The derivatives of an exponential polynomial at a point are bounded by its first N, geometrically in the order
ProvedTranscendence.expPoly_iteratedDeriv_leLet , over a finite index set, with and each zero or of degree less than . Put . If for every at a point , with , then for every
The frequencies need not be distinct.
Proof idea. Induction on . If , every is zero. Otherwise pick with and peel that frequency: has the same shape, with lowered by one. Its first derivatives at are at most , so by induction . Then gives the bound for by induction on , because .
What it is for. With the Taylor series at it gives FourExp.expPoly_value_le_derivs in a few lines, in place of the binomial bookkeeping of that node's first formal proof.
Novelty. None is claimed. The function satisfies the linear differential equation of order whose characteristic polynomial is , and the bound is the elementary induction along it.
import Mathlib
open Finset Polynomial
namespace Transcendence
/-- Derivatives of an exponential polynomial `f(z) = ∑ⱼ Pⱼ(z) e^{wⱼ z}`, with `|wⱼ| ≤ W` and `deg Pⱼ < qⱼ`,
`∑ qⱼ = N`: if the first `N` derivatives at a point `c` are at most `D`, then
`‖f⁽ⁿ⁾(c)‖ ≤ D (W + 1)^{n + N}` for every `n`. -/
theorem expPoly_iteratedDeriv_le {ι : Type*} [Fintype ι] (w : ι → ℂ) (W : ℝ) (hW0 : 0 ≤ W)
(hW : ∀ j, ‖w j‖ ≤ W) (c : ℂ) (N : ℕ) (q : ι → ℕ) (P : ι → ℂ[X])
(hP : ∀ j, P j = 0 ∨ (P j).natDegree < q j) (hN : ∑ j, q j = N) (D : ℝ) (hD0 : 0 ≤ D)
(hD : ∀ s < N, ‖iteratedDeriv s (fun z => ∑ j, (P j).eval z * Complex.exp (w j * z)) c‖ ≤ D)
(n : ℕ) :
‖iteratedDeriv n (fun z => ∑ j, (P j).eval z * Complex.exp (w j * z)) c‖
≤ D * (W + 1) ^ (n + N) := by
sorry
end Transcendence