The future of computational catalysis

The future of computational heterogeneous catalysis is shaped by machine learning in two different but equally important areas: (i) development of atomistic potentials that closely approximate DFT and wavefunction based ab initio methods (MP2, CCSD(T)), but are computationally more efficient, and (ii) finding structure and reactivity descriptors for predicting catalytic materials and reactions. Machine Learning Potentials will enable improved sampling of the potential energy surface (PES) for reaction conditions, but they will not do better than the data on which they are trained. Therefore, this perspective focusses on ways of improving the quality of the PES beyond the generalized gradient approximation (GGA) climbing the Jacob's ladder of functionals in DFT up to RPA and using wavefunction methods such as MP2 and CCSD(T). It is problem solving in close collaboration with experiment that has made computational methodology relevant and this remains an important aspect of data science methods in heterogeneous catalysis.