Massive discovery of crystal structures across dimensionalities by leveraging vector quantization

Abstract Discovering new functional crystalline materials through computational methods remains a challenge in materials science. We introduce VQCrystal, a deep learning framework leveraging discrete latent representations to overcome key limitations to crystal generation and inverse design. VQCrystal employs a hierarchical VQ-VAE architecture to encode global and atom-level crystal features, coupled with an inter-atomic potential model and a genetic algorithm to realize property-targeted inverse design. Benchmark evaluations on diverse datasets demonstrate VQCrystal’s capabilities in representation learning and crystal discovery. We further apply VQCrystal for both 3D and 2D material design. For 3D materials, the density-functional theory validation confirmed that 62.22% of bandgaps and 99% of formation energies of the 56 filtered materials matched the target range. 437 generated materials were validated as existing entries in the full MP-20 database outside the training set. For 2D materials, 73.91% of 23 filtered structures exhibited high stability with formation energies below -1 eV/atom.