Maximum over disclosed outcomes
OpenQAlgorithms.Papers.Independent.Intakenext100.P77.classical_disclosure_boundLet and let be jointly independent random variables with finite alphabets on a normalized finite probability space. Let be any finite-valued random variable on the same space; it may depend arbitrarily on the family. Information is Shannon mutual information in bits. Write for preceding coordinates and for all other coordinates.
Maximizing over coordinates and positive-probability outcomes of the other coordinates gives
The finite maximum is nonempty under normalization. Null outcomes do not define conditional experiments. This permits a pointwise bound after all other key bits are disclosed.
import Definitions.Def_QAlgorithms_Papers_Independent_Intakenext100_P77
noncomputable section
namespace QAlgorithms.Papers.Independent.Intakenext100.P77
open scoped BigOperators ComplexConjugate ComplexOrder
open QAlgorithms
-- The repair instruction authorizes imports of pinned accepted definitions.
-- These are definitions only; no imported paper theorem is used here.
open QAlgorithms.Papers.Independent.Dependency50.P47 (Law)
open QAlgorithms.Papers.Independent.Information100.P04 (Measurement)
open QAlgorithms.Papers.Independent.Information100.P27 (shannon)
open QAlgorithms.Papers.Independent.Information100.P39 (pushLaw)
open QAlgorithms.Papers.Independent.Support50.P16 (trivialMeasurement)
attribute [local instance] Classical.propDecidable Classical.decEq
/-- Lemma H.2, PDF p.24: maximize additionally over positive-probability
outcomes of the other coordinates. Null outcomes are not experiments. -/
theorem classical_disclosure_bound {Omega E : Type}
[Fintype Omega] [Fintype E] {m : ℕ} (hm : 1 ≤ m)
{A : Fin m → Type} [∀ i, Fintype (A i)] (p : Law Omega)
(key : (i : Fin m) → Omega → A i) (eve : Omega → E)
(hind : JointIndependent p.prob key) :
classicalMutualInfo p.prob (fun w i => key i w) eve ≤
m * disclosureMaximum p.prob key eve := by sorry
end QAlgorithms.Papers.Independent.Intakenext100.P77
end
Read-back
What the Lean code literally says, in plain math · gpt-6.1-sol
Let be finite types, let , let each , , be finite, let be nonnegative with , and let and . Assume joint independence: for every tuple ,
For a finite-valued function , write ; define , where and . For and any value , define to equal on that fibre and zero elsewhere, with division by zero equal to zero. Then
The maximum excludes disclosure values of zero weight; its defining convention is zero for an empty collection, although the stated probability and ensure that this collection is nonempty.
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.