Common grids, shifts and refinement-invariant diagrams

Finite-constructible real modules can be placed on a common finite grid without changing their real endpoint diagrams. The grid may add observations before, between and after the old critical values. Positive evaluation shifts move critical values left. These constructions supply grid data for comparing a module with a shifted target.

Grids and constructibility

A grid is a strictly increasing c : Fin n ↪o ℝ, including n=0. FiniteConstructible.cellIndex c t counts critical values at or below t. Its cells are a zero left tail, left-closed/right-open cells between successive values, and an unbounded final cell. No sentinel real grade represents the zero cell. The reconstruction API defines IsConstructible c M using a zero left tail and isomorphic structure maps within each cell.

Constructibility results work for M : PModule.{u,0,w} R ℝ over a ring, with independent coefficient and carrier universes. Barcode and dimension results below use a field k : Type u, carrier universe u, and finite dimensionality at the original critical fibers.

Refines c d means that the range of c is contained in the range of d. It derives a unique increasing embedding e of critical indices with d(e i)=c i. The function commonGrid c d orders the finite union of their critical-value sets. Shared critical values occur once in the union grid; equal barcode occurrences retain their multiplicity.

If two parameters have the same d-cell, every c-critical value has the same comparison with both parameters, so their c-cells agree. Therefore IsConstructible.refine transfers constructibility from c to d. A parameter below every d-value is also below every c-value, preserving the zero left tail.

Positive evaluation shifts

For finite nonnegative ε : ℝ≥0, the shift satisfies (SεM)(t)=M(t+ε). The grid shiftGrid ε c has values c_i-ε, and

\[ \operatorname{cellIndex}(c-\varepsilon,t) =\operatorname{cellIndex}(c,t+\varepsilon). \]

This identity proves IsConstructible.shift for the shifted module, including negative shifted critical values, exact critical parameters and the whole right tail. At ε=0 the shifted grid equals c. Empty grids select no stage.

For M constructible on c and N constructible on d, common_shift_constructible places both M and SεN on the constructed commonGrid c (shiftGrid ε d). No common-grid witness is supplied as input.

Finiteness and the diagram on a refinement

Suppose M is c-constructible and its original critical fibers are finite dimensional. The reconstruction component identifies each real fiber with its finite sample fiber, which is either zero or an original critical fiber. Thus finiteDimensional_of_isConstructible derives finiteness everywhere, including at all added d-critical values.

The existing moduleDiagram c M assigns real endpoints to the canonical rank barcode of the critical samples. Its terminal index means the constant essential tail. Under Refines c d, moduleDiagram_refinement proves equality of the two diagrams. Finite-index barcodes on c and d have different types and are related by monotone sampling; the equality is of their real endpoint multisets.

Let A be the coarse sample with its initial zero stage. Sample its reconstruction at the fine critical parameters using

\[ q(j)=\operatorname{cellIndex}(c,d_j)\in\operatorname{Fin}(n+1). \]

Whiskering the reconstruction isomorphism identifies this module with the fine samples of M. Finite rank-barcode invariance and rankBarcode_reindex therefore restrict each original occurrence along q.

For an old positive observed endpoint r, with 1≤r≤n, its endpoint cut is exactly the position e(r-1) of c_(r-1) in d. At terminal endpoint r=n+1 the cut is m, the fine grid size. Every old barcode birth is positive because the padded zero stage contributes no bars. Strict increase of e makes every restricted occurrence nonempty. Consequently both finite endpoints recover the same real values, while terminal death remains infinity even when d adds later observations. Applying this identity occurrence by occurrence through Multiset.filterMap also preserves duplicates.

Boundary consumers

A rational module reconstructed on [1,3] has two copies of [1,3) and one essential bar [3,∞). Refining to [-2,1,2,3,8] gives exactly the same real endpoint multiset. The new observation at 8 creates no finite death for the essential occurrence. The grid [1,2] fails refinement because it omits 3.

At ε=2 the original critical values become [-1,1]. The exact first boundary belongs to its new cell, and the last cell still contains time 100. The union with [1,3] has three distinct values because 1 is shared. Zero shift and refinement of the zero module from an empty grid are also checked in TDA.Persistence.ConstructibleRefinementExamples.

Source correspondence and scope

Bauer–Lesnick v4, §2.1, supplies the functor, interval and multiset-barcode semantics. Section 3.1 evaluates a shift at t+ε; the barcode correspondence preceding Theorem 6.4 in §6.1 moves its endpoints left by ε, with the inverse correspondence adding ε. The zero-left/right-continuous finite grid model and its refinement theorem are library constructions proved using existing finite reconstruction and rank results. They are not attributed to a numbered source refinement theorem or to an inspected proof of general infinite decomposition. Source and rights metadata are recorded in BL15-v4.

The defect and sampling API proves kernel/cokernel constructibility for morphisms between modules on a shared grid and transfers real annihilation to finite sampled defects. It uses naturality and within-cell isomorphisms, without arbitrary-submodule closure. The shift-diagram formula moves endpoints by −ε and returns occurrence labels by +ε. The finite-constructible isometry constructs the original-diagram ε-cost matching and proves equality of the extended distances. The finite quantitative induced matching retains its finite-dimensionality and endpoint hypotheses.