The tail of the Taylor series at 0 of an entire function bounded by M on the ball of radius R is at most M (r/R)^T / (1 − r/R) on the ball of radius r
ProvedTranscendence.taylor_tail_leLet be a complex normed space, let be analytic at every point, with on the closed ball of radius about , and let . For every with and every ,
where is the -th Fréchet derivative.
In the proof of Proposition 4.10 (Transcendence.siegel_small_values_of_count) it bounds the tail of the Taylor expansion at the origin; on with the sup norm, the balls are polydiscs.
It is one step of the proof of Baker's theorem formalised in this mission, which follows Chapter 4 of Waldschmidt's book: Baker's theorem from the criterion of Schneider–Lang for Cartesian products, as observed by Bertrand and Masser (1980).
Novelty. None: the bound for the tail in the proof of Lemma 4.13 of Waldschmidt's book (pp. 134–135), with the factor in place of the book's . The book bounds the tail by Schwarz's lemma and Parseval's formula; here Cauchy's inequality on the line through bounds each term, and the terms are summed as a geometric series. The contribution of this node is the formal proof.
import Mathlib
namespace Transcendence
/-- **The tail of a Taylor series** (Waldschmidt, *Diophantine Approximation on Linear Algebraic
Groups*, the tail half of Lemma 4.13, with Cauchy's inequality in place of Schwarz's lemma). Let `G` be an
entire function on a complex normed space `E`, bounded by `M` on the closed ball of radius `R`, and let
`0 < r < R`. For `‖z‖ ≤ r`, the Taylor polynomial of order `< T` of `G` at `0`, whose `k`-th term is
`D^k G(0)(z, …, z) / k!`, approximates `G z` within `M (r/R)^T / (1 - r/R)`. -/
theorem taylor_tail_le {E : Type*} [NormedAddCommGroup E] [NormedSpace ℂ E] {G : E → ℂ}
(hG : AnalyticOnNhd ℂ G Set.univ) {r R M : ℝ} (hr : 0 < r) (hrR : r < R)
(hM : ∀ x ∈ Metric.closedBall (0 : E) R, ‖G x‖ ≤ M) {z : E} (hz : ‖z‖ ≤ r) (T : ℕ) :
‖G z - ∑ k ∈ Finset.range T, (k.factorial : ℂ)⁻¹ * iteratedFDeriv ℂ k G 0 (fun _ => z)‖ ≤
M * (r / R) ^ T / (1 - r / R) := by
sorry
end Transcendence