Abstract
Despite recent advances of data acquisition and algorithms development, machine learning (ML) faces tremendous challenges to being adopted in practical catalyst design, largely due to its limited generalizability and poor explainability. Herein, we develop a theory-infused neural network (TinNet) approach that integrates deep learning algorithms with the well-established d-band theory of chemisorption for reactivity prediction of transition-metal surfaces. With simple adsorbates (e.g., *OH, *O, and *N) at active site ensembles as representative descriptor species, we demonstrate that the TinNet is on par with purely data-driven ML methods in prediction performance while being inherently interpretable. Incorporation of scientific knowledge of physical interactions into learning from data sheds further light on the nature of chemical bonding and opens up new avenues for ML discovery of novel motifs with desired catalytic properties.
Highlights
Despite recent advances of data acquisition and algorithms development, machine learning (ML) faces tremendous challenges to being adopted in practical catalyst design, largely due to its limited generalizability and poor explainability
We present a theory-infused neural network (TinNet) approach to predicting chemical reactivity of transition-metal surfaces and, more importantly, to extracting physical insights into the nature of chemical bonding that can be translated into catalyst design strategies
By learning from ab initio adsorption properties with deep learning algorithms, e.g., convolutional neural networks, while respecting the well-established d-band theory of chemisorption in architecture design, the TinNet can be applied for a broad range of d-block metal sites and naturally encodes physical aspects of bonding interactions, inheriting the merits of both worlds
Summary
Despite recent advances of data acquisition and algorithms development, machine learning (ML) faces tremendous challenges to being adopted in practical catalyst design, largely due to its limited generalizability and poor explainability. Rapid discovery of structural motifs with kinetics-favorable descriptor values, for example using quantum-chemical calculations, is appealing while remaining a daunting task due to the formidable computational cost in accurately solving the many-electron Schrödinger equation In this aspect, the d-band theory of chemisorption pioneered by Hammer and Nørskov[2,3,4,5,6] has been widely used for understanding reactivity trends of d-block metals[7,8] and, to some extent, their compounds[9]. By learning from ab initio adsorption properties with deep learning algorithms, e.g., convolutional neural networks, while respecting the well-established d-band theory of chemisorption in architecture design, the TinNet can be applied for a broad range of d-block metal sites and naturally encodes physical aspects of bonding interactions, inheriting the merits of both worlds. The TinNet achieves prediction performance on par with purely regression-based ML methods, especially for out-of-sample systems with unseen structural and electronic features and enables physical interpretation, paving the path toward ML discovery of novel motifs with desired catalytic properties
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
