Robustness of "cut and splice" genetic algorithms in the structural optimization of atomic clusters

TL;DR

This paper investigates the robustness of 'cut and splice' genetic algorithms in optimizing atomic cluster structures, highlighting how mutation strategies improve stability across different population sizes, crucial for complex first-principles calculations.

Contribution

It demonstrates that incorporating twinning mutation moves enhances the robustness of genetic algorithms in atomic cluster optimization, especially over varying population sizes.

Findings

Twinning mutation improves algorithm robustness.

Stable performance over a wide range of population sizes.

Relevance for first-principles based sampling.

Abstract

We return to the geometry optimization problem of Lennard-Jones clusters to analyze the performance dependence of "cut and splice" genetic algorithms (GAs) on the employed population size. We generally find that admixing twinning mutation moves leads to an improved robustness of the algorithm efficiency with respect to this a priori unknown technical parameter. The resulting very stable performance of the corresponding mutation+mating GA implementation over a wide range of population sizes is an important feature when addressing unknown systems with computationally involved first-principles based GA sampling.