(706e) First Principles Prediction of Active Sites for Bimetallic Catalysts

It has been well established that bimetallic catalysts exhibit properties which are not an interpolation of the parent metals. As a result, they offer great promise for developing superior, non-precious metal catalysts. Their structure is tunable using modern synthesis techniques, which offers a way to optimize activity and selectivity for the chemistry of interest. Although some studies have attempted to interrogate their structure¹ and make inferences about the active site, it is difficult to pinpoint which sites contribute to the chemistry.

First principles modeling allows for detailed analysis of the catalyst surface. Kinetic Monte Carlo (KMC) studies have been able to estimate activity for complex structures². However, the inverse problem of finding the structures which maximize activity is largely unexplored. To this end, we employ optimization techniques to screen the high-dimensional space of possible bimetallic structures using a suitable metric of merit. The binding energies of surface species on sites with dynamically generated metal coordination environments are computed using a Hamiltonian cluster expansion. Density Functional Theory (DFT) is used in conjunction with Bayesian statistics³ to obtain training data for the cluster expansion.

The approach is applied to the Hydrogen Evolution Reaction (HER) on Ti-Cu as well as ammonia synthesis on Ni-Pt. For both systems, we show that there are novel, non-intuitive structures whose performance is superior to previously explored structures. This work opens the possibility of atomic center design using first principles methods.