Universal Interatomic Potentials as Configuration-Space Generators for One-Shot and Iterative Fine-Tuning of Ab Initio-Accurate Material-Specific Models
Abstract
Universal machine-learning interatomic potentials (MLIPs) are rapidly becoming general-purpose tools for atomistic simulation, but their role in quantitative materials modeling when reactive events are involved remains unsettled. We compare five universal MLIPs across seven chemically diverse systems and find that strong performance on standard benchmarks does not guarantee accurate predictions of target observables. In particular, zero-shot models do not reliably reproduce reactive, transport, or high-barrier processes, exemplified here in particular by the sulfur-vacancy jump in MoS2. We therefore propose a practical alternative: universal MLIPs are used to generate long molecular dynamics trajectories, the resulting configurations are sub-sampled and relabeled with DFT, and material-specific MLIPs are subsequently trained or fine-tuned on the resulting first-principles datasets. This workflow converts universal models into efficient configuration-space generators while retaining ab initio reference labels for training. Across the tested systems, 2,000 DFT-recalculated structures are often sufficient to obtain accurate fine-tuned or trained-from-scratch models. For the most challenging case, iterative self-training progressively refines the sampled configuration space and recovers the DFT MoS2 potential energy profile with only $600$ first-principles calculations in total. The resulting workflow enables the generation of $1$ ns ab initio-quality trajectories - including training data generation and model creation - within three days.
Versions
Related
- Correcting the Liquid Densities of Universal Machine-Learning Interatomic Potentials Using Local-Number-Density Repulsion
- Reference Conventions and Structure Types, Rather Than 4f Occupancy, Shape Lanthanide Errors in Universal Machine-Learning Interatomic Potentials
- Can Universal Machine-Learning Interatomic Potentials Capture Small Polarons? A Geometry-Based Benchmark in Transition-Metal Oxides
- Bias in Universal Machine-Learned Interatomic Potentials and Its Effects on Fine-Tuning
- From sparse quantum-computing data to atomistic simulation with universal machine-learning interatomic potentials
BibTeX
@misc{hanseroth2026universal,
title = {{Universal Interatomic Potentials as Configuration-Space Generators for One-Shot and Iterative Fine-Tuning of Ab Initio-Accurate Material-Specific Models}},
author = {Jonas Hänseroth and Aaron Flötotto and Christian Dreßler},
howpublished = {arXiv},
year = {2026},
month = jun,
doi = {10.48550/arxiv.2606.23214},
eprint = {2606.23214},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2606.23214},
}
RIS
TY - UNPB
TI - Universal Interatomic Potentials as Configuration-Space Generators for One-Shot and Iterative Fine-Tuning of Ab Initio-Accurate Material-Specific Models
AU - Jonas Hänseroth
AU - Aaron Flötotto
AU - Christian Dreßler
PY - 2026
DA - 2026/06/22
PB - arXiv
DO - 10.48550/arxiv.2606.23214
UR - https://arxiv.org/abs/2606.23214
AB - Universal machine-learning interatomic potentials (MLIPs) are rapidly becoming general-purpose tools for atomistic simulation, but their role in quantitative materials modeling when reactive events are involved remains unsettled. We compare five universal MLIPs across seven chemically diverse systems and find that strong performance on standard benchmarks does not guarantee accurate predictions of target observables. In particular, zero-shot models do not reliably reproduce reactive, transport, or high-barrier processes, exemplified here in particular by the sulfur-vacancy jump in MoS_2. We therefore propose a practical alternative: universal MLIPs are used to generate long molecular dynamics trajectories, the resulting configurations are sub-sampled and relabeled with DFT, and material-specific MLIPs are subsequently trained or fine-tuned on the resulting first-principles datasets. This workflow converts universal models into efficient configuration-space generators while retaining ab initio reference labels for training. Across the tested systems, 2,000 DFT-recalculated structures are often sufficient to obtain accurate fine-tuned or trained-from-scratch models. For the most challenging case, iterative self-training progressively refines the sampled configuration space and recovers the DFT MoS_2 potential energy profile with only $600$ first-principles calculations in total. The resulting workflow enables the generation of $1$ ns ab initio-quality trajectories - including training data generation and model creation - within three days.
ER -