Bound for the fractional-part zeta integral
ProvedZeroFreeBridge.zeta_repr_integral_boundarda-rh65k-transfernumber-theoryriemann-zeta
If has , then
Here is the real fractional part and the integral is over .
Preamble
import Mathlib
/- THE BRIDGE: 3 Re L(s) + 4 Re L(s+it) + Re L(s+2it) >= 0 for Re s > 1,
where L(Lambda, .) = -zeta'/zeta is the von Mangoldt L-series.
The Mertens nonnegative-cosine certificate (3 + 4 cos + cos 2 = 2(1+cos)^2, from
TrigNonneg), proven in-kernel to meet the actual Dirichlet series of -zeta'/zeta.
This is the KEY INEQUALITY of the classical zero-free region -- the positivity that,
combined with an (unformalized, research-scale) zeta growth bound, yields
zeta(s) != 0 for Re s > 1 - c/log|t|. A certificate FEEDING the region, NOT the
region and NOT a proof of RH. conjecture1_proved = False. -/
open scoped Real
open Filter Topology MeasureTheory
section
namespace ZeroFreeBridge
end ZeroFreeBridge
open ZeroFreeBridge
/- ===================================================================================
RESIDUE OF THE LOGARITHMIC DERIVATIVE = ORDER. A general complex-analysis lemma
(NOT RH-specific): for f meromorphic at z0 of order n (a zero if n>0, a pole if n<0),
(z - z0) * logDeriv f z -> n as z -> z0 within the punctured neighbourhood.
Mathlib v4.32.0 has ONLY the analytic simple-zero case
(AnalyticAt.tendsto_mul_logDeriv_simple_zero, n=1); the general-order version is a
genuine gap here (present only on Mathlib master). Built from the local factorization
meromorphicOrderAt_eq_int_iff (f =ᶠ (z-z0)^n • g, g analytic, g z0 != 0).
=================================================================================== -/
/- The same positivity, restated LITERALLY about zeta's logarithmic derivative -zeta'/zeta,
via Mathlib's LSeries(vonMangoldt) s = -zeta'(s)/zeta(s) (Re s > 1). This is the exact
form the classical zero-free-region argument uses: 3 Re(-zeta'/zeta)(sigma) + 4 Re(...)(sigma+it)
+ Re(...)(sigma+2it) >= 0. Still NOT a proof of RH -- the region needs the log-derivative ->
product step and the zeta growth bound (both unformalized, research-scale). -/
/- ===================================================================================
THE BRIDGE-POWERED BOUNDARY CONTRADICTION (classical de la Vallee Poussin core).
The Mertens positivity 3 Re(-zeta'/zeta)(sigma) + 4 Re(...)(sigma+it) + Re(...)(sigma+2it) >= 0
(proven above), multiplied by (sigma-1) > 0 and taken to sigma -> 1+, forces
3*1 - 4k - k' >= 0, where +1 is the residue of -zeta'/zeta at the simple pole s=1 and
-k, -k' are the residues at 1+it, 1+2it (k = order of a zero at 1+it, k' >= 0 at 1+2it).
With k >= 1 this is 3 - 4k - k' <= -1 < 0, impossible: so zeta has NO zero of order >= 1
at 1+it. This is exactly how the classical zero-free-region boundary (zeta(1+it) != 0)
is forced by the 3-4-1 inequality -- routed here through -zeta'/zeta (a different internal
path from Mathlib's product-route riemannZeta_ne_zero_of_one_le_re).
The three residue LIMITS are taken as hypotheses (hpole/hz1/hz2): each is the real-line
restriction of residue_logDeriv (proven above: (z-z0)*logDeriv f z -> order) applied to
zeta at 1, 1+it, 1+2it -- given zeta's meromorphy/orders there. Discharging them fully needs
zeta's simple-pole handle at s=1 (a v4.32.0 API gap, via completedRiemannZeta) + order
extraction + the .re/real-ray plumbing. conjecture1_proved = False; NOT a proof of RH --
this is the boundary (c=0) edge of the classical region, which Mathlib already has. -/
/- ===================================================================================
IMPROVED ZERO-FREE CERTIFICATE (degree 3). The de la Vallee Poussin polynomial
3 + 4cos + cos2 = 2(1+cos)^2 is NOT optimal. Optimizing the leading-order zero-free
functional F(P) = (sqrt a_1 - sqrt a_0)^2 / sum_{k>=1} a_k over NONNEGATIVE cosine
polynomials (a_k >= 0) is exactly the Mossinghoff-Trudgian 2015 program (J. Number
Theory 157; improved region constant R_0 = 5.573412). The (1+cos)^n family gives clean
Fejer-Riesz certificates capturing most of the gain. Here n = 3:
20 + 30 cos + 12 cos2 + 2 cos3 = 8 (1 + cos)^3 >= 0, a_1 = 30 > a_0 = 20,
with F = 0.02296 vs 0.01436 for de la Vallee Poussin (1.60x wider region, leading order).
Generator UNTRUSTED, Lean kernel sole arbiter. This still only FEEDS the classical region
CONSTANT -- it improves the constant, NOT the Vinogradov-Korobov rate 1/(log t)^{2/3}, and
is NOT a proof of RH. conjecture1_proved = False.
=================================================================================== -/
/- ===================================================================================
GENERAL DEGREE-n CERTIFICATE. Everything above is the n=2 (Mertens) and n=3 instances of
a single statement: ANY pointwise-nonnegative cosine polynomial
P(φ) = sum_{k < N} a k * cos (k φ) >= 0
yields the zero-free-region positivity on -zeta'/zeta. This subsumes the (1+cos)^n family
(`one_add_cos_pow_nonneg`, whose cosine coefficients are all >= 0) and every Mossinghoff-Trudgian
optimal polynomial -- the whole certificate cone, at once, for arbitrary degree.
Generator UNTRUSTED, Lean kernel sole arbiter. Still FEEDS the classical region constant only,
NOT the Vinogradov-Korobov rate, NOT a proof of RH. conjecture1_proved = False.
=================================================================================== -/
/- ===================================================================================
THE HINGE. The general positivity feeds the classical boundary conclusion: ANY admissible
certificate forces `ζ(1+it) ≠ 0`. "Admissible" = coefficients `a_k ≥ 0` (residue upper bounds),
pointwise-nonnegative cosine polynomial (positivity), and the single inequality `a 0 < a 1`.
That last condition is EXACTLY the hinge: the pole residue `+a 0` at `s=1` loses to a zero
residue `-a 1·m` (m ≥ 1) at `1+it` precisely when `a 1 > a 0`. The Fejer bound `a_1 < 2 a_0`
caps how far this can be pushed. Still the c=0 boundary (`ζ(1+it) ≠ 0` is already in Mathlib),
NOT a proof of RH. conjecture1_proved = False.
=================================================================================== -/
/- ===================================================================================
MAGNITUDE LAYER (Layer 2) -- first brick. The reassessment
(docs/RH_ZERO_FREE_REASSESSMENT_2026-08-30) located the frontier BEYOND the (Fejer-capped)
positivity layer in the MAGNITUDE bound |zeta(sigma+it)|. This is the sigma>1 BASE CASE: the
elementary Dirichlet-triangle bound |zeta(s)| <= sum_n (n+1)^{-Re s} = zeta(Re s). It is NOT
the frontier -- the zero-free-region improvement needs |zeta| growth INSIDE the critical strip
(|zeta(sigma+it)| << |t|^{1-sigma}, or << log|t| near sigma=1), which needs Euler-Maclaurin / the
{x}-integral continuation (the harder next piece, Mathlib-API dependent). This brick is the honest
base: |zeta| is finite for sigma>1, and the RHS blows up like 1/(sigma-1) as sigma->1+, which is
exactly why the strip needs the finer bound. A different KIND of object from the positivity certs
(a magnitude bound, not a positivity), opening Layer 2. conjecture1_proved = False. -/
/- The STRIP growth bound, reduced to its two analytic inputs. Research (agentic campaign,
docs/RH_ZERO_FREE_REASSESSMENT §8) confirmed Mathlib v4.32.0 has NO strip |zeta| bound and NO
complex fractional-part integral representation, so the crude strip growth bound
|zeta(sigma+it)| <= ||s||/||s-1|| + ||s||/sigma (<< |t|, for 0 < sigma < 1)
must be assembled from two analytic facts still to be formalized (the identified remaining work):
(R) the representation zeta(s) = s/(s-1) - s * I with I = integral_1^inf {x} x^{-s-1} dx
(Re s > 0; via Abel summation + the identity theorem -- a genuine Mathlib gap-filler), and
(B) the integral bound ||I|| <= 1/Re s (from |{x}| <= 1 and integral_1^inf x^{-sigma-1} = 1/sigma).
This theorem does the (high-confidence) ASSEMBLY, taking (R),(B) as hypotheses -- exactly the style
of `zeta_boundary_contradiction`, which likewise takes its analytic limits as hypotheses. Crude
growth only (~|t|), NOT the sharp |t|^{1-sigma} nor the log|t| that feeds the region; and the VK rate
needs VMVT (absent from Mathlib). conjecture1_proved = False. -/
Formal statement
theorem ZeroFreeBridge.zeta_repr_integral_bound {s : ℂ} (hs : 0 < s.re) :
‖∫ x in Set.Ioi (1 : ℝ), ((Int.fract x : ℝ) : ℂ) / (x : ℂ) ^ (s + 1)‖ ≤ 1 / s.re := by sorry
Source