Machine learning interatomic potential can infer electrical response

2025-04-07

AI Summary

This work addresses the fundamental challenge in machine-learned interatomic potentials (MLIPs): accurately predicting electrostatic responses—specifically polarization tensors and Born effective charge (BEC) tensors—under external electric fields, using only energy and force labels from first-principles data, without requiring explicit charge, polarization, or BEC annotations. We propose a long-range MLIP framework coupled with implicit Ewald summation and leverage automatic differentiation of tensor-valued physical responses to enable end-to-end differentiable modeling of electrostatic properties. The method unifies predictions across diverse electric-field-driven phenomena, including infrared spectra, ionic conductivity, and ferroelectric phase transitions. Validated on liquid water (infrared absorption), superionic ice (ionic conductivity), and PbTiO₃ (ferroelectric hysteresis), our approach achieves accuracy comparable to density functional theory (DFT) while reducing computational cost by three orders of magnitude.

Technology Category

• Machine Learning: Matrix & Tensor Methods
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Natural Language Processing: (Large) Language Models

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Modeling the response of material and chemical systems to electric fields remains a longstanding challenge. Machine learning interatomic potentials (MLIPs) offer an efficient and scalable alternative to quantum mechanical methods but do not by themselves incorporate electrical response. Here, we show that polarization and Born effective charge (BEC) tensors can be directly extracted from long-range MLIPs within the Latent Ewald Summation (LES) framework, solely by learning from energy and force data. Using this approach, we predict the infrared spectra of bulk water under zero or finite external electric fields, ionic conductivities of high-pressure superionic ice, and the phase transition and hysteresis in ferroelectric PbTiO$_3$ perovskite. This work thus extends the capability of MLIPs to predict electrical response--without training on charges or polarization or BECs--and enables accurate modeling of electric-field-driven processes in diverse systems at scale.

Problem

• Modeling material response to electric fields efficiently
• Extracting polarization and BEC tensors from MLIPs
• Predicting electrical properties without charge training data

Innovation

• Extract polarization from long-range MLIPs
• Predict infrared spectra under electric fields
• Model phase transitions without charge training