Where Should Physics Enter a Molecular Crystal Generator?

2026-09-28

AI Summary

This study addresses the geometric and packing violations that arise when generative models predict molecular crystal structures. To this end, we propose CrystAF, a framework that systematically investigates how the timing of physical signal integration affects generation quality. Leveraging all-atom flow matching and UMA interatomic potentials, this work introduces a complementary strategy that combines post-training to learn physical preferences with inference-time correction of residual constraints, enabling cross-architecture transferability. The proposed approach significantly enhances molecular validity and crystal packing quality while preserving sampling efficiency and substantially reducing deployment costs.

Abstract

Generative models make molecular crystal structure prediction fast, but their samples still exhibit geometric and packing violations. Physics can be introduced during training, post-training, or inference, yet these choices are rarely compared with the generator and physical signal held fixed. We introduce CrystAF, an all-atom crystal flow-map generation model, and use it with the UMA interatomic potential to systematically study where physics should enter. Post-training learns physical preferences directly into CrystAF, improving molecular validity and crystal packing while leaving sampling unchanged: physics is paid for once during training rather than repeatedly at deployment. In contrast, UMA relaxation is effective at repairing local clashes but makes generation 6--26$\times$ slower, while learning from relaxed targets provides little benefit. These routes are complementary rather than competing. Physics-informed post-training first shifts the generated distribution toward more physically reasonable structures, after which inexpensive inference-time corrections further remove clashes and restore stereochemistry that the generator cannot represent. Importantly, the same post-training strategy also improves the multi-step all-atom Clari-M and rigid-body MolCrystalFlow generators, demonstrating transfer across architectures and representations. Together, our results suggest a simple principle: learn reusable physical alignment into the generator, and reserve inference-time physics for residual constraints that are better corrected than learned.

Problem

• molecular crystal generation
• physics-informed models
• crystal structure prediction
• generative models

Innovation

• Molecular crystal generation
• Flow matching
• Physics-informed post-training
• Interatomic potential
• Generative model alignment