Physically transparent and predictive models that quantify structure-property relationships of materials are important in many fields, including catalysis. Many of the high-profile applications of machine learning in catalysis and materials science have focused on using black-box models such as neural networks and Gaussian process regressors for the fast and accurate prediction of target properties such as material stability or adsorbate binding energies. Extracting meaningful physical insights from these black-box models has proven challenging, as the internal logic of these black-box models is not readily interpretable due to their high complexity. Interpretable ML methods that merge the predictive capacity of black-box models with the physical interpretability of physics-based models offer an alternative to black-box models. Interpretable models are desirable because they can lead to theories and hypotheses that advance our scientific knowledge and accelerate progress beyond solely enabling high-throughput screening.<br/><br/>In this talk, I will show how interpretable machine learning can help understand and clarify structure-property relationships of catalysts and materials. I will first present how the Sure Independence Screening and Sparsifying Operator (SISSO) algorithm can find simple descriptors that predict the stability of perovskite oxides and halides. SISSO finds a physically meaningful descriptor that predicts the stability of perovskite oxide and halide materials with superior performance compared to the well-known Goldschmidt tolerance factor. Second, I will show how subgroup discovery can find and describe local patterns in heterogeneous materials data, giving unique insight that global models would miss. Third, I will present the use of supervised generalized additive models (GAMs) and principal component analysis (PCA) to establish relationships between the geometric structure, electronic structure, and chemisorption strength of adsorbates on alloys. By comparing the GAM-derived chemisorption models and the PCA analysis to previously established electronic-structure models, the critical physical parameters that control the chemisorption process on metal and alloy surfaces are elucidated.

Interpretable machine learning to elucidate structure-property relationships of catalysts and materials

Fall 2022 abstracts, posters, and presentations will launch in mid-September.

ACS Fall 2022

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