An Accurate Charge-Aware Machine-Learning Interatomic Potential for the Reduction of Li-Ion Battery Electrolytes in...

Area of Science:

• Computational chemistry
• Materials science
• Electrochemistry

Background:

• Machine learning interatomic potentials (MLIPs) offer ab initio accuracy for complex system simulations.
• Solid electrolyte interphase (SEI) formation in Li-ion batteries (LIBs) is crucial but poorly understood.
• Classical force fields struggle with the bonding and electron transfer complexities in electrochemical processes.

Purpose of the Study:

• To develop and validate MLIPs for accurate simulation of electrochemical processes in LIB electrolytes.
• To address the challenge of training MLIPs for systems with multiple oxidation states, like those in batteries.
• To investigate electrolyte reduction mechanisms and electron transfer during the initial charge cycle.

Main Methods:

• Utilized the MPNICE MLIP architecture, featuring message passing and iterative charge equilibration.
• Trained models on reduced and unreduced potential energy surfaces for LIB-relevant electrolytes.
• Developed strategies for sampling and training on off-center radicals (OCRs) and addressed limitations of global charge equilibration (Qeq).

Main Results:

• Achieved high accuracy (within 1 kcal/mol) in training MLIPs for electrolyte systems.
• Successfully trained models capable of simulating systems with different oxidation states.
• Demonstrated effective methods for handling anion radicals and mitigating charge delocalization issues.

Conclusions:

• MPNICE enables accurate atomistic simulations of electrochemical processes relevant to LIB SEI formation.
• The study provides new insights into electrolyte reduction and realistic simulation of condensed-phase electron transfer.
• This approach advances the capability of MLIPs for complex battery chemistry simulations.