Computational Design of Nano-clusters by Property-Based Cascade Genetic Algorithms: Tuning the Electronic Properties of (TiO2)n Clusters

For complex open systems such as atomic clusters, defected surfaces, structured over-layers adsorbed on inorganic surfaces, human intuition for predicting relevant structures is likely incomplete or even misleading. Thus, an unbiased algorithm is required for the global optimization. In order to obtain both extensive and accurate sampling of the configurational space - we have developed a massively parallel cascade genetic algorithm (cGA^)[1-3]. The term “cascade” refers to a multi-stepped procedure involving increasing levels of accuracy for the evaluation of the globally-optimized quantity (usually total energy of the system). Typically, a cGA starts with classical force field and goes up to density functional theory (DFT) with hybrid functionals. This development has already been applied and successful in various inter-disciplinary fields in materials science^[4-6].
         Recently, we have extended our cGA-implementation to property based potential energy surface (PES)-scanning to address the famous “inverse problem” of materials science, i.e. how to computationally design materials/structures with the desired electronic properties as opposed to calculating the properties of the given material/structure. In this talk, I shall discuss the fundamental challenges behind this implementation in the context of computational design of cluster-based nano-catalysts^[7].