Prediction and Experimental Verification of Electrolyte Solvation Structure from an OMol25-Trained Interatomic...

Area of Science:

• Computational chemistry
• Materials science
• Electrochemistry

Background:

• Understanding electrolyte solvation and ion correlations is key for next-generation batteries.
• Accurate, fast, and transferable atomistic simulations are needed for electrolyte design.
• Machine learning interatomic potentials (MLIPs) offer near-DFT accuracy at significantly higher speeds.

Purpose of the Study:

• To assess MLIPs trained on materials data versus the Open Molecules 2025 (OMol25) dataset for Na-ion battery electrolyte simulations.
• To evaluate the accuracy of MLIPs in predicting nanoscale structural organization and ion-solvation characteristics.
• To analyze trends in solvation structure under varying conditions using validated MLIPs.

Main Methods:

• Trained MLIPs on large, diverse datasets, including OMol25, which specifically samples electrolyte configurations.
• Integrated computational modeling with experimental validation (densities, X-ray structure factors).
• Employed the Universal Model of Atoms (UMA-OMol) trained on OMol25 for molecular dynamics simulations.

Main Results:

• The OMol25-trained UMA-OMol model showed superior agreement with experimental densities and X-ray structure factors compared to models trained solely on inorganic materials.
• Systematic trends in solvation structure were analyzed, revealing temperature effects on solvation heterogeneity and ion pairing.
• Subtle changes in glyme solvent topology significantly impacted ion correlations and solvation structure.

Conclusions:

• OMol25-trained MLIPs provide a practical and accurate approach for high-throughput electrolyte simulations.
• These MLIPs surpass classical force fields and direct DFT methods for electrolyte modeling.
• This work accelerates the development of next-generation Na-ion battery electrolytes and other advanced energy storage systems.