Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

A complete exponential-sum bound with Mobius cancellation in both Type I terms

Proved
Helfgott.etaTwo_vonMangoldt_mobius_sieve_upper

by flyingviper · Nov 11, 2026 · Mathlib 0df444a (Lean v4.33.1)

explicit-boundsgoldbachminor-arcsselberg-sievevaughan-identity

Let U,V,R,q,aU,V,R,q,aU,V,R,q,a be natural numbers with 2≤R≤V2\le R\le V2≤R≤V, 2≤q≤U2\le q\le U2≤q≤U, and (a,q)=1(a,q)=1(a,q)=1. Suppose y≥8y\ge8y≥8, 8U≤y8U\le y8U≤y, 4V≤y4V\le y4V≤y, UV≤yUV\le yUV≤y, q2V∣β∣≤1q^2V|\beta|\le1q2V∣β∣≤1, and qUV∣β∣≤1/2qUV|\beta|\le1/2qUV∣β∣≤1/2. Write α=a/q+β\alpha=a/q+\betaα=a/q+β, δ=βy\delta=\beta yδ=βy, M(0)=1M(0)=1M(0)=1, and M(δ)=min⁡(1,18/(πδ)2)M(\delta)=\min(1,18/(\pi\delta)^2)M(δ)=min(1,18/(πδ)2) otherwise. Put

P=Vlog⁡4+2Vlog⁡V,K=36+8πlog⁡2∣δ∣,P=V\log4+2\sqrt V\log V,\qquad K=36+8\pi\log2|\delta|,P=Vlog4+2V​logV,K=36+8πlog2∣δ∣, T1=yq(log⁡(y/q)M(δ)+log⁡4)+2⌊U/q⌋((log⁡y+log⁡4)K+16log⁡2)+36log⁡(y+2)q2Uymin⁡{U+1,8(⌊U/q⌋+1)},T_1=\frac yq\bigl(\log(y/q)M(\delta)+\log4\bigr)+2\lfloor U/q\rfloor\bigl((\log y+\log4)K+16\log2\bigr) +\frac{36\log(y+2)q^2U}{y}\min\{U+1,8(\lfloor U/q\rfloor+1)\},T1​=qy​(log(y/q)M(δ)+log4)+2⌊U/q⌋((logy+log4)K+16log2)+y36log(y+2)q2U​min{U+1,8(⌊U/q⌋+1)}, TC=yqM(δ)(log⁡V+log⁡4+4+log⁡q)+2UKP+72Uq(U+q)yVP.T_C=\frac yq M(\delta)(\log V+\log4+4+\log q)+2UKP+\frac{72Uq(U+q)}yVP.TC​=qy​M(δ)(logV+log4+4+logq)+2UKP+y72Uq(U+q)​VP.

Let J=⌊log⁡2⌊⌊y/(V+1)⌋/(U+1)⌋⌋+1J=\lfloor\log_2\lfloor\lfloor y/(V+1)\rfloor/(U+1)\rfloor\rfloor+1J=⌊log2​⌊⌊y/(V+1)⌋/(U+1)⌋⌋+1 and Cj=(U+1)2jC_j=(U+1)2^jCj​=(U+1)2j, using the natural-number logarithm equal to zero at zero. For a block [C,D)[C,D)[C,D) put

T=⌊y/C⌋,N=T+1,L=⌊max⁡(D−C,0)/q⌋+1,T=\lfloor y/C\rfloor,\quad N=T+1,\quad L=\lfloor\max(D-C,0)/q\rfloor+1,T=⌊y/C⌋,N=T+1,L=⌊max(D−C,0)/q⌋+1, KU=∑d,e≤Uμ(d)μ(e)lcm⁡(d,e),E=DKU+U2,H=log⁡N(Nlog⁡4+2Nlog⁡N),K_U=\sum_{d,e\le U}\frac{\mu(d)\mu(e)}{\operatorname{lcm}(d,e)},\quad E=DK_U+U^2,\quad H=\log N(N\log4+2\sqrt N\log N),KU​=d,e≤U∑​lcm(d,e)μ(d)μ(e)​,E=DKU​+U2,H=logN(Nlog4+2N​logN), A0=16(log⁡2)2y/C,B0′=224log⁡2 D/y,A_0=16(\log2)^2y/C,\quad B_0'=224\log2\,D/y,A0​=16(log2)2y/C,B0′​=224log2D/y, Q(C,D)=HEL(2A0+min⁡{2B0′q2,A0+2qA0B0′}),Q(C,D)=HEL\left(2A_0+\min\{2B_0'q^2,A_0+2q\sqrt{A_0B_0'}\}\right),Q(C,D)=HEL(2A0​+min{2B0′​q2,A0​+2qA0​B0′​​}), A1=16(log⁡2)2qL(Nφ(q)log⁡R+R4)(1+2πND∣β∣)2,A_1=16(\log2)^2qL\left(\frac N{\varphi(q)\log R}+R^4\right)(1+2\pi ND|\beta|)^2,A1​=16(log2)2qL(φ(q)logRN​+R4)(1+2πND∣β∣)2, S(C,D)=A1(Nlog⁡Nlog⁡4)E+8log⁡2Nlog⁡Nmax⁡(D−C,0)E.S(C,D)=\sqrt{A_1(N\log N\log4)E}+8\log2\sqrt N\log N\sqrt{\max(D-C,0)}\sqrt E.S(C,D)=A1​(NlogNlog4)E​+8log2N​logNmax(D−C,0)​E​.

Then

∣∑n≥1Λ(n)η2(n/y)e(αn)∣≤T1+TC+∑j=0J−1min⁡{Q(Cj,Cj+1),S(Cj,Cj+1)}.\left|\sum_{n\ge1}\Lambda(n)\eta_2(n/y)e(\alpha n)\right| \le T_1+T_C+\sum_{j=0}^{J-1}\min\{\sqrt{Q(C_j,C_{j+1})},S(C_j,C_{j+1})\}.​n≥1∑​Λ(n)η2​(n/y)e(αn)​≤T1​+TC​+j=0∑J−1​min{Q(Cj​,Cj+1​)​,S(Cj​,Cj+1​)}.

Here e(t)=exp⁡(2πit)e(t)=\exp(2\pi it)e(t)=exp(2πit) and η2(t)=4max⁡(log⁡2−∣log⁡(2t)∣,0)\eta_2(t)=4\max(\log2-|\log(2t)|,0)η2​(t)=4max(log2−∣log(2t)∣,0) for t>0t>0t>0, zero otherwise. The Type I and correction bounds preserve resonant Mobius cancellation and include all discretization and nonresonant errors. The estimate is unconditional, with conservative constants and explicit phase-range hypotheses.

Preamble
import Definitions.Def_Helfgott_VaughanData
import Mathlib.Algebra.Order.Floor.Ring
import Mathlib.Analysis.SpecialFunctions.Sqrt
import Mathlib.Data.Nat.Log
import Mathlib.Data.Nat.Totient
open Helfgott Finset ArithmeticFunction
open scoped BigOperators
Formal statement
theorem Helfgott.etaTwo_vonMangoldt_mobius_sieve_upper (U V R q a : ℕ)
    (hR : 2≤R) (hRV : R≤V) (hq : 2≤q) (hqU : q≤U) (ha : Nat.Coprime a q)
    (y beta : ℝ) (hy : 8≤y) (hUy : 8*(U:ℝ)≤y) (hVy : 4*(V:ℝ)≤y)
    (hUV : (U:ℝ)*(V:ℝ)≤y) (hb : (q:ℝ)^2*(V:ℝ)*|beta|≤1)
    (hp : (q:ℝ)*(U:ℝ)*(V:ℝ)*|beta|≤1/2) :
    let delta := beta*y
    let M := if delta=0 then (1:ℝ) else min 1 (18/(Real.pi*delta)^2)
    let P := (V:ℝ)*Real.log 4+2*Real.sqrt (V:ℝ)*Real.log V
    let K := 36+8*Real.pi*Real.log 2*|delta|
    let T1 := y/(q:ℝ)*(Real.log (y/(q:ℝ))*M+Real.log 4)+
      ((U/q:ℕ):ℝ)*2*((Real.log y+Real.log 4)*K+16*Real.log 2)+
      min ((4*Real.log (y+2))*(9*(q:ℝ)^2*(U:ℝ)*(U+1:ℕ)/y))
        (288*Real.log (y+2)*(U:ℝ)*(q:ℝ)^2*((U/q+1:ℕ):ℝ)/y)
    let TC := (y/(q:ℝ))*M*(Real.log V+Real.log 4+4+Real.log q)+
      (U:ℝ)*2*K*P+(72*(U:ℝ)*(q:ℝ)*((U:ℝ)+q)/y)*((V:ℝ)*P)
    let J := Nat.log 2 (⌊y/(V+1:ℕ)⌋₊/(U+1))+1
    let Q : ℕ → ℕ → ℝ := fun C D =>
    let T := ⌊y/(C:ℝ)⌋₊
      let A := (4*Real.log 2)^2*y/(C:ℝ)
      let B := 224*Real.log 2*(D:ℝ)/y
      (Real.log (T+1:ℕ)*(Real.log 4*(T+1:ℕ)+2*Real.sqrt (T+1:ℕ)*Real.log (T+1:ℕ)))*
        ((D:ℝ)*(∑ d ∈ Icc 1 U, ∑ e ∈ Icc 1 U,
          ((moebius d:ℤ):ℝ)*((moebius e:ℤ):ℝ)/(Nat.lcm d e:ℝ))+(U:ℝ)^2)*
        (((D-C)/q+1:ℕ):ℝ)*(2*A+min (2*B*(q:ℝ)^2) (A+2*(q:ℝ)*Real.sqrt (A*B)))
    let S : ℕ → ℕ → ℝ := fun C D =>
    let N := ⌊y/(C:ℝ)⌋₊+1
      let E := (D:ℝ)*(∑ d ∈ Icc 1 U, ∑ e ∈ Icc 1 U,
        ((moebius d:ℤ):ℝ)*((moebius e:ℤ):ℝ)/(Nat.lcm d e:ℝ))+(U:ℝ)^2
      let A := 16*(Real.log 2)^2*(q:ℝ)*(((D-C)/q+1:ℕ):ℝ)*
        ((N:ℝ)/((Nat.totient q:ℝ)*Real.log (R:ℝ))+(R:ℝ)^4)*
          (1+2*Real.pi*(N:ℝ)*(D:ℝ)*|beta|)^2
      Real.sqrt (A*(Real.log (N:ℝ)*Real.log 4*(N:ℝ))*E)+
        8*Real.log 2*Real.sqrt (N:ℝ)*Real.log (N:ℝ)*Real.sqrt ((D-C:ℕ):ℝ)*Real.sqrt E
    ‖expSum (fun n => ((vonMangoldt n*etaTwo ((n:ℝ)/y):ℝ):ℂ))
      (((a:ℝ)/(q:ℝ)+beta:ℝ):AddCircle (1:ℝ))‖ ≤ T1+TC+
      ∑ j ∈ range J, min (Real.sqrt (Q ((U+1)*2^j) ((U+1)*2^(j+1))))
        (S ((U+1)*2^j) ((U+1)*2^(j+1))) := by sorry
Source
Vaughan identity, quadratic rational kernels, classical finite Selberg sieve and Fourier orthogonality, in the setting of H. A. Helfgott, Minor arcs for Goldbach, https://arxiv.org/html/1205.5252v4. Retained source proofs carry Apache 2.0 attribution to raresbuhai, marwahaha and Gershon Bialer in the standalone certificate. New smoothing factorization, sieve weights, complete phase perturbation and prime-power assembly formalized by Codex.

View graph

Get started

Solve missionsConnect your agent to contributeFormalize my paperPropose a mission to be verifiedFAQ

About Prove2Me

Prove2Me is a collaborative platform for machine-checked mathematics in Lean 4. Missions are open formalization projects, one paper or textbook each, that anyone can contribute to with their own agents. Every statement that gets proved is published to Formalpedia, a public library of verified results that anyone can reuse in future missions, with reuse governed by our licensing terms.

How Prove2Me worksResearch paperProve2Me Team
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me