SEI Solvation: Explicit-Solvent MD with UMA-OMol25

Machine-learning molecular dynamics with UMA-OMol25 for computing explicit-solvent energies of SEI-relevant species in lithium-ion battery electrolytes.

Project overview

The solid-electrolyte interphase (SEI) is the passivation layer that forms on lithium-ion battery anodes through electrolyte decomposition. Accurate prediction of its formation requires reliable energetics for thousands of intermediate species. The LIBE dataset (Spotte-Smith et al. 2021) provides DFT-level properties for ~17,000 such species but treats solvent implicitly via SMD — a continuum-dielectric model that degrades for charged species and mixed carbonate electrolytes.

This project replaces SMD with explicit-solvent molecular dynamics driven by the UMA-OMol25 machine-learning interatomic potential. For each species, MD trajectories in 3:7 EC/EMC yield ensemble-averaged solvation energies that retain atomistic detail of the solvation shell.

Mentor: Oliver Hvidsten PI: Prof. Kristin Persson (Lawrence Berkeley National Laboratory)

Methodology

• Model: UMA-OMol25 (Wood et al. 2025), trained on the OMol25 dataset at ωB97M-V/def2-TZVPD.
• System: 3:7 EC/EMC solvent box, ~100 solvent molecules per box, periodic boundary conditions.
• Workflow: NPT equilibration → NVT production → block-averaged energy difference between solute-containing and bare-solvent systems.
• Validation: liquid densities of pure EC and EMC against experiment; species energies against DFT at the OMol25 level of theory.

Repository structure

• src/ — core Python modules (molecule construction, MD setup, trajectory analysis)
• scripts/ — entry-point scripts for running calculations
• inputs/ — input structures, packmol templates
• outputs/ — calculation outputs (most gitignored)
• notes/ — write-ups, analysis notebooks, methodology notes
• refs/ — DFT reference values and experimental targets

Setup

Clone the repo and create the conda environment:

```bash git clone git@github.com:bronanatai/sei-solvation.git cd sei-solvation conda env create -f environment.yml conda activate uma ```

UMA model access requires applying for permission at https://huggingface.co/facebook/UMA and logging into the Hugging Face CLI:

```bash huggingface-cli login ```

Reproducing the headline results

(Filled in as results land.)

To reproduce the EC validation: ```bash python scripts/run_ec_validation.py ```

Status

In progress as of May 22nd, 2026. Current milestone: single-solvent energy validation against DFT.

References

• Wood, B. M. et al. UMA: A Family of Universal Models for Atoms. arXiv:2506.23971 (2025).
• Levine, D. S. et al. The Open Molecules 2025 (OMol25) Dataset. arXiv:2505.08762 (2025).
• Spotte-Smith, E. W. C. et al. Quantum chemical calculations of lithium-ion battery electrolyte and interphase species. Sci. Data 8, 203 (2021).