WorksMathematics
Stochastic Euler Products in the Paley-Wiener Space
Abstract
General sufficient conditions, in terms of stochastic Euler products, are given for a realization of a point process to have finite excess or finite deficiency in the Paley-Wiener space.
To Andrei Yurevich Zaitsev, with profound respect
Introduction
Almost every realization of a determinantal point process associated with an orthogonal projection that is locally of trace class is a uniqueness set for the range of the projection [7]. Equivalently, the family of reproducing kernels of the space defining our determinantal process, evaluated at the points of a realization of the process, is almost surely complete. In the special case of the sine process, defined by the projection onto the Paley–Wiener space of square-integrable functions whose Fourier transforms are supported in the interval , the Completeness Theorem is due to Ghosh [16]; in the case of discrete phase spaces, it is due to Lyons [23]. The proof of completeness given in [7] relies essentially on the determinantal structure.
The Completeness Theorem from [7] raises the question of the minimality of our point sets—in other words, whether completeness is preserved when a particle is removed from our random configuration. Minimality clearly holds in the finite-dimensional case. However, for processes with infinitely many particles, minimality fails: the sine process almost surely has excess one [8]. Almost every realization of the sine process remains complete after one particle is removed, but after two particles are removed the resulting point set is the zero set of a nonzero function in the Paley–Wiener space. The second statement—that the excess is at most one—was established in [8] for all determinantal processes with integrable kernels such that
The aim of this note is to give general sufficient conditions ensuring that a realization of a point process belongs to the Paley–Wiener space; that it belongs to this space after a finite (deterministic!) number of particles are removed; and, conversely, sufficient conditions ensuring that a realization does not belong to the Paley–Wiener space.
Informally, the results of the note can be summarized as follows. Suppose initially that our stochastic Euler product grows at infinity no faster than . If a nontrivial limit theorem holds for the logarithm of the absolute value of the stochastic Euler product, then, with positive probability, realizations of our point process are uniqueness sets for the Paley–Wiener space. If, on the other hand, the Euler product itself converges to a random distribution, then, with positive probability, realizations of our point process have excess at least one. While this note is devoted to the Paley–Wiener space only, its methods are quite general, and should be applicable to more general integral transforms, such as, for instance, the Hankel (cf., e.g., [2]), the Airy, the Kontorovich-Lebedev transform [21] as well as the Gorbunov transform in the theory of confluent hypergeometric functions [17].
We now turn to the precise statements. Let denote the space of configurations on . Recall that a configuration is called a uniqueness set, or equivalently a complete set, for the Paley–Wiener space if a function in the Paley–Wiener space is uniquely determined by its restriction to —or, equivalently, if a function in the Paley–Wiener space that vanishes on is identically zero. A uniqueness set is minimal if it ceases to be a uniqueness set once a particle is removed. A uniqueness set has excess if, after removing distinct particles, it remains complete and becomes minimal. A configuration has deficiency if adding distinct particles makes it a complete minimal set. Finally, we say that a configuration has finite deviation if it has either finite excess or finite deficiency.
Let denote the set of configurations such that
(1)
is defined in the principal-value sense;
(2)
Every complete minimal set in the Paley–Wiener space belongs to ((Theorem 6, p. 544, in Levin [22])).
For configurations and numbers , , define the Euler product
and set . If there is an such that
and the restriction of the Euler product to the real axis is square-integrable, then is not complete in the Paley–Wiener space.
Remark 1. The Paley–Wiener Theorem in the form of Plancherel and Pólya ((cf. Theorem II, p. 226, in [27])) gives a slightly stronger statement: if is an entire function of exponential type at most , that is, if
and the restriction of the Euler product to the real axis is square-integrable, then is not complete in the Paley–Wiener space. Nevertheless, the Euler products of realizations of all point processes considered in this paper almost surely satisfy the more restrictive condition (2) as well.
The converse is not true in general: the fact that the Euler product is not square-integrable does not by itself imply that the configuration is complete in the Paley–Wiener space. The set of even integers provides a counterexample. Nonetheless, if the Euler product has regular behaviour on the imaginary axis, then its excess or deficiency can indeed be calculated in terms of .
Proposition 1. Let be a configuration such that its Euler product satisfies inequality (2). Suppose, in addition, that for some , the Euler product satisfies
Then the following statements hold:
(1) If
then is complete and minimal in the Paley–Wiener space.
(2) If is such that
then has excess in the Paley–Wiener space.
(3) If is such that
then has deficiency in the Paley–Wiener space.
Proof. It suffices to prove the first statement; the second and the third follow. By the Paley–Wiener Theorem, if , then
so the set is not complete in the Paley–Wiener space. Now suppose, to the contrary, that there exists a nonzero function such that . We follow B.Ya. Levin, cf. [22 Proposition (г), p. 542]. Set . By definition, is of zero exponential type. In view of (4), has polynomial growth on the imaginary axis, and hence the Phragmén–Lindelöf principle [25, 26] implies that is a polynomial. But the Euler product itself is not square-integrable on the real line, which is a contradiction. The Proposition is proved. □
Observe that, in order to define the absolute value of the Euler product on the imaginary axis, we need only the convergence of the sum of the inverse squares of the particles—that is, only the second condition in the definition of . Moreover, the infimum in (4) may be taken over integer points only:
Proposition 2. For every configuration such that
and every , we have
This is clear.
Now let be a point process such that . The random entire function is called the stochastic Euler product associated with the point process . Our first aim is to give simple sufficient conditions under which the stochastic Euler product almost surely enters the Paley–Wiener space after a finite (deterministic!) number of particles are removed and, conversely, leaves the Paley–Wiener space after a finite (deterministic!) number of particles are added.
Proposition 3. Suppose that, for some and , the following estimates hold:
Then, for every with , for -almost every configuration and every choice of distinct particles , we have
In particular, for -almost every configuration , either is itself incomplete in , or it has excess at most in .
Proposition 3 directly implies
Corollary 1. Suppose that estimate (6) holds and, in addition, that for some and every ,
Then, for every with , for -almost every configuration and every choice of distinct particles , we have
In particular, for -almost every configuration , either is itself incomplete in , or it has excess at most in the space .
To obtain an upper bound on the deficiency of a random configuration , we will need stronger assumptions.
We begin with a simple general observation.
Proposition 4. If has finite deviation, then is a uniqueness set in the Paley–Wiener space if and only if .
Indeed, the statement directly follows from Theorem 6, p. 544, in Levin [22].
We say that a point process has bounded deviation if there exists an such that, for -almost every configuration, either its excess is at most or its deficiency is at most .
Proposition 5. Suppose that estimate (6) holds and, in addition, that for some , , , and , and for every , , with , we have
Then the point process has bounded deviation.
Observe that, in order to obtain estimate (8), it is sufficient to have the estimate
provided that .
We now give a sufficient condition ensuring that the Euler products associated with realizations of our point process fail to belong to the Paley–Wiener space with positive probability.
Theorem 1. Let , and suppose that the limit
exists in the space of distributions on and is nondegenerate in the sense that there exists a nonnegative infinitely differentiable function with compact support on the positive half-line such that . Assume, furthermore, that the limit
exists in the space of distributions on . Then we have
Under the additional assumption that the point process has bounded deviation we also have
Recall that two configurations are called symmetrically equivalent if there exists a compact subset of the phase space such that the two point sets coincide outside this compact set and have the same number of particles inside it. In the special case of the real line, two configurations are symmetrically equivalent if and only if a compactly supported diffeomorphism of the line takes one configuration to the other. Clearly, two symmetrically equivalent configurations on the line have the same excess (or deficiency) in the Paley–Wiener space. A Borel subset of the space of configurations that can be represented as a union of symmetric equivalence classes is called a symmetric event. The completion of the sigma-algebra of all symmetric events is called the symmetric sigma-algebra of our point process. Theorem 1 now yields the following corollary.
Corollary 2. Suppose that the point process has bounded deviation and satisfies the assumptions of Theorem 1. If the symmetric sigma-algebra of the point process is trivial, then almost all its realizations are uniqueness sets in the Paley–Wiener space .
Remark 2. The fact that the excess of the sine process is equal to 1 follows immediately from Theorem 1; a detailed proof will be given in the sequel to this note. Recently, Bourgade, Sodin, and Zeitouni announced an exact computation of the excess of the sine- process for all values of the parameter . It would be very interesting to establish the Bourgade–Sodin–Zeitouni Theorem using the elementary methods developed in this note.
Stochastic Euler products of point processes have been studied, in particular, in [3], [9], [11], [12], [24], and [32].
Proofs of Propositions 3 and 5
Proof of Proposition 3
Proof of Proposition 3. First, recall that
and, by the Paley–Wiener Theorem in the form of Plancherel and Pólya, an entire function belongs to if and only if
and
Remark 3. For the purposes of this note, however, it suffices to use the original Paley–Wiener condition
and its polynomially weighted extension (2).
We now verify the conditions of the Paley-Wiener Theorem for the Euler product associated with our configuration.
Let satisfy . Choose the points of our configuration that are closest to the origin, and set . The map is Borel measurable. First observe that, under our assumptions, the Euler product satisfies condition (11) almost surely with respect to . Indeed, by (6) and the Borel–Cantelli Lemma, for -almost every we have
which implies (11).
Condition (10) for follows directly from (7).
It follows that, almost surely with respect to , the function belongs to the Paley–Wiener space .
If a function belongs to the Paley–Wiener space and
then the function also belongs to the Paley–Wiener space. Our assumptions on imply that, for every set of distinct particles of the configuration , the function associated with the configuration is square-integrable on the real line:
for -almost every configuration . It follows that belongs to the Paley–Wiener class , as required. □
We turn to the proof of Proposition 5.
Proof of Proposition 5
Let . From the Plancherel–Pólya Discrete Sampling Theorem [28] it directly follows that if the point set is obtained from by removing particles (in order to ensure Borel choice, remove positive particles closest to ), then . Suppose that , , and
Then, for -almost every , we have
and, by Proposition 2, also
The Proposition now follows directly from Proposition 1.
Non-convergence in probability ensures completeness
Proposition 6. Let be a point process on satisfying (6) and (8). Suppose, moreover, that there exist , , and a sequence such that
Then .
Proof. By Proposition 5, the deviation of is almost surely finite. Suppose, to the contrary, that . By Proposition 4, it follows that almost surely. Consequently,
almost surely and, a fortiori, in probability. However, the convergence in probability to zero contradicts the assumption. The proof is complete.
Proof of Theorem 1
Proof. Under the assumptions of the Theorem, there exists a nonnegative infinitely differentiable function with compact support such that
and . Of course, we also have
Our assumptions imply the existence of a positive limit
as well as the existence of the limit
By the Cauchy–Bunyakovsky–Schwarz Inequality, we have . The Paley–Zygmund Inequality now implies that there exist constants and , depending only on and , such that
Observing that
we obtain
Consequently, there exist , , and such that
which directly implies, also in the case (indeed, if a random function is square-integrable almost surely, then the integrals of its square over intervals far from zero converge in probability to zero, which contradicts (12)), that
The proof is complete. □
Acknowledgements
Acknowledgements. I am deeply grateful to A. S. Arakcheev, A. A. Borichev, M. A. Borovikov, S. M. Gorbunov, M. L. Sodin, and I. I. Zavolokin for useful discussions.
The author is a winner of the “Leader” competition of the BASIS Foundation for the Advancement of Theoretical Physics and Mathematics, and would like to thank the jury and sponsors.
Part of this paper was written in the framework of a research group at the Sirius Mathematical Centre. I am deeply grateful to the Sirius Mathematical Centre for its warm hospitality.
Steklov Mathematical Institute of Russian Academy of Sciences, Moscow
Email address: [email protected]
References
- [2]N. I. Akhiezer, On a generalization of the Fourier transform and the Paley–Wiener theorem. Doklady Akademii Nauk SSSR 96 (1954), no. 4, 889–892.
- [3]T. Assiotis, Random entire functions from random polynomials with real zeros, Advances in Mathematics 410 (2022), article 108701.arxiv.org/abs/2202.03362
- [7]A. Bufetov, Y. Qiu, A. Shamov, Kernels of conditional determinantal measures and the Lyons–Peres completeness conjecture, arXiv:1612.0675, JEMS, 23(5), 2021, 1477–1519.
- [8]A. I. Bufetov, The sine-process has excess one, arXiv:1906.02098, 2019.
- [9]A. I. Bufetov, Quasi-symmetries of determinantal point processes. Ann. Probab. 46 (2018), 956–1003.arxiv.org/abs/1409.2068
- [11]R. Chhaibi, J. Najnudel and A. Nikeghbali, A limiting random analytic function related to the CUE, preprint, 2014. arXiv:1403.7814.arxiv.org/abs/1403.7814
- [12]R. Chhaibi, E. Hovhannisyan, J. Najnudel, A. Nikeghbali and B. Rodgers, The limiting characteristic polynomial of classical random matrix ensembles, Annales Henri Poincaré 20 (2019), 1093–1119.
- [16]S. Ghosh, Determinantal processes and completeness of random exponentials: the critical case. Probab. Theory Relat. Fields 163 (2015), 643–665.arxiv.org/abs/1211.2435
- [17]S. M. Gorbunov, Unitary transform diagonalizing the confluent hypergeometric kernel, Sb. Math. 216 (2025), no. 12, 1643–1663.DOI
- [21]M. I. Kontorovich and N. N. Lebedev, On the method of solving certain problems of mathematical physics. Journal of Experimental and Theoretical Physics (JETP) 8 (1938), no. 10–11, 1192–1206.
- [22]B. Ya. Levin, Distribution of Zeros of Entire Functions, GITTL, Moscow, 1956.
- [23]R. Lyons, Determinantal probability measures. Publ. Math. Inst. Hautes Études Sci. (98) (2003), 167–212.
- [24]J. Najnudel and A. Nikeghbali, Convergence of random holomorphic functions with real zeros and extensions of the stochastic zeta function, preprint, 2022, revised 2024. arXiv:2202.04284.arxiv.org/abs/2202.04284
- [25]E. Phragmén (1904), “Sur une extension d’un théorème classique de la théorie des fonctions”, Acta Mathematica 28 (1904), 351–368.DOI
- [26]L. E. Phragmén and E. Lindelöf (1908), “Sur une extension d’un principe classique de l’analyse et sur quelques propriétés des fonctions monogènes dans le voisinage d’un point singulier”, Acta Math. 31 (1), 381–406.
- [28]M. Plancherel, G. Pólya, Fonctions entières et intégrales de Fourier multiples. II., Commentarii Mathematici Helvetici 10 (1937), no. 2, 110–163.
- [32]B. Valkó and B. Virág, The many faces of the stochastic zeta function, Geometric and Functional Analysis 32 (2022), 1160–1231.DOI