A priori lattice parameter prediction without DFT

Develop an accurate method to estimate the lattice parameters of crystalline materials a priori, without performing density functional theory calculations, to enable reliable initialization of hypothetical structures in high-throughput dataset generation and active learning workflows.

Background

For generating large numbers of hypothetical crystal structures by elemental substitution into prototypes, an initial guess of the lattice parameters is required to avoid severe distortions or nonphysical configurations prior to relaxation. The authors tested heuristic approaches based on atomic radii but emphasize that a robust, generally accurate predictor is still lacking.

An accurate non-DFT estimator would reduce computational overhead, improve convergence, and enhance the quality of sampling for off-equilibrium datasets intended for training universal ML interatomic potentials.