1: \documentclass[a4paper,10pt]{article}
2: %
3: \usepackage[latin1]{inputenc}
4: %\usepackage[italian]{babel}
5: \usepackage{graphicx}
6: %
7: %
8: \begin{document}
9: \begin{titlepage}
10: \begin{center}
11: \Large{\bf{Canonical solution of a system of long-range
12: interacting rotators on a lattice}}
13: \end{center}
14: \vspace{2cm}
15: \begin{center}
16: \large{Alessandro Campa$^1$, Andrea Giansanti$^{2, \S}$
17: and Daniele Moroni$^2$}
18: \end{center}
19: \vspace{1cm}
20: \begin{center}
21: \normalsize{$^1$Physics Laboratory, Istituto Superiore di Sanit\`a and
22: INFN Sezione di Roma1, Gruppo Collegato Sanit\`a\\Viale Regina
23: Elena 299, 00161 Roma, Italy}
24: \end{center}
25: \begin{center}
26: \normalsize{$^2$Physics Department, Universit\`a di Roma ``La Sapienza'' and
27: INFM Unit\`a di Roma1, \\ Piazzale Aldo Moro 2, 00185 Roma, Italy}
28: \end{center}
29: \vspace{0.5cm}
30: \begin{center}
31: \emph{(13 April 2000, accepted for publication in Physical Review E)}
32: \end{center}
33: \vspace{1cm}
34: \begin{center}
35: \large{\bf {Abstract}}
36: \end{center}
37: \vspace{1cm}
38: \small{
39: The canonical partition function of a system of rotators (classical
40: X-Y spins) on a lattice, coupled by terms decaying as the inverse of
41: their distance to the power $\alpha$, is analytically computed. It is
42: also shown how to
43: compute a rescaling function that allows to reduce the model, for any
44: $d$-dimensional lattice and for any $\alpha<d$, to the mean field
45: ($\alpha=0$) model.
46: }
47: {\bf PACS}: 05.20.-y, 05.70.Ce, 05.10.-a
48:
49: \vspace{2cm}
50: $^{\S}$ \small{Author to whom correspondence should be addressed; electronic address:
51: Andrea.Giansanti@roma1.infn.it}
52: \end{titlepage}
53: \section{Introduction}
54: Let us consider the following classical hamiltonian model of a system
55: of rotators:
56: \begin{equation} \label{num1}
57: H=\frac{1}{2} \sum_{i=1}^N L_i^2 \, + \, \frac{1}{2}
58: \sum_{i,j=1}^N\left[1-\cos(\theta_i-\theta_j)\right]=K+V\, .
59: \end{equation}
60: The potential energy $V$ is not thermodynamically stable and the ensemble
61: averaged energy
62: density $U=\langle \frac{H}{N}\rangle$ diverges in the thermodynamic
63: limit (TL) \cite{ruelle}.
64: If the potential energy term is divided by $N$, then the
65: energy density becomes intensive and it
66: is bounded as $N$ goes to infinity.
67:
68: Indeed, dynamics and thermodynamics of the $1/N$ rescaled model
69: has been extensively investigated \cite{LRR}; in particular, Ruffo and
70: Antoni, who called it the hamiltonian mean field X-Y model (HMF),
71: solved it in the canonical ensemble, and compared the
72: theoretical caloric ($T\, vs\, U$) and magnetization ($M\, vs\, U$)
73: curves with those obtained from a microcanonical simulation \cite{AR}.
74:
75: Here we consider a generalization of model (\ref{num1}):
76: \begin{equation} \label{num2}
77: H=\frac{1}{2} \sum_{i=1}^{N} L_i^2 \, + \, \frac{1}{2}
78: \sum_{i\neq j}^{N} \frac {1-\cos(\theta_i-\theta_j)}
79: {r_{ij}^\alpha} \, .
80: \end{equation}
81: The rotators are placed at the sites of a lattice
82: and the interaction between rotators $i$ and $j$ decays as the inverse of
83: their distance to the power $\alpha$.
84:
85: A onedimensional version of model (\ref{num2}) has been studied by
86: Anteneodo and Tsallis \cite{AT}, who have numerically measured the largest
87: Lyapounov exponent, as a function of $N$ and $\alpha$.
88: Through a rescaling factor $N^{*}=\frac {N^{1-\alpha}-1}{1-\alpha}$
89: Anteneodo and Tsallis showed that their results
90: coincide with those previously obtained for the HMF ($\alpha=0$) model;
91: this rescaling could then give a well defined TL to model (\ref{num2}).
92:
93: In a recent paper Tamarit and Anteneodo, using a rescaling
94: factor $\tilde{N}= 2^{\alpha}\frac {N^{1-\alpha}-1}{1-\alpha}$, have shown
95: that the caloric and magnetization curves of model (\ref{num2}) in one
96: dimension collapse onto the curves
97: of the HMF model \cite{TA}.
98: This universality emerges plotting $T/\tilde{N}$ as a function of
99: $H/N\tilde{N}$
100: and $M$ as a function of $H/N\tilde{N}$, from molecular dynamics simulation
101: of model (\ref{num2}) for different $N$ and $\alpha$ values. These authors
102: conjecture that
103: the results they obtained in the onedimensional case might be general,
104: valid in any dimension $d$ and for $\alpha <d$, as suggested also
105: in \cite{TA}.
106: \section{Partition function}
107: In this work, inspired by \cite{AR} and \cite{TA}, we analytically compute
108: the partition function of an $\tilde{N}$-rescaled model (\ref{num2})
109: for any $d$ and $\alpha <d$.
110: In formula (\ref{exactn}) we give the right expression of the rescaling
111: function $\tilde{N}$, to obtain universal state curves for all lattice
112: models with long range ($\alpha < d$) interactions.
113:
114: Let us now rewrite the rescaled version of Hamiltonian (\ref{num2}):
115: \begin{eqnarray}\label{num3}
116: H&=&\frac{1}{2} \sum_{i=1}^N L_i^2 \, + \, \frac{1}{2 \tilde{N}}
117: \sum_{i,j=1}^N \frac {1-\cos(\theta_i-\theta_j)}
118: {r_{ij}^\alpha} \nonumber \\
119: &&- h_x \sum_{i=1}^N m_{ix} - h_y\sum_{i=1}^N m_{iy} \, ,
120: \end{eqnarray}
121: where we have introduced an external magnetic field
122: $\mathbf{h}=(h_x,h_y)$ of modulus $h$, that makes possible to compute
123: the magnetization.
124: The indexes $i,j$ label the sites of a $d$-dimensional generic
125: lattice; $r_{ij}$ is the distance between them, with periodic
126: boundary conditions and nearest image convention (the definition of
127: $r_{ii}$ will be given shortly); $\alpha \geq 0$.
128: At each site a classical rotator (X-Y spin) of unit
129: momentum of inertia is represented by conjugate canonical
130: coordinates $(L_i,\theta_i)$, where the $L_i$'s are angular
131: momenta, and the $\theta_i$'s $\in [0,2\pi )$ are the angles of
132: rotation on a family of parallel planes,
133: each one defined at each lattice point; $x$ and $y$ refer to the
134: components of boldface twodimensional vectors defined over these planes.
135: To each lattice site a spin vector
136: \begin{equation}
137: \mathbf{m}_i=(m_{ix},m_{iy})=(\cos \theta_i,\sin \theta_i)
138: \end{equation}
139: is associated, and the total magnetization is given by:
140: \begin{equation}
141: \mathbf{M}=(M_x,M_y)=\frac{1}{N}\sum_{i=1}^N\mathbf{m}_i \, .
142: \end{equation}
143: Note in (\ref{num3}) the rescaling factor $\tilde{N}$ in front of the
144: potential energy term, now written as a free double sum over both indexes.
145: $\tilde{N}$ should be regarded as an unknown function of
146: $N,\alpha,d$ and the geometry of the lattice, with the
147: fundamental property of making
148: \begin{equation}\label{Ntildedefi}
149: \frac{1}{\tilde{N}}\sum_{j,j \neq i} \frac{1}{r_{ij}^\alpha}
150: \end{equation}
151: an intensive quantity; this guarantees the thermodynamic stability of
152: the potential. We also note that the sum in (\ref{Ntildedefi}) is
153: independent of the origin $i$ because of periodic conditions.
154: To reproduce the usual HMF it is also necessary that
155: $\tilde{N}(N,\alpha=0,d)=N$.
156: The constraint $i\neq j$ over the double sum is removed defining
157: $r_{ii}^\alpha = 1/b$, a finite number. Since the numerator
158: $1-\cos(\theta_i-\theta_j)$ is zero for $i=j$ the choice of $b$ is
159: free. The removal of the constraint allows to introduce the distance
160: matrix $R'_{ij}=\frac{1}{r_{ij}^\alpha}$; the diagonalization of
161: such matrix is the key point to obtain, in the computation of the
162: partition function, known integrals in the variables $\theta_i$.
163:
164: As usual the partition function factorizes in a kinetic
165: part $Z_K=\left(\frac{2 \pi}{\beta} \right)^{\frac{N}{2}}$, where
166: $\beta=1/k_B T$, and a potential part $Z_V$. After defining
167: $R_{ij}=\frac{\beta}{2 \tilde{N}} R'_{ij}$,
168: $\mathbf{B}=\beta \mathbf{h}$, $C=\exp\left(-\frac{\beta}{2 \tilde{N}}
169: \sum_{ij} \frac {1}{r_{ij}^\alpha}\right)$, the potential part
170: can be written as:
171: \begin{equation}\label{Zmatr}
172: Z_V=C \int_{-\pi}^{\pi}
173: d^N\theta\, \exp\left[\sum_{i,j,\mu}m_{i\mu}R_{ij}
174: m_{j\mu} + \sum_iB_{\mu}m_{i\mu} \right],
175: \end{equation}
176: where $\mu=x,y$. Diagonalizing the symmetric matrix $R=(R_{ij})$ with the
177: unitary matrix
178: $U$ such that $R=U^TDU$, $D=(R_i\delta_{ij})$, where $R_i$ are the
179: eigenvalues of $R$, we can write the first part of the exponent in
180: (\ref{Zmatr}) as:
181: \begin{equation}\label{diago}
182: \sum_{ij}\left(m_{ix}R_{ij}m_{jx} + m_{iy}R_{ij}m_{jy}\right) =
183: \sum_i\left(n_{ix}^2R_i + n_{iy}^2R_i\right),
184: \end{equation}
185: where $n_{i\mu}=\sum_jU_{ij}m_{j\mu}$. In order
186: to apply the gaussian transformation:
187: \begin{equation}\label{gauss}
188: e^{aS^2}=\frac{1}{\sqrt{4\pi a}}\int_{-\infty}^{+\infty}dz e^{-\frac{z^2}{4a}+Sz}
189: \,\,\,\, a>0
190: \end{equation}
191: to each term of the sum in the right hand side of (\ref{diago}), each
192: $R_i$ must be positive. The spectrum can be explicitly
193: computed using a $d$-dimensional Fourier transform of matrix $R$, the
194: eigenvalues being labelled by vectors of the reciprocal lattice. These
195: eigenvalues are trivially related to those of matrix $R'$. A study of the
196: spectrum of $R'$ in the limit $N \rightarrow \infty$ and for $b=0$ shows
197: that: when $\alpha > d$ each element of the spectrum converges to a finite
198: quantity, the least eigenvalues being negative and of order one in modulus;
199: when $\alpha < d$ a part of the spectrum converges to a finite quantity,
200: another part diverges to $+ \infty$, at most as $\tilde{N}$. However this
201: last part consists of a
202: fraction of the total number of eigenvalues which goes to zero in the limit
203: $N \rightarrow \infty$. The least eigenvalue is still negative and of order
204: one in modulus.
205: Then part of the spectrum is negative, but it is easily seen that it is
206: shifted by $b$. Thus calling $p$ the least eigenvalues of $R'$ for $b=0$
207: and choosing
208: \begin{equation}\label{defbi}
209: b=-p + \epsilon \quad \epsilon > 0 \quad,
210: \end{equation}
211: we have that with this $b$ the whole spectrum of $R'$ (and therefore that
212: of $R$) becomes positive. Then for each $i=1,\cdots,N$,
213: $\mu=x,y$ we can apply (\ref{gauss}) with the correspondence
214: $a\rightarrow R_i$, $S\rightarrow n_{i\mu}$, $z\rightarrow z_{i\mu}$.
215: Performing the integrals over variables $\theta_i$ and using the
216: transformation $z_{i\mu}=2\sum_j(UR)_{ij}\Psi_{j\mu}$ with Jacobian
217: $2^N\det R$, we can rewrite the partition function as:
218: \begin{eqnarray}
219: Z&=&CZ_K\frac{\det R}{\pi^N}\int_{-\infty}^{+\infty} d^N\Psi_x
220: d^N\Psi_y \\
221: && e^{ N\left[-\sum_{ij\mu} \Psi_{i\mu} \frac{R_{ij}}
222: {N}\Psi_{j\mu} +\frac{1}{N} \sum_l \ln \left( 2 \pi I_0\left( |2 \sum_j R_{lj}
223: \mathbf{\Psi}_j+\mathbf{B}|\right) \right) \right] } \nonumber
224: \end{eqnarray}
225: where $I_0$ is the zeroth order modified Bessel function.
226: The isolation of the $N$ factor in the exponential prepares the object
227: for the use of the saddle point method. The quantity in square brackets
228: is intensive. Double sums in the first two terms are compensated by
229: $R/N=(\beta/2 N \tilde{N}) R'$ and the last sum has $1/N$ in front of
230: it. The argument of $I_0$ is also intensive because involves a
231: term of the form $\sum_j R_{lj}=(\beta / 2 \tilde{N}) \sum_j R'_{lj}$.
232: If we call $f(w)$ the function in square brackets, where
233: $w=(\Psi_{1x},\cdots,\Psi_{Nx},\Psi_{1y},\cdots,\Psi_{Ny})$, then
234: the application of the method requires the following three conditions:
235: $f(w)$ admits a stationary point $w_0$; $w_0$ is a simple stationary
236: point, i.e., $\det He f|_0\neq 0$, where $He f|_0$ is the hessian matrix
237: of $f$ in $w_0$; the path of integration can be
238: deformed (generally going into ${\cal C}^{2N}$) into a path that
239: passes through $w_0$ following the steepest descent of $f(w)$ and such
240: that $f(w)<f(w_0)$ throughout the all path. If the point $w_0$ is a
241: maximum no deformation is necessary and the method is also called the
242: Laplace method. Since, as we show below, $w_0$ is indeed a real-valued
243: maximum, we readily obtain for the free energy per particle $F$:
244: \begin{eqnarray}\label{fren}
245: -\beta F&=&\lim_{N\rightarrow \infty}\frac{\ln Z}{N}=
246: \lim_{N\rightarrow \infty} \{ \frac{1}{2}\ln
247: \left( \frac{2\pi}{\beta}\right) -\frac{\beta}{2\tilde{N}}\sum_{j}
248: \frac{1}{r_{ij}^\alpha} \nonumber \\
249: &&+ \max_{w}[f(w)] +\frac{1}{N}\ln \frac{\det R}{\sqrt
250: {\det \left( -\frac{N}{2} He f|_0\right)}} \} \, .
251: \end{eqnarray}
252: The stationary point $w_0$ is given by the vector
253: $(\Psi_x,\cdots,\Psi_x,\Psi_y,\cdots,\Psi_y)$, homogeneous on the lattice
254: sites. Defining $\mathbf{\Psi}=
255: (\Psi_x,\Psi_y)$, its direction is that of $\mathbf{B}$, and its modulus
256: $\Psi$ is given by the solution of:
257: \begin{equation}\label{psieq}
258: \Psi =\frac{I_1}{I_0}\left( \beta\left[A\Psi +h\right] \right) \, ,
259: \end{equation}
260: with
261: \begin{equation}\label{ascal}
262: A=\frac{1}{\tilde{N}} \sum_j R'_{ij}=\frac{1}{\tilde{N}} \left[ b +
263: \sum_{j\neq i}\frac{1}{r_{ij}^\alpha} \right] \, ,
264: \end{equation}
265: and where $I_1$ is the first order modified Bessel function. In (\ref{ascal})
266: $A$ does not depend on $i$ because of the periodic boundary conditions.
267: We note that when $h=0$ we have
268: infinitely many degenerate solutions, since only the modulus $\Psi$ is
269: determined. Evaluation of the elements of the hessian matrix at the
270: stationary point gives:
271: \begin{equation}\label{hesmat}
272: \left. -\frac{N}{2} \frac{\partial^2 f}{\partial \Psi_{i\mu}
273: \partial \Psi_{j\nu}}
274: \right|_0 =
275: R_{ij} \delta_{\mu \nu} - (R^2)_{ij} g_{\mu \nu}(w_0)
276: \end{equation}
277: where we do not give the explicit expression of $g_{\mu \nu}(w_0)$.
278: As we will see shortly, the eigenvalues analysis of the hessian matrix
279: (\ref{hesmat}) shows that the stationary point $w_0$ is a maximum. Then,
280: Laplace method applies and Eq. (\ref{fren}) is valid. However, only in the
281: long range case ($\alpha<d$) the last term in the rightmost side of
282: (\ref{fren}) is zero; when $\alpha>d$ its expression does not appear to
283: be manageable. We will comment on this point later. Restricting then to
284: $\alpha<d$, and computing the derivative of (\ref{fren}) with
285: respect to the magnetic field we find that the magnetization
286: $M=\langle|\mathbf{M}|\rangle$ is given by the solution $\Psi$ of
287: (\ref{psieq}). Then the internal energy $U$ is given by:
288: \begin{equation}\label{energy}
289: U=\frac{\partial (\beta F)}{\partial \beta} = \frac{1}{2\beta} +
290: \frac{A}{2}(1-M^2) -hM \, .
291: \end{equation}
292: Equations (\ref{psieq}) and (\ref{energy}) are the same as those
293: of HMF, as soon as a proper $\tilde{N}$ rescaling gives
294: \begin{equation}\label{rescal}
295: A=\frac{1}{\tilde{N}} \sum_j R'_{ij}=\frac{1}{\tilde{N}} \left[
296: b + \sum_{j\neq i}\frac{1}{r_{ij}^\alpha} \right] = 1 \, .
297: \end{equation}
298: Now, from equations (\ref{hesmat}) and (\ref{rescal}), and calling
299: $\lambda_n$ the eigenvalues of $R'$, we find that, choosing $\mathbf{B}$
300: along one of the coordinate axes, the eigenvalues of the hessian matrix
301: at the stationary point are given by:
302: \begin{eqnarray}\label{eighes}
303: \chi^{(1)}_n&=&\frac{\beta}{2}\frac{\lambda_n}{\tilde{N}}\left[1-
304: \left(\beta-\Psi^2\beta-\frac{\Psi}{\Psi+h}\right)
305: \frac{\lambda_n}{\tilde{N}}\right] \\
306: \chi^{(2)}_n&=&\frac{\beta}{2}\frac{\lambda_n}{\tilde{N}}\left[1-
307: \frac{\Psi}{\Psi+h}\frac{\lambda_n}{\tilde{N}}\right] \quad \quad
308: n=1,\cdots,N
309: \nonumber
310: \end{eqnarray}
311: Following our previous analysis we have that:
312: \begin{equation}
313: \frac{\epsilon}{\tilde{N}}\leq \frac{\lambda_n}{\tilde{N}}\leq 1 \, .
314: \end{equation}
315: Then we immediately see that $\chi^{(2)}_n$ are all positive for any
316: $\beta$ and $h$; for $\chi^{(1)}_n$ we need to include $\Psi (\beta,h)$ from
317: (\ref{psieq}). We have checked numerically that the quantity in round brackets
318: in (\ref{eighes}) is always smaller than $1$, and therefore $\chi^{(1)}_n$
319: are also all positive. From (\ref{eighes}) we can derive an expression for
320: the determinant of matrix (\ref{hesmat}). It is given by:
321: \begin{eqnarray}\label{dethes}
322: \frac{1}{N}&&\ln \det \left( -\frac{N}{2} He f|_0\right)=\frac{2}{N}\ln \det R
323: \nonumber \\ +&&\frac{1}{N}\sum_{n=1}^N \{ \ln \left[1-
324: \left(\beta-\Psi^2\beta-\frac{\Psi}{\Psi+h}\right)
325: \frac{\lambda_n}{\tilde{N}}\right]\nonumber \\ +&&\ln \left[1-
326: \frac{\Psi}{\Psi+h} \frac{\lambda_n}{\tilde{N}}\right]
327: \} \, .
328: \end{eqnarray}
329: When $\alpha<d$ most of $\frac{\lambda_n}{\tilde{N}}$ go to zero for
330: $N\rightarrow \infty$, then the sum in (\ref{dethes}) is
331: effectively constituted by the terms with the remaining
332: $\frac{\lambda_n}{\tilde{N}}$. These terms are a fraction of $N$ that, as
333: we already pointed out, goes to zero when $N\rightarrow \infty$. If we call
334: $N'(N)$ this fraction, then the sum in (\ref{dethes}) can be bounded
335: from above by $\frac{N'}{N}c \rightarrow 0$ for $N\rightarrow \infty$, where
336: $c$ is a finite number. Therefore the last term in (\ref{fren}) is zero.
337: When $\alpha>d$ all terms contribute to the sum in (\ref{dethes}), and we
338: can not give a meaningful expression for (\ref{fren}). At the end of the
339: calculations we can let $\epsilon \rightarrow 0$ in (\ref{defbi}).
340:
341: Then we have shown that any model with $\alpha<d$ on any lattice is
342: equivalent to HMF. From (\ref{rescal}) we get an exact expression
343: for $\tilde{N}$:
344: \begin{equation}\label{exactn}
345: \tilde{N} = -p + \sum_{j\neq i}\frac{1}{r_{ij}^\alpha} \, .
346: \end{equation}
347: We have made a microcanonical simulation of Hamiltonian
348: (\ref{num3}) on a threedimensional simple cubic lattice
349: in zero magnetic field, using a fourth
350: order simplectic algorithm \cite{yosh} with time step $0.02$, selected to have
351: relative energy fluctuations not exceeding $1/10^6$. We have chosen a fixed
352: $N=343=7^3$, and have simulated various energy densities
353: $H/N$ and various $\alpha<3$. In Fig. 1 we show that the numerical
354: caloric curves collapse onto the universal HMF curve. The kind of results
355: shown in \cite{TA} for a onedimensional lattice, where a slightly
356: different $\tilde{N}$ has been used.
357:
358: \begin{figure}[htbp]
359: \begin{center}
360: \includegraphics[width=12cm]{fig1.eps}
361: \end{center}
362: \caption{The full line gives the canonical theoretical caloric curve
363: (temperature $T$ $vs$ energy density $U$) for
364: long range rotators compared with the microcanonical simulation of a
365: threedimensional simple cubic lattice for three different $\alpha$ values:
366: $0.75$ (open circles), $1.5$ (diamonds) and $2.25$ (crosses). Note that
367: in spite of the size of the system, still not very large (side with $7$
368: lattice sites), the results already follow very well the theoretical
369: curve.}
370: \label{Fig. 1}
371: \end{figure}
372:
373: \section{Conclusions}
374: Going back to the beginning of our discussion:
375: it is now clear that model (\ref{num2}) completely reduces to
376: model (\ref{num1}) for $\alpha=0$. In model (\ref{num1}) the range of the
377: interactions is infinite; each rotator interacts with
378: all the others and with the same intensity. To get
379: a well defined TL it is sufficient to divide
380: $V$ in (\ref{num1}) by $N$, the total numbers of rotators.
381: It is then possible to compute caloric and magnetization
382: curves \cite{AR}; the spatial arrangement of the rotators has no effect on
383: them since the intensity of the interaction is the same for each couple
384: of rotators. In this work
385: we have shown that, when considering model (\ref{num2}), it is possible
386: to take into account the spatial $d$-dimensional arrangement
387: of the rotators and the decaying of their mutual interaction
388: through a factor $\tilde{N}$, which is computable for any
389: periodic lattice and any $\alpha < d$. Dividing by $\tilde{N}$
390: the potential energy in (\ref{num2}), the model gets a well defined
391: TL and it is possible to compute state curves which become those
392: of the HMF model with a proper normalization of the constant $A$
393: in (\ref{rescal}). The HMF ($\alpha =0$) model has revealed peculiar
394: equilibrium and nonequilibrium properties \cite{LRR}, namely:
395: ensemble inequivalence, metastability, collective oscillations,
396: anomalous diffusion and interesting chaotic properties, both in
397: the ferromagnetic and antiferromagnetic case. On the basis of
398: the thermodynamical equivalence here established it would be
399: interesting to investigate the $\alpha$ dependence of all these
400: properties. The study of the Lyapounov exponents in \cite{AT}
401: is the first in this direction.
402: \section{Aknowledgments}
403: A. G. warmly thanks C. Tsallis for having suggested the study
404: of the long range interacting rotators.
405:
406:
407: \begin{thebibliography}{9}
408: \bibitem{ruelle}
409: D. Ruelle, {\it Statistical mechanics: rigorous results},
410: (Addison-Wesley, New York 1989).
411: \bibitem{LRR}
412: V. Latora, A. Rapisarda, and S. Ruffo, cond-mat/0001010,
413: to appear in Prog. Theor. Phys. Supplement.
414: \bibitem{AR}
415: M. Antoni and S. Ruffo, Phys. Rev. E {\bf 52}, 2361 (1995).
416: \bibitem{AT}
417: C. Anteneodo and C. Tsallis, Phys. Rev. Lett. {\bf 80}, 5313 (1998).
418: \bibitem{TA}
419: F. Tamarit and C. Anteneodo, Phys. Rev. Lett. {\bf 84}, 208 (2000).
420: \bibitem{yosh}
421: H. Yoshida, Phys. Lett. A {\bf 150}, 262 (1990).
422: \end{thebibliography}
423: \end{document}
424:
425:
426: