Reproduce “High-order Equivariant Flow Matching for Density Functional Theory Hamiltonian Prediction” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Seongsu Kim, Nayoung Kim, Dongwoo Kim, Sungsoo Ahn
QHFLOW introduces a generative model leveraging flow matching and high-order SE(3)-equivariance to predict Kohn-Sham Hamiltonians from molecular geometries. This approach effectively models the distribution of Hamiltonians, reducing prediction error by up to 73% on the MD17 dataset and accelerating Density Functional Theory (DFT) calculations by 54% on the QH9-id benchmark.
Key insights
- Modeling Hamiltonian prediction as a generative problem via flow matching allows for capturing the complex, structured distributions of Hamiltonians, outperforming deterministic regression in physical fidelity.
- Rigorous enforcement of SE(3)-equivariance throughout the continuous-time flow matching process, combined with symmetry-aware priors, is crucial for robust and generalizable predictions.
- A targeted energy alignment fine-tuning step (WA-FT) can effectively enhance the accuracy of critical electronic properties like HOMO, LUMO, and the energy gap, without negatively impacting overall Hamiltonian prediction.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.