Lean source theorem: sq_tsum_le
ProvedErdosProblems.Shared.DirichletPole.sq_tsum_lelean-sharedsupport-lemma
Under the stated nonnegative Dirichlet-convolution bounds and a uniform bound B on weighted partial sums of b, the square of the Dirichlet series sum for a at any real s≥1 is at most B² times the corresponding sum for m, provided both series are summable.
Preamble
import Definitions.Def_ErdosProblems_Shared_DirichletPoleComparison import Mathlib.Analysis.SpecialFunctions.Pow.Real import Mathlib.NumberTheory.LSeries.Convolution /-! # Two simple poles exclude a squared comparison of Dirichlet series Let `a, g, b, m : ℕ → ℝ` be nonnegative with * `a ≤ g ⋆ b` and `g ⋆ g ≤ m` coefficientwise (Dirichlet convolution, `n ≠ 0`), * `g ≤ a`, and the partial sums of `∑ b(n) / n` bounded by `B`. If the Dirichlet series of `a` and of `m` both have a simple pole at `s = 1` (`(s - 1) L(a, s) → r_a > 0` and `(s - 1) L(m, s) → r_m ≠ 0` as `s → 1⁺`), this is impossible. Indeed, for real `s > 1` near `1`, `L(a, s) ^ 2 ≤ (L(g, s) L(b, s)) ^ 2 ≤ B ^ 2 L(g ⋆ g, s) ≤ B ^ 2 L(m, s)`, so `((s - 1) L(a, s)) ^ 2 ≤ B ^ 2 (s - 1) · (s - 1) L(m, s)`; the left side tends to `r_a ^ 2 > 0` and the right side to `0`. This is the analytic core of the classical proof that a non-square element of a number field is a non-square modulo infinitely many degree-one primes: `a` counts the ideals of the base field, `m` those of the quadratic extension, `g` the ideals built from split degree-one primes and `b` the rest. Summability of the two series near `s = 1` is not assumed: it follows from the nonzero limits, since a divergent series has `tsum = 0`. -/ noncomputable section open Filter Topology open scoped LSeries.notation open ErdosProblems.Shared.DirichletPole
Formal statement
theorem ErdosProblems.Shared.DirichletPole.sq_tsum_le {a g b m : ℕ → ℝ}
(ha : ∀ n, 0 ≤ a n) (hg : ∀ n, 0 ≤ g n) (hb : ∀ n, 0 ≤ b n)
(hga : ∀ n, g n ≤ a n)
(h1 : ∀ n, n ≠ 0 → a n ≤ dconv g b n)
(h2 : ∀ n, n ≠ 0 → dconv g g n ≤ m n)
(B : ℝ) (h3 : ∀ X : ℕ, ∑ n ∈ Finset.range X, b n / n ≤ B)
{s : ℝ} (hs : 1 ≤ s) (hsa : Summable (rterm a s)) (hsm : Summable (rterm m s)) :
(∑' n, rterm a s n) ^ 2 ≤ B ^ 2 * ∑' n, rterm m s n := by sorry
Source
Lean source (Apache-2.0): https://github.com/wcook04/plectis-erdos-lean/blob/4fe59e0191169606942baae0949365070b7f419c/ErdosProblems/Shared/DirichletPoleComparison.lean#L108-L143
Related paper by Will Cook (CC-BY-4.0): https://github.com/wcook04/plectis-erdos/blob/6917e15ec4abc2623512254da93221e446eeb707/paper/243/erdos-243-reciprocal-tail-rigidity.tex#L1-L58
Paper's AI-assistance disclosure: https://github.com/wcook04/plectis-erdos/blob/6917e15ec4abc2623512254da93221e446eeb707/paper/paper-house-style.sty#L180-L188
Original Erdős problem and Koizumi prior work are distinguished in the paper: https://github.com/wcook04/plectis-erdos/blob/6917e15ec4abc2623512254da93221e446eeb707/paper/243/erdos-243-reciprocal-tail-rigidity.tex#L57-L110