Data-Driven Lithium Migration Barrier Prediction 

Lithium-ion migration barriers govern ionic conductivity in cathode and solid-state battery materials. Accurate estimation of migration barriers typically requires Density Functional Theory (DFT) nudged elastic band (NEB) calculations, which are computationally expensive and difficult to scale across large materials spaces.

The goal of this project was to develop a machine learning model that:

1. Learns migration barrier physics from large-scale approximate datasets,

2. Transfers this knowledge to high-fidelity DFT data,

3. Achieves accurate DFT-level predictions in the small-data regime.

By pretraining on 122,000 approximate BVSE-NEB migration barriers and fine-tuning on 1,681 DFT-NEB calculations, we reduced DFT test error by 51% compared to training on DFT alone.

This work demonstrates that large-scale approximate physics can serve as an effective pretraining corpus for high-fidelity quantum predictions.