betaFeedback

WorksMathematics

Local Nevanlinna embedding for Hardy space of Dirichlet series below critical exponent

Abstract

In this paper, we investigate a problem posed by Saksman and Seip [ Contemp. Math., 2016] on whether functions in the Hardy space of Dirichlet series Hp\mathcal{H}^p, 0<p<20<p<2, are locally in the Nevanlinna class of the critical half-plane C1/2\mathbb C_{1/2}. We show that the local Nevanlinna embedding fails below critical exponent p=2p=2 by constructing Hp\mathcal{H}^p-functions whose local logarithmic mass becomes unbounded as ℜs→1/2+\real s\to 1/2^+. This also clarifies the distinction between global Jessen-type logarithmic control and local boundary Nevanlinna behavior for Dirichlet series.

Introduction

Hardy spaces of Dirichlet series form a natural meeting point of analytic function theory, harmonic analysis, and complex analysis. If f(s)=∑n=1∞ann−sf(s) = \sum_{n=1}^{\infty} a_n n^{-s} is a Dirichlet series, then the Bohr’s correspondence identifies ff with a holomorphic function on the infinite-dimensional polydisk by writing n=∏jpjνj(n)n = \prod_j p_j^{\nu_j(n)} and associating to ff the formal power series Bf(z)=∑n=1∞anzν(n)\mathcal{B}f(z) = \sum_{n=1}^{\infty} a_n z^{\nu(n)}. For 0<p<∞0 < p < \infty, the Hardy space Hp\mathcal{H}^p is obtained as the completion of Dirichlet polynomials with respect to the norm

∥f∥Hp=∥Bf∥Hp(T∞).\|f\|_{\mathcal{H}^p} = \|\mathcal{B}f\|_{H^p(\mathbb{T}^{\infty})}.

Equivalently, for Dirichlet polynomials this norm is represented by the global vertical mean

∥f∥Hpp=lim⁡T→∞12T∫−TT∣f(it)∣p dt.\|f\|_{\mathcal{H}^p}^p = \lim_{T\to\infty} \frac{1}{2T}\int_{-T}^{T} |f(it)|^p \,dt.

A basic question is to understand how this global mean norm controls the behavior of ff on finite portions of vertical lines. The natural boundary for this problem is the critical line Re⁡(s)=1/2\operatorname{Re}(s) = 1/2. Indeed, point-evaluation functionals are bounded on Hp\mathcal{H}^p throughout the half-plane C1/2\mathbb{C}_{1/2}, where for any θ∈R\theta\in\mathbb{R}, we denote

Cθ:={s∈C:Re⁡(s)>θ}.\mathbb{C}_{\theta} := \{s \in\mathbb{C}: \operatorname{Re}(s) > \theta\}.

and the corresponding evaluation norm grows without bound as Re⁡(s)→1/2+\operatorname{Re}(s) \to1/2^{+}. Thus the line Re⁡s=1/2\operatorname{Re} s = 1/2 is the endpoint at which one may ask whether the global Hp\mathcal{H}^p-norm gives a uniform local trace estimate. The recent formulation emphasizes that restriction to this line amounts, under the Bohr transform, to sampling an infinite-dimensional Hardy function along a single one-dimensional prime-flow orbit, whereas the Hp\mathcal{H}^p-norm averages over all prime coordinates.

This leads to the local embedding problem: determine for which p>0p>0 there exists a constant Cp>0C_p>0 such that every Dirichlet polynomial PP satisfies

sup⁡τ∈R∫ττ+1∣P(12+it)∣p dt≤Cp∥P∥Hpp.(1)\sup_{\tau\in\mathbb{R}}\int_{\tau}^{\tau+1}\left|P\left(\frac{1}{2}+it\right)\right|^{p}\,dt\le C_p\|P\|_{\mathcal{H}^{p}}^{p}. \tag*{(1)}

The essential feature of this estimate is that the constant CpC_p must be independent of the number and choice of prime variables appearing in PP. In this sense, the local embedding problem is a genuinely dimension-free trace problem.

The local embedding problem was formulated explicitly by Hedenmalam, Lindqvist and Seip [11 Theorem 4.11] and later Saksman and Seip [16 Section 3] recorded as Problem 1.2 in their list of open problems for Dirichlet series [17 Problem 2.1]. Subsequent investigations revealed several equivalent formulations of the problem. Bayart and Brevig [2] demonstrated that the local embedding property is equivalent to both the conformally invariant half-plane embedding and the boundedness of all characteristic zero symbols in the Gordon-Hedenmalm class [9] of composition operators; notably, the boundedness of a single canonical operator suffices. Olsen and Saksman [14] identified local embedding with the reverse transfer of local Carleson measures, proving that this transfer may be tested on a distinguished family of atomic measures whose masses are proportional to their distance from the critical boundary. Furthermore, for 2≤p<∞2\le p<\infty, Saksman and Seip [16] established that the local embedding problem is equivalent to a strengthened Carleson-type ergodic identity on the critical distinguished boundary, required to hold for every starting point. Brevig, Ortega-Cerdà, and Seip [5] highlighted the local embedding problem as “perhaps the most important open problem” concerning the Hardy spaces Hp\mathcal{H}^{p}.

Historically, the estimate was known at p=2p=2, where the Hilbert-space structure and orthogonality of Dirichlet monomials play a decisive role. By applying the p=2p=2 result to integer powers, one also obtains the inequality for every even exponent p=2kp=2k. In contrast, Harper’s low-moment estimates for random multiplicative functions imply that the local embedding inequality fails for every 0<p<20<p<2. Thus, for a long time, the unresolved range consisted of the non-even exponents p>2p>2. This gap was recently closed by Chen, Fang, Guo, Hou, Shao, and Zhou [8], who proved a dimension-free local trace estimate for every 2<p<∞2<p<\infty. Combined with the classical p=2p=2 case and the known failure for p<2p<2, their result gives the sharp classification

LEP holds if, and only if, p≥2p\ge2.

Hence p=2p=2 is the critical exponent for the local embedding problem. The significance of the local embedding problem extends well beyond the estimate itself. It is closely connected with boundary behavior of Hardy space of Dirichlet series, Carleson measure transfer, critical-line sampling, composition operators, and the geometry of the prime flow. The recent work shows that the local embedding property is equivalent, in the appropriate exponent ranges, to several classical trace and Carleson-measure formulations. At the same time, failure of the local LpL^{p}-embedding below p=2p=2 raises a subtler question: although ∣f∣p|f|^{p} may fail to admit uniform local control, can the much weaker logarithmic quantity log⁡+∣f∣\log^{+}|f| still remain locally integrable near the critical line? This question leads naturally to the problem posed by Saksman and Seip [17] concerning local Nevanlinna behavior of Hp\mathcal{H}^{p}-functions for p<2p<2. They imposed the following problem:

Problem 1.2. [17 Problem 2.4] Are elements of Hp\mathcal{H}^{p} for 0<p<20<p<2 locally in the Nevanlinna class?

Since, for p>0p>0

log⁡+x≤xpp,x>0,\log^{+}x\le\frac{x^{p}}{p},\qquad x>0,

the local LpL^{p}-embedding implies local logarithmic control whenever it holds. Below the critical exponent p=2p=2, however, the LpL^{p}-embedding breaks down, and it becomes meaningful to ask whether the weaker Nevanlinna-type trace can nevertheless survive. Thus the local Nevanlinna problem may be viewed as a natural logarithmic counterpart of the local embedding problem and provides a finer test of boundary regularity in Hardy spaces of Dirichlet series.

Let Hol⁡(Ω)\operatorname{Hol}(\Omega) denote the space of holomorphic functions on a domain Ω⊂C\Omega\subset\mathbb{C}, endowed with the topology of uniform convergence on compact subsets. A function f∈Hol⁡(Cθ)f\in\operatorname{Hol}(\mathbb{C}_{\theta}) belongs to the local Nevanlinna class, denoted as f∈Nloc(Cθ)f\in N_{\mathrm{loc}}(\mathbb{C}_{\theta}), if for every t0∈Rt_{0}\in\mathbb{R} there exist δ,r>0\delta,r>0 and a positive harmonic function hh on the local rectangle

Qθ(t0;δ,r):={s=σ+it∈C:θ<σ<θ+δ, ∣t−t0∣<r}Q_{\theta}(t_{0};\delta,r):=\{s=\sigma+it\in\mathbb{C}:\theta<\sigma<\theta+\delta,\ |t-t_{0}|<r\}

such that

log⁡+∣f(s)∣≤h(s)for all s∈Qθ(t0;δ,r).\log^{+}|f(s)|\le h(s)\qquad\text{for all }s\in Q_{\theta}(t_{0};\delta,r).

Equivalently, f∈Nloc(Cθ)f\in N_{\mathrm{loc}}(\mathbb{C}_{\theta}) if log⁡+∣f∣\log^{+}|f| admits a harmonic majorant in a neighborhood, relative to Cθ\mathbb{C}_{\theta}, of every finite boundary segment of the line Re⁡(s)=θ\operatorname{Re}(s)=\theta.

A useful consequence of the definition is the following logarithmic trace criterion, that if f∈Nloc(Cθ)f\in N_{\mathrm{loc}}(\mathbb{C}_{\theta}), then for every compact interval I⋐RI\Subset\mathbb{R}, after possibly shrinking the surrounding rectangle, there exists εI>0\varepsilon_I>0 such that

sup⁡0<ε<εI∫Ilog⁡+∣f(θ+ε+it)∣ dt<∞.\sup_{0<\varepsilon<\varepsilon_I}\int_I\log^{+}|f(\theta+\varepsilon+it)|\,\mathrm{d}t<\infty.

Conversely, if the uniform estimate

sup⁡0<ε<ε0∫I∗log⁡+∣f(θ+ε+it)∣ dt<∞\sup_{0<\varepsilon<\varepsilon_0}\int_{I^{*}}\log^{+}|f(\theta+\varepsilon+it)|\,\mathrm{d}t<\infty

holds on a slightly larger open interval I∗⊃II^{*}\supset I, then log⁡+∣f∣\log^{+}|f| admits a positive harmonic majorant on a smaller boundary rectangle over II. Thus, up to standard enlargement or shrinking of intervals, the harmonic-majorant definition and the local logarithmic trace condition are equivalent.

Recently, Guo, Ni and Zhou [10] introduced the Nevanlinna class in the infinite many variables. It is well-known that functions of infinitely many variables and Dirichlet series are intimately linked via the Bohr correspondence. In connection with composition operators on Dirichlet series spaces, Brevig and Perfekt [6] introduced the following notion of the Nevanlinna class for a Dirichlet series ff.

Definition 1.1. [6] A Dirichlet series ff with uniform abscissa σu(f)≤0\sigma_u(f)\le0 is defined in the Nevanlinna class of Dirichlet series Nu\mathcal{N}_u if the following condition holds:

lim sup⁡σ→0+lim⁡T→∞12T∫−TTlog⁡+∣f(σ+it)∣ dt<∞,(2)\limsup_{\sigma\to0^{+}}\lim_{T\to\infty}\frac{1}{2T}\int_{-T}^{T}\log^{+}|f(\sigma+it)|\,\mathrm{d}t<\infty, \tag*{(2)}

where log⁡+(x)=max⁡{log⁡(x),0}\log^{+}(x)=\max\{\log(x),0\}.

The abscissa of uniform convergence of a Dirichlet series ff, denoted by σu(f)\sigma_u(f), is defined by

σu(f):=inf⁡{θ∈R:f(s)=∑n=1∞ann−s converges uniformly on C‾θ}.\sigma_u(f):=\inf\left\{\theta\in\mathbb{R}:f(s)=\sum_{n=1}^{\infty}a_n n^{-s}\text{ converges uniformly on }\overline{\mathbb{C}}_{\theta}\right\}.

If σu(f)≤0\sigma_u(f) \le0 and f∈Hpf \in\mathcal{H}^{p}, then f∈Nuf \in\mathcal{N}_u, which follows from the elementary inequality

log⁡+t≤1ptp\log^{+} t \le\frac{1}{p}t^{p}

(valid for t≥0t \ge0 and p>0p > 0). We emphasize that the condition σu(f)≤0\sigma_u(f) \le0 is essential; without it, the class Nu\mathcal{N}_u can not be properly defined. For example, consider the Dirichlet series

g(s)=∑n=2∞n−snlog⁡n.g(s) = \sum_{n=2}^{\infty} \frac{n^{-s}}{\sqrt{n}\log n}.

It is easy to see that g∈H2g \in\mathcal{H}^{2}, whereas σu(g)=1/2\sigma_u(g) = 1/2. The inequalities

log⁡+(x)≤log⁡(1+x)≤log⁡(2)+log⁡+(x),x≥0\log^{+}(x) \le\log(1+x) \le\log(2) + \log^{+}(x), \qquad x \ge0

along with (1.3) ensures that f∈Nuf \in\mathcal{N}_u if, and only if,

∥f∥0:=lim sup⁡σ→0+lim⁡T→∞12T∫−TTlog⁡(1+∣f(σ+it)∣) dt<∞,f∈Nu.(3)\|f\|_0 := \limsup_{\sigma\to0^+} \lim_{T \to\infty} \frac{1}{2T}\int_{-T}^{T} \log(1+|f(\sigma+it)|)\,dt < \infty,\qquad f \in\mathcal{N}_u. \tag*{(3)}

Here, the class Nu\mathcal{N}_u is an FF-space (see for instance [18]); that is, a topological vector space whose topology is induced by a complete, translation-invariant metric defined by

d0(f,g)=∥f−g∥0for f,g∈Nu.d_0(f,g) = \|f-g\|_0 \qquad\text{for } f,g \in\mathcal{N}_u.

Since ff is a Dirichlet series with σu(f)≤0\sigma_u(f) \le0 and h(x)=log⁡(1+x)h(x) = \log(1+x) is continuous and non-decreasing on [0,∞)[0,\infty), the composition t↦log⁡(1+∣f(it)∣)t \mapsto\log(1+|f(it)|) is almost periodic (see, e.g., [3, 15]). Consequently, the limit in (1.4) exists. In contrast, Guo et al. [10] defined the Nevanlinna-Dirichlet class N\mathcal{N} as the completion of Nu\mathcal{N}_u under the metric ∥⋅∥0\|\cdot\|_0 given in (1.4). Under the canonical embedding of Nu\mathcal{N}_u into N\mathcal{N}, the topology induced on Nu\mathcal{N}_u by the completion coincides with the topology induced by ∥⋅∥0\|\cdot\|_0.

The novelty of the present work lies in identifying and resolving the gap between global Nevanlinna-type control for Dirichlet series and local boundary Nevanlinna behavior on the critical half-plane. More precisely, for 0<p<20 < p < 2, we show that membership in the Hardy space Hp\mathcal{H}^{p} does not in general imply local Nevanlinna regularity in C1/2\mathbb{C}_{1/2}, thereby giving a negative answer to Saksman-Seip’s Problem 2.4. The argument exploits Harper-type low-moment estimates together with a gliding-hump construction to produce an Hp\mathcal{H}^{p} Dirichlet series whose logarithmic mass becomes unbounded on a fixed finite interval as the line Re⁡s=1/2\operatorname{Re}s = 1/2 is approached. After translation, this also yields a separation between the global Nevanlinna class Nu\mathcal{N}_u and local classical Nevanlinna behavior, showing that global logarithmic control need not imply local boundary control. In this way, the result identifies p=2p = 2 as a natural threshold not only for local LpL^{p} trace estimates, but also for universal local Nevanlinna behavior of Hardy-Dirichlet series.

The paper is organized as follows. Section 2 presents our main results concerning the negative resolution to Problem 1.2, along with several key supporting lemmas largely inspired by the work of Harper [13]. Section 3 opens with the proof of Theorem 2.1, followed by a detailed investigation linking the local Nevanlinna embedding problem (LNEP) to global and local logarithmic mean control, as well as to the original problem of Saksman and Seip (Problem 1.2).

Notation. Throughout this article, CC denotes a positive constant which may vary from line to line. When necessary, we write that C=C(x)C = C(x) to indicate that the constant depends on a parameter xx. If ff and gg are two real functions defined on the same set SS, the notation f≲gf \lesssim g (or f≲pgf \lesssim_p g) means that there exists a constant C≥1C \ge1 such that C−1f(x)≤g(x)≤Cf(x)C^{-1}f(x) \le g(x) \le Cf(x) (or C−1(p)f(x)≤g(x)≤C(p)f(x)C^{-1}(p)f(x) \le g(x) \le C(p)f(x)), for all xx, while f(x)≲g(x)f(x) \lesssim g(x) (or f(x)≲pg(x)f(x) \lesssim_p g(x)) signifies that there exists a constant C>0C > 0 such that f(x)≤Cg(x)f(x) \le Cg(x) (or f(x)≤C(p)g(x)f(x) \le C(p)g(x)), for all xx. Also, the symbol Op(1)O_p(1) indicates a sequence (an)n=1∞(a_n)_{n=1}^{\infty} is bounded by a constant depending exclusively on the parameter pp; that is, there exist C(p)>0C(p) > 0 and N0(p)∈NN_0(p) \in\mathbb{N} such that ∣an∣≤C(p)|a_n| \le C(p) for all n≥N0(p)n \ge N_0(p).

Main Results and Supporting Lemmas

This section starts with our primary result, which provides a negative resolution to Problem 1.2, followed by a few key supporting lemmas required for its proof.

Theorem 2.1. Let 0<p<20 < p < 2. Then there exists a Dirichlet series f(s)=∑n=1∞ann−s∈Hpf(s) = \sum_{n=1}^{\infty} a_n n^{-s} \in\mathcal{H}^p such that

lim sup⁡ε→0+∫01log⁡+∣f(12+ε+it)∣ dt=+∞.\limsup_{\varepsilon\to0^+} \int_0^1 \log^+ \left|f\left(\frac{1}{2}+\varepsilon+it\right)\right|\,dt=+\infty.

Consequently, ff does not belong locally to the Nevanlinna class in the half-plane C1/2\mathbb{C}_{1/2}. In particular,

Hp⊄Nloc(C1/2),0<p<2,\mathcal{H}^p \not\subset\mathcal{N}_{\mathrm{loc}}(\mathbb{C}_{1/2}), \qquad0 < p < 2,

thereby answering Problem 2.4 of Saksman and Seip in the negative.

Remark 2.1. For every 0<p<20 < p < 2, there exists a Dirichlet series g(s)=∑n=1∞bnn−sg(s) = \sum_{n=1}^{\infty} b_n n^{-s} satisfying σu(g)≤0\sigma_u(g) \le0 and g∈Nug \in\mathcal{N}_u, such that g∉Nloc(C0)g \notin\mathcal{N}_{\mathrm{loc}}(\mathbb{C}_0), where C0={s∈C:Re⁡(s)>0}\mathbb{C}_0 = \{s \in\mathbb{C}: \operatorname{Re}(s) > 0\}. More precisely,

lim sup⁡σ→0+∫01log⁡+∣g(σ+it)∣ dt=+∞,\limsup_{\sigma\to0^+} \int_0^1 \log^+|g(\sigma+it)|\,dt=+\infty,

so that Nu⊄Nloc(C0)\mathcal{N}_u \not\subset\mathcal{N}_{\mathrm{loc}}(\mathbb{C}_0). The function gg is obtained from ff in Theorem 2.1 via the translation g(s)=f(s+1/2)g(s)=f(s+1/2), highlighting the sharp distinction between global Jessen-type logarithmic control Nu\mathcal{N}_u and local logarithmic boundary trace control near a finite segment of the imaginary axis.

Lemma 2.1. Let Q={s=σ+it∈C:1/2<σ<1/2+δ, a<t<b}Q=\{s=\sigma+it\in\mathbb{C}:1/2<\sigma<1/2+\delta,\ a<t<b\}. If f∈Hol⁡(Q)f\in\operatorname{Hol}(Q) and log⁡+∣f∣\log^+|f| admits a harmonic majorant in QQ, then for every compact sub interval I⊂(a,b)I\subset(a,b) and every 0<δ′<δ0<\delta'<\delta,

sup⁡0<ε<δ′∫Ilog⁡+∣f(12+ε+it)∣ dt<+∞.\sup_{0<\varepsilon<\delta'}\int_I\log^+\left|f\left(\frac{1}{2}+\varepsilon+it\right)\right|\,dt<+\infty.

Proof. Let f∈Hol⁡(Q)f\in\operatorname{Hol}(Q). The local Nevanlinna membership f∈Nloc(C1/2)f\in\mathcal{N}_{\mathrm{loc}}(\mathbb{C}_{1/2}) near a bounded interval of the critical boundary line Re⁡(s)=1/2\operatorname{Re}(s)=1/2 is equivalent to the restriction property f∣Q∈N(Q)f|_Q\in\mathcal{N}(Q). This condition holds precisely when

sup⁡0<ε<δ∫Ilog⁡+∣f(12+ε+it)∣ dt<+∞\sup_{0<\varepsilon<\delta}\int_I\log^+\left|f\left(\frac{1}{2}+\varepsilon+it\right)\right|\,dt<+\infty

for every compact subinterval I⊂(a,b)I\subset(a,b) and some δ>0\delta>0. Using the elementary inequality log⁡+x≤xp/p\log^+x\le x^p/p for x>0x>0 and p>0p>0, we observe that

sup⁡0<ε<δ∫Ilog⁡+∣f(12+ε+it)∣ dt≤1psup⁡0<ε<δ∫I∣f(12+ε+it)∣p dt≤Cp,I∥f∥Hpp.\sup_{0<\varepsilon<\delta}\int_I\log^+\left|f\left(\frac{1}{2}+\varepsilon+it\right)\right|\,dt \le\frac{1}{p}\sup_{0<\varepsilon<\delta}\int_I\left|f\left(\frac{1}{2}+\varepsilon+it\right)\right|^p\,dt \le C_{p,I}\|f\|_{\mathcal{H}^p}^p.

Consequently, the Local Embedding Problem (LEP) directly implies the Local Nevanlinna Embedding Problem (LNEP). □

Moreover, the argument above breaks down for p<2p<2. Consequently, if there exists a sequence εk→0+\varepsilon_k\to0^+ such that

∫Ilog⁡+∣f(12+εk+it)∣ dt⟶+∞,\int_I \log^+\left|f\left(\frac{1}{2}+\varepsilon_k+it\right)\right|\,dt\longrightarrow+\infty,

then ff fails to belong locally to the Nevanlinna class near the boundary segment {1/2+it:t∈I}\{1/2+it:t\in I\}.

Lemma 2.2. Fix 0<p<20<p<2, and define DX(s)=∑n≤Xn−sD_X(s)=\sum_{n\le X}n^{-s} and QX(s)=DX(s)/∥DX∥HpQ_X(s)=D_X(s)/\|D_X\|_{\mathcal H^p}, so that ∥QX∥Hp=1\|Q_X\|_{\mathcal H^p}=1. By Harper’s low-moment estimates, ∥DX∥Hp≍pX/(log⁡log⁡X)1/4\|D_X\|_{\mathcal H^p}\asymp_p\sqrt{X}/(\log\log X)^{1/4} as X→+∞X\to+\infty. Setting εX=1/log⁡X\varepsilon_X=1/\log X, we have

inf⁡0≤t≤1∣QX(12+εX+it)∣≳p(log⁡log⁡X)1/4⟶+∞.(4)\inf_{0\le t\le1}\left|Q_X\left(\frac{1}{2}+\varepsilon_X+it\right)\right|\gtrsim_p(\log\log X)^{1/4}\longrightarrow+\infty. \tag*{(4)}

In particular,

∫01log⁡(1+∣QX(12+εX+it)∣) dt⟶+∞as X→+∞.\int_0^1\log\left(1+\left|Q_X\left(\frac{1}{2}+\varepsilon_X+it\right)\right|\right)\,dt\longrightarrow+\infty\quad\text{as }X\to+\infty.

Proof. Let DX(s)=∑n≤Xn−sD_X(s)=\sum_{n\le X}n^{-s}, and write p=2qp=2q with 0<q<10<q<1. By Harper’s moment bounds [13 Theorem 1],

E∣∑n≤Xf(n)∣q≲(X1+(1−q)log⁡log⁡X)q,(5)\mathbb{E}\left|\sum_{n\le X}f(n)\right|^q\lesssim\left(\frac{X}{1+(1-q)\sqrt{\log\log X}}\right)^q, \tag*{(5)}

where f(n)f(n) is a Steinhaus random multiplicative function defined by f(n):=∏pa∥nf(p)af(n):=\prod_{p^a\parallel n}f(p)^a for all n∈Nn\in\mathbb{N}, with pa∥np^a\parallel n denoting the exact prime power dividing nn. Under the Bohr transform, the expectation in (5) corresponds to the H2q\mathcal H^{2q}-norm of DXD_X. Consequently, for p=2q<2p=2q<2,

∥DX∥Hp≍p(X1+(1−p/2)log⁡log⁡X)p/2,\|D_X\|_{\mathcal H^p}\asymp_p\left(\frac{X}{1+(1-p/2)\sqrt{\log\log X}}\right)^{p/2},

which yields the asymptotic behavior

∥DX∥Hp≲pX(log⁡log⁡X)1/4as X→+∞.(6)\|D_X\|_{\mathcal H^p}\lesssim_p\frac{\sqrt{X}}{(\log\log X)^{1/4}}\quad\text{as }X\to+\infty. \tag*{(6)}

On the critical line Re⁡(s)=1/2\operatorname{Re}(s)=1/2, partial summation and Riemann–Stieltjes integration give

DX(12+it)=X1/2−it1/2−it+O(1),0≤t≤1,D_X\left(\frac{1}{2}+it\right)=\frac{X^{1/2-it}}{1/2-it}+O(1),\qquad0\le t\le1,

from which it follows that

∣DX(12+it)∣≍X,uniformly for 0≤t≤1.(7)\left|D_X\left(\frac{1}{2}+it\right)\right|\asymp\sqrt{X},\qquad\text{uniformly for }0\le t\le1. \tag*{(7)}

By normal family estimates, replacing 1/21/2 with 1/2+εX1/2+\varepsilon_X where εX=1/log⁡X\varepsilon_X=1/\log X preserves the bound ∣DX(1/2+εX+it)∣≍X|D_X(1/2+\varepsilon_X+it)|\asymp\sqrt{X}.

Normalizing QX(s)=DX(s)/∥DX∥HpQ_X(s)=D_X(s)/\|D_X\|_{\mathcal H^p} so that ∥QX∥Hp=1\|Q_X\|_{\mathcal H^p}=1, we combine (6) and (7) to obtain

∣QX(12+εX+it)∣≳p(log⁡log⁡X)1/4,uniformly for 0≤t≤1.\left|Q_X\left(\frac{1}{2}+\varepsilon_X+it\right)\right|\gtrsim_p(\log\log X)^{1/4},\qquad\text{uniformly for }0\le t\le1.

Taking logarithms on both sides yields

∫01log⁡(1+∣QX(12+εX+it)∣) dt=14log⁡log⁡log⁡X+Op(1),(8)\int_{0}^{1}\log\left(1+\left|Q_{X}\left(\frac{1}{2}+\varepsilon_{X}+it\right)\right|\right)\,dt=\frac{1}{4}\log\log\log X+O_{p}(1), \tag*{(8)}

which tends to +∞+\infty as X→+∞X\to+\infty. This completes the proof. □

Lemma 2.3. For 0<p<∞0<p<\infty and σ>1/2\sigma>1/2, point evaluations on Hp\mathcal{H}^{p} satisfy ∣f(σ+it)∣≤ζ(2σ)1/p∥f∥Hp|f(\sigma+it)|\leq\zeta(2\sigma)^{1/p}\|f\|_{\mathcal{H}^{p}} uniformly in t∈Rt\in\mathbb{R}. Consequently, for every ε>0\varepsilon>0,

sup⁡t∈R∣f(12+ε+it)∣≤ζ(1+2ε)1/p∥f∥Hp,\sup_{t\in\mathbb{R}}\left|f\left(\frac{1}{2}+\varepsilon+it\right)\right|\leq\zeta(1+2\varepsilon)^{1/p}\|f\|_{\mathcal{H}^{p}},

which controls the tail-block contributions in the gliding-hump construction along vertical lines Re⁡(s)=1/2+εk\operatorname{Re}(s)=1/2+\varepsilon_{k}.

The proof of Lemma 2.3 is straightforward and arises directly from the point evaluations of the function f(1/2+ε+it)f(1/2+\varepsilon+it), applying the Cauchy–Schwarz inequality effectively.

Lemma 2.4. There exist a sequence (ak)k=1∞∈ℓ2(a_{k})_{k=1}^{\infty}\in\ell^{2} with ak→0+a_{k}\to0^{+} and a rapidly increasing sequence of integers Xk→∞X_{k}\to\infty such that the series

f(s)=∑k=1∞akQXk(s)f(s)=\sum_{k=1}^{\infty}a_{k}Q_{X_{k}}(s)

converges in Hp\mathcal{H}^{p}. Setting εk=1/log⁡Xk\varepsilon_{k}=1/\log X_{k}, the sequences (ak)(a_{k}) and (Xk)(X_{k}) can be chosen recursively so that

inf⁡0≤t≤1∣f(12+εk+it)∣≥kfor all k≥1.\inf_{0\leq t\leq1}\left|f\left(\frac{1}{2}+\varepsilon_{k}+it\right)\right|\geq k\qquad\text{for all }k\geq1.

Consequently, ∫01log⁡+∣f(1/2+εk+it)∣ dt≥log⁡k\int_{0}^{1}\log^{+}|f(1/2+\varepsilon_{k}+it)|\,dt\geq\log k, yielding

lim sup⁡ε→0+∫01log⁡+∣f(12+ε+it)∣ dt=+∞.\limsup_{\varepsilon\to0^{+}}\int_{0}^{1}\log^{+}\left|f\left(\frac{1}{2}+\varepsilon+it\right)\right|\,dt=+\infty.

The technique used in Lemma 2.4 is known as gliding-hump technique. This technique converts a sequence of singular test polynomials into a single function f=∑k=1∞fk∈Hpf=\sum_{k=1}^{\infty}f_{k}\in\mathcal{H}^{p} exhibiting unbounded localized behavior along a sequence of scales. Each block fk∈Hpf_{k}\in\mathcal{H}^{p} is chosen to be small in the ambient norm ∥fk∥Hp\|f_{k}\|_{\mathcal{H}^{p}}, ensuring series convergence f∈Hpf\in\mathcal{H}^{p}. Moreover, each block is exceptionally large at the newly selected kk-th scale, and negligible at all previously established scales j<kj<k. By ensuring fkf_{k} dominates both the prior partial sum and the future tail at scale kk, the resulting limit ff displays infinitely many bad scales simultaneously, upgrading quantitative norm blowups into a qualitative counterexample.

Proof of Lemma 2.4. Assume the hypotheses of Lemma 2.2. From (2.1), the normalized Harper blocks satisfy

Ap(X):=inf⁡0≤t≤1∣QX(12+εX+it)∣≥Cp(log⁡log⁡X)1/4⟶+∞as X→+∞,(9)A_{p}(X):=\inf_{0\leq t\leq1}\left|Q_{X}\left(\frac{1}{2}+\varepsilon_{X}+it\right)\right|\geq C_{p}(\log\log X)^{1/4}\longrightarrow+\infty\quad\text{as }X\to+\infty, \tag*{(9)}

where εX=1/log⁡X\varepsilon_{X}=1/\log X. We construct a function f∈Hpf\in\mathcal{H}^{p} by setting

f(s):=∑k=1∞akQXk(s).f(s):=\sum_{k=1}^{\infty}a_{k}Q_{X_{k}}(s).

Let r=min⁡{1,p}r=\min\{1,p\}. By choosing coefficients ak>0a_{k}>0 such that ∑k=1∞akr<+∞\sum_{k=1}^{\infty}a_{k}^{r}<+\infty, the quasi-norm triangle inequality in Hp\mathcal{H}^{p} guarantees the convergence of ∑k=1∞akQXk\sum_{k=1}^{\infty}a_{k}Q_{X_{k}} in Hp\mathcal{H}^{p}, ensuring f∈Hpf \in\mathcal{H}^{p}. We construct the sequences (ak)k=1∞(a_k)_{k=1}^{\infty} and (Xk)k=1∞(X_k)_{k=1}^{\infty} inductively. Suppose a1,X1,…,ak−1,Xk−1a_1,X_1,\ldots,a_{k-1},X_{k-1} have been chosen, and define the partial sum

Fk−1(s):=∑j<kajQXj(s).F_{k-1}(s):=\sum_{j<k}a_jQ_{X_j}(s).

Since Fk−1F_{k-1} is a Dirichlet polynomial, it is bounded on the compact rectangle [1/2,3/2]×[0,1][1/2,3/2]\times[0,1]; hence,

Mk−1:=sup⁡0≤u≤10≤t≤1∣Fk−1(12+u+it)∣<+∞.M_{k-1}:=\sup_{\substack{0\leq u\leq1\\0\leq t\leq1}}\left|F_{k-1}\left(\frac{1}{2}+u+it\right)\right|<+\infty.

Next, choose ak>0a_k>0 sufficiently small such that akr≤2−ka_k^r\leq2^{-k} and

Cp(εj)ak≤2−kfor all j<k,C_p(\varepsilon_j)a_k\leq2^{-k}\qquad\text{for all }j<k,

where Cp(ε)=ζ(1+2ε)1/pC_p(\varepsilon)=\zeta(1+2\varepsilon)^{1/p} denotes the point-evaluation bound on Re⁡(s)=1/2+ε\operatorname{Re}(s)=1/2+\varepsilon from Lemma 2.3. With aka_k fixed, choose Xk∈NX_k\in\mathbb{N} so large that εk=1/log⁡Xk<εk−1/2\varepsilon_k=1/\log X_k<\varepsilon_{k-1}/2 and the kk-th block dominates both Fk−1F_{k-1} and the future tail:

akAp(Xk)≥4(Mk−1+k+1).a_kA_p(X_k)\geq4(M_{k-1}+k+1).

Now evaluate f(s)f(s) at s=1/2+εk+its=1/2+\varepsilon_k+it for 0≤t≤10\leq t\leq1, and decompose ff into

f(12+εk+it)=Fk−1(12+εk+it)+akQXk(12+εk+it)+Rk(12+εk+it),f\left(\frac{1}{2}+\varepsilon_k+it\right)=F_{k-1}\left(\frac{1}{2}+\varepsilon_k+it\right)+a_kQ_{X_k}\left(\frac{1}{2}+\varepsilon_k+it\right)+R_k\left(\frac{1}{2}+\varepsilon_k+it\right),

where Rk(s):=∑j>kajQXj(s)R_k(s):=\sum_{j>k}a_jQ_{X_j}(s). By choice of Mk−1M_{k-1}, we have ∣Fk−1(1/2+εk+it)∣≤Mk−1\left|F_{k-1}(1/2+\varepsilon_k+it)\right|\leq M_{k-1}. By Lemma 2.3 and our bound on aja_j, the tail RkR_k satisfies

∣Rk(12+εk+it)∣≤∑j>kajCp(εk)≤∑j>k2−j<1.\left|R_k\left(\frac{1}{2}+\varepsilon_k+it\right)\right|\leq\sum_{j>k}a_jC_p(\varepsilon_k)\leq\sum_{j>k}2^{-j}<1.

Applying the reverse triangle inequality yields, for all 0≤t≤10\leq t\leq1,

∣f(12+εk+it)∣≥ak∣QXk(12+εk+it)∣−∣Fk−1(12+εk+it)∣−∣Rk(12+εk+it)∣\left|f\left(\frac{1}{2}+\varepsilon_k+it\right)\right|\geq a_k\left|Q_{X_k}\left(\frac{1}{2}+\varepsilon_k+it\right)\right|-\left|F_{k-1}\left(\frac{1}{2}+\varepsilon_k+it\right)\right|-\left|R_k\left(\frac{1}{2}+\varepsilon_k+it\right)\right|
≥4(Mk−1+k+1)−Mk−1−1>k.\geq4(M_{k-1}+k+1)-M_{k-1}-1>k.

Integrating the positive part of the logarithm yields

∫01log⁡+∣f(12+εk+it)∣ dt≥log⁡k.\int_0^1\log^{+}\left|f\left(\frac{1}{2}+\varepsilon_k+it\right)\right|\,dt\geq\log k.

Since εk↓0\varepsilon_k\downarrow0 as k→∞k\to\infty, taking the limit superior establishes

lim sup⁡ε→0+∫01log⁡+∣f(12+ε+it)∣ dt=+∞.(10)\limsup_{\varepsilon\to0+}\int_0^1\log^{+}\left|f\left(\frac{1}{2}+\varepsilon+it\right)\right|\,dt=+\infty. \tag*{(10)}

which implies f∉Nloc(C1/2)f\notin N_{\mathrm{loc}}(\mathbb{C}_{1/2}). This completes the proof. □\square

Lemma 2.5. Let g(s)=f(s+1/2)g(s)=f(s+1/2), where f∈Hpf\in\mathcal{H}^{p} is the Dirichlet series constructed in Lemma 2.4. Then g∈Hpg\in\mathcal{H}^{p} and its uniform convergence abscissa satisfies σu(g)≤0\sigma_u(g)\leq0, placing gg in the Nevanlinna class Nu\mathcal{N}_u. Since

lim sup⁡ε→0+∫01log⁡+∣g(ε+it)∣ dt=+∞,\limsup_{\varepsilon\to0+}\int_0^1\log^{+}|g(\varepsilon+it)|\,dt=+\infty,

it follows that g∉Nloc(C0)g\notin N_{\mathrm{loc}}(\mathbb{C}_0), establishing that Nu⊄Nloc(C0)\mathcal{N}_u\not\subset N_{\mathrm{loc}}(\mathbb{C}_0).

Proof. Set g(s):=f(s+1/2)g(s) := f(s+1/2), where f∈Hpf \in\mathcal{H}^{p} is the Dirichlet series constructed in Lemma 2.4. Expanding gg as a Dirichlet series yields

g(s)=∑k=1∞ak∥DXk∥Hp(∑n≤Xkn−1/2n−s)=∑n=1∞cnn−s,g(s) = \sum_{k=1}^{\infty} \frac{a_k}{\lVert D_{X_k}\rVert_{\mathcal{H}^{p}}}\left(\sum_{n\leq X_k}n^{-1/2}n^{-s}\right)=\sum_{n=1}^{\infty}c_n n^{-s},

where the coefficients are given by

cn=1n∑{k:Xk≥n}ak∥DXk∥Hp≥0.c_n=\frac{1}{\sqrt{n}}\sum_{\{k:X_k\geq n\}}\frac{a_k}{\lVert D_{X_k}\rVert_{\mathcal{H}^{p}}}\geq0.

Since f∈Hpf \in\mathcal{H}^{p}, its uniform convergence abscissa satisfies σu(f)≤1/2\sigma_u(f)\leq1/2. The translated function gg therefore satisfies σu(g)=σu(f)−1/2≤0\sigma_u(g)=\sigma_u(f)-1/2\leq0, placing gg in the Nevanlinna class Nu\mathcal{N}_u. By construction, g(ε+it)=f(1/2+ε+it)g(\varepsilon+it)=f(1/2+\varepsilon+it). Applying (2.7) from Lemma 2.4, we obtain

lim sup⁡ε→0+∫01log⁡+∣g(ε+it)∣ dt=+∞.\limsup_{\varepsilon\to0+}\int_0^1\log^{+}|g(\varepsilon+it)|\,dt=+\infty.

Consequently, log⁡+∣g∣\log^{+}|g| cannot admit a local harmonic majorant near the imaginary axis Re⁡(s)=0\operatorname{Re}(s)=0. Hence g∉Nloc(C0)g\notin\mathcal{N}_{\mathrm{loc}}(\mathbb{C}_0), which establishes that Nu⊄Nloc(C0)\mathcal{N}_u\not\subset\mathcal{N}_{\mathrm{loc}}(\mathbb{C}_0).

Remark 2.2. The construction in Lemma 2.5 demonstrates that the Hardy space Hp\mathcal{H}^{p} of Dirichlet series fails to lie locally in the Nevanlinna class Nloc(C1/2)\mathcal{N}_{\mathrm{loc}}(\mathbb{C}_{1/2}) whenever 0<p<20<p<2. Combining this negative result for 0<p<20<p<2 with the local LpL^{p}-embedding theorem for p≥2p\geq2, and applying the elementary inequality log⁡+x≤xp/p\log^{+}x\leq x^{p}/p for x>0x>0, we establish the following sharp threshold classification for local Nevanlinna boundary behavior: Let 0<p<+∞0<p<+\infty. Then

Hp⊂Nloc(C1/2)⟺p≥2.\mathcal{H}^{p}\subset\mathcal{N}_{\mathrm{loc}}(\mathbb{C}_{1/2})\Longleftrightarrow p\geq2.

Consequently, p=2p=2 is the critical exponent governing local Nevanlinna boundary behavior in Hardy spaces of Dirichlet series.

Proof of Theorem 2.1 and Global Vs Local Logarithmic Mean Control

In [1], we posed the following open problem regarding quantitative logarithmic trace estimates:

Problem 3.1. [1] Does there exist a constant C>0C>0 such that, for every g∈Nug\in\mathcal{N}_u with σu(g)≤0\sigma_u(g)\leq0,

sup⁡σ>0sup⁡τ∈R∫ττ+1log⁡(1+∣g(σ+it)∣) dt≤C∥g∥0?(11)\sup_{\sigma>0}\sup_{\tau\in\mathbb{R}}\int_{\tau}^{\tau+1}\log(1+|g(\sigma+it)|)\,dt\leq C\lVert g\rVert_0? \tag*{(11)}

We refer to condition (11) as the quantitative Nu\mathcal{N}_u local Nevanlinna embedding problem (LNEP). This naturally leads to the following underlying question:

Question 3.3. Does global mean logarithmic control imply uniform local logarithmic control on short interval segments?

For f∈Nuf\in\mathcal{N}_u, define the global logarithmic mean (Jessen-type) by

Jf(σ):=lim⁡T→∞12T∫−TTlog⁡+∣f(σ+it)∣ dt.J_f(\sigma):=\lim_{T\to\infty}\frac{1}{2T}\int_{-T}^{T}\log^{+}|f(\sigma+it)|\,dt.

By classical results of Jessen [12], this limit exists for every σ>0\sigma> 0, and the norm defining the Nu\mathcal{N}_u metric is given by

∥f∥0:=lim sup⁡σ→0+Jf(σ)<+∞.\|f\|_0 := \limsup_{\sigma\to0^+} J_f(\sigma) < +\infty.

The corresponding short-interval local mass quantity is defined as

Lf(σ):=sup⁡τ∈R∫ττ+1log⁡(1+∣f(σ+it)∣) dt,L_f(\sigma) := \sup_{\tau\in\mathbb{R}} \int_{\tau}^{\tau+1} \log(1 + |f(\sigma+ it)|)\,dt,

with the uniform local norm given by L(f):=sup⁡σ>0Lf(σ)L(f) := \sup_{\sigma>0} L_f(\sigma). Problem 3.1 reduces to testing the implication

∥f∥0<∞⟹?L(f)<+∞,\|f\|_0 < \infty\mathrel{\overset{?}{\Longrightarrow}} L(f) < +\infty,

or, quantitatively, asking whether there exists a universal constant C>0C > 0 such that L(f)≤C∥f∥0L(f) \le C\|f\|_0.

Control of Jf(σ)J_f(\sigma) does not automatically yield control of short-interval masses. Indeed, setting uσ(t):=log⁡(1+∣f(σ+it)∣)u_\sigma(t) := \log(1 + |f(\sigma+ it)|) and Aσ(τ):=∫ττ+1uσ(t) dtA_\sigma(\tau) := \int_{\tau}^{\tau+1} u_\sigma(t)\,dt, Fubini’s theorem gives

lim⁡T→∞12T∫−TTAσ(τ) dτ=Jf(σ).\lim_{T\to\infty} \frac{1}{2T}\int_{-T}^{T} A_\sigma(\tau)\,d\tau= J_f(\sigma).

Thus, Jf(σ)J_f(\sigma) controls only the average over τ\tau of the unit-interval masses Aσ(τ)A_\sigma(\tau), whereas Lf(σ)=sup⁡τ∈RAσ(τ)L_f(\sigma) = \sup_{\tau\in\mathbb{R}} A_\sigma(\tau) requires peak control. For arbitrary non-negative functions, mean control implies no upper bound on the supreme local mass.

When σu(f)≤0\sigma_u(f) \le0, the function t↦f(σ+it)t \mapsto f(\sigma+ it) is almost periodic for each fixed σ>0\sigma> 0, ensuring uσ(t)u_\sigma(t) is almost periodic and thus Lf(σ)<+∞L_f(\sigma) < +\infty for each fixed σ\sigma. The central obstruction lies in whether these bounds remain uniform as σ→0+\sigma\to0^+. A potential pathology arises if Jf(σ)≤CJ_f(\sigma) \le C uniformly for small σ>0\sigma> 0, while simultaneously Lf(σk)→+∞L_f(\sigma_k) \to+\infty along some sequence σk→0+\sigma_k \to0^+. Using the Harper block polynomials, we establish in the next section that this pathology indeed occurs, answering Question 3.3 in the negative.

Finally, we note that finiteness of L(f)L(f) requires a uniform bound across all translation intervals τ∈R\tau\in\mathbb{R}. By contrast, membership in Nloc(C0)N_{\mathrm{loc}}(\mathbb{C}_0) requires logarithmic control only locally near each compact boundary segment. Consequently, while L(f)<+∞L(f) < +\infty implies f∈Nloc(C0)f \in N_{\mathrm{loc}}(\mathbb{C}_0), local membership does not guarantee a uniform constant independent of τ\tau.

Let f∈Hpf \in H^p and define the translation g(s):=f(s+1/2)g(s) := f(s + 1/2). As established, its uniform convergence abscissa satisfies σu(g)≤σu(f)−1/2≤0\sigma_u(g) \le\sigma_u(f) - 1/2 \le0, placing g∈Nug \in\mathcal{N}_u. If the quantitative Local Nevanlinna Embedding Problem (LNEP) held affirmatively, applying (11) to gg would yield

sup⁡ε>0sup⁡τ∈R∫ττ+1log⁡(1+∣f(12+ε+it)∣) dt≤C∥g∥0<+∞.(12)\sup_{\varepsilon>0}\sup_{\tau\in\mathbb{R}}\int_{\tau}^{\tau+1}\log\left(1+\left|f\left(\frac{1}{2}+\varepsilon+it\right)\right|\right)\,dt \le C\|g\|_0 < +\infty. \tag*{(12)}

Condition (11) in Problem 3.1 is strictly stronger than the local logarithmic boundary condition required for membership in Nloc(C1/2)N_{\mathrm{loc}}(\mathbb{C}_{1/2}). Indeed, since any compact interval I⊂RI \subset\mathbb{R} can be covered by finitely many unit intervals [τk,τk+1][\tau_k,\tau_k+1], (3.4) immediately implies

sup⁡0<ε<ε0∫Ilog⁡+∣f(12+ε+it)∣ dt<+∞.\sup_{0<\varepsilon<\varepsilon_0}\int_I\log^+\left|f\left(\frac{1}{2}+\varepsilon+it\right)\right|\,dt < +\infty.

Consequently, an affirmative resolution to Problem 3.1 concerning LNEP would imply an affirmative solution to Saksman and Seip’s Problem 1.2. Crucially, this logical implication holds for all p>0p > 0, not merely p<2p < 2. The contrapositive of this reduction is particularly decisive:

Theorem 3.1. A negative resolution to Problem 1.2 implies a negative resolution to Problem 3.1 (LNEP).

Proof. Suppose f∈Hpf \in\mathcal{H}^p (0<p<20 < p < 2) is the Dirichlet series produced by the gliding-hump construction (Lemma 2.4), associated with sequence εk→0+\varepsilon_k \to0^+ satisfying

∫01log⁡(1+∣f(12+εk+it)∣) dt⟶+∞as k→+∞.\int_0^1 \log\left(1+\left|f\left(\frac{1}{2}+\varepsilon_k+it\right)\right|\right)\,dt \longrightarrow+\infty\quad\text{as } k \to+\infty.

Define the translated function g(s):=f(s+1/2)g(s) := f(s+1/2). Since σu(g)≤σu(f)−1/2≤0\sigma_u(g) \le\sigma_u(f)-1/2 \le0, we have g∈Nug \in\mathcal{N}_u, which guarantees ∥g∥0<+∞\|g\|_0 < +\infty. However,

∫01log⁡(1+∣g(εk+it)∣) dt⟶+∞as k→+∞.\int_0^1 \log(1+|g(\varepsilon_k+it)|)\,dt \longrightarrow+\infty\quad\text{as } k \to+\infty.

Since the interval [0,1][0,1] is a subset of the unit translation interval τ=0\tau=0, taking the supremum over all ε>0\varepsilon>0 and τ∈R\tau\in\mathbb{R} yields

sup⁡ε>0sup⁡τ∈R∫ττ+1log⁡(1+∣g(ε+it)∣) dt=+∞.\sup_{\varepsilon>0}\sup_{\tau\in\mathbb{R}}\int_\tau^{\tau+1}\log(1+|g(\varepsilon+it)|)\,dt=+\infty.

Consequently, no universal constant C>0C>0 can exist satisfying inequality (11) in Problem 3.1, providing a negative answer to LNEP. □

Moreover, the following proof resolves Problem 3.1 in the negative, establishing the failure of the quantitative local Nevanlinna embedding for Nu\mathcal{N}_u.

Proof of failure of Problem 3.1. Choose disjoint pairs of prime numbers (pj,qj)(p_j,q_j) satisfying pj<qj<2pjp_j<q_j<2p_j recursively via Bertrand’s postulate. For each m∈Nm\in\mathbb{N}, define the Dirichlet polynomial

Bm(s)=e−m/2∏j=1m(pj−s+qj−s).(13)B_m(s)=e^{-m/2}\prod_{j=1}^{m}(p_j^{-s}+q_j^{-s}). \tag*{(13)}

Since each BmB_m is a Dirichlet polynomial, its uniform convergence abscissa satisfies σu(Bm)=−∞\sigma_u(B_m)=-\infty, so Bm∈NuB_m\in\mathcal{N}_u.

The Bohr lift of Bm(s)B_m(s) to the compact abelian group T∞\mathbb{T}^{\infty} is given by

B(Bm)(z)=e−m/2∏j=1m(zpj+zqj).\mathcal{B}(B_m)(z)=e^{-m/2}\prod_{j=1}^{m}(z_{p_j}+z_{q_j}).

Applying the elementary bound log⁡(1+x)≤x\log(1+x)\le x for x≥0x\ge0, the global logarithmic norm satisfies

∥Bm∥Nu≤e−m/2∏j=1m∫T2∣z+w∣ dm(z) dm(w)=(4πe)m.(14)\begin{aligned} \|B_m\|_{\mathcal{N}_u} &\le e^{-m/2}\prod_{j=1}^{m}\int_{\mathbb{T}^2}|z+w|\,dm(z)\,dm(w)\\ &=\left(\frac{4}{\pi\sqrt{e}}\right)^m . \tag*{(14)} \end{aligned}

where we have used the standard torus integral evaluation ∫T2∣z+w∣ dm(z) dm(w)=4/π\int_{\mathbb{T}^2}|z+w|\,dm(z)\,dm(w)=4/\pi. Because 4/(πe)<14/(\pi\sqrt{e})<1, the right-hand side of (14) satisfies ∥Bm∥0→0\|B_m\|_0\to0 as m→+∞m\to+\infty.

On the other hand, along the imaginary axis Re⁡(s)=0\operatorname{Re}(s)=0 for 0≤t≤10\leq t\leq1, each factor satisfies

∣pj−it+qj−it∣=2∣cos⁡(t2log⁡qjpj)∣.\left|p_j^{-it}+q_j^{-it}\right|=2\left|\cos\left(\frac{t}{2}\log\frac{q_j}{p_j}\right)\right|.

Since 1<qj/pj<21<q_j/p_j<2, we obtain the lower bound ∣pj−it+qj−it∣≥2cos⁡(log⁡22)≈1.881=:L\left|p_j^{-it}+q_j^{-it}\right|\geq2\cos\left(\frac{\log2}{2}\right)\approx1.881=:L. Consequently, ∣Bm(it)∣≥(L/e)m\left|B_m(it)\right|\geq(L/\sqrt{e})^m uniformly for 0≤t≤10\leq t\leq1, which implies

∫01log⁡(1+∣Bm(it)∣) dt≥log⁡(1+(Le)m)⟶+∞as m⟶+∞.\int_0^1\log(1+\left|B_m(it)\right|)\,\mathrm{d}t\geq\log\left(1+\left(\frac{L}{\sqrt{e}}\right)^m\right)\longrightarrow+\infty\quad\text{as }m\longrightarrow+\infty.

since L/e>1L/\sqrt{e}>1. By continuity, for each m∈Nm\in\mathbb{N} we can select σm>0\sigma_m>0 sufficiently small such that

inf⁡0≤t≤1∣Bm(σm+it)∣≥(L1e)m\inf_{0\leq t\leq1}\left|B_m(\sigma_m+it)\right|\geq\left(\frac{L_1}{\sqrt{e}}\right)^m

for a fixed constant L1L_1 with e<L1<L\sqrt{e}<L_1<L. Thus, as m⟶+∞m\longrightarrow+\infty,

∫01log⁡(1+∣Bm(σm+it)∣) dt⟶+∞,\int_0^1\log(1+\left|B_m(\sigma_m+it)\right|)\,\mathrm{d}t\longrightarrow+\infty,

while ∥Bm∥0⟶0\|B_m\|_0\longrightarrow0. Therefore, there exists no universal constant C>0C>0 such that

sup⁡σ>0sup⁡τ∈R∫ττ+1log⁡(1+∣f(σ+it)∣) dt≤C∥f∥0\sup_{\sigma>0}\sup_{\tau\in\mathbb{R}}\int_\tau^{\tau+1}\log(1+\left|f(\sigma+it)\right|)\,\mathrm{d}t\leq C\|f\|_0

holds for all f∈Nuf\in\mathcal{N}_u, answering Problem 3.1 (LNEP) in the negative. □

Remark 3.1. The underlying chain of logical implications is structured as follows:

LNEP (3.1)⟹Nu⊂Nloc(C0)⟹Hp⊂Nloc(C1/2)⟹Affirmative Problem 1.2.\mathrm{LNEP}\ (3.1)\Longrightarrow\mathcal{N}_u\subset\mathcal{N}_{\mathrm{loc}}(C_0)\Longrightarrow\mathcal{H}^p\subset\mathcal{N}_{\mathrm{loc}}(C_{1/2})\Longrightarrow\text{Affirmative Problem 1.2}.

In general, no reverse implications hold. First, the failure of quantitative LNEP guarantees a sequence (gn)⊂Nu(g_n)\subset\mathcal{N}_u satisfying

sup⁡ε>0sup⁡τ∈R∫ττ+1log⁡(1+∣gn(ε+it)∣) dt∥gn∥Nu⟶+∞,\frac{\sup_{\varepsilon>0}\sup_{\tau\in\mathbb{R}}\int_\tau^{\tau+1}\log(1+\left|g_n(\varepsilon+it)\right|)\,\mathrm{d}t}{\|g_n\|_{\mathcal{N}_u}}\longrightarrow+\infty,

even though each gng_n individually may lie in Nloc(C0)\mathcal{N}_{\mathrm{loc}}(C_0). Second, Saksman and Seip’s question [17] is restricted to Hp\mathcal{H}^p-derived Dirichlet series, whereas LNEP encompasses all of Nu\mathcal{N}_u.

Proof of Theorem 2.1. The Harper polynomial estimates establish that the quantitative local embedding fails for 0<p<20<p<2. However, because each individual Harper polynomial QXQ_X is entire, it individually satisfies the local Nevanlinna embedding condition. To upgrade this quantitative blowup to a genuine qualitative counterexample in Hp\mathcal{H}^p, we deploy a Baire category argument.

Fix the compact interval I=[0,1]I=[0,1]. For m,n∈Nm,n\in\mathbb{N}, define the subset

Em,n:={f∈Hp:sup⁡0<ε<1/n∫01log⁡(1+∣f(12+ε+it)∣) dt≤m}.E_{m,n}:=\left\{f\in\mathcal{H}^p:\sup_{0<\varepsilon<1/n}\int_0^1\log\left(1+\left|f\left(\frac{1}{2}+\varepsilon+it\right)\right|\right)\,\mathrm{d}t\leq m\right\}.

We first claim that each Em,nE_{m,n} is closed in Hp\mathcal{H}^p. Indeed, let (fj)j=1∞⊂Em,n(f_j)_{j=1}^{\infty}\subset E_{m,n} be a sequence converging to ff in Hp\mathcal{H}^p. By Lemma 2.3, point evaluations are bounded on compact subsets of the half-plane Re⁡(s)>1/2\operatorname{Re}(s) > 1/2. Thus fj(1/2+ε+it)→f(1/2+ε+it)f_j(1/2+\varepsilon+it) \to f(1/2+\varepsilon+it) uniformly for t∈[0,1]t \in[0,1] for each fixed ε>0\varepsilon> 0. By Fatou’s lemma (or uniform convergence), the functional

f↦∫01log⁡(1+∣f(12+ε+it)∣) dtf \mapsto\int_0^1 \log\left(1+\left|f\left(\frac{1}{2}+\varepsilon+it\right)\right|\right)\,dt

is continuous on Hp\mathcal{H}^p. Taking the supremum over 0<ε<1/n0 < \varepsilon< 1/n yields a lower semicontinuous functional, which guarantees that Em,nE_{m,n} is closed.

Suppose now, for contradiction, that every f∈Hpf \in\mathcal{H}^p lies locally in Nloc(C1/2)N_{\mathrm{loc}}(\mathbb{C}_{1/2}). By definition, log⁡+∣f∣\log^+ |f| admits a local harmonic majorant near each compact boundary segment. Thus for every f∈Hpf \in\mathcal{H}^p, there exist integers m,n∈Nm,n \in\mathbb{N} such that f∈Em,nf \in E_{m,n}. Consequently, we obtain the countable exhaustion

Hp=⋃m,n=1∞Em,n.\mathcal{H}^p=\bigcup_{m,n=1}^{\infty}E_{m,n}.

Since Hp\mathcal{H}^p is a complete metric space (a Banach space for p≥1p \ge1 and a quasi-Banach F-space for 0<p<10 < p < 1), the Baire Category Theorem asserts that at least one set Em,nE_{m,n} must contain an open ball. That is, there exist g∈Em,ng \in E_{m,n} and r>0r > 0 such that g+h∈Em,ng+h \in E_{m,n} whenever ∥h∥Hp<r\lVert h\rVert_{\mathcal{H}^p} < r. Using the elementary inequality 1+∣h∣≤(1+∣g+h∣)(1+∣g∣)1+|h| \le(1+|g+h|)(1+|g|), taking logarithms yields

log⁡(1+∣h∣)≤log⁡(1+∣g+h∣)+log⁡(1+∣g∣).\log(1+|h|) \le\log(1+|g+h|)+\log(1+|g|).

Integrating over t∈[0,1]t \in[0,1] and using g,g+h∈Em,ng,g+h \in E_{m,n}, we deduce that for all ∥h∥Hp<r\lVert h\rVert_{\mathcal{H}^p}<r,

sup⁡0<ε<1/n∫01log⁡(1+∣h(12+ε+it)∣) dt≤2m.(15)\sup_{0<\varepsilon<1/n}\int_0^1\log\left(1+\left|h\left(\frac{1}{2}+\varepsilon+it\right)\right|\right)\,dt\le2m. \tag*{(15)}

By scaling, choose c>0c>0 sufficiently small so that ∥cf∥Hp<r\lVert cf\rVert_{\mathcal{H}^p}<r whenever ∥f∥Hp≤1\lVert f\rVert_{\mathcal{H}^p}\le1. Substituting h=cfh=cf into (15) yields ∫01log⁡(1+c∣f∣) dt≤2m\int_0^1\log(1+c|f|)\,dt\le2m. Applying the identity log⁡(1+∣f∣)≤log⁡+(1/c)+log⁡(1+c∣f∣)\log(1+|f|)\le\log^+(1/c)+\log(1+c|f|), we obtain

sup⁡∥f∥Hp≤1sup⁡0<ε<1/n∫01log⁡(1+∣f(12+ε+it)∣) dt<+∞.(16)\sup_{\lVert f\rVert_{\mathcal{H}^p}\le1}\sup_{0<\varepsilon<1/n}\int_0^1\log\left(1+\left|f\left(\frac{1}{2}+\varepsilon+it\right)\right|\right)\,dt<+\infty. \tag*{(16)}

Restricting (16) to the subspace of Dirichlet polynomials P⊂Hp\mathcal{P}\subset\mathcal{H}^p and taking the boundary limit ε↓0\varepsilon\downarrow0, we conclude that

sup⁡P∈P∥P∥Hp≤1∫01log⁡(1+∣P(12+it)∣) dt<+∞.(17)\sup_{\substack{P\in\mathcal{P}\\ \lVert P\rVert_{\mathcal{H}^p}\le1}}\int_0^1\log\left(1+\left|P\left(\frac{1}{2}+it\right)\right|\right)\,dt<+\infty. \tag*{(17)}

However, the sequence of normalized Harper polynomials (QX)X≥3⊂P(Q_X)_{X\ge3}\subset\mathcal{P} satisfies ∥QX∥Hp=1\lVert Q_X\rVert_{\mathcal{H}^p}=1 while ∫01log⁡(1+∣QX(1/2+it)∣) dt→+∞\int_0^1\log(1+|Q_X(1/2+it)|)\,dt\to+\infty as X→+∞X\to+\infty (from (8)), directly contradicting (17).

Thus, the assumption Hp⊂Nloc(C1/2)\mathcal{H}^p\subset N_{\mathrm{loc}}(\mathbb{C}_{1/2}) is false. For every 0<p<20<p<2, there exists a function f∈Hpf\in\mathcal{H}^p which fails to belong to Nloc(C1/2)N_{\mathrm{loc}}(\mathbb{C}_{1/2}), resolving Problem 1.2 in the negative.

Answer to Question 3.3. Fix 0<p<20<p<2. Let DX(s):=∑n≤Xn−sD_X(s):=\sum_{n\le X}n^{-s} denote the partial sums of the Riemann zeta function, and let HX:=∥DX∥HpH_X:=\lVert D_X\rVert_{\mathcal{H}^p}. Define the normalized polynomial

PX(s):=DX(s)HX,P_X(s):=\frac{D_X(s)}{H_X},

so that ∥PX∥Hp=1\lVert P_X\rVert_{\mathcal{H}^p}=1. By Harper’s low-moment estimates [13], the norm HXH_X satisfies

HX≳X(1+(1−p/2)log⁡log⁡X)1/2.H_X \gtrsim\frac{\sqrt{X}}{\left(1+(1-p/2)\sqrt{\log\log X}\right)^{1/2}}.

Define the translated function gX(s):=PX(s+1/2)g_X(s):=P_X(s+1/2). Since gXg_X is a Dirichlet polynomial, its uniform convergence abscissa satisfies σu(gX)=−∞\sigma_u(g_X)=-\infty, placing gX∈Nug_X\in\mathcal{N}_u.

We first demonstrate that the global logarithmic norm ∥gX∥0\lVert g_X\rVert_0 is uniformly bounded in XX. Since right translations are contractive on Hp\mathcal{H}^p, we have

lim⁡T→∞12T∫−TT∣gX(σ+it)∣p dt≤∥PX∥Hpp=1\lim_{T\to\infty}\frac{1}{2T}\int_{-T}^{T}|g_X(\sigma+it)|^p\,dt\leq\lVert P_X\rVert_{\mathcal{H}^p}^p=1

for all σ>0\sigma>0.

Applying the elementary inequalities log⁡(1+x)≤log⁡2+log⁡+x≤log⁡2+xp/p\log(1+x)\leq\log2+\log^+x\leq\log2+x^p/p for x≥0x\geq0, we obtain

JgX(σ)≤log⁡2+1p.J_{g_X}(\sigma)\leq\log2+\frac{1}{p}.

Taking the limit superior as σ→0+\sigma\to0^+ yields ∥gX∥0≤log⁡2+1/p<+∞\lVert g_X\rVert_0\leq\log2+1/p<+\infty uniformly for all X≥3X\geq3. We now show that the local short-interval mass L(gX)L(g_X) explodes as X→+∞X\to+\infty. Set σX:=1/log⁡X\sigma_X:=1/\log X. By partial summation, for 0≤t≤10\leq t\leq1 we have

DX(12+σX+it)=X1/2−σX−it1/2−σX−it+O(1).D_X\left(\frac{1}{2}+\sigma_X+it\right)=\frac{X^{1/2-\sigma_X-it}}{1/2-\sigma_X-it}+O(1).

Since X−σX=X−1/log⁡X=e−1X^{-\sigma_X}=X^{-1/\log X}=e^{-1}, evaluating the modulus yields the lower bound

inf⁡0≤t≤1∣gX(σX+it)∣≥Cp(1+(1−p/2)log⁡log⁡X)1/2≥Cp′(log⁡log⁡X)1/4,\inf_{0\leq t\leq1}|g_X(\sigma_X+it)|\geq C_p\left(1+(1-p/2)\sqrt{\log\log X}\right)^{1/2}\geq C'_p(\log\log X)^{1/4},

for some constant Cp′C'_p. Integrating over the short interval [0,1][0,1] gives

L(gX)≥∫01log⁡(1+∣gX(σX+it)∣) dt≥log⁡(1+Cp′(log⁡log⁡X)1/4)=14log⁡log⁡log⁡X+Op(1)⟶+∞as X→+∞.\begin{aligned} L(g_X)&\geq\int_0^1\log(1+|g_X(\sigma_X+it)|)\,dt\\ &\geq\log\left(1+C'_p(\log\log X)^{1/4}\right)\\ &=\frac{1}{4}\log\log\log X+O_p(1)\longrightarrow+\infty\quad\text{as }X\to+\infty. \end{aligned}

Thus, sup⁡XL(gX)=+∞\sup_X L(g_X)=+\infty while sup⁡X∥gX∥0<+∞\sup_X\lVert g_X\rVert_0<+\infty. Consequently, no universal constant C>0C>0 can exist satisfying L(f)≤C∥f∥0L(f)\leq C\lVert f\rVert_0 for all f∈Nuf\in\mathcal{N}_u, establishing the failure of global-to-local logarithmic control and resolving Question 3.3 in the negative. □\square

Remark 3.2. The Harper block sequence (gX)X≥3(g_X)_{X\geq3} provides a simpler and more transparent refutation of the quantitative embedding inequality L(f)≤C∥f∥0L(f)\leq C\lVert f\rVert_0 than the full gliding-hump argument. Indeed, because

sup⁡∥f∥0≤CpL(f)=+∞,\sup_{\lVert f\rVert_0\leq C_p}L(f)=+\infty,

no universal constant C>0C>0 can satisfy L(f)≤C∥f∥0L(f)\leq C\lVert f\rVert_0 across Nu\mathcal{N}_u. However, because each gXg_X is a Dirichlet polynomial, L(gX)<+∞L(g_X)<+\infty for every individual XX. Consequently, the sequence (gX)X≥3(g_X)_{X\geq3} alone does not establish the existence of an element g∈Nug\in\mathcal{N}_u for which L(g)=+∞L(g)=+\infty. To obtain this stronger qualitative statement, the gliding-hump construction is indispensable. By concatenating successively larger, well-separated

Harper blocks into a single Dirichlet series g=∑kakGXkg = \sum_k a_k G_{X_k}, the construction yields a function g∈Nug \in\mathcal{N}_u satisfying ∥g∥0<+∞\lVert g\rVert_0 < +\infty, while simultaneously

∫01log⁡(1+∣g(σk+it)∣) dt⟶+∞as k→+∞\int_0^1 \log\left(1+\left|g\left(\sigma_k+it\right)\right|\right)\,\mathrm{d}t \longrightarrow+\infty\quad\text{as } k \to+\infty

along a sequence σk→0+\sigma_k \to0^+. This provides the definitive qualitative proof that global mean logarithmic control fails to imply uniform short-interval logarithmic control.

Acknowledgment: The second named author is supported by CSIR-UGC Fellowship, New Delhi, India.

Vasudevarao Allu, Department of Mathematics, School of Basic Sciences, Indian Institute of Technology Bhubaneswar, Bhubaneswar-752051, Odisha, India. Email address: [email protected]

Dipon Kumar Mondal, Department of Mathematics, School of Basic Sciences, Indian Institute of Technology Bhubaneswar, Bhubaneswar-752051, Odisha, India. Email address: [email protected]

References

  1. [1]V. Allu and D. K. Mondal, Composition operators and Carlson embeddings for the Nevanlinna class of Dirichlet series, https://arXiv:2609.17133v1, (2026).arxiv.org/abs/2609.17133
  2. [2]F. Bayart and O. F. Brevig, Composition operators and embedding theorems for some function spaces of Dirichlet series, Math. Z. 293 (2019), 989–1004.
  3. [3]A. S. Besicovitch, Almost periodic functions, Dover publications, Inc., New York, (1955).
  4. [5]O. F. Brevig, J. O. Cerdà and K. Seip, Idempotents Fourier multipliers acting contractively on Hp spaces, Geom. Funct. Anal. 31 (2021), 1377–1413.
  5. [6]O. F. Brevig and K. M. Perfekt, A mean counting function for Dirichlet series and compact composition operators, Adv. Math. 385 (2021), paper no. 107775, 48 pp.DOI
  6. [8]B. Chen, X. Fang, F. Guo, S. Hou, Y. Shao and Q. Zhou, The local embedding problem for Hardy spaces of Dirichlet series, https://arXiv:2609.11560v2, (2026).arxiv.org/abs/2609.11560
  7. [9]J. Gördon and H. Hedenmalm, The Composition operators on the space of Dirichlet series with square summable coefficients, Michigan. Math. J. 46(2) (1999), 313–329.DOI
  8. [10]K. Guo, J. Ni and Q. Zhou, Nevanlinna class, Dirichlet series and Szegő’s problem, Ann. Inst. Fourier (Grenoble) 75(4) (2025), 1603–1647.
  9. [11]H. Hedenmalm, P. Lindqvist and K. Seip, A Hilbert space of Dirichlet series and systems of dilated functions in L²(0,1), Duke Math. J. 86(1) (1997), 1–37.DOI
  10. [12]B. Jessen and H. Tornehave, Mean motions and zeroes of almost periodic functions, Acta Math. 77 (1945), 137–279.DOI
  11. [13]A. J. Harper, Moments of random multiplicative functions: I. Low moments, better than squareroot cancellation, and critical multiplicative chaos, Forum Math. Pi 8 (2020), e1, 95.arxiv.org/abs/1703.06654
  12. [14]J. F. Olsen and E. Saksman, On the boundary behaviour of the Hardy spaces of Dirichlet series and a frame bounded estimate, J. Reine Angew. Math. 663 (2012), 33–66.arxiv.org/abs/0907.2708
  13. [15]H. Queffélec and M. Queffélec, Diophantine approximation and Dirichlet series, Texts and Readings in Mathematics, Springer 80 (2020).
  14. [16]E. Saksman and K. Seip, Integral means and boundary limits of Dirichlet series, Bull. London Math. Soc. 41(3) (2009), 411–422.
  15. [17]E. Saksman and K. Seip, Some open questions in analysis for Dirichlet series, in Recent Progress on Operator Theory and Approximation in Spaces of Analytic Functions (C. Bénéteau, A. A. Condori, C. Liaw, W. T. Ross and A. A. Sola, eds.), Contemp. Math. 679, American Mathematical Society, Providence, RI, (2016), 179–191.arxiv.org/abs/1601.01616
  16. [18]J. H. Shapiro and A. L. Shields, Unusual topological properties of the Nevanlinna class, Amer. J. Math. 97(4) (1975), 915–936.DOI

Paper details

vlnc:W_urtfb3zy
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
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