[Paper Review] Some New Results on the Kinetic Ising Model in a Pure Phase

T. Bodineau, Fabio Martinelli|ArXiv.org|Feb 8, 2002

Markov Chains and Monte Carlo Methods10 references4 citations

TL;DR

This paper establishes polynomial lower bounds for the spectral gap and logarithmic Sobolev constant in the kinetic Ising model under plus boundary conditions in dimensions $d \geq 2$, using a novel test function inspired by H.T. Yau. It proves the inverse spectral gap grows at least as $N$ (up to log corrections) in $d=2$, and the logarithmic Sobolev constant as $N^2$, supporting the conjecture that spin auto-correlations decay as a stretched exponential in two dimensions.

ABSTRACT

We consider a general class of Glauber dynamics reversible with respect to the standard Ising model in $\bbZ^d$ with zero external field and inverse temperature $\gb$ strictly larger than the critical value $\gb_c$ in dimension 2 or the so called ``slab threshold'' $\hat \b_c$ in dimension $d \geq 3$. We first prove that the inverse spectral gap in a large cube of side $N$ with plus boundary conditions is, apart from logarithmic corrections, larger than $N$ in $d=2$ while the logarithmic Sobolev constant is instead larger than $N^2$ in any dimension. Such a result substantially improves over all the previous existing bounds and agrees with a similar computations obtained in the framework of a one dimensional toy model based on mean curvature motion. The proof, based on a suggestion made by H.T. Yau some years ago, explicitly constructs a subtle test function which forces a large droplet of the minus phase inside the plus phase. The relevant bounds for general $d\ge 2$ are then obtained via a careful use of the recent $\bbL^1$--approach to the Wulff construction. Finally we prove that in $d=2$ the probability that two independent initial configurations, distributed according to the infinite volume plus phase and evolving under any coupling, agree at the origin at time $t$ is bounded from below by a stretched exponential $\exp(-\sqrt{t})$, again apart from logarithmic corrections. Such a result should be considered as a first step toward a rigorous proof that, as conjectured by Fisher and Huse some years ago, the equilibrium time auto-correlation of the spin at the origin decays as a stretched exponential in $d=2$.

Motivation & Objective

• To establish rigorous lower bounds for the spectral gap and logarithmic Sobolev constant in the Glauber dynamics of the Ising model in a pure phase.
• To understand the slow relaxation mechanism driven by interface motion in the phase coexistence regime.
• To provide a mathematical foundation for the conjecture by Fisher and Huse that spin auto-correlations decay as a stretched exponential in $d=2$.
• To validate the relevance of mean curvature motion heuristics in the dynamics of droplet shrinking via a one-dimensional toy model.

Proposed method

• Construction of a subtle test function that forces a large droplet of the minus phase inside the plus phase, inspired by H.T. Yau’s suggestion.
• Use of the ${\mathbb{L}}^{1}$–approach to the Wulff construction to extend results from $d=2$ to general $d \geq 2$.
• Application of Hardy-type inequalities to derive sharp bounds on the inverse spectral gap and logarithmic Sobolev constant in a one-dimensional birth-death process modeling droplet volume evolution.
• Analysis of the equilibrium measure $\mu(x) \propto \exp(-x^{(d-1)/d})$ for droplet volume $x$, with birth rate $b(x) \propto x^{(d-1)/d}$ and death rate derived from detailed balance.
• Use of variational formulas for Poincaré and logarithmic Sobolev inequalities, bounded via the test function to extract polynomial scaling in system size $N$.
• Comparison of the toy model’s dynamics to mean curvature motion, showing drift $\sim -x^{2\alpha - 1}$ with $\alpha = (d-1)/d$, consistent with curvature-driven shrinking.

Experimental results

Research questions

• RQ1 What is the asymptotic behavior of the inverse spectral gap for the Glauber dynamics of the Ising model in a large box with plus boundary conditions in $d=2$?
• RQ2 How does the logarithmic Sobolev constant scale with system size $N$ in dimensions $d \geq 2$?
• RQ3 Can the dynamics of droplet shrinking under Glauber dynamics be rigorously linked to mean curvature motion in the phase coexistence regime?
• RQ4 Does the spin auto-correlation function at the origin decay as a stretched exponential in $d=2$, as conjectured by Fisher and Huse?
• RQ5 Can a one-dimensional birth-death process model the volume evolution of a droplet in a way that reproduces the scaling of the spectral and log-Sobolev constants?

Key findings

• In $d=2$, the inverse spectral gap is bounded from below by $cN / (\log N)^k$ for some $k>0$, indicating relaxation time grows at least linearly with system size $N$.
• In any dimension $d \geq 2$, the logarithmic Sobolev constant is bounded from below by $cN^2 / (\log N)^k$, consistent with mean curvature-driven interface motion.
• For $d \geq 3$, the inverse spectral gap is bounded from above by a constant independent of $N$, suggesting faster relaxation than in $d=2$.
• In the one-dimensional toy model, the logarithmic Sobolev constant scales as $\Theta(N^2)$, and the inverse spectral gap as $\Theta(N)$ in $d=2$, matching the main results.
• The probability that two independent configurations evolve to agree at the origin at time $t$ is bounded below by $\exp(-\sqrt{t})$ (up to log corrections), supporting the stretched exponential decay of spin auto-correlations in $d=2$.
• The test function construction successfully captures the energy cost of forming a large droplet, enabling the derivation of polynomial lower bounds from equilibrium phase segregation estimates.