Abstract
Lithium metal batteries offer superior volumetric and gravimetric specific capacities compared to those based on traditional graphite anodes. Although advancements in solid-state electrolytes address safety concerns, challenges remain, particularly regarding interphase formation in lithium metal anodes. This work presents a computational framework based on high-throughput first-principles density functional theory and machine-learning interatomic potentials (MLIPs) including automated iterative, active learning to enable robust computational exploration of interphase formation between lithium metal anodes and an inorganic solid-state electrolyte. As a demonstration, we apply the framework to a Li/Li(7)P(3)S(11) interface and find that it accurately identifies the experimentally observed, thermodynamically stable interphase products as well as their overall spatial arrangement within a heterogeneous, amorphous layered structure, with Li(2)S domains of nanocrystallinity. Our simulations show two stages, a fast and slow diffusion reaction regime, that corroborate the relative phase formation rate of Li (x) P, Li(2)S, and Li(3)P. Using the Onsager transport theory, we capture time-dependent ionic diffusion within the reacting interface, including cross-correlation effects. We found that cross-correlation effects between Li-P and P-S ionic motion significantly influence P-ion diffusion, making it highly sensitive to the local environment and potentially leading to "kinetic trapping" of Li-P phases. The passivation of the interface is shown as the ionic fluxes all approach zero, effectively halting interphase growth.