Finite information theory for route-sensitive context lifting
DefinitionTranscendenceTowerInformationTheoryThis bundle defines the finite probabilistic objects used to quantify uncertainty in the route-sensitive context-lifting example.
The logical alternatives are represented by two routed-resolution labels mapped to the blocking sets and . Probability is additional modeling data: the bundle defines explicit finite mass functions, Shannon entropy in nats, a uniform prior over direct and routed transfer, the resulting four-cell route/output joint mass, its marginals and conditional masses, and mutual information as a finite sum.
It also defines a separate abstract binary loss channel with survival parameter and fresh-stage composition through a newly sampled intermediate state.
These definitions do not assign probabilities to nonmonotonic logic in general. They support one finite experiment whose logical minimal blocks are classified by the accompanying theorem.
/-
Finite definitions for the platform-shaped transcendence-tower
information-theory result. The
solution proves the logical classification and every nontrivial probabilistic
claim from these explicit finite models.
-/
import Definitions.Def_TranscendenceTowerLiftConflicts
import Mathlib.Analysis.SpecialFunctions.BinaryEntropy
set_option autoImplicit false
open scoped BigOperators
open TranscendenceTower TranscendenceTower.ContextTheories
TranscendenceTower.Lifting
namespace TranscendenceTower.InformationTheoryPlatform
noncomputable section
/-- Labels for the two routed minimal-block resolutions. -/
inductive RoutedResolution where
| dropP
| dropQ
deriving DecidableEq
instance : Fintype RoutedResolution where
elems := {.dropP, .dropQ}
complete r := by cases r <;> simp
/-- Exchanging the two routed resolutions. -/
def swapResolution : RoutedResolution → RoutedResolution
| .dropP => .dropQ
| .dropQ => .dropP
/-- The blocking set selected by each routed-resolution label. -/
def routedBlock : RoutedResolution → Set L0
| .dropP => {pA}
| .dropQ => {qA}
/-- The sole direct-resolution label and its empty blocking set. -/
abbrev DirectResolution := Unit
def directBlock (_ : DirectResolution) : Set L0 := ∅
/-- A finite real-valued probability mass function. -/
def IsFiniteDistribution {α : Type} [Fintype α] (mass : α → ℝ) : Prop :=
(∀ x, 0 ≤ mass x) ∧ ∑ x, mass x = 1
/-- The Shannon summand, in nats. -/
def shannonTerm (mass : ℝ) : ℝ := mass * Real.log mass⁻¹
/-- Shannon entropy of a finite mass function, in nats. -/
def finiteEntropy {α : Type} [Fintype α] (mass : α → ℝ) : ℝ :=
∑ x, shannonTerm (mass x)
/-- The singleton mass on the direct-resolution label. -/
def directSingletonMass (_ : DirectResolution) : ℝ := 1
/-- The explicitly imposed fair mass on routed-resolution labels. -/
def uniformRoutedSelectionMass (_ : RoutedResolution) : ℝ := 1 / 2
/-- The direct and routed transfer modes in the finite experiment. -/
inductive Route where
| direct
| viaMid
deriving DecidableEq
instance : Fintype Route where
elems := {.direct, .viaMid}
complete r := by cases r <;> simp
/-- The uniform prior on direct and routed transfer. -/
def uniformRoutePrior : Route → ℝ
| .direct => 1 / 2
| .viaMid => 1 / 2
/-- `true` denotes survival of the distinguished fact. -/
def routeChannel : Route → Bool → ℝ
| .direct, true => 1
| .direct, false => 0
| .viaMid, true => 1 / 2
| .viaMid, false => 1 / 2
/-- The explicit four-cell route/output joint law. -/
def jointMass (route : Route) (output : Bool) : ℝ :=
uniformRoutePrior route * routeChannel route output
/-- Marginals obtained by summing the explicit joint law. -/
def routeMarginal (route : Route) : ℝ := ∑ output : Bool, jointMass route output
def outputMarginal (output : Bool) : ℝ := ∑ route : Route, jointMass route output
/-- Output mass conditional on a route, obtained from the joint law. -/
def routeConditionalMass (route : Route) (output : Bool) : ℝ :=
jointMass route output / routeMarginal route
/-- Conditional output entropy calculated from the joint law. -/
def uniformRouteConditionalEntropy : ℝ :=
∑ route : Route, routeMarginal route * finiteEntropy (routeConditionalMass route)
/-- Entropy of the output marginal derived from the joint law. -/
def uniformRouteOutputEntropy : ℝ := finiteEntropy outputMarginal
/-- Finite mutual information of a joint law and its marginals. -/
def finiteMutualInformation (joint : Route → Bool → ℝ)
(first : Route → ℝ) (second : Bool → ℝ) : ℝ :=
∑ route : Route, ∑ output : Bool,
if joint route output = 0 then 0 else
joint route output * Real.log (joint route output / (first route * second output))
/-- Mutual information of the explicit route/output joint law. -/
def uniformRouteMutualInformation : ℝ :=
finiteMutualInformation jointMass routeMarginal outputMarginal
/-- An abstract binary loss-only channel with survival parameter `s`. -/
def lossChannel (s : ℝ) (input output : Bool) : ℝ :=
if input then if output then s else 1 - s else if output then 0 else 1
/-- Sequential composition through a freshly sampled intermediate Boolean state. -/
def composeLossChannel (s t : ℝ) (input output : Bool) : ℝ :=
∑ middle : Bool, lossChannel s input middle * lossChannel t middle output
/-- Trivial finite-enumeration identities used by the solution. -/
theorem sum_resolutions {M : Type} [AddCommMonoid M] (f : RoutedResolution → M) :
(∑ r : RoutedResolution, f r) = f .dropP + f .dropQ := by
change (∑ r ∈ ({.dropP, .dropQ} : Finset RoutedResolution), f r) = _
simp
theorem sum_routes {M : Type} [AddCommMonoid M] (f : Route → M) :
(∑ r : Route, f r) = f .direct + f .viaMid := by
change (∑ r ∈ ({.direct, .viaMid} : Finset Route), f r) = _
simp
theorem sum_bools {M : Type} [AddCommMonoid M] (f : Bool → M) :
(∑ b : Bool, f b) = f false + f true := by
change (∑ b ∈ ({true, false} : Finset Bool), f b) = _
simp [add_comm]
theorem sum_unit {M : Type} [AddCommMonoid M] (f : Unit → M) :
(∑ x : Unit, f x) = f () := by
change (∑ x ∈ ({()} : Finset Unit), f x) = _
simp
end
end TranscendenceTower.InformationTheoryPlatform