Research map / Probabilistic and scientific modeling / Molecules and proteins
Towards a Universally Transferable Acceleration Method for Density Functional Theory
Zhe Liu, Yuyan Ni, Zhichen Pu, Qiming Sun, Siyuan Liu, Wen Yan
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| predict electron density coefficients instead of the Hamiltonian as initial guess (ablation) | QHNet · SCFbench OOD test | RIC: 179.47% -> 73.26% | Improved |
| predict density coefficients instead of the density matrix (ablation) | NequIP-L / QHNet · SCFbench OOD test | RIC: 91.69% -> 66.68% | Improved |
| scale NequIP model size from S to L (ablation) | NequIP density coefficients (jfit) · SCFbench OOD test | RIC: 89.46% -> 66.68% | Improved |
| use larger ETB beta=1.5 auxiliary basis for predicted coefficients (ablation) | NequIP-L · SCFbench OOD test | RIC: 66.68% -> 81.33% | Hurt |
| use a larger auxiliary basis for ground-truth density guess (ablation) | ground-truth density coefficients · D-Glucose SCF cycles | SCF iteration ratio: 53.8 -> 38.5 | Improved |
| add WALoss to Hamiltonian prediction training (ablation) | QHNet · SCFbench OOD test | RIC: 179.47% -> 170.57% to 183.31% | No clear effect |
| use flow-matching QHFlow instead of QHNet for Hamiltonian prediction (ablation) | Hamiltonian prediction · SCFbench OOD test | RIC: 179.47% -> 146.97% | Improved |
| explicitly normalize predicted density to N_e (ablation) | NequIP-L jfit · SCFbench OOD test | RIC: 66.68% -> 67.81% | Hurt |
| transfer model trained on PBE/def2-SVP to other XC functionals | NequIP-L jfit · SCFbench OOD test | RIC: 66.68% -> 71.22% (BLYP), 86.45% (SCAN), 85.51% (PBE0) | Mixed |
| transfer to larger orbital basis sets | NequIP-L jfit · SCFbench OOD test | RIC: 66.68% -> 75.24% (def2-TZVP), 75.98% (def2-QZVP) | Mixed |
| apply density-coefficient model to molecules of 100-200 atoms | NequIP-L jfit · QMugs 100-200 atoms | convergence rate: 100% convergence, RIC 0.73-0.82; Hamiltonian/density matrix near 0% | Improved |
| compute ground-truth coefficients via Coulomb-metric fitting rather than L2 projection (ablation) | density coefficient targets · SCF acceleration | RIC | No clear effect |
Closest papers by meaning
- Enhancing the Scalability and Applicability of Kohn-Sham Hamiltonians for Molecular Systems (ICLR 2025 · similarity 0.91)
- Self-Consistency Training for Density-Functional-Theory Hamiltonian Prediction (ICML 2024 · similarity 0.89)
- A Recipe for Charge Density Prediction (NeurIPS 2024 · similarity 0.89)
- Efficient and Scalable Density Functional Theory Hamiltonian Prediction through Adaptive Sparsity (ICML 2025 · similarity 0.89)
- High-order Equivariant Flow Matching for Density Functional Theory Hamiltonian Prediction (NeurIPS 2025 · similarity 0.89)
- Infusing Self-Consistency into Density Functional Theory Hamiltonian Prediction via Deep Equilibrium Models (NeurIPS 2024 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
