(479a) Data-Driven Acceleration of Catalyst Modeling

Computational catalysis has made many contributions to our understanding of catalyst reactivity and selectivity. A key method in this field has been density functional theory. A critical challenge in using density functional theory though is that it is computationally expensive, which limits the size of system that can be simulated, and the number of computations that can be done. Although it is now routine to do 1000s of calculations with even hundreds of atoms, some simulation methods like molecular dynamics and Monte Carlo can require millions of calculations on systems with thousands of atoms. In this talk, I will talk about how we leverage the DFT data we can compute with ideas and tools from machine learning to speed up DFT calculations, enabling more data to be generated. I will also show how can use this data to build surrogate models that can be used in Monte Carlo simulations to model alloy surface segregation in the dilute limit, and with adsorbates. These simulations require large systems to reach the dilute limit, and statistical averaging that is not possible with DFT. By judicious coarse graining, and development of machine learning models we are able to estimate upper bounds on the bulk composition that maximizes the number of single atom sites without having too many dimer sites. We will illustrate this for an Ag-Pd alloy in the dilute Pd limit in the presence of acrolein.