Gamma-phase approximation and its convergence condition
DefinitionArdaRH65K_ThetaValuearda-rh65k-transfernumber-theoryriemann-zeta
For real and nonnegative integer , define
The formal convention is . The bundled convergence condition says that for with , the sequence tends to a real for which .
Definition code
import Mathlib.Algebra.Ring.GeomSum
import Mathlib.Analysis.Calculus.Deriv.MeanValue
import Mathlib.Analysis.SpecialFunctions.Complex.Arg
import Mathlib.Analysis.SpecialFunctions.Gamma.Beta
import Mathlib.Analysis.SpecialFunctions.Integrals.Basic
import Mathlib.Analysis.SpecialFunctions.Trigonometric.ArctanDeriv
/- ThetaValue.lean -- ANDÚRIL A4 θ-value instrument: toward a kernel box for
φ(t) = arg Γℝ(1/2+it) = the Riemann–Siegel θ, with NO numeric hypotheses.
## THE ROUTE (Euler/Weierstrass limit form, γ-free).
Γℝ(s) = π^{-s/2}·Γ(s/2), so with `s = 1/2+it`, `t = 14`:
φ(14) = arg Γℝ(1/2+14i) = −7·log π + Im log Γ(1/4 + 7i).
(Numerically φ(14) = θ_RS(14) ≈ −1.7829487; verified in the driver against mpmath's
`siegeltheta(14)` and against `arg Γℝ` directly.)
For `Im log Γ(z)` with `z = x+iy` (`x = 1/4`, `y = 7`) we use Mathlib's Gauss product
`GammaSeq z n = n^z·n!/∏_{k=0}^n (z+k) → Γ(z)` (`Complex.GammaSeq_tendsto_Gamma`). Its
principal-log has imaginary part (this file's `imLn_formula`, PROVED):
Im L_n = y·log n − Σ_{k=0}^n arctan(y/(x+k)),
L_n := z·log n + log(n!) − Σ_{k=0}^n log(z+k) (all principal complex logs),
and `exp(L_n) = GammaSeq z n` (this file's `exp_Ln_eq_gammaSeq`, PROVED). Each `arctan`
term is boxed by the reusable `ArctanTaylor.arctan_bracket` instrument (no ψ, no Γ anchor).
## WHAT IS PROVED HERE (all axiom-clean, no numeric hypotheses).
* `arg_eq_arctan_of_re_pos` : `w.arg = arctan(im w/re w)` for `re w > 0` (the per-factor atom).
* `imLn_formula` : the exact `Im L_n` formula above, GENERAL in `(x,y,n)`, `x>0`.
* `exp_Ln_eq_gammaSeq` : `exp(L_n) = GammaSeq z n`, the bridge to the Gauss product.
Together with `ArctanTaylor.arctan_bracket` these are a COMPLETE finite kernel pipeline: for
any `n` they reduce `Im L_n` to a rational interval with no numeric hypotheses.
## THE PRECISE REMAINING GAP (named Props, honestly stated -- see mission "land D1 + the gap").
Two obligations separate the proven finite `Im L_n` from the target `φ(14)` box:
* `ThetaGap.ConvergenceObligation` (branch bridge) : `imLnVal x y n → Λ` for a real `Λ` that
is a genuine argument of `Γ(z)` (`Γ(z) = ‖Γ(z)‖·exp(IΛ)`). IMPORTANT honest correction,
driver-verified: `Im L_n` converges to the CONTINUOUS unwrapped value `Im loggamma(1/4+7i)
≈ 6.230`, which is `arg Γ(z) + 2π` — NOT the principal `(Complex.log (Γ z)).im ≈ −0.053`.
(`Im L_n` sweeps from `−2.93` at `n=1` up through `6.23`, crossing far outside `(−π,π)`.) So
the correct target is `Λ` characterized by the exp identity, not a principal-branch equality.
This is the one genuine limit-interchange (from `exp_Ln_eq_gammaSeq` +
`Complex.GammaSeq_tendsto_Gamma` + `exp` continuity, tracking the winding); a `Tendsto`
obligation, NOT a numeric hypothesis.
* `ThetaGap.RateObligation` (rate) : `|Λ − imLnVal x y n| ≤ C(x,y)/n` with an EXPLICIT
elementary `C` -- provable by the integral-comparison harmonic-tail bound (`1/(2n)`-type
`∑ 1/(x+k)` vs `log`) plus the cubic `arctan(u)=u−r`, `0<r<u³/3` residual, using only
elementary calculus in Mathlib and this file's arctan corollaries. Driver: at `n=2000`,
honest `k≤30`, box width ≈ 9.4e-4 < the 2e-3 target.
With both, `φ(14) = −7·log π + Λ` (continuous branch) is boxed by `−7·[log π box] +
[imLnVal box] ± [rate]` -- a kernel interval of width ≤ 2e-3, no hypotheses. `log π` is boxed
by the corpus verified-log pattern, exactly as `TrigReduceOperating` boxes `log 2`. The shape
of the final consumption is recorded in `ThetaGap.PhiTargetShape`.
conjecture1_proved = False. The finite Im-part pipeline of the θ-value, kernel-clean; the θ
VALUE box itself awaits the two named Props above (rate + branch bridge). NOT a proof of RH.
-/
open Complex Real
namespace ThetaValue
/-! ### The per-factor atom: `arg` of a right-half-plane point is an `arctan`. -/
/-! ### The exact imaginary part of the finite log-Weierstrass sum `L_n`. -/
/-! ### The bridge to the Gauss product: `exp(L_n) = GammaSeq z n`. -/
/-! ### Concrete instrument demonstration: a kernel box for a θ-series arctan term.
These show the `ArctanTaylor` instrument produces REAL numeric boxes with no hypotheses --
the atomic step the θ-series scale-up repeats per term. We box the two `n₀ = 1` terms
(`k = 0, 1`) of the `x = 1/4, y = 7` series: `arctan(7/(1/4)) = arctan 28` and
`arctan(7/(1/4+1)) = arctan(28/5)`. -/
end ThetaValue
/-! ## The precise remaining gap -- named `Prop`s, honestly stated. -/
namespace ThetaGap
open ThetaValue
/-- The finite θ-series partial value `Im L_n = y·log n − Σ_{k=0}^n arctan(y/(x+k))`. -/
noncomputable def imLnVal (x y : ℝ) (n : ℕ) : ℝ :=
y * Real.log n - ∑ k ∈ Finset.range (n+1), Real.arctan (y/(x+k))
/-- **CONVERGENCE + LIMIT CHARACTERIZATION (the remaining core).** There is a real limit `Λ`
(the CONTINUOUS `arg Γℝ`-companion, i.e. the unwrapped `Im log Γ(z)`) with:
(i) `imLnVal x y n → Λ` (`Filter.Tendsto`), and
(ii) `Γ(z) = ‖Γ(z)‖ · exp(I·Λ)` (Λ is a genuine argument of `Γ(z)`, mod `2π`).
NOTE the honest correction (driver-verified): `Im L_n` converges to the CONTINUOUS log value
(`Im loggamma(z) ≈ 6.230` at `z=1/4+7i`), which is `arg Γ(z) + 2π` — it is NOT the principal
`(Complex.log (Γ z)).im`. Characterizing `Λ` by (ii) captures the correct branch without
committing to a wrong principal-branch identity. Condition (i) is the genuine
limit-interchange (from `exp_Ln_eq_gammaSeq` + `Complex.GammaSeq_tendsto_Gamma` + `exp`
continuity, tracking the winding); (ii) fixes the branch. Neither is a numeric hypothesis. -/
def ConvergenceObligation : Prop :=
∀ (x y : ℝ), 0 < x → Complex.Gamma ((x:ℂ) + (y:ℂ)*I) ≠ 0 →
∃ Λ : ℝ,
Filter.Tendsto (imLnVal x y) Filter.atTop (nhds Λ) ∧
Complex.Gamma ((x:ℂ) + (y:ℂ)*I)
= (‖Complex.Gamma ((x:ℂ) + (y:ℂ)*I)‖ : ℂ) * Complex.exp (Complex.I * (Λ:ℂ))
end ThetaGap
def ArdaRH65K_ThetaValue : Nat := 0
Source