Bound on annealing performance from stochastic thermodynamics, with application to simulated annealing

Yutong Luo, Yi-Zheng Zhen, Xiangjing Liu, Daniel Ebler, Oscar Dahlsten

Research output: Journal Publications and ReviewsRGC 21 - Publication in refereed journalpeer-review

2 Citations (Scopus)
55 Downloads (CityUHK Scholars)

Abstract

Annealing is the process of gradually lowering the temperature of a system to guide it towards its lowest energy states. In an accompanying paper [Y. Luo et al., Phys. Rev. E 108, L052105 (2023)], we derived a general bound on annealing performance by connecting annealing with stochastic thermodynamics tools, including a speed limit on state transformation from entropy production. We here describe the derivation of the general bound in detail. In addition, we analyze the case of simulated annealing with Glauber dynamics in depth. We show how to bound the two case-specific quantities appearing in the bound, namely the activity, a measure of the number of microstate jumps, and the change in relative entropy between the state and the instantaneous thermal state, which is due to temperature variation. We exemplify the arguments by numerical simulations on the Sherrington-Kirkpatrick (SK) model of spin glasses. © 2023 American Physical Society.
Original languageEnglish
Article number054119
JournalPhysical Review E
Volume108
Issue number5
Online published13 Nov 2023
DOIs
Publication statusPublished - Nov 2023

Publisher's Copyright Statement

  • COPYRIGHT TERMS OF DEPOSITED FINAL PUBLISHED VERSION FILE: Luo, Y., Zhen, Y-Z., Liu, X., Ebler, D., & Dahlsten, O. (2023). Bound on annealing performance from stochastic thermodynamics, with application to simulated annealing. Physical Review E, 108(5), Article 054119. https://doi.org/10.1103/PhysRevE.108.054119. The copyright of this article is owned by American Physical Society.

Fingerprint

Dive into the research topics of 'Bound on annealing performance from stochastic thermodynamics, with application to simulated annealing'. Together they form a unique fingerprint.

Cite this