English

Optimization by Move--Class Deflation

Statistical Mechanics 2007-05-23 v1 Disordered Systems and Neural Networks

Abstract

A new approach to combinatorial optimization based on systematic move-class deflation is proposed. The algorithm combines heuristics of genetic algorithms and simulated annealing, and is mainly entropy-driven. It is tested on two problems known to be NP hard, namely the problem of finding ground states of the SK spin--glass and of the 3-DD ±J\pm J spin-glass. The algorithm is sensitive to properties of phase spaces of complex systems other than those explored by simulated annealing, and it may therefore also be used as a diagnostic instrument. Moreover, dynamic freezing transitions, which are well known to hamper the performance of simulated annealing in the large system limit are not encountered by the present setup.

Keywords

Cite

@article{arxiv.cond-mat/9805137,
  title  = {Optimization by Move--Class Deflation},
  author = {Reimer Kuehn and Yu-Cheng Lin and Gerhard Poeppel},
  journal= {arXiv preprint arXiv:cond-mat/9805137},
  year   = {2007}
}

Comments

16 pages, 10 eps figures