(474a) Automated Generation of Metal-Doped Amorphous Silica Clusters

Single-site metal-doped amorphous materials are widely used in heterogeneous catalytic processes, particularly in industrial applications. Unfortunately, the rational design of these materials is hindered by the irregularity of the support, as it prevents the experimental characterization of the various single-site metal structures. Thus, we aim to characterize the number of possible active sites using electronic structure theory. Further difficulties arise, however, as millions of structurally different active sites must be sampled to estimate a rate constant, making a brute-force approach impractical. To address this issue, we have developed an algorithm that automatically generates metal-doped amorphous silica clusters from slabs of amorphous silica, also automatically made using molecular dynamic simulations. This cluster generation algorithm is combined with machine learning to develop training sets on the fly while also using the clusters to guide the sampling so that it is focused on active sites with low activation energies. This combination of algorithms may be tested by automatically generating Nb(IV) doped amorphous silica clusters and investigating their ethylene epoxidation reactivity, as the mechanism is already known, to derive the predictive ability of our combination of algorithms.