Accelerated High-Throughput Screening for Solid-State Multivalent Conductors Using Machine Learning Interatomic Potentials

Multivalent-ion batteries present considerable advantages over lithium-ion technologies, including lower costs and potentially higher energy densities. However, realizing efficient multivalent-ion conduction in solid-state materials remains a significant challenge, primarily due to the high charge density of multivalent ions, which restricts ion mobility compared to their monovalent counterparts. Additionally, high-throughput screening of multivalent conductors is impeded by the high migration barriers of these ions, leading to increased computational costs.
In this study, we leverage universal Machine Learning Interatomic Potentials¹ (uMLIP) in conjunction with the Approximate Nudged Elastic Band² (ApproxNEB) method to systematically construct migration graphs and predict percolation barriers, thereby reducing the computational cost of high-throughput screening. Our initial dataset consists of 1,316 and 1,204 candidate materials that already contain Ca and Mg ions, respectively, alongside 4,781 materials without mobile species and therefore unknown interstitial site locations. Key material properties, including composition, band gaps, thermodynamic stability, ion migration barriers, dopability, electrochemical stability, and diffusivity, were evaluated to identify promising candidates for solid-state multivalent conductors.
Our results show that uMLIP underestimates barriers relative to Density Functional Theory (DFT) but remains effective in identifying high-barrier materials with a low rate of false negatives. Furthermore, uMLIP calculations reveal a significant dependence of Mg-ion migration on the local coordination environment, whereas Ca-ion migration exhibits comparatively less sensitivity to the local environment. These findings highlight the distinct behaviors of multivalent ions in solid-state frameworks and underscore the efficacy of machine learning-based approaches in accelerating the discovery of viable multivalent conductors.

1. I. Batatia, D. P. Kovács, G. N. C. Simm, C. Ortner, and G. Csányi, arXiv:2206.07697.
2. Z. Rong, D. Kitchaev, P. Canepa, W. Huang, and G. Ceder, J. Chem. Phys. 145, 074112 (2016).