Browse Data ESRA-MLP-0007 MLIP Public
Juno Nam
0000-0002-9495-8599
Rafael Gomez-Bombarelli
Flow matching for accelerated simulation of atomic transport in crystalline materials
2025_JunoNam_RafaelGomezBombarelli_Nat.Mach.Intell._s42256-025-01125-4 ← click to see all uploads from this cell
10.1038/s42256-025-01125-4 ↗
2026-08-14 15:15:43
3ce0d8c2-079e-4193-9f91-d67402248b30
How to Cite
1 — Cite the dataset paper
↗ View paper
2 — Cite the ESRA platform
Galib, M. et al. (2026). ESRA: Energy Storage Research Assistant. Argonne National Laboratory. https://github.com/MusannaGalib/esra-platform
CC BY 4.0  ·  Data shared under Creative Commons Attribution 4.0

Location on Eagle / Perlmutter

/global/cfs/cdirs/m4845/gsharing/rafael-gomez-bombarelli/flow-matching-for-accelerated-simulation/2025-junonam-rafaelgomezbombarelli-natma/2025_JunoNam_RafaelGomezBombarelli_Nat.Mach.Intell._s42256-025-01125-4_COMPLETE_METADATA/mlip/ESRA-MLP-0007/
Open in Globus File Manager →

Linked Uploads

Type Group Date Project Link
MD Rafael Gomez-Bombarelli 2026-08-14 15:15 Flow matching for accelerated simulation of atomic transport in crystalline materials same cell

Experiment Metadata

Model Name LiFlow v0.1.0
Model Type Universal dataset: MACE-MP-0 small model NVT MD (pre-trained on DFT data). AIMD reference (LPS): DFT/VASP . AIMD reference (LGPS): DFT/VASP. Not a traditional MLIP — LiFlow learns displacement distributions, not energies/forces
Visibility public
Publication Doi 10.1038/s42256-025-01125-4
Institution Code ESRA
Target Properties conditional distribution of atomic displacements ; LiFlow predicts displacements (not energies or forces)
Dft Training Level Universal dataset: PBE (from MACE-MP-0)AIMD reference (LGPS/LPS): PBE
Training Dataset Name Universal MLIP dataset: 4,186 Li-containing structures × 4 temperatures × 25 ps NVT MD trajectories (MACE-MP-0); 153 GB on Zenodo. Also: LPS AIMD dataset and LGPS AIMD datase
Training Dataset Size 3,767 trajectories (90% of 4,186; composition-based train/test split); validation set sampled from training portion. Up to ~16,744 total trajectories (4 temperatures)