betaFeedback

WorksMathematics

Extreme values of derivatives of Dedekind zeta functions on the 11-line

Abstract

In this paper, we establish lower bounds for extreme values of derivatives of Dedekind zeta functions on the 11-line over arbitrary fixed number fields. Our result extends Yang's corresponding estimate for the Riemann zeta function in 2024, with the leading constant involving residues of Dedekind zeta functions and a moment of the Dickman function. A key ingredient is a uniform asymptotic formula for weighted friable sums involving the coefficients of Dedekind zeta functions.

Introduction

The study of extreme values of the Riemann zeta function ζ(s)\zeta(s) has a long history. The 1-line is the boundary of absolute convergence of its Dirichlet series and Euler product, and larger values on this line reflect the combined contributions of primes. Understanding these values also provides a model for studying the distribution of values of more general zeta and LL-functions.

Granville and Soundararajan [12] proved that there exist arbitrarily large tt such that

∣ζ(1+it)∣≥eγ(log⁡2t+log⁡3t−log⁡4t−C),|\zeta(1+it)| \ge e^\gamma(\log_{2}t+\log_{3}t-\log_{4}t-C),

where C>0C>0 is an absolute constant and γ\gamma denotes the Euler-Mascheroni constant. Here and throughout this paper, we write log⁡j\log_j for the jj-th iterated logarithm. They also conjecture about the maximum of ∣ζ(1+it)∣|\zeta(1+it)|. Subsequently, Aistleitner, Mahatab and Munsch [1] improved the above lower bound by removing the log⁡4t\log_4 t term, up to an absolute constant.

We now turn to Dedekind zeta functions associated with number fields. Let F/Q\mathbb{F}/\mathbb{Q} be a fixed number field of degree dd, and let ζF(s)\zeta_{\mathbb{F}}(s) denote its Dedekind zeta function. We also write

κF=Res⁡s=1 ζF(s).\kappa_{\mathbb{F}}=\underset{s=1}{\operatorname{Res}}\,\zeta_{\mathbb{F}}(s).

In 2021, Dixit and Mahatab [8 Corollary 1.4] established a lower bound for extreme values of Dedekind zeta functions on the 1-line. More precisely, there exists a constant CF>0C_{\mathbb{F}}>0 such that

max⁡t∈[T,T]∣ζF(1+it)∣≥eγκF(log⁡2T+log⁡3T−CF)\max_{t\in[\sqrt{T},T]} \left|\zeta_{\mathbb{F}}(1+it)\right|\geq e^{\gamma\kappa_{\mathbb{F}}(\log_2 T+\log_3 T-C_{\mathbb{F}})}

for all sufficiently large TT. Away from the 1-line, Balakrishnan [2] obtained extreme values of Dedekind zeta functions of quadratic fields in the critical strip. More recently, Bondarenko, Darbar, Hagen, Heap and Seip [3] established lower bounds for cyclotomic fields on the critical line.

Turning to derivatives, Yang [20] established lower bounds for extreme values of derivatives of the Riemann zeta function, which were subsequently improved on the 1-line by Dong and Wei [10]. Let ρ\rho denote the Dickman function, and put

Yj:=∫0∞ujρ(u) du.Y_j:=\int_0^\infty u^j\rho(u)\,du.

More recently, Yang [21 Theorem 1] proved that, as T→∞T\to\infty, uniformly for all positive integers ℓ≤log⁡3T/log⁡4T\ell\leq\log_3 T/\log_4 T, we have

max⁡t∈[T,2T]∣ζ(ℓ)(1+it)∣≥(Yℓ+o(1))(log⁡2T)ℓ+1.\max_{t\in[T,2T]}\left|\zeta^{(\ell)}(1+it)\right|\geq\left(Y_\ell+o(1)\right)(\log_2 T)^{\ell+1}.

For related results concerning derivatives of the Dedekind zeta function of a cyclotomic field on and near the critical line, see [16]. Further results on extreme values of derivatives of the Riemann zeta function and Dirichlet LL-functions can be found in [4, 9, 14, 22].

In the present paper, we extend Yang’s result on the 1-line to Dedekind zeta functions of arbitrary fixed number fields. Our main result is as follows.

Theorem 1.1. Let F/Q\mathbb{F}/\mathbb{Q} be a fixed number field. Then, as T→∞T\to\infty, uniformly for all positive integers ℓ≤(log⁡3T)/(log⁡4T)\ell\leq(\log_3 T)/(\log_4 T), we have

max⁡t∈[T,2T]∣ζF(ℓ)(1+it)∣≥(κFYℓ+oF(1))(log⁡2T)ℓ+1.\max_{t\in[T,2T]}\left|\zeta_{\mathbb{F}}^{(\ell)}(1+it)\right|\geq\left(\kappa_{\mathbb{F}}Y_\ell+o_{\mathbb{F}}(1)\right)(\log_2 T)^{\ell+1}.

To explain the main difficulty in extending Yang’s result, for each positive integer nn, define

aF(n):=#{a⊆OF:Na=n},a_{\mathbb{F}}(n):=\#\{\mathfrak{a}\subseteq\mathcal{O}_{\mathbb{F}}:N\mathfrak{a}=n\},

where OF\mathcal{O}_F denotes the ring of integers of FF and Na⁡\operatorname{Na} is the absolute norm of the integral ideal a\mathfrak{a}. Then, for Re⁡(s)>1\operatorname{Re}(s)>1, we have

ζF(s)=∑n=1∞aF(n)nsand(−1)ℓζF(ℓ)(s)=∑n=1∞aF(n)(log⁡n)ℓns.\zeta_F(s)=\sum_{n=1}^{\infty}\frac{a_F(n)}{n^s}\quad\text{and}\quad(-1)^\ell\zeta_F^{(\ell)}(s)=\sum_{n=1}^{\infty}\frac{a_F(n)(\log n)^\ell}{n^s}.

Although aF(n)a_F(n) is multiplicative, the coefficients aF(n)(log⁡n)ℓa_F(n)(\log n)^\ell are not so far as ℓ≥1\ell\geq1. Consequently, the Euler product argument underlying the long resonance method of Aistleitner, Mahatab and Munsch [1] cannot be directly applied to derivatives. Instead, we employ Soundararajan’s method [18] with a resonator supported on the divisors of a suitably chosen integer. The resulting lower bound is reduced to the estimation of weighted friable sums involving aF(n)a_F(n), for which we establish a uniform asymptotic formula.

Although Theorem 1.1 is stated for positive integers ℓ\ell, the same argument also applies to ℓ=0\ell=0. Since

Y0=∫0∞ρ(u) du=eγ,Y_0=\int_0^\infty\rho(u)\,du=e^\gamma,

we obtain

max⁡t∈[T,2T]∣ζF(1+it)∣≥(eγκF+oF(1))log⁡2T.\max_{t\in[T,2T]}|\zeta_F(1+it)|\geq\left(e^\gamma\kappa_F+o_F(1)\right)\log_2 T.

This agrees with the leading term of the result of Dixit and Mahatab [8], whose lower bound also contains a secondary term of order log⁡3T\log_3 T, albeit over a different range of tt.

The range of ℓ\ell in Theorem 1.1 can be slightly extended. In fact, for any fixed 0<θ<10<\theta<1, the same conclusion holds uniformly for

1≤ℓ≤log⁡3Tlog⁡4T−θlog⁡5T.1\leq\ell\leq\frac{\log_3 T}{\log_4 T-\theta\log_5 T}.

Indeed, the arguments of Lemmas 2.3 and 2.4, together with Stirling’s formula, remain valid in this range. We retain the simpler bound in Theorem 1.1 for convenience.

Theorem 1.1 also yields several consequences for special classes of number fields. When F=QF=\mathbb{Q}, we have κF=1\kappa_F=1 and hence recover [21 Theorem 1]. For a fixed quadratic field F=Q(D)F=\mathbb{Q}(\sqrt{D}) with fundamental discriminant DD, let χD\chi_D denote the corresponding quadratic Dirichlet character. The factorization

ζQ(D)(s)=ζ(s)L(s,χD)\zeta_{\mathbb{Q}(\sqrt{D})}(s)=\zeta(s)L(s,\chi_D)

gives κF=L(1,χD)\kappa_F=L(1,\chi_D). Consequently,

max⁡t∈[T,2T]∣ζQ(D)(ℓ)(1+it)∣≥(L(1,χD)Yℓ+oD(1))(log⁡2T)ℓ+1.\max_{t\in[T,2T]}\left|\zeta_{\mathbb{Q}(\sqrt{D})}^{(\ell)}(1+it)\right|\geq\left(L(1,\chi_D)Y_\ell+o_D(1)\right)(\log_2 T)^{\ell+1}.

The arithmetic meaning of L(1,χD)L(1,\chi_D) is given by the analytic class number formula. Let hDh_D denote the class number of Q(D)\mathbb{Q}(\sqrt{D}). Then,

L(1,χD)={2πhDwD∣D∣,D<0,2hDRDD,D>0,L(1,\chi_D)= \begin{cases} \frac{2\pi h_D}{w_D\sqrt{|D|}}, & D<0,\\[6pt] \frac{2h_D R_D}{\sqrt{D}}, & D>0, \end{cases}

where wDw_D is the number of roots of unity and RDR_D is the regulator; see [5]. Thus, the leading constant in our lower bound reflects the normalized class number in the imaginary quadratic case, and the normalized product of the class number and regulator in the real quadratic case. For the distribution and extreme values of L(1,χD)L(1,\chi_D), see [11]. For instance, when F=Q(i)\mathbb{F}=\mathbb{Q}(i) and F=Q(−3)\mathbb{F}=\mathbb{Q}(\sqrt{-3}), we have

κQ(i)=π4andκQ(−3)=π33,\kappa_{\mathbb{Q}(i)}=\frac{\pi}{4} \qquad\text{and}\qquad \kappa_{\mathbb{Q}(\sqrt{-3})}=\frac{\pi}{3\sqrt{3}},

respectively.

For a cyclotomic field F=Q(ωm)\mathbb{F}=\mathbb{Q}(\omega_m), where ωm\omega_m is a primitive mm-th root of unity and m≥3m\ge3, let χ∗\chi^* denote the primitive Dirichlet character inducing a character χ\chi modulo mm. By the factorization of the Dedekind zeta function, we have

κQ(ωm)=∏χ(modm)χ≠χ0L(1,χ∗).\kappa_{\mathbb{Q}(\omega_m)} = \prod_{\substack{\chi\;(\bmod m)\\ \chi\ne\chi_0}} L(1,\chi^*).

In particular, when m=qm=q is prime, every non-principal character modulo qq is primitive. Thus, Theorem 1.1 gives the corresponding lower bound with an explicit leading constant for cyclotomic fields.

Preliminary lemmas

In this section, we present several lemmas. We begin with a classical estimate due to Landau [15].

Lemma 2.1. Let F/Q\mathbb{F}/\mathbb{Q} be a fixed number field of degree dd, and let AF(x)A_{\mathbb{F}}(x) denote the number of nonzero integral ideals of F\mathbb{F} with norm not exceeding xx. Then for x≥2x\ge2, we have

AF(x):=∑n≤xaF(n)=κFx+OF(x1−δ),A_{\mathbb{F}}(x):=\sum_{n\le x}a_{\mathbb{F}}(n)=\kappa_{\mathbb{F}}x+O_{\mathbb{F}}\left(x^{1-\delta}\right),

where δ=δF=2/(d+1)\delta=\delta_{\mathbb{F}}=2/(d+1).

Proof. This follows from [15], Satz 210. □

The following lemma gives an asymptotic formula, which approximates derivatives of ζF(s)\zeta_{\mathbb{F}}(s) on the 1-line by a truncated sum.

Lemma 2.2. Let F/QF/\mathbb{Q} be a fixed number field of degree dd. When TT is sufficiently large, uniformly for all positive integers ℓ≤(log⁡3T)/(log⁡4T)\ell\le(\log_{3} T)/(\log_{4} T) and T≤t≤2TT \le t \le2T, we have

(−1)ℓζF(ℓ)(1+it)=∑n≤Td+1aF(n)(log⁡n)ℓn1+it+OF(T−1+o(1)).(-1)^{\ell}\zeta_{F}^{(\ell)}(1+it)=\sum_{n\le T^{d+1}}\frac{a_{F}(n)(\log n)^{\ell}}{n^{1+it}}+O_{F}\left(T^{-1+o(1)}\right).

Proof. Let EF(x):=AF(x)−κFxE_{F}(x):=A_{F}(x)-\kappa_{F}x. By Lemma 2.1, we have

EF(x)≪Fx1−δ,(1)E_{F}(x)\ll_{F}x^{1-\delta}, \tag*{(1)}

where δ=2/(d+1)\delta=2/(d+1). For Re⁡(s)>1\operatorname{Re}(s)>1, partial summation gives

ζF(s)=∑n≤XaF(n)ns+κFX1−ss−1−EF(X)Xs+s∫X∞EF(u)us+1 du,(2)\zeta_{F}(s)=\sum_{n\le X}\frac{a_{F}(n)}{n^{s}}+\frac{\kappa_{F}X^{1-s}}{s-1}-\frac{E_{F}(X)}{X^{s}}+s\int_{X}^{\infty}\frac{E_{F}(u)}{u^{s+1}}\,du, \tag*{(2)}

where X=TαX=T^{\alpha} with α>0\alpha>0 to be chosen. In fact, by (1), the integral on the right-hand side of (2.2 converges absolutely for Re⁡(s)>1−δ\operatorname{Re}(s)>1-\delta. Hence, by analytic continuation, (2) holds for Re⁡(s)>1−δ\operatorname{Re}(s)>1-\delta, apart from the pole at s=1s=1. Differentiating (2) ℓ\ell times and multiplying by (−1)ℓ(-1)^{\ell} gives

(−1)ℓζF(ℓ)(s)=∑n≤XaF(n)(log⁡n)ℓns+κFX1−s∑j=0ℓ(ℓj)j!(log⁡X)ℓ−j(s−1)j+1−EF(X)(log⁡X)ℓXs+s∫X∞EF(u)(log⁡u)ℓus+1 du−ℓ∫X∞EF(u)(log⁡u)ℓ−1us+1 du.(3)(-1)^{\ell}\zeta_{F}^{(\ell)}(s)=\sum_{n\le X}\frac{a_{F}(n)(\log n)^{\ell}}{n^{s}}+\kappa_{F}X^{1-s}\sum_{j=0}^{\ell}\binom{\ell}{j}\frac{j!(\log X)^{\ell-j}}{(s-1)^{j+1}}-\frac{E_{F}(X)(\log X)^{\ell}}{X^{s}}+s\int_{X}^{\infty}\frac{E_{F}(u)(\log u)^{\ell}}{u^{s+1}}\,du-\ell\int_{X}^{\infty}\frac{E_{F}(u)(\log u)^{\ell-1}}{u^{s+1}}\,du. \tag*{(3)}

Now let s=1+its=1+it, where T≤t≤2TT\le t\le2T. We estimate the last four terms on the right-hand side of (3), respectively. Since

(ℓj)j!=ℓ!(ℓ−j)!≤ℓj,\binom{\ell}{j}j!=\frac{\ell!}{(\ell-j)!}\le\ell^{j},

we have

∣κFX−it∑j=0ℓ(ℓj)j!(log⁡X)ℓ−j(it)j+1∣≤κF(log⁡X)ℓT∑j=0ℓ(ℓTlog⁡X)j≪F(log⁡X)ℓT.(4)\left|\kappa_{F}X^{-it}\sum_{j=0}^{\ell}\binom{\ell}{j}\frac{j!(\log X)^{\ell-j}}{(it)^{j+1}}\right|\le\frac{\kappa_{F}(\log X)^{\ell}}{T}\sum_{j=0}^{\ell}\left(\frac{\ell}{T\log X}\right)^{j}\ll_{F}\frac{(\log X)^{\ell}}{T}. \tag*{(4)}

In the last step, we use ℓ≤(log⁡3T)/(log⁡4T)=o(log⁡X)\ell\le(\log_{3}T)/(\log_{4}T)=o(\log X). Moreover, Lemma 2.1 gives

∣EF(X)(log⁡X)ℓX1+it∣≪FX−δ(log⁡X)ℓ.(5)\left|\frac{E_{F}(X)(\log X)^{\ell}}{X^{1+it}}\right|\ll_{F}X^{-\delta}(\log X)^{\ell}. \tag*{(5)}

To estimate the integrals, observe that, for 0≤j≤ℓ0\le j\le\ell and ℓ=o(log⁡X)\ell=o(\log X),

∫X∞(log⁡u)ju1+δ du=X−δ(log⁡X)j∫0∞(1+vlog⁡X)je−δv dv≪FX−δ(log⁡X)j.\int_{X}^{\infty}\frac{(\log u)^{j}}{u^{1+\delta}}\,du=X^{-\delta}(\log X)^{j}\int_{0}^{\infty}\left(1+\frac{v}{\log X}\right)^{j}e^{-\delta v}\,dv\ll_{F}X^{-\delta}(\log X)^{j}.

Consequently, we have

∣s∫X∞EF(u)(log⁡u)ℓus+1 du∣≪FTX−δ(log⁡X)ℓ,(6)\left|s\int_{X}^{\infty}\frac{E_{F}(u)(\log u)^{\ell}}{u^{s+1}}\,du\right|\ll_{F}TX^{-\delta}(\log X)^{\ell}, \tag*{(6)}

and

∣ℓ∫X∞EF(u)(log⁡u)ℓ−1us+1 du∣≪FℓX−δ(log⁡X)ℓ−1≪FX−δ(log⁡X)ℓ.(7)\left|\ell\int_X^\infty\frac{E_{\mathrm F}(u)(\log u)^{\ell-1}}{u^{s+1}}\,du\right| \ll_{\mathrm F} \ell X^{-\delta}(\log X)^{\ell-1} \ll_{\mathrm F} X^{-\delta}(\log X)^\ell. \tag*{(7)}

Combining (4), (5), (6), (7) with (3), we get

(−1)ℓζF(ℓ)(1+it)=∑n≤XaF(n)(log⁡n)ℓn1+it+OF((log⁡X)ℓ(T−1+TX−δ)).(8)(-1)^\ell\zeta_{\mathrm F}^{(\ell)}(1+it)=\sum_{n\le X}\frac{a_{\mathrm F}(n)(\log n)^\ell}{n^{1+it}}+O_{\mathrm F}\left((\log X)^\ell(T^{-1}+TX^{-\delta})\right). \tag*{(8)}

Set α=d+1\alpha=d+1 so that TX−δ=T−1TX^{-\delta}=T^{-1}. On the other hand, it follows from

log⁡(log⁡X)ℓ=ℓlog⁡((d+1)log⁡T)≪Flog⁡2Tlog⁡3Tlog⁡4T=o(log⁡T)\log(\log X)^\ell=\ell\log((d+1)\log T)\ll_{\mathrm F}\frac{\log_2 T\log_3 T}{\log_4 T}=o(\log T)

that (log⁡X)ℓ=To(1)(\log X)^\ell=T^{o(1)} holds. Substituting these estimates into (8) yields that

(−1)ℓζF(ℓ)(1+it)=∑n≤Td+1aF(n)(log⁡n)ℓn1+it+OF(T−1+o(1))(-1)^\ell\zeta_{\mathrm F}^{(\ell)}(1+it)=\sum_{n\le T^{d+1}}\frac{a_{\mathrm F}(n)(\log n)^\ell}{n^{1+it}}+O_{\mathrm F}(T^{-1+o(1)})

holds uniformly for all positive integers ℓ≤(log⁡3T)/(log⁡4T)\ell\le(\log_3 T)/(\log_4 T) and T≤t≤2TT\le t\le2T. This completes the proof. □

The following result provides a basic estimate, which will be used repeatedly in the subsequent proof.

Lemma 2.3. Let A,C>0A,C>0 be fixed. As T→∞T\to\infty, we have

sup⁡0≤j≤log⁡3Tlog⁡4T+1(log⁡3T)AYjlog⁡2T=o(1)andsup⁡0≤j≤log⁡3Tlog⁡4T+1(log⁡3T)ACj!log⁡2T=o(1),\sup_{0\le j\le\frac{\log_3 T}{\log_4 T}+1}\frac{(\log_3 T)^A Y_j}{\log_2 T}=o(1) \qquad\text{and}\qquad \sup_{0\le j\le\frac{\log_3 T}{\log_4 T}+1}\frac{(\log_3 T)^{AC}j!}{\log_2 T}=o(1),

where both supremum are taken over nonnegative integers jj.

Proof. By [13 Lemma 2.6], the Laplace transform of the Dickman function is given by

ρ^(s)=∫0∞ρ(u)e−su du=exp⁡(γ+∫0se−z−1z dz),s∈C.\widehat{\rho}(s)=\int_0^\infty\rho(u)e^{-su}\,du=\exp\left(\gamma+\int_0^s\frac{e^{-z}-1}{z}\,dz\right),\qquad s\in\mathbb{C}.

Taking s=−1s=-1, we obtain ∫0∞euρ(u) du<∞\int_0^\infty e^u\rho(u)\,du<\infty. Since uj≤j!euu^j\le j!e^u for u≥0u\ge0, it follows that Yj≪j!Y_j\ll j!. On the other hand, by Stirling’s formula, uniformly for 1≤j≤(log⁡3T)/(log⁡4T)+11\le j\le(\log_3 T)/(\log_4 T)+1, we have

log⁡(log⁡3T)ACj!log⁡2T≤Alog⁡4T+jlog⁡j−log⁡3T+OC(j)≤−log⁡3Tlog⁡5Tlog⁡4T+OA,C(log⁡3Tlog⁡4T+log⁡4T)→−∞.\begin{aligned} \log\frac{(\log_3 T)^{AC}j!}{\log_2 T}\le A\log_4 T+j\log j-\log_3 T+O_C(j) \\ &\le-\frac{\log_3 T\log_5 T}{\log_4 T}+O_{A,C}\left(\frac{\log_3 T}{\log_4 T}+\log_4 T\right)\to-\infty. \end{aligned}

The case j=0j=0 is immediate. Both assertions follow. □

The following lemma provides uniform estimates for the weighted Dickman function and its tail integral.

Lemma 2.4. Suppose that U∼log⁡3TU \sim\log_{3} T as T→∞T \to\infty. Then uniformly for all positive integers ℓ≤(log⁡3T)/(log⁡4T)\ell\le(\log_{3} T)/(\log_{4} T), we have Uℓρ(U)=o(1)U^{\ell}\rho(U) = o(1), and

∫U∞uℓρ(u) du=o(1).\int_{U}^{\infty} u^{\ell}\rho(u)\,\mathrm{d}u = o(1).

Proof. de Bruijn [6] gives that

log⁡ρ(u)=−u(log⁡u+log⁡2(u+2)−1+O(log⁡2(u+2)log⁡(u+2))).\log\rho(u) = -u\left(\log u + \log_{2}(u+2) - 1 + O\left(\frac{\log_{2}(u+2)}{\log(u+2)}\right)\right).

as u→∞u \to\infty. Hence, for all sufficiently large uu, we have ρ(u)≤exp⁡((−ulog⁡u)/2)\rho(u) \le\exp((-u\log u)/2). Since U∼log⁡3TU \sim\log_{3} T and ℓ≤log⁡3T/log⁡4T\ell\le\log_{3} T/\log_{4} T, we have ℓ≤U/4\ell\le U/4 for sufficiently large TT. Hence, for u≥Uu \ge U,

uℓρ(u)≤exp⁡((ℓ−u2)log⁡u)≤exp⁡(−14ulog⁡u).u^{\ell}\rho(u) \le\exp\left(\left(\ell-\frac{u}{2}\right)\log u\right) \le\exp\left(-\frac{1}{4}u\log u\right).

Taking u=Uu=U gives Uℓρ(U)=o(1)U^{\ell}\rho(U)=o(1). Moreover,

∫U∞uℓρ(u) du≤∫U∞exp⁡(−14ulog⁡u) du=o(1).\int_{U}^{\infty} u^{\ell}\rho(u)\,\mathrm{d}u \le\int_{U}^{\infty} \exp\left(-\frac{1}{4}u\log u\right)\,\mathrm{d}u=o(1).

Both estimates hold uniformly for ℓ≤(log⁡3T)/(log⁡4T)\ell\le(\log_{3} T)/(\log_{4} T), completing the proof. □

Weighted friable sums

For x≥1x \ge1 and y≥2y \ge2, we define the weighted friable sum

ΨF(x,y):=∑n≤xP+(n)≤yaF(n),\Psi_{\mathrm{F}}(x,y) := \sum_{\substack{n\le x\\P^{+}(n)\le y}} a_{\mathrm{F}}(n),

where P+(n)P^{+}(n) is the largest prime factor of nn, with P+(1)=1P^{+}(1)=1. To establish an asymptotic formula for ΨF(x,y)\Psi_{\mathrm{F}}(x,y), we first verify that aFa_{\mathrm{F}} belongs to the class of arithmetic functions considered by de la Bretèche and Tenenbaum [7].

Lemma 3.1. Let F/Q\mathbb{F}/\mathbb{Q} be a fixed number field of degree dd. There exists a constant cF>0c_{\mathrm{F}}>0 such that

aF∈H(1,d+1;13,cF,112),a_{\mathrm{F}}\in\mathcal{H}\left(1,d+1;\frac{1}{3},c_{\mathrm{F}},\frac{1}{12}\right),

where H(z,κ;β,c,δ)\mathcal{H}(z,\kappa;\beta,c,\delta) denotes the class of arithmetic functions introduced in [7].

Proof. Write

ζF(s)=ζ(s)BF(s),BF(s)=ζF(s)ζ(s).\zeta_{\mathrm{F}}(s)=\zeta(s)B_{\mathrm{F}}(s),\qquad B_{\mathrm{F}}(s)=\frac{\zeta_{\mathrm{F}}(s)}{\zeta(s)}.

Since both ζF(s)\zeta_{\mathrm{F}}(s) and ζ(s)\zeta(s) have simple poles at s=1s=1, we have BF(1)=κFB_{\mathrm{F}}(1)=\kappa_{\mathrm{F}}. We now verify the conditions required in [7]. First, we take

β=13,δ=112,β+δ=512<35.\beta=\frac{1}{3},\qquad\delta=\frac{1}{12},\qquad\beta+\delta=\frac{5}{12}<\frac{3}{5}.

By the classical zero-free region for ζ(s)\zeta(s) (see [19 Theorem 3.8]), the function BF(s)\mathcal{B}_{\mathrm{F}}(s) admits a holomorphic continuation to

σ>1−cF(1+log⁡+∣t∣)7/5\sigma> 1-\frac{c_{\mathrm{F}}}{(1+\log^{+}|t|)^{7/5}}

for some sufficiently small constant cF>0c_{\mathrm{F}}>0. Moreover, by [19 Theorem 3.11], we have

1ζ(s)≪log⁡∣t∣(9)\frac{1}{\zeta(s)} \ll\log|t| \tag*{(9)}

in the corresponding zero-free region for sufficiently large ∣t∣|t|. On the other hand, Scourfield [17 Lemma 2.6] gives

ζF(s)≪F(log⁡∣t∣)2/3log⁡2∣t∣(10)\zeta_{\mathrm{F}}(s) \ll_{\mathrm{F}} (\log|t|)^{2/3}\log_{2}|t| \tag*{(10)}

in a region containing the one considered above. Consequently, by (9) and (10),

BF(s)=ζF(s)ζ(s)≪F(log⁡∣t∣)5/3log⁡2∣t∣≪F(1+∣t∣)11/12,\mathcal{B}_{\mathrm{F}}(s)=\frac{\zeta_{\mathrm{F}}(s)}{\zeta(s)}\ll_{\mathrm{F}}(\log|t|)^{5/3}\log_{2}|t|\ll_{\mathrm{F}}(1+|t|)^{11/12},

uniformly in this region. Thus, the required analytic condition is satisfied.

Next, write

BF(s)=∑n=1∞bF(n)ns,\mathcal{B}_{\mathrm{F}}(s)=\sum_{n=1}^{\infty}\frac{b_{\mathrm{F}}(n)}{n^{s}},

where bF=aF∗μb_{\mathrm{F}}=a_{\mathrm{F}}*\mu. For y≥2y\ge2, set

BF(s;y):=∑n≥1P+(n)≤ybF(n)ns.\mathcal{B}_{\mathrm{F}}(s;y):=\sum_{\substack{n\ge1\\P^{+}(n)\le y}}\frac{b_{\mathrm{F}}(n)}{n^{s}}.

By the Euler product,

BF(s;y)=∏p≤y(1−p−s)∏p≤yp∣p(1−(Np)−s)−1.\mathcal{B}_{\mathrm{F}}(s;y)=\prod_{p\le y}(1-p^{-s})\prod_{\substack{\mathfrak{p}\le y\\\mathfrak{p}\mid p}}(1-(N\mathfrak{p})^{-s})^{-1}.

To estimate this product, for E=F,Q\mathrm{E}=\mathrm{F},\mathrm{Q}, define

ZE(s;y)=∏Np≤y(1−(Np)−s)−1.Z_{\mathrm{E}}(s;y)=\prod_{N\mathfrak{p}\le y}(1-(N\mathfrak{p})^{-s})^{-1}.

Fix a sufficiently small η∈(0,1/60)\eta\in(0,1/60). By [17 Lemma 4.1], for E=F,Q\mathrm{E}=\mathrm{F},\mathrm{Q}, we have

ZE(s;y)=ζE(s)(s−1)log⁡y ρ^((s−1)log⁡y)(1+OE(e−(log⁡y)3/5−η))Z_{\mathrm{E}}(s;y)=\zeta_{\mathrm{E}}(s)(s-1)\log y\,\widehat{\rho}((s-1)\log y)\left(1+O_{\mathrm{E}}\left(e^{-(\log y)^{3/5-\eta}}\right)\right)

uniformly for

σ≥1−(log⁡y)−2/5−η,∣t∣≤e(log⁡y)3/5−η,\sigma\ge1-(\log y)^{-2/5-\eta},\qquad|t|\le e^{(\log y)^{3/5-\eta}},

where ρ^\widehat{\rho} denotes the Laplace transform of the Dickman function. Then, taking the quotient gives

ZF(s;y)ZQ(s;y)=BF(s)(1+OF(e−(log⁡y)3/5−η)).(11)\frac{Z_{\mathrm{F}}(s;y)}{Z_{\mathrm{Q}}(s;y)}=\mathcal{B}_{\mathrm{F}}(s)\left(1+O_{\mathrm{F}}\left(e^{-(\log y)^{3/5-\eta}}\right)\right). \tag*{(11)}

On the other hand,

BF(s;y)=ZF(s;y)ZQ(s;y)∏p∣pNp>y(1−(Np)−s)−1.\mathcal{B}_{F}(s;y)=\frac{Z_{F}(s;y)}{Z_{\mathbb{Q}}(s;y)}\prod_{\substack{\mathfrak p\mid p\\ \mathrm{N}\mathfrak p>y}} \left(1-(\mathrm{N}\mathfrak p)^{-s}\right)^{-1}.

For σ≥3/4\sigma\ge3/4, we have

log⁡∏p∣pNp>y(1−(Np)−s)−1≪F∑f=2d∑p>y1/fp−3f/4≪Fy−1/4.\log\prod_{\substack{\mathfrak p\mid p\\ \mathrm{N}\mathfrak p>y}} \left(1-(\mathrm{N}\mathfrak p)^{-s}\right)^{-1}\ll_{F}\sum_{f=2}^{d}\sum_{p>y^{1/f}}p^{-3f/4}\ll_{F}y^{-1/4}.

Consequently, combining this with (11), we obtain

BF(s;y)=BF(s)(1+OF(e−(log⁡y)3/5−η+y−1/4)).\mathcal{B}_{F}(s;y)=\mathcal{B}_{F}(s)\left(1+O_{F}\left(e^{-(\log y)^{3/5-\eta}}+y^{-1/4}\right)\right).

Since 5/12+η<3/55/12+\eta<3/5, we have

BF(s;y)=BF(s)+OF(e−(log⁡y)5/12).\mathcal{B}_{F}(s;y)=\mathcal{B}_{F}(s)+O_{F}\left(e^{-(\log y)^{5/12}}\right).

uniformly for the above region. This establishes the required approximation for the truncated Dirichlet series.

Finally, since 0≤aF(n)≤τd(n)≤τd+1(n)0\le a_{F}(n)\le\tau_{d}(n)\le\tau_{d+1}(n) and

∑n=1∞τd+1(n)ns=ζ(s)d+1,\sum_{n=1}^{\infty}\frac{\tau_{d+1}(n)}{n^{s}}=\zeta(s)^{d+1},

we may take

aF†(n)=τd+1(n)a_{F}^{\dagger}(n)=\tau_{d+1}(n)

as a majorant, where τk(n)\tau_{k}(n) denotes the kk-fold divisor function. The majorant condition is therefore satisfied. This completes the proof. □

The following two propositions provide the estimates for weighted friable sums needed in the proof of Theorem 1.1.

Proposition 3.2. Let F/QF/\mathbb{Q} be a fixed number field, and let u=log⁡x/log⁡yu=\log x/\log y. Then, uniformly for x≥y≥exp⁡((log⁡x)2/3)x\ge y\ge\exp((\log x)^{2/3}), we have

ΨF(x,y)=κFxρ(u)+OF(xρ(u)log⁡(2u)log⁡y).\Psi_{F}(x,y)=\kappa_{F}x\rho(u)+O_{F}\left(x\rho(u)\frac{\log(2u)}{\log y}\right).

Proof. By Lemma 3.1, we apply [7 Corollary 1.3 (ii)] with z=1z=1. Following the notation of [7], we have ψ1(u)=ρ(u)\psi_{1}(u)=\rho(u) and R1(u)≍ρ(u)R_{1}(u)\asymp\rho(u). Moreover,

a0(aF)=lim⁡w→0wζF(1+w)1+w=κF.a_{0}(a_{F})=\lim_{w\to0}\frac{w\zeta_{F}(1+w)}{1+w}=\kappa_{F}.

Since ρ(u)≍1\rho(u)\asymp1 for 1≤u≤21\le u\le2, the additional error term in [7 Corollary 1.3 (ii)], supported near u=1u=1, can be absorbed into the stated error term. This completes the proof. □

Let

b:=⌊(log⁡2T)3⌋,y:=log⁡T40b,L:=exp⁡(log⁡2Tlog⁡3T).b := \left\lfloor(\log_2 T)^3 \right\rfloor,\qquad y := \frac{\log T}{40b},\qquad L := \exp(\log_2 T\log_3 T).

With these choices, we obtain the following asymptotic formula.

Proposition 3.3. Let F/Q\mathbb{F}/\mathbb{Q} be a fixed number field. As T→∞T \to\infty, uniformly for all positive integers ℓ≤(log⁡3T)/(log⁡4T)\ell\le(\log_3 T)/(\log_4 T), we have

∑n<LP+(n)≤yaF(n)(log⁡n)ℓn=(κFYℓ+oF(1))(log⁡2T)ℓ+1.\sum_{\substack{n<L\\ P^{+}(n)\le y}} \frac{a_{\mathbb{F}}(n)(\log n)^\ell}{n} = \left(\kappa_{\mathbb{F}}Y_\ell+o_{\mathbb{F}}(1)\right)(\log_2 T)^{\ell+1}.

Proof. By partial summation,

S:=SF,ℓ(L,y)=∑n≤LP+(n)≤yaF(n)(log⁡n)ℓn=(log⁡L)ℓLΨF(L,y)+∫1LΨF(x,y)(log⁡x)ℓ−ℓ(log⁡x)ℓ−1x2 dx.(12)\begin{aligned} S := S_{\mathbb{F},\ell}(L,y) = \sum_{\substack{n\le L\\ P^{+}(n)\le y}} \frac{a_{\mathbb{F}}(n)(\log n)^\ell}{n} \\ &= \frac{(\log L)^\ell}{L}\Psi_{\mathbb{F}}(L,y)+\int_1^L \Psi_{\mathbb{F}}(x,y)\frac{(\log x)^\ell-\ell(\log x)^{\ell-1}}{x^2}\,dx. \tag*{(12)} \end{aligned}

When 1≤x≤y1\le x\le y, by Lemma 2.1, we have

ΨF(x,y)=AF(x)=κFx+OF(x1−2d+1).\Psi_{\mathbb{F}}(x,y)=A_{\mathbb{F}}(x)=\kappa_{\mathbb{F}}x+O_{\mathbb{F}}\left(x^{1-\frac{2}{d+1}}\right).

On the other hand, when y≤x≤Ly\le x\le L, since (log⁡L)2/3=o(log⁡2T)(\log L)^{2/3}=o(\log_2 T), Proposition 3.2 applies uniformly for this range. Thus, for 1≤x≤L1\le x\le L,

ΨF(x,y)=κFxρ(log⁡xlog⁡y)+EF(x).(13)\Psi_{\mathbb{F}}(x,y)=\kappa_{\mathbb{F}}x\rho\left(\frac{\log x}{\log y}\right)+\mathcal{E}_{\mathbb{F}}(x). \tag*{(13)}

where

EF(x)≪F{x1−2d+1,if 1≤x≤y,xρ(log⁡xlog⁡y)log⁡(2log⁡x/log⁡y)log⁡y,if y≤x≤L.\mathcal{E}_{\mathbb{F}}(x)\ll_{\mathbb{F}} \begin{cases} x^{1-\frac{2}{d+1}}, & \text{if }1\le x\le y,\\ x\rho\left(\frac{\log x}{\log y}\right)\frac{\log(2\log x/\log y)}{\log y}, & \text{if }y\le x\le L. \end{cases}

Substituting (13) into (12) yields that

S=κF(log⁡L)ℓρ(log⁡Llog⁡y)+κF∫1Lρ(log⁡xlog⁡y)(log⁡x)ℓ−ℓ(log⁡x)ℓ−1x dx+E,(14)S=\kappa_{\mathbb{F}}(\log L)^\ell\rho\left(\frac{\log L}{\log y}\right)+\kappa_{\mathbb{F}}\int_1^L\rho\left(\frac{\log x}{\log y}\right)\frac{(\log x)^\ell-\ell(\log x)^{\ell-1}}{x}\,dx+\mathcal{E}, \tag*{(14)}

where

E=OF(CFℓℓ!+(log⁡L)ℓlog⁡yρ(log⁡Llog⁡y)log⁡(2log⁡Llog⁡y)+OF(1log⁡y∫yLρ(log⁡xlog⁡y)log⁡(2log⁡xlog⁡y)(log⁡x)ℓ+ℓ(log⁡x)ℓ−1x dx),\begin{aligned} \mathcal{E}=O_{\mathbb{F}}\left(C_{\mathbb{F}}^\ell\ell!+\frac{(\log L)^\ell}{\log y}\rho\left(\frac{\log L}{\log y}\right)\log\left(\frac{2\log L}{\log y}\right)\right. \\ &+O_{\mathbb{F}}(\frac{1}{\log y}\int_y^L\rho(\frac{\log x}{\log y})\log(\frac{2\log x}{\log y})\frac{(\log x)^\ell+\ell(\log x)^{\ell-1}}{x}\,dx), \end{aligned}

and CF>0C_{\mathbb{F}}>0 depends only on F\mathbb{F}. Making the change of variables u=(log⁡x)/(log⁡y)u=(\log x)/(\log y), the first two terms on the right-hand side of the above identity equal

κF(log⁡y)ℓUℓρ(U)+κF(log⁡y)ℓ+1∫0Uuℓρ(u) du−κFℓ(log⁡y)ℓ∫0Uuℓ−1ρ(u) du,\kappa_{\mathbb{F}}(\log y)^\ell U^\ell\rho(U)+\kappa_{\mathbb{F}}(\log y)^{\ell+1}\int_0^U u^\ell\rho(u)\,du-\kappa_{\mathbb{F}}\ell(\log y)^\ell\int_0^U u^{\ell-1}\rho(u)\,du,

where U=(log⁡L)/(log⁡y)U = (\log L)/(\log y). Note that U=(1+o(1))log⁡3TU = (1 + o(1))\log_{3} T, by using Lemma 2.3 and Lemma 2.4, this equals

(κFYℓ+oF(1))(log⁡y)ℓ+1.(15)(\kappa_F Y_{\ell} + o_F(1))(\log y)^{\ell+1}. \tag*{(15)}

It remains to estimate the error term E\mathcal{E}. In fact, it satisfies

∣E∣(log⁡y)ℓ+1≪FCFℓℓ!(log⁡y)ℓ+1+Yℓ+Yℓ+1log⁡y+ℓ(Yℓ−1+Yℓ)(log⁡y)2+Uℓρ(U)log⁡(2U)(log⁡y)2.\frac{|\mathcal{E}|}{(\log y)^{\ell+1}} \ll_F \frac{C_F^{\ell}\ell!}{(\log y)^{\ell+1}} + \frac{Y_{\ell}+Y_{\ell+1}}{\log y} + \frac{\ell(Y_{\ell-1}+Y_{\ell})}{(\log y)^2} + \frac{U^{\ell}\rho(U)\log(2U)}{(\log y)^2}.

By repeatedly applying Lemma 2.3, one deduces, for large yy,

E=o((log⁡y)ℓ+1).(16)\mathcal{E} = o((\log y)^{\ell+1}). \tag*{(16)}

By the definitions of bb and yy, we have log⁡y=log⁡2T−3log⁡3T−log⁡40+o(1)\log y = \log_{2} T - 3\log_{3} T - \log40 + o(1). Consequently, uniformly for ℓ≤(log⁡3T)/(log⁡4T)\ell\le(\log_{3} T)/(\log_{4} T),

∣(log⁡ylog⁡2T)ℓ+1−1∣≪(ℓ+1)log⁡3Tlog⁡2T.\left|\left(\frac{\log y}{\log_{2} T}\right)^{\ell+1}-1\right| \ll\frac{(\ell+1)\log_{3} T}{\log_{2} T}.

It follows from Lemma 2.3, with A=2A=2, that

Yℓ∣(log⁡ylog⁡2T)ℓ+1−1∣≪Yℓ(ℓ+1)log⁡3Tlog⁡2T=o(1).Y_{\ell}\left|\left(\frac{\log y}{\log_{2} T}\right)^{\ell+1}-1\right| \ll\frac{Y_{\ell}(\ell+1)\log_{3} T}{\log_{2} T}=o(1).

Therefore, combining (15) and (16), we obtain

S=(κFYℓ+oF(1))(log⁡2T)ℓ+1S = (\kappa_F Y_{\ell} + o_F(1))(\log_{2} T)^{\ell+1}

uniformly for all positive integers ℓ≤(log⁡3T)/(log⁡4T)\ell\le(\log_{3} T)/(\log_{4} T). This completes the proof. □

Proof of Theorem 1.1

In this section, we prove Theorem 1.1 using an approach similar to Yang [20, Section 4]. Let N=T1/4N=T^{1/4} and X=Td+1X=T^{d+1}. For a function r:N→{0,1}r:\mathbb{N}\to\{0,1\}, define the resonator

R(t)=∑n≤Nr(n)n−it.\mathcal{R}(t)=\sum_{n\le N}r(n)n^{-it}.

As in [18], let Φ\Phi be a smooth function compactly supported in [1,2][1,2] to satisfy 0≤Φ(t)≤10\le\Phi(t)\le1 always and Φ(t)=1\Phi(t)=1 for 5/4<t<7/45/4<t<7/4. Repeated integration by parts shows that its Fourier transform Φ~(ξ)≪v∣ξ∣−v\widetilde{\Phi}(\xi)\ll_v |\xi|^{-v} holds for every positive integer vv.

Consider the integrals

I0:=I0(R,T)=∫R∣R(t)∣2Φ(tT) dt,I_0:=I_0(\mathcal{R},T)=\int_{\mathbb{R}}|\mathcal{R}(t)|^2\Phi\left(\frac{t}{T}\right)\,dt,

and

I1:=I1(R,T)=∫R(−1)ℓζSF(ℓ)(1+it)∣R(t)∣2Φ(tT) dt.I_1:=I_1(\mathcal{R},T)=\int_{\mathbb{R}}(-1)^{\ell}\zeta_{\mathrm{S}_F}^{(\ell)}(1+it)|\mathcal{R}(t)|^2\Phi\left(\frac{t}{T}\right)\,dt.

Clearly, we obtain

max⁡t∈[T,2T]∣ζSF(ℓ)(1+it)∣≥∣I1∣I0.(17)\max_{t\in[T,2T]}\left|\zeta_{\mathrm{S}_F}^{(\ell)}(1+it)\right|\ge\frac{|I_1|}{I_0}. \tag*{(17)}

Following the argument in [18 p. 471], the Cauchy–Schwarz inequality and the rapid decay of Φ~\widetilde{\Phi} give

I0=TΦ~(0)(1+O(T−1))∑n≤N∣r(n)∣2.(18)I_0=T\widetilde{\Phi}(0)(1+O(T^{-1}))\sum_{n\leq N}|r(n)|^2. \tag*{(18)}

On the other hand, Lemma 2.2 gives

I1=∫R∑k≤XaF(k)(log⁡k)ℓk1+it∣R(t)∣2Φ(tT) dt+OF(T−1+o(1)I0).I_1=\int_{\mathbb{R}}\sum_{k\leq X}\frac{a_{\mathbb{F}}(k)(\log k)^{\ell}}{k^{1+it}}|\mathcal{R}(t)|^2\Phi\left(\frac{t}{T}\right)\,\mathrm{d}t+O_{\mathbb{F}}(T^{-1+o(1)}I_0).

Expanding ∣R(t)∣2|\mathcal{R}(t)|^2, then we obtain

I1=T∑k≤XaF(k)(log⁡k)ℓk∑m,n≤Nr(m)r(n)‾Φ~(Tlog⁡kmn)+OF(T−1+o(1)I0).I_1=T\sum_{k\leq X}\frac{a_{\mathbb{F}}(k)(\log k)^{\ell}}{k}\sum_{m,n\leq N}r(m)\overline{r(n)}\widetilde{\Phi}\left(T\log\frac{km}{n}\right)+O_{\mathbb{F}}(T^{-1+o(1)}I_0).

For km≠nkm\ne n, since n≤Nn\leq N and km,nkm,n are distinct positive integers,

∣log⁡kmn∣≥1N+1.\left|\log\frac{km}{n}\right|\geq\frac{1}{N+1}.

Moreover, using aF(k)≤τd(k)a_{\mathbb{F}}(k)\leq\tau_d(k), we obtain

∑k≤XaF(k)(log⁡k)ℓk≤(log⁡X)ℓ∑k≤Xτd(k)k≪F(log⁡X)ℓ+d.\sum_{k\leq X}\frac{a_{\mathbb{F}}(k)(\log k)^{\ell}}{k}\leq(\log X)^{\ell}\sum_{k\leq X}\frac{\tau_d(k)}{k}\ll_{\mathbb{F}}(\log X)^{\ell+d}.

Here, we use the classical estimate for the divisor function, namely

∑k≤Xτd(k)k≪d(log⁡X)d.\sum_{k\leq X}\frac{\tau_d(k)}{k}\ll_d(\log X)^d.

Since X=Td+1X=T^{d+1} and ℓ≤(log⁡3T)/(log⁡4T)\ell\leq(\log_3 T)/(\log_4 T), it follows that

∑k≤XaF(k)(log⁡k)ℓk=To(1).\sum_{k\leq X}\frac{a_{\mathbb{F}}(k)(\log k)^{\ell}}{k}=T^{o(1)}.

Consequently, by the rapid decay of Φ~\widetilde{\Phi}, the contribution of all terms with km≠nkm\ne n is negligible. More precisely, by applying the Cauchy–Schwarz inequality, we obtain

I1=TΦ~(0)∑km=n≤NaF(k)(log⁡k)ℓkr(m)r(n)‾+OF(To(1)∑n≤N∣r(n)∣2).(19)I_1=T\widetilde{\Phi}(0)\sum_{km=n\leq N}\frac{a_{\mathbb{F}}(k)(\log k)^{\ell}}{k}r(m)\overline{r(n)}+O_{\mathbb{F}}\left(T^{o(1)}\sum_{n\leq N}|r(n)|^2\right). \tag*{(19)}

Since aF(k)a_{\mathbb{F}}(k) and r(n)r(n) are both non-negative, by (17), we have

max⁡t∈[T,2T]∣ζF(ℓ)(1+it)∣≥(1+O(T−1))Qℓ(r)+OF(T−1+o(1)),(20)\max_{t\in[T,2T]}\left|\zeta_{\mathbb{F}}^{(\ell)}(1+it)\right|\geq(1+O(T^{-1}))Q_{\ell}(r)+O_{\mathbb{F}}(T^{-1+o(1)}), \tag*{(20)}

where

Qℓ(r):=Qℓ(r,F,N)=∣∑km=n≤NaF(k)(log⁡k)ℓkr(m)r(n)‾∣/(∑n≤N∣r(n)∣2).Q_{\ell}(r):=Q_{\ell}(r,\mathbb{F},N)=\left|\sum_{km=n\leq N}\frac{a_{\mathbb{F}}(k)(\log k)^{\ell}}{k}r(m)\overline{r(n)}\right|\bigg/\left(\sum_{n\leq N}|r(n)|^2\right).

We next focus on obtaining an effective lower bound for Qℓ(r)Q_{\ell}(r). To this end, using the notation introduced before Proposition 3.3 in § 3, we define

P(y,b):=∏p≤ypb−1andM:={n:n∣P(y,b)}.\mathcal{P}(y,b):=\prod_{p\le y}p^{b-1}\quad\text{and}\quad\mathcal{M}:=\{n:n\mid\mathcal{P}(y,b)\}.

By the prime number theorem, we have P(y,b)≤N\mathcal{P}(y,b)\le N. Take r(n)r(n) to be the characteristic function of the set M\mathcal{M}. In this case,

∑n≤N∣r(n)∣2=∣M∣,\sum_{n\le N}|r(n)|^2=|\mathcal{M}|,

and

Qℓ(r)=1∣M∣∑n∈Mk∣naF(k)(log⁡k)ℓk=∑k∈MaF(k)(log⁡k)ℓk∑n∈Mk∣n1∣M∣.(21)Q_{\ell}(r)=\frac{1}{|\mathcal{M}|}\sum_{\substack{n\in\mathcal{M}\\k\mid n}}\frac{a_{\mathrm{F}}(k)(\log k)^{\ell}}{k} =\sum_{k\in\mathcal{M}}\frac{a_{\mathrm{F}}(k)(\log k)^{\ell}}{k}\sum_{\substack{n\in\mathcal{M}\\k\mid n}}\frac{1}{|\mathcal{M}|}. \tag*{(21)}

Define K\mathcal{K} as the set of positive integers kk satisfying k≤Lk\le L and P+(k)≤yP^{+}(k)\le y. Here L=exp⁡(log⁡2Tlog⁡3T)L=\exp(\log_{2}T\log_{3}T), which is as defined before. Set

K1={k∈K:Ω(k)≤b(log⁡2T)2}.\mathcal{K}_{1}=\left\{k\in\mathcal{K}:\Omega(k)\le\frac{b}{(\log_{2}T)^{2}}\right\}.

where Ω(k)\Omega(k) denotes the number of prime factors of kk, counted with multiplicity. It is clear that K1⊆M\mathcal{K}_{1}\subseteq\mathcal{M}. Furthermore, set K2=K∖K1\mathcal{K}_{2}=\mathcal{K}\setminus\mathcal{K}_{1}.

For any k∈K1k\in\mathcal{K}_{1}, write

k=∏p≤ypαp.k=\prod_{p\le y}p^{\alpha_{p}}.

Since

∑p≤yαp=Ω(k)≤b(log⁡2T)2,\sum_{p\le y}\alpha_{p}=\Omega(k)\le\frac{b}{(\log_{2}T)^{2}},

for the inner sum in the right-hand side of (4.5), we have

∑n∈Mk∣n1∣M∣=∏p≤y(1−αpb)≥1−Ω(k)b≥1−1(log⁡2T)2.\sum_{\substack{n\in\mathcal{M}\\k\mid n}}\frac{1}{|\mathcal{M}|} =\prod_{p\le y}\left(1-\frac{\alpha_{p}}{b}\right) \ge1-\frac{\Omega(k)}{b} \ge1-\frac{1}{(\log_{2}T)^{2}}.

Substituting this lower bound into (4.5) gives

Qℓ(r)≥(1−1(log⁡2T)2)∑k∈K1aF(k)(log⁡k)ℓk.(22)Q_{\ell}(r)\ge\left(1-\frac{1}{(\log_{2}T)^{2}}\right)\sum_{k\in\mathcal{K}_{1}}\frac{a_{\mathrm{F}}(k)(\log k)^{\ell}}{k}. \tag*{(22)}

We now treat the contribution from the set K2\mathcal{K}_{2}. To this end, fix ε∈(0,log⁡2)\varepsilon\in(0,\log2). By Rankin’s trick and aF(k)≤τd(k)a_{\mathrm{F}}(k)\le\tau_{d}(k), we have

∑k∈K2aF(k)(log⁡k)ℓk≤(log⁡L)ℓe−εb/(log⁡2T)2∑P+(k)≤yaF(k)eεΩ(k)k≤(log⁡L)ℓe−εb/(log⁡2T)2∏p≤y(1−eεp)−d.\begin{aligned} \sum_{k\in\mathcal{K}_{2}}\frac{a_{\mathrm{F}}(k)(\log k)^{\ell}}{k} \le(\log L)^{\ell}e^{-\varepsilon b/(\log_{2}T)^{2}}\sum_{P^{+}(k)\le y}\frac{a_{\mathrm{F}}(k)e^{\varepsilon\Omega(k)}}{k} \\ &\le(\log L)^{\ell}e^{-\varepsilon b/(\log_{2}T)^{2}}\prod_{p\le y}\left(1-\frac{e^{\varepsilon}}{p}\right)^{-d}. \end{aligned}

Then, by using Mertens’ theorem,

∑k∈K2aF(k)(log⁡k)ℓk≪F,ε(log⁡L)ℓe−εb/(log⁡2T)2(log⁡y)dε.\sum_{k\in\mathcal{K}_2}\frac{a_F(k)(\log k)^\ell}{k}\ll_{F,\varepsilon}(\log L)^\ell e^{-\varepsilon b/(\log_2 T)^2}(\log y)^{d\varepsilon}.

Since log⁡2y=O(log⁡3T)=o(log⁡2T)\log_2 y=O(\log_3 T)=o(\log_2 T),

b(log⁡2T)2=(1+o(1))log⁡2Tandℓlog⁡2L≪(log⁡3T)2log⁡4T=o(log⁡2T),\frac{b}{(\log_2 T)^2}=(1+o(1))\log_2 T\quad\text{and}\quad\ell\log_2 L\ll\frac{(\log_3 T)^2}{\log_4 T}=o(\log_2 T),

we obtain

∑k∈K2aF(k)(log⁡k)ℓk≪F,εexp⁡(−εlog⁡2T+o(log⁡2T))=oF(1).(23)\sum_{k\in\mathcal{K}_2}\frac{a_F(k)(\log k)^\ell}{k}\ll_{F,\varepsilon}\exp(-\varepsilon\log_2 T+o(\log_2 T))=o_F(1). \tag*{(23)}

By the definition of K\mathcal{K} and the construction of K2\mathcal{K}_2, (23) yields

∑k∈K1aF(k)(log⁡k)ℓk=∑k≤LP+(k)≤yaF(k)(log⁡k)ℓk+oF(1).\sum_{k\in\mathcal{K}_1}\frac{a_F(k)(\log k)^\ell}{k}=\sum_{\substack{k\leq L\\ P^+(k)\leq y}}\frac{a_F(k)(\log k)^\ell}{k}+o_F(1).

Furthermore, by using Proposition 3.3, we obtain

∑k∈K1aF(k)(log⁡k)ℓk=∑k≤LP+(k)≤yaF(k)(log⁡k)ℓk+oF(1)=(κFYℓ+oF(1))(log⁡2T)ℓ+1.(24)\sum_{k\in\mathcal{K}_1}\frac{a_F(k)(\log k)^\ell}{k}=\sum_{\substack{k\leq L\\ P^+(k)\leq y}}\frac{a_F(k)(\log k)^\ell}{k}+o_F(1)=(\kappa_FY_\ell+o_F(1))(\log_2 T)^{\ell+1}. \tag*{(24)}

Moreover, Lemma 2.3 gives

Yℓ(log⁡2T)2=o(1)\frac{Y_\ell}{(\log_2 T)^2}=o(1)

uniformly for ℓ≤(log⁡3T)/(log⁡4T)\ell\leq(\log_3 T)/(\log_4 T). Thus, combining this estimate with (22) and (24) gives

Qℓ(r)≥(κFYℓ+oF(1))(log⁡2T)ℓ+1.Q_\ell(r)\geq(\kappa_FY_\ell+o_F(1))(\log_2 T)^{\ell+1}.

Finally, substituting this into (20) completes the proof of Theorem 1.1.

Acknowledgments

Zikang Dong was supported by the National Natural Science Foundation of China (Grant No. 1240011770). Qiyu Yang was supported by the Natural Science Foundation of Henan Province (Grant No. 252300421782) and the National Natural Science Foundation of China (Grant No. 12601011). Guang-Liang Zhou was supported by the National Natural Science Foundation of China (Grant No. 12401009).

References

References

  1. [1]C. Aistleitner, K. Mahatab, and M. Munsch, Extreme values of the Riemann zeta function on the 1-line, Int. Math. Res. Not. IMRN, 2019 (2019), pp. 6924–6932.arxiv.org/abs/1703.08315
  2. [2]U. Balakrishnan, Extreme values of the Dedekind zeta function, Acta Arith., 46 (1986), pp. 199–209.DOI
  3. [3]A. Bondarenko, P. Darbar, M. V. Hagen, W. Heap, and K. Seip, A dichotomy for extreme values of zeta and Dirichlet L-functions, Bull. Lond. Math. Soc., 55 (2023), pp. 2963–2975.DOI
  4. [4]X. Cui, Z. Peng, Y. Song, and S. Zhao, Extreme values of derivatives of Dirichlet L-functions, Bull. Aust. Math. Soc., (2026), pp. 1–8.arxiv.org/abs/2607.04084
  5. [5]H. Davenport, Multiplicative Number Theory, vol. 74 of Graduate Texts in Mathematics, Springer, 3rd ed., 2000. Revised by H. L. Montgomery.
  6. [6]N. G. de Bruijn, The asymptotic behaviour of a function occurring in the theory of primes, J. Indian Math. Soc., 15 (1951), pp. 25–32.
  7. [7]R. de la Bretèche and G. Tenenbaum, Friable averages of complex arithmetic functions, Funct. Approx. Comment. Math., 71 (2024), pp. 87–119.arxiv.org/abs/2305.06486
  8. [8]A. B. Dixit and K. Mahatab, Large values of L-functions on the 1-line, Bull. Aust. Math. Soc., 103 (2021), pp. 230–243.DOI
  9. [9]Z. Dong, Y. Song, W. Wang, and H. Zhang, On derivatives of zeta and L-functions, Ramanujan J., 66 (2025). Article No.5, 16 pp.DOI
  10. [10]Z. Dong and B. Wei, A note on the large values of |ζ^(ℓ)(1 + it)|, Bull. Aust. Math. Soc., 108 (2023), pp. 217–223.
  11. [11]A. Granville and K. Soundararajan, The distribution of values of L(1, χd), Geom. Funct. Anal., 13 (2003), pp. 992–1028.
  12. [12]A. Granville and K. Soundararajan, Extreme values of |ζ(1 + it)|, In The Riemann zeta function and related themes: papers in honour of Professor K. Ramanujan, vol. 2 of Ramanujan Math. Soc. Lect. Notes Ser., Ramanujan Math. Soc., Mysore, 2006, pp. 65–80.DOI
  13. [13]A. Hildebrand and G. Tenenbaum, Integers without large prime factors, J. Théor. Nombres Bordeaux, 5 (1993), pp. 411–484.DOI
  14. [14]A. Kalmyнин, Omega-theorems for the Riemann zeta function and its derivatives near the line Re(s) = 1, Acta Arith., 186 (2018), pp. 201–217.
  15. [15]E. Landau, Einführung in die elementare und analytische Theorie der algebraischen Zahlen und der Ideale, Chelsea Publishing Company, New York, 1949.
  16. [16]Z. Li, Y. Song, Q. Yang, and S. Zhao, Extreme values of derivatives of the Dedekind zeta function of a cyclotomic field, J. Aust. Math. Soc., (2026), pp. 1–27.arxiv.org/abs/2506.12342
  17. [17]E. J. Scourfield, On ideals free of large prime factors, J. Théor. Nombres Bordeaux, 16 (2004), pp. 733–772.
  18. [18]K. Soundararajan, Extreme values of zeta and L-functions, Math. Ann., 342 (2008), pp. 467–486.arxiv.org/abs/0708.3990
  19. [19]E. C. Titchmarsh, The Theory of the Riemann Zeta-Function, Oxford University Press, New York, 2 ed., 1986. Edited and with a preface by D. R. Heath-Brown.
  20. [20]D. Yang, Extreme values of derivatives of the Riemann zeta function, Mathematika, 68 (2022), pp. 486–510.DOI
  21. [21]——, Extreme values of derivatives of zeta and L-functions, Bull. Lond. Math. Soc., 56 (2024), pp. 79–95.DOI
  22. [22]Q. Yang and S. Zhao, Large values of derivatives of the Riemann zeta function on vertical homogeneous progressions, J. Math. Anal. Appl., 560 (2026). No. 130487, 12pp.arxiv.org/abs/2510.18576

Paper details

vlnc:W_tz98skwk
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC0-1.0
Usage

views · 0

downloads · 0

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