Bridging Atomistic and Mesoscale Lithium Transport via Machine-Learned Force Fields and Markov State Models

Area of Science:

• Computational materials science
• Solid-state chemistry
• Energy storage

Background:

• Lithium diffusion in silicon anodes is critical for battery performance but challenging to simulate due to high energy barriers.
• Atomistic simulations like ab initio molecular dynamics (AIMD) are limited by timescale, hindering the capture of rare diffusion events.

Purpose of the Study:

• To establish a multiscale computational workflow for accurate and efficient modeling of lithium transport in silicon-based battery anodes.
• To bridge atomistic mechanisms with mesoscale diffusion phenomena using advanced simulation techniques.

Main Methods:

• Developed machine-learned force fields (MLFFs) trained on AIMD data for near-DFT accuracy.
• Employed MLFFs to perform large-scale molecular dynamics simulations up to tens of nanoseconds.
• Constructed Markov state models (MSMs) from simulation trajectories to analyze lithium-jump statistics and predict diffusion coefficients.

Main Results:

• MLFFs enabled simulations significantly longer than AIMD, capturing essential diffusion events.
• MSMs quantitatively reproduced lithium diffusivities in crystalline Li-Si phases with high accuracy.
• Achieved uncertainties in diffusion coefficients an order of magnitude smaller than traditional AIMD simulations.

Conclusions:

• The AIMD → MLFF → MSM workflow provides a robust and transferable method for quantitative transport modeling in battery materials.
• This multiscale approach is applicable to amorphous structures, defect-mediated diffusion, and other solid-state anode materials.
• The methodology paves the way for accelerated discovery and optimization of advanced battery anode technologies.