Machine-Learning-Accelerated\nDFT Conformal Sampling\nof Catalytic Processes

Computational modeling of catalytic processes at gas/solid\ninterfaces\nplays an increasingly important role in chemistry, enabling accelerated\nmaterials and process optimization and rational design. However, efficiency,\naccuracy, thoroughness, and throughput must be enhanced to maximize\nits practical impact. By combining interpolation of DFT energetics\nvia highly accurate Machine-Learning Potentials with conformal techniques\nfor building the training database, we present here an original approach\n(that we name Conformal Sampling of Catalytic Processes, CSCP), to\naccelerate and achieve an accurate and thorough sampling of novel\nsystems by exporting existing information on a worked-out case. We\nuse methanol decomposition (of interest in the field of hydrogen production\nand storage) as a test catalytic reaction. Starting from worked-out\nPt-based systems, we show that after only two iterations of active-learning\nCSCP is able to provide reaction energy diagrams for a set of 7 diverse\nsystems (Pd, Ni, Au, Ag, Cu, Co, Fe) leading to DFT-accuracy-level\npredictions. Cases exhibiting a change in adsorption sites and mechanisms\nare also successfully reproduced as tests of catalytic path modification.\nThe CSCP approach thus offers itself as an operative tool to fully\ntake advantage of accumulated information to achieve high-throughput\nsampling of catalytic processes.