Almost-sure continuity of the Girsanov drift-shifted process under the tilted measure
ProvedAppliedSDE.SarkkaSolin.girsanov_shifted_path_continuous_coreapplied-sdebrownianchange-of-measuregirsanovstochastic-calculus
After exponential change of measure Q, the drift-corrected Brownian path has an almost-sure continuous version. Transfer the original Brownian almost-sure path continuity through Q absolutely continuous with respect to P, and use time-local integrability of the progressive drift. This is the path-regularity part of Girsanov.
Preamble
import Mathlib import Definitions.Def_KLSModular_LevyStochCalc_Brownian_Filtered import Definitions.Def_KLSModular_LevyStochCalc_Brownian_ItoL2CompletionMaster open MeasureTheory ProbabilityTheory Filter LevyStochCalc.Brownian open scoped ENNReal NNReal Topology
Formal statement
theorem AppliedSDE.SarkkaSolin.girsanov_shifted_path_continuous_core
{Ω : Type*} [MeasurableSpace Ω]
(P : Measure Ω) [IsProbabilityMeasure P]
(W : LevyStochCalc.Brownian.BrownianMotion P)
(ℱ : Filtration ℝ ‹MeasurableSpace Ω›)
(hW : LevyStochCalc.Brownian.IsBrownianFiltration W ℱ)
(T : ℝ) (hT : 0 < T)
(θ : Ω → ℝ → ℝ)
(hθmeas : Measurable (Function.uncurry θ))
(hθprog : LevyStochCalc.Probability.ProgressivelyMeasurable ℱ θ)
(hθq : ∀ (t : ℝ), 0 < t →
∫⁻ ω, ∫⁻ s in Set.Icc (0 : ℝ) t,
(‖θ ω s‖₊ : ℝ≥0∞) ^ 2 ∂volume ∂P < ⊤)
(hθzero : ∀ (ω : Ω) (s : ℝ), T ≤ s → θ ω s = 0)
(Z : Ω → ℝ)
(hZ : ∀ ω : Ω,
Z ω = Real.exp (
LevyStochCalc.Brownian.Ito.stochasticIntegralBrownian W ℱ hW θ hθmeas hθprog hθq T ω -
(1 / 2 : ℝ) * ∫ s in (0 : ℝ)..T, θ ω s ^ 2))
(hZmean : ∫ ω, Z ω ∂P = 1) :
let Q : Measure Ω := P.withDensity (fun ω => ENNReal.ofReal (Z ω))
let shifted : ℝ≥0 → Ω → ℝ := fun t ω =>
W.W (t : ℝ) ω - ∫ s in (0 : ℝ)..(t : ℝ), θ ω s
(∀ᵐ ω ∂Q, Continuous (fun t : ℝ≥0 => shifted t ω)) := by
sorrySource
Sarkka and Solin (2019) Girsanov theorem, Chapter 7; Mathlib IsBrownianReal extends IsPreBrownianReal with the cont field.