Machine learning enabled screening of heterogeneous catalysts based on electronic structure calculations has the potential to significantly accelerate the discovery of novel heterogeneous catalysts for energy applications such as Fischer-Tropsch synthesis, methanation process, etc. We present a Δ-learning algorithm for high throughput screening of bimetallic alloy catalysts for Fischer-Tropsch synthesis process on closed packed 111 facets of 37 pure metals, 1232 L12 alloys, and 101 facets of 666 L10 alloys. We carry out high throughput electronic structure calculations of adsorption of five important atomic species (i.e., C, N, O, H, and S) on all the alloy surfaces which comprises of more than 65, 000 BEEF-vdw calculations. We uploaded this dataset on www.Catalysis-Hub.org to make it publicly accessible for researchers all across the world for high throughput catalysis research. Next, using a novel machine learning framework based on scaling relations and connectivity matrix-based fingerprints, we developed a machine learning model to predict the adsorption energies of multi-atomic adsorbates, e.g., CHx, OHx, SHx etc. We show that this hybrid model provides a significant improvement in the prediction of residual for adsorption energy in comparison to the connectivity-based fingerprints or scaling relations alone. In addition, we used the regression model predictions and uncertainties in our model to guide the acquisition of new data points to improve its predictability. As a result, we can exhaustively search the potentially relevant space for a given reaction system with minimal number of expensive quantum chemical calculations.

Utilizing machine learning for high throughput screening of bimetallic alloy catalysts for Fischer-Tropsch synthesis

2019 Spring National Meeting

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.