Real shift diagrams and occurrence untranslation

Positive evaluation shifts move a real persistence module's barcode to the left. The critical restrictions prove this formula, and the resulting occurrence bijections let a matching's shifted target return to its original diagram. The construction retains finite endpoint pairs, essential tails and all copies of repeated bars.

Modules, grids and endpoint conventions

Fix a field k : Type u, a module M : PModule.{u,0,u} k ℝ and an increasing grid c : Fin n ↪o ℝ. The diagram moduleDiagram c M hfd is obtained from the restriction of M to its critical fibers, assumed finite dimensional by hfd. Its intervals are [birth,death): the left endpoint is included, the finite right endpoint is excluded, and death at the finite chain's terminal endpoint represents an essential tail with death ⊤. An empty grid selects no critical stage. The padded zero stage of reconstruction receives no real grade.

A finite diagram is a multiset of endpoint pairs. Its finite and essential occurrence types retain separate labels for repeated pairs or births. The bottleneck matching API uses a partial injection on finite occurrences and a total equivalence on essential occurrences. Finite and essential occurrences cannot be cross-matched.

Positive evaluation and the diagram formula

For ε : ℝ≥0, the shift evaluates at t+ε, so a bar [b,d) becomes [b−ε,d−ε) and [b,∞) becomes [b−ε,∞). The existing translateDiagram D a adds a to both finite endpoints, or to an essential birth while preserving ⊤. Thus the diagram translation parameter is −ε:

\[ D_{c-\varepsilon}(S_\varepsilon M) =\operatorname{translateDiagram}(D_c(M),-\varepsilon). \]

shiftedCriticalIso compares the restrictions on shiftGrid ε c and c. At every critical index their real parameters agree by (c_i−ε)+ε=c_i. Applying M to this index comparison gives a natural isomorphism, not just equality of dimensions. Finiteness of each shifted critical fiber follows from the same index equality. Rank-barcode invariance under this natural isomorphism gives the same finite interval multiset; ofIndexBars_shiftGrid then moves each real endpoint. Its finite-death and terminal-death cases preserve the half-open and essential conventions. moduleDiagram_shift is the resulting formula. It needs no constructibility assumption on the native shifted grid.

If M is constructible on c, its shifted module is constructible on shiftGrid ε c. For any refinement e of that shifted grid, the refinement theorem derives finite dimensionality at every added critical fiber and preserves the real diagram. moduleDiagram_shift_refinement therefore proves the same formula on e. In particular e may be the constructed common grid with another source grid; no finiteness witness at those new observations is assumed. Translating this shifted diagram by +ε recovers D_c(M).

Occurrence bijections and matchings

finiteTranslationEquiv D a and essentialTranslationEquiv D a send the original occurrence labels bijectively to those of translateDiagram D a. Their value theorems give (b+a,d+a) and b+a. Their inverse value theorems give (b−a,d−a) and b−a. Consequently the inverse for a=−ε adds ε, which is the barcode correspondence usually denoted by rε.

These are equivalences of labels, including all copies with equal values. They do not deduplicate a multiset. At ε=0 the endpoint values are unchanged; a choice of labels for repeated values need not be definitionally the identity.

translatedMatching D a is a complete matching made from these bijections. Every finite target is selected, so no finite label is sent to the diagonal. Its maximum cost is at most |a|: each finite pair and each essential birth moves exactly that amount, while an empty maximum is zero. exists_shift_matching and exists_shift_refinement_matching apply this construction to the module diagrams above, with every finite endpoint value, finite-target coverage, essential birth and cost bound proved. They compare M with its own shift and require no supplied matching or endpoint estimate.

For an existing matching from D to translateDiagram E a, OccurrenceMatching.untranslateTarget returns the target labels to E by the inverse equivalences. A finite source is matched to j exactly when the old matching selected the translated occurrence of j. Unmatched sources are the same; the selected target range is transported by that same bijection. The essential target birth changes from b to b−a. This relabelling is an operation on a matching. Translating only its target can change its pair costs, so it supplies no cost-invariance claim.

Boundary consumers

A finite-barcode pullback has two copies of [1,3) and one essential bar [3,∞). Its positive evaluation shift by 2 has two copies of [-1,1) and one essential bar [1,∞). The consumers check both exact finite endpoints, the essential birth and unbounded tail, and the preserved multiplicities. Observations before, inside and after the shifted grid retain these diagrams using derived finiteness. The occurrence inverse has the stated endpoint values, and untranslating the constructed matching recovers each original label. Parameter zero and the empty grid are also checked in TDA.Persistence.ConstructibleShiftBarcodeExamples.

Source correspondence and the algebraic-stability dependency

Bauer–Lesnick v4, §§2.1–2.3, defines interval modules and multiset barcode occurrences, including decorated endpoints. Section 3.1 defines the positive-evaluation shift and its leftward barcode displacement. The bijection immediately preceding Theorem 6.4 in §6.1 adds ε back to shifted endpoints. The formula and inverse here are a checked specialization to the library's finite, zero-left, right-continuous half-open convention. The grid/rank proof, occurrence representation and target-relabel operation are library constructions. Source and rights metadata are in BL15-v4.

The real induced matching constructs a matching into the shifted target, with exact 2ε endpoint and unmatched-length bounds and a total essential equivalence. The isometry construction applies the +ε target inverse, proves the ε pair/diagonal cost bounds and takes the distance infimum to obtain d_B=d_I for finite-constructible modules. The shift-specific construction above establishes its own complete matching and cost bound. General infinite decomposition and the source's omitted duality proof are outside its scope.