One integration-by-parts step for a Bernoulli sawtooth
ProvedZetaReflection.em_saw_step_cpowarda-rh65k-transfernumber-theoryriemann-zeta
Let , , and let , with on and interval-integrable there. Let be the periodic Bernoulli sawtooth and the Bernoulli function. Then
This raises the sawtooth order on one unit interval.
Preamble
import Definitions.Def_ArdaRH65K_Turing121_EMZeta
import Definitions.Def_ArdaRH65K_Turing121_EMZetaComplex
import Definitions.Def_ArdaRH65K_Turing121_EMZetaTail
import Mathlib.Analysis.Analytic.Uniqueness
import Mathlib.Analysis.SpecialFunctions.ImproperIntegrals
import Mathlib.Analysis.SpecialFunctions.Pow.Deriv
import Mathlib.MeasureTheory.Function.Floor
import Mathlib.MeasureTheory.Integral.IntegralEqImproper
import Mathlib.MeasureTheory.Integral.IntervalIntegral.IntegrationByParts
import Mathlib.NumberTheory.LSeries.RiemannZeta
import Mathlib.NumberTheory.ZetaValues
/- EMZetaTail.lean -- A2 Theorem 1, TAIL track: the order-K Euler-Maclaurin tail engine.
The blocker of PROGRAM ANDÚRIL's reflection track. Builds on `EMZeta.lean` /
`EMZetaComplex.lean` (the K=1 chain). The K=1 remainder envelope `‖s‖/(2·Re s)` proven in
`EMZetaComplex.em_zeta_strip_enclosure` is sign-LOOSE (≈14 at s=1/2+14i), so the G2 reflected
band could not be made sign-tight from it. This file raises the Euler-Maclaurin order and cuts
the sum at a finite `N`, producing a remainder that DECAYS like `N^{-Re s - K + 1}`, small enough
to make `gLine` boxes sign-tight.
Architecture (bottom-up), ALL kernel-clean (AxiomGuardEMTail):
F. Order-2/3 periodized "saw" Bernoulli sup bounds: `|sawBernoulli 2| ≤ 1/6`,
`|sawBernoulli 3| ≤ 1/12` (via `B₃ = y(1−y)(1/2−y)`), and `bernoulliFun_three` closed form.
G. `em_saw_step` — the one IBP step raising the saw order by one over a unit cell (the proven
form of `EMZetaWip.em_saw_step_wip`), via the continuous polynomial surrogate
`B_{k+1}(x−m)/(k+1)`. Endpoint saw values `B_{k+1}(0)`, `B_{k+1}(1)` (equal for `k+1≠1`).
G'. `em_saw_step_window` — sum `em_saw_step` over `[M,N]`; interior boundary terms telescope
(`bernoulliFun_endpoints_eq_of_ne_one`), leaving only `B_{k+1}(0)(fk N − fk M)/(k+1)`.
H. `em_tail_integral_bound` — the reusable engine: `‖∫_N^∞ saw_k · c·x^{−s−k}‖`
≤ `B·‖c‖·N^{−(Re s+k−1)}/(Re s+k−1)`, the `N`-decay that makes gLine boxes sign-tight.
I. `em_tail3_bound` (order-3, `B = 1/12`) and `em_tail3_number` — THE NUMBER: at the G2 pilot
`s = 1/2 + 14i`, `N = 200`, the order-3 remainder is `≤ 10⁻³` (kernel-decided). Clears the
go/no-go the K=1 envelope `‖s‖/(2·Re s) ≈ 14` failed.
REMAINING for the full ζ-order-K identity (handoff): tie `em_saw_step_window` to `riemannZeta` by
raising the K=1 remainder `∫_1^∞ saw₁·(−s x^{−s−1})` (from `EMZetaComplex.em_zeta_strip`) to
order 3 over `[N,∞)` — apply `em_saw_step_window` on `[N,M]` (complex cpow orders 2,3, derivative
coefficients `s(s+1)…`), then `M → ∞` (boundary terms → 0; integral → `∫_N^∞`), reusing the
`intervalIntegral_tendsto_integral_Ioi` limit pattern from `em_zeta_cpow`. The remainder BOUND at
the endpoint is already `em_tail3_bound`; only the identity plumbing remains.
conjecture1_proved = False. This is a classical analysis lemma (Euler-Maclaurin tail bound),
NOT a proof of RH.
-/
open MeasureTheory intervalIntegral Set Filter Topology Complex
open scoped Real
section
namespace ZetaReflection
end ZetaReflection
open ZetaReflection
/-! ## F. Order-k periodized Bernoulli: derivative tower and explicit sup bounds. -/
/-! ## G. The saw-order-raising IBP step over a unit cell.
The antiderivative tower: `sawBernoulli (k+1) / (k+1)` has derivative `sawBernoulli k` on the
OPEN cell `(m, m+1)` (from `antideriv_bernoulliFun`, since `Int.fract` has derivative `1` off
ℤ and `bernoulliFun (k+1)/(k+1)` has derivative `bernoulliFun k`). IBP with this antiderivative
turns `∫ sawBernoulli k · fk` into a boundary term plus `∫ sawBernoulli (k+1) · fk'`, exactly
one order higher. The endpoint saw values are `bernoulliFun (k+1) 0` and `bernoulliFun (k+1) 1`
(the cell is `[m, m+1]`), which for `k+1 ≠ 1` are EQUAL by `bernoulliFun_endpoints_eq_of_ne_one`
— the mechanism that makes the interior Bernoulli boundary terms telescope to zero. -/
/-! ## G'. Summed saw-order-raising over an integer window `[M, N]`.
Summing `em_saw_step` over the cells `[M,M+1], …, [N-1,N]` and telescoping via
`sum_integral_adjacent_intervals_Ico`, the interior boundary terms cancel (for `k+1 ≠ 1` the
saw endpoint values `B_{k+1}(0) = B_{k+1}(1)` agree, so adjacent cells contribute
`+B_{k+1}(0)·f(j)` and `−B_{k+1}(0)·f(j)`), leaving only the two outer endpoints:
∫_M^N saw_k · fk
= B_{k+1}(0)·(fk N − fk M)/(k+1) − (∫_M^N saw_{k+1} · fk1)/(k+1).
(Here `k ≥ 1` so `k + 1 ≠ 1`, which is what makes the telescoping clean.) -/
/-! ## H. The order-K tail remainder integrals and their explicit `N`-decaying bounds.
The consumer needs a remainder over the TAIL `[N, ∞)` whose norm decays in `N`. We work with
the complex integrand `sawBernoulli k x · (c · (x:ℂ)^(-s-k))` where `c` is the falling-factorial
coefficient `(-1)^k s(s+1)…(s+k-1)` (the k-th derivative coefficient of `x^{-s}`). The norm
bound is the clean product form
‖∫_N^∞ saw_k · c·x^{-s-k}‖ ≤ Bsup_k · ‖c‖ · N^{-Re s - k + 1} / (Re s + k - 1),
from `‖saw_k‖ ≤ Bsup_k`, `‖x^{-s-k}‖ = x^{-Re s - k}`, and `∫_N^∞ x^{-Re s -k} = N^{-Re s -k+1}/(Re s+k-1)`. -/
/-! ## H'. The SHIFTED-BASE tail engine `(w + x)^{-a}` (the Binet / Γℝ substrate).
The base-1 engine `em_tail_integral_bound` bounds saw-weighted `x^{-s-k}` tails. The Binet
remainder `μ_K(w)` (Stirling series for `ψ`) is, up to the finite Bernoulli peel, an
Euler–Maclaurin tail of the trigamma power series `Σ (w+n)^{-2-k}` — a SHIFTED-BASE `(w+x)^{-a}`
tail. This subsection generalizes the engine to that base. Crucially the exponent is kept REAL
(`a : ℝ`, `a > 1`) — which is exactly the form the trigamma tail needs (`a = 2 + k ∈ ℕ`) — so
the norm is `‖(w+x)^{-a}‖ = ‖w+x‖^{-a}` (`Complex.norm_cpow_real`, no `exp(arg·im)` factor), and
the honest `N`-decay is driven by the base lower bound `‖w+x‖ ≥ Re(w+x) = Re w + x ≥ Re w + N`. -/
/-! ## I. The K = 3 instance and THE NUMBER (the go/no-go for full reflection).
The order-3 tail remainder integrand is `sawBernoulli 3 x · (c₃(s) · x^{-s-3})` where
`c₃(s) = -s(s+1)(s+2)` is the coefficient of `x^{-s-3}` in `d³/dx³(x^{-s})`. `em_tail_integral_bound`
bounds this integral for ANY complex `c`; here we record it with the honest falling-factorial
coefficient and the proven saw-3 sup bound `1/12`, then evaluate the numeric envelope at the G2
pilot point `s = 1/2 + 14i` and certify a tail cut making it `< 10⁻³` (the K = 1 envelope
`‖s‖/(2·Re s) ≈ 14` did NOT — see EMZetaComplex.em_zeta_strip_enclosure). -/
/-! ## J. The order-2 windowed EM identity for the ζ integrand (complex).
We raise the K=1 windowed complex EM identity (`euler_maclaurin_one_window_cpow` applied to
`f x = (x:ℂ)^(-s)`) to order 2 using `em_saw_step_window` (real coefficients, applied to the
real and imaginary parts — but here directly to the ℂ integrand via linearity of the identity
over the real saw). The order-2 remainder integrand carries the coefficient `c₂(s) = s(s+1)`.
We give the complex σ-direction derivatives to order 2 (the inputs `em_saw_step_window` needs),
then the windowed order-2 identity. This is the reusable step; iterating once more gives order 3,
and the `[N,∞)` limit (handoff) yields the ζ representation. -/
/-! ### ℂ-valued saw-order-raising over a window (for the ζ integrand).
The ℂ lift of `em_saw_step` and `em_saw_step_window`. The saw factor `sawBernoulli k x` stays
REAL (cast to ℂ); `fk, fk1` are ℂ-valued. Proof structure is identical to the real versions
with `linear_combination` in place of `linarith` for the final ℂ arithmetic. -/
Formal statement
theorem ZetaReflection.em_saw_step_cpow (k : ℕ) (m : ℤ) (fk fk1 : ℝ → ℂ)
(hd : ∀ x ∈ Icc (m : ℝ) (m + 1), HasDerivAt fk (fk1 x) x)
(hi : IntervalIntegrable fk1 volume (m : ℝ) (m + 1)) :
(∫ x in (m : ℝ)..(m + 1), (sawBernoulli k x : ℂ) * fk x)
= ((bernoulliFun (k + 1) 1 : ℂ) * fk (m + 1) - (bernoulliFun (k + 1) 0 : ℂ) * fk m) / (k + 1)
- (∫ x in (m : ℝ)..(m + 1), (sawBernoulli (k + 1) x : ℂ) * fk1 x) / (k + 1) := by sorry
Source