AutoSamplingTheory.TechnicalLemmas.StochasticProcesses.CanonicalRawLocalizationL2
3 named declarations scanned from AutoSamplingTheory/TechnicalLemmas/StochasticProcesses/CanonicalRawLocalizationL2.lean.
Declarations
theorem AutoSamplingTheory.TechnicalLemmas.StochasticProcesses.CanonicalRawLocalizationL2.rawStoppedTimeLintegral_le Partial Not mapped
- On each sample path, the literal source stopped integrand has nonnegative Lebesgue energy at most the canonical level `n+1`.
theorem rawStoppedTimeLintegral_le
(hUsual : SatisfiesUsualConditions filtration mu)
(eta : LocalProgressiveL2Integrand filtration mu T)
(n : ℕ) (omega : Omega) :
(∫⁻ t in Icc (0 : ℝ≥0) T,
ENNReal.ofReal
((Localization.stoppedIntegrand eta.process
(fun w =>
(canonicalRawLocalizingTime hUsual eta n w : WithTop ℝ≥0))
t omega) ^ 2)
∂TimeMeasure.nnrealLebesgue) ≤ ENNReal.ofReal (n + 1 : ℝ) := by
change
(∫⁻ t,
ENNReal.ofReal
((Localization.stoppedIntegrand eta.process
(fun w =>
(canonicalRawLocalizingTime hUsual eta n w : WithTop ℝ≥0))
t omega) ^ 2)
∂(TimeMeasure.nnrealLebesgue.restrict (Icc (0 : ℝ≥0) T))) ≤
ENNReal.ofReal (n + 1 : ℝ)
rw [← TimeMeasure.upTo_eq_restrict_nnrealLebesgue T]
have hae := stoppedIntegrand_ae_eq_energyStoppedIntegrand
hUsual eta n omega
have hsq :
(fun t => ENNReal.ofReal
((Localization.stoppedIntegrand eta.process
(fun w =>
(canonicalRawLocalizingTime hUsual eta n w : WithTop ℝ≥0))
t omega) ^ 2)) =ᵐ[TimeMeasure.upTo T]
(fun t => ENNReal.ofReal
((energyStoppedIntegrand hUsual eta (n + 1 : ℝ) t omega) ^ 2)) := by
filter_upwards [hae] with t ht
rw [ht]
rw [lintegral_congr_ae hsq]
rw [← ofReal_integral_eq_lintegral_ofReal
(sectionSquare_integrable hUsual eta (n + 1 : ℝ) omega)
(ae_of_all _ fun _ => sq_nonneg _)]
exact ENNReal.ofReal_le_ofReal
(integral_energyStoppedIntegrand_sq_le hUsual eta (by positivity) omega)
/-- Integrating the pathwise `n+1` bound over a probability space gives the
finite expected stopped energy required by Definition 1.1.12. -/
AutoSamplingTheory/TechnicalLemmas/StochasticProcesses/CanonicalRawLocalizationL2.lean:29published source at ed709a7f320d
theorem AutoSamplingTheory.TechnicalLemmas.StochasticProcesses.CanonicalRawLocalizationL2.rawStoppedProductEnergy_lt_top Partial Not mapped
- Integrating the pathwise `n+1` bound over a probability space gives the finite expected stopped energy required by Definition 1.1.12.
theorem rawStoppedProductEnergy_lt_top
[IsProbabilityMeasure mu]
(hUsual : SatisfiesUsualConditions filtration mu)
(eta : LocalProgressiveL2Integrand filtration mu T)
(n : ℕ) :
(∫⁻ omega, ∫⁻ t in Icc (0 : ℝ≥0) T,
ENNReal.ofReal
((Localization.stoppedIntegrand eta.process
(fun w =>
(canonicalRawLocalizingTime hUsual eta n w : WithTop ℝ≥0))
t omega) ^ 2)
∂TimeMeasure.nnrealLebesgue ∂mu) < ∞ := by
calc
(∫⁻ omega, ∫⁻ t in Icc (0 : ℝ≥0) T,
ENNReal.ofReal
((Localization.stoppedIntegrand eta.process
(fun w =>
(canonicalRawLocalizingTime hUsual eta n w : WithTop ℝ≥0))
t omega) ^ 2)
∂TimeMeasure.nnrealLebesgue ∂mu) ≤
∫⁻ _omega, ENNReal.ofReal (n + 1 : ℝ) ∂mu := by
apply lintegral_mono
intro omega
exact rawStoppedTimeLintegral_le hUsual eta n omega
_ = ENNReal.ofReal (n + 1 : ℝ) := by simp
_ < ∞ := ENNReal.ofReal_lt_top
/-- Chewi Proposition 1.1.13 in the exact repository-native source contract:
the canonical raw times form a `Localization.IsLocalizingSequence` for the
original progressive locally square-integrable integrand. -/
AutoSamplingTheory/TechnicalLemmas/StochasticProcesses/CanonicalRawLocalizationL2.lean:71published source at ed709a7f320d
theorem AutoSamplingTheory.TechnicalLemmas.StochasticProcesses.CanonicalRawLocalizationL2.canonicalRaw_isLocalizingSequence Partial Not mapped
- Chewi Proposition 1.1.13 in the exact repository-native source contract: the canonical raw times form a `Localization.IsLocalizingSequence` for the original progressive locally square-integrable integrand.
theorem canonicalRaw_isLocalizingSequence
[IsProbabilityMeasure mu]
(hUsual : SatisfiesUsualConditions filtration mu)
(eta : LocalProgressiveL2Integrand filtration mu T) :
Localization.IsLocalizingSequence eta.process filtration mu T
(fun n omega =>
(canonicalRawLocalizingTime hUsual eta n omega : WithTop ℝ≥0)) := by
refine ⟨eta.progressive, ?_, ?_, ?_, ?_⟩
· intro n
exact canonicalRawLocalizingTime_isStoppingTime hUsual eta n
· intro n k hnk omega
simpa only [WithTop.coe_le_coe] using
((canonicalRawLocalizingTime_mono hUsual eta) hnk omega)
· intro n
exact rawStoppedProductEnergy_lt_top hUsual eta n
· exact canonicalRawLocalizingTime_tendsto_terminal_ae hUsual eta
end CanonicalRawLocalizationL2
end StochasticProcesses
end TechnicalLemmas
end AutoSamplingTheory
AutoSamplingTheory/TechnicalLemmas/StochasticProcesses/CanonicalRawLocalizationL2.lean:101published source at ed709a7f320d
Complete Lean source
AutoSamplingTheory/TechnicalLemmas/StochasticProcesses/CanonicalRawLocalizationL2.lean — full module with imports and context
import AutoSamplingTheory.TechnicalLemmas.StochasticProcesses.CanonicalRawLocalization
/-!
# L2 certificate for the source-facing canonical localizing sequence
This file closes the semantic gap between Chewi's raw stopped integrand
`eta_t 1_{t <= tau_n}` and the completed representative used internally by
the global L2 Ito construction. The two agree almost everywhere in time,
so the completed pathwise energy estimate transfers to the literal raw
localizing-sequence predicate.
-/
namespace AutoSamplingTheory
namespace TechnicalLemmas
namespace StochasticProcesses
namespace CanonicalRawLocalizationL2
open Filter MeasureTheory Set
open scoped ENNReal NNReal Topology
open ProgressiveL2 LocalProgressiveL2 CanonicalRawLocalization
EnergyStoppedIntegrand
variable {Omega : Type*} {m : MeasurableSpace Omega}
{filtration : Filtration ℝ≥0 m} {mu : Measure Omega} {T : ℝ≥0}
/-- On each sample path, the literal source stopped integrand has nonnegative
Lebesgue energy at most the canonical level `n+1`. -/
theorem rawStoppedTimeLintegral_le
(hUsual : SatisfiesUsualConditions filtration mu)
(eta : LocalProgressiveL2Integrand filtration mu T)
(n : ℕ) (omega : Omega) :
(∫⁻ t in Icc (0 : ℝ≥0) T,
ENNReal.ofReal
((Localization.stoppedIntegrand eta.process
(fun w =>
(canonicalRawLocalizingTime hUsual eta n w : WithTop ℝ≥0))
t omega) ^ 2)
∂TimeMeasure.nnrealLebesgue) ≤ ENNReal.ofReal (n + 1 : ℝ) := by
change
(∫⁻ t,
ENNReal.ofReal
((Localization.stoppedIntegrand eta.process
(fun w =>
(canonicalRawLocalizingTime hUsual eta n w : WithTop ℝ≥0))
t omega) ^ 2)
∂(TimeMeasure.nnrealLebesgue.restrict (Icc (0 : ℝ≥0) T))) ≤
ENNReal.ofReal (n + 1 : ℝ)
rw [← TimeMeasure.upTo_eq_restrict_nnrealLebesgue T]
have hae := stoppedIntegrand_ae_eq_energyStoppedIntegrand
hUsual eta n omega
have hsq :
(fun t => ENNReal.ofReal
((Localization.stoppedIntegrand eta.process
(fun w =>
(canonicalRawLocalizingTime hUsual eta n w : WithTop ℝ≥0))
t omega) ^ 2)) =ᵐ[TimeMeasure.upTo T]
(fun t => ENNReal.ofReal
((energyStoppedIntegrand hUsual eta (n + 1 : ℝ) t omega) ^ 2)) := by
filter_upwards [hae] with t ht
rw [ht]
rw [lintegral_congr_ae hsq]
rw [← ofReal_integral_eq_lintegral_ofReal
(sectionSquare_integrable hUsual eta (n + 1 : ℝ) omega)
(ae_of_all _ fun _ => sq_nonneg _)]
exact ENNReal.ofReal_le_ofReal
(integral_energyStoppedIntegrand_sq_le hUsual eta (by positivity) omega)
/-- Integrating the pathwise `n+1` bound over a probability space gives the
finite expected stopped energy required by Definition 1.1.12. -/
theorem rawStoppedProductEnergy_lt_top
[IsProbabilityMeasure mu]
(hUsual : SatisfiesUsualConditions filtration mu)
(eta : LocalProgressiveL2Integrand filtration mu T)
(n : ℕ) :
(∫⁻ omega, ∫⁻ t in Icc (0 : ℝ≥0) T,
ENNReal.ofReal
((Localization.stoppedIntegrand eta.process
(fun w =>
(canonicalRawLocalizingTime hUsual eta n w : WithTop ℝ≥0))
t omega) ^ 2)
∂TimeMeasure.nnrealLebesgue ∂mu) < ∞ := by
calc
(∫⁻ omega, ∫⁻ t in Icc (0 : ℝ≥0) T,
ENNReal.ofReal
((Localization.stoppedIntegrand eta.process
(fun w =>
(canonicalRawLocalizingTime hUsual eta n w : WithTop ℝ≥0))
t omega) ^ 2)
∂TimeMeasure.nnrealLebesgue ∂mu) ≤
∫⁻ _omega, ENNReal.ofReal (n + 1 : ℝ) ∂mu := by
apply lintegral_mono
intro omega
exact rawStoppedTimeLintegral_le hUsual eta n omega
_ = ENNReal.ofReal (n + 1 : ℝ) := by simp
_ < ∞ := ENNReal.ofReal_lt_top
/-- Chewi Proposition 1.1.13 in the exact repository-native source contract:
the canonical raw times form a `Localization.IsLocalizingSequence` for the
original progressive locally square-integrable integrand. -/
theorem canonicalRaw_isLocalizingSequence
[IsProbabilityMeasure mu]
(hUsual : SatisfiesUsualConditions filtration mu)
(eta : LocalProgressiveL2Integrand filtration mu T) :
Localization.IsLocalizingSequence eta.process filtration mu T
(fun n omega =>
(canonicalRawLocalizingTime hUsual eta n omega : WithTop ℝ≥0)) := by
refine ⟨eta.progressive, ?_, ?_, ?_, ?_⟩
· intro n
exact canonicalRawLocalizingTime_isStoppingTime hUsual eta n
· intro n k hnk omega
simpa only [WithTop.coe_le_coe] using
((canonicalRawLocalizingTime_mono hUsual eta) hnk omega)
· intro n
exact rawStoppedProductEnergy_lt_top hUsual eta n
· exact canonicalRawLocalizingTime_tendsto_terminal_ae hUsual eta
end CanonicalRawLocalizationL2
end StochasticProcesses
end TechnicalLemmas
end AutoSamplingTheory