vibe-qcDocumentation

Complete three-centre image domain and reciprocal pair density

This continuation constructs the absolute convergence missing from the single-image analytic fitting contract. It formalizes full-space localized Gaussian integrals. Identification with a primitive-cell Bloch pairing also requires a valid tiling/unfolding theorem, phase convention and cell normalization; those are not silently inferred from the localized formula. The physical motivation is Ye and Berkelbach (2021), DOI 10.1063/5.0046617, Eqs. 11-13 and SI Section S4, especially S6-S17, read locally. Their two AO-image labels and relative-image Fourier sum motivate the domains here. We do not claim to verify every approximation or unit-cell convention in that reference.

Definitions and explicit hypotheses

Coordinates and exponents use the atomic-unit convention of the shell-image bounds. Write q_eta(z)=erfc(eta|z|)/|z|, eta>0, and L:Z³ -> R³ for an additive lattice map satisfying |L(n)|>=sigma |n|, sigma>0. Let h,f,g be continuous complex functions with finite centred Gaussian envelopes

|h(x)| <= E exp(-(a/2)|x-O|²),
|f(y)| <= C exp(-a|y-P|²),
|g(y)| <= D exp(-a|y-Q|²),

where a>0 and E,C,D>=0. A sufficiently small common positive exponent can satisfy these conditions for finite contracted positive-exponent AOs and auxiliary functions. Their stored normalization and all finite signed coefficients belong in the envelope constants. No unit charge is assumed. Choose 0<c<=min(a/12,eta²/3).

The mathematical convention in threeTerm is

J(m,n) = integral dx dy h(x) conjugate(f(y-L(m)) g(y-L(n))) q_eta(x-y).

For the physical bra-conjugated auxiliary and ordered AO pair, instantiate h=conjugate(auxiliary), f=phi_mu, g=conjugate(phi_nu). The reciprocal pair is then conjugate(f) conjugate(g)=conjugate(phi_mu) phi_nu. This explicit instantiation avoids silently dropping complex conjugation. The real s-shell native fixtures test image and phase contractions, not general complex AO kernels.

Joint absolute convergence

Let r=n-m, U(s,t)=m(s)[2 pi+m(t)], m(s)=(pi/s)^(3/2), and K=e+2/sqrt(pi). The unit-bohr contact split in U has the same numerical atomic-unit meaning as in the preceding supplement. The overlap bound supplies exp(-a|L(r)+Q-P|²/4) and leaves a Gaussian of exponent a/2 at P+L(m). Applying the screened two-centre bound to that Gaussian and h supplies the independent base-image decay:

integral |threeTerm(m,m+r)|
  <= M exp(-c|L(m)+P-O|²) exp(-(a/4)|L(r)+Q-P|²),
M = E C D K U(a/4,a/4).

Both displaced lattice Gaussians are summable by the lattice lower bound. Their product is a summable majorant on Z³ x Z³. This constructs absolute six-dimensional integrability for every term and absolute summability of all image-integral norms. It also proves convergence of the complex integral series, allows either iterated summation order and permits the sum/integral interchange. Bounded coefficients |w(m,n)|<=1 preserve these results. This is not an assumption of convergence based on a small finite numerical remainder.

Bijection, phases and finite screens

The bijection (m,n) <-> (m,r=n-m) has inverse (m,r) -> (m,m+r). For positive Bloch characters chi_k(n)=exp(i k.L(n)), the proof derives unit modulus, multiplicativity and

conjugate(chi_k(m)) chi_l(m+r) = chi_(l-k)(m) chi_l(r).

At k=l the base-image transfer character is one. With x and y both translated by L(m), the full-space integral satisfies

J(m,m+r) = integral dx dy h(x+L(m))
                conjugate(f(y) g(y-L(r))) q_eta(x-y).

Thus the second independent image family can be evaluated as translated auxiliary panels. Fixing that auxiliary panel to m=0 is not the same sum. The finite absolute domain S x T transforms to {(m,r):m in S and m+r in T}. Replacing it by S x T again changes the selected terms. The finite-domain theorem retains this exact transported predicate, including boundary ties. Floating selector uncertainty still requires the earlier symmetric-difference budget.

For one relative AO pair b_r(y)=f(y)g(y-L(r)), the weighted analytic base-image sum is also the product-space screened-kernel pairing

sum_m w(m) integral h(x) conjugate(b_r(y-L(m))) q_eta(x-y)
  = integral h(x) conjugate(b_r(y)) sum_m w(m) q_eta(x-y-L(m)).

three_transfer_kernel proves this identity with the absolute-integral summability constructed above. In the physical character case w=chi_(l-k). The Fourier-space Ewald interpretation additionally needs the matched angular dual/transfer, physical zero-mode exclusion and the prescribed background/seam. The current theorem does not repair a mismatched SR/LR producer or prove primitive-cell unfolding.

Reciprocal AO-pair coefficient and its truncation

For any |w(r)|<=1 and Fourier test vector K, the overlap envelope gives an absolute mass at most

C D m(a/2) exp(-(a/4)|L(r)+Q-P|²).

The Fourier plane wave has unit modulus. Absolute summability therefore proves

integral [sum_r w(r) f(y) g(y-L(r))] exp(-i K.y) dy
  = sum_r w(r) integral f(y) g(y-L(r)) exp(-i K.y) dy.

The Lean eriPlane has positive sign; use its argument -K for the native negative Fourier sign. Use conjugate(f),conjugate(g) for the physical pair specified above, and w=chi_l for the positive ket-image phase. The same identity holds for any finite list with exactly the same selected labels. For a complete closed selector |L(r)+Q-P|<=R, the omitted coefficient has the previous shellTail bound with M=C D m(a/2), c=a/4, H=|Q-P| and the lattice lower bound sigma. No reciprocal-vector magnitude is needed in this conservative bound. A home-cell transform contains only r=0 and does not supply the full pair coefficient.

Kernel-checked statements

VibeQCAICCM/ThreePeriodic.lean adds 27 audited claims, bringing the project to 602 claims, including 520 AICCM claims. Dependencies and the transitive axiom policy are unchanged.

Statements

Obligation

three_image_bijection, three_domain, three_finite_reindex, three_reindex

Complete and transported finite image domains

three_joint_bound, three_gaussian_sum, three_l1_sum, three_images_absolute

Constructed joint Gaussian majorant and absolute summability

three_raw_absolute, three_integrals_summable, three_absolute_interchange, three_weighted_interchange, three_sum_orders

Absolute-coordinate convergence, interchanges and both iterated orders

three_phase, three_character_norm, three_character_add, three_transfer_character, three_gamma_character

Bloch phase and transfer convention

three_auxiliary_unfold, three_finite_unfold, three_periodic_unfold, three_transfer_kernel

Translated auxiliary panels and screened transfer-kernel identity

three_pair_mass, three_pair_fourier_mass, three_pair_fourier, three_pair_fourier_finite, three_pair_fourier_tail

Reciprocal pair coefficient, finite correspondence and image tail

The series identities must be used with the separately constructed summability hypotheses: Lean totalizes an undefined tsum, and a bare reindexing identity alone is not a convergence theorem.

Implementation bridge and remaining obligations

The checked native SR routine still fixes the auxiliary shell and first AO shell and translates only the second AO shell. Summing separately captured auxiliary panels is a verification construction for the new domain, not a production implementation change. The native Bloch AO-pair Fourier helper ao_pair_fourier_transform_bloch_ss, also used by ao_pair_fourier_transform_bloch, does include the supplied relative images and positive ket phase. In contrast, standalone rsgdf_lr_3c_tensor calls the home-cell transform. Other fitting routes have different callers and must be checked on their own evidence.

formal/capture_aiccm_three_periodic.py records native panels and Fourier values. The independent standard-library checker uses Gaussian product and screened convolution formulas, transported integer labels and complex phases. It reuses the preceding primitive convolution helper, whose radial checks were retained in the guarded-ERI tranche. Controls omit auxiliary images, replace the transported domain with a new rectangle, choose the wrong relative phase, omit relative Fourier images or reverse the Fourier sign. A finite panel refinement is recorded as a difference, not an infinite-domain certificate. No fitted RHF/MP2 energy or ORCA/CRYSTAL acceptance is claimed for these integral fixtures.

Remaining work includes primitive-cell tiling/unfolding and its normalization, matching the actual periodic SR/LR producer domains, certified selection and numerical evaluation, and propagation through whitening under its metric stability hypotheses. The earlier finite-fit RHF/MP2 bounds remain conditional on these contracts. This continuation does not establish complete formal verification of either production engine.

Executed implementation checks

The qualified 0.18.11 runtime (source 5c89075026d59eb6b8235554594d9ae4f1831f2a, native SHA-256 499575ab1782406d9ab85a664b74a368a7f300feb8a40c746fadcbb3266bd0d5) completed the capture through vq on 2026-11-12 UTC. Independent Gaussian formulas agree with 608 native short-range entries within 1.34e-15 (tolerance 3e-11), and 36 complex Bloch Fourier entries within 4.45e-16 (tolerance 3e-12). Native auxiliary panels symmetrize the two AO directions; the reference explicitly reconstructs that operation. The maximum raw finite-panel directional difference is 4.53e-7. Increasing the auxiliary cube from 27 to 125 panels changes its off-diagonal sum by 1.655e-4; this difference is not an infinite-tail certificate.

The transported finite double sum agrees within 2.23e-16. Controls that omit auxiliary images, use a fresh rectangle after reindexing, or use the wrong relative phase change the tested contraction by 0.1981, 9.568e-5 and 0.08763, respectively. Omitting relative Fourier images, conjugating the ket phase and reversing the Fourier sign give discrepancies 0.1301, 0.1438 and 0.9404. These are ordinary floating-point checks, not certified roundoff enclosures. Captured operands, measured results and all helper hashes are retained in both paper source archives. Replay with:

python3 formal/check_aiccm_three_periodic.py three-periodic-input.json replay.json