InformationTheory.Shannon.BroadcastChannel.Marton.MutualCovering
Second-moment core of the mutual covering lemma #
Marton's mutual covering lemma asserts that among the M₁ * M₂ pairs formed from two
independently drawn codebooks, at least one pair is jointly typical. The standard proof
is a second moment argument, and its analytic core does not mention typicality at all:
it only needs a measurable set S in the product alphabet and the fact that indicator
variables attached to two pairs sharing no index are independent.
This file develops that core for an abstract S, in two strengths. The crude estimate uses
no information about S beyond p, and its conclusion needs both M₁ p and M₂ p to be
large. The sharpened estimate additionally assumes that every conditional slice of S has
mass at most qbar, and its conclusion splits into a term needing only the product M₁ M₂ p
to be large plus two terms carrying the ratio qbar / p.
Main definitions #
codebookFamily X Y— the two codebooks interleaved into one family indexed byFin M₁ ⊕ Fin M₂, so that mutual independence of allM₁ + M₂codewords is a singleiIndepFunhypothesis.pairIndicator X Y S p— indicator of "the codeword pair indexed byplands inS".pairCount X Y S— number of codeword pairs landing inS.pairProb μX μY S— probability that one independently drawn pair lands inS.sharedFstSet S/sharedSndSet S— the triples of codewords realizing two pairs that share one index.
Main statements #
integral_pairCount— the first momentE[A] = M₁ * M₂ * p.covariance_pairIndicator_eq_zero— covariance vanishes for pairs sharing no index.variance_pairCount_le— the resulting variance boundVar[A] ≤ M₁ M₂ (M₁ + M₂) p.meas_pairCount_eq_zero_le— Chebyshev's inequality applied to the above:P(A = 0) ≤ (M₁ + M₂) / (M₁ M₂ p).covariance_pairIndicator_shared_le— under a uniform slice boundqbar, the covariance of two pairs sharing one index is at mostqbar * p.variance_pairCount_le'andmeas_pairCount_eq_zero_le'— the sharpened variance boundVar[A] ≤ M₁ M₂ p + M₁ M₂ (M₁ + M₂) qbar pand the estimate it yields,P(A = 0) ≤ 1 / (M₁ M₂ p) + qbar / (M₁ p) + qbar / (M₂ p).
InformationTheory.Shannon.BroadcastChannel.Marton.codebookFamily
sourceThe two codebooks interleaved into a single family indexed by Fin M₁ ⊕ Fin M₂, each
entry padded by a constant in the unused coordinate so that the whole family shares the
codomain α × β. Padding by a constant leaves the generated σ-algebra unchanged, so
iIndepFun (codebookFamily X Y) μ says exactly that the M₁ + M₂ codewords are mutually
independent.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Used by
InformationTheory.Shannon.BroadcastChannel.Marton.codebookFamily_inl
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.Marton.codebookFamily_inr
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.Marton.pairIndicator
sourceIndicator of the event that the codeword pair indexed by p lands in S.
Equations
Instances For
Used by
InformationTheory.Shannon.BroadcastChannel.Marton.pairCount
sourceNumber of codeword pairs landing in S.
Equations
- InformationTheory.Shannon.BroadcastChannel.Marton.pairCount X Y S ω = ∑ p : Fin M₁ × Fin M₂, InformationTheory.Shannon.BroadcastChannel.Marton.pairIndicator X Y S p ω
Instances For
Used by
InformationTheory.Shannon.BroadcastChannel.Marton.pairProb
sourceProbability that one independently drawn codeword pair lands in S.
Equations
- InformationTheory.Shannon.BroadcastChannel.Marton.pairProb μX μY S = ((μX.prod μY) S).toReal
Instances For
Used by
InformationTheory.Shannon.BroadcastChannel.Marton.measurable_codebookFamily
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.Marton.measurable_pairIndicator
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.Marton.pairIndicator_nonneg
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.Marton.pairIndicator_le_one
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.Marton.memLp_pairIndicator
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.Marton.memLp_pairCount
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.Marton.indepFun_codewordPair
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.Marton.indepFun_pairIndicator
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.Marton.covariance_pairIndicator_eq_zero
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.Marton.indepFun_codeword
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.Marton.map_codewordPair
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.Marton.integral_pairIndicator
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.Marton.integral_pairCount
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.Marton.covariance_pairIndicator_le
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.Marton.variance_pairCount_le
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.Marton.meas_pairCount_eq_zero_le
sourceSecond-moment estimate behind Marton's mutual covering lemma: the probability that no
codeword pair lands in S is at most (M₁ + M₂) / (M₁ M₂ p), where p is the probability
that one independently drawn pair lands in S.
Used by
Sharpened variance bound under a uniform slice bound #
variance_pairCount_le bounds the covariance of two pairs sharing one index by the crude
pairProb μX μY S. If instead every conditional slice of S has mass at most qbar, that
covariance is bounded by qbar * pairProb μX μY S, and the resulting Chebyshev estimate
splits into a diagonal term 1 / (M₁ M₂ p) plus shared-index terms carrying a factor qbar.
Two slice bounds are needed, one in each coordinate: the α-slices control the pairs sharing
their second index and the β-slices control the pairs sharing their first index. Neither
bound implies the other.
InformationTheory.Shannon.BroadcastChannel.Marton.map_codewordTripleFst
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.Marton.map_codewordTripleSnd
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.Marton.variance_pairCount_le'
sourceSharpened variance bound: the diagonal terms contribute M₁ M₂ p, and the pairs sharing one
index contribute at most qbar * p each.
@audit:ok
Used by
InformationTheory.Shannon.BroadcastChannel.Marton.meas_pairCount_eq_zero_le'
sourceSharpened second-moment estimate behind Marton's mutual covering lemma. When every
conditional slice of S has mass at most qbar, the probability that no codeword pair lands
in S splits into a term governed by the product M₁ M₂ and two terms carrying qbar.
@audit:ok
Used by
A canonical ambient realizing the hypotheses #
meas_pairCount_eq_zero_le is stated over an abstract probability space. This section
exhibits the i.i.d. codebook ambient on which its hypotheses hold, so that the abstract
statement is known to be non-vacuous and downstream users have a template to instantiate.
InformationTheory.Shannon.BroadcastChannel.Marton.ambientFactor
sourceLaw of one entry of the interleaved codebook family on the canonical ambient: an
α-codeword carries μX in its first coordinate and a point mass in the padded one.
Equations
- InformationTheory.Shannon.BroadcastChannel.Marton.ambientFactor μX μY M₁ M₂ (Sum.inl val) = μX.prod (MeasureTheory.Measure.dirac (Classical.arbitrary β))
- InformationTheory.Shannon.BroadcastChannel.Marton.ambientFactor μX μY M₁ M₂ (Sum.inr val) = (MeasureTheory.Measure.dirac (Classical.arbitrary α)).prod μY
Instances For
Used by
InformationTheory.Shannon.BroadcastChannel.Marton.instIsProbabilityMeasureAmbientFactor
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.Marton.ambient
sourceCanonical ambient measure: the M₁ + M₂ codewords are drawn independently.
Equations
Instances For
Used by
InformationTheory.Shannon.BroadcastChannel.Marton.instIsProbabilityMeasureAmbient
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.Marton.ambientX
sourceFirst codebook read off the canonical ambient.
Equations
- InformationTheory.Shannon.BroadcastChannel.Marton.ambientX M₁ M₂ i ω = (ω (Sum.inl i)).1
Instances For
Used by
InformationTheory.Shannon.BroadcastChannel.Marton.ambientY
sourceSecond codebook read off the canonical ambient.
Equations
- InformationTheory.Shannon.BroadcastChannel.Marton.ambientY M₁ M₂ j ω = (ω (Sum.inr j)).2
Instances For
Used by
InformationTheory.Shannon.BroadcastChannel.Marton.ambientPad
sourceThe padding map attached to one index of the interleaved family.
Equations
- InformationTheory.Shannon.BroadcastChannel.Marton.ambientPad M₁ M₂ (Sum.inl val) = fun (r : α × β) => (r.1, Classical.arbitrary β)
- InformationTheory.Shannon.BroadcastChannel.Marton.ambientPad M₁ M₂ (Sum.inr val) = fun (r : α × β) => (Classical.arbitrary α, r.2)
Instances For
Used by
InformationTheory.Shannon.BroadcastChannel.Marton.measurable_ambientX
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.Marton.measurable_ambientY
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.Marton.measurable_ambientPad
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.Marton.iIndepFun_codebookFamily_ambient
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.Marton.map_ambientX
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.Marton.map_ambientY
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.Marton.meas_pairCount_ambient_eq_zero_le
sourceThe mutual covering estimate on the canonical i.i.d. codebook ambient.