General Learning of the Electric Response of Inorganic Materials

Abstract

We introduce \texttt{MACE-Field}, a field-aware, $O(3)$-equivariant interatomic potential that learns a single electric enthalpy functional $\mathcal{F}(\{\mathbf{R}\},\mathbf{E})$ and obtains $\mathbf{P}$, $Z^{\ast }$, and $\mathbf{\alpha }$ by exact differentiation. A uniform field couples to latent equivariant features inside the \texttt{MACE} backbone, while the scalar energy readout preserves Maxwell reciprocity, the acoustic sum rule, and crystal tensor symmetries by construction. Because this coupling is a plug-in on top of standard \texttt{MACE}, existing energy/force foundation models can be upgraded to become field-aware. Benchmarked against semilocal DFT/DFPT reference data, a directly trained cross-chemistry ferroelectric model reproduces the same-branch Berry-phase and spontaneous polarisations across diverse inorganic crystals. Starting from the multihead foundation model \texttt{mace-mp-mh-0} and its OMAT-PBE head, joint fine-tuning on dielectric, ferroelectric, and replay data yields \texttt{MACE-Field-MH-0} foundation models, which predict $Z^{\ast }$, $\mathbf{\alpha }$, derived dielectric constants, and cross-chemistry polarisation trends with fidelity that captures branch-resolved polarisation and spontaneous-polarisation, while retaining strong force-field accuracy. Further, single-material \texttt{MACE-Field} models and \texttt{MACE-Field-MH-0} reproduce \ce{BaTiO3} hysteresis loops and $\alpha$-quartz infrared, Raman, and dielectric spectra from finite-field molecular dynamics, comparable to DFPT. These results show that a simple, physics-informed field coupling can endow atomistic foundation models with transferable dielectric and ferroelectric response, while targeted single-material training remains advantageous for the most quantitative spectroscopic predictions.

arXiv comment22 pages, 12 figures, 43 equations

References in corpus (22)

• Physics of thin-film ferroelectric oxides
• E(3)-Equivariant Graph Neural Networks for Data-Efficient and Accurate Interatomic Potentials
• Graph Networks as a Universal Machine Learning Framework for Molecules and Crystals
• PhysNet: A Neural Network for Predicting Energies, Forces, Dipole Moments and Partial Charges
• First principles calculations of the Shift Current Bulk Photovoltaic Effect in Ferroelectrics
• Molecular ferroelectric contributions to anomalous hysteresis in hybrid perovskite solar cells
• A First-Principles Approach to Insulators in Finite Electric Fields
• Ferroelectric Materials for Solar Energy Conversion: Photoferroics Revisited
• Benchmarking Materials Property Prediction Methods: The Matbench Test Set and Automatminer Reference Algorithm
• Design principles for shift current photovoltaics
• Symmetry-Adapted Machine-Learning for Tensorial Properties of Atomistic Systems
• Berry-phase treatment of the homogeneous electric field perturbation in insulators
• A deep potential model with long-range electrostatic interactions
• Incorporating long-range physics in atomic-scale machine learning
• Accuracy of generalized gradient approximation functionals for density functional perturbation theory calculations
• Accurate molecular polarizabilities with coupled-cluster theory and machine learning
• Deep neural network for the dielectric response of insulators
• Operators in Machine Learning: Response Properties in Chemical Space
• Predicting molecular dipole moments by combining atomic partial charges and atomic dipoles
• Modelling the dielectric constants of crystals using machine learning
• User Applications Driven by the Community Contribution Framework MPContribs in the Materials Project
• Robust model benchmarking and bias-imbalance in data-driven materials science: a case study on MODNet