Observable-targeting global cluster structure optimization

Abstract

Traditionally, global cluster structure optimization is done by minimizing energy. As an alternative, we propose minimizing the difference between actual experimental observables and their simulated counterparts. To validate and explain this approach, test cases for small clusters are shown. Additionally, an application to real-life data for a larger cluster illustrates the advantages of this method: it provides direct links between properties and structure, and avoids problems both with insufficient accuracy in theoretical energy-ordering and with non-equilibrium conditions in experiment.