Minimal-complement cardinalities in finite grids

For a lattice S, minimalComplement p is the set of minimal elements outside the principal downset ↓p. For a finite lattice, complementCard p counts those elements and maxComplementCard takes the maximum over p. These are minimal generators of the complement, not covers above p.

For an arbitrary product of lattices with an OrderBot supplied on each factor, mem_minimalComplement_pi_iff_of_orderBot classifies membership using those caller-supplied bottoms: one coordinate is a minimal complement in its factor and every other coordinate is bottom. It requires no finiteness or linearity; each supplied OrderBot already provides its factor bottom and inhabitant, so no separate Nonempty assumption is needed. This classification does not bound the number of minimal complements by the number of factors, since a lattice factor can have multiple incomparable minimal complement values.

The compatibility theorem mem_minimalComplement_pi_iff retains the original finite-index and finite nonempty linear-factor assumptions and canonical factor bottoms from the private factor enumeration Fintype.ofFinite (T i) and its Fintype.toOrderBot instances. In a finite product of nonempty finite linear orders, the unique active coordinate gives an injection into the index set and proves the bound by the number of factors, including the empty product and singleton factors.

Adjoining a new top requires a correction: the old top has the singleton complement consisting of the new top, while the new top itself has empty complement. An old non-top point retains precisely the old minimal complements. Consequently the maximum for WithTop S is at most max 1 (maxComplementCard S).

For the existing interval-pair subtype PairPoset S = {q : WithTop S × WithTop S // q.1 ≤ q.2}, a point (j,k) has two disjoint minimal-complement families. The first consists of diagonal pairs (a,a) for a ∈ minimalComplement j with a ≤ k. The second consists of (⊥,b) for b ∈ minimalComplement k. The exact count is the restricted diagonal-family cardinality plus complementCard k; bounding the first family by complementCard j gives the pair maximum bound 2 * max 1 (maxComplementCard S).

For an actual CoordinateGrid n with every coordinate factor nonempty, the proved maxima are at most n on G.Point, at most n on WithTop G.Point when n ≥ 1, and at most 2*n on PairPoset G.Point when n ≥ 1. The explicit positive-dimension hypothesis accounts for the new-top singleton when the underlying point lattice has only one element.

Combining the point-grid count with the existing finite-injective Hom dimension theorem proves HasProjectiveDimensionLE (Hom(HookSum,M)) (n - 1) for every pointwise finite-dimensional ordinary injective M on the grid, over any field. Every supplied certified minimal rank-projective resolution of M has zero terms in degrees j > n - 1. The result supplies a combinatorial input for later incidence-algebra estimates; it does not claim a global-dimension bound or the 2n−2 persistence estimate.

The source context is BOOS24-v3, §2.4 and Notations 5.2 and 5.9, Proposition 5.14, and the minimal-complement count used in Theorem 5.16. This formalization proves the displayed finite-product, adjoined-top, and actual interval-pair counts directly, along with the ordinary-injective Hom corollary above. It does not establish the remaining conclusions of Theorem 5.16.

The implementation is in TDA.Persistence.FiniteLatticeComplementCardinality and TDA.Persistence.FiniteGridInjectiveRankDimension; the focused fixtures are in TDA.Persistence.FiniteGridComplementCardinalityExamples.

For any field and n ≥ 1, the 2*n-2 degree bound for certified minimal rank-projective resolutions of originally finitely presented real-coordinate modules is proved separately in FiniteRankProjectiveResolutionDimension; see the real-coordinate degree bound. It asserts vanishing of zero-padded terms above that degree, independently of the stored length.