Understanding the mass transport in electrolyte solutions is critical for the development of next generation battery systems. Computational methods such as molecular dynamics (MD) simulations have been used to study the transport of Li+ ions during charge and discharge cycles in lithium-ion batteries.^[1] However, in real battery systems the transport of ions is affected by the formation of a solid electrolyte interphase (SEI). But, in classical MD simulations the formation of a SEI cannot be described. The reason for this is that the potential energy surface (PES) of a system is modeled with fixed functional forms known as classical force fields (FFs). Based on their fixed form, classical FFs fail to model chemical reactions necessary to capture SEI formation. Machine Learning Interatomic Potentials (MLIPs) are an alternative with greater flexibility. MLIPs model the PES of a system based on Gaussian processes or artificial neural networks. Based on the enhanced flexibility, MLIPs can account for effects abundant in classical FFs such as chemical reactions.^[2] However, since there is no free lunch, the gain in flexibility and accuracy comes with an increase in computational cost. This limits the scalability of MD simulations with MLIPs to systems beyond thousands of atoms and simulation times beyond hundreds of picoseconds.^[3] This prohibits the accurate computation of the long-time mass transport in electrolyte solution necessary to compare computational with experimental results.^[4] With classical FFs on the other side one can easily reach system sizes of millions of atoms and simulation times up to microseconds. To compensate the shortcoming of either approach and to accurately model mass transport in electrolyte solutions, we propose to couple a classical FF with a MLIP in a sequential manner. The goal is to reach simulation times typically for classical MD simulation to capture the long-time behavior and keeping the accuracy and capabilities of MLIPs by passing information from one model to another.

[1] Muralidharan, A. et al., Scientific Reports 8, 10736 (2018)

[3] Mondal, A. et al., Journal of Chemical Theory and Computation 8, 19, 14, 4584–4595 (2023)

[2] Friedrich, P. et al., Nature Materials 20, 750–761 (2021)

[4] Yao, N. et al., Chemical Reviews 122, 10970−11021 (2022)