Proof of (4.1) — (b/(a+b))^{a+b} e^a ≤ exp[−a²/(2(a+b))] for a, b > 0
ProvedFreedmanTail.Bernstein.bennett_le_gaussbennett-inequalitybernstein-inequalityp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1real-analysis
For all real numbers and ,
This is the second inequality of Theorem (4.1), which the paper leaves to "calculus": it converts the Bennett form of the bound into the Bernstein (Gaussian-type) form .
Formalization Note is a real power of a number in .
Preamble
import Mathlib
Formal statement
namespace FreedmanTail.Bernstein
/-- Freedman (1975), proof of (4.1) Theorem, p. 108 ("the second follows by calculus"):
`(b/(a+b))^{a+b} e^a ≤ exp[−a²/(2(a+b))]` for positive `a, b` (real power). -/
theorem bennett_le_gauss (a b : ℝ) (ha : 0 < a) (hb : 0 < b) :
(b / (a + b)) ^ (a + b) * Real.exp a ≤ Real.exp (-(a ^ 2) / (2 * (a + b))) := by sorry
end FreedmanTail.Bernstein
Source
Freedman, On Tail Probabilities for Martingales, Ann. Probab. 3 (1975), p. 108 (PDF p. 9), proof of (4.1) Theorem: 'the second follows by calculus'
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.