CBE Doctoral Dissertation Defense: “Identifying Material Fingerprints of Relevance to Understand Adsorbate-Surface Interactions Using First Principles Modeling and Machine Learning” (Genesis Quiles-Galarza)

Adsorption of chemical species on surfaces of materials is one of the critical phenomenon governing the reactivity and activity of the material for surface and interface driven chemical reactions. At the core of the analytical $d$-band adsorption model is the correlation between the adsorption energy of a chemical species (molecule or reaction intermediate) on the metal surface and an electronic material property, namely the d-band center from the density of states (DOS). Although very successful, the d-band model has its limitations and cannot be applied to all materials. Therefore, efforts have been devoted to discover material fingerprints that can be used to describe adsorption of chemical species on more complex surfaces and materials. Herein, we use first principles methods (density functional theory, DFT) and machine learning (ML) to elucidate what kind of material fingerprints or features are needed to describe the interaction between an adsorbate and the surface of a metal material, a two-dimensional (2D) transition metal carbide and nitride compound materials known as MXenes, and a hybrid (molecular catalyst supported on a heterogeneous surface) catalyst material. The ML models used in this study belong either to the “black-box” or “glass-box” category, enabling not only prediction of the adsorption energy with small errors, but also allow insight into the material physics governing the adsorbate-surface interaction. These ML studies indisputably demonstrate that the electronic fingerprints of the material are the most critical features in reliably determining the adsorbate-surface interactions.For metals, we confirm the findings of the analytical d-band model by achieving adsorption formulas with contributions from both the sp and d-DOS bands, as well as multiple higher order contributions. For MXenes, we find that the adsorbate-surface interaction is complex with significant contributions from the terminating functional group atom, specifically their sp-DOS band features. Generally, our studies shows that the nature of the adsorbate-surface interactions cannot be fully captured by single or simple linear correlations between the adsorbate energy and a materials feature, but instead require higher order, multi-dimensional feature combinations. These findings imply that further investigations are needed to develop physically-sound, multi-dimensional features which could be used as descriptors to predict adsorbate-surface interactions with an accuracy comparable to that of DFT.