Distilling First-Principles Accuracy into Compact Machine Learning Potentials for Condensed-Phase Chemistry
4 affiliations
- 1Initiative for Computational Catalysis, Flatiron Institute, New York, New York 10010, United States
- 2Max Planck Institute for Polymer Research, Mainz 55128, Germany
- 3Department of Physics and Astronomy, University College London, 7-19 Gordon St, London WC1H 0AH, U.K.
- 4Thomas Young Centre and London Centre for Nanotechnology, 9 Gordon St, London WC1H 0AH, U.K.
Abstract
Accurate machine learning interatomic potentials (MLIPs) have made first-principles-quality potential energy surfaces increasingly accessible for condensed-phase chemistry, but their inference cost can still limit the sampling needed to compute experimentally relevant observables. In this work, we use knowledge distillation to construct compact “student” models that retain the accuracy of much larger “teacher” models obtained through either finetuning of foundation models or Δ-learning. The resulting students reduce production simulation cost by approximately an order of magnitude, making high-accuracy sampling practical for challenging condensed-phase problems. We demonstrate this across three problems of increasing sampling complexity: finite-temperature NPT simulations of ice Ih, classical and path-integral simulations of liquid water over 240–370 K, and path-integral umbrella-sampling simulations of water dissociation at the anatase TiO2(101)/water interface. When compared, the distilled students reproduce the target observables of their teachers more reliably than equally compact models trained only on the available reference data. The liquid water student, distilled from a Δ-learned CCSD(T)-quality teacher, reproduces thermodynamic, structural, transport, and nuclear quantum effects over the full temperature range studied. At the TiO2/water interface, knowledge distillation makes PIMD umbrella sampling practical and shows that nuclear quantum effects lower the dissociation barrier by approximately 2 kcal/mol and shift the molecular–dissociated free energy difference into closer agreement with recent solid-state 17O NMR measurements. Our work demonstrates how knowledge distillation can make accurate MLIPs practical for the sampling methods needed to connect condensed-phase reaction thermodynamics with experiment, notably for interfacial chemistry and catalysis.
Related
- Distilling Large-Scale Machine Learning Interatomic Potentials to Predict Long-Timescale Electrolyte Transport Properties
- Quantum-Classical Effective Fragment Potential Embedding for Condensed-Phase Quantum Chemistry
- A Divide-and-Conquer Embedded Correlated Wavefunction Transfer Learning Approach for Chemically Accurate Condensed-Phase Interatomic Potentials: Liquid Water as a Test Case.
- Foundation interatomic potentials learn the practical chemistry they were never taught
- Toward Chemical Accuracy in Biomolecular Simulations through Data-Driven Many-Body Potentials: III. A Library of Amino Acid Building Blocks
BibTeX
@article{chen2026distilling,
title = {{Distilling First-Principles Accuracy into Compact Machine Learning Potentials for Condensed-Phase Chemistry}},
author = {Sijia Chen and Niamh O’Neill and Benjamin X. Shi and Venkat Kapil},
journal = {Journal of Chemical Theory and Computation},
year = {2026},
month = oct,
doi = {10.1021/acs.jctc.6c01131},
url = {https://doi.org/10.1021/acs.jctc.6c01131},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Distilling First-Principles Accuracy into Compact Machine Learning Potentials for Condensed-Phase Chemistry
AU - Sijia Chen
AU - Niamh O’Neill
AU - Benjamin X. Shi
AU - Venkat Kapil
PY - 2026
DA - 2026/10/09
JO - Journal of Chemical Theory and Computation
DO - 10.1021/acs.jctc.6c01131
UR - https://doi.org/10.1021/acs.jctc.6c01131
AB - Accurate machine learning interatomic potentials (MLIPs) have made first-principles-quality potential energy surfaces increasingly accessible for condensed-phase chemistry, but their inference cost can still limit the sampling needed to compute experimentally relevant observables. In this work, we use knowledge distillation to construct compact “student” models that retain the accuracy of much larger “teacher” models obtained through either finetuning of foundation models or Δ-learning. The resulting students reduce production simulation cost by approximately an order of magnitude, making high-accuracy sampling practical for challenging condensed-phase problems. We demonstrate this across three problems of increasing sampling complexity: finite-temperature NPT simulations of ice Ih, classical and path-integral simulations of liquid water over 240–370 K, and path-integral umbrella-sampling simulations of water dissociation at the anatase TiO2(101)/water interface. When compared, the distilled students reproduce the target observables of their teachers more reliably than equally compact models trained only on the available reference data. The liquid water student, distilled from a Δ-learned CCSD(T)-quality teacher, reproduces thermodynamic, structural, transport, and nuclear quantum effects over the full temperature range studied. At the TiO2/water interface, knowledge distillation makes PIMD umbrella sampling practical and shows that nuclear quantum effects lower the dissociation barrier by approximately 2 kcal/mol and shift the molecular–dissociated free energy difference into closer agreement with recent solid-state 17O NMR measurements. Our work demonstrates how knowledge distillation can make accurate MLIPs practical for the sampling methods needed to connect condensed-phase reaction thermodynamics with experiment, notably for interfacial chemistry and catalysis.
SN - 1549-9618
SN - 1549-9626
ER -