What is validated
Implemented ≠ validated. Working code is not a checked number.
This is the most important page in this documentation, and the one to read before trusting any result.
The authority is the wiki
The project wiki's VALIDATION.md is the authority on what each capability's
numbers are checked against, and where they are known to fail. It grades each
capability — proven / smoke / stub — rather than presenting them as a flat
list of equals.
This documentation describes what exists. It is not a claim that everything here is equally trustworthy, and the feature list on these pages should not be read as a validation claim.
Why the distinction is drawn so sharply
A quantum chemistry code can produce a plausible number in many ways that are wrong:
- a non-converged SCF returned as an ordinary result, because convergence is a flag rather than an error
- a method missing a physical term it does not mention — TDDFT's XC-kernel response is exactly this case, which is why it now warns
- a fallback model silently substituted for one atom in a molecule, changing a partitioning without changing the shape of the output
- a screening or truncation threshold that happens to be safe for the test system and not for yours
None of these look like failures. All of them have occurred and been fixed in this codebase. The remedy is not optimism — it is grading each capability separately and saying which ones are checked against ground truth.
Some anchors
Where numbers are checked, they are checked against external references rather than self-consistency:
| Capability | Anchor |
|---|---|
| CCSD(T) | H2O/cc-pVDZ matches PySCF to ~1e-6 Ha |
| G0W0@HF | matches MOLGW to ~5 meV |
| Gradients | validated against finite differences |
| RI-MP2 extensivity | 2e-12 Ha on a separated dimer |
| COSX exchange | water/cc-pVDZ dense-grid limit 3.3e-7 vs direct K; SCF dE 4.9e-6 Ha (water/cc-pVDZ), 1.7e-4 Ha (butane/def2-SVP), 1.2e-4 Ha (butane/def2-TZVP) at (50,110)+fit; butane/def2-QZVP K 137 s vs 400 s default direct J+K (one thread; LinK ratios withdrawn pending re-measurement after #50); density-driven screen t=1e-7 changes K by <= 2e-6 (water..octane); alkane_4..20/def2-SVP one-thread tail (C12-C20): A-build ~N^1.5, full K ~N^2.2 (dense half-transform GEMMs), slower than analytic exchange at SVP; Coulomb-only, no gradients. Open-shell (UHF/ROHF) wiring anchored 2026-09-08: CH3 doublet/cc-pVDZ SCF dE 1.96e-5 Ha (UHF) / 1.97e-5 Ha (ROHF) vs direct at (50,110)+fit, same iteration count; LinK over the same open-shell path is exact (UHF dE 0 bitwise, ROHF 1.4e-14 Ha) |
These are the figures stated in the repository itself. The wiki's
VALIDATION.md carries the full per-capability grading, including benchmark
sweeps (GW100 and others) whose numbers are not reproduced here — quoting a
benchmark MAE from memory rather than from the record is exactly the kind of
unchecked claim this page exists to discourage.
Known negatives
Reported rather than omitted:
- AO-sparse Laplace SOS-MP2 — truncation radius tracks molecular diameter instead of saturating; not a reduced-scaling path
- Local MP2 — J build still dense-from-RI; no scaling claim made
- TDHF/RPAx \( C_6 \) — ~60% low regardless of gap
- TDDFT with a DFT reference — omits the \( (ia|f_{xc}|jb) \) kernel response; warns at run time
- Hessians / frequencies — not implemented; the mpqc4 libint2 export lacks second-derivative integrals
Testing discipline
Some properties are pinned by tests rather than asserted in prose:
- Bit-identity across thread counts for reductions and permutations —
results do not depend on
RAYON_NUM_THREADS - ERI 8-fold permutational symmetry — verified against the engine, not assumed, since the MP2 code exploits it to compute only ~1/8 of quartets
- Size-extensivity and rotational invariance of total energies
- Memory guards in both directions — a starved budget must be refused and an ample budget must still run, because an over-estimating guard is also a bug
New guards are mutation-tested: a deliberate defect is injected and the test confirmed to fail before the guard is trusted. This has caught guards that passed while proving nothing.