%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%% %%% %%% ON THE MICROSCOPIC THEORY OF PHASE COEXISTENCE %%% %%% %%% %%% Author: Salvador MIRACLE SOLE %%% %%% %%% %%% This file contains the manuscript, in plain TEX, %%% %%% of a contribution to the XIII SITGES CONFERENCE %%% %%% "25 Years of Non-Equilibrium Statistical Mechanics" %%% %%% June 1994 %%% %%% %%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \magnification 1200 \hsize 17truecm \vsize 24truecm \font\ten=cmbx10 at 13pt \font\twelve=cmbx10 \font\eightrm=cmr8 \font\ninerm=cmr9 \font\pc=cmcsc10 \rm \font\baton=cmss10 \baselineskip 15pt %\hsize=122mm %\vsize=193mm \parindent=3mm \def\relatif{\ \hbox{{\rm Z}\kern-.4em\hbox{\rm Z}}} \def\Reel{I\!\!R} { \centerline{{\twelve Centre de Physique Th\'eorique}\footnote{$^{\star}$}{\eightrm Unit\'e Propre de Recherche 7061}{\twelve - CNRS - Luminy, Case 907}} \centerline{{\twelve F--13288 Marseille Cedex 9 - France}} \vskip 4truecm \centerline{{\ten ON THE MICROSCOPIC THEORY}} \centerline{{\ten OF PHASE COEXISTENCE}\footnote{$^{\star\star}$}{\eightrm Contribution to the XIII Sitges Conference on Statistical Mechanics, Sitges-Barcelona, Spain, June 1994}} \bigskip \centerline{ {\bf Salvador Miracle Sol\'e} } \vskip 2truecm \centerline{\bf Abstract} \medskip Some rigorous results concerning the microscopic theory of interfaces and crystal shapes in classical lattice systems are reported. \vskip 3truecm %\noindent Number of figures : 1 \bigskip \noindent October 1994 \noindent CPT-94/P.3077 \bigskip \noindent anonymous ftp or gopher : cpt.univ-mrs.fr \footline={} \vfill\eject } \noindent{\pc 1. Introduction} \medskip \noindent It is known that the equilibrium shape of a crystal is obtained, according to the Gibbs thermodynamic theory, by minimizing the total surface free energy associated to the crystal-medium interface, and that this shape is given by the Wulff construction, provided one knows the anisotropic surface tension (or interfacial free energy per unit area). It is therefore important, even if a complete microscopic derivation of the Wulff construction within statistical mechanics has been proved only for some two-dimensional lattice models (see the recent work by Dobrushin {\it et al.} [1,2] and also Ref.\thinspace 3), to study the properties of the surface tension $\tau ({\bf n})$, as a function of the unit vector ${\bf n}$ which specifies the orientation of the interface with respect to the crystal axes. In the first approximation the crystal can be modelled by a lattice gas. In these notes, we shall present some new rigorous results on this subject which relate, in particular, to the problem of the appearance of plane facets in the Wulff equilibrium shape. For this purpose several aspects of the microscopic theory of interfaces are analysed, and another important quantity in this theory, the step free energy, is investigated (a complete version of this work will appear later [4]). In the last Section we shall report on some recent developments related to the theory of crystal growth. \bigskip \noindent{\pc 2. Gibbs states and interfaces} \medskip \noindent First we recall some classical results about the Gibbs states of the Ising model (to some extent these results were already discussed in Ref.\thinspace 5). The model is defined on the cubic lattice ${\cal L}=\relatif^3$, with configuration space $\Omega = \{ -1, 1\}^{\cal L}$. The value $\sigma (i)$ is the spin at the site $i$. The energy of a configuration $\sigma_{\Lambda} = \{\sigma(i), i\in \Lambda \}$, in a finite subset $\Lambda \subset {\cal L} $, under the boundary conditions ${\bar \sigma}\in\Omega$, is $$ H_{\Lambda}(\sigma_{\Lambda}\mid\bar{\sigma}) = - \sum_{\langle i,j \rangle \cap \Lambda \not= \emptyset} \sigma (i)\sigma (j) $$ where $\langle i,j \rangle$ are pairs of nearest neighbour sites and $\sigma (i) = \bar{\sigma} (i) $ if $ i \not\in \Lambda $. The partition function, at the inverse temperature $\beta=1/kT$, is given by $$ Z^{\bar\sigma}(\Lambda) =\sum_{\sigma_{\Lambda}}\exp \big(-\beta H_{\Lambda}(\sigma_{\Lambda}\mid{\bar\sigma})\big) $$ It is known that this model presents, at low temperatures $T0$, we expect this interface, which at $T=0$ coincides with the plane $i_3=-1/2$, to be modified by deformations. It can be described by means of its defects, or excitations, with respect to the interface at $T=0$. These defects, called {\it walls}, form the boundaries (which may have some width), between the smooth plane portions of the interface. In this way the interface structure, with its probability distribution in the corresponding Gibbs state, may then be interpreted as a ``gas of walls'' on a two-dimensional lattice. Using the Peierls method, Dobrushin [6] proved the dilute character of this gas at low temperatures, which means that the interface is essentially flat (or rigid). The considered boundary conditions yield indeed a non translation invariant Gibbs state. Furthermore, cluster expansion techniques have been applied by Bricmont {\it et al.} [7,8], to study the interface structure in this case (see also Ref.\thinspace 9). The same analysis applied to the two-dimensional model shows a different behaviour at low temperatures. In this case the walls belong to a one-dimensional lattice, and Gallavotti [10] proved that the microscopic interface undergoes large fluctuations of order $\sqrt{L_1}$. The interface does not survive in the thermodynamic limit, $\Lambda\to\infty$, and the corresponding Gibbs state is translation invariant. Moreover, the interface structure can be studied by means of a cluster expansion for any orientation of the interface (see also Ref.\thinspace 11). Such a problem in the three-dimensional case leads to very difficult problems of random surfaces. This is one of the serious difficulties which face the attempts to generalise the work by Dobrushin {\it et al.} [1] to the three-dimensional Ising model, since a very accurate description of the microscopic interface for any orientation ${\bf n}$ is needed in this work. \vfill\eject \noindent {\pc 3. The surface tension} \medskip \noindent The free energy, per unit area, due to the presence of the interface, is the surface tension. It can be defined by $$ \tau({\bf n})= \lim_{L_1,L_{2}\rightarrow\infty} \, \lim_{L_3 \rightarrow\infty} \, -{{n_1} \over {\beta L_1 L_2}} \ln \, {Z^{(\pm,{\bf n})}(\Lambda)\over Z^{(+)}(\Lambda)} $$ Notice that in this expression the volume contributions proportional to the free energy of the coexisting phases, as well as the boundary effects, cancel, and only the contributions to the free energy of the interface are left. \medskip \noindent \pc Theorem 1. \it The thermodynamic limit $\tau ({\bf n})$, of the interfacial free energy per unit area, exists, and, as a function of ${\bf n}$, extends by positive homogeneity to a convex function $f({\bf x})=\vert {\bf x} \vert \, \tau ({\bf x}/ \vert {\bf x} \vert) $ defined for any vector ${\bf x}\in \Reel^3$. \rm \medskip A proof of these statements was given in Ref.\thinspace 12 using correlation inequalities (this being the reason for their general validity). Moreover, we know (from Refs.\thinspace 13 and 14 and the convexity condition) that $\tau ({\bf n})$ is strictly positive for $T 0$ if $T < T_c^{d=2}$, was proved in Ref.\thinspace 18, as well as that $\tau^{\rm step} = 0$ if $T \ge T_c$. Since then, however, it appears to be no proof of the fact that $T_R < T_c$. At present one is able to study rigorously the roughening transition only for some simplified models of the microscopic interface. Thus, Fr\"ohlich and Spencer [19] have proved this transition for the SOS (solid-on-solid) model, and several restricted SOS models, which are exactly solvable, have also been studied in this context (these models are reviewed in Refs.\thinspace 20 and 21). In order to define the step free energy we consider the box $\Lambda$ as above and and introduce the ({\eightrm step},{\bf m}) boundary conditions, associated to the unit vectors $ {\bf m} = (\cos \phi,\sin \phi) \in \Reel^2 $, by $$ {\bar \sigma }(i) = \cases{ 1 &if $ i>0 $ or if $ i_3=0 $ and $ i_1m_1+i_2m_2\ge 0 $\cr -1 &otherwise \cr} $$ Then, the {\it step free energy}, for a step orthogonal to ${\bf m}$ (such that $m_2\ne0$), is $$ \tau ^{\rm step}(\phi ) = \lim _{L_1 \to \infty } \lim _{L_2 \to \infty } \lim _{L_3 \to \infty } - { {\cos \phi}\over {\beta L_1} } \ \ln \ { {Z^{({\rm step},{\bf m})}(\Lambda )}\over {Z^{(\pm,{\bf n}_0)}(\Lambda )} } $$ Clearly, this expression represents the residual free energy due to the considered step, per unit length. When considering the configurations under the ({\eightrm step},{\bf m}) boundary conditions, the step may be viewed as a defect on the rigid interface. It is in fact, a long wall going from one side to the other of the box $\Lambda$. A more careful description of it can be obtained as follows. At $T=0$, the step parallel to the axis (i. e., for ${\bf m}=(0,1)$) is a perfectly straight step of height 1. At a low temperature $T>0$, some deformations appear, connected by straight portions of height 1. The step structure, with its probability distribution in the corresponding Gibbs state, can then be described as a ``gas'' of these defects (to be called {\it step-jumps}), on a one-dimensional lattice. This description, somehow similar to the description of the interface of the two-dimensional Ising model used by Gallavotti [10], is valid, in fact, for any orientation ${\bf m}$ of the step. It can be shown that the gas of step-jumps is a dilute gas at low temperature and, as a consequence of this fact, cluster expansion techniques can be applied in order to study the step structure. Actually, the step-jumps are not independent since the rest of the system produces an effective interaction between them. Nevertheless, this interaction can be treated by applying the low temperature expansion, in terms of walls, for the rigid interface, to the regions of the interface lying at both sides of the step. {}From this analysis one gets the following result. \medskip \noindent \pc Theorem 3. \it If the temperature is low enough (i.e., if $T\le T_0 $ where $T_0 > 0$ is a given constant), then the step free energy $ \tau^{\rm step} $, exists in the thermodynamic limit, and extends by positive homogeneity to a strictly convex function. Moreover, it can be expressed in terms of an analytic function of $T$, for which a convergent power series expansion can be obtained from the above mentioned cluster expansion. \rm \medskip In fact, $$ \eqalignno{ \tau^{\rm step}({\bf m}) =\ \ 2J (|m_1|+|m_2|) &- (1/\beta) \big( (|m_1|+|m_2|) \ln (|m_1|+|m_2|) \cr &- |m_1|\ln|m_1| - |m_2|\ln|m_2| \big) - (1/\beta) \varphi_{\bf m} (\beta) \cr} $$ where $ \varphi_{\bf m} $ is an analytic function of $z=e^{-2\beta}$, for $|z| \le e^{-2\beta_0}$. The first two terms in this expression, which represent the main contributions for $T\to 0$, come from the ground state of the system under the considered boundary conditions. The first term can be recognised as the residual energy of the step at zero temperature and, the second term, as $-(1/\beta)$ times the entropy of this ground state. The same two terms occur in the surface tension of the two-dimensional Ising model (see Ref.\thinspace 22). By considering the lowest energy excitations, it can be seen that $\varphi_{\bf m}$ is $O(e^{-4\beta})$, and also, that the first term in which this series differs from the series associated to the surface tension of the two-dimensional Ising model, is $O(e^{-12\beta})$. \bigskip \noindent {\pc 5. Facets in the equilibrium crystal} \medskip \noindent The roughness of an interface should be apparent when considering the shape of the equilibrium crystal associated with the system. One knows that a typical equilibrium crystal at low temperatures has smooth plane facets linked by rounded edges and corners. The area of a particular facet decreases as the temperature is raised and the facet finally disappears at a temperature characteristic of its orientation. The reader will find information and references on equilibrium crystals in the review articles of Refs.\thinspace 20, 21, 23 and 24. It can be argued, as discussed below, that the roughening transition corresponds to the disappearance of the facet whose orientation is the same as that of the considered interface. The exactly solvable SOS models mentioned above, for which the function $\tau({\bf n})$ has been computed, are interesting examples of this behaviour (this subject has been reviewed in Ref.\thinspace 12, Chapter VII). For the three-dimensional Ising model, Bricmont {\it et al.} [25] have proved a correlation inequality which establish $\tau^{\rm step}$ as a lower bound to the one-sided derivative $\partial\tau(\theta) / \partial\theta |_{\theta=0^+} $ (here $\tau^{\rm step}=\tau^{\rm step}(0,1)$ and $\tau(\theta)=\tau(0,\sin\theta,\cos\theta)$). Thus $\tau^{\rm step}>0$ implies a kink in $\tau(\theta)$ at $\theta=0$ and, according to the Wulff construction, a facet is expected. In fact, $ \tau^{\rm step} $ should be equal to this derivative. This is reasonable, since the increment in surface tension of an interface tilted by an angle $\theta$, with respect to the surface tension of the rigid interface, can be approximately identified, for $\theta$ small, with the free energy of a ``gas of steps'' (the density of the steps being proportional to $\theta$). And, again, if the interaction between the steps can be neglected, the free energy of this gas can be approximated by the sum of the individual free energies of the steps. As a result of the methods described in Section 3, it is possible to study the statistical mechanics of this ``gas of steps'', and to derive the following result. \medskip \noindent \pc Theorem 4. \it For $TJ_2>0$, in the presence of a very small magnetic field $h>0$, is studied. Let $\Lambda$ be a square box of side $L$, with periodic boundary conditions, and let $H(\sigma)$ denote the energy of a configuration $\sigma\in\Omega_\Lambda=\{-1,1\}^\Lambda$. We suppose that the volume is sufficiently large, $L>(2J_1/h)^3$. A discrete time {\it stochastic dynamics} is then considered for this model. Namely, the Metropolis dynamics defined by the following updating rule: Given a configuration $\sigma$ at time $t$ one first chooses randomly a site $i\in\Lambda$ with uniform probability $1/\vert\Lambda\vert$. Then one flips the spin at site $i$ with probability $$ \exp ( -\beta \max\{H(\sigma^{(i)}) - H(\sigma)\thinspace, 0\}) $$ where $\sigma^{(i)}(j)=\sigma(j)$, whenever $j\ne i$, and $\sigma^{(i)}(j)= - \sigma(j)$, for $j=i$. This dynamics is {\it reversible} with respect to the Gibbs measure. The nucleation from a metastable state is studied for this model in the limit of very low temperatures ($h$ fixed). It turns out that the critical nucleus, as well as the configurations on a typical path to it, {\it differ from the Wulff shape} of an equilibrium droplet. The critical droplet is in fact a square of side $\ell^*=[2J_2/h]+1$ ($[\cdot]$ denotes the integer part), while the Wulff shape is a rectangle of sides proportional to $J_1,J_2$ (agreement could be expected, however, in the more customary region $T$ fixed, small, and $h\to0$). A {\it path} of the process is a sequence $\omega=\sigma_0,\sigma_1,\dots,\sigma_t,\dots$ of configurations in $\Omega_\Lambda$. We suppose that the process starts at the configuration $\sigma_0=(-)$ (all spins $\sigma_0(i)$ in $\Lambda$ equal to $-1$). We are interested in the first passage {}from the configuration $(-)$ to the configuration $(+)$, which takes place between the moments $\tau_{(-)}=\max \{t<\tau_{(+)} \mid \sigma_t=(-)\}$ and $\tau_{(+)}=\min \{t \mid \sigma_t=(+)\}$. The configurations $r(\ell_1,\ell_2)$, which have a rectangle of sides $\ell_1,\ell_2$ as unique Peierls contour, play a particular role in the process. They correspond to the local minima of the energy (in the sense that one spin flip increases the energy). Now, the probability that starting from a given local minimum $Q$ the system goes to a neighbouring local minimum $Q'$, is determined by the energy barrier $H(S)-H(Q)$, where $S$ is any configuration at which the energy on a path from $Q$ to $Q'$ reaches its maximum, but with the path chosen to minimalise it. In other words, the configurations in $S$ are the {\it local saddle points} for which the minimax $$ \min_{\omega: Q\to Q'} \max_{\sigma\in\omega} H(\sigma) $$ is attained (here $\omega: Q\to Q'$ denotes a generic path with successive spin flips starting from a configuration in $Q$ and ending at $Q'$). The considered probability is proportional to $\exp[-\beta(H(S)-H(Q))]$. On the other side, the system in the local minimum $Q$ is likely to ``stay'' in its basin of attraction for a time of order $\exp[\beta(H({\bar S})-H(Q))]$, where ${\bar S}$ is the local saddle point with lowest energy through which it can escape from the local minimum $Q$, not necessarily in ``the direction'' of $Q'$. These are the basic mechanisms which determine the local dynamics. The task is then to find the class of paths which describe the most probable evolution. Let us consider the probability of reaching a {\it global saddle point}, defined by the same minimax condition extended to all paths from the configuration $(-)$ to the configuration $(+)$. These configurations give rise to the {\it critical nucleus}. It can be seen that the set of all global saddle points coincides with the set $\cal P$ of all configurations having as unique contour a rectangle, of sides $\ell^*,\ell^*-1$, or $\ell^*-1,\ell^*$, with a unit square attached to one of its longer sides. The relative energy of any ${\bar\sigma}\in{\cal P}$ is $$ \Gamma = H({\bar\sigma})-H((-))=2(J_1+J_2)\ell^* - h ((\ell^*)^2 - \ell^* + 1) $$ It is proved that the first excursion from $(-)$ to $(+)$ passes through a configuration from ${\cal P}$ and the time needed for this to happen is of the order $\exp(\beta\Gamma)$. Introducing the time $\tau_{\cal P}=\min\{t>\tau_{(-)} \mid \sigma_t\in {\cal P}\}$, the precise statement can be formulated as follows. \medskip \noindent \pc Theorem 5. \it We have $$ \lim_{T\to 0} {\rm Prob}\ [\tau_{\cal P}<\tau_{(+)}] = 1 $$ and, moreover, for any $\epsilon>0$, $$ \lim_{T\to 0} {\rm Prob}\ [\exp(\beta(\Gamma-\epsilon)) < \tau_{\cal P} < \exp(\beta(\Gamma+\epsilon))] = 1 $$ \rm In addition, from the arguments in the proof of this result, one is getting very detailed information about a typical path followed by the process $\sigma_t$ during its first excursion from $(-)$ to $(+)$. \medskip \itemitem{\it a)} {\it First it passes through a monotonously growing sequence of subcritical rectangles $r(\ell_1,\ell_2)$, such that $\vert\ell_1 -\ell_2\vert = 0$ or $1$, up to the critical square $r(\ell^*,\ell^*)$.} \itemitem{\it b)} {\it After the vertical edge stays constant at the value $\ell^*$ while the horizontal edge grows up to $L$.} {\it Finally the vertical edge grows from $\ell^*$ to $L$.} \medskip The precise statements involve the notion of $\epsilon$-typical path, that is determined not only in terms of geometrical properties, but also with specified times of passage (by means of bounds analogous to those used above for $\tau_{\cal P}$) through certain configurations. The path is an $\epsilon$-typical path (for any given $\epsilon>0$) with a probability which tends to 1 when $T\to 0$. Finally, let us mention that Schonmann [32] has recently discussed the regime in which the temperature is kept fixed and the field $h>0$ is scaled to zero. As conjectured in Ref.\thinspace 33, for the Ising model in any dimension $d\ge 2$, if the temperature is low enough, the relaxation time goes in this regime as an exponential of $1/h^{d-1}$. Moreover, before a time which grows also as an exponential of $1/h^{d-1}$ the system stays in a metastable situation. \bigskip \noindent {\pc Acknowledgements}: It is a pleasure to thank Roman Koteck\'y for very valuable discussions. \bigskip \noindent {\pc References} \medskip \parindent=6mm \item{1.} R.L. Dobrushin, R. Koteck\'y and S.B. Shlosman: {The Wulff construction: a global shape from local interactions.} Providence: Amer. Math. Soc. 1992. \item{2.} R.L. Dobrushin, R. Koteck\'y and S.B. Shlosman: {J. Stat. Phys.} {\bf 72}, 1 (1993) \item{3.} C.E. Pfister: {Helv. Phys. Acta} {\bf 64}, 953 (1991) \item{4.} S. Miracle-Sole: Surface tension, step free energy and facets in the equilibrium crystal. Marseille: Preprint, 1994. \item{5.} S. Miracle-Sole: In: {Critical phenomena, Sitges international school}. J. Brey and R.B. Jones eds. Berlin: Springer, 1976. \item{6.} R.L. Dobrushin: {Theory Prob. Appl.} {\bf 17}, 582 (1972) \item{7.} J. Bricmont, J.L. Lebowitz, C.E. Pfister and E. Olivieri: {Commun. Math. Phys.} {\bf 66}, 1 (1979). \item{8.} J. Bricmont, J.L. Lebowitz and C.E. Pfister: {Commun. Math. Phys.} {\bf 66}, 21 (1979); {Commun. Math. Phys.} {\bf 69}, 267 (1979) \item{9.} P. Holick\'y, R. Koteck\'y and M. Zahradn\'\i k: {J. Stat. Phys.} {\bf 50}, 755 (1988) \item{10.} G. Gallavotti: {Commun. Math. Phys.} {\bf 27}, 103 (1972) \item{11.} J. Bricmont, J.L. Lebowitz and C.E. Pfister: {J. Stat. Phys.} {\bf 26}, 313 (1981) \item{12.} A. Messager, S. Miracle-Sole and J. Ruiz: {J. Stat. Phys.} {\bf 67}, 449-470 (1992) \item{13.} J. Bricmont, J.L. Lebowitz and C.E. Pfister: {Ann. Acad. Sci. New York} {\bf 337}, 214 (1980) \item{14.} J.L. Lebowitz and C.E. Pfister: {Phys. Rev. Lett.} {\bf 46}, 1031 (1981) \item{15.} R.L. Dobrushin and S.B. Shlosman: In: Ideas and methods in mathematical analysis, stochastics and applications. S. Albeverio, S.E. Fenstad, H. Holden and T. Lindstrom eds. Cambridge: University Press, 1991 \item{16.} J.D. Weeks, G.H. Gilmer and H.J. Leamy: {Phys. Rev. Lett.} {\bf 31}, 549 (1973) \item{17.} H. van Beijeren: {Commun. Math. Phys.} {\bf 40}, 1 (1975) \item{18.} J. Bricmont, J.R. Fontaine and J.L. Lebowitz: {J. Stat. Phys.} {\bf 29}, 193 (1982) \item{19.} J. Fr\"ohlich and T. Spencer: {Commun. Math. Phys.} {\bf 81}, 527 (1981) \item{20.} H. van Beijeren and I. Nolden: In: {Topics in current Physics, Vol. 43}. W. Schommers and P. von Blackenhagen eds. Berlin: Springer 1987. \item{21.} D.B. Abraham: In: {Phase transitions and critical phenomena, Vol. 10}. C. Domb and J.L. Lebowitz eds. London: Academic Press 1986. \item{22.} J.E. Avron, H. van Beijeren, L.S. Shulman and R.K.P. Zia: {J. Phys. A: Mat. Gen.} {\bf 15}, L81 (1982) \item{23.} C. Rottman and M. Wortis: {Phys. Rep.} {\bf 103}, 59 (1984) \item{24.} R. Koteck\'y: In: {IX international congress of mathematical Physics}. B. Simon A. Truman and I.M. Davies eds. Bristol: Adam Hilger 1889. \item{25.} J. Bricmont, A. El Mellouki and J. Fr\"ohlich: {J. Stat. Phys.} {\bf 42}, 743 (1986) \item{26.} S. Miracle-Sole: {In preparation} \item{27.} R. Koteck\'y and E. Olivieri: {J. Stat. Phys.} {\bf 70}, 1121 (1993). \item{28.} R. Koteck\'y and E. Olivieri: {J. Stat. Phys.} {\bf 75}, 409 (1994) \item{29.} E.J. Neves and R.H. Schonmann: {Commun. Math. Phys.} {\bf 137}, 209 (1991) \item{30.} R.H. Schonmann: {Commun. Math. Phys.} {\bf 147}, 231 (1991) \item{31.} M. Cassandro, A. Galves, E. Olivieri and M.E. Vares: {J. Stat. Phys.} {\bf 35}, 603 (1984) \item{32.} R.H. Schonmann: In: {Probability and phase transitions}. G. Grimmet, ed. Dordrecht: Kluwer Academic Publishers, 1994. \item{33.} M. Aizenman and J.L. Lebowitz: {J. Phys. A: Math. Gen.} {\bf 21}, 3801 (1988) \end