Theorem 8.1.2 — the whole random walk embedded in a Brownian motion
OpenDurrettProbability.skorokhod_embedding_walkLet be a probability measure on with mean zero and variance one: and , both integrals absolutely convergent.
Then there exist a probability space, a Brownian motion on it, an increasing family of -fields containing the Brownian past , and times such that
- each is a stopping time for , and with everywhere;
- the process has the law of the partial-sum process of an i.i.d. sequence with step law — that is, the push-forward of on the sequence space equals the push-forward of under the infinite product measure ;
- the gaps are independent, and each has the law of .
Iterating Skorokhod's representation puts the entire walk inside one Brownian motion, read at a sequence of stopping times whose increments are i.i.d.
Formalization Note As in Theorem 8.1.1 the probability space is produced rather than assumed, for the same reason: the construction needs an i.i.d. sequence of auxiliary pairs independent of the Brownian motion.
The book's "" is read as equality of the laws of the whole processes on sequence space, not as equality of one-dimensional marginals for each ; the latter would be strictly weaker and would not support the proof of Donsker's theorem, where the whole embedded path is compared with the Brownian one. Indexing starts at with the empty sum, matching .
The gaps are taken as real numbers, differences of the coercions of and ; since is monotone these are non-negative. Independence is joint independence of the whole family, not merely pairwise.
The hypotheses are satisfiable — any centred law of variance one, for instance the standard Gaussian or the symmetric two-point law — so the statement is not vacuous.
import Mathlib import Definitions.Def_DurrettProbability_Brownian import Definitions.Def_DurrettProbability_Donsker open Filter MeasureTheory ProbabilityTheory open scoped NNReal Topology
namespace DurrettProbability
theorem skorokhod_embedding_walk (μ : Measure ℝ) [IsProbabilityMeasure μ]
(hint : Integrable id μ) (hmean : ∫ x, x ∂μ = 0)
(hsq : Integrable (fun x : ℝ => x ^ 2) μ) (hvar : ∫ x : ℝ, x ^ 2 ∂μ = 1) :
∃ (Ω : Type) (mΩ : MeasurableSpace Ω) (P : @Measure Ω mΩ) (_ : IsProbabilityMeasure P)
(B : ℝ≥0 → Ω → ℝ) (F : ℝ≥0 → MeasurableSpace Ω) (T : ℕ → Ω → ℝ≥0),
IsBrownianReal B P ∧
(∀ s t : ℝ≥0, s ≤ t → F s ≤ F t) ∧ (∀ t, F t ≤ mΩ) ∧ (∀ t, pastSigma B t ≤ F t) ∧
(∀ n, ∀ t : ℝ≥0, MeasurableSet[F t] {ω | T n ω ≤ t}) ∧
T 0 = 0 ∧ (∀ n ω, T n ω ≤ T (n + 1) ω) ∧
Measure.map (fun ω (n : ℕ) => B (T n ω) ω) P
= Measure.map (fun (x : ℕ → ℝ) (n : ℕ) => ∑ k ∈ Finset.range n, x k)
(Measure.infinitePi fun _ : ℕ => μ) ∧
iIndepFun (fun n ω => (T (n + 1) ω : ℝ) - (T n ω : ℝ)) P ∧
(∀ n, IdentDistrib (fun ω => (T (n + 1) ω : ℝ) - (T n ω : ℝ))
(fun ω => (T 1 ω : ℝ) - (T 0 ω : ℝ)) P P) := by sorry
end DurrettProbabilityConfirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.