Engineered Atomic Structures — MLIP Zoo
Collection
DeepMD/MACE potentials and training data for SSE, Na-ion, Li-S, and cathode battery materials research. GPU-trained today! • 10 items • Updated • 1
Li-Metal Coating Screen: AlF₃ / MgF₂ / Al₂O₃ / SbF₃ / ZnCl₂ (DeepMD)
What this is
2 of 5 DeepMD-kit potentials from a comparative screen of candidate protective coating materials for a Li-metal anode interface: AlF₃, MgF₂, Al₂O₃, SbF₃, ZnCl₂. Each was trained on AIMD data for that coating's Li-metal interface slab (300/1000/3000 K sampling), then used for longer MLMD screening across 300–450 K. The two included here (ZnCl₂, MgF₂) are the two with the lowest validation force RMSE of the five — not necessarily the two best coatings scientifically (AlF₃ is the current transport front-runner in this project; ask if you want it swapped in instead).
This is a comparative screen, not one material — the point is to rank candidates against each other on the same footing (same interface construction, same training/validation protocol) before committing to one for deeper study.
Files & Validation
| File | RMSE energy (eV/atom) | RMSE force (eV/Å) | Training steps | Size |
|---|---|---|---|---|
model/ZnCl2_Li.pb |
0.00435 | 0.0747 | 1,000,000 | 73.0 MB |
model/MgF2_Li.pb |
0.00573 | 0.103 | 1,000,000 | 58.9 MB |
RMSE values are validation-set (held-out) energy/force error, read directly from each run's DeepMD-kit lcurve.out at its final training step — not re-derived or estimated.
Elements involved
Al, Mg, Sb, Zn (coating cations) · F, Cl, O (coating anions) · Li (mobile ion, shared framework across all five).
Training pipeline
AIMD interface slabs (VASP, PBE, 300/1000/3000 K) → DeepMD-kit train/freeze/ compress → MLMD screening (300–450 K, 7 temperatures). Produced by HPCA (github.com/selvachandrasekaranselvaraj/hpca).
Citation
Selva Chandrasekaran Selvaraj, University of Illinois Chicago.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support