New ML models improve electronic ground-state prediction with physics-aligned training

Researchers have developed new machine learning models for electronic ground-state prediction that significantly improve generalization capabilities beyond current methods. By incorporating physical constraints derived from Kohn-Sham Density Functional Theory (KS-DFT) and introducing novel training techniques like OrthoNormal-Loss (ON-Loss) and Grassmann Restricted Occupied-Orbital Training (GROOT), these models achieve substantial reductions in error rates for energy and force predictions. The new approach demonstrates effectiveness in extrapolating from smaller datasets to larger ones and shows promise for applications in reactive chemistry. AI

IMPACT These physics-aligned models could accelerate scientific discovery in materials science and drug development by providing more accurate and generalizable predictions.

RANK_REASON The item is a research paper detailing novel methods for machine learning models in physics. [lever_c_demoted from research: ic=1 ai=1.0]

Read on Hugging Face Daily Papers →

• Grassmann Restricted Occupied-Orbital Training
• Hamiltonian GSMs
• Hugging Face
• Kohn-Sham Density Functional Theory with Complex, Spin-Restricted Orbitals: Accessing a New Class of Densities without the Symmetry Dilemma
• MLIPs
• OrthoNormal-Loss
• QM40
• QM9
• QMugs
• Residual Optimal-gauge Conditioning-aware KS-Eq. Training

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →