Exchangeable relational laws: size-vector marginals and consistency (AHK umbrella #103, R2a) #
The finite-marginal layer of the generic exchangeable-law theory (issue #105). A generic
exchangeable law is given by a size-vector-indexed family of probability marginals that
is consistent under arbitrary sortwise injections — the multi-sorted, arbitrary-arity
analogue of Graphon.ExchangeableGraphLaw.
Crucially the finite marginals are ProbabilityMeasure (RelStructure S (Vfinite n)), not
PMF: with countably many relation symbols a finite vertex carrier still has infinitely
many Boolean coordinates, so RelStructure S (Vfinite n) can be an (uncountable) standard-Borel
space and the law can be non-atomic. (D2 recovers PMFs in the finite-signature case.)
RelSignature.exists_const_ge— becauseS.Srtis finite, the diagonal size vectors are cofinal (every size vector is bounded by a constant one);RelSignature.RelExchangeableLaw— the structure: probability marginals + arbitrary sortwise-injection consistency;RelExchangeableLaw.marginal_map_restrictLE/marginal_map_perm— consistency under the standard size-vector inclusion, and finite exchangeability (permutation invariance) of the marginals, both special cases.
The compactness-based extension to an infinite law is R2b; uniqueness (via
ext_of_map_restrictFin) and the finite/infinite equivalence are R2c.
Diagonal cofinality: since S.Srt is finite, every size vector is bounded above by
a constant (diagonal) size vector — so diagonal marginals suffice for the extension.
An exchangeable relational law: a size-vector-indexed family of probability marginals
on the finite structure spaces, consistent under every sortwise injection
∀ s, Fin (n s) ↪ Fin (m s) (the m-marginal restricts to the n-marginal).
- marginal (n : S.Srt → ℕ) : MeasureTheory.ProbabilityMeasure (S.RelStructure (Vfinite n))
The probability marginal on structures over the size vector
n. - consistent {n m : S.Srt → ℕ} (e : (s : S.Srt) → Fin (n s) ↪ Fin (m s)) : MeasureTheory.Measure.map (RelStructure.restrict e) ↑(self.marginal m) = ↑(self.marginal n)
Consistency under every sortwise injection.
Instances For
Consistency under the standard size-vector inclusion n ≤ m.
Finite exchangeability: each marginal is invariant under the sortwise permutations of
its finite vertex sets — the special case n = m, e a permutation.