C[0,1] as a measurable space, the polygonal interpolation of a random walk, and its rescaled path
DefinitionDurrettProbability_DonskerThe path-space apparatus of section 8.1 of Durrett, Probability: Theory and Examples.
The space. is the continuous real functions on , carrying the topology of uniform convergence — the compact-open topology, which agrees with the uniform one because the domain is compact — and its Borel -algebra. Both the measurable-space structure and the fact that it is the Borel structure of that topology are supplied here; the ambient library has neither for a space of continuous maps.
The interpolation. For a sequence of real random variables and ,
At an integer this is , and on it interpolates linearly between consecutive partial sums, so it is Durrett's . Written this way it is a finite sum of continuous functions of .
The rescaled path. for , an element of .
Restriction of a path. For , its restriction to as an element of when is continuous, and the zero path otherwise; applied to a process this gives the -valued random variable .
Formalization Note Indexing runs from zero, so the partial sum is and the interpolation of the first steps is determined on ; beyond it is constant at , which is irrelevant because the rescaled path only evaluates it on . At the rescaling divides by , which in the reals is ; that value plays no role in a limit along .
The restriction map is total, returning the zero path on a discontinuous argument rather than being undefined. Every statement in this mission assumes each path of the process is continuous, so the branch is never taken; this is a formalization convenience, and it is also what makes "the Brownian path as a random element of " a well-formed object at all.
import Mathlib
import Definitions.Def_DurrettProbability_Brownian
open Filter MeasureTheory ProbabilityTheory
open scoped NNReal Topology
namespace DurrettProbability
/-- `C[0,1]`, the paths of Donsker's theorem, with the uniform topology and its Borel σ-algebra.
Durrett, *Probability: Theory and Examples*, Theorem 8.1.4. -/
abbrev PathSpace : Type := C(Set.Icc (0 : ℝ) 1, ℝ)
noncomputable instance : MeasurableSpace PathSpace := borel _
instance : BorelSpace PathSpace := ⟨rfl⟩
variable {Ω : Type*}
/-- The polygonal interpolation `S(u)` of the first `n` steps of the walk `X`: it agrees with
`S_m = X_1 + ⋯ + X_m` at integer times `m ≤ n` and is linear in between. Writing it as
`∑_{k<n} X_k · clamp(u - k)` makes it a finite sum of continuous functions.
Durrett, *Probability: Theory and Examples*, section 8.1. -/
noncomputable def walkInterp (X : ℕ → Ω → ℝ) (n : ℕ) (ω : Ω) (u : ℝ) : ℝ :=
∑ k ∈ Finset.range n, X k ω * max 0 (min 1 (u - k))
/-- The rescaled polygonal path `t ↦ S(nt)/√n` on `[0,1]`, the random element of `C[0,1]` whose
laws Donsker's theorem says converge. Durrett, Theorem 8.1.4. -/
noncomputable def walkPath (X : ℕ → Ω → ℝ) (n : ℕ) (ω : Ω) : PathSpace :=
⟨fun t => walkInterp X n ω (n * (t : ℝ)) / Real.sqrt n, by unfold walkInterp; fun_prop⟩
open Classical in
/-- The restriction of a path `f : [0,∞) → ℝ` to `[0,1]` as an element of `C[0,1]`, and `0` when
`f` is not continuous. The junk branch is never reached in the statements below, which assume
every path of the process is continuous. -/
noncomputable def toPath (f : ℝ≥0 → ℝ) : PathSpace :=
if h : Continuous f then
⟨fun t => f ⟨t, t.2.1⟩, h.comp (by fun_prop)⟩
else 0
/-- A Brownian motion viewed as a random element of `C[0,1]`. -/
noncomputable def brownianPath (B : ℝ≥0 → Ω → ℝ) (ω : Ω) : PathSpace := toPath (fun t => B t ω)
end DurrettProbability