Convexity step, p. 767 — g′(x/n)x/n ≥ g(x/n) − g(0), hence e^{hx/n−g(x/n)} ≥ e^{−g(0)} for h = g′(x/n)
ProvedNagaevLD.GenMoment.convexity_stepLet have, on , a positive nondecreasing derivative . Let be an integer, , and . Then
and consequently
The first inequality expresses the convexity of (its graph lies above the tangent line at , evaluated at ). The second is the form used to prove (2.48), which in turn lets the negative part of be absorbed into the factor .
Formalization Note The paper prints the first inequality with a strict "". It fails when is constant on — for example , a case the paper itself discusses on p. 768 — so the non-strict inequality is stated; the proof of Theorem 2.5 only uses the second inequality, which the paper prints with "". The hypotheses on are imposed on only. The implicit of the paper is explicit in Lean.
import Mathlib import Definitions.Def_NagaevLD_GenMoment_Setting open MeasureTheory ProbabilityTheory
namespace NagaevLD.GenMoment
/-- Convexity step of the proof of Theorem 2.5, p. 767: `g′(x/n)·x/n ≥ g(x/n) − g(0)` and,
for `h = g′(x/n)`, `e^{hx/n − g(x/n)} ≥ e^{−g(0)}`. (The page prints a strict `>` in the first
inequality; it fails when `g′` is constant on `[0, x/n]`, so the non-strict form is stated.) -/
theorem convexity_step (g g' : ℝ → ℝ) (hg : ∀ u : ℝ, 0 ≤ u → HasDerivAt g (g' u) u)
(hg'pos : ∀ u : ℝ, 0 ≤ u → 0 < g' u) (hg'mono : MonotoneOn g' (Set.Ici 0))
(n : ℕ) (hn : 0 < n) (x : ℝ) (hx : 0 < x) (h : ℝ) (hh : h = g' (x / n)) :
g (x / n) - g 0 ≤ g' (x / n) * (x / n) ∧
Real.exp (-g 0) ≤ Real.exp (h * (x / n) - g (x / n)) := by sorry
end NagaevLD.GenMoment
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.