Machine-Learning-Accelerated Conformal Sampling of Methanol Catalytic Conversion on Bimetallic Systems

Abstract

We present an approach to accelerate the construction of reaction energy diagrams and mechanistic pathways of novel bimetallic catalytic systems by exploiting information on a known case, and we test it on the CuPd system for the catalytic decomposition of methanol. Based on machine-learning and conformal techniques for building training databases, our proposal realizes the multicomponent extension of the conformal sampling of catalytic process (CSCP) approach and maintains its same characteristic features of accuracy, efficiency, and throughput, thus in principle enabling a high-throughput screening of catalytic processes across general alloy compositions. Moreover, the so-derived CSCP reactive MLIP describes equally well the pure (Cu, Pd) and bimetallic (CuPd) catalysts, thus enabling a high-throughput screening for the given catalytic process.