InformationTheory.Probability.TwoSidedExtension.Core
Shifted finite-dimensional marginals #
InformationTheory.Shannon.TwoSided.shiftAmount
sourceThe minimal non-negative integer N that makes every element of J non-negative
after shifting by N. For empty J we return 0.
Equations
Instances For
Used by
InformationTheory.Shannon.TwoSided.zero_le_shiftAmount_add
sourceFor any sufficient shift N, every j ∈ J satisfies 0 ≤ j + N.
Used by
InformationTheory.Shannon.TwoSided.obsZ
sourceThe joint observation at ℤ-indexed times, parametrized by a shift N:
obsZ μ p N J ω j := X (T^[(j + N).toNat] ω). When N ≥ shiftAmount J, this
records the genuine block (X_{j₀+N}, X_{j₁+N}, …) evaluated at the points of J.
Equations
- InformationTheory.Shannon.TwoSided.obsZ μ p N J ω j = p.obs (↑j + ↑N).toNat ω
Instances For
Used by
InformationTheory.Shannon.TwoSided.measurable_obsZ
sourceThe joint observation obsZ μ p N J is measurable.
Used by
InformationTheory.Shannon.TwoSided.shiftedMarginal
sourceThe finite-dimensional marginal of the (yet-to-be-built) two-sided extension at
the index set J : Finset ℤ. Defined as the law of obsZ at the canonical shift
shiftAmount J; the value is independent of the chosen shift by stationarity
(shiftedMarginal_eq_of_shift).
Equations
Instances For
Used by
InformationTheory.Shannon.TwoSided.obsZ_succ_shift
sourceFor two non-negative shifts N₁ ≤ N₂ both valid for J, the joint observation
under N₂ is the same as under N₁ precomposed with T^[N₂ - N₁].
Used by
InformationTheory.Shannon.TwoSided.map_obsZ_succ
sourceN-invariance of obsZ under push-forward (stationarity, single-shift step).
For any valid shift N, pushing forward by obsZ μ p (N + k) J agrees with pushing
forward by obsZ μ p N J.
Used by
InformationTheory.Shannon.TwoSided.map_obsZ_eq_of_shift
sourceN-invariance of the pushforward law (general form): any two valid shifts yield the same pushforward measure.
Used by
InformationTheory.Shannon.TwoSided.shiftedMarginal_eq_of_shift
sourceThe shifted marginal coincides with the pushforward of obsZ for any valid shift,
not just the canonical shiftAmount J.
Used by
InformationTheory.Shannon.TwoSided.instIsProbabilityMeasure_shiftedMarginal
sourceEach shifted marginal is a probability measure (pushforward of a probability measure under a measurable map).
Used by
InformationTheory.Shannon.TwoSided.isProjectiveMeasureFamily_shiftedMarginal
sourceProjective consistency: for J ⊆ I, restricting shiftedMarginal μ p I along
the inclusion gives shiftedMarginal μ p J.
Used by
Projective consistency + σ-additivity #
InformationTheory.Shannon.TwoSided.stationaryContent
sourceThe AddContent on measurable cylinders of ℤ → α induced by the shifted
finite-dimensional marginals.
Equations
Instances For
Used by
InformationTheory.Shannon.TwoSided.stationaryContent_tendsto_zero
sourceσ-additivity input: for any antitone sequence of measurable cylinders with empty intersection, the content tends to 0.
The proof uses the finite-type Cantor argument: equipping α with the discrete
topology turns ℤ → α into a compact Hausdorff space (Pi.compactSpace from finite
discrete factors), every cylinder cylinder I S is closed (preimage of a finite, hence
closed, set), and every closed subset of a compact space is compact. Cantor's
intersection theorem for sequences (IsCompact.nonempty_iInter_of_sequence_…) then
gives ∃ N, A N = ∅, after which stationaryContent (A n) = 0 for n ≥ N.
Used by
InformationTheory.Shannon.TwoSided.stationaryContent_isSigmaSubadditive
sourcestationaryContent is σ-subadditive (from tendsto_zero).
Used by
InformationTheory.Shannon.TwoSided.μZ
sourceThe two-sided extension μ_ℤ : Measure (ℤ → α) obtained from the stationary
process (Ω, T, μ, X) by Carathéodory extension of the stationaryContent.
Equations
Instances For
Used by
InformationTheory.Shannon.TwoSided.μZ_cylinder
sourceOn measurable cylinders, μZ agrees with shiftedMarginal.
Used by
InformationTheory.Shannon.TwoSided.isProjectiveLimit_μZ
sourceμZ is the projective limit of the shifted marginals.
Used by
InformationTheory.Shannon.TwoSided.instIsProbabilityMeasure_μZ
sourceμZ is a probability measure.
Used by
InformationTheory.Shannon.TwoSided.shiftZ
sourceThe two-sided shift σ : (ℤ → α) → (ℤ → α), σ x i := x (i + 1).
Equations
- InformationTheory.Shannon.TwoSided.shiftZ x i = x (i + 1)
Instances For
Used by
InformationTheory.Shannon.TwoSided.shiftZSymm
sourceThe inverse shift, σ⁻¹ x i := x (i - 1).
Equations
- InformationTheory.Shannon.TwoSided.shiftZSymm x i = x (i - 1)
Instances For
Used by
InformationTheory.Shannon.TwoSided.measurable_shiftZ
sourceThe shift is measurable.
Used by
InformationTheory.Shannon.TwoSided.measurable_shiftZSymm
sourceThe inverse shift is measurable.
Used by
InformationTheory.Shannon.TwoSided.measurePreserving_shiftZ
sourceThe two-sided shift preserves μZ.
Proof outline: both μZ μ p and (μZ μ p).map shiftZ are projective limits of
the family shiftedMarginal μ p of finite-dimensional marginals. By
IsProjectiveLimit.unique they are equal. The projective property of the
pushforward reduces to the shift-invariance of the marginals: for any
I : Finset ℤ, the marginal at I + 1 (relabeled through the bijection · + 1)
equals the marginal at I, which is exactly shiftedMarginal_eq_of_shift.
Used by
InformationTheory.Shannon.TwoSided.leftInverse_shiftZSymm
sourceshiftZSymm is the left-inverse of shiftZ.
Used by
InformationTheory.Shannon.TwoSided.rightInverse_shiftZSymm
sourceshiftZSymm is the right-inverse of shiftZ.
Used by
Coupling with the one-sided side #
InformationTheory.Shannon.TwoSided.forwardEmbed
sourceThe forward (one-sided) embedding Ω → (ℕ → α), ω ↦ (X (T^[i] ω))_i.
Equations
- InformationTheory.Shannon.TwoSided.forwardEmbed μ p ω i = p.obs i ω
Instances For
Used by
InformationTheory.Shannon.TwoSided.measurable_forwardEmbed
sourceThe forward embedding is measurable.
Used by
InformationTheory.Shannon.TwoSided.μZ_nat_proj_eq
sourceThe ℕ-projection of μZ agrees with the law of forwardEmbed.
Both sides are probability measures on ℕ → α. We verify equality on the
π-system of measurable cylinders using MeasureTheory.ext_of_generate_finite,
then compute each cylinder's measure: on the LHS via μZ_cylinder after
rewriting the preimage as a ℤ-cylinder, and on the RHS via the definition
of forwardEmbed together with shiftedMarginal_eq_of_shift at shift 0.
Used by
InformationTheory.Shannon.TwoSided.μZ_block_cylinder_eq
sourceBlock cylinder under μZ equals the block-law on the one-sided side.
Used by
ergodic_shiftZ #
Two-sided shift ergodicity via cylinder approximation + ℕ-factor transfer.
For a shiftZ-invariant measurable set A ⊆ (ℤ → α):
- Approximate
Aby a finite cylindert(over someF ⊆ ℤ) up to ε in symmetric difference — viaMeasureTheory.exists_measure_symmDiff_lt_of_generateFrom_isSetRingonmeasurableCylinders(a set ring generating the product σ-algebra). - For
klarge enough thatF + k ⊆ ℕ, the shifted cylindershiftZ^[k]⁻¹' tis a cylinder over a nonnegative index set, hence lies incylinderEvents {i : ℤ | 0 ≤ i}. - By measure-preservation of
shiftZ^[k]and shift-invariance ofA,μZ(A Δ shiftZ^[k]⁻¹' t) = μZ(A Δ t) < ε, soAis approximable to within any ε by sets incylinderEvents {i : ℤ | 0 ≤ i}. - Hence
AisμZ-a.e. equal to someA' = natProj⁻¹' Bfor measurableB, andBisshiftN-invariant mod-null onμ.map forwardEmbed. (ℕ → α, shiftN, μ.map forwardEmbed)is ergodic (forwardergodic_of_ergodic_semiconjfrom(Ω, T, μ)viaforwardEmbed).- Conclude
μZ(A) = μN(B) ∈ {0, 1}.
InformationTheory.Shannon.TwoSided.shiftN
sourceThe one-sided (ℕ-indexed) shift σ_ℕ : (ℕ → α) → (ℕ → α), σ_ℕ y i := y (i+1).
Equations
- InformationTheory.Shannon.TwoSided.shiftN y i = y (i + 1)
Instances For
Used by
InformationTheory.Shannon.TwoSided.measurable_shiftN
sourceThe forward shift on ℕ → α is measurable.
Used by
InformationTheory.Shannon.TwoSided.forwardEmbed_semiconj
sourceThe forward embedding semiconjugates T and shiftN.
Used by
InformationTheory.Shannon.TwoSided.natProj
sourceEquations
Instances For
Used by
InformationTheory.Shannon.TwoSided.measurable_natProj
sourcenatProj is measurable.
Used by
InformationTheory.Shannon.TwoSided.natProj_semiconj
sourceUsed by
InformationTheory.Shannon.TwoSided.measurePreserving_natProj
sourcenatProj is measure-preserving from μZ to μ.map forwardEmbed.
Used by
InformationTheory.Shannon.TwoSided.measurePreserving_forwardEmbed
sourceforwardEmbed is measure-preserving from μ to its pushforward.
Used by
InformationTheory.Shannon.TwoSided.ergodic_shiftN
sourceThe forward shift on (ℕ → α, μ.map forwardEmbed) is ergodic when the underlying
process is ergodic. Direct application of ergodic_of_ergodic_semiconj with
forwardEmbed as the measure-preserving semiconjugacy from (Ω, T, μ).
Used by
InformationTheory.Shannon.TwoSided.shiftZ_iterate_apply
sourceshiftZ iterated k times: shiftZ^[k] x i = x (i + k).
Used by
InformationTheory.Shannon.TwoSided.shiftZ_iterate_preimage_cylinder
sourcePreimage of a measurable cylinder over F under shiftZ^[k] is a measurable
cylinder over F + k = F.image (· + k).
Used by
InformationTheory.Shannon.TwoSided.shiftZ_iterate_preimage_mem_measurableCylinders
sourcePreimage under shiftZ^[k] preserves measurable cylinders: the preimage of
a measurableCylinders element is again in measurableCylinders.
Used by
InformationTheory.Shannon.TwoSided.shiftZ_iterate_preimage_cylinder_in_pos
sourceAfter shifting by k ≥ -F.min (when F is nonempty), the cylinder lies
in cylinderEvents {i : ℤ | 0 ≤ i} (the positive-index σ-algebra). The witness
index set is F.image (· + k), all of whose elements are ≥ 0.
Used by
InformationTheory.Shannon.TwoSided.posSigma
sourceThe positive-index σ-algebra on ℤ → α, i.e., the σ-algebra of events
depending only on coordinates i ≥ 0.
Equations
Instances For
Used by
InformationTheory.Shannon.TwoSided.posSigma_le_pi
sourceposSigma ≤ pi.
Used by
InformationTheory.Shannon.TwoSided.measurable_natProj_posSigma
sourceUsed by
InformationTheory.Shannon.TwoSided.exists_preimage_natProj_of_posSigma
sourceA posSigma-measurable set is the natProj-preimage of a measurable set
in ℕ → α. This is the factoring lemma: positive-index events factor through natProj.
Used by
InformationTheory.Shannon.TwoSided.ergodic_shiftZ
sourceThe two-sided shift is ergodic when the underlying process is ergodic.
By cylinder approximation + ℕ-factor transfer. For any
shiftZ-invariant measurable set A, approximate A to within ε by a
measurable cylinder, then shift to move the index set into the nonnegative
half-line. This yields a posSigma-measurable approximation, so A is μZ-a.e.
equal to a posSigma-measurable set A' = natProj⁻¹ B. By ergodicity of
shiftN on μ.map forwardEmbed, B is either null or co-null, hence so is A.