Electronic structure and reactive molecular dynamics simulations for modelling solid-state batteries

Abstract

Solid-state batteries exhibit superior theoretical energy density, significantly enhanced safety characteristics, and prolonged cycling stability, establishing them as a pivotal focus in next-generation energy materials research. Compared to conventional liquid electrolyte batteries, solid-state batteries eliminate safety hazards associated with flammable liquid electrolytes while promising higher energy densities and extended operational lifespans—attributes of paramount importance for meeting the future demands of electric vehicles and renewable energy storage applications. Nevertheless, their commercialization trajectory remains impeded by multiple challenges, including interface stability, ionic transport mechanisms, and materials compatibility issues. In solid-state battery research, computational methodologies have emerged as powerful instruments for elucidating fundamental mechanisms and accelerating technological advancement. Electronic structure calculations provide precise insights into band structures and interfacial properties of materials, while molecular dynamics simulations capture atomic-scale dynamic processes. This thesis presents innovative computational modelling techniques focused on these two critical domains, offering novel perspectives for solid-state battery investigation.

In the realm of electronic structure methods, this research establishes an innovative density-functional tight-binding parameterization scheme for accurately modelling band alignment at solid-state battery interfaces. Through the integration of genetic algorithms with explicit band alignment constraints, this approach successfully reproduces the electronic structure of a typical all solid-state battery system (Li-Li₂PO₂N-LiCoO₂) in accordance with density functional theory calculations. The developed parameter set demonstrates excellent transferability across diverse lithium-containing materials and various battery charging states, providing a computational tool that balances efficiency and accuracy for interface design.

In the domain of molecular dynamics simulations, this thesis develops a universal machine learning potential framework that incorporates equivariant graph neural networks with polarizable long-range interactions. This methodological innovation overcomes a critical limitation of existing machine learning potentials by accurately capturing electrostatic and dispersion effects beyond the typical cutoff radius. The pretrained universal model, encompassing elements up to plutonium, exhibits exceptional performance in predicting various materials properties, including bulk modulus, ionic diffusivity, and phase transitions. This framework enables large-scale reactive molecular dynamics simulations of solid-state battery interfaces, revealing unprecedented insights into the formation mechanisms of the solid-electrolyte interphase. Simulations of Li₃PS₄/Li and Li₆PS₅Cl/Li interfaces elucidate the atomic-level processes involved in interface stabilization, including the crystallization of Li₂S and the formation of amorphous Li₃P/Li₂S regions, findings that align closely with experimental observations. These methodological developments provide computational tools for investigating critical phenomena in solid-state batteries across relevant length and time scales, contributing to the theoretical design of improved energy storage technologies.