Binet bound from a digamma asymptotic anchor
ProvedZeroFreeBridge.norm_digamma_sub_log_le_of_anchorarda-rh65k-transfernumber-theoryriemann-zeta
Let satisfy , and suppose as through natural numbers. Then
Here is the complex digamma function and is the principal logarithm.
Preamble
import Definitions.Def_ArdaRH65K_Turing121_DlvpTheta
import Mathlib
import Theorems.Thm_ZeroFreeBridge_log_succ_sub_log
import Theorems.Thm_ZeroFreeBridge_sum_norm_inv_sub_log_le
/- TURING RUNG T2, brick 1: the Riemann–Siegel theta function, defined BRANCH-CUT-FREE as the
integral of an explicit real integrand (TURING_METHOD_SCOPING §3, rung T2).
Classically `θ(t) = Im logΓ(1/4 + it/2) − (t/2)·log π` with a continuously-tracked branch of
`logΓ` — awkward to formalize. Instead we DEFINE
`θ(t) := ∫_0^t ( (1/2)·Re ψ(1/4 + i·u/2) − (1/2)·log π ) du`,
where `ψ = Complex.digamma = logDeriv Γ` (Mathlib). This agrees with the classical `θ`
(both vanish at `0` and have the same derivative `θ'(u) = (1/2)·Re ψ(1/4 + iu/2) − (1/2)·log π`)
and needs NO branch tracking: the integrand is a plain continuous real function of `u`.
This brick: the ray `1/4 + (u/2)·i` avoids all Γ-poles (`Re = 1/4 > 0`), Γ is ANALYTIC there
(differentiable on the open right half-plane ⟹ analytic), `ψ` is continuous on the ray
(`deriv Γ / Γ`, with `Γ ≠ 0` by `Gamma_ne_zero_of_re_pos`), hence the integrand is continuous,
`θ` is well-defined with `θ(0) = 0` and `θ' = integrand` (FTC for continuous integrands).
Later bricks (T2 cont.): effective two-sided bounds `θ(t) = (t/2)log(t/2π) − t/2 − π/8 +
O*(explicit/t)` via Binet-type digamma estimates; then T3 wires `θ` into the rectangle
argument identity `N(T) = θ(T)/π + 1 + S(T)`.
conjecture1_proved = False. -/
open Complex
section
namespace ZeroFreeBridge
end ZeroFreeBridge
open ZeroFreeBridge
/-! ## Brick 2: the Archimedean bridge — `θ'` IS the `Gammaℝ` log-derivative on the critical line.
`Gammaℝ s = π^(-s/2)·Γ(s/2)` is the Archimedean factor of the completed zeta
(`Λ = Gammaℝ · ζ` up to the `s(s-1)` polynomial), the object the T3 rectangle identity
splits off of `logDeriv Λ`. We compute `logDeriv Gammaℝ s = -(log π)/2 + (1/2)·ψ(s/2)`
on `Re(s/2) > 0` and specialize to the critical line `s = 1/2 + iu`, where `s/2 = thetaRay u`:
the θ integrand is EXACTLY `Re (logDeriv Gammaℝ (1/2 + iu))`. This hard-wires the
branch-cut-free `θ` into the counting formula's Archimedean term. -/
/-! ## Brick 3a: the θ-asymptotics skeleton (T2 rung, first sub-brick).
Target (brick 3 complete): `θ(t) = (t/2)·log(t/(2π)) − t/2 − π/8 + O*(explicit/t)`.
This sub-brick ships the three UNCONDITIONAL pieces every derivation route needs:
* `digamma_shift` — the iterated recursion `ψ(s+N) = ψ(s) + Σ_{k<N} (s+k)⁻¹` (from Mathlib's
single-step `digamma_apply_add_one`), which moves the evaluation point right, where the
Stirling comparison is easy;
* `norm_inv_sub_log_one_add_inv_le` — the telescoping step bound
`‖s⁻¹ − log(1+s⁻¹)‖ ≤ ‖s⁻¹‖²·(1−‖s⁻¹‖)⁻¹/2` (Mathlib `norm_log_one_add_sub_self_le`):
the summand of the Binet-series `ψ(s) − log s = −Σ_{k≥0} [(s+k)⁻¹ − log(1+(s+k)⁻¹)] + …`
is quadratically small, so the telescoped series converges with an explicit `O(1/‖s‖)` tail;
* `thetaMain` + `thetaMain_hasDerivAt` — the main term and its derivative
`(log(t/(2π)))/2`, matching `θ'`'s leading behaviour, so brick 3c can compare the two via
FTC on `[t₀, t]`.
Remaining sub-bricks: 3b = sum the telescoped series (Summable + tail bound ⟹ effective
`|ψ(s) − log s| ≤ C/‖s‖` on `Re s ≥ 1`, plus the anchor `ψ(s+N) − log(s+N) → 0`); 3c = integrate
`θ' − thetaMain'` and fix the `−π/8` constant. conjecture1_proved = False. -/
/-! ## Brick 3b: the telescoped Binet series — `|ψ(s) − log s| ≤ 1/(Re s − 1)` modulo the anchor.
Telescoping `d(s) = ψ(s) − log s` with `d(s+1) − d(s) = s⁻¹ − log(1+s⁻¹)` (branch-safe on
`Re s > 0` since `|arg| < π/2` on the right half-plane) gives the FINITE identity
`d(s) = d(s+N) − Σ_{k<N} [(s+k)⁻¹ − log(1+(s+k)⁻¹)]`; the 3a step bound + an elementary
inline telescoping estimate `Σ_{k<N} 1/(x+k)² ≤ 1/(x−1)` bound the sum UNIFORMLY in `N`;
letting `N → ∞` against the anchor `d(s+N) → 0` (the ONE remaining analytic input, carried as
an explicit hypothesis — brick 3b′ discharges it) yields the effective Binet bound.
conjecture1_proved = False. -/
Formal statement
theorem ZeroFreeBridge.norm_digamma_sub_log_le_of_anchor {s : ℂ} (hs : 2 ≤ s.re)
(hanchor : Filter.Tendsto
(fun N : ℕ => Complex.digamma (s + N) - Complex.log (s + N))
Filter.atTop (nhds 0)) :
‖Complex.digamma s - Complex.log s‖ ≤ 1 / (s.re - 1) := by sorry
Source