New training signal improves machine-learned Hamiltonian models for DFT

Researchers have developed a new training signal called cumulative local density of states (cumulative LDOS) to improve machine-learned Hamiltonian models used in density functional theory (DFT). This method addresses limitations in existing approaches by analyzing prediction errors in the eigenbasis of the reference Hamiltonian. By fine-tuning three Hamiltonian networks using FermiScope, the cumulative LDOS signal significantly reduces errors in HOMO-LUMO gap predictions, even for out-of-distribution data, outperforming other methods like wavefunction-alignment loss. AI

IMPACT Enhances the accuracy and efficiency of computational chemistry simulations, potentially accelerating materials discovery and drug design.

RANK_REASON The cluster contains a research paper detailing a new method for improving machine learning models in computational physics. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• cumulative local density of states
• density functional theory
• FermiScope
• HOMO-LUMO gap
• Kohn–Sham equations

• paper
• infra

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