A degree-two rank-projective witness in two real coordinates
Let k be any field. No characteristic, finiteness, or algebraic-closure
assumption is needed. The finite model lives on the actual grid
G : CoordinateGrid 2 whose two coordinate sets are both {0,1,2}. Write
H(x,y) for the library's finite-death hook, with support
{ p | x ≤ p and not (y ≤ p) }.
Thus a hook is not the bounded rectangle or half-open square between its
endpoints. In particular, H((1,1),(2,2)) contains the points (1,2) and
(2,1) and dies at (2,2).
The finite module and its hook terms
Name the grid points
e=(2,0) g=(2,1) h=(2,2)
b=(1,0) c=(1,1) f=(1,2)
o=(0,0) a=(0,1) d=(0,2)
The module finiteModule is defined independently from a resolution. Its
support is
S = ((↑a ∪ ↑b) ∩ ↓c) = {a,b,c}.
It has one copy of k at those three points, zero elsewhere, and identity
structure maps from a to c and from b to c. This is a pointwise
finite-dimensional module on a finite poset.
The concrete resolution uses the following actual finite hooks:
A = H(a,d) B = H(a,g) C = H(b,f) D = H(b,e)
E = H(a,h) F = H(b,h) J = H(c,f) K = H(c,g)
L = H(c,h)
P₀ = A ⊕ B ⊕ C ⊕ D
P₁ = E ⊕ F ⊕ J ⊕ K
P₂ = L
Each summand of P₀ maps to finiteModule by sending its birth unit to the
unit at a or b. The resulting augmentation q₀ : P₀ → finiteModule is
surjective, and its actual categorical kernel is K₁ := kernel q₀.
The next quotient q₁ : P₁ → K₁ is defined by lifting the following map
through that actual kernel:
E F J K
A 1 0 1 0
B -1 0 0 1
C 0 1 -1 0
D 0 -1 0 -1
The degree-two map j : L → P₁ has column (1,-1,-1,1) in the order
(E,F,J,K). These are the literal maps d1 and j, built from the
repository's normalized hook maps. The selected length-three resolution has
kernel objects (finiteModule,K₁,L,0), terms (P₀,P₁,L), quotients
(q₀,q₁,id_L), and inclusions (kernel.ι q₀,j,0).
Rank exactness and minimality
Both nontrivial rows are proved rank exact from the maps. At every grid point,
q₀ is surjective and every quotient vector killed by a structure map has a
lift in P₀ killed at the same target. This supplies the actual-kernel
rank-exact row for q₀.
For the second row, the fiber at c of K₁ consists of the coefficient
vectors (A,B,C,D) satisfying A+B+C+D=0; one preimage under q₁ is
(-B,-D,A+B,0). The kernel of q₁ is exactly the image of the given j,
including at the boundary fibers f and g. For same-death lifting, a vector
at c dying toward f is a multiple of A-C and lifts through J; one
dying toward g is a multiple of B-D and lifts through K. The other
nonzero syzygy fibers propagate until h, where their E or F lifts die.
These pointwise facts give the row's rank additivity through the existing
same-death criterion. The last row is the split identity row on L and zero.
Each term is rank projective. Within each four-hook family, all off-diagonal
Hom spaces vanish; every endomorphism of a summand is a scalar identity.
Therefore every endomorphism of P₀ or P₁ fixed by its quotient is
diagonal with every scalar equal to one, since each quotient component is
nonzero at birth. This proves that q₀ and q₁ are rank-projective covers.
The final quotient is the identity cover. Hence the concrete resolution is a
certified minimal finite rank-projective resolution, rather than just a
rank-exact complex of the stated length.
The unit of L at c is nonzero, and the prescribed j is nonzero over
every field, including characteristic two. Termwise uniqueness of certified
minimal resolutions then shows that every minimal finite resolution of
finiteModule has nonzero term two and zero terms above two, including after
arbitrary terminal zero padding.
There is a separate ordinary projective-dimension calculation for the actual finite hook-generator Hom object:
projectiveDimension
((FiniteRankProjectiveHomExact.homFunctor (S := G.Point)).obj finiteModule) = 2.
This is a dimension over the actual opposite endomorphism ring of the finite
hook generator after applying the Hom functor. It is not a claim about the
ordinary functor-category projective dimension of finiteModule.
The real-coordinate module
Define realModule literally as G.lan.obj finiteModule, the actual left Kan
extension along the grid embedding. Its original
GradedFree.FinitelyPresented.{u,0,0} predicate is derived from the
finite-dimensional fibers of the finite-poset module and the existing
finite-grid Kan preservation theorem. No resolution or finite-presentation
assumption is used to define the module.
Applying the actual Kan functor to the selected finite resolution gives a
concrete certified minimal resolution of realModule. The second term is
identified with the real hook H((1,1),(2,2)) by the Kan comparison for
H(c,h). This real hook is supported where both coordinates are at least one
and the point does not lie above (2,2). It therefore continues along the
upper arm (1,t) for t ≥ 1 and the right arm (t,1) for t ≥ 1, while its
birth unit dies at (2,2).
For every certified minimal resolution of this real module, term two is
nonzero and all terms above degree two are zero. These conclusions use the
real term isomorphism and zero-padded termwise uniqueness, so they do not
depend on the supplied resolution's stored length. The module is thus a
concrete two-coordinate equality case for the general 2n-2 upper bound.
The construction proves no lower bound in every dimension, computes no
ordinary projective dimension for the module itself, and includes no signed
stability or primary-source comparison claim. The reusable theory is in
TDA.Persistence.RankResolutionDegreeTwo; the rational and characteristic-two
consumers are in the separate fixture module
TDA.Persistence.RankResolutionDegreeTwoExamples, under the declaration
namespace TDA.Persistence.RankResolutionDegreeTwo.Examples.
The actual degree-one simplicial homology of a literal four-vertex face filtration is naturally isomorphic on the finite grid to this independently defined module; its real left Kan extension is naturally isomorphic on all real grades. The signed-chain calculation, grid isomorphism, and real Kan comparison are documented in the simplicial realization note.