vibe-qcDocumentation

Finite-fixture occupied-ladder singles dressings

The experimental linear_occupied_ladder_terms action qualifies the two ooov singles dressings at fixed doubles on the accepted four-band BIPOLE (3,1,1) fixture. The complete occupied-family action now combines these terms with the separately exposed oooo terms, cubic dressings and full half expression. CCSD iteration remains unavailable.

For target pair \(p=(ij)\), define the full diagonal input \(d_i=C_{ii}t_i\) and the transported ordered double

\[B_{mn}^{(p)}=(C_p^T S C_{mn})T_{mn}(C_p^T S C_{mn})^T.\]

Every source double uses its own pair frame and the AO overlap \(S\). The two selected contributions are

\[X_p=\sum_{mn}B_{mn}^{(p)}(m i|n d_j),\qquad Y_p=\sum_{mn}B_{mn}^{(p)}(m d_i|n j).\]

Both coefficients are +1. All 36 ordered occupied pairs participate: six diagonal pairs once and thirty off-diagonal ordered pairs once. No half, exchange, orbit or cell multiplier enters. The contracted singles retain their original diagonal frames; projecting them into \(C_p\) before the integral contraction changes the equations. Pair reversal exchanges X and Y and transposes them. Individual diagonal X/Y matrices need not be symmetric. In complex Bloch coordinates the first and third integral columns are conjugated, the second and fourth are not.

In the shared cpp/src/dlpno_ccsd.cpp occupied block these are Boo_i.transpose()*Wj and (Boo_j.transpose()*Wi).transpose() contracted with tproj. The full native block contracts them with \(\tau_{mn}^{(p)}=B_{mn}^{(p)}+s_ms_n^T\), where \(s_m=C_p^TSC_{mm}t_m\). The additional cubic terms containing \(s_ms_n^T\) are outside this action. Under common double scaling \(\alpha\) and separate home singles scaling \(\beta_i\), X scales as \(\alpha\beta_j\) and Y as \(\alpha\beta_i\). Both vanish at zero singles or zero doubles.

Stanton et al., Sec. II, Eqs. (2), (6), (10) give the parent residual, occupied intermediate and tau. The spatial coefficients above follow the shared closed-shell spin integration. Riplinger and Neese, Sec. II C give the pair overlap construction, and McClain et al., Sec. 4, Eqs. (23)-(24) give the periodic integral convention. Signed finite BIPOLE panels replace auxiliary factors without a positive-Gram assumption.

Provider and resource contract

vibeqc.dlpno.bipole_occupied_singles consumes the existing result-owned terminal generation and authenticated pair build. Its seven input doubles and two real home-diagonal singles use the accepted unequal domains and rank-two/rank-three PNOs. The x, y, total results are immutable; placed(spaces,i,j) applies the exchanged-leg reversal rule. occupied_mixed_transition exposes a normalized (m* i|n* e) witness and returns zero outside its conserved momentum sector.

One exclusive action lease visits 256 domains and 6,912 normalized elements. Two seven-by-six-by-six occupied intermediates are accumulated before streaming one projected source-pair double at a time. Production does not allocate dense AO ERIs or a target-by-source inventory of projected doubles. resource_census() counts live input/copy arrays, AO/Bloch singles, occupied columns, scalar weights, projection products, accumulators, immutable copies and retained qualified contributions. The numerical census fits a 64 KiB subreservation of the existing 1 MiB transform/output slot. The total reservation is 42,991,616 bytes (8 MiB domain, 32 MiB panel, 1 MiB transforms). This is not a process-RSS bound. Work includes setup, each domain, native elements and 65,536 consumer arithmetic units per element, for 231,276,511,232 units. Exact limits pass; one byte or work unit below refuses before contraction. Signed transfers, omitted G=0 and finite erfc subtraction remain unchanged.

Qualification and assembly boundary

The executable regression builds its independent dense AO-supercell oracle before invoking the production action. Independent placed PAO/PNO bases and metric rotations carry each double and single from its own frame. Each term and sum is checked at every component of all 36 ordered targets (174 components), with a predeclared absolute gate of 1e-10 Ha. It checks zero amplitudes, double scaling, separate and joint home-singles scaling, exchanged-leg reversal, stale state/generations, changed frames, and resource boundaries. Premature target projection, omitted source transpose, incorrect half weight and missing reversal leg exchange must exceed the gate. Normalized ooov witnesses cover each signed transfer; all 864 forbidden band components pass a 1e-12 Ha gate.

The complete native occupied increment is occupied_ladder - bare. The harness derives the bare oooo*T2, oooo*ss, cubic ooov*ss, and full half-weighted pieces independently to isolate the selected linear terms. It also checks the entire occupied oracle against the actual shared evaluator through a signed spectral resolution of the AO operator. That reference-only check does not qualify the other production pieces. Keeping the nonzero bare snapshot is a deliberately failing boundary control. External molecular total energies are not matched references for this isolated finite-source tensor.

The partial production assembly adds X and Y to the already qualified full half-weighted expression, whose algebra also occurs in the occupied block. Both actions run sequentially at their own caps; no fused pass is claimed. The full virtual ladder remains separately qualified and must not be added inside the occupied snapshot.

The subsequent cubic occupied action qualifies both ooov*ss dressings using this same authenticated provider, including their separate amplitude projections, cubic scaling controls and native isolation. Its expanded partial sum adds those two cubic terms to X, Y and the full half expression above. All 36 ordered targets and 174 components pass at 1e-10 Ha; the expanded partial-sum maximum error is 9.1681e-16 Ha. This action’s X, Y, linear sum, half and earlier partial outputs are exactly unchanged. The complete occupied-family qualification adds oooo*T2 and oooo*ss, preserves these earlier arrays exactly and passes the isolated native occupied-minus-bare snapshot. No occupied term remains unqualified on this finite fixture. R2 Fock dressings, rings, R1, iteration, triples and materials acceptance remain open.