Kernel and cokernel shortness from a real interleaving

An ε-interleaving forces the kernel and cokernel of either comparison morphism to have zero transition across 2ε. This supplies an algebraic input to induced matching estimates without assuming a matching or a barcode decomposition.

Objects and parameters

Fix a ring R : Type u and real-indexed modules M N : PModule.{u,0,w} R ℝ. The carrier universe w is independent of u. The shift parameter ε : ℝ≥0 is finite, and the real shift has positive evaluation direction: (SεN)(t)=N(t+ε).

Let i : RealInterleaving ε M N, with natural transformations f=i.φ : M → SεN and g=i.ψ : N → SεM. Their component equations, before the canonical addition transport, are

\[ g_{t+\varepsilon}f_t=M[t\le(t+\varepsilon)+\varepsilon],\qquad f_{t+\varepsilon}g_t=N[t\le(t+\varepsilon)+\varepsilon]. \]

Write K=kernel f and C=cokernel f for mathlib's objects in the functor category. Their inclusion κ:K→M and projection π:SεN→C are natural transformations. Write η_A^δ for the transition RealShift.transition δ A. A module is δ-trivial when this morphism is zero.

The two annihilation statements

In TDA.Persistence.RealInterleavingShortness, the theorems RealInterleaving.kernel_transition_eq_zero and RealInterleaving.cokernel_transition_eq_zero prove

\[ \eta_K^{2\varepsilon}=0,\qquad \eta_C^{2\varepsilon}=0. \]

No finite-dimensionality, constructibility or chosen decomposition is needed. The same statements for g follow by applying them to i.symm.

For the kernel, naturality of κ expresses its transition followed by the later inclusion as κ_t followed by the transition of M. The first interleaving equation factors that transition through f_t. Since f_t κ_t=0, cancellation of the later monomorphism gives the zero kernel transition.

For the cokernel, precompose its transition with the earlier epimorphism π_t. Naturality identifies the result with the transition of SεN, followed by the later projection. Apply the second interleaving equation at t+ε: the transition of N from t+ε to ((t+ε)+ε)+ε factors through f_((t+ε)+ε). The cokernel relation kills this final factor. Cancellation of π_t gives the zero cokernel transition.

Both arguments first reach the nested index (t+ε)+ε. Applying the functor map along (t+ε)+ε=t+(ε+ε) transports these zero maps to the declared double-shift transition. This keeps the shifted target in the cokernel argument and the addition identification explicit.

Sharp threshold and parameter zero

The rational interval module [0,2) has zero two-transition. Thus zero forward and reverse maps define a one-interleaving of this nonzero module with itself. Its kernel is isomorphic to [0,2), and its cokernel to the shifted interval [-1,1), by the categorical zero-morphism isomorphisms. The one-transitions of both defects are nonzero: evaluate at 0 for the kernel and at -1 for the cokernel. Both retain coefficient one. Consequently the factor two cannot uniformly be replaced by one, including at the half-open length-equals-two boundary.

At parameter zero a natural isomorphism of the constant integer module gives a zero-interleaving. The proved shortness and the canonical zero-shift unit imply that both defects are zero objects, while the original constant module has nonzero fibers.

These consumers are in TDA.Persistence.RealInterleavingShortnessExamples.

Source correspondence and remaining bridge

Bauer–Lesnick v4, Lemma 6.3, states the 2ε-triviality of the kernel and cokernel of an interleaving morphism; its short proof refers directly to the interleaving definition in §3.1. The source's ε-triviality in §6 is exactly vanishing of the transition morphism. The field-valued statement is a specialization of these Lean results, which also cover ring-valued modules and impose no dimension hypothesis. The proof above establishes this extension using categorical kernels and cokernels. Source and rights metadata are recorded in BL15-v4.

Lemma 6.2 is the separate duality statement, whose proof the source omits; this argument does not use it. The constructible defect and sampling bridge derives defect constructibility on a common grid and transfers this annihilation to the finite sampled kernels and cokernels. The finite quantitative induced matching retains its finite-dimensionality and endpoint hypotheses. The shift-diagram correspondence returns shifted occurrence labels by +ε. The real induced matching consumes these derived hypotheses and proves exact endpoint and long-bar bounds into the shifted target. The finite-constructible isometry returns that target by +ε, proves ε-cost and takes the distance infimum to obtain equality.