openAI-math-dataset / lean /OAI /NumberTheory /DukePrimeDegree /Entropy /CountableFutureFiberEntropy.lean
roseking's picture
Add lean source and binary files (part 103)
a44267b verified
Raw History Blame Contribute Delete
4.28 kB
import OAI.NumberTheory.DukePrimeDegree.Entropy.CountableFutureSigma
import OAI.NumberTheory.DukePrimeDegree.Geometry.CountableRootPrefixAtoms
import OAI.NumberTheory.DukePrimeDegree.Geometry.RootPrefixProductKernel
namespace OAI
/-! # Actual leaf entropy conditioned on a countable future
Finite time-and-label approximations identify the conditional entropy of
any finite present label with the entropy of the actual leaf kernel
restricted to its positive-mass countable future atom.
-/
noncomputable section
open MeasureTheory Set
namespace DukePrimeDegree
open PrimeDegreePackets
theorem conditionalEntropy_countableFutureFiber
{S V Y : Type*} [MeasurableSpace S] [MeasurableSpace V]
[StandardBorelSpace V] [Nonempty V] [MeasurableSpace Y]
(ρ : Measure (S × V)) [IsFiniteMeasure ρ]
{M : ℕ} {P : Y → Fin M} (hP : Measurable P)
{Q : Y → ℕ} (hQ : Measurable Q) {T : Y → Y} (hT : Measurable T)
{F : S × V → Y} (hF : Measurable F)
(hpos : ∀ᵐ z ∂ρ, ρ.condKernel z.1 (countableFutureFiber T Q F z) ≠ 0) :
conditionalSigmaEntropy ρ (P ∘ F)
(MeasurableSpace.comap Prod.fst (inferInstance : MeasurableSpace S) ⊔
MeasurableSpace.comap F (⨆ k : ℕ, observationSigma (fun y => Q (T^[k+1] y)))) =
∫ z, observationEntropy
((anchorNormalize (ρ.condKernel z.1) (countableFutureFiber T Q F z)).restrict
(countableFutureFiber T Q F z)) (fun t => P (F (z.1,t))) ∂ρ := by
rw [←transverseCountableFutureSigma_eq_sup T Q F]
simpa only [countableFutureFiber_eq_iInter,Function.comp_def] using
conditionalEntropy_infinite_label_restriction ρ (hP.comp hF)
(fun m z => countableFutureCode T Q m (F z))
(fun m => (countableFutureCode_measurable hT hQ m).comp hF)
(transverse_countableFutureCode_monotone T Q F)
(countableFutureCode_fibers_antitone T Q F)
(by simpa only [countableFutureFiber_eq_iInter] using hpos)
namespace RootPrefixBox
open scoped Topology
variable {d : ℕ} {P₀ : Finset (Fin (d+1) × Fin (d+1))} {D : Finset (Fin (d+1))}
{hP₀ : InitialRootPattern P₀} {c : EliminationChain P₀ D}
{hf : c.finalEntries = allPositiveRoots (d+1)}
{V : Type*} [TopologicalSpace V] [SecondCountableTopology V]
[MeasurableSpace V] [BorelSpace V]
{C : RootPrefixCoordinates (d+1) P₀ V} {g : G (d+1)}
theorem conditionalEntropy_countableFuture_intrinsicProduct
(B : RootPrefixBox C c hP₀ hf g) (R : ℝ) (hBR : B.time ⊆ c.remainingWindow R 0)
(μ : Measure (X (d+1))) [IsProbabilityMeasure μ]
[ErgodicSMul (Multiplicative (LogCoordinates d)) (X (d+1)) μ]
{M : ℕ} {P : X (d+1) → Fin M} (hP : Measurable P)
{Q : X (d+1) → ℕ} (hQ : Measurable Q)
{T : X (d+1) → X (d+1)} (hT : Measurable T)
(hgood : ∀ᵐ x ∂μ, ∀ᶠ v in 𝓝 (completeRootChain (d+1)).zeroTime, ∀ k : ℕ,
Q (T^[k] (rightAction x ((completeRootChain (d+1)).word v)))=Q (T^[k] x)) :
let F := C.timeParam c hP₀ hf g
conditionalSigmaEntropy (B.coordinateMeasure μ) (P ∘ F)
(MeasurableSpace.comap Prod.fst (inferInstance : MeasurableSpace (UpperRootTransverse (d+1) × V)) ⊔
MeasurableSpace.comap F (⨆ k : ℕ, observationSigma (fun y => Q (T^[k+1] y)))) =
∫ z, observationEntropy
((anchorNormalize (B.intrinsicProductConditional μ z) (countableFutureFiber T Q F z)).restrict
(countableFutureFiber T Q F z)) (fun t => P (F (z.1,t))) ∂B.coordinateMeasure μ := by
let F := C.timeParam c hP₀ hf g
have hpos : ∀ᵐ z ∂B.coordinateMeasure μ,
(B.coordinateMeasure μ).condKernel z.1 (countableFutureFiber T Q F z) ≠ 0 := by
filter_upwards [B.ae_wholeFutureAtom_pos μ T Q hgood] with z hz
apply ne_of_gt
apply hz.trans_le
apply measure_mono
intro t ht k
exact ht (k+1)
have he := conditionalEntropy_countableFutureFiber (B.coordinateMeasure μ) hP hQ hT
(C.timeParam_localHomeomorph c hP₀ hf g).continuous.measurable hpos
refine he.trans (integral_congr_ae ?_)
filter_upwards [B.ae_conditionalKernel_eq_intrinsicProduct R hBR μ] with z hz
change (B.coordinateMeasure μ).condKernel z.1 = B.intrinsicProductConditional μ z at hz
rw [hz]
end RootPrefixBox
end DukePrimeDegree
end
end OAI