vibe-qcDocumentation

Finite identities of the correlated methods on the torus

AICCM/CCM remains experimental. Three Lean files add finite algebraic statements for the correlated methods of the two AICCM papers. They follow the author decision of 2026-11-14 on the scope of the proof claim: the finite identities of coupled cluster are formalised, and truncation errors of coupled cluster stay numerical. No quantum-chemistry calculation is involved.

Coupled cluster: VibeQCAICCM/CCFinite.lean (26 claims)

The setting is abstract. Det is a finite index set, the determinants of the N-electron space. H is a matrix on it, o the reference index, and the wave operator S is any matrix with a two-sided inverse Sinv. Second quantisation, the exponential form S = exp(T), the truncation to singles and doubles and the triples correction are not formalised. Every statement therefore holds for any invertible wave operator.

Definitions: ccBar = Sinv * H * S, ccEnergy = ccBar o o, ccResidual mu = ccBar mu o, and ccComplement o, the projector onto all mu different from o.

Claims

Statement

cc_exact_eigen, cc_state_nonzero, cc_energy_eigenvalue, cc_eigen_solves

If every residual off the reference vanishes, S e_o is a nonzero eigenvector of H with eigenvalue ccEnergy, and conversely. This is the full-configuration limit of the complete excitation manifold.

cc_shift_bar, cc_shift_energy, cc_shift_invariant

Adding a constant z on the N-electron space shifts the energy by z and leaves every residual unchanged. The exchange constant in its matched form is such a constant.

cc_rotation_bar, cc_rotation_energy, cc_rotation_residual, cc_rotation_projected

Under an invertible rotation U that fixes the reference, the energy is unchanged, the residual vector becomes U r, and projected equations with a projector that commutes with U are the same equations.

cc_projection_weaker, cc_projection_complete, cc_complement_apply

Equations on a smaller manifold follow from those on a larger one. With the projector onto everything but the reference the projected equations are the full equations: the zero-threshold limit.

cc_union_contains, cc_union_projector_product, cc_union_projector_absorb, cc_union_projected_weaker

A pair or triple space spanned inside a metric-orthonormal union space is left unchanged by the union projector, and equations projected on the union imply those projected on the contained space.

cc_energy_projected, cc_energy_same_state, cc_intertwine, cc_bar_reference, cc_residual_column, cc_energy_component, cc_invertible_kernel, cc_rotation_inverse

Supporting identities.

Correspondence: Paper 1 eq:ccsd and its zero-threshold paragraph, Paper 2 Eqs. dlpno_r1, dlpno_r2, ccsd_energy and the complete-domain paragraph. Not covered: that orbital rotations inside the occupied and virtual spaces induce a U of this kind, the singles-and-doubles truncation, existence or convergence of a solution, the (T) energy, and any error bound for a truncated or local manifold.

MP2 under rotations inside degenerate levels: VibeQCAICCM/Mp2Rotation.lean (11 claims)

The canonical orbitals of the real supercell are unitary mixtures, inside degenerate levels, of the Bloch orbitals. rot_mp2_invariant proves that the closed-shell MP2 energy

sum_{iajb} conj(g_iajb) (2 g_iajb - g_ibja) / (e_i + e_j - e_a - e_b)

is unchanged when the integrals are rotated by a unitary A on the occupied indices and a unitary B on the virtual indices, each block-diagonal with respect to the orbital energies. The proof flattens the integrals to a vector, writes the energy as a sesquilinear form with the exchange operator 2 - P and the diagonal of inverse denominators, and shows that the rotation is unitary and commutes with both (rot_matrix_unitary, rot_matrix_weight_commute, rot_matrix_exchange_commute, rot_sesquilinear_invariant). No hypothesis of nonzero denominators is needed.

Correspondence: Paper 1 Section 4.1, the statement that the real-torus MP2 energy equals the character-mesh MP2 energy. Together with the per-pair character identities of ChiMP2.lean this closes the finite part of that statement. Not covered: that the two orbital sets are related by such a rotation for a given calculation, which is a property of the SCF solutions.

Active-space energy and the exchange constant: VibeQCAICCM/ActiveSpace.lean (9 claims)

actEnergy h Wc Wa Dc Da is the core energy, plus the active one-body energy with the inactive Fock matrix built with the kernel Wc, plus the active two-body energy with the kernel Wa, for a core density Dc and an active density Da in the single-determinant limit of the active part.

Claim

Statement

act_split, act_split_energy

With one kernel the Hartree-Fock energy of Dc + Da equals actEnergy.

act_empty

An empty active space gives the Hartree-Fock energy of the core density.

act_matched

With the fold tensor and the exchange constant in both kernels, the total energy of Dc + Da is actEnergy plus the nuclear term.

act_unmatched_defect

If the inactive Fock matrix carries the exchange constant and the active integrals do not, the energy misses xi/4 Da . (S x S) Da.

act_defect_idempotent, act_defect_electrons, act_unmatched_idempotent, act_unmatched_fold

For a closed-shell active density, A S A = 2 A, that defect is xi/2 times the number of active electrons.

Correspondence: the matched convention of the exchange constant decided for coupled cluster and for the active-space Hamiltonian (code issues 1182 and 1201). Not covered: a correlated active wavefunction, whose two-body energy is not a quadratic form in a one-particle density.

Check

python formal/check_proofs.py with the pinned Lake. The allowed axioms are propext, Classical.choice and Quot.sound.