(699a) Catkit: Symmetry Methods for Automated Generation of Catalytic Structures

Cheminformatics have dramatically accelerated the rate at which we discover useful properties of chemical species, originally for the purpose of drug discovery. This acceleration is made possible by taking advantage of bond symmetries present in organic molecules, allowing computers to systematically enumerate through an otherwise infinite search space. These techniques are also easily extend to other forms of symmetry, such as the space group symmetries present in bulk systems.

Since the systems in heterogeneous catalysis contain the complexity of both of these chemical domains, high-throughput techniques remain difficult to automate effectively for catalyst design. In this work, we propose a methodology which combines space group and bond symmetry techniques general enough to enumerate a large number of possible catalytic systems. This includes surfaces of any miller index produced from a provided bulk system, unique adsorption site identification, and creation of 3D structures bonded by one or multiple sites. This technique also produces a graph structure which provides the basis for all machine-learning fingerprints which require connectivity information; ideal for accurate descriptors which are not dependent on information gathered by expensive techniques such as Density Functional Theory.

We demonstrate these techniques for the enumeration of unique adsorption configurations across a wide range of alloy systems and catalytically relevant adsorbates. Utilizing these enumerated structures we show how the graph of each can be easily utilized to construct a variety of fingerprints as a basis for machine-learning accelerated predictions of adsorption energies. Finally, we discuss our Python implementation of these algorithms in the open source software package known as CatKit, intended to make high-throughput screening techniques easier to implement for the community.