Elucidating Lithium-Ion Dynamics in Highly Concentrated Electrolytes with Markov State Models

Abstract

Abstract Highly concentrated electrolytes (HCEs) have attracted considerable attention for lithium batteries owing to their wide electrochemical stability windows and favorable interfacial properties, which originate from their anion-rich solvation structures. The dynamics of lithium ions play a key role for HCEs performance; however, a clear mechanistic understanding of lithium-ion (Li+) transport within percolating ionic networks of HCEs remains elusive. Here, we performed a molecular dynamics simulation on a lithium bis(fluorosulfonyl)imide/1,2-dimethoxyethane (LiFSI/DME) HCE system and then constructed a Markov state model (MSM) using a graph convolutional neural network architecture named GDyNet to investigate the dynamics of lithium ions. The MSM classifies lithium ions into three distinct metastable states based on their local coordination environments: nonsolvated state (State 0), moderately solvated state (State 1), and highly solvated state (State 2), which are kinetically well separated and exhibit distinct dynamic behaviors. Li+ transport in LiFSI/DME HCE is dominated by the self-transitions within each state rather than interstate transitions. During the slow interconversion among the three metastable states, FSI– anions in the coordination shells undergo frequent exchange. The moderately solvated state exhibits the fastest diffusion, followed by the highly solvated state, and the nonsolvated state exhibits the slowest. By integrating the Green–Kubo relation and MSMs, we quantified the relative contributions of these state transitions to Li+ conduction. The transitions from the nonsolvated state contribute negligibly to Li+ conduction, and the dominant contributions originate from the moderately and highly solvated states. The insights obtained from the MSM provide an in-depth mechanistic understanding of Li+ dynamics in HCEs.