For several decades, atomic simulations of lithium-ion battery electrolytes have been conducted to elucidate lithium-ion solvation structures and the fundamental ion conduction mechanisms. This research aims to deepen the understanding of atomic-scale...
For several decades, atomic simulations of lithium-ion battery electrolytes have been conducted to elucidate lithium-ion solvation structures and the fundamental ion conduction mechanisms. This research aims to deepen the understanding of atomic-scale dynamics within these battery electrolytes. Especially, molecular dynamics (MD) simulations employing polarizable force fields have demonstrated enhanced accuracy for highly charged electrolyte systems when compared to conventional non-polarizable force fields. However, traditional methods of analyzing trajectories typically rely on time-averaging atomic properties, which can suppress or diminish the visibility of local structural variations and subtle dynamics.
In this dissertation, a detailed investigation of the lithium-ion solvation structure and dynamics in a variety of (perfluoroalkylsulfonyl)imide-based ionic liquid electrolytes is presented, utilizing a deep learning-assisted analysis of MD simulations performed with a polarizable force field. A deep learning methodology known as Graph Dynamical Networks (GDyNets) is employed to categorize local solvation structures into distinct states and to derive key state dynamics as a Markov state model.
Initially, polarizable molecular dynamics simulations of the lithium bis(trifluoromethanesulfonyl)imide / N-butyl-N-methylpyrrolidinium bis(trifluoromethanesulfonyl)imide (LiTFSI/PYR14TFSI) ionic liquid electrolyte are analyzed and learned by a GDyNet model. Lithium atoms in the simulation trajectories are designated as target atoms and categorized into three distinct states by the GDyNet model. Properties specific to each state are defined by weighting the state probabilities of the target atoms based on radial, spatial, and cluster distribution functions. The identified states include a Li+ ion coordinated by two and three TFSI− ions, and a Li cluster composed of multiple Li+ and TFSI− ions. Large clusters containing three or more Li ions are also captured within the Li cluster state. Eigenvalue decomposition of the Markov state model reveals two significant transitions among these states. The fastest mechanism of lithium-ion transport is identified through the mean-squared displacements of the states and transitions. The influence of the lithium to ionic liquid (Li:IL) ratio is examined by applying the GDyNet model to systems with varying Li:IL ratios. Furthermore, an additional investigation using a 5-state GDyNet model uncovers three distinct Li cluster states.
Secondly, to explore the impact of anion length and symmetry, a broader range of ionic liquid electrolytes is investigated using the same methodology. LiX/PYR14X electrolytes featuring bis(pentafluoroethanesulfonyl)imide (X = BETI) and (nonafluorobutanesulfonyl)(trifluoromethanesulfonyl)imide (X = IM14) anions are studied. BETI- is characterized by longer perfluoroalkyl chains compared to TFSI-, while IM14- possesses laterally asymmetric perfluoroalkyl chains. Using the deep learning-assisted analysis, four major Li+ states are identified in both BETI and IM14 systems: [LiX3]2−, [LiX2]−, and two Li cluster states differentiated by short and long Li-Li distances. The population of [LiX3]2− decreases (TFSI > IM14 > BETI) while [LiX2]− increases (TFSI < IM14 < BETI) due to steric hindrance among perfluoroalkyl chains. Anions with at least one short perfluoroalkyl chain (TFSI-, IM14-) more readily form Li-O-Li interactions and larger Li-anion clusters than BETI. Transition timescales generally increase with anion mass/size (TFSI < BETI < IM14), and structural motion is faster than vehicular motion. At higher Li concentrations, the interplay between anion size and asymmetry becomes evident, influencing transport properties differently for BETI and IM14 electrolytes.
This work integrates state-of-the-art atomistic simulation with deep learning techniques to unveil subdivided microscopic structures and their transition dynamics across diverse combinations of battery electrolytes. This approach thus facilitates the discovery of novel and specialized battery materials and enables in-depth analysis of complex molecular modeling results. The ability to interpret the intricate relationships between electrolyte composition, microscopic structure, and dynamic behavior at the atomic scale provides a powerful tool for the rational design of next-generation energy storage systems.