Machine learning-aided first principles predictions of redox potentials
A study conducted by Ryosuke Jinnouchi et al., in collaboration with Universität Wien and VASP Software GmbH, was published in the npj Computational Materials.
First-principles calculations, based on the physical principle of the wave equation, are attempts to predict a wide range of material properties solely through computer simulations. One property that has been challenging to calculate using this method is the redox potential. The redox potential is a fundamental electrochemical property that predicts how and at what potential a material might change. It is crucial when selecting materials for electrochemical devices such as batteries, fuel cells, electrolysis, and artificial photosynthesis. However, first-principles prediction of redox potential has been difficult, requiring computational resources equivalent to 1 million core hours (114 years of computation on a single-core processor). This study devised a method to accurately obtain first-principles calculation results in 1/50th of the conventional time by progressively enhancing the accuracy from rough approximation models to high-precision models with the aid of machine learning. This approach is expected to be the foundation of future first-principles electrochemistry.
Title: Machine Learning-aided First-principles Calculations of Redox Potentials
Authors: Jinnouchi, R., Karsai, F., Kresse, G.
Journal Name: npj Computational Materials
Published: May 20, 2024
https://doi.org/10.1038/s41524-024-01295-6