Universality results for kinetically constrained spin models in two dimensions

TL;DR

This paper establishes universality classes for infection times in two-dimensional kinetically constrained spin models, revealing different behaviors from bootstrap percolation and confirming some conjectures about their sharpness.

Contribution

It identifies new universality classes for KCM infection times and proves universal bounds, advancing understanding of their glassy dynamics and differences from bootstrap percolation.

Findings

Universal upper bounds on mean infection time within each class

Different universality classes from bootstrap percolation

Sharpness of bounds confirmed in specific models like Duarte

Abstract

Kinetically constrained models (KCM) are reversible interacting particle systems on $\mathbb Z^d$ with continuous time Markov dynamics of Glauber type, which represent a natural stochastic (and non-monotone) counterpart of the family of cellular automata known as $\mathcal U$-bootstrap percolation. KCM also display some of the peculiar features of the so-called "glassy dynamics", and as such they are extensively used in the physics literature to model the liquid-glass transition, a major and longstanding open problem in condensed matter physics. We consider two-dimensional KCM with update rule $\mathcal U$, and focus on proving universality results for the mean infection time of the origin, in the same spirit as those recently established in the setting of $\mathcal U$-bootstrap percolation. We first identify what we believe are the correct universality classes, which turn out to be different from those of $\mathcal U$-bootstrap percolation. We then prove universal upper bounds on the mean infection time within each class, which we conjecture to be sharp up to logarithmic corrections. In certain cases, including all supercritical models, and the well-known Duarte model, our conjecture has recently been confirmed in [MMT]. In fact, in these cases our upper bound is sharp up to a constant factor in the exponent. For certain classes of update rules, it turns out that the infection time of the KCM diverges much faster than for the corresponding $\mathcal U$-bootstrap process when the equilibrium density of infected sites goes to zero. This is due to the occurrence of energy barriers which determine the dominant behaviour for KCM, but which do not matter for the monotone bootstrap dynamics.