Optimization by thermal cycling

TL;DR

This paper introduces a novel optimization algorithm that uses cyclic heating and quenching, demonstrating superior efficiency over traditional simulated annealing and competitive speed with genetic algorithms in solving the traveling salesman problem.

Contribution

The paper presents a new thermal cycling optimization method that outperforms standard simulated annealing and matches the speed of advanced genetic local search algorithms.

Findings

Outperforms traditional simulated annealing in efficiency.

Competitive in speed with recent genetic local search algorithms.

Effective when applied to an archive of samples rather than a single sample.

Abstract

An optimization algorithm is presented which consists of cyclically heating and quenching by Metropolis and local search procedures, respectively. It works particularly well when it is applied to an archive of samples instead of to a single one. We demonstrate for the traveling salesman problem that this algorithm is far more efficient than usual simulated annealing; our implementation can compete concerning speed with recent, very fast genetic local search algorithms.