Utilizing machine learning and phonon density of states for innovative approaches to design and optimize high-performance solid-state Mg-ion electrolytes

Abstract

This study presents the development of machine learning (ML) model for the enhancement of ionic conductivity in solid-state electrolytes for Mg-ion batteries (MIBs), utilizing phonon density of states (PhDOS) as tool for predicting ionic conductivity. Data comprising compounds with PhDOS calculations are sourced from the Materials Project (MP) and processed to extract frequencies and PhDOS values, enabling the calculation of the total phonon band center, a practical proxy for ionic conductivity. Learning process involving three ML algorithms (Extra Random Trees (XT), Gradient Boosting (GB), Extreme Gradient Boosting (XGB)) is conducted using these band centers. Cross-validation results indicated the superior accuracy of XT-model where its performance, is validated through density functional theory (DFT)-based phonon calculations on Mg2Ge, Mg2Gd, MgSe2 and Mg3P2. The XT-model was further employed to predict total phonon band centers of ∼9 K Mg compounds obtained from MP with no previous PhDOS calculations containing Mg–O, Mg–S and Mg–Se compounds which show that the band center median of Mg–Se materials is the lowest of the systems. Additionally, Nudged Elastic Band (NEB) and Ab initio Molecular Dynamics (AIMD) calculations is used to calculate the migration energy of MgSe2 which shows potential as an Mg-ion solid electrolyte when compared to other Mg electrolytes.