Application of methods of quasicrystals theory to entire functions of exponential growth Sergii Yu.Favorov Abstract Abstract. Let f be an entire almost periodic function with zeros in a horizontal strip of finite width; for example, any exponential polynomial with purely imaginary exponents is such a function. Let \mu be the measure on the set of zeros of f whose masses coincide with multiplicities of zeros. We define the Fourier transform in the sense of distributions for \mu and prove that it is a pure point measure on {\mathbb{R}} whose complex masses correspond to coefficients of Dirichlet series of the logarithmic derivative of f . Bases on this description and Meyer’s theorem on quasicrystals, we give a simple necessary and sufficient condition for f to be a finite product of sines. AMS Mathematics Subject Classification: 42A75, 42A38, 52C23 Keywords: exponential polynomial, zero set in a strip, almost periodic function, Fourier quasicrystal 1. Introduction In [ 22 ] A.Olevskii and A.Ulanovskii established 1-1 connection between zeros of real–rooted exponential polynomials and Fourier quasicrystals. They proved that for the exponential polynomial (1) 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}},\quad N\in{\mathbb{N}}. with a zero set A\subset{\mathbb{R}} the measure (2) \mu_{A}=\sum_{n}\delta_{a_{n}},\quad A=\{a_{n}\}, is the Fourier quasicrystal, and conversely, every set A\subset{\mathbb{R}} for which the measure ( 2 ) is a Fourier quasicrystal, is the zero set of some exponential polynomial ( 1 ). Note that previously a nontrivial example of the Fourier quasicrystal of the form ( 2 ) whose support has only a finite intersection with any arithmetical progression was found by P.Kurasov and P.Sarnak [ 14 ] . Recall that a complex or real measure \mu is called the Fourier quasicrystal if it is tempered distribution, has locally finite support (that is, its intersection with any compact set is finite), its Fourier transform in the sense of distributions is also is a measure with locally finite support, and moreover, both measures |\mu| , |\hat{\mu}| are temperate distributions. Here and below |\nu|(E) means the variation of the complex measure \nu on the set E . Recently, Fourier quasicrystals and connected with them real-rooted exponential polynomials are studied very actively (see, for example, [ 21 , 15 , 2 ] ). In fact, Fourier quasicrystals are the form of Poisson formulas (see below), the latter were used in particular by D. Radchenko and M. Viazovska in [ 23 ] . L.Alon, A.Cohen, K.Vinzant [ 1 ] established a connection between real–rooted exponential polynomials and Lee-Yang polynomials of many variables. In [ 8 ] the result of Olevskii and Ulanovskii was generalized to Dirichlet series, it was found a full description of zero sets of absolutely convergent real–rooted Dirichlet series (3) Q(z)=\sum_{\omega\in\Omega}q_{\omega}e^{2\pi i\omega z},\qquad q_{\omega}\in{\mathbb{C}},\quad\omega\in{\mathbb{R}},\quad\sum_{\omega\in\Omega}|q_{\omega}|<\infty, with bounded \Omega . F.Goncalves [ 12 ] adapted the techniques of Olevskii and Ulanovskii for study of arbitrary tempered measures on {\mathbb{R}} with a pure point Fourier transform and got similar results for such measures. In [ 10 ] we applied the above description of Fourier quasicrystals and Meyer’s theorem on quasicrystals [ 19 ] to prove a simple criterion for the exponential polynomials ( 1 ) and Dirichlet series ( 3 ) to be a finite product of sines. But there we assumed that these polynomials and series have only real zeros. In [ 9 ] we applied the method of Olevskii and Ulanovskii to measures of the form ( 2 ) where we replace the condition A\subset{\mathbb{R}} with A\subset\{z=x+iy:\,|y|\leq H\} . We slightly changed the definition of the Fourier transform of a measure, introduced the measure \hat{\mu}_{A}^{c} , and proved that the above results on zeros of exponential sums and Dirichlet series are valid in this case as well. Therefore it is natural to use this result to remove the condition of reality of zeros in [ 10 ] . Moreover, we would like to get the above criterion for the wider class of entire almost periodic functions. To achieve this result, two obstacles must be overcome. First, the Dirichlet series of a holomorphic almost periodic function need not converge. Second, the inverse Fourier transform of the measure \hat{\mu}_{A}^{c} is not the measure \mu_{A} , which makes direct application of Meyer’s theorem impossible. Solving these problems, we obtain the following: Theorem . Let f(z) be an entire almost periodic function of exponential growth with zeros in a strip S_{(\alpha,\beta)}=\{z\in{\mathbb{C}}:\,\alpha<\Im z<\beta\},\qquad-\infty<\alpha\leq 0\leq\beta<\infty, and (4) f^{\prime}(z)/f(z)\sim\sum_{\gamma\in\Gamma^{*}}h^{+}_{\gamma}e^{2\pi i\gamma z},\ z\in S_{(\beta,\infty)},\qquad f^{\prime}(z)/f(z)\sim\sum_{\gamma\in\Gamma_{*}}h^{-}_{\gamma}e^{2\pi i\gamma z},\ z\in S_{(-\infty,\alpha)}. Then the condition (5) \sum_{|\gamma|1, where K is a constant, is necessary and sufficient for the identity (6) f(z)=Ce^{iaz}\prod_{j=1}^{J}\sin(\alpha_{j}z+\beta_{j}) with some C\in{\mathbb{C}},\ J\in{\mathbb{N}},\ a,\,\alpha_{j},\,\beta_{j}\in{\mathbb{R}} . Remark 1 . Obviously, this is the case when support of \mu is a finite union of arithmetic progressions. There are many papers (e.g., [ 6 ] , [ 7 ] , [ 16 ] ) where various conditions have been found for support of \mu to be just like that. But in these papers, the distances between points of support of \mu or between points of support of its Fourier transform \hat{\mu} are assumed to be uniformly separated from zero. The exception is Meyer’s theorem [ 19 ] , where support of \mu is an arbitrary locally finite set and \hat{\mu} is a slowly increasing Radon measure with arbitrary support. The proof of our theorem is based on it. Remark 2 . The representations ( 4 ) for every entire almost periodic function of exponential growth with zeros in the strip S_{(\alpha,\beta)} will be proved below. Almost all papers related to Fourier quasicrystals use (in explicit or implicit form) properties of almost periodic functions and measures. In section 2 we give definitions of these objects and present some of their properties. In section 3 we recall some definitions related to the Fourier transform of functions and measures. In section 4 we give a proof of the theorem. 2. Almost periodic functions and measures Our proof of the Theorem is based on some results of the classical theory of almost periodic functions (see [ 3 ] - [ 5 ] , [ 18 ] , [ 17 , Ch.6] ). Definition 1 . A continuous function f(x) on the real line {\mathbb{R}} is almost periodic if for any \varepsilon>0 the set of \varepsilon -almost periods E_{\varepsilon}=\{\tau\in{\mathbb{R}}:\,\sup_{x\in{\mathbb{R}}}|f(x+\tau)-f(x)|<\varepsilon\} is relatively dense, i.e., E_{\varepsilon}\cap(x,x+L)\neq\emptyset for all x\in{\mathbb{R}} and some L depending on \varepsilon ; a continuous function g(z) on the open strip S_{(\alpha,\beta)}=\{z=x+iy\in{\mathbb{C}}:\,-\infty\leq\alpha0 the set of \varepsilon -almost periods E_{\alpha^{\prime},\,\beta^{\prime},\varepsilon}=\{\tau\in{\mathbb{R}}:\,\sup_{x\in{\mathbb{R}},\alpha^{\prime}\leq y\leq\beta^{\prime}}|g(x+\tau+iy)-g(x+iy)|<\varepsilon\} is relatively dense with L depending on \varepsilon,\alpha^{\prime},\beta^{\prime} . Note that every exponential sum ( 1 ) and absolutely convergent Dirichlet series ( 3 ) are almost periodic in the whole plane S_{(-\infty,+\infty)} . Properties of almost periodic functions : a) a finite sum, a product, and a linear combination of almost periodic functions on {\mathbb{R}} is also almost periodic on {\mathbb{R}} ; a finite sum, a product, and a linear combination of almost periodic functions in a strip S_{(\alpha,\beta)} is also almost periodic in S_{(\alpha,\beta)} , b) each almost periodic function f(x) on {\mathbb{R}} is bounded and has unbounded support unless f(x)\not\equiv 0 ; each almost periodic function g(z) on S_{(\alpha,\beta)} is bounded on every substrip S_{(\alpha^{\prime},\beta^{\prime})},\ \alpha<\alpha^{\prime}<\beta^{\prime}<\beta , and has unbounded support unless g(z)\not\equiv 0 , c) the Fourier series corresponds to each almost periodic function f(x) on {\mathbb{R}} f(x)\sim\sum_{\omega\in\Omega}a_{\omega}e^{2\pi i\omega x}, where the spectrum \Omega=\{\omega:\,a_{\omega}\neq 0\} is a countable subset of {\mathbb{R}} ; the Fourier series corresponds to each almost periodic function g(z) in S_{(\alpha,\beta)} g(x+iy)\sim\sum_{\omega\in\Omega}a_{\omega}(y)e^{2\pi i\omega x},\quad\alpha0 and \alpha<\alpha^{\prime}<\beta^{\prime}<\beta there is m=m(\varepsilon,\alpha^{\prime},\beta^{\prime})>0 such that the conditions z\in S_{(\alpha^{\prime}\beta^{\prime})} and \dist(z,A)>\varepsilon implies |g(z)|\geq m. k) for an almost periodic in S_{(\alpha,\beta)} function g(z) the numbers \#\{z=x+iy\in S_{(\alpha^{\prime},\beta^{\prime})}\,:\,t\leq x\leq t+1,\,g(z)=0\},\quad\alpha<\alpha^{\prime}<\beta^{\prime}<\beta, are bounded uniformly in t\in{\mathbb{R}} ; all zeros are counted according to their multiplicities, hence multiplicities of zeros are uniformly bounded in every substrip S_{(\alpha^{\prime},\beta^{\prime})} . Here we do not give a general definition of almost periodic measures on a strip (see [ 24 ] , [ 11 ] ), but restrict ourselves to the case of measures on {\mathbb{R}} ( [ 20 ] ). Definition 2 . A measure \mu on {\mathbb{R}} is almost periodic (in the sense of distributions) if the function \mu\star\phi(t)=\int\psi(t-x)\mu(dx) is almost periodic for any \psi\in\mathcal{D} . It is easily seen that any linear combinations of almost periodic measures is almost periodic, and the support of every almost periodic measure \mu\not\equiv 0 is unbounded. 3. Some properties of the Fourier transform Let \mathcal{D} be the space of C^{\infty} -functions with compact support on {\mathbb{R}} , \mathcal{S} be the Schwartz space of all C^{\infty} -functions \varphi with finite norms \mathcal{N}_{k}(\varphi)=\sup_{x\in{\mathbb{R}}}(1+|x|)^{k}\max_{0\leq m\leq k}|\varphi^{(m)}(x)|\quad\forall\,k=0,1,2,\dots, and the space of tempered distributions \mathcal{S}^{\prime} be the space of all linear functional on \mathcal{S} , which continuous with respect to the topology generated by these norms. The Fourier transform (8) \hat{\varphi}(x)=\int_{{\mathbb{R}}}\varphi(t)e^{-2\pi ixt}dt is a continuous bijection of \mathcal{S} onto \mathcal{S} , and for \varphi\in\mathcal{S} (9) \varphi(t)=\int_{{\mathbb{R}}}\hat{\varphi}(x)e^{2\pi ixt}dx, (see, e.g., [ 26 ] ). For every \Phi\in\mathcal{S}^{\prime} its Fourier transform \hat{\Phi} is defined by the equality (10) \hat{\Phi}(\varphi)=\Phi(\hat{\varphi})\quad\forall\,\varphi\in\mathcal{D}. Since \mathcal{D} is a dense subset of \mathcal{S} , we see that ( 10 ) is valid for all \varphi\in\mathcal{S} . Therefore, \hat{\Phi}\in\mathcal{S}^{\prime} for all \Phi\in\mathcal{S}^{\prime} . It follows from ( 8 )–( 10 ) that \widehat{\hat{\Phi}}(x)=\Phi(-x). When \Phi and \hat{\Phi} are tempered measure, and \Phi=\sum_{\lambda\in\Lambda}a_{\lambda}\delta_{\lambda},\qquad\hat{\Phi}=\sum_{\gamma\in\Gamma}b_{\gamma}\delta_{\gamma} with countable \Lambda,\,\Gamma , then equality ( 10 ) has the form \sum_{\gamma\in\Gamma}b_{\gamma}\varphi(\gamma)=\sum_{\lambda\in\Lambda}a_{\lambda}\hat{\varphi}(\lambda),\qquad\forall\varphi\in\mathcal{D}, and is called generalized Poisson formula . For \varphi\in\mathcal{D} the function (11) \hat{\varphi}^{c}(z)=\int_{{\mathbb{R}}}\varphi(t)e^{-2\pi izt}dt=\widehat{(\varphi(t)e^{2\pi yt})}(x),\quad z=x+iy, is the entire extension of \hat{\varphi} . It is easy to check that for every k\in{\mathbb{N}} and s>0 uniformly in y\in[-s,s] (12) \hat{\varphi}^{c}(x+iy)=O(|x|^{-k}),\qquad|x|\to\infty, In the case of \supp\mu\subset\mathcal{S}_{(\alpha,\beta)},-\infty<\alpha<\beta<\infty , and \mu such that (13) \log\left(|\mu|\{z\in\mathcal{S}_{(\alpha,\beta)}:\,|\Re z|0 there are m=m(\varepsilon,\alpha,\beta,\eta)>0 and sequences L_{k}\to+\infty,\,L^{\prime}_{k}\to-\infty such that |f(x+iy)|>m\quad\text{for}\quad x=L_{k}\quad\text{or}\quad x=L^{\prime}_{k},\quad\alpha-\eta\leq y\leq\beta+\eta. Let \varphi\in\mathcal{D} , and \hat{\varphi}^{c}(z) be defined in ( 11 ). Consider the integrals I_{k} of the function \hat{\varphi}^{c}(z)f^{\prime}(z)f^{-1}(z) over the boundaries of the rectangles \Pi_{k}=\{z=x+iy:\,L^{\prime}_{k}0 \sum_{\gamma\in\Gamma^{*}\setminus\{0\}}|h_{\gamma}^{+}|e^{-2\pi\gamma\varepsilon}\leq\int_{0}^{\infty}e^{-2\pi r\varepsilon}R(dr)=2\pi\varepsilon\int_{0}^{\infty}R(r)e^{-2\pi r\varepsilon}dt<\infty. Therefore the Dirichlet series \sum_{\gamma\in\Gamma^{*}}h_{\gamma}^{+}e^{2\pi i\gamma z} absolutely converges on every half-plane \Im z>\varepsilon for all \varepsilon>0 . Hence this sum is the analytic extension of the function f^{\prime}(z)/f(z) , and f(z)\neq 0 for y>0 . By the same way we get f(z)\neq 0 for y<0 . So, A\subset{\mathbb{R}} , and the measure \hat{\mu}_{A}^{c} coincides with the measure \hat{\mu}_{A} . By ( 15 ), we also have for any \varphi\in\mathcal{D} (17) \int\varphi(t-s)\mu_{A}(ds)=\int\hat{\varphi}(x)e^{2\pi itx}\hat{\mu}_{A}(dx)=\sum_{\gamma\in\Gamma^{*}}\frac{ih^{+}_{\gamma}}{2\pi}\hat{\varphi}(\gamma)e^{2\pi it\gamma}-\sum_{\gamma\in\Gamma_{*}}\frac{ih^{-}_{\gamma}}{2\pi}\hat{\varphi}(\gamma)e^{2\pi it\gamma}. Since \sup_{|\gamma|