Quantitative finite induced matchings from kernel and cokernel shortness

Modules, occurrences and translations

Fix a field k : Type u and pointwise finite-dimensional modules M N : PModule.{u,0,u} k (Fin s). The input is a natural transformation f : M ⟶ N. The field and module carriers share their universe. The canonical barcodes contain half-open intervals [birth,death) with birth < death ≤ s. Occurrences retain their original list indices, including separate labels for repeated interval values.

A translation theta : Fin s →o Fin s is monotone and inflationary: hTheta : ∀ i, i ≤ theta i. It may identify stages or fix them. TranslationTrivial theta hTheta K means that every structure map K(i → theta i) is zero. An interval [b,d) is long when theta(b) < d; equality belongs to the short case.

Birth is an observed stage, packaged by FiniteInterval.birthStage. Death may instead be the terminal endpoint s. Define its extension by

thetaPlus(d) = theta(d),  d < s,
thetaPlus(s) = s.

FiniteInterval.translatedDeath implements this convention. Its value depends only on death, and identity translation leaves every endpoint unchanged. The terminal endpoint records finite observed survival and is never used as a Fin s stage. The empty chain supplies no occurrence and selects no stage.

Kernel shortness

finiteKernel f is mathlib's categorical kernel in the functor category. finiteKernelFiberIso identifies its fiber with ker(f_i) and takes the kernel inclusion to the subtype inclusion. Its finite dimensionality follows from the source subspace. Naturality of the inclusion and its component monomorphisms prove finiteKernel_map_eq_zero_iff. Thus kernel_translationTrivial_iff gives

TranslationTrivial theta hTheta (finiteKernel f)
  ↔ ∀ i, ker(f_i) ≤ ker M(i → theta i).

The right side is a proved criterion for the kernel, rather than a substitute object or an assumed barcode condition. The analogous cokernel criterion is explained in the mono translation estimate.

Shifted survival counts for an epimorphism

First suppose p : M ⟶ N is componentwise surjective and its kernel is translation-trivial. Let S_M(b,c) count bars born exactly at b that survive at c. The result is

S_M(b,theta(c)) ≤ S_N(b,c),    b ≤ c < s.

The proof uses linear maps. Kernel shortness makes M(c → theta c) kill ker(p_c). The first isomorphism theorem therefore descends this transition through the surjection p_c to a map N_c → M_(theta c). For any earlier stage a ≤ c, naturality identifies the image of N(a → c)(p_a x) with M(a → theta c)(x).

Consequently the descended map restricts to a surjection

image N(b → c) → image M(b → theta c).

For positive birth, surjectivity at b−1 and the same naturality equation send the preceding target image into the preceding source image. Passing to these image quotients gives a genuine surjection. Their dimensions are the same-birth survival counts, so the dimension comparison proves the inequality. Birth zero compares the images directly, without forming a predecessor. birthSuffixCount_translated_le_of_kernelTrivial exposes the result.

The canonical epi estimate

The canonical epi embedding goes from target occurrences back to source occurrences. It preserves birth and matches equal ordinals in each birth group, ordered by descending death and original index. For a target occurrence J and its source I, epiEmbedding_death_bounds_of_kernelTrivial proves

birth(I) = birth(J),
death(J) ≤ death(I) ≤ thetaPlus(death(J)).

At a finite target death d, if the source survived theta(d), its ordinal would lie below the translated source survival count. The target ordinal lies at or above its survival count at d. Equal ordinals and the shifted inequality contradict these two bounds. At terminal death, death(I) ≤ s supplies the upper bound directly.

Taking c=b puts every long source ordinal below the target birth-group size. epiEmbedding_covers_long_of_kernelTrivial therefore gives exactly one target preimage for every long source. The unmatched-source result is its contrapositive. Short source bars may also be matched. Terminal source bars are always long for an endomap of Fin s, so kernel shortness forces their coverage too.

Quantitative matching through the image

For an arbitrary f, the image factorization is an epi followed by a mono. Corestriction preserves the component kernel; the image inclusion has the same range as f. The proved kernel/range criteria therefore transfer shortness to the respective factors, without choosing new kernel or cokernel objects as hypotheses.

Let inducedMap f i = some j, and write I and J for the source and target intervals. Kernel shortness for thetaK and cokernel shortness for thetaC give

birth(J) ≤ birth(I) ≤ thetaC(birth(J)),
birth(I) < death(J) ≤ death(I) ≤ thetaKPlus(death(J)).

The unique intermediate image occurrence has source birth and target death. This connects the two factor estimates to the partial injection. inducedMap_endpoint_bounds_of_shortness combines them. The one-sided birth/death theorems require only their respective shortness hypothesis. Independently, every kernel-long source has exactly one selected target, and every cokernel-long target has exactly one selected source. Unmatched occurrences satisfy the corresponding shortness bound.

Consumers include truncations attaining the finite death bound, a positive-birth quotient, a noninjective translation, terminal coverage, distinct duplicate labels, identity translation and the empty chain. Threshold-short sources occur both matched and unmatched. A nonzero kernel at a fixed point fails shortness. The genuine overlap map [1,3) → [0,2) has nonzero kernel and cokernel, attains both endpoint upper bounds and exercises both coverage directions.

Source correspondence and limits

Manifestation, attribution and rights: BL15-v4. Bauer–Lesnick v4, §6, formal Theorem 6.1 controls matched births and long targets by cokernel triviality, and matched deaths and long sources by kernel triviality. Its part (iii) is obtained by dualizing (ii), with Proposition 5.5 and Lemma 6.2 giving an alternative route. The canonical epi order is in §4.1 and Theorem 4.2(ii).

The arbitrary finite monotone translation theorem is a library extension. The direct quotient-surjection proof here replaces the source's duality route; no linear-dual or order-reversal adapter is supplied. The real induced matching converts the rounded finite bounds and partial injection to exact real endpoint estimates, long-bar coverage and a total essential equivalence. The common-grid sampled defects supply its shortness inputs. Its interleaving consumer reaches the shifted target; the finite-constructible isometry completes +ε untranslation, ε-cost and the distance-infimum comparison. The matching retains the qualitative construction and its restriction of composition laws to monos and epis.