Enabling accurate modelling of materials for a solid electrolyte interphase in lithium-ion batteries using effective...

Area of Science:

• Materials Science
• Computational Chemistry
• Battery Technology

Background:

• Accurate modeling of the solid electrolyte interphase (SEI) in lithium-ion batteries is crucial but challenging due to its complex structure.
• Traditional atomistic simulations require accurate interatomic potentials, which are difficult to evaluate for mixed-material SEI systems.

Purpose of the Study:

• To demonstrate the effectiveness of machine learning interatomic potentials (MLIPs) for modeling SEI structural and dynamic properties.
• To develop a scalable computational workflow for SEI analysis that overcomes limitations of conventional density functional theory (DFT) methods.

Main Methods:

• Utilized moment tensor potentials (MTPs) trained on amorphous structures and density functional theory (DFT) calculations.
• Employed active learning loops for efficient sampling of molecular dynamics (MD) trajectories.
• Validated MLIP models against experimental and theoretical data for SEI-relevant materials like Li2CO3 and Li2EDC.

Main Results:

• Trained MTP models accurately predicted structural properties (lattice parameters, elastic constants, phonon spectra) of SEI materials.
• Dynamical properties and energy barriers were accurately captured, showing finite temperature effects.
• Identified dominant lithium diffusion mechanisms (vacancy, interstitial, Frenkel pair) in Li2CO3, consistent with DFT.
• Demonstrated that MLIP training datasets can improve graph neural network (GNN) potentials.

Conclusions:

• Developed a scalable machine learning workflow for SEI modeling, enabling larger time and length scale simulations.
• The MLIP approach provides a reliable and efficient method for understanding SEI behavior in lithium-ion batteries.
• This work facilitates advanced battery material design and performance optimization.