Quantitative finite mono matchings under monotone translations
Modules, intervals and translations
Fix a field k : Type u and pointwise finite-dimensional modules
M N : PModule.{u,0,u} k (Fin s). The field and module carriers share their universe.
A natural transformation f : M ⟶ N has injective components.
Its canonical mono embedding pairs occurrence labels in
the two rank barcodes, preserving death and the ordinal at that death.
Repeated interval values keep their original list labels.
Intervals use [birth,death) with birth < death ≤ s.
FiniteInterval.birthStage packages the birth as an index in Fin s.
Let theta : Fin s →o Fin s be monotone and suppose hTheta : ∀ i, i ≤ theta i.
It may identify stages and need not advance a fixed number of indices.
Call a target interval [b,d) long when (theta b).val < d.
Equality belongs to the short case because the interval excludes its death.
Shortness of the cokernel
finiteCokernel f is mathlib's categorical cokernel in the functor category.
finiteCokernelFiberIso identifies its fiber at i with
N_i / range(f_i). The identification sends the cokernel projection to
the quotient map, and target finite dimensionality gives cokernel finite dimensionality.
The structure maps are those induced by the original natural transformation.
TranslationTrivial theta hTheta K says that every structure map
K(i → theta i) is zero. In particular, it imposes no barcode or matching condition.
cokernel_translationTrivial_iff proves the equivalent image criterion:
TranslationTrivial theta hTheta (finiteCokernel f)
↔ ∀ i, range N(i → theta i) ≤ range f_(theta i).
The equivalence uses the natural cokernel projection, its component epimorphisms, and the concrete range-quotient identification. Thus the image criterion is a consequence of the shortness hypothesis.
Shifted prefix counts
Write D_M(b,d) for deathPrefixCount (rankBarcode M) b d: the number of
occurrences born by b and dying exactly at d. Cokernel shortness gives
D_N(b,d) ≤ D_M(theta b,d), (theta b).val < d ≤ s.
deathPrefixCount_le_translated_of_cokernelTrivial proves this inequality.
For d < s, the counts are dimensions of the image/kernel intersections
at d−1. A target vector born by b lifts at theta b by the image criterion.
Transporting the lift to d−1 produces the target vector by naturality.
Its image at d is zero; injectivity of f_d forces the source lift to die there too.
This embeds the target subspace in the image of the translated source subspace,
so dimension monotonicity gives the count comparison.
For d=s, the count is the final structure-map image. The same vector lift
compares these images without forming a stage s or a kernel there.
The inequalities justify every predecessor and finite index.
Bounds and coverage for the canonical embedding
For a source occurrence I and its canonical target J, the result is
death(I) = death(J),
birth(J) ≤ birth(I) ≤ theta(birth(J)).
monoEmbedding_birth_bounds_of_cokernelTrivial supplies both birth bounds.
For long J, the shifted prefix comparison and equal ordinals exclude a source
birth after the translated target birth. For short J, the upper bound follows
directly from birth(I) < death(I) = death(J) ≤ theta(birth(J)).
monoEmbedding_covers_long_of_cokernelTrivial gives exactly one source occurrence
for every long target occurrence. Its ordinal lies below the translated source
prefix count and hence below the source death-group cardinality.
monoEmbedding_unmatched_short_of_cokernelTrivial states the contrapositive:
every unmatched target is short. Short targets can also be matched.
The consumers use a nonidentity translation identifying stages, a delayed-birth inclusion attaining the upper bound, a threshold-equality target left unmatched, duplicate long bars, terminal coverage, the identity translation and an empty chain. A nonzero quotient at a fixed point fails the shortness condition.
Terminal observations and later real grids
Terminal death s records finite observed survival. It is always long for an
endomap of Fin s, so the shortness hypothesis forces terminal target coverage.
No unobserved zero stage is appended. The empty chain forms no stage at all.
The separate constant-tail real realization is what
interprets a surviving final fiber as an essential tail.
For a later real-grid connection, the intended index at a critical value c_i
is the last c_j at or below c_i+epsilon. This floor convention can fix a stage
when the next critical value is far away. Testing only sample pairs separated by
at least epsilon would miss a nonzero map inside such a cell.
This real-grid transport, refinement and real endpoint estimate remain unformalized.
Source correspondence and limits
Manifestation, attribution and rights: BL15-v4. Bauer–Lesnick v4, §6, Theorem 6.1(ii) and its proof give the additive-real cokernel image criterion, long-target coverage and birth bound. The proof sandwiches a delayed image inside the image of the morphism and uses canonical mono composition; its long/short cases motivate the corresponding split here.
The theorem for an arbitrary monotone inflationary finite theta is a library extension.
Its direct shifted-prefix dimension proof replaces the delayed-image barcode calculation.
The formal source theorem and the target-image condition in its proof determine the
object types; the inconsistent source symbol in the preceding delayed-image sentence
does not change the target structure map used here.
No full version comparison or general erratum claim is made.
Kernel/epi estimates and quantitative image-factorization assembly
provide the complementary death bounds and long-source coverage. Real-grid
compatibility, conversion to a distance matching and d_B ≤ d_I remain separate results.