Constructible defects and shortness on critical samples

An interleaving of finite-constructible real modules supplies finite sampled kernels and cokernels with a proved shortness bound. The bridge uses natural transformations, categorical defects and within-cell isomorphisms. It constructs the common grid and finite translation used in the conclusion.

Modules, grids and the two samples

Fix a finite strictly increasing grid c : Fin n ↪o ℝ. The predicate IsConstructible c M requires a zero left tail and isomorphic structure maps within each half-open cell, including the unbounded final cell. The reconstruction API counts these cells with cellIndex c. A parameter exactly at a critical value belongs to its new cell.

Critical sampling is reindex c.toOrderHom M on Fin n, with a morphism sampled by reindexNatTrans c.toOrderHom f. This is the sample used by moduleDiagram. The distinct function FiniteConstructible.sample c M adds a zero initial stage for reconstruction. The critical sample here has no padded stage, sentinel real grade or appended terminal zero. An empty critical grid gives an empty sample; constructibility on that grid also forces the real module to be pointwise zero.

Constructibility and natural defect comparisons work over a ring, with independent coefficient and module-carrier universes. The finite TranslationTrivial wrappers use the existing coefficient-field and same-carrier universe convention. None of these shortness results requires finite dimensionality.

Kernels and cokernels

Let f : M → N be a natural transformation, with M and N constructible on c. The results IsConstructible.kernel and IsConstructible.cokernel prove constructibility of mathlib's functor-category kernel f and cokernel f. They do not assume a chosen defect model.

On the left tail, the kernel inclusion is a monomorphism into a zero fiber, and the cokernel projection is an epimorphism out of a zero fiber. Thus both defect fibers are zero. Within a cell, naturality gives a commuting square whose source and target structure maps are isomorphisms. Evaluation preserves kernels and cokernels; their induced maps for that square are isomorphisms. The defect structure maps agree with these induced maps through the evaluation comparisons, so they are isomorphisms as well.

Left whiskering by the critical-grid functor preserves kernels and cokernels. The natural isomorphisms criticalKernelIso and criticalCokernelIso therefore identify the critical sample of each real defect with the defect of the sampled morphism. Their compatibility theorems retain the kernel inclusion and cokernel projection. This closure argument applies to a morphism between modules on a common grid; it makes no claim that arbitrary submodules are constructible.

Translating and rounding critical indices

Fix a finite nonnegative real δ : ℝ≥0. The positive-evaluation shift sends a parameter t to t+δ. For every critical index i, the value c_i+δ lies on or after c_i, so its cell is positive. Define

\[ \theta_\delta(i)=\operatorname{cellIndex}(c,c_i+\delta)-1. \]

The function sampleTranslation c δ is this monotone inflationary endomap of Fin n. Its selected critical value satisfies

\[ c_i\le c_{\theta_\delta(i)}\le c_i+\delta, \qquad \operatorname{cellIndex}(c,c_{\theta_\delta(i)}) =\operatorname{cellIndex}(c,c_i+\delta). \]

Thus rounding selects the start of the translated parameter's cell. It can fix an interior index when δ is smaller than the next gap. Exact critical boundaries advance to the critical value itself. At δ=0 the endomap is the identity. The last observed index is fixed for every finite δ; its real cell remains unbounded. No barcode terminal endpoint is introduced as an observed stage.

Transferring annihilation

Suppose A is c-constructible and its real transition transition δ A is zero. The map from c_i to c_i+δ factors as

\[ A(c_i)\longrightarrow A(c_{\theta_\delta(i)}) \longrightarrow A(c_i+\delta). \]

The second map is a within-cell isomorphism. Cancelling it from the zero composite proves that the first map is zero. This is critical_structureMap_eq_zero, expressed as critical_translationTrivial over a field. Cancellation is essential when the translation rounds down.

Apply this to the real kernel or cokernel of f, then transport through the natural sampling comparison. The results criticalKernel_translationTrivial and criticalCokernel_translationTrivial prove shortness of the finite defect of the sampled morphism. Their generic real zero-transition hypothesis is explicit.

For an ε-interleaving of M and N on possibly different grids c and d, interleaving_critical_defects constructs e = commonGrid c (shiftGrid ε d). The grid results put M and the shifted target on e. The theorem derives constructibility of both defects and both finite sampled shortness statements for sampleTranslation e (ε+ε). The real annihilation comes from the interleaving equations; no shortness, matching or decomposition witness is an input to this theorem.

Boundary consumers and remaining scope

An one-interleaving obtained from zero comparison maps on [0,2) has nonzero kernel and cokernel. Its constructed common grid includes a negative shifted critical value. Both sampled defects satisfy the double-shift bound, while the sampled kernel fails the shorter parameter because its real map from 0 to 1 preserves coefficient one. The sampling comparisons retain their inclusion and projection in this example.

The grid [-3,0,10] checks strict rounding, an interior fixed index, an exact translated boundary and the fixed last observation. A genuine constructible essential tail rejects every finite sampled shortness bound: its last self-map is an isomorphism on a nonzero fiber. Parameter zero has zero defects even when the original module is nonzero. Empty-grid sampling is also checked. These consumers are in TDA.Persistence.ConstructibleDefectExamples.

Bauer–Lesnick v4, §2.1, gives the functor and natural-transformation conventions; §3.1 defines positive shifts and transitions. Section 6 defines epsilon-triviality by the zero transition, and Lemma 6.3 derives double-shift defect annihilation from an interleaving. Its field-valued case is a checked specialization of the existing ring-valued annihilation theorem. Finite-grid defect closure, rounded critical translation and the sampling bridge above are library constructions using mathlib's limit comparisons, rather than numbered source sampling theorems. Source and rights metadata are in BL15-v4.

The real induced matching consumes these sampled defects and converts the finite partial injection into real labels, with exact endpoint, essential and long-bar coverage bounds. Its interleaving consumer constructs a matching into the shifted target. The finite-constructible isometry returns that target by +ε, proves ε-cost and takes the distance infimum to establish equality. General infinite decomposition and the omitted Lemma 6.2 proof are outside these sampling results.