Free energy landscapes, dynamics and the edge of chaos in mean-field models of spin glasses

TL;DR

This paper investigates the free energy landscapes and dynamics of mean-field spin-glass models, revealing differences between TAP and naive mean-field equations, and explores the edge of chaos as a method to find metastable states.

Contribution

It compares TAP and NMF equations in spin glasses, analyzes their free energy landscapes, and introduces an iterative scheme to locate metastable states near the edge of chaos.

Findings

TAP states form pairs of minima and saddle points with specific scaling laws.

Barriers between states scale as N^{1/3} for pure states.

Iterative method effectively finds metastable states near the edge of chaos.

Abstract

Metastable states in Ising spin-glass models are studied by finding iterative solutions of mean-field equations for the local magnetizations. Two different equations are studied: the TAP equations which are exact for the SK model, and the simpler `naive-mean-field' (NMF) equations. The free-energy landscapes that emerge are very different. For the TAP equations, the numerical studies confirm the analytical results of Aspelmeier et al., which predict that TAP states consist of close pairs of minima and index-one (one unstable direction) saddle points, while for the NMF equations saddle points with large indices are found. For TAP the barrier height between a minimum and its nearby saddle point scales as (f-f_0)^{-1/3} where f is the free energy per spin of the solution and f_0 is the equilibrium free energy per spin. This means that for `pure states', for which f-f_0 is of order 1/N, the barriers scale as N^{1/3}, but between states for which f-f_0 is of order one the barriers are finite and also small so such metastable states will be of limited physical significance. For the NMF equations there are saddles of index K and we can demonstrate that their complexity Sigma_K scales as a function of K/N. We have also employed an iterative scheme with a free parameter that can be adjusted to bring the system of equations close to the `edge of chaos'. Both for the TAP and NME equations it is possible with this approach to find metastable states whose free energy per spin is close to f_0. As N increases, it becomes harder and harder to find solutions near the edge of chaos, but nevertheless the results which can be obtained are competitive with those achieved by more time-consuming computing methods and suggest that this method may be of general utility.