Documentation

TauCeti.Probability.Process.PathLaw.FiniteMarginals

Finite-dimensional marginal uniqueness #

A finite measure on path space ℕ → α is determined by its finite prefix marginals: any measure agreeing with it on every prefix projection (prefixProj α n, the projection to the first n coordinates) is equal to it. This is a thin ℕ-prefix wrapper over Mathlib's projective-limit machinery (IsProjectiveLimit.unique), not new measure theory.

The public API:

The paired form reduces to the unpaired one rather than reproving it. Mathlib's IsProjectiveLimit is stated for dependent products ∀ i, α i, and T × (ℕ → α) is not of that shape — but replicating the first factor at every coordinate, (t, x) ↦ fun n => (t, x n), embeds it in ℕ → T × α, which is. That map has an explicit measurable left inverse, so it is injective on measures, and each of its prefix marginals is read off the paired prefix marginal one step longer.

Both apply directly to probability measures, since IsProbabilityMeasure provides IsFiniteMeasure, so no separate probability-measure theorem is needed.

Finite-marginal uniqueness. Two measures on ℕ → α, with μ finite, that have the same law under every finite prefix projection prefixProj α n are equal. (Finiteness of ν is not needed: projective-limit uniqueness only requires the prefix-marginal family, supplied by μ, to be finite.)

Finite-marginal uniqueness, setwise form: two measures on ℕ → α, with μ finite, agreeing on every measurable prefix-cylinder are equal. It assumes only μ is finite; ν's finiteness is forced by the conclusion.

def TauCeti.Probability.prefixProjPair (T : Type u_2) (α : Type u_3) (n : ℕ) :
T × (ℕ → α) → T × (Fin n → α)

The prefix map onto the first n path coordinates, keeping the first factor.

Equations
Instances For
    @[simp]
    theorem TauCeti.Probability.prefixProjPair_apply {T : Type u_2} {α : Type u_3} (n : ℕ) (q : T × (ℕ → α)) :
    prefixProjPair T α n q = (q.1, fun (i : Fin n) => q.2 ↑i)

    Applying the paired prefix map.

    The paired prefix projection is measurable: it keeps the first factor and reads finitely many path coordinates.

    Paired prefix marginals #

    theorem TauCeti.Probability.prefixProjPair_comp {T : Type u_2} {α : Type u_3} {m n : ℕ} (hmn : m ≤ n) :
    prefixProjPair T α m = (fun (r : T × (Fin n → α)) => (r.1, fun (i : Fin m) => r.2 (Fin.castLE hmn i))) ∘ prefixProjPair T α n

    Longer prefixes refine shorter ones.

    Paired finite-marginal uniqueness. Two measures on T × (ℕ → α) that agree under every prefix projection — keeping the T coordinate — are equal.

    As with measure_eq_of_prefixProj_map_eq, only one of the two measures need be assumed finite: the n = 0 projection already forces the total masses to agree, since prefixProjPair retains the first factor even at the empty prefix.