Generative model improves scalable all-atom crystal structure creation
GLASS: Global Latent Aggregation with Slot-based Set Decoding for Scalable All-Atom Crystal Generation
Machine Learning
Summary
Creating detailed 3D models of crystals with many atoms is hard because it's tricky to keep track of which atom corresponds to which position. The authors identify this as a core challenge in existing methods that treat atoms individually. They propose GLASS, which uses a global overview of the structure and builds all atoms at once, avoiding the problem of tracking atom positions during generation. Their approach works well on standard crystal datasets and can generate larger, complex metal-organic frameworks with more atoms than before.
What this means in practice
- •For materials design teams: Generate larger and more accurate atomic models of complex crystal structures to accelerate material discovery without predefining building blocks or composition.
- •For computational chemistry groups: Produce scalable all-atom crystal samples for simulations that need structures of challenging large metal-organic frameworks without manual topology input.
Authors
Hendrik Kraß, Seyed Mohamad Moosavi, Mathias Niepert
Abstract
Generative models for crystals enable the discovery of novel structures, but scaling all-atom generation to larger systems such as metal--organic frameworks remains challenging. We connect this difficulty to the correspondence problem of particle-space generation. Even on a single fixed target set, index-free permutation-equivariant particle flows require substantially more training for reliable generation as set size and density increase, under both independent and optimal-transport couplings. To resolve this challenge, we introduce GLASS---Global Latent Aggregation with Slot-based Set Decoding, which encodes structures in a permutation-invariant global latent space and learns their distribution via flow matching. A learned-slot decoder constructs all atoms in parallel, removing atom-wise correspondence from generative transport. On MP20, GLASS is competitive with particle-space models, and flow training can reach the validity of the training data at every structure size. On a QMOF subset, GLASS generates MOFs with up to 150 atoms per unit cell without conditioning on building blocks, topology, or composition, and approaches the structural validity of the training data. On both datasets, flow training exposes a validity--novelty tradeoff, and MOF novelty remains limited by autoencoder generalization on the available data. These results show that separating correspondence assignment from generative transport provides a simple route toward high-validity generation of larger atomistic systems.