A Fixed-Point Neural Operator for Size- and Functional-Transferable Hamiltonian Prediction
Researchers have developed HamEvo, a novel neural operator designed to accelerate density functional theory (DFT) calculations by predicting Kohn-Sham Hamiltonians. This method achieves significant error reductions of 35-49% compared to existing baselines and can predict molecular orbital energies with high accuracy. HamEvo demonstrates impressive scalability, extending its capabilities to larger molecules with minimal fine-tuning and offering inference speeds up to 242 times faster than conventional DFT. AI
IMPACT Accelerates scientific discovery by enabling faster and more accurate molecular simulations.