Right-module base change along the diagonal quotient

Let k be a field, S a finite lattice, A the interval-pair incidence algebra, I its diagonal ideal, and B = A/I. Put R = Aᵐᵒᵖ, C = Bᵐᵒᵖ, and let f : R →+* C be the opposite of the actual quotient projection. Modules in ModuleCat R and ModuleCat C are right modules over A and B, respectively. Write Res for restriction along f and E for the quotient-extension functor extensionFunctor.

For M : ModuleCat R, form the R-submodule

K(M) = span_R { a • m | a ∈ ker f, m ∈ M }.

The quotient M/K(M) is annihilated by ker f. Factoring its actual R-action through the surjection f therefore equips it with a C-action. This defines extensionFunctor; its public class and map formulas identify the C-action with the original action after applying f.

The functor is left adjoint to restriction along f. The adjunction is constructed from the actual hom-set equivalence: an R-linear map into a restricted C-module kills K(M) and factors uniquely through M/K(M). Restriction preserves epimorphisms because its underlying maps remain surjective. The adjunction criterion consequently proves that extension preserves projectives; its left-adjoint property proves preservation of finite colimits, hence right exactness. No preservation of arbitrary monomorphisms or short exact rows is claimed.

For every N : ModuleCat C, the natural counit isomorphism extensionFunctor (restrictScalars N) ≅ N sends the class of n to n and has inverse n ↦ [n]. The actual regular right module R extends to the regular right module C by the quotient map. For each strict vertex v, restriction of the existing quotient vertex module gives an actual R-module; its action is the pulled-back character, and it is simple because f is surjective. Its extension is the original quotient vertex module by the counit. The canonical comparison ((extensionFunctor.leftDerived 0).obj (restrictScalars Vᵥ)) ≅ Vᵥ is C-linear.

The kernel J = ker f is idempotent: quotientKernel_mul_self proves J * J = J from the actual diagonal ideal and its idempotent generator. This identity shows that E carries every short exact row with restricted quotient-module cokernel to a short exact row. The restriction on the cokernel is essential to this argument.

For every quotient module N and every projective resolution P of Res(N), extensionMappedResolution_exactAt_one proves that E(P) is exact at degree one. Hence extension_leftDerived_one_isZero gives the actual categorical vanishing IsZero ((E.leftDerived 1).obj (Res(N))). The theorem rightVertexExtension_leftDerived_one_isZero specializes this result to each strict vertex. The previous degree-zero comparison remains the canonical C-linear isomorphism ((E.leftDerived 0).obj (Res(Vᵥ))) ≅ Vᵥ.

For an interval pair q = (a,b) with a,b ∈ WithTop S and a ≤ b (so either endpoint may be ⊤), put d(q) = (a,a), and write δ_{p,r} for the actual incidence matrix unit at a comparable pair p ≤ r of interval-pair indices. For a right A-module M, define the additive group G(M) = ∏_q {m ∈ M | m · δ_{d(q),d(q)} = m} and the assembly τ_M((m_q)) = Σ_q m_q · δ_{d(q),q}, with both product and sum indexed by all interval pairs. The functor G is exact: project a preimage onto the fixed subgroup in each coordinate. The image of τ_M is the actual quotient relation submodule, and τ_M is injective for projective M, by regular-module columns, arbitrary free modules, and the projective splitting. The higher-degree cycle lift uses these exactness, image, and injectivity properties.

For every right-incidence module M, every actual projective resolution P of M, and every n ≥ 2, extensionMappedResolution_exactAt proves that E(P) is exact at degree n. Consequently, extension_leftDerived_isZero_atLeastTwo proves the actual categorical statement IsZero ((E.leftDerived n).obj M) in that range. For a restricted quotient module Res(N), this higher result combines with the degree-one theorem to give exactness of every positive degree and vanishing of every positive left-derived value. The same conclusion specializes to each strict vertex.

The range is deliberately positive: on restricted modules the unaugmented mapped resolution has the existing degree-zero comparison with N, which may be nonzero. No degree-one result for an arbitrary right-incidence module, exactness at degree zero, numerical dimension bound, 2n−2 bound, or stability statement is established here.

For every actual projective resolution P of Res(N), extensionProjectiveResolution N P is an actual projective resolution of N. Its complex is literally E(P), and its augmentation is E(P.π.f 0) followed by the counit. The public computation laws identify its terms, differentials, and representative-class augmentation; any supplied termwise zero-above-degree condition on P also transfers to this augmented resolution. The unaugmented complex E(P) still need not be exact at degree zero.

For every d : ℕ, extension_hasProjectiveDimensionLE_restricted transfers an actual bound HasProjectiveDimensionLE (Res(N)) d to N. The extended-natural comparison projectiveDimension_le_restriction also covers zero objects and infinite projective dimension. Its proof uses the diagonal-column assembly criterion: extension carries a short exact row to a short exact row when assembly is injective on the right endpoint, and assembly is injective on the restriction of every quotient module because its fixed-column group is zero. The induction uses ordinary projective epimorphisms and dimension shifting; it does not require minimal or essential projective covers. This is a right-module projective-dimension comparison, not a numerical bound from the incidence algebra or a 2n−2 persistence estimate.

The quotient extension is the right-module base-change construction in BOOS24-v3, §2.4, Definition 5.4 and Notation 5.9. Its idempotent-kernel factorization formalizes the degree-one mechanism for the restricted-cokernel row. Separately, the actual diagonal-column construction gives the higher-degree projective-ideal/column mechanism of Lemma 5.6 for this interval-pair quotient. The underived vertex comparison and projective-dimension comparison specialize the right-module setup of Lemma 5.7; the finite-length and numerical conclusions of that source are not asserted here.

The separate vertex-tensor vanishing in the resolution and derived comparison uses the other derived variable and is not a premise here; no derived tensor comparison is asserted. The Bool and singleton examples exercise the zero and regular modules, projectivity, vertex and degree-zero comparisons, the counit class formula, naturality for a nonidentity endomorphism, and the degree-one and degree-two consumers. The projective-resolution examples also check bottom projective dimension, transport at the projective boundary, the augmented class formula, and supplied termwise support bounds on an actual strict-vertex resolution.