feat: CAS active-space support for 40-58 qubit molecules (Issue #21 Phase 1) by thc1006 · Pull Request #22 · QuantumNoLab/qvartools

thc1006 · 2026-03-30T01:20:09Z

Summary

Closes Phase 1 of #21. Adds CASSCF/CASCI support to compute_molecular_integrals and 12 new CAS molecules to the registry, enabling 40-58 qubit systems.

Key changes:

compute_molecular_integrals(..., cas=(nelecas, ncas), casci=False) — new params
_compute_cas_integrals() helper — CASSCF/CASCI with auto-fallback at >50M configs
nuclear_repulsion = e_core (frozen-core), NOT mol.energy_nuc() — verified: N₂ CAS(10,8) e_core=-77.38 vs E_nuc=23.57
Cr₂ factory with fix_spin_(ss=0) to enforce singlet (prevents septet convergence)
Benzene CAS(6,15) via CASCI (ncas=15 auto-triggers CASCI)
joblib cache bypassed for CAS integrals (CASSCF is non-deterministic)
12 new registry entries: N₂-CAS ×5, Cr₂ ×6, Benzene ×1

Files changed (4):

src/qvartools/hamiltonians/integrals.py — cas/casci params + _compute_cas_integrals
src/qvartools/molecules/registry.py — 3 factory functions + 12 registry entries
tests/test_hamiltonians/test_cas_integrals.py — 19 tests (new)
tests/test_hamiltonians/test_cas_factories.py — 10 tests (new)

Test plan

19 CAS integral tests (shapes, e_core, electron counts, CASCI fallback, cache bypass)
10 factory/registry tests (completeness, is_cas flag, N₂-CAS creation)
ruff check + ruff format clean
354 existing tests pass (no regression)
N₂ CAS(10,12) end-to-end: 24Q Hamiltonian created successfully (89s)

Remaining Phases (separate PRs)

Phase 2: Sparse diag path (remove 10K limit) — unblocks >10K basis experiments
Phase 3: CAS-aware FCI + pipeline scripts — unblocks experiment scripts
Phase 4: Parameter auto-scaling — accuracy tuning

…uantumNoLab#21 Phase 1) Add CASSCF/CASCI support to compute_molecular_integrals via `cas` and `casci` parameters. When `cas=(nelecas, ncas)` is provided, runs CASSCF (or CASCI for ncas>=15 / >50M configs) and returns active-space integrals with nuclear_repulsion=e_core (frozen-core energy, NOT mol.energy_nuc()). New molecule factories: - _make_n2_cas: N₂ CAS with auto-CASCI for large active spaces - _make_cr2: Cr₂ with fix_spin_(ss=0) for singlet ground state - _make_benzene: Benzene CAS(6,15) Registry additions (12 new entries, 24→24 total): - N₂-CAS(10,12/15/17/20/26): 24-52 qubits - Cr₂ + Cr₂-CAS(12,18/20/26/28/29): 24-58 qubits - Benzene CAS(6,15): 30 qubits Cache: CAS integrals bypass joblib cache (CASSCF non-deterministic). TDD: 21 new tests (19 integrals + 2 registry completeness). All verified: N₂ CAS(10,8) e_core=-77.38 Ha (not E_nuc=23.57 Ha).

Copilot

Pull request overview

Adds CAS active-space integral generation and registers new large-qubit benchmark molecules, enabling 40–58 qubit workflows via PySCF CASSCF/CASCI integration.

Changes:

Extend compute_molecular_integrals / cached_compute_molecular_integrals with cas=(nelecas, ncas) and casci to produce active-space integrals (with auto-CASCI switching for large spaces).
Add CAS-capable molecule factories + 12 CAS registry entries (N₂-CAS variants, Cr₂ variants, Benzene).
Add new PySCF-marked slow tests for CAS integrals and CAS molecule factories/registry entries.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 12 comments.

File	Description
`src/qvartools/hamiltonians/integrals.py`	Introduces CAS/CASCI codepath and cache-bypass behavior for CAS in the cached helper.
`src/qvartools/molecules/registry.py`	Adds CAS geometries/factories and registers N₂/Cr₂/Benzene CAS entries.
`tests/test_hamiltonians/test_cas_integrals.py`	Adds CAS integral tests (shapes, e_core semantics, CASCI fallback, cache behavior).
`tests/test_hamiltonians/test_cas_factories.py`	Adds tests for new CAS factories and registry completeness.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

- Fix test_is_cas_flag_present: lowercase key comparison (was "CAS"→"cas") - Fix test_valid_config_count: use integrals fixture, not hardcoded value - Fix typo: TestCacheBypasForCAS → TestCacheBypassForCAS - Update docstring: RHF/CASSCF non-convergence warns, not raises - Add CAS electron/orbital validation in _compute_cas_integrals - Use shared _FCI_CONFIG_LIMIT constant in _make_cr2 (was hardcoded 50M) - Add ROHF non-convergence warning in _make_cr2 - Add comment explaining N₂ CAS 1.10 Å vs full-space 1.0977 Å

Copilot

Pull request overview

Copilot reviewed 4 out of 5 changed files in this pull request and generated 1 comment.

Comments suppressed due to low confidence (1)

src/qvartools/molecules/registry.py:34

registry.py imports _FCI_CONFIG_LIMIT from qvartools.hamiltonians.integrals. Since the name is underscore-prefixed (private), this creates a fragile cross-module dependency and makes refactors harder. Consider exporting a public constant (e.g. FCI_CONFIG_LIMIT) or moving the threshold into a shared config module, or avoid importing it here by delegating the CASCI decision entirely to compute_molecular_integrals.

from qvartools.hamiltonians.integrals import _FCI_CONFIG_LIMIT
from qvartools.hamiltonians.molecular import (
    MolecularHamiltonian,
    compute_molecular_integrals,
)

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Previously deferred 3 items — all now fixed: #1: Cr₂ singlet test: assert e_core in [-2040, -2030] Ha range (septet would give different e_core, catches missing fix_spin_) QuantumNoLab#4: get_integral_cache() now returns CAS-aware wrapper that bypasses joblib when cas is not None (was a real cache-escape bug) QuantumNoLab#7: Add test_cache_actually_bypassed_for_cas using mock.patch to verify compute_molecular_integrals is called directly Also from new comment: QuantumNoLab#13: Remove dead tuple/list branch for nelecas in _compute_cas_integrals (public API is tuple[int, int], nelecas is always int)

…mNoLab#22 audit) - Issue 7: Replace module-level pytest.importorskip("pyscf") with importlib.util.find_spec check so TestCASParameterSignature runs without PySCF installed - Issue 8: Add Raises section to _compute_cas_integrals docstring - Issue 9: Add full NumPy-style docstrings (Parameters, Returns, Raises) to _make_n2_cas, _make_cr2, _make_benzene factory functions - Issue 10: Update CHANGELOG.md [Unreleased] > Added with CAS entries - Issue 11: Update AGENTS.md Section 4 molecule table from 12 to 24 molecules (12 full-space + 12 CAS active-space)

…chk)

Merge with upstream/main brought in _MOLECULE_INFO_REGISTRY consistency check (from PR QuantumNoLab#18). Our 12 CAS molecules were in MOLECULE_REGISTRY but missing from _MOLECULE_INFO_REGISTRY, causing import-time RuntimeError.

Copilot

Pull request overview

Copilot reviewed 6 out of 6 changed files in this pull request and generated 3 comments.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

George930502 · 2026-03-30T14:05:38Z

Code review

Found 2 issues:

fix_spin_(ss=0) is only applied on the CASSCF branch, not the CASCI branch. For Cr2 entries with ncas >= 15 (cr2-cas(12,18), cr2-cas(12,20), cr2-cas(12,26), cr2-cas(12,28), cr2-cas(12,29)), the CASCI path is taken and spin enforcement is skipped entirely. The function's own docstring warns: "Without fix_spin_, CASSCF converges to the septet (S=3) instead of the singlet ground state." (bug due to conditional spin enforcement logic)

https://github.com/George930502/qvartools/blob/a96803ae306444443db32a8bbf89073df967e9c8/src/qvartools/molecules/registry.py#L527-L535

12 new CAS molecules added but README.md molecule table not updated. AGENTS.md Section 9 "Adding a New Molecule" step 4 says: "Update README.md molecule table"

https://github.com/George930502/qvartools/blob/a96803ae306444443db32a8bbf89073df967e9c8/AGENTS.md#L647-L653

🤖 Generated with Claude Code

_{- If this code review was useful, please react with 👍. Otherwise, react with 👎.}

George930502

Code review: 2 inline issues found.

…oLab#23, QuantumNoLab#24 merged

7 fixes from owner code reviews on PRs QuantumNoLab#20/QuantumNoLab#22/QuantumNoLab#23: 1. Cr₂ CASCI path now has fix_spin_(ss=0) — was missing, causing large CAS (ncas>=15) to converge to septet instead of singlet 2. README.md molecule table updated with all 24 molecules (12 full-space + 12 CAS active-space) 3. AGENTS.md Section 5 error handling: 10K → 50K for matrix_elements_fast 4. test_sparse_diag.py: add @pytest.mark.pyscf to 3 test classes that use h2_hamiltonian fixture (importorskip alone is insufficient) 5. Validation ordering: check shape/binary on raw initial_basis BEFORE .to(dtype=torch.long) cast (fail-fast, error shows original shape) 6. float("inf") → RuntimeError when all batches produce non-finite energies (convention per AGENTS.md: non-finite = RuntimeError) 7. ADR-004 TDD plan: note _MOLECULE_INFO_REGISTRY sync requirement

fix: address owner review findings across merged PRs #20/#22/#23

Copilot AI review requested due to automatic review settings March 30, 2026 01:20

Copilot started reviewing on behalf of thc1006 March 30, 2026 01:20 View session

Copilot AI reviewed Mar 30, 2026

View reviewed changes

thc1006 added 3 commits March 30, 2026 09:25

style: apply ruff format to test_cas_integrals.py

c323d16

fix: remove accidentally committed PySCF temp files, update .gitignore

4e3c2aa

thc1006 requested a review from Copilot March 30, 2026 01:29

Copilot started reviewing on behalf of thc1006 March 30, 2026 01:29 View session

Copilot AI reviewed Mar 30, 2026

View reviewed changes

Comment thread src/qvartools/hamiltonians/integrals.py Outdated

thc1006 added 3 commits March 30, 2026 09:35

merge: resolve .gitignore conflict with upstream/main (use broader *.…

d1acf7b

…chk)

thc1006 requested a review from Copilot March 30, 2026 10:33

Copilot started reviewing on behalf of thc1006 March 30, 2026 10:33 View session

fix: add 12 CAS molecules to _MOLECULE_INFO_REGISTRY

de0fa50

Merge with upstream/main brought in _MOLECULE_INFO_REGISTRY consistency check (from PR QuantumNoLab#18). Our 12 CAS molecules were in MOLECULE_REGISTRY but missing from _MOLECULE_INFO_REGISTRY, causing import-time RuntimeError.

Copilot AI reviewed Mar 30, 2026

View reviewed changes

Comment thread src/qvartools/molecules/registry.py

Comment thread src/qvartools/hamiltonians/integrals.py Outdated

Comment thread tests/test_hamiltonians/test_cas_integrals.py

fix: document ncas>=15 auto-CASCI trigger in docstring

a96803a

thc1006 requested a review from George930502 March 30, 2026 10:53

George930502 reviewed Mar 30, 2026

View reviewed changes

Comment thread src/qvartools/molecules/registry.py

Comment thread AGENTS.md

George930502 added 4 commits March 31, 2026 04:29

fix: improve cache-bypass test to verify joblib is actually skipped

85d9112

style: fix ruff formatting in test_cas_integrals.py

76c9e8f

merge: resolve CHANGELOG conflict with main

3263020

merge: resolve CHANGELOG conflict after PRs QuantumNoLab#20, QuantumN…

06a2628

…oLab#23, QuantumNoLab#24 merged

George930502 merged commit f91101c into QuantumNoLab:main Mar 31, 2026
5 checks passed

thc1006 deleted the feat/cas-active-space branch March 31, 2026 09:27

This was referenced Mar 31, 2026

fix: address owner review findings across merged PRs #20/#22/#23 #26

Merged

feat: CAS active space support for 40–58 qubit molecules (ADR-004) #21

Closed

George930502 added a commit that referenced this pull request Mar 31, 2026

Merge pull request #26 from thc1006/fix/owner-review-round2

0b75fd2

fix: address owner review findings across merged PRs #20/#22/#23

Conversation

thc1006 commented Mar 30, 2026

Summary

Test plan

Remaining Phases (separate PRs)

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Uh oh!

George930502 commented Mar 30, 2026

Code review

Uh oh!

George930502 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants