InformationTheory.Shannon.Pinsker.Sharp
Sharp Pinsker inequality (constant 1/√2) #
For probability measures P, Q on a finite alphabet α with P ≪ Q, the sharp Cover–Thomas
form of Pinsker's inequality: tvNorm P Q ≤ √((klDiv P Q).toReal / 2). The total-variation
norm tvNorm is shared with the weak form in Pinsker/Basic.lean.
Main statements #
klFun_sharp_lower— the pointwise sharp bound3 * (t - 1)^2 ≤ 2 * (t + 2) * klFun t.tvNorm_le_sqrt_klDiv_div_two— the sharp Pinsker inequalitytvNorm P Q ≤ √((klDiv P Q).toReal / 2).
Implementation notes #
The pointwise bound is obtained from the auxiliary H t := 2 (t + 2) · klFun t - 3 (t - 1)^2,
whose second derivative 4 (log t + 1/t - 1) is nonnegative on (0, ∞) and which has a minimum
of 0 at t = 1. The global bound follows by a per-element application together with the
Cauchy–Schwarz step and ∑ (p + 2q) = 3.
References #
- T. M. Cover and J. A. Thomas, Elements of Information Theory (2nd ed.), Wiley, 2006.
Pointwise sharp Pinsker inequality #
InformationTheory.Shannon.PinskerSharp.klFun_sharp_lower
sourceThe pointwise sharp Pinsker inequality 3 · (t - 1)^2 ≤ 2 · (t + 2) · klFun t for t ≥ 0.
Used by
Sharp Pinsker inequality #
InformationTheory.Shannon.PinskerSharp.tvNorm_le_sqrt_klDiv_div_two
sourcePinsker's inequality (sharp form, constant 1/√2): for probability measures P ≪ Q on a
finite alphabet, tvNorm P Q ≤ √((klDiv P Q).toReal / 2).