Evaluating Predicted Densities, Hamiltonians, and Density Matrices as Periodic SCF Initializers
Abstract
Learned electronic states are usually evaluated by prediction error, even though their intended use is to accelerate the self-consistent-field (SCF) loop of density functional theory (DFT). We ask whether lower offline error actually yields a better SCF initializer. We construct ρHD-43K, a 43,851-crystal DFT corpus with aligned charge-density, Hamiltonian, and density-matrix labels, and extend the solver workflow to inject all three predicted states. The closed-loop benchmark compares direct converged-state prediction against residual prediction from solver-native references on a frozen test set of non-magnetic crystals, using paired convergence, iteration, and SCF-loop timing measurements. Injecting the exact converged density matrix or Hamiltonian as an oracle upper bound cuts the median SCF count from 16 to one, revealing substantial acceleration headroom. Among learned inputs, direct charge-density prediction (Charge3Net-E3) accelerates about 91% of paired crystals, saves a median of three SCF iterations, and yields a 1.18× SCF-loop speedup. In contrast, the tested residual-density and matrix initializers do not consistently improve over the standard superposition-of-atomic-densities baseline. These results establish that target-space accuracy alone is insufficient: a learned initializer must also be compatible with the nonlinear solver trajectory, and its utility must be measured in the loop.
Related
- Exact joint eigenvalue densities of non-Hermitian random matrices are Calogero scattering states
- Phase structure and observables at high densities from first principles QCD
- Low-x parton densities accounting for absorptive effects and exclusive J/ψ and Υ data from the LHC
- Conservation laws for a class of generic Hamiltonians
- Ab initio spin Hamiltonians and magnetism of Ce and Yb triangular-lattice compounds
BibTeX
@misc{chen2026evaluating,
title = {{Evaluating Predicted Densities, Hamiltonians, and Density Matrices as Periodic SCF Initializers}},
author = {Pin Chen and Jiang Li and Yutong Lu},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.15151},
eprint = {2609.15151},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.15151},
}
RIS
TY - UNPB
TI - Evaluating Predicted Densities, Hamiltonians, and Density Matrices as Periodic SCF Initializers
AU - Pin Chen
AU - Jiang Li
AU - Yutong Lu
PY - 2026
DA - 2026/09/14
PB - arXiv
DO - 10.48550/arxiv.2609.15151
UR - https://arxiv.org/abs/2609.15151
AB - Learned electronic states are usually evaluated by prediction error, even though their intended use is to accelerate the self-consistent-field (SCF) loop of density functional theory (DFT). We ask whether lower offline error actually yields a better SCF initializer. We construct ρHD-43K, a 43,851-crystal DFT corpus with aligned charge-density, Hamiltonian, and density-matrix labels, and extend the solver workflow to inject all three predicted states. The closed-loop benchmark compares direct converged-state prediction against residual prediction from solver-native references on a frozen test set of non-magnetic crystals, using paired convergence, iteration, and SCF-loop timing measurements. Injecting the exact converged density matrix or Hamiltonian as an oracle upper bound cuts the median SCF count from 16 to one, revealing substantial acceleration headroom. Among learned inputs, direct charge-density prediction (Charge3Net-E3) accelerates about 91% of paired crystals, saves a median of three SCF iterations, and yields a 1.18× SCF-loop speedup. In contrast, the tested residual-density and matrix initializers do not consistently improve over the standard superposition-of-atomic-densities baseline. These results establish that target-space accuracy alone is insufficient: a learned initializer must also be compatible with the nonlinear solver trajectory, and its utility must be measured in the loop.
ER -