Saved in:
Bibliographic Details
Main Authors: Barzegar, Amin, Hamze, Firas, Amey, Christopher, Machta, Jonathan
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2402.14717
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917549687439360
author Barzegar, Amin
Hamze, Firas
Amey, Christopher
Machta, Jonathan
author_facet Barzegar, Amin
Hamze, Firas
Amey, Christopher
Machta, Jonathan
contents Annealing algorithms such as simulated annealing and population annealing are widely used both for sampling the Gibbs distribution and solving optimization problems (i.e. finding ground states). For both statistical mechanics and optimization, additional parameters beyond temperature are often needed such as chemical potentials, external fields or Lagrange multipliers enforcing constraints. In this paper we derive a formalism for optimal annealing schedules in multidimensional parameter spaces using methods from non-equilibrium statistical mechanics. The results are closely related to work on optimal control of thermodynamic systems [Sivak and Crooks, PRL 108, 190602 (2012)]. Within the formalism, we compare the efficiency of population annealing and multiple weighted runs of simulated annealing ("annealed importance sampling") and discuss the effects of non-ergodicity on both algorithms. Theoretical results are supported by numerical simulations of spin glasses.
format Preprint
id arxiv_https___arxiv_org_abs_2402_14717
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Optimal schedules for annealing algorithms
Barzegar, Amin
Hamze, Firas
Amey, Christopher
Machta, Jonathan
Statistical Mechanics
Computational Physics
Annealing algorithms such as simulated annealing and population annealing are widely used both for sampling the Gibbs distribution and solving optimization problems (i.e. finding ground states). For both statistical mechanics and optimization, additional parameters beyond temperature are often needed such as chemical potentials, external fields or Lagrange multipliers enforcing constraints. In this paper we derive a formalism for optimal annealing schedules in multidimensional parameter spaces using methods from non-equilibrium statistical mechanics. The results are closely related to work on optimal control of thermodynamic systems [Sivak and Crooks, PRL 108, 190602 (2012)]. Within the formalism, we compare the efficiency of population annealing and multiple weighted runs of simulated annealing ("annealed importance sampling") and discuss the effects of non-ergodicity on both algorithms. Theoretical results are supported by numerical simulations of spin glasses.
title Optimal schedules for annealing algorithms
topic Statistical Mechanics
Computational Physics
url https://arxiv.org/abs/2402.14717