Machine learning to tame divergent density functional approximations: a new path to consensus materials design principles

• DOI: 10.1039/d1sc03701c
• Publication Date: 2021-09-02T13:01:03Z
• Authors: Chenru Duan; Shuxin Chen; Michael G. Taylor; Fang Liu; Heather J. Kulik
• Keywords: Chemical Physics (physics.chem-ph); FOS: Computer and information sciences; Condensed Matter - Materials Science; Computer Science - Machine Learning; Materials Science (cond-mat.mtrl-sci); FOS: Physical sciences; 01 natural sciences; Machine Learning (cs.LG); 0104 chemical sciences; Chemistry; Physics - Chemical Physics; 0103 physical sciences

Virtual high-throughput screening (VHTS) with density functional theory (DFT) and machine-learning (ML)-acceleration is essential in rapid materials discovery. By necessity, efficient DFT-based workflows are carried out a single approximation (DFA). Nevertheless, properties evaluated different DFAs can be expected to disagree for cases challenging electronic structure (e.g., open-shell transition-metal complexes, TMCs) which most needed accurate benchmarks often unavailable. To quantify the effect of DFA bias, we introduce an approach rapidly obtain property predictions from 23 representative spanning multiple families, "rungs" semi-local double hybrid) basis sets on over 2000 TMCs. Although computed values spin state splitting frontier orbital gap) differ by DFA, high linear correlations persist across all DFAs. We train independent ML models each observe convergent trends feature importance, providing DFA-invariant, universal design rules. devise strategy artificial neural network (ANN) informed use them predict spin-splitting energy) 187k requiring consensus ANN-predicted properties, improve correspondence computational lead compounds literature-mined, experimental typically employed single-DFA approach.