InformationTheory.Shannon.BroadcastChannel.OuterBoundUV.Region
Broadcast channel — the UV outer region as a subset of the plane #
The four information slots of the UV outer bound are functionals of a five-tuple law
(U, V, X, Y₁, Y₂) and do not mention the channel. A region defined as the union of the
resulting quadrilaterals over all five-tuple laws would therefore be the whole plane: a law
that copies the input into the outputs makes every slot as large as the input alphabet allows,
whatever the channel is. The union is therefore indexed by the laws whose output pair is
generated from the input letter by the channel and by nothing else, which is one composition
product identity, IsUVChannelLaw.
Main definitions #
IsUVChannelLaw W ν— the conditional law of the output pair(Y₁, Y₂)given the two auxiliaries and the input letter(U, V, X)isW X. This says at once that the output law is the channel and that the auxiliaries reach the outputs only through the input letter.uvLawOfInput W ρ— the channel law generated by a law of the auxiliaries and the input letter.uvRegion ν— the quadrilateral cut out by the four information slots of a five-tuple law.bcOuterRegionUV W— the UV outer region: the closure of the union ofuvRegion νover the channel lawsνon a fixed pair of countable auxiliary alphabets.uvConstLaw W x₀— the channel law with constant auxiliaries and constant input letterx₀, which witnesses that the union is indexed by a nonempty family.uvOutputCopiesInputLaw,uvAuxCopiesOutputLaw— two five-tuple laws that the channel constraint rejects, over the channelsuvBlindChannelanduvFairBitChannel.
Main statements #
bcOuterRegionUV_isClosed— the region is closed.bcOuterRegionUV_isLowerSet— the region is a lower set, so a rate pair below one of its points belongs to it as well.bcOuterRegionUV_nonempty— the region is nonempty, witnessed byuvConstLaw, so the union is indexed by a nonempty family of channel laws.isUVChannelLaw_iff— a law is a channel law exactly when it is a law of the auxiliaries and the input letter pushed through the channel, which describes the index of the union directly.IsUVChannelLaw.map_input_output— a channel law has the channel joint(ν.map X) ⊗ₘ Was its input-output pair law, which is the constraint a law copying the input letter into the outputs violates.IsUVChannelLaw.map_auxiliary_input_output— dropping the second auxiliary leaves the law of the first auxiliary, the input letter and the two outputs in the shape a superposition ensemble has: the law of the pair(U, X)pushed through the channel.IsUVChannelLaw.isMarkovChain_UV_X_Y— the two auxiliaries reach the output pair through the input letter only, withIsUVChannelLaw.isMarkovChain_U_X_Y₁andIsUVChannelLaw.isMarkovChain_V_X_Y₁reading one auxiliary and one output off it.not_isUVChannelLaw_uvOutputCopiesInputLawandnot_isUVChannelLaw_uvAuxCopiesOutputLaw— the constraint rejects two structurally different degenerate laws, one whose outputs copy the input letter and one whose auxiliary copies an output over a one-letter input alphabet.IsUVChannelLaw.smul,IsUVChannelLaw.add— mixtures of channel laws are channel laws, so averaging the letter laws of a code stays inside the index of the union.IsUVChannelLaw.map_auxiliaries— re-encoding the two auxiliary alphabets keeps a channel law a channel law, which is how a law on the auxiliaries of a code reaches the fixed ones.IsUVChannelLaw.swap_auxiliaries— exchanging the two auxiliaries keeps a channel law a channel law, so a law indexing them in the opposite order indexes the union too.
Implementation notes #
IsUVChannelLaw is one composition-product identity between two pushforwards of ν, rather
than a conjunction of "the output pair is distributed by the channel" and "the auxiliaries are
conditionally independent of the output pair given the input letter". A single identity is the
shape the Measure.map and Measure.compProd lemmas consume, so the mixture, re-encoding and
marginalization lemmas are each a rewrite chain; the first conjunct is recovered from it as
IsUVChannelLaw.map_input_output, and the second is the statement that the conditional law is
read at the input coordinate only.
uvLawOfInput is the construction appearing on the right of isUVChannelLaw_iff: together with
uvLawOfInput_isUVChannelLaw it says that the channel laws are exactly the laws obtained by
pushing a law of (U, V, X) through the channel, which is how a witness such as uvConstLaw is
built and shown to index the union.
Both auxiliary alphabets of bcOuterRegionUV are fixed to ℕ instead of being quantified over
countable types, so the union ranges over measures rather than over types. The closure is taken
because a union of intersections of closed half-planes need not be closed, and because the
operational region is itself a closure.
uvRegion imposes no sign constraint on the rate pair, matching the operational region, which
contains nonpositive pairs; imposing one would exclude pairs the operational region contains.
Both regions being lower sets is what carries the inclusion off the first quadrant, so no
intersection with it is needed.
Channel laws of a five-tuple #
The constraint and its characterization #
InformationTheory.Shannon.BroadcastChannel.IsUVChannelLaw
sourceA five-tuple law (U, V, X, Y₁, Y₂) is a channel law for W when the conditional law of the
output pair given the two auxiliaries and the input letter is W X: pushing the law forward to
the pair ((U, V, X), (Y₁, Y₂)) gives the composition product of the (U, V, X) marginal with
W read at the input coordinate.
The identity carries both constraints that keep the region proper. Taking the (U, V) component
of the first factor away leaves the input-output pair law (ν.map X) ⊗ₘ W, so the outputs are
distributed by the channel; keeping it says that the conditional law does not depend on the
auxiliaries, so they act on the outputs only through the input letter.
The identity pins the law exactly: it holds if and only if ν is the composition product of its
own (U, V, X) marginal with the channel read at the input letter, so the union is indexed by
the laws obtained from an arbitrary law of (U, V, X) through the channel and by nothing else.
@audit:ok
Equations
- One or more equations did not get rendered due to their size.
Instances For
Used by
InformationTheory.Shannon.BroadcastChannel.isUVChannelLaw_iff
sourceUsed by
Mixtures, re-encodings and marginals #
InformationTheory.Shannon.BroadcastChannel.IsUVChannelLaw.smul
source@audit:ok
Used by
InformationTheory.Shannon.BroadcastChannel.IsUVChannelLaw.add
source@audit:ok
Used by
InformationTheory.Shannon.BroadcastChannel.IsUVChannelLaw.finsetSum
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.IsUVChannelLaw.map_auxiliaries
source@audit:ok
Used by
InformationTheory.Shannon.BroadcastChannel.IsUVChannelLaw.swap_auxiliaries
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.IsUVChannelLaw.map_input_output
source@audit:ok
Used by
InformationTheory.Shannon.BroadcastChannel.IsUVChannelLaw.map_auxiliary_input_output
sourceUsed by
The Markov chains carried by a channel law #
InformationTheory.Shannon.BroadcastChannel.IsUVChannelLaw.isMarkovChain_UV_X_Y
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.IsUVChannelLaw.isMarkovChain_U_X_Y₁
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.IsUVChannelLaw.isMarkovChain_V_X_Y₁
sourceUsed by
The law generated by a law of the auxiliaries and the input #
InformationTheory.Shannon.BroadcastChannel.uvLawOfInput
sourceThe channel law generated by an arbitrary law of the auxiliaries and the input letter.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Used by
InformationTheory.Shannon.BroadcastChannel.uvLawOfInput_isProbabilityMeasure
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.uvLawOfInput_isUVChannelLaw
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.uvLawOfInput_map_aux_input
sourceUsed by
The UV outer region #
InformationTheory.Shannon.BroadcastChannel.uvRegion
sourceThe quadrilateral of a five-tuple law: the rate pairs satisfying the two corner bounds and
the two sum-rate bounds of InBCOuterRegionUV at the four information slots of the law. No sign
constraint is imposed, matching the operational region, which contains nonpositive rate pairs.
@audit:ok
Equations
- One or more equations did not get rendered due to their size.
Instances For
Used by
InformationTheory.Shannon.BroadcastChannel.bcOuterRegionUV
sourceThe UV (Nair–El Gamal) outer region of a broadcast channel: the closure of the union of the
quadrilaterals uvRegion ν over the channel laws ν of W.
Both auxiliary alphabets are fixed to ℕ, which quantifies over every countable auxiliary
without quantifying over types. The closure is taken because a union of closed half-plane
intersections need not be closed, and because the operational region is itself a closure.
@audit:ok
Equations
- One or more equations did not get rendered due to their size.
Instances For
Used by
InformationTheory.Shannon.BroadcastChannel.bcOuterRegionUV_isClosed
source@audit:ok
Used by
InformationTheory.Shannon.BroadcastChannel.uvRegion_isLowerSet
source@audit:ok
Used by
InformationTheory.Shannon.BroadcastChannel.bcOuterRegionUV_isLowerSet
sourceThe UV outer region is a lower set: a rate pair below a point of the region is again in the region. Each quadrilateral bounds the two rates and their sum from above, and both the union and the closure preserve that. @audit:ok
Used by
InformationTheory.Shannon.BroadcastChannel.uvConstLaw
sourceThe five-tuple law with constant auxiliaries and a constant input letter x₀, whose output
pair is drawn from W x₀.
@audit:ok
Equations
Instances For
Used by
InformationTheory.Shannon.BroadcastChannel.uvConstLaw_isProbabilityMeasure
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.uvConstLaw_isUVChannelLaw
source@audit:ok
Used by
InformationTheory.Shannon.BroadcastChannel.bcOuterRegionUV_nonempty
source@audit:ok
Used by
The channel constraint is not vacuous #
The information slots do not mention the channel, so the constraint is what stops the union from exhausting the plane, and it has to reject two structurally different families of laws.
A law that copies the input letter into both outputs carries a full input alphabet of information
in every slot no matter which channel indexes the region, and such laws exist over every alphabet,
so the output law has to be pinned to the channel. uvOutputCopiesInputLaw copies a fair bit
into both outputs and is rejected over the channel that always outputs (false, false).
Pinning the output law alone would still leave the auxiliaries free to read the outputs directly,
which manufactures information about a receiver that the input letter does not carry.
uvAuxCopiesOutputLaw lives over a one-letter input alphabet and has the channel joint as its
input-output pair law, so it meets the first constraint, yet its first auxiliary is the output
bit, and it is rejected as well.
InformationTheory.Shannon.BroadcastChannel.uvBlindChannel
sourceThe broadcast channel over a binary input alphabet whose two receivers always read false,
so that no input letter is visible at either output.
Equations
Instances For
Used by
InformationTheory.Shannon.BroadcastChannel.uvBlindChannel_isMarkovKernel
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.uvOutputCopiesInputLaw
sourceThe five-tuple law that draws a fair bit and copies it into the input letter, into both outputs and into both auxiliaries.
Equations
- InformationTheory.Shannon.BroadcastChannel.uvOutputCopiesInputLaw = MeasureTheory.Measure.map (fun (b : Bool) => (b, b, b, b, b)) ((↑(Fintype.card Bool))⁻¹ • MeasureTheory.Measure.count)
Instances For
Used by
InformationTheory.Shannon.BroadcastChannel.uvOutputCopiesInputLaw_isProbabilityMeasure
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.not_isUVChannelLaw_uvOutputCopiesInputLaw
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.uvFairBitPair
sourceThe law of two copies of a fair bit.
Equations
Instances For
Used by
InformationTheory.Shannon.BroadcastChannel.uvFairBitPair_isProbabilityMeasure
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.uvFairBitChannel
sourceThe broadcast channel over a one-letter input alphabet that sends the same fair bit to both receivers, so that its input carries no information at all.
Equations
Instances For
Used by
InformationTheory.Shannon.BroadcastChannel.uvFairBitChannel_isMarkovKernel
sourceUsed by
InformationTheory.Shannon.BroadcastChannel.uvAuxCopiesOutputLaw
sourceThe five-tuple law over uvFairBitChannel whose first auxiliary is the common output bit and
whose second auxiliary is constant.