TL;DR
N2 reference systems use HF (RHF) canonical orbitals by default. Switching to CASSCF-optimized orbitals lowers ground-state energy by ~80 mHa for the same active space, with no algorithmic change. Required for fair comparison against published SQD/NQS-SCI work that uses CASSCF orbitals.
Evidence
| System |
HF orbital best |
CASSCF orbital best |
Δ |
| N2-CAS(10,20) HCI gold |
-109.21473 |
-109.30609 |
86 mHa |
| N2-CAS(10,20) HI-NQS-SQD |
-109.21460 |
-109.29549 |
81 mHa |
Source: nano4 historical bench_orbital_fair_163794.log (job 163794, 2026-04-13)
Why removed/missing
qvartools/molecules/registry.py builds Hamiltonians directly from PySCF SCF integrals — never invokes CASSCF orbital rotation. SCI baseline (-109.21034) shipped as reference is the HF-orbital SCI, not the gold standard.
Literature support
Proposal
Add orbital_basis: "rhf" | "casscf" field to MOLECULE_REGISTRY entries. When casscf:
- Run PySCF CASSCF preprocessing (cache result) — typical wall ~7000s for CAS(10,20)
- Use CASSCF MO coefficients to rotate one- and two-electron integrals
- Build MolecularHamiltonian from rotated integrals
"n2-cas(10,20)": {
"factory": _make_n2_cas_10_20,
"orbital_basis": "casscf", # NEW; default "rhf" for compat
"casscf_cache_path": "~/.cache/qvartools/casscf/n2_cas_10_20.npz",
}
Impact
- Direct +80 mHa improvement on N2 benchmarks
- Required to compare with GTNN-SCI, ADAPT-QSCI, Robledo-Moreno SQD
- Without this, qvartools pipelines are systematically worse than published baselines by ~80 mHa
Effort: 1 week
- PySCF CASSCF wrapper + caching
- Integral rotation
- Registry field + factory updates
- Migration tests on H2/BeH2/NH3 (small) before scaling to 40Q
TL;DR
N2 reference systems use HF (RHF) canonical orbitals by default. Switching to CASSCF-optimized orbitals lowers ground-state energy by ~80 mHa for the same active space, with no algorithmic change. Required for fair comparison against published SQD/NQS-SCI work that uses CASSCF orbitals.
Evidence
Source: nano4 historical bench_orbital_fair_163794.log (job 163794, 2026-04-13)
Why removed/missing
qvartools/molecules/registry.py builds Hamiltonians directly from PySCF SCF integrals — never invokes CASSCF orbital rotation. SCI baseline (-109.21034) shipped as reference is the HF-orbital SCI, not the gold standard.
Literature support
Proposal
Add
orbital_basis: "rhf" | "casscf"field to MOLECULE_REGISTRY entries. Whencasscf:Impact
Effort: 1 week