(598c) DFT Calculations and Machine Learning Approach to Predict Catalytic Properties of Nanoscale Electrocatalysts in Solution for Clean Fuel Generation

Innovative design of functional nanomaterials has been prolonged challenge to scientific community and industrial sector, especially regarding with structural stability on the thermodynamic and electrochemical aspects. It is due to the key functionality of the materials in wide range of applications, such as Li-air batteries, Photovoltaic devices, and fuel cells. Advanced experimental tools and computational theories attempted to deploy various candidates, but their performances in operations did not meet the practical level requirements.

For example, one of the major huddles toward to the solution is how to accurately and quickly explore the gigantic material space with varying the structure and composition of catalysts to identify promising catalysts and propose design principles for the performance optimization. Over the last several decades, first-principles calculations have been considerably developed for especially the objective through an efficient analysis of large-scale database. It is noteworthy that the approach enabled an identification of a key descriptor on the catalytic activity and the high-throughput screening of the best candidate beyond the conventional Pt. Unfortunately, most of the candidates were not applied to the real devices, due to the serious structural degradation issue via chemical or/and electrochemical reaction paths. It means that just electronic structure-level analysis for the catalytic activity may not be enough for designing high functional catalysts. Probably, the systematic and consistent incorporation of hybrid methodologies should be the paradigm. For example, establishment of big data, material informatics based on neural network or machine learning technique, and multi-scale simulations such as molecular dynamics (MD) or Monte Carlo (MC) simulations can be the relevant procedure.

This presentation introduces machine learning driven computational framework for high throughput screen of electrocatalysts for hydrogen fuel generation.