Real induced occurrence matchings

A morphism between finite-constructible real persistence modules gives an occurrence matching with exact real endpoint and long-bar bounds when its kernel and cokernel have zero transitions at the specified parameters. The construction retains multiplicities and essential tails. An interleaving supplies these hypotheses for its comparison into the shifted target.

Modules, defects and occurrences

Fix a field k : Type u and modules M N : PModule.{u,0,u} k ℝ constructible on a common finite increasing grid c : Fin n ↪o ℝ. Assume finite dimensionality only at the original critical fibers. Write D_M=moduleDiagram c M and D_N=moduleDiagram c N for their critical rank-barcode diagrams. The constructible convention is a zero left tail, left-closed/right-open finite strata and a constant final tail. A terminal index death n represents real death ⊤.

For f : M → N and finite δK δC : ℝ≥0, assume the transition of kernel f across δK and of cokernel f across δC is zero. The sampling comparison derives shortness of the sampled finite defects. No chosen decomposition, matching or endpoint estimate is supplied.

A diagram is a multiset. An occurrence label contains both an endpoint value and an index among its copies. Finite occurrence labels and essential birth labels form a disjoint sum. An OccurrenceMatching consists of a partial injection on finite labels and a total equivalence on essential labels.

The induced map on real labels

RealBarcode.indexOccurrenceEquiv identifies finite rank-barcode list labels with the finite/essential sum of the real endpoint diagram. Equality of their value multisets proves a value-preserving equivalence; equal endpoint values retain distinct labels. FiniteBarcode.RealInduced.sumMap conjugates the finite induced partial injection by these equivalences. Its selected-pair theorem and injectivity refer to that same map.

Let source [b,d) be selected with target [b′,d′). For finite deaths, sumMap_bounds proves

\[ b'\le b\le b'+\delta_C,\qquad b<d'\le d\le d'+\delta_K. \]

The finite quantitative bounds place each upper endpoint below the critical stage selected by the rounded translation. That stage has real value at most the original value plus δ. Monotonicity of c therefore gives the displayed real bounds directly, even on an irregular grid. No mesh-size term is added. The argument also works at δ=0 and when the rounded translation fixes or identifies indices.

The finite death upper bound forces the source death to be finite whenever the selected target death is finite. The opposite classification is excluded by the qualitative death order. Thus selected finite and essential labels cannot cross. An unmatched essential source would have ⊤≤b+δK, contradicting its finite right side; the unmatched-target estimate gives the analogous contradiction for essential targets. Partial injectivity then makes the selected essential map a total equivalence, including every repeated birth.

FiniteBarcode.RealInduced.matching splits this map into the required finite partial injection and essential equivalence. Its finite and essential pair theorems retain the exact selected-pair relation. The theorem matching_bounds derives RealMatchingBounds δK δC, which collects the endpoint conclusions and the following unmatched-length bounds:

\[ d-b\le\delta_K\quad\text{for every unmatched finite source},\qquad d'-b'\le\delta_C\quad\text{for every unmatched finite target}. \]

Consequently every source longer than δK and every target longer than δC has exactly one partner. Equality is the short case: a threshold-equality bar may be matched or unmatched. Essential occurrences are always covered in both directions.

Constructible morphisms and interleavings

FiniteConstructible.inducedOccurrenceMatching constructs the matching of D_M with D_N from the real defect hypotheses above. inducedOccurrenceMatching_bounds derives every field of its bounds object. Only critical-fiber finite dimensionality is supplied; finite sampled shortness is obtained from constructibility and defect annihilation.

For an ε-interleaving of M on c and N on another grid d, interleaving_shifted_induced_matching constructs the common grid of c and shiftGrid ε d. The interleaving equations give kernel and cokernel 2ε-annihilation for M → SεN. Refinement derives finiteness at the new observations. The shift-diagram formula identifies the resulting target with translateDiagram D_N (−ε). The output is a matching into that shifted diagram with RealMatchingBounds (ε+ε) (ε+ε), including its essential equivalence and strict long-bar coverage. No common-grid, new-fiber-finiteness, shortness or matching witness is an input.

The finite-constructible isometry construction applies the +ε target inverse and proves every finite, essential and diagonal cost contribution is at most ε. It then takes the interleaving-parameter infimum and combines the reverse bound into equality. The shifted-target bounds here are its derived input, with no extra matching or cost hypothesis on the original interleaving.

Concrete boundary cases

On the grid [0,1,3,4], a genuine overlap map from [1,4) to [0,3) has nonzero kernel and cokernel and attains both displacement-one upper bounds. The gap of length two adds no error. A map erasing [0,1) leaves its sole real source unmatched at threshold one. Identity maps retain threshold-short bars, two distinct copies of [0,3) and two essential copies born at one. They also give unique strict-long coverage, essential coverage at every finite threshold and the parameter-zero result. The empty grid constructs the empty matching without choosing a stage.

The zero-comparison one-interleaving on [0,2) derives a matching into its shifted target with both bounds equal to two. These consumers are in RealInducedMatchingExamples, FiniteInducedTranslationExamples and ConstructibleDefectExamples.

Source correspondence and limits

Bauer–Lesnick v4, §§2.1–2.3 and §3.1, supplies the functor, multiset-occurrence, endpoint-decoration and shift conventions. Formal Theorem 6.1(ii) controls births and long targets from cokernel triviality; (iii) controls deaths and long sources from kernel triviality. Their combined finite, half-open specialization is proved here. The finite quantitative proof uses a direct quotient-surjection argument rather than the source's duality route. The source omits Lemma 6.2's proof; no verification of that omitted argument is claimed.

In the proof of Theorem 6.1(ii), both the v4 HTML and PDF define a submodule of N using an image labelled with M, then use the transition of N in the following cokernel argument. The formal theorem and the target maps of N determine the specialization used here. This notation inconsistency does not alter the library's hypotheses or endpoint inequalities.

Lemma 6.3 supplies double-shift defect annihilation. The occurrence correspondence immediately before Theorem 6.4 adds ε back to shifted endpoints, and its proof converts the induced bounds to an ε-matching. The shifted-target part is established here; the linked isometry theorem completes its return and cost argument on the finite-constructible domain. Finite-grid sampling, real occurrence relabelling and the derived essential equivalence are library constructions implementing the specialization. General decorated intervals, infinite or q-tame decompositions and full isometry are outside this result. Source and rights metadata are recorded in BL15-v4.