betaFeedback

WorksMathematics

The exceptional set of Goldbach numbers among polynomial values

Abstract

Let ϕϕ be a fixed polynomial with integer coefficients, positive leading coefficient and degree k≥2k\geq2. We prove that the values 2ϕ(n)2ϕ(n), 1≤n≤N1\leq n\leq N, can be written as the sum of two primes, apart from at most Nαk+o(1)N^{α_k+o(1)} exceptional indices, where αk<1α_k<1. For quadratic polynomials the exponent is 1/21/2, for cubic polynomials 5/85/8, for quartic polynomials 3/43/4, and for k>4k>4 it is 1−1/(k+1)1-1/(k+1). The proof combines a fourth-moment estimate over primes with mean-value estimates over arbitrary subsets of polynomial sequences.

Introduction

Let P\mathcal{P} denote the set of primes. Althöfer and Pintz proved very recently in a joint work [PA2026] that the exceptional set in Goldbach’s problem,

E(X):=#{n∈N:n≤X, 2∣n, n≠p1+p2 for all p1,p2∈P},(1)E(X) := \#\{n \in\mathbb{N} : n \le X,\ 2 \mid n,\ n \ne p_1 + p_2\text{ for all }p_1,p_2 \in\mathcal{P}\}, \tag*{(1)}

satisfies

E(X)≪εX7/12+ε(2)E(X) \ll_{\varepsilon} X^{7/12+\varepsilon} \tag*{(2)}

(ε>0\varepsilon> 0).

2010 Mathematics Subject Classification: Primary 11P32, Secondary 11P55.

Keywords and phrases: Goldbach problem. Exceptional Goldbach numbers among polynomial values.

This was improved a little later by H. Li and Z. Liu [4] to

E(X)≪εX1/2+ε.(3)E(X) \ll_{\varepsilon} X^{1/2+\varepsilon}. \tag*{(3)}

The analytic arguments in these works use a fixed zero-free half-plane for Dirichlet LL-functions. The result of OpenAI quoted as Theorem A below supplies the required hypothesis. We use these external results in the precise forms stated in § 2. Under the Generalized Riemann Hypothesis (GRH) for Dirichlet LL-functions, the classical result of Hardy and Littlewood [HL1924] already gives the exponent in (3).

Throughout the paper, ϕ∈Z[x]\phi\in\mathbb{Z}[x] is fixed, has degree k≥2k \ge2 and has positive leading coefficient. Put

Ek,ϕ(N):={n∈N:n≤N, 2ϕ(n)≠p1+p2 for all p1,p2∈P},Ek,ϕ(N):=#Ek,ϕ(N).(4)\begin{aligned} \mathcal{E}_{k,\phi}(N) &:= \{n \in\mathbb{N}: n \le N,\ 2\phi(n) \ne p_1+p_2\text{ for all }p_1,p_2 \in\mathcal{P}\},\\ E_{k,\phi}(N) &:= \#\mathcal{E}_{k,\phi}(N). \tag*{(4)} \end{aligned}

The finitely many indices for which ϕ(n)≤0\phi(n) \le0 are included in this definition and cause no difficulty. Since the values have size Oϕ(Nk)O_{\phi}(N^k), a direct application of (3) gives no nontrivial power bound for (4) when k≥2k \ge2. Perelli proved in 1996 [Per1996] that, for every $A > 0,

Ek,ϕ(N)≪ϕ,AN(log⁡N)−A.(5)E_{k,\phi}(N) \ll_{\phi,A} N(\log N)^{-A}. \tag*{(5)}

This was improved by Brüdern, Kawada and Wooley [1] to

Ek,ϕ(N)≪ϕN1−c0/k,(6)E_{k,\phi}(N) \ll_{\phi} N^{1-c_0/k}, \tag*{(6)}

with a small absolute constant c0>0c_0 > 0. Our goal is to show a stronger explicit estimate for every k≥2k \ge2.

Theorem 1. Let ϕ∈Z[x]\phi\in\mathbb{Z}[x] be a fixed polynomial of degree k≥2k \ge2 with positive leading coefficient. For every ε>0\varepsilon> 0 and N≥2N \ge2,

Ek,ϕ(N)≪ϕ,εNαk+ε,E_{k,\phi}(N) \ll_{\phi,\varepsilon} N^{\alpha_k+\varepsilon},

where α2=1/2\alpha_2 = 1/2, α3=5/8\alpha_3 = 5/8, α4=3/4\alpha_4 = 3/4, and αk=1−1/(k+1)\alpha_k = 1 - 1/(k+1) for k>4k > 4.

We use the real smooth weight ww specified in § 2 of [4], chosen in logarithmic coordinates from the Gevrey class of order 2. In particular, w∈Cc∞((0,1))w \in C_c^\infty((0,1)), 0≤w≤10 \le w \le1, and w(u)=1w(u)=1 for 1/8≤u≤7/81/8 \le u \le7/8. We use the circle method with

Sw(α):=∑p≤Xw(p/X)log⁡p e(pα),e(u)=e2πiu,L=log⁡X,(7)S_w(\alpha) := \sum_{p \le X} w(p/X)\log p\, e(p\alpha), \qquad e(u)=e^{2\pi i u}, \qquad L=\log X, \tag*{(7)}

and with polynomial sums of the form

fk(α):=∑1≤n≤Ne(2ϕ(n)α).(8)f_k(\alpha) := \sum_{1 \le n \le N} e(2\phi(n)\alpha). \tag*{(8)}

Here NN measures indices and XX measures polynomial values. In the proof we work with subsets for which X/2<2ϕ(n)≤XX/2 < 2\phi(n) \le X; on an index interval (N/2,N](N/2,N] all the required scales satisfy X≍ϕNkX \asymp_\phi N^k.

Write T=R/Z\mathbb{T}=\mathbb{R}/\mathbb{Z} and define the major and minor arcs on this circle by

M(P)=⋃q≤P ⋃amodq(a,q)=1{α∈T:∥α−aq∥≤PqX},m(P)=T∖M(P).(9)\mathfrak{M}(P)=\bigcup_{q\le P}\ \bigcup_{\substack{a \bmod q\\(a,q)=1}}\left\{\alpha\in\mathbb{T}:\left\lVert\alpha-\frac{a}{q}\right\rVert\le\frac{P}{qX}\right\}, \qquad \mathfrak{m}(P)=\mathbb{T}\setminus\mathfrak{M}(P). \tag*{(9)}

The norm denotes distance to the nearest integer. We choose P=XκP=X^\kappa with a fixed κ<1/2\kappa<1/2, so that the individual major arcs are disjoint for sufficiently large XX. We usually omit PP from the notation.

For even m∈[X/2,X]m\in[X/2,X], our goal is to obtain

R(m):=∑p+p′=mp,p′≤Xw(p/X)w(p′/X)log⁡plog⁡p′>0(10)R(m):=\sum_{\substack{p+p'=m\\p,p'\le X}}w(p/X)w(p'/X)\log p\log p'>0 \tag*{(10)}

when m=2ϕ(n)m=2\phi(n), apart from a small exceptional set of indices. Put

R1(m)=∫MSw(α)2e(−mα) dα,R2(m)=∫mSw(α)2e(−mα) dα.(11)\begin{aligned} R_1(m)&=\int_{\mathfrak{M}}S_w(\alpha)^2e(-m\alpha)\,d\alpha,\\ R_2(m)&=\int_{\mathfrak{m}}S_w(\alpha)^2e(-m\alpha)\,d\alpha. \tag*{(11)} \end{aligned}

Orthogonality gives

R(m)=∫01Sw(α)2e(−mα) dα=R1(m)+R2(m).(12)R(m)=\int_0^1S_w(\alpha)^2e(-m\alpha)\,d\alpha=R_1(m)+R_2(m). \tag*{(12)}

For the simple weight w=1w=1 on [0,1][0,1], the classical goal is an asymptotic formula of the type

R1∗(m)=(1+o(1))S(m)m,(13)R_1^*(m)=(1+o(1))\mathcal{S}(m)m, \tag*{(13)}

where, for even positive mm,

S(m)=2C0∏p∣mp>2(1+1p−2),C0=∏p>2(1−1(p−1)2)=0.66016….(14)\mathcal{S}(m)=2C_0\prod_{\substack{p\mid m\\p>2}}\left(1+\frac{1}{p-2}\right),\qquad C_0=\prod_{p>2}\left(1-\frac{1}{(p-1)^2}\right)=0.66016\ldots. \tag*{(14)}

In the present weighted setting the corresponding goal is

R1(m)=(1+o(1))S(m)Iw(m),(15)R_1(m)=(1+o(1))\mathcal{S}(m)I_w(m), \tag*{(15)}

where

Iw(m):=∑r∈Zw(r/X)w((m−r)/X)≥m−X/4−1≫X.(16)I_w(m):=\sum_{r\in\mathbb{Z}}w(r/X)w((m-r)/X)\ge m-X/4-1\gg X. \tag*{(16)}

Indeed, both weights are one for integers r∈[X/8,m−X/8]r\in[X/8,m-X/8]. For our purposes it is sufficient to show, uniformly on the indicated range of even mm, that

R1(m)≥cS(m)m(17)R_1(m)\ge c\mathcal{S}(m)m \tag*{(17)}

with a fixed c>0c>0.

Auxiliary results for exponential sums over primes

Let θ\theta denote the supremum of the real parts of the nontrivial zeros of all primitive Dirichlet LL-functions, including the Riemann zeta function. We use the following result of OpenAI [5 OAI2026b].

Theorem A. θ≤7/8\theta\le7/8.

For the weight fixed in § 1 and the arcs (9), Li and Liu [4], Theorem 1 and Proposition 4.1, give the following two results.

Theorem B. If θ<11/12\theta< 11/12, there is a fixed κ=κ(θ)∈(4/9,9/20)\kappa= \kappa(\theta) \in(4/9, 9/20) such that, with P=XκP = X^\kappa,

∫m∣Sw(α)∣4 dα≪θ,wX5/2(log⁡X)100.(18)\int_{\mathfrak{m}} |S_w(\alpha)|^4\,d\alpha\ll_{\theta,w} X^{5/2}(\log X)^{100}. \tag*{(18)}

Theorem C. If θ<1\theta< 1, then for every fixed κ∈(0,9/20)\kappa\in(0, 9/20) there is c>0c > 0 such that, with P=XκP = X^\kappa, uniformly for even m∈[X/2,X]m \in[X/2, X],

R1(m)=∫mSw(α)2e(−mα) dα=S(m)Iw(m)+Ow,θ,κ(X1−c).(19)R_1(m) = \int_{\mathfrak{m}} S_w(\alpha)^2 e(-m\alpha)\,d\alpha= \mathfrak{S}(m)I_w(m) + O_{w,\theta,\kappa}(X^{1-c}). \tag*{(19)}

Remark. Pintz’s explicit formula [Pin2023] expresses the major-arc contribution in terms of the pole at one and possible zeros near the line Re⁡s=1\operatorname{Re} s = 1. A fixed zero-free half-plane excludes the latter zeros when XX is sufficiently large. For the precise range κ<9/20\kappa< 9/20 needed here we use Theorem C. We choose the same κ\kappa in Theorems B and C.

Since S(m)≥2C0>0\mathfrak{S}(m) \ge2C_0 > 0, Theorem C and (16) imply (17) for sufficiently large XX. The integral R1(m)R_1(m) is real: the arcs are invariant under α↦−α\alpha\mapsto-\alpha on T\mathbb{T}, and Sw(−α)=Sw(α)‾S_w(-\alpha) = \overline{S_w(\alpha)}. The proof below only requires the hypothesis θ<11/12\theta< 11/12; Theorem A supplies it.

Auxiliary results for exponential sums over polynomial values

To estimate the cardinality of the exceptional set, we use sums over its subsets. For A⊆{1,…,M}A \subseteq\{1,\ldots,M\} put

KA(α):=∑m∈Ae(2ϕ(m)α),Z=#A.(20)K_A(\alpha) := \sum_{m\in A} e(2\phi(m)\alpha), \qquad Z = \# A. \tag*{(20)}

We require moment estimates that are uniform in the choice of AA. The first follows from Wooley’s weighted form of Vinogradov’s mean-value theorem [Woo2019], Corollary 1.4.

Theorem D. Let k≥1k \ge1 and s=k(k+1)/2s = k(k+1)/2. For every sequence b1,…,bMb_1,\ldots,b_M of complex numbers and every ε>0\varepsilon> 0,

∫Tk∣∑m=1Mbme(β1m+β2m2+⋯+βkmk)∣2s dβ≪k,εMε(∑m=1M∣bm∣2)s.(21)\int_{\mathbb{T}^k} \left|\sum_{m=1}^{M} b_m e(\beta_1m+\beta_2m^2+\cdots+\beta_km^k)\right|^{2s}\,d\boldsymbol{\beta} \ll_{k,\varepsilon} M^\varepsilon\left(\sum_{m=1}^{M}|b_m|^2\right)^s. \tag*{(21)}

We use the following consequence.

Corollary E. Let ψ∈Z[x]\psi\in\mathbb{Z}[x] have degree k≥1k \ge1 and nonzero leading coefficient. Put

q=k(k+1),s=q/2,D=k(k−1)/2.q = k(k+1), \qquad s = q/2, \qquad D = k(k-1)/2.

For every A⊆{1,…,M}\mathcal{A} \subseteq\{1,\ldots,M\}, Z=#AZ = \#\mathcal{A}, and every ε>0\varepsilon> 0,

∫T∣∑m∈Ae(αψ(m))∣q dα≪k,εMD+εZs.(22)\int_{\mathbb{T}} \left|\sum_{m\in\mathcal{A}} e(\alpha\psi(m))\right|^q\,d\alpha\ll_{k,\varepsilon} M^{D+\varepsilon}Z^s. \tag*{(22)}

The implied constant is independent of the coefficients of ψ\psi.

Proof. Let

F(β)=∑m∈Ae(β1m+⋯+βkmk),J=∫Tk∣F(β)∣2s dβ.(23)F(\boldsymbol{\beta}) = \sum_{m\in\mathcal{A}} e(\beta_1m+\cdots+\beta_km^k), \qquad J = \int_{\mathbb{T}^k}|F(\boldsymbol{\beta})|^{2s}\,d\boldsymbol{\beta}. \tag*{(23)}

By Theorem D, J≪k,εMεZsJ \ll_{k,\varepsilon} M^\varepsilon Z^s. For h=(h1,…,hk)∈Zk\mathbf{h}=(h_1,\ldots,h_k)\in\mathbb{Z}^k, let N(h)N(\mathbf{h}) count the ordered pairs (x,y)∈As×As(\mathbf{x},\mathbf{y})\in\mathcal{A}^s\times\mathcal{A}^s such that

∑i=1s(xij−yij)=hj(1≤j≤k).\sum_{i=1}^{s}(x_i^j-y_i^j)=h_j \qquad(1\le j\le k).

Orthogonality gives

N(h)=∫Tk∣F(β)∣2se(−β⋅h) dβ,0≤N(h)≤J.N(\mathbf{h})=\int_{\mathbb{T}^k}|F(\boldsymbol{\beta})|^{2s}e(-\boldsymbol{\beta}\cdot\mathbf{h})\,d\boldsymbol{\beta}, \qquad0\le N(\mathbf{h})\le J.

Write ψ(m)=akmk+⋯+a0\psi(m)=a_km^k+\cdots+a_0. Since there are equally many xix_i and yiy_i, the constant coefficient cancels. A second application of orthogonality yields

∫T∣∑m∈Ae(αψ(m))∣2s dα=∑h∈Zka1h1+⋯+akhk=0N(h).(24)\int_{\mathbb{T}}\left|\sum_{m\in\mathcal{A}}e(\alpha\psi(m))\right|^{2s}\,d\alpha = \sum_{\substack{\mathbf{h}\in\mathbb{Z}^k\\a_1h_1+\cdots+a_kh_k=0}}N(\mathbf{h}). \tag*{(24)}

A nonzero summand has ∣hj∣≤sMj|h_j|\le sM^j. Once h1,…,hk−1h_1,\ldots,h_{k-1} are fixed, the relation in (24) determines at most one integer hkh_k, because ak≠0a_k\ne0. Thus there are at most

∏j=1k−1(2sMj+1)≪kM1+⋯+(k−1)=MD\prod_{j=1}^{k-1}(2sM^j+1)\ll_k M^{1+\cdots+(k-1)}=M^D

potentially nonzero summands. For k=1k=1 the product is empty and equals one. Bounding each summand by JJ proves (3.3).

For k=3k=3 and k=4k=4 we use the following result of Hughes and Wooley [3], Theorem 4.1, equation (4.2). It is valid for every degree at least three.

Theorem F. If ψ∈Z[x]\psi\in\mathbb{Z}[x] has degree at least three, A⊆{1,…,M}\mathcal{A} \subseteq\{1,\ldots,M\} and Z=#AZ=\#\mathcal{A}, then

∫01∣∑m∈Ae(αψ(m))∣8 dα≪ψ,εM1+εZ4.(25)\int_0^1 \left|\sum_{m\in\mathcal{A}} e(\alpha\psi(m))\right|^8 \,\mathrm{d}\alpha\ll_{\psi,\varepsilon} M^{1+\varepsilon}Z^4. \tag*{(25)}

For degree two an elementary divisor argument suffices.

Lemma G. If ψ∈Z[x]\psi\in\mathbb{Z}[x] has degree two, A⊆{1,…,M}\mathcal{A} \subseteq\{1,\ldots,M\} and Z=#AZ=\#\mathcal{A}, then

∫01∣∑m∈Ae(αψ(m))∣4 dα≪ψ,εMεZ2.(26)\int_0^1 \left|\sum_{m\in\mathcal{A}} e(\alpha\psi(m))\right|^4 \,\mathrm{d}\alpha\ll_{\psi,\varepsilon} M^\varepsilon Z^2. \tag*{(26)}

Proof. Write ψ(m)=am2+bm+c\psi(m)=am^2+bm+c, with a≠0a\ne0, and put

r(t)=#{(x,y)∈A2:ψ(x)−ψ(y)=t}.r(t)=\#\{(x,y)\in\mathcal{A}^2:\psi(x)-\psi(y)=t\}.

Orthogonality gives

∫01∣∑m∈Ae(αψ(m))∣4 dα=∑t∈Zr(t)2,∑tr(t)=Z2.(27)\int_0^1 \left|\sum_{m\in\mathcal{A}} e(\alpha\psi(m))\right|^4 \,\mathrm{d}\alpha=\sum_{t\in\mathbb{Z}}r(t)^2,\qquad\sum_t r(t)=Z^2. \tag*{(27)}

For t≠0t\ne0 the factorization

(x−y)(a(x+y)+b)=t(28)(x-y)(a(x+y)+b)=t \tag*{(28)}

gives at most 2d(∣t∣)2d(|t|) choices for the signed divisor z=x−yz=x-y. Once zz is fixed, the other factor is fixed, and

a(2x−z)+b=t/z(29)a(2x-z)+b=t/z \tag*{(29)}

determines at most one xx. Since ∣t∣≪ψM2|t|\ll_{\psi} M^2, the divisor bound, with a suitably smaller exponent, gives r(t)≪ψ,εMεr(t)\ll_{\psi,\varepsilon} M^\varepsilon for t≠0t\ne0. For t=0t=0, either x=yx=y or a(x+y)+b=0a(x+y)+b=0, whence r(0)≤2Zr(0)\le2Z. Consequently,

∑tr(t)2≤4Z2+Oψ,ε(Mε)∑t≠0r(t)≪ψ,εMεZ2.\sum_t r(t)^2\le4Z^2+O_{\psi,\varepsilon}(M^\varepsilon)\sum_{t\ne0}r(t)\ll_{\psi,\varepsilon} M^\varepsilon Z^2.

Remark. The diagonal solutions give a lower bound Z2Z^2 for the moment in (3.7). Thus its dependence on ZZ is sharp, apart from the factor MεM^\varepsilon.

Proof of Theorem 1

Fix ϕ\phi and let NN be sufficiently large in terms of its coefficients. For indices N/2<n≤NN/2<n\le N, positivity of the leading coefficient gives

c1Nk≤2ϕ(n)≤c2Nkc_1N^k\le2\phi(n)\le c_2N^k

with constants c1,c2>0c_1,c_2>0 depending only on ϕ\phi. This value interval is covered by Oϕ(1)O_\phi(1) consecutive dyadic intervals (X/2,X](X/2,X], all with X≍ϕNkX\asymp_\phi N^k. For one such interval put

A={n∈Ek,ϕ(N):N/2<n≤N, X/2<2ϕ(n)≤X},Z=#A.\mathcal{A}=\{n\in\mathcal{E}_{k,\phi}(N):N/2<n\le N,\ X/2<2\phi(n)\le X\},\qquad Z=\#\mathcal{A}.

At this scale choose P=XκP=X^\kappa with the fixed κ\kappa of Theorem B. Theorem C applies to the same arcs. For every n∈An\in\mathcal{A}, R(2ϕ(n))=0R(2\phi(n))=0, and therefore

R2(2ϕ(n))=−R1(2ϕ(n))=−(1+o(1))S(2ϕ(n))Iw(2ϕ(n))≤−cX(30)R_2(2\phi(n))=-R_1(2\phi(n))=-(1+o(1))\mathfrak{S}(2\phi(n))I_w(2\phi(n))\le-cX \tag*{(30)}

for a fixed c>0c>0, uniformly in nn. Define

K(α)=∑n∈Ae(−2ϕ(n)α).(31)K(\alpha)=\sum_{n\in\mathcal{A}}e(-2\phi(n)\alpha). \tag*{(31)}

Summing (4.1), interchanging the finite sum and the integral, and taking absolute values gives

ZX≪∣∫mSw(α)2K(α) dα∣≤∫m∣Sw(α)∣2∣K(α)∣ dα.(32)ZX\ll\left|\int_{\mathfrak{m}}S_w(\alpha)^2K(\alpha)\,d\alpha\right|\le\int_{\mathfrak{m}}|S_w(\alpha)|^2|K(\alpha)|\,d\alpha. \tag*{(32)}

Suppose that, for some q≥4q\ge4 and b≥0b\ge0, we have

(∫01∣K(α)∣q dα)1/q≪ϕ,εNb/q+εZ1/2.(33)\left(\int_0^1|K(\alpha)|^q\,d\alpha\right)^{1/q}\ll_{\phi,\varepsilon}N^{b/q+\varepsilon}Z^{1/2}. \tag*{(33)}

All results of § 3 apply with ψ=−2ϕ\psi=-2\phi. Lemma G gives q=4q=4, b=0b=0 for k=2k=2. Theorem F gives q=8q=8, b=1b=1 for k=3,4k=3,4. Corollary E gives q=k(k+1)q=k(k+1), b=k(k−1)/2=q/2−kb=k(k-1)/2=q/2-k in general. We write

∣Sw(α)∣2=(∣Sw(α)∣4)1/q(∣Sw(α)∣2)1−2/q(34)|S_w(\alpha)|^2=(|S_w(\alpha)|^4)^{1/q}(|S_w(\alpha)|^2)^{1-2/q} \tag*{(34)}

and apply Hölder’s inequality to these two factors and ∣K∣|K|, with exponents qq, q/(q−2)q/(q-2) and qq, respectively. Parseval’s identity and the elementary estimate ∑p≤Xlog⁡p≪X\sum_{p\le X}\log p \ll X give

∫01∣Sw(α)∣2 dα=∑p≤Xw(p/X)2(log⁡p)2≪XL.\int_0^1 |S_w(\alpha)|^2\,\mathrm{d}\alpha=\sum_{p\le X}w(p/X)^2(\log p)^2\ll XL.

Together with Theorem B and (33), this yields

ZX≪(∫01∣Sw∣4)1/q(∫01∣Sw∣2)1−2/q(∫01∣K∣q)1/q≪ϕ,εX1+1/(2q)L1+98/qNb/q+εZ1/2.(35)\begin{aligned} ZX&\ll\left(\int_0^1|S_w|^4\right)^{1/q}\left(\int_0^1|S_w|^2\right)^{1-2/q}\left(\int_0^1|K|^q\right)^{1/q}\\ &\ll_{\phi,\varepsilon}X^{1+1/(2q)}L^{1+98/q}N^{b/q+\varepsilon}Z^{1/2}. \tag*{(35)} \end{aligned}

If Z=0Z=0 there is nothing to prove. Otherwise divide by XZ1/2XZ^{1/2} and square. Since X≍ϕNkX\asymp_\phi N^k, and since ε\varepsilon may be chosen arbitrarily small, we obtain

Z≪ϕ,εN(k+2b)/q+ε.(36)Z\ll_{\phi,\varepsilon}N^{(k+2b)/q+\varepsilon}. \tag*{(36)}

Thus the quantity to minimize among the available moment estimates is (k+2b)/q(k+2b)/q.

For k=2k=2, the choices q=4q=4, b=0b=0 give α2=1/2\alpha_2=1/2. For k=3,4k=3,4, the choices q=8q=8, b=1b=1 give αk=(k+2)/8\alpha_k=(k+2)/8. For k>4k>4, the choices q=k(k+1)q=k(k+1), 2b=q−2k2b=q-2k give αk=1−k/q=1−1/(k+1)\alpha_k=1-k/q=1-1/(k+1). Summing over the Oϕ(1)O_\phi(1) value intervals proves the required bound for the exceptional indices in (N/2,N](N/2,N]. Summing next over the index intervals (N/2j+1,N/2j](N/2^{j+1},N/2^j], and absorbing the finitely many initial indices into the implied constant, proves Theorem 1. □

Application to other thin sequences

Theorem 1 gives related thin sequences containing only finitely many Goldbach exceptional numbers. For example, consider the values 2mk2m^k with m=⌊a(n+β)r⌋m=\lfloor a(n+\beta)^r\rfloor, where a>0a>0 is fixed. For almost every β∈[0,1]\beta\in[0,1] there are only finitely many exceptions when r>1/(1−αk)r>1/(1-\alpha_k). Thus it suffices to take r>2r>2 for k=2k=2, r>8/3r>8/3 for k=3k=3, r>4r>4 for k=4k=4, and r>k+1r>k+1 for k>4k>4. No parity restriction on mm is needed.

Put

Bk={m∈N:2mk≠p+q for all primes p,q}.B_k=\{m\in\mathbb{N}:2m^k\ne p+q\text{ for all primes }p,q\}.

Assume, for a fixed 0≤ck≤10 \le c_k \le1, that for every ε>0\varepsilon> 0

#(Bk∩[1,Y])≪k,εYck+ε(Y≥2).(37)\#(B_k \cap[1,Y]) \ll_{k,\varepsilon} Y^{c_k+\varepsilon} \qquad(Y \ge2). \tag*{(37)}

Proposition. Fix a>0a > 0 and r>1/(1−ck)r > 1/(1-c_k), with 0≤ck≤10 \le c_k \le1 as in (37). For almost every β∈[0,1]\beta\in[0,1], every sufficiently large nn satisfies

2⌊a(n+β)r⌋k=pn+qn(pn,qn prime).(38)2\lfloor a(n+\beta)^r \rfloor^k = p_n+q_n \qquad(p_n,q_n\text{ prime}). \tag*{(38)}

Moreover, for Aa,r,β={2⌊a(n+β)r⌋k:n≥1}A_{a,r,\beta}=\{2\lfloor a(n+\beta)^r\rfloor^k:n\ge1\}, the counting asymptotic

#(Aa,r,β∩[1,X])∼(X2ak)1/(kr)\#(A_{a,r,\beta}\cap[1,X])\sim\left(\frac{X}{2a^k}\right)^{1/(kr)}

holds for every fixed β∈[0,1]\beta\in[0,1].

Proof. Put mn(β)=⌊a(n+β)r⌋m_n(\beta)=\lfloor a(n+\beta)^r\rfloor and Bk,n={β∈[0,1]:mn(β)∈Bk}B_{k,n}=\{\beta\in[0,1]:m_n(\beta)\in B_k\}. The assumption on rr implies r>1r>1. For T≤n<2TT\le n<2T, the derivative of a(n+β)ra(n+\beta)^r with respect to β\beta is ≍a,rTr−1\asymp_{a,r}T^{r-1}. Hence, for every integer bb,

meas⁡{β∈[0,1]:mn(β)=b}≪a,rT1−r.(39)\operatorname{meas}\{\beta\in[0,1]:m_n(\beta)=b\}\ll_{a,r}T^{1-r}. \tag*{(39)}

The intervals [anr,a(n+1)r][an^r,a(n+1)^r] have disjoint interiors. Their lengths tend to infinity since r>1r>1. For sufficiently large TT, each unit interval [b,b+1)[b,b+1) therefore meets at most two of them, apart from irrelevant endpoints. It follows from (37) and (39) that

∑T≤n<2Tmeas⁡(Bk,n)≪a,rT1−r#(Bk∩[1,Ca,rTr])\sum_{T\le n<2T}\operatorname{meas}(B_{k,n})\ll_{a,r}T^{1-r}\#(B_k\cap[1,C_{a,r}T^r])
≪a,r,k,εT1−r+r(ck+ε)=T1−r(1−ck−ε).(40)\ll_{a,r,k,\varepsilon}T^{1-r+r(c_k+\varepsilon)} =T^{1-r(1-c_k-\varepsilon)}. \tag*{(40)}

Choose ε>0\varepsilon>0 so that r(1−ck−ε)>1r(1-c_k-\varepsilon)>1. Summation over T=2jT=2^j gives ∑nmeas⁡(Bk,n)<∞\sum_n\operatorname{meas}(B_{k,n})<\infty. The first Borel–Cantelli lemma proves (38); no independence assumption is required. Also mn(β)≥1m_n(\beta)\ge1 for all sufficiently large nn, so that the omission of 00 from BkB_k is harmless. Finally, mn(β)∼anrm_n(\beta)\sim an^r and the values are eventually strictly increasing. Inverting 2mn(β)k∼2aknkr2m_n(\beta)^k\sim2a^kn^{kr} proves the counting asymptotic.

Remark. The measure argument does not provide an explicit admissible β\beta. For example, can one exhibit a particular β∈[0,1]\beta\in[0,1] for which

2⌊(n+β)3⌋2(41)2\left\lfloor(n+\beta)^3\right\rfloor^2 \tag*{(41)}

is the sum of two primes for every sufficiently large nn? § 5 proves this for almost every β\beta, by taking k=2k=2, a=1a=1 and r=3r=3. The particularly natural choice β=0\beta=0 would assert that every sufficiently large number 2n62n^6 is Goldbach, a substantially stronger conclusion than the exceptional-set estimates proved in this paper. We do not know any triple (a,k,r)(a,k,r) with a>0a>0, integer k>0k>0, r>0r>0 for which we could specify a value β∈(0,1]\beta\in(0,1] such that the expression (38) would be a sum of two primes for every value of nn.

AI assistance. OpenAI Astra was used for careful auditing. Its checks concerned the proof, references and notation. It also assisted with the revision and with formulating the above remark.

Ingo Althöfer Faculty of Mathematics and Computer Science Friedrich-Schiller University Jena, D-07740 Jena, Germany E-mail: [email protected]

Dijia Chen Department of Mathematics, University of Wisconsin–Madison Madison, WI 53706, USA E-mail: [email protected]

Alberto Perelli Dipartimento di Matematica, Università di Genova Via Dodecaneso 35, 16146 Genova, Italy E-mail: [email protected]

János Pintz HUN-REN Alfréd Rényi Institute of Mathematics Réáltanoda u. 13–15, H-1053 Budapest, Hungary E-mail: [email protected]

References

  1. [1]J. Brüder[n], K. Kawada and T. D. Wooley, Additive representation in thin sequences, II: The binary Goldbach problem, Mathematika 47 (2000), 117–125.
  2. [3]K. Hughes and T. D. Wooley, Discrete restriction for (x, x³) and related topics, International Mathematics Research Notices 2022, no. 20, 15612–15631. doi:10.1093/imrn/rnab113. Theorem 4.1 and Conjecture 4.2.DOI
  3. [4]H. Li and Z. Liu, Fourth moments of prime exponential sums and exceptional sets for sums and differences of primes, arXiv:2610.26756v1, 19 October 2026. Theorem 1, Proposition 4.1 and Section 2.arxiv.org/abs/2610.26756
  4. [5]OpenAI, The Quasi-Riemann Hypothesis: A Zero-Free Half-Plane Re s > 7/8, OpenAI Math Release preprint, 30 September 2026, family 003, Theorem 1.1. Archived repository version at commit fd4aeeb2ee4f.

Paper details

vlnc:W_xkptvma5
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 10

downloads · 1

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents