Ni-CeO2 nanoparticles (NPs) are promising nanocatalysts for water splitting and water gas shift reactions due to the ability of ceria to temporarily donate oxygen to the catalytic reactions and accept oxygen after the reaction is completed. Therefore, elucidating how different properties of the Ni-Ceria NPs, such as size, composition, and electronic structure, relate to the activity and selectivity of the catalytic reaction is of crucial importance for the development of novel catalysts.

In this work we use machine learning (ML) regression and its uncertainty -- an active-learning (AL) strategy -- for the global optimization of Ce(y-x)NixO(2y-x) (x=1, 2, 3; y=4) nanoparticles guided by density functional theory (DFT) calculations. The method allows the learned structure-energy relationship to improve iteratively when more data are obtained from DFT calculations of promising structures indicated by the AL.

The progress of the on-the-fly AL method from different independent AL runs are reported. The electronic structure and magnetic properties of the “putative” GM found by AL are analyzed by plots of the density of states, the spin density, the electron localization function and by population analysis, such as Mulliken, and Bader analysis. Additionally, further investigation of the NPs by mass scaled parallel-tempering Born Oppenheimer molecular dynamic (PT-BOMD) resulted in the same GM structures found by AL, demonstrating the robustness of our AL search to learn from small datasets and assist in the global optimization of complex electronic structure systems, such as Ni-Ceria NPs.