Insights into Segregation and Aggregation in Dilute Atom Alloy Catalysts Using DFT and Machine Learning

Abstract

Dilute atom alloys (DAAs) are an important class of heterogeneous catalysts due to their ability to precisely tune the activity and selectivity of reactions. DAA catalysts typically consist of a small quantity of metal solute in a metal host. Key considerations in the stability of DAA catalysts are the segregation and aggregation energy. In this work, we report a systematic theoretical study of segregation and aggregation energies of DAA catalysts composed of 3d, 4d, and 5d transition metals. To investigate the nature of DAAs, we analyzed both Bader charge and density of states, as well as formation energies, to identify the most stable DAA configuration for a given alloy. We further applied regression-based, tree-based, and neural network machine learning (ML) models to gain physics-based insights in predicting segregation and aggregation energies based on readily available atomic and bulk features. We found that the d-band filling of the solute and host, nearest neighbor distance of the host, and d-band width of the solute determine the segregation energy, whereas the Pauling electronegativity of the host and solute, nearest neighbor distance of the host, and cohesive energy of host determine aggregation energy. Our findings provide crucial insights for DAA catalyst design.