Lemma 10.8.1 —
ProvedWildeQIT.PinskerRefined.lemma_10_8_1entropy-inequalitiesinformation-theoryp2o-gran-per-chapterp2o-plan-bookp2o-v1pinsker
Let . Then
The left side is the binary relative entropy of with respect to in nats, with the conventions and for ; for and it is and the inequality holds trivially, as the book notes. This is the two-point case of Pinsker's inequality, to which the general case is reduced by a coarse-graining channel.
Formalization Note The left side is written as with the published relEntropy (bits, extended reals, off support) on the two-point alphabet Fin 2, where is an arbitrary probability distribution on Fin 2 (so ranges over ) and .
Differences from the source:
- encoding: a ln(a/b) + (1 − a) ln((1 − a)/(1 − b)) for a, b ∈ [0, 1], +∞ when b ∈ {0, 1} and a ≠ b → ln 2 · relEntropy p ![b, 1 − b] in EReal, with p any FinDist on Fin 2 and a = p(0); reason: relEntropy carries the book’s support convention (Definition 10.5.2) so the endpoints b ∈ {0, 1} keep their extended values; ln 2 converts bits to nats; every a ∈ [0, 1] is p(0) for p = (a, 1 − a)
Preamble
import Definitions.Def_WildeQIT_relEntropy
Formal statement
namespace WildeQIT.PinskerRefined
open WildeQIT
/-- **Lemma 10.8.1**, Wilde, *From Classical to Quantum Shannon Theory*, arXiv:1106.1445v8, p. 311: for `a, b ∈ [0, 1]`,
`a ln(a/b) + (1 − a) ln((1 − a)/(1 − b)) ≥ 2 (a − b)²`.
**Formalization Note** The left side is the binary relative entropy in nats, with the book's
conventions at `b ∈ {0, 1}` (`0 ln(0/·) = 0`, and `+∞` when `a ≠ b = 0` or `a ≠ b = 1`, the case the
book calls trivial). It is stated as `ln 2 · D(p‖q)` with the published `relEntropy` (bits, `EReal`,
`+∞` off support, Definition 10.5.2) on the two-point alphabet `Fin 2`, where `p = (a, 1 − a)` is any
probability distribution on `Fin 2` (`a = p(0)`, so `a` ranges over all of `[0, 1]`) and
`q = (b, 1 − b)`. In `EReal`, `ln 2 · (+∞) = +∞`. -/
theorem lemma_10_8_1 (p : FinDist (Fin 2)) (b : ℝ) (hb0 : 0 ≤ b) (hb1 : b ≤ 1) :
((2 * (p.prob 0 - b) ^ 2 : ℝ) : EReal) ≤
((Real.log 2 : ℝ) : EReal) * relEntropy p ![b, 1 - b] := by sorry
end WildeQIT.PinskerRefined
Source
Wilde, From Classical to Quantum Shannon Theory, arXiv:1106.1445v8, Lemma 10.8.1, p. 311
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.