Analogues of Fourier quasicrystals for a strip

Abstract. We study a certain family of discrete measures with unit masses on a horizontal strip as an analogue of Fourier quasicrystals on the real line. We prove a one-to-one correspondence between supports of measures from this family and zero sets of exponential polynomials with imaginary frequencies. This result is the special case of a general result on measures whose supports correspond to zero sets of absolutely convergent Dirichlet series with bounded spectrum.

AMS Mathematics Subject Classification: 30B50, 32A60, 52C23

Keywords: Fourier quasicrystal, almost periodic set, zero set, exponential polynomial, Dirichlet series

1. Introduction

A measure $\mu$ with locally finite support (that is, its introsection with any compact set is finite) is called a Fourier quasicrystal if $\mu$ is a temperate distribution, its Fourier transform in sense of distributions $\hat{\mu}$ is also a measure with locally finite support, and both measures $|\mu|$, $|\hat{\mu}|$ are temperate distributions. Here and below $|\nu|(E)$ is the variation of the complex measure $\nu$ on the set $E$.

In [14], [15] A.Olevskii and A.Ulanovskii proved that a measure $\mu_{A}$ of the form

(1) $$\mu_{A}=\sum_{n}\delta_{a_{n}},\quad A=\{a_{n}\}\subset{\mathbb{R}},$$

is a Fourier quasicrystal if and only if its support $A=\{a_{n}\}$ is a zero set of an exponential polynomial

(2) $$P(z)=\sum_{1\leq j\leq N}q_{j}e^{2\pi i\omega_{j}z},\qquad q_{j}\in{\mathbb{C}},\quad\omega_{j}\in{\mathbb{R}}.$$

Here $\delta_{x}$ means the unit mass at the point $x$. We suppose also that $A=\{a_{n}\}$ is a multiset, where every point $a_{n}$ has a finite multiplicity in $A$.

Fourier quasicrystals are currently being studied very actively. Many works are devoted to the study of the properties of Fourier quasicrystals, see, for example, the collections of papers [2], [16], in particular, the basic work [12]. Fourier quasicrystals find application in modern physics, since they are used as mathematical models of certain atomic structures. A nontrivial example of a Fourier quasicrystal of the form (1), which is not a finite union of shifts of an arithmetical progression, was found by P.Kurasov and P.Sarnak [10]. Note also the paper [8], which actually considers pairs $(\mu,\hat{\mu})$.

In [5], [6] we studied zero sets $A=\{a_{n}\}\subset{\mathbb{R}}$ of absolutely convergent Dirichlet series

(3) $$Q(z)=\sum_{j}q_{j}e^{2\pi i\omega_{j}z},\qquad q_{j}\in{\mathbb{C}},\quad\omega_{j}\in{\mathbb{R}},\quad\sum_{j}|q_{j}|<\infty,$$

with bounded spectra $\Omega=\{\omega_{n}\}$. Such sums are natural extensions of the finite sums (2). We proved that the Fourier transform of the corresponding measure $\mu_{A}$ is also a measure of the form

(4) $$\hat{\mu}_{A}=\sum_{\gamma\in\Gamma}b_{\gamma}\delta_{\gamma},\quad\Gamma\subset{\mathbb{R}}\ \text{ is countable}.$$

Besides, we found a necessary and sufficient condition on $\hat{\mu}_{A}$ for the set $A$ to be the zero set of some Dirichlet series (3).

The zero sets $A=\{a_{n}\}$ of exponential polynomials (2) and Dirichlet series (3) with zeros $a_{n}$ either on the real line or on a horizontal strip of finite width is an almost periodic set in the sense of M.Krein and B.Levin [11, App.VI]. In particular, $A$ is translation bounded, i.e., quantities

$$\#\{n:\,t<\mbox{Re }a_{n}<t+1\}$$

are bounded uniformly in $t\in{\mathbb{R}}$ (see Section 3). Also, the measure $|\hat{\mu}|$ on ${\mathbb{R}}$ is temperate if and only if the variation $|\hat{\mu}|(-r,r)$ grows polynomially (see [4]). Have this in mind, we can reformulate the above result of Olevskii and Ulanovskii in the following way:

A locally finite set $A=\{a_{n}\}\subset{\mathbb{R}}$ is the zero set of an exponential polynomial (2) if and only if $A$ is almost periodic, $\hat{\mu}_{A}$ has the form (4) with locally finite support $\{\gamma\in{\mathbb{R}}:\,b_{\gamma}\neq 0\}$, and

$$\log\sum_{|\gamma|<r}|b_{\gamma}|=O(\log r).$$

In the present paper we consider measures of the form (1), but we replace the condition $A\subset{\mathbb{R}}$ with $A\subset\mathcal{S}_{H}:=\{z=x+iy:\,|y|\leq H\}$. We slightly change the definition of $\hat{\mu}_{A}$ and prove that the above results on zeros of exponential sums and Dirichlet series are valid in this case as well.

Theorem 1.

Let $A=\{a_{n}\}\subset\mathcal{S}_{H}$ be the zero set of some Dirichlet series (3) and

(5) $$\mu_{A}=\sum_{n}\delta_{a_{n}}.$$

Then $A$ is an almost periodic set and $\hat{\mu}_{A}$ is a pure point measure (4) such that the coefficients $b_{\gamma}$ satisfy conditions

(6) $$\log\sum_{|\gamma|<r}|b_{\gamma}|=O(r)\quad\text{as}\quad r\to\infty,$$

and

(7) $$\sum_{0<|\gamma|<1}|b_{\gamma}/\gamma|<\infty.$$

If $A$ is the zero set of an exponential polynomial (2), then $\Gamma$ is locally finite.

Theorem 2.

Let $A=\{a_{n}\}\subset\mathcal{S}_{H}$ be an almost periodic set, the measure $\mu_{A}$ define in (5), and the Fourier transform $\hat{\mu}_{A}$ be a pure point measure (4) with coefficients $b_{\gamma}$ which satisfy (6) and (7). Then there is the Dirichlet series (3) with bounded spectrum and the zero set $A$. If, in addition, $\Gamma$ is locally finite (in which case the condition (7) is satisfied automatically), then the Dirichlet series is an exponential polynomial (2).

Remark. In both theorems the condition ”$A$ is almost periodic” can be replaced with ”$A$ is translation bounded” (see Section 3).

The article is structured as follows.

In Section 2 we give auxiliary results on Fourier transform and Dirichlet series.

In Section 3 we present some results on almost periodic measures and sets.

In Section 4 we consider entire functions of exponential growth with an almost periodic zero set and prove some formulas for their logarithmic derivatives.

In Sections 5 and 6 we prove Theorem 1 and 2, respectively.

2. Preliminary results

Fourier transform

By $S({\mathbb{R}})$ denote the Schwartz space of functions $\varphi\in C^{\infty}({\mathbb{R}})$ with the finite norms

$$N_{n,m}(\varphi)=\sup_{{\mathbb{R}}}\max_{k\leq m}|(1+|x|^{n})\varphi^{(k)}(x)|,\quad n,m=0,1,2,\dots$$

These norms generate the topology of the projective limit on $S({\mathbb{R}})$. The set of $C^{\infty}$-functions with compact supports is dense in $S({\mathbb{R}})$.

The Fourier transform is defined for $\varphi\in S({\mathbb{R}})$ (in general for $\varphi\in L^{1}({\mathbb{R}})$) by the equality

$$\hat{\varphi}(x)=\int_{{\mathbb{R}}}\varphi(t)e^{-2\pi ixt}dt.$$

If $\varphi$ satisfies the estimate

(8) $$|\varphi(t)|\leq Ce^{-H^{\prime}|t|},\quad H^{\prime}>H,$$

then

(9) $$\hat{\varphi}(z)=\int_{{\mathbb{R}}}\varphi(t)e^{-2\pi izt}dt$$

is the holomorphic extension of $\hat{\varphi}$ to a neighborhood of $\mathcal{S}_{H}$. If $\varphi\in S({\mathbb{R}})$ has compact support and $\phi\in L^{1}({\mathbb{R}})$ satisfies (8), then for $z\in\mathcal{S}_{H}$ we have

(10) $$\widehat{\varphi\star\phi}(z)=\hat{\varphi}(z)\hat{\phi}(z).$$

The following formula for the inverse Fourier transform

(11) $$\int_{\mathbb{R}}\hat{\varphi}(x+iy)e^{2\pi ixt}dx=\varphi(t)e^{2\pi yt},\quad z=x+iy\in\mathcal{S}_{H},$$

holds for $\varphi\in S({\mathbb{R}})$ satisfying (8).

The Fourier transform of a measure $\mu$ with support in $\mathcal{S}_{H}$ is defined by setting

$$(\hat{\mu},\varphi)=(\mu,\hat{\varphi})$$

for all $\varphi\in S({\mathbb{R}})$ satisfying (8). In the general case $\hat{\mu}$ is a distribution.

The Fourier transform is a bijection of $S({\mathbb{R}})$ onto itself, hence $\hat{\mu}$ is well-defined.

Dirichlet series on ${\mathbb{R}}$

Denote by $\mathcal{H}$ the algebra of all Dirichlet series

$$Q(x)=\sum_{n}q_{n}e^{2\pi i\omega_{n}x},\,\omega_{n}\in{\mathbb{R}},$$

with finite Wiener’s norm $\|Q\|_{W}=\sum_{n}|q_{n}|$. Spectrum of $Q\in\mathcal{H}$ is the set $\operatorname{sp}Q=\{\omega_{n}:\,q_{n}\neq 0\}$. For each $Q\in\mathcal{H}$ and each analytic function $h(z)$ on a neighborhood of the set $\overline{\{Q(x):\,x\in{\mathbb{R}}\}}$ we have $h(Q(x))\in\mathcal{H}$ (see [18], Ch.VI). In particular, for $Q\in\mathcal{H}$ we have $\exp Q\in\mathcal{H}$; if $Q\in\mathcal{H}$ and $\inf_{\mathbb{R}}|Q(x)|>0$, then $1/Q\in\mathcal{H}$.

We conclude this section with a useful lemma.

Lemma.

If $Q\in\mathcal{H}$ and has nonnegative spectrum, then $\exp Q\in\mathcal{H}$ also has nonnegative spectrum; if, in addition, the spectrum of $Q$ is locally finite, the same is valid for the spectrum of $\exp Q$.

If $Q\in\mathcal{H}$, spectrum $Q$ is strictly positive, and $\sup_{\mathbb{R}}|Q(x)|<1$, then $\log(1+Q)\in\mathcal{H}$ and has strictly positive spectrum; if, in addition, the spectrum of $Q$ is locally finite, the same is valid for the spectrum of $\log(1+Q)$.

Proof. Let $\operatorname{sp}Q\subset[0,+\infty)$ and $r<\infty$. Clearly, $\operatorname{sp}Q^{n}\subset[0,+\infty)$. Equality

$$\exp{Q(x)}=\sum_{n=0}^{\infty}Q^{n}(x)/n!$$

implies that $\operatorname{sp}\exp Q\subset[0,+\infty)$ and

(12) $$\operatorname{sp}\exp Q\cap[0,r)\subset\cup_{n}\operatorname{sp}Q^{n}\cap[0,r).$$

Now let $\operatorname{sp}Q$ be locally finite. Fix $r<\infty$. Then $\operatorname{sp}Q^{n}\cap(0,r)$ is a subset of various sums of at most $n$ numbers from $\operatorname{sp}Q$ (numbers can be repeated). Since these sums contain no number from $\operatorname{sp}Q\cap[r,\infty)$, we see that the family of such sums is finite. On the other hand, let $\omega^{\prime}=\min\{\operatorname{sp}Q\setminus\{0\}\}$. If $N>r/\omega^{\prime}$, then $\operatorname{sp}Q^{p}\cap(0,r)$ for all $p$ contains no sums of $n>N$ numbers from $\operatorname{sp}Q\setminus\{0\}$. It follows from (12) that $\operatorname{sp}\exp Q$ is locally finite. For the function $\log(1+P)$ we use the representation

$$\phantom{XXXXXXXXXXXXXX}\log(1+P(x))=\sum_{n=1}^{\infty}(-1)^{n}P^{n}(x)/n.\phantom{XXXXXXXXXXXXXXXX}\leavevmode\nobreak\ \hfill\rule{7.0pt}{7.0pt}$$

3. Almost periodic measures and sets on a strip

Definition 1 (cf.[1], App.II). A continuous function $g(z)$ on the open strip

$$S_{(\alpha,\beta)}=\{z=x+iy\in{\mathbb{C}}:\,-\infty\leq\alpha<y<\beta\leq+\infty\}$$

is almost periodic if for any $\eta\in(0,(\beta-\alpha)/2)$ and $\varepsilon>0$ the set of $\varepsilon$-almost periods

$$E_{\eta,\varepsilon}=\{\tau\in{\mathbb{R}}:\,\sup_{x\in{\mathbb{R}},\alpha+\eta\leq y\leq\beta-\eta}|g(x+\tau+iy)-g(x+iy)|<\varepsilon\}$$

is relatively dense, i.e., $E_{\eta,\varepsilon}\cap(x,x+L)\neq\emptyset$ for all $x\in{\mathbb{R}}$ and some $L$ depending on $\varepsilon$ and $\eta$.

Note that Dirichlet series (3) extends as an entire function which is almost periodic on the whole plane ${\mathbb{C}}=S_{(-\infty,+\infty)}$.

Definition 2 ([17]). A measure $\mu$ is almost periodic on the open strip $S_{(\alpha,\beta)}$ if for any $\eta>0$ and any continuous function $\varphi(z)$ with compact support such that $\varphi(x+iy)=0$ for $|y|>\eta$ the convolution $\varphi\star\mu(z)=\int\varphi(z-w)\mu(dw)$ is an almost periodic function on the substrip $\{z\in{\mathbb{C}}:\,\alpha+\eta<y<\beta-\eta\}$.

In our investigation we consider only almost periodic measures with supports in closed horizontal strips of finite width. In this case the definition of almost periodic measures can be simplified:

Definition 3([17]). A measure $\mu$ with support in the closed strip $S_{[\alpha,\beta]}=\{z=x+iy:-\infty<\alpha\leq y\leq\beta<\infty\}$ is almost periodic if for any continuous function $\varphi(z)$ with compact support the convolution $\varphi\star\mu(z)=\int\varphi(z-w)\mu(dw)$ is an almost periodic function on ${\mathbb{C}}$.

Definition 4. A measure $\mu$ on a closed horizontal strip $S_{[\alpha,\beta]}$ is translation bounded if

$$\sup_{t\in{\mathbb{R}}}|\mu|\{z\in S_{[a,b]}:\,t<\mbox{Re }z<t+1\}<\infty.$$

Clearly, a locally finite multiset $A=\{a_{n}\}$ is translation bounded iff the measure $\mu_{A}=\sum_{n}\delta_{a_{n}}$ is translation bounded.

Proposition 1.

Every almost periodic complex measure $\mu$ on $S_{[a,b]}$ is translation bounded.

This proposition was proved in [17] for the wider class of almost periodic distributions. Here we are only interested complex measures on a closed strip. For convenience of riders, here we give its proof, which is very simple in this case.

Proof. Let $\varphi$ be any function from the Banach space $X$ of all continuous functions $\phi$ on $[0,1]\times[a,b]$ such that $\phi(0+iy)=\phi(1+iy)=0$ for all $y\in[a,\,b]$. The function

$$(\varphi\star\mu)(z)=\int_{\zeta\in S_{[a,b]}}\varphi(\zeta)\mu_{z}(d\zeta)=(\mu_{z},\varphi)\quad\text{with}\quad\mu_{z}(\cdot)=\mu(z-\cdot)$$

is almost periodic in ${\mathbb{C}}$, hence it is uniformly bounded on $\mathcal{S}_{H}$, and for all fix $z\in\mathcal{S}_{H}$ actions $(\mu_{z},\varphi)$ are bounded on $X$. Applying Banach–Steinhaus Theorem, we obtain the assertion of Proposition 1.    

The converse assertion for $\operatorname{supp}\mu\subset{\mathbb{R}}$ and for a pure point measure $\hat{\mu}$ was proved by Y.Meyer in [13], Lemma 5.8; see also [6]. The proof can be transferred without changes to measures on closed strips of bounded width.

Proposition 2.

Let a complex measure $\mu$ be translation bounded on $S_{[a,b]}$ and its Fourier transform $\hat{\mu}$ be a pure point measure. Then $\mu$ is an almost periodic measure.

Propositions 1 and 2 explain the Remark after Theorem 2.

Definition 5 (see [7]). A locally finite set $A=\{a_{n}\}\subset S_{[\alpha,\beta]}$ (or $\subset S_{(\alpha,\beta)}$) is almost periodic if the measure $\mu_{A}=\sum_{n}\delta_{a_{n}}$ is almost periodic on the same strip. If $\alpha=\beta=0$ we obtain the definition of almost periodic sets in ${\mathbb{R}}$.

The original definition, due to M.Krein and B.Levin [11, App. VI], follows:

Definition 6. A locally finite set $A=\{a_{n}\}_{n\in{\mathbb{Z}}}\subset S_{[\alpha,\beta]}$ is almost periodic if for any $\varepsilon>0$ the set of its $\varepsilon$-almost periods

$$E_{\varepsilon}=\{\tau\in{\mathbb{R}}:\,\exists\ \text{ a bijection }\sigma:{\mathbb{Z}}\to{\mathbb{Z}}\quad\text{such that}\ \sup_{n}|a_{n}+\tau-a_{\sigma(n)}|<\varepsilon\}$$

has a nonempty intersection with every interval $(x,x+L_{\varepsilon})$.

The generalization of this definition to almost periodic sets in an open strip, in particular in ${\mathbb{C}}$, due to H.Tornehave [19]. In [7] it was proved that Definition 6 and Tornhave’s one are equivalent to Definition 5.

Note that the zero set of any holomorphic almost periodic function on an open strip is an almost periodic set in this strip. The converse is not true; the connection between almost periodic sets and zeros of holomorphic almost periodic functions is rather complicated. A complete description is given in [3] in terms of Chern cohomologies. But in our article we will deal only with sets that a priori lie in a horizontal strip of finite width. In this case any almost periodic set is the zero set of an entire almost periodic function [7].

The zero set of Dirichlet series (3) is almost periodic; it lies in a horizontal strip of finite width if and only if $\sup$ and $\inf$ of spectrum $Q$ are bounded and belong to this spectrum (see [11], Ch.VI, Con.2). In particular, the zero set of each exponential polynomial (2) lies in a strip of this type.

Clearly, for every almost periodic set $A=\{a_{n}\}_{n\in{\mathbb{Z}}}\subset\mathcal{S}_{H}$ the set $A^{\prime}=\{\mbox{Re }a_{n}\}$ is almost periodic in ${\mathbb{R}}$. Hence Theorem 1 from [6] implies the following proposition:

Proposition 3.

For every almost periodic set $A=\{a_{n}\}_{n\in{\mathbb{Z}}}\subset\mathcal{S}_{H}$ under condition $\mbox{Re }a_{n}\leq\mbox{Re }a_{n+1}$ there is a bounded mapping $\phi:\,Z\to{\mathbb{C}}$ such that

$$a_{n}=n/d+\phi(n)$$

with a density $d>0$.

4. Entire functions with almost periodic zeros

By Hadamard’s Theorem, every entire function $g(z),\,z\in{\mathbb{C}}$, of exponential growth (i.e., $\log|g(z)|\leq O(|z|)$ as $|z|\to\infty$) with zeros $a_{n}\in{\mathbb{C}}\setminus\{0\}$ has the form

(13) $$g(z)=e^{Lz}\prod_{n}(1-z/a_{n})e^{z/a_{n}},\quad L\in{\mathbb{C}}.$$

Note that each $a\in{\mathbb{C}}$ can be repeated any finite number of times in the sequence $\{a_{n}\}$.

Further, zeros of $g$ satisfy the conditions

(14) $$\#\{n:\,|a_{n}|\leq r\}=O(r)\quad\text{as}\quad r\to\infty$$

and

(15) $$\sum_{n:\,|a_{n}|\leq r,a_{n}\neq 0}\frac{1}{a_{n}}=O(1)\quad\text{as}\quad r\to\infty.$$

On the other hand, if a sequence $A=\{a_{n}\}_{n\in{\mathbb{Z}}}\subset{\mathbb{C}}\setminus\{0\}$ satisfies these conditions, then the function (13) is an entire function of exponential growth (Lindelöf’s Theorem (see [9])).

Proposition 4.

Let $A=\{a_{n}\}_{n\in{\mathbb{Z}}}\subset\mathcal{S}_{H}\setminus\{0\}$ be an almost periodic set. Then the points $a_{n}$ satisfy conditions (14), (15), and the infinite product

(16) $$f(z)=(1-z/a_{0})\prod_{n\in{\mathbb{N}}}(1-z/a_{n})(1-z/a_{-n})$$

converges uniformly on compact sets in ${\mathbb{C}}$. Moreover, the sum

(17) $$\frac{f^{\prime}(z)}{f(z)}=\frac{1}{z-a_{0}}+\sum_{n\in{\mathbb{N}}}\left[\frac{1}{z-a_{n}}+\frac{1}{z-a_{-n}}\right]=\\ \frac{1}{x+iy-\phi(0)}+\sum_{n\in{\mathbb{N}}}\left[\frac{1}{x+iy-n/d-\phi(n)}+\frac{1}{x+iy+n/d-\phi(-n)}\right]$$

converges absolutely and uniformly for $x$ from any bounded set and $|y|>M+1$ with $M=\sup_{n\in{\mathbb{Z}}}|\phi(n)|$.

Proof. By Proposition 3, we can renumber $a_{n}$ such that $a_{n}=n/d+\phi(n)$, $n\in{\mathbb{Z}}$, with a bounded complex-valued function $\phi(n)$. Therefore the sum

(18) $$\sum_{n\in{\mathbb{N}}}\left[\frac{1}{a_{n}}+\frac{1}{a_{-n}}\right]=\sum_{n\in{\mathbb{N}}}\left[\frac{1}{n/d+\phi(n)}+\frac{1}{-n/d+\phi(-n)}\right]$$

converges absolutely, and the sum

$$\sum_{n\in{\mathbb{N}}}\left[\log(1-z/a_{n})+\log(1-z/a_{-n})\right]$$

converges uniformly on compact subsets of ${\mathbb{C}}$. The sums

$$\sum_{n\in{\mathbb{Z}},|a_{n}|<r}\frac{1}{a_{n}}\quad\mbox{and}\quad\sum_{n\in{\mathbb{Z}},0<|n/d|<r}\frac{1}{a_{n}}$$

differ for a uniformly bounded (with respect to $r$) number of terms, and each of these terms tends to $0$ as $r\to\infty$. Therefore the first sum has a finite limit as $r\to\infty$, and we obtain representation (16) and condition (15). Condition (14) is evident.

Since the sum

(19) $$\sum_{|n/d|<|x|+M+1}\left|\frac{1}{x+iy-n/d-\phi(n)}\right|+\sum_{n/d\geq|x|+M+1}\left|\frac{1}{x+iy-n/d-\phi(n)}+\frac{1}{x+iy+n/d-\phi(-n)}\right|$$

is uniformly bounded for $|x|\leq C$ and $|y|>M+1$, we obtain (17).    

Proposition 5.

Let $A$ be an almost periodic set in $\mathcal{S}_{H}\setminus\{0\}$, and the distribution $\hat{\mu}_{A}$ defined in Section 2 is a pure point measure (4) such that

(20) $$\sum_{|\gamma|<r}|b_{\gamma}|<Ce^{2\pi Kr}$$

with a constant $K<\infty$. Then

(21) $$f^{\prime}(\zeta)/f(\zeta)=-2\pi i\sum_{\gamma\in\Gamma\cap(0,+\infty)}b_{\gamma}e^{2\pi i\gamma\zeta}-\pi ib_{0},$$

and the sum in the right-hand side absolutely converge on every horizontal line with $\mbox{Im }\zeta>\max\{H,K\}$.

Proof. Fix $\zeta\in{\mathbb{C}},\,\mbox{Im }\zeta>H$. Set $e^{+}_{\zeta}(t)=-2\pi ie^{2\pi it\zeta}$ for $t>0$ and $e^{+}_{\zeta}(t)=0$ for $t\leq 0$. Clearly, its Fourier transform

$$\hat{e}^{+}_{\zeta}(z)=-\int_{0}^{\infty}2\pi ie^{2\pi it\zeta}e^{-2\pi izt}dt=1/(\zeta-z)$$

is defined for $z\in\mathcal{S}_{H}$. Let $\varphi(t)$ be any even nonnegative $C^{\infty}$-function such that $\operatorname{supp}\varphi\subset(-1,1)$ and $\int\varphi(t)dt=1$. Set $\varphi_{\varepsilon}(t)=\varepsilon^{-1}\varphi(t/\varepsilon)$ for $\varepsilon>0$, The function $e^{+}_{\zeta}\star\varphi_{\varepsilon}$ satisfies (8), therefore (10) yields

$$(\hat{\mu}_{A},e^{+}_{\zeta}\star\varphi_{\varepsilon})(t))=(\mu_{A},\hat{e}^{+}_{\zeta}(z)\hat{\varphi}_{\varepsilon}(z)).$$

We have

$$(\mu_{A}(z),\hat{e}^{+}_{\zeta}(z)\hat{\varphi}_{\varepsilon}(z))=\frac{\hat{\varphi}(\varepsilon a_{0})}{\zeta-a_{0}}+\sum_{n\in{\mathbb{N}}}\left[\frac{\hat{\varphi}(\varepsilon a_{n})}{\zeta-a_{n}}+\frac{\hat{\varphi}(\varepsilon a_{-n})}{\zeta-a_{-n}}\right].$$

Clearly, $|\hat{\varphi}_{\varepsilon}(z)|\leq e^{2\pi H}$ and $\hat{\varphi}_{\varepsilon}(z)\to 1$ as $\varepsilon\to 0$. Taking into account (19), we can apply Lebesgue’s Dominate Convergence Theorem. Hence,

(22) $$(\mu_{A}(z),\hat{e}^{+}_{\zeta}(z)\hat{\varphi}_{\varepsilon}(z))\to\frac{1}{\zeta-a_{0}}+\sum_{n\in{\mathbb{N}}}\left[\frac{1}{\zeta-a_{n}}+\frac{1}{\zeta-a_{-n}}\right]\qquad(\varepsilon\to 0).$$

Set

$$n(s)=\sum_{\gamma\in\Gamma:\,0<\gamma<s}|b_{\gamma}|.$$

By (20), $n(s)\leq Ce^{2\pi Ks}$. We obtain for $y>K$

(23) $$\sum_{\gamma\geq r}|b_{\gamma}|e^{-2\pi\gamma y}=\int_{r}^{\infty}e^{-2\pi sy}n(ds)\leq\lim_{T\to\infty}n(T)e^{-2\pi Ty}+2\pi y\int_{r}^{\infty}e^{-2\pi sy}n(s)ds<\infty.$$

Therefore the series $\sum_{\gamma>0}b_{\gamma}e^{2\pi i\gamma\zeta}$ absolutely and uniformly converges on the line $\eta=\mbox{Im }\zeta=\operatorname{const}>K$.

Further, we have for $\varepsilon>0$

(24) $$\frac{i}{2\pi}(\hat{\mu}_{A}(t),e^{+}_{\zeta}\star\varphi_{\varepsilon}(t))=\sum_{\gamma\in\Gamma}b_{\gamma}e^{2\pi i\gamma\zeta}\int_{-\varepsilon}^{\gamma}e^{-2\pi is\zeta}\varphi_{\varepsilon}(s)ds.$$

It is easy to see that no integral in the right-hand side exceeds $e^{2\pi\eta\varepsilon}$, and for $\varepsilon\to 0$

$$\int_{-\varepsilon}^{\gamma}e^{-2\pi is\zeta}\varphi_{\varepsilon}(s)ds=\int_{-1}^{\min\{\gamma/\varepsilon,1\}}e^{-2\pi i\varepsilon s\zeta}\varphi(s)ds\to\begin{cases}1,&\text{if }\gamma>0,\\ 1/2,&\text{if }\gamma=0,\end{cases}$$

$$\int_{-\varepsilon}^{\gamma}e^{-2\pi is\zeta}\varphi_{\varepsilon}(s)ds=\int_{-1}^{\max\{\gamma/\varepsilon,-1\}}e^{-2\pi i\varepsilon s\zeta}\varphi(s)ds\to 0,\quad\text{if }\gamma<0.$$

Using Lebesgue’s Dominate Convergence Theorem, we obtain from (24)

$$(\hat{\mu}_{A}(t),e^{+}_{\zeta}\star\varphi_{\varepsilon}(t))\to-2\pi i\sum_{\gamma\in\Gamma\cap(0,+\infty)}b_{\gamma}e^{2\pi i\gamma\zeta}-\pi ib_{0}\qquad(\varepsilon\to 0).$$

(17) and (22) yield (21).    

5. Proof of Theorem 1

Since the zero set $A$ of the Dirichlet series (3) lies in a strip of finite width, we get $\sup\Gamma\in\Gamma$ and $\inf\Gamma\in\Gamma$. Without lose of generality suppose $\sup\Gamma=\kappa,\,\inf\Gamma=-\kappa$. We have $q_{-\kappa}\neq 0$. Then

$$Q(z)=q_{-\kappa}e^{-2\pi i\kappa z}(1+H(z)),\qquad H(z)=\sum_{\omega\in\Gamma\setminus\{-\kappa\}}\frac{q_{\omega}}{q_{-\kappa}e^{2\pi(\kappa+\omega)y}}e^{2\pi i(\kappa+\omega)x},\quad z=x+iy.$$

Taking into account that $\sum_{\omega}|q_{\omega}|<\infty$, we choose a finite number of elements $\omega_{1},\dots,\omega_{N}\in\Gamma\setminus\{-\kappa\}$ and then $s^{*}>H$ such that

$$\sum_{\omega\in\Gamma\setminus\{-\kappa,\omega_{1},\dots,\omega_{N}\}}|q_{\omega}/q_{-\kappa}|<1/3,\qquad\sum_{j=1}^{N}e^{-2\pi(\omega_{j}+\kappa)s^{*}}|q_{\omega_{j}}/q_{-\kappa}|<1/3.$$

So $\sup_{\mathbb{R}}|H(x+is^{*})|<2/3$, and by Lemma,

$$\log(1+H(x+is^{*}))\in\mathcal{H}$$

and

(25) $$\log(1+H(x+is^{*}))=\sum_{\gamma\in\Gamma^{*}}p_{\gamma}e^{2\pi i\gamma x}\quad\text{with}\quad\Gamma^{*}\subset(0,+\infty),\quad\sum_{o\in\Gamma^{*}}|p_{\gamma}|<\infty.$$

Therefore,

(26) $$Q^{\prime}(x+is^{*})/Q(x+is^{*})=\sum_{\gamma\in\Gamma^{*}}2\pi i\gamma p_{\gamma}e^{2\pi i\gamma x}-2\pi i\kappa.$$

Since $Q$ is an almost periodic function, we can apply Lemma 1 from Ch.6, [11] and for any $\varepsilon>0$ and $s<\infty$ find a number $m=m(\varepsilon,s)>0$ such that

(27) $$|Q(z)|\geq m\quad\mbox{for}\quad|\mbox{Im }z|\leq s\quad\mbox{and}\quad\operatorname{dist}(z,A)\geq\varepsilon.$$

Hence,

$$\inf_{x\in{\mathbb{R}}}|Q(x\pm is)|>0$$

for every $s>H$, and $1/Q\in\mathcal{H}$. Since $\Gamma$ is bounded, we get $Q^{\prime}(x+is^{*})\in\mathcal{H}$ and $(Q^{\prime}/Q)(x+is^{*})\in\mathcal{H}$. Hence,

(28) $$\sum_{o\in\Gamma^{*}}|\gamma p_{\gamma}|<\infty.$$

The same arguments show that for some countable set $\Gamma_{*}\subset(-\infty,0)$ and $s_{*}$ large enough

(29) $$Q^{\prime}(x-is_{*})/Q(x-is_{*})=\sum_{\gamma\in\Gamma_{*}}2\pi i\gamma\tilde{p}_{\gamma}e^{2\pi i\gamma x}+2\pi i\kappa,\qquad\sum_{\gamma\in\Gamma_{*}}|\tilde{p}_{\gamma}|<\infty,\qquad\sum_{\gamma\in\Gamma_{*}}|\gamma\tilde{p}_{\gamma}|<\infty.$$

In what follows we set $s=\max\{s^{*},s_{*}\}$.

Since $Q$ is an almost periodic function, we see that its zero set $A\subset\mathcal{S}_{H}$ is almost periodic. Therefore by Proposition 1, the numbers $\#\{a_{n}\in A:\,x<\mbox{Re }a_{n}<x+1\}$ are bounded uniformly in $x\in{\mathbb{R}}$. Consequently, for $\varepsilon$ small enough every interval $(x,x+1)$ contains a number $L$ such that $\operatorname{dist}(z,A)\geq\varepsilon$ for all points of the segment $[L-iH,L+iH]$. By (27), there exist $m=m(s,\varepsilon)>0$ and two sequences $L_{k}\to+\infty,\,L^{\prime}_{k}\to-\infty$ such that

$$|Q(x+iy)|\geq m>0\quad\text{for}\quad x=L_{k}\quad\text{or}\quad x=L^{\prime}_{k},\quad|y|\leq s.$$

Let $\varphi\in\ S({\mathbb{R}})$ satisfy (8) with $H^{\prime}>s$. Its Fourier transform (9) is holomorphic in $\mathcal{S}_{H}$ and $|x|^{k}\hat{\varphi}(x+iy)\to 0$ as $|x|\to\infty$ for every $k>0$ uniformly with respect to $|y|\leq s$. Therefore, integrals of the function $\hat{\varphi}(z)Q^{\prime}(z)Q^{-1}(z)$ over boundaries of the rectangles $\{z:\,L^{\prime}_{k}<x<L_{k},-s<y<s\}$ tend to

(30) $$\int_{-\infty}^{+\infty}\hat{\varphi}(x-is)Q^{\prime}(x-is)Q^{-1}(x-is)dx-\int_{-\infty}^{+\infty}\hat{\varphi}(x+is)Q^{\prime}(x+is)Q^{-1}(x+is)dx=:I_{1}-I_{2}.$$

Using the Theorem on residues, we get

(31) $$\frac{I_{1}-I_{2}}{2\pi i}=\sum_{\lambda:Q(\lambda)=0}\operatorname{Res}_{\lambda}\hat{\varphi}(z)\frac{Q^{\prime}(z)}{Q(z)}=\sum_{\lambda:Q(\lambda)=0}a(\lambda)\hat{\varphi}(\lambda)=(\mu_{A},\hat{\varphi}),$$

where $a(\lambda)$ is the multiplicity of the zero $Q$ at the point $\lambda$. Change the order of integration and summation here. Using (26), (29) and the equality (11) for the inverse Fourier transform, we obtain

(32) $$\frac{I_{1}-I_{2}}{2\pi i}=\sum_{\gamma\in\Gamma_{*}}\gamma\tilde{p}_{\gamma}\int_{-\infty}^{+\infty}\hat{\varphi}(x-is)e^{2\pi i\gamma x}dx-\sum_{\gamma\in\Gamma^{*}}\gamma p_{\gamma}\int_{-\infty}^{+\infty}\hat{\varphi}(x+is)e^{2\pi i\gamma x}dx\\ +\kappa\int_{-\infty}^{+\infty}\hat{\varphi}(x-is)dx+\kappa\int_{-\infty}^{+\infty}\hat{\varphi}(x+is)dx=\sum_{\gamma\in\Gamma_{*}}e^{-2\pi\gamma s}\gamma\tilde{p}_{\gamma}\varphi(\gamma)-\sum_{\gamma\in\Gamma^{*}}e^{2\pi\gamma s}\gamma p_{\gamma}\varphi(\gamma)+2\kappa\varphi(0).$$

Set

$$\Gamma=\Gamma_{*}\cup\Gamma^{*},\quad b_{\gamma}=-\gamma p_{\gamma}e^{-2\pi\gamma s}\quad\text{for}\ \gamma\in\Gamma^{*},\quad b_{\gamma}=\gamma\tilde{p}_{\gamma}e^{2\pi\gamma s}\quad\text{for}\ \gamma\in\Gamma_{*},\quad b_{0}=2\kappa.$$

It follows from (25) and (29) that $\nu=\sum_{\gamma\in\Gamma}b_{\gamma}\delta_{\gamma}$ is a measure on ${\mathbb{R}}$. (31) and (32) yield that for any $\varphi\in S({\mathbb{R}})$ satisfying (8)

$$(\nu,\varphi)=\sum_{\gamma\in\Gamma}b_{\gamma}\varphi(\gamma)=(\mu_{A},\hat{\varphi})=(\hat{\mu}_{A},\varphi).$$

Therefore, $\nu=\hat{\mu}_{A}$.

Now suppose that $\operatorname{supp}\varphi\subset(-r,r)$ for some $r<\infty$. We get from (31)

$$|(\hat{\mu},\varphi)|=|(\mu_{A},\hat{\varphi})|\leq e^{2\pi rs}\left(\sum_{\gamma\in\Gamma_{*}}|\gamma\tilde{p}_{\gamma}|+\sum_{\gamma\in\Gamma^{*}}|\gamma p_{\gamma}|+2\kappa\right)\sup_{|y|\leq r}|\varphi(y)|.$$

Taking into account (28) and (29), we obtain the estimate

$$|(\hat{\mu}_{A},\varphi)|\leq Ce^{2\pi sr}\sup_{(-r,r)}|\varphi(y)|$$

It remains correct for all continuous functions on $[-r,r]$ which equals zero at the points $\pm r$. Therefore,

$$|\hat{\mu}_{A}|(-r,r)\leq Ce^{2\pi sr}.$$

Property (6) is proved. Property (7) follows from (25), (29), and the definition of $b_{\gamma}$.

If $Q$ is an exponential polynomial, then it follows from Lemma that $\Gamma^{*}$ and $\Gamma_{*}$ are locally finite.    

Remark. Without loss of generality assume $0\not\in A$. Then the functions (16) is an entire function of exponential growth with the zero set $A$, and $Q(z)$ is the same. Consequently, $Q(z)=e^{\alpha z+i\beta z}f(z),\,\alpha,\,\beta\in{\mathbb{R}}$. This equality and Proposition 5 yield

$$\frac{Q^{\prime}(x+is)}{Q(x+is)}+2\pi i\kappa=\frac{f^{\prime}(x+is)}{f(x+is)}+\alpha+i\beta+2\pi i\kappa=-2\pi i\sum_{\gamma\in\Gamma\cap{\mathbb{R}}_{+}}\frac{b_{\gamma}}{e^{2\pi\gamma s}}e^{2\pi i\gamma x}+i(\beta+2\pi\kappa-\pi b_{0})+\alpha,$$

Since $\operatorname{sp}\{Q^{\prime}/Q+2\varphi i\kappa\}\subset(0,+\infty)$, we obtain $\alpha=0$, $\beta=\pi b_{0}-2\pi\kappa$, and $Q(z)=Cf(z)e^{(\pi b_{0}-2\pi\kappa)iz}$.

6. Proof of Theorem 2

Let $A\subset\mathcal{S}_{H}\setminus\{0\}$ be an almost periodic set, $\mu_{A}$ be defined in (5), and the pure point measure $\hat{\mu}_{A}$ satisfy (6) and (7). Without loss of generality assume $0\not\in A$. Integrate (21) over the segment $[iy,iy+x]$ with $y$ large enough and change the order of summation and integration. We get

$$\log f(x+iy)-\log f(iy)=\int_{iy}^{x+iy}\frac{f^{\prime}(\zeta)}{f(\zeta)}d\zeta=-\sum_{\gamma\in\Gamma\cap(0,+\infty)}b_{\gamma}\frac{(e^{2\pi i\gamma x}-1)e^{-2\pi\gamma y}}{\gamma}-ib_{0}\pi x.$$

Hence,

(33) $$\|\log f(x+iy)+ib_{0}\pi x\|_{W}\leq\sum_{0<\gamma<1}\frac{2|b_{\gamma}|}{\gamma e^{2\pi\gamma y}}+\sum_{\gamma\geq 1}\frac{2|b_{\gamma}|}{\gamma e^{2\pi\gamma y}}+|\log|f(iy)||.$$

It follows from (7) that the first sum in the right-hand side converges, and (23) implies that the second one converges as well. We see that (33) is finite.

Redefine $y=y_{0}$. The function

$$f(x+iy_{0})e^{ib_{0}\pi x}=\exp\{\log f(x+iy_{0})+ib_{0}\pi x\}$$

belongs to $\mathcal{H}$. Therefore,

$$f(x+iy_{0})e^{ib_{0}\pi x}=\sum_{\omega\in\Omega}\beta_{\omega}e^{2\pi i\omega x},\qquad\sum_{\omega\in\Omega}|\beta_{\omega}|<\infty$$

with $\beta_{\omega}\in{\mathbb{C}}$ and a countable spectrum $\Omega$. The entire function $f(z+iy_{0})e^{ib_{0}\pi z}$ has the exponential growth with zeros in a horizontal strip of bounded width. It follows from [11, §1, Ch.VI], that $\Omega$ is bounded. Hence the function

$$f(z)=\sum_{\omega\in\Omega}\beta_{\omega}e^{\pi(2\omega-b_{0})y_{0}}e^{\pi i(2\omega-b_{0})z}$$

is also a Dirichlet series and $\|f\|_{W}<\infty$.

If $\Gamma$ is locally finite, then the function $\log f(x+iy)+ib_{0}\pi x\in\mathcal{H}$ has nonnegative locally finite spectrum. By Lemma, the same is valid for the function $f(x+iy_{0})e^{ib_{0}\pi x}$. Since $\Omega$ is bounded, we see that only a finite number of coefficients $\beta_{\omega}$ does not vanish. Hence, $f(z)$ is an exponential polynomial. Theorem 2 is proved.