Gluing of the path-component functor

Statement

Let X and Y be topological spaces, f : C(X, Y), W an open set of Y, and U : ι → Opens Y a family of opens with U i ≤ W for every i and W ≤ ⨆ i, U i. Write C_f for the open-set path-component functor componentFunctor f (reeb-components.md). If maps a i : C_f(U i) → A agree after both inclusions from every overlap,

a i (C_f.map (U i ⊓ U j ≤ U i) c) = a j (C_f.map (U i ⊓ U j ≤ U j) c)

for all i, j and every class c over U i ⊓ U j, then there is a unique g : C_f(W) → A with g ∘ C_f(U i ≤ W) = a i for every i (TDA.Reeb.componentFunctor_existsUnique_glue). The test object A lies in an arbitrary Sort. Agreement is on classes over the overlap, not on labels of components.

The empty cover is included. It forces C_f(W) to be empty (Examples.emptyCover_isEmpty), and then the unique g is the map out of the empty type.

Colimit forms

The statement is derived from the elementary form of the cosheaf condition (cosheaf.md):

  • componentFunctor_exists_map_eq: every class over W comes from a member. A point of the class has its image in some U i (Opens.mem_iSup).
  • componentFunctor_map_eq_iff_eqvGen: two local classes have the same image over W iff they are related by the equivalence closure of overlapRel C_f U. The forward direction takes the chain of componentFunctor_reflTransGen_of_mk_eq and turns each shared chain point into an overlap step with overlapRel_mk. The backward direction holds because both inclusions of a class over an overlap send it to the class of the same point over W.

Hence C_f sends every cocone over the pairwise-intersection diagram whose apex is covered by the members to a colimit (componentFunctor_isColimit). The cases coverCocone U hU and mathlib's Pairwise.cocone U give componentFunctor_isColimit_coverCocone and componentFunctor_isPairwiseCosheaf: C_f satisfies IsPairwiseCosheaf.{w} for every cover-index universe w.

Naturality

For Z : Type u, g : C(Z, Y) and h : C(X, Z) with g ∘ h = f, gluing commutes with the induced transformation componentNatTrans h: if G glues a family a for C_g, and G' glues the family a i ∘ (componentNatTrans h).app (U i) for C_f, then G' = G ∘ (componentNatTrans h).app W (componentFunctor_glue_natTrans). The fixture Examples.collapse_glue_identifies applies this to the collapse of the discrete two-point space over the one-point base. The glue for the collapsed map identifies two classes that are distinct over the two-point space.

Universes and index sorts

The carrier X : Type u and the base Y : Type v are independent, and the cover index ι : Type w is independent of both. The functor lands in Type u, so the colimit forms test only cocones with apex in Type u; the raw statement covers every Sort. A cover indexed by an arbitrary Sort is reindexed through PLift, which preserves the union (Examples.existsUnique_glue_sort). No smallness assumption is used.

Acceptance cases

  • Two overlapping intervals (-∞, 2/3) and (1/3, ∞) covering ℝ: the class of 0 over the first and the class of 1 over the second are related through overlap steps (Examples.twoOpens_classes_related), and the colimit form holds (Examples.twoOpens_isColimit).
  • Disjoint cover of ℝ \ {0} by (-∞, 0) and (0, ∞): an overlap step never changes the member, so the classes of -1 and 1 over the union differ (Examples.negPos_classes_ne). Gluing makes no false identification.
  • Empty cover (Examples.emptyCover_isEmpty, Examples.emptyCover_existsUnique_glue) and empty open with a nonempty cover (Examples.bot_isColimit).
  • Duplicate indices, ⊤ twice: the two local classes of one point are related (Examples.dupCover_related), and the colimit form holds (Examples.dupCover_isColimit).
  • Naturality at the collapse (Examples.collapse_glue_identifies).

Source correspondence

de Silva, Munch and Patel prove that the path-component pre-cosheaf of an ℝ-space is a cosheaf (Categorified Reeb Graphs, Proposition 3.15), in the universal-property form (3.5)–(3.6) of §3.2. Their proof defines the glued map on the class of a point y through any member containing f(y), shows independence of the member by the consistency condition, and shows independence of the point by local constancy along a path. The Lean proof follows the same argument in factored form. Independence of the member is the overlap step, independence of the point is the chain of the path-local cover lemma (path-cover.md), and the glued map is chosen through a representative. Uniqueness holds because every class comes from a member.

Library generalizations, chosen by the library and not printed in the paper: an arbitrary base space instead of ℝ, arbitrary opens and open covers instead of open intervals and interval covers, and covers indexed by any type or, through PLift, any sort. The empty cover is explicit in Lean and implicit in the source's convention that the value at the empty set is empty. Restricting to open intervals of ℝ recovers the source statement. The result concerns path components of inverse images. The source notes that it fails for connected components (Example 3.20); no constructibility or finiteness is assumed or concluded.

Sources: de Silva, Munch and Patel, Categorified Reeb Graphs.