AI Simulates Battery Chemistry To Unlock Faster-Charging Lithium Metal Power

Scientists are increasingly focused on understanding the complex processes governing solid-electrolyte interphase (SEI) formation within lithium metal batteries, a critical factor limiting performance and longevity. Syed Mustafa Shah, Mohammed Lemaalem, and Anh T. Ngo, from the University of Illinois Chicago and Argonne National Laboratory et al., present a novel Deep Potential-based molecular dynamics framework to accurately model early-stage SEI nucleation at lithium metal interfaces. This research is significant because it provides detailed insights into the behaviour of concentrated LiTFSI and LiPF6 electrolytes, revealing how electrolyte composition directly influences SEI characteristics such as thickness, growth kinetics, and chemical composition, ultimately correlating with observed cycling stability and rate capability.

Deep potential modelling elucidates initial solid-electrolyte interphase development in lithium metal batteries

Scientists have developed a new molecular dynamics framework, leveraging deep potential machine learning, to accurately model the complex early stages of solid-electrolyte interphase formation in lithium metal batteries. This breakthrough addresses a critical challenge in battery research, experimentally resolving the dynamic processes occurring at the lithium metal interface during battery operation.
The work focuses on understanding how different electrolyte compositions influence the formation of the SEI, a crucial layer determining battery performance and longevity. Researchers trained the Deep Potential model on extensive ab initio datasets, achieving a level of accuracy previously unattainable with classical molecular dynamics simulations.

This innovative approach allows for detailed observation of the initial nucleation of the SEI, revealing the elementary reduction events and bond rearrangements that govern its growth. Validation against experimental data confirms the model’s ability to predict both macroscopic transport properties, such as ionic conductivity and viscosity, and the subtle structural details of electrolyte solvation.

Specifically, simulations demonstrate that 3.5 M LiTFSI/DMC induces spontaneous, thermally activated reduction reactions, leading to the rapid growth of a thick, anion-derived SEI enriched in oxygen and fluorine-containing species. Conversely, electrolytes containing 1.5-2.5 M LiTFSI/DMC and 1 M LiPF6/EMC/DMC/EC result in thinner, LiF-dominated interphases that develop at a slower rate.

These findings establish a fundamental connection between electrolyte concentration and SEI architecture, highlighting a trade-off between growth rate and composition. The research establishes the atomistic origins of initial SEI architecture, providing a predictive tool for designing electrolytes that stabilize the lithium metal anode and enhance battery performance. The potential’s accuracy was confirmed by low root mean square error (RMSE) for both force and energy, alongside high coefficients of determination (R2 ≈0.96, 0.99), and excellent agreement between atomic-level molecular dynamics (AIMD) and MLMD radial distribution functions.

Simulation protocols for classical molecular dynamics (CMD) and methodologies for calculating structural, transport, and interfacial properties are detailed in supplementary sections. Radial distribution functions, gαβ(r), were computed for both LiTFSI and LiPF6 electrolytes, revealing broadly similar solvation structures between MLMD and CMD, but with notable quantitative differences in the first coordination shell.

For 1.5 M LiTFSI, MLMD predicted a first Li, O peak at 1.95 Å, while CMD overestimated the separation at 2.05 Å, a trend persisting with increasing concentration to 2.2 Å and 2.0 Å for MLMD versus 2.25, 2.30 Å for CMD at 3.5 M. These findings suggest that classical force fields systematically overestimate Li, O separations, underestimating the tighter coordination captured by MLMD.

Conversely, Li, F and Li, N correlations were well reproduced by CMD, demonstrating reasonable description of secondary, longer-range interactions. Cumulative coordination numbers, Nαβ(r), further quantified these differences, with MLMD showing a steeper rise and higher values in the first solvation shell for LiTFSI, indicating denser local packing.

Ionic conductivity, ion self-diffusion coefficients, and viscosity were calculated, demonstrating that MLMD markedly improves agreement with experimental data compared to CMD across all electrolyte systems. The 1 M LiPF6 electrolyte exhibited the highest diffusivities, while LiTFSI/DMC electrolytes displayed progressively slower ion motion with increasing concentration from 1.5 to 3.5 M, reflecting a transition to a crowded ionic environment. The study revealed that 3.5 M LiTFSI/DMC induces spontaneous, thermally activated reduction reactions, resulting in rapidly growing, thick anion-derived SEIs enriched in oxygen and fluorine-containing species.

Conversely, electrolytes comprising 1.5, 2.5 M LiTFSI/DMC and 1 M LiPF6/EMC/DMC/EC formed thinner, LiF-dominated interphases exhibiting slower growth kinetics. The modelling results align with experimental observations demonstrating that 3.5 M LiTFSI enhances cycling stability and rate capability, while lower concentrations lead to weaker passivation.

Deep Potential models were constructed for LiTFSI in DMC and LiPF6 in EC/EMC/DMC, trained on extensive ab initio molecular dynamics datasets and validated against experimental benchmarks. The framework outperforms classical molecular dynamics in predicting transport properties and uniquely enables atomistic resolution of SEI nucleation.

Radial distribution functions for LiTFSI and LiPF6 electrolytes showed that MLMD and CMD yield broadly similar solvation structures, but quantitative differences emerge in the first coordination shell. For 1.5 M LiTFSI, the MLMD predicted first-shell Li, O peak at 1.95 Å, while CMD overestimated the separation at 2.05 Å.

At 2.5 M LiTFSI, MLMD predicted a Li, O peak at 2.2 Å, and at 3.5 M LiTFSI, the peak was at 2.0 Å, whereas CMD consistently yielded longer distances in the 2.25, 2.30 Å range. These findings suggest that the classical force field systematically overextends Li, O separations, underestimating the tighter coordination captured by MLMD.

Electrolyte composition dictates solid-electrolyte interphase morphology and lithium metal anode stability

Researchers have developed a molecular dynamics framework to model the formation of the solid-electrolyte interphase in lithium metal batteries with improved accuracy. This framework, trained using data from detailed quantum mechanical calculations and validated against experimental measurements, allows for the observation of early-stage SEI nucleation at lithium metal interfaces.

The simulations reveal that electrolyte composition significantly influences SEI formation, with concentrated LiTFSI/DMC inducing rapid growth of thick, anion-derived SEI layers. Conversely, lower concentrations of LiTFSI/DMC and LiPF6/EMC/DMC/EC result in thinner, LiF-rich interphases that develop more slowly.

These findings align with experimental results demonstrating that 3.5 M LiTFSI enhances both cycling stability and rate capability, while lower concentrations lead to weaker passivation of the lithium metal anode. The modelling approach efficiently captures electrolyte transport and the mechanisms of early-stage SEI formation, demonstrating the importance of considering the interplay between bulk solvation and interfacial reactivity when predicting electrolyte performance.

The timescale of the simulations, approximately one nanosecond, captures the initial nucleation phase, although long-term SEI evolution remains beyond its scope. Furthermore, the current study acknowledges that future work should decouple the effects of salt and solvent variations to provide a more detailed understanding of their individual contributions. This computational method offers a pathway for discovering electrolytes that inherently stabilise the lithium metal anode, accelerating the development of high-performance batteries.

👉 More information
🗞 Predictive Machine Learning Molecular Dynamics of SEI Formation in Concentrated LiTFSI and LiPF6 Electrolytes for Lithium Metal Batteries
🧠 ArXiv: https://arxiv.org/abs/2602.05141