theorem as well as the constraint analysis. However, our arguments and analysis will be valid for the set of operators: $`Q_{AB},Q_{AD},W`$ as well.
$`\mathrm{\Pi }^{\mu \nu }=H^{0\mu \nu }`$ and the equations of motion are: $`_\mu H^{\mu \nu \lambda }=0`$. Thus, it is clear that $`\mathrm{\Pi }^{0i}0`$ is the primary constraint and the secondary constraint is nothing but the equation of motion w.r.t. $`B_{0i}`$ field, i.e., $`_jH^{oij}_j\mathrm{\Pi }^{ij}0`$. Both these constraints are first-class in the language of Dirac and they imply the existence of a gauge symmetry in the theory. For the consistent quantization of this theory, it is essential that $`\mathrm{\Pi }^{0i}|phys>=0,_j\mathrm{\Pi }^{ij}|phys>=0`$ (Dirac’s presecription). We shall see that exactly these constraints will appear when we shall demand: $`Q_B|phys>=0`$ (for the Lagrangian density (3.1)). Its dual description will emerge from the requirement: $`Q_D|phys>=0`$.
It can be readily seen that the requirement $`Q_B|phys>=0`$, for the Lagrangian density (3.1), leads to the following constraints on the theory:
$$\begin{array}{ccc}\mathrm{\Pi }^{0i}(=B^i)|phys>=0\hfill & & (_\rho B^{\rho i}^i\varphi _1)|phys>=0\hfill \\ _j\mathrm{\Pi }^{ij}(=_0B^i)|phys>=0\hfill & & (\epsilon ^{oijk}_j_k)|phys>=0\hfill \end{array}$$
$`(6.4)`$
where the expression for $`Q_B`$ has been taken from eqn. (3.10) and equations of motion from (3.6) have been used for the above derivation. Furthermore, it has been assumed here that the ghost fields, present in the expression for $`Q_B`$, do not lead to any constraints on the physical states of the theory. It is evident that in the above equation, we retrieve the constraints of the original gauge theory described by the Lagrangian density (1.2). Now the requirement $`Q_D|phys>=0`$ leads to
$$\begin{array}{ccc}(^i)|phys>=0\hfill & & (\frac{1}{2}\epsilon ^{i\nu \lambda \xi }_\nu B_{\lambda \xi }^i\varphi _2)|phys>=0\hfill \\ (_0^i)|phys>=0\hfill & & (+\epsilon ^{oijk}_jB_k)|phys>=0.\hfill \end{array}$$
$`(6.5)`$
Exploiting eqn. (3.7) of the duality transformations for the bosonic part of the Lagrangian density, it can be checked that the above constraints in (6.5) are just the ‘dual description’ of the constraints obtained in (6.4), though they appear different.
It will be noticed that even though the auxiliary field $`B_0`$ is present in the expression for $`Q_B`$, we have not written $`Q_B|phys>=0`$ implies $`B_0|phys>=0`$. This is because of the fact that $`B_0`$ is a conserved quantity and it remains the same w.r.t. time evolution. In fact, it can be easily seen that the quantity: $`I_0=d^3xB_0`$ is a time evolution invariant operator due to equations of motion in (3.6) (i.e., $`_0B_0=_iB_i`$). Thus, $`B_0|phys>=0`$ is a trivial constraint on the theory. Similarly, we have not concluded from the restriction $`Q_D|phys>=0`$, the obvious constraint $`_0|phys>=0`$ as there is no evolution for the $`_0`$ field due to $`=0`$ (cf. (3.6)). Strictly speaking, however, these constraints should be incorporated in (6.4) and (6.5) respectively. In fact, these finally imply that $`\mathrm{\Pi }_{\varphi _1}(=B_0)|phys>=0`$ and $`\mathrm{\Pi }_{\varphi _2}(=_0)|phys>=0`$. More precisely, the constraints $`B_0|phys>=0,B_i|phys>=0`$ and its counterpart $`_0|phys>=0,_i|phys>=0`$ together imply that:
$$\begin{array}{ccc}(_\mu )|phys>=0\hfill & & (\frac{1}{2}\epsilon ^{\mu \nu \lambda \xi }_\nu B_{\lambda \xi }^\mu \varphi _2)|phys>=0\hfill \\ (B_\mu )|phys>=0\hfill & & (_\rho B^{\rho \mu }^\mu \varphi _1)|phys>=0.\hfill \end{array}$$
$`(6.6)`$
This shows that the total gauge-fixing term $`(^\rho B_{\rho \mu }_\mu \varphi _1`$) and its dual annihilate the physical states of the theory. These conditions gauge away some of the degrees of freedom of the $`B_{\mu \nu }`$ gauge field. It is straightforward to see that the constraints $`W|phys>=0`$ does not lead to any new restrictions on the physical state. In fact, it encompasses both the constraints given in eqns. (6.4) and (6.5) due to $`Q_B|phys>=0`$ and $`Q_D|phys>=0`$. This is due to the fact that $`W=\{Q_B,Q_D\}`$ and $`W|phys>=0`$ implies that $`Q_B|phys>=0`$ and $`Q_D|phys>=0`$ which are in some sense, unique solutions to the constraint $`W|phys>=0`$. It should be recalled that in the discussion of the de Rham cohomology operators and the Hodge decomposition theorem, one says that the definition of the harmonic form $`h`$ ($`\mathrm{\Delta }h=0`$) implies that $`h`$ is closed ($`dh=0`$) and co-closed ($`\delta h=0`$) together (see, e.g., Refs. ). We note that similar conclusions can be drawn here from the properties of the set of local and conserved charges $`W,Q_B`$ and $`Q_D`$ (or the set $`W,Q_{AD}`$ and $`Q_{AB}`$).
7 Summary and discussion
We have shown that the BRST invariant two-form gauge theory in four ($`3+1`$) dimensions has an additional nilpotent symmetry, called the dual BRST, which keeps the gauge fixing term invariant. The anti-commutator of both the nilpotent charges (viz., $`Q_B`$ and $`Q_D`$) is the generator ($`W`$) of a bosonic symmetry transformation, under which, the ghost terms remain invariant. We can see the parallel between the BRST and the dual-BRST symmetry: The nilpotent (anti)BRST symmetry transformations leave the kinetic energy term (more precisely the curvature term) of the free Abelian 2-form gauge theory invariant. On the other hand, it is the gauge-fixing term that remains invariant under the (anti)dual BRST symmetry transformations. Another parallel is: Like the BRST invariant Lagrangian density (3.1) can be written as the sum of kinetic energy- and the BRST exact terms i.e, $`_{KE}+\frac{1}{\eta }\delta _B(F)`$ (where $`F`$ is a function of the local fields), in the same way, we can also express (3.1) as the sum of gauge-fixing and the co-BRST exact parts i.e, $`_{GF}+\frac{1}{\eta }\delta _D(G)`$, namely;
$$\begin{array}{ccc}_D\hfill & =& B_\nu (_\mu B^{\mu \nu }^\nu \varphi _1)+\frac{1}{\eta }\delta _D\left(G\right)\hfill \\ G\hfill & =& \frac{1}{2}C_\mu ^\mu \frac{1}{6}\epsilon _{\mu \nu \lambda \sigma }C^\mu H^{\nu \lambda \sigma }(^\mu C_\mu +\lambda )\varphi _2(_\mu \overline{C}^\mu +\rho )\beta .\hfill \end{array}$$
$`(7.1)`$
We have exploited the above symmetries to construct a field theoretical model for the Hodge theory on the four dimensional Minkowskian manifold where all the de Rham cohomology operators $`(d,\delta ,\mathrm{\Delta })`$ have their counterparts as the conserved and nilpotent charges (corresponding to $`d`$ and $`\delta `$) and the bosonic conserved charge $`W`$ (corresponding to the Laplacian operator $`\mathrm{\Delta }`$) for the BRST invariant version of the free 2-form Abelian gauge theory. All these charges are local and they generate the symmetry transformations for the Lagrangian density of this theory. In the framework of the BRST formalism, it turns out that the analogue of the Laplacian operator (i.e., $`W`$) can be represented in two different ways ($`W=\{Q_B,Q_D\}=\{Q_{AB},Q_{AD}\}`$). Thus, $`d`$ and $`\delta `$ have two representations (i.e., $`dQ_B,Q_{AD},\delta Q_D,Q_{AB}`$) in terms of the nilpotent charges.
The bosonic symmetry generator W (anticommutator of $`Q_B`$ and $`Q_D`$) turns out to be the Casimir operator for the extended BRST algebra. Under the transformation, generated by the Casimir operator, all the fermionic fields either do not transform or transform by a vector gauge transformation (e.g., $`\delta _WC_\mu =i\kappa _\mu \lambda ,\delta _W\overline{C}_\mu =i\kappa _\mu \rho `$). It will also be noticed that all the gauge-fixing terms, for the bosonic as well as the fermionic fields, transform to the equations of motion under this transformations (cf. eqn. (4.1)). There exists a discrete symmetry transformation in the theory (cf. eqns. (2.7) and (3.7)) which behaves like the analogue of the Hodge $``$ operation. In fact, it relates the nilpotent transformations $`\delta _D`$ and $`\delta _B`$ in a similar fashion as there exists a relationship between the dual-exterior derivative $`\delta `$ ($`\delta =\pm d`$) and the exterior derivative $`d`$ in differential geometry.
We summarise the main results : (i) We have found out a possible mapping between the de Rahm cohomology operators of differential geometry and the symmetry generators of a $`3+1`$ dimensional field theoretical model for the Hodge theory. (ii) We have shown the existence of a mapping between Hodge $``$ operation and the discrete transformations on the fields of the theory. Both these mappings can be concisely expressed as
$$\begin{array}{ccc}\mathrm{Exterior}\mathrm{derivative}d\hfill & & Q_B,Q_{AD}\hfill \\ \mathrm{Co}\mathrm{exterior}\mathrm{derivative}\delta \hfill & & Q_D,Q_{AB}\hfill \\ \mathrm{Laplacian}\mathrm{\Delta }\hfill & & W=\{Q_B,Q_D\}=\{Q_{AB},Q_{AD}\}\hfill \\ \mathrm{Hodge}\mathrm{operation}\hfill & & \mathrm{symmetry}\mathrm{transformations}(2.7)and(3.7).\hfill \end{array}$$
$`(7.2)`$
(iii) The constraints, emerging from $`Q_B|phys>=0`$ and $`Q_D|phys>=0`$, are related to each-other due to the existence of the discrete duality transformations (3.7) for the bosonic part of the Lagrangian density (3.1). (iv) We see that the Lapalcian operator $`W`$ does not go to zero on-shell. This property was claimed to be one of the salient features of the topological field theory in Refs. \[18-20\] where the topological nature of the free 2D (non)Abelian 1-form gauge theory was established. Furthermore, we are unable to express the Lagrangian density (3.1) as the sum of BRST- and dual BRST invariant parts. This, in turn, implies that the energy-momentum tensor can also be not expressed as the sum of BRST- and dual BRST anticommutators. In addition, we are unable to obtain the topological invariants of the theory under consideration. Thus, 2-form free Abelian gauge theory in 4D does not mimic all the features of the free 2D 1-form gauge theory as a field theoretical model for the Hodge theory.
It will be interesting to explore the possibility of extending our investigations to the case of interacting (non)Abelian two-form gauge theory where matter fields are also present. The existence of new symmetries and their generalizations might turn out to be useful in the context of the proof for the renormalizability of such theories. It would be useful if we could discuss the $`BF`$ (non)Abelian gauge theory, in the framework of BRST cohomology and Hodge decomposition theorem where the 1-form gauge fields and 2-form gauge fields are coupled to each-other in a topologically invariant way. Its further extension to include matter fields is another workable problem. These are some of the issues which are under investigations and our results would be reported in our future publications.
Acknowledgements
One of us (EH) would like to thank the Director, SNBNCBS, Calcutta for the warm hospitality extended to him during his visit to the Centre where a part of this work was done. He also thanks U.G.C., Govt. of India, for the financial support through S.R.F. scheme. Fruitful conversations with A. Lahiri are gratefully acknowledged. EH and MS thank V. Srinivasan for encouragement.
|
warning/0004/gr-qc0004034.html
|
ar5iv
|
text
|
# Infinite hierarchies of exact solutions of the Einstein and Einstein–Maxwell equations for interacting waves and inhomogeneous cosmologies
## Introduction
A number of solution-generating techniques are known which provide tools for the construction of vacuum and electrovacuum solutions of Einstein’s equations for space-times with symmetries. These methods are based on the integrability of the symmetry reduced Einstein equations (viz. the Ernst equations). However, most of them were primarily designed to construct exact stationary axisymmetric solutions for which an additional regularity condition should be satisfied on the axis. This condition does not apply to interacting waves or cosmological models as considered here.
Apart from the completely linearizable subcase of Einstein–Rosen vacuum gravitational waves, the only techniques which provide nontrivial tools for the construction of solutions for the dynamical case are the vacuum Belinskii–Zakharov inverse-scattering method , the so called “monodromy transform” approach , and the group-theoretical approach recently developed by Hauser and Ernst . In particular, the methods of enable the construction of soliton perturbations of homogeneous cosmological models and some specific solutions for wave interaction regions. For example, the Khan–Penrose or Nutku–Halil solutions for the interaction region for colliding impulsive gravitational waves on a Minkowski background formally turn out to be two-soliton solutions on a symmetric Kasner background.
Here we consider the monodromy transform approach and the linear singular integral equations which arise in this context as an alternative form of the reduced Einstein equations. We present a new method for the solution of these equations which gives rise to infinite hierarchies of exact solutions. Among many other solutions, these include the particular cases mentioned above together with other soliton solutions on the symmetric Kasner background and their non-soliton extensions.
## Integral equation form of reduced Einstein equations
According to methods developed in , any solution of the Ernst equations can be constructed from the solution of the linear singular integral equation
$$\frac{1}{\pi i}_L\frac{[\lambda ]_\zeta }{\zeta \tau }(\tau ,\zeta )\text{}(\xi ,\eta ,\zeta )𝑑\zeta =\text{k}(\tau )$$
(1)
considered here for the hyperbolic case only. The parameters $`\xi `$, $`\eta `$ are two real null space-time coordinates, e.g. $`(\xi ,\eta )=(x+t,xt)`$. These coordinates span some local region in the neighbourhood of some initial regular space-time point $`P_0`$: $`\xi =\xi _0`$, $`\eta =\eta _0`$, in which local solutions of the reduced Einstein equations are considered.
The integration in (1) is performed along the path $`L`$ on the spectral plane $`w`$ which consists of two disconnected parts $`L_+`$ and $`L_{}`$. In the hyperbolic case, these are chosen as the segments of the real axis in the $`w`$-plane, which go from $`w=\xi _0`$ to $`w=\xi `$, and from $`w=\eta _0`$ to $`w=\eta `$ respectively.(We choose $`\xi _0\eta _0`$ and take $`\xi `$ and $`\eta `$ sufficiently close to $`\xi _0`$ and $`\eta _0`$ that the segments $`L_\pm `$ do not overlap.)
The integral in (1) splits into two, one of which possesses a singular kernel of Cauchy type and should be understood as a Cauchy principal value integral. The integration parameter $`\zeta `$ and a parameter $`\tau `$ span both of the contours $`L_+`$ and $`L_{}`$. Sometimes it will be convenient to introduce suffices: $`\zeta _+,\tau _+L_+`$ and $`\zeta _{},\tau _{}L_{}`$.
In the integrand in (1), $`[\lambda ]_\zeta =\frac{1}{2}(\lambda _{\mathrm{left}}\lambda _{\mathrm{right}})`$. This represents the jump on the contour, i.e. half of the difference between left and right limit values at the point $`\zeta L_+`$ or $`\zeta L_{}`$ of some “standard” function $`\lambda (\xi ,\eta ,w)`$. This function is a product of two functions $`\lambda (\xi ,\eta ,w)=\lambda _+(\xi ,w)\lambda _{}(\eta ,w)`$ given by
$$\lambda _+=\sqrt{\frac{w\xi }{w\xi _0}},\lambda _{}=\sqrt{\frac{w\eta }{w\eta _0}},$$
(2)
with the additional conditions $`\lambda _+|_{w=\mathrm{}}=\lambda _{}|_{w=\mathrm{}}=1`$. Each of these functions is an analytic function on the whole spectral plane $`w`$ apart from the cut $`L_+`$ or $`L_{}`$ respectively, whose endpoints are the branching points of the corresponding function.
In the equations (1), the three-dimensional complex vector function $`\text{}(\xi ,\eta ,\zeta )`$ is unknown, and the right hand side $`\text{k}(\tau )`$ is a three-dimensional complex vector function of the spectral parameter which may be taken to be
$$\text{k}(w)=\{1,𝐮(w),\text{v}(w)\},$$
(3)
where $`𝐮(w)`$ and $`\text{v}(w)`$ are arbitrary functions. The kernel of the integral in (1) is a scalar function $`(\tau ,\zeta )`$ given by
$`(\tau ,\zeta )=1+i(\zeta \beta _0)(𝐮(\tau )𝐮^{}(\zeta ))+\alpha _0^2𝐮(\tau )𝐮^{}(\zeta )`$
$`4(\zeta \xi _0)(\zeta \eta _0)\text{v}(\tau )\text{v}^{}(\zeta )`$ (4)
where the dagger denotes complex conjugation: e.g. $`𝐮^{}(w)\overline{𝐮(\overline{w})}`$. The additional constants in (4) are $`\alpha _0=(\xi _0\eta _0)/2`$ and $`\beta _0=(\xi _0+\eta _0)/2`$.
It is important to emphasize that the integral equations (1), and hence the functions $`𝐮(w)`$, $`\text{v}(w)`$ and $`\text{}(\xi ,\eta ,w)`$, only need to be evaluated on the two cuts $`L_+`$ and $`L_{}`$ in the spectral plane. Thus all the above vector and scalar functions of the spectral parameter are actually determined by pairs of functions which represent their values on these contours. For convenience we shall denote the values of these functions on $`L_\pm `$ by the corresponding suffices:
$$\{\text{u}(w),\text{v}(w)\}=\{\begin{array}{c}\{\text{u}_+(w),\text{v}_+(w)\},wL_+\\ \{\text{u}_{}(w),\text{v}_{}(w)\},wL_{}\end{array}$$
(5)
Thus, in (1) written in a more explicit form, we actually have two unknown vector functions $`\text{}_\pm `$. For any of these suffixed functions we can use also an alternative definition, for example,
$$\text{}(\xi ,\eta ,\tau _\pm )\text{}_\pm (\xi ,\eta ,\tau ).$$
Using this notation, it is convenient to split the integral in (1) into separate integrals over $`L_+`$ and $`L_{}`$ and to consider separately the cases $`\tau =\tau _+L_+`$ and $`\tau =\tau _{}L_{}`$. It is also convenient to denote the four scalar kernels which appear in the integrands of (1) in the form
$`(\tau _+,\zeta _+)_{++}(\tau ,\zeta ),`$ $`(\tau _+,\zeta _{})_+(\tau ,\zeta ),`$
$`(\tau _{},\zeta _+)_+(\tau ,\zeta ),`$ $`(\tau _{},\zeta _{})_{}(\tau ,\zeta )`$
where the functions $`_{++}(\tau ,\zeta )`$, $`_+(\tau ,\zeta )`$, $`_+(\tau ,\zeta )`$ and $`_{}(\tau ,\zeta )`$ can be determined explicitly in terms of the four functions $`𝐮_\pm (w)`$ and $`\text{v}_\pm (w)`$ using (4).
To conclude our description of the structure of the master integral equations, we recall that the four functions $`𝐮_\pm (w)`$ and $`\text{v}_\pm (w)`$ appearing in (3) and (5) play a significant role in the entire construction. They determine completely the coefficients of the integral equations in the electrovacuum case. In the vacuum case there are only two such functions $`𝐮_\pm (w)`$, as $`\text{v}_\pm (w)0`$. As shown in , they characterize unambiguously every individual solution of the Ernst equations. Moreover, the singular integral equations (1) possess a unique solution for any given choice of analytical functions $`𝐮_\pm (w)`$ and $`\text{v}_\pm (w)`$.
We recall now also, that the general local solution of the hyperbolic Ernst equations can be expressed by quadratures in terms of the solution of (1)
$`=1{\displaystyle \frac{2}{\pi }}{\displaystyle _L}[\lambda ]_\zeta \left[1i(\zeta \beta _0)𝐮^{}(\zeta )\right]\text{}^{[u]}(\xi ,\eta ,\zeta )𝑑\zeta `$
$`\mathrm{\Phi }={\displaystyle \frac{2}{\pi }}{\displaystyle _L}[\lambda ]_\zeta \left[1i(\zeta \beta _0)𝐮^{}(\zeta )\right]\text{}^{[v]}(\xi ,\eta ,\zeta )𝑑\zeta `$ (6)
where $`\text{}^{[u]}`$ and $`\text{}^{[v]}`$, in some association with the definition (3), denote respectively the second and third components of the vector solutions of the master integral equation (1), corresponding to a given choice of the monodromy data functions $`𝐮_\pm (w)`$ and $`\text{v}_\pm (w)`$. In a more explicit form, each of the the integrals in (6) should be split into two integrals evaluated over $`L_+`$ and $`L_{}`$.
## New hierarchies of solutions
Here we will construct a class of hyperbolic solutions that is determined by the rational monodromy data
$$\text{u}_\pm (w)=\frac{U_\pm (w)}{Q_\pm (w)},\text{v}_\pm (w)=\frac{V_\pm (w)}{Q_\pm (w)}$$
(7)
where $`U_+(w)`$, $`V_+(w)`$, $`Q_+(w)`$ and $`U_{}(w)`$, $`V_{}(w)`$, $`Q_{}(w)`$ are arbitrary complex polynomials, provided $`\text{u}_+(w)`$, $`\text{v}_+(w)`$ and $`\text{u}_{}(w)`$, $`\text{v}_{}(w)`$ do not have poles on $`L_+`$ and $`L_{}`$ respectively.
For what follows, it is convenient to calculate some auxiliary polynomials of two variables – we introduce the four polynomials $`P_{\pm \pm }(\tau ,\zeta )`$ defined by the relations
$$_{\pm \dot{\pm }}(\tau ,\zeta )=\frac{P_{\pm \dot{\pm }}(\tau ,\zeta )}{Q_\pm (\tau )Q_{\dot{\pm }}^{}(\zeta )},$$
(8)
and four polynomials $`R_{\pm \pm }(\tau ,\zeta )`$ defined from them by
$$R_{\pm \dot{\pm }}(\tau ,\zeta )=\frac{P_{\pm \dot{\pm }}(\tau ,\zeta )P_{\pm \dot{\pm }}(\zeta ,\zeta )}{\zeta \tau }.$$
(9)
In these definitions there are two sets of suffices, denoted as dotted and undotted, which should each be taken to be the same. Finally, it is convenient to introduce a redefinition of the unknown functions
$`\text{}_+(\zeta )={\displaystyle \frac{\lambda _{}^1(\zeta )Q_+^{}(\zeta )}{P_{++}(\zeta ,\zeta )}}\stackrel{~}{\text{}}_+(\zeta ),`$
$`\text{}_{}(\zeta )={\displaystyle \frac{\lambda _+^1(\zeta )Q_{}^{}(\zeta )}{P_{}(\zeta ,\zeta )}}\stackrel{~}{\text{}}_{}(\zeta ).`$ (10)
Hereafter we do not show explicitly the arguments $`\xi `$ and $`\eta `$ of $`\text{}_\pm `$ and $`\lambda `$ or the suffices $`\pm `$ at the points $`\zeta `$ and $`\tau `$, unless it is necessary.
A direct substitution of (7) into equations (1) with the use of (8)–(10) leads to the following convenient form of linear equations with polynomial right hand sides
$`{\displaystyle \frac{1}{\pi i}}{\displaystyle \underset{\xi _0}{\overset{\xi }{}}}{\displaystyle \frac{[\lambda _+]_\zeta }{\zeta \tau _+}}\stackrel{~}{\text{}}_+(\zeta )𝑑\zeta `$ $`={\displaystyle \frac{1}{\pi i}}{\displaystyle \underset{\xi _0}{\overset{\xi }{}}}[\lambda _+]_\zeta {\displaystyle \frac{R_{++}(\tau _+,\zeta )}{P_{++}(\zeta ,\zeta )}}\stackrel{~}{\text{}}_+(\zeta )𝑑\zeta `$
$`{\displaystyle \frac{1}{\pi i}}{\displaystyle \underset{\eta _0}{\overset{\eta }{}}}[\lambda _{}]_\zeta {\displaystyle \frac{R_+(\tau _+,\zeta )}{P_{}(\zeta ,\zeta )}}\stackrel{~}{\text{}}_{}(\zeta )𝑑\zeta +\left(\begin{array}{c}Q_+(\tau _+)\\ U_+(\tau _+)\\ V_+(\tau _+)\end{array}\right),`$
$`{\displaystyle \frac{1}{\pi i}}{\displaystyle \underset{\eta _0}{\overset{\eta }{}}}{\displaystyle \frac{[\lambda _{}]_\zeta }{\zeta \tau _{}}}\stackrel{~}{\text{}}_{}(\zeta )𝑑\zeta `$ $`={\displaystyle \frac{1}{\pi i}}{\displaystyle \underset{\eta _0}{\overset{\eta }{}}}[\lambda _{}]_\zeta {\displaystyle \frac{R_{}(\tau _{},\zeta )}{P_{}(\zeta ,\zeta )}}\stackrel{~}{\text{}}_{}(\zeta )𝑑\zeta `$
$`{\displaystyle \frac{1}{\pi i}}{\displaystyle \underset{\xi _0}{\overset{\xi }{}}}[\lambda _+]_\zeta {\displaystyle \frac{R_+(\tau _{},\zeta )}{P_{++}(\zeta ,\zeta )}}\stackrel{~}{\text{}}_+(\zeta )𝑑\zeta +\left(\begin{array}{c}Q_{}(\tau _{})\\ U_{}(\tau _{})\\ V_{}(\tau _{})\end{array}\right)`$
if we impose constraints on the coefficients of the rational functions (7) such that
$$P_+(\zeta ,\zeta )=P_+(\zeta ,\zeta )=0.$$
(12)
This leads to a large class of explicit solutions $`\stackrel{~}{\text{}}_\pm (\xi ,\eta ,\tau )`$ of (LABEL:lineqs) that are regular on the cuts $`L_\pm `$. However, the solution of the Ernst equations needs the solutions $`\text{}_\pm (\xi ,\eta ,\tau )`$ of (1) to be regular on the cuts $`L_\pm `$. Fortunately, all additional singularities (poles) of $`\text{}_+(\xi ,\eta ,\tau )`$ on $`L_+`$ and $`\text{}_{}(\xi ,\eta ,\tau )`$ on $`L_{}`$, which arise from the denominators in (10), can be avoided by the additional restrictions that $`\text{u}_+(\eta _0)=i/\alpha _0`$ and $`\text{u}_{}(\xi _0)=i/\alpha _0`$. We therefore specify
$`\text{u}_+(w)={\displaystyle \frac{i}{\alpha _0}}+(w\eta _0){\displaystyle \frac{C_+(w)}{Q_+(w)}}`$
$`\text{u}_{}(w)={\displaystyle \frac{i}{\alpha _0}}+(w\xi _0){\displaystyle \frac{C_{}(w)}{Q_{}(w)}}`$ (13)
where $`C_+(w)`$, $`C_{}(w)`$, $`Q_+(w)`$ and $`Q_{}(w)`$ are arbitrary polynomials. With these, the ansatz (12) leads to the constraint $`C_{}(w)=B(w)C_+^{}(w)4iA(w)V_+^{}(w)`$ and, for the polynomials in (7), the general solution of (12) reads
$`U_+(w)={\displaystyle \frac{i}{\alpha _0}}Q_+(w)+(w\eta _0)C_+(w)`$
$`U_{}(w)=B(w)\left({\displaystyle \frac{i}{\alpha _0}}Q_+^{}(w)+(w\beta _0)C_+^{}(w)\right)4i(w\xi _0)A(w)V_+^{}(w)`$
$`V_{}(w)=A(w)\left(Q_+^{}(w)i\alpha _0^2C_+^{}(w)\right)`$
$`Q_{}(w)=B(w)\left(Q_+^{}(w)i\alpha _0^2C_+^{}(w)\right)`$ (14)
where the polynomials $`A(w)`$, $`B(w)`$, $`C_+(w)`$, $`V_+(w)`$ and $`Q_+(w)`$ can be chosen arbitrarily, provided the corresponding functions $`\text{u}_\pm (w)`$, $`\text{v}_\pm (w)`$ have no poles on the cuts $`L_+`$ and $`L_{}`$ respectively. The vacuum case, which occurs when $`A(w)=V_+(w)=0`$ and $`B(w)=1`$, yields simpler expressions which involve just two arbitrary polynomials $`C_+(w)`$ and $`Q_+(w)`$.
Returning to (New hierarchies of solutions), we note that the integral operators in the left hand sides can be inverted using the Poincaré–Bertrand formula for singular integrals
$$\frac{1}{\pi i}_L\frac{[\lambda ]_\zeta }{\zeta \tau }\text{}(\zeta )𝑑\zeta =f(\tau )\text{}(\tau )=\frac{1}{\pi i}_L\frac{[\lambda ^1]_\zeta }{\zeta \tau }f(\zeta )𝑑\zeta .$$
(15)
This can be applied to the integrals over $`L_+`$ (using $`\lambda _+`$), or over $`L_{}`$ (using $`\lambda _{}`$).
Since the right hand sides of (New hierarchies of solutions) are polynomials in $`\tau `$, the inversion (15) leads to the solution in the form
$$\stackrel{~}{\text{}}_\pm (\tau )=\underset{k=0}{\overset{N_\pm }{}}\left(\begin{array}{c}\stackrel{~}{q}_{k\pm }\\ \stackrel{~}{u}_{k\pm }\\ \stackrel{~}{v}_{k\pm }\end{array}\right)Z_{k\pm }(\tau )$$
(16)
where $`N_+`$ and $`N_{}`$ are the maxima of the degrees of the polynomials $`U_+`$, $`V_+`$, $`Q_+`$ and $`U_{}`$, $`V_{}`$, $`Q_{}`$ respectively, $`\stackrel{~}{u}_{k\pm }`$, $`\stackrel{~}{v}_{k\pm }`$, $`\stackrel{~}{q}_{k\pm }`$ are unknown $`\tau `$-independent functions of $`\xi `$ and $`\eta `$, and $`Z_{k\pm }(\tau )`$ are “standard” functions given by
$$Z_{k\pm }(\tau )=\frac{1}{\pi i}_{L_\pm }\frac{[\lambda _\pm ^1]_\zeta }{\zeta \tau }\zeta ^k𝑑\zeta .$$
(17)
All these functions (integrals) can be evaluated as the residues of their integrands at $`\zeta =\mathrm{}`$ are polynomials in $`\tau `$ of degree $`k`$.
We note now, that the vector integral equations (1) decouple into three pairs of equations – one pair for each of the three components of $`\stackrel{~}{\text{}}_+`$ and the corresponding component of $`\stackrel{~}{\text{}}_{}`$. All these pairs of equations possess the same kernels but different right hand sides. Therefore, substituting the expressions (16) into (New hierarchies of solutions) and using (9) with (12), we get three decoupled algebraic systems, each of order $`(N+2)\times (N+2)`$ where $`N=N_++N_{}`$ and for the sets of unknowns $`\stackrel{~}{q}_{k\pm }`$, $`\stackrel{~}{u}_{k\pm }`$, $`\stackrel{~}{v}_{k\pm }`$ respectively. However, in view of (6), we need the solutions of two of these systems only:
$$\begin{array}{c}\underset{B=0}{\overset{N+1}{}}𝒟_{AB}\stackrel{~}{u}_B=u_A,\\ \underset{B=0}{\overset{N+1}{}}𝒟_{AB}\stackrel{~}{v}_B=v_A\end{array}𝒟=\left(\begin{array}{cc}D_{++}& D_+\\ D_+& D_{}\end{array}\right)$$
(18)
where the indices $`A,B=0,1,\mathrm{}N+1`$. The column vectors $`u_A`$, $`v_A`$ (shown below as rows) are composed of the coefficients of the polynomials $`U_\pm (\zeta )`$ and $`V_\pm (\zeta )`$:
$`u_A=\{u_{0+},u_{1+},\mathrm{},u_{N_+},u_0,u_1,\mathrm{},u_N_{}\}`$
$`v_A=\{v_{0+},v_{1+},\mathrm{},v_{N_+},v_0,v_1,\mathrm{},v_N_{}\}.`$ (19)
Similarly, we combine the coefficients $`\stackrel{~}{u}_{k\pm }`$, $`\stackrel{~}{v}_{k\pm }`$ in (16) to form the column vectors (rows)
$`\stackrel{~}{u}_A(\xi ,\eta )=\{\stackrel{~}{u}_{0+},\stackrel{~}{u}_{1+},\mathrm{},\stackrel{~}{u}_{N_+},\stackrel{~}{u}_0,\stackrel{~}{u}_1,\mathrm{},\stackrel{~}{u}_N_{}\}`$
$`\stackrel{~}{v}_A(\xi ,\eta )=\{\stackrel{~}{v}_{0+},\stackrel{~}{v}_{1+},\mathrm{},\stackrel{~}{v}_{N_+},\stackrel{~}{v}_0,\stackrel{~}{v}_1,\mathrm{},\stackrel{~}{v}_N_{}\}`$ (20)
The matrix $`𝒟`$ consists of the blocks $`D_{++}`$, $`D_+`$, $`D_+`$, $`D_{}`$ of orders $`(N_++1)\times (N_++1)`$, $`(N_++1)\times (N_{}+1)`$, $`(N_{}+1)\times (N_++1)`$ and $`(N_{}+1)\times (N_{}+1)`$ respectively. Their components are determined by the integrals:
$`(D_{++})_{kl}(\xi )=\delta _{kl}+{\displaystyle \frac{1}{\pi i}}{\displaystyle \underset{\xi _0}{\overset{\xi }{}}}[\lambda _+]_\zeta {\displaystyle \frac{(R_{++})_k(\zeta )}{P_{++}(\zeta ,\zeta )}}Z_{l+}(\zeta )𝑑\zeta `$
$`(D_+)_{kl}(\eta )={\displaystyle \frac{1}{\pi i}}{\displaystyle \underset{\eta _0}{\overset{\eta }{}}}[\lambda _{}]_\zeta {\displaystyle \frac{(R_+)_k(\zeta )}{P_{}(\zeta ,\zeta )}}Z_l(\zeta )𝑑\zeta `$
$`(D_+)_{kl}(\xi )={\displaystyle \frac{1}{\pi i}}{\displaystyle \underset{\xi _0}{\overset{\xi }{}}}[\lambda _+]_\zeta {\displaystyle \frac{(R_+)_k(\zeta )}{P_{++}(\zeta ,\zeta )}}Z_{l+}(\zeta )𝑑\zeta `$ (21)
$`(D_{})_{kl}(\eta )=\delta _{kl}+{\displaystyle \frac{1}{\pi i}}{\displaystyle \underset{\eta _0}{\overset{\eta }{}}}[\lambda _{}]_\zeta {\displaystyle \frac{(R_{})_k(\zeta )}{P_{}(\zeta ,\zeta )}}Z_l(\zeta )𝑑\zeta `$
where $`(R_{\pm \pm })_k`$ are the coefficients in the expansions $`R_{+\pm }(\tau ,\zeta )=_{k=0}^{N_+}(R_{+\pm })_k(\zeta )\tau ^k`$ and $`R_\pm (\tau ,\zeta )=_{k=0}^N_{}(R_\pm )_k(\zeta )\tau ^k`$.
To calculate the final expressions for the Ernst potentials, we need to evaluate the additional sets of integrals
$`J_{k+}(\xi )={\displaystyle \frac{1}{\pi i}}{\displaystyle \underset{\xi _0}{\overset{\xi }{}}}[\lambda _+]_\zeta {\displaystyle \frac{Q_+^{}(\zeta )i(\zeta \beta _0)U_+^{}(\zeta )}{P_{++}(\zeta ,\zeta )}}Z_{k+}(\zeta )𝑑\zeta `$
$`J_k(\eta )={\displaystyle \frac{1}{\pi i}}{\displaystyle \underset{\eta _0}{\overset{\eta }{}}}[\lambda _{}]_\zeta {\displaystyle \frac{Q_{}^{}(\zeta )i(\zeta \beta _0)U_{}^{}(\zeta )}{P_{}(\zeta ,\zeta )}}Z_k(\zeta )𝑑\zeta ,`$
and to combine them into one row vector
$$J_A=\{J_{0+},J_{1+},\mathrm{},J_{N_+},J_0,J_1,\mathrm{},J_N_{}\}.$$
(22)
Let us also define two additional $`(N+2)\times (N+2)`$ matrices
$$𝒢_{AB}=𝒟_{AB}2iu_AJ_B,_{AB}=𝒟_{AB}2iv_AJ_B.$$
(23)
All integrals determining the components of the matrices $`𝒢_{AB}`$, $`_{AB}`$ and $`𝒟_{AB}`$ can be evaluated in terms of the residues of their integrands at the zeros of $`P_{++}(w,w)`$ and $`P_{}(w,w)`$ and at $`w=\mathrm{}`$. We then have
$$=\frac{det𝒢_{AB}}{det𝒟_{AB}},\mathrm{\Phi }=\frac{det_{AB}}{det𝒟_{AB}},$$
(24)
which are the final expressions for the Ernst potentials. These solutions generally possess essentially nonlinear properties. They are not trivial time-dependent analogues of any stationary axisymmetric solutions with regular axis of symmetry which have different structures of monodromy data. The expressions (24) generally are not rational functions of $`\xi `$, $`\eta `$.
When evaluating explicit examples, it may be noted that solutions with a diagonal metric occur when $`𝐮_\pm ^{}=𝐮_\pm `$. The plane symmetric (type D) Kasner metric with $`=\alpha /\alpha _0`$ is obtained using the constants $`𝐮_+=i/\alpha _0`$, $`𝐮_{}=i/\alpha _0`$ and $`𝐯_\pm =0`$. The Khan–Penrose solution for colliding plane impulsive gravitational waves is obtained with $`\text{v}_+(w)=\text{v}_{}(w)=0`$ and
$$𝐮_+(w)=ik_+\frac{wa_+}{wb_+},𝐮_{}(w)=ik_{}\frac{wa_{}}{wb_{}}$$
(25)
when the constants $`a_\pm `$, $`b_\pm `$ and $`k_\pm `$ are real. The nondiagonal Nutku–Halil solution for non-colinear impulsive waves is obtained from the same expression using complex constants. This explicitly demonstrates that the above method is applicable to both the linear and nonlinear cases.
This work was partly supported by the EPSRC and by the grants 99-01-01150 and 99-02-18415 from the RFBR.
|
warning/0004/astro-ph0004094.html
|
ar5iv
|
text
|
# Magnetohydrodynamic Instabilities in Shearing, Rotating, Stratified Winds and Disks
## 1 Introduction
The ubiquity of energetic molecular outflows and atomic jets from young stellar objects (YSOs) ranging from deeply embedded infrared sources to classical T Tauri stars suggests that they are an inescapable by-product of star formation (e.g., Richer et al., 2000, and references therein). Winds from YSO disks play an important role in shedding angular momentum carried by inflowing material, thereby permitting further accretion in order for the central objects to attain stellar dimensions (Hartmann & MacGregor, 1982; Pudritz & Norman, 1986; Shu et al., 1988). These winds may also have strong effects on the dynamical evolution of the parent cloud by providing a source of turbulent energy (Norman & Silk, 1980), and may help to determine the final masses of stars by reversing the infall of surrounding gas (Shu, Adams, & Lizano, 1987). Therefore, understanding the physics of protostellar winds is of essential importance to the theory of star formation.
Among the various scenarios regarding the origin and nature of the protostellar winds, the most promising is magnetohydrodynamic (MHD) models in which winds are driven by the interaction of the centrifugal force with open magnetic fields threading rapidly rotating disks. These magnetocentrifugally driven wind models can account for observed high mass and momentum losses (Lada, 1985; Bachiller, 1996) and the kinematic and structural characteristics of bipolar molecular outflows in general (Li & Shu, 1996), as well as in specific cases (e.g., HH111, cf. Nagar et al., 1997). It is, however, still controversial whether the wind originates only from a small magnetosphere-disk interaction region near the central star (Shu et al., 1994, 2000), or whether it emanates from an extended region of the disk (following the seminal model of Blandford & Payne 1982; see, e.g., K$`\ddot{\mathrm{o}}`$nigl & Pudritz 2000).
Although the role of magnetic fields in driving protostellar winds is by now well established (at least theoretically), their complementary role in governing the stability properties of winds is less well explored. In addition, azimuthal and vertical shear within winds may also affect their stability properties. Questions of wind stability are potentially important for both large scale and small-scale phenomena. These include understanding the role of magnetic fields and velocity shear in (a) helping winds to propagate over enormous distances (up to a factor $`10^6`$ in dynamic range) through the ISM in parsec-scale giant HH flows (Reipurth, Bally, & Devine, 1997); (b) creating bright HH “knots” spaced throughout optical jets (Hartigan et al., 2000); (c) governing the angular extent of the emergent wind and establishing the momentum distribution for driving molecular outflows; and (d) converting large-scale ordered flow energy to jet heating through small-scale instabilities. In addition, it is important to study the dynamical properties of large classes of theoretical wind solutions to test whether they are stable equilibria which can represent real astronomical systems, or whether they are unstable equilibria which should rarely be observed in nature because they evolve rapidly into other configurations.
Previous studies of time-dependent behavior in steadily-input MHD winds have focussed primarily on the Kelvin-Helmholtz instabilities driven by the interaction of the wind surface layer with the ambient medium (see, e.g., the studies of Appl & Camenzind, 1992; Hardee et al., 1992; Rosen et al., 1999, and references therein). Generally, heavy jets containing strong toroidal fields are relatively resistant to these instabilities. In addition to “driven” instabilities resulting from boundary conditions, winds may also be subject to “free” instabilities in their interiors. Whether a given wind solution is internally unstable must depend on the velocity shear, magnetic geometry, and internal stratification. One route to studying internal wind instabilities is via time-dependent numerical simulations of winds. Although such studies (e.g., Ouyed & Pudritz, 1997, and references therein) have yielded intriguing results on the development of episodic knots in MHD winds, the computational demands in carrying out simulations precludes extensive exploration of parameter space, large spatial dynamic range, or very long-term integration. In addition, some of the time-dependent internal features found in simulations may be introduced by particular choices of inflow boundary conditions that are inconsistent with a steady-state flow, rather than occurring as a result of intrinsic instability of the wind.
Due to the importance of gaseous accretion disks in a wide variety of astrophysical systems, major attention has focussed on disk dynamics, and, in particular, the role of instability-driven turbulence in angular momentum transport (see, e.g., Balbus & Hawley, 1998; Stone et al., 2000, for reviews). Saturated magnetorotational instabilities (hereafter MRIs; Balbus & Hawley, 1991, 1992; Hawley & Balbus, 1991; Hawley, Gammie, & Balbus, 1995) represent perhaps the most important local dynamical process affecting disk evolution. Magnetized disk winds share many generic properties with disks, so it is interesting to investigate the potential importance of MRIs in winds.
In this work, we investigate the internal stability of rotating, magnetized protostellar winds to (primarily) local shear and buoyancy modes. We also extend previous studies of local MRIs in accretion disks. The fundamental difference between “wind” and “disk” systems in our idealized models is in the absence or presence of gravity as a confining force. These systems may also be distinguished by the geometry of the magnetic field, with toroidally-dominant fields expected in the wind case, but either poloidal or toroidal fields possible for the disk case. Our most general analysis and results apply to cold flows, but we also perform separate calculations (see §7) including thermal effects which specialize to local analysis of MRIs.
Whether a wind originates from a narrow boundary layer or an extended radial region, the radial expansion of the flow will lead to shear in the azimuthal velocity field of the asymptotic state. The total specific angular momentum of the flow, $`J=R(v_\varphi B_{\mathrm{pol}}B_\varphi /(4\pi \rho v_{\mathrm{pol}}))`$, is conserved along streamlines (where $`B_{\mathrm{pol}}`$, $`B_\varphi `$ and $`v_{\mathrm{pol}}`$, $`v_\varphi `$ are the poloidal, toroidal components of the magnetic field and the flow velocity). If the Alfv$`\stackrel{´}{\mathrm{e}}`$n mach number $`M_\mathrm{A}v_{\mathrm{pol}}/v_{\mathrm{A},\mathrm{pol}}1`$, the kinetic part dominates the specific angular momentum and $`v_\varphi J/R`$. Thus, the angular velocity $`\mathrm{\Omega }=v_\varphi /RJ/R^2`$ in the asymptotic wind will have a gradient $`d\mathrm{ln}\mathrm{\Omega }/d\mathrm{ln}R=d\mathrm{ln}J/d\mathrm{ln}R2`$. If the wind comes from a boundary layer, then $`d\mathrm{ln}J/d\mathrm{ln}R`$ may be small; if the wind originates from a large region with self-similar scalings, then $`d\mathrm{ln}J/d\mathrm{ln}R=1/2`$. In either case, $`d\mathrm{ln}\mathrm{\Omega }/d\mathrm{ln}R`$ is expected to be a negative, order-unity quantity for wind systems. For thin disk systems (i.e., negligible pressure support), a Keplerian radial profile $`d\mathrm{ln}\mathrm{\Omega }/d\mathrm{ln}R=3/2`$ is expected if the central mass is dominant. For the analysis of this paper (except where noted otherwise), we adopt $`d\mathrm{ln}\mathrm{\Omega }/d\mathrm{ln}R=3/2`$ for both “wind” and “disk” systems, but our qualitative results are insensitive to this assumption. As discussed below, order-unity radial logarithmic gradients may also be expected in the magnetic field strengths; we allow for a range of magnetic gradients.
Significant shear may also exist in the poloidal velocities of jets if they originate from an extended radial region. The asymptotic outflow speed $`v_{0\mathrm{z}}`$ generally scales linearly with the rotational speed at the footpoint $`R_{\mathrm{foot}}`$ of the streamline, so that
$$\frac{\mathrm{ln}v_{0\mathrm{z}}}{\mathrm{ln}R}=\frac{\mathrm{ln}v_{0\mathrm{z}}}{\mathrm{ln}v_{\mathrm{foot}}}\frac{\mathrm{ln}v_{\mathrm{foot}}}{\mathrm{ln}R_{\mathrm{foot}}}\frac{\mathrm{ln}R_{\mathrm{foot}}}{\mathrm{ln}R}\frac{1}{2}\frac{\mathrm{ln}R_{\mathrm{foot}}}{\mathrm{ln}R}.$$
If the range of footpoint radii is small compared to the range of asymptotic radii (as for a wind from a boundary layer), then $`|\mathrm{ln}v_{0\mathrm{z}}/\mathrm{ln}R|1`$; if the radial ranges are comparable, then $`\mathrm{ln}v_{0\mathrm{z}}/\mathrm{ln}R`$ is negative and order unity. For disks, the vertical velocity shear is negligible. We allow for a range of vertical shear rates in the present analysis.
Our analysis consists of developing and solving sets of linearized MHD equations for several general classes of background flows. Both axisymmetric and non-axisymmetric disturbances are explored. Even in linearized form, MHD problems present considerable technical challenges. Thus, instead of attacking sophisticated sets of steady-state wind solutions, for many specific examples we will take as an unperturbed configuration one of the power-law cylindrical equilibrium solutions recently identified by Ostriker (1997) as asymptotic states of self-similar disk winds. These have density $`\rho R^q`$, $`𝔹R^{(1+q)/2}`$, $`𝕧R^{1/2}`$, and sound speed $`c_\mathrm{s}=0`$. Although these adopted initial configurations are relatively simple, they retain general asymptotic characteristics of MHD disk winds in the sense that they have both azimuthal and vertical magnetic field and velocity components with arbitrary ratios, and all the physical variables have radial gradients. These gradients may also be thought of as representing local scalings within a more complex overall stratification. We also include models without vertical motion but with significant equilibrium gravity to study stability in magnetized astrophysical disks. In our analysis of MRIs, we use equilibria with uniform $`𝔹,\rho `$, and $`c_\mathrm{s}0`$, and take $`\mathrm{\Omega }R^a`$ with arbitrary $`a`$.
Because the systems we are studying contain significant azimuthal shear, an arbitrary initial spatial planform is not maintained indefinitely. When $`\mathrm{\Omega }^{}0`$ and the azimuthal wavenumber $`m0`$ (and/or when $`v_{0\mathrm{z}}^{}0`$ and the vertical wavenumber $`k_\mathrm{z}0`$; the prime represents a differentiation with respect to $`R`$), spatial wavefunctions may become increasingly radially corrugated in time due to the kinematic shearing of the wavefronts imposed in the initial conditions. If we describe the radial spatial wavefunction in terms of the amplitudes of Fourier coefficients with radial wavenumbers $`k_\mathrm{R}`$, this corresponds at late times to a secular increase in the amplitudes of large-$`k_\mathrm{R}`$ terms and decrease in the amplitudes of small-$`k_\mathrm{R}`$ terms. As we shall show, all the disturbances we identify are stabilized at sufficiently large $`k_\mathrm{R}`$. Thus, if $`m\mathrm{\Omega }^{}`$ and/or $`k_\mathrm{z}v_{0\mathrm{z}}^{}0`$, the net amplification factor for any arbitrary initial disturbance is limited by the rate of kinematic growth of radial corrugation compared to the growth rate of any dynamically-driven instabilities.
In previous work, two complementary analytical methods have been used to study small-amplitude disturbances in shearing astrophysical systems. One approach adopts the “shearing-sheet” formalism, and integrates the local, time-dependent, linearized wave equations directly to obtain the evolutionary behavior of shearing wavelets treated as an initial-value problem (Goldreich & Lynden-Bell, 1965; Julian & Toomre, 1966; Balbus & Hawley, 1992). An alternative analytical approach uses WKB techniques to derive dispersion relations for spatial Fourier modes, superpositions of which represent local wavefunctions (e.g., Shu, 1974, 1992; Ryu & Goodman, 1992; Terquem & Papaloizou, 1996). This paper includes analyses using both approaches, and also introduces a hybrid technique which we term a “coherent wavelet” analysis. We adopt the “modal” strategy in order to identify characteristic instantaneous growth rates and physical mechanisms for a wide variety of spatial disturbances. By considering the time over which a spatial pattern is altered by shear, we can estimate the net amplification factor of a given initial modal disturbance. We use the shearing-sheet formalism for studying magnetorotational instabilities, which are cut off at relatively small values of $`Rk_\mathrm{R}/m`$ (whereas the modal analysis applies to large $`Rk_\mathrm{R}/m`$), and also for studying buoyancy instabilities in the high-$`m`$ regime where modes are short-lived. We show that the results obtained from the shearing-sheet integrations in both cases are in excellent agreement with the predictions of a coherent wavelet analysis, in which time-dependent growth rates $`\gamma (t)`$ are computed by time-localizing the shearing-sheet equations and solving an analytic dispersion relation.
The organization of this paper is as follows: We begin by studying instabilities in cold, magnetized winds. In §2, the basic MHD equations and the specific adopted wind equilibria are described. In §3, we analyze the stability of winds to the simplest perturbation with $`k_\mathrm{z}=m=0`$, where $`k_\mathrm{z}`$ and $`m`$ are respectively the vertical and azimuthal wavenumbers of the perturbation. We term these the “fundamental modes”; we present solutions for stable and unstable global modes under the assumption of free Lagrangian boundary conditions. The modal analysis and general local dispersion relation for cold flows with arbitrary $`k_\mathrm{z}`$ and $`m`$ are presented in §4. We present numerical solutions of the dispersion relation for both axisymmetric and non-axisymmetric perturbations in §5. In §6, we classify the unstable or overstable modes and provide the physical interpretation for each mode. Next, we include (variable) thermal pressure terms to compare the susceptibility of cold winds vs. warm disks to shear instabilities. In §7 we analyze the axisymmetric Balbus-Hawley instability of poloidal fields and the non-axisymmetric MRI of toroidal fields, discuss the respective instability mechanisms, and provide the corresponding instability criteria. Here, we use the coherent wavelet technique to compute growth rates, and compare with direct shearing-sheet integrations. The generalized instability criteria and net amplification factors for the magnetorotational disturbances with both toroidal and poloidal background fields are also derived. Finally in §8, we summarize and discuss conclusions of the present work.
## 2 Basic Equations and Cylindrical Equilibrium for Cold Wind
We begin with the ideal MHD equations
$$\frac{\rho }{t}+(\rho 𝕧)=0,$$
(1)
$$\frac{𝕧}{t}+𝕧𝕧=\frac{1}{4\pi \rho }(\times 𝔹)\times 𝔹\frac{P}{\rho }\mathrm{\Phi }_\mathrm{G},$$
(2)
$$\frac{𝔹}{t}=\times (𝕧\times 𝔹),$$
(3)
and
$$𝔹=0,$$
(4)
where $`\rho `$ is the density, $`𝕧`$ is the fluid velocity, $`𝔹`$ is the magnetic field, $`P`$ is the thermal pressure, and $`\mathrm{\Phi }_\mathrm{G}𝕘`$ is the gravitational force due to a central object. We ignore self-gravity in the flow.
We now consider cold, magnetized cylindrical flows. Since the flow velocity in disk winds is always supersonic except in the vicinity of the disk where the material is lifted by the thermal pressure (Blandford & Payne, 1982), the thermal pressure term in eq. (2) can generally be neglected compared to magnetic stress. Except for investigations of generalized MRIs (§7), we shall drop the thermal pressure term. We adopt standard cylindrical coordinates $`(R,\varphi ,z)`$.
By assuming that $`v_\mathrm{R}=B_\mathrm{R}=0`$ and all variables are independent of $`z`$, we have a general equilibrium condition from eq. (2)
$$\mathrm{\Omega }^2R\frac{v_\varphi ^2}{R}=\frac{1}{4\pi \rho }\left(\frac{B_\varphi ^2}{R}+𝔹𝔹^{}\right)+g_\mathrm{R},$$
(5)
where a prime denotes differentiation with respect to $`R`$. At a large distance from the origin, the gravitational force due to the central source can also be ignored on the grounds that magnetic and centrifugal forces far exceed it. In this case, the magnetic hoop stress acting inward is the only force that balances the outward centrifugal force and outward magnetic pressure gradient force (under the assumption that the magnetic field strength decreases outward).
As an initial equilibrium configuration of the wind, for specific cases we will adopt the asymptotic solutions for cylindrically symmetric axial flows presented by Ostriker (1997). All variables have power-law dependences on $`R`$: $`\rho R^q,B_\varphi B_\mathrm{z}R^{(1+q)/2}`$, and $`v_\varphi v_\mathrm{z}R^{1/2}`$. We define the local pitch angle $`i`$ of the magnetic fields as $`i\mathrm{tan}^1(B_\mathrm{z}/B_\varphi )`$. Neglecting $`g_\mathrm{R}`$ in eq. (5), radial momentum balance requires
$$v_\varphi ^2=\frac{v_\mathrm{A}^2}{2}\left(\mathrm{cos}2iq\right),$$
(6)
where $`v_\mathrm{A}`$ is the local Alfv$`\stackrel{´}{\mathrm{e}}`$n speed defined by
$$v_\mathrm{A}^2v_{\mathrm{A}\varphi }^2+v_{\mathrm{Az}}^2\mathrm{with}v_{\mathrm{A}\varphi }\frac{B_\varphi }{\sqrt{4\pi \rho }}\mathrm{and}v_{\mathrm{Az}}\frac{B_\mathrm{z}}{\sqrt{4\pi \rho }}.$$
It is obvious from eq. (6) that there would be no such power-law solutions if $`q>1`$. This is because when $`q>1`$, the magnetic field has so steep a gradient that the corresponding pressure force always exceeds the tension. Therefore, to ensure force balance and cylindrical collimation in winds with power-law profiles, the magnetic field strength must decline with $`R`$ more slowly than $`R^1`$.
We can define an angular velocity $`\mathrm{\Omega }_\mathrm{f}\mathrm{\Omega }v_{\mathrm{pol}}B_\varphi /(RB_{\mathrm{pol}})`$ as that of a rotating frame in which the flow of winds is parallel to the local field line. $`\mathrm{\Omega }_\mathrm{f}`$ is the rotation rate of the magnetic field lines thought of as rigid wires. In such a frame, the family of solutions can be completely described in terms of scaled values of the specific angular momentum $`j`$, the Bernoulli constant $`e`$, and $`q`$, where
$$j\frac{\mathrm{\Omega }}{\mathrm{\Omega }_\mathrm{f}}\left(1\frac{(v_{\mathrm{A}\varphi }/R\mathrm{\Omega })^2}{1\mathrm{\Omega }_\mathrm{f}/\mathrm{\Omega }}\right)\mathrm{and}e\frac{1}{(R\mathrm{\Omega }_\mathrm{f})^2}\left(\frac{1}{2}𝕧^2+\mathrm{\Phi }_\mathrm{G}R^2\mathrm{\Omega }_\mathrm{f}\mathrm{\Omega }\right),$$
(Ostriker, 1997). The condition for a super-Alfv$`\stackrel{´}{\mathrm{e}}`$nic outflow velocity requires $`0<j1`$. Generally speaking, the pitch angle $`i`$ does not depend on $`q`$, although one can parameterize $`i`$ in terms of $`q`$, $`e`$, and $`j`$. However, for flows originating from a Kepler-rotating disk, angular momentum and energy conservation requirements limit the range of $`i`$ available to an equilibrium (asymptotic) magnetic field configuration. Utilizing eqs. (15) to (23) of Ostriker (1997) one can show that the maximum, over all permitted values of $`e`$ and $`j`$, pitch angle $`i_{\mathrm{max}}`$ is given by
$$\mathrm{tan}^2i_{\mathrm{max}}=\left(\frac{4}{3+q}\right)^21,$$
(7)
which is attained when $`e=0`$ and $`j=1`$. If $`j>1`$, the streamline never reaches the Alfv$`\stackrel{´}{\mathrm{e}}`$n radius.
## 3 Fundamental Mode
### 3.1 Dynamical Equations
We first consider the response of the equilibrium state when small, axisymmetric perturbations with an infinite wavelength along a vertical direction are imposed. We term the waves with $`k_\mathrm{z}=0`$ and $`m=0`$ the “fundamental modes” analogous to eigenfunctions of oscillations without any node. Let the subscripts 0 and 1 denote the equilibrium and perturbed states, respectively. Linearizing the set of the dynamical equations (1) to (4), we may write
$$\frac{\rho _1}{t}=\frac{1}{R}\frac{}{R}\left(R\rho _0v_{1\mathrm{R}}\right),$$
(8)
$$\frac{v_{1\mathrm{R}}}{t}=2\mathrm{\Omega }v_{1\varphi }\frac{1}{4\pi \rho _0}\left\{\frac{2B_{0\varphi }B_{1\varphi }}{R}+\frac{}{R}\left(𝔹_0𝔹_1\right)\frac{\rho _1}{\rho _0}\left(\frac{B_{0\varphi }^2}{R}+𝔹_0𝔹_0^{}\right)\right\},$$
(9)
$$\frac{v_{1\varphi }}{t}=\frac{\kappa ^2}{2\mathrm{\Omega }}v_{1\mathrm{R}},$$
(10)
$$\frac{v_{1\mathrm{z}}}{t}=v_{0\mathrm{z}}^{}v_{1\mathrm{R}},$$
(11)
$$\frac{B_{1\varphi }}{t}=\frac{}{R}\left(B_{0\varphi }v_{1\mathrm{R}}\right),$$
(12)
$$\frac{B_{1\mathrm{z}}}{t}=\frac{1}{R}\frac{}{R}\left(RB_{0\mathrm{z}}v_{1\mathrm{R}}\right),$$
(13)
and $`B_{1\mathrm{R}}=0`$. In eq. (10), $`\kappa `$ stands for the epicycle frequency
$$\kappa ^2\frac{1}{R^3}\frac{d}{dR}(R^4\mathrm{\Omega }^2).$$
Combining the perturbed equations (8)$``$(13) and eliminating all other variables in favor of the perturbed radial velocity $`v_{1\mathrm{R}}`$, one obtains the wave equation
$`{\displaystyle \frac{1}{v_\mathrm{A}^2}}{\displaystyle \frac{^2v_{1\mathrm{R}}}{t^2}}`$ $`=`$ $`{\displaystyle \frac{^2v_{1\mathrm{R}}}{R^2}}+{\displaystyle \frac{d\mathrm{ln}(RB_0^2)}{dR}}{\displaystyle \frac{v_{1\mathrm{R}}}{R}}`$ (14a)
$``$ $`\left[{\displaystyle \frac{\kappa ^2}{v_\mathrm{A}^2}}+{\displaystyle \frac{1}{R^2}}\left(1+\mathrm{cos}^2i{\displaystyle \frac{d\mathrm{ln}\rho _0}{d\mathrm{ln}R}}\right){\displaystyle \frac{\rho _0}{RB_0^2}}{\displaystyle \frac{d}{dR}}\left({\displaystyle \frac{R}{\rho _0}}𝔹_0𝔹_0^{}\right)\right]v_{1\mathrm{R}}.`$
For power-law profiles, this becomes
$$\frac{1}{v_\mathrm{A}^2}\frac{^2v_{1\mathrm{R}}}{t^2}=\frac{^2v_{1\mathrm{R}}}{R^2}\frac{q}{R}\frac{v_{1\mathrm{R}}}{R}\left[\frac{\kappa ^2}{v_\mathrm{A}^2}+\frac{1}{R^2}\left(\frac{1q}{2}q\mathrm{cos}^2i\right)\right]v_{1\mathrm{R}}.$$
()
To transform eqs. (14) to the $`\mathrm{Schr}\ddot{\mathrm{o}}\mathrm{dinger}`$ form, we define a new independent variable $`\mathrm{\Psi }`$ through
$$v_{1\mathrm{R}}=\frac{\mathrm{\Psi }(R)}{\sqrt{RB_0^2}}e^{i\omega t};\mathrm{or}v_{1\mathrm{R}}=R^{q/2}\mathrm{\Psi }(R)e^{i\omega t}$$
(15)
for the power-law case. Then, we have
$$\frac{d^2\mathrm{\Psi }}{dR^2}+K^2(R)\mathrm{\Psi }=0,$$
(16)
with $`K(R)`$ defined by
$$K^2(R)\frac{\omega ^2\kappa ^2}{v_\mathrm{A}^2}\frac{3}{4R^2}\frac{d\mathrm{ln}\rho _0}{dR}\left(\frac{\mathrm{cos}^2i}{R}+\frac{𝔹_0𝔹_0^{}}{B_0^2}\right)+\left(\frac{𝔹_0𝔹_0^{}}{B_0^2}\right)^2,$$
()
or
$$K^2(R)\frac{\omega ^2\kappa ^2}{v_\mathrm{A}^2}\frac{1}{R^2}\left(\frac{1}{2}+\frac{q^2}{4}q\mathrm{cos}^2i\right)$$
()
for the power-law case. Local (WKB) solutions to eq. (16) have $`\mathrm{\Psi }e^{i{\scriptscriptstyle k_\mathrm{R}𝑑R}}`$ with $`Rk_\mathrm{R}1`$ and $`dk_\mathrm{R}/dRk_\mathrm{R}^2`$. In this case, $`\mathrm{\Psi }^{\prime \prime }k_\mathrm{R}^2\mathrm{\Psi }`$, and we can use $`K^2(R)=k_\mathrm{R}^2`$ to write a local dispersion relation
$$\omega ^2=v_\mathrm{A}^2k_\mathrm{R}^2,$$
(18)
which corresponds to MHD fast modes propagating along the radial direction. When $`|\omega |^2`$ is comparable to or smaller than $`v_\mathrm{A}^2/R^2`$, however, modes are not localized, and solutions must be sought as a global problem subject to boundary conditions.
### 3.2 Global Analysis for the Fundamental Modes
In the previous section we showed that there is no short wavelength (local) unstable fundamental mode with $`k_\mathrm{z}=m=0`$ in self-similar MHD disk winds. Here, we present the results of a global normal-mode analysis performed with carefully chosen boundary conditions, and adopting the power-law equilibrium. Define the dimensionless radial variable $`rR/R_\mathrm{e}`$, and dimensionless parameters $`\alpha (q1)\mathrm{cos}^2i+q(2q)/4`$ and $`\sigma ^2\omega ^2R_\mathrm{e}^2/v_\mathrm{A}^2(R_\mathrm{e})`$, with $`R_\mathrm{e}`$ being the position of the unperturbed outer edge of the wind. Then eq. (16) can be cast into the form
$$\frac{d^2\mathrm{\Psi }}{dr^2}+\left(\frac{\alpha }{r^2}+\sigma ^2r\right)\mathrm{\Psi }=0.$$
(19)
It is not difficult to show that the general solutions of eq. (19) are
$$\mathrm{\Psi }=\{\begin{array}{cc}A\sqrt{r}J_\nu (2\sigma r^{3/2}/3)+B\sqrt{r}Y_\nu (2\sigma r^{3/2}/3),\hfill & \mathrm{if}\sigma ^2>0,(20\mathrm{a})\hfill \\ & \\ C\sqrt{r}I_\nu (2|\sigma |r^{3/2}/3)+D\sqrt{r}K_\nu (2|\sigma |r^{3/2}/3),\hfill & \mathrm{if}\sigma ^2<0,(20\mathrm{b})\hfill \end{array}$$
with $`3\nu \sqrt{14\alpha }=\sqrt{(1q)^2+4(1q)\mathrm{cos}^2i}`$. In eqs. (20), $`J_\nu `$, $`Y_\nu `$, and $`I_\nu `$, $`K_\nu `$ are the ordinary and modified Bessel functions of the 1st and 2nd kinds, respectively, and the coefficients $`A,B,C`$, and $`D`$ are constants to be determined from imposed boundary conditions.
Let us consider the case of a free Lagrangian boundary at which the total pressure due to initial and perturbed fields balances with a fixed external pressure at both inner and outer edges, which is equal to the unperturbed magnetic pressure. If the total pressure at an edge of an outflow is different from the external pressure, the boundary itself will move until a new balance exists. To first order, this condition of constant pressure at the boundary is written
$$\frac{1}{2}\frac{dB_0^2}{dR}\frac{R_\mathrm{b}}{t}+𝔹_0\frac{𝔹_1}{t}=0,$$
(21)
where $`R_\mathrm{b}`$ is the location of the perturbed boundary. The first term of eq. (21) represents the change in the total pressure due to the boundary movement, while the second term arises from perturbed magnetic pressure itself. All quantities are evaluated at the unperturbed boundary position. Using eqs. (12), (13), and (15), and using $`R_\mathrm{b}/t=v_{1\mathrm{R}}`$ at boundaries, we find the desired boundary conditions are
$$\frac{d\mathrm{\Psi }}{dr}+\frac{q/2+\mathrm{sin}^2i}{r}\mathrm{\Psi }=0,\mathrm{at}r=r_\mathrm{i}\mathrm{and}\mathrm{\hspace{0.33em}\hspace{0.33em}1},$$
(22)
where $`r_\mathrm{i}R_\mathrm{i}/R_\mathrm{e}`$ is the normalized distance of an inner boundary from the axis. Together with the boundary conditions (22), eq. (19) forms a Sturm-Liouville system. By employing the variational principle one can show that $`\sigma ^2`$ is real and that $`\sigma ^2(\mathrm{\Psi })`$ is stationary subject to an arbitrary variation of $`\mathrm{\Psi }`$.
When $`\sigma ^2>0`$ (stable modes), the oscillatory properties of $`J_\nu `$ and $`Y_\nu `$ guarantee the existence of discrete eigenvalues $`\sigma _\mathrm{n}`$ with $`n`$ denoting the number of nodes in the corresponding eigenfunction $`\mathrm{\Psi }_\mathrm{n}`$. The resulting eigenvalues for $`r_\mathrm{i}=10^1`$ and $`10^4`$, and $`0<i<i_{\mathrm{max}}(q)`$ are plotted in Fig. 1. Only a few cases with small $`n`$ are shown. Eigenvalues associated with different $`i`$’s fill each shaded area completely. When $`r_\mathrm{i}=10^4`$, eigenfunctions which link inner and outer boundaries have to extend across enormous changes in density and magnetic field strengths. In this case, $`B/A1`$ and eigenvalues become rather insensitive to the local properties such as $`q`$ and $`i`$. When $`r_\mathrm{i}=10^1`$, however, the wind mimics a slender hollow cylinder. The variation in density and field strengths over radius is slight, causing eigenvalues to be sensitive to $`i`$ and $`q`$. In addition, the narrow width of the wind changes the number of nodes in eigenfunctions. When $`q>0.5`$, for example, the eigenfunctions with $`r_\mathrm{i}=10^1`$ have almost the same eigenvalues as, but one more node than, the $`r_\mathrm{i}=10^4`$ case, as seen in Fig. 1.
When $`r_\mathrm{i}1`$ and $`\sigma ^21`$, the asymptotic solutions to eqs. (20a) and (22) gives $`\sigma _\mathrm{n}=3\pi n/2+3\pi (2\nu +1)/8`$. These are plotted with dotted lines in Fig. 1b, and show good agreement with the values calculated without any assumption (even for $`n=0`$). The case with $`q=0.5`$ and $`i=0`$ is special, because the slope of the eigenfunction at the inner boundary is $`1/4`$, which automatically satisfies the boundary condition (22). In this case the asymptotic eigenvalues are $`\sigma _\mathrm{n}=3\pi n/2`$, drawn as filled circles in Fig. 1b. Eigenvalues have no upper bound as $`n\mathrm{}`$, which is a general property of solutions to a Sturm-Liouville equation (Morse & Feshbach, 1953).
Now consider the unstable global solutions with $`\sigma ^2<0`$. Let $`\psi _1`$ and $`\psi _2`$ be the two linearly independent solutions of $`\mathrm{\Psi }`$: $`\psi _1\sqrt{r}I_\nu (2|\sigma |r^{3/2}/3)`$ and $`\psi _2\sqrt{r}K_\nu (2|\sigma |r^{3/2}/3)`$ such that $`\mathrm{\Psi }=C\psi _1+D\psi _2`$. Because $`\psi _1`$ is a monotonically increasing function of $`r`$ (i.e., $`\psi _1,\psi _1^{}>0`$ always) and $`\mathrm{\Psi }`$ must have a negative logarithmic slope at the inner and outer boundaries (cf. eq. ), $`\psi _1`$ alone can not constitute eigenfunctions for global modes. In addition, since $`\psi _1`$ increases exponentially for a large value of $`|\sigma |r^{3/2}`$, while $`\psi _2,\psi _2^{}0`$, the outer free Lagrangian boundary condition requires $`C/D0`$. Although $`C`$ is not strictly zero when $`|\sigma |`$ has a relatively small value, the contribution of $`\psi _1`$ to global solutions near the inner boundary is negligibly small. Thus, unstable eigenvalues, if they exist, are essentially determined by the inner boundary condition imposed on $`\psi _2`$.
As we move inward from the outer boundary, $`\psi _2`$ rapidly increases asymptoting to
$$\psi _2r^{(13\nu )/2}\left[1\frac{\pi \nu |\sigma |^{2\nu }}{3^{2\nu }\mathrm{sin}(\pi \nu )\mathrm{\Gamma }^2(\nu +1)}r^{3\nu }+\mathrm{}\right],$$
(23)
for $`r1`$ (cf. Abramowitz & Stegun, 1965), where $`\mathrm{\Gamma }(\nu +1)`$ is a Gamma function. In fact, $`(13\nu )/2`$ is the maximum logarithmic slope $`\psi _2(r)`$ can ever attain. From the inner boundary constraint (22), the existence of unstable global solutions is guaranteed if $`(13\nu )/2>(q/2+\mathrm{sin}^2i)`$, or, equivalently
$$q>1\frac{(1+\mathrm{sin}^2i)^2}{2},$$
(24)
is satisfied. Eq. (24) is the global instability criterion for the fundamental modes of self-similar, cold, magnetized winds, subject to the free boundary conditions expressed by eq. (22). By putting $`C=0`$ and neglecting higher order terms in $`\psi _2`$, we derive from eqs. (22) and (23) the approximate, analytic expression for the eigenvalues of the global instability
$$|\sigma |r_\mathrm{i}^{3/2}=\frac{|\omega |R_\mathrm{i}}{v_\mathrm{A}(R_\mathrm{i})}=3\left[\frac{\mathrm{sin}(\pi \nu )\mathrm{\Gamma }^2(\nu +1)}{\pi \nu }\frac{(13\nu +q+2\mathrm{sin}^2i)}{(1+3\nu +q+2\mathrm{sin}^2i)}\right]^{1/2\nu },$$
(25)
which again shows that $`|\sigma |`$ has a positive real value if eq. (24) holds.
In Fig. 2 we plot the approximate growth rates for unstable modes from eq. (25) as dotted lines, as well as the exact growth rates numerically computed for $`r_\mathrm{i}=10^4`$ (thin solid lines) and for $`r_\mathrm{i}=0.1`$ (dashed lines). The curves shown are for $`i=0^\mathrm{o},5^\mathrm{o},\mathrm{},35^\mathrm{o},40^\mathrm{o}`$ from right to left, and the uppermost thick lines are for $`i_{\mathrm{max}}`$ calculated from eq. (7). Note that varying the width of outflow via $`r_\mathrm{i}`$ yields very little change in the plotted solutions: $`r_\mathrm{i}`$-dependence of the growth rates appears mainly through the product $`|\sigma |r_\mathrm{i}^{3/2}`$. Eq. (25) gives accurate growth rates for relatively small values of $`|\sigma |r_\mathrm{i}^{3/2}`$, while its estimates deviate up to $`16\%`$ from the exact values as $`|\sigma |r_\mathrm{i}^{3/2}`$ becomes larger. In this case we need to include next order terms in $`\psi _2`$ (cf. eq. ) to obtain more accurate results. For a given set of equilibrium parameters, we note that whereas there exist an infinite set of stable eigenmodes, there is (at most) a unique unstable eigenmode. Noting $`|\omega ||\sigma |v_\mathrm{A}(R_\mathrm{e})/R_\mathrm{e}=|\sigma |r_\mathrm{i}^{3/2}v_\mathrm{A}(R_\mathrm{i})/R_\mathrm{i}`$, we expect from Fig. 2 that the system is typically globally unstable within $``$5 crossing times of Alfv$`\stackrel{´}{\mathrm{e}}`$n waves at the inner boundary.
Once the ratios of the coefficients and the eigenfrequencies are found for the fundamental modes, one can easily construct radial solutions for the perturbed variables: $`\rho _1`$, $`v_{1\mathrm{R}}`$, $`v_{1\varphi }`$, and $`B_{1\varphi }`$. These are plotted in Fig. 3 for $`i=0^\mathrm{o}`$ and $`r_\mathrm{i}=10^3`$. Fig. 3a corresponds to a stable case with $`q=0.4`$ and $`\sigma _0=2.42`$, while Fig. 3b depicts an unstable case with $`q=0.6`$ and $`|\sigma |r_\mathrm{i}^{3/2}=0.11`$. Although normalization is arbitrary, we note that for the unstable modes, the negative radial velocity case drives the entire system into a more stable configuration (with lower magnetic energy) when the equilibrium magnetic field is predominantly toroidal. This can be shown as follows: Let $`\delta M`$, $`\delta E_\mathrm{B}`$, and $`\delta \mathrm{\Phi }_\mathrm{B}`$ denote the mass, magnetic energy, and toroidal magnetic flux per unit height in a local flux tube. Then we have $`\delta E_\mathrm{B}=(\pi /2)(\delta \mathrm{\Phi }_\mathrm{B}/\delta M)^2\delta M\rho R^2`$. For a given flux tube, $`\delta M`$ and $`\delta \mathrm{\Phi }_\mathrm{B}/\delta M`$ are constant in time and $`\delta M>0`$. Thus, $`\mathrm{sgn}d(\delta E_\mathrm{B})/dt=\mathrm{sgn}d(\rho R^2)/dt=\mathrm{sgn}[\rho _0R^2(v_{1\mathrm{R}}/Rv_{1\mathrm{R}}/R)]`$ from the equation of continuity. If $`v_{1\mathrm{R}}/R`$ dominates $`v_{1\mathrm{R}}/R`$ and $`v_{1\mathrm{R}}<0`$, then $`\mathrm{sgn}d(\delta E_\mathrm{B})/dt<0`$; magnetic energy decreases with time, meaning that the system evolves into a more stable state. Thus we scale $`v_{1\mathrm{R}}/v_{0\varphi }=1`$ at $`r=1`$ for Fig. 3a and $`v_{1\mathrm{R}}/v_{0\varphi }=1`$ at $`r=r_\mathrm{i}`$ for Fig. 3b, respectively. Note that stable eigenfunctions have their largest amplitude near the outer boundary, while the inner, high density region is nearly static during oscillation. Unstable eigenfunctions, on the other hand, are almost zero except in the region close to the inner boundary. The respective inner-region vs. outer-region predominance of unstable vs. stable eigenfunctions reflects the respective characteristic frequencies as well: the inner-wind unstable modes grow at large rates comparable to Alfv$`\stackrel{´}{\mathrm{e}}`$n frequencies in the interior, whereas outer-wind stable modes oscillate at low frequencies comparable to the Alfv$`\stackrel{´}{\mathrm{e}}`$n frequencies in the exterior.
We remark that there is no globally unstable fundamental mode when one adopts rigid boundaries with $`\mathrm{\Psi }(r)=0`$ at both $`r=r_\mathrm{i}`$ and $`r=1`$, instead of the free Lagrangian boundaries, since both $`\psi _1`$ and $`\psi _2`$ are monotonic functions of $`R`$. If $`\mathrm{\Psi }^{}(r)=0`$ is imposed at both boundaries (cf. Dubrulle & Knobloch, 1993), however, we still have unstable fundamental modes with a different instability criterion<sup>1</sup><sup>1</sup>1 In this case, eq. (24) would become $`(13\nu )/4=\alpha >0`$, corresponding to $`R^2(K^2(R)\omega ^2/v_\mathrm{A}^2)>0`$. and different growth rates.
We discuss the significance of fundamental modes to protostellar outflows in §8.3.
## 4 Local Analysis for Cold Winds
We now consider general non-axisymmetric Eulerian perturbations with small amplitudes. Neglecting the effects of thermal pressure and external gravity due to a central object, we linearize eqs. (1)$``$(4)
$$\frac{d\rho _1}{dt}=\frac{1}{R}\frac{}{R}(R\rho _0v_{1\mathrm{R}})\frac{1}{R}\frac{}{\varphi }(\rho _0v_{1\varphi })\frac{}{z}(\rho _0v_{1z}),$$
(26)
$$\frac{dv_{1\mathrm{R}}}{dt}=2\mathrm{\Omega }v_{1\varphi }+\frac{1}{4\pi \rho _0}\left[\frac{2B_{0\varphi }B_{1\varphi }}{R}+\frac{B_{0\varphi }}{R}\frac{B_{1\mathrm{R}}}{\varphi }+B_{0\mathrm{z}}\frac{B_{1\mathrm{R}}}{z}\frac{}{R}(𝔹_0𝔹_1)+\frac{\rho _1}{\rho _0}\left(\frac{B_{0\varphi }^2}{R}+𝔹_0𝔹_0^{}\right)\right],$$
(27)
$$\frac{dv_{1\varphi }}{dt}=\frac{\kappa ^2}{2\mathrm{\Omega }}v_{1\mathrm{R}}+\frac{1}{4\pi \rho _0}\left[\left(B_{0\varphi }^{}+\frac{B_{0\varphi }}{R}\right)B_{1\mathrm{R}}+B_{0\mathrm{z}}\left(\frac{B_{1\varphi }}{z}\frac{1}{R}\frac{B_{1\mathrm{z}}}{\varphi }\right)\right],$$
(28)
$$\frac{dv_{1\mathrm{z}}}{dt}=v_{0\mathrm{z}}^{}v_{1\mathrm{R}}+\frac{1}{4\pi \rho _0}\left[B_{0\mathrm{z}}^{}B_{1\mathrm{R}}+B_{0\varphi }\left(\frac{1}{R}\frac{B_{1\mathrm{z}}}{\varphi }\frac{B_{1\varphi }}{z}\right)\right],$$
(29)
$$\frac{dB_{1\mathrm{R}}}{dt}=\frac{B_{0\varphi }}{R}\frac{v_{1\mathrm{R}}}{\varphi }+B_{0\mathrm{z}}\frac{v_{1\mathrm{R}}}{z},$$
(30)
$$\frac{dB_{1\varphi }}{dt}=R\mathrm{\Omega }^{}B_{1\mathrm{R}}\frac{}{R}(B_{0\varphi }v_{1\mathrm{R}})+B_{0\mathrm{z}}\frac{v_{1\varphi }}{z}B_{0\varphi }\frac{v_{1\mathrm{z}}}{z},$$
(31)
$$\frac{dB_{1\mathrm{z}}}{dt}=v_{0\mathrm{z}}^{}B_{1\mathrm{R}}\frac{1}{R}\frac{}{R}(RB_{0\mathrm{z}}v_{1\mathrm{R}})+\frac{B_{0\varphi }}{R}\frac{v_{1\mathrm{z}}}{\varphi }\frac{B_{0\mathrm{z}}}{R}\frac{v_{1\varphi }}{\varphi },$$
(32)
where the Lagrangian time derivative is denoted by
$$\frac{d}{dt}\frac{}{t}+\mathrm{\Omega }(R)\frac{}{\varphi }+v_{0\mathrm{z}}\frac{}{z},$$
(33)
and again the subscripts 0 and 1 indicate the equilibrium and perturbation variables, respectively.
Since all the coefficients of the perturbed variables in eqs. (26)$``$(32) do not depend on the coordinates $`\varphi `$ and $`z`$, we may look for solutions having sinusoidal dependence on $`\varphi `$ and $`z`$. Furthermore, if there exist any normal modes, we can write eigenfunctions in the form
$$\chi _1(R,\varphi ,z,t)=\chi _1(R)e^{i(m\varphi +k_\mathrm{z}z\omega t)},$$
(34)
where $`\chi _1`$ refers to any physical variable of perturbations. Substituting eq. (34) into the set of eqs. (26)$``$(32) and eliminating all other variables in terms of the radial Lagrangian displacement $`\xi _\mathrm{R}v_{1\mathrm{R}}/i\stackrel{~}{\omega }`$ with a Doppler shifted frequency
$$\stackrel{~}{\omega }\omega m\mathrm{\Omega }k_\mathrm{z}v_{0\mathrm{z}},$$
we obtain the second order differential equation
$$\frac{d^2\xi _\mathrm{R}}{dR^2}+\frac{d}{dR}\mathrm{ln}\left(RB_0^2\frac{\stackrel{~}{\omega }_\mathrm{A}^2}{\stackrel{~}{\omega }_\mathrm{F}^2}\right)\frac{d\xi _\mathrm{R}}{dR}+\frac{H(R)}{\stackrel{~}{\omega }_\mathrm{A}^2}\xi _\mathrm{R}=0,$$
(35)
where
$`H(R)`$ $``$ $`\stackrel{~}{\omega }_\mathrm{F}^2\left\{{\displaystyle \frac{\stackrel{~}{\omega }_\mathrm{A}^2\kappa ^2}{v_\mathrm{A}^2}}F_\mathrm{B}{\displaystyle \frac{d\mathrm{ln}\rho _0}{dR}}{\displaystyle \frac{1}{R^2}}+{\displaystyle \frac{1}{RB_0^2}}{\displaystyle \frac{d}{dR}}\left(R𝔹_0𝔹_0^{}\right)\right\}`$
$``$ $`4\mathrm{\Omega }\left\{\stackrel{~}{\omega }\left({\displaystyle \frac{m}{R}}F_\mathrm{B}+{\displaystyle \frac{B_{0\mathrm{z}}}{RB_0}}G_+\right)+\mathrm{\Omega }k^2{\displaystyle \frac{B_{0\mathrm{z}}^2}{B_0^2}}v_\mathrm{A}^2k^2F_\mathrm{B}{\displaystyle \frac{B_{0\varphi }(𝕜𝔹_0)}{\stackrel{~}{\omega }B_0^2}}\right\}`$
$``$ $`\left({\displaystyle \frac{d}{dR}}\mathrm{ln}{\displaystyle \frac{\stackrel{~}{\omega }_\mathrm{F}^2}{RB_0^2}}{\displaystyle \frac{d}{dR}}\right)\left\{v_\mathrm{A}^2F_\mathrm{B}G_{}^2+2\mathrm{\Omega }G_{}\stackrel{~}{\omega }{\displaystyle \frac{B_{0\mathrm{z}}}{B_0}}+{\displaystyle \frac{v_\mathrm{A}^2}{R}}G_+G_{}\right\}`$
$``$ $`v_\mathrm{A}^2\left\{k^2F_\mathrm{B}^2{\displaystyle \frac{\stackrel{~}{\omega }_\mathrm{A}^2}{\stackrel{~}{\omega }^2}}+{\displaystyle \frac{2}{R}}F_\mathrm{B}G_+G_{}+{\displaystyle \frac{G_+^2}{R^2}}\right\},`$ ()
with
$$F_\mathrm{B}\frac{1}{B_0^2}\left(\frac{B_{0\varphi }^2}{R}+𝔹_0𝔹_{0}^{}{}_{}{}^{}\right),G_\pm \frac{1}{B_0}\left(\frac{m}{R}B_{0\mathrm{z}}\pm k_\mathrm{z}B_{0\varphi }\right),$$
()
$$\stackrel{~}{\omega }_\mathrm{A}^2\stackrel{~}{\omega }^2v_\mathrm{A}^2(𝕜𝕓_0)^2,\stackrel{~}{\omega }_\mathrm{F}^2\stackrel{~}{\omega }^2v_\mathrm{A}^2k^2,\mathrm{and}𝕜(0,\frac{m}{R},k_\mathrm{z}).$$
Here, $`F_\mathrm{B}`$ represents the equilibrium magnetic force, and $`\stackrel{~}{\omega }_\mathrm{A}`$ and $`\stackrel{~}{\omega }_\mathrm{F}`$ are frequencies connected to the Alfv$`\stackrel{´}{\mathrm{e}}`$nic and fast magnetosonic modes in cold MHD fluids, respectively. $`𝕓_0𝔹_0/|𝔹_0|`$ is the unit vector along an equilibrium field direction, and finally $`𝕜`$ is a vector wavenumber. When $`m=k_\mathrm{z}=0`$, only the terms in the first bracket in the definition of $`H(R)`$ do not vanish, recovering the radial wave equation (eq. , \[17a\]) for the fundamental mode.
The second order differential equation (35) has a singularity at $`\stackrel{~}{\omega }_\mathrm{A}^2=0`$, but if we treat a fully general problem including the thermal effects of compressible gas, we will find another singularity (a so called cusp singularity) at the positions where Doppler shifted frequencies of traveling waves match with slow MHD wave frequencies of the medium (cf. Roberts, 1985). For an incompressible medium, $`\stackrel{~}{\omega }_\mathrm{A}^2=0`$ singularities are often referred to as shear Alfv$`\stackrel{´}{\mathrm{e}}`$n singularities where because of resonances the characteristics of waves propagating radially would be modified to be either absorbed into or amplified by background medium, if considered as a boundary value problem (Ross et al., 1982; Curry & Pudritz, 1996). As pointed out by Appert et al. (1974), the locations with $`\stackrel{~}{\omega }_\mathrm{F}^2=0`$ in eq. (35) are not singularities; these cut-off points in our local analyses appear as resonance waves with frequencies having relatively small imaginary parts, suggesting potential attenuation or amplification of amplitudes.
To remove the second term in eq. (35) we further define
$$\xi _\mathrm{R}\mathrm{\Psi }\left(RB_0^2\frac{\stackrel{~}{\omega }_\mathrm{A}^2}{\stackrel{~}{\omega }_\mathrm{F}^2}\right)^{1/2}.$$
(37)
Then eq. (35) is reduced to the standard $`\mathrm{Schr}\ddot{\mathrm{o}}\mathrm{dinger}`$ form of eq. (16), with generalized $`K^2(R)`$ defined by
$$K^2(R)\frac{H(R)}{\stackrel{~}{\omega }_\mathrm{A}^2}\frac{1}{2}\frac{d^2}{dR^2}\mathrm{ln}\left(RB_0^2\frac{\stackrel{~}{\omega }_\mathrm{A}^2}{\stackrel{~}{\omega }_\mathrm{F}^2}\right)\frac{1}{4}\left[\frac{d}{dR}\mathrm{ln}\left(RB_0^2\frac{\stackrel{~}{\omega }_\mathrm{A}^2}{\stackrel{~}{\omega }_\mathrm{F}^2}\right)\right]^2.$$
(38)
Generally speaking, $`K(R)`$ is a function of $`R`$ for fixed values of $`m`$, $`k_\mathrm{z}`$, and $`\omega `$. However, we can still consider the behavior in a local sense near some fixed $`R_\mathrm{o}`$, such that $`K`$ is close to $`K(R_\mathrm{o})`$. This is mathematically formalized as described in Lin et al. (1993) and Terquem & Papaloizou (1996). Let us consider in the nonuniform background the spatially localized wave packet of the form
$$\mathrm{\Psi }=\psi (RR_\mathrm{o})e^{ik_\mathrm{R}(RR_\mathrm{o})}+O(\frac{1}{k_\mathrm{R}}),$$
where $`\psi (r)`$ is a function which is non-zero only in a small neighborhood of $`rRR_\mathrm{o}=0`$. The scale over which $`\psi (r)`$ varies significantly must tend to zero as $`k_\mathrm{R}\mathrm{}`$, but no faster than $`k_\mathrm{R}^1`$. Then, to leading order, $`d^2\mathrm{\Psi }/dR^2k_\mathrm{R}^2\mathrm{\Psi }`$, and the solution $`k_\mathrm{R}^2=K^2(R,\stackrel{~}{\omega },m,k_\mathrm{z})`$ of the $`\mathrm{Schr}\ddot{\mathrm{o}}\mathrm{dinger}`$ equation yields a local dispersion relation with the right hand side evaluated at a reference point $`R_\mathrm{o}`$, provided $`k_\mathrm{R}`$ is limited to a sufficiently large value (i.e., $`R_\mathrm{o}k_\mathrm{R}1`$). We may invert this dispersion relation to find $`\stackrel{~}{\omega }=W(m,k_\mathrm{z},k_\mathrm{R},R)`$, so that $`k_\mathrm{R}`$ now plays the role of an independent parameter and the dispersion relation yields the Doppler-shifted frequency $`\stackrel{~}{\omega }`$ of a wave near a position $`R_\mathrm{o}`$ having local wavevector $`𝕜`$. This is equivalent to a standard WKB approximation in the radial direction.
Solution of the local dispersion relation near $`R_\mathrm{o}`$ yields
$$\omega =W(m,k_\mathrm{z},k_\mathrm{R},R_\mathrm{o})+m\mathrm{\Omega }(R_\mathrm{o})+k_\mathrm{z}v_{0\mathrm{z}}(R_\mathrm{o})+𝒪(r),$$
(39)
where the $`𝒪(r)`$ term is $`(m\mathrm{\Omega }^{}(R_\mathrm{o})+k_\mathrm{z}v_{0\mathrm{z}}^{}(R_\mathrm{o}))r`$. Defining $`dk_\mathrm{R}/dR(W/R)/(W/k_\mathrm{R})k_\mathrm{R}/R`$, the WKB condition $`|dk_\mathrm{R}/dR|k_\mathrm{R}^2`$ will be satisfied for $`k_\mathrm{R}R1`$. For normal mode solutions, the $`𝒪(r)`$ term in $`\omega `$ must provide a negligible contribution to the phase; this requires that we must have $`|m\mathrm{\Omega }^{}(R_\mathrm{o})+k_\mathrm{z}v_{0\mathrm{z}}^{}(R_\mathrm{o})|tk_\mathrm{R}`$. For axisymmetric modes with negligible vertical shear, this is always satisfied. However, for $`m0`$ disturbances, or flows with non-negligible $`k_\mathrm{z}v_{0\mathrm{z}}^{}`$, spatially localized wavepackets maintain a characteristic radial wavenumber for only a limited time, altering their spatial pattern because of the background shear. For a wavepacket with initial wavenumber $`k_\mathrm{R}(0)`$, the radial wavenumber at time $`t`$ becomes, upon inclusion of $`t`$ times the $`𝒪(r)`$ term in $`\omega `$ in the phase,
$$k_\mathrm{R}(t)=k_\mathrm{R}(0)(m\mathrm{\Omega }^{}+k_\mathrm{z}v_{0\mathrm{z}}^{})t.$$
(40)
Thus, for example, with $`k_\mathrm{z}=0`$, the pitch $`\mathrm{tan}pm/Rk_\mathrm{R}`$ of a spiral pattern changes by a fraction $`ϵ=|dk_\mathrm{R}|/k_\mathrm{R}`$ over time $`t=ϵk_\mathrm{R}/|m\mathrm{\Omega }^{}|`$. If $`Rk_\mathrm{R}m`$, the pattern changes slowly compared to the orbit time. Among nonaxisymmetric disturbances, the wavepackets with low $`m/Rk_\mathrm{R}`$ have the largest temporal range for which they remain close to normal modes of the system. In the following two sections, we present solutions for the growth rates of unstable disturbances determined from a local modal analysis (i.e., producing solutions $`\stackrel{~}{\omega }=W(𝕜,R_\mathrm{o}))`$, with the understanding that when $`|m\mathrm{\Omega }^{}+k_\mathrm{z}v_{0\mathrm{z}}^{}|0`$, the modal growth (i.e., $`e^{|\stackrel{~}{\omega }|t}`$) with fixed pattern holds only for a limited time. In assessing the potential of the non-axisymmetric instabilities we shall identify to affect flow dynamics, we will consider their total amplification over times $`<k_\mathrm{R}R/m\mathrm{\Omega }`$, for which the spatial pattern changes little.
For simplicity let us define dimensionless variables
$$\sigma \stackrel{~}{\omega }R_\mathrm{o}/v_\mathrm{A}(R_\mathrm{o}),x_\mathrm{z}k_\mathrm{z}R_\mathrm{o},x_\mathrm{R}k_\mathrm{R}R_\mathrm{o},\kappa R_\mathrm{o}\kappa _\mathrm{o}/v_\mathrm{A}(R_\mathrm{o}),$$
$$\mathrm{\Omega }R_\mathrm{o}\mathrm{\Omega }_\mathrm{o}/v_\mathrm{A}(R_\mathrm{o}),\mathrm{and}\zeta R_\mathrm{o}v_{0\mathrm{z}}^{}(R_\mathrm{o})/v_\mathrm{A}(R_\mathrm{o}).$$
Here, $`\zeta `$ measures the amount of shear in the vertical velocity of the winds, and the “o” subscripts in the equilibrium epicyclic and rotation frequencies denote evaluation at the reference point $`R_\mathrm{o}`$. We adopt the power-law equilibria of §2. We now organize the terms in eq. (38) finally to get a 12th-degree polynomial
$$0=\sigma ^{12}+\underset{j=0}{\overset{10}{}}f_j(q,i,x_\mathrm{R},x_\mathrm{z},m,\mathrm{\Omega },\zeta )\sigma ^j.$$
(41)
The functional dependences of the coefficient $`f_j`$’s on the parameters are so complicated that it is not illuminating to write down the whole expression here. We may obtain more simplified forms for $`f_j`$’s by sorting out terms and taking the limit of $`x_\mathrm{R}1`$. Because there are various interesting modes which demand different regimes of parameters, however, we keep all the terms as the general local dispersion relation (with $`x_\mathrm{R}`$ large) and compute numerical growth rates by solving eq. (41) for $`\sigma `$ as a function of the other variables. We present these numerical results in §5. In §6, we will classify individual (either unstable or overstable) modes and provide their limiting dispersion relations.
Terquem & Papaloizou (1996) considered only incompressible modes, for disk applications where thermal pressure is considerable, by taking a divergence-free displacement vector as a perturbation eigenvector, thus obtaining a 4th-order polynomial. Our dispersion relation, for applications to supersonic MHD winds in which thermal pressure is negligible hence motions are compressible, contains information about all possible, either oscillating or unstable, modes of cold MHD winds.
## 5 Numerical Solutions of Modal Dispersion Relation
The derived dispersion relation (41) is a 12th-order polynomial with real coefficients, indicating that solutions appear as complex conjugate pairs. Solutions having non-zero real and imaginary parts are overstable modes, and solutions with vanishing real parts are unstable modes. In this section, we present both types of solutions which are consistent with the local analysis by fixing $`x_\mathrm{R}=10`$. As we shall show, both higher values of $`x_\mathrm{R}`$ (more spatially localized) and lower values of $`x_\mathrm{R}`$ (less spatially localized) give qualitatively the same family of solutions as with $`x_\mathrm{R}=10`$.
### 5.1 Axisymmetric Modes of Instabilities
First, we consider the axisymmetric case with $`m=0`$. For 4 selected sets of parameters, we plot the real and imaginary parts of the unstable and overstable modes in Fig. 4. A Keplerian rotation gradient with $`\kappa ^2=\mathrm{\Omega }^2`$ is assumed and vertical shear is neglected except in $`\stackrel{~}{\omega }`$. We take $`\mathrm{\Omega }`$ as arbitrary rather than using the relation (6), by allowing that the gravitational force from a central object also contributes to the equilibrium rotation velocity. Then, from eq. (5), the normalized angular velocity becomes
$$\mathrm{\Omega }^2=\kappa ^2=\mathrm{cos}^2i\frac{1+q}{2}+G_\mathrm{R},$$
(42)
where $`G_\mathrm{R}g_\mathrm{R}(R_\mathrm{o})R_\mathrm{o}/v_\mathrm{A}^2(R_\mathrm{o})>0`$ is the normalized gravitational acceleration. Note that for $`1+q>0`$ (i.e., magnetic fields decreasing outward), equilibrium solutions with $`i`$ approaching 90<sup>o</sup> require non-zero gravity (because hoop stresses do not confine a primarily-poloidal flow). Also note that as $`G_\mathrm{R}`$ strengthens, the initial equilibrium is maintained by the balance between centrifugal and gravitational forces, implying that the magnetic force is negligible.
The behavior of the solutions shown in Fig. 4 (and similar behavior for other parameters) allows us to identify 4 different axisymmetric mode families: a toroidal resonance mode (TR), an axisymmetric toroidal buoyancy mode (ATB), a poloidal buoyancy mode (PB), and a Balbus-Hawley (BH) mode. One (TR) of these is an overstable mode and the others (ATB, PB, and BH) are purely growing modes.
Fig. 4a and 4b correspond to a disk wind at large distance from the source, where magnetic fields are dominantly toroidal (small $`i`$) and centrifugal force balances magnetic force (small $`G_\mathrm{R}`$), while Fig. 4c and 4d correspond to an accretion disk or inner part of a wind where magnetic fields are poloidal (large $`i`$) and centrifugal force is balanced by the gravity from a central object (relatively large $`G_\mathrm{R}`$). In each frame, solid and dotted lines represent the imaginary and real parts of the frequencies, respectively. Fig. 4a shows the TR mode which splits into two branches in the presence of (arbitrarily small) poloidal magnetic fields (Fig. 4b and 4c). This TR mode is not a generic instability mode because it has a far larger real part (associated with ordinary MHD oscillations), indicating an overstability. With the presence of poloidal field components, there exist two different types of buoyancy modes, namely ATB and PB modes. When the pitch angle of the magnetic field is relatively small, the buoyancy instabilities are driven by the interplay of the centrifugal force with the hoop stress of toroidal fields, so we call these ATB modes. Since, as explained in §6.1.2, the ATB modes need non-zero poloidal fields as well to be unstable, they disappear when $`i=0`$. On the other hand, the PB instability modes arise when the fields are predominantly poloidal so that the pressure gradient forces of poloidal fields and the gravity from a central object are main driving forces, similar in character to the Parker instability. ATB and PB are pure instability modes with Re($`\sigma `$)=0, as shown in Fig. 4. These instability modes operate even in the arbitrarily high-$`k_\mathrm{z}`$ limit because of our cold MHD assumption; otherwise sound waves would stabilize short wavelength perturbations, as they do in the Parker instability. Fig. 4d shows that the BH instability mode appears when $`G_\mathrm{R}1`$, corresponding to dynamically weak magnetic fields in the equilibrium; with reduced $`G_\mathrm{R}`$ (also shown in Fig. 4d), BH is stabilized by radial MHD wave motions when $`x_\mathrm{R}`$ is large. As discussed in §6.1.3 and §7.1, one interesting finding in our work is that the compressible axisymmetric BH mode is strongly suppressed even for $`G_\mathrm{R}`$ large when the toroidal field is sufficiently strong; in a cold, Kepler-rotating MHD flow, it is fully stabilized when the pitch angle $`i<30^\mathrm{o}`$ (see §6.1.3).
Fig. 5 shows how the characteristics of unstable/overstable modes change as $`i`$ and $`G_\mathrm{R}`$ vary for the fixed values of $`x_\mathrm{z}=4`$, $`x_\mathrm{R}=10`$, and $`q=\zeta =0`$. For a pure toroidal field configuration with $`i=0^\mathrm{o}`$, we observe only overstable TR modes that are almost independent of $`G_\mathrm{R}`$. As $`i`$ increases, ATB emerges but is stabilized by rotation with $`G_\mathrm{R}`$ large. When $`i=45^\mathrm{o}`$ and $`q=0`$, the buoyancy mode disappears because with these parameters the net force from the background magnetic fields vanishes (cf. eqs. , , and ). When $`i>30^\mathrm{o}`$ the BH mode strengthens as $`G_\mathrm{R}`$ increases. This is because in our normalization higher values of $`G_\mathrm{R}`$ correspond to weaker equilibrium magnetic fields, with which the BH instability operates efficiently. At a pure poloidal configuration of magnetic fields, BH and PB modes remain unstable (Fig. 5d). Dotted lines at very small $`G_\mathrm{R}`$ in Fig. 5c and 5d mark the minimum value of $`G_\mathrm{R}`$, available for given values of $`q`$ and $`i`$, below which no initial equilibrium exists (cf. eq. ).
### 5.2 Non-Axisymmetric Modes of Instabilities
When non-axisymmetric perturbations are applied, the cold MHD system responds with 3 more modes which are either unstable or overstable, in addition to the axisymmetric modes. We shall refer to these as non-axisymmetric toroidal buoyancy (NTB), geometric poloidal buoyancy (GPB), and poloidal resonance (PR) modes. The PR modes are MHD waves which have non-zero azimuthal wavenumbers and become overstable when there is a radial gradient of the axial field, analogous to the TR modes. In addition to the above modes, systems with toroidal magnetic field configurations and non-zero sound speed are also subject to significant non-axisymmetric magnetorotational instability (NMRI) modes. Unlike the other three non-axisymmetric modes, the NMRI mode arises due to a differential rotation with $`d\mathrm{\Omega }/dR<0`$, where $`\mathrm{\Omega }`$ is an angular velocity of the rotation. More than anything else, the fact that the NMRI in $`B_\varphi `$-dominated systems needs a finite sound speed to be unstable distinguishes it from the axisymmetric Balbus-Hawley instability, which can be unstable regardless of temperature for purely axial fields. As we shall discuss later, the basic mechanism for the onset of NMRI is quite different from that of axisymmetric BH instability. We reserve the discussion of NMRI modes for §7, concentrating here on numerical results for our basic cold MHD system.
Fig. 6 shows the unstable and overstable solutions of the dispersion relations for two combinations of selected parameters: Fig. 6a corresponds to a disk wind with small $`i`$ and small $`G_\mathrm{R}`$, while Fig. 6b is for the near-disk case with large $`i`$ and large $`G_\mathrm{R}`$. We assume a Keplerian rotation and take an arbitrary $`\mathrm{\Omega }`$ once again using eq. (42). For all cases, we chose $`x_\mathrm{R}=10`$, $`q=\zeta =0`$, and $`m=1`$, and confirmed that changes to these parameters do not appreciably affect the qualitative results. Solid and dotted lines in Fig. 6 represent imaginary and real parts of the normalized wave frequencies, respectively. When $`i=G_\mathrm{R}=0`$, Fig. 6a shows the presence of unstable NTB and overstable TR modes which are split by the non-axisymmetry. NTB modes are nearly like PB modes in their physical basis and have an almost constant growth rate over a wide range of $`x_\mathrm{z}`$. But they depend sensitively on the logarithmic gradients of the density and magnetic structures (i.e., $`q`$; see Fig. 7). For an intermediate value of $`i`$, both TR and PR modes coexist. At some wavenumber $`x_\mathrm{z}`$, they combine to simply vanish, but overall they give rise to complicated behavior of Im($`\sigma `$). When $`i=90^\mathrm{o}`$, we observe three unstable GPB, BH, and PB modes, and overstable PR modes (Fig. 6b). GPB modes are driven by a buoyancy force together with the geometrical effect. Note that the real parts of PR modes are linearly proportional to the vertical wavenumber $`x_\mathrm{z}`$, as TR modes are, indicating that they are really overstable modes. Since $`x_\mathrm{R}m`$, however, there exists only a small contribution from non-axisymmetric effects to the axisymmetric BH and PB modes (cf. Fig. 4d). Remember that when $`mx_\mathrm{R}`$, the normal mode assumption rapidly breaks down because such high-$`m`$ modes lose their spatial pattern very quickly; we investigate $`mx_\mathrm{R}`$ cases using different methods in §§7 and 8.
Fig. 7 shows how the characteristics of the buoyancy modes change as $`x_\mathrm{z}`$, $`i`$, and $`q`$ vary. When $`G_\mathrm{R}`$=0, an initial equilibrium exists only for a limited range of $`i<i_{\mathrm{crit}}\mathrm{cos}^1\sqrt{(1+q)/2}`$ from eq. (42), with toroidal field components dominating over poloidal field components. In Fig. 7, therefore, the unstable modes with $`i<i_{\mathrm{crit}}`$ correspond to toroidal buoyancy modes, while poloidal buoyancy modes have $`i>i_{\mathrm{crit}}`$. Generally speaking, with the assumption of extremely cold medium, smaller-scale buoyancy modes with high $`x_\mathrm{z}`$ have larger growth rates. When $`i`$ is very small, as seen in Fig. 7, ATB modes are stable because they need the aid of poloidal fields to be unstable, while NTB modes become unstable for all $`i<i_{\mathrm{crit}}`$. This reflects the physically different driving mechanisms between ATB and NTB instabilities. PB modes become more unstable with higher $`q`$ (steeper background gradients), while ATB/NTB modes are more efficient with smaller $`q`$. Greater instability is simply associated with higher background magnetic force in the respective cases (cf. the initial equilibrium condition ).
The $`k_\mathrm{R}`$-dependence of the unstable/overstable modes are summarized in Fig. 8. Here we fix $`x_\mathrm{z}=2`$ for all cases and choose $`q=0.8`$, $`G_\mathrm{R}=0`$, and small $`i`$ for Fig. 8a and 8c, corresponding to disk wind-like systems, and $`q=0`$, $`G_\mathrm{R}=5`$, and large $`i`$ for Fig. 8b and 8d, corresponding to accretion disks or disk winds near their sources. The BH instability modes are completely suppressed by MHD waves when $`x_\mathrm{R}3`$; we will show that this is consistent with the prediction of the asymptotic dispersion relation. All the other modes extend with smaller growth rates to larger $`x_\mathrm{R}`$, with Im($`\sigma )x_\mathrm{R}^1`$, which we will show agrees well with the asymptotic dispersion relations (43), (45), (50), and (52) for the PB, ATB, TR, and NTB modes, respectively. For the PR modes the asymptotic dispersion relation (56), showing Im($`\sigma )x_\mathrm{R}^{2/3}`$, is valid only when $`Rk_\mathrm{z}m`$, which is not consistent with the parameters adopted in Fig. 8d. When $`k_\mathrm{R}k_\mathrm{z}m/R`$, one can confirm analytically that the PR modes also behave as Im($`\sigma )x_\mathrm{R}^1`$. In the shearing-wavelet point of view with eq. (40), Fig. 8 shows that kinematic shear arising from the background flows ultimately stabilizes both unstable and overstable modes, as $`k_\mathrm{R}`$ grows secularly increases in time. Although the local approximation breaks down if $`x_\mathrm{R}`$ is not large, Fig. 8 indicates that the BH mode exists and may show interesting behavior for small $`x_\mathrm{R}`$. In addition, Fig. 8 also suggests larger growth rates when $`x_\mathrm{R}`$ is small for other modes, although the assumptions of this section of a radially-local, slowly-changing pattern are not self-consistent when $`x_\mathrm{R}`$ is small. To study dynamical growth of disturbances which occurs when $`x_\mathrm{R}m`$, we use direct integrations of the shearing-sheet equations. We present these results in §8.2 (for the NTB modes) and §§7.2 and 7.3 (for the NMRI modes and generalized MRIs).
## 6 Mode Classification
The cold MHD system we are investigating has 8 distinct local modes with Im$`(\sigma )>0`$. Some of them (TR and PR) have larger Re$`(\sigma )`$ corresponding to overstability, while the others (PB, ATB, BH, NTB, GPB, and NMRI) have negligible Re$`(\sigma )`$, indicating pure instability. The NMRI modes do not appear in the numerical solutions because of the cold MHD assumption we made. Detailed discussion of the NMRI modes will be separately given in §7.2. In this section we describe the physical nature of the individual cold-fluid modes and present the respective dispersion relations under some limiting approximations.
### 6.1 Axisymmetric Modes
#### 6.1.1 Poloidal Buoyancy Mode
Consider a system with pure axial fields. If gravitational forces are large, then they may balance the combined outward radial centrifugal force and pressure gradient force of outward-decreasing $`B_{0\mathrm{z}}(R)`$; otherwise, if $`g_\mathrm{R}=0`$, then the strength in the magnetic fields must increase outward for an equilibrium to exist. In an initial state, at any point in the system the magnetic pressure force acting outward is balanced by the difference between gravity and the centrifugal force acting inward. If perturbed, a denser fluid element experiences reduced magnetic forces but unchanged centrifugal and gravitational forces per unit mass, and thus it would tend to sink radially inward dragging the field line with it; a lighter fluid element would correspondingly tend to float outward. Then, in a frozen-in-field condition, the neighboring gas finds itself on sloping lines of force and thus slides inward to add its weight and to cause field lines to bend more, expediting the instability. This poloidal buoyancy mode is analogous to the Parker instability (Parker, 1966), with the driving force role of external gravity in Parker’s instability replaced by combination of gravity and the centrifugal force in the PB. The PB mode can occur for both axisymmetric and non-axisymmetric disturbances.
Putting $`B_{0\varphi }=m=0`$ and considering short wavelength perturbations with $`v_{\mathrm{Az}}^2k_\mathrm{z}^2\stackrel{~}{\omega }^2`$ in eq. (38), one can find the dispersion relation for the poloidal buoyancy mode is
$$\stackrel{~}{\omega }^2=\left(\frac{1+q}{2}\right)^2\frac{v_{\mathrm{Az}}^2k_\mathrm{z}^2}{R^2(k_\mathrm{z}^2+k_\mathrm{R}^2)}=\left(\frac{|g_\mathrm{R}R\mathrm{\Omega }^2|^2}{v_{\mathrm{Az}}^2}\right)\frac{k_\mathrm{z}^2}{k_\mathrm{z}^2+k_\mathrm{R}^2}.$$
(43)
Eq. (43) states that there is no preferred length scale as long as $`k_\mathrm{z}`$ is large. However, the inclusion of thermal effects would stabilize the PB mode with shorter wavelengths, as in the Parker instability<sup>2</sup><sup>2</sup>2Also by taking a local approximation and by neglecting density stratification and the effects of thermal and cosmic ray pressures, one can simplify eq. (III.12) of Parker (1966) to get the asymptotic ($`𝕜\mathrm{}`$) dispersion relation
$$\omega ^2=\left(\frac{g^2}{v_\mathrm{A}^2}\right)\frac{k_{}^2}{k_{}^2+k_{}^2},$$
where $`g`$ is the gravity perpendicular to the galactic plane, $`v_\mathrm{A}`$ is the Alfv$`\stackrel{´}{\mathrm{e}}`$n speed of initial fields parallel to the galactic plane, and $`k_{}`$ and $`k_{}`$ are perturbation wavenumbers in the respective directions parallel and perpendicular to the galactic disk and magnetic field. Comparing the above with eq. (43), we may write $`g_{\mathrm{eff}}g_\mathrm{R}R\mathrm{\Omega }^2`$ for the PB modes, with wavenumber correspondence $`k_\mathrm{z}k_{}`$ and $`k_\mathrm{R}k_{}`$. Also, sufficiently large $`k_\mathrm{R}k_\mathrm{z}`$ stabilizes the mode.
#### 6.1.2 Axisymmetric Toroidal Buoyancy Mode
Now consider a system with weak poloidal but strong toroidal field components and negligible gravity. When magnetic fields are predominantly toroidal (i.e., when $`i<\mathrm{cos}^1\sqrt{(1+q)/2}`$ from eq. ), an initial equilibrium state is maintained by the balance mainly between the centrifugal force acting outward, and magnetic hoop stresses which act inward. With sinusoidal density perturbations with $`k_\mathrm{z}`$ imposed on the equilibrium, a heavier blob of material would tend to float radially outward under the action of unchanged centrifugal forces per unit mass but reduced specific magnetic forces; a lighter element would correspondingly tend to sink. The radial motions of the heavier and lighter blobs are in opposite directions and thus cause the poloidal field lines to bend, creating radial perturbed fields.
The azimuthal fluid motion is slightly accelerated by the tension force exerted by the initial toroidal and the perturbed radial fields (cf. $`B_{0\varphi }B_{1\mathrm{R}}/R`$ term in eq. , associated with spiral magnetic field line projections in the $`z`$=constant plane). This causes the initial poloidal component of field lines to bend now in the azimuthal direction, creating bands of perturbed azimuthal fields with signs alternating in the $`\widehat{𝕫}`$ direction. The resulting total azimuthal fields are distributed in such a way that the heavier (lighter) blob in the initial perturbation has a lower (higher) toroidal field strength. Induced motions due to the vertical magnetic pressure gradient force carry the matter from under dense to over dense regions, closing the loop and amplifying the initial perturbation.
By setting $`m=v_{0\mathrm{z}}=0`$ and taking the $`v_\mathrm{A}^2k_\mathrm{z}^2\stackrel{~}{\omega }^2`$ limit, we obtain from eq. (38) the following dispersion relation
$$0=\stackrel{~}{\omega }^4\left[v_{\mathrm{Az}}^2\left(k_\mathrm{z}^2+k_\mathrm{R}^2\right)+\kappa ^24\mathrm{\Omega }^2\frac{B_{0\mathrm{z}}^2}{B_0^2}\right]\stackrel{~}{\omega }^24\mathrm{\Omega }v_{\mathrm{A}\varphi }v_{\mathrm{Az}}F_\mathrm{B}k_\mathrm{z}\stackrel{~}{\omega }v_\mathrm{A}^2v_{\mathrm{Az}}^2F_\mathrm{B}^2k_\mathrm{z}^2,$$
(44)
with $`F_\mathrm{B}=(\mathrm{cos}^2i(1+q)/2)/R`$. When $`v_{\mathrm{A}\varphi }=0`$ (i.e., with pure poloidal fields), eq. (44) immediately recovers eq. (43), the limiting dispersion relation for PB modes. On the other hand, if $`v_{\mathrm{Az}}=0`$, there is no unstable ATB mode, clearly demonstrating that ATB modes operate by bending poloidal field lines. From Fig. 5b, we note that ATB modes are stabilized by rotation (larger $`G_\mathrm{R}`$ corresponds to stronger rotation). It can be shown from eq. (44) that when $`Rk_\mathrm{z},Rk_\mathrm{R}\stackrel{~}{\omega }`$, the critical wavenumbers are $`(k_\mathrm{z}^2+k_\mathrm{R}^2)_{\mathrm{crit}}=(d\mathrm{\Omega }^2/d\mathrm{ln}R)/(v_\mathrm{A}^2\mathrm{sin}^2i)`$, below which the system is stable against the ATB modes. For $`k_\mathrm{z}k_{\mathrm{z},\mathrm{crit}}`$, eq. (44) is further reduced to
$$\stackrel{~}{\omega }^2=\left(\mathrm{cos}^2i\frac{1+q}{2}\right)^2\frac{v_\mathrm{A}^2k_\mathrm{z}^2}{R^2(k_\mathrm{z}^2+k_\mathrm{R}^2)}=\left(\frac{|g_\mathrm{R}R\mathrm{\Omega }^2|^2}{v_\mathrm{A}^2}\right)\frac{k_\mathrm{z}^2}{k_\mathrm{z}^2+k_\mathrm{R}^2}.$$
(45)
Since ATB instabilities are axisymmetric modes, they can persist without being disturbed by the kinematic growth of $`k_\mathrm{R}`$ due to shear, if $`v_{0\mathrm{z}}^{}=0`$.
Among well-known plasma modes, the pinch or sausage mode of a plasma column is most similar to the ATB in overall geometry and effect. Both are axisymmetric and require the radial tension force from predominantly toroidal magnetic fields to drive the instability. For both the plasma pinch mode and the ATB of cold cylindrical winds, the net effect is that matter tends to be ejected radially in bands alternating with contracting magnetic field loops. However, in pinch modes, the plasma is generally unmagnetized and surrounded by external toroidal fields, and axial fields tend to suppress the instability. In the ATB, on the other hand, internal toroidal magnetic fields permeate the fluid, and non-zero axial fields are required for instability.
#### 6.1.3 Compressible Balbus-Hawley Mode
In the presence of axial magnetic fields, a differentially rotating disk is unstable to an axisymmetric incompressible perturbation (Balbus & Hawley 1991; see also Velikhov 1959 and Chandrasekhar 1960). Because this Balbus-Hawley instability<sup>3</sup><sup>3</sup>3Often referred to magnetorotational instability, or briefly, MRI. has a rapid growth time (comparable to the local rate of rotation) and exists for arbitrarily weak magnetic field strength, it is believed to provide a powerful mechanism for the generation of the effective viscosity in astrophysical accretion disks. Through numerical simulations, Hawley & Balbus (1991) argued that the roles of compressibility and toroidal fields are not significant as long as the total field strength is subthermal. Also, Blaes & Balbus (1994) studied the effect of toroidal fields on the compressible axisymmetric BH instability and showed that toroidal fields do not modify the instability criterion, while reducing growth rates slightly if $`v_{\mathrm{A}\varphi }<c_\mathrm{s}`$. We find the striking result that under extremely cold conditions (i.e., $`v_\mathrm{A}c_\mathrm{s}`$), compressibility prohibits the axisymmetric BH instability from occurring if the toroidal fields are as strong as the poloidal fields.
By taking the weak magnetic field limit ($`\mathrm{\Omega }v_\mathrm{A}k_\mathrm{R},v_\mathrm{A}k_\mathrm{z}`$) and $`m=0`$, we obtain from eq. (38) the following dispersion relation for the compressible axisymmetric BH instability in a cold MHD flow
$$\omega ^4\omega ^2[v_\mathrm{A}^2(k_\mathrm{z}^2+k_\mathrm{R}^2)+v_{\mathrm{Az}}^2k_\mathrm{z}^2+\kappa ^2]+v_\mathrm{A}^2k_\mathrm{z}^2[v_{\mathrm{Az}}^2(k_\mathrm{z}^2+k_\mathrm{R}^2)+\kappa ^24\mathrm{\Omega }^2\mathrm{sin}^2i]=0,$$
(46)
and thus from the last term in eq. (46) we obtain the instability criterion
$$v_{\mathrm{Az}}^2(k_\mathrm{z}^2+k_\mathrm{R}^2)+\kappa ^24\mathrm{\Omega }^2\mathrm{sin}^2i<0.$$
(47)
With an $`\mathrm{\Omega }R^a`$ rotation profile, eq. (47) implies that if $`\mathrm{sin}^2i<1a/2`$, we anticipate no BH instability in a cold flow. For a Keplerian rotation law with $`a=3/2`$, for instance, no axisymmetric BH instability occurs if $`i<30^\mathrm{o}`$!; when the magnetic field strength is superthermal, the inclusion of toroidal fields tends to suppress the growth of the BH instability. With a steeper rotation profile (as would occur, for example, in winds from boundary layers), there is an increase in the range of $`i`$ for which a system is BH-unstable.
We defer the full discussion on the BH instability until §7.1, where we explicitly include pressure terms in the dynamical equations.
#### 6.1.4 Toroidal Resonance Mode
Consider a system having pure toroidal fields without rotation. If the initial fields are homogeneous in space, magnetosonic waves, driven solely by magnetic pressure (with the assumption of the cold medium), would propagate without any interruption in the plane whose normal is perpendicular to the magnetic field direction. In an inhomogeneous medium, however, MHD waves no longer maintain a sinusoidal planform, and the characteristics of the waves change through the interaction with the background medium. The amplitudes of the waves may sometimes increase as they propagate, or sometimes they may become evanescent and decay at a resonance, or even may be trapped between two resonance points (cf. Rae & Roberts, 1982). In such a strongly structured medium, the classification of MHD waves is not in general possible.
Our local treatment of MHD waves can provide some insight on the amplification or evanescence of propagating MHD waves in a structured medium. For $`\stackrel{~}{\omega }_\mathrm{F}^20`$ and $`B_{0\mathrm{z}}=\mathrm{\Omega }=0`$, the local wave equation (35) can be simplified as
$$\frac{d^2\xi _\mathrm{R}}{dR^2}\frac{d\mathrm{ln}\stackrel{~}{\omega }_\mathrm{F}^2}{dR}\frac{d\xi _\mathrm{R}}{dR}+\frac{\stackrel{~}{\omega }_\mathrm{F}^2}{v_{\mathrm{A}\varphi }^2}\xi _\mathrm{R}=0.$$
(48)
Again we define $`\xi _\mathrm{R}(\stackrel{~}{\omega }_\mathrm{F}^2)^{1/2}\mathrm{\Psi }`$, then eq. (48) takes the form of eq. (16) with $`k_\mathrm{R}K(R)`$ defined by
$$k_\mathrm{R}^2\frac{\stackrel{~}{\omega }_\mathrm{F}^2}{v_{\mathrm{A}\varphi }^2}\frac{3}{4}\left(\frac{v_{\mathrm{A}\varphi }^2k_\mathrm{z}^2}{R\stackrel{~}{\omega }_\mathrm{F}^2}\right)^2,$$
(49)
where $`v_{\mathrm{A}\varphi }R^{1/2}`$ was assumed. Thus, considering limiting cases of $`k_\mathrm{z}`$ and $`k_\mathrm{R}`$, one can find the dispersion relation for this mode near the resonance frequencies (i.e., $`\stackrel{~}{\omega }v_{\mathrm{A}\varphi }k_\mathrm{z}`$)
$$\stackrel{~}{\omega }^2=\{\begin{array}{cc}v_{\mathrm{A}\varphi }^2k_\mathrm{z}^2\left[1+(3/4)^{1/2}e^{\pm \pi i/2}(Rk_\mathrm{R})^1\right],\hfill & \mathrm{for}Rk_\mathrm{R}Rk_\mathrm{z}1,\hfill \\ & \\ v_{\mathrm{A}\varphi }^2k_\mathrm{z}^2\left[1+(3/4)^{1/3}e^{\pm 2\pi i/3}(Rk_\mathrm{z})^{2/3}\right],\hfill & \mathrm{for}Rk_\mathrm{z}Rk_\mathrm{R}1,\hfill \end{array}$$
(50)
showing that the imaginary part of toroidal resonance mode vanishes quickly as $`k_\mathrm{R}\mathrm{}`$, while its real part gets bigger as $`k_\mathrm{z}`$ increases. Therefore, it is not adequate to regard TR modes as a true local instability mode. Though the TR mode is not a local instability, it suggests potential for waves to have global instabilities in which the magnetosonic resonance ($`\stackrel{~}{\omega }=v_{\mathrm{A}\varphi }k_\mathrm{z}`$) plays a similar role to the Lindblad resonance in rotating disks. Thus, waves of fixed frequency propagating with a radial component of $`𝕜`$ into their magnetosonic resonances may be amplified or reflected.
The modification of traveling waves due to the inhomogeneity of the medium is mediated through the magnetic pressure. A similar effect would occur when hydrodynamic waves propagate into an inhomogeneous medium.
### 6.2 Non-Axisymmetric Modes
#### 6.2.1 Non-Axisymmetric Toroidal Buoyancy Mode
The non-axisymmetric toroidal buoyancy mode is very similar to the PB mode in its physical mechanism, in spite of the different field geometry. For toroidal-field dominated cases, an equilibrium can exist with the net inward magnetic stresses balancing the outward centrifugal force. When the system is perturbed non-axisymmetrically, the instability would develop similarly to PB modes, as described in §6.1.1. For the $`B_\varphi `$-dominated case, however, over-dense regions float outward and under-dense regions sink, because the inward magnetic forces are enhanced when the density drops.
Setting $`B_{0\mathrm{z}}=v_{0\mathrm{z}}=0`$ and assuming $`Rk_\mathrm{z}m,Rk_\mathrm{R}`$, one can show that the general dispersion relation (38) is reduced to the following quartic equation in terms of $`\stackrel{~}{\omega }`$
$$0=\stackrel{~}{\omega }^4\left[v_{\mathrm{A}\varphi }^2\left(\frac{m^2}{R^2}+k_\mathrm{R}^2\right)+\kappa ^2\right]\stackrel{~}{\omega }^24\mathrm{\Omega }\frac{m}{R}F_\mathrm{B}v_{\mathrm{A}\varphi }^2\stackrel{~}{\omega }v_{\mathrm{A}\varphi }^4F_\mathrm{B}^2\left(\frac{m}{R}\right)^2,$$
(51)
with $`F_\mathrm{B}=(1q)/2R`$, The negative last term in eq. (51) guarantees the existence of unstable NTB modes. The third term (caused by the coupling of the rotation with the background fields) tends to stabilize NTB modes. Thus, if
$$\frac{m^2}{R^2}+k_\mathrm{R}^2<\frac{(4\mathrm{\Omega }^2\kappa ^2)}{v_{\mathrm{A}\varphi }^2},$$
there is no unstable NTB mode. Note that for wind equilibria with $`B_{0\mathrm{z}}=0`$, from eq. (6) the RHS of the above equals $`3(1q)/2R^2`$; thus the NTB instability will be present at all $`m`$ when $`1/3<q<1`$.
In the limit of large $`m`$, we obtain the asymptotic dispersion relation for the NTB mode
$$\stackrel{~}{\omega }^2=\left(\frac{1q}{2}\right)^2\frac{v_{\mathrm{A}\varphi }^2m^2}{R^2(m^2+R^2k_\mathrm{R}^2)}=\left(\frac{|g_\mathrm{R}R\mathrm{\Omega }^2|^2}{v_{\mathrm{A}\varphi }^2}\right)\frac{m^2}{m^2+R^2k_\mathrm{R}^2}.$$
(52)
Here for the second equality, eq. (6) with $`i=0^\mathrm{o}`$ is used. Eq. (52) is akin to eq. (43), the dispersion relation for the PB mode, and to eq. (45), the dispersion relation for the ATB mode, reflecting the common origin in buoyancy forces of all three. In fact, Fig. 7 clearly shows how the various buoyancy modes extend and smoothly join at intermediate pitch angles.
#### 6.2.2 Geometric Poloidal Buoyancy Mode
Now suppose a system with pure vertical fields. When perturbed azimuthally, a fluid element becomes over dense and tends to move inward due to the decreased background magnetic pressure force per unit mass if $`0<q<1`$. This geometrically converging motion of fluid increases density and field strength by factors of $`(1q)`$ and $`(1q)/2`$, respectively. On the other hand, the magnetic field enhancement induces diverging motions of the fluid in the azimuthal direction by building up a pressure gradient, tending to lower the density. The net effect of these two processes is a density increase by a factor of $`(1q)/2`$, accelerating the inward motion of the heavier element. When $`mRk_\mathrm{z}`$, the dispersion relation for this GPB mode is found to be
$$\stackrel{~}{\omega }^2=\left(\frac{1q^2}{4}\right)\frac{v_{\mathrm{Az}}^2m^2}{R^2(m^2+R^2k_\mathrm{R}^2)}.$$
(53)
When $`q=1`$, there is no instability. This is because the initial configuration of the density and the field is such that the mass and magnetic flux contained in a thin ring with the thickness $`dR`$ and the radius $`R`$ are constant over $`R`$, and no gain from the geometrical effect is possible.
#### 6.2.3 Poloidal Resonance Mode
The physical basis for the poloidal resonance mode is quite similar to that of the toroidal resonance mode. The only difference between them is the background field geometry. In the presence of pure axial fields, MHD waves with non-zero $`m`$ are easily influenced by radial magnetic pressure gradients.
To derive the dispersion relation near the resonance frequencies (i.e., $`\stackrel{~}{\omega }_\mathrm{F}0`$), let us suppose a system with pure axial fields and neglect the vertical velocity shear. The system is also assumed to rotate slowly enough that the effects of rotation may not be important in the wave dynamics (i.e., $`mv_{\mathrm{Az}}R\mathrm{\Omega }`$). For $`\stackrel{~}{\omega }^2v_{\mathrm{Az}}^2(k_\mathrm{z}^2+m^2/R^2)`$, we are left from eq. (35) with
$$\frac{d^2\xi _\mathrm{R}}{dR^2}+\frac{d}{dR}\mathrm{ln}\left(\frac{\stackrel{~}{\omega }_\mathrm{A}^2}{\stackrel{~}{\omega }_\mathrm{F}^2}\right)\frac{d\xi _\mathrm{R}}{dR}+\frac{\stackrel{~}{\omega }_\mathrm{F}^2}{v_{\mathrm{Az}}^2}\xi _\mathrm{R}=0.$$
(54)
We now define $`\xi _\mathrm{R}(\stackrel{~}{\omega }_\mathrm{F}^2/\stackrel{~}{\omega }_\mathrm{A}^2)^{1/2}\mathrm{\Psi }`$ to simplify eq. (54) into eq. (16) with $`k_\mathrm{R}K(R)`$ defined by
$$k_\mathrm{R}^2\frac{\stackrel{~}{\omega }_\mathrm{F}^2}{v_{\mathrm{Az}}^2}\frac{m^2v_{\mathrm{Az}}^2}{R^2\stackrel{~}{\omega }_\mathrm{F}^2}\left(\frac{d\mathrm{ln}\stackrel{~}{\omega }_\mathrm{F}^2}{dR}\right)^2,$$
(55)
where we took the limit of $`Rk_\mathrm{z}m`$ and assumed $`v_{\mathrm{Az}}R^{1/2}`$. Solving eq. (55) for two limits of $`k_\mathrm{R}`$, we obtain the dispersion relation near the resonance frequencies
$$\stackrel{~}{\omega }^2=\{\begin{array}{cc}v_{\mathrm{Az}}^2k_\mathrm{z}^2\left[1+e^{\pi i/3}(ma_{\mathrm{sh}})^{2/3}(R^2k_\mathrm{R}k_\mathrm{z})^{2/3}\right],\hfill & \mathrm{for}Rk_\mathrm{R}Rk_\mathrm{z}m,\hfill \\ & \\ v_{\mathrm{Az}}^2k_\mathrm{z}^2\left[1\pm e^{\pi i/2}(ma_{\mathrm{sh}})^{1/2}(Rk_\mathrm{z})^1\right],\hfill & \mathrm{for}Rk_\mathrm{z}Rk_\mathrm{R}m,\hfill \end{array}$$
(56)
where $`a_{\mathrm{sh}}1\pm 3m\mathrm{\Omega }/v_{\mathrm{Az}}k_\mathrm{z}`$, showing again a rapidly declining imaginary part as $`k_\mathrm{R}`$ increases, at which the local approximation is valid. Thus, just like TR modes, PR modes are not strictly local instability modes.
## 7 Magnetorotational Instability
### 7.1 Axisymmetric BH Instability
In the preceding section, we briefly discussed the axisymmetric BH instability in a cold, differentially rotating medium and found that the BH instability can be suppressed by the azimuthal component of magnetic fields, if the medium is cold enough. Incompressibility has generally been adopted in the study of the BH instability in an accretion disk on the grounds that in such a system the magnetic fields are subthermal and thus acoustic waves can maintain the incompressible condition over many rotation periods. For magnetocentrifugally driven winds, however, sound waves play a minor role in controlling the dynamics and thus the incompressible approximation is inapplicable. In addition, since an initial equilibrium is attained through the balance between the centrifugal and magnetic forces (cf. eq. ), the Alfv$`\stackrel{´}{\mathrm{e}}`$n crossing time scale is comparable to the rotation time scale ($`v_\mathrm{A}R\mathrm{\Omega }`$); in this case, the fields are not weak and the unstable range of wavenumbers becomes narrow.
We generalize the previous discussion of the axisymmetric compressible BH instability by explicitly including the thermal pressure terms in the momentum equation and exploring the role of compressibility to the development of the Balbus-Hawley instability. We consider a cylindrical flow threaded by both vertical and azimuthal magnetic fields, ignoring the radial variations in the initial configuration except $`\mathrm{\Omega }=\mathrm{\Omega }(R)`$ and neglecting the vertical velocity. We assume the medium is isothermal and take the WKB ($`Rk_\mathrm{z}1`$) approximation. Through the standard approach to linear analyses, we arrive at the dispersion relation for the compressible version of the BH instability
$$(\omega _\mathrm{A}^2\kappa ^2)f(\omega ^2)=k_\mathrm{R}^2\omega _\mathrm{A}^2((c_\mathrm{s}^2+v_\mathrm{A}^2)\omega ^2c_\mathrm{s}^2v_{\mathrm{Az}}^2k_\mathrm{z}^2)+4\mathrm{\Omega }^2v_{\mathrm{Az}}^2k_\mathrm{z}^2(\omega ^2c_\mathrm{s}^2k_\mathrm{z}^2),$$
(57)
where $`c_\mathrm{s}`$ is the isothermal sound speed of the medium, $`\omega _\mathrm{A}^2\omega ^2v_{\mathrm{Az}}^2k_\mathrm{z}^2`$, and $`f(\omega ^2)`$ is defined by
$$f(\omega ^2)\omega ^4\omega ^2(c_\mathrm{s}^2+v_\mathrm{A}^2)k_\mathrm{z}^2+c_\mathrm{s}^2v_{\mathrm{Az}}^2k_\mathrm{z}^4.$$
Eq. (57) is a sixth-order equation for $`\omega `$ with only even terms. When $`k_\mathrm{R}=0`$, eq. (57) is identical to eq. (64) of Blaes & Balbus (1994) or eq. (99) of Balbus & Hawley (1998). Now let us take the two opposite limits of $`c_\mathrm{s}`$ to obtain the following dispersion relations
$`\omega _\mathrm{A}^4(\kappa ^2+v_\mathrm{A}^2k_\mathrm{R}^2+v_{\mathrm{A}\varphi }^2k_\mathrm{z}^2)\omega _\mathrm{A}^2+(\kappa ^2v_{\mathrm{A}\varphi }^24\mathrm{\Omega }^2v_{\mathrm{Az}}^2)k_\mathrm{z}^2`$ $`=`$ $`0,\mathrm{for}c_\mathrm{s}0,`$ (58a)
$`(1+k_\mathrm{R}^2/k_\mathrm{z}^2)\omega _\mathrm{A}^4\kappa ^2\omega _\mathrm{A}^24\mathrm{\Omega }^2v_{\mathrm{Az}}^2k_\mathrm{z}^2`$ $`=`$ $`0,\mathrm{for}c_\mathrm{s}\mathrm{},`$ (58b)
and the corresponding instability criteria<sup>4</sup><sup>4</sup>4In fact, from eq. (57) the formal instability criterion (59b) is generic for any value of $`c_\mathrm{s}0`$; it may be written as $`v_{\mathrm{Az}}^2(k_\mathrm{z}^2+k_\mathrm{R}^2)+\kappa ^24\mathrm{\Omega }^2<0`$. However, when $`c_\mathrm{s}/v_\mathrm{A}1`$, growth rates for small $`i`$ are very low.
$`v_{\mathrm{Az}}^2(k_\mathrm{z}^2+k_\mathrm{R}^2)+\kappa ^24\mathrm{\Omega }^2\mathrm{sin}^2i`$ $`<`$ $`0,\mathrm{for}c_\mathrm{s}0,`$ (59a)
$`v_{\mathrm{Az}}^2(k_\mathrm{z}^2+k_\mathrm{R}^2)+d\mathrm{\Omega }^2/d\mathrm{ln}R`$ $`<`$ $`0,\mathrm{for}c_\mathrm{s}\mathrm{}.`$ (59b)
Note that eq. (58b) is the same as the original dispersion relation of the incompressible BH instability (eq. \[2.9\] of Balbus & Hawley 1991 without the Brunt-$`\mathrm{V}\ddot{\mathrm{a}}\mathrm{is}\ddot{\mathrm{a}}\mathrm{l}\ddot{\mathrm{a}}`$ frequency). The instability criterion (59a) in the extremely compressible limit depends explicitly on the local pitch angle, showing that as $`i`$ departs from $`90^\mathrm{o}`$ the instability becomes gradually confined to smaller values of $`k_\mathrm{z}`$. For a cold Keplerian flow, no instability occurs when $`i`$ is smaller than $`30^\mathrm{o}`$.
To examine what role thermal pressure plays to the growth of the BH instability and why the instability criterion depends on $`i`$, we plot the unstable solutions of eq. (57) as functions of $`q_\mathrm{A}(𝕜𝕧_\mathrm{A})/\mathrm{\Omega }`$ $`(=k_\mathrm{z}v_{\mathrm{Az}}/\mathrm{\Omega }`$ for the axisymmetric case) and $`\beta c_\mathrm{s}^2/v_\mathrm{A}^2`$ in Fig. 9. For the time being, we confine our discussion to the $`k_\mathrm{R}=0`$ case. When $`i=90^\mathrm{o}`$, the instability criterion from eqs. (59a,b) is $`v_\mathrm{A}^2k_\mathrm{z}^2<d\mathrm{\Omega }^2/d\mathrm{ln}R`$ and the growth rate is independent of $`\beta `$, implying that the compressibility does not alter the instability (Fig. 9a). This can be understood as follows: when magnetic fields are mainly axial, sound waves propagating along a vertical direction decouple completely from the magnetic fields and are undisturbed by rotation. But transverse MHD waves which are intrinsically incompressible are influenced by rotation to become ultimately unstable for a range of $`k_\mathrm{z}`$ when $`d\mathrm{\Omega }/dR<0`$. Therefore, $`i=90^\mathrm{o}`$ is a very special case.
On the other hand, when both vertical and azimuthal fields are present, toroidal perturbed fields generated by an initial azimuthal displacement or by sheared motion following a radial perturbation of the initial axial fields tend to cause vertical oscillations, but in a cold assumption, mainly due to the magnetic field gradient terms, $`B_{0\varphi }(B_{1\varphi }/z)`$. This oscillatory vertical motion tries to distribute the perturbed fields as uniformly as possible, thereby tending to suppress the growth of the disturbances. However, the vertical magnetic pressure gradients are not strong enough to create significant vertical motions if thermal pressure is large: a compressed region tends to expand vertically but with little change in the strength of the toroidal fields, thus providing a favorable condition for the development of the BH instability. This explains why higher $`\beta `$ cases have higher growth rates at fixed $`i`$, and why the growth rate decreases as $`i`$ decreases at fixed $`\beta `$ (Fig. 9b$``$9d).
Although the instability criterion (59b) is completely independent of the strength of the azimuthal fields provided that $`\beta 0`$, indicating as noted by Blaes & Balbus (1994) that to all orders, azimuthal fields do not modify the stability criterion, the corresponding growth rates drop progressively as $`\beta `$ decreases if $`i90^\mathrm{o}`$. When $`\beta 1`$, any change of an inclination angle $`i`$ from $`90^\mathrm{o}`$ does not bring significant reductions in growth rates, implying that the characteristics of the instability are essentially the same as the pure poloidal case. If $`\beta 1`$, however, we observe dramatic stabilizing effects from toroidal fields, as illustrated in Fig. 9.
A few comments should be devoted to the effect of $`k_\mathrm{R}`$. Radial wave motions do nothing but add another restoring force to perturbations. This in turn means that thermal pressure has a stabilizing influence on the growth of the BH instability. Thus there are two competing processes of thermal pressure: thermal pressure associated with vertical wave motion promotes the BH instability, while thermal pressure controlling radial motion opposes it. It turns out that for $`i90^\mathrm{o}`$ the former process always dominates. For $`i=90^\mathrm{o}`$, only the latter effect exists, giving higher growth rates for smaller $`\beta `$, when $`k_\mathrm{R}0`$.
Notice the stabilizing effect of $`k_\mathrm{R}`$ in eqs. (59). If the background vertical flow has significant shear, the local radial wavenumber would increase with time (cf. eq. ), suppressing the instability. Thus when $`v_{0\mathrm{z}}^{}0`$, the compressible BH instability will exhibit a transient growth, as must happen to all modes if $`k_\mathrm{z}v_{0\mathrm{z}}^{}0`$ and/or $`m\mathrm{\Omega }^{}0`$.
In conclusion, we have found that compressibility has a stabilizing effect on the axisymmetric BH instability. Even though its effect is small if the sound speed is super-Alfv$`\stackrel{´}{\mathrm{e}}`$nic, compressibility must be considered whenever the Alfv$`\stackrel{´}{\mathrm{e}}`$n speed is comparable to or even exceeds the thermal sound speed, as is expected in winds and also in disk coronae (cf. Miller & Stone, 1999).
The above discussion applies only for axisymmetric perturbations. It was also found that an accretion disk with purely toroidal fields is subject to non-axisymmetric instability (Balbus & Hawley, 1992; Terquem & Papaloizou, 1996), but we will show in the following section that the physical role of compressibility in that case is completely different, in spite of the same quantitative instability criteria.
### 7.2 Non-Axisymmetric MRI: Coherent Wavelet Analysis
Balbus & Hawley (1992) found that a differentially rotating disk of incompressible fluid with embedded toroidal magnetic field is unstable to non-axisymmetric perturbations. Adopting shearing sheet coordinates (see below), they integrated a set of the perturbed equations and showed that perturbations with an intermediate azimuthal wavenumber $`m`$ can exhibit transient, but enormous growth over a time scale of several percent of $`\mathrm{\Omega }^1`$.
An alternative approach was taken by Terquem & Papaloizou (1996) to study a similar instability to that identified by Balbus & Hawley (1992). They solved the problem using the local WKB approximation. They started from a general compressible equation of state, but subsequently they supposed divergence-free poloidal Lagrangian displacements, which made their treatment essentially incompressible. They derived a sufficient condition for the instability which is exactly the same form as that of axisymmetric BH instability (i.e., $`d\mathrm{\Omega }^2/d\mathrm{ln}R<0`$). Noting that azimuthal shear is the main driving mechanism and bending of the field lines provides a stabilizing restoring force, they suggested the non-axisymmetric instability of toroidal magnetic fields might resemble the original BH instability.
We argue in this work that the underlying physical mechanisms for non-axisymmetric toroidal-$`𝐁`$ MRI (which we refer to as “NMRI”) and axisymmetric poloidal-$`𝐁`$ MRI (which we refer to as “BH”) are in fact quite different from each other. In this section, we analyze the NMRI by looking at “coherent wavelet” solutions in which every physical variable, localized in both space and time, oscillates or grows with the same space-time dependence, and provide quantitative results in detail.
#### 7.2.1 Localization in Space and Time
We begin by considering a shearing, rotating disk with uniform density, and magnetic fields with only an azimuthal component. We ignore any unperturbed vertical motion in the medium. We include thermal pressure effects with an isothermal equation of state to obtain the explicit dependence of the NMRI on the temperature, but neglect effects of cylindrical geometry. This configuration is the same as Balbus & Hawley’s (1992), except that they considered only the incompressible case with the Boussinesq approximation, and allowed for vertical equilibrium gradients yielding buoyant oscillations. Adopting the shearing sheet coordinates $`(\stackrel{~}{R},\stackrel{~}{\varphi },\stackrel{~}{z})`$ such that $`\stackrel{~}{R}=R`$, $`\stackrel{~}{\varphi }=\varphi \mathrm{\Omega }(R)(tt_\mathrm{o})`$, and $`\stackrel{~}{z}=z`$ (Goldreich & Lynden-Bell, 1965; Julian & Toomre, 1966; Balbus & Hawley, 1992), we consider the time development of an initial plane-wave disturbance which preserves sinusoidal variation in the local rest frame of the equilibrium shearing, rotating flow
$$\chi _1(R,\varphi ,z,t)=\chi _1(t)e^{im\stackrel{~}{\varphi }+ik_\mathrm{z}\stackrel{~}{z}+ik_\mathrm{R}(t_\mathrm{o})\stackrel{~}{R}},$$
(60)
where $`k_\mathrm{R}(t_\mathrm{o})`$ is a radial wavenumber at a fiducial time $`t=t_\mathrm{o}`$. The linearized form of the MHD eqs. (1)$``$(4), can be written in dimensionless form as
$$\frac{d\alpha }{d\tau }=q_\mathrm{R}u_{1\mathrm{R}}q_\mathrm{m}u_{1\varphi }q_\mathrm{z}u_{1\mathrm{z}},$$
(61)
$$\frac{du_{1\mathrm{R}}}{d\tau }=2u_{1\varphi }q_\mathrm{m}b_\mathrm{R}+q_\mathrm{R}(\beta \alpha +b_\varphi ),$$
(62)
$$\frac{du_{1\varphi }}{d\tau }=\frac{\kappa ^2}{2\mathrm{\Omega }^2}u_{1\mathrm{R}}+\beta q_\mathrm{m}\alpha ,$$
(63)
$$\frac{du_{1\mathrm{z}}}{d\tau }=q_\mathrm{m}b_\mathrm{z}+q_\mathrm{z}(\beta \alpha +b_\varphi ),$$
(64)
$$\frac{db_\mathrm{R}}{d\tau }=q_\mathrm{m}u_{1\mathrm{R}},$$
(65)
$$\frac{db_\varphi }{d\tau }=\frac{d\mathrm{ln}\mathrm{\Omega }}{d\mathrm{ln}R}b_\mathrm{R}q_\mathrm{z}u_{1\mathrm{z}}q_\mathrm{R}u_{1\mathrm{R}},$$
(66)
$$\frac{db_\mathrm{z}}{d\tau }=q_\mathrm{m}u_{1\mathrm{z}},$$
(67)
where the dimensionless Lagrangian derivative is denoted by
$$\frac{d}{d\tau }=\frac{1}{\mathrm{\Omega }}\frac{}{t}+\frac{}{\varphi }.$$
(68)
In eqs. (61)$``$(68), all perturbed variables are dimensionless and defined by $`\alpha \rho _1/\rho _0`$, $`𝕦_\mathrm{𝟙}i𝕧_\mathrm{𝟙}/v_{\mathrm{A}\varphi }`$, $`𝕓𝔹_\mathrm{𝟙}/B_{0\varphi }`$, and $`\tau t\mathrm{\Omega }`$, and dimensionless parameters are $`\beta c_\mathrm{s}^2/v_{\mathrm{A}\varphi }^2`$, $`q_\mathrm{m}v_{\mathrm{A}\varphi }m/R\mathrm{\Omega }`$, $`q_\mathrm{z}v_{\mathrm{A}\varphi }k_\mathrm{z}/\mathrm{\Omega }`$, and
$$q_\mathrm{R}(\tau )\frac{v_{\mathrm{A}\varphi }k_\mathrm{R}(t)}{\mathrm{\Omega }}=\frac{v_{\mathrm{A}\varphi }}{\mathrm{\Omega }}\left[k_\mathrm{R}(t_\mathrm{o})m(tt_\mathrm{o})\frac{d\mathrm{\Omega }}{dR}\right]=mt\frac{v_{\mathrm{A}\varphi }}{\mathrm{\Omega }}\frac{d\mathrm{\Omega }}{dR}=\tau q_\mathrm{m}\frac{d\mathrm{ln}\mathrm{\Omega }}{d\mathrm{ln}R},$$
(69)
where the third equality holds when $`t_\mathrm{o}k_\mathrm{R}(t_\mathrm{o})/m\mathrm{\Omega }^{}`$. Eqs. (65)$``$(67) yield the divergence free condition for the perturbed magnetic fields.
Since $`q_\mathrm{R}`$ has a $`\tau `$-dependence, the linear system of eqs. (61)$``$(67) does not form an eigenvalue problem; kinematics of shear wrap a given disturbance by increasing its radial wavenumber linearly with time. In the original shearing sheet formalism, the fate of a system exposed to perturbations is analyzed through direct integrations of linearized equations over time. In doing so, one may observe transient amplification or decay of applied disturbances depending on their stability. One can say that a system is unstable if some physical variables grow sufficiently over certain time scales. The efficiency of instability for a system is identified by computing the response of the system to variation of parameters input to temporal integrations. This approach was adopted by Balbus & Hawley (1992) in their identification of the NMRI.
Here, we instead analyze the NMRI by proceeding one more step from the original shearing sheet formalism to find solutions which are localized in time as well as in space. First, we note that there exist two distinct time scales: the growth time of instabilities determined by the inverse of the dimensionless instantaneous growth rate,
$$\gamma (\tau )\frac{d}{d\tau }\mathrm{ln}\chi _1(\tau )=\frac{1}{\mathrm{\Omega }}\frac{d}{dt}\mathrm{ln}\chi _1(t),$$
(70)
and the dimensionless shearing time, $`(d\mathrm{ln}q_\mathrm{R}/d\tau )^1=q_\mathrm{R}|q_\mathrm{m}d\mathrm{ln}\mathrm{\Omega }/d\mathrm{ln}R|^1`$, as a typical time scale of the linear growth of the radial wavenumber. If the shearing time is much longer than the growth time, that is, if $`q_\mathrm{m}|d\mathrm{ln}\mathrm{\Omega }/d\mathrm{ln}R|/(\gamma q_\mathrm{R})1`$ (the “weak shear limit”), the time dependence of $`q_\mathrm{R}`$ in eq. (69) can be neglected, and thus normal mode solutions having an exponential or oscillatory behavior can be sought. Shu (1974) applied this technique to investigate the effects of a differential rotation on the Parker instability. Also, Ryu & Goodman (1992) obtained an algebraic dispersion relation for the convective instability in differentially rotating disks, by assuming that $`q_\mathrm{R}`$ is time-independent.
Because the convective and the Parker instabilities arise from hydrodynamic and magnetic buoyancy effects, respectively, independent of the rotation of a disk, one can always find a regime in which the weak shear limit is applicable. In some cases, however, as for example in the axisymmetric poloidal or the non-axisymmetric toroidal MRIs with weak magnetic fields, the instabilities result directly from a differential rotation with $`\mathrm{\Omega }^{}<0`$. In such cases, peak growth rates are of the same order as rotational frequencies (Balbus & Hawley, 1998), and thus the weak shear is not a good approximation for these non-axisymmetric instabilities.
However, we can still look for coherent solutions in which all perturbed variables vary as $`e^{\gamma \tau }`$ with time, provided the variation of the instantaneous growth rate $`\gamma (\tau )`$ over the growth time $`\gamma ^1`$ is relatively small, i.e.,
$$\left|\frac{d\mathrm{ln}\gamma (\tau )}{d\tau }\right|\gamma (\tau ).$$
(71)
We refer to the solutions under this approximation as “coherent wavelet solutions” because all physical quantities localized in both space and time grow at the same instantaneous rate. If the condition (71) holds, the changes in $`\gamma (\tau )`$ can be neglected over a short time interval, and the set of dynamical equations (61)$``$(67) constitutes an eigensystem instantaneously. This is equivalent to the WKB method in the time dimension. Since $`\gamma ^1d\mathrm{ln}\gamma (\tau )/d\tau =q_\mathrm{m}(d\mathrm{ln}\mathrm{\Omega }/d\mathrm{ln}R)(d\mathrm{ln}\gamma /d\mathrm{ln}q_\mathrm{R})/(\gamma q_\mathrm{R})`$, eq. (71) is satisfied if either $`q_\mathrm{m}|d\mathrm{ln}\mathrm{\Omega }/d\mathrm{ln}R|/(\gamma q_\mathrm{R})1`$ (the weak shear limit), or $`|d\mathrm{ln}\gamma /d\mathrm{ln}q_\mathrm{R}|1`$ (instantaneous growth rates are relatively insensitive to the radial wavenumber); the condition (71) is less restrictive and in fact is the generalization of the weak shear limit. Of course, we need to check the self-consistency of this coherent wavelet approximation by examining a posteriori whether resulting solutions satisfy the condition (71). For incompressible media, Balbus & Hawley (1992) mapped the regime of instability in \[$`(𝐤𝐯_\mathrm{A})^2`$, $`|𝐤|/k_\mathrm{z}`$\] space using WKB methods similar to those we adopt.
#### 7.2.2 Coherent Wavelet Dispersion Relation
Upon substituting eq. (70) into eqs. (61)$``$(67) and applying the approximation (71) so that $`d\chi _1/d\tau \gamma \chi _1`$, one can form a matrix equation $`\gamma Q=Q`$, where $`Q=(\alpha ,u_{1\mathrm{R}},u_{1\varphi },u_{1\mathrm{z}},b_\mathrm{R},b_\varphi ,b_\mathrm{z})^\mathrm{T}`$ is a column vector and $``$ is a $`7\times `$7 matrix whose components are determined by the coefficients of $`Q`$ in the right hand sides of eqs. (61)$``$(67). By solving the condition det($`\gamma )=0`$, where $``$ is the identity matrix, we obtain a seventh order polynomial in $`\gamma `$. As a further approximation, however, if at least one of the conditions, $`q_\mathrm{z}q_\mathrm{R}`$, $`\gamma \tau `$, or $`\gamma \tau 1`$, is satisfied, all even order terms that depend linearly on $`\tau `$ and $`q_\mathrm{m}`$ but are independent of $`q_\mathrm{z}`$, can be neglected compared to the remaining terms. The first two conditions apply when the radial wavenumber is not significant, either because disturbances are highly localized in the vertical direction ($`q_\mathrm{z}q_\mathrm{R}`$) or simply because we are looking at modal behaviors at the time $`\tau 0`$, while the third condition holds when net amplification of perturbations is large. This simplification yields a trivial solution $`\gamma =0`$ (this arises from the fact that perturbed magnetic fields, $`b_\mathrm{R}`$, $`b_\varphi `$, and $`b_\mathrm{z}`$, are linearly dependent via the divergence-free condition) and a third-order polynomial in $`\gamma ^2`$ which is the resulting instantaneous dispersion relation for NMRI
$`0=\gamma ^6`$ $`+`$ $`\gamma ^4\left[(1+\beta )q^2+q_\mathrm{m}^2+{\displaystyle \frac{\kappa ^2}{\mathrm{\Omega }^2}}\right]`$ ()
$`+`$ $`\gamma ^2\left[(1+2\beta )q^2q_\mathrm{m}^2+{\displaystyle \frac{\kappa ^2}{\mathrm{\Omega }^2}}(q_\mathrm{m}^2+(1+\beta )q_\mathrm{z}^2)\right]+\beta q_\mathrm{m}^2\left[q^2q_\mathrm{m}^2+{\displaystyle \frac{d\mathrm{ln}\mathrm{\Omega }^2}{d\mathrm{ln}R}}q_\mathrm{z}^2\right],`$
where the amplitude of the total wavenumber defined by $`q^2(\tau )q_\mathrm{R}(\tau )^2+q_\mathrm{m}^2+q_\mathrm{z}^2`$ is a function of $`\tau `$ through eq. (69). Combining eqs. (69) and (72), one can evaluate a local, instantaneous growth rate at a given time.
With vanishing magnetic fields and thermal pressure, we would obtain from eq. (72) stable epicyclic oscillations. In the limit of strong magnetic fields and no rotation, eq. (72) is immediately reduced to $`(\gamma ^2+q_\mathrm{m}^2)(\gamma ^4+(1+\beta )q^2\gamma ^2+\beta q_\mathrm{m}^2q^2)=0`$, the usual dispersion relations for the Alfv$`\stackrel{´}{\mathrm{e}}`$n waves and the fast and slow MHD waves in a medium embedded with toroidal magnetic fields. In the presence of rotation with non-vanishing but weak fields, however, these Alfv$`\stackrel{´}{\mathrm{e}}`$n and MHD modes are coupled to exhibit generally complex modal behaviors. They can be stable or unstable depending on the parameters, but it is always a slow MHD wave that becomes unstable because it has the lowest frequency so that there is a plenty of time during which destabilizing forces (centrifugal forces for NMRI) act on it. We have instantaneously growing solutions with real positive values of $`\gamma `$ provided that the last term in eq. (72) is negative. Thus, when $`\beta q_\mathrm{m}0`$, the local, instantaneous instability criterion in terms of the dimensionless variables for the NMRI can be written
$$q_\mathrm{m}^2\left(1+\frac{q_\mathrm{R}(\tau )^2+q_\mathrm{m}^2}{q_\mathrm{z}^2}\right)+\frac{d\mathrm{ln}\mathrm{\Omega }^2}{d\mathrm{ln}R}<0,$$
(73)
demonstrating that $`d\mathrm{\Omega }^2/d\mathrm{ln}R=\kappa ^24\mathrm{\Omega }^2<0`$ is indeed a sufficient condition for the instability to arise when the magnetic field strength is negligible (Balbus & Hawley, 1992; Foglizzo & Tagger, 1995; Terquem & Papaloizou, 1996). Eq. (73) recovers the results of Balbus & Hawley (1992) for the instability regime for toroidal-field NMRI. The result of eq. (73) can be compared to the poloidal-field BH instability criterion of eq. (59b). Notice that the NMRI (with $`B_{0\mathrm{z}}=0`$) vanishes completely as $`\beta 0`$, while the axisymmetric BH instability still exists even when $`\beta =0`$ (see §7.1).
Although they share the same instability criterion, the operating mechanisms for the NMRI of toroidal fields are quite different from the axisymmetric BH instability of poloidal fields. Both arise via destabilization of the slow mode. The NMRI mode, just as the poloidal BH mode, depends on shear to generate azimuthal fields from radial perturbations of the background fields. But there is more to the story. The key mechanism for the NMRI instability lies in the vertical MHD wave motions driven by the gradient of the total (initial plus perturbed) azimuthal fields, as schematically illustrated in Fig. 10. Since we suppose perturbations which are sinusoidal in both vertical and azimuthal directions, the perturbed azimuthal fields are also periodic in both directions. Rapid vertical motions with high $`k_\mathrm{z}`$, generated by $`B_{0\varphi }(B_{1\varphi }/z)`$ stress, would produce over- and under-dense regions which regularly alternate along the azimuthal direction (Fig. 10a). And then, azimuthal fluid motions are induced, according to the equation of continuity, from over-dense regions to under-dense regions (Fig. 10b). Depending on the direction ($`\widehat{\varphi }`$) of these induced motions, the coriolis and/or centrifugal force would alter the paths, radially inward or outward (Fig. 10c). Under the condition of field freezing, these radial motions would produce radial magnetic fields with a small amplitude from the background toroidal fields (Fig. 10d). These radial fields would in turn be sheared out to generate (positive or negative) perturbed azimuthal fields, due to the differential rotation of the background flows. When $`d\mathrm{\Omega }/dR<0`$, the resulting azimuthal fields from initial and perturbed ones would be distributed (Fig. 10e) such that they reinforce the applied initial vertical perturbations (Figs. 10a and 10f), implying the MRI; the entire system would just oscillate with rotation-modified MHD frequencies if $`d\mathrm{\Omega }/dR>0`$. This explains how the NMRI operates.
When $`k_\mathrm{z}`$ is large, the stabilization of the NMRI occurs when a magnetic tension from radially bent field lines exceeds the centrifugal or coriolis force (Figs. 10b and 10c). Shear Alfv$`\stackrel{´}{\mathrm{e}}`$n waves with radial polarization can suppress the instability if the field lines are sufficiently strong or if the azimuthal wavenumber is large enough, as expressed by the dimensionless parameter $`q_\mathrm{m}^2`$ outside the parentheses in eq. (73). When $`q_\mathrm{z}q_\mathrm{R}`$, on the other hand, MHD waves propagating along the radial direction stabilize the NMRI, as indicated by the terms inside parentheses in eq. (73); $`q_\mathrm{R}(\tau )`$ clearly reflects the stabilizing effect of the background shear.
The maximum instantaneous growth rate is achieved when $`q_\mathrm{R}0`$. In this case, eq. (73) implies instability if
$$q_\mathrm{m}^2<q_{\mathrm{m},\mathrm{crit}}^2\frac{1}{2}\left[q_\mathrm{z}^2+\sqrt{q_\mathrm{z}^44q_\mathrm{z}^2\frac{d\mathrm{ln}\mathrm{\Omega }^2}{d\mathrm{ln}R}}\right].$$
Note that $`q_{\mathrm{m},\mathrm{crit}}^2d\mathrm{ln}\mathrm{\Omega }^2/d\mathrm{ln}R`$, for $`q_\mathrm{z}1`$, while $`q_{\mathrm{m},\mathrm{crit}}^2q_\mathrm{z}\sqrt{d\mathrm{ln}\mathrm{\Omega }^2/d\mathrm{ln}R}`$, for $`q_\mathrm{z}1`$. Numerical solutions of eq. (72) with $`q_\mathrm{R}=0`$ are presented in Fig. 11. As both eq. (73) and Fig. 11 show, the maximum growth rates for toroidal-field background states are attained when $`q_\mathrm{z}\mathrm{}`$, which is a sharp contrast to the axisymmetric poloidal-field BH instability that has fastest growing mode at moderate $`q_\mathrm{z}`$’s (cf. Fig. 9 and see also discussion in Balbus & Hawley, 1998). But both forms of the MRI have the same maximum growth rates at the same $`q_\mathrm{A}(𝕜𝕧_\mathrm{A})/\mathrm{\Omega }`$. Fig. 11b shows how growth rates depend on the sound speed. As the sound speed increases, the medium becomes more unstable. This reflects the incompressible nature of the NMRI. Even though the marginally critical wavenumber is independent of temperature (for $`\beta 0`$), the virulence of the instability is greatly inhibited as $`\beta `$ decreases. For $`q_\mathrm{m}1`$, one can find from eq. (72) the temperature dependence of the limiting growth rate
$$\gamma =q_\mathrm{m}\sqrt{\frac{\beta }{\kappa ^2(1+\beta )}\frac{d\mathrm{\Omega }^2}{d\mathrm{ln}R}},$$
(74)
or $`\gamma =q_\mathrm{m}\sqrt{3\beta /(1+\beta )}`$ for a Keplerian rotation. Eq. (74) gives slopes of the growth rates for small $`q_\mathrm{m}`$ (Fig. 11). For magnetocentrifugally driven winds which are as cold as $`\beta <0.01`$, the NMRI is not expected to play a significant role; the growth rate in dimensional units is $`\sqrt{3}c_\mathrm{s}m/R`$.
When the medium is incompressible ($`\beta \mathrm{}`$), eq. (72) allows the analytic expression for the instantaneous growth rate for pure toroidal-field background states,
$$\gamma ^2=\{\begin{array}{cc}\frac{q_\mathrm{z}^2}{2q^2}\frac{\kappa ^2}{\mathrm{\Omega }^2}\left[\sqrt{1+16\frac{q^2q_\mathrm{m}^2}{q_\mathrm{z}^2}\frac{\mathrm{\Omega }^4}{\kappa ^4}}1\right]q_\mathrm{m}^2,\hfill & \mathrm{if}\frac{q^2q_\mathrm{m}^2}{q_\mathrm{z}^2}+\frac{d\mathrm{ln}\mathrm{\Omega }^2}{d\mathrm{ln}R}<0,\hfill \\ & \\ 0.\hfill & \mathrm{otherwise},\hfill \end{array}$$
(75)
When $`q_\mathrm{z}1`$, one can derive the maximum growth rate $`\gamma _{\mathrm{max}}=|d\mathrm{ln}\mathrm{\Omega }/d\mathrm{ln}R|/2`$, which is achieved when $`q_{\mathrm{m},\mathrm{max}}^2=(d\mathrm{ln}\mathrm{\Omega }^2/d\mathrm{ln}R)/2\gamma _{\mathrm{max}}^2`$. It can be shown from eq. (72) or (75) that $`d\gamma ^2/d\tau q_\mathrm{m}^3q_\mathrm{R}/q^20`$ as $`q_\mathrm{z}\mathrm{}`$. This proves that the coherent wavelet approximation is self-consistent for the NMRI with high $`q_\mathrm{z}`$.
#### 7.2.3 Comparison With the Shearing Sheet Formalism
In order to compare the coherent wavelet solutions for toroidal-field NMRI with the results from the shearing sheet approximation, we directly integrate eqs. (61)$``$(67) over time, with given sets of initial conditions. In Fig. 12, we display the time evolution of all perturbed variables for $`q_\mathrm{m}=0.1`$, $`q_\mathrm{z}=1`$, and $`\beta =100`$, which are the same parameters as chosen for Fig. 3 of Balbus & Hawley (1992). We adopt a Keplerian rotation profile in what follows. The initial amplitudes are 0.1 for every variable except $`b_\mathrm{R}=0.01`$ and $`b_\mathrm{z}=0.4`$, and the initial $`\stackrel{~}{\tau }`$, where the orbit number $`\stackrel{~}{\tau }\tau /2\pi =t\mathrm{\Omega }/2\pi `$, is allowed to be determined from the divergence free condition of the initial, perturbed magnetic fields. When $`\stackrel{~}{\tau }<20`$, the system responds with MHD wave motions before they start to grow. During this relaxation stage, fast MHD modes having large $`|q_\mathrm{R}|`$ are nearly longitudinal acoustic waves, affecting $`u_{1\mathrm{R}}`$ and $`\alpha `$, while $`u_{1\varphi }`$ and $`b_\varphi `$ are mostly influenced by transverse slow modes. As time increases, $`|q_\mathrm{R}|`$ gradually decreases, permitting rotational shear to affect the overall dynamics. Once the condition (73) is satisfied, shear drives the slow modes to be unstable, following the process illustrated in Fig. 10. Even though the growth of disturbances shows a transient nature due to the kinematic growth of $`q_\mathrm{R}`$, the net amplification is about 9 orders of magnitude, a bit higher than Balbus & Hawley’s result. This is because the integration interval in Balbus & Hawley covered a slightly smaller part of the unstable time range. At later time when $`q_\mathrm{R}`$ has a large value, the system exhibits stable oscillations with the slow MHD wave frequency. Fig. 12 also shows the predicted amplification magnitude (thick solid line) from the coherent wavelet approximation (see below).
In Fig. 13, we plot the numerical growth rates for each variable calculated from Fig. 12 based on the direct numerical integrations in the shearing sheet formalism, together with the growth rate of the corresponding coherent wavelet solution. Here, a dimensionless instantaneous growth rate $`\stackrel{~}{\gamma }`$ as a function of $`\stackrel{~}{\tau }`$ is defined through $`10^{\stackrel{~}{\gamma }\stackrel{~}{\tau }}=e^{\gamma \tau }`$, (or $`\stackrel{~}{\gamma }(\stackrel{~}{\tau })=2\pi \gamma (\tau )\mathrm{log}e)`$. Note that the heavy solid line for $`\stackrel{~}{\gamma }`$ drawn from eq. (72) fits well with various curves computed from the direct numerical integrations. The instantaneous growth rates are almost symmetric with respect to their maxima near $`\stackrel{~}{\tau }=0`$, as expected. Growth of the modes occurs only when $`|\stackrel{~}{\tau }|<18.3`$, which is in good agreement with the results of the direct integrations, demonstrating the validity of the coherent wavelet approximation.
We define a dimensionless amplification magnitude as
$$\mathrm{\Gamma }(\stackrel{~}{\tau })_{\mathrm{}}^{\stackrel{~}{\tau }}\stackrel{~}{\gamma }(\stackrel{~}{\tau }^{})𝑑\stackrel{~}{\tau }^{}=\mathrm{log}e_{\mathrm{}}^\tau \gamma (\tau ^{})𝑑\tau ^{}.$$
(76)
Then, $`\mathrm{\Gamma }(\stackrel{~}{\tau })`$ is an order of magnitude measurement of the amplification of an unstable mode during the time interval ($`\mathrm{},\stackrel{~}{\tau }`$). The total amplification is given by $`10^{\mathrm{\Gamma }(\mathrm{})}`$. When eq. (75) is substituted, the analytic evaluation of the integral in eq. (76) is not an easy task. In view of a shape of $`\stackrel{~}{\gamma }(\stackrel{~}{\tau })`$ (Fig. 13), we further approximate $`\stackrel{~}{\gamma }`$ with a simple form
$$\stackrel{~}{\gamma }=\{\begin{array}{cc}\stackrel{~}{\gamma }_\mathrm{o}(1|\stackrel{~}{\tau }|/\stackrel{~}{\tau }_\mathrm{c})^{1q_\mathrm{m}/2},\hfill & \mathrm{if}|\stackrel{~}{\tau }|<\stackrel{~}{\tau }_\mathrm{c},\hfill \\ 0,\hfill & \mathrm{otherwise},\hfill \end{array}$$
(77)
where
$$\stackrel{~}{\gamma }_\mathrm{o}\sqrt{2}\pi \mathrm{log}e\left\{\frac{q_\mathrm{z}^2}{q_\mathrm{m}^2+q_\mathrm{z}^2}\frac{\kappa ^2}{\mathrm{\Omega }^2}\left[\sqrt{1+16\frac{(q_\mathrm{m}^2+q_\mathrm{z}^2)q_\mathrm{m}^2}{q_\mathrm{z}^2}\frac{\mathrm{\Omega }^4}{\kappa ^4}}1\right]2q_\mathrm{m}^2\right\}^{1/2},$$
()
and the termination epoch of growth $`\stackrel{~}{\tau }_\mathrm{c}`$ is defined by
$$\stackrel{~}{\tau }_\mathrm{c}\frac{1}{2\pi q_\mathrm{m}}\left|\frac{d\mathrm{ln}\mathrm{\Omega }}{d\mathrm{ln}R}\right|^1\sqrt{\left(\frac{d\mathrm{ln}\mathrm{\Omega }^2}{d\mathrm{ln}R}\right)\frac{q_\mathrm{z}^2}{q_\mathrm{m}^2}q_\mathrm{m}^2q_\mathrm{z}^2}.$$
()
Notice that eq. (77) is valid only if $`\stackrel{~}{\tau }_\mathrm{c}`$ is real, that is, only if the condition (73) is satisfied. From eqs. (76) and (77), the total amplification magnitude is easily found to be
$$\mathrm{\Gamma }(\mathrm{})=\frac{4\stackrel{~}{\gamma }_\mathrm{o}\stackrel{~}{\tau }_\mathrm{c}}{4q_\mathrm{m}},$$
(79)
which is illustrated with solid contours in Fig. 14. Also shown with dotted contours are the direct results from numeral integration of eq. (75), which are in excellent agreement with $`\mathrm{\Gamma }(\mathrm{})`$. The thick contour is the locus of $`\stackrel{~}{\tau }_\mathrm{c}=0`$, demarcating the stable and unstable regions: the system is stable at the right hand side of the thick contour. In the $`q_\mathrm{m}q_\mathrm{z}`$ plane, the total amplification tends to be greater as $`q_\mathrm{z}`$ becomes larger and as $`q_\mathrm{m}`$ becomes smaller. This is because the NMRI with $`B_{0\mathrm{z}}=0`$ acquires maximum instantaneous growth rates at $`q_\mathrm{z}=\mathrm{}`$ (Fig. 11a) and because the shearing time is longer with smaller $`q_\mathrm{m}`$ (cf. eq. \[78b\]). For comparison, we also include in Fig. 14 the results from the shearing sheet equations for four parameter sets: ($`q_\mathrm{m}`$, $`q_\mathrm{z}`$) = (0.03, 0.1), (0.1, 1), (1, 10), and ($`\sqrt{2}`$, 100$`\sqrt{2}`$), and $`\beta `$=100 for all cases: these are marked with dots on $`q_\mathrm{m}q_\mathrm{z}`$ plane, labeled by the respective exact and estimated (in parentheses) amplification magnitudes. Note that all of the estimated amplification magnitudes are within 5% of the results of direct shearing sheet integrations. This indicates that eq. (79) is an excellent analytic estimate for the amplifications of incompressible NMRI modes.
### 7.3 Generalized MRI
Motivated by the success of the coherent wavelet method in finding the solutions of the NMRI with purely toroidal background fields, we now generalize both the axisymmetric BH and NMRI instabilities by considering non-axisymmetric perturbations applied to the rotating medium threaded by both vertical and azimuthal magnetic fields. We include the effect of thermal pressure and allow the angular velocity $`\mathrm{\Omega }`$ to be a function of $`R`$, but ignore any other radial variations in the initial state. We adopt the shearing sheet coordinates as before, and linearize eqs. (1)$``$(4). After applying perturbations in the form of eq. (60), we assume that the perturbations evolve with time as $`e^{\gamma (t)t}`$ with the coherent wavelet condition (i.e., $`d\mathrm{ln}\gamma (t)/dt\gamma (t)`$). Following the same procedure as §7.2, we obtain the general instantaneous dispersion relation for the MRI (now written in dimensional form)
$`0=\gamma ^6`$ $`+`$ $`\gamma ^4\left[(c_\mathrm{s}^2+v_\mathrm{A}^2)k^2+\kappa ^2+(𝕜𝕧_\mathrm{A})^2\right]`$ ()
$`+`$ $`\gamma ^2\left[(2c_\mathrm{s}^2+v_\mathrm{A}^2)(𝕜𝕧_\mathrm{A})^2k^2+\kappa ^2\left(c_\mathrm{s}^2k_\mathrm{z}^2+v_{\mathrm{A}\varphi }^2({\displaystyle \frac{m^2}{R^2}}+k_\mathrm{z}^2)\right)+(𝕜𝕧_\mathrm{A})k_\mathrm{z}v_{\mathrm{Az}}{\displaystyle \frac{d\mathrm{\Omega }^2}{d\mathrm{ln}R}}\right]`$
$`+`$ $`c_\mathrm{s}^2(𝕜𝕧_\mathrm{A})^2\left[(𝕜𝕧_\mathrm{A})^2k^2+{\displaystyle \frac{d\mathrm{\Omega }^2}{d\mathrm{ln}R}}k_\mathrm{z}^2\right],`$
where $`k^2k_\mathrm{R}^2(t)+m^2/R^2+k_\mathrm{z}^2`$ with the radial wavenumber defined by $`k_\mathrm{R}(t)=mtd\mathrm{\Omega }/dR`$ when we choose $`t_\mathrm{o}=k_\mathrm{R}(t_\mathrm{o})/m\mathrm{\Omega }^{}`$. When either $`m=0`$ (axisymmetric case) or $`v_{\mathrm{Az}}=0`$ (pure toroidal field case), eq. (80) becomes identical respectively with eq. (57) for the BH modes or eq. (72) for the NMRI modes.
From eq. (80), we obtain the instantaneous instability criteria for the generalized MRI modes
$`v_\mathrm{A}^2k^2(t)(𝕜𝕧_\mathrm{A})^2+\kappa ^2v_{\mathrm{A}\varphi }^2\left({\displaystyle \frac{m^2}{R^2}}+k_\mathrm{z}^2\right)+(𝕜𝕧_\mathrm{A})v_{\mathrm{Az}}k_\mathrm{z}{\displaystyle \frac{d\mathrm{\Omega }^2}{d\mathrm{ln}R}}`$ $`<\mathrm{\hspace{0.33em}0},\mathrm{for}c_\mathrm{s}=0,`$ (81a)
$`(𝕜𝕧_\mathrm{A})^2\left(1+{\displaystyle \frac{k_\mathrm{R}^2(t)+m^2/R^2}{k_\mathrm{z}^2}}\right)+{\displaystyle \frac{d\mathrm{\Omega }^2}{d\mathrm{ln}R}}`$ $`<\mathrm{\hspace{0.33em}0},\mathrm{for}c_\mathrm{s}0,`$ (81b)
which are obviously the generalizations of eqs. (59) and (73). It can also be shown that when $`i=90^\mathrm{o}`$, both equations (81a) and (81b) become identically $`v_{\mathrm{Az}}^2(k_\mathrm{R}^2(t)+m^2/R^2+k_\mathrm{z}^2)+d\mathrm{\Omega }^2/d\mathrm{ln}R<0`$. With an $`\mathrm{\Omega }R^a`$ rotation profile, eq. (81a) gives the sufficient condition for the instability in an extremely cold medium: $`\mathrm{sin}i>(42a)/(4a)`$. Keplerian flows for example become unstable only if $`i24^\mathrm{o}`$, indicating that cold, $`B_\varphi `$-dominated media are not subject to the generalized non-axisymmetric MRI disturbances, just as we found earlier that axisymmetric BH modes are also stable in cold flows for small $`i`$. We remark that the case with $`a2`$, as potentially possible in MHD winds from boundary layers, can just barely satisfy the cold medium instability criterion for $`i0`$. Note that unlike the NMRI mode with $`v_{\mathrm{Az}}=0`$, maximum growth rates in the $`c_\mathrm{s}0`$ case are not achieved as $`k_\mathrm{z}\mathrm{}`$. In fact, high-$`k_\mathrm{z}`$ or high-$`m`$ disturbances are efficiently stabilized by Alfv$`\stackrel{´}{\mathrm{e}}`$n and/or MHD waves whenever both poloidal and toroidal fields are present. However small they may be, therefore, inclusion of poloidal fields would yield a different result from the case with pure toroidal fields (this point was previously noted by Balbus & Hawley 1998). Again the stabilizing effect of kinematic shear appears through the time dependence of $`k_\mathrm{R}^2(t)`$, when $`m0`$.
Fig. 15a shows how the compressible BH modes are stabilized by azimuthal magnetosonic waves. Here, we confine consideration to the radial wavenumber $`k_\mathrm{R}=0`$. As $`q_\mathrm{m}(=v_\mathrm{A}m/R\mathrm{\Omega })`$ increases, both the growth rates and the ranges in $`q_\mathrm{z}(=v_\mathrm{A}k_\mathrm{z}/\mathrm{\Omega })`$ of unstable modes decrease. This is because if $`q_\mathrm{m}0`$, azimuthally displaced material feels relatively stronger restoring forces due to both thermal and magnetic pressures of the medium as well as stronger tension forces from bent field lines. Non-axisymmetric poloidal-field BH instability modes become stabilized with increasing values of $`m`$. When $`q_\mathrm{m}>\sqrt{d\mathrm{ln}\mathrm{\Omega }^2/d\mathrm{ln}R}`$ ($`=\sqrt{3}`$ for a Keplerian rotation), the instability is strictly cut off, even when the effect of kinematic shear is not taken into account.
We remark that the role of temperature of the medium to the BH instability is different between axisymmetric (with $`q_\mathrm{m}=0`$ and $`i90^\mathrm{o}`$; see Fig. 9) and non-axisymmetric (with $`q_\mathrm{m}0`$ and $`i=90^\mathrm{o}`$; see Fig. 15) cases. When $`i=90^\mathrm{o}`$, as already explained in §7.1, the axisymmetric BH instability with $`q_\mathrm{m}=0`$ is independent of $`\beta `$, because only Alfv$`\stackrel{´}{\mathrm{e}}`$n and sound waves exist and they do not interact with each other. When $`q_\mathrm{m}=0`$ and $`i90^\mathrm{o}`$, magnetic pressure induces vertical MHD wave motions which tend to stabilize the system when $`\beta `$ is small. If $`\beta 1`$, however, the vertical wave motions become nearly acoustic, leaving the toroidal component of perturbed fields unaffected and permitting higher growth rates. When $`q_\mathrm{m}0`$ and $`i=90^\mathrm{o}`$, on the other hand, the coupling of thermal pressure with magnetic pressure occurs through azimuthal MHD motions, and the growth rate depends only weakly on $`\beta `$.
Fig. 15b shows loci of equi-growth rate on the $`q_\mathrm{z}q_\mathrm{m}`$ plane for $`i=10^\mathrm{o}`$ and $`\beta =0.01`$ (dotted contours) and $`\beta =100`$ (thin solid contours). For $`0<q_\mathrm{m}<1.17`$, there exist upper and lower critical vertical wavenumbers, $`q_{\mathrm{z},\mathrm{u}}`$ and $`q_{\mathrm{z},\mathrm{l}}`$ such that the system is unstable with $`q_{\mathrm{z},\mathrm{l}}<q_\mathrm{z}<q_{\mathrm{z},\mathrm{u}}`$. When $`q_\mathrm{z}>q_{\mathrm{z},\mathrm{u}}`$, disturbances are stabilized by MHD waves propagating mainly along vertical direction, while perturbations with $`q_\mathrm{z}<q_{\mathrm{z},\mathrm{l}}`$ approach stable Alfv$`\stackrel{´}{\mathrm{e}}`$n waves. Each contour has a slope of $`\mathrm{tan}i`$ ($`=0.18`$ for $`i=10^\mathrm{o}`$) at both ends. Note that dotted contours with lower $`\beta `$ are labeled with much smaller growth rates than solid ones with higher $`\beta `$, even though they are similar in shape. Compared to Fig. 9 or Fig. 15a, this implies that the NMRI instability with a toroidal field configuration is more sensitive to temperature than the axisymmetric/non-axisymmetric BH instability with poloidal fields.
When $`\beta \mathrm{}`$, from eq. (80) we have the instantaneous growth rates for the generalized MRI modes
$$\gamma ^2=\frac{\kappa ^2k_\mathrm{z}^2}{2k^2}\left[\sqrt{1+16(𝐤𝐯_\mathrm{A})^2\frac{k^2\mathrm{\Omega }^2}{k_\mathrm{z}^2\kappa ^4}}1\right](𝐤𝐯_\mathrm{A})^2,$$
(82)
for $`(𝐤𝐯_\mathrm{A})^2k^2/k_\mathrm{z}^2+d\mathrm{\Omega }^2/d\mathrm{ln}R<0`$, which is also a generalization of eq. (75). With weak magnetic field strength, one can show from eq. (82) that $`\gamma ^3d\gamma ^2/dtmk_\mathrm{R}(𝐤𝐯_\mathrm{A})/(\mathrm{\Omega }k_\mathrm{z}^2)0`$ as $`(𝐤𝐯_\mathrm{A})/\mathrm{\Omega }0`$. Thus, we see that for the generalized MRIs, the coherent wavelet approach is self-consistent in the weak field limit. When the field strength is moderate, on the other hand, we obtain $`d\gamma ^2/dtm\mathrm{\Omega }^{}k_\mathrm{R}(𝐤𝐯_\mathrm{A})^2/k^2`$. Since $`\gamma \mathrm{\Omega }(𝐤𝐯_\mathrm{A})`$ in this case, the coherent wavelet condition is met only when $`m|\mathrm{\Omega }^{}|k_\mathrm{R}\gamma `$ (the weak shear limit). Of course, the predominantly toroidal-field case that becomes unstable with $`k_\mathrm{z}1`$ also satisfies the coherent wavelet condition, since $`k^2`$ becomes arbitrarily large without increasing the $`(𝐤𝐯_\mathrm{A})`$-term, as discussed in §7.2.2.
Comparing eq. (75) with eq. (82), we note that the incompressible MRI can be generalized simply by replacing the dimensionless azimuthal wavenumber $`q_\mathrm{m}`$ with $`(𝐤𝐯_\mathrm{A})/\mathrm{\Omega }`$. Therefore, we can write the net amplification magnitude for the generalized incompressible MRI as
$$\mathrm{\Gamma }(\mathrm{})=\frac{4\mathrm{log}e}{4q_\mathrm{A}}\gamma _\mathrm{o}t_\mathrm{c}$$
(83)
where the dimensional peak growth rate $`\gamma _\mathrm{o}`$ and the cut-off time of the instability $`t_\mathrm{c}`$ are defined by
$$\frac{\gamma _\mathrm{o}^2}{\mathrm{\Omega }^2}\frac{\kappa ^2\mathrm{cos}^2\theta }{2\mathrm{\Omega }^2}\left[\sqrt{1+16\frac{q_\mathrm{A}^2}{\mathrm{cos}^2\theta }\frac{\mathrm{\Omega }^4}{\kappa ^4}}1\right]q_\mathrm{A}^2$$
(84)
and
$$t_\mathrm{c}\frac{1}{\mathrm{sin}\theta }\left|\frac{d\mathrm{\Omega }}{d\mathrm{ln}R}\right|^1\sqrt{\frac{d\mathrm{ln}\mathrm{\Omega }^2}{d\mathrm{ln}R}\frac{\mathrm{cos}^2\theta }{q_\mathrm{A}^2}1},$$
(85)
respectively. The net amplification can thus be completely determined by the two parameters: the dimensionless wavenumber $`q_\mathrm{A}(𝐤𝐯_\mathrm{A})/\mathrm{\Omega }`$ projected in the direction of initial equilibrium magnetic fields, and the angle $`\theta \mathrm{tan}^1(m/Rk_\mathrm{z})`$ of the wavenumber vector with respect to the vertical axis. Note that $`\gamma _\mathrm{o}0`$ with vanishing $`q_\mathrm{A}`$, while $`t_\mathrm{c}\mathrm{}`$ as $`\theta 0`$, indicating that low-$`m`$ instabilities show higher net amplifications than high-$`m`$ disturbances as long as $`q_\mathrm{A}0`$. The total amplification magnitudes, eq. (83), are plotted in Fig. 16 with thin solid contours. For comparison, we also plot the numerical results from eqs. (76) and (82) with dotted contours. We assume a Keplerian rotation profile. The heavy curve with $`q_\mathrm{A}^2=3\mathrm{cos}^2\theta `$ draws the locus of the marginal stability. In the limit of a weak magnetic field strength (i.e., $`q_\mathrm{A}0`$), it can be shown from eqs. (83)$``$(85) that $`\mathrm{\Gamma }(\mathrm{})=(2\mathrm{log}e)\mathrm{\Omega }/(\kappa \mathrm{tan}\theta )`$, inversely proportional to $`\theta `$ (for $`\theta 1`$) but independent of $`q_\mathrm{A}`$, as illustrated in Fig. 16. Also shown in Fig. 16 are the results from the direct temporal integrations of shearing-sheet equations with $`\beta =100`$ as filled circles (for $`i=90^\mathrm{o}`$), filled triangles (for $`i=30^\mathrm{o}`$), and open circles (for $`i=0^\mathrm{o}`$), labeled by the respective exact and estimated (in parentheses) amplification magnitudes. These two results agree very well, implying that the coherent wavelet approach indeed provides excellent approximations to the solutions for amplification of generalized MRIs.
From Fig. 16, it is apparent that it is the locally near-axisymmetric (in the sense $`m/Rk_\mathrm{z}=\mathrm{tan}\theta 1`$) disturbances that experience maximum amplification, with the amplification magnitude only weakly dependent on $`q_\mathrm{A}=(𝐤𝐯_\mathrm{A})/\mathrm{\Omega }`$ within the unstable regime ($`q_\mathrm{A}1`$). The increase in amplification factor with $`Rk_\mathrm{z}/m`$ predicted from linear theory may in part explain the larger amplitudes of power spectra for modes with larger $`\widehat{𝐤}\widehat{𝐳}`$ measured from nonlinear simulations of the saturated MRI (cf. Hawley, Gammie, & Balbus, 1995). In addition, for the case of pure toroidal fields, Fig. 16 suggests only low amplification factors unless $`k_\mathrm{z}`$ is very large, which may help explain why Hawley, Gammie, & Balbus (1995) found lower magnetic field saturation amplitudes in cases with initial $`B_\mathrm{z}=0`$.
## 8 Summary and Discussion
### 8.1 General Conclusions Based on Linearized Analysis
Through linear analyses of the ideal MHD equations, we have explored the stability of shearing, rotating flows to a wide range of (primarily local) disturbances. The chief motivation for this study is to characterize the internal instabilities that could develop in disk winds that emanate from an extended region of a differentially rotating protostellar disk around a young star. The dynamics of such winds has inspired intensive theoretical effort because they may be responsible for observed YSO jets and outflows. In our analysis, we include both results based on generic density, magnetic field, and flow profiles, and results which adopt as initial equilibrium configurations the power-law asymptotic solutions of self-confined cylindrically symmetric winds presented by Ostriker (1997): $`\rho R^q,B_\varphi B_\mathrm{z}R^{(1+q)/2}`$, and $`v_\varphi v_\mathrm{z}R^{1/2}`$. For most of our analysis (§§2-6), the flows were assumed to be cold enough that thermal effects can be ignored compared with magnetic forces. To make contact with other studies of shear-induced MHD instabilities in rotating disks, we also consider stability of specific models which include non-zero thermal pressure (§7). For the lowest-order “fundamental” modes, we employ a normal-mode analysis with free Lagrangian boundary conditions to find eigenvalues and eigenfunctions of both stable and unstable modes (§3). For higher-order modes, we employ three different local techniques to study growth of unstable disturbances: In §§4-6, we present numerical and analytic solutions of dispersion relations obtained from normal mode analyses in the $`Rk_\mathrm{R}1`$ WKB limit. These are exact for $`m\mathrm{\Omega }^{}+k_\mathrm{z}v_{0\mathrm{z}}^{}=0`$ disturbances and are valid for a limited time for weakly-shearing circumstances where $`k_\mathrm{R}m|\mathrm{\Omega }^{}/\mathrm{\Omega }|,k_\mathrm{z}|Rv_{0\mathrm{z}}^{}/v_{0\mathrm{z}}|`$ (see also §8.2 below). In §7, we employ temporal integrations of the shearing-sheet equations to study MRI modes (which are cut off for $`Rk_\mathrm{R}1`$). We also introduce, in §7.2, a “coherent wavelet” formalism which adapts modal analyses for situations where shear is considerable (i.e., small $`Rk_\mathrm{R}/m`$); the coherent wavelet analysis is equivalent to a WKB approach in the temporal domain. We include a comparison of results from the shearing-sheet and coherent-wavelet techniques applied to MRIs, in §7.2.3.
Applying these techniques we have identified a total of nine different unstable or overstable families of disturbances that occur for a wide range of flow parameters: five (FM, BH, ATB, PB, and TR) of them are axisymmetric and the other four (NTB, GPB, PR, and NMRI) are non-axisymmetric. Table 1 summarizes the properties of these modes. The main general conclusions drawn from the analysis in this work can be summarized as follows:
(1) Systems having a primarily azimuthal magnetic field, for example, disk winds far from their source, are susceptible to the fundamental (FM), axisymmetric (ATB) and non-axisymmetric (NTB) toroidal buoyancy, non-axisymmetric magnetorotational instability (NMRI) and toroidal resonance (TR) modes. Unstable fundamental modes (see §3.2) are concentrated in the central parts of jets, and occur in $`B_\varphi `$-dominated flows when the logarithmic gradient of the magnetic field is steeper than $`0.75`$ (cf. eqs. and ). Growth rates of unstable FM are comparable to inner-wind Alfv$`\stackrel{´}{\mathrm{e}}`$n frequencies. Long wavelength modes with large amplitudes at large radii are all stable, for power-law wind profiles. The TR mode (see §6.1.4) is an overstability, with growth suppressed when $`k_\mathrm{R}`$ increases through shear of the vertical velocity, simply becoming oscillatory MHD waves. The axisymmetric toroidal buoyancy mode (ATB; §6.1.2) is activated initially by the buoyancy force and subsequently by bending poloidal magnetic fields. In geometrical form, it is locally similar to the sausage mode of a plasma column confined by toroidal fields, and leads to radial mixing. Because growth rates are larger on smaller scales, ATB can contribute to the generation of local turbulence in disk winds. The non-axisymmetric toroidal buoyancy mode (NTB; see §6.2.1) is much like the Parker instability, but with the centrifugal force replacing the role of external gravity. Although the normal-mode analysis for NTB has the largest temporal validity at small $`m/Rk_\mathrm{R}`$, the instantaneous growth rate increases with increasing $`m/Rk_\mathrm{R}`$ (cf. eq. and Fig. 8c). We thus return to the NTB in §8.2, below, applying time-dependent techniques to study the $`Rk_\mathrm{R}/m1`$ limit. Because the NTB is present whenever radial magnetic forces are non-zero, it may be important in promoting radial mixing. Both of the toroidal buoyancy instabilities require non-zero magnetic forces in the equilibrium state. The rarefied and cold conditions of disk winds do not favor the development of the NMRI (see §7.2.2 and §7.3). Like the original (poloidal field) magnetorotational (BH) instability, NMRI requires $`d\mathrm{ln}\mathrm{\Omega }^2/d\mathrm{ln}R<0`$, but also requires a relatively incompressible medium, as is provided by the relatively dense and warm ($`c_\mathrm{s}v_\mathrm{A}`$) conditions in an accretion disk. We show the NMRI vanishes in the limit of $`c_\mathrm{s}/v_\mathrm{A}0`$. We further discuss perturbations in cold, $`B_\varphi `$-dominated flows in §§8.2 and 8.3, below.
(2) Systems having primarily axial magnetic fields, for example, accretion disks or winds very near their origin, are susceptible to the Balbus-Hawley (BH), poloidal buoyancy (PB and GPB), and poloidal resonance (PR) modes. The well-known axisymmetric Balbus-Hawley instability (BH; see §6.1.3 and §7.1) is the most efficient member of the family of magnetorotational instabilities (MRIs; see §7.3). It will work to produce channel flows, eventually generating fully-developed MHD turbulence through coupling to non-axisymmetric disturbances in the non-linear stage. Driven by background magnetic pressure and the centrifugal force, the axisymmetric poloidal buoyancy mode (PB; see §6.1.1) requires a gradient in the magnetic field strength to be unstable. If the field distribution is steep enough, the poloidal buoyancy modes would also work effectively to generate radial mixing and turbulence over much smaller scales than the BH instability. Because of their overstable characteristics, the impact on the system of poloidal resonance modes (PR; see §6.2.3) would be best evaluated with a global rather than local formalism. Configurations with shallower background magnetic gradients $`(q<1)`$ are also subject to a non-axisymmetric poloidal buoyancy instability (GPB; see §6.2.2) which arises in part from geometric effects.
(3) In distinction to the original, incompressible, axisymmetric BH instability, we found that the compressible axisymmetric BH mode is strongly stabilized by the presence of an azimuthal magnetic field if the medium has substantially sub-Alfv$`\stackrel{´}{\mathrm{e}}`$nic sound speeds. For example, in a cold rotating flow with $`\mathrm{\Omega }R^{3/2}`$, the axisymmetric BH instability would be completely suppressed if the local pitch angle $`i\mathrm{tan}^1(B_\mathrm{z}/B_\varphi )`$ is less than 30<sup>o</sup> (cf. eq. ). In an incompressible medium (as provided by a disk with $`c_\mathrm{s}v_\mathrm{A}`$), faster sound waves preserve perturbed toroidal fields from being dispersed by MHD wave motions, thereby providing a favorable condition for the BH instability. When the field configuration is purely poloidal, the compressible BH instability is identical with its incompressible counterpart, independent of temperature (cf. eqs. and ).
(4) Even though they share the same instability criterion (cf. eqs. \[59b\] and ), the operating mechanisms for the NMRI of purely toroidal $`𝔹`$-fields is entirely different from the axisymmetric BH instability of primarily poloidal $`𝔹`$-fields. In the NMRI (see §7.2), vertical MHD wave motions driven by magnetic pressure play an essential role in the feedback loop for induced radial disturbances, while the axisymmetric BH instability tends to be stabilized by vertical wave motions. Faster sound speeds produce higher growth rates in both instabilities, but for different reasons: in the NMRI by activating azimuthal fluid motions preceded by the vertical MHD wave motions; in the BH instability by maintaining the perturbed azimuthal fields generated by shear (when $`B_{0\varphi }0`$). Because of their non-axisymmetric nature, the NMRI has a transient growth, stabilized by the growth of $`k_\mathrm{R}`$ from kinematic azimuthal shear. For the NMRI mode, we show explicitly by comparison to direct temporal integrations of the shearing sheet equations that the growth rate at $`k_\mathrm{R}=0`$ can be used to provide a good estimate of the net amplification magnitude (see §7.2.3).
(5) The coherent wavelet formalism we develop (§7.2) may be used to compute instability criteria and net amplification factors for generalized MRI disturbances with arbitrary magnetic field and wavevector orientations (§7.3). Eq. (80) gives the instantaneous dispersion relation for generalized MRIs. For strongly compressible flows ($`c_\mathrm{s}/v_\mathrm{A}0`$), instability does not occur in $`B_\varphi `$-dominated configurations (cf. eq. \[81a\]); in this case, flows with an $`\mathrm{\Omega }R^{3/2}`$ rotation law can be unstable only when the magnetic pitch angle $`i>24^\mathrm{o}`$. Because MHD disk winds generally have very small pitch angles, this result has the important implication that such winds will not be subject to the development of strong internal turbulence that occurs as a consequence of nonlinear MRIs in disks. The absence of MRIs in cold, $`B_\varphi `$-dominated winds may be crucial in enabling them to propagate over large distances from their sources. High-$`k_\mathrm{z}`$ and/or high-$`m`$ modes of the generalized MRI are stabilized by MHD waves, which is a sharp contrast with the NMRI of purely toroidal fields in which maximum growth rates are attained at $`k_\mathrm{z}\mathrm{}`$. For incompressible flows, the amplification factor for all MRIs can be written analytically in terms of $`q_\mathrm{A}=(𝐤𝐯_\mathrm{A})/\mathrm{\Omega }`$ and $`\theta =\mathrm{tan}^1(m/Rk_\mathrm{z})`$ (eqs. $``$); within the unstable regime ($`q_\mathrm{A}<|d\mathrm{ln}\mathrm{\Omega }/d\mathrm{ln}R|^{1/2}`$, from eq. \[81b\]), the amplification is $`\mathrm{exp}[2\mathrm{\Omega }/(\kappa \mathrm{tan}\theta )]`$, favoring “locally-axisymmetric” disturbances.
### 8.2 Effect of Shear on Dynamical Growth of Buoyancy Instabilities
Apart from the results of §7 where we adopted the shearing sheet formulation of the dynamical equations to study MRIs, the results in this work have been elicited on the basis of the local normal mode analyses. As described in §4, these modes may have a limited range of temporal validity, due to the effects of background shear. For axisymmetric disturbances with negligible vertical shear (i.e., $`m\mathrm{\Omega }^{}+k_\mathrm{z}v_{0\mathrm{z}}^{}0`$), the results presented in §5 and §6 are acceptable for all time; the modes with pure imaginary $`\omega `$ will show an exponential growth without interruption over arbitrarily long time until nonlinearity sets in. However, for non-axisymmetric disturbances, or for flows with non-negligible vertical shear, unstable modes identified in §5 and §6 are not purely growing. As time evolves, the differential velocities build up the radial wavenumber through the kinematic shear (cf. eq. ), which in turn tends to stop the further growth of disturbances. This can be seen directly through the suppression of instabilities in the local analysis when $`k_\mathrm{R}`$ is large (cf. Fig. 8). The characteristic time for the wave pattern to change by a fraction $`ϵ`$ is $`t=ϵk_\mathrm{R}/|m\mathrm{\Omega }^{}+k_\mathrm{z}v_{0\mathrm{z}}^{}|`$; over this interval, the disturbance will be amplified by a factor $`\mathrm{exp}(ϵk_\mathrm{R}\mathrm{Im}(\stackrel{~}{\omega })/|m\mathrm{\Omega }^{}+k_\mathrm{z}v_{0\mathrm{z}}^{}|)`$. When $`k_\mathrm{R}m/R,k_\mathrm{z}`$, Fig. 8 shows that Im($`\stackrel{~}{\omega })k_\mathrm{R}^1`$, so that the net amplification factor is nearly independent of $`k_\mathrm{R}`$. Since, however, Im($`\stackrel{~}{\omega }`$) is not larger than $`|m\mathrm{\Omega }^{}+k_\mathrm{z}v_{0\mathrm{z}}^{}|k_\mathrm{R}^1`$, only order-unity amplification can be expected for disturbances which are consistent with the requirements for quasi-steady normal mode analysis.
Because the normal-mode dispersion relations indicate larger values of growth rates when $`Rk_\mathrm{R}/m`$ is small (which is however not self-consistent with the WKB treatment), it is desirable to extend investigation to allow for $`Rk_\mathrm{R}/m`$ small. The coherent wavelet formalism used for the MRI in §7 suggests that when $`m1`$ (or $`Rk_\mathrm{z}1`$ for $`v_{0\mathrm{z}}^{}0`$ cases), this can be done by regarding $`k_\mathrm{R}`$ as a time-dependent variable according to eq. (40) and using the asymptotic dispersion relations of §6 (i.e., eq. for PB, eq. for ATB, eq. for NTB, and eq. for GPB). To verify this argument, we specifically consider the NTB modes (which are one of the chief instabilities in $`B_\varphi `$-dominated winds) and compare the results with the shearing sheet temporal integrations. For the latter, we set $`B_\mathrm{z}=0`$ and integrate eqs. (26)$``$(31) in time, setting all of the coefficients to constant values. The resulting instantaneous growth rates and time evolutions of variables are plotted in Fig. 17 as functions of the normalized time $`\tau =t\mathrm{\Omega }`$. We omitted the $`k_\mathrm{R}`$-dependent term in eq. (26) in order to remove rapid oscillations arising from a phase mismatch between the density and radial velocity; the amplitude evolution is independent of this term. We also neglected the vertical velocity shear and and selected $`q=k_\mathrm{R}(0)=k_\mathrm{z}=0`$, $`R\mathrm{\Omega }=0.1v_{\mathrm{A}\varphi }`$, and $`m=100`$. As initial conditions, we chose 0.1 for every variable except $`b_\mathrm{R}=0.01`$ and integrated the system of the linearized equations. Various curves are computed from direct numerical integrations of shearing sheet equations, while the heavy solid lines are drawn from the normal mode solution, eqs. (52) and (86) (see below), after taking allowance for the time dependence of $`k_\mathrm{R}`$. The rapid fluctuations of the perturbed variables for $`\tau <0`$ are due to MHD waves with high $`|k_\mathrm{R}|`$, disappearing after variables grow substantially. Again, most of growth occurs over a relatively short period of time near $`k_\mathrm{R}0`$. Note an excellent agreement between the results from two different approaches; we have also obtained similar results with integration from other initial conditions. This confirms that our normal mode results can also be applied to high-$`m`$ disturbances if $`k_\mathrm{R}`$ is allowed to vary with time.
Using eq. (52) with $`k_\mathrm{R}(t)`$ from eq. (40), we integrate $`\stackrel{~}{\omega }`$ over time to estimate the net amplification for the NTB modes
$$\frac{\chi _1(t)}{\chi _1(0)}=\mathrm{exp}_0^t\mathrm{Im}(\stackrel{~}{\omega })𝑑t=\left[\frac{3}{2}\mathrm{\Omega }t+\sqrt{1+\left(\frac{3}{2}\mathrm{\Omega }t\right)^2}\right]^{(1q)v_{\mathrm{A}\varphi }/3R\mathrm{\Omega }},$$
(86)
where we put $`v_{0\mathrm{z}}^{}=k_\mathrm{R}(0)=0`$ and assume a Keplerian rotation. Thus instead of an exponential growth, at later time of evolution we have the power-law growth due to the kinematic shear. This behavior is distinct from the MRI modes, which are strictly stable for large enough $`k_\mathrm{R}`$. However, the continued local growth of buoyancy perturbations is offset by the role of kinematic shear in mixing phases of disturbances. Considering waves on $`z`$=constant plane in a square with sides $`L`$, the maximum averaged contrast in any variable relative to the mean value is $`(\lambda /L)\chi _1(t)`$, where $`\lambda `$ is the local wavenumber of the waves. With $`\lambda k_\mathrm{R}(t)^1`$ and using eqs. (6) and (86), the average contrast for the NTB modes evolve with time as $`(\mathrm{\Omega }t)^{\sqrt{2(1q)}/31}`$, vanishing as $`t\mathrm{}`$ for $`0<q<1`$. From eq. (86), amplification factors are essentially scale-free. Although there may be a significant growth of the NTB modes on large scales, their dynamical effect on small scales is limited by the phase mixing due to shear.
In the presence of vertical shear, the evolution of ATB modes is also affected by the kinematic growth of $`k_\mathrm{R}`$ although they are axisymmetric modes. The net amplification of the ATB modes follows a power-law growth as that of the NTB mods does. In fact, eq. (86) with $`\mathrm{\Omega }`$ replaced by $`2v_{0\mathrm{z}}^{}/3`$ gives the temporal behavior of the net amplification for the ATB modes.
### 8.3 Discussion of Applications to Protostellar Winds
In order for a disk wind to overcome the gravitational barrier due to a central object and to be centrifugally launched from the surface of a Keplerian disk, the poloidal components of field lines should thread the disk at an angle of 30<sup>o</sup> or more from the axis (Blandford & Payne, 1982). Once material starts to flow outward along such field lines, it is accelerated primarily in the radial direction by the centrifugal force or by the pressure gradient in the toroidal field. Beyond the Alfv$`\stackrel{´}{\mathrm{e}}`$n surface where the local, poloidal component of the flow velocity is equal to that of the Alfv$`\stackrel{´}{\mathrm{e}}`$n wave velocity, the magnetic field is not strong enough to play a role of “a rigid wire”, and the inertia of gas becomes important, winding up the field lines to be progressively more toroidal. In this process, the azimuthal flow velocity decreases below the corotation value. With the increase in the azimuthal component of the magnetic field, the associated hoop stress provides the collimation of the outflow and causes the streamlines to bend upward. The radial flow velocity of the outflow is still positive, although it decreases gradually, eventually becoming zero at the cylindrical asymptotic limit. The power-law solutions (with $`v_\mathrm{R}=B_\mathrm{R}=0`$) we adopted for many specific cases represent the asymptotic limit of each streamline.
Ostriker (1997) presented self-similar steady solutions for disk winds with cylindrical asymptotics and gave the asymptotic fluid and Alfv$`\stackrel{´}{\mathrm{e}}`$n speeds and the location of the asymptotic streamlines, characterized by $`q`$ together with $`R_\mathrm{A}/R_1`$ or $`R_0/R_1`$, where $`R_0,R_\mathrm{A}`$, and $`R_1`$ denote the radii of the footpoint, the Alfv$`\stackrel{´}{\mathrm{e}}`$n surface, and the asymptote of each given streamline, respectively. Typical numerical values for those solutions are $`\mathrm{\Omega }=0.2\mathrm{\Omega }_0`$, $`v_{\mathrm{A}\varphi }/R=0.42\mathrm{\Omega }_0`$, and $`v_{\mathrm{A}\varphi }/v_\mathrm{z}=6`$ for $`q=0.5`$, and $`\mathrm{\Omega }=0.1\mathrm{\Omega }_0`$ and $`v_{\mathrm{A}\varphi }/R=0.45\mathrm{\Omega }_0`$ for $`q=0.9`$, where $`\mathrm{\Omega }_0`$ is the Keplerian rotation rate at the streamline footpoint.
Using these values we can estimate the growth times of the global fundamental mode and the fastest growing (with $`k_\mathrm{R}`$ near 0) toroidal buoyancy modes. The foregoing analysis suggests that these disturbances will play the most significant dynamical role, given the ineffectiveness of MRIs in cold, $`B_\varphi `$-dominated flows. We define the time to grow by $`\mathrm{\Gamma }`$ orders of magnitude as $`t_\mathrm{\Gamma }`$. For FM, $`t_{\mathrm{\Gamma },\mathrm{FM}}=\mathrm{\Gamma }/(|\omega |\mathrm{log}e)`$, so from Fig. 2 and eq. (25) we have for $`q=0.9`$
$$t_{\mathrm{\Gamma },\mathrm{FM}}\frac{14\mathrm{\Gamma }}{\mathrm{\Omega }_{0,\mathrm{i}}}=25\mathrm{\Gamma }\mathrm{days}\left(\frac{M}{M_{}}\right)^{1/2}\left(\frac{R_{0,\mathrm{i}}}{0.1\mathrm{AU}}\right)^{3/2}.$$
For ATB and NTB, from eqs. (45) and (52), we have for $`i0`$ and $`q=0.5`$
$$t_{\mathrm{\Gamma },\mathrm{ATB}}\frac{10^\mathrm{\Gamma }}{0.21\mathrm{\Omega }_0}=24\times 10^\mathrm{\Gamma }\mathrm{yrs}\left(\frac{M}{M_{}}\right)^{1/2}\left(\frac{R_0}{10\mathrm{AU}}\right)^{3/2}\mathrm{and}$$
$$t_{\mathrm{\Gamma },\mathrm{NTB}}\frac{10^{1.4\mathrm{\Gamma }}}{3\mathrm{\Omega }_0}=1.7\times 10^{1.4\mathrm{\Gamma }}\mathrm{yrs}\left(\frac{M}{M_{}}\right)^{1/2}\left(\frac{R_0}{10\mathrm{AU}}\right)^{3/2},$$
respectively. Here, $`M`$ is the mass of the central star and $`\mathrm{\Omega }_{0,\mathrm{i}}`$ is the angular speed of a disk at the footpoint $`R_{0,\mathrm{i}}`$ of the innermost streamline of winds. The fact that the growth of the FM by a factor $`10^\mathrm{\Gamma }`$ occurs within $`\mathrm{\Gamma }`$ times the rotation period of the disk at the inner radius, far shorter than the lifetime of winds ($`10^410^5`$ yrs), suggests that the FM mode is dynamically important in the evolution of the disk winds. When $`q`$ is small, the radial turbulent mixing of the wind, caused by both axisymmetric and non-axisymmetric toroidal buoyancy modes over a relatively short time, is likely to cascade down into arbitrarily smaller scales to dissipate when the microscopic processes such as magnetic reconnection are included. The released energy in the dissipation processes may heat up the flow, potentially making a significant contribution to the heating of protostellar winds and jets. Because the growth rates of buoyancy modes are proportional to the equilibrium magnetic forces (cf. eqs. and ), winds that have approached a force-free magnetic configuration will not be subject to the ATB and NTB instabilities.
The global fundamental mode affects only the inner region of the disk winds (e.g., the central tenth for the model shown in Fig. 3). The logarithmic density gradient $`\mathrm{ln}\rho /\mathrm{ln}R`$ changes relative to the equilibrium value by $`(\rho _1/\rho _0)/\mathrm{ln}R`$. Fig. 3 shows that as a consequence of the fundamental mode, the very central region becomes more steeply stratified, a surrounding concentric region less steeply stratified, and the balance (most of the wind) remains nearly unchanged. Thus, the FM tends to enhance jetlike structure in the central parts of winds. In addition, because of their tendency to compress interior gas via the FM mechanism, disk winds may help to collimate any interior flows into narrow, fast jets, even when the disk winds themselves have relatively slow motion (cf. Ostriker, 1997).
What do the present results imply about the likely radial extent of protostellar winds? First, we note that observed optical jets are unlikely to be isolated structures, because if so they would be significantly overpressured relative to the ambient medium: Since magnetocentrifugal jet models typically predict internal Alfv$`\stackrel{´}{\mathrm{e}}`$n speeds comparable to their flow speeds in the range $`150400`$ km s<sup>-1</sup>, they have strong internal magnetic pressure $`P_{\mathrm{wind}}=B_\varphi ^2/8\pi \rho _\mathrm{w}v_\mathrm{A}^2/22.8\times 10^7`$ ergs cm<sup>-3</sup>, which is about 6 orders of magnitude greater than the gas pressure of ambient medium, $`P_{\mathrm{ext}}=c_\mathrm{s}^2\rho _{\mathrm{ext}}1.3\times 10^{13}`$ ergs cm<sup>-3</sup>. Here as reference values we adopted $`\rho _\mathrm{w}=350m_\mathrm{H}`$ cm<sup>-3</sup>, $`\rho _{\mathrm{ext}}=200m_\mathrm{H}`$ cm<sup>-3</sup>, $`v_\mathrm{A}=310`$ km s<sup>-1</sup>, and $`c_\mathrm{s}=0.2`$ km s<sup>-1</sup> with $`m_\mathrm{H}`$ being the mass of a hydrogen atom (Hartigan et al., 1999). Thus the pressure imbalance at the outer boundary of the jet would cause either the wind as a whole or only its surface layer to expand until a new balance is attained. Based on the results of this paper, if the magnetic field at the base of the wind is stratified less steeply than $`R^1`$, then perturbations of the outer parts of the wind are stable. As a consequence, only the surface layers of such winds would expand in order to achieve a pressure-balanced condition with the ambient medium. If, on the other hand, the wind’s magnetic field is stratified more steeply than $`R^1`$ at its base, no equilibrium is even possible; the wind would expand as a whole to fill the entire $`4\pi `$ steradians, with the inner parts having higher density observable as a narrow optical jet (cf. Shu et al., 1994; Shang et al., 1998). Numerical simulations presently underway (Lee et al., 2000) support previous work indicating that protostellar winds with a wide-angle component are better able to produce observed molecular outflow structures than purely jetlike winds (see also Li & Shu, 1996; Ostriker, 1997, 1998; Matzner & McKee, 1999) but further studies are required to determine just how distributed in angle the wind momentum should be - i.e., to discriminate between “fully-expanded” and “surface-expanded” models. Recent observations (see, e.g., Richer et al., 2000) showing a correlation in molecular outflow kinematics with age - with extremely high velocity, highly-collimated flows seen only in the youngest sources - may indicate an underlying temporal evolution from more-collimated to more-expanded protostellar winds.
We acknowledge a stimulating report from an anonymous referee, and helpful comments from N. Turner and S. Balbus.
|
warning/0004/cond-mat0004104.html
|
ar5iv
|
text
|
# Periodic Ground States in the Neutral Falicov-Kimball Model in Two Dimensions
## Appendix
In this appendix we derive eq. (4) which expresses the sixth and eighth order terms in the Hamiltonian in terms of the number of ions of types A,B and C. Recall that the Hamiltonian at a given order is given in terms of walks by
$$H_{2m}=\underset{\gamma :|\gamma |=2m}{}w(\gamma )$$
(20)
This may be rewritten as
$$H_{2m}=\underset{X}{}c_{m,X}S(X)$$
(21)
where the sum is over subsets of the lattice and $`S(X)`$ is $`1`$ if the configuration $`S`$ has an ion at every site in $`X`$ and 0 otherwise.
Recall that $`I(S)`$ denotes the sets of sites at which there is an ion for the configuration $`S`$. If $`S(X)`$ is to give a non zero contribution to $`H_6(S)+H_8(S)`$, $`X`$ must be a subset of $`I(S)`$ which is contained in a closed nearest neighbor walk of six or eight steps. Let $`S`$ correspond to a tiling by squares and parallelograms. Then the only subsets $`XI(S)`$ which can contribute at orders 6 and 8 are those shown, up to rotations and reflections, in figure 12.
We will refer to an ion and the eight ions arranged about it as in figure 4 as the “local arrangement” of the ion. Recall that the local arrangement about an ion in $`I(S)`$ must look like one of the three shown in figure 4. So for any $`X`$ which gives a nonzero contribution to $`H_6(S)+H_8(S)`$, there is at least one ion $`x`$ such that $`X`$ is a subset of the local arrangement about $`x`$. There may be more than one such $`x`$. For example, if $`X`$ consists of a single ion, $`X`$ will be contained in 9 local arrangements. To compensate for this over counting, we define $`\stackrel{~}{c}_{m,X}=\frac{c_{m,X}}{\kappa (X)}`$ where $`\kappa (X)`$ is the number of ions whose local arrangement contains $`X`$. Now fix a type A ion and let $`L`$ denote the local arrangement about it. Then for even $`m8`$ we have
$$e_m^A=\underset{XL}{}\stackrel{~}{c}_{m,X}$$
This may be written as
$$e_m^A=\underset{\stackrel{Xfigure\text{12}:}{XL}}{}\tau _{A,X}\stackrel{~}{c}_{m,X}$$
where “$`Xfigure\text{12}`$” means that $`X`$ is a rotation or reflection of one of the arrangements from figure 12, and $`\tau _{A,X}`$ is the number of translates of $`X`$ appearing in the local configuration about a type A ion. $`e_m^B`$ and $`e_m^C`$ are given by the same formula with $`L`$ taken to be the local arrangement of a type B or C ion respectively.
Table 1 gives the coeffecients $`c_{6,X}`$ and $`c_{8,X}`$ for each $`X`$ in figure 12. The last three columns give the numbers $`\frac{\tau _{A,X}}{\kappa (X)}`$, $`\frac{\tau _{B,X}}{\kappa (X)}`$, and $`\frac{\tau _{C,X}}{\kappa (X)}`$. So we see, for example, that $`e_6^A=64\times 1384\times 2+96\times 2674\times 4=3200`$.
We should note that one can use the above type of argument to show that an equation of the form (4) must hold. The values of $`e_m^A,e_m^B`$ and $`e_m^C`$ can then be computed by carrying out the perturbation theory for $`H_m(S)`$ for several different periodic configurations $`S`$ and then using the results to set up a linear system of equations for $`e_m^A,e_m^B`$ and $`e_m^C`$. We have done this as an independent check on the calculation above.
Acknowledgements: This work was supported by NSF grant DMS-9623509. It was part of Karl Haller’s Ph.D. dissertation .
|
warning/0004/astro-ph0004214.html
|
ar5iv
|
text
|
# Central cusp due to a super-massive black hole in axisymmetric models of elliptical galaxies
## 1 Introduction
HST observations have shown that a central density cusp is present in most, if not all, elliptical galaxies (Lauer et al. 1995, and companion papers). Furthermore the slope of the density cusp has shown a dichotomy between high mass (luminosity) systems and low mass (luminosity) ones, analogous to what has been found for other observational properties (Bender et al. 1989). Low mass ellipticals tend to host steep cusps, the radial profile of their luminosity $`\rho (r)`$ having a logarithmic slope $`\gamma \mathrm{d}log(\rho )/\mathrm{d}log(r)1.9\pm 0.5`$. On the other hand, luminous ellipticals have shallower cusps, with $`\gamma 0.8\pm 0.5`$ (Gebhard et al. 1996).
The most popular explanation for the formation of a density cusp is a central super-massive black hole (BH)– see Richstone et al. (1998), Kormendy & Richstone (1995) for reviews. Such BHs are believed to be present in a large fraction (possibly close to 1, Haenelt & Rees 1993, Tremaine 1997) of present day galaxies. Their influence on a collisionless galactic nucleus was first addressed by Peebles (1972), Young (1980), Goodman & Binney (1984) using semi-analytic models.
All these models suppose an isolated galaxy, at the center of which a BH grows by gas accretion. The slope of the cusp produced in such models is within the range observed for the less massive E galaxies. In this paper we shall also consider the formation of a cusp in an isolated galaxy, as we are interested in the origin for the steep cusp ($`\gamma 1.9`$) of low luminosity ellipticals. On the other hand, numerical simulations (Makino & Ebisuzaki, 1996, Quinlan & Hernquist 1997, Nakano & Makino 1999) have shown that a BH sinking within the core of an elliptical galaxy, or a binary of BHs produce a shallow cusp, similar to that observed for massive E galaxies. Such a BH binary could result from a merger. This picture is therefore compatible with the general belief (e.g. Nieto & Bender 1989) that the apparent existence of two classes of ellipticals is related to a more important role played by merging and interactions in the history of massive E galaxies.
All the semi-analytic models of cusp formation due to a central BH use the adiabatic invariance of actions to derive the distribution function after a BH has slowly grown (see Young 1980). Since actions are explicitly known for spherical systems, analytical models have been derived in spherical symmetry. Based on such quasi-analytic computations for spherical models, Quinlan et al. (1995) investigated the consequences of the initial density profile on the slope of the cusp, and addressed the influence of a velocity anisotropy.
For non-spherical systems – except for Stäckel potentials – an expression for the actions is unknown, and therefore models based on the conservation of actions have not been computed hitherto. In this paper, we propose to use numerical simulations with high central resolution, based on the PP method (Leeuwin et al. 1993), to investigate the cusp formed in axisymmetric systems by the growth of a BH.
Many low luminosity galaxies display little evidence for triaxiality, but are close to axisymmetry; they can be modeled as two-integral models, with $`f(E,L_z)`$. Such a distribution function (d.f.) implies velocity dispersions obeying $`\sigma _R=\sigma _z`$, meaning that flattening would be due to an excess of azimuthal motion (which increases the net rotation, or the tangential anisotropy). This does not preclude that three-integral models would not do as well, or better (e.g. review by Merritt, 1998). Many of those galaxies however exhibit kinematic features consistent with isotropic rotators models: at least for them, an excess of tangential motion is plausibly the main support for their flattening.
The BH itself may have perturbed orbits sufficiently to erase the part of memory for initial conditions that corresponds to conservation of a third integral (Norman et al. 1985, Gerhard & Binney 1985, Merritt & Quinlan 1998). In this case axisymmetry is a consequence of the BH growth. Nevertheless, one still ought to investigate the scenario in which, for some of these galaxies, the presently observed axisymmetry already existed before the BH growth.
This is the goal of this paper, where we will study the cusp generated by the growth of a central BH within axisymmetric systems having various degrees of rotation. If a BH grows in a roughly axisymmetric galaxy, do the initial flattening and rotation have a sizeable effect on the evolution, so that it deviates from that of the well known spherical case? Could we in certain cases be able, eventually with more detailed observational data, to infer from the properties of presently observed cusps any information about the pre-black hole galaxy? Should one be cautious in certain cases about deriving the BH mass using the spherical adiabatic model?
This paper is organized as follows: first, the existing models are briefly recalled in section 2. We then specify the initial conditions (section 3) and the numerical techniques used for this work (section 4). Results for non-rotating and rotating models are given in sections 5 and 6, respectively. We conclude in section 7.
## 2 Previous models for galactic cusps
Previous models are based on a few basic assumptions, and ours will follow suit. The growth of the BH is supposed to be due to gas accretion, and thus does not deplete the original stellar component. The existence of the gaseous component is otherwise neglected, so that the models follow the evolution of a purely stellar component. The growth time $`t_{BH}`$ of the black hole is taken to be at maximum $`10^7`$ Yrs. This is long enough with respect to the stellar dynamical times, so that we can compare our models with the adiabatic computations, but not too long for obvious computation time reasons. It is furthermore reasonnable if compared to the ‘Salpeter time’, or timescale for the mass accretion onto a black hole:
$$t_S=M_{BH}/\dot{M}_{BH}ϵ\times 4\times 10^8Yrs.$$
(1)
In the above, $`ϵ`$ is the radiative efficiency, which is most often $`ϵ0.1`$ in current models of accretion disks (cf. Richstone et al. 1998), or in estimates derived from quasars number counts (van der Marel 1997).
Furthermore the models neglect collisions. It has been estimated in the past that only some of the spiral galaxies bulges are dense enough to experience significant two-body interactions (see e.g. Kormendy 1988), although these estimates were based on the belief that elliptical galaxies had cores. Let us suppose for instance that up to $`0.11\%`$ of an E galaxy mass, corresponding to say $`N10^610^9`$ stars, is enclosed in the central $`10`$ pc, where velocity is of order a few times $`100`$ km/s. The crossing time in these regions should be $`T_{cr}10^5\left(\frac{R(\mathrm{pc})}{10}\right)\left(\frac{100}{\sigma (\mathrm{km}/\mathrm{s})}\right)`$ Yrs. Therefore the relaxation time would be (e.g. Binney & Tremaine 1987): $`t_{rel}N/(8\mathrm{l}\mathrm{n}N)t_{cr}10^9\left(\frac{R(\mathrm{pc})}{10}\right)\left(\frac{100}{\sigma (\mathrm{km}/\mathrm{s})}\right)\left(\frac{N}{10^6}\right)`$ Yrs. Stellar collisions are liable to play some role over the age of the universe in such a system with an already highly concentrated density. For the models we consider here, however, a collisionless model is justified, given that the initial galaxy has a large core, and that the BH grows in $`10^7`$ Yrs. Nevertheless it should be kept in mind that collisional processes could play a role in the evolution of the center-most part after the cusp has developed.
The spatial extent of the cusp is given approximately by the influence radius $`r_I`$ of the BH, which is the distance up to which the BH potential dominates. This is estimated as:
$$\frac{M_{BH}}{r_I}\sigma _0^2,$$
(2)
where $`\sigma _0`$ is the central velocity dispersion, and $`M_{BH}`$ is the BH mass.
### 2.1 Analytical models: spherical adiabatic models
Analytical models assume that the central BH grows over time scales that are long compared to dynamical times in the central region. The final d.f. can then be deduced from adiabatic conservation of actions (Peebles 1972, Young 1980).
In an isotropic system having initially a core, the cusp is predicted to have a slope $`\gamma =3/2`$ (Young 1980; see also Tremaine 1997). This value, however, holds if the BH mass is less than roughly the core mass; when larger, the self-gravity of the cusp becomes important and the slope increases up to values $`2`$ (e.g. Young 1980). For centres that are initially ‘non-analytical’ – i.e. whose density profile has a non-zero derivative at the origin– Quinlan et al. (1995) derive larger slopes $`\gamma `$, with higher values for stronger initial cusps.
Goodman & Binney (1984) provide the following simple picture for the way orbits are transformed as the cusp grows. The fact that radial action is conserved plays an especially important role in the shape evolution of elongated orbits. It means that the radial excursion $`r_{max}r_{min}`$ does not change much, but the potential center, which is originally at the orbit center (roughly harmonic potential), is finally at one of its foci (keplerian potential). Thus for elongated orbits eccentricity has decreased, which means also that the motion is more circularly biased. The adiabatic analytical models already cited predict indeed a moderate tangential anisotropy. This trend for rounder orbits is also present in numerical results by Norman et al. (1985), who studied how the shape of orbits close to a growing point mass is modified.
Initial anisotropies in the stellar velocities appear to have a weaker influence than an initial cusp (Quinlan et al. 1995). Due, however, to the small number of available analytical anisotropic models, this problem could not be thoroughly examined. One of the few anisotropic cases for which the cusp slope has been semi-analytically evaluated is a spherical model fully made of circular orbits. From conservation of energy and angular momentum, the final logarithmic slope is found to be $`\gamma =9/4`$ (Quinlan et al. 1995 <sup>1</sup><sup>1</sup>1 using their notations in which initially $`\rho (r)r^C`$, the value for the slope given in their Eq. (17) should actually be $`C=3\frac{3}{4\gamma }`$ ), significantly higher than for the isotropic case ($`\gamma =3/2`$). Therefore we could expect that in a model with some intermediate degree of tangentially biased motion, $`3/2<\gamma <9/4`$. Since our axisymmetric models are flattened in their centre by an excess of tangential motion, it is interesting to ask whether such an anisotropy will have a sizeable influence on the cusp slope. We may also ask whether a systematic rotation will bring any changes. These are some of the questions we will try to answer in this paper.
For the very few galaxies which had been observed ten years ago with sufficient central resolution, observations showed rotation velocities rising towards the center. This prompted attempts to compute the amount of angular momentum brought to the center by the forming cusp (Lee & Goodman 1989, hereafter LG89 ; see also Cipollina & Bertin 1994, hereafter CB94). These works, however, had to assume a spherical potential in order to derive analytical solutions. One aim of this paper is to address this problem anew without any such approximation. In sections 5 & 6, we investigate the influence of initial rotation for self-consistent axisymmetric models. Our motivations do not follow those of LG89, because higher quality, more recent observations no longer show evidence for more rotation than dispersion in the centre (see e.g. van der Marel et al. 1997, Joseph et al. 2000 for HST observations of M32, and Kormendy et al. 1998, regarding NGC 3377).
### 2.2 Numerical methods
The pioneering simulation by Norman et al. (1985) did not search for any detailed feature in the center, but was meant instead to measure the shape changes caused at moderate and large radii by a central growing BH. Due to the limited numerical capacity available at the time, their results displayed substantial particle noise, but showed already the trend towards axisymmetrization. Merritt & Quinlan (1998) recompute this problem with a considerably improved accuracy, obtained by using a larger number of particles ($`N1.1\times 10^5`$ instead of $`N=2\times 10^4`$ for Norman et al. 1985), and by enhancing the central particle density. To this end, they replace each strongly bound particle by a number of less massive particles distributed over its orbit (see §4.4).
Sigurdsson et al. (1995) built a numerical code to follow the growth of the cusp itself, and tested it for the spherical case. Their code makes use of the self-consistent field method (Hernquist & Ostriker 1992). This is formulated as a perturbation method, in the sense that it uses a family of density-potential functions, of which the zero-order term corresponds, or is close, to the initial model. At each time step during the simulation, the coefficients of the expansion for the density yield the new potential. The less the system evolves, the fewer are the terms required in the expansion to provide a given level of accuracy. In the implementation by Sigurdsson et al. (1995), the number of particles in the central parts is enhanced by using a mass spectrum for the particles. This is achieved by considering for the particles distribution, instead of the initial d.f. $`f_0(E)`$, a distribution $`f_0(E)/m(E,L)`$, such that the mass $`m`$ of a particle is less than unity if the pericenter of its orbit is smaller than some radius, and is unity if it is larger. This way, orbits reaching close to the centre are represented by more numerous, lighter particles. A numerical simulation based on this code, and using $`512,000`$ particles, has been performed by van der Marel et al. (1997), in order to check the stability of their dynamical model for M32, i.e. an axisymmetric model including a central supermassive BH.
In the present paper, we use a numerical perturbation technique which allows to concentrate particles at the center of the model, without entailing at the same time high particle-noise for the potential at larger radii (§4). In addition, we perform a subdivision of the central particles when necessary, to further increase the resolution at the center. These numerical simulations are used here in order to extend the models of BH induced cusps to non-spherical systems.
## 3 The initial axisymmetric model
### 3.1 Axisymmetric, non-rotating model
Fully analytical axisymmetric models for elliptical galaxies are scarce (Lynden-Bell 1962, Dejonghe 1986, Evans 1994, cf. discussion in Hunter & Qian 1993). Most of them are sums –often infinite sums – of Fricke terms. We take here the simple model by Lynden-Bell (1962), which uses only two Fricke terms:
$`f_0(E,L_z)=`$ $`F_1E^{7/2}+F_2L_z^2E^{13/2}`$ $`\mathrm{if}E>0`$ (3)
$`=`$ $`0`$ $`\mathrm{otherwise}`$
where $`E`$ is the binding energy $`E=\mathrm{\Phi }_0v^2/2`$ and $`L_z`$ is the component of angular momentum about the symmetry axis. This distribution corresponds to a flattened Plummer potential:
$$\mathrm{\Phi }_0(R,z)=\frac{M_0}{((R^2+z^2+a_P^2)^22b_P^2R^2)^{1/4}},$$
(4)
where we have used cylindrical coordinates. In the above, $`M_0`$ is the total mass of the model, $`a_P`$ and $`b_P`$ are two parameters, and $`F_1`$ and $`F_2`$ are normalizing numerical constants, which expression is given by Lynden-Bell (1962):
$$\begin{array}{cc}F_1\hfill & =\frac{\sqrt{2}}{4\pi ^{3/2}}\frac{5!}{7/2!}\frac{3a_P^22b_P^2}{M^4}\hfill \\ F_2\hfill & =\frac{\sqrt{2}}{4\pi ^{3/2}}\frac{9!}{13/2!}\frac{5b_P^2(2a_P^2b_P^2)}{M^8}\hfill \end{array}$$
We take in the following $`a_P=4`$ and $`M_0=1`$ in the units of the simulation. These are such that the mass unit is $`10^{11}M_{}`$, the length unit is $`1`$ kpc, and $`G=1`$.
The core radius is usually defined as the radius at which the surface brightness $`\mathrm{\Sigma }(R,z)`$ has decreased to half its central value $`\mathrm{\Sigma }_0`$. This corresponds for the present model to $`R_{1/2}0.76a_P`$. The mass enclosed within the $`\mathrm{\Sigma }=cst`$ isodensity level passing at $`(R_{1/2},z=0)`$ is $`M_{core}0.2M_0`$.
The flattening decreases with radius, so that the models tend towards spherical symmetry at large radii. The larger $`b_P`$, the larger the central flattening. $`b_p`$ however must be smaller than $`b_{Pmax}0.64a_p`$ otherwise the isodensity contours have a dimple on the z-axis. We shall consider a model close to the model with maximum flattening, by taking $`b_P=0.55a_P`$. For this model the axis ratio is about constant, with value $`c/a0.55`$ (Fig. 1) within the region where the cusp will develop (see above the definition of the influence radius),
Since the d.f. depends only on $`E`$ and $`L_z`$, the velocity dispersions obey $`\sigma _R=\sigma _z`$. The flattening is sustained by an excess of motion in the $`\varphi `$ direction, with respect to motion in the $`R`$ or $`z`$ directions. If velocity fields that are moderately biased towards tangential motion can affect the final density cusp, we should witness it in our computations.
The distribution function in Eq. 3.1 is even in $`L_z`$ and corresponds to a non-rotating system. In this case (cf. Lynden-Bell 1962):
$`\sigma _R^2=`$ $`{\displaystyle \frac{\mathrm{\Phi }_0}{6}}\left(10.2{\displaystyle \frac{CR^2\mathrm{\Phi }_0^9}{\rho _0}}\right)=\sigma _z^2`$ (5)
$`\sigma _\varphi ^2=`$ $`{\displaystyle \frac{\varphi _0}{6}}\left(1+0.4{\displaystyle \frac{CR^2\mathrm{\Phi }_0^9}{\rho _0}}\right)`$ $`,`$ (6)
with $`C=5b_P^2(2a_P^2b_P^2)/(4\pi M_0^8)`$.
Fig. 1 displays the isocontours for the density, as well as the profile along $`R`$ for the velocity dispersions along each direction.
### 3.2 Rotating models
Models with positive net rotation are obtained by changing the sign of $`L_z`$ for a fraction of the orbits having negative $`L_z`$. If $`\eta `$ is the fraction of orbits being reversed, we obtain:
$$f^r(E,L_z)=\{\genfrac{}{}{0pt}{}{f_0(E,L_z)(1\eta ),L_z<0}{f_0(E,L_z)(1+\eta ),L_z>0}$$
(7)
In order however to avoid a discontinuity at $`L_z=0`$, the distribution function must have $`f(E,L_z)=f(E,L_z)`$ for $`L_z0`$. We therefore take for initial model:
$$f_0^r(E,L_z)f_0(E,L_z)\times (1+\eta p(L_z)),$$
(8)
where $`p(L_z)`$ is some function having value $`\pm 1`$, except for a small range around the origin, where it goes to zero. A simple choice is:
$$p(L_z)=\{\begin{array}{cc}\mathrm{sin}\left(\frac{\pi }{2}\frac{L_z}{L_m}\right),\hfill & \hfill |L_z|<L_m\\ \mathrm{sign}(L_z),\hfill & \hfill \mathrm{otherwise}\end{array}$$
(9)
Here $`L_m`$ is a parameter which determines the extent of the $`L_z`$ region where rotation is smaller than $`\eta `$. Its choice is of special interest for what follows.
The simplest choice for $`L_m`$ is a constant, independent of energy (cf. CB94). However, by taking some function $`L_m(E)`$, we can build models where, for any given energy, a constant fraction of orbits contribute to rotation. This is similar to the case studied by LG89. We investigate both types of models, as we know from comparing these two papers that they yield different results.
We thus consider two kinds of initial distributions, that use respectively:
$`p_1(L_z)=`$ $`p(L_z)`$ $`\mathrm{with}L_m=L_0`$ (10)
$`p_2(L_z,E)=`$ $`p(L_z;E)`$ $`\mathrm{with}L_m=xL_c(E),0<x<1`$ (11)
In the above definition, $`L_c(E)`$ is the angular momentum for the circular orbit with energy $`E`$, whereas $`L_0`$ is some constant, which is conveniently expressed in units of $`l_0=a_P\sqrt{\mathrm{\Phi }_0(0)}`$.
The mean tangential velocity is given by:
$$v_\varphi (R,z)=\frac{d^3vf_0^r(E,L_z)(L_z/R)}{d^3vf_0^r(E,L_z)}.$$
(12)
The effective rotation is measured by:
$$\eta _e=\frac{d^3rd^3vf(E,L_z)L_z}{d^3rd^3vf(E,L_Z)|Lz|},$$
(13)
which is in practice slightly smaller than $`\eta `$ in the definition of $`f_0^r`$ given by Eq. (8) (and exactly $`\eta `$ in the limit $`L_m0`$).
## 4 Numerical method
A crucial feature for any $`N`$-body code aimed at studying the density cusp is of course the central resolution. Most particles should ideally orbit within a few times the ‘influence radius’ $`r_I`$ of the BH (Eq. 2). In this paper, we use a scheme which is formulated as a perturbation method, and allows to choose freely the sampling in phase-space. The main advantage of our scheme is that even if we have few particles outside $`r_I`$, where evolution is weak, the potential is fairly well represented (since it is dominated by the analytical $`\mathrm{\Phi }_0`$ term, which has zero particle-noise).
### 4.1 The ‘perturbation particles’ scheme
The PP scheme has been described in detail elsewhere (Leeuwin et al. 1993). In this scheme, one writes the collisionless Boltzmann equation (CBE) as a perturbation problem. The unperturbed state corresponds to the initial analytical model, while the perturbation corresponds to the evolution of the system. Thus, the system d.f. is formally written as $`f=f_0+f_1(t)`$, where $`f_1(t)`$ is found by integrating explicitly the CBE along orbits in an N-body realization. The choice of the orbits to be integrated during the run –that is to say the initial distribution of particles $`f_S(𝐫,𝐯)`$– is in principle arbitrary, but one should aim at a good sampling of phase-space regions where large perturbations are expected. The function $`f_S`$ is chosen to be a function of the system classical integrals (here $`E,L_z`$), so that in the absence of time evolution the sampling of phase-space is stationary.
If the system potential varies with time, $`f_S`$ can not be made a stationary function. However, both $`f_S`$ and $`f`$ obey the CBE equation. Therefore, noting the phase-space coordinate $`w(𝐫,𝐯)`$, we may simply record the initial value $`f_S(w(t_0))`$ for each particle, since along each orbit $`f_S(t)=f_S(t_0)`$. The mass attributed to the particle running on that orbit is weighted according to the local phase-space density:
$$m_1(t)=\frac{f_1(w(t))}{f_S(w(t_0))},$$
(14)
as explained in Leeuwin et al. (1993) and Wachlin et al. (1993).
On the other hand, the CBE is explicitly solved for $`f_1`$: the accuracy of orbits integration can then be monitored through the relative variations of $`f`$. The perturbation along an orbit is given by:
$$\frac{\mathrm{d}f_1(t)}{\mathrm{d}t}=[f_0,\mathrm{\Phi }_{BH}+\mathrm{\Phi }_{s.g.}]$$
(15)
where $`\mathrm{\Phi }_{BH}=M_{BH}/r`$, and $`\mathrm{\Phi }_{s.g.}`$ is the self- gravitating potential of the cusp.
The computation of moments for $`f_1`$ amounts in fact to a Monte-Carlo integral with sampling $`f_S`$. For instance, the perturbed total mass is evaluated as:
$$M_1f_1(w)d^6w\mathrm{\Sigma }_{i=1}^N\frac{f_1(w_i)}{f_S(w_i)}\mathrm{\Sigma }_{i=1}^Nm_{1i}.$$
(16)
To compute local values of the density field, or of any other field, the sum over the particles is of course limited to a small region of space where the quantity of interest is roughly constant. Moments of the total distribution $`f=f_0+f_1`$ are given by:
$$<V^n>=\frac{(f_0+f_1)V^nd^6w}{(f_0+f_1)d^6w},$$
(17)
where $`V`$ represents any component of the velocity vector. This is estimated as:
$`<V^n>(𝐫)`$ $`=`$ $`{\displaystyle \frac{f_0V^nd^6w+f_1V^nd^6w}{\rho _0(𝐫)+\rho _1(𝐫)}}`$ (18)
$``$ $`{\displaystyle \frac{\rho _0<V^n>_0+\mathrm{\Sigma }_i\frac{f_1(w_i)}{f_S(w_i)}V_i^n}{\rho _0+\rho _1}}`$
$`=`$ $`{\displaystyle \frac{\rho _0<V^n>_0+\mathrm{\Sigma }_im_{1i}V_i^n}{\rho _0+\rho _1}}`$
The sampling distribution $`f_S`$ is most efficient for the evaluation of the above Monte-Carlo integrals, when it is most similar to the integrants. The integration accuracy can be checked by monitoring the total mass in the perturbation, which should remain equal to zero. A convenient measure for the associated error is $`M_1/M_1`$, where we note $`M_1\mathrm{\Sigma }_i|m_{1i}|`$.
The self-gravity of the cusp corresponds to the force due to the perturbation mass. Thus, a particle $`j`$ will be subject to a self-consistent force term:
$$\mathrm{\Phi }_{s.g.}(𝐫_j)\mathrm{\Sigma }_{ij}\frac{m_{1i}}{|𝐫_i𝐫_j|^3}(𝐫_i𝐫_j).$$
(19)
### 4.2 Orbit integration
Close to the BH, the force gradients are of course very large. Some orbits therefore demand extremely short time-steps during small time intervals. For such problems, adaptative time-steps is the most convenient scheme. In a 1D simulation (Leeuwin & Dejonghe 1998) that we ran to check whether the PP method could be advantageous for the present problem, we have experimented block time-steps integration (using either a leap-frog, or a RK4 scheme). In a block time-step scheme, the particles are sorted into groups at regular time intervals, and each group advanced with its smallest time-step until the next sorting out. This makes the scheme a costly one, since a set of particles may be advanced with a needlessly tiny timestep.
Symplectic methods (Wisdom & Holman 1991, Saha & Tremaine 1992), because they conserve phase-space volume, would in principle be useful to guarantee conservation of our sampling phase-space density (Binney, private communication). Unfortunately, symplectic integrators with adaptive time-steps do not yet exist (see Duncan, Levison & Lee 1998 for a version with block time-steps).
We use in this work the ODEINT routine due to Press et al. (1986), which is an adaptative individual time-steps scheme, based on a 4th-order Runge-Kutta time interpolation. The conservation of $`f`$ along orbits is better than $`10^4`$ over each entire simulation.
### 4.3 Force computation with GRAPE machines
The BH mass was updated at each time-step. This is straightforward since the time dependence is explicit. Moreover, within the BH influence radius where the BH force term dominates, a precise evaluation is required.
The self-gravitating force (Eq. 19) is computed by direct summation of the individual particles contributions, using a GRAPE machine. A small modification of the standard software was necessary since the PP masses can be either positive or negative, a possibility not foreseen for GRAPE . Furthermore, timesteps within the cusp are very small, so that the cost of re-evaluating by direct summation the inter-particles force at each time-step for the most bound orbits would be prohibitive. Since the BH grows very slowly ($`10^7`$ Yrs) compared to the central dynamical times ($`10^5`$ Yrs, see §2.1), the force field due to the cusp also changes slowly with respect to individual crossing times. Moreover, it is negligible at large radii where dynamical times are larger (see Fig. 6). Therefore the self-consistent force field was evaluated using GRAPE only at regular time intervals $`\mathrm{\Delta }t`$. To derive the self-gravity force experienced by each particle at intermediate times, we proceed as follows:
* The self-consistent forces are evaluated using GRAPE at regular time intervals $`\mathrm{\Delta }t`$. This yields the three components $`(F_{xi},F_{yi},F_{zi})`$ of the force for each particle $`i`$. We infer for each particle the two components $`(F_{Ri},F_{zi})`$ along $`R`$ and $`z`$ in cylindrical coordinates.
* Using a cell-in-cloud (CIC, cf. Hockney & Eastwood 1981) scheme in polar coordinates, we compute the force field $`(F_R,F_z)`$ on a grid in $`R,z`$ (we average over the azimuthal angle $`\varphi `$).
* At each time-step within $`\mathrm{\Delta }t`$, the force $`(F_R,F_z)(r_i(t))`$ at the current position of each particle is interpolated from the grid values (with again a CIC).
The grid has $`20\times 20`$ cells, and extends in both directions from $`x_1=r_I/200`$ to $`x_{20}=4a_p`$. Points are spaced with a logarithmic increment:
$$x_{i+1}=x_i+dx_ik^a,$$
(20)
where $`a`$ and $`k`$ are two real parameters numerically determined after the choice of $`x_1,dx_1`$ and $`x_{20}`$. We have taken $`dx_1=x_1`$.
For the force evaluation with GRAPE we use a softening length $`ϵ=10^4a_P`$. The force is computed $`1000`$ times during the BH growth. At each time-step, the value for the self-gravity force is linearly interpolated from the grid values, using again a CIC scheme.
A run using $`3\times 10^5`$ particles, and the parameters for the standard case defined hereafter (§5.1), takes approximately $`2`$ days on the Marseille 5 board GRAPE-3AF system, and somewhat longer on the GRAPE-4 system with 41 chips (see Athanassoula et al. 1998 for the characteristics of the Marseille GRAPE -3AF system, and Makino et al. 1997, Kawai et al. 1997 for the characteristics of the GRAPE -4 system). For $`N=300,000`$ particles, one force evaluation by GRAPE -3AF takes approximately $`140`$ s, therefore $`38`$h in the simulation are devoted to the computation of the self-gravity force field.
### 4.4 Choice of the sampling distribution
In order to ensure a stationary distribution (in the absence of any perturbation), $`f_s`$ must be a function of the isolating integrals of motion for the unperturbed potential $`\mathrm{\Phi }_0`$. We are limited in the choice of analytical distribution functions $`f_s(E,L_z)`$, as we were for $`f_0(E,L_z)`$, because such models are scarce in the iterature. The known d.f.’s for axisymmetric systems are in general expressible as Fricke series, for which the self-consistent density can be calculated. The simplest choice is thus to consider a term from a Fricke series
$$f_s(E,L_z)=FE^\alpha L_z^\beta ,$$
(21)
where $`E=\mathrm{\Phi }_0\frac{v^2}{2}`$ and $`F`$ is a normalizing constant. The corresponding density (see e.g. Dejonghe & de Zeeuw 1988) is given by:
$$\rho _s(R,z)=AFR^\beta \mathrm{\Phi }_0^{\alpha +\beta /2+3/2},$$
(22)
with $`A=\frac{\mathrm{\Gamma }(\alpha +1)\mathrm{\Gamma }(\beta /2+1/2)}{\mathrm{\Gamma }(\alpha +\beta /2+1/2)}`$. We can not generate a density of particles with a strong cusp using such a term. Models with a cusp in $`R`$ can in principle be built by choosing $`1<\beta <0`$, however (i) their cusp in $`R`$ has a rather faint logarithmic slope $`1<\beta <0`$; (ii) these are d.f.’s with a cusp in $`L_z`$, so their sampling of velocity space is very inhomogeneous. This is actually less unfavourable than it seems, since it favours nearly-radial orbits, which are actually those which experience the strongest perturbation (see §5.5). More freedom is available regarding the density decay at large radii. A steeper decline of density, with respect to the unperturbed model, follows from taking $`\alpha >7/2+\beta /2`$, as can be easily verified. We are not worried by the low corresponding particle density at large radii, since the potential there will be dominated by the analytical term.
Here we experiment with two very different Fricke terms for $`f_s`$, in order to make sure that the choice of the sampling does not affect the measure of the cusp slope, which is our main objective. We use:
1. $`f_{S1}`$: $`\beta =0,\alpha =8`$.
This is a model with a core, but its density falls much faster than the mass density of the unperturbed model. The central particle density is enhanced by splitting particles according to their energy and angular momentum, in a way similar to Merritt & Quinlan (1998). First particles are binned according to their binding energy $`E_0`$. For the most bound particles, we evaluate their pericenter $`R_{min}(E,L_z)`$. If $`R_{min}`$ is smaller than the final BH influence radius (see §4.3), the orbit is selected for splitting. It is then integrated, within the unperturbed potential, for sufficient dynamical times. A number $`n`$ of phase-space positions along the orbit are recorded. Then the initial particle $`(r(t_0),v(t_0))`$ is replaced by $`n`$ particles, each of which is given one of the $`n`$ recorded positions and a statistical weight $`\mathrm{\Gamma }=1/[nf_S(r(t_0),v(t_0))]`$. Usually, out of 20 equally sized bins in energy, we consider for splitting the $`j=17\mathrm{}20`$ bins, take $`n=(j17)^2`$, and repeat the whole procedure a second time (with $`n=(j15)`$).
2. $`f_{S2}`$: $`\beta =0.9,\alpha =7`$.
This choice ensures a central density with cusp $`R^{0.9}`$, which proves an effective way to increase the central number of particles. We do not perform any splitting. On the other hand the sampling is very inhomogeneous in $`R`$ and $`z`$, as already mentioned.
The two particles distributions arising from the two sampling distributions just described are displayed on Fig. 2.
## 5 Cusps in non-rotating models
### 5.1 Cases considered
We assume a point mass is growing at the center of the model, by gas accretion, without depleting the stellar component (as in Young 1980, Quinlan et al. 1995, CB94, LG89). The BH mass grows over a time $`t_{BH}`$ from $`M=0`$ at $`t=0`$, to the final value $`M_{BH}`$ following (cf. Merritt & Quinlan 1998):
$$M(t)=M_{BH}\left(\frac{t}{t_{BH}}\right)^2\left(32\frac{t}{t_{BH}}\right).$$
(23)
For most of the runs discussed hereafter, we took $`t_{BH}=10^7`$ Yrs.
The potential due to the BH is modeled using a Plummer potential with smoothing length $`ϵ_{BH}`$:
$$\mathrm{\Phi }_{BH}=M/\sqrt{r^2+ϵ_{BH}^2}.$$
(24)
We can choose $`ϵ_{BH}`$ by imposing e.g.: $`ϵ_{BH}r_I`$. For instance, $`ϵ_{BH}10^3r_I`$. For the standard case, where $`r_I8\times 10^2`$, we have taken $`ϵ_{BH}=5\times 10^5`$, and verified that this value was small enough (§5.2).
We consider as the standard case one where the BH has a mass of $`2\%`$ the galaxy mass. This is supposed to be roughly representative of observational data. However, our flattened Plummer model does not have a realistic density profile, so that a more meaningful figure may be the ratio of the BH mass to the initial core mass. The latter is roughly $`0.2M_0`$ (see §3.1).
In addition, we discuss below experiments with higher BH masses, and different growth times. The summary of all the cases considered is to be found in Table 3. We will also consider, in the next section, models having initially some rotation. Those will be summarized in Table 4 of that section.
We display on Fig. 3 the results for the standard case. The sampling distribution used was $`f_{S1}`$. For that case, and other ones where the BH mass is very small, statistical noise is reduced by adding $`10`$ snapshots taken within a small time interval after $`t_{BH}`$. This is much less useful for the runs with larger BH masses (see below, Table 3). A polar grid has been used to produce the graphs, with $`20`$ points logarithmically spaced along $`r`$ (based on Eq. 20), and $`10`$ points equally spaced within $`[0,\pi /2]`$ along $`\theta `$, which is the angle from the $`z`$ axis. The smallest value $`r_1`$ of the $`r`$\- grid is chosen so that the total snapshot has at least $`1,000`$ particles with $`r<r_1`$.
Quantities associated to the particles correspond to perturbed quantities. The related grid quantities $`𝐚_{1,kl}`$ at grid points $`(r_k,\theta _l)`$ are derived by applying a CIC scheme in the polar geometry. The total fields are finally recovered by adding the corresponding unperturbed analytical terms evaluated on the grid points. Thus for the total density:
$$\rho _{kl}=\rho _0(r_k,\theta _l)+\rho _{1,kl}.$$
(25)
Similarly the velocity components and dispersions are derived by adding the analytical terms, using Eq. 18, for $`V=V_R,V_\varphi ,`$ and $`V_z`$, and $`n=1,2`$. Fig. 3 shows isolevel contours for the total density at the end of the simulation, as well as for the total velocity dispersion $`\sigma ^2=\sigma _R^2+\sigma _z^2+\sigma _\varphi ^2`$. The dash-dotted curves show the initial, unperturbed quantities. Also displayed (first row) are the curves $`\rho _{k,1}`$ and $`\rho _{k,10}`$, which give approximately the profiles of the total density along $`R`$ and $`z`$, respectively. The same is drawn on the bottom row for the square total dispersion $`\sigma ^2`$. The dotted segments on these graphs indicate a line with logarithmic slope respectively $`3/2`$ for the density, and $`1`$ for the square total dispersion.
Therefore, the figure shows that even for the less massive BHs considered in this work, the central cusp is well resolved in two-dimensions. The maps also show that the central region has become very nearly spherical; the final axis-ratio is displayed on Fig. 4 as a function of the major-axis radius, scaled to $`r_I`$. By eye inspection of 3, the slope for the velocity is very similar to what is expected theoretically; also the density cusp appears to be similar to what is predicted for an initially spherical model. The cusp slopes will be measured with more care and discussed further in §5.4. We now turn to some runs performed in order to check our computations.
### 5.2 Numerical checks
A run was first made for $`b_P=0`$, in order to check that the slope obtained for a spherical potential agrees with the analytical estimates for this case. Its slope is measured, in the way we will explain in §5.4, both for the density and the velocity dispersion. We measure respectively $`\gamma _\rho 1.52\pm 0.02`$ for the density, in good agreement with the adiabatic model (and the simulation by Sigurdsson et al. 1995), and $`\gamma _{\sigma ^2}0.94\pm 0.05`$, very close to unity, as expected in the nearly Keplerian potential produced around the BH.
We also made some numerical checks for the standard axisymmetric run. First of all, the simulation was pursued after the BH growth time for an equal duration, in order to verify that a stationary model had been reached. The influence of the smoothing length $`ϵ_{BH}`$ used in the BH potential was tested, by comparing runs using either a larger ($`ϵ_{BH}=10^4a_P`$) or a smaller ($`ϵ_{BH}=2\times 10^5a_P`$) smoothing length. The larger value produced a slightly shallower cusp, but results were unaffected for the smaller value. Therefore we use $`ϵ_{BH}=5\times 10^5a_P`$ in all subsequent runs –including those with larger BH masses, where a larger value of $`ϵ_{BH}`$ could have sufficed.
We have experimented with both the $`f_S`$ described in §4.4. The final model obtained, for the parameters of the standard case, but using $`f_{S2}`$ instead of $`f_{S1}`$, is shown on Fig. 5. Results can be seen, by comparing to Fig. 3, to be extremely similar to those obtained when using the other sampling distribution, in spite of the fact that $`f_{S1}`$ and $`f_{S2}`$ are very different functions. The first sampling ($`f_{S1}`$) turned in practice to yield results with apparently somewhat less particle-noise for the density. This is probably due to the fact that $`f_{S1}`$ corresponds to a particle distribution which is similar along the $`R`$ and $`z`$ axes, while $`f_{S2}`$ has a cusp only in $`R`$. This may for instance explain why the spherically averaged cusp has more mass for $`f_{S1}`$ than for $`f_{S2}`$ (see Fig. 8). Some perturbation mass may be less well sampled in the $`z`$ direction. The differences, however, remain marginal. The two sampling functions give estimates for the logarithmic slope of the cusp (see 5.4) that can not be distinguished, within the error bars. Therefore we are confident that our results are not affected by the initial particle distribution.
A few runs were also performed in order to make sure that the results do not depend significantly on numerical parameters, such as the spacing of the grid used for the force evaluation (§4.3), the total number of force computations by GRAPE 4.3), or the choice that we made for the number of particles $`N=3\times 10^5`$.
### 5.3 Importance of the cusp self-gravity
The perturbed central densities are high, but within small volumes. Therefore the mass in the cusp is not necessarily very large. Applying Eq. 2 to the Lynden-Bell model we find for the influence radius:
$$r_I2a_P\frac{M_{BH}}{M_0}.$$
(26)
For the standard run $`M_{BH}/M_0=0.02`$, thence $`r_I0.16`$. Initially, the mass enclosed in this radius is roughly (neglecting the effects of flattening): $`M_0(r_I)M_0(r_I/\sqrt{r_I^2+a_P^2})^32\times 10^5M_0`$. At the end of the simulation, we find it is: $`M(r_I)=M_0(r_I)+M_1(r_I)=M_0(r_I)+\mathrm{\Sigma }_{(r_i<r_I)}m_{1i}2\times 10^5+6\times 10^46\times 10^4M_0`$. Therefore the mass within the cusp remains a small fraction of the system mass.
On the other hand, the system has been affected by the central mass in a region much larger than the cusp itself. This can be seen from the radial distribution of the perturbation mass density, displayed on Fig. 6 for the standard case. For $`r2`$, $`\frac{dM_1(r)}{dr}<0`$, corresponding to orbits that have been requisitioned in order to build the cusp; this negative perturbation extends far from the center. As a consequence, the cumulated mass $`M_1(r)`$ decreases after $`r2`$. It is $`10\%`$ of its maximum value at $`r5`$. Outside this radius the cusp force is practically zero (Fig. 6). The orbital participation will be further described below (§5.5).
We may estimate the self-consistent force (due to the cusp itself) $`F_{s.g.}=\mathrm{\Phi }_1`$ within the influence radius, where we assume $`\rho _1(R,z)Kr^{3/2}`$. A reasonable rough guess for $`K`$ is such that $`\rho _1(r_I)\rho _0(r_I)`$. This yields a roughly correct ‘calibration’ for the central value, in agreement with the numerical value at the end of the simulation. Fig. 6 displays the different radial force terms: $`F_0=_R\mathrm{\Phi }_0`$, $`F_1=_R\mathrm{\Phi }_1`$, $`F_{BH}`$. Within a substantial fraction of the initial core radius ($`0.75a_P=3`$), the unperturbed force is negligible. The BH dominates the central region, within roughly $`a_P/4`$, by a factor scaling from $`10^4`$ at $`R0`$ to $`10^2`$ at $`Rr_I`$. The self-consistent force contribution reaches at most $`1\%`$ for $`M_{BH}/M_0=0.02`$. Therefore we have neglected it for most runs with this BH mass. The self-gravity contributes at most for $`2\%`$ when $`M_{BH}/M_0=0.03`$ (see Table 3). This is still very small, therefore we have also neglected the cusp self-gravity for models with $`M_{BH}/M_0=0.03`$.
The self-consistent term is not negligible, however, for the larger mass ratio $`M_{BH}/M_0=0.2`$ considered in this work, since it contributes for up to $`15\%`$ of the total force at $`Ra_P/4`$ (Fig. 7).
### 5.4 Measure of the cusp slope
The Poisson noise due to the discretisation into particles varies with the number of particles as $`1/\sqrt{N}`$. To reduce it, we continue the simulation for another $`0.5\times 10^7`$ Yrs, and draw 10 snapshots, at regular time intervals. The snapshots are then merged, and this snapshot of $`10N3\times 10^6`$ particles is analyzed. This is especially useful for the runs with $`M_{BH}/M_0=0.02`$, and, to a lesser extent, $`M_{BH}/M_0=0.03`$.
The central region (within $`r_I`$) is very nearly spherical, as can be seen from the 2D contour map of $`\rho (R,z)`$ (Fig. 3). Therefore, in order to measure the cusp slope, we can integrate over the $`\theta `$ angle of the polar coordinates $`(r,\theta )`$. A logarithmic grid in $`r`$ is built, using Eq. 20 and such that the sphere with radius the first grid point $`r_1`$ encloses $`1000`$ particles. The spherically symmetrized density and velocity dispersions are evaluated at the grid points, and these grid values used to derive the corresponding logarithmic slopes.
The logarithmic slopes $`\gamma _\rho `$ for $`\rho (r)`$, and $`\gamma _\sigma ^2`$ for $`\sigma ^2(r)`$, are evaluated by computing the respective average logarithmic derivative over the grid points in some interval $`I_r=[0,r_{max}]`$. To estimate the error in this measure, we also compute the slope for each individual snapshot, and infer the mean deviation.
The radius $`r_{max}`$ should in theory be $`r_{max}r_I`$. In practice, we take the maximum interval in which the radial profile appears as a straight line when plotted in logarithmic axes. This leads actually to consider a slightly smaller interval than $`[0,r_I]`$ for $`\sigma ^2(r)`$, in order to avoid the transition region from cusp to core (see Fig. 8). Including grid points that belong to this transition region would obviously lead to a systematic under-estimate for the cusp slope.
This point is illustrated in Table 2, for the standard case with $`M_{BH}/M_0=0.02`$. This is the case where the measure depends the most on the choice of $`I_r`$. Indeed, the less massive the BH, the fewer the grid points contained within $`r_I`$. As a consequence, each of the grid points has a larger statistical weight in the average slope.
Of course, the value of the slope for $`\sigma ^2`$ must be $`1`$ in the region where the BH dominates the dynamics. As can be seen from the table, the effects of excluding successively $`r_N`$, $`r_{N1}`$ etc. from the $`N`$ grid points within $`[0,r_I]`$ are, for $`\gamma _{\sigma ^2}`$: (i) to increase the average slope estimate (ii) to increase the mean deviation around the average value, as fewer data points are made available.
If we exclude the $`2`$ points lying within the cusp/core transition, the logarithmic slope measured for $`\sigma ^2`$ is very nearly equal to unity, as it should be.
As for the density, there are up to 8 points that lie well within the cusp (Fig. 8). Our density profile deviates slightly from a pure power-law at the 2 inner points. Therefore we obtain a better estimate of the slope in the linear section by taking into account the full set of points within $`r_I`$. With these grid points, we find for the standard case:
* using $`f_{S1}`$ : $`\gamma _\rho 1.46\pm 0.06`$
Fig. 8 also displays the density and velocity profiles obtained when the sampling distribution is $`f_{S2}`$. The central density value is slightly smaller than with the previous case, probably because the sampling is not as efficient due to its strong spherical asymmetry. The value measured for the density logarithmic slope is now:
* using $`f_{S2}`$ : $`\gamma _\rho 1.35\pm 0.15`$,
This value is somewhat smaller than the value for the other sampling, but still consistent with it. We therefore conclude that the two sampling distributions tested, in spite of their different behaviour in the centre, yield similar results. We therefore believe the results are not an artifact of the method used. The value of $`\gamma _\rho `$ is consistent, within its error bars, with the value given in the literature for the spherical case, $`\gamma =3/2`$. We find no evidence for a significantly larger, nor smaller, slope.
The results found for experiments using a larger final BH mass are discussed in §5.7. For such cases, the dispersion in the measures is very small (see Table 3), as the number of points useful for the measure is larger.
### 5.5 Orbital response to the BH
When we use the PP method, we have direct access to the distribution functions $`f_0,f_1`$ and $`f(E,L_z)=f_0+f_1`$. Those are available with a large signal-to-noise ratio, even for the case where the BH mass is only $`1\%`$ of the model mass, that we analyse in this paragraph.
The initial model can be viewed by plotting the surface $`f_0(E_0,L_z)`$, with $`E_0=\mathrm{\Phi }_0v^2/2`$ the unperturbed energy. Since $`\mathrm{\Phi }_0(r=0)=0.25`$, $`0<E_0<0.25`$. This surface, drawn from a set of initial conditions for the particles, is shown on Fig. 9 (central panel). We use a regular grid in $`E_0`$ and $`L_z`$ to bin the particles, and sum the masses $`m_{0i}=f_0(w_i)/f_S(w_i)`$ within each cell of the grid. It could of course have been drawn analytically, but, as it is, the figure shows how smooth the numerical realization is. A map using a linear grey scale is shown on top of the surface. At fixed energy, the d.f. increases with increasing $`L_z`$, as can be seen from the figure. This is related to the excess of tangential motion supporting the flattening.
In a similar way, using snapshots at the end of the standard run, we can build the surfaces $`f_1(E,L_z)`$ and $`f(E,Lz)`$. We plot in fact two slightly different maps. (i) A map of $`f_1`$ as a function of the initial values of $`E_0`$ and $`L_z`$$`L_z`$ is anyway conserved during the BH growth– reveals which orbits have been most perturbed. (ii) On the other hand, a map for $`f(E,L_z)=f_0(E,L_z)+f_1(E,L_z)`$ (with $`E=\mathrm{\Phi }_0+\mathrm{\Phi }_{BH}+\mathrm{\Phi }_{s.g.}\frac{v^2}{2}`$ the total final energy) yields information about the orbital structure in the final model. In practise and for the sake of simplicity, we neglect the self-gravity of the cusp (we have shown in §5.3 that the self-gravity contribution is small for $`M_{BH}/M_00.03`$), and plot $`f(E^{},L_z)`$, where $`E^{}\mathrm{\Phi }_0+\mathrm{\Phi }_{BH}\frac{v^2}{2}`$.
The map for $`f_1`$ is show on Fig. 9 (left panel). Positive values of $`f_1`$ are displayed in grey shades darker than the background (with white contour lines), whereas negative values appear as lighter shades (with black contour lines). The map shows that orbits having $`E_0\mathrm{\Phi }_0`$ have been the most perturbed. Also, at a given energy, the perturbation is more important for orbits with small angular momentum $`|L_z|`$. For the most bound orbits ($`E_0\mathrm{\Phi }_0`$), $`f_1>0`$, reflecting the fact that they now belong to phase-space regions more populated than initially. By contrast, a negative perturbation ($`f_1<0`$) can be found for smaller values of $`E_0`$, and $`L_z0`$. These orbits have an apocenter located at a radius much larger than the influence radius. The fact that the perturbation extends much further than the influence radius was already seen in Fig. 65.3).
We have used 10 snapshots, issued at the end of the simulation and merged together, in order to produce the surface $`f(E^{},L_z)`$ shown on the right panel of Fig. 9. The final total distribution shows most clearly that only orbits with $`L_z0`$ have a final energy $`E^{}>\mathrm{\Phi }_0(r=0)`$ (Fig. 9). These are orbits such that $`\mathrm{\Phi }_{BH}\frac{v^2}{2}>0`$, i.e. orbits that have become bound to the BH potential. Such orbits must provide an important contribution to the cusp. However, the fact that nearly radial orbits are more efficiently attracted into the cusp, does not imply radially biased motion within the cusp, since these orbits get rounder (see also CB94). In models with a central cusp, constructed either with, or without, a BH, Dehnen (1995) and Dehnen & Gerhard (1994) find that such orbits typically have $`L_zL_c(E)`$.
The final d.f. appears remarkably constant at high binding energies ($`f(E)cst`$ for $`E>\mathrm{\Phi }_0(0)`$ and $`Lz0`$). Tremaine et al. (1994) have built analytical models for spherical systems with a central density cusp and a central BH. For a cusp with slope $`\gamma =3/2`$, their d.f. tends to a constant for high binding energies, in a way very similar to our results.
The final d.f. can therefore be said ‘degenerate’ at high binding energies, in the sense that different energy levels have the same probability. Such a degeneracy is also typical of violently relaxed systems (see Lynden-Bell 1967, Chavanis & Sommeira 1998). Our results therefore support those by Stiavelli (1998), who has shown that the observable signatures of an adiabatically grown BH, on the one hand, and of violent relaxation around a pre-existing BH, on the other hand, are very similar. We have shown that this is due to the fact the distribution functions themselves have very similar behaviour at high energies.
### 5.6 Influence of growth time
The influence of the BH growth time on the final cusp has been checked, by experimenting with shorter times ($`T_{BH}=10^6`$, $`10^5`$ Yrs), for the case $`M_{BH}/M_0=0.03`$. The dynamical times within the cusp are close to $`10^5`$ Yrs, therefore the assumption of adiabaticity does not hold for the latter case. Moreover, some orbits may not have had time to rearrange into an equilibrium during the BH growth. Therefore all simulations are pursued after the BH has reached its final mass, so that the total time is $`10^7`$ Yrs for all cases.
Results for the logarithmic slopes are reported in Table 3. In good agreement with Sigurdsson et al. (1995), we found that the adiabatic model is still roughly valid for growth times of order $`10T_{dyn}`$. Nevertheless small growth times introduce some differences, and already for $`T_{BH}=10^6`$ Yrs, there is a tendency for the cusp to spread over a larger diameter. Whereas the density profile follows a power-law in a smaller region around the center, the transition zone around $`r_I`$ is much larger than in the case $`T_{BH}=10^7`$ Yrs. There is therefore a trend for the average cusp slope to be slightly smaller than $`3/2`$.
This trend towards a shallower cusp is more evident in the case where $`T_{BH}=10^5`$ Yrs (Fig. 10). Actions are not expected to be conserved in this case where $`T_{grow}T_{dyn}`$, as orbits are deflected by the rapidly varying central potential. Probably as a result of such deflections, we find an excess of radial velocity dispersion ($`\sigma _R\sigma _z>\sigma _\varphi `$) in the central region, as found also by Sigurdsson et al. (1995) for experiments with a similar growth time. Less orbits settle in the vicinity of the BH, and the cusp appears less strong than in the adiabatic case, with a logarithmic slope $`\gamma _\rho 0.9`$.
### 5.7 Varying the BH mass
We also experiment with BH masses which are a larger fraction of the model mass (see Table 3). The mass of the dark matter concentrations betrayed by a central cusp in early-type galaxies is evaluated to be at most $`2\%`$ of the galaxy mass (e.g. Richstone et al. 1998). Maybe more significant for the cusp morphology and kinematics, is the ratio between the BH mass and the initial core mass of the galaxy.
Semi-analytic works (Young 1980, Quinlan et al. 1995) predict that the density profile of the cusp remains roughly self-similar, as long as the final values of the BH mass are smaller than the mass within the initial core. On the other hand, for $`M_{BH}`$ larger than the core mass, the cusp is steeper, with $`\gamma _\rho `$ approaching a value of $`2`$ when the mass ratio is $`M_{BH}/M_{core}1`$. Of course no variation is seen in the velocity dispersion cusp, which is always what we expect in an essentially keplerian potential ($`\gamma _{\sigma ^2}=1`$).
The cases reported in Table 3 correspond to a BH with mass equal respectively to $`0.05`$, $`0.15`$ and $`1`$ times the core mass $`M_{core}`$. The first case has already been discussed in §5.4; the two other cases are illustrated on Fig. 11, which displays the spherically symmetrized profiles for the density and the velocity dispersions.
As can be seen from this Figure, the number of points within the power-law cusp is much larger than for the standard case. For a BH whose mass is a substantial fraction of the core mass, the statistics on the results is therefore much better than in the standard case $`M_{BH}/M_{core}=0.1`$.
A single snapshot (with $`N320,000`$ particles) suffices to trace clearly the cusp, and measure its slope with good accuracy. The dispersion in this measure is very small, as can be seen in the results summarized in Table 3. Fig. 12 shows in more detail the case with $`M_{BH}=0.2M_0`$, corresponding to $`M_{BH}/M_{core}1`$. It was plotted using one snapshot.
The logarithmic slope $`\gamma _\rho `$ that we measure for the total density is an increasing function of the final BH mass (see Table 3). It takes values within $`[3/2,2]`$, very much as expected from the spherical adiabatic model. Fig. 13 compares the density profiles obtained for the different $`M_{BH}/M_{core}`$ values considered.
We also ran a simulation where the final BH mass was $`2M_{core}`$. Although results seemed very reasonable in all the cases considered, the error in the mass conservation increases in this case to non-negligible values ($`M_1/M_120\%`$). The computation is therefore at the limit of credibility for the method. The perturbation technique used here is indeed better suited to small or moderate values of $`M_{BH}/M_0`$, rather than larger ones. The reason is not that the method is limited to the linear regime: the computation is fully non-linear, since we compute perturbed orbits within the full potential. The reason is rather that the sampling is more difficult to control for larger perturbations. The non-conservation of the total mass reflects the sampling inadequacies. For higher mass ratios than those considered in this work, a standard $`N`$-body technique should probably be preferred.
## 6 Cusps in rotating systems
In this section, we investigate the effects of an initial rotation of the host galaxy. LG89 and CB94 both study the rotation brought to central regions by a growing cusp, but find different results. Is the galaxy response somehow affected in these works by the approximation of a spherical potential? Here we can study this case in its true geometry, without any such approximation. Moreover, we would like to clarify which difference in these works is responsible for the difference in the final velocity field. The net rotation within the influence radius is weak in CB94, unless the BH has a mass much larger than the initial core mass. On the other hand, LG89 obtain a cusp in the rotation velocity for any initial BH mass. We therefore investigate in turn models that are built similarly to those considered by CB94, and LG89.
### 6.1 Models with $`p_1(L_z)`$
We first experiment with rotating models built using $`p(E,L_z)=p_1(L_z)`$ (Eq. 11). The motivation for considering this kind of rotation tapering for small $`L_z`$, is obviously its simplicity. Furthermore, such models are not unplausible if a galaxy has formed through a major merger. Violent relaxation in the central regions can erase net rotation for very bound orbits, whereas less bound orbits may still exhibit a preference for one sense of rotation. An example of merger remnant with this sort of angular momentum distribution is found in numerical simulations by Barnes & Hernquist (1996; their Fig. 17).
The parameters in $`p_1`$ have been chosen as follows:
$$L_m=0.2l_0\eta =0.5.$$
(27)
The corresponding rotation curve is displayed on Fig. 14.
For each BH mass considered, the slope of the cusp is the same, within the error bars, as in the corresponding non-rotating case. These results are summarized in Table 4.
The initial distribution $`f_0(E_0,L_z)`$ is displayed on the middle panel of Fig. 15, for a mass ratio $`M_{BH}/M_0=0.01`$, corresponding to $`M_{BH}/M_{core}0.05`$. Again, even for this small final BH mass, the distribution function exhibits little noise. Preference for the positive sense of rotation is reflected by the distortion of the surface $`f_0(E_0,L_z)`$, which is higher for $`L_z>0`$ values.
The left panel of Fig. 15 shows the perturbation $`f_1`$ as a function of the initial integrals of motion $`E_0,L_z`$. The grey shades and contour lines are as in the left panel of Fig. 9. Again, perturbation is strongest for large $`E_0`$ values, and small $`L_z`$ values. The peak of positive perturbation (at $`EE_0`$) is roughly even in $`L_z`$, corresponding to an energy range where the initial distribution $`f_0`$ is itself roughly even in $`L_z`$. We have indeed considered a model like the model in CB94, where the net rotation is mainly due to orbits having $`L_zL_0`$, with $`L_0`$ some finite value. For orbits with high binding energy, $`L_c(E)`$ is smaller than $`L_0`$, so that practically none of them can contribute to the net rotation. On the other hand, for smaller energies $`E_0`$, the perturbation $`f_1`$ is odd with respect to $`L_z`$. This is a consequence of the odd term $`\eta p(L_z)`$ that has been added to the distribution function $`f_0`$ in order to construct $`f_0^r`$ – see Eq. 8.
The final total distribution function $`f(E,L_z)`$ is displayed on the same figure (right panel). At high energies, the distortion of this surface towards positive $`L_z`$ has slightly decreased in amplitude, corresponding to the negative perturbation $`f_1`$ for the positive $`L_z`$. The strip of particles having $`E>\mathrm{\Phi }_0(0)`$ is very narrow around $`L_z=0`$, as can be best seen on the grey scale map on top of the right panel. Thus, only orbits with $`L_z0`$ become bound to the cusp. As a consequence, orbits confined to the central regions do not create a significant global rotation.
Fig. 16 shows the rotation at the end of the BH growth for different mass ratios. Similar to results by CB94, we find that a moderate BH mass does not bring a significant rotation within the influence radius. This result therefore was not affected by simplifications in the analytical derivation of CB94. Our results are indeed very close to those of CB94, as can be seen by comparing our Fig. 16 and their Fig. 3.
As the BH mass increases well above the core mass, its influence radius eventually overcomes the radius corresponding to the circular orbit with angular momentum $`L_0`$, and rotation becomes more important in the centre. However, for the mass ratios we consider (up to $`M_{BH}/M_{core}1`$), we find that no cusp is produced in the rotation velocity.
### 6.2 Models with $`p_2(E,L_z)`$
We now experiment with a rotating model that uses $`p(E,L_z)=p_2(L_z/L_c(E))`$ given by Eq. 11. This model is similar to that by LG89, with their parameter $`\omega `$ corresponding roughly to $`\pi /(2x)`$ in our models. We obtain a model with an initial rotation very close to the rotation for the models with $`p_1`$ previously considered, by taking:
$$x=0.2,\eta =0.5.$$
(28)
The corresponding rotation profile along the $`R`$ axis can be seen on Fig. 14.
In the models now considered, an identical fraction of orbits, at any energy level, contributes to the net rotation. Orbits having initially $`E\mathrm{\Phi }_0(0)`$ have $`L_c(E)0`$, so that even orbits with very small values of $`L_z`$ contribute to the net rotation. These orbits are those which will bring rotation within the cusp. A map for $`f_1(E_0(t=0),L_z)`$ (see §5.5), displayed on Fig. 17, shows indeed that, for orbits with high initial energies, the perturbation has grown positive for $`L_z>0`$, and negative for $`L_z<0`$. Therefore a rotational velocity has efficiently built within the cusp.
The effects on the final distribution function are best seen from Fig. 18. We have plotted the case with$`M_{BH}/M_0=0.03`$, rather than the standard case with $`M_{BH}/M_0=0.01`$, because the structure within the total final d.f. is more easily visible. The fact that rotation has been dragged within the density cusp is disclosed by the bent of the surface strip of particles with high binding energies.
This explains why a cusp in rotation velocity is now produced for any BH mass considered, in agreement with the models described by LG89. The cusp follows from the fact that, initially, rotation was present at all levels of energy, including high energies. The rotation velocity profiles along the $`R`$ axis are shown on Fig. 19, for different masses of the BH. The different curves are labeled as in Fig. 16.
Therefore we have shown that the rotation in the cusp region depends on the orbital structure that existed in the galaxy prior to the BH growth. The formation of a cusp in $`V_{rot}`$ around a BH that grows adiabatically is sensitive to the way the initial d.f. depends on $`L_z`$. If rotation is not present initially at high binding energies, only very massive BHs –with respect to the initial core mass – will produce an observable signature onto the $`V_{rot}`$ profile.
There has been some debate about the role played by a nuclear disk onto the light profile in the center of E galaxies (see Jaffe et al. 1994, and Lauer et al. 1995). Such a disk would of course demand a different model for the rotation profile. The suggestion that the power-law galaxies (those with steep cusps) owed their inner luminosity profile to the presence of a disk has been much weakened by the fact that Lauer et al. (1995) find little evidence for inner disks in the power-law E’s of their sample. More evidence for central disks, at the scale of $`10`$ pc, has been collected for SO galaxies (van den Bosch et al. 1994, Scorza & van den Bosch 1998), or giant E’s with shallow cusps (Forbes 1994). Obviously the statistics on the detection of such disks is still too small, and the dynamical models far from entering such detail.
## 7 Conclusion
This paper aims at investigating the cusp induced in elliptical galaxies by the growth of a central, supermassive black hole. The spherical case is the only one to have been studied in detail in the literature previously, using an adiabatic model. In this paper we study the cusp due to a growing BH by numerical means, in order to free ourselves from the assumption of spherical symmetry.
We thus investigate the possible influence of flattening and rotation, by considering a simple axisymmetric model. For our models, which have a substantial central flattening, we do not find any sizeable signature on the cusp slope with respect to the spherical case. We deduce from this result that a reasonable degree of tangential anisotropy in the stellar velocities, which sustains the flattening, has little effect on the cusp slope. The spherical adiabatic models appear very robust with respect to the geometry of the initial galaxy. We find a cusp slope of $`1.5`$ in the models where the BH mass is less than the initial core mass, whereas the observed preferred value for low mass E’s is larger, around $`1.9`$. This fact remains to be explained. One possibility is that the BH grew in a galaxy with initially a very small core, as suggested by van der Marel (1999), or even no core at all. The observational trends (steeper cusps for low luminosity E galaxies, and smaller cores for less massive core Ellipticals) are roughly consistent with this suggestion (see van der Marel 1999). Violent relaxation during a merger or a gravitationnal collapse favours highly concentrated systems (Farouki et al. 1983; see also Chavanis & Sommeira 1998), specially if dissipation occurs (Udry 1993). Gas or stellar collisions around the BH may also have favoured steeper cusps in low $`L`$ galaxies. The relaxation time is indeed shorter for less massive E galaxies as they have denser centers (cf. Magorrian & Tremaine 1999) and we know that the cusp induced in a collisional system is higher than in the collisionless models, with $`\gamma _\rho `$ around $`7/4`$ (cf. Duncan & Shapiro 1983).
The formation of a cusp in $`V_{rot}`$ around a BH that grows adiabatically appears sensitive to the way the initial d.f. depends on $`L_z`$. If rotation is not present initially at high binding energies, only very massive BHs will produce a signature in the $`V_{rot}`$ profile. Such an initial orbital distribution could for instance be expected after mergers involving an efficient violent relaxation in the central region. Stellar kinematic observations with the required resolution are extremely scarce, although STIS observations (e.g. Joseph et al. 2000) may soon improve on this situation. Existing data for M32 (Joseph et al. 2000) and for the E5 galaxy NGC3377 (Kormendy et al. 1998) show an increase of rotation velocity within the inner few arcsec. This is again consistent with the interpretation that in power-law elliptical galaxies the BH mass exceeds the initial core mass (eventually zero).
On the other hand, a cusp in rotation is produced, whatever the BH/core mass ratio, when initially the fraction of orbits contributing to rotation is non-zero at high binding energies. Evidence for such a cusp is probably present in HST observations of the SO galaxy NGC 4342 (Cretton & van de Bosch 1999).
Let us finally note that the spherical adiabatic model yields values for the mass of dark matter concentration similar to more elaborate dynamical modeling (van der Marel 1999). This however does not preclude that central BHs pre-existed to their host galaxy formation, since the adiabatic growth scenario predicts final galaxy models very similar to those produced by violent relaxation around a pre-existing BH (as demonstrated by Stiavelli 1998, and by our own results on the distribution function).
Acknowledgments
We are grateful to J.-C. Lambert for his extremely efficient and cheerful help with implementing the code on the GRAPE machines. We thank P.-H. Chavanis and A. Bosma for stimulating discussions. F.L. acknowledges gratefully a Marie-Curie fellowship, as well as hospitality at the Marseille observatory, during this work. We would also like to thank IGRAP, the INSU/CNRS and the University of Aix-Marseille I for funds to develop the computing facilities used for the calculations in this paper.
References
Athanassoula, E., Bosma, A., Lambert, J.-C., Makino, J., 1998 MNRAS 293, 369
Barnes, J., Hernquist, L. 1996, ApJ 471, 115
Bender, R. et al. 1989, A&A 217, 35
Binney, J., Tremaine, S. 1987. ‘Galactic Dynamics’, Princeton Univ. Press, Princeton, NJ
Chavanis, P.-H., Sommeira, J. 1998, MNRAS 296, 569
Cipollina, M., Bertin, G. 1994, A&A 288, 43 (CB94)
Cretton, N., van den Bosch, F. C. 1999, ApJ 514, 704
Dejonghe, H. 1986,Phys. Rep., 133, 218
Dejonghe, H., de Zeeuw, T. 1988, ApJ 329, 720
Dehnen, W. 1995, MNRAS 274, 919
Dehnen, W., Gerhard, O. 1994, MNRAS 268, 1019-1032.
Duncan, M. J., Shapiro, S. L. 1983, Nature 262, 743
Duncan, M. J., Levison, H. F., Lee, M.-H. 1998, AJ 116, 51
Evans, N. W. 1994, MNRAS 267, 333-360.
Farouki, R., Shapiro, S., Duncan, M. 1983, ApJ 265, 597
Forbes, D. A. 1994, AJ 107, 2017
Gebhardt, K. et al. 1996, AJ 112 , 105
Gerhard, 0. & Binney, J. 1985, MNRAS 216, 467
Goodman, J., Binney, J. 1984, MNRAS 207, 511
Haehnelt, M. G., Rees, M. J. 1993, MNRAS 263, 168
Hernquist, L., Ostriker, J. P., ApJ 386, 362
Hockney, R., W., Eastwood, J. W., 1981. ‘Computer Simulations Using Particles’, McGraw-Hill
Hunter, C, Qian, E. 1993, MNRAS 262, 401
Jaffe, W., Ford, H. C., O’Connell, R. W., van den Bosch, F. C., Ferrarese, L. 1994, AJ 108, 1567
Joseph, C. L., Merritt, D., Olling, R., Valluri, M. et al. 2000 in preparation
Kawai, A., Fukushige, T., Taiji, M., Makino, J., Sugimoto, D. 1997, PASP 49, 607
Kormendy, J. 1988. In ‘Supermassive Black Holes’, ed. M. Kafatos (Cambridge: Cambridge Univ. Press)
Kormendy, J., Bender, R., Evans, A. S., Richstone, D. 1998, AJ 115, 1823
Kormendy, J., Richstone, D. 1995, ARA&A 33, 581
Lauer, T. et al. 1995, AJ 110, 2622
Lee, M. H., Goodman, J. 1989, ApJ 343, 594 (LG89)
Leeuwin, F., Combes, F., Binney, J. 1993, MNRAS 262, 1013
Leeuwin, F., Dejonghe, H. 1998. In ‘Galaxy Dynamics’, ed. D. Merritt, M. Valluri & J. Sellwood (ASP Conf. Series)
Lynden-Bell, D., 1962, MNRAS 123, 457
Lynden-Bell, D., 1967, MNRAS 136, 101
Magorrian, J., Tremaine, S. 1999, MNRAS 309, 447
Makino, J., Ebisuzaki, T. 1996, ApJ 465, 527
Makino, J., Taiji, M., Ebisuzaki, T., Sugimoto, D. 1997, ApJ 480, 432
Merritt, D. 1998. In ‘Galaxy Dynamics’, ed. D. Merritt, M. Valluri & J. Sellwood (ASP Conf. Series)
Merritt, D., Quinlan, G. 1998, ApJ 498, 625
Nakano, T., Makino, J. 1999, ApJ 510, 155
Nieto, J.-L., Bender, R. 1989, A&A 215, 266
Norman, C.A., May, A., van Albada, T. S. 1985, ApJ 296, 20
Peebles, P. J. E. 1972, Gen. Rel. Grav. 3, 61
Press, W. H., Flannery, B. P., Teukolsky, S. A., Vetterling, W. T., 1986. ‘Numerical Recipes’, Cambridge Univ. Press
Quinlan, G. D., Hernquist, L. 1997, New Ast. 2, 533
Quinlan, G. D., Hernquist, L., Sigurdsson, S. 1995, ApJ 440, 554
Richstone, D. et al. 1998, Nature 395, 14
Saha, P., Tremaine, S. 1992, AJ 104, 1633
Scorza, C., van den Bosch, F. C. 1998, MNRAS 300, 469
Sigurdsson, S., Hernquist, L., Quinlan, G. D. 1995, ApJ 446, 75-85.
Stiavelli, M. 1998, ApJ 495, 91
Tremaine, S. 1997. In ‘Unsolved Problems in Astrophysics’, ed. J. Ostriker (Princeton)
Tremaine, S. et al. 1994, AJ 107, 634
Udry, S. 1993, A&A 268, 35
van den Bosch, F. C., Ferrarese, L., Jaffe, W., Ford, H., O’Connell, R. W. 1994, AJ 108, 1579
van der Marel, R. P. 1999, AJ 117, 744
van der Marel, R. P. 1997. In ‘Galaxy Interactions at Low and High Redshift’, IAU Symp. 186, eds. D. Sanders & J. Barnes (Kluwer)
van der Marel, R., de Zeeuw, T., Rix, H.-W. 1997, ApJ 488, 702
Wachlin, F. C., Rybicki, G. B., Muzzio, J. C. 1993, MNRAS 262, 1007
Wisdom, J., Holman, M. 1991, AJ 102, 1528
Young, P. 1980, ApJ 242, 1232
|
warning/0004/hep-ph0004206.html
|
ar5iv
|
text
|
# Anatomy of the differential gluon structure function of the proton from the experimental data on 𝐹_{2𝑝}(𝑥,𝑄²)
## 1 Introduction: Why unintegrated gluon structure functions?
The familiar objects from Gribov-Lipatov-Dokshitzer-Altarelli-Parisi (DGLAP) evolution description of deep inelastic scattering (DIS) are quark, antiquark and gluon distribution functions $`q_i(x,Q^2),\overline{q}(x,Q^2),g(x,Q^2)`$ (hereafter $`x,Q^2`$ are the standard DIS variables). At small $`x`$ they describe the integral flux of partons with the lightcone momentum $`x`$ in units of the target momentum and transverse momentum squared $`Q^2`$ and form the basis of the highly sophisticated description of hard scattering processes in terms of collinear partons . On the other hand, at very small $`x`$ the object of the Balitskii-Fadin-Kuraev-Lipatov evolution equation is the differential gluon structure function (DGSF) of the target
$$(x,Q^2)=\frac{G(x,Q^2)}{\mathrm{log}Q^2}$$
(1)
(evidently the related unintegrated distributions can be defined also for charged partons). For instance, it is precisely DGSF of the target proton which emerges in the familiar color dipole picture of inclusive DIS at small $`x`$ and diffractive DIS into dijets . Another familiar example is the QCD calculation of helicity amplitudes of diffractive DIS into continuum and production of vector mesons . DGSF’s are custom-tailored for QCD treatment of hard processes, when one needs to keep track of the transverse momentum of gluons neglected in the standard collinear approximation .
In the past two decades DGLAP phenomenology of DIS has become a big industry and several groups — notably GRV , CTEQ & MRS and some other — keep continuously incorporating new experimental data and providing the high energy community with updates of the parton distribution functions supplemented with the interpolation routines facilitating practical applications. On the other hand, there are several pertinent issues — the onset of the purely perturbative QCD treatment of DIS and the impact of soft mechanisms of photoabsorption on the proton structure function in the region of large $`Q^2`$ being top ones on the list — which can not be answered within the DGLAP approach itself because DGLAP evolution is obviously hampered at moderate to small $`Q^2`$. The related issue is to what extent the soft mechanisms of photoabsorption can bias the $`Q^2`$ dependence of the proton structure function and, consequently, determination of the gluon density from scaling violations. We recall here the recent dispute on the applicability of the DGLAP analysis at $`Q^2\text{ }<`$ 2–4 GeV<sup>2</sup> triggered by the so-called Caldwell’s plot . Arguably the $`𝜿`$-factorization formalism of DGSF in which the interesting observables are expanded in interactions of gluons of transverse momentum $`𝜿`$ changing from soft to hard is better suited to look into the issue of soft-hard interface. At last but not the least, neglecting the transverse momentum $`𝜿`$ of gluons is a questionable approximation in evaluation of production cross sections of jets or hadrons with large transverse momentum. It is distressing, then, that convenient parameterizations of DGSF are not yet available in the literature.
In this communication we report a simple phenomenological determination of the DGSF of the proton at small $`x`$. We analyze $`x`$ and $`Q^2`$ dependence of the proton structure function $`F_{2p}(x,Q^2)`$ in the framework of the $`𝜿`$-factorization approach, which is closely related to the color dipole factorization. In the formulation of our Ansatz for $`(x,𝜿^2)`$ we take advantage of large body of the early work on color dipole factorization and follow a very pragmatic strategy first applied in : (i) for hard gluons with large $`𝜿`$ we make as much use as possible of the existing DGLAP parameterizations of $`G(x,𝜿^2)`$, (ii) for the extrapolation of hard gluon densities to small $`𝜿^2`$ we use an Ansatz which correctly describes the color gauge invariance constraints on radiation of soft perturbative gluons by colour singlet targets, (iii) as suggested by color dipole phenomenology, we supplement the density of gluons with small $`𝜿^2`$ by nonperturbative soft component, (iv) as suggested by the soft-hard diffusion inherent to the BFKL evolution, we allow for propagation of the predominantly hard-interaction driven small-$`x`$ rise of DGSF into the soft region invoking plausible soft-to-hard interpolations. The last two components of DGSF are parameterized following the modern wisdom on the infrared freezing of the QCD coupling and short propagation radius of perturbative gluons. Having specified the infrared regularization, we can apply the resulting $`(x,𝜿^2)`$ to evaluation of the photoabsorption cross section in the whole range of small to hard $`Q^2`$.
The practical realization of the above strategy is expounded as follows: The subject of section 2 is a pedagogical introduction into the concept of DGSF on an example of Fermi-Weizsäcker-Williams photons in QED. Taking the electrically neutral positronium as a target, we explain important constraints imposed by gauge invariance on DGSF at small $`\kappa ^2`$. In section 3 we present the $`𝜿`$-factorization approach, which is the basis of our analysis of small-$`x`$ DIS in terms of DGSF. We also comment on the connection between the standard DGLAP analysis of DIS and $`𝜿`$-factorization and property of soft-to-hard and hard-to-soft diffusion inherent to $`𝜿`$-factorization. In section 4 we formulate our Ansatz for DGSF. The results of determination of DGSF from the experimental data on the proton structure function $`F_{2p}(x,Q^2)`$ and real photoabsorption cross section are presented in section 5. In section 6 we discuss an anatomy of the so-determined DGSF in the momentum space and comment on the interplay of soft and hard components in DGSF, integrated gluon SF and proton structure function $`F_{2p}(x,Q^2)`$. In section 7 we focus on effective intercepts of $`x`$-dependence, and the systematics of their change from DGSF to integrated gluon SF to $`F_{2p}(x,Q^2)`$, which illustrates nicely the gross features of soft-to-hard and hard-to-soft diffusion pertinent to BFKL physics. The subject of section 8 is a comparison of integrated gluon distributions from $`𝜿`$-factorization and conventional DGLAP analysis of the proton structure function. As anticipated, the two distributions diverge substantially at very small $`x`$ and small to moderate $`Q^2`$. In section 9 discuss in more detail how different observables — the scaling violations $`F_{2p}(x,Q^2)/\mathrm{log}Q^2`$, the longitudinal structure function $`F_L(x,Q^2)`$ and charm structure function $`F_2^{c\overline{c}}(x.Q^2)`$ — probe the DGSF. In section 10 we summarize our major findings.
## 2 Differential density of gauge bosons: the QED example
For the pedagogical introduction we recall the celebrated Fermi-Weizsäcker-Williams approximation in QED, which is the well known precursor of the parton model (for the review see ). Here high energy reactions in the Coulomb field of a charged particle are treated as collisions with equivalent transversely polarized photons — partons of the charged particle, fig.1. The familiar flux of comoving equivalent transverse soft photons carrying a lightcone fraction $`x_\gamma 1`$ of the momentum of a relativistic particle, let it be the electron, reads
$$dn_e^\gamma =\frac{\alpha _{em}}{\pi }\frac{𝜿^2d𝜿^2}{(𝜿^2+\kappa _z^2)^2}\frac{dx_\gamma }{x_\gamma }\frac{\alpha _{em}}{\pi }\frac{d𝜿^2}{𝜿^2}\frac{dx_\gamma }{x_\gamma },$$
(2)
Here $`𝜿`$ is photon transverse momentum and $`\kappa _z=m_ex_\gamma `$ is the photon longitudinal momentum in the electron Breit frame. The origin of $`𝜿^2`$ in the numerator is in the current conservation, i.e. gauge invariance. Then the unintegrated photon structure function of the electron is by definition
$$_\gamma (x_\gamma ,𝜿^2)=\frac{G_\gamma }{\mathrm{log}𝜿^2}=x_\gamma \frac{dn_e^\gamma }{dx_\gamma d\mathrm{log}𝜿^2}=\frac{\alpha _{em}}{\pi }\left(\frac{𝜿^2}{𝜿^2+\kappa _z^2}\right)^2.$$
(3)
If the relativistic particle is a positronium, fig. 2, destructive interference of electromagnetic fields of the electron and positron must be taken into account. Specifically, for soft photons with the wavelength $`\lambda =\frac{1}{\kappa }a_\mathrm{P}`$, where $`a_\mathrm{P}`$ is the positronium Bohr radius, the electromagnetic fields of an electron and positron cancel each other and the flux of photons vanishes, whereas for $`\lambda a_\mathrm{P}`$ the flux of photons will be twice that for a single electron. The above properties are quantified by the formula
$$_\gamma ^\mathrm{P}(x_\gamma ,𝜿^2)=2\frac{\alpha _{em}}{\pi }\left(\frac{𝜿^2}{𝜿^2+\kappa _z^2}\right)^2V(\kappa ),$$
(4)
where the factor 2 is a number of charged particles in the positronium and corresponds to the Feynman diagrams of fig. 2a, 2b. The vertex function $`V(\kappa )`$ is expressed in terms of the two-body formfactor of the positronium,
$$V(\kappa )=1F_2(𝜿,𝜿)=1\mathrm{P}|\mathrm{exp}(i𝜿(𝐫_{}𝐫_+))|\mathrm{P},$$
(5)
where $`𝐫_{}𝐫_+`$ is the spatial separation of $`e^+`$ and $`e^{}`$ in the positronium. The two-body formfactor $`F_2(𝜿,𝜿)`$ describes the destructive interference of electromagnetic fields of the electron and positron and corresponds to the Feynman diagrams of fig. 2c, 2d. It vanishes for large enough $`\kappa \text{ }>a_\mathrm{P}^1`$, leaving us with $`V(\kappa )=2`$, whereas for soft gluons one has
$$V(\kappa )𝜿^2a_\mathrm{P}^2$$
(6)
One can say that the law (6) is driven by electromagnetic gauge invariance, which guarantees that long wave photons decouple from the charge neutral system.
Finally, recall that the derivation of the differential flux of transverse polarized photons would equally hold if the photons were massive vector bosons interacting with the conserved current, the only change being in the propagator. For instance, for the charge neutral source one finds
$$_V^\mathrm{P}(x_V,𝜿^2)=\frac{\alpha _{em}}{\pi }\left(\frac{𝜿^2}{𝜿^2+m_V^2}\right)^2V(\kappa ).$$
(7)
Recall that the massive vector fields are Yukawa-Debye screened with the screening radius
$$R_c=m_V^1.$$
(8)
To the lowest in QED perturbation theory the two exchanged photons in figs.12 do not interact and we shall often refer to (7) as the Born approximation for the differential vector boson structure function. One can regard (7) as a minimal model for soft $`\kappa `$ behavior of differential structure function for Yukawa-Debye screened vector bosons.
## 3 The insight into the differential density of gluons
### 3.1 Modeling virtual photoabsorption in QCD
The quantity which is measured in deep inelastic leptoproduction is the total cross section of photoabsorption $`\gamma _\mu ^{}pX`$ summed over all hadronic final states $`X`$, where $`\mu ,\nu =\pm 1,0`$ are helicities of $`(T)`$ transverse and $`(L)`$ longitudinal virtual photons. One usually starts with the imaginary part of the amplitude $`A_{\mu \nu }`$ of forward Compton scattering $`\gamma _\mu ^{}p\gamma _\nu ^{}p^{}`$, which by optical theorem gives the total cross cross section of photoabsorption of virtual photons
$`\sigma _T^{\gamma ^{}p}(x_{bj},Q^2)={\displaystyle \frac{1}{\sqrt{(W^2+Q^2m_p^2)^2+4Q^2m_p^2}}}\mathrm{Im}A_{\pm \pm },`$ (9)
$$\sigma _L^{\gamma ^{}p}(x_{bj},Q^2)=\frac{1}{\sqrt{(W^2+Q^2m_p^2)^2+4Q^2m_p^2}}\mathrm{Im}A_{00},$$
(10)
where $`W`$ is the total energy in the $`\gamma ^{}p`$ c.m.s., $`m_p`$ is the proton mass, $`Q^2`$ is the virtuality of the photon and $`x_{bj}=Q^2/(Q^2+W^2m_p^2)`$ is the Bjorken variable.
In perturbative QCD (pQCD) one models virtual photoabsorption in terms of the multiple production of gluons, quarks and antiquarks (fig. 3). The experimental integration over the full phase space of hadronic states $`X`$ is substituted in the pQCD calculation by integration over the whole phase space of QCD partons
$$|M_𝒳|^2𝑑\tau _𝒳\underset{n}{}|M_n|^2_0^1\frac{dx_i}{x_i}d^2𝜿_i,$$
(11)
where the integration over the transverse momenta of partons goes over the whole allowed region
$$0\kappa _i^2\frac{1}{4}W^2=\frac{Q^2(1x)}{4x}.$$
(12)
The core of the so-called DGLAP approximation is an observation that at finite $`x`$ the dominant contribution to the multiparton production cross sections comes from a tiny part of the phase space
$`1x_1x_2\mathrm{}x_{n1}x_nx`$ $`,`$ (13)
$`0\kappa _1^2\kappa _2^2\mathrm{}\kappa _{n1}^2k^2Q^2`$ $`,`$
in which the upper limit of integration over transverse momenta of partons is much smaller than the kinematical limit (12). At very small $`x`$ this limitation of the transverse phase space becomes much too restrictive and the DGLAP approximation is doomed to failure.
Hereafter we focus on how lifting the restrictions on the transverse phase space changes our understanding of the gluon structure function of the nucleon at very small $`x`$, that is, very large $`\frac{1}{x}`$ . In this kinematical region the gluon density $`g(x,Q^2)`$ is much higher than the density of charged partons $`q(x,Q^2),\overline{q}(x,Q^2)`$. As Fadin, Kuraev and Lipatov have shown, to the leading $`\mathrm{log}\frac{1}{x}`$ (LL$`\frac{1}{x}`$) approximation the dominant contribution to photoabsorption comes in this regime from multigluon final states of fig. 3; alternatively, to the LL$`\frac{1}{x}`$ splitting of gluons into gluons dominates the splitting of gluons into $`q\overline{q}`$ pairs. As a matter of fact, for the purposes of the present analysis we do not need the full BFKL dynamics, in the $`𝜿`$-factorization only the $`q\overline{q}`$ loop is treated explicitly to the LL$`\frac{1}{x}`$ approximation. In this regime the Compton scattering can be viewed as an interaction of the nucleon with the lightcone $`q\overline{q}`$ Fock states of the photon via the exchange by gluons, fig. 4, and the Compton scattering amplitude takes the form
$$A_{\nu \mu }=\mathrm{\Psi }_{\nu ,\lambda \overline{\lambda }}^{}A_{q\overline{q}}\mathrm{\Psi }_{\mu ,\lambda \overline{\lambda }}$$
(14)
Here $`\mathrm{\Psi }_{\mu ,\lambda \overline{\lambda }}`$ is the $`Q^2`$ and $`q,\overline{q}`$ helicity $`\lambda ,\overline{\lambda }`$ dependent lightcone wave function of the photon and the QCD pomeron exchange $`q\overline{q}`$-proton scattering kernel $`A_{q\overline{q}}`$ does not depend on, and conserves exactly, the $`q,\overline{q}`$ helicities, summation over which is understood in (14).
The resummation of diagrams of fig. 3 defines the unintegrated gluon structure function of the target, which is represented in diagrams of fig. 4 as the dashed blob. The calculation of the forward Compton scattering amplitudes ($`𝚫=0`$) is straightforward and gives the $`𝜿`$-factorization formulas for photoabsorption cross sections
$`\sigma _T(x_{bj},Q^2)={\displaystyle \frac{\alpha _{em}}{\pi }}{\displaystyle \underset{f}{}}e_f^2{\displaystyle _0^1}𝑑z{\displaystyle d^2𝐤\frac{d^2𝜿}{𝜿^4}\alpha _S(q^2)(x_g,𝜿^2)}`$
$`\{[z^2+(1z)^2]({\displaystyle \frac{𝐤}{𝐤^2+\epsilon ^2}}{\displaystyle \frac{𝐤𝜿}{(𝐤𝜿)^2+\epsilon ^2}})^2`$
$`+m_f^2({\displaystyle \frac{1}{𝐤^2+\epsilon ^2}}{\displaystyle \frac{1}{(𝐤𝜿)^2+\epsilon ^2}})^2\}`$ (15)
$`\sigma _L(x_{Bj},Q^2)={\displaystyle \frac{\alpha _{em}}{\pi }}{\displaystyle \underset{f}{}}e_f^2{\displaystyle _0^1}𝑑z{\displaystyle d^2𝐤\frac{d^2𝜿}{𝜿^4}\alpha _S(q^2)(x_g,𝜿^2)}`$
$`4Q^2z^2(1z)^2\left({\displaystyle \frac{1}{𝐤^2+\epsilon ^2}}{\displaystyle \frac{1}{(𝐤𝜿)^2+\epsilon ^2}}\right)^2`$ (16)
Here $`m_f`$ and $`e_f`$ are the mass and charge of the quark $`f=u,d,s,c,b,..`$,
$$\epsilon ^2=z(1z)Q^2+m_f^2,$$
(17)
the QCD running coupling $`\alpha _S(q^2)`$ enters the integrand at the largest relevant virtuality,
$$q^2=\mathrm{max}\{\epsilon ^2+𝐤^2,𝜿^2\},$$
(18)
and the density of gluons enters at
$$x_g=\frac{Q^2+M_t^2}{W^2+Q^2}=x_{bj}\left(1+\frac{M_t^2}{Q^2}\right).$$
(19)
Here $`M_t`$ is the transverse mass of the produced $`q\overline{q}`$ pair in the photon-gluon fusion $`\gamma ^{}gq\overline{q}`$:
$$M_t^2=\frac{m_f^2+𝐤^2}{1z}+\frac{m_f^2+(𝐤𝜿)^2}{z}.$$
(20)
For longitudinal photons only transitions $`\gamma _Lq_\lambda \overline{q}_{\overline{\lambda }}`$ into states with $`\lambda +\overline{\lambda }=0`$ are allowed. In $`\sigma _T`$ the terms $`m_f^2`$ are the contribution of states with $`\lambda +\overline{\lambda }=\mu `$, whereas the dominant contribution in the scaling regime of $`Q^2m_f^2`$ comes from the transitions $`\gamma _Tq_\lambda \overline{q}_{\overline{\lambda }}`$ into states $`\lambda +\overline{\lambda }=0`$, when the helicity of the photon is transferred to the angular momentum of the quark-antiquark pair. The corresponding transition amplitudes are $`𝐤,𝐤\pm 𝜿`$, for more discussion see .
No restrictions on the transverse momentum in the $`q\overline{q}`$ loop, $`𝐤`$, and gluon momentum, $`𝜿`$, are imposed in the representation (15), (16). This representation was contained essentially in the classic Fadin, Kuraev, Lipatov papers of mid-70’s, in the recent literature it is sometimes referred to as the $`𝜿`$-factorization.
We note that Eqs. (15), (16) are for forward diagonal Compton scattering, but similar representation in terms of the unintegrated gluons structure function holds also for the off-forward Compton scattering at finite momentum transfer $`𝚫`$, off-diagonal Compton scattering when the virtualities of the initial and final state photons are different, $`Q_f^2Q_i^2`$, including the timelike photons and vector mesons, $`Q_f^2=m_V^2`$, in the final state.
The photoabsorption cross sections define the dimensionless structure functions
$$F_{T,L}(x_{bj},Q^2)=\frac{Q^2}{4\pi ^2\alpha _{em}}\sigma _{T,L}$$
(21)
and $`F_2=F_T+F_L`$, which admit the familiar pQCD parton model interpretation
$$F_T(x_{bj},Q^2)=\underset{f=u,d,s,c,b,..}{}e_f^2[q_f(x_{bj},Q^2)+\overline{q}_f(x_{bj},Q^2)],$$
(22)
where $`q_f(x_{bj},Q^2),\overline{q}_f(x_{bj},Q^2)`$ are the integrated density of quarks and antiquarks carrying the fraction $`x_{bj}`$ of the lightcone momentum of the target and with transverse momenta $`Q`$. Hereafter we suppress the subscript $`\mathrm{"}bj\mathrm{"}`$.
### 3.2 Where $`𝜿_{}`$-factorization meets DGLAP factorization
Recall the familiar DGLAP equation for scaling violations at small $`x`$,
$$\frac{dF_2(x,Q^2)}{d\mathrm{log}Q^2}=\underset{f}{}e_f^2\frac{\alpha _S(Q^2)}{2\pi }_x^1𝑑y[y^2+(1y)^2]G(\frac{x}{y},Q^2)\frac{\alpha _S(Q^2)}{3\pi }G(2x,Q^2)\underset{f}{}e_f^2,$$
(23)
where for the sake of simplicity we only consider light flavours. Upon integration we find
$$F_2(x,Q^2)\underset{f}{}e_f^2_0^{Q^2}\frac{d\overline{Q}^2}{\overline{Q}^2}\frac{\alpha _S(\overline{Q}^2)}{3\pi }G(2x,\overline{Q}^2).$$
(24)
In order to see the correspondence between the $`𝜿_{}`$-factorization and DGLAP factorization it is instructive to follow the derivation of (24) from the $`𝜿_{}`$-representation (15).
First, separate the $`\kappa ^2`$-integration in (15) into the DGLAP part of the gluon phase space $`\kappa ^2\text{ }<\overline{Q}^2=ϵ^2+k^2`$ and beyond-DGLAP region $`\kappa ^2\text{ }>\overline{Q}^2`$. One readily finds
$$\left(\frac{𝐤}{𝐤^2+\epsilon ^2}\frac{𝐤𝜿}{(𝐤𝜿)^2+\epsilon ^2}\right)^2=\{\begin{array}{ccc}\left(\frac{2z^2(1z)^2Q^4}{\overline{Q}^8}\frac{2z(1z)Q^2}{\overline{Q}^6}+\frac{1}{\overline{Q}^4}\right)𝜿^2\hfill & \text{if}& \hfill 𝜿^2\overline{Q}^2\\ \left(\frac{1}{\overline{Q}^2}\frac{z(1z)Q^2}{\overline{Q}^4}\right),\hfill & \text{if}& \hfill 𝜿^2\text{ }>\overline{Q}^2\end{array}$$
(25)
Consider first the contribution from the DGLAP part of the phase space $`\kappa ^2\text{ }<\overline{Q}^2`$. Notice that because of the factor $`𝜿^2`$ in (25), the straightforward $`𝜿^2`$ integration of the DGLAP component yields $`G(x_g,\overline{Q}^2)`$ and $`\overline{Q}^2`$ is precisely the pQCD hard scale for the gluonic transverse momentum scale:
$`{\displaystyle _0^{\overline{Q}^2}}{\displaystyle \frac{d𝜿^2}{𝜿^2}}\alpha _S(q^2)(x_g,𝜿^2)\left({\displaystyle \frac{𝐤}{𝐤^2+\epsilon ^2}}{\displaystyle \frac{𝐤𝜿}{(𝐤𝜿)^2+\epsilon ^2}}\right)^2`$
$`=\left({\displaystyle \frac{2z^2(1z)^2Q^4}{\overline{Q}^8}}{\displaystyle \frac{2z(1z)Q^2}{\overline{Q}^6}}+{\displaystyle \frac{1}{\overline{Q}^4}}\right)G(x_g,\overline{Q}^2)`$ (26)
The contribution from the beyond-DGLAP region of the phase space can be evaluated as
$`{\displaystyle _{\overline{Q}^2}^{\mathrm{}}}{\displaystyle \frac{d𝜿^2}{𝜿^4}}\alpha _S(q^2)(x_g,𝜿^2)\left({\displaystyle \frac{1}{\overline{Q}^2}}{\displaystyle \frac{z(1z)Q^2}{\overline{Q}^4}}\right)=\left({\displaystyle \frac{1}{\overline{Q}^4}}{\displaystyle \frac{z(1z)Q^2}{\overline{Q}^6}}\right)(x_g,\overline{Q}^2)I(x_g,\overline{Q}^2)`$
$`=\left({\displaystyle \frac{2z^2(1z)^2Q^4}{\overline{Q}^8}}{\displaystyle \frac{2z(1z)Q^2}{\overline{Q}^6}}+{\displaystyle \frac{1}{\overline{Q}^4}}\right)(x_g,\overline{Q}^2)\mathrm{log}C_2(x_g,\overline{Q}^2,z).`$ (27)
The latter form of (27) allows to conveniently combine (26) and (27) rescaling the hard scale in the GSF
$$G(x_g,\overline{Q}^2)+(x_g,\overline{Q}^2)\mathrm{log}C_2(x_g,\overline{Q}^2,z)=G(x_g,C_2(x_g,\overline{Q}^2,z)\overline{Q}^2).$$
(28)
Here the exact value of $`I(x_g,\overline{Q}^2)1`$ depends on the rate of the $`𝜿^2`$-rise of $`(x_g,𝜿^2)`$. At small $`x_g`$ and small to moderate $`\overline{Q}^2`$ one finds $`I(x_g,\overline{Q}^2)`$ substantially larger than 1 and $`C_2(x_g,\overline{Q}^2,z)1`$, see more discussion below in section 9.
Now change from $`d𝐤^2`$ integration to $`d\overline{Q}^2`$ and again split the $`z`$,$`Q^2`$ integration into the DGLAP part of the phase space $`\overline{Q}^2\frac{1}{4}Q^2`$, where either $`z<\frac{\overline{Q}^2}{Q^2}`$ or $`1z<\frac{\overline{Q}^2}{Q^2}`$, and the beyond-DGLAP region $`\overline{Q}^2\text{ }>\frac{1}{4}Q^2`$, where $`0<z<1`$. As a result one finds
$`{\displaystyle 𝑑z[z^2+(1z)^2]\left(\frac{2z^2(1z)^2Q^4}{\overline{Q}^8}\frac{2z(1z)Q^2}{\overline{Q}^6}+\frac{1}{\overline{Q}^4}\right)}`$
$`=\{\begin{array}{ccc}\frac{4}{3\overline{Q}^2Q^2},\hfill & \text{if}& \hfill \overline{Q}^2\frac{1}{4}Q^2\\ \left(2A_2\frac{Q^4}{\overline{Q}^8}2A_1\frac{Q^2}{\overline{Q}^6}+A_0\frac{1}{\overline{Q}^4}\right),\hfill & \text{if}& \hfill \overline{Q}^2\text{ }>\frac{1}{4}Q^2\end{array}`$ (31)
where
$$A_m=_0^1𝑑z[z^2+(1z)^2]z^m(1z)^m$$
(32)
Let $`\overline{C}_2`$ be $`C_2(x_g,\overline{Q}^2,z)`$ at a mean point. Notice also that $`M_t^2Q^2`$, so that $`x_g2x`$. Then the contribution from the DGLAP phase space of $`\overline{Q}^2`$ can be cast in precisely the form (24)
$$F_2(x,Q^2)|_{DGLAP}\underset{f}{}e_f^2_0^{\frac{\overline{C_2}}{4}Q^2}\frac{d\overline{Q}^2}{\overline{Q}^2}\frac{\alpha _S(\overline{Q}^2)}{3\pi }G(2x,\overline{Q}^2).$$
(33)
The beyond-DGLAP region of the phase space gives the extra contribution of the form
$`\mathrm{\Delta }F_2(x,Q^2)|_{nonDGLAP}{\displaystyle \underset{f}{}}e_f^2{\displaystyle \frac{\alpha _S(Q^2)}{3\pi }}{\displaystyle _{Q^2}^{\mathrm{}}}{\displaystyle \frac{d\overline{Q}^2}{\overline{Q}^2}}{\displaystyle \frac{Q^2}{\overline{Q}^2}}G(2x,\overline{Q}^2)`$
$`{\displaystyle \underset{f}{}}e_f^2{\displaystyle \frac{\alpha _S(Q^2)}{3\pi }}G(2x,Q^2).`$ (34)
Eqs.(33) and (34) immediately reveal the phenomenological consequences of lifting the DGLAP restrictions in the transverse momenta integration. Indeed, the DGLAP approach respects the following strict inequalities
$$𝜿^2𝐤^2\text{and}𝐤^2Q^2.$$
(35)
As we just saw, removing the first limitation effectively shifted the upper limit in the $`\overline{Q}^2`$ integral to $`\frac{\overline{C}_2}{4}Q^2Q^2`$, while lifting the second constraint led to an additional, purely non-DGLAP contribution. Although both of these corrections lack one leading log-$`Q^2`$ factor they are numerically substantial. As a matter of fact, in section 9 we show that $`\overline{C}_28`$.
The above analysis suggests that the DGLAP and $`𝜿`$-factorization schemes converge logarithmically at large $`Q^2`$. However, in order to reproduce the result (33) and (34) for the full phase space by the conventional DGLAP contribution (24) from the restricted phase space (13) one has to ask for DGLAP gluon density $`G_{pt}(x,Q^2)`$ larger than the integrated GSF in the $`𝜿`$-factorization scheme and the difference may be quite substantial in the domain of strong scaling violations.
### 3.3 The different evolution paths: soft-to-hard diffusion and vice versa
The above discussion of the contributions to the total cross section from the DGLAP and non-DGLAP parts of the phase space can conveniently be cast in the form of the Huygens principle. To the standard DGLAP leading log$`Q^2`$ (LL$`Q^2`$) approximation one only considers the contribution from the restricted part of the available transverse phase space (13). The familiar Huygens principle for the homogeneous DGLAP LL$`Q^2`$ evaluation of parton densities in the $`x_{bj}`$-$`Q^2`$ plane is illustrated in fig. 5a: one starts with the boundary condition $`p(x,Q_0^2)`$ as a function of $`x`$ at fixed $`Q_0^2`$, the evolution paths $`(z,\stackrel{~}{Q}^2)`$ for the calculation of $`p(x,Q^2)`$ shown in fig. 5a are confined to a rectangle $`xz1,Q_0^2\stackrel{~}{Q}^2Q^2`$, the evolution is unidirectional in the sense that there is no feedback on the $`x`$-dependence of $`p(x,Q_1^2)`$ from the $`x`$-dependence of $`p(x,Q_2^2)`$ at $`Q_2^2Q_1^2`$. In fig. 5a we show some examples of evolution paths which are kinematically allowed but neglected in the DGLAP approximation. Starting with about flat or slowly rising $`G(x,Q_0^2)`$ one finds that the larger $`Q^2`$, the steeper the small-$`x`$ rise of $`G(x,Q^2)`$.
At $`x1`$ the DGLAP contribution from the restricted transverse phase space (13) no longer dominates the multiparton production cross sections, the restriction (13) must be lifted and the contribution to the cross section from small $`𝜿_i^2`$ and large $`𝜿_i^2\text{ }>Q^2`$ can no longer be neglected. The Huygens principle for the homogeneous BFKL evolution is illustrated in fig. 5b: one starts with the boundary condition $`(x_0,Q^2)`$ as a function of $`Q^2`$ at fixed $`x_01`$, the evolution paths $`(z,\stackrel{~}{Q}^2)`$ for the calculation of $`p(x,Q^2)`$ are confined to a stripe $`xzx_0`$, in contrast the the unidirectional DGLAP evolution one can say that under BFKL evolution the small-$`x`$ behaviour of $`p(x,Q^2)`$ at large $`Q^2`$ is fed partly by the $`x`$-dependence of soft $`p(x,Q^2)`$ at larger $`x`$ and vice versa. The most dramatic consequence of this soft-to-hard and hard-to-soft diffusion which can not be eliminated is that at very small $`x`$ the $`x`$-dependence of the gluon structure in the soft and hard regions will eventually be the same:
$$\underset{\frac{1}{x}\mathrm{}}{lim}G(x,Q^2)=G(Q^2)\left(\frac{1}{x}\right)^{\mathrm{\Delta }_{𝐈𝐏}}.$$
(36)
The rate of such a hard-to-soft diffusion is evidently sensitive to the infrared regularization of pQCD, the model estimates show that in the HERA range of $`x`$ it is very slow .
## 4 The Ansatz for differential gluon structure function
The major insight into parameterization of DGSF comes from early experience with color dipole phenomenology of small-$`x`$ DIS. In color dipole approach, which is closely related to $`𝜿`$-factorization, the principal quantity is the total cross section of interaction of the $`q\overline{q}`$ color dipole $`𝐫`$ with the proton target
$$\sigma (x,r)=\frac{\pi ^2r^2}{3}\frac{d𝜿^2}{𝜿^2}\frac{4[1J_0(\kappa r)]}{(\kappa r)^2}\alpha _S\left(\mathrm{max}\{𝜿^2,\frac{A}{r^2}\}\right)(x,𝜿^2),$$
(37)
which for very small color dipoles can be approximated by
$$\sigma (x,r)=\frac{\pi ^2r^2}{3}\alpha _S\left(\frac{A}{r^2}\right)G(x,\frac{A}{r^2}),$$
(38)
where $`A10`$ comes from properties of the Bessel function $`J_0(z)`$. The phenomenological properties of the dipole cross section are well understood, for extraction of $`\sigma (x,r)`$ from the experimental data see . The known dipole size dependence of $`\sigma (x,r)`$ serves as a constrain on the possible $`𝜿^2`$-dependence of $`(x,𝜿^2)`$.
As we argued in section 3.2, DGLAP fits are likely to overestimate $`_{hard}(x,𝜿^2)`$ at moderate $`𝜿^2`$. Still, approximation (38) does a good job when the hardness $`A/r^2`$ is very large, and at large $`Q^2`$ we can arguably approximate the DGSF by the direct differentiation of available fits (GRV, CTEQ, MRS, …) to the integrated gluon structure function $`G_{pt}(x,Q^2)`$:
$$_{pt}(x,𝜿^2)\frac{G_{pt}(x,𝜿^2)}{\mathrm{log}𝜿^2}$$
(39)
Hereafter the subscript $`pt`$ serves as a reminder that these gluon distributions were obtained from the pQCD evolution analyses of the proton structure function and cross sections of related hard processes.
The available DGLAP fits are only applicable at $`𝜿^2Q_c^2`$, see table 1 for the values of $`Q_c^2`$, in the extrapolation to soft region $`𝜿^2Q_c^2`$ we are bound to educated guess. To this end recall that perturbative gluons are confined and do not propagate to large distances; recent fits to the lattice QCD data suggest Yukawa-Debye screening of perturbative color fields with propagation/screening radius $`R_c0.27fm`$. Incidentally, precisely this value of $`R_c`$ for Yukawa screened colour fields has been used since 1994 in the very successful color dipole phenomenology of small-$`x`$ DIS . Furthermore, important finding of is a good quantitative description of the rising component of the proton structure function starting with the Yukawa-screened perturbative two-gluon exchange as a boundary condition for the color dipole BFKL evolution.
The above suggests that $`𝜿^2`$ dependence of perturbative hard $`_{hard}(x,𝜿^2)`$ in the soft region $`𝜿^2Q_c^2`$ is similar to the Yukawa-screened flux of photons in the positron, cf. eq. (4), with $`\alpha _{em}`$ replaced by the running strong coupling of quarks $`C_F\alpha _S(𝜿^2)`$ and with factor $`N_c`$ instead of 2 leptons in the positronium, for the early discussion see ,
$$_{pt}^{(B)}(𝜿^2)=C_FN_c\frac{\alpha _s(𝜿^2)}{\pi }\left(\frac{𝜿^2}{𝜿^2+\mu _{pt}^2}\right)^2V_N(𝜿),$$
(40)
Here $`\mu _{pt}=\frac{1}{R_c}=0.75`$ GeV is the inverse Yukawa screening radius and must not be interpreted as a gluon mass; more sophisticated forms of screening can well be considered. Following we impose also the infrared freezing of strong coupling: $`\alpha _S(𝜿^2)0.82`$; recently the concept of freezing coupling has become very popular, for the review see .
The vertex function $`V_N(𝜿)`$ describes the decoupling of soft gluons, $`𝜿\frac{1}{R_p}`$, from color neutral proton and has the same structure as in eq. (5). In the nonrelativistic oscillator model for the nucleon one can relate the two-quark form factor of the nucleon to the single-quark form factor,
$$F_2(\stackrel{}{𝜿},\stackrel{}{𝜿})=F_1\left(\frac{2N_c}{N_c1}𝜿^2\right).$$
(41)
To the extent that $`R_c^2R_p^2`$ the detailed functional form of $`F_2(\stackrel{}{𝜿},\stackrel{}{𝜿})`$ is not crucial, the simple relation (41) will be used also for a more realistic dipole approximation
$$F_1(𝜿^2)=\frac{1}{(1+\frac{𝜿^2}{\mathrm{\Lambda }^2})^2}.$$
(42)
The gluon probed radius of the proton and the charge radius of the proton can be somewhat different and $`\mathrm{\Lambda }1`$ GeV must be regarded as a free parameter. Anticipating the forthcoming discussion of the diffraction slope in vector meson production we put $`\mathrm{\Lambda }=1`$ GeV.
As discussed above, the hard-to-soft diffusion makes the DGSF rising at small-$`x`$ even in the soft region. We model this hard-to-soft diffusion by matching the $`𝜿^2`$ dependence (40) to the DGLAP fit $`_{pt}(x,Q_c^2)`$ at the soft-hard interface $`Q_c^2`$ and assigning to $`_{hard}(x,𝜿^2)`$ in the region of $`𝜿^2Q_c^2`$ the same $`x`$-dependence as shown by the DGLAP fit $`_{pt}(x,Q_c^2)`$, i.e.,
$$_{hard}(x,𝜿^2)=_{pt}^{(B)}(𝜿^2)\frac{_{pt}(x,Q_c^2)}{_{pt}^{(B)}(Q_c^2)}\theta (Q_c^2𝜿^2)+_{pt}(x,𝜿^2)\theta (𝜿^2Q_c^2).$$
(43)
Because the accepted propagation radius $`R_c0.3`$ fm for perturbative gluons is short compared to a typical range of strong interaction, the dipole cross section (37) evaluated with the DGSF (43) would miss an interaction strength in the soft region, for large color dipoles.
In Ref. interaction of large dipoles has been modeled by the non-perturbative, soft mechanism with energy-independent dipole cross section, whose specific form has been driven by early analysis of the exchange by two nonperturbative gluons. More recently several closely related models for $`\sigma _{\mathrm{soft}}(r)`$ have appeared in the literature, see for instance models for dipole-dipole scattering via polarization of non-perturbative QCD vacuum and the model of soft-hard two-component pomeron . In order to reproduce the required interaction strength for large dipoles, we introduce the genuinely soft, nonperturbative component of DGSF which we parameterize as
$$_{soft}^{(B)}(x,𝜿^2)=a_{soft}C_FN_c\frac{\alpha _s(𝜿^2)}{\pi }\left(\frac{𝜿^2}{𝜿^2+\mu _{soft}^2}\right)^2V_N(𝜿).$$
(44)
The principal point about this non-perturbative component of DGSF is that it must not be subjected to pQCD evolution. Thus the arguments about the hard-to-soft diffusion driven rise of perturbative DGSF even at small $`𝜿^2`$ do not apply to the non-perturbative DGSF and we take it energy-independent one. Such a nonperturbative component of DGSF would have certain high-$`𝜿^2`$ tail which should not extend too far. The desired suppression of soft DGSF at large-$`𝜿^2`$ and of hard DGSF (43) at moderate and small $`𝜿^2`$ can be achieved by the extrapolation of the form suggested in
$$(x,𝜿^2)=_{soft}^{(B)}(x,𝜿^2)\frac{𝜿_s^2}{𝜿^2+𝜿_s^2}+_{hard}(x,𝜿^2)\frac{𝜿^2}{𝜿^2+𝜿_h^2}$$
(45)
The above described Ansatz for DGSF must be regarded as a poor man’s approximation. The separation of small-$`𝜿^2`$ DGSF into the genuine nonperturbative component and small-$`𝜿^2`$ tail of the hard perturbative DGSF is not unique. Specifically, we attributed to the latter the same small-$`x`$ rise as in the DGLAP fits at $`Q_c^2`$ while one can not exclude that the hard DGSF has a small $`x`$-independent component. The issue of soft-hard separation must be addressed in dynamical models for infrared regularization of perturbative QCD. As we shall see below, in section 4.4, the soft component of the above described Ansatz is about twice larger than the soft component used in the early color dipole phenomenology .
The $`𝜿`$-factorization formulas (15) and (16) correspond to the full-phase space extension of the LO DGLAP approach at small $`x`$. For this reason our Ansätze for $`_{hard}(x,Q^2)`$ will be based on LO DGLAP fits to the gluon structure function of the proton $`G_{pt}(x,Q^2)`$. We consider the GRV98LO , CTEQ4L, version 4.6 and MRS LO 1998 parameterizations. We take the liberty of referring to our Anzätze for DGSF based on those LO DGLAP input as D-GRV, D-CTEQ and D-MRS parameterizations, respectively.
Our formulas (13), (14) describe the sea component of the proton structure function. Arguably these LL$`\frac{1}{x}`$ formulas are applicable at $`x\text{ }<x_0=1÷310^2`$. At large $`Q^2`$ the experimentally attainable values of $`x`$ are not so small. In order to give a crude idea on finite-energy effects at moderately small $`x`$, we stretch our fits to $`x\text{ }>x_0`$ multiplying the above Ansatz for DGSF by the purely phenomenological factor $`(1x)^5`$ motivated by the familiar large-$`x`$ behaviour of DGLAP parameterizations of the gluon structure function of the proton. We also add to the sea components (13), (14) the contribution from DIS on valence quarks borrowing the parameterizations from the respective GRV, CTEQ and MRS fits. The latter are only available for $`Q^2Q_c^2`$. At $`x\text{ }<10^2`$ this valence contribution is small and fades away rapidly with decreasing $`x`$, for instance see .
## 5 Determination of the differential gluon structure function of the proton
### 5.1 The parameters of DGFS for different DGLAP inputs
Our goal is a determination of small-$`x`$ DGSF in the whole range of $`𝜿^2`$ by adjusting the relevant parameters to the experimental data on small-$`x`$ $`F_{2p}(x,Q^2)`$ in the whole available region of $`Q^2`$ as well as the real photoabsorption cross section. The theoretical calculation of these observables is based on Eqs. (15), (16), (45).
The parameters which we did not try adjusting but borrowed from early work in the color dipole picture are $`R_c=0.27`$ fm, i.e., $`\mu _{pt}=0.75`$ GeV and the frozen value of the LO QCD coupling with $`\mathrm{\Lambda }_{QCD}=0.2`$ GeV:
$$\alpha _S(Q^2)=\mathrm{min}\{0.82,\frac{4\pi }{\beta _0\mathrm{log}\frac{Q^2}{\mathrm{\Lambda }_{QCD}^2}}\}.$$
(46)
We recall that the GRV, MRS and CTEQ fits to GSF start the DGLAP evolution at quite a different soft-to-hard interface $`Q_c^2`$ and diverge quite a lot, especially at moderate and small $`𝜿^2`$. The value of $`Q_c^2`$ is borrowed from these fits and is not a free parameter.
The adjustable parameters are $`\mu _{soft}`$, $`a_{soft}`$, $`m_{u,d}`$, $`𝜿_s^2`$ and $`𝜿_h^2`$. We take $`m_s=m_{u,d}+0.15`$GeV and $`m_c=1.5`$ GeV. The rôle of these parameters is as follows. The quark mass $`m_{u,d}`$ defines the transverse size of the $`q\overline{q}=u\overline{u},d\overline{d}`$ Fock state of the photon, whereas $`\mu _{soft}^2`$ controls the $`r`$-dependence of, and in conjunction with $`a_{soft}`$ sets the scale for, the dipole cross section for large size $`q\overline{q}`$ dipoles in the photon. The both $`m_{u,d}`$ and $`\mu _{soft}`$ have clear physical meaning and we have certain insight into their variation range form the early work on color dipole phenomenology of DIS. The magnitude of the dipole cross section at large and moderately small dipole size depends also on the soft-to-hard interpolation of DGSF, which is somewhat different for different LO DGLAP inputs for $`G_{pt}(x,Q^2)`$. This difference of DGLAP inputs can be corrected for by adjusting $`\mu _{soft}^2`$ and the value of $`𝜿_h^2`$. Because of soft-to-hard and hard-to-soft diffusion the DGLAP fits are expected to fail at small $`x`$, we allow for $`x`$ dependence of $`𝜿_h^2`$.
Evidently, roughly equal values of $`F_{2p}(x,Q^2)`$ can be obtained for somewhat smaller $`(x,Q^2)`$ at the expense of taking smaller $`m_{u,d}`$ and vise versa. Therefore, though the quark mass does not explicitly enter the parameterization for $`(x,𝜿^2)`$, the preferred value of $`m_{u,d}`$ turns out to be correlated with the Ansatz for DGSF, i.e. each particular parameterization of DGSF implies certain $`m_{u,d}`$. In what follows we set $`a_{soft}=2`$, $`𝜿_s^2=3.0`$ GeV<sup>2</sup>, $`m_{u,d}=0.22`$ GeV, so that only $`𝜿_h^2`$ and $`\mu _{soft}`$ varied from one DGLAP input to another, see Table 1. The soft components of the D-GRV and D-CTEQ parameterizations turn out identical. The eye-ball fits are sufficient for the purposes of the present exploratory study. The parameters found are similar to those used in where the focus has been on the description of diffractive DIS.
Table 1. The parameters of differential gluon structure function for different DGLAP inputs.
| | D-GRV | D-MRS | D-CTEQ |
| --- | --- | --- | --- |
| LO DGLAP input | GRV98LO | CTEQ4L(v.4.6) | MRS-LO-1998 |
| $`Q_c^2`$, GeV<sup>2</sup> | 0.895 | 1.37 | 3.26 |
| $`\kappa _h^2`$, GeV<sup>2</sup> | $`\left(1+0.0018\mathrm{log}^4\frac{1}{x}\right)^{1/2}`$ | $`\left(1+0.038\mathrm{log}^2\frac{1}{x}\right)^{1/2}`$ | $`\left(1+0.047\mathrm{log}^2\frac{1}{x}\right)^{1/2}`$ |
| $`\mu _{soft}`$, GeV | 0.1 | 0.07 | 0.1 |
One minor problem encountered in numerical differentiation of all three parameterizations for $`G_{pt}(x,Q^2)`$ was the seesaw $`𝜿^2`$-behavior of the resulting DGSF (39), which was an artifact of the grid interpolation routines. Although this seesaw behavior of DGSF is smoothed out in integral observables like $`G(x,Q^2)`$ or $`F_{2p}(x,Q^2)`$, we still preferred to remove the unphysical seesaw cusps and have smooth DGSF. This was achieved by calculation DGSF from (39) at the center of each interval of the $`Q^2`$-grid and further interpolation of the results between these points. By integration of the so-smoothed $`_{pt}(x,Q^2)`$ one recovers the input $`G_{pt}(x,Q^2)`$. The values of $`Q_c^2`$ cited in Table 1 corresponds to centers of the first bin of the corresponding $`Q^2`$-grid.
### 5.2 The description of the proton structure function $`F_{2p}(x,Q^2)`$
We focus on the sea dominated leading log$`\frac{1}{x}`$ region of $`x<10^2`$. The practical calculation of the proton structure function involves the two running arguments of DGSF: $`x_g`$ and $`𝜿^2`$. We recall that in the standard collinear DGLAP approximation one has $`𝜿^2k^2Q^2`$ and $`x_g2x`$, see eq. (23). Within the $`𝜿`$-factorization one finds that the dominant contribution to $`F_{2p}(x,Q^2)`$ comes from $`M_t^2Q^2`$ with little contribution from $`M_t^2\text{ }>Q^2`$. Because at small $`x_g`$ the $`x_g`$ dependence of $`(x_g,Q^2)`$ is rather steep, we take into account the $`x_gx_{bj}`$ relationship (19). Anticipating the results on effective intercepts to be reported in section 7, we notice that for all practical purposes one can neglect the impact of $`𝜿`$ on the relationship (19), which simplifies greatly the numerical analysis. Indeed, the $`x_g`$ dependence of $`(x_g,𝜿^2)`$ is important only at large $`𝜿^2`$, which contribute to $`F_{2p}(x,Q^2)`$ only at large $`Q^2`$; but the larger $`Q^2`$, the better holds the DGLAP ordering $`𝜿^2k^2,Q^2`$. Although at small to moderate $`Q^2`$ the DGLAP the ordering breaks down, the $`x_g`$ dependence of $`(x_g,𝜿^2)`$ is negligible weak here.
As we shall discuss in more detail below, achieving a good agreement from small to moderate to large $`Q^2`$ is a highly nontrivial task, because strong modification of the soft contribution to $`(x,Q^2)`$ unavoidably echos in the integral quantity $`G_D(x,Q^2)`$ throughout the whole range of $`Q^2`$ and shall affect the calculated structure function from small to moderate to large $`Q^2`$.
The quality of achieved description of the experimental data on the small-$`x`$ proton structure function is illustrated by figs. 67. The data shown include recent HERA data (ZEUS , ZEUS shifted vertex , ZEUS BPC , H1 , H1 shifted vertex ), FNAL E665 experiment and CERN NMC experiment . When plotting the E665 and NMC data, we took the liberty of shifting the data points from the reported values of $`Q^2`$ to the closest $`Q^2`$ boxes for which the HERA data are available. For $`Q^2<Q_c^2=0.9`$ GeV<sup>2</sup> the parameterizations for valence distributions are not available and our curves show only the sea component of $`F_{2p}(x,Q^2)`$, at larger $`Q^2`$ the valence component is included.
At $`x<10^2`$ the accuracy of our D-GRV description of the proton structure function is commensurate to that of the accuracy of standard LO GRV fits. In order not to cram the figures with nearly overlapping curves, we show the results for D-GRV parameterization. The situation with D-CTEQ and D-MRS is very similar, which is seen in fig. 8, where we show on a larger scale simultaneously the results from the D-GRV, D-CTEQ and D-MRS DGSFs for several selected values of $`Q^2`$. Here at large $`Q^2`$ we show separately the contribution from valence quarks. The difference between the results for $`F_{2p}(x,Q^2)`$ for different DGLAP inputs is marginal for all the practical purposes, see also a comparison of the results for $`\sigma ^{\gamma p}`$ for different DGLAP inputs in fig. 9.
### 5.3 Real photoabsorption cross section $`\sigma ^{\gamma p}`$.
In the limiting case of $`Q^2=0`$ the relevant observable is the real photoabsorption cross section $`\sigma ^{\gamma p}`$. Although the Bjorken variable is meaningless at very small $`Q^2`$, the gluon variable $`x_g`$ remains well defined at $`Q^2=0`$, see eq. (19). In fig. 9 we present our results alongside with the results of the direct measurements of $`\sigma ^{\gamma p}`$ and the results of extrapolation of virtual photoabsorption cross sections to $`Q^2=0`$, for the summary of the experimental data see . We emphasize that we reproduce well the observed magnitude and pattern of the energy dependence of $`\sigma ^{\gamma p}`$ in an approach with the manifestly energy-independent soft contribution to the total cross section (which is shown separately in fig. 9). We recall that our parameterizations for $`(x,Q^2)`$ give identical soft cross sections for the GRV and CTEQ inputs (see table 1). The barely visible decrease of $`\sigma _{soft}^{\gamma p}`$ towards small $`W`$ is a manifestation of $`(1x)^5`$ large-$`x`$ behaviour of gluon densities. The extension to lower energies requires introduction of the secondary reggeon exchanges which goes beyond the subject of this study.
In our scenario the energy dependence of $`\sigma ^{\gamma p}`$ is entirely due to the $`x_g`$-dependent hard component $`_{hard}(x_g,Q^2)`$ and as such this rise of the total cross section for soft reaction can be regarded as driven entirely by very substantial hard-to-soft diffusion. Such a scenario has repeatedly been discussed earlier . Time and time again we shall see similar effects of hard-to-soft diffusion and vise versa. Notice that hard-to-soft diffusion is a straightforward consequence of full phase space calculation of partonic cross sections and we do not see any possibility for decoupling of hard gluon contribution from the total cross sections of any soft interaction, whose generic example is the real photoabsorption.
## 6 Properties of differential gluon structure function in the momentum space
### 6.1 Soft/hard decomposition of DGSF
Now we focus on the $`x`$ and $`𝜿^2`$ behavior of the so-determined DGSF starting for the reference with the D-GRV parameterization. The same pattern holds for DGSF based on CTEQ and MRS DGLAP inputs, see below. In figs. 10 and 11 we plot the differential gluon density $`(x_g,Q^2)`$, while in fig. 12 we show the integrated gluon density
$$G_D(x,Q^2)=_0^{Q^2}\frac{d\kappa ^2}{\kappa ^2}(x,𝜿^2).$$
(47)
Here the subscript D is a reminder that the integrated $`G_D(x,Q^2)`$ is derived from DGSF. As such it must not be confused with the DGLAP parameterizations $`G_{pt}(x,Q^2)`$ supplied with the subscript $`pt`$.
Figs. 10 and 11 illustrate the interplay of the nonperturbative soft component of DGSF and perturbative hard contribution supplemented with the above described continuation into $`𝜿^2Q_c^2`$ at various $`x`$ and $`𝜿^2`$. The soft and hard contributions are shown by dashed and dotted lines respectively; their sum is given by solid line.
Apart from the large-$`x`$ suppression factor $`(1x)^5`$ our non-perturbative soft component does not depend on $`x`$. At a not so small $`x=10^2`$ it dominates the soft region of $`𝜿^2\text{ }<1÷2`$ GeV<sup>2</sup>, the hard component takes over at higher $`𝜿^2`$. The soft-hard crossover point is close to $`\mu _{pt}^2`$ but because of the hard-to-soft diffusion it moves with decreasing $`x`$ to a gradually smaller $`Q^2`$.
### 6.2 Soft/hard decomposition of the integrated gluon structure function
The rôle of the soft component if further illustrated by fig. 12 where we show the integrated gluon density (47) and its soft and hard components $`G_{soft}(x,Q^2)`$ and $`G_{hard}(x,Q^2)`$, respectively. The soft contribution $`G_{soft}(x,Q^2)`$ is a dominant feature of the integrated gluon density $`G_D(x,Q^2)`$ for $`Q^2\text{ }<1`$ GeV<sup>2</sup>. It builds up rapidly with $`Q^2`$ and receives the major contribution from the region $`𝜿^20.3÷0.5`$ GeV<sup>2</sup>. Our Ansatz for $`_{soft}(x,𝜿^2)`$ is such that it starts decreasing already at $`𝜿^20.2`$ GeV<sup>2</sup> and vanishes rapidly beyond $`\kappa ^2\text{ }>\kappa _{soft}^2`$, see figs. 10,11. Still the residual rise of the soft gluon density beyond $`Q^20.5`$ GeV<sup>2</sup> is substantial: $`G_{soft}(x,Q^2)`$ rises by about the factor two before it flattens at large $`Q^2`$. We emphasize that $`G_{soft}(Q^2)`$ being finite at large $`Q^2`$ is quite natural — a decrease of $`G_{soft}(Q^2)`$ at large $`Q^2`$ only is possible if $`_{soft}(Q^2)`$ becomes negative valued at large $`Q^2`$, which does not seem to be a viable option.
At moderately small $`x=10^2`$ the scaling violations are still weak and the soft contribution $`G_{soft}(x,Q^2)`$ remains a substantial part, about one half, of integrated GSF $`G_D(x,Q^2)`$ at all $`Q^2`$. At very small $`x\text{ }<10^3`$ the scaling violations in the gluon structure function are strong and $`G_{hard}(x,Q^2)G_{soft}(x,Q^2)`$ starting from $`Q^2`$ 1-2 GeV<sup>2</sup>.
### 6.3 Soft/hard decomposition of the proton structure function $`F_2(x,Q^2)`$
Eqs. (15), (16) define the soft/hard decomposition of the proton structure function. In fig. 13 we show $`F_{2p}^{hard}(x,Q^2)`$ and $`F_{2p}^{soft}(x,Q^2)`$ as a function of $`Q^2`$ for the two representative values of $`x`$. Notice how significance of soft component as a function of $`Q^2`$ rises from fully differential $`(x,Q^2)`$ to integrated $`G_D(x,Q^2)`$ to doubly integrated $`F_{2p}^{soft}(x,Q^2)`$. At a moderately small $`x10^3`$ the soft contribution is a dominant part of $`F_{2p}(x,Q^2)`$, although the rapidly rising hard component $`F_{2p}^{hard}(x,Q^2)`$ gradually takes over at smaller $`x`$.
Notice that not only does $`F_{2p}^{soft}(x,Q^2)`$ not vanish at large $`Q^2`$, but also it rises slowly with $`Q^2`$ as
$$F_{2p}^{soft}(x,Q^2)e_f^2\frac{4G_{soft}(Q^2)}{3\beta _0}\mathrm{log}\frac{1}{\alpha _S(Q^2)}.$$
(48)
Again, the decrease of $`F_{2p}^{soft}(x,Q^2)`$ with $`Q^2`$ would only be possible at the expense of unphysical negative valued $`G_{soft}(Q^2)`$ at large $`Q^2`$.
## 7 DGSF in the $`x`$-space: effective intercepts and hard-to-soft diffusion
It is instructive to look at the change of the $`x`$-dependence from the differential gluon structure function $`(x,Q^2)`$ to integrated gluon structure function $`G_D(x,Q^2)`$ to proton structure function $`F_{2p}(x,Q^2)`$. It is customary to parameterize the $`x`$ dependence of various structure functions by the effective intercept. For instance, for the effective intercept $`\tau _{eff}`$ for differential gluon structure function is defined by the parameterization
$$(x,𝜿^2)\left(\frac{1}{x}\right)^{\tau _{eff}(𝜿^2)}.$$
(49)
One can define the related intercepts $`\tau _{hard}`$ for the hard component $`_{hard}(x,Q^2)`$. Notice, that in our Ansatz $`\tau _{soft}0`$.
The power law (49) is only a crude approximation to the actual $`x`$ dependence of DGSF and the effective intercept $`\tau _{eff}`$ will evidently depend on the range of fitted $`x`$. To be more definitive, for the purposes of the present discussion we define the effective intercept as
$$\tau _{eff}(Q^2)=\frac{\mathrm{log}[(x_2,𝜿^2)/(x_1,𝜿^2)]}{\mathrm{log}(x_1/x_2)}$$
(50)
taking $`x_2=10^5`$ and $`x_1=10^3`$. The effective intercept $`\tau _{hard}(Q^2)`$ is defined by (50) in terms of $`_{hard}(x,Q^2)`$.
One can define the related intercepts $`\lambda _{eff},\lambda _{hard}`$ for the integrated gluon structure function $`G_D(x,Q^2)`$:
$$G_D(x,Q^2)\left(\frac{1}{x}\right)^{\lambda _{eff}(Q^2)}.$$
(51)
In the case of $`F_{2p}(x,Q^2)`$ we define the intercept $`\mathrm{\Delta }(Q^2)`$ in terms of the variable $`\overline{x}`$ defined as
$$\overline{x}=\frac{Q^2+M_V^2}{W^2+Q^2}x_g,$$
(52)
where $`M_V`$ is the mass of the ground state vector meson in the considered flavor channel. Such a replacement allows one to treat on equal footing $`Q^2\text{ }<1`$ GeV<sup>2</sup>, where the formally defined Bjorken variable $`x_{bj}`$ can no longer be interpreted as a lightcone momentum carried by charged partons. For the purposes of the direct comparison with $`\tau (Q^2),\lambda (Q^2)`$ and in order to avoid biases caused by the valence structure function, here we focus on intercepts $`\mathrm{\Delta }_{eff},\mathrm{\Delta }_{hard}`$ for the sea component of the proton structure function $`F_{2p}^{sea}(x,Q^2)`$:
$$F_{2p}^{sea}(x,Q^2)\left(\frac{1}{\overline{x}}\right)^{\mathrm{\Delta }_{eff}(Q^2)}.$$
(53)
The results for the effective intercepts are shown in figs. 14, 15 and 16.
In our simplified hard-to-soft extrapolation of $`_{hard}(x,Q^2)`$ we attribute to $`_{hard}(x,Q^2)`$ at $`Q^2Q_c^2`$ the same $`x`$-dependence as at $`Q^2=Q_c^2`$ modulo to slight modifications for the $`x`$-dependence of $`𝜿_h^2`$. This gives the cusp in $`\tau _{hard}(Q^2)`$ at $`Q^2=Q_c^2`$, i.e., the first derivative of $`\tau _{hard}(Q^2)`$ is discontinuous at $`Q^2=Q_c^2`$.
A comparison of fig. 11 with fig. 12 and further with fig. 13 shows clearly that only in DGSF $`(x,Q^2)`$ the effect of the soft component is concentrated at small $`Q^2`$, in integrated $`G_D(x,Q^2)`$ and especially in the proton structure function $`F_{2p}(x,Q^2)`$ the impact of the soft component extends to much larger $`Q^2`$. The larger the soft contribution, the stronger is the reduction of $`\tau _{eff}`$ from $`\tau _{hard}`$ and so forth, the pattern which is evident from fig. 14a to 14b to 14c, see also figs. 15 and 16.
The change of effective intercepts from differential $`(x,Q^2)`$ to integrated $`G_D(x,Q^2)`$ is straightforward, the principal effect is that $`\lambda _{hard}(Q^2)<\tau _{hard}(Q^2)`$ and $`\lambda _{eff}(Q^2)<\tau _{eff}(Q^2)`$ which reflects the growing importance of soft component in $`G_D(x,Q^2)`$. The change of effective intercepts from $`(x,Q^2)`$ and $`G_D(x,Q^2)`$ to $`F_{2p}(x,Q^2)`$ is less trivial and exhibits two dramatic consequences of the hard-to-soft and soft-to-hard diffusion. If the standard DGLAP contribution (24) were all, then the change from the intercept $`\lambda (Q^2)`$ for integrated gluon density to the intercept $`\mathrm{\Delta }(Q^2)`$ for the proton structure function $`F_{2p}(x,Q^2)`$ would have been similar to the change from $`\tau (Q^2)`$ to $`\lambda (Q^2)`$, i.e., the effective intercept $`\mathrm{\Delta }_{eff}(Q^2)`$ would have been close to zero for $`Q^2\text{ }<1`$ GeV<sup>2</sup>. However, by virtue of the hard-to-soft diffusion phenomenon inherent to the $`𝜿`$-factorization, $`F_{2p}(x,Q^2)`$ receives a contribution from gluons with $`𝜿^2>Q^2`$, which enhances substantially $`\mathrm{\Delta }_{hard}(Q^2)`$ and $`\mathrm{\Delta }_{eff}(Q^2)`$. The net result is that at small to moderately large $`Q^2`$ we find $`\mathrm{\Delta }_{hard}(Q^2)>\lambda _{hard}(Q^2)`$ and $`\mathrm{\Delta }_{eff}(Q^2)>\lambda _{eff}(Q^2)`$. As we emphasized above in sections 5.3, the rise of real photoabsorption cross section is precisely of the same origin.
The second effect is a dramatic flattening of effective hard intercept, $`\mathrm{\Delta }_{hard}(Q^2)`$, over the whole range of $`Q^2`$. For all three DGLAP inputs $`\mathrm{\Delta }_{hard}(Q^2)`$ flattens at approximately the same $`\mathrm{\Delta }_{hard}0.4`$.
The whole set of figs. 1416 also shows that the systematics of intercepts in the hard region of $`Q^2>Q_c^2`$ is nearly identical for all the three DGLAP inputs. In the soft region we have a slight inequality $`\tau _{hard}(𝜿^2)|_{DMRS}>\tau _{hard}(𝜿^2)|_{DGRV}`$, which can be readily attributed to a slight inequality $`Q_c^2(MRS)>Q_c^2(GRV)`$. In the case of CTEQ4L(v.4.6) input the value of $`Q_c^2(CTEQ)`$ is substantially larger then $`Q_c^2(MRS),Q_c^2(GRV)`$. In the range $`Q_c^2(MRS),Q_c^2(GRV)<𝜿^2<Q_c^2(CTEQ)`$ the effective intercept $`\tau _{hard}(𝜿^2)`$ rises steeply with $`𝜿^2`$. This explains a why in the soft region $`\tau _{hard}(𝜿^2)|_{CTEQ}`$ is significantly larger than for the D-GRV and D-MRS parameterizations. The difference among intercepts for the three parameterizations decreases gradually from differential $`(x,𝜿^2)`$ to integrated $`G_D(x,Q^2)`$ gluon density to the proton structure function $`F_{2p}(x,Q^2)`$.
Finally, in fig. 17 we compare our results for $`\mathrm{\Delta }_{eff}(Q^2)`$ with the recent experimental data from ZEUS collaboration . In the experimental fit the range of $`x=[x_{max},x_{min}]`$ varies from point to point, in our evaluation of $`\mathrm{\Delta }_{eff}`$ from eq. (57) we mimicked the experimental procedure taking $`\overline{x}_2=x_{max}`$ and $`\overline{x}_1=x_{min}`$. This explains the somewhat irregular $`Q^2`$ dependence. The experimental data include both sea and valence components. At $`Q^2>Q_c^2(GRV)=0.9`$ GeV<sup>2</sup> we included the valence component of the structure function taking the GRV98LO parameterization. For CTEQ4L(v.4.6) and MRS-LO-1998 the values of $`Q_c^2`$ are substantially larger. However, the valence component is a small correction and we took a liberty of evaluating the valence contribution $`F_{2p}^{val}(x,Q^2)`$ for $`Q_c^2(GRV)<Q^2<Q_c^2(MRS),Q_c^2(CTEQ)`$. The overall agreement with experiment is good. Difference among the three parameterization is marginal and can of course be traced back to figs. 1416.
## 8 How the gluon densities of $`𝜿`$-factorization differ from DGLAP gluon densities
It is instructive also to compare our results for integrated GSF (47) with the conventional DGLAP fit $`G_{pt}(x,Q^2)`$. In fig. 18 we present such a comparison between our integrated D-GRV distribution (the solid curves) and the GRV98LO distribution (the dashed curves). As was anticipated in section 3.2, at very large $`Q^2`$ the two gluon distributions converge. We also anticipated that at small $`x`$ and moderate $`Q^2`$ the DGLAP gluon structure functions $`G_{pt}(x,Q^2)`$ are substantially larger than the result of integration of DGSF, see eq. (47). At $`x=10^5`$ they differ by as much as the factor two-three over a broad range of $`Q^2\text{ }<100`$ GeV<sup>2</sup>. The difference between integrated DGSF and the DGLAP fit decreases gradually at large $`x`$, and is only marginal at $`x=10^2`$.
Recall the substantial divergence of the GRV, MRS and CTEQ gluons structure functions of DGLAP approximation $`G_{pt}(x,Q^2)`$ at small and moderate $`Q^2`$. Contrary to that, the $`𝜿`$-factorization D-GRV, D-CTEQ and D-MRS gluon structure functions $`G_D(x,Q^2)`$ are nearly identical. We demonstrate this property in fig. 19 where we show integrated $`G_D(x,Q^2)`$ and their DGLAP counterparts $`G_{pt}(x,Q^2)`$ for the three parameterizations at two typical values of $`x`$. Because of an essentially unified treatment of the region of $`𝜿^2Q_c^2`$ and strong constraint on DGSF in this region from the experimental data at small $`Q^2`$, such a convergence of D-GRV, D-CTEQ and D-MRS DGSF’s is not unexpected.
One can also compare the effective intercepts for our integrated GSF $`G_D(x,Q^2)`$ with those obtained from DGLAP gluon distributions $`G_{pt}(x,Q^2)`$. Fig.20 shows large scattering of $`\lambda _{eff}^{(pt)}(Q^2)`$ from one DGLAP input to another. At the same time, this divergence of different DGLAP input parameterizations is washed out to a large extent in the $`𝜿`$-factorization description of physical observables (see also 17).
## 9 How different observables probe the DGSF $`(x,Q^2)`$
The issue we address in this section is how different observables map the $`𝜿^2`$ dependence of $`(x_g,𝜿^2)`$. We expand on the qualitative discussion in section 3.2 and corroborate it with numerical analysis following the discussion in . . We start with the two closely related quantities — longitudinal structure function $`F_L(x,Q^2)`$ and scaling violations $`F_2(x,Q^2)/\mathrm{log}Q^2`$ — and proceed to $`F_{2p}(x,Q^2)`$ and the charm structure function of the proton $`F_{2p}^{c\overline{c}}(x,Q^2)`$. This mapping is best studied if in (15) and (16) we integrate first over $`𝐤`$ and $`z`$. In order to focus on the $`𝜿^2`$ dependence we prefer presenting different observables in terms of $`(2x,𝜿^2)`$ and $`G_D(2x,𝜿^2)`$
$`F_L(x,Q^2)={\displaystyle \frac{\alpha _S(Q^2)}{3\pi }}{\displaystyle e_f^2\frac{d𝜿^2}{𝜿^2}\mathrm{\Theta }_L^{(f\overline{f})}(Q^2,𝜿^2)(2x,𝜿^2)},`$ (54)
$`{\displaystyle \frac{F_2(x,Q^2)}{\mathrm{log}Q^2}}={\displaystyle \frac{\alpha _S(Q^2)}{3\pi }}{\displaystyle e_f^2\frac{d𝜿^2}{𝜿^2}\mathrm{\Theta }_2^{(f\overline{f})}(Q^2,𝜿^2)(2x,𝜿^2)}.`$ (55)
In the numerical calculation of $`F_L(x,Q^2)`$ starting from eq. (16) we have $`x_g`$ and $`𝜿^2`$ as the two running arguments of $`(x_g,𝜿^2)`$. As discussed above, the mean value of $`x_g`$ is close to $`2x`$, but the exact relationship depends on $`𝜿^2`$. The $`𝐤,z`$ integration amounts to averaging of $`(x_g,𝜿^2)`$ over certain range of $`x_g`$. The result of this averaging is for the most part controlled by the effective intercept $`\tau _{eff}(𝜿^2)`$:
$$(x_g,𝜿^2)=(2x,𝜿^2)\left(\frac{2x}{x_g}\right)^{\tau _{eff}(𝜿^2)}=r(𝜿^2)(2x,𝜿^2).$$
(56)
Because the derivative of $`\tau _{eff}(𝜿^2)`$ changes rapidly around $`𝜿^2=Q_c^2`$, the rescaling factor $`r(𝜿^2)`$ also has a rapid variation of the derivative at $`𝜿^2=Q_c^2`$, which in the due turn generates the rapid change of derivatives of $`\mathrm{\Theta }_{L,2}^{(f\overline{f})}(Q^2,𝜿^2)`$ around $`𝜿^2=Q_c^2`$. As far as the mapping of differential $`(2x,𝜿^2)`$ is concerned, this is an entirely marginal effect. However, if we look at the mapping of integrated gluon structure function $`G_D(x,Q^2)`$, which is derived from (54), (55) by integration by parts:
$`F_L(x,Q^2)={\displaystyle \frac{\alpha _S(Q^2)}{3\pi }}{\displaystyle e_f^2\frac{d𝜿^2}{𝜿^2}\frac{\mathrm{\Theta }_L^{(f\overline{f})}(Q^2,𝜿^2)}{\mathrm{log}𝜿^2}G_D(2x,𝜿^2)},`$ (57)
$`{\displaystyle \frac{F_2(x,Q^2)}{\mathrm{log}Q^2}}={\displaystyle \frac{\alpha _S(Q^2)}{3\pi }}{\displaystyle e_f^2\frac{d𝜿^2}{𝜿^2}\frac{\mathrm{\Theta }_2^{(f\overline{f})}(Q^2,𝜿^2)}{\mathrm{log}𝜿^2}G_D(2x,𝜿^2)},`$ (58)
then the weight functions $`\mathrm{\Theta }_{2,L}^{(f\overline{f})}(Q^2,𝜿^2)/\mathrm{log}𝜿^2`$ will exhibit a slightly irregular behaviour around $`𝜿^2=Q_c^2`$. Evidently, such an irregularity appears in any region of fast variation of $`\tau _{eff}(𝜿^2)`$; in our simplified model it is somewhat amplified by the cusp-like $`𝜿^2`$ dependence of $`\tau _{eff}(𝜿^2)`$.
Finally, starting from (58) one obtains a useful representation for how the proton structure function $`F_{2p}(x,Q^2)`$ maps the integrated gluon structure function:
$`F_{2p}(x,Q^2)={\displaystyle _0^{Q^2}}{\displaystyle \frac{dq^2}{q^2}}{\displaystyle \frac{\alpha _S(q^2)}{3\pi }}{\displaystyle e_f^2\frac{d𝜿^2}{𝜿^2}\frac{\mathrm{\Theta }_2^{(f\overline{f})}(q^2,𝜿^2)}{\mathrm{log}𝜿^2}G_D(2x,𝜿^2)}`$
$`={\displaystyle \frac{1}{3\pi }}{\displaystyle e_f^2\frac{d𝜿^2}{𝜿^2}W_2^{(f\overline{f})}(Q^2,𝜿^2)\alpha _S(𝜿^2)G_D(2x,𝜿^2)}`$ (59)
In figs. 21 and 22 we show the weight functions $`\mathrm{\Theta }_L`$ and $`\mathrm{\Theta }_2`$. Evidently, for light flavours and very large $`Q^2`$ they can be approximated by step-functions
$$\mathrm{\Theta }_{L,2}^{(f\overline{f})}(Q^2,k^2)\theta (C_{L,2}Q^2𝜿^2),$$
(60)
where the scale factors $`C_L\frac{1}{2}`$ and $`C_22`$ can be readily read from figures, for the related discussion see . Note that the value $`C_22`$ corresponds to $`\overline{C}_28`$ introduced in Section 3.2. Recall that the development of the plateau-like behaviour of $`\mathrm{\Theta }_L`$ and $`\mathrm{\Theta }_2`$ which extends to $`𝜿^2Q^2`$ signals the onset of the leading log$`Q^2`$ approximation. For large $`Q^2`$ in the approximation (60) the $`𝜿^2`$ integration can be carried out explicitly and $`F_L(x,Q^2)G_D(2x,C_LQ^2)`$. Similarly, $`F_2(x,Q^2)/\mathrm{log}Q^2G_D(2x,C_2Q^2)`$, cf. eq. (28).
Still better idea on how $`F_L`$ and scaling violations map the integrated GSF is given by figs. 23,24, where we show results for $`\mathrm{\Theta }(f\overline{f})_L/\mathrm{log}𝜿^2`$ and $`W_2^{(f\overline{f})}`$. The first quantity is sharply peaked at $`\kappa ^2C_LQ^2`$. The second quantity visibly develops a plateau at large $`Q^2`$. As can be easily seen, scaling violations do receive a substantial contribution from the beyond-DGLAP region of $`𝜿^2>Q^2`$.
Because of the heavy mass, the case of the charm structure function $`F_{2p}^{c\overline{c}}(x,Q^2)`$ is somewhat special. Figs. 23 and 24 show weak sensitivity of $`F_{2p}^{c\overline{c}}(x,Q^2)`$ to a soft component of $`(x_{bj},𝜿^2)`$ which has an obvious origin: long-wavelength soft gluons with $`\kappa \text{ }<m_c`$ decouple from the color neutral $`c\overline{c}`$ Fock state of the photon which has a small transverse size $`\text{ }<\frac{1}{m_c}`$. Our results for $`F_{2p}^{c\overline{c}}(x,Q^2)`$ are shown in fig. 25, the agreement with the recent precision experimental data from ZEUS is good.
## 10 Summary and outlook
We present the first parameterization of differential gluon structure function $`(x,Q^2)`$ of the proton inherent to the $`𝜿`$-factorization approach to small-$`x`$ DIS. The form of the parameterization is driven by color gauge invariance constraints for soft $`Q^2`$, early ideas from color dipole phenomenology on the necessity of nonperturbative soft mechanism for interaction of large color dipoles and by matching to the derivative of familiar DGLAP fits $`G_{pt}(x,Q^2)`$. The latter condition is not imperative, though, and can be relaxed; in this exploratory study we simply wanted to take advantage of the insight on $`G_{pt}(x,Q^2)`$ from early DGLAP approximation studies on scaling violations. The parameters of $`(x,Q^2)`$ have been tuned to the experimental data on $`F_{2p}`$ in the low-$`x`$ ($`x\text{ }<0.01`$) domain and throughout the entire $`Q^2`$ region as well as on real photoabsorption cross section $`\sigma _{tot}^{\gamma p}`$. Differential gluon structure function $`(x,Q^2)`$ is the principal input for pQCD calculation of many diffractive processes and we anticipate that the consistent use of our parameterizations shall reduce the uncertainties of calculations of cross section of such processes as diffractive DIS into vector mesons and continuum.
Our results allow to address several interesting issues. First, our Ansätze for $`(x,Q^2)`$ have been so constructed as to ensure the convergence of $`G_D(x,Q^2)`$ — the integral of $`(x,Q^2)`$ — to the corresponding large $`Q^2`$ DGLAP input $`G_{pt}(x,Q^2)`$. We notice that both gluon distributions provide a comparable description of the same set of the experimental data on the proton structure function, the only difference being that in the $`𝜿`$-factorization we lift the DGLAP limitation on the transverse phase space of quarks and antiquarks. We find very slow convergence of, and numerically very large difference between, the $`𝜿`$-factorization distribution $`G_D(x,Q^2)`$ and the DGLAP fit $`G_{pt}(x,Q^2)`$. As anticipated, the divergence of the two distributions is especially large at small-$`x`$ and persists even in the hard region up to $`Q^210÷100`$ GeV<sup>2</sup> at $`x=10^5`$. We interpret this divergence as a signal of breaking of the DGLAP approximation which arguably gets poorer at smaller $`x`$. The second finding is a numerically very strong impact of soft gluons on the integrated gluons structure function $`G_D(x,Q^2)`$ and the proton structure function $`F_{2p}(x,Q^2)`$. It is not unexpected in view of the early work on color dipole phenomenology of small-$`x`$ DIS, but the evaluation of the soft component of integrated gluon structure function is reported here for the first time. In conjunction with the strong departure of the $`𝜿`$-factorization distribution $`G_D(x,Q^2)`$ from the DGLAP fit $`G_{pt}(x,Q^2)`$ it serves as an warning against unwarranted application of DGLAP evolution at $`Q^2`$ in the range of several GeV<sup>2</sup>.
The phenomenologically most interesting finding is the anatomy of the rising component of the proton structure function from the Regge theory point of view. We notice that effective intercepts $`\tau _{hard}(Q^2)`$ and $`\lambda _{hard}`$ for hard components of the differential and integrated gluon distributions are lively functions of $`Q^2`$ which vary quite rapidly with $`Q^2`$ from $`0.1`$ at small $`Q^2`$ to $`0.6`$ at $`Q^210^3`$ GeV<sup>2</sup>. In the Regge theory language this evidently implies that hard component of neither $`(x,Q^2)`$ nor $`G_D(x,Q^2)`$ is dominated by a single Regge pole exchange and a contribution from several hard Regge poles with broad spacing of intercepts is called upon. However, an approximately flat $`Q^2`$ dependence of $`\mathrm{\Delta }_{hard}(Q^2)`$ shows that the hard component of the proton structure function can well be approximated by a single Regge pole with intercept $`\mathrm{\Delta }_{hard}0.4`$. Such a scenario in which a contribution of subleading BFKL-Regge poles to $`F_{2p}(x,Q^2)`$ is suppressed dynamically because of the nodal properties of gluon distributions for subleading BFKL-Regge poles has been encountered earlier in the color dipole BFKL approach . The intercept $`\mathrm{\Delta }_{hard}(Q^2)`$ found in the present analysis is remarkably close the intercept of the leading BFKL-Regge pole $`\mathrm{\Delta }_{𝐈𝐏}=0.4`$ found in the color dipole approach in 1994 . For the related two-pomeron phenomenology of DIS see also . From the point of view of $`𝜿`$-factorization, the hard-to-soft diffusion is a unique mechanism by which an approximate constancy of $`\mathrm{\Delta }_{hard}(Q^2)`$ derives from a very rapidly changing $`\tau _{hard}(Q^2)`$. Fourth, the same hard-to-soft diffusion provides a mechanism for the rise of the real photoabsorption cross section $`\sigma ^{\gamma p}`$ in a model with the manifestly energy independent soft cross section. We emphasize that the hard-to-soft diffusion is a generic phenomenon and we do not see any possibility for decoupling of hard contribution from photoabsorption at $`Q^2=0`$.
We restricted ourselves to a purely phenomenological determination of differential gluon distributions from the experimental data on $`F_{2p}(x,Q^2)`$ which is sufficient for major applications of the $`𝜿`$-factorization technique. Whether the so-determined hard components of $`(x,Q^2)`$ and $`G_D(x,Q^2)`$ do satisfy the dynamical evolution equations and what is the onset of DGLAP regime will be addressed elsewhere.
IPI wishes to thank Prof. J.Speth for his hospitality at Forschungszentrum Jülich. This work was partly supported by the grant INTAS 97-30494.
|
warning/0004/hep-th0004132.html
|
ar5iv
|
text
|
# QED in external field with space-time uniform invariants. Exact solutions
## I Introduction
Exact solutions of the relativistic wave equations (Klein-Gordon or Dirac equation) in an external electromagnetic field as well as Green functions of those equations are very important in QED with such a field. Special complete sets of the exact solutions and special kinds of the Green functions allow one to calculate different quantum effects, for example, particles scattering, pair creation and so on, in zero order with respect to the radiative corrections, but taking the interaction with the external field into account exactly . They may serve also as a basis for the perturbation theory with respect to the radiative interaction, in which the external field is taken into account exactly (so called generalized Furry picture) . In four-dimensional space-time the above exact solutions were studied for different configurations of the external field in numerous papers and books (see, for example, . Among all the configurations of the external electromagnetic fields, which admit the exact solutions in $`d=4`$, one is especially important due to the fact that it corresponds to a wide class of real physical situations. It is a combination of constant uniform electric and magnetic fields and plane wave field. The exact solutions can be found if there exist a reference frame in which the constant electric and magnetic fields, and the wave 3-vector are collinear. First, the exact solutions of Klein-Gordon and Dirac equation in such a configuration were found in . Some of the Green functions of scalar field in the same configuration were studied in the papers . Using those exact solutions and Green functions different quantum effects were calculated . Calculation of Green functions of spinor field was not present in the literature. Moreover, lately, field theoretical models in dimensions different from $`d=4`$ attract attention due to various reasons. One can mention here models in $`2+1`$ dimensions which probably describe planar physical phenomenon, and models in $`d>4`$ dimensions in connection with Kaluza-Klein ideas (see e.g. . That is why in the present article we are going to study exact solutions and Green functions (both scalar and spinor case) in $`d`$-dimensional QED with an external electromagnetic field, which may be considered as a generalization of the above mentioned special configuration in four dimensions. At the same time we also present new results in four dimensions. A general definition of such a configuration valid for any dimensions of space-time is the following: an external electromagnetic field with constant and homogeneous field invariants. Also, an important motivation to consider exact solutions and Green functions in such a configuration of the external electromagnetic field is the fact that formally problems in some configurations of an external gravitational field lead to the same kind of equations, so that the former exact solutions can be used in the latter case after some simple identifications. Such a reduction may be done, for example, for De Sitter and FRW metrics , for four fermionic models in an external field and so on.
In a sense, the paper can be considered as a continuation and generalization of the one , where only an electric-like external field was considered. The latter paper can help the reader to better understand the logic of the exact solutions constructed in this more complicated present case. Thus, here we sometimes omit technical details and explanations which can be extracted from the above mentioned previous paper.
The paper is organized as follows: In Sec. II we describe first the external electromagnetic field under consideration. Then, we present the so called in and out complete sets of exact solutions of Dirac and Klein-Gordon equations in the external field. The cases of even and odd dimensions have to be considered separately, they are essentially different. We attract attention to the asymmetry of the quasienergy spectrum, which appears in odd dimensions. We spend enough time to prove strictly the $`in`$ and $`out`$ classification of the exact solutions as well as to prove the completeness and orthogonality relations. Then in Sec. III we construct different Green functions in the form of sums over the exact solutions and present them in the form of contour integrals over the Fock-Schwinger proper time. Among them are the causal Green function, $`inin`$ and $`outout`$ Green functions, the commutator function and so on. One ought to say that the spinor case is treated first, even in $`d=4`$ dimensions. As physical applications we consider in Sec. IV the calculations of different quantum effects related to the vacuum instability in the external field. For example, we present mean values of particles created from the vacuum, the probability for the vacuum remaining a vacuum, the effective action, the expectation values of the current and energy-momentum tensor. We calculate the latter quantities by means of both $`in`$ and $`out`$ sets of the exact solutions.
## II Complete sets of exact solutions
The Dirac equation in an external electromagnetic field with potentials $`A_\mu (x)`$ in $`d`$ dimensions has the form ($`\mathrm{}=c=1`$)
$$\left(𝒫_\mu \gamma ^\mu m\right)\psi (x)=0,𝒫_\mu =i_\mu qA_\mu (x),$$
(1)
where $`\psi (x)`$ is a $`2^{[\frac{d}{2}]}`$-component column, $`\gamma ^\mu `$ are $`\gamma `$-matrices in $`d`$ dimensions ,
$`[\gamma ^\mu ,\gamma ^\nu ]_+=2\eta ^{\mu \nu },\eta ^{\mu \nu }=\mathrm{diag}(\underset{d}{\underset{}{1,1,1,\mathrm{}}}),d=D+1,`$
and $`x=(x^\mu )=(x^0,𝐱),𝐱=(x^i),\mu =0,1,\mathrm{},D,i=1,\mathrm{},D`$. The time-independent scalar product of the solutions of the Dirac equation may be chosen in the conventional form
$$(\psi ,\psi ^{})=\overline{\psi }(x)\gamma ^0\psi ^{}(x)𝑑𝐱.$$
(2)
As usual, it is convenient to present $`\psi (x)`$ in the form
$$\psi (x)=\left(𝒫_\mu \gamma ^\mu +m\right)\varphi (x).$$
(3)
Then the functions $`\varphi `$ have to obey the squared Dirac equation in $`d`$ dimensions,
$$\left(𝒫^2m^2\frac{q}{2}\sigma ^{\mu \nu }_{\mu \nu }\right)\varphi (x)=0,_{\mu \nu }=_\mu A_\nu (x)_\nu A_\mu (x),\sigma ^{\mu \nu }=\frac{i}{2}[\gamma ^\mu ,\gamma ^\nu ].$$
(4)
To construct the above mentioned generalized Furry picture in QED with an external field one has to find special sets of exact solutions of Eq. (1), namely, two complete and orthonormal sets of solution: $`\left\{{}_{\pm }{}^{}\psi _{\left\{n\right\}}^{}(x)\right\}`$ which describes particles (+) and antiparticles ($``$) in the initial time instant ($`x^0\mathrm{})`$, and $`\left\{{}_{}{}^{\pm }\psi _{\left\{n\right\}}^{}(x)\right\}`$ which describes particles (+) and antiparticles ($``$) in the final time instant ($`x^0+\mathrm{}).`$ According to the general approach such solutions obey the following asymptotic conditions
$`H_{o.p.}(x^0){}_{\zeta }{}^{}\psi _{\left\{n\right\}}^{}(x)={}_{\zeta }{}^{}\epsilon {}_{\zeta }{}^{}\psi _{\left\{n\right\}}^{}(x),,\mathrm{sgn}{}_{\zeta }{}^{}\epsilon =\zeta ,x^0\mathrm{},`$ (5)
$`H_{o.p.}(x^0){}_{}{}^{\zeta }\psi _{\left\{l\right\}}^{}(x)={}_{}{}^{\zeta }\epsilon {}_{}{}^{\zeta }\psi _{\left\{l\right\}}^{}(x),\mathrm{sgn}{}_{}{}^{\zeta }\epsilon =\zeta ,x^0+\mathrm{},`$ (6)
where $`\zeta ,\left\{n\right\}`$ and $`\zeta ,\left\{l\right\}`$ are complete sets of quantum numbers which characterize solutions $`{}_{\zeta }{}^{}\psi _{\left\{n\right\}}^{}(x)`$ and $`{}_{}{}^{\zeta }\psi _{\left\{l\right\}}^{}(x)`$ respectively, $`H_{o.p.}=\gamma ^0(m\gamma ^i𝒫_i)`$ is a one-particle Dirac Hamiltonian in convenient external field gauge $`A_0(x)=0`$; $`{}_{}{}^{+}\epsilon `$, $`{}_{+}{}^{}\epsilon `$ are particle quasi-energies and $`|^{}\epsilon |`$ and $`|_{}\epsilon |`$ are antiparticles quasi-energies. All the information about the processes of particles scattering and creation by an external field (in zeroth order with respect to the radiative corrections) can be extracted from the decomposition coefficients (matrices) $`G({}_{\zeta }{}^{}|{}_{}{}^{\zeta ^{}})`$,
$${}_{}{}^{\zeta }\psi (x)={}_{+}{}^{}\psi (x)G({}_{+}{}^{}|{}_{}{}^{\zeta })+{}_{}{}^{}\psi (x)G({}_{}{}^{}|{}_{}{}^{\zeta }).$$
(7)
The matrices $`G({}_{\zeta }{}^{}|{}_{}{}^{\zeta ^{}})`$ obey the following relations,
$`G({}_{\zeta }{}^{}|{}_{}{}^{+})G({}_{\zeta }{}^{}|{}_{}{}^{+})^{}+G({}_{\zeta }{}^{}|{}_{}{}^{})G({}_{\zeta }{}^{}|{}_{}{}^{})^{}=𝐈,`$ (8)
$`G({}_{+}{}^{}|{}_{}{}^{+})G({}_{}{}^{}|{}_{}{}^{+})^{}+G({}_{+}{}^{}|{}_{}{}^{})G({}_{}{}^{}|{}_{}{}^{})^{}=0,`$ (9)
where $`𝐈`$ is the identity matrix.
Let us describe the external electromagnetic field in which we are going to construct the exact solutions. Such a field is a generalization of the corresponding field in $`d=4`$, which is a combination of the constant uniform field with the plane wave field and which admits there exact solutions. One can describe such a field in $`d=4`$ in arbitrary reference frame saying that both its field invariants do not depend on space-time coordinates. In $`d>4`$ there exist, generally speaking, more independent field invariants. Besides $`I_1=1/2_{\mu \nu }^{\mu \nu }`$, those are all possible independent invariant combinations which may be constructed from the field tensor $`_{\mu \nu }`$ and Levi- Civita tensor. It is easy to see that there exist $`\left[d/2\right]`$ such invariants. We define the external field under consideration in arbitrary dimensions in the same manner, all its field invariants have to be constant and uniform. One can see that such a field is a combination of a constant uniform field $`F_{\mu \nu }`$ and a plane-wave field $`f_{\mu \nu }(nx)`$
$$_{\mu \nu }=F_{\mu \nu }+f_{\mu \nu }(nx),$$
(10)
where $`n_\mu `$ is an real isotropic vector, $`n_\mu n^\mu =0`$. It is an eigenvector of the tensor $`F_{\mu \nu },`$ $`F_{\mu \nu }n^\nu =n_\mu ,,`$ and $`f_{\mu \nu }(nx)`$ is a transverse field with respect to $`n_\mu ,`$ $`n^\mu f_{\mu \nu }(nx)=f_{\mu \nu }(nx)n^\nu =0.`$ If all of the invariants are equal to zero (it is only possible for $`d>2`$), only a plane-wave field configuration is possible. If some of the invariants are not equal to zero, then a constant, uniform component $`F_{\mu \nu },`$ is not zero. The field (10) is free. Such a field is of special interest due to the fact that QED with a free classical field can be treated as exact QED (without external fields) but with some special (coherent) initial photon states .
If the eigenvalue $``$ is not zero, the external field violates the vacuum stability (creating particles). One can find an inertial frame where the matrix $`_{\mu \nu }`$ has a simple form
$`F_{\mu \nu }=F_{\mu \nu }^{}+F_{\mu \nu }^{},`$ (11)
$`F_{\mu \nu }^{}=E\left(\delta _\mu ^0\delta _\nu ^D\delta _\mu ^D\delta _\nu ^0\right),`$ (12)
$`f_{\mu \nu }\left(nx\right)={\displaystyle \underset{k=1}{\overset{D1}{}}}\left(n_\mu \delta _\nu ^kn_\nu \delta _\mu ^k\right)\dot{f}_k\left(x_{}\right),\dot{f}_k(x_{})={\displaystyle \frac{df_k(x_{})}{dx_{}}},`$ (13)
$`n_\mu =\delta _\mu ^0\delta _\mu ^D,x_{}=nx=x^0x^D.`$ (14)
If $`d>2`$ is even then
$`F_{\mu \nu }^{}={\displaystyle \underset{j=1}{\overset{\left(d2\right)/2}{}}}H_j(\delta _\mu ^{2j}\delta _\nu ^{2j1}\delta _\nu ^{2j}\delta _\mu ^{2j1}),`$
and if $`d=2`$ the fields $`F_{\mu \nu }^{}`$ and $`f_{\mu \nu }\left(nx\right)`$ are absent. In the case $`d`$ is odd, and $`0,`$
$`F_{\mu \nu }^{}={\displaystyle \underset{j=1}{\overset{\left(d3\right)/2}{}}}H_j(\delta _\mu ^{2j}\delta _\nu ^{2j1}\delta _\nu ^{2j}\delta _\mu ^{2j1})\mathrm{if}\text{ }d>3,\mathrm{and}F_{\mu \nu }^{}=0\text{if }d=3.`$
The same formula is valued if $`=0`$, but at least one of the imaginary eigenvalues of $`F_{\mu \nu }`$ is zero. Here $`f_k(x_{})`$ are arbitrary functions of $`x_{}`$. If $`=0`$, and all imaginary eigenvalues of $`F_{\mu \nu }`$ are not equal to zero (all field invariants are not equal to zero) we have
$`E=0,F_{\mu \nu }^{}={\displaystyle \underset{j=1}{\overset{\left(d1\right)/2}{}}}H_j(\delta _\mu ^{2j}\delta _\nu ^{2j1}\delta _\nu ^{2j}\delta _\mu ^{2j1}),f_k\left(x_{}\right)=0\mathrm{for}\mathrm{all}k.`$
In the reference frame under consideration $`=E.`$
To realize the external electromagnetic field of the above form we select the following potentials:
$`A_\mu \left(x\right)=A_\mu ^E\left(x\right)+A_\mu ^H\left(x\right)+f_\mu \left(x_{}\right),`$ (15)
$`A_\mu ^E\left(x\right)=Ex^0\delta _\mu ^D,A_i^H=H_jx_{i+1}\delta _{i,2j1},j=1,\mathrm{},[d/2]1,i=1,\mathrm{},D1,`$ (16)
$`f_\mu \left(x_{}\right)=0\mathrm{if}\mu =0,D.`$ (17)
Below we present linearly independent sets of solutions of the squared Dirac equation, which correspond to the particles in the initial time instance and to the antiparticles in the final time instant:
$`{}_{+}{}^{}\varphi _{p_{},n,\xi ,r}^{}(x)`$ $`=`$ $`{}_{+}{}^{}\varphi _{p_{},n,r}^{}(x^0,x^D){}_{+}{}^{}u_{\xi ,r}^{}(x_{})\varphi _{n,r}(x_{}),`$ (18)
$`\varphi _{n,r}(x_{})`$ $`=`$ $`\varphi _{p_1,n_1}(x_1,x_2)\varphi _{p_3,n_2}(x_3,x_4)\mathrm{}\varphi _{p_{d3},n_{(d2)/2}}(x_{D2},x_{D1}),\text{if }d\text{ is even},`$ (19)
$`\varphi _{n,r}(x_{})`$ $`=`$ $`\varphi _{p_1,n_1}(x_1,x_2)\varphi _{p_3,n_2}(x_3,x_4)\mathrm{}\varphi _{p_{d2},n_{(d3)/2}}(x_{D3},x_{D2})`$ (20)
$`(2\pi )^{1/2}\mathrm{exp}(ip_{D1}x^{D1})\text{if }d\text{ is odd }\mathrm{and}H_{(d1)/2}=0,`$ (21)
$`{}_{+}{}^{}\varphi _{p_{},n,r}^{}(x^0,x^D)`$ $`=`$ $`(4\pi )^{1/2}\mathrm{exp}\{{\displaystyle \frac{i}{2}}(qE(x_{}^2/2x_D^2)p_{}x_++\lambda \mathrm{ln}(\stackrel{~}{\pi }_{}))+`$ (23)
$`{}_{+}{}^{}J(x_{}){}_{+}{}^{}K_{}^{\mu }(x_{})\pi _\mu )\},`$
$`{}_{+}{}^{}J\left(x_{}\right)`$ $`=`$ $`{\displaystyle \frac{1}{2qE}}{\displaystyle _{\mathrm{}}^\pi _{}}qf\left({\displaystyle \frac{p_{}\tau }{qE}}\right)\left[qf\left({\displaystyle \frac{p_{}\tau }{qE}}\right)+qF_+^{}K\left({\displaystyle \frac{p_{}\tau }{qE}}\right)\right]\tau ^1𝑑\tau ,`$ (24)
$`{}_{+}{}^{}K\left(x_{}\right)`$ $`=`$ $`{\displaystyle \frac{1}{qE}}{\displaystyle _{\mathrm{}}^\pi _{}}\mathrm{exp}\left\{{\displaystyle \frac{F}{E}}\mathrm{ln}{\displaystyle \frac{\pi _{}}{\tau }}\right\}qf\left({\displaystyle \frac{p_{}\tau }{qE}}\right)\tau ^1𝑑\tau ,`$ (25)
$`{}_{+}{}^{}u_{\xi ,r}^{}(x_{})=`$ $`\left({\displaystyle \frac{1}{2\pi _{}}}(1+\gamma ^0\gamma ^D)+{\displaystyle \frac{1}{2}}(1\gamma ^0\gamma ^D)+{\displaystyle \frac{1}{2\pi _{}}}(\gamma ^0\gamma ^D)\gamma qf(x_{})\right)v_{\xi ,r},`$ (27)
$`x_+=x^0+x^D,\stackrel{~}{\pi }_{}=\pi _{}/\sqrt{qE},`$
where $`p_{},`$ $`n=(n_1,n_2,\mathrm{},n_{[d/2]1};p_1,p_3,\mathrm{},p_{2[(d1)/2]1}),\xi `$ and $`r=(r_1,r_2,\mathrm{},r_{[d/2]1})`$ is a complete set of quantum numbers. Among them $`p_{},`$ $`p_j`$ are momenta of the continuous spectrum, $`n_j`$ are integer quantum numbers, $`\xi =\pm 1`$ and $`r_j=\pm 1`$ are spin quantum numbers; the momentum $`p_{}`$ is the eigenvalue of the operator $`2i\frac{}{x_+}`$; if $`qE>0`$ is chosen then the signs $``$ assigned to the functions $`{}_{+}{}^{}\varphi _{p_{},n,\xi ,r}^{}(x)`$ are matched with those of the kinetic momentum $`\pi _{}=p_{}qEx_{}`$ at $`x_{}\pm \mathrm{}`$, and
$`x_{}^\mu =0\mathrm{if}\mu =0,D,x_{}^\mu =x^\mu \mathrm{if}\mu =1,\mathrm{},D1,`$ (28)
$`\pi _\mu =0\mathrm{if}\mu =0,D,\pi _\mu =i{\displaystyle \frac{}{x^\mu }}qA_\mu ^H(x)\mathrm{if}\mu =1,\mathrm{},D1,`$ (29)
$`qE\lambda =m^2+{\displaystyle \underset{j=1}{\overset{[d/2]1}{}}}\omega _j+\omega _0,\omega _0=\{\begin{array}{cc}0,\hfill & d\text{is even}\hfill \\ p_{d2}^2,\hfill & d\text{is odd},\hfill \end{array}`$ (32)
$`\omega _j=\{\begin{array}{cc}|qH_j|(2n_j+1r_j),\hfill & H_j0\hfill \\ p_{2j1}^2+p_{2j}^2,\hfill & H_j=0\hfill \end{array}.`$ (35)
Each function $`\varphi _{p_{2j1},n_j}(x_{2j1},x_{2j})`$ obeys the following equations
$`(\pi _{2j1}^2+\pi _{2j}^2\omega _j)\varphi _{p_{2j1},n_j}(x_{2j1},x_{2j})`$ $`=`$ $`0,`$
$`\left(\pi _{2j1}p_{2j1}\right)\varphi _{p_{2j1},n_j}(x_{2j1},x_{2j})`$ $`=`$ $`0.`$
If $`H_j0,`$ a solution of these equations is
$`\varphi _{p_{2j1},n_j}(x_{2j1},x_{2j})=`$
$`\left({\displaystyle \frac{\sqrt{|qH_j|}}{2^{n_j+1}\pi ^{\frac{3}{2}}n_j!}}\right)^{1/2}\mathrm{exp}\left\{ip_{2j1}x^{2j1}{\displaystyle \frac{|qH_j|}{2}}\left(x^{2j}+{\displaystyle \frac{p^{2j1}}{qH_j}}\right)^2\right\}_{n_j}\left[\sqrt{|qH_j|}\left(x^{2j}+{\displaystyle \frac{p^{2j1}}{qH_j}}\right)\right],`$
where $`_{n_j}(x)`$ are the Hermite polynomial with integer $`n_j=0,1,\mathrm{}`$. If $`H_j=0,`$ the discrete quantum numbers $`n_j`$ have to be replaced by the momenta $`p_{2j},`$ and the corresponding function has the form
$`\varphi _{p_{2j1},p_{2j}}(x_{2j1},x_{2j})=(2\pi )^1\mathrm{exp}\left\{i\left(p_{2j1}x^{2j1}+p_{2j}x^{2j}\right)\right\}.`$
The symbol $`\mathrm{ln}\left(\stackrel{~}{\pi }_{}\right)`$ means the principal branch of the logarithm, $`\mathrm{ln}\left(\stackrel{~}{\pi }_{}\right)=\mathrm{ln}\left|\stackrel{~}{\pi }_{}\right|+i\pi \mathrm{\Theta }\left(\pm \stackrel{~}{\pi }_{}\right),`$ while the integration paths in the $`\tau `$-plane, as well as the arguments $`\pi _{},`$ are shown in FIG.1 and FIG.2 for the functions $`{}_{+}{}^{}K\left(x_{}\right),`$ $`{}_{+}{}^{}J\left(x_{}\right)`$ and $`{}_{}{}^{}K\left(x_{}\right),`$ $`{}_{}{}^{}J\left(x_{}\right)`$ respectively.
One assumes the functions $`f_k(x_{})`$ obey the requirement that $`{}_{+}{}^{}K\left(x_{}\right)`$ and $`{}_{+}{}^{}J\left(x_{}\right)`$ are analytic functions, and in particular, the functions $`f_k(x_{})`$ vanish at $`x_{}\pm \mathrm{}`$ quite rapidly.
The solution has a different form if $`d`$ is odd, $`=0`$, and all the imaginary eigenvalues of $`F_{\mu \nu }`$ are not equal to zero (in this case a plane-wave field is absent),
$`{}_{+}{}^{}\varphi _{n,\xi ,r}^{}(x)`$ $`=`$ $`{}_{+}{}^{}\varphi _{n,r}^{}(x^0)\varphi _{n,r}(𝐱)v_{\xi ,r},`$ (36)
$`\varphi _{n,r}(𝐱)`$ $`=`$ $`\varphi _{p_1,n_1}(x_1,x_2)\varphi _{p_3,n_2}(x_3,x_4)\mathrm{}\varphi _{p_{d2},n_{(d1)/2}}(x_{D1},x_D),`$ (37)
$`{}_{+}{}^{}\varphi _{n,r}^{}(x^0)`$ $`=`$ $`c\mathrm{exp}\left(\pm i\left|{}_{}{}^{}\epsilon _{nr}^{}\right|x^0\right),\left|{}_{}{}^{}\epsilon _{nr}^{}\right|=\sqrt{m^2+{\displaystyle \underset{j=1}{\overset{(d1)/2}{}}}\omega _j},`$ (38)
where $`c`$ is a normalization constant.
Here $`v_{\xi ,r}`$ are some constant orthonormal spinors, $`v_{\xi ,r}^{}v_{\xi ,r^{}}=\delta _{r,r^{}}.`$ Equation (4) allows one to subject these spinors to some supplementary conditions,
$$\mathrm{\Xi }_\pm v_{1,r}=0,\mathrm{\Xi }_\pm =\frac{1}{2}(1\pm \gamma ^0\gamma ^D),\mathrm{rank}\mathrm{\Xi }_\pm =J_{(d)}=2^{[\frac{d}{2}]1};$$
(39)
$$R_jv_{\xi ,r}=r_jv_{\xi ,r},d4,R_j=i\gamma ^{2j1}\gamma ^{2j}.$$
(40)
If $`d=2,3`$, the independent quantum number $`r`$ does not appear.
One can verify the solutions of the Dirac equation with different $`\xi `$, namely, $`\left(𝒫_\mu \gamma ^\mu +m\right){}_{+}{}^{}\varphi _{p_{},n,+1,r}^{}(x)`$ and $`\left(𝒫_\mu \gamma ^\mu +m\right){}_{+}{}^{}\varphi _{p_{},n,1,r}^{}(x),`$ or $`\left(𝒫_\mu \gamma ^\mu +m\right){}_{}{}^{}\varphi _{p_{},n,+1,r}^{}(x)`$ and $`\left(𝒫_\mu \gamma ^\mu +m\right){}_{}{}^{}\varphi _{p_{},n,1,r}^{}(x)`$ are linearly dependent for each sign ”$`+`$” or ”$``$”. It means, in fact, that the spin projections of a particle (+) and an antiparticle ($``$) can take on only $`J_{(d)}`$ values. Thus, to construct the complete sets it is sufficient to use only the following solutions:
$${}_{+}{}^{}\psi _{p_{},n,r}^{}(x)=\left(𝒫_\mu \gamma ^\mu +m\right){}_{+}{}^{}\varphi _{p_{},n,+1,r}^{}(x).$$
(41)
In particular, if $`d=2,3`$, there is only one spin projection, with only one spinor $`v_{+1,r}=\frac{1}{\sqrt{2}}\left(\begin{array}{c}1\\ i\end{array}\right)`$. To get the case where $`E=0`$ one has to consider a limit in the solution $`{}_{+}{}^{}\psi _{p_{},n,r}^{}(x)`$, such that $`\left(qE\right)^1\mathrm{ln}\left(\stackrel{~}{\pi }_{}\right)x_{}/p_{}`$, and $`p_{}>0`$ for $`\{{}_{+}{}^{}\psi _{p_{},n,r}^{}(x)\}`$ whereas $`p_{}<0`$ for $`\{{}_{}{}^{}\psi _{p_{},n,r}^{}(x)\}`$.
Using the solutions (36) we may construct the Dirac spinors in a slightly different way,
$`{}_{+}{}^{}\psi _{n,\overline{r}}^{}(x)`$ $`=`$ $`\left(𝒫_\mu \gamma ^\mu +m\right)_+^{}\varphi _{n,\overline{r}}(x),`$ (42)
$`{}_{+}{}^{}\varphi _{n,\overline{r}}^{}(x)`$ $`=`$ $`\left({}_{+}{}^{}\varphi _{n,+1,r}^{}(x)+\mathrm{sgn}\left(qH_{(d1)/2}\right)_+^{}\varphi _{n,1,r}(x)\right),\overline{r}=(r_1,r_{2,}\mathrm{},r_{(d3)/2},\mathrm{sgn}\left(qH_{(d1)/2}\right)),`$ (43)
where
$`R_j{}_{+}{}^{}\varphi _{n,\overline{r}}^{}(x)=\overline{r}_j{}_{+}{}^{}\varphi _{n,\overline{r}}^{}(x),j=1,\mathrm{\hspace{0.33em}2},\mathrm{},(d1)/2,`$
and $`{}_{+}{}^{}\varphi _{n,\overline{r}}^{}(x)`$ are eigenvectors of the time-independent operator $`m^2𝒫^i𝒫_i+\frac{q}{2}\sigma ^{\mu \nu }F_{\mu \nu }.`$
The solutions $`{}_{+}{}^{}\psi _{n,\overline{r}}^{}(x)`$ from (42) describe particles (+) and $`{}_{}{}^{}\psi _{n,\overline{r}}^{}(x)`$ describe antiparticles ($``$) in any time instant. They form a complete and orthonormal set of solutions. There is an interesting asymmetry in energy spectrums of the such particles and antiparticles in odd dimensions. If all quantum numbers $`n_j=0`$, and $`r_j=\mathrm{sgn}\left(qH_j\right),`$ then $`\gamma 𝐏_+^{}\psi _{n,\overline{r}}(x)=0,`$ and the ground state of the particle has energy $`{}_{+}{}^{}\epsilon _{0}^{}=m`$, but the energy of the ground state of the antiparticle is dependent of the magnetic field and different: $`\left|{}_{}{}^{}\epsilon _{0}^{}\right|=\sqrt{m^2+2\mathrm{min}\left|qH_j\right|}`$.
In the $`d=4`$ case the solutions similar to (41) were found first in . In case $`E=0`$ they coincide with ones in , and if $`H=0`$ they coincide with well-known Wolkov form . In the case $`E=0,`$ the solutions form a complete and orthonormal set, moreover $`{}_{+}{}^{}\psi _{p_{},n,r}^{}(x)`$ describe particles and $`{}_{}{}^{}\psi _{p_{},n,r}^{}(x)`$ describe antiparticles in any time instant. It is no longer if $`E0`$. However, solutions (41) can be used to construct new complete sets which can be classified by the correct way.
One can form two complete and orthonormal sets of the solutions:$`\left\{{}_{\pm }{}^{}\psi _{p_{},n,r}^{}(x)\right\}`$ and $`{}_{}{}^{\pm }\psi _{p_{},n,r}^{}(x)`$ using (41) and additional sets
$`{}_{}{}^{+}\psi _{p_{},n,r}^{}(x)`$ $`=`$ $`\mathrm{\Theta }(\pi _{})\left({}_{+}{}^{}\psi (x)G({}_{+}{}^{}|{}_{}{}^{+})\right)_{p_{},n,r},`$ (44)
$`{}_{}{}^{}\psi _{p_{},n,r}^{}(x)`$ $`=`$ $`\mathrm{\Theta }(\pi _{})\left({}_{}{}^{}\psi (x)G({}_{}{}^{}|{}_{}{}^{})^{}\right)_{p_{},n,r},`$ (45)
where $`G({}_{\zeta }{}^{}|{}_{}{}^{\zeta ^{}})`$ obey the relations (9) and, in particular, $`G({}_{+}{}^{}|{}_{}{}^{})`$ are decomposition coefficients of $`{}_{}{}^{}\psi _{p_{},n,r}^{}(x)`$ solutions in $`{}_{+}{}^{}\psi _{p_{},n,r}^{}(x)`$ solutions,
$$G({}_{+}{}^{}|{}_{}{}^{})_{p_{},n,r,p_{}^{},n^{},r^{}}=({}_{+}{}^{}\psi _{p_{},n,r}^{},{}_{}{}^{}\psi _{p_{}^{},n^{},r^{}}^{}).$$
(46)
The corresponding Klein-Gordon solutions follow from (18) by the replacement $`u_{\xi ,r}(x_{})=\mathrm{exp}\left\{\frac{1}{2}\mathrm{ln}\left(\stackrel{~}{\pi }_{}\right)\right\}.`$
The orthonormality of all the solutions can be verified by using the following integral transformations:
$`{}_{\pm }{}^{}\psi _{p_{},n,r}^{}(x)=(2\pi qE)^{1/2}{\displaystyle _{\mathrm{}}^+\mathrm{}}M^{}(p_D,p_{}){}_{\pm }{}^{}\psi _{p_D,n,r}^{}(x)𝑑p_D,`$ (47)
$`{}_{\pm }{}^{}\psi _{p_D,n,r}^{}(x)=(2\pi qE)^{1/2}{\displaystyle _{\mathrm{}}^+\mathrm{}}M(p_D,p_{}){}_{\pm }{}^{}\psi _{p_{},n,r}^{}(x)𝑑p_{},`$ (48)
$`M(p_D,p_{})=\mathrm{exp}\left\{{\displaystyle \frac{i}{4qE}}((p_{}2p_D)^22(p_D)^2)\right\},`$ (49)
$`{\displaystyle _{\mathrm{}}^+\mathrm{}}M^{}(p_D,p_{}^{})M(p_D,p_{})𝑑p_D=2\pi qE\delta (p_{}p_{}^{}).`$ (50)
The same relation is valid for functions with ($`\pm `$) indices above.
The saddle points $`\pi _{}=2(qEx_0p_D)`$ give the main contribution to the integrals (48) with $`{}_{+}{}^{}\psi _{p_{},n,r}^{}(x)`$ functions at $`x_0\pm \mathrm{}`$ (this was first found in Ref. for the $`d=4`$ case without magnetic field). Since the plane wave vanishes in $`{}_{+}{}^{}\psi _{p_{},n,r}^{}(x)`$ at $`\pi _{}\pm \mathrm{},`$ relations (48) reduce at $`x_0\pm \mathrm{}`$ to the space-time uniform field case and can be presented by use of formulas from as follows
$`{}_{+}{}^{}\psi _{p_D,n,r}^{}(x)`$ $`=`$ $`\left(𝒫_\mu \gamma ^\mu +m\right){}_{+}{}^{}\varphi _{p_D,n,r}^{}(x),x_0\pm \mathrm{},`$ (51)
$`{}_{+}{}^{}\varphi _{p_D,n,r}^{}(x)`$ $`=`$ $`{}_{+}{}^{}\varphi _{n,r}^{}(x^0)\mathrm{exp}\left(ip_Dx^D\right)\varphi _{n,r}(x_{})v_{+1,r},`$ (52)
$`{}_{+}{}^{}\varphi _{n,r}^{}(x^0)`$ $`=`$ $`CD_{\nu 1}[\pm (1i)\xi ],`$ (53)
$`\xi =\left(qEx^0p_D\right)/\sqrt{qE},\nu =i\lambda /2,C=(4\pi qE)^{1/2}\mathrm{exp}\{(\pi /2+i\mathrm{ln}2)\lambda /4\}(i),`$
where the function $`\varphi _{n,r}(x_{})`$ was defined in (18), and, $`D_\nu (z)`$ is the Weber parabolic cylinder function . Such solutions were discussed in . One can now verify the orthonormality relations of the sets$`\left\{{}_{+}{}^{}\psi _{p_D,n,r}^{}(x)\right\}`$ and $`\left\{{}_{}{}^{}\psi _{p_D,n,r}^{}(x)\right\}.`$ Using transformation (47) one gets the orthonormality relations of sets $`\left\{{}_{+}{}^{}\psi _{p_{},n,r}^{}(x)\right\}`$ and $`\left\{{}_{}{}^{}\psi _{p_{},n,r}^{}(x)\right\}`$ as well. Using the explicit forms of solutions (41), (45), and relations (9) one can derive the following relations,
$`{}_{}{}^{+}\psi (x)G({}_{+}{}^{}|{}_{}{}^{+})^{}`$ $`=`$ $`{}_{+}{}^{}\psi (x){}_{}{}^{}\psi (x)G({}_{+}{}^{}|{}_{}{}^{})^{},`$ (54)
$`{}_{}{}^{}\psi (x)G({}_{}{}^{}|{}_{}{}^{})`$ $`=`$ $`{}_{}{}^{}\psi (x)_+\psi (x)G({}_{+}{}^{}|{}_{}{}^{}).`$ (55)
By means of the latter one can get the orthonormality relations for both sets of the solutions,
$`\left({}_{\zeta }{}^{}\psi _{p_{},n,r}^{},_\zeta ^{}\psi _{p_{}^{},n^{},r^{}}\right)=\delta _{\zeta \zeta ^{}}\delta _{rr^{}}\delta _{nn^{}}\delta (p_{}p_{}^{}),`$ (56)
$`\left({}_{}{}^{\zeta }\psi _{p_{},n,r}^{},^\zeta ^{}\psi _{p_{}^{},n^{},r^{}}\right)=\delta _{\zeta \zeta ^{}}\delta _{rr^{}}\delta _{nn^{}}\delta (p_{}p_{}^{}),`$ (57)
where $`\zeta ,\zeta ^{}=\pm `$, $`\delta _{nn^{}}`$ is the Kronecker symbol for the discrete spectrum and the $`\delta `$-function for the continuous one. Here $`r=r^{}=+1`$ if $`d=2,3.`$
To solve a problem of the ($`\pm `$) classification of the solutions one needs to study their asymptotic behavior at $`x_0\pm \mathrm{}.`$ From the asymptotic forms (51) it follows that the asymptotic of the quasienergies of these solution: $`{}_{+}{}^{}\epsilon =qEx^0,`$ is positive and $`{}_{}{}^{}\epsilon =qEx^0`$ is negative. Thus, the solutions $`{}_{+}{}^{}\psi _{p_D,n,r}^{}(x)`$ describe particles at $`x_0\mathrm{},`$ and the solutions $`{}_{}{}^{}\psi _{p_D,n,r}^{}(x)`$ describe antiparticles at $`x_0+\mathrm{}.`$ Since the solutions $`{}_{}{}^{}\psi _{p_{},n,r}^{}(x)`$ are orthogonal to $`{}_{+}{}^{}\psi _{p_{},n,r}^{}(x),`$ they describe antiparticles at $`x_0\mathrm{},`$ and solutions $`{}_{}{}^{+}\psi _{p_{},n,r}^{}(x)`$ describe particles at $`x_0+\mathrm{}`$ since they are orthogonal to $`{}_{}{}^{}\psi _{p_{},n,r}^{}(x).`$ One can verify this by taking into account that the main contribution to integrals (48) at $`x_0\pm \mathrm{}`$ for $`{}_{}{}^{+}\psi _{p,r}^{}(x)`$ is given by point $`\pi _{}=0`$. In this limit the contribution of the plane wave does not depend on $`x`$ and the results of transformation (48) are proportional to a superposition of the solutions in a constant uniform field ,
$`{}_{}{}^{+}\psi _{p_D,n,r}^{}(x)`$ $`=`$ $`{\displaystyle \underset{n,r}{}}a_{n^{}r^{}}\left(𝒫_\mu \gamma ^\mu +m\right){}_{}{}^{+}\varphi _{p_D,n^{},r^{}}^{}(x),x_0\pm \mathrm{},`$ (58)
$`{}_{}{}^{+}\varphi _{p_D,n,r}^{}(x)`$ $`=`$ $`{}_{}{}^{+}\varphi _{n,r}^{}(x^0)\mathrm{exp}\left(ip_Dx^D\right)\varphi _{n,r}(x_{})v_{+1r},`$ (59)
$`{}_{}{}^{+}\varphi _{n,r}^{}(x^0)`$ $`=`$ $`CD_\nu [\pm (1+i)\xi ],`$ (60)
where $`a_{nr\text{ }}`$ are some coefficients dependent on the plane-wave form. From the asymptotic representations (58) one can see that the asymptotic of the quasienergies of these solutions: $`{}_{}{}^{+}\epsilon =qEx^0`$ is positive and $`{}_{}{}^{}\epsilon =qEx^0`$ is negative.
Both sets $`\{{}_{\pm }{}^{}\psi _{p_{},n,r}^{}(x)\}`$ and $`\{{}_{}{}^{\pm }\psi _{p_{},n,r}^{}(x)\}`$ are orthonormal and complete at any time instant. The form of the commutation function, which will be present in the next section, can serve as direct proof of the last statement. In $`d=4`$ such complete sets of the solutions were first found in . If the plane wave is absent and $`E=0`$ one can get the solutions with defined energies from $`{}_{+}{}^{}\psi _{p_D,n,r}^{}(x)`$ (51) considering the next limit in $`{}_{+}{}^{}\varphi _{n,r}^{}(x^0)`$:
$`{}_{+}{}^{}\varphi _{n,r}^{}(x^0)`$ $``$ $`\left[{}_{}{}^{}\epsilon _{nr}^{}(_{}\epsilon _{nr}+p_D)\right]^1\mathrm{exp}(\pm i|{}_{}{}^{}\epsilon _{nr}^{}\left|x^0\right),`$
$`\left|{}_{}{}^{}\epsilon _{nr}^{}\right|`$ $`=`$ $`\sqrt{m^2+{\displaystyle \underset{j=1}{\overset{\left[(d1)/2\right]}{}}}\omega _j+\omega _0^{}},\omega _0^{}=\{\begin{array}{c}p_D^2,d\mathrm{is}\mathrm{even}\\ 0,d\mathrm{is}\mathrm{odd}\end{array},`$
where $`\omega _j`$ is defined in (32). In this case one can choose $`{}_{}{}^{+}\psi _{p_D,n,r}^{}(x)=_+\psi _{p_D,n,r}(x)`$ and $`{}_{}{}^{}\psi _{p_D,n,r}^{}(x)=^{}\psi _{p_D,n,r}(x).`$
Let us select some important properties of the solution. One can see that the matrix elements $`G({}_{\zeta }{}^{}|{}_{}{}^{\zeta ^{}})`$ are diagonal with respect to continuous quantum numbers and spin quantum numbers,
$$G({}_{\zeta }{}^{}|{}_{}{}^{\zeta ^{}})_{p_{},n,r,p_{}^{},n^{},r^{}}=\delta _{rr^{}}\delta (p_{}p_{}^{})\delta (p_1p_1^{})\mathrm{}\delta (p_{2\left[(d1)/2\right]1}p_{2\left[(d1)/2\right]1}^{})g({}_{\zeta }{}^{}|{}_{}{}^{\zeta ^{}})_{nn^{}}.$$
(62)
The solutions $`{}_{+}{}^{}\psi _{p_{},n,r}^{}(x)`$ satisfy the orthonormality conditions on the null-plane,
$${}_{+}{}^{}\overline{\psi }_{p_{},n,r}^{}(x)\mathrm{\Xi }_{}{}_{+}{}^{}\psi _{p_{}^{},n^{},r^{}}^{}(x)𝑑x_+𝑑x^1\mathrm{}𝑑x^{D1}=\delta _{rr^{}}\delta _{nn^{}}\delta (p_{}p_{}^{}),\pi _{}>0.$$
(63)
Since $`{}_{}{}^{+}\psi _{p_{},n,r}^{}(x)=0`$ for $`\pi _{}<0`$ and relations (55) take place, one gets the following representations for $`G\left({}_{+}{}^{}|_{}^{}\right)G({}_{+}{}^{}|{}_{}{}^{})^{}`$ and $`G\left({}_{+}{}^{}|_{}^{}\right)^{}G({}_{+}{}^{}|{}_{}{}^{})`$ matrices:
$`\left(G({}_{+}{}^{}|{}_{}{}^{})G({}_{+}{}^{}|{}_{}{}^{})^{}\right)_{p_{},n,r,p_{}^{},n^{},r^{}}`$ $`=`$ $`{\displaystyle {}_{+}{}^{}\overline{\psi }_{p_{},n,r}^{}(x)\mathrm{\Xi }_{}{}_{+}{}^{}\psi _{p_{}^{},n^{},r^{}}^{}(x)𝑑x_+𝑑x^1\mathrm{}𝑑x^{D1}},\pi _{}<0,`$ (64)
$`\left(G({}_{+}{}^{}|{}_{}{}^{})^{}G({}_{+}{}^{}|{}_{}{}^{})\right)_{p_{},n,r,p_{}^{},n^{},r^{}}`$ $`=`$ $`{\displaystyle {}_{}{}^{}\overline{\psi }_{p_{},n,r}^{}(x)\mathrm{\Xi }_{}{}_{}{}^{}\psi _{p_{}^{},n^{},r^{}}^{}(x)𝑑x_+𝑑x^1\mathrm{}𝑑x^{D1}},\pi _{}>0.`$ (65)
Thus one can calculate $`{}_{+}{}^{}𝒟=g\left({}_{+}{}^{}|_{}^{}\right)g({}_{+}{}^{}|{}_{}{}^{})^{}`$ and $`{}_{}{}^{}𝒟=g\left({}_{+}{}^{}|_{}^{}\right)^{}g({}_{+}{}^{}|{}_{}{}^{})`$ matrices using the following integrals with the solutions of the squared Dirac equation:
$${}_{+}{}^{}𝒟_{nn^{}}^{}={}_{+}{}^{}\varphi _{p_{},n,+1,r}^{}(x){}_{+}{}^{}\varphi _{p_{},n^{},+1,r}^{}(x)𝑑x^2𝑑x^4\mathrm{}𝑑x^{2\left[(d1)/2\right]},\pi _{}<0.$$
(66)
## III Green functions
The perturbation theory with respect to the radiative interaction for the matrix elements of the processes also has the usual Feynman structure in an external field creating pairs . The Feynman diagrams have to be calculated by means of the causal propagator
$$S^c(x,x^{})=c_v^1i<0,out|T\psi (x)\overline{\psi }(x^{})|0,in>,c_v=<0,out|0,in>,$$
(67)
where $`\psi (x)`$ is quantum spinor field in the generalized Furry representation, satisfying the Dirac equation (1), $`|0,in>`$ and $`|0,out>`$ are the initial and the final vacuum in this representation, and $`c_v`$ is the vacuum to vacuum transition amplitude. The propagator $`S^c(x,x^{})`$ obeys the equation
$$\left(𝒫_\mu \gamma ^\mu m\right)S^c(x,x^{})=\delta ^{(d)}(xx^{}),$$
(68)
and is a Green function of the equation. Another important singular function is the commutation function
$$S(x,x^{})=i[\psi (x),\overline{\psi }(x^{})]_+.$$
(69)
It obeys the homogeneous Dirac equation (1) and the initial condition
$$S(x,x^{})|_{x_0=x_0^{}}=i\gamma ^0\delta (𝐱𝐱^{}).$$
(70)
The commutation function $`S(x,x^{})`$ is at the same time the propagation function of the Dirac equation, i.e. it connects solutions of the equation in two different time instants.
QED with unstable vacuum has a number of peculiarities. Thus, for instance, in the calculation of the expectation values and the total probabilities Green functions of different types from (67) appear :
$`S_{in}^c(x,x^{})`$ $`=`$ $`i<0,in|T\psi (x)\overline{\psi }(x^{})|0,in>,`$ (71)
$`S_{in}^{}(x,x^{})`$ $`=`$ $`i<0,in|\psi (x)\overline{\psi }(x^{})|0,in>,`$ (72)
$`S_{in}^+(x,x^{})`$ $`=`$ $`i<0,in|\overline{\psi }(x^{})\psi (x)|0,in>,`$ (73)
$`S_{in}^{\overline{c}}(x,x^{})`$ $`=`$ $`i<0,in|\psi (x)\overline{\psi }(x^{})T|0,in>,`$ (74)
$`S_{out}^c(x,x^{})`$ $`=`$ $`i<0,out|T\psi (x)\overline{\psi }(x^{})|0,out>,`$ (75)
where the symbol of the $`T`$-product acts on both sides: it orders the field operators to the right of its and antiorders them to the left. The functions $`S_{in}^c(x,x^{}),`$ $`S_{out}^c(x,x^{})`$ obey Eq. (68), $`S^{}(x,x^{})`$ satisfies Eq. (1) and $`S_{in}^{\overline{c}}(x,x^{})`$ obeys the equation
$$\left(𝒫_\mu \gamma ^\mu m\right)S_{in}^{\overline{c}}(x,x^{})=\delta ^{(d)}(xx^{}).$$
(76)
As well, all these different kinds of the Green functions are used to represent various matrix elements of operators of the current and energy-momentum tensor, and effective action beginning with zeroth order with respect to radiative interaction.
Solutions (41) and (45) with quantum number $`p_{}`$ are especially adapted to calculate all the Green functions. One can express the Green functions via solutions (41) and (45) :
$`S^c(x,x^{})`$ $`=`$ $`\theta \left(x_0x_0^{}\right)S^{}(x,x^{})\theta \left(x_0^{}x_0\right)S^+(x,x^{}),`$ (77)
$`S(x,x^{})`$ $`=`$ $`S^{}(x,x^{})+S^+(x,x^{}),`$ (78)
$`S^{}(x,x^{})`$ $`=`$ $`i{\displaystyle _{\mathrm{}}^+\mathrm{}}dp_{}{\displaystyle \underset{nr\{n_j^{}\}}{}}{}_{}{}^{+}\psi _{p_,n,r}^{}\left(x\right)g\left({}_{+}{}^{}|_{}^{+}\right)_{nn^{}}^1{}_{+}{}^{}\overline{\psi }_{p_{},n^{},r}^{}\left(x^{}\right),`$ (79)
$`S^+(x,x^{})`$ $`=`$ $`i{\displaystyle _{\mathrm{}}^+\mathrm{}}dp_{}{\displaystyle \underset{nr\{n_j^{}\}}{}}{}_{}{}^{}\psi _{p_{},n,r}^{}\left(x\right)\left[g\left({}_{}{}^{}|_{}^{\mathrm{\_}}\right)^1\right]_{nn^{}}^{}{}_{}{}^{}\overline{\psi }_{p_{},n^{},r}^{}\left(x^{}\right),`$ (80)
$`S_{in}^c(x,x^{})`$ $`=`$ $`\theta \left(x_0x_0^{}\right)S_{in}^{}(x,x^{})\theta \left(x_0^{}x_0\right)S_{in}^+(x,x^{}),`$ (81)
$`S_{in}^{\overline{c}}(x,x^{})`$ $`=`$ $`\theta \left(x_0^{}x_0\right)S_{in}^{}(x,x^{})\theta \left(x_0x_0^{}\right)S_{in}^+(x,x^{}),`$ (82)
$$S_{in}^{}(x,x^{})=i_{\mathrm{}}^+\mathrm{}𝑑p_{}\underset{nr}{}{}_{\pm }{}^{}\psi _{p_{},n,r}^{}\left(x\right)_\pm \overline{\psi }_{p_{},n,r}\left(x^{}\right),$$
(83)
$`S_{out}^c(x,x^{})`$ $`=`$ $`\theta \left(x_0x_0^{}\right)S_{out}^{}(x,x^{})\theta \left(x_0^{}x_0\right)S_{out}^+(x,x^{}),`$ (84)
$`S_{out}^{}(x,x^{})`$ $`=`$ $`i{\displaystyle _{\mathrm{}}^+\mathrm{}}𝑑p_{}{\displaystyle \underset{nr}{\overset{\pm }{}}}\psi _{p_{},n,r}\left(x\right)^\pm \overline{\psi }_{p_{},n,r}\left(x^{}\right).`$ (85)
where all $`p_j^{}=p_j\text{}`$the symbol $`_{nr}`$ means the summation over all discrete quantum numbers $`n_j,r_j`$ and the integration over all continuous $`p_j`$, and the symbol $`_{\{n_j\}}`$ means the summation over all discrete quantum numbers $`n_j`$ only. Using the relations between the Green functions and between the matrices $`G({}_{\zeta }{}^{}|{}_{}{}^{\zeta ^{}})`$ one can present the functions $`S^{},`$ $`S_{in}^{}`$ and $`S_{out}^{}`$ as follows
$`\pm S^{}(x,x^{})=S^c(x,x^{})\pm \theta ((x_0x_0^{}))S(x,x^{}),`$ (86)
$`\pm S_{in}^{}(x,x^{})=S_{in}^c(x,x^{})\pm \theta ((x_0x_0^{}))S(x,x^{}),`$ (87)
$`\pm S_{out}^{}(x,x^{})=S_{out}^c(x,x^{})\pm \theta ((x_0x_0^{}))S(x,x^{}),`$ (88)
$`S_{in}^c(x,x^{})=S^c(x,x^{})S^a(x,x^{}),`$ (89)
$`S_{out}^c(x,x^{})=S^c(x,x^{})S^p(x,x^{}),`$ (90)
$`S^a(x,x^{})=i{\displaystyle _{\mathrm{}}^+\mathrm{}}dp_{}{\displaystyle \underset{nr\{n_j^{}\}}{}}{}_{}{}^{}\psi _{p_{},n,r}^{}(x)\left[g(_+|^{})g(_{}|^{})^1\right]_{nn^{}}^{}{}_{+}{}^{}\overline{\psi }_{p_{},n^{},r}^{}(x^{}),`$ (91)
$`S^p(x,x^{})=i{\displaystyle _{\mathrm{}}^+\mathrm{}}dp_{}{\displaystyle \underset{nr\{n_j^{}\}}{\overset{+}{}}}\psi _{p_{},n,r}(x)\left[g(_+|^+)^1g(_+|^{})\right]_{nn^{}}{}_{}{}^{}\overline{\psi }_{p_{},n^{},r}^{}(x^{}).`$ (92)
To calculate all the types of Green functions it is sufficient to take sums in $`S^\pm (x,x^{})`$ and $`S^{a,p}(x,x^{})`$ only. That will be done below.
It follows from (45)
$`S^\pm (x,x^{})`$ $`=`$ $`{\displaystyle _{\mathrm{}}^+\mathrm{}}\mathrm{\Theta }\left(\pi _{}\right)_+^{}Y(x,x^{},p_{})𝑑p_{},`$ (93)
$`{}_{+}{}^{}Y(x,x^{},p_{})`$ $`=`$ $`i{\displaystyle \underset{nr}{}}{}_{+}{}^{}\psi _{p_,n,r}^{}\left(x\right)_+^{}\overline{\psi }_{p_{},n,r}\left(x^{}\right).`$ (94)
Also, taking into account Eqs. (55) one gets
$`S^a(x,x^{})`$ $`=`$ $`{\displaystyle _{\mathrm{}}^+\mathrm{}}\mathrm{\Theta }\left(\pi _{}\right)_+Y(x,x^{},p_{})𝑑p_{},`$ (95)
$`S^p(x,x^{})`$ $`=`$ $`{\displaystyle _{\mathrm{}}^+\mathrm{}}\mathrm{\Theta }\left(\pi _{}\right)^{}Y(x,x^{},p_{})𝑑p_{},`$ (96)
Due to the fact that $`{}_{+}{}^{}\psi _{p_,n,r}^{}\left(x\right)`$ and $`{}_{}{}^{}\psi _{p_,n,r}^{}\left(x\right)`$ have similar form (41), the sums in (94) can be taken in the same way. Since
$`{\displaystyle \underset{r}{}}v_{+1,r}v_{+1,r}^+=\mathrm{\Xi }_+\mathrm{if}d>3,\mathrm{and}v_{+1}v_{+1}^+=\mathrm{\Xi }_+\mathrm{if}d=2,3,`$
the summation over the spin quantum numbers can be done in (94) to get
$`{}_{+}{}^{}Y(x,x^{},p_{})`$ $`=`$ $`\left[\gamma ^0+\left(\gamma _{}𝒫_{}+m\right){\displaystyle \frac{1}{\pi _{}}}\right]\mathrm{\Xi }_+\mathrm{exp}\left(i{\displaystyle \frac{q}{2}}\sigma ^{\mu \nu }F_{\mu \nu }^{}{}_{+}{}^{}a\right)\left[\gamma ^0+\left(\gamma _{}𝒫_{}^{}{}_{}{}^{}+m\right){\displaystyle \frac{1}{\pi _{}^{}}}\right]`$ (98)
$`\gamma ^0\mathrm{exp}\left\{{\displaystyle \frac{1}{2}}\left[\mathrm{ln}\left(\stackrel{~}{\pi }_{}\right)+\left(\mathrm{ln}\left(\stackrel{~}{\pi }_{}^{}\right)\right)^{}\right]\right\}\sqrt{qE}_+^{}\stackrel{~}{f}^{(0)}(x,x^{},p_{}),`$
$`{}_{+}{}^{}\stackrel{~}{f}_{}^{(0)}(x,x^{},p_{})`$ $`=`$ $`i{\displaystyle \underset{n}{}}{}_{+}{}^{}\phi _{p_,n,r}^{}\left(x\right)_+^{}\phi _{p_{},n,r}^{}\left(x^{}\right),`$ (99)
$`{}_{+}{}^{}\phi _{p_,n,r}^{}\left(x\right)`$ $`=`$ $`\mathrm{exp}\left\{{\displaystyle \frac{1}{2}}\mathrm{ln}\left(\stackrel{~}{\pi }_{}\right)\right\}{}_{+}{}^{}\varphi _{p_{},n,r}^{}(x^0,x^D)\varphi _{n,r}(x_{}),`$ (100)
$`{}_{+}{}^{}a`$ $`=`$ $`\left(2qE\right)^1\left[\left(\mathrm{ln}\left(\stackrel{~}{\pi }_{}^{}\right)\right)^{}\mathrm{ln}\left(\stackrel{~}{\pi }_{}\right)\right],`$ (101)
$`\stackrel{~}{\pi }_{}^{}`$ $`=`$ $`\stackrel{~}{\pi }_{}+\sqrt{qE}y_{},y_\mu =x_\mu x_\mu ^{},`$ (102)
$`𝒫_\mu `$ $`=`$ $`0\mathrm{if}\mu =0,D,𝒫_\mu =𝒫_\mu \mathrm{if}k=1,\mathrm{},D1,`$ (104)
$`𝒫_{}^{}{}_{\mu }{}^{}=i{\displaystyle \frac{}{x^{}^\mu }}qA_\mu (x^{}).`$
It is convenient to make the replacement
$`u=\left(2qE\right)^1\left[\left(\mathrm{ln}\left(\stackrel{~}{\pi }_{}^{}\right)\right)^{}\mathrm{ln}\left(\tau \right)\right]`$
in the functions $`{}_{+}{}^{}K(x_{})`$ and $`{}_{+}{}^{}J(x_{})`$, and the one
$`u=\left(2qE\right)^1\left[\left(\mathrm{ln}\left(\stackrel{~}{\pi }_{}^{}\right)\right)^{}\mathrm{ln}\tau \right]`$
in the functions $`{}_{+}{}^{}K_{}^{}(x_{}^{})`$ and $`{}_{+}{}^{}J_{}^{}(x_{}^{})`$ (from $`{}_{+}{}^{}\varphi _{p_{},n,r}^{}(x^0,x^D)`$ and $`{}_{+}{}^{}\varphi _{p_{},n,r}^{}(x^0,x^D)`$ defined in (18)). We do not discuss the integration paths over the variable $`u,`$ since it is natural to consider the plane-wave potential $`f(x_{})`$ to be an entire function on $`x_\text{ }`$, in that case the integrals $`{}_{+}{}^{}K(x_{})`$, $`{}_{+}{}^{}J(x_{})`$, $`{}_{+}{}^{}K_{}^{}(x_{}^{})`$ and $`{}_{+}{}^{}J_{}^{}(x_{}^{})`$ are well defined by the integration limits. For different forms of the plane-wave potential the integration paths over $`u`$ are no longer arbitrary, however, they easily are extracted from the integration paths shown in Fig.1 and Fig.2. Then, one can present some combinations involved in (99) in the following way:
$`{}_{+}{}^{}K\left(x_{}\right)_+^{}K^{}\left(x_{}^{}\right)`$ $`=`$ $`l\left({}_{+}{}^{}a\right)+2\left(e^{2qF_+^{}a}1\right){\displaystyle _{{}_{+}{}^{}b}^0}e^{2qFu}qf\left(x_{}\left(u\right)\right)𝑑u,`$ (105)
$`{}_{+}{}^{}K\left(x_{}\right)+_+^{}K^{}\left(x_{}^{}\right)`$ $`=`$ $`l\left({}_{+}{}^{}a\right)+2\left(e^{2qF_+^{}a}+1\right){\displaystyle _{{}_{+}{}^{}b}^0}e^{2qFu}qf\left(x_{}\left(u\right)\right)𝑑u,`$ (106)
$`{}_{+}{}^{}J\left(x_{}\right)_+^{}J^{}\left(x_{}^{}\right)`$ $`=`$ $`\mathrm{\Phi }\left({}_{+}{}^{}a\right)+2{\displaystyle _0^{{}_{+}{}^{}a}}qf\left(x_{}\left(u\right)\right)e^{2qFu}𝑑uqF{\displaystyle _{{}_{+}{}^{}b}^0}e^{2qFu^{}}qf\left(x_{}\left(u^{}\right)\right)𝑑u^{},`$ (107)
where
$`\mathrm{\Phi }\left({}_{+}{}^{}a\right)`$ $`=`$ $`{\displaystyle _0^{{}_{+}{}^{}a}}qf\left(x_{}\left(u\right)\right)\left[qf\left(x_{}\left(u\right)\right)+qFl\left(u\right)\right]𝑑u,`$ (108)
$`l\left(u\right)`$ $`=`$ $`2{\displaystyle _0^u}e^{2qF\left(uu^{}\right)}qf\left(x_{}\left(u^{}\right)\right)𝑑u^{},`$ (109)
$`{}_{+}{}^{}b`$ $`=`$ $`\mathrm{}i\pi /\left(2qE\right)\mathrm{\Theta }\left(\pm \pi _{}^{}\right),x_{}(u)=x_{}^{}+y_{}{\displaystyle \frac{1\mathrm{exp}\left\{2qEu\right\}}{1\mathrm{exp}\left\{2qE_+^{}a\right\}}}.`$ (110)
The summation over $`n_j`$ in (99) can be performed using the Möller formula . After integration over $`p_j`$ in (99), applying the operators $`\mathrm{exp}\left\{i_+^{}K\left(x_{}\right)\pi _{}\right\}`$ and $`\mathrm{exp}\{i_+^{}K\left(x_{}\right)\pi _{}^{}\}^{}`$, and using relations (106), we obtain
$`{}_{+}{}^{}\stackrel{~}{f}_{}^{(0)}(x,x^{},p_{})=\mathrm{exp}\{iq\mathrm{\Lambda }im_+^2{}_{}{}^{}a`$ (112)
$`{\displaystyle \frac{1}{2}}[\mathrm{ln}(\stackrel{~}{\pi }_{})+\left(\mathrm{ln}(\stackrel{~}{\pi }_{}^{})\right)^{}]i{\displaystyle \frac{\pi _{}+\pi _{}^{}}{4}}y_+\}h(_+^{}a),`$
$`h(_+^{}a)`$ $`=`$ $`\mathrm{exp}\{i\mathrm{\Phi }(_+^{}a)i{\displaystyle \frac{1}{4}}(y+l(_+^{}a))qF^{}\mathrm{coth}(qF_+^{}{}_{}{}^{}a)(y+l(_+^{}a))+{\displaystyle \frac{i}{2}}yqF^{}l(s)\}Z_{(d)},`$ (113)
where
$`Z_{(d)}`$ $`=`$ $`c_d{\displaystyle \underset{j=1}{\overset{(d2)/2}{}}}\left({\displaystyle \frac{qH_j}{\mathrm{sin}(qH_js)}}\right),d\text{ is even},`$ (114)
$`Z_{(d)}`$ $`=`$ $`c_ds^{1/2}{\displaystyle \underset{j=1}{\overset{(d3)/2}{}}}\left({\displaystyle \frac{qH_j}{\mathrm{sin}(qH_js)}}\right),d\text{ is odd},`$ (115)
$`c_d`$ $`=`$ $`(4\pi )^{d/2}\mathrm{exp}\left\{i\pi (d4)/4\right\},`$ (116)
and
$$\mathrm{\Lambda }=_x^{}^x\left(A_\mu ^E+A_\mu ^H\right)𝑑x^\mu .$$
(117)
Here $`A_\mu ^E+A_\mu ^H`$ is a potential of the constant uniform field $`F_{\mu \nu },`$ and the integral is taken along the line.
Let us remark that after convenient gauge transformation of electric constant field potentials the functions $`{}_{+}{}^{}\stackrel{~}{f}_{}^{(0)}(x,x^{},p_{})`$ obeys the Klein-Gordon equation,
$$\left(\pi _{}2i\frac{}{x_{}}iqE+𝒫_{}^2m^2\right)\mathrm{exp}\left\{\frac{iqE}{2}\left(\frac{x_{}^2}{2}x_D^2\right)\right\}{}_{+}{}^{}\stackrel{~}{f}_{}^{(0)}(x,x^{},p_{})=0.$$
(118)
Taking into account the relations
$`\pi _{}e^{iq\mathrm{\Lambda }}`$ $`=`$ $`e^{iq\mathrm{\Lambda }}\left(i{\displaystyle \frac{}{x_{}}}+{\displaystyle \frac{1}{2}}qFy_{}\right),`$ (119)
$`\pi _{}^{}e^{iq\mathrm{\Lambda }}`$ $`=`$ $`e^{iq\mathrm{\Lambda }}\left(i{\displaystyle \frac{}{x_{}^{}}}{\displaystyle \frac{1}{2}}qFy_{}\right),`$ (120)
$`\mathrm{exp}\left(iq\sigma ^{0D}F_{0D}{}_{+}{}^{}a\right)`$ $`=`$ $`\mathrm{cosh}\left(qE_+^{}a\right)i\sigma ^{0D}\mathrm{sinh}\left(qE_+^{}a\right),`$ (121)
$`\mathrm{exp}\left(i{\displaystyle \frac{q}{2}}\sigma ^{\mu \nu }F_{\mu \nu }^{(j)}{}_{+}{}^{}a\right)`$ $`=`$ $`\mathrm{cos}\left(qH_j{}_{+}{}^{}a\right)+iR_j\mathrm{sin}\left(qH_j{}_{+}{}^{}a\right),`$ (122)
where
$`F_{\mu \nu }^{(j)}=H_j(\delta _\mu ^{2j}\delta _\nu ^{2j1}\delta _\nu ^{2j}\delta _\mu ^{2j1}),`$
and matrix $`R_j`$ is defined by (40), one can get relation
$`\mathrm{exp}\left(i{\displaystyle \frac{q}{2}}\sigma ^{\mu \nu }F_{\mu \nu }^{}{}_{+}{}^{}a\right)\gamma \pi _{}^{}{}_{+}{}^{}\stackrel{~}{f}_{}^{(0)}(x,x^{},p_{})=`$ (123)
$`(\gamma \pi _{}+\gamma qF^{}l\left({}_{+}{}^{}a\right))\mathrm{exp}(i{\displaystyle \frac{q}{2}}\sigma ^{\mu \nu }F_{\mu \nu }^{}{}_{+}{}^{}a)_+^{}\stackrel{~}{f}^{(0)}(x,x^{},p_{}).`$ (124)
By using formulas (122), (124) and Eq. (118) one can transform expression (98) to the following form,
$`{}_{+}{}^{}Y(x,x^{},p_{})`$ $`=`$ $`\left(\gamma 𝒫+m\right)_+^{}\stackrel{~}{f}(x,x^{},p_{}),`$ (125)
$`{}_{+}{}^{}\stackrel{~}{f}(x,x^{},p_{})`$ $`=`$ $`[\mathrm{exp}(i{\displaystyle \frac{q}{2}}\sigma ^{\mu \nu }F_{\mu \nu }{}_{+}{}^{}a){\displaystyle \frac{1}{2}}(\gamma ^0\gamma ^D)\gamma {\displaystyle _0^{{}_{+}{}^{}a}}e^{qF(_+^{}a2u)}q{\displaystyle \frac{df(x_{}(u))}{du}}du`$ (127)
$`\mathrm{exp}\{{\displaystyle \frac{1}{2}}[\mathrm{ln}(\stackrel{~}{\pi }_{})+\left(\mathrm{ln}(\stackrel{~}{\pi }_{}^{})\right)^{}]\}\left]_+^{}\stackrel{~}{f}^{(0)}\right(x,x^{},p_{}).`$
In the external field under consideration the real proper time $`S`$ is a function of $`\pi _{}`$ and $`\pi _{}^{}`$ because the classical equation of motion has a form $`\pi _{}^1dx_{}=m^1dS`$. Thus if $`y_{}0`$ one can transform the $`p_{}`$ integration in Green functions into integration over the Fock-Schwinger proper time by making the change of the variable,
$$s={}_{+}{}^{}a.$$
(128)
Then, one gets the following representations for the Green functions:
$`S^{,a,p}(x,x^{})`$ $`=`$ $`(\gamma 𝒫+m)\mathrm{\Delta }^{,a,p}(x,x^{}),`$ (129)
$`\mathrm{\Delta }^\pm (x,x^{})`$ $`=`$ $`{\displaystyle _{\mathrm{\Gamma }_c}}f(x,x^{},s)𝑑s\mathrm{\Theta }(\pm y_{}){\displaystyle _{\mathrm{\Gamma }_c\mathrm{\Gamma }_2\mathrm{\Gamma }_1}}f(x,x^{},s)𝑑s,`$ (130)
$`\mathrm{\Delta }^a(x,x^{})`$ $`=`$ $`{\displaystyle _{\mathrm{\Gamma }_a}}f(x,x^{},s)𝑑s+\mathrm{\Theta }(y_{}){\displaystyle _{\mathrm{\Gamma }_2+\mathrm{\Gamma }_3\mathrm{\Gamma }_a}}f(x,x^{},s)𝑑s.`$ (131)
$`\mathrm{\Delta }^p(x,x^{})`$ $`=`$ $`{\displaystyle _{\mathrm{\Gamma }_a}}f(x,x^{},s)𝑑s+\mathrm{\Theta }(y_{}){\displaystyle _{\mathrm{\Gamma }_1^a}}f(x,x^{},s)𝑑s.`$ (132)
All the contours of the integrals are shown on Fig. 3. The contours $`\mathrm{\Gamma }_c`$ and $`\mathrm{\Gamma }_1`$ are placed below the singular points on the real axis everywhere outside of the origin, and
$`f(x,x^{},s)=[\mathrm{exp}(i{\displaystyle \frac{q}{2}}\sigma ^{\mu \nu }F_{\mu \nu }s)`$ (134)
$`+(n\gamma )\gamma {\displaystyle _0^s}e^{qF(s2u)}{\displaystyle \frac{df((nx(u)))}{du}}du{\displaystyle \frac{\mathrm{sinh}(qEs)}{E(ny)}}]f^{(0)}(x,x^{},s),`$
$`f^{(0)}(x,x^{},s)=\mathrm{exp}\{iq\mathrm{\Lambda }\}{\displaystyle \frac{qE}{\mathrm{sinh}(qEs)}}\mathrm{exp}\{im^2s+i\mathrm{\Phi }(s)`$ (136)
$`i{\displaystyle \frac{1}{4}}(y+l(s))qF\mathrm{coth}(qFs)(y+l(s))+{\displaystyle \frac{i}{2}}yqFl(s)\}Z_{(d)},`$
where $`Z_{(d)}`$ is defined in (115). The function $`f^{(0)}(x,x^{},s)`$ has two singular points on the
complex region between the contours $`\mathrm{\Gamma }_c\mathrm{\Gamma }_1`$ and $`\mathrm{\Gamma }_a\mathrm{\Gamma }_3.`$ They are situated at the imaginary axis: $`s_0=0,`$ and $`qEs_1=i\pi `$. One can transform the contours $`\mathrm{\Gamma }_c\mathrm{\Gamma }_2\mathrm{\Gamma }_1`$ into $`\mathrm{\Gamma }`$ and $`\mathrm{\Gamma }_2+\mathrm{\Gamma }_3\mathrm{\Gamma }_a`$ into $`\mathrm{\Gamma }_1^a`$ (see Fig. 4) with radii tending to zero. Since
$`{\displaystyle _\mathrm{\Gamma }}f(x,x^{},s)𝑑s=0\text{if}y_\mu y^\mu <0,{\displaystyle _{\mathrm{\Gamma }_1^a}}f(x,x^{},s)𝑑s=0\text{if}y_0^2>y_D^2,`$
one can rewrite (130),(131) and (132) as follows
$`S^{}(x,x^{})`$ $`=`$ $`(\gamma 𝒫+m)\mathrm{\Delta }^{}(x,x^{}),`$ (137)
$`\mathrm{\Delta }^\pm (x,x^{})`$ $`=`$ $`{\displaystyle _{\mathrm{\Gamma }^c}}f(x,x^{},s)𝑑s\mathrm{\Theta }(\pm y_0){\displaystyle _\mathrm{\Gamma }}f(x,x^{},s)𝑑s,`$ (138)
$`S^{a,p}(x,x^{})`$ $`=`$ $`(\gamma 𝒫+m)\mathrm{\Delta }^{a,p}(x,x^{}),`$ (139)
$`\mathrm{\Delta }^a(x,x^{})`$ $`=`$ $`{\displaystyle _{\mathrm{\Gamma }^a}}f(x,x^{},s)𝑑s+\mathrm{\Theta }(y^D){\displaystyle _{\mathrm{\Gamma }_1^a}}f(x,x^{},s)𝑑s,`$ (140)
$`\mathrm{\Delta }^p(x,x^{})`$ $`=`$ $`{\displaystyle _{\mathrm{\Gamma }^a}}f(x,x^{},s)𝑑s+\mathrm{\Theta }(y^D){\displaystyle _{\mathrm{\Gamma }_1^a}}f(x,x^{},s)𝑑s.`$ (141)
One can check that these expressions are valid for arbitrary $`x`$ and $`x^{}`$. To prove this one need to verify that the functions $`S^\pm (x,x^{})`$ and $`S^{a,p}(x,x^{}),`$ which are presented via integrals (137) and (139), and by means of representations (93) and (96), are the same solutions of the Dirac equation for any $`x`$ and $`x^{}`$. Thus it is enough, to check first that expressions (137) and (139) obey the Dirac equation for any $`x`$ and $`x^{}`$. Then, one has to prove that the Cauchy conditions for distributions (137) and (139) coincide at $`x_0=x_0^{}`$ with (93) and (96) respectively. As to $`S^\pm (x,x^{}),`$ one can use the proper time representations for $`S^c(x,x^{})`$ and $`S(x,x^{})`$ functions, which follow from (137), (77) and (78),
$`S^c(x,x^{})`$ $`=`$ $`(\gamma 𝒫+m)\mathrm{\Delta }^c(x,x^{}),`$ (142)
$`\mathrm{\Delta }^c(x,x^{})`$ $`=`$ $`{\displaystyle _{\mathrm{\Gamma }_c}}f(x,x^{},s)𝑑s,`$ (143)
$`S(x,x^{})`$ $`=`$ $`(\gamma 𝒫+m)\mathrm{\Delta }(x,x^{}),`$ (144)
$`\mathrm{\Delta }(x,x^{})`$ $`=`$ $`\text{sign}(x_0x_0^{}){\displaystyle _\mathrm{\Gamma }}f(x,x^{},s)𝑑s.`$ (145)
One can see that $`f(x,x^{},s)`$ obeys the following equations,
$`i{\displaystyle \frac{d}{ds}}f(x,x^{},s)=\left(𝒫^2m^2{\displaystyle \frac{q}{2}}\sigma ^{\mu \nu }_{\mu \nu }\right)f(x,x^{},s),`$ (146)
$`\underset{s+0}{lim}f(x,x^{},s)=i\delta ^{(d)}(xx^{}).`$ (147)
Thus $`f(x,x^{},s)`$ is the Fock-Schwinger function . So, representation (143) of causal Green function has the well-known Schwinger form . The concrete representation (145) of the commutation function has the same form as the general representation . Similar to one can select all singularities of this function and see that it is continuous at $`x_0x_0^{}`$. In $`d=4`$ case one can transform this representation to the Fock form . The analysis of space-time singularities in the integrals (139) over the contour $`\mathrm{\Gamma }_1^a`$ can be done in a similar manner to . In the $`d=4`$ the representations (137) - (145) coincide with the ones found in . To find the proper time representation we used the special behavior of solutions (41). But one has the different form of solutions (42) if $`d`$ is odd, $`E=0`$ and all the imaginary eigenvalues $`H_j`$ of the field tensor are not equal to zero. In this case since solutions (42) can represent a special case of the $`d+1`$ general form (41), one does not need to calculate Green functions of the case independently and representations (137), (143) and (145) are valid if one puts $`E=0`$, $`f((nx))=0`$ and replaces $`qE/\mathrm{sinh}(qEs)qH_{(d1)/2}/\mathrm{sin}\left(qH_{(d1)/2}\right)`$ in formulas (134) and (136) (the functions $`S^{a,p}(x,x^{})`$ are equal to zero).
## IV Some physical applications
All the information about the processes of particle creation, annihilation, and scattering in an external field (without radiative corrections) can be extracted from the matrices $`G({}_{\zeta }{}^{}|{}_{}{}^{\zeta ^{}})`$ (7). These matrices define a canonical transformation between in and out creation and annihilation operators in the generalized Furry representation ,
$`a^{}(out)=a^{}(in)G({}_{+}{}^{}|{}_{}{}^{+})+b(in)G({}_{}{}^{}|{}_{}{}^{+}),`$ (148)
$`b(out)=a^{}(in)G({}_{+}{}^{}|{}_{}{}^{})+b(in)G({}_{}{}^{}|{}_{}{}^{}).`$ (149)
Here $`a_{\{n\}}^{}(in)`$, $`b_{\{n\}}^{}(in)`$, $`a_{\{n\}}(in)`$, $`b_{\{n\}}(in)`$ are creation and annihilation operators of in-particles and antiparticles respectively and $`a_{\{n\}}^{}(out)`$,$`b_{\{n\}}^{}(out)`$, $`a_{\{n\}}(out),b_{\{n\}}(out)`$ are ones of out-particles and antiparticles, $`\{n\}`$ are possible quantum numbers. For example, let us calculate the mean numbers of antiparticles created (which are also equal to the numbers of pairs created) by the external field from the in-vacuum $`|0,in>`$ with a given quantum number $`p_D,n,r`$. By using relations (149) and (48) one finds representation of this quantity:
$`N_{p_D,n,r}=<0,in|b_{p_D,n,r}^{}(out)b_{p_D,n,r}(out)|0,in>=`$ (150)
$`\left(G({}_{+}{}^{}|{}_{}{}^{})^{}G({}_{+}{}^{}|{}_{}{}^{})\right)_{p_D,n,r,p_D^{},n^{},r}|_{p_D=p_D^{},n=n^{}},`$ (151)
$`\left(G({}_{+}{}^{}|{}_{}{}^{})^{}G({}_{+}{}^{}|{}_{}{}^{})\right)_{p_D,n,r,p_D^{},n^{},r}=`$ (152)
$`{\displaystyle _{\mathrm{}}^+\mathrm{}}\left(G({}_{+}{}^{}|{}_{}{}^{})G({}_{+}{}^{}|{}_{}{}^{})^{}\right)_{p_{},n,r,p_{},n^{},r}M^{}(p_D,p_{})M(p_D^{},p_{}){\displaystyle \frac{dp_{}}{2\pi qE}}.`$ (153)
where the standard space coordinate volume regularization was used, so that $`\delta (p_jp_j^{})\delta _{p_j,p_j^{}}`$. Then, by using formulas (65) and (66) one gets
$$N_{p_D,n,r}=_{\mathrm{}}^+\mathrm{}{}_{}{}^{}𝒟_{nn}^{}M^{}(p_D,p_{})M(p_D^{},p_{})\frac{dp_{}}{2\pi qE}|_{p_D=p_D^{}}.$$
(154)
If $`L_D`$ is the length of the correspondent edge of the space box then the maximum wave length of the plane wave is $`2L_D`$. Thus, using the Fourier-series expansion of $`{}_{}{}^{}𝒟`$ one gets
$$N_{p_D,n,r}=\frac{1}{L_D}_0^{L_D}{}_{}{}^{}𝒟_{nn}^{}𝑑p_{}.$$
(155)
One can calculate the quantities $`{}_{+}{}^{}𝒟_{nn^{}}^{}`$, given by Eq. (66), taking into account that the operator $`\pi _{}`$ is Hermitian, and
$`e^{iK^{}\pi _{}}e^{iK\pi _{}}=\mathrm{exp}\left\{{\displaystyle \frac{i}{2}}K^{}qFKi\left(KK^{}\right)\pi _{}\right\}.`$
Then all the integrals over $`x^j`$ in form (66) can be expressed in terms of the Laguerre polynomials . In the special case when $`n=n^{}`$ one has
$`{}_{}{}^{}𝒟_{nn}^{}=\mathrm{exp}\left\{\pi \lambda \text{Im}{}_{}{}^{}K(x_{})\left(qf(p_{}/qE)+qF\text{Re}{}_{}{}^{}K(x_{})\right){\displaystyle \underset{j=1}{\overset{[d/2]1}{}}}h_j\right\}{\displaystyle \underset{j=1}{\overset{[d/2]1}{}}}L_{n_j}(2h_j),`$ (156)
$`h_j=|qH_j|((\text{Im}{}_{}{}^{}K_{2j1}^{}(x_{}))^2+(\text{Im}{}_{}{}^{}K_{2j}^{}(x_{}))^2),\pi _{}>0.`$ (157)
Remember that we are discussing the case in which the constant electric field acts for an infinite time. However, one can analyze the problem in finite times $`T=x_{out}^0x_{in}^0`$, acting similar to . Then the mean numbers of pairs created by the external field $`N_{p_D,n,r}`$ are the same if time $`T`$ is large enough: $`\sqrt{qE}T1,\sqrt{qE}T\lambda \text{and}qET|p_D|.`$
Summing over the quantum numbers in (155), one can find the total number of pairs created from the vacuum. Using standard regularization with respect to the $`(d1)`$-dimensional spatial volume $`V_{(d1)}`$ and special regularization with respect to time $`T`$ of acting of a constant electric field , where $`𝑑p_DN_{p_D,n,r}=qETN_{p_D,n,r}`$, one gets
$`N=V_{(d1)}n^{cr},`$ (158)
$`n^{cr}=J_{(d)}{\displaystyle \frac{Tm^2\beta (1)}{2^{(d1)}\pi ^{d/2}}}{\displaystyle \frac{E}{E_c}}\mathrm{exp}\left\{\pi {\displaystyle \frac{E_c}{E}}\right\},`$ (159)
where $`n^{cr}`$ is the number density of the created pairs for time $`T`$, and the coefficient $`\beta (1)`$ is defined by the next formula as a special case on $`l=1`$:
$`\beta (l)={\displaystyle \underset{j=1}{\overset{(d2)/2}{}}}\left\{qH_j\mathrm{coth}(l\pi H_j/E)\right\},d\mathrm{is}\mathrm{even},`$ (160)
$`\beta (l)=\left({\displaystyle \frac{m^2E}{n\pi E_c}}\right)^{\frac{1}{2}}{\displaystyle \underset{j=1}{\overset{(d3)/2}{}}}\left\{qH_j\mathrm{coth}(l\pi H_j/E)\right\},d\mathrm{is}\mathrm{odd}.`$ (161)
Here, $`E_c=m^2/|q|`$ is the characteristic value of a constant electric field strength. This quantity $`N`$ does not depend on the parameters of plane-wave field and is the same as the number of pairs created in a constant and uniform field . The corresponding formulas for the $`d=4`$ case were first written in , and, in fact, can be derived easily from the Schwinger formulas .
By using proper-time kernel $`f(x,x^{},s)`$ (134) one can construct the $`d`$ dimensional form of the Schwinger out-in effective action
$$\mathrm{\Gamma }_{outin}=\frac{1}{2}\text{tr}\left\{𝑑x_0^{\mathrm{}}s^1f(x,x,s)𝑑s\right\}.$$
(162)
It is not dependent on the parameters of the plane-wave field. And it is not amazing because the effective action is a function of the field invariants only which do not depend on the plane wave. Taking into account formulas (122) one can find the trace in (162) representation
$$\rho (s)=\text{tr}\left\{\mathrm{exp}\left(i\frac{q}{2}\sigma ^{\mu \nu }F_{\mu \nu }s\right)\right\}=2^{[d/2]}\mathrm{cosh}(qEs)\underset{j=1}{\overset{[(d2)/2]}{}}\mathrm{cos}(qH_js).$$
(163)
Then, one calculates the probability for a vacuum to remain a vacuum by using Schwinger method ,
$`P_v=\mathrm{exp}\left(2\text{Im}\mathrm{\Gamma }_{outin}\right)=\mathrm{exp}\left\{\mu N\right\},`$ (164)
$`\mu ={\displaystyle \underset{l=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{\beta (l+1)}{(l+1)\beta (1)}}\mathrm{exp}\left\{l\pi {\displaystyle \frac{E_c}{E}}\right\},`$ (165)
where $`\beta (l)`$ is defined by (161). This result coincides with the result from , and in $`d=4`$ the results coincide with Refs. .
Let the operator of the current of the Dirac field operator $`\psi (x)`$ have the form
$$j_\mu =\frac{q}{2}[\overline{\psi }(x),\gamma _\mu \psi (x)],$$
(166)
and operator of metric energy-momentum tensor (EMT) of the Dirac field operator has the form
$`T_{\mu \nu }={\displaystyle \frac{1}{2}}\left(T_{\mu \nu }^{can}+T_{\nu \mu }^{can}\right),,`$ (167)
$`T_{\mu \nu }^{can}={\displaystyle \frac{1}{4}}\left\{[\overline{\psi }(x),\gamma _\mu 𝒫_\nu \psi (x)]+[𝒫_\nu ^{}\overline{\psi }(x),\gamma _\mu \psi (x)]\right\},`$ (168)
where $`T_{\mu \nu }^{can}`$ is the canonical EMT operator. We are going to discuss the following matrix elements with these operators:
$`<`$ $`j_\mu >^c=<0,out|j_\mu |0,in>c_v^1,`$ (169)
$`<`$ $`T_{\mu \nu }>^c=<0,out|T_{\mu \nu }|0,in>c_v^1,`$ (170)
$`<`$ $`j_\mu >^{in}=<0,in|j_\mu |0,in>,`$ (171)
$`<`$ $`T_{\mu \nu }>^{in}=<0,in|T_{\mu \nu }|0,in>,`$ (172)
$`<`$ $`j_\mu >^{out}=<0,out|j_\mu |0,out>,`$ (173)
$`<`$ $`T_{\mu \nu }>^{out}=<0,out|T_{\mu \nu }|0,out>.`$ (174)
Using the Green functions which were found before, one can present these matrix elements in the following form:
$`<`$ $`j_\mu >^c=iq\text{tr}\left\{\gamma _\mu S^c(x,x)\right\}=iq\text{tr}\left\{\gamma _\mu \gamma ^\kappa 𝒫_\kappa \mathrm{\Delta }^c(x,x^{})\right\}|_{x=x^{}},`$ (175)
$`<`$ $`T_{\mu \nu }>^c=i/4\text{tr}\left\{(\gamma _\mu (𝒫_\nu +𝒫_{}^{}{}_{\nu }{}^{})+\gamma _\nu (𝒫_\mu +𝒫_{}^{}{}_{\mu }{}^{}))S^c(x,x^{})\right\}|_{x=x^{}}`$ (176)
$`=`$ $`i\text{tr}\left\{B_{\mu \nu }\mathrm{\Delta }^c(x,x^{})\right\}|_{x=x^{}},`$ (177)
$`<`$ $`j_\mu >^{in}=<j_\mu >^c<j_\mu >^a,`$ (178)
$`<`$ $`j_\mu >^{out}=<j_\mu >^c<j_\mu >^p,`$ (179)
$`<`$ $`T_{\mu \nu }>^{in}=<T_{\mu \nu }>^c<T_{\mu \nu }>^a,`$ (180)
$`<`$ $`T_{\mu \nu }>^{out}=<T_{\mu \nu }>^c<T_{\mu \nu }>^p,`$ (181)
$`<`$ $`j_\mu >^{(a,p)}=iq\text{tr}\left\{\gamma _\mu \gamma ^\kappa 𝒫_\kappa \mathrm{\Delta }^{a,p}(x,x^{})\right\}|_{x=x^{}},`$ (182)
$`<`$ $`T_{\mu \nu }>^{(a,p)}=i\text{tr}\left\{B_{\mu \nu }\mathrm{\Delta }^{a,p}(x,x^{})\right\}|_{x=x^{}},`$ (184)
$`B_{\mu \nu }=1/4\left\{\gamma _\mu \left(𝒫_\nu +𝒫_{}^{}{}_{\nu }{}^{}\right)+\gamma _\nu \left(𝒫_\mu +𝒫_{}^{}{}_{\mu }{}^{}\right)\right\}\gamma ^\kappa 𝒫_\kappa ,`$
where the Green functions are given by Eqs. (139), (143) and the relation
$`\mathrm{\Delta }^c(x,x)={\displaystyle \frac{1}{2}}\left[\mathrm{\Delta }^{}(x,x)\mathrm{\Delta }^+(x,x)\right]`$
is used. It is convenient to represent $`<j_\mu >^{a,p}`$ and $`<T_{\mu \nu }>^{a,p}`$ as follows:
$`<j_\mu >^a=<j_\mu >^{(1)}+<j_\mu >^{(2)},`$ (185)
$`<j_\mu >^p=<j_\mu >^{(1)}<j_\mu >^{(2)},`$ (186)
$`<T_{\mu \nu }>^a=<T_{\mu \nu }>^{(1)}+<T_{\mu \nu }>^{(2)},`$ (187)
$`<T_{\mu \nu }>^p=<T_{\mu \nu }>^{(1)}<T_{\mu \nu }>^{(2)},`$ (188)
where
$`<`$ $`j_\mu >^{(1)}=iq\text{tr}\left\{\gamma _\mu \gamma ^\kappa 𝒫_\kappa \mathrm{\Delta }^{(1)}(x,x^{})\right\}|_{x=x^{}},`$ (189)
$`<`$ $`T_{\mu \nu }>^{(1)}=i\text{tr}\left\{B_{\mu \nu }\mathrm{\Delta }^{(1)}(x,x^{})\right\}|_{x=x^{}},`$ (191)
$`\mathrm{\Delta }^{(1)}(x,x^{})={\displaystyle \frac{1}{2}}{\displaystyle _{\mathrm{\Gamma }_3+\mathrm{\Gamma }_2+\mathrm{\Gamma }_a}}f(x,x^{},s)𝑑s,`$
and all contributions with derivatives of $`\mathrm{\Theta }(\pm y^D)`$ functions, which are formally divergent, are included in terms $`<j_\mu >^{(2)}`$ and $`<T_{\mu \nu }>^{(2)}`$. The nature of such divergences is connected with infinite time $`T`$ of acting of a constant electric field and was discussed in (see also below).
The components $`<j_\mu >^{(1,2)}`$ and $`<T_{\mu \nu }>^{(1,2)}`$ can not be calculated in the framework of the perturbation theory with respect to the external background or in the framework of the WKB method. Among them only the term $`<j_\mu >^{in}`$ was calculated before in $`d=4`$ . Only expression (177) for $`<T_{\mu \nu }>^c`$ has to be regularized and renormalized because of the ultraviolet divergences. Expression (175) for the term $`<j_\mu >^c`$ is finite after the regularization lifting and equal to zero. The terms $`<j_\mu >^{(1)}`$ and $`<T_{\mu \nu }>^{(1)}`$ are also finite, and $`<j_\mu >^{(1)}=0.`$ The terms $`<j_\mu >^{(2)}`$ and $`<T_{\mu \nu }>^{(2)}`$ have to be regularized with respect to time $`T`$ of acting of a constant electric field and do not have standard ultraviolet divergences. That is consistent with the fact that the ultraviolet divergences have a local nature and result (as in the theory without external field) from the leading local terms at $`s+0`$. The nonzero contributions to the expressions $`<j_\mu >^{(2)}`$ and $`<T_{\mu \nu }>^{(1,2)}`$ are related to global features of the theory and indicate the vacuum instability.
At asymptotic region $`x_0=T/2+\mathrm{}`$ the densities of current and the EMT of particles created are
$`j_\mu ^{cr}={\displaystyle \frac{𝑑𝐱\left(<j_\mu >^{in}<j_\mu >^{out}\right)}{𝑑𝐱}},`$ (192)
$`T_{\mu \nu }^{cr}={\displaystyle \frac{𝑑𝐱\left(<T_{\mu \nu }>^{in}<T_{\mu \nu }>^{out}\right)}{𝑑𝐱}},`$ (193)
according to definitions (171) - (174).
Then, using representations (178) - (184), and taking into account terms $`<j_\mu >^{(2)}`$ and $`<T_{\mu \nu }>^{(2)}`$ which are uniform, one gets from (192) and (193),
$`j_\mu ^{cr}=2<j_\mu >^{(2)},`$ (194)
$`T_{\mu \nu }^{cr}=2<T_{\mu \nu }>^{(2)}.`$ (195)
Using special regularization with respect to time $`T`$ of acting of a constant electric field we can interpret these divergent terms and, correct to first order of $`\sqrt{qE}T`$ , obtain
$`<`$ $`j_\mu >^{cr}=2|q|n^{cr}\delta _\mu ^D,`$ (196)
$`<`$ $`T_{00}>^{cr}=<T_{DD}>^{cr}=qETn^{cr}.`$ (197)
Other components of $`<T_{\mu \nu }>^{cr}`$ are of the order of $`\mathrm{ln}(\sqrt{qE}T)`$.
To study the backreaction of particles created on the electromagnetic field and metrics one needs the expressions $`<j_\mu >^{in}`$ and $`<T_{\mu \nu }>^{in}.`$ We found the form of $`<j_\mu >^{in}=<j_\mu >^{(2)}=1/2<j_\mu >^{cr}`$ and $`<T_{\mu \nu }>^{(2)}=1/2<T_{\mu \nu }>^{cr}.`$ Let us calculate other terms in expressions (180). By using formulas (122) one can find traces in (177) and (184) representations. One has the next nonzero traces: one presented by expression (163) and
$`\text{tr}\left\{\gamma ^0\gamma ^D\mathrm{exp}\left(i{\displaystyle \frac{q}{2}}\sigma ^{\mu \nu }F_{\mu \nu }s\right)\right\}=\mathrm{tanh}(qEs)\rho (s),`$ (198)
$`\text{tr}\left\{\gamma ^{2j}\gamma ^{2j1}\mathrm{exp}\left(i{\displaystyle \frac{q}{2}}\sigma ^{\mu \nu }F_{\mu \nu }s\right)\right\}=\mathrm{tan}(qH_js)\rho (s).`$ (199)
Then, nonzero components of $`<T_{\mu \nu }>^c`$ and $`<T_{\mu \nu }>^{(1)}`$ are
$`<`$ $`T_{\mu \nu }>^c={\displaystyle _{\mathrm{\Gamma }_c}}\tau _{\mu \nu }(s)ds,`$ (200)
$`<`$ $`T_{\mu \nu }>^{(1)}={\displaystyle \frac{1}{2}}{\displaystyle _{\mathrm{\Gamma }_3+\mathrm{\Gamma }_2+\mathrm{\Gamma }_a}}\tau _{\mu \nu }(s)ds,`$ (203)
$`\tau _{\mu \nu }(s)=b_\mu (s)\rho (s)f^{(0)}(x,x,s)\text{if}\mu =\nu ,`$
$`b_D(s)=b_0(s)={\displaystyle \frac{qE}{\mathrm{sinh}(2qEs)}},b_{2j}(s)=b_{2j1}(s)={\displaystyle \frac{qH_j}{\mathrm{sin}(2qH_js)}}.`$
## V Acknowledgments
D.M.G. thanks the Brazilian foundations CNPq for support. S.P.G. thanks the Brazilian foundation CAPES for support, and the Department of Physics of UEL and Department of Mathematical Physics of USP for their hospitality.
|
warning/0004/gr-qc0004025.html
|
ar5iv
|
text
|
# Negative mode problem in false vacuum decay with gravity Talk given at Constrained Dynamics and Quantum Gravity 99, Villasimius, (Sardinia, Italy), September 14-18, 1999. To apper in the Proceedings.
## 1 INTRODUCTION
The first order phase transitions might play an important role in the Early Universe . They proceed via nucleation of the bubbles of true vacuum in metastable (false) one and subsequent growth of the bubbles. The false vacuum decay is usually discussed in frame of self-interacting scalar field theory. This process is described by the $`O(4)`$symmetric bounce solution of the Euclidean equations of motion. Value of the Euclidean action at the bounce gives leading exponential factor in decay rate. The perturbations about the bounce solution define the one loop corrections to the bubble nucleation rate and determine quantum state of materialized bubble . It is remarkable that in the spectrum of small perturbations about the bounce in flat spacetime there is exactly one negative mode . This mode is responsible for making correction to the ground-state energy imaginary, thus justifying decay interpretation.
When gravity is included the model contains gauge degrees of freedom and to extract information about the spectrum one should find proper reduction to physical variables. In the Euclidean gravity the presence of the gauge degrees of freedom manifests itself in the well known conformal factor problem. In pure gravity it is possible to cure this problem either at the level of prescription , or via careful gauge fixing . For scalar perturbations in the theory of a scalar field coupled to gravity it is the only spatially homogeneous modes, which suffer from the conformal factor problem (see e.g. ). On the other hand the negative mode is expected exactly in this problematic sector of scalar homogeneous perturbations, which consists of coupled scalar field and metric perturbations. The task is to reduce this coupled system to a system with a single physical variable. Note that there is no problem in elimination of unphysical variables in perturbations about the flat Friedmann-Robertson-Walker (FRW) type model with scalar field . Problem arises for perturbations in the theory of scalar field in closed FRW type Universe.
There are two possible ways of reduction of coupled system of perturbations: either to eliminate metric perturbations or perturbations of scalar field. The approach based on elimination of scalar field perturbations shows no negative mode in the spectrum of remining variable . On the other hand the recent investigation indicates presence of a single negative mode in the approach based on elimination of gravitational perturbations. This indicates on lack of full understanding of the role of gauge-fixing procedure in description of false vacuum decay with gravity and need for further investigation of this subject.
The aim of present contribution is to discuss and compare existing approaches to the negative mode problem. The rest of the article is organized as follows. In the next section we recall main formulas relevant to false vacuum decay in flat spacetime. In Sec. 3 we discuss inclusion of gravity, remind some important classical solutions of Euclidean equations of motion and derive quadratic action for scalar perturbations about $`O(4)`$symmetric background solutions. We will restrict ourselves to the most problematic sector of $`O(4)`$symmetric perturbations. In Sec. 4 we analyse three different reduction schemes. Sec. 5 contains the concluding remarks.
## 2 FALSE VACUUM DECAY IN FLAT SPACETIME
Let’s consider self-interacting scalar field $`\varphi `$ in flat spacetime and assume that the scalar field potential $`V(\varphi )`$ has shape shown in Fig.1, with local minimum (false vacuum) at $`\varphi =\varphi _+`$ and absolute minimum (true vacuum) at $`\varphi =\varphi _{}`$. The false vacuum decay is described by the Euclidean path integral
$$<\varphi _{}|e^{HT}|\varphi _+>=e^{{\scriptscriptstyle (i\pi _\varphi {\scriptscriptstyle \frac{d\varphi }{d\tau }}H)𝑑\tau }}𝑑\varphi 𝑑\pi _\varphi ,$$
(1)
where $`T`$ is a large positive number and $`H`$ is the scalar field Hamiltonian
$`H(\pi ,\varphi )={\displaystyle }[{\displaystyle \frac{\pi (𝐱)^2}{2}}+{\displaystyle \frac{1}{2}}_n\varphi (𝐱)^n\varphi (𝐱)`$
$`+V(\varphi )]d^3x,n=1,2,3.`$ (2)
Integrating Eq.(1) over $`\pi _\varphi `$ one gets
$$<\varphi _{}|e^{HT}|\varphi _+>=e^{S_E}𝑑\varphi ,$$
(3)
where $`S_E`$ is the usual Euclidean action
$$S_E=[\frac{1}{2}_\mu \varphi (x)^\mu \varphi (x)+V(\varphi )]d^3x𝑑\tau ,$$
(4)
with $`\mu =0,1,2,3`$. In the limit $`T\mathrm{}`$ the l.h.s. of Eq.(3) contains information about the lowest energy eigenvalue and wave function. The r.h.s. can be calculated in the WKB approximation. The energy $`E_0`$ of the lowest state gets correction due to tunneling phenomenon
$$E=E_0Ke^B,$$
(5)
with
$$B=S_E(\varphi _{bounce})S_E(\varphi _+),$$
(6)
where $`\varphi _{bounce}`$ is the $`O(4)`$ symmetric euclidean bounce solution with the lowest action, which interpolates between true and false vacua. The factor $`K`$ is obtained by the Gaussian integration of exponent of quadratic action of small perturbations about the bounce
$$K=\frac{B^2}{4\pi ^2}\left(\frac{det^{}[^2+V^{\prime \prime }(\varphi )]}{det[^2+V^{\prime \prime }(\varphi _+)]}\right)^{1/2}.$$
(7)
The operator $`^2+V^{\prime \prime }(\varphi )`$ has zero modes and $`det^{}`$ means that the determinant computed with the zero eigenvalues omitted. It turns out that there is exactly one mode with the negative eigenvalue in the spectrum of small perturbations about the bounce solution. This makes $`K`$ and hence the energy shift in the Eq.(5) purely imaginary and supports the false vacuum decay interpretation. Decay rate per unit volume $`𝒱`$, per unit time is given by
$$\frac{\mathrm{\Gamma }}{𝒱}=|K|e^B.$$
(8)
Note that the functional integral Eq.(1) is written for unconstrained (physical) degrees of freedom. If one has $`m`$ first class constraints
$$C_\alpha (\varphi ,\pi )0,\alpha =1,\mathrm{},m,$$
(9)
then according to general recipy one should choose some gauge fixing conditions
$$\chi _\beta (\varphi ,\pi )=0,\beta =1,\mathrm{},m,$$
(10)
and modify the functional integral, Eq.(1), as follows
$$e^{{\scriptscriptstyle (i\pi _\varphi {\scriptscriptstyle \frac{d\varphi }{d\tau }}H)𝑑\tau }}\mathrm{\Delta }\delta (C_\alpha )\delta (\chi ^\beta )𝑑\varphi 𝑑\pi _\varphi ,$$
(11)
with the Faddeev-Popov determinant
$$\mathrm{\Delta }det\{C_\alpha ,\chi ^\beta \}0.$$
(12)
## 3 INCLUSION OF GRAVITY
The Euclidean action of system composed of scalar field minimally coupled to gravity is
$$S=d^4x\sqrt{g}\left[\frac{1}{2\kappa }R+\frac{1}{2}_\mu \varphi ^\mu \varphi +V(\varphi )\right],$$
(13)
where $`\kappa =8\pi G`$ is the reduced Newton’s constant.
Since we are interested in solutions having $`O(4)`$symmetry, the metric is parametrised as follows
$$ds^2=d\sigma ^2+a^2(\sigma )\gamma _{ij}dx^idx^j,\varphi =\varphi (\sigma ),$$
(14)
where $`\gamma _{ij}`$ is the three-dimensional metric on the constant curvature space sections. The field equations are
$`\ddot{a}+{\displaystyle \frac{\kappa a}{3}}(\dot{\varphi }^2+V(\varphi ))=0,`$ (15)
$`\dot{a}^2𝒦{\displaystyle \frac{\kappa a^2}{3}}({\displaystyle \frac{\dot{\varphi }^2}{2}}V(\varphi ))=0,`$ (16)
$`\ddot{\phi }+3{\displaystyle \frac{\dot{a}}{a}}\dot{\phi }{\displaystyle \frac{\delta V}{\delta \phi }}=0,`$ (17)
where a dot denotes a derivative with respect to proper time $`\sigma `$ and $`𝒦`$ is the curvature parameter, which has the values 1, 0, -1 for closed, flat and open universes respectively. In what follows we are interested in closed case, but we do not specify the value of $`𝒦`$ now in order to see how it enters in the final expressions.
### 3.1 Euclidean solutions
There are few celebrated solutions of these equations, which play an important role in Euclidean quantum gravity.
1. The Hawking-Moss solution is a 4-sphere corresponding to scalar field sitting on the top of the potential barrier
$$\varphi (\sigma )=\varphi _{top},a(\sigma )=_{top}^1sin(_{top}\sigma ),$$
(18)
with $`_{top}=\sqrt{\kappa V(\varphi _{top})/3}`$. Note that the Euclidean time $`\sigma `$ varies in finite interval $`\sigma =(0,\sigma _f)`$.
2. The Coleman-De Luccia bounce is a deformed 4-sphere . It starts with some $`\varphi =\varphi _0`$ at $`\sigma =0`$ close to $`\varphi _{}`$, stops at $`\sigma =\sigma _f`$ close to $`\varphi _+`$ and obeys the regularity conditions
$$a(0)=\dot{\varphi }(0)=0,a(\sigma _f)=\dot{\varphi }(\sigma _f)=0.$$
(19)
3. Relaxing the regularity conditions at the second zero of $`a`$ one gets the Hawking-Turok singular instanton . It was introduced to describe creation of an open inflationary Universe.
Our main goal is to investigate spectrum of small perturbations about the Euclidean solutions and, in particular, to determine presence of a negative mode.
### 3.2 Quadratic action of scalar $`O(4)`$ symmetric perturbations
Investigation of perturbations would be convenient to perform in conformal frame. We expand the metric and the scalar field over a $`O(4)`$symmetric background
$`ds^2`$ $`=`$ $`a(\tau )^2[(1+2A(\tau ))d\tau ^2`$
$`+`$ $`\gamma _{ij}(12\mathrm{\Psi }(\tau ))dx^idx^j],`$
$`\varphi `$ $`=`$ $`\phi (\tau )+\mathrm{\Phi }(\tau ),`$ (20)
where $`\tau `$ is the conformal time, $`a`$ and $`\phi `$ are the background field values and $`A,\mathrm{\Psi }`$ and $`\mathrm{\Phi }`$ are small perturbations.
The background field equations in the conformal time are
$`^2^{}𝒦={\displaystyle \frac{\kappa }{2}}\phi {}_{}{}^{}{}_{}{}^{2},`$ (21)
$`2^{}+^2𝒦={\displaystyle \frac{\kappa }{2}}(\phi ^2+2a^2V(\phi )),`$ (22)
$`\phi {}_{}{}^{\prime \prime }+2\phi {}_{}{}^{}a^2{\displaystyle \frac{\delta V}{\delta \phi }}=0,`$ (23)
where a prime denotes a derivative with respect to $`\tau `$ and $`:=a^{}/a`$.
Expanding the total action, keeping terms of the second order in perturbations and using the background equations, we find
$$S=S^{(0)}+S^{(2)},$$
(24)
where $`S^{(0)}`$ is the action for the background solution and $`S^{(2)}`$ is quadratic in perturbations with the Lagrangian for scalar $`O(4)`$symmetric perturbations
$`{}_{}{}^{(s)}={\displaystyle \frac{1}{2\kappa }}a^2\sqrt{\gamma }[6\mathrm{\Psi }^2+6𝒦\mathrm{\Psi }^2`$
$`+\kappa (\mathrm{\Phi }^2+a^2{\displaystyle \frac{\delta ^2V}{\delta \phi \delta \phi }}\mathrm{\Phi }^2+6\phi ^{}\mathrm{\Psi }^{}\mathrm{\Phi })`$
$`(2\kappa \phi ^{}\mathrm{\Phi }^{}2\kappa a^2{\displaystyle \frac{\delta V}{\delta \phi }}\mathrm{\Phi }+12\mathrm{\Psi }^{}+12𝒦\mathrm{\Psi })A`$
$`2(^{}+2^2+𝒦)A^2].`$ (25)
Note that the variation with respect to $`A`$ gives the first order (constraint) equation
$`2\kappa \phi ^{}\mathrm{\Phi }^{}2\kappa a^2{\displaystyle \frac{\delta V}{\delta \phi }}\mathrm{\Phi }+12\mathrm{\Psi }^{}+12𝒦\mathrm{\Psi }`$
$`+4(^{}+2^2+𝒦)A=0.`$ (26)
## 4 THREE APPROACHES TO THE REDUCTION
To obtain the unconstrained system corresponding to the degenerate Lagrangian (3.2) we will follow the conventional Dirac formulation of generalized Hamiltonian dynamics . Calculating the canonical momenta
$`\mathrm{\Pi }_\mathrm{\Psi }:=i{\displaystyle \frac{\delta ^{(s)}}{\delta \mathrm{\Psi }^{}}}=i{\displaystyle \frac{6a^2\sqrt{\gamma }}{\kappa }}\left(\mathrm{\Psi }^{}+{\displaystyle \frac{\kappa }{2}}\phi ^{}\mathrm{\Phi }A\right),`$
$`\mathrm{\Pi }_\mathrm{\Phi }:=i{\displaystyle \frac{\delta ^{(s)}}{\delta \mathrm{\Phi }^{}}}=ia^2\sqrt{\gamma }\left(\mathrm{\Phi }^{}\phi ^{}A\right),`$
$`\mathrm{\Pi }_A:=i{\displaystyle \frac{\delta ^{(s)}}{\delta A^{}}}=0,`$ (27)
we find the primary constraint $`C_1:=\mathrm{\Pi }_A=0`$.
Thus the evolution is governed by the total Hamiltonian
$$H_T=H_C+u_1(\tau )C_1,$$
(28)
with arbitrary function $`u_1(\tau )`$ and canonical Hamiltonian
$`H_C={\displaystyle \frac{\kappa }{12a^2\sqrt{\gamma }}}\mathrm{\Pi }_\mathrm{\Psi }^2+{\displaystyle \frac{1}{2a^2\sqrt{\gamma }}}\mathrm{\Pi }_\mathrm{\Phi }^2+{\displaystyle \frac{i\kappa \phi ^{}}{2}}\mathrm{\Pi }_\mathrm{\Psi }\mathrm{\Phi }`$
$`+a^2\sqrt{\gamma }\left[{\displaystyle \frac{3𝒦}{\kappa }}\mathrm{\Psi }^2+{\displaystyle \frac{1}{2}}\left(a^2{\displaystyle \frac{\delta ^2V}{\delta \phi \delta \phi }}+{\displaystyle \frac{3}{2}}\kappa \phi ^2\right)\mathrm{\Phi }^2\right]`$
$`+A[i\phi ^{}\mathrm{\Pi }_\mathrm{\Phi }i\mathrm{\Pi }_\mathrm{\Psi }`$
$`+a^2\sqrt{\gamma }((a^2{\displaystyle \frac{\delta V}{\delta \phi }}3\phi ^{})\mathrm{\Phi }{\displaystyle \frac{6𝒦}{\kappa }}\mathrm{\Psi })].`$ (29)
Conservation of primary constraint gives the secondary constraint
$`C_2`$ $`=`$ $`i\phi ^{}\mathrm{\Pi }_\mathrm{\Phi }i\mathrm{\Pi }_\mathrm{\Psi }`$ (30)
$`+`$ $`a^2\sqrt{\gamma }\left((a^2{\displaystyle \frac{\delta V}{\delta \phi }}3\phi ^{})\mathrm{\Phi }{\displaystyle \frac{6𝒦}{\kappa }}\mathrm{\Psi }\right).`$
The primary and secondary constraints are first class one and there are no ternary constraints.
Under the infinitesimal shift $`\tau \tau +\lambda `$ these constraints generate the gauge transformations
$`\delta \mathrm{\Psi }`$ $`=`$ $`\lambda ,\delta \mathrm{\Pi }_\mathrm{\Psi }={\displaystyle \frac{i6a^2\sqrt{\gamma }𝒦}{\kappa }}\lambda ,`$
$`\delta \mathrm{\Phi }`$ $`=`$ $`\phi {}_{}{}^{}\lambda ,\delta \mathrm{\Pi }_\mathrm{\Phi }=ia^2\sqrt{\gamma }(\phi {}_{}{}^{\prime \prime }\phi {}_{}{}^{})\lambda ,`$
$`\delta A`$ $`=`$ $`\lambda {}_{}{}^{}+\lambda .`$ (31)
The existence of constraints in the system as usually means the presence of unphysical degrees of freedom and that the whole system has no unique dynamics. To find the physical variables with unique evolution one should fix the gauge and solve the constraints. There are two possible general strategies: either to eliminate the gravitational degrees of freedom ($`\mathrm{\Pi }_\mathrm{\Psi }`$ and $`\mathrm{\Psi }`$) or perturbations of scalar field ($`\mathrm{\Pi }_\mathrm{\Phi }`$ and $`\mathrm{\Phi }`$).
### 4.1 The I approach
In the first investigation on this subject the Lagrangian approach was used . The gauge was fixed by the condition
$$\mathrm{\Psi }=0.$$
(32)
Using this gauge condition and eliminating $`A`$ with the help of the constraint equation Eq.(3.2) we obtain the unconstrained quadratic action in the form
$`S_{LRT}^{(2)}={\displaystyle }{\displaystyle \frac{a^4\sqrt{\gamma }}{2Q_{LRT}}}[{\displaystyle \frac{^2}{a^2}}\mathrm{\Phi }^2{\displaystyle \frac{\kappa \phi ^{}}{3}}{\displaystyle \frac{\delta V}{\delta \phi }}\mathrm{\Phi }^{}\mathrm{\Phi }`$
$`+({\displaystyle \frac{\kappa a^2}{6}}{\displaystyle \frac{\delta V}{\delta \phi }}+Q_{LRT}{\displaystyle \frac{\delta ^2V}{\delta \phi \delta \phi }})\mathrm{\Phi }^2]d\tau d^3x,`$ (33)
with
$$Q_{LRT}:=(^{}+2^2+𝒦)/3=^2\frac{\kappa \phi _{}^{}{}_{}{}^{2}}{6}.$$
(34)
We see that the action Eq.(4.1) has correct overall sign if $`Q_{LRT}>0`$. As it was shown in under certain choice of parameters the bounce can develop a region with negative $`Q_{LRT}`$. The main conclusion in was that the region with $`Q_{LRT}<0`$ leads to catastrophic particles creation and is pathological.
However the gauge fixing Eq.(32) has problems where $`a^{}=0`$. This can be seen in the Hamiltonian formalism, while the Faddeev-Popov determinant is proportional to $`a^{}`$.
### 4.2 The II approach
Eliminating $`\mathrm{\Pi }_\mathrm{\Phi }`$ using the constraint Eq.(30) and working in the gauge invariant variables
$`𝚿`$ $`=`$ $`\mathrm{\Psi }+{\displaystyle \frac{}{\phi ^{}}}\mathrm{\Phi },`$
$`𝚷_𝚿`$ $`=`$ $`\mathrm{\Pi }_\mathrm{\Psi }+{\displaystyle \frac{6i𝒦a^2\sqrt{\gamma }}{\kappa \phi ^{}}}\mathrm{\Phi },`$ (35)
one obtains the reduced hamiltonian
$`H^{}={\displaystyle \frac{\kappa }{12a^2\sqrt{\gamma }}}𝚷_{𝚿}^{}{}_{}{}^{2}+{\displaystyle \frac{a^2\sqrt{\gamma }3𝒦}{\kappa }}𝚿^\mathrm{𝟐}`$
$`{\displaystyle \frac{2a^2\sqrt{\gamma }}{\phi _{}^{}{}_{}{}^{2}}}\left({\displaystyle \frac{3𝒦}{\kappa }}𝚿+{\displaystyle \frac{i}{2a^2\sqrt{\gamma }}}𝚷_𝚿\right)^2.`$ (36)
Performing the canonical transformation
$`𝚿={\displaystyle \frac{i\kappa \phi ^{}}{4}}\stackrel{~}{q}{\displaystyle \frac{}{3𝒦a^2\sqrt{\gamma }\phi ^{}}}\stackrel{~}{p},`$
$`𝚷_𝚿={\displaystyle \frac{3𝒦a^2\sqrt{\gamma }\phi ^{}}{2}}\stackrel{~}{q}{\displaystyle \frac{2i}{\kappa \phi ^{}}}\stackrel{~}{p},`$ (37)
and solving for the momenta $`\stackrel{~}{p}`$ one obtains the quadratic part of the Euclidean action
$$S^{(2)}=\frac{(14𝒦)}{2}\left[(\frac{dq}{d\tau })^2+Uq^2\right]\sqrt{\gamma }d^3x𝑑\tau ,$$
(38)
with a potential $`U`$ depending on the background fields
$$U=\frac{\kappa }{2}\phi _{}^{}{}_{}{}^{2}+\phi ^{}(\frac{1}{\phi ^{}})^{\prime \prime }+14𝒦.$$
(39)
and $`q=a\stackrel{~}{q}`$.
We see that quadratic action for the homogeneous harmonic has “wrong” overall sign. To overcome this problem it was suggested that analytic continuation
$$qiq$$
(40)
is performed while integrating over this mode (compare ).
The equation for the mode functions, which diagonalize the action (38), has form of the Schrödinger equation
$$\frac{d^2}{d\tau ^2}q+Uq=Eq.$$
(41)
Note that the singularities $`{\displaystyle \frac{1}{\phi ^{}}}`$, which appear in this approach do not allow one to investigate straightforwardly the perturbations about the Hawking-Moss solution (with $`\phi =\phi _{top}=const`$).
The spectrum about the Coleman-De Luccia bounce was investigated and no negative mode theorem was proven .
The Hawking-Turok instanton was investigated and it was found that for monotonous scalar field potentials there is no negative mode in the spectrum. On the other hand it was found that the negative mode is present in “exotic” cases when the Hawking-Turok instanton “overshoots” the Coleman-De Luccia bounce. This means that in this case the Hawking-Turok instanton is unstable and has larger action then the corresponding Coleman-De Luccia bounce.
Note that the formulation in terms of the gauge invariant variables Eq.(4.2) corresponds to the gauge choice $`\chi _{GMST}:=\mathrm{\Phi }=0`$ (compare ). Hence the Faddeev-Popov determinant is $`\phi ^{}`$ and this approach is singular for certain configurations.
### 4.3 The III approach
Having in mind that i) spherically symmetric gravity has no propagating degrees of freedom and ii) in the limit $`\kappa 0`$ we should recover scalar field theory in flat space, we prefer an approach based on elimination of gravitational perturbations . At the same time from the Eq.(30) we see that solving constraint we divide either to $`\phi ^{}`$ or to $``$ or to $`a^2`$. Since $`\phi ^{}`$ and $`a^{}`$ might have zero(s) by solving constraints we might introduce extra singularities in quadratic action. For closed universe the scale factor $`a`$ is vanishing for some $`\tau =\tau _\pm `$. Since we are interested in perturbations which vanish for $`\tau \tau _\pm `$, the coordinate singularity $`a(\tau _\pm )=0`$ looks most harmless.
We use the following gauge fixing conditions
$$\chi _1:=A=0,\chi _2:=\frac{\kappa }{6𝒦a^2\sqrt{\gamma }}\mathrm{\Pi }_\mathrm{\Psi }=0,$$
(42)
which obey the following canonical algebra:
$`\{C_i,C_j\}=\{\chi _i,\chi _j\}=0,\{\chi _i,C_j\}=\delta _{ij}.`$ (43)
As a next step we consider all constraints in a strong sense and define the physical Hamiltonian as
$$H^{}:=H_C|_{\chi _i=0,C_i=0}.$$
(44)
For the physical Hamiltonian we find
$`H^{}`$ $`=`$ $`{\displaystyle \frac{1}{2a^2\sqrt{\gamma }}}\left(1{\displaystyle \frac{\kappa }{6𝒦}}\phi ^2\right)\mathrm{\Pi }_\mathrm{\Phi }^2`$ (45)
$`+`$ $`{\displaystyle \frac{i\kappa \phi ^{}}{6𝒦}}(a^2{\displaystyle \frac{\delta V}{\delta \phi }}3\phi ^{})\mathrm{\Pi }_\mathrm{\Phi }\mathrm{\Phi }`$
$`+`$ $`{\displaystyle \frac{1}{2}}a^2\sqrt{\gamma }[{\displaystyle \frac{\kappa }{6𝒦}}(a^2{\displaystyle \frac{\delta V}{\delta \phi }}3\phi ^{})^2`$
$`+`$ $`(a^2{\displaystyle \frac{\delta ^2V}{\delta \phi \delta \phi }}+{\displaystyle \frac{3}{2}}\kappa \phi ^2)]\mathrm{\Phi }^2,`$
which corresponds to the following unconstrained quadratic action for one physical dynamical degree of freedom
$`S^{(2)}={\displaystyle }{\displaystyle \frac{a^2\sqrt{\gamma }}{2Q}}[\mathrm{\Phi }^2{\displaystyle \frac{\kappa \phi ^{}}{3𝒦}}(a^2{\displaystyle \frac{\delta V}{\delta \phi }}3\phi ^{})\mathrm{\Phi }^{}\mathrm{\Phi }`$
$`+({\displaystyle \frac{\kappa }{6𝒦}}(a^2{\displaystyle \frac{\delta V}{\delta \phi }}3\phi ^{})^2`$
$`+Q(a^2{\displaystyle \frac{\delta ^2V}{\delta \phi \delta \phi }}+{\displaystyle \frac{3}{2}}\kappa \phi ^2))\mathrm{\Phi }^2]d\tau d^3x,`$ (46)
with
$$Q:=\left(1\frac{\kappa }{6𝒦}\phi ^2\right).$$
(47)
In what follows we will consider the background configurations for which the factor $`Q`$ is positive definite.
Introducing a new variable $`q=a/\sqrt{Q}\mathrm{\Phi }`$ and integrating by parts we obtain the unconstrained quadratic action in the form
$$S^{(2)}=\frac{1}{2}\left(q^2+W[a(\tau ),\phi (\tau )]q^2\right)𝑑\tau \sqrt{\gamma }d^3x,$$
(48)
with frequency $`W`$ whose time dependence is determined by the background solutions
$`W[a(\tau ),\phi (\tau )]={\displaystyle \frac{a^2}{Q}}{\displaystyle \frac{\delta ^2V}{\delta \phi \delta \phi }}+{\displaystyle \frac{\kappa a^4}{2𝒦Q^2}}({\displaystyle \frac{\delta V}{\delta \phi }})^2`$ (49)
$``$ $`{\displaystyle \frac{2\kappa \phi ^{}a^2}{𝒦Q^2}}{\displaystyle \frac{\delta V}{\delta \phi }}{\displaystyle \frac{10^2}{Q}}+{\displaystyle \frac{12^2}{Q^2}}`$
$`+`$ $`{\displaystyle \frac{8}{Q}}𝒦6𝒦3𝒦Q.`$
Note that the obtained action allows us to consider the slowly varying scalar field, $`\phi ^{}0`$, and vanishing gravity, $`\kappa 0`$, limits.
The equation for the mode functions, which diagonalize the action Eq.(48), has form of the Schrödinger equation
$$\frac{d^2}{d\tau ^2}q+W[a(\tau ),\phi (\tau )]q=Eq,$$
(50)
with the potential Eq.(49). Having this equation, one can investigate the negative mode problem for concrete cases numerically. We investigated the case with the scalar field potential
$$V(\phi )=\frac{m^2}{2}(\phi ^2(\phi v)^2+B\phi ^4).$$
(51)
This potential<sup>1</sup><sup>1</sup>1$`\kappa =1`$ units are used. for $`m^2=2,B=0.12`$ and $`v=0.5`$ has local maximum at $`\phi _{top}=0.31250`$, and local minimum (false vacuum) at $`\phi _{false}=0.3571428`$. For this potential there exists the Coleman-De Luccia bounce solution with $`\phi _0=0.1123579`$. The quantum mechanical Schrödinger equation Eq.(50) was solved numerically for this potential and exactly one bound state was found. Note that the factor $`Q`$ was positive definite for this case.
Within this approach it is problematic to investigate the Hawking-Turok instanton, while scalar field $`\phi `$ runs away and eventually the factor $`Q`$ becomes negative.
We found that the Schrödinger equation Eq.(50) for the Hawking-Moss background solution has six boundstates. This number is in perfect agreement with the analytic formula for the eigenvalues :
$$\lambda _n=n(n+3)_{top}^2+V^{\prime \prime }(\phi _{top}),n=0,1,2\mathrm{}$$
(52)
while for our example
$$\frac{V^{\prime \prime }(\phi _{top})}{_{top}^2}=40.96.$$
(53)
Presence of many negative modes about the Hawking-Moss solution means that the corresponding Coleman-De Luccia bounce for the given potential and set of parameters gives dominant contribution to tunneling .
Note that this approach based on gauge fixing Eq.(42) can be equally well formulated in the gauge invariant variables
$`𝚽`$ $`=`$ $`\mathrm{\Phi }{\displaystyle \frac{i\kappa \phi ^{}}{6a^2\sqrt{\gamma }𝒦}}\mathrm{\Pi }_\mathrm{\Psi },`$
$`𝚷_𝚽`$ $`=`$ $`\mathrm{\Pi }_\mathrm{\Phi }+{\displaystyle \frac{\kappa }{6𝒦}}(\phi {}_{}{}^{\prime \prime }\phi ^{})\mathrm{\Pi }_\mathrm{\Psi }.`$ (54)
## 5 CONCLUDING REMARKS
There are two important types of the Euclidean solutions in flat spacetime: instantons and bounces. The instanton describes quantum-mechanical mixing between the equal energy states, whereas the bounce describes decay of a metastable state. There are no negative modes in the spectrum of perturbations about the instantons (only zero modes). The bounce has a single negative mode. When gravity is included plenty of new euclidean solutions are found. Presence or absence of a negative mode might be a good indicator for understanding to which type each solution belongs: instanton (mixing) or bounce (decay).
The approach based on elimination of gravitational degrees of freedom shows no negative mode in the spectrum. It was suggested that the $`i`$ which comes from the conformal rotation Eq.(40) plays the same role as a negative mode. This argument has a weak point. The conformal rotation Eq.(40) is the same for all euclidean background solutions, while the imaginary shift in energy is peculiarity of an unstable state. Moreover, for the Giddings-Strominger wormhole a negative mode was found on top of conformal rotation .
The approach based on elimination of gravitational degrees of freedom indicates the presence of a negative mode. It has smooth $`\kappa 0`$ limit and can be straightforwardly used for the Hawking-Moss solution (with $`\phi 0`$). At the same time there are still many open questions. A detailed numerical study of different cases is needed. Another issue which needs further investigation is an understanding of the role of configurations with the negative factor $`Q`$: is it real physical instability or it is the breakdown of the reduction scheme.
I am deeply thankful to the organizers of QG99 for splendid hospitality and for creating charming atmosphere for discussions during the meeting.
Note added: After this contribution was essentially completed, further progress in investigation of negative mode problem was made. The results are summarized in the revised version of .
|
warning/0004/physics0004026.html
|
ar5iv
|
text
|
# Untitled Document
SOMMERFELD PARTICLE
IN STATIC MAGNETIC FIELD:
TUNNELING AND DELAYED UNTWISTING IN CYCLOTRON
Alexander A. Vlasov
High Energy and Quantum Theory
Department of Physics
Moscow State University
Moscow, 119899
Russia
Motion of a charged particle with finite size, described by Sommerfeld model, in static magnetic field has two peculiar features: 1.) there is the effect of tunneling - Sommerfeld particle overcomes the barrier and finds itself in the forbidden, from classical point of view, area; 2.) the untwisting of trajectory in cyclotron for Sommerfeld particle is strongly delayed compared to that of a classical particle.
03.50.De
Here we continue our investigation of peculiar features of motion of Sommerfeld particle . Let us remind that long time ago Sommerfeld proposed a model of a charged particle of finite size - sphere with uniform surface charge $`Q`$ and mechanical mass $`m`$. In nonrelativistic approximation such sphere obeys the equation (see also ):
$$m\dot{\stackrel{}{v}}=\stackrel{}{F}_{ext}+\eta \left[\stackrel{}{v}(t2a/c)\stackrel{}{v}(t)\right]$$
$`(1)`$
here $`a`$ \- radius of the sphere, $`\eta =\frac{Q^2}{3ca^2},\stackrel{}{v}=d\stackrel{}{R}/dt,\stackrel{}{R}`$ \- coordinate of the center of the shell, $`\stackrel{}{F}_{ext}`$ \- some external force.
This model is a good tool to consider effects of radiation reaction of a charged particle of finite size, free of problems of classical point-like Lorentz-Dirac description. A. If Sommerfeld particle moves in the external static magnetic field $`\stackrel{}{H}`$, the force $`\stackrel{}{F}_{ext}=𝑑\stackrel{}{r}\rho [\dot{\stackrel{}{R}},\stackrel{}{H}]`$ for $`\rho =Q\delta (|\stackrel{}{r}\stackrel{}{R}|a)/4\pi a^2`$ has the form
$$F_{ext}=\frac{Q}{c}[\dot{\stackrel{}{R}},\stackrel{}{H}]$$
If magnetic field has non-zero values only in the shell of finite size $`S`$ ( $`0<Y<S`$, $`\stackrel{}{H}`$ is parallel to $`z`$-axis, $`\stackrel{}{R}=(X,Y,0)`$ ), then, as the particle has finite size $`2a`$, force $`\stackrel{}{F}_{ext}`$ must be multiplied by the factor $`f`$:
$$f=\{\begin{array}{cc}0,& Y<a;\\ \frac{Y}{2a}+\frac{1}{2},& a<Y<a;\\ 1,& a<Y<Sa;\\ \frac{SY}{2a}+\frac{1}{2},& Sa<Y<S+a;\\ 0,& S+a<y;\end{array}$$
$`(2)`$
For dimensionless variables $`x=X/M,y=Y/M,\tau =ct/M`$ ($`M`$ -scale factor) equation (1) takes the form
$$\ddot{y}=K\left[\dot{y}(\tau d)\dot{y}(\tau )\right]\lambda \dot{x}f,$$
$$\ddot{x}=K\left[\dot{x}(\tau d)\dot{x}(\tau )\right]+\lambda \dot{y}f,$$
$`(3)`$
here
$$f=\{\begin{array}{cc}0,& y<\frac{d}{2};\\ \frac{y}{d}+\frac{1}{2},& \frac{d}{2}<y<\frac{d}{2};\\ 1,& \frac{d}{2}<y<L\frac{d}{2};\\ \frac{Ly}{d}+\frac{1}{2},& L\frac{d}{2}<y<L+\frac{d}{2};\\ 0,& L+\frac{d}{2}<y;\end{array}$$
$`(4)`$
and
$$K=\frac{Q^2M}{3a^2mc^2},\lambda =\frac{QHM}{mc^2},d=\frac{2a}{M},L=\frac{S}{M}.$$
Classical analog of equation (3) for point-like particle without radiation reaction reads
$$\ddot{y}=\lambda \dot{x}g,$$
$$\ddot{x}=\lambda \dot{y}g,$$
$`(5)`$
here
$$g=\{\begin{array}{cc}0,& y<0;\\ 1,& 0<y<L;\\ 0,& L<y;\end{array}$$
$`(6)`$
For initial conditions $`x(0)=0,y(0)=0,\dot{x}(0)=0,\dot{y}(0)=v`$ solution of (5) is
$$x=\frac{v}{\lambda }+\frac{v}{\lambda }\mathrm{cos}(\lambda \tau ),$$
$$y=\frac{v}{\lambda }\mathrm{sin}(\lambda \tau )(0<y<L)$$
$`(7)`$
We see that for initial velocities $`v`$ smaller, then the critical velocity $`v_{cr}=\lambda L`$, particle trajectory (half-circle) lies inside the shell, i.e. particle cannot overcome the barrier. If $`L=10^4,\lambda =10^4`$ then $`v_{cr}=1`$.
We numerically investigated the particle motion governed by equation (3) for the following values of initial velocity:
$$v=0.43,v=0.44$$
and for
$$L=10^4,\lambda =10^4,d=1.0,K=4/(3d^2),$$
i.e. particle is of electron size and mass, magnetic field approximately equals $`10^{12}`$ gauss and $`S5,610^9`$sm.
The result is shown on Fig. A, compared with classical trajectory, governed by (7) with $`v=0.44`$. Horisontal axis is $`x`$ and vertical axis is $`y`$.
The effect of tunneling for Sommerfeld particle is vividly seen: velocity $`v=0.44`$ is smaller then the critical $`v_{cr}=1`$, but the particle overcomes the barrier and finds itself in the forbidden from classical point of view area $`y>L=10^4`$. B.
If magnetic field is parallel to $`z`$-axis for $`y<0`$ and $`y>L`$ and equals to zero for $`0<y<L`$, and for $`0<y<L`$ there is static electric field $`E`$, parallel to $`y`$\- axis in such a way, that it is always collinear to $`y`$-component of particle velocity (i.e. particle is always accelerates in the clearance $`0<y<L`$), then there is a model of cyclotron.
Equation of motion for Sommerfeld particle in cyclotron reads
$$\ddot{y}=K\left[\dot{y}(\tau d)\dot{y}(\tau )\right]\lambda \dot{x}f+ϵSgn(\dot{y})(1f),$$
$$\ddot{x}=K\left[\dot{x}(\tau d)\dot{x}(\tau )\right]+\lambda \dot{y}f,$$
$`(8)`$
here
$$ϵ=\frac{QEM}{mc^2}$$
Classical analog of (8) one can construct replacing in (8) $`K`$ by zero and $`f`$ by $`g`$ (6):
$$\ddot{y}=\lambda \dot{x}g+ϵSgn(\dot{y})(1g),$$
$$\ddot{x}=\lambda \dot{y}g,$$
$`(9)`$
Initial conditions are:
$$x(0)=y(0)=\dot{x}(0)=\dot{y}(0)=0$$
Due to classical equation of motion without radiation reaction (9) particle moves along untwisting trajectory. Total increase of kinetic energy $`W_c=(\dot{x})^2/2+(\dot{y})^2/2`$ of particle is $`NeL`$:
$$W_c=NϵL$$
where $`N`$ \- is the total number of passing of particle through the accelerating field $`E`$.
If $`N=10,ϵ=\lambda =10^7,L=10^5`$, then
$$W_c=10^1.$$
We numerically calculated the particle motion governed by equation (8) with zero initial conditions for the following values of parameters:
$$L=10^5,\lambda =10^7=ϵ,d=0.3,K=2.0,$$
i.e. particle is of electron size and mass, magnetic field approximately equals to $`8.110^7`$ gauss and electric field produces in the clearance potential difference equal to $`10^4`$ eV.
The results of calculations are shown on Fig. B.1 - classical case and on Fig. B.2 - case of Sommerfeld particle. Horisontal axis is $`x\lambda `$ and vertical axis is $`y\lambda `$.
We see that for the same ”time” $`\tau 10^8`$ (i.e $`t10^4sec`$) classical particle (without radiation reaction) made $`N=10`$ passings through the accelerating field $`E`$ with total energy increase $`W_c=10^1`$, while Sommerfeld particle made only $`N=6`$ passings with total energy increase $`W_s=0.0375`$ ( $`W_c`$ for $`N=6`$ is equal to $`0.06`$ ). Thus untwisting of trajectory for Sommerfeld particle is strongly delayed compared to that of a classical one.
Delay in energy increase falls mainly on the moments of passing through the clearance. It can be explained by difference in accelerations in electric field (proportional to $`ϵ10^7`$) and in magnetic field (proportional to $`v\lambda 10^8`$ ) as flux of radiating energy is proportional to square of acceleration.
REFERENCES
1. Alexander A.Vlasov, physics/9905050, physics/9911059.
2. A.Sommerfeld, Gottingen Nachrichten, 29 (1904), 363 (1904), 201 (1905).
3. L.Page, Phys.Rev., 11, 377 (1918). T.Erber, Fortschr. Phys., 9, 343 (1961). P.Pearle in ”Electromagnetism”,ed. D.Tepliz, (Plenum, N.Y., 1982), p.211. A.Yaghjian, ”Relativistic Dynamics of a Charged Sphere”. Lecture Notes in Physics, 11 (Springer-Verlag, Berlin, 1992).
v=0.43 Sommerfeld: v=0.44 Classic: v=0.44 2.61e0 1.80e3 3.60e3 5.40e3 7.20e3 9.00e3 1.08e4 1.26e4 1.44e4 1.62e4 1.80e4 0.00e0 1.80e3 3.60e3 5.40e3 7.20e3 9.00e3 1.08e4 1.26e4 1.44e4 1.62e4 1.80e4
Fig. A
-6.00e-1 -4.80e-1 -3.60e-1 -2.40e-1 -1.20e-1 0.00e0 1.20e-1 2.40e-1 3.60e-1 4.80e-1 6.00e-1 -6.00e-1 -4.80e-1 -3.60e-1 -2.40e-1 -1.20e-1 0.00e0 1.20e-1 2.40e-1 3.60e-1 4.80e-1 6.00e-1
Fig. B.1.
-6.00e-1 -4.80e-1 -3.60e-1 -2.40e-1 -1.20e-1 0.00e0 1.20e-1 2.40e-1 3.60e-1 4.80e-1 6.00e-1 -6.00e-1 -4.80e-1 -3.60e-1 -2.40e-1 -1.20e-1 0.00e0 1.20e-1 2.40e-1 3.60e-1 4.80e-1 6.00e-1
Fig. B.2
|
warning/0004/hep-th0004116.html
|
ar5iv
|
text
|
# Symmetry Origin of Infrared Divergence Cancellation in Topologically Massive Yang-Mills Theory
## Abstract
We manifestly verify that the miraculous cancellation of the infrared divergence of the topologically massive Yang-Mills theory in the Landau gauge is completely determined by a new vector symmetry existing in its large topological mass limit, the Chern-Simons theory. Furthermore, we show that the cancellation theorem proposed by Pisarski and Rao is an inevitable consequence of this new vector symmetry.
PACS: 11.10.Kk, 11.15.-q, 03.70.tk
Keywords: Landau vector supersymmetry; infrared divergence; cancellation theorem; large topological mass limit.
There had been an upsurge in the perturbative investigation of 2+1-dimensional non-Abelian Chern-Simons field theory following a seminal work by Witten, in which a miraculous relation between quantum Chern-Simons theory and two-dimensional conformal invariant WZW model was found. Various regularization schemes were utilized in the perturbative calculation. Now looking back what remarkable features have been revealed by the perturbative theory, it seems to us that there are three points. In addition to the famous finite renormalization of the gauge coupling, one is an explicit confirmation on an earlier assumption by Jackiw that the Chern-Simons theory is the large topological mass limit of the topologically massive Yang-Mills theory (TMYM) , in which the Chern-Simons term can provide a topological mass to the Yang-Mills field without recourse to the Higgs mechanism; The other is the exposure of a new vector symmetry existing only in the Landau gauge fixed Chern-Simons theory, which similar to the BRST symmetry puts the gauge and ghost fields in a nonlinear multiplet and is thus entitled the Landau vector supersymmetry. It was verified that this new symmetry has non-trivial dynamical effects: it protects the Chern-Simons theory from getting quantum correction. In particular, it is intimately related to the ambiguity of the finite renormalization of the gauge coupling constant: if the regularization method adopted in a concrete perturbative calculation preserves the BRST symmetry but violates the Landau vector supersymmetry, the coupling constant receives a finite renormalization, while reversely the coupling constant will keep its classical value. There exists no regularization schemes that can preserve both of the Landau vector supersymmetry and the BRST symmetry simultaneously.
It was well known that in the Landau gauge TMYM is ultraviolet finite and its beta function vanishes identically. Consequently, there is no dynamical generated mass scale and the form factor of a Green function can only be the form of $`f(p^2/m^2)`$ with $`p`$ being certain external momentumThe mass dimension of the coupling constant can be removed by a re-definition of the fields, so the only massive parameter left is the topological mass.. Thus the fact that the Chern-Simons theory is the large topological mass limit of TMYM means that it can be defined as the infrared limit of TMYM. An equivalent statement is that TMYM is infrared finite. On the other side, the Landau vector supersymmetry arises in the Landau gauge-fixed Chern-Simons theory. In view of these two aspects, we cannot help thinking that the Landau vector supersymmetry, which arises in the low-energy limit of TMYM, probably has some connection with the cancellation of the infrared divergence. The aim of this letter is to give an explicit verification.
The cancellation of the infrared divergence in TMYM was investigated in detail in a beautiful paper by Pisarski and Rao. They first observed that in the one-loop vacuum polarization tensor, the infrared divergences come from the gauge parameter dependent part of the gluon propagator, the ghost loop amplitude, and the combination of the parity-odd pieces from each three-gluon vertex and the gluon propagator in the gluon loop contribution. However, under the choice of the Landau gauge, the gauge dependent part vanishes, the combination of parity-odd pieces in the gluon loop amplitude and the ghost loop contribution unexpectedly cancel in the infrared limit. The same phenomenon happens for the one-loop three-gluon vertex. Based on these observations, Pisarski and Rao concluded that the Chern-Simons part is completely responsible for the cancellation of the infrared divergences in TMYM, despite that their ultraviolet behaviour is greatly different. More concretely, the cancellation of the infrared divergence is determined by the odd parity property of the gauge-fixed Chern-Simons theory, and only in the Landau gauge the Chern-Simons theory can present this feature. This conclusion render them to formulate a cancellation theorem to explain the miraculous cancellation of the infrared divergence in TMYM: In Landau gauge the Chern-Simons theory has only a finite renormalization and its quantum effective action takes the same form as the classical one. The cancellation of infrared divergences contributed by the ghost field and the parity-odd parts of the gauge field is entirely dominated by this theorem.
Despite that this theorem gets the essence of the infrared divergence cancellation, but it looks far from being regarded as the origin of a dynamical phenomenon. Especially, the cancellation takes place at the level of the Green function, this theorem is not clear enough to show how the cancellation occurs among the Green functions. There must be certain symmetry hidden behind this empirical theorem, which will relate various Green functions through the Ward identities, to implement the cancellation of the infrared divergence. Thus we place the hope on the Landau vector supersymmetry to play such a role.
The classical action of TMYM in Euclidean space-time is
$`S`$ $`=`$ $`im{\displaystyle d^3xϵ_{\mu \nu \rho }\left(\frac{1}{2}A_\mu ^a_\nu A_\rho ^a+\frac{1}{3!}gf^{abc}A_\mu ^aA_\nu ^bA_\rho ^c\right)}+{\displaystyle \frac{1}{4}}{\displaystyle d^3xF_{\mu \nu }^aF_{\mu \nu }^a},`$ (1)
where it is assumed that $`m>0`$. To remove the mass dimension of the gauge coupling so that it is explicit that the Chern-Simons theory can be regarded as the large topological mass limit of TMYM at classical level, we rescale the field and the coupling, $`Am^{1/2}A`$ and $`gm^{1/2}g`$. Consequently, the classical action (1) becomes
$`S`$ $`=`$ $`i{\displaystyle d^3xϵ_{\mu \nu \rho }\left(\frac{1}{2}A_\mu ^a_\nu A_\rho ^a+\frac{1}{3!}gf^{abc}A_\mu ^aA_\nu ^bA_\rho ^c\right)}+{\displaystyle \frac{1}{4m}}{\displaystyle d^3xF_{\mu \nu }^aF_{\mu \nu }^a}.`$ (2)
In the Lorentz gauge-fixing condition, the gauge-fixing and the ghost field parts of the total classical effective action are as the following,
$`S_g={\displaystyle d^3x\left[B^a_\mu A_\mu ^a_\mu \overline{c}^a\left(_\mu c^a+gf^{abc}A_\mu ^bc^c\right)+\frac{\xi }{2}(B^a)^2\right]}.`$ (3)
As in the usual gauge theory, the gauge-fixed action has the well known BRST symmetry,
$`\delta A_\mu ^a=D_\mu ^{ab}c^b,\delta c^a={\displaystyle \frac{g}{2}}f^{abc}c^bc^c,\delta \overline{c}^a=B^a,\delta B^a=0.`$ (4)
However, in the Landau gauge choice, $`\xi =0`$, the large topological mass limit of TMYM, the gauge-fixed Chern-Simons theory has a new BRST-like vector symmetry ,
$`V_\mu A_\nu ^a=iϵ_{\mu \nu \rho }_\rho c^a,V_\mu B^a=(D_\mu c)^a,V_\mu c^a=0,V_\mu \overline{c}^a=A_\mu ^a.`$ (5)
At quantum level, writing out formally the generating functional with the inclusion of the external sources for the fields and their variations under the supersymmetry (5), one can derive the Ward identities corresponding to the Landau vector supersymmetry of the Chern-Simons theory in a standard way. In the topologically massive Yang-Mills theory this symmetry is explicitly broken by the Yang-Mills term, but the broken Ward identity can still be written out. Here we take a shortcut to derive the needed identities. Consider a general functional $`F[\mathrm{\Phi }]`$ of the field $`\mathrm{\Phi }=(A_\mu ^a,B^a,c^a,\overline{c}^a)`$, the invariance of the quantum observable
$`F[\mathrm{\Phi }]={\displaystyle 𝒟\mathrm{\Phi }F[\mathrm{\Phi }]e^{S[\mathrm{\Phi }]}}`$ (6)
under an arbitrary infinitesimal transformation $`\mathrm{\Phi }\mathrm{\Phi }+\delta \mathrm{\Phi }`$ yields the following identity,
$`{\displaystyle \frac{F[\mathrm{\Phi }]}{\mathrm{\Phi }}}\delta \mathrm{\Phi }\delta S[\mathrm{\Phi }]F[\mathrm{\Phi }]=0.`$ (7)
Choosing the functions $`F[\mathrm{\Phi }]=A_\mu ^a(x)\overline{c}^b(y)`$ and $`F[\mathrm{\Phi }]=A_\mu ^a(x)A_\nu ^b(y)\overline{c}^c(z)`$ in TMYM, respectively, the Landau vector supersymmetry transformation (5) leads to
$`A_\mu ^a(x)A_\nu ^b(y)=iϵ_{\mu \nu \rho }_\rho ^xc^a(x)\overline{c}^b(y)+A_\mu ^a(x)\overline{c}^b(y)V_\nu S_{YM},`$ (8)
and
$`A_\mu ^a(x)A_\nu ^b(y)A_\rho ^c(z)`$ $`=`$ $`iϵ_{\mu \rho \lambda }_\lambda ^xc^a(x)A_\nu ^b(y)\overline{c}^c(z)+iϵ_{\nu \rho \lambda }_\lambda ^yA_\mu ^a(x)c^b(y)\overline{c}^c(z)`$ (10)
$`+A_\mu ^a(x)A_\nu ^b(y)\overline{c}^c(z)V_\rho S_{YM}.`$
A straightforward calculation gives
$`V_\mu S_{YM}`$ $`=`$ $`𝒪_\mu ^{(0)}+𝒪_\mu ^{(1)}+𝒪_\mu ^{(2)};`$ (11)
$`𝒪_\mu ^{(0)}`$ $`=`$ $`{\displaystyle \frac{i}{m}}{\displaystyle d^3xϵ_{\mu \nu \rho }_\rho c^a\mathrm{}A_\nu ^a};`$ (12)
$`𝒪_\mu ^{(1)}`$ $`=`$ $`{\displaystyle \frac{i}{m}}gf^{abc}{\displaystyle d^3xϵ_{\mu \nu \rho }_\rho c^a\left[2A_\lambda ^b_\lambda A_\nu ^c+A_\lambda ^b_\nu A_\lambda ^c\left(_\lambda A_\lambda ^b\right)A_\nu ^c\right]};`$ (13)
$`𝒪_\mu ^{(2)}`$ $`=`$ $`{\displaystyle \frac{i}{m}}g^2f^{abc}f^{ade}{\displaystyle d^3xϵ_{\mu \nu \rho }A_\lambda ^b_\rho c^cA_\lambda ^dA_\nu ^e}.`$ (14)
Eq.(8) shows that the broken Landau vector supersymmetry has established a direct relation between the gauge field and the ghost field propagators. In momentum space, it reads
$`D_{\mu \nu }(p)=ϵ_{\mu \nu \rho }p_\rho S(p)+D_{\mu \rho }(p)\mathrm{\Omega }_{\rho \nu }(p)S(p),`$ (15)
where $`D_{\mu \nu }(p)`$ and $`S(p)`$ are the propagators of gauge and ghost fields in momentum space, respectively; the composite vertex $`\mathrm{\Omega }(p)`$ at tree level comes from $`𝒪_\mu ^{(0)}`$. Eq.(15) formally leads to
$`ϵ_{\mu \nu \rho }p_\rho D_{\lambda \mu }^1(p)S(p)=\delta _{\lambda \nu }\mathrm{\Omega }_{\lambda \nu }(p)S(p).`$ (16)
Eq.(10) relates the three-point function of the gauge field to the three-point function of the gauge field and the ghost fields. In momentum space it is
$`D_{\mu \alpha }(p)D_{\nu \beta }(q)D_{\rho \gamma }(r)\mathrm{\Gamma }_{\alpha \beta \gamma }^{abc}(p,q,r)`$ $`=`$ $`ϵ_{\mu \rho \lambda }p_\lambda S(p)D_{\nu \beta }(q)S(r)\mathrm{\Gamma }_\beta (p,r,q)`$ (19)
$`ϵ_{\nu \rho \lambda }q_\lambda D_{\mu \alpha }(p)S(q)S(r)\mathrm{\Gamma }_\alpha (q,r,p)`$
$`+D_{\mu \alpha }(p)D_{\nu \beta }(q)S(r)\stackrel{~}{\mathrm{\Gamma }}_{\alpha \beta \rho }^{abc}(p,q,r),`$
$`p+q+r`$ $`=`$ $`0.`$ (20)
where $`\mathrm{\Gamma }_{\mu \nu \sigma }^{abc}`$ and $`\mathrm{\Gamma }_\mu ^{abc}`$ are the three-gluon vertex and the ghost-ghost-gluon vertex, respectively; $`\stackrel{~}{\mathrm{\Gamma }}_{\mu \nu \rho }^{abc}(p,q,r)`$ at tree-level comes from the composite operator $`𝒪_\nu ^{(1)}`$. Extracting the one particle irreducible (1PI) part, we obtain a relation between the three-gluon vertex and the ghost-ghost-gluon vertex,
$`D_{\rho \sigma }(r)\mathrm{\Gamma }_{\mu \nu \sigma }^{abc}(p,q,r)`$ $`=`$ $`ϵ_{\alpha \rho \lambda }p_\lambda D_{\mu \alpha }^1(p)S(p)S(r)\mathrm{\Gamma }_\nu ^{acb}(p,r,q)`$ (22)
$`ϵ_{\alpha \rho \lambda }q_\lambda D_{\nu \alpha }^1(q)S(q)S(r)\mathrm{\Gamma }_\nu ^{bca}(q,r,p)+S(r)\stackrel{~}{\mathrm{\Gamma }}_{\mu \nu \rho }^{abc}(p,q,r),`$
$`p+q+r`$ $`=`$ $`0.`$ (23)
The identities (15) and (23) at tree-level can be easily verified with the bare propagators and vertices. The substitution of (16) into (23) further yields
$`D_{\rho \sigma }(r)\mathrm{\Gamma }_{\mu \nu \sigma }^{abc}(p,q,r)`$ $`=`$ $`\delta _{\mu \rho }S(r)\mathrm{\Gamma }_\nu ^{abc}(p,r,q)\delta _{\nu \rho }S(r)\mathrm{\Gamma }_\mu ^{abc}(q,r,p)`$ (26)
$`\mathrm{\Omega }_{\mu \rho }(p)S(p)S(r)\mathrm{\Gamma }_\nu ^{abc}(p,r,q)+\mathrm{\Omega }_{\nu \rho }(q)S(q)S(r)\mathrm{\Gamma }_\mu ^{abc}(q,r,p)`$
$`+S(r)\stackrel{~}{\mathrm{\Gamma }}_{\mu \nu \rho }^{abc}(p,q,r).`$
In the following, we shall show explicitly how the identity (26), born out of the Landau vector supersymmetry, enforces the cancellation of infrared divergence. Consider the vacuum polarization tensor contributed from the skeleton diagrams of the ghost and the gluon loops, we have
$`\mathrm{\Pi }_{\mu \nu }^{ab}(p)`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \frac{d^3k}{(2\pi )^3}\mathrm{\Gamma }_{\mu \lambda \rho }^{acd}(p,k,kp)D_{\lambda \beta }(k)\mathrm{\Gamma }_{\nu \alpha \beta }^{bdc}(p,k+p,k)D_{\alpha \rho }(k+p)}`$ (28)
$`{\displaystyle \frac{d^3k}{(2\pi )^3}\mathrm{\Gamma }_\mu ^{acd}(p,k,kp)S(k)\mathrm{\Gamma }_\nu ^{bdc}(p,k+p,k)S(k+p)}.`$
Despite that Eq.(28) is written as the form of one-loop Feynman diagram amplitude, it can actually be regarded as a vacuum polarization tensor at any higher order since the insertion of the vertices and propagators in these two one-loop skeletons can be chosen to be any order as desired. The first term in the R.H.S. of Eq.(28) is the contribution from the skeleton of gluon loop and the second one from the ghost loop skeleton, these are the only two sources of the infrared divergence. Substituting Eq.(26) into Eq.(28), we obtain
$`\mathrm{\Pi }_{\mu \nu }^{ab}(p)`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle }{\displaystyle \frac{d^3k}{(2\pi )^3}}\{S(kp)S(k)[\delta _{\mu \alpha }\mathrm{\Gamma }_\lambda ^{acd}(p,kp,k)\delta _{\lambda \alpha }\mathrm{\Gamma }_\mu ^{acd}(k,kp,p)`$ (33)
$`\mathrm{\Omega }_{\mu \alpha }(p)S(p)\mathrm{\Gamma }_\lambda ^{acd}(p,kp,k)+\mathrm{\Omega }_{\lambda \alpha }(k)S(k)\mathrm{\Gamma }_\mu ^{acd}(k,kp,p)`$
$`+\stackrel{~}{\mathrm{\Gamma }}_{\mu \lambda \alpha }^{acd}(p,k,kp)\left]\right[\delta _{\nu \lambda }\mathrm{\Gamma }_\alpha ^{bdc}(p,k,k+p)\delta _{\lambda \alpha }\mathrm{\Gamma }_\mu ^{bdc}(k+p,k,p)`$
$`\mathrm{\Omega }_{\nu \lambda }(p)S(p)\mathrm{\Gamma }_\alpha ^{bdc}(p,k,k+p)+\mathrm{\Omega }_{\lambda \alpha }(k+p)S(k+p)\mathrm{\Gamma }_\nu ^{bdc}(k+p,k,p)`$
$`+\stackrel{~}{\mathrm{\Gamma }}_{\mu \alpha \lambda }^{bdc}(p,k+p,k)]\}`$
$``$ $`{\displaystyle \frac{d^3k}{(2\pi )^3}\mathrm{\Gamma }_\mu ^{acd}(p,k,kp)S(k)\mathrm{\Gamma }_\nu ^{bdc}(p,k+p,k)S(k+p)}`$ (34)
$`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle }{\displaystyle \frac{d^3k}{(2\pi )^3}}S(k)S(kp)[\mathrm{\Gamma }_\mu ^{bdc}(p,k,k+p)\mathrm{\Gamma }_\nu ^{acd}(p,kp,k)`$ (44)
$`\mathrm{\Gamma }_\mu ^{acd}(p,kp,k)\mathrm{\Gamma }_\nu ^{bdc}(k+p,k,p)\mathrm{\Gamma }_\mu ^{acd}(k,kp,p)\mathrm{\Gamma }_\nu ^{bdc}(p,k,k+p)`$
$`+3\mathrm{\Gamma }_\mu ^{acd}(k,kp,p)\mathrm{\Gamma }_\nu ^{bdc}(k+p,k,p)]`$
$`+{\displaystyle \frac{1}{2}}{\displaystyle }{\displaystyle \frac{d^3k}{(2\pi )^3}}S(k)S(kp)\{[\delta _{\mu \alpha }\mathrm{\Gamma }_\lambda ^{acd}(p,kp,k)\delta _{\lambda \alpha }\mathrm{\Gamma }_\mu ^{acd}(k,kp,p)]`$
$`\times \left[\mathrm{\Omega }_{\lambda \alpha }(k+p)S(k+p)\mathrm{\Gamma }_\nu ^{bdc}(k+p,k,p)\mathrm{\Omega }_{\nu \lambda }(p)S(p)\mathrm{\Gamma }_\alpha ^{bdc}(p,k,k+p)\right]`$
$`+\left[\mathrm{\Omega }_{\lambda \alpha }(k)S(k)\mathrm{\Gamma }_\mu ^{acd}(k,kp,p)\mathrm{\Omega }_{\mu \alpha }(p)S(p)\mathrm{\Gamma }_\lambda ^{acd}(p,kp,k)\right]`$
$`\times [\delta _{\nu \lambda }\mathrm{\Gamma }_\alpha ^{bdc}(p,k,k+p)\delta _{\alpha \lambda }\mathrm{\Gamma }_\nu ^{bdc}(k+p,k,p)]\}`$
$`+{\displaystyle \frac{1}{2}}{\displaystyle }{\displaystyle \frac{d^3k}{(2\pi )^3}}S(k)S(kp)\{[\delta _{\mu \alpha }\mathrm{\Gamma }_\lambda ^{acd}(p,kp,k)\delta _{\lambda \alpha }\mathrm{\Gamma }_\mu ^{acd}(k,kp,p)]`$
$`\times \stackrel{~}{\mathrm{\Gamma }}_{\nu \alpha \lambda }^{bdc}(p,k+p,k)+\stackrel{~}{\mathrm{\Gamma }}_{\mu \lambda \alpha }^{acd}(p,k,kp)[\delta _{\nu \lambda }\mathrm{\Gamma }_\alpha ^{bdc}(p,k,k+p)`$
$`\delta _{\lambda \alpha }\mathrm{\Gamma }_\nu ^{bdc}(k+p,k,p)]\}`$
$`+`$ $`{\displaystyle \frac{1}{2}}{\displaystyle }{\displaystyle \frac{d^3k}{(2\pi )^3}}S(k)S(kp)\{[\mathrm{\Omega }_{\lambda \alpha }(k)S(k)\mathrm{\Gamma }_\mu ^{acd}(k,kp,p)`$ (48)
$`\mathrm{\Omega }_{\mu \alpha }(p)S(p)\mathrm{\Gamma }_\lambda ^{acd}(p,kp,k)]\stackrel{~}{\mathrm{\Gamma }}^{bdc}_{\nu \alpha \lambda }(p,k+p,k)`$
$`+\stackrel{~}{\mathrm{\Gamma }}_{\mu \lambda \alpha }^{acd}(p,k,kp)[\mathrm{\Omega }_{\lambda \alpha }(k+p)S(k)\mathrm{\Gamma }_\nu ^{bdc}(k+p,k,p)`$
$`\mathrm{\Omega }_{\nu \lambda }(p)S(p)\mathrm{\Gamma }_\alpha ^{bdc}(p,k,k+p)]\}`$
$`+`$ $`{\displaystyle \frac{1}{2}}{\displaystyle }{\displaystyle \frac{d^3k}{(2\pi )^3}}S(k)S(kp)\{[\mathrm{\Omega }_{\lambda \alpha }(k)S(k)\mathrm{\Gamma }_\mu ^{acd}(k,kp,p)`$ (51)
$`\mathrm{\Omega }_{\mu \alpha }(p)S(p)\mathrm{\Gamma }_\lambda ^{acd}(p,kp,k)\left]\right[\mathrm{\Omega }_{\lambda \alpha }(k+p)S(k+p)\mathrm{\Gamma }_\nu ^{bdc}(k+p,k,p)`$
$`\mathrm{\Omega }_{\nu \lambda }(p)S(p)\mathrm{\Gamma }_\alpha ^{bdc}(p,k,k+p)]\}`$
$`+`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \frac{d^3k}{(2\pi )^3}S(k)S(kp)\stackrel{~}{\mathrm{\Gamma }}_{\mu \lambda \alpha }^{acd}(p,k,kp)\stackrel{~}{\mathrm{\Gamma }}_{\mu \alpha \lambda }^{bdc}(p,k+p,k)}`$ (52)
$``$ $`{\displaystyle \frac{d^3k}{(2\pi )^3}S(k)S(k+p)\mathrm{\Gamma }_\mu ^{acd}(p,k,kp)\mathrm{\Gamma }_\nu ^{bdc}(p,k+p,k)}.`$ (53)
The first term in the expansion of Eq.(53) is only associated with the contribution from the ghost propagators and the ghost-ghost-gluon vertices, and hence it is the term in the gluon loop skeleton that contains the infrared divergence. The global gauge symmetry requires that the ghost-ghost-gluon vertex at any order must be of the form,
$`\mathrm{\Gamma }_\mu ^{abc}(p,q,r)=f^{abc}\mathrm{\Gamma }_\mu (p,q,r).`$ (54)
Moreover, the Ward identities from the BRST symmetry given in Eq.(4) determines the ghost field propagator and the ghost-ghost-gluon vertex are,
$`S(p)`$ $`=`$ $`{\displaystyle \frac{1}{p^2\left[1+\mathrm{\Sigma }(p^2/m^2)\right]}};`$ (55)
$`\mathrm{\Gamma }_\nu ^{abc}(p,q,r)`$ $`=`$ $`p_\mu \gamma _{\mu \nu }^{abc}(p,q,r),p+q+r=0,`$ (56)
where $`\gamma _{\mu \nu }^{abc}`$ is the composite ghost-gluon vertex function of the following general form,
$`\gamma _{\mu \nu }^{abc}(p,q,r)`$ $`=`$ $`f^{abc}(ϵ_{\mu \nu \rho }p_\rho A_1+ϵ_{\mu \nu \rho }q_\rho A_2+ϵ_{\mu \lambda \rho }p_\lambda q_\rho p_\nu A_3+ϵ_{\mu \lambda \rho }p_\lambda q_\rho q_\nu A_4`$ (59)
$`+ϵ_{\nu \lambda \rho }p_\lambda q_\rho p_\mu A_5+ϵ_{\nu \lambda \rho }p_\lambda q_\rho q_\mu A_6+g_{\mu \nu }A_7+p_\mu p_\nu A_8+p_\mu q_\nu A_9`$
$`+p_\nu q_\mu A_{10}+q_\mu q_\nu A_{11}),`$
$`A_i`$ $``$ $`A_i(p^2,q^2,r^2,m),i=1,2,\mathrm{},11;p+q+r=0.`$ (60)
Using Eqs.(54)–(60), we can easily find that up to a constant coefficient term the first term of Eq.(53) cancels with the contribution from the ghost loop skeleton (i.e. the last term) in the low-energy limit,
$`\underset{p^20}{lim}{\displaystyle }{\displaystyle \frac{d^3k}{(2\pi )^3}}{\displaystyle \frac{1}{2}}C_V\delta ^{ab}S(k)S(kp)[\mathrm{\Gamma }_\mu (p,kp,k)\mathrm{\Gamma }_\nu (k+p,k,p)`$ (61)
$`+\mathrm{\Gamma }_\mu (k,kp,p)\mathrm{\Gamma }_\nu (p,k,k+p)+2\mathrm{\Gamma }_\mu (p,k,kp)\mathrm{\Gamma }_\nu (p,k+p,k)`$ (62)
$`\mathrm{\Gamma }_\mu (p,k,k+p)\mathrm{\Gamma }_\nu (p,kp,k)3\mathrm{\Gamma }_\mu (k,kp,p)\mathrm{\Gamma }_\nu (k+p,k,p)]=0,`$ (63)
where $`C_V`$ is the quadratic Casimir operator in the adjoint representation of the gauge group, $`f^{acd}f^{bcd}=C_V\delta ^{ab}`$.
To complete the proof that the infrared divergence indeed cancels in Eq.(53), we need to show that the remained terms are also free from infrared divergence. The remained terms is composed of the ghost field propagator, the ghost-ghost-gluon vertex and the composite vertices relevant to the Landau vector supersymmetric variation of the Yang-Mills term. The identity (15) yields<sup>§</sup><sup>§</sup>§ In fact, Eq.(15) yields two possibilities, $`p_\mu \mathrm{\Omega }_{\mu \nu }(p)=0`$ or proportional to $`p_\nu `$, but the tree level and one-loop results, $`\mathrm{\Omega }_{\mu \nu }^{(0)}(p)=p^2/mϵ_{\mu \nu \rho }p_\rho `$ and $`\mathrm{\Omega }_{\mu \nu }^{(1)}(p)=3/(4\pi )g^2C_V(p^2\delta _{\mu \nu }p_\mu p_\nu )`$, exclude the second one.
$`p_\mu \mathrm{\Omega }_{\mu \nu }(p)=0,`$ (64)
and hence $`\mathrm{\Omega }_{\mu \nu }`$ should be the following general form,
$`\mathrm{\Omega }_{\mu \nu }(p)=mϵ_{\mu \nu \rho }p_\rho B_1\left({\displaystyle \frac{p^2}{m^2}}\right)+(p^2\delta _{\mu \nu }p_\mu p_\nu )B_2\left({\displaystyle \frac{p^2}{m^2}}\right).`$ (65)
As for the composite ghost-gluon-gluon vertex associated with the Landau vector supersymmetry transformation, $`\stackrel{~}{\mathrm{\Gamma }}_{\mu \nu \rho }(p,q,r)`$, the identity (20) leads to a relation between $`\stackrel{~}{\mathrm{\Gamma }}_{\mu \nu \rho }`$ and the ghost-ghost-gluon vertex,
$`r_\rho \stackrel{~}{\mathrm{\Gamma }}_{\mu \nu \rho }(p,q,r)=ϵ_{\alpha \lambda \rho }p_\lambda q_\rho \left[S(p)D_{\mu \alpha }^1(p)\mathrm{\Gamma }_\nu (p,r,q)S(q)D_{\nu \alpha }^1(q)\mathrm{\Gamma }_\mu (q,r,p)\right].`$ (66)
Thus the infrared divergence-free of the ghost-ghost-gluon vertex $`\mathrm{\Gamma }_\mu `$ implies that $`\stackrel{~}{\mathrm{\Gamma }}_{\mu \nu \rho }`$ should also has no infrared divergence. The other argument in favour of the infrared divergence-free of $`\stackrel{~}{\mathrm{\Gamma }}_{\mu \nu \rho }`$ is the observation that at tree-level it is relevant to the three-gluon vertex of the three-dimensional Yang-Mills theory,
$`\stackrel{~}{\mathrm{\Gamma }}_{\mu \nu \rho }^{(0)}(p,q,r)`$ $`=`$ $`{\displaystyle \frac{i}{m}}ϵ_{\nu \lambda \sigma }r_\sigma \left[g_{\mu \lambda }(pq)_\rho +g_{\lambda \rho }(qr)_\mu +g_{\rho \mu }(rp)_\lambda \right]`$ (67)
$`=`$ $`{\displaystyle \frac{i}{m}}ϵ_{\nu \lambda \sigma }r_\sigma \mathrm{\Gamma }_{(YM)\mu \lambda \rho }^{(0)}(p,q,r),`$ (68)
where $`\mathrm{\Gamma }_{(YM)\mu \lambda \rho }^{(0)}`$ is the tree-level three-gluon vertex of the Yang-Mills part of TMYM. There is no infrared divergence in three dimensional Yang-Mills theory due to its superrenormalizability. The relation (68) will probably be modified at quantum level by quantum correction, but we still think that it gives a favourable support on the infrared divergence-free of $`\stackrel{~}{\mathrm{\Gamma }}_{\mu \nu \rho }`$.
The information collected in Eqs.(56), (60), (65), (66) and (68) on propagators and vertices enable us to use the inductive method to prove the infrared divergence-free of the remained terms of Eq.(53). The procedure is as following. We first show by concrete calculation that in dimensional regularization the terms other than the first and the last ones in Eq.(53) have the $`p^20`$ limit, then assume that this is satisfied at the $`n`$th order; After a lengthy analysis we find that at $`n+1`$ order the remained terms of Eq.(53) indeed have no infrared divergence. Based on this observation, we conclude that the cancellation of the infrared divergence has occurred in the whole vacuum polarization tensor.
Similar analysis are applied to the quantum three-gluon vertex. It is shown that with Eqs.(26) the infrared divergences coming from the ghost loop skeleton and the pure $`ϵ`$-parts of the gluon loop skeleton cancelIn concrete perturbative calculation, depending on the choice of a regularization scheme, Eq.(74) at one-loop may have a finite term, $`C_1ϵ_{\mu \nu \rho }`$$`+C_2/m[\delta _{\mu \nu }(pq)_\rho `$$`+\delta _{\nu \rho }(qr)_\mu +\delta _{\rho \mu }(rp)_\mu ]`$ with $`C_1`$ and $`C_2`$ being constants.,
$`\underset{p^2,q^2,r^20}{lim}\left[\mathrm{\Gamma }_{(gl)\mu \nu \rho }^{abc}(p,q,r)+\mathrm{\Gamma }_{(gh)\mu \nu \rho }^{abc}(p,q,r)\right]`$ (69)
$`=`$ $`\underset{p^2,q^2,r^20}{lim}\{{\displaystyle }{\displaystyle \frac{d^3k}{(2\pi )^3}}\mathrm{\Gamma }_{\mu \mu _1\mu _2}^{aa_1a_2}(p,pk,k)D_{\mu _2\rho _1}(k)\mathrm{\Gamma }_{\rho \rho _1\rho _2}^{ca_2b_1}(pq,k,p+q+k)`$ (71)
$`\times D_{\rho _2\nu _1}(p+q+k)\mathrm{\Gamma }_{\nu \nu _1\nu _2}^{bb_1a_1}(q,pqk,p+k)D_{\nu _2\mu _1}(p+k)`$
$``$ $`{\displaystyle }{\displaystyle \frac{d^3k}{(2\pi )^3}}[\mathrm{\Gamma }_\mu ^{aa_1a_2}(p,pk,k)S(k)\mathrm{\Gamma }_\rho ^{ca_2b_1}(pq,k,p+q+k)S(k+p+q)`$ (74)
$`\times \mathrm{\Gamma }_\nu ^{bb_1a_1}(q,kpq,k+p)S(k+p)+\mathrm{\Gamma }_\mu ^{aa_1a_2}(p,k+q,kpq)S(k+q)`$
$`\times \mathrm{\Gamma }_\nu ^{bb_1a_1}(q,k,kq)S(k)\mathrm{\Gamma }_\rho ^{ca_2b_1}(pq,k+p+q,k)S(k+p+q)]\}=0,`$
the three-gluon vertex function is thus infrared finite. In above equation, $`\mathrm{\Gamma }_{(gl)\mu \nu \rho }^{abc}`$ and $`\mathrm{\Gamma }_{(gh)\mu \nu \rho }^{abc}`$ denote the quantum three-gluon vertex contributed by the gluon loop skeleton and the ghost loop skeleton, respectively.
Having proved that the Landau vector supersymmetry protects the topologically massive Yang-Mills theory from getting the infrared divergence, in the following we shall have a look at the relation between the Landau vector supersymmetry and the cancellation theorem proposed by Pisarski and Rao.
First, the infrared cancellation theorem makes it possible for the Landau vector supersymmetry to persist after the quantum correction. The main content of Pisarski and Rao’s infrared divergence cancellation theorem is that in the Landau gauge the Chern-Simons term of TMYM keeps its classical form up to a finite wave function and vertex renormalization, i.e. the quantum effective action of Chern-Simons theory is
$`\mathrm{\Gamma }_{CS}[A,c,\overline{c},B]`$ $`\stackrel{\xi =0}{=}`$ $`i{\displaystyle d^3xZ_Aϵ_{\mu \nu \rho }\left(\frac{1}{2}A_\mu ^a_\nu A_\rho ^a+\frac{1}{3!}\frac{Z_V}{Z_A}g_Rf^{abc}A_\mu ^aA_\nu ^bA_\rho ^c\right)}`$ (76)
$`{\displaystyle d^3x\left[B^a_\mu A_\mu ^a+Z_c_\mu \overline{c}^a\left(_\mu c^a+\frac{Z_V^{}}{Z_c}g_Rf^{abc}A_\mu ^bc^c\right)\right]}`$
$`=`$ $`i{\displaystyle d^3xZ_Aϵ_{\mu \nu \rho }\left(\frac{1}{2}A_\mu ^a_\nu A_\rho ^a+\frac{1}{3!}Z_A^{1/2}gf^{abc}A_\mu ^aA_\nu ^bA_\rho ^c\right)}`$ (78)
$`{\displaystyle d^3x\left[B^a_\mu A_\mu ^a+Z_c_\mu \overline{c}^a\left(_\mu c^a+Z_A^{1/2}gf^{abc}A_\mu ^bc^c\right)\right]},`$
where the fields are all renormalized ones; $`g`$ and $`g_R`$ are the bare and the renormalized gauge couplings, respectively; $`Z_A`$ and $`Z_g`$ are the wave function renormalization constants for the gauge and ghost fields, respectively; $`Z_V`$ and $`Z_V^{}`$ are the vertex renormalization constants for the three-gluon vertex and the ghost-ghost-gluon field vertex, respectively. Further, the Slavnov-Taylor identity from the quantum BRST symmetry requires that $`Z_V/Z_A=Z_V^{}/Z_c`$ . The form (78) clearly shows that that the quantum Chern-Simons action has the following renormalized Landau vector supersymmetry,
$`V_{R\mu }A_\nu ^a`$ $`=`$ $`iZ_A^{1/2}Z_c^{1/2}ϵ_{\mu \nu \rho }_\rho c^a,V_{R\mu }c^a=0,V_{R\mu }\overline{c}^a=Z_c^{1/2}Z_A^{1/2}A_\mu ^a,`$ (79)
$`V_{R\mu }B^a`$ $`=`$ $`Z_A^{1/2}Z_c^{1/2}_\mu c^aZ_V^{}Z_A^{1/2}Z_c^{1/2}g_Rf^{abc}A_\mu ^bc^c`$ (80)
$`=`$ $`Z_A^{1/2}Z_c^{1/2}_\mu c^aZ_VZ_c^{1/2}Z_A^{1/2}g_Rf^{abc}A_\mu ^bc^c.`$ (81)
Since the Landau vectors supersymmetry transformation changes parity, so any parity-even terms generated from the quantum correction such as $`(F_{\mu \nu })^2`$ etc will break the Landau vector supersymmetry. Therefore, the above infrared divergence cancellation theorem guarantees that Landau vector supersymmetry is completely anomaly free.
On the other side, there had appeared several different ways proving that once the Landau vector supersymmetry is imposed, the Ward identity corresponding to it will enforce that the quantum Chern-Simons action to be the form of (78). The first method is combining the Landau vector supersymmetry and the BRST transformation invariance together to form a $`N=1`$ supersymmetry algebra, then with respect to this superalgebra, arranging the energy-momentum tensor, the ghost number current, the BRST current and a tensor current corresponding to the Landau vector supersymmetry into a supermultiplet. With the fact that the anomalies of the component currents in a supermultiplet also constitute a supermultiplet, it was shown by making use of various Ward identities that the quantum Chern-Simons theory is scale invariant, its beta function and the anomalous dimensions of every fields vanish identically. This means that the quantum Chern-Simons term keeps its classical form. The second way is considering the Ward identity implied by the Landau vector supersymmetry and the Schwinger-Dyson equations for the propagators of the ghost and gauge fields. This method elegantly manifested the vanishing of the radiative correction to the Chern-Simons theory. In addition, the concrete perturbative calculation in a regularization scheme preserving the Landau vector supersymmetry also explicitly confirms that the Chern-Simons action indeed receives no quantum correction. In fact, even one chooses a regularization schemes violating the Landau vector supersymmetry in the perturbative Chern-Simons theory, one typical example of which is the hybrid regularization of higher covariant derivative plus dimensional regularization with the Yang-Mills action as the higher covariant derivative term, the explicit calculation shows that after the regulator is removed the theory still keeps its classical form up to a finite renormalization of the gauge coupling. Thus the cancellation theorem is an inevitable consequence of the Landau vector supersymmetry. Based on above arguments considered from both sides, we conclude that the existence of Landau vector supersymmetry of Chern-Simons theory is equivalent to the cancellation cancellation theorem proposed by Pisarski and Rao.
In summary, we have verified that the infrared divergence cancellation of topologically massive Yang-Mills theory in the Landau gauge originates completely from a new vector symmetry in the gauge-fixed Chern-Simons term. This is an unusual property of TMYM. It is well known that in four-dimensional case, the infrared divergences arise in a massless field theory. They cancel out in the scattering cross sections when the degeneracy of a real particle with massless particle states is considered as summarized by the celebrated Kinoshita-Lee-Nauenberg theorem, where cancellation has no explicit relevance with the symmetries of the theory. While here in three-dimensional TMYM, the cancellation of the infrared divergence can adhere to certain symmetry of the theory. Thus it is significant to point out this exotic feature.
Furthermore, we have analyzed the relation between the Landau vector supersymmetry and the infrared divergence cancellation theorem proposed earlier by Pisarski and Rao. On one hand, the cancelation theorem ensures that the Landau vector supersymmetry will persist after quantum correction; On the other hand, the Landau vector supersymmetry determine that infrared divergences are doomed to cancel. Consequently, Chern-Simons theory can be defined as the large topological mass limit of TMYM and Pisarski and Rao’s cancellation theorem arises. However, there is a difference between the Landau vector supersymmetry and the cancellation theorem: the Landau vector supersymmetry has displayed the origin of the infrared divergence cancellation of TMYM in the Landau gauge, while the cancellation theorem has only emphasized the consequence of the infrared divergence cancellation .
###### Acknowledgements.
This work is supported by the Natural Sciences and Engineering Research Council of Canada. I would like to thank Professor G. Kunstatter for useful discussions and Professors M. Chaichian and R. Kobes for encouragement and help.
|
warning/0004/hep-th0004166.html
|
ar5iv
|
text
|
# Cosmological dynamics on the brane
## I Introduction
Einstein’s theory of general relativity breaks down at high enough energies, and is likely to be a limit of a more general theory. Recent developments in string theory indicate that gravity may be a truly higher-dimensional theory, becoming effectively 4-dimensional at lower energies. These exciting theoretical developments need to be accompanied by efforts to test such higher-dimensional theories against their cosmological and astrophysical implications. In that spirit, we investigate here a particular class of models, showing how their dynamical properties generalize those of Einstein’s theory, and discussing the broad implications of these generalizations for cosmological dynamics.
In many higher-dimensional gravity theories inspired by string theory, the matter fields are confined to a 3-brane in $`1+3+d`$ dimensions, while the gravitational field can propagate also in the $`d`$ extra dimensions (see, e.g., ). It is not necessary for the $`d`$ extra dimensions to be small, or even finite: recently Randall and Sundrum have shown that for $`d=1`$, gravity can be localized on a single 3-brane even when the fifth dimension is infinite (see also ). An elegant geometric formulation and generalization of the Randall-Sundrum scenario has been given by Shiromizu, Maeda and Sasaki . The Friedmann equation on the brane in these models is modified by both high-energy matter terms and a term carrying nonlocal bulk effects onto the brane. The Friedmann brane models have been extensively investigated (see, e.g., ), and inflationary scalar perturbations in these models have also been considered . The models are compatible with observations subject to reasonable constraints on the parameters. A broader study of cosmological dynamics, i.e., for induced metrics more general than the simple Friedmann case, has not been done. In particular, the analysis of perturbed Friedmann models also remains to be done. (Considerable work has been done on perturbations of flat brane metrics; see, e.g., .)
In this paper, we initiate a study of nonlinear and perturbed cosmological dynamics in Randall-Sundrum-type brane-world models, generalizing some important results in general relativity. We find the bulk corrections to the propagation and constraint equations, using the covariant Lagrangian approach . This approach is well-suited to identifying the geometric and physical quantities that determine inhomogeneity and anisotropy on the brane, and it is also the basis for a gauge-invariant perturbation theory . Our first task is to identify and interpret the covariant physical content of the bulk effects on the brane. Local effects lead to quadratic corrections of the density, pressure and energy flux. The nonlocal effects of the free gravitational field in the bulk are transmitted by a Weyl projection term, which we decompose into energy density, energy flux and anisotropic stress parts. We calculate the gravitational (tidal) and non-gravitational acceleration of fluid world-lines, finding the role of the nonlocal energy density in localization of gravity, and showing how the world-lines have a non-gravitational acceleration off the brane at high energies. During inflation, the acceleration is directed towards the brane.
We derive the propagation (‘conservation’) equations governing the nonlocal energy density and flux parts; the evolution of the anisotropic stress part is not determined on the brane. These nonlocal terms also enter into crucial dynamical equations, such as the Raychaudhuri equation and the shear propagation equation, and can lead to important changes from the general relativistic case. For example, it is possible via the nonlocal term to avoid the initial singularity in a nonrotating model without cosmological constant. Nonlocal effects also mean that isotropy of the cosmic microwave background (CMB) may no longer guarantee a Friedmann geometry.
The covariant nonlinear equations lead to a covariant and gauge-invariant description of perturbations on the brane. We derive the equations governing adiabatic scalar perturbations, which are not in general closed on the brane, because of nonlocal effects. However, on super-Hubble scales, the density perturbations satisfy a decoupled third-order equation, with an additional nonlocal degree of freedom, and can therefore be evaluated by brane observers. Tensor perturbations cannot be determined by brane observers on any scales. The local bulk effects tend to enhance tensor perturbations during non-inflationary expansion and suppress them during inflation. Nonlocal bulk effects can in principle act either way. Vorticity on the brane decays as in general relativity, but bulk effects act as a source for the gravito-magnetic field, and hence vector perturbations, on the brane.
Our results remain incomplete in one fundamental aspect, i.e., we do not provide a description of the gravitational field in the bulk, but confine our investigations to effects that can be measured by brane-observers. In order to fill this gap, one would need to study the off-brane derivatives of the curvature, which are given in general in . This is an important topic for further research.
The 5-dimensional field equations are Einstein’s equations, with a (negative) bulk cosmological constant $`\stackrel{~}{\mathrm{\Lambda }}`$ and brane energy-momentum as source:
$$\stackrel{~}{G}_{AB}=\stackrel{~}{\kappa }^2\left[\stackrel{~}{\mathrm{\Lambda }}\stackrel{~}{g}_{AB}+\delta (\chi )\left\{\lambda g_{AB}+T_{AB}\right\}\right].$$
(1)
The tildes denote the bulk (5-dimensional) generalization of standard general relativity quantities, and $`\stackrel{~}{\kappa }^2=8\pi /\stackrel{~}{M}_\mathrm{p}^3`$, where $`\stackrel{~}{M}_\mathrm{p}`$ is the fundamental 5-dimensional Planck mass, which is typically much less than the effective Planck mass on the brane, $`M_\mathrm{p}=1.2\times 10^{19}`$ GeV. The brane is given by $`\chi =0`$, so that a natural choice of coordinates is $`x^A=(x^\mu ,\chi )`$, where $`x^\mu =(t,x^i)`$ are spacetime coordinates on the brane. The brane tension is $`\lambda `$, and $`g_{AB}=\stackrel{~}{g}_{AB}n_An_B`$ is the induced metric on the brane, with $`n_A`$ the spacelike unit normal to the brane. Matter fields confined to the brane make up the brane energy-momentum tensor $`T_{AB}`$ (with $`T_{AB}n^B=0`$).
Although it is usually assumed that the spacetime is exactly anti-de Sitter in the absence of a brane ($`\lambda =0=T_{AB}`$), this is not necessarily the case. The brane-free bulk metric can be any solution of $`\stackrel{~}{G}_{AB}=\stackrel{~}{\kappa }^2\stackrel{~}{\mathrm{\Lambda }}\stackrel{~}{g}_{AB}`$, including non-conformally flat solutions. When the induced metric on the brane is Friedmann, then the 5-dimensional Schwarzschild-anti-de Sitter metric is a solution of Eq. (1. However, more general bulk metrics are in principle possible.
The field equations induced on the brane are derived via an elegant geometric approach by Shiromizu et al. , using the Gauss-Codazzi equations, matching conditions and $`Z_2`$ symmetry. The result is a modification of the standard Einstein equations, with the new terms carrying bulk effects onto the brane:<sup>*</sup><sup>*</sup>* For clarity and consistency, we have changed the notation of .
$$G_{\mu \nu }=\mathrm{\Lambda }g_{\mu \nu }+\kappa ^2T_{\mu \nu }+\stackrel{~}{\kappa }^4S_{\mu \nu }_{\mu \nu },$$
(2)
where $`\kappa ^2=8\pi /M_\mathrm{p}^2`$. The energy scales are related to each other via
$$\lambda =6\frac{\kappa ^2}{\stackrel{~}{\kappa }^4},\mathrm{\Lambda }=\frac{4\pi }{\stackrel{~}{M}_\mathrm{p}^3}\left[\stackrel{~}{\mathrm{\Lambda }}+\left(\frac{4\pi }{3\stackrel{~}{M}_\mathrm{p}^{\mathrm{\hspace{0.17em}3}}}\right)\lambda ^2\right].$$
(3)
The bulk corrections to the Einstein equations on the brane are of two forms: firstly, the matter fields contribute local quadratic energy-momentum corrections via the tensor $`S_{\mu \nu }`$, and secondly, there are nonlocal effects from the free gravitational field in the bulk, transmitted via the projection $`_{\mu \nu }`$ of the bulk Weyl tensor. The matter corrections are given by
$$S_{\mu \nu }=\frac{1}{12}T_\alpha {}_{}{}^{\alpha }T_{\mu \nu }^{}\frac{1}{4}T_{\mu \alpha }T^\alpha {}_{\nu }{}^{}+\frac{1}{24}g_{\mu \nu }[3T_{\alpha \beta }T^{\alpha \beta }(T_\alpha {}_{}{}^{\alpha })^2].$$
(4)
The projection of the bulk Weyl tensor is This projection is called the ‘electric’ part of the Weyl tensor in , but the term is potentially misleading, since the electric part is associated with projection on a timelike vector , and $`n^A`$ is spacelike. $`_{\mu \nu }`$ should not be confused with the electric part of the brane Weyl tensor, $`E_{\mu \nu }`$, defined below.
$$_{AB}=\stackrel{~}{C}_{ACBD}n^Cn^D,$$
(5)
which is symmetric (Square) round brackets enclosing indices denote (anti-)symmetrization. and traceless ($`_{[AB]}=0=_A^A`$) and without components orthogonal to the brane, so that $`_{AB}n^B=0`$ and $`_{AB}_{\mu \nu }\delta _A{}_{}{}^{\mu }\delta _{B}^{}^\nu `$ as $`\chi 0`$. The Weyl tensor $`\stackrel{~}{C}_{ABCD}`$ represents the free, nonlocal gravitational field in the bulk. The local part of the bulk gravitational field is the Einstein tensor $`\stackrel{~}{G}_{AB}`$, which is determined via the bulk field equations (1). Thus $`_{\mu \nu }`$ transmits nonlocal gravitational degrees of freedom from the bulk to the brane, including tidal (or Coulomb), gravito-magnetic and transverse traceless (gravitational wave) effects.
## II Covariant decomposition of bulk effects
We now provide a covariant decomposition of the bulk correction tensors given by Shiromizu et al. .
### A Local bulk effects
For any matter fields (scalar fields, perfect fluids, kinetic gases, dissipative fluids, etc.), including a combination of different fields, the general form of the brane energy-momentum tensor can be covariantly given as
$$T_{\mu \nu }=\rho u_\mu u_\nu +ph_{\mu \nu }+\pi _{\mu \nu }+2q_{(\mu }u_{\nu )}.$$
(6)
The decomposition is irreducible for any chosen 4-velocity $`u^\mu `$. Here $`\rho `$ and $`p`$ are the energy density and isotropic pressure, and $`h_{\mu \nu }=g_{\mu \nu }+u_\mu u_\nu `$ projects orthogonal to $`u^\mu `$. The energy flux obeys $`q_\mu =q_\mu `$, and the anisotropic stress obeys $`\pi _{\mu \nu }=\pi _{\mu \nu }`$, where angle brackets denote the projected, symmetric and tracefree part:
$`V_\mu =h_\mu {}_{}{}^{\nu }V_{\nu }^{},W_{\mu \nu }=\left[h_{(\mu }{}_{}{}^{\alpha }h_{\nu )}^{}{}_{}{}^{\beta }\frac{1}{3}h^{\alpha \beta }h_{\mu \nu }\right]W_{\alpha \beta }.`$
Equations (4) and (6) imply the irreducible decomposition
$`S_{\mu \nu }`$ $`=`$ $`\frac{1}{24}\left[2\rho ^23\pi _{\alpha \beta }\pi ^{\alpha \beta }\right]u_\mu u_\nu +\frac{1}{24}\left[2\rho ^2+4\rho p+\pi _{\alpha \beta }\pi ^{\alpha \beta }4q_\alpha q^\alpha \right]h_{\mu \nu }`$ (8)
$`\frac{1}{12}(\rho +2p)\pi _{\mu \nu }+\pi _{\alpha \mu }\pi _\nu {}_{}{}^{\alpha }+q_\mu q_\nu +\frac{1}{3}\rho q_{(\mu }u_{\nu )}\frac{1}{12}q^\alpha \pi _{\alpha (\mu }u_{\nu )}.`$
For a perfect fluid or minimally-coupled scalar field
$`S_{\mu \nu }=\frac{1}{12}\rho ^2u_\mu u_\nu +\frac{1}{12}\rho \left(\rho +2p\right)h_{\mu \nu },`$in agreement with . The quadratic energy-momentum corrections to standard general relativity will be significant for $`\stackrel{~}{\kappa }^4\rho ^2\kappa ^2\rho `$, i.e., in the high-energy regime
$`\rho \lambda \left({\displaystyle \frac{\stackrel{~}{M}_\mathrm{p}}{M_\mathrm{p}}}\right)^2\stackrel{~}{M}_\mathrm{p}^{\mathrm{\hspace{0.17em}4}}.`$
### B Nonlocal bulk effects
The symmetry properties of $`_{\mu \nu }`$ imply that in general we can decompose it irreducibly with respect to a chosen 4-velocity field $`u^\mu `$ as
$$_{\mu \nu }=\left(\frac{\stackrel{~}{\kappa }}{\kappa }\right)^4\left[𝒰\left(u_\mu u_\nu +\frac{1}{3}h_{\mu \nu }\right)+𝒫_{\mu \nu }+2𝒬_{(\mu }u_{\nu )}\right].$$
(9)
The factor $`(\stackrel{~}{\kappa }/\kappa )^4`$ is introduced for dimensional reasons. Here
$`𝒰=\left({\displaystyle \frac{\kappa }{\stackrel{~}{\kappa }}}\right)^4_{\mu \nu }u^\mu u^\nu `$is an effective nonlocal energy density on the brane, arising from the free gravitational field in the bulk. This nonlocal energy density need not be positive (see below). There is an effective nonlocal anisotropic stress
$`𝒫_{\mu \nu }=\left({\displaystyle \frac{\kappa }{\stackrel{~}{\kappa }}}\right)^4_{\mu \nu }`$on the brane, arising from the free gravitational field in the bulk, and
$`𝒬_\mu =\left({\displaystyle \frac{\kappa }{\stackrel{~}{\kappa }}}\right)^4_{\mu \nu }u^\nu `$is an effective nonlocal energy flux on the brane, arising from the free gravitational field in the bulk.
If the induced metric on the brane is flat, and the bulk is anti-de Sitter, as in the original Randall-Sundrum scenario , then $`_{\mu \nu }=0`$. Treating this as a background, it follows that tensor (transverse traceless) perturbations on the brane arising from nonlocal bulk degrees of freedom are given by
$$𝒰=0=𝒬_\mu ,\text{D}^\nu 𝒫_{\mu \nu }=0,$$
(10)
where $`\text{D}_\mu `$ is the totally projected part of the brane covariant derivative:
$`\text{D}_\mu F^\alpha \mathrm{}{}_{\mathrm{}\beta }{}^{}=h_\mu {}_{}{}^{\nu }h_{}^{\alpha }{}_{\gamma }{}^{}\mathrm{}h_\beta {}_{}{}^{\delta }_{\nu }^{}F^\gamma \mathrm{}{}_{\mathrm{}\delta }{}^{}.`$Equation (10) provides a covariant characterization of brane tensor perturbations on an anti-de Sitter background.
In cosmology, the background induced metric is not flat, but a spatially homogeneous and isotropic Friedmann model, for which
$$\text{D}_\mu 𝒰=𝒬_\mu =𝒫_{\mu \nu }=0.$$
(11)
Thus for a perturbed Friedmann model, the nonlocal bulk effects are covariantly and gauge-invariantly described by the first-order quantities $`\text{D}_\mu 𝒰`$, $`𝒬_\mu `$, $`𝒫_{\mu \nu }`$. Since $`𝒰0`$ in general in the Friedmann background , it follows that for tensor perturbations on the brane
$$\text{D}_\mu 𝒰=0=𝒬_\mu ,\text{D}^\nu 𝒫_{\mu \nu }=0.$$
(12)
Scalar perturbations (Coulomb-like bulk effects) will be characterized by
$$𝒬_\mu =\text{D}_\mu 𝒬,𝒫_{\mu \nu }=\text{D}_\mu \text{D}_\nu 𝒫,$$
(13)
for some scalars $`𝒬`$ and $`𝒫`$, while for vector perturbations (gravito-magnetic-like bulk effects)
$$\text{D}^\mu 𝒬_\mu =0,𝒫_{\mu \nu }=\text{D}_\mu 𝒫_\nu ,\text{D}^\mu 𝒫_\mu =0.$$
(14)
### C Gravitational and non-gravitational acceleration
In order to find the role of bulk effects in tidal acceleration on the brane, we start from the relation
$`_{\mu \nu }u^\mu u^\nu =\underset{\chi 0}{lim}\stackrel{~}{𝒞}_{ABCD}\stackrel{~}{u}^An^B\stackrel{~}{u}^Cn^D,`$where $`\stackrel{~}{u}^A`$ is an extension off the brane of the 4-velocity (with $`\stackrel{~}{u}^An_A=0`$). The tidal acceleration in the $`n^A`$ direction measured by comoving observers is $`n_A\stackrel{~}{R}^A{}_{BCD}{}^{}\stackrel{~}{u}_{}^{B}n^C\stackrel{~}{u}^D`$. Now
$`\stackrel{~}{R}_{ABCD}=\stackrel{~}{C}_{ABCD}+\frac{2}{3}\left\{\stackrel{~}{g}_{A[C}\stackrel{~}{R}_{D]B}+\stackrel{~}{g}_{B[D}\stackrel{~}{R}_{C]A}\right\}\frac{1}{6}\stackrel{~}{R}\stackrel{~}{g}_{A[C}\stackrel{~}{g}_{D]B},`$so that by the field equation (1) (and recalling that $`T_{AB}n^B=0`$),
$`\stackrel{~}{R}_{ABCD}n^A\stackrel{~}{u}^Bn^C\stackrel{~}{u}^D=_{AB}\stackrel{~}{u}^A\stackrel{~}{u}^B+\frac{1}{6}\stackrel{~}{\kappa }^2\stackrel{~}{\mathrm{\Lambda }}.`$Thus the comoving tidal acceleration on the brane, in the off-brane direction, is
$$\left(\frac{\stackrel{~}{\kappa }}{\kappa }\right)^4𝒰+\frac{1}{6}\stackrel{~}{\kappa }^2\stackrel{~}{\mathrm{\Lambda }}.$$
(15)
Since $`\stackrel{~}{\mathrm{\Lambda }}`$ is negative, it contributes to acceleration towards the brane. This reflects the confining role of the negative bulk cosmological constant on the gravitational field in the warped metric models of Randall-Sundrum type. Equation (15) shows that localization of the gravitational field near the brane is enhanced by negative $`𝒰`$, while positive $`𝒰`$ acts against confinement, by contributing to tidal acceleration away from the brane. This picture is consistent with a Newtonian interpretation, where the gravitational field carries negative energy density.
The non-gravitational acceleration of fluid world-lines on the brane is
$`\stackrel{~}{A}^A=\stackrel{~}{u}^B\stackrel{~}{}_B\stackrel{~}{u}^A\text{ at }\chi =0.`$Locally, near the brane, the metric may be written as
$`d\stackrel{~}{s}^2=d\chi ^2+g_{\mu \nu }(x^\alpha ,\chi )dx^\mu dx^\nu ,`$so that
$`\stackrel{~}{\mathrm{\Gamma }}^A{}_{\mu \nu }{}^{}(x,0)=\mathrm{\Gamma }^\alpha {}_{\mu \nu }{}^{}(x)\delta _\alpha {}_{}{}^{A}\frac{1}{2}g_{\mu \nu ,\chi }(x,0)\delta _\chi {}_{}{}^{A}.`$This allows us to express the 5-dimensional acceleration $`\stackrel{~}{A}^A`$ in terms of the 4-dimensional acceleration $`A^\mu =u^\nu _\nu u^\mu `$ on the brane. First, we use the covariant form $`g_{\mu \nu ,\chi }=_𝐧g_{\mu \nu }`$, where $`_𝐧`$ is the Lie derivative along $`n^A`$. Then we use the expression for the extrinsic curvature of the brane
$`K_{\mu \nu }^+=\frac{1}{2}\underset{\chi 0^+}{lim}_𝐧g_{\mu \nu },`$which leads to
$$\stackrel{~}{A}^A(x,0^+)=A^\mu (x)\delta _\mu {}_{}{}^{A}K_{\mu \nu }^+(x)u^\mu (x)u^\nu (x)n^A(x,0^+),$$
(16)
on the brane. The extrinsic curvature is given in terms of the brane tension and energy-momentum by
$`K_{\mu \nu }^+=\frac{1}{6}\stackrel{~}{\kappa }^2\left[\lambda g_{\mu \nu }+3T_{\mu \nu }+(\rho 3p)g_{\mu \nu }\right].`$Substituting in Eq. (16), we find
$$\stackrel{~}{A}^A(x,0^+)=A^\mu (x)\delta _\mu {}_{}{}^{A}+\frac{1}{6}\stackrel{~}{\kappa }^2[2\rho (x)+3p(x)\lambda ]n^A(x,0^+).$$
(17)
It follows that $`n_A\stackrel{~}{A}^A`$ is nonzero on the brane, i.e., there is a non-gravitational acceleration of fluid world-lines orthogonal to the brane. The direction depends on the sign of $`2\rho +3p\lambda `$ : for $`2\rho +3p\lambda >0`$, the transverse acceleration is away from the brane, while for $`2\rho +3p\lambda <0`$, it is towards the brane. If the pressure is positive, then at high energies $`2\rho +3p\lambda >0`$, so that either the brane must accelerate, or there must be a non-gravitational mechanism for keeping matter on the brane.
Inflationary expansion, when pressure is negative, can change this situation. Inflation on the brane is characterized by
$$p<\left(\frac{\lambda +2\rho }{\lambda +\rho }\right)\frac{\rho }{3},$$
(18)
and this condition implies that $`2\rho +3p\lambda <0`$. Thus during inflation on the brane, the transverse acceleration is towards the brane; inflation acts as a non-gravitational mechanism keeping matter on the brane.
### D Effective total energy-momentum tensor
All the bulk corrections may be consolidated into effective total energy density, pressure, anisotropic stress and energy flux, as follows. The modified Einstein equations take the standard Einstein form with a redefined energy-momentum tensor:
$$G_{\mu \nu }=\mathrm{\Lambda }g_{\mu \nu }+\kappa ^2T_{\mu \nu }^{\mathrm{tot}},$$
(19)
where
$$T_{\mu \nu }^{\mathrm{tot}}=T_{\mu \nu }+\frac{\stackrel{~}{\kappa }^4}{\kappa ^2}S_{\mu \nu }\frac{1}{\kappa ^2}_{\mu \nu }.$$
(20)
Then
$`\rho ^{\mathrm{tot}}`$ $`=`$ $`\rho +{\displaystyle \frac{\stackrel{~}{\kappa }^4}{\kappa ^6}}\left[{\displaystyle \frac{\kappa ^4}{24}}\left(2\rho ^23\pi _{\mu \nu }\pi ^{\mu \nu }\right)+𝒰\right]`$ (21)
$`p^{\mathrm{tot}}`$ $`=`$ $`p+{\displaystyle \frac{\stackrel{~}{\kappa }^4}{\kappa ^6}}\left[{\displaystyle \frac{\kappa ^4}{24}}\left(2\rho ^2+4\rho p+\pi _{\mu \nu }\pi ^{\mu \nu }4q_\mu q^\mu \right)+\frac{1}{3}𝒰\right]`$ (22)
$`\pi _{\mu \nu }^{\mathrm{tot}}`$ $`=`$ $`\pi _{\mu \nu }+{\displaystyle \frac{\stackrel{~}{\kappa }^4}{\kappa ^6}}[{\displaystyle \frac{\kappa ^4}{12}}\{(\rho +3p)\pi _{\mu \nu }+\pi _{\alpha \mu }\pi _\nu {}_{}{}^{\alpha }+q_\mu q_\nu \}+𝒫_{\mu \nu }]`$ (23)
$`q_\mu ^{\mathrm{tot}}`$ $`=`$ $`q_\mu +{\displaystyle \frac{\stackrel{~}{\kappa }^4}{\kappa ^6}}\left[{\displaystyle \frac{\kappa ^4}{24}}\left(4\rho q_\mu \pi _{\mu \nu }q^\nu \right)+𝒬_\mu \right].`$ (24)
(Note that $`\stackrel{~}{\kappa }^4/\kappa ^6`$ is dimensionless.)
These general expressions simplify in the case of a perfect fluid (or minimally coupled scalar field, or isotropic one-particle distribution function), i.e., for $`q_\mu =0=\pi _{\mu \nu }`$. However, the total energy flux and anisotropic stress do not vanish in this case in general:
$`q_\mu ^{\mathrm{tot}}={\displaystyle \frac{\stackrel{~}{\kappa }^4}{\kappa ^6}}𝒬_\mu ,\pi _{\mu \nu }^{\mathrm{tot}}={\displaystyle \frac{\stackrel{~}{\kappa }^4}{\kappa ^6}}𝒫_{\mu \nu }.`$Thus nonlocal bulk effects can contribute to effective imperfect fluid terms even when the matter on the brane has perfect fluid form.
## III Local and nonlocal conservation equations
As a consequence of the form of the bulk energy-momentum tensor in Eq. (1) and of $`Z_2`$ symmetry, it follows that the brane energy-momentum tensor separately satisfies the conservation equations, i.e.,
$$^\nu T_{\mu \nu }=0.$$
(25)
Then the Bianchi identities on the brane imply that the projected Weyl tensor obeys the constraint
$$^\mu _{\mu \nu }=\stackrel{~}{\kappa }^4^\mu S_{\mu \nu },$$
(26)
which shows that its longitudinal part is sourced by quadratic energy-momentum terms, including spatial gradients and time derivatives. Thus evolution and inhomogeneity in the matter fields generates nonlocal Coulomb-like gravitational effects in the bulk, which ‘backreact’ on the brane.
The brane energy-momentum tensor and the consolidated effective energy-momentum tensor are both conserved separately. Conservation of $`T_{\mu \nu }`$ gives the standard general relativity energy and momentum conservation equations
$`\dot{\rho }+\mathrm{\Theta }(\rho +p)+\text{D}^\mu q_\mu +2A^\mu q_\mu +\sigma ^{\mu \nu }\pi _{\mu \nu }=0,`$ (27)
$`\dot{q}_\mu +\frac{4}{3}\mathrm{\Theta }q_\mu +\text{D}_\mu p+(\rho +p)A_\mu +\text{D}^\nu \pi _{\mu \nu }+A^\nu \pi _{\mu \nu }+\sigma _{\mu \nu }q^\nu [\omega ,q]_\mu =0.`$ (28)
A dot denotes $`u^\nu _\nu `$, $`\mathrm{\Theta }=\text{D}^\mu u_\mu `$ is the volume expansion rate of the $`u^\mu `$ congruence, $`A_\mu =\dot{u}_\mu =A_\mu `$ is its 4-acceleration, $`\sigma _{\mu \nu }=\text{D}_\mu u_\nu `$ is its shear rate, and $`\omega _\mu =\frac{1}{2}\text{curl}u_\mu =\omega _\mu `$ is its vorticity rate. The covariant spatial curl is given by
$`\text{curl}V_\mu =\epsilon _{\mu \alpha \beta }\text{D}^\alpha V^\beta ,\text{curl}W_{\mu \nu }=\epsilon _{\alpha \beta (\mu }\text{D}^\alpha W^\beta {}_{\nu )}{}^{},`$where $`\epsilon _{\mu \nu \sigma }`$ is the projected alternating tensor. The covariant cross-product is
$`[V,Y]_\mu =\epsilon _{\mu \alpha \beta }V^\alpha Y^\beta .`$
The conservation of $`T_{\mu \nu }^{\mathrm{tot}}`$ gives, upon using Eqs. (21)–(28), a propagation equation for the nonlocal energy density $`𝒰`$:
$`\dot{𝒰}+\frac{4}{3}\mathrm{\Theta }𝒰+\text{D}^\mu 𝒬_\mu +2A^\mu 𝒬_\mu +\sigma ^{\mu \nu }𝒫_{\mu \nu }`$ (29)
$`=\frac{1}{24}\kappa ^4[6\pi ^{\mu \nu }\dot{\pi }_{\mu \nu }+6(\rho +p)\sigma ^{\mu \nu }\pi _{\mu \nu }+2\mathrm{\Theta }(2q^\mu q_\mu \pi ^{\mu \nu }\pi _{\mu \nu })`$ (30)
$`+2A^\mu q^\nu \pi _{\mu \nu }+4q^\mu \text{D}_\mu \rho +q^\mu \text{D}^\nu \pi _{\mu \nu }+\pi ^{\mu \nu }\text{D}_\mu q_\nu 2\sigma ^{\mu \nu }\pi _{\alpha \mu }\pi _\nu {}_{}{}^{\alpha }2\sigma ^{\mu \nu }q_\mu q_\nu ],`$ (31)
and a propagation equation for the nonlocal energy flux $`𝒬_\mu `$:
$`\dot{𝒬}_\mu +\frac{4}{3}\mathrm{\Theta }𝒬_\mu +\frac{1}{3}\text{D}_\mu 𝒰+\frac{4}{3}𝒰A_\mu +\text{D}^\nu 𝒫_{\mu \nu }+A^\nu 𝒫_{\mu \nu }+\sigma _{\mu \nu }𝒬^\nu [\omega ,𝒬]_\mu `$ (32)
$`=\frac{1}{24}\kappa ^4[4(\rho +p)\text{D}_\mu \rho +q^\nu \dot{\pi }_{\mu \nu }+\pi _\mu {}_{}{}^{\nu }\text{D}_{\nu }^{}(2\rho +5p)+6(\rho +p)\text{D}^\nu \pi _{\mu \nu }`$ (33)
$`\frac{2}{3}\pi ^{\alpha \beta }\left(\text{D}_\mu \pi _{\alpha \beta }+3\text{D}_\alpha \pi _{\beta \mu }\right)3\pi _{\mu \alpha }\text{D}_\beta \pi ^{\alpha \beta }+\frac{28}{3}q^\nu \text{D}_\mu q_\nu `$ (34)
$`+4\rho A^\nu \pi _{\mu \nu }3\pi _{\mu \alpha }A_\beta \pi ^{\alpha \beta }+\frac{8}{3}A_\mu \pi ^{\alpha \beta }\pi _{\alpha \beta }\pi _{\mu \alpha }\sigma ^{\alpha \beta }q_\beta +\sigma _{\mu \alpha }\pi ^{\alpha \beta }q_\beta +\pi _{\mu \nu }[\omega ,q]^\nu `$ (35)
$`\epsilon _{\mu \alpha \beta }\omega ^\alpha \pi ^{\beta \nu }q_\nu +4(\rho +p)\mathrm{\Theta }q_\mu +6q_\mu A^\nu q_\nu +\frac{14}{3}A_\mu q^\nu q_\nu +4q_\nu \sigma ^{\alpha \beta }\pi _{\alpha \beta }].`$ (36)
These equations may be thought of as conservation equations on the brane for the nonlocal energy density and energy flux due to the free gravitational field in the bulk. In general, the 4 independent equations determine 4 of the 9 independent components of $`_{\mu \nu }`$ on the brane. What is missing, is an evolution equation for $`𝒫_{\mu \nu }`$ (which has up to 5 independent components). Thus in general, the projection of the 5-dimensional field equations onto the brane does not lead to a closed system. Nor could we expect this to be the case, since there are bulk degrees of freedom whose impact on the brane cannot be predicted by brane observers. These degrees of freedom could arise from propagating gravity waves in the bulk, possibly generated by inhomogeneity on the brane itself. The point is that waves which penetrate the 5th dimension are governed by off-brane bulk dynamical equations. Our decomposition of $`_{\mu \nu }`$ has shown that the evolution of the nonlocal energy density and flux (associated with the scalar and vector parts of $`_{\mu \nu }`$) is determined on the brane, while the evolution of the nonlocal anisotropic stress (associated with the tensor part of $`_{\mu \nu }`$) is not.
If the nonlocal anisotropic stress contribution from the bulk field vanishes, i.e., if
$`𝒫_{\mu \nu }=0,`$then the evolution of $`_{\mu \nu }`$ is determined by Eqs. (31) and (36). A special case of this arises when the induced metric on the brane is Friedmann, i.e., when Eq. (11) holds. Then $`_{\mu \nu }`$ has only 1 independent component, $`𝒰`$, and it is determined by Eq. (31) (see below), with Eq. (36) reducing to $`0=0`$. Another case when the equations close on the brane is when the brane is static (see ).
In the perfect fluid case, the conservation equations (27)–(36) reduce to:
$`\dot{\rho }+\mathrm{\Theta }(\rho +p)=0,`$ (37)
$`\text{D}_\mu p+(\rho +p)A_\mu =0.`$ (38)
For a minimally-coupled scalar field, $`\rho =\frac{1}{2}_\mu \phi ^\mu \phi +V(\phi )`$ and $`p=\frac{1}{2}_\mu \phi ^\mu \phi V(\phi )`$. In the adiabatic case, Eq. (38) gives
$$A_\mu =\frac{c_\mathrm{s}^2}{\rho +p}\text{D}_\mu \rho ,c_\mathrm{s}^2=\frac{\dot{p}}{\dot{\rho }}.$$
(39)
The nonlocal conservation equations (31) and (36) reduce to
$`\dot{𝒰}+\frac{4}{3}\mathrm{\Theta }𝒰+\text{D}^\mu 𝒬_\mu +2A^\mu 𝒬_\mu +\sigma ^{\mu \nu }𝒫_{\mu \nu }=0,`$ (40)
$`\dot{𝒬}_\mu +\frac{4}{3}\mathrm{\Theta }𝒬_\mu +\frac{1}{3}\text{D}_\mu 𝒰+\frac{4}{3}𝒰A_\mu +\text{D}^\nu 𝒫_{\mu \nu }+A^\nu 𝒫_{\mu \nu }+\sigma _{\mu \nu }𝒬^\nu [\omega ,𝒬]_\mu `$ (41)
$`=\frac{1}{6}\kappa ^4(\rho +p)\text{D}_\mu \rho .`$ (42)
Equation (42) shows that if $`_{\mu \nu }=0`$ and the brane energy-momentum tensor has perfect fluid form, then the density $`\rho `$ must be homogeneous . The converse does not hold, i.e., homogeneous density does not in general imply vanishing $`_{\mu \nu }`$. This is readily apparent from Eq. (42). A simple example is provided by the Friedmann case: Eq. (42) is trivially satisfied, while Eq. (40) gives the ‘dark radiation’ solution
$$𝒰=𝒰_o\left(\frac{a_o}{a}\right)^4.$$
(43)
A simple generalization of the Friedmann case is the purely Coulomb-like case, $`𝒬_\mu =0=𝒫_{\mu \nu }`$. Equation (43) still holds, but with $`a`$ an average scale factor, which is in general inhomogeneous. Equation (42) reduces to a constraint on the acceleration. Local momentum conservation already provides the constraint in Eq. (39). It follows that in the purely Coulomb-like case, the spatial gradient of the nonlocal energy density must be proportional to that of the local energy density:
$`\text{D}_\mu 𝒰=\left[{\displaystyle \frac{8c_\mathrm{s}^2𝒰\kappa ^4(\rho +p)^2}{2(\rho +p)}}\right]\text{D}_\mu \rho .`$
Linearization about a Friedmann background does not change Eqs. (37) and (38), but Eqs. (40) and (42) lead to
$`\dot{𝒰}+\frac{4}{3}\mathrm{\Theta }𝒰+\text{D}^\mu 𝒬_\mu =0,`$ (44)
$`\dot{𝒬}_\mu +4H𝒬_\mu +\frac{1}{3}\text{D}_\mu 𝒰+\frac{4}{3}𝒰A_\mu +\text{D}^\nu 𝒫_{\mu \nu }=\frac{1}{6}\kappa ^4(\rho +p)\text{D}_\mu \rho ,`$ (45)
where $`H`$ is the Hubble rate in the background. The nonlocal tensor mode, which satisfies $`\text{D}^\nu 𝒫_{\mu \nu }=0𝒫_{\mu \nu }`$, does not enter the nonlocal conservation equations.
## IV Propagation and constraint equations
Equations (27)–(36) are propagation equations for the local and nonlocal energy density and flux. The remaining covariant equations on the brane are the propagation and constraint equations for the kinematic quantities and the free gravitational field on the brane. The kinematic quantities govern the relative motion of neighboring fundamental world-lines, and describe the universal expansion and its local anisotropies. The locally free gravitational field on the brane is given by the brane Weyl tensor $`C_{\mu \nu \alpha \beta }`$. This splits irreducibly for a given $`u^\mu `$ into the gravito-electric and gravito-magnetic fields on the brane:
$`E_{\mu \nu }=C_{\mu \alpha \nu \beta }u^\alpha u^\beta =E_{\mu \nu },H_{\mu \nu }=\frac{1}{2}\epsilon _{\mu \alpha \beta }C^{\alpha \beta }{}_{\nu \gamma }{}^{}u_{}^{\gamma }=H_{\mu \nu },`$where $`E_{\mu \nu }`$ must not be confused with $`_{\mu \nu }`$. The Ricci identity for $`u^\mu `$ and the Bianchi identities $`^\beta C_{\mu \nu \alpha \beta }=_{[\mu }(R_{\nu ]\alpha }+\frac{1}{6}Rg_{\nu ]\alpha })`$ produce the fundamental evolution and constraint equations governing the above covariant quantities . Einstein’s equations are incorporated via the algebraic replacement of the Ricci tensor $`R_{\mu \nu }`$ by the effective total energy-momentum tensor, according to Eq. (19). These are derived directly from the standard general relativity versions (see, e.g., ) by simply replacing the energy-momentum tensor terms $`\rho ,\mathrm{}`$ by $`\rho ^{\mathrm{tot}},\mathrm{}`$. The result for a general imperfect fluid is given in Appendix A.
### A Nonlinear equations
For a perfect fluid or minimally-coupled scalar field, the equations in Appendix A reduce to the following.
Expansion propagation (generalized Raychaudhuri equation):
$`\dot{\mathrm{\Theta }}+\frac{1}{3}\mathrm{\Theta }^2+\sigma _{\mu \nu }\sigma ^{\mu \nu }2\omega _\mu \omega ^\mu \mathrm{D}^\mu A_\mu +A_\mu A^\mu +\frac{1}{2}\kappa ^2(\rho +3p)\mathrm{\Lambda }`$ (46)
$`={\displaystyle \frac{1}{12}}\left({\displaystyle \frac{\stackrel{~}{\kappa }}{\kappa }}\right)^4\left[\kappa ^4\rho (2\rho +3p)+12𝒰\right].`$ (47)
Vorticity propagation:
$$\dot{\omega }_\mu +\frac{2}{3}\mathrm{\Theta }\omega _\mu +\frac{1}{2}\text{curl}A_\mu \sigma _{\mu \nu }\omega ^\nu =0.$$
(48)
Shear propagation:
$$\dot{\sigma }_{\mu \nu }+\frac{2}{3}\mathrm{\Theta }\sigma _{\mu \nu }+E_{\mu \nu }\text{D}_\mu A_\nu +\sigma _{\alpha \mu }\sigma _\nu {}_{}{}^{\alpha }+\omega _\mu \omega _\nu A_\mu A_\nu =\frac{1}{2}\left(\frac{\stackrel{~}{\kappa }}{\kappa }\right)^4𝒫_{\mu \nu }.$$
(49)
Gravito-electric propagation:
$`\dot{E}_{\mu \nu }+\mathrm{\Theta }E_{\mu \nu }\text{curl}H_{\mu \nu }+\frac{1}{2}\kappa ^2(\rho +p)\sigma _{\mu \nu }`$ (50)
$`2A^\alpha \epsilon _{\alpha \beta (\mu }H_{\nu )}{}_{}{}^{\beta }3\sigma _{\alpha \mu }E_\nu {}_{}{}^{\alpha }+\omega ^\alpha \epsilon _{\alpha \beta (\mu }E_{\nu )}^\beta `$ (51)
$`={\displaystyle \frac{1}{12}}\left({\displaystyle \frac{\stackrel{~}{\kappa }}{\kappa }}\right)^4[\{\kappa ^4\rho (\rho +p)+8𝒰\}\sigma _{\mu \nu }+6\dot{𝒫}_{\mu \nu }+2\mathrm{\Theta }𝒫_{\mu \nu }+6\text{D}_\mu 𝒬_\nu `$ (52)
$`+12A_\mu 𝒬_\nu +6\sigma ^\alpha {}_{\mu }{}^{}𝒫_{\nu \alpha }^{}+6\omega ^\alpha \epsilon _{\alpha \beta (\mu }𝒫_{\nu )}{}_{}{}^{\beta }].`$ (53)
Gravito-magnetic propagation:
$`\dot{H}_{\mu \nu }+\mathrm{\Theta }H_{\mu \nu }+\text{curl}E_{\mu \nu }3\sigma _{\alpha \mu }H_\nu {}_{}{}^{\alpha }+\omega ^\alpha \epsilon _{\alpha \beta (\mu }H_{\nu )}{}_{}{}^{\beta }+2A^\alpha \epsilon _{\alpha \beta (\mu }E_{\nu )}^\beta `$ (54)
$`={\displaystyle \frac{1}{2}}\left({\displaystyle \frac{\stackrel{~}{\kappa }}{\kappa }}\right)^4\left[\text{curl}𝒫_{\mu \nu }3\omega _\mu 𝒬_\nu +\sigma ^\alpha {}_{(\mu }{}^{}\epsilon _{\nu )\alpha \beta }^{}𝒬^\beta \right].`$ (55)
Vorticity constraint:
$$\text{D}^\mu \omega _\mu A^\mu \omega _\mu =0.$$
(56)
Shear constraint:
$$\text{D}^\nu \sigma _{\mu \nu }\text{curl}\omega _\mu \frac{2}{3}\text{D}_\mu \mathrm{\Theta }+2[\omega ,A]_\mu =\left(\frac{\stackrel{~}{\kappa }}{\kappa }\right)^4𝒬_\mu .$$
(57)
Gravito-magnetic constraint:
$$\text{curl}\sigma _{\mu \nu }+\text{D}_\mu \omega _\nu H_{\mu \nu }+2A_\mu \omega _\nu =0.$$
(58)
Gravito-electric divergence:
$`\text{D}^\nu E_{\mu \nu }\frac{1}{3}\kappa ^2\text{D}_\mu \rho [\sigma ,H]_\mu +3H_{\mu \nu }\omega ^\nu `$ (59)
$`={\displaystyle \frac{1}{18}}\left({\displaystyle \frac{\stackrel{~}{\kappa }}{\kappa }}\right)^4\left\{\kappa ^4\rho \text{D}_\mu \rho +6\text{D}_\mu 𝒰6\mathrm{\Theta }𝒬_\mu 9\text{D}^\nu 𝒫_{\mu \nu }+9\sigma _{\mu \nu }𝒬^\nu 27[\omega ,𝒬]_\mu \right\}.`$ (60)
Gravito-magnetic divergence:
$`\text{D}^\nu H_{\mu \nu }\kappa ^2(\rho +p)\omega _\mu +[\sigma ,E]_\mu 3E_{\mu \nu }\omega ^\nu `$ (61)
$`={\displaystyle \frac{1}{6}}\left({\displaystyle \frac{\stackrel{~}{\kappa }}{\kappa }}\right)^4\left\{\kappa ^4\rho (\rho +p)\omega _\mu 3\text{curl}𝒬_\mu +8𝒰\omega _\mu 3[\sigma ,𝒫]_\mu 3𝒫_{\mu \nu }\omega ^\nu \right\}.`$ (62)
Here the covariant tensor commutator is
$`[W,Z]_\mu =\epsilon _{\mu \alpha \beta }W^\alpha {}_{\gamma }{}^{}Z_{}^{\beta \gamma }.`$The standard 4-dimensional general relativity results are regained by setting all right hand sides to zero in Eqs. (47)–(62).
Together with Eqs. (37)–(42), these equations govern the dynamics of the matter and gravitational fields on the brane, incorporating both the local (quadratic energy-momentum) and nonlocal (projected Weyl) effects from the bulk. These effects give rise to important new driving and source terms in the propagation and constraint equations. The vorticity propagation and constraint, and the gravito-magnetic constraint have no explicit bulk effects, but all other equations do. Local and nonlocal energy density are driving terms in the expansion propagation, and note that these are the terms that determine the gravitational and non-gravitational acceleration transverse to the brane. The spatial gradients of local and nonlocal energy density provide sources for the gravito-electric field. The nonlocal anisotropic stress is a driving term in the propagation of shear and the gravito-electric/ -magnetic fields, and the nonlocal energy flux is a source for shear and the gravito-magnetic field.
In general the system of equations is not closed: there is no evolution equation for the nonlocal anisotropic stress $`𝒫_{\mu \nu }`$, which carries the tensor modes in perturbed solutions. If we set $`_{\mu \nu }=0`$ to close the system, i.e., if we allow only local matter effects from the fifth dimension, then, as noted above, the density is forced to be homogeneous. This is clearly too restrictive. A less restrictive way of closing the system is to assume that the nonlocal anisotropic stress vanishes, i.e., $`𝒫_{\mu \nu }=0`$. However, this rules out tensor modes arising from the free field in the bulk, and also limits the scalar and vector modes, which will in general also be present in $`𝒫_{\mu \nu }`$.
### B Linearized equations
We have derived the exact nonlinear equations that govern gravitational dynamics on the brane as seen by brane observers. These equations hold for any geometry of the brane, and they are fully covariant on the brane. In particular, this means that we can linearize the equations by taking a suitable limit, and without starting from a given background solution. In this way we avoid the need for choosing coordinates, and we deal directly with covariant physical and geometric quantities, rather than metric components.
The limiting case of the background Friedmann brane is characterized by the vanishing of all inhomogeneous and anisotropic quantities. These quantities are then first-order of smallness in the linearization scheme, and since they vanish in the background, they are gauge-invariant . The standard general relativity scheme is modified by the additional degrees of freedom arising from bulk effects. In particular, the generalized Friedmann equation on the brane is
$$H^2=\frac{1}{3}\mathrm{\Lambda }+\frac{1}{3}\kappa ^2\rho \frac{K}{a^2}+\frac{1}{36}\stackrel{~}{\kappa }^4\rho ^2+\frac{1}{3}\left(\frac{\stackrel{~}{\kappa }}{\kappa }\right)^4𝒰_o\left(\frac{a_o}{a}\right)^4,$$
(63)
where $`K=0,\pm 1`$.
Linearization about a Friedmann brane model of the propagation and constraint equations leads to the reduced system:
$`\dot{\mathrm{\Theta }}+\frac{1}{3}\mathrm{\Theta }^2\mathrm{D}^\mu A_\mu +\frac{1}{2}\kappa ^2(\rho +3p)\mathrm{\Lambda }={\displaystyle \frac{1}{12}}\left({\displaystyle \frac{\stackrel{~}{\kappa }}{\kappa }}\right)^4\left[\kappa ^4\rho (2\rho +3p)+12𝒰\right],`$ (64)
$`\dot{\omega }_\mu +2H\omega _\mu +\frac{1}{2}\text{curl}A_\mu =0,`$ (65)
$`\dot{\sigma }_{\mu \nu }+2H\sigma _{\mu \nu }+E_{\mu \nu }\text{D}_\mu A_\nu ={\displaystyle \frac{1}{2}}\left({\displaystyle \frac{\stackrel{~}{\kappa }}{\kappa }}\right)^4𝒫_{\mu \nu },`$ (66)
$`\dot{E}_{\mu \nu }+3HE_{\mu \nu }\text{curl}H_{\mu \nu }+\frac{1}{2}\kappa ^2(\rho +p)\sigma _{\mu \nu }`$ (67)
$`={\displaystyle \frac{1}{12}}\left({\displaystyle \frac{\stackrel{~}{\kappa }}{\kappa }}\right)^4\left[\left\{\kappa ^4\rho (\rho +p)+8𝒰\right\}\sigma _{\mu \nu }+6\text{D}_\mu 𝒬_\nu +6\dot{𝒫}_{\mu \nu }+6H𝒫_{\mu \nu }\right],`$ (68)
$`\dot{H}_{\mu \nu }+3HH_{\mu \nu }+\text{curl}E_{\mu \nu }={\displaystyle \frac{1}{2}}\left({\displaystyle \frac{\stackrel{~}{\kappa }}{\kappa }}\right)^4\text{curl}𝒫_{\mu \nu },`$ (69)
$`\text{D}^\mu \omega _\mu =0,`$ (70)
$`\text{D}^\nu \sigma _{\mu \nu }\text{curl}\omega _\mu \frac{2}{3}\text{D}_\mu \mathrm{\Theta }=\left({\displaystyle \frac{\stackrel{~}{\kappa }}{\kappa }}\right)^4𝒬_\mu ,`$ (71)
$`\text{curl}\sigma _{\mu \nu }+\text{D}_\mu \omega _\nu H_{\mu \nu }=0,`$ (72)
$`\text{D}^\nu E_{\mu \nu }\frac{1}{3}\kappa ^2\text{D}_\mu \rho ={\displaystyle \frac{1}{18}}\left({\displaystyle \frac{\stackrel{~}{\kappa }}{\kappa }}\right)^4\left[\kappa ^4\rho \text{D}_\mu \rho +6\text{D}_\mu 𝒰18H𝒬_\mu 9\text{D}^\nu 𝒫_{\mu \nu }\right],`$ (73)
$`\text{D}^\nu H_{\mu \nu }\kappa ^2(\rho +p)\omega _\mu ={\displaystyle \frac{1}{6}}\left({\displaystyle \frac{\stackrel{~}{\kappa }}{\kappa }}\right)^4\left[\left\{\kappa ^4\rho (\rho +p)+8𝒰\right\}\omega _\mu 3\text{curl}𝒬_\mu \right],`$ (74)
These equations, together with the linearized conservation equations (37), (38), (44) and (45), are the basis for a covariant and gauge-invariant description of perturbations on the brane. The local bulk effects (quadratic energy-momentum effects) are purely scalar, as is the nonlocal energy density. The nonlocal energy flux has in general scalar and vector modes:
$$𝒬_\mu =\text{D}_\mu 𝒬+\overline{𝒬}_\mu ,$$
(75)
and scalar, vector and tensor modes enter the nonlocal anisotropic stress:
$$𝒫_{\mu \nu }=\text{D}_\mu \text{D}_\nu 𝒫+\text{D}_\mu \overline{𝒫}_\nu +\overline{𝒫}_{\mu \nu }.$$
(76)
In these equations, an overbar denotes a transverse (divergence-free) quantity:
$`\text{D}^\mu \overline{𝒬}_\mu =0=\text{D}^\mu \overline{𝒫}_\mu ,\text{D}^\nu \overline{𝒫}_{\mu \nu }=0.`$
## V Nonlinear and perturbative dynamics
Bulk effects introduce new degrees of freedom into the dynamics on the brane, subject to the additional nonlocal ‘conservation’ equations (40) and (42). Standard results in general relativity may or may not continue to hold under this higher-dimensional modification. We now use the conservation, propagation and constraint equations to generalize some standard results of 4-dimensional general relativity.
### A CMB isotropy and brane homogeneity
In standard general relativity, the isotropy of the CMB radiation has crucial implications for the homogeneity of the universe. If all fundamental observers after last scattering observe an isotropic CMB, then it follows from a theorem of Ehlers, Geren and Sachs that the universe must have a homogeneous Friedmann geometry. This has been generalized to the almost-isotropic case , providing a foundation for the perturbative analysis of CMB anisotropies. The Ehlers-Geren-Sachs theorem is based on the collisionless Boltzmann equation, and on the kinematic/dynamic characterization:
$$A_\mu =\omega _\mu =\sigma _{\mu \nu }=0\mathrm{\Theta }\text{ and }q_\mu =\pi _{\mu \nu }=0\text{ Friedmann geometry.}$$
(77)
Bulk effects do not change the Boltzmann equation, but they do mean that the Friedmann characterization is no longer in general true on the brane. While the gravito-magnetic constraint Eq. (58) still leads to $`H_{\mu \nu }=0`$, the shear propagation equation (49) no longer forces $`E_{\mu \nu }=0`$, because of the nonlocal term $`𝒫_{\mu \nu }`$, so that the intrinsic metric need not be conformally flat.
A consistent solution on the brane of the system of (nonlinear) conservation, propagation and constraint equations can be given as follows. We take
$`\text{D}_\mu \rho =\text{D}_\mu \rho _\mathrm{r}=\text{D}_\mu 𝒰=\text{D}_\mu \mathrm{\Theta }=0,𝒬_\mu =0=\text{D}^\nu 𝒫_{\mu \nu }.`$Then the system of equations reduces to the consistent set
$`\dot{\rho }+\mathrm{\Theta }\rho `$ $`=`$ $`0,`$
$`\dot{\rho }_\mathrm{r}+\frac{4}{3}\mathrm{\Theta }\rho _\mathrm{r}`$ $`=`$ $`0,`$
$`\dot{𝒰}+\frac{4}{3}\mathrm{\Theta }𝒰`$ $`=`$ $`0,`$
$`\dot{\mathrm{\Theta }}+\frac{1}{3}\mathrm{\Theta }^2+\frac{1}{2}\kappa ^2(\rho +2\rho _\mathrm{r})\mathrm{\Lambda }`$ $`=`$ $`{\displaystyle \frac{1}{12}}\left({\displaystyle \frac{\stackrel{~}{\kappa }}{\kappa }}\right)^4\left[\kappa ^4(\rho +\rho _\mathrm{r})(2\rho +3\rho _\mathrm{r})+12𝒰\right],`$
$`\dot{𝒫}_{\mu \nu }+\frac{2}{3}\mathrm{\Theta }𝒫_{\mu \nu }`$ $`=`$ $`0,`$
$`E_{\mu \nu }`$ $`=`$ $`\frac{1}{2}\left({\displaystyle \frac{\stackrel{~}{\kappa }}{\kappa }}\right)^4𝒫_{\mu \nu }.`$
In general, provided that the propagation equation for $`𝒫_{\mu \nu }`$ is consistent with the bulk geometry, $`E_{\mu \nu }`$ need not be zero, so that the brane geometry need not be Friedmann, although it is asymptotically Friedmann.
Thus it is in principle possible via nonlocal bulk effects that isotropic CMB does not force the brane metric to be Friedmann.
### B Generalized gravitational collapse
Another important question is how the higher-dimensional bulk effects modify the picture of gravitational collapse/ singularities, which depends on Raychaudhuri’s equation .
The generalized Raychaudhuri equation (47) governs gravitational collapse and initial singularity behavior on the brane. The local energy density and pressure corrections,
$`\frac{1}{12}\stackrel{~}{\kappa }^4\rho (2\rho +3p),`$further enhance the tendency to collapse, if $`2\rho +3p>0`$. This condition will be satisfied in thermal collapse (or time-reversed expansion), but it is violated during very high-energy inflation ($`\rho \lambda `$), by Eq. (18), and in that case the local bulk term acts to further accelerate expansion. This is consistent with the results given in . Thus local bulk effects at high energy reinforce the formation of singularities during thermal collapse, as predicted in general relativity, while further accelerating expansion during high-energy inflation.
The nonlocal term,
$`\left({\displaystyle \frac{\stackrel{~}{\kappa }}{\kappa }}\right)^4𝒰,`$can act either way, depending on its sign. As shown from Eq. (15), a negative $`𝒰`$ enhances the localization of the gravitational field on the brane. In this case, the effect of $`𝒰`$ is to counteract gravitational collapse. A positive $`𝒰`$ acts against localization, and also reinforces the tendency to collapse.
If higher-dimensional corrections to Einstein’s theory tend to prevent singularities, then the effective energy density $`𝒰`$ on the brane of the free gravitational field in the bulk should be negative. In this case, $`𝒰`$ also acts to reinforce confinement of the gravitational field to the brane.
### C Cosmological scalar perturbations
The linearized equations on the brane derived in the previous two sections encompass scalar, vector and tensor modes. In order to covariantly (and locally) separate out the scalar modes, we impose the condition that all perturbative quantities are spatial gradients of scalars, i.e.,
$`V_\mu =\text{D}_\mu V,W_{\mu \nu }=\text{D}_\mu \text{D}_\nu W.`$The identities in Appendix B, the vorticity constraint equation (70) and the gravito-magnetic constraint equation (72) then show that
$$\text{curl}V_\mu =0=\text{curl}W_{\mu \nu },\text{D}^\nu W_{\mu \nu }=\frac{2}{3}\text{D}^2(\text{D}_\mu W),\omega _\mu =0=H_{\mu \nu },$$
(78)
as in standard general relativity.
If we choose the fundamental 4-velocity $`u^\mu `$ such that $`\sigma _{\mu \nu }=0`$, which is the covariant analogue of the longitudinal or conformal Newtonian gauge in metric-based perturbation theory (see for further discussion), then the shear propagation equation (66) becomes a constraint determining the brane tidal tensor:
$`E_{\mu \nu }=\text{D}_\mu \text{D}_\nu \left[\mathrm{\Phi }+{\displaystyle \frac{1}{2}}\left({\displaystyle \frac{\stackrel{~}{\kappa }}{\kappa }}\right)^4𝒫\right].`$Here $`\mathrm{\Phi }`$ is the relativistic generalization of the gravitational potential, defined by $`A_\mu =\text{D}_\mu \mathrm{\Phi }`$, and $`𝒫`$ is the potential for the nonlocal anisotropic stress, defined in Eq. (76). It follows that nonlocal bulk effects lead to a change in the gravitational tidal potential (in longitudinal-like gauge):
$$\mathrm{\Phi }\mathrm{\Phi }+\frac{1}{2}\left(\frac{\stackrel{~}{\kappa }}{\kappa }\right)^4𝒫.$$
(79)
In the general case, i.e., when $`u^\mu `$ is not chosen to give $`\sigma _{\mu \nu }=0`$, this simple relation does not hold.
In order to derive the equations governing density perturbations in the general case, we define the density and expansion gradients (as in )
$$\mathrm{\Delta }_\mu =\frac{a}{\rho }\text{D}_\mu \rho ,Z_\mu =a\text{D}_\mu \mathrm{\Theta },$$
(80)
and the (dimensionless) gradients describing inhomogeneity in the nonlocal quantities:
$$U_\mu =\frac{a}{\rho }\text{D}_\mu 𝒰,Q_\mu =\frac{1}{\rho }\text{D}_\mu 𝒬,P_\mu =\frac{1}{a\rho }\text{D}_\mu 𝒫,$$
(81)
where $`𝒬`$ is defined in Eq. (75). Then we take the spatial gradient of the energy conservation equations (37) and (44) and the generalized Raychaudhuri equation (64), using the identities in Appendix B and the adiabatic equation (39). We arrive at the following system of equations:
$`\dot{\mathrm{\Delta }}_\mu `$ $`=`$ $`3wH\mathrm{\Delta }_\mu (1+w)Z_\mu ,`$ (82)
$`\dot{Z}_\mu `$ $`=`$ $`2HZ_\mu \left({\displaystyle \frac{c_\mathrm{s}^2}{1+w}}\right)\text{D}^2\mathrm{\Delta }_\mu \left({\displaystyle \frac{\stackrel{~}{\kappa }}{\kappa }}\right)^4\rho U_\mu `$ (84)
$`\frac{1}{2}\rho \left[\kappa ^2+\frac{1}{6}\stackrel{~}{\kappa }^4(4+3w)\rho \left({\displaystyle \frac{\stackrel{~}{\kappa }}{\kappa }}\right)^4\left({\displaystyle \frac{2c_\mathrm{s}^2}{1+w}}\right){\displaystyle \frac{𝒰}{\rho }}\right]\mathrm{\Delta }_\mu ,`$
$`\dot{U}_\mu `$ $`=`$ $`(3w1)HU_\mu \left({\displaystyle \frac{4c_\mathrm{s}^2}{1+w}}\right)\left({\displaystyle \frac{𝒰}{\rho }}\right)H\mathrm{\Delta }_\mu \left({\displaystyle \frac{4𝒰}{3\rho }}\right)Z_\mu a\text{D}^2Q_\mu ,`$ (85)
$`\dot{Q}_\mu `$ $`=`$ $`(13w)HQ_\mu {\displaystyle \frac{1}{3a}}U_\mu \frac{2}{3}a\text{D}^2P_\mu +{\displaystyle \frac{1}{6a}}\left[\left({\displaystyle \frac{8c_\mathrm{s}^2}{1+w}}\right){\displaystyle \frac{𝒰}{\rho }}\kappa ^4\rho ^2(1+w)\right]\mathrm{\Delta }_\mu ,`$ (86)
where $`w=p/\rho `$. The background Friedmann equation (63) relates $`H`$ and $`\rho `$, with $`𝒰`$ in the background given by Eq. (43).
In standard general relativity, only the first two equations apply, with $`\stackrel{~}{\kappa }`$ set to zero in Eq. (84). In this case we can decouple the density perturbations via a second-order equation for $`\mathrm{\Delta }_\mu `$, whose independent solutions are adiabatic growing and decaying modes. Nonlocal bulk effects introduce important changes to this simple picture. First, we note that there is no equation for $`\dot{P}_\mu `$, so that in general, scalar perturbations on the brane cannot be predicted by brane observers without additional information from the unobservable bulk.
However, there is a very important exception to this, arising from the fact that $`Q_\mu `$ and $`P_\mu `$ only occur in Eqs. (82)–(85) via the Laplacian terms $`\text{D}^2Q_\mu `$ and $`\text{D}^2P_\mu `$, and the latter term is the only occurrence of $`P_\mu `$ in the system. It follows that on super-Hubble scales, the system does close, and brane observers can predict scalar perturbations from initial conditions intrinsic to the brane. The system reduces to 3 coupled equations in $`\mathrm{\Delta }_\mu `$, $`Z_\mu `$ and $`U_\mu `$, plus an equation for $`Q_\mu `$, which is determined once the other 3 quantities are solved for. One can decouple the density perturbations via a third-order equation for $`\mathrm{\Delta }_\mu `$. The nonlocal energy density plays the role of a non-interacting radiation fluid with the same velocity as the ordinary fluid, and inhomogeneity in $`𝒰`$ introduces an additional entropy-like scalar mode. As may have been expected, this additional mode is absent during radiation-domination; in this case Eqs. (82) and (85) show that
$`\dot{U}_\mu ={\displaystyle \frac{𝒰_o}{\rho _o}}\dot{\mathrm{\Delta }}_\mu .`$
In principle, it is straightforward to solve the coupled system on super-Hubble scales, although numerical techniques will be necessary. We do not attempt particular solutions here. However, it may be instructive to see the decoupled third-order equation for density perturbations during matter-domination, on a flat background:
$$\ddot{\mathrm{\Delta }}_\mu ^{^{}}+2H\ddot{\mathrm{\Delta }}_\mu +\left[\frac{4}{3}\left(\frac{\stackrel{~}{\kappa }}{\kappa }\right)^4𝒰\frac{7}{6}\kappa ^2\rho \frac{2}{9}\stackrel{~}{\kappa }^4\rho ^2\frac{8}{3}\mathrm{\Lambda }\right]\dot{\mathrm{\Delta }}_\mu +\left[\frac{1}{2}\kappa ^2\frac{2}{3}\stackrel{~}{\kappa }^4\rho \right]\rho H\mathrm{\Delta }_\mu =0.$$
(87)
### D Cosmological tensor perturbations
Tensor perturbations are covariantly characterized by
$`\text{D}_\mu f=0,A_\mu =\omega _\mu =𝒬_\mu =0,\text{D}^\nu W_{\mu \nu }=0,`$where $`f=\rho ,p,\mathrm{\Theta },𝒰`$, and $`W_{\mu \nu }=\sigma _{\mu \nu },E_{\mu \nu },H_{\mu \nu },𝒫_{\mu \nu }`$. Then all the conservation equations reduce to background equations, and the system of linearized propagation and constraint equations in the previous section reduces to:
$`\dot{\overline{\sigma }}_{\mu \nu }+2H\overline{\sigma }_{\mu \nu }+E_{\mu \nu }^{}=0,`$ (88)
$`\dot{E}_{\mu \nu }^{}+3HE_{\mu \nu }^{}\text{curl}\overline{H}_{\mu \nu }+\frac{1}{2}\kappa ^2(\rho +p)\overline{\sigma }_{\mu \nu }`$ (89)
$`={\displaystyle \frac{1}{12}}\left({\displaystyle \frac{\stackrel{~}{\kappa }}{\kappa }}\right)^4\left[\left\{\kappa ^4\rho (\rho +p)+8𝒰\right\}\overline{\sigma }_{\mu \nu }+12\left(\dot{\overline{𝒫}}_{\mu \nu }+2H\overline{𝒫}_{\mu \nu }\right)\right],`$ (90)
$`\dot{\overline{H}}_{\mu \nu }+3H\overline{H}_{\mu \nu }+\text{curl}E_{\mu \nu }^{}=0,`$ (91)
$`\text{curl}\overline{\sigma }_{\mu \nu }\overline{H}_{\mu \nu }=0,`$ (92)
where the overbars denote transverse tensors, and
$`E_{\mu \nu }^{}=\overline{E}_{\mu \nu }{\displaystyle \frac{1}{2}}\left({\displaystyle \frac{\stackrel{~}{\kappa }}{\kappa }}\right)^4\overline{𝒫}_{\mu \nu }.`$Since there is no equation for $`\dot{\overline{𝒫}}_{\mu \nu }`$, the system of equations does not close on the brane: brane observers cannot evaluate tensor perturbations on the brane without additional information from the unobservable bulk. This remains true on super-Hubble scales, unlike the scalar perturbation case.
Equations (88) and (92) show that the shear is a gravito-potential for $`E_{\mu \nu }^{}`$ and $`\overline{H}_{\mu \nu }`$. Using the identities in Appendix B, we can derive the following covariant wave equation for the shear:
$`\text{D}^2\overline{\sigma }_{\mu \nu }\ddot{\overline{\sigma }}_{\mu \nu }5H\dot{\overline{\sigma }}_{\mu \nu }\left[2\mathrm{\Lambda }+\frac{1}{2}\kappa ^2(\rho 3p)\frac{1}{12}\stackrel{~}{\kappa }^4\rho (\rho +3p)\right]\overline{\sigma }_{\mu \nu }`$ (93)
$`=\left({\displaystyle \frac{\stackrel{~}{\kappa }}{\kappa }}\right)^4\left[\dot{\overline{𝒫}}_{\mu \nu }+2H\overline{𝒫}_{\mu \nu }\right].`$ (94)
For adiabatic tensor perturbations in standard general relativity, the right hand side falls away. Nonlocal bulk effects provide driving terms that are like anisotropic stress terms in general relativity . In the latter case however, the evolution of anisotropic stress is determined by the Boltzmann equation or other intrinsic physics.
What we can conclude from Eq. (94) is that the local bulk effects enhance tensor perturbations for non-inflationary expansion, and suppress them during inflation, since inflation implies $`\rho +3p<0`$ (at all energy scales) by Eq. (18). The nature of the nonlocal bulk effects carried by $`\overline{𝒫}_{\mu \nu }`$ requires knowledge of the off-brane dynamics of $`_{AB}`$, which we have not considered.
### E Cosmological vector perturbations
The linearized vorticity propagation equation (65) does not carry any bulk effects, and vorticity decays as in standard general relativity. However, the gravito-magnetic divergence equation (74) shows that the local and nonlocal energy density and the nonlocal energy flux provide additional sources for the brane gravito-magnetic field:
$`\text{D}^\nu H_{\mu \nu }=\kappa ^2(\rho +p)\omega _\mu +{\displaystyle \frac{1}{6}}\left({\displaystyle \frac{\stackrel{~}{\kappa }}{\kappa }}\right)^4\left[\left\{\kappa ^4\rho (\rho +p)+8𝒰\right\}\omega _\mu 3\text{curl}𝒬_\mu \right].`$In standard general relativity, it is necessary to increase the angular momentum density $`\kappa ^2(\rho +p)\omega _\mu `$ in order to increase the gravito-magnetic field, but bulk effects allow an increased gravito-magnetic field without this. In particular, unlike in general relativity, it is possible to source vector perturbations even when the vorticity vanishes, since $`\text{curl}𝒬_\mu `$ may be nonzero.
## VI Conclusion
By adopting a covariant approach based on physical and geometrical quantities that are in principle measurable by brane-observers, we have given a comprehensive analysis of intrinsic cosmological dynamics in Randall-Sundrum-type brane-worlds. This has allowed us to carefully delineate what can and can not be predicted by brane observers without additional information from the unobservable bulk.
Our main result is probably that scalar perturbations on super-Hubble scales can be evaluated intrinsically on the brane. This is a generalization of the result given in , i.e., the evaluation of adiabatic scalar perturbations on super-Hubble scales during inflation on the brane, done under the assumption that $`_{\mu \nu }`$ may be neglected.
We showed that tensor perturbations can not be evaluated intrinsically on any scales. This is not surprising, since the ability of gravitational waves to penetrate the 5th dimension inevitably introduces unpredictability from the standpoint of observers confined to the brane. Vector perturbations on the brane can be generated even in the absence of vorticity, via the curl of the nonlocal energy flux.
Our nonperturbative results include:
(i) Calculating the gravitational and non-gravitational acceleration felt by brane observers, allowing us to provide covariant characterizations of gravity and matter localization, and showing in particular that the non-gravitational off-brane acceleration of fundamental world-lines is towards the brane during inflation.
(ii) Showing how bulk effects can disrupt the relation between isotropy of the CMB and spatial isotropy and homogeneity on the brane.
(iii) Showing how bulk effects modify the dynamics of gravitational collapse and singularity formation.
We have derived the exact nonlinear equations governing cosmological dynamics on the brane, and the corresponding covariant and gauge-invariant linearized equations. Further implications of these equations could usefully be pursued. In particular, an important topic for further research is the calculation of scalar perturbations on very large scales, and the investigation of limits imposed by observations.
However, the major further step required, and not undertaken here, is to complete the picture by investigating the dynamical equations of the gravitational field off the brane. A starting point is provided by the general equations given in , which determine
$`_𝐧_{AB},_𝐧_{ABC},_𝐧R_{ABCD},`$where $`R_{ABCD}`$ is the 4-dimensional Riemann tensor, and
$`_{ABC}=g_A{}_{}{}^{D}g_{B}^{}{}_{}{}^{E}\stackrel{~}{C}_{DECF}^{}n^F.`$However, it may turn out to be more useful to develop an alternative decomposition of the bulk Weyl tensor, along a timelike direction $`\stackrel{~}{u}^A`$ rather than a spatial direction $`n^A`$. Intuitively, this may provide a more direct and transparent route to the evolution equation for $`𝒫_{\mu \nu }`$, whose absence leads to unpredictability on the brane. Such a timelike decomposition requires a generalization to higher dimensions of the 4-dimensional decomposition of the Weyl tensor . The generalization has been given in .
The complete and closed system of dynamical equations would allow us to develop more systematic and probing tests of the Randall-Sundrum-type models against observational constraints. Despite the appealing geometric and particle-physics properties of such models, it is their confrontation with cosmological observational tests that will provide a more decisive arbiter as to whether they are viable generalizations of Einstein’s theory.
Note: Since this work was completed, a number of papers has appeared, setting up the 5-dimensional formalism (metric-based) of bulk perturbations , and gravitational waves produced during inflation on the brane have also been studied .
Acknowledgements:
I would like to thank Bruce Bassett, Marco Bruni, Chris Clarkson, Naresh Dadhich, George Ellis, Jose Senovilla and David Wands for helpful discussions and comments.
## A General propagation and constraint equations
For a general, imperfect energy-momentum tensor, as in Eq. (6), the propagation and constraint equations (47)–(62) are generalized to:
Propagation:
$`\dot{\mathrm{\Theta }}+\frac{1}{3}\mathrm{\Theta }^2+\sigma _{\mu \nu }\sigma ^{\mu \nu }2\omega _\mu \omega ^\mu \mathrm{D}^\mu A_\mu +A_\mu A^\mu +\frac{1}{2}\kappa ^2(\rho +3p)\mathrm{\Lambda }`$ (A1)
$`={\displaystyle \frac{1}{12}}\left({\displaystyle \frac{\stackrel{~}{\kappa }}{\kappa }}\right)^4\left[\kappa ^4\left(2\rho ^2+3\rho p3q_\mu q^\mu \right)+12𝒰\right],`$ (A2)
(A3)
$`\dot{\omega }_\mu +\frac{2}{3}\mathrm{\Theta }\omega _\mu +\frac{1}{2}\text{curl}A_\mu \sigma _{\mu \nu }\omega ^\nu =0,`$ (A4)
(A5)
$`\dot{\sigma }_{\mu \nu }+\frac{2}{3}\mathrm{\Theta }\sigma _{\mu \nu }+E_{\mu \nu }\frac{1}{2}\kappa ^2\pi _{\mu \nu }\text{D}_\mu A_\nu +\sigma _{\alpha \mu }\sigma _\nu {}_{}{}^{\alpha }+\omega _\mu \omega _\nu A_\mu A_\nu `$ (A6)
$`={\displaystyle \frac{1}{24}}\left({\displaystyle \frac{\stackrel{~}{\kappa }}{\kappa }}\right)^4[\kappa ^4\{(\rho +3p)\pi _{\mu \nu }+\pi _{\alpha \mu }\pi _\nu {}_{}{}^{\alpha }+q_\mu q_\nu \}+12𝒫_{\mu \nu }],`$ (A7)
(A8)
$`\dot{E}_{\mu \nu }+\mathrm{\Theta }E_{\mu \nu }\text{curl}H_{\mu \nu }+\frac{1}{2}\kappa ^2(\rho +p)\sigma _{\mu \nu }+\frac{1}{2}\kappa ^2\dot{\pi }_{\mu \nu }+\frac{1}{2}\kappa ^2\text{D}_\mu q_\nu +\frac{1}{6}\kappa ^2\mathrm{\Theta }\pi _{\mu \nu }+\kappa ^2A_\mu q_\nu `$ (A9)
$`2A^\alpha \epsilon _{\alpha \beta (\mu }H_{\nu )}{}_{}{}^{\beta }3\sigma _{\alpha \mu }E_\nu {}_{}{}^{\alpha }+\omega ^\alpha \epsilon _{\alpha \beta (\mu }E_{\nu )}{}_{}{}^{\beta }+\frac{1}{2}\kappa ^2\sigma ^\alpha {}_{\mu }{}^{}\pi _{\nu \alpha }^{}+\frac{1}{2}\kappa ^2\omega ^\alpha \epsilon _{\alpha \beta (\mu }\pi _{\nu )}^\beta `$ (A10)
$`={\displaystyle \frac{1}{72}}\left({\displaystyle \frac{\stackrel{~}{\kappa }}{\kappa }}\right)^4[\kappa ^4\{3(2\rho ^2+2\rho p\pi _{\alpha \beta }\pi ^{\alpha \beta }2q_\alpha q^\alpha )\sigma _{\mu \nu }+3(\dot{\rho }+\dot{p})\pi _{\mu \nu }+3(\rho +3p)\dot{\pi }_{\mu \nu }`$ (A11)
$`6\pi _{\alpha \mu }\dot{\pi }_\nu {}_{}{}^{\alpha }6q_\mu \dot{q}_\nu +\frac{3}{2}\text{D}_\mu (\pi _{\nu \alpha }q^\alpha 4\rho q_\nu )+\mathrm{\Theta }([\rho +3p]\pi _{\mu \nu }\pi _{\alpha \mu }\pi _\nu {}_{}{}^{\alpha }q_\mu q_\nu )`$ (A12)
$`3A_\mu (4\rho q_\nu \pi _{\nu \alpha }q^\alpha )+3(\rho +3p)\sigma ^\alpha {}_{\mu }{}^{}\pi _{\nu \alpha }^{}3\sigma ^\alpha {}_{\mu }{}^{}(\pi ^\beta {}_{\nu }{}^{}\pi _{\alpha \beta }^{}+q_\nu q_\alpha )`$ (A13)
$`+3(\rho +3p)\omega ^\alpha \epsilon _{\alpha \beta (\mu }\pi _{\nu )}{}_{}{}^{\beta }3\omega _\alpha \epsilon ^{\alpha \beta }{}_{(\mu }{}^{}(\pi ^\gamma {}_{\nu )}{}^{}\pi _{\beta \gamma }^{}+q_{\nu )}q_\beta )\}48𝒰\sigma _{\mu \nu }36\dot{𝒫}_{\mu \nu }`$ (A14)
$`36\text{D}_\mu 𝒬_\nu 12\mathrm{\Theta }𝒫_{\mu \nu }72A_\mu 𝒬_\nu 36\sigma ^\alpha {}_{\mu }{}^{}𝒫_{\nu \alpha }^{}36\omega ^\alpha \epsilon _{\alpha \beta (\mu }𝒫_{\nu )}{}_{}{}^{\beta }],`$ (A15)
(A16)
$`\dot{H}_{\mu \nu }+\mathrm{\Theta }H_{\mu \nu }+\text{curl}E_{\mu \nu }\frac{1}{2}\kappa ^2\text{curl}\pi _{\mu \nu }3\sigma _{\alpha \mu }H_\nu {}_{}{}^{\alpha }+\omega ^\alpha \epsilon _{\alpha \beta (\mu }H_{\nu )}^\beta `$ (A17)
$`+2A^\alpha \epsilon _{\alpha \beta (\mu }E_{\nu )}{}_{}{}^{\beta }+\frac{3}{2}\kappa ^2\omega _\mu q_\nu \frac{1}{2}\kappa ^2\sigma ^\alpha {}_{(\mu }{}^{}\epsilon _{\nu )\alpha \beta }^{}q^\beta `$ (A18)
$`={\displaystyle \frac{1}{48}}\left({\displaystyle \frac{\stackrel{~}{\kappa }}{\kappa }}\right)^4[\kappa ^4\{2\text{curl}((\rho +3p)\pi _{\mu \nu }\pi _{\alpha \mu }\pi _\nu {}_{}{}^{\alpha }q_\mu q_\nu )12\rho \omega _\mu q_\nu +4\omega _\mu \pi _{\nu \alpha }q^\alpha `$ (A19)
$`+\sigma ^\alpha {}_{(\mu }{}^{}\epsilon _{\nu )\alpha \beta }^{}(4\rho q^\beta \pi ^{\beta \gamma }q_\gamma )\}+24\text{curl}𝒫_{\mu \nu }72\omega _\mu 𝒬_\nu +24\sigma ^\alpha {}_{(\mu }{}^{}\epsilon _{\nu )\alpha \beta }^{}𝒬^\beta ].`$ (A20)
Constraint:
$`\text{D}^\mu \omega _\mu A^\mu \omega _\mu =0,`$ (A21)
(A22)
$`\text{D}^\nu \sigma _{\mu \nu }\text{curl}\omega _\mu \frac{2}{3}\text{D}_\mu \mathrm{\Theta }+\kappa ^2q_\mu +2[\omega ,A]_\mu ={\displaystyle \frac{1}{24}}\left({\displaystyle \frac{\stackrel{~}{\kappa }}{\kappa }}\right)^4\left[\kappa ^4\left(4\rho q_\mu \pi _{\mu \nu }q^\nu \right)+24𝒬_\mu \right],`$ (A23)
(A24)
$`\text{curl}\sigma _{\mu \nu }+\text{D}_\mu \omega _\nu H_{\mu \nu }+2A_\mu \omega _\nu =0,`$ (A25)
(A26)
$`\text{D}^\nu E_{\mu \nu }+\frac{1}{2}\kappa ^2\text{D}^\nu \pi _{\mu \nu }\frac{1}{3}\kappa ^2\text{D}_\mu \rho +\frac{1}{3}\kappa ^2\mathrm{\Theta }q_\mu [\sigma ,H]_\mu +3H_{\mu \nu }\omega ^\nu \frac{1}{2}\kappa ^2\sigma _{\mu \nu }q^\nu +\frac{3}{2}\kappa ^2[\omega ,q]_\mu `$ (A27)
$`={\displaystyle \frac{1}{48}}\left({\displaystyle \frac{\stackrel{~}{\kappa }}{\kappa }}\right)^4[\kappa ^4\{\frac{2}{3}\mathrm{\Theta }(\pi _{\mu \nu }q^\nu 4\rho q_\mu )+2\text{D}^\nu ((\rho +3p)\pi _{\mu \nu }\pi _{\alpha \mu }\pi _\nu {}_{}{}^{\alpha }q_\mu q_\nu )`$ (A28)
$`+\frac{8}{3}\rho \text{D}_\mu \rho 4\pi ^{\alpha \beta }\text{D}_\mu \pi _{\alpha \beta }+\sigma _{\mu \nu }(4\rho q^\nu \pi ^{\nu \alpha }q_\alpha )+3\epsilon _{\mu \alpha \beta }\omega ^\alpha \pi ^{\beta \gamma }q_\gamma 4\rho [\omega ,q]_\mu \}`$ (A29)
$`+16\text{D}_\mu 𝒰16\mathrm{\Theta }𝒬_\mu 24\text{D}^\nu 𝒫_{\mu \nu }+24\sigma _{\mu \nu }𝒬^\nu 72[\omega ,𝒬]_\mu ],`$ (A30)
(A31)
$`\text{D}^\nu H_{\mu \nu }+\frac{1}{2}\kappa ^2\text{curl}q_\mu \kappa ^2(\rho +p)\omega _\mu +[\sigma ,E]_\mu +\frac{1}{2}\kappa ^2[\sigma ,\pi ]_\mu 3E_{\mu \nu }\omega ^\nu +\frac{1}{2}\kappa ^2\pi _{\mu \nu }\omega ^\nu `$ (A32)
$`={\displaystyle \frac{1}{48}}\left({\displaystyle \frac{\stackrel{~}{\kappa }}{\kappa }}\right)^4[\kappa ^4\{\text{curl}(\pi _{\mu \nu }q^\nu 4\rho q_\mu )+4(2\rho ^2+2\rho p\pi _{\alpha \beta }\pi ^{\alpha \beta }2q_\alpha q^\alpha )\omega _\mu `$ (A33)
$`2\epsilon _{\mu \alpha \beta }\sigma ^\alpha {}_{\gamma }{}^{}(\pi _\nu {}_{}{}^{\beta }\pi _{}^{\gamma \nu }+q^\beta q^\gamma )+2((\rho +3p)\pi _{\mu \nu }\pi _{\alpha \mu }\pi _\nu {}_{}{}^{\alpha }q_\mu q_\nu )\omega ^\nu `$ (A34)
$`+2(\rho +3p)[\sigma ,\pi ]_\mu \}24\text{curl}𝒬_\mu +64𝒰\omega _\mu 24[\sigma ,𝒫]_\mu 24𝒫_{\mu \nu }\omega ^\nu ].`$ (A35)
The 4-dimensional general relativistic results are regained by setting all the right hand sides of these equations to 0.
## B Differential identities
On a flat Friedmann background, the following covariant linearized identities hold :
$`\text{D}_\mu \dot{f}`$ $`=`$ $`(\text{D}_\mu f)^{}+H\text{D}_\mu f\dot{f}A_\mu ,`$ (B1)
$`\text{D}^2(\text{D}_\mu f)`$ $`=`$ $`\text{D}_\mu (\text{D}^2f)+2\dot{f}\omega _\mu ,`$ (B2)
$`(\text{D}^2f)^{}`$ $`=`$ $`\text{D}^2\dot{f}2H\text{D}^2f+\dot{f}\text{D}^\mu A_\mu ,`$ (B3)
$`\text{curl}\text{D}_\mu f`$ $`=`$ $`2\dot{f}\omega _\mu ,`$ (B4)
$`\text{curl}\text{D}_\mu \text{D}_\nu f`$ $`=`$ $`0,`$ (B5)
$`(\text{D}_\mu V_\nu )^{}`$ $`=`$ $`\text{D}_\mu \dot{V}_\nu H\text{D}_\mu V_\nu ,`$ (B6)
$`\text{D}_{[\mu }\text{D}_{\nu ]}V_\alpha `$ $`=`$ $`0=\text{D}_{[\mu }\text{D}_{\nu ]}W_{\alpha \beta },`$ (B7)
$`\text{D}^\mu \text{curl}V_\mu `$ $`=`$ $`0,`$ (B8)
$`\text{D}^\nu \text{D}_\mu V_\nu `$ $`=`$ $`\frac{1}{2}\text{D}^2V_\mu +\frac{1}{6}\text{D}_\mu (\text{D}^\nu V_\nu ),`$ (B9)
$`\text{curl}\text{D}_\mu V_\nu `$ $`=`$ $`\frac{1}{2}\text{D}_\mu \text{curl}V_\nu ,`$ (B10)
$`\text{curl}\text{curl}V_\mu `$ $`=`$ $`\text{D}^2V_\mu +\text{D}_\mu (\text{D}^\nu V_\nu ),`$ (B11)
$`(\text{D}_\mu W_{\alpha \beta })^{}`$ $`=`$ $`\text{D}_\mu \dot{W}_{\alpha \beta }H\text{D}_\mu W_{\alpha \beta },`$ (B12)
$`\text{D}^\nu \text{curl}W_{\mu \nu }`$ $`=`$ $`\frac{1}{2}\text{curl}(\text{D}^\nu W_{\mu \nu }),`$ (B13)
$`\text{curl}\text{curl}W_{\mu \nu }`$ $`=`$ $`\text{D}^2W_{\mu \nu }+\frac{3}{2}\text{D}_\mu \text{D}^\alpha W_{\nu \alpha },`$ (B14)
where $`V_\mu =V_\mu `$ and $`W_{\mu \nu }=W_{\mu \nu }`$ vanish in the background.
|
warning/0004/cond-mat0004016.html
|
ar5iv
|
text
|
# Excitation spectrum of three dressed Bose-Einstein condensates
## Abstract
We study quantum dynamics of three dressed Bose-Einstein condensates in a high-Q cavity. The quasiparticle excitation spectrum of this system is found numerically. The stability of the quasiparticle excitation is analyzed. It is shown that there exist instabilities in the excitation spectrum.
\]
Recently, multicomponent Bose-Einstein condensates (MBEC’s) have attracted much attention due to the recent experimental realizations of two and three component Bose-Einstein condensates (BEC’s) . MBEC’s offer new degrees of freedom, which give rise to a rich set of new phenomena that do not exist in a single condensate. Especially, the interaction of BEC’s with light leads to fascinating effects including the extreme slowing down of the speed of light and matter-wave four-wave mixing . Goldstein, Wright and Meystre (GWM) proposed an approach to produce MBEC’s, called dressed BEC’s, inside a high-$`Q`$ multimode cavity, and investigated the quasiparticle excitation spectrum for the case of two dressed BEC’s which dress one-photon. The purpose of this letter is to study the quasiparticle excitation spectrum for a system of three dressed BEC’s which dress two photons. We will investigate quantum dynamics of the three dressed BEC’s, and calculate numerically their elementary excitations.
Consider the GWM model of dressed BEC’s , which consists of a condensate which interacts with two counter-propagating modes by a high-Q ring cavity with $`n_1`$ photons being in the mode 1 and $`n_2`$ photons being in the mode 2. We consider a state composing $`n=n_1+n_2`$ photons and $`N`$ atoms which comprises the condensate confined by an external trapping potential $`U(𝐫)`$. The state of the system can be written $`|\mathrm{\Psi }_{N,n}=_{n_1,n_2}𝑑r_1\mathrm{}𝑑r_Nf_{n_1,n_2}(𝐫_1,\mathrm{},r_N,t)\mathrm{\Psi }^{}(𝐫_1)\mathrm{}\mathrm{\Psi }^{}(𝐫_N)\times |0,n_1,n_2`$, where $`\mathrm{\Psi }^{}(𝐫_i)`$ are the atomic field operators, $`f_{n_1,n_2}(𝐫_1,\mathrm{},r_N,t)`$ is the many-particle Schrödinger wave function for the condensate, which can be factorized as a product of Hartree wave functions $`\varphi _{n_1,n_2}`$ representing states which the atoms occupy, $`f_{n_1,n_2}(𝐫_1,\mathrm{},r_N,t)=_{i=1}^N\varphi _{n_1,n_2}(𝐫_i,t)`$.
Let us consider such a trapping potential $`U(𝐫)=m\omega _{}^2(𝐫_{}^2+\lambda ^2z)^2`$, where $`r=(𝐫_{},z)`$, $`𝐫_{}`$ being the transverse position coordinate, $`\omega _{}`$ the transverse angular frequency of the trap, and $`\lambda =\omega _z/\omega _L`$ is the ratio of the longitudinal to transverse frequencies. For the cigar structure condensate , it is reasonable to assume that $`\lambda 1`$, and the transverse structure of the condensate is determined by the ground state solution of the transverse potential $`\upsilon (𝐫_{})`$. Then the Hartree wave function can be expressed as $`\varphi _{n_1,n_2}(𝐫,t)=\upsilon (𝐫_{})e^{i\omega _{}t}\psi _{n_1,n_2}(z,t)`$. From the Hartree variational pronciple and after integrating over the transverse coordinate $`r_{}`$, one obtains Gross-Pitaeviskii equations (GPE’s)
$`i\dot{\psi }_{n_1,n_2}(\xi ,\tau )`$ $`=`$ $`H_L\psi _{n_1,n_2}(\xi ,\tau )`$ (4)
$`+\eta \psi _{n_1,n_2}(\xi ,\tau )^2\psi _{n_1,n_2}(\xi ,\tau )`$
$`+g[\sqrt{n_1(n_2+1)}\psi _{n_11,n_2+1}(\xi ,\tau )`$
$`+\sqrt{(n_1+1)n_2}\psi _{n_1+1,n_21}(\xi ,\tau )],`$
where $`H_L=^2/\xi ^2+\xi ^2/4`$, with $`\xi =z/a_z`$, $`a_z=\sqrt{\mathrm{}/2m\omega _z}`$, $`\tau =\omega _zt`$, $`g`$ and $`\eta `$ are the linear and nonlinear coupling coefficients, respectively.
The dressed BEC’s are the eigenstates of Eq.(1), and are quantum superposition states of states with different photon numbers $`(n_1,n_2)`$. We consider the case of three dressed BEC’s. In this case, only the states with (2,0), (1,1) and (0,2) are concerned. The coupled GPE’s (1) reduce to
$`i\dot{\psi }_{2,0}(\xi ,\tau )`$ $`=`$ $`H_L\psi _{2,0}(\xi ,\tau )_+\sqrt{2}g\psi _{1,1}(\xi ,\tau )`$ (5)
$`+`$ $`\eta \psi _{2,0}(\xi ,\tau )^2\psi _{2,0}(\xi ,\tau ),`$ (6)
$`i\dot{\psi }_{1,1}(\xi ,\tau )`$ $`=`$ $`H_L\psi _{1,1}(\xi ,\tau )_+\sqrt{2}g[\psi _{2,0}(\xi ,\tau )`$ (7)
$`+`$ $`\psi _{0,2}(\xi ,\tau )]+\eta \psi _{1,1}(\xi ,\tau )^2\psi _{1,1}(\xi ,\tau ),`$ (8)
$`i\dot{\psi }_{0,2}(\xi ,\tau )`$ $`=`$ $`H_L\psi _{0,2}(\xi ,\tau )_+\sqrt{2}g\psi _{1,1}(\xi ,\tau )`$ (9)
$`+`$ $`\eta \psi _{0,2}(\xi ,\tau )^2\psi _{0,2}(\xi ,\tau ).`$ (10)
The exact analytic stationary solutions are generally not available for the coupling GPE’s of the dressed BEC’s. We here consider their solutions under the Thomas-Fermi approximation (TFA), in which the non-linear interaction term dominates over the kinetic-energy term . Setting $`\psi _{n_1,n_2}(\xi ,\tau )=e^{i\mu \tau }\theta _{n_1,n_2}(\xi )`$ with $`\mu `$ being the chemical potential, under the TFA, coupling GPE’s (2) reduce to
$`\mu \theta _{2,0}(\xi )`$ $`=`$ $`{\displaystyle \frac{1}{4}}\xi ^2\theta _{2,0}(\xi )+\sqrt{2}g\theta _{1,1}(\xi )+\eta \theta _{2,0}(\xi )^2\theta _{2,0}(\xi ),`$ (11)
$`\mu \theta _{0,2}(\xi )`$ $`=`$ $`{\displaystyle \frac{1}{4}}\xi ^2\theta _{0,2}(\xi )+\sqrt{2}g\theta _{1,1}(\xi )+\eta \theta _{0,2}(\xi )^2\theta _{0,2}(\xi ),`$ (12)
$`\mu \theta _{1,1}(\xi )`$ $`=`$ $`{\displaystyle \frac{1}{4}}\xi ^2\theta _{1,1}(\xi )+\sqrt{2}g[\theta _{0,2}(\xi )+\theta _{2,0}(\xi )]`$ (14)
$`+\eta \theta _{1,1}(\xi )^2\theta _{1,1}(\xi ).`$
For simplicity, we consider only solutions of two cases for $`\theta _{2,0}(\xi )`$ and $`\theta _{0,2}(\xi )`$: (a) Out-of-phase solution with $`\theta _{2,0}(\xi )=\theta _{0,2}(\xi )`$; (b) In-phase solution with $`\theta _{2,0}(\xi )=\theta _{0,2}(\xi )`$.
(a) Out-of-phase solution. In this case, $`\theta _{2,0}(\xi )=\theta _{0,2}(\xi )`$, from Eqs.(3) we obtain the following dressed state nonzero solutions
$$\theta _{2,0}(\xi )=\sqrt{\frac{4\mu \xi ^2}{4\eta ^2}},\theta _{1,1}(\xi )=0,4\mu \xi ^2.$$
(15)
Let $`\xi _m`$ be the boundary of the dressed condensates at which the TFA solutions vanish. From Eq.(4) it follows that the chemical potential $`\mu =\frac{1}{4}\xi _m^2`$. The value of $`\xi _m`$ is determined by the normalization condition of the condensate wave function $`_{n_1,n_2}_{\xi _m}^{\xi _m}𝑑\xi \theta _{n_1,n_2}(\xi )^2=1`$, which leads to $`\xi _m=[3\eta /2]^{1/3}`$. So we get the following out-of-phase solution
$$\theta _{2,0}(\xi )^2=\theta _{0,2}(\xi )^2=\frac{\xi _m^2\xi ^2}{4\eta },\theta _{1,1}(\xi )^2=0.$$
(16)
It is interesting to note that the out-of-phase solution of the three dressed BEC’s are the same as those of two dressed BEC’s found in Ref. due to the vanishness of $`\theta _{1,1}(\xi )`$. This reflects the fact that the three dressed BEC system can transit to a two dressed BEC system. Physically, this is reasonable, because a three dressed BEC system naturally becomes a two dressed BEC system when one of the three dressed condensates vanishes.
(b) In-phase solution. In the case, $`\theta _{2,0}(\xi )=\theta _{0,2}(\xi )`$, Eqs.(2) reduce to two equations:
$`(\mu {\displaystyle \frac{1}{4}}\xi ^2)\theta _{2,0}(\xi )`$ $`=`$ $`\sqrt{2}g\theta _{1,1}(\xi )+\eta \theta _{2,0}(\xi )^2\theta _{2,0}(\xi ),`$ (17)
$`(\mu {\displaystyle \frac{1}{4}}\xi ^2)\theta _{1,1}(\xi )`$ $`=`$ $`2\sqrt{2}g\theta _{2,0}(\xi )`$ (19)
$`+\eta \theta _{1,1}(\xi )^2\theta _{1,1}(\xi ),`$
which can be solved numerically. Assume $`\xi _m`$ to be the boundary of the dressed condensates, from Eqs.(6) we then obtain the chemical potential $`\mu ^\pm =\pm 2g+\xi _m^2/4`$. Here the value of $`\xi _m`$ depends on the normalization condition $`_{n_1,n_2}_{\xi _m}^{\xi _m}𝑑\xi \theta _{n_1,n_2}(\xi )^2=1`$. For example, when $`g=7.29`$ and $`\eta =80`$, we numerically find that $`\xi _m^+=4.584`$ and $`\xi _m^{}=4.430`$ coresponding to the chemical potential being $`\mu ^+`$ and $`\mu ^{}`$, respectively.
We now find the elementary excitations of the system by linearizing Eqs.(6) around the dressed state solutions: $`\psi _{n_1,n_2}(\xi ,\tau )=e^{i\mu \tau }[\theta _{n_1,n_2}(\xi )+u_{n_1,n_2}(\xi )e^{i\omega \tau }+v_{n_1,n_2}^{}(\xi )e^{i\omega \tau }]`$, where$`(n_1,n_2)=(2,0)`$, $`(1,1)`$ and $`(0,2)`$, $`u(\xi )`$ and $`v(\xi )`$ represent small perturbations arround the dressed state with energies $`\mu \pm \omega `$. Substituting the expression of $`\psi _{n_1,n_2}(\xi ,\tau )`$ into Eqs.(6), in first order, we get the system of six equations for the linearized perturbations $`u(\xi )`$ and $`v(\xi )`$:
$`\omega u_{2,0}(\xi )`$ $`=`$ $`[H_L+2\eta \theta _{2,0}(\xi )^2\mu ]u_{2,0}(\xi )`$ (21)
$`+\sqrt{2}gu_{1,1}(\xi )+\eta \theta _{2,0}(\xi )^2v_{2,0}(\xi ),`$
$`\omega u_{1,1}(\xi )`$ $`=`$ $`[H_L+2\eta \theta _{1,1}(\xi )^2\mu ]u_{1,1}(\xi )`$ (23)
$`+\sqrt{2}g[u_{2,0}(\xi )+u_{0,2}(\xi )]+\eta \theta _{1,1}(\xi )^2v_{1,1}(\xi ),`$
$`\omega u_{0,2}(\xi )`$ $`=`$ $`[H_L+2\eta \theta _{0,2}(\xi )^2\mu ]u_{0,2}(\xi )`$ (25)
$`+\sqrt{2}gu_{1,1}(\xi )+\eta \theta _{0,2}(\xi )^2v_{0,2}(\xi ),`$
$`\omega v_{2,0}(\xi )`$ $`=`$ $`[H_L+2\eta \theta _{2,0}(\xi )^2\mu ]v_{2,0}(\xi )`$ (27)
$`+\sqrt{2}gv_{1,1}(\xi )+\eta \theta _{2,0}(\xi )^2u_{2,0}(\xi ),`$
$`\omega v_{1,1}(\xi )`$ $`=`$ $`[H_L+2\eta \theta _{1,1}(\xi )^2\mu ]v_{1,1}(\xi )`$ (29)
$`+\sqrt{2}g[v_{2,0}(\xi )+v_{0,2}(\xi )]+\eta \theta _{1,1}(\xi )^2u_{1,1}(\xi ),`$
$`\omega v_{0,2}(\xi )`$ $`=`$ $`[H_L+2\eta \theta _{0,2}(\xi )^2\mu ]v_{0,2}(\xi )`$ (31)
$`+\sqrt{2}gv_{1,1}(\xi )+\eta \theta _{0,2}(\xi )^2u_{0,2}(\xi ).`$
The normal modes of the above coupled equations are identical to the elemantary excitations determined via Bogoliubov method. In order to obtain the elementary excitations, we expand them in terms of eigenfunctions of the trapping potential $`q_\nu (\xi )`$ as
$`u_{n_1,n_2}(\xi )`$ $`=`$ $`{\displaystyle \underset{\nu }{}}b_{n_1,n_2}^\nu q_\nu (\xi ),`$ (32)
$`v_{n_1,n_2}(\xi )`$ $`=`$ $`{\displaystyle \underset{\nu }{}}c_{n_1,n_2}^\nu q_\nu (\xi ),`$ (33)
where $`q_\nu (\xi )`$ satisfies $`H_Lq_\nu (\xi )=(\nu +\frac{1}{2})q_\nu (\xi )`$.
It is difficult to get an exact solution for the normal modes of Eqs.(7). In what follows, we give numerical solution of them by using the TFA. For the out-of-phase solution (4), we note that $`\theta _{2,0}(\xi )^2=\theta _{0,2}(\xi )^2\theta _{2,0}(\xi =0)^2\xi _m^2/4\eta `$ for normal modes localized close to the center of the trapping potential due to $`\xi _m1`$.
Substitution of Eqs.(8) into Eqs.(7) gives a system of linear equations for the coefficients $`b_{n_1,n_2}^\nu `$ and $`c_{n_1,n_2}^\nu `$. The nonzero solution condition of the system of linear equations gives rise to the excitation spectrum of the elementary excitations. In the Fig. 1(a) we have plotted the excitation spectrum in terms of the index $`\nu `$ of the linear oscillator mode for the out-of-phase solution when $`\eta =80`$, $`g=7.29`$ and $`\eta _m=4.9324`$. Fig. 1(b) is the enlargement of the part of $`10<\omega ^2<20`$ for the curve $`c`$. From Fig.1(a) we see that in the region of $`0\nu 2`$, there exists one type of excitation spectrum, which is stable due to $`\omega ^2>0`$. In the region of $`\nu >2`$ we can observe three types of excitation spectrum. Two of them indicated by curve $`a`$ and $`b`$ are stable excitations. The third one indicated by curve $`c`$ is instable in the region of $`14<\nu <18`$ since $`\omega ^2`$ is negative. This is explicitly indicated in Fig. 1(b) through the enlargement of the curve $`c`$.
Similarly, we can numerically express the excitation spectrum for the in-phase solution in terms of the index $`\nu `$ of the linear oscillator mode. In Fig. 2, we give rise to the excitation spectrum for the in-phase solution when $`\eta =80`$, $`g=7.29`$ and $`\eta _m=4.932`$, where we have set that $`\theta _{2,0}(\xi )^2=\theta _{0,2}(\xi )^2\theta _{2,0}(\xi =0)^20.0336`$, $`\theta _{1,1}(\xi )^2\theta _{1,1}(\xi =0)^20.0925`$ for the branch $`\mu ^+`$, and $`\theta _{2,0}(\xi )^2=\theta _{0,2}(\xi )^20.04925`$, $`\theta _{1,1}(\xi )^2=\theta _{1,1}(0)^20.0775`$ for the branch $`\mu ^{}`$, which lead to $`\xi _m^+=4.584`$ and $`\xi _m^{}=4.431`$, respectively. From Fig. 2(a), we see that there exist different excitation spectrum for three different regions of $`\nu `$. In the first region $`0<\nu <8`$, we can find three kinds of excitation spectrum, and all of them are stable. In the second region $`8<\nu <13`$, there is only one kind of excitation spectrum which is unstable (i.e., $`\omega ^2<0`$). In the third region $`\nu >13`$, there exist three kinds of excitation spectrum. One of them is stable, and other two are unstable. These unstable regimes are explicitly indicated in Fig. 2(b), which is the enlargement of the curves in the vicinity of the point A in Fig. 2(a). From Fig. 2(c), we see that $`\omega ^2`$ has two positive real number solutions and one pure imaginary solution which is not indicated in the figure. these means that there does not exist solutions of $`\omega ^2<0`$. Therefore, the excitation spectrum of this case is always stable.
This work is supported in part by the National Climbing Program of China and NSF of China, the Excellence Young-Teacher Foundation of the Educational Department of China, and ECF and STF of Hunan Province. The authors thank Dr.Changsong Zhou for his help in preparing figures of the paper.
* Corresponding author.
Figure Captions
FIG.1. (a) Excitation spectrum $`\omega ^2`$ for the out-of-phase solution for $`\eta =80`$, $`g=7.29`$, and $`\eta _m=4.932`$. (b) The enlargement of the part of $`10<\omega ^2<20`$ for the curve $`c`$.
FIG.2. (a) Excitation spectrum $`\omega ^2`$ for the in-phase solution with $`\xi _m^+=4.584`$. (b) The enlargement of the curves in the vicinity of the point A in Fig.1(a). (c) Excitation spectrum $`\omega ^2`$ for the in-phase solution with $`\xi _m^{}=4.431`$. Here we set $`\eta =80`$, $`g=7.29`$.
|
warning/0004/cond-mat0004466.html
|
ar5iv
|
text
|
# Slow dynamics of Ising models with energy barriers
## I Introduction
A lot of efforts have been devoted in the last twenty years to understanding the behaviour of various glassy and disordered systems . Such systems, which include conventional glasses, spin glasses, amorphous semiconductors, and many others are of great importance both experimental and theoretical. However, despite intensive research, our understanding of such systems is still limited. For example, even the very nature of the glassy phase in spin-glasses is still a very controversial issue . Although they are much more abundant, conventional glasses seem to pose even a greater puzzle. Why do supercooled liquids fall out of equilibrium at a more or less well defined temperature? Why do they collapse into the glassy state when the cooling is fast enough and into the crystalline phase when the cooling is slow? These fundamental questions still await definitive answers. One of the important problems in physics of conventional glasses is the continuing lack of a satisfactory microscopic model of such systems. In this respect the situation is much better for spin-glasses where it is commonly accepted that models containing quenched disorder correctly describe physics of such systems. Lattice realizations of such models are a particularly valuable source of information about spin-glasses . The most realistic models of conventional glasses, so-called off-lattice models, still constitute an enormous computational challenge although progress in this field is also being made .
A model of conventional glasses should be capable of describing (at least) three phases: liquid, glass and crystal. The actual state of the system should be determined by control parameter(s) (e.g., temperature) and possibly also its history. Since the glass is regarded as a liquid trapped during the falling out of equilibrium, the model should possess such a trapping mechanism. In spin glasses the trapping mechanism is related with energy barriers generated by quenched disorder . On the other hand we do not expect the quenched disorder to be a relevant factor in conventional glasses because models with strong quenched disorder are unlikely to exhibit periodic solutions (which are needed for the model to be in the crystal phase). Recently, various lattice models, which do not contain quenched disorder, were studied which have some features of conventional glasses. Some of these models are infinite-dimensional and their thermodynamical properties can be found exactly . There are also finite dimensional models whose dynamics exhibit some glassy behaviour
Recently, it has been shown that the three dimensional Ising model with the four-spin (plaquette) interaction also exhibits some glassy features . This model undergoes a first-order phase transition between low-temperature (crystal) and high-temperature (liquid) phases. However, when conventional simulation techniques are used, the transition is screened by a very strong metastability during heating as well as cooling. For temperatures lower than the limit of metastability of the liquid phase, the model has a very slow coarsening dynamics. In addition, the zero-temperature characteristic length increases very slowly as a function of the inverse cooling rate, which is also an expected property of glasses. Further evidence of the glassy behaviour in this model has been recently reported by Swift et al. . They have shown that the glassy transition coincides with the divergence of a certain relaxation time and that aging properties of the model are also typical of glassy systems. They have also observed that some time correlation functions may decay as stretched exponentials.
These results strongly suggest that the model with four-spin interactions might describe important aspects of the glassy transition. It would be interesting to find which properties of this model are responsible for such a behaviour. It has been already suggested that the trapping mechanism might be related with diverging energy barriers. These barriers would arise in this model basically due to the same mechanism as in a model with competing nearest-neighbour and next-nearest-neighbour interactions examined by Shore et al. (the SS model for short). However, the behaviour of the SS model is not fully consistent with our conception of glasses since it orders too quickly under cooling . It was also suggested that the difference in the behaviour of the SS and four-spin models might be related with the degeneracy of the ground state in the four-spin model. This degeneracy might lead to some entropy barriers, which would be responsible for the strong metastability of the liquid phase.
In the present paper, using Monte Carlo simulations, we examine a certain class of three-dimensional Ising models which generate energy barriers. These models are described by the following Hamiltonian
$$H=J_1\underset{<i,j>}{}S_iS_jJ_2\underset{<<i,j>>}{}S_iS_jJ_4\underset{[i,j,k,l]}{}S_iS_jS_kS_l.$$
(1)
In the above expression $`<..>`$ and $`<<..>>`$ denote pairs of nearest and next-nearest neighbours, respectively, and $`[i,j,k,l]`$ stands for summation over elementary plaquettes. In general, these models have double degenerate ground state and our simulations suggest that the dynamical properties in this case are similar to the SS model. However, when the interaction constants are such that the model has a strongly degenerate ground state (gonihedric case), the dynamical properties change. Simulations suggest that two types of dynamical behaviour appear. In the first type the model behaves similarly to the already described four-spin model. In the second type, the glassy transition appears to coincide with the thermodynamic transition.
Such behaviour gives rise to the following questions: why a glassy transition appears only in certain systems with slow dynamics and what is its nature. Analysis of the ground-state structure and thermodynamic properties of models studied here prompts the following conjecture, which, if confirmed, would constitute an important result of the present paper: at the glassy transition the domain walls lose their surface tension, and, as a result, the glassy phase is composed of tensionless domains. Although based on the analysis of Ising models, we hope that such an interpretation might shed some light on the nature of the glassy transition in more realistic systems also. Moreover, such an interpretation of glassy phase is in accord with some recent hypotheses concerning the nature of the glassy phase in spin glasses .
In section II we discuss briefly the properties of the four-spin model. In section III we present the results of our simulations for the gonihedric case. The case of the doubly-degenerate ground state is discussed in section IV. Section V contains a summary of our results and some arguments on the nature of the glassy transition.
## II Ising model with plaquette interactions
This model corresponds to the case $`J_1=J_2=0,J_4=1`$, and has been already studied using Cluster Variational Method and Monte Carlo simulations .
Clearly, the ferromagnetic configuration is a ground-state configuration of this model. It is also easy to realize that flipping coplanar spins does not change the energy. Thus any configuration obtained from the ferromagnetic configuration by flipping coplanar spins is also a ground-state configuration. Moreover any combination of such coplanar flippings (even for crossing planes) does not increase the energy. Simple analysis along these lines shows that for the model on the lattice of the linear size $`L`$ the degeneracy of the ground state is equal to $`2^{3L}`$. Although ground state of this model is strongly degenerate its ground-state entropy is zero.
The model undergoes a first-order thermodynamic transition at $`T=T_\mathrm{c}3.6`$ which is, however, screened by very strong metastability . As a result, when heated or cooled, the transition observed in simulations is shifted to $`T3.9`$ or $`T3.4`$, respectively. Transitions at these spinodals are accompanied by peaks in the specific heat.
The low-temperature spinodal $`T3.4`$ seems to coincide with the glassy transition. Below this temperature the model exhibits very slow coarsening dynamics as well as aging properties which are characteristic of glassy systems . A certain characteristic time, which governs the relaxation of energy-energy correlation functions, also seems to diverge at this temperature . In addition, the behaviour of the model under continuous cooling supports the glassy-transition interpretation of this temperature .
## III Gonihedric Ising model
### A Ground state and thermodynamics
It has already been suggested that the slow dynamics of the four-spin model might be related with energy barriers generated in that model . These barriers arise due to the shape-dependence of the energy of excitations: it is not only the size of an excitation which determines its energy but also its shape. Such shape dependence appears also in the SS model.
Are there any other models which could have a similar property? In our opinion, the shape dependence of energy of excitations should be rather a robust feature of Ising-type models. It is only in some specific cases, like the standard nearest-neighbour case, when this energy does not depend on the shape of an excitation. In particular, energy barriers appear in model (1). The Hamiltonian of this model is quite general and it includes both the four-spin model and the Shore et al.’s model ($`J_4=0,J_1>0,J_2<0`$).
In the present section we examine a class of models described by this Hamiltonian, namely gonihedric models . These models correspond to the following choice of interaction constants: $`J_1=2k,J_2=\frac{k}{2}`$ and $`J_4=\frac{1}{2}(k1)`$. Gonihedric models have a strongly degenerate ground state. In addition to the ferromagnetic ground state any configuration obtained by flipping coplanar spins also minimizes the Hamiltonian. Any combination of such flips does not increases the energy, provided that flipping planes do not cross. As a particular example of such a ground-state we can mention lamellar configurations where e.g., every second plane of spins is flipped. Although lamellar structures constitute a legitimate ground-state, they do not survive at finite temperature as shown by Cirillo et al. using the Cluster Variational Method . We will return to this feature in the last section.
For $`k=0`$ the gonihedric model is equivalent to the four-spin Ising model. In this case the model has an additional symmetry which implies a larger degeneracy of the ground state since the flipping planes can now cross. As a result we obtain that antiferromagnetic configurations belong to the ground state. Further analysis of differences between the $`k=0`$ and the $`k0`$ cases is postponed to the last section.
Gonihedric models are expected to undergo a thermodynamic transition which for $`k<k_{\mathrm{tr}}`$ is of first order and for $`k>k_{\mathrm{tr}}`$ is of second order. Only very rough estimations of $`k_{\mathrm{tr}}(0.5)`$ are known . In this section we analyze dynamical properties of the gonihedric model for $`k=2`$, i.e., for a value with a continuous transition. Our results were obtained using a standard Monte Carlo method with random sequential update using Metropolis algorithm . Some details can be found elsewhere .
To find the thermodynamic transitions we measured the specific heat. Our simulations, which were made for various linear sizes $`L`$ up to $`L=40`$, locate the peak at $`T_\mathrm{c}2.35`$, which is a good agreement with the Cluster Variational Method estimation . The absence of hysteresis effects confirms that the transition at $`T=T_\mathrm{c}`$ is continuous. (Although the nature of the thermodynamic transition in gonihedric models is an interesting and still open problem, its further analysis is not an objective of the present paper).
### B Dynamics
An important indication of glassy dynamics is a slow evolution of a random quench. For usual models with nonconservative dynamics one expects that the characteristic length $`l`$ increases with time $`t`$ as $`lt^{1/2}`$. However, in glassy systems $`l`$ should increase much more slowly and presumably only logarithmically with time ($`l\mathrm{ln}t`$). Such a behaviour most likely appears in the SS model and in the four-spin model. In the following we present the results of our simulations of the evolution of quenches in the $`k=2`$ model. We measured the excess energy $`\delta E=EE_{\mathrm{}}`$, where $`E_{\mathrm{}}`$ is the equilibrium energy. Our results for temperatures $`T<T_\mathrm{c}`$ are shown in Fig. 1. To relate the characteristic length with the energy excess we can employ the frequently used relation
$$l1/\delta E.$$
(2)
With this identification from Fig. 1 we infer that for all the examined temperatures the asymptotic increase of $`l`$ is much slower than $`t^{1/2}`$. At the end of this section we will argue that the relation (2) most likely does not hold for this model and Fig. 1 actually suggest that the increase of $`l`$ is slower than $`t^{1/4}`$. Since there are no theoretical arguments for such a slow algebraic increase in our opinion it is quite plausible that asymptotically we have $`l\mathrm{ln}t`$. Such a slow increase of $`l`$ is most likely due to energy barriers.
An alternative technique to examine the dynamics of the model is to measure characteristic times of certain processes. For example, one can measure the average time $`\tau `$ needed for the inversion of a cubic like excitation. In this method, which was already applied to similar models , one prepares the system of the size $`L`$ with fixed boundary conditions and interior spins which are opposite to the boundary spins. One expects that after some time, the system will invert the interior spins. For a two-spin Ising model or SS model above the corner rounding transition, $`\tau L^2`$, which indicates a relatively fast dynamics (naive inversion of this relation gives $`lt^{1/2}`$). On the other hand, in the SS model below the corner rounding transition and in the four-spin model $`\tau `$ increases much faster, presumably exponentially, with $`L`$.
We measured the time needed for magnetization of the interior spins to reach the equilibrium value at a given temperature and the results are shown in Fig. 2. These results show that $`\tau `$ even at the highest examined temperature increases faster than $`L^2`$. This is a potential indication of an exponential increase $`\tau a^L(a>1)`$ in the entire low-temperature phase.
Additional confirmation of such behaviour is obtained from simulations of this model under the continuous cooling. Similarly to simulations of the four-spin model , we relax the random sample at a temperature $`T_0>T_\mathrm{c}`$ and then continuously lower the temperature according to the formula $`T(t)=T_0rt`$, where $`r`$ is the cooling rate. When the temperature is reduced below the critical point $`T_\mathrm{c}`$ the growth of order begins. The slower the cooling the more ordered is the system at the end of the cooling , i.e., at $`T=0`$ (see Fig. 3). To quantify the zero-temperature order we measure the excess energy $`\delta E`$ at $`T=0`$ and the result is shown in Fig. 4.
Using the relation (2) this data suggests that asymptotically $`lr^{1/2}`$. Such a relation, which indicates that the growth of order is relatively fast, holds for the two-spin Ising model and also for the SS model . However, this conclusion is based on the validity of the relation (2) and, similarly to the four-spin model , we want to argue that this relation does not hold. Our argument refers to the following property of all gonihedric models: the energy of cubic-like excitations scales as their linear size $`L`$ . Let us recall that in two-spin Ising model, this energy scales as the area of the excitation ($`L^2`$). Provided that the final configuration is composed of such domains of the size $`L`$ and using a simple dimensional argument we obtain
$$l\frac{1}{(\delta E)^{1/2}}.$$
(3)
A direct confirmation of the assumption about the structure of the configuration at the of the cooling process comes from visual inspection. In Fig. 5 we can see an example of single-layer configuration. One can clearly see cubic-like (i.e., non-rough) domains whose energy scales linearly with their size. Using the relation (3) Fig. 4 shows that the zero-temperature characteristic length scales as $`r^{1/4}`$ which is much slower than in the two-spin Ising model but faster than in the four-spin model.
In this section we used three independent techniques to probe the dynamics of the gonihedric model in the case of continuous thermodynamic transition $`k=2`$. Domain coarsening suggest that the model has a slow dynamics up to, at least, the temperature $`T=1.9`$ ($`T_\mathrm{c}2.35`$). An analysis of the size dependence of the characteristic time $`\tau `$ suggests that cubic-like domains remain non-rough at least up to the temperature $`T=2.1`$ Thus, a slow-dynamics regime is most likely extended up this temperature. Since this is very close to the critical point is it not unlikely that a slow-dynamics regime actually covers the whole low-temperature phase. The behaviour of the model under cooling confirms such a scenario: if there would be a certain temperature $`T_0<T_\mathrm{c}`$ such that for $`T_0<T<T_\mathrm{c}`$ the dynamics would be fast then for the slow cooling the growth of order would be dominated by the time spent in this temperature interval and we would have $`lr^{1/2}`$. Such a scenario takes place in the SS model . The growth of order in our $`k=2`$ model is much slower $`lr^{1/4}`$ and excludes the existence of such a temperature $`T_0`$ (unless it is very close to $`T_\mathrm{c}`$ and our simulations are not sufficient to detect the true asymptotic behaviour). Let us also notice that Shore et al. also analyzed certain SOS model for which (by necessity) $`T_0=T_\mathrm{c}`$. Using some scaling arguments they have shown that for this model one should have $`lr^{1/4}`$, which is in agreement with our numerical result.
## IV Off-gonihedric Ising model
The gonihedric case corresponds to a certain choice of interaction constants in the Hamiltonian (1). As we have noted this choice has important implications: the ground state is strongly degenerate and energy of excitations scale as their linear size and not as their area. In the present section we examine what is going on when the interaction constants of model (1) deviate from the gonihedric case .
As a particular example we choose: $`J_1=6,J_2=1,J_4=1/2`$, which differs from the gonihedric case $`k=2`$ by a modified nearest-neighbour coupling $`J_1`$. Such a model has a double-degenerate (ferromagnetic) ground state and our rough estimation of the critical temperature is $`T_\mathrm{c}12.5`$.
To examine the dynamics of this model we used the same techniques as described in the previous section. First, we examined the coarsening behaviour of this model. At low temperature (up to $`T3.0`$) we observed a very slow decrease of the energy toward the ground state value. Our data, which we do not present suggests that for such temperatures $`l`$ most likely increases logarithmically with time. However, above this temperature dynamics becomes much faster and presumably the characteristic length increases as $`lt^{1/2}`$.
Such behaviour is confirmed by the measurements of the characteristic time $`\tau `$ defined in the same way as in the previous section. Results of simulations are shown in Fig. 6. They indicate that for temperature $`T=6`$ and 8 the characteristic time $`\tau `$ increases as $`L^2`$ similarly to the SS model above the corner rounding transition.
The above results indicate that the dynamical behaviour of the model is very similar to the SS model. Namely, in the low temperature regime the model has a slow dynamics and rapidly (faster than $`L^2`$) increasing characteristic time $`\tau `$. However, within the ordered phase (i.e., for $`T<T_\mathrm{c}12.5`$ there is also a high temperature regime where dynamics is much faster. Presumably, in this regime the dynamics is similar to other nonconservative systems with scalar order parameter . Similarity between the model examined in this section and the SS model is in our opinion related with the structure of the ground state: both models have only double degenerate (ferromagnetic) ground states. It is well known in the statistical mechanics that degeneracy of the ground state plays a very important role in determining the thermodynamic behaviour of models (e.g., critical behaviour). It is probable that this is also an important factor for determining the dynamical behaviour of models.
## V Nature of the glassy phase
In the present paper we have examined the dynamics of models described by the Hamiltonian (1). Depending on the interaction constants we can distinguish three types of behaviour which are schematically shown in Fig. 7.
(a) Double-degenerate ground state
This, the most typical situation, appears in the off-gonihedric model studied in previous section and also in the SS model whose dynamics has been already examined in great details . The dynamics of the model a low temperatures ($`T<T_\mathrm{c}`$) has two regimes separated by a certain temperature $`T_{\mathrm{cr}}`$. For $`T<T_{\mathrm{cr}}`$ the model has slow dynamics with most likely logarithmically increasing characteristic length $`l`$. Such behaviour is related with the fact that at such temperature the model is below the corner-roughening transition . As a result, an evolving quench develops complicated structures of cubic-like excitations which are very stable and effectively block further coarsening dynamics . At $`T=T_{\mathrm{cr}}`$ the model undergoes the corner rounding transition and the blocking mechanism is no longer effective. As a result the fast (standard) dynamics is restored.
(b) Gonihedric case with continuous transition
In this case the entire low-temperature phase has slow dynamics, whose origin is similar to the case (a). Namely, a quench develops cubic-like structures which block further coarsening dynamics. The degeneracy of the ground state seems to be the most important difference between this and the off-gonihedric case. Thus, we relate the disappearance of the corner-rounding transition (or maybe its overlap with $`T_\mathrm{c}`$) with the infinite degeneracy of the ground state.
(c) Four-spin model
As we already mentioned, the four-spin model undergoes a dynamic transition which exhibits a lot of similarities with the glassy transition. It might be interesting to examine whether such a behaviour appears only at $`k=0`$ or persists also for some other (small) values of $`k`$.
Glassy transition: loss of surface tension
Why does this transition exist in the four-spin case and not in the other cases? It was already suggested that the difference between the four-spin model and the SS model is related to the degeneracy of the ground state . However, the above analysis of the gonihedric model with $`k=2`$ shows that the infinite degeneracy of the ground state is not sufficient for the model to have a glassy transition (of course, we limit our analysis to models which can generate diverging energy barriers and thus have slow coarsening dynamics). Why does the gonihedric case $`k=2`$ differ from the $`k=0`$ case (i.e. the four-spin model)? Both models have strongly degenerate ground state. The degeneracy equals $`2^{3L}`$ for $`k=0`$ and $`2^L`$ for $`k=2`$. Since in both cases degeneracy increase exponentially with the linear system size this difference does not seem important. In our opinion, however, the difference in the dynamical behaviour is related with the ground state structure of these models. As we already mentioned, for $`k=0`$ the flipping planes, which generate various ground-state configurations might cross. As a result, in addition to ferromagnetic-like configurations we obtain antiferromagnetic-like ones. For the $`k0`$ such crossings are not allowed and only ferromagnetic-like configurations are possible (we consider lamellar configurations also as ferromagnetic-like).
This difference has important implications. Let us note that for the four-spin model in addition to tensionless domain walls, which appear for example when a cubic ferromagnetic ’up’ domain is surrounded by ’down’ one, there are tensionful ones too. As an example of such a domain wall we can consider an antiferromagnetic domain surrounded by ferromagnetic one . At first sight it does not seem to be much different from the $`k=2`$ case. Indeed, when one considers a lamellar configuration where successive layers are of opposite sign (see Fig. 8) which is surrounded by a ferromagnetic domain than the excess energy of such a configuration scales as the area of the wall (i.e., the domain wall is tensionful). There is, however, an important question: do such configurations affect coarsening dynamics or, in other words, are they spontaneously generated in sufficient amounts? In our opinion the answer to this question is negative. Our first argument comes simply from the visual inspection of the snapshot configuration. In Fig. 5 one can see relatively large ferromagnetic-like domains but there is no indication of lamellar ones. The second argument comes from the Cluster Variational Method calculations by Cirillo et al. who have shown that the lamellar structures are equivalent (i.e., of the same energy) to the ferromagnetic one but only for the ground state. At non-zero temperatures they are always metastable. These arguments show why such configurations are not spontaneously generated during the evolution of the quench. They imply that for $`k=2`$ the dominant domain walls which exist at the late-time evolution of the quench are tensionless.
On the other hand, for the four-spin model, antiferromagnetic structures are fully equivalent to the ferromagnetic ones. In the liquid phase both ferromagnetic and antiferromagnetic domains are intertwined and form very complicated structures. As already noticed , tensionful domains usually have lowest energy barriers and the system can relatively easily remove the interior domains. On the contrary, tensionless domain walls have large energy barriers and their dynamics is much slower. It means that in the liquid phase dynamics is dominated by dynamics of tensionful domains and thus resembles the dynamics of two-spin Ising models.
It is in our opinion very likely that upon lowering the temperature the system will undergo a phase transition which will eliminate tensionful domain walls. Below that transition the energy of the system would be located mainly in tensionless domain walls. It means that at this transition the antiferromagnetic-ferromagnetic symmetry of the model would be spontaneously broken. In other words, at this transition the system selects a dominant type of domains, whether ferromagnetic or antiferromagnetic. This, rather novel, type of symmetry breaking is an essential ingredient of the transition which we tentatively identify as the glassy transition.
Finally, let us note that the idea that the glassy phase consists of a complicated mixture of tensionless domain walls appeared recently in the context of spin glasses . It suggests that, at least at the geometrical level, spin glasses and ordinary glasses might exhibit a lot of similarities. Their further explorations is, however, left as a future problem.
## ACKNOWLEDGMENTS
This work was partially supported by the EC IHP network “Discrete Random Geometries: From Solid State Physics to Quantum Gravity” HPRN-CT-1999-000161 and the ESF network “Geometry and Disorder: From Membranes to Quantum Gravity”. The work of DJ and DE was also partially supported by an Acciones Integradas grant.
|
warning/0004/math0004146.html
|
ar5iv
|
text
|
# Embeddings of ℓ_𝒑 into non-commutative spaces
## 1. Introduction
The study of rearrangement invariant Banach spaces of measurable functions is a classical theme. Several studies have been devoted on characterizations of subspaces of rearrangement invariant spaces. Recently, the theory of rearrangement invariant Banach spaces of measurable operators affiliated with semi-finite von Neumann algebra have emerged as the natural non-commutative generalizations of Köthe functions spaces. This theory, which is based on the theory of non-commutative integration introduced by Segal (see ), replaces the classical duality $`(L_{\mathrm{}}(\mu ),L_1(\mu ))`$ by the duality between a semi-finite von Neumann algebra and its predual. It provides a unified approach to the study of unitary ideals and rearrangement invariant spaces. Several authors have considered these non-commutative spaces of measurable operators (see for instance, , , , and ).
The purpose of the present paper is to examine the subspaces of symmetric spaces of measurable operators in which the norm topology and the measure topology coincide and subspaces generated by disjointly supported basic sequences. The classical spaces $`L_p(\mu )`$ are of central inportance and results in their structures go back to the work of Banach. Since their introduction by Lorentz in 1950, the Lorentz function spaces $`L_{p,q}`$ have been found to be of special interests in many aspects of analysis and probability theory. In and , Carothers and Dilworth studied the spaces $`L_{p,q}[0,1]`$ and $`L_{p,q}[0,\mathrm{})`$. They proved, among other things, that for some appropriate values of the indices $`p`$ and $`q`$, $`L_{p,q}[0,\mathrm{})`$ does not contain $`\mathrm{}_p`$. Presisely:
###### Theorem 1.1.
Let $`0<p<\mathrm{}`$, $`0<q<\mathrm{}`$, $`pq`$, $`p2`$. Then $`\mathrm{}_p`$ does not embed into $`L_{p,q}[0,\mathrm{})`$.
Motivated by Theorem 1.1, we examine the subspace structure of the non commutative Lorentz spaces $`L_{p,q}(,\tau )`$, where $`(,\tau )`$ is a semi-finite von Neumann algebra. The principal result of the present paper is that the main results of and extend to the non-commutative settings. The initial basic question, that led to the consideration of these Lorentz spaces, is the question of embeddings of $`\mathrm{}_p`$ into $`L_p(,\tau )`$. Clearly, any disjointly supported basic sequence in $`L_p(,\tau )`$ is isomorphic to $`\mathrm{}_p`$. For the commutative case, Rosenthal proved in that if $`(\mathrm{\Omega },\mathrm{\Sigma },\mu )`$ is a $`\sigma `$-finite measure space, $`1p<2`$ and $`X`$ be a subspace of $`L_p(\mathrm{\Omega },\mathrm{\Sigma },\mu )`$ containing $`\mathrm{}_p`$ then the norm topology and the measure topology do not coincide on $`X`$. For $`0<p<1`$, the same result can be found implicitely in a paper of Kalton . This implies that for $`0<p<2`$, any basic sequence in $`L_p(\mathrm{\Omega },\mathrm{\Sigma },\mu )`$ that is equivalent to $`\mathrm{}_p`$ is essentially a perturbation of a disjointly supported basic sequence. We establish, as applications of our results on Lorentz spaces, that Kalton and Rosenthal’s results extend to $`L_p(,\tau )`$.
The paper is organized as follows. In Section 2 below, we gather some necessary definitions and present some basic facts conserning symmetric spaces of measurable operators that will be needed throughout. In particular, we prove a Kadec-Pełczyński type dichotomy for basic sequences in symmetric spaces of measurable operators of Rademacher type 2, generalizing a result of Sukochev (see ). The final section is devoted entirely to the study of subspaces of Lorentz spaces and its applications. In particular, Theorem 1.1 is generalized to the non-commutative case. Our approach relies on a disjointification techniques based on the non-commutative analogue of the Khintchine’s inequalities of Lust-Piquard and Pisier ( and ).
## 2. Definitions and Preliminaries
We begin by recalling some definitions and facts about function spaces. Let $`E`$ be a complex quasi-Banach lattice. If $`0<p<\mathrm{}`$, then $`E`$ is said to be $`p`$-convex (respectively $`p`$-concave) if there exists a constant $`C>0`$ such that all finite sequence $`\{x_n\}`$ in $`E`$,
$$\left(|x_n|^p\right)^{\frac{1}{p}}_EC\left(x_n^p\right)^{\frac{1}{p}}$$
$$\left(\text{resp.}\left(|x_n|^p\right)^{\frac{1}{p}}_EC^1\left(x_n^p\right)^{\frac{1}{p}}\right).$$
The least constant $`C`$ is called the $`p`$-convexity (respectively $`p`$-concavity) constant of $`E`$ and is denoted by $`M^{(p)}(E)`$ (respectively $`M_{(p)}(E)`$).
For $`0<p<\mathrm{}`$, $`E^{(p)}`$ will denote the quasi-Banach lattice defined by
$$E^{(p)}:=\{x:|x|^pE\}$$
equipped with
$$x_{E^{(p)}}=|x|^p_E^{1/p}.$$
It is easy to verify that if $`E`$ is $`\alpha `$-convex and $`q`$-concave then $`E^{(p)}`$ is $`\alpha p`$-convex and $`qp`$-concave with $`M^{(\alpha p)}(E^{(p)})M^{(\alpha )}(E)^{1/p}`$ and $`M_{(qp)}(E^{(p)})M_{(q)}(E)^{1/p}`$. Consequently, if $`E`$ is $`\alpha `$-convex then $`E^{(1/\alpha )}`$ is $`1`$-convex and therefore can be equivalently renormed to be a Banach lattice ().
The quasi-Banach lattice $`E`$ is said to satisfy a lower $`q`$-estimate (respectively upper $`p`$-estimate) if there exists a positive constant $`C>0`$ such that for all finite sequence of mutually disjoint elements of $`E`$
$$(x_n_E^q)^{1/q}Cx_n_E$$
$$\left(\text{resp.}(x_n_E^p)^{1/p}C^1x_n_E\right).$$
We denote by $``$ a semi-finite von Neumann algebra on the Hilbert space $``$, with a fixed faithful and normal semi-finite trace $`\tau `$. The identity in $``$ is denoted by $`\mathrm{𝟏}`$, and we denote by $`^p`$ the set of all projections in $``$. A linear operator $`x`$:dom$`(x)`$, with domain dom$`(x)`$, is called affiliated with $``$ if $`ux=xu`$ for all unitary $`u`$ in the commutant $`^{}`$ of $`.`$ The closed and densely defined operator $`x`$, affiliated with $``$ , is called $`\tau `$-measurable if for every $`ϵ>0`$ there exists $`p^p`$ such the $`p()`$dom$`(x)`$ and $`\tau (\mathrm{𝟏}p)<ϵ`$. With the sum and product defined as the respective closures of the algebraic sum and product, $`\stackrel{~}{}`$ is a \*-algebra. For standard facts concerning von Neumann algebras, we refer to and . We recall the notion of generalized singular value function . Given a self-adjoint operator $`x`$ in $``$ we denote by $`e^x()`$ the spectral measure of $`x`$. Now assume that $`x\stackrel{~}{}`$. Then $`e^{|x|}(B)`$ for all Borel sets $`B`$, and there exists $`s>0`$ such that $`\tau (e^{|x|}(s,\mathrm{}))<\mathrm{}`$. For $`x\stackrel{~}{}`$ and $`t0`$ we define
$$\mu _t(x)=inf\{s0:\tau (e^{|x|}(s,\mathrm{}))t\}.$$
The function $`\mu (x):[0,\mathrm{})[0,\mathrm{}]`$ is called the generalized singular value function (or decreasing rearrangement) of $`x`$; note that $`\mu _t(x)<\mathrm{}`$ for all $`t>0`$. Suppose that $`a>0`$. If we consider $`=L_{\mathrm{}}([0,a),m)`$, where $`m`$ denotes Lebesgue measure on the interval $`[0,a)`$, as an Abelian von Neumann algebra acting via multiplication on the Hilbert space $`=L_2([0,a),m)`$, with the trace given by integration with respect to $`m`$, it is easy to see that $`\stackrel{~}{}`$ consists of all measurable functions on $`[0,a)`$ which are bounded except on a set of finite measure. Further, if $`f\stackrel{~}{}`$, then the generalized singular value function $`\mu (f)`$ is precisely the classical non-increasing rearrangement of the function $`|f|`$. On the other hand, if $`(,\tau )`$ is the space of all bounded linear operators in some Hilbert space equipped with the canonical trace $`tr`$, then $`\stackrel{~}{}=`$ and, if $`x`$ is compact, then the generalized singular value function $`\mu (x)`$ may be identified in a natural manner with the sequence $`\{\mu _n(x)\}_{n=0}^{\mathrm{}}`$ of singular values of $`|x|=\sqrt{x^{}x}`$, repeated according to multiplicity and arranged in non-increasing order. By $`L_0([0,a),m)`$ we denote the space of all $``$-valued Lebesgue measurable functions on the interval $`[0,a)`$ (with identification $`m`$-a.e.). A quasi-Banach space $`(E,__E)`$, where $`EL_0([0,a),m)`$ is called a rearrangement-invariant Banach function space on the interval $`[0,a)`$, if it follows from $`fE,gL_0([0,a),m)`$ and $`\mu (g)\mu (f)`$ that $`gE`$ and $`g_Ef_E`$. If $`(E,_E)`$ is a rearrangement-invariant quasi-Banach function space on $`[0,a)`$, then $`E`$ is said to be symmetric if $`f,gE`$ and $`gf`$ imply that $`g_Ef_E`$. Here $`gf`$ denotes submajorization in the sense of Hardy-Littlewood-Polya :
$$_0^t\mu _s(g)𝑑s_0^t\mu _s(f)𝑑s,\mathrm{for}\mathrm{all}t>0.$$
The general theory of rearrangement-invariant spaces may be found in and .
Given a semi-finite von Neumann algebra $`(,\tau )`$ and a symmetric quasi-Banach function space $`(E,__E)`$ on $`([0,\tau (\mathrm{𝟏})),m)`$, we define the non-commutative space $`E(,\tau )`$ by setting
$$E(,\tau ):=\{x\stackrel{~}{}:\mu (x)E\}\text{with}$$
$$x_{_{E(,\tau )}}:=\mu (x)__E\text{for}xE(,\tau ).$$
It is known that if $`E`$ is $`\alpha `$-convex for some $`0<\alpha <\mathrm{}`$ with $`M^{(\alpha )}(E)=1`$ then $`_{_{E(,\tau )}}`$ is a norm for $`\alpha 1`$ and an $`\alpha `$-norm if $`0<\alpha <1`$. The space $`(E(,\tau ),_{_{E(,\tau )}})`$ is a $`\alpha `$-Banach space. Moreover, the inclusions
$$L_\alpha (,\tau )E(,\tau )L_\alpha (,\tau )+.$$
hold with continuous embeddings. We remark that if $`0<p<\mathrm{}`$ and $`E=L_p([0,\tau (\mathrm{𝟏}))`$ then $`E(,\tau )`$ coincides with the definition of $`L_p(,\tau )`$ as in and . In particular, if $`=()`$ with the standard trace then these $`L_p`$-spaces are precisely the Schatten classes $`𝒞_p`$.
We recall that the topology defined by the metric on $`\stackrel{~}{}`$ obtained by setting
$$d(x,y)=inf\{t0:\mu _t(xy)t\},\text{for}x,y\stackrel{~}{},$$
is called the measure topology. It is well-known that a net $`(x_\alpha )_{\alpha I}`$ in $`\stackrel{~}{}`$ converge to $`x\stackrel{~}{}`$ in measure topology if and only if for every $`ϵ>0`$, $`\delta >0`$, there exists $`\alpha _0I`$ such that whenever $`\alpha \alpha _0`$, there exists a projection $`p^p`$ such that
$$(x_\alpha x)p_{}<ϵ\text{and}\tau (\mathrm{𝟏}p)<\delta .$$
It was shown in that $`(\stackrel{~}{},d)`$ is a complete metric, Hausdorff, topological -algebra. For $`x\stackrel{~}{}`$, the right and left support projections of $`x`$ are denoted by $`r(x)`$ and $`l(x)`$ respectively. Operators $`x,y\stackrel{~}{}`$ are said to be right (respectively, left) disjointly supported if $`r(x)r(y)=0`$ (respectively, $`l(x)l(y)=0`$).
###### Definition 2.1.
Let $`E`$ be a symmetric quasi-Banach function space on $`[0,\tau (\mathrm{𝟏}))`$. We say that a subspace $`X`$ of $`E(,\tau )`$ is strongly embedded into $`E(,\tau )`$ if the $`_{E(,\tau )}`$-topology and the measure topology on $`X`$ coincide.
The following definition was introduced in as an analogue of the uniform integrability of family of functions.
###### Definition 2.2.
Let $`E`$ be a symmetric quasi-Banach function space on $`[0,\tau (\mathrm{𝟏}))`$. A bounded subset $`K`$ of $`E(,\tau )`$ is said to be $`E`$-uniformly-integrable if $`\underset{n\mathrm{}}{lim}\underset{xK}{sup}e_nxe_n_{E(,\tau )}=0`$ for every decreasing sequence $`\{e_n\}_{n=1}^{\mathrm{}}`$ of projections with $`e_n_n0`$.
A non-commutative extension of the Kadec-Pełczyński subsequence decomposition relative to the above notion of uniform integrability was considered in (see Theorem 3.1, Theorem 3.7 and Corollary 3.8) and will be used repeatedly throughout this paper.
For the remaining of this section, we will present some results, some of which may be of independent interests, that we will need in the sequel.
The following proposition is essentially due to Sukochev ().
###### Proposition 2.3.
Let $`E`$ be $`\alpha `$-convex with constant $`1`$ and assume that $`E`$ is order continuous. Let $`\{x_n\}_{n=1}^{\mathrm{}}`$ be a basic sequence in $`E(,\tau )`$ such that $`\{x_n\}_{n=1}^{\mathrm{}}`$ is both right and left disjointly supported then $`\{x_n\}_{n=1}^{\mathrm{}}`$ is equivalent to a disjointly supported basic sequence in $`E`$.
###### Proof.
For each $`n1`$, let $`q_n:=l(x_n)`$ and $`p_n:=r(x_n)`$ be the left and right support projection of $`x_n`$ respectively. Both sequences $`\{q_n\}_{n=1}^{\mathrm{}}`$ and $`\{p_n\}_{n=1}^{\mathrm{}}`$ are mutually disjoint and for every $`n1`$, $`x_n=q_nx_np_n`$. For any finite sequence of scalars $`\{a_i\}_{i=1}^n`$,
$$\begin{array}{cc}\hfill \left|\underset{i=1}{\overset{n}{}}a_ix_i\right|^2& =\left(\underset{i=1}{\overset{n}{}}\overline{a}_ip_ix_i^{}q_i\right)\left(\underset{i=1}{\overset{n}{}}a_iq_ix_ip_i\right)\hfill \\ & =\underset{i=1}{\overset{n}{}}|a_i|^2p_ix_i^{}q_ix_ip_i\hfill \\ & =\left|\underset{i=1}{\overset{n}{}}a_i|x_i|\right|^2.\hfill \end{array}$$
Note that $`\{|x_i|\}_{i=1}^{\mathrm{}}`$ is disjointly supported by the projections $`\{p_i\}_{i=1}^{\mathrm{}}`$. For each $`i1`$, the semi-finiteness of $`p_i`$ implies that the family $`\{e_\beta \}_\beta `$ of all projections in $`p_ip_i`$ of finite trace satisfies $`0e_\beta _\beta p_i`$. Since $`E`$ is order-continuous, it follows that $`e_\beta |x_i|e_\beta |x_i|_\beta 0`$.
For each $`i1`$, choose a projection $`\stackrel{~}{p}_ip_i`$ such that $`\tau (\stackrel{~}{p}_i)<\mathrm{}`$ and $`\stackrel{~}{p}_i|x_i|\stackrel{~}{p}_i|x_i|^\alpha 2^i`$.
Claim: The sequence $`\{\stackrel{~}{p}_i|x_i|\stackrel{~}{p}_i\}_{i=1}^{\mathrm{}}`$ is equivalent to $`\{|x_i|\}_{i=1}^{\mathrm{}}`$.
Let $`p=_{i=1}^{\mathrm{}}\stackrel{~}{p}_i`$. For any $`x=_{i=1}^{\mathrm{}}a_i|x_i|\overline{\text{span}}\{|x_i|,i1\}`$, we have $`_{i=1}^{\mathrm{}}a_i\stackrel{~}{p}_i|x_i|\stackrel{~}{p}_i=p(_{i=1}^{\mathrm{}}a_i|x_i|)p`$ so the series $`_{i=1}^{\mathrm{}}a_i\stackrel{~}{p}_i|x_i|\stackrel{~}{p}_i`$ is convergent whenever $`_{i=1}^{\mathrm{}}a_i|x_i|`$ does. Conversely, if $`\{a_n\}_{n=1}^{\mathrm{}}`$ is a bounded sequence of scalars such that $`_{i=1}^{\mathrm{}}a_i\stackrel{~}{p}_i|x_i|\stackrel{~}{p}_i`$ is convergent, then for any subset $`S`$ of $``$,
$$\begin{array}{cc}\hfill \underset{iS}{}a_i|x_i|_{E(,\tau )}^\alpha & \underset{iS}{}a_i(\stackrel{~}{p}_i|x_i|\stackrel{~}{p}_i)_{E(,\tau )}^\alpha +\underset{iS}{}a_i(\stackrel{~}{p}_i|x_i|\stackrel{~}{p}_i|x_i|)_{E(,\tau )}^\alpha \hfill \\ & \underset{iS}{sup}|a_i|^\alpha \underset{iS}{}2^i+\underset{iS}{}a_i\stackrel{~}{p}_i|x_i|\stackrel{~}{p}_i_{E(,\tau )}^\alpha .\hfill \end{array}$$
This shows that the series $`_{i=1}^{\mathrm{}}a_i|x_i|`$ is convergent. Let $`C_1`$ and $`C_2`$ be positive constants so that for any finite sequence of scalars $`\{a_i\}_{i=1}^n`$,
$$C_1\underset{i=1}{\overset{n}{}}a_i|x_i|_{E(,\tau )}\underset{i=1}{\overset{n}{}}a_i\stackrel{~}{p}_i|x_i|\stackrel{~}{p}_i_{E(,\tau )}C_2\underset{i=1}{\overset{n}{}}a_i|x_i|_{E(,\tau )}.$$
If $`\alpha _1=0`$ and $`\alpha _n=\underset{i=1}{\overset{n}{}}\tau (\stackrel{~}{p}_i)<\mathrm{}`$, set $`f_n:=\mu _{()\alpha _n}(\stackrel{~}{p}_n|x_n|\stackrel{~}{p}_n)`$ for $`n1`$. The sequence $`\{f_n\}_{n=1}^{\mathrm{}}`$ is disjointly supported in $`E(0,\tau (\mathrm{𝟏}))`$ and $`\{f_n\}_{n=1}^{\mathrm{}}`$ is isometrically isomorphic to $`\{\stackrel{~}{p}_n|x_n|\stackrel{~}{p}_n\}_{n=1}^{\mathrm{}}`$. For any finite sequence of scalars $`\{a_i\}_{i=1}^n`$,
$$\begin{array}{cc}\hfill C_1\underset{i=1}{\overset{n}{}}a_ix_i_{E(,\tau )}& =C_1\underset{i=1}{\overset{n}{}}a_i|x_i|_{E(,\tau )}\hfill \\ & \underset{i=1}{\overset{n}{}}a_i\stackrel{~}{p}_i|x_i|\stackrel{~}{p}_i_{E(,\tau )}\hfill \\ & =\underset{i=1}{\overset{n}{}}a_if_i_E\hfill \\ & C_2\underset{i=1}{\overset{n}{}}a_ix_i_{E(,\tau )}.\hfill \end{array}$$
The proof is complete. ∎
###### Proposition 2.4.
Let $`E`$ be an order continuous symmetric quasi-Banach function space on $`[0,\tau (\mathrm{𝟏}))`$ that is $`\alpha `$-convex with constant $`1`$ for some $`0<\alpha 1`$ and suppose that $`E`$ satisfies a lower $`q`$-estimate with constant $`1`$ for some $`q\alpha `$. If $`X`$ is a subspace of $`E(,\tau )`$ then either $`X`$ is strongly embedded into $`E(,\tau )`$ or there exists a normalized basic sequence $`\{y_n\}_{n=1}^{\mathrm{}}`$ in $`X`$, a mutually disjoint sequence of projections $`\{e_n\}_{n=1}^{\mathrm{}}`$ in $``$ such that
$$\underset{n\mathrm{}}{lim}y_ne_ny_ne_n_{E(,\tau )}=0.$$
In particular, $`\{y_n\}_{n=1}^{\mathrm{}}`$ has a subsequence that is equivalent to a disjointly supported basic sequence in $`E`$. Moreover, if $`X`$ has a basis then the sequence $`\{y_n\}_{n=1}^{\mathrm{}}`$ can be chosen to be a block basis of the basis of $`X`$.
###### Proof.
Assume that $`X`$ is not strongly embedded into $`E(,\tau )`$ and set $`j:E(,\tau )\stackrel{~}{}`$ the natural inclusion. Since $`X`$ is not strongly embedded into $`E(,\tau )`$, the restriction $`j|_X`$ is not an isomorphism. There exists a sequence $`\{y_n\}_{n=1}^{\mathrm{}}`$ in the unit sphere of $`X`$ which converges to zero in measure. Note that the bounded set $`\{y_n,n1\}`$ cannot be $`E`$-uniformly integrable. Applying the non-commutative Kadec-Pełczyński subsequence decomposition to the sequence $`\{y_n\}_{n=1}^{\mathrm{}}`$ on $`E(,\tau )`$ (see Theorem 3.7), there exist a subsequence of $`\{y_n\}_{n=1}^{\mathrm{}}`$ (which we will denote again by $`\{y_n\}_{n=1}^{\mathrm{}}`$ for simplicity) and a mutually disjoint sequence of projections $`\{e_n\}_{n=1}^{\mathrm{}}`$ in $``$ such that the set $`\{y_ne_ny_ne_n,n1\}`$ is $`E`$-uniformly integrable and since $`\{y_ne_ny_ne_n\}_{n=1}^{\mathrm{}}`$ converges to zero in measure, we get that $`lim_n\mathrm{}y_ne_ny_ne_n_{E(,\tau )}=0`$.
Assume now that $`X`$ has a basis $`\{x_n\}_{n=1}^{\mathrm{}}`$. We will show that the sequence $`\{y_n\}_{n=1}^{\mathrm{}}`$ above can be chosen to be a block basis of $`\{x_n\}_{n=1}^{\mathrm{}}`$. In fact since $`j(B_X)`$ cannot be a neighborhood of zero for the (relative) measure topology on $`X`$, for every $`ϵ>0`$, $`B_\stackrel{~}{}(0,ϵ)XB_X`$ (where $`B_\stackrel{~}{}(0,ϵ)`$ denotes the ball centered at zero and with radius $`ϵ`$ relative to the metric of the measure topology). Denote by $`\pi _n`$ the projection $`X`$ onto $`\overline{\text{span}}\{x_k,kn\}`$. Fix $`z_1S_XB_\stackrel{~}{}(0,2^1)`$ and choose $`k_11`$ so that $`z_1\pi _{k_1}(z_1)<2^1`$. The restriction of $`j`$ on $`(Id\pi _{k_1})(X)`$ cannot be an isomorphism. As above, one can choose $`z_2S_XB_\stackrel{~}{}(0,2^2)`$ and $`\pi _{k_1}(z_2)=0`$. Inductively, one can construct a sequence $`\{z_n\}_{n=1}^{\mathrm{}}`$ in $`S_X`$ and a strictly increasing sequence of integers $`\{k_n\}_{n=1}^{\mathrm{}}`$ such that:
* $`z_nB_\stackrel{~}{}(0,2^n)`$ for all $`n1`$;
* $`z_n\pi _{k_n}(z_n)<2^n`$ for all $`n1`$;
* $`(Id\pi _{k_n})(z_{n+1})=0`$ for all $`n1`$.
Set $`y_n:=\pi _{k_n}(z_n)`$ for all $`n1`$. Clearly $`\{y_n\}_{n=1}^{\mathrm{}}`$ is a block basic sequence, $`y_n^\alpha 12^{n\alpha }`$ for all $`n1`$ and $`\{y_n\}_{n=1}^{\mathrm{}}`$ converges to zero in measure. The proof is complete. ∎
The next result can be viewed as a non-commutative analogue of Proposition 1.c.10 of (p.39). Below, $`\{r_n()\}_{n=1}^{\mathrm{}}`$ denote the sequence of the Rademacher functions.
###### Proposition 2.5.
Let $`E`$ be a symmetric Banach function space on $`[0,\tau (\mathrm{𝟏}))`$. Assume that $`E`$ is order continuous and satisfy the Fatou property. Let $`\{x_n\}_{n=1}^{\mathrm{}}`$ be a sequence in $`E(,\tau )`$ with:
* $`x_n=1`$ for all $`n1`$;
* there exists a projection $`e`$ with $`\tau (e)<\mathrm{}`$ and $`ex_n=x_n`$ for all $`n1`$.
Then either there exists a constant $`C>0`$ such that for every choice of scalars $`\{a_n\}_{n=1}^{\mathrm{}}`$, we have
$$_0^1\underset{i=1}{\overset{n}{}}r_i(t)a_ix_i_{E(,\tau )}𝑑tC\left(\underset{i=1}{\overset{n}{}}|a_i|^2\right)^{\frac{1}{2}}$$
for every $`n1`$, or $`\{x_n\}_{n=1}^{\mathrm{}}`$ has a subsequence $`\{x_{n_j}\}_{j=1}^{\mathrm{}}`$ which is an unconditional basic sequence equivalent to a disjoint element of $`E`$.
###### Proof.
For $`xE(,\tau )`$, we set (as in ),
$$\sigma (x,ϵ):=\chi _{[ϵx_{E(,\tau )},\mathrm{})}(|x|)$$
and
$$M_{E(,\tau )}(ϵ):=\{xE(,\tau ),\tau (\sigma (x,ϵ))ϵ\}.$$
Assume first that for every $`ϵ>0`$, there exists $`n_ϵ`$ such that $`|x_n^{}|M_{E(,\tau )}(ϵ)`$. We remark that $`|x_n^{}|`$ is supported by the finite projection $`e`$. There exists a subsequence $`\{x_{n_j}\}_{j=1}^{\mathrm{}}`$ such that $`\{|x_{n_j}^{}|\}_{j=1}^{\mathrm{}}`$ converges to zero in measure. In particular, $`\{x_{n_j}\}_{j=1}^{\mathrm{}}`$ converge to zero in measure. By the Kadec-Pełczyński subsequence decomposition on $`E(,\tau )`$ (see , Corollary 3.8), there exists a further subsequence (which we will denote again by $`\{x_{n_j}\}_{j=1}^{\mathrm{}}`$) and a disjoint sequence of projections $`\{e_j\}_{j=1}^{\mathrm{}}`$ so that the set $`\{x_{n_j}e_jx_{n_j}e_j,j1\}`$ is $`E`$-uniformly integrable so by (Proposition 2.8),
$$\underset{j\mathrm{}}{lim}x_{n_j}e_jx_{n_j}e_j=0$$
which shows that a subsequence of $`\{x_{n_j}\}_{j=1}^{\mathrm{}}`$ can be taken to be equivalent to a disjoint sequence of $`E`$.
On the other hand, if $`\{|x_n^{}|,n1\}M_{E(,\tau )}(ϵ)`$ for some $`ϵ>0`$ then
$$\begin{array}{cc}\hfill 1=x& =|x_n^{}|\hfill \\ & |x_n^{}|_{L_1(,\tau )+}\hfill \\ & \left(\text{max}(1,\tau (e))\right)^1|x_n^{}|_{L_1(,\tau )}\hfill \\ & ϵ\left(\text{max}(1,\tau (e))\right)^1\tau \left(\sigma (|x_n^{}|,ϵ)\right)\hfill \\ & ϵ^2\left(\text{max}(1,\tau (e))\right)^1\hfill \end{array}$$
So for every $`n1`$, $`x_n_1=x_n^{}_1=|x_n^{}|_1ϵ^2\left(\text{max}(1,\tau (e))\right)^1`$. Since $`L_1(,\tau )`$ is of cotype 2 (), there exists $`A_1>0`$ such that
$$\begin{array}{cc}\hfill _0^1\underset{i=1}{\overset{n}{}}r_i(t)a_ix_i_{E(,\tau )}𝑑t& =_0^1e\left(\underset{i=1}{\overset{n}{}}r_i(t)a_ix_i\right)_{E(,\tau )}𝑑t\hfill \\ & _0^1e\left(\underset{i=1}{\overset{n}{}}r_i(t)a_ix_i\right)_{L_1(,\tau )+}𝑑t\hfill \\ & \left(\text{max}(1,\tau (e))\right)^1_0^1e\left(\underset{i=1}{\overset{n}{}}r_i(t)a_ix_i\right)_{L_1(,\tau )}𝑑t\hfill \\ & A_1\left(\text{max}(1,\tau (e))\right)^1\left(\underset{i=1}{\overset{n}{}}|a_i|^2x_i_1^2\right)^{1/2}\hfill \\ & A_1ϵ^2\left(\text{max}(1,\tau (e))\right)^2\left(\underset{i=1}{\overset{n}{}}|a_i|^2\right)^{1/2}.\hfill \end{array}$$
The proof is complete. ∎
###### Remarks 2.6.
We do not know if condition $`(ii)`$ can be removed. The same conclusion holds if $`(ii)`$ is replaced by: there exists a projection $`e`$ with $`\tau (e)<\mathrm{}`$ and $`x_ne=x_n`$ for all $`n1`$.
The following theorem gives a description of symmetric basic sequences in some symmetric space of measurable operators with type 2 and can be viewed as semi-finite generalization of Theorem 2.4 of .
###### Theorem 2.7.
Let $`E`$ be an order continuous rearrangement invariant Banach function space on $`[0,\tau (\mathrm{𝟏}))`$ with the Fatou property and assume that $`E(,\tau )`$ is of type 2. Then every symmetric basic sequence in $`E(,\tau )`$ either has a block basic sequence equivalent to a disjointly supported sequence in $`E`$ or is equivalent to $`\mathrm{}_2`$.
If $`\tau (\mathrm{𝟏})<\mathrm{}`$, the theorem is a simple corollary of the above proposition with the word “block basic sequence” replaced by “subsequence”.
For the general case, choose a mutually orthogonal family $`\{f_i\}_{iI}`$ of projections in $``$ with $`_{iI}f_i=\mathrm{𝟏}`$ for the strong operator topology and $`\tau (f_i)<\mathrm{}`$ for all $`iI`$. Let $`\{x_n\}_{n=1}^{\mathrm{}}`$ be a symmetric basic sequence in $`E(,\tau )`$. Using a similar argument as in , one can get an at most countable subset $`\{f_k\}_{k=1}^{\mathrm{}}`$ of $`\{f_i\}_{iI}`$ such that for each $`f_i`$ outside of $`\{f_k\}_{k=1}^{\mathrm{}}`$ and $`n1`$, $`f_ix_n=x_nf_i=0`$. Let $`f=_kf_k`$. For every $`n1`$, we have $`fx_n=x_nf=xn`$. Replacing $``$ by $`ff`$ and $`\tau `$ by its restriction on $`ff`$, we may assume that $`f=\mathrm{𝟏}`$. For every $`n1`$, set $`e_n:=_{k=1}^nf_k`$. The sequence $`\{e_n\}_{n=1}^{\mathrm{}}`$ is such that $`e_n_n\mathrm{𝟏}`$ and $`\tau (e_n)<\mathrm{}`$ for all $`n1`$. Let $`X:=\overline{\text{span}}\{x_n,n1\}`$ and for $`a`$, let $`aX:=\{ax,xX\}`$ and $`Xa:=\{xa,xX\}`$.
###### Lemma 2.8.
If for every $`n1`$, $`X`$ is not isomorphic to $`e_nX`$ then there exists a normalized block basic sequence $`\{y_k\}_{k=1}^{\mathrm{}}`$ of $`\{x_n\}_{n=1}^{\mathrm{}}`$ and a strictly increasing sequence of integers $`\{n_k\}_{k=1}^{\mathrm{}}`$ so that
$$y_k(e_{n_k}e_{n_{k1}})y_k<2^k,\text{for}k1.$$
Similarly, if for every $`n1`$, $`X`$ is not isomorphic to $`Xe_n`$ then there exists a normalized block basic sequence $`\{y_k\}_{k=1}^{\mathrm{}}`$ of $`\{x_n\}_{n=1}^{\mathrm{}}`$ and a strictly increasing sequence of integers $`\{n_k\}_{k=1}^{\mathrm{}}`$ so that
$$y_ky_k(e_{n_k}e_{n_{k1}})<2^k,\text{for}k1.$$
###### Proof.
Inductively, we will construct a sequence $`\{y_k\}_{k=1}^{\mathrm{}}`$ in the unit sphere of $`X`$, strictly increasing sequences of integers $`\{m_k\}_{k=1}^{\mathrm{}}`$ and $`\{n_k\}_{k=1}^{\mathrm{}}`$ such that:
* $`y_k\text{span}\{x_n,m_{k1}<nm_k\}`$ for all $`k1`$;
* $`e_{n_{k_1}}y_k<2^{(k+1)}`$ for all $`k1`$;
* $`y_ke_{n_k}y_k<2^{(k+1)}`$ for all $`k1`$.
Fix $`y_1`$ a finitely supported vector in $`S_X`$ and let $`m_11`$ so that $`y_1\text{span}\{x_n,nm_1\}`$. Since $`(\mathrm{𝟏}e_n)_n0`$, there exists $`n_1`$ such that $`y_1e_{n_1}y_1<2^1`$.
Assume that the construction is done for $`1,2,\mathrm{},(j1)`$. Let $`X_j=\overline{\text{span}}\{x_n,nm_{j1}\}`$. Since $`X_j`$ is not isomorphic to $`e_{n_{j1}}X_j`$, there exist $`y_jS_{X_j}`$ such that $`e_{n_{j1}}y_j<2^{(j+1)}`$. By perturbation, we can assume that $`y_j`$ is finitely supported. If we fix $`n_j>n_{j1}`$ so that $`y_je_{n_j}y_j<2^{(j+1)}`$ then $`y_j(e_{n_j}e_{n_{j1}})y_j<2^j`$ and the lemma follows. ∎
To prove the theorem, assumme first the there exist $`n_01`$ such that $`X`$ is isomorphic to $`e_{n_0}X`$. Since $`\tau (e_{n_0})<\mathrm{}`$, the sequence $`\{e_{n_0}x_n\}_{n=1}^{\mathrm{}}`$ satisfies the assumptions of Proposition 2.5. Since $`E(,\tau )`$ has type 2, either $`\{e_{n_0}x_n\}_{n=1}^{\mathrm{}}`$ is equivalent to $`\mathrm{}_2`$ or there exists a subsequence $`\{e_{n_0}x_{n_j}\}_{j=1}^{\mathrm{}}`$ which is equivalent to a sequence of disjoint elements of $`E`$ and by isomorphism, the theorem follows.
Assume now that for every $`n1`$, $`X`$ is not isomorphic to $`e_nX`$. By the above lemma, there exist a normalized block basic sequence $`\{y_k\}_{k=1}^{\mathrm{}}`$ and a strictly increasing sequence of integers $`\{n_k\}_{k=1}^{\mathrm{}}`$ so that for every $`k1`$,
(2.1)
$$y_k(e_{n_k}e_{n_{k1}})y_k<2^k.$$
Let $`Y:=\overline{\text{span}}\{(e_{n_k}e_{n_{k1}})y_k,k1\}`$. As above, if there exists $`m_0`$ such that $`Y`$ is isomorphic to $`Ye_{m_0}`$, then the conclusion follows. Otherwise, there exist a block basic sequence $`\{z_k\}_{k=1}^{\mathrm{}}`$ of $`\{(e_{n_k}e_{n_{k1}})y_k\}_{k=1}^{\mathrm{}}`$ and a strictly increasing sequence of integers $`\{m_k\}_{k=1}^{\mathrm{}}`$ such that for every $`k1`$,
(2.2)
$$z_kz_k(e_{m_k}e_{m_{k1}})<2^k.$$
We remark that since the sequence $`\{z_k\}_{k=1}^{\mathrm{}}`$ is a block basic sequence of $`\{(e_{n_k}e_{n_{k1}})y_k\}_{k=2}^{\mathrm{}}`$, there exists a sequence $`\{q_k\}_{k=1}^{\mathrm{}}`$ of mutually disjoint projections such that for every $`k1`$, $`z_k=q_kz_k`$. Therefore, the sequence $`\{z_k(e_{m_k}e_{m_{k1}})\}_{k=2}^{\mathrm{}}`$ is both right and left disjointly supported and hence is equivalent to a disjointly supported sequence in $`E`$. By (2.2), we conclude that $`\{z_k\}_{k=1}^{\mathrm{}}`$ has a subsequence that is equivalent to a disjointly supported sequence in $`E`$ (see for instance, , Theorem 9 p.46). Since $`\{z_k\}_{k=1}^{\mathrm{}}`$ is a block basic sequence of $`\{(e_{n_k}e_{n_{k1}})y_k\}_{k=2}^{\mathrm{}}`$, inequality (2.1) shows that the corresponding block of $`\{y_k\}_{k=1}^{\mathrm{}}`$ is equivalent to $`\{z_k\}_{k=1}^{\mathrm{}}`$. The proof of the theorem is complete. ∎
## 3. Subspaces of Lorentz spaces
In this section, we will specialize to the concrete case of Lorentz spaces. We begin by recalling some definitions and basic facts about Lorentz spaces.
For $`0<p<\mathrm{}`$, $`0<q<\mathrm{}`$, and $`I=[0,1]`$ or $`[0,\mathrm{})`$, the Lorentz function space $`L_{p,q}(I)`$ is the space of all (classes of) Lebesgue measurable functions $`f`$ on $`I`$ for which $`f_{p,q}<\mathrm{}`$, where
(3.1)
$$\begin{array}{cc}\hfill f_{p,q}& =\left(_I\mu _t^q(f)d(t^{q/p})\right)^{1/q},q<\mathrm{},\hfill \\ & =\underset{tI}{sup}t^{1/p}\mu _t(f),q=\mathrm{}.\hfill \end{array}$$
Clearly, $`L_{p,p}(I)=L_p(I)`$ for any $`p>0`$. It is well known that for $`1qp<\mathrm{}`$, (3.1) defines a norm under which $`L_{p,q}(I)`$ is a separable rearrangement invariant Banach function space; otherwise, (3.1) defines a quasi-norm on $`L_{p,q}(I)`$ ( which is known to be equivalent to a norm if $`1<p<q<\mathrm{}`$). The following lemma was observed in . It contains the technical ingredients for the construction of the non-commutative counterparts.
###### Lemma 3.1.
Let $`0<p<\mathrm{}`$, $`0<q<\mathrm{}`$.
* If $`q<p`$, then $`L_{p,q}(I)`$ is $`q`$-convex with constant $`1`$ and satisfies a lower $`p`$-estimate with constant $`1`$;
* $`L_{p,q}(I)`$ satisfies an upper $`r`$-estimate and lower $`s`$-estimate (with some constant $`C`$) where $`r=\text{min}(p,q)`$ and $`s=\text{max}(p,q)`$.
For $`0<p<q<\mathrm{}`$, $`L_{p,q}(I)`$ can be equivalently renormed to be a quasi-Banach lattice that is $`\gamma `$-convex (for $`\gamma <p`$) with constant $`1`$ and satisfies a lower $`q`$-estimate of constant $`1`$. Hence for any $`0<p,q<\mathrm{}`$, we can define the non-commutative space $`L_{p,q}(,\tau )`$ as in Section 2. Since we are only interested in isomorphic properties, we will use the quasi-norm defined in (3.1). All results from sectio 2 apply to $`L_{p,q}(,\tau )`$ with appropriate values of $`p`$ and $`q`$.
For any given $`0<p<\mathrm{}`$ and $`0<q\mathrm{}`$, it is well known that the space $`L_{p,q}(I)`$ is equal (up to an equivalent quasi-norm) to the spaces $`(L_{p_1}(I),L_{p_2}(I))_{\theta ,q}`$ constructed using the real interpolation method where $`0<p_1<p_2<\mathrm{}`$, $`0<\theta <1`$ and $`1/p=(1\theta )/p_1+\theta /p_2`$. This was extended to the non-commutative setting by Xu ().
###### Lemma 3.2.
For $`0<p_1,p_2,q<\mathrm{}`$ and $`0<\theta <1`$ then
$$(L_{p_1}(,\tau ),L_{p_2}(,\tau ))_{\theta ,q}=L_{p,q}(,\tau )$$
(with equivalent quasi-norms) where $`1/p=(1\theta )/p_1+\theta /p_2`$.
The following result can be found in (Lemma 2.4 and the remark after Theorem 2.5).
###### Lemma 3.3.
Let $`0<p<\mathrm{}`$ and $`0<q<\mathrm{}`$. Let $`\{f_n\}_{n=1}^{\mathrm{}}`$ be a normalized disjointly supported sequence in $`L_{p,q}[0,\mathrm{})`$. Then $`\overline{\text{span}}\{f_n,n1\}`$ contains a copy of $`\mathrm{}_q`$.
A non-commutative extension follows directely from Proposition 2.3.
###### Proposition 3.4.
Let $`0<p<\mathrm{}`$ and $`0<q<\mathrm{}`$. Let $`\{x_n\}_{n=1}^{\mathrm{}}`$ be a normalized basic sequence in $`L_{p,q}(,\tau )`$. If $`\{x_n\}_{n=1}^{\mathrm{}}`$ is both right and left disjointly supported then $`\overline{\text{span}}\{x_n,n1\}`$ contains a copy of $`\mathrm{}_q`$.
The next result is an analogue of Theorem 2.5 of and is a direct application of Proposition 2.4 and Proposition 3.4
###### Theorem 3.5.
Let $`0<p<\mathrm{}`$, $`0<q<\mathrm{}`$, and let $`X`$ be a subspace of $`L_{p,q}(,\tau )`$. Then either $`X`$ is strongly embedded into $`L_{p,q}(,\tau )`$ or $`X`$ contains a copy of $`\mathrm{}_q`$.
For the next result, we need to fix some notation. Let $`𝒩`$ be a von Neumann algebra on a given Hilbert space $`H`$ with semi-finite trace $`\phi `$. Define $`[𝒩]`$ to be the von Neumann algebra over $`\mathrm{}_2(H)`$ as follows:
$$[𝒩]:=\{(a_{ij})_{ij};i,j,a_{ij}𝒩,(a_{ij})_{ij}_{B(\mathrm{}^2(H))}<\mathrm{}\}.$$
equipped with the trace $`[\phi ]((a_{ij})_{ij})=_{i=1}^{\mathrm{}}\phi (a_{ii})`$. It is clear that $`([𝒩],[\phi ])`$ is a semi-finite von Neumann algebra acting on $`\mathrm{}_2(H)`$. Let $`\{y_k\}_{k=1}^{\mathrm{}}`$ be a sequence in $`𝒩`$. For each $`k1`$, we define $`[y_k]=([y_k]_{ij})_{ij}`$ by setting: $`[y_k]_{1,k}=y_k`$ and $`[y_k]_{ij}=0`$ for $`(i,j)(1,k)`$ i.e for $`k1`$,
$$[y_k]:=\left(\begin{array}{cccccc}0& \mathrm{}& 0& y_k& 0& \mathrm{}\\ 0& \mathrm{}& 0& 0& 0& \mathrm{}\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\end{array}\right).$$
This amounts to placing the sequence $`\{y_k\}_{k=1}^{\mathrm{}}`$ in the first row of an infinite matrix.
###### Lemma 3.6.
Let $`0<p<2`$ and $`\{y_k\}_{k=1}^{\mathrm{}}`$ be a sequence in $`L_{p,q}(𝒩,\phi )`$. There exists an absolute constant $`C`$ such that for every choice of scalars $`\{a_k\}_{k=1}^{\mathrm{}}`$,
(3.2)
$$_0^1\underset{k=1}{\overset{n}{}}r_k(t)a_ky_k_{L_{p,q}(𝒩,\phi )}^2𝑑tC\text{min}\{\underset{k=1}{\overset{n}{}}a_k[y_k]_{L_{p,q}([𝒩],[\phi ])}^2,\underset{k=1}{\overset{n}{}}\overline{a}_k[y_k^{}]_{L_{p,q}([𝒩],[\phi ])}^2\}$$
for every $`n1`$.
###### Proof.
We claim first that for $`0<p<2`$, there exists an absolute constant $`A`$ such that:
(3.3)
$$\left(_0^1\underset{k=1}{\overset{n}{}}r_k(t)a_ky_k_{L_p(𝒩,\phi )}^2𝑑t\right)^{\frac{1}{2}}\sqrt{A}(\underset{k=1}{\overset{n}{}}|a_k|^2y_ky_k^{})^{\frac{1}{2}}_{L_p(𝒩,\phi )}.$$
Indeed, for $`1p<2`$, (3.3) follows from non-commutative Khintchine inequality (see Corollary III-4 and Remark III-6). For $`0<p<1`$, we remark that for any given finite sequence $`\{x_k\}_{k=1}^n`$ in $`L_p(𝒩,\phi )`$,
$$\underset{k=1}{\overset{n}{}}x_k_{L_p(𝒩,\phi )}^2A(\underset{k=1}{\overset{n}{}}x_kx_k^{})^{\frac{1}{2}}_{L_p(𝒩,\phi )}$$
for some fixed constant $`A`$. We observe that
$$\begin{array}{cc}\hfill |\underset{k=1}{\overset{n}{}}r_k(t)[x_k]^{}|^2& =\left(\underset{k=1}{\overset{n}{}}r_k(t)[y_k]\right).\left(\underset{k=1}{\overset{n}{}}r_k(t)[x_k]^{}\right)\hfill \\ & =\begin{array}{cc}\left(\begin{array}{ccccc}r_1(t)x_1& r_2(t)x_2& \mathrm{}& r_n(t)x_n& \mathrm{}\\ 0& 0& \mathrm{}& 0& \mathrm{}\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\end{array}\right)& \left(\begin{array}{ccc}r_1(t)x_1^{}& 0& \mathrm{}\\ r_2(t)x_2^{}& 0& \mathrm{}\\ \mathrm{}& \mathrm{}& \mathrm{}\end{array}\right)\end{array}\hfill \\ & =\left(\begin{array}{ccc}_{k=1}^nx_kx_k^{}& 0& \mathrm{}\\ 0& 0& \mathrm{}\\ \mathrm{}& \mathrm{}& \mathrm{}\end{array}\right)\hfill \end{array}$$
so
$$|\underset{k=1}{\overset{n}{}}r_k(t)[x_k]^{}|^p=\left(\begin{array}{ccc}\left(_{k=1}^nx_kx_k^{}\right)^{\frac{p}{2}}& 0& \mathrm{}\\ 0& 0& \mathrm{}& \\ \mathrm{}& \mathrm{}& \mathrm{}\end{array}\right)$$
which implies that
$$(\underset{k=1}{\overset{n}{}}x_k^{}x_k)^{\frac{1}{2}}_{L_p(𝒩,\phi )}^2=_0^1\underset{k=1}{\overset{n}{}}r_k(t)[x_k]^{}_{L_p([𝒩],[\phi ])}^2𝑑t$$
and since $`L_p([𝒩],[\phi ])`$ is of cotype $`2`$(), there exists a constant $`A>0`$ so that,
$$\begin{array}{cc}\hfill A(\underset{k=1}{\overset{n}{}}x_kx_k^{})^{\frac{1}{2}}_{L_p(𝒩,\phi )}^2& \underset{i=1}{\overset{n}{}}[x_k]^{}_{L_p([𝒩],[\phi ])}^2\hfill \\ & =\underset{k=1}{\overset{n}{}}x_k_{L_p(𝒩,\phi )}^2.\hfill \end{array}$$
For each $`\theta \{1,1\}^n`$, set $`s_\theta =\underset{k=1}{\overset{n}{}}\theta _ka_ky_k`$. Applying the above inequality to the finite sequence $`\{s_\theta \}_\theta `$, we get that
$$\begin{array}{cc}\hfill _0^1\underset{k=1}{\overset{n}{}}r_k(t)a_ky_k_{L_p(𝒩,\phi )}^2𝑑t& =2^n\underset{\theta \{1,1\}^n}{}s_\theta _{L_p(𝒩,\phi )}^2\hfill \\ & 2^nA(\underset{\theta \{1,1\}^n}{}s_\theta s_\theta ^{})^{\frac{1}{2}}_{L_p(𝒩,\phi )}^2\hfill \\ & =A2^n(\underset{\theta \{1,1\}^n}{}\underset{i,k}{}\theta _i\theta _k\overline{a_i}a_ky_ky_i^{})^{\frac{1}{2}}_{L_p(𝒩,\phi )}^2\hfill \\ & =A(_0^1\underset{i,k}{}r_k(t)r_i(t)\overline{a_i}a_ky_ky_i^{})^{\frac{1}{2}}_{L_p(𝒩,\phi )}\hfill \\ & =A(\underset{k=1}{\overset{n}{}}|a_k|^2y_ky_k^{})^{\frac{1}{2}}_{L_p(𝒩,\phi )}^2\hfill \end{array}$$
and therefore (3.3) is verified.
For the proof of the lemma, we note as above that
$$|\underset{k=1}{\overset{n}{}}\overline{a}_k[y_k]^{}|^p=\left(\begin{array}{ccc}\left(_{k=1}^n|a_k|^2y_ky_k^{}\right)^{\frac{p}{2}}& 0& \mathrm{}\\ 0& 0& \mathrm{}& \\ \mathrm{}& \mathrm{}& \mathrm{}\end{array}\right)$$
therefore
$$\underset{k=1}{\overset{n}{}}\overline{a}_k[y_k]^{}_{L_p([𝒩],[\phi ])}=(\underset{k=1}{\overset{n}{}}|a_k|^2y_ky_k^{})^{\frac{1}{2}}_{L_p(𝒩,\phi )},$$
hence
$$\underset{k=1}{\overset{n}{}}a_k[y_k]_{L_p([𝒩],[\phi ])}=(\underset{k=1}{\overset{n}{}}|a_k|^2y_ky_k^{})^{\frac{1}{2}}_{L_p(𝒩,\phi )}$$
and by (3.3),
$$_0^1\underset{k=1}{\overset{n}{}}r_k(t)a_ky_k_{L_p(𝒩,\phi )}^2𝑑tA\underset{k=1}{\overset{n}{}}a_k[y_k]_{L_p([𝒩],[\phi ])}^2.$$
###### Sublemma 3.7.
For every $`0<p<1`$, the map $`(a_{ij})_{ij}\underset{k}{}r_k()a_{1k}`$ is bounded as a linear map from $`L_p([𝒩],[\phi ])`$ into $`L_2([0,1],L_p(𝒩,\phi ))`$.
Let $`a=(a_{ij})_{ij}`$ be an element of $`L_p([𝒩],[\phi ])`$ and consider $`|a^{}|^2=(b_{ij})_{ij}`$. Clearly, $`b_{11}=_{k=1}^{\mathrm{}}a_{1k}a_{1k}^{}`$. Let $`e`$ be the projection in $`[𝒩]`$ defined by $`e=(e_{ij})_{ij}`$ with $`e_{11}=\mathrm{𝟏}`$ and $`e_{ij}=0`$ for $`(i,j)(1,1)`$. We have
$$e|a^{}|^2e=\left(\begin{array}{ccc}_{k=1}^{\mathrm{}}a_{1k}a_{1k}^{}& 0& \mathrm{}\\ 0& 0& \mathrm{}& \\ \mathrm{}& \mathrm{}& \mathrm{}\end{array}\right)$$
so $`e|a^{}|^2e_{L_{p/2}([𝒩],[\phi ])}=(_{k=1}^{\mathrm{}}a_{1k}a_{1k}^{})^{1/2}_{L_p(𝒩,\phi )}^2`$ and as above,
$$\begin{array}{cc}\hfill _0^1\underset{k=1}{\overset{\mathrm{}}{}}r_k(t)a_{1k}_{L_p(𝒩,\phi )}^2𝑑t& A\underset{k=1}{\overset{\mathrm{}}{}}[a_{1k}]_{L_p([𝒩],[\phi ])}^2\hfill \\ & =A(\underset{k=1}{\overset{\mathrm{}}{}}a_{1k}a_{1k}^{})^{1/2}_{L_p(𝒩,\phi )}^2\hfill \\ & =Ae|a^{}|^2e_{L_{p/2}([𝒩],[\phi ])}\hfill \\ & Aa_{L_p([𝒩],[\phi ])}^2.\hfill \end{array}$$
and the sublemma follows.
By interpolation, it is also a bounded map from $`L_{p,q}([𝒩],[\phi ])`$ into $`L_2([0,1],L_{p,q}(𝒩,\phi ))`$ which shows in particular that there exists an absolute constant $`C`$ such that
$$_0^1\underset{k=1}{\overset{n}{}}r_k(t)a_ky_k_{L_{p,q}(𝒩,\phi )}^2𝑑tC\underset{k=1}{\overset{n}{}}a_k[y_k]_{L_{p,q}([𝒩],[\phi ])}^2.$$
By taking adjoints, the other inequality follows. The proof of the lemma is complete. ∎
Our next result is a disjointification of sequences in $`L_{p,q}(,\tau )`$ and could be of independent interest.
###### Proposition 3.8.
Let $`(,\tau )`$ be a semi-finite von Neumann algebra. There exists a semi-finite von Neumann algebra $`𝒮`$ equipped with a faithful normal semi-finite trace $`\omega `$ with the following properties:
* $``$ is a von Neumann subalgebra of $`𝒮`$;
* $`\tau `$ is the restriction of $`\omega `$ on $``$;
* For $`0<p<2`$ and $`0<q<\mathrm{}`$, there exists a constant $`K`$ such that for any given basic sequence $`\{x_n\}_{n=1}^{\mathrm{}}`$ in $`L_{p,q}(,\tau )`$, there exists a left and right disjointly supported sequence $`\{s_n\}_{n=1}^{\mathrm{}}`$ in $`L_{p,q}(𝒮,\omega )`$ such that for any choice of scalars $`\{a_k\}_{k=1}^{\mathrm{}}`$ and $`n1`$,
$$_0^1\underset{k=1}{\overset{n}{}}a_kr_k(t)x_k_{L_{p,q}(,\tau )}^2𝑑tK\underset{k=1}{\overset{n}{}}a_ks_k_{L_{p,q}(𝒮,\omega )}^2.$$
###### Proof.
Using the notation above, let $`𝒩=[]`$, $`\phi =[\tau ]`$. Clearly, $`(𝒩,\phi )`$ is a semi-finite von Neumann algebra on the Hilbert space $`H=\mathrm{}_2()`$. Set $`𝒮=[𝒩]`$ and $`\omega =[\phi ]`$. As above, $``$ can be identified as a von Neumann subalgebra of $`𝒮`$ with $`\tau `$ being the restriction of $`\omega `$ on $``$.
Let $`\{x_n\}_{n=1}^{\mathrm{}}`$ be a basic sequence in $`L_{p,q}(,\tau )`$. Consider the sequence $`\{[x_n]\}_{n=1}^{\mathrm{}}`$ in $`𝒩=[]`$.
Claim: The sequence $`\{[x_n]\}_{n=1}^{\mathrm{}}`$ is right disjointly supported.
To verify this claim, recall that elements of $`𝒩`$ are infinite matrices with entries in $``$. For $`n1`$, let $`\pi _n=(a_{ij})_{ij}`$ with $`a_{n,n}=\mathrm{𝟏}`$ and $`a_{i,j}=0`$ for $`(i,j)(n,n)`$. Clearly $`\{\pi _n\}_{n=1}^{\mathrm{}}`$is a mutually disjoint sequence of projection in $`𝒩`$ and for each $`n1`$, $`[x_n]\pi _n=[x_n]`$.
For each $`n1`$, let $`z_n=[x_n]L_{p,q}(𝒩,\phi )`$ and consider the sequence $`\{s_n\}_{n=1}^{\mathrm{}}`$ in $`L_{p,q}(𝒮,\omega )`$ defined by $`s_n:=[z_n^{}]^{}`$.
Claim: The sequence $`\{s_n\}_{n=1}^{\mathrm{}}`$ is left and right disjointly supported.
First note that, as above, the sequence $`\{[z_n^{}]\}_{n=1}^{\mathrm{}}`$ is right disjointly supported so its adjoints $`\{s_n\}_{n=1}^{\mathrm{}}`$ is left disjointly supported. To prove that it is right disjointly supported, consider the following sequence $`\{e_n\}_{n=1}^{\mathrm{}}`$ in $`𝒮`$: $`e_n=(a_{ij}^{(n)})_{ij}`$ where $`a_{11}^{(n)}=\pi _n`$ and $`a_{ij}^{(n)}=0`$ for $`(i,j)(1,1)`$.
It is clear that the $`e_n`$’s are projections in $`𝒮`$ and since $`\{\pi _n\}_{n=1}^{\mathrm{}}`$ is mutually disjoint in $`𝒩`$, $`\{e_n\}_{n=1}^{\mathrm{}}`$ is mutually disjoint and one can see that for every $`n1`$, $`s_ne_n=s_n`$.
To complete the proof, we use Lemma 3.6,
$$\begin{array}{cc}\hfill _0^1\underset{k=1}{\overset{n}{}}r_k(t)a_kx_k_{L_{p,q}(,\tau )}^2𝑑t& C\underset{k=1}{\overset{n}{}}a_k[x_k]_{L_{p,q}(𝒩,\phi )}^2\hfill \\ & =C_0^1\underset{k=1}{\overset{n}{}}r_k(t)a_k[x_k]_{L_{p,q}(𝒩,\phi )}^2𝑑t\hfill \\ & =C_0^1\underset{k=1}{\overset{n}{}}r_k(t)a_kz_k_{L_{p,q}(𝒩,\phi )}^2𝑑t.\hfill \end{array}$$
Applying Lemma 3.6 on the von Neumann algebra $`𝒩`$,
$$\begin{array}{cc}\hfill _0^1\underset{k=1}{\overset{n}{}}r_k(t)a_kx_k_{L_{p,q}(,\tau )}^2𝑑t& C^2\underset{k=1}{\overset{n}{}}\overline{a_k}[z_k^{}]_{L_{p,q}(𝒮,\omega )}^2\hfill \\ & =C^2\underset{k=1}{\overset{n}{}}a_k[z_k^{}]^{}_{L_{p,q}(𝒮,\omega )}^2\hfill \\ & =C^2\underset{k=1}{\overset{n}{}}a_ks_k_{L_{p,q}(𝒮,\omega )}^2.\hfill \end{array}$$
The proof is complete ∎
We may now state the main result of this paper.
###### Theorem 3.9.
Let $`0<p<\mathrm{}`$, $`0<q<\mathrm{}`$, $`pq`$ and $`p2`$. Then $`\mathrm{}_p`$ does not embed into $`L_{p,q}(,\tau )`$. In particular, the Lorentz-Schatten ideals $`S_{p,q}`$ does not contain $`\mathrm{}_p`$.
The proof will be divided into several cases. First, notice that since $`pq`$, Proposition 3.5 shows that every subspace of $`L_{p,q}(,\tau )`$ equivalent to $`\mathrm{}_p`$ (and therefore not containing any copy of $`\mathrm{}_q`$) is strongly embedded into $`L_{p,q}(,\tau )`$. Fix $`r>q`$ then $`_{p,r}C_{p,q}`$, where $`C`$ is a constant depending only on $`p`$, $`q`$ and $`r`$ (see for instance , Proposition 4.2 p.217). In particular, there exists a continuous inclusion from $`L_{p,q}(,\tau )`$ into $`L_{p,r}(,\tau )`$ and if $`X`$ is a strongly embedded subspace of $`L_{p,q}(,\tau )`$ then $`X`$ is isomorphic to a subspace of $`L_{p,r}(,\tau )`$ so without loss of generality, we can assume that $`p<q`$ and $`1<q`$.
Case $`0<p<q<\mathrm{}`$ and $`p<2`$.
Assume that there exists a sequence $`\{x_n\}_{n=1}^{\mathrm{}}`$ that is $`M`$-equivalent to $`\mathrm{}_p`$ in $`L_{p,q}(,\tau )`$ and consider the disjoint sequence $`\{y_n\}_{n=1}^{\mathrm{}}`$ in $`L_{p,q}(𝒮,\omega )`$ as in Proposition 3.8. For every finite sequence of scalars $`\{a_n\}`$, we have:
$$\begin{array}{cc}\hfill \left(\underset{n}{}|a_n|^p\right)^{\frac{1}{p}}& M\left(_0^1\underset{n}{}r_n(t)a_nx_n_{L_{p,q}(,\tau )}^2𝑑t\right)^{\frac{1}{2}}\hfill \\ & M.\sqrt{K}\underset{n}{}a_ny_n_{L_{p,q}(𝒮,\omega )}\hfill \\ & N.M.\sqrt{K}\underset{n}{}a_n\phi _n_{L_{p,q}(0,\mathrm{})}\hfill \end{array}$$
where $`\{\phi _k\}_{k=1}^{\mathrm{}}`$ is a disjoint sequence in $`L_{p,q}[0,\mathrm{})`$ and $`N>0`$. Since $`p<q`$, the space $`L_{p,q}[0,\mathrm{})`$ satisfies an upper $`p`$-estimate hence there exists constants $`C_1`$ and $`C_2`$ such that
$$C_1\left(\underset{k=1}{\overset{n}{}}|a_k|^p\right)^{\frac{1}{p}}\underset{k=1}{\overset{n}{}}a_k\phi _k_{L_{p,q}[0,\tau (\mathrm{𝟏}))}C_2\left(\underset{k=1}{\overset{n}{}}|a_k|^p\right)^{\frac{1}{p}}.$$
But this is a contradiction since $`\overline{\text{span}}\{\phi _k,k1\}`$ contains a copy of $`\mathrm{}_q`$. ∎
Case $`2<p<q<\mathrm{}`$.
We remark that $`L_{p,q}(,\tau )`$ is of type 2 (see for instance, , Proposition 2g.22 p.230). Assume that there exists a sequence $`\{x_n\}_{n=1}^{\mathrm{}}`$ in $`L_{p,q}(,\tau )`$ that is equivalent to $`\mathrm{}_p`$. Since $`p2`$, Theorem 2.7 and Proposition 2.3 imply that $`\{x_n\}_{n=1}^{\mathrm{}}`$ contains a block basic sequence $`\{y_n\}_{n=1}^{\mathrm{}}`$ that is equivalent to a disjointly supported normalized sequence in $`L_{p,q}[0,\tau (\mathrm{𝟏}))`$ so $`\overline{\text{span}}\{y_n,n1\}`$ does not contain $`\mathrm{}_p`$. Contradiction.
The following application characterizes strongly embedded subspaces in $`L_p(,\tau )`$. It generalizes results of Rosenthal and Kalton on $`L_p[0,1]`$.
###### Corollary 3.10.
Let $`0<p<\mathrm{}`$, $`p2`$ and $`X`$ be a subspace of $`L_p(,\tau )`$. Then the following are equivalent:
* $`X`$ contains $`\mathrm{}_p`$;
* $`X`$ is not strongly embedded into $`L_p(,\tau )`$.
###### Proof.
Let $`X`$ be a subspace of $`L_p(,\tau )`$ and assume that $`X`$ contains $`\mathrm{}_p`$. Since for $`p<q`$, as above, $`_{p,q}C_p`$, for some constant $`C`$ (see Proposition 4.2 p.217). There exists an inclusion map from $`L_p(,\tau )`$ into $`L_{p,q}(,\tau )`$. If $`X`$ is strongly embedded into $`L_p(,\tau )`$, then $`X`$ is isomorphic to a subspace of $`L_{p,q}(,\tau )`$. In particular $`\mathrm{}_p`$ embeds into $`L_{p,q}(,\tau )`$. Contradiction.
The converse is a direct consequence of Theorem 3.5. ∎
The next result is known for copies of $`\mathrm{}_1`$ in preduals of von Neumann algebras.
###### Corollary 3.11.
Let $`1p<\mathrm{}`$, $`p2`$. If $`\{x_n\}_{n=1}^{\mathrm{}}`$ is a sequence in $`L_p(,\tau )`$ that is equivalent to $`\mathrm{}_p`$ and $`\{\epsilon _n\}_{n=1}^{\mathrm{}}`$ is a sequence in the interval $`(0,1)`$ with $`\epsilon _n_n0`$, then there exists a block basis $`\{y_n\}_{n=1}^{\mathrm{}}`$ of $`\{x_n\}_{n=1}^{\mathrm{}}`$ such that:
$$\left(\underset{n}{}|a_n|^p\right)^{\frac{1}{p}}\left(\underset{n}{}|a_n|^p\epsilon _{n}^{}{}_{}{}^{p}\right)^{\frac{1}{p}}\underset{n}{}a_ny_n\left(\underset{n}{}|a_n|^p\right)^{\frac{1}{p}}+\left(\underset{n}{}|a_n|^p\epsilon _{n}^{}{}_{}{}^{p}\right)^{\frac{1}{p}}$$
for all finite sequence $`(a_n)_n`$ of scalars. In particular, for every $`k1`$, the sequence $`\{y_n\}_{n=k}^{\mathrm{}}`$ is $`(1+\epsilon _k)`$-equivalent to $`\mathrm{}_p`$.
###### Proof.
Since $`\mathrm{}_p`$ is not strongly embedded into $`L_p(,\tau )`$, Proposition 2.4 implies the existance of a block basic sequence $`\{z_n\}_{n=1}^{\mathrm{}}`$ of $`\{x_n\}_{n=1}^{\mathrm{}}`$ and a sequence $`\{p_n\}_{n=1}^{\mathrm{}}`$ of mutually disjoint projections in $``$ such that:
$$\underset{n\mathrm{}}{lim}z_np_nz_np_n=0.$$
Note that $`lim\; inf_n\mathrm{}p_nz_np_n>0`$. By taking a subsequence (if necessary), we will assume that for every $`n1`$,
$$\frac{z_np_nz_np_n}{p_nz_np_n}\epsilon _n2^n.$$
For $`n1`$, set $`y_n:=z_n/p_nz_np_n`$. If $`(a_n)_n`$ is a finite sequence of scalars then
$$\begin{array}{cc}\hfill \underset{n}{}a_ny_n& \underset{n}{}|a_n|y_np_nz_np_n+\left(\underset{n}{}|a_n|^p\right)^{1/p}\hfill \\ & \left(\underset{n}{}|a_n|^p\epsilon _n^p\right)^{1/p}\left(\underset{n}{}2^{nq}\right)^{1/q}+\left(\underset{n}{}|a_n|^p\right)^{1/p}\hfill \end{array}$$
where $`1/p+1/q=1`$. This shows that
$$\underset{n}{}a_ny_n\left(\underset{n}{}|a_n|^p\right)^{1/p}+\left(\underset{n}{}|a_n|^p\epsilon _n^p\right)^{1/p}.$$
The other inequality can be obtain with similar estimate. ∎
Acknowledgement. The author wishes to thank F. Lust-Piquard and G. Pisier for insightful discussions conserning the non-commutative Khintchine’s inequalities.
|
warning/0004/hep-ex0004021.html
|
ar5iv
|
text
|
# DETERMINATION OF THE QCD COUPLING 𝛼ₛ
## 1 Preface
The coupling strength $`\alpha _\mathrm{s}`$ is the basic free parameter of Quantum Chromodynamics (QCD), the theory of the Strong Interaction which is one of the four fundamental forces of nature. QCD describes the interaction of quarks through the exchange of an octet of massless vector gauge bosons, the gluons, using similar concepts as known from Quantum Electrodynamics, QED. QCD, however, is more complex than QED because quarks and gluons, the analogues to electrons and photons in QED, are not observed as free particles but are confined inside hadrons.
Confinement implies that the coupling strength $`\alpha _\mathrm{s}`$, the analogue to the fine structure constant $`\alpha `$ in QED, becomes large in the regime of large-distance or low-momentum transfer interactions<sup>1</sup><sup>1</sup>1 In the world of quantum physics, “large” distances $`\mathrm{\Delta }s`$ correspond to $`\mathrm{\Delta }s>`$ 1 fm, “low” momentum transfers to $`Q<`$ 1 GeV/c…. Conversely, quarks and gluons are probed to behave like free particles, for short time intervals<sup>2</sup><sup>2</sup>2 … and “short” time intervals correspond to $`\mathrm{\Delta }t<10^{24}`$ s., in high-energy or short-distance reactions; they are said to be “asymptotically free”, i.e. $`\alpha _\mathrm{s}`$ 0 for momentum transfers $`Q\mathrm{}`$.
Within QCD, the phenomenology of confinement and of asymptotic freedom is realized by introducing a new quantum number, called “colour charge”. Quarks carry one out of three different colour charges, while hadrons are colourless bound states of 3 quarks or 3 antiquarks (“baryons”), or of a quark and an anti-quark (“mesons”). Gluons, in contrast to photons which do not carry (electrical) charge by themselves, have two colour charges. This concept leads to the process of gluon self-interaction, which in turn, through the effect of gluon vacuum polarization, gives rise to asymptotic freedom, i.e. the decrease of $`\alpha _\mathrm{s}`$ with increasing momentum transfer.
As in the case of QED, QCD predicts the energy dependence of $`\alpha _\mathrm{s}`$, while the actual value of $`\alpha _\mathrm{s}`$, at a given energy or four momentum transfer scale<sup>3</sup><sup>3</sup>3Here and in the following, a system of units is utilized where the speed of light and Planck’s constant are put to unity, $`c=\mathrm{}=1`$, such that energies, momenta and masses are all given in units of GeV. $`Q`$, is not predicted but must be determined from experiment.
Determining $`\alpha _\mathrm{s}`$ at a specific energy scale $`Q`$ is therefore a fundamental measurement, to be compared with measurements of the electromagnetic coupling $`\alpha `$, of the elementary electric charge, or of the gravitational constant. Testing QCD as such, however, requires the measurement of $`\alpha _\mathrm{s}`$ at least at two different energy scales, and/or at different processes: one measurement fixes the free parameter and thus provides accurate predictions for the value of $`\alpha _\mathrm{s}`$ at other energy scales and/or at other processes.
In general, $`\alpha _\mathrm{s}`$ can be determined in dynamic particle reactions involving in- or outgoing quarks and gluons, which manifest themselves as hadrons. Examples of Feynman diagrams describing hadronic final states in deep inelastic lepton-nucleon scattering (DIS), electron-positron annihilation ($`\mathrm{e}^+\mathrm{e}^{}`$), hadron collisions and quarkonia decays are shown in figure 1.
In this report, at the turn of the millennium, the current status of measurements of $`\alpha _\mathrm{s}`$ is reviewed. Theoretical basics of $`\alpha _\mathrm{s}`$ and QCD are given in Section 2. Measurements of $`\alpha _\mathrm{s}`$ from deep inelastic scattering, from $`\mathrm{e}^+\mathrm{e}^{}`$ annihilation processes, from hadron colliders and from heavy quarkonia decays are discussed in Sections 3 to 6, respectively. A global summary of these results, a determination of the world average value of $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})`$ and quantitative studies of the energy dependence of $`\alpha _\mathrm{s}`$ are presented in Section 7. Section 8 concludes and gives an outlook to future developments.
## 2 QCD and $`\alpha _\mathrm{s}`$: basic theoretical predictions
The concepts of QCD are described in many text books and review articles, see e.g. references . A brief review of the basics of perturbative QCD and of the coupling strength $`\alpha _\mathrm{s}`$, like the concepts of renormalization, asymptotic freedom and confinement, the $`\mathrm{\Lambda }`$ parameter, the treatment of quark masses and thresholds, perturbative predictions of physical observables and renormalization scale dependence, and of nonperturbative methods like lattice calculations will be presented in the following subsections.
### 2.1 Renormalization
In quantum field theories like QCD and QED, dimensionless physical quantities $``$ can be expressed by a perturbation series in powers of the coupling parameter $`\alpha _\mathrm{s}`$ or $`\alpha `$, respectively. Consider $``$ depending on $`\alpha _\mathrm{s}`$ and on a single energy scale $`Q`$. This scale shall be larger than any other relevant, dimensionful parameter such as quark masses. In the following, these masses are therefore set to zero.
When calculating $``$ as a perturbation series in $`\alpha _\mathrm{s}`$, ultraviolet divergencies occur. Because $``$ must retain physical values, these divergencies are removed by a procedure called “renormalization”. This introduces a second mass or energy scale, $`\mu `$, which represents the point at which the subtraction to remove the ultraviolet divergencies is actually performed. As a consequence of this procedure, $``$ and $`\alpha _\mathrm{s}`$ become functions of the renormalization scale $`\mu `$. Since $``$ is dimensionless, we assume that it only depends on the ratio $`Q^2/\mu ^2`$ and on the renormalized coupling $`\alpha _\mathrm{s}(\mu ^2)`$:
$$(Q^2/\mu ^2,\alpha _\mathrm{s});\alpha _\mathrm{s}\alpha _\mathrm{s}(\mu ^2).$$
Because the choice of $`\mu `$ is arbitrary, however, $``$ cannot depend on $`\mu `$, for a fixed value of the coupling, such that
$$\mu ^2\frac{\mathrm{d}}{\mathrm{d}\mu ^2}(Q^2/\mu ^2,\alpha _\mathrm{s})=\left(\mu ^2\frac{}{\mu ^2}+\mu ^2\frac{\alpha _\mathrm{s}}{\mu ^2}\frac{}{\alpha _\mathrm{s}}\right)=^!0,$$
(1)
where the convention of multiplying the whole equation with $`\mu ^2`$ is applied in order to keep the expression dimensionless. Equation 1 implies that any explicit dependence of $``$ on $`\mu `$ must be cancelled by an appropriate $`\mu `$-dependence of $`\alpha _\mathrm{s}`$. It would therefore be natural to identify the renormalization scale with the physical energy scale of the process, $`\mu ^2=Q^2`$, eliminating the uncomfortable presence of a second and unspecified scale. In this case, $`\alpha _\mathrm{s}`$ transforms to the “running coupling constant” $`\alpha _\mathrm{s}(Q^2)`$, and the energy dependence of $``$ enters only through the energy dependence of $`\alpha _\mathrm{s}(Q^2)`$.
### 2.2 $`\alpha _\mathrm{s}`$ and its energy dependence
While QCD does not predict the actual size of $`\alpha _\mathrm{s}`$ at a particular energy scale, its energy dependence is precisely determined. If the renormalized coupling $`\alpha _\mathrm{s}(\mu ^2)`$ can be fixed (i.e. measured) at a given scale $`\mu ^2`$, QCD definitely predicts the size of $`\alpha _\mathrm{s}`$ at any other energy scale $`Q^2`$ through the renormalization group equation
$$Q^2\frac{\alpha _\mathrm{s}(Q^2)}{Q^2}=\beta \left(\alpha _\mathrm{s}(Q^2)\right).$$
(2)
The perturbative expansion of the $`\beta `$ function, including higher order loop corrections to the bare vertices of the theory, is calculated to complete 4-loop approximation :
$$\beta (\alpha _\mathrm{s}(Q^2))=\beta _0\alpha _\mathrm{s}^2(Q^2)\beta _1\alpha _\mathrm{s}^3(Q^2)\beta _2\alpha _\mathrm{s}^4(Q^2)\beta _3\alpha _\mathrm{s}^5(Q^2)+𝒪(\alpha _\mathrm{s}^6),$$
(3)
where
$`\beta _0`$ $`=`$ $`{\displaystyle \frac{332N_f}{12\pi }},`$
$`\beta _1`$ $`=`$ $`{\displaystyle \frac{15319N_f}{24\pi ^2}},`$
$`\beta _2`$ $`=`$ $`{\displaystyle \frac{7713915099N_f+325N_f^2}{3456\pi ^3}},`$
$`\beta _3`$ $``$ $`{\displaystyle \frac{292436946.3N_f+405.089N_f^2+1.49931N_f^3}{256\pi ^4}},`$ (4)
and $`N_f`$ is the number of active quark flavours at the energy scale $`Q`$. The numerical constants in equation 2.2 are functions of the group constants $`C_A`$ and $`C_F`$, which for QCD — exhibiting $`SU(3)`$ symmetry — have values of $`C_A=3`$ and $`C_F=4/3`$ (see e.g. for more details). $`\beta _0`$ and $`\beta _1`$ are independent of the renormalization scheme, while all higher order $`\beta `$ coefficients are scheme dependent.
### 2.3 Asymptotic freedom and confinement
A solution of equation 3 in 1-loop approximation, i.e. neglecting $`\beta _1`$ and higher order terms, is
$$\alpha _\mathrm{s}(Q^2)=\frac{\alpha _\mathrm{s}(\mu ^2)}{1+\alpha _\mathrm{s}(\mu ^2)\beta _0\mathrm{ln}\frac{Q^2}{\mu ^2}}.$$
(5)
Apart from giving a relation between $`\alpha _\mathrm{s}(Q^2)`$ and $`\alpha _\mathrm{s}(\mu ^2)`$, equation 5 also demonstrates the property of asymptotic freedom: if $`Q^2`$ becomes large and $`\beta _0`$ is positive, i.e. if $`N_f<17`$, $`\alpha _\mathrm{s}(Q^2)`$ will decrease to zero.
Likewise, equation 5 indicates that $`\alpha _\mathrm{s}(Q^2)`$ grows to large values and actually diverges to infinity at small $`Q^2`$: for instance, with $`\alpha _\mathrm{s}(\mu ^2M_{\mathrm{Z}^0}^2)=0.12`$ and for typical values of $`N_f=2\mathrm{}5`$, $`\alpha _\mathrm{s}(Q^2)`$ exceeds unity for $`Q^2𝒪(100\mathrm{MeV}\mathrm{}1\mathrm{GeV})`$. Clearly, this is the region where perturbative expansions in $`\alpha _\mathrm{s}`$ are not meaningful anymore, and we may regard energy scales of $`\mu ^2`$ and $`Q^2`$ below the order of 1 GeV as the nonperturbative region where confinement sets in, and where equations 3 and 5 cannot be applied.
Including $`\beta _1`$ and higher order terms, similar but more complicated relations for $`\alpha _\mathrm{s}(Q^2)`$, as a function of $`\alpha _\mathrm{s}(\mu ^2)`$ and of $`\mathrm{ln}\frac{Q^2}{\mu ^2}`$ as in equation 5, emerge. They can be solved numerically, such that for a given value of $`\alpha _\mathrm{s}(\mu ^2)`$, choosing a suitable reference scale like the mass of the $`\mathrm{Z}^0`$ boson, $`\mu =M_{\mathrm{Z}^0}`$, $`\alpha _\mathrm{s}(Q^2)`$ can be accurately determined at any energy scale $`Q^21\mathrm{GeV}^2`$.
### 2.4 The $`\mathrm{\Lambda }`$ parameter
Alternatively, as another parametrization of $`\alpha _\mathrm{s}(Q^2)`$, a parameter called $`\mathrm{\Lambda }`$ is introduced as a constant of integration of the $`\beta `$-function, such that, in leading order,
$$\alpha _\mathrm{s}(Q^2)=\frac{1}{\beta _0\mathrm{ln}(Q^2/\mathrm{\Lambda }^2)}.$$
(6)
Equations 5 and 6 are equivalent with each other if
$$\mathrm{\Lambda }^2=\frac{\mu ^2}{e^{1/\left(\beta _0\alpha _\mathrm{s}(\mu ^2)\right)}}.$$
Hence, $`\alpha _\mathrm{s}(\mu ^2)`$ is replaced by a suitable choice of the $`\mathrm{\Lambda }`$ parameter, which technically is identical to the energy scale $`Q`$ where $`\alpha _\mathrm{s}(Q^2)`$ diverges to infinity, $`\alpha _\mathrm{s}(Q^2)\mathrm{}`$ for $`Q^2\mathrm{\Lambda }^2`$. To give a numerical example, $`\mathrm{\Lambda }0.1`$ GeV for $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0}91.2\mathrm{GeV})=0.12`$ and $`N_f`$ = 5.
The parametrization of the running coupling $`\alpha _\mathrm{s}(Q^2)`$ with $`\mathrm{\Lambda }`$ instead of $`\alpha _\mathrm{s}(\mu ^2)`$ has become a common standard, see e.g. reference , and will also be adapted in this review. While being a convenient and well-used choice, however, this parametrization has several shortcomings which the user should be aware of.
First, requiring that $`\alpha _\mathrm{s}(Q^2)`$ must be continuous when crossing a quark threshold<sup>4</sup><sup>4</sup>4Strictly speaking, physical observables $``$ rather than $`\alpha _\mathrm{s}`$ must be continuous, which may lead to small discontinuities in $`\alpha _\mathrm{s}(Q^2)`$ at quark thresholds in finite order perturbation theory; see section 2.5., $`\mathrm{\Lambda }`$ actually depends on the number of active quark flavours. Secondly, $`\mathrm{\Lambda }`$ depends on the renormalization scheme, see e.g. reference . In this review, the so-called “modified minimal subtraction scheme” ($`\overline{\text{MS}}`$) will be adopted, which also has become a common standard . $`\mathrm{\Lambda }`$ will therefore be labelled $`\mathrm{\Lambda }_{\overline{\text{MS}}}^{(N_f)}`$ to indicate these peculiarities.
In complete 4-loop approximation and using the $`\mathrm{\Lambda }`$-parametrization, the running coupling is thus given by
$`\alpha _\mathrm{s}(Q^2)`$ $`=`$ $`{\displaystyle \frac{1}{\beta _0L}}{\displaystyle \frac{1}{\beta _0^3L^2}}\beta _1\mathrm{ln}L`$
$`+`$ $`{\displaystyle \frac{1}{\beta _0^3L^3}}\left({\displaystyle \frac{\beta _1^2}{\beta _0^2}}\left(\mathrm{ln}^2L\mathrm{ln}L1\right)+{\displaystyle \frac{\beta _2}{\beta _0}}\right)`$
$`+`$ $`{\displaystyle \frac{1}{\beta _0^4L^4}}\left({\displaystyle \frac{\beta _1^3}{\beta _0^3}}\left(\mathrm{ln}^3L+{\displaystyle \frac{5}{2}}\mathrm{ln}^2L+2\mathrm{ln}L{\displaystyle \frac{1}{2}}\right)3{\displaystyle \frac{\beta _1\beta _2}{\beta _0^2}}\mathrm{ln}L+{\displaystyle \frac{\beta _3}{2\beta _0}}\right)`$
(7)
where $`L=Q^2/\mathrm{\Lambda }_{\overline{\text{MS}}}^2`$. The first line of equation 2.4 includes the 1- and the 2-loop coefficients, the second line is the 3-loop and the third line is the 4-loop correction, respectively.
The functional form of $`\alpha _\mathrm{s}(Q)`$, for the 1-, the 2- and the 3-loop approximation of equation 2.4, each with $`\mathrm{\Lambda }=0.220`$ GeV, is shown in figure 2(a). As can be seen, there is an almost 15% decrease of $`\alpha _\mathrm{s}`$ when changing from 1-loop to 2-loop approximation, for the same value of $`\mathrm{\Lambda }`$. The difference between the 2-loop and the 3-loop prediction is only about 1-2%, and is less than 0.01% between the 3-loop and the 4-loop presentation which cannot be resolved in the figure.
The fractional difference in the energy dependence of $`\alpha _\mathrm{s}`$, $`\frac{(\alpha _\mathrm{s}^{(4loop)}\alpha _\mathrm{s}^{(nloop)})}{\alpha _\mathrm{s}^{(4loop)}}`$, for $`n`$ = 1, 2 and 3, is presented in figure 2(b). Here, in contrast to figure 2(a), the values of $`\mathrm{\Lambda }_{\overline{\text{MS}}}`$ were chosen such that $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})=0.119`$ in each order, i.e., $`\mathrm{\Lambda }_{\overline{\text{MS}}}=93`$ MeV (1-loop), $`\mathrm{\Lambda }_{\overline{\text{MS}}}=239`$ MeV (2-loop), and $`\mathrm{\Lambda }_{\overline{\text{MS}}}=220`$ MeV (3- and 4-loop). Only the 1-loop approximation shows sizeable differences of up to several per cent, in the energy and parameter range chosen, while the 2- and 3-loop approximation already reproduce the energy dependence of the best, i.e. 4-loop, prediction quite accurately.
### 2.5 Quark masses and thresholds
So far in this discussion, finite quark masses $`m_q`$ were neglected, assuming that both the physical and the renormalization scales $`Q^2`$ and $`\mu ^2`$, respectively, are larger than any other relevant energy or mass scale involved in the problem. This is, however, not entirely correct, since there are several QCD studies and $`\alpha _\mathrm{s}`$ determinations at energy scales around the charm- and bottom-quark masses of about 1.5 and 4.7 GeV, respectively.
Finite quark masses may have two major effects on actual QCD studies: Firstly, quark masses will alter the perturbative predictions of observables $``$. While phase space effects which are introduced by massive quarks can often be studied using hadronization models and Monte Carlo simulation techniques, explicit quark mass corrections in higher than leading perturbative order are available only for jet production and for total hadronic cross sections in $`\mathrm{e}^+\mathrm{e}^{}`$ annihilation.
Secondly, any quark-mass dependence of $``$ will add another term $`\mu ^2\frac{m}{\mu ^2}\frac{}{m}`$ to equation 1, which leads to energy-dependent, running quark masses, $`m_q(Q^2)`$, in a similar way as the running coupling $`\alpha _\mathrm{s}(Q^2)`$ was obtained, see e.g. reference .
In addition to these effects, $`\alpha _\mathrm{s}`$ indirectly also depends on the quark masses, through the dependence of the $`\beta `$ coefficients on the effective number of quarks flavours, $`N_f`$, with $`m_q\mu `$. Constructing an effective theory for, say, ($`N_f`$-1) quark flavours which must be consistent with the $`N_f`$ quark flavours theory at the heavy quark threshold $`\mu ^{(N_f)}𝒪(m_q)`$, results in matching conditions for the $`\alpha _\mathrm{s}`$ values of the ($`N_f`$-1)- and the $`N_f`$-quark flavours theories .
In leading and in next-to-leading order, the matching condition is $`\alpha _\mathrm{s}^{(N_f1)}=\alpha _\mathrm{s}^{N_f}`$. In higher orders and the $`\overline{MS}`$ scheme, however, nontrivial matching conditions apply . Formally these are, if the energy evolution of $`\alpha _\mathrm{s}`$ is performed in $`n^{th}`$ order, of order ($`n1`$).
The matching scale $`\mu ^{(N_f)}`$ can be chosen in terms of the (running) $`\overline{MS}`$ mass $`m_q(\mu _q)`$, or of the constant, so-called pole mass $`M_q`$. For both cases, the relevant matching conditions in NNLO are given in . These expressions have a particluarly simple form for the choice<sup>5</sup><sup>5</sup>5The results of reference are also valid for other relations between $`\mu ^{(N_f)}`$ and $`m_q`$ or $`M_q`$, as e.g. $`\mu ^{(N_f)}=2M_q`$. For 3-loop matching, however, practical differences due to the freedom of this choice are negligible. $`\mu ^{(N_f)}=m_q(m_q)`$ or $`\mu ^{(N_f)}=M_q`$. In this report, the latter choice will be used to perform 3-loop matching at the heavy quark pole masses, in which case the matching condition reads, with $`a=\alpha _\mathrm{s}^{(N_f)}/\pi `$ and $`a^{}=\alpha _\mathrm{s}^{(N_f1)}/\pi `$:
$$\frac{a^{}}{a}=1+C_2a^2+C_3a^3,$$
(8)
where $`C_2=0.291667`$ and $`C_3=5.32389+(N_f1)0.26247`$ .
The 4-loop prediction for the running $`\alpha _\mathrm{s}`$, using equation 2.4 with $`\mathrm{\Lambda }_{\overline{\text{MS}}}^{(N_f=5)}`$ = 220 MeV and 3-loop matching at the charm- and bottom-quark pole masses, $`\mu _c^{(N_f=4)}=M_c=1.5`$ GeV and $`\mu _b^{(N_f=5)}=M_b=4.7`$ GeV, is illustrated in figure 3a (full line). Small discontinuities at the quark thresholds can be seen, such that $`\alpha _\mathrm{s}^{(N_f1)}<\alpha _\mathrm{s}^{(N_f)}`$ by about 2 per mille at the bottom- and about 1 per cent at the charm-quark threshold. The corresponding values of $`\mathrm{\Lambda }_{\overline{\text{MS}}}`$ are $`\mathrm{\Lambda }_{\overline{\text{MS}}}^{(N_f=4)}=305`$ MeV and $`\mathrm{\Lambda }_{\overline{\text{MS}}}^{(N_f=3)}=346`$ MeV . Comparison with the 4-loop prediction, without applying threshold matching and for $`\mathrm{\Lambda }_{\overline{\text{MS}}}^{(N_f=5)}`$ = 220 MeV and $`N_f`$ = 5 throughout (dashed line) demonstrates that, in spite of the discontinuities, the matched calculation shows a steeper rise towards smaller energies because of the larger values of $`\mathrm{\Lambda }_{\overline{\text{MS}}}^{(N_f=4)}`$ and $`\mathrm{\Lambda }_{\overline{\text{MS}}}^{(N_f=3)}`$.
The size of discontinuities and the changes of slopes are more clearly demonstrated in figure 3b, where the fractional difference between the two curves from figure 3a, i.e. between the matched and the unmatched calculation, is presented. Note that the step function of $`\alpha _\mathrm{s}`$ is not an effect which can be measured — the steps are artifacts of the truncated perturbation theory and the requirement that predictions for observables at energy scales around the matching point must be consistent and independent of the two possible choices of (neighbouring) values of $`N_f`$.
### 2.6 Perturbative predictions of physical quantities
In practice, $`\alpha _\mathrm{s}`$ is not an ‘observable’ by itself. Values of $`\alpha _\mathrm{s}(\mu ^2)`$ are determined from measurements of observables $``$ for which perturbative predictions exist. These are usually given by a power series in $`\alpha _\mathrm{s}(\mu ^2)`$, like
$`(Q^2)`$ $`=`$ $`P_l{\displaystyle \underset{n}{}}R_n\alpha _\mathrm{s}^n`$ (9)
$`=`$ $`P_l\left(R_0+R_1\alpha _\mathrm{s}(\mu ^2)+R_2(Q^2/\mu ^2)\alpha _\mathrm{s}^2(\mu ^2)+\mathrm{}\right),`$
where $`R_n`$ are the $`n_{th}`$ order coefficients of the perturbation series and $`P_lR_0`$ denotes the lowest-order value of $``$.
For processes which involve gluons already in lowest order perturbation theory, $`P_l`$ itself may include (powers of) $`\alpha _\mathrm{s}`$. For instance, this happens in case of the hadronic decay width of heavy Quarkonia, $`\mathrm{\Gamma }(\mathrm{{\rm Y}}ggghadrons)`$ for which $`P_l\alpha _\mathrm{s}^3`$. If no gluons are involved in lowest order, as e.g. in $`\mathrm{e}^+\mathrm{e}^{}q\overline{q}hadrons`$ or in deep inelastic scattering processes, $`P_lR_0`$ is a constant and the usual choice is $`P_l1`$. $`R_0`$ is called the lowest order coefficient and $`R_1`$ is the leading order (LO) coefficient. Following this naming convention, $`R_2`$ is the next-to-leading order (NLO) and $`R_3`$ is the next-to-next-to-leading order (NNLO) coefficient.
QCD calculations in NLO perturbation theory are available for many observables $``$ in high energy particle reactions; calculations including the complete NNLO are available for some totally inclusive quantities, like the total hadronic cross section in $`\mathrm{e}^+\mathrm{e}^{}hadrons`$, moments and sum rules of structure functions in deep inelastic scattering processes and the hadronic decay width of the $`\tau `$ lepton. The complicated nature of QCD, due to the process of gluon self-coupling and the resulting large number of Feynman diagrams in higher orders of perturbation theory, so far limited the number of QCD calculations in complete NNLO.
An alternative approach to calculating higher order corrections is based on the resummation of leading logarithms which arise from soft and collinear singularities in gluon emission . In such a case, the effective expansion parameter is $`\alpha _\mathrm{s}L^2`$ rather than $`\alpha _\mathrm{s}`$, where $`L=\mathrm{ln}(1/)`$ and $``$ is some generic observable which tends to zero in lowest order. For small values of $``$, $`\alpha _\mathrm{s}L^2`$ becomes large, and therefore these terms should be known to all orders in $`\alpha _\mathrm{s}`$ if a reliable prediction of $``$ is to be obtained. For certain observables it has proved possible to sum up both the leading and next-to-leading logarithms, which is referred to as the ‘Next-to-Leading Log Approximation’ or NLLA.
For observables $``$ for which resummation is possible, the $`cumulative`$ cross-section $`\mathrm{\Sigma }()`$ may be written as
$$\mathrm{\Sigma }()_0^{}\frac{1}{\sigma }\frac{d\sigma }{d}𝑑=C(\alpha _\mathrm{s})\mathrm{exp}\left[G(\alpha _\mathrm{s},L)\right]+D(\alpha _\mathrm{s},),$$
(10)
where $`D(\alpha _\mathrm{s},)`$ is a remainder function which should vanish as $`0`$, and
$`C(\alpha _\mathrm{s})`$ $`=`$ $`1+{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}C_n\widehat{\alpha }_s^n`$ (11)
$`G(\alpha _\mathrm{s},L)`$ $`=`$ $`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}{\displaystyle \underset{m=1}{\overset{n+1}{}}}G_{nm}\widehat{\alpha }_s^nL^m`$
$``$ $`Lg_1(\alpha _\mathrm{s}L)+g_2(\alpha _\mathrm{s}L)+\alpha _\mathrm{s}g_3(\alpha _\mathrm{s}L)+\alpha _\mathrm{s}^2g_4(\alpha _\mathrm{s}L)\mathrm{},`$
with $`\widehat{\alpha }_s(\alpha _\mathrm{s}/2\pi )`$. The functions $`Lg_1`$ and $`g_2`$ represent the sums of the leading and next-to-leading logarithms respectively, to all orders in $`\alpha _\mathrm{s}`$.
Resummation of leading and next-to-leading logarithms provides predictions which include leading terms of all perturbative orders; however, they are not complete in the sense of fixed-order predictions because the latter also include sub-leading logarithms and non-logarithmic terms. This is illustrated in Table 1, where the NLLA calculations provide the sum of all terms of the first two columns, while NLO ($`𝒪(\alpha _\mathrm{s}^2)`$) calculations yield the sums of the terms in the first two rows.
If the maximum available information from both resummed and fixed order calculations shall be utilized, they must be added such that double counting of terms which are common to both these calculations is avoided. This can be achieved involving several (approximate) ‘matching schemes’. For instance, terms to $`𝒪(\alpha _\mathrm{s}^2)`$ in the NLLA expression for $`\mathrm{ln}\mathrm{\Sigma }()`$ (c.f. table 1) can be removed and replaced by the full expression in exact $`𝒪(\alpha _\mathrm{s}^2)`$. This procedure is called ‘$`\mathrm{ln}R`$ matching’; it results in ‘resummed $`𝒪(\alpha _\mathrm{s}^2)`$’ predictions which should be superior to the fixed $`𝒪(\alpha _\mathrm{s}^2)`$ case because more higher order correction terms are included.
The perturbative order up to which QCD predictions for different processes and observables are available, is indicated in Table 6, at the end of this review.
### 2.7 Renormalization scale dependence - infinite discomfort in finite order.
The principal independence of a physical observable $``$ from the choice of the renormalization scale $`\mu `$ was expressed in equation 1. Replacing $`\alpha _\mathrm{s}`$ by $`\alpha _\mathrm{s}(\mu ^2)`$, using equation 2, and inserting the perturbative expansion of $``$ (equation 9) into equation 1 results in
$`0=\mu ^2{\displaystyle \frac{R_0}{\mu ^2}}+\alpha _\mathrm{s}(\mu ^2)\mu ^2{\displaystyle \frac{R_1}{\mu ^2}}`$ $`+`$ $`\alpha _\mathrm{s}^2(\mu ^2)\left[\mu ^2{\displaystyle \frac{R_2}{\mu ^2}}R_1\beta _0\right]`$ (12)
$`+`$ $`\alpha _\mathrm{s}^3(\mu ^2)\left[\mu ^2{\displaystyle \frac{R_3}{\mu ^2}}[R_1\beta _1+2R_2\beta _0]\right]`$
$`+`$ $`𝒪(\alpha _\mathrm{s}^4).`$
Solving this relation requires that the coefficients of $`\alpha _\mathrm{s}^n(\mu ^2)`$ vanish for each order $`n`$. With an appropriate choice of integration limits one thus obtains
$`R_0`$ $`=\mathrm{const}.,`$ (13)
$`R_1`$ $`=\mathrm{const}.,`$
$`R_2`$ $`\left({\displaystyle \frac{Q^2}{\mu ^2}}\right)=R_2(1)\beta _0R_1\mathrm{ln}{\displaystyle \frac{Q^2}{\mu ^2}},`$
$`R_3`$ $`\left({\displaystyle \frac{Q^2}{\mu ^2}}\right)=R_3(1)[2R_2(1)\beta _0+R_1\beta _1]\mathrm{ln}{\displaystyle \frac{Q^2}{\mu ^2}}+R_1\beta _0^2\mathrm{ln}^2{\displaystyle \frac{Q^2}{\mu ^2}}`$
as a solution of equation 12.
In other words, invariance of the complete perturbation series against the choice of the renormalization scale $`\mu ^2`$ implies that the coefficients $`R_n`$, except $`R_0`$ and $`R_1`$, explicitly depend on $`\mu ^2`$. In infinite order, the renormalization scale dependence of $`\alpha _\mathrm{s}`$ and of the coefficients $`R_n`$ cancel; in any finite (truncated) order, however, the cancellation is not perfect, such that all realistic perturbative QCD predictions include an explicit dependence on the choice of the renormalization scale.
This renormalization scale dependence is most pronounced in leading order QCD because $`R_1`$ does not depend on $`\mu `$ and thus, there is no cancellation of the (logarithmic) scale dependence of $`\alpha _\mathrm{s}(\mu ^2)`$ at all. Only in next-to-leading and higher orders, the scale dependence of the coefficients $`R_n`$, for $`n2`$, partly cancels that of $`\alpha _\mathrm{s}(\mu ^2)`$. In general, the degree of cancellation improves with the inclusion of higher orders in the perturbation series of $``$.
A practical example of the scale dependence of $`\alpha _\mathrm{s}`$, determined from the measured value of the scaled hadronic decay width of the $`\mathrm{Z}^0`$ boson ,
$$R_\mathrm{Z}=\frac{\mathrm{\Gamma }(\mathrm{Z}^0\mathrm{hadrons})}{\mathrm{\Gamma }(\mathrm{Z}^0\mathrm{leptons})}=20.768\pm 0.0024,$$
(14)
is shown in figure 4. QCD predictions of $`R_\mathrm{Z}`$ are available in complete NNLO . Including further corrections like quark mass and non-factorizable electroweak and QCD effects, these predictions can be parametrised to
$$R_\mathrm{Z}=19.934\left[1+1.045\frac{\alpha _\mathrm{s}(\mu )}{\pi }+0.94\left[\frac{\alpha _\mathrm{s}(\mu )}{\pi }\right]^215\left[\frac{\alpha _\mathrm{s}(\mu )}{\pi }\right]^3\right],$$
(15)
for $`\mu M_{\mathrm{Z}^0}`$; see section 4.3 for more details. In order to demonstrate the scale dependence of $`\alpha _\mathrm{s}`$, $`\alpha _\mathrm{s}(\mu )`$ is determined from equations 14 and 15 in LO, in NLO and in NNLO, as a function of the scaled renormalization scale, $`x_\mu =\mu /M_{\mathrm{Z}^0}`$, using the appropriate, scale dependent QCD coefficients $`R_n`$ according to equation 13. $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})`$ is then calculated from $`\alpha _\mathrm{s}(\mu )`$ using equation 2.4 in 1-loop expansion for the LO, in 2-loop for the NLO and in 3-loop for the NNLO case.
The resulting scale dependence of $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})`$ is displayed in figure 4(a); the scale dependence of the QCD coefficients $`R_1`$, $`R_2`$ and $`R_3`$ is given in figure 4(b). The scale dependence of $`\alpha _\mathrm{s}`$ is visibly reduced if higher orders are included: limiting renormalization scales to factors from 1/5 to 5 of the “physical” scale $`QM_{\mathrm{Z}^0}`$, results in changes of $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})`$ of about 40% in LO, 8% in NLO and 2.5% in NNLO. Beyond these limits, however, the scale dependence even in NNLO QCD diverges, such that it seems not to be meaningful to consider yet larger ranges of scale factors $`x_\mu `$ \- although there is no rule, from first principles, which limits scale variations to a definite range.
There are several proposals to optimize or to fix the renormalization scale in finite order perturbation theory. An intuitive approach is to apply the general scale insensitivity requirement, equation 1, which is strictly valid in infinite order, to calculations in any finite (beyond leading) order. This is Stevenson’s principle of minimal sensitivity (PMS), where, for each observable $``$, the optimal scale is defined by d$`/\mathrm{d}\mu ^2=0`$. The PMS solution for $`R_\mathrm{Z}`$ in NLO is marked in figure 4(a) \[filled circle\].
In the “effective charge” method (EC) of Grunberg , all higher (i.e. beyond leading) order coefficients $`R_n`$ vanish. The EC solution in NLO is also marked in figure 4(a) \[open circle\].
Brodsky, Lepage and Mackenzie (BLM) propose to choose $`\mu `$ such that the NLO coefficient $`R_2`$ is independent of $`N_f`$, the number of active quark flavours. The BLM solution is marked in figure 4(a) \[open rhombus\].
PMS, EC and BLM all provide unambiguous solutions in NLO; in the case of $`R_\mathrm{Z}`$, they are very close to each other but the NNLO prediction does not exactly match these results: the value of $`\alpha _\mathrm{s}`$ from “optimized” NLO can only be reached in NNLO at very small values of $`x_\mu `$.
While in NLO QCD it is sufficient to vary the renormalization scale in order to assess the full spectrum of variation offered by the renormalization procedure, in NNLO $`both`$ scale- and scheme-dependences must be taken into account. PMS, EC and BLM can then be generalized to NNLO and beyond, however do not lead to unambiguous results anymore. The whole issue of theoretical scale optimizations was and still is vividly discussed in the literature, see e.g. .
Another approach is motivated by an experimental point of view: ideally, the optimal choice of $`\mu `$ in a given order of perturbation theory should reproduce or, at least, closely approach the unknown all-order result. Measurements of observables should inherently include all orders of perturbation theory; therefore it may be possible to extract the optimal choice of $`\mu `$ in a given, finite order calculation from fits to the experimental data. Indeed, experimental scale optimization is possible in cases where $`\alpha _\mathrm{s}`$ is determined from differential $`distributions`$ of observables, and often leads to a remarkable consistency between results obtained from different observables .
So far, none of the methods described above was generally accepted as $`the`$ preferred method to optimize or fix renormalistaion scales in finite order calculations. Instead, in experimental determinations of $`\alpha _\mathrm{s}`$, systematic uncertainties due to unknown higher order contributions are usually defined by allowing the renormalization scale $`\mu `$ to vary in “reasonable” ranges.
Unfortunately, there is no common agreement on what should be a reasonable range, and the significance and interpretation of such procedures is unclear. Furthermore, as indicated above, scale changes alone are not sufficient to investigate higher order uncertainties in NNLO predictions, where renormalization scheme dependences should also be accounted for.
### 2.8 Lattice QCD
At large distances or low momentum transfers, $`\alpha _\mathrm{s}`$ becomes large and application of perturbation theory becomes inappropriate. In this regime, nonperturbative methods must be used to describe processes of the Strong Interaction. Lattice QCD is one of the most developed nonperturbative methods which is used to calculate, for instance, hadron masses, hadron mass splittings and QCD matrix elements. In Lattice QCD, field operators are applied on a discrete, 4-dimensional Euclidean space-time of hypercubes with side length $`a`$.
Energy levels of heavy quarkonia systems ($`Q\overline{Q}`$) calculated using lattice QCD depend on the heavy quark mass $`M_Q`$ and on $`\alpha _\mathrm{s}`$. Comparison with measurements of quarkonia mass splittings then allows to determine $`\alpha _\mathrm{s}`$, which at this point is not given in the $`\overline{\text{MS}}`$ renormalization scheme but is based on other suitable definitions of $`\alpha _\mathrm{s}`$ on the lattice. Conversion from the lattice to the $`\overline{\text{MS}}`$ coupling at high energies can be done using an expansion in second or in third order perturbation theory. A review of methods to determine $`\alpha _\mathrm{s}`$ from lattice investigations is given in
So far, calculations exist which either neglect light quark loop contributions ($`N_f`$=0; “quenched approximation”) or which include two light quark flavours ($`N_f`$=2); the latter allow extrapolation to $`N_f`$=3. Uncertainties in this extrapolation, the limited order to which the conversion to the $`\overline{\text{MS}}`$ scheme is known, limited Monte-Carlo statistics and corrections for light quark masses lead to theoretical uncertainties of final ($`\overline{\text{MS}}`$) values of $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})`$.
## 3 Results from deep inelasting scattering processes
Measurements of the violation of Bjorken scaling in deep inelastic lepton-nucleon scattering belong to the earliest methods to determine $`\alpha _\mathrm{s}`$. The first significant determinations of $`\alpha _\mathrm{s}`$, i.e. those based at least on next-to-leading order (NLO) perturbative QCD prediction, started to emerge in 1979 .
Today, a large number of results is available, from data in the energy ($`Q^2`$) range of a few to several thousand GeV<sup>2</sup>, using electron-, muon- and neutrino-beams on various fixed target materials (H, D, C, Fe and CaCO<sub>3</sub>), as well as electron-proton or positron-proton colliding beams at HERA. In addition to scaling violations of structure functions, $`\alpha _\mathrm{s}`$ is also determined from moments of structure functions, from QCD sum rules and — as in $`\mathrm{e}^+\mathrm{e}^{}`$ annihilation — from hadronic jet production and event shapes.
### 3.1 Scaling violations and moments of structure functions
Cross sections of physical processes in lepton-nucleon scattering and in hadron-hadron collisions depend on the quark- and gluon-densities in the nucleon. Assuming factorization between short-distance, hard scattering processes which can be calculated using QCD perturbation theory, and low-energy or long-range processes which are not accessible by perturbative methods, such cross sections are parametrized by a set of structure functions $`F_i`$ ($`i`$= 1,2,3). The transition between the long- and the short-range regimes is defined by an arbitrary factorization scale $`\mu _f`$, which — in general — is independent from the renormalization scale $`\mu `$, but has similar features as the latter: the higher order coefficients of the perturbative QCD series for physical cross sections depend on $`\mu _f`$ in such a way that the cross section to all orders must be independent of $`\mu _f`$, i.e. $`\sigma /\mu _f=0`$. To simplify application of theory to experimental measurements, the assumption $`\mu _f=\mu `$ is usually made, with $`\mu Q`$ as the standard choice of scales.
In the naive quark-parton model, i.e. neglecting gluons and QCD, the differential cross sections for electromagnetic charged lepton (electron or muon) or weak neutrino-proton scattering off an unpolarized proton target are written
$`{\displaystyle \frac{\mathrm{d}^2\sigma ^{em}}{\mathrm{d}x\mathrm{d}y}}`$ $``$ $`{\displaystyle \frac{1+(1y)^2}{2}}2xF_1+(1y)(F_2^{em}2xF_1^{em}){\displaystyle \frac{M}{2E}}xyF_2^{em};`$
$`{\displaystyle \frac{\mathrm{d}^2\sigma ^\nu }{\mathrm{d}x\mathrm{d}y}}`$ $``$ $`\left(1y{\displaystyle \frac{M}{2E}}xy\right)F_2^\nu +y^2xF_1^\nu +y\left(1{\displaystyle \frac{1}{2}}y\right)xF_3^\nu ,`$ (16)
where $`x=\frac{Q^2}{2M(EE^{})}`$ is the momentum fraction of the nucleon carried by the struck parton, $`y=1E^{}/E`$, $`Q^2`$ is the negative quadratic momentum transfer in the scattering process, and $`M`$, $`E`$ and $`E^{}`$ are the mass of the proton and the lepton energies before and after the scattering, respectively, in the rest frame of the proton. In the quark-parton model, these structure functions consist of combinations of the quark- and antiquark densities $`q(x)`$ and $`\overline{q}(x)`$ for both valence- (u,d) and sea-quarks (s,c).
In QCD, the gluon content of the proton as well as higher order diagrams describing photon/gluon scattering, $`\gamma gq\overline{q}`$, and gluon radiation off quarks must be taken into account. Quark- and gluon-densities, the structure functions $`F_i`$ and physical cross sections become energy ($`Q^2`$) dependent. QCD thus predicts, departing from the naive quark-parton model, scaling violations in physical cross sections, which are associated with the radiation of gluons. While perturbative QCD cannot predict the functional form of parton densities and structure functions, their energy evolution is described by the so-called DGLAP equations .
Structure functions contain, apart from terms whose energy dependence is given by perturbative QCD, which so far is known up to complete NLO , so-called “higher twist” contributions (HT). The leading higher twist terms are proportional to $`1/Q^2`$; they are numerically important at low $`Q^2<𝒪`$(few GeV$`{}_{}{}^{2})`$ and at very large $`x1`$.
Historically, precise results of $`\alpha _\mathrm{s}`$, from the logarithmic slopes of $`F_2(x,Q^2)`$ and in NLO QCD, were obtained from a combined analysis of SLAC and BCDMS data, in a $`Q^2`$ range from 0.5 to 260 GeV<sup>2</sup>, giving $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})=0.113\pm 0.005`$. The error includes theoretical (scale) uncertainties of $`\pm 0.004`$ . Higher twist terms as well as the gluon distribution were simultaneously determined in this analysis.
The CCFR colaboration obtained $`\alpha _\mathrm{s}`$ from $`\nu `$-nucleon scattering and a fit to the non-singlet structure function $`F_3(x,Q^2)`$ , which is independent of the poorly known gluon distribution. In order to increase statistical precision, $`F_3`$ was substituted by $`F_2`$ at large $`x`$ where gluons do not contribute much, giving $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})=0.111\pm 0.002(\mathrm{stat}.)\pm 0.003(\mathrm{sys}.)\pm 0.004(\mathrm{theo}.)`$.
These two results were, for several years, the most significant from deep inelastic lepton-nucleon scattering. Because they were numerically smaller than typical values obtained from $`\mathrm{e}^+\mathrm{e}^{}`$ annihilation, $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})0.120`$ (see Section 4), speculations about possible reasons for such differences arose. These speculations came to a halt, at least partly, when the CCFR collaboration corrected their previous result — due to a new energy calibration of the detector — to $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})=0.119\pm 0.005`$ . Further results from $`F_2`$, in a study of the HERA data at small $`x`$ and $`Q^2<100\mathrm{GeV}^2`$, based on NLO QCD including summations of all leading and subleading logarithms of $`Q^2`$ and $`1/x`$, lead to $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})=0.120\pm 0.005\pm 0.009`$, where the first error is experimental and the second theoretical. Finally, a reanalysis of the SLAC, BCDMS and NMC data on $`F_2`$, taking proper account of point-to-point correlations, resulted in an increase of $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})`$ from 0.113 to 0.118 . A continuation of this analysis, including recent HERA data and careful studies of the effects of higher twist terms, was reported to result in $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})=0.116\pm 0.003`$ .
Significant progress in this field was achieved with the availability of NNLO QCD predictions for the non-singlet structure function $`F_3`$ as well as for moments of $`F_2`$. A new analysis of the CCFR data, now in NNLO QCD and from $`xF_3`$ alone, resulted in
$$\alpha _\mathrm{s}(M_{\mathrm{Z}^0})=0.118\pm 0.002(\mathrm{stat}.)\pm 0.005(\mathrm{sys}.)\pm 0.003(\mathrm{theo}.);$$
see figure 5. This value is taken as the currently most significant result from structure functions in neutrino-nucleon deep inelastic scattering, and is included in the final summary, see Table 6. Another recent study was based on Bernstein polynomials and moments of $`\mathrm{F}_2`$ from electron- and muon-scattering data, including fixed target as well as HERA colliding beam data in the $`Q^2`$ range of 2.5 to 230 GeV<sup>2</sup>. The result was
$$\alpha _\mathrm{s}(M_{\mathrm{Z}^0})=0.1172\pm 0.0017\pm 0.0017,$$
where the first error is experimental and the second theoretical. The theoretical error includes higher twist effects and an estimate of the NNNLO corrections. This result is taken as the final value from $`F_2`$ in DIS and is added to the summary in section 7.
From scaling violations of polarized structure functions, based on data from SLAC and SMC, $`\alpha _\mathrm{s}`$ was determined in NLO QCD, resulting in
$$\alpha _\mathrm{s}(M_{\mathrm{Z}^0})=0.120_{0.005}^{+0.004}(\mathrm{exp}.)_{0.006}^{+0.009}(\mathrm{theo}.),$$
which is also added to the final summary.
### 3.2 Sum Rules
Perturbative corrections to two inclusive measurements, namely the Gross-Llewellyn Smith (GLS) sum rule for deep inelastic neutrino scattering and the Bjorken sum rule for polarized structure functions, have been determined to complete NNLO QCD . The GLS sum rule,
$$\mathrm{GLS}=_0^1F_3(x,Q^2)dx3(1\frac{\alpha _\mathrm{s}}{\pi }+\mathrm{}),$$
(17)
when fitted to data of the CCFR collaboration at $`Q^2=3\mathrm{GeV}^2`$, resulted in $`\alpha _\mathrm{s}(1.73\mathrm{GeV})=0.32\pm 0.05`$. A recent update of the data and analysis from CCFR , in the $`Q^2`$ range from 1 to 15.5 GeV<sup>2</sup>, gave
$$\alpha _\mathrm{s}(1.73\mathrm{GeV})=0.28\pm 0.035(\mathrm{stat}.)\pm 0.050(\mathrm{sys}.)_{0.030}^{+0.035}(\mathrm{theo}.).$$
The systematic uncertainty is dominated by the extrapolation of the GLS integral to the regions $`x<0.01`$ where no measurements exist, and to $`x>0.5`$ which is substituted by $`F_2`$ from SLAC data, see figure 6. The theoretical error is dominated by uncertainties in the higher twist corrections.
The Bjorken polarized sum rule determines that
$$_0^1\left[g_1^p(x)g_1^n(x)\right]dx=\frac{1}{3}\left|\frac{g_A}{g_V}\right|\left(1\frac{\alpha _\mathrm{s}}{\pi }+\mathrm{}\right),$$
(18)
where the polarized structure functions $`g_1(x)`$ for protons (superscript $`p`$) and neutrons ($`n`$) are derived from the difference of cross sections for parallel and antiparallel polarized targets and ($`\mu `$ or e) beams; $`g_A`$ and $`g_V`$ are the constants of the neutron weak decay, $`g_A/g_V=1.26`$. A study of the CERN SMC and the SLAC E142 data obtained
$$\alpha _\mathrm{s}(1.58\mathrm{GeV})=0.375_{0.081}^{+0.062}.$$
No explicit corrections for nonperturbative higher twist where applied to derive this result, however an estimate of the size of the $`𝒪(\alpha _\mathrm{s}^4)`$ terms was taken into account.
### 3.3 Jet Production and Event Shapes
Observables parametrizing hadronic event shapes and jet production rates are the classical inputs for $`\alpha _\mathrm{s}`$ studies in $`\mathrm{e}^+\mathrm{e}^{}`$ annihilation. In recent years, these observables were also studied in high energetic electron- and positron-proton collisons at HERA, leading to significant determinations of $`\alpha _\mathrm{s}`$.
In detail, inclusive as well as differential jet production rates were studied in the energy range of $`Q^210`$ up to 10000 GeV<sup>2</sup> , using variants of jet algorithms from $`\mathrm{e}^+\mathrm{e}^{}`$ annihilation which are decribed, in more detail, in section 4. In leading order $`\alpha _\mathrm{s}`$, 2 + 1 jet events in deep inelastic $`ep`$ scattering arise from photon-gluon fusion and from QCD compton processes. The term ‘2 + 1 jet’ denotes events where two resolved jets can be identified, in addition to the beam jet from the remnants of the incoming proton. Previous NLO QCD predictions which were shown to be imprecise are now replaced by more recent calculations .
The results of $`\alpha _\mathrm{s}`$ from jet production at HERA can be summarized to
$$\alpha _\mathrm{s}(M_{\mathrm{Z}^0})=0.118\pm 0.002(\mathrm{stat}.)\pm 0.008(\mathrm{sys}.)\pm 0.007(\mathrm{theo}.).$$
The systematic error contains uncertainties from using different jet algorithms and hadronization models, and the theoretical error is dominated by uncertainties from structure functions and from scale variations.
So far, studies of $`\alpha _\mathrm{s}`$ from hadronic event shape distributions at HERA are based on fits to the energy ($`Q^2`$) evolution of mean values of shape distributions, using NLO QCD predictions together with parametrizations of power corrections $`1/Q^p`$ . These methods do not yet lead to consistent results of $`\alpha _\mathrm{s}`$ from different shape observables — at least not to the degree of precision which is obtained from “classical” studies of $`\alpha _\mathrm{s}`$ from event shape distributions, as e.g. in $`\mathrm{e}^+\mathrm{e}^{}`$ annihilation; see Section 4.1 for further discussion. Awaiting further progress in theoretical understanding of power corrections to event shapes, the $`\alpha _\mathrm{s}`$ results of these studies will not yet be included in the final summary of $`\alpha _\mathrm{s}`$.
## 4 Results from $`\mathrm{e}^+\mathrm{e}^{}`$ annihilation
In $`\mathrm{e}^+\mathrm{e}^{}`$ annihilation reactions, $`\alpha _\mathrm{s}`$ is classically determined from hadronic event shapes, jet production rates and energy correlations for which complete NLO QCD calculations are available . The first data analysis of this type, from jet rates observed at the PETRA $`\mathrm{e}^+\mathrm{e}^{}`$ collider, emerged in 1982 . Reviews of early results on $`\alpha _\mathrm{s}`$ from $`\mathrm{e}^+\mathrm{e}^{}`$ annihilations can be found e.g. in references .
More recently, this field was extensively covered by the experiments at the LEP and SLC colliders, see e.g. for reviews of the first few years of LEP operation. Using data from LEP and from earlier $`\mathrm{e}^+\mathrm{e}^{}`$ colliders at lower c.m. energies, $`\alpha _\mathrm{s}`$ can be determined from scaling violations of fragmentation functions. Most importantly, precise determinations of $`\alpha _\mathrm{s}`$ are now obtained from the hadronic partial width of $`\mathrm{Z}^0`$ decays, from overall fits to precision electroweak data and from hadronic branching fractions of $`\tau `$ lepton decays, which are calculated in complete NNLO QCD.
### 4.1 Event shapes, jet rates and energy correlations
#### 4.1.1 Observables in (resummed) NLO QCD
Hadronic event shape variables, jet rates and energy correlations are tools to study both the amount of gluon radiation and details of the hadronization process. The definitions of observables which are applied to hadronic final states of $`\mathrm{e}^+\mathrm{e}^{}`$ annihilations, like Thrust, Thrust Major and Minor, Oblateness, jet masses, the jet broadening measures, energy correlations and jet production rates, are summarized elsewhere; see e.g. . Only two of these, the Thrust observable and the JADE jet algorithm shall be described here in some more detail:
The Thrust $`T`$ is the normalized sum of the momentum components $`\stackrel{}{p}_i\stackrel{}{n}`$ of all particles $`i`$ of a given event along a specific axis $`\stackrel{}{n}`$; the axis is chosen such that $`T`$ is maximized:
$$T=\mathrm{max}\left(\frac{_i|\stackrel{}{p}_i\stackrel{}{n}|}{_i|\stackrel{}{p}_i|}\right).$$
Thrust assumes values of $`T=1`$ for perfectly aligned momentum vectors, i.e. for an ideal event $`\mathrm{e}^+\mathrm{e}^{}q\overline{q}`$ with two narrow back-to-back jets, down to $`T=0.5`$ for a completely spherical distribution of momentum vectors in the limit of events with many gluons radiated off the initial $`q\overline{q}`$ pair.
Within the JADE jet algorithm , the scaled pair mass of two resolvable jets $`i`$ and $`j`$, $`y_{ij}=M_{ij}^2/E_{vis}^2`$, is required to exceed a threshold value $`y_{cut}`$, where $`E_{vis}`$ is the sum of the measured energies of all particles of an event. In a recursive process, the pair of particles or clusters of particles $`n`$ and $`m`$ with the smallest value of $`y_{nm}`$ is replaced by (or “recombined” into) a single jet or cluster $`k`$ with four-momentum $`p_k=p_n+p_m`$, as long as $`y_{nm}<y_{\mathrm{cut}}`$. The procedure is repeated until all pair masses $`y_{ij}`$ are larger than the jet resolution parameter $`y_{\mathrm{cut}}`$, and the remaining clusters of particles are called jets.
Several jet recombination schemes and definitions of $`M_{ij}`$ exist ; the original JADE scheme with $`M_{ij}^2=2E_iE_j(1\mathrm{cos}\theta _{ij})`$, where $`E_i`$ and $`E_j`$ are the energies of the particles and $`\theta _{ij}`$ is the angle between them, and the “Durham” scheme with $`M_{ij}^2=2\mathrm{min}(E_i^2,E_j^2)(1\mathrm{cos}\theta _{ij})`$, were most widely used at LEP, due to their superior features like small sensitivity to hadronization and particle mass effects .
QCD predictions for hadronic event shapes, of jet production rates and of energy correlations are available in complete NLO . Differential distributions of such observables $`y`$ are parametrized as
$$\frac{1}{\sigma _0}\frac{\mathrm{d}\sigma }{\mathrm{d}y}=R_1(y)\alpha _\mathrm{s}(\mu ^2)+R_2(y,Q^2/\mu ^2)\alpha _\mathrm{s}^2(\mu ^2),$$
(19)
where $`\sigma _0`$ is the total hadronic cross section in leading order. In addition, for some of the observables, resummation of the leading and next-to-leading logarithms (NLLA) is available which can be matched with the NLO expressions (resummed NLO); see section 2.6.
In a typical study, measured distributions are corrected for effects of limited detector acceptance and resolution. Fits of QCD predictions to the data are performed after applying hadronization corrections, as predicted from hadronization models, to the data or to the theoretical predictions. Systematic uncertainties due to these corrections, from renormalization scale variations and from different matching procedures, are studied and included in the overall error of the fitted value of $`\alpha _\mathrm{s}`$. No common agreement exists, however, about the range of variations used to estimate these uncertainties.
Early determinations of $`\alpha _\mathrm{s}`$ from event shapes, jet rates and energy correlations obtained from experiments at the PETRA and PEP colliders were summarized to $`\alpha _\mathrm{s}=0.14\pm 0.02`$ at $`Q^2=E_{cm}^2(34\mathrm{GeV})^2`$ , in NLO QCD. The results which led to this average did not include estimates of theoretical uncertainties; however the error was determined from the scatter of the single results which were based on different observables, and thus gave a first estimate of theoretical uncertainties.
#### 4.1.2 Results in resummed NLO
A reanalysis of PETRA data, using refined analysis techniques, resummed NLO QCD calculations, modern model calculations and including new observables, quite similar to recent analyses at LEP, resulted in
$`\alpha _\mathrm{s}\left(22\mathrm{GeV}\right)`$ $`=`$ $`0.161\pm 0.009\pm _{0.006}^{+0.014},`$
$`\alpha _\mathrm{s}\left(35\mathrm{GeV}\right)`$ $`=`$ $`0.145\pm 0.002_{0.007}^{+0.012},\mathrm{and}`$
$`\alpha _\mathrm{s}\left(44\mathrm{GeV}\right)`$ $`=`$ $`0.139\pm 0.004_{0.007}^{+0.010},`$
where the first errors are statistical and experimental systematic, added in quadrature, and the second are theoretical uncertainties. These values are taken as the final results in the PETRA energy range and are included in the summary table 6. Similar analyses from experiments at the TRISTAN collider, around c.m. energies of 58 GeV, gave
$$\alpha _\mathrm{s}\left(58\mathrm{GeV}\right)=0.132\pm 0.008,$$
which is also included in the summary.
At the LEP collider, all four experiments (ALEPH, DELPHI, L3 and OPAL) have contributed a multitude of studies based on the high statistics data samples around the $`\mathrm{Z}^0`$ resonance (LEP-I; $`E_{cm}91.2`$ GeV) and at the higher energies of the LEP-II running phase ($`E_{cm}133,161,172,183`$ and 189 GeV). The SLD experiment at the SLAC Linear Collider (SLC) contributed similar studies at $`E_{cm}M_{\mathrm{Z}^0}`$. As an example of the precise description of data by the CQD fits, the Thrust distributions as measured by ALEPH at LEP-I and LEP-II energies are shown in figure 7, together with the corresponding fits of the resummed NLO QCD calculations.
The most current results from LEP and SLC, based on resummed NLO QCD calculations, are taken from and are combined to obtain one single value of $`\alpha _\mathrm{s}`$ at each c.m. energy. This is done by calculating weighted averages of the results quoted by each experiment, with the inverse quadratic total error taken as the weight. The experimental uncertainties are combined assuming that no correlation exists between the experiments. Theoretical uncertainties are assumed to be common to all experiments, however different methods and definitions were used to estimate these in each case. Therefore, a linear average of the theoretical uncertainties quoted by the experiments is used to define the theoretical error of the combined results. This gives
$`\alpha _\mathrm{s}\left(91.2\mathrm{GeV}\right)`$ $`=`$ $`0.121\pm 0.001\pm 0.006,`$
$`\alpha _\mathrm{s}\left(133\mathrm{GeV}\right)`$ $`=`$ $`0.113\pm 0.003\pm 0.006,`$
$`\alpha _\mathrm{s}\left(161\mathrm{GeV}\right)`$ $`=`$ $`0.109\pm 0.004\pm 0.005,`$
$`\alpha _\mathrm{s}\left(172\mathrm{GeV}\right)`$ $`=`$ $`0.104\pm 0.004\pm 0.005,`$
$`\alpha _\mathrm{s}\left(183\mathrm{GeV}\right)`$ $`=`$ $`0.109\pm 0.002\pm 0.004,`$
$`\alpha _\mathrm{s}\left(189\mathrm{GeV}\right)`$ $`=`$ $`0.110\pm 0.001\pm 0.004,`$
where the first errors are experimental and the second theoretical. These results are included in the final summary table 6. Note, however, that some of the averages at LEP-II energies contain results which are still preliminary, see .
The energy dependence of $`\alpha _\mathrm{s}`$ from studies of event shape observables is clearly seen in the results presented above. This is also demonstrated in a dedicated analysis of the L3 Collaboration which includes a study of radiative events recorded at LEP-I. Such events are effectively produced at reduced hadronic centre of mass energies; they can therefore be used to determine $`\alpha _\mathrm{s}`$ at energy scales below the nominal collider energy. The statistical precision at these reduced energies is, however, limited.
The results of the L3 study are shown in figure 8, demonstrating the agreement of the data with the QCD expectation of a running coupling. Note that, in order to judge the significance for the running, only the innermost, experimental errors must be taken into account, since theoretical uncertainties are highly correlated between the different data points. The running of $`\alpha _\mathrm{s}`$ was also demonstrated in a recent study of jet production rates at PETRA and at LEP energies, using the same consistent analysis method and data from two experiments comprising similar detector techniques, JADE and OPAL .
Before the advent of resummed QCD calculations for event shape observables, determinations of $`\alpha _\mathrm{s}`$ were done in pure NLO, see e.g. . Studies of theoretical uncertainties typically included scale variations from $`x_\mu =1`$ down to the best fit values of $`x_\mu `$, from two-parameter fits, or the optimized scales given by the methods discussed in Section 2.7 . These ranges were quite large, down to $`x_\mu 0.01`$, which lead to rather large scale uncertainties in case of some of the observables. Resummed NLO results, however, had smaller dependencies on scale variations, as theoretically expected, which is why they developed as a standard in $`\alpha _\mathrm{s}`$ determinations from event shape studies.
#### 4.1.3 Pure NLO results
Fits of pure NLO QCD to experimental event shape and jet rate distributions are known to provide a consistent description of the data if - in addition to $`\alpha _\mathrm{s}`$ \- also the renormalization scale is treated as a free parameter of the fit . While the fitted scale factor $`x_\mu `$ assumes different values for different observables, indicating that the amount of unknown, higher order contributions is different in each case, the resulting value of $`\alpha _\mathrm{s}`$ appears to be “universal” and in much better agreement than applying a single and fixed choice of scale .
These findings are corroborated in a recent re-analysis of DELPHI data at LEP-I , which also demonstrates that experimentally optimized NLO calculations can provide a more consistent description of the data than resummed NLO at fixed scale $`x_\mu =1`$. Defining a scale range, for each observable, of a factor of 2 around the experimental fit value of $`x_\mu ^2`$, results in $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})=0.1174\pm 0.0026`$, which includes both experimental and theoretical uncertainties. This method of optimizing renormalization scales and defining the size of the remaining theoretical uncertainties seems logical from an experimental point of view, see the discussion in Section 2.7, but carries intrinsic theoretical problems:
Firstly, the resulting scale factors are rather small, down to $`𝒪`$(0.01) or smaller, such that large logarithms appear in the NLO coefficients, see equation 13. Secondly, there is no common agreement about the significance of $`one`$ overall scale for each distribution because theoretical scale optimizations predict $`x_\mu `$ to depend on the value of the observable itself. The issue of experimental scale optimization therefore requires further investigation and discussion.
#### 4.1.4 Power corrections
In the past few years, analytical approaches were pursued to approximate nonperturbative hadronization effects by means of perturbative methods, introducing a universal, non-perturbative parameter
$$\alpha _0(\mu _I)=\frac{1}{\mu _I}_0^{\mu _I}dk\alpha _\mathrm{s}(k)$$
to parametrize the unknown behaviour of $`\alpha _\mathrm{s}(Q)`$ below a certain infrared matching scale $`\mu _I`$ . Divergent soft gluon contributions to the perturbative predictions of event shapes are removed and, as a consequence of these techniques, lead to corrections which are proportional to powers of $`1/Q`$. Power corrections are regarded as an alternative approach to describe hadronization effects on event shape distributions, instead of using phenomenological hadronization models as in the studies discussed in the previous sections.
The energy dependence of event shape distributions and of their integrated mean values, in the c.m. energy range from 14 to 183 GeV, were analysed and compared with perturbative QCD calculations plus added power corrections . Two-parameter fits of $`\alpha _0`$ and $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})`$ to the data provide a consistent description of the shapes and the energy evolution of the data; however the scatter between results from different shape observables is larger than the typical uncertainties would suggest. The non-perturbative parameter $`\alpha _0`$ turns out to be universal for all studied observables within about 20%, and values of $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})`$ seem to be close to but systematically smaller than those from fits using conventional methods of hadronization corrections .
Power corrections, as an analytical ansatz to compute nonperturbative QCD corrections to event shape observables, offer a rather promising alternative to the use of phenomenological hadronization models in determinations of $`\alpha _\mathrm{s}`$; however more experience, confidence and further developments are necessary before actual fit results of $`\alpha _\mathrm{s}`$ can supersede the results from “classical” event shape analyses described above.
### 4.2 $`\alpha _\mathrm{s}`$ from scaling violations of fragmentation functions
The total cross section of $`\mathrm{e}^+\mathrm{e}^{}`$ annihilations into charged hadrons $`h`$, $`\mathrm{e}^+\mathrm{e}^{}h+X`$, can be factorized into a perturbative and a nonperturbative regime, describing processes of hard gluon radiation and of hadronization, respectively. The perturbative part can be calculated in terms of so-called coefficient functions, while the non-perturbative part is parametrized by phenomenological fragmentations functions. The separation of these two regimes is performed at an energy scale $`\mu _f`$, the factorisation scale, which — as in DIS — enters as a second, arbitrary energy scale, in addition to the renormalization scale $`\mu `$. According to the factorization theorem, the cross section can then be written :
$$\frac{\mathrm{d}\sigma }{\mathrm{d}x}(\mathrm{e}^+\mathrm{e}^{}h+X)=\underset{i}{}_x^1\frac{\mathrm{d}z}{z}C_i(z,\alpha _\mathrm{s}(\mu ^2),\mu ^2/Q^2)D_i(x/z,\mu _f^2),$$
(20)
where $`C_i`$ are the coefficient functions for creating a parton with flavour $`i`$ and momentum fraction $`z=p_{parton}/p_{beam}`$, $`D_i`$ represent the probability that parton $`i`$ fragments into a hadron $`h`$ with momentum fraction $`x/z`$, and $`x=p_{hadron}/p_{beam}`$.
In lowest order, the coefficient functions $`C_i`$ are given by the electroweak couplings for quarks; they vanish for gluons. Higher order QCD corrections apply to the quark coefficient functions and lead to finite functions for gluons, too. These corrections are known up to complete NLO , however only the LO corrections were available and used in currently existing analyses.
The fragmentation functions $`D_i`$ are not given by perturbation theory; however — as in the case of parton densities and structure functions (c.f. Section 3.1) — their energy dependence is described by QCD via complicated integro-differential equations and the DGLAP evolution functions .
The ALEPH and the DELPHI collaborations at LEP have both published detailed analyses of the charged hadron fragmentation functions, using data samples from the PETRA, PEP and LEP colliders spanning c.m. energy ranges from 14 to 91.2 GeV. At LEP, ALEPH and DELPHI extracted differential $`x`$ distributions separately for initial b-, c- and uds-quark events, and obtained the corresponding gluon-jet distribution from tagged gluon jet event samples. Assuming a certain parametrization of the “bare” fragmentation functions at a “starting energy scale” $`Q_0`$, $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})`$, the free parameters of the starting fragmentation functions for gluons and the different quark flavours, as well as a parameter allowing for additional, nonperturbative power law corrections, are determined in fits to all experimental $`x`$-distributions.
Combining both the ALEPH and the DELPHI results, which are based on comparable data sets and analysis strategies, leads to
$$\alpha _\mathrm{s}(M_{\mathrm{Z}^0})=0.125_{0.007}^{+0.006}(\mathrm{exp}.)\pm 0.009(\mathrm{theo}.)$$
where the theoretical uncertainty includes variations of both the factorization and the renormalization scales<sup>6</sup><sup>6</sup>6 DELPHI chose $`\mu _f\mu `$ and varied both within 1/2 to 2 $`E_{cm}`$; ALEPH varied each independently within $`1/\sqrt{e}`$ to $`\sqrt{e}E_{cm}`$ and added both uncertainties in quadrature. Here, the uncertainties quoted by DELPHI are taken because they are less restrictive in limiting the range of scales..
### 4.3 $`\alpha _\mathrm{s}`$ from total hadronic cross sections
Because of its inclusive nature, the total cross section of the process $`\mathrm{e}^+\mathrm{e}^{}`$ hadrons was the first quantity for which QCD corrections up to complete NNLO were known . For c.m. energies in the $`\mathrm{e}^+\mathrm{e}^{}`$ continuum, far below the $`\mathrm{Z}^0`$ pole and for $`\mu E_{cm}`$, the normalized cross section is given by
$$R_\gamma =\frac{\sigma (\mathrm{e}^+\mathrm{e}^{}\mathrm{hadrons})}{\sigma (\mathrm{e}^+\mathrm{e}^{}\mu ^+\mu ^{})}=3\underset{i}{}q_i^2\left(1+\frac{\alpha _\mathrm{s}}{\pi }+1.441\left(\frac{\alpha _\mathrm{s}}{\pi }\right)^212.8\left(\frac{\alpha _\mathrm{s}}{\pi }\right)^3\right),$$
(21)
where $`q_i`$ are the electrical charges of quark flavours $`i`$ which are produced, i.e. for which $`2M_qE_{cm}`$.
Because of the above functional form of $`R_\gamma (\alpha _\mathrm{s})`$, $`relative`$ errors of $`R_\gamma `$ lead to $`absolute`$ errors in $`\alpha _\mathrm{s}`$ of about the same size, $`\mathrm{\Delta }R/R\mathrm{\Delta }\alpha _\mathrm{s}`$, such that precise measurements of $`R`$ still lead to rather large errors in $`\alpha _\mathrm{s}`$. As another complication, electroweak corrections due to real $`\mathrm{Z}^0`$ exchange and due to $`\gamma \mathrm{Z}^0`$ interference must be applied to the data in the c.m. energy range above 20 GeV. These corrections are negligible at $`E_{cm}`$ = 14 GeV, but at $`E_{cm}`$ =46 GeV they amount to about the same size as the QCD corrections .
A combination of the cross section measurements of all four PETRA experiments, in the c.m. energy range of 14 to 46.8 GeV, accounting for correlated experimental uncertainties, electroweak corrections and leading order initial state radiation effects, resulted in $`\alpha _\mathrm{s}(34\mathrm{GeV})=0.169\pm 0.025`$ , in NLO QCD. Applying the full NNLO QCD prediction which was not yet available at that time results in $`\alpha _\mathrm{s}(34\mathrm{GeV})=0.175\pm 0.028`$ or — equivalently — in $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})=0.143\pm 0.018`$. Two groups have analysed and combined $`\mathrm{e}^+\mathrm{e}^{}`$ hadronic cross sections measured in the c.m. energy ranges from 7 to 57 GeV and from 2.65 to 52 GeV . They both used an initially erroneous value of the NNLO QCD coefficient of $`R_\gamma `$, +64.8 instead of -12.8; when corrected for this mistake, they result in $`\alpha _\mathrm{s}(34\mathrm{GeV})=0.165\pm 0.022`$ and in $`\alpha _\mathrm{s}(31.6\mathrm{GeV})=0.158\pm 0.019`$ . Combining the two gives $`\alpha _\mathrm{s}(34\mathrm{GeV})=0.160\pm 0.019(\mathrm{exp}.+\mathrm{sys}.)`$ or $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})=0.133\pm 0.013`$.
A re-analysis of PETRA and TRISTAN data, in the c.m. energy range of 20 to 65 GeV, took better account of higher order QED and electroweak corrections and used the mass of the $`\mathrm{Z}^0`$ measured from LEP experiments. In NLO QCD, this study resulted in $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})=0.124\pm 0.021`$, where the error includes experimental and systematical uncertainties, added in quadrature. Re-applying the NNLO QCD correction this gives
$$\alpha _\mathrm{s}(42.4\mathrm{GeV})=0.175\pm 0.028\mathrm{or}\alpha _\mathrm{s}(M_{\mathrm{Z}^0})=0.126\pm 0.022,$$
which is included in the final summary section 7. A more recent determination of $`\alpha _\mathrm{s}`$ from the total $`\mathrm{e}^+\mathrm{e}^{}`$ hadronic cross section measured by CLEO at $`E_{cm}=10.52`$ GeV resulted in
$$\alpha _\mathrm{s}(10.52\mathrm{GeV})=0.130_{0.029}^{+0.021}$$
and is also added to the final summary.
On top and around the $`\mathrm{Z}^0`$ resonance, the LEP experiments have collected large statistics data samples which allow accurate determinations of $`\alpha _\mathrm{s}`$. At the $`\mathrm{Z}^0`$ pole, equation 21 must be modified according to the dominant electroweak couplings of the $`\mathrm{Z}^0`$ to the quarks, resulting in NNLO QCD predictions of the hadronic decay width of the $`\mathrm{Z}^0`$ . Quark mass corrections in NLO and partly to NNLO , non-factorizable electroweak and QCD corrections , top quark effects and other electroweak corrections apply in addition, rendering numerical expressions for $`R_\mathrm{Z}=\mathrm{\Gamma }(\mathrm{Z}^0\mathrm{hadrons})/\mathrm{\Gamma }(\mathrm{Z}^0\mathrm{leptons})`$ rather complicated; see e.g. ref. for a comprehensive report about QCD corrections on $`R_\gamma `$ and $`R_\mathrm{Z}`$.
In this review, a recent parametrisation of the NNLO QCD prediction of $`R_\mathrm{Z}`$ , including all known corrections indicated above, is applied to determine $`\alpha _\mathrm{s}`$ from $`R_\mathrm{Z}`$. This parametrisation, as given in equation 15, is also used by the LEP collaborations in their combined studies of electroweak precision data . The coefficients given in equation 15 are calculated for a Higgs mass $`M_H`$ of 300 GeV, a top quark mass $`M_t`$ of 174.1 GeV and for $`M_{\mathrm{Z}^0}`$ = 91.19 GeV. With the latest combined LEP result, $`R_\mathrm{Z}=20.768\pm 0.024`$ , this gives $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})=0.124\pm 0.004(\mathrm{exp}.)`$.
Errors in $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})`$ from different sources, namely from changes of $`M_{\mathrm{Z}^0}`$ and $`M_t`$ within their current uncertainties, from $`M_H`$ in the range of 100 GeV (the current experimental lower limit from LEP) to 1000 GeV, and from renormalization scale uncertainties, varying $`\mu `$ from 1/4 to 4$`M_{\mathrm{Z}^0}`$, are given in Table 2. The estimate of the $`scheme`$ dependence, which in NNLO must be considered in addition to the renormalization $`scale`$ dependence, was taken from reference . Neglecting the tiny errors from $`M_{\mathrm{Z}^0}`$ and from $`M_t`$, the overall theoretical uncertainty on $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})`$ from $`R_\mathrm{Z}`$ is estimated to be $`\pm 0.002`$ from $`M_H`$ and $`M_t`$ and $`{}_{0.001}{}^{}{}_{}{}^{+0.003}`$ from higher order QCD contributions. Therefore, the final result from $`R_\mathrm{Z}`$, to be added to the overall summary, is
$$\alpha _\mathrm{s}(M_{\mathrm{Z}^0})=0.124\pm 0.004(\mathrm{exp}.)\pm 0.002(M_H,M_t)_{0.001}^{+0.003}(\mathrm{QCD}).$$
Note that the precision of this result crucially depends on the assumption that the predictions of the electroweak Standard Model are strictly valid; small deviations from these predictions can produce large systematic shifts of $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})`$ from $`R_\mathrm{Z}`$. The LEP data, however, are in excellent overall agreement with the Standard Model predictions , such that the uncertainties quoted above seem realistic.
A combined fit to all data from LEP-I and LEP-II, including the measurement of the mass of the W boson, $`M_W`$, and all measured cross sections and asymmetries, instead of $`R_\mathrm{Z}`$ alone, results in $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})=0.120\pm 0.003(\mathrm{exp}.)`$; a fit to all data including those from $`p\overline{p}`$ collider and lepton-nucleon scattering experiments gives $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})=0.118\pm 0.003(\mathrm{exp}.)`$ . These fits include simultaneous determinations of $`M_H`$ and $`M_t`$, such that only the QCD uncertainty must be added. Because of their complicated nature, however, these results are only mentioned for completeness, but are not taken to replace the above value of $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})`$ from $`R_\mathrm{Z}`$ for the final summary.
### 4.4 $`\alpha _\mathrm{s}`$ from $`\tau `$ decays
An important quantity to determine $`\alpha _\mathrm{s}`$ from measurements at small energy scales is the normalized hadronic branching fraction of $`\tau `$ lepton decays,
$$R_\tau =\frac{\mathrm{\Gamma }(\tau \mathrm{hadrons}\nu _\tau )}{\mathrm{\Gamma }(\tau \mathrm{e}\nu _e\nu _\tau )},$$
(22)
which is predicted to be
$$R_\tau =3.058(1.001+\delta _{\mathrm{pert}}+\delta _{\mathrm{nonpert}}).$$
(23)
Here, $`\delta _{\mathrm{pert}}`$ and $`\delta _{\mathrm{nonpert}}`$ are perturbative and nonperturbative QCD corrections; $`\delta _{\mathrm{pert}}`$ was calculated to complete $`𝒪(\alpha _\mathrm{s}^3)`$ and is of similar structure to the one for $`R_\mathrm{Z}`$ :
$$\delta _{\mathrm{pert}}=\frac{\alpha _\mathrm{s}(M_\tau )}{\pi }+5.20\left(\frac{\alpha _\mathrm{s}(M_\tau )}{\pi }\right)^2+26.37\left(\frac{\alpha _\mathrm{s}(M_\tau )}{\pi }\right)^3,$$
(24)
and parts of the $`4^{th}`$ order coefficient are also known . Based on the operator product expansion (OPE) , the nonperturbative correction was estimated to be small , $`\delta _{\mathrm{nonpert}}=0.007\pm 0.004`$.
The most comprehensive determinations of $`\alpha _\mathrm{s}`$ from $`\tau `$ decays are based on recent studies from LEP, making use of the large data statistics available at LEP-I. The ALEPH and the OPAL Collaborations presented measurements of the vector and the axial-vector contributions to the differential hadronic mass distributions of $`\tau `$ decays, which allow simultaneous determination of $`\alpha _\mathrm{s}`$ and of the nonperturbative corrections (in terms of the OPE). These corrections were found to be small and to largely cancel in the total sum of $`R_\tau `$, in good agreement with the theoretical estimates.
The final results of $`\alpha _\mathrm{s}(M_\tau )`$ are listed in Table 3, obtained for different variants of the NNLO QCD predictions: fixed order perturbation theory (FOPT) , contour improved perturbation theory (CIPT) , expressing $`\delta _{\mathrm{pert}}`$ by contour integrals in the complex $`s`$-plane, and renormalon chain improved perturbation theory (RCPT) , where leading terms of the $`\beta `$-functions are resummed by inserting so-called renormalon chains, i.e. gluon lines with many loop insertions. The two groups agree well on their $`\alpha _\mathrm{s}`$ results and — approximately — on the estimated uncertainties, however different theoretical approaches give systematic differences in $`\alpha _\mathrm{s}`$. The FOPT results seem to represent the mean of these theoretical approaches. Therefore the final result from $`R_\tau `$, to be included in the final summary section 7, is taken to be
$$\alpha _\mathrm{s}(M_\tau )=0.323\pm 0.005(\mathrm{exp}.)\pm 0.030(\mathrm{theo}.),$$
where the average between ALEPH and OPAL was taken and an additional error of $`\pm 0.020`$, accommodating the shift between different theoretical approaches, was added in quadrature<sup>7</sup><sup>7</sup>7In another study of the impact of different theoretical variants of the NNLO QCD expectation for $`R_\tau `$, it was concluded that the overall theoretical uncertainty on $`\alpha _\mathrm{s}(M_\tau )`$ is, at best, $`\pm 0.05`$ , which is larger than the respective error quoted above. to the theoretical uncertainties given by ALEPH and OPAL. When extrapolated to the energy scale $`M_{\mathrm{Z}^0}`$, using the 4-loop $`\beta `$-function and 3-loop matching at the bottom quark pole mass, this results in $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})=0.1181\pm 0.0007(\mathrm{exp}.)\pm 0.0030(\mathrm{theo}.)`$.
## 5 Results from hadron colliders
Significant determinations of $`\alpha _\mathrm{s}`$ from hadron collider data are obtained from $`b\overline{b}`$ production cross sections, from prompt photon production, from inclusive jet production cross sections and from the ratio of $`W+(1jet)`$ and $`W+(0jet)`$ production cross section, all of which are calculated in complete NLO QCD. The latter topic, $`W+jet`$ production, will not be discussed further here, because early measurements , giving $`\alpha _\mathrm{s}(M_W)=0.123\pm 0.025`$, were put in question by new analyses showing bad disagreement between data and QCD, see e.g. .
Hard scattering cross sections initiated by two hadrons with four-momenta $`P_1`$ and $`P_2`$ can be written as
$$\sigma (P_1,P_2)=\underset{i,j}{}dx_1dx_2f_i(x_1,\mu _f^2)f_j(x_2,\mu _f^2)\sigma _{ij}(p_1,p_2,\alpha _\mathrm{s}(\mu ^2),Q^2/\mu _f^2),$$
(25)
where $`p_1=x_1P_1`$ and $`p_2=x_2P_2`$ are the momenta of the interacting partons, $`f_i`$ and $`f_j`$ are the QCD quark and gluon distributions, $`Q`$ is the characteristic scale of the hard scattering, and $`\sigma _{ij}`$ is the short-distance cross section of the hard scattering between partons of type $`i`$ and $`j`$. This parametrization again is based — as in DIS — on the assumption of factorization between the short- and the long-range regimes of the scattering process, the transition between both being defined at the factorization scale $`\mu _f`$.
In general, determinations of $`\alpha _\mathrm{s}`$ from hadron-hadron-collisions are less precise than those from $`\mathrm{e}^+\mathrm{e}^{}`$ annihilation or deep inelastic scattering processes, due to larger uncertainties associated with incoming hadrons: parton distributions and soft remnants from spectator partons, which do not participate in the hard scattering process, substantially add to the overall uncertainties.
### 5.1 $`\alpha _\mathrm{s}`$ from $`b\overline{b}`$ cross sections
The first theoretically well-defined determination of $`\alpha _\mathrm{s}`$ from a purely hadronic production process was presented by the UA1 collaboration , obtained from a measurement of the cross section of the process $`p\overline{p}b\overline{b}X`$ for which NLO QCD predictions exist . $`b`$-quarks were detected through their semileptonic decays into muons which yield high transverse momenta $`p_T`$ with respect to the beam axis. A strong correlation between the decay muons and the original $`b`$-quark allows determination of the $`b`$-quark production cross section without application of a jet algorithm, thus avoiding systematic effects from spectator partons (or the “underlying event”). The LO QCD contribution is of $`𝒪(\alpha _\mathrm{s}^2)`$ and leads to a back-to-back (in azimuth) $`b\overline{b}`$ configuration; virtual corrections to this state and the emission of a third (gluon) jet are of $`𝒪(\alpha _\mathrm{s}^3)`$, predicted in NLO QCD.
Comparison of the measured cross-section for 2-body final states with NLO QCD predictions yielded
$$\alpha _\mathrm{s}(20\mathrm{GeV})=0.145_{0.010}^{+0.012}(\mathrm{exp}.)_{0.016}^{+0.013}(\mathrm{theo}.),$$
where the theoretical error includes uncertainties due to different sets of structure functions, renormalization/factorization scale uncertainties and the b-quark mass .
### 5.2 $`\alpha _\mathrm{s}`$ from prompt photon production
Production of high transverse momentum direct (“prompt”) photons in hadron collisions is well suited to test perturbative QCD and to determine $`\alpha _\mathrm{s}`$, because photons, in contrast to quarks, do not hadronize and their energies and directions can — in general — be measured with higher accuracy than those of hadron jets. In leading order, however, prompt photon production is of $`𝒪(\alpha \alpha _\mathrm{s})`$, compared to $`𝒪(\alpha _\mathrm{s}^2)`$ for hadron jets, and therefore suffers from relatively small production cross sections. In addition, there is a sizeable background of photons from $`\pi ^0`$ and $`\eta `$ decays, such that quantitative tests of QCD from prompt photon production are not trivial, from an experimental point of view.
Using complete $`𝒪(\alpha \alpha _\mathrm{s}^2)`$ QCD predictions , the UA6 collaboration determined $`\alpha _\mathrm{s}`$ from a measurement of the cross sections difference $`\sigma (p\overline{p}\gamma X)\sigma (pp\gamma X)`$ , where the poorly known contributions of the sea quarks and the gluon distributions in the proton cancel. The result is
$$\alpha _\mathrm{s}(24.3\mathrm{GeV})=0.135\pm 0.006(\mathrm{exp}.)_{0.005}^{+0.011}(\mathrm{theo}.),$$
where the theoretical error includes uncertainties from the scale choice and from variation of the parton distribution functions.
### 5.3 $`\alpha _\mathrm{s}`$ from inclusive jet cross sections
The definition and reconstruction of particle jets in hadron collisions traditionally has followed other strategies than those used in $`\mathrm{e}^+\mathrm{e}^{}`$ annihilation. In hadron collisions, so-called cone jet finders are employed which allow particles, ideally those which originate from the proton remnants, not to be associated with any of the reconstructed jets — in contrast to the clustering algorithms used in $`\mathrm{e}^+\mathrm{e}^{}`$ annihilation where $`all`$ particles are assigned to jets; c.f. section 4.1. Nowadays, almost all of the jet studies at hadron collider experiments follow the “Snowmass” definition of jets . Here, jets are defined by concentrations of transverse energy $`E_T=|E\mathrm{sin}\theta |`$ in cones of radius
$$R=\sqrt{(\mathrm{\Delta }\eta )^2+(\delta \varphi )^2},$$
where $`\eta =\mathrm{ln}\mathrm{tan}(\theta /2)`$ is the pseudorapidity, $`\varphi `$ is the azimuthal and $`\theta `$ is the polar angle of a particle or an energy cluster in the calorimeter of the detector, measured w.r.t. the point of beam crossing.
Giele, Glover and Yu determined $`\alpha _\mathrm{s}`$ by fitting the NLO parton level Monte Carlo JETRAD , which is based on the QCD matrix elements of reference and the MRSA’ parton density functions , to the single inclusive jet cross sections measured by CDF for a cone size of $`R=0.7`$. In each bin of $`E_T`$, $`\alpha _\mathrm{s}`$ is determined for the scale choice $`\mu E_T`$. The resulting values of $`\alpha _\mathrm{s}(E_T)`$ are displayed in figure 9, where the error bars represent the combined statistical and theoretical uncertainties; the size of an additional, overall (experimental) systematic error is also indicated. The results are in good agreement with the QCD expectation of the running $`\alpha _\mathrm{s}`$. A corresponding overall QCD fit, in NLO, gives
$$\alpha _\mathrm{s}(M_{\mathrm{Z}^0})=0.121\pm 0.001(\mathrm{stat}.)\pm 0.008(\mathrm{sys}.)\pm 0.005(\mathrm{theo}.)\pm 0.002(\mathrm{pdf})$$
where the theoretical error was obtained from a variation of $`\mu `$ between 0.5 and $`2E_T`$, and the last error represents the uncertainty from using different parton density functions.
This study was basically meant to introduce a general method and possibility to determine $`\alpha _\mathrm{s}`$ from hadron collider jet cross sections, rather than to present a complete and final analysis. For instance, the parton density functions used in this study were extracted from (mainly deep inelastic scattering) data for a fixed input value of $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})=0.113`$, while a coherent determination of $`\alpha _\mathrm{s}`$ should allow $`\alpha _\mathrm{s}`$ to vary in the density functions, too. Although the data statistics — and hopefully also the experimental systematic errors — must have improved significantly with respect to the previous data sample of reference , and although there exist new parametrizations of density functions for different input values of $`\alpha _\mathrm{s}`$, no update of the measurement of $`\alpha _\mathrm{s}`$ from hadron collider jets was published so far. Therefore the above result is retained in the final summary — keeping in mind, however, that the quoted uncertainties constitute lower limits rather than a complete assessment of the overall error.
## 6 Results from heavy quarkonia decays and masses
The mass spectra and partial decay widths of heavy quark-antiquark bound states are a good testing ground for QCD. For quark masses $`m_Q\mathrm{\Lambda }_{QCD}`$, the short- and long-range effects on the decay widths can be factorized, and the short-range part can be calculated by perturbative QCD. Mass splittings of heavy quarkonia states can be calculated using nonperturbative methods like lattice gauge theory, which also provide means to extract values of $`\alpha _\mathrm{s}`$.
### 6.1 $`\alpha _\mathrm{s}`$ from quarkonia decay branching fractions
Partial decay widths of heavy quarkonia, like $`\mathrm{\Gamma }^{\mu \mu }=\mathrm{\Gamma }(Q\overline{Q}\mu ^+\mu ^{})`$, $`\mathrm{\Gamma }^{\gamma gg}`$ and $`\mathrm{\Gamma }^{ggg}`$ are calculated in NLO QCD . The expressions contain the (unknown) radial wave function at the origin, which can be eliminated by forming the ratios
$`R_\mu `$ $``$ $`{\displaystyle \frac{\mathrm{\Gamma }^{ggg}}{\mathrm{\Gamma }^{\mu \mu }}}={\displaystyle \frac{10(\pi ^29)}{9\pi }}{\displaystyle \frac{\alpha _\mathrm{s}^3(\mu ^2)}{\alpha ^2}}\left(1+\left[0.46.3\mathrm{ln}\left({\displaystyle \frac{m_b^2}{\mu ^2}}\right)\right]{\displaystyle \frac{\alpha _\mathrm{s}(\mu ^2)}{\pi }}\right)`$
$`R_\gamma `$ $``$ $`{\displaystyle \frac{\mathrm{\Gamma }^{\gamma gg}}{\mathrm{\Gamma }^{ggg}}}={\displaystyle \frac{4}{5}}{\displaystyle \frac{\alpha }{\alpha _\mathrm{s}(\mu ^2)}}\left(1\left[2.62.1\mathrm{ln}\left({\displaystyle \frac{m_b^2}{\mu ^2}}\right)\right]{\displaystyle \frac{\alpha _\mathrm{s}(\mu ^2)}{\pi }}\right).`$ (26)
An early, comprehensive fit of $`\alpha _\mathrm{s}`$ from the $`J/\mathrm{\Psi }`$ and the $`\mathrm{{\rm Y}}`$ branching ratios was presented in reference , resulting in $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})=0.113_{0.005}^{+0.007}`$. This study included estimates of the higher order QCD uncertainties and of relativistic corrections.
More recently, sum rules for the $`\mathrm{{\rm Y}}`$ system were analysed with resummation of QCD Coulomb effects which are responsible for the growth of the perturbative coefficients in NLO QCD , resulting in $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})=0.118\pm 0.006`$. Similar methods were already applied in ealier studies but have led to conflicting results. Application of NNLO QCD predictions resulted in
$$\alpha _\mathrm{s}(M_{\mathrm{Z}^0})=0.118\pm 0.006,$$
which is included in the final summary of $`\alpha _\mathrm{s}`$. It should be noted that further studies of $`\mathrm{{\rm Y}}`$ sum rules in NNLO QCD obtained similar results of $`\alpha _\mathrm{s}`$, however with a much more conservative estimate of the theoretical uncertainties — the resulting, large uncertainties of $`\alpha _\mathrm{s}`$ may indicate that $`\mathrm{{\rm Y}}`$ sum rules do not appear to provide competitive results of $`\alpha _\mathrm{s}`$.
### 6.2 $`\alpha _\mathrm{s}`$ from lattice calculations of mass splittings
Early determinations of $`\alpha _\mathrm{s}`$ from heavy quarkonia mass splittings were based on quenched approximations of lattice QCD calculations, neglecting light quark flavour loops ($`N_f=0`$). When evolved to the $`\overline{\text{MS}}`$ coupling at the scale $`M_{\mathrm{Z}^0}`$, these methods led to $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})=0.105\pm 0.004`$ , where the error included statistical as well as estimates of systematic errors.
Refined nonrelativistic QCD calculations with $`N_f=0`$ and $`N_f=2`$ allowed extrapolation to the physically correct number of light quarks, $`N_f=3`$, and included advanced (3-loop) perturbative extrapolation from the lattice to the $`\overline{\text{MS}}`$ coupling. A detailed study of various mass splittings of $`\mathrm{{\rm Y}}`$ ($`b\overline{b}`$) states, which are precisely known from corresponding measurements, was presented in . Technically, the lattice calculation for a given value of the “bare” lattice coupling $`\alpha _{lat}`$ gave a value for the dimensionless quantity $`a\mathrm{\Sigma }`$, where $`a`$ is the lattice spacing and $`\mathrm{\Sigma }`$ is the mass splitting of suitable quarkonium states. From this result, divided by the measured value of $`\mathrm{\Sigma }`$, the lattice spacing $`a`$ was obtained. From QCD perturbation theory, the $`\overline{\text{MS}}`$ coupling $`\alpha _\mathrm{s}`$ at the energy scale $`a^1`$ is given as a function of $`\alpha _{lat}`$ . Extraction of $`\alpha _\mathrm{s}(a^1)`$ and extrapolation to $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})`$ resulted in $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})=0.1174\pm 0.0024`$ , where the error included statistical as well as systematic uncertainties which include variations of light quark masses and estimates of truncation errors in the extraction of the $`\overline{\text{MS}}`$ coupling using perturbation theory.
More recently, a study based on similar lattice calculations and the same $`\mathrm{{\rm Y}}`$ mass splittings as in reference , however using a different discretization scheme, derived a lower value, $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})=0.1118\pm 0.0017`$ . This is about three standard deviations smaller than the result from reference , which led to the conclusion that the “true” systematic uncertainty must be three to four times larger than estimated before.
Following this suggestion, the result on $`\alpha _\mathrm{s}`$ from nonperturbative lattice calculations to be considered in the final summary section is chosen to be the average of the two results discussed above, with the overall uncertainty increased by a factor of 3, giving
$$\alpha _\mathrm{s}(M_{\mathrm{Z}^0})=0.115\pm 0.006.$$
## 7 Summing up …
A summary of the $`\alpha _\mathrm{s}`$ measurements discussed in the previous sections is presented in table 6 at the end of this review. The results are given, if applicable, at the relevant energy scale $`Q`$ of the process, $`\alpha _\mathrm{s}(Q)`$, and at the standard “reference” scale of the $`\mathrm{Z}^0`$ mass, $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})`$. The conversion between these two cases was done using the 4-loop QCD expression for the running $`\alpha _\mathrm{s}`$, equation 2.4, with 3-loop matching at the c- and b-quark pole masses of 1.5 and 4.7 GeV, respectively, as discussed in section 2.5. The splitting of the overall uncertainties of $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})`$, $`\mathrm{\Delta }\alpha _\mathrm{s}(M_{\mathrm{Z}^0})`$, into experimental and theoretical errors is given in the $`5^{th}`$ and the $`6^{th}`$ columns of table 6, and the last column indicates the level of theoretical calculations on which these results are based.
The results for $`\alpha _\mathrm{s}(Q)`$, given in the $`3^{rd}`$ column of table 6, are presented in figure 10, together with fits of the 4-loop QCD prediction for the running $`\alpha _\mathrm{s}`$ (equation 2.4) with 3-loop matching at the quark pole masses. Results which were obtained from data in large ranges of $`Q`$ are not displayed in this figure. The data are in very good agreement with the theoretical expectation, and prove the running of $`\alpha _\mathrm{s}`$ with high significance. The latter point will be analysed in more detail in section 7.2.
The values of $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})`$ are presented in figure 11. Within their assigned total erros, all results agree well with each other and with a weighted average value of $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})=0.1184`$, which is indicated by the vertical line in figure 11. The overall $`\chi ^2`$ of all results with this average is 7.2 for 25 degrees of freedom, which indicates that the individual errors must be highly correlated. Therefore, the determination of the average value of $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})`$ and its overall uncertainty requires special treatment and discussion.
### 7.1 World average $`\overline{\alpha _\mathrm{s}}(M_{\mathrm{Z}^0})`$ and its overall uncertainty $`\mathrm{\Delta }\overline{\alpha _\mathrm{s}}`$
The errors of most $`\alpha _\mathrm{s}`$ results are dominated by theoretical uncertainties, which are estimated using a variety of different methods and definitions. The significance of these nongaussian errors is largely unclear. Furthermore, there are large correlations between different results, due to common theoretical uncertainties, as e.g. for event shape measurements in $`\mathrm{e}^+\mathrm{e}^{}`$ annihilations. Correlations between $`\alpha _\mathrm{s}`$ determinations from different processes, such as DIS and $`\mathrm{e}^+\mathrm{e}^{}`$ annihilations, or between different procedures and observables used within the same class of processes may be present, too. Standard statistical methods therefore do not apply when averaging these results. Several methods are employed to derive an estimate of the average value $`\overline{\alpha _\mathrm{s}}(M_{\mathrm{Z}^0})`$ and its overall uncertainty, $`\mathrm{\Delta }\overline{\alpha _\mathrm{s}}`$. The results are summarized in table 4:
* An error weighted average and an “optimized correlation” error is calculated from the error covariance matrix, assuming an overall correlation factor between the total errors of all measurements. This factor is adjusted so that the overall $`\chi ^2`$ equals one per degree of freedom . The resulting mean values, overall uncertainties and optimized correlation factors are given in columns 3 to 5 of Table 4, respectively.
* For illustrative purposes only, an overall error is calculated assuming that all measurements are entirely uncorrelated and all quoted errors are gaussian. The results are displayed in column 6.
* The simple, unweighted root mean squared of the mean values of all measurements is calculated and shown in column 7, labelled “simple rms”.
* Assuming that each result of $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})`$ has a rectangular-shaped rather than a gaussian probability distribution, the resulting weights (the inverse of the square of the total error) are summed up in a histogram, and the resulting $`rms`$ of that distribution is quoted as “rms box” .
All of these methods have certain advantages but also include inherent problems. The “simple rms” indicates the scatter of all results around their common mean, but does not depend on the individual errors quoted for each measurement. The “box rms”, which takes account of the errors and of their nongaussian nature, was criticized as being too conservative an estimate of the overall uncertainty of $`\alpha _\mathrm{s}`$. The “optimized correlation” method — in the absence of a detailed knowledge of these correlations — over-simplifies by the assumption of one overall correlation factor. Moreover, if correlations are present, $`\chi ^2`$ does not have the same mathematical and probabilistic meaning as in the case of uncorrelated data. In the extreme, $`\chi ^2`$ may even be negative.
With these reservations in mind, all four methods do provide some estimate of $`\mathrm{\Delta }\overline{\alpha _\mathrm{s}}`$. Apart from the method to calculate $`\mathrm{\Delta }\overline{\alpha _\mathrm{s}}`$, the result also depends on the significance of the data included in the averaging process: in all cases except the “uncorrelated” error estimate, $`\mathrm{\Delta }\overline{\alpha _\mathrm{s}}`$ is largest if all data are included, and tends to smaller values if the averaging is restricted to results with errors $`\mathrm{\Delta }\alpha _\mathrm{s}\mathrm{\Delta }\alpha _\mathrm{s}^{(max)}`$, i.e. if only the most significant results are taken into account. This is demonstrated in rows 1 to 5 of table 4. The dependence of $`\mathrm{\Delta }\overline{\alpha _\mathrm{s}}`$ on $`\mathrm{\Delta }\alpha _\mathrm{s}^{(max)}`$ is illustrated in figure 12b, where the rightmost results correspond to the first row of table 4 which includes all 26 $`\alpha _\mathrm{s}`$ measurements summarized in table 6. Figure 12a illustrates the distribution of $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})`$ and the total errors of all these measurements.
On first sight it seems logical to restrict the determination of $`\overline{\alpha _\mathrm{s}}(M_{\mathrm{Z}^0})`$, and especially of $`\mathrm{\Delta }\overline{\alpha _\mathrm{s}}`$, to the most significant data, if the inclusion of less significant measurements only enlarges $`\mathrm{\Delta }\overline{\alpha _\mathrm{s}}`$. Taken to the extreme, one may even be tempted to take the one result with the smallest quoted error as the final world average value of $`\overline{\alpha _\mathrm{s}}(M_{\mathrm{Z}^0})`$. However, the errors on $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})`$ estimated in individual studies are very often lower limits because unknown and additional systematic effects can only increase the total error. In some cases, small total errors can be due to fortunate coincidences, to ignorance, over-optimism and/or neglect of certain error sources. In order to ensure and to test consistency of the results, it is therefore mandatory not to rely on a single determination alone but to possibly include several different, significant measurements in the averaging process.
The averaging procedures discussed so far include results which are based on different orders and types of QCD calculations, namely on lattice gauge theory and on perturbation theory in NLO, resummed NLO and in NNLO. Comparing and averaging these different types of results is nevertheless justified because all of them include estimates of the relevant uncertainties, such that they should be compatible with a common average within their given errors. In order to achieve the highest precision and confidence in a combined average value $`\overline{\alpha _\mathrm{s}}(M_{\mathrm{Z}^0})`$, however, it may be beneficial not to include results which are, for instance, based on calculations in lower than the maximum available order of perturbation theory.
In this sense, the averaging procedure is applied only to those results which are based on complete NNLO QCD perturbation theory. Only very recently the number, the precision and the diversity of such measurements reached a level where such a restriction still provides a solid basis for a meaningful averaging process which includes sufficient freedom for internal consistency checks. Table 6 contains 9 measurements which are based on NNLO QCD; the results of averaging these are given in rows 6 to 9 of table 4, for different subsamples defined by selecting measurements with a total error $`\mathrm{\Delta }\alpha _\mathrm{s}\mathrm{\Delta }\alpha _\mathrm{s}^{(max)}`$. The combined errors, $`\mathrm{\Delta }\overline{\alpha _\mathrm{s}}`$, using the optimized correlation method, are displayed in figure 12b.
Requiring $`\mathrm{\Delta }\alpha _\mathrm{s}0.008`$ rejects those $`\alpha _\mathrm{s}`$ measurements which are dominated by large experimental uncertainties, leaving the 5 most significant NNLO results from which one obtains, using the optimized correlation method,
$$\overline{\alpha _\mathrm{s}}(M_{\mathrm{Z}^0})=0.1184\pm 0.0031.$$
This value is taken as the currently best estimate of the world average of $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})`$. According to equations 2.4 and 8, in 4-loop approximation and with 3-loop threshold matching at $`M_b=4.7`$ GeV, this corresponds to $`\mathrm{\Lambda }_{\overline{\text{MS}}}^{(N_f=5)}=\left(213_{35}^{+38}\right)`$ MeV and $`\mathrm{\Lambda }_{\overline{\text{MS}}}^{(N_f=4)}=\left(296_{44}^{+46}\right)`$ MeV.
Each of the 26 single results summarized in table 6 is compatible with this average, to within about one standard deviation of its assigned uncertainty or less. In order to investigate possible systematic deviations or trends between and within subsets of these measurements, the averaging procedure was repeated for deep inelastic scattering data, for $`\mathrm{e}^+\mathrm{e}^{}`$ annihilation and for hadron collider data alone, see rows 10 to 12 of table 4, and for 3 energy ranges as shown in rows 13 to 15. Within the overall uncertainties (e.g. those derived from the optimized correlation method), all results agree well with each other and with the world average value derived above. Small but insignificant systematic differences between DIS and $`\mathrm{e}^+\mathrm{e}^{}`$ results and between those obtained from low and from high energy data may be visible; these may well be accidental or may be caused, for example, by different methods of treating renormalization and factorization scales.
### 7.2 Quantifying the running of $`\alpha _\mathrm{s}`$: determination of $`\beta _0`$, of $`N_c`$ and of the functional $`Q`$-dependence
The precision of the experimental results, the large energy range for which data are available, and the good agreement of the measurements with the QCD expectation of a running $`\alpha _\mathrm{s}`$, see figure 10, suggest actually fitting the functional form of the energy dependence from the data. For this purpose, a subset of data is selected which ensures maximal independence between the data chosen and which is based on the most significant measurements in the largest possible energy range (c.f. table 6 and figure 10), namely:
* $`\alpha _\mathrm{s}`$ from $`\tau `$ decays: $`\alpha _\mathrm{s}(1.778\mathrm{GeV})=0.323\pm 0.030`$;
* $`\alpha _\mathrm{s}`$ from moments of $`F_2`$: $`\alpha _\mathrm{s}(2.96\mathrm{GeV})=0.252\pm 0.011`$;
* $`\alpha _\mathrm{s}`$ from scaling violations of $`F_3`$: $`\alpha _\mathrm{s}(5.0\mathrm{GeV})=0.214\pm 0.021`$;
* $`\alpha _\mathrm{s}`$ from the hadronic width of the $`\mathrm{Z}^0`$: $`\alpha _\mathrm{s}(91.2\mathrm{GeV})=0.124\pm 0.005`$;
* $`\alpha _\mathrm{s}`$ from hadronic event shapes: $`\alpha _\mathrm{s}(189\mathrm{GeV})=0.110\pm 0.004`$.
All these results are based on complete NNLO QCD predictions — with the exception of the latter which includes resummed NLO calculations — and represent the most precise $`\alpha _\mathrm{s}`$ results of their class. The nature of their experimental and theoretical uncertainties ensures a maximum of independence between them.
Table 5 summarizes some of the functional fits which were performed with these selected data. The general idea was to obtain qualitative measures for the functional form of the energy dependence of $`\alpha _\mathrm{s}`$, as well as numeric fit values for the coefficient $`\beta _0`$ of the QCD $`\beta `$-function, see equation 3, and eventually for one of the main parameters of QCD, the number of colour degrees of freedom $`N_cC_A`$ which is an integral part of $`\beta _0`$:
$$\beta _0=\frac{11C_A2N_f}{12\pi }.$$
(27)
Because of its simplicity and in order to avoid too many QCD-inspired biases, the functional form of the leading order QCD expression for the running $`\alpha _\mathrm{s}`$ ( c.f. equation 2.4), without quark threshold matching, was fitted to the data. Higher order corrections to the running affect $`\alpha _\mathrm{s}`$ at the smallest energy scales, $`QM_\tau `$, by less than $`10\%`$, which is well within the error of $`\alpha _\mathrm{s}`$ in this energy regime.
The results of this fit are given in the first row of table 5, with $`B1/\beta _0=1.62\pm 0.09`$, and $`C\mathrm{\Lambda }=0.125\pm 0.032`$ (in units of GeV). The $`\chi ^2`$ of this fit is 0.63 for 3 degrees of freedom, which corresponds to a probability of 0.89. Within QCD, with $`N_c=3`$ and $`N_f=5`$, $`1/\beta _0=1.64`$, which is in excellent agreement with the fit value of the parameter $`B`$. $`\mathrm{\Lambda }=0.125`$ GeV corresponds, in LO QCD, to $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})=0.124`$ and to $`\alpha _\mathrm{s}(M_\tau )=0.31`$ (without threshold matching and with $`N_f=5`$ throughout). From the fit value of $`B1/\beta _0`$ one can derive, according to equation 27 and for $`N_f=5`$ quark flavours,
$$N_c=3.03\pm 0.12,$$
which is an excellent and precise verification of the QCD group structure which implies $`N_c=3`$. Alternatively, because this functional fit actually constrains the ratio $`B/n`$, where $`n`$ is the power of $`Q/C`$ inside the logarithm, the fit determines, when $`B`$ is fixed to its QCD value of 1.64, that $`n=2.03\pm 0.12`$.
Other functional forms of the energy dependence of $`\alpha _\mathrm{s}`$, which are not predicted by any consistent theory but which are added to demonstrate the significance of the logarithmic decrease of $`\alpha _\mathrm{s}(Q^2)`$, are fitted and presented in the last two rows of table 5: a straight-line fit is clearly excluded by the unacceptable $`\chi ^2`$ of 78 for 3 d.o.f., while a $`1/Q`$ energy dependence is disfavoured but cannot be excluded with the current data.
## 8 Conclusions
This review of experimental determinations of the coupling parameter of the Strong Interaction, $`\alpha _\mathrm{s}`$, summarized a topical and still ongoing field of activities in the high energy physics community. Due to impressive theoretical developments and experimental efforts during the past 10 years, $`\alpha _\mathrm{s}`$ could be determined from a large variety of physical observables and processes with steadily increasing precision. Current state-of-the-art measurements reach experimental uncertainties down to a few per cent, and theoretical calculations which are complete up to NNLO perturbative QCD obey higher order uncertainties of a few per cent in $`\alpha _\mathrm{s}(M_{\mathrm{Z}^0})`$, too. While not all of the available studies reach both these levels of precision, the most significant determinations of $`\alpha _\mathrm{s}`$, based on complete NNLO QCD calculations, can be summarized to a new world average value of
$$\overline{\alpha _\mathrm{s}}(M_{\mathrm{Z}^0})=0.1184\pm 0.0031.$$
The overall uncertainty is slightly larger than the smallest total errors quoted for some of the measurements, which is due to the following facts:
* The errors of single results contain different estimates of theoretical uncertainties, which are not uniquely defined.
* Many (if not all) results rely on assumptions which ultimately cannot be proven, like factorization between perturbative and nonperturbative effects, the nature and size of the latter, neglect of quark masses, exact realization of the electro-weak theory including the existence of a Standard Model Higgs boson, reaching of the continuum limit in lattice calculations etc.; the quoted theoretical uncertainties must therefore rather be viewed as lower limits instead of complete estimates.
* Many of the results and their theoretical uncertainties are highly correlated, however mostly to an unknown degree. This applies not only to their quoted errors, but also to the unproven assumptions mentioned under the previous item.
A realistic determination of the world average value of $`\alpha _\mathrm{s}`$ must therefore be based on comparing several different measurements of similar precision. It must account for possible correlations between them and allow for underestimates and fortunate cancellations of errors in some cases. In the absence of any “exact” method to account for all these effects, the “optimized correlation” method was used to calculate the world average $`\overline{\alpha _\mathrm{s}}(M_{\mathrm{Z}^0})`$ and its overall remaining uncertainty as given above. The total error depends on the averaging method chosen - alternative procedures or preselections of data resulted in values ranging from $`\pm 0.0022`$ to $`\pm 0.0057`$.
All values of $`\alpha _\mathrm{s}`$ summarized in table 6 agree well, within their quoted uncertainties, with the world average. Averages obtained from various subsamples of these data, like those from deep inelastic scattering, from $`\mathrm{e}^+\mathrm{e}^{}`$ annihilations and from hadron colliders, as well as subsamples from different energy ranges, do not show significant biases or shifts. In fact, the measured energy dependence of $`\alpha _\mathrm{s}(Q^2)`$ is in excellent agreement with the QCD expectation and significantly proves the running of $`\alpha _\mathrm{s}`$.
Functional fits to data in the energy range from 1.78 to 189 GeV provide evidence for a logarithmic (in contrast to e.g. a linear) energy dependence, and the fitted logarithmic slope — if interpreted in terms of $`\beta _0`$, the LO coefficient of the QCD $`\beta `$-function — provides an accurate value for the number of colour degrees of freedom, $`N_c=3.03\pm 0.12`$. This is not actually a measurement of $`N_c`$, because many of the $`\alpha _\mathrm{s}`$ determinations are based on theoretical predictions of observables which inherently include the QCD value of $`N_c=3`$. Nevertheless, the fit result of $`N_c`$ from the measured energy dependence of $`\alpha _\mathrm{s}`$ constitutes an important consistency check between data and QCD.
The total uncertainty of $`\overline{\alpha _\mathrm{s}}(M_{\mathrm{Z}^0})`$ quoted above is $`2.6\%`$. This precision is a remarkable success. The error on $`\overline{\alpha _\mathrm{s}}(M_{\mathrm{Z}^0})`$, however, is much larger than those on other fundamental “constants” of nature, like the fine structure constant $`\alpha `$, the weak mixing angle $`\mathrm{sin}^2\theta _W`$ or the gravitational constant .
Any further reduction of the uncertainties on $`\alpha _\mathrm{s}`$ will require large theoretical as well as experimental efforts:
* New and higher order QCD calculations at least for some observables will be mandatory to understand, specify and possibly reduce theoretical uncertainties; most wishful candidates would be complete NNLO calculations for jet rates and hadronic event shapes in $`\mathrm{e}^+\mathrm{e}^{}`$ annihilation and in DIS, which are well understood experimentally.
* The precision and theoretical understanding of parton density functions and their correlations with extracted values of $`\alpha _\mathrm{s}`$ must be significantly improved. Future high statistics and high energy runs of the Tevatron hadron collider will hopefully provide this important input. New theoretical developments to understand and predict structure functions from basic principles, for instance through application of classical string theory, see e.g. reference , are mandatory.
* So far, the neglect of finite (heavy) quark masses in almost all higher order QCD calculations potentially limits the reliability especially of results based on data which are close to these thresholds. While phase-space effects can be experimentally examined, only complete higher order QCD calculations for massive quarks, see e.g. reference , can close this gap of potentially large systematic uncertainties.
* The rôle of nonperturbative, long distance effects and the application of factorization in many processes and $`\alpha _\mathrm{s}`$ determinations must be further investigated and understood. Power corrections replacing or supplementing hadronization models, see e.g. references , or higher twist corrections from lattice QCD, see e.g. reference , are promising new developments which have the potential for new, significant insights and improvements.
The physics of hadronic interactions at high energy colliders is rich and colourful. Much has been achieved in the past, however there are still many fundamental and open questions. Further and more precise determinations of the strong coupling parameter $`\alpha _\mathrm{s}`$ will continue to concern and employ many motivated scientists in the future.
## Acknowledgments
I am grateful to O. Biebel, A. Hoang, A.L. Kataev, J.H. Kühn, B. Webber and P. Weisz for many interesting discussions, suggestions and comments.
|
warning/0004/cond-mat0004288.html
|
ar5iv
|
text
|
# Bosonization Rules for Electron-Hole Systems - II
## I Bilinear Sea-boson Correspondence
Here we write down the formal correspondence between Fermi bilinears and sea-bosons suitably generalised to two-component systems. The correspondence presented here is necessarily approximate but one that conforms to the spirit of the random-phase approximation, which in the two-component system corresponds to the exciton approximation. However, just as in the one-component case, new physics could be extracted by suitably generalising the notion of the random-phase approximation, we find that here too we may extract new physics by generalising the notion of the exciton approximation so that one goes beyond the dilute limit and includes experimentally relevent situations such as those in which a large number of real carriers are created by external fields. The discussion presented here follows closely the discussion in the response to the comment on our previous work. Let us write the Fermi bilinear sea-boson correspondence with spin. ($`𝐪0`$)
$`c_{𝐤+𝐪/2,\sigma }^{}c_{𝐤𝐪/2,\sigma ^{^{}}}=\mathrm{\Lambda }_{𝐤,\sigma }(𝐪,\sigma ^{^{}})a_{𝐤,\sigma }(𝐪,\sigma ^{^{}})+a_{𝐤,\sigma ^{^{}}}^{}(𝐪,\sigma )\mathrm{\Lambda }_{𝐤,\sigma ^{^{}}}(𝐪,\sigma )`$
$`+{\displaystyle \underset{𝐪_1\sigma _1}{}}a_{𝐤+𝐪/\mathrm{𝟐}𝐪_\mathrm{𝟏}/\mathrm{𝟐}\sigma _1}^{}(𝐪_\mathrm{𝟏}\sigma )a_{𝐤𝐪_\mathrm{𝟏}/\mathrm{𝟐}\sigma _1}(𝐪_\mathrm{𝟏}𝐪\sigma ^{^{}})`$
$$\underset{𝐪_1\sigma _1}{}a_{𝐤𝐪/\mathrm{𝟐}+𝐪_\mathrm{𝟏}/\mathrm{𝟐}\sigma ^{^{}}}^{}(𝐪_\mathrm{𝟏}\sigma _1)a_{𝐤+𝐪_\mathrm{𝟏}/\mathrm{𝟐}\sigma }(𝐪_\mathrm{𝟏}𝐪\sigma _1)$$
(1)
Here,
$$\mathrm{\Lambda }_{𝐤,\sigma }(𝐪,\sigma ^{^{}})=\sqrt{\overline{n}_{𝐤+𝐪/2,\sigma }(1\overline{n}_{𝐤𝐪/2,\sigma ^{^{}}})}$$
(2)
Let us make the following identifications,
$$c_𝐤=c_𝐤$$
(3)
$$c_𝐤=d_𝐤^{}$$
(4)
Taking a cue from the one-component case let us now argue that( for both $`𝐪=0`$ and $`𝐪0`$)
$`d_{𝐤𝐪/2}c_{𝐤𝐪/2}\sqrt{(1\overline{n}_e(𝐤𝐪/2))(1\overline{n}_h(𝐤𝐪/2))}a_𝐤(𝐪)`$
$$+\sqrt{\overline{n}_e(𝐤𝐪/2)\overline{n}_h(𝐤𝐪/2)}a_𝐤^{}(𝐪)$$
(5)
$$\mathrm{\Lambda }_1(𝐤,𝐪)=\sqrt{(1\overline{n}_e(𝐤𝐪/2))(1\overline{n}_h(𝐤𝐪/2))}$$
(6)
$$\mathrm{\Lambda }_2(𝐤,𝐪)=\sqrt{\overline{n}_e(𝐤𝐪/2)\overline{n}_h(𝐤𝐪/2)}$$
(7)
The above form in Eq.( 5) automatically satisfies,
$$[d_{𝐤𝐪/2}c_{𝐤𝐪/2},d_{𝐤^{^{}}𝐪^{^{}}/2}c_{𝐤^{^{}}𝐪^{^{}}/2}]=0$$
(8)
Next we would like the following to happen (for both $`𝐪=0`$ and $`𝐪0`$), as it does in the Fermi language,
$`[d_{𝐤𝐪/2}c_{𝐤𝐪/2},c_{𝐤^{^{}}𝐪^{^{}}/2}^{}d_{𝐤^{^{}}𝐪^{^{}}/2}^{}]=`$
$`\delta _{𝐤,𝐤^{^{}}}\delta _{𝐪,𝐪^{^{}}}d_{𝐤^{^{}}𝐪^{^{}}/2}^{}d_{𝐤𝐪/2}\delta _{𝐤𝐪/2,𝐤^{^{}}𝐪^{^{}}/2}c_{𝐤^{^{}}𝐪^{^{}}/2}^{}c_{𝐤𝐪/2}\delta _{𝐤+𝐪/2,𝐤^{^{}}+𝐪^{^{}}/2}`$
$$\delta _{𝐤,𝐤^{^{}}}\delta _{𝐪,𝐪^{^{}}}(1d_{𝐤𝐪/2}^{}d_{𝐤𝐪/2}c_{𝐤𝐪/2}^{}c_{𝐤𝐪/2})$$
(9)
$$(1\overline{n}_e(𝐤𝐪/2))(1\overline{n}_h(𝐤𝐪/2))\overline{n}_e(𝐤𝐪/2)\overline{n}_h(𝐤𝐪/2)=1d_{𝐤𝐪/2}^{}d_{𝐤𝐪/2}c_{𝐤𝐪/2}^{}c_{𝐤𝐪/2}$$
(10)
The expectation value is with respect to the full interacting ground state including(especially including) the external fields that allow for significant real populations to be generated. Let us now argue(again inspired by the one-component system),
$$d_𝐤^{}d_𝐤=n_h^{(0)}(𝐤)\underset{𝐪_1\sigma _1}{}a_{𝐤𝐪_1/2\sigma _1}^{}(𝐪_1)a_{𝐤𝐪_1/2\sigma _1}(𝐪_1)+\underset{𝐪_1\sigma _1}{}a_{𝐤+𝐪_1/2}^{}(𝐪_1\sigma _1)a_{𝐤+𝐪_1/2}(𝐪_1\sigma _1)$$
(11)
and,
$$c_𝐤^{}c_𝐤=n_e^{(0)}(𝐤)+\underset{𝐪_1\sigma _1}{}a_{𝐤𝐪_1/2\sigma _1}^{}(𝐪_1)a_{𝐤𝐪_1/2\sigma _1}(𝐪_1)\underset{𝐪_1\sigma _1}{}a_{𝐤+𝐪_1/2}^{}(𝐪_1\sigma _1)a_{𝐤+𝐪_1/2}(𝐪_1\sigma _1)$$
(12)
Here $`n_h^{(0)}(𝐤)`$ and $`n_e^{(0)}(𝐤)`$ account for possible doping in the system. In other words, the presence of excess charge. These two put together obey the attractive identity(charge conservation),
$$\underset{𝐤}{}c_𝐤^{}c_𝐤\underset{𝐤}{}d_𝐤^{}d_𝐤=Q=\underset{𝐤}{}[n_e^{(0)}(𝐤)n_h^{(0)}(𝐤)]$$
(13)
Now we move on to the off-diagonal(in the indices) parts($`𝐪0`$),
$$d_{𝐤+𝐪/2}^{}d_{𝐤𝐪/2}=\mathrm{\Lambda }_h(𝐤,𝐪)a_𝐤(𝐪)\mathrm{\Lambda }_h(𝐤,𝐪)a_𝐤^{}(𝐪)$$
(14)
$$c_{𝐤+𝐪/2}^{}c_{𝐤𝐪/2}=\mathrm{\Lambda }_e(𝐤,𝐪)a_𝐤(𝐪)+\mathrm{\Lambda }_e(𝐤,𝐪)a_𝐤^{}(𝐪)$$
(15)
Assuming that the ground states(of the non-interacting system) are annhilated by the sea-bosons $`a_𝐤(𝐪)`$ and $`a_𝐤(𝐪)`$
$$a_𝐤(𝐪)|Free=0,\text{ }a_𝐤(𝐪)|Free=0$$
(16)
This means,
$$c_{𝐤+𝐪/2}^{}c_{𝐤𝐪/2}c_{𝐤𝐪/2}^{}c_{𝐤+𝐪/2}=\overline{n}_e(𝐤+𝐪/2)(1\overline{n}_e(𝐤𝐪/2))=\mathrm{\Lambda }_e^2(𝐤,𝐪)$$
(17)
and similarly for the holes.
$$\mathrm{\Lambda }_e(𝐤,𝐪)=\sqrt{\overline{n}_e(𝐤+𝐪/2)(1\overline{n}_e(𝐤𝐪/2))}$$
(18)
$$\mathrm{\Lambda }_h(𝐤,𝐪)=\sqrt{\overline{n}_h(𝐤+𝐪/2)(1\overline{n}_h(𝐤𝐪/2))}$$
(19)
Then we make a leap of faith and suggest that the same should hold even when there are inteactions present and even when external fields are present. That is, the $`\overline{n}_h(𝐤)=d_𝐤^{}d_𝐤`$ now represents the expectation value with respect to the full interacting ground state. Lastly we would like to point out the internal self-consistency of this approach by computing the commutator bettween the diagonal and the off-diagonal bilinears. We find much to our relief that no matter what the choices for the cofficients $`\mathrm{\Lambda }_1`$, $`\mathrm{\Lambda }_2`$, $`\mathrm{\Lambda }_e`$ and $`\mathrm{\Lambda }_h`$ are, we recover the following exact identities.
$$[d_{𝐤𝐪/2}c_{𝐤𝐪/2},c_𝐩^{}c_𝐩]=d_{𝐤𝐪/2}c_{𝐤𝐪/2}\delta _{𝐩,𝐤𝐪/2}$$
(20)
$$[d_{𝐤𝐪/2}c_{𝐤𝐪/2},d_𝐩^{}d_𝐩]=d_{𝐤𝐪/2}c_{𝐤𝐪/2}\delta _{𝐩,𝐤+𝐪/2}$$
(21)
$$[c_{𝐤+𝐪/2}^{}c_{𝐤𝐪/2},c_𝐩^{}c_𝐩]=c_{𝐤+𝐪/2}^{}c_{𝐤𝐪/2}(\delta _{𝐩,𝐤𝐪/2}\delta _{𝐩,𝐤+𝐪/2})$$
(22)
$$[d_{𝐤+𝐪/2}^{}d_{𝐤𝐪/2},d_𝐩^{}d_𝐩]=d_{𝐤+𝐪/2}^{}d_{𝐤𝐪/2}(\delta _{𝐩,𝐤+𝐪/2}\delta _{𝐩,𝐤𝐪/2})$$
(23)
Thus we have written down a potentially useful set of identities. Let us write down the hamiltonian of free electrons and holes. In the Fermi language it is,
$$H_{free}=\underset{𝐤}{}ϵ^e(𝐤)c_𝐤^{}c_𝐤+\underset{𝐤}{}ϵ^h(𝐤)d_𝐤^{}d_𝐤+\underset{𝐪}{}\mathrm{\Omega }_{LO}b_𝐪^{}b_𝐪$$
(24)
The kinetic energy of the LO-phonon modes is also included. In the sea-boson language it may be expressed as follows :
$`H_{free}={\displaystyle \underset{\mathrm{𝐤𝐪}\mathrm{𝟎}}{}}(ϵ^e(𝐤𝐪/2)+ϵ^h(𝐤+𝐪/2))a_𝐤^{}(𝐪)a_𝐤(𝐪)`$
$`{\displaystyle \underset{𝐤}{}}({\displaystyle \frac{k^2}{2\mu }}+E_g)a_𝐤^{}(\mathrm{𝟎})a_𝐤(\mathrm{𝟎})+{\displaystyle \underset{\mathrm{𝐤𝐪}\mathrm{𝟎}}{}}(ϵ^h(𝐤𝐪/2)ϵ^h(𝐤+𝐪/2))a_𝐤^{}(𝐪)a_𝐤(𝐪)`$
$`+{\displaystyle \underset{\mathrm{𝐤𝐪}\mathrm{𝟎}}{}}(ϵ^h(𝐤𝐪/2)+ϵ^e(𝐤+𝐪/2))a_𝐤^{}(𝐪)a_𝐤(𝐪)+{\displaystyle \underset{𝐤}{}}({\displaystyle \frac{k^2}{2\mu }}+E_g)a_𝐤^{}(\mathrm{𝟎})a_𝐤(\mathrm{𝟎})`$
$$+\underset{\mathrm{𝐤𝐪}\mathrm{𝟎}}{}(ϵ^e(𝐤+𝐪/2)ϵ^e(𝐤𝐪/2))a_𝐤^{}(𝐪)a_𝐤(𝐪)$$
(25)
Again it may be noted that objects such as $`a_𝐤(\mathrm{𝟎})`$ and $`a_𝐤(\mathrm{𝟎})`$ are omitted from the formalism. The fact that the Fermi bilinears all evolve properly with respect to this hamiltonian is apparent without the need to perform any calculations. This is a strong indication that we are on the right track. It may puzzle the reader that we have included an object such as $`a_{𝐤,}(\mathrm{𝟎},)`$ in the above formula. This is due to the following reason. The commutation rule $`[d_𝐤c_𝐤,c_𝐤^{^{}}^{}d_𝐤^{^{}}^{}]`$ does not come out right if we don’t. Let us now write down some typical interaction terms.
$$H_{eh}=\underset{𝐪0}{}\frac{v_{eh}(𝐪)}{V}\underset{𝐤,𝐤^{^{}}}{}c_{𝐤+𝐪/2}^{}d_{𝐤^{^{}}𝐪/2}^{}d_{𝐤^{^{}}+𝐪/2}c_{𝐤𝐪/2}$$
(26)
This may be recast in the sea-boson language as follows,
$`H_{eh}={\displaystyle \underset{𝐪0}{}}{\displaystyle \frac{v_{eh}(𝐪)}{V}}{\displaystyle \underset{𝐤,𝐤^{^{}}}{}}[\mathrm{\Lambda }_1(𝐤/2+𝐤^{^{}}/2+𝐪/2,𝐤^{^{}}𝐤)a_{𝐤/2+𝐤^{^{}}/2+𝐪/2}^{}(𝐤𝐤^{^{}})+\mathrm{\Lambda }_2(𝐤/2+𝐤^{^{}}/2+𝐪/2,𝐤^{^{}}𝐤)a_{𝐤/2+𝐤^{^{}}/2+𝐪/2}(𝐤^{^{}}𝐤)]`$
$$[\mathrm{\Lambda }_1(𝐤/2+𝐤^{^{}}/2𝐪/2,𝐤^{^{}}𝐤)a_{𝐤/2+𝐤^{^{}}/2𝐪/2}(𝐤𝐤^{^{}})+\mathrm{\Lambda }_2(𝐤/2+𝐤^{^{}}/2𝐪/2,𝐤^{^{}}𝐤)a_{𝐤/2+𝐤^{^{}}/2𝐪/2}^{}(𝐤^{^{}}𝐤)]$$
(27)
Let us now try and write down the e-e/h-h repulsion terms(let us now focus on an undoped system),
$$H_{ee}=\underset{𝐪0}{}\frac{v(𝐪)}{2V}\rho ^e(𝐪)\rho ^e(𝐪)$$
(28)
$$H_{hh}=\underset{𝐪0}{}\frac{v(𝐪)}{2V}\rho ^h(𝐪)\rho ^h(𝐪)$$
(29)
$$\rho ^e(𝐪)=\underset{𝐤}{}\mathrm{\Lambda }_e(𝐤,𝐪)a_𝐤(𝐪)+\underset{𝐤}{}\mathrm{\Lambda }_e(𝐤,𝐪)a_𝐤^{}(𝐪)$$
(30)
$$\rho ^h(𝐪)=\underset{𝐤}{}\mathrm{\Lambda }_h(𝐤,𝐪)a_𝐤(𝐪)\underset{𝐤}{}\mathrm{\Lambda }_h(𝐤,𝐪)a_𝐤^{}(𝐪)$$
(31)
The coupling to phonons may be written as follows,
$$H_{ph}=\underset{𝐪0}{}\frac{M_𝐪}{\sqrt{V}}(b_𝐪+b_𝐪^{})(\rho ^e(𝐪)\rho ^h(𝐪))$$
(32)
It may be seen that only in the presence of real charge distributions do the electron-electron/hole-hole repuslion and coupling to phonons contribute appreciably to the hamiltonian. This means that in the undoped case in the absence of external fields we expect only the excitonic contribution broadened perhaps only via coupling to photons(which is ignored here). External fields, especially pump fields above the band gap cause significant real populations of carriers and these in turn relax by emitting phonons and through Coulomb scattering. Thus the formalism we have written down is simple and ideal for the study of these systems. The coupling to external fields may be written as,
$`H_{ext}(t)=({\displaystyle \frac{|e|}{\mu c}})\stackrel{}{A}_{ext}(t).\stackrel{}{p}_{vc}{\displaystyle \underset{𝐤}{}}[\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})a_𝐤(\mathrm{𝟎})+\mathrm{\Lambda }_2(𝐤,\mathrm{𝟎})a_𝐤^{}(\mathrm{𝟎})]`$
$$+(\frac{|e|}{\mu c})\stackrel{}{A}_{ext}^{}(t).\stackrel{}{p}_{vc}\underset{𝐤}{}[\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})a_𝐤^{}(\mathrm{𝟎})+\mathrm{\Lambda }_2(𝐤,\mathrm{𝟎})a_𝐤(\mathrm{𝟎})]$$
(33)
Let us now write down the various equations of motion of this system.
$`i{\displaystyle \frac{}{t}}a_𝐤(\mathrm{𝟎})=(ϵ^e(𝐤)+ϵ^h(𝐤))a_𝐤(\mathrm{𝟎}){\displaystyle \underset{𝐐0}{}}{\displaystyle \frac{v_{eh}(𝐐)}{V}}a_{𝐤𝐐}(\mathrm{𝟎})`$
$$+\underset{𝐐0}{}\frac{v_{eh}(𝐐)}{V}[(1\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})\mathrm{\Lambda }_1(𝐤𝐐,\mathrm{𝟎}))a_{𝐤𝐐}(\mathrm{𝟎})\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})\mathrm{\Lambda }_2(𝐤𝐐,\mathrm{𝟎})a_{𝐤𝐐}^{}(\mathrm{𝟎})]+(\frac{|e|}{\mu c})\stackrel{}{A}_{ext}^{}(t).\stackrel{}{p}_{vc}\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})$$
(34)
$`i{\displaystyle \frac{}{t}}a_𝐤(\mathrm{𝟎})=(ϵ^e(𝐤)+ϵ^h(𝐤))a_𝐤(\mathrm{𝟎})`$
$`{\displaystyle \underset{𝐐0}{}}{\displaystyle \frac{v_{eh}(𝐐)}{V}}\mathrm{\Lambda }_2(𝐤,\mathrm{𝟎})[\mathrm{\Lambda }_1(𝐤+𝐐,\mathrm{𝟎})a_{𝐤+𝐐}^{}(\mathrm{𝟎})+\mathrm{\Lambda }_2(𝐤+𝐐,\mathrm{𝟎})a_{𝐤+𝐐}(\mathrm{𝟎})]`$
$$+(\frac{|e|}{\mu c})\stackrel{}{A}_{ext}(t).\stackrel{}{p}_{vc}\mathrm{\Lambda }_2(𝐤,\mathrm{𝟎})$$
(35)
$$i\frac{}{t}a_𝐤(𝐪)=\frac{𝐤.𝐪}{m_e}a_𝐤(𝐪)+\frac{v(𝐪)}{V}\mathrm{\Lambda }_e(𝐤,𝐪)\rho ^{(e)}(𝐪)+\frac{M_𝐪}{\sqrt{V}}X_𝐪\mathrm{\Lambda }_e(𝐤,𝐪)$$
(36)
$$i\frac{}{t}a_𝐤(𝐪)=\frac{𝐤.𝐪}{m_h}a_𝐤(𝐪)\frac{v(𝐪)}{V}\mathrm{\Lambda }_h(𝐤,𝐪)\rho ^{(h)}(𝐪)+\frac{M_𝐪}{\sqrt{V}}X_𝐪\mathrm{\Lambda }_h(𝐤,𝐪)$$
(37)
$$i\frac{}{t}a_𝐤^{}(𝐪)=\frac{𝐤.𝐪}{m_e}a_𝐤^{}(𝐪)\frac{v(𝐪)}{V}\mathrm{\Lambda }_e(𝐤,𝐪)\rho ^{(e)}(𝐪)\frac{M_𝐪}{\sqrt{V}}X_𝐪\mathrm{\Lambda }_e(𝐤,𝐪)$$
(38)
$$i\frac{}{t}a_𝐤^{}(𝐪)=\frac{𝐤.𝐪}{m_h}a_𝐤^{}(𝐪)+\frac{v(𝐪)}{V}\mathrm{\Lambda }_h(𝐤,𝐪)\rho ^{(h)}(𝐪)\frac{M_𝐪}{\sqrt{V}}X_𝐪\mathrm{\Lambda }_h(𝐤,𝐪)$$
(39)
$$i\frac{}{t}X_𝐪=(2i\mathrm{\Omega }_{LO})P_𝐪$$
(40)
$$i\frac{}{t}P_𝐪=\frac{\mathrm{\Omega }_{LO}}{2i}X_𝐪i\frac{M_𝐪}{\sqrt{V}}(\rho ^{(e)}(𝐪)\rho ^{(h)}(𝐪))$$
(41)
The last four equations of motion only affect the electron and hole populations but do not impact directly upon the polarization or induced currents. This means that electron-electron and hole-hole repulsion and electron-phonon interaction change the distributions of electrons and holes and the electron-hole attraction determines the absorption spectrum. Let us now write down the electron and hole populations,
$`\overline{n}_h(𝐤)=a_𝐤^{}(\mathrm{𝟎})a_𝐤(\mathrm{𝟎})+a_𝐤^{}(\mathrm{𝟎})a_𝐤(\mathrm{𝟎})`$
$$\underset{𝐪0}{}a_{𝐤𝐪/2}^{}(𝐪)a_{𝐤𝐪/2}(𝐪)+\underset{𝐪0}{}a_{𝐤+𝐪/2}^{}(𝐪)a_{𝐤+𝐪/2}(𝐪)$$
(42)
$`\overline{n}_e(𝐤)=a_𝐤^{}(\mathrm{𝟎})a_𝐤(\mathrm{𝟎})a_𝐤^{}(\mathrm{𝟎})a_𝐤(\mathrm{𝟎})`$
$$+\underset{𝐪0}{}a_{𝐤𝐪/2}^{}(𝐪)a_{𝐤𝐪/2}(𝐪)\underset{𝐪0}{}a_{𝐤+𝐪/2}^{}(𝐪)a_{𝐤+𝐪/2}(𝐪)$$
(43)
In the above sets of equations we have ignored the contribution from objects such as $`a_𝐤(𝐪)`$ with $`𝐪0`$. The reason being that these contributions are difficult to deal with. The practical consequences of this assumption means that we have to restric our attention to large $`𝐤`$. Namely that we must ensure that $`𝐤`$ in the above equation is much larger than any inverse length-scale in the problem. Thus we expect our theory to be poor for $`𝐤`$ small. This is in fact the case as we shall soon find out. The analysis including this large $`𝐪`$ effect Let us now introduce several propagators. These are going to be useful in ascertaining the influence carrier-carrier repulsion and carrier-phonon interactions have on the populations of electrons and holes. We can see from Eq.( 42) and Eq.( 43), the manner in which the populations of electrons and holes evolve. If we ignore the terms that involve the sea-bosons $`a_{𝐤\sigma }(𝐪\sigma )`$ thern we see that electrons and holes have the same distribution determined solely by the external fields. That is, so long as relaxation processess are ignored we find simple and intuitively appealing formulas for the distributions. It is also worth pointing out that this situation is entirely analogous(perhaps even equivalent) to the Semiconductor Bloch Equations(SBE). There we find that if one ignores terms beyond the Hartree-Fock approximation, then the populations of electrons and holes are indentical even though the effective masses are different. However once relaxation processes begin, the distributions respond appropriately and the system must be solved self-consistently. A quantity such as $`a_{𝐤\sigma }^{}(𝐪\sigma )a_{𝐤\sigma }(𝐪\sigma )`$ is zero when the electrons and holes have ”ideal” momentum distributions(that is, identically zero for undoped systems). However, when they start to acquire non-zero values due to external fields the above quantity also begins to acquire a non-zero value and the whole system proceeds to evolve accordingly. Let us now introduce several Green functions.
$$G_{11}(𝐤,𝐤^{^{}};𝐪)=iT\text{ }a_𝐤(𝐪,t)a_𝐤^{^{}}^{}(𝐪,0)$$
(44)
$$G_{12}(𝐤,𝐤^{^{}};𝐪)=iT\text{ }a_𝐤^{}(𝐪,t)a_𝐤^{^{}}^{}(𝐪,0)$$
(45)
$$G_{21}(𝐤,𝐤^{^{}};𝐪)=iT\text{ }a_𝐤^{}(𝐪,t)a_𝐤^{^{}}^{}(𝐪,0)$$
(46)
$$G_{22}(𝐤,𝐤^{^{}};𝐪)=iT\text{ }a_𝐤(𝐪,t)a_𝐤^{^{}}^{}(𝐪,0)$$
(47)
$$G_{X\sigma }(𝐤;𝐪)=iX_𝐪(t)a_{𝐤\sigma }^{}(𝐪\sigma ,0)$$
(48)
$$G_{P\sigma }(𝐤;𝐪)=iP_𝐪(t)a_{𝐤\sigma }^{}(𝐪\sigma ,0)$$
(49)
$`i{\displaystyle \frac{}{t}}G_{11}(𝐤,𝐤^{^{}};𝐪,t)=\delta (t)\delta _{𝐤,𝐤^{^{}}}+{\displaystyle \frac{𝐤.𝐪}{m_e}}G_{11}(𝐤,𝐤^{^{}};𝐪,t)i{\displaystyle \frac{v(𝐪)}{V}}\mathrm{\Lambda }_e(𝐤,𝐪)T\text{ }\rho ^{(e)}(𝐪,t)a_𝐤^{^{}}^{}(𝐪,0)`$
$$+\frac{M_𝐪}{\sqrt{V}}\mathrm{\Lambda }_e(𝐤,𝐪)G_X(𝐤^{^{}};𝐪,t)$$
(50)
$`i{\displaystyle \frac{}{t}}G_{12}(𝐤,𝐤^{^{}};𝐪,t)={\displaystyle \frac{𝐤.𝐪}{m_e}}G_{12}(𝐤,𝐤^{^{}};𝐪,t)+i{\displaystyle \frac{v(𝐪)}{V}}\mathrm{\Lambda }_e(𝐤,𝐪)T\text{ }\rho ^{(e)}(𝐪,t)a_𝐤^{^{}}^{}(𝐪,0)`$
$$\frac{M_𝐪}{\sqrt{V}}\mathrm{\Lambda }_e(𝐤,𝐪)G_X(𝐤^{^{}};𝐪,t)$$
(51)
$`i{\displaystyle \frac{}{t}}G_{22}(𝐤,𝐤^{^{}};𝐪,t)=\delta (t)\delta _{𝐤,𝐤^{^{}}}{\displaystyle \frac{𝐤.𝐪}{m_h}}G_{22}(𝐤,𝐤^{^{}};𝐪,t)+i{\displaystyle \frac{v(𝐪)}{V}}\mathrm{\Lambda }_h(𝐤,𝐪)T\text{ }\rho ^{(h)}(𝐪,t)a_𝐤^{^{}}^{}(𝐪,0)`$
$$+\frac{M_𝐪}{\sqrt{V}}\mathrm{\Lambda }_h(𝐤,𝐪)G_X(𝐤^{^{}};𝐪,t)$$
(52)
$`i{\displaystyle \frac{}{t}}G_{21}(𝐤,𝐤^{^{}};𝐪,t)={\displaystyle \frac{𝐤.𝐪}{m_h}}G_{21}(𝐤,𝐤^{^{}};𝐪,t)i{\displaystyle \frac{v(𝐪)}{V}}\mathrm{\Lambda }_h(𝐤,𝐪)T\text{ }\rho ^{(h)}(𝐪,t)a_𝐤^{^{}}^{}(𝐪,0)`$
$$\frac{M_𝐪}{\sqrt{V}}\mathrm{\Lambda }_h(𝐤,𝐪)G_X(𝐤^{^{}};𝐪,t)$$
(53)
$$i\frac{}{t}G_{X\sigma }(𝐤;𝐪,t)=(2i\mathrm{\Omega }_{LO})G_{P\sigma }(𝐤;𝐪,t)$$
(54)
$$i\frac{}{t}G_P(𝐤;𝐪,t)=(\frac{\mathrm{\Omega }_{LO}}{2i})G_X(𝐤;𝐪,t)\frac{M_𝐪}{\sqrt{V}}T\text{ }\rho ^{(e)}(𝐪,t)a_𝐤^{}(𝐪,0)$$
(55)
$$i\frac{}{t}G_P(𝐤;𝐪,t)=(\frac{\mathrm{\Omega }_{LO}}{2i})G_X(𝐤;𝐪,t)+\frac{M_𝐪}{\sqrt{V}}T\text{ }\rho ^{(h)}(𝐪,t)a_𝐤^{}(𝐪,0)$$
(56)
In order to solve this system we have to expand the Green functions in terms of Matsubara frequencies. We introduce a temperature just for ease of doing calculations. In the end we shall go to the zero temperature limit as this is the regime when the interpretations are the cleanest (here $`z_n=2\pi n/\beta `$ and $`\beta =1/k_BT`$ ).
$$(iz_n\frac{𝐤.𝐪}{m_e})G_{11}(𝐤,𝐤^{^{}};𝐪,z_n)=\frac{1}{i\beta }\delta _{𝐤,𝐤^{^{}}}i\frac{\stackrel{~}{v}(𝐪,z_n)}{V}\mathrm{\Lambda }_e(𝐤,𝐪)T\text{ }\rho ^{(e)}(𝐪,z_n)a_𝐤^{^{}}^{}(𝐪,0)$$
(57)
$$(iz_n\frac{𝐤.𝐪}{m_e})G_{12}(𝐤,𝐤^{^{}};𝐪,z_n)=i\frac{\stackrel{~}{v}(𝐪,z_n)}{V}\mathrm{\Lambda }_e(𝐤,𝐪)T\text{ }\rho ^{(e)}(𝐪,z_n)a_𝐤^{^{}}^{}(𝐪,0)$$
(58)
$$(iz_n+\frac{𝐤.𝐪}{m_h})G_{22}(𝐤,𝐤^{^{}};𝐪,z_n)=\frac{1}{i\beta }\delta _{𝐤,𝐤^{^{}}}+i\frac{\stackrel{~}{v}(𝐪)}{V}\mathrm{\Lambda }_h(𝐤,𝐪)T\text{ }\rho ^{(h)}(𝐪,z_n)a_𝐤^{^{}}^{}(𝐪,0)$$
(59)
$$(iz_n+\frac{𝐤.𝐪}{m_h})G_{21}(𝐤,𝐤^{^{}};𝐪,z_n)=i\frac{\stackrel{~}{v}(𝐪,z_n)}{V}\mathrm{\Lambda }_h(𝐤,𝐪)T\text{ }\rho ^{(h)}(𝐪,z_n)a_𝐤^{^{}}^{}(𝐪,0)$$
(60)
$$\stackrel{~}{v}(𝐪,z_n)=v(𝐪)\frac{2\text{ }M_𝐪^2\mathrm{\Omega }_{LO}}{z_n^2+\mathrm{\Omega }_{LO}^2}$$
(61)
$$iT\text{ }\rho ^{(e)}(𝐪,z_n)a_𝐤^{^{}}^{}(𝐪,0)=\underset{𝐩}{}\mathrm{\Lambda }_e(𝐩,𝐪)G_{11}(𝐩,𝐤^{^{}};𝐪,z_n)+\underset{𝐩}{}\mathrm{\Lambda }_e(𝐩,𝐪)G_{12}(𝐩,𝐤^{^{}};𝐪,z_n)$$
(62)
$$iT\text{ }\rho ^{(h)}(𝐪,z_n)a_𝐤^{^{}}^{}(𝐪,0)=\underset{𝐩}{}\mathrm{\Lambda }_h(𝐩,𝐪)G_{22}(𝐩,𝐤^{^{}};𝐪,z_n)\underset{𝐩}{}\mathrm{\Lambda }_h(𝐩,𝐪)G_{21}(𝐩,𝐤^{^{}};𝐪,z_n)$$
(63)
$`iT\text{ }\rho ^{(e)}(𝐪,z_n)a_𝐤^{^{}}^{}(𝐪,0)=({\displaystyle \frac{1}{i\beta }}){\displaystyle \frac{\mathrm{\Lambda }_e(𝐤^{^{}},𝐪)}{iz_n\frac{𝐤^{^{}}.𝐪}{m_e}}}`$
$$+i\frac{\stackrel{~}{v}(𝐪,z_n)}{V}\underset{𝐩}{}\frac{\mathrm{\Lambda }_e^2(𝐩,𝐪)\mathrm{\Lambda }_e^2(𝐩,𝐪)}{iz_n\frac{𝐩.𝐪}{m_e}}T\text{ }\rho ^{(e)}(𝐪,z_n)a_𝐤^{^{}}^{}(𝐪,0)$$
(64)
$$T\text{ }\rho ^{(e)}(𝐪,z_n)a_𝐤^{^{}}^{}(𝐪,0)=(\frac{1}{i\beta })\frac{i\mathrm{\Lambda }_e(𝐤^{^{}},𝐪)}{iz_n\frac{𝐤^{^{}}.𝐪}{m_e}}\frac{1}{ϵ^{(e)}(𝐪,iz_n)}$$
(65)
where,
$$ϵ^{(e)}(𝐪,iz_n)=1+\frac{\stackrel{~}{v}(𝐪,z_n)}{V}\underset{𝐩}{}\frac{\overline{n}^{(e)}(𝐩+𝐪/2)\overline{n}^{(e)}(𝐩𝐪/2)}{iz_n\frac{𝐩.𝐪}{m_e}}$$
(66)
$$T\text{ }\rho ^{(h)}(𝐪,z_n)a_𝐤^{^{}}^{}(𝐪,0)=(\frac{1}{i\beta })\frac{i\mathrm{\Lambda }_h(𝐤^{^{}},𝐪)}{iz_n+\frac{𝐤^{^{}}.𝐪}{m_h}}\frac{1}{ϵ^{(h)}(𝐪,iz_n)}$$
(67)
where,
$$ϵ^{(h)}(𝐪,iz_n)=1+\frac{\stackrel{~}{v}(𝐪,z_n)}{V}\underset{𝐩}{}\frac{\overline{n}^{(h)}(𝐩+𝐪/2)\overline{n}^{(h)}(𝐩𝐪/2)}{iz_n\frac{𝐩.𝐪}{m_h}}$$
(68)
Therefore,
$$a_𝐤^{}(𝐪)a_𝐤(𝐪)=(\frac{1}{i\beta })(\frac{1}{V})\underset{n}{}\frac{\stackrel{~}{v}(𝐪,z_n)}{ϵ^{(e)}(𝐪,iz_n)}\frac{i\mathrm{\Lambda }_e^2(𝐤,𝐪)}{(iz_n\frac{𝐤.𝐪}{m_e})^2}$$
(69)
$$a_𝐤^{}(𝐪)a_𝐤(𝐪)=(\frac{1}{i\beta })(\frac{1}{V})\underset{n}{}\frac{\stackrel{~}{v}(𝐪,z_n)}{ϵ^{(h)}(𝐪,iz_n)}\frac{i\mathrm{\Lambda }_h^2(𝐤,𝐪)}{(iz_n+\frac{𝐤.𝐪}{m_h})^2}$$
(70)
Then when we go to the zero-temperature limit we have to integrate over all $`n`$. In Fig.1 we see the pole structure of the above equations. Call $`i\text{ }z_n=i\text{ }z`$. Then,
$$a_𝐤^{}(𝐪)a_𝐤(𝐪)=(\frac{1}{i\beta })(\frac{1}{V})\frac{\beta }{2\pi }_Cdz\text{ }\frac{\stackrel{~}{v}(𝐪,z)}{ϵ^{(e)}(𝐪,iz)}\frac{i\mathrm{\Lambda }_e^2(𝐤,𝐪)}{(iz\frac{𝐤.𝐪}{m_e})^2}$$
(71)
In Fig.1 we see the pole structure of the above contour integral. Let us assume that $`𝐤.𝐪>0`$ then the pole $`z=i\text{ }𝐤.𝐪/m_e`$ is in the lower half-plane. We have to close the contour in such a way that this pole is excluded from consideration, since if it were included we would have a formula for $`a_𝐤(𝐪)a_𝐤^{}(𝐪)`$ rather than $`a_𝐤^{}(𝐪)a_𝐤(𝐪)`$. Therefore we have to close the contuor in the upper half-plane(and $`C=C1`$). Now, if we count the number of poles in the integrand we find that first of all, the zeros of the dielectric function that lie on the positive imaginary axis of the z-plane(how many zeros are there, is an important question which we shall address subsequently) contribute. Then it seems at first sight that even the poles of $`\stackrel{~}{v}(𝐪,z)`$ contribute. The poles of this function lie at $`\pm i\text{ }\mathrm{\Omega }_{LO}`$. However, upon closer examination we find that this is not the case. The poles of $`\stackrel{~}{v}(𝐪,z)`$ are also poles of $`ϵ^{(e)}(𝐪,iz)`$ and the two cancel. Thus the only poles that contribute are the zeros of the dielectric function that lie on the positive imaginary axis. How many such zeros are there ? If one counts only the collective modes then one arrives at the conclusion that there are only two, one corresponding to the plasmon(modified by phonons) and the other corresponding to phonons(modified by Coulomb interactions). There is another mode that is equally important, indeed it would be a serious mistake to ignore this contribution, namely the particle-hole mode. We have encountered this problem before. In our earlier article we presented an argument that shows how one may incorporate the particle-hole mode. In retrospect it seems that the approach presented there is not a good one, although it serves well to illustrate the importance of the particle-hole mode. Here we shall take the point of view that all energies are allowed as zeros of the dielectric function(for each $`𝐪`$) but each comes with a weight corresponding to the strength of the dynamical structure factor at that energy. Thus for small $`𝐪`$ we recover naturally the collective modes but for larger $`𝐪`$ we start summing the particle-hole modes as well. There is really is no rigorous justification for this point of view except that it is physically well-motivated.
$$a_𝐤^{}(𝐪)a_𝐤(𝐪)=(\frac{1}{i})(\frac{1}{V})_C\frac{dz}{2\pi \text{ }i}\text{ }\frac{\stackrel{~}{v}(𝐪,z)}{ϵ^{(e)}(𝐪,iz)}\frac{\mathrm{\Lambda }_e^2(𝐤,𝐪)}{(iz\frac{𝐤.𝐪}{m_e})^2}$$
(72)
The poles are $`z=i\text{ }\omega _I^{(e)}(𝐪)`$, $`\omega _I^{(e)}(𝐪)>0`$ satisfies $`ϵ^{(e)}(𝐪,\omega _I^{(e)})=0`$.
$$a_𝐤^{}(𝐪)a_𝐤(𝐪)=(\frac{1}{i})(\frac{1}{V})\underset{I}{}\frac{\stackrel{~}{v}(𝐪,i\text{ }\omega _I^{(e)})}{\frac{}{z}|_{z=i\text{ }\omega _I^{(e)}(𝐪)}ϵ^{(e)}(𝐪,iz)}\frac{\mathrm{\Lambda }_e^2(𝐤,𝐪)}{(\omega _I^{(e)}(𝐪)+\frac{𝐤.𝐪}{m_e})^2}$$
(73)
Let us first evaluate this. Similarly, one may write for holes,
$$a_𝐤^{}(𝐪)a_𝐤(𝐪)=(\frac{1}{i})(\frac{1}{V})\underset{I}{}\frac{\stackrel{~}{v}(𝐪,i\text{ }\omega _I^{(h)})}{\frac{}{z}|_{z=i\text{ }\omega _I^{(h)}(𝐪)}ϵ^{(h)}(𝐪,iz)}\frac{\mathrm{\Lambda }_h^2(𝐤,𝐪)}{(\omega _I^{(h)}(𝐪)\frac{𝐤.𝐪}{m_h})^2}$$
(74)
Let us now evaluate these quantities more explicitly.
$`{\displaystyle \frac{}{z}}|_{z=i\omega _I}ϵ(𝐪,iz)=P(𝐪,iz)|_{z=i\omega _I}{\displaystyle \frac{}{z}}|_{z=i\omega _I}\stackrel{~}{v}(𝐪,z)\stackrel{~}{v}(𝐪,z)|_{z=i\omega _I}{\displaystyle \frac{}{z}}|_{z=i\omega _I}P(𝐪,iz)`$
$`={\displaystyle \frac{1}{\stackrel{~}{v}(𝐪,z)}}|_{z=i\omega _I}{\displaystyle \frac{}{z}}|_{z=i\omega _I}\stackrel{~}{v}(𝐪,z)\stackrel{~}{v}(𝐪,z)|_{z=i\omega _I}{\displaystyle \frac{}{z}}|_{z=i\omega _I}P(𝐪,iz)`$Since,
$$\stackrel{~}{v}(𝐪,z)=v(𝐪)\frac{2\mathrm{\Omega }_{LO}M_𝐪^2}{z^2+\mathrm{\Omega }_{LO}^2}$$
(75)
$$\frac{}{z}|_{z=i\omega _I}P(𝐪,iz)=\frac{i}{V}\underset{𝐤}{}\frac{\overline{n}_{𝐤𝐪/2}\overline{n}_{𝐤+𝐪/2}}{(\omega _I\frac{𝐤.𝐪}{m})^2}$$
(76)
$$\frac{}{z}|_{z=i\omega _I}\stackrel{~}{v}(𝐪,z)=\frac{4i\omega _I\mathrm{\Omega }_{LO}M_𝐪^2}{(\omega _I^2\mathrm{\Omega }_{LO}^2)^2}$$
(77)
$$\frac{}{z}|_{z=i\omega _I}ϵ(𝐪,iz)=i\frac{V}{\stackrel{~}{v}(𝐪,i\omega _I)}\{\frac{M_𝐪^2}{V}\frac{4\mathrm{\Omega }_{LO}\omega _I}{(\omega _I^2\mathrm{\Omega }_{LO}^2)^2}+(\frac{\stackrel{~}{v}(𝐪,i\omega _I)}{V})^2\underset{𝐤}{}\frac{\overline{n}_{𝐤𝐪/2}\overline{n}_{𝐤+𝐪/2}}{(\omega _I\frac{𝐤.𝐪}{m})^2}\}$$
(78)
As we pointed our just a while ago, it is necessary that we interpret the sum over $`I`$ in a special manner so that we are able to recover both the collective as well as the particle-hole modes. The way this is done is through the following identification,
$$\underset{I,𝐪}{}f(𝐪,\omega _I)=\underset{𝐪}{}_0^{\mathrm{}}𝑑\omega \text{ }W(𝐪,\omega )f(𝐪,\omega )$$
(79)
where the weight is the dynamical structure factor normalised to unity.
$$W(𝐪,\omega )=\frac{S(𝐪,\omega )}{_0^{\mathrm{}}𝑑\omega \text{ }S(𝐪,\omega )}$$
(80)
The dynamical structure factor is defined to be the the dynamical density-density correlation function fourier-transformed divided by the total number of particles. Let us first write down,
$$T\text{ }\rho ^{(e)}(𝐪,t)a_𝐤^{^{}}^{}(𝐪,0)=(\frac{1}{i\beta })\underset{n}{}e^{z_nt}\frac{i\mathrm{\Lambda }_e(𝐤^{^{}},𝐪)}{iz_n\frac{𝐤^{^{}}.𝐪}{m_e}}\frac{1}{ϵ^{(e)}(𝐪,iz_n)}$$
(81)
Since $`Im(t)[0,\beta ]`$, if $`Im(t)<0`$ then,
$`\rho ^{(e)}(𝐪,t)a_𝐤^{^{}}^{}(𝐪,0)=({\displaystyle \frac{1}{i\beta }})({\displaystyle \frac{\beta }{2\pi }}){\displaystyle _C_{}}\text{ }dz\text{ }e^{zt}\text{ }{\displaystyle \frac{i\mathrm{\Lambda }_e(𝐤^{^{}},𝐪)}{iz\frac{𝐤^{^{}}.𝐪}{m_e}}}{\displaystyle \frac{1}{ϵ^{(e)}(𝐪,iz)}}`$
$$=(\frac{i}{2\pi })_{\mathrm{}}^+\mathrm{}𝑑x\text{ }e^{xt}\text{ }\frac{i\mathrm{\Lambda }_e(𝐤^{^{}},𝐪)}{ix\frac{𝐤^{^{}}.𝐪}{m_e}}\frac{1}{ϵ^{(e)}(𝐪,ix)}$$
(82)
If $`Im(t)>0`$ then,
$`a_𝐤^{^{}}^{}(𝐪,0)\rho ^{(e)}(𝐪,t)=({\displaystyle \frac{1}{i\beta }})({\displaystyle \frac{\beta }{2\pi }}){\displaystyle _{C_+}}\text{ }dz\text{ }e^{zt}\text{ }{\displaystyle \frac{i\mathrm{\Lambda }_e(𝐤^{^{}},𝐪)}{iz\frac{𝐤^{^{}}.𝐪}{m_e}}}{\displaystyle \frac{1}{ϵ^{(e)}(𝐪,iz)}}`$
$$=(\frac{i}{2\pi })_{\mathrm{}}^+\mathrm{}\text{ }𝑑x\text{ }e^{xt}\text{ }\frac{i\mathrm{\Lambda }_e(𝐤^{^{}},𝐪)}{ix\frac{𝐤^{^{}}.𝐪}{m_e}}\frac{1}{ϵ^{(e)}(𝐪,ix)}$$
(83)
Here $`C_+(C_{})`$ is the semi-circle in the upper(lower) half plane. Let us now take the complex conjugate of the above equation. $`Im(t)>0`$ implies $`Im(t^{})<0`$. Therefore if $`Im(t^{})<0`$,
$$\rho ^{(e)}(𝐪,t^{})a_𝐤^{^{}}(𝐪,0)=(\frac{i}{2\pi })_{\mathrm{}}^+\mathrm{}\text{ }dx\text{ }e^{xt^{}}\text{ }\frac{i\mathrm{\Lambda }_e(𝐤^{^{}},𝐪)}{ix\frac{𝐤^{^{}}.𝐪}{m_e}}\frac{1}{ϵ^{(e)}(𝐪,ix)}$$
(84)
Define,
$$\rho ^{(e,a)}(𝐪,0)=\underset{𝐤^{^{}}}{}\mathrm{\Lambda }_e(𝐤^{^{}},𝐪)a_𝐤^{^{}}^{}(𝐪,0)$$
(85)
$$\rho ^{(e,b)}(𝐪,0)=\underset{𝐤^{^{}}}{}\mathrm{\Lambda }_e(𝐤^{^{}},𝐪)a_𝐤^{^{}}(𝐪,0)$$
(86)
therefore,
$$\rho ^{(e)}(𝐪,0)=\rho ^{(e,a)}(𝐪,0)+\rho ^{(e,b)}(𝐪,0)$$
(87)
Then we have $`Im(t)<0`$,
$$\rho ^{(e)}(𝐪,t)\rho ^{(e)}(𝐪,0)=(\frac{i}{2\pi })_{\mathrm{}}^+\mathrm{}𝑑x\text{ }e^{xt}\text{ }\underset{𝐤^{^{}}}{}\frac{i(\overline{n}_e(𝐤^{^{}}𝐪/2)\overline{n}_e(𝐤^{^{}}+𝐪/2))}{ix\frac{𝐤^{^{}}.𝐪}{m_e}}\frac{1}{ϵ^{(e)}(𝐪,ix)}$$
(88)
For $`Im(t^{})>0`$,
$$\rho ^{(e)}(𝐪,0)\rho ^{(e)}(𝐪,t^{})=(\frac{i}{2\pi })_{\mathrm{}}^+\mathrm{}𝑑x\text{ }e^{xt^{}}\text{ }\underset{𝐤^{^{}}}{}\frac{i(\overline{n}_e(𝐤^{^{}}+𝐪/2)\overline{n}_e(𝐤^{^{}}𝐪/2))}{ix+\frac{𝐤^{^{}}.𝐪}{m_e}}\frac{1}{ϵ^{(e)}(𝐪,ix)}$$
(89)
Combining these two,
$`T\rho ^{(e)}(𝐪,t)\rho ^{(e)}(𝐪,0)=({\displaystyle \frac{i}{2\pi }}){\displaystyle _{\mathrm{}}^+\mathrm{}}𝑑x\text{ }e^{xt}\text{ }{\displaystyle \underset{𝐤^{^{}}}{}}{\displaystyle \frac{i(\overline{n}_e(𝐤^{^{}}𝐪/2)\overline{n}_e(𝐤^{^{}}+𝐪/2))}{ix\frac{𝐤^{^{}}.𝐪}{m_e}}}{\displaystyle \frac{1}{ϵ^{(e)}(𝐪,ix)}}`$
$$=(\frac{V}{2\pi })_{\mathrm{}}^+\mathrm{}𝑑x\text{ }e^{xt}\text{ }\frac{P_e(𝐪,ix)}{ϵ^{(e)}(𝐪,ix)}$$
(90)
Define the Green function,
$$𝒟(𝐪,t)=T\rho ^{(e)}(𝐪,t)\rho ^{(e)}(𝐪,0)$$
(91)
then,
$$𝒟_{ret}(𝐪,\omega )=V\frac{P_e^{ret}(𝐪,\omega )}{ϵ_{ret}^{(e)}(𝐪,\omega )}$$
(92)
The corresponding spectral function is the dynamical structure factor,
$$N_e\text{ }S(𝐪,\omega )=2Im(𝒟_{ret}(𝐪,\omega ))$$
(93)
$$ϵ_r^{(e)}(𝐪,\omega )=1v^r(𝐪,\omega )P_e^r(𝐪,\omega )+v^i(𝐪,\omega )P_e^i(𝐪,\omega )$$
(94)
$$ϵ_i^{(e)}(𝐪,\omega )=v^r(𝐪,\omega )P_e^i(𝐪,\omega )v^i(𝐪,\omega )P_e^r(𝐪,\omega )$$
(95)
This procedure ensures that we correctly incorporate both the collective (for small $`𝐪`$) and the particle-hole modes. There is an alternative approach that comes to mind. That is the method of exact diagonalisation. Consider the hamiltonian,
$`H^{^{}}={\displaystyle \underset{𝐤,𝐪}{}}{\displaystyle \frac{𝐤.𝐪}{m_e}}a_𝐤^{}(𝐪)a_𝐤(𝐪){\displaystyle \underset{𝐤,𝐪}{}}{\displaystyle \frac{𝐤.𝐪}{m_h}}a_𝐤^{}(𝐪)a_𝐤(𝐪)+{\displaystyle \underset{𝐪}{}}\mathrm{\Omega }_{LO}b_𝐪^{}b_𝐪`$
$`+{\displaystyle \underset{𝐪\mathrm{𝟎}}{}}{\displaystyle \frac{v(𝐪)}{2V}}{\displaystyle \underset{𝐤,𝐤^{^{}}}{}}[\mathrm{\Lambda }_e(𝐤,𝐪)a_𝐤(𝐪)+\mathrm{\Lambda }_e(𝐤,𝐪)a_𝐤^{}(𝐪)][\mathrm{\Lambda }_e(𝐤^{^{}},𝐪)a_𝐤^{^{}}(𝐪)+\mathrm{\Lambda }_e(𝐤^{^{}},𝐪)a_𝐤^{^{}}^{}(𝐪)]`$
$`+{\displaystyle \underset{𝐪\mathrm{𝟎}}{}}{\displaystyle \frac{v(𝐪)}{2V}}{\displaystyle \underset{𝐤,𝐤^{^{}}}{}}[\mathrm{\Lambda }_h(𝐤,𝐪)a_𝐤(𝐪)+\mathrm{\Lambda }_h(𝐤,𝐪)a_𝐤^{}(𝐪)][\mathrm{\Lambda }_h(𝐤^{^{}},𝐪)a_𝐤^{^{}}(𝐪)+\mathrm{\Lambda }_h(𝐤^{^{}},𝐪)a_𝐤^{^{}}^{}(𝐪)]`$
$$+\underset{𝐪\mathrm{𝟎}}{}\frac{M_𝐪}{\sqrt{V}}(b_𝐪+b_𝐪^{})[\mathrm{\Lambda }_e(𝐤,𝐪)a_𝐤(𝐪)+\mathrm{\Lambda }_e(𝐤,𝐪)a_𝐤^{}(𝐪)+\mathrm{\Lambda }_h(𝐤,𝐪)a_𝐤(𝐪)+\mathrm{\Lambda }_h(𝐤,𝐪)a_𝐤^{}(𝐪)]$$
(96)
In order to diagonalise this we proceed as follows. Let us postulate the existence of dressesd sea-bosons $`d_{I\sigma }(𝐪)`$ such that,
$$H^{^{}}=\underset{I,𝐪,\sigma }{}\omega _{I\sigma }(𝐪)d_{I\sigma }^{}(𝐪)d_{I\sigma }(𝐪)$$
(97)
Now for some notation. $`\sigma =e,h`$ (correspondingly $`\sigma =,`$, furthermore, $`S(\sigma )`$ is such that $`S()=+1`$ and $`S()=1`$).
$$a_{𝐤\sigma }(𝐪\sigma )=\underset{I}{}[a_{𝐤\sigma }(𝐪\sigma ),d_{I\sigma }^{}(𝐪)]d_{I\sigma }(𝐪)\underset{I}{}[a_{𝐤\sigma }(𝐪\sigma ),d_{I\sigma }(𝐪)]d_{I\sigma }^{}(𝐪)$$
(98)
$$b_𝐪=\underset{I,\sigma }{}[b_𝐪,d_{I\sigma }^{}(𝐪)]d_{I\sigma }(𝐪)\underset{I,\sigma }{}[b_𝐪,d_{I\sigma }(𝐪)]d_{I\sigma }^{}(𝐪)$$
(99)
The inverse relation is,
$`d_{I\sigma }(𝐪)={\displaystyle \underset{𝐤}{}}[d_{I\sigma }(𝐪),a_{𝐤\sigma }^{}(𝐪\sigma )]a_{𝐤\sigma }(𝐪\sigma ){\displaystyle \underset{𝐤}{}}[d_{I\sigma }(𝐪),a_{𝐤\sigma }(𝐪\sigma )]a_{𝐤\sigma }^{}(𝐪\sigma )`$
$$+[d_{I\sigma }(𝐪),b_𝐪^{}]b_𝐪[d_{I\sigma }(𝐪),b_𝐪]b_𝐪^{}$$
(100)
Therefore,
$`\omega _{I\sigma }(𝐪)d_{I\sigma }(𝐪)={\displaystyle \underset{𝐤}{}}S(\sigma ){\displaystyle \frac{𝐤.𝐪}{m_\sigma }}[d_{I\sigma }(𝐪),a_{𝐤\sigma }^{}(𝐪\sigma )]a_{𝐤\sigma }(𝐪\sigma )S(\sigma ){\displaystyle \underset{𝐤}{}}{\displaystyle \frac{𝐤.𝐪}{m_\sigma }}[d_{I\sigma }(𝐪),a_{𝐤\sigma }(𝐪\sigma )]a_{𝐤\sigma }^{}(𝐪\sigma )`$
$`+{\displaystyle \frac{v(𝐪)}{V}}{\displaystyle \underset{𝐤,𝐤^{^{}}}{}}[\mathrm{\Lambda }_\sigma (𝐤,𝐪)[d_{I\sigma }(𝐪),a_{𝐤\sigma }(𝐪\sigma )]+\mathrm{\Lambda }_\sigma (𝐤,𝐪)[d_{I\sigma }(𝐪),a_{𝐤\sigma }^{}(𝐪\sigma )]][\mathrm{\Lambda }_\sigma (𝐤^{^{}},𝐪)a_{𝐤^{^{}}\sigma }(𝐪\sigma )+\mathrm{\Lambda }_\sigma (𝐤^{^{}},𝐪)a_{𝐤^{^{}}\sigma }^{}(𝐪\sigma )]`$
$`+\mathrm{\Omega }_{LO}[d_{I\sigma }(𝐪),b_𝐪^{}]b_𝐪+\mathrm{\Omega }_{LO}[d_{I\sigma }(𝐪),b_𝐪]b_𝐪^{}`$
$`+{\displaystyle \frac{M_𝐪}{\sqrt{V}}}([d_{I\sigma }(𝐪),b_𝐪]+[d_{I\sigma }(𝐪),b_𝐪^{}]){\displaystyle \underset{𝐤}{}}[\mathrm{\Lambda }_\sigma (𝐤,𝐪)a_{𝐤\sigma }(𝐪\sigma )+\mathrm{\Lambda }_\sigma (𝐤,𝐪)a_{𝐤\sigma }^{}(𝐪\sigma )]`$
$$+\frac{M_𝐪}{\sqrt{V}}(b_𝐪+b_𝐪^{})\underset{𝐤}{}(\mathrm{\Lambda }_\sigma (𝐤,𝐪)[d_{I\sigma }(𝐪),a_{𝐤\sigma }(𝐪\sigma )]+\mathrm{\Lambda }_\sigma (𝐤,𝐪)[d_{I\sigma }(𝐪),a_{𝐤\sigma }^{}(𝐪\sigma )]$$
(101)
$$(\omega _{I\sigma }(𝐪)S(\sigma )\frac{𝐤.𝐪}{m_\sigma })[d_{I\sigma }(𝐪),a_{𝐤\sigma }^{}(𝐪\sigma )]=S(\sigma )\frac{\stackrel{~}{v}(𝐪,I\sigma )}{V}\mathrm{\Lambda }_\sigma (𝐤,𝐪)\rho ^{(\sigma )}(𝐪,I)$$
(102)
$$(\omega _{I\sigma }(𝐪)S(\sigma )\frac{𝐤.𝐪}{m_\sigma })[d_{I\sigma }(𝐪),a_{𝐤\sigma }(𝐪\sigma )]=S(\sigma )\frac{\stackrel{~}{v}(𝐪,I\sigma )}{V}\mathrm{\Lambda }_\sigma (𝐤,𝐪)\rho ^{(\sigma )}(𝐪,I)$$
(103)
$$\rho ^{(e)}(𝐪,I)=\underset{𝐤}{}\mathrm{\Lambda }_e(𝐤,𝐪)[d_I(𝐪),a_𝐤(𝐪)]+\underset{𝐤}{}\mathrm{\Lambda }_e(𝐤,𝐪)[d_I(𝐪),a_𝐤^{}(𝐪)]$$
(104)
$$\rho ^{(h)}(𝐪,I)=\underset{𝐤}{}\mathrm{\Lambda }_h(𝐤,𝐪)[d_I(𝐪),a_𝐤(𝐪)]\underset{𝐤}{}\mathrm{\Lambda }_h(𝐤,𝐪)[d_I(𝐪),a_𝐤^{}(𝐪)]$$
(105)
$$(\omega _{I\sigma }(𝐪)\mathrm{\Omega }_{LO})[d_{I\sigma }(𝐪),b_𝐪^{}]=S(\sigma )\frac{M_𝐪}{\sqrt{V}}\rho ^{(\sigma )}(𝐪,I)$$
(106)
$$(\omega _{I\sigma }(𝐪)+\mathrm{\Omega }_{LO})[d_{I\sigma }(𝐪),b_𝐪]=S(\sigma )\frac{M_𝐪}{\sqrt{V}}\rho ^{(\sigma )}(𝐪,I)$$
(107)
From this we have the following fact that $`\omega _I`$ are zeros of the dielectric function,
$$ϵ^{(e)}(𝐪,\omega _I^{(e)})=0$$
(108)
$$ϵ^{(h)}(𝐪,\omega _I^{(h)})=0$$
(109)
$$ϵ^{(e,h)}(𝐪,\omega )=1v(𝐪,\omega )P^{(e,h)}(𝐪,\omega )$$
(110)
and,
$$v(𝐪,\omega )=v(𝐪)+\frac{2\mathrm{\Omega }_{LO}M_𝐪^2}{\omega ^2\mathrm{\Omega }_{LO}^2}$$
(111)
and $`P^{(e,h)}(𝐪,\omega )`$ is the usual RPA-polarization bubble. If we now make use of the fact that $`[d_{I\sigma }(𝐪),d_{I\sigma }^{}(𝐪)]=1`$ then we have,
$`{\displaystyle \underset{𝐤}{}}|[d_{I\sigma }(𝐪),a_{𝐤\sigma }^{}(𝐪\sigma )]|^2{\displaystyle \underset{𝐤}{}}|[d_{I\sigma }(𝐪),a_{𝐤\sigma }(𝐪\sigma )]|^2`$
$$+|[d_{I\sigma }(𝐪),b_𝐪^{}]|^2|[d_{I\sigma }(𝐪),b_𝐪]|^2=1$$
(112)
$`{\displaystyle \underset{𝐤}{}}{\displaystyle \frac{\mathrm{\Lambda }_\sigma ^2(𝐤,𝐪)\mathrm{\Lambda }_\sigma ^2(𝐤,𝐪)}{(\omega _{I\sigma }(𝐪)S(\sigma )\frac{𝐤.𝐪}{m_\sigma })^2}}[\rho ^{(\sigma )}(𝐪,I)]^2({\displaystyle \frac{v(𝐪,I\sigma )}{V}})^2`$
$$+\frac{M_𝐪^2}{V}[\rho ^{(\sigma )}(𝐪,I)]^2\frac{4\omega _{I\sigma }(𝐪)\mathrm{\Omega }_{LO}}{(\omega _{I\sigma }^2(𝐪)\mathrm{\Omega }_{LO}^2)^2}=1$$
(113)
Therefore,
$$\rho ^{(e)}(𝐪,I)=\{(\frac{v(𝐪,I,e)}{V})^2\underset{𝐤}{}\frac{\overline{n}^{(e)}(𝐤𝐪/2)\overline{n}^{(e)}(𝐤+𝐪/2)}{(\omega _{I,e}(𝐪)\frac{𝐤.𝐪}{m_e})^2}+\frac{M_𝐪^2}{V}\frac{4\omega _{I,e}(𝐪)\mathrm{\Omega }_{LO}}{(\omega _{I,e}^2(𝐪)\mathrm{\Omega }_{LO}^2)^2}\}^{\frac{1}{2}}$$
(114)
$$\rho ^{(h)}(𝐪,I)=\{(\frac{v(𝐪,I,h)}{V})^2\underset{𝐤}{}\frac{\overline{n}^{(h)}(𝐤+𝐪/2)\overline{n}^{(h)}(𝐤𝐪/2)}{(\omega _{I,h}(𝐪)+\frac{𝐤.𝐪}{m_h})^2}+\frac{M_𝐪^2}{V}\frac{4\omega _{I,h}(𝐪)\mathrm{\Omega }_{LO}}{(\omega _{I,h}^2(𝐪)\mathrm{\Omega }_{LO}^2)^2}\}^{\frac{1}{2}}$$
(115)
Therefore,
$$a_{𝐤\sigma }^{}(𝐪\sigma )a_{𝐤\sigma }(𝐪\sigma )=\underset{I}{}([a_{𝐤\sigma }(𝐪\sigma ),d_{I\sigma }(𝐪)])^2=\underset{I}{}(\frac{v(𝐪,I\sigma )}{V})^2\mathrm{\Lambda }_\sigma ^2(𝐤,𝐪)\frac{\rho ^{2\sigma }(𝐪,I)}{(\omega _{I\sigma }(𝐪)+S(\sigma )\frac{𝐤.𝐪}{m_\sigma })^2}$$
(116)
$$a_𝐤^{}(𝐪)a_𝐤(𝐪)=\underset{I}{}(\frac{v(𝐪,I,e)}{V})^2\mathrm{\Lambda }_e^2(𝐤,𝐪)\frac{(\rho ^{(e)}(𝐪,I))^2}{(\omega _{I,e}(𝐪)+\frac{𝐤.𝐪}{m_e})^2}$$
(117)
$$a_𝐤^{}(𝐪)a_𝐤(𝐪)=\underset{I}{}(\frac{v(𝐪,I,h)}{V})^2\mathrm{\Lambda }_h^2(𝐤,𝐪)\frac{(\rho ^{(h)}(𝐪,I))^2}{(\omega _{I,h}(𝐪)\frac{𝐤.𝐪}{m_h})^2}$$
(118)
After some algebra it is clear that Eqs.( 73) and ( 74) are identical to Eqs.( 117) and ( 118) respectively. Let us now solve the fundamental equations namely Eq.( 34) and Eq.( 35). For this we first would like to decompose the various fields in the exciton basis.
$$a_𝐤(\mathrm{𝟎})=\underset{I}{}\stackrel{~}{\phi }_I(𝐤)e^{iϵ_I\text{ }t}\stackrel{~}{D}_I$$
(119)
$$a_𝐤(\mathrm{𝟎})=e^{i(ϵ^e(𝐤)+ϵ^h(𝐤))\text{ }t}\stackrel{~}{a}_𝐤(\mathrm{𝟎})$$
(120)
This means we may rewrite these equations as follows :
$`i{\displaystyle \frac{}{t}}\stackrel{~}{D}_I(t)=({\displaystyle \frac{|e|}{\mu c}})e^{iϵ_I\text{ }t}\stackrel{}{A}_{ext}^{}(t).\stackrel{}{p}_{vc}{\displaystyle \underset{𝐤}{}}\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})\stackrel{~}{\phi }_I^{}(𝐤)`$
$`+{\displaystyle \underset{𝐐0}{}}{\displaystyle \frac{v_{eh}(𝐐)}{V}}[{\displaystyle \underset{𝐤,J}{}}\stackrel{~}{\phi }_I^{}(𝐤)(1\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})\mathrm{\Lambda }_1(𝐤𝐐,\mathrm{𝟎}))\stackrel{~}{\phi }_J(𝐤𝐐)e^{i(ϵ_Iϵ_J)\text{ }t}\stackrel{~}{D}_J(t)`$
$$\underset{𝐤}{}\stackrel{~}{\phi }_I^{}(𝐤)\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})\mathrm{\Lambda }_2(𝐤𝐐,\mathrm{𝟎})e^{iϵ_I\text{ }t}e^{i(ϵ^e(𝐤𝐐)+ϵ^h(𝐤𝐐))t}\stackrel{~}{a}_{𝐤𝐐}^{}(\mathrm{𝟎})]$$
(121)
$`i{\displaystyle \frac{}{t}}\stackrel{~}{a}_𝐤(\mathrm{𝟎})={\displaystyle \underset{𝐐0}{}}{\displaystyle \frac{v_{eh}(𝐐)}{V}}\mathrm{\Lambda }_2(𝐤,\mathrm{𝟎})[\mathrm{\Lambda }_1(𝐤+𝐐,\mathrm{𝟎})\stackrel{~}{\phi }_J^{}(𝐤+𝐐)e^{iϵ_J\text{ }t}e^{i(ϵ^e(𝐤)+ϵ^h(𝐤))\text{ }t}\stackrel{~}{D}_J^{}`$
$$+\mathrm{\Lambda }_2(𝐤+𝐐,\mathrm{𝟎})e^{i(ϵ^e(𝐤)+ϵ^h(𝐤))\text{ }t}e^{i(ϵ^e(𝐤+𝐐)+ϵ^h(𝐤+𝐐))\text{ }t}\stackrel{~}{a}_{𝐤+𝐐}(\mathrm{𝟎})]+(\frac{|e|}{\mu c})\stackrel{}{A}_{ext}(t).\stackrel{}{p}_{vc}\text{ }\mathrm{\Lambda }_2(𝐤,\mathrm{𝟎})e^{i(ϵ^e(𝐤)+ϵ^h(𝐤))\text{ }t}$$
(122)
$`i{\displaystyle \frac{}{t}}\stackrel{~}{D}_0(t)=({\displaystyle \frac{|e|}{\mu c}})e^{iϵ_0\text{ }t}\stackrel{}{A}_{ext}^{}(t).\stackrel{}{p}_{vc}{\displaystyle \underset{𝐤}{}}\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})\stackrel{~}{\phi }_0^{}(𝐤)`$
$`+{\displaystyle \underset{𝐤^{^{}}𝐤}{}}{\displaystyle \frac{v_{eh}(𝐤𝐤^{^{}})}{V}}\stackrel{~}{\phi }_0^{}(𝐤)(1\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})\mathrm{\Lambda }_1(𝐤^{^{}},\mathrm{𝟎}))\stackrel{~}{\phi }_0(𝐤^{^{}})\stackrel{~}{D}_0(t)`$
$`+{\displaystyle \underset{𝐤^{^{}}𝐤}{}}{\displaystyle \frac{v_{eh}(𝐤𝐤^{^{}})}{V}}[\stackrel{~}{\phi }_0^{}(𝐤)(1\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})\mathrm{\Lambda }_1(𝐤^{^{}},\mathrm{𝟎}))e^{i(ϵ_0ϵ_𝐤^{^{}})\text{ }t}\stackrel{~}{D}_𝐤^{^{}}(t)`$
$$\stackrel{~}{\phi }_0^{}(𝐤)\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})\mathrm{\Lambda }_2(𝐤^{^{}},\mathrm{𝟎})e^{i(ϵ_0ϵ_𝐤^{^{}})\text{ }t}\stackrel{~}{a}_𝐤^{^{}}^{}(\mathrm{𝟎})]$$
(123)
$`i{\displaystyle \frac{}{t}}\stackrel{~}{D}_𝐤(t)=({\displaystyle \frac{|e|}{\mu c}})e^{iϵ_𝐤\text{ }t}\stackrel{}{A}_{ext}^{}(t).\stackrel{}{p}_{vc}\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})`$
$`+{\displaystyle \underset{𝐤^{^{}}𝐤}{}}{\displaystyle \frac{v_{eh}(𝐤𝐤^{^{}})}{V}}(1\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})\mathrm{\Lambda }_1(𝐤^{^{}},\mathrm{𝟎}))\stackrel{~}{\phi }_0(𝐤^{^{}})e^{i(ϵ_𝐤ϵ_0)\text{ }t}\stackrel{~}{D}_0(t)`$
$`+{\displaystyle \underset{𝐤^{^{}}𝐤}{}}{\displaystyle \frac{v_{eh}(𝐤𝐤^{^{}})}{V}}[(1\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})\mathrm{\Lambda }_1(𝐤^{^{}},\mathrm{𝟎}))e^{i(ϵ_𝐤ϵ_𝐤^{^{}})\text{ }t}\stackrel{~}{D}_𝐤^{^{}}(t)`$
$$\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})\mathrm{\Lambda }_2(𝐤^{^{}},\mathrm{𝟎})e^{i(ϵ_𝐤ϵ_𝐤^{^{}})\text{ }t}\stackrel{~}{a}_𝐤^{^{}}^{}(\mathrm{𝟎})]$$
(124)
$`i{\displaystyle \frac{}{t}}\stackrel{~}{a}_𝐤^{}(\mathrm{𝟎})={\displaystyle \underset{𝐤^{^{}}𝐤}{}}{\displaystyle \frac{v_{eh}(𝐤𝐤^{^{}})}{V}}\mathrm{\Lambda }_2(𝐤,\mathrm{𝟎})\mathrm{\Lambda }_1(𝐤^{^{}},\mathrm{𝟎})\stackrel{~}{\phi }_0(𝐤^{^{}})e^{i(ϵ_0ϵ_𝐤)\text{ }t}\stackrel{~}{D}_0(t)`$
$`+{\displaystyle \underset{𝐤^{^{}}𝐤}{}}{\displaystyle \frac{v_{eh}(𝐤𝐤^{^{}})}{V}}\mathrm{\Lambda }_2(𝐤,\mathrm{𝟎})[\mathrm{\Lambda }_1(𝐤^{^{}},\mathrm{𝟎})e^{i(ϵ_𝐤^{^{}}ϵ_𝐤)\text{ }t}\stackrel{~}{D}_𝐤^{^{}}(t)`$
$$+\mathrm{\Lambda }_2(𝐤^{^{}},\mathrm{𝟎})e^{i(ϵ_𝐤ϵ_𝐤^{^{}})t}\stackrel{~}{a}_𝐤^{^{}}^{}(\mathrm{𝟎})](\frac{|e|}{\mu c})\stackrel{}{A}^{}_{ext}(t).\stackrel{}{p}_{vc}\text{ }\mathrm{\Lambda }_2(𝐤,\mathrm{𝟎})e^{iϵ_𝐤\text{ }t}$$
(125)
$$\stackrel{~}{\phi }_0(𝐤)=\frac{1}{\sqrt{V}}(4\pi )\sqrt{\frac{1}{\pi a_X^3}}\frac{2\text{ }/a_X}{(1/a_X^2+k^2)^2}$$
(126)
Furthermore,
$`\overline{n}_e(𝐤)=\overline{n}_0(𝐤)+{\displaystyle \underset{I,𝐪}{}}({\displaystyle \frac{v(𝐪,I,e)}{V}})^2\overline{n}_e(𝐤𝐪)(1\overline{n}_e(𝐤)){\displaystyle \frac{(\rho ^{(e)}(𝐪,I))^2}{(\omega _{I,e}(𝐪)+\frac{𝐤.𝐪}{m_e}\frac{q^2}{2m_e})^2}}`$
$$\underset{I,𝐪}{}(\frac{v(𝐪,I,e)}{V})^2\overline{n}_e(𝐤)(1\overline{n}_e(𝐤+𝐪))\frac{(\rho ^{(e)}(𝐪,I))^2}{(\omega _{I,e}(𝐪)+\frac{𝐤.𝐪}{m_e}+\frac{q^2}{2m_e})^2}$$
(127)
$`\overline{n}_h(𝐤)=\overline{n}_0(𝐤){\displaystyle \underset{I,𝐪}{}}({\displaystyle \frac{v(𝐪,I,h)}{V}})^2\overline{n}_h(𝐤)(1\overline{n}_h(𝐤𝐪)){\displaystyle \frac{(\rho ^{(h)}(𝐪,I))^2}{(\omega _{I,h}(𝐪)\frac{𝐤.𝐪}{m_h}+\frac{q^2}{2m_h})^2}}`$
$$+\underset{I,𝐪}{}(\frac{v(𝐪,I,h)}{V})^2\overline{n}_h(𝐤+𝐪)(1\overline{n}_h(𝐤))\frac{(\rho ^{(h)}(𝐪,I))^2}{(\omega _{I,h}(𝐪)\frac{𝐤.𝐪}{m_h}\frac{q^2}{2m_h})^2}$$
(128)
$`\overline{n}_0(𝐤)=|\phi _0(𝐤)|^2\stackrel{~}{D}_0^{}(t)\stackrel{~}{D}_0(t)+\stackrel{~}{D}_𝐤^{}(t)\stackrel{~}{D}_𝐤(t)`$
$$+\phi _0(𝐤)e^{i(ϵ_𝐤ϵ_0)t}\stackrel{~}{D}_𝐤^{}(t)\stackrel{~}{D}_0(t)+\phi _0(𝐤)e^{i(ϵ_𝐤ϵ_0)t}\stackrel{~}{D}_0^{}(t)\stackrel{~}{D}_𝐤(t)\stackrel{~}{a}_𝐤^{}(\mathrm{𝟎})\stackrel{~}{a}_𝐤(\mathrm{𝟎})$$
(129)
or,
$`\overline{n}_0(𝐤)=|\phi _0(𝐤)|^2(\stackrel{~}{D}_0^r(t)\stackrel{~}{D}_0^r(t)+\stackrel{~}{D}_0^i(t)\stackrel{~}{D}_0^i(t))+\stackrel{~}{D}_𝐤^r(t)\stackrel{~}{D}_𝐤^r(t)+\stackrel{~}{D}_𝐤^i(t)\stackrel{~}{D}_𝐤^i(t)`$
$`+2\phi _0(𝐤)cos((ϵ_𝐤ϵ_0)t)[\stackrel{~}{D}_𝐤^r(t)\stackrel{~}{D}_0^r(t)+\stackrel{~}{D}_𝐤^i(t)\stackrel{~}{D}_0^i(t)]`$
$$2\phi _0(𝐤)sin((ϵ_𝐤ϵ_0)t)[\stackrel{~}{D}_𝐤^r(t)\stackrel{~}{D}_0^i(t)\stackrel{~}{D}_𝐤^i(t)\stackrel{~}{D}_0^r(t)]\stackrel{~}{a}_𝐤^r(\mathrm{𝟎})\stackrel{~}{a}_𝐤^r(\mathrm{𝟎})\stackrel{~}{a}_𝐤^i(\mathrm{𝟎})\stackrel{~}{a}_𝐤^i(\mathrm{𝟎})$$
(130)
$$\overline{n}_e(𝐤)=\overline{n}_0(𝐤)A_e(𝐤)+(1\overline{n}_0(𝐤))B_e(𝐤)$$
(131)
$$\overline{n}_h(𝐤)=\overline{n}_0(𝐤)A_h(𝐤)+(1\overline{n}_0(𝐤))B_h(𝐤)$$
(132)
$$A_e(𝐤)=\frac{1}{1+\frac{T_2^e(𝐤)}{1+T_1^e(𝐤)}}$$
(133)
$$B_e(𝐤)=\frac{1}{1+\frac{1+T_2^e(𝐤)}{T_1^e(𝐤)}}$$
(134)
$$T_1^e(𝐤)=\underset{I,𝐪}{}(\frac{v(𝐪,I,e)}{V})^2\overline{n}_e(𝐤𝐪)\frac{(\rho ^{(e)}(𝐪,I))^2}{(\omega _{I,e}(𝐪)+\frac{𝐤.𝐪}{m_e}\frac{q^2}{2m_e})^2}$$
(135)
$$T_2^e(𝐤)=\underset{I,𝐪}{}(\frac{v(𝐪,I,e)}{V})^2(1\overline{n}_e(𝐤+𝐪))\frac{(\rho ^{(e)}(𝐪,I))^2}{(\omega _{I,e}(𝐪)+\frac{𝐤.𝐪}{m_e}+\frac{q^2}{2m_e})^2}$$
(136)
$$A_h(𝐤)=\frac{1}{1+\frac{T_1^h(𝐤)}{1+T_2^h(𝐤)}}$$
(137)
$$B_h(𝐤)=\frac{1}{1+\frac{1+T_1^h(𝐤)}{T_2^h(𝐤)}}$$
(138)
$$T_1^h(𝐤)=\underset{I,𝐪}{}(\frac{v(𝐪,I,h)}{V})^2(1\overline{n}_h(𝐤𝐪))\frac{(\rho ^{(h)}(𝐪,I))^2}{(\omega _{I,h}(𝐪)\frac{𝐤.𝐪}{m_h}+\frac{q^2}{2m_h})^2}$$
(139)
$$T_2^h(𝐤)=\underset{I,𝐪}{}(\frac{v(𝐪,I,h)}{V})^2\overline{n}_h(𝐤+𝐪)\frac{(\rho ^{(h)}(𝐪,I))^2}{(\omega _{I,h}(𝐪)\frac{𝐤.𝐪}{m_h}\frac{q^2}{2m_h})^2}$$
(140)
$`{\displaystyle \frac{}{t}}\stackrel{~}{D}_0^r(t)=({\displaystyle \frac{|e|}{\mu c}})A_X^i(ϵ_0,t)p_{vc}{\displaystyle \underset{𝐤}{}}\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})\stackrel{~}{\phi }_0(𝐤)`$
$`+{\displaystyle \underset{𝐤^{^{}}𝐤}{}}{\displaystyle \frac{v_{eh}(𝐤𝐤^{^{}})}{V}}\stackrel{~}{\phi }_0(𝐤)(1\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})\mathrm{\Lambda }_1(𝐤^{^{}},\mathrm{𝟎}))\stackrel{~}{\phi }_0(𝐤^{^{}})\stackrel{~}{D}_0^i(t)`$
$`+{\displaystyle \underset{𝐤^{^{}}𝐤}{}}{\displaystyle \frac{v_{eh}(𝐤𝐤^{^{}})}{V}}sin((ϵ_0ϵ_𝐤^{^{}})\text{ }t)[\stackrel{~}{\phi }_0(𝐤)(1\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})\mathrm{\Lambda }_1(𝐤^{^{}},\mathrm{𝟎}))\stackrel{~}{D}_𝐤^{^{}}^r(t)`$
$`\stackrel{~}{\phi }_0(𝐤)\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})\mathrm{\Lambda }_2(𝐤^{^{}},\mathrm{𝟎})\stackrel{~}{a}_𝐤^{^{}}^r(\mathrm{𝟎})]`$
$`+{\displaystyle \underset{𝐤^{^{}}𝐤}{}}{\displaystyle \frac{v_{eh}(𝐤𝐤^{^{}})}{V}}cos((ϵ_0ϵ_𝐤^{^{}})\text{ }t)[\stackrel{~}{\phi }_0(𝐤)(1\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})\mathrm{\Lambda }_1(𝐤^{^{}},\mathrm{𝟎}))\stackrel{~}{D}_𝐤^{^{}}^i(t)`$
$$+\stackrel{~}{\phi }_0(𝐤)\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})\mathrm{\Lambda }_2(𝐤^{^{}},\mathrm{𝟎})\stackrel{~}{a}_𝐤^{^{}}^i(\mathrm{𝟎})]$$
(141)
$`{\displaystyle \frac{}{t}}\stackrel{~}{D}_0^i(t)=({\displaystyle \frac{|e|}{\mu c}})A_X^r(ϵ_0,t)p_{vc}{\displaystyle \underset{𝐤}{}}\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})\stackrel{~}{\phi }_0(𝐤)`$
$`{\displaystyle \underset{𝐤^{^{}}𝐤}{}}{\displaystyle \frac{v_{eh}(𝐤𝐤^{^{}})}{V}}\stackrel{~}{\phi }_0(𝐤)(1\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})\mathrm{\Lambda }_1(𝐤^{^{}},\mathrm{𝟎}))\stackrel{~}{\phi }_0(𝐤^{^{}})\stackrel{~}{D}_0^r(t)`$
$`{\displaystyle \underset{𝐤^{^{}}𝐤}{}}{\displaystyle \frac{v_{eh}(𝐤𝐤^{^{}})}{V}}cos((ϵ_0ϵ_𝐤^{^{}})\text{ }t)[\stackrel{~}{\phi }_0(𝐤)(1\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})\mathrm{\Lambda }_1(𝐤^{^{}},\mathrm{𝟎}))\stackrel{~}{D}_𝐤^{^{}}^r(t)`$
$`\stackrel{~}{\phi }_0(𝐤)\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})\mathrm{\Lambda }_2(𝐤^{^{}},\mathrm{𝟎})\stackrel{~}{a}_𝐤^{^{}}^r(\mathrm{𝟎})]`$
$`+{\displaystyle \underset{𝐤^{^{}}𝐤}{}}{\displaystyle \frac{v_{eh}(𝐤𝐤^{^{}})}{V}}sin((ϵ_0ϵ_𝐤^{^{}})\text{ }t)[\stackrel{~}{\phi }_0(𝐤)(1\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})\mathrm{\Lambda }_1(𝐤^{^{}},\mathrm{𝟎}))\stackrel{~}{D}_𝐤^{^{}}^i(t)`$
$$+\stackrel{~}{\phi }_0(𝐤)\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})\mathrm{\Lambda }_2(𝐤^{^{}},\mathrm{𝟎})\stackrel{~}{a}_𝐤^{^{}}^i(\mathrm{𝟎})]$$
(142)
$`{\displaystyle \frac{}{t}}\stackrel{~}{D}_𝐤^r(t)=({\displaystyle \frac{|e|}{\mu c}})A_X^i(ϵ_𝐤,t)p_{vc}\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})`$
$`+{\displaystyle \underset{𝐤^{^{}}𝐤}{}}{\displaystyle \frac{v_{eh}(𝐤𝐤^{^{}})}{V}}(1\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})\mathrm{\Lambda }_1(𝐤^{^{}},\mathrm{𝟎}))\stackrel{~}{\phi }_0(𝐤^{^{}})[sin((ϵ_𝐤ϵ_0)\text{ }t)\stackrel{~}{D}_0^r(t)+cos((ϵ_𝐤ϵ_0)\text{ }t)\stackrel{~}{D}_0^i(t)]`$
$`+{\displaystyle \underset{𝐤^{^{}}𝐤}{}}{\displaystyle \frac{v_{eh}(𝐤𝐤^{^{}})}{V}}[(1\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})\mathrm{\Lambda }_1(𝐤^{^{}},\mathrm{𝟎}))\{sin((ϵ_𝐤ϵ_𝐤^{^{}})\text{ }t)\stackrel{~}{D}_𝐤^{^{}}^r(t)+cos((ϵ_𝐤ϵ_𝐤^{^{}})\text{ }t)\stackrel{~}{D}_𝐤^{^{}}^i(t)\}`$
$$\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})\mathrm{\Lambda }_2(𝐤^{^{}},\mathrm{𝟎})\{sin((ϵ_𝐤ϵ_𝐤^{^{}})\text{ }t)\stackrel{~}{a}_𝐤^{^{}}^r(\mathrm{𝟎})cos((ϵ_𝐤ϵ_𝐤^{^{}})\text{ }t)\stackrel{~}{a}_𝐤^{^{}}^i(\mathrm{𝟎})\}]$$
(143)
$`{\displaystyle \frac{}{t}}\stackrel{~}{D}_𝐤^i(t)=({\displaystyle \frac{|e|}{\mu c}})A_X^r(ϵ_𝐤,t)p_{vc}\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})`$
$`+{\displaystyle \underset{𝐤^{^{}}𝐤}{}}{\displaystyle \frac{v_{eh}(𝐤𝐤^{^{}})}{V}}(1\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})\mathrm{\Lambda }_1(𝐤^{^{}},\mathrm{𝟎}))\stackrel{~}{\phi }_0(𝐤^{^{}})\{cos((ϵ_𝐤ϵ_0)\text{ }t)\stackrel{~}{D}_0^r(t)sin((ϵ_𝐤ϵ_0)\text{ }t)\stackrel{~}{D}_0^i(t)\}`$
$`+{\displaystyle \underset{𝐤^{^{}}𝐤}{}}{\displaystyle \frac{v_{eh}(𝐤𝐤^{^{}})}{V}}[(1\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})\mathrm{\Lambda }_1(𝐤^{^{}},\mathrm{𝟎}))\{cos((ϵ_𝐤ϵ_𝐤^{^{}})\text{ }t)\stackrel{~}{D}_𝐤^{^{}}^r(t)sin((ϵ_𝐤ϵ_𝐤^{^{}})\text{ }t)\stackrel{~}{D}_𝐤^{^{}}^i(t)\}`$
$$\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})\mathrm{\Lambda }_2(𝐤^{^{}},\mathrm{𝟎})\{cos((ϵ_𝐤ϵ_𝐤^{^{}})\text{ }t)\stackrel{~}{a}_𝐤^{^{}}^r(\mathrm{𝟎})+sin((ϵ_𝐤ϵ_𝐤^{^{}})\text{ }t)\stackrel{~}{a}_𝐤^{^{}}^i(\mathrm{𝟎})]$$
(144)
$`{\displaystyle \frac{}{t}}\stackrel{~}{a}_𝐤^i(\mathrm{𝟎})={\displaystyle \underset{𝐤^{^{}}𝐤}{}}{\displaystyle \frac{v_{eh}(𝐤𝐤^{^{}})}{V}}\mathrm{\Lambda }_2(𝐤,\mathrm{𝟎})\mathrm{\Lambda }_1(𝐤^{^{}},\mathrm{𝟎})\stackrel{~}{\phi }_0(𝐤^{^{}})\{cos((ϵ_0ϵ_𝐤)\text{ }t)\stackrel{~}{D}_0^r(t)+sin((ϵ_0ϵ_𝐤)\text{ }t)\stackrel{~}{D}_0^i(t)\}`$
$`+{\displaystyle \underset{𝐤^{^{}}𝐤}{}}{\displaystyle \frac{v_{eh}(𝐤𝐤^{^{}})}{V}}\mathrm{\Lambda }_2(𝐤,\mathrm{𝟎})[\mathrm{\Lambda }_1(𝐤^{^{}},\mathrm{𝟎})\{cos((ϵ_𝐤^{^{}}ϵ_𝐤)\text{ }t)\stackrel{~}{D}_𝐤^{^{}}^r(t)+sin((ϵ_𝐤^{^{}}ϵ_𝐤)\text{ }t)\stackrel{~}{D}_𝐤^{^{}}^i(t)\}`$
$$+\mathrm{\Lambda }_2(𝐤^{^{}},\mathrm{𝟎})\{cos((ϵ_𝐤ϵ_𝐤^{^{}})t)\stackrel{~}{a}_𝐤^{^{}}^r(\mathrm{𝟎})+sin((ϵ_𝐤ϵ_𝐤^{^{}})t)\stackrel{~}{a}_𝐤^{^{}}^i(\mathrm{𝟎})](\frac{|e|}{\mu c})A_X^r(ϵ_𝐤,t)p_{vc}\text{ }\mathrm{\Lambda }_2(𝐤,\mathrm{𝟎})$$
(145)
$`{\displaystyle \frac{}{t}}\stackrel{~}{a}_𝐤^r(\mathrm{𝟎})={\displaystyle \underset{𝐤^{^{}}𝐤}{}}{\displaystyle \frac{v_{eh}(𝐤𝐤^{^{}})}{V}}\mathrm{\Lambda }_2(𝐤,\mathrm{𝟎})\mathrm{\Lambda }_1(𝐤^{^{}},\mathrm{𝟎})\stackrel{~}{\phi }_0(𝐤^{^{}})\{sin((ϵ_0ϵ_𝐤)\text{ }t)\stackrel{~}{D}_0^r(t)+cos((ϵ_0ϵ_𝐤)\text{ }t)\stackrel{~}{D}_0^i(t)\}`$
$`+{\displaystyle \underset{𝐤^{^{}}𝐤}{}}{\displaystyle \frac{v_{eh}(𝐤𝐤^{^{}})}{V}}\mathrm{\Lambda }_2(𝐤,\mathrm{𝟎})[\mathrm{\Lambda }_1(𝐤^{^{}},\mathrm{𝟎})\{sin((ϵ_𝐤^{^{}}ϵ_𝐤)\text{ }t)\stackrel{~}{D}_𝐤^{^{}}^r(t)+cos((ϵ_𝐤^{^{}}ϵ_𝐤)\text{ }t)\stackrel{~}{D}_𝐤^{^{}}^i(t)\}`$
$`+\mathrm{\Lambda }_2(𝐤^{^{}},\mathrm{𝟎})\{sin((ϵ_𝐤ϵ_𝐤^{^{}})t)\stackrel{~}{a}_𝐤^{^{}}^r(\mathrm{𝟎})cos((ϵ_𝐤ϵ_𝐤^{^{}})t)\stackrel{~}{a}_𝐤^{^{}}^i(\mathrm{𝟎})\}]`$
$$+(\frac{|e|}{\mu c})A_X^i(ϵ_𝐤,t)p_{vc}\text{ }\mathrm{\Lambda }_2(𝐤,\mathrm{𝟎})$$
(146)
In order to simplify the calculations further, let us define,
$`P_i(𝐪,\omega )=({\displaystyle \frac{1}{4\pi }}){\displaystyle _0^{\mathrm{}}}dk\text{ }k^2\text{ }\overline{n}(𝐤)({\displaystyle \frac{m}{|k||q|}})[\theta (\omega {\displaystyle \frac{|k||q|}{m}}{\displaystyle \frac{q^2}{2m}})\theta (\omega +{\displaystyle \frac{|k||q|}{m}}{\displaystyle \frac{q^2}{2m}})`$
$$\theta (\omega \frac{|k||q|}{m}+\frac{q^2}{2m})+\theta (\omega +\frac{|k||q|}{m}+\frac{q^2}{2m})]$$
(147)
$$P_r(𝐪,\omega )=_0^{\mathrm{}}\frac{d\omega ^{^{}}}{\pi }\text{ }P_i(𝐪,\omega ^{^{}})(\frac{2\omega ^{^{}}}{\omega ^{}_{}{}^{}2\omega ^2})$$
(148)
$$S(𝐪,\omega )\stackrel{~}{}\frac{P_i(𝐪,\omega )}{(1v(𝐪,\omega )P_r(𝐪,\omega ))^2+v^2(𝐪,\omega )(P_i(𝐪,\omega ))^2}$$
(149)
$$\rho (𝐪,\omega )=\{(\frac{v(𝐪,\omega )}{V})^2V_0^{\mathrm{}}\frac{d\omega ^{^{}}}{\pi }P_i(𝐪,\omega ^{^{}})\frac{4\omega \omega ^{^{}}}{(\omega ^{}_{}{}^{}2\omega ^2)^2}+\frac{M_𝐪^2}{V}\frac{4\omega \mathrm{\Omega }_{LO}}{(\omega ^2\mathrm{\Omega }_{LO}^2)^2}\}^{\frac{1}{2}}$$
(150)
## II Comparison with Semiconductor Bloch Equations
The equations presented in the previous section namely Eqs.( 121) and ( 122) are intended as alternatives to the usual Semiconductor Bloch equations used to study semiconductors. Let us now try to solve this system with a pulse field. That is, we apply an external field with central frequency $`\omega _X`$ and assume it lasts for a time $`\tau _X`$ and therefore we have a spread in frequency $`\mathrm{\Gamma }_X=2\pi /\tau _X`$. We would like to see the evolution of the polarization and populations in this case. This exercise also enables us to compare our results with those of the SBEs and ascertain where the differences lie. To this end let us set,
$$\stackrel{}{A}_{ext}(t)=\widehat{p}_{vc}A_X(0,t)$$
(151)
where,
$$A_X(E,t)=A_0_{\mathrm{}}^{\mathrm{}}𝑑\omega \text{ }\frac{\mathrm{\Gamma }_X/\pi }{(\omega \omega _X)^2+\mathrm{\Gamma }_X^2}e^{i(\omega E)\text{ }t}$$
(152)
For comparison the SBE is reproduced below.
$`g_{hh}(𝐤t)=i\overline{n}_h(𝐤)=id_𝐤^{}d_𝐤,\text{ }=i\text{ }f(𝐤)g_{ee}(𝐤t)=i\overline{n}_e(𝐤)=ic_𝐤^{}c_𝐤`$
$`g_{he}(𝐤t)=id_𝐤(t)c_𝐤(t)=i\text{ }p(𝐤)`$
$`i{\displaystyle \frac{}{t}}g_{hh}(𝐤t)=2\text{ }Re(\mathrm{\Omega }(𝐤t)g_{he}^{}(𝐤t))+R_{hh}(𝐤t)`$
$`i{\displaystyle \frac{}{t}}g_{he}(𝐤t)=\mathrm{\Omega }(𝐤t)(i2g_{hh}(𝐤t))+(ϵ_h(𝐤)+ϵ_c(𝐤)2\mathrm{\Sigma }(𝐤t))g_{he}(𝐤t)+R_{he}(𝐤t)`$
$`\mathrm{\Omega }(𝐤t)={\displaystyle \frac{|e|}{\mu c}}𝐀_{ext}^{}(t).𝐩_{\mathrm{𝐯𝐜}}i{\displaystyle \underset{𝐤^{^{}}}{}}v_{𝐤𝐤^{^{}}}g_{he}(𝐤^{^{}}t)`$
$$\mathrm{\Sigma }(𝐤t)=i\underset{𝐤^{^{}}}{}v_{𝐤𝐤^{^{}}}g_{hh}(𝐤^{^{}}t)$$
(153)
Define,
$$\mathrm{\Omega }(𝐤,t)=\stackrel{~}{\mathrm{\Omega }}(𝐤,t)e^{i(k^2/2\mu +E_g)t},\text{ }p(𝐤,t)=\stackrel{~}{p}(𝐤,t)e^{i(k^2/2\mu +E_g)t}$$
(154)
$$\frac{f(𝐤)}{t}=2\stackrel{~}{\mathrm{\Omega }}_R(𝐤,t)\stackrel{~}{p}_I(𝐤)2\stackrel{~}{\mathrm{\Omega }}_I(𝐤,t)\stackrel{~}{p}_R(𝐤)$$
(155)
$$\frac{\stackrel{~}{p}_R(𝐤)}{t}=\stackrel{~}{\mathrm{\Omega }}_R(𝐤,t)(12f(𝐤))2\mathrm{\Sigma }(𝐤,t)\stackrel{~}{p}_I(𝐤,t)$$
(156)
$$\frac{\stackrel{~}{p}_I(𝐤)}{t}=\stackrel{~}{\mathrm{\Omega }}_I(𝐤,t)(12f(𝐤))+2\mathrm{\Sigma }(𝐤,t)\stackrel{~}{p}_R(𝐤,t)$$
(157)
$$\stackrel{~}{\mathrm{\Omega }}_R(𝐤,t)=(\frac{|e|}{\mu c})\stackrel{~}{A}_{ext}(t)p_{vc}\text{ }cos((k^2/2\mu +E_g\omega _X)t)+\underset{𝐤^{^{}}𝐤}{}v_{𝐤𝐤^{^{}}}[\stackrel{~}{p}_R(𝐤^{^{}})cos((k^2k^{}_{}{}^{}2)t/2\mu )\stackrel{~}{p}_I(𝐤^{^{}})sin((k^2k^{}_{}{}^{}2)t/2\mu )]$$
(158)
$$\stackrel{~}{\mathrm{\Omega }}_I(𝐤,t)=(\frac{|e|}{\mu c})\stackrel{~}{A}_{ext}(t)p_{vc}\text{ }sin((k^2/2\mu +E_g\omega _X)t)+\underset{𝐤^{^{}}𝐤}{}v_{𝐤𝐤^{^{}}}[\stackrel{~}{p}_I(𝐤^{^{}})cos((k^2k^{}_{}{}^{}2)t/2\mu )+\stackrel{~}{p}_R(𝐤^{^{}})sin((k^2k^{}_{}{}^{}2)t/2\mu )]$$
(159)
$$\mathrm{\Sigma }(𝐤,t)=\underset{𝐤^{^{}}𝐤}{}v_{𝐤𝐤^{^{}}}f(𝐤^{^{}})$$
(160)
## III Optical Conductivity
First define the total polarization,
$$P(t)=\underset{𝐤}{}d_𝐤c_𝐤$$
(161)
From this we may obtain the Fourier component,
$$\stackrel{~}{P}(\omega )=_{\mathrm{}}^{\mathrm{}}𝑑t\text{ }P(t)\text{ }e^{i\omega t}$$
(162)
Now since,
$$A_\tau (t)=A_\tau \text{ }e^{i\text{ }\omega _X\text{ }\tau }\delta (t\tau )$$
(163)
$$E(t)=\frac{A_\tau (t)}{t},\text{ }A_\tau e^{i\omega _X\text{ }\tau }\text{ }\delta ^{^{}}(t\tau )=_{\mathrm{}}^{\mathrm{}}\frac{d\omega }{2\pi }\text{ }e^{i\omega t}E(\omega )$$
(164)
$$E(\omega )=(i\omega )A_\tau \text{ }e^{i(\omega \omega _X)\tau }$$
(165)
Since,
$$j(\omega )=p_{vc}\stackrel{~}{P}(\omega )$$
(166)
We have
$$\sigma (\omega )=p_{vc}e^{i(\omega \omega _X)\tau }Lim_{A_\tau 0}\frac{(\stackrel{~}{P}(\omega ,A_\tau )\stackrel{~}{P}(\omega ,0))}{(i\omega )A_\tau }$$
(167)
Here,
$`\stackrel{~}{P}(\omega )={\displaystyle \underset{𝐤}{}}{\displaystyle _{\mathrm{}}^\tau }𝑑t\text{ }\sqrt{(1\overline{n}_e(𝐤))(1\overline{n}_h(𝐤))}[\phi _0(𝐤)\stackrel{~}{D}_0^{()}(t)e^{i(\omega ϵ_0)t}+\stackrel{~}{D}_𝐤^{()}(t)e^{i(\omega ϵ_𝐤)t}]`$
$`+{\displaystyle \underset{𝐤}{}}{\displaystyle _{\mathrm{}}^\tau }𝑑t\text{ }\sqrt{\overline{n}_e(𝐤)\overline{n}_h(𝐤)}e^{i(\omega ϵ_𝐤)t}\stackrel{~}{a}_𝐤^{}(0,,)`$
$`+{\displaystyle \underset{𝐤}{}}{\displaystyle _\tau ^{\mathrm{}}}𝑑t\text{ }\sqrt{(1\overline{n}_e(𝐤))(1\overline{n}_h(𝐤))}[\phi _0(𝐤)\stackrel{~}{D}_0^{(+)}(t)e^{i(\omega ϵ_0)t}+\stackrel{~}{D}_𝐤^{(+)}(t)e^{i(\omega ϵ_𝐤)t}]`$
$$+\underset{𝐤}{}_\tau ^{\mathrm{}}𝑑t\text{ }\sqrt{\overline{n}_e(𝐤)\overline{n}_h(𝐤)}e^{i(\omega ϵ_𝐤)t}\stackrel{~}{a}_𝐤^{}(0,,+)$$
(168)
$$i(\stackrel{~}{D}_0^{(+)}(\tau )\stackrel{~}{D}_0^{()}(\tau ))=(\frac{|e|}{\mu c})e^{i(ϵ_0\omega _X)\tau }A_\tau p_{vc}\underset{𝐤}{}\phi _0(𝐤)\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})$$
(169)
$$i(\stackrel{~}{D}_𝐤^{(+)}(\tau )\stackrel{~}{D}_𝐤^{()}(\tau ))=(\frac{|e|}{\mu c})e^{i(\frac{k^2}{2\mu }+E_g\omega _X)\tau }A_\tau p_{vc}\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})$$
(170)
$$i(\stackrel{~}{a}_𝐤(\mathrm{𝟎},+)\stackrel{~}{a}_𝐤(\mathrm{𝟎},))=(\frac{|e|}{\mu c})e^{i(\frac{k^2}{2\mu }+E_g\omega _X)\tau }A_\tau p_{vc}\mathrm{\Lambda }_2(𝐤,\mathrm{𝟎})$$
(171)
$`\stackrel{~}{P}_R(\omega )={\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{4\pi k^2}{(2\pi )^3}}𝑑k\text{ }{\displaystyle _{\mathrm{}}^\tau }𝑑t\text{ }\sqrt{(1\overline{n}_e(𝐤))(1\overline{n}_h(𝐤))}`$
$`[\phi _0(𝐤)\stackrel{~}{D}_0^{(),R}(t)cos((\omega ϵ_0)t)\phi _0(𝐤)\stackrel{~}{D}_0^{(),I}(t)sin((\omega ϵ_0)t)+\stackrel{~}{D}_𝐤^{(),R}(t)cos((\omega ϵ_𝐤)t)\stackrel{~}{D}_𝐤^{(),I}(t)sin((\omega ϵ_𝐤)t)]`$
$`+{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{4\pi k^2}{(2\pi )^3}}𝑑k\text{ }{\displaystyle _{\mathrm{}}^\tau }𝑑t\text{ }\sqrt{\overline{n}_e(𝐤)\overline{n}_h(𝐤)}cos((\omega ϵ_𝐤)t)\stackrel{~}{a}_𝐤^R(0,,)`$
$`+{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{4\pi k^2}{(2\pi )^3}}𝑑k\text{ }{\displaystyle _{\mathrm{}}^\tau }𝑑t\text{ }\sqrt{\overline{n}_e(𝐤)\overline{n}_h(𝐤)}sin((\omega ϵ_𝐤)t)\stackrel{~}{a}_𝐤^I(0,,)`$
$`+{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{4\pi k^2}{(2\pi )^3}}𝑑k\text{ }{\displaystyle _\tau ^{\mathrm{}}}𝑑t\text{ }\sqrt{(1\overline{n}_e(𝐤))(1\overline{n}_h(𝐤))}`$
$`[\phi _0(𝐤)\stackrel{~}{D}_0^{(+),R}(t)cos((\omega ϵ_0)t)\phi _0(𝐤)\stackrel{~}{D}_0^{(+),I}(t)sin((\omega ϵ_0)t)+\stackrel{~}{D}_𝐤^{(+),R}(t)cos((\omega ϵ_𝐤)t)\stackrel{~}{D}_𝐤^{(+),I}(t)sin((\omega ϵ_𝐤)t)]`$
$`+{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{4\pi k^2}{(2\pi )^3}}𝑑k\text{ }{\displaystyle _\tau ^{\mathrm{}}}𝑑t\text{ }\sqrt{\overline{n}_e(𝐤)\overline{n}_h(𝐤)}cos((\omega ϵ_𝐤)t)\stackrel{~}{a}_𝐤^R(0,,+)`$
$$+_0^{\mathrm{}}\frac{4\pi k^2}{(2\pi )^3}𝑑k\text{ }_\tau ^{\mathrm{}}𝑑t\text{ }\sqrt{\overline{n}_e(𝐤)\overline{n}_h(𝐤)}sin((\omega ϵ_𝐤)t)\stackrel{~}{a}_𝐤^I(0,,+)$$
(172)
$`\stackrel{~}{P}_I(\omega )={\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{4\pi k^2}{(2\pi )^3}}𝑑k\text{ }{\displaystyle _{\mathrm{}}^\tau }𝑑t\text{ }\sqrt{(1\overline{n}_e(𝐤))(1\overline{n}_h(𝐤))}`$
$`[\phi _0(𝐤)\stackrel{~}{D}_0^{(),R}(t)sin((\omega ϵ_0)t)+\phi _0(𝐤)\stackrel{~}{D}_0^{(),I}(t)cos((\omega ϵ_0)t)+\stackrel{~}{D}_𝐤^{(),R}(t)sin((\omega ϵ_𝐤)t)+\stackrel{~}{D}_𝐤^{(),I}(t)cos((\omega ϵ_𝐤)t)]`$
$`+{\displaystyle _{\mathrm{}}^\tau }𝑑t\text{ }{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{4\pi k^2}{(2\pi )^3}}𝑑k\text{ }\sqrt{\overline{n}_e(𝐤)\overline{n}_h(𝐤)}sin((\omega ϵ_𝐤)t)\stackrel{~}{a}_𝐤^R(0,,)`$
$`{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{4\pi k^2}{(2\pi )^3}}𝑑k\text{ }{\displaystyle _{\mathrm{}}^\tau }𝑑t\text{ }\sqrt{\overline{n}_e(𝐤)\overline{n}_h(𝐤)}cos((\omega ϵ_𝐤)t)\stackrel{~}{a}_𝐤^I(0,,)`$
$`+{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{4\pi k^2}{(2\pi )^3}}𝑑k\text{ }{\displaystyle _\tau ^{\mathrm{}}}𝑑t\text{ }\sqrt{(1\overline{n}_e(𝐤))(1\overline{n}_h(𝐤))}`$
$`[\phi _0(𝐤)\stackrel{~}{D}_0^{(+),R}(t)sin((\omega ϵ_0)t)+\phi _0(𝐤)\stackrel{~}{D}_0^{(+),I}(t)cos((\omega ϵ_0)t)+\stackrel{~}{D}_𝐤^{(+),R}(t)sin((\omega ϵ_𝐤)t)+\stackrel{~}{D}_𝐤^{(+),I}(t)cos((\omega ϵ_𝐤)t)]`$
$`+{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{4\pi k^2}{(2\pi )^3}}𝑑k\text{ }{\displaystyle _\tau ^{\mathrm{}}}𝑑t\text{ }\sqrt{\overline{n}_e(𝐤)\overline{n}_h(𝐤)}sin((\omega ϵ_𝐤)t)\stackrel{~}{a}_𝐤^R(0,,+)`$
$$_0^{\mathrm{}}\frac{4\pi k^2}{(2\pi )^3}𝑑k\text{ }_\tau ^{\mathrm{}}𝑑t\text{ }\sqrt{\overline{n}_e(𝐤)\overline{n}_h(𝐤)}cos((\omega ϵ_𝐤)t)\stackrel{~}{a}_𝐤^I(0,,+)$$
(173)
$$\stackrel{~}{D}_0^{(+),R}(\tau )\stackrel{~}{D}_0^{(),R}(\tau )=(\frac{|e|}{\mu c})A_\tau p_{vc}sin((ϵ_0\omega _X)\tau )_0^{\mathrm{}}\frac{4\pi k^2}{(2\pi )^3}𝑑k\text{ }\phi _0(𝐤)\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})$$
(174)
$$\stackrel{~}{D}_0^{(+),I}(\tau )\stackrel{~}{D}_0^{(),I}(\tau )=(\frac{|e|}{\mu c})A_\tau p_{vc}cos((ϵ_0\omega _X)\tau )_0^{\mathrm{}}\frac{4\pi k^2}{(2\pi )^3}𝑑k\text{ }\phi _0(𝐤)\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})$$
(175)
$$\stackrel{~}{D}_𝐤^{(+),R}(\tau )\stackrel{~}{D}_𝐤^{(),R}(\tau )=(\frac{|e|}{\mu c})A_\tau p_{vc}sin((ϵ_𝐤\omega _X)\tau )\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})$$
(176)
$$\stackrel{~}{D}_𝐤^{(+),I}(\tau )\stackrel{~}{D}_𝐤^{(),I}(\tau )=(\frac{|e|}{\mu c})A_\tau p_{vc}cos((ϵ_𝐤\omega _X)\tau )\mathrm{\Lambda }_1(𝐤,\mathrm{𝟎})$$
(177)
$$\stackrel{~}{a}_𝐤^R(\mathrm{𝟎},,+)\stackrel{~}{a}_𝐤^R(\mathrm{𝟎},,)=(\frac{|e|}{\mu c})A_\tau p_{vc}sin((ϵ_𝐤\omega _X)\tau )\mathrm{\Lambda }_2(𝐤,\mathrm{𝟎})$$
(178)
$$\stackrel{~}{a}_𝐤^I(\mathrm{𝟎},,+)\stackrel{~}{a}_𝐤^I(\mathrm{𝟎},,)=(\frac{|e|}{\mu c})A_\tau p_{vc}cos((ϵ_𝐤\omega _X)\tau )\mathrm{\Lambda }_2(𝐤,\mathrm{𝟎})$$
(179)
$`\stackrel{~}{P}(\omega )={\displaystyle \underset{𝐤}{}}{\displaystyle _{\mathrm{}}^\tau }𝑑t\text{ }\sqrt{(1\overline{n}_e(𝐤))(1\overline{n}_h(𝐤))}[\phi _0(𝐤)\stackrel{~}{D}_0^{()}(t)e^{i(\omega ϵ_0)t}+\stackrel{~}{D}_𝐤^{()}(t)e^{i(\omega ϵ_𝐤)t}]`$
$`+{\displaystyle \underset{𝐤}{}}{\displaystyle _{\mathrm{}}^\tau }𝑑t\text{ }\sqrt{\overline{n}_e(𝐤)\overline{n}_h(𝐤)}e^{i(\omega ϵ_𝐤)t}\stackrel{~}{a}_𝐤^{}(0,,)`$
$`+{\displaystyle \underset{𝐤}{}}{\displaystyle _\tau ^{\mathrm{}}}𝑑t\text{ }\sqrt{(1\overline{n}_e(𝐤))(1\overline{n}_h(𝐤))}[\phi _0(𝐤)\stackrel{~}{D}_0^{(+)}(t)e^{i(\omega ϵ_0)t}+\stackrel{~}{D}_𝐤^{(+)}(t)e^{i(\omega ϵ_𝐤)t}]`$
$$+\underset{𝐤}{}_\tau ^{\mathrm{}}𝑑t\text{ }\sqrt{\overline{n}_e(𝐤)\overline{n}_h(𝐤)}e^{i(\omega ϵ_𝐤)t}\stackrel{~}{a}_𝐤^{}(0,,+)$$
(180)
The Optical Conductivity
$$Re(\sigma (\omega ))=|e|\text{ }Lt_{A_\tau 0}\frac{cos((\omega \omega _X)\tau )\stackrel{~}{P}_I(\omega ,A_\tau )sin((\omega \omega _X)\tau )\stackrel{~}{P}_R(\omega ,A_\tau )cos((\omega \omega _X)\tau )\stackrel{~}{P}_I(\omega ,0)+sin((\omega \omega _X)\tau )\stackrel{~}{P}_R(\omega ,0)}{\omega A_\tau }$$
(181)
A quick check of dimensions. In units of $`\mathrm{}=c=1`$, all masses are of dimension inverse length we take to be centeimeter. All times are in centimeters. Charge is dimensionless. Since $`\frac{|e|}{\mu c}A_\tau p_{vc}\text{ }\tau `$ is dimensionless, it follows that $`[A_\tau ]=L^1`$. It is easy to check that $`[Re(\sigma (\omega ))]=L^1`$ as it should be.
## IV Results and Discussion
The equations written down in the previous sections have to be solved numerically. It is worthwhile to point out some pitfalls and problems. Let us first focus on the SBEs. As we pointed out in our earlier work involving the SBE’s the sums over $`𝐤^{^{}}`$ have to be carried out in a special manner so as to avoid potential divergences.
$`{\displaystyle \underset{𝐤𝐤^{^{}}}{}}{\displaystyle \frac{v_{𝐤𝐤^{^{}}}}{V}}f(k^{^{}})={\displaystyle \frac{(4\pi e^2)}{(2\pi )^2}}{\displaystyle _0^{k_{max}}}𝑑k^{^{}}\text{ }({\displaystyle \frac{k^{^{}}}{k}})ln(\left|{\displaystyle \frac{(k^{^{}}+k)}{(k^{^{}}k)}}\right|)(f(k^{^{}})f(k))`$
$$+\frac{(4\pi e^2)}{(2\pi )^2}f(k)(\frac{1}{k})[\frac{1}{2}(k_{max}^2k^2)ln(\left|\frac{(k_{max}+k)}{(k_{max}k)}\right|)+k\text{ }k_{max}]$$
(182)
Further, it was suggested by Binder et.al. that we should use a momentum cutoff $`k_{max}=12/a_{Bohr}`$, where $`a_{Bohr}`$ is the exciton Bohr radius. The justification for this stems from the fact that beyond this cutoff the probability of the electron existing is negligible. This assertion is true only if the pump field frequency is below or equal the band-gap. When the frequency is well above the band gap the situation is less clear and care must be taken inorder not to lose features that may be present at high momenta. Unfortunately we have found that even when the pump field has a frequency equal to the band gap the prescription of Binder et.al has some problems. In particular, we have found that if we try and sneak a peek at the form of the distribution for $`k>>12/a_{Bohr}`$, we find a periodic pattern suggesting therefore that electrons can exist at (arbitrarily) high momenta long after a pump field whose frequency is at the band gap is switched off (after a time long enough so that we may still meaningfully talk of a well-defined frequency). This is a paradoxical and counterintutive result that has been gloosed over by the pioneers . The arbitrary cutoff of Binder et.al. shold not be taken too seriously. In order to make more sense out of all this we have to claim that the SBE’s produce the correct momentum distributions only for small enough $`k`$, and we have to use some judgement as to where we should cutoff the distributions.
The sea-boson analogs of the SBE’s written down above have their own numerical problems. First is the fact that even in the two-component case the sea-boson technique works well only when $`q<<k`$. Since we have chosen to study only $`𝐪=0`$, parts of the hamiltonian, it seems that we are in good shape. However we find that even then there is a cutoff small $`k=k_{min}`$ cutoff below which the momentum distributions become unphysical (larger than unity). This is true if we use the formula in Eq.( 130) $`\overline{n}_e(𝐤)=\overline{n}_h(𝐤)=\overline{n}_0(𝐤)`$. Further we find that this identification is the analog of the SBE. It is comforting to know that the sea-boson technique is equivalent to the SBE in some limit. In the SBEs, the momentum distributions of the electron and holes are identical even if the effective masses are very different. This is due to the fact that the SBEs neglect the collision terms responsible for the asymmetry that we would otherwise expect. Similarly, the sea-boson equations at the level of Eq.( 130) neglect the repulsion and phonon terms. However the SBEs do include repulsion at the Hartree-Fock level, therefore the analogy between the two is not exact. In Fig.1 we see how far we may take this analogy between the SBEs and the sea-boson equations. The approach toward unphysical behaviour for the small $`k`$ limit of the momentum distribution obtained using the sea-boson equations is also seen. When we include the effects of repulsion and phonons, the answers change quite dramatically. In fact they are so very different from the SBE results that we have decided not to publish them. It will take some more time before a thorough analysis is completed and all the ramifications are explored. For now we shall assume that the momentum distribution is that given by $`\overline{n}_0(𝐤)`$ or that given by the SBEs. Let us first write down some formulas that relate the real part of the conductivity to the absorption coefficient. We may expect the two to have qualitatively similar features. However just to be sure and so that we don’t make any mistakes having gotten this far, let us write down the formulas. They are a combination of the formulas from the text by Haug and Koch and the one by Manah. The transverse dielectric function may be decomposed as follows.
$$ϵ(\omega )=ϵ_1(\omega )+i\text{ }ϵ_2(\omega )$$
(183)
The abosrption coefficient, refractive index and the real part of the conductivity are given by(in units $`\mathrm{}=c=1`$),
$$\alpha (\omega )=\frac{\omega }{n(\omega )}ϵ_2(\omega )$$
(184)
$$n(\omega )=\{\frac{1}{2}(ϵ_1(\omega )+\sqrt{ϵ_1^2(\omega )+ϵ_2^2(\omega )})\}^{\frac{1}{2}}$$
(185)
The real part of the conductivity is,
$$Re(\sigma (\omega ))=\frac{\omega }{4\pi }ϵ_2(\omega )$$
(186)
It is better not to use the Kramers-Kronig relations as our answer for the real part of the conductivity is undetermined upto a factor(actually it has no reason to, it just so happens that the magnitude does not agree with observations). We may write down a formula for the imaginary part of the conductivity just as we did the real part.
$$Im(\sigma (\omega ))=|e|Lt_{A_\tau 0}\frac{sin((\omega \omega _X)\tau )\stackrel{~}{P}_I(\omega ,A_\tau )cos((\omega \omega _X)\tau )\stackrel{~}{P}_R(\omega ,A_\tau )+sin((\omega \omega _X)\tau )\stackrel{~}{P}_I(\omega ,0)+cos((\omega \omega _X)\tau )\stackrel{~}{P}_R(\omega ,0)}{\omega A_\tau }$$
(187)
$$Im(\sigma (\omega ))=\frac{\omega }{4\pi }(ϵ_1(\omega )1)$$
(188)
Therefore we may deduce the optical dielectric function and hence the absorption coefficient. All this would not be neceesary if the refractive index was close to unity and then $`ϵ_2`$ is negligible in comparison with $`ϵ_1`$ for all frequencies. Then the real part of the conductivity would be proportional to the absorption coefficient. Let us compare the experimental magnitude of the absorption coefficient and the energy $`\omega `$. We find according to the experiments of Song’s group, $`|\alpha |10^5cm^1`$ whereas $`\omega 2.0\pi /(352\times 10^7cm)=1.78\times 10^5cm^1`$. we can see that these two quantites are comparable to each other suggesting thereby that the real part of the dielectric function is not close to unity.
|
warning/0004/astro-ph0004117.html
|
ar5iv
|
text
|
# Population and Size Distribution of Small Jovian Trojan Asteroids
## 1 Introduction
The Jovian Trojans are asteroidal objects confined to two swarms in Jupiter’s orbit, leading and trailing the planet by $`60^{}`$ of longitude (known as the L4 and L5 Trojans, respectively). The first recognized Jovian Trojan (588 Achilles), discovered in 1906 by Max Wolf, was taken as providing observational confirmation of Lagrange’s prediction of stable orbits at the triangular points. Currently, 132 Jovian Trojans have been numbered while another 125 await permanent designations. These objects follow loose orbits that librate around the L4 and L5 points with periods near 150 years. Recent work has shown that Trojan orbits are destabilized by collisional ejection (for which the loss rate of bodies larger than 1 km in diameter is estimated at $``$$`10^3`$ yr<sup>-1</sup>; Marzari et al. 1997) and, to a lesser extent, by dynamical chaos (corresponding loss rate $``$$`6\times 10^5`$ yr<sup>-1</sup>; Levison, Shoemaker and Shoemaker 1997). The implication is that the Trojans must either be the remnants of a much more substantial initial population of trapped bodies or that these objects are continually replenished from an unidentified external source.
The origin of the Trojans is a subject of much conjecture. The principal dynamical problem concerns the nature of the dissipation needed to stabilize objects in weakly bound orbits librating about L4 and L5. Schemes under consideration include capture of near-Jupiter planetesimals by gas drag in an early phase of the solar nebula (Peale 1993), stabilization of planetesimals near the L4 and L5 points due to the rapidly increasing mass of Jupiter in the late stages of its growth (Marzari and Scholl 1998), and collisional dissipation followed by capture of asteroidal fragments (Shoemaker et al. 1989). Physical observations provide only limited clues about the source of the Trojans. The optical (Jewitt and Luu 1990; Fitzsimmons et al. 1994) and near-infrared (Luu, Jewitt and Cloutis 1994; Dumas, Owen and Barruci 1998) reflection spectra appear featureless, and are reminiscent of the spectra of the nuclei of short-period comets. Like cometary nuclei, the Trojans have very low ($``$4%) visual albedos (Cruikshank 1977; Tedesco 1989) that suggest carbonized surface compositions. If the Trojans formed near or beyond Jupiter’s orbit, temperatures were probably low enough for water to exist as solid ice (rather than vapor, as in the inner nebula). This fact has led to the suggestion that the Trojans might possess ice-rich interiors equivalent to those of the cometary nuclei, a possibility which is not contradicted by any available observations (Jewitt 1996).
In this paper, we discuss the results of an optical survey taken in the direction of the L4 Jovian swarm. The survey differs from most previous work on these objects in two main respects. First, it is based on the use of a digital (CCD) detector instead of photographic plates and so has relatively high sensitivity to faint (small) Jovian Trojans. Second, the parameters of the survey are extremely well known as a consequence of the relative ease with which digital data may be calibrated (compared to non-linear, analog photographic data). Therefore, we are able to measure the statistical properties of the L4 Trojans with greater confidence than would be possible with photographic data. A preliminary abstract describing this work (Chen et al. 1997) is superceded by the present report.
## 2 Observations and Data Reduction
The present observations were taken as part of a study of the Kuiper Belt, the main results of which are already published (Jewitt, Luu and Trujillo 1998). Here we present observations taken UT 1996 Oct. 7 to 15 at the $`f/10`$ Cassegrain focus of the University of Hawaii 2.2-meter telescope with a 8192$`\times `$8192 pixel CCD mosaic (hereafter called 8k). The 8k consists of eight 2048$`\times `$4096 pixel Loral chips with 15 $`\mu `$m pixels and gaps between chips of $``$1 mm. The pixels were binned 3$`\times `$3 in order to reduce the readout time (from approximately 6 minutes to 1 minute) while maintaining Nyquist sampling of the images. The binned image scale was $`0.405\pm 0.002^{\prime \prime }`$/pixel yielding a field of view 18.4’$`\times `$18.4’ (0.094 deg<sup>2</sup>). Typical image quality (including contributions from the intrinsic seeing, wind shake and tracking oscillations during the unguided integrations) varied from 0.8” to 1.0” Full Width at Half Maximum (FWHM), meaning that the images were Nyquist sampled. The images were taken with a 150-sec integration time through a specially optimized VR filter (bandwidth 5000 Å to 7000 Å; Jewitt, Luu and Trujillo 1998). Each sky position was imaged at three epochs, with a separation between epochs of about 1 hour. In total, we observed 20 deg<sup>2</sup> of sky in the direction of L4.
The data were flattened using a median combination of dithered images of the evening twilight sky. Observations of photometric standard stars (Landolt 1992) were used to calibrate the sensitivity of each chip. By defining an A0 star to have $`m_{VR}=V=R=0`$, an object of solar color ($`VR=0.35`$) has $`Vm_{VR}+0.2`$. We adopted the latter relation to transform our VR magnitude to standard $`V`$ magnitude in this work. Note that Trojan asteroids display a wide range of optical colors, from nearly solar to very red ($`VR0.6`$: Jewitt and Luu 1990, Fitzsimmons et al. 1994), leading to the introduction of small, color-dependent corrections to the $`V`$ vs. $`m_{VR}`$ relation. In addition, some of the 8k CCDs were of locally inferior photometric quality. Together, these effects introduce an inherent uncertainty in the absolute photometric accuracy of about 0.2 mag. Trojans were identified using the MODS detection program (Trujillo and Jewitt 1998). We determined the detection efficiency of MODS by searching for artificial objects added to real data. The efficiency is adequately fitted by the function
$$e=e_{\mathrm{max}}\left[1\frac{1}{2}\mathrm{exp}\left(\frac{m_Vm_V(50)}{\sigma _V}\right)\right]$$
(1)
where $`e_{\mathrm{max}}`$ is the maximum detection efficiency, $`m_V`$ is the Trojan magnitude, $`m_V(50)`$ is the magnitude at which the detection efficiency equals $`e_{\mathrm{max}}/2`$, and $`\sigma _V`$ measures the width of the band of decreasing detection efficiency. When including the trailing loss due to the motion of the Trojans during the integration, we found $`e_{\mathrm{max}}=0.86\pm 0.01`$, $`m_V(50)=22.47\pm 0.05`$ and $`\sigma _V=1.13\pm 0.01`$. Variations in the seeing within the data are small and affect $`m_V(50)`$ by at most $`0.1`$ mag. (i.e., less than the formal uncertainty in the absolute photometry).
We took all observations near opposition, where the rate of retrograde motion across the sky, $`\omega `$ \[<sup>′′</sup>/hr\], is inversely related to the heliocentric distance. With this constraint we could not observe the L4 point directly, but instead mapped areas at a range of angular distances from the Lagrangian point (Figure 1). In addition, because we could not afford to interfere with our primary (Kuiper Belt) observational program, we secured no follow-up astrometry of Trojan candidates with which to determine orbital elements. Instead, we used the sky-plane angular speed to distinguish Trojans from main-belt asteroids. The main-belt asteroids move westward at rates generally higher than the more distant Trojans, permitting us to separate the two types of object on the basis of speed. Figure 2a shows the apparent velocities in RA and Dec of numbered main-belt asteroids and previously known Trojans in the direction of our observations on 1996 October 10. The curves in Figure 2a show the loci of points having total angular motions $`\omega =22^{\prime \prime }`$/hr and $`\omega =24^{\prime \prime }`$/hr. At $`\omega >24^{\prime \prime }`$/hr we find exclusively main-belt asteroids (Fig. 2a). Roughly equal numbers of main-belt and Trojan asteroids appear in the range $`22\omega 24^{\prime \prime }`$/hr. On the other hand, a large majority ($``$90%) of the objects with $`\omega 22^{\prime \prime }`$/hr are Trojans. Therefore, $`\omega 22^{\prime \prime }`$/hr constitutes our operational definition of Trojans in this survey. This definition is clearly not perfect, but it is sufficiently robust that we can make statistical identifications of the Jovian Trojans in our data. Figure 2b shows all 93 Trojan candidates flagged by the detection software. Of these, only the 4th object in Table 1 has position and motion consistent with a previously known Trojan asteroid (6020 P-L). The best evidence that we have indeed obtained a sample dominated by Trojans, as opposed to foreground main-belt asteroids, is provided by the sky-plane surface density distribution of the 93 identified objects (Figure 3). This distribution is peaked towards L4 in a manner incompatible with the azimuthally uniform distribution of the main-belt asteroids.
## 3 Discussion
(a) Luminosity Function
Photometry was performed using a circular aperture $`4.7^{\prime \prime }`$ in projected diameter, with sky subtraction from a contiguous annulus $`3^{\prime \prime }`$ wide (Table 1). This aperture was selected by trial and error to give a stable measure of the flux while minimizing photometric noise from the background sky. The statistical photometric uncertainty is $`\pm `$0.2 mag at the faint end and less than $`\pm `$0.1 mag at the bright end of the magnitude distribution. For comparison with other work, it is useful to employ the absolute $`V`$ magnitude, $`V(1,1,0)=V5\mathrm{log}_{10}(R\mathrm{\Delta })`$, where $`R`$ \[AU\] and $`\mathrm{\Delta }`$ \[AU\] are the heliocentric and geocentric distances, respectively. The correction to zero phase angle has been ignored (since the phase angles near opposition are small). The distances $`R`$ and $`\mathrm{\Delta }`$ are not accurately known for the individual Trojan asteroids. We adopt the mean distances of the numbered L4 Trojans at the epoch of observation, namely $`R=5.1\pm 0.2`$ AU and $`\mathrm{\Delta }=4.1\pm 0.2`$ AU. Here, the quoted errors are $`1\sigma `$ standard deviations and the dispersion in $`R`$ and $`\mathrm{\Delta }`$ results from the finite extent of the L4 swarm along the line of sight. With these values we obtain $`V(1,1,0)=V(6.60\pm 0.24)`$. For reference, we further compute the Trojan radii, $`r`$ \[km\], from the relation
$$r=\sqrt{\frac{2.24\times 10^{16+0.4(V_{\mathrm{Sun}}V(1,1,0))}}{p_V}}$$
(2)
where $`p_V`$ is the geometric albedo at the $`V`$ wavelength and $`V_{\mathrm{Sun}}=26.74`$ is the apparent $`V`$ magnitude of the sun. The mean geometric albedo of the Trojans recorded by the IRAS satellite (Tedesco 1989) is $`p_V=0.040\pm 0.005`$ (however, the quoted statistical uncertainty is probably smaller than inherent systematic uncertainties due to assumptions made in the radiometric modelling of the IRAS data). With this $`p_V`$, we obtain
$$r_{0.04}[\mathrm{km}]=10^{0.2(24.23V)}$$
(3)
(Figure 4). A Trojan at the 50% detection threshhold $`V=22.5`$ has $`V(1,1,0)=15.9`$ and $`r_{0.04}`$ = 2.2 km. The brightest (faintest) Trojan detected in the present survey has $`V=17.7(23.4)`$ corresponding approximately to $`r_{0.04}=20.2`$ km (1.5 km). The distance variation across the diameter of the L4 swarm introduces an uncertainty to the derived radii of about $`\pm 10`$%.
Figure 5 shows the cumulative luminosity function (CLF) computed from the present data. The “$`\times `$” marks in the figure show the distribution of the raw counts while the filled circles show the distribution corrected for the detection efficiency. We have not included a correction for contamination of the Trojan sample by main-belt interlopers. As noted above, this is a small effect whose inclusion would decrease the estimated surface densities at all magnitudes by about 10%. Error bars in Figure 5 were estimated from Poisson statistics. The luminosity function is taken to be of the form
$$N(V)\mathrm{d}V=10^{\alpha (VV_0)}\mathrm{d}V$$
(4)
A least-squares fit to the differential magnitude distribution gives slope parameter $`\alpha =0.40\pm 0.05`$. The three lines in Figure 5 have gradients $`\alpha 1\sigma ,\alpha `$, and $`\alpha +1\sigma `$, where $`1\sigma =0.05`$. We assume that the radii of Trojans follow a differential power law distribution such that the number of objects having radii in the range $`rr+\mathrm{d}r`$ is $`n(r)\mathrm{d}r=\mathrm{\Gamma }r^q\mathrm{d}r`$, where $`\mathrm{\Gamma }`$ and $`q`$ are constants. For objects all located at a single heliocentric and geocentric distance, $`\alpha `$ and $`q`$ are related by $`q=5\alpha +1`$ (Irwin et al. 1995). Thus, the present data suggest $`q=3.0\pm 0.3`$ in the radius range $`2.2r_{0.04}20`$ km. For comparison, Shoemaker et al. 1989 estimated $`\alpha =0.433`$ for $`10.25B(1,0)14`$ mag (corresponding to $`q=3.17`$ in the approximate range $`4r_{0.04}40`$ km, assuming $`BV0.65`$), but did not state their uncertainty. We consider these determinations to be in good agreement.
The difference between $`q`$ measured here and the canonical $`q=3.5`$ distribution produced by collisional shattering (Dohnanyi 1969) is statistically insignificant. In any case, unmodeled effects will cause the Trojan distribution to differ from a Dohnanyi power law. For example, the velocity of ejection of collision fragments varies inversely with fragment size (as $`r^{1/2}`$, Nakamura and Fujiwara 1991). Following collisional production the small Trojans should preferentially escape from the L4 region, leading to a distribution flatter than the Dohnanyi power law (i.e., $`q<3.5`$). We consider it likely that the small Trojan asteroids are collisionally produced fragments of once larger bodies (c.f. Marzari et al. 1997).
(b) Inclination-Frequency Distribution
We are able to measure the position angles of the proper-motion vectors of the Trojans with an uncertainty of about $`\pm 1^{}`$. The proper-motion vectors cannot be accurately converted to orbital inclinations without a fuller knowledge of the orbits than we possess. The approximate relation between the direction of apparent motion and the true inclination is given by
$$\mathrm{tan}\theta _a=\frac{\mathrm{tan}(i)}{\sqrt{R[1+\mathrm{tan}^2(i)]}1}$$
(5)
where $`\theta _a`$ is the angle between the apparent direction of motion and the projected ecliptic, $`i`$ is the true orbital inclination, and $`R`$ is the semi-major axis of the orbit. In deriving Eq. (5) we have assumed that the orbital eccentricities are zero, that the proper motion is the vector difference of the intrinsic motion of the Trojans from the orbital motion of earth and that the observations are taken at opposition. In the limit $`i90^{}`$, $`\mathrm{tan}\theta _aR^{1/2}`$ so that with $`R=5.2`$ AU we find $`\theta _a24^{}`$. One object in Table 1 has $`\theta _a>24^{}`$ (#26). Presumably, it is a main belt asteroid whose sky-plane velocity falls fortuitously within the Trojan domain.
The distribution of intrinsic inclinations (Figure 6a) has a mean $`i=7.4^{}\pm 0.7^{}`$, median $`i=6.2^{}`$. Trojans with large inclinations spend most of their time away from the ecliptic, leading to a bias against their detection. The bias correction varies approximately in inverse proportion to the orbital inclination. Figure 6b shows the inclination distribution after correction by the factor $`1/i`$ and normalization to Figure 6a in the range $`12i14^{}`$. The bias-corrected mean inclination is $`\overline{i}=13.7^{}\pm 0.5^{}`$ which compares favorably with Shoemaker et al.’s 1989 best estimate for the mean inclination of the larger Trojans ($`\overline{i}=17.7^{}`$).
Observers of Trojan asteroids two decades ago suspected that “there is a possibility that the inclinations are bimodal… with groups separated by a minimum at $`i13^{}`$” (Degewij and van Houten 1979). There is indeed an apparent lack of Trojans in the corrected inclination distribution with $`i=14^{}\pm 1^{}`$, giving a bimodal appearance (Fig. 6b). However, inspection of the raw data in Figure 6a shows that the local minimum is statistically insignificant. Furthermore, the inclination distribution of numbered and un-numbered L4 Trojans (from an electronic list maintained by Brian Marsden at the Minor Planet Center) shows no evidence for bimodality (Figure 7). Therefore, we conclude that the data provide no compelling evidence for a bimodal distribution of inclinations.
(c) Size and Content of the L4 Trojan Swarm
Figure 3 shows the variation of the surface density, $`\mathrm{\Sigma }(\theta )`$ \[deg<sup>-2</sup>\], of L4 Trojans with angular distance, $`\theta `$, from the L4 point along the ecliptic. The data can be fitted by a Gaussian function
$$\mathrm{\Sigma }(\theta )d\theta =\mathrm{\Sigma }_0+\mathrm{\Sigma }_1\mathrm{exp}\left[\frac{\theta ^2}{2\sigma _T^2}\right]d\theta ,(15^{}\theta 60^{})$$
(6)
with $`\mathrm{\Sigma }_0=1.1\pm 0.4`$ \[deg<sup>-2</sup>\], $`\mathrm{\Sigma }_1=27.7\pm 4.6`$ \[deg<sup>-2</sup>\] and $`\sigma _T=11.2^{}\pm 0.9^{}`$ \[deg<sup>-2</sup>\]. The apparent FWHM of the L4 swarm measured along the ecliptic is $`26.4^{}\pm 2.1^{}`$, with measurable surface density for $`\theta 40^{}`$. The linear size corresponding to FWHM = $`26.4^{}`$ is $``$2.4 AU. For comparison, Holman and Wisdom (1992) found that the theoretical stable zones of the Jupiter Lagrange point have an angular radius of about $`35^{}`$. The angular half width of the swarm along the ecliptic ($`13.2^{}\pm 1.0^{}`$) is comparable to the bias-corrected mean inclination ($`13.7^{}\pm 0.5^{}`$), meaning that we may take the projected outline of the swarm as approximately circular in the plane of the sky.
The number of L4 Trojans within angle $`\theta _{\mathrm{max}}`$ of L4 is then
$$N(\theta _{\mathrm{max}})=_0^{}^{\theta _{\mathrm{max}}}2\pi \mathrm{sin}(\theta )\mathrm{\Sigma }(\theta )d\theta .$$
(7)
We plot solutions to Eq. (7) in Figure 8. The amplitude of libration about L4 ranges up to $`60^{}`$ (Shoemaker et al. 1989). Accordingly, we obtain the nominal population estimate $`N(60^{})=3.4\times 10^4`$ ($`V22.5,r_{0.04}2.2`$ km), plotted in Figure 8 as Model A. The uncertainty on this number may be estimated in two ways. Uniformly increasing (decreasing) the fitted parameters $`\mathrm{\Sigma }_0`$, $`\mathrm{\Sigma }_1`$ and $`\sigma _T`$ by $`1\sigma `$ changes $`N`$ by $`\pm 35`$% (c.f. Figure 8). Systematic errors may also affect $`N(\theta _{\mathrm{max}})`$, particularly since the surface density at $`\theta 15^{}`$ is not measured in our survey. However, we find these errors to be small. If, to consider an extreme case, we arbitrarily (and unphysically) assume that $`\mathrm{\Sigma }(\theta 15^{})=0`$, we obtain (from Eq. 7) $`N(60^{})=2.1\times 10^4`$, still only 40% less than the nominal estimate. We conclude that $`N(\theta _{\mathrm{max}})`$ is uncertain to within a factor of order 2.
The cumulative luminosity function is replotted in Figure 9, including 132 numbered and 125 unnumbered Trojan asteroids (hollow circles) from orbital element catalogs maintained by the Minor Planet Center. We used $`V(1,1,0)=H+0.36`$ to correct the catalog magnitudes to the $`V`$ band magnitudes employed here. In making the comparison between the number of Trojans deduced from the present survey (Eq. 8) with those from the Minor Planet lists, we have corrected the former by a factor of two, to account for the unobserved L5 swarm. (Early suspicions that L4 might be more populated than L5 have not been borne out by recent data, supporting our application of a factor of two, Shoemaker et al. 1989). Curvature of the CLF at $`V(1,1,0)9.5(r_{0.04}42`$ km) indicates observational incompleteness in the Minor Planet Trojan sample, as does the fact that the catalog asteroids are less numerous than those of the present survey by 1 to 2 orders of magnitude in the common $`11V(1,1,0)14`$ mag range (Figure 9). Thirty seven objects have $`V(1,1,0)<9.5`$. Their effective CLF has slope $`\alpha =0.89\pm 0.15`$, significantly steeper than measured from the fainter objects of the 8k survey. The implied size distribution index is $`q=5.5\pm 0.9(V(1,1,0)<9.5,r_{0.04}42`$ km).
The data of Figure 9 are adequately fitted by the following differential size distributions:
$$n_1\left(r_{0.04}\right)\mathrm{d}r_{0.04}=1.5\times 10^6\left(\frac{1\mathrm{km}}{r_{0.04}}\right)^{3.0\pm 0.3}\mathrm{d}r_{0.04}(2.2r_{0.04}20\mathrm{km})$$
(8)
and
$$n_2\left(r_{0.04}\right)\mathrm{d}r_{0.04}=3.5\times 10^9\left(\frac{1\mathrm{km}}{r_{0.04}}\right)^{5.5\pm 0.9}\mathrm{d}r_{0.04}(r_{0.04}42\mathrm{km}).$$
(9)
The corresponding integral distributions are
$$N(>r_{0.04})=1.6\times 10^5\left(\frac{1\mathrm{km}}{r_{0.04}}\right)^{2.0\pm 0.3}(2.2r_{0.04}20\mathrm{km})$$
(10)
and
$$N(>r_{0.04})=7.8\times 10^8\left(\frac{1\mathrm{km}}{r_{0.04}}\right)^{4.5\pm 0.9}(r_{0.04}42\mathrm{km})$$
(11)
From Eq. (10) we find the number of L4 Trojans with $`r_{0.04}5`$ km is $`N6400`$, to within a factor of order 2. For comparison, there were 5700 numbered and 1100 unnumbered main-belt asteroids with $`r_{0.04}5`$ km as of 1999 July 29. These estimates validate the assertion by Shoemaker et al. (1989) to the effect that the populations of the main-belt and the Trojan swarms are of the same order. The number of L4 Trojans with $`r_{0.04}1`$ km is $`1.6\times 10^5`$, from Eq. (11).
The most straightforward explanation of the slope differences (Eq. 8 and 9 and Figure 9) is that the large objects represent a primordial population while Trojans smaller than a critical radius, $`r_c`$, are produced from the larger ones by collisional shattering (Shoemaker et al. 1989, Marzari et al. 1996). By equating Eqs. (10) and (11) we find $`r_c30`$ km (corresponding to $`V(1,1,0)=10.2\pm 0.5`$ (Figure 9)). Binzel and Sauter (1992) reported that Trojans with $`r_{0.04}>45`$ km have a larger mean lightcurve amplitude (a measure of elongated body shape) than their low albedo main-belt counterparts. They suggested that this might mark the primordial/fragment transition size, with larger bodies retaining the aspherical forms in which they were created (we note that this explanation is clearly non-unique). Inspection of their Figure 21 shows that the transition radius defined in this way is uncertain to within a factor of 2 and fully compatible with $`r_c30`$ km as found here. We conclude that two independently measured physical parameters (the size distribution and the lightcurve amplitude distribution) show evidence for a change near $`r_c30`$ km to 40 km.
The total mass of Trojans is
$$M_T=_0^{r_c}\frac{4}{3}\pi \rho n_1(r)dr+_{r_c}^{\mathrm{}}\frac{4}{3}\pi \rho n_2(r)dr$$
(12)
where $`n_1`$ and $`n_2`$ are from Eqs. (8) and (9) and $`\rho =2000`$ kg m<sup>-3</sup> is the assumed bulk density. We find $`M_T5\times 10^{20}`$ kg $`9\times 10^5M_{\mathrm{Earth}}`$, equivalent to a 400 km radius sphere having the same density. Dynamical calculations show that escaped Trojans would be quickly scattered into orbits indistinguishable from those of some short-period comets (Marzari et al. 1995). Therefore it is of interest to compare $`M_T`$ with the mass of short-period comets delivered to the inner solar system over the past 4.5 Gyr. The rate of supply of short period comets is $`f10^2`$ yr<sup>-1</sup> (Fernandez 1985). The size and mass distributions of the cometary nuclei have not been adequately measured. We assume that the nuclei follow a differential size distribution with index $`q=3`$ and minimum and maximum radii $`r_1=0.5`$ and $`r_2=30`$ km. The mass-weighted mean radius drawn from this distribution is $`\overline{r}(2r_2r_1^2)^{1/3}=2.5`$ km. A small number of well-measured cometary nuclei have radii $``$ few km (Jewitt 1996), consistent with this estimate. The delivered mass of short-period comets is then $`M_C4\pi \rho \overline{r}^3fT/3`$, where $`T=4.5\times 10^9`$ yr is the age of the solar system. We find $`M_C6\times 10^{21}`$ kg, corresponding to $`M_C10M_T`$. This mass, if taken at face value, makes it unlikely that the Trojan swarms could be the dominant source of the comets. It is entirely possible, however, that a fraction of the short-period comets could be escaped Trojans. A definitive estimate of this fraction will require better knowledge of the cometary parameters ($`r_1`$, $`r_2`$, $`q`$, $`f`$, and density) than we now possess, as well as detailed understanding of the physics of collisional ejection from the Trojan swarms.
The present results were extracted from data taken for an independent (Kuiper Belt) purpose. They serve to give an idea of the power of modern CCD arrays on a telescope of rather modest diameter. Much more could be learned from a survey specifically targeting the Trojan asteroids and including astrometric follow-up, so that orbital elements can be determined for individual objects. Future work should focus on a more complete digital survey of both L4 and L5 swarms in order to determine the total population and size distribution with greater confidence. Observations taken away from the ecliptic will provide a better measure of the high inclination objects. Observations to fainter limiting magnitudes will allow us to probe the sub-kilometer population. Carefully planned observations will produce stronger constraints on the collisional and dynamical states of the Jovian Trojans, leading ultimately to a deeper understanding of these enigmatic bodies.
## 4 Summary
1. The luminosity function of the Jovian L4 Trojans has slope of $`0.40\pm 0.05`$ in the magnitude range $`18.0V22.5`$, corresponding to objects with radii $`2r_{0.04}20`$ km (where $`r_{0.04}`$ is the radius derived assuming a geometric albedo of 0.04). The corresponding differential power law size distribution index is $`q=3.0\pm 0.3`$. This is consistent with the slope expected for a collisionally shattered population ($`q3.5`$; Dohnanyi 1969) within $`2\sigma `$ (95%) confidence and suggests that the small Trojans are collisional fragments of larger bodies.
2. The brighter (larger) Trojans follow a $`q=5.5\pm 0.9`$ differential power law distribution.
3. The apparent FWHM of the L4 swarm is $`26^{}\pm 2^{}`$, measured along the ecliptic.
4. The distribution of inclinations of the Trojans, when corrected for observational bias, has a mean $`13.7^{}\pm 0.5^{}`$.
5. About $`1.6\times 10^5`$ L4 Trojans are bigger than 1 km radius. Their combined mass is of order $`5\times 10^{20}`$ kg ($`9\times 10^5M_{\mathrm{Earth}}`$), assuming bulk density $`\rho =2000`$ kg m<sup>-3</sup>.
Acknowledgements
We thank John Dvorak for operation of the UH telescope. Gerry Luppino, Mark Metzger, Richard Wainscoat and Pui Hin Rhoads helped us with the camera and the computer set-up. We benefitted from digital lists of Trojan orbital elements maintained by Brian Marsden and David Tholen. Scott Sheppard and an anonymous referee provided helpful comments. This research was supported by grants from NASA to DCJ.
FIGURE CAPTIONS
Figure 1. The sky coverage of the present survey is marked by squares. The L4 Lagrangian point is marked by a large filled circle and the ecliptic is shown for reference.
Figure 2. (a) Apparent angular velocities of numbered main-belt asteroids (empty circles) and Trojans (filled circles) projected in the direction of observation on 1996 October 10. Marked curves show angular velocities $`\omega =22^{\prime \prime }`$/hr and 24<sup>′′</sup>/hr. (b) Objects detected by MODS having $`\omega 24^{\prime \prime }`$/hr: those with $`\omega 22^{\prime \prime }`$/hr are taken to be Trojans for the purpose of this study. All objects moving faster than $`\omega =24^{\prime \prime }`$/hr are main-belt asteroids and were ignored by the MODS software.
Figure 3. Surface density variation of the observed L4 swarm along the ecliptic. A Gaussian distribution with $`\sigma _T=11.2\pm 0.9`$ (corresponding to FWHM is $`26.4^{}\pm 2.1^{}`$) fits the data.
Figure 4. Magnitude vs. size relation for Trojan asteroids (Eq. 2).
Figure 5. Cumulative magnitude-frequency distribution of detected Trojans. Crosses mark the original data; filled circles indicate the distribution after correction for the detection efficiency. The straight lines have gradients $`\alpha `$ = 0.35, 0.40 and 0.45.
Figure 6. (a) Apparent distribution of inclinations of Trojans found in the present survey. (b) Same as (a) but corrected for inclination bias and normalized at $`12^{}i14^{}`$.
Figure 7. Inclination distribution of the L4 Trojans (numbered + unnumbered).
Figure 8. Cumulative number of Trojans with radii $`2.2`$ km (the smallest objects detected in the present survey) as a function of angular distance from the L4 point. Model A shows Eq. (7) with $`\mathrm{\Sigma }_0=1.1\pm 0.4`$, $`\mathrm{\Sigma }_1=27.7\pm 4.6`$ and $`\sigma _T=11.2\pm 0.9`$ in the range $`0^{}\theta 60^{}`$. Model B is the same as Model A except that $`\mathrm{\Sigma }_0=\mathrm{\Sigma }_10`$ for $`\theta 15^{}`$. Uncertainties on both curves show the effect of forcing $`\mathrm{\Sigma }_0`$, $`\mathrm{\Sigma }_1`$ and $`\sigma _T`$ to $`+1\sigma `$ and $`1\sigma `$ from the best-fit values.
Figure 9. Cumulative luminosity function from this work (filled circles) and from 257 cataloged Trojans detected in earlier surveys (empty circles). Counts from the present survey have been doubled to account for the unobserved L5 swarm.
|
warning/0004/astro-ph0004204.html
|
ar5iv
|
text
|
# THE EVOLUTION OF ROTATING STARS
## 1 INTRODUCTION
Stellar rotation is an example of an astronomical domain which has been studied for several centuries and where the developments are rather slow. A short historical review since the discovery of the solar rotation by Galileo Galilei is given by Tassoul (1978). Few of the early works apply to real stars, since in general gaseous configurations were not considered and no account was given to radiative energy transport. The equations of rotating stars in radiative equilibrium were first considered by Milne (1923), von Zeipel (1924) and Eddington (1925); see also Tassoul (1990) for a more recent history. From the early days of stellar evolution, the studies of rotation and evolution have been closely associated. Soon after the first models showing that Main–Sequence (MS) stars move further into the giant and supergiant region (Sandage & Schwarzschild 1952), rotation was used as a major test for the evolution. Oke & Greenstein (1954) and Sandage (1955) found that the observed rotational velocities were consistent with the proposed evolutionary sequence.
Stellar evolution, as other fields of Science, proceeds using as a guideline the principle of Occam’s razor, which says that the explanation relying on the smallest number of hypotheses is usually the one to be preferred. Thus, due to the many well known successes of the theory of stellar evolution, rotation was and is still generally considered as only a second order effect. However, over recent years a number of serious discrepancies between current models and observations have been noticed. They concern particularly the helium and nitrogen abundances in massive O– and B–type stars and in giants and supergiants, as well as the distribution of stars in the HR diagram at various metallicities. The observations show that the role of rotation has been largely overlooked. All the model outputs (tracks in the HR diagram, lifetimes, actual masses, surface abundances, nucleosynthetic yields, supernova precursors, etc…) are greatly influenced by rotation, thus it turns out that stellar evolution is basically a function of mass M, metallicity Z and angular velocity $`\mathrm{\Omega }`$.
There are a number of reviews concerning stellar rotation, for example Strittmatter (1969), Fricke & Kippenhahn (1972), Tassoul (1978, 1990), Zahn (1983, 1994) and Pinsonneault (1997). Here, we focus on rotation in Upper MS stars, where the effects are likely the largest ones. The consequences for blue, yellow and red supergiants, Wolf–Rayet (W–R) stars, as well as for red giants and Asymptotic Giant Branch (AGB) stars are also examined. The rotation of low mass stars, where spin–down due to magnetic coupling between the wind and the central body is important, has been treated in a recent review (Pinsonneault 1997); rotation and magnetic activity were also reviewed by Hartmann and Noyes (1987). The role of rotation in pre–Main Sequence evolution with accretion disks has been discussed by Bodenheimer (1995).
## 2 BASIC PHYSICAL EFFECTS OF ROTATION
### 2.1 HYDROSTATIC EFFECTS
In a rotating star, the centrifugal forces reduce the effective gravity according to the latitude and introduce deviations from sphericity. The four equations of stellar structure need to be modified. The idea of the original method devised by Kippenhahn & Thomas (1970) and applied in most subsequent works (Endal & Sofia 1976; Pinsonneault et al 1989, 1990, 1991; Fliegner & Langer 1995; Heger et al 2000) is to replace the usual spherical eulerian or lagrangian coordinates by new coordinates characterizing the equipotentials. This method applies when the effective gravity can be derived from a potential, i.e. when the problem is conservative, which occurs for solid body rotation or for constant rotation on cylinders centered on the axis of rotation. If so, the structural variables $`P,T,\rho ,\mathrm{}`$ are constant on an equipotential $`\mathrm{\Psi }=\mathrm{\Phi }+\frac{1}{2}\mathrm{\Omega }^2r^2\mathrm{sin}^2\vartheta `$, where $`\mathrm{\Phi }`$ is the gravitational potential, $`\vartheta `$ the colatitude and $`\mathrm{\Omega }`$ the angular velocity. Thus the problem can be kept one–dimensional which is a major advantage. However the internal rotation generally evolves towards rotation laws which are non conservative. In that case the above method is not physically consistent. Unfortunately it has been and is still used by most authors.
A particularly interesting case of differential rotation is that with $`\mathrm{\Omega }`$ constant on isobars (Zahn 1992). This case is called “shellular rotation” and it is often approximated by $`\mathrm{\Omega }=\mathrm{\Omega }(r)`$, which is valid at low rotation. It is supported by the study of turbulence in the Sun and stars (Spiegel & Zahn 1992; Zahn 1992). Such a law is due to the fact that the turbulence is very anisotropic, with a much stronger, geostrophic–like transport in the horizontal direction than in the vertical one, where stabilisation is favoured by the stable temperature gradient. The horizontal turbulence enforces an essentially constant rotation rate on isobars, thus producing the above rotation law. The star models are essentially one–dimensional, which enormously simplifies the computations. Shellular rotation is likely to occur in fast rotators as well as in slow ones.
The equations of stellar structure can be written consistently for a differentially rotating star, if the rotation law is shellular. Then, the isobaric surfaces satisfy the same equation $`\mathrm{\Psi }=const.`$ as the equipotentials of the conservative case (Meynet & Maeder 1997) and thus the equations of stellar structure can be written as a function of the coordinate of an isobar, either in the lagrangian or eulerian form. Let us emphasize that in general the hydrostatic effects of rotation have only very small effects of the order of a few percents on the internal evolution (Faulkner et al 1968; Kippenhahn & Thomas 1970). Recent two–dimensional models including the hydrostatic effects of rotation confirm the smallness of these effects (Shindo et al 1997).
The above potential $`\mathrm{\Psi }`$ describes the shape of the star in the conservative case for the so–called Roche model, where the distorsions of the gravitational potential are neglected. However, the stellar surface deviates from a surface given by $`\mathrm{\Psi }=const.`$ in case of non–conservative rotation law (Kippenhahn, 1977).
### 2.2 THE VON ZEIPEL THEOREM
The von Zeipel theorem (1924) is essential to predict the distribution of temperature at the surface of a rotating star. It applies to the conservative case and states that the local radiative flux $`\stackrel{}{F}`$ is proportional to the local effective gravity $`\stackrel{}{g_{\mathrm{eff}}}`$, which is the sum of the gravity and centrifugal force, if the star is not close to the Eddington limit,
$$\stackrel{}{F}=\frac{L(P)}{4\pi GM_{}(P)}\stackrel{}{g_{\mathrm{eff}}}$$
(1)
with $`M_{}(P)=M(1\frac{\mathrm{\Omega }^2}{2\pi G\overline{\rho }})`$; $`L(P)`$ is the luminosity on an isobar and $`\overline{\rho }`$ the mean internal density. Thus, the local effective temperature on the surface of a rotating star varies like $`T_{\mathrm{eff}}(\vartheta )g_{\mathrm{eff}}(\vartheta )^{\frac{1}{4}}`$. This shows that the spectrum of a rotating star is in fact a composite spectrum made of local atmospheres of different gravity and $`T_{\mathrm{eff}}`$. If it is meaningful to define an average, a reasonable choice is to take $`T_{\mathrm{eff}}^4=L/(\sigma S(\mathrm{\Omega }))`$, where $`\sigma `$ is Stefan’s constant and $`S(\mathrm{\Omega })`$ the total actual stellar surface. In the case of non–conservative rotation law, the corrections to the von Zeipel theorem depend on the opacity law and on the degree of differential rotation, but they are likely to be small, i.e. $`1\%`$ in current cases of shellular rotation (Kippenhahn 1977; Maeder 1999a). There are some discussions (Heger et al 2000; Langer et al 1999) whether von Zeipel must really be used or not. We would like to emphasize that this is just a mere and unescapable consequence of Newton’s law and basic thermodynamics (Sect. 5.3).
### 2.3 TRANSPORT OF ANGULAR MOMENTUM AND CHEMICAL ELEMENTS
Inside a rotating star the angular momentum is transported by convection, turbulent diffusion and meridional circulation. The equation of transport has been derived by Jeans (1928), Tassoul (1978), Chaboyer & Zahn (1992), Zahn (1992). For shellular rotation, the equation of transport of angular momentum in the vertical direction is (in lagrangian coordinates)
$`\rho {\displaystyle \frac{d}{dt}}\left(r^2\mathrm{\Omega }\right)_{M_r}={\displaystyle \frac{1}{5r^2}}{\displaystyle \frac{}{r}}\left(\rho r^4\mathrm{\Omega }U(r)\right)+{\displaystyle \frac{1}{r^2}}{\displaystyle \frac{}{r}}\left(\rho Dr^4{\displaystyle \frac{\mathrm{\Omega }}{r}}\right).`$ (2)
$`\mathrm{\Omega }(r)`$ is the mean angular velocity at level r, $`U(r)`$ the vertical component of the meridional circulation velocity and $`D`$ the diffusion coefficient due to the sum of the various turbulent diffusion processes (Sect. 2.5). The factor $`\frac{1}{5}`$ comes from the integration in latitude. If both $`U(r)`$ and $`D`$ are zero, we just have the local conservation of the angular momentum $`r^2\mathrm{\Omega }=const.`$, for a fluid element in case of contraction or expansion. The solution of eqn. 2 gives the *non–stationary solution* of the problem. The rotation law is not arbitrary chosen, but is allowed to evolve with time as a result of transport by meridional circulation, diffusion processes and contraction or expansion. In turn, the differential rotation built–up by these processes generates some turbulence and meridional circulation, which are themselves functions of the rotation law. This coupling provides a feedback, and the self–consistent solution for the evolution of $`\mathrm{\Omega }(r)`$ has to be found (Zahn 1992).
Some characteristic times can be associated, to both the processes of meridional circulation and diffusion,
$$t_{\mathrm{circ}}\frac{R}{U},t_{\mathrm{diff}}\frac{R^2}{D}.$$
(3)
These timescales are essential quantities, because the comparison with the nuclear timescales will show the relative importance of the transport processes in the considered nuclear phases. Equation 2 also admits a *stationary solution*, when one of the above characteristic times is short with respect to the nuclear evolution time, a situation which only occurs at the beginning of the MS:
$$U(r)=\frac{5D}{\mathrm{\Omega }}\frac{\mathrm{\Omega }}{r}.$$
(4)
This equation (Randers 1941; Zahn 1992) expresses that the (inward) flux of angular momentum transported by meridional circulation is equal to the (outward) diffusive flux of angular momentum. As a matter of fact, this solution is equivalent to considering local conservation of angular momentum (Urpin et al 1996).
Instead of eqn. 2, the transport of angular momentum by circulation is often treated as a diffusion process (Endal & Sofia 1978; Pinsonneault et al 1989, 1990; Langer 1991a; Fliegner & Langer 1995; Chaboyer et al 1995ab; Heger et al 2000). We see from eqn. 2 that the term with $`U`$ (advection) is functionally not the same as the term with D (diffusion). Physically advection and diffusion are quite different: diffusion brings a quantity from where there is a lot to other places where there is little. This is not necessarily the case for advection. As an example, the circulation of money in the world is not a diffusive process, but rather an advective one ! Let us make clear that circulation with a positive value of $`U(r)`$, i.e. rising along the polar axis and descending at the equator, is as a matter of fact making an *inward* transport of angular momentum. If this process were treated as a diffusive function of $`\frac{\mathrm{\Omega }}{r}`$, even the sign of the effect may be wrong.
A differential equation like eqn. 2 is subject to boundary conditions at the edge of the core and at the stellar surface. At both places, this condition is usually $`\frac{\mathrm{\Omega }}{r}=0`$, with in addition the assumptions of solid body rotation for the convective core and $`U=0`$ at the surface (Talon et al 1997; Denissenkov et al 1999). If there is magnetic coupling at the surface (Hartmann & Noyes 1987; Pinsonneault et al 1989, 1990; Pinsonneault 1997) or mass loss by stellar winds, the surface condition must be modified accordingly (Maeder 1999a). Various asymptotic regimes for the angular momentum transport can be considered (Zahn 1992) depending on the presence of a wind with or without magnetic coupling.
TRANSPORT OF CHEMICAL ELEMENTS
The transport of chemical elements is also governed by a diffusion–advection equation like eqn. 2 (Endal & Sofia 1978; Schatzman et al 1981; Langer 1991a, 1992; Heger et al 2000). However, if the horizontal component of the turbulent diffusion is large, the vertical advection of the elements (and not that of the angular momentum) can be treated as a simple diffusion (Chaboyer & Zahn 1992) with a diffusion coefficient $`D_{\mathrm{eff}}`$,
$$D_{\mathrm{eff}}=\frac{rU(r)^2}{30D_h},$$
(5)
where $`D_h`$ is the coefficient of horizontal turbulence, for which the estimate is $`D_h=|rU(r)|`$ (Zahn 1992). Eqn. 5 expresses that the vertical advection of chemical elements is severely inhibited by the strong horizontal turbulence characterized by $`D_h`$. Thus, the change of the mass fraction $`X_i`$ of the chemical species i is simply
$$\left(\frac{dX_i}{dt}\right)_{M_r}=\left(\frac{}{M_r}\right)_t\left[(4\pi r^2\rho )^2D_{\mathrm{mix}}\left(\frac{X_i}{M_r}\right)_t\right]+\left(\frac{dX_i}{dt}\right)_{\mathrm{nucl}}.$$
(6)
The second term on the right accounts for composition changes due to nuclear reactions. The coefficient $`D_{\mathrm{mix}}`$ is the sum $`D_{\mathrm{mix}}=D+D_{\mathrm{eff}}`$, where D is the term appearing in expression 2 and $`D_{\mathrm{eff}}`$ accounts for the combined effect of advection and horizontal turbulence. The characteristic time for the mixing of chemical elements is therefore
$$t_{\mathrm{mix}}\frac{R^2}{D_{\mathrm{mix}}}.$$
(7)
Noticeably, the characteristic time for chemical mixing is not $`t_{\mathrm{circ}}`$ given by eqn. 3, as has been generally considered (Schwarzschild 1958). This makes the mixing of the chemical elements much slower, since $`D_{\mathrm{eff}}`$ is very much reduced. In this context, we recall that several authors have reduced by arbitrary factors, up to 30 or 100, the effect of the transport of chemicals in order to better fit the observed surface compositions (Pinsonneault et al 1989, 1991; Chaboyer et al 1995ab; Heger et al 2000). This is no longer necessary with the more appropriate expressions given above.
### 2.4 MERIDIONAL CIRCULATION
Meridional circulation arises from the local breakdown of radiative equilibrium in a rotating star (Vogt 1925; Eddington 1925). In a uniformly rotating star, the equipotentials are closer to each other along the polar axis than along the equatorial axis. Thus, according to von Zeipel’s theorem, the heating on an equipotential is generally higher in the polar direction than in the equatorial direction, which thus drives a large scale circulation rising at the pole and descending at the equator. This problem has been studied for about 75 years (see reviews by Tassoul 1978 or Zahn 1983). The classical formulation (Sweet 1950; Mestel 1953, 1965; Kippenhahn & Weigert 1990) for rigid rotation predicts a value of the vertical velocity of the Eddington–Sweet circulation
$$U_{ES}=\frac{8}{3}\omega ^2\frac{L}{gM}\frac{\gamma 1}{\gamma }\frac{1}{_{\mathrm{ad}}}\left(1\frac{\mathrm{\Omega }^2}{2\pi G\rho }\right),$$
(8)
with $`\omega ^2=\frac{\mathrm{\Omega }^2r^3}{GM_r}`$ the local ratio of centrifugal force to gravity and $`\gamma `$ the ratio of the specific heats $`C_P/C_V`$. The term $`\frac{\mathrm{\Omega }^2}{2\pi G\rho }`$, often called the Gratton–Öpik term (Tassoul 1990), predicts that $`U_{ES}`$ becomes negative at the stellar surface due to the presence of the term $`1/\rho `$. This means an inverse circulation, i.e. descending at the pole and rising at the equator. The dependence in $`\frac{1}{\rho }`$ also makes $`U(r)`$ diverge at the surface. This has led to some controversies on what is limiting $`U(r)`$ (Tassoul & Tassoul 1982, 1995; Zahn 1983; Tassoul 1990). The timescale for circulation mixing defined in eqn. 3 becomes with the above Eddington–Sweet velocity
$$t_{\mathrm{ES}}t_{\mathrm{KH}}\frac{g}{\mathrm{\Omega }^2R},$$
(9)
where $`g`$ is the surface gravity and $`R`$ the stellar radius. Even for modest rotation velocities, $`t_{\mathrm{ES}}`$ is much shorter than the MS lifetime (Schwarzschild 1958; see also Denissenkov et al 1999), so that most stars should be mixed, if this timescale were applicable. However, the presence of $`\mu `$–gradients was not taken into account in the above expressions. When this is done, rotation is found to allow circulation only above a certain rotation limit that depends on the value of the $`\mu `$–gradient (Mestel 1965; Kippenhahn 1974; Kippenhahn & Weigert 1990).
The velocity of the meridional circulation in the case of shellular rotation was derived by Zahn (1992), who considered the effects of the latitude–dependent $`\mu `$–distribution (Mestel 1953, 1965). Even more important are the effects of the vertical $`\mu `$–gradient $`_\mu `$ and of the horizontal turbulence (Maeder & Zahn 1998). Contrary to the conclusions of the previous works, the $`\mu `$–gradients were shown not to introduce a velocity threshold for the existence of the circulation, but to progressively reduce the circulation when $`_\mu `$ increases. One has then
$`U(r)`$ $`=`$ $`{\displaystyle \frac{P}{\rho gC_PT[_{\mathrm{ad}}+(\phi /\delta )_\mu ]}}\left\{{\displaystyle \frac{L}{M_{}}}(E_\mathrm{\Omega }+E_\mu )\right\},`$ (10)
where P is the pressure, $`C_P`$ the specific heat, $`E_\mathrm{\Omega }`$ and $`E_\mu `$ are terms depending on the $`\mathrm{\Omega }`$– and $`\mu `$–distributions respectively, up to the third order derivatives. Since the derivative of $`U(r)`$ appears in eqn. 2, we see that the consistent solution of the problem is of fourth order (Zahn 1992). This makes the numerical solution difficult (Talon et al 1997; Denissenkov et al 1999; Meynet & Maeder 2000). While the classical solution predicts an infinite velocity at the interface between a radiative and a semiconvective zone with an inverse circulation in the semiconvective zone, expression 10 gives a continuity of the solution with no change of sign. In evolutionary models, the term $`_\mu `$ in eqn. 10 may be one or two orders of magnitude larger than $`_{\mathrm{ad}}`$ in some layers, so that $`U(r)`$ may be reduced by the same ratio. This considerably increases the characteristic time $`t_{\mathrm{circ}}`$ with respect to the classical estimate $`t_{\mathrm{ES}}`$.
### 2.5 INSTABILITIES AND TRANSPORT
The subject of the instabilities in moving plasmas is a field in itself. Here we limit ourselves to a short description of the main instabilities currently considered influential in the evolution of Upper MS stars, (Endal & Sofia 1978; Zahn 1983, 1992; Pinsonneault et al 1989; Heger et al 2000).
CONVECTIVE AND SOLBERG-HØILAND INSTABILITY
In a rotating star, the Ledoux or Schwarzschild criteria for convective instability should be replaced by the Solberg–Høiland criterion (Kippenhahn & Weigert 1990). This criterion accounts for the difference of the centrifugal force for an adiabatically displaced fluid element; the condition for dynamical stability is
$$N^2+\frac{1}{s^3}\frac{d(s^2\mathrm{\Omega })^2}{ds}0.$$
(11)
The Brunt-V is l frequency $`N^2`$ is given by $`N^2=\frac{g\delta }{H_p}\left(_{\mathrm{ad}}+\frac{\phi }{\delta }_\mu \right)`$, where the various symbols have their usual meaning (Kippenhahn & Weigert 1990), and the term $`s`$ is the distance to the rotation axis. For no rotation, the Ledoux criterion is recovered. If the thermal effects are ignored, we just recover Rayleigh’s criterion for stability, which says that the specific angular momentum $`s^2\mathrm{\Omega }`$ must increase with the distance to the rotation axis. In practice, convective stability is reinforced by rotation. In the absence of rotation, a zone located beween the places where $`=_{\mathrm{ad}}+_\mu `$ and $`=_{\mathrm{ad}}`$ is called semiconvective. There, non–adiabatic effects can drive growing oscillatory instabilities (Kato 1966; Kippenhahn & Weigert 1990). An appropriate diffusion coefficient describing the transport in such zones has been derived by Langer et al (1983). However there is no diffusion coefficient yet available for semiconvective mixing in the presence of rotation.
The assumption of solid body rotation is generally made in convective regions, owing to the strong turbulent coupling. However, the collisions or scattering of convective blobs influence the rotation law in convective regions (Kumar et al 1995): solid body rotation only occurs for an isotropic scattering. For some forms of anisotropic scattering, an outward rising rotation profile such as that observed in the Sun, can be produced. Two–dimensional models of rotating stars (Deupree 1995, 1998) also show that the angular velocity in convective cores is not uniform, but it decreases with distance from the center, being about constant on cylinders. A considerable overshoot is obtained by Deupree (1998), amounting to about 0.35 $`H_P`$, where $`H_P`$ is the local pressure scale height. Similar conclusions are obtained by Toomre (1994), who also finds penetrative convection at the edge of the convective core of rotating A–type stars. The very large Reynolds number characterizing stellar turbulence prevents direct numerical simulations of the Navier–Stokes equation, so that some new specific methods have been proposed to study convective turbulence (Canuto 1994; Canuto et al 1996; Canuto & Dubovikov 1997) and its interplay with differential rotation (Canuto et al 1994; Canuto 1998); these last results have not yet been applied to evolutionary models in rotation.
SHEAR INSTABILITIES: DYNAMICAL AND SECULAR
In a radiative zone, shear due to differential rotation is likely to be a very efficient mixing process. Indeed shear instability grows on a dynamical timescale that is of the order of the rotation period (Zahn 1992, 1994). Stability is maintained when the Richardson number $`Ri`$ is larger than a critical value $`Ri_{\mathrm{cr}}`$
$$Ri=\frac{N^2}{\left(\frac{dV}{dz}\right)^2}>Ri_{\mathrm{cr}}=\frac{1}{4},$$
(12)
where $`V`$ is the horizontal velocity and $`z`$ the vertical coordinate. Equation 12 means that the restoring force of the density gradient is larger than the excess energy $`\frac{1}{4}\left(\frac{dV}{dz}\right)^2`$ present in the differentially rotating layers (Chandrasekhar 1961). In eqn. 12, heat exchanges are ignored and the criterion refers to the “dynamical shear instability” (Endal & Sofia 1978; Kippenhahn & Weigert 1990).
When thermal dissipation is significant, the restoring force of buoyancy is reduced and the instability occurs more easily (Endal & Sofia 1978), its timescale is however longer, being the thermal timescale. This case is sometimes referred to as “secular shear instability”. For small thermal effects ($`Pe1`$), a factor equal to $`Pe`$ appears as multiplying $`N^2`$ in eqn. 12 (Zahn 1974). The number $`Pe`$ is the ratio of the thermal cooling time to the dynamical time, i.e. $`Pe=\frac{v\mathrm{}}{K}`$ where $`v`$ and $`\mathrm{}`$ are the characteristic velocity and length scales, and $`K=(4acT^3)/(3C_P\kappa \rho ^2)`$ is the thermal diffusivity. $`Pe`$ varies typically from $`10^9`$ in deep interior to $`10^2`$ in outer layers (Cox & Giuli 1968). For general values of $`Pe`$, a more general expression of the Richardson’s criterion can be found (Maeder 1995; Maeder & Meynet 1996); it is consistent with the case of low $`Pe`$ treated by Zahn (1974). The problem of the Richardson criterion has also been considered by Canuto (1998), who suggests that for $`Pe>1`$, i.e. negligible radiative losses $`Ri_{\mathrm{cr}}1`$ and for $`Pe<1`$, i.e. important radiative losses $`Ri_{\mathrm{cr}}Pe^1`$. Thus, similar dependences with respect to $`Pe`$ are obtained, but Canuto (1998) finds that turbulence may exist beyond the $`\frac{1}{4}`$ limit in eqn. 12.
Many authors have shown that the $`\mu `$–gradients appear to inhibit the mixing too much with respect to what is required by the observations (Chaboyer et al 1995ab; Meynet & Maeder 1997; Heger et al 2000). Changing $`Ri_{\mathrm{cr}}`$ from $`\frac{1}{4}`$ to $`1`$ does not solve the problem, the difference being a matter of one or two orders of magnitude. For example, instead of using a gradient $`_\mu `$, some authors write $`f_\mu _\mu `$ with a factor $`f_\mu =0.05`$ or even smaller (Heger et al 2000; see also Chaboyer et al 1995ab). Most of the zone external to the convective core, where the $`\mu `$–gradient inhibits mixing, is as a matter of fact semiconvective and is thus subject to thermal instability anyway. This has led to the hypothesis that the excess energy in the shear is degraded by turbulence on the thermal timescale, changing the entropy gradient and consequently the $`\mu `$–gradient (Maeder 1997). This gives a diffusion coefficient $`D_{\mathrm{shear}}`$, which tends towards the diffusion coefficient for semiconvection by Langer et al (1983) when shear is negligible, and towards the value $`D_{\mathrm{shear}}=(K/N^2)(\mathrm{\Omega }\frac{d\mathrm{ln}\mathrm{\Omega }}{d\mathrm{ln}r})^2`$ given by Zahn (1992) when semiconvection is negligible. Another proposition has been made by Talon & Zahn (1997), who take into account the homogeneizing effect of the horizontal diffusion on the restoring force produced by the $`\mu `$–gradient. This also reduces the excessive stabilizing effect of the $`\mu `$–gradient. Both the above suggestions lead to an acceptable amount of mixing in view of the observations. We stress that the Reynolds condition $`D_{\mathrm{shear}}\frac{1}{3}\nu Re_c`$ must be satisfied when the medium is turbulent, where $`Re_c`$ is the critical Reynolds number (Zahn 1992; Denissenkov et al 1999) and $`\nu `$ is the total viscosity (radiative + molecular).
Globally, we may expect the secular instability to work during the MS phase, where the $`\mathrm{\Omega }`$–gradients are small and the lifetimes are long, while the dynamical instability could play a role in the advanced stages.
OTHER INSTABILITIES
Baroclinicity, i.e. the non–coincidence of the equipotentials and surface of constant $`\rho `$, generates various instabilities in the case of non–conservative rotation laws. Some instabilities are axisymmetric, like the GSF instability (Goldreich & Schubert 1967; Fricke 1968; Korycansky 1991). The GSF instability is created by fluid elements displaced between the directions of constant angular momentum and of the rotational axis. Stability demands a uniform or constant rotation on cylinders, which is incompatible with shellular rotation. The GSF instability thus favours solid–body rotation, on a time–scale of the order of that of the meridional circulation (Endal & Sofia 1978). This instability is however inhibited by the $`\mu `$–gradients (Knobloch & Spruit 1983), nevertheless Heger et al (2000) find it to play a role near the end of the helium–burning phase. Another axisymmetric instability is the ABCD instability (Knobloch & Spruit 1983). Fluid elements displaced between the surfaces of constant $`P`$ and $`T`$ create the ABCD instability, a kind of horizontal convection. The ABCD instability is an oscillatory one and its efficiency is difficult to estimate for now. Non–axisymmetric instabilities, like salt–fingers, may also occur (Spruit & Knobloch 1984). They are not efficient when rotation is low. However in the case of fast rotation they may occur everywhere in rotating stars, so that one–dimensional models are likely to be an unsatisfactory idealization in this case.
The study of the transport of angular momentum by gravity waves has been stimulated by the finding of an almost solid body rotation for most of the radiative interior of the Sun (Schatzman 1993; Montalban 1994; Kumar & Quataert 1997; Zahn et al 1997; Talon & Zahn 1998). Gravity waves are supposed to transport angular momentum from the external convective layers to the radiative interior. However, Ringot (1998) has recently shown that quasi–solid rotation of the radiative zone of the Sun cannot be a direct consequence of the action of gravity waves. Thus, even in the Sun the question remains open.
In Upper MS stars, we could expect gravity waves to be generated by turbulent motions in the convective core (Denissenkov et al 1999). The momentum will be deposited where the Doppler shift of the waves due to differential rotation is equal to the initial wave frequency. From the work by Montalban & Schatzman (1996), we know that in general the deposition of energy decreases very quickly away from the boundaries of a convective zone. The same is found by Denissenkov et al (1999), who show that uniform rotation sustained by gravity waves is limited to the very inner radiative envelope; the size of the region of uniform rotation enforced by gravity waves likely increases with stellar mass. Only angular momentum may be directly transported by gravity waves and not chemical elements. Nevertheless, the transport of momentum by waves, which reduces the differential rotation, could also influence indirectly the distribution of chemical elements in stars.
### 2.6 MASS LOSS AND ROTATION
Mass loss by stellar winds is a dominant effect in the evolution of Upper MS stars (Chiosi & Maeder 1986). The mass loss rates currently applied in stellar models are based on the observations (de Jager et al 1988; Lamers & Cassinelli 1996). A significant growth of the mass flux of OB stars with rotation, i.e. by 2–3 powers of 10, was found by Vardya (1985). Nieuwenhuijzen & de Jager (1988) suggested that the correlation found by Vardya mainly reflects the distributions of the mass loss rates $`\dot{M}`$ and of the rotation velocities $`v_{\mathrm{rot}}`$ over the HR diagram. After trying to disentangle the effects of $`L`$, $`T_{\mathrm{eff}}`$ and $`v_{\mathrm{rot}}`$, they found that the $`\dot{M}`$-rates seem to increase only slightly with rotation for O– and B–type stars. The result by Vardya might not be incorrect, since when the data for OB stars by Nieuwenhuijzen & de Jager (1988) are considered, a correlation of the mass fluxes with $`v_{\mathrm{rot}}`$ is noticeable. These authors also point out that the equatorial $`\dot{M}`$–rates of Be–stars are larger by a factor $`10^2`$. Since Be–stars are essentially B–stars with fast rotation, a single description of the large changes of the $`\dot{M}`$–rates from the low to the high values of $`v_{\mathrm{rot}}`$ should be considered.
On the theoretical side, Pauldrach et al (1986) and Poe & Friend (1986) find a very weak change of the $`\dot{M}`$–rates with $`v_{\mathrm{rot}}`$ for O–stars: the increase amounts to about 30% for $`v_{\mathrm{rot}}`$ = 350 km/sec. Friend & Abbott (1986) find an increase of the $`\dot{M}`$–rates which can be fitted by the relation (Langer 1998; Heger & Langer 1998)
$$\dot{M}(v_{\mathrm{rot}})=\dot{M}(v_{\mathrm{rot}}=0)\left(\frac{1}{1\frac{v_{\mathrm{rot}}}{v_{\mathrm{crit}}}}\right)^\xi $$
(13)
with $`\xi `$ = 0.43; this expression is often used in evolutionary models.
The previous wind models of rotating stars are incomplete since they do not account for the von Zeipel theorem. The gravity darkening at the equator leads to a reduction of the equatorial mass flux (Owocki et al 1996; Owocki & Gayley 1997, 1998). This leads to very different predictions for the wind morphology than those of the current wind–compressed disk model by Bjorkman & Cassinelli (1993), which is currently advocated to explain disk formation. Equatorial disks may however form quite naturally around rotating stars. The theory of radiative winds, with revised expressions of the von Zeipel theorem and of the Eddington factor, has been applied to rotating stars (Maeder 1999a). There are two main sources of wind anisotropies: 1. The “$`g_{\mathrm{eff}}`$–effect” which favours polar ejection, since the polar caps of a rotating star are hotter. 2. The “opacity or $`\kappa `$–effect”, which favours an equatorial ejection, when the opacity is large enough at the equator due to an opacity law which increases rapidly with decreasing temperature. In O–type stars, since opacity is due mainly to the T–independent electron scattering, the $`g_{\mathrm{eff}}`$–effect is likely to dominate and to raise a fast highly ionized polar wind. In B– and later type stars, where a T–growing opacity is present in the external layers, the opacity effect should favour a dense equatorial wind and ring formation, with low terminal velocities and low ionization.
The so–called B\[e\] stars (Zickgraf 1999) are known to show both a fast, highly ionized polar wind and a slow, dense, low ionized equatorial ejection and may be a template of the $`g_{\mathrm{eff}}`$– and $`\kappa `$–effects. At some values of T, the ionization equilibrium of the stellar wind changes rather abruptly and so does the opacity as well as the force–multipliers which characterize the opacities (Kudritzki et al 1989; Lamers 1997; Lamers 1999). Such transitions, called the bi–stability of stellar winds by Lamers, may favour strong anisotropies of the winds, and even create some symmetrical rings at the latitude where an opacity–peak occurs on the T–varying surface of the star.
It could be thought at first sight that wind anisotropies have no direct consequences for stellar evolution. This is not at all the case. Like magnetic coupling for low mass stars, the anisotropic mass loss removes selectively the angular momentum (Maeder 1999a) and influences the further evolution. Winds through polar caps, as likely in O–stars, remove very little angular momentum, while equatorial mass loss removes a lot of angular momentum from the stellar surface.
## 3 MAIN SEQUENCE EVOLUTION OF ROTATING STARS
### 3.1 EVOLUTION OF THE INTERNAL ROTATION
As a result of transport processes, contraction and expansion, the stars should be differentially rotating, with a strong horizontal turbulence enforcing a rotation law of the form $`\mathrm{\Omega }`$ constant on isobars (Zahn 1992). The whole problem must be treated self–consistently, because the differential rotation in turn determines the behaviours of the meridional circulation and turbulence, which themselves contribute to differential rotation. There are various approximations to treat the above physical problem (Pinsonneault et al 1989, 1991; Chaboyer et al 1995ab; Langer 1998; Heger & Langer 1998; Heger et al 2000). In some works, the assumption of rigid rotation is made, while in other works advection is treated as a diffusion with the risk that even the sign of the effect is the wrong one! Some authors, in order to fit the observations, introduce several efficiency factors $`f_\mu ,f_c`$, etc… The problem is that the sensitivity of the results to these many efficiency factors is as large, or even larger than the sensitivity to rotation.
A simplification has been applied by Urpin et al (1996), who assume equilibrium between the outward transport of angular momentum by diffusion and the inward transport by circulation, as also suggested by Zahn (1992). This is the stationary case discussed in Sect. 2.3 for which eqn. 4 applies. The values of $`U(r)`$ are always positive and the circulation has only one loop. Urpin et al (1996) point out that the stationary distribution of $`\mathrm{\Omega }`$ arranges itself so as to reduce $`U(r)`$ to a minimum value over the bulk of the star. This makes the values of $`U(r)`$ of the order of $`10^510^6`$ cm/s, quite insufficient to produce any efficient mixing in a 20 M star.
The initial non–stationary approach to equilibrium in 10 and 30 M stars has been studied by Denissenkov et al (1999). In a very short time of about 1% of the MS lifetime $`t_{MS}`$, $`\mathrm{\Omega }(r)`$ converges towards a profile with a small degree of differential rotation and with very small values of $`U(r)`$ (Urpin et al 1996). The circulation shows two cells, an internal one rising along the polar axis and an external one descending at the pole. The evolution is not calculated, but it is noted that the timescale $`t_{\mathrm{circ}}`$ (which behaves like $`\mathrm{\Omega }^2`$) is very short with respect to the MS lifetime for most Upper MS stars and that the ratio $`\frac{t_{\mathrm{mix}}}{t_{MS}}1`$ (eqn. 7). It is noticeable that $`t_{\mathrm{circ}}t_{\mathrm{mix}}`$, so that no efficiency factors are needed to reduce the mixing of chemical elements compared to the transport of angular momentum. This reduction naturally results from the effect of the horizontal turbulence.
The full evolution of the rotation law has been studied with the non–stationary scheme for a 9 M star by Talon et al (1997), and for stars from 5 to 120 M including the effects of mass loss by Meynet & Maeder (2000). The very fast initial convergences of $`\mathrm{\Omega }(r)`$ and of $`U(r)`$ are confirmed. However, after convergence the asymptotic state of $`U(r)`$ does not correspond to the stationary approximation. In the full solution, $`U(r)`$ changes sign in the external region and thus transports some angular momentum outward, which is not the case in the stationary solution. Also, contrary to the classical result of the Eddington–Sweet circulation (eqn. 8), it is found that $`U(r)`$ depends very little on the initial rotation.
Fig. 1 shows the evolution of $`\mathrm{\Omega }(r)`$ during MS evolution of a 20 M star. Mass loss at the stellar surface removes a substantial fraction of the total angular momentum, which makes $`\mathrm{\Omega }(r)`$ decrease with time everywhere in the star. The outer zone with inverse circulation progressively deepens during MS evolution, because a growing part of the outer layers has lower densities. This inverse circulation contributes to the outward transport of angular momentum. The deepening of the inverse circulation also has the consequence that the stationary and non–stationary solutions differ more and more as the evolution proceeds, since no inverse circulation is predicted by the stationary solution. This shows that the stationary solutions are too simplified and that outward and inward transport never reach exact equilibrium, contrary to the initial expectations.
Fig. 2 shows the various diffusion coefficients inside a 20 M star when the hydrogen mass fraction at the center is equal to 0.20. We notice that in general $`KD_hD_{\mathrm{shear}}D_{\mathrm{eff}}`$. This confirms the basic hypothesis of a large $`D_h`$ necessary for the validity of the assumption of shellular rotation.
The characteristic time $`t_{\mathrm{mix}}`$ of the mixing processes is of the same order as the lifetime $`t_\mathrm{H}`$ of the H–burning phase for the Upper MS stars (Maeder 1987). Indeed, if shear mixing is the dominant mixing process, the timescale is $`t_{\mathrm{mix}}\frac{R^2}{D_{\mathrm{shear}}}`$, and for a given degree of differential rotation it behaves like $`t_{\mathrm{mix}}K^1`$, which itself goes like about $`M^{1.7}`$. The timescale $`t_\mathrm{H}`$ behaves like $`M^{0.7}`$ for $`M15`$ M (Maeder 1998). Thus, for larger masses $`t_{\mathrm{mix}}`$ tends to decrease much faster than $`t_\mathrm{H}`$ and mixing processes grow in importance. From the end of MS evolution when $`X_c0.05`$, central contraction starts dominating the evolution of the central $`\mathrm{\Omega }(r)`$, which grows quickly until core collapse. During these post–MS phases, the average value of $`t_{\mathrm{mix}}`$ will be longer than the nuclear lifetimes. Thus, the rotational mixing processes during these phases are likely to be globally unimportant (Heger et al 2000; Meynet & Maeder 2000). Nevertheless, it is likely that in some regions of a rotating star in the advanced stages, as a result of extreme central contraction, the $`\mathrm{\Omega }`$–gradient may become so large that some other local instabilities develop leading to fast mixing.
### 3.2 EVOLUTION OF $`V_{\mathrm{rot}}`$. THE CASE OF Be STARS.
The evolution of the surface rotational velocities $`v_{\mathrm{rot}}`$ at the equator is a consequence of the processes discussed above. Let us first consider the two extreme cases of coupling and of no coupling between adjacent layers, first examined in the early works by Oke & Greenstein (1954) and Sandage (1955): a) the case of rigid rotation and b) the case of local conservation. a) For rigid rotation, $`v_{\mathrm{rot}}`$ remains nearly constant during the MS phase (Fliegner & Langer 1995). This is because the effects of core contraction and envelope expansion nearly compensate for one another. b) For local conservation of the angular momentum, one has $`v_{\mathrm{rot}}=\mathrm{\Omega }RR^1`$, while the critical velocity changes like $`v_{\mathrm{crit}}R^{\frac{1}{2}}`$. Thus, the inflation of the stellar radius $`R`$ during evolution makes rotation less and less critical. The opposite effect may occur during a bluewards crossing of the HR diagram and then critical rotation may be reached (Sect. 5.2). Endal & Sofia (1979) have shown that the ratio $`v_{\mathrm{rot}}(casea)/v_{\mathrm{rot}}(caseb)1.8`$ before the crossing of the HR diagram for intermediate mass stars. The truth generally lies between the two cases, closer to the rigid case during the MS phase since transport processes have more time to proceed. In post–MS phases, and in particular during the fast crossings of the HR diagram, the evolution timescale is short, so that little transport occurs and the evolution of $`v_{\mathrm{rot}}`$ closely resembles that of local conservation.
Stars in solid body rotation may reach the break–up limit before the end of the MS phase even for moderate initial $`v_{\mathrm{rot}}`$ (Sackmann & Anand 1970; Langer 1997). However this is a consequence of the simplified assumption of solid body rotation. With diffusion and transport, it is less easy for the star to reach the break–up limit. For a 20 M model without mass loss (dotted line in Fig. 3 and 4), $`v_{\mathrm{rot}}`$ grows and the ratio $`\frac{\mathrm{\Omega }}{\mathrm{\Omega }_{\mathrm{crit}}}`$ may become close to 1 before the end of the MS phase. For stars with M $`<`$ 15 M, where mass loss is small, the ratio $`\frac{\mathrm{\Omega }}{\mathrm{\Omega }_{\mathrm{crit}}}`$ also increases during the MS phase and the break–up or $`\mathrm{\Omega }`$–limit may be reached during the MS phase. If so, the mass loss should then become very intense, until the velocity again becomes subcritical. It is somehow paradoxical that no or small mass loss rates during the bulk of the MS phase (as at low metallicity Z) may lead to very high mass loss at the end of the MS evolution for rotating stars. This may be the cause of some ejection processes as in Luminous Blue Variables (LBV), B\[e\] and Be stars. We may wonder whether the higher relative number of Be stars observed in lower Z regions (Maeder et al 1999) is just a consequence of the lower average mass loss in lower metallicity regions or whether this is related to star formation.
If mass loss is important during MS evolution, $`v_{\mathrm{rot}}`$ decreases substantially (Fig. 3 and 4). Therefore it is not surprising that Be–stars, which likely are close to break–up (Slettebak 1966) do not form among O–type stars, but mainly among B–type stars, with a relative maximum at type B3. Indeed, to form a Be star it is probably not necessary that the break–up limit is exactly reached. The conditions for an equatorial ejection responsible for the Be spectral features occur when the $`\kappa `$–effect is important (Sect. 2.6), which requires that the equatorial regions of the fastly rotating star are below the bi–stability limit. Of course, the higher the rotation, the higher the equatorial mass loss will be.
Any magnetic coupling between the star and the wind would dramatically reduce $`v_{\mathrm{rot}}`$. However, such a coupling does not seem to be important in general, except for Bp and Ap stars. MacGregor et al (1992) show that, even in the presence of a small magnetic field of 100 G, the rotation velocities of OB stars should be much lower than observed. This result is in agreement with that of Mathys (1999), who finds no detectable magnetic field in hot stars.
EFFECTS OF MASS LOSS ON ROTATION
Mass loss by stellar winds drastically reduces $`v_{\mathrm{rot}}`$ during the evolution (Packet et al 1980; Langer & Heger 1998; Figs. 3 and 4). Even if isotropic, the stellar winds carry away quite a lot of angular momentum, and this is even more important in the case of equatorial mass loss. The new surface layers then have a lower $`v_{\mathrm{rot}}`$ as a result of expansion and redistribution.
With the simplified assumption of solid body rotation for a 60 M model with mass loss, Langer (1997, 1998) finds a convergence of $`v_{\mathrm{rot}}`$ towards the critical value (the $`\mathrm{\Omega }`$–limit), before the end of the MS phase, for all initial velocities above 100 km/s. The overall result is that the final velocities are the same, all being at the critical limit, while the final MS masses strongly depend on the initial rotation. This convergence towards the $`\mathrm{\Omega }`$–limit clearly results also from the simplified assumption of rigid rotation, which grossly exagerates the coupling of the surface layers.
Fig. 3 illustrates the decrease in $`v_{\mathrm{rot}}`$ during evolution when the various transport mechanisms are followed in detail. The decrease in $`v_{\mathrm{rot}}`$ is much larger for larger initial stellar masses, because mass loss is larger for them. The same is true for $`\frac{\mathrm{\Omega }}{\mathrm{\Omega }_{\mathrm{crit}}}`$ (Fig. 4). For $`M40`$ M the velocities $`v_{\mathrm{rot}}`$ will remain largely subcritical for all initial velocities, except during the overall contraction phase at the end of MS evolution. For a given initial mass, the resulting scatter of $`v_{\mathrm{rot}}`$ should be smaller at the end of the MS phase, but this is not a convergence towards the $`\mathrm{\Omega }`$–limit as for models with solid body rotation (Langer 1997, 1998). The final masses are of course lower for larger rotation, since mass loss is enhanced by rotation, this results in a large scatter of masses and $`v_{\mathrm{rot}}`$ at a given luminosity.
COMPARISON WITH OBSERVATIONS
Only a few comparisons between the observed $`v_{\mathrm{rot}}`$ and the model predictions have been made until now. Conti & Ebbets (1977) found that $`v_{\mathrm{rot}}`$ in O–type giants and supergiants is not as low as expected from models with rotation conserved in shells. This fact and the relative absence of low rotators among the evolved O–stars led them to conclude that another line broadening mechanism, such as macroturbulence, should be present in these objects. The same conclusion based on similar arguments was supported by Penny (1996) and Howarth et al (1997). This conclusion needs to be checked further, since the decrease predicted by the new models is not as fast as for local conservation.
It is well–known that the average value of $`v_{\mathrm{rot}}`$ increases from the early O–type to the early B–type stars (Slettebak 1970). This may be the signature of the effect of higher losses of mass and angular momentum in the more massive stars (Penny 1996), which leads to a lower average rotation in the course of the MS phase. We notice that the increase of $`v_{\mathrm{rot}}`$ from O to B stars is larger for the stars of class IV than for the stars of class V (Fukuda 1982), a fact which is expected since the difference due to mass loss is more visible near the end of the MS phase (Fig. 3). The stars of luminosity class I (Fukuda 1982) show a fast decline of the average $`v_{\mathrm{rot}}`$ from the O–type to the B–type stars. As the supergiants of class I originate from the most massive stars, which evolve at about constant luminosity, this last effect could be related to the fact that for a given initial mass, $`v_{\mathrm{rot}}`$ declines strongly as the star moves away from the MS (Langer 1998).
### 3.3 THE HR DIAGRAM, LIFETIMES AND ISOCHRONES
As always in stellar evolution, the shape of the tracks is closely related to the internal distribution of the mean molecular weight $`\mu `$. All results show that the convective cores are slightly increased by rotation (Maeder 1987, 1998; Langer 1992; Talon et al 1997; Meynet 1998, 1999; Heger et al 2000). The height of the $`\mu `$–discontinuity at the edge of the core is reduced and there is a mild composition gradient built up from the core to the surface, which may then be slightly enriched in helium and nitrogen. For low or moderate rotation, the convective core shrinks as usual during MS evolution, while for high masses ($`M40`$ M) and large initial rotations ($`\frac{\mathrm{\Omega }}{\mathrm{\Omega }_{\mathrm{crit}}}0.5`$), the convective core grows in mass during evolution. These behaviours, i.e. reduction or growth of the core, determine whether the star will follow respectively the usual redwards MS tracks in the HR diagram, or whether it will bifurcate to the blue (cf. Maeder 1987; Langer 1992) towards the classical tracks of homogenous evolution (Schwarzschild 1958) and likely produce W–R stars (Sect. 5.4). Also for O– and B–type stars, fast rotation increases the He–content of the envelope and the decrease of the opacity also favours a bluewards track.
Fig. 5 shows the overall HR diagram for rotating and non–rotating stars during the MS phase and slightly beyond. The atmospheric distortions produce a shift to the red in the HR diagram by several 0.01 in (B-V), with on the average only a small change of luminosity (Maeder & Peytremann 1970; Collins & Sonneborn 1977). During MS evolution, the luminosity of the rotating stars grows faster and the tracks extend farther away from the ZAMS, as in the case of a moderate overshooting (Maeder 1987; Langer 1992; Sofia et al 1994; Talon et al 1997). This effect introduces a significant scatter in the mass–luminosity relation (Meynet 1998), in the sense that fast rotators are overluminous with respect to their actual masses. This may explain some of the discrepancies between the evolutionary masses and the direct mass estimates in some binaries (Penny et al 1999). In this context, we recall that for a decade a severe mass discrepancy beween spectroscopic and evolutionary masses was claimed by some authors (Groenewegen et al 1989; Kudritzki et al 1992; Herrero et al 1992). Most of the problem has collapsed and was shown to be a result of the proximity of O–stars to the Eddington limit (Lamers & Leitherer 1993; Herrero et al 1999) and the large effect of metal line blanketing not usually accounted for in the atmosphere models of massive stars (Lanz et al 1996).
There is little difference between tracks with $`v_{\mathrm{rot}}`$ = 200 or 300 km/s (Meynet & Maeder 2000; see also Talon et al 1997). If the effects behaved like $`v_{\mathrm{rot}}^2`$, there would be larger differences. This saturation effect occurs because outward transport of angular momentum by shears are larger when rotation is larger, also a larger rotation produces more mass loss, which further reduces rotation during the evolution.
LIFETIMES AND ISOCHRONES
The lifetimes $`t_\mathrm{H}`$ in the H–burning phase grow only moderately because more nuclear fuel is available, but at the same time the luminosity is larger. The net result is an increase by about 20 to 30 % for an initial velocity of 200 km/s (Talon et al 1997; Meynet & Maeder 2000). This influences the isochrones and the age determinations. As an example, for $`v_{\mathrm{rot}}`$ = 200 km/s, the isochrone of log age = 7.0 is the same as that of log age = 6.90 without rotation (Meynet 2000). Thus, accounting for rotation could lead to ages larger by about 25 % for O– and early B–type stars. However, since the cluster ages are generally determined on the basis of the blue envelope of the observed sequence, where most low rotators lie (Maeder 1971), it is likely that the effect in current age determinations is rather small. If a bluewards track occurs, the larger core and mixing lead to much longer lifetimes in the H–burning phase. In this case, the fitting of time–lines becomes hazardous.
## 4 ROTATION AND CHEMICAL ABUNDANCES
### 4.1 OBSERVATIONS
The chemical abundances are a very powerful test of internal evolution and they give strong evidence in favour of some additional mixing processes in O– and B–type stars, in supergiants and in red giants of lower masses.
ABUNDANCES IN MS O–AND B–TYPE STARS
Many evidences of He– and N–excesses in O–type and early B–type stars have been reported over the last decade (Gies & Lambert 1992; Herrero et al 1992, 1998; Kilian 1992; Kendall et al 1995, 1996; Lyubimkov 1996, 1998). We can extract the following main points: 1.– The OBN stars show significant He– and N–excesses. OBN stars are more frequent among stars above 40 M (Walborn 1988; Schönberner et al 1988; Herrero et al 1992). 2.– All fast rotators among the O–stars show some He–excesses (Herrero et al 1992, 1998, 1999; Lyubimkov 1996). 3.– Although rather controversial initially, there seems to be an increase of the He– and N–abundances with the relative age (i.e. the fraction $`t/t_{\mathrm{MS}}`$ of the MS lifetime spent) for the early B–type stars (Lyubimkov 1991, 1996; Gies & Lambert 1992, see note added in proof; Denissenkov 1994). Lyubimkov (1996) suggests a sharp rise from $`(He/H)`$ = 0.08 – 0.10 to 0.20 in number for O–stars when $`t/t_{\mathrm{MS}}0.50.7`$, while for B–type stars the corresponding value rises to 0.12 – 0.14. As to nitrogen, its abundance is estimated to rise to about 3 times for a 14 M and to 2 for a 10 M star. An increase of the N–abundance by a factor 2 – 3 for an O–star with the average $`v_{\mathrm{rot}}`$ of 200 km/s is the order of magnitude typically considered as a constraint for recent stellar models (Heger et al 2000). 4.– The boron abundances in five B–stars on the MS have been found to be smaller by at least a factor 3 or 4 than the cosmic meteoritic value (Venn et al 1996). The boron depletion occurs in stars which also show N–excess and this is supporting the idea that rotational mixing occurs throughout the star (Fliegner et al 1996).
ABUNDANCES IN SUPERGIANTS
The main observations are the following ones: 1.– He– and N–excesses seem to be the rule among OB–supergiants (Walborn 1988). According to Walborn, only the small group of the “peculiar” OBC stars has the normal cosmic abundances. An excess of He, sometimes called the “helium discrepancy”, and corresponding excesses of N have been found by a number of authors (Voels et al 1989; Lennon et al 1991; Gies and Lambert 1992; Herrero et al 1992, 1999 ; Smith and Howarth 1994; Venn 1995ab; Crowther 1997; McEarlan 1998; McEarlan et al 1999). As shown by these last authors, the determination of the helium abundance also depends on the adopted value for microturbulence. Villamariz & Herrero (1999) and Herrero et al (1999) point out however that the helium discrepancy is only reduced, but not solved when microturbulence is accounted for. 2.– Evidence of highly CNO processed material is present for B–supergiants in the range 20 – 40 M, (McEarlan et al 1999). Values of \[N/H\] (i.e. the difference in log with respect to the solar values) amounting to 0.6 dex have been found for B–supergiants around 20 M (Venn 1995ab). Such values are in agreement with the enrichments found in the ejecta of SN 1987A (Fransson et al 1989). 3.– The values of \[N/H\] for galactic A–type supergiants around 12 M lie between 0 and 0.4 dex (Venn 1995ab, 1999). All these values are globally consistent with the above results of Lyubimkov (1996). Takeda & Takada-Hidai (1995) have suggested that these excesses are larger for larger masses, a result in agreement with theory and also recently confirmed by McEarlan et al (1999). For the A–type supergiants in the SMC, the N/H excesses are much larger spanning a range \[N/H\] = 0 to 1.2 dex (Venn et al 1998; Venn 1999). 4.– Na–excesses have been found in yellow supergiants (Boyarchuk & Lyubimkov 1983) and the overabundances also seem to be larger for higher mass stars (Sasselov 1986). Two different explanations have been proposed, one based on the reaction <sup>22</sup>Ne(p,$`\gamma )^{23}`$Na (Denissenkov 1994), the other one based on <sup>20</sup>Ne(p,$`\gamma )^{21}`$Na (Prantzos et al 1991), with some additional mixing processes in both cases. The latter reaction seems to have a too low rate, while the first one may work (Denissenkov 1994). The important point is that the observed Na excesses imply some mixing from the deep interior to the surface. 5.– There are only very few abundance determinations in yellow and red supergiants. Some excesses of N with respect to C and O have been found by Luck (1978). Barbuy et al (1996) found both N–enrichments and normal compositions among the slow rotating F–G supergiants. Isotopic ratios <sup>13</sup>C/<sup>12</sup>C, <sup>17</sup>O/<sup>16</sup>O and <sup>18</sup>O/<sup>17</sup>O for red supergiants would provide very useful information. However, the dilution factor in the convective envelope of red giants and supergiants is so large that it is not possible from the rare data available to make any conclusion about the presence of additional mixing (Maeder 1987; Dearborn 1992; Denissenkov 1994; El Eid 1994).
### 4.2 COMPARISONS OF MODELS AND OBSERVATIONS
MASSIVE STARS IN MS AND POST–MS PHASES
Let us first recall that from the comparison of $`t_{\mathrm{mix}}`$ and $`t_\mathrm{H}`$, it is clear that mixing processes are more efficient in more massive stars. For the intermediate mass stars of the B– and A–types, there is no global mixing currently predicted. Often, the comparisons with the observed abundance excesses for O– and B–type stars are used to adjust some efficiency factors in the models (Pinsonneault et al 1989; Weiss et al 1988; Weiss 1994; Chaboyer et al 1995ab; Heger et al 2000). Although not fully consistent, these approaches are useful to appreciate the importance of the various possible effects. The old prescriptions of Zahn (1983) were applied by Maeder (1987), Langer (1992) and Eryurt et al (1994) and they led to some surface He– and N–enrichments.
The prescriptions by Zahn (1992) were applied to the evolution of a 9 M star (Talon et al 1997). They found essentially no He–enrichment and a moderate enhancement (factor $`2`$) of N at the stellar surface, for an initial velocity of 300 km/s. Fig. 6 illustrates the changes of the N/H ratios from the ZAMS to the red supergiant stage for 20 and 25 M stars (Meynet & Maeder 2000). For non–rotating stars, the surface enrichment in nitrogen only occurs when the star reaches the red supergiant phase; there, CNO elements are dredged–up by deep convection. For rotating stars, N–excesses occur already during the MS phase and they are larger for high rotation and initial stellar masses. At the end of the MS phase, for solar metallicity Z = 0.02, the predicted excesses amount to factors 3 and 4 for initial $`v_{\mathrm{rot}}`$ = 200 and 300 km/s respectively. At lower metallicity, the N–enrichment during the MS phase is smaller, likely due to the lower mass loss; however, there is a very large increase (up to a factor of $`10`$) for late B–type supergiants, because the star spends a lot of time in the blue phase and mixing processes have time to work. The predictions of Fig. 6 are in agreement with the observed excesses for galactic B– and A–type supergiants (Venn 1995ab; Venn 1998). Also the very large excesses observed for A–type supergiants in the SMC (Venn 1998, 1999) are remarkably well accounted for.
QUESTIONS ABOUT NITROGEN
Many studies of galactic halo stars, blue compact galaxies and highly redshifted galaxies have revealed the need for initial production of primary N in addition to the current secondary nitrogen (cf. Edmunds & Pagel 1978; Matteucci 1986; Pettini et al 1995; Thuan et al 1995; Centurion et al 1998; Pilyugin 1999; Henry & Worthey 1999). The early production of N in the evolution of galaxies seems to imply that some N is produced in massive stars (Matteucci 1986; Thuan et al 1995). The problem is that the usual stellar models do not show such a production without adhoc assumptions. Models of rotating stars allow us to clearly identify the conditions for the production of primary N: the star must have an He–burning core and a thick and long–lived H–burning shell, then diffusion and transport of new <sup>12</sup>C from the core to the shell may generate some primary <sup>14</sup>N. W–R stars do not seem favourable, since the H–shell does not live long enough, being quickly extinguished and removed by mass loss. Low metallicity supergiants that have not suffered large mass loss are a very favourable site (Meynet & Maeder 2000), especially if rotation is faster at low metallicities.
### 4.3 RED GIANTS AND AGB STARS
For MS stars in the range of 1.5 to $`10`$ M, there is no evidence of extra–mixing, however there are interesting indications for red giants. The study of <sup>12</sup>C/<sup>13</sup>C in cluster red giants (Gilroy 1989) shows that stars between 2.2 M and 7 M have ratios close to the standard predictions without mixing. However, red giants below 2.2 M show <sup>12</sup>C/<sup>13</sup>C ratios much lower than the predictions, indicating some extra–mixing (Gilroy 1989; see also Harris et al 1988); the lower the mass, the higher the mixing. On the red giant branch of M67, there are indications of extra–mixing for stars brighter than $`\mathrm{log}L/L_{}1.0`$, where the first dredge–up occurs (Gilroy & Brown 1991).
From the data on M67, it appears that extra–mixing is only efficient when the H–burning shell reaches the $`\mu `$–discontinuity left by the inwards progression of the outer convective zone (Charbonnel 1994). Prior to this stage, the $`\mu `$–gradient created by the first dredge–up acts as a barrier to any mixing below the convective envelope (Charbonnel et al 1998). The $`\mu `$–gradient necessary to prevent mixing is found to be in agreement with that expected to stop the meridional circulation. For stars with $`M2.2`$ M, helium ignition occurs in non–degenerate cores, i.e. early enough so that the H–shell does not reach the border left by the outer convective zone and there is always a $`\mu `$–gradient high enough to prevent mixing (Charbonnel et al 1998). Boothroyd & Sackmann (1999) confirm that extra–mixing and the associated CNO nuclear processing (cool bottom processing, CBP) occur when the H–burning shell erases the $`\mu `$–barrier established by the first dredge–up and they predict that the effects of the CBP behave like $`M^2`$ and $`Z^1`$. Further observations of red giants with various masses are very much needed to confirm the relative absence of enrichment for masses $`2.2`$ M.
For the more advanced stages of intermediate mass stars, the critical questions concern nucleosynthesis and the processes leading to the production of the s–elements in AGB stars (Iben 1999). Rotation appears to allow the formation of larger degenerate cores (Sackmann & Weidemann 1972; Maeder 1974), then the C/O core mass is further increased during the early AGB phases (Dominguez et al 1996). The large $`\mathrm{\Omega }`$–gradients between the bottom of the convective envelope and the H–burning shell can drive mixing, mainly by the GSF instability and to a lesser extent by shear and meridional circulation between the H– and <sup>12</sup>C–rich layers during the third dredge–up in AGB stars (Langer et al 1999). The neutron production by <sup>13</sup>C($`\alpha `$,n)<sup>16</sup>O between the thermal pulses is favourable to the production of s–elements. For further studies on the role of rotation in AGB stars, the exact treatment of the instabilities in regions of steep $`\mathrm{\Omega }`$– and $`\mu `$–gradients will play a crucial role.
## 5 POST–MS EVOLUTION WITH ROTATION
The post–MS evolution of rotating stars differs from that of non–rotating stars for three main reasons : 1) the structure at He–ignition is different due to the rotationally induced mixing during the previous H–burning phase. In rotating stars, the He–cores are more massive (Sreenivasan & Wilson 1985b; Sofia et al 1994) and the radiative envelope is enriched in CNO–burning products (Maeder 1987; Heger et al 2000; Meynet & Maeder 2000). 2) The mass loss rates are increased by rotation (Friend & Abbott 1986; Langer 1998; Sect 2.6). 3) Rotational transport mechanisms may also operate in the interior during the post–MS phases. However in massive stars, the timescales for mixing and circulation, $`t_{\mathrm{mix}}`$ and $`t_{\mathrm{circ}}`$, are much larger than the evolutionary timescale by one to two orders of magnitudes during the He–burning phase (Endal & Sofia 1978) and even larger in the post He–burning phases (Heger et al 2000). Thus these processes will globally have small effects during these stages. However, due to very high angular velocity gradients occuring locally, some instabilities may appear on much smaller timescales (Endal & Sofia 1978; Deupree 1995).
### 5.1 INTERNAL EFFECTS
The fast contraction of the core and expansion of the envelope which follows the end of the MS phase produces an acceleration of $`\mathrm{\Omega }`$ in the inner regions and a slowing down in the outer layers (Kippenhahn et al 1970; Endal & Sofia 1978; Talon et al 1997; Heger et al 2000; Meynet & Maeder 2000). Typically for a 20 M model with an initial $`v_{\mathrm{rot}}=300`$ km/s (Fig. 5), the ratio of the central to surface angular velocity never exceeds 5 during the MS phase, while it increases up to 10<sup>5</sup> or 10<sup>6</sup> during the He–burning phase (Sect. 5.5).
At the beginning of the He–burning phase, the He–cores in rotating models are more massive by about 15% for $`v_{\mathrm{rot}}=300`$ km/s. The further evolution of the convective core depends on the adopted criterion for convection. If the Ledoux criterion is used, the growth of the convective core is prevented by the $`\mu `$–barriers and it remains small. Above it several small convective zones appear, each separated by semiconvective layers (e.g. Langer 1991c; Heger et al 2000). If the Schwarzschild criterion is used, the convective core simply grows in mass. For both cases, the rotational effects depend on their sensitivity to the $`\mu `$–gradients. As an example, models with the Ledoux criterion show that when rotational mixing is artificially made insensitive to $`\mu `$–gradients, the shear mixing efficiently operates in semiconvective regions and considerably enlarges the final C/O core masses (Heger et al 2000). Typically the C/O core mass increases from a value of 1.77 M in the 15 M non–rotating model to a value of 3.4 M for an initial $`v_{\mathrm{rot}}=200`$ km/s. For a similar rotating model using the Schwarzschild criterion and incorporating the inhibiting effect of the $`\mu `$–gradients, Meynet & Maeder (2000) obtain a C/O core mass of 2.9 M to be compared with the value of 2 M obtained in the non–rotating model. The treatment of the $`\mu `$–gradient is thus critical, since the C/O core masses play a key role in determining the stellar remnants as well as the chemical yields (Sect. 5.5).
### 5.2 EVOLUTION IN THE HR DIAGRAM, LIFETIMES, ROTATIONAL VELOCITIES
Rotating as well as non–rotating models with initial masses between 9 and 40 M at solar metallicity evolve towards the red supergiant (RSG) stage after the MS phase (Kippenhahn et al 1970; Sofia et al 1994; Heger 1998; Meynet & Maeder 2000). Due to the larger He–cores, the rotating stars have higher luminosities, as long as mass loss is not too large. The initial distribution of the rotational velocities will thus introduce some scatter in the luminosities of the supergiants originating from the same initial mass. The lower the sensitivity of the mixing processes to the $`\mu `$–gradients, the greater the scatter. For initial $`v_{\mathrm{rot}}`$ between 0 and 300 km/s, the difference will be of the order of 0.25 mag (Fig. 5).
In rotating stars of initial mass between 20 and 40 M, due to the large cores, the quantity of nuclear fuel is larger, but the luminosities are also higher and thus the He–burning lifetimes change slightly; as an example for initial $`v_{\mathrm{rot}}`$ = 200 – 300 km/s, the changes are less than 5%. The ratios $`t_{\mathrm{He}}/t_\mathrm{H}`$ of the He to H–burning lifetimes are not very sensitive to rotation and they remain around 10% (Heger 1998; Meynet & Maeder 2000).
THE NUMBER RATIO OF BLUE TO RED SUPERGIANTS
The variation with metallicity Z of the number of blue and red supergiants (RSG) is important in relation to the nature of the supernova progenitors in different environments (cf. Langer 1991bc) and population synthesis (e.g. Cervino & Mas-Hesse 1994; Origlia et al 1999). The observations show that the number ratio (B/R) of blue to red supergiants increases steeply with Z. Cowley et al (1979) examined the variation of the B/R ratio across the Large Magellanic Cloud (LMC) and found that it increases by a factor 1.8 when the metallicity is larger by a factor 1.2. For M<sub>bol</sub> between -7.5 and -8.5, the B/R ratio is up to 40 or more in inner Galactic regions and only about 4 in the SMC (Humphreys & McElroy 1984). A difference in the B/R ratio of an order of magnitude between the Galaxy and the Small Magellanic Cloud (SMC) was also found from cluster data (Meylan & Maeder 1982). Langer & Maeder (1995) compared different stellar models with the observations and concluded that most massive star models have problems reproducing this observed trend.
A part of this difficulty certainly arises from the fact that supergiants are often close to a neutral state between a blue and a red location in the HR diagram. Even small changes in mass loss, in convection or other mixing processes greatly affect the evolution and the balance between the red and the blue locations (Stothers & Chin 1973, 1975, 1979, 1992ab; Maeder 1981; Brunish et al 1986; Maeder & Meynet 1989; Arnett 1991; Chin & Stothers 1991; Langer 1991bc, 1992; Salasnich et al 1999). As said by Kippenhan & Weigert (1990), “the present phase is a sort of magnifying glass, revealing relentlessly the faults of calculations of earlier phases.”
The choice of the criterion for convection plays a key role, particularly when the mass loss rates are small. Models with the Ledoux criterion, with or without semiconvection, predict at low metallicity (Z between 0.002 – 0.004) both red and blue supergiants. However, when the metallicity increases, the B/R ratio decreases in contradiction with the observed trend (Stothers & Chin 1992a; Brocato & Castellani 1993; Langer & Maeder 1995). Models with the Schwarzschild criterion, with or without overshooting, can more or less reproduce the observed B/R ratio in the solar neighborhood. However, they predict no or very few red supergiants at the metallicity of the SMC, while many are observed (Brunish et al 1986; Schaller et al 1992; Bressan et al 1993; Fagotto et al 1994).
When the mass loss rates are low (i.e. at low Z), a large intermediate convective zone forms in the vicinity of the H–burning shell, homogenizing part of the star and maintaining it as a blue supergiant (Stothers & Chin 1979; Maeder 1981). For larger mass loss rates, the intermediate convective zone is drastically reduced and the formation of RSG favoured. A further increase of the mass loss rates may bring the star back to the blue. When the He–core encompasses more than some critical mass fraction $`q_c`$ of the total mass (Chiosi et al 1978; Maeder 1981), the star moves to the blue and becomes either a blue supergiant or a W–R star (e.g. Schaller et al 1992; Salasnich et al 1999; Stothers & Chin 1999). The critical mass fraction $`q_c`$ is equal to $`67\%`$ at 60 M, $`77\%`$ at 30 M and 97% at 15 M (Maeder 1981). This agrees with the investigations made for lower masses by Giannone et al (1968).
Rotation mainly affects the B/R ratio through its effect on the interior structure and on the mass loss rates. Maeder (1987), Sofia et al (1994) and Talon et al (1997) show that the effect of additional mixing due to rotational instabilities in some respect mimics that of a small amount of convective overshoot, which does not favour the formation of RSG at low Z. However fast rotation implies also higher mass loss rates by stellar winds and in general this favours the formation of RSG. In view of these two opposite effects, it is still uncertain whether rotation may solve the B/R problem. Due to the initial distribution of rotational velocities, one expects a scatter of the mass loss rates and therefore different evolutionary scenarios for a given initial mass star (Sreenivasan & Wilson 1985a). However, by producing high mass loss rates even at low metallicity, rotation may help to resolve the B/R problem.
Is rotation responsible for the observed characteristics of the blue progenitor of the SN 1987A ? The presence of the ring structures around SN 1987A (Burrows et al 1995; Meaburn et al 1995) which likely result from axisymmetric inhomogeneities in the stellar winds ejected by the progenitor (Eriguchi et al 1992; LLoyd et al 1995; Martin & Arnett 1995), and the high level of nitrogen enhancements in the circumstellar material (Fransson et al 1989; Panagia et al 1996; Lundqvist & Fransson 1996) are features which may be explained at least in part by rotation. Woosley et al (1998) suggest that any mechanism that reduces the helium core while simultaneously increasing helium in the envelope would favour a blue supernova progenitor.
EVOLUTION OF THE SURFACE VELOCITIES
As already stated in Sect. 3.2, as the star evolves from the blue toward the RSG stage, the surface velocities quickly decrease (Endal & Sofia 1979; Langer 1998). For the stars shown in Fig. 5, velocities between 20 – 50 km/s are obtained when $`\mathrm{log}T_{\mathrm{eff}}=4.0`$. Values of the order of 1 km/s are reached at the RSG stage. Observations confirm this rapid decline of the surface velocities (Rosendhal 1970; Fukuda 1982). The values at $`\mathrm{log}T_{\mathrm{eff}}=4.0`$ are in good agreement with the recent determinations of rotational velocities for galactic A–type supergiants by Verdugo et al (1999).
When the star evolves back to the blue from the RSG stage, as it is the case for the rotating 12 M model shown on Fig. 5, the rotational velocity approaches the break–up velocity (Heger & Langer 1998; Meynet & Maeder 2000). This behavior results from the stellar contraction which concentrate a large fraction of the angular momentum of the star (previously contained in the extended convective envelope of the RSG) in the outer few hundredths of a solar mass. At the maximum extension of the blue loop, the equatorial velocity at the surface of the 12 M star (Fig. 5) reaches values as high as 150 km/s. As the star evolves back towards the RSG stage, the surface velocity declines again to about 2 km/s. When the star crosses the Cepheid instability strip its surface velocity is between 10 and 20 km/s, well inside the observed range (Kraft et al 1959; Kraft 1966; Schmidt-Kaler 1982).
The increase of the surface velocity occurs every time a star leaves the Hayashi line to hotter zones of the HRD (Heger & Langer 1998). In most cases, there are observational evidences for axisymmetric circumstellar matter: disks around T Tauri stars (e.g. Guilloteau & Dutrey 1998), bi–polar planetary nebulae (e.g. Garcia-Segura et al 1999), structures around SN 1987A (e.g. Meaburn et al 1995) and rings around W–R stars (e.g. Marston 1997).
### 5.3 THE $`\mathrm{\Omega }`$, $`\mathrm{\Gamma }`$ LIMITS AND THE LBV’s
Over recent years, the problem of the very luminous stars close to the Eddington limit and reaching the break–up limit (Langer 1997, 1998) has been discussed in relation to the Luminous Blue Variables (LBV) and their origin (see Davidson et al 1989; de Jager & Nieuwenhuijzen 1992; Nota & Lamers 1997). The LBV, also called the Hubble–Sandage Variables and S Dor Variables, are extreme OB supergiants with $`\mathrm{log}L/L_{}6.0`$ and $`T_{\mathrm{eff}}`$ between about 10 000 and 30 000 K (Humphreys 1989). Only a few of them are known in the Milky Way, among which $`\eta `$ Carinae (Davidson & Humphreys 1997; Davidson et al 1997). They experience giant outbursts with shell ejections. Often they show surrounding bi–polar nebulae (Nota et al 1997). Many models and types of instabilities have been proposed to explain the LBV outbursts (de Jager & Nieuwenhuijzen 1992; Stothers & Chin 1993, 1996, 1997; Nota & Lamers 1997).
PHYSICS OF THE BREAK–UP LIMIT
The first problem concerns the expression of the break–up limit for stars close to the Eddington limit, i.e. for the brightest supergiants. When the radiation field is strong, the radiative acceleration $`\stackrel{}{g_{\mathrm{rad}}}`$ must be accounted for in the total acceleration
$$\stackrel{}{g_{\mathrm{tot}}}=\stackrel{}{g_{\mathrm{grav}}}+\stackrel{}{g_{\mathrm{rot}}}+\stackrel{}{g_{\mathrm{rad}}}=\stackrel{}{g_{\mathrm{eff}}}+\stackrel{}{g_{\mathrm{rad}}},$$
(14)
whith a modulus $`g_{\mathrm{rad}}=(\kappa L/4\pi cR^2)`$, $`R`$ being the equatorial radius. Thus, the break–up velocity obtained when $`\stackrel{}{g_{\mathrm{tot}}}=0`$ is found to be $`v_{\mathrm{crit}}^2=\frac{GM}{R}(1\mathrm{\Gamma })`$ (Langer 1997, 1998 ; Langer & Heger 1998; Lamers 1997), where $`\mathrm{\Gamma }`$ is the Eddington factor $`\mathrm{\Gamma }=\kappa L/(4\pi cGM)`$. For the most luminous stars, $`\mathrm{\Gamma }1`$ and thus the critical velocity tends towards zero. This has led Langer (1997, 1998) to conclude that for any initial rotation, the critical limit is reached before the Eddington limit. Therefore, Langer claims that one should rather speak of an $`\mathrm{\Omega }`$–limit for LBV stars rather than of a $`\mathrm{\Gamma }`$–limit.
Glatzel (1998) has suggested that the $`\mathrm{\Omega }`$–limit is an artifact due to the absence of von Zeipel’s relation in the expression of $`g_{\mathrm{rad}}`$. Indeed, with von Zeipel’s relation the radiative flux tends towards zero when the resulting gravity is zero. Thus the critical velocity is just $`v_{\mathrm{crit}}^2=\frac{GM}{R}`$, while rotation reduces (up to 40 %; Glatzel 1998) the limiting luminosity. Stothers (1999) also considers that fast rotation reduces the limiting luminosity.
For stars close to the Eddington limit, convection may develop in the outer layers (Langer 1997; Glatzel & Kiriakidis 1998). This is however not an objection to the application of the von Zeipel theorem, since most of the flux is carried by radiation at the surface. Another possible objection (Langer et al 1999) is that according to a generalization of the von Zeipel theorem by Kippenhahn (1977), the radiative flux at the equator may be reduced or increased depending on the internal rotation law. However, the deviations from von Zeipel’s theorem are negligible in the current cases of models with shellular rotation (Maeder 1999a). Thus, a study of the physical conditions, of the critical velocity and of the instabilities in rotating stars close to the Eddington limit is still very needed.
THE EVOLUTION OF LBV’s
According to the evolutionary models at high masses (Schaller et al 1992; Stothers & Chin 1996; Salasnich et al 1999), there are three possible ways for very massive stars to reach the $`\mathrm{\Omega }`$–limit in the HR diagram. 1.– Stars with very large initial mass and high rotation, especially if their $`v_{\mathrm{rot}}`$ is increased by blueward evolution during the MS phase, may reach the $`\mathrm{\Omega }`$–limit in the blue part of the HR diagram. Some fast rotators may reach the break–up limit during the overall contraction phase at the end of the MS, as shown for the 60 M (Fig. 4). If the mass loss for O–stars is mainly bipolar (Maeder 1999a), the reduction of $`v_{\mathrm{rot}}`$ during the MS phase may be smaller. For smaller mass loss rates as in lower metallicity galaxies, $`v_{\mathrm{crit}}`$ could possibly be reached earlier in evolution. The star $`\eta `$ Carinae shows evidence that the $`\mathrm{\Omega }`$–limit is reached in the blue and is likely at the end of its MS phase or beyond in view of its surface composition (Davidson et al 1986; Viotti et al 1989). 2.– After the end of the MS phase, when the star evolves redwards in the HR diagram, the value of $`\frac{\mathrm{\Omega }}{\mathrm{\Omega }_{\mathrm{crit}}}`$ becomes quite small, since the star evolves with essentially local conservation of angular momentum. Thus, rotation is less important. However, the $`\mathrm{\Gamma }`$–limit without rotation lies at a much lower luminosity there (Lamers 1997; Ulmer & Fitzpatrick 1998), so that the $`\mathrm{\Omega }`$–limit may be reached by the very massive stars during their redwards crossing of the HR diagram. 3.– When stars leave the red supergiant phase, either on blue loops or evolving towards the W–R stage, the ratio $`\frac{\mathrm{\Omega }}{\mathrm{\Omega }_{\mathrm{crit}}}`$ increases quite a lot. This is due to conservation of angular momentum in retreating convective envelopes, which contributes to strongly spin up the bluewards evolving stars (Langer 1998). Thus, the $`\mathrm{\Omega }`$–limit may also be reached from the red side. This possibility is particularly interesting since the observed CNO abundances in some nebulae around LBV stars are the same as in red supergiants, which suggests that some LBV may originate from red supergiants (Smith 1997). A similar conclusion was obtained by Waters (1997), who found evidence in some LBV nebulae of crystalline forms of silicates, with composition similar to that of red supergiants.
THE NEBULAE AROUND LBV’s: SIGNATURE OF ROTATION ?
Almost all nebulae around LBV stars show a bipolar structure (Nota et al 1995; Nota & Clampin 1997). This might be related to binarity (Damineli 1996; Damineli et al 1997). Most models invoke collisions of winds of different velocities and densities, emitted at different phases of their evolution. In some cases, an equatorial density enhancement is assumed before the outburst (Frank et al 1995; see also Nota et al 1995), while other models assume rather arbitrary non–spherical winds or a ring–like structure interacting with a previous spherical wind (Dwarkadas & Balik 1998; Frank et al 1998). The models by Garcia–Segura et al (1996, 1997) and Langer et al (1999) consider three phases in the formation of the nebula for $`\eta `$ Carinae. In both the pre– and post–outburst phases, the star has the spherical fast and low density wind typical of a blue supergiant. At the break–up limit, the star is assumed to have a slow dense wind concentrated in the equatorial plane. The bi–polar structure then arises because the shell ejected in the third phase expands more easily into the lower density at the pole. Langer et al (1999) assume that the equatorial enhancement during the outburst results from the wind compressed disk model (Bjorkman & Cassinelli 1993), which does not apply if the von Zeipel theorem is used (Owocki & Gayley 1997, 1998). Nevertheless, we note that due to the “$`\kappa `$–effect” at the break–up limit (Maeder 1999a), a strong equatorial ejection occurs quite naturally, characterized by a high density and a low velocity, as required by the above colliding wind model of Langer et al (1999).
### 5.4 ROTATION AND W–R STAR FORMATION
GENERALITIES
Recent reviews on the Wolf–Rayet (W–R) phenomenon have been presented by Abbott & Conti (1987), van der Hucht (1992), Maeder & Conti (1994) and Willis (1999). Wolf–Rayet stars are bare cores of initially massive stars (Lamers et al 1991). Their original H–rich envelope has been removed by stellar winds or through a Roche lobe overflow in a close binary system. Observationally, most W–R stars appear to originate from stars initially more massive than about 40 M (Conti et al 1983; Conti 1984; Humphreys et al 1985; Tutukov & Yungelson 1985), however a few stars may originate from initial masses as low as 15–25 M (Thé et al 1982; Schild & Maeder 1984; Hamann et al 1993; Hamann & Koesterke 1998a; Massey & Johnson 1998). The stars enter the W–R phase as WN stars, i.e. with surface abundances representative of equilibrium CNO processed material. If the peeling off proceeds deep enough the star may enter the WC phase, during which the He–burning products appear at the surface.
Many observed features are well reproduced by current stellar models. Typically good agreement is obtained between the observed and predicted values for the surface abundances of WN stars (Crowther et al 1995; Hamann & Koesterke 1998a). This indicates the general correctness of our understanding of the CNO cycle and of the relevant nuclear data (Maeder 1983), but is not a test of the model structure. For WC stars, comparisons with observed surface abundances also generally show a good agreement (Willis 1991; Maeder & Meynet 1994). In particular, the strong surface Ne–enrichments predicted by the models of WC stars have been confirmed by ISO observations (Willis et al 1997, 1998; Morris et al 1999; Dessart et al 1999).
The star number ratios W–R/O, W–R/RSG, WN/WC show a strong correlation with metallicity (Azzopardi et al 1988; Smith 1988; Maeder 1991; Maeder & Meynet 1994; Massey & Jonhson 1998). For instance, the W–R/O number ratio increases with the metallicity Z of the parent galaxy. Despite many other claims (Bertelli & Chiosi 1981, 1982; Garmany et al 1982; Armandroff & Massey 1985; Massey 1985; Massey et al 1986), the main cause is metallicity Z, which through stellar winds influences stellar evolution and thus the W–R lifetimes (Smith 1973; Maeder et al 1980; Moffat & Shara 1983). The higher the metallicity, the stronger is the mass loss by stellar winds and thus the earlier is the entry in the W–R phase for a given star; also the minimum initial mass for forming a W–R star is lowered.
REMAINING PROBLEMS WITH W–R STARS
Despite these successes, observations indicate some remaining problems: 1.– It is possible to reproduce the W–R/O and WN/WC number ratios observed in the Milky Way and in various galaxies of the Local Group, only by using models with mass loss rates enhanced by a factor of two during the MS and WNL phases (Maeder & Meynet 1994). The relative populations of WN and WC stars observed in young starburst regions are also better reproduced when models with high mass loss rates are used (Meynet 1995; Schaerer et al 1999). This is not satisfactory, since clumping in the winds of hot star tends to reduce by a factor 2 – 3 the observed mass loss rates (Nugis et al 1998; Hamann & Koesterke 1998b). 2.– The lower limit for the luminosities of WN stars (around $`\mathrm{log}L/L_{}5.0`$; Hamann & Koesterke 1998a) is fainter than predicted by standard evolutionary tracks. Massey & Johnson (1998) find that the presence of luminous red supergiants (RSG) and W–R stars is well correlated for the OB associations in M31 and M33, suggesting that some stars with mass $``$ 15 M go through both the RSG and W–R phases. 3.– For WN stars, there is a continuous transition from high H–surface abundances (0.4–0.5 in mass fraction) to hydrogen–free atmospheres, while standard models predict an abrupt transition (Langer et al 1994; Hamann & Koesterke 1998a; Fig. 7, Meynet & Maeder 2000). 4.– Smith & Maeder (1998) show that, besides the mass, a second parameter affecting the mass loss rates and terminal velocities of the wind is necessary to characterize the hydrogen–free WN stars. 5.– Standard models do not reproduce the observed number of stars in the transition WN/WC phase, characterized by spectra with both H– and He–burning products. These models predict indeed an abrupt transition from WN to WC stars, since the He–core is growing and thus building up a steep chemical discontinuity at its outer edge (e.g. Schaller et al 1992). Thus, almost no ($`<`$ 1%) stars with intermediate characteristics of WN and WC stars are predicted. However, 4 – 5% of the W–R stars are in such a transition phase (Conti & Massey 1989; van der Hucht 1999), showing that some extra–mixing is at work (Langer 1991b).
ROTATION AND THE FORMATION OF W–R STARS
Rotation may affect the formation and properties of W–R stars in several ways (Sreenivasan & Wilson 1982, 1985a; Maeder 1987; Fliegner & Langer 1995; Maeder 1999b; Meynet 1999):
1.– Surface abundances characteristic of the WNL stars may appear in a rotating star, not only as a result of the mass loss which uncovers inner layers, but also as a result of mixing in radiative zones. The same remark applies to the entry into the WC phase.
2.– Rotation may imply different evolutionary scenarios. Before becoming a W–R star, the non–rotating 60 M model at solar metallicity is likely to go through a short LBV phase after the H–exhaustion in its core. In the case of fast rotation, the star may enter the W–R phase, still burning hydrogen in its core (Maeder 1987; Fliegner & Langer 1995; Meynet 1999), thus skipping the LBV phase and spending more time in the W–R phase.
3.– Rotation favours the formation of W–R stars from lower initial mass both through its effects on the mass loss rates (Sreenivasan & Wilson 1982; Sect. 2.6) and on the mixing. Typically for the non–rotating models shown in Fig. 5, the minimum mass for W–R star formation is betwen 35–40 M. It decreases to about 25 M for initial $`v_{\mathrm{rot}}=300`$ km/s. This effect may help to explain the low luminous WN stars reported by Hamann & Koesterke (1998a). It also favours the entry into the W–R phase from the RSG stage.
4.– During the WN phase, the surface abundances are different. Indeed as a consequence of the first point, the N/C, N/O ratios obtained at the surface of the rotating WN models may have not yet reached the full nuclear equilibrium in contrast with the non–rotating case where nuclear equilibrium is reached as soon as the star enters the WN phase. The CNO ratios are however close to the equilibrium values (Fig. 7). During the transition WN/WC phase, nitrogen enhancements can be observed simultaneously with carbon and neon enhancements. After this transition phase, the <sup>22</sup>Ne enhancement reaches more or less the same high equilibrium level whatever the initial angular velocity, in agreement with the determinations of the neon abundance at the surface of WC stars (Willis et al 1997).
5.– Higher rotational velocities lead to longer W–R lifetimes. As an example for a 60 M model (Fig. 7), the W–R lifetime is increased by more than a factor 3 when rotation is included. The durations of the WN and of the transition WN/WC phases are increased. The ratio of the lifetimes of the WC to the WN phase is reduced.
6.– High rotations lead to less luminous WC stars. This is because a fast rotating star enters the W–R stage earlier in its evolution and thus begins to loose large amounts of mass early. Therefore, fast rotators enter the WC phase with a small mass and a low luminosity; the final masses are also smaller.
Thus rotation could remove or at least alleviate the above mentioned problems. The need to enhance the mass loss rates to reproduce the observed W–R/O number ratios no longer appears necessary. Rotation also implies effects which cannot be reproduced by an increase of the mass loss rate. In particular, mixing induced by rotation produces milder chemical gradients and leads to a more progressive decrease of the hydrogen abundance at the surface of WN stars (Fig. 7).
ARE W–R STARS FAST OR SLOW ROTATORS ?
Direct attempts to measure the rotational velocity of W–R stars have been performed only for a few cases: Massey (1980) and Koenigsberger (1990) obtain $`v\mathrm{sin}i500`$ km/s for WR138. However the binary nature of this object (Annuk 1991) blurs this picture, since the origin of this high velocity might be the O–type companion. The second case, WR3 with $`v\mathrm{sin}i`$ 150 – 200 km/s, looks more promising (Massey & Conti 1981) since the broadened absorption lines move in phase with the W–R emission lines (Moffat et al 1986).
There is some indirect evidence pointing towards the existence of some axisymmetric features around W–R stars (see Drissen 1992; Marchenko 1994). For instance Arnal (1992) has mapped the environment of 6 W–R stars at a frequency of 1.42 MHz and found that all the HI cavities around these have an elongated shape with a mean major–to–minor axis ratio of about 2.2. Other evidences have been found by Schulte–Ladbeck et al (1992), Miller & Chu (1993). According to Harries et al (1998), about 15% of W–R stars have anisotropic winds. They suggest that the main cause of the wind anisotropy is equatorial density enhancements produced by fast rotation rates and estimate the rotational velocities to be about 10 – 20% of the break–up velocity.
The surface velocities of W–R stars depend mainly on the initial velocity and on the amount of angular momentum lost during the previous stages. This amount will depend on the exact evolutionary sequence followed; in particular, the questions are, whether the star has passed through the RSG stage and what were the anisotropies of the stellar winds. Some other effects may also intervene, for instance the possible presence of a magnetic field (e.g. Cassinelli 1992; Sreenivasan & Wilson 1982; Sect. 3.2).
For the 60 M model shown on Fig. 7 with $`v_{\mathrm{rot}}`$ = 300 km/s, computed assuming spherically symmetric winds and no magnetic fields, the surface velocity ranges between 20 and 40 km/s, i.e. between 3 and 6% of the break–up velocity during most of the WNL phase. At the beginning of the core He–burning phase, the He–core contracts, the small H–rich envelope expands and the surface velocity reaches the break–up limit. Huge mass loss rates ensue, which eject about 3 M of material forming an anisotropic nebula with abundances characteristic of CNO–equilibrium. When the star has lost sufficient angular momentum, it drops below the break–up limit and pursues its evolution with a nearly constant surface velocity around 40 km/s. The contraction of the W–R star at the very end of the He–burning phase may again increase $`v_{\mathrm{rot}}`$, but it remains far below the break–up limit ($`\mathrm{\Omega }/\mathrm{\Omega }_{\mathrm{crit}}`$ 5%).
### 5.5 LATE STAGES, REMNANTS AND CHEMICAL YIELDS
THE POST He–BURNING PHASES
The masses of the C/O cores are larger in rotating stars, that do not evolve through a W–R phase (Sofia et al 1994; Heger et al 2000). As an example, at the end of the He–burning phase, the C/O core mass in a rotating 20 M model with an initial $`v_{\mathrm{rot}}=`$ 300 km/s is 5.7 M (Meynet & Maeder 2000). The value in a non–rotating model is 3.8 M. Thus, a rotating 20 M star will have a behavior during the late stages similar to that of a non–rotating 25 M star.
It is also interesting to notice that due to the larger He–cores, the <sup>12</sup>C($`\alpha `$,$`\gamma `$)<sup>16</sup>O reaction is more active at the end of the He–burning phase; therefore the fraction of carbon left in the C/O core decreases with respect to that in the non–rotating model (by about a factor 2.5 in the above example). This leads to an increase in the oxygen yield. Moreover since the carbon burning phase is considerably reduced, the stellar core has less time to remove its entropy through heavy neutrino losses, favouring the formation of black holes (Woosley 1986). According to Fryer (1999) the lower mass limit for black hole formation is likely lowered by rotation (see also Fryer & Heger 1999).
After central He–exhaustion, the He–shell ignites and the above layers expand leading to a decrease of the strength of the H–burning shell. The smaller thermal gradient near the H–shell favours the mixing of chemical elements there (Heger et al 2000). Protons as well as nitrogen are brought from the H–shell down into the underlaying He–rich layers. At the same time, due to the contraction of the C/O core, the temperature at the bottom of the He–shell increases and the overlying convective zone extends in mass, engulfing the region rotationally enriched in protons and nitrogen. These species then burn very rapidly, on a timescale shorter than the convective turn–over time; thus, they have nearly completely disappeared before reaching the bottom of the He–convective zone (Heger 1998). The large extension of the He–burning shell has important consequences for nucleosynthesis. This extension is larger for larger core masses, i.e. for large initial mass and/or for large rotational velocities.
When evolution proceeds further, the rotating core speeds up more and more, possibly becoming unstable with respect to nonaxial symmetric perturbations (Kippenhahn et al 1970; Ostriker & Bodenheimer 1973; Tassoul 1978). However, the results obtained by Heger et al (2000) for stars with initial $`v_{\mathrm{rot}}`$ between 200 and 300 km/s suggest that, before the core collapse, the ratio of the rotational to the potential energy is lower than required for such instabilities to occur.
ROTATIONAL PERIODS OF PULSARS
According to the models by Heger et al (2000) and Meynet & Maeder (2000), at their birth neutron stars (NS) should have rotation periods of about 0.6 ms, being nearly at the break–up rate. What is striking is that these periods are much smaller than the measured periods for young pulsars, which are around 20 – 150 ms (Marshall et al 1998). This means that the models have between $``$ 20 and 100 times more specific angular momentum than found in the young neutron stars. Various effects may be responsible for this excess rotation in the models. The efficiency of some rotationally induced mixing processes may have been underestimated or some important transport mechanism may still be missing. Kippenhahn et al (1970) speculated about the possibility for rapidly rotating dense cores to shed some mass into the envelope at its equator, in a way similar to rapidly rotating stars shedding mass into the circumstellar envelope. The equatorial mass loss by anisotropic stellar winds heavily modifies the surface boundary conditions and may remove a huge amount of angular momentum (Maeder 1999a). Other braking mechanisms, such as the removal of angular momentum from the convective core by gravity waves (Denissenkov et al 1999; Sect. 2.5) or through a magnetic field (Spruit & Phinney 1998) may also be invoked. As pointed out by Fricke & Kippenhahn (1972), the coupling of the core and envelope cannot be complete, because with solid body rotation at all time, the core would rotate too slowly (P $``$ 650 ms) to form pulsars with the observed periods.
The evacuation of the excess angular momentum could also have occured during the formation of a NS. The NS could also have been born spinning at break–up velocity and have been very efficiently slowed down during the first years. However, as discussed by Hardorp (1974), there are arguments suggesting that NS have never been near break–up (Ruderman 1972), and the study of the Crab pulsar supports this view (Trimble & Rees 1970). Indeed in this case, the release of such an important amount of rotational energy, if not emitted in the form of a $`\gamma `$–ray burst or through gravitational waves, would have shown up in the expansion energy of the nebula and in the optical light during historical times, which is not reported. The original rotation period of the Crab pulsar at birth is estimated to be 5 ms, still one order of magnitude from the break–up period (Hardorp 1974). Therefore, the stellar core must have been spinning slowly before its collapse.
If such a large angular momentum is embarrassing at present for explaining the observed rotating periods of young pulsars, it may give some support to the “collapsar” model proposed by Woosley (1993; MacFadyen & Woosley 1999) for the $`\gamma `$–ray bursts. A “collapsar” is a black hole formed by the incomplete explosion of a rapidly rotating massive star. The rapid rotation is necessary to allow the formation of an accretion disk outside the black hole. The accretion disk efficiently transforms the gravitational binding energy into heat which can then power a highly relativistic jet. The burst and its afterglow in various wavelengths is attributed to the jet and its interactions with the external medium. The models by Heger et al (2000) and Meynet & Maeder (2000) have enough angular momentum to support matter in a stable disk outside a black hole and thus could offer interesting progenitors for this kind of evolution, if it exists.
THE CHEMICAL YIELDS
Rotation affects the chemical yields in many ways. The larger He–cores obtained in rotating models at core collapse imply larger production of helium and other $`\alpha `$-nuclei elements (Heger 1998). This is by far the most important effect of rotation on the chemical yields. Also, by enhancing the mass loss rates and by making the formation of W–R stars easier, rotation favours the enrichment of the interstellar medium by stellar winds (Maeder 1992). Indeed, the stronger the winds, the richer the ejecta in helium and carbon and the lower in oxygen.
The rotational diffusion during the H–burning phase enriches the outer layers in CNO processed elements (Maeder 1987; Fliegner & Langer 1995; Meynet 1998; Heger 1998). Some <sup>14</sup>N is extracted from the core and saved from further destruction. The same can be said for <sup>17</sup>O and <sup>26</sup>Al, a radioisotope with a half–life of 0.72 Myr. The mixing in the envelope of rotating stars also leads to a faster depletion of the temperature sensistive light isotopes, for instance lithium and boron (Fliegner et al 1996).
The presence of <sup>26</sup>Al in the interstellar medium is responsible for the diffuse galactic emission observed at 1.8 MeV (e.g. Oberlack et al 1996). If the nucleosynthetic sites of this element appear to be the massive stars (Prantzos & Diehl 1996), it is still not yet clear how the production is shared between the supernovae and the W–R stars and how it is affected by rotation and binarity. For stellar masses between 12 and 15 M, the lifetimes are much longer than that of <sup>26</sup>Al and therefore most of the <sup>26</sup>Al produced during central H–burning and partially mixed in the envelope has decayed at the time of the supernova explosion. Thus, for this mass range, rotation does not seem to bring important changes (Heger 1998). However, when the star is massive enough to go through the W–R phase, the stellar winds may remove <sup>26</sup>Al–enriched layers at a much earlier stage. In that case, rotation may substantially increase the quantity of <sup>26</sup>Al injected in the interstellar medium (Langer et al 1995).
The convective zone associated with the He–shell in rotating models transports H–burning products to the He–burning shell (Heger 1998). The injection of protons and nitrogen into a He–burning zone opens new channels of nucleosynthesis (Jorissen & Arnould 1989). In particular, this enhances the s–process and the formation of <sup>14</sup>C, <sup>18</sup>O and <sup>19</sup>F. Since these elements are produced just before the core collapse they can survive until the supernova explosion. The injection of protons into a He–burning zone may also be responsible for primary <sup>14</sup>N production through <sup>12</sup>C(p,$`\gamma )^{13}`$N$`(\beta ^+)^{13}`$C(p,$`\gamma )^{14}`$N, but this primary nitrogen is rapidly destroyed to produce <sup>18</sup>O. As discussed in Sect. 4.2, stars with an He–burning core and a thick and long–lived H–burning shell seem to be more favourable sites for a primary <sup>14</sup>N production. Another important effect of the growth of the He–shell in these late stages is the production shortly before core–collapse of some <sup>15</sup>N. In non–rotating models, this element is destroyed, while in rotating ones it is synthesized (Heger 1998). The very low <sup>14</sup>N/<sup>15</sup>N ratios measured in star–forming regions of the LMC and in the core of the (post–)starburst galaxy NGC 4945 (Chin et al 1999) support an origin of <sup>15</sup>N in massive stars.
## 6 PERSPECTIVES
We hope to have shown that rotation is unavoidable for a proper understanding and modelling of the evolution for the Upper MS stars.
The way to further progress goes through more studies of the physical effects of rotation, in particular of the various instabilities which can produce mixing of the elements and transport of angular momentum, both in the early and advanced phases of evolution. The influence of rotation on mass loss is also critical. In this respect, the shape and composition of the asymmetric nebulae observed around many massive stars provide interesting constraints on the models of rotating stars.
ACKNOWLEDGEMENTS: We express our best thanks to
Dr. Laura Fullton for her most useful advices on the manuscript.
|
warning/0004/hep-th0004059.html
|
ar5iv
|
text
|
# Fate of the classical false vacuum
## I Introduction
The reach of equilibrium from a metastable state covers a large number of interesting effects from instabilities observed in the mixed phase of first order phase transitions to the inflation in the early Universe . The final state is reached in an irreversible process, the description of this relaxation relies on out-of-equilibrium dynamics of field theories.
The decay of metastable states is usually discussed in the framework of the nucleation scenario . It has been implemented in the form of saddle point expansions in classical , and quantum systems . This large amplitude instability is the first of the possible instabilities, suggested by mean field analysis, which consists of the creation of a bubble of the true vacuum larger than the critical size, embedded into the false one.
Another possibility, the instability against fluctuations with infinitesimal amplitude leads to the spinodal phase separation. A recent observation made it clear that soft fluctuations of these inhomogeneous unstable modes generate in equilibrium the Maxwell construction by their tree-level contribution to the renormalization group flow . The fluctuation induced flatness of the effective potential suggests the dominance of spinodal phase separation in equilibrium.
Since the type of instability one observes, might depend essentially on the time scale of the observation, a detailed investigation of the time evolution can separate the effects of the two kinds of instabilities. This is made possible by large scale computer simulations of the thermalisation process in closed systems.
Another question, left open by the Maxwell construction in equilibrium , concerns the structure of the vacuum with spontaneous symmetry breaking. In fact, the effective potential is related to the probability distribution of the order parameter and the Maxwell-cut applied to the former suggests that the latter is also degenerate in the mixed phase. Either we accept that the vacua with spontaneously broken symmetry correspond overwhelmingly to the mixed phase or a dynamical mechanism is seeked to eliminate the mixed phase from among the final states of the time evolution.
In cosmology different slow-roll scenarios of inflation are being considered. Recent studies of the dynamics of inflaton fields with large number of components (large $`N`$ limit) displayed for a first time a dynamical version of the Maxwell construction . The Hartree type solution of the quantum dynamical equations leads to the conclusion that the order parameter might get rest with finite probability at any value smaller than the position of the stable minimum of the tree level effective potential, corresponding to the stabilisation of a mixed state.
Detailed investigations of the thermalisation phenomenon were performed also in noisy-damped systems, coupled to an external heat bath . The relaxation to thermal equilibrium of the space averaged scalar field (the order parameter) starting from metastable initial values has been thoroughly investigated. In these simulations the damping coefficient is treated as an external control parameter. Using the numerical solution of the corresponding Langevin-equations the validity range of the analytical results for the homogenous nucleation mechanism has been explored.
In this paper we focus on an alternative description of the decay process of the metastable vacuum state. The process is described exclusively in terms of the homogenous order parameter (OP) mode. The evolution of the OP is studied in interaction with the rest of the system as described by the reversible dynamical equations of motion of the full system. Careful analysis of its dynamics allows us to explore the effects of both kinds of the above mentioned instabilities. The transition of the order parameter from the metastable to the stable vacuum is induced by a homogenous external “magnetic” field, whose strength is systematically reduced. No random noise is introduced to represent any external heat bath, the friction coefficient of the effective order parameter dynamics is determined internally.
Our results offer a ”dualistic” resolution of the competition between the nucleation and the spinodal phase separation mechanisms in establishing the true equilibrium. On the one hand, we find that the statistical features of the decay of the false vacuum agree with the results obtained by expanding around the critical bubble. The microscopic mapping of the field configurations during the relaxation supports the bubble creation scenario. Alternatively, the effective OP-theory displays the presence of soft modes and produces dynamically a Maxwell-cut when the time dependence of the transition trajectory is described in the effective OP theory. We find that the larger is the system the smaller is the external field which is able to produce the instability. For infinite systems an infinitesimal field pushes the system through the Maxwell-cut, where no force is experienced by the OP. Therefore it will not stop before reaching the true homogenous ground state passing by the mixed states with constant velocity.
Our model, a spacelike lattice regulated classical scalar $`\mathrm{\Phi }^4`$ field theory in its broken symmetry phase is introduced in Section 2. In Section 3 we describe the evolution of the system starting from order parameter values near a metastable point which relaxes first to this state under the combined effect of parametric resonances and spinodal instabilities. The second stage of the transition to the stable ground state is the actual focus of our paper.
Characteristic intervals of the observed order parameter trajectory are reinterpreted as being the solutions of some effective point-particle equation of motion, which displays dissipation effects explicitly. Our approach can be understood also as the real-time version of the lowest mode approximation used for the estimation of finite size dependences in Euclidean field theory . In this sense our approach can be considered also as the numerical implementation of a real time renormalisation group strategy.
One of our principal goals is to reconstruct the thermodynamics of the classical ”OP-ensemble” on the (meta)stable branches of the OP-trajectory (Section 4). Its dissipative dynamical equations near equilibria will be established. On the transition trajectory we shall elaborate on the presence of the Maxwell construction in the effective dynamics describing the motion after nucleation (Section 5). The statistical aspects of the approach to the equilibrium are established for reference and comparison in an Appendix. The conclusions of this investigation are summarised in Section 6.
The results of this study can be usefully compared with classical investigations of metastability and nucleation in the kinetic Ising model . This system has first order dissipative dynamics by its definition. Still several relaxation features of the kinetic Ising model are comparable to our findings, since the “numerical experiment” performed in the two models are essentially the same. Especially, the relaxation function of is simply related to the order parameter we focus our attention. In both cases in the mechanism of the bubble growth the aggregation of spontanously generated small size regions of the true ground state to its surface plays important role.
## II Classical cut-off $`\mathrm{\Phi }^4`$-theory on lattice
The energy functional of a classical system in a two-dimensional box of size $`L_d`$ coupled to an external magnetic field of strength $`h_d`$ is of the following form:
$$E_d=d^2x_d\left[\frac{1}{2}\left(\frac{d\mathrm{\Phi }_d}{dt_d}\right)^2+\frac{1}{2}\left(_d\mathrm{\Phi }_d\right)^2+\frac{1}{2}m^2\mathrm{\Phi }_d^2+\frac{1}{24}\lambda \mathrm{\Phi }_d^4h_d\mathrm{\Phi }_d\right].$$
(1)
The index $`d`$ is introduced to distinguish the dimensionfull quantities from the dimensionless ones, defined by the relations (for $`m^2<0`$):
$`t=t_d|m|,x=x_d|m|,`$ (2)
$`\mathrm{\Phi }=\sqrt{{\displaystyle \frac{\lambda }{6}}}{\displaystyle \frac{1}{|m|}}\mathrm{\Phi }_d,h=\sqrt{{\displaystyle \frac{\lambda }{6}}}{\displaystyle \frac{1}{|m|^3}}h_d.`$ (3)
For the spatial discretisation one introduces a lattice of size
$$L_d=Na_d=Na\frac{1}{|m|}.$$
(4)
The energy functional of the lattice system can be written as
$`E{\displaystyle \frac{\lambda }{6|m|^2}}E_d={\displaystyle \frac{a^2}{a_t^2}}{\displaystyle \underset{𝐧}{}}[{\displaystyle \frac{1}{2}}(\mathrm{\Phi }_𝐧(t)\mathrm{\Phi }_𝐧(ta_t))^2+{\displaystyle \frac{a_t^2}{2a^2}}{\displaystyle \underset{𝐢}{}}(\mathrm{\Phi }_{𝐧+\widehat{𝐢}}\mathrm{\Phi }_𝐧)^2`$ (5)
$`{\displaystyle \frac{a_t^2}{2}}\mathrm{\Phi }_𝐧^2+{\displaystyle \frac{a_t^2}{4}}\mathrm{\Phi }_𝐧^4a_t^2h\mathrm{\Phi }_𝐧].`$ (6)
(Here we have introduced the dimensionless time-step $`a_t`$, which should be chosen much smaller than $`a`$, and $`𝐧`$ denotes the lattice site vectors.) The equation of motion to be solved numerically is the following:
$`\mathrm{\Phi }_𝐧(t+a_t)+\mathrm{\Phi }_𝐧(ta_t)2\mathrm{\Phi }_𝐧(t){\displaystyle \frac{a_t^2}{a^2}}{\displaystyle \underset{i}{}}(\mathrm{\Phi }_{𝐧+\widehat{𝐢}}(t)+\mathrm{\Phi }_{𝐧\widehat{𝐢}}(t)2\mathrm{\Phi }_𝐧(t))`$ (7)
$`+a_t^2(\mathrm{\Phi }_𝐧+\mathrm{\Phi }_𝐧^3h)=0.`$ (8)
The initial conditions for Eq.(8) were chosen as
$$\dot{\mathrm{\Phi }}_𝐧(t=0)=0,\mathrm{\Phi }_𝐧(t=0)=\mathrm{\Phi }_0+\xi \mathrm{\Phi }_1.$$
(9)
The random variable $`\xi `$ is distributed evenly on the interval $`(1/2,1/2)`$. Therefore the starting OP-value is $`\mathrm{\Phi }_0`$. The energy density $`E/Na^2`$ is controlled through the magnitude of $`\mathrm{\Phi }_1`$. In this study we have chosen $`\mathrm{\Phi }_0=0.815`$ and $`\mathrm{\Phi }_1=4/\sqrt{6}`$. The latter corresponds to a temperature value $`T_i=0.57`$ in the metastable equilibrium (from Eq.(14)). This value is much below the critical temperature of the system ($`T_c1.5T_i`$). It has been checked that at this energy density all other choices of $`\mathrm{\Phi }_0>0`$, for fixed $`h`$, find a unique metastable equilibrium.
Eq.(8) was solved with $`a=1`$ and with typical $`a_t`$ values in the range (0.01-0.09). It has been checked that the statistical characteristics of the time evolution is not sensitive to the variation of $`a_t`$, though the “release” time (the moment of the transition from metastability to the true ground state) in any single run with given initial conditions might change considerably under the variation of $`a_t/a`$. Three lattice sizes were systematically explored: $`N=64,128,256`$. Several single runs were performed also for $`N=512`$ and $`N=1000`$ with the aim to analyze in more detail some self-averaging physical quantities on different portions of the trajectory under the assumption of the ergodicity of the system. The magnetic field $`h`$ inducing the transition was varied in the range $`h(0,0.08)/\sqrt{6}`$. For the reconstruction of the effective potential felt by the OP also positive values were chosen up to $`h=0.5/\sqrt{6}`$. The smaller the value of $`|h|`$ was fixed, the longer the “release” times have grown on the average. For this reason also the runs were prolongated with decreasing $`|h|`$, and for the smallest $`|h|`$ the length of a run reached up to $`(10^610^7)|m|^1`$ until the transition took place.
For a careful comparison of our transition statistics with the generally used statistical nucleation theory, and also for understanding the systematics of its change when $`h`$ has been diminished a large number of ($`h`$, $`N`$) pairs were used in this analysis. Altogether 24 422 transition events have been recorded (for $`N=64`$: 16 908, $`N=128`$: 2 903, $`N=256`$: 4 411). For the largest systems at the smallest $`h`$ the event rate was 1-2/day/ 400 MHz-processor.
## III Time-history of the order parameter
A typical OP-history is displayed in Fig.1. In the same figure we show also the history of the OP mean square (MS)-fluctuation ($`\mathrm{\Phi }^2\mathrm{\Phi }^2`$) and of its third moment ($`(\mathrm{\Phi }\mathrm{\Phi })^3`$). The evolution of the non-zero $`𝐤`$ modes is demonstrated in Fig.2, where the averaged kinetic energy content of the $`|𝐤|<2.5`$ and $`|𝐤|>2.5`$ regions is followed. Although the separation value is somewhat arbitrary, namely it divides into two nearly equal groups the spatial frequencies available in the lattice system, the figure demonstrates the most important features of the evolution of the power in the low-$`|𝐤|`$ and high-$`|𝐤|`$ modes.
In general, five qualitatively distinct parts of the trajectory can be distinguished, although some of the first three might be missing for some initial configurations and/or magnetic field strengths.
The OP-motion usually starts with large amplitude damped oscillations. The “white noise” initial condition of Eq.(9) corresponds to a $`𝐤`$-independent Fourier amplitude distribution, therefore the initial distribution of the kinetic energy is $`\omega ^2(|𝐤|)`$. During this period, in the power spectrum of the kinetic energy, first a single sharp peak shows up at a resonating $`|𝐤|`$-value ($`|𝐤|1.5`$), which breaks up into several peaks ($`|𝐤|<1`$) at later times due to the non-linear interaction of the modes. At the end of the first period the whole $`|𝐤|<1`$ range gets increased power, the $`|𝐤|>1.5`$ part of the power spectrum does not seem to change.
Next a slow, almost linear (modulated) decrease of the OP follows. At the same time its MS fluctuation increases linearly. In Fig.2 an energy flow towards the low-$`|𝐤|`$ part can be observed, which proceeds through the excitation of single modes in this part. As a result the minimum of the effective potential is continously shifted to smaller $`\mathrm{\Phi }`$-values, as if the temperature would gradually increase (see the left part of Fig.1). The full (microscopic) kinetic energy density shows in this period less than 5% variation. In view of the picture based on the Maxwell construction it is rather surprising that independently of the initial conditions the OP converges towards a well-defined absolute value, depending only on the total energy density.
On the third portion the average value of the OP and its moments stay constant (see the right picture in Fig.1). The average energy content of the low and high-$`|𝐤|`$ part of the spectra is nearly the same. This suggests the establishment of a sort of thermal (meta)equilibrium.
In terms of the terminology introduced for the inflation, the first period leading to this (quasi) -stationary state can be called preheating, and the second reheating. Directly before the moment of the transition to the true vacuum a peak appears in the power spectrum in the narrow neighbourhood of $`𝐤=0`$ with varying position in time. On the snapshots of the real space configurations a set of randomly distributed small bubbles of the true ground state appear with a radius increasing in time, until one of the bubbles exceeds the critical size.
The fourth portion of the motion is the transition itself. The value of the OP MS-fluctuation increases by about a factor of three and the temporal width of this transient increase measures very well the transition time. The transition time decreases on larger lattices, the height of the jump in the OP-fluctuation is not sensitive to the lattice size. Also the third reduced moment shows a characteristic variation. An increase of the temperature proceeds smoothly during the transition of the OP to its stable value (Fig.2). The slight separation of the two curves in Fig.2 gives a feeling on the degree of uniformity of the temperature variation of the different modes.
The last portion of the trajectory represents stable (thermalised) oscillations around the true ground state. Here a complete equilibration of the power spectrum can be observed (see Fig.2) corresponding to a somewhat increased temperature $`(T0.6)`$.
## IV Motion near the (meta)stable point
Our analysis of the motion around the (meta)stable value of the order parameter explores the consequences of assuming the ergodicity hypothesis for a sufficiently long finite time interval, after the system has already reached the equilibrium. The equilibrium is characterised by a limiting probability density in the configuration space. The averaging with this density should provide the same value as the one yielded by a single long time evolution when subsequent configurations are used in constructing the statistics. Our “statistical system” is now a single degree of freedom, the order parameter of the lattice system, interacting with all other ($`k0`$) modes.
The effective equation of motion can be thought to result from the application of a “molecular dynamical renormalization group”, to our microscopical equations. The blocking in space is performed by projecting the field configuration $`\mathrm{\Phi }(𝐱,t)`$ on the OP $`\mathrm{\Phi }(t)`$. It represents the infrared (IR) end point of such a blocking whose effective theory is now reconstructed from the actual time dependence found numerically. Combining ergodicity of the full system with the renormalization group (RG) concept we arrive to the conclusion that ensemble averages of any OP-function coincide with time averages of the same function.
In view of the RG concept we look for an effective equation of motion where the value of OP is determined exclusively by its values preceding in time. Assuming the absence of long memory effects a “gradient” expansion in time can be envisaged, leading to a local differential equation. We introduce into this phenomenological “Newton-type” equation a term violating time-reversal invariance. We are not able to derive it from the original system, we only wish to test its presence. The sustained motion of OP, however, requires, in this case the presence of a random “force” term, too. The deterministic part of the “force” is expected to be related to the equilibrium effective potential, since this object determines the stationary probability distribution for the OP near equilibrium.
The above considerations lead us to write down the equation of motion, which represents a linear relation between the acceleration and the velocity of the order parameter:
$$\ddot{\mathrm{\Phi }}_d+\eta _d(\mathrm{\Phi }_d)\dot{\mathrm{\Phi }}_dh_d+\frac{dV_{\text{eff}}(\mathrm{\Phi }_d)}{d\mathrm{\Phi }_d}=\zeta _d,$$
(10)
where $`\zeta _d`$ is a noise term. In the corresponding dimensionless equation of motion the following new rescaled quantities will appear:
$$\eta _d=|m|\eta ,\zeta _d=\zeta |m|^3\sqrt{\frac{6}{\lambda }}.$$
(11)
The fitting procedure for the coefficients on the left hand side of Eq. (10) was the following. For a given interval of time $`I_t`$ the region of the order parameter space visited by the system $`\left\{\mathrm{\Phi }(t)\right|tI_t\}`$ was divided into small bins. Having defined the time set $`T_{\mathrm{\Phi }_b}=\left\{tI_t\right|\mathrm{\Phi }_b<\mathrm{\Phi }(t)<\mathrm{\Phi }_b+\mathrm{\Delta }\mathrm{\Phi }\}`$ corresponding to a given bin, the linear relation $`\ddot{\mathrm{\Phi }}(t_b)=\eta (\mathrm{\Phi }_b)\dot{\mathrm{\Phi }}(t_b)f(\mathrm{\Phi }_b)`$ was fitted with the method of least squares using the $`\dot{\mathrm{\Phi }}(t_b)`$, $`\ddot{\mathrm{\Phi }}(t_b)`$ data measured at time moments $`t_b`$ belonging to $`T_{\mathrm{\Phi }_b}`$. We have obtained in this way the coefficient functions $`\eta (\mathrm{\Phi }_b),f(\mathrm{\Phi }_b)`$. Once the functions $`\eta (\mathrm{\Phi })`$ and $`f(\mathrm{\Phi })h+V_{\text{eff}}^{}(\mathrm{\Phi })`$ are determined, we evaluate for each time $`t`$ the expression $`\ddot{\mathrm{\Phi }}(t)+\eta (\mathrm{\Phi }(t))\dot{\mathrm{\Phi }}(t)+f(\mathrm{\Phi }(t))`$. Its actual value determines the random noise function $`\zeta (t)`$, whose statistical features (autocorrelation) should be extracted from the data.
The effective force $`f(\mathrm{\Phi })`$ calculated from the time-average of the oscillatory motion around the equilibrium, is expected to agree with the force coming from the theoretically determined finite temperature effective potential calculated perturbatively in the cut-off two-dimensional field theory for some appropriately chosen value of the temperature . With one-loop accuracy the expected equality reads:
$$f(\mathrm{\Phi })_{measured}=h+\frac{d}{d\mathrm{\Phi }}\left[V(\mathrm{\Phi })+\frac{T}{8\pi }V^{^{\prime \prime }}(\mathrm{\Phi })\left(1+\mathrm{log}\frac{\mathrm{\Lambda }^2}{V^{^{\prime \prime }}(\mathrm{\Phi })}\right)\right]_{theory}+𝒪(T^2).$$
(12)
The expressions on the right hand side of this equation are connected to the dimensionfull quantities of the original one-loop computation in the following way:
$$T=\frac{\lambda }{6|m|^2}T_d,V(\mathrm{\Phi })=\frac{1}{2}\mathrm{\Phi }^2+\frac{1}{4}\mathrm{\Phi }^4=\frac{\lambda }{6|m|^2}V_d,\mathrm{\Lambda }=\frac{\pi }{a}.$$
(13)
The dimensionless temperature is defined by the time-average of the kinetic energy based on the assumption that in the effective theory of the OP it has the usual expression in terms of $`\dot{\mathrm{\Phi }}(t)`$:
$$\underset{t\mathrm{}}{lim}\frac{1}{t}_0^t𝑑t^{}\frac{1}{2}(\dot{\mathrm{\Phi }}(t^{},x))^2\frac{1}{2}\overline{\dot{\mathrm{\Phi }}^2}^t=\frac{T}{2}.$$
(14)
One should note that this definition in the dimensionfull version implies a “Boltzmann-constant”, $`|m|^2`$ multiplying $`T_d`$.
By measuring the order parameter average for different values of the external field $`h`$ we obtain the magnetisation curve of the system. In the numerical work we restarted the computation at different values of the external magnetic field but one might as well change $`h`$ adiabatically and measure the force law at each (quasi)equilibrium point. The results should agree in the stable regime, up to the phenomenon of the hysteresis. The resulting curve can be viewed as the numerical Legendre transformation by identifying the external source $`h`$ with the derivative of the effective potential,
$$h+V_{\text{eff}}^{}(\mathrm{\Phi }_{h,measured})=0.$$
(15)
It was found that $`f(\mathrm{\Phi })_{measured}`$ is always vanishing at $`\mathrm{\Phi }=\mathrm{\Phi }_{h,measured}`$, thus the force acting on the order parameter at the equilibrium position $`\mathrm{\Phi }_h`$ induced by the external source $`h`$ is indeed always $`V_{eff}^{}(\mathrm{\Phi }_h)`$.
The two sides of the relation (12) are shown in Fig.3. To the values of the external field used in preparing Fig.3 ($`h\sqrt{6}=0.5,0.02,0,0.5,1,2)`$ single runs were selected by the requirement that the system stayed in the metastable vacuum up to the time $`10^6`$. The force shown in Fig.3 demonstrates that not only the equilibrium positions but also the fluctuations of the OP are governed by the static effective potential. Similar measurements, performed on $`100\times 100`$ lattices showed no finite size effects in the stable regime, $`h>0`$. The error of $`f(\mathrm{\Phi })`$ is not shown in the Figure, since the typical values ($`0.004`$ for $`T=1/3`$ and $`0.002`$ for $`T=1/30`$) are too small to be displayed.
Another piece in the effective equation of motion (10) is the friction term whose presence indicates the dynamical breakdown of the time inversion symmetry. The friction coefficient $`\eta (\mathrm{\Phi })`$ proves clearly non-vanishing and shows only weak dependence on the actual value of the OP around its equilibrium position. The breakdown of the time-reflection symmetry in a closed system must arise only in presence of infinitely many degrees of freedom. Till then only statements on the Poincaré-time can be made. Thus our non-zero results for $`\eta `$ require further clarifications.
The point is that there are two types of infinities, controlled by the temporal UV and the IR cutoffs, respectively. The spontaneous symmetry breaking is driven by the IR modes, and the non-trivial minima of the potential energy arise from the presence of infinitely many degrees of freedom in the IR (thermodynamical limit). On the contrary, the dynamical symmetry breaking is the result of the effects of the derivative terms in the action and the infinitely many UV modes (continuum limit). The breakdown of the time inversion, being related to a time-derivative term in the effective equation of motion, should come from the UV, the short time behaviour of the system.
In fact, one expects no friction when the UV cutoff, $`a_t`$ is so small, that not enough energy can be dissipated during such a short time. In quantitative terms one should have
$$a_t>\tau =\frac{2\pi }{\sqrt{\frac{8}{a^2}+2}}+𝒪(T),$$
(16)
where $`\tau `$ is the time scale of the fastest mode in the system. The right hand side relation of Eq.(16) gives an estimate of the maximal frequency from the free dispersion relation $`p_0^2=4(\mathrm{sin}^2p_xa/2+\mathrm{sin}^2p_ya/2)/a^2+M^2(T)`$ of the lattice hamiltonian system for fixed $`a`$ in the limit $`a_t0`$. The fast modes absorb energy from the OP in a single time step for $`a_t>\tau `$, and the friction term should appear in the effective equation of motion.
With help of the temporal blocking
$$\mathrm{\Phi }(t)\mathrm{\Phi }_n(t)=\frac{1}{n}\underset{k=n/2}{\overset{n/21}{}}\mathrm{\Phi }(t+ka_t)$$
(17)
one can construct the trajectories $`\mathrm{\Phi }_n(t)`$ corresponding to larger values of the time cut-off, $`b_t=na_t`$. Such blocking was performed in time up to $`n=2000`$ and a discrete time equation of motion of the form Eq.(10) was reconstructed for the blocked trajectory $`\mathrm{\Phi }_n(t)`$. The non-trivial dependence of $`\eta `$ on the temporal cutoff $`a_t`$ is shown in Fig. 4.
One can distinguish two regimes separated by a crossover apparently independent of the lattice size, located at $`b_t0.2\tau `$. A scaling behavior is observed on the UV side, where $`\eta (b_t)`$ tends to zero with a critical exponent close to 1 and the force being constant. This result should be independent of the actual blocking details. In the other regime, on the IR side $`\eta `$ goes over first into a $`b_t`$-independent regime, corresponding to the saturation of the energy transfer from the OP and giving a stable, microscopic definition of the friction coefficient. Finally, in the far IR part a qualitatively different oscillatory behaviour sets in.
The location of the crossover from the UV scaling regime to the plateau can be understood by writing the fluctuation dissipation theorem in the corresponding discretised form: $`\zeta ^2=2\eta T/b_t`$. The linearly increasing regime of $`\eta (b_t)`$ implies constant second moment for the noise. When $`\eta (b_t)`$ reaches the plateau the second moment of the noise decreases like $`1/b_t`$. The crossover therefore is located at the autocorrelation time scale of the noise.
A qualitative interpretation of the oscillatory IR regime can be based on the observed small amplitude beating phenomenon in the OP trajectory. This can be recognized by closer naked eye inspection of the left side of Fig.1, which persists further also on the right side. It is reflected in the OP autocorrelation function, too, since in the course of the blocking an interference effect occurs on the right hand side of Eq.(17) due to this regularity. This feature is relevant to the value of $`\eta (b_t)`$, responsible for the decay of all kinds of fluctuations.
The appareance of peaks in $`\eta (b_t)`$ at both the maximal destructive and constructive interferences can be modeled semi-quantitatively by identifying the beating part of the OP-motion $`\text{Re}\delta \mathrm{\Phi }=\text{Re}\delta \mathrm{\Phi }_0\mathrm{exp}(i\omega t)`$ with the stationary solution of a single weakly damped driven harmonic oscillator, $`\ddot{\delta \mathrm{\Phi }}+\eta \dot{\delta \mathrm{\Phi }}+\omega _0^2\delta \mathrm{\Phi }=f_0\mathrm{exp}(i\omega t)`$. The blocking acts on the trajectory $`\delta \mathrm{\Phi }(t)`$ as $`\delta \mathrm{\Phi }(t)u\delta \mathrm{\Phi }(t)`$, where $`u=(\mathrm{exp}(i\omega b_t)1)/i\omega b_t`$. It does not change the relative phase of the driving force and of the blocked oscillation amplitude, leading to a relation between the parameters of the original and the blocked equation of motion:
$$\omega ^2+i\eta \omega +\omega _0^2=\omega ^2u^2+i\overline{\eta }\omega u+\overline{\omega }_0^2.$$
(18)
The friction coefficient for the blocked trajectory turns out to be $`\overline{\eta }=(\eta +\omega \text{Im}(u^2))/\text{Re}(u)`$. It is easy to see that $`\text{Re}(u)`$ is vanishing at maximal constructive and destructive interferences. This provides singularities in $`\overline{\eta }`$. Whenever $`\text{Re}(u)=0`$ we have $`\text{Im}(u^2)=0`$ and numerator changes sign in the vicinity of the singularity. Thus $`\overline{\eta }>0`$ apart for a short time interval around the singularities where the non-harmonic features should stabilise the fluctuations and keep $`\overline{\eta }>0`$, as observed in our simulation.
The $`a`$ dependence appearing in Fig.4 arises from the following two effects. One is that for larger $`a`$ the maximal oscillation frequency is smaller and the time resolution of the system becomes cruder. Another is that larger $`a`$ represent bigger physical volume, many more soft modes and less harmonic system, which tends to invalidate the simple picture based on a single harmonic mode. As a result the effects of oscillatory nature will be smeared, as one clearly recognizes in the figure.
The quality of any proposed deterministic equation of motion (e.g. equations similar to Eq.(10) with zero on the right hand side), can be judged by the amplitude and the autocorrelation of its error term, the noise term $`\zeta `$. The amplitude of the noise was found at least two-three order of magnitude below the average level of the force as fitted to Eq. (10). The autocorrelation function of the noise of Eq. (10) appeared to be local, approximately of the form $`\zeta _d(t)\zeta _d(t^{})\delta ^{\prime \prime }(tt^{})`$. The status of the fluctuation-dissipation theorem will be investigated for more complex field theoretical systems in future investigations.
The selfconsistency of the definition of the temperature in Eq.(14) and also the establishment of the thermal equilibrium can be tested further by plotting histograms for the following quantities:
$`E_k`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left({\displaystyle \frac{d\mathrm{\Phi }}{dt}}\right)^2,`$ (19)
$`E_p`$ $`=`$ $`{\displaystyle \frac{\mu ^2}{2}}(\mathrm{\Phi }\mathrm{\Phi }_h)^2,`$ (20)
$`E_t`$ $`=`$ $`E_k+E_p,`$ (21)
where $`\mu ^2`$ is the slope of the force as the function of the order parameter, determined numerically. Typical results for the energy-histograms are shown in Fig.5. It shows perfect agreement of all slopes and good agreement with the expectations based on equipartition of the energy between the kinetic and the potential parts. We have checked for several temperatures, that the temperature determined by these histograms and from the $`|𝐤|`$-spectra of the kinetic energy agree. This piece of information is parallel to the recent detailed investigations of the thermalisation in $`(1+1)`$-dimensional $`\mathrm{\Phi }^4`$-theories .
It is worthwhile noting that the relaxation time from a given initial condition to the thermally distributed state like in this figure takes at least one order of magnitude longer time for systems in the symmetric phase compared to the spontaneously broken case. At higher energy density (temperature) one would expect larger collision frequency, therefore shorter thermalisation time. The opposite result hints to the importance of slow, soft modes, whose presence is due to the symmetry breaking mechanism. We shall argue in the next Section that these modes are responsible for the realization of the Maxwell-cut in the potential term of the effective equation of motion for the OP. What we find remarkable is that these modes are present not only in the mixed phase but also near (meta)stable equilibria, among the dynamical fluctuations around the ordered vacuum in contrast to the massive perturbative excitation spectrum in the equilibrium.
We complete the analysis of this section with the remark, that after the transition to the stable vacuum shortly a thermalised distribution is recovered for the OP at somewhat increased temperature which agrees with the value given by the equipartition.
## V Jump from the metastable to the stable vacuum
As it has been emphasized in the Introduction, there are (at least) two different descriptions of the transition from the metastable state to the stable one. In the first approach an expansion around the lower pass, the critical bubble (bounce) configuration yields a detailed space-time picture and the transition rate by means of the analytic continuation of the potential experienced by the OP near the (meta)stable position .
Another possibility is to provide for the OP a probabilistic description, obtained by the elimination of all other degrees of freedom in some kind of blocking procedure. This description is based usually on some underlying Master equation for the probability distribution of the OP and the resulting Fokker-Planck equation . The transition to the stable state appears in this approach formally as a tunneling solution of the Fokker-Planck equation. The probabilistic feature of the dynamics of the OP is supposed to arise from assuming a statistical ensemble of the initial conditions.
In our simulation we find results analogous with the predictions of the probabilistic description by analysing the OP-motion starting from a single, well defined initial condition. One might wonder at this point if it is possible to understand the probabilistic tunneling of the OP by following the system from a unique initial condition. The self averaging in time can not be used for this argument since such a transition occurs only once during the evolution in a (quasi)irreversible manner. We have to develop a third approach to the metastable$``$stable transition.
In general, the effective equation of motion for $`\mathrm{\Phi }(t)`$ reflects the typical landscape of the microscopic potential energy functional around the actual point $`\mathrm{\Phi }(𝐱,t)`$ in the configuration space. The classical origin of what appears as a tunneling on the Fokker-Planck level must be the arrival of $`\mathrm{\Phi }(𝐱,t)`$ to the vicinity of some narrow valley opening up towards the stable vacuum. In traversing this valley the landscape changes and the typical fluctuations will be different from those felt in the metastable regime. The constants parametrising the effective equation of motion must reflect this change.
Our goal in this Section is to construct an effective description of the transition to the stable state by carefully tracing the time evolution of the OP. This will be achieved by projecting the microscopic equation of motion onto the homogeneous mode and phenomenologically parametrising it similarly to Eq.(10).
As long as the system is far from the narrow valley of the instability the force is time independent and agrees with the force derived from the perturbative effective potential according to the part of Fig. 3 corresponding to $`\mathrm{\Phi }>0.9`$. When the system arrives close to the entrance of the unstable valley, $`(\mathrm{\Phi }0.8)`$, the soft modes start to be important. This is reflected in the slight glitch in the leftmost piece of the measured force law in Fig. 3. A sequence of glitches results in a situation depicted in Fig. 6. It shows the force acting on the OP in three successive time intervals preceding the event of tunneling for $`h=0.04/\sqrt{6}`$. The $`f(\mathrm{\Phi })`$ curves determined using Eq. (10) in the disjoint time subintervals coincide within error bars for all, but the last one. In this last interval preceding directly the transition towards the direction of negative $`\mathrm{\Phi }`$ values, the fitted force bends down and its average becomes (a small positive) constant. This is characteristic feature of the instant when the system finds the entrance into the unstable potential valley. The vanishing of the force is an indication for the dynamical realization of the Maxwell-cut. The OP moves fast through the valley and the method of fitting the trajectory to Eq. (10) for finding the force fails due to the insufficient statistics.
The fluctuation moments depicted in Fig. 1 tell a bit more about this region. The increased values of the moments, the renormalized coupling constants in Wilsonian sense at vanishing momentum, indicate the enhanced importance of the soft interactions as the OP tunnels through the mean field potential barrier. This softening makes the OP fluctuating with larger amplitude. The valley of instability is in a surprising manner flatter than the typical landscape around equilibrium. The flatness along the motion of the OP (the mode $`𝐤=0`$ in momentum space) comes from the Maxwell-cut. The average curvature of the potential in the transverse directions, i.e. for modes with $`|𝐤|0`$, can be estimated by the second functional derivative of the two dimensional effective action with cutoff $`|𝐤|`$, which can be taken as $`𝐤^2+V_𝐤^{\prime \prime }(\mathrm{\Phi })`$. The increased MS fluctuation of the OP corresponds to a decrease in $`V_{𝐤=0}^{\prime \prime }(\mathrm{\Phi })`$ in the valley. It pushes down $`V_𝐤^{\prime \prime }(\mathrm{\Phi })`$ also for small non-vanishing $`|𝐤|`$, in the low momentum regime which is expected to be the most influenced by the changing landscape.
The lesson of Fig. 6 is that the potential itself should be considered as a fluctuating quantity. This helps to translate into more quantitative terms the above qualitative points. We discuss the projection of the microscopic equation of motion (8) onto the zero momentum sector,
$`0`$ $`=`$ $`\ddot{\mathrm{\Phi }}\mathrm{\Phi }+\mathrm{\Phi }^3+3\overline{\phi ^2}^V\mathrm{\Phi }+\overline{\phi ^3}^Vh`$ (22)
$``$ $`\ddot{\mathrm{\Phi }}+V_{\text{inst}}^{}(\mathrm{\Phi }),`$ (23)
where the symbol $`\overline{\phi ^n}^V`$ means the space average of $`\phi ^n`$, ($`\overline{\phi }^V=0,\mathrm{\Phi }(t,𝐱)=\mathrm{\Phi }(t)+\phi (𝐱,t)`$). The instant potential introduced in the second line contains a deterministic piece, which is the sum of the tree-level potential and of the slowly varying part the second and third moments (eg. $`(\overline{\phi ^n}^V)_{det}`$). This last feature clearly appears graphycally in Fig.1 and a simple model based on a two-phase model of the transition period will be constructed below to account for it. The remaining oscillating pieces of the moments provide the probabilistic fluctuating contribution to the instant potential:
$$V_{\text{inst}}(\mathrm{\Phi })=h\mathrm{\Phi }\frac{1}{2}\mathrm{\Phi }^2+\frac{1}{4}\mathrm{\Phi }^4+(\overline{\phi ^3}^V)_{det}\mathrm{\Phi }+\frac{3}{2}(\overline{\phi ^2}^V)_{det}\mathrm{\Phi }^2+\zeta _0\mathrm{\Phi }+\frac{3}{2}\zeta _1\mathrm{\Phi }^2.$$
(24)
The additive ($`\zeta _0`$) and the multiplicative ($`\zeta _1`$) noises are given by the differences
$`\zeta _0(t)`$ $`=`$ $`\overline{\phi ^3}^V(t)(\overline{\phi ^3}^V)_{det},`$ (25)
$`\zeta _1(t)`$ $`=`$ $`\overline{\phi ^2}^V(t)(\overline{\phi ^2}^V)_{det}.`$ (26)
Note that no friction terms can be introduced in a natural way into the system (23-26). Therefore we face the intriguing question, how irreversibility is realised in such a system. The time-correlation matrix $`\overline{\zeta _i(t)\zeta _j(t+\tau )}^t`$ appears to us to be the key object for its investigation, to which we plan to return in the future.
In order to gain more insight how this works we build into Eq.(23) the consequences of the mixed two-phase picture of the phase transformation. The microscopical basis for this picture is provided by the thermal nucleation whose quantitative discussion is given for completeness in the Appendix.
More specifically, we assume that the space can be splitted into sharp domains (neglecting the thickness of the walls in between), where the field is the sum of the constant background values $`\mathrm{\Phi }_{0\pm }`$ and the fluctuations $`\stackrel{~}{\phi }_\pm `$ around it,
$$\mathrm{\Phi }_\pm (𝐱,t)=\mathrm{\Phi }_{0\pm }+\stackrel{~}{\phi }_\pm (𝐱,t).$$
(27)
We assume local equilibrium in both phases, based on the smooth evolution of the temperature as displayed in Fig.2.
The actual value of the order parameter is determined by the surface ratio $`p(t)`$ occupied by the stable phase:
$$\mathrm{\Phi }(t)=p(t)\mathrm{\Phi }_0+(1p(t))\mathrm{\Phi }_{0+}.$$
(28)
Simple calculation then yields
$`(\overline{\phi ^2}^V)_{det}(t)`$ $`=`$ $`{\displaystyle \frac{\mathrm{\Phi }_{0+}\mathrm{\Phi }(t)}{\mathrm{\Phi }_{0+}\mathrm{\Phi }_0}}\left(\overline{\mathrm{\Phi }_{}^2(𝐱,t)}^V\overline{\mathrm{\Phi }_+^2(𝐱,t)}^V\right)+\mathrm{\Phi }_{0+}^2\mathrm{\Phi }^2(t)+\overline{\stackrel{~}{\phi }_{+}^{}{}_{}{}^{2}}^V,`$ (29)
$`(\overline{\phi ^3}^V)_{det}(t)`$ $`=`$ $`{\displaystyle \frac{\mathrm{\Phi }_{0+}\mathrm{\Phi }(t)}{\mathrm{\Phi }_{0+}\mathrm{\Phi }_0}}\left(\overline{\mathrm{\Phi }_{}^3(𝐱,t)}^V\overline{\mathrm{\Phi }_+^3(𝐱,t)}^V\right)+\overline{\mathrm{\Phi }_+^3(𝐱,t)}^V3\mathrm{\Phi }(t)(\overline{\phi ^2}^V)_{det}(t)\mathrm{\Phi }^3(t),`$ (30)
where the volume averages should be read off the corresponding equilibria on the two sides of the transition. If one takes the values of $`\mathrm{\Phi }_{0\pm },\overline{\stackrel{~}{\phi }_\pm ^n}^V`$ from the respective equilibria determined in the same simulation, a quite accurate description of the shape of the two fluctuation moments arises in the whole transition region and its close neighbourhood using $`\mathrm{\Phi }(t)`$ to parametrize their $`t`$-dependence. (Note that $`\overline{\mathrm{\Phi }_\pm ^n(x,t)}^V`$ does not depend on time.) The deterministic part of the moments are therefore well-defined functions of $`\mathrm{\Phi }`$. In this way, a simple explicit construction can be given for the effective noisy equation of the OP-motion, if the above pieces are supplemented by the correlation characteristics of the two kinds of noises.
The RMS (root mean square) fluctuations found numerically on the transition part of the trajectory and its close neighbourhood are the following:
$$\sqrt{\overline{\zeta _0^2(t)}^t}=0.0063(10),\sqrt{\overline{\zeta _1^2(t)}^t}=0.0054(10),\text{for}N=64$$
(31)
(its magnitude increases with the size of the system). The magnitude of their equilibrium cross correlation was found $`10^5`$.
As a corollary of this construction one can demonstrate the absence of the deterministic part of the acceleration of the order parameter in the transition period.
Substituting Eq.(30) into Eq.(23), one finds for the deterministic part of the force,
$$f\left(\mathrm{\Phi }\right)=\left(1+\frac{\overline{\mathrm{\Phi }_+^3(𝐱,t)}^V\overline{\mathrm{\Phi }_{}^3(𝐱,t)}^V}{\mathrm{\Phi }_{0+}\mathrm{\Phi }_0}\right)\mathrm{\Phi }+\frac{\mathrm{\Phi }_{0+}\overline{\mathrm{\Phi }_{}^3(𝐱,t)}^V\mathrm{\Phi }_0\overline{\mathrm{\Phi }_+^3(𝐱,t)}^V}{\mathrm{\Phi }_{0+}\mathrm{\Phi }_0}h.$$
(32)
The average of the equations of motion in the respective equilibria,
$$\mathrm{\Phi }_\pm ^3(𝐱,t)\mathrm{\Phi }_{0\pm }h=0$$
(33)
implies the vanishing of the deterministic force in Eq. (32), when exploiting the equality of the volume and the ensemble average in this case. Eq. (23) transforms into
$$\ddot{\mathrm{\Phi }}(t)+\zeta _0(t)+3\zeta _1(t)\mathrm{\Phi }(t)=0.$$
(34)
This is the dynamical realization of the Maxwell-construction holding when the mixed phase model with local equilibrium is valid.
If the force is calculated from the full instant potential, $`V(\mathrm{\Phi })_{\text{inst}}`$ in Eq. (24), it depends parametrically on the moments $`\overline{\phi ^2}^V(t)`$ and $`\overline{\phi ^3}^V(t)`$. The approximate trajectory $`\mathrm{\Phi }_{\text{inst}}(t)`$, defined by minimising $`V(\mathrm{\Phi })_{\text{inst}}`$ with respect to $`\mathrm{\Phi }`$, where the moments are taken from the numerically determined time evolution, reproduced accurately the observed OP, $`\mathrm{\Phi }(t)`$ with the following RMS/unit time:
$$\sqrt{\overline{(\mathrm{\Phi }(t)\mathrm{\Phi }_{\text{inst}}(t))^2}^t}=0.0014(5),\text{for}N=64.$$
(35)
This construction interpretes the OP-trajectory as a continous deformation of the instant potential with the OP ”sitting” permanently in the actual minimum. Notice that such motion is possible only if the OP continously undergoes some sort of dissipation.
The vanishing of the acceleration was checked by comparing the computer generated OP trajectory in the transition region with a ballistic motion in a viscous medium. In particular it was tested that the ratio $`h/\dot{\mathrm{\Phi }}`$ is nearly time and $`h`$ independent for small enough $`h`$. The dynamical friction measured by the above ratio tends to a constant with decreasing $`h`$ at fixed lattice size. Although the error for its value in each individual transition is rather small, the central values obtained in different runs fluctuate quite strongly, which leads eventually to a large error of the mean calculated as an average of runs with different initial configurations having the same energy density.
It is worth to note that the order of magnitude of these “renormalized”, i.e. IR determined values of the friction coefficient agree with the peak value in Fig. 4 for $`N=64`$. The “running” friction coefficient determined by the blocking in time, however, drops as $`b_t`$ is increased and takes extremely small values at $`b_t20`$, the average time length of the ballistic fits for the transition. This serves an example for the dependence of the renormalization group flow on the details of the blocking. It remains to be understood whether the agreement between the peak value and the ballistic fit is an accident or follows from the internal dynamics.
## VI Conclusions and future directions
In this paper we have investigated in detail the decay of the false vacuum in a classical lattice field theory based exclusively on the effective theory of the order parameter. Two versions of the effective theory were reconstructed from its trajectory derived from the microscopical equations of the theory. The first refers to the (meta)stable branch of the motion. The second one which takes into account the existence of a mixed phase during the transition period describes very well the transition together with its neighbourhood. The first equation has the form of conventional mechanical motion taking place in a dissipative noisy environment. The dissipation, the dynamical breakdown of the time inversion symmetry was found only for times longer then the minimal microscopic time scale of the system, the autocorrelation time of the noise.
As a corollary we find that the effective phenomenological OP-theory for the decay of the false vacuum through a narrow but flat valley demonstrates the presence of a dynamical Maxwell-construction.
The next possible directions of extending this work include the investigation of the effect of the quantum fluctuations on the large amplitude thermalization of the one-component $`\mathrm{\Phi }^4`$ theory . The effective equation of motion for the order parameter of a quantum field theory can be studied by breaking up the field $`\widehat{\mathrm{\Phi }}(𝐱,t)`$ into a “classical” and a quantum part:
$$\widehat{\mathrm{\Phi }}(𝐱,t)=\mathrm{\Phi }_{\text{cl}}(𝐱,t)+\widehat{\varphi }(𝐱,t),\widehat{\mathrm{\Phi }}(𝐱,t)=\mathrm{\Phi }_{\text{cl}}(𝐱,t).$$
(36)
The quantum effects enter the equation of $`\mathrm{\Phi }_{\text{cl}}(𝐱,t)`$ through $`\widehat{\varphi }^n(𝐱,t)`$, $`n=2,3`$ analogously to Eq. (23). The method of the mode-function expansion displays explicitly the way the quantum effects $`(\mathrm{})`$ enter the “classical” equations of motion. In this method one solves the time evolution of the coefficient functions of the expansion
$$\widehat{\varphi }(𝐱,t)=\underset{𝐐}{}\left[f_𝐐(𝐱,t)a_𝐐+f_𝐐^{}(𝐱,t)a_𝐐^+\right],$$
(37)
where $`[a_𝐐,a_𝐐^{}^+]=\delta _{\mathrm{𝐐𝐐}^{}}`$. The initial conditions are fixed by requiring the canonical commutation relations to be fulfilled at $`t=0`$:
$$f_𝐐(𝐱,0)=\sqrt{\frac{\mathrm{}}{2\omega _𝐐V}}e^{i\mathrm{𝐐𝐱}},\dot{f}_𝐐(𝐱,0)=i\sqrt{\frac{\mathrm{}\omega _𝐐}{2V}}e^{i\mathrm{𝐐𝐱}}.$$
(38)
The quantum effects relative to Eq. (8) show up as $`𝒪(\mathrm{})`$ corrections to the coefficients of the effective equation of motion of $`\mathrm{\Phi }_{\text{cl}}(𝐱,t)`$. The adequate form of the effective equation of motion for $`\overline{\mathrm{\Phi }_{\text{cl}}(𝐱,t)}^V`$ represents a further challenge .
The extension of our results to classical field theories with continous internal symmetry describing the dynamics of the inflaton field coupled to the Higgs+Gauge system is also of actual interest.
## Appendix: The nucleation picture
In this section we analyse the transition using the more conventional statistical approach of thermal nucleation theory.
From the study (on lattices up to $`N=512`$) of the detailed microscopical field configurations it turned out that the phase transformation starts by the nucleation of a single bubble of the stable ($`\mathrm{\Phi }<0`$) phase. Late-coming further large bubbles are aggregated to it as well as the small size (consisting of $`n_{bubble}<5060`$ joint sites) bubbles. Following the nucleation the expansion rate of the large bubble governs the rate of change of the order parameter. To a very good approximation each individual transition could have been characterised by a constant value of $`\dot{\mathrm{\Phi }}`$ in this interval. Two mechanisms are known to lead to this behavior. The first is scattering of hard waves (“particles”) off the bubble wall, while in the second the expansion velocity is limited by the diffusive aggregation of smaller bubbles . The latter process seem to be the dominant in the kinetic Ising model .
The statistics of the “release” time of the supercritical bubble from the metastable state shows at first sight rather peculiar characteristics. The binned histogram for larger values of $`h`$ shows an asymmetric peaked structure which apparently deviates from the exponential distribution characterising the thermal nucleation scenario (Fig.7). The very early transitions ($`t<t_{max}`$) seem to be suppressed for small values of $`h`$. We expect they correspond to transitions, which happen before the system reaches the metastable state.
The fall-off for $`t>t_{max}`$ starts nearly exponentially, but the histogram develops a very long tail when $`h0`$. The bigger sample is used for estimating the probability distribution, the more suppressed is the weight of these events in the normalised distribution. In practice this means a longer time interval where a good linear fit can be obtained to the log-linear histogram. Eventually, the separation of a clean exponential signal is possible, and the slope can be compared with predictions of the nucleation theory.
The nucleation rate is proportional to the volume of the system $`(N^2)`$. In Fig.8 we see evidence for the size independence of the transition rate per unit surface. The logarithm of this quantity can be estimated following the standard nucleation theory :
$$\mathrm{ln}\mathrm{\Gamma }=K\frac{S_2}{T}.$$
(39)
$`S_2`$ is determined approximately by the action of the bounce solution, connecting the two vacua. In this investigation we have studied at fixed temperature the $`h`$-dependence of $`S_2/T`$, assuming the $`h`$-independence of $`K`$. In the thin wall approximation the following expression is used based on the tree level potential:
$$(S_2)_d=2\pi \overline{r}^2h\sqrt{\frac{6}{\lambda }}|m|+8\sqrt{2}\pi \overline{r}\frac{|m|^3}{\lambda },$$
(40)
(the first term on the right hand side corresponds to the volume energy of a bubble of radius $`\overline{r}`$, the second one to its surface energy). For the critical bubble size a very simple expression is obtained for the exponent of the rate in terms of dimensionless quantities:
$$\frac{S_2(\text{thin wall})}{T}=\frac{16\pi }{\sqrt{6}}\frac{|m|^5}{\lambda ^{3/2}}\frac{1}{h_dT_d}=\frac{4\pi }{9}\frac{1}{hT}.$$
(41)
The linear dependence on $`1/h`$ is fulfilled in our numerical calculations very well, but the predicted action is more than one order of magnitude larger than what can be derived from the slope of Fig.8: $`S_2/T(\text{measured})0.1/(hT)`$. If one relaxes the thin wall approximation and solves numerically the two-dimensional bounce equation directly for several $`h`$, one finds $`S_2(\text{bounce})/T=0.78/(hT)`$.
Further improvement can be obtained by applying the temperature corrected effective potential in the bounce equation. We have determined the parameters of the $`T`$-dependent potential directly from our numerical calculation in the following way. The restoring force was measured for a certain $`h`$ around both the stable and the metastable minima as decribed in section 4. Next an interpolating fit has been constructed of the form: $`h_{\text{eff}}+m_{\text{eff}}^2\mathrm{\Phi }+\lambda _3\mathrm{\Phi }^2+\lambda _{\text{eff}}/6\mathrm{\Phi }^3`$. It turned out that in the best fit $`\lambda _30`$ is fulfilled always, while the values of the other coefficients are not too far from their tree level values. Then a bounce solution can be built on the corresponding (real!) potential which includes the temperature corrections. The final result is quite close to the measured value of the rate logarithm: $`S_2(T,\text{bounce})/T=0.29/(hT)`$. The same real interpolation can be built on the neighbourhood of the minima of the 2-loop $`T`$-dependent effective potential, leading to $`S_2(T,\text{bounce})/T=0.2/(hT)`$. By common experience in the surface tension simulations a factor of 2-3 difference in $`S_2`$ is expected to arise relative to the mean field theory.
We conclude, that the finite temperature corrections are important for quantitative treatment of the nucleation rate within the thermal nucleation theory.
Next we turn to the discussion of the possible nucleation threshold at small $`h`$. The average “release time” increases for fixed lattice size with decreasing $`h`$ (Fig.9). On a lattice of fixed size we found small $`h`$ values for which we could not detect any transition, what makes very probable the existence of a threshold value of the external field $`h_{th}(N)`$. We did not attempt to locate this value beyond the simple hyperbolic fits to the few largest $`t_{release}`$ values. The value of $`h_{th}(N)`$ remains stable when the smallest $`h`$ is excluded from the fit, therefore we conclude that a threshold magnetic field exists similar to the case of the kinetic Ising model . The value of $`h_{th}`$ decreases when the lattice size is increased. Intuitively we expect $`h(\mathrm{})=0`$, but with the three lattices studied by us this conjecture cannot be demonstrated.
## Acknowledgements
The authors are grateful for valuable discussions to C. Gagne, J. Hajdu, A. Jakovác, Z. Rácz and H.J. de Vega. They acknowledge the use of computing resources generously provided by the University of Bielefeld and the Dept. of Computer Science and the Inst. for Theoretical Physics of the Eötvös University. This research has been supported by CNRS and the Hungarian Science Fund (OTKA).
|
warning/0004/cond-mat0004229.html
|
ar5iv
|
text
|
# Axial phase of quantum fluids in nanotubes
## I Introduction
The physical properties of carbon nanotubes are expected to manifest a variety of intriguing one-dimensional (1d) and quasi-1d behaviors . We, and other groups, have found that fluids absorbed within these tubes ought to exhibit an “effective” dimensionality which is sensitive to the geometry, species, number ($`N`$) of particles and temperature ($`T`$). For example, in the case of isolated nanotubes, the ideal gas regime of very low $`N`$, is predicted to show crossover between 1d and 2d behavior as $`T`$ increases, due to the excitation of azimuthal motion of the atoms adsorbed on the inner wall of the tubes . For a bundle of nanotubes, in contrast, it has been predicted and found experimentally that low $`N`$ gases which have small diameter are adsorbed strongly in interstitial channels, between tubes, in which the gases display 1d motion . Larger diameter gases adsorb preferentially within the tubes.
The novel collective properties of the absorbed interacting fluid at higher $`N`$ are now being explored. In this paper we consider the evolution, as a function of $`N`$, of the structure and properties of dense quantum fluids within individual nanotubes at $`T=0`$. Consider He for specificity. Above the ideal gas regime of density, there appears a regime of density in which liquid <sup>4</sup>He is adsorbed on the walls of the tube; we call this a cylindrical “shell” phase . At somewhat higher density this fluid should solidify; we expect that this film is similar to the much studied incommensurate monolayer solid film on graphite . For values of N below a threshold value $`N_a`$ there exists only this shell phase. We predict here a transition in which there appears, for $`N>N_a`$, an axial phase of this fluid, signified by the presence of a fluid confined to the vicinity of the tube axis. This axial phase transition is qualitatively similar to two other transitions familiar in the field of adsorption: capillary condensation (CC) and the layering transition . While the latter (2d) transition is known to occur at finite $`T`$, the appearance of the axial phase is a genuine thermodynamic transition only at $`T=0`$ because the geometry is 1d, so that fluctuations eliminate the transitions at higher $`T`$. This feature is also the case for a CC transition in a single pore, but often in experiments the effect of interactions with other pores is such as to permit a CC transition at nonzero $`T`$ . Similar behavior of the axial phase transition should occur in a nanotube bundle (“rope”), but we have not yet investigated that problem. Also, the same phenomenon is expected to occur for a classical system since nothing specifically related to quantum effects plays a paramount role .
We have investigated the axial phase with three distinct methods. The next section presents Hartree model calculations for four systems: <sup>4</sup>He, <sup>3</sup>He, their mixtures, and H<sub>2</sub>. These calculations rely extensively on the use of other workers’ calculations and data for 2d quantum systems . Section III presents results of Path Integral Monte Carlo studies of H<sub>2</sub>. Section IV reports results for <sup>3</sup>He-<sup>4</sup>He mixtures, obtained with the Hartree approach. Section V describes the results obtained with Density Functional calculations applied to <sup>4</sup>He . The comparison between our various approaches is interesting, as is that between the present data and “exact” results for <sup>4</sup>He in 1d . Section VI summarizes our results and conclusions.
## II Hartree Model Calculations for Helium and Hydrogen
Our goal is to determine the threshold coverage for formation of the axial phase. The general approach described in this section is analogous to that used in previous studies of the problem of layer promotion . The assumption is made that, above threshold, atoms (or molecules in the case of H<sub>2</sub>) form two coexisting phases, the axial phase and the shell phase. These are characterized by 1d densities $`N_a/L`$ and $`N_s/L`$, respectively. The condition for equilibrium is that the chemical potentials of the two phases coincide:
$$\mu _{axial}=\mu _{shell}$$
(1)
The axial phase is thought of as a 1d phase affected by the “external” Hartree potential provided by the shell phase and the host nanotube. Its chemical potential is assumed to satisfy
$$\mu _{axial}(N_a/L)=ϵ_a+\mu _{1d}(N_a/L)$$
(2)
The 1d approximation ought to be valid here in that we focus on the regime when the rms displacement transverse to the axis is small compared to the axial phase interparticle spacing and the axial and shell atoms are well separated spatially. Here $`\mu _{1d}(N_a/L)`$ is the chemical potential of a 1d fluid at density N<sub>a</sub>/L and $`ϵ_a`$ is the eigenvalue of the atoms in the Hartree potential. At the threshold for forming the axial phase, $`\mu _{1d}`$ assumes its lowest value, the ground state cohesive energy of the 1d fluid. This energy has been found previously for <sup>4</sup>He to be of order 2 mK, which is negligible small in the present context, so we ignore it for <sup>4</sup>He and the other cases studied here .
The energy $`ϵ_a`$ was computed from a numerical solution of the Schrödinger equation for ground state (zero angular momentum) atomic motion in the Hartree potential:
$$V_{tot}(r)=V_C(r)+\theta V_H(r)$$
(3)
where $`V_C(r)`$ is the potential near the axis due to the carbon atoms alone computed in the assumption of smooth tube walls as in previous studies . The second term is a Hartree interaction, where $`\theta `$ is the density of gas atoms/molecules in the shell, and $`\theta V_H(r)`$ is the interaction due to the shell:
$$V_H(r)=𝑑𝐫^{}|\mathrm{\Psi }_s(𝐫^{})|^2U_{gg}(|𝐫𝐫^{}|)$$
(4)
where $`\mathrm{\Psi }_s(𝐫^{})`$ is the wave function of gas atoms/molecules in the shell and $`U_{gg}`$ is the gas-gas interaction. We make the further approximation of assuming that the atoms in the shell phase are narrowly confined in the radial direction:
$$|\mathrm{\Psi }_s(𝐫)|^2\delta (rR)$$
(5)
where $`R`$ is the radius of the cylindrical surface near which the atoms are situated. Then the evaluation of $`V_H`$ reduces to integrations over the longitudinal and azimuthal coordinates, as shown in Ref. .
The ground state eigenfunction of the Schrödinger equation for an atom close to the axis is of the form $`\mathrm{\Psi }_a(𝐫)=f(r)exp(ikz)`$, where $`r`$ and $`z`$ are the cylindrical coordinates of the atom’s position vector $`𝐫`$. The eigenvalue $`ϵ_a`$ is then determined from the differential equation satisfied by the ground state radial wave function, $`f(r)`$:
$$\frac{d^2f}{dr^2}+\frac{1}{r}\frac{df}{dr}+\frac{2m}{\mathrm{}^2}\left[ϵ_aV_{tot}(r)\right]f(r)=0$$
(6)
We solve this equation numerically.
Figures 1 and 2 display for the He isotopes the two chemical potentials appearing in Eq. 1; their curves’ crossing occurs at the threshold at which the axial phase appears. The abscissa in these figures is a 2d density of the shell phase:
$$\theta =N_s/(2\pi RL)$$
(7)
As seen in figures 1 and 2, the eigenvalue $`ϵ_a`$ is a nearly linear function of the shell phase density in the neighborhood of the transition, as expected from a perturbation theory of the Hartree interaction, based on Eq. 3.
We have made the simplest plausible assumption for the function $`\mu _{shell}(\theta )`$ \- that it coincides with the 2d chemical potential on graphite, apart from an additive constant arising from the stronger substrate attraction and resulting larger binding energy in the tubes:
$$\mu _{shell}(\theta )=\mu _{graphite}(\theta )+\mathrm{\Delta }E_b$$
(8)
Here $`\mathrm{\Delta }E_b`$ is the difference in binding energy between the result for the nanotubes and that for graphite . The justification for this implicit neglect of the effect of curvature appears in the Appendix. Our values of these functions $`\mu _{shell}(\theta )`$ are taken from theoretical studies of Ref. which are consistent with experiments reported in Ref. .
The key properties of the shell and axial phases at threshold are displayed in Tables I and II and Fig. 3 for the isotopes <sup>4</sup>He and <sup>3</sup>He, respectively. There are no significant differences between the qualitative behaviors of the two isotopes. Consider, for example, the case of <sup>4</sup>He within nanotube having a radius of $`R_C=6`$ Å. Note that the He shell radius, $`R=3.06`$ Å, is about 3 Å closer to the axis than is the carbon shell; $`R`$ increases by ca. 1 Å as $`R_C`$ increases by 1 Å. As $`R_C`$ increases, the energy levels of the shell ($`E_0`$) and axial states at threshold increase monotonically. The threshold density $`\theta _c`$ of the shell state is insensitive to this change, remaining close to the known value 0.115 Å<sup>-2</sup> for monolayer completion density on graphite . However, the threshold chemical potential $`\mu _c`$ varies much more; it increases by about 20 K for each 1 Å change in $`R`$; for $`R_C`$ = 8 Å, $`\mu _c`$ approaches the value -35 K for monolayer completion on graphite . Note that the potential energy responsible for the axial state changes by an even larger amount than $`\mu _c`$. The minimum value of the total potential energy, $`V_{tot}(r_{min})`$, changes by 90 K as $`R_C`$ increases by 2 Å, while $`\mu _c`$ changes by only 40 K. The reason is that the zero point kinetic energy also changes significantly, as can be understood from the uncertainty principle. The kinetic energy is much larger at $`R_C=6`$ Å, for which state the root mean square radial coordinate is $`r_{rms}=0.41`$ Å, than at $`R_C=8`$ Å, for which $`r_{rms}=2`$ Å. Note from the table that the carbon shell’s contribution to the potential energy $`V_C(r)`$ is typically 40% of the total potential energy while the He shell contribution is about 60%; these proportions change slowly as $`R_C`$ changes. Finally, we note the qualitative change in the axial state wave function seen in Fig. 3. For the case of $`R_C=8`$ Å, the axial state’s probability density is no longer confined to the immediate vicinity of the axis, exhibiting a maximum near $`r=2`$ Å. One observable property which is determined by this spread is the momentum distribution function, which ought to be much narrower in the case of such a dispersed wave function. We intend to consider this topic in future work.
Figures 4 and 5 and Table III present the results of analogous calculations for H<sub>2</sub> adsorption in nanotubes of varying size. The data corresponding to $`R_C=8`$ Å look qualitatively similar to those for He except with different numerical values, of course. As for <sup>4</sup>He, the threshold values of coverage and chemical potential are similar to those for monolayer completion on graphite ($`\theta _c=0.094`$ Å<sup>-2</sup>, $`\mu _c=244`$ K) .
For $`R_C=7`$ Å, the axial phase threshold value of the chemical potential for H<sub>2</sub>, i.e. the crossover seen in Fig. 4, moves to a much lower value ($`\mu _c=385`$ K) than for $`R_C=8`$ Å (-261 K). For the smallest case studied, $`R_C=6`$ Å, we find no crossing of the chemical potential curves. This means simply that nanotubes which are so small do not produce an axial phase, according to our calculations.
## III Path Integral Simulations for Hydrogen
As a check on the validity of the Hartree calculations we have performed molecular simulations of hydrogen adsorption in nanotubes, using the path integral formalism implemented in the grand canonical ensemble. The algorithm we have used follows the previous work of Wang and Johnson . The Silvera-Goldman isotropic interaction potential was used for the H<sub>2</sub>-H<sub>2</sub> potential. The nanotube-hydrogen potential used in the path integral simulations was the same as that used in the Hartree calculations. The path integral calculations were carried out at finite temperature, so we do not expect complete agreement with the Hartree method. In particular, the shell-axial first order phase change will disappear in favor of a continuous transition as the axial phase is populated. Calculations were performed at 5 and 10 K for nanotubes of radius 5, 6, 7, and 8 Å. In the path integral formalism each quantum molecule is replaced by a classical ring polymer containing some number of beads. The accuracy of the simulations depends on the number of beads used in each ring, with the results becoming exact as the number of beads per ring approaches infinity. We found that 50 beads per ring were sufficient to obtain convergent results at both 5 and 10 K.
The shell and axial phase densities of hydrogen in the 7 Å tube are plotted in figure 6 as a function of the chemical potential. The shell phase density remains fairly constant while the axial phase increases dramatically at a chemical potential of about $`375`$ K, in reasonable agreement with the Hartree value of $`\mu _c=385`$ K given in Table III. The results at a temperature of 5 K (not shown) are very similar, but the population of the axial phase commences at a somewhat higher chemical potential. The shell and axial phase densities for the 6 Å radius nanotube are shown in figure 7. The path integral calculations demonstrate that population of the axial phase does occur, with the onset at a chemical potential of about $`325`$ K. This is inconsistent with the Hartree calculations, which do not show the formation of an axial phase at any chemical potential for this small value of R.
Figure 8 is a plot of the density profile in the 6 Å nanotube at three different chemical potentials. Note from figure 8 that the location of the shell phase is pushed progressively outward in response to the increased population of the axial phase. The disagreement between the Hartree and path integral calculations can be ascribed to the fact that the Hartree model does not account for the shift in the location of the shell phase due to the presence of the axial phase, and therefore the Hartree model misses the axial phase observed in the simulations. The density profile for the 7 Å tube is plotted in figure 9 for a series of chemical potentials. Note that that shell phase is essentially unperturbed by the presence of the axial phase in the 7 Å nanotube.
The coverage and density profiles for the 8 Å nanotube are plotted in figures 10 and 11, respectively. Note that the onset of axial phase population occurs at a higher chemical potential (larger bulk pressure) than for either the 6 or 7 Å nanotube. This is somewhat surprising given that the shell phase must be compressed in order to make way for the axial phase in the 6 Å tube. The onset of population of the axial phase in the 8 Å nanotube occurs at about $`\mu =298`$ K, in fair agreement with the Hartree value of $`\mu _c=261`$ K from Table III. Note from figure 11 that it is evident that the 8 Å nanotube does not actually have an axial phase, but rather a second shell phase, centered about 2 Å from the center of the tube. This agrees with the Hartree calculations of figure 5.
## IV Helium Mixtures
We now consider the case of isotopic mixtures of helium adsorbed at medium to high density within the tube. In the analogous situation on the graphite basal plane, there is complete isotopic separation at $`T=0`$ . We expect the same behavior in the nanotubes at coverages below that required for producing the axial phase. In this case, the criterion determining the distribution of the two phases on the surface is equality of their spreading pressures. Figure 12 presents the coexisting 2d densities of these phases, as derived from application of this criterion to the equation of state data of Bruch and Figure 13 shows a schematic view of the mixture configuration, determined by computing the threshold for the axial phase. We found that <sup>3</sup>He atoms first go to the axial region surrounded by the <sup>4</sup>He atoms at virtually the same spreading pressure at which <sup>4</sup>He atoms begin to go on the axis. The preferential binding in the <sup>4</sup>He region is a consequence of the higher 2d shell density at <sup>3</sup>He-<sup>4</sup>He coexistence. Our numerical results for this onset condition appear in Figure 14 and Table IV.
## V Density Functional Study of <sup>4</sup>He
We consider that the energy of an ensemble of $`N`$ helium atoms in the nanotube has the form
$$E[\rho ]=E_0[\rho ]+𝑑𝐫\rho (r)V_C(r)$$
(9)
where $`V_C(r)`$ is the adsorption potential due to all the carbon atoms in the nanotube evaluated at point r and $`E_0[\rho ]`$ is the Finite Range Density Functional (FRDF) that represents the energy of inhomogeneous <sup>4</sup>He. The form of $`E_0[\rho ]`$ and the discussion of formalism can be found in Ref. . Such density functionals have previously been applied to the study of adsorbed films and impurity atoms in liquid He.
At zero temperature, the density of the $`N`$ helium atoms is $`\rho (r)=N|\mathrm{\Psi }(r)|^2`$, where $`\mathrm{\Psi }(r)`$ is the single particle ground state wave function that minimizes the energy $`E[\rho ]`$. The minimization procedure leads to the following Hartree-Fock equation for $`\mathrm{\Psi }`$,
$$((\mathrm{}^2/2m)^2+U[\rho ])\mathrm{\Psi }=\mu \mathrm{\Psi }$$
(10)
Here $`U[\rho ]`$ is the effective single particle interaction and is equal to
$$U[\rho ]=\delta E/\delta \rho .$$
(11)
The ground state wave function, $`\mathrm{\Psi }`$, depends only on the distance to the axis of the nanotube, $`r`$, and equation 10 becomes one-dimensional. We compute $`\mathrm{\Psi }`$ and $`\mu `$ by solving equation 10 self-consistently. The results are displayed in figures 15-17 , for the 6 Å nanotube.
In figure 15 we plot the density as a function of the radial coordinate $`r`$ for different values of the chemical potential. For small $`\mu `$ the density has only one peak centered at $`R=3.06`$ Å. As long as the value of $`\mu `$ increases, the height of this peak grows, reaching densities much larger than the saturation density of bulk <sup>4</sup>He (0.022 Å<sup>-3</sup>), suggesting the formation of a solid phase (which is not accurately described by our method). For larger $`\mu `$ the axial peak becomes appreciable and continues to grow as $`\mu `$ increases, while the shell peak remains constant. The picture agrees very well with the model we proposed in section II.
We evaluated the number of atoms in the axial region, $`N_a`$, and in the shell region, $`N_s`$, by integrating each of the peaks of the density. In figure 16 we plot the axial density $`N_a/L`$ and the shell density $`\theta =N_s/(L2\pi R)`$ vs. $`\mu `$. As we can see $`N_a/L`$ is non-zero for $`\mu `$ greater than $`\mu _c`$ = -108 K. This axial phase threshold corresponds to $`N/L`$=2.2 Å<sup>-1</sup> and to $`\theta _c`$=0.113 Å<sup>-2</sup> in agreement with our previous Hartree model’s prediction (see table I).
The energy per atom and the chemical potential $`\mu `$ are displayed in figure 17. Note a kink in the $`\mu `$ curve at the axial phase threshold. This effect can be attributed to the presence of a “gas-like” axial phase, since $`d\mu /dN`$ is proportional to the inverse of the compressibility.
## VI Conclusions
In this work we have presented various calculations pertaining to a hypothetical quasi-one-dimensional axial phase of quantum fluids adsorbed in nanotubes. Such a phase ought to be thermodynamically distinct from the high density cylindrical shell phase at lower density. Our calculations use an approximate Hartree model for H<sub>2</sub> and the He isotopes. Low temperature path integral calculations for H<sub>2</sub> are qualitatively consistent with the Hartree results, apart from the case of very small R. For <sup>4</sup>He, a density functional model yields results which are also consistent with those of the Hartree method. We find such an axial phase in all of these cases, except for very small tubes containing H<sub>2</sub>. We note that similar behavior is expected for classical fluids consisting of small atoms, e.g. Ne and Ar. Both our methodology and the results presented here are qualitatively similar to those obtained previously for the problem of determining the monolayer completion of quantum gases adsorbed on graphite . Quantitatively, however, there is a big difference: the threshold chemical potential values are lower for small tubes than for a flat surface, due to the higher effective coordination of the particles in small tubes.
We have not addressed here one of the most important questions: what are the properties of this axial phase? We expect these to be entirely different from those of the second layer on graphite because of the difference in dimensionality. In particular, we anticipate the behavior to be that of the so-called “Luttinger liquid” in one dimension. Among the most remarkable of these properties for fermions are the absence of a fermi surface and the dominant role of spin and density fluctuations at low T. We hope that relevant calculations and experiments are soon forthcoming.
###### Acknowledgements.
We are grateful to M. J. Bojan, M. Boninsegni, M. W. H. Chan, V. H. Crespi, P. C. Eklund, R. B. Hallock, S. Hernandez, W. A. Steele and K. Williams for helpful discussions. This research has been supported by ARO, NSF, the University of Buenos Aires, and the Petroleum Research Fund of the American Chemical Society. APPENDIX Curvature correction to energy
We here compute the leading curvature correction to the potential energy $`P(R)`$ per atom located on a cylindrical surface of infinite length and radius $`R`$. It is typical in such cases to find an expansion of the form:
$$\frac{P(R)}{P(\mathrm{})}1+\left(\frac{b}{R}\right)^2$$
(12)
where $`b`$ is a characteristic length of the system, assumed to be small compared to $`R`$ in the present asymptotic limit. We shall confirm this expectation here.
It is reasonable to identify the curvature correction as arising primarily from the attractive part of the interatomic interaction since that experiences a larger distance scale than the repulsive part of the interaction and so becomes comparable to $`R`$ first as the graphene sheet is curved. The attraction varies as
$$V(r)\frac{C}{r^6}$$
(13)
where $`C`$ is the interatomic dispersion coefficient. For the case of atoms having 2d density $`\theta `$,
$$2\frac{P(R)}{C\theta }=R𝑑z𝑑\phi \frac{1}{[z^2+(2Rsin(\phi /2)^2]^3}.$$
(14)
where the 2 on the left side avoids double-counting and we bear in mind that eventually we will need a small distance cutoff ($`r=a`$) in order to get a finite answer. The planar limit, $`R=\mathrm{}`$, involves only small $`\phi `$ contributions, leading to
$$2\frac{P(\mathrm{})}{C\theta }=𝑑z𝑑x(z^2+x^2)^3=\frac{\pi }{2a^4}$$
(15)
The difference $`\mathrm{\Delta }P`$ between this value and that of P at finite $`R`$ may be evaluated with an asymptotic expansion. We find
$$8\frac{\mathrm{\Delta }P}{C\theta }=\frac{3\pi }{8a^2}$$
(16)
so that the ratio
$$\frac{\mathrm{\Delta }P}{P}=\frac{3}{16}\left(\frac{a}{R}\right)^2$$
(17)
confirming the anticipated curvature dependence. A plausible near-neighbor cutoff is $`a=(\pi /\theta )^{1/2}2`$ Å. For a nanotube of $`R_C=7`$ Å, the film is situated at $`R4`$ Å. Then the curvature correction to the energy per atom is about 5%.
FIG. 1
FIG. 2
FIG. 3
FIG. 4
FIG. 5
FIG. 6
FIG. 7
FIG. 8
FIG. 9
FIG. 10
FIG. 11
FIG. 12
FIG. 13
FIG. 14
FIG. 15
FIG. 16
FIG. 17
|
warning/0004/hep-ph0004077.html
|
ar5iv
|
text
|
# Do Neutrinos Decay?*footnote **footnote *Presented at the 5th International Conference on Physics Potential & Development of μ⁺μ⁻ Colliders, San Francisco, December 1999.
## I Introduction
It is generally agreed that most probably neutrinos have non-zero masses and non-trivial mixings. This belief is based primarily on the evidence for neutrino mixings and oscillations from the data on atmospheric neutrinos and on solar neutrinosbarger .
If this is true, then in general the heavier neutrinos are expected to decay into the lighter ones via flavor changing processes. The only questions are (a) whether the lifetimes are short enough to be phenomenologically interesting (or are they too long?) and (b) what are the dominant decay modes.
Throughout the following discussion, to be specific, I will assume that the neutrino masses are at most of order of a few eV. There are interesting things to be done with heavier decaying neutrinos which we will not discuss here todaykawasaki .
## II Radiative Decays
For eV neutrinos, the only radiative decay modes possible are $`\nu _i\nu _j+\gamma `$. They can occur at one loop level in SM (Standard Model). The decay rate is given bypetcov :
$$\mathrm{\Gamma }=\frac{9}{16}\frac{\alpha }{\pi }\frac{G_F^2}{128\pi ^3}\frac{\left(\delta m_{ij}^2\right)^3}{m_i}\left|\underset{\alpha }{}U_{i\alpha }^{}U_{\alpha j}\left(\frac{m_\alpha ^2}{m_W^2}\right)\right|^2$$
(1)
where $`\delta m_{ij}^2=m_i^2m_j^2`$ and $`\alpha `$ runs over $`e,\mu `$ and $`\tau `$. When $`m_im_j,m_iO(eV)`$ and for maximal mixing $`(4U_{i\alpha }^2U_{\alpha j}^2)O(1)`$ and $`(\alpha \tau )`$ one obtains for $`\mathrm{\Gamma }`$
$$\mathrm{\Gamma }_{SM}10^{45}sec^1$$
(2)
which is far too small to be interesting. If the dominant contribution is from electron (rather than $`\tau `$) then there is enhancement in matter (due to the electrons present) by a factornieves
$$10^{24}\left(\frac{\rho _e}{10^{24}(cc)^1}\right)\left(\frac{1eV}{m_i}\right)^4F$$
(3)
where the function $`F`$ tends to $`4m_i/E_i`$ for relativistic $`\nu _i^{}s_i`$ and this enhancement can be as large as $`10^{16}`$.
The decay mode $`\nu _i\nu _j+\gamma `$ comes from an effective coupling which can be written as:
$$\left(\frac{e}{m_i+m_j}\right)\overline{\psi }_j\sigma _{\mu \nu }(C+D\gamma _5)\psi _iF{}_{\mu }{}^{}\nu $$
(4)
Let us define $`k_{ij}`$ as
$`k_{ij}`$ $`=`$ $`\left({\displaystyle \frac{e}{m_i+m_j}}\right)\sqrt{C^2+D^2}`$
$``$ $`k_0\mu _B`$
where $`\mu _B=e/2m_e`$. Since the experimental bounds on $`\mu _{\nu i}`$, the magnetic moments of neutrinos, come from reactions such as $`\nu _eee\mathrm{`}\mathrm{`}\nu ^{\prime \prime }`$ which are not sensitive to the final state neutrinos; the bounds apply to both diagonal as well as transition magnetic moments and so can be used to limit $`k_0^i`$ and the corresponding lifetimes. The current bounds areparticle :
$`k_0^e<10^{10}`$ (6)
$`k_0^\mu <\mathrm{7.4.10}^{10}`$
$`k_0^\tau <\mathrm{5.4.10}^7`$
For $`m_im_j`$, the decay rate for $`\nu _i\nu _j+\gamma `$ is given by
$$\mathrm{\Gamma }=\frac{\alpha }{2m_e^2}m_i^3k_0^2$$
(7)
This, in turn, gives indirect bounds on radiative decay lifetimes for $`O(eV)`$ neutrinos of:
$`\tau _{\nu _e}>5.10^{18}\text{sec}`$ (8)
$`\tau _{\nu _\mu }>5.10^{16}\text{sec}`$
$`\tau _{\nu _\tau }>2.10^{11}\text{sec}`$
There is one caveat in deducing these bounds. Namely, the form factors C and D are evaluated at $`q^2O(eV^2)`$ in the decay matrix elements whereas in the scattering from which the bounds are derived, they are evaluated at $`q^2O(MeV^2)`$. Thus, some extrapolation is necessary. It can be argued that, barring some bizarre behaviour, this is justifiedfrere .
There are other bounds on radiative lifetimes of neutrinos which are all based on non-observation of the final state $`\gamma `$-ray. The most direct observational bounds from reactor and accelerators areparticle :
$`\tau (\nu _e)>300s`$ (9)
$`\tau (\nu _\mu )>15.4s`$
There are other more indirect boundsparticle . From cosmology:
$$\tau >2.10^{21}s.$$
(10)
From x-ray and $`\gamma `$-ray fluxes:
$$\tau >7.10^9s.$$
(11)
Here we have set $`m_iO(eV)`$. All these bounds depend on assuming that $`m_im_j`$ in the mode $`\nu _i\nu _j+\gamma `$. The bounds are not valid if there is near degeneracy; since in this case the $`\gamma `$-ray can be soft.
## III The Sciama Model
Dennis Sciama proposed an intriguing and imaginative scenario for radiative decay of a neutrino about 10 years agosciama . It went thru several metamorphoses; starting out as a decay mode $`\nu _\tau \nu _\mu +\gamma `$ and ending up most recently as $`\nu _{st}\nu _{st}^{}+\gamma `$mohapatra .
The proposal was this: (a) suppose the decay $`\nu _\alpha \nu _\beta +\gamma `$ takes place, (b) the mass of $`\nu _\alpha (m_{\nu \alpha })`$ is $`27.4`$ eV, and $`m_{\nu _\alpha }m_{\nu _\beta },`$ (c) thus $`E_\gamma 13.7`$ eV for a slow moving $`\nu _\alpha `$; (d) the lifetime by this mode for $`\nu _\alpha `$ is about $`2.10^{23}`$ sec. These properties are consistent with all the bounds. The purpose was to account for the anomalous amount of ionised $`H`$ in interstellar space. The energy of the decay photon is just right to ionise hyrogen; and the lifetime is chosen so as to provide the required amount of ionization.
After a lengthy innings (of 10 years), it appears that the Sciama model is finally ruled out. The most recent evidence against it is the non-observation of the line at 911 Å(corresponding to 13.7 eV) with the predicted intensitybowyer .
## IV Invisible Decays
A decay mode with essentially invisible final states which does not involve any new particles is the three body neutrino decay mode. The decay is $`\nu _i\nu _j\nu _j\overline{\nu }_j`$. This decay, like the radiative mode, can occur at one loop level in SM. With a mass pattern $`m_im_j`$ the decay rate can be written as
$$\mathrm{\Gamma }=\frac{ϵ^2G_F^2m_j^5}{192\pi ^3}$$
(12)
In the SM at one loop level, with the internal $`\tau `$ dominating, the value of $`ϵ^2`$ is given bylee
$$ϵ_{SM}^2=\frac{3}{16}\left(\frac{\alpha }{\pi }\right)^2\left(\frac{m_\tau }{m_W}\right)^4\left\{\mathrm{}n\left(\frac{m_\tau ^2}{m_w^2}\right)\right\}^2\left(U_{\tau j}U_{\tau i}^{}\right)^2$$
(13)
With maximal mixing $`ϵ_{SM}^23.10^{12}`$. Even if $`ϵ`$ were as large as 1 with new physics contributions; it only gives a value for $`\mathrm{\Gamma }`$ of $`5.10^{35}sec^1`$. Hence, this decay mode will not yield decay rates large enough to be of interest. It is worth noting that the current experimental bound on $`ϵ`$ is quite poor: $`ϵ<O(100)`$bilenky .
## V $`\nu _{\alpha _L}\nu _{\beta _L}+\chi `$
If the only new particle introduced is an I=0, L=0, J=0, massless particle such as a Goldstone boson (familon), then a new decay mode is possible:
$$\nu _{\alpha _L}\nu _{\beta _L}+\chi $$
(14)
which arises from a coupling of the form
$$g_p\overline{\psi }_{\beta _L}\gamma _\mu \psi _{\alpha _L}_\mu \chi $$
(15)
with a decay rate
$$\mathrm{\Gamma }=\frac{g_p^2m_\alpha ^3}{16\pi }$$
(16)
$`SU(2)_L`$ symmetry predicts a similar coupling for the charged leptons with decay mode $`\mathrm{}_\alpha \mathrm{}_\beta +\chi `$. Thus the $`\nu _\alpha `$ lifetime is related to the B.R. $`(\mathrm{}_\alpha \mathrm{}_\beta +\chi )`$ by
$$\tau _{\nu _\alpha }=\frac{\tau _\mathrm{}\alpha }{B.R.(\mathrm{}_\alpha \mathrm{}_\beta +\chi )}\left(\frac{m_{\nu _\alpha }}{m_\mathrm{}_\alpha }\right)^3$$
(17)
The current bounds on $`\mu `$ and $`\tau `$ branching ratios areparticle ; jodidio
$`B.R.(\mu e\chi )<2.10^6`$ (18)
$`B.R.(\tau \mu \chi )<7.10^6`$
and lead to
$`\tau _{\nu _\mu }>10^{24}s`$ (19)
$`\tau _{\nu _\tau }>10^{20}s.`$
Coupling to a scalar field with $`I=1/2`$, $`L=0`$ would have the same constraint as above; and in addition would require fine tuning to avoid mixing with the SM Higgs.
## VI Fast Invisible Decays
The only possibility for fast invisible decays of neutrinos seems to lie with Majoron models. There are two classes of models; the I=1 Gelmini-Roncadelligelmini majoron and the I=0 Chikasige-Mohapatra-Pecceichikasige majoron. In general, one can choose the majoron to be a mixture of the two; furthermore the coupling can be to flavor as well as sterile neutrinos. The effective interaction is of the form:
$$g_\alpha \overline{\nu }_\beta ^c\nu _\alpha J$$
(20)
giving rise to decay:
$$\nu _\alpha \overline{\nu }_\beta +J$$
(21)
where $`J`$ is a massless $`J=0L=2`$ particle; $`\nu _\alpha `$ and $`\nu _\beta `$ are mass eigenstates which may be mixtures of flavor and sterile neutrinos. Models of this kind which can give rise to fast neutrino decays and satisfy the bounds below have been discussed by Valle, Joshipura and othersvalle . These models are unconstrained by $`\mu `$ and $`\tau `$ decays which do not arise due to the $`\mathrm{\Delta }L=2`$ nature of the coupling. The I=I coupling is constrained by the bound on the invisible $`Z`$ widthkawasaki ; and requires that the Majoron be a mixture of I=1 and I=0. The couplings of $`\nu _\mu `$ and $`\nu _e(g_\mu `$ and $`g_e)`$ are constrained by the limits on multi-body $`\pi ,k`$ decays $`\pi \mu \nu \nu \nu `$ and $`K\mu \nu \nu \nu `$ and on $`\mu e`$ university violation in $`\pi `$ and K decaysbarger1 .
Granting that models with fast, invisible decays of neutrinos can be constructed, can such decay modes be responsible for any observed neutrino anomaly?
We assume a component of $`\nu _\alpha ,`$ i.e., $`\nu _2`$, to be the only unstable state, with a rest-frame lifetime $`\tau _0`$, and we assume two flavor mixing, for simplicity:
$$\nu _\alpha =cos\theta \nu _2+sin\theta \nu _1$$
(22)
with $`m_2>m_1`$. From Eq. (2) with an unstable $`\nu _2`$, the $`\nu _\alpha `$ survival probability is
$`P_{\alpha \alpha }`$ $`=`$ $`sin^4\theta +cos^4\theta \mathrm{exp}(\alpha L/E)`$
$`+`$ $`2sin^2\theta cos^2\theta \mathrm{exp}(\alpha L/2E)cos(\delta m^2L/2E),`$
where $`\delta m^2=m_2^2m_1^2`$ and $`\alpha =m_2/\tau _0`$. Since we are attempting to explain neutrino data without oscillations there are two appropriate limits of interest. One is when the $`\delta m^2`$ is so large that the cosine term averages to 0. Then the survival probability becomes
$$P_{\mu \mu }=sin^4\theta +cos^4\theta \mathrm{exp}(\alpha L/E)$$
(24)
Let this be called decay scenario A. The other possibility is when $`\delta m^2`$ is so small that the cosine term is 1, leading to a survival probability of
$$P_{\mu \mu }=(sin^2\theta +cos^2\theta \mathrm{exp}(\alpha L/2E))^2$$
(25)
corresponding to decay scenario B. Decay models for both kinds of scenarios can be constructed; although they require fine tuning and are not particularly elegant.
The possibility of solar neutrinos decaying to explain the discrepancy is a very old suggestionpakvasa . The most recent analysis of the current solar neutrino data finds that no good fit can be found: $`U_{ei}0.6`$ and $`\tau _\nu `$ (E=10 MeV) $`6`$ to 27 sec. come closestacker . The fits become acceptable only if the suppression of the solar neutrinos is energy independent as proposed by several authorsharrison (which is possible if the Homestake data are excluded from the fit). The above conclusions are valid for both the decay scenarios A as well as B.
For atmospheric neutrinos, assuming neutrino decay scenario A, it was found that it is possible to choose $`\theta `$ and $`\alpha `$ to provide a good fit to the Super-Kamiokande L/E distributions of $`\nu _\mu `$ events and $`\nu _\mu /\nu _e`$ event ratiobarger2 . The best-fit values of the two parameters are $`cos^2\theta 0.87`$ and $`\alpha 1GeV/D_E,`$ where $`D_E=12800`$ km is the diameter of the Earth. This best-fit $`\alpha `$ value corresponds to a rest-frame $`\nu _2`$ lifetime of
$$\tau _0=m_2/\alpha \frac{m_2}{(1eV)}\times 10^{10}s.$$
(26)
However, it was then shown that the fit to the higher energy events in Super-K (especially the upcoming muons) is quite poorlipari . The reason that the inclusion of high energy upcoming muon events makes the fits poorer is very simple. In the decay A scenario, due to time dilation, the decay is suppressed at high energy and there is hardly any depletion of $`\nu _\mu ^{}s`$, contrary to observation.
Turning to decay scenario B, consider the following possibilitybarger3 . The three states $`\nu _\mu ,\nu _\tau ,\nu _s`$ (where $`\nu _s`$ is a sterile neutrino) are related to the mass eigenstates $`\nu _2\nu _3\nu _4`$ by the approximate mixing matrix.
$$\left(\begin{array}{c}\nu _\mu \\ \nu _\tau \\ \nu _s\end{array}\right)=\left(\begin{array}{ccc}\mathrm{cos}\theta & \mathrm{sin}\theta & 0\\ \mathrm{sin}\theta & \mathrm{cos}\theta & 0\\ 0& 0& 1\end{array}\right)\left(\begin{array}{c}\nu _2\\ \nu _3\\ \nu _4\end{array}\right)$$
(27)
and the decay is $`\nu _2\overline{\nu }_4+J`$. The electron neutrino, which we identify with $`\nu _1`$, cannot mix very much with the other three because of the more stringent bounds on its couplingsbarger1 , and thus our preferred solution for solar neutrinos would be small angle matter oscillations.
In this case the $`\delta m_{23}^2`$ in Eq. (23) is not related to the $`\delta m_{24}^2`$ in the decay, and can be very small, say $`<10^4\mathrm{eV}^2`$ (to ensure that oscillations play no role in the atmospheric neutrinos). Then the oscillating term is 1 and $`P(\nu _\mu \nu _\mu )`$ is given by Eq. (25).
In order to compare the predictions of this model with the standard $`\nu _\mu \nu _\tau `$ oscillation model, we have calculated with Monte Carlo methods the event rates for contained, semi-contained and upward-going (passing and stopping) muons in the Super-K detector, in the absence of ‘new physics’, and modifying the muon neutrino flux according to the decay or oscillation probabilities discussed above. We have then compared our predictions with the SuperK data fukuda , calculating a $`\chi ^2`$ to quantify the agreement (or disagreement) between data and calculations. In performing our fit (see Ref. barger3 for details) we do not take into account any systematic uncertainty, but we allow the absolute flux normalization to vary as a free parameter $`\beta `$.
The decay model of Equation (24) above gives a very good fit with a minimum $`\chi ^2=33.7`$ (32 d.o.f.) for the choice of parameters
$$\tau _\nu /m_\nu =63\mathrm{km}/\mathrm{GeV},\mathrm{cos}^2\theta =0.30$$
(28)
and normalization $`\beta =1.17`$.
In Fig. 1 we compare the best fits of the two models considered (oscillations and decay) with the SuperK data. In the figure we show (as data points with statistical error bars) the ratios between the SuperK data and the Monte Carlo predictions calculated in the absence of oscillations or other form of ‘new physics’ beyond the standard model. In the six panels we show separately the data on $`e`$-like and $`\mu `$-like events in the sub-GeV and multi-GeV samples, and on stopping and passing upward-going muon events. The solid (dashed) histograms correspond to the best fits for the decay model ($`\nu _\mu \nu _\tau `$ oscillations). One can see that the best fits of the two models are of comparable quality. The reason for the similarity of the results obtained in the two models can be understood by looking at Fig. 2, where we show the survival probability $`P(\nu _\mu \nu _\mu )`$ of muon neutrinos as a function of $`L/E_\nu `$ for the two models using the best fit parameters. In the case of the neutrino decay model (thick curve) the probability $`P(\nu _\mu \nu _\mu )`$ monotonically decreases from unity to an asymptotic value $`\mathrm{sin}^4\theta 0.49`$. In the case of oscillations the probability has a sinusoidal behaviour in $`L/E_\nu `$. The two functional forms seem very different; however, taking into account the resolution in $`L/E_\nu `$, the two forms are hardly distinguishable. In fact, in the large $`L/E_\nu `$ region, the oscillations are averaged out and the survival probability there can be well approximated with 0.5 (for maximal mixing). In the region of small $`L/E_\nu `$ both probabilities approach unity. In the region $`L/E_\nu `$ around 400 km/GeV, where the probability for the neutrino oscillation model has the first minimum, the two curves are most easily distinguishable, at least in principle.
Decay Model
There are two decay possibilities that can be considered: (a) $`\nu _2`$ decays to $`\overline{\nu }_4`$ which is dominantly $`\nu _s`$ with $`\nu _2`$ and $`\nu _3`$ mixtures of $`\nu _\mu `$ and $`\nu _\tau `$, as in Eq. (27), and (b) $`\nu _2`$ decays into $`\overline{\nu }_4`$ which is dominantly $`\overline{\nu }_\tau `$ and $`\nu _2`$ and $`\nu _3`$ are mixtures of $`\nu _\mu `$ and $`\nu _s`$. In both cases the decay interaction has to be of the form
$$_{int}=g_{24}\overline{\nu _{4_L}^c}\nu _{2_L}J+h.c.$$
(29)
where $`J`$ is a Majoron field that is dominantly iso-singlet (this avoids any conflict with the invisible width of the $`Z`$). Viable models for both the above cases can be constructed valle ; joshipura . However, case (b) needs additional iso-triplet light scalars which cause potential problems with Big Bang Nucleosynthesis (BBN), and there is some preliminary evidence from SuperK against $`\nu _\mu `$$`\nu _s`$ mixing kajita . Hence we only consider case (a), i.e. $`\nu _2\overline{\nu }_4+J`$ with $`\nu _4\nu _s`$, as implicit in Eq. (27). With this interaction, the $`\nu _2`$ rest-lifetime is given by
$$\tau _2=\frac{16\pi }{g^2}\frac{m_2}{\delta m^2(1+x)^2},$$
(30)
where $`\delta m^2=m_2^2m_4^2`$ and $`x=m_4/m_2(0<x<1)`$. From the value of $`\alpha ^1=\tau _2/m_2=63`$ km/GeV found in the fit and for $`x=0`$, we have
$$g^2\delta m^20.16\mathrm{eV}^2$$
(31)
Combining this with the bound on $`g^2`$ from $`K\mu `$ decays of $`g^2<2.4\times 10^4`$ barger1 we have
$$\delta m^2>650\mathrm{eV}^2.$$
(32)
Even with a generous interpretation of the uncertainties in the fit, this $`\delta m^2`$ implies a minimum mass difference in the range of about 25 eV. Then $`\nu _2`$ and $`\nu _3`$ are nearly degenerate with masses $`\stackrel{}{>}𝒪`$(25 eV) and $`\nu _4`$ is relatively light. We assume that a similar coupling of $`\nu _3`$ to $`\nu _4`$ and J is somewhat weaker leading to a significantly longer lifetime for $`\nu _3`$, and the instability of $`\nu _3`$ is irrelevant for the analysis of the atmospheric neutrino data.
For the atmospheric neutrinos in SuperK, two kinds of tests have been proposed to distinguish between $`\nu _\mu `$$`\nu _\tau `$ oscillations and $`\nu _\mu `$$`\nu _s`$ oscillations. One is based on the fact that matter effects are present for $`\nu _\mu `$$`\nu _s`$ oscillationsbdppw but are nearly absent for $`\nu _\mu `$$`\nu _\tau `$ oscillationspanta leading to differences in the zenith angle distributions due to matter effects on upgoing neutrinos smirnov . The other is the fact that the neutral current rate will be affected in $`\nu _\mu `$$`\nu _s`$ oscillations but not for $`\nu _\mu `$$`\nu _\tau `$ oscillations as can be measured in events with single $`\pi ^0`$’s vissani . In these tests our decay scenario will behave as a hybrid in that there is no matter effect but there is some effect in neutral current rates.
Long-Baseline Experiments
The survival probability of $`\nu _\mu `$ as a function of $`L/E`$ is given in Eq. (1). The conversion probability into $`\nu _\tau `$ is given by
$$P(\nu _\mu \nu _\tau )=\mathrm{sin}^2\theta \mathrm{cos}^2\theta (1e^{\alpha L/2E})^2.$$
(33)
This result differs from $`1P(\nu _\mu \nu _\mu )`$ and hence is different from $`\nu _\mu `$$`\nu _\tau `$ oscillations. Furthermore, $`P(\nu _\mu \nu _\mu )+P(\nu _\mu \nu _\tau )`$ is not 1 but is given by
$$P(\nu _\mu \nu _\mu )+P(\nu _\mu \nu _\tau )=1\mathrm{cos}^2\theta (1e^{\alpha L/E})$$
(34)
and determines the amount by which the predicted neutral-current rates are affected compared to the no oscillations (or the $`\nu _\mu `$$`\nu _\tau `$ oscillations) case. In Fig. 3 we give the results for $`P(\nu _\mu \nu _\mu )`$, $`P(\nu _\mu \nu _\tau )`$ and $`P(\nu _\mu \nu _\mu )+P(\nu _\mu \nu _\tau )`$ for the decay model and compare them to the $`\nu _\mu `$$`\nu _\tau `$ oscillations, for both the K2Kwho and MINOSminos (or the corresponding European projectNGS ) long-baseline experiments, with the oscillation and decay parameters as determined in the fits above.
The K2K experiment, already underway, has a low energy beam $`E_\nu 1\text{}2`$ GeV and a baseline $`L=250`$ km. The MINOS experiment will have 3 different beams, with average energies $`E_\nu =3,`$ 6 and 12 GeV and a baseline $`L=732`$ km. The approximate $`L/E_\nu `$ ranges are thus 125–250 km/GeV for K2K and 50–250 km/GeV for MINOS. The comparisons in Figure 3 show that the energy dependence of $`\nu _\mu `$ survival probability and the neutral current rate can both distinguish between the decay and the oscillation models. MINOS and the European project may also have $`\tau `$ detection capabilities that would allow additional tests.
Big Bang Nucleosynthesis
The decay of $`\nu _2`$ is sufficiently fast that all the neutrinos ($`\nu _e,\nu _\mu ,\nu _\tau ,\nu _s`$) and the Majoron may be expected to equilibrate in the early universe before the primordial neutrinos decouple. When they achieve thermal equilibrium each Majorana neutrino contributes $`N_\nu =1`$ and the Majoron contributes $`N_\nu =4/7`$ kolb , giving and effective number of light neutrinos $`N_\nu =4\frac{4}{7}`$ at the time of Big Bang Nucleosynthesis. From the observed primordial abundances of <sup>4</sup>He and <sup>6</sup>Li, upper limits on $`N_\nu `$ are inferred, but these depend on which data are usedolive ; lisi ; burles . Conservatively, the upper limit to $`N_\nu `$ could extend up to 5.3 (or even to 6 if <sup>7</sup>Li is depleted in halo starsolive ).
Cosmic Neutrino Fluxes
Since we expect both $`\nu _2`$ and $`\nu _3`$ to decay, neutrino beams from distant sources (such as Supernovae, active galactic nuclei and gamma-ray bursters) should contain only $`\nu _e`$ and $`\overline{\nu }_e`$ but no $`\nu _\mu `$, $`\overline{\nu }_\mu `$, $`\nu _\tau `$ and $`\overline{\nu }_\tau `$. This is a very strong prediction of our decay scenario. We can compare the very different expectations for neutrino flavor mixes from very distant sources such as AGN’s or GRB’s. Let us suppose that at the source the flux ratios are typical of a beam dump, a reasonable assumption: $`N_{\nu _e}:N_{\nu _\mu }:N_{\nu _\tau }=1:2:0`$. Then, for the conventional oscillation scenario, when all the $`\delta m^2`$’s satisfy $`\delta m^2L/4E>>1`$, it turns out curiously enough that for a wide variety of choices of neutrino mixing matrices, the final flavor mix is the same, namely: $`N_{\nu _e}:N_{\nu _\mu }:N_{\nu _\tau }=1:1:1`$. In the case of the decay B scenario, as mentioned here, we have $`N_{\nu _e}:N_{\nu _\mu }:N_{\nu _\tau }=1:0:0.`$ The two are quite distinct. Techniques for determining these flavor mixes in future KM3 neutrino telescopes have been proposedlearned .
Reactor and Accelerator Limits
The $`\nu _e`$ is essentially decoupled from the decay state $`\nu _2`$ so the null observations from the CHOOZ reactor are satisfiedchooz . The mixings of $`\nu _\mu `$ and $`\nu _\tau `$ with $`\nu _s`$ and $`\nu _e`$ are very small, so there is no conflict with stringent accelerator limits on flavor oscillations with large $`\delta m^2`$ zuber .
Summary
In summary, neutrino decay remains a viable alternative to neutrino oscillations as an explanation of the atmospheric neutrino anomaly. The model consists of two nearly degenerate mass eigenstates $`\nu _2`$, $`\nu _3`$ with mass separation $`\stackrel{}{>}𝒪(25`$ eV) from another nearly degenerate pair $`\nu _1`$, $`\nu _4`$. The $`\nu _\mu `$ and $`\nu _\tau `$ flavors are approximately composed of $`\nu _2`$ and $`\nu _3`$, with a mixing angle $`\theta _{23}57^{}`$. The state $`\nu _2`$ is unstable, decaying to $`\overline{\nu }_4`$ and a Majoron with a lifetime $`\tau _210^{12}`$ sec. The electron neutrino $`\nu _e`$ and a sterile neutrino $`\nu _s`$ have negligible mixing with $`\nu _\mu ,\nu _\tau `$ and are approximate mass eigenstates ($`\nu _e\nu _1,\nu _s\nu _4)`$, with a small mixing angle $`\theta _{14}`$ and a $`\delta m_{41}^210^5\mathrm{eV}^2`$ to explain the solar neutrino anomaly. The states $`\nu _3`$ and $`\nu _4`$ are also unstable, but with $`\nu _3`$ lifetime somewhat longer and $`\nu _4`$ lifetime much longer than the $`\nu _2`$ lifetime. This decay scenario is difficult to distinguish from oscillations because of the smearing in both L and $`E_\nu `$ in atmospheric neutrino events. However, long-baseline experiments, where $`L`$ is fixed, should be able to establish whether the dependence of $`L/E_\nu `$ is exponential or sinusoidal. In this scenario only $`\nu _1`$ is stable. Thus, neutrinos of supernovae or of extra galactic origin would be almost entirely $`\nu _e`$. The contribution of the electron neutrinos and the Majorons to the cosmological energy density $`\mathrm{\Omega }`$ is negligible and not relevant for large scale structure formation.
## VII Conclusion
In general, when neutrinos have masses and do mix; they can decay as well as oscillate. Some neutrino anomalies may be caused by one or the other or both. Non-oscillatory explanations have to be ruled out experimentally. Eventually, invisible neutrino decays can be severely constrained when data from Long Baseline experiments with good resolution in L and E become available.
## Acknowledgements
I thank the organizers for the invitation to give this talk and for selecting the title. I thank all my collaborators: A. Acker, V. Barger, J. G. Learned, P. Lipari, M. Lusignoli and T.J. Weiler. This work is partially supported by the U.S.D.O.E. under grant DOE-FG-03-94ER40833.
|
warning/0004/hep-th0004135.html
|
ar5iv
|
text
|
# References
On an effective action of monopoles in Abelian projections of Yang-Mills theory
Sergei V. Shabanov <sup>1</sup><sup>1</sup>1on leave from Laboratory of Theoretical Physics, JINR, Dubna, Russia.
Department of Mathematics, University of Florida, Gainesville, FL-32611, USA.
## Abstract
A path integral description of an effective action of monopoles in Abelian projections of Yang-Mills theories is discussed and used to establish a projection independence of the effective action. A dynamic regime in which the effective dynamics may contain massive solitonic excitations is described.
Numerical simulations of lattice Yang-Mills theories show that certain topological defects, which occur upon a partial gauge fixing , play an important role in the infrared (nonperturbative) dynamics of Yang-Mills fields. A typical numerical experiment of this kind would involve the following steps. Given a set of Yang-Mills field configurations generated with the probability $`e^{S_W}`$, where $`S_W`$ is the Wilson action, one performs a gauge transformation of each configuration from the Wilson ensemble so that after the gauge transformation the configuration satisfies a certain gauge condition that breaks the original gauge group to its maximal Abelian subgroup. The most popular gauge is the so called maximal Abelian gauge which is achieved by minimizing the $`L_2`$ norm of non-Abelian components of Yang-Mills connections by means of suitable gauge transformations . Clearly, this procedure would leave the maximal Abelian subgroup of the gauge group unbroken. After the gauge fixing (or the Abelian projection), the non-Abelian components are removed from every configuration and only the Abelian components of the gauge-fixed Wilson ensemble are used to compute expectation values of some gauge invariant operators like, for instance, the Wilson loop. An interesting feature of such a numerical experiment is that, despite a substantial reduction of the degrees of freedom (removing the non-Abelian components of connections), the string tension obtained from the Abelian (gauge-fixed) ensemble is 92% of the full string tension (computed in the original Wilson ensemble) . This is known as the Abelian dominance.
Since this phenomenon does not occur if non-Abelian components are removed before the gauge fixing, it is clear that some relevant degrees of freedom have been transferred from non-Abelian components to the Abelian ones upon the gauge transformation which has been used to impose the gauge in the Wilson ensemble. Hence, the effective Abelian theory cannot be a usual Maxwell theory. The topology of the gauge group and its maximal Abelian subgroup are different. Therefore any gauge fixing which breaks SU(N) to U(1)<sup>N-1</sup> would have singularities. In particular, there are connections for which the gauge transformation that minimizes the $`L_2`$ norm of their non-Abelian components cannot be regular everywhere in spacetime. It is easy to show that after the gauge transformation the Abelian components of such connections would contain Dirac monopoles as topological defects ,. Furthermore one can separate numerically monopole (singular) and photon (regular) parts of the Abelian fields and use only the “monopole” ensemble to compute the expectation value of the Wilson loop and the string tension. The “monopole” string tension differs from the “Abelian” one only by 5% . Thus, the topological defects play the major role in nonperturbative Yang-Mills theory in the maximal Abelian gauge. There is a strong numerical evidence that this occurs in other Abelian projection gauges . This phenomenon is called the monopole dominance.
It is well known that the physical configuration space, the space of connections modulo gauge transformations (the orbit space), has a non-Euclidean geometry and topology. When computing the functional integral over the physical configuration space, one usually uses some local coordinates on it. The coordinates on the orbit space can be obtained, for example, by a gauge fixing. Thanks to the nontrivial topology of gauge orbits, the coordinate system does not exist globally. Therefore any description based on local coordinates will always exhibit singularities. Although a “physical” interpretation of the coordinate singularities depends on the choice of coordinates (or, frankly speaking, on the choice of the gauge), they are inevitable in any coordinate description. Moreover, they have to be taken into account when computing the functional integral in order to obtain a correct (gauge invariant) spectrum of physical excitations in the theory . It should be stressed that, though the coordinate singularities do depend on the choice of local coordinates, their inevitable existence is essentially due to a non-Euclidean structure of the physical configuration (or phase) space which is gauge independent . That is, the geometry of the orbit space reveal itself through singularities in any coordinate description of the dynamics on the orbit space.
Consider a local function $`\varphi (A)`$ of the Yang-Mills connection $`A_\mu `$ which transforms in the adjoint representation under gauge transformations: $`\varphi (A)\mathrm{\Omega }\varphi (A)\mathrm{\Omega }^{}`$. An Abelian projection can be made by the gauge transformation $`A_\mu A_\mu ^\mathrm{\Omega }=\mathrm{\Omega }A_\mu \mathrm{\Omega }^{}+i\mathrm{\Omega }_\mu \mathrm{\Omega }^{}`$ where for every $`A_\mu `$ the gauge group element $`\mathrm{\Omega }`$ is determined by the condition that $`\varphi (A^\mathrm{\Omega })=\mathrm{\Omega }\varphi (A)\mathrm{\Omega }^{}`$ is an element of the Cartan algebra ($`\mathrm{\Omega }`$ diagonalizes $`\varphi (A)`$ in a matrix representation). In other words, we impose the gauge condition that off-diagonal components of $`\varphi (A)`$ should be zero for every $`A_\mu `$. The Faddeev-Popov determinant in this gauge is easy to compute $`\mathrm{\Delta }_{FP}[A]=_xdet(\mathrm{ad}\varphi (A))^2`$, where the adjoint operator $`\mathrm{ad}\varphi `$ acts on any element $`y`$ of the Lie algebra as $`\mathrm{ad}\varphi y=i[\varphi ,y]`$. When computing the determinant, the zero modes of the operator $`\mathrm{ad}\varphi `$ associated with the residual Abelian gauge symmetry must be removed. Note that the maximal Abelian subgroup of the gauge group is isomorphic to the stationary group of $`\varphi `$. One can show that the gauge group element $`\mathrm{\Omega }`$ is singular whenever $`det(\mathrm{ad}\varphi (A))^2`$ vanishes at some points of spacetime and the Abelian components of the projected configuration $`A_\mu ^\mathrm{\Omega }`$ would contain Dirac monopoles at those points. Thus, if an Abelian projection gauge is used to establish local coordinates on the gauge orbit space, the “monopole” configurations would generally appear as singular points of this coordinate system where the Faddeev-Popov determinant vanishes.
The results of the aforementioned numerical simulations show that the dynamics of Yang-Mills field configurations, that look like Dirac magnetic monopoles after an Abelian projection, captures essential features of quantum Yang-Mills theory at large distances (in the infrared limit). A theoretical challenge is to derive an effective theory for such degrees of freedom from the first principles. The conventional functional integral in the Abelian projection gauge cannot be used as a starting point because it is ill-defined ($`\mathrm{\Delta }_{FP}=0`$) at relevant configurations. Our goal is to develop the path integral formalism in which this problem is circumvented.
We begin with a trivial observation that the coordinate singularities depend on the choice of a gauge and by changing the gauge they can be moved away from a configuration space region of interest. Suppose we have a general parameterization of the Yang-Mills connections which, upon an Abelian projection, become purely Abelian configurations containing monopole-like topological defects. An effective action of such configurations can be computed by the functional integral in a gauge in which the “monopoles” configurations are no longer coordinate singularities. A simple analogy can be made with an ordinary integral over a sphere. If the origin of the coordinate system is chosen to be on the north pole of the sphere, then the south pole is the singular point (the point where two geodesics through the origin intersect again). Suppose the integrand is such that the stationary point approximation near the south pole gives a good estimate of the integral. Clearly, to compute the integral, it is more convenient to change the coordinates so that the singular point will be away from the south pole (e.g., by moving the origin to the south pole).
We shall elaborate this idea with an example of SU(2). A generalization to the case of SU(N) is straightforward . Let $`𝒏_0=(0,0,1)`$ be a unit isotopic vector. We are looking for connections $`𝑨_\mu `$ which can be transformed to configurations of the form $`𝒏_0A_\mu `$ by a suitable gauge transformation. Clearly, such connections would have six functional parameters, four in $`A_\mu `$ and the other two are associated with parameters of gauge transformations from SU(2)/U(1), where the group U(1)$``$SO(2) is the stationary group of $`𝒏_0`$. The latter two parameters are unified into a unit isotopic vector $`𝒏`$, $`𝒏^2=1`$. By analyzing generic (singular) gauge transformations of the connection $`𝒏_0A_\mu `$, we find that the connection
$$𝑨_\mu =𝒏\times _\mu 𝒏+𝒏C_\mu $$
(1)
can be made purely Abelian by a suitable gauge transformation, $`𝑨_\mu 𝒏_0(C_\mu +C_\mu ^m)`$, where $`C_\mu ^m=𝒏_0(_\mu 𝝃\times 𝝃)`$ and $`𝝃=(\mathrm{sin}(\theta /2)\mathrm{cos}\phi ,\mathrm{sin}(\theta /2)\mathrm{sin}\phi ,\mathrm{cos}(\theta /2))`$ if $`𝐧=(\mathrm{sin}\theta \mathrm{cos}\phi `$,$`\mathrm{sin}\theta `$ $`\mathrm{sin}\phi `$, $`\mathrm{cos}\theta )`$. If we take, for example, $`𝒏=𝒙/r`$, $`r^2=𝒙^2`$, the first term in (1) will be a vector potential of the Wu-Yang monopole. Upon the Abelian projection it produces an addition $`C_\mu ^m`$ to the regular Maxwell potential $`C_\mu `$ which is the vector potential of the Dirac magnetic monopole localized at the origin and with the Dirac string extended along a negative part of the $`z`$-axis. In a similar way one can also parameterize “Abelian” connections in the SU(N) gauge theory which would generate monopoles upon an Abelian projection.
To develop an effective theory for the field $`𝒏`$, one needs a change of variables in the space of connections. That is, six more functional variables have to be added to the connection (1):
$$𝑨_\mu =𝒏\times _\mu 𝒏+𝒏C_\mu +𝑾_\mu ,$$
(2)
where $`𝑾_\mu 𝒏=0`$ and $`𝑾_\mu `$ contain only six independent variables. There are infinitely many ways to parameterize $`𝑾_\mu `$ by six variables. It can be done either implicitly or explicitly . In general, one can say that $`𝑾_\mu `$ should satisfy two more conditions
$$𝝌(𝑾,𝒏,C)=0,𝒏𝝌0.$$
(3)
For example, one can choose
$$𝝌=_\mu 𝐖_\mu +C_\mu 𝐧\times 𝐖_\mu +𝐧(_\mu 𝐧𝐖_\mu )=0.$$
(4)
The condition (4) means that $`_\mu 𝑾_\mu =0`$ where the covariant derivative $`_\mu `$ is taken for the connection (1). An example of an explicit parameterization can be found if we unify six independent components of $`𝑾_\mu `$ into an antisymmetric tensor $`W_{\mu \nu }=W_{\nu \mu }`$. Then one can set
$$𝑾_\mu =W_{\mu \nu }𝒏\times _\nu 𝒏.$$
(5)
It is easy to find $`𝝌(𝑾,𝒏)=0`$ whose solution is given by (5). The analysis can be extended to the SU(N) case .
It is rather straightforward to show that the “monopole” configurations of the Wilson ensemble in the maximal Abelian gauge are described by the vector potential $`C_\mu ^m(𝒏)`$ introduced after Eq.(1) if $`𝑾_\mu `$ in the change of variables (2) satisfies the condition (4). In general, for any Abelian projection one can find a parameterization of $`𝑾_\mu `$ such that the monopole defects are always described by $`C_\mu ^m(𝒏)`$ . In the new variables (2), an Abelian projection is described by the gauge $`𝒏=𝒏_0`$ which is singular if $`C_\mu ^m(𝒏)`$ carries Dirac monopoles, i.e., this gauge does not exist everywhere in spacetime. Moreover the dynamics favors configurations which become coordinate singularities in this gauge.
Since (2) is a change of variables, the Wilson ensemble $`𝑨_\mu `$ can be used to generate the ensembles of $`C_\mu [𝑨]`$, $`𝑾_\mu [𝑨]`$ and $`𝒏[𝑨]`$. The inverse transformation is nonlocal so $`𝒏`$ is a nonlocal functional of $`𝑨_\mu `$. In principle, the effective action of $`𝒏`$ can be computed numerically by means of the inverse Monte-Carlo method . Given an ensemble of $`𝒏`$ one could try to find the probability which generates it. This method has already been used to compute an effective action of the monopole current in lattice gauge theories .
In a theoretical analysis, the change of variables (2) in the functional integral allows one to avoid the singularities of the Faddeev-Popov action in an Abelian projection gauge. The idea is to choose a gauge so that the Faddeev-Popov determinant does not vanish for the “monopole” configurations $`𝑨_\mu =𝒏\times _\mu 𝒏`$. To compute the Faddeev-Popov determinant, one has to find the gauge transformation law in the new variables. An infinitesimal gauge transformation of the SU(2) connection reads
$$\delta 𝑨_\mu =_\mu (𝑨)𝝋=_\mu 𝝋+𝑨_\mu \times 𝝋.$$
(6)
From (2) we infer
$$C_\mu =𝒏𝑨_\mu ,𝑾_\mu =𝒏\times _\mu (𝑨)𝒏.$$
(7)
Substituting these relations into (3) and solving them for $`𝒏`$ (two equations for two independent variables in $`𝒏`$), we find $`𝒏=𝒏(𝑨)`$. The latter together with (7) specifies the inverse change of variables. Let $`\delta 𝒏`$ be an infinitesimal gauge transformation of $`𝒏`$. Then from (7) and (6) we deduce that
$`\delta C_\mu `$ $`=`$ $`𝑨_\mu (\delta 𝒏𝒏\times 𝝋)+𝒏_\mu 𝝋,`$ (8)
$`\delta 𝑾_\mu `$ $`=`$ $`𝑾_\mu \times 𝝋𝒏[𝑾_\mu (\delta 𝒏𝒏\times 𝝋)]+𝒏\times _\mu (\delta 𝒏𝒏\times 𝝋),`$ (9)
where we have used that $`𝒏\delta 𝒏=0`$. An explicit form of $`\delta 𝒏`$ can be found from the equation $`\delta \overline{𝝌}(𝒏,𝑨)=0`$ where $`\overline{𝝌}(𝒏,𝑨)`$ is obtained by a substitution of (7) into $`𝝌(𝑾,C,𝒏)`$. We emphasize that $`\delta 𝒏`$ is determined by the choice of $`𝝌`$ and so are $`\delta C_\mu `$ and $`\delta 𝑾_\mu `$.
Let us assume we have a gauge (e.g., a Lorentz gauge) in which the “monopole” configurations $`𝒏\times _\mu 𝒏`$ are not coordinate singularities, that is, the Faddeev-Popov determinant $`\mathrm{\Delta }_{FP}[𝑨]`$ does not vanish for $`𝑨_\mu =𝒏\times _\mu 𝒏`$. This would allow us to develop a perturbation theory for computing an effective action of the field $`𝒏`$. After the change of variables (2) in the integral
$$𝒵D𝑨_\mu \mathrm{\Delta }_{FP}[𝑨]e^S,$$
(10)
where $`S`$ is the Yang-Mills action with a gauge fixing term, we can integrate out $`C_\mu `$ and $`𝑾_\mu `$ by perturbation theory. However, the Jacobian of the change of variables may also be a source of singularities if it vanishes for the “monopole” configurations. To compute the Jacobian, consider the identity
$$1=D𝒏\mathrm{\Delta }[𝑨,𝒏]\delta (\overline{𝝌}),\mathrm{\Delta }[𝑨,𝒏]=det[\delta \overline{𝝌}/\delta 𝒏].$$
(11)
Next, we insert the identity into (10) and change the integration variables $`𝑨_\mu `$ $`𝒏C_\mu +𝑾_\mu `$ with a generic $`𝑾_\mu `$ perpendicular to $`𝒏`$ so that $`D𝑨_\mu DC_\mu D𝑾_\mu `$. Finally, we shift the integration variables $`𝑾_\mu 𝑾_\mu +𝒏\times _\mu 𝒏`$ with the result
$$𝒵D𝒏DC_\mu D𝑾_\mu \mathrm{\Delta }[𝑨,𝒏]\mathrm{\Delta }_{FP}[𝑨]\delta [𝝌(𝑾,C,𝒏)]e^S.$$
(12)
where $`𝑨_\mu `$ is to be replaced by (2). This completes the path integral representation of Yang-Mills theory in the new variables (2). The determinant $`\mathrm{\Delta }[𝑨,𝒏]`$ depends on the choice of $`𝝌`$. The choice of $`𝝌`$ must be such that $`\mathrm{\Delta }[𝑨,𝒏]`$ does not vanish for $`𝑨_\mu =𝒏\times _\mu 𝒏`$.
Various choices of $`𝝌`$ can be regarded as gauge fixing conditions for the gauge symmetry associated with a reparameterization of $`𝑾_\mu `$. Note that Eq.(2) contains 14 functions in the right-hand side, while there are only 12 components in $`𝑨_\mu `$. Therefore the gauge transformations (6) would, in general, be induced by five-parametric transformations of the new variables. There are two-parametric transformations of the new variables under which $`𝑨_\mu `$ remains invariant. Precisely this gauge freedom is fixed by (3) and by the corresponding delta function in (12).
The delta function in (12) can be put into the exponential by means of the usual procedure of averaging over the gauge condition with the result $`SS+𝑑x𝝌^2/2`$. The determinants can also be lifted up to the exponential by introducing the ghost variables. The effective action obtained in such a way will be invariant under five-parametric BRST transformations. This symmetry can be used to show that the effective action of the field $`𝒏`$ does not depend on the choice of $`𝝌`$ in every order of perturbation theory for a rather large class of $`𝝌`$’s that includes all Abelian projections in which topological defects occur in the the Abelian components (i.e., they are monopoles).
It is noteworthy that one can give rather general arguments that the gradient expansion of the effective action should have the form (in the leading order)
$$S_{eff}=𝑑x\left\{m^2(_\mu 𝐧)^2+H_{\mu \nu }^2\right\},$$
(13)
where $`H_{\mu \nu }=𝐧(_\mu 𝐧\times _\nu 𝐧)`$ and $`m`$ is a mass scale. The field theory (13) contains stable knot solitons. It might therefore be possible to describe a nonperturbative (glueball) spectrum of Yang-Mills theory by means of quantum theory of knot solitons . The mass scale $`m^2`$ can be expressed as an expectation value of the antisymmeric field $`W_{\mu \nu }`$ introduced in (5)
$$\left(_\mu W_{\nu \sigma }_\nu W_{\mu \sigma }\right)\left(_\mu W_{\nu \lambda }_\nu W_{\mu \lambda }\right)m^2\delta _{\sigma \lambda }.$$
(14)
The above procedure of constructing a path integral for the “monopole” degrees of freedom can be generalized to the SU(N) gauge group .
|
warning/0004/hep-th0004142.html
|
ar5iv
|
text
|
# The M-theory dual of a 3 dimensional theory with reduced supersymmetry
## I Introduction
In the framework of the AdS-CFT duality and of brane polarization , Polchinski and Strassler found a supergravity dual of a confining gauge theory by perturbing $`AdS_5\times S_5`$ with a 3 form field background. The AdS-CFT duality allowed them to extract information about a 4 dimensional $`𝒩=1`$ gauge theory. In particular they found a mapping between the gauge theory vacua and states corresponding to the D3 branes being polarized into NS5 and D5 branes.
We can apply the same philosophy in order to obtain the supergravity dual of a theory coming from perturbing the 3 dimensional $`𝒩=8`$ theory living on $`N`$ M2 branes. The approach is similar, but at some points subtle differences between string theory and M theory come to play a role.
As discussed in , this 3 dimensional theory is obtained in the IR (strongly coupled) limit of a 3 dimensional $`𝒩=8`$ SYM. We know that the strongly coupled theory has 8 scalars and 8 Majorana fermions. Of these, 6 scalars and 6 fermions already can be paired into 3 hypermultiplets in the UV (the D2 brane theory). In the strongly coupled limit, SO(8) symmetry is restored, so the other scalar pairs up with the dualized $`A_\mu `$ and the 2 other fermions into a hypermultiplet. We can give masses to these 4 hypermultiplets (which appear as fermion masses in the Lagrangian), preserving $`𝒩=2`$ supersymmetry.
The fermions transform in the 8 of the SO(8) R-symmetry group. Thus, a fermion mass transforms in the $`\mathrm{𝟑𝟓}_{}`$ of SO(8). By the AdS-CFT duality, giving mass to the fermions corresponds to turning on a nonnormalizable mode of the antiselfdual 4-form field strength in the 8 dimensional transverse space of the M2 branes.
In it was observed that in a background p+3 form field, Dp branes become “polarized”. The polarization was understood in 2 ways. In one picture the configuration with the D-branes spread on an $`S^2`$ with a p+2 brane charge was energetically favored to the configuration with the D-branes in the center in a background p+3 form field. In the other picture, the nonabelian scalars describing the position of the brane become noncommutative, which resulted in a p+2 brane charge. This was worked out in more detail in . In the 3 brane case, for example, noncommutative configurations of the 3 chiral multiplets describe the polarization into a 3+2=5 brane.
Based on the above, we expect M2 branes to polarized also, when placed in a field configuration which couples with a higher brane (which can only be the M5 brane). We can understand M2 brane polarization easily in the first way - an M5 brane of geometry $`R^3\times S^3`$ with M2 brane charge will have a supersymmetric minimum at a nonzero radius. This is what most of this paper will be on.
Unfortunately the degrees of freedom of the M2 brane are not known, so the second picture is elusive. The weakly coupled theory is irrelevant. It describes the polarization of D2 branes into D4 branes, which is a different subject to be treated on its own .
We can present at most a speculation of this type: for D3 branes, when we give mass to the 3 chiral multiplets the vevs become noncommutative. This can be interpreted as polarization into a 2-dimensional higher object. We expect on intuitive grounds that when given mass, 4 hypers becoming “noncommutative” (whatever that means if they are not matrices) represent somehow the polarization of a M2 brane into a 3 dimensional higher object. We will present a bit of support for this picture in chapter V.
## II Perturbations of $`AdS_4\times S^7`$
As it is by now standard lore, in the framework of the AdS-CFT duality, to each local CFT operator of dimension $`\mathrm{\Delta }`$ correspond one normalizable and one nonnormalizable solution of the supergravity field equation. The coefficient of the nonnormalizable solution corresponds to the coefficient of the operator in the Hamiltonian, while the coefficient of the normalizable one corresponds to the vev of this operator .
According to the AdS-CFT conjecture the conformal field theory living on a $`N`$ M2 branes in dual to M theory living in the geometry they create, for very large $`N`$. This geometry is
$$ds^2=Z^{2/3}\eta _{\mu \nu }dx^\mu dx^\nu +Z^{1/3}dx^idx^i$$
$$C_3^0=\frac{1}{Z}dx^0dx^1dx^2,F_4^0=dC_3^0,$$
$`(1)`$
where $`\mu ,\nu =0,1,2`$, $`i,j=3,\mathrm{},10`$. For the case when the branes are coincident, the geometry becomes $`AdS_4\times S^7`$, and:
$$Z=\frac{R^6}{r^6},R^6=32\pi ^2NM_{11}^6,$$
$`(2)`$
where $`M_{11}`$ is the 11-dimensional Planck mass.
We are interested in turning on a bulk field which corresponds to a fermion mass. The fermions in the theory living on the M2 branes transform in the 8 of the SO(8) R symmetry group. The operators $`\lambda _i\lambda _j\delta _{ij}\lambda ^2/8`$ are chiral and transforms in the $`\mathrm{𝟑𝟓}_{}`$ of this group. Thus, their dimension does not change, and their coefficient, transforming also in the $`\mathrm{𝟑𝟓}_{}`$ is a fermion mass . Therefore, the bulk field which we have to turn on is a field strength $`F_4^1`$ oriented perpendicular to the brane, and transforming in the $`\mathrm{𝟑𝟓}_{}`$ of SO(8) - which is anti self dual tensors.
The 11d supergravity 4-form field strength satisfies the equation of motion:
$$d_{11}F_4=\frac{1}{2}F_4F_4$$
$`(3)`$
The total field strength will contain both the background $`F_4^0`$ and the perturbation $`F_4^1`$. Thus $`F_4=F_4^0+F_4^1`$. We can reduce the 11 dimensional Hodge dual to an 8-dimensional one:
$$_{11}F_4^1=Z^1(F_4^1)dx^0dx^1dx^2$$
$`(4)`$
Combining (1)(3) and (4) we obtain the equation of motion for the perturbed field:
$$d[Z^1(F_4^1F_4^1)]=0.$$
$`(5)`$
Since $`F_4^1`$ can be written as the exterior derivative of a vector potential: $`F_4^1=dC_3^1`$, the Bianchi identity is simply $`dF_4^1=0`$.
### A Tensor Spherical Harmonics
As we explained in the previous section, a fermion mass corresponds to an anti-self-dual 4-tensor SUGRA background. Other operators which may be of interest in this theory transform in $`\mathrm{𝟑𝟓}_+`$ and correspond to antisymmetric self-dual 4-tensors backgrounds. Thus
$$T_{ijkl}\frac{1}{4!}ϵ_{ijkl}^{mnop}T_{mnop}=\pm T_{ijkl},$$
$`(6)`$
where the $`+`$ and $``$ are for $`\mathrm{𝟑𝟓}_+`$ respectively $`\mathrm{𝟑𝟓}_{}`$.To make an (anti) self dual 4-tensor field on the space transverse to the M2 brane transforming in the same way, we can use $`T_4`$ or combine it with the radius vector to form
$$V_{mnpq}=\frac{x^r}{r^2}[x_mT_{rnpq}+x_nT_{mrpq}+x_pT_{mnrq}+x_qT_{mnpr}].$$
$`(7)`$
The forms $`T_4`$ and $`V_4`$ are normalized:
$$T_4=\frac{1}{4!}T_{mnpq}dx^mdx^ndx^pdx^q,V_4=\frac{1}{4!}V_{mnpq}dx^mdx^ndx^pdx^q.$$
$`(8)`$
In addition we define
$$S_3=\frac{1}{3!}T_{mnpq}x^mdx^ndx^pdx^q.$$
$`(8)`$
Since the general perturbation will be a combination of $`T_4`$ and $`V_4`$ multiplied by a power of $`r`$, the following relations will be useful:
$$dS_3=4T_4,d(\mathrm{ln}r)S_3=V_4,d(r^pS_3)=r^p(4T_4+pV_4),$$
$$dT_4=0,dV_4=4d(\mathrm{ln}r)T_4,drV_4=0$$
$$T_4=\pm T_4,V_4=\pm (T_4V_4).$$
$`(9)`$
In order to relate fermion masses to tensors it is convenient to group the 8 fermions and the 8 transverse coordinates in complex pairs:
$$z^1=x^3+ix^7,z^2=x^4+ix^8,z^3=x^5+ix^9,z^4=x^6+ix^{10},$$
$`(10a)`$
Similarly the fermions can be “complexified”:
$$\mathrm{\Lambda }^1=\lambda ^1+i\lambda ^2,\mathrm{\Lambda }^2=\lambda ^3+i\lambda ^4,\mathrm{\Lambda }^3=\lambda ^5+i\lambda ^6,\mathrm{\Lambda }^4=\lambda ^7+i\lambda ^8.$$
$`(10b)`$
Under a rotation $`z^ie^{i\varphi _i}z^i`$ the fermions transform as :
$$\mathrm{\Lambda }^1e^{i(\varphi _1+\varphi _2+\varphi _3+\varphi _4)/2}\mathrm{\Lambda }^1$$
$$\mathrm{\Lambda }^2e^{i(\varphi _1\varphi _2\varphi _3+\varphi _4)/2}\mathrm{\Lambda }^2$$
$$\mathrm{\Lambda }^3e^{i(\varphi _1+\varphi _2\varphi _3+\varphi _4)/2}\mathrm{\Lambda }^3$$
$$\mathrm{\Lambda }^4e^{i(\varphi _1+\varphi _2+\varphi _3\varphi _4)/2}\mathrm{\Lambda }^4$$
$`(11)`$
If we give masses to the 4 complex fermions we preserve $`𝒩=2`$ supersymmetry . Thus the Lagrangian will be perturbed with:
$$\mathrm{\Delta }L=\mathrm{Re}(m_1\mathrm{\Lambda }_1^2+m_2\mathrm{\Lambda }_2^2+m_3\mathrm{\Lambda }_3^2+m_4\mathrm{\Lambda }_4^2)$$
$`(12)`$
This perturbation transforms under SO(8) like:
$$T=\mathrm{Re}(m_1d\overline{z}^1dz^2dz^3dz^4+m_2dz^1d\overline{z}^2dz^3dz^4+$$
$$+m_3dz^1dz^2d\overline{z}^3dz^4+m_4dz^1dz^2dz^3d\overline{z}^4)$$
$`(13)`$
Regardless of the masses, this perturbation is invariant under the discrete $`Z_2`$ symmetry:
$$z_1i\overline{z}_1,z_2i\overline{z}_2,z_3i\overline{z}_3,z_4i\overline{z}_4.$$
$`(14)`$
This symmetry has to do with the fact that our tensors are anti self dual. We will first be exploring the SO(4) symmetric configuration: $`m^1=m^2=m^3=m^4=m`$. We can easily check that in this case only two components of $`T`$ will be nonzero: $`T_{3456}=T_{\mathrm{789\; 10}}=4m`$. In chapter V we will be exploring generalizations to unequal masses.
### B Linearized Perturbations
As discussed in the previous subsection, the general form of the perturbation is
$$F_4^1=r^p(aT_4+bV_4)$$
$`(15)`$
We can use the Bianchi identity and (9) to simplify this to:
$$F_4^1=(a/4)r^p(4T_4+pV_4)=(a/4)d(r^pS_3).$$
Using the equation of motion (5) and (9) we obtain after a few steps:
$$p^2+14p+2424=0.$$
$`(16)`$
We can see that for $`\mathrm{𝟑𝟓}_{}`$ there are 2 solutions
$$p=6;F_4^1(R/r)^6[4T_46V_4],$$
$`(17a)`$
$$p=8;F_4^1(R/r)^8[4T_48V_4],$$
$`(17b)`$
Translating to an inertial frame, and remembering that the $`AdS_4`$ radius $`u=r^2`$, we can see that the first perturbation is nonnormalizable, and corresponds to turning on a fermion mass in the gauge theory, while the second one is normalizable and corresponds to the vev of $`\lambda \lambda `$. These perturbations correspond to a field which is $`AdS_4`$ pseudoscalar and $`S^7`$ 3-tensor, and which satisfies an $`S^7`$ equation of “self duality in odd dimension” type . The mass perturbation corresponds to:
$$F_4^1=\alpha (R/r)^6[4T_46V_4]=d\left(\alpha (R/r)^6S_3\right),$$
$`(18)`$
where $`\alpha `$ is the numerical constant that relates the boundary theory mass to the coefficient of the nonnormalizable bulk mode. From now on and throughout this paper we will be absorbing $`\alpha `$ into $`m`$.
## III M5 Brane Probes
In this section we consider a test M5 brane in the $`AdS_4\times S^7`$ geometry, perturbed with $`F_4^1`$ flux. This is a relatively simple problem which contains most of the “meat” of the more complicated problem - that of the M2 branes becoming polarized into M5 branes.
We will be first examining the case of 4 equal masses, which has an SO(4) symmetry between the 4 complex scalars. Our test M5 brane has the geometry $`R^3\times S^3`$, and has M2 brane charge $`n`$. Thus 3 of its directions are parallel to the M2 branes which create the geometry, while the other 3 are wrapped on an $`S^3`$ inside $`S^7`$.
Turning on a M2 brane charge (parallel to that of the source M2 branes) on the M5 brane is done by turning on a 3 form field strength flux $`F_3`$ on $`S^3`$. This can be done, but it is not so straightforward. The 3 form field strength on the M5 brane is self dual - so naively turning on a flux on the $`S^3`$ cannot be done without turning on a flux in the other 3 directions. Moreover, we are in a background of $`C_3`$, so the gauge invariant object in the brane theory is not $`F_3`$ but $`F_3C_3`$. Thus before proceeding with the computation we need to have a thorough understanding of the M5 brane action and of how self duality is achieved.
### A The M5 Brane Action
The action of an M5 brane is more complicated than that of D-branes because, as we said, the theory on the M5 brane contains a self dual 2 form field. We will only be interested in the bosonic part of this action. There are 2 approaches at writing an action for such a field.
The first approach, by Pasti Sorokin and Tonin consists in combining in a clever way the 3-form field with an auxiliary scalar field $`a`$ to form the action:
$$S_{PST}=d^6x[\sqrt{det(g_{mn}+i\stackrel{~}{H}_{mn})}+\sqrt{g}\frac{1}{4_ra^ra}_ma(H)^{mnp}H_{npq}^qa]$$
$$\left[C^6+\frac{1}{2}FC^3\right].$$
$`(19)`$
Here $`C^6`$ and $`C^3`$ are the pullbacks of the M-theory forms, $`F=dB`$ is the field strength living on the brane, $`H_{mnp}F_{mnp}C_{mnp}`$, $``$ represents the Hodge dual, and $`\stackrel{~}{H}_{mn}=(H)_{mnp}\frac{^pa}{\sqrt{_ra^ra}}`$.
The action has a Lorentz invariant form, and the self duality of $`H`$ is forced when integrating out the auxiliary field. The first term looks like a Born -Infeld term (and reduces to the normal Born - Infeld term for a D4 brane), the second term is a mixed term (which reduces to a part of the Wess-Zumino term of a D4 brane, but which unlike normal Wess-Zumino terms is not zero in the absence of background fields). The third term is a Wess-Zumino term. Using this approach the relative normalizations of the 3 terms in the action, and the generalized formula for background fields (19) can be easily found. Nevertheless, in order to compute anything using the first 2 terms one has to fix some of the gauge symmetries.
The second approach, by Perry and Schwarz consists in picking a special direction, and thus maintaining only 5-d explicit Lorenz invariance (although the theory secretly is 6-d Lorentz invariant). The 6d metric $`G_{\widehat{\mu }\widehat{\nu }}`$ contains 5d pieces $`G_{\mu \nu },G_{\mu 5}`$ and $`G_{55}`$. Hatted indices denote 6 dimensional quantities, and unhatted ones represent 5 dimensional ones. The self dual antisymmetric tensor is represented by a $`5\times 5`$ antisymmetric tensor $`B_{\mu \nu }`$, and its curl $`F_{\mu \nu \rho }=_\mu B_{\nu \rho }+_\nu B_{\rho \mu }+_\rho B_{\mu \nu }`$. The action obtained is
$$S_{PS}=d^6x(L_1+L_2+L_3),$$
where
$$L_1=\sqrt{det\left(G_{\widehat{\mu }\widehat{\nu }}+\frac{iG_{\widehat{\mu }\rho }G_{\widehat{\nu }\lambda }\stackrel{~}{H}^{\rho \lambda }}{\sqrt{G_5}}\right)}$$
$$L_2=\frac{1}{4}\stackrel{~}{H}^{\mu \nu }_5B_{\mu \nu },$$
$$L_3=\frac{1}{8}ϵ_{\mu \nu \rho \sigma \lambda }\frac{G^{5\rho }}{G^{55}}\stackrel{~}{H}^{\mu \nu }\stackrel{~}{H}^{\sigma \lambda },$$
$`(20)`$
where $`\stackrel{~}{H}^{\mu \nu }=\frac{1}{3!}ϵ^{\mu \nu \rho \sigma \lambda }F_{\rho \sigma \lambda }`$, and $`G_5=detG_{\mu \nu }`$, and the $`ϵ`$ symbol is purely numerical.
This action can be used directly for explicit computations, and can be obtained from the PST action with no external background upon fixing $`_\mu a=\delta _\mu ^5`$, and $`B_{\mu 5}=0`$ . $`L_1`$ is obtained from the Born-Infeld term, and $`L_2+L_3`$ are obtained from the mixed term. Self duality (which in this approach appears as an equation of motion) in the limit of a free theory in the gauge $`B_{\mu 5}=0`$ is $`\stackrel{~}{H}^{\mu \nu }=_5B_{\mu \nu }`$. Note that $`\stackrel{~}{H}^{\mu \nu }`$ is not a tensor, since $`ϵ`$ is numeric
If the theory is interacting, the self duality relation is more complicated, and can be found in its full splendor in . For the cases we are interested in, where only 345 and 012 fields are turned on, the equations simplify to give:
$$_5B_{\mu \nu }=K_{\mu \nu }=\frac{\sqrt{G}G_{\mu \mu ^{}}G_{\nu \nu ^{}}\stackrel{~}{H}^{\mu ^{}\nu ^{}}}{G_5\sqrt{1+H_{\mu \nu \rho }H^{\mu \nu \rho }}}.$$
$`(21)`$
In the language of , if only 345 and 012 fields are turned on, $`z_1^2=2z_2`$, which makes the formulas simplify. Under this assumption, also
$$L_1=\sqrt{G}\sqrt{1+H_{\mu \nu \rho }H^{\mu \nu \rho }}.$$
In the action is given for a general gravitational background, but not for a background with a 3-form field turned on. The PST action however is given in the presence of a background 3-form field, but it is hard to use for explicit computations. Fortunately, we know how to obtain the PS action from the PST action without a background field. Therefore, we expect to obtain a generalization of the PS action by gauge fixing the PST action with background field. This can be easily done, and the only change in the PS action is : $`_5B_{\mu \nu }_5B_{\mu \nu }C_{5\mu \nu }`$, $`\stackrel{~}{H}^{\mu \nu }\frac{1}{3!}ϵ^{\mu \nu \rho \sigma \lambda }(F_{\rho \sigma \lambda }C_{\rho \sigma \lambda })`$.
### B A Toy Problem with the M5 brane action
In this section we will do a toy problem in which the interplay of the 2 formulations of the M5 brane action is shown. Let us consider a flat M5 brane extended in the 012345 directions in flat 11d, in a background of $`C_{012}`$ and $`C_{345}`$. We turn on a nonzero $`F_{345}`$, and we select “2” as our special direction in the PS action. We call $`f=F_{345}C_{345}`$. Thus $`\stackrel{~}{H}^{01}=f`$, and the Born Infeld term is $`L_{BI}=\sqrt{1+f^2}`$. The equation of motion (21) gives:
$$_2B_{01}C_{201}=\frac{f}{\sqrt{1+f^2}},$$
$`(22)`$
and thus
$$L_2=\frac{f^2}{2\sqrt{1+f^2}}.$$
Since the metric is diagonal, $`L_3=0`$. We can read off the Wess-Zumino term from the PST action. In the gauge we chose, $`F_{012}=_2B_{01}`$ is given by (22). Note that the action goes like $`f^2`$ for small $`f`$ and like $`f`$ for large $`f`$. This is also characteristic to the D-brane action. Note that applying naively the weak coupling version of the self-duality would give us an action growing like $`f^2`$ for large $`f`$, which is nonphysical (it does not reduce to Born Infeld).
Let us try to get some intuition about the physics of the problem. For small $`f`$ and no background fields, turning on a flux in the 345 direction induces the turning on of a flux in the 012 of the same magnitude (22). What this tells us is that if we dissolve an M2 brane in an M5 brane their fields force (by the SUGRA equations of motion) the appearance of a field which couples with an orthogonal M2 brane. However if $`f`$ is very large, (22) tells us that the dual M2 brane charge asymptotes to 1.
### C The M5 brane probe
We consider a large number of M2 brane along the $`012`$ directions and an M5 brane with 3 directions wrapped on an $`S^3`$. Since we know the effect of rotations in the 3-7, 4-8, 5-9, and respectively 7-10 planes on all the fields, we can assume the plane of the sphere to be $`3456`$. Let us denote by $`\widehat{ϵ}_{ijkl}`$ the numerical antisymmetric tensor restricted to the $`3456`$ plane. We also give the M5 brane an M2 brane charge $`n`$, by turning on a 3-form field strength along $`S^3`$:
$$F_3=\frac{4\pi n}{M_{11}^3r^43!}\widehat{ϵ}_{ijkl}x^idx^jdx^kdx^l.$$
$`(23)`$
We assume $`n<N`$, so the effect of the M2 brane charge of the probe on the background can be ignored. From (12) (13) and (18) we find that $`T_{3456}=3m`$, and thus
$$C_3^1=\left(\frac{R}{r}\right)^6S_3=3m\left(\frac{R}{r}\right)^6\frac{\widehat{ϵ}_{ijkl}}{3!}x^idx^jdx^kdx^l.$$
$`(24)`$
For further reference we can also express $`C_{\theta \varphi \alpha }^1`$ and $`F_{\theta \varphi \alpha }`$ using the angles of the 3 sphere, by noticing that $`\widehat{ϵ}_{ijkl}x^idx^jdx^kdx^l|_{S^3}=3!r^4\mathrm{sin}^2\theta \mathrm{sin}\varphi d\theta d\varphi d\alpha `$. $`C_3^0`$ is given by (1). The M theory 6 form is the dual of the 3 form and can be found using
$$dC_6\frac{1}{2}C_3F_4=dF_7=dF_4.$$
$`(25)`$
Using (4), (17) and the relations in (9), we obtain
$$dC_6=[Z^1(F_4^1F_4^1)+\frac{1}{2}d(Z^1C^1)]dx^0dx^1dx^2=0.$$
$`(26)`$
Thus the first term in the Wess-Zumino action gives no contribution. This is different from , where the nonzero 6-form background gave one of the leading contributions to the action.
Since we have spherical symmetry, the value of the action will be the same at every point on the 3-sphere. To make the computation more explicit we concentrate on the point $`x^6=r`$, and we chose “2” as our special direction. Thus:
$$H_{345}=H_{}=\left[\frac{4\pi n}{M_{11}^3r^3}3mR\frac{R^5}{r^5}\right]=\frac{A}{r^3}C_{345},$$
$`(27)`$
where $`A4\pi n/M_{11}^3`$ We are interested in the limit when the M2 brane charge of the M5 brane is bigger than its M5 brane charge. This means $`n>>\sqrt{N}`$. Therefore we expect the first term in $`H_{345}`$ to be dominant. We separate the 6 dimensional metric into perpendicular and parallel parts and denote by $`G_{}`$ and $`G_{}`$ their respective determinants. Using the equation of motion (21) we obtain
$$_2B_{01}C_{201}=H_{}G_{}^1\frac{\sqrt{G_{}G_{}}}{\sqrt{1+H_{}^2G_{}^1}}=H_{345}G^{33}G^{44}G^{55}\frac{\sqrt{detG_{\widehat{\mu }\widehat{\nu }}}}{\sqrt{1+H_{345}H^{345}}}$$
$`(28)`$
Since $`C_{012}`$ is known, (28) gives us the value of $`F_{012}`$, which couples with $`C_{345}`$ in the Wess Zumino term. For large M2 brane charge , $`C_{201}`$ is very close to the right hand side of (28), and basically the Wess-Zumino term containing $`F_{012}`$ is negligible.
This is very interesting, and definitely not a coincidence. What we discovered is that the equations of motion of an M5 brane with relatively large M2 brane charge in a geometry given by (1), give rise to a “dual” 3 form equal to the background 3-form field of this geometry, for any M2 brane charge which is large enough. Note that this is independent of $`Z`$, and even of the shape of the M5 brane (the $`G_{}`$’s cancel out). Thus the M5 brane Lagrangian somehow knows about the M-theory equations of motion. This is an interesting, if somewhat not expected connection which deserves further study.
We have all the pieces needed to compute the full potential for an M5 brane in this geometry. We also integrate over the sphere, which will give us a potential energy per unit length. The relevant parts of the potential are:
$$\frac{S_{BI}}{2\pi ^2V}=r^3\sqrt{\frac{1}{Z}\left[1+\frac{H_{345}^2}{Z}\right]}\frac{A}{Z}+\frac{r^6}{2A}+\frac{r^3C_{345}}{Z}.$$
$`(29a)`$
$$\frac{S_{mixed}}{2\pi ^2V}=\frac{r^3}{2}\frac{H_{345}H^{345}\sqrt{G}}{\sqrt{1+H_{345}H^{345}}}\frac{A}{2Z}+\frac{r^6}{4A}\frac{r^3C_{345}}{2Z}.$$
$`(29b)`$
$$\frac{S_{WZ}}{2\pi ^2V}=\frac{r^3}{2}[2C_{012345}^6+C_{012}F_{345}F_{012}C_{345}]\frac{A}{2Z},$$
$`(29c)`$
where the approximation is in the large $`n`$ limit. We included the $`C^6`$ component which in this case is 0, in order to see how it combines with $`C_{345}`$ in this action.
As expected, the dominant contributions of the 3 actions come from the M2 brane charge of the M5 brane, and they cancel. The second terms in (29a) and (29b) represent the gravitational energy of the M5 brane. In the absence of a background $`C_3`$ they would cause the M5 brane to collapse on the stack of M2 branes.
Since we only worked at order $`m`$ in perturbation theory, there can be another term proportional to $`m^2A`$ which has the same relevance as the first 2. Indeed, we expect an order $`m^2`$ correction in $`C_{012}`$, via (3). Since $`C_{012}`$ couples with $`F_{345}`$, we can see that this correction will be relevant. It can also be easily seen that the contribution of this term to the potential goes like $`r^2`$.
Thus, the dominant part of the potential is:
$$\frac{S}{2\pi ^2V}=\frac{3r^6}{4A}\frac{4m}{2}r^4+cAm^2r^2,$$
$`(30)`$
where $`c`$ is not yet determined. Since we have 4 supercharges, $`c`$ can be computed easily. Indeed, the potential is the square of the derivative of the superpotential, so $`c`$ is obtained by simply completing the square. Before proceeding with this we need to write the potential in a more general form. As we mentioned at the beginning of this chapter, we restricted our attention the 3456 plane. The general SO(4) invariant brane configuration is obtained by rotating the sphere of radius $`r`$ in the 3-7, 4-8, … planes by the same angle $`\theta `$. This configuration is parametrized by $`z=re^{i\theta }`$. Examining the effect of rotation on the terms we had, we see that the action will be:
$$\frac{S}{2\pi ^2V}=\frac{3|z|^6}{4A}2m\mathrm{Re}(z^3\overline{z})+\frac{4|z|^2m^2A}{3}=\frac{3}{4A}|z^3+4zmA/3|^2,$$
$`(31)`$
where the last term was obtained by completing the square. This has a supersymmetric minimum in the 3456 plane at
$$r^2=\frac{4mA}{3}.$$
$`(32)`$
There is however an extra case to consider. Our mass perturbation (13) is invariant under (14), which flips the M5 brane from the 3456 plane to the 879 10 plane. Thus, there will also exist a supersymmetric minimum corresponding to an anti M5 brane of the same radius, in the 789 10 plane. We will have more comments on the superpotential which generates the action (31) in chapter V.
This is what we advertised: a test M5 brane in the background formed by M2 branes with $`F_4`$ flux has a ground state at nonzero $`r`$. As we mentioned, the equivalent string theory picture can be interpreted in 2 ways - as D3 brane polarization or as a ground state for a D5 brane with D3 brane charge. Unfortunately, no one has studied the polarizability of M2 branes, so we can only assume it takes place by analogy with the string theory case.
## IV The Full Problem - Warped Geometry
As we have seen in the previous chapter, turning on fermion masses polarizes the M2 brane. We now consider the case of the $`N`$ M2 branes distributed uniformly on one or several 3-spheres, with M5 brane charges. When the 5 brane charges are relatively small, the background geometry will still be given by (1), but $`Z`$ will be different. Since we will lose most of the symmetry, it appears that the problem will be far harder than the one with a probe M5 brane. Fortunately, like in , the action does not change. so there is no more work to do. However, here the “conspiracy” which makes this work is far more unexpected.
### A The Geometry
As known from time immemorial, the geometry created by a distribution of M2 branes is still given by (2), but with $`Z`$ being the superposition of the harmonic functions sourced by each brane. If for example the M2 branes are spread over a 3-sphere of radius $`r_0`$ in the $`3456`$ plane, the new Z-factor will be
$$Z=\frac{2}{\pi }_0^\pi \frac{R^6\mathrm{sin}^2\theta d\theta }{(r_2^2+r_1^2+r_0^22r_0r_1\mathrm{cos}\theta )^3}=\frac{R^6}{(r_2^2+(r_1+r_0)^2)^{\frac{3}{2}}(r_2^2+(r_1r_0)^2)^{\frac{3}{2}}},$$
$`(33)`$
where $`r_1`$ and $`r_2`$ are the radii in the $`3456`$ and respectively $`\mathrm{789\; 10}`$ planes. When the M2 branes are distributed over several such spheres, in the $`3456`$ and $`\mathrm{789\; 10}`$ planes, Z will be a sum of such terms, properly normalized.
The field equation for an antisymmetric antiselfdual perturbation is again:
$$d[Z^1(F_4^1F_4^1)]=0.$$
$`(34)`$
and the Bianchi identity is $`dF_4^1=0`$. The behavior of the solution at infinity is given by the boundary theory, and is the same as for trivial $`Z`$. There is also a magnetic source corresponding to the 5 brane, but this creates a normalizable mode, which is subleading at $`\mathrm{}`$. We can perform the same clever trick as in . From (34) we can derive:
$$d[Z^1(F_4^1F_4^1)]=0.$$
$`(35)`$
Therefore, by the Hodge decomposition we derive that $`[Z^1(F_4^1F_4^1)]`$ is harmonic, and thus equal to its value at $`\mathrm{}`$:
$$Z^1(F_4^1F_4^1)=2T_4.$$
$`(36)`$
In particular, $`C_3^1`$ and $`C_6`$ will change, but (26) implies that the combination $`[C_6+\frac{1}{2}C_3^1C_3^0]`$ will not change. This same combination appears in the M5 brane action (29), namely $`C_{012345}\frac{1}{2Z}C_{345}`$. We are definitely seeing a “conspiracy” - the factor of $`1/2`$ came from both the Born-Infeld and the mixed term of the M5 brane Lagrangian, and provides exactly the combination which is unchanged when we change $`Z`$.
There is one more thing to consider, the effect of the M5 brane charge on itself. The M5 brane is a magnetic source for the 3-form field, and it appears as a source in the right hand side of the Bianchi identity. Nevertheless, the M5 brane only couples with the combination (36), which remains unchanged, and thus it is unaffected by itself.
### B The solutions
Let us consider first the potential felt by a probe M5 brane with M2 brane charge in the geometry created by several shells of M2 branes. This is still given by (29) but with a different $`Z`$. The leading contributions cancel as usually. The first 2 terms in (30) will not change, and by supersymmetry, the third term will not change either. Therefore, the potential will be independent of the distribution of the sources.
We would like now to consider the potential felt by the full set of M2 branes. This can be found by bringing the branes one by one from $`\mathrm{}`$. In our case, like in the potential felt by a brane does not depend on the distribution of the others, so as explained there, the potential is the same as in the probe case.
If the 4 masses are equal, a general ground state is a configuration consisting of M2 brane 3-sphere shells with charges $`n_b`$ in one of the planes $`3456`$ and $`\mathrm{789\; 10}`$. The potential will be:
$$\frac{S}{2\pi ^2V}=\underset{b}{}\frac{3}{4A_b}\left|z_b^3\frac{4Q_bz_bmA_b}{3}\right|^2,$$
$`(37)`$
where $`Q_b`$ is by convention 1 for M5 and -1 for anti M5 branes, and $`A_b4\pi n_b/M_{11}^3`$. There is however one more condition which we ignored. In order for the geometry to be valid, the M5 brane charge density of the shells should be smaller than their M2 brane charge. This means:
$$n_b>>\sqrt{N}$$
$`(38)`$
Thus the possible configurations will be given by distributions satisfying (38).
We can see that there is a large number of discrete vacua, corresponding to combinations of charges $`n_b`$, adding to $`N`$ and satisfying (38), in both the 3456 and the 789 10 planes. It is a straightforward exercise to compute the normalizable modes created by the M5 branes, in each of these vacua. The coefficient of these normalizable modes gives the value of a condensate which contains the fermion condensate and its supersymmetric partners. The vacua will be distinguished by the values of these condensates. The $`Z_2`$ symmetry will relate the vacua with M5 branes replaced by anti M5 branes ($`Q_bQ_b`$).
Unfortunately, we cannot interpret the vacua in any way as corresponding to broken gauge symmetries some of which are restored when the M5 branes are coincident, since the theory living on $`N`$ M2 branes has mysterious degrees of freedom. Just for fun we may observe that a relation similar to (82,83) in also holds here. Namely, if $`n_1\times n_2=N`$, descriptions with $`n_1`$ coincident M5 branes and with $`n_2`$ coincident anti M5 branes have complementary ranges of validity. A speculative mind may see in this a sign of some duality, but we will refrain from further commenting on that.
## V Unequal masses
We can try to generalize the previous construction for the case of unequal masses. Since we will only be interested in the limits when one or two masses go to 0, and since want to keep the presentation simple, we will keep $`m_3=m_4m`$ and vary $`m_1`$ and $`m_2`$. Unlike the previous case, where the general configurations was SO(4) invariant, here we only have SO(2) symmetry, so we expect the M2 branes to become polarized into an ellipsoid with 2 equal axes. Again we can restrict to the 3456 plane and then obtain more general configurations by phase rotations. The ellipsoid will be parametrized:
$$x^3=ar\mathrm{cos}\theta $$
$$x^4=br\mathrm{sin}\theta \mathrm{cos}\varphi $$
$$x^5=r\mathrm{sin}\theta \mathrm{sin}\varphi \mathrm{cos}\alpha $$
$$x^6=r\mathrm{sin}\theta \mathrm{sin}\varphi \mathrm{sin}\alpha $$
$`(39)`$
From (23) and (24), we see that on the ellipsoid $`C_{\theta \varphi \alpha }^1`$ is multiplied by $`ab`$ and $`F_{\theta \varphi \alpha }`$ is unchanged. Also in the $`\theta \varphi \alpha `$ coordinates
$$G_{}=r^6Z\mathrm{sin}^4\theta \mathrm{sin}^2\varphi (a^2\mathrm{sin}^2\theta \mathrm{cos}^2\varphi +a^2b^2\mathrm{sin}^2\theta \mathrm{sin}^2\varphi +b^2\mathrm{cos}^2\theta )$$
$`(40)`$
We noted that for any distribution of brane, the M5 brane equations of motion create a dual field almost equal to the background $`C_{012}`$, so there will be no new contribution from the Wess-Zumino term. The dominant terms of the potential do not depend on $`G_{}`$, so they will have the same value as before, and they will cancel as usually. The terms which before were proportional to $`r^6/A`$ will now be variable on the ellipsoid, and their value will be:
$$V_{r^6}_{E^3}\sqrt{G_{}}\frac{G_{}}{H_{\theta \varphi \alpha }}$$
$`(41)`$
What one might have been afraid of (the integrand proportional to $`\sqrt{G_{}}`$ \- which would have made the integral elliptic) does not happen. The integral can be easily worked out to give:
$$\frac{S_{r^6}}{2\pi ^2V}=\frac{3r^6}{4A}\frac{a^2+b^2+2a^2b^2}{4}$$
$`(42)`$
The term proportional with $`C_3`$ will be multiplied by $`ab`$, and will change also because of (12). Putting back the phases corresponding to rotations, and remembering that $`|z_3|=|z_4|=r`$, $`|z_1|=ar`$, and $`|z_2|=br`$ we obtain the potential to be:
$$\frac{S}{2\pi ^2V}=\frac{3}{16A}\left(2|z|^2|z_1|^2|z_2|^2+|z|^4|z_1|^2+|z|^4|z_2|^2\right)\frac{1}{2}\mathrm{Re}(2mz_1z_2z\overline{z}+m_2z_1\overline{z}_2zz+m_1\overline{z}_1z_2zz)$$
$$+\frac{A}{3}(2m^2|z|^2+m_1^2|z_1|^2+m_2^2|z_2|^2).$$
$`(43)`$
where the second line was added to complete the square, as required by supersymmetry. We can illustrate better the SUSY nature of this action by writing it as:
$$\frac{S}{2\pi ^2V}=\frac{3}{16A}(2|zz_1z_24mAz/3|^2+|z^2z_24m_1Az_1/3|^2+|z^2z_14m_2Az_2/3|^2)$$
$`(44)`$
This potential has a supersymmetric minimum at:
$$z_1^2=\frac{4A}{3}\sqrt{\frac{m^2m_2}{m_1}}z_2^2=\frac{4A}{3}\sqrt{\frac{m^2m_1}{m_2}}z^2=\frac{4A}{3}\sqrt{\frac{m_1m_2m}{m}}$$
$`(45)`$
Thus, the branes will polarize into an ellipsoid in the 3456, respectively 879 10 planes with the axes given above.
We will now try to write a few comments about the superpotential corresponding to the action(44). We can see that if we assume our fields to be the complex scalars $`z_i`$, we can write a superpotential of the form
$$Wz_1z_2z_3z_4\frac{2A}{3}\underset{i}{}m_iz_i^2$$
$`(46)`$
In the case of polarized D-branes, a similar superpotential came from a superpotential originally of the form:
$$W\mathrm{tr}(\mathrm{\Phi }_1[\mathrm{\Phi }_2,\mathrm{\Phi }_3]),$$
$`(47)`$
perturbed with a mass term. The $`\mathrm{\Phi }_i`$ were the $`N\times N`$ scalar fields on the brane, corresponding to position in spacetime. The mass terms forced the ground state $`\mathrm{\Phi }_i`$ to become noncommutative, which corresponded to polarization of the Dp brane into a D(p+2) brane.
In our case however we do not know what form the scalars representing the position of the M2 brane have (they are definitely not $`N\times N`$ matrices). Therefore, we do not know what form the equivalent of (46) will have. What we do know is that it will involve all the 4 scalar fields, and that adding mass terms will cause the ground state fields to become “noncommutative” (more rigorously speaking - to modify the ground state fields so that the term which contains all the 4 fields will be nonzero). This can give an intuitive picture for the polarization of the M2 brane into a brane with 3 extra dimensions.
We also observe that the superpotential (46) has a “classical” form. More precisely, it looks like the superpotential of a theory with 4 massive hypermultiplets, and does not contain nonperturbative terms. Since we are at very strong coupling and we do not know the degrees of freedom of the theory, we can only suggest that this happens because of $`N`$ being large.
We can also make a few comments about cases when some of the masses go to 0. Using (45) we can see that if we take one mass to 0, the ellipsoid degenerates into a line of very long length, which corresponds to the theory having a moduli space. Intuitively we can see that if $`z_1`$ has no mass, it is a modulus. Nevertheless, this moduli space is not protected by supersymmetry, and can in principle be lifted by corrections. If we take $`m_1,m_20`$ keeping $`m_3=m_4=m`$, we restore $`𝒩=4`$. The ellipsoid degenerates into a pancake of radius $`r^2=\sqrt{4Am^2/3}`$. It would be interesting to give an interpretation for this in the framework of theories with 8 supercharges.
## VI More about the theory on the brane
Since not too much is known about the $`𝒩=2`$ theory whose dual we constructed, we can only use the duality one way: to interpret the possible M-theory configurations from the point of view of the $`𝒩=2`$ theory.
### A Domain Walls
Since our theory has multiple vacua, they can be separated by domain walls. Let us consider the domain wall between the vacuum corresponding to all the 2-branes polarized into one M5 brane, and the vacuum with 2 M5 branes of charges $`n_1`$ and $`n_2`$. Since the first 3-sphere has radius $`r\sqrt{N}`$, and the concentric 3-spheres will have radii $`r_i\sqrt{n_i}`$, they will both bend and meet at an intermediate radius $`r_0`$. By charge conservation, another M5 brane should come out of the junction. Therefore, the domain wall should correspond to a M5 brane filling the 3456 ball of radius $`r_0`$ and extended in the 01 directions. Since the M5 brane has 2 longitudinal and 4 transverse directions, it feels no warp factor, and its tension will be
$$\tau _1=\frac{M_{11}^6}{(2\pi )^5}\frac{\pi ^2r_0^4}{2}m^2N^2.$$
$`(48)`$
The tension will have another piece, $`\tau _2`$, which comes from the bending of the branes.
In the case of vacua with the same number of M5 branes, the branes will just bend into each other, and there will be no object required by charge conservation to fill the 4-ball. Thus the tension will only have one piece coming from the bending of the branes. It is not hard to compute $`\tau _2`$, although we have not done it here.
We can compare $`\tau _1`$ with the tension of a supersymmetric domain wall, which is given by the difference of the superpotentials in the 2 phases. Using (46) and (37) we can see that for two generic vacua
$$\tau _{DW}|\mathrm{\Delta }W|m^2N^2.$$
$`(49)`$
This has the same dependence on $`m`$ and $`N`$ as $`\tau _1`$. It will be an interesting exercise to show that the construction with bent M5 branes reproduces the superpotential calculation for supersymmetric domain walls. Even if the exact normalization of the superpotential is not known, the matching of the dependence of the tension on the different $`n_i`$’s would be a beautiful result.
Since naively the superpotentials are the same in 2 vacua related by the $`Z_2`$ symmetry, it may appear from (49) that the tension of the domain wall between them is 0. Similarily, (49) implies that the tension between vacua characterized by M2 brane charges $`n_i`$ and $`n_i^{}`$ is zero, if $`n_i=n_i^{}=N`$ and $`n_i^2=n_i^2`$ . This appears to contradict the expectation that the bending tension $`\tau _20`$. Nevertheless, since we do not know the relative sign of the superpotentials in the 2 phases, and we also do not know if the domain walls are supersymmetric, there is no contradiction.
### B Condensates, Instantons, etc.
The only candidate for an instanton (an object with spacetime dimension 0) is an M2 brane wrapped on the $`S^3`$. Nevertheless, this configuration is unstable. The wrapped M2 brane can attach to the M5 brane and slide off.
As we mentioned, the coefficient of the normalizable mode created by the brane configurations gives the value of the condensate which contains the fermion bilinear and its supersymmetric partners. This can be straightforwardly computed, although it is not done here. The other objects which may exist in this theory also need a more thorough investigation.
## VII Conclusions and future directions
Most of the conclusions of this construction are identical to the conclusions of , and probably the main one is that M theory resolves the naked singularity on might have expected to obtain by turning on fermion masses.
Lots of things remain to be done and understood. The tension of the domain walls, as well as their shape are well within our reach. A bit hard is to get an idea about what do the plethora of vacua of the theory represent.
Another interesting thing which we observed was the interplay of the 3 pieces of the M5 brane Lagrangian, and how the turning on of a large flux along the $`S^3`$ induced via the M5-brane equations of motion a dual field equal to the background 3-form. This is an interesting connection between the equations of motion of the M5-brane and of 11d supergravity which deserves further study. We should also mention that this is to our knowledge the first time when the M5 brane bosonic actions (both by Pasti, Sorokin and Tonin, and by Perry and Schwarz) were used in a direct calculation. The cancellation of main contributions in the potential (as required by supersymmetry), and the fact that the subleading terms reproduce a supersymmetric potential are nontrivial consistency checks for these actions.
The case when one or two hypermultiplet masses are brought to 0 also awaits a more thorough investigation. More can be said about the moduli of these theories, and possible connections can be made with the vast literature on theories with 8 supercharges.
There is also a relatively well developed subject dealing with generalizations of the $`AdS_4\times S^7`$ duality to different less supersymmetric versions corresponding to various 7-manifolds. In particular the perturbation which preserves $`G_2`$ symmetry (and which is a combination of the self-dual and anti self dual field strengths) has been identified as giving rise to the flow from the $`𝒩=8,AdS_4\times S^7`$ vacuum to the $`G_2`$ invariant Englert vacuum. The tools developed in this paper can be used to learn more about that flow.
The theory whose M-theory dual we found still remains very mysterious. Nevertheless we were able using this construction to understand quite a few things about it. It is the (probably overoptimistic) hope of the author that this approach may bring us closer to a more complete understanding of this theory and of the M-theory degrees of freedom.
ACKNOWLEGEMENTS: I’m deeply indebted to Joe Polchinski for his help and guidance through this project, and to Costin Popescu for his help in understanding the M5 brane action. I also profited from very stimulating conversations with Duiliu-Emanuel Diaconescu, Alex Buchel, Aleksey Nudelman, Andrew Frey, Mitesh Patel, and Simion Hellerman. This work was supported in part by NSF grant PHY97-22022.
|
warning/0004/hep-ph0004066.html
|
ar5iv
|
text
|
# A POSSIBLE TWO-COMPONENT STRUCTURE OF THE NON-PERTURBATIVE POMERON
Abstract
We propose a QCD-inspired two-component Pomeron form which gives an excellent description of the $`pp,\pi p,Kp,\gamma p`$ and $`\gamma \gamma `$ total cross-sections. Our fit has a better $`\chi ^2/dof`$ for a smaller number of parameters as compared with the PDG fit. Our 2-Pomeron form is fully compatible with weak Regge exchange-degeneracy, universality, Regge factorization and the generalized vector dominance model.
LPNHE 00-02 April 2000
40 years after its introduction and in spite of very important advances in QCD, the Pomeron remains an open problem. In particular, the non-perturbative structure of the Pomeron is still controversial.
The most popular model of the non-perturbative Pomeron is, of course, the Donnachie-Landshoff (DL) model . The total cross-sections for $`pp`$ and $`\overline{p}p`$ scattering are parametrized in terms of five parameters :
$$\sigma _{pp}=X_{pp}s^\epsilon +Y_{pp}s^\eta $$
(1)
$$\sigma _{\overline{p}p}=X_{pp}s^\epsilon +Y_{\overline{p}p}s^\eta $$
(2)
where
$$\epsilon =\alpha _P(0)1$$
(3)
and
$$\eta =1\alpha _R(0);$$
(4)
$`\alpha _P(0)`$ is the Pomeron-intercept, $`\alpha _R(0)`$ is the effective non-leading exchange-degenerate Regge intercept and $`X,Y`$ the corresponding Regge residues. An overall scale factor $`s_0=1`$ GeV<sup>2</sup> is implicitely present in eqs. (1)-(2). The key-parameters $`\epsilon `$ and $`\alpha _R(0)`$ have the following values :
$$\epsilon =0.0808$$
(5)
and
$$\alpha _R(0)=0.5475.$$
(6)
The $`pp`$ data are well reproduced. It was therefore tempting to use the DL form to the simultaneous study of all existing total cross-sections. It is precisely what was done by PDG in the last edition of the ”Review of Particle Physics” , .
The total cross-sections $`\sigma `$ are parametrized in refs. 3 and 4 in the variant of a non-exchange-degenerate DL form :
$$\sigma _{AB}=X_{AB}s^\epsilon +Y_{1AB}s^{\eta _1}Y_{2AB}s^{\eta _2},$$
(7)
$$\sigma _{\overline{A}B}=X_{AB}s^\epsilon +Y_{1AB}s^{\eta _1}+Y_{2AB}s^{\eta _2},$$
(8)
where
$$\eta _1=1\alpha _{R_+}(0),\eta _2=1\alpha _R_{}(0),$$
(9)
$`\alpha _{R_+}(0)`$ and $`\alpha _R_{}(0)`$ being the Regge intercepts of the non-leading Regge trajectory $`R_+`$ in the even-under-crossing amplitude and $`R_{}`$ in the odd-under-crossing amplitude respectively. $`X,Y_1,Y_2`$ are the corresponding Regge residues. There are 16 parameters for fitting 271 experimental points involving 8 reactions : $`\overline{p}p,pp,\pi ^\pm p,K^\pm p,\gamma p`$ and $`\gamma \gamma `$. The overall $`\chi ^2`$ is excellent : $`\chi ^2/dof=0.93`$<sup>1</sup><sup>1</sup>1A bigger value $`\chi ^2/dof=1.02`$, corresponding to 383 experimental points and $`\epsilon =0.0933`$, is quoted in table 1 of ref. 4 because real parts are also included in the respective fits (see text for a discussion of this option).. The key-parameter $`\epsilon `$ has now the value 0.0900.
The problem with the form of refs. 3 and 4 is the bad violation of the weak exchange-degeneracy (i.e. $`\alpha _{R_+}(0)=\alpha _R_{}(0))`$, namely
$$\alpha _{R_+}(0)\alpha _R_{}(0)0.2.$$
However, the masses of the resonances, as published in the ”Review of Particle Physics” , clearly indicate that the weak exchange-degeneracy is respected. As seen from fig. 1a) the 10 resonances belonging to the 4 different $`I^G(J^{PC})`$ families $`\rho \omega f_2a_2`$ are compatible with a unique linear exchange-degenerate Regge trajectory
$$\alpha (t)=\alpha (0)+\alpha ^{}t$$
(10)
with
$$\alpha (0)=0.48$$
(11)
and
$$\alpha ^{}=0.88(GeV/c)^2.$$
(12)
The numerical values (11)-(12) are extracted just by plugging in (10) the masses and the spins of $`\rho _1(770)`$ and $`\rho _3(1690)`$ resonances.
Remarkably enough, the same $`\alpha (0)`$ value (11) is compatible with the $`\mathrm{\Delta }\sigma `$ data for the total cross-section differences
$$\mathrm{\Delta }\sigma _{AB}\sigma _{\overline{A}B}\sigma _{AB}=2Y_{2AB}s^{\alpha _R_{}(0)1}$$
(13)
or
$$\text{ln}[s\mathrm{\Delta }\sigma _{AB}]=\text{ln}(2Y_{2AB})+\alpha _R_{}(0)\text{ln}s.$$
(14)
The $`\mathrm{\Delta }\sigma `$ data for $`pp,Kp`$ and $`\pi p`$ and $`\sqrt{s}\underset{}{>}6`$ GeV shown in the log-log plot of fig. 1b) are all compatible with the straight lines of eq. 14 , the slopes of which are precisely given by the $`\alpha _R_{}(0)`$ value of eq. (11).
These indications in favour of the weak exchange-degeneracy, coming both from the resonance and scattering region, is too striking to be a mere coincidence. One can therefore wonder if something is inadequate in the parametrization (7-8).
A first problem can come from the fact that the ratio $`\rho (s,t=0)=\text{Re}F(s,t=0)/\text{Im}F(s,t=0)`$ has been included into the PDG fit together with the total cross-sections. As it is known, the determination of this $`\rho `$ parameter is semi-theoretical : its value is obtained through an extrapolation of the elastic amplitude to t=0 using a theoretical model. The result is very sensitive to these theoretical assumptions (see, for example, ). If we try to redo the minimization using the total cross-sections only, the violation of the weak Regge exchange-degeneracy persists, as already noted in . In the following we will minimize with different analytic forms but using the total cross-sections only.
An important problem may come from the form of the Pomeron. The non-perturbative Pomeron is certainly much more complex than a simple pole, which violates the unitarity. It surely includes cuts associated with multiexchanges which restore unitarity. We do not know the exact form of these complicate singularities. But we can try to mimic them by a 2-component Pomeron, a Pomeron built from two Regge singularities.
The perturbative Pomeron has also a complex form. The BFKL Pomeron is not a simple pole but rather a complicate cut or an accumulation of poles close to $`J=1`$. Also, very recently, detailed calculations in the perturbative QCD indicate, in fact, the existence of a 2-component Pomeron. Namely, in LLA, beside the BFKL Pomeron associated with 2-gluon exchange and corresponding to an intercept $`\alpha _p^{2g}(0)>1`$, one finds a new Pomeron associated with the 3-gluon exchange with $`C=+1`$ and corresponding to an intercept $`\alpha _p^{3g}(0)=1`$ ; the 3-gluon Pomeron is exchange-degenerate with the 3-gluon $`C=1`$ Odderon .
Inspired by these considerations, we explore in this paper the possibility of a 2-component Pomeron in the non-perturbative sector, namely a Pomeron built from two poles. We propose the new analytic forms for the total cross-sections :
$$\begin{array}{ccc}\hfill \sigma _{pp}& =& Z_{pp}+Xs^\epsilon +(Y_1^{pp}Y_2^{pp})s^{\alpha (0)1}\hfill \\ \hfill \sigma _{\overline{p}p}& =& Z_{pp}+Xs^\epsilon +(Y_1^{pp}+Y_2^{pp})s^{\alpha (0)1}\hfill \\ \hfill \sigma _{\pi ^+p}& =& Z_{\pi p}+Xs^\epsilon +(Y_1^{\pi p}Y_2^{\pi p})s^{\alpha (0)1}\hfill \\ \hfill \sigma _{\pi ^{}p}& =& Z_{\pi p}+Xs^\epsilon +(Y_1^{\pi p}+Y_2^{\pi p})s^{\alpha (0)1}\hfill \\ \hfill \sigma _{K^+p}& =& Z_{Kp}+Xs^\epsilon +(Y_1^{Kp}Y_2^{Kp})s^{\alpha (0)1}\hfill \\ \hfill \sigma _{K^{}p}& =& Z_{Kp}+Xs^\epsilon +(Y_1^{Kp}+Y_2^{Kp})s^{\alpha (0)1}\hfill \\ \hfill \sigma _{\gamma p}& =& \delta Z_{pp}+\delta Xs^\epsilon +Y_1^{\gamma p}s^{\alpha (0)1}\hfill \\ \hfill \sigma _{\gamma \gamma }& =& \delta ^2Z_{pp}+\delta ^2Xs^\epsilon +Y_1^{\gamma \gamma }s^{\alpha (0)1}\hfill \end{array}$$
(15)
where $`\alpha (0)`$ is fixed to the value $`\alpha (0)=0.48`$ as given by resonance masses and a scale factor $`s_0=1`$ GeV<sup>2</sup> is implicitely supposed.
The Pomeron in eqs. (15) has 2 components : the $`X`$-component corresponds to a Regge intercept bigger than 1 ($`\epsilon >0`$) and the $`Z`$-component corresponds to an intercept exactly localised at 1. We suppose that the $`X`$-component is fully universal (its coupling is the same in all hadron-hadron reactions, as well as the energy behaviour $`s^\epsilon `$), while the $`Z`$-component is not fully universal. It is tempting to interpret the $`X`$-component as the gluonic component of the non-perturbative Pomeron and the $`Z`$-component as its flavour-dependent non-perturbative component. It is interesting to note that the possibility of a fully universal Pomeron was already considered in literature .
Of course, there is no double counting. In the framework of the $`S`$-matrix Theory , there are 2 solutions of the Reggeon calculus : a critical Pomeron with intercept equal to 1, leading asymptotically to a $`(\text{ln}s)^\eta (\eta <2)`$ behaviour of the total cross-sections, and a supercritical Pomeron with intercept higher than 1, connected, at asymptotic energies, to the Froissart $`\text{ln}^2s`$ behaviour of the total cross-sections. In other words, the $`X`$ and $`Z`$ components correspond to 2 different Regge singularities.
Both components are supposed to obey the Regge factorization property. This is realized via the $`\delta `$-parameter in eqs. (15) for the $`pp,\gamma p`$ and$`\gamma \gamma `$ processes.
Finally, the secondary Regge pole of intercept $`\alpha (0)`$ corresponds to an exchange-degenerate trajectory, in agreement with our previous discussion.
The forms (15) involve n=14 parameters (to be compared with the PDG value n=16). The values of the parameters in eq. (15) are given in table 1. The corresponding $`\chi ^2`$ value is excellent : $`\chi ^2/dof=0.86`$ to be compared with the PDG value $`\chi ^2/dof=0.93`$. The quality of the fit is illustrated in fig. 2.
The value $`\epsilon =0.132`$ (see table 1) is certainly bigger than the DL value $`\epsilon =0.081`$ or the PDG value $`\epsilon =0.093`$ and it appears as being in between the effective Pomeron intercept value 1.1 and the bare Pomeron intercept value 1.2 . However a direct comparison of different $`\epsilon `$ values is not yet significative : all the existing data other than $`\sigma `$ have first to be refitted by using a 2-component Pomeron amplitude.
Note that the residue of the non-leading Regge trajectory $`Y_1^{\gamma \gamma }`$, numerically close to 0, is not well determined : its weight in the minimization is negligible due to the low precision of the low energy $`\gamma \gamma `$ data.
Let us also note that the value $`0.30310^2`$ of the $`\delta `$-parameter is perfectly compatible with the generalized vector-dominance model .
It is interesting to explore the relative importance of $`X`$ and $`Z`$ components in $`\sigma `$, by plotting the ratio $`R`$ (see fig. 3)
$$R=\frac{Xs^\epsilon }{Z}.$$
(16)
It can be seen from fig. 3 that the $`X`$-component acts like an asymptotic component. However the asymptoticity is clearly delayed : $`X`$ dominates over the $`Z`$ component only in the TeV region. The only exception is the $`Kp`$ scattering, where the asymptoticity occurs already in the ISR region of energies.
By using the analytic forms (15) and the values of the parameters given in table 1, one can make detailed predictions for $`\sigma `$ at high energies, in particular in the RHIC, LHC and cosmic-rays regions of energies - see table 2. However, one should not consider too seriously such predictions : unitarization will certainly introduce important corrections at high energies. The $`X`$-component, as the DL Pomeron, violates unitarity.
In conclusion, we propose a QCD-inspired analytic form of the Pomeron, a 2-pole Pomeron form, which gives an excellent fit to the $`pp,\pi p,Kp,\gamma p`$ and $`\gamma \gamma `$ total cross-sections. Compared to the PDG fit with a simple Pomeron-pole, our fit has a better $`\chi ^2/dof`$ with a smaller number of parameters. This 2-pole Pomeron form has the advantage to be fully compatible with the weak Regge exchange-degeneracy , universality, Regge factorization and the generalized vector dominance model.
The theoretical and phenomenological implications of the 2-component Pomeron are important and therefore they should be explored in the future in a detailed way.
Acknowledgements. We thank Prof. Vladimir Ezhela for kindly putting to our disposal the PDG data and for very fruitful discussions.We also thank Prof. Alphonse Capella for useful remarks.
| $`\epsilon `$ | $`\delta ,10^2`$ | X | $`Z_{pp}`$ | $`Z_{\pi p}`$ | $`Z_{Kp}`$ |
| --- | --- | --- | --- | --- | --- |
| 0.132 | 0.303 | 7.572 | 20.251 | 5.283 | 2.208 |
| $`Y_1^{pp}`$ | $`Y_2^{pp}`$ | $`Y_1^{\pi p}`$ | $`Y_2^{\pi p}`$ | $`Y_1^{Kp}`$ | $`Y_2^{Kp}`$ | $`Y_1^{\gamma p}`$ | $`Y_1^{\gamma \gamma }`$ |
| --- | --- | --- | --- | --- | --- | --- | --- |
| 74.811 | 29.918 | 48.972 | 6.028 | 34.483 | 11.935 | 0.121 | $`0`$ |
Table 1 : The values of the parameters in the analytic forms (15).
$`\epsilon `$ and $`\delta `$ are pure numbers. The rest of the parameters are in mb.
| $`\sqrt{s},GeV`$ | $`\sigma _{\overline{p}p}`$ | $`\sigma _{pp}`$ | $`\sigma _{\pi ^+p}`$ | $`\sigma _{\pi ^{}p}`$ | $`\sigma _{K^+p}`$ | $`\sigma _{K^{}p}`$ | $`\sigma _{\gamma p}`$ | $`\sigma _{\gamma \gamma },10^3`$ |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| 100 | 46.8 | 46.3 | 31.4 | 31.3 | 28.2 | 28.0 | 0.140 | 0.421 |
| 200 | 51.5 | 51.2 | 36.3 | 36.3 | 33.2 | 33.1 | 0.155 | 0.469 |
| 300 | 54.8 | 54.7 | 39.7 | 39.7 | 36.6 | 36.6 | 0.166 | 0.501 |
| 400 | 57.5 | 57.4 | 42.4 | 42.4 | 39.3 | 39.3 | 0.174 | 0.525 |
| 500 | 59.7 | 59.6 | 44.6 | 44.6 | 41.5 | 41.5 | 0.180 | 0.546 |
| 600 | 61.6 | 61.5 | 46.6 | 46.6 | 43.5 | 43.5 | 0.186 | 0.564 |
| 1800 | 75.4 | 75.4 | 60.4 | 60.4 | 57.3 | 57.3 | 0.228 | 0.692 |
| 12000 | 111 | 111 | 96.4 | 96.4 | 93.3 | 93.3 | 0.337 | 1.02 |
| 30000 | 136 | 136 | 121 | 121 | 118 | 118 | 0.413 | 1.25 |
Table 2 : Extrapolation of the analytic forms (15) at high energies.
$`\sigma `$ are given in mb.
|
warning/0004/hep-th0004071.html
|
ar5iv
|
text
|
# Perturbative Chern-Simons Theory on Noncommutative ℝ³
## 1 Introduction
It is well known that at the Planck scale the flat space-time must be modified drastically leading to a noncommutative space-time. The corresponding field theories have to be formulated in the framework of noncommutative geometry .
In this short letter we discuss the Chern-Simons theory with respect to a non-Abelian $`U(N)`$-gauge group on noncommutative $`^3`$. In the corresponding $`\theta `$-deformed field theory we are able to establish the usual BRST-symmetry and the topological vector supersymmetry (VSUSY) . The vector-like generators of this VSUSY give rise together with the BRST-operator to a Wess-Zumino-like anticommutation relation which closes on-shell on the space-time translations.
At the classical level the Ward-identity characterizing the VSUSY is linearly broken in the quantum fields. These breakings are induced by the external sources needed for the description of the nonlinear terms of the BRST-transformations.
It is well known that in the commutative case the perturbative finiteness of the Chern-Simons model is governed by the VSUSY within the algebraic renormalization procedure .
In the presented notes we are able to show that the Chern-Simons model is finite and independent of the deformation parameter $`\theta _{\mu \nu }`$ at least at the one loop level. Additionally, the perturbative calculations are in agreement with the restrictions coming from the VSUSY.
The letter is organized as follows. Section 2 gives a short presentation of $`\theta `$-deformed field theory. In section 3 we present the Chern-Simons theory as a $`\theta `$-deformed field theory on a noncommutative $`^3`$. The BRST-transformation and the VSUSY are introduced at the tree level. Section 4 is devoted to sketch the perturbative calculation at the one loop level.
## 2 $`\theta `$-deformed Field Theory
The noncommutative $`^3`$ is defined as the algebra $`𝒜_x`$ generated by $`\widehat{x}_\mu `$, $`\mu =1,2,3`$ satisfying the commutation relation
$$[\widehat{x}_\mu ,\widehat{x}_\nu ]=\mathrm{i}\theta _{\mu \nu },$$
(1)
where $`\theta _{\mu \nu }`$ is a real, constant, antisymmetric matrix with rank 3. For a function $`f(x)`$ on ordinary $`^3`$ one associates an element $`W(f)`$ of the algebra $`𝒜_x`$ by
$$W(f)=\frac{1}{(2\pi )^{\frac{3}{2}}}d^3ke^{\mathrm{i}k_\mu \widehat{x}_\mu }\stackrel{~}{f}(k),$$
(2)
where $`\stackrel{~}{f}(k)`$ is the Fourier transform of $`f(x)`$
$$\stackrel{~}{f}(k)=\frac{1}{(2\pi )^{\frac{3}{2}}}d^3xe^{\mathrm{i}k_\mu x_\mu }f(x).$$
(3)
Using relation (2) the star product on $`^3`$ is defined by
$$W(f)W(g)=W(fg),$$
(4)
and one finds
$$(fg)(x)=\frac{d^3k}{(2\pi )^{\frac{3}{2}}}\frac{d^3p}{(2\pi )^{\frac{3}{2}}}e^{\mathrm{i}(k_\mu +p_\mu )x_\mu }e^{\frac{\mathrm{i}}{2}\theta _{\mu \nu }k_\mu p_\nu }\stackrel{~}{f}(k)\stackrel{~}{g}(p).$$
(5)
Relations (2) and (4) render a representation of functions on the algebra $`𝒜_x`$ , and thus allow us to examine gauge theory on $`𝒜_x`$ by considering the counterpart on ordinary $`^3`$ with the usual product replaced by the star product. From (5) follows
$$d^3x(fg)(x)=d^3xf(x)g(x).$$
(6)
Similarly, one gets for a triple product
$`\left(fgh\right)(x)`$ $`=`$ $`{\displaystyle \frac{d^3p_1}{(2\pi )^{\frac{3}{2}}}\frac{d^3p_2}{(2\pi )^{\frac{3}{2}}}\frac{d^3p_3}{(2\pi )^{\frac{3}{2}}}e^{\frac{\mathrm{i}}{2}\theta _{\mu \nu }\left[(p_1)_\mu (p_2)_\nu +(p_1)_\mu (p_3)_\nu +(p_2)_\mu (p_3)_\nu \right]}}`$ (7)
$`\times e^{\mathrm{i}(p_1+p_2+p_3)_\mu x_\mu }\stackrel{~}{f}(p_1)\stackrel{~}{g}(p_2)\stackrel{~}{h}(p_3).`$
With $`n`$ arbitrary fields $`\varphi _k`$ one can show the validity of cyclic permutations
$`{\displaystyle _^3}d^3x\left(\varphi _1\varphi _2\mathrm{}\varphi _n\right)(x)`$ $`=`$ $`(1)^{g_1(g_2+\mathrm{}+g_n)}{\displaystyle _^3}d^3x\left(\varphi _2\mathrm{}\varphi _n\varphi _1\right)(x),`$ (8)
where $`g_i`$ is the total grading of the field $`\varphi _i`$. Additionally, one can show that the functional differentiation holds too, i.e.
$`{\displaystyle \frac{\delta }{\delta \varphi _1(y)}}{\displaystyle _^3}d^3x\left(\varphi _1\varphi _2\mathrm{}\varphi _n\right)(x)`$ $`=`$ $`\left(\varphi _2\mathrm{}\varphi _n\right)(y).`$ (9)
Equations (8) and (9) imply now that the off-shell algebra of the relevant functional operators describing the symmetry content of the model remains valid.
## 3 BRST-symmetry and Linear VSUSY
With a Lie-algebra valued gauge field $`A_\mu (x)=A_\mu ^a(x)T^a`$, where $`T^a`$ are the $`N^2`$ generators of the U($`N`$)-group,<sup>1</sup><sup>1</sup>1Not all gauge groups are realizable in noncommutative field theory $`a=0,1,\mathrm{},(N^21)`$
$$\left(T^0\right)_{mn}=\left(\frac{2}{N}\right)^{\frac{1}{2}}\delta _{mn},\left(T^a\right)_{mn}$$
with
$$[T^a,T^b]=2\mathrm{i}f^{abc}T^c,\{T^a,T^b\}=2d^{abc}T^c,\text{Tr}\left(T^aT^b\right)=2\delta ^{ab},$$
(10)
one can define the gauge-invariant classical metric independent action as
$$\mathrm{\Sigma }_{cl}=d^3x\frac{1}{2}ϵ_{\mu \nu \rho }\text{Tr}\left(A_\mu _\nu A_\rho \frac{2\mathrm{i}}{3}A_\mu A_\nu A_\rho \right),$$
(11)
which is invariant under the infinitesimal gauge transformation
$$\delta A_\mu =_\mu \lambda +\mathrm{i}\left(\lambda A_\mu A_\mu \lambda \right)D_\mu \lambda ,$$
(12)
where $`\lambda `$ is a Lie-algebra valued gauge parameter. The corresponding field strength is given by
$$F_{\mu \nu }=_\mu A_\nu _\nu A_\mu \mathrm{i}\left(A_\mu A_\nu A_\nu A_\mu \right).$$
(13)
In order to quantize the model within the BRST-scheme, the gauge-symmetry is replaced by the nilpotent BRST-symmetry
$$sA_\mu =D_\mu c,sc=\mathrm{i}cc,$$
(14)
where $`c`$ is the anticommuting Faddeev-Popov ghost field. Within the quantization procedure a BRST-invariant gauge-fixing must be introduced in the following manner
$$\mathrm{\Sigma }_{gf}=sd^3x\text{Tr}\left(\overline{c}_\mu A_\mu \right),$$
(15)
with
$$s\overline{c}=B,sB=0,$$
(16)
where $`\overline{c}`$ is the antighost field and $`B`$ the multiplier field implementing the Landau gauge. The gauge fixing part of the action of course depends on the metric, chosen here as the flat Euclidian one $`\delta _{\mu \nu }`$. The total action is now
$$\mathrm{\Sigma }=\mathrm{\Sigma }_{cl}+\mathrm{\Sigma }_{gf}.$$
(17)
Besides the BRST-invariance, (17) possesses an additional global supersymmetry, whose generators carry the Lorentz index .
$`\delta _\nu A_\tau `$ $`=`$ $`ϵ_{\mu \nu \tau }_\mu \overline{c},`$
$`\delta _\nu c`$ $`=`$ $`A_\nu ,`$
$`\delta _\nu B`$ $`=`$ $`_\nu \overline{c},`$
$`\delta _\nu \overline{c}`$ $`=`$ $`0.`$ (18)
The operator $`\delta _\nu `$ gives rise together with the BRST-operator to the anticommutation relations
$`\delta _\nu \delta _\tau +\delta _\tau \delta _\nu `$ $`=`$ $`0,`$ (19)
$`\delta _\nu s+s\delta _\nu `$ $`=`$ $`_\nu +\text{(eq. of motion)},`$ (20)
which close on-shell on space time translations. In order to describe the symmetry content of the model with respect to the BRST-symmetry and the linear VSUSY, one has to write down the corresponding Ward-identities (WI) at the classical level. In order to carry out this procedure, one introduces external unquantized sources for the nonlinear pieces of the BRST-transformations
$$\mathrm{\Sigma }_{ext}=d^3x\text{Tr}\left(\rho _\mu sA_\mu +\sigma sc\right).$$
(21)
Then the total corresponding tree-level action is given by
$`\mathrm{\Sigma }`$ $`=`$ $`\mathrm{\Sigma }_{cl}+\mathrm{\Sigma }_{gf}+\mathrm{\Sigma }_{ext}`$ (22)
$`=`$ $`{\displaystyle d^3x\frac{1}{2}ϵ_{\mu \nu \rho }\text{Tr}\left(A_\mu _\nu A_\rho \frac{2\mathrm{i}}{3}A_\mu A_\nu A_\rho \right)}`$
$`+`$ $`{\displaystyle d^3x\text{Tr}\left(B_\mu A_\mu \overline{c}_\mu D_\mu c\right)}`$
$`+`$ $`{\displaystyle d^3x\text{Tr}\left(\rho _\mu D_\mu c+\mathrm{i}\sigma cc\right)}.`$
The total action satisfies the nonlinear, $`\theta `$-deformed Slavnov-Taylor identity
$$𝒮\left(\mathrm{\Sigma }\right)=0,$$
(23)
where
$$𝒮\left(\mathrm{\Sigma }\right)=d^3x\text{Tr}\left(\frac{\delta \mathrm{\Sigma }}{\delta \rho _\mu }\frac{\delta \mathrm{\Sigma }}{\delta A_\mu }+\frac{\delta \mathrm{\Sigma }}{\delta \sigma }\frac{\delta \mathrm{\Sigma }}{\delta c}+B\frac{\delta \mathrm{\Sigma }}{\delta \overline{c}}\right),$$
(24)
which describes the BRST-symmetry. The VSUSY is characterized by the following integrated functional differential operator
$`𝒲_\mu `$ $`=`$ $`{\displaystyle d^3x\text{Tr}\left(ϵ_{\nu \mu \tau }\left(_\nu \overline{c}+\rho _\nu \right)\frac{\delta }{\delta A_\tau }A_\mu \frac{\delta }{\delta c}_\mu \overline{c}\frac{\delta }{\delta B}\sigma \frac{\delta }{\delta \rho _\mu }\right)}.`$ (25)
The WI operator (25) yields the linearly broken supersymmetry
$$𝒲_\mu \mathrm{\Sigma }=\mathrm{\Delta }_\mu ^{cl},$$
(26)
with
$$\mathrm{\Delta }_\mu ^{cl}=d^3x\text{Tr}\left(ϵ_{\nu \mu \tau }\rho _\nu _\tau B+\rho _\tau _\mu A_\tau \sigma _\mu c\right),$$
(27)
where $`\mathrm{\Delta }_\mu ^{cl}`$ is linear in the quantum fields.
## 4 One-loop Calculations
In order to check the one-loop UV and IR behaviour of the noncommutative Chern-Simons model, one needs the corresponding Feynman rules. For the various propagators of the model only the bilinear part of the full action (22) is relevant. However, due to the fact that noncommutativity does not affect bilinear parts of the action (6), the propagators are the usual ones . In momentum space one therefore has
$`\stackrel{~}{G}^{c\overline{c},ab}(k)`$ $`=`$ $`{\displaystyle \frac{\delta ^{ab}}{(2\pi )^{\frac{3}{2}}}}{\displaystyle \frac{1}{k^2}},`$ (28)
$`\stackrel{~}{G}_{\mu \nu }^{AA,ab}(k)`$ $`=`$ $`{\displaystyle \frac{\mathrm{i}\delta ^{ab}}{(2\pi )^{\frac{3}{2}}}}{\displaystyle \frac{ϵ_{\mu \nu \lambda }k_\lambda }{k^2}},`$ (29)
$`\stackrel{~}{G}_\mu ^{AB,ab}(k)`$ $`=`$ $`{\displaystyle \frac{\mathrm{i}\delta ^{ab}}{(2\pi )^{\frac{3}{2}}}}{\displaystyle \frac{k_\mu }{k^2}}.`$ (30)
The interaction piece of the action (22) determines the corresponding Feynman-rules for the three gluon vertex and the ghost-antighost-gluon vertex:
$`\mathrm{\Sigma }_{int}`$ $`=`$ $`{\displaystyle d^3x\text{Tr}\left(\frac{\mathrm{i}}{3}ϵ_{\mu \nu \rho }A_\mu A_\nu A_\rho +\mathrm{i}_\mu \overline{c}\left(cA_\mu A_\mu c\right)\right)}`$ (31)
$`=`$ $`\text{Tr}^{abc}{\displaystyle d^3x\left(\frac{\mathrm{i}}{3}ϵ_{\mu \nu \rho }A_\mu ^aA_\nu ^bA_\rho ^c+\mathrm{i}_\mu \overline{c}^ac^bA_\mu ^c\mathrm{i}_\mu \overline{c}^aA_\mu ^bc^c\right)},`$
where $`\text{Tr}^{abc}=\text{Tr}\left(T^aT^bT^c\right)`$. From (10) follows
$$\text{Tr}^{abc}=2\left(\mathrm{i}f^{abc}+d^{abc}\right),$$
(32)
and therefore the corresponding Feynman rules for the interaction vertices in momentum space become
$`\stackrel{~}{V}_{\mu \nu \rho }^{AAA,abc}(p_1,p_2,p_3)={\displaystyle \frac{1}{(2\pi )^{\frac{3}{2}}}}ϵ_{\mu \nu \rho }\mathrm{\Omega }_\theta ^{abc}(p_2,p_3),`$ (33)
$`\stackrel{~}{V}_\mu ^{cA\overline{c},abc}(p_1,p_2,p_3)={\displaystyle \frac{1}{(2\pi )^{\frac{3}{2}}}}\mathrm{i}(p_3)_\mu \mathrm{\Omega }_\theta ^{abc}(p_2,p_3),`$ (34)
where $`\mathrm{\Omega }_\theta ^{abc}(p_2,p_3)`$ is defined by
$$\mathrm{\Omega }_\theta ^{abc}(p_2,p_3)=4\left[f^{abc}\text{cos}\left(\frac{1}{2}p_2\times p_3\right)d^{abc}\text{sin}\left(\frac{1}{2}p_2\times p_3\right)\right],$$
(35)
with $`p_2\times p_3=\theta _{\mu \nu }(p_2)_\mu (p_3)_\nu `$. In order to carry out explicit calculations the following contractions of the structure constants $`f^{abc}`$ and $`d^{abc}`$ are useful
$`f^{abc}f^{dbc}`$ $`=`$ $`N(\delta ^{ad}\delta ^{a0}\delta ^{d0}),`$ (36)
$`d^{abc}d^{dbc}`$ $`=`$ $`N(\delta ^{ad}+\delta ^{a0}\delta ^{d0}).`$ (37)
From the commutation relations we have additionally,
$`f^{abc}`$ $`=`$ $`{\displaystyle \frac{1}{4\mathrm{i}}}\text{Tr}\left([T^a,T^b]T^c\right),`$
$`d^{abc}`$ $`=`$ $`{\displaystyle \frac{1}{4}}\text{Tr}\left(\{T^a,T^b\}T^c\right).`$ (38)
In order to verify the relations (36) and (37) one has to use the following completeness relation for generators $`T^a`$ of the $`U(N)`$ gauge group
$$\left(T^a\right)_{mn}\left(T^a\right)_{m^{}n^{}}=2\delta _{mn^{}}\delta _{nm^{}},m,n=1,\mathrm{},N.$$
(39)
Before doing explicitly the one loop analysis we want to stress that already at the tree level the consequences of the VSUSY WI are fulfilled. In doing a Legendre transformation one gets from the WI (26) that the propagator of the gluon field and the ghost-antighost-field propagator are related. In momentum space one obtains
$$ϵ_{\mu \nu \lambda }\mathrm{i}k_\lambda \stackrel{~}{G}^{c\overline{c},ab}(k)=\stackrel{~}{G}_{\mu \nu }^{AA,ab}(k).$$
(40)
For the three point vertices the following WI follow directly from (26)
$$\mathrm{i}\left(p_3\right)_\nu ϵ_{\mu \nu \tau }\stackrel{~}{V}_{\lambda \sigma \tau }^{AAA,abc}(p_1,p_2,p_3)=\delta _{\mu \sigma }\stackrel{~}{V}_\lambda ^{cA\overline{c},abc}(p_1,p_2,p_3)\delta _{\mu \lambda }\stackrel{~}{V}_\sigma ^{cA\overline{c},abc}(p_1,p_2,p_3).$$
(41)
These relations are indeed responsible for the one loop perturbative finiteness. In fact, one finds that the gluon loop and the ghost loop contribution to the gluon self energy $`\mathrm{\Pi }_{\mu \nu }^{ab}`$ are equal up to a sign, and thus cancel. A similar result is found for the one loop vertex corrections which therefore also vanish. Therefore, the loop cancellation present in the ordinary Chern-Simons model persists in the noncommutative case, i.e. the noncommutative Chern-Simons model is finite to first loop order.
In the following we will explicitely perform the one loop analysis of the gluon self energy graph (a) of fig.1 and show that the ’non-planar’ contribution arising due to the noncommutativity is finite. Using the above Feynman rules one calculates the following expression for the gluon self-energy with an internal gluon field,
$$\mathrm{\Pi }_{\mu \nu }^{ab}(p,p)=\frac{1}{(2\pi )^6}\frac{d^3k}{(2\pi )^{\frac{3}{2}}}\frac{\left(k_+\right)_\mu \left(k_{}\right)_\nu +\left(k_+\right)_\nu \left(k_{}\right)_\mu }{k_+^2k_{}^2}\mathrm{\Omega }_\theta ^{acd}(k_+,k_{})\mathrm{\Omega }_\theta ^{cbd}(p,k_{}),$$
(42)
where $`k_+=k+\frac{p}{2}`$ and $`k_{}=k\frac{p}{2}`$. Using now (36) and (37) one gets
$$\mathrm{\Pi }_{\mu \nu }^{ab}(p,p)=\frac{16N}{(2\pi )^6}\frac{d^3k}{(2\pi )^{\frac{3}{2}}}\frac{\left(k_+\right)_\mu \left(k_{}\right)_\nu +\left(k_+\right)_\nu \left(k_{}\right)_\mu }{k_+^2k_{}^2}\left[\delta ^{ab}\delta ^{a0}\delta ^{b0}\frac{1}{2}\left(e^{\mathrm{i}p\times k}+e^{\mathrm{i}p\times k}\right)\right].$$
(43)
The first ’planar’ term is just the result which corresponds to the commutative model. From the result (43) it is also seen that one has to use a $`U(N)`$ gauge group in order to have a nontrivial result. Now one can use the result of with an appropriate regularization scheme in order to discuss the $`\theta `$-independent expression of eq. (43). The further discussion of the $`\theta `$-dependent, ’non-planar’ part of (43) can be done with the techniques presented already in .
Using Schwinger parametrization
$$\frac{1}{k_+^2}=_0^{\mathrm{}}𝑑\alpha _+e^{\alpha _+k_+^2}$$
(44)
we discuss now
$$\mathrm{\Pi }_{\mu \nu }^{}(p,p)=\frac{d^3k}{(2\pi )^{\frac{3}{2}}}\frac{\left(k_+\right)_\mu \left(k_{}\right)_\nu +\left(k_+\right)_\nu \left(k_{}\right)_\mu }{k_+^2k_{}^2}\left(e^{\mathrm{i}p\times k}+e^{\mathrm{i}p\times k}\right),$$
(45)
and get
$`\mathrm{\Pi }_{\mu \nu }^{}(p,p)`$ $`=`$ $`{\displaystyle \frac{d^3k}{(2\pi )^{\frac{3}{2}}}_0^{\mathrm{}}𝑑\alpha _+_0^{\mathrm{}}𝑑\alpha _{}\left(\left(k_+\right)_\mu \left(k_{}\right)_\nu +\left(k_+\right)_\nu \left(k_{}\right)_\mu \right)e^{\alpha _+k_+^2\alpha _{}k_{}^2}}`$ (46)
$`\times \left(e^{\mathrm{i}p\times k}+e^{\mathrm{i}p\times k}\right).`$
In defining $`\stackrel{~}{p}_\mu =p_\lambda \theta _{\lambda \mu }`$ and
$$I_{\mu \nu }(\eta ,p)=\left\{\left(\frac{}{y_\mu }\frac{}{z_\nu }+\frac{}{y_\nu }\frac{}{z_\mu }\right)\frac{d^3k}{(2\pi )^{\frac{3}{2}}}e^{\alpha _+\left(k+\frac{p}{2}\right)^2\alpha _{}\left(k\frac{p}{2}\right)^2+\mathrm{i}\eta \stackrel{~}{p}ky\left(k+\frac{p}{2}\right)z\left(k\frac{p}{2}\right)}\right\}|_{y=z=0}.$$
(47)
Eq. (46) can be written as
$$\mathrm{\Pi }_{\mu \nu }^{}(p,p)=_0^{\mathrm{}}𝑑\alpha _+_0^{\mathrm{}}𝑑\alpha _{}\left(I_{\mu \nu }(1,p)+I_{\mu \nu }(1,p)\right).$$
(48)
In order to carry out the $`k`$-integration one introduces new variables
$$k_\lambda ^{}=k_\lambda +\frac{B_\lambda }{\alpha _++\alpha _{}}k_\lambda +\frac{B_\lambda }{\beta },$$
(49)
with
$$B_\lambda =\frac{1}{2}\left[\left(\alpha _+\alpha _{}\right)p_\lambda \mathrm{i}\eta \stackrel{~}{p}_\lambda +\left(y+z\right)_\lambda \right],$$
(50)
in order to use gaussian integration
$$\frac{d^3k^{}}{(2\pi )^{\frac{3}{2}}}e^{\beta k_{}^{}{}_{}{}^{2}}=\frac{1}{(2\beta )^{\frac{3}{2}}}.$$
(51)
In this way one gets for (47)
$$I_{\mu \nu }(\eta ,p)=\left\{\left(\frac{}{y_\mu }\frac{}{z_\nu }+\frac{}{y_\nu }\frac{}{z_\mu }\right)\frac{1}{(2\beta )^{\frac{3}{2}}}e^{\frac{B^2}{\beta }\frac{\beta p^2}{4}(yz)\frac{p}{2}}\right\}|_{y=z=0}.$$
(52)
Doing explicitly the differentiation one finds
$$I_{\mu \nu }(\eta ,p)=\frac{1}{(2\beta )^{\frac{3}{2}}}e^{\frac{\alpha _+\alpha _{}}{\beta }p^2\frac{\eta ^2}{4\beta }\stackrel{~}{p}^2}\left\{\frac{1}{\beta }\delta _{\mu \nu }2\left(\frac{\alpha _+\alpha _{}}{\beta ^2}p_\mu p_\nu +\eta ^2\frac{\stackrel{~}{p}_\mu \stackrel{~}{p}_\nu }{4\beta ^2}\right)+\text{terms linear in }\eta \right\},$$
(53)
implying the following result
$$\mathrm{\Pi }_{\mu \nu }^{}(p,p)=2_0^{\mathrm{}}𝑑\alpha _+_0^{\mathrm{}}𝑑\alpha _{}e^{\frac{\alpha _+\alpha _{}}{\beta }p^2\frac{\stackrel{~}{p}^2}{4\beta }}\frac{1}{(2\beta )^{\frac{3}{2}}}\left\{\frac{\delta _{\mu \nu }}{\beta }2\left(\frac{\alpha _+\alpha _{}}{\beta ^2}p_\mu p_\nu +\frac{\stackrel{~}{p}_\mu \stackrel{~}{p}_\nu }{2\beta ^2}\right)\right\}.$$
(54)
It remains to study the parametric integration (54). One has two different types of integrals (with $`\alpha =1,2`$)
$$I_\alpha ^{(i)}=_0^{\mathrm{}}𝑑\alpha _+_0^{\mathrm{}}𝑑\alpha _{}\frac{1}{\beta ^{\frac{3}{2}+\alpha }}e^{\frac{\alpha _+\alpha _{}}{\beta }p^2\frac{\stackrel{~}{p}^2}{4\beta }}.$$
(55)
$$I^{(ii)}=_0^{\mathrm{}}𝑑\alpha _+_0^{\mathrm{}}𝑑\alpha _{}\frac{\alpha _+\alpha _{}}{\beta ^{\frac{7}{2}}}e^{\frac{\alpha _+\alpha _{}}{\beta }p^2\frac{\stackrel{~}{p}^2}{4\beta }}.$$
(56)
With the reparametrization
$`\alpha _+`$ $`=`$ $`\xi \lambda ,`$
$`\alpha _{}`$ $`=`$ $`(1\xi )\lambda ,`$
$`\rho `$ $`=`$ $`\xi (1\xi )p^2\lambda ,`$ (57)
one arrives at
$$I_\alpha ^{(i)}=_0^1𝑑\xi \left(\xi (1\xi )p^2\right)^{\left(\alpha \frac{1}{2}\right)}_0^{\mathrm{}}𝑑\rho \frac{1}{\rho ^{\left(\frac{1}{2}+\alpha \right)}}e^{\left(\rho +\frac{\xi (1\xi )p^2\stackrel{~}{p}^2}{4\rho }\right)},$$
(58)
and
$$I^{(ii)}=\left(\frac{1}{p^2}\right)^{\frac{1}{2}}_0^1𝑑\xi \left[\xi (1\xi )\right]^{\frac{1}{2}}_0^{\mathrm{}}\frac{d\rho }{\rho ^{\frac{1}{2}}}e^{\left(\rho +\frac{\xi (1\xi )p^2\stackrel{~}{p}^2}{4\rho }\right)}$$
(59)
The $`\rho `$-integration is carried out with the formula
$$_0^{\mathrm{}}\frac{dx}{x^{1\nu }}e^{\gamma x\frac{\beta }{x}}=2\left(\frac{\beta }{\gamma }\right)^{\frac{\nu }{2}}K_\nu \left(2\sqrt{\beta \gamma }\right),$$
(60)
where $`K_\nu `$ is the modified Bessel function . In this way one obtains
$$I_\alpha ^{(i)}=2\left(\frac{4p^2}{\stackrel{~}{p}^2}\right)^{\left(\alpha \frac{1}{2}\right)\frac{1}{2}}_0^1𝑑\xi \left[\xi (1\xi )\right]^{\left(\alpha \frac{1}{2}\right)\frac{1}{2}}K_{\frac{1}{2}\alpha }\left(z\right).$$
(61)
where $`z=(\xi (1\xi )p^2\stackrel{~}{p}^2)^{\frac{1}{2}}`$. Additionally, one has
$$I^{(ii)}=\left(\frac{1}{p^2}\right)^{\frac{1}{2}}\sqrt{2}_0^1𝑑\xi \left[\xi (1\xi )\right]^{\frac{1}{2}}\left[\xi (1\xi )p^2\stackrel{~}{p}^2\right]^{\frac{1}{4}}K_{\frac{1}{2}}(z).$$
(62)
With
$$K_{\frac{1}{2}}(z)=K_{\frac{1}{2}}(z)=\left(\frac{\pi }{2z}\right)^{\frac{1}{2}}e^z,$$
(63)
and
$$K_{\nu 1}(z)K_{\nu +1}(z)=\frac{2\nu }{z}K_\nu (z),$$
(64)
one can show that $`I_1^{(i)}`$ behaves as
$$I_1^{(i)}\left(\frac{1}{\stackrel{~}{p}^2}\right)^{\frac{1}{2}}\times \text{finite},$$
(65)
which shows the usual IR-singularity for $`\stackrel{~}{p}0`$. Whereas $`I_2^{(i)}`$ develops two contributions of the form
$$I_2^{(i)}\left(p^2\right)^{\frac{1}{2}}\left(\frac{1}{\stackrel{~}{p}^2}\right)\times \text{finite}+\left(\frac{1}{\stackrel{~}{p}^2}\right)^{\frac{3}{2}}\times \text{finite}.$$
(66)
These last terms must be multiplied by $`\stackrel{~}{p}_\mu \stackrel{~}{p}_\nu `$. Thus the second term of (66) reproduces an IR-singularity for $`\stackrel{~}{p}0`$, whereas the first term leads to IR-regular, $`\theta `$-independent matrix elements.
A similar calculation for $`I^{(ii)}`$ yields
$$I^{(ii)}\left(\frac{1}{p^2}\right)^{\frac{1}{2}}\times \text{finite},$$
(67)
a $`\theta `$-independent result. By setting $`\theta =0`$ the second term of (43) will produce a term proportional to $`\delta ^{0a}\delta ^{0b}`$, (67). This term combines with the first part of (43) to reproduce the result of the commutative $`SU(N)`$ model.
## 5 Conclusion and Outlook
In this short note we have demonstrated the finiteness of the $`\theta `$-deformed Chern-Simons theory and its $`\theta _{\mu \nu }`$-independence at the one-loop level.
Additionally, the obtained results are in agreement with the restrictions coming from the VSUSY. In the limit $`\theta _{\mu \nu }0`$ one gets the old results of . Furthermore, since no radiative corrections are present, the model is regular in this limit (see also and ).
In a forthcoming paper we will discuss further topological field models of Schwarz-type and Witten-type.
## 6 Acknowledgements
It is our pleasure to thank H. Grosse, H. Hüffel, R. Wulkenhaar and C. Rupp for helpful discussions in a weekly seminar.
|
warning/0004/hep-th0004129.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
It has long been thought that ’t Hooft’s $`N_c\mathrm{}`$ limit of $`SU(N_c)`$ gauge theory might be usefully described by some sort of string theory. However, there is an apparently devastating argument, that this “QCD String” (a.k.a. a tower of glueballs) is not fundamental string (of “string theory”): the graviton appears in the spectrum of the latter, contradicting the well-known folk theorem forbidding massless spin 2 bound states in a Poincaré invariant quantum field theory<sup>§</sup><sup>§</sup>§Incidentally, the conceptual problems with the formulation of string theory (and its low energy limit, quantum gravity) would disappear if string (including its graviton) were a composite structure arising in ordinary flat space quantum field theory. Sakharov proposed this idea for quantum gravity long ago , and it was later vigorously explored by Adler and Zee . Early in this game, it was realized that the underlying flat space theory cannot be Poincaré invariant, because of this same folk theorem. My suggestion that string be a composite of string bits also evades the theorem because the bits live in at least one less space dimension than string, so the space-time symmetry of the underlying string bit dynamics is only a subgroup of the Poincaré group of string..
A way to evade this argument has been shown by the conjectured equivalence between classical IIB superstring theory on an AdS<sub>5</sub> background and $`𝒩=4`$ supersymmetric $`SU(\mathrm{})`$ Yang-Mills on flat 4 dimensional Minkowski space-time . The point is that in this example the graviton lives in 5 space-time dimensions and the flat space-time global symmetry (Poincaré(3,1)) is only a subgroup of Poincaré(4,1), which is realized locally, not globally. Thus the “massless” 5 dimensional graviton is a composite of the quanta of a flat-space quantum field theory in 4 dimensional space-time. There is no massless spin 2 particle in this 4d quantum field theory and no folk theorems are violated. In the $`𝒩=4`$ conformally invariant example, the projection of the 5d graviton onto the 4d Minkowski boundary of AdS<sub>5</sub> is a multi-gluon continuum state. But if the mechanism can be extended to the non-conformally invariant gauge theory describing the gluon sector of QCD, the graviton 4d remnant would presumably be a massive spin 2 glueball.
To move these statements beyond conjecture, one clearly needs to establish the “dual” description starting from either of the supposedly equivalent theories. I think it is clear that the best starting point for such a project is the flat space quantum field theory. Unlike the previous conjectured dualities in string theory, which asserted the equivalence of pairs of poorly defined theories, this duality asserts the equivalence of a poorly defined theory (string or quantum gravity) to a perfectly well defined theory (asymptotically free or conformally invariant quantum gauge theory on flat 4d space-time). Indeed, I am inclined to regard this “duality” as more analogous to the alternate descriptions of superconductivity given by BCS theory on the one-hand and Landau-Ginzburg theory on the other. If this metaphor holds, the flat space quantum field theory should be embraced as the long sought microscopic formulation of string/quantum gravity.
As ’t Hooft showed in his pioneering paper , the $`N_c\mathrm{}`$ limit of $`SU(N_c)`$ Yang-Mills theory reduces to a certain sum of planar Feynman diagrams. Elegant techniques, involving the explicit elimination of the off-diagonal matrix elements of the matrix field, have been used to obtain this limit in matrix theories of extremely low space-time dimension (namely D=0,1) , but these methods have failed to deal with theories with space-time dimension $`D>1`$. At the moment, I see no better approach to the $`D>1`$ case than setting up a framework to carry out the direct summation of planar graphs. In the mid-1970’s, motivated by the success of light-cone quantization of string theory , I proposed that planar diagram sums be carried out by using light-cone parameterization $`x^\pm (t\pm z)/\sqrt{2}`$ and that a convenient way to digitize the sum was to discretize the momentum conjugate to $`x^{}`$, $`P^+=lm`$ with $`l=1,2,\mathrm{}`$, and imaginary light-cone time $`ix^+=ka`$, with $`k=1,2,\mathrm{}`$. In those first papers I restricted attention to scalar field theories, but Brower, Giles, and I soon made a first attempt to extend the approach to QCD . In our setup, the strong ’t Hooft coupling limit $`N_cg^2\mathrm{}`$ favors the fishnet diagrams that lead to a light-cone string interpretation. Of course, by its very nature a strong coupling limit probes the microscopic details of discretization, and can at best show only rough qualitative resemblance to the continuum theory. Even so, there were a number of loose ends and unsatisfactory features of this first discretization of QCD which needed to be addressed.
Motivated by the goal of discovering a more definitive string description of large $`N_c`$ QCD Klaus Bering, Joel Rozowsky, and I set out to remedy these shortcomings, and in this talk I would like to tell you about the results of our efforts . As you will see, we have obtained a much improved discretization setup, but have just begun to explore its usefulness in capturing a string picture of the sum of planar diagrams.
## 2 String on a Light-Front
An evolving string sweeps out a world sheet $`x^\mu (\sigma ^1,\sigma ^0)`$ in space-time. One can choose the parameters so that $`x^+=\sigma ^0`$, and so that $`𝒫^+`$, the density of $`P^+`$, is a constant $`T_0`$. Then evolution in $`x^+`$ is generated by the Hamiltonian
$`HP^{}={\displaystyle _0^{P^+/T_0}}𝑑\sigma \left[{\displaystyle \frac{𝒫^2}{2T_0}}+{\displaystyle \frac{T_0}{2}}𝐱^2\right],`$ (2.1)
where for brevity we have called $`\sigma ^1=\sigma `$ and the prime denotes differentiation with respect to $`\sigma `$. Of course $`𝒫(\sigma )`$ is the momentum conjugate to $`𝐱(\sigma )`$. A key novelty here is that $`P^+/T_0`$ measures the quantity of string present, and its interpretation as a component of momentum is secondary and derivative. In this way the string is seen to enjoy a Galilei invariant dynamics as it moves only in the $`d2`$ dimensional transverse space.
As shown by Mandelstam , interactions are easily introduced by using the path history form of quantum mechanics and including histories in which strings break and join. Technically this is accomplished by first obtaining the imaginary time $`ix^+\tau `$ path integral representation of $`f|e^{\beta H}|i`$ for free string. The action in this case is just
$`S^{Free}(\beta )={\displaystyle \frac{T_0}{2}}{\displaystyle _0^\beta }𝑑\tau {\displaystyle _0^{P^+/T_0}}𝑑\sigma \left[\dot{𝐱}^2+𝐱^2\right].`$ (2.2)
Here the action is seen to be an integral over a simply connected rectangular domain for an open free string and a cylinder for a closed string. A diagram describing an arbitrary number of splits and joins is obtained by allowing some number of cuts, each at constant $`\sigma `$ but of varying length, within the domain. The ends of these cuts mark the splitting or joining points, and $`𝐱`$ is discontinuous across them. Calling the generic such domain $`\mathrm{\Sigma }`$, the complete amplitude is then given by
$$=\underset{\mathrm{\Sigma }}{}𝒟𝐱e^{S(\beta ,\mathrm{\Sigma })}.$$
(2.3)
## 3 Discretization
In order to give a nonperturbative definition of the path integrals appearing in the previous section, Roscoe Giles and I introduced a lattice version of the domains $`\mathrm{\Sigma }`$. It was only necessary to discretize $`\tau `$ and $`\sigma `$. So we set $`\beta =(N+1)a`$ and $`P^+=MmMaT_0`$, with $`M`$ and $`N`$ fixed positive integers. Then $`𝐱(\sigma ,\tau )`$ is replaced by $`𝐱_{lk}`$, and the functional integration by ordinary integrals. Finally, the action is simply replaced by
$$S=\frac{T_0}{2}\underset{L\mathrm{\Sigma }}{}\mathrm{\Delta }𝐱_L^2,$$
(3.1)
where $`L`$ labels a link on the lattice. Links between nearest neighbor sites in the $`\tau `$ direction are all present, but those in the $`\sigma `$ direction are occasionally absent reflecting the possibility of splits and joins. It is a highly nontrivial fact that this apparently noncovariant setup turns out, after the continuum limit, to be fully compatible with Poincaré invariance in the critical dimension.
## 4 Feynman Diagrams on a Light-Front
To understand how Feynman diagrams look in light-cone parameterization, consider the mixed representation of the scalar field propagator:
$$\mathrm{\Delta }(𝐩,p^+,ix^+)=\theta (x^+p^+)\frac{e^{ix^+(𝐩^2+\mu ^2)/2p^+}}{2|p^+|}.$$
(4.1)
For simplicity we may establish the convention that each line propagates forward in $`x^+`$, and correspondingly $`p^+>0`$. Also we may pass to imaginary time, $`\tau =ix^+>0`$ and then write
$$\mathrm{\Delta }(𝐩,p^+,\tau )=\frac{e^{\tau (𝐩^2+\mu ^2)/2p^+}}{2p^+},\stackrel{~}{\mathrm{\Delta }}(𝐱,p^+,\tau )=\left(\frac{p^+}{2\pi \tau }\right)^{d/2}\frac{e^{p^+𝐱^2/2\tau \mu ^2\tau /2p^+}}{2p^+}$$
(4.2)
where the second form is in transverse coordinate representation and $`d=D2`$ is the dimensionality of transverse space.
Discretizing $`\tau =ka`$ and $`p^+=lm=laT_0`$, $`k,l=1,2,\mathrm{}`$, the coordinate propagator becomesThe discretization of $`\tau `$ provides a universal ultraviolet cutoff, and every diagram will therefore be finite. In contrast, the DLCQ industry keeps time continuous, and must regulate ultraviolet divergences in some other fashion.
$$\stackrel{~}{\mathrm{\Delta }}_{lk}(𝐱)=\left(\frac{lT_0}{2k\pi }\right)^{d/2}\frac{e^{T_0(l/2k)𝐱^2+k\mu ^2/2lT_0}}{2lm}.$$
(4.3)
Comparing to the previous section, we see that we can crudely think of a planar Feynman diagram as a (coarsely) discretized world sheet, with a dynamical link dependent string tension $`T_{lk}=(l/k)T_0`$. Each link has its own independent $`k,l`$ which are each summed over all positive integers. Of course only the large fishnet diagrams will bear any actual resemblance to a continuous world sheet! The sum over all planar diagrams would then define the QCD string dynamics as including an average over all such discretizations ranging from coarse to fine. Also note that a good world sheet path integral should have (effectively) only positive weights. For instance, for $`\lambda \varphi ^4`$ theory each vertex contributes a minus sign if $`\lambda >0`$. A good world sheet interpretation requires $`\lambda <0`$, the attractive unstable sign. Gauge theories have vertices of both signs, complicating a straightforward world sheet interpretation.
To illustrate the effect of our discretization on a standard diagrammatic calculation, consider the sum of those 2 to 2 scattering diagrams in $`\lambda \varphi ^4`$ theory shown in Fig. 1. We work in the transverse center of mass frame.
Assume that the (discrete) $`P^+/m`$ of the initial (final) particles is $`l,Ml`$ ($`r,Mr`$). Let $`E=M𝐩^2/2l(Ml)`$ be the initial energy ($`P^{}`$). Fix the discrete time of the first vertex at $`0`$, and sum over all diagrams in which the final particles both propagate to time $`k`$, multiply by $`e^{kaE}`$ and sum over all $`k`$ from 1 to $`\mathrm{}`$. The result for the off-energy shell S-matrix is then
$`S_{fi}`$ $`=`$ $`\delta _{rl}\delta (𝐩𝐩^{})+{\displaystyle \frac{\lambda }{32\pi ^3\sqrt{l(Ml)r(Mr)}}}`$ (4.4)
$`\left(e^{aE+M𝐩^2/2r(Mr)T_0}1\right)^1\left(1{\displaystyle \frac{\lambda (M1)}{16\pi ^2M}}\mathrm{ln}(1e^{aE})\right)^1`$
Compared to standard formal scattering theory, we see that instead of a factor $`1/(E_fEiϵ)`$ which acts in wavepackets like the standard energy conserving delta function, the use of discrete time has rendered this as $`1/(e^{aE_faE}1)`$. This replacement is easy to understand because with discrete imaginary time, the amplitudes should be periodic in $`E`$ with period $`2\pi i/a`$. It is thus apparent that the scattering amplitude should be identified with the coefficient of this factor. For $`\lambda <0`$, the scattering amplitude shows a bound state pole at a real negative value of $`E`$:
$$E_BB=\frac{1}{a}\mathrm{ln}\left[1e^{16\pi ^2M/\lambda (M1)}\right].$$
In the continuum limit, $`M\mathrm{},a0`$, one can make the pole location stay finite by tuning $`\lambda `$ to vanish logarithmically as $`a0`$, showing dimensional transmutation in an asymptotically free theory.
## 5 Discretization Setup for Yang-Mills Field Theory
The discretization of QCD initially attempted by Brower, Giles, and me , was based on a literal transcription of the Feynman rules in light-cone gauge. The transverse gauge field can be described in the complex basis $`(A_1\pm iA_2)`$ when it takes on the guise of a complex scalar field, described diagrammatically by attaching an arrow to each line of a Feynman diagram. The primitive quartic vertex conserves arrows, but the cubic vertices can act as sources or sinks of arrows. The longitudinal gauge field $`A_+`$ does not propagate and can be integrated out to yield an induced quartic vertex, which depends upon the $`P^+`$ values of the incoming legs in a singular way:
$$\mathrm{\Gamma }_{\mathrm{induced}}^4=g^2\frac{(P_1^++P_1^+)(P_2^++P_2^+)}{(P_1^+P_1^+)^2}.$$
(5.1)
Upon discretization, we adopted the drastic prescription of simply dropping the infinite contribution at $`P_1^+=P_1^+`$. Furthermore all tadpole diagrams had to be dropped, because of our insistence that no line propagate 0 time steps. Then the strong coupling limit singled out large planar diagrams involving only the primitive quartic couplings, thus leading to an evaluation similar to the $`\varphi ^4`$ example of the previous section. Unfortunately, these quartic couplings have mixed signs: an attractive interaction between gluons of parallel spin and repulsive between gluons of antiparallel spin. This ferromagnetic interaction pattern meant that our discretization led to a formal strong coupling limit in which the only long string that could form in the limit would have huge total spin. The essential problem is that attractive interactions between gluons of opposite spin arise in QCD from gluon exchange , and the discretization chosen in prevents anti-ferromagnetic gluon exchange from competing at strong coupling with the ferromagnetic quartic interaction. This, together with our drastic prescription for all of the $`P^+=0`$ ills of light-cone quantization, points to the need for a more refined discretized model to adequately describe the strong coupling behavior of large $`N_c`$ QCD.
### 5.1 Improved Discretization Rules
Clearly what is needed is a prescription that either enhances strong coupling gluon exchange or suppresses the strong coupling quartic interactions. We found it most natural to arrange the latter by abolishing all quartic interactions, primitive and induced, and replacing them with the exchange of short lived fictitious particles. This is shown for the primitive quartic interaction in Fig.2.
The dashed line is associated with the fictitious two-form propagator
$$\mathrm{\Delta }_{2\mathrm{form}}=h_ke^{k𝐐^2/2lT_0},\underset{k=1}{\overset{\mathrm{}}{}}h_k=1,$$
(5.2)
where the $`h_k`$ are tunable parameters which are required to vanish rapidly with $`k=1,2,\mathrm{}`$, the number of time steps propagated. We treat the induced quartic interaction in a similar fashion, giving the non-dynamical field $`A_+`$ a short-time propagator
$$\mathrm{\Delta }_+=f_ke^{k𝐐^2/2lT_0},\underset{k=1}{\overset{\mathrm{}}{}}f_k=1.$$
(5.3)
The presence of the tunable parameters $`f_k`$ and $`h_k`$ is very welcome, because they can be adjusted to arrange the cancelation of cut-off artifacts that can typically spoil Poincaré invariance in the continuum limit. As a bonus, we find that our prescription provides the appealing interpretation of tadpole diagrams indicated in Fig. 3.
The first two diagrams cancel exactly (which is fortunate since they can’t be drawn in our discrete model!) leaving the third diagram which can be drawn. Our complete set of Feynman rules is neatly summarized in Fig. 4 taken from our paper .
Note that to avoid clutter we have suppressed the double line notation so these rules are completely sufficient for all graphs of planar or cylindrical topology ($`N_c\mathrm{}`$). For general diagrams, the double line notation must be restored in order to properly account for the $`1/N_c`$ corrections.
An easy way to understand why a set of rules with only cubic vertices is possible, is to apply light-cone gauge to the Yang-Mills Lagrangian in first order form. The upshot is the Lagrangian
$``$ $`=`$ $`{\displaystyle \frac{1}{2}}\mathrm{Tr}A_kA_k+{\displaystyle \frac{1}{2}}\mathrm{Tr}\widehat{A}^2+\mathrm{Tr}\varphi _{kl}^22ig\mathrm{Tr}{\displaystyle \frac{1}{_{}}}A_k[_{}A_n_kA_n_kA_n_{}A_n]`$ (5.4)
$`ig\mathrm{Tr}{\displaystyle \frac{1}{_{}}}\widehat{A}[A_k,_{}A_k]ig\mathrm{Tr}\varphi _{kn}[A_k,A_n],`$
where $`\widehat{A}_{}A_+𝐀`$. This makes it clear why we called the fictitious scalar represented by the dashed lines a 2-form: it is a (pseudo) scalar only in 3 + 1 dimensions.
### 5.2 One Loop Self Energy
Quite apart from its use as a facilitator for a strong coupling expansion, our discretization can also serve as a novel way to regulate the diagrams of weak coupling perturbation theory. To illustrate this aspect, we quote the result for the one loop gluon self-energy, with discretization in place:
$`\mathrm{\Pi }_2(Q^2)`$ $`=`$ $`{\displaystyle \frac{g^2N_c}{4\pi ^2}}[{\displaystyle \underset{k=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{u^k}{k^2}}(4M[\psi (M)+\gamma ]{\displaystyle \frac{(M1)(11M1)}{3M}})`$ (5.5)
$`{\displaystyle \underset{k=1}{\overset{\mathrm{}}{}}}u^k{\displaystyle \underset{l=1}{\overset{M1}{}}}{\displaystyle \frac{lh_k(l)}{Mk}}{\displaystyle \underset{k=1}{\overset{\mathrm{}}{}}}f_k{\displaystyle \frac{u^k}{k}}(4M[\psi (M)+\gamma ]{\displaystyle \frac{7(M1)}{2}})],`$
where $`u=e^{Q^2/2MT_0}`$. In order to cancel lattice artifacts in the continuum limit $`M\mathrm{}`$, we find the constraints
$`{\displaystyle \underset{k=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{f_k}{k}}={\displaystyle \frac{\pi ^2}{6}},{\displaystyle \underset{k=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{h_k(l)}{k}}={\displaystyle \frac{\pi ^2}{18}}\left(1{\displaystyle \frac{1}{l}}\right).`$ (5.6)
Then we obtain
$`\mathrm{\Pi }_2{\displaystyle \frac{g^2N_c}{16\pi ^2}}{\displaystyle \frac{Q^2}{T_0}}\left\{\left[8(\mathrm{ln}M+\gamma ){\displaystyle \frac{22}{3}}\right]\mathrm{ln}{\displaystyle \frac{Q^2}{2MT_0}}+{\displaystyle \frac{4}{3}}\right\}.`$ (5.7)
Here $`2MT_0`$ functions as a uv cut-off. With this understanding our result for $`\mathrm{\Pi }_2`$ agrees exactly with the known light-cone gauge result . Notice that the parameters $`f_k,h_k`$ which specify our discretization enter weak coupling physics only through their moments, for example $`_kf_k/k`$. In contrast, the strong coupling limit is sensitive to the values of these parameters at low $`k`$. Thus the two limits give complementary constraints on these parameters.
## 6 Concluding Remarks
Our main aim in developing this discretization formalism is to establish a framework for handling the sum of all the planar diagrams of $`N_c\mathrm{}`$ gauge theories. As yet we don’t have any dramatic results to report. However we have begun studying how the machinery works in simpler situations than full-blown gauge theories. In our paper we worked out the sum of the densest (strong coupling “fishnet”) planar diagrams of $`\mathrm{Tr}\varphi ^3`$ scalar field theory. As expected the result leads to the light-cone quantized bosonic string (with all of its usual pathologies, including the tachyon). The presence of tachyons is not particularly surprising, since the energy density of the theory is unbounded below. We have not made analogous progress on the corresponding diagrams of gauge theories. But we have studied the latter theory in the sectors with $`M=2`$. This is not particularly difficult, since the limitation on $`M`$ reduces the sum of all possible diagrams to a geometric series. Nevertheless, it is interesting, for example, that the $`M=2`$ gluon propagator summed to all orders in perturbation theory displays no additional poles beyond that of the massless gluon itself. In contrast the $`M=2`$ propagator for the fictitious 2-form field shows a bound state pole at sufficiently strong coupling $`g^2N_c/8\pi ^2>18.28`$. This indicates that the 2-form field may be particularly important for the understanding of the strong coupling limit.
Another relatively simple testing ground for our formalism is quantum field theory in low space-time dimensions, the simplest being the ’t Hooft model (QCD in one space and one time dimensions). Rozowsky and I are just finishing up a study of this model. Since this model is well understood even in the continuum limit, we used it mainly as a test of how our model approaches the continuum theory. One interesting feature of our simultaneous discretization of $`P^+`$ and $`x^+`$ is that one can approach the continuum in different directions. For example the approach to continuum at fixed $`T_0m/a`$ is different from the approach studied in conventional DLCQ. The latter keep $`x^+`$ continuous throughout (in our language this means taking $`a0`$ first followed by $`m0`$). We have confirmed that the same continuum physics emerges in both cases.
I would like to conclude this talk with some remarks on longer term prospects and goals. Our renewed efforts to sum planar graphs have been directly stimulated by the ADS/CFT duality proposed in the last couple of years. This duality in turn was recognized to be a higher dimensional realization of ’t Hooft’s concept of holography : the vision that a consistent quantum theory of gravity requires our apparently 3 dimensional spatial world to be 2 dimensional. I have advocated string bits as a way to realize holography in ’t Hooft’s original 2 dimensional sense: the two dimensions being the transverse dimensions of light-cone string. However, the QCD gluons of this talk really live in 3 dimensions in spite of their description on the light-cone: the longitudinal dimension hasn’t disappeared. Rather, it has been disguised as a variable Newtonian mass. (In the ADS/CFT duality this third dimension gets interpreted as a fifth dimension, whence holography is the statement that a 4+1 dimensional effective theory arises from a 3+1 dimensional quantum field theory.) The defining character of string bits is that they have a fixed Newtonian mass, in sharp contrast to the gluons we have been describing. To understand 3+1 gauge theories as part of a string bit theory, a gluon with $`P^+=lm`$ must in reality be a composite of $`l`$ string bits: the gluon vertices would then be effective fission/fusion amplitudes as in nuclear physics, rather than fundamental interactions.
Acknowledgments: Most of the work described in this talk was done in collaboration with Klaus Bering and Joel Rozowsky, whom I thank for their essential contributions and insights.
|
warning/0004/hep-ph0004225.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
35 years ago Hagedorn has written a paper of fundamental importance which nowadays is very often cited in various contexts of statistical mechanics, particle physics, and string theory. In this paper he developed a statistical description of momentum spectra of particles produced in collider experiments. At a fundamental level, of course, the underlying theory for this is quantum chromodynamics. However, for the hadronization cascade phenomenological models have to be used in addition to QCD. The complexity inherent in this process is so immense that usually one has to perform Monte Carlo simulations in order to explain the experimental measurements of cross sections.
Hagedorn’s successful approach was to see this problem from a thermodynamic point of view (of course QCD was not known at the time he wrote the paper). He devoped a theory that is nowadays regularly in use do describe e.g. heavy ion collisions . The basic assumption is that in the scattering region the density of states grows so rapidly that the temperature cannot exceed a certain maximum temperature, the Hagedorn temperature $`T_0`$. This is similar to a first oder phase transition of e.g. boiling water, just that the Hagedorn temperature describes ’boiling’ nuclear matter at about $`T_0200`$ MeV. This state of matter is closely related to what is nowadays called quark gluon plasma. For some theoretical approaches to transverse momentum spectra, see e.g. . As already mentioned, the Hagedorn phase transition is also of fundamental interest in string theories .
While for collider experiments (e.g. $`e^+e^{}`$ annihilation) the Hagedorn theory yields a good description for relatively small center of mass enegies ($`E<10`$ GeV), it fails at large energies. Indeed, Hagedorn’s approach predicts an exponential decay of differential cross sections for large transverse momenta, whereas in experiments one observes non-exponential behaviour for large energies ($`E>10`$ GeV) . For this reason one usually restricts the comparison of the experimental results at higher energies to Monte Carlo simulations, which reasonably well reproduce the experimental data. But apparently, for a deeper understanding from a statistical mechanics point of view there is the need to generalize Hagedorn’s original ideas.
In this paper we show that it is possible to extend Hagedorn’s theory in such a way that it correctly describes the experimental findings at large energies as well. The basic input for this is the recently developed formalism of non-extensive statistical mechanics , a generalization of ordinary statistical mechanics suitable for multifractal and self-similar systems with long-range interactions. The formalism has been introduced by Tsallis 12 years ago and has in the mean time been shown to be very successful to describe e.g. systems exhibiting turbulent behaviour or Hamiltonian systems with long-range interactions . Several other physical applications were described in . Very recently, evidence was provided that the new formalism has applications in particle physics as well. Walton and Rafelski studied a Fokker Planck equation describing charmed quarks in a thermal quark-gluon plasma and showed that Tsallis statistics is relevant. Wilk and Wlodarczyk provided evidence that the distribution of depths of vertices of hadrons originating from cosmic ray cascades, as measured in emulsion chamber experiments, follows Tsallis statistics. Alberico, Lavagno and Quarati successfully analysed heavy ion collisions using Tsallis statistics. Bediaga, Curado and Miranda showed that differential cross sections in $`e^+e^{}`$ annihilation experiments can be very very well fitted using Tsallis statistics. In the present paper we will work out the theory underlying this.
To effectively describe the complex QCD interactions and hadronization cascade by a a thermodynamic model, it seems reasonable to use the Tsallis formalism (which contains ordinary statistical mechanics as a special case), since this formalism is specially designed to include selfsimilar systems and systems with long-range interactions. What happens in a collider experiment is expected to fall into this category, since QCD forces are strong at large distances. Moreover, the self-similarity of the scattering process was already recognized by Hagedorn, who described the various possible particle states as a ’fireball’ and who defined a fireball as follows : A fireball is
\*… a statistical equilibrium of an undetermined number of all kinds of fireballs, each of which in turn is considered to be…
(back to \*)
Clearly, nowadays we would call this a self-similarity assumption.
In this paper we will combine Hagedorn’s and Tsallis’ approach. We will introduce a suitable generalized grand canonical partition function. The predictions following from this generalized thermodynamic model can be evaluated analytically (at least in a certain approximation). We will also take into account the multiplicity and suggest a concrete dependence of the non-extensitivity parameter $`q`$ on the energy $`E`$. The result will be a concrete formula for differential cross sections of transverse momenta $`p_T`$ where no free fitting parameters are left. Our formula turns out to be in excellent agreement with experimental measurements in $`e^+e^{}`$ annihilation experiments. This turns out to be true for the entire range of center of mass energies that have been probed in experiments. Our analytical formulas are also in good agreement with the curves obtained from Monte Carlo simulations, thus suggesting that the complexity inherent in these simulations is effectively reproduced by the simple thermodynamic model considered here.
This paper is organized as follows. In section 2 we shortly review the main results of Hagedorn’s theory. In section 3 we shortly review Tsallis’ theory. In section 4 we combine both. We will write down the generalized grand canonical partition function, including both fermions and bosons. In section 5 we will consider a large $`p_T`$ approximation, which allows for a simple analytical treatment. We will numerically show that this approximation is a good one even for relatively small values of the transverse momentum. We derive the relevant form of the probability density and calculate all moments of transverse momenta. In section 6 we investigate the (weak) dependence of the Hagedorn temperature $`T_0`$ on the non-extensitivity parameter $`q`$. In section 7 we analyse the energy dependence of the parameter $`q`$. In section 8 we take into account the multiplicity and derive the final formula for the differential cross section. Finally, in section 9 we compare our theoretical results with the experimentally measured results of the TASSO and DELPHI collaboration. Our concluding remarks are given in section 10.
## 2 Hagedorn’s theory
Hagedorn’s theory essentially predicts that the differential cross section in a scattering experiment is given by
$$\frac{1}{\sigma }\frac{d\sigma }{dp_T}=cp_T_0^{\mathrm{}}𝑑xe^{\beta \sqrt{x^2+\mu ^2}}$$
(1)
Here $`p_T`$ is the transverse momentum, $`\mu =\sqrt{p_T^2+m^2}`$ is the transverse mass, the integration variable $`x`$ stands for the longitudinal momentum, $`\beta =1/(kT_0)`$ denotes the inverse Hagedorn temperature, and $`c`$ is some constant. The factor $`e^{\beta \sqrt{x^2+\mu ^2}}`$ is immediately recognized as a Boltzmann factor with energy given by the relativistic energy-momentum relation. In the following, we will set the Boltzmann constant $`k`$ equal to 1.
The Hagedorn temperature $`T_0`$ is independent of the center of mass energy $`E`$ of the beam and typically has a a value of about 100-300 MeV. The physical idea underlying Hagedorn’s approach is that $`T_0`$ describes a kind of ’boiling temperature’ of nuclear matter, which cannot be further increased by external energy transfer. Any further increase of energy just produces new states of particles rather than an increase of temperature.
The integral in eq. (1) can be further evaluated to yield
$$\frac{1}{\sigma }\frac{d\sigma }{dp_T}=cp_T\mu K_1(\beta \mu ),$$
(2)
where $`K_1`$ is the modified Hankel function. For $`p_T`$ large compared to both $`T_0`$ and $`m`$ one obtains from the asymptotic behaviour of the Hankel function the approximate formula
$$\frac{1}{\sigma }\frac{d\sigma }{dp_T}p_T^{3/2}e^{\beta p_T}.$$
(3)
One sees that the differential cross section decays exponentially for large values of $`p_T`$. While this exponential decay is indeed observed for collider experiments with relatively small center of mass energies ($`E<10`$ GeV), clear deviations from exponential decay have been observed at higher energies . Here the Hagedorn theory is not valid any more and the dependence of the differential cross section on the transverse momentum is more complicated. Indeed, for large $`p_T`$ polynomial decay is observed. For example, the ZEUS colloboration has fitted their measurements obtained in $`ep`$ collision experiments by an empirical power law of the form $`(1+constp_T)^\alpha `$, with exponent $`\alpha `$ measured as $`\alpha =5.8\pm 0.5`$.
In the following sections we will show that it is straightforward to extend the Hagedorn theory in such a way that it decribes experiments at high energies ($`E>10`$ GeV) as well. The basic tool for this is the recently developed formalism of non-extensive statistical mechanics. It will just lead to asymptotic power laws of the above form.
## 3 Tsallis’ theory
The formalism of non-extensive statistical mechanics is a generalization of the ordinary formalism of statistical mechanics . Wheras ordinary statistical mechanics is derived by extremizing the Boltzmann-Gibbs entropy $`S=_ip_i\mathrm{ln}p_i`$ (subject to constraints), in non-extensive statistical mechanics the more general Tsallis entropies
$$S_q=\frac{1}{q1}\left(1\underset{i}{}p_i^q\right)$$
(4)
are extremized. The $`p_i`$ are probabilities associated with the microstates of a physical system, and $`q`$ is the non-extensivity parameter. The ordinary Boltzmann-Gibbs entropy is obtained in the limit $`q1`$.
The Tsallis entropies are closely related to the Rényi information measures and have similarly nice properties as the Boltzmann-Gibbs entropy has. They are positive, concave, take on their extremum for the uniform distribution, and preserve the Legendre transform structure of thermodynamics. However, they are not additive for independent subsystems (hence the name ’non-extensive’ statistical mechanics). In the mean time a lot of physical applications have been reported for the formalism with $`q1`$. Examples are 3-dimensional fully developed hydrodynamic turbulence , 2-dimensional turbulence in pure electron plasmas , Hamiltonian systems with long-range interactions , granular systems , systems with strange non-chaotic attractors , and peculiar velocities in Sc galaxies .
Given some set of probabilities $`p_i`$ one can proceed to another set of probabilities $`P_i`$ defined by
$$P_i=\frac{p_i^q}{_ip_i^q}.$$
(5)
These distributions are called escort distributions . Extremizing $`S_q`$ under the energy constraint
$$\underset{i}{}P_iϵ_i=U_q,$$
(6)
where the $`ϵ_i`$ are the energy levels of the microstates, the probabilities $`P_i`$ come out of the extremization procedure as
$$P_i=\frac{1}{Z_q}(1+(q1)\beta ϵ_i)^{\frac{q}{q1}},$$
(7)
where
$$Z_q=\underset{i}{}(1+(q1)\beta ϵ_i)^{\frac{q}{q1}}$$
(8)
is the partition function and $`\beta =1/T`$ is a suitable inverse temperature variable (depending on $`U_q`$). In the limit $`q1`$, ordinary statistical mechanics is recovered, and the $`P_i`$ just reduce to the ordinary canonical distributions $`P_i=\frac{1}{Z}e^{\beta ϵ_i}`$. For $`q1`$, on the other hand, they can be regarded as generalized versions of the canonical ensemble, describing probability distributions in a complex non-extensive system at inverse temperature $`\beta `$.
## 4 Combining Hagedorn’s and Tsallis’ theory
### 4.1 States of factorizing probabilities
To generalize Hagedorn’s theory we first have to decide on how to introduce grand canonical partition functions in non-extensive statistical mechanics. The problem is non-trivial, as one immediately recognizes from just considering a system of $`N`$ independent particles. In fact, it must first be defined what one means by independence in the non-extensive approach. The most plausible definition is that for independent particles probabilities should factorize. Let us consider such a distinguished factorized state where each joint probability is given by products of single-particle Tsallis distributions. This means, the joint probability to $`P_{i_1,i_2,\mathrm{},i_N}`$ to observe particle 1 in energy state $`ϵ_{i_1}`$, particle 2 in energy state $`ϵ_{i_2}`$, and so on is given by
$$P_{i_1,i_2,\mathrm{},i_N}=\frac{1}{Z}\underset{j=1}{\overset{N}{}}(1+(q1)\beta ϵ_{i_j})^{\frac{q}{q1}}.$$
(9)
At the same time, we could also describe our system by the Tsallis distribution formed with the total energy (the Hamiltonian $`H(i_1,i_2,\mathrm{},i_N)`$ of the system)
$$P_{i_1,i_2,\mathrm{},i_N}=\frac{1}{Z}(1+(q1)\beta H(i_1,\mathrm{},i_N))^{\frac{q}{q1}}.$$
(10)
Equating (9) and (10) we see that the total energy of the system is given by
$$1+(q1)\beta H=\underset{j=1}{\overset{N}{}}(1+(q1)\beta ϵ_{i_j}),$$
(11)
which we may write as
$$H=\underset{j}{}ϵ_{i_j}+(q1)\beta \underset{j,k}{}ϵ_{i_j}ϵ_{i_k}+(q1)^2\beta ^2\underset{j,k,l}{}ϵ_{i_j}ϵ_{i_k}ϵ_{i_l}+\mathrm{}$$
(12)
(all summation indices are pairwise different). This means that for the unique particle state where probabilities factorize the total energy of the system is not the sum of the single particle energies, provided $`q1`$. In other words, if seen from the energy point of view, the particles are formally interacting with a coupling constant $`(q1)\beta `$ although the probabilities factorize. This fact is not too surprising— the formalism of non-extensive statistical mechanics is of course designed to describe systems with (long-range) interactions, and also the entropy is non-additive.
We can also invert the above statement. If we consider a Hamiltonian that is just the sum of the single particle energies, then the probabilities do not factorize provided $`q1`$. This is a well-known statement in non-extensive statistical mechanics.
In the following, we will generalize Hagedorn’s theory using the unique states of factorizing probabilities—thus implicitly introducing interactions between the particles from the energy point of view.
### 4.2 Statistical mechanics of the fireball
We will consider particles of different types and label the particle types by an index $`j`$. Each particle can be in a certain momentum state labelled by the index $`i`$. The energy associated with this state is
$$ϵ_{ij}=\sqrt{𝐩_i^2+m_j^2}$$
(13)
(the relativistic energy-momentum relation). Let us define a non-extensive Boltzmann factor by
$$x_{ij}=(1+(q1)\beta ϵ_{ij})^{\frac{q}{q1}}.$$
(14)
It approaches the ordinary Boltzmann factor $`e^{\beta ϵ_{ij}}`$ for $`q1`$. We now very much follow Hagedorn’s original ideas, replacing the ordinary Boltzmann factor by the generalized one. The generalized grand canonical partition function is introduced as
$$Z=\underset{(\nu )}{}\underset{ij}{}x_{ij}^{\nu _{ij}}$$
(15)
Here $`\nu _{ij}`$ denotes the number of particles of type $`j`$ in momentum state $`i`$. The sum $`_{(\nu )}`$ stands for a summation over all possible particle numbers. For bosons one has $`\nu _{ij}=0,1,2,\mathrm{},\mathrm{}`$, whereas for fermions one has $`\nu _{ij}=0,1`$. It follows that for bosons
$$\underset{\nu _{ij}}{}x_{ij}^{\nu _{ij}}=\frac{1}{1x_{ij}}(bosons)$$
(16)
whereas for fermions
$$\underset{\nu _{ij}}{}x_{ij}^{\nu _{ij}}=1+x_{ij}(fermions)$$
(17)
Hence the partition function can be written as
$$Z=\underset{ij}{}\frac{1}{1x_{ij}}\underset{i^{}j^{}}{}(1+x_{i^{}j^{}}),$$
(18)
where the prime labels fermionic particles. For the logarithm of the partition function we obtain
$$\mathrm{log}Z=\underset{ij}{}\mathrm{log}(1x_{ij})+\underset{i^{}j^{}}{}\mathrm{log}(1+x_{i^{}j^{}}).$$
(19)
One may actually proceed to continuous variables by replacing
$$\underset{i}{}[\mathrm{}]_0^{\mathrm{}}\frac{V_04\pi p^2}{h^3}[\mathrm{}]𝑑p=\frac{V_0}{2\pi ^2}_0^{\mathrm{}}p^2[\mathrm{}]𝑑p(\mathrm{}=1)$$
(20)
($`V_0`$: volume of the interaction region) and
$$\underset{j}{}[\mathrm{}]_0^{\mathrm{}}\rho (m)[\mathrm{}]𝑑m,$$
(21)
where $`\rho (m)`$ is the mass spectrum.
Let us now calculate the average occupation number of a particle of species $`j`$ in the momentum state $`i`$. We obtain
$`\overline{\nu _{ij}}=x_{ij}{\displaystyle \frac{}{x_{ij}}}\mathrm{log}Z`$ $`=`$ $`{\displaystyle \frac{x_{ij}}{1\pm x_{ij}}}`$ (22)
$`=`$ $`{\displaystyle \frac{1}{(1+(q1)\beta ϵ_{ij})^{\frac{q}{q1}}\pm 1}}`$
where the $``$ sign is for bosons and the $`+`$ sign for fermions.
In order to single out a particular particle of mass $`m_0`$, one can formally work with the mass spectrum $`\rho (m)=\delta (mm_0)`$. To obain the probability to observe a particle of mass $`m_0`$ in a certain momentum state, we have to multiply the average occupation number with the available volume in momentum space. An infinitesimal volume in momentum space can be written as
$$dp_xdp_ydp_z=dp_Lp_T\mathrm{sin}\theta dp_Td\theta $$
(23)
where $`p_T=\sqrt{p_y^2+p_z^2}`$ is the transverse monentum and $`p_x=p_L`$ is the longitutinal one. By integrating over all $`\theta `$ and $`p_L`$ one finally arrives at a probability density $`w(p_T)`$ of transverse momenta given by
$$w(p_T)=constp_T_0^{\mathrm{}}𝑑p_L\frac{1}{(1+(q1)\beta \sqrt{p_T^2+p_L^2+m_0^2})^{\frac{q}{q1}}\pm 1}.$$
(24)
Since the Hagedorn temperature is rather small (of the order of the $`\pi `$ mass), under normal circumstances one has $`\beta \sqrt{p_T^2+p_L^2+m_0^2}>>1`$, and hence the $`\pm 1`$ can be neglected if $`q`$ is close to 1. One thus obtains for both fermions and bosons the statistics
$$w(p_T)constp_T_0^{\mathrm{}}𝑑p_L(1+(q1)\beta \sqrt{p_T^2+p_L^2+m_0^2})^{\frac{q}{q1}}$$
(25)
which, if our theory is correct, should determine the $`p_T`$ dependence of experimentally measured particle spectra. The differential cross section $`\sigma ^1d\sigma /dp_T`$ is expected to be proportional to $`w(p_T)`$.
## 5 Large $`p_T`$ approximation
### 5.1 The differential cross section
To further proceed with analytic calculations, one may actually perform one further approximation step. This is the generalization of the step of going from eq. (1) to eq. (3) in Hagedorn’s original theory, which is a good approximation for large $`p_T`$. Since for our applications $`q`$ is close to 1, one expects that a similar step is possible in the more general non-extensive theory.
Let us write the formula for the differential cross section in the form
$$\frac{1}{\sigma }\frac{d\sigma }{dp_T}=cu_0^{\mathrm{}}𝑑x\left(1+(q1)\sqrt{x^2+u^2+m_\beta ^2}\right)^{\frac{q}{q1}}.$$
(26)
Here $`x=p_L/T_0`$, $`u=p_T/T_0`$ and $`m_\beta :=m_0/T_0`$ are the longitudinal momentum, transverse momentum and mass in units of the Hagedorn temperature $`T_0`$, respectively. $`c`$ is a suitable constant. Let us look at this formula for large values of $`p_T`$. If $`u`$ is very large, we can approximate $`\sqrt{x^2+u^2+m_\beta ^2}=u\sqrt{1+(x^2+m_\beta ^2)/u^2}u+(x^2+m_\beta ^2)/(2u)`$. Of course, for this to be true the integration variable $`x`$ should not be too large, but for large $`x`$ the integrand is small anyway and yields only a very small contribution to the cross section. Moreover, since $`u`$ is large we may also neglect the mass term $`m_\beta ^2`$, arriving at the aprroximation
$$\frac{1}{\sigma }\frac{d\sigma }{dp_T}cu_0^{\mathrm{}}𝑑x\left(1+(q1)\left(u+\frac{x^2}{2u}\right)\right)^{\frac{q}{q1}}$$
(27)
Although this may look like a rather crude approximation, in practice it is quite a good one. This is illustrated in Fig. 1, which shows the right hand sides of eq. (26) and eq. (27) for $`q=1,1.1,1.2`$. The lines corresponding to the exact expression (26) and the approximation (27) can hardly be distinguished. The range of $`u`$ and $`q`$ is similar to what we will use later for the comparison with experimental measurements. Since the Hagedorn temperature $`T_0`$ is about 120 MeV, and since very often pions are produced, we have chosen in formula (26) $`m_\beta ^2=(m_\pi /T_0)^21.3`$. The errors made in eq. (27) by neglecting $`m_\beta `$ and by approximating the square root work in opposite directions and almost cancel each other for $`q1.1`$.
Within this approximation we can now easily proceed by analytical means. We may write eq. (27) as
$$\frac{1}{\sigma }\frac{d\sigma }{dp_T}=cu\left(1+(q1)u\right)^{\frac{q}{q1}}_0^{\mathrm{}}𝑑x\left(1+\frac{q1}{2u(1+(q1)u)}x^2\right)^{\frac{q}{q1}}$$
(28)
Substituting
$$t:=\sqrt{\frac{q1}{2u(1+(q1)u)}}x$$
(29)
and using the general formula
$$_0^{\mathrm{}}\frac{t^{2x1}}{(1+t^2)^{x+y}}𝑑t=\frac{1}{2}B(x,y)(Rex>0,Rey>0),$$
(30)
where
$$B(x,y)=\frac{\mathrm{\Gamma }(x)\mathrm{\Gamma }(y)}{\mathrm{\Gamma }(x+y)}$$
(31)
denotes the beta-function, one arrives at
$$\frac{1}{\sigma }\frac{d\sigma }{dp_T}=c(2(q1))^{1/2}B(\frac{1}{2},\frac{q}{q1}\frac{1}{2})u^{3/2}\left(1+(q1)u\right)^{\frac{q}{q1}+\frac{1}{2}}.$$
(32)
This formula, with suitably determined $`c`$, $`q`$, $`T_0`$, will turn out to be in very good agreement with experimentally measured cross sections.
### 5.2 Normalization
Let us quite generally consider a probability density with the above $`u`$-dependence
$$p(u)=\frac{1}{Z_q}u^{3/2}\left(1+(q1)u\right)^{\frac{q}{q1}+\frac{1}{2}}$$
(33)
Normalization yields
$$Z_q=_0^{\mathrm{}}u^{3/2}\left(1+(q1)u\right)^{\frac{q}{q1}+\frac{1}{2}}𝑑u.$$
(34)
Substituting $`t^2:=(q1)u`$ the integral can be brought into the form (30), and one obtains
$$Z_q=(q1)^{5/2}B(\frac{5}{2},\frac{q}{q1}3)$$
(35)
Note that the beta function further simplifies if $`\frac{q}{q1}3`$ is an integer. For $`n𝐍`$ one generally has
$$B(x,n)=\frac{(n1)!}{x(x+1)(x+2)\mathrm{}(x+n1)},$$
(36)
in our case $`x=\frac{5}{2}`$.
### 5.3 Moments
Let us now evaluate the moments of $`u`$ defined by
$$u^m=_0^{\mathrm{}}u^mp(u)𝑑u=\frac{1}{Z_q}_0^{\mathrm{}}u^{\frac{3}{2}+m}\left(1+(q1)u\right)^{\frac{q}{q1}+\frac{1}{2}}𝑑u.$$
(37)
Again substituting $`t^2=(q1)u`$ the integral can be evaluated to give
$`u^m`$ $`=`$ $`{\displaystyle \frac{1}{(q1)^m}}{\displaystyle \frac{B(\frac{5}{2}+m,\frac{q}{q1}3m)}{B(\frac{5}{2},\frac{q}{q1}3)}}`$ (38)
$`=`$ $`{\displaystyle \frac{1}{(q1)^m}}{\displaystyle \frac{\mathrm{\Gamma }\left(\frac{5}{2}+m\right)\mathrm{\Gamma }\left(\frac{q}{q1}3m\right)}{\mathrm{\Gamma }\left(\frac{5}{2}\right)\mathrm{\Gamma }\left(\frac{q}{q1}3\right)}}`$
Generally the Gamma function satisfies
$$\mathrm{\Gamma }(x+1)=x\mathrm{\Gamma }(x)$$
(39)
which one may iterate to obtain
$$\mathrm{\Gamma }(x+m)=\mathrm{\Gamma }(x)\underset{j=0}{\overset{m1}{}}(x+j).$$
(40)
Using this in eq. (38) one finally arrives at
$$u^m=\frac{1}{2^m}\underset{j=0}{\overset{m1}{}}\frac{5+2j}{4+j(3+j)q}.$$
(41)
In particular, one obtains for the average of $`u`$
$$u=\frac{1}{2}\frac{5}{43q}$$
(42)
and for the second and third moment
$`u^2`$ $`=`$ $`{\displaystyle \frac{1}{4}}{\displaystyle \frac{5}{43q}}{\displaystyle \frac{7}{54q}}`$ (43)
$`u^3`$ $`=`$ $`{\displaystyle \frac{1}{8}}{\displaystyle \frac{5}{43q}}{\displaystyle \frac{7}{54q}}{\displaystyle \frac{9}{65q}}.`$ (44)
The variance is given by
$$\sigma ^2:=u^2u^2=\frac{5}{4}\frac{3q}{(43q)^2(54q)}.$$
(45)
Note that the moments obey the simple recurrence relation
$$u^{m+1}=u^m\frac{5+2m}{4+m(3+m)q}.$$
(46)
## 6 $`q`$-dependence of the Hagedorn temperature
A fundamental property of Hagedorn’s theory is the fact that the Hagedorn temperature $`T_0`$ is independent of the center of mass energy $`E`$. Now, in the generalized theory we have a new parameter $`q`$ and it is a priori not clear if and how $`T_0`$ depends on $`q`$.
However, looking at eq. (12) we recognize that if $`q`$ increases from 1 to slightly larger values the (formal) interaction energy of our system increases. This increase in interaction energy must come from somewhere and is expected to be taken from the (finite volume) heat bath of the fireball. Thus the effective temperature of the bath is expected to slightly decrease with increasing $`q`$.
To get a rough estimate of this effect, let us work within the approximation scheme of the previous section. Differentiating eq. (33) with respect to $`u`$ one immediately sees that the distribition $`p(u)`$ has a maximum at
$$u^{}=\frac{p_T^{}}{T_0}=\frac{3}{3q}.$$
(47)
On the other hand, the experimentally measured cross sections always appear to have their maximum at roughly the same value of the transverse momentum $`p_T`$, namely at
$$p_T^{}180MeV,$$
(48)
independent of the beam energy $`E`$. This implies that in the generalized thermodynamic approach the effective Hagedorn temperature $`T_0`$ will become (slightly) $`q`$-dependent.
$$T_0=\left(1\frac{q}{3}\right)p_T^{}$$
(49)
The variation of $`T_0`$ with energy $`E`$ is actually very weak. In the next section we will consider two extreme cases, namely $`q1`$ for $`E0`$ and $`q\frac{11}{9}`$ for $`E\mathrm{}`$. The first case corresponds to $`T_0=120`$ MeV, the second case to $`T_0=107`$ MeV. This is only a very small variation.
## 7 Energy dependence of $`q`$
We still have to decide how the non-extensitivity parameter $`q`$ depends on the center of mass energy $`E`$ of the beam. We will present some theoretical arguments, which are well supported by the experimental data.
### 7.1 Plausible value for $`q_{max}`$
Let us define $`\alpha `$ to be the power of the term $`(1+(q1)u)^1`$ in eq. (33)
$$\alpha =\frac{q}{q1}\frac{1}{2}.$$
(50)
Clearly, for small energies $`E`$ Hagedorn’s theory ($`q=1`$) is valid:
$$E0\alpha \mathrm{}q1$$
(51)
For increasing $`E`$, $`\alpha `$ should decrease. For example, the ZEUS collaboration measures $`\alpha =5.8\pm 0.5`$ at medium energies . However, $`\alpha `$ cannot become arbitrarily small, because we must have a finite variance of $`u`$ (otherwise statistical mechanics does not make sense). For large $`u`$, we can certainly use the large $`p_T`$ approximation of section 5. Asymptotically the density $`p(u)`$ decays as
$$p(u)u^{\alpha +3/2}$$
(52)
Thus $`u^2p(u)`$ decays as $`u^{\alpha +7/2}`$ and hence $`u^2=𝑑uu^2p(u)`$ only exists if
$$\alpha +\frac{7}{2}<1\alpha >\frac{9}{2}q<\frac{5}{4}$$
(53)
(for $`\alpha =9/2`$ there is a logarithmic divergence). This is a strict upper bound on $`q`$.
If, in addition, we postulate that the distinguished limit theory corresponding to the largest possible value of $`q`$ should be analytic in $`\sqrt{u}`$, then only integer and half-integer values of $`\alpha `$ are allowed. This leads to
$$\alpha _{min}=5q_{max}=\frac{11}{9}=1.222$$
(54)
as the smallest possible value of $`\alpha `$, respectively largest value of $`q`$.
### 7.2 A smooth dependence on $`q`$
It is reasonable to assume (and supported by the experimental results) that the smallest possible value of the exponent $`\alpha `$ is taken on for the largest possible energy $`E`$. Hence we have the two limit cases
$`E0`$ $``$ $`\alpha \mathrm{}`$ (55)
$`E\mathrm{}`$ $``$ $`\alpha \alpha _{min}=5.`$ (56)
A smooth monotonously decreasing interpolation between these limit cases is given by the formula
$$\alpha (E)=\frac{5}{1e^{E/E_0}},$$
(57)
where $`E_0`$ is some suitable energy scale. It turns out that the experimental data are perfectly fitted if we choose
$$E_0=\frac{1}{2}m_Z=45.6GeV,$$
(58)
$`m_Z`$ being the mass of the $`Z^0`$ boson.
Whether this coincidence of the relaxation parameter $`E_0`$ with half of the $`Z^0`$ mass is a random effect or whether there is physical meaning behind this is still an open question. What is clear is the following. At center of mass energies of the order 90 GeV often a massive elektroweak gauge boson $`Z^0`$ is produced. If a $`Z^0`$ at rest decays into a quark pair $`q\overline{q}`$, then each quark has the initial energy $`\frac{1}{2}m_Z`$, which is input to the hadronization cascade. It is obvious that at this energy scale there is a rapid change in the behaviour of the fireball due to the occurence of the electroweak bosons. The same mass scale now seems to mark the crossover scale between ordinary Hagedorn thermostatistics, valid for $`E<<\frac{1}{2}m_Z`$, and generalized thermostatistics with $`qq_{max}`$, valid for $`E>>\frac{1}{2}m_Z`$.
Eq. (57) can equivalently be written as
$$q(E)=\frac{11e^{E/E_0}}{9+e^{E/E_0}}.$$
(59)
## 8 Energy dependence of the multiplicity
The differential cross section $`\sigma ^1d\sigma /dp_T`$ measured in experiments is not a normalized probability density. This is due to the fact that it is dependent on the multiplicity $`M`$ (the average number of produced charged particles). Also, it has dimension $`GeV^1`$, whereas the probability density $`p(u)`$ is dimensionless. All this, however, is just a question of normalization.
To proceed from $`p(u)`$ to $`\sigma ^1d\sigma /dp_T`$, we should multiply $`p(u)`$ with the multiplicity $`M`$. Also, in order to give the correct dimension of a cross section, we should multiply with $`T_0^1`$. Thus we arrive at the formula
$$\frac{1}{\sigma }\frac{d\sigma }{dp_T}=\frac{1}{T_0}Mp(u).$$
(60)
The multiplicity $`M`$ as a function of the beam energy $`E`$ has been independently measured in many experiments. A good fit of the experimental data in the relevant energy region is the formula
$$M=\left(\frac{E}{T_0^{q=1}}\right)^{5/11}T_0^{q=1}=120MeV$$
(61)
(see Fig. 2, data as collected in ). It is remarkable that the constant in front of this power law is 1 if $`E`$ is measured in units of the Hagedorn Temperature $`T_0^{q=1}`$. Apparently, the Hagedorn temperature is a very appropriate portion of energy for our statistical approach. Essentially, the scaling law (61) says that only a certain fraction of the logarithm of the energy $`E`$ (described by the scaling exponent $`5/110.45`$) is used to increase the number of charged particles.
## 9 Comparison with experimentally measured cross sections
We are now in a position to directly compare with experimental measurements of cross sections. All parameters of the generalized Hagedorn theory (the Hagedorn temperature $`T_0(q)`$, the non-extensitivity parameter $`q(E)`$ and the normalization constant of the cross section) have been discussed in the previous sections and concrete equations have been derived.
Formula (60) with $`p(u)`$ given by eq.(33), $`q(E)`$ given by eq. (59), $`T_0(q)`$ given by eq.(49) and multiplicity $`M(E)`$ given by eq. (61) turns out to very well reproduce the experimental results of cross sections for all energies $`E`$. This is illustrated in Fig. 3, which shows experimental cross sections measured by the TASSO and DELPHI collaborations (, data as collected in ) as well as our theoretical prediction. For all curves we have chosen the same universal parameters $`p_T^{}=180`$ MeV and $`E_0=\frac{1}{2}m_Z`$, so we do not use any energy-dependent fitting parameters.
The agreement is remarkably good. Hence the statistical approach presented in this paper qualifies as a simple thermodynamic model that explains the experimental data quite well. In particular, for the first time analytical formulas are obtained that correctly describe the measured cross sections. The quality of agreement is at least as good as that of Monte Carlo simulations, in spite of the fact that Monte Carlo simulations usually use a large number of free parameters. For small $`p_T`$, the agreement seems even to be slightly better than that of Monte Carlo simulations (see e.g. for typical Monte Carlo results).
Our approach yields $`q`$-values of similar order of magnitude as the ones used in the fits in . In our theoretical approach all parameters are now given by concrete formulas. Some of these formulas, most importantly eq. (59), have been derived empirically and should thus be further checked and possibly refined by further experimental measurements.
## 10 Conclusion
In this paper we have developed a thermodynamic model describing the statistics of transverse momenta of particles produced in high-energy collisions. Our approach is based on a generalization of Hagedorn’s theory using Tsallis’ formalism of non-extensive statistical mechanics. Hagedorn’s original theory is recovered for small center of mass energies, whereas for larger energies deviations from ordinary Boltzmann-Gibbs statistics become relevant. The crossover energy scale between the two regimes $`E0`$, $`q1`$ and $`E\mathrm{}`$, $`q\frac{11}{9}`$ is given by half of the $`Z^0`$ mass.
At large energies the generalized thermodynamic theory implies that cross sections decay with a power law. This power law is indeed observed in experiments. Our theory allows us to analytically evaluate moments of arbitrary order (within a certain approximation). We obtain formulas for differential cross sections that are in very good agreement with experimentally measured cross sections in annihilation experiments.
We suggest that future analysis of the experimental data should concentrate on precision measurements of the energy dependence of the parameter $`q`$. For this the mass spectrum of produced particles should carefully be taken into account. Generally, it appears that high-energy collider experiments do not only yield valuable information on particle physics, but they may also be regarded as ideal test grounds to verify new ideas from statistical mechanics.
### Acknowledgement
This research was supported in part by the National Science Foundation under Grant No. PHY94-07194. The author gratefully acknowledges support by a Royal Society Leverhulme Trust Senior Research Fellowship.
### Figure captions
Fig. 1 Comparison between the exact formula (26) with $`m_\beta ^2=\left(m_\pi /T_0\right)^2=1.3`$ and the approximation (27) for various values of $`q`$. The constant $`c`$ was set equal to 1 for all $`q`$.
Fig. 2 Dependence of the multiplicity $`M`$ on the center of mass energy $`E`$. The figure shows the experimental data and a straight line that corresponds to the scaling law (61).
Fig. 3 Differential cross section as a function of the transverse momentum $`p_T`$ for various center of mass energies $`E`$. The data correspond to measurements of the TASSO ($`E44`$ GeV) and DELPHI ($`E91`$ GeV) collaboration. The solid lines are given by the analytic formula (60).
|
warning/0004/gr-qc0004009.html
|
ar5iv
|
text
|
# On the equations of motion of point-particle binaries at the third post-Newtonian order
## Abstract
We investigate the dynamics of two point-like particles through the third post-Newtonian (3PN) approximation of general relativity. The infinite self-field of each point-mass is regularized by means of Hadamard’s concept of “partie finie”. Distributional forms associated with the regularization are used systematically in the computation. We determine the stress-energy tensor of point-like particles compatible with the previous regularization. The Einstein field equations in harmonic coordinates are iterated to the 3PN order. The 3PN equations of motion are Lorentz-invariant and admit a conserved energy (neglecting the 2.5PN radiation reaction). They depend on an undetermined coefficient, in agreement with an earlier result of Jaranowski and Schäfer. This suggests an incompleteness of the formalism (in this stage of development) at the 3PN order. In this paper we present the equations of motion in the center-of-mass frame and in the case of circular orbits.
A cardinal problem in Gravitational Physics is that of the dynamics of binary systems of point particles. In general relativity, this problem is tackled by means of the post-Newtonian approximation, or formal expansion when the speed of light $`c`$ goes to infinity. By definition, the $`n`$PN approximation refers to the terms in the equations of motion that are smaller than the Newtonian force by a factor of order $`1/c^{2n}`$. For the motion of two non-spinning point particles, the 1PN approximation was obtained first by Lorentz and Droste . Subsequently, Einstein, Infeld and Hoffmann re-derived the 1PN order using their famous “surface-integral” method. In the eighties, Damour and Deruelle , starting from a “post-Minkowskian” iteration scheme developed by Bel et al , were able to compute the equations of motion up to the 2.5PN order, at which the gravitational-radiation reaction effects first take place. The motivation was to firmly establishing the rate at which the orbit of the binary pulsar PSR 1913+16 decays because of gravitational-radiation emission. The 2.5PN approximation was then obtained by Schäfer using an “ADM Hamiltonian” approach initiated by Ohta et al . Furthermore, Kopeikin et al derived the same result within their “extended-body” method (without any need of a regularization). More recently, the 2.5PN equations of motion as well as 2.5PN gravitational field were derived by Blanchet, Faye and Ponsot applying a direct “post-Newtonian” iteration of the field equations. Finally, Jaranowski and Schäfer investigated within the Hamiltonian approach the 3PN order and found some ambiguities linked to the regularization of the self-field of point masses. As for them, the 3.5PN terms in the equations of motion are well-known ; they are associated with higher-order radiation reaction effects.
The motivation for working out the 3PN equations of motion is not the timing of the binary pulsar anymore, but the detection of gravitational radiation by future experiments such as LIGO and VIRGO. Indeed, the 3PN equations are needed in particular to construct accurate 3.5PN templates for detecting and analyzing the waves generated by inspiralling compact binaries . Currently, we know the complete templates up to the 2.5PN order , plus the contribution therein of non-linear effects to the 3.5PN order , plus all the terms in the vanishing mass-ratio limit to the 5.5PN order . In this Letter, we outline our derivation of the 3PN equations of motion, based on the direct post-Newtonian approach of , and we present the result in the case, appropriate to inspiralling compact binaries, of circular orbits. We confirm by means of a well-defined regularization à la Hadamard the finding of Jaranowski and Schäfer that there remains at the 3PN order an undetermined coefficient appearing in front of a quartically non-linear term (proportional to $`G^4`$).
Consider the class $``$ of functions $`F(𝐱)`$ that are smooth on $`^3`$ except at two isolated points $`𝐲_1`$ and $`𝐲_2`$, around which they admit a power-like singular expansion of the form
$$n,F(𝐱)=\underset{a_0an}{}r_1^a\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{a}{}^{}(𝐧_1)+o(r_1^n)\text{when }r_10,$$
(1)
where $`r_1=|𝐱𝐲_1|`$, $`𝐧_1=(𝐱𝐲_1)/r_1`$, and where the powers $`a`$ are supposed to be real, to range in discrete steps: $`a(a_i)_i`$, and to be bounded from below: $`a_0a`$. The coefficients $`{}_{1}{}^{}f_{a}^{}`$ of the various powers of $`r_1`$ in this expansion are smooth functions of the unit vector $`𝐧_1`$. We refer to the coefficients $`{}_{1}{}^{}f_{a}^{}`$ with $`a<0`$ as the singular coefficients of $`F`$ around 1; their number is always finite. Moreover, we have the same type of expansion around the other point (when $`r_20`$). The Hadamard “partie finie” of $`F`$ at the location of the singular point 1 is equal to the angular average of the zeroth-order coefficient in (1), i.e.
$$(F)_1=\frac{d\mathrm{\Omega }_1}{4\pi }\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{0}{}^{}(𝐧_1),$$
(2)
with $`d\mathrm{\Omega }_1=d\mathrm{\Omega }(𝐧_1)`$ the usual solid angle element. The partie finie is “non-distributive” in the sense that $`(FG)_1(F)_1(G)_1`$ in general. Besides (2), we define also the partie finie ($`\mathrm{Pf}`$) of the divergent integral $`d^3𝐱F`$, assuming that $`F`$ decreases sufficiently rapidly when $`|𝐱|+\mathrm{}`$ so that the divergencies come only from the singular points 1 and 2. With full generality ,
$`\mathrm{Pf}_{s_1,s_2}{\displaystyle d^3𝐱F}`$ $`=`$ $`\underset{s0}{lim}\{{\displaystyle _{^3_1(s)_2(s)}}d^3𝐱F`$ (4)
$`+4\pi {\displaystyle \underset{a+3<0}{}}{\displaystyle \frac{s^{a+3}}{a+3}}\left({\displaystyle \frac{F}{r_1^a}}\right)_1+4\pi \mathrm{ln}\left({\displaystyle \frac{s}{s_1}}\right)\left(r_1^3F\right)_1+12\}.`$
The first term is the finite integral over $`^3`$ deprived from the two spherical balls $`_1(s)`$ and $`_2(s)`$ with radius $`s`$ and centred on the two singularities. The extra terms are such that they exactly cancel out the divergent part of the integral when $`s0`$ (the notation $`12`$ indicates the same extra terms but referring to the other singularity point). The logarithmic terms depend on two arbitrary positive constants $`s_1`$ and $`s_2`$ that come from the arbitrariness in the choice of unit length for measuring $`s`$; hence, the partie finie depends on both $`s_1`$ and $`s_2`$ (as indicated by the notation $`\mathrm{Pf}_{s_1,s_2}`$). Applying (4) to the case of a gradient $`_iF`$, we find
$$\mathrm{Pf}d^3𝐱_iF=4\pi (n_1^ir_1^2F)_1+12.$$
(5)
In words, the integral of a gradient is equal to the sum of the surface integrals surrounding the two singularities, in the limit where the surface areas shrink to zero and following the regularization (2). Thus, the integral of a gradient is not zero in general, which shows that the “ordinary” derivative $`_iF`$ is not adequate for applying to point-particles a formalism initially valid for continuous sources, since in the latter case the integral of a gradient does never contribute. To define a “better” notion of derivative, we must construct the distributional forms associated with the functions in the class $``$.
For any $`F`$, we consider the “pseudo-function” $`\mathrm{Pf}F`$ defined as the linear form on $``$ such that $`G`$, $`<\mathrm{Pf}F,G>=\mathrm{Pf}d^3𝐱FG`$, the duality bracket denoting here the result of the action of $`\mathrm{Pf}F`$ on the function $`G`$. The product of pseudo-functions is defined to be the ordinary pointwise product that we use in Physics, i.e. $`\mathrm{Pf}F.\mathrm{Pf}G=\mathrm{Pf}(FG)`$. With the help of the Riesz delta-function $`\delta _\epsilon (𝐱𝐲_1)=\frac{\epsilon (1\epsilon )}{4\pi }r_1^{\epsilon 3}`$, which belongs to the class $``$, we construct the pseudo-function $`\mathrm{Pf}\delta _1`$ (in the limit $`\epsilon 0`$); by definition: $`F`$, $`<\mathrm{Pf}\delta _1,F>=(F)_1`$. Clearly $`\mathrm{Pf}\delta _1`$ generalizes the standard Dirac distribution $`\delta _1\delta (𝐱𝐲_1)`$ to the case of the Hadamard regularization of the functions in $``$. Furthermore, consistently with the product of pseudo-functions, we construct the object $`\mathrm{Pf}(F\delta _1)`$ which is such that $`G`$, $`<\mathrm{Pf}(F\delta _1),G>=(FG)_1`$. A trivial consequence of the non-distributivity of the Hadamard partie finie is that $`\mathrm{Pf}(F\delta _1)(F)_1\mathrm{Pf}\delta _1`$ in general cases. The derivative of the pseudo-function $`\mathrm{Pf}F`$ is then obtained from the requirements that (i) the “rule of integration by parts” is satisfied, i.e. $`F,G`$, $`<_i(\mathrm{Pf}F),G>=<_i(\mathrm{Pf}G),F>`$, (ii) the derivative reduces to the “ordinary” one in the case where all the singular coefficients of $`F`$ vanish. These requirements imply in particular that $`<_i(\mathrm{Pf}F),1>=0`$, i.e. the integral of a gradient is zero. A derivative operator satisfying (i) and (ii) is given by
$$_i(\mathrm{Pf}F)=\mathrm{Pf}(_iF+4\pi n_1^i[\frac{1}{2}r_1\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{1}{}^{}+\underset{k0}{}\frac{1}{r_1^k}\stackrel{}{f}\genfrac{}{}{0pt}{}{}{1}{}_{2k}{}^{}]\delta _1+12)$$
(6)
(assuming for simplicity that the $`{}_{1}{}^{}f_{a}^{}`$’s have $`a`$). This derivative reduces to the standard distributional derivative of Schwartz when applied on smooth functions with compact support. We refer to for the construction of the most general derivative operator satisfying (i), (ii) and, in addition, (iii) the rule of commutation of derivatives \[not obeyed by (6)\]. One can show however that it does not satisfy in general the Leibniz rule for the derivative of a product. The derivative (6) is sufficient in the derivation of the results below. See for details about the Hadamard regularization and the associated pseudo-functions.
In the post-Newtonian application we are led to consider the partie finie, in the sense of (4), of the Poisson integral of $`F`$, i.e. $`\mathrm{Pf}d^3𝐱F(𝐱)/|𝐱𝐱^{}|`$; more specifically, we are interested in the regularized value, in the sense of (2), of the latter Poisson integral at the location of the singular point 1, i.e. when $`r_1^{}=|𝐱^{}𝐲_1|0`$. We obtain
$`\left(\mathrm{Pf}_{s_1,s_2}{\displaystyle \frac{d^3𝐱}{|𝐱𝐱^{}|}F}\right)_1`$ $`=`$ $`\mathrm{Pf}_{s_1,s_2}{\displaystyle \frac{d^3𝐱}{r_1}F}4\pi \left[\mathrm{ln}\left({\displaystyle \frac{r_1^{}}{s_1}}\right)1\right]\left(r_1^2F\right)_1`$ (8)
$`=`$ $`4\pi \mathrm{ln}\left({\displaystyle \frac{r_{12}}{r_1^{}}}\right)\left(r_1^2F\right)_1+4\pi \mathrm{ln}\left({\displaystyle \frac{r_{12}}{s_2}}\right)\left({\displaystyle \frac{r_2^3}{r_1}}F\right)_2+\mathrm{}`$ (9)
(with $`r_{12}=|𝐲_1𝐲_2|`$). The first term in (8) represents simply what we get by replacing formally $`𝐱^{}`$ by $`𝐲_1`$ inside the integrand of the Poisson integral. The second term is due to the presence of some logarithms $`\mathrm{ln}r_1^{}`$ in the expansion of the integral. (An adaptation of the previous formalism, detailed in , is needed to take these logarithms into account, as well as the presence of the integrable singularity $`𝐱^{}`$.) As, at last, the $`\mathrm{ln}r_1^{}`$ can be gauged away, we regard it as a constant, taking some finite value (even though $`r_1^{}0`$). We check, on the other hand, that the true constant $`s_1`$ cancels out between the two terms of (8), so that the result depends only on $`\mathrm{ln}r_1^{}`$ and $`\mathrm{ln}s_2`$. The complete dependence of the partie finie on these constants is shown in (9), with the convention that the dots indicate the terms independent of the constants.
The Einstein field equations relaxed by the condition of harmonic coordinates \[i.e. $`_\nu h^{\mu \nu }=0`$ with $`h^{\mu \nu }=\sqrt{g}g^{\mu \nu }\eta ^{\mu \nu }`$; $`g=\mathrm{det}g_{\mu \nu }`$; $`\eta ^{\mu \nu }=`$diag$`(1,1,1,1)`$\] read as
$$\mathrm{}h^{\mu \nu }=\frac{16\pi G}{c^4}(g)T^{\mu \nu }+\mathrm{\Lambda }^{\mu \nu }[h,h,^2h],$$
(10)
where $`T^{\mu \nu }`$ is the matter stress-energy tensor and $`\mathrm{\Lambda }^{\mu \nu }`$ a complicated functional of $`h`$ which is at least of order $`O(h^2)`$ (and where $`\mathrm{}=\eta ^{\mu \nu }_\mu _\nu `$). We start by constructing a post-Newtonian solution of (10), initially valid in the case of a continuous (“fluid”) matter tensor $`T^{\mu \nu }`$, and parametrized by some appropriate potentials. We define a “Newtonian” potential $`V=\mathrm{}_R^1[4\pi G\sigma ]`$ where $`\mathrm{}_R^1`$ denotes the standard retarded integral and $`\sigma =(T^{00}+T^{ii})/c^2`$; we also introduce a 1PN “gravitomagnetic” potential $`V_i=\mathrm{}_R^1[4\pi G\sigma _i]`$ where $`\sigma _i=T^{0i}/c`$; some 2PN potentials $`\widehat{X}`$, $`\widehat{R}_i`$ and $`\widehat{W}_{ij}`$, e.g. $`\widehat{W}_{ij}=\mathrm{}_R^1\left[4\pi G(\sigma _{ij}\delta _{ij}\sigma _{kk})_iV_jV\right]`$ where $`\sigma _{ij}=T^{ij}`$; and finally some 3PN potentials $`\widehat{T}`$, $`\widehat{Y}_i`$ and $`\widehat{Z}_{ij}`$. In particular, the potential $`\widehat{W}_{ij}`$ generates the non-linear term $`\mathrm{}_R^1[\widehat{W}_{ij}_{ij}V]`$, involving a cubic ($`G^3`$) contribution, which is part of the potential $`\widehat{X}`$ (many other cubic terms are contained in $`\widehat{T}`$ and $`\widehat{Y}_i`$). With a specific choice of potentials we can arrange that all the quartic ($`G^4`$) terms in the metric appear in “all-integrated” form. Since $`V`$ is dominantly Newtonian, it needs to be evaluated at the 3PN order but, for instance, the term $`\mathrm{}_R^1[\widehat{W}_{ij}_{ij}V]`$, inside the 2PN potential $`\widehat{X}`$, needs only a relative 1PN precision. The metric is expressed as a functional of all these potentials; and with our particular choice of potentials, it turns out not to be too complicated.
An important point is now to determine the expression of the matter stress-energy tensor $`T^{\mu \nu }`$ appropriate to the description of point-particles. We demand that the dynamics of point-masses follows from the variation, with respect to the metric, of the action
$$I_{\mathrm{point}\mathrm{particle}}=m_1c_{\mathrm{}}^+\mathrm{}𝑑t\sqrt{(g_{\mu \nu })_1v_1^\mu v_1^\nu }+12,$$
(11)
where $`v_1^\mu =(c,d𝐲_1/dt)`$ is the coordinate velocity of particle 1. We can check that to the 3PN order all the metric coefficients $`g_{\mu \nu }`$ belong to $``$ (treating $`\mathrm{ln}r_1^{}`$ as a constant); so $`(g_{\mu \nu })_1`$ in (11) denotes the value of the metric at 1 in the sense of (2) \[or, rather, in the sense of a Lorentz-covariant Hadamard regularization defined below\]. The stationarity of the action with respect to a metric variation within the class $``$ (i.e. $`\delta g_{\mu \nu }`$) yields the stress-energy tensor
$$T_{\mathrm{point}\mathrm{particle}}^{\mu \nu }=\frac{m_1v_1^\mu v_1^\nu }{\sqrt{(g_{\rho \sigma })_1v_1^\rho v_1^\sigma /c^2}}\mathrm{Pf}\left(\frac{\delta _1}{\sqrt{g}}\right)+12,$$
(12)
where the pseudo-function $`\mathrm{Pf}(\frac{1}{\sqrt{g}}\delta _1)`$ is of the type $`\mathrm{Pf}(F\delta _1)`$ defined before. \[From the rule of multiplication of pseudo-functions we find that the matter source term in (10) involves the pseudo-function $`\mathrm{Pf}(\sqrt{g}\delta _1)`$.\] To obtain the equations of motion of the particle 1 we integrate the matter equations of motion $`_\nu T_{\mathrm{point}\mathrm{particle}}^{\mu \nu }=0`$ over a volume surrounding 1 (exclusively), and use the properties of pseudo-functions. The equations turn out to have the same form as the geodesic equations, not with respect to some smooth background but with respect to the regularized metric generated by the two bodies. Namely,
$$\frac{d}{dt}\left(\frac{(g_{\mu \nu })_1v_1^\nu }{\sqrt{(g_{\rho \sigma })_1v_1^\rho v_1^\sigma }}\right)=\frac{1}{2}\frac{(_\mu g_{\nu \lambda })_1v_1^\nu v_1^\lambda }{\sqrt{(g_{\rho \sigma })_1v_1^\rho v_1^\sigma }},$$
(13)
where all the quantities at 1 are evaluated using the regularization. Let us emphasize that the equations of motion (13) are derived from the specific expression (12) of the stress-energy tensor; had we used another expression, e.g. by replacing $`\mathrm{Pf}(\frac{1}{\sqrt{g}}\delta _1)(\frac{1}{\sqrt{g}})_1\mathrm{Pf}\delta _1`$ inside (12) (which is forbidden by the non-distributivity of Hadamard’s partie finie), we would have obtained some different-looking, and a priori uncorrect, equations.
The regularization (2) is defined stricto sensu within the spatial slice $`t=`$const, and therefore should prevent, at some stage, the equations of motion from being Lorentz invariant (recall that the harmonic-gauge condition preserves the Lorentz invariance). It is known that regularizing within the slice $`t=`$const yields the correct, Lorentz-invariant, equations of motion up to the 2PN level . We find that the breakdown of the Lorentz invariance due to the regularization (2) occurs precisely at the 3PN order. Therefore, starting at this order, we must in fact apply a Lorentz-covariant regularization in (11)-(13). Evidently, the good thing to do is to apply the Hadamard regularization in the frame at which the particle is instantaneously at rest. Let us consider the Lorentz boost $`x^\mu =\mathrm{\Lambda }_\nu ^\mu (𝐕)x^\nu `$, where $`𝐕`$ denotes the constant boost velocity. We replace all the quantities in the original frame by their equivalent expressions, developed to the 3PN order, in the new frame. Notably, the trajectories $`𝐲_1(t)`$, $`𝐲_2(t)`$ and velocities $`𝐯_1(t)`$, $`𝐯_2(t)`$ are replaced by certain functionals of $`𝐱^{}`$ and the new trajectories $`𝐲_1^{}(t^{})`$, $`𝐲_2^{}(t^{})`$ and velocities $`𝐯_1^{}(t^{})`$, $`𝐯_2^{}(t^{})`$ (where $`t^{}=x^0/c`$). We apply the Hadamard regularization within the slice $`t^{}=`$const, keeping $`𝐕`$ as a constant “spectator” vector. Finally, we re-express all the quantities back into the original frame at the point 1 ($`r_10`$), under the condition that $`𝐯_1^{}(t^{})=0`$ and (equivalently) $`𝐕=𝐯_1(t)`$. This ensures that the new frame is indeed the rest frame of the particule 1 at the instant $`t`$. Now the 3PN equations of motion are Lorentz invariant.
All the potentials $`V`$, $`V_i`$, $`\widehat{W}_{ij}`$, $`\mathrm{}`$ and their gradients are computed at the point 1, using the regularization of Poisson-type integrals defined by formulas like (8). All the derivatives appearing inside the non-linear sources of the potentials are considered as distributional and evaluated following the prescription (6). We carefully take into account the fact that the distributional derivative does not obey the Leibniz rule (it does satisfy it only in an “integrated” sense, thanks to the rule of integration by parts). An important feature of the equations at the 3PN order is the occurence of some logarithms. From (9) we know that they are necessarily of the type $`\mathrm{ln}(r_{12}/r_1^{})`$ and $`\mathrm{ln}(r_{12}/s_2)`$ in the equations of motion of body 1; interestingly, the $`\mathrm{ln}(r_{12}/s_2)`$ appears only in a quartic-interaction term proportional to $`G^4m_1m_2^3`$. Thus, at this stage, the 3PN equations of 1 depend on the constants $`\mathrm{ln}r_1^{}`$ and $`\mathrm{ln}s_2`$ (and idem for the equations of 2). Under the form we obtain them, the equations do not yet admit a conserved energy (of course we are speaking only about the conservative part of the acceleration, which excludes the radiation-reaction potential at 2.5PN order). However, we find that a conserved energy exists if and only if the logarithmic ratios $`\mathrm{ln}(r_2^{}/s_2)`$ and $`\mathrm{ln}(r_1^{}/s_1)`$ are adjusted in such a way that
$$\mathrm{ln}\left(\frac{r_2^{}}{s_2}\right)=\frac{159}{308}+\lambda \frac{m_1+m_2}{m_2}\text{and }12,$$
(14)
where $`\lambda `$ is a single numerical constant. If (and only if) the condition (14) is realized, the equations admit an energy and, in fact, a Lagrangian formulation; in this case, they depend on some arbitrary constant $`\lambda `$. The dependence upon the masses in (14) is a priori allowed. Therefore, the formalism introduces at this point an undetermined constant $`\lambda `$. \[Using (14), the equations of motion depend also on the constants $`\mathrm{ln}r_1^{}`$ and $`\mathrm{ln}r_2^{}`$, but it can be checked that the latter dependence is pure gauge.\]
Finally, having in view the application to inspiralling compact binaries, we present the 3PN relative acceleration and center-of-mass energy in the case of circular orbits. The acceleration reads as
$$\frac{d𝐯_{12}}{dt}=\omega ^2𝐲_{12}+\frac{1}{c^5}𝐅_{\mathrm{reac}}+O\left(\frac{1}{c^7}\right),$$
(15)
where $`𝐲_{12}=𝐲_1𝐲_2`$ is the relative separation in harmonic coordinates, $`𝐯_{12}=d𝐲_{12}/dt`$ the relative velocity, and $`𝐅_{\mathrm{reac}}=\frac{32}{5}\frac{G^3m^3\nu }{r_{12}^4}𝐯_{12}`$ the standard radiation-reaction force at the 2.5PN order. The mass parameters are $`m=m_1+m_2`$, $`\mu =\frac{m_1m_2}{m}`$ and $`\nu =\frac{\mu }{m}`$. The content of the 3PN approximation in (15) lies in the relation between the orbital frequency $`\omega `$ and the coordinate distance $`r_{12}=|𝐲_{12}|`$. With $`\gamma =\frac{Gm}{r_{12}c^2}`$ denoting a small post-Newtonian parameter, we get
$`\omega ^2={\displaystyle \frac{Gm}{r_{12}^3}}\{1`$ $`+`$ $`\left(3+\nu \right)\gamma +\left(6+{\displaystyle \frac{41}{4}}\nu +\nu ^2\right)\gamma ^2`$ (16)
$`+`$ $`(10+[{\displaystyle \frac{67759}{840}}+{\displaystyle \frac{41}{64}}\pi ^2+22\mathrm{ln}\left({\displaystyle \frac{r_{12}}{r_0^{}}}\right)+{\displaystyle \frac{44}{3}}\lambda ]\nu +{\displaystyle \frac{19}{2}}\nu ^2+\nu ^3)\gamma ^3+O(\gamma ^4)\}.`$ (17)
The logarithm at 3PN depends on a constant $`r_0^{}`$ defined as the “logarithmic” barycenter of the two constants $`r_1^{}`$ and $`r_2^{}`$, namely $`\mathrm{ln}r_0^{}=\frac{m_1}{m}\mathrm{ln}r_1^{}+\frac{m_2}{m}\mathrm{ln}r_2^{}`$. The constant $`r_0^{}`$ can be eliminated by a change of coordinates. The center-of-mass energy $`E`$ of the particles, such that $`\frac{dE}{dt}=0`$ as a consequence of the conservative equations of motion (neglecting $`𝐅_{\mathrm{reac}}`$), is obtained as
$`E={\displaystyle \frac{1}{2}}\mu c^2\gamma \{1`$ $`+`$ $`\left({\displaystyle \frac{7}{4}}+{\displaystyle \frac{1}{4}}\nu \right)\gamma +\left({\displaystyle \frac{7}{8}}+{\displaystyle \frac{49}{8}}\nu +{\displaystyle \frac{1}{8}}\nu ^2\right)\gamma ^2`$ (18)
$`+`$ $`({\displaystyle \frac{235}{64}}+[{\displaystyle \frac{106301}{6720}}{\displaystyle \frac{123}{64}}\pi ^2+{\displaystyle \frac{22}{3}}\mathrm{ln}\left({\displaystyle \frac{r_{12}}{r_0^{}}}\right){\displaystyle \frac{22}{3}}\lambda ]\nu +{\displaystyle \frac{27}{32}}\nu ^2+{\displaystyle \frac{5}{64}}\nu ^3)\gamma ^3+O(\gamma ^4)\}.`$ (19)
At last, by substituting the expression of $`\gamma `$ in terms of the orbital frequency $`\omega `$ following from the inverse of (16), we find that the 3PN energy in invariant form is given by
$`E={\displaystyle \frac{1}{2}}\mu c^2x\{1`$ $`+`$ $`\left({\displaystyle \frac{3}{4}}{\displaystyle \frac{1}{12}}\nu \right)x+\left({\displaystyle \frac{27}{8}}+{\displaystyle \frac{19}{8}}\nu {\displaystyle \frac{1}{24}}\nu ^2\right)x^2`$ (20)
$`+`$ $`({\displaystyle \frac{675}{64}}+[{\displaystyle \frac{209323}{4032}}{\displaystyle \frac{205}{96}}\pi ^2{\displaystyle \frac{110}{9}}\lambda ]\nu {\displaystyle \frac{155}{96}}\nu ^2{\displaystyle \frac{35}{5184}}\nu ^3)x^3+O(x^4)\},`$ (21)
with $`x=(\frac{Gm\omega }{c^3})^{2/3}`$. In the form (20) the logarithm $`\mathrm{ln}(r_{12}/r_0^{})`$ cancels out. We can compare directly this result with the one obtained by Jaranowski and Schäfer (see the equation (5.13) in ). We find that there is perfect agreement provided that $`\omega _{\mathrm{static}}=\frac{11}{3}\lambda \frac{1987}{840}`$ and $`\omega _{\mathrm{kinetic}}=\frac{41}{24}`$, where $`\omega _{\mathrm{static}}`$ and $`\omega _{\mathrm{kinetic}}`$ are the two “ambiguous” parameters found by Jaranowski and Schäfer. Thus, our undetermined constant $`\lambda `$ defined by (14) is related to the ambiguous parameter $`\omega _{\mathrm{static}}`$, while the other ambiguity $`\omega _{\mathrm{kinetic}}`$ takes a unique value . Let us stress that in the present formalism we do not meet any ambiguity in the sense of Jaranowski and Schäfer. Rather, the formalism is well-defined thanks in particular to the rules we employ for handling the pseudo-functions associated with the Hadamard regularization . All the integrals encountered in the problem have been given a precise mathematical sense, and are computed by means of a uniquely defined prescription. Yet, the appearance of the undetermined constant $`\lambda `$ suggests that the present formalism might be physically incomplete, at least in this present stage of development. Notice that the constant $`\lambda `$ enters only the term proportional to $`G^4m_1^2m_2^2(m_1+m_2)`$ in the 3PN energy, and that for general orbits, the energy contains also 164 other terms which are all uniquely determined. The details of these calculations will be published elsewhere.
|
warning/0004/cond-mat0004125.html
|
ar5iv
|
text
|
# Exact Solution of Return Hysteresis Loops in One Dimensional Random Field Ising Model at Zero Temperature
## I Introduction
Hysteresis is observed in any material which is driven by a cyclic force faster than it can equilibrate. It has practical importance, and old scientific interest renewed by present focus of statistical mechanics on nonequilibrium phenomena. There have been many theoretical studies of hysteresis recently, and also simulations and experiments . So far only exact calculations of hysteresis are limited to the random field Ising model (RFIM), in one dimension and on a Bethe lattice, at zero temperature, when the driving field changes infinitely slowly, and the system evolves from a saturated state. These limitations are forced by our analytical abilities, but make reasonable simplifications of physical systems in a regime where temperature effects on hysteresis are small. The ferromagnetic RFIM model with single spin flip dynamics at zero temperature has been proposed as a model of the Barkhausen noise by Sethna et al . It covers other phenomena as well including athermal martensitic transformations, fluid flow in porous media, and pinning of flux lines in superconductors. The difficulty (in one dimension!) of an exact solution of this model lies in the analytical treatment of quenched disorder. Even at a mean field level, the analysis of quenched disorder can involve a formalism (e.g. replica method) which belies the transparency of numerical simulations. We have used probabilistic methods to solve the antiferromagnetic RFIM in one dimension , and the ferromagnetic RFIM on a Bethe lattice as well . As indicated above, these solutions were restricted to the case where the system evolves from an initial state with all spins parallel to each other. In the present paper we are able to lift this restriction for the ferromagnetic RFIM in one dimension. We present exact solutions of return hysteresis loops starting anywhere on the parent loop. This brings the probabilistic method of solution to a level of maturity where its application to other problems appears plausible.
## II Starting with a saturated state
The one dimensional random field Ising model is characterized by the Hamiltonian,
$$H=J\underset{i}{}s_is_{i+1}\underset{i}{}h_is_ih\underset{i}{}s_i$$
(1)
Here $`s_i=\pm 1`$ are the Ising spins, $`h_i`$ is the quenched random field drawn from a continuous probability distribution $`p(h_i)`$, and h is the external field. The zero temperature dynamics amounts to flipping a spin only if it lowers the energy of the system. It normally causes an avalanche, i.e. a large number of neighboring spins have to be flipped before the system comes to a stable state. We keep the applied field fixed during an avalanche, and raise it afterwards until the next avalanche occurs.
The ferromagnetic RFIM ($`J0`$) has two important properties. It is abelian, i.e. the stable state after an avalanche does not depend upon the order in which the spins flip during an avalanche. And it has return point memory, i.e. the stable state in a slowly changing field $`h`$ depends only on the state where this field was last reversed. In the special case when we start at $`h=\mathrm{}`$, and raise the field monotonically, the state at $`h`$ does not depend on the rate of increase in $`h`$. Large rates of increase result in fewer but larger avalanches, and small rates in more numerous but smaller avalanches. The final state remains the same. We exploit this property in determining the stable state at $`h`$ through a single large avalanche from the initial state at $`h=\mathrm{}`$. The abelian property tells us that during this avalanche, whether a spin at site $`i`$ flips or not depends on the quenched field $`h_i`$ on the site and the number of nearest neighbors n ($`n=0,1,2`$) which have flipped up before it, but not on the order in which the neighbors flipped. This probability is given by,
$$p_n(h)=\text{ prob}[h_i+2(n1)J+h]0=_{2(1n)J}^{\mathrm{}}p(h_i)𝑑h_i$$
We now need to calculate the probability that a nearest neighbor of a site i flips up before site i. Let us denote the conditional probability that site i+1 (or site i-1) flips up before site i by $`P^{}(h)`$. There are many ways in which the site i+1 could be up, and we must sum over all the possibilities to calculate $`P^{}(h)`$. If site i is down, and site i+1 is up, a spin at site i+m ($`m1`$) must have flipped up before any of its neighbors were up, and then the spins from i+m to i+1 must have flipped up. Summing over these cases, we get
$$P^{}(h)=\frac{p_0(h)}{1[p_1(h)p_0(h)]}$$
The probability than an arbitrary site is up at field h is given by,
$`p(h)=[P^{}(h)]^2p_2(h)+2P^{}(h)[1P^{}(h)]p_1(h)`$ (2)
$`+[1P^{}(h)]^2p_0(h)`$ (3)
The magnetization per spin m(h) is related to p(h) by the simple equation $`m(h)=2p(h)1`$. The lower half of the large hysteresis loop in Figure 1 shows $`m(h)`$ for a Gaussian distribution of the quenched field, and in Figure 2 for a rectangular distribution. The upper half of the main loop in each case has been obtained by symmetry $`m_u(h)=m(h)`$.
## III Reversing the applied field
Reversing the applied field from $`h=+\mathrm{}`$ does not constitute a new problem because the upper half of the large hysteresis loop shown in Figure 1 can be obtained from the lower half by symmetry. However, reversing the applied field from any other point constitutes a new and somewhat more difficult problem. The reason is that in a starting state at a finite field $`h`$, whether the spin at a site is flipped or not depends in a nontrivial way on the random field at that site as well as on neighboring sites. The state is thus ”strongly correlated”, and it is difficult to do perturbation theory about this state.
For an arbitrary starting state on the lower hysteresis loop, the spins which can initiate a downward avalanche have to be separated into at least nine different categories; three categories depending on the number of nearest neighbors which are up in the starting state (n=0,1,2; these are the number of up neighbors of an up spin after the upward avalanche has settled at the point of return), and three categories depending on the number of up neighbors just before it flips up during an avalanche. The number of up neighbors before turning up in an avalanche remains important even after the avalanche because it determines the a posteriori distribution of the quenched field on the up spins in the stable state at the point of return. We also use three more categories characterized by the number of up neighbors just before a spin turns down in a downward avalanche (this number is different from the one at the starting point of the reverse trajectory).
We start backtracking from an arbitrary applied field $`h`$ on the lower loop, and come down to $`h^{}`$ ( $`h^{}h`$). We want the magnetization at $`h^{}`$. Obviously, spins can only flip down on the reverse trajectory, and therefore we focus on spins which are up at $`h`$ but turn down at $`h^{}`$. We divide the up spins at $`h`$ into three basic categories characterizing the range of their random field, and how they turned up on the lower hysteresis loop. Spins in category-0 have $`h_i2Jh`$. These spins could turn up at $`h`$ even if none of their neighbors were up to help them. Spins in category-1 have $`hh_i2Jh`$, and spins in category-2 have $`2Jhh_ih`$. No spin could be up at $`h`$ if it has $`h_i2Jh`$. How the spins turn down at $`h^{}`$ on the reverse trajectory is determined by the random field on a spin and the number of up neighbors it has just before it turns down. The three basic categories listed above were determined by the number of up neighbors just before a spin turned up during an upward avalanche at $`h`$. After that avalanche has settled, the number of up neighbors may increase. Thus each of the basic categories can be further divided into three categories characterized by the number of up neighbors after the avalanche. Some of these sub-categories may be empty. For example, a spin of category-2 which is up at $`h`$ necessarily has both neighbors up. Spins of category-1 could have one or both neighbors up. Spins of category-0 could have zero, one, or two neighbors up at $`h`$. When the applied field is reversed, spins of category-2 with both neighbors up are as susceptible to turn down as spins of category-1 with one neighbor up because the net field in both cases lies in the same range.
In the first instance, we consider a restricted range of the reversed field: $`h2Jh^{}h`$. In this range, the only spins which could turn down are spins of category-2 with two neighbors up, spins of category-1 with one neighbor up, and spins of category-0 with zero neighbors up. We add the contributions from these three categories, and subtract the sum from the number of up spins at $`h`$. This gives us the magnetization at $`h^{}`$. Consider the spins of category-2 first; their fraction at $`h`$ is equal to $`[P^{}(h)]^2[p_2(h)p_1(h)]`$. The factor $`[P^{}(h)]`$ gives the probability that a nearest neighbor of a spin is up on the lower hysteresis loop before that spin is relaxed. Thus $`[P^{}(h)]^2`$ is the probability that both neighbors of the spin are up before it is relaxed. The factor $`[p_2(h)p_1(h)]`$ gives the probability that the spin turns up if two neighbors are up but not if only one neighbor is up. Thus, the fraction of category-2 spins which turn down at $`h^{}`$ on the return loop is given by,
$$q_r^2(h,h^{})=[P^{}(h)]^2[p_2(h)p_2(h^{})].$$
Now we take up the spins of category-1. In the initial state at $`h`$, category-1 spins come in two sub-categories; (i) spins with one neighbor up, and (ii) spins with two neighbors up. In the restricted range of the reversed field ($`h2Jh^{}h`$), spins in sub-category (ii) can not turn down spontaneously. However they can turn down in an avalanche, if the avalanche puts one of their neighbors in category (ii) and it turns down. An avalanche can start with a category-1 spin which has one neighbor down in the starting state at $`h`$. This occurs with the probability $`f(h)`$ given by,
$$f(h)=\{1p_2(h)\}[P^{}(h)]+\{1p_1(h)\}[1P^{}(h)]$$
The above equation can be understood as follows. Suppose, the spin at site i is up, and $`f(h)`$ denotes the probability that the spin at site i+1 is down. Before the spin at site i+1 is relaxed, the spin at site i+2 is up with the probability $`[P^{}(h)]`$, and down with the probability $`[1P^{}(h)]`$. The probability that the spin stays down in the two cases even after it is relaxed is given by $`\{1p_2(h)\}`$ and $`\{1p_1(h)\}`$ respectively. The probability for the spin at i+1 to flip down at $`h^{}`$ is equal to $`[p_1(h)p_1(h^{})]`$. After it flips down, the spin at i-1 can also flip down with the same probability if it belongs to category-1 and the spin at i-2 is up. Thus an avalanche can start. The avalanche will go on till it meets a category-1 spin which does not flip down at $`h^{}`$ or it meets a category-0 spin which has an up neighbor on the other side. The probability that a nearest neighbor of an up spin is down at $`h^{}`$ is given by,
$$q_a(h,h^{})=\frac{f(h)}{1[p_1(h)p_1(h^{})]}$$
Here, $`f(h)`$ is the probability that the neighbor was already down in the initial state. The other factor is the sum of an infinite series which accounts for avalanches of various sizes which may bring the neighbor down.
An avalanche can also be started by a spin of category-2 flipping down. This gives another term,
$$q_b(h,h^{})=\frac{[p_2(h)p_2(h^{})][P^{}(h)]}{1[p_1(h)p_1(h^{})]}$$
The numerator in the above equation can be understood as follows. Suppose the spins at sites i, i+1, and i+2 are up and site i+1 belongs to category-2. $`[P^{}(h)]`$ is the probability that site i+2 was up before site i+1 was relaxed at $`h`$. The numerator gives the probability that the right side neighbor of the up spin at site i flips down at $`h^{}`$. The denominator takes care of any possible avalanches started by the flipping down of a category-2 site. The total probability that a nearest neighbor of an up spin is down at $`h^{}`$ is equal to $`q_a+q_b`$. We also need the probability that a nearest neighbor of an up spin is up before that spin is relaxed at $`h^{}`$. This is equal to the probability that the neighbor in question was up on the lower hysteresis loop before the site was relaxed at $`h`$, i.e. it is equal to $`P^{}(h)`$. With this knowledge, we are now in a position to write the fraction of category-1 spins which turn down on the return loop at $`h^{}`$. We get,
$$q_r^1(h,h^{})=2[P^{}(h)][q^a(h,h^{})+q_b(h,h^{})][p_1(h)p_1(h^{})]$$
Spins of category-1 can not have both nearest neighbors down. The reason is that this class of spins are flipped up during an avalanche on the lower hysteresis loop. Therefore they must be connected by up spins to a spin of category-0 on one side at least. A spin of category-0 can not turn down if it has at least one neighbor up. However, if both neighbors of a spin of category-0 are down at $`h^{}`$, it may turn down. The fraction of such spins is given by,
$$q_r^0(h,h^{})=[q^a(h,h^{})+q_b(h,h^{})]^2[p_0(h)p_0(h^{})]$$
We are now in a position to write the magnetization on the return loop in range $`[h2Jh^{}h]`$. We get, $`m^{}(h^{})=2p^{}(h^{})1`$, where
$$p^{}(h^{})=p(h)q_r^2(h,h^{})q_r^1(h,h^{})q_r^0(h,h^{})$$
(4)
The key to getting the return magnetization beyond the range considered above is to note that the state of the system on the reverse trajectory at $`h^{}=h2J`$ is the same as would be obtained from the initial state at $`h^{}=+\mathrm{}`$. If the initial state at $`h^{}=\mathrm{}`$ is exposed to an applied field $`h2J`$, spins with $`h_ih`$ will flip down spontaneously and start avalanches where the adjacent spins in the range $`hh_i2Jh`$ will flip down. When this avalanche is finished, the remaining up spins will belong to three categories: (i) spins with $`h_i2Jh`$ with one neighbor up, (ii) spins with $`h_i4Jh`$ with no neighbors up, and (iii) spins with $`h_ih`$ with two neighbors up. This is precisely the state obtained at the end of the reverse trajectory obtained above. Therefore, the reverse trajectory in the range $`h^{}h2J`$ merges with the upper half of the big hysteresis loop.
## IV Reversing the field again
The magnetization in reversed field merges with the upper half of the big hysteresis loop when the field falls below $`h2J`$. Pulling up the field from below $`h2J`$ can be related by symmetry to the problem of the return loop analyzed in the previous section. We need not repeat this calculation. However, if the reversed field is re-reversed before it reaches $`h2J`$, we have a new problem on our hands which we now analyze.
We turn back the field at $`h^{}`$. Our object is to obtain the magnetization at an arbitrary value $`h^{\prime \prime }`$ ($`h^{}h^{\prime \prime }h)`$ on the lower half of the return loop. Essentially, we are looking at the same strings of spins which turned down in the previous section, but now they turn up from the other end. If a spin is down on the lower half of the return loop, it must have been down at end of the upper half as well. The reason is that on the lower half, spins can only flip up, none can flip down. Thus the probability that a nearest neighbor of a down spin is down on the lower return loop is equal to $`q_a(h,h^{})+q_b(h,h^{})`$. The probability that the nearest neighbor is up increases steadily as more spins flip up on the lower half. First, let us look at the probability of an up neighbor at the start of the lower return loop. Consider three adjacent sites: i-1, i, and i+1. Given that site i+1 is down, we want the probability that site i is up. It follows from the previous section that if site i is up at $`h^{}`$, it must be a spin of category-0, or there must be a string of up spins to the left of i containing a spin of category-0. Spins of category-0 are up with probability unity if they are adjacent to an up spin, otherwise they have to have a quenched field in excess of $`4Jh`$. Thus the probability that site $`i`$ is up and is a spin of category-0 is equal to $`[1(q_a+q_b)]p_0(h)+(q_a+q_b)p_0(h^{\prime \prime })`$. The probability that site $`i`$ is up and not a spin of category-0 is equal to $`[P^{}(h)][p_1(h^{})p_0(h)]`$. Putting it together, the probability that a nearest neighbor of a down spin is up on the lower return loop before that neighbor is relaxed is given by:
$$p_{rr}(h,h^{},h^{\prime \prime })=\frac{a}{1[p_1(h^{\prime \prime })p_1(h^{})]}$$
where,
$`a=[p_1(h^{})p_0(h)]P^{}(h)+[1(q_a+q_b)]p_0(h)`$ (5)
$`+(q_a+q_b)p_0(h^{\prime \prime })`$ (6)
The magnetization on the lower return loop is given by $`m^{\prime \prime }(h^{\prime \prime })=2p^{\prime \prime }(h^{\prime \prime })1`$, where
$`p^{\prime \prime }(h^{\prime \prime })=p^{}(h^{})+(q_a+q_b)^2[p_0(h^{\prime \prime })p_0(h^{})]`$ (7)
$`+2(q_a+q_b)p_{rr}(h,h^{},h^{\prime \prime })[p_1(h^{\prime \prime })p_1(h^{})]`$ (8)
$`+p_{rr}^2(h,h^{},h^{\prime \prime })[p_2(h^{\prime \prime })p_0(h^{})]`$ (9)
As may be expected, the analytical results agree quite well with numerical simulations of the model. Figure 1 shows a comparison for a Gaussian distribution of the random field, and Figure 2 for a rectangular distribution. Analytical expressions are shown by continuous lines. Simulations for a chain of $`1000`$ spins (averaged over $`1000`$ different realizations of the random field distribution) are indistinguishable from the analytical expressions, but these are shown by large symbols at sparse intervals for visual convenience.
## V Concluding remarks
The nonequilibrium response of RFIM at zero temperature is related to experimentally measurable quantities in several diverse systems. It has been calculated analytically in one dimension using probabilistic methods, and checked against numerical simulations of the model. It remains for the future to apply the present method in higher dimensions, although it should be qualitatively similar.
I thank Deepak Dhar for critical reading of the manuscript.
Caption Figure 1: Hysteresis loop (filled squares) between two saturated states for a Gaussian random field (mean=0, variance=1, J=1). Two excursions from the lower half are shown; $`h=1`$ to $`h^{}=1`$ and back (open squares), and $`h=1`$ to $`h^{}=.6`$ and back (open circles).
Caption Figure 2: Hysteresis loop (filled squares) for a rectangular distribution of the random field of width 6 (J=1). Return loop (open squares) shows an excursion from the lower half ($`h=1.5`$ to $`h^{}=.5`$ and back).
|
warning/0004/hep-ph0004135.html
|
ar5iv
|
text
|
# The decay 𝐻→𝛾𝛾 in Multi-Higgs doublet models
## Abstract
We study the dominant decays of the lightest Higgs boson in models with 2 and 3 Higgs doublets, for the case when its couplings to fermions are absent at tree-level. It is found that the branching ratio for the decay $`H\gamma \gamma `$ is above the one into fermion pairs, which is evaluated also at the 1-loop level.
The search for the Higgs boson of the standard model (SM) , is the most important test of the symmetry breaking and mass generation of the theory. Current limits on the SM Higgs mass coming from direct searches for the Higgs boson, specifically from the study of the reaction $`e^+e^{}Z(Z^{}ff)h`$ at LEPI. The combined limit of the four experiments on the Higgs mass is $`m_h>65.4`$ GeV . At LEPII with the total energy $`\sqrt{s}=130200`$ GeV, the dominant Higgs production process is $`e^+e^{}hZ`$, where the final state particles in the analysed Higgs boson channels are $`e^+e^{}(Zqq,bb,\nu \nu ,\tau \tau ,e^+e^{},\mu \mu )(hbb,\tau \tau )`$. Combined limit of the four experiments with $`\sqrt{s}=195.6`$ GeV gives $`m_h102.6`$ GeV at $`95\%`$ C.L.. LEPII running with the total energy $`200`$ GeV will be able to discover standard Higgs boson with a mass up to $`107`$ GeV .
On the other hand, indirect bound on the Higgs mass can be obtained from precision electroweak measurements. Although the sensitivity to the Higgs boson mass through radiative corrections is only logarithmic, the increasing precision in the measurement of electroweak observables allows to derive constraints on $`m_h`$, around of $`m_h=71_{42}^{+75}\pm 5`$ GeV . Other constraints coming from tree level unitarity in $`W_LW_L`$ scattering, $`m_h1`$ TeV , validity of perturbation theory, $`m_h930`$ GeV , and the analisys done of vacuum stability, $`m_h120`$ GeV . On the other hand, the search techniques for intermediate and heavy Higgs boson are known, and their implementation requires the next generation of hadron colliders LHC . One of the most important reactions for the search for the Higgs boson at LHC is $`pp(h\gamma \gamma )`$ which is the most promising one for the search in the region $`100m_h140`$ GeV .
In this paper we shall study the Higgs sector for two particular models, which contain 2 and 3 Higgs doublets respectively, for the case when some Higgs boson couples only to gauge bosons but not to fermions, at tree-level. In this case the decay into $`\gamma \gamma `$ is expected to dominate for the lower part of the intermediate mass region, however the decays into fermion pairs can be generated also at the 1-loop level, and one needs to include their contribution into the total width, in order to know the precise values of branching ratios, which to our knowledge has not been done in the literature .
We shall discuss first the Lagrangian for the model with two doublets,then the dominant branching ratios for the intermediate mass range are evaluated, and we also discuss the limits on the Higgs mass that can be obtained from the LEP data. Later on, it will be explained how to obtain the corresponding results for the 3-Higgs doublet model.
The extension of the SM with two Higgs doublets has been studied in great detail before . In terms of components the two doublets are written as: $`\mathrm{\Phi }_1=(\varphi _1^+,\varphi _1^0)^T`$, $`\mathrm{\Phi }_2=(\varphi _2^+,\varphi _2^0)^T`$. It happens that the Yukawa coupling of the Higgses with the fermions can be choosen in two ways, usually denoted as models I and II. In model I, one doublet is used to generate masses for both the U- and D-type quarks, whereas in model II one doublet generates the masses of the U-type quarks and the second one generates the masses of the D-type quarks. We shall consider here only model I.
After diagonalizing the general Higgs potential, one gets the scalar mass eigenstates, which include one charged pair ($`H^\pm `$), two CP-even scalars ($`h^0`$, $`H^0`$, with $`m_H>m_h`$), and one CP-odd scalar ($`A^0`$). Thus, the free parameters are the scalar masses, the mixing angle $`\alpha `$ and the ratio of vev’s $`\mathrm{tan}\beta =v_2/v_1`$. The mass eigenstates can be written in terms of the components of the Higgs doublets; for the CP-even scalars for example, one has:
$$H^0=2^{1/2}[(Re\varphi _1^0v_1)\mathrm{cos}\alpha +(Re\varphi _2^0v_2)\mathrm{sin}\alpha ]$$
(1)
$$h^0=2^{1/2}[(Re\varphi _1^0v_1)\mathrm{sin}\alpha +(Re\varphi _2^0v_2)\mathrm{cos}\alpha ]$$
(2)
The interaction of fermions with Higgs bosons in model I are given by the Yukawa Lagrangian,
$`L=`$ $``$ $`{\displaystyle \frac{g}{2m_W\mathrm{sin}\beta }}[\overline{D}M_DD+\overline{U}M_UU](H^0\mathrm{sin}\alpha +h^0\mathrm{cos}\alpha )`$ (3)
$`+`$ $`{\displaystyle \frac{ig\mathrm{cot}\beta }{2m_W}}[\overline{D}M_D\gamma _5D+\overline{U}M_U\gamma _5U]A^0`$
$`+`$ $`{\displaystyle \frac{g\mathrm{cot}\beta }{2^{3/2}m_W}}(H^+\overline{U}[M_UKP_LKM_DP_R]D+h.c.)`$
where $`P_{R,L}=(1\pm \gamma _5)/2`$, $`M_{U,D}`$ are the diagonal mass matrices of the U- and D-type quarks, $`K`$ is the Kobayashi-Maskawa mixing matrix. It happens for this type of models that the important coupling $`h^0VV`$ ($`V=Z,W`$) is proportional to $`\mathrm{sin}(\beta \alpha )`$. Moreover, if the mixing angle takes the value $`\alpha =\pi /2`$, then there is no mixing among $`H^0`$ and $`h^0`$, and one has that $`h^0`$ interact only with the gauge bosons, with the coupling proportional to $`\mathrm{sin}(\beta \pi /2)=\mathrm{cos}\beta `$.
The dominant decays of $`h^0`$ in the mass range $`m_h>2m_Z`$ are into $`WW`$, $`ZZ`$; whereas for the intermediate mass range (80 GeV $`<m_h<2m_Z`$) the allowed decays are into $`\gamma \gamma `$, $`Z\gamma `$ , $`WW^{}`$, $`ZZ^{}`$ , which will compete also with the decays into fermion pairs, generated at the 1-loop level.
The decay width into photon pairs can be written as:
$$\mathrm{\Gamma }(h^0\gamma \gamma )=\mathrm{cos}^2\beta \mathrm{\Gamma }_{sm}^W(h^0\gamma \gamma ),$$
(4)
where $`\mathrm{\Gamma }_{sm}^W`$ denotes the W-loop contribution to the decay width of the SM Higgs boson; the decay width for $`h^0Z+\gamma `$ has also the same form, and it will be bellow the one for $`h^0\gamma \gamma `$, as in the SM case, thus we shall not discuss it further here. Similarly, we find that the decays into $`WW^{}`$ and $`ZZ^{}`$ can be written in the same form, namely
$$\mathrm{\Gamma }(h^0VV^{})=\mathrm{cos}^2\beta \mathrm{\Gamma }(\varphi _{sm}^0VV^{})$$
(5)
Finally, the expression for the decay width into fermion pairs ($`h^0f\overline{f}`$), that results after one evaluates the 1-loop amplitude is written as follows:
$$\mathrm{\Gamma }(h^0f\overline{f})=\frac{G_F\alpha ^2\pi }{2\sqrt{2}\mathrm{sin}^4\theta _W}m_hm_f^2\mathrm{cos}^2\beta F(m_h,m_i,m_W)$$
(6)
where $`m_i`$ is the mass of the fermion that enters in the loop, and $`F(m_h,m_i,m_W)`$ is a function that arises from the loop integration, which is written as follows,
$$F=[|F_1|^2+|F_2|^2]|K_if|^2,$$
(7)
where,
$`F_1`$ $`=`$ $`4m_W^2C_{12}+m_h^2(C_0C_{12}+C_{23}C_{11})+m_i^2(C_{12}C_0),`$
$`F_2`$ $`=`$ $`4m_W^2(C_0C_{11}+2C_{12})m_h^2(C_0+C_{11}+C_{12}+C_{23})`$ (8)
$`+m_i^2(2C_{11}C_0),`$
where $`C_{ij}=C_{ij}(m_h,m_i,m_W)`$ can be written in terms of the scalar integral $`C_0`$, as discussed in . From this expression one notices that the width is again proportional to $`m_f`$, which will suppress the width, thus only the heaviest fermions will contribute significantly. This result can be understood easily if one follows the chirality in the graphs, which need a mass insertion to be different from zero. In the following analysis we include only the contribution from the top quark to the loop, which is the dominant one. The resulting branching ratios are presented in fig. 1, where one can see that the decay into a photon pair dominates for mass values of the Higgs up to about $`m_Z`$.
On the other hand, if one considers a model which allows Higgs-fermion couplings, and one assumes that there are no Flavour Changing Neutral Currents (FCNC) mediated by the neutral Higgs bosons, then the couplings $`h^0ff`$ are proportional to $`m_f`$, and then the production rates will be highly suppressed; whereas if one allows for the presence of FCNC, then the Higgs-fermions couplings are not neccessarly proportional to the fermion masses , and the cross-section can be large.
One can use the experimental results to constrain the mass of the Higgs boson for this kind of models. This can be done using the LEP bound $`BR(Z\nu \nu \gamma \gamma )<10^6`$ , which can be written for this model as,
$$BR(Z\nu \nu \gamma \gamma )=BR(Z\nu \nu +h^0)BR(h^0\gamma \gamma ),$$
(9)
which depends only on $`m_h`$ and $`\mathrm{tan}\beta `$.
Fig. 2 shows the excluded region in the plane $`\mathrm{tan}\beta m_h`$, obtained from the previous equation. It is important to point out that this is the first bound that is obtained for a model of this type, which eventhough is valid only for an specific value of $`\alpha `$, it does not depends on the remmaining parameters of the general Higgs potential. For low values of $`\mathrm{tan}\beta `$ the limit on the Higgs mass is $`m_h>91`$ GeV, which is similar to the one obtained for the SM Higgs.
Finally, we have also evaluated the branching ratios of Higgs bosons within the context of a model with 3-Higgs doublets, where 2 of them behave like the doublets of model II in the 2-doublet case. The third doublet couples only to vector bosons, and we find again that the dominant decay in the intermediate mass range, is into photon pairs.
The scalar potential for this model, which allows for the existence of one CP-even Higgs boson that does not couples at fermions, is the following,
$$V=V(\mathrm{\Phi }_1,\mathrm{\Phi }_2)+V(\mathrm{\Phi }_3)$$
(10)
where $`V(\mathrm{\Phi }_1,\mathrm{\Phi }_2)`$ is the two-Higgs doublet model , whereas $`\mathrm{\Phi }_3`$ contains the Higgs scalars that do not mix with the other Higgs bosons; each of the doublets can be written as $`\mathrm{\Phi }_i=[\varphi _i^+,h_i+v_i+i\eta _i]`$. Then, $`h_3`$ can be choosen as the light Higgs that only couples to gauge bosons, whose coupling is $`g_{h_3VV}=\mathrm{sin}\gamma g_{\varphi _{sm}^0VV}`$ (where $`\mathrm{tan}\gamma =v_3/(v_1^2+v_2^2+v_3^2)^{1/2}`$).
We have derived all the relevant Feynman rules of the model, needed to evaluate the decay widths at tree-level and 1-loop, and the final results is that all the branching ratios of $`h_3`$ can be obtained from the ones obtained previously for the two Higgs doublets model I, just by replacing $`\mathrm{cos}\beta \mathrm{sin}\gamma `$. Thus, the previous limits on the Higgs mass apply also for this scalar. Models with N-doublets have been analyzed in the literature too , and it is possible to translate our limits for such models by taking the appropriate limit.
In summary, we find that the scalar sectors studied in this paper have an interesting phenomenology in their own. And, it is possible to use the LEP results to put limits on the Higgs mass, which are comparable to the ones obtained for the SM Higgs for low values of $`\mathrm{tan}\beta `$, namely $`m_h>91`$ GeV.
At $`e^+e^{}`$ machines with TeV CM-energies, it will be possible to produce the Higgs boson of these models by WW fusion, and also in association with Z. The production of a Higgs by photon-photon fusion, could be also important too, unfortunately we found that at LEP energies the event rate is far bellow detectability. On the other hand, the phenomenological consequences of these models at hadron colliders are also interesting. Because of the absence of Higgs-fermions couplings, it will not be possible to produce the Higgs in association with top pairs, and neither by gluon fusion (which will occur only at the 2-loop level), which have proved to be usefull for the SM case. Thus, the main production mechanism will be in association with W/Z, however since the decays into gauge bosons will dominate, its detection could be feasible. Clearly, a detailed study is needed in order to determine the detection feasibilities of these modes, which is beyond the scope of this work.
We acknowledge discussions with L. Diaz-Cruz. This work has been finantialy supported by COLCIENCIAS(Colombia) and CONACyT (Mexico).
Figure Caption
Figure 1: Branching ratios for the decay of the Higgs boson; $`h\gamma \gamma `$: solid , $`hb\overline{b}`$: dot-dash, $`hWW^{}`$: dashes, $`hZZ^{}`$: dots.
Figure 2: Regions in the plane $`\mathrm{tan}\beta m_h`$ excluded by the LEP results (shaded), for the models discussed in the text.
|
warning/0004/hep-th0004087.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
The derivation of exact results in $`𝒩=1`$supersymmetric gauge theories based on low energy effective superpotentials and holomorphy was pioneered in , then new wave of development was initiated by Seiberg, see for review. Additional input was provided by the Seiberg-Witten solution of $`𝒩=2`$supersymmetric gauge theories with and without matter .
The key feature of the $`𝒩=2`$theory is the existence of the Coulomb branch where the vacuum expectation value of the adjoint scalar serves as a modulus . The solution is described in terms of Riemann surfaces and the Coulomb branch parametrizes the moduli space of their complex structures. The simplest way to break $`𝒩=2`$supersymmetry (SUSY) down to $`𝒩=1`$amounts to giving a nonvanishing mass $`\mu `$ to the chiral $`𝒩=1`$superfield in the adjoint representation. This field is a partner to the gauge fields in the $`𝒩=2`$supermultiplet. At small values of $`\mu `$ the theory is close to its $`𝒩=2`$counterpart while at large $`\mu `$ the adjoint matter decouples and the pure $`𝒩=1`$theory emerges. The emerging theory at large $`\mu `$ is close to supersymmetric QCD (SQCD) but does not coincide with it. A trace of the massive adjoint remains in the effective theory in the form of nonrenormalizable quartic terms in the superpotential which are suppressed by $`1/\mu `$. Although in the $`𝒩=1`$theory the degeneracy on the Coulomb branch is lifted by the superpotential, memory of the structure of the Riemann surfaces remains. Namely, the vanishing of the discriminant of the Riemann surface defines the set of vacua in the corresponding $`𝒩=1`$theory .
Different vacua are distinguished by the values of chiral condensates, such as the gluino condensate $`\mathrm{Tr}\lambda \lambda `$ and the condensate of fundamental matter $`\stackrel{~}{Q}Q`$. Generically, the latter can be found in SQCD using the effective superpotential, while the gluino condensate can be evaluated using the Konishi anomaly which relates the two condensates (see Ref. for a review). To obtain the condensate in pure $`𝒩=1`$Yang-Mills theory one has to start with the massive SQCD and use the holomorphy to decouple massive matter. Recently, some points concerning formation of the condensate and identification of the relevant field configurations were clarified in
The brane picture provides another approach to the problem. The brane configurations for $`𝒩=1`$theories with different matter content are known and the recipe for calculating minima of the superpotentials has been formulated . The key point concerning the brane configurations is that to break SUSY down to $`𝒩=1`$one has to rotate the $`𝒩=2`$picture. However only configurations which correspond to the vanishing of the discriminant can be rotated which means that at any value of the adjoint mass these points remain intact. The superpotentials calculated from brane configurations are in correspondence with field theory expectations.
In this paper we consider an $`𝒩=1`$theory with both adjoint and fundamental matter and limit ourselves to the most tractable case of SU(2) gauge group with one fundamental flavor and one multiplet in the adjoint representation. Our strategy is as follows: First, we integrate out the adjoint matter to get SQCD-like effective superpotential for the fundamental matter. The only nonperturbative input in this effective superpotential is given by the Affleck-Dine-Seiberg superpotential generated by one instanton . Difference with pure SQCD is due to the tree level nonrenormalizable term generated by the heavy adjoint exchange, mentioned above. Similarly to SQCD, the effective superpotential together with the Konishi relations unambiguously fixes condensates of fundamental and adjoint matter as well as the gaugino condensates in all three vacua of the theory.
We then compare the condensate of the adjoint matter with points in the $`u`$ plane corresponding to the vanishing of the discriminant defined by Seiberg-Witten solution in $`𝒩=2`$theory and find a complete match. Our results for matter and gaugino condensates are consistent with those obtained by the ‘integrating in’ method and can be viewed as an independent confirmation of this method. What is specific to our approach is that we start from the weak coupling regime where the notion of an effective Lagrangian is well defined, and then use holomorphy to extend results for chiral condensates into strong coupling.
We subsequently determine monopole, dyon, and charge condensates following the Seiberg-Witten approach, i.e. considering effective superpotentials near singularities on the Coulomb branch of the $`𝒩=2`$theory. Again, holomorphy allows us to extend our results to the domain of the “hard” $`𝒩=2`$breaking. This extension include not only the mass term of adjoint but also breaking of $`𝒩=2`$in Yukawa couplings.
Our next step is the study chiral condensates in the Argyres-Douglas (AD) points. These points were originally introduced in the moduli/parameter space of $`𝒩=2`$theories as points where two singularities on the Coulomb branch coalesce . It is believed that the theory in the AD point flows in the infrared to a nontrivial superconformal theory. The notion of the AD point continues to make sense even when the $`𝒩=2`$theory is broken to $`𝒩=1`$; in the $`𝒩=1`$theory it is the point in parameter space where two vacua collide.
In particular, we consider the AD point where the monopole and charge vacua collide at a particular value of the mass of the fundamental flavor. Our key result is that both monopole and charge condensates vanish at the AD point <sup>1</sup><sup>1</sup>1Vanishing of condensates for coalescing vacua was mentioned by Douglas and Shenker in the context of SU($`N`$) theories without fundamental matter for $`N3`$. Note, that it was before the notion of the AD point was introduced . . We interpret this as deconfinement of both electric and magnetic charges at the AD point. It provides evidence that the theory at the AD point remains superconformal even after strong breaking of $`𝒩=2`$to $`𝒩=1`$. Argyres and Douglas conjectured this in their consideration of SU(3) theory
Let us recall that the condensation of monopoles ensures confinement of quarks in the monopole vacuum , while the condensation of charges provides confinement of monopoles in the charge vacuum. As shown by ’t Hooft it is impossible for these two phenomena to coexist. This apparently leads to a paradoxical situation in the AD point where the monopole and charge vacua collide. Our result resolves this paradox.
This paradox is a part of more general problem: whether there is an uniquely defined theory at the AD point. Indeed, when two vacua collide the Witten index of the emerging effective theory at the AD point is fixed, namely there are two bosonic vacuum states. The question is whether there is any physical quantity which could serve as an order parameter differentiating these two vacua. The continuity of chiral condensates in the AD point we find shows that these condensates are not playing this role. The same continuity also leads to vanishing tension for domain walls interpolating between colliding vacua when we approach the AD point. We discuss if these domain walls could serve as a signal of two vacua in the AD point.
The paper is organized as follows. In Sec. 2 we dwell on the calculation of matter and gaugino condensates, while monopole, charge and dyon condensates are considered in Sec. 3. In Sec. 4 we briefly discuss a definition of the theory at the AD point and the related problem of domain walls. Our results are discussed in Sec. 5.
## 2 Matter and gaugino condensates
### 2.1 Effective superpotential and condensates
We consider a $`𝒩=1`$theory with SU(2) gauge group where the matter sector consists of the adjoint field $`\mathrm{\Phi }_\beta ^\alpha =\mathrm{\Phi }^a(\tau ^a/2)_\beta ^\alpha `$ ($`\alpha ,\beta =1,2;a=1,2,3`$), and two fundamental fields $`Q_f^\alpha `$ $`(f=1,2)`$ describing one flavor. The general renormalizable superpotential for this theory has the form,
$$𝒲=\mu \mathrm{Tr}\mathrm{\Phi }^2+\frac{m}{2}Q_f^\alpha Q_\alpha ^f+\frac{1}{\sqrt{2}}h^{fg}Q_{\alpha f}\mathrm{\Phi }_\beta ^\alpha Q_g^\beta .$$
(1)
Here the parameters $`\mu `$ and $`m`$ are related to the masses of the adjoint and fundamental fields, $`m_\mathrm{\Phi }=\mu /Z_\mathrm{\Phi }`$, $`m_Q=m/Z_Q`$, by the corresponding $`Z`$ factors in the kinetic terms. Having in mind normalization appropriate for the $`𝒩=2`$case we choose for bare parameters $`Z_\mathrm{\Phi }^0=1/g_0^2`$, $`Z_Q^0=1`$. The matrix of Yukawa couplings $`h^{fg}`$ is symmetric, and summation over color indices $`\alpha ,\beta =1,2`$ is explicit. Unbroken $`𝒩=2`$SUSY appears when $`\mu =0`$ and $`deth=1`$.
To obtain an effective theory similar to SQCD we integrate out the adjoint field $`\mathrm{\Phi }`$ implying that $`m_\mathrm{\Phi }m_Q`$. In the classical approximation this integration reduces to the substitution
$$\mathrm{\Phi }_\beta ^\alpha =\frac{1}{2\sqrt{2}\mu }h^{fg}\left(Q_{\beta f}Q_g^\alpha \frac{1}{2}\delta _\beta ^\alpha Q_{\gamma f}Q_g^\gamma \right),$$
(2)
which follows from $`𝒲/\mathrm{\Phi }=0`$. What is the effect of quantum corrections on the effective superpotential? It is well known from the study of SQCD that perturbative loops do not contribute and nonperturbative effects are exhausted by the Affleck-Dine-Seiberg (ADS) superpotential generated by one instanton . The effective superpotential then is
$$𝒲_{\mathrm{eff}}=mV\frac{(deth)}{4\mu }V^2+\frac{\mu ^2\mathrm{\Lambda }_1^3}{4V}$$
(3)
where the gauge and subflavor invariant chiral field $`V`$ is defined as
$$V=\frac{1}{2}Q_f^\alpha Q_\alpha ^f.$$
(4)
The first two terms in Eq. (3) appear on the tree level after substitution (2) into Eq. (1) while the third nonperturbative one is the ADS superpotential. The scale parameter $`\mathrm{\Lambda }_1`$ is given in terms of the mass of Pauli-Villars regulator $`M_{\mathrm{PV}}`$ and the bare coupling $`g_0`$ (plus the vacuum angle $`\theta _0`$) as
$$\mathrm{\Lambda }_1^3=4M_{\mathrm{PV}}^3\mathrm{exp}\left(\frac{8\pi ^2}{g_0^2}+i\theta _0\right).$$
(5)
The coefficient $`\mu ^2\mathrm{\Lambda }_1^3/4`$ in the ADS superpotential is equivalent to $`\mathrm{\Lambda }_{\mathrm{SQCD}}^5`$ in SQCD. The factor $`\mu ^2`$ in the coefficient reflects four fermionic zero modes of the adjoint field, see e.g. Ref. for details.
The only term in the superpotential (3) which differentiates it from the SQCD case is the second term which is due to tree level exchange by the adjoint field. At $`h=0`$ it vanishes and we are back to the known SQCD case with two vacua and a Higgs phase for small $`m`$.
When $`deth`$ is nonvanishing we have three vacua, marked by the vevs of the lowest component of $`V`$,
$$v=V.$$
(6)
These vevs are roots of the algebraic equation $`\mathrm{d}𝒲_{\mathrm{eff}}/\mathrm{d}v=0`$ which has the form
$$m\frac{(deth)}{2}\frac{v}{\mu }\frac{\mathrm{\Lambda }_1^3}{4}\left(\frac{\mu }{v}\right)^2=0.$$
(7)
This equation shows, in particular, that although the second term in the superpotential (3) seems to be suppressed at large $`\mu `$ it turns out to be of the same order as the ADS term. From Eq. (7) it is also clear that the dependence on $`\mu `$ is given by the scaling $`v\mu `$.
To see the dependence on the other parameters let us substitute $`v`$ by the dimensionless variable $`\kappa `$ defined by the relation
$$v=\mu \sqrt{\frac{\mathrm{\Lambda }_1^3}{4m}}\kappa .$$
(8)
Then Eq. (7), when rewritten in terms of $`\kappa `$,
$$1\sigma \kappa \frac{1}{\kappa ^2}=0$$
(9)
is governed by the dimensionless parameter $`\sigma `$,
$$\sigma =\frac{(deth)}{4}\left(\frac{\mathrm{\Lambda }_1}{m}\right)^{3/2}.$$
(10)
We see that the two parameters $`m`$ and $`deth`$ enter only as $`m(deth)^{2/3}`$. The dependence of $`v`$ on $`\mu `$ is linear as we discussed above.
The particular dependence of condensate $`v`$ on the parameters $`\mu `$, $`m`$ and $`deth`$ follows from the $`R`$ symmetries of the theory. Following Seiberg one can consider $`\mu `$, $`m`$ and $`deth`$ as background fields and identify nonanomalous $`R`$ symmetries which prove the dependence discussed above. Classically, there are three U(1) symmetries in the theory associated with the three fermion fields (gaugino, adjoint and fundamental fermions). In the quantum theory one can organize two nonanomalous combinations (a symmetry is nonanomalous if it does not transform the scale $`\mathrm{\Lambda }_1`$, associated with regulators).
The charges of the fields and parameters of the theory under these two U(1) symmetries are shown in Table 1.
The first of these symmetries U$`{}_{J}{}^{}(1)`$ is a subgroup of the global SU$`{}_{R}{}^{}(2)`$ group related to the $`𝒩=2`$ superalgebra . This explains the zero charge of the coupling $`h`$ with respect to this symmetry. The symmetry U$`{}_{J}{}^{}(1)`$ fixes the $`\mu `$ dependence of condensates. Namely, it is given by a power of $`\mu `$ equal to half the U$`{}_{J}{}^{}(1)`$ charge of the condensate. In particular, the field $`V`$ has U$`{}_{J}{}^{}(1)`$ charge equal to 2 which ensures that $`v\mu `$. Thus, we can use holomorphy to extend results to arbitrary values of $`\mu `$.
The second nonanomalous symmetry U$`{}_{R}{}^{}(1)`$ is similar to the $`R`$ symmetry of Ref. extended to include the adjoint field. As a consequence, for a given chiral field $`X`$
$$X=\mu ^{Q_J/2}m^{Q_R/4}\mathrm{\Lambda }_1^{d_X(Q_J/2)(Q_R/4)}f_X(\sigma ),$$
(11)
where $`Q_J`$, $`Q_R`$ are the U$`{}_{J}{}^{}(1)`$, U$`{}_{R}{}^{}(1)`$ charges of the field $`X`$, $`d_X`$ is its dimension, and $`f_X`$ is an arbitrary function of the dimensionless parameter $`\sigma `$ defined by Eq. (10). This parameter is neutral under both U(1)’s. The equation (8) is an example of the general relation (11) with $`Q_J=2`$, $`Q_R=2`$ and $`f_V=\kappa (\sigma )/2`$.
The important benefit of the consideration above is that in a theory with $`𝒩=2`$SUSY strongly broken by large $`\mu `$ and $`deth1`$ we can still relate chiral condensates with those in softly broken $`𝒩=2`$where $`deth=1`$ and $`\mu `$ is small.
Here is an example. When $`\sigma 0`$ two roots of Eq. (9) are $`\kappa _{1,2}=\pm 1`$ and the third one goes to infinity as $`\kappa _3=1/\sigma `$. For two finite roots one can suggest dual interpretations. Firstly, taking $`h=0`$, one can relate them to two vacua of SQCD in the Higgs phase. Second, for $`deth=1`$ (which is its $`𝒩=2`$value) one can make $`\sigma `$ small by taking the limit of large $`m`$. But this limit should bring us to the monopole and dyon vacua of softly broken $`𝒩=2`$SYM. The naming of vacua refers to the particle whose mass vanishes in the corresponding vacuum.
To verify this interesting mapping we need to determine the vev
$$u=U=\mathrm{Tr}\mathrm{\Phi }^2,$$
(12)
which can be accomplished using the set of Konishi anomalies. Generic equation for an arbitrary matter field $`Q`$ looks as follows (we are using the notation of the review ):
$$\frac{1}{4}\overline{D}^2J_Q=Q\frac{𝒲}{Q}+T(R)\frac{\mathrm{Tr}W^2}{8\pi ^2},$$
(13)
where $`T(R)`$ is the Casimir in the matter representation. The left hand side is a total derivative in superspace so its average over any supersymmetric vacuum vanishes. In our case this results in two relations for the condensates,
$`{\displaystyle \frac{m}{2}}Q_f^\alpha Q_\alpha ^f+{\displaystyle \frac{1}{\sqrt{2}}}h^{fg}Q_{\alpha f}\mathrm{\Phi }_\beta ^\alpha Q_g^\beta +{\displaystyle \frac{1}{2}}{\displaystyle \frac{\mathrm{Tr}W^2}{8\pi ^2}}=0`$
$`2\mu \mathrm{Tr}\mathrm{\Phi }^2+{\displaystyle \frac{1}{\sqrt{2}}}h^{fg}Q_{\alpha f}\mathrm{\Phi }_\beta ^\alpha Q_g^\beta +2{\displaystyle \frac{\mathrm{Tr}W^2}{8\pi ^2}}=0`$ (14)
From the first relation, after the substitution in (2) and comparing with Eq. (7), we find an expression for gluino condensate
$$s=\frac{\mathrm{Tr}\lambda ^2}{16\pi ^2}=\frac{\mathrm{Tr}W^2}{16\pi ^2}=\frac{\mu ^2\mathrm{\Lambda }_1^3}{4v}.$$
(15)
This is consistent with the general expression $`[T_GT(R)]\mathrm{Tr}\lambda ^2/16\pi ^2`$ for the nonperturbative ADS piece of the superpotential (3), see . Combining the two relations in (14) we can express the condensate $`u`$ in terms of $`v`$,
$$u=\frac{1}{2\mu }\left(mv+3s\right)=\frac{1}{2\mu }\left(mv+\frac{3}{4}\frac{\mu ^2\mathrm{\Lambda }_1^3}{v}\right)=\frac{\sqrt{m\mathrm{\Lambda }_1^3}}{4}\left(\kappa +\frac{3}{\kappa }\right).$$
(16)
Now we see that in the limit of large $`m`$ two vacua $`\kappa =\pm 1`$ are in perfect correspondence with $`u=\pm \mathrm{\Lambda }_0^2`$ for the monopole and dyon vacua of $`𝒩=2`$SYM. Indeed, $`\mathrm{\Lambda }_0^4=m\mathrm{\Lambda }_1^3`$ is the correct relation between the scale parameters of the theories.
For the third vacuum at large $`m`$ the value $`u=m^2/(deth)`$ corresponds on the Coulomb branch to the so called charge vacuum, where some fundamental fields become massless. Moreover, the correspondence with $`𝒩=2`$results can be demonstrated for the three vacua at any value of $`m`$. To this end we use the relation (16) and Eq. (9) to derive the following equation for $`u`$,
$$(deth)u^3m^2u^2\frac{9}{8}(deth)m\mathrm{\Lambda }_1^3u+m^3\mathrm{\Lambda }_1^3+\frac{27}{2^8}(deth)^2\mathrm{\Lambda }_1^3=0.$$
(17)
The three roots of this equation are the vevs of $`\mathrm{Tr}\mathrm{\Phi }^2`$ in the corresponding vacua.
How does this look from $`𝒩=2`$side? The Riemann surface governing the Seiberg-Witten solution is given by the curve
$$y^2=x^3ux^2+\frac{1}{4}\mathrm{\Lambda }_1^3mx\frac{1}{64}\mathrm{\Lambda }_1^6.$$
(18)
Singularities of the metric, i.e. the points in the u-plane where the discriminant of the curve vanishes are defined by two equations, $`y^2=0`$ and $`\mathrm{d}y^2/\mathrm{d}x=0`$,
$$x^3ux^2+\frac{1}{4}\mathrm{\Lambda }_1^3mx\frac{1}{64}\mathrm{\Lambda }_1^6=0,3x^22ux+\frac{1}{4}\mathrm{\Lambda }_1^3m=0,$$
(19)
which lead to
$$u^3m^2u^2\frac{9}{8}m\mathrm{\Lambda }_1^3u+m^3\mathrm{\Lambda }_1^3+\frac{27}{2^8}\mathrm{\Lambda }_1^3=0.$$
(20)
We see that this is a particular case of the $`𝒩=1`$equation (17) at $`deth=1`$.
Moreover, when $`deth`$ is not equal to its $`𝒩=2`$value $`(1)`$ Eq. (17) coincides with Eq. (20) after the rescaling
$$u=(deth)^{1/3}u^{},m=(deth)^{2/3}m^{},v=(deth)^{1/3}v^{}.$$
(21)
This is in agreement with the master parameter $`\sigma `$ which contains the product $`m^{3/2}deth`$ and the nonanomalous U(1) symmetries we discussed above. In other words, breaking of $`𝒩=2`$by Yukawa couplings does not influence consideration of the chiral condensates modulus the rescaling (21).
The consideration above shows that the only nonperturbative input needed to determine the chiral condensates is provided by the one-instanton ADS superpotential. This means that any reference to the $`𝒩=2`$limit is not crucial at all, i.e. in regard to these condensates the exact Seiberg-Witten solution of $`𝒩=2`$is equivalent to the ADS superpotential.
The relations for the condensates we have derived are not new, they were obtained in by the ‘integrating in’ procedure introduced in . Our approach which is based on ‘integrating out’, plus the Konishi relations, can be viewed as an independent proof of the ‘integrating in’ procedure.
What we see as an advantage of our approach it is that, within a certain range of parameters, the superpotential (3) gives a complete description of the low energy physics. Indeed, when the mass $`m_V`$ of the field $`V`$,
$$m_V=2m\left(23\sigma \kappa \right),$$
(22)
is much less than the other masses, such as $`m_\mathrm{\Phi }=g^2\mu `$ and $`m_W=|g^2v|^{1/2}`$, we are in the weakly coupled Higgs phase and enjoy full theoretical control. The Konishi relations help to determine the condensates of heavy fields in this phase. Holomorphy then allows for continuation of these results for the condensates to strong coupling.
At strong coupling the superpotential (3), like other versions of the Veneziano-Yankielowicz Lagrangians , does not describe the low energy physics. For example, it contains no light monopole degrees of freedom near the monopole vacuum point at small $`\mu `$. Moreover, there is no single local superpotential which could describe mutually nonlocal degrees of freedom which become light in different regions of the moduli space of the theory. At strong coupling the superpotential (3) is equivalent to the effective superpotential of the ‘integrating in’ procedure and can be viewed as a shorthand equation that gives the values of the condensates.
### 2.2 Matter and gaugino condensates in the limit of large mass
Here we summarize the results for matter, $`v=Q_f^\alpha Q_\alpha ^f/2`$, $`u=\mathrm{Tr}\mathrm{\Phi }^2`$, and gaugino, $`s=\mathrm{Tr}\lambda ^2/16\pi ^2`$, condensates in the limit where the parameter $`\sigma `$ defined by Eq. (10) is small. This can be achieved in the limit of large $`m`$ if the Yukawa coupling is fixed, or by taking $`deth`$ to be small otherwise. In the charge vacuum:
$`v_C={\displaystyle \frac{2\mu m}{(deth)}}\left(1+𝒪(\sigma ^2)\right),`$
$`u_C={\displaystyle \frac{m^2}{(deth)}}\left(1+𝒪(\sigma ^2)\right),`$
$`s_C={\displaystyle \frac{\mu \mathrm{\Lambda }_1^3(deth)}{8m}}\left(1+𝒪(\sigma ^2)\right).`$ (23)
In the monopole and dyon vacua:
$`v_{M,D}=\pm \mu \sqrt{{\displaystyle \frac{\mathrm{\Lambda }_1^3}{4m}}}\left(1+𝒪(\sigma )\right),`$
$`u_{M,D}=\pm \sqrt{\mathrm{\Lambda }_1^3m}\left(1+𝒪(\sigma )\right),`$
$`s_{M,D}=\pm {\displaystyle \frac{1}{2}}\mu \sqrt{\mathrm{\Lambda }_1^3m}\left(1+𝒪(\sigma )\right).`$ (24)
The upper sign refers to the monopole vacuum, while the lower one is for the dyon vacuum. As discussed above we can interpret these vacua also as the two vacua of the Higgs phase in SQCD. To this end we need to consider the limit of small $`deth`$ and $`m\mathrm{\Lambda }_{\mathrm{SQCD}}`$ with the identification
$$\mathrm{\Lambda }_{\mathrm{SQCD}}^5=\frac{1}{4}\mu ^2\mathrm{\Lambda }_1^3$$
(25)
### 2.3 Small mass limit
The limit of massless fundamentals $`m0`$ corresponds to $`\sigma \mathrm{}`$. In this limit the three vacua are related by a $`Z_3`$ symmetry ,
$`v={\displaystyle \frac{\mu \mathrm{\Lambda }_1}{(2deth)^{1/3}}}e^{2\pi ik/3}\left(1+𝒪(\sigma ^{2/3})\right),(k=0,\pm 1),`$
$`u={\displaystyle \frac{3}{8}}\mathrm{\Lambda }_1^2\left(2deth\right)^{1/3}e^{2\pi ik/3}\left(1+𝒪(\sigma ^{2/3})\right),(k=0,\pm 1),`$
$`s={\displaystyle \frac{1}{4}}\mu \mathrm{\Lambda }_1^2\left(2deth\right)^{1/3}e^{2\pi ik/3}\left(1+𝒪(\sigma ^{2/3})\right),(k=0,\pm 1)`$ (26)
Note that the massless limit exists due to the nonvanishing Yukawa coupling. When $`h0`$ we are back to the runaway vacua of massless SQCD.
### 2.4 Argyres-Douglas points
When the mass $`m`$ changes from large to small values we interpolate between the two quite different structures of vacua shown above. Let us consider this transition when, for definiteness, $`deth=1`$ and $`m`$ is real and positive and changes from large to small values. At large positive $`m`$ all the vacua are situated at real values of $`u`$, from Eqs. (23, 24) we see that the dyon vacuum is at negative $`u`$, the monopole vacuum is at positive $`u`$, and the charge vacuum is also at positive, but much larger, values of $`u`$. When $`m`$ diminishes then at some point the monopole and charge vacua collide on the real axis of $`u`$ and subsequently go more off to complex values producing the $`Z_3`$ picture at small $`m`$.
The point in the parameter manifold where the two vacua coincide is the AD point . In the SU(2) theory these points were studied in . Mutually non-local states, say charges and monopoles, becomes massless at these points. On the Coulomb branch of the $`𝒩=2`$theory these points correspond to a non-trivial conformal field theory . Here we study the $`𝒩=1`$SUSY theory, where $`𝒩=2`$is broken by the mass term for the adjoint matter as well as by the difference of the Yukawa coupling from its $`𝒩=2`$value. Collisions of two vacua still occur in this theory. In this subsection we find the values of $`m`$ at which AD points appear and calculate the values of the condensates at this point. In the next section we study what happens to the confinement of charges in the monopole point at non-zero $`\mu `$ once we approach the AD point.
First, let us work out the AD values of $`m`$, generalizing the consideration in . Coalescence of two roots for $`v`$ means that together with Eq. (7) the derivative of its left-hand-side should also vanish,
$$m\frac{(deth)}{2}\frac{v}{\mu }\frac{\mathrm{\Lambda }_1^3}{4}\left(\frac{\mu }{v}\right)^2=0,(deth)+\mathrm{\Lambda }_1^3\left(\frac{\mu }{v}\right)^3=0.$$
(27)
This system is consistent only at three values of $`m=m_{\mathrm{AD}}`$,
$$m_{\mathrm{AD}}=\frac{3}{4}\omega \mathrm{\Lambda }_1(deth)^{2/3},\omega =\mathrm{e}^{2\pi in/3}(n=0,\pm 1),$$
(28)
related by $`Z_3`$ symmetry. The condensates at the AD vacuum are
$`v_{\mathrm{AD}}=\omega {\displaystyle \frac{\mu \mathrm{\Lambda }_1}{(deth)^{1/3}}},`$
$`u_{\mathrm{AD}}=\omega ^1{\displaystyle \frac{3}{4}}\mathrm{\Lambda }_1^2(deth)^{1/3},`$
$`s_{\mathrm{AD}}=\omega ^1{\displaystyle \frac{1}{4}}\mu \mathrm{\Lambda }_1^2(deth)^{1/3}.`$ (29)
## 3 Dyon condensates
In this section we calculate various dyon condensates at the three vacua of the theory. As discussed above, holomorphy allows us to find these condensates starting from a consideration on the Coulomb branch in $`𝒩=2`$near the singularities associated with a given massless dyon. Namely, we calculate the monopole condensate near the monopole point, the charge condensate near the charge point and the dyon $`(n_m,n_e)=(1,1)`$ condensate near the point where this dyon is light. Although we start with small values of the adjoint mass parameter $`\mu `$, our results for condensates are exact for any $`\mu `$ as well as for any value of $`deth`$.
### 3.1 Monopole condensate.
Let us start with calculation of the monopole condensate near the monopole point. Near this point the effective low energy description of our theory can be given in terms of $`𝒩=2`$dual QED . It includes a light monopole hypermultiplet interacting with a vector (dual) photon multiplet in the same way as electric charges interact with ordinary photons. Following Seiberg and Witten we write down the effective superpotential in the following form,
$$W=\sqrt{2}\stackrel{~}{M}MA_D+\mu U,$$
(30)
where $`A_D`$ is a neutral chiral field (it is a part of the $`𝒩=2`$dual photon multiplet in the $`𝒩=2`$theory) and $`U=\mathrm{Tr}\mathrm{\Phi }^2`$ considered as a function of $`A_D`$. The second term breaks $`𝒩=2`$supersymmetry down to $`𝒩=1`$.
Varying this superpotential with respect to $`A_D`$, $`M`$ and $`\stackrel{~}{M}`$ we find that $`A_D=0`$, i.e. the monopole mass vanishes, and
$$\stackrel{~}{M}M=\frac{\mu }{\sqrt{2}}\frac{\mathrm{d}u}{\mathrm{d}a_D}|_{a_D=0}.$$
(31)
The condition $`A_D=0`$ means that the Coulomb branch near the monopole point, where the monopole mass vanishes, shrinks to a single vacuum state at the singularity while Eq. (31) determines the value of monopole condensate. Below we consider $`a_D`$ as a function of $`u`$. The value of $`u`$, $`u=u_M`$, at the monopole vacuum was determined in the previous section.
The non-zero value of the monopole condensate ensures U(1) confinement for charges via the formation of Abrikosov-Nielsen-Olesen vortices. Let us work out the r.h.s. of Eq. (31) to determine the $`\mu `$ and $`m`$ dependence of the monopole condensate. From the exact Seiberg-Witten solution , we have
$$\frac{\mathrm{d}a_D}{\mathrm{d}u}=\frac{\sqrt{2}}{8\pi }_\gamma \frac{\mathrm{d}x}{y(x)}.$$
(32)
Here for $`y(x)`$ given by Eq. (18) we use the form
$$y^2=(xe_0)(xe_{})(xe_+).$$
(33)
The integration contour $`\gamma `$ in the $`x`$ plane circles around two branch points $`e_+`$ and $`e_{}`$ of $`y(x)`$. At the monopole vacuum, when $`u=u_M`$, two branch points $`e_+`$ and $`e_{}`$ coincide, $`e_+=e_{}=e`$ and the integral (32) is given by the residue at $`x=e`$.
$$\frac{\mathrm{d}a_D}{\mathrm{d}u}(u_M)=\frac{i\sqrt{2}}{4\sqrt{ee_0}}.$$
(34)
The value of $`ee_0`$ (equal at $`u=u_M`$ to $`(1/2)\mathrm{d}^2(y^2)/\mathrm{d}x^2`$ ) is fixed by the equation $`\mathrm{d}(y^2)/\mathrm{d}x=0`$,
$$ee_0=\sqrt{u_M^2\frac{3}{4}m\mathrm{\Lambda }_1^3}.$$
(35)
Substituting this into the expression for the monopole condensate (31) we get finally
$$\stackrel{~}{M}M=2i\mu \left(u_M^2\frac{3}{4}m\mathrm{\Lambda }_1^3\right)^{1/4}.$$
(36)
To test the result let us consider first the limit of a large masses $`m`$ for the fundamental matter. As in Sec. 2.1 this limit can be viewed as a RG flow to pure Yang-Mills theory with the identification
$$\mathrm{\Lambda }_0^4=m\mathrm{\Lambda }_1^3,$$
(37)
where $`\mathrm{\Lambda }_0`$ is the scale of the $`𝒩=2`$Yang-Mills theory. In this limit we have $`u_M=\mathrm{\Lambda }_0^2`$. Then Eq. (36) gives
$$\stackrel{~}{M}M=\sqrt{2}i\mu \mathrm{\Lambda }_0,$$
(38)
which coincides with the Seiberg-Witten result . This ensures monopole condensation and charge confinement in the monopole point at large $`m`$.
Notice, that in the derivation above $`𝒩=2`$was not broken by the Yukawa coupling, i.e. we assume $`deth=1`$. The result, however, can be easily generalized to arbitrary $`deth`$ by means of U(1) symmetries considered above. The U$`{}_{R}{}^{}(1)`$ charge of $`\stackrel{~}{M}M`$ is equal to one. Indeed, the coefficient of the $`\sqrt{2}\stackrel{~}{M}MA_D`$ term in the superpotential (30) which is equal to one in the $`𝒩=2`$limit remains the same when $`𝒩=2`$is broken down to $`𝒩=1`$. It follows from U(1) symmetries together with decoupling of fundamental matter at large $`m`$. As a result we see that the Eq. (36) for the monopole condensate remains valid for arbitrary $`deth`$.
It is instructive to rewrite the result (36) for the monopole condensate in terms of $`v`$,
$$\stackrel{~}{M}M=i\left[(deth)v^2\frac{\mu ^3\mathrm{\Lambda }_1^3}{v}\right]^{1/2}=i\left[(deth)v^24\mu s\right]^{1/2},$$
(39)
where we also show a form which uses the gluino condensate $`s`$. It follows from the expression (16) for $`u`$ and Eq. (7) for $`v`$. It is interesting to observe that at $`m\mathrm{\Lambda }_1`$, when the value of $`v`$ is large and the nonperturbative term in (39) can be neglected, the monopole condensate reduces to that of the quark, $`\stackrel{~}{M}Miv\sqrt{deth}`$. Another interesting limit is the SQCD one when $`h0`$. In this limit the nonperturbative term in Eq. (39) dominates, and the square of the monopole condensate reduces to the gluino condensate.
Now let us address the question: what happens with the monopole condensate when we reduce $`m`$ and approach the AD point? The AD point corresponds to a particular value of $`m`$ which ensures coalescence of the monopole and charge singularities in the $`u`$ plane. Near the monopole point we have condensation of monopoles and confinement of charges while near the charge point we have condensation of charges and confinement of monopoles. As shown by ’t Hooft these two phenomena cannot happen simultaneously . The question is: what happens when monopole and charge points collide in the $`u`$ plane?
The monopole condensate at the AD point is given by Eq. (36). When $`m`$ and $`u`$ are substituted by $`m_{AD}`$ and $`u_{AD}`$ from Eqs. (28) and (29), we get
$$\stackrel{~}{M}M_{AD}=0.$$
(40)
We see that the monopole condensate goes to zero at the AD point. Our derivation makes it clear why it happens. At the AD point all three roots of $`y^2`$ become degenerate, $`e_+=e_{}=e_0`$, so the monopole condensate which is proportional to $`\sqrt{ee_0}`$ naturally vanishes.
In the next subsection we calculate the charge condensate in the charge point and show that it also goes to zero as $`m`$ approaches its AD value (28). Thus we interpret the AD point as a deconfinement point for both monopoles and charges.
### 3.2 Charge and dyon condensates
In this subsection we use the same method to calculate values for the charge and dyon condensates near the charge and dyon points respectively. We first consider $`m`$ above its AD value (28) and then continue our results to values of $`m`$ below $`m_{AD}`$. In particular, in the limit $`m=0`$ we recover $`Z_3`$ symmetry.
Let us start with the charge condensate. At $`\mu =0`$, $`deth=1`$ and large $`m`$ the effective theory near the charge point
$$a=\sqrt{2}m$$
(41)
on the Coulomb branch is $`𝒩=2`$QED. Here $`a`$ is the neutral scalar, the partner of photon in the $`𝒩=2`$supermultiplet. Half the degrees of freedom in color doublets become massless whereas the other half acquire large a mass $`2m`$. The massless fields form one hypermultiplet $`\stackrel{~}{Q}_+,Q_+`$ of charged particles in the effective electrodynamics. Once we add the mass term for the adjoint matter the effective superpotential near the charge point becomes
$$𝒲=\frac{1}{\sqrt{2}}\stackrel{~}{Q}_+Q_+A+m\stackrel{~}{Q}_+Q_++\mu U$$
(42)
Minimizing this superpotential we get condition (41) as well as
$$\stackrel{~}{Q}_+Q_+=\sqrt{2}\mu \frac{\mathrm{d}u}{\mathrm{d}a}|_{a=\sqrt{2}m}.$$
(43)
Now, following the same steps which led us from (31) to (36), we get
$$\sqrt{deth}\stackrel{~}{Q}_+Q_+=2\mu (u_C^2\frac{3}{4}m\mathrm{\Lambda }_1^3)^{1/4},$$
(44)
where we include a generalization to arbitrary $`deth`$. We choose to consider
$`\sqrt{deth}\stackrel{~}{Q}_+Q_+`$ because it has the U<sub>R</sub>(1) charge equal to one, similar to the $`\stackrel{~}{M}M`$ condensate considered above. By $`u_C`$ we denote the position of the charge vacuum in the $`u`$ plane. At large $`m`$ $`u_C=m^2/(deth)`$, see Eq. (23), and
$$\stackrel{~}{Q}_+Q_+=\frac{2\mu m}{(deth)}\left(1+𝒪(\sigma ^2)\right).$$
(45)
Holomorphy allows us to extend the result (44) to arbitrary $`m`$ and $`deth`$. So we can use Eq. (44) to find the charge condensate at the AD point. Using Eqs. (28) and (29) we see that the charge condensate vanishes at the AD point in the same manner the monopole condensate does. As it was mentioned we interpret this as deconfinement for both charges and monopoles.
As with the monopole condensate, we can also relate the charge condensate with the quark vev $`v`$,
$$\stackrel{~}{Q}_+Q_+=\left[v^2\frac{\mu ^3\mathrm{\Lambda }_1^3}{v(deth)}\right]^{1/2}=\left[v^2\frac{4\mu s}{(deth)}\right]^{1/2},$$
(46)
This expression differs from the one for the monopole condensate only by a phase factor. The coincidence of the charge condensate with the quark one at large $`v`$, i.e. at weak coupling, is natural. The difference is due to nonperturbative effects and is similar to the difference between $`a^2/2`$ and $`u`$ on the Coulomb branch of the $`𝒩=2`$theory. At strong coupling the difference is not small. In particular, the charge condensate vanishes at the AD point while the quark condensate remains finite.
Now let us work out the dyon condensate. More generally let us introduce the dyon field $`D_i`$, $`i=1,2,3`$, which stands for charge, monopole and $`(1,1)`$ dyon,
$$D_i=\{(deth)^{1/4}Q_+,M,D\}.$$
(47)
The arguments of the previous subsection which led us to the result (36) for monopole condensate gives for $`\stackrel{~}{D}_iD_i`$
$$\stackrel{~}{D}_iD_i=2i\zeta _i\mu \left(u_i^2\frac{3}{4}m\mathrm{\Lambda }_1^3\right)^{1/4},$$
(48)
where $`u_i`$ is the position of the i-th point in the $`u`$ plane and the $`\zeta _i`$ are phase factors.
For the monopole condensate at real values of $`m`$ larger than the $`m_{\mathrm{AD}}`$ Eq. (36) gives
$$\zeta _M=1,$$
(49)
while for the charge condensate from Eq. (44) we have
$$\zeta _C=i.$$
(50)
In fact one can fix the charge phase factor by imposing the condition that the charge condensate should approach the value $`2m\mu `$ in the large $`m`$ limit. For the dyon the phase factor is
$$\zeta _D=i.$$
(51)
At the particular AD point we have chosen the monopole and charge condensates vanish, while the dyon condensate remains non-zero, see (48). Below the AD point, condensates are still given by Eq. (48), but the charge and monopole phase factors can change <sup>2</sup><sup>2</sup>2Note that the quantum numbers of the “charge” and “monopole” are also transformed, see . The dyon phase factor (51) does not change when we move through the AD point because the dyon condensate does not vanish at this point.
In the limit $`m=0`$ we should recover the $`Z_3`$-symmetry for the values of condensates. From Eq. (48) it is clear that the absolute values of all three condensates are equal because the values of the three roots $`u_i`$ are on the circle in the $`u`$ plane, see (26). Imposing the requirement of $`Z_3`$ symmetry at $`m=0`$ we can fix the unknown phase factors $`\zeta _C`$ and $`\zeta _M`$ below the AD point using the value (51) for dyon. This gives
$$\zeta _C=i,\zeta _M=i.$$
(52)
### 3.3 Photino and gaugino condensates
The gaugino condensate $`\mathrm{Tr}\lambda ^2`$ we found in the previous section can be viewed as a sum of the condensates for charged gauginos and the photino,
$$\mathrm{Tr}\lambda ^2=\lambda ^+\lambda ^{}+\frac{1}{2}\lambda ^3\lambda ^3$$
(53)
In gauge invariant form the photino condensate can be associated with
$$(\mathrm{Tr}W\mathrm{\Phi })^2$$
(54)
We argue here that the photino condensate vanishes so that the gaugino condensate is solely due to the charged gluino.
Let us start with the Coulomb branch in the $`𝒩=2`$theory. All gaugino condensates vanish in $`𝒩=2`$for a simple reason: $`\lambda ^2`$ is not the lowest component in the corresponding $`𝒩=2`$supermultiplet. When the perturbation $`\mu U`$ which breaks $`𝒩=2`$is added to the superpotential the gaugino condensate is proportional to $`\mu `$. However, the term $`\mu U`$ in the superpotential does not break $`𝒩=2`$SUSY in the effective QED. Consider, for example, the monopole vacuum. The corresponding effective superpotential is given by Eq. (30), where in the expansion of $`U`$ as function of $`A_D`$ it is sufficient to retain only linear term. It was shown in that the perturbation linear in $`A_D`$ does not break $`𝒩=2`$in the effective QED. An immediate consequence of this observation is that the photino condensate continues to vanish.
## 4 The Argyres-Douglas point: how well is the theory defined
As discussed in the Introduction, at the AD point we encounter the problem of not having a uniquely defined vacuum state. Indeed, when the mass parameter $`m`$ approaches its AD value $`m_{AD}`$ we deal with two vacuum states which can be distinguished by values of the chiral condensates. It is unlikely that the number of states with zero energy will change when we reach the AD point, it is very similar to the Witten index. However, the continuity of the chiral condensates we obtained above shows that they are no longer parameters which differentiate the two states once we reach the AD point.
This does not prove the absence of a relevant order parameter so the quest can be continued. A natural possibility to consider is a domain walls interpolating between colliding vacua. In the case of BPS domain walls their tension is given by the central charge ,
$$T_{ab}=2|𝒲_{\mathrm{eff}}(v_a)𝒲_{\mathrm{eff}}(v_b)|$$
(55)
where $`a`$,$`b`$ label the colliding vacua. The central charge here is expressed via values of exact superpotential (3) in corresponding vacua. The continuity of the condensate $`v`$ shows that the domain wall becomes tensionless at the AD point, $`T(mm_{\mathrm{AD}})^{3/2}`$ when $`mm_{\mathrm{AD}}`$. If such a domain wall were observable at the AD point it could serve as a signal of two vacua.
We argue, however, that this domain wall is not observable in continuum limit. The crucial point is that the wall is built out of massless fields, therefore its thickness is infinite at the AD point. This makes it impossible to observe this tensionless wall in any physical experiment of a limited spatial scale.
In the conclusion of this section let us review briefly the brane construction of $`𝒩=1`$vacua. Gauge theories are realized on brane worldvolumes. Brane configurations responsible for $`𝒩=1`$theories were suggested in and a derivation of domain wall tensions from analysis of Riemann surfaces (which is similar to the calculation of the masses of BPS particles in $`𝒩=2`$theories) can be found in . The brane configuration for the $`𝒩=1`$theory with one flavor and SU(2) gauge group is described by Riemann surface embedded into three dimensional complex space $`C^3`$ parametrized by three variables $`t`$, $`v`$ and $`w`$. The embedding is given by the following equations
$`v+m={\displaystyle \frac{(ww_+)(ww_{})}{\mu w}},t=\mu ^2w(ww_+)(ww_{});`$ (56)
$`w_++{\displaystyle \frac{1}{2}}w_{}+m\mu =0,w_{}^2w_+=(\mu \mathrm{\Lambda }_1)^3.`$
with free parameters $`\mu ,m,\mathrm{\Lambda }`$. The tension of the walls, which have the interpretation of the M5 branes wrapping three-cycle with the boundaries on the Riemann surface above can be calculated by integrating the holomorphic three-form over this cycle
$$T=dv\mathrm{d}w\mathrm{d}(\mathrm{log}t)$$
(57)
Let us consider the geometry of the brane configuration near the AD point. It was shown recently that the AD point corresponds to a singular Calabi-Yau 3-manifold which is resolved if one adds particular perturbation. Since the tension is defined by integration of the holomorphic 3-form around the resolved singularity the tensionless wall has a geometrical interpretation as the M5 brane wrapping this vanishing cycle. Actually, the curves can be considered as fibered over the complex $`m`$ plane and the AD singularities correspond to the appearance of vanishing cycles in the fiber in a manner quite similar to the Seiberg-Witten solution of $`𝒩=2`$theories where vanishing cycles correspond to massless BPS particles.
## 5 Conclusions
The approach of this work is similar to that used in SQCD. Namely, we integrate out the adjoint field which leads, in some range of parameters, to an SQCD-like effective superpotential. This superpotential describes the low energy theory at weak coupling where we have full theoretical control. The nonperturbative part is given by the ADS superpotential generated at the one instanton level. The adjoint field shows up only as an extra (as compared with SQCD) nonrenormalizable term quartic in the fundamental fields.
Results for chiral condensates of matter and gaugino fields are continued into the range of a small adjoint mass where we find a complete matching with the $`𝒩=2`$Seiberg-Witten solution. The Argyres-Douglas points introduced in $`𝒩=2`$theories are shown to exists in the $`𝒩=1`$theory as well. Although the bulk of our results for matter and gaugino condensates overlaps with what is known in the literature we think that our approach clarifies some aspects of duality in $`𝒩=1`$theories.
We then analyze monopole, charge and dyon condensates departing from the Coulomb branch of the $`𝒩=2`$theory. This resulted in explicit relations between these condensates and those of the fundamental matter. The most interesting phenomenon occurs at the AD point: when the monopole and charge vacua collide both the monopole and charge condensates vanish. We interpret this as a deconfinement of electric and magnetic charges at the AD point. Vanishing of condensates signals that the theory at this point becomes superconformal.
In our approach we see straightforwardly that the one-instanton generated ADS superpotential is the only nonperturbative input needed to fix all chiral condensates. The general nature of this statement is seen from our derivation which relates polynomial coefficients in the Seiberg-Witten curve to the ADS superpotential.
Let us mention a relation to finite-dimensional integrable systems. It was recognized that $`𝒩=2`$theories are governed by finite-dimensional integrable systems. The integrable system responsible for $`𝒩=2`$SQCD was identified with the nonhomogenious XXX spin chain . After perturbation to the $`𝒩=1`$theory the Hamiltonian of the integrable system is expected to coincide with the superpotential of corresponding $`𝒩=1`$theory. This has been confirmed by direct calculation in the pure $`𝒩=2`$gauge theory as well in the theory with a massive adjoint multiplet . It would be very interesting to find a similar connection between spin chain Hamiltonians and superpotentials in the $`𝒩=1`$SQCD. One more point to be clarified is the meaning of the AD point within approach based on integrability. Since the quark mass is identified as a value of spin one might expect that at particular spin values corresponding to the AD mass, the XXX spin chain would have additional symmetries similar to superconformal ones. We hope to discuss these points in more details elsewhere.
In this paper we considered only the SU(2) theory with one flavor postponing the generic $`N_c`$, $`N_f`$ case for a separate publication. The most interesting problem in the generic situation involves Seiberg IR duality of the electric SU$`(N_c)`$ theory with $`N_f`$ flavors and the magnetic SU$`(N_fN_c)`$ theory. In generic case of nondegenerate fundamental masses we expect deconfinement at the AD points. A degeneracy in fundamental masses leads to the appearance of Higgs branches. The approach of the present paper can be applied to this case as well. However, since Higgs branches do not disappear at the AD points we do not expect deconfinement to occur in this case .
### Acknowledgments
Authors are grateful to P. Argyres, A. Hanany, K. Konishi, A. Marshakov, S. Rudaz, A. Ritz, and M. Shifman for helpful discussions. Part of this work was done when two of the authors, A.V. and A.Y., participated in the SUSY99 program organized by the Institute for Theoretical Physics at Santa Barbara. A.V. and A.Y. are thankful to ITP for hospitality and support from NSF by the grant PHY 94-07194. A.G. and A.Y. thank the Theoretical Physics Institute at the University of Minnesota where this work was initiated for support. A.G. thanks J. Ambjorn for hospitality at the Niels Bohr Institute where part of this work has been done.
The work of A.G. is supported in part by the grant INTAS-97-0103, A. V. is supported in part by DOE under the grant DE-FG02-94ER40823, and A.Y. is supported in part by Russian Foundation for Basic Research under the grant 99-02-16576.
|
warning/0004/hep-th0004203.html
|
ar5iv
|
text
|
# References
On Pure Lattice Chern-Simons Gauge Theories
F. Berruto<sup>(a)</sup>, M.C. Diamantini$`^{(b)^{}}`$ and P. Sodano<sup>(a)</sup>
a)Dipartimento di Fisica and Sezione I.N.F.N., Università di Perugia, Via A. Pascoli I-06123 Perugia, Italy
b)Department of Theoretical Physics, Oxford University, 1 Keble Rd, Oxford UK
DFUPG-19-00
Abstract
<sup>*</sup><sup>*</sup>footnotetext: Supported by a Swiss National Science Foundation fellowship.
We revisit the lattice formulation of the Abelian Chern-Simons model defined on an infinite Euclidean lattice. We point out that any gauge invariant, local and parity odd Abelian quadratic form exhibits, in addition to the zero eigenvalue associated with the gauge invariance and to the physical zero mode at $`\stackrel{}{p}=\stackrel{}{0}`$ due to translational invariance, a set of extra zero eigenvalues inside the Brillouin zone. For the Abelian Chern-Simons theory, which is linear in the derivative, this proliferation of zero modes is reminiscent of the Nielsen-Ninomiya no-go theorem for fermions. A gauge invariant, local and parity even term such as the Maxwell action leads to the elimination of the extra zeros by opening a gap with a mechanism similar to that leading to Wilson fermions on the lattice.
It is by now well known that in odd space-time dimensions, there is the possibility of adding a gauge invariant, topological Chern-Simons (CS) term to the gauge field action. The CS term breaks both the parity and time-reversal symmetries and, when coupled with a Maxwell or Yang-Mills term, leads to massive gauge excitations . For an Abelian model in three space-time dimensions, the pure CS Lagrangian is defined as
$$_{CS}=\frac{k}{2}A_\mu ϵ^{\mu \alpha \nu }_\alpha A_\nu ,$$
(1)
where $`k`$ is a dimensionless coupling constant.
The pure CS theory is a topological field theory . It is exactly solvable and it is used to compute topological invariants of three manifolds, the knot invariants for links embedded in three-manifolds . As a model for physical phenomena, being dominant at large distances, the CS action may be used as a low energy effective field theory for condensed matter systems such as the fractional quantum Hall effect or Josephson junction arrays . Of great interest is also the relationship of CS theories to conformal field theories in two dimensions .
While in the continuum the pure CS theory is exactly solvable, things are quite different on the lattice. As originally shown by Fröhlich and Marchetti , the kernel defining the CS action exhibits a set of zeroes which are not due to gauge invariance; thus, the theory is not integrable even after gauge fixing. The action (1) is of first order in the derivatives, and the appearence of extra zeros in its lattice formulation is reminescent of the “doubling” of fermions on the lattice . While, for fermion models, the doubling made for many years impossible to formulate chiral gauge theories on the lattice (for recent development on this subject see ), the extra zeroes of the CS action make only the definition of a parity odd theory on the lattice sick. In the following we revisit the lattice formulation of the Euclidean version of the Abelian CS model defined by the Lagrangian (1). Previous studies of pure CS theory on the lattice have been carried out using the Hamiltonian formalism in , by introducing a mixed CS action with two gauge fields with opposite parity or by means of two gauge fields living on the links of two dual lattices (thereby obtaining in both cases a parity even action) .
In this letter we shall show that, due to the Poincaré lemma recently proved on the lattice by Lüscher , the non-integrability of the CS kernel is a general feature of any gauge-invariant, local and parity odd gauge theory on a lattice. Moreover, again as a consequence of the Poincaré lemma, we shall show that the only gauge invariant regularization of the CS action may be obtained by adding to it a parity even term; of course, the most physical choice is the Maxwell term. Since the addition of a Maxwell term regularizes the CS action, the presence of the extra zeros in the CS action did not cause any problem in previous investigations of the Maxwell-CS action on the lattice .
We consider an infinite Euclidean cubic lattice with lattice spacing $`a`$, which we set to unity ($`a=1`$). We shall denote a lattice site by the vector $`\stackrel{}{x}`$ and a link between $`\stackrel{}{x}`$ and $`\stackrel{}{x}+\widehat{\mu }`$ ($`\mu =0,1,2`$) by $`(\stackrel{}{x},\widehat{\mu })`$. Forward and backward difference operators are given by $`d_\mu f(\stackrel{}{x})=f(\stackrel{}{x}+\widehat{\mu })f(\stackrel{}{x})=(S_\mu 1)f(\stackrel{}{x})`$ and $`\widehat{d}_\mu f(\stackrel{}{x})=f(\stackrel{}{x})f(\stackrel{}{x}\widehat{\mu })=(1S_\mu ^1)f(\stackrel{}{x})`$, where $`S_\mu f(\stackrel{}{x})=f(\stackrel{}{x}+\widehat{\mu })`$, $`S_\mu ^1f(\stackrel{}{x})=f(\stackrel{}{x}\widehat{\mu })`$ are the forward and backward shift operators respectively. Summation by parts on the lattice interchanges the forward and backward derivatives: $`_\stackrel{}{x}f(\stackrel{}{x})d_\mu g(\stackrel{}{x})=_\stackrel{}{x}\widehat{d}_\mu f(\stackrel{}{x})g(\stackrel{}{x})`$.
The lattice Fourier transformation of the gauge field $`A_\mu `$ is given by
$$A_\mu (\stackrel{}{x})=_{}\frac{d^3p}{(2\pi )^3}e^{i\stackrel{}{p}\stackrel{}{x}}e^{ip_\mu /2}A_\mu (\stackrel{}{p}).$$
(2)
Due to the phase factor $`e^{ip_\mu /2}`$, $`A_\mu (\stackrel{}{p})`$ is antiperiodic if $`p_\mu p_\mu +2\pi (2n+1)`$, with $`n`$ integer. The integration over momenta in eq.(2) is restricted to the Brillouin zone $`=\left\{p_\mu \right|\pi p_\mu \pi ,\mu =0,1,2\}`$. Under parity, which on an Euclidean cubic lattice corresponds to the simultaneous inversion of all three directions, $`A_\mu (\stackrel{}{x})A_\mu (\stackrel{}{x}\widehat{\mu })`$ and $`A_\mu (\stackrel{}{p})A_\mu (\stackrel{}{p})`$.
The CS action on an Euclidean lattice derived by Frölich and Marchetti is:
$$S=\underset{\stackrel{}{x}}{}A_\mu (\stackrel{}{x})\stackrel{~}{K}_{\mu \nu }(\stackrel{}{x}\stackrel{}{y})A_\nu (\stackrel{}{y}),$$
(3)
where $`\stackrel{~}{K}_{\mu \nu }=K_{\mu \nu }+\widehat{K}_{\mu \nu }`$, and
$`K_{\mu \nu }(\stackrel{}{x}\stackrel{}{y})`$ $`=`$ $`S_\mu ^\stackrel{}{y}ϵ_{\mu \alpha \nu }d_\alpha ^\stackrel{}{y}\delta _{\stackrel{}{x},\stackrel{}{y}},`$ (4)
$`\widehat{K}_{\mu \nu }(\stackrel{}{x}\stackrel{}{y})`$ $`=`$ $`S_\nu ^{1,\stackrel{}{y}}ϵ_{\mu \alpha \nu }\widehat{d}_\alpha ^\stackrel{}{y}\delta _{\stackrel{}{x},\stackrel{}{y}}.`$ (5)
$`K`$ and $`\widehat{K}`$ are exchanged by summation by parts.
Both $`K`$ and $`\widehat{K}`$ define a gauge invariant and parity odd kernel. In momentum space, apart from the zero due to gauge invariance, $`K`$ and $`\widehat{K}`$ have eigenvalues $`\lambda (p)=\widehat{\lambda }^{}(p)=\pm 2e^{i_{\mu =0}^2p_\mu /2}\sqrt{_{\mu =0}^2\mathrm{sin}^2p_\mu /2}`$ and exhibit no extra zeroes apart from the one at zero momentum, associated with translational invariance. The kernels (4) and (5), after gauge fixing, both define an integrable CS action. However, since (3) is a quadratic form in $`A_\mu (\stackrel{}{x})`$, the kernel should be symmetric under the simultaneous exchange of $`\mu \nu ,(\stackrel{}{x})(\stackrel{}{y})`$ (according to , we call this Bose symmetry); it is easy to see that only the linear combination $`\stackrel{~}{K}=K+\widehat{K}`$ respects this symmetry and thus provides an acceptable definition of the lattice CS action. This has far reaching consequences on the integrability of the action (3) since, in momentum space, the operator $`\stackrel{~}{K}(p)=K(p)+\widehat{K}(p)`$ has, apart from the zero mode associated with gauge invariance, eigenvalues given by
$$\stackrel{~}{\lambda }(p)=\pm 2\sqrt{1+\mathrm{cos}\underset{\mu =0}{\overset{2}{}}p_\mu }\sqrt{3\underset{\mu =0}{\overset{2}{}}\mathrm{cos}p_\mu }.$$
(6)
One gets $`\stackrel{~}{\lambda }=0`$ whenever $`\mathrm{cos}_{\mu =0}^2p_\mu =1`$, $`i.e.`$ when $`_{\mu =0}^2p_\mu =(2n+1)\pi `$, which defines planes of zeros with co-dimension 1 in the Brillouin zone: the CS action (3) is thus not integrable. The properties of $`K`$ and $`\widehat{K}`$ parallel the ones of the forward and backward derivatives, which in momentum space read $`d_\mu e^{ip_\mu /2}\widehat{p}_\mu `$ and $`\widehat{d}_\mu e^{ip_\mu /2}\widehat{p}_\mu `$ with $`\widehat{p}_\mu =2\mathrm{sin}p_\mu /2`$: they do not have extra zeroes inside the Brillouin zone, but their linear combination $`d+\widehat{d}2\mathrm{cos}(p_\mu /2)\widehat{p}_\mu `$ has zeros at the border of the Brillouin zone $`p_\mu =\pm \pi `$.
The appearance of the extra zeroes is not due to the specific form of the kernel in (3). In fact, let us consider an action given by
$$S=\underset{\stackrel{}{x},\stackrel{}{y}}{}A_\mu (\stackrel{}{x})G_{\mu \nu }(\stackrel{}{x}\stackrel{}{y})A_\nu (\stackrel{}{y}).$$
(7)
We shall now prove that, under the assumptions that
i) (7) is local on the lattice;
ii) (7) is gauge invariant: $`\widehat{d_\mu ^x}G_{\mu \nu }(\stackrel{}{x}\stackrel{}{y})=\widehat{d_\nu ^y}G_{\mu \nu }(\stackrel{}{x}\stackrel{}{y})=0`$;
iii) (7) is odd under parity;
(7) is not integrable.
The fact that $`S`$ is a quadratic form implies that $`G_{\mu \nu }(\stackrel{}{x}\stackrel{}{y})=G_{\nu \mu }(\stackrel{}{y}\stackrel{}{x})`$ (Bose symmetry). By Fourier transforming eq.(7), one gets
$$_{}\frac{d^3p}{(2\pi )^3}A_\mu (\stackrel{}{p})\stackrel{~}{G}_{\mu \nu }(\stackrel{}{p})A_\nu (\stackrel{}{p})$$
(8)
with $`\stackrel{~}{G}_{\mu \nu }(\stackrel{}{p})=e^{ip_\mu /2}G_{\mu \nu }(\stackrel{}{p})e^{ip_\nu /2}`$ (no sum over $`\mu `$ and $`\nu `$). Note that, in order for $`G_{\mu \nu }(\stackrel{}{p})`$ to be a periodic function of the momentum $`\stackrel{}{p}`$, $`\stackrel{~}{G}_{\mu \nu }(\stackrel{}{p})`$ must be antiperiodic in $`p_\mu p_\mu +2\pi (2n+1)`$ and $`p_\nu p_\nu +2\pi (2n+1)`$, with $`n`$ integer. Moreover, $`\stackrel{~}{G}_{\mu \nu }(\stackrel{}{p})`$ must be a periodic function of the component of the three-momentum different from $`\mu `$ and $`\nu `$, since no extra phase is present in the definition of $`\stackrel{~}{G}_{\mu \nu }(\stackrel{}{p})`$. These properties of periodicity and antiperiodicity are crucial in the proof of the non-integrability of (7).
The kernel $`\stackrel{~}{G}_{\mu \nu }(\stackrel{}{p})`$ is such that
$$\stackrel{~}{G}_{\mu \nu }(\stackrel{}{p})=\stackrel{~}{G}_{\nu \mu }(\stackrel{}{p})$$
(9)
due to the Bose symmetry. In momentum space the assumptions i)-iii) imply that $`\stackrel{~}{G}_{\mu \nu }(\stackrel{}{p})`$ is an analytic function (from locality) satisfying to
$$\widehat{p}_\mu \stackrel{~}{G}_{\mu \nu }(\stackrel{}{p})=0$$
(10)
from gauge invariance, and to
$$\stackrel{~}{G}_{\mu \nu }(\stackrel{}{p})=\stackrel{~}{G}_{\nu \mu }(\stackrel{}{p})$$
(11)
from parity oddness. As a consequence of (10), using the Poincaré lemma, one may write
$$\stackrel{~}{G}_{\mu \nu }(\stackrel{}{p})=ϵ_{\rho \mu \nu }\widehat{p}_\rho f(\stackrel{}{p}),$$
(12)
where $`f(\stackrel{}{p})`$ is an analytic function of $`\stackrel{}{p}`$. Since $`\stackrel{~}{G}_{\mu \nu }(\stackrel{}{p})`$ must be periodic in the component of the momentum different from $`\mu `$ and $`\nu `$, one has that $`f(\stackrel{}{p})`$ must be antiperiodic in $`p_\rho p_\rho +2\pi (2n+1)`$. Moreover, since all the dipendence of $`\stackrel{~}{G}_{\mu \nu }(\stackrel{}{p})`$ on $`p_\mu `$ and $`p_\nu `$ is carried by $`f(\stackrel{}{p})`$, the latter must also be antiperiodic with respect to $`p_\mu `$ and $`p_\nu `$: in order for the kernel $`G_{\mu \nu }(\stackrel{}{p})`$ to be a periodic function of the momenta, $`f(\stackrel{}{p})`$ must satisfy to
$$f(p_0+2\pi (2n+1),p_1+2\pi (2n+1),p_2+2\pi (2n+1))=f(p_0,p_1,p_2).$$
(13)
Due to eqs.(11,12) one has that $`f(\stackrel{}{p})=f(\stackrel{}{p})`$. From eq.(13), one may easily check that for $`p_0=p_1=p_2=\pm \pi `$ one gets
$$f(\pm \pi ,\pm \pi ,\pm \pi )=f(\pm \pi ,\pm \pi ,\pm \pi )=0$$
(14)
(this is also true when any two component of the momentum are equal to zero and one is equal to $`\pm \pi `$). Since the spectrum of $`G_{\mu \nu }(\stackrel{}{p})`$ is given by $`G(\stackrel{}{p})=\pm |f(\stackrel{}{p})|\sqrt{_{\mu =0}^2\widehat{p}_\mu ^2}`$ (apart from the zero due to gauge invariance), eq.(14) implies that the kernel $`G(\stackrel{}{p})`$ exhibits extra zeroes at the edges of the Brillouin zone and is thus not integrable. Note that the presence of the extra zeros is completely independent on the nature, complex or real, of the function $`f(\stackrel{}{p})`$. This observation is pertinent since, in Euclidean space-time, the pure CS action is purely immaginary.
Relaxing the assumption iii) one may study the general form of a gauge invariant local action in three dimensions. With the help of the Poincaré lemma it is easy to show that the kernel $`\stackrel{~}{G}_{\mu \nu }(\stackrel{}{p})`$ can be divided into the sum of parity even and parity odd terms. Since, due to locality, $`\stackrel{~}{G}_{\mu \nu }(\stackrel{}{p})`$ is an analytic function of $`\stackrel{}{p}`$, it may be expanded in Taylor series: all the terms having even power of the momenta are parity even, while the terms with odd power of the momenta are parity odd. The terms with the lowest number of derivatives in this expansion are the CS term defined in and the Maxwell term, whose kernel on the lattice is:
$$M_{\mu \nu }=\mathrm{}\delta _{\mu \nu }+d_\mu \widehat{d}_\nu =K_{\mu \rho }\widehat{K}_{\rho \nu },$$
(15)
where $`\mathrm{}=_{\mu =0}^2d_\mu \widehat{d}_\mu `$ is the Laplacean in three dimensions. Since all the parity odd terms fullfill the assumptions of the theorem they generate extra zeroes in the spectrum. The only gauge invariant way to regularize the CS action is then the addition of a parity even term such as the Maxwell term.
For the Maxwell-CS theory on the lattice the kernel $`\mathrm{\Gamma }_{\mu \nu }`$ may be written as
$$\mathrm{\Gamma }_{\mu \nu }=\frac{1}{4e^2}M_{\mu \nu }+ikG_{\mu \nu }.$$
(16)
In (16) $`k`$ is dimensionless and $`e^2`$ has the dimension of a mass; the Maxwell term is an irrelevant operator and the CS action dominates in the infrared region. The Fourier transform of $`\mathrm{\Gamma }_{\mu \nu }`$, apart from a zero mode due to gauge invariance, has eigenvalues given by
$$\lambda _{MCS}(\stackrel{}{p})=\frac{1}{2e^2}\underset{\mu =0}{\overset{2}{}}(1\mathrm{cos}p_\mu )+ikG(\stackrel{}{p}),$$
(17)
and, as it stands, it is free from extra zeroes in the Brillouin zone since the first term in (17), which is the Fourier transform of the Maxwell kernel, is zero only at zero momentum, and at the corners of the Brillouin zone, $`p_\mu =\pm \pi ,\mu =0,1,2`$, takes the value $`\lambda _{MCS}(\stackrel{}{p})=3/e^2`$.
Since the CS action is purely immaginary, the addition of the Maxwell term is used also in the continuum theory to provide a proper definition of the functional integral in the partition function of the pure CS theory. The CS limit is reached also there by taking the limit $`e^2\mathrm{}`$ after Gaussian integration.
The regularization of the extra zeros in the CS action by adding a Maxwell term and thereby opening a gap in the fermion spectrum is similar to the mechanism of the Wilson fermion where a gap is opened and the energy does not have secondary minima at the non-zero corners of the Brillouin zone. As in the case of the Wilson fermions , the regularization is done by means of an irrelevant operator and the continuum limit $`a0`$ is not changed by this addition. Moreover, as the Wilson action explicitly breaks chiral symmetry, the action obtained after the addition of the Maxwell term is not anymore defined under parity.
A key result of our analysis is a no-go theorem in the lattice regularization of the pure CS theory, if one requires locality, gauge invariance and parity oddness on the lattice. Clearly all the arguments needed for the proof rely on the definition of the parity transformation for the gauge field, i.e. on $`A_\mu (\stackrel{}{x})A_\mu (\stackrel{}{x}\widehat{\mu })`$. One possible way out is to define a new parity transformation under which the CS action is still odd but the kernel is integrable. The new definition of the parity transformation should then play a role analogous to the Ginsparg-Wilson relation in the definition of chiral gauge theories on the lattice. When the Ginsparg-Wilson relation holds, the lattice fermion action has an exact chiral symmetry and the no-go theorem of Nielsen and Ninomiya is avoided. It is pertinent to point out that, while for chiral gauge theories the problem of the non-perturbative regularization of the theory in a way preserving the symmetry is related to the presence, at the quantum level, of anomalies, for the CS action the theory in the continuum is well defined and solvable. This hints to the fact that the non-integrability on the lattice of the CS action is only a lattice artifact, due mainly to the existence of two derivatives giving rise to the two kernels $`K`$ and $`\widehat{K}`$.
Acknoledgments We are particularly in debt with M. Lüscher for many helpful discussions. We also greatly benefited from conversation on topics relevant to the subject of this paper with G. Grignani, G.W. Semenoff and C. A. Trugenberger. M.C.D. and F.B. thank the Theory Division of C.E.R.N., where part of this work was performed, for the warm hospitality. This work was partially supported by grants from M.U.R.S.T. and I.N.F.N..
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.