Nature of ‎Metal-Support Interaction Discovered by Interpretable Machine ‎Learning

Metal catalysts supported on oxides play a paramount role in numerous industrial reactions. ‎Modulating metal-support interaction is a key strategy to boost ‎catalytic productivity and ‎stability; however, the nature of metal-support interaction and quantification remain major ‎unsolved problems. By ‎leveraging interpretable ‎machine learning, domain knowledge, and ‎experimental data available, we discover a physical metal-support interaction equation ‎applicable ‎to metal nanoparticles and adatoms on oxides, and oxide films on metals. Though ‎metal-oxygen interaction dominates metal-support interaction and determines the metal ‎composition effect, metal-metal interaction delineates the support effect. This ensures a principle ‎of strong metal-metal interaction for encapsulation of suboxide over metal ‎nanoparticles, ‎substantiated comprehensively by molecular dynamics simulations and ‎previous experiments. The ‎developed theory provides valuable insights and guidance in engineering the metal-support ‎systems. ‎