Quantum-Enhanced Machine Learning Framework for Photocatalytic Materials Discovery: Integrating Quantum Chemical Descriptors and Many-Body Electronic Structure Theory

Abstract

Photocatalytic materials belong to a large class of quantum materials, whose electronic structure controls charge carriers' behavior and reaction paths via quantum mechanics. To produce superior photocatalytic quantum materials, we recommend a quantum-enhanced machine learning scheme that blends multi-body estimates of electronic structure with advanced machine learning techniques. The effects of quantum coherence lengths, electron-phonon coupling strengths, exciton binding energies, as well as spin-orbit coupling obtained from density functional theory and several body perturbation theory calculations will be discussed. The database is based on a screened library of 8,642 photocatalytic materials that have been experimentally characterized in H 2 evolution reactions and contains 152 descriptors relating to electronic correlations, quantum confinement effects, and spin-dependent properties. A cross-validation report shows promising predictive results (R 2 = 0.8, MAE = 0.28 log units), and specifically accounting for quantum effects improves accuracy by 18% over classical descriptors. Through a high-throughput investigation of 45,000 hypothetical compositions, where DFT and GW-BSE validation on the 23 candidates revealed 75% agreement within quantum uncertainty bounds. Quantum coherence has been shown to have a major effect on photocatalytic charge separation, with optimal coherence lengths of 15-35 nm being highly associated with increased production. Our quantum-mechanical findings show that many body interactions, such as exciton dissociation dynamics and the contribution of quantum tunneling to charge transfer, can predict photocatalytic responses in new emerging quantum materials.