Open Materials Generation with Stochastic Interpolants

The discovery of new materials is essential for enabling technological advancements. Computational approaches for predicting novel materials must effectively learn the manifold of stable crystal structures within an infinite design space. We introduce Open Materials Generation (OMatG), a unifying framework for the generative design and discovery of inorganic crystalline materials. OMatG employs stochastic interpolants (SI) to bridge an arbitrary base distribution to the target distribution of inorganic crystals via a broad class of tunable stochastic processes, encompassing both diffusion models and flow matching as special cases. In this work, we adapt the SI framework by integrating an equivariant graph representation of crystal structures and extending it to account for periodic boundary conditions in unit cell representations. Additionally, we couple the SI flow over spatial coordinates and lattice vectors with discrete flow matching for atomic species. We benchmark OMatG's performance on two tasks: Crystal Structure Prediction (CSP) for specified compositions, and de novo generation (DNG) aimed at discovering stable, novel, and unique structures. In our ground-up implementation of OMatG, we refine and extend both CSP and DNG metrics compared to previous works. OMatG establishes a new state of the art in generative modeling for materials discovery, outperforming purely flow-based and diffusion-based implementations. These results underscore the importance of designing flexible deep learning frameworks to accelerate progress in materials science. The OMatG code is available at https://github.com/FERMat-ML/OMatG.

The number of possible inorganic crystal structures is vast, and traditional simulations are too slow to explore them efficiently. We present OMatG, an open-source AI framework for generating realistic crystal structures. It uses stochastic interpolants—a flexible and highly customizable approach that transforms random atomic arrangements into stable, physically valid crystals. The model architecture enforces essential crystal properties, such as periodicity, and jointly learns atomic positions and element types. OMatG can accurately reconstruct known crystals from their chemical formulas and also propose entirely new, chemically plausible candidate crystals. In benchmarks, it surpasses previous AI methods in both accuracy and the discovery of unique, low-energy crystals. By reducing reliance on slow and costly trial-and-error searches, OMatG accelerates the search for new materials. This capability can help drive innovation in fields such as electronics, batteries, and clean energy.