Ferromagnetic phase transition and Bose-Einstein condensation in spinor Bose gases
Extracted source text ยท arXiv cond-mat/0307063
\documentclass[prl,twocolumn,showpacs,preprintnumbers,amsmath,amssymb]{revtex4}
%\documentclass[twocolumn,showpacs,amsmath,amssymb]{revtex4}
\usepackage{graphicx}% Include figure files
\usepackage{dcolumn}% Align table columns on decimal point
\usepackage{bm}% bold math
\begin{document}
\input{epsf}
\preprint{}
\title{Ferromagnetic phase transition and Bose-Einstein condensation in spinor
Bose gases}
\author{Qiang Gu and Richard A. Klemm}
\affiliation{Max-Planck-Institut f\"{u}r Physik komplexer Systeme,
N\"{o}thnitzer Stra{\ss}e 38, 01187 Dresden, Germany}
\date{\today}
\begin{abstract}
Phase transitions in spinor Bose gases with ferromagnetic (FM) couplings are
studied via mean-field theory. We show that an infinitesimal value of the
coupling can induce a FM phase transition at a finite temperature always above
the critical temperature of Bose-Einstein condensation. This contrasts sharply
with the case of Fermi gases, in which the Stoner coupling $I_s$ can not lead
to a FM phase transition unless it is larger than a threshold value $I_0$. The
FM coupling also increases the critical temperatures of both the ferromagnetic
transition and the Bose-Einstein condensation.
\end{abstract}
\pacs{05.30.Jp, 03.75.Mn, 75.10.Lp, 75.30.Kz}
\maketitle
The magnetism of Fermi (electron) gases has long been a research topic in solid
state physics. Although many open questions remain, the magnetic properties of
Fermi gases have been well understood\cite{mohn}. The Fermi surface plays an
important role in determining their magnetism. For example, a magnetization
density ${\overline M}$ in Fermi gases increases the band energy by splitting
the Fermi surfaces for spin-up and spin-down particles. As a result, Fermi
gases mainly behave as Pauli paramagnets in the absence of an exchange
interaction. If an effective ferromagnetic (FM) exchange $I_s$ is present,
electron gases can exhibit ferromagnetism. Within the framework of the Stoner
theory, $I_s$ results in a negative molecular field energy when ${\overline M}$
is finite. If $I_s>I_0$, the Stoner threshold, the value of the molecular field
energy becomes larger than that of the increase of the band energy induced by
${\overline M}$, and then a FM ground state is energetically favored.
The magnetism of Bose gases has been less studied. But since the
realization of Bose-Einstein condensation (BEC) in ultracold atomic
gases\cite{cornell}, more and more attention has been attracted to this topic,
because the constituent atoms, such as $^{87}$Rb, $^{23}$Na, and $^7$Li
usually have (hyperfine) spin degrees of freedom and thus a magnetic moment
$m$. $m$ can be large in atoms such as $^{52}$Cr\cite{weinstein}, for which
$m=6\mu_B$, where $\mu_B$ is the Bohr magneton. Since 1998, atomic
gases can be confined and cooled in purely optical traps in which their spin
degrees of freedom remain active, and therefore investigating their magnetic
properties becomes experimentally possible\cite{ketterle}.
Ferromagnetism in spinor bosons without any spin-dependent interactions has
already been theoretically studied. As opposed to the case of fermions, the
ground states of spinor bosons are degenerate and a ferromagnetic state is
among the ground states\cite{suto,yang}. Furthermore, the spinor Bose gas is
rather apt to be magnetized by an external magnetic field even at a finite
temperature $T$, as long as $T<T_c$, the BEC critical
temperature\cite{yamada,caramico,Simkin,gu}.
In ultracold atomic gases, a Heisenberg-like exchange interaction is usually
present, arising from spin-flip scattering processes. For example, the
effective two-body interaction in spin-$1$ atoms was given by\cite{ho},
\begin{eqnarray}
V = ~:\int d^3{\bf r}
\left[ \frac {c_0}2 n^2({\bf r})
+ \frac {c_s}2 {\bf S}({\bf r})\cdot {\bf S}({\bf r}) \right]: ,
\end{eqnarray}
where $n({\bf r})=\sum_{\sigma}\psi_{\sigma}^\dag ({\bf r})
\psi_{\sigma}({\bf r})$,
$S^\alpha ({\bf r}) = \sum_{\sigma\sigma^{\prime}}
\psi_{\sigma}^\dag ({\bf r}) S^\alpha_{\sigma\sigma^{\prime}}
\psi_{\sigma^{\prime}}({\bf r})$ with $S^\alpha$
($\alpha=x,y,z$) being $3\times 3$ spin matrices, $\psi_\sigma({\bf r})$ is a
field annihilation operator for an
atom in the hyperfine state $|F=1,\sigma\rangle$ at point ${\bf r}$, and
$\sigma=1,0,-1$. The normal ordering removes the spin-spin interaction of a
particle with itself. In the special case of $^{87}$Rb, the exchange is
ferromagnetic ($c_s<0$)\cite{burke,pu}.
Ferromagnetic couplings can also be generated by ``superexchange" processes in
correlated spinor boson systems described by the boson Hubbard model. Recently,
this model was used to describe Bose gases moving in a periodic optical
lattice\cite{greiner}, and a lot of work has been done to discuss its
properties\cite{fisher}. It has been shown that the spinor boson Hubbard model
can be reduced to the {\sl ferromagnetic} Heisenberg model, provided the
on-site repulsion is large enough\cite{yang},
\begin{eqnarray}
H = - I_H \sum_{{\langle ij \rangle}}{\bf S}_i\cdot {\bf S}_j ,
\end{eqnarray}
where $I_H$ is the Heisenberg exchange constant.
The presence of FM couplings enriches the properties
of spinor bosons significantly. One can naturally imagine that a FM phase
transition could be induced, at a temperature plausibly depending on the
energy scale of the coupling. Thus, an interesting question arises as to how
the FM transition interplays with the BEC. We focus on this question in this
paper. Without loss of generality, we study a {\sl homogeneous} spinor Bose gas
before considering trapped gases. Our theoretical results indicate that the
FM phase transition takes place at a finite temperature $T_F$ {\sl always
above $T_c$}, regardless of the size of the coupling. This contrasts sharply
with the case of fermions. Figure 1 shows schematically the
relation between $T_F$ and $I_s$ for both Bose and Fermi gases. More recently,
Isoshima {\it et al.}\cite{isoshima} and Huang {\it et al.}\cite{huang2}
studied BEC in trapped $F=1$ spinor bosons with FM couplings, but the FM
transition was not considered.
\begin{figure}
\center{\epsfxsize=65mm \epsfysize=40mm \epsfbox{fig1.eps}}
\caption{
\label{fig:epsart}
Schematic relations between transition temperatures and FM couplings $I_s$.
$T^b_F$ and $T^f_F$ represent the FM transition temperature for Bose and Fermi
gases respectively. $T_c$ and $T_0$ denote the BEC critical temperature for
spinor bosons with and without couplings respectively. $I_0$ is the Stoner
threshold.}
\end{figure}
{\sl The model.}--
Following the Stoner theory for fermion gases\cite{mohn}, a molecular field
$H_m = I_s {\overline M}$ is introduced to describe the effective FM coupling
between bosons, and the energy shift arising from the molecular field is
\begin{eqnarray}
\epsilon^m_i = - H_m S^z_i = - I_s {\overline M} S^z_i .
\end{eqnarray}
The molecular field contains all the spin-dependent interactions. It can be
derived from the exchange interactions via the mean-field approximation.
In order to avoid the spin-spin interaction of a particle with itself, we treat
the particles as being on some kind of ``lattice",
\begin{eqnarray}
-\sum_{\langle ij \rangle}{\bf S}_i\cdot {\bf S}_j &\approx&
- \sum_{\langle ij \rangle} ( \langle{\bf S}_i\rangle \cdot {\bf S}_j
+ {\bf S}_i \cdot \langle{\bf S}_j\rangle
-\langle{\bf S}_i\rangle \cdot \langle{\bf S}_j\rangle ) \nonumber\\
&=&- Z {\overline M}\sum_{i} S^z_i + \frac 12 Z{\cal N} {\overline M}^2 ,
\end{eqnarray}
where $Z$ is the effective ``coordination number", which for a gas is an
irrelevant dimensionless parameter of order unity, and ${\cal N}$ is the number
of sites. Therefore $I_s=Z I_H$ for the Hamiltonian in Eq. (2).
${\bf {\overline M}}_i=\langle{\bf S}_i\rangle$ serves as the ferromagnetic
order parameter. We investigate a spinor Bose gas with hyperfine
spin-$F$ and set the boson magnetic moment to unity. It is convenient to choose
$\langle{\bf S}_i\rangle=(0,0,{\overline M}_i)$ and
${\overline M}_i=\langle{S_i^z}\rangle
= \sum_{\sigma}\sigma \langle \psi_{i\sigma}^\dag \psi_{i\sigma} \rangle$ where
$\psi_{i\sigma}$ annihilates a boson with spin quantum number $\sigma$ at site
$i$. We take ${\overline M}_i={\overline M}$ for a homogeneous boson gas.
Our main purpose is to show how the FM coupling affects the
properties of spinor bosons. So we neglect spin-independent interactions for
simplicity. Then the effective Hamiltonian for the grand canonical ensemble
reads
\begin{eqnarray}
H-N\mu= \sum_{{\bf k}\sigma}
[ \epsilon_{\bf k} - \mu - \sigma (H_m + H_e) ] n_{{\bf k}\sigma} ,
\end{eqnarray}
where $\epsilon_{\bf k}=\hbar^2k^2/2m^*$ is the kinetic energy of free
particles with mass $m^*$, $\mu$ is the chemical potential, and $H_e$ is the
external magnetic field. Since the Hamiltonian is diagonal, we may calculate the
grand thermodynamical potential
\begin{eqnarray}
\Omega &=& -\frac 1{\beta}{\rm ln} {\cal Z}
= -\frac 1{\beta} {\rm ln} {\rm Tr}{ e^{-\beta(H-N\mu)} } ,
\end{eqnarray}
where ${\cal Z} = {\rm Tr} \{ {\rm exp}[-\beta(H-N\mu)] \}$ is the partition
function, and $\beta=1/(k_BT)$. The density of particles is
\begin{eqnarray}
{\overline n}
= - \frac 1V \left( \frac {\partial\Omega}{\partial \mu} \right)_{T,V}
= \frac 1V\sum_{{\bf k}\sigma}
\langle n_{{\bf k}\sigma} \rangle ,
\end{eqnarray}
where $V$ is the volume of the system, $N$ is the total number of particles,
and ${\overline n}=N/V$. ${\overline M}$ is determined self-consistently by
\begin{eqnarray}
{\overline M}
= -\frac 1V \left( \frac {\partial\Omega}{\partial H_e} \right)_{T,V}
= \frac 1V\sum_{{\bf k}\sigma}
\sigma \langle n_{{\bf k}\sigma} \rangle .
\end{eqnarray}
The external magnetic field $H_e$ is set to zero in the following calculations.
We now consider only $F=1$ bosons. Eqs. (7) and (8) lead to the basic equations
determining the phase diagram of the spin-$1$ boson system,
\begin{subequations}
\begin{eqnarray}
1 &=& n_0 + t^{\frac 32} \left[
f_{\frac 32}(a) + f_{\frac 32}(a+b) + f_{\frac 32}(a+2b) \right] , \\
M &=& n_0 +
t^{\frac 32}\left[ f_{\frac 32}(a) - f_{\frac 32}(a+2b) \right] ,
\end{eqnarray}
\end{subequations}
where $a=-(\mu+H_m)/(k_B T)$, $b=H_m/(k_B T)$, the reduced temperature and
coupling are given by $t={k_B T m^*}/{( 2\pi\hbar^2{\overline n}^{2/3} )}$
and $I={I_s {\overline n}^{1/3} m^*}/{( 2\pi\hbar^2 )}$, respectively,
${\overline n}_0$ is the condensate density,
$n_0={\overline n}_0/{\overline n}$ is the condensate fraction,
$M={\overline M}/{\overline n}$ is the normalized magnetization, and
$f_{s}(a)$ is the polylogarithm function defined as\cite{robinson}
\begin{eqnarray}
f_{s}(a) \equiv {\rm Li}_s( e^{-a})
= \sum_{p=1}^{\infty}\frac {\left(e^{-a}\right)^p}{p^s} .
\end{eqnarray}
We note that $f_s(0)=\zeta(s)$, the Riemann zeta function.
Eq. (9a) is the standard formula for the density of bosons, by which the
reduced BEC critical temperature $t_c$ can be determined\cite{huang}. $a>0$ for
$t>t_c$ and $a\to 0$ as $t\to t_c$ from above. ${\overline n}_0>0$ and $a=0$
for $t<t_c$. In deriving Eq. (9b), we assume that only the spin-$1$ component
of the bosons can condense. This assumption is discussed in the following.
\begin{figure}
\center{\epsfxsize=65mm \epsfysize=50mm \epsfbox{fig2.eps}}
\caption{
\label{fig:epsart}
Reduced temperature $t$ vs. reduced FM coupling $I$ phase diagram of spin-1
Bose gases. The paramagnetic normal phase (PM), ferromagnetic normal phase
(FM), and coexisting phase of ferromagnetism and Bose condensation (FM+BEC) are
indicated. The solid ($t_c$) and dotted ($t_F$) curves are results from
Eqs. (15) and (11), respectively, and diamonds ($t_c^*$) and circles ($t_F^*$)
are obtained from Eqs. (9). The insets show the differences between $t_c^*$
and $t_c$, and between $t_F^*$ and $t_F$, respectively, versus $I$.}
\end{figure}
{\sl Preliminary analyses.}--
Assuming a FM phase transition is induced by $I$ below the reduced transition
temperature $t_F$, it is of special interest to determine the relation between
$t_F$ and $I$. We first suppose $I$ is very large so that $t_F>t_c$. Provided
that the FM transition is {\sl continuous}, i.e., $b\to 0$ with $t \to t_F$,
Eqs. (9) become
\begin{subequations}
\begin{eqnarray}
1 &=& 3 t_F^{\frac 32} f_{\frac 32}(a_F) ,\\
1 &=& 2 I t_F^{\frac 12} f_{\frac 12}(a_F) .
\end{eqnarray}
\end{subequations}
where $a_F=a(t_F)$. Equations (11) define a relation between $t_F$ and $I$. For
a given $t_F$, $I$ is given by
\begin{eqnarray}
I = \frac { \left[ 3f_{\frac 32}(a_F) \right]^{\frac 13} }
{2 f_{\frac 12}(a_F) } .
\end{eqnarray}
$I$ is a monotically decreasing function of $a_F$. As $a_F\to 0$,
$f_{\frac 32}(a_F)\to \zeta(3/2)\approx 2.612$, and
$f_{\frac 12}(a_F)\approx \sqrt{\pi/a_F}$\cite{robinson}. So for small values
of $I$ and $a_F$ we have
\begin{eqnarray}
a_F \approx {4\pi} t_0 I^2 ,
\end{eqnarray}
where $t_0=1/[3\zeta(3/2)]^{2/3}$ is the reduced BEC critical temperature for
the Bose gas with $I=0$. Since $a_F$ is always larger than zero as long as $I$
is finite, Eq. (13) implies that {\sl an infinitesimal FM coupling can induce a
FM phase transition at a finite temperature above the BEC critical
temperature}, because $a_F$ is finite only for $t>t_c$.
We now calculate the asymptotic behavior of the relation between $t_F$ and
$I$ for small couplings: $I\ll 1$. From Eqs. (11) we find\cite{robinson}
\begin{eqnarray}
\frac {t_F}{t_0} \approx 1 + 8\pi t^2_0 I .
\end{eqnarray}
This equation shows that {\sl the FM transition temperature increases with
the FM coupling}.
By decreasing $t$ further, the Bose gas will undergo BEC. Due to the molecular
field, the spin-1 bosons have the lowest free energy. So the condensate
contains only the spin-1 component and the other two boson spin components
remain in their excited states. Since ${\overline n}_0=0$ at the BEC critical
temperature, the self-consistent Eqs. (9) reduce to
\begin{subequations}
\begin{eqnarray}
1 &=& t_c^{\frac 32} \left[ \zeta(\frac 32)
+ f_{\frac 32}(b_c) + f_{\frac 32}(2b_c) \right] , \\
\frac {b_c} I &=&
t_c^{\frac 12}\left[ \zeta(\frac 32) - f_{\frac 32}(2b_c) \right] ,
\end{eqnarray}
\end{subequations}
where $b_c=b(t_c)$. These equations define the relation between $t_c$ and $I$.
Since the FM phase transition occurs above $t_c$, the system has a finite
magnetization at $t_c$. When $I\ll 1$ and $b_c$ is very small, Eq. (15a)
leads to
\begin{eqnarray}
\frac {t_c}{t_0} \approx
1 + \frac {4(1+\sqrt{2})\sqrt{\pi b_c}}{9\zeta(\frac32)} .
\end{eqnarray}
Equation (16) is similar to the expression of the BEC critical temperature in
an external magnetic field\cite{Simkin}, but here $b_c\approx 8\pi t_0I^2$ is
not the external field, but is a self-consistently determined quantity
proportional to the spontaneous magnetization.
Substituting $b_c$ into Eq. (16), we find
\begin{eqnarray}
\frac {t_c}{t_0} \approx
1 + \frac 83 (2+\sqrt{2})\pi t^2_0 I .
\end{eqnarray}
This equation indicates that {\sl the FM coupling also increases the BEC
critical temperature}.
\begin{figure}
\center{\epsfxsize=65mm \epsfysize=45mm \epsfbox{fig3.eps}}
\caption{
\label{fig:epsart}
Plots of the normalized magnetization $M_c$ at reduced BEC critical temperature
$t_c$ vs. reduced FM coupling $I$. The solid curve denotes $M_c$ at $t_c$
obtained from Eqs. (15), and the diamonds represent $M_c$ at $t_c^*$ obtained
from Eqs. (9). Inset: plots of $t_F^*-t_c$ (inverted triangles), and $t_F-t_c$
(solid curve). The dotted curve is a guide to the eye.}
\end{figure}
\begin{figure}
\center{\epsfxsize=65mm \epsfysize=70mm \epsfbox{fig4.eps}}
\caption{
\label{fig:epsart}
Order parameters of FM phase ($M$) and Bose condensed phase ($n_0$) versus the
reduced temperature $t$ for the reduced FM couplings $I=0.5,0.4,0.3$, from top
to bottom. The diamonds mark the points where BEC takes place. At $t_F^*$
(circle), $M$ drops discontinuously from $M_F=M(t_F^*)$ to zero.}
\end{figure}
{\sl Numerical results.}--
Comparing Eqs. (14) and (17), we see that $t_F$ is slightly smaller than $t_c$
for small FM couplings. This is unphysical and might be attributed to
the assumption that the FM transition is continuous. To investigate this point, we
solved the self-consistent Eqs. (9) numerically. The results obtained
indicate that both the FM transition and the BEC become discontinuous if the FM
coupling is sufficiently small.
Fig. 2 shows the phase diagram of spin-1 Bose gases.
We first discuss the Bose-Einstein condensation. For $I\gtrsim I_c\approx 0.2$,
$t_c=t_c^*$, as shown in the left inset, and both $a$ and ${\overline n}_0$
are exactly determined to be
zero at $t_c$, while $M$ has a finite value: $M(t_c)=M_c$. In this case, the
results from Eqs. (15) agree perfectly with those obtained from Eqs. (9). With
increasing $I$, $t_c$($t^*_c$) tends to the upper limit value
$1/[\zeta(3/2)]^{2/3}$ which is the reduced BEC critical temperature of scalar
bosons. Nevertheless, when $I<I_c$, the BEC becomes a discontinuous phase
transition: ${\overline n}_0$ has a small but nonzero value at $t^*_c$. In this
case, $t^*_c$ is slightly larger than $t_c$, as the left inset shows. The fact
that the Bose gas has a finite magnetization at
$t_c$($t^*_c$) implies that a FM transition has taken place at a higher
temperature as stated above. Fig. 3 shows the normalized magnetization $M_c$ at
$t_c$($t^*_c$) versus the reduced FM coupling $I$.
Fig. 4 plots the order parameters of both phases. The normalized magnetization
$M$ drops quickly with $t$ increasing above $t_c$($t^*_c$), and tends to zero
at the reduced FM transition temperature $t^*_F$. $t^*_F=t_F$ when
$I \gtrsim I_F\approx 0.35$, as shown in the right inset of Fig. 2. The smaller
$I$ is, the more quickly the magnetization drops. When $I < I_F$, $M$ does not
disappear continuously but drops from a finite value $M_F=M(t^*_F)\neq 0$ to
zero abruptly at $t^*_F$, with $t^*_F>t_F$ as shown in the right inset of
Fig. 2. This signals that the FM transition becomes first order.
$M_F\approx 0.11$ for $I=0.3$. In this case Eqs. (11) no longer hold. We
determine $t^*_F$ for the coupling down to $I=0.22$ and the results show that
$t^*_F>t_c$, although $t_F<t_c$ when $I\lesssim 0.24$, as shown in the inset of
Fig. 3. Hence the FM transition occurs at a higher temperature than the BEC.
However, for small $I$, $t^*_F$ is so close to $t^*_c$ that it is very
difficult to solve Eqs. (9) for the FM normal phase.
It is worth noting that the FM transition is continuous for large couplings.
This point is consistent with the Weiss molecular field theory for classical
particles, in which the FM transition is continuous\cite{mohn}, because for
large couplings, the FM transition occurs at a relatively high temperature,
when the Bose statistics reduces to Boltzmann statistics.
In conclusion, we studied the ferromagnetic phase transition and
Bose-Einstein condensation in spinor Bose gases with ferromagnetic couplings
via mean-field theory. We showed that the coupling, regardless of its
magnitude, induces a ferromagnetic phase transition at a temperature always
above the critical temperature of Bose-Einstein condensation. Moreover, the
ferromagnetic coupling also increases the critical temperatures of both phase
transitions.
R.A.K acknowledges partial support from the Max-Planck-Institut f\"ur Chemische
Physik fester Stoffe.
\begin{references}
\bibitem{mohn} P. Mohn, {\sl Magnetism in the Solid State: an Introduction}
(Springer- Verlag, Berlin, 2003)
\bibitem{cornell} M.H. Anderson {\it et al}., Science {\bf 269}, 198 (1995);
K.B. Davis {\it et al}., Phys. Rev. Lett. {\bf 75}, 3969 (1995);
C.C. Bradley {\it et al}., {\sl ibid}. {\bf 75}, 1687 (1995).
\bibitem{weinstein} J.D. Weinstein {\it et al}., Phys. Rev. A {\bf 57},
R3173 (1998).
\bibitem{ketterle} D.M. Stamper-Kurn {\it et al}., Phys. Rev. Lett. {\bf 80},
2027 (1998); J. Stenger {\it et al}., Nature {\bf 396}, 345 (1998).
\bibitem{suto} A. S\"{u}t\H{o}, J. Phys. A {\bf 26}, 4689 (1993):
E. Eisenberg and E.H. Lieb, Phys. Rev. Lett. {\bf 89}, 220403 (2002).
\bibitem{yang} K. Yang and Y.-Q. Li, Int. J. Mod. Phys. B {\bf 17}, 1027 (2003).
\bibitem{yamada} K. Yamada, Prog. Theor. Phys. {\bf 67}, 443 (1982).
\bibitem{caramico} A. Caramico D'Auria, L. De Cesare, and I. Rabuffo,
Physica A {\bf 225}, 363 (1996).
\bibitem{Simkin}M.V. Simkin and E.G.D. Cohen, Phys. Rev. A {\bf 59}, 1528
(1999).
\bibitem{gu} Q. Gu, Phys. Rev. A (in production).
\bibitem{ho} T.L. Ho, Phys. Rev. Lett. {\bf 81}, 742 (1998);
T. Ohmi and K. Machida, J. Phys. Soc. Jpn. {\bf 67}, 1822 (1998).
\bibitem{burke} J.P. Burke, Jr. and J.L. Bohn, Phys. Rev. A {\bf 59},
1303 (1999).
\bibitem{pu} H. Pu, W. Zhang, and P. Meystre, Phys. Rev. Lett. {\bf 87},
140405 (2001); W. Zhang {\it et al}., {\sl ibid}. {\bf 88}, 060401 (2002).
\bibitem{greiner} M. Greiner {\it et al}., Nature {\bf 415}, 39 (2002).
\bibitem{fisher} M.P.A. Fisher {\it et al}., Phys. Rev. B. {\bf 40}, 546 (1989);
G.G. Batrouni {\it et al.}, Phys. Rev. Lett. {\bf 89}, 117203 (2002).
\bibitem{isoshima} T. Isoshima, T. Ohmi, and K. Machida, J. Phys. Soc. Jpn.
{\bf 69}, 3864 (2000).
\bibitem{huang2} W.-J. Huang, S.-C. Gou, and Y.-C. Tsai, Phys. Rev. A {\bf 65},
063610 (2002).
\bibitem{robinson} J.E. Robinson, Phys. Rev. {\bf 83}, 678 (1951); M.H. Lee,
Phys. Rev. E {\bf 56}, 3909 (1997).
\bibitem{huang} K. Huang, {\sl Statistical Physics} (John Wiley \& Sons, New
York, 1987).
\end{references}
\end{document}