Improved cluster structure optimization: Hybridizing evolutionary algorithms with local heat pulses

Abstract

Cluster structure optimization (CSO) refers to finding the globally minimal cluster structure with respect to a specific model and quality criterion, and is a computationally extraordinarily hard problem. Here we report a successful hybridization of evolutionary algorithms (EAs) with local heat pulses (LHPs). We describe the algorithm's implementation and assess its performance with hard benchmark CSO cases. EA-LHP showed superior performance compared to regular EAs. Additionally, the EA-LHP hybrid is an unbiased, general CSO algorithm requiring no system-specific solution knowledge. These are compelling arguments for a wider future use of EA-LHP in CSO.