\documentstyle[amsfonts,12pt]{report}
\begin{document}
\begin{center}
{\bf Weak versus strong uniqueness of Gibbs measures:}\\
{\bf a regular short-range example}\\
\mbox{}\\
\mbox{}\\
M. Campanino\\
{\em Dipartimento di Matematica}\\
{\em Universit\'a degli Studi di Bologna}\\
{\em piazza di Porta S. Donato 5}\\
{\em I-40127 Bologna}\\
{\em Italy}\\
A.C.D. van Enter\\
{\em Institute for Theoretical Physics}\\
{\em Nijenborgh 4}\\
{\em 9747 AG Groningen}\\
{\em The Netherlands}
\end{center}
\vspace{3cm}
\noindent
{\bf Abstract:} We provide an example of a nearest neighbor random model on a
regular lattice which has a unique (disordered) Gibbs state for every
boundary condition, almost surely, although by choosing interaction-dependent
boundary conditions one can obtain different Gibbs states.
\newpage
\noindent
The correct treatment and interpretation of boundary conditions for
statistical mechanical systems is a subtle issue, especially for random
models [COvE, vE1, vE2, vEF, vEG, FH1, FH2, FH3, GNS, HF, N, NS, Oz, Z].
For instance, the notion of a plurality of pure states with an ultrametric
structure as in Parisi's proposal, as well as the mechanism by which they can
be obtained, requires a careful treatment of boundary conditions. This issue
is of great interest in spin-glass theory [MPV]. Another example is the
distinction between ,,weak'' and ,,strong'' uniqueness of Gibbs measures for
random spin systems. We say that strong uniqueness applies to a model if for
almost all choices of the interaction only one Gibbs measure exists, while
weak uniqueness holds if for any boundary condition chosen independently of
the interaction, the thermodynamic limit of the Gibbs measure is the same for
almost all interactions. Weak uniqueness was first explicitly introduced and discussed
in [COvE] where it was proved to hold for one-dimensional Ising models with
random interactions decaying as $1/r^{\alpha}$ with $\alpha > 1$;
strong uniqueness is known to hold only for $\alpha > 3/2$ ([K]). The
distinction between interaction-dependent versus interaction-independent
boundary conditions was earlier discussed in [vEF, vEG, CCST], in a more
implicit way.
Up till now there have been examples of systems which are weakly, but not
strongly, unique for spin-glass models on the Bethe lattice ([CCST, BRZ]) in
the temperature range between the ferromagnetic and the spin-glass transition
temperatures and for extreme-long-range (square summable, but non-summable)
spin-glass models ([GNS, FZ]) at high temperatures. The usual interpretation
of these results is that in these examples interaction-dependent boundary
conditions should be dismissed as ,,unphysical''.
Here we present for the first time a nearest neighbor example on a regular
lattice of this phenomenon. The interpretation of interaction-dependent
boundary conditions as ,,unphysical'' seems somewhat tenuous in this case.
The model we consider is the $q$-state nearest-neighbor Potts model on ${\mathbb{Z}}^d$
at $T_c$, for high $q$ (and $d \geq 2$) with a one-pattern Hopfield-type
site disorder
\[
H=- \sum_{\langle i,j \rangle} \delta (\xi_i \sigma_i, \xi_j \sigma_j)
\hspace{4cm} (1)
\]
where $\sigma_i \in \{ 1, ..., q\}$ and $\xi_i$ is a random permutation of the
$q$ Potts states at site $i$ [Ka, Ga].
After a gauge transformation the model is equivalent to a Potts ferromagnet,
for which it is well known [KS, BKL] that at $T_c$ there is a first-order
phase transition in the temperature variable. At $T_c$ $q$ ordered states
(Gibbs measures) coexist with a disordered one. Any fixed boundary condition,
due to the random gauge transformation, is equivalent to taking a random
choice for the boundary condition for the ferromagnet. Thus it is sufficient
to argue that a sequence of finite volume Gibbs measures for the
ferromagnetic model with randomly chosen boundary conditions converges to the
disordered state. To do this we invoke the proof of the first-order
transition due to Bricmont-Kuroda-Lebowitz [BKL]. This proof is based on
Pirogov-Sinai theory, that is a sophisticated contour argument. The
disordered state is a small perturbation of the ,,restricted ensemble'' of
configurations in which all neighboring spins are different (small for $q$
high enough). We observe that the density of pairs of neighboring spins on
the boundary which are equal is close to $\frac{1}{q}$ with large probability,
and they are ``sparse'', if the spins on the boundary
are independently chosen on each boundary site, with probability
$\frac{1}{q}$ for each of the $q$ Potts states. This means that the boundary
condition is a small random perturbation of a ,,purely restricted ensemble''
boundary condition. Therefore, a probabilistic contour argument, as in [CCF],
will lead to the desired result, that almost surely with respect to the
random boundary conditions the thermodynamic limit measure will be the
disordered state.
As we mentioned before, this implies the weak uniqueness of the
gauge-transformed random Potts-Hopfield model.
Intuitively, it is of course plausible that uncorrelated random boundary
conditions are the most ,,disordered'' ones. We note that for any extremal
Gibbs measure it is true that almost all (with respect to this {\em same\/}
Gibbs measure) boundary conditions recover the original measure [G]. What we
have considered here, is almost all boundary conditions with respect to
symmetric product measure which is the most random prescription in some sense
(for example, this measure has the largest entropy).
\bigskip
\noindent
{\bf Acknowledgements:}\
This paper
was begun during a visit of M. C. to the university of Groningen.
Part of the research of A. C. D. van Enter was made possible by a
fellowship of the Royal Dutch Academy of Arts and Sciences (K.N.A.W.).
This work was supported by EU-contract CHRX-CT93-0411.
\bigskip
\noindent
{\bf References:}
\begin{itemize}
\item[{[BKL]}]
J.Bricmont, K. Kuroda and J.L. Lebowitz; Comm.Math.Phys.\ {\bf 101}: 501 (1985)
\item[{[BRZ]}]
P.M. Blekher, J. Ruiz and V.A. Zagrebnov; Marseille Preprint, Austin
Mathematical Physics Archives 94-296.
\item[{[COvE]}]
M. Campanino, E. Olivieri and A.C.D. van Enter; Comm.Math.\ {\bf 108}: 241-255
(1987)
\item[{[CCF]}]
J.T. Chayes, L. Chayes and J. Fr\"ohlich; Comm.Math.Phys.\ {\bf 100}: 399 (1985)
\item[{[CCST]}]
J.T. Chayes, L. Chayes, J.P. Sethna, D.J. Thouless: Comm.Math.Phys.\ {\bf 106}:
41 (1986)
\item[{[vE1]}
]A.C.D. van Enter; J.Stat.Phys.\ {\bf 60}: 275 (1990)
\item[{[vE2]}]
A.C.D. van Enter; J.Phys.A {\bf 21}: 1781 (1988)
\item[{[vEF]}]
A.C.D. van Enter and J. Fr\"ohlich; Comm.Math.Phys.\ {\bf 98}: 425 (1985)
\item[{[vEG]}]
A.C.D. van Enter and R.B. Griffiths; Comm.Math.Phys.\ {\bf 90}: 319 (1983)
\item[{[FH1]}]
D.S. Fisher and D.A. Huse; Phys.Rev.Lett.\ {\bf 56}: 1601 (1986)
\item[{[FH2]}]
D.S. Fisher and D.A. Huse; J.Phys.A {\bf 20}: L1005 (1987)
\item[{[FH3]}]
D.S. Fisher and D.A. Huse; Phys.Rev.B {\bf 38}: 386 (1988)
\item[{[FZ]}]
J. Fr\"ohlich and B. Zegarlinski; Comm.Math.Phys.\ {\bf 110}: 121
\item[{[G]}]
H.-O. Georgii; Comm.Math.Phys.\ {\bf 32}: 107 (1973)
\item[{[Ga]}]
V. Gayrard; J.Stat.Phys.\ {\bf 68}: 977 (1992)
\item[{[GNS]}]
A. Gandolfi, C.M. Newman and D.L. Stein; Comm.Math.Phys.\ {\bf 157}: 371 (1993)
\item[{[HF]}]
D.A Huse and D.S. Fisher; J.Phys.A {\bf 20}: L997 (1987)
\item[{[K]}]
K. M. Khanin; Theor. Math. Phys. {\bf 15 }: 445 (1980)
\item[{[Ka]}]
I. Kanter; Phys.Rev.A {\bf 37}: 2739 (1988)
\item[{[KS]}]
R. Koteck\'y and S.B. Shlosman; Comm.Math.Phys.\ {\bf 83}: 493 (1982)
\item[{[MPV]}]
M. M\'ezard, G. Parisi and M.A. Virasoro; Spin Glass Theory and Beyond, World
Scientific, Singapore-New Jersey-Hong Kong (1987)
\item[{[N]}]
C.M. Newman; in Probability and Phase transition, Ed. G. Grimmett, 247 (1994),
NATO ASI (420, Kluwer Dordrecht, Boston, London)
\item[{[NS]}]
C.M. Newman and D.L. Stein; Phys.Rev.B {\bf 46}: 973 (1992)
\item[{[Oz]}]
Y. Ozeki; J.Phys.Soc.Japan {\bf 62}: 2641 (1993)
\item[{[Z]}]
B. Zegarlinski; Rev.Math.Phys.\ {\bf 6}: 115 (1994)
\end{itemize}
\end{document}