Scalar testing bounds from entropy and Rényi inequalities
DefinitionCRCD_ChannelContinuity_Testingchannel-renyi-continuity-jwang226quantum-information
Let be real numbers, where is a testing payoff and . If , , and , then
The same conclusion holds when the entropy inequality is assumed only in the case , provided additionally that . If , , and
then . A companion formulation assumes this logarithmic inequality only when ; the conclusion then holds without a separate positivity premise on . All logarithms in these testing bounds have base two.
Definition code
import Mathlib.Analysis.SpecialFunctions.Log.Base
import Mathlib.Order.ConditionallyCompleteLattice.Basic
import Mathlib.Tactic.FieldSimp
import Mathlib.Tactic.Linarith
import Mathlib.Tactic.NormNum
import Mathlib.Tactic.Ring
/-
Copyright (c) 2026 Jinzhao Wang. All rights reserved.
Released under Apache 2.0 license as described in the file LICENSE.
Authors: Jinzhao Wang (AI-assisted formalization)
-/
/-!
# Scalar reductions of the binary testing inequalities
These lemmas prove the algebra, base-two exponentiation, and passage to a
supremum in the manuscript's testing lemma. The quantum binary data-processing
bounds are explicit hypotheses; this file does not identify abstract scalars
with quantum relative entropies.
-/
namespace ChannelContinuity
/-- The relative-entropy binary testing inequality bounds the testing payoff.
The real variable `n` allows direct specialization to positive natural block
sizes without making the elementary calculation depend on coercions. -/
theorem weak_testing_bound {n r d p objective : ℝ}
(hn : 0 < n) (hr : 0 < r)
(hdata : (n * r) * p - 1 ≤ n * d) (hobjective : objective ≤ p) :
objective ≤ d / r + 1 / (n * r) := by
have hp : p ≤ (n * d + 1) / (n * r) :=
(le_div_iff₀ (mul_pos hn hr)).2 (by nlinarith)
calc
objective ≤ p := hobjective
_ ≤ (n * d + 1) / (n * r) := hp
_ = d / r + 1 / (n * r) := by
field_simp [ne_of_gt hn, ne_of_gt hr]
/-- Base-two exponentiation turns the Rényi binary data-processing inequality
into the high-rate exponential estimate. Positivity of `n` or of the rate gap
is not needed for this algebraic implication. -/
theorem high_testing_bound {n alpha a ell p objective : ℝ}
(halpha : 1 < alpha) (hp : 0 < p)
(hdata : n * ell + (1 / (alpha - 1)) * Real.logb 2 p ≤ n * a)
(hobjective : objective ≤ p) :
objective ≤ (2 : ℝ) ^ (-n * (alpha - 1) * (ell - a)) := by
have ha : 0 < alpha - 1 := sub_pos.mpr halpha
have hlogdiv : Real.logb 2 p / (alpha - 1) ≤ n * a - n * ell := by
rw [div_eq_mul_inv]
simpa only [one_div, mul_comm] using
(show (1 / (alpha - 1)) * Real.logb 2 p ≤ n * a - n * ell by linarith)
have hlog := (div_le_iff₀ ha).1 hlogdiv
have hpbound : p ≤ (2 : ℝ) ^ (-n * (alpha - 1) * (ell - a)) := by
apply (Real.logb_le_iff_le_rpow (by norm_num : (1 : ℝ) < 2) hp).1
nlinarith
exact hobjective.trans hpbound
/-- In the hockey-stick argument the binary inequality is only needed for
strictly positive payoff. Nonpositive payoffs already satisfy the bound. -/
theorem high_testing_bound_of_positive_payoff {n alpha a ell p objective : ℝ}
(halpha : 1 < alpha) (hobjective : objective ≤ p)
(hdata : 0 < objective →
n * ell + (1 / (alpha - 1)) * Real.logb 2 p ≤ n * a) :
objective ≤ (2 : ℝ) ^ (-n * (alpha - 1) * (ell - a)) := by
by_cases hpos : 0 < objective
· exact high_testing_bound halpha (hpos.trans_le hobjective) (hdata hpos) hobjective
· exact (le_of_not_gt hpos).trans (Real.rpow_nonneg (by norm_num) _)
/-- The weak estimate likewise only needs binary data processing on testers
whose hockey-stick payoff is positive. -/
theorem weak_testing_bound_of_positive_payoff {n r d p objective : ℝ}
(hn : 0 < n) (hr : 0 < r) (hd : 0 ≤ d) (hobjective : objective ≤ p)
(hdata : 0 < objective → (n * r) * p - 1 ≤ n * d) :
objective ≤ d / r + 1 / (n * r) := by
by_cases hpos : 0 < objective
· exact weak_testing_bound hn hr (hdata hpos) hobjective
· apply (le_of_not_gt hpos).trans
exact add_nonneg (div_nonneg hd hr.le)
(div_nonneg zero_le_one (mul_pos hn hr).le)
end ChannelContinuity
Source