Ordinary projective dimension on finite lattices

Let k be an arbitrary field and S a finite bounded lattice. Write D(S) = FiniteInjectiveRankDimension.maxComplementCard (S := S), the maximum number of minimal elements outside a principal downset. The module TDA.Persistence.OrdinaryFiniteLatticeDimension proves that every actual pointwise finite-dimensional persistence functor M : PModule k S has ordinary Ext-based projective dimension at most D(S). The category is the persistence functor category itself; the result is not an estimate for arbitrary modules over the incidence algebra.

For a ≤ b, the existing principal-upset inclusion HookCokernel.inclusion is identified with the singleton face map, whose injectivity is already proved. Its actual cokernel is the finite hook, so the resulting categorical short-exact row and projectivity of both principal upsets give ordinary projective dimension at most one for each finite hook. The infinite hooks are projective. The existing occurrence-indexed classification of rank-projectives then gives the same bound for every pointwise finite-dimensional rank-projective.

An actual finite rank-projective resolution with m middle terms gives the loose ordinary bound PD(M) ≤ m: its pointwise exact rows become categorical short-exact rows, each middle term has ordinary projective dimension at most one, and induction proceeds from the terminal zero object. The stored count is only a sufficient bound; it is neither a shortestness statement nor an upper bound on every supplied resolution's raw length. Applying this result to the existing principal-downset resolution gives PD(I_p) ≤ complementCard(p) for p ≠ ⊤. At ⊤, the existing isomorphism to a projective principal upset gives the sharper degree-zero bound.

Every pointwise finite-dimensional ordinary injective embeds into the existing finite occurrence-indexed biproduct of principal-downset injectives. Its actual retraction and finite-biproduct closure give ordinary dimension at most D(S). To remove any bound on a caller's module, the proof first exposes the count of one already constructed finite rank-projective resolution, 3 * Fintype.card S + 2, and obtains a temporary uniform bound for all such modules. It then descends that bound one step at a time using the actual injective-presentation row M → I(M) → Q(M), with I(M) ordinary injective and Q(M) pointwise finite-dimensional. The temporary bound is discharged in the final theorem.

For an actual coordinate grid with nonempty factors, the existing minimal- complement count yields ordinary projective dimension at most n; for the actual interval-pair poset it yields at most 2n when n ≥ 1. These are representation-category bounds. The ordinary dimension of the Hom module used in minimal relative rank-resolution comparisons is a different invariant, and this theorem does not imply relative resolution-term vanishing. No arbitrary incidence-module equivalence or Hom-cokernel bound follows from this representation theorem alone. The finite-dimensional actual incidence-module comparison and the actual Hom-module/Hom-cokernel dimensions are proved separately; see ordinary projective dimensions of actual incidence modules and finite-grid Hom-module and Hom-cokernel dimensions. This representation-category theorem does not establish a global-dimension bound for arbitrary incidence modules, the real-coordinate transfer, or the final stability result.

The proof is a bounded representation-category component of the arguments in BOOS24-v3, including the ordinary-dimension step around Theorem 5.16. It uses the library's actual hook, finite-resolution, injective-presentation and dimension-shift constructions; it does not claim the source's broader incidence-algebra conclusion.