Documentation

CslibFrontier.Foundations.Analysis.SpecialFunctions.ExpBounds

Polynomial bounds on the real exponential #

Two Taylor-type bounds on the real exponential for nonnegative arguments. They are the scalar inequalities behind multiplicative Chernoff bounds and the Johnson–Lindenstrauss lemma.

CslibFrontier.Foundations.Analysis.SpecialFunctions.ExpLogQuadratic has the weaker Real.exp_neg_le_one_sub_add_sq, exp (-y) ≤ 1 - y + y², valid for all -1 ≤ y.

Main results #

theorem Real.exp_neg_le_one_sub_add_sq_div_two {x : ℝ} (hx : 0 ≤ x) :
exp (-x) ≤ 1 - x + x ^ 2 / 2

exp (-x) ≤ 1 - x + x² / 2 for 0 ≤ x.

theorem Real.one_add_le_exp_sub_sq_div_two_add_cube_div_three {x : ℝ} (hx : 0 ≤ x) :
1 + x ≤ exp (x - x ^ 2 / 2 + x ^ 3 / 3)

1 + x ≤ exp (x - x² / 2 + x³ / 3) for 0 ≤ x; equivalently log (1 + x) ≤ x - x² / 2 + x³ / 3.

exp (-2) ≤ 1 / 3.

theorem Real.two_pow_le_exp (n : ℕ) :
2 ^ n ≤ exp ↑n

2 ^ n ≤ exp n, since 2 ≤ exp 1.

theorem Real.one_sub_pow_le_exp {p c : ℝ} {R : ℕ} (hp1 : p ≤ 1) (hc : c ≤ ↑R * p) :
(1 - p) ^ R ≤ exp (-c)

(1 - p) ^ R ≤ exp (-c) once c ≤ R p.

theorem Real.pow_le_one_third {p δ : ℝ} {k : ℕ} (hp : 0 ≤ p) (hpδ : p ≤ 1 - δ) (hk : 2 ≤ δ * ↑k) :
p ^ k ≤ 1 / 3

Amplification bound: if 0 ≤ p ≤ 1 - δ and 2 ≤ δ k, then p ^ k ≤ 1 / 3.