Crystal Structure Prediction

Crystal structure prediction (CSP) seeks the three-dimensional arrangements of atoms or molecules that form crystals, helping researchers identify materials and polymorphs before they are made. Methods include energy-landscape searches using density functional theory or machine-learned interatomic potentials, and generative models such as diffusion and flow matching that predict coordinates, lattices, and symmetry; molecular CSP also models molecular orientations and packing. Some approaches infer structures from powder or electron diffraction, while benchmarks assess structural matches, stability, novelty, and recovery of experimental forms. A central tension is that fast models can perform well on standard benchmarks yet rely on known structural prototypes, so generalization to unfamiliar structures and noisy experimental data remains a challenge.