Foundation-Model Supervision of Distilled Potentials for Million-Atom Molecular Dynamics
Abstract
Machine-learned interatomic potentials distilled from foundation models enable fast large-scale molecular dynamics (MD), but they can fail locally in chemical environments that their training data did not sample. System-averaged error metrics and small random audits often miss such sparse or isolated failures. Here we use an exact equivariant foundation model, MACE-MPA-0, to check every atom of million-atom production simulations. Reorganizing two-layer MACE inference into a receiver-chunked three-pass schedule that keeps no per-edge intermediates resident reduces the marginal memory per atom 340-fold, evaluates up to 2.86×107 atoms on a single H200 GPU, and supervises 106-atom trajectories in LAMMPS at 1.3% of the run time. Distilled MACE students run four to nine times slower than NEP89 with 0.95-1.28 times the teacher's force error against density functional theory (DFT). In molten SiO2, whole-system checks uncover a persistent tail of under-coordinated atoms hidden by the system-averaged error and an isolated, unphysical O-O contact that a 1% random audit misses 94% of the time. On-the-fly retraining on teacher-labelled 13-Å clusters, with the force loss restricted to their inner 6 Å, removes three quarters of the failure tail and the kilobar-level pressure biases in SiO2 and compressed water; the repair converges to the teacher, not to DFT.
Related
- How Far Can Foundation Models Be Distilled for Efficient Jet Tagging?
- Strain-Induced BCC-to-FCC Transformation in Tungsten Uncovered by Molecular Dynamics Simulations with Machine Learning Potentials
- Melting Behavior and Phase Stability of CaO from Neural Network Potentials: a Molecular Dynamics Study
- A latent-space extrapolation grade built into graph atomic cluster expansion foundation potentials
- FP-DeErr: Application-Oriented Error Decomposition for Foundation Potentials
BibTeX
@misc{liu2026foundationmodel,
title = {{Foundation-Model Supervision of Distilled Potentials for Million-Atom Molecular Dynamics}},
author = {Xiaoqing Liu and Yangshuai Wang and Lei Shen},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.23371},
eprint = {2610.23371},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.23371},
}
RIS
TY - UNPB
TI - Foundation-Model Supervision of Distilled Potentials for Million-Atom Molecular Dynamics
AU - Xiaoqing Liu
AU - Yangshuai Wang
AU - Lei Shen
PY - 2026
DA - 2026/10/16
PB - arXiv
DO - 10.48550/arxiv.2610.23371
UR - https://arxiv.org/abs/2610.23371
AB - Machine-learned interatomic potentials distilled from foundation models enable fast large-scale molecular dynamics (MD), but they can fail locally in chemical environments that their training data did not sample. System-averaged error metrics and small random audits often miss such sparse or isolated failures. Here we use an exact equivariant foundation model, MACE-MPA-0, to check every atom of million-atom production simulations. Reorganizing two-layer MACE inference into a receiver-chunked three-pass schedule that keeps no per-edge intermediates resident reduces the marginal memory per atom 340-fold, evaluates up to 2.86×10^7 atoms on a single H200 GPU, and supervises 10^6-atom trajectories in LAMMPS at 1.3% of the run time. Distilled MACE students run four to nine times slower than NEP89 with 0.95-1.28 times the teacher's force error against density functional theory (DFT). In molten SiO_2, whole-system checks uncover a persistent tail of under-coordinated atoms hidden by the system-averaged error and an isolated, unphysical O-O contact that a 1% random audit misses 94% of the time. On-the-fly retraining on teacher-labelled 13-Å clusters, with the force loss restricted to their inner 6 Å, removes three quarters of the failure tail and the kilobar-level pressure biases in SiO_2 and compressed water; the repair converges to the teacher, not to DFT.
ER -