Classical and machine learning driven molecular dynamics in battery research

초록

Molecular dynamics (MD) simulation has long served as a foundational tool for probing atomistic phenomena in chemistry, physics, and materials science. As computational power and algorithms have advanced, MD has become essential for elucidating complex mechanisms in electrochemical systems, particularly in batteries, where experimental access to interfacial dynamics, solvation structures, and transport pathways remains challenging. Conventional molecular dynamics (CMD), based on Newtonian mechanics and empirically parameterized force fields, offers a physically interpretable and computationally efficient framework but is limited by accuracy, transferability, and the treatment of polarization and long-range interactions. In contrast, machine learning driven molecular dynamics (MLMD) leverages data-driven interatomic potentials trained on quantum mechanical or experimental data to achieve near-ab initio accuracy with substantial computational acceleration. Recent developments, including neural network potentials, Gaussian approximation potentials, moment tensor potentials, and graph-based frameworks, have enabled accurate modeling of lithium, sodium, and magnesium battery materials across liquid, solid, and interfacial phases. This review surveys representative CMD and MLMD studies in battery research, highlighting simulation methodologies, force field developments, and their key findings in understanding ion transport, interfacial stability, and reaction dynamics. Finally, we discuss remaining challenges-such as long-range force treatment, transferability, data diversity, and uncertainty quantification- and outline strategic pathways for the integration of physics-based constraints, active learning, and hybrid multiscale modeling to guide the next generation of molecular simulations for battery materials discovery and design.

키워드