Chi route correspondence and capability matrix¶
Version 2, baseline 2026-11-14. Status: chi is experimental; D93 fitted/post-HF remains not-qualified. This page is generated from chi_routes.json by python formal/check_chi_routes.py --write and extends the formal inventory for roadmap #1007.
Six statuses are kept apart for every route. A proof about supplied finite operators does not show that native integrals produce those operators. A source check does not show that a native calculation ran. A native calculation that agrees with itself is not accuracy against an external reference, and a measured resource figure is not a scaling claim. No status here promotes a label.
proof: Lean claims that apply to the route: none, partial, or supplied-operators (exact statements about supplied finite operators).
correspondence: Whether a production function was compared with those claims: none, source-check (pure source and NumPy), native-check (formal script run on an identified runtime).
source: landed, library-only, or refused at the front door.
native: Recorded complete native calculations: none-recorded, tracker-only (verdict held in the named issue and not re-derived here), recorded-pass-internal (same-Hamiltonian only), recorded-fail.
accuracy: Scientific accuracy against a matched external reference. Every row is not-qualified.
cost: Cost and scaling: not-qualified, or bounded-measurement (measured resources exist, no scaling claim).
Summary¶
Route |
Proof |
Correspondence |
Source |
Native |
Accuracy |
Cost |
Roadmap row |
|---|---|---|---|---|---|---|---|
supplied-operators |
native-check |
landed |
recorded-pass-internal |
not-qualified |
bounded-measurement |
B |
|
supplied-operators |
none |
landed |
recorded-fail |
not-qualified |
not-qualified |
B |
|
supplied-operators |
none |
landed |
recorded-fail |
not-qualified |
not-qualified |
B |
|
none |
none |
landed |
none-recorded |
not-qualified |
not-qualified |
B |
|
supplied-operators |
none |
landed |
tracker-only |
not-qualified |
not-qualified |
B |
|
supplied-operators |
native-check |
landed |
recorded-pass-internal |
not-qualified |
not-qualified |
B |
|
supplied-operators |
none |
landed |
none-recorded |
not-qualified |
not-qualified |
B |
|
partial |
source-check |
landed |
recorded-fail |
not-qualified |
not-qualified |
B |
|
partial |
source-check |
landed |
tracker-only |
not-qualified |
not-qualified |
C |
|
partial |
source-check |
landed |
none-recorded |
not-qualified |
not-qualified |
C |
|
partial |
source-check |
landed |
none-recorded |
not-qualified |
not-qualified |
C |
|
partial |
source-check |
landed |
none-recorded |
not-qualified |
not-qualified |
C |
|
partial |
source-check |
landed |
tracker-only |
not-qualified |
not-qualified |
C |
|
partial |
source-check |
landed |
tracker-only |
not-qualified |
not-qualified |
C |
|
none |
none |
landed |
none-recorded |
not-qualified |
not-qualified |
C |
|
partial |
source-check |
landed |
tracker-only |
not-qualified |
not-qualified |
D |
|
partial |
source-check |
landed |
tracker-only |
not-qualified |
not-qualified |
D |
|
partial |
source-check |
landed |
tracker-only |
not-qualified |
not-qualified |
D |
|
supplied-operators |
native-check |
landed |
recorded-pass-internal |
not-qualified |
bounded-measurement |
E |
|
partial |
none |
landed |
none-recorded |
not-qualified |
not-qualified |
E |
|
partial |
source-check |
landed |
recorded-fail |
not-qualified |
not-qualified |
E |
|
none |
none |
landed |
recorded-pass-internal |
not-qualified |
not-qualified |
E |
|
none |
none |
landed |
none-recorded |
not-qualified |
not-qualified |
E |
|
none |
source-check |
landed |
recorded-pass-internal |
not-qualified |
not-qualified |
E |
|
partial |
source-check |
library-only |
recorded-fail |
not-qualified |
not-qualified |
G |
|
none |
source-check |
library-only |
recorded-pass-internal |
not-qualified |
not-qualified |
E to G |
|
supplied-operators |
native-check |
landed |
recorded-pass-internal |
not-qualified |
bounded-measurement |
F |
|
none |
none |
refused |
none-recorded |
not-qualified |
not-qualified |
G |
|
partial |
none |
landed |
tracker-only |
not-qualified |
not-qualified |
B |
|
partial |
none |
landed |
tracker-only |
not-qualified |
not-qualified |
later release |
|
none |
none |
landed |
tracker-only |
not-qualified |
not-qualified |
later release |
Refused for every route¶
Cells that are not 3D periodic:
_require_supported_scf_dimensioninpython/vibeqc/periodic/chi/scf.py.Charged effective cells and wrong multiplicity for closed-shell drivers:
_validate_closed_shell_probleminpython/vibeqc/periodic/chi/scf.py.Smearing and fractional occupations on open-shell drivers:
_validate_open_shell_probleminpython/vibeqc/periodic/chi/scf.py.Non-integer or character-dependent occupations on explicit-support HF/KS:
_require_hf_integer_stateinpython/vibeqc/periodic/chi/scf.py.Explicit nuclear image radius that changes a normalized retained state by more than 1e-3 Ha (#1212):
fitted_nuclear_image_checkinpython/vibeqc/periodic_fitted_support.py.Unvalidated unrestricted level shifts:
_require_zero_unrestricted_level_shiftinpython/vibeqc/periodic/chi/scf.py.Quadratic SCF fallback:
_require_inactive_quadratic_fallbackinpython/vibeqc/periodic/chi/scf.py.Symmetry integral acceleration:
_build_symmetry_planinpython/vibeqc/periodic/chi/scf.py.Unknown correlation spellings, including bare triples labels:
resolve_aiccm_correlationinpython/vibeqc/periodic_jk_method.py.Correlation on KS references, non-RI backends or loose SCF thresholds:
require_public_correlation_requestinpython/vibeqc/periodic/chi/correlation.py.Derivative requests outside the admitted RHF, MP2 and CCSD(T) surfaces:
require_chi_derivative_requestinpython/vibeqc/periodic/chi/optimize.py.
Hartree-Fock¶
RHF, RI backend, classic physical-pair source¶
Selector: method='aiccm', variant='chi', aiccm_backend='ri', scf_reference='rhf', explicit ChiCoulombControls
Hamiltonian and approximations: Finite Born-von Karman Coulomb Hamiltonian on the complete unshifted character mesh; neutral cell, zero-mean background, BvK Ewald exchange zero mode (exchange_q0=’bvk-ewald’). Classic physical-pair RI factors with the retained auxiliary metric projector.
Domain: Closed shell, all-electron or scalar ECP, 3D, integer occupations, zero temperature.
Production entry points: periodic/chi/scf.py::run_aiccm2026dev_b_rhf, periodic_k_gdf.py::_chi_rebuild_hf_acceptance.
Admission guards: periodic/chi/scf.py::_validate_closed_shell_problem, periodic/chi/scf.py::_require_supported_scf_dimension, periodic/chi/scf.py::_require_hf_integer_state, periodic_k_gdf.py::_chi_require_integer_occupations, periodic_fitted_support.py::fitted_nuclear_image_check, periodic_k_gdf.py::_chi_diagonalize_character.
Lean claims: bvk_rhf_fock_iff, bvk_fock_stationary_iff, bvk_quadratic_fock, bvk_orbital_stationary_iff, bvk_uhf_closed_shell_fock, bvk_uhf_closed_shell_ledger, bvk_uhf_retained_stationary_iff, bvk_retained_unprojected_counterexample, chi_projected_hermitian, chi_projection_skew_le_twice_error, chi_projection_skew_with_ao_defect, chi_projection_skew_split. Modules: Stationarity.lean, SpinStationarity.lean, Energy.lean, ComplexFit.lean, ChiProjection.lean.
Hypotheses and limits: Supplied Hermitian blocks; Hermitian overlap; reciprocal real-linear Coulomb and exchange maps; uniform character weights; integer metric-projector occupations. Native integrals, the fit, roundoff and SCF convergence are outside the theorems. For one Hermitian AO Fock and any complex rectangular X, the Frobenius skew of a supplied projected matrix is at most twice its distance from exact X^H F X. This gives no upper error bound. For an imperfect AO block, add ||X||_F squared times ||F-F^H||_F to the bound on projected skew. This separates upstream skew from arithmetic in the projection; no BLAS error model is assumed. The exact signed split attributes the projected skew to the transported AO defect and the skew of projection arithmetic. These matrices may cancel even when both norms are nonzero.
Checks: check_aiccm_stationarity.py, check_chi_uhf_stationarity.py, check_chi_uhf_native.py, check_chi_reduced_rank_native.py, check_chi_nuclear_image.py, check_chi_projection.py. Tests: test_chi_uhf_stationarity.py, test_chi_fitted_uhf.py, test_chi_hf_reference.py, test_chi_nuclear_image_check.py, test_chi_reduced_rank_reference.py.
Evidence:
Paper 2 packet 2026-11-09-chi-lih-rocksalt-rhf-v0.18.11: LiH one character, internal PASS, stability not assessed.
Paper 2 packet 2026-11-10-chi-lih-rocksalt-n2-rhf-completed-build-1fc985dc: LiH (2,2,2), accepted reference on a mixed pinned runtime.
Paper 2 packet 2026-11-09-lih-rocksalt-krhf-reference-pyscf-2.14.0: external reference unmatched, first unmatched operator is the one-electron Hamiltonian.
Paper 2 packet 2026-11-14-chi-uhf-native-build-c48994c2: closed-shell hydrogen cell on (3,1,1); factor reciprocity and energy slopes on complex characters PASS.
Paper 2 packet 2026-11-14-chi-reduced-rank-native-build-a4c86bcf: two dense reduced-rank cells accepted on the retained residual on a runtime containing the repair; the discriminating cell is refused on a control runtime without it.
Paper 2 packet 2026-11-14-chi-nuclear-image-native-build-54f73150: the nuclear image check of the explicit electron-nuclear operator (#1212) refuses the diffuse-basis collapse and a short radius on a compact basis; four accepted energies are bit-identical to the control. Internal PASS against frozen gates.
Paper 2 control on source 23e96cc3 for #1211: Hermitian AO Fock, projected skew 3.69e-9 Ha against a 9.34e-10 Ha gate, retained rank 6 of 6. No energy returned. Exact rational projection of the captured binary64 inputs: forward error 1.93e-9 Ha, transpose without conjugation differs by 1.21e3 Ha. The forward error has a Hermitian component of 5.74e-10 Ha and an anti-Hermitian component of 1.84e-9 Ha. Symmetrizing the projection would hide the latter and leave the former.
Paper 2 packet 2026-11-14-chi-projection-guard-build-5f829cf3: #1211 diagnostic PASS on native job df2b694410ee (one attempt). Character 1 retained rank 6/6, frame norm 3073.5582, condition 8646.8006, AO skew zero, projected skew 1.56558e-9 Ha above 9.33993e-10 Ha. Independent stored-operand replay matched the printed values; exact forward error 9.64676e-10 Ha. No accepted density or energy.
Refused: Smearing, fractional occupations, charged cells, 1D/2D cells, shifted meshes.
Remaining gaps:
External accuracy is unmatched; no matched periodic reference exists.
Orbital stability and global minimality are not assessed.
Native reduced-rank acceptance covers two constructed cells. On the three-character helium/hydrogen cell (#1211) both pinned runtimes refused a full-rank character after projection. The measured diagnostic explains the refusal but does not extend the admitted domain.
Sealed real-torus and translation references (reference.py::_admit_state) refuse a reduced-rank state by name, because those consumers use the complete AO Fock; canonical MP2 and restart validation admit it on the projected condition (#1194).
The native slope and reciprocity check on this complex-character RI route covers one small hydrogen cell only.
Owner: chi HF. Depends on: matched external reference (#1007 row A/B).
RHF, RI backend, short-range (RSGDF) source¶
Selector: aiccm_backend='ri', fit_source='sr'
Hamiltonian and approximations: Finite Born-von Karman Coulomb Hamiltonian on the complete unshifted character mesh; neutral cell, zero-mean background, BvK Ewald exchange zero mode (exchange_q0=’bvk-ewald’). Short-range fitted source with a separate long-range part.
Domain: Closed shell, all-electron, 3D.
Production entry points: periodic/chi/scf.py::run_aiccm2026dev_b_rhf.
Admission guards: periodic/chi/scf.py::_validate_closed_shell_problem.
Lean claims: bvk_rhf_fock_iff, bvk_fock_stationary_iff, bvk_quadratic_fock, bvk_orbital_stationary_iff. Modules: ScreenedPotential.lean, DomainTails.lean.
Hypotheses and limits: Supplied Hermitian blocks; Hermitian overlap; reciprocal real-linear Coulomb and exchange maps; uniform character weights; integer metric-projector occupations. Native integrals, the fit, roundoff and SCF convergence are outside the theorems.
Checks: check_aiccm_screened_potential.py, check_aiccm_domain_tails.py. Tests: test_chi_fit_source.py, test_chi_fitted_controls.py.
Evidence:
#453: a recorded LiH/STO-3G SR algebra FAIL (exchange reciprocity and J contraction) and negative-metric refusals; issue open.
Refused: SR post-HF, SR ROHF, SR ECP.
Remaining gaps:
SR reciprocity failure is unresolved; the reciprocity hypothesis of the stationarity theorems is not met on the recorded case.
Owner: #453 verifier. Depends on: #453.
RHF, direct four-centre backend¶
Selector: aiccm_backend='four_center'
Hamiltonian and approximations: Finite Born-von Karman Coulomb Hamiltonian on the complete unshifted character mesh; neutral cell, zero-mean background, BvK Ewald exchange zero mode (exchange_q0=’bvk-ewald’). Direct lattice-summed integrals through the BIPOLE driver; no fit.
Domain: Closed shell, all-electron, 3D.
Production entry points: periodic/chi/scf.py::run_aiccm2026dev_b_rhf, periodic/chi/symmetry.py::build_aiccm2026dev_b_evaluated_state.
Admission guards: periodic/chi/scf.py::_reject_four_center_aux_basis, periodic/chi/scf.py::_validate_closed_shell_problem.
Lean claims: bvk_rhf_fock_iff, bvk_fock_stationary_iff, bvk_quadratic_fock, bvk_orbital_stationary_iff. Modules: Stationarity.lean, GuardedERI.lean.
Hypotheses and limits: Supplied Hermitian blocks; Hermitian overlap; reciprocal real-linear Coulomb and exchange maps; uniform character weights; integer metric-projector occupations. Native integrals, the fit, roundoff and SCF convergence are outside the theorems.
Checks: check_aiccm_guarded_eri.py. Tests: test_chi_fock_state.py, test_chi_physical_support.py.
Evidence:
#451: public manifest replay criterion failed (serialized support identity); issue open.
Refused: fit_source other than auto, aux_basis.
Remaining gaps:
The final acceptance residual comes from the BIPOLE driver; its retained-space form was not examined here.
Under D93 only this unfitted backend can become eligible for qualification; fitted and post-HF rows stay not-qualified.
Owner: #451 owner, BIPOLE owner. Depends on: #451.
RHF, RIJCOSX backend¶
Selector: aiccm_backend='rijcosx'
Hamiltonian and approximations: Finite Born-von Karman Coulomb Hamiltonian on the complete unshifted character mesh; neutral cell, zero-mean background, BvK Ewald exchange zero mode (exchange_q0=’bvk-ewald’). RI Coulomb with a seminumerical exchange correction.
Domain: Closed shell, at least two cells.
Production entry points: periodic/chi/scf.py::run_aiccm2026dev_b_rhf.
Admission guards: periodic/chi/symmetry.py::build_aiccm2026dev_b_evaluated_state.
Lean claims: none. Modules: none.
Hypotheses and limits: None. The corrected operator is not the physical Fock of its energy, so the stationarity theorems do not apply.
Checks: none. Tests: test_chi_fock_state.py.
Refused: Evaluated-state export, correlation references and restart refuse this backend.
Remaining gaps:
No variational statement exists for the corrected operator.
Owner: unowned. Depends on: none.
RHF, RI backend on the MANTA integral engine¶
Selector: aiccm_backend='ri', integral_engine='manta'
Hamiltonian and approximations: Finite Born-von Karman Coulomb Hamiltonian on the complete unshifted character mesh; neutral cell, zero-mean background, BvK Ewald exchange zero mode (exchange_q0=’bvk-ewald’). MANTA one-electron operators and whole-mesh RI J/K, zero mode excluded and restored by chi.
Domain: Closed shell, all-electron, 3D, energy only.
Production entry points: periodic/chi/manta_backend.py::admit_chi_integral_engine.
Admission guards: periodic/chi/manta_backend.py::require_chi_manta_rhf_request.
Lean claims: bvk_rhf_fock_iff, bvk_fock_stationary_iff, bvk_quadratic_fock, bvk_orbital_stationary_iff. Modules: none.
Hypotheses and limits: Supplied Hermitian blocks; Hermitian overlap; reciprocal real-linear Coulomb and exchange maps; uniform character weights; integer metric-projector occupations. Native integrals, the fit, roundoff and SCF convergence are outside the theorems.
Checks: none. Tests: test_chi_manta_backend.py.
Evidence:
#633: same-operator comparison against the MANTA whole-mesh RI-RHF recorded by its owner.
Refused: UHF/ROHF/KS, correlation, gradients, ECP, explicit controls, KPoints objects.
Remaining gaps:
A distinct producer; no chi evidence transfers from the Libint RI rows.
Owner: #633 owner. Depends on: #633.
UHF, RI backend, classic physical-pair source¶
Selector: scf_reference='uhf', aiccm_backend='ri', explicit ChiCoulombControls
Hamiltonian and approximations: Finite Born-von Karman Coulomb Hamiltonian on the complete unshifted character mesh; neutral cell, zero-mean background, BvK Ewald exchange zero mode (exchange_q0=’bvk-ewald’). Classic physical-pair RI factors with the retained auxiliary metric projector. One Coulomb map of the total density and one exchange map per spin.
Domain: Integer per-spin occupations fixed at every character; high- or low-spin by the requested multiplicity.
Production entry points: periodic/chi/scf.py::run_aiccm2026dev_b_uhf, periodic_k_gdf.py::_chi_rebuild_hf_acceptance.
Admission guards: periodic/chi/scf.py::_validate_open_shell_problem, periodic/chi/scf.py::_require_zero_unrestricted_level_shift, periodic/chi/scf.py::_require_hf_integer_state, periodic_k_gdf.py::_chi_require_integer_occupations, periodic_fitted_support.py::fitted_nuclear_image_check.
Lean claims: bvk_uhf_quadratic_fock, bvk_uhf_stationary_iff, bvk_uhf_fock_iff, bvk_uhf_half_fock_ledger, bvk_seam_shifted_exchange_reciprocal, bvk_uhf_retained_stationary_iff, bvk_retained_unprojected_counterexample, bvk_spin_rotation_feasible, bvk_uhf_orbital_stationary_iff, bvk_seam_exchange_reciprocal, bvk_uhf_closed_shell_fock, bvk_uhf_closed_shell_ledger, bvk_retained_residual, bvk_retained_density, bvk_uhf_retained_fock. Modules: SpinStationarity.lean.
Hypotheses and limits: Supplied Hermitian blocks; Hermitian overlap; reciprocal real-linear Coulomb and exchange maps; uniform character weights; integer metric-projector occupations. Native integrals, the fit, roundoff and SCF convergence are outside the theorems.
Checks: check_chi_uhf_stationarity.py, check_chi_uhf_native.py. Tests: test_chi_uhf_stationarity.py, test_chi_fitted_uhf.py, test_chi_direct_uhf.py, test_chi_nuclear_image_check.py.
Evidence:
Paper 2 packet 2026-11-14-chi-uhf-native-build-c48994c2: hydroxyl doublet, hydrogen triplet and closed-shell collapse on (3,1,1), STO-3G; independent spin-Fock rebuild, ledger, factor reciprocity and energy slopes PASS on a runtime that predates the acceptance repair, and again in packet 2026-11-14-chi-reduced-rank-native-build-a4c86bcf on a runtime that contains it, together with the synthetic check in import mode.
No other native chi UHF calculation is recorded in the Paper 2 evidence catalog.
Refused: Smearing, fractional or character-dependent occupations, unrestricted level shifts without explicit controls.
Remaining gaps:
The native check covers three small complete-rank classic cells; larger cells, other bases and the short-range source are not covered.
No native reduced-rank UHF calculation is recorded; the reduced-rank native cells are closed shell.
Stability and spin contamination are not assessed.
Owner: chi HF. Depends on: larger native cases.
UHF, four-centre and RIJCOSX backends¶
Selector: scf_reference='uhf', aiccm_backend in ('four_center','rijcosx')
Hamiltonian and approximations: Finite Born-von Karman Coulomb Hamiltonian on the complete unshifted character mesh; neutral cell, zero-mean background, BvK Ewald exchange zero mode (exchange_q0=’bvk-ewald’).
Domain: As the UHF RI row.
Production entry points: periodic/chi/scf.py::run_aiccm2026dev_b_uhf.
Admission guards: periodic/chi/scf.py::_validate_open_shell_problem.
Lean claims: bvk_uhf_stationary_iff. Modules: SpinStationarity.lean.
Hypotheses and limits: Supplied Hermitian blocks; Hermitian overlap; reciprocal real-linear Coulomb and exchange maps; uniform character weights; integer metric-projector occupations. Native integrals, the fit, roundoff and SCF convergence are outside the theorems.
Checks: none. Tests: test_chi_direct_uhf.py.
Refused: MANTA UHF.
Remaining gaps:
No correspondence check for the direct driver’s final acceptance; RIJCOSX has no variational statement.
Owner: unowned. Depends on: hf-rhf-four-center.
ROHF, RI backend, classic source¶
Selector: scf_reference='rohf', aiccm_backend='ri', gdf_method='rsgdf', explicit physical_pairs controls
Hamiltonian and approximations: Finite Born-von Karman Coulomb Hamiltonian on the complete unshifted character mesh; neutral cell, zero-mean background, BvK Ewald exchange zero mode (exchange_q0=’bvk-ewald’). Classic physical-pair RI factors with the retained auxiliary metric projector. One shared spatial orbital set with closed, open and virtual categories.
Domain: High-spin, integer occupations, zero Fock mixing and level shift, symmetry off.
Production entry points: periodic/chi/rohf.py::run_chi_rohf, periodic/chi/scf.py::run_aiccm2026dev_b_rohf.
Admission guards: periodic/chi/rohf.py::validate_rohf_state, periodic_fitted_support.py::fitted_nuclear_image_check, periodic/chi/rohf.py::_require_retained_seed.
Lean claims: chi_rohf_gradient_entry, chi_rohf_gradient_zero_iff. Modules: ChiROHF.lean.
Hypotheses and limits: Supplied Hermitian spin Focks in one common retained orbital frame; nested integer alpha/beta occupations (closed, singly occupied, virtual). The shared commutator vanishes iff the three independent blocks vanish. No native source, nonlinear derivative, floating point or solver convergence theorem.
Checks: check_chi_rohf_blocks.py. Tests: test_chi_rohf_reference.py, test_chi_rohf_public.py, test_chi_nuclear_image_check.py.
Evidence:
#523: source landed, scientific acceptance pending with the validator.
Optional initial-coefficient API: synthetic rank-one counterexample with identical unit metric norms but energies 0.5 and 1 Ha; out-of-span seed now refused, in-span complex rotations accepted. Not native evidence.
Fixed-operator source oracle: three characters, four AOs, three retained orbitals, all 18 real/imaginary inter-category directions; the two extrapolated step pairs match to 1.66e-12 Ha. Equal and opposite nonzero spin gradients cancel on the stationary control.
Paper 2 LiH2 doublet control, source 23e96cc3, mesh (1,1,1), no frozen core: 16 MiB trace cap refused before integration. No ROHF energy, RMP2 singles or doubles; no physical verdict.
Refused: Four-centre, RIJCOSX, SR source, MANTA, level shift, mixing, ROHF-CC.
Remaining gaps:
The algebraic shared gradient is checked; a proof deriving it from the nonlinear physical energy along feasible ROHF paths remains open.
The gradient is computed in C^H F_sigma C and requires equal retained ranks. Native three-block derivative checks, independent of the orbital optimizer, are still owed; the optional seed guard has source evidence only.
Owner: #523 validator. Depends on: #523.
Kohn-Sham¶
RKS, LDA, explicit controls¶
Selector: functional=..., chi_xc_controls with support_policy (direct) or coulomb_controls (RI); RKS, LDA
Hamiltonian and approximations: Finite Born-von Karman Coulomb Hamiltonian on the complete unshifted character mesh; neutral cell, zero-mean background, BvK Ewald exchange zero mode (exchange_q0=’bvk-ewald’). Semilocal exchange-correlation energy on the declared cell grid and density-image support.
Domain: Neutral all-electron 3D, symmetry off; RKS LDA consumes density.
Production entry points: periodic/chi/scf.py::_run_chi_semilocal, periodic/chi/scf.py::run_aiccm2026dev_b_rks, periodic/chi/xc.py::functional_identity, periodic/chi/xc.py::_evaluate, periodic/chi/xc.py::_functional.
Admission guards: periodic/chi/xc.py::functional_identity, periodic_fitted_support.py::fitted_nuclear_image_check, periodic/chi/xc.py::_density.
Lean claims: chi_grid_rho_adjoint, chi_grid_rho_line, chi_lda_grid_deriv. Modules: ChiGrid.lean.
Hypotheses and limits: Fixed complex AO rows, character and quadrature weights; the scalar local energy density per unit volume is differentiable at each sampled density with supplied derivative. The weighted derivative pairs the AO density direction with V(k) = sum_g v(g) de(g) u(g,k)^H u(g,k). This is the discretized local potential (Kohn and Sham, Sec. II, DOI 10.1103/PhysRev.140.A1133); the quadrature is held fixed (Becke, DOI 10.1063/1.454033).
Checks: test_chi_xc.py. Tests: test_chi_xc.py, test_chi_direct_ks.py, test_chi_ks_reference.py, test_chi_nuclear_image_check.py.
Evidence:
#471: H2 direct Gamma LDA fixed-density potential and derivative ladders (0c07d5230991, source 4c160c36), cold SCF (455c3236a398, 562b013f), and prepared/reuse/public checks (16998a1dd450 and 72213faa8c7a, 74d97990).
Refused: External, double-hybrid, VV10 and Laplacian functionals; RIJCOSX; ECP.
Remaining gaps:
No theorem identifies the native AO rows, grid, Libxc derivatives or their floating-point errors. The scalar derivative does not establish spin, GGA or meta-GGA correspondence.
Keep density-image, partition, radial/angular grid and SCF convergence separate. See #1169 C471-1 through C471-9, including the fitted SCF and outstanding spin/class rows.
Owner: #1169 scheduled campaign; #471 scoped receipts. Depends on: B rows.
RKS, GGA, explicit controls¶
Selector: functional=..., chi_xc_controls with support_policy (direct) or coulomb_controls (RI); RKS, GGA
Hamiltonian and approximations: Finite Born-von Karman Coulomb Hamiltonian on the complete unshifted character mesh; neutral cell, zero-mean background, BvK Ewald exchange zero mode (exchange_q0=’bvk-ewald’). Semilocal exchange-correlation energy on the declared cell grid and density-image support.
Domain: Neutral all-electron 3D, symmetry off; RKS GGA consumes density and gradient invariants.
Production entry points: periodic/chi/scf.py::_run_chi_semilocal, periodic/chi/scf.py::run_aiccm2026dev_b_rks, periodic/chi/xc.py::functional_identity, periodic/chi/xc.py::_evaluate, periodic/chi/xc.py::_functional.
Admission guards: periodic/chi/xc.py::functional_identity, periodic_fitted_support.py::fitted_nuclear_image_check, periodic/chi/xc.py::_density.
Lean claims: chi_grid_rho_adjoint, chi_grid_rho_line, chi_grid_gradient_adjoint, ks_rks_gradient_flow, chi_semilocal_grid_deriv. Modules: ChiGrid.lean, KohnSham.lean.
Hypotheses and limits: Fixed complex AO and Cartesian derivative rows, fixed quadrature, Hermitian density variations. The AO gradient adjoint gives the two conjugate outer products. The separate local invariant derivative supplies the gradient coefficients. The finite sum chain rule assembles all supplied field kernels into one potential per spin and character at the same density; the local derivative is supplied in density, Cartesian gradient and (if used) kinetic coordinates.
Checks: test_chi_xc.py. Tests: test_chi_xc.py, test_chi_direct_ks.py, test_chi_ks_reference.py, test_chi_nuclear_image_check.py.
Evidence:
No standalone native qualification for this spin and functional class is identified in the #471 closing ledger. Other spin/class receipts do not transfer.
Refused: External, double-hybrid, VV10 and Laplacian functionals; RIJCOSX; ECP.
Remaining gaps:
No theorem derives Libxc derivatives, AO rows, image selection or quadrature from the native inputs; finite-precision and physical accuracy remain open.
Keep density-image, partition, radial/angular grid and SCF convergence separate. See #1169 C471-1 through C471-9, including the fitted SCF and outstanding spin/class rows.
Owner: #1169 scheduled campaign; #471 scoped receipts. Depends on: B rows.
RKS, MGGA, explicit controls¶
Selector: functional=..., chi_xc_controls with support_policy (direct) or coulomb_controls (RI); RKS, MGGA
Hamiltonian and approximations: Finite Born-von Karman Coulomb Hamiltonian on the complete unshifted character mesh; neutral cell, zero-mean background, BvK Ewald exchange zero mode (exchange_q0=’bvk-ewald’). Semilocal exchange-correlation energy on the declared cell grid and density-image support.
Domain: Neutral all-electron 3D, symmetry off; RKS MGGA consumes density, gradient invariants and kinetic density tau.
Production entry points: periodic/chi/scf.py::_run_chi_semilocal, periodic/chi/scf.py::run_aiccm2026dev_b_rks, periodic/chi/xc.py::functional_identity, periodic/chi/xc.py::_evaluate, periodic/chi/xc.py::_functional.
Admission guards: periodic/chi/xc.py::functional_identity, periodic_fitted_support.py::fitted_nuclear_image_check, periodic/chi/xc.py::_density.
Lean claims: chi_grid_rho_adjoint, chi_grid_rho_line, chi_grid_gradient_adjoint, ks_rks_gradient_flow, chi_grid_tau_adjoint, chi_semilocal_grid_deriv. Modules: ChiGrid.lean, KohnSham.lean.
Hypotheses and limits: Fixed complex AO and Cartesian derivative rows, fixed quadrature, Hermitian density variations. The AO gradient adjoint gives the two conjugate outer products. The separate local invariant derivative supplies the gradient coefficients. For positive kinetic density, tau contributes one half of the sum of derivative-row Gram matrices (Furness et al., DOI 10.1021/acs.jpclett.0c02405). The finite sum chain rule assembles all supplied field kernels into one potential per spin and character at the same density; the local derivative is supplied in density, Cartesian gradient and (if used) kinetic coordinates.
Checks: test_chi_xc.py. Tests: test_chi_xc.py, test_chi_direct_ks.py, test_chi_ks_reference.py, test_chi_nuclear_image_check.py.
Evidence:
No standalone native qualification for this spin and functional class is identified in the #471 closing ledger. Other spin/class receipts do not transfer.
Refused: External, double-hybrid, VV10 and Laplacian functionals; RIJCOSX; ECP.
Remaining gaps:
No theorem derives Libxc derivatives, AO rows, image selection or quadrature from the native inputs; finite-precision and physical accuracy remain open.
Keep density-image, partition, radial/angular grid and SCF convergence separate. See #1169 C471-1 through C471-9, including the fitted SCF and outstanding spin/class rows.
Owner: #1169 scheduled campaign; #471 scoped receipts. Depends on: B rows.
UKS, LDA, explicit controls¶
Selector: functional=..., chi_xc_controls with support_policy (direct) or coulomb_controls (RI); UKS, LDA
Hamiltonian and approximations: Finite Born-von Karman Coulomb Hamiltonian on the complete unshifted character mesh; neutral cell, zero-mean background, BvK Ewald exchange zero mode (exchange_q0=’bvk-ewald’). Semilocal exchange-correlation energy on the declared cell grid and density-image support.
Domain: Neutral all-electron 3D, symmetry off; UKS LDA consumes density.
Production entry points: periodic/chi/scf.py::_run_chi_semilocal, periodic/chi/scf.py::run_aiccm2026dev_b_uks, periodic/chi/xc.py::functional_identity, periodic/chi/xc.py::_evaluate, periodic/chi/xc.py::_functional.
Admission guards: periodic/chi/xc.py::functional_identity, periodic_fitted_support.py::fitted_nuclear_image_check, periodic/chi/xc.py::_density.
Lean claims: chi_grid_rho_adjoint, chi_grid_rho_line, chi_spin_lda_grid_deriv. Modules: ChiGrid.lean.
Hypotheses and limits: Fixed finite complex AO rows, character and quadrature weights. At the actual spin densities the supplied local energy per unit volume has a joint derivative with the declared spin coefficients. Independent spin variations then pair with their separate AO potentials; mixed spin dependence is allowed.
Checks: test_chi_xc.py. Tests: test_chi_xc.py, test_chi_direct_ks.py, test_chi_ks_reference.py, test_chi_nuclear_image_check.py.
Evidence:
No standalone native qualification for this spin and functional class is identified in the #471 closing ledger. Other spin/class receipts do not transfer.
Refused: External, double-hybrid, VV10 and Laplacian functionals; RIJCOSX; ECP.
Remaining gaps:
Native grid, AO, Libxc spin coefficients, floating-point errors and a matched physical reference remain unproved or unqualified.
Keep density-image, partition, radial/angular grid and SCF convergence separate. See #1169 C471-1 through C471-9, including the fitted SCF and outstanding spin/class rows.
Owner: #1169 scheduled campaign; #471 scoped receipts. Depends on: B rows.
UKS, GGA, explicit controls¶
Selector: functional=..., chi_xc_controls with support_policy (direct) or coulomb_controls (RI); UKS, GGA
Hamiltonian and approximations: Finite Born-von Karman Coulomb Hamiltonian on the complete unshifted character mesh; neutral cell, zero-mean background, BvK Ewald exchange zero mode (exchange_q0=’bvk-ewald’). Semilocal exchange-correlation energy on the declared cell grid and density-image support.
Domain: Neutral all-electron 3D, symmetry off; UKS GGA consumes density and gradient invariants.
Production entry points: periodic/chi/scf.py::_run_chi_semilocal, periodic/chi/scf.py::run_aiccm2026dev_b_uks, periodic/chi/xc.py::functional_identity, periodic/chi/xc.py::_evaluate, periodic/chi/xc.py::_functional.
Admission guards: periodic/chi/xc.py::functional_identity, periodic_fitted_support.py::fitted_nuclear_image_check, periodic/chi/xc.py::_density.
Lean claims: chi_grid_rho_adjoint, chi_grid_rho_line, chi_grid_gradient_adjoint, ks_uks_gradient_flow, chi_semilocal_grid_deriv. Modules: ChiGrid.lean, KohnSham.lean.
Hypotheses and limits: Fixed complex AO and Cartesian derivative rows, fixed quadrature, Hermitian density variations. The AO gradient adjoint gives the two conjugate outer products. The separate local invariant derivative supplies the gradient coefficients. The finite sum chain rule assembles all supplied field kernels into one potential per spin and character at the same density; the local derivative is supplied in density, Cartesian gradient and (if used) kinetic coordinates.
Checks: test_chi_xc.py. Tests: test_chi_xc.py, test_chi_direct_ks.py, test_chi_ks_reference.py, test_chi_nuclear_image_check.py.
Evidence:
#471: H2 PBE UKS on (3,1,1), fixed-density and spin-gradient negative (1a0c9df665ac, 562b013f), covariance (77759635577e, same source), direct cold SCF (9d33a565c6ba, 74d97990).
Refused: External, double-hybrid, VV10 and Laplacian functionals; RIJCOSX; ECP.
Remaining gaps:
No theorem derives Libxc derivatives, AO rows, image selection or quadrature from the native inputs; finite-precision and physical accuracy remain open.
Keep density-image, partition, radial/angular grid and SCF convergence separate. See #1169 C471-1 through C471-9, including the fitted SCF and outstanding spin/class rows.
Owner: #1169 scheduled campaign; #471 scoped receipts. Depends on: B rows.
UKS, MGGA, explicit controls¶
Selector: functional=..., chi_xc_controls with support_policy (direct) or coulomb_controls (RI); UKS, MGGA
Hamiltonian and approximations: Finite Born-von Karman Coulomb Hamiltonian on the complete unshifted character mesh; neutral cell, zero-mean background, BvK Ewald exchange zero mode (exchange_q0=’bvk-ewald’). Semilocal exchange-correlation energy on the declared cell grid and density-image support.
Domain: Neutral all-electron 3D, symmetry off; UKS MGGA consumes density, gradient invariants and kinetic density tau.
Production entry points: periodic/chi/scf.py::_run_chi_semilocal, periodic/chi/scf.py::run_aiccm2026dev_b_uks, periodic/chi/xc.py::functional_identity, periodic/chi/xc.py::_evaluate, periodic/chi/xc.py::_functional.
Admission guards: periodic/chi/xc.py::functional_identity, periodic_fitted_support.py::fitted_nuclear_image_check, periodic/chi/xc.py::_density.
Lean claims: chi_grid_rho_adjoint, chi_grid_rho_line, chi_grid_gradient_adjoint, ks_uks_gradient_flow, chi_grid_tau_adjoint, chi_semilocal_grid_deriv. Modules: ChiGrid.lean, KohnSham.lean.
Hypotheses and limits: Fixed complex AO and Cartesian derivative rows, fixed quadrature, Hermitian density variations. The AO gradient adjoint gives the two conjugate outer products. The separate local invariant derivative supplies the gradient coefficients. For positive kinetic density, tau contributes one half of the sum of derivative-row Gram matrices (Furness et al., DOI 10.1021/acs.jpclett.0c02405). The finite sum chain rule assembles all supplied field kernels into one potential per spin and character at the same density; the local derivative is supplied in density, Cartesian gradient and (if used) kinetic coordinates.
Checks: test_chi_xc.py. Tests: test_chi_xc.py, test_chi_direct_ks.py, test_chi_ks_reference.py, test_chi_nuclear_image_check.py.
Evidence:
#471: H2 r2SCAN UKS (3,1,1), fixed-density derivative and doubled-tau negative (591fd77f42fe, 562b013f). No consumer SCF qualification from this row.
Refused: External, double-hybrid, VV10 and Laplacian functionals; RIJCOSX; ECP.
Remaining gaps:
No theorem derives Libxc derivatives, AO rows, image selection or quadrature from the native inputs; finite-precision and physical accuracy remain open.
Keep density-image, partition, radial/angular grid and SCF convergence separate. See #1169 C471-1 through C471-9, including the fitted SCF and outstanding spin/class rows.
Owner: #1169 scheduled campaign; #471 scoped receipts. Depends on: B rows.
RKS/UKS through the functional-only path¶
Selector: functional=... without explicit chi controls
Hamiltonian and approximations: Finite Born-von Karman Coulomb Hamiltonian on the complete unshifted character mesh; neutral cell, zero-mean background, BvK Ewald exchange zero mode (exchange_q0=’bvk-ewald’). XC built by the shared multi-k drivers.
Domain: 3D; RI and RIJCOSX need at least two cells.
Production entry points: periodic/chi/scf.py::run_aiccm2026dev_b_rks, periodic/chi/scf.py::run_aiccm2026dev_b_uks.
Admission guards: periodic/chi/scf.py::_validate_closed_shell_problem, periodic/chi/scf.py::_validate_open_shell_problem.
Lean claims: none. Modules: none.
Hypotheses and limits: None.
Checks: none. Tests: test_chi_xc.py.
Refused: Range-separated functionals on RI and RIJCOSX.
Remaining gaps:
No executed-support record; not a candidate for qualification before the explicit path.
Owner: unowned. Depends on: ks-semilocal-explicit.
Hybrids¶
Global hybrids, explicit controls¶
Selector: functional=<hybrid>, explicit chi controls; A full Coulomb fraction, with zero short-range coefficient.
Hamiltonian and approximations: Finite Born-von Karman Coulomb Hamiltonian on the complete unshifted character mesh; neutral cell, zero-mean background, BvK Ewald exchange zero mode (exchange_q0=’bvk-ewald’). Exact exchange c_full/r + c_sr erfc(omega r)/r with kernel-specific zero modes.
Domain: One screening range; non-negative endpoint fractions; restricted and unrestricted. A full Coulomb fraction, with zero short-range coefficient.
Production entry points: periodic/chi/xc.py::exchange_assembly, periodic/chi/scf.py::_run_chi_semilocal.
Admission guards: periodic/chi/xc.py::exchange_assembly, periodic_fitted_support.py::fitted_nuclear_image_check.
Lean claims: bvk_seam_shifted_exchange_reciprocal, chi_hybrid_exchange_fock, bvk_uhf_closed_shell_ledger. Modules: ScreenedPotential.lean, SpinStationarity.lean.
Hypotheses and limits: Supplied reciprocal, real-linear Coulomb and exchange maps on identical finite character blocks, including the declared zero-mode terms. The weighted full and screened exchange contribution differentiates to minus the weighted exchange of each separate spin density. Equal spin densities give the closed-shell ledger. The semilocal grid term is separate.
Checks: check_aiccm_screened_potential.py, test_chi_hybrid_contract.py. Tests: test_chi_hybrid_contract.py, test_chi_hybrid_direct.py, test_chi_hybrid_fitted.py, test_chi_hybrid_native.py, test_chi_nuclear_image_check.py.
Evidence:
#480: direct Gamma RKS with 0.25 full exchange against a Gaussian oracle (2add4dd7c2f6, 493bbba2): Fock 7.3e-15 Ha, energy 1.1e-14 Ha. No complete hybrid SCF verdict.
Refused: Other kernels, more than one screening range, negative endpoint fractions.
Remaining gaps:
The weighted spin derivative does not construct native kernel tensors, establish complementarity of separately fitted kernels, match the functional decomposition or qualify a native SCF energy.
The acceptance rebuild adds no BvK zero-mode term on the exchange_builder path; the builder must own the kernel-specific zero mode. Not verified here.
#1169 C480-1 through C480-6: semilocal component, consumer SCF, fitted supports and matched convergence are separate gates.
Owner: #1169 scheduled campaign; #480 scoped receipts. Depends on: B and C rows.
Short-range screened hybrids, explicit controls¶
Selector: functional=<hybrid>, explicit chi controls; Short-range erfc exchange with zero long-distance fraction.
Hamiltonian and approximations: Finite Born-von Karman Coulomb Hamiltonian on the complete unshifted character mesh; neutral cell, zero-mean background, BvK Ewald exchange zero mode (exchange_q0=’bvk-ewald’). Exact exchange c_full/r + c_sr erfc(omega r)/r with kernel-specific zero modes.
Domain: One screening range; non-negative endpoint fractions; restricted and unrestricted. Short-range erfc exchange with zero long-distance fraction.
Production entry points: periodic/chi/xc.py::exchange_assembly, periodic/chi/scf.py::_run_chi_semilocal.
Admission guards: periodic/chi/xc.py::exchange_assembly, periodic_fitted_support.py::fitted_nuclear_image_check.
Lean claims: bvk_seam_shifted_exchange_reciprocal, chi_hybrid_exchange_fock, bvk_uhf_closed_shell_ledger. Modules: ScreenedPotential.lean, SpinStationarity.lean.
Hypotheses and limits: Supplied reciprocal, real-linear Coulomb and exchange maps on identical finite character blocks, including the declared zero-mode terms. The weighted full and screened exchange contribution differentiates to minus the weighted exchange of each separate spin density. Equal spin densities give the closed-shell ledger. The semilocal grid term is separate.
Checks: check_aiccm_screened_potential.py, test_chi_hybrid_contract.py. Tests: test_chi_hybrid_contract.py, test_chi_hybrid_direct.py, test_chi_hybrid_fitted.py, test_chi_hybrid_native.py, test_chi_nuclear_image_check.py.
Evidence:
#480: native primitive-s component suite (df16787b754b, source 3de6d12a), with separate erfc factors. Component evidence does not qualify a complete screened-hybrid consumer SCF.
#480: remaining direct HSE06 rows are C480-1 under #1169. A signed-CAM or global mixture is not HSE06.
Refused: Other kernels, more than one screening range, negative endpoint fractions.
Remaining gaps:
The weighted spin derivative does not construct native kernel tensors, establish complementarity of separately fitted kernels, match the functional decomposition or qualify a native SCF energy.
The acceptance rebuild adds no BvK zero-mode term on the exchange_builder path; the builder must own the kernel-specific zero mode. Not verified here.
#1169 C480-1 through C480-6: semilocal component, consumer SCF, fitted supports and matched convergence are separate gates.
Owner: #1169 scheduled campaign; #480 scoped receipts. Depends on: B and C rows.
Range-separated hybrids, explicit controls¶
Selector: functional=<hybrid>, explicit chi controls; Different short- and long-distance fractions, with a nonzero long-distance fraction.
Hamiltonian and approximations: Finite Born-von Karman Coulomb Hamiltonian on the complete unshifted character mesh; neutral cell, zero-mean background, BvK Ewald exchange zero mode (exchange_q0=’bvk-ewald’). Exact exchange c_full/r + c_sr erfc(omega r)/r with kernel-specific zero modes.
Domain: One screening range; non-negative endpoint fractions; restricted and unrestricted. Different short- and long-distance fractions, with a nonzero long-distance fraction.
Production entry points: periodic/chi/xc.py::exchange_assembly, periodic/chi/scf.py::_run_chi_semilocal.
Admission guards: periodic/chi/xc.py::exchange_assembly, periodic_fitted_support.py::fitted_nuclear_image_check.
Lean claims: bvk_seam_shifted_exchange_reciprocal, chi_hybrid_exchange_fock, bvk_uhf_closed_shell_ledger, chi_cam_operator_partition. Modules: ScreenedPotential.lean, SpinStationarity.lean.
Hypotheses and limits: Supplied reciprocal, real-linear Coulomb and exchange maps on identical finite character blocks, including the declared zero-mode terms. The weighted full and screened exchange contribution differentiates to minus the weighted exchange of each separate spin density. Equal spin densities give the closed-shell ledger. The semilocal grid term is separate. CAM (Yanai et al., Sec. 2, DOI 10.1016/j.cplett.2004.06.011): the full-plus-short and full-plus-long descriptions coincide only when full equals short plus long on the same operator domain; distinct nonlinear fits need not be complementary.
Checks: check_aiccm_screened_potential.py, test_chi_hybrid_contract.py. Tests: test_chi_hybrid_contract.py, test_chi_hybrid_direct.py, test_chi_hybrid_fitted.py, test_chi_hybrid_native.py, test_chi_nuclear_image_check.py.
Evidence:
#480: direct UKS signed CAM (0.65 full, -0.46 erfc at omega 0.4) on Gamma and (3,1,1), 749f287c4324 and 27e15d9924e6, source 9ce5b7d9. Covariance on 454e0540 (a028b848a2b3). Large-omega refinement applies only to its endpoint fixture. No complete hybrid SCF verdict.
Refused: Other kernels, more than one screening range, negative endpoint fractions.
Remaining gaps:
The weighted spin derivative does not construct native kernel tensors, establish complementarity of separately fitted kernels, match the functional decomposition or qualify a native SCF energy.
The acceptance rebuild adds no BvK zero-mode term on the exchange_builder path; the builder must own the kernel-specific zero mode. Not verified here.
#1169 C480-1 through C480-6: semilocal component, consumer SCF, fitted supports and matched convergence are separate gates.
Owner: #1169 scheduled campaign; #480 scoped receipts. Depends on: B and C rows.
Canonical correlation¶
RI-MP2 on a sealed classic RHF reference¶
Selector: correlation='mp2', scf_reference='rhf'
Hamiltonian and approximations: Finite Born-von Karman Coulomb Hamiltonian on the complete unshifted character mesh; neutral cell, zero-mean background, BvK Ewald exchange zero mode (exchange_q0=’bvk-ewald’). Classic physical-pair RI factors with the retained auxiliary metric projector. Same-source opposite-transfer fitted integrals and unshifted orbital-energy denominators.
Domain: Closed shell; optional frozen core on classic RHF; HF reference only.
Production entry points: periodic/chi/correlation.py::run_chi_mp2, periodic/chi/canonical_mp2.py::evaluate_same_source_mp2.
Admission guards: periodic/chi/correlation.py::require_public_correlation_request, periodic/chi/correlation.py::require_public_mp2_request, periodic/chi/correlation.py::require_public_mp2_controls, periodic/chi/correlation.py::validate_public_mp2_options, periodic/chi/canonical_mp2.py::_spin_state.
Lean claims: chi_mp2_energy, chi_fitted_integral, chi_transfer_sum, chi_cell_eri, chi_real_cell, chi_mesh_scale, chi_same_spin, chi_den_swap, bvk_uhf_retained_stationary_iff, bvk_retained_unprojected_counterexample. Modules: ChiMP2.lean, Amplitudes.lean, SeamSpectrum.lean.
Hypotheses and limits: Supplied factors and canonical orbitals; nonzero denominators; momentum conservation on the complete mesh.
Checks: check_chi_mp2.py, check_chi_mp2_contracts.py, check_chi_reduced_rank_mp2_native.py. Tests: test_chi_canonical_mp2.py, test_chi_mp2_checkpoint.py, test_chi_reduced_rank_reference.py.
Evidence:
Paper 2 formalization chi-mp2-results.json: H2 (3,1,1) same-fit PASS on v0.18.11.
Paper 2 packets 2026-11-09-chi-lih-rocksalt-mp2-retry-build-b3a828ec and 2026-11-10-chi-lih-rocksalt-n2-mp2-build-1fc985dc: LiH canonical rows.
#492 closed.
Paper 2 packet 2026-11-14-chi-reduced-rank-mp2-control-build-a4c86bcf: on a runtime before the projected admission, complete-rank MP2 equals an independent sum and the reduced-rank reference is refused as not stationary (control).
Paper 2 packet 2026-11-14-chi-reduced-rank-mp2-native-build-d54ee9fb: on a runtime containing the projected admission, the same frozen input admits the reduced-rank reference and production MP2 equals the independent sum within 3e-15 Ha per cell.
Refused: KS and hybrid references, SR frozen core, smearing, MOM, restart combined with optimize.
Remaining gaps:
External correlation accuracy is unmatched (the HF operator is unmatched).
Reduced-rank references are admitted on the projected canonical condition since #1194; the native evidence is one constructed Gamma-only closed-shell cell, with no truncated space on complex characters and no unrestricted case.
Owner: chi correlation. Depends on: hf-rhf-ri-classic.
UMP2 on a sealed classic UHF reference¶
Selector: correlation='mp2', scf_reference='uhf'
Hamiltonian and approximations: Finite Born-von Karman Coulomb Hamiltonian on the complete unshifted character mesh; neutral cell, zero-mean background, BvK Ewald exchange zero mode (exchange_q0=’bvk-ewald’). Classic physical-pair RI factors with the retained auxiliary metric projector.
Domain: Classic source only; frozen core must be zero.
Production entry points: periodic/chi/correlation.py::run_chi_mp2, periodic/chi/canonical_mp2.py::evaluate_same_source_mp2.
Admission guards: periodic/chi/correlation.py::require_public_mp2_request, periodic/chi/correlation.py::validate_public_mp2_options.
Lean claims: chi_same_spin, chi_den_swap, chi_fitted_integral. Modules: ChiMP2.lean.
Hypotheses and limits: Same-spin and transfer identities only; no theorem for the full spin-resolved energy.
Checks: none. Tests: test_chi_canonical_mp2.py.
Refused: SR UMP2, frozen core.
Remaining gaps:
No native UMP2 evidence; depends on a qualified UHF row.
Owner: chi correlation. Depends on: hf-uhf-ri-classic.
Knowles ROHF-RMP2¶
Selector: correlation='mp2', scf_reference='rohf'
Hamiltonian and approximations: Finite Born-von Karman Coulomb Hamiltonian on the complete unshifted character mesh; neutral cell, zero-mean background, BvK Ewald exchange zero mode (exchange_q0=’bvk-ewald’). Classic physical-pair RI factors with the retained auxiliary metric projector. Spin-block semicanonical partition with singles.
Domain: Classic source; shared frozen closed-core projector.
Production entry points: periodic/chi/canonical_rohf_mp2.py::evaluate_same_source_rohf_mp2, periodic/chi/canonical_rohf_mp2.py::_single_term.
Admission guards: periodic/chi/correlation.py::validate_public_mp2_options, periodic/chi/canonical_rohf_mp2.py::_semicanonical_channels.
Lean claims: chi_rmp2_single_square, chi_rmp2_single_minimum. Modules: ChiROHF.lean.
Hypotheses and limits: One supplied same-spin, same-character occupied-to-virtual coupling f with strictly positive semicanonical gap. The unique minimizer of the complex single-excitation quadratic is -f/gap and its value is -|f|^2/gap (Knowles et al., Sec. 2, Eqs. 15-16, DOI 10.1016/S0009-2614(91)85118-G). Hermitian spin Fock permits either orientation of the matrix element in the energy.
Checks: test_chi_rohf_mp2.py, check_chi_rohf_blocks.py. Tests: test_chi_rohf_mp2.py.
Evidence:
#523: source landed, validator acceptance pending.
Paper 2 LiH2 doublet control, source 23e96cc3, mesh (1,1,1), no frozen core: 16 MiB trace cap refused before integration. No ROHF energy, RMP2 singles or doubles; no physical verdict.
Supplied-Fock control: three complex-character blocks in a rectangular retained overlap frame; opposite-spin closed-virtual Fock couplings cancel at shared ROHF stationarity. The independently known nonzero spin singles nevertheless add. No native energy is implied.
Refused: SR ROHF, ROHF-CC.
Remaining gaps:
The singles variational identity assumes the spin partition and positive gap. No theorem derives the entire Knowles Hamiltonian, frozen core, spin-block eigenproblem, double-excitation couplings or native reference.
Owner: #523 validator. Depends on: hf-rohf-ri-classic.
CCSD and canonical-spin-orbital-hermitian-(T)/v1 on RHF¶
Selector: correlation='ccsd', scf_reference='rhf'; triples through ChiCCOptions.compute_triples
Hamiltonian and approximations: Finite Born-von Karman Coulomb Hamiltonian on the complete unshifted character mesh; neutral cell, zero-mean background, BvK Ewald exchange zero mode (exchange_q0=’bvk-ewald’). Classic physical-pair RI factors with the retained auxiliary metric projector. Projected similarity-transformed equations in packed spin orbitals; triples C* (C+D)/Delta over unique triples. The exchange constant enters the orbital energies and the integrals with matched indices (ChiCCOptions.coulomb_zero_mode=’matched-eri’, the default since the author decision of 2026-11-14; ‘hf-energy-only’ is the earlier default by name).
Domain: Classic full-Coulomb factors; equal frozen counts; converged energy and residual required.
Production entry points: periodic/chi/correlation.py::run_chi_ccsd, periodic/chi/canonical_cc.py::evaluate_same_source_cc.
Admission guards: periodic/chi/correlation.py::validate_public_cc_options, periodic/chi/correlation.py::require_public_correlation_request.
Lean claims: none. Modules: none.
Hypotheses and limits: None. No Lean statement of the projected CC equations, the residual or the triples contraction.
Checks: check_chi_cc_defaults.py. Tests: test_chi_cc_equations.py, test_chi_cc_checkpoint.py, test_chi_cc_zero_mode.py, test_chi_reduced_rank_reference.py.
Evidence:
#518: source landed, validator jobs and acceptance pending.
Paper 2 packet 2026-11-14-chi-cc-defaults-native-build-6d82354f: guarded native run of the defaults of #1181 (matched exchange constant on the canonical route; projected residual and spin-adapted pair density in the local recipe). Internal PASS against frozen gates on a second host; defaults equal the control’s named forms within 3.4e-14 Ha/cell.
Refused: ROHF, KS and hybrid references; near-singular denominators; unconverged continuation. The range-separated source is route correlation-rhf-sr.
Remaining gaps:
Independent determinant-space and nonzero-triples native receipts are not recorded in the Paper 2 catalog.
The complex conjugation in the triples expression is a project adaptation of the real formula.
Reduced-rank references are admitted on the projected canonical condition since #1194, on synthetic evidence only.
A reference with a reciprocal tail is refused, although canonical RHF MP2 admits it and the local routes accept it; no canonical CCSD(T) exists on the reference of the Paper 2 LiH ladder (#1248).
Owner: #518 validator. Depends on: mp2-rhf-classic.
UCCSD and the same named triples on UHF¶
Selector: correlation='ccsd', scf_reference='uhf'
Hamiltonian and approximations: Finite Born-von Karman Coulomb Hamiltonian on the complete unshifted character mesh; neutral cell, zero-mean background, BvK Ewald exchange zero mode (exchange_q0=’bvk-ewald’). Classic physical-pair RI factors with the retained auxiliary metric projector. The exchange constant enters the orbital energies and the integrals with matched indices (ChiCCOptions.coulomb_zero_mode=’matched-eri’, the default since the author decision of 2026-11-14; ‘hf-energy-only’ is the earlier default by name).
Domain: Classic source; independent alpha and beta frozen counts.
Production entry points: periodic/chi/correlation.py::run_chi_ccsd, periodic/chi/canonical_cc.py::evaluate_same_source_cc.
Admission guards: periodic/chi/correlation.py::validate_public_cc_options.
Lean claims: none. Modules: none.
Hypotheses and limits: None.
Checks: none. Tests: test_chi_cc_equations.py, test_chi_reduced_rank_reference.py.
Evidence:
#521: source landed, validator acceptance pending.
Refused: ROHF-CC, SR.
Remaining gaps:
Depends on a qualified UHF reference row.
Reduced-rank references are admitted on the projected canonical condition since #1194, on synthetic evidence only.
Owner: #521 validator. Depends on: hf-uhf-ri-classic.
RHF correlation on one range-separated fit¶
Selector: correlation in ('mp2', 'ccsd', 'dlpno-mp2') with scf_reference='rhf', fit_source='sr'; run_aiccm2026dev_b_ccsd(_t) and _dlpno_ccsd(_t) with reference=ChiHFReference and fit_source='sr'
Hamiltonian and approximations: Finite Born-von Karman Coulomb Hamiltonian on the complete unshifted character mesh; neutral cell, zero-mean background, BvK Ewald exchange zero mode (exchange_q0=’bvk-ewald’). One range-separated physical-pair RI fit (short-range real-space part, long-range reciprocal part; the split is numerical, the kernel is the full Coulomb kernel) serves the reference and every correlation step; its factors are in the canonical auxiliary frame. Canonical routes contract factors within one transfer; the local routes fold the same factors to the real torus. Exchange constant in the orbital energies only or matched in the integrals, as on the classic source.
Domain: Closed shell, all-electron, 3D, neutral; explicit ChiCoulombControls with auxiliary_cutoff_bohr and omega_bohr_inv; energy only.
Production entry points: periodic/chi/correlation.py::run_chi_mp2, periodic/chi/correlation.py::run_chi_ccsd, periodic/chi/correlation.py::run_chi_local_mp2_job, periodic/chi/posthf.py::run_aiccm2026dev_b_dlpno_ccsd_t.
Admission guards: periodic/chi/correlation.py::require_public_cc_admission, periodic/chi/correlation.py::require_public_local_mp2_admission, periodic/chi/cc_zero_mode.py::reference_madelung, periodic/chi/posthf.py::_adapt_explicit_cc_reference.
Lean claims: none. Modules: none.
Hypotheses and limits: None apply as executed. The chi MP2 claims are stated for one fitted tensor with supplied operators and do not distinguish how it was fitted; no claim covers CCSD or triples, and the reciprocity hypothesis of the stationarity theorems is not met on the recorded case of the range-separated reference route.
Checks: check_chi_sr_correlation.py. Tests: test_chi_sr_correlation.py, test_chi_canonical_mp2.py.
Evidence:
Source check on the shared native core (H2 and LiH, STO-3G, small meshes): on one sealed range-separated reference the canonical CCSD(T) and the local complete-domain CCSD(T) agree within 3e-14 Ha/cell in both exchange conventions, untruncated local MP2 equals canonical MP2 within 4e-10, the fold to the real torus has residuals below 2e-13, and the exchange constant recovered from the Fock equals the lattice value. No vq receipt yet.
Paper 2 packet 2026-11-15-chi-sr-correlation-control-build-a5641d5b: control record on a runtime before the admission (the range-separated coupled-cluster and local cases are refused there; the range-separated RHF and canonical MP2 and all classic rows are recorded) and frozen gates for the guarded run.
Paper 2 packets 2026-11-15-chi-sr-fit-ladder-lih-n1-build-d2eabad4, 2026-11-15-chi-sr-fit-lih-n2-build-d2eabad4 (preserved FAIL of the 28 bohr auxiliary cutoff) and 2026-11-15-chi-sr-fit-lih-n2-extension-build-d2eabad4: range-separated RHF and canonical MP2 on LiH/pob-DZVP-rev2 at meshes (1,1,1) and (2,2,2) through vq on a runtime before the admission. The auxiliary cutoff converges at 28 bohr at one character and at 36 bohr at mesh (2,2,2) (within 4e-11 Ha/cell of 44 to 60 bohr); the RHF stage is 17 and 13.9 times faster than with the classic fit. A measurement of cost and control convergence; it qualifies neither.
Paper 2 packet 2026-11-15-chi-sr-correlation-native-build-309ec106: guarded native run of the admission against the frozen gates, internal PASS (job 3b18e45f32ee, source 309ec1067). On one sealed range-separated reference (H2 and LiH, STO-3G) the local complete-domain CCSD(T) equals the canonical route within 3e-14 Ha/cell in both exchange conventions, untruncated local MP2 equals canonical MP2 within 4e-10, the fold residual is below 2e-13, and the control of the runtime before the admission repeats within 2.1e-13.
Refused: SR UCCSD(T) and SR UMP2 (unrestricted range-separated references). Persistent coupled-cluster and MP2 restart files on the range-separated source. A wrapper call whose fit_source does not name the executed source of its reference; the library’s own range-separated SCF as a post-HF source (fit_source=’sr’ or ‘auto’ without reference=). A range-separated fit whose metric has negative modes (the producer asks for converged cutoffs).
Remaining gaps:
The range-separated reference route itself has a recorded native failure and open acceptance (#453, route hf-rhf-ri-sr); correlation on it inherits that status.
The library default (fit_source=’auto’) still mixes a range-separated SCF with a classic correlation fit; one fit there is the next step of #1188.
No canonical-against-local comparison on a cell with d functions or a (2,2,2) mesh on this source.
The converged auxiliary cutoff depends on the mesh (28 bohr at one character, 36 bohr at (2,2,2) on LiH); no cutoff rule for nonzero transfers exists.
On LiH/pob-DZVP-rev2 the range-separated fit is below the classic fit by 1.1e-7 and 1.2e-6 (RHF) and by 1.07e-5 and 9.2e-6 (MP2) Ha/cell at meshes (1,1,1) and (2,2,2); the cause is not determined.
Owner: #1188. Depends on: hf-rhf-ri-sr.
Local correlation¶
Retained explicit-support CCSD(T) recipes (library)¶
Selector: run_aiccm2026dev_b_ccsd_t and DLPNO variants with reference=ChiHFReference
Hamiltonian and approximations: Finite Born-von Karman Coulomb Hamiltonian on the complete unshifted character mesh; neutral cell, zero-mean background, BvK Ewald exchange zero mode (exchange_q0=’bvk-ewald’). Classic physical-pair RI factors with the retained auxiliary metric projector. Real-torus inverse transform of the sealed factors; triples mode exact, local or t1 by option.
Domain: Untailed classic RI-RHF, all-electron, no frozen core.
Production entry points: periodic/chi/posthf.py::run_aiccm2026dev_b_ccsd_t, periodic/chi/posthf.py::run_aiccm2026dev_b_dlpno_ccsd_t.
Admission guards: periodic/chi/posthf.py::_require_posthf_source.
Lean claims: chi_real_cell, chi_inverse_entries. Modules: ChiMP2.lean.
Hypotheses and limits: Inverse-transform identities for supplied factors only. Canonical (T), localized T0 and iterative T1 are distinct and none is proved.
Checks: check_chi_retained_cc.py. Tests: test_chi_retained_cc_adapter.py, test_chi_retained_cc_recipe.py, test_chi_retained_cc_harness.py.
Evidence:
Paper 2 packet 2026-11-13-chi-cc-native-admission-build-1e6469a8: RHF converged, CC admission rejected the requested versus executed auxiliary label; A/B/C unexercised.
Paper 2 packet 2026-11-13-chi-cc-checkpoint-replay-build-a4c86bcf: auxiliary identity passed, chunk-backed eigenvalue provenance refused. The producer-preserving serialization repair landed at 3e4e9d803.
Paper 2 packet 2026-11-14-chi-cc-checkpoint-replay-build-ed1973b0 (article f799269): original checkpoint, no SCF; A/B/C energies and independent residuals met their limits, but the fixed-rotation shortcut missed its sensitivity floor. Complete-space qualification unexercised.
Refused: ECP electron accounting, frozen core, tailed sources, truncation through the canonical wrapper.
Remaining gaps:
Native complete-space acceptance, including the fixed-amplitude shortcut discriminator, belongs to the retained CC owner. Source and partial stage results do not qualify A/B/C or PNO truncation.
Owner: active CC admission owner. Depends on: complete frozen native gate ledger and owner verdict.
Library post-HF routes on their own SCF (default path)¶
Selector: run_aiccm2026dev_b_mp2, _dlpno_mp2, _ccsd(_t), _dlpno_ccsd(_t), _uccsd(_t) without reference=
Hamiltonian and approximations: Finite Born-von Karman Coulomb Hamiltonian on the complete unshifted character mesh, folded to the real torus. With fit_source=’auto’ the reference comes from the range-separated fit and the correlation integrals from a separately built classic fit (two fitted Hamiltonians, #1188); the classic support is the origin-centred index ball. The coupled-cluster recipe uses the projected residual (residual_domain=’pno’) and the pair density of the spin-adapted amplitudes (pno_norm=’mp2’) at tcut_pno=1e-7 (#1181, author decisions of 2026-11-14) and the default triples step in the complete virtual space. By default the exchange constant enters the orbital energies only (the ‘hf-energy-only’ convention). The restricted coupled-cluster entries take coulomb_zero_mode=’matched-eri’ by name: the amplitude equations use the occupied Fock without the constant and the (T) denominators keep the reference orbital energies (LocalCCSDOptions.triples_occupied_shift); the unrestricted local entries take the same keyword, each spin Fock with its own occupied projector (DLPNOUCCSDPilotOptions.triples_occupied_shift).
Domain: All-electron 3D, neutral, no frozen core on the localized route; RHF or UHF from the route’s own SCF.
Production entry points: periodic/chi/posthf.py::run_aiccm2026dev_b_dlpno_ccsd_t, periodic/chi/posthf.py::run_aiccm2026dev_b_mp2, periodic/chi/posthf.py::run_aiccm2026dev_b_dlpno_mp2, periodic/chi/posthf.py::run_aiccm2026dev_b_uccsd_t.
Admission guards: periodic/chi/posthf.py::_require_converged_cc, periodic/chi/posthf.py::_require_integer_gapped_reference, periodic/chi/posthf.py::_require_torus_domain_criterion, periodic/chi/posthf.py::_require_posthf_source.
Lean claims: none. Modules: none.
Hypotheses and limits: None apply as executed. The chi MP2 claims assume one fitted tensor, which the default two-fit path does not supply; no claim covers CCSD, triples or the bare pair density of the recipe.
Checks: check_chi_posthf_guards.py, check_chi_cc_defaults.py, check_chi_local_cc_zero_mode.py, check_chi_local_ucc_zero_mode.py. Tests: test_chi_posthf_guards.py, test_periodic_aiccm2026dev_b_posthf.py, test_chi_local_cc_zero_mode.py.
Evidence:
Paper 2 packet 2026-11-13-chi-recipe-decision-measurements-build-f7463a61: exploratory local values on the hydrogen fixture (pair minus pno CCSD -3.49e-08 Ha/cell; two fits minus one -8.4e-13 Ha/cell in MP2). Not vq receipts.
vq control job on the dev runtime at source c48994c268c6 (before the guards): an unconverged run was returned with a CCSD(T) label and E(T) = 0, and truncated DOI domains ran. The guarded run passed (Paper 2 packet 2026-11-14-chi-posthf-guards-native-build-d54ee9fb): the unconverged cases and the truncated DOI domains are refused and the converged energies are unchanged.
Paper 2 packet 2026-11-14-chi-local-cc-zero-mode-control-build-54f73150: control record and frozen gates for the matched exchange constant on the restricted local entries. On the changed source the complete-domain and zero-threshold local CCSD(T) reproduce the canonical matched route on one sealed reference within 7e-14 Ha/cell (source evidence).
Paper 2 packet 2026-11-15-chi-local-cc-zero-mode-native-build-d2eabad4: guarded native run of the matched exchange constant on the restricted local entries. Internal PASS against the frozen (amended) gates on a second host; the complete-domain local CCSD(T) equals the canonical matched route within 3e-15 Ha/cell.
Paper 2 packet 2026-11-15-chi-local-ucc-zero-mode-native-build-d2eabad4: guarded native run of the matched exchange constant on the unrestricted local entries. Internal PASS against the frozen gates; the complete-domain route equals a spin-orbital CCSD(T) with the constant in its integrals within 2.4e-12 Ha/cell (CCSD) and 3e-15 ((T)) on two open-shell hydrogen cells.
Paper 2 packet 2026-11-14-chi-cc-defaults-native-build-6d82354f: guarded native run of the defaults of #1181 (matched exchange constant on the canonical route; projected residual and spin-adapted pair density in the local recipe). Internal PASS against frozen gates on a second host; defaults equal the control’s named forms within 3.4e-14 Ha/cell.
Refused: Unconverged coupled-cluster amplitudes, and a triples label without an executed triples step (#1180). A reference without integer, character-independent occupations or without a gap above 1e-12 Ha (#1193). Truncated domains with a criterion other than Mulliken (#1184). Explicit SR correlation source, pair-distance screening, local fitting, frozen core on the localized route, 1D and 2D cells.
Remaining gaps:
The triples step of the recipe differs from Paper 2 Eqs. 26 and 28; by the author decision of 2026-11-14 the paper follows the code (open item O47). The residual domain follows the decision since 8e963cba3.
The default mixes two fits and does not evaluate the fit-response bound (#1188, open item O46). By the author decision of 2026-11-14 one range-separated fit serves both steps; that is available on sealed explicit references (route correlation-rhf-sr) and not yet on this default path.
Frozen core and ECPs are refused on the localized route (#1193).
Centroid aliasing and the minimum image of the shared local engine (#1184) are untouched; occupied-distance screening stays refused here.
The refusal of a reference without integer occupations or a gap (#1193) has synthetic tests only.
The matched exchange convention is a named option of the restricted and unrestricted local coupled-cluster entries and not yet their default (author decision of 2026-11-14 on the exchange constant).
With localized occupied orbitals the unrestricted local entries evaluate the (T) with the diagonal of the localized occupied Fock; at zero threshold their (T) is not the complete-domain (T) (#1246).
The explicit-reference path tests the route keyword, not the executed tail of the reference, for its untailed guard; a tailed sealed reference is accepted (#1248).
Owner: chi lane. Depends on: author decisions on O46 and O47.
Translation-local MP2 (front-door ‘dlpno-mp2’)¶
Selector: correlation='dlpno-mp2' with explicit ChiLocalMP2Options(tcut_mkn, tcut_pno)
Hamiltonian and approximations: Finite Born-von Karman Coulomb Hamiltonian on the complete unshifted character mesh; neutral cell, zero-mean background, BvK Ewald exchange zero mode (exchange_q0=’bvk-ewald’). Classic physical-pair RI factors with the retained auxiliary metric projector. Shared translation-native local MP2 on the sealed physical-pair factors, pno_norm=’mp2’.
Domain: RHF, classic, neutral 3D singlet; odd meshes for PNO truncation.
Production entry points: periodic/chi/correlation.py::run_chi_local_mp2_job, periodic/chi/local_mp2.py::run_chi_local_mp2.
Admission guards: periodic/chi/correlation.py::require_public_correlation_request.
Lean claims: chi_coupled_energy_bound, chi_coupled_tail_bound, chi_coupled_solvable. Modules: ChiCoupled.lean, ChiPairDensity.lean, ChiPNOTrace.lean, ChiPNOOptimum.lean, ChiPairEnergyBound.lean, ChiMargin.lean.
Hypotheses and limits: Supplied pair operators with a stability margin; bounds are for MP2 only and are not monotonic guarantees for combined truncations.
Checks: check_chi_coupled.py, check_chi_pair_density.py, check_chi_pno_trace.py, check_chi_pno_optimum.py, check_chi_pair_energy_bound.py. Tests: test_chi_translation_mp2.py, test_chi_dlpno_front_door.py.
Evidence:
Paper 2 packets 2026-11-10-chi-lih-rocksalt-n2-mp2-b-build-1fc985dc (full space equals canonical, PASS) and 2026-11-09 LiH rows: PNO 1e-6 missed its 1e-5 target, PNO 1e-7 met it.
#1109 front-door wiring; #572/#591 bounded cost measurement.
Refused: UHF/ROHF/KS references, non-classic sources, checkpoints, PNO truncation on even meshes. Reduced-rank references, by name.
Remaining gaps:
Single-axis domain, pair and PNO truncation rows on a nontrivial mesh are incomplete.
A complete vq calculation against the canonical same-source value and Paper 2 O10 is owed.
Owner: chi producer with shared DLPNO owners. Depends on: mp2-rhf-classic.
Translation-local CCSD and triples¶
Selector: correlation='dlpno-ccsd'
Hamiltonian and approximations: Not defined on the front door.
Domain: None.
Production entry points: periodic/chi/correlation.py::require_public_correlation_request.
Admission guards: periodic/chi/correlation.py::require_public_correlation_request.
Lean claims: none. Modules: none.
Hypotheses and limits: None.
Checks: none. Tests: test_chi_dlpno_front_door.py.
Refused: Refused before the SCF, naming the shared CC residual engine (#586) and triples (#587).
Remaining gaps:
Consumes shared interfaces when their owners deliver them; no chi work starts on #586.
Owner: shared DLPNO owners. Depends on: #586, #587.
Restart¶
Persistent HF reference restart¶
Selector: chi restart with a saved accepted reference
Hamiltonian and approximations: The saved state must rebuild the retained-source physical Fock and energy.
Domain: Classic RI RHF/UHF/ROHF in 3D, unsmeared, non-KS; all-electron v1 and scalar-ECP v2 records.
Production entry points: periodic/chi/checkpoint.py::save_chi_hf_reference, periodic/chi/checkpoint.py::load_chi_hf_reference.
Admission guards: periodic/chi/checkpoint.py::_validate_physical, periodic/chi/checkpoint.py::_validate_overlap_space.
Lean claims: bvk_uhf_half_fock_ledger. Modules: none.
Hypotheses and limits: The half-Fock energy form; tamper and mismatch rejection are tested, not proved.
Checks: none. Tests: test_chi_hf_checkpoint.py, test_chi_restart_public.py, test_chi_ecp_checkpoint.py, test_chi_reduced_rank_reference.py.
Evidence:
#524 closed with bounded evidence; Paper 2 packet 2026-11-10-chi-overlap-policy-fix-build-b8fc07dd (no native reduced-rank reference).
Refused: KS, SR, smeared states, restart combined with optimize, shared checkpoint paths.
Remaining gaps:
Reduced-rank states are validated on the projected conditions (#1194); no native reduced-rank handoff is recorded.
Owner: chi HF. Depends on: hf-rhf-ri-classic.
Derivatives¶
Analytic nuclear gradients and fixed-cell optimisation¶
Selector: chi_gradient=True or optimize=True on RHF, MP2 and CCSD(T) surfaces
Hamiltonian and approximations: Derivatives of the RI-RHF, RI-MP2 and CCSD(T) energies on explicit classic controls.
Domain: RHF only, RI, 3D, no dispersion, tight SCF thresholds.
Production entry points: periodic/chi/optimize.py::require_chi_derivative_request, periodic/chi/gradient.py::compute_aiccm2026dev_b_gradient, periodic/chi/mp2_gradient.py::compute_aiccm2026dev_b_mp2_gradient, periodic/chi/cc_gradient.py::compute_aiccm2026dev_b_cc_gradient.
Admission guards: periodic/chi/optimize.py::require_chi_derivative_request.
Lean claims: bvk_rhf_fock_iff, bvk_fock_stationary_iff, bvk_quadratic_fock, bvk_orbital_stationary_iff. Modules: FitResponse.lean.
Hypotheses and limits: Stationarity is a hypothesis of the gradient derivation. Pulay, grid, support, metric and rank-boundary derivative terms are not proved.
Checks: none. Tests: test_chi_ri_rhf_gradient.py, test_chi_mp2_gradient.py, test_chi_cc_gradient.py, test_chi_front_door_gradients.py.
Evidence:
#1106, #1108, #1139 under #1101: finite-difference checks by their owners.
Refused: Cell optimisation, UHF/ROHF/KS, ECP, MANTA, other backends, translation-local MP2. Reduced-rank references, by name: rank-boundary derivative terms are not implemented.
Remaining gaps:
Single-point evidence does not cover derivatives; stress and response are not advertised.
Owner: #1101 coordinator. Depends on: B and E rows.
ECP¶
Scalar ECP Hamiltonians¶
Selector: ECP basis on classic RI RHF/UHF/ROHF with physical_pairs controls
Hamiltonian and approximations: Finite Born-von Karman Coulomb Hamiltonian on the complete unshifted character mesh; neutral cell, zero-mean background, BvK Ewald exchange zero mode (exchange_q0=’bvk-ewald’). Classic physical-pair RI factors with the retained auxiliary metric projector. Semilocal scalar ECP added to the one-electron operator; valence electron count.
Domain: Neutral 3D; canonical and local MP2 and CC accept ECP references.
Production entry points: periodic/chi/ecp.py::ecp_contract, periodic/chi/ecp_hamiltonian.py::require_ecp_support.
Admission guards: periodic/chi/ecp.py::validate_ecp_reference.
Lean claims: none. Modules: none.
Hypotheses and limits: None.
Checks: none. Tests: test_chi_ecp_public.py, test_chi_ecp_hamiltonian.py, test_chi_ecp_operator.py.
Evidence:
#530 closed with bounded evidence.
Refused: Spin-orbit ECPs, SR, KS, direct support, retained CC, gradients, MANTA.
Remaining gaps:
ECP evidence does not transfer between sources.
Owner: chi HF. Depends on: hf-rhf-ri-classic.