A broad review that explains DFT and machine learning from the ground up, then maps how the two have been combined across chemistry and materials science up to late 2020.
DFT calculations account for a substantial share of academic high-performance computing, so making them cheaper or extending their reach has obvious value. This review starts with accessible introductions to both density functional theory and machine learning, so that readers from either community can follow the argument.
The survey then covers research articles in chemistry and materials science that use machine learning alongside DFT, with literature through December 2020. The authors organize the field into main research threads, highlight the results that proved influential, and use citation analysis to suggest where the field is moving. Their outlook stresses that learned acceleration can both reduce the resources a calculation needs and enable simulations of larger systems, with consequences for in-silico materials discovery and the search for new reaction pathways.