Restriction independence and vertex-tail triviality (AHK umbrella, R3b / #106) #
The generic equivalence between dissociation and restriction independence, and the implication to vertex-tail triviality, for exchangeable relational laws:
InfiniteRelExchangeableLaw.RestrictionIndependent— for every block size the initial and after-block σ-algebras are independent;InfiniteRelExchangeableLaw.VertexTailTrivial— every vertex-tail event has law-measure0or1;isDissociated_iff_restrictionIndependent— dissociation ↔ restriction independence: independence of the comap σ-algebras of the two block maps is the block-pair map factorization (IndepFun_iff_Indep+indepFun_iff_map_prod_eq_prod_map_map), and the finite tail windows exhaust the after-block σ-algebra (R3a);RestrictionIndependent.vertexTailTrivial— a vertex-tail event is independent of every initial σ-algebra, hence (the initial σ-algebras generating, R3a) of itself.VertexTailTrivial.isDissociated/isDissociated_iff_vertexTailTrivial— the closing arrow, representation-free: condition the initial-event indicator on successively later diagonal tail algebras; Lévy's downward theorem (Graphon/LevyDownward.lean) converges the conditional expectations to the vertex-tail one, which tail triviality makes a.e. constant; exchangeability keeps the joint mass with an arbitrarily far window constant (law_map_restrict_pair); in the limit the block mass factorizes exactly. (The undirected proof instead rides the graphon mixture representation; the generic theory keeps the characterization representation-free, with R5's Dirac-mixing as a later corollary.)
Window monotonicity #
The finite tail windows are monotone in the window size.
The two properties #
Restriction independence: for every block size, the structure on the initial block is independent of the structure on the remaining vertices.
Equations
- M.RestrictionIndependent = ∀ (k : S.Srt → ℕ), ProbabilityTheory.Indep (RelSignature.RelStructure.initialAlgebra k) (RelSignature.RelStructure.tailAlgebra k) ↑M.law
Instances For
Vertex-tail triviality: every vertex-tail event has law-measure 0 or 1.
Equations
- M.VertexTailTrivial = ∀ (s : Set (S.RelStructure S.Vinfinite)), MeasurableSet s → ↑M.law s = 0 ∨ ↑M.law s = 1
Instances For
Dissociation ↔ restriction independence #
Finite-window independence from dissociation: comap-σ-algebra independence of the initial block and a tail window is exactly the block-pair map factorization.
Dissociation implies restriction independence: the finite windows exhaust the after-block σ-algebra.
Restriction independence implies dissociation: restrict the independence to a window and read it as the block-pair map factorization.
Dissociation ↔ restriction independence (R3b).
Restriction independence implies vertex-tail triviality #
Restriction independence implies vertex-tail triviality: a vertex-tail event is independent of every initial σ-algebra, hence — the initial σ-algebras generating the Borel σ-algebra — independent of itself.
Dissociation implies vertex-tail triviality (R3b chain).
Vertex-tail triviality implies dissociation (the closing arrow) #
Vertex-tail triviality implies dissociation (representation-free): condition the initial-event indicator on successively later diagonal tail algebras; Lévy's downward theorem converges the conditional expectations to the vertex-tail one, which tail triviality makes a.e. constant; exchangeability keeps the joint mass with an arbitrarily far window constant; in the limit the block mass factorizes exactly.
Dissociation ↔ vertex-tail triviality (R3b complete, representation-free).