Distilling latent electrostatics from foundation machine learning interatomic potentials

2026-06-12

AI Summary

This work addresses the limitations of current machine learning interatomic potential (MLIP) models, which lack explicit electrostatic interactions and are computationally expensive, thereby struggling to accurately capture long-range electrical responses. The authors propose Latent Ewald Summation (LES), a method that leverages unsupervised knowledge distillation from energy and force predictions of multiple base MLIP teacher models to extract Born effective charge tensors and infrared spectral responses. This approach enables, for the first time, a systematic evaluation of electrical fidelity in MLIPs, revealing that it is primarily governed by the level of density functional theory (DFT) used in training rather than model architecture. After fine-tuning with minimal high-accuracy DFT data, lightweight student models significantly improve structural and infrared predictions in challenging systems such as liquid water, concentrated hydrochloric acid, and TiO₂–water interfaces, effectively transforming general-purpose MLIPs into efficient electrostatically responsive potentials.

Abstract

Foundation machine learning interatomic potentials (MLIPs) have enabled atomistic simulations across broad regions of chemical and materials space, but many remain computationally expensive and lack explicit electrostatics, limiting their use for systems governed by long-range interactions and electrical response. Previously, we introduced Latent Ewald Summation (LES), which learns latent atomic charges and long-range electrostatics from density functional theory (DFT) energy and force labels alone. Here, we use LES to extract electrostatics that are latent in foundation models: energies and forces predicted by a teacher model are used to train a lightweight LES-augmented student MLIP, with optional fine-tuning on additional DFT data. The resulting models reduce computational cost while providing access to Born effective charge tensors, and infrared spectra. We benchmark student models distilled from a broad set of foundation MLIPs, including UMA, MACE, Orb, eSEN, GemNet-OC, PET, and EquiformerV2-based models, against experimental infrared spectra for liquid water, concentrated hydrochloric acid, and the anatase TiO2(101)-water interface. Across these systems, electrostatic response can be extracted from most foundation MLIPs. The benchmark further shows that the underlying DFT level and dataset used to train the teacher model play a larger role than architecture in determining electrostatic and spectroscopic accuracy. For the TiO2-water interface, fine-tuning with a modest amount of higher-level DFT data improves structural and infrared predictions. LES-based distillation therefore provides a practical route for converting foundation MLIPs into efficient, electrically responsive models, while also testing the physical fidelity encoded in foundation models.

Problem

• machine learning interatomic potentials
• electrostatics
• long-range interactions
• electrical response
• computational cost

Innovation

• Latent Ewald Summation
• machine learning interatomic potentials
• electrostatic distillation
• infrared spectra
• Born effective charges