Apr 8, 2025
10:30am - 11:00am
Summit, Level 3, Room 340
Matthew Witman1,Lauren Way1,Catalin Spataru1,Anuj Goyal2,Stephan Lany3,Andrew Rowberg4,Anthony McDaniel1,Tyra Douglas1,Sean Bishop1
Sandia National Laboratories1,Indian Institute of Technology Hyderabad2,National Renewable Energy Laboratory3,Lawrence Livermore National Laboratory4
Matthew Witman1,Lauren Way1,Catalin Spataru1,Anuj Goyal2,Stephan Lany3,Andrew Rowberg4,Anthony McDaniel1,Tyra Douglas1,Sean Bishop1
Sandia National Laboratories1,Indian Institute of Technology Hyderabad2,National Renewable Energy Laboratory3,Lawrence Livermore National Laboratory4
The formation and migration of crystallographic defects dictate material properties and performance for a plethora of technological applications. Density functional theory (DFT)-based calculations are a powerful computational technique for predicting defect formation energy and migration energy barriers, yet they become prohibitively expensive for the high-throughput screening necessitated by large-scale materials discovery campaigns. Without introducing hand-crafted (i.e., chemistry- or structure-specific) descriptors, we propose a generalized deep learning approach to train graph neural network surrogate models, using only the host structure as input, to predict both vacancy formation and migration energy. With sufficient training data, computationally efficient and simultaneous inference of vacancy defect thermodynamics
and kinetics can be obtained to compute temperature-dependent diffusivities and to down-select candidates for more thorough DFT analysis or experiments. Thus, as we specifically demonstrate for potential water-splitting materials, candidates with desired defect thermodynamics, kinetics, and host stability properties can be more rapidly targeted from open-source databases of known materials. For top predicted candidates, we perform more precise DFT calculations to confirm machine learning predictions and, finally, experimental validation studies on their water-splitting behavior.