MACE-MDP: A General Dipole and Polarizability Model for Organic Molecules and Materials

Molecular electronic response properties, including dipole moments and polarizabilities, govern vibrational spectroscopy and interactions with electric fields, thereby providing important insight into structure, intermolecular interactions, and function.While first-principles evaluation of these properties carries substantial computational overhead, predicting them with machine learning remains challenging because largescale training data spanning molecular and condensed-phase environments are scarce.In this work, we introduce SPICE-, a new dataset comprising over 1.8 million electricfield response calculations across molecular and condensed-phase environments, and MACE-MDP, a model trained to jointly predict molecular dipole moments and fully anisotropic polarizability tensors.MACE-MDP achieves first-principles accuracy at substantially reduced computational cost and exhibits transferability from small molecules and dimers to oligomers and molecular crystals, including accurate prediction of infrared and Raman spectra as well as reliable polymorph identification.This opens the door to high-throughput spectroscopic screening, accelerated molecular discovery, and rapid identification of crystal polymorphs in complex chemical environments.