Finite induced matchings through the image

Modules and labeled barcodes

Fix a field k : Type u, a finite chain Fin s, and pointwise finite-dimensional modules M N : PModule.{u,0,u} k (Fin s). The input is an arbitrary natural transformation f : M ⟶ N. The field and module carriers share their universe.

The canonical barcode rankBarcode M is a multiset of half-open intervals [birth,death) with natural endpoints birth < death ≤ s. BarcodeOccurrence s B is the index type of B.toList, and occurrence B i is the interval at that index. Equal interval values retain distinct labels. Death s records survival through the final observed stage.

The image factorization

finiteImage f has fiber range (f.app i).hom at stage i. Naturality sends each such range into the next one, so restricting the structure maps of N gives its functor structure. The image fibers inherit finite dimensionality from the target fibers by the existing submodule instance.

toFiniteImage f corestricts each component to its range and is componentwise surjective. fromFiniteImage f includes each range in the target fiber and is componentwise injective. Both transformations are natural and their composite is f, as proved by toFiniteImage_comp_fromFiniteImage.

Write A = O(rankBarcode (finiteImage f)), B = O(rankBarcode M) and C = O(rankBarcode N) for the three occurrence types. The existing epi embedding and mono embedding give embeddings

E : A ↪ B,    J : A ↪ C.

They are exposed as imageSourceEmbedding and imageTargetEmbedding. E preserves birth and sends the image bar to a source bar dying no earlier. J preserves death and sends it to a target bar born no later. Their canonical orders use the original occurrence indices to distinguish tied copies.

The partial injection

inducedDomain f is exactly range E. imageOccurrenceEquivDomain identifies A with this domain by Equiv.ofInjective. Taking its inverse and then applying J defines inducedMatching f, an embedding of the domain subtype into C. Thus its inverse step is defined precisely where an intermediate occurrence exists.

inducedMap f : B → Option C evaluates this partial injection on every source label. It returns none on unselected labels. inducedMap_eq_some_iff_unique characterizes its graph:

inducedMap f i = some j  ↔  ∃! a : A, E(a) = i and J(a) = j.

Consequently its domain is exactly range E and its range is exactly range J. inducedMap_injective proves that equal selected values have equal source labels. imageOccurrenceEquivPairs is an equivalence from A to the subtype of matched pairs (i,j). This retains intermediate multiplicities even when several bars have equal endpoint values. The construction also works with empty occurrence sets and on the empty chain: no stage or unmatched output value is chosen.

Endpoints and specializations

For a matched source interval I, target interval K and their intermediate image interval H, the epi and mono properties give

birth(H) = birth(I),    death(H) = death(K),
birth(K) ≤ birth(I) < death(K) ≤ death(I).

The strict middle inequality is the nonempty-interval property of H. inducedMap_intermediate_endpoints and inducedMap_endpoints expose these facts. In this half-open convention, H therefore has the endpoints of I ∩ K. Terminal target survival forces terminal source survival by the upper bound death ≤ s.

For componentwise injective f, the map to its image is an isomorphism. For componentwise surjective f, the inclusion from its image is an isomorphism. Isomorphism invariance identifies the corresponding canonical barcodes. Transport along that equality preserves the original list indices and both group ordinals; the mono and epi embeddings across the isomorphic factor are inverse transports. This proves the specializations of the constructed map:

  • inducedMap_of_injective is some (monoEmbedding f hf i) on every source label.
  • inducedMap_of_surjective_iff is the reverse graph of epiEmbedding f hf.
  • inducedMap_id matches every occurrence to itself.

Consumers include a rank-one endomorphism between modules with two repeated bars, whose image supplies exactly one matched pair; a zero map selecting none; an overlap map [1,3) → [0,2); duplicate identity labels; terminal bars; and an empty chain.

Source correspondence and limits

Manifestation, attribution and rights: BL15-v4. The image-factorization construction specializes Bauer–Lesnick v4, §5, and the qualitative endpoint result specializes Proposition 5.3 and its proof. The intermediate-occurrence equivalence implements the matched-pair bijection used in Proposition 5.4's proof; the full barcode-reconstruction statement of that proposition is not asserted here. The range-valued functor, domain subtype, Option map and explicit list-label transports are library representation choices.

Composition laws concern the mono and epi subcategories, as in Proposition 5.7. The general induced matching need not preserve composition; see Example 5.6. The distance-specific OccurrenceMatching has a separate total essential equivalence, so it is not the interface for this construction on all finite-chain occurrences. Finite mono/cokernel translation estimates bound the source birth and cover long targets for component injections. Quantitative induced matching transfers kernel and cokernel shortness through the image factors, bounds both endpoints and uniquely covers the corresponding long source and target occurrences. Constructible closure, real-grid/shift compatibility, distance-matching conversion and d_B ≤ d_I remain separate results.