BODY %FONTS \font\twelverm=cmr12 \font\twelvei=cmmi12 \font\twelvesy=cmsy10 \font\twelvebf=cmbx12 \font\twelvett=cmtt12 \font\twelveit=cmti12 \font\twelvesl=cmsl12 \font\ninerm=cmr9 \font\ninei=cmmi9 \font\ninesy=cmsy9 \font\ninebf=cmbx9 \font\ninett=cmtt9 \font\nineit=cmti9 \font\ninesl=cmsl9 \font\eightrm=cmr8 \font\eighti=cmmi8 \font\eightsy=cmsy8 \font\eightbf=cmbx8 \font\eighttt=cmtt8 \font\eightit=cmti8 \font\eightsl=cmsl8 \font\sixrm=cmr6 \font\sixi=cmmi6 \font\sixsy=cmsy6 \font\sixbf=cmbx6 \catcode`@=11 % we will access private macros of plain TeX (carefully) \newskip\ttglue %MACRO TWELVEPOINT \def\twelvepoint{\def\rm{\fam0\twelverm}% switch to 12-point type \textfont0=\twelverm \scriptfont0=\ninerm \scriptscriptfont0=\sevenrm \textfont1=\twelvei \scriptfont1=\ninei \scriptscriptfont1=\seveni \textfont2=\twelvesy \scriptfont2=\ninesy \scriptscriptfont2=\sevensy \textfont3=\tenex \scriptfont3=\tenex \scriptscriptfont3=\tenex \textfont\itfam=\twelveit \def\it{\fam\itfam\twelveit}% \textfont\slfam=\twelvesl \def\sl{\fam\slfam\twelvesl}% \textfont\ttfam=\twelvett \def\tt{\fam\ttfam\twelvett}% \textfont\bffam=\twelvebf \scriptfont\bffam=\ninebf \scriptscriptfont\bffam=\sevenbf \def\bf{\fam\bffam\twelvebf}% \tt \ttglue=.5em plus.25em minus.15em \normalbaselineskip=15pt \setbox\strutbox=\hbox{\vrule height10pt depth5pt width0pt}% \let\sc=\tenrm \let\big=\twelvebig \normalbaselines\rm} %MACRO TENPOINT \def\tenpoint{\def\rm{\fam0\tenrm}% switch to 10-point type \textfont0=\tenrm \scriptfont0=\sevenrm \scriptscriptfont0=\fiverm \textfont1=\teni \scriptfont1=\seveni \scriptscriptfont1=\fivei \textfont2=\tensy \scriptfont2=\sevensy \scriptscriptfont2=\fivesy \textfont3=\tenex \scriptfont3=\tenex \scriptscriptfont3=\tenex \textfont\itfam=\tenit \def\it{\fam\itfam\tenit}% \textfont\slfam=\tensl \def\sl{\fam\slfam\tensl}% \textfont\ttfam=\tentt \def\tt{\fam\ttfam\tentt}% \textfont\bffam=\tenbf \scriptfont\bffam=\sevenbf \scriptscriptfont\bffam=\fivebf \def\bf{\fam\bffam\tenbf}% \tt \ttglue=.5em plus.25em minus.15em \normalbaselineskip=12pt \setbox\strutbox=\hbox{\vrule height8.5pt depth3.5pt width0pt}% \let\sc=\eightrm \let\big=\tenbig \normalbaselines\rm} %MACRO NINEPOINT \def\ninepoint{\def\rm{\fam0\ninerm}% switch to 9-point type \textfont0=\ninerm \scriptfont0=\sixrm \scriptscriptfont0=\fiverm \textfont1=\ninei \scriptfont1=\sixi \scriptscriptfont1=\fivei \textfont2=\ninesy \scriptfont2=\sixsy \scriptscriptfont2=\fivesy \textfont3=\tenex \scriptfont3=\tenex \scriptscriptfont3=\tenex \textfont\itfam=\nineit \def\it{\fam\itfam\nineit}% \textfont\slfam=\ninesl \def\sl{\fam\slfam\ninesl}% \textfont\ttfam=\ninett \def\tt{\fam\ttfam\ninett}% \textfont\bffam=\ninebf \scriptfont\bffam=\sixbf \scriptscriptfont\bffam=\fivebf \def\bf{\fam\bffam\ninebf}% \tt \ttglue=.5em plus.25em minus.15em \normalbaselineskip=11pt \setbox\strutbox=\hbox{\vrule height8pt depth3pt width0pt}% \let\sc=\sevenrm \let\big=\ninebig \normalbaselines\rm} %MACRO EIGHTPOINT \def\eightpoint{\def\rm{\fam0\eightrm}% switch to 8-point type \textfont0=\eightrm \scriptfont0=\sixrm \scriptscriptfont0=\fiverm \textfont1=\eighti \scriptfont1=\sixi \scriptscriptfont1=\fivei \textfont2=\eightsy \scriptfont2=\sixsy \scriptscriptfont2=\fivesy \textfont3=\tenex \scriptfont3=\tenex \scriptscriptfont3=\tenex \textfont\itfam=\eightit \def\it{\fam\itfam\eightit}% \textfont\slfam=\eightsl \def\sl{\fam\slfam\eightsl}% \textfont\ttfam=\eighttt \def\tt{\fam\ttfam\eighttt}% \textfont\bffam=\eightbf \scriptfont\bffam=\sixbf \scriptscriptfont\bffam=\fivebf \def\bf{\fam\bffam\eightbf}% \tt \ttglue=.5em plus.25em minus.15em \normalbaselineskip=9pt \setbox\strutbox=\hbox{\vrule height7pt depth2pt width0pt}% \let\sc=\sixrm \let\big=\eightbig \normalbaselines\rm} %MACRO BIG \def\twelvebig#1{{\hbox{$\textfont0=\twelverm\textfont2=\twelvesy \left#1\vbox to10pt{}\right.\n@space$}}} \def\tenbig#1{{\hbox{$\left#1\vbox to8.5pt{}\right.\n@space$}}} \def\ninebig#1{{\hbox{$\textfont0=\tenrm\textfont2=\tensy \left#1\vbox to7.25pt{}\right.\n@space$}}} \def\eightbig#1{{\hbox{$\textfont0=\ninerm\textfont2=\ninesy \left#1\vbox to6.5pt{}\right.\n@space$}}} \def\tenmath{\tenpoint\fam-1 } %for 10-point math in 9-point territory %%%%%%%%%%%%%%% FORMATO \magnification=\magstephalf\hoffset=0.cm \voffset=1truecm\hsize=16.5truecm \vsize=21.truecm \baselineskip=14pt plus0.1pt minus0.1pt \parindent=12pt \lineskip=4pt\lineskiplimit=0.1pt \parskip=0.1pt plus1pt \def\ds{\displaystyle}\def\st{\scriptstyle}\def\sst{\scriptscriptstyle} \font\seven=cmr7 %%%%%%%%%%%%%%%% GRECO \let\a=\alpha \let\b=\beta \let\c=\chi \let\d=\delta \let\e=\varepsilon \let\f=\varphi \let\g=\gamma \let\h=\eta \let\k=\kappa \let\l=\lambda \let\m=\mu \let\n=\nu \let\o=\omega \let\p=\pi \let\ph=\varphi \let\r=\rho \let\s=\sigma \let\t=\tau \let\th=\vartheta \let\y=\upsilon \let\x=\xi \let\z=\zeta \let\D=\Delta \let\F=\Phi \let\G=\Gamma \let\L=\Lambda \let\Th=\Theta \let\O=\Omega \let\P=\Pi \let\Ps=\Psi \let\Si=\Sigma \let\X=\Xi \let\Y=\Upsilon %%%%%%%%%%%%%%% DEFINIZIONI LOCALI \let\ciao=\bye \def\fiat{{}} \def\pagina{{\vfill\eject}} \def\\{\noindent} \def\bra#1{{\langle#1|}} \def\ket#1{{|#1\rangle}} \def\ie{\hbox{\it i.e.\ }} \let\ig=\int \let\io=\infty \let\i=\infty \let\dpr=\partial \def\V#1{\vec#1} \def\Dp{\V\dpr} \def\tende#1{\vtop{\ialign{##\crcr\rightarrowfill\crcr \noalign{\kern-1pt\nointerlineskip} \hskip3.pt${\scriptstyle #1}$\hskip3.pt\crcr}}} \def\otto{{\kern-1.truept\leftarrow\kern-5.truept\to\kern-1.truept}} \def\Z{{\bf Z^2}} \def\supnorm#1{\vert#1\vert_\infty} \def\norm#1{\vert\vert#1\vert\vert} \def\R{{\cal R}} \def\eop{\hfill\bbox$\,$} \def\bbox{\vrule height 1.5ex width 0.6em depth 0ex } %%%%%%%%%%%%%%%%%%%%% Numerazione pagine \def\data{\number\day/\ifcase\month\or gennaio \or febbraio \or marzo \or aprile \or maggio \or giugno \or luglio \or agosto \or settembre \or ottobre \or novembre \or dicembre \fi/\number\year} %%\newcount\tempo %%\tempo=\number\time\divide\tempo by 60} \setbox200\hbox{$\scriptscriptstyle \data $} \newcount\pgn \pgn=1 \def\foglio{\number\numsec:\number\pgn \global\advance\pgn by 1} \def\foglioa{A\number\numsec:\number\pgn \global\advance\pgn by 1} %\footline={\rlap{\hbox{\copy200}\ $\st[\number\pageno]$}\hss\tenrm %\foglio\hss} %\footline={\rlap{\hbox{\copy200}\ $\st[\number\pageno]$}\hss\tenrm %\foglioa\hss} % %%%%%%%%%%%%%%%%% EQUAZIONI CON NOMI SIMBOLICI %%% %%% per assegnare un nome simbolico ad una equazione basta %%% scrivere \Eq(...) o, in \eqalignno, \eq(...) o, %%% nelle appendici, \Eqa(...) o \eqa(...): %%% dentro le parentesi e al posto dei ... %%% si puo' scrivere qualsiasi commento; %%% per assegnare un nome simbolico ad una figura, basta scrivere %%% \geq(...); per avere i nomi %%% simbolici segnati a sinistra delle formule e delle figure si deve %%% dichiarare il documento come bozza, iniziando il testo con %%% \BOZZA. Sinonimi \Eq,\EQ,\EQS; \eq,\eqs; \Eqa,\Eqas;\eqa,\eqas. %%% All' inizio di ogni paragrafo si devono definire il %%% numero del paragrafo e della prima formula dichiarando %%% \numsec=... \numfor=... (brevetto Eckmannn); all'inizio del lavoro %%% bisogna porre \numfig=1 (il numero delle figure non contiene la sezione. %%% Si possono citare formule o figure seguenti; le corrispondenze fra nomi %%% simbolici e numeri effettivi sono memorizzate nel file \jobname.aux, che %%% viene letto all'inizio, se gia' presente. E' possibile citare anche %%% formule o figure che appaiano in altri files, purche' sia presente il %%% corrispondente file .aux; basta includere all'inizio l'istruzione %%% \include{nomefile} %%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \global\newcount\numsec\global\newcount\numfor \global\newcount\numfig \gdef\profonditastruttura{\dp\strutbox} \def\senondefinito#1{\expandafter\ifx\csname#1\endcsname\relax} \def\SIA #1,#2,#3 {\senondefinito{#1#2} \expandafter\xdef\csname #1#2\endcsname{#3} \else \write16{???? ma #1,#2 e' gia' stato definito !!!!} \fi} \def\etichetta(#1){(\veroparagrafo.\veraformula) \SIA e,#1,(\veroparagrafo.\veraformula) \global\advance\numfor by 1 \write15{\string\FU (#1){\equ(#1)}} \write16{ EQ \equ(#1) == #1 }} \def \FU(#1)#2{\SIA fu,#1,#2 } \def\etichettaa(#1){(A\veroparagrafo.\veraformula) \SIA e,#1,(A\veroparagrafo.\veraformula) \global\advance\numfor by 1 \write15{\string\FU (#1){\equ(#1)}} \write16{ EQ \equ(#1) == #1 }} \def\getichetta(#1){Fig. \verafigura \SIA e,#1,{\verafigura} \global\advance\numfig by 1 \write15{\string\FU (#1){\equ(#1)}} \write16{ Fig. \equ(#1) ha simbolo #1 }} \newdimen\gwidth \def\BOZZA{ \def\alato(##1){ {\vtop to \profonditastruttura{\baselineskip \profonditastruttura\vss \rlap{\kern-\hsize\kern-1.2truecm{$\scriptstyle##1$}}}}} \def\galato(##1){ \gwidth=\hsize \divide\gwidth by 2 {\vtop to \profonditastruttura{\baselineskip \profonditastruttura\vss \rlap{\kern-\gwidth\kern-1.2truecm{$\scriptstyle##1$}}}}} } \def\alato(#1){} \def\galato(#1){} \def\veroparagrafo{\number\numsec}\def\veraformula{\number\numfor} \def\verafigura{\number\numfig} %\def\geq(#1){\getichetta(#1)\galato(#1)} \def\Eq(#1){\eqno{\etichetta(#1)\alato(#1)}} \def\eq(#1){\etichetta(#1)\alato(#1)} \def\Eqa(#1){\eqno{\etichettaa(#1)\alato(#1)}} \def\eqa(#1){\etichettaa(#1)\alato(#1)} \def\eqv(#1){\senondefinito{fu#1}$\clubsuit$#1\else\csname fu#1\endcsname\fi} \def\equ(#1){\senondefinito{e#1}\eqv(#1)\else\csname e#1\endcsname\fi} \let\EQS=\Eq\let\EQ=\Eq \let\eqs=\eq \let\Eqas=\Eqa \let\eqas=\eqa %%%%%%%%%%%%%%%%%% Numerazione verso il futuro ed eventuali paragrafi %%%%%%% precedenti non inseriti nel file da compilare \def\include#1{ \openin13=#1.aux \ifeof13 \relax \else \input #1.aux \closein13 \fi} \openin14=\jobname.aux \ifeof14 \relax \else \input \jobname.aux \closein14 \fi \openout15=\jobname.aux %%%%%%%%%%%%%%%%%%%%%%%%%%%% %\BOZZA \footline={\rlap{\hbox{\copy200}\ $\st[\number\pageno]$}\hss\tenrm \foglio\hss} \def\refj#1#2#3#4#5#6#7{\parindent 2.2em \item{[{\bf #1}]}{\rm #2,} {\it #3\/} {\rm #4} {\bf #5} {\rm #6} {(\rm #7)}} %\input formato.tex \numsec=0\numfor=1 \tolerance=10000 \font\ttlfnt=cmcsc10 scaled 1200 %small caps \font\bit=cmbxti10 %bold italic text mode % Author. Initials then last name in upper and lower case % Point after initials % \def\author#1 {\vskip 18pt\tolerance=10000 \noindent\centerline{\ttlfnt#1}\vskip 1cm} % % Address % \def\address#1 {\vskip 4pt\tolerance=10000 \noindent #1\vskip 0.5cm} % % Abstract % \def\abstract#1 { \noindent{\bf Abstract.\ }#1\par} % %\vskip 1cm \centerline{\ttlfnt On the Two Dimensional Dynamical Ising Model} \centerline{\ttlfnt In the Phase Coexistence Region}\vskip 0.5cm \author{F. Martinelli $^{\dag}$ $^{\ddag}$} \address{\ninerm \dag Dipartimento di Matematica, III Universit\`a di Roma, Italy \hfill\break{\ddag Istituto per le Applicazioni del Calcolo "Mauro Picone", CNR, Roma Italy}\hfill\break{ e-mail: martin@mat.uniroma1.it}} \abstract{\ninerm We consider a Glauber dynamics reversible with respect to the two dimensional Ising model in a finite square of side $L$, in the absence of an external field and at large inverse temperature $\beta$. We first consider the gap in the spectrum of the generator of the dynamics in two different cases: with plus and open boundary condition. We prove that, when the symmetry under global spin flip is broken by the boundary conditions, the gap is much larger than the case in which the symmetry is present. For this latter we compute exactly the asymptotics of \hskip 0.5cm $-{1\over \beta L}\log (\hbox{gap})$\hskip 0.5cm as $L\to\infty$ and show that it coincides with the surface tension along one of the coordinat axes. As a consequence we are able to study quite precisely the large deviations in time of the magnetization and to obtain an upper bound on the spin-spin time correlation in the infinite volume plus phase. Our results establish a connection between the dynamical large deviations and those of the equilibrium Gibbs measure studied by Shlosman in the framework of the rigorous description of the Wulff shape for the Ising model. Finally we show that, in the case of open boundary conditions, it is possible to rescale the time with $L$ in such a way that, as $L\to \infty$, the finite dimensional distributions of the time rescaled magnetization converge to those of a symmetric continuous time Markov chain on the two state space $\{-m^*(\beta ),m^*(\beta )\}$, $m^*(\beta )$ being the spontaneous magnetization. Our methods rely upon a novel combination of techniques for bounding from below the gap of symmetric Markov chains on complicate graphs, developed by Jerrum and Sinclair in their Markov chain approach to hard computational problems, and the idea of introducing "block Glauber dynamics" instead of the standard single site dynamics, in order to put in evidence more effectively the effect of the boundary conditions in the approach to equilibrium.} \vskip 1cm\noindent {\eightrm Work partially supported by grant SC1-CT91-0695 of the Commission of European Communities} \vfill \eject %\input formato.tex \tolerance=10000 \numsec=0 \numfor=1 {\bf Section 0}\par \centerline{\bf Introduction }\bigskip In the last years there have been very important progresses in the rigorous analysis of Glauber dynamics (see section 1 for a precise definition) for classical lattice spin systems when the thermodynamic parameters are such that the static system, described by the usual Gibbs measure $$\mu\,=\, {\exp (-\beta H)\over Z}$$ does not undergo a phase transition in the thermodynamic limit.\par In particular we refer the reader to the series of papers by Stroock and Zegarlinski (see [SZ] and references therein), by Olivieri and myself (see [MO1], [MO2]), Olivieri, Schonmann and myself [MOS], Yau and Shin Lin [SLY] for the proof, under various mixing conditions on the Gibbs measure $\mu$, of the exponential (in time) relaxation to equilibrium, represented by $\mu$ itself, in finite or infinite volume, of the associated Glauber dynamics, and to the works by Schonmann (see [Sch1] and references therein), Kotecky and Olivieri [KO1], [K02], Scoppola [Sc] for detail description of the metastable behaviour of Glauber dynamics for Ising type models close to the line of first order phase transition.\par It is important to emphasize that some of the results in the above works cover most of the one phase region, going sometimes, e.g. in ferromagnetic systems, arbitrary close to the critical point (see [MO1] and [MOS]). \par A natural question arises as to what happens when the thermodynamic parameters are such that we do have a phase transition in the thermodynamic limit. To be more concrete, let us consider the usual Ising model in d-dimensions, $d\geq 2$, in the absence of an external field, described by the formal (normalized) energy function: $$H(\s )\;=\;-{1\over 2}\sum_{x,y\in {\bf Z}^d\atop \vert x-y\vert =1}(\s (x)\s (y)-1)\qquad \s\in \{-1,1\}^{{\bf Z}^d}\Eq(0.1)$$ and let us suppose that the inverse temperature $\beta$ is larger (actually in all rigorous results much larger) than the critical value $\beta_c$.\par Then, as it is well known (see e.g. [Li]), any associated infinite volume Glauber dynamics is not ergodic and it is rather natural to ask how this absence of ergodicity is reflected if we look at the dynamics in a finite, but large cube $V_L$ of side $L$, where ergodicity is never broken.\par A first partial answer was provided by Thomas [T] few years ago. He showed that, if the symmetry of $H(\s )$ under global spin flip is not broken by the boundary conditions on the exterior of the cube $V_L$, then the relaxation time to equilibrium, that in a first approximation can be taken equal to the inverse of the gap in the spectrum of the generator $L_{V_L}$ of the dynamics, diverges, as $L\to \infty$, at least as an exponential of the {\it surface} $L^{d-1}$.\par The reason for such a result is the following. When the symmetry is not broken e.g. when boundary conditions are open (i.e. absent) or periodic, then the energy landscape determined by the function $H(\s )$ has only two absolute minima, corresponding to the two configurations identically equal to either $+1$ or to $-1$. Thus the dynamics started e.g. from all minuses, in order to relax to equilibrium, has to reach the neighborood of the opposite minimum by necessarily crossing the set of configurations of zero magnetization. Since the Gibbs measure gives to the latter a weight of the order of a negative exponential of the {\it surface} (see e.g. [Sch2]) a kind of bottleneck is present and the result follows by a rather simple argument (see the first part of the proof of theorem 4.1 below).\par The same reasoning also suggests that, if the symmetry is broken by the boundary conditions, e.g. by fixing equal to $+1$ all spins outside $V_L$, then the relaxation time should be much shorter than in the previous case since there should be no bottlenecks to cross. Equilibrium is, in this case, induced by the boundary conditions by means of some sort of plus spins wave, initially attached to the boundary and shrinking to zero as time goes on.\par The interesting but unproven conjecture, is that, at least in two dimensions with plus boundary conditions, the relaxation time will diverge, as $L\to \infty$, like $L^{2}$. The proof of the above conjecture would have some very nice consequences on the equal site time correlations function of the infinite volume dynamics started in the plus phase, for which Fisher and Huse [FH] predicted, using the above conjecture, a stretched exponential decay of the form $\exp (-\sqrt {t})$ (see also [Og] for numerical simulations and [M] for further discussion).\par In this paper we consider the above and other related questions for the two dimensional Ising model at very low temperature without external field. For some less precise results in arbitrary dimensions see the remark after theorem 4.2\par In section 3 we prove a {\it lower} bound on the gap in the spectrum of the generator $L_{V_L}$ of the Glauber dynamics with $+$ boundary conditions of the form: $$\hbox{gap}(L_{V_L})\;\geq\;\exp (-C\beta L^{{1\over 2}+\epsilon})\qquad \e\in (0,{1\over 2}]\Eq(0.2)$$ which, although gives a bound on the relaxation time which is far from the conjectured $L^{2}$ law, is in any case much larger than the {\it upper bound } obtained by Thomas without the plus boundary conditions.\par As a consequence we derive an upper bound of the form $$\exp (-\log (t)^\a)\qquad \a\in [0,2)$$ on the equal site time correlation function of the infinite volume dynamics started in the plus phase.\par In section 4 we compute exactly the asymptotics of the gap with open boundary conditions. More precisely we obtain, for any $\e \in (0,{1\over 4}]$, any $\b$ large enough and any $L$: $$\exp (-\beta \t (\beta )L\,-\,C\beta L^{{1\over 2}+\e})\,\leq\,\hbox{gap}(L_{V_L})\,\leq\, \exp (-\beta \t (\beta )L\,+\,C\beta L^{{1\over 2}+\e})$$ where $\t (\b )$ is the surface tension in the direction of e.g. the horizontal axis. As a byproduct of the proof of this result, we show that the bound \equ(0.2) is valid even if the plus boundary conditions are added on only one side of the square $V_L$.\par The proofs of the above two results follow two very similar steps:\bigskip\noindent Step 1: we prove the sought result for a generalized Glauber dynamics in which single sites are replaced by suitable blocks. This mean that, given apriori a covering $\{Q_i\}$ of $V_L$, at each updating of the dynamics the spin configuration is changed in only one block $Q_i$ and there it is replaced by the equilibrium Gibbs measure of the block given the configuration outside it. It turns out that a convenient choice of the blocks in our case consists of long and thin overlapping rectangles with basis $L$ and height $L^{{1\over 2}+\e}$, $0<\e<<1$.\medskip\noindent Step 2: we relate the gap of the single site Glauber dynamics to that of the generalized block dynamics in such a way that the estimates obtained in step 1 are not significantly changed.\bigskip The above way to attack the problem is not entirely new; it was in fact introduced long ago by Holley [H] to prove exponential convergence to equilibrium in the one phase region. One has in fact that, if the system is away from the phase transition region and if the bloks are overlapping large enough (depending on the thermodynamic parameters) cubes, the blocks Glauber dynamics behaves as a very high temperature single site Glauber dynamics, i.e. an almost independent systems for which the discussion of the approach to equilibrium is a relatively easy task (see section 4 of [MO2]).\par While we accomplish the first step via a very natural probabilistic construction, the second, rather crucial, step is carried out via the application to our contest of a clever geometric technique introduced by Jerrum and Sinclair [JS1], [JS2], Sinclair [Si] (see also [DS]), to estimate from {\it below} the gap in the spectrum of symmetric Markov chain on complicate graphs. They invented their technology while working on a stochastic algorithm approach to compute the partition function $Z$ of the Ising model and the permanent of a large matrix in a time {\it polynomial} in the size of the problem.\par Such a technique, which is illustrated in our case in a self contained way in section 2, gives in a very natural way a {\it lower} bound on the gap of the generator of the dynamics in a rectangle $R$ with shortest side $l$, with or without boundary conditions, of the form: $$\hbox{gap}(L_R)\;\geq\;\exp (-\beta C l)$$ Moreover, if the blocks $Q_i$ of the generalized Glauber dynamics are a suitable translations of the rectangle $R$, then: $$\hbox{gap(Glauber)}\;\geq\;\exp (-\beta C l)\hbox{gap(Generalized Glauber)}$$ It is worthwhile to mention that our proof of step 1 is constructive in the sense that it indicates how actually the system reaches equilibrium: by simply propagating the plus boundary conditions in the bulk if these are present and the initial configuration is e.g. all minuses, or by creating inside the starting phase, via a large fluctuation, an almost horizontal (or vertical) interface close e.g. to the bottom side of $V_L$ which afterwards rigidly moves to the opposite side until the other phase has invaded the whole volume.\par Once we have a precise control on the relaxation time with open boundary conditions, we can study in details the large fluctuations of the magnetization $$m(\s_t)\,=\,{1\over L^2}\sum_x\s_t (x)$$ by considering for example the hitting time $\t_\rho$ of the set $$M_\rho\;=\;\{\,\s ;\;m(\s )\,=\,\rho\,\}\qquad \rho\,\in\,\{-m^*(\b ),m^*(\b )\}$$ with $m^*(\b )$ the spontaneous magnetization.\par In section 5 we show that, if the starting configuration is distributed according to the equilibrium measure restricted to the "phase" of positive magnetization or if it is identically equal to plus one, then the expected value $E(\t_\rho )$ of the hitting time $\t_\rho$ is of the order: $$E(\t_\rho )\;\approx\;\exp( \b L\psi (\rho\vee 0 ))$$ where the rate function $\psi (\rho )$ is the same as for the static problem: $$\mu_{V_L} (m(\s )\,=\,\rho )\;\approx\;\exp( -\b L\psi (\rho))$$ and it has been computed by Shlosman [Sh] in the framework of the rigorous description of the Wulff shape for the Ising model carried out by Dobrushin,Koteck\'y and Shlosman [DKS]. We also show that the hitting time $\t_\rho$ rescaled by roughly its average converges, as $L\to\infty$, to an exponential time of mean one.\par It is important to outline that the typical configurations of the equilibrium Gibbs measure under the condition $\{m(\s )\,=\,\rho \}$ have a very precise geometric structure related to the Wulff shape with open boundary conditions (see [Sh]). Thus, the fact that the rate function for $E(\t_\rho )$ is the same as in the static problem, suggests that, when the system, started in the positively magnetized "phase", reaches for the first time the set $M_\rho$, it does it by forming a droplet of the right volume and with the correct Wulff shape. We hope to come back in a future work on this and related problems.\par A key step in the discussion of the above problems is the proof, based on the results of section 3, that the relaxation time inside a single "phase" is much shorter than the typical values of the hitting time $\t_\rho$ (see proposition 5.2 for a precise statement).\par This last result indicates that the gap in the spectrum of the generator restricted to the invariant subspace of the functions even with respect to global spin flip is much larger than the true gap; unfortunately we do not have any precise statement in this direction.\par Finally in section 6 we complete the analysis of the time evolution of the magnetization by showing that, if the time is scaled with $L$ in such a way that on the new unit of time the system is likely to have jumped from one phase to the other, then the finite dimensional distributions of the time scaled magnetization converge, as $L\to\infty$, to those of a continuous time Markov chain on the two states space $\{-m^*(\b ),m^*(\b )\}$ with unitary jump rate.\par The rest of the paper contains a preliminary section, section 1, where all the necessary definitions are given together with the required results on Wulff shape, cluster expansion and so forth. The proofs of various technical results for the Ising model have been collected in an appendix.\bigskip\noindent {\bf Acknowledgments}\par I am particularly in debt with Pablo Ferrari, who was involved in this work at its early stages, for many constructive discussions and comments concerning the material in sections 5 and 6, particularly theorems 5.2 and 6.1. I would also like to warmly thank A.S Sznitman, E.Bolthausen and J.Moser for their kind invitation to E.T.H. where the part of this work was carried out, and H.Kesten, G.Grimmet and M. Barlow for inviting me at the Newton Institute in Cambridge, within the special program "Random Spatial Processes", where I had the opportunity to discuss about this work and its possible extensions with S.Shlosman, R.Kotecky, A. Mazel and R.Schonmann. \pagina %\input formato.tex \numsec=1 \numfor=1 {\bf Section 1}\par \centerline{\bf Preliminaries}\bigskip In this section we precisely define the model and the random dynamics that will be the object of study in the next sections. \bigskip $\S 1$\hskip 1cm The Ising Model in a Finite Set.\medskip\noindent Let $\Z$ be the usual two dimensional square lattice with sites $x\,=\,(x_1,x_2)$, equipped with the norm $\norm{x}\,=\,\vert x_1\vert \,+\,\vert x_2\vert$. We will sometime consider $\Z$ as a graph with vertices the sites $x\in \Z$ and edges all pairs of sites $x$ and $y$ such that $\norm{x-y}\,=\,1$. We will use the notation $\sigma$ to denote a generic element of the set $\O_{\Z}\;=\;\{-1,+1\}^{\Z}$; whenever $V\subset \Z$ we use the notation $\s_V\;=\,\{\s (x),\;x\in V\}$ to denote the restriction of $\s$ to the set $V$ and $\O_V$ to denote the set of them. \par Given $V\subset \Z$, we define the interior and exterior boundaries of $V$ as : $$\partial_{int}V\,\equiv\,\{\,x\,\in V\,;\;\exists \,y\notin V\,;\quad \norm{x-y}\,=\,1\}$$ $$\partial_{ext}V\,\equiv\,\{\,x\,\notin V\,;\;\exists \,y\in V\,;\quad \norm{x-y}\,=\,1\}$$ and the boundary $\partial V$ as: $$\partial V\;=\;\{(x,y);\;x\in \,\partial_{int}V,\;y\,\in\,\partial_{ext}V\;\quad \norm{x-y}\,=\,1\,\}$$ We also denote by $\vert V\vert $ the cardinality of $V$.\par Next, for any finite subste $V$ of $\Z$, we define the energy $H_V^{U^{\partial V},\t}(\s_V )$ of a configuration $\s_V\,\in \,\O_V$ with boundary conditions $\tau$ outside $V$, $\t \in \{-1,+1\}^{\Z}$, and boundary coupling $0\,\leq\,U^{\partial V}(x,y)\,\leq\,1$, $(x,y)\,\in \,\partial V$, as: $$H_V^{U^{\partial V},\t}(\s_V )\;=\;$$ $$-{1\over 2}\sum_{\vbox{\eightpoint{\hbox{$x,y\in V$} \hbox{$\norm{x-y}=1$}}}}(\s_V(x)\s_V(y)\,-\,1)\;-\;\sum_ {(x,y)\,\in \,\partial V} U^{\partial V}(x,y)(\s_V(x)\t (y)\,-\,1)\Eq(1.1)$$ and the associated Gibbs probability measure at inverse temperature $\beta$: $$\mu_V^{U^{\partial V},\t}(\s_V)\;=\;Z(V,U^{\partial V},\t)^{-1}\, \exp (-\beta H_V^{U^{\partial V},\t}(\s_V)) \Eq(1.2)$$ where the partition function $ Z(V,U^{\partial V},\t)$ is given by $$Z(V,U^{\partial V},\t)\;=\;\sum_{\s_V} \exp (-\beta H_V^{U^{\partial V},\t}(\s_V))\Eq(1.3)$$ If the boundary condition $\t$ is the special configuration $\t (x)\,=\,1\;\forall \;x\in \Z$, then in all our notation the superscript $\t$ will be replaced by a simple $+$. Notice that the $-1$ appearing in the definition of the energy $H_V^{U^{\partial V},\t}(\s_V )$ fixes equal to zero the energy with plus boundary conditions of the configuration identically equal to plus one.\par We also set, for any function $f\,:\,\O_V\,\to\,{\bf R}$, $$\mu_V^{U^{\partial V},\t}(f)\,=\,\sum_{\s_V} \mu_V^{U^{\partial V},\t}(\s_V)f(\s_V)$$ Although, for technical reasons, it will be convenient to consider cases in which the boundary coupling $U^{\partial V}(x,y)$ does depend on $x$ and $y$ and it is for example equal to plus one along some parts of the external boundary of $V$ and positive but very weak along some other parts of the boundary, the most typical choices of $U^{\partial V}$ will be either $U^{\partial V}$ identically equal to one, in which case the Gibbs measure \equ(1.2) is the usual Ising model in the set $V$ with $\t$ boundary conditions, or $U^{\partial V}$ identically equal to zero which corresponds to the Ising model with open boundary conditions. In both cases the (cumbersome) notation $\mu_V^{U^{\partial V},\t}$, $Z(V,U^{\partial V},\t)$ will be replaced by the more natural ones $\mu_V^{\t}$, $\mu_V^{\emptyset}$, $Z(V,\t)$, $Z(V,\emptyset )$ respectively.\par As a next step we recall some monotonicity properties enjoied by the Gibbs measure $\mu_V^{U^{\partial V},\t}$, which easily follow from the well known FKG inequalities (see [FKG]), which will play a crucial role in the next sections.\par Given two configurations $\t_1$, $\t_2$ in $\O_{\Z}$, we say that $\t_1\,\leq \,\t_2$ iff $$\t_1 (x)\,\leq \,\t_2(x)\;\;\forall \; x\in \Z$$ and similarly for $\s_V, \,\s_V'\;\in \O_V$. Then, for any pair of finite subsets $V_1\, \subset \,V_2 $, any pair of boundary coupling $U_1^{\partial V_1}(x,y),\,U_2^{\partial V_1}(x,y)$, and boundary conditions $\t_1,\,\t_2$ such that $$U_1^{\partial V_1}(x,y)\t_1(y)\,\leq \, U_2^{\partial V_1}(x,y)\t_2(y)\quad \forall \;(x,y)\,\in \, \partial V_1$$ and any function $f\,:\,\O_{V_1}\,\to\,{\bf R}$ which is increasing with respect to the above partial order, we have: $$\mu_{V_1}^{U_1^{\partial V_1},\t_1}(f)\,\leq\, \mu_{V_1}^{U_2^{\partial V_1},\t_2}(f)\Eq(1.4)$$ $$ \mu_{V_2}^{U^{\partial V_2},\t_2}(f)\,\leq\, \mu_{V_1}^{+}(f)\Eq(1.5)$$ \bigskip $\S 2$\hskip 1cm Contours and Cluster Expansion.\medskip\noindent In this second paragraph we recall, for the reader's convenience, a version of the cluster expansion for the partition function $Z(V,U^{\partial V},+)$ valid under some restrictions on the boundary coupling $U^{\partial V}$, which will turn out to be quite essential in the next sections. The material that follows has been adapted to our situation, in which $U^{\partial V}$ is not necessarily identically equal to one, from sections 3.8, 3.9 of [DKS].\par To begin with, let us recall the definition of Peierls contours for a generic configuration $\s$ which is identically equal to +1 outside a finite region.\par If we denote by $\Z^*$ the dual lattice of $\Z$, we call a {\it bond} any segment in ${\bf R^2}$ connecting two neighboring sites of $\Z^*$. Then we say that two sites $x$ and $y$ in $\Z$ are separated by the bond $h$ if their distance (as sites in ${\bf R^2}$) from $h$ is equal to $1\over 2$. Given $\s\in \O_{\Z}$ we denote by $\Gamma (\s )$ the collection of all bonds separating sites $x$ and $y$ in $\Z$ where $\s (x)\neq \s (y)$. If moreover we use the convention that any pair of orthogonal bonds that intersect in a given site $x^*$ of the dual lattice $\Z^*$ are a {\it linked pair of bonds} iff they are both on the same side of the forty-five degrees line across $x^*$, then we immediately see that $\Gamma (\s )$ splits up in a unique way in a collection of closed contours $\Gamma_1 (\s )\, ,\Gamma_2 (\s ),\dots\,\Gamma_i (\s )\dots$ where a closed contour is a sequence $e_o,\,e_1,\,e_2\,\dots e_n$ of bonds such that:\medskip \item{i)} $e_i\,\neq \,e_j$ for all $i$ and $j$ with the exception of $i=0$ and $j=n$ for which $e_o\,=\,e_n$. \item{ii)} for all $i$ the bonds $e_i$ and $e_{i+1}$ have a common vertex in $\Z^*$. \item{iii)} if $e_i$, $e_{i+1}$, $e_j$, $e_{j+1}$ intersect at a given site $x^*$ then both $e_i$, $e_{i+1}$ and $e_j$, $e_{j+1}$ are linked pairs of bonds.\medskip The length $\vert \G\vert$ of a contour is simply the number of bonds in $\G$. Given a contour $\G$, we denote by $\D \G$ the set of sites in $\Z$ such that either their distance (in ${\bf R^2}$) from $\G$ is $1\over 2$ or their distance from the set of vertices of $\Z^*$ where two non-linked pair of bonds of $\G$ meet is equal to $1\over \sqrt{2}$. \par Since we can always identify any finite set $V\subset \Z$ with the bounded set $\tilde V\subset {\bf R^2}$ obtained by considering the union of all unit closed squares centered at each site in $V$, with an abuse of notation we will write for a generic closed contour $\G$: $\G \subset V$ if $\G \subset \tilde V$ and $\G \cap V$ for the set of bonds of $\G (\s )$ contained in $\tilde V$.\par Finally, given a boundary condition $\t$ on the external boundary of a finite region $V$, we can associate to any element $\s_V$ the configuration $\s^{(\t +)}\in \O_{\Z}$ equal to $\s_V$ inside $V$, equal to $\t$ on $\partial_{ext}V$ and equal to +1 outside $V\cup \partial_{ext}V $. Then, via the previous construction, we can associate in a unique way to $\s_V$ the {\it finite} collection of closed contours $\G (\s^{(\t +)} )$ that, for simplicity, will be referred to as $\G^\t (\s_V )$. If we consider $\G^\t (\s_V )\cap V$, then it will consists of the union of some closed contours, in the sequel refered to as the closed contours of $\s_V$ under the boundary condition $\t$, and some open polygonal curves that will be refered to as the open contours of $\s_V$ under the boundary condition $\t$, where an open polygonal line is a sequence of distinct bonds $e_o,\,e_1,\,e_2\,\dots e_n$ satisfying ii) and iii) above.\par Notice that, by construction, the first and last bond of an open contour necessarily separate at least one site in $\partial_{int}V$.\par\par Let us now assume that $$\a_V\,\equiv \,\min_{(x,y)\,\in \,\partial V}U^{\partial V}(x,y)\,>\,0\Eq(1.6)$$ Then, if for a given closed contour $\G$ we write $\G^{\partial V}$ for the sets of bonds in $\G$ that separate two sites $(x,y)\,\in \,\partial V$ and we set $U^{\partial V}(h )\,\equiv\, U^{\partial V}(x,y)$ for any pair $(x,y)\,\in \,\partial V$ that are separated by $h\in \G^{\partial V}$, a simple computation shows that: $$H_V^{U^{\partial V},+}(\s_V )\;=\;2\sum_{\G\in \G^+ (\s_V)}\{\vert \G\vert \,-\,\sum_{h\in \G^{\partial V}}(1-U^{\partial V}(h ))\}\Eq(1.7)$$ Thus the partition function can be written as: $$Z(V,U^{\partial V},+)\;=\;\sum_{\s_V} \exp(-\beta (2\sum_{\G\in \G^+ (\s_V)}\{\vert \G\vert \,-\,\sum_{h\in \G^{\partial V}}(1-U^{\partial V}(h ))\}))\Eq(1.8)$$ We can rewrite \equ(1.8) in a more suitable form by introducing the notion of compatibility between different contours. \par We say that the contours $\G_1,\dots\G_n$ in $V$ are {\it compatible} if there exists $\s_V\in \O_V$ such that $\G^+ (\s_V)\,=\,\{\G_1,\dots\G_n\}$ and we denote by ${\cal C}_V$ the set of them. Then, if we denote by $z_V^{U^{\partial V}}(\G )$ the weight of a single contour $\G$: $$z_V^{U^{\partial V}}(\G )\;=\;\exp(-\beta 2\{\vert \G\vert \,-\,\sum_{h\in \G^{\partial V}}(1-U^{\partial V}(h ))\}))\Eq(1.9)$$ \equ(1.8) can be written as : $$Z(V,U^{\partial V},+)\,=\,\sum_{{\cal G}\in {\cal C}_V}\prod_{\G\in {\cal G}}z_V^{U^{\partial V}}(\G )\Eq(1.10)$$ Then the main result of the cluster expansion that is needed in the present paper can be stated as follows :\bigskip {\bf Proposition 1.1}\par {\it Assume that there exists a constant $\a\in [0,1)$ such that: $$z_V^{U^{\partial V}}(\G )\,\leq \,\exp (-2\beta \vert \G\vert (1-\a ))\quad \forall \, \G\in {\cal C}_V$$ Then there exists $\beta_o \,=\,\beta_o(\a )$ such that for all $\beta \geq \beta_o$ the logarithm of the partition function $Z(V,U^{\partial V},+)$ can be written as: $$\log (Z(V,U^{\partial V},+))\,=\,\sum_{\L\subset V} \Phi^{U^{\partial V},+} (\L )$$ where the coefficients $\Phi^{U^{\partial V},+} (\L )$ satisfy the following two basic properties:\medskip \item{1)} $$\Phi^{U^{\partial V},+} (\L )\,=\, \Phi^{+} (\L )\qquad \hbox{if } \;\partial_{ext}\L\subset V$$ where $\Phi^{+} (\L )$ is the coefficient associated to the set $\L\subset V$ when the boundary coupling $U^{\partial V}$ is identically equal to one.\par \item{2)} For all $\L\subset V$, $\vert \L\vert \geq 2$: $$\eqalign{\vert \Phi^{U^{\partial V},+} (\phantom{\{}\L \phantom{\{})\vert \,&\leq\,\exp (-2(1-\a )[\beta \,-\,\beta_o]d(\L ))\cr\vert\Phi^{U^{\partial V},+} (\{x\} )\vert \,&\leq\,\exp (-8(1-\a )[\beta \,-\,\beta_o])}$$ where, for all connected (in the sense of subgraphs of the graph $\Z$) $\L\subset V$, $d(\L )$ is the length of the smallest connected set of bonds from $\bar \L\, \equiv\,\{$ all bonds in $V$ that separates at least one site in $\L\,\}$ containing all the bonds separating sites in $\partial_{int}\L$ from sites in $\partial_{ext}\L$. If $\L$ is not connected $d(\L )\,=\,+\infty$. }\bigskip $\S 3$\hskip 1cm Surface Tension and Wulff Shape.\medskip\noindent We conclude our short review of the Ising model by recalling the definition of the surface tension $\t_\beta({\bf \vec n })$ and of the associated Wulff shape. Again we follow the basic reference [DKS].\par Let us fix a direction ${\bf \vec n \,\in \, }S^1$ ($S^1$ being the unit circle) and let us define the boundary condition $\t^{\bf \vec n}$ as follows: $$\eqalign{t^{\bf \vec n}(x)\,&=\, +1 \quad\hbox{if}\quad (x,{\bf \vec n})\,>\,0\cr t^{\bf \vec n}(x)\,&=\, -1 \qquad \hbox{otherwise}}$$ where $(x,{\bf \vec n})$ denotes the usual scalar product in $\bf R^2$.\par Let also $V_{N,M}$ be the rectangle $\{x\in \Z\,;\;\;-N\leq x_1\leq N\,;\;-M\leq x_2\leq M\;\}$. Then we define the surface tension with respect to a surface orthogonal to the direction $\bf \vec n$, $\t_\beta({\bf \vec n})$, as: $$\t_\beta({\bf \vec n})\,\equiv\,\lim_{N\to \infty} \lim_{M\to \infty}{1\over \beta d(N,{\bf \vec n})} \log ({Z(V_{N,M},t^{\bf \vec n}) \over Z(V_{N,M},+)})\Eq(1.11)$$ where $d(N,{\bf \vec n})$ is the length of the segment $$\{x\,;\quad (x,{\bf \vec n})\,=\,0\, ,\quad -N\leq x_1\leq N\;\}$$ We will simply write $\tau_\beta$ to denote the surface tension associated to the direction ${\bf \vec n}\,=\,(1,0)$.\par For a proof of the existence of the limit \equ(1.11) when $\beta$ is large enough see Theorem 1.15 in [DKS].\par We now define the Wulff shape $W\, \subset \, {\bf R^2}$ as: $$W\,=\,\{x\in {\bf R^2};\;\vert (x,{\bf \vec n})\vert \,\leq \, \l\t_\beta({\bf \vec n})\quad \forall \;{\bf \vec n}\,\}\Eq(1.12)$$ where the constant $\l$ is chosen in such a way that the area of $W$ is equal to 1. The following fundamental result has been proved in [DKS] (see also [Pf]):\bigskip {\bf Theorem 1.1}\par {\it Let, for any closed, piecewise smooth curve $\g$ in ${\bf R^2}$, the Wulff functional $W_\t(\g )$ on $\g$ be given by: $$W_\t(\g )\;=\;\int_{\g}ds\,\t_\beta({\bf \vec n({\it s})})$$ where ${\bf \vec n({\it s})}$ is the normal vector at the point $s$ on the curve $\g$. Then, if we denote by $\partial W$ the closed curve encircling the Wulff shape $W$, we have: $$W_\t(\g )\;\geq \;W_\t(\partial W )$$ for any closed curve $\g$ which encloses an area equal to one, and equality holds iff $\g$ is a translate of the curve $\partial W$.}\bigskip $\S 3$\hskip 1cm A Class of Block-Glauber Dynamics for The Ising Model.\medskip\noindent In this paragraph we define, for a given finite set $V\subset \Z$, boundary condition $\t\in \O_\Z$ and boundary coupling $U^{\partial V}$, a class of Markov processes on $\O_V$ which are all reversible with respect to the Gibbs measure $\mu_V^{U^{\partial V},\t}$.\par Although the main object of study in this work is any standard (e.g Metropolis or Heath Bath) {\it single spin flip} Markov process, reversible with respect to the Gibbs measure of the Ising model, we found very convenient to introduce, as a technical tool, auxiliary Markov processes for which, in each updating of the dynamics, a whole collection of dynamical variables (i.e. spins $\s_V(x)$) are changed instead of just one. Each one of these auxiliary Markov processes will be indexed by a certain covering of the set $V$ by {\it blocks} (i.e. subsets of $V$) and at a given updating only the spins inside a particular block will be changed.\par More precisely, let $\{Q_i\}_{i=1\dots n}$ be a covering of $V$ and let $$\eqalign{U^{\partial Q_i}(x,y)\,&=\,1\phantom{^{\partial V}(x,y)} \qquad \hbox{if }(x,y)\,\in \,\partial Q_i\setminus\{\partial Q_i\cap \partial V\}\cr U^{\partial Q_i}(x,y)\,&=\,U^{\partial V}(x,y)\qquad \hbox{if }(x,y)\,\in \,\partial Q_i\cap \partial V}\Eq(1.12bis)$$ Then we define the generator $L^{\{Q_i\},\t ,U^{\partial V}}$ of the Markov process $\s_t^{\{Q_i\},\t ,U^{\partial V}}$ indexed by the covering $\{Q_i\}_{i=1\dots n}$ by: $$(\,L^{\{Q_i\},\t ,U^{\partial V}}f\,) (\s_V )\;=\;\sum_{i}\sum_{\eta\in \O_{Q_i}} \mu_{Q_i}^{U^{\partial Q_i},(\t\s_V)}(\eta )\,[f(\s_V^{\eta}\;-\;f(\s_V )\,]\Eq(1.13)$$ where $(\t\s_V)$ denotes the configuration in $\O_\Z$ equal to $\t$ outside $V$ and to $\s_V$ inside $V$, while $\s_V^{\eta}$ is the configuration in $\O_V$ equal to $\eta$ in $Q_i$ and to $\s_{V\setminus Q_i}$ in $V\setminus Q_i$. Most of the times we will refer to the Markov process generated by $L^{\{Q_i\},\t ,U^{\partial V}}$ as the $\{Q_i\}$-dynamics. \par A concrete way to construct the $\{Q_i\}$-dynamics starting from a configuration $\s\,\equiv\,\s_V$ is to choose with rate $n$, ($n$ is the cardinality of the covering), a particular element $Q_i$ of the covering, and to replace the restriction to $Q_i$ of the configuration $\s$ with a configuration $\eta\,\in \,\O_{Q_i}$ with probability $\mu_{Q_i}^{U^{\partial Q_i},(\t\s )}(\eta )$.\par The particular case in which the elements $Q_i$ of the covering are the sites $x$ of the set $V$ is known in the literature as the {\it Heat Bath process} (HB-dynamics in the sequel) and it is a particular example of a Glauber dynamics for the Ising model , that is a Markov process on $\O_V$ with generator $L^{\t ,U^{\partial V}}$ of the form: $$(\,L^{\t ,U^{\partial V}}f\,) (\s_V )\;=\;\sum_{x\in V}\sum_{a\in \{-1,+1\}} c_x^{\t ,U^{\partial V}}(\s_V ,a)\,[f(\s_V^{x,a}\;-\;f(\s_V )\,]\Eq(1.14)$$ where $\s_V^{x,a}$ is obtained from $\s_V$ by substituting the value $\s_V(x)$ with $a$ and the jump rates $c_x^{\t ,U^{\partial V}}(\s_V ,a)$ satisfy the {\it detailed balance condition}: $$\mu_V^{U^{\partial V},\t}(\s_V)\,c_x^{\t ,U^{\partial V}}(\s_V ,a)\;=\; \mu_V^{U^{\partial V},\t}(\s_V^{x,a})\,c_x^{\t ,U^{\partial V}}(\s_V^{x,a},\s_V(x))\Eq(1.14bis)$$ and a short range condition: $$c_x^{\t ,U^{\partial V}}(\s_V ,a)\,=\,c_x^{\t ,U^{\partial V}}(\eta_V ,a)\quad\hbox{if }\;\s_V(y)\,=\,\eta_V(y)\quad\forall \;\norm{x-y}\,\leq\, R$$ for some finite $R$. \par As it is easy to check, the $\{Q_i\}$-dynamics is a (continuous time) Markov chain on $\O_V$, reversible with respect to the Gibbs measure $\mu_V^{U^{\partial V},\t}$; in other words $L^{\{Q_i\},\t ,U^{\partial V}}$ is symmetric in the Hilbert space $L^2(\O_V,\,d\mu_V^{U^{\partial V},\t})$ with real non positive eigenvalues $$0\,=\,\l_o(\{Q_i\},\t ,U^{\partial V})\,>\,- \l_1(\{Q_i\},\t ,U^{\partial V})\,\geq\,\dots\,\geq \, -\, \l_{k}(\{Q_i\},\t ,U^{\partial V});\quad k\,=\,2^{\vert V\vert}$$ The absolute value of the first negative eigenvalue, $\l_1(\{Q_i\},\t ,U^{\partial V})$, will be of special value for us and it will be denoted by $\hbox{gap}_V(\{Q_i\},\t ,U^{\partial V})$ or by $\hbox{gap}_V(HB,\t ,U^{\partial V})$ if the dynamics under consideration is the Heat-Bath.\par The following variational characterization of the gap will be particulaly useful in the sequel. Let, for any $f\,\in \, L^2(\O_V,\,d\mu_V^{U^{\partial V},\t})$, ${\cal E}(f,f)$ be the Dirichlet form associated to the generator $L^{\{Q_i\},\t ,U^{\partial V}}$: $${\cal E}(f,f)\,=\,{1\over 2}\sum_i\sum_{\s_V}\sum_{\eta\in \O_{Q_i}} \mu_V^{U^{\partial V},\t}(\s_V)\mu_{Q_i}^{U^{\partial Q_i},(\t\s_V)}(\eta )\,[f(\s_V^{\eta})\;-\;f(\s_V )\,]^2\Eq(1.14tris)$$ Then: $$\hbox{gap}_V(\{Q_i\}\,=\,\inf_{\vbox{\eightpoint{\hbox{$f\,\in \, L^2(\O_V,\,d\mu_V^{U^{\partial V},\t})$} \hbox{$$}}}}{{\cal E}(f,f)\over Var(f)}\Eq(1.14quatris)$$ where $$Var(f)\,=\,{1\over 2}\sum_{\s ,\h}\mu_V^{U^{\partial V},\t}(\s ) \mu_V^{U^{\partial V},\t}(\h )[f(\s )\,-\,f(\h )]^2$$ {\bf Remark} Using the above variational characterization of the gap, it is very easy to check that, if we consider a general Glauber dynamics defined as in \equ(1.14) with jump rates bounded above and below uniformly in $\s_V$ and in $V$, then the corresponding gap can be bounded from above and from below by $\hbox{gap}_V(HB,\t ,U^{\partial V})$ multiplied by two suitable constants. \bigskip The following simple estimate, which follows from elementary $L^2$ consideration, illustrates the role played by the $\hbox{gap}(\{Q_i\},\t ,U^{\partial V})$ in the approach to the invariant measure $\mu_V^{U^{\partial V},\t}$ of the distribution $P_{V,\eta_V}^{\{Q_i\},\t ,U^{\partial V}}(t)$ of the $\{Q_i\}$-dynamics at time t starting from $\eta_V$ at time $t=0$: $$\norm{P_{V,\eta_V}^{\{Q_i\},\t ,U^{\partial V}}\;- \;\mu_V^{U^{\partial V},\t}}\;\leq \;{\exp (-t \hbox{ gap}(\{Q_i\},\t ,U^{\partial V}))\over 2[\mu_V^{U^{\partial V},\t}(\eta_V)]^{1\over 2}}\Eq(1.15)$$ where, for two arbitrary probability measures $\nu$ and $\mu$ on $\O_V$, $\norm{\nu\;-\;\mu}$ denotes their variation distance.\bigskip {\bf Remark} It is worthwhile to observe that \equ(1.15) can be a very bad estimate since the denominator $[\mu_V^{U^{\partial V},\t}(\eta_V)]^{1\over 2}$ is of order $\exp (-c\beta\,\vert V\vert )$ for some constant $c$. There are situations, for example when $\beta$ is smaller than the critical value $\beta_{c}$, in which the factor $[\mu_V^{U^{\partial V},\t}(\eta_V)]^{-{1\over 2}} $ in \equ(1.15) can be replaced by $c\vert V\vert$ for some constant $c$ (see e.g [SZ],[MO1] and [MO2]). However, in a phase transition regime, $\beta \,>\, \beta_c$, the gap can be very small, something like $\exp (-cL)$ if $V$ is a square of side $L$ with open boundary conditions (see section 4), and therefore the possible improvement in the denominator from $\exp (-c\beta\,\vert V\vert )$ to some negative power of $\vert V\vert$ is negligible.\bigskip $\S 4$\hskip 1cm Coupling for the $\{Q_i\}$-Dynamics .\medskip\noindent We conclude this preparatory section by discussing a useful coupling for the $\{Q_i\}$-dynamics that will be essential in the forthcoming sections.\par Let, for any finite set $V$, $\t^{(1)},\,\t^{(2)}\,\dots\,t^{(N)}$ be $N\,\leq \,2^{\vert \partial_{ext}V\vert }$ boundary conditions on the external boundary of $V$, and let $\nu_{V}^{ \t^{(1)},\,\t^{(2)}\,\dots\,t^{(N)}}$ be the unique invariant probability measure on $(\O_{V})^N$ ($N$ copies of $\O_{V}$) of the following ergodic Markov process:\bigskip \item{i)} With rate $\vert V\vert$ one chooses a site $x\,\in \,V$ and, given $x$, a random number $\xi_x\,\in \,[0,1]$ with a uniform distribution. $$\Eq(1.15bis)$$ \item{ii)}For $k\,=\,1\dots N$ the value of the spin at x in the $k^{th}$ component of the initial configuration $\tilde \s_V\;\equiv \;\{\s^{(1)}_V\dots \s^{(N)}_V\}$, $\s^{(k)}_V\,\in \, \O_{V}$, is replaced by $+1$ if $$\xi_x\,\leq\,\mu_{\{x\}}^{U^{\partial\{x\}},(\t^{(k)}\s^{(k)}_V)}(+1)\Eq(1.15tris)$$ and by $-1$ if the opposite inequality holds. Here $U^{\partial\{x\}}$ is defined as in \equ(1.12bis) but with $Q_i$ replaced by $\{x\}$.\bigskip\noindent The above algorithm is of course nothing more than an explicit way to realize on a common probability space the HB-dynamics in $V$ with different boundary conditions $\t^{(1)},\,\t^{(2)}\,\dotsJ\,t^{(N)}$.\par Using this observation one can explicitely check that the measure $\nu_{V}^{ \t^{(1)},\,\t^{(2)}\,\dots\,t^{(N)}}$ enjoyes the following properties: $$\sum_{\vbox{\eightpoint{\hbox{$\eta^{(1)},\dots\eta^{(k-1)}$} \hbox{$\eta^{(k+1)}\dots\eta^{(N)}$}}}} \nu_{V}^{ \t^{(1)},\,\t^{(2)}\,\dots \,t^{(N)}} (\eta^{(1)},\dots\eta^{(k-1)},\eta^{(k)}, \eta^{(k+1)}\dots\eta^{(N)} )\;=\;\mu_{V}^{U^{\partial V},(\t^{(k)})}(\eta^{(k)} )\Eq(1.16)$$ $$\nu_{V}^{ \t^{(1)},\,\t^{(2)}\,\dotsJ\,t^{(N)}} (\eta^{(k)}\,\leq\,\eta^{(j)} )\;=\;1\quad \hbox{if } \t^{(k)}\,\leq \t^{(j)}\Eq(1.17)$$ Given now a finite set $V$, a boundary condition $\t$ and a covering $\{Q_i\}_{i=1}^{n}$, let, for each $i=1\dots n$, $\t^{(1)},\,\t^{(2)}\,\dots\,t^{(N)}$ be an arbitrary enumeration of all the possible boundary conditions on the external boundary of $Q_i$ which agree with $\t$ on $\partial_{ext}Q_i\cap \partial_{ext} V$, and let $$\nu_{Q_i}\,\equiv \, \nu_{Q_i}^{ \t^{(1)},\,\t^{(2)}\,\dots\,t^{(N)}}$$ Using the measures $\nu_{Q_i}$, we can mimick the algorithm \equ(1.15bis), \equ(1.15tris), to realize on a common probability space the Markov processes $\s_t^{\{Q_i\},\t ,U^{\partial V}}$ starting from an arbitrary initial condition $\s_V$ as follows:\medskip \item{a)} With rate $n$ ($n$ is the cardinality of the covering) we choose one of the $Q_i$'s. $$\Eq(1.18)$$ \item{b)} For all $k=1\,\dots N$, the configurations $\s_V$ which agree with $\t^{(k)}$ on the external boundary of $Q_i$ are updated to $\s_V^{\eta^{(k)}}$, $\eta^{(k)}\,\in \, \O_{Q_i}$, and the joint probability of $\eta^{(1)},\,\eta^{(2)},\dots \eta^{(N)}$ is $\nu_{Q_i}(\eta^{(1)},\,\eta^{(2)},\dots \eta^{(N)})$. $$\Eq(1.19)$$ It is clear that, because of \equ(1.16) above, a) and b) give the right law for the evolution of any given initial configuration $\s_V$. Moreover, because of \equ(1.17), it also follows that any ordered set of initial conditions $\s_{V}^1\leq\s_{V}^2\leq\dots\leq\s_{V}^k$ will remain orderded for any future time $t$. We will refer to this last property as monotonicity in the initial configuration.\pagina %\input formato.tex \numsec=2 \numfor=1 {\bf Section 2}\par \centerline{\bf Geometric Bounds On the Gap}\bigskip In this section we establish two basic estimates on the gap which, besides being interesting by themselves, will play a crucial role in the determination of the exact asymptotics in the thermodynamic limit of the gap of the HB-dynamics in a finite square with open boundary conditions. The first estimate relates $\hbox{gap}_V(HB,\t ,U^{\partial V})$ to $\hbox{gap}_V(\{Q_i\},\t ,U^{\partial V})$ when $V$ is a rectangle $V_{N,M}$ $$V_{N,M}\,=\,\{x;\;-N\leq x_1\leq N;\;-M\leq x_2\leq M\}$$ with, say, $M\,\leq\,N$ and the covering $\{Q_i\}$ consists of rectangles: $$Q_i\,=\,\{\;x\in \Z;\;-N\,\leq \,x_1\,\leq\,N\quad i{l\over 2}\,\leq\,x_2\,\leq\,(i+2){l\over 2}\;\}$$ with $l\over 2$ and $2M/l\;$ integers, $i\,=\,-{2M\over l}\dots {2M\over l}-2$. The estimate shows that the ratio $$\hbox{gap}_V(HB,\t ,U^{\partial V})\over \hbox{gap}_V(\{Q_i\},\t ,U^{\partial V})$$ is bounded from below by a suitable exponential of the {\it short} side $l$. More precisely :\bigskip {\bf Theorem 2.1}\par {\it Let $V$ and $\{Q_i\}$ be as above. Then for any boundary coupling $U^{\partial V}$ and any boundary condition $\t$ we have: $$\hbox{\rm gap}_{V}(HB,\t ,U^{\partial V})\;\geq\; {1\over 2\vert Q_i\vert}{\exp (-4\beta)\over \exp (-4\beta)+\exp (+4\beta)}\,\exp (-4\beta (l\,+\,1))\,\hbox{\rm gap}_V(\{Q_i\},\t ,U^{\partial V})$$} \bigskip {\bf Remark} The above theorem remains valid also if the covering of the set V consisted of rectangles $Q_i$ with longest side smaller than that of $V_{N,M}$. However, for reasons that will become clear in the next section, the above choice of the covering is very sensible in the low temperature regime. It will also become clear at the end of the proof of the theorem that our method allows one to relate the gap of the HB-dynamics to that of the $\{Q_i\}$-dynamics for arbitrary geometric shapes of the elements of the covering. This generality is however not needed in the present paper.\bigskip As a corollary we obtain, in the same setting as above, that $\hbox{gap}_V(HB,\t ,U^{\partial V})$ is not smaller than a negative exponential of the shortest side $M$. More precisely we have:\bigskip {\bf Corollary 2.1}\par {\it For any boundary coupling $U^{\partial V}$ and any boundary condition $\t$ we have: $$\hbox{\rm gap}_{V}(HB,\t ,U^{\partial V})\;\geq\; {1\over 2\vert V\vert}{\exp (-4\beta)\over \exp (-4\beta)+\exp (+4\beta)}\,\exp (-4\beta (2M\,+\,1))$$} \bigskip {\bf Proof of the corollary}\par Let us take in theorem 2.1 the shortest side $l$ of the elements $Q_i$ of the covering equal to $2M$ so that the covering consists of just the rectangle $V_{N,M}$ itself. Then the generator $L^{\{Q_i\},\t ,U^{\partial V}}$ restricted to the space of functions of mean zero (i.e. orthogonal to the constant functions) becomes minus the identity, so that $\hbox{gap}_V(\{Q_i\},\t ,U^{\partial V})\,=\,1$, and the corollary follows from theorem 2.1. \bigskip {\bf Remark} The estimate described in the corollary is a very bad one for temperatures above the critical one (that is $\beta\,<\,\beta_c$), since in this case it has been recently proved by Olivieri, Schonmann and myself [MOS] that the gap is bounded away from zero uniformly in $N$ and $M$. However, at low temperature, when the infinite volume dynamics is not ergodic, it gives the right dependence on the size of the set $V_{N,M}$, namely a negative exponential of the surface and not of the volume $\vert V_{N,M}\vert$, but the constant in the exponential is wrong by a factor 2 even in the limit $\beta\,\to \,\infty$. A more precise bound will be discussed in the next section.\par The proof of the corollary represents also the first, actually rather trivial, example of the role played by the $\{Q_i\}$-dynamics: in an approach to a gap estimate this latter may be considerably simpler than the single spin dynamics. In particular one may try to attack the problem of finding a lower bound on the gap of the HB-dynamics by first proving lower bounds on the gap of the $\{Q_i\}$-dynamics and then, using Theorem 2.1 above, transfer the bound to the HB-dynamics. This idea played an important role in the analysis of the approach to equilibrium in general Glauber dynamics in the one phase region (see e.g. [H], [SZ], [MO1]). However its application in the phase transition region seems to be new. \bigskip {\bf Proof of Theorem 2.1}\par The proof is an application in our contest of some geometric techniques developed few years ago in order to bound from below the gap of symmetric Markov chains on complicate graphs (see e.g. [LS], [JS1],[JS2], [Si], [DS]). We will make use in particular of some beautiful ideas introduced by Jerrum and Sinclair in their study of rapid mixing properties of Markov chains arising in some hard computational problems. The way these techniques apply to spin dynamics like the Glauber dynamics was discussed for the first time in some unpublished notes of mine and used recently, in a slightly different form, by Schonmann in his study of metastability for the Ising model [Sch1]. \par In what follows we will omit for simplicity in all the notation the boundary condition $\t$, the boundary coupling $U^{\partial V}$ and the volume $V$. Thus for example the Gibbs measure $\mu_V^{U^{\partial V},\t}$ will become $\mu$, the conditional Gibbs measure on $Q_i$, $\mu_{Q_i}^{U^{\partial Q_i},(\t\s)_V}(\eta )$, $\mu_{Q_i}^{\s}(\eta )$ and similarly for the generators of the HB and $\{Q_i\}$ dynamics together with their gaps. \par We start by introducing the set of {\it canonical paths} in $\O_V$ between configurations $\s$ and $\s '$ with $\s \,\neq\, \s '$, that are connected by just one single jump of the $\{Q_i\}$-dynamics, that is $\s '\,=\,\s^{\eta}$ for some $i$ and some $\eta\in \O_{Q_i}$. We adopt the convention that, if the $\s ,\,\s '$ can be connected by the updating either of $Q_i$ or of $Q_{i+1}$, due to their mutual overlap, then we think of $\s '$ as arising from the updating of $Q_i$.\par Let us first order the sites in each rectangle $Q_i$ as follows: $$x\;< \;y\quad \hbox{iff}\quad x_1\,<\,y_1\; \hbox{ or } x_1\,=\,y_1\; \hbox{ and }x_2\,<\,y_2$$ Given now $\sigma\,\in \O_V$ and $\eta\in \O_{Q_i}$ we define the path $\gamma (\sigma ,\s^{\eta})$ as the sequence of configurations obtained from $\sigma$ by adjusting one by one, in increasing order, the values of its spins in $Q_i$ to those of the spins of $\s^{\eta}$. More precisely, if $x_1\,,\,.....\,x_n$ are the sites in $Q_i$, ordered as above, such that $\sigma (x_i)\,\neq\,\s^{\eta}(x_i)$, then we define $\gamma (\sigma ,\s^{\eta})\,=\,\{\sigma^o\,...\,\sigma^n\}$ where $\sigma^i$, $i=1\dots n$, is the configuration equal to: $$\eqalign{\sigma^i(x)\, &=\,\s^{\eta}(x)\quad\hskip 1cm \hbox{iff } x\,\leq\,x_i\cr \sigma^i(x)\,&=\,\s (x)\quad\hskip 1cm \hbox{iff } x\,>\, x_i}\Eq(2.1)$$ and $\sigma^o\,=\,\sigma$.\par Next, for any allowed transition of the HB-dynamics $$\bar\sigma\,\to\,\bar\s^{x,a}\quad x\,\in \,Q_i, \quad a\,=\,-\bar\s (x)$$ we set $$\eqalign{e\,&=\,(\bar\sigma,\bar\sigma^{x,a} ) \cr Q(e)\,&=\,\mu(\bar\sigma)\mu_x^{\bar\sigma} (a)}\Eq(2.2)$$ and we say that the transition $e$ belongs to the canonical path $\g$, $e\,\in \g$, if, for some index $i$, $(\bar \s,\bar\sigma^{x,a} )\,=\,(\s^i,\s^{i+1})$. Finally we define the constant $\rho$ as : $$\rho \,=\,\sup_{i,e} \sum_{\s\,,\eta\atop e\in \gamma (\sigma,\s^{\eta})} {\mu (\sigma)\mu_{Q_i}^{\s}(\eta)\over Q(e)}\Eq(2.3)$$ Then we have: $$\hbox{gap}_{V}(HB)\;\geq\; {1\over 2\vert Q_i\vert}{1\over \rho }\,\hbox{gap}_V(\{Q_i\})\Eq(2.4)$$ Although the proof of \equ(2.4) can be found in [Si], we reproduce it below because of its simplicity.\par Using the variational principle for any $f\,\in\,L^2(\O_V,\,d\mu )$ we have : $$Var(f)\,\leq\,\hbox{gap}_V(\{Q_i\})^{-1}{1\over 2}\sum_{\s ,i}\sum_{\eta}\mu (\s )\mu_{Q_i}^{\s }(\eta)[f(\s^{\eta})\,-\,f(\s )]^2\,=$$ $$=\;\hbox{gap}_V(\{Q_i\})^{-1}{1\over 2}\sum_{\s ,i}\sum_{\eta}\mu (\s )\mu_{Q_i}^{\s }(\eta)[\sum_{j=1\dots n}(f(\s^{j}\,-\,f(\s^{j-1}))]^2\;\Eq(2.5)$$ where $\gamma (\s\,,\s^{\eta})\,=\,\{\s^o\,\s^1\dots\s^n\}$ is the canonical path going from $\s$ to $\s^{\eta}$.\par Using the Schwartz inequality, the fact that the length $n$ of the path is smaller than $\vert Q_i\vert$ and the definition of $\rho$, we can bound from above the r.h.s. of \equ(2.5) by: $$\hbox{gap}_V(\{Q_i\})^{-1}\rho\,\vert Q_i\vert {1\over 2}\sum_{\s,\,i}\sum_{x\in Q_i}\sum_{a\in \{-1,1\}}\mu (\s )\mu_{x}^{\s }(a)[f(\s^{x,a}\,-\,f(\s )]^2\;\leq\;$$ $$\leq\;2\hbox{gap}_V(\{Q_i\})^{-1}\rho\,\vert Q_i\vert\,{\cal E}_{HB}(f,f)\Eq(2.6)$$ where ${\cal E}_{HB}(f,f)$ is the Dirichlet form of the HB-dynamics and the factor $2$ in the first inequality comes from the fact that most of the sites belong to two elements ofthe covering. Thus, if we combine \equ(2.5) with \equ(2.6) and (1.18), we get: $$\hbox{gap}_V(HB)\,\geq\,{\hbox{gap}_V(\{Q_i\})\over 2\rho\,\vert Q_i\vert}$$ and in order prove the theorem, we only need to estimate from above the constant $\rho$ by: $$\rho\;\leq\;{\exp (4\beta)+\exp (-4\beta) \over \exp (-4\beta)}\,\exp (4\beta (l\,+\,1))\Eq(2.7)$$ uniformly in the boundary condition $\t$ and in the boundary coupling $U^{\partial V}$.\par Apparently this is not an easy problem since we have to count how many canonical paths use a given allowed transition $e$ = $(\bar \sigma,\bar\sigma^{x,a} )$. It is precisely at this stage that Jerrum and Sinclair's lovely ideas become essential.\par Given a transition $e\,=\,(\bar\s\,,\bar\s^{x,a})$ we define an {\it injective $\Phi$} mapping from the set of all the canonical paths that use the transition $e$, $\G (e)$, to $\O_V$ as follows: $$\eqalign{\Phi(\gamma (\sigma ,\s^{\eta}))(y)\,&=\, \sigma (y) \phantom{\h}\qquad \forall \,y\,\in \, Q_i,\quad y\,< \, x\cr \Phi(\gamma (\sigma ,\s^{\eta}))(y)\,&=\, \s^{\eta} (y) \qquad\forall \,y\,\in \, Q_i,\quad y\,\geq \, x\cr \Phi(\gamma (\sigma ,\s^{\eta}))(y)\,&=\, \sigma (y)\phantom{\h} \qquad \forall \,y\,\notin \, Q_i }\Eq(2.7bis)$$ where the index $i$ labels the rectangle associatesd to the path $\g (\s ,\s^\h )$.\par It is clear that $\Phi$ is iniective. In fact the knowledege of the transition {\it e}, that is of $x$ and $\bar \s$, {\it and } of $\xi\,\equiv\,\Phi(\gamma (\sigma ,\s^{\eta}))$ allow us to reconstruct completely the initial and final configurations $\sigma$ and $\s^{\eta}$ and thus the path itself, simply by observing that, for example: $$\eqalign{\s (y)\,&=\,\bar \s (y)\quad \forall \;y\notin Q_i\cr \s (y)\,&=\,\bar \s (y)\quad \forall \;y\in Q_i\quad y\,\geq\,x\cr\s (y)\,&=\,\xi (y)\quad \forall \;y\in Q_i\quad y\,< \,x}\Eq(2.8)$$ and similarly for $\s^{\eta}$.\par Let now $c_o$ be the smallest constant such that for any canonical path $\gamma (\sigma ,\s^{\eta})$ in $\Gamma (e)$ the following bound holds : $$\mu_{Q_i}^{\s}(\Phi (\g ))\,Q(e)\; \geq \; {1\over c_o}\mu (\sigma)\mu_{Q_i}^\s(\eta)\Eq(2.9)$$ Then we have $$\rho\; \leq \; c_o\Eq(2.10)$$ Using \equ(2.9) we can in fact estimate the r.h.s. of \equ(2.3) by: $$c_o\sup_{e,i}\sum_{\gamma\,\in \,\G (e)}\mu_{Q_i}^{\s} (\Phi (\gamma ))\Eq(2.11)$$ Since the map $\Phi$ is injective and $\mu$ is a probability measure, the sum in \equ(2.11) is not greater than one and \equ(2.10) follows.\par In order to estimate the constant $c_o$, let, for $x\,\in \,Q_i$, $\partial_x$ be the set of bonds in $Q_i$ which separates sites in $Q_i$ smaller or equal than $x$ from sites in $Q_i$ larger than $x$. Clearly, by construction, $\partial_x$ consists of two vertical segments joined by a single horizontal bond $h$ at distance $1\over 2$ from $x$ and placed below it if $x_1\,\geq\,-N\,+\,1 $, and by a single vertical segment plus an horizontal bond as above if $x_1\,=\,-N$. Let also, for any pair configurations $\s_1$, $\s_2$ which agree outside $Q_i$, $H_{\partial_x}(\sigma_1,\sigma_2)$ be the interaction through $\partial_x$ of a configuration $\s\,\in \, \O_V$ which is equal to $\sigma_1$ ($\sigma_2$) to the left (right) of $\partial_x$. More precisely: $$H_{\partial_x}(\sigma_1,\sigma_2)\;=\; -\sum_{y\,\leq\, x\,<\,z\atop z\in Q_i,\,\norm{y-z}\,=\,1}(\sigma_1(y)\sigma_2(z)-1)\Eq(2.12)$$ Clearly $\vert H_{\partial_x}(\sigma_1,\sigma_2)\vert $ is bounded from above by $2(l\,+\,1)$. Then, by direct inspection: $${\mu (\sigma)\mu_{Q_i}^\s(\eta)\over \mu_{Q_i}^{\s} (\Phi (\g ))\mu (\bar \s )}\,=\,$$ $$=\;\exp (-\beta [H_{\partial_x}(\s,\s )\,+\, H_{\partial_x}(\s^{\eta},\s^{\eta})\,-\,H_{\partial_x}(\bar\s,\bar\s )\,-\,H_{\partial_x}(\Phi (\g ),\Phi (\g ))])\Eq(2.13)$$ for any boundary condition $\t$ and boundary coupling $U^{\partial V}$. In turn \equ(2.13), together with \equ(2.12) and the observation that $$\mu_x^{\bar \s}(a)\,\geq\,{\exp (-4\beta)\over \exp (-4\beta)+\exp (+4\beta)}\quad \forall\; x,\;\bar \s,\;\t,\;U^{\partial V}$$ implies that the l.h.s of \equ(2.13) is smaller than $${\exp (4\beta)+\exp (-4\beta)\over \exp (-4\beta)}\,\exp (4\beta (l\,+\,1))\mu_x^{\bar\s}(a)\Eq(2.14)$$ that is $$\mu_{Q_i}^{\s} (\Phi (\g ))\,Q(e)\; \geq \;{\exp (4\beta)+\exp (-4\beta)\over \exp (-4\beta)}\,\exp (4\beta (l\,+\,1)) \mu (\sigma)\mu_{Q_i}^\s(\eta)\Eq(2.15)$$ Thus the constant $c_o$ can be taken equal to $$c_o\;=\;{\exp (4\beta)+\exp (-4\beta)\over \exp (-4\beta)}\,\exp (4\beta (l\,+\,1))$$ Using \equ(2.7), \equ(2.10), the theorem follows.\bigskip {\bf Remark} It is amusing to observe that, if one applies the above construction to the one dimensional case for which the set $\partial_x$ consists of just {\it one } bond, $H_{\partial_x}(\sigma_1,\sigma_2)$ can be bounded by a constant independent of $\s_1$, $\s_2$ and of the dimension of $V$, even if the energy (1.1) of a configuration $\s_V$ is replaced by a more general expression like: $$H(\s_V)\,=\,-{1\over 2}\sum_{x,y\in V}J(\norm{x-y})\s_V(x)\s_V(y)\quad + \hbox{b.c.}$$ provided that the long range potential $J(\norm{x-y})$ decays faster than $\norm{x-y}^{-{2+\e}}$ for some $\e\,>0$. Therefore in this case the gap of the corresponding Heat Bath dynamics in a segment of length $L$ in $\bf Z$ has a lower bound which is only proportional to $L^{-1}$ without any negative exponential of $L$.\par On the other hand it is known that a long range potential $J(\norm{x-y})$ with a fast decay as above is not able to induce any phase transition, the reason being that the energy between two semi-infinite lines is finite uniformly in the spin configuration.\par Thus, in some sense, the above geometric contruction is able to capture, at least at the level of the exponential, some (but certainly not all) of the physical aspects of the presence (or of the absence) of a phase transition in the Ising model at low temperature.\bigskip \pagina %\input formato.tex \numsec=3 \numfor=1 \tolerance =15000 {\bf Section 3}\par \centerline{\bf A Lower Bound On the Gap With + Boundary Conditions} \centerline{\bf and Its Application} \bigskip In this section we consider the HB-dynamics in a square $V\equiv V_L$: $$V_L\,=\,\{\,x\,\in \, \Z\,;\quad 0\,\leq\,x_i\,\leq\,L\quad i=1,2\,\}$$ with full plus boundary conditions, that is $$\eqalign{\t (x)\,&=\,+1\; \;\forall \,x\in \Z\cr U^{\partial V_L}(x,y)\,&=\,+1\quad \forall \;(x,y)\,\in \, \partial V_L}$$ and very large $\beta$.\par We show that, due precisely to the presence of the plus boundary conditions, the gap of the HB-dynamics is much larger, as $L\to \infty$, than its value with {\it open} boundary conditions (see also the discussion in the introduction). As a simple consequence we show that the equal site time correlations of the infinite volume process started in the plus phase decay faster than any inverse power of the time.\par Before stating and discussing our main result let us fix few more convenient notation. We will denote by $(+)$ and $(-)$ the two extreme configurations in $\O_{V_L}$ identically equal to plus and minus one respectively and, for any rectangle $R$, by $\mu_R^{\t_1,\t_2,\t_3,\t_4}$, the Gibbs measure on $R$ with the boundary conditions $\t_1,\t_2,\t_3,\t_4$ on the external boundary of its four sides ordered clockwise starting from the bottom side. We use the usual convention that, if one of the configurations $\t_i$ is identically equal to +1 or -1, then we replace it by a + or a $-$ sign. Thus for example $\t_1,+,-,+$ means $\t_1$ boundary conditions on the bottom side, plus boundary conditions on the vertical ones and minus boundary condition on the top one. Whenever confusion does not arise we will also omit the subfix $V$ in the notation $\s_V$.\par We finally denote by $\mu^+$ the infinite volume Gibbs state obtained as the limit as $L\to \infty$ of finite volume Gibbs states $\mu_V^+$ with plus boundary conditions, by $m^*(\b )\,=\,\mu^+( \s (0))$ the spontaneous magnetization and by $\s_t$ the infinite volume Heat Bath dynamics started from the (infinite volume) configuration $\s$ (see [Li] for the existence of such process).\par We can now state the main results : \bigskip {\bf Theorem 3.1}\par {\it Let $\e\,\in \, (0,{1\over 2})$ be given. Then there exists $\beta_o\,<\,+\infty$ and $C\,<\,+\infty$ such that for any $\beta\,\geq\,\beta_o$ and any integer $L$: $$\hbox{gap}_{V_L}(HB,\,+)\;\geq\;\exp (-C\beta L^{{1\over 2}+\e})$$} {\bf Theorem 3.2}\par {\it Let $\a\,\in \,[0,2)$ be given. Then there exists $\beta_o\,<\,+\infty$ and $C\,<\,+\infty$ such that for any $\beta\,\geq\,\beta_o$ $$0\,\leq\,\int d\mu^+(\s )\s (0)E(\s_t(0))\,-\,(m^*(\b ))^2\,\leq\,C\exp (-(log(t))^\a)\quad \forall \;t$$} {\bf Proof of Theorem 3.1}\par Let $l\,=\,2[L^{{1\over 2}+\e}]$ and let us suppose, without loss of generality, that $ N\,\equiv \,{2L\over l}\,-\,1$ is an integer; for $i\,=\,1\dots N$, we define $Q_i$ to be the rectangle: $$Q_i\,=\,\{\;x\in V_L;\,0\,\leq \,x_1\,\leq\,L,\quad (i-1){l\over 2}\,\leq\,x_2\,\leq\,(i+1){l\over 2}\;\}$$ Then, using theorem 2.1, we have that: $$\hbox{gap}_{V_L}(HB,\,+)\;\geq\; {1\over \vert Q_i\vert}{\exp (-4\beta)\over \exp (-4\beta)+\exp (+4\beta)}\,\exp (-4\beta (l\,+\,1))\,\hbox{gap}_{V_L}(\{Q_i\},\,+)\Eq(3.1)$$ It remains to show that the $\{Q_i\}$-dynamics has a "large" gap, where "large" means, for example, larger than $\exp (-L^{{1+\e\over 2}})$ .\par To prove this result we will show that, with very large probability, under the coupling for the $\{Q_i\}$-dynamics described at the end of section 1, the two extreme configurations, $(+)$ and $(-)$ become identical in a time smaller than $\exp (L^{{1+\e\over 2}}) $.\par The intuitive reason for that, which also explains our apparently strange choice of the length $l$ of the short side of $Q_i$, is the following.\par Let us suppose that we start with the two extreme configurations and that we update one after the other in increasing order of $i$ the rectangles $Q_i$. In the first updating of $Q_1$ we have to replace $(+)_{Q_1}$ and $(-)_{Q_1}$ with two configurations $\eta_{Q_1}^+$ and $\eta_{Q_1}^-$ distributed according to $\mu_{Q_1}^{+,+,+,+}$ and $\mu_{Q_1}^{+,+,-,+}$ respectively. It is a relatively easy matter to show (see Proposition 3.1 below) that, for large enough $\beta$, due to our choice of $l$ and to the fact that in two dimensions the fluctuations of an interface separating plus spins from minus spins are of the order of the {\it square root} of the length of the interface, it is possible to couple the two measures $\mu_{Q_1}^{+,+,+,+}$, $\mu_{Q_1}^{+,+,-,+}$ in such a way that, with probability much larger than $1\,-\,{1\over N}$, the two configurations $\eta_{Q_1}^+$ and $\eta_{Q_1}^-$ are identical in a large portion of $Q_1$, e.g. for all $x\in Q_1$ with $x_2\,\leq\,{3l\over 4}$, and in particular on the external boundary of the bottom side of $Q_2$. Moreover, with large probability, both $\eta_{Q_1}^+$ and $\eta_{Q_1}^-$ will be mostly +1 on the external boundary of the bottom side of $Q_2$. Thus the second updating in $Q_2$ will be very similar to the first one in $Q_1$ with the exception that now the boundary conditions on the external boundary of the bottom side of $Q_2$ will not be identically equal to + but only approximately. \par As we will show below this fact, with probability much larger than $1\,-\,{1\over N}$, does not really matter and one can, at least in a first appproximation, consider the + boundary conditions also on the bottom side of $Q_2$. In this approximation the second updating will be statistically equal to the first one and, with large probability, it will force $(+)^{\eta_{Q_1}^+}$ and $(-)^{\eta_{Q_1}^-}$ to agree also in $3\over 4$ of $Q_2$ without introducing any new discrepancy between them in the previous region of agreement $$\{\,x\in Q_1;\;x_2\,\leq\,{3l\over 4}\,\}$$ In such a way, after the first two updatings, the evoluted of $(+)$ and $(-)$ will agree in the set $$\{\,x\in V_L;\;\;0\leq x_2\leq {5\over 4}l\,\}$$ By iterating this procedure N times we can glue together $(+)$ and $(-)$ in N steps with a probability of order one.\par Since the probability of {\it not} having within time $t$ a sequence of N updatings exactly in the order needed above is roughly of order $$\exp (-N^{-N}{t\over N})\,<<\,1\quad \hbox{if e.g. }\quad t\,=\,\exp (L^{{1+\e\over 2}})\quad L>>1$$ we can conclude that the time the $\{Q_i\}$-dynamics needs to relax to equilibrium should not be larger than $\exp (L^{{1+\e\over 2}})$ for $L$ large enough.\par Let us start with the technicalities. Let $R$ be the rectangle $$R\,=\,\{\;x\in \Z;\,0\,\leq \,x_1\,\leq\,L_1\quad 0\,\leq\,x_2\,\leq\,L_2\;\}$$ with $L_1\,\geq \,L_2\,\geq\,L_1^{{1\over 2}+\e}$. \bigskip {\bf Proposition 3.1}\par {\it Let $m\,>\,0$ and $\e\,\in \, (0,{1\over 2})$ be given. Then there exists $\beta_o\,\equiv\,\beta_o(\e,m )$ independent of $R$ such that for all $\beta \,\geq \,\beta_o$ and all $x\, \in \,R$ with $x_2\,\leq \,{3\over 4}L_2$, we have: $$\mu_R^{+,+,+,+}(\s (x)=1)\;-\;\mu_R^{+,+,-,+}(\s (x)=1)\;\leq \;\exp (-mL_1^{2\e})$$} The above result will actually be given in a greater generality than that required here, see proposition 4.1. The proof of proposition 4.1 has been collected with some other similar results for the Ising model in appendix 2.\medskip The second result that we need is an estimate on the probability of not seeing within time $t$ a sequence of updatings of the $\{Q_i\}$-dynamics with the correct order decribed above.\bigskip {\bf Lemma 3.1}\par {\it Let us call $S_N\;\equiv\;\{t_1\,\dots \,t_N\}$, $N\,=\,{2L\over l}\,-\,1$, an ordered sequence of updatings if for any $i=1,\dots N$:\medskip \item{i)} at time $t_i$ the dynamics updates the rectangle $Q_i$. \item{ii)} there are no updatings between times $t_i$ and $t_{i+1}$.\medskip Then, for any $N$ large enough (independent of $t$): $$P(\hbox{ there exists no ordered sequence in} \;[0,t]\,)\;\leq\;\exp (-{tN^{-N}\over 2})$$ }\bigskip {\bf Proof}\par Given that $t_1\,\dots \,t_N$ are $N$ consecutive updatings, the probability that $S_N\;\equiv\;\{t_1\,\dots \,t_N\}$ is an ordered sequence is clearly $N^{-N}$ since the probability of choosing a specific rectangle is $1\over N$. Let now $\nu_t$ denotes the total number of updatings within time $t$. By construction the process $\nu_t$ is a Poisson process of parameter $tN$. Therefore we can estimate the probability appearing in the Lemma by: $$P(\hbox{ there exists no ordered sequence in} \;[0,t]\,)\;\leq\;$$ $$\leq \;\sum_{k=0}^{+\infty}{e^{-tN}(tN)^k\over k!}(1\,-\,N^{-N})^{[{k\over N}]}\;\leq$$ $$\leq\;2e^{-tN(1\,-\,(1\,-\,N^{-N})^{1\over N})}\Eq(3.2)$$ which is smaller, for $N$ large enough, than $$\exp (-{tN^{-N}\over 2})\Eq(3.3)$$ Let now $S_N\;\equiv\;\{t_1\,\dots \,t_N\}$ be a fixed ordered sequence with $t_1\,=\,0$, let $\s^{\{Q_i\},+}_{t_i}$ be the evoluted at time $t_i$ of the initial configuration $\s$, let $R_i$ be the rectangle $$R_i\;=\;\{ x\in \cup_{j\leq i}Q_j;\;x_2\leq (i+1){l\over 2}-[{l\over 4}]\}$$ and let, for $ i=1\dots N-1$, $A_i(x)$, $A_i$, be the events: $$\eqalign{A_i(x)\;&=\;\{\,(+)^{\{Q_i\},+}_{t_i}(x)\, \neq\,(-)^{\{Q_i\},+}_{t_i}(x)\,\}\cr A_i\phantom{(x)}\;&=\;\bigcup_{\{ x\in R_i\}}A_i(x)\cr A_N\phantom{(x)}\;&=\;\bigcup_{\{ x\in V_L\}}A_N(x)}\Eq(3.4)$$ and let $q_i\,=\,P(A_i)$. Then we have: $$q_{n+1}\,\leq \,q_n\;+\;P(\,A_{n+1}\cap A_n^c\,)\,\leq\,\sum_{n=1}^{N-1}P(\,A_{n+1}\cap A_n^c\,)\;+\;P(A_1)\Eq(3.5)$$ where $A_n^c$ is the complement set of $A_n$.\par Then the term $P(\,A_{n+1}\cap A_n^c\,)$ in the r.h.s. of \equ(3.5) can be estimated by: $$P(\,A_{n+1}\cap A_n^c\,)\;\leq$$ $$\sum_{\vbox{\eightpoint{\hbox{$x\in R_{n+1}\cap Q_{n+1}$} \hbox{$\s\in \O_V$}}}}\mu_V^+(\s ) P(\,A_{n+1}(x)\cap [\cap_{y\in R_n}\{(+)^{\{Q_i\},+}_{t_n}(y)\,=\,(-)^{\{Q_i\},+}_{t_n}(y)\,=\,\s ^{\{Q_i\},+}_{t_n}(y)\}]\;)\Eq(3.6)$$ In the derivation of \equ(3.6) we used the fact that at time $t_{n+1}$ we update only the set $Q_{n+1}$ and that, under the coupling described in $\S 4$ of section 1, for any time $t$ and any configuration $\s$, $(-)^{\{Q_i\},+}_t\,\leq\,\s_t^{\{Q_i\},+}\,\leq\,(+)^{\{Q_i\},+}_t$.\par In turn, if we denote by $E$ the expectation over the random configuration $\s^{\{Q_i\},+}_{t_n}$, then a given term in the sum appearing in the r.h.s. of \equ(3.6) can be estimated from above by: $$\mu_V^+(\s )\, E [\mu_{Q_{n+1}}^{\s^{\{Q_i\},+}_{t_n},+,(+)^{\{Q_i\},+}_{t_n},+}(\eta (x)=1)- \mu_{Q_{n+1}}^{\s^{\{Q_i\},+}_{t_n},+,(-)^{\{Q_i\},+}_{t_n},+}(\eta (x)=1)]\,=$$ $$=\,[\mu_V^+(\s )\, E\mu_{Q_{n+1}}^{\s^{\{Q_i\},+}_{t_n},+,+,+}(\eta (x)=1)\,-\,\mu_V^+(\s )E\mu_{Q_{n+1}}^{\s^{\{Q_i\},+}_{t_n},+,-,+}(\eta (x)=1)]\Eq(3.7)$$ since $(+)^{\{Q_i\},+}_{t_n}$ and $(-)^{\{Q_i\},+}_{t_n}$ are, respectively, identically equal to plus one and minus one on the external boundary of the top of $Q_{n+1}$ because the sequence $S_N$ is ordered.\par Let us consider the term $$\sum_{\s\in \O_V}\mu_V^+(\s )\, E\mu_{Q_{n+1}}^{\s^{\{Q_i\},+}_{t_n},+,+,+}(\eta (x)=1)\Eq(3.8)$$ Since the $\{Q_i\}$-dynamics is reversible with respect to $\mu_V^+(\s )$, the distribution of $\s^{\{Q_i\},+}_{t_n}$, given that $\s$ is distributed according to $\mu_V^+(\s )$, will of course be again $\mu_V^+(\s )$. Therefore \equ(3.8) will be equal to: $$\sum_{\s\in \O_V}\mu_V^+(\s )\, \mu_{Q_{n+1}}^{\s,+,+,+}(\eta (x)=1)\,\leq\,\sum_{\s\in \O_{R_{n+1}\cup Q_{n+1}}} \mu_{R_{n+1}\cup Q_{n+1}}^{+,+,+,+}(\s )\, \mu_{Q_{n+1}}^{\s,+,+,+}(\eta (x)=1)\Eq(3.9)$$ where we used once more the monotonicity (1.5).\par By the DLR property of the Gibbs measure $\mu_{R_{n+1}\cup Q_{n+1}}^{+,+,+,+}$, the r.h.s. of \equ(3.9) is just $$\mu_{R_{n+1}\cup Q_{n+1}}^{+,+,+,+}(\s (x)\,=\,1)\Eq(3.10)$$ Similarly we obtain that the term $$\sum_{\s\in \O_V}\mu_V^+(\s )\, E\mu_{Q_{n+1}}^{\s^{\{Q_i\},+}_{t_n},+,-,+}(\eta (x)=1)$$ is bounded from below by $$\mu_{R_{n+1}\cup Q_{n+1}}^{+,+,-,+}(\s (x)\,=\,1)\Eq(3.11)$$ In conclusion, using \equ(3.10), \equ(3.11) and Proposition (3.1), we get that, for any $n$, the r.h.s. of \equ(3.6) is bounded from above by: $$\mu_{R_{n+1}\cup Q_{n+1}}^{+,+,+,+}(\s (x)\,=\,1)\,-\, \mu_{R_{n+1}\cup Q_{n+1}}^{+,+,-,+}(\s (x)\,=\,1)\,\leq\, L^2\exp (-mL^{2\e})\Eq(3.12)$$ for a suitable constant $m\,\equiv\,m(\beta )$ which diverges as $\beta\to \infty$ . Similarly one estimates $P(A_1)$.\par Therefore we get: $$q_N\,\leq \,NL^2\exp (-mL^{2\e})\Eq(3.13)$$ We are now in a position to conclude the proof of the theorem.\par Given a sequence $S_N\;\equiv\;\{t_1\,\dots \,t_N\}$ of updatings we say that $S_N$ is a {\it good } sequence iff $S_N$ is ordered and the event $A_N^c$ occured at the end of the sequence. Because of \equ(3.13) we know that the probability that an ordered sequence is also a good sequence is larger than $$1\,-\,NL^2\exp (-mL^{2\e})\,>\,{1\over 2}$$ for $L$ large enough. Thus, using Lemma 3.1, we get that if $T\,=\,\exp ( L^{{1+\e\over 2} })$ and $L$ is large enough: $$P(\hbox{ there exists a good sequence in} \;[0,T]\,)\;\geq\;{1\over 3}\Eq(3.14)$$ We conclude by observing that, if there exists a good sequence in $[0,t]$, then, by monotonicity (see $\S 4$ section 1), the evoluted at the end of the sequence of $(+)$ and of $(-)$ will be identical. Therefore we can estimate $P((+)^{\{Q_i\},+}_t\,\neq\,(-)^{\{Q_i\},+}_t)$ by $$P((+)^{\{Q_i\},+}_t\,\neq\,(-)^{\{Q_i\},+}_t)\,\leq\,({2\over 3})^{[{t\over T}]}\Eq(3.15)$$ which immediately implies that $$ \hbox{gap}_{V_L}(\{Q_i\},\,+)\,\geq\,T^{-1}log({3\over 2})\,= \, \exp (-L^{{1+\e\over 2} })\log ({3\over 2})\Eq(3.16)$$ Clearly \equ(3.16) together with \equ(3.1) prove the theorem.\bigskip {\bf Proof of Theorem 3.2}\par The first inequality, namely $$0\,\leq\,\int d\mu^+(\s )\s (0)E(\s_t(0))\,-\,(m^*(\b ))^2\Eq(3.17)$$ follows immediately from the FKG inequality applied to $\mu^+$ and the fact that the infinite volume Heat Bath dynamics is reversible with respect to $\mu^+$.\par In order to obtain the upper bound we write the r.h.s. as: $$\int d\mu^+(\s )(\s (0)+1)E(\s_t(0))\,-\,(m^*(\b ))^2\,-\,m^*(\b )\Eq(3.17bis)$$ and we observe that, by the monotonicity (1.24), (1.25), for any $L$ and any $t$: $$E(\s_t(0))\,\leq\, E_{V_L,+}^{+}(\s_t(0))\Eq(3.18)$$ where $E_{V_L,+}^{+}$ denotes the expectation over the HB-dynamics in $V_L$ with plus boundary conditions starting from the configuration identically equal to plus one.\par In turn, the r.h.s. of \equ(3.18) can be bounded above, using the estimate (1.19), by: $$E_{V_L,+}^{+}(\s_t(0))\,\leq\,\mu_{V_L}^+(\s (0))\;+\;\exp ( C\b L^2\,-\,t\hbox{gap}(HB,V_L,+))\Eq(3.19)$$ If we plug \equ(3.18) into \equ(3.17bis) and we use \equ(3.19), we obtain that the r.h.s. of \equ(3.17) is bounded above by: $$(\mu_{V_L}^+(\s (0))\,-\,m^*(\b ))(m^*(\b )+1)\;+\;2\exp ( C\b L^2\,-\,t\hbox{gap}(HB,V_L,+))\Eq(3.20)$$ As it is well known $$0\,\leq \,\mu_{V_L}^+(\s (0))\,-\,m^*(\b )\,\leq\,C_1\exp (-mL)\Eq(3.21)$$ for any large enough $\b$ where $C_1$ and $m$ are suitable constants with $m\to \infty$ as $\b\to \infty$.\par We now choose the size $L$ depending on $t$ as: $$L\;=\;[{\log (t)\over 2C(\a )\b}]^\a\Eq(3.22)$$ where $C(\a )$ is the constant appearing in theorem 3.1 for the value $\e\,=\,{2-\a\over 2\a }$ and we apply theorem 3.1 to get that the r.h.s. of \equ(3.20) is bounded from above by: $$C_1\exp (-m[{\log (t)\over 2C(\a )\b}]^\a )\;+\;C_2\exp (-{\sqrt (t)\over 2})\Eq(3.23)$$ for all $\b$ large enough, where $C_2$ is a suitable constant.\par Clearly \equ(3.23) proves the theorem.\pagina %\input formato.tex \numsec=4 \numfor=1 {\bf Section 4}\par \centerline{\bf Asymptotics of the Gap With Open Boundary Conditions}\bigskip In this section we again consider the HB-dynamics in a square $V\,\equiv\,V_L$ of side $L$ at very low temperature, but this time with open boundary conditions, that is $$U^{\partial V_L}(x,y)\,=\,0\quad \forall \;(x,y)\,\in \, \partial V_L$$ In this case the two extremal configurations, $(+)$ and $(-)$, are the only absolute minima of the energy $H_V^\emptyset(\s_V)$ and they are related one to the other by a global spin flip.\par We show that, due precisely to the above symmetry, the gap of the HB-dynamics is much smaller, as $L\to \infty$, than its value with plus boundary conditions. More precisely we obtain that the gap is of the order of $\exp (-\beta \t_\b L)$, where $\t_\b$ is the surface tension defined in (1.11) with respect to an interface parallel to one of the coordinate axes.\par Since the proof of the main result of the present section (see theorem 4.1 below) will mimick as close as possible the proof of theorem 3.1, we will keep the same notation of section 3 with the following modification.\par Let $R$ be a rectangle and let us suppose that we have a boundary coupling $U^{\partial R}$ which is constant on each of the four components of $\partial R$ ordered clockwise starting from the bottom. Let us denote by $0\leq \delta_i\leq 1,\quad i=1\dots 4$, the value of the boundary coupling on the $i\hbox{-th}$ side of $R$. Then we will write $\mu_R^{\delta_1\t_1,\delta_2\t_2,\delta_3\t_3,\delta_4\t_4}$, to denote the corresponding Gibbs measure on $R$ with the boundary conditions $\t_1,\t_2,\t_3,\t_4$. As usual, if one the $\delta_i$'s is equal to one it will be omitted in the notation, while if it is zero the corresponding term $\delta_i\t_i$ will be replaced by $\emptyset$. Thus for example ($\t_1,\delta +,\emptyset ,\delta +$) means $\t_1$ boundary conditions on the bottom side, plus boundary conditions on the vertical ones coupled to the interior of $R$ by a constant boundary coupling equal to $\delta$ and open boundary condition on the top one.\par As in section 3, whenever confusion does not arise, we will omit the subfix $V$ in the notation $\s_V$.\par Let us now state the main result : \bigskip {\bf Theorem 4.1}\par {\it Let $\e\,\in \, (0,1/4)$ be given. Then there exit $\beta_o\,<\,+\infty$ and $C\,<\,+\infty$ such that for any $\beta\,\geq\,\beta_o$ and any integer $L$: $$\exp (-\beta \t_\b L\,-\,C\beta L^{{1\over 2}+\e})\,\leq\,\hbox{\rm gap}_{V_L}(HB,\,\emptyset)\,\leq\, \exp (-\beta \t_\b L\,+\,C\beta L^{{1\over 2}+\e})$$} {\bf Proof}\medskip {\bf Upper Bound}. \par\noindent The idea behind the upper bound is very simple and intuitive: when the system starts from a typical configuration of the Gibbs measure $\mu_V^\emptyset$ it has a magnetization $m$ approximately equal to either $+m^*(\beta )$ or $-m^*(\beta )$, where $m^*(\beta )$ is the value of the spontaneous magnetization at inverse temperature $\beta$ in the infinite volume limit. Therefore, in order to reach the equilibrium where the expected value of the magnetization is zero by symmetry, the process has to hit the set of configurations of zero magnetization. Since the probability starting at equilibrium to have at a given time $t$ zero magnetization is equal to $\mu_V^\emptyset(m=0)$, one expects the relaxation time to equilibrium, which is roughly the inverse of the gap, to be at least as large as the inverse of $\mu_V^\emptyset(m=0)$. That is actually correct and the argument, thanks to a basic result of Shlosman (see theorem 4.2 below), gives a correct upper bound.\par Let us implement the above idea. Without a true loss of generality we may assume that $L^2$ is odd. We also denote by $m(\s )$ the total magnetization of the configuration $\s\,\in \,\O_V$: $$m(\s )\;=\; \sum_{x\in V}\s (x)$$ and by $<\,;\,>$ the scalar product in $L^2(\O_V,d\mu_V^\emptyset )$.\par If we recall that the generator of the dynamics, $L^\emptyset$, is selfadjoint on $L^2(\O_V,d\mu_V^\emptyset )$, we get that: $$\;\leq$$ $$\leq \;\exp(-\hbox{gap}_{V_L}(HB,\,\emptyset)t)\,\leq\, L^4\exp(-\hbox{gap}_{V_L}(HB,\,\emptyset)t)\Eq(4.1)$$ since, by simmetry, $\,=\,0$. \par On the other hand, again by symmetry $$(\exp (tL^{\emptyset})m)(\s )\,=\,-(\exp (tL^{\emptyset})m)(-\s )\Eq(4.2)$$ so that: $$\;=\;2\int_{\s ;\,m(\s )\geq 0}d\mu_V^\emptyset (\s )m(\s )(\exp (tL^{\emptyset})m)(\s )\Eq(4.3)$$ If we denote by $T^{(m<0)}(\s )$ the first hitting time of the set $\{\,m(\s )\,<\,0\,\}$ for the HB-dynamics in $V$ starting at time $t=0$ from the configuration $\s$, we get that, for configurations $\s$ with positive magnetization, $(\exp (tL^{\emptyset})m)(\s )$ can bounded from below by $$(\exp (tL^{\emptyset})m)(\s )\,\geq \,P(T^{(m<0)}(\s )\,>\,t)\;-\;L^2P(T^{(m<0)}(\s )\,\leq\,t)\,=$$ $$=\, 1\,-\,(L^2+1)P(T^{(m<0)}(\s )\,\leq\,t)\Eq(4.4)$$ since $\inf_{\s ;m(\s )\geq 0}m(\s )\,=\,1$ in view of our condition that $L^2$ is odd .\par A rather standard computation in the theory of Glauber dynamics that uses the invariance of the measure $\mu_V^\emptyset (\s )$ and the fact that $$P(\nu_t\,\geq \,2L^2t)\;\leq\; \exp (-KL^2t)$$ for a suitable constant $K$, where $\nu_t$ is the number of updatings within time $t$, shows that $$\sum_{\s}\mu_V^\emptyset (\s )P(T^{(m<0)}(\s )\,\leq\,t)\;\leq \; 2L^2t\mu_V^\emptyset (m(\s )\,=\,1)\;+\;\exp (-KL^2t)\Eq(4.5)$$ If we insert \equ(4.4) and \equ(4.5) in \equ(4.3), we get that: $$\;\geq$$ $$\geq\; 2\mu_V^\emptyset (m(\s )\geq 0)\;-\;4(L^2+1)L^2t\mu_V^\emptyset (m(\s )\,=\,1)\,-\,2(L^2+1)\exp (-KL^2t)\Eq(4.6)$$ By simmetry $\mu_V^\emptyset (m(\s )\geq 0)\,=\,{1\over 2}$, so that, for all $L$ large enough and all $$1\,\leq \,t\,\leq \, [16(L^2+1)L^2t\mu_V^\emptyset (m(\s )\,=\,1)]^{-1}\Eq(4.6bis)$$ the r.h.s of \equ(4.6) is greater than ${1\over 4}$.\par If we combine this result with \equ(4.1) we obtain: $${1\over 4}\;\leq\;L^4\exp(-\hbox{gap}_{V_L}(HB,\,\emptyset)t)\quad \forall\; 1\,\leq \,t\,\leq \, [16(L^2+1)L^2t\mu_V^\emptyset (m(\s )\,=\,1)]^{-1}\Eq(4.6tris)$$ We use at this point a fundamental result due to Shlosman (see Theorem 3 in [Sh]) in his study of the Wulff shape in a finite square with periodic or open boundary conditions:\bigskip {\bf Theorem 4.2 } (Shlosman)\par {\it There exists $\beta_o$ such that for any $\beta\,\geq \,\beta_o$ and any sequence of integers $\rho_L$, $L\,\in \,{\bf N}$, satisfying $$\lim_{L\to \infty}{\rho_L\over L^2}\;=\;\rho\;\in \;(0,m^*(\beta ))\qquad \rho_L\,-\,L^2\;=\;\hbox{ mod } \,2$$ the limit $$\psi (\rho )\;=\; \lim_{L\to \infty}-{1\over \beta L}\log (\mu_V^\emptyset (m(\s )\,=\,\rho_L))$$ exists and it is given by: $$\eqalign{\psi (\rho )\;&=\;{1\over 2}w\sqrt{{m^*(\beta )\,-\,\vert \rho\vert\over 2m^*(\b )} }\quad \vert \rho\vert\,\geq\,\rho_1,\cr \psi (\rho )\;&=\;{1\over 2}w\sqrt{m^*(\beta )\,- \,\rho_1}\quad \vert \rho\vert\,\leq\,\rho_1}$$ where the constant $w$ is the value of the Wulff functional $W_\t$ on the Wulff curve $\partial W$ (see theorem 1.1) and the singularity point $\rho_1$ satisfies the equation $${1\over 2}w \sqrt{{m^*(\beta )\,-\,\vert \rho_1\vert\over 2m^*(\b )}}\;=\;\t_\beta$$} {\bf Warning} Due to some misprints, the formula for $\psi (\rho )$ in [Sh] appears with $1\over 2$ and $\sqrt{{m^*(\beta )\,-\,\vert \rho\vert\over 2m^*(\b )} }$ replaced by $1\over 4$ and by $\sqrt{m^*(\beta )\,-\,\vert \rho\vert}$ respectively.\bigskip {\bf Remark} Given $\e\,\in \,(0,{1\over 4})$ and $\beta$ large enough, it is possible to show, using the methods of [DKS], that the above limit is approached, as $L\to \infty$, at least as fast as $L^{-{1\over 2}+\e}$.\bigskip By plugging in \equ(4.6tris) the result of theorem 4.2 and its strengthening mentioned in the remark above , we immediately obtain the required upper bound on the gap. \bigskip {\bf Remark} Actually the above reasoning leads to an upper bound on the gap which is a negative exponential of the surface in {\it any} dimension $d\geq 2$ if we use the estimate of Schonmann [Sch2]: $$\mu_V^\emptyset (m(\s )\,=\,0)\,\leq \,\exp (-c(\beta )L^{d-1})$$ for a suitable constant $c$. Moreover it is possible to show (see [CCSch]) that in two dimensions the above estimate is valid for {\it any} $\b$ larger than the critical value $\b_c$. Therefore, using corollary 2.1 and the above observation, we get that in $d=2$ for any $\b\,>\,\b_c$ there exist two constants $c_1$ and $c_2$ such that for any $L$ large enough: $$\exp (-c_1L)\,\leq\,\hbox{gap}_{V_L}(HB,\emptyset )\,\leq\,\exp (-c_2L)$$ It would be nice to show that at least one of the two constants is equal to $\b\t_\b$.\par We finally notice that it was possible to follow a slightly different proof by using in the variational characterization of the gap the trial function $$f(\s )\,=\,\chi (m(\s )>0)\,-\,\chi (m(\s )<0)$$ $\chi (A)$ being the characteristic function of the event $A$, and then exploiting Shlosman's result.\bigskip \medskip {\bf Lower Bound} \par\noindent We start by replacing the open boundary conditions on $\partial V$ by very weak {\it plus} boundary conditions. More precisely, let $$\delta\;=\;L^{-{1\over 2}}$$ and let us consider a constant boundary coupling $U^{\partial V}(x,y)$: $$U^{\partial V}(x,y)\,=\,\delta\qquad \forall \;(x,y)\in \partial V$$ Then, in the notation established in section 1 $\S 3$ and at the beginning of the present section, we trivially have for any $a\in \{-1,+1\}$: $$\eqalign{\exp (-8\beta \d L)\mu_V^{\delta +,\delta +,\delta +,\delta +} (\s )\,&\leq\, \mu_V^\emptyset (\s )\;\leq\;\exp (8\beta \d L)\mu_V^{\delta +,\delta +,\delta +,\delta +}(\s )\cr \exp (-8\beta\d L)\mu_{\{x\}}^{U^{\partial x},(\s ,\delta +)} (a)\,&\leq\, \mu_{\{x\}}^{(\s ,\emptyset)} (a)\;\leq\;\exp (8\beta \d L)\mu_{\{x\}}^{U^{\partial x},(\s , \delta +)} (a)\,}\Eq(4.7)$$ where $\mu_{\{x\}}^{(\s , \emptyset)}$ is the conditional probability of having the value $a$ for $\s (x)$ given that outside $V$ there are open boundary conditions and that the configuration in $V\setminus \{x\}$ is $\s$. Similarly for $\mu_{\{x\}}^{U^{\partial x},(\s , \delta +)} (a)$.\par It is immediate to check, using the variational characterization of the gap in term of the Dirichlet form (1.17), that \equ(4.7) implies the following bound on $\hbox{gap}_{V_L}(HB,\,\emptyset)$ in terms of $\hbox{gap}_{V_L}(HB,\,+,\, \delta)$: $$\hbox{gap}_{V_L}(HB,\,\emptyset)\;\geq\; \exp (-32\beta \d L)\hbox{gap}_{V_L}(HB,\,+,\, \delta)\Eq(4.8)$$ It is therefore sufficient to establish the correct lower bound with "$\delta$+" boundary conditions.\par To this purpose we proceed exactly as in section 3, namely we consider the $\{Q_i\}$-dynamics with $Q_i$ as in the proof of theorem 3.1 and estimate $\hbox{gap}_{V_L}(HB,\,+,\, \delta)$ by: $$\hbox{gap}_{V_L}(HB,\,+,\, \delta)\;\geq\; {1\over \vert Q_i\vert}{\exp (-4\beta)\over \exp (-4\beta)+\exp (+4\beta)} \,\exp (-4\beta (l\,+\,1))\,\hbox{gap}_{V_L}(\{Q_i\},\,+,\,\delta)\Eq(4.9))$$ where $l\;=\;2[L^{{1\over 2}+\e}]$.\par The main difference now with the reasoning behind the proof of theorem 3.1 is the following.\par\noindent When we start from the two extremal configurations $(+)$ and $(-)$ at the beginning of an ordered sequence $S_N$ and we update the first rectangle $Q_1$, we replace $(+)_{Q_1}$ and $(-)_{Q_1}$ with two configurations $\eta_{Q_1}^+$ and $\eta_{Q_1}^-$ distributed according to $\mu_{Q_1}^{\delta +,\delta +,+,\delta +}$ and $\mu_{Q_1}^{\delta +,\delta +,-,\delta +}$ respectively. Contrary to the "full" (i.e. $\delta \,=\,1$) plus boundary conditions discussed in theorem 3.1, the measure $\mu_{Q_1}^{\delta +,\delta +,-,\delta +}$ is {\it not} concentrated for $\beta$ large on configurations which resemble those of the plus phase, at least far from the top side, but instead, due precisely to the "full" minus boundary condition on the top side, on configurations in which the spins are mostly minus one with little islands of plus spins. Therefore, with large probability, the first updating of the ordered sequence will {\it not} force $(+)$ and $(-)$ to agree in a large portion (e.g. $3\over 4$ ) of $Q_1$.\par We notice, however, that with very small probability the two new configurations $(+)^{\eta_{Q_1}^+}$ and $(-)^{\eta_{Q_1}^-}$ {\it will} agree in, say, $3\over 4$ of $Q_1$, if for example the interface in the configuration $\eta_{Q_1}^-$ separating the minus spins on the top side of $Q_1$ from the plus spins in the rest of the boundary instead of being in its typical position, namely close to the bottom side of $Q_1$, is very close to the top one. It turns out that the probability in question is at least of the order of $\exp (-\beta \t_\beta L )$ . Once this rare event has occurred then, in the second updating, we will have to consider the Gibbs measures $\mu_{Q_1}^{\eta_{Q_1}^+,\delta +,+,\delta +}$ and $\mu_{Q_1}^{\eta_{Q_1}^+,\delta +,-,\delta +}$ which, if we approximate, as we did in the introduction to the proof of theorem 3.1, the boundary condition $\eta_{Q_1}^+$ with a "full" plus, become $\mu_{Q_1}^{+,\delta +,+,\delta +}$ and $\mu_{Q_1}^{+,\delta +,-,\delta +}$.\par Now the situation is very different from the first updating and much more similar to the case treated in the proof of theorem 3.1. In fact, in the Gibbs measure $\mu_{Q_1}^{+,\delta +,-,\delta +}$, the "full" plus boundary condition on the bottom side compensate exactly the "full" minus boundary condition on the top one and therefore the "phase" (that is the structure of the typical configurations) is decided by the lateral "$\delta$+" boundary conditions. Since the typical fluctuations of the interface separating the minus spin of the top from the plus spins at the bottom are of order $\sqrt L\,<<\,l$, and since $\d l\,= \,L^\e\,>>\,1$, one can conclude (see Proposition 4.1 below) that the above two Gibbs measures are very similar in, say, $3\over 4$ of $Q_1$. Thus the second updating will, with large probability, enlarge the region of agreement between the evoluted of $(+)$ and $(-)$ to $$\{\,x\in V\quad 0\leq x_2\leq {5\over 4}l\,\}$$ Iterating this procedure, we see that an ordered sequence $S_N\,=\,\{t_1,\dots \t_N\}$ will typically glue together $(+)$ and $(-)$ with the last updating at time $t_N$, provided that in the first one, at time $t_1$, a very rare event of probability of order $\exp (-\beta \t_\beta L )$ has occured.\par Clearly the above reasoning implies that the relaxation time to equilibrium for the $\{Q_i\}$-dynamics should be at most of order $\exp (+\beta \t_\beta L )$ and therefore, using \equ(4.9), the required lower bound would follow.\par Let us implement the above program. We start by giving a generalization to the case of $\delta +$ lateral boundary conditions of proposition 3.1. As in section 3, let $R$ be a rectangle $$R\,=\,\{\;x\in \Z;\,0\,\leq \,x_1\,\leq\,L_1\quad 0\,\leq\,x_2\,\leq\,L_2\;\}$$ with $L_1\,\geq \,L_2\,\geq\,L_1^{{1\over 2}+\e}$. Then we have: \bigskip {\bf Proposition 4.1}\par {\it Let $m\,>\,0$ and $\e\,\in \, (0,{1\over 2})$ be given and let $\d\,=\,L_1^{-{1\over 2}}$. Then there exists $\beta_o\,\equiv\,\beta_o(\e,m )$ such that for all $\beta \,\geq \,\beta_o$ and all $x\,=\,(x_1,x_2)\, \in \,R$ with $x_2\,\leq \,{3\over 4}L_2$ we have: $$\mu_R^{+,\delta +,+,\delta +}(\s (x)=1)\;-\;\mu_R^{+,\delta +,-,\delta +}(\s (x)=1)\;\leq \;\exp (-m L_1^{\e})$$ Moreover, if $R$ and $R'$ are two rectangles as above with the same basis $L_1$ but different heights $L_1\,\geq \,L_2\,\geq\,L_2'\,\geq \,L_1^{{1\over 2}+\e}$, then for all $x\,=\,(x_1,x_2)\, \in \,R$ with, for example, $x_2\,\leq \,{1\over 16}L_2'$, we have: $$\mu_{R'}^{\delta +,\delta +,+,\delta +}(\s (x)=1)\;-\;\mu_R^{\delta +,\delta +,+,\delta +}(\s (x)=1)\;\leq \;\exp (-m L_1^{{1\over 2}+\e})$$} For a proof see appendix 1.\par There is an interesting corollary to the above proposition that can be viewed as a generalization of theorem 3.1 to the case when we have open boundary conditions on three sides of the square $V_L$ and full $+$ boundary conditions on the remaining one.\bigskip {\bf Corollary 4.1}\par {\it Let $\e\,\in \, (0,{1\over 2})$ be given. Then there exit $\beta_o\,<\,+\infty$ and $C\,<\,+\infty$ such that for any $\beta\,\geq\,\beta_o$ and any integer $L$: $$\hbox{gap}_{V_L}(HB,\,\emptyset ,\emptyset ,+,\emptyset )\;\geq\;\exp (-C\beta L^{{1\over 2}+\e})$$} {\bf Proof}\par We use \equ(4.8) to replace the open boundary conditions on the three sides by $\d +$ boundary conditions. Then we can repeat word by word the proof of theorem 3.1, with proposition 3.1 replaced by proposition 4.1 .\bigskip The second new result that we need is as follows.\par For a given rectangle $R$ as above and $\s\,\in \, \O_R$, let $\G^{+,+,-,+} (\s)$ be the family of contours of $\s$ with boundary condition $\t$ having the constant sign $+,+,-,+$ on the external boundary of the four sides of $R$ ordered in the usual way. As one can immediately check, under the above boundary conditions there exists only one open contour that will be denoted by $\Gamma_{R,open}^{+,+,-,+}(\s )$.\par We then define the event ${\cal A}_R^{+,+,-,+}$ as: $$ {\cal A}_R^{+,+,-,+}\;=\;\{\s ;\quad \Gamma_{R,open}^{+,+,-,+}(\s )\,\subset\, \{\;x\in R;\;x_2\,>\,{13L_2\over 16}\}\;\} \Eq(4.10)$$ {\bf Proposition 4.2} \par {\it In the hypotheses of proposition 4.1 there exists a positive constant $C$ independent of $\beta$ and $L_1$ such that: $$\mu_{R}^{\delta +,\delta +,-,\delta +}({\cal A}_R^{+,+,-,+})\;\geq\;\exp (-\beta \t_\beta L_1\,-\,C\beta L_1^{\e} )$$} For a proof see appendix 1.\medskip We are now in a position to complete the proof of the lower bound.\par As a first step and for reasons that will appear clear later in the proof, it is convenient to modify slightly the coupling for the $Q_i$-dynamics. More precisely we use the same algorithm described (1.24), (1.25), but with a modified coupled measure $\tilde \nu_{Q_1}$ for the first rectangle $Q_1$. The measure $\tilde \nu_{Q_1}$, that will be obtained from the old one $ \nu_{Q_1}$ via "surgery" (see for instance [DSh]) on a suitable subset of $Q_1$, will however still enjoy the monotonicity property described at the end section 1.\par Let $\tilde R_1$ be the rectangle: $$\tilde R_1\,=\, \{\,x\in Q_1;\quad x_2\,\leq \,l-{3l\over 16}\,\}$$ and let $\nu_{Q_1}^{\t^{(1)}\dots \t^{(N)}}$ be the measure on $(\O_{Q_1})^N$, $N\,=\,2^L$, constructed in $\S 4$ of section 1, with boundary conditions $\delta +$ on the bottom and lateral sides of $Q_1$ and $\t^{(1)}\dots \t^{(N)}$ on the top side of $Q_1$.\par We then construct the new measure $\tilde \nu_{Q_1}$ on $\O_{Q_1}^{N}$ as follows.\par Given $N$ configurations $\s^{(1)}\dots \s^{(N)}$ in $\O_{\cal L}$, where $\cal L$ denotes the external boundary of the top side of $\tilde R_1$, let $\nu_{Q_1\setminus {\cal L}}^{\s^{(1)}\dots \s^{(N)};\t^{(1)}\dots \t^{(N)}}$ be the measure constructed according to (1.20), (1.21) for the set $Q_1\setminus {\cal L}$ and boundary conditions:\medskip \item{} $\delta +$ on the bottom and lateral sides of $Q_1$ \item{} $\s^{(1)}\dots \s^{(N)}$ on $\cal L$ \item{} $\t^{(1)}\dots \t^{(N)}$ on the top side of $Q_1$, where $\t^{(1)}\dots \t^{(N)}$ are {\it all} possible configurations on the external boundary of the top side of $Q_1$.\medskip\noindent It is very important to notice that $\nu_{Q_1\setminus {\cal L}}^{\s^{(1)}\dots \s^{(N)};\t^{(1)}\dots \t^{(N)}}$ is a product of the measures $$\nu_{\tilde R_1}^{\s^{(1)}\dots \s^{(N)}}\; \hbox{ and }\; \nu_{Q_1\setminus \{\tilde R_1\cup {\cal L}\}}^{\s^{(1)}\dots \s^{(N)};\t^{(1)}\dots \t^{(N)}}$$ where, for notation convenience, we have omitted to indicate the fixed $\delta +$ boundary conditions on the bottom and lateral sides of $Q_1$.\par Finally, given $N$ configurations $\tilde \s^{(1)}\dots \tilde \s^{(N)}$ in $\O_{Q_1}$, we set: $$\tilde \nu_{Q_1}^{\t^{(1)}\dots \t^{(N)}}(\tilde \s^{(1)}\dots \tilde \s^{(N)})\,=\,\sum_{\s^{(1)}\dots \s^{(N)}}\nu_{Q_1}^{\t^{(1)}\dots \t^{(N)}}(\s^{(1)}\dots \s^{(N)})\,T(\s^{(1)}\dots \s^{(N)};\;\tilde \s^{(1)}\dots \tilde \s^{(N)})\Eq(4.10bis)$$ where $$\eqalign{T(\s^{(1)}\dots \s^{(N)};\;\tilde \s^{(1)}\dots \tilde \s^{(N)})\;&=\;\nu_{Q_1\setminus {\cal L}}^{\s^{(1)}\dots \s^{(N)};\t^{(1)}\dots \t^{(N)}}( \tilde \s^{(1)}_{Q_1\setminus {\cal L}}\dots \tilde \s^{(N)}_{Q_1\setminus {\cal L}}) \quad \hbox{if } \tilde \s^{(i)}_{\cal L}\,=\,\s^{(i)}_{\cal L}\cr T(\s^{(1)}\dots \s^{(N)};\;\tilde \s^{(1)}\dots \tilde \s^{(N)})\;&=\;0\quad \hbox{otherwise}}\Eq(4.10tris)$$ It is easy to see, using the DLR equations, that if the event $A$ depends only on the $k^{th}$ configuration, $\tilde \s^{(k)}$, then $$\tilde \nu_{Q_1}^{\t^{(1)}\dots \t^{(N)}}(A)\,=\,\mu_{Q_1}^{\delta +,\delta +,\t^{(k)},\delta +}(A)\Eq(4.10quatris)$$ and moreover that, if the event $A\subset (\O_{Q_1})^N$ depends {\it only} on the values of the spins in $Q_1\setminus {\cal L}$, then: $$\tilde \nu_{Q_1}^{\t^{(1)}\dots \t^{(N)}}(A)\,=\,\sum_{\s^{(1)}\dots \s^{(N)}}\nu_{Q_1}(\s^{(1)}\dots \s^{(N)})\nu_{Q_1\setminus {\cal L}}^{\s^{(1)}\dots \s^{(N)};\t^{(1)}\dots \t^{(N)}}( A)\Eq(4.10five)$$ Finally, it is immediate to check, using the monotonicity (1.23) of the measures $\nu_{Q_1}^{\t^{(1)}\dots \t^{(N)}}$ and $\nu_{Q_1\setminus {\cal L}}^{\s^{(1)}\dots \s^{(N)};\t^{(1)}\dots \t^{(N)}}$, that (1.23) holds true also for $\tilde \nu_{Q_1}^{\t^{(1)}\dots \t^{(N)}}$. This fact implies, in particular, that if we use the coupling (1.24), (1.25) with the measures $\tilde \nu_{Q_1}$, $ \nu_{Q_2}$, ... $ \nu_{Q_n}$, then, under this new coupling, any ordered set of initial configurations will stay ordered at any future time. \medskip Let now $S_N\;\equiv\;\{t_1\,\dots \,t_N\}$ be a fixed ordered sequence with $t_1\,=\,0$, let $A_i(x)$, $A_i$, be the events defined in (3.4) and let $q_i\,=\,P(A_i\vert {\cal B})$ where $${\cal B}\,=\,\{((-)^{\{Q_i\},\delta +}_{t_1} )_{Q_1}\;\in \;{\cal A}_{Q_1}^{+,+,-,+}\,\}\Eq(4.11)$$ As in section 3 we have: $$q_{n+1}\,\leq \,q_n\;+\;P(\,A_{n+1}\cap A_n^c\,\vert {\cal B}) \Eq(4.12)$$ Let us estimate the second term in the r.h.s. of \equ(4.12). As in section 3 we let $$R_n\;=\;\{ x\in \cup_{j\leq n}Q_j;\;x_2\leq (n+1){l\over 2}-[{l\over 4}]\}$$ and $$D\,=\,\cup_{j\geq 2}Q_j$$ We observe that, since the sequence $S_N$ is ordered, if $A_n^c$ has occurred for some $n\geq 1$ then necessarily $(-)^{\{Q_i\},\delta +}_{t_1}$ and $(+)^{\{Q_i\},\delta +}_{t_1}$ are both equal on the external boundary of the bottom side of $Q_2$ to a common configuration that we call $\t$. Again because of the ordering of the sequence $S_N$, the next updatings at time $t_i$, $i\,\geq\, 2$, will not modify $\t$ on the external boundary of the bottom side of $Q_2$ and therefore they will be reversible with respect to the Gibbs measure $\mu_D^{\t,\d +,\d +,\d +}$ on $\O_D$, .\par Thus, following section 3 (see (3.6) ... (3.9)), we can bound $P(\,A_{n+1}\cap A_n^c\,\vert {\cal B})$ by: $$\sum_\t\tilde \nu_{Q_1}^{\t^{(1)}\dots \t^{(N)}}((-)^{\{Q_i\},\delta +}_{t_1}= (+)^{\{Q_i\},\delta +}_{t_1}=\t\vert{\cal A}_{Q_1}^{+,+,-,+} )\, F_1(\t )\,\leq\,$$ $$\sum_\t\mu_{Q_1}^{\d +,\d +,-,\d +}(\t\vert{\cal A}_{Q_1}^{+,+,-,+} )\, F_1(\t )$$ where $$F_1(\t )\,\equiv\, \sum_{x\in R_{n+1}\cap Q_{n+1}\atop\s\in \O_D}\mu_D^{\t,\d +,\d +,\d +}(\s )\,[ \mu_{Q_{n+1}}^{\s,\d +,+,\d +}(\eta (x)=1)\,-\, \mu_{Q_{n+1}}^{\s,\d +,-,\d +}(\eta (x)=1)]\Eq(4.15)$$ As in (3.9)$\dots$(3.11) we get, by monotonicity and the DLR equations, that $F_1(\t )$ is bounded from above by: $$F_2(\t )\,=\, \sum_{\vbox{\eightpoint{\hbox{$x\in R_{n+1}\cap Q_{n+1}$}}}}[ \mu_{R_{n+1}\cup Q_{n+1}\setminus Q_1}^{\t,\d +, +,\d +}(\s )(\eta (x)=1)\,-\,\mu_{R_{n+1}\cup Q_{n+1}\setminus Q_1}^{\t,\d +, -,\d +}(\s )(\eta (x)=1) ]\Eq(4.17)$$ In conclusion we have shown that: $$P(\,A_{n+1}\cap A_n^c\,\vert {\cal B})\,\leq\,\sum_\t\mu_{Q_1}^{\d +,\d +,-,\d +} (\t\vert {\cal A}_{Q_1}^{+,+,-,+} )\, F_2(\t )\Eq(4.17bis)$$ In order to prove that \equ(4.17bis) is very small, we need a last result on the Ising model which shows that, conditional to the event ${\cal A}_{Q_1}^{+,+,-,+}$, the projection (or relativization) of the measure $\mu_{Q_1}^{\d +,\d +,-,\d +}$ on the external boundary of the bottom side of $Q_2$ is, in some sense, very close to the same projection both of the measure $\mu_{R_{n+1}\cup Q_{n+1}}^{\d +,\d +, +,\d +}$ and of the measure $\mu_{R_{n+1}\cup Q_{n+1}}^{ +,\d +, -,\d +}$. More precisely: \bigskip {\bf Proposition 4.3 } \par {\it Let $m\,>\,0$ and $\e\,\in \, (0,{1\over 2})$ be given. Then there exists $\beta_o\,\equiv\,\beta_o(\e ,m )$ such that for all $\beta \,\geq \,\beta_o$ we have:\bigskip\noindent {\bf a)} $$\vert \sum_\t\mu_{Q_1}^{\d +,\d +,-,\d +}(\t\vert {{\cal A}_{Q_1}^{+,+,-,+}} )F_2(\t ) \,-\, \sum_\t \mu_{R_{n+1}\cup Q_{n+1}}^{\d +,\d +, +,\d +}(\t )F_2(\t )\vert \;\leq $$ $$\leq \;\exp (-m L^{{1\over 2}+\e})$$ \noindent {\bf b)} $$\vert \sum_\t\mu_{Q_1}^{\d +,\d +,-,\d +}(\t\vert {{\cal A}_{Q_1}^{+,+,-,+}} )F_2(\t ) \,-\, \sum_\t \mu_{R_{n+1}\cup Q_{n+1}}^{+,\d +, -,\d +}(\t )F_2(\t )\vert \;\leq $$ $$\leq \;\exp (-mL^\e)$$} For a proof see appendix 1.\par Using proposition 4.3 and the DLR equations for $\mu_{R_{n+1}\cup Q_{n+1}}^{+,\d +, -,\d +}$ and $\mu_{R_{n+1}\cup Q_{n+1}}^{\d +,\d +, +,\d +}$, we get that \equ(4.17bis) is bounded from above by: $$2\exp (-mL^\e)\;+\;\sum_{\vbox{\eightpoint{\hbox{$x\in R_{n+1}\cap Q_{n+1}$}}}}[\mu_{R_{n+1}\cup Q_{n+1}}^{\d +,\d +, +,\d +}(\eta (x)=1)\;-\;\mu_{R_{n+1}\cup Q_{n+1}}^{+,\d +, -,\d +}(\eta (x)=1)]\Eq(4.18)$$ In turn, using the fact that $$\mu_{R_{n+1}\cup Q_{n+1}}^{\d +,\d +, +,\d +}(\eta (x)=1)\;\leq\;\mu_{R_{n+1}\cup Q_{n+1}}^{+,\d +, +,\d +}(\eta (x)=1)$$ and applying proposition 4.1 to the rectangle $R_{n+1}\cup Q_{n+1}$, we get that \equ(4.18) can be bounded from above by: $$3\exp (-mL^\e)\Eq(4.19)$$ for any given $m\,>\,0$ and $\e\,\in\,(0,{1\over 2})$, provided that $\beta$ is large enough depending on $m$ and $\e$.\par In conclusion we have shown that: $$P(\,A_{n+1}\cap A_n^c\,\vert {\cal B})\,\leq \, 3\exp (-mL^{\e}) \Eq(4.20)$$ In order to conclude that $q_N$ is small, we need to control the first term $q_1$ since $$q_N\,\leq\,q_1\;+\;\sum_n^{N-1}P(\,A_{n+1}\cap A_n^c\,\vert {\cal B})$$ {\bf Proposition 4.4} \par {Let $m\,>\,0$ and $\e\,\in \, (0,{1\over 2})$ be given. Then there exists $\beta_o\,\equiv\,\beta_o(\e,m )$ such that for all $\beta \,\geq \,\beta_o$ we have: $$q_1\,\leq\,\exp (-mL^{{1\over 2}+\e})$$} {\bf Proof}\par Let $\nu\,\equiv\,\tilde \nu_{Q_1}$ be the measure on $\O_{Q_1}^{2^L}$ constructed in \equ(4.10bis). By monotonicity in the initial configuration and by the definition of the event $A_1$, we can estimate $q_1$ from above by: $$q_1\,\leq\,\sum_{x\in R_1}[\nu (\s^{(N)}(x)\,=\,1\vert \{ \s^{(1)}\in {\cal A}_{Q_1}^{+,+,-,+}\})\,-\, \nu (\s^{(1)}(x)\,=\,1\vert \{ \s^{(1)}\in {\cal A}_{Q_1}^{+,+,-,+}\}) ]\Eq(4.21)$$ where we used the convention that $\s^{(1)}$ and $\s^{(N)}$ are the components of a generic configuration $\tilde \s\,\in \,\O_{Q_1}^{2^L}$ corresponding respectively to the minimal ($-$) and maximal ($+$) boundary condition on the top side of $Q_1$.\par Let us examine separately each one of the two terms appearing in the sum in the r.h.s. of \equ(4.21).\par Because of \equ(4.10quatris), the second term $\nu (\s^{(1)}(x)\,=\,1\vert \{ \s^{(1)}\in {\cal A}_{Q_1}\} )$ is equal to: $$\nu (\s^{(1)}(x)\,=\,1\vert \{ \s^{(1)}\in {\cal A}_{Q_1}^{+,+,-,+}\})\,=\, \mu_{Q_1}^{\delta +,\delta +,-,\delta +}(\s (x)=1\vert {\cal A}_{Q_1}^{+,+,-,+})\Eq(4.22)$$ Since the event ${\cal A}_{Q_1}^{+,+,-,+}$ implies that the entire unique open contour of the configuration $\s$ is {\it outside} $ R_1$, it is immediate to check, using the monotonicity of the Gibbs measure with respect to an increase of the boundary conditions, that: $$\mu_{Q_1}^{\delta +,\delta +,-,\delta +}(\s (x)=1\vert {\cal A}_{Q_1}^{+,+,-,+})\,\geq \, \mu_{Q_1}^{\delta +,\delta +,+,\delta +}(\s (x)=1)\Eq(4.23)$$ Let us now consider the first term $$\nu (\s^{(N)}(x)\,=\,1\vert \{ \s^{(1)}\in {\cal A}_{Q_1}^{+,+,-,+}\})\,=\, {\nu (\s^{(N)}(x)\,=\,1\cap \{ \s^{(1)}\in {\cal A}_{Q_1}^{+,+,-,+}\})\over \mu_{Q_1}^{\delta +,\delta +,-,\delta +}({\cal A}_{Q_1}^{+,+,-,+})}\Eq(4.24)$$ where we used, once more, \equ(4.10quatris).\par We observe that the event $\{\s (x)\,=\,1\}$, $x\in R_1$, and ${\cal A}_{Q_1}^{+,+,-,+}$ depend {\it only} on the spins in $R_1\subset \tilde R_1$ and $Q_1\setminus \{\tilde R_1\cup{\cal L}\}$ respectively, where $\tilde R_1$ and $\cal L$ have been defined right after proposition 4.2 .\par Therefore, using \equ(4.10five) and the fact that $\nu_{Q_1\setminus {\cal L}}^{\s^{(1)}\dots \s^{(N)};\t^{(1)}\dots \t^{(N)}}$ is a product of the measures $\nu_{\tilde R_1}^{\s^{(1)}\dots \s^{(N)}}$ and $\nu_{Q_1\setminus \{\tilde R_1\cup {\cal L}\}}^{\s^{(1)}\dots \s^{(N)};\t^{(1)}\dots \t^{(N)}}$, we get: $$\nu (\s^{(N)}(x)\,=\,1\cap \{ \s^{(1)}\in {\cal A}_{Q_1}^{+,+,-,+}\})\,=\,$$ $$\sum_{\s^{(1)}\dots \s^{(N)}}\nu_{Q_1}(\s^{(1)}\dots \s^{(N)})\nu_{\tilde R_1}^{\s^{(1)}\dots \s^{(N)}}(\tilde \s^{(N)}(x)\,=\,1) \nu_{Q_1\setminus \{\tilde R_1\cup {\cal L}\}}^{\s^{(1)}\dots \s^{(N)};\t^{(1)}\dots \t^{(N)}}(\tilde \s^{(1)}\,\in\, {\cal A}_{Q_1}^{+,+,-,+})\Eq(4.25)$$ We now observe that, because of (1.22) applied to $\nu_{\tilde R_1}^{\s^{(1)}\dots \s^{(N)}}$, $$\nu_{\tilde R_1}^{\s^{(1)}\dots \s^{(N)}}(\tilde \s^{(N)}(x)\,=\,1)\,=\, \mu_{\tilde R_1}^{\delta +,\delta +,\s^{(N)},\delta +}(\s (x)\,=\,1)\,\leq\, \mu_{\tilde R_1}^{\delta +,\delta +,+,\delta +}(\s (x)\,=\,1)\Eq(4.26)$$ so that the r.h.s. of \equ(4.25) becomes smaller than: $$\mu_{\tilde R_1}^{\delta +,\delta +,+,\delta +}(\s (x)\,=\,1) \nu (\s^{(1)}\in {\cal A}_{Q_1}^{+,+,-,+})\,=\,$$ $$\mu_{\tilde R_1}^{\delta +,\delta +,+,\delta +}(\s (x)\,=\,1) \mu_{Q_1}^{\delta +,\delta +,-,\delta +}(\s\in {\cal A}_{Q_1}^{+,+,-,+})\Eq(4.27)$$ where we used once more \equ(4.10quatris) to write $$\nu (\s^{(1)}\in {\cal A}_{Q_1}^{+,+,-,+})\,=\, \mu_{Q_1}^{\delta +,\delta +,-,\delta +}(\s\in {\cal A}_{Q_1}^{+,+,-,+})$$ In conclusion, from \equ(4.24)...\equ(4.27), we get that $$\nu (\s^{(N)}(x)\,=\,1\vert \{ \s^{(1)}\in {\cal A}_{Q_1}^{+,+,-,+}\})\,\leq\, \mu_{\tilde R_1}^{\delta +,\delta +,+,\delta +}(\s (x)\,=\,1)\Eq(4.28)$$ Combining finally \equ(4.23) and \equ(4.28) we bound from above the sum in \equ(4.21) by: $$\sum_{x\in R_1}[\mu_{\tilde R_1}^{\delta +,\delta +,+,\delta +}(\s (x)\,=\,1)\,-\, \mu_{ Q_1}^{\delta +,\delta +,+,\delta +}(\s (x)\,=\,1)]\,\leq\,$$ $$\exp (-mL^{{1\over 2}+\e})\Eq(4.29)$$ for any given $m$, provided that $\beta$ is large enough. In the derivation of the last inequality in \equ(4.29) we use part ii) of Proposition 4.1 and the definition of $\tilde R_1$.\bigskip If we now use proposition 4.4 together with \equ(4.20), we get that $$P(A_N\,\vert {\cal B})\,\leq\,3N\exp (-m L^{\e})\Eq(4.30)$$ We are now in a position to conclude the proof of the theorem. Given a sequence $S_N\;\equiv\;\{t_1\,\dots \,t_N\}$ of updatings we say that $S_N$ is a {\it good } sequence iff $S_N$ is ordered and the event $A_N^c$ occured at the end of the sequence. Using \equ(4.30) together with propositions 4.2, we conclude that the probability that an ordered sequence is also a good sequence is larger than $$[1\,-\,N\exp (-mL^{\e})]P({\cal B})\;\geq\;{1\over 2}\exp (-\beta L\t_\beta\,-\,C\beta L^{{1\over 2}+\e}))$$ for $L$ large enough and some constant $C$.\par Thus, using Lemma 3.1, we get that, if $T\,=\,\exp (+\beta L\t_\beta\,+\,2C\beta L^{{1\over 2}+\e}))$ and $L$ is large enough: $$P(\hbox{ there exists a good sequence in} \;[0,T]\,)\;\geq\;{1\over 3}\Eq(4.31)$$ As in section 3 \equ(4.31) immediately implies that $$ \hbox{gap}_{V_L}(\{Q_i\},\,\emptyset)\,\geq\, \exp (-\beta L\t_\beta\,-\,3C\beta L^{{1\over 2}+\e}))\Eq(4.32)$$ Clearly \equ(4.32) together with \equ(4.8) and \equ(4.9) proves the correct lower bound.\par The proof of the theorem is completed. \pagina \numsec=5 \numfor=1 {\bf Section 5}\par \centerline{\bf Rare Excursions of the Magnetization }\bigskip In this section we apply the results obtained in the previous sections to study in detail the time evolution of the magnetization $m(\s_t)$ of the process. In particular we will analyze the large fluctuations of the observable $m(\s_t)$ and prove some asymptotic results close in spirit to the results obtained by Shlosman for the static problem (see Theorem 4.2).\par The setting will be that of section 4, namely the HB-dynamics in a square $V\equiv V_L$ of side L with open boundary conditions. Although the case with $+$ boundary conditions could be treated as well without any significant modification, we decided to omit it in order not to burden too much the reader.\par Let $\rho_L$, $L\in {\bf N}$, be a sequence of integers such that: $$\lim_{L\to \infty}{\rho_L\over L^2}\;=\;\rho\;\in \;(-m^*(\beta ),m^*(\beta ))\qquad \rho_L\,-\,L^2\;=\;\hbox{ mod } \,2$$ where, as usual, $m^*(\beta )$ denotes the spontaneous magnetization, and let $\t_{\rho_L}$ be the stopping time: $$\t_{\rho_L}\,=\,\inf \{t\geq 0;\quad m(\s_t ) \,\leq\,\rho_L\,\}\Eq(5.1)$$ Then our two main results can be stated as follows:\bigskip {\bf Theorem 5.1}\par {\it There exists $\beta_o$ such that for any $\beta\,\geq \,\beta_o$ and any $\rho_L$ as above: $$\lim_{L\to \infty}{1\over \beta L}\log (\sum_{\s \atop m(\s )>0}\mu_V^\emptyset (\s )E_\s(\t_{\rho_L}))\; =\;\psi (\rho\vee 0 )$$ where the symbol $E_\s$ denotes the expectation over the HB-dynamics starting from the configuration $\s$ and the function $\psi (\rho )$ is given in theorem 4.2 .\par The same asymptotics holds if instead of starting from equilibrium with positive magnetization we start from the configuration identically equal to all pluses .}\bigskip {\bf Theorem 5.2}\par {\it There exists $\beta_o$ such that for any $\beta\,\geq \,\beta_o$ and any $\rho_L$ as above, there exist numbers $\{a_L\}_{L\in {\bf N}}$ such that for any $t\,>\,0$ : \item{{\bf a})} $$\lim_{L\to \infty}{1\over \beta L}\log (a_L)\;=\;\psi (\rho\vee 0 )$$ \item{{\bf b})}$$ \lim_{L\to\infty}\sum_{\s \atop m(\s )>0}\mu_V^\emptyset (\s )P_\s(\t_{\rho_L}\,>\,ta_L)\;=\;\exp (-t)$$ An analogous result holds if instead of starting from equilibrium we start from the configuration identically equal to all pluses .}\bigskip {\bf Remark} Theorem 5.2 says that, under a suitable rescaling determined by the numbers $a_L\,\approx\,\exp (\beta L\psi (\rho ))$, the stopping time $\t_{\rho_L}$ started at equilibrium with positive magnetization becomes essentially {\it unpredictable} i.e. it can be thought of as the (random) number of independent attempts, each of which has a probability of success of the order of $\exp (-\beta L\psi (\rho ))$, that one has to make before seeing a success. \par For results with the magnetization density $\rho$ outside the region $(-m^*(\beta ),+m^*(\beta ))$ we refer the reader to the paper by Lebowitz and Schonmann [LSch]).\bigskip {\bf Proof of Theorem 5.1}\medskip We start by proving a lower bound of the right order when we start from the measure $\mu_V^\emptyset$ restricted to the configurations of positive magnetization.\par Clearly for such class of configurations: $$\t_{\rho_L}\,\geq\,\t_{\rho_L\vee 0}$$ so that it is enough to prove a correct lower bound only for $\rho\,\in\,[0,m^*(\beta ))$.\par For any positive $T$ we can write: $$\sum_{\s \atop m(\s )>0}\mu_V^\emptyset (\s )E_\s(\t_{\rho_L})\;\geq\; T\sum_{\s \atop m(\s )>0}\mu_V^\emptyset (\s )(1\,-\,P_\s(\t_{\rho_L}\,\leq\,T))\,\geq\,$$ $${T\over 2}\,-\, T\sum_{\s \atop m(\s )>0}\mu_V^\emptyset (\s )P_\s(\t_{\rho_L}\,\leq\,T)\Eq(5.2)$$ where we used the symmetry of the Gibbs measure under global spin flip.\par As in section 4 (see (4.4) and (4.5)) the sum in the r.h.s. of \equ(5.2) can be estimated from above by: $$2L^2T\mu_V^\emptyset(m(\s )\,=\,\rho_L)\,+\,\exp (-KL^2T)\Eq(5.3)$$ for a suitable constant $K$.\par We now take the time $T$ of the form $$T\,=\,\exp (\beta L(\psi (\rho )-\d))\Eq(5.4)$$ where $\d$ is any fixed small number independent of $L$. If we recall theorem 4.2, we get that, with this choice of $T$ \equ(5.3) goes to zero as $L$ gets large. This fact together with \equ(5.2) implies that $$\lim_{L\to \infty}{1\over \beta L}\log (\sum_{\s \atop m(\s )>0}\mu_V^\emptyset (\s )E_\s(\t_{\rho_L}))\;\geq\;\psi (\rho )-\d\quad \forall \;\rho\in \,[0,m^*_\beta )\Eq(5.5)$$ Since $\d$ can be taken arbitrarily small (after the limit $L\to \infty$) the required lower bound follows. It is also clear that the same lower bound applies also to $E_+(\t_{\rho_L})$, that is when the starting configuration is identically equal to plus one, since, because of monotonicity in the initial configuration, $$E_+(\t_{\rho_L})\,\geq\,\sum_{\s \atop m(\s )>0}\mu_V^\emptyset (\s )E_\s(\t_{\rho_L})$$ In order to prove an upper bound we have to distinguish between two cases: $$\eqalign{\hbox{Case 1}\qquad\rho\;&\in\;(-m^*(\beta ),\rho_1]\cr\hbox{Case 2}\qquad\rho\;&\in\;(\rho_1,m^*(\beta ))}$$ where $\rho_1$ is the singularity point of the function $\psi (\rho )$ defined in theorem 4.2 .\par Let us begin with the first one.\par Clearly $$\sum_{\s \atop m(\s )>0}\mu_V^\emptyset (\s )E_\s(\t_{\rho_L})\,=\,\sum_{\s \atop m(\s )>0}\mu_V^\emptyset (\s )\sum_nP_\s(\t_{\rho_L}\,\geq\,n)\,\leq\,\sum_nP_+(\t_{\rho_L}\,\geq\,n)\Eq(5.6)$$ since, by monotonicity in the initial configuration, $$P_\s(\t_{\rho_L}\,\geq\,n)\,\leq\,P_+(\t_{\rho_L}\,\geq\,n)\quad \forall\;n\quad \forall \;\s\Eq(5.6bis)$$ In turn, for any integer $N$, it follows from the Markov property and \equ(5.6bis) that: $$P_+(\t_{\rho_L}\,\geq\,n)\,\leq\,P_+(\t_{\rho_L}\,\geq\,N)^{[{n\over N}]}\Eq(5.7)$$ Let us therefore estimate $P_+(\t_{\rho_L}\,\geq\,N)$. We write $$P_+(\t_{\rho_L}\,\geq\,N)\,=\,P_+(\int_0^Ndt \chi (m(\s_t)\,\geq\,\rho_L)\,=\,N)\,\leq\,{\int_0^Ndt P_+(m(\s_t)\,\geq\,\rho_L)\over N}\Eq(5.8)$$ where $\chi (A)$ denotes the characteristic function of the event $A$ and we used the generalized Chebyshev inequality in order to get the last inequality.\par Let us now choose the integers $N$, $N_o$ equal to $$\eqalign{N\;&=\;\exp (\beta L(\psi (0)\,+\,2\d ))\cr N_o\;&=\; \exp (\beta L(\psi (0)\,+\,\d ))\qquad \d\,<<\,1}$$ Then we can estimate the integral in the r.h.s. of \equ(5.8) by $${\int_0^Ndt P_+(m(\s_t)\,\geq\,\rho_L)\over N}\,\leq\,$$ $${N_o\over N}\;+\;\mu_V^\emptyset (m(\s )\,\geq\,\rho_L )\;+\; {\int_{N_o}^Ndt [P_+(m(\s_t)\,\geq\,\rho_L) \,-\,\mu_V^\emptyset (m(\s )\,\geq\,\rho_L )]\over N}\Eq(5.9)$$ Let us examine separately each one of the three terms in the r.h.s. of \equ(5.9) in the limit as $L\to \infty$. The first term goes to zero by construction. The second term converges to $1\over 2$ because of theorem 4.2 . The third term also goes to zero for $\beta$ large enough, if we use theorem 4.1, the basic estimate (1.19), the fact that $\psi (0)\,=\,\t_\beta$ and our choice of the integer $N_o$.\par In conclusion we have shown that, for all $\beta$ large enough and all large enough $L$ $$P_+(\t_{\rho_L}\,\geq\,N)\,\leq\, {2\over 3}\Eq(5.10)$$ Clearly \equ(5.10) together with \equ(5.6), \equ(5.7) prove that $$\sum_{\s \atop m(\s )>0}\mu_V^\emptyset (\s )E_\s(\t_{\rho_L})\,\leq\,E_+(\t_{\rho_L})\,\leq\,3N\;=\;3\exp (\beta L(\psi (0)\,+\,2\d ))\Eq(5.11)$$ which establishes the correct upper bound in the limit $L\to\infty$ due to the arbitrariness of $\d$ also for the case when the starting configuration is identically equal to plus one.\par Let us now treat the (more difficult) second case $\rho\;\in\;[\rho_1,m^*(\beta ))$.\par First of all we bound $$\sum_{\s \atop m(\s )>0}\mu_V^\emptyset (\s )E_\s(\t_{\rho_L})$$ by the same quantity but computed for the HB-dynamics in $V$ with extra $+$ boundary conditions on the top horizontal side and starting from all pluses: $$\sum_{\s \atop m(\s )>0}\mu_V^\emptyset (\s )E_\s(\t_{\rho_L})\,\leq\, E_+^{\emptyset,\emptyset,+,\emptyset}(\t_{\rho_L})\Eq(5.12)$$ with self explanatory notation. The reason for introducing on only one side of $V$ extra $+$ boundary conditions is the following. For large L, the relaxation time to equilibrium ($\equiv \hbox{gap}(HB,\emptyset,\emptyset,+,\emptyset )^{-1}$) with the indicated boundary conditions is of the order of $\exp (C\beta L^{{1\over 2}+\e})$ (see Corollary 4.1); therefore the relaxation time is much smaller than the inverse of the equilibrium measure of the hitting set $\{\s;\; m(\s )\leq \rho_L\}$, $$\mu_V^{\emptyset,\emptyset,+,\emptyset}(m(\s )\leq \rho_L)^{-1}\,\geq\, \exp (\beta c(\rho ) L)\; ;\quad c(\rho )>0$$ It thus follows from a standard argument (see e.g. [A]) that: $$E_+^{\emptyset,\emptyset,+,\emptyset}(\t_{\rho_L})\,\leq\, \mu_V^{\emptyset,\emptyset,+,\emptyset}(m(\s )\leq \rho_L)^{-1}\exp (\beta \d L)\qquad \d<<1\Eq(5.13)$$ Thus one needs, in strict analogy with theorem 4.2, to estmate from below $$\mu_V^{\emptyset,\emptyset,+,\emptyset}(m(\s )\leq \rho_L)$$ as $L\to \,\infty$. This is the content of the next proposition:\bigskip {\bf Proposition 5.1}\par {\it Let $\rho_L$ be as above. Then there exists $\beta_o$ such that for any $\beta\,\geq \,\beta_o$ and any given positive $\d$ $$\mu_V^{\emptyset,\emptyset,+,\emptyset}(m(\s )\leq \rho_L)\,\geq\, \exp(-\beta (\psi (\rho)+\d) L)$$ for all $L$ large enough, where $\psi (\rho )$ is as in theorem 4.2 .}\bigskip The proposition can be proved by exactly the same methods developed in [DKS] (see also [PF]) and employed by Shlosman in [Sh] in his proof of Theorem 4.2; the proof is however lengthy and therefore it is not included in this work.\par It is possible to give a convincing explanation why the extra $+$ boundary conditions on the top side of $V$ do not affect the aymptotics (or at least a lower bound) of $$\mu_V^{\emptyset,\emptyset,+,\emptyset}(m(\s )\leq \rho_L)$$ In [Sh] (see [DKS] for full details in the case of periodic boundary conditions and [Pf] in the case of plus boundary conditions) the asymptotics of $\mu_V^{\emptyset,\emptyset,\emptyset ,\emptyset}(m(\s )= \rho_L)$ is derived by proving that the typical structure of the set of configurations under the event $\{m(\s )= \rho_L\}$ is as follows:\medskip \item{1)} In case $\rho \,>\,\rho_1$ there is a bubble $W^{o}_\rho$ of the minus phase close to one of the four corners of $V$, while in $V\setminus W^{o}_\rho$ one has the plus phase. The shape of $W^{o}_\rho$ is that of the intersection with $V$ of the rescaled Wulff shape $2\sqrt{{m^*_\beta -\rho\over 2m^*_\beta}}W$ of total volume $4{m^*_\beta -\rho\over 2m^*_\beta}$ and centered at one of the corners of $V$. It is clear from the results in [DKS] (see ch. 5) that the probability (with open boundary conditions) for the above situation to occur is of the order $$\exp (-\beta L{1\over 2} \sqrt{{m^*_\beta -\rho\over 2m^*_\beta}}w)\;=\; \exp(-\beta \psi (\rho ) L)$$ where $w$ is the Wulff functional computed on the Wulff curve $\partial W$. \item{2)} In case $\rho\,\leq\, \rho_1$, where $\rho_1$ is as in theorem 4.2, it is more convenient to divide the volume $V$ into roughly two rectangles, with the correct volumes determined by $\rho$, by means of a (roughly) straight horizontal line. It is clear that in this other case the probability is of the order of $$\exp (-\beta \t_\beta L)$$ for any $\rho\,\leq\,\rho_1$.\medskip In the first case, we can impose to our configuration to have a unique "large" contour exactly like the one described above, at distance greater than $cL$ from the top side of $V$, where $c>{1\over 2}$ is a suitable constant depending on $\rho$. Since the coefficients $\Phi^{\emptyset,\emptyset,+,\emptyset}(\L )$ of the cluster expansion of the partition function decay exponentially fast in the "size" of the set $\L$, it is not difficult to see that in this way one obtains a lower bound on \hfill $\mu_V^{\emptyset,\emptyset,+,\emptyset}(m(\s )\leq \rho_L)$ which, apart from minor corrections that are adsorbed in the $\d$ appearing in the proposition, is like the one obtained without the extra plus boundary condition on the top side.\bigskip It is clear that if we plug the statement of the proposition into \equ(5.13) we get the required upper bound. The proof of the theorem is complete.\bigskip {\bf Proof of Theorem 5.2}\par Let us define the numbers $a_L$ by the following condition: $$\sum_{\s ;\,\atop m(\s )>0}\mu_V^{\emptyset}(\s )P_\s(\t_{\rho_L}\,\geq\,a_L)\;=\;\hbox{e}^{-1}\Eq(5.14)$$ and let $f_L(t)$ be given by: $$f_L(t)\,=\,\sum_{\s ;\,\atop m(\s )>0}\mu_V^{\emptyset}(\s )P_\s(\t_{\rho_L}\,>\,a_Lt)\Eq(5.15)$$ In order to prove the theorem it is enough, using the normalization \equ(5.14), to show that: $$\lim_{L\to \infty}\vert f_L(t+s)\,-\,f_L(t)f_L(s)\vert\;=0\Eq(5.16)$$ and that the asymptotics of the number $a_L$, as $L\to \infty$, is the right one. Because of \equ(5.2) applied to $T\,=\,a_L$, we immediately get that: $$a_L\;\leq\;e\sum_{\s ;\,\atop m(\s )>0}\mu_V^{\emptyset}(\s )E_\s(\t_{\rho_L})\Eq(5.17)$$ In order to obtain a lower bound on $a_L$ we observe that, using the argument employed in section 4 (see e.g. (4.4), (4.5)): $$1-e^{-1}\,=\,\sum_{\s\atop m(\s )>0}\mu_V^{\emptyset}(\s )P_\s(\t_{\rho_L}\,<\,a_L)\,\leq$$ $$2L^2(a_L\vee 1)\mu_V^\emptyset(m(\s )\,=\,(\rho_L\vee 0))\,+\,\exp (-KL^2(a_L\vee 1))\Eq(5.18)$$ for a suitable constant $K$. Thus $$a_L\,\geq\,{1-e^{-1}\over 4L^2\mu_V^\emptyset(m(\s )\,=\,(\rho_L\vee 0))}\Eq(5.19)$$ for large $L$. Clearly \equ(5.17) and \equ(5.19) together with theorems 5.1, 4.2 prove the first part of the theorem.\par Let us turn to the proof of \equ(5.16). We observe that, because of the definition of the stopping time $\t_{\rho_L}$, it trivially follows that: $$f_L(t)\,=\,\sum_{\s }\mu_V^\emptyset(\s )P_\s(\t_{\rho_L}>a_Lt) \,-\,\sum_{\s \atop \rho_L\leq m(\s )\leq 0}\mu_V^\emptyset(\s )P_\s(\t_{\rho_L}>a_Lt)\,\equiv \,\bar f_L(t)\,-\,\e_L$$ Clearly,using theorem 4.2 $\e_L$ goes to zero exponentially fast in $L$.\par Using the reversibility of the dynamics with respect to the Gibbs measure $\mu_V^\emptyset$, we can write $\bar f_L(t+s)$ as: $$\bar f_L(t+s)\,=\,\sum_{\s }\mu_V^\emptyset(\s ) P_\s(\t_{\rho_L}>a_Lt) P_\s(\t_{\rho_L}>a_Ls)\,=\,$$ $$ \sum_{\s ;\atop m(\s )>0 }\mu_V^\emptyset(\s ) P_\s(\t_{\rho_L}>a_Lt) P_\s(\t_{\rho_L}>a_Ls)\,+\,\e_L$$ so that the difference $ \vert f_L(t+s)\,-\, f_L(t) f_L(s)\vert$ can be estimated from above by: $$\vert \sum_{\s , \h \atop\,m(\s )>0,m(\h )>0}\mu_V^\emptyset(\s )\mu_V^\emptyset(\h ) P_\s(\t_{\rho_L}>a_Lt)[ P_\s(\t_{\rho_L}>a_Ls)\,-\, P_\h(\t_{\rho_L}>a_Ls)]\vert \,+\,2\e_L\Eq(5.20)$$ If we now couple the HB-dynamics starting from $\s$ and $\h$ together in the way described in section 1, we can estimate the first term in \equ(5.20) by $$\sum_{\s , \h \atop\,m(\s )>0,m(\h )>0}\mu_V^\emptyset(\s )\mu_V^\emptyset(\h ) P(\t_{\rho_L}(\s )\neq \t_{\rho_L}(\h ))\Eq(5.21)$$ where, with an abuse of notation, $P$ denotes the probability measure of the coupled process, $\t_{\rho_L}(\s )$ and $\t_{\rho_L}(\h )$ the stopping times starting from $\s$ and $\h$ respectively.\par The idea behind the estimate of $P(\t_{\rho_L}(\s )\neq \t_{\rho_L}(\h ))$ (see below) is at this point very natural: when the two starting configurations $\s$ and $\h$ are both chosen at random with respect to the Gibbs measure $\mu_V^\emptyset$ restricted to the "phase" $\{m\,>\,0\}$, then in a time scale $T_o$, which is much shorter than the typical time scale of $\t_{\rho_L}(\s )$ and $\t_{\rho_L}(\h )$, the two configurations become identical with very large probability. The reason for this quick loss of memory inside the "phase" $\{m\,>\,0\}$, in contrast to the smallness of the gap (see Theorem 4.1), has to be found in the fact (see the proof of proposition 5.2 below) that, starting in equilibrium with positive magnetization, with large probability the HB-dynamics in $V$ with open boundary conditions cannot be distinguished, at a given site $x\in V$, from the HB-dynamics in $V$ with an extra $+$ boundary condition on one of the sides of $V$. This latter looses memory of the initial condition much faster than the dynamics with open boundary conditions (see corollary 4.1) and the result follows.\par Let us start with the technicalities. Let $\e\in (0,{1\over 2})$ be given and let $T_o\,=\,\exp (\beta L^{{1\over 2}+\e })$. Then we estimate \equ(5.21) by: $$2\sum_{\s , \atop m(\s )>0}\mu_V^\emptyset(\s )P_\s(\t_{\rho_L}\,<\,T_o)\;+\; 2\sum_{\s , \atop m(\s )>0}\mu_V^\emptyset(\s )P(\s_{T_o}\neq (+)_{T_o})\Eq(5.22)$$ where $(+)_{T_o}$ is the evoluted at time $T_o$ of the configuration identically equal to $+1$.\par We know already (see \equ(5.18)) that the first term in \equ(5.22) goes to zero as $L\to \infty$ provided that $\beta$ is large enough. The second term is controled by the following new result:\bigskip {\bf Proposition 5.2}\par {\it Let $\e\,\in \, (0,{1\over 2})$ and $m\,>\,0$ be given. Then there exist $\beta_o\,<\,+\infty$ and $C\,<\,+\infty$ such that for any $\beta\,\geq\,\beta_o$, any integer $L$ and any time $t\;\geq\;\exp (C\beta L^{{1\over 2}+\e })$: $$\sum_{\s \atop m(\s )>0}\mu_V^\emptyset(\s )P((+)_t\neq\s_t)\;\leq\;\exp (-mL)$$ }\bigskip It is clear that the above proposition concludes the proof of Theorem 5.2 in the case the starting configuration is distributed according to the restriction of the Gibbs measure to the set $\{\s;\;m(\s )>0\}$. A similar argument can be repeated if the starting configuration is identically equal to $+1$.\bigskip {\bf Proof of proposition 5.2}\par Since $$P((+)_{t+s}\neq\s_{t+s})\,\leq\,P((+)_t\neq\s_t)\qquad \forall \;s\,\geq \,0$$ it is sufficient to prove the result for the fixed time $t_o\,=\,\exp (C\beta L^{{1\over 2}+\e })$. We first estimate $P((+)_{t_o}\neq\s_{t_o})$ by $$P((+)_{t_o}\neq\s_{t_o})\;\leq\;\sum_{x\in V}P((+)_{t_o}(x)\neq\s_{t_o}(x))$$ Given now $x\in V$ let us uppose, without loss of generality, that the top horizontal side of $V$ is such that its distance from $x$ is greater or equal than $L\over 2$. Let also $(+)_{t_o}^{\emptyset ,\emptyset ,+,\emptyset }$ be the evoluted at time ${t_o}$ of the configuration $(+)$ under the HB-dynamics in $V$ with $(\emptyset ,\emptyset ,+,\emptyset )$ boundary conditions on $\partial_{ext} V$. Then, by monotonicity, we have: $$P((+)_{t_o}(x)\neq\s_{t_o}(x))\;\leq\;P((+)_{t_o}^{\emptyset ,\emptyset ,+,\emptyset }(x)\neq\s_{t_o}(x))\;=\;$$ $$P((+)_{t_o}^{\emptyset ,\emptyset ,+,\emptyset}(x)\,=\,+1)\,-\, P(\s_{t_o} (x)=+1)\Eq(5.23)$$ Thus $$\eqalign{&\phantom{\sum_{x\in V}}\sum_{\s \atop m(\s )>0}\mu_V^\emptyset(\s )P((+)_{t_o}\neq\s_{t_o})\;\leq\;\cr\sum_{x\in V}&[\sum_{\s \atop m(\s )>0}\mu_V^\emptyset(\s )(P((+)_{t_o}^{\emptyset ,\emptyset ,+,\emptyset}(x)\,=\,+1)\,-\, P(\s_{t_o} (x)=+1))]\,=\,\cr\sum_{x\in V}&[{1\over 2}P((+)_{t_o}^{\emptyset ,\emptyset ,+,\emptyset}(x)=+1)\;-\;\sum_{\s \atop m(\s )>0}\mu_V^\emptyset(\s )P(\s_{t_o} (x)=+1)\,]}\Eq(5.23bis)$$ Let us first treat the term $P((+)_{t_o}^{\emptyset ,\emptyset ,+,\emptyset}(x)=+1)$. Using (1.19), Corollary 4.1 and our choice of the time ${t_o}$, we get that: $$0\,\leq\,P((+)_{t_o}^{\emptyset ,\emptyset ,+,\emptyset}(x)\,=\,+1)\,-\, \mu_V^{\emptyset ,\emptyset ,+,\emptyset}(\s (x)=+1)\,\leq\,$$ $$\exp (C'L^2\,-\,{t_o}\exp (-C\beta L^{{1+\e\over 2}}))\,\leq\,{1\over 3}\exp (-mL)\Eq(5.24)$$ for any given $m>0$ and any $L\in {\bf N}$, provided that $\beta$ is large enough.\par As far as the second term in the square parenthesis in the r.h.s. of \equ(5.23bis) is concerned, we write: $$\sum_{\s \atop m(\s )>0}\mu_V^\emptyset(\s )P(\s_{t_o} (x)=+1)\;\geq\;\sum_{\s \atop m(\s )>0}\mu_V^\emptyset(\s )P(\s_{t_o} (x)=+1\cap m(\s_{t_o} )>0)\;=\;$$ $$\sum_{\s}\mu_V^\emptyset(\s )P(\s_{t_o} (x)=+1\cap m(\s_{t_o} )>0)\;-\;\sum_{\s ;\;m(\s )\leq 0}\mu_V^\emptyset(\s )P(\s_{t_o} (x)=+1\cap m(\s_{t_o} )>0)\;=\;$$ $$\mu_V^\emptyset(\s (x)=1\cap m(\s )\geq 0)\;-\; \sum_{\s ;\;m(\s )< 0}\mu_V^\emptyset(\s )P(\s_{t_o} (x)=+1\cap m(\s_{t_o} )\geq 0)\Eq(5.25)$$ where we used the invariance of the measure $\mu_V^\emptyset$.\par The last term in the r.h.s. of \equ(5.25) can be bounded from above by: $$\sum_{\s }\mu_V^\emptyset(\s )P(\hbox{there exists }s\leq {t_o};\; m(\s_s )=0)\,\leq\,2L^2{t_o}\mu_V^\emptyset(m(\s )\,=\,0)\,+\,\exp (-KL^2{t_o})\Eq(5.26)$$ for a suitable constant $K$ and large enough $\beta$, by the argument illustrated in section 4 (see (4.5)).\par Clearly, because of our choice of ${t_o}$ and of theorem 4.2, the r.h.s. of \equ(5.26) is smaller than ${1\over 3}\exp (-mL)$ for any given $m$ provided $\beta$ is large enough.\par In conclusion we have shown that: $${1\over 2}P((+)_{t_o}^{\emptyset ,\emptyset ,+,\emptyset}(x)\;-\;\sum_{\s \atop m(\s )>0}\mu_V^\emptyset(\s )P(\s_{t_o} (x)=+1)\,\leq\,$$ $${1\over 2}\vert\mu_V^{\emptyset ,\emptyset ,+,\emptyset}(\s (x)=+1)\,-\,\mu_V^\emptyset(\s (x)=1\vert m(\s )\geq 0)\vert \;+\;{2\over 3} \exp (-mL)\Eq(5.27)$$ for any $L$, provided that $\beta$ is large enough.\par In order to complete the proof we need a last, rather obvious result on the Ising model, whose proof is an exercise in the cluster expansion and it is therefore omitted.\bigskip {\bf Lemma 5.1}\par {\it Given $m>0$ there exists $\beta_o$ such that for all $\beta \geq \beta_o$ and all $L$: $$\vert\mu_V^{\emptyset ,\emptyset ,+,\emptyset}(\s (x)=+1)\,-\,\mu_V^\emptyset(\s (x)=1\vert m(\s )\geq 0)\vert\;\leq {1\over 3}\exp (-mL)$$}\bigskip If we apply the lemma to \equ(5.27) we obtain: $$\sum_{x\in V}\sum_{\s \atop m(\s )>0}\mu_V^\emptyset(\s )P((+)_t\neq\s_t)\;\leq\;L^2\exp (-mL)$$ for any given $m>0$ and any $L\in {\bf N}$, provided that $\beta$ is large enough. The proposition is proved. \pagina %\input formato.tex \def\muo{\mu_{V_L}^\emptyset} \def\Pr{ P^{\mu}} \def \rh#1{\rho (\tilde\s_{#1})} \tolerance=10000 \numsec=6 \numfor=1 {\bf Section 6}\par \centerline{\bf Markov Chain Description of the Time Rescaled Magnetization}\bigskip In this final section we work in the same setting and notation of the previous two sections and we consider the normalized magnetization $$\rho (\s_t)\,=\,{1\over \vert V_L\vert}\sum_{x\in V_L}\s_t (x)$$ of the process started at equilibrium (or from one of the two extreme configurations $(+)$ or $(-)$).\par We show that it is possible to rescale the time $t$ by a multiplicative factor $t_L$ depending on the side $L$ of the square $V_L$, in such a way that, as $L\to \infty$, the finite dimensional distributions of the rescaled process $$ \rho (\tilde \s_{t})\,=\,\rho (\s_{t_Lt})\Eq(6.1.0)$$ converge to those of a continuous time Markov chain on the set $\{-m^*(\b ),m^*(\b )\}$ with unitary jump rate for both states.\par >From what we just said, it is clear that the speeding factor $t_L$ must be determined essentially by the condition that: $$\int_{m(\s )>0}d\muo (\s )P(\rho (\s_{t})\,\approx \,-m^*(\b ))\;\approx\; {p\over 2}$$ where $$p\;=\;{1\over 2}(1-\exp (-2))\Eq(6.1)$$ is the probability that a continuous time Markov chain with unitary jump rate on $\{-1,1\}$ starting at time $t=0$ in $+1$ is, at time $t=1$, in the state $-1$.\par It is also clear from the results of section 4 and section 5 that $$t_L\;\approx\;\exp (\b \t_\b L)$$ for large $L$.\par Let us state more precisely our result. We denote by $M$ the two state space $\{-m^*(\b ),m^*(\b )\}$ and by $Y_t$ a continuous time Markov chain on $M$ with unitary jump rate for both states. Clearly the invariant measure $\nu$ of the chain $Y_t$ is uniform over $M$. Let also, for any given $\e\,\in\,(0,{m^*(\b )\over 2})$, $t_L\,\equiv \,t_L(\e )$ be the such that: $$\Pr (\rh{0}\,\geq\, m^*(\b )-\e\,;\,\rh{1}\,\leq\, -m^*(\b )+\e)\;=\;{p\over 2}\Eq(6.2)$$ where $p$ is given by \equ(6.1) and $\Pr$ denotes the probability over the HB-dynamics started from the equilibrium distribution $\muo$. Then we have :\bigskip {\bf Theorem 6.1}\par {For any $\b$ large enough, any $\e$ as above and for any choice of times $t_1\,<\,t_2\,<\, \dots\,<\,t_k$ and numbers $m_i\,\in \, M$, $i=1\dots k$ : $$\lim_{L\to \infty}\Pr (\vert \rh{t_1}-m_1\vert <\e,\dots ,\vert \rh{t_k}-m_k\vert <\e )\;=\;P^{\nu }(Y_{t_1}=m_1,\dots ,Y(t_k)=m_k)\Eq(6.3)$$ where $P^{\nu }$ denotes the probability of the chain $Y_t$ with initial distribution the invariant measure $\nu$. Moreover $$\lim_{L\to \infty}{1\over \b L}\log (t_L)\,=\,\t_\b$$}\bigskip {\bf Proof}\par The second part follows immediately from the results of section 4 and section 5.\par As far as the first part is concerned, it is well known that $$\lim_{L\to \infty}\muo (\vert\rho (\s )\,-\,m^*(\b )\,\vert\,\leq \, \d)\;=\;{1\over 2}\quad \forall \,\d>0\Eq(6.2bis)$$ and similarly, by the symmetry under global spin flip, for $m^*(\b )$ replaced by $-m*(\b )$. Hence, if the limit in the l.h.s. of \equ(6.3) exists for fixed $t_1\,<\,t_2\,<\, \dots\,<\,t_k$ and arbitrary choice of $m_i\,\in \, M$, $i=1\dots k$, it must be a probability measure on $M^k$.\par We will prove \equ(6.3) by showing that the limit along any convergent subsequence is equal to the r.h.s. of \equ(6.3). The key step in our argument is to prove that, asympotically as $L\to \infty$, the variables $\rh{t_1},\dots ,\rh{t_k}$ enjoy the Markov property. This is the content of the following proposition. For notation convenience we denote by $A_{t_i}(m_i)$ the event $\vert \rh{t_i}-m_i\vert <\e$.\bigskip {\bf Lemma 6.1}\par {\it In the same hypotheses of theorem 6.1 the difference $$\Pr (A_{t_1}(m_1),.. ,A_{t_k}(m_k))-\Pr (A_{t_1}(m_1),.. ,A_{t_{k-1}}(m_{k-1})) 2\Pr (A_{t_{k-1}}(m_{k-1}),A_{t_k}(m_k))$$ tends to zero as $L\to \infty$.}\bigskip Before giving the proof of the above key result we complete the proof of theorem 6.1\par Using the lemma and \equ(6.2bis) it is clearly enough to prove that: $$\lim_{L\to \infty}\Pr (\vert \rh{0}+m^*(\b )\vert \leq \e;\vert\rh{t}-m^*(\b )\vert \leq \e)\,=$$ $$=\, P^{\nu }(Y_{0}=-m^*(\b );Y_t=m^*(\b ))\,=\,(1-(1-2p)^t)\Eq(6.4)$$ Let us first consider times $t$ of the form $t\,=\,{1\over m},\;m\in {\bf N}$ and let us define by $a({1\over m})$ any limit of the l.h.s. of \equ(6.4) computed for such $t$. From the lemma applied to times $t_i\,=\,{i\over m}\;i=1\dots m$ and the fact that, by construction $$a(1)\;=\;{p\over 2}$$ one immediately gets $$a({1\over m})\,=\,(1-(1-2p)^{1\over m})\Eq(6.5)$$ Once we know the value of $a({1\over m})$ we can repeat the same argument to show that \equ(6.4) holds also for rational times of the form $t\,=\,{n\over m}$. In order to extend \equ(6.4) to all times $t$, it is sufficient to prove for example that, if $\bar a(t)$ and $\underline a(t)$ denote the $\limsup$ and $\liminf$ of the l.h.s. of \equ(6.4), then both of them are non decreasing function of $t$.\par For this purpose and denoting by $a_L(t)$ the l.h.s. of \equ(6.4), we immediately obtain from the lemma and \equ(6.2bis) that $a_L(t+s)$ satisfies the equation: $$a_L(t+s)\,=\,a_L(t)\, +\,a_L(s)(1-4a_L(t))\,+\,r_L\Eq(6.6)$$ where $\lim_{L\to\infty}r_L\;=\;0$.\par We now observe that $$a_L(t)\;\leq\;{1\over 4}\;+\;r'_L\Eq(6.7)$$ where, as before, $\lim_{L\to\infty}r_L'\;=\;0$.\par In fact, again because of \equ(6.2bis) $$\Pr (\vert \rh{t}-m^*(\b )\vert \leq \e;\,\vert \rh{0}+m^*(\b )\vert \leq \e)\,=\,{1\over 2}\,-\,\Pr(\rh{t}>0;\,\rh{0}>0)\;+\;r'_L$$ and, by the F.K.G. property of the measure $\Pr$ (see e.g. [Li]): $$\Pr(\rh{t}>0;\,\rh{0}>0)\,\geq\,\Pr(\rh{t}>0)\Pr(\rh{0}>0)\,=\,{1\over 4}$$ If we insert \equ(6.7) we obtain: $$a_L(t+s)\,\geq\,a_L(t)\,+\,r_L\,-\,4r'_L\Eq(6.8)$$ Clearly \equ(6.8) shows that $\bar a(t)$ and $\underline a(t)$ are non decreasing function of $t$ and thus \equ(6.4) holds for all $t$.\bigskip {\bf Proof of Lemma 6.1}\par Using the reversibility we can write: $$\Pr (A_{t_1}(m_1) \,\dots\,A_{t_k}(m_k))\,=$$ $$=\;\int_{A_0(m_{k-1})} d\muo (\s)P_\s (A_{\vert t_k-t_{k-1}\vert}(m_k))P_\s(A_{\vert t_{k-1}-t_1\vert}(m_1),\dots ,A_{\vert t_{k-2}-t_{k-1}\vert}(m_{k-2}))\Eq(6.9)$$ where $P_\s$ denotes the probability measure on the HB-dynamics starting from $\s$.\par We now compare the r.h.s. of \equ(6.9) with the quantity: $$(\muo (A_0(m_{k-1}))^{-1}\Pr (A_{t_1}(m_1),.. ,A_{t_{k-1}}(m_{k-1})) \Pr (A_{t_{k-1}}(m_{k-1}),A_{t_k}(m_k)) \Eq(6.10)$$ Using the stationarity of the measure $\Pr$ and reversibility, we can write their difference as: $$(\muo (A_0(m_{k-1}))^{-1}\int\int_{A_o(m_{k-1})\atop A_o(m_{k-1})} d\muo (\s )d\muo (\h )G(\h ,\s ) \Eq(6.11)$$ where $$\eqalign{G(\h ,\s )&=\cr [P_\h (A_{\vert t_k-t_{k-1}\vert}(m_k))-P_\s (A_{\vert t_k-t_{k-1}\vert}(m_k))]& P_\s(A_{\vert t_{k-1}-t_1\vert}(m_1),\dots ,A_{\vert t_{k-2}-t_{k-1}\vert}(m_{k-2}))}\Eq(6.12)$$ Using the coupling described in section 1 and the symmetry under global spin flip, the absolute value of \equ(6.12) can be estimated from above by: $$2\int\int_{m(\s )>0\atop m(\h )>0} d\muo (\s )d\muo (\h )P(\tilde\s_{t_k-t_{k-1}}\neq \tilde\h_{t_k-t_{k-1}})\Eq(6.12bis)$$ which tends to zero as $L\to \infty$ because of proposition 5.2 . We have in fact that $$t_L[t_k-t_{k-1}]\,>>\,\exp (C\b L^{{1\over 2}+\e})$$ because of the second part of theorem 6.1.\par The statement of the lemma now follows from \equ(6.2bis) since, in \equ(6.10) we can safely replace the factor $(\muo (A_0(m_{k-1}))^{-1}$ with $2$. The proof is complete.\pagina %\input formato.tex \numsec=1 \numfor=1 \centerline{\bf Appendix 1}\bigskip In this appendix we prove propositions 4.1, 4.2, 4.3. Since the proof of proposition 3.1 is very similar, although much simpler, than that of proposition 4.1, we decided, for shortness, to omit it.\medskip\noindent {\bf Proof of Proposition 4.1}\medskip Let us fix $\e\,\in \, (0,{1\over 2})$ and a rectangle $R$: $$R\,=\,\{\;x\in \Z;\,0\,\leq \,x_1\,\leq\,L_1\quad 0\,\leq\,x_2\,\leq\,L_2\;\}$$ with $L_1\,\geq \,L_2\,\geq\,L_1^{{1\over 2}+\e}$.\par If ${\cal A}_R^{+,+,-,+}$ is the event described in (4.12): $$ {\cal A}_R^{+,+,-,+}\;=\;\{\s ;\quad \Gamma_{open}(\s )\,\subset\, \{\;x\in R;\;x_2\,\geq\,{13L_2\over 16}\,\}\;\}$$ we can write: $$\eqalign{\mu_R^{+,\delta +,-,\delta +}(\s (x)=1)\;&=\cr \mu_R^{+,\delta +,-,\delta +}(\s (x)=1\vert {\cal A}_R^{+,+,-,+})\mu_R^{+,\delta +,-,\delta +}({\cal A}_R)\,&+\,\mu_R^{+,\delta +,-,\delta +}(\s (x)=1\cap ({\cal A}_R^{+,+,-,+})^c)}$$ where $ ({\cal A}_R^{+,+,-,+})^c$ is just the complement event.\par Since $$ \mu_R^{+,\delta +,-,\delta +}(\s (x)=1\vert {\cal A}_R)\,\geq \, \mu_R^{+,\delta +,+,\delta +}(\s (x)=1)\Eqa(1)$$ (see 4.27), we obtain that the difference $$\mu_R^{+,\delta +,+,\delta +}(\s (x)=1)\;-\;\mu_R^{+,\delta +,-,\delta +}(\s (x)=1)$$ can be bounded from above by $$\mu_R^{+,\delta +,-,\delta +}(({\cal A}_R^{+,+,-,+})^c)\Eqa(2)$$ In order to estimate the above probability, we first observe that the event $({\cal A}_R^{+,+,-,+})^c$ is a decreasing event (in the sense that its characteristic function is a non increasing function of the configuration). Therefore, if ${\hat R}$ is the new rectangle $${\hat R}\,=\,\{\;x\in \Z;\,0\,\leq \,x_1\,\leq\,L_1,\quad -{L_2\over 16}\,\leq\,x_2\,\leq\,L_2\;\}$$ $\t$ is the configuration: $$\eqalign{\t (x)\,&=\,+1\quad \forall \,x\in \partial_{ext}{\hat R}\quad\hbox{with }x_2\,\leq\,L_2-{L_2\over 16}\cr \t (x)\,&=\,-1\quad \hbox{otherwise}}$$ and $$\eqalign{ U^{\partial {\hat R}}(x,y)\,&=\,\delta \qquad \forall \,(x,y)\in \partial {\hat R}\quad \hbox{with }x_1\,=\,0\;\hbox{or }L_1,\;\hbox{and }\,0\,\leq\,x_2\,\leq\,L_2-{L_2\over 16}\cr U^{\partial {\hat R}}(x,y)\,&=\,1\qquad \hbox{otherwise}}$$ then $$\mu_R^{+,\delta +,-,\delta +}(({\cal A}_R^{+,+,-,+})^c)\,\leq\, \mu_{{\hat R}}^{U^{\partial {\hat R}},\t}(({\cal A}_R^{+,+,-,+})^c)\Eqa(3)$$ If we denote by $\G^\t_{{\hat R},open}(\s )$ the (unique) open contour of $\s\in \O_{{\hat R}}$ under the $\t$ boundary conditions described above, it is immediate to check that: $$({\cal A}_R^{+,+,-,+})^c\, \subset \,({\cal A}^\t_{{\hat R}})^c\,\equiv\,\{\s ;\;\G^\t_{{\hat R},open}(\s )\cap \{x\in {\hat R};\;x_2<{13\over 16}L_2\}\neq \emptyset\}$$ so that $$\mu_{{\hat R}}^{U^{\partial {\hat R}},\t}(({\cal A}_R^{+,+,-,+})^c)\,\leq\, \mu_{{\hat R}}^{U^{\partial {\hat R}},\t}(({\cal A}^\t_{{\hat R}})^c))\Eqa(3bis)$$ For simplicity in the sequel we will denote the measure $ \mu_R^{U^{\partial {\hat R}},\t}$ by $P$.\par Let us now order the bonds in $\G_{{\hat R},open}^\t(\s )$ from left to right and let us denote by $e_{k_1}$, $e_{k_1+n}$, the smallest, respectively the largest, bond in $\G_{{\hat R},open}^\t(\s )$ such that no site in the portion of the exterior boundary of the left, respectively right, lateral side of ${\hat R}$ where the boundary coupling is $\d$ is separated by one of the bonds $e\in\G_{{\hat R},open}^\t(\s )$ with $e\,\geq\,e_{k_1}$, respectively $e\,\leq\,e_{k_1+n}$.\medskip We will denote by $\gamma\,=\,e_{k_1}\dots e_{k_1+n}$ the portion of the open contour $\G_{{\hat R},open}^\t(\s )$, $\s\in \O_{{\hat R}}$, between $e_{k_1}$ and $e_{k_1+n}$ and by ${\cal F}$ the set of them.\par Notice that, by construction, $\g$ is itself an open polygonal line and that the first, respectively the last, bond in $\g$ separates at least one site in the internal boundary of the left, respectively right vertical side of ${\hat R}$. Moreover, if we denote by $h_\g$ be the horizontal line in $\bf R^2$ containing the middle point of the first bond $e_1$ of $\g$, then $h_\g$ is at distance at least ${L_2\over 16}-1$ from the horizontal portion of the boundary of ${\hat R}$ . Let also $$d(\g )\;=\;\hbox{dist}(\g,h_\g )$$ We now define the event $\cal C$ as: $${\cal C}\;=\;\{\s\, ;\; d(\g )\;\leq\;{L_2\over 32}\}\Eqa(4)$$ Then we estimate \equ(2) by $$P(({\cal A}_{{\hat R}}^{\t})^c)\,\leq\, P({\cal C}^c)\,+\, P(({\cal A}_{{\hat R}}^{+,+,-,+})^c\cap {\cal C})\Eqa(5)$$ {\bf Lemma A.1}\par {\it Given $m\,>\,0$ there exists $\beta (\e,m)$ such that for all $\beta\,\geq \,\beta (\e,m)$ $$ P({\cal C}^c)\,\leq\, \exp (-mL_1^{2\e})$$} {\bf Proof}\par Given $\gamma$, the set ${\hat R}$ can be written as the disjoint union of three sets: $${\hat R}\,=\, \D\g\cup R^+_\g\cup R^-_\g$$ where $\D\g$ has been defined in section 1 and $R^+_\g$, $R^-_\g$ lay, in a natural way, below and above $\g$ respectively.\par Associated to the set $R^+_\g$ we consider the partition function $Z(R^+_\g,U^{\partial {\hat R}},\t )$ with $\t$ boundary condition and boundary coupling $U^{\partial {\hat R}}$ on $\partial_{ext}R^+_\g\cap \partial_{ext}{\hat R}$ and $+$ boundary condition on $\partial_{ext}R^+_\g\cap\D\g$; similarly for $Z(R^-_\g,U^{\partial {\hat R}},\t )$.\par We can now write: $$P({\cal C}^c)\,=\, {Z({\hat R},U^{\partial {\hat R}},+)\over Z({\hat R},U^{\partial {\hat R}},\t)}\sum_{\g ;d(\g )\geq {L_2\over 32}\atop \g \in {\cal F}}\exp (-2\beta \vert \g \vert ) {Z(R^+_\g,U^{\partial {\hat R}}, \t)Z(R^-_\g,U^{\partial {\hat R}},\t )\over Z({\hat R},U^{\partial {\hat R}},+)}\Eqa(5.a)$$ Unfortunately we cannot yet use the cluster expansion described in section 1 to simplify the above ratio of partition functions since, although in $Z(R^+_\g,U^{\partial {\hat R}},\t)$ the boundary condition is constantly equal to $+1$ because of a) in the definition of $\g$, so that $$Z(R^+_\g,U^{\partial {\hat R}},\t)\;=\;Z(R^+_\g,U^{\partial {\hat R}}, +)$$ in $Z(R^-_\g,U^{\partial {\hat R}},\t )$ the lateral boundary condition may change sign. However a trivial and rough comparison shows that: $$\exp (-4\beta \d L_2)\,\leq\,{Z(R^-_\g,U^{\partial {\hat R}},\t )\over Z(R^-_\g,U^{\partial {\hat R}},- )}\,\leq\,\exp (+4\beta \d L_2)\Eqa(5bis)$$ Therefore the r.h.s. of \equ(5.a) is bounded from above by: $$\exp (+8\beta \d L_2){\sum_{\g ;d(\g )\geq {L_2\over 32}\atop \g \in {\cal F}} \exp (-2\beta \vert \g \vert) {Z(R^+_\g,U^{\partial {\hat R}}, +)Z(R^-_\g,U^{\partial {\hat R}},- )\over Z({\hat R},U^{\partial {\hat R}},+)}\over \sum_{\g\subset {\hat R}\atop \g \in {\cal F}} {Z(R^+_\g,U^{\partial {\hat R}}, +)Z(R^-_\g,U^{\partial {\hat R}},- )\over Z({\hat R},U^{\partial {\hat R}},+)}}\Eqa(6)$$ We observe at this point that each one of the partition functions $$Z({\hat R},U^{\partial {\hat R}},+)\quad Z(R^+_\g,U^{\partial {\hat R}}, +)\quad Z(R^-_\g,U^{\partial {\hat R}},- )$$ can be written as in (1.10), with weights that satisfy the condition of proposition 1.1 with constant $\a\,=\,{1\over 2}$. Therefore, following [DKS], we can apply proposition 1.1 to write: $${Z(R^+_\g,U^{\partial {\hat R}}, +)Z(R^-_\g,U^{\partial {\hat R}},- )\over Z(R,U^{\partial {\hat R}},+)}\,=\,\exp (-\sum_{\vbox{\eightpoint{ \hbox{$\L\subset {\hat R}$}\hbox{$\L\cap \D\g\neq \emptyset$}}}}\Phi^{U^{\partial {\hat R}}, +}(\L ))\Eqa(7)$$ so that \equ(6)becomes: $$\exp (16\beta \d L_2) {\sum_{\g ;d(\g )\geq {L_2\over 32}\atop \g\in {\cal F}}\exp (-2\beta \vert \g\vert ) \exp (-\sum_{\L\subset {\hat R}\atop \L\cap \D\g\neq \emptyset}\Phi^{U^{\partial {\hat R}}, +}(\L ))\over \sum_{\g\subset {\hat R}\atop \g\in {\cal F}} \exp (-2\beta \vert \g\vert ) \exp (-\sum_{\L\subset {\hat R}\atop \L\cap \D\g\neq \emptyset}\Phi^{U^{\partial {\hat R}}, +}(\L )) }\Eqa(9)$$ We can use at this point two basic results in [DKS] (see the proposition and the theorem in section 4.14 and section 4.16 respectively) to conclude that, since $ L_1\,\geq\, L_2\,\geq\,L_1^{{1\over 2}+\e}$, for any given $m\,>\,0$, the ratio between the two sums in \equ(9) is smaller than $$\exp (-m{L_2^2\over L_1})\,=\,\exp (-mL_1^{2\e})$$ provided that $\beta$ is large enough.\par\noindent The lemma is proved.\bigskip We now turn to the estimate of the second term in the r.h.s. of \equ(5), $P((({\cal A}_{{\hat R}}^{\t})^c\cap {\cal C})$.\par Let $l\,=\,{L_2\over 16}$ (we are assuming for simplicity that $L_2\over 16$ is an integer) and let, for $i\,=\,0\dots N\,=\,32$, $R_i$ be the rectangle: $$R_i\,=\,\{\;x\in R;\,0\,\leq \,x_1\,\leq\,L_1,\quad -{L_2\over 16}+i{l\over 2}\,\leq\,x_2\,\leq\,-{L_2\over 16}+(i+2){l\over 2}\;\}$$ Then we define $P_i$ as: $$P_i\,=\,P(\{\g\,\subset\,R_i\}\cap {\cal C})$$ and we estimate from above $P((({\cal A}_{{\hat R}}^{\t})^c\cap {\cal C})$ by: $$P((({\cal A}_{{\hat R}}^{\t})^c\cap {\cal C})\,\leq\,\sum_{i=1\dots N-5}P(\{\g\subset R_i\}\cap {\cal C})\Eqa(9.1)$$ In \equ(9.1) we used the fact that, if the event $(({\cal A}_{{\hat R}}^{\t})^c\cap {\cal C}$ occurs, then, by construction, $\g$ is entirely contained in some $R_i$ because of ${\cal C}$, with the index $i\,\neq\, N,\,\dots N-4$ because of $(({\cal A}_{{\hat R}}^{\t})^c$ and $i\neq 0$ again because of $\cal C$.\par In order to estimate each term in the r.h.s. of \equ(9.1) we proceed in a slightly different way depending whether $i=1$ or $i\,>\,1$ the reason being that in the case $i=1$ the poygonal line $\g$ is very close to the discontinuity point of the lateral boundary coupling.\par Let us first consider the case $i\geq 2$. In this case we bound from above the ratio: ${P_i\over P_{N-2}}$ uniformly in $i\,=\,2\dots N-5$. If we use the representation \equ(5.a) for the probability of a given $\g$, we may write: $${P_i\over P_{N-2}}\,=\,{\sum_{\g\subset R_i;\, d(\g )\leq {L_2\over 32}}\exp (-2\beta \vert \g \vert ) Z(R^+_\g,U^{\partial {\hat R}}, \t)Z(R^-_\g,U^{\partial {\hat R}},\t ) \over \sum_{\g\subset R_{N-2};\, d(\g )\leq {L_2\over 32}}\exp (-2\beta \vert \g \vert ) Z(R^+_\g,U^{\partial {\hat R}}, \t)Z(R^-_\g,U^{\partial {\hat R}},\t )}\Eqa(10)$$ Given $\g\subset R_i$, let $F_i(\g )$ be its image under a vertical translation in $\bf R^2$ by an amount $(N-2-i){l\over 2}$.\par Then clearly $F_i$ establish a bijection between the $\g$ in $R_i$ and those in $R_{N-2}$, so that the r.h.s. is estimated from above by: $$\sup_{\g\subset R_i;\, d(\g )\leq {L_2\over 32}}{Z(R^+_\g,U^{\partial {\hat R}}, \t)Z(R^-_\g,U^{\partial {\hat R}}, \t)\over Z(R^+_{F_i(\g )},U^{\partial {\hat R}}, \t) Z(R^-_{F_i(\g )},U^{\partial {\hat R}}, \t)}\Eqa(11)$$ which we write as: $$\sup_{\g\subset R_i;\, d(\g )\leq {L_2\over 32}}{Z(R^+_\g,U^{\partial {\hat R}}, +)Z(R^-_\g,U^{\partial {\hat R}}, -)\over Z(R^+_{F_i(\g )},U^{\partial {\hat R}}, +) Z(R^-_{F_i(\g )},U^{\partial {\hat R}}, -)}\, {Z(R^-_\g,U^{\partial {\hat R}}, \t)Z(R^-_{F_i(\g )},U^{\partial {\hat R}}, -)\over Z(R^-_\g,U^{\partial {\hat R}}, -)Z(R^-_{F_i(\g )},U^{\partial {\hat R}}, \t)}\Eqa(12)$$ Let us consider the first ratio $${Z(R^+_\g,U^{\partial {\hat R}}, +)Z(R^-_\g,U^{\partial {\hat R}}, -)\over Z(R^+_{F_i(\g )},U^{\partial {\hat R}}, +) Z(R^-_{F_i(\g )},U^{\partial {\hat R}}, -)}\Eqa(13)$$ If we divide numerator and denominator by $Z({\hat R},U^{\partial {\hat R}},+)$ and we use \equ(7), we get that \equ(13) is equal to: $$\exp (-\sum_{\vbox{\eightpoint{ \hbox{$\L\subset {\hat R}$}\hbox{$\L\cap \D\g\neq \emptyset$}}}}\Phi^{U^{\partial {\hat R}}, +}(\L )\,+\,\sum_{\vbox{\eightpoint{ \hbox{$\L'\subset {\hat R}$}\hbox{$\L'\cap \D F_i(\g )\neq \emptyset$}}}}\Phi^{U^{\partial {\hat R}}, +}(\L' ))\Eqa(14)$$ Notice that, for any pair $\L\subset {\hat R}$, $\L'\subset {\hat R}$ that intersect neither the horizontal part of $\partial {\hat R}$ nor the lateral portion where $U^{\partial {\hat R}}\,=\,1$ and are one the translated of the other: $$\L'\,=\,F_i(\L )$$ we have: $$ \Phi^{U^{\partial {\hat R}}, +}(\L' )\;=\;\Phi^{U^{\partial {\hat R}}, +}(\L )\Eqa(15)$$ by the very definition of the coefficients $\Phi^{U^{\partial {\hat R}}, +}(\L )$.\par Therefore the difference between the two sums appearing in \equ(14) becomes simply $$\sum^\g_{\L,\L'} \Phi^{U^{\partial {\hat R}}, +}(\L' )\,-\,\Phi^{U^{\partial {\hat R}}, +}(\L )\Eqa(16)$$ where $ \sum^\g_{\L,\L'}$ is a shorthand notation for the sum over all pairs $\L$, $\L'$ which intersect $\D\g$ and $\D F_i(\g )$ respectively, and are such that one of the above two requirements is violated by $\L$ or $F_i( \L )$ and by $\L'$ or $F_i^{-1}( \L' )$, where $F_i^{-1}$ is the inverse of $F_i$.\par Since $\g\subset R_i$, $F_i(\g )\subset R_{N-2}$ and $d(\g )\,\leq \,{L_2\over 32}$, we can bound from above \equ(16), uniformly in $i=2\dots N-5$, by: $$\sum_{\L\cap R_i\neq \emptyset\atop \L\cap \{\partial {\hat R}\setminus \partial R_i\}\neq \emptyset }\vert \Phi^{U^{\partial {\hat R}}, +}(\L )\vert \;+\; \sum_{\L'\cap R_{N-2}\neq \emptyset \atop \L'\cap \{ \partial {\hat R}\setminus \partial R_{N-2}\}\neq \emptyset }\vert \Phi^{U^{\partial {\hat R}}, +}(\L' )\vert \;\leq \;C \Eqa(17)$$ for a suitable constant $C$ independent of ${\hat R}$.\par In \equ(17) we used the exponential decay of $\Phi^{U^{\partial {\hat R}}, +}(\L )$ in the "size" $d(\L )$ of $\L$, the fact that the distance between the horizontal part of the boundaries of ${\hat R}$, $R_i$, $R_{N-2}$ is, by construction and because $ d(\g )\leq {L_2\over 32}$, at least $L_2\over 32$ and the fact that the boundary coupling $U^{\partial {\hat R}}$ is equal to $\d$ on the lateral boundary of $R_i$, $i=2\dots N-5$ by construction.\par Let us consider the second ratio in \equ(12) $${Z(R^-_\g,U^{\partial {\hat R}}, \t)Z(R^-_{F_i(\g )},U^{\partial {\hat R}}, -)\over Z(R^-_\g,U^{\partial {\hat R}}, -)Z(R^-_{F_i(\g )},U^{\partial {\hat R}}, \t)}$$ Using Jensen inequality we obtain: $$\eqalign{{Z(R^-_\g,U^{\partial {\hat R}}, \t)\over Z(R^-_\g,U^{\partial {\hat R}}, -)}\;&\leq \; \exp (2\beta\delta\sum_{(x,y)\in \partial R^-_\g;U^{\partial {\hat R}}(x,y)=\d}<\s (x)>^\t)\cr {Z(R^-_{F_i(\g )},U^{\partial {\hat R}}, -)\over Z(R^-_{F_i(\g )},U^{\partial {\hat R}}, \t)}\;&\leq \; \exp (-2\beta\d\sum_{(x,y)\in \partial R^-_{F_i(\g )};U^{\partial {\hat R}}(x,y)=\d}<\s (x)>^-)}\Eqa(18)$$ where $<\s (x)>^\t$ is a shorthand notation for the average of the spin $\s (x)$ in the Gibbs measure $\mu_{R^-_\g}^{U^{\partial {\hat R}},\t}$ and similarly for $<\s (x)>^-$.\par A simple Peierls argument shows that $$<\s (x)>^\t\;\leq\;-1\,+\,k\qquad \forall\;x\in \,\partial_{int}R^-_\g;\;U^{\partial {\hat R}}(x,y)=\d $$ with $k\,\to\,0$ as $\beta\,\to \, \infty$, so that, from \equ(18) we obtain that $${Z(R^-_\g,U^{\partial {\hat R}}, \t)Z(R^-_{F_i(\g )},U^{\partial {\hat R}}, -)\over Z(R^-_\g,U^{\partial {\hat R}}, -)Z(R^-_{F_i(\g )},U^{\partial {\hat R}}, \t)}\,\leq\,\exp (-\beta \d l\,+\,2\beta\d k L_2)\Eqa(19)$$ Finally, combining \equ(17) and \equ(19), we obtain that: $$P_i\;\leq\;C\exp (-\beta \d l\,+\,2\beta\d k L_2)\quad \forall \; i=2\dots N-5\Eqa(20)$$ Let us now treat the case $i=1$ by estimating from above the ratio $P_1\over P_{N-1}$. We define the map $F_1$ to be simply the clockwise rotation of $\pi$ around the center of the rectangle ${\hat R}$ and we proceed as before. In this case, by symmetry, the ratio \equ(13) with $F_i(\g )$ replaced by $F_1(\g )$ is equal to one and the rest of the argument does not change.\par In conclusion, since $l$ is proportional to $L_2$ and $k$ is very small for large $\beta$, we get that, for any $m\,>\,0$ $$P({\cal A}^c_{{\hat R}}\cap {\cal C})\,\leq \,\exp (-m L_1^{\e})\Eqa(21)$$ Thus, combining together \equ(3), \equ(3bis), Lemma A.1 and \equ(21) we get the first part of the proposition. \medskip The second part follows immediately by a standard Peierls argument.\par The proposition is proved.\bigskip {\bf Proof of Proposition 4.2}\medskip Following the proof of proposition 4.1 let $\t$ be the configuration: $$\eqalign{\t (x)\,&=\,+1\quad \forall \,x\in \partial_{ext}R\quad\hbox{with }x_2\,\leq\,L_2-{L_2\over 16}\cr \t (x)\,&=\,-1\quad \hbox{otherwise}}$$ and let $$\eqalign{ U^{\partial R}(x,y)\,&=\,\delta \qquad \forall \,(x,y)\in \partial R\quad \hbox{with }\,0\,\leq\,x_2\,<\,L_2\cr U^{\partial R}(x,y)\,&=\,1\qquad \hbox{otherwise}}$$ Let also $S$ be the cigar-shaped neighborhood of the segment of the horizontal line at height $L_2-{L_2\over 16}$ and joining the two vertical sides of $R$: $$S\,=\,\{(x_1,x_2)\in R;\;\vert x_2-(L_2-{L_2\over 16})\vert \,\leq\,({x_1(L_1-x_1)\over L_1})^{{1+\e\over 2}}\}\Eqa(22)$$ Notice that, for large values of $L_1$, the region $S$ is at distance at least $L_2\over 32$ from the upper horizontal side of $R$.\par Then it is immediate to see that $$\mu_R^{\delta +,\delta +,-,\delta +}(({\cal A}_R^{+,+,-,+}))\,\geq\, \mu_{R}^{U^{\partial R},\t}({\cal S}_R^{\t})\Eqa(23)$$ where $${\cal S}_R^{\t}\,=\,\{\s ;\;\G_{R,open}^{\t}\,\subset\,S\}$$ Let ${\cal F}_R$ be the set of all possible configurations of $\G_{R,open}^{\t}$. As in \equ(5.a) we write: $$\mu_{R}^{U^{\partial R},\t}({\cal S}_R^{\t})\,=\, {Z(R,U^{\partial R},-)\over Z(R,U^{\partial R},\t)}\sum_{\G\in {\cal F}_R;\,\G\subset S }\exp (-2\beta \vert \G \vert ) {Z(R^+_\G,U^{\partial R}, \t)Z(R^-_\G,U^{\partial R},\t )\over Z(R,U^{\partial R},-)}\Eqa(24)$$ where $R^+_\G$ and $R^-_\G$ are defined as in the proof of proposition 4.1 .\par The ratio ${Z(R,U^{\partial R},-)\over Z(R,U^{\partial R},\t)}$ is clearly bounded from below by $$\exp (-\delta (2L_2+L_1)\beta )\Eqa(24bis)$$ Notice that, since the polygonal line $\G$ is contained in the region $S$, the boundary conditions in the partitions functions $Z(R^+_\G,U^{\partial R}, \t)$ and $Z(R^-_\G,U^{\partial R}, \t)$ are, by construction, $+$ and $-$ respectively. Therefore, using the representation (1.10), the ratio in the second factor in \equ(24) can be written as: $${Z(R^+_\G,U^{\partial R}, \t)Z(R^-_\G,U^{\partial R},\t )\over Z(R,U^{\partial R},-)}\,=\, \exp (-\sum_{\L\subset R\atop\L\cap \D\g\neq \emptyset}\Phi^{U^{\partial R}, +}(\L ))\Eqa(25)$$ Using proposition 1.1 and the fact that the polygonal line $\G$ is contained in the region $S$, it is easy to see that: $$\sum_{\L\subset R\atop \L\cap \D\g\neq \emptyset}\Phi^{U^{\partial R}, +}(\L )\,\leq\, \sum_{\L\cap \D\g\neq \emptyset}\Phi^{+}(\L ) \,+\,C_o$$ for a suitable constant $C_o$ independent of $L_1$.\par Thus \equ(24) can be estimated from below by $$\exp (-C_o\,-\,3\d \b L_1)\sum_{\G\in {\cal F}_R;\,\G\subset S }\exp (-2\beta \vert \G \vert -\sum_{\L\subset R\atop \L\cap \D\g\neq \emptyset}\Phi^{ +}(\L ))\Eqa(26)$$ We use at this point the fundamental result of [DKS] (see section 4.16 ) which says that: $$ \sum_{\G\in {\cal F}_R;\,\G\subset S }\exp (-2\beta \vert \G \vert -\sum_{\L\subset R\atop\L\cap \D\g\neq \emptyset}\Phi^{U^{\partial R}, +}(\L ))\,\geq\,$$ $$ \exp (- \beta L_1\t_\beta- C(\log (L_1))^{max(6,{2\over \e})})\Eqa(27)$$ If we combine together \equ(23) \equ(24), \equ(24bis), \equ(26) and \equ(27) we finally get the result.\bigskip {\bf Proof of Proposition 4.3}\medskip It is easy to show that the expression appearing in part a) is bounded from above by: $$\vert F_2\vert _\infty \sum_{x\in \partial_{ext}(\hbox{bottom side of }{Q_2})} \mu_{Q_1}^{\d +,\d +,-,\d +}(\t (x)=1\vert {\cal A}_{Q_1}^{+,+,-,+})\,-\, \mu_{R_{n+1}\cup Q_{n+1}}^{\d +,\d +,+,\d +}(\t (x)=1) \Eqa(28)$$ By monotonicity $$\mu_{Q_1}^{\d +,\d +,-,\d +}(\t (x)=1\vert {\cal A}_{Q_1}^{+,+,-,+})\,\leq\, \mu_{R_1}^{\d +,\d +,+,\d +}(\t (x)=1)$$ A standard Peierls argument shows that, for each $x\in \partial_{ext} (\hbox{bottom side of }{Q_2})$ and any given positive $m$ $$0\,\leq \,\mu_{R_1}^{\d +,\d +,+,\d +}(\t (x)=1)\,-\, \mu_{R_{n+1}\cup Q_{n+1}}^{\d +,\d +,+,\d +}(\t (x)=1)\,\leq\, \exp (-m L^{{1\over 2}+\e})\Eqa(29)$$ provided that $\beta$ is large enough.\par Clearly \equ(29) proves part a) since $$\vert F_2\vert _\infty \,\leq\,2L^2$$ Part b) follows immediately from part a) and proposition 4.1 applied to the rectangle $R_{n+1}\cup Q_{n+1}$. The proposition is proved.\pagina %\input formato.tex \centerline{\bf References}\bigskip\noindent \tolerance 10000 \refj{A}{D.Aldous}{Markov chains with almost exponential hitting times}{Stoch. Proc. Appl.}{13}{305}{1982} \refj{CCSch}{J.Chayes, L.Chayes, R.H.Schonmann}{Exponential decay of connectivities in the two-dimensional Ising model.} {Journ. Stat. Phys.}{49}{433}{1987} \refj{DSh}{R.L. Dobrushin, S. Shlosman}{Constructive Criterion for the Uniqueness of Gibbs fields.} {Statistical Physics and Dynamical Systems, Fritz, J., Jaffe, A. and Sz\'asz, D., eds., Birkhauser} {}{347}{1985} \refj{DKS}{R.Dobrushin, R.Koteck\'y, S.Shlosman}{Wulff Construction. A Global Shape From Local Interaction}{Translation of Math. Monographs, AMS}{104}{}{1992} \refj{DS}{P.Diaconis, D.W.Stroock}{Geometric Bounds on the Eigenvalues of Markov Chains}{The Annals of Appl. Prob.}{1}{No1, 36}{1991} \refj{FH}{D.Fisher, D.Huse}{}{Phys.Rev.B}{35}{6841}{1987} \refj{FKG}{C.M.Fortuin, P.W.Kasteleyn, J.Ginibre}{Correlations Inequalities on Some Partially Ordered Sets}{Comm. Math. Phys.}{22}{89}{1971} \refj{H}{R.Holley}{Possible Rates of Convergence in Finite Range Attractive Spin Systems}{Contemp.Math.}{41}{215}{1985} \refj{KO1}{R.Koteck\'y, E.Olivieri}{Droplet Dynamics for Asymmetric Ising Models}{Journ. Stat. Phys.}{70}{1121}{1993} \refj{KO2}{R.Koteck\'y,E.Olivieri}{Shapes of Growing Droplets: a Model of Escape from a Metastable Phase}{preprint }{}{}{1993} \refj{JS1}{M.Jerrum, A.Sinclair}{Approximating the Permanent}{Siam Journ.Comput.}{18}{1149}{1989} \refj{JS2}{M.Jerrum, A.Sinclair}{Polynomial-Time Approximation Algorithms for the Ising Model}{SIAM Journ. Comput.}{22}{1087}{1993} \refj{Li}{T.Ligget}{Interacting Particle Systems}{Springer-Verlag}{}{}{1985} \refj{LS}{G.Lawler, A.Sokal}{Bounds on the $L^2$ Spectrum for Markov Chains and Markov Processes: a Generalization of Cheeger Inequality}{Trans. Amer.Math. Soc.}{309}{557}{1988} \refj{LSch}{J.Lebowitz, R.Schonmann}{On the asymptotics of occurrence times of rare events for stochastic spin systems}{Journ.Stat.Phys.}{48}{727}{1987} \refj{M}{F.Martinelli}{Low Temperature Stochastic Spin Dynamics: Convergence to Equilibrium, Metastability and Phase Segregation}{in Proc. X Intern. Conf. Math. Phys. Leipzig}{}{K.Schmudgen ed. Springer Verlag}{1991} \refj{MO1}{F.Martinelli, E.Olivieri}{Approach to Equilibrium of Glauber Dynamics in the One Phase Region I: The Attractive Case} {Comm.Math.Phys. in press}{}{}{1992} \refj{MO2}{F.Martinelli, E.Olivieri}{Approach to Equilibrium of Glauber Dynamics in the One Phase Region II: The General Case}{Comm.Math.Phys. in press}{}{}{1992} \refj{MOS}{F.Martinelli, E.Olivieri, R.Schonmann}{For Gibbs state of 2D Lattice spin Systems Weak Mixing Implies Strong Mixing}{Comm.Math.Phys. in press}{}{}{1992} \refj{Og}{Ogielski}{Dynamics of fluctuations in the order phase of kinetic Ising model}{Phys. Rev. Lett.}{36,No 13}{7315}{1987} \refj{Pf}{C.Pfister}{Large Deviations and Phase Separation in the Two Dimensional Ising Model}{Elv. Phys. Acta}{64}{953}{1991} \refj{Sc}{E.Scoppola}{Renormalization Group for Markov Chains and Application to Metastability}{Journ. Stat. Phys. in press}{}{}{1993} \refj{Sch1}{R.Schonmann}{Slow Droplet Driven Relaxation of Stochastic Ising Models in the Vicinity of the Phase Coexistence Region}{Comm.Math.Phys. in press}{}{}{1993} \refj{Sch2}{R.Schonmann}{Second Order Large Deviations Estimates for Ferromagnetic Systems in the Phase Coexistence region}{Comm.Math.Phys.}{112}{409}{1987} \refj{Sh}{S.Shlosman}{The Droplet in the Tube: A Case of Phase Transition in the Canonical Ensemble}{Comm.Math.Phys.}{125}{81}{1989} \refj{Si}{A.Sinclair}{Improved Bounds for Mixing Rates of Markov Chains and Multicommodity Flow} {Combinatorics, Probability and Computing} {1} {351}{1992} \refj{SLY}{ShengLin Lu, H.T.Yau}{Spectral Gap and Logarithmic Sobolev Inequality for Kawasaki and Glauber Dynamics}{Comm.Math.Phys.}{156}{399}{1993} \refj{SZ}{D.W.Stroock, B. Zegarlinski}{The Logarithmic Sobolev Inequality for Discrete Spin Systems on a Lattice}{Comm. Math.Phys.}{149}{175}{1992} \refj{T}{L.Thomas}{Bounds on the Mass Gap for the Finite Volume Stochastic Ising Models at Low Temperatures}{Comm.Math.Phys.}{126}{1}{1989} \end ENDBODY