Distilling First-Principles Accuracy into Compact Machine Learning Potentials for Condensed-Phase Chemistry

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.