Answers are grounded in this paper and cite their sources.
Machine learning study demonstrates accelerated density functional theory calculations across molecular and solid systems, indicating enhanced efficiency for computational materials screening.
Density functional theory (DFT) is the workhorse in computational materials science and quantum chemistry, widely used for systems ranging from molecules to surfaces and bulk systems. The electronic properties are determined from the solution of the Kohn-Sham equations, which is obtained iteratively, starting from a best-guess initial charge density. If the number of atoms in the simulation box increases to a few hundred atoms and a large class of compounds is required to be screened, the computational effort becomes significant even when linear-scaling DFT codes are employed. We introduce the mcv2v-cGAN model based on conditional generative-adversarial networks (cGANs), which performs a multi-channel voxel-to-voxel mapping from readily available structural information to the final self-consistent charge density. In order to preserve the DFT-level accuracy, the predicted density is used as a better initial guess density, which renders a substantial decrease in the number of steps in the self-consistent loop. The mcv2v-cGAN model succeeds in making accurate predictions for the self-consistent charge densities and derived observables, using as test cases extended sets of organic molecules and defective silicon supercells. Moreover, a surrogate model for molecular dynamics using the predicted charge densities produced consistent trajectories while significantly reducing the computational effort.
No takes yet. Share an insight, caveat, or question.
Pantis-Simut et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: