The central saddle segment and gamma error for a fixed Mellin coefficient
DefinitionDeBruijnNewman_Dobner_SaddleFor the existing Riemann gamma factor
define the relative error after retaining the leading linear term by
For real , a positive integer , and a point of height , put
The central Gaussian–Mellin contribution is
Here is the existing definition. The logarithm of a complex number is the principal logarithm. The segment is oriented upward; its contour Jacobian is included by cancellation with the usual prefactor.
Formalization Note. These quantities are gammaLinearError, mellinWindow, mellinSaddlePoint, and centralMellinTerm. The natural-number index represents . The definitions are total; analytic theorems impose negative time and sufficiently large positive height. The center used here is the leading saddle , a fixed-index variant of the paper's corrected saddle with .
import Definitions.Def_DeBruijnNewman_Dobner_Mellin
open MeasureTheory Set
namespace DeBruijnNewman.Dobner
/-- Relative error after retaining the leading linear term of `log gamma`.
The local estimate for this error is a separate theorem. -/
noncomputable def gammaLinearError (s z : ℂ) : ℂ :=
gammaFactor z /
(gammaFactor s * Complex.exp
((1 / 2 : ℂ) * Complex.log (s / ((2 * Real.pi : ℝ) : ℂ)) * (z - s))) - 1
/-- Height of the central segment in Dobner's contour argument. -/
noncomputable def mellinWindow (y : ℝ) : ℝ := y ^ (2 / 3 : ℝ)
/-- The leading saddle for a fixed coefficient, parametrized vertically.
Its real shift is `|t| log(n+1) / 2`. -/
noncomputable def mellinSaddlePoint (t : ℝ) (s : ℂ) (n : ℕ) (u : ℝ) : ℂ :=
s + ((|t| / 2 * Real.log ((n : ℝ) + 1) : ℝ) : ℂ) + (u : ℂ) * Complex.I
/-- The contribution of the finite segment through the leading saddle.
The contour Jacobian cancels the factor `1/i` in the contour prefactor. -/
noncomputable def centralMellinTerm (t : ℝ) (s : ℂ) (n : ℕ) : ℂ :=
(1 / (Real.sqrt (Real.pi * |t|) : ℂ)) *
∫ u in Icc (-mellinWindow s.im) (mellinWindow s.im),
gammaFactor (mellinSaddlePoint t s n u) *
Complex.exp
((J t s - mellinSaddlePoint t s n u) ^ 2 / ((|t| : ℝ) : ℂ) -
mellinSaddlePoint t s n u * (Real.log ((n : ℝ) + 1) : ℂ))
end DeBruijnNewman.Dobner