Absolute standard hydrogen electrode potential and redox potentials of atoms and molecules: machine learning aided first principles calculations

TL;DR

This paper develops a machine learning-enhanced first-principles approach to accurately predict redox potentials and the absolute standard hydrogen electrode potential, achieving high accuracy across various molecules and ions.

Contribution

It introduces a hybrid functional with 25% exact exchange combined with machine learning techniques for precise electrochemical property predictions.

Findings

Achieves an average error of 80 mV in redox potential predictions.

Successfully predicts potentials for seven diverse redox couples.

Demonstrates the effectiveness of machine learning in first-principles electrochemistry calculations.

Abstract

Constructing a self-consistent first-principles framework that accurately predicts the properties of electron transfer reactions through finite-temperature molecular dynamics simulations is a dream of theoretical electrochemists and physical chemists. Yet, predicting even the absolute standard hydrogen electrode potential, the most fundamental reference for electrode potentials, proves to be extremely challenging. Here, we show that a hybrid functional incorporating 25 % exact exchange enables quantitative predictions when statistically accurate phase-space sampling is achieved via thermodynamic integrations and thermodynamic perturbation theory calculations, utilizing machine-learned force fields and $\Delta$-machine learning models. The application to seven redox couples, including molecules and transition metal ions, demonstrates that the hybrid functional can predict redox potentials across a wide range of potentials with an average error of 80 mV.