0% found this document useful (1 vote)
516 views313 pages

TRIGONOMETRIA

MATEMATICA
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (1 vote)
516 views313 pages

TRIGONOMETRIA

MATEMATICA
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 313

Andrei Raigorodskii

Michael Th. Rassias  Editors

Trigonometric
Sums and Their
Applications
Trigonometric Sums and Their Applications
Andrei Raigorodskii • Michael Th. Rassias
Editors

Trigonometric Sums
and Their Applications
Editors
Andrei Raigorodskii Michael Th. Rassias
Moscow Institute of Physics and Institute of Mathematics
Technology University of Zurich
Dolgoprudny, Russia Zurich, Switzerland
Moscow State University Moscow Institute of Physics and
Moscow, Russia Technology
Dolgoprudny, Russia
Buryat State University
Ulan-Ude Russia Institute for Advanced Study
Program in Interdisciplinary Studies
Caucasus Mathematical Center
Princeton, NJ, USA
Adyghe State University, Maykop, Russia

ISBN 978-3-030-37903-2 ISBN 978-3-030-37904-9 (eBook)


https://doi.org/10.1007/978-3-030-37904-9

Mathematics Subject Classification (2010): 42-XX, 43-XX, 44-XX, 26-XX, 30-XX, 32-XX, 33-XX,
35-XX, 40-XX, 41-XX, 46-XX, 47-XX, 65-XX

© Springer Nature Switzerland AG 2020


This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors, and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, expressed or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Switzerland AG.
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface

This volume is devoted to the study of trigonometric and exponential sums with their
various applications and interconnections with other mathematical objects. These
types of sums play an essential role in a broad variety of mathematical areas of
research, including real and complex analysis, analytic number theory, functional
analysis, approximation theory, harmonic analysis, and analytic inequalities with
best constants. More specifically, the volume deals with Marcinkiewicz-Zygmund
inequalities at Hermite zeros and their airy function cousins, polynomials with
constrained coefficients, nonnegative sine polynomials, inequalities for weighted
trigonometric sums, Dedekind- and Hardy-type sums and trigonometric sums
induced by quadrature formulas, best orthogonal trigonometric approximations
of classes of infinitely differentiable functions, trigonometric functions related to
the Riemann zeta function and the Nyman-Beurling criterion for the Riemann
hypothesis, half-discrete Hilbert-type inequalities in the whole plane with the kernel
of hyperbolic secant function, reverse Hilbert-type integral inequalities with the
kernel of hyperbolic cotangent function, sets with small Wiener norm, double-
sided Taylor’s approximations and their applications in the theory of trigonometric
inequalities, norm inequalities for generalized Laplace transforms, and the first
derivative of Hardy’s Z-function.
The papers have been contributed by eminent experts in the corresponding
domains, who have presented the state of the art in the problems treated. The present
volume is expected to be a valuable source for both graduate students and research
mathematicians as well as physicists and engineers. We would like to express our
warmest thanks to all the authors of papers in this volume who contributed in this
collective effort. Last but not least, we would like to extend our appreciation to the
Springer staff for their valuable help throughout the publication process of this work.

Moscow, Russia Andrei Raigorodskii


Zurich, Switzerland Michael Th. Rassias

v
Contents

On a Category of Cotangent Sums Related to the Nyman-Beurling


Criterion for the Riemann Hypothesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Nikita Derevyanko, Kirill Kovalenko, and Maksim Zhukovskii
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Nyman–Beurling Criterion for the Riemann Hypothesis . . . . . . . . . . . . 2
1.2 The Cotangent Sum’s Applications to Problems Related to
the Riemann Hypothesis. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2 Central Properties of the Cotangent Sum c0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.1 Ellipse . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3 The Maximum of c0 in Rational Numbers in Short Intervals . . . . . . . . . . . . . . 15
4 The Function g(x) and Moments of c0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
5 Dedekind Sums . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
6 Sums Appearing in the Nyman-Beurling Criterion for the Riemann
Hypothesis Containing the Möbius Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Recent Progress in the Study of Polynomials with Constrained
Coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
Tamás Erdélyi
1 Ultraflat Sequences of Unimodular Polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2 More Recent Results on Ultraflat Sequences of Unimodular
Polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3 Flatness of Conjugate-Reciprocal Unimodular Polynomials . . . . . . . . . . . . . . 38
4 Average Lq Norm of Littlewood Polynomials on the Unit Circle . . . . . . . . . 40
5 Rudin-Shapiro Polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
6 Mahler Measure and Moments of the Rudin-Shapiro Polynomials . . . . . . . 43
7 Lemmas for Theorem 6.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
8 Saffari’s Conjecture on the Shapiro Polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . 45
9 Consequences of Saffari’s Conjecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
10 Open Problems Related to the Rudin-Shapiro Polynomials . . . . . . . . . . . . . . . 49
11 On the Size of the Fekete Polynomials on the Unit Circle . . . . . . . . . . . . . . . . . 50

vii
viii Contents

12 Unimodular Zeros of Self-Reciprocal Polynomials


with Coefficients in a Finite Set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
13 Bourgain’s L1 Problem and Related Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
Reference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
Classes of Nonnegative Sine Polynomials. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
Horst Alzer and Man Kam Kwong
1 Introduction and Statement of Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
2 Lemmas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3 Proof of Theorem 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4 Proof of Theorem 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
5 Proof of Theorem 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
6 Proof of Theorem 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
7 Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
Inequalities for Weighted Trigonometric Sums . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
Horst Alzer and Omran Kouba
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
2 A Technical Lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
3 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
Norm Inequalities for Generalized Laplace Transforms . . . . . . . . . . . . . . . . . . . . 97
J. C. Kuang
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
2 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
3 Proof of Theorem 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
4 The Discrete Versions of the Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
5 Some Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
On Marcinkiewicz-Zygmund Inequalities at Hermite Zeros
and Their Airy Function Cousins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
D. S. Lubinsky
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
2 Proof of Theorems 1.3 and 1.4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
The Maximum of Cotangent Sums Related to the Nyman-Beurling
Criterion for the Riemann Hypothesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
Helmut Maier, Michael Th. Rassias, and Andrei Raigorodskii
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
2 Exponential Sums over Primes in Finite Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
3 Other Preliminary Lemmas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
4 Proof of Theorem 1.5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
Contents ix

Double-Sided Taylor’s Approximations and Their Applications in


Theory of Trigonometric Inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
Branko Malešević, Tatjana Lutovac, Marija Rašajski, and Bojan Banjac
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
2 An Overview of the Results Related to Double-Sided TAYLOR’s
Approximations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
3 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
3.1 Generalization of Statement 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
3.2 An Improvement of Statement 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
The Second Moment of the First Derivative of Hardy’s Z-Function . . . . . . . 169
Maxim A. Korolev and Andrei V. Shubin
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
2 Auxiliary Lemmas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
3 Proof of the Main Theorem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
Dedekind and Hardy Type Sums and Trigonometric Sums Induced
by Quadrature Formulas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
Gradimir V. Milovanović and Yilmaz Simsek
1 Introduction and Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
2 Lambert and Eisenstein Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
2.1 Further Remarks and Observations for Eisenstein Series . . . . . . . . . . . . 189
3 Dedekind Sums . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
3.1 Some Others Formulas for the Dedekind Sums . . . . . . . . . . . . . . . . . . . . . . 196
4 Hardy Sums . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
5 Dedekind Type Daehee-Changhee (DC) Sums . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
6 Trigonometric Representation of the DC-Sums . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
7 DC-Sums Related to Special Functions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
8 Reciprocity Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
9 Sums Obtained from Gauss-Chebyshev Quadratures . . . . . . . . . . . . . . . . . . . . . . 216
10 Sums Obtained from Trigonometric Quadrature Rules . . . . . . . . . . . . . . . . . . . . 223
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
On a Half-Discrete Hilbert-Type Inequality in the Whole Plane
with the Kernel of Hyperbolic Secant Function Related to the
Hurwitz Zeta Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
Michael Th. Rassias, Bicheng Yang and Andrei Raigorodskii
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
2 Weight Functions and Some Lemmas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
3 Main Results. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
4 Operator Expressions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
5 Two Kinds of Equivalent Reverse Inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258
x Contents

A Remark on Sets with Small Wiener Norm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261


I.D. Shkredov
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
2 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
3 On the Multiplicative Energy of Sets with Small Wiener Norm . . . . . . . . . . . 265
4 On the Quantity M+ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
Order Estimates of Best Orthogonal Trigonometric Approximations
of Classes of Infinitely Differentiable Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
Tetiana A. Stepanyuk
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
2 Best Orthogonal Trigonometric Approximations of the Classes
ψ
Lβ,p , 1 < p < ∞, in the Metric of Space L∞ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276
3 Best Orthogonal Trigonometric Approximations of the Classes
ψ
Lβ,1 in the Metric of Space L∞ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
4 Best Orthogonal Trigonometric Approximations of the Classes
ψ
Lβ,1 in the Metric of Spaces Ls , 1 < s < ∞. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286
Equivalent Conditions of a Reverse Hilbert-Type Integral
Inequality with the Kernel of Hyperbolic Cotangent Function
Related to the Riemann Zeta Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
Bicheng Yang
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
2 An Example and Two Lemmas. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
3 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296
4 Some Corollaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 304
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 304

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
On a Category of Cotangent Sums
Related to the Nyman-Beurling Criterion
for the Riemann Hypothesis

Nikita Derevyanko, Kirill Kovalenko, and Maksim Zhukovskii

Abstract The purpose of the present paper is to provide a general overview of


a variety of results related to a category of cotangent sums which have been
proven to be associated to the so-called Nyman-Beurling criterion for the Riemann
Hypothesis. These sums are also related to the Estermann Zeta function.

Keywords Cotangent sums · Riemann zeta function · Vasyunin sums ·


Estermann zeta function · Riemann Hypothesis · Dedekind sums ·
Nyman-Beurling-Báez-Duarte criterio

1 Introduction

This paper is focused on applications of certain cotangent sums to various problems


related to the Riemann Hypothesis. The expression for the trigonometric sums in
question is the following

Definition 1.1
  
b−1  
r m π mr
c0 := − cot , (1)
b b b
m=1

where r, b ∈ N, b ≥ 2, 1 ≤ r ≤ b and (r, b) = 1.

N. Derevyanko
Moscow Institute of Physics and Technology, Dolgoprudny, Russia
e-mail: [email protected]
K. Kovalenko
National Research University Higher School of Economics, Moscow, Russia
e-mail: [email protected]
M. Zhukovskii ()
Laboratory of Advanced Combinatorics and Network Applications, Moscow Institute
of Physics and Technology, Dolgoprudny, Russia

© Springer Nature Switzerland AG 2020 1


A. Raigorodskii, M. T. Rassias (eds.), Trigonometric Sums and Their Applications,
https://doi.org/10.1007/978-3-030-37904-9_1
2 N. Derevyanko et al.

These sums were thoroughly studied for the first time in the important paper of S.
Bettin and J. Conrey [7] in which they establish a reciprocity formula for these sums
among other properties. We will provide more details about their work – among
other results – in Section 2.

1.1 Nyman–Beurling Criterion for the Riemann Hypothesis

There are many interesting results concerning these cotangent sums, but initially
we will present some general information about the Riemann Hypothesis and
some related problems. Moreover, our aim is to provide motivation for the use of
cotangent sums in these problems.
In this paper we shall denote a complex variable by s = σ + it, where σ and t
are the real and imaginary part of s respectively.
Definition 1.2 The Riemann zeta function is a function of the complex variable s
defined in the half-plane {σ > 1} by the absolutely convergent series

 1
ζ (s) := . (2)
ns
n=1

As shown by B. Riemann, ζ (s) extends to C as a meromorphic function with


only a simple pole at s = 1, with the residue 1, and satisfies the functional equation
πs 
ζ (s) = 2s π s−1 sin (1 − s)ζ (1 − s). (3)
2
For negative integers, one has a convenient representation of the Riemann zeta
function in terms of Bernoulli numbers:
Bn+1
ζ (−n) = (−1)n , for n ≥ 0.
n+1

By the above formula one can easily deduce that ζ (s) vanishes when s is a negative
even integer because Bm = 0 for all odd m other than 1. The negative even integers
are called trivial zeros of the Riemann zeta function. All other complex points where
ζ (s) vanishes are called non-trivial zeros of the Riemann zeta function, and they play
a significant role in the distribution of primes.
The actual connection with the distribution of prime numbers was observed in
Riemann’s 1859 paper. It is in this paper that Riemann proposed his well known
hypothesis.
On a Category of Cotangent Sums Related to the Nyman-Beurling Criterion for. . . 3

Hypothesis 1.1 (Riemann) The Riemann zeta function ζ (s) attains its non-trivial
zeros only in complex points with σ = 12 . The line on the complex plane given by
the equation σ = 12 is usually called “critical”.

The Nyman–Beurling–Baez–Duarte–Vasyunin (also simply known as Nyman-


Beurling) approach to the Riemann Hypothesis enables us to associate the study of
cotangent sums and the Riemann Hypothesis through the following theorem:
Theorem 1.1 The Riemann Hypothesis is true if and only if

lim dN = 0,
N →+∞

where
 +∞ 
    2
1 
dN2 = inf 1 − ζ 1 + it DN 1 + it  dt
(4)
DN 2π  2 2  1
+ t2
−∞ 4

and the infimum is taken over all Dirichlet polynomials


N
an
DN (s) = , an ∈ C. (5)
ns
n=1

In his paper [2], B. Bagchi used a slightly different formulation of Theorem 1.1.
In order to state it, we have to introduce some definitions.
Definition 1.3 The Hardy space H 2 () is the Hilbert space of all analytic functions
F on the half-plane  (we define it for a right half-plane {σ > σ0 } of the complex
plane) such that
 +∞
1
F  := sup
2
|F (σ + it)|2 dt < ∞.
σ >σ0 2π −∞

Everywhere in this section we will use  = {σ > 12 }.


Definition 1.4 For 0 ≤ λ ≤ 1, let Fλ ∈ H 2 () be defined by

ζ (s)
Fλ (s) = (λs − λ) , s ∈ ,
s

and for l = 1, 2, 3, . . ., let Gl ∈ H 2 () be defined by Gl = F 1 , i.e.


l

ζ (s)
Gl (s) = (l −s − l −1 ) , s ∈ .
s

Also, let E ∈ H 2 () be defined by


4 N. Derevyanko et al.

1
E(s) = , s ∈ .
s
Now we can state the reformulation of Theorem 1.1 which was used in paper [2].
Theorem 1.2 The following statements are equivalent:
(1) The Riemann Hypothesis is true;
(2) E belongs to the closed linear span of the set {Gl : l = 1, 2, 3 . . .};
(3) E belongs to the closed linear span of the set {Fλ : 0 ≤ λ ≤ 1}.
The plan of the proof is to verify three implications: 1 → 2, 2 → 3 and 3 → 1.
The first implication is the most challenging of all three. It is proven using
some famous results obtained under the assumption that the Riemann Hypothesis
is true, among which are Littlewood’s Theorem 1.3 and the Lindelöf Hypothe-
sis 1.2, and some standard techniques of functional analysis, particularly concerning
convergence in the norm. More details can be found in the original paper by
B. Bagchi [2].

Hypothesis 1.2 (Lindelöf) If the Riemann Hypothesis is true, then


 
1
∀ε > 0, ζ + it = O(t ε ).
2

Remark A very interesting and novel approach to the Lindelöf hypothesis is


presented by A. Fokas [11].
Theorem 1.3 (Littlewood) If the following conditions are satisfied:

• limr→1− ∞ n=0 r cn = a, ∀i ci ∈ C, a ∈ C,
n

• cn = O( n1 ),
then


cn = a.
n=0

The second implication follows from the embedding

{Gl : l = 1, 2, 3 . . .} ⊂ {Fλ : 0 ≤ λ ≤ 1}.

To prove the third implication (3 → 1), suppose that the Riemann Hypothesis is
false. Then ∃s0 = σ0 + it0 : ζ (s0 ) = 0 and σ0 = 12 , which implies that ∀0 ≤ λ ≤
1 Fλ (s0 ) = 0. That together with statement 3 gives that E(s0 ) = s10 = 0, i.e. 0 = 1.
This contradiction completes the proof.
On a Category of Cotangent Sums Related to the Nyman-Beurling Criterion for. . . 5

1.2 The Cotangent Sum’s Applications to Problems Related to


the Riemann Hypothesis

The main motivation behind the study of the cotangent sum (1) follows from
Theorem 1.1, which constitutes an equivalent form of the Riemann Hypothesis.
Asymptotics for dN (4) under the assumption that the Riemann Hypothesis is
true have been studied in several papers. S. Bettin, J. Conrey and D. Farmer in [9]
obtained the following result.
Theorem 1.4 If the Riemann Hypothesis is true and if
 1 3
T 2 −δ
|ζ (ρ)|2
|I m(ρ)|≤T

for some δ > 0, with the sum on the left hand side taken over all distinct zeros ρ of
the Riemann zeta function with imaginery part less than or equal to T , then
 +∞ 
    2
1  2 + γ − log 4π
1 − ζ 1 + it VN 1 + it  dt

2π  2 2  1
+ t2 log N
−∞ 4

for

N 
 
log n μ(n)
VN (s) := 1− . (6)
log N ns
n=1

We should mention that in the sequel γ stands for the Euler–Mascheroni constant.
Also, here μ is the Möbius function.
Also, from results of [9] it follows that under some restrictions, the infimum
from (4) is attained for DN = VN .
Nevertheless, it is interesting to obtain an unconditional estimate for dN .
In order to proceed further, we shall study equation (4) in more detail. In
particular, we can expand the square in the integral:
 +∞  1
  
1

dN = inf 1−ζ + it DN + it
DN −∞ 2 2
   
1 1 dt
−ζ − it DN + it
4 +t
2 2 1 2

 +∞   2   2

 1  
  1 
 dt 
+ ζ 2 + it  DN 2 + it  1 + t 2 .
−∞ 4
6 N. Derevyanko et al.

The integral in the second summand can be expressed as


   2  it
 +∞  
ar a¯b r − 21 − 12
b ζ 1 + it  r dt
, (7)
 2  b 1
+ t2
1≤r,b≤N −∞ 4

where ai are from the Definition (5) of DN .


Therefore, the integral
  2  it
∞  1 
ζ ( + it) r dt
 2  b 1
+ t2
−∞ 4

plays an important role in the Nyman-Beurling criterion for the Riemann Hypoth-
esis. Moreover, one can prove that this integral can be expressed via the so-called
Vasyunin sum.
Definition 1.5 The Vasyunin sum is defined as follows:
  b−1

r  mr  π mr 
V := cot , (8)
b b b
m=1

where {x} = x − x, x ∈ R.


The following proposition holds true:
Proposition 1.1
   2  it
1  ∞ 
ζ 1 + it  r dt
2π(rb)1/2  2  b 1
+ t2
−∞ 4
      
log 2π − γ 1 1 b−r r π r b
= + + log − V +V . (9)
2 r b 2rb b 2rb b r

One can note that the only non-explicit function on the right hand side of this
formula is the Vasyunin sum.
The next equation connects this result with the cotangent sums in question.
Proposition 1.2 It holds that
   
r r̄
V = −c0 , (10)
b b

where r̄ is such that r̄r ≡ 1(mod b).


The cotangent sum c0 can also be used to describe some special values of the
Estermann zeta function.
On a Category of Cotangent Sums Related to the Nyman-Beurling Criterion for. . . 7

Definition 1.6 The Estermann zeta function E(s, br , α) is defined by the Dirichlet
series
  
r σα (n) exp( 2πbinr )
E s, , α = , (11)
b ns
n≥1

where Re s > Re α + 1, b ≥ 1, (r, b) = 1, and



σα (n) = dα. (12)
d|n

One can show that the Estermann zeta function E(s, br , α) satisfies the following
functional equation:
   1+α−2s
r 1 b
E s, , α = (1 − s)(1 + α − s)
b π 2π
 πα   

× cos E 1 + α − s, , α
2 b
   
πα r̄
− cos π s − E 1 + α − s, − , α , (13)
2 b

where r is such that r̄r ≡ 1(mod b).


Properties of E(0, br , 0) were used by R. Balasubramanian, J. Conrey, and D.
Heath-Brown [3] to prove an asymptotic formula for
   2   2
T    
I= ζ 1 + it  A 1 + it  dt, (14)
 2   2 
0

where A(s) is a Dirichlet polynomial.


Asymptotic results for the function I as well as other functions of this type, are
useful for estimating a lower bound for the portion of zeros of the Riemann zeta
function ζ (s) on the critical line.
The following result of M. Ishibashi from [13] concerning E(s, br , α) for s =
0, provides the connection of the Estermann zeta function with the cotangent sum
c0 ( br ), simply by setting α = 0.
Theorem 1.5 (Ishibashi) Let b ≥ 2, 1 ≤ r ≤ b, (r, b) = 1, α ∈ N ∪ {0}. Then
for even α, it holds that

 r   i α+1 
b−1
m  π mr  1
E 0, , α = − cot(α) + δα,0 , (15)
b 2 b b 4
m=1

where δα,0 is the Kronecker delta function.


8 N. Derevyanko et al.

For odd α, it holds that


 r  Bα+1
E 0, , α = . (16)
b 2(α + 1)

In the special case when r = b = 1, we have

(−1)α+1 Bα+1
E(0, 1, α) = ,
2(α + 1)

where Bm is the m-th Bernoulli number and B2m+1 = 0,

(−1)m+1 2(2m)!
B2m = ζ (2m), f or m ≥ 1.
(2π )2m

Thus for b ≥ 2, 1 ≤ r ≤ b, (r, b) = 1, it follows that


 r  1  
i r
E 0, , 0 = + c0 , (17)
b 4 2 b

where c0 ( br ) is our cotangent sum from (1).

2 Central Properties of the Cotangent Sum c0

The function c0 is thoroughly studied in the papers of S. Bettin and J. Conrey [7,
8], where they have established a reciprocity formula for it, which encapsulates
important information about the behaviour of these sums. However, before we state
the formula itself, we shall give several definitions.
For a ∈ C and Im (s) > 0, consider


Sa (s) := σa (n)e(ns)e2π ins ,
n=1

2
Ea (s) := 1 + Sa (s).
ζ (−a)

It is worth mentioning that for a = 2k +1, k ∈ Z≥1 , Ea is the well known Eisenstein
series of weight 2k + 2.
On a Category of Cotangent Sums Related to the Nyman-Beurling Criterion for. . . 9

Definition 2.1 For a ∈ C and I m(s) > 0, define the function


 
1 1
ψa (s) := Ea+1 (s) − Ea+1 − . (18)
s a+1 s

For a = 2k, k ≥ 2, the function ψa (s) is equal to zero, because of the modularity
property of the Eisenstein series. Unfortunately, it is not true for other values of a,
but the functions ψa (s) have some remarkable properties, which were described in
detail by S. Bettin and J. Conrey.
Now we can state the theorem proven in paper [7].
Theorem 2.1 The function c0 satisfies the following reciprocity formula:
     
r b b 1 i r
c0 + c0 − = ψ0 . (19)
b r r 2π r 2 b

This result implies that the value of c0 ( br ) can be computed within a prescribed
accuracy in a polynomial of log b.
S. Bettin and J. Conrey highlighted that the reciprocity formula from 2.1 is very
similar to that of the Dedekind sum 5.1. We will consider Dedekind sums in more
detail in section 5 of this paper.
In [8] the result for c0 was generalized for the sums
  
b−1    
r π mr m
ca := ba cot ζ − a, , (20)
b b b
m=1

where ζ (s, x) is the Hurwitz zeta function.


After the introduction of the sum c0 by Bettin and Conrey and the proof of their
reciprocity formula, these sums have also been very extensively studied in a number
of works by H. Maier and M. Th. Rassias by the use of different techniques.
We shall now state some crucial results concerning the cotangent sum c0 .
In [22], M. Th. Rassias proved – using elementary techniques – the following
asymptotic formula:
Theorem 2.2 For b ≥ 2, b ∈ N, we have
 
1 1 b
c0 = b log b − (log 2π − γ ) + O(1). (21)
b π π

Subsequently in [18], M. Th. Rassias and H. Maier established an improvement,


or rather an asymptotic expansion, of Theorem 2.2.
Theorem 2.3 Let b, n ∈ N, b ≥ 6N, with N =  n2  + 1. There exist absolute
real constants A1 , A2 ≥ 1 and absolute real constants El , where l ∈ N, with
|El | ≤ (A1 l)2l , such that for each n ∈ N we have
10 N. Derevyanko et al.

 
1 
n
1 1 b
c0 = b log b − (log 2π − γ ) − + El b−l + Rn∗ (b), (22)
b π π π
l=1

where

|Rn∗ | ≤ (A2 n)4n b−(n−1) .

One could obtain Theorems 2.2 and 2.3 by the use of the reciprocity formula of
Bettin and Conrey, but the proofs of Maier and Rassias follow a different method.
The proof of Theorems 2.2 and 2.3 in [22] and [18], respectively, were obtained
using a common underlying idea proposed in [22]. First of all, one can obtain the
following relation between sums of cotangents and sums with fractional parts:

 b(1 − 2{a/b}) 
b  π m  sin( 2π m a)
= cot b
. (23)
a b a
a≥1 a≥1 m=1
ba ba

This relation provided the following proposition:


Proposition 2.1 For every positive integer b ≥ 2, we have
 
1 1  b(1 − 2{a/b})
c0 = . (24)
b π a
a≥1
ba

Then the difficulty lies in obtaining a good approximation of the sum S(L; b)
defined by

 1 a
S(L; b) := 2b . (25)
a b
1≤a≤L

The difference between the estimates from Theorems 2.2 and 2.3 is that stronger
approximation techniques were applied in [18] to obtain more information about
S(L; b). Namely, the generalized Euler summation formula (26) was used to
improve the result of Theorem 2.2.
Definition 2.2 If f is a function that is differentiable at least (2N + 1) times in
[0, Z], let
 Z
1
rN (f, Z) = (u − u + B)(2N +1) f (2N +1) (u)du,
(2N + 1)! 0

with the following notation:


On a Category of Cotangent Sums Related to the Nyman-Beurling Criterion for. . . 11


2N +1  
(2N +1) (2N +1) 2N + 1
(u−u+B) = ((u−u)+B) := (u−u)j B2N +1−j ,
j
j =0

where B2j are the Bernoulli numbers.


Theorem 2.4 (Generalized Euler summation formula) Let f be (2N + 1) times
differentiable in the interval [0, Z]. Then


Z 
f (0) + f (Z) Z
f (ν) = + f (u)du
2 0
ν=0


N
B2j
+ (f (2j −1) (Z) − f (2j −1) (0)) + rN (f, Z). (26)
(2j )!
j =1

Particularly, H. Maier and M. Th. Rassias used Theorem 2.4 to obtain the
following new representation for S(L; b).
Theorem 2.5 For N ∈ N, we have
  
(k + 1)b − 1 1
S(L; b) = 2b k log + F1 (k, b)
kb − 1 2
k≤L/b

  
B2j 
N
L
+2b kF2j (k, b) + 2brN f, , (27)
2j b
j =1 k≤L/b

where the function f satisfies



1
u, if u ≥ 1
f (u) =
0, if u = 0

and f ∈ C ∞ ([0, ∞)) with f (j ) (0) = 0 for j ≤ 2N + 1.


The new form of S(L; b) from (27) leads essentially to the proof of Theorem 2.3.
Furthermore, H. Maier and M. Th. Rassias obtained even more interesting results
concerning c0 ( br ) for a fixed arbitrary positive integer value of r and for large integer
values of b, which give us a deeper understanding of our cotangent sum for almost
all values of r and b.
Proposition 2.2 For r, b ∈ N with (r, b) = 1, it holds that
     
r 1 1 1 r
c0 = c0 − Q , (28)
b r b r b

where
12 N. Derevyanko et al.

  b−1  
r π mr rm
Q = cot .
b b b
m=1

Theorem 2.6 Let r, b0 ∈ N be fixed, with (b0 , r) = 1. Let b denote a positive


integer with b ≡ b0 (mod r). Then, there exists a constant C1 = C1 (r, b0 ), with
C1 (1, b0 ) = 0, such that
 
r 1 b
c0 = b log b − (log 2π − γ ) + C1 b + O(1) (29)
b πr πr

for large integer values of b.


We would like to mention that one could also prove the result of Theorem 2.6 via
the techniques introduced by Bettin and Conrey.

2.1 Ellipse
 
It is interesting to mention that if one examines the graph of c0 br for hundreds of
integer values of b by the use of MATLAB, the resulting Figs. 1 and 2 always have
a shape similar to an ellipse.
In 2014 H. Maier and M. Th. Rassias (and M. Th Rassias in his PhD thesis
[23]) [18] tried to explain this phenomenon and obtained an important result (later

Fig. 1 axis Ox : r, axis Oy : c0 , b = 1021

Fig. 2 axis Ox : r, axis Oy : c0 , b = 1357


On a Category of Cotangent Sums Related to the Nyman-Beurling Criterion for. . . 13

generalized by S. Bettin in [6]), which establishes the equidistribution of certain


normalized cotangent sums with respect to a positive measure. The result of Maier
and Rassias is presented in the following theorem.
Definition 2.3 For z ∈ R, let

F (z) = meas{x ∈ [0, 1] : g(x) ≤ z},

where “meas” denotes the Lebesgue measure,

+∞
 1 − 2{lx}
g(x) := ,
l
l=1

and

C0 (R) = {f ∈ C(R) : ∀ε > 0, ∃ a compact set K ⊂ R, such that |f (x)| < ε, ∀x ∈


/ K}.

Remark The convergence of the above series has been investigated by R.Bretèche
and G.Tenenbaum (see Theorem 4.2). It depends on the partial fraction expansion
of the number x.
Theorem 2.7
(i) F is a continuous function of z.
(ii) Let A0 , A1 be fixed constants, such that 1/2 < A0 < A1 < 1. Let also
 1  g(x) 2k
Hk = dx,
0 π

so Hk is a positive constant depending only on k, k ∈ N.


There is a unique positive measure μ on R with the following properties:
(a) For α < β ∈ R we have

μ([α, β]) = (A1 − A0 )(F (β) − F (α)).

(b)
 
(A1 − A0 )Hk/2 , for even k
x dμ =
k
0, otherwise.

(c) For all f ∈ C0 (R), we have


14 N. Derevyanko et al.

    
1 1 r
lim f c0 = f dμ,
b→+∞ φ(b) b b
r:(r,b)=1,
A0 b≤r≤A1 b

where φ(·) denotes the Euler phi-function.


As mentioned above, this result was later generalized by the use of a different
method by S. Bettin in [6].
Outline of the proof of Theorem 2.7. In [18] H. Maier and M. Th. Rassias proved
Theorem 2.3, which constitutes an improvement of their earlier
  Theorem 2.2.
Additionally, they investigated the cotangent sum c0 br for a fixed arbitrary
positive integer value of r and for large integer values of b and proved Theorem 2.6
as well as the following results
Theorem 2.8 Let k ∈ N be fixed. Let also A0 , A1 be fixed constants such that
1
2 < A0 < A1 < 1. Then there exists a constant Ek > 0, depending only on k,
such that
(a)

  2k
r
Q = Ek (A12k+1 − A02k+1 )b4k φ(b)(1 + o(1)) (b → +∞),
b
r:(r,b)=1
A0 b=r=A1 b

(b)

  2k−1
r
Q = o(b4k−2 φ(b)) (b → +∞),
b
r:(r,b)=1
A0 b=r=A1 b

(c)

  2k
r
c0 = Hk (A1 − A0 )b2k φ(b)(1 + o(1)) (b → +∞),
b
r:(r,b)=1
A0 b=r=A1 b

(d)

  2k−1
r
c0 = o(b2k−1 φ(b)) (b → +∞),
b
r:(r,b)=1
A0 b=r=A1 b

with
On a Category of Cotangent Sums Related to the Nyman-Beurling Criterion for. . . 15

Hk
Ek = .
(2k + 1)

Using the method of moments,   one can deduce detailed information about the
distribution of the values of c0 br , where A0 b ≤ r ≤ A1 b and b → +∞. Namely,
one can prove Theorem 2.7.

3 The Maximum of c0 in Rational Numbers in Short


Intervals

In this section we consider some results about the maximum of c0 in rational


numbers in short intervals. More precicely, consider the following definition:
Definition 3.1 Let 0 < A0 < 1, 0 < C < 1/2. For b ∈ N we set

 := (b, C) = b−C .

We define
  
 
M(b, C, A0 ) := max c0 r .
A0 b≤r<(A0 +)b  b 

In [19] H. Maier and M. Th. Rassias proved the following theorems.


Theorem 3.1 Let D satisfy 0 < D < 1
2 − C. Then we have for sufficiently large b:

D
M(b, C, A0 ) ≥ b log b.
π
An important part of the proof is the following key proposition.
Proposition 3.1 Let a0 ; a1 , a2 , . . . , an  be the continued fraction expansion of r̄
b ∈
Q. Moreover, let uvll be the l-th partial quotient of br̄ . Then
   (−1)l  1   
r vl−1
c0 = −b +ψ .
b vl π vl vl
1≤l≤n

Here ψ is an analytic function satisfying

log(2π x) − γ
ψ(x) = − + O(log x), (x → 0).
πx
The proposition was proven in [6] by S. Bettin.
16 N. Derevyanko et al.

Definition 3.2 Let  be as in Definition 3.1 and  > 0. We set

N (b, , ) := #{r : A0 b ≤ r < (A0 + )b, |r̄| = b}.

Another key proposition, proven in [19], is the following:


Proposition 3.2 Let ε < 0 be such that

1
D+ε < − C.
2
Set

 := b−(D+ε) .

Then for sufficiently large b it holds

N(b, , ) > 0.

s
ui
Let vi i=1 be the sequence of partial fractions of such br̄ . From

r̄ 1
≥ ≥
b v1 + 1

we obtain

v1 + 1 ≥ −1 .

Then by Proposition 3.2 we have


  1  
vl−1

+ψ < 2ε log b, for b ≥ b0 (ε).
π vl vl
l>1

Therefore,
  
 
c0 r  ≥ 2 log(−1 (1 + o(1))) b → +∞.
 b  π
This proves Theorem 3.1.
Theorem 3.2 Let C be as in Theorem 3.1 and let D satisfy 0 < D < 2 − C − E,
where E ≥ 0 is a fixed constant. Let B be sufficiently large. Then we have

D
M(b, C, A0 ) ≤ b log b
π

for all b with B ≤ b < 2B, with at most B E exceptions.


On a Category of Cotangent Sums Related to the Nyman-Beurling Criterion for. . . 17

Proof We will need the following proposition.


Proposition 3.3 Let ε > 0, B ≥ Bε, B < b ≤ 2B. For 1 ≤ r < b, (r, b) = 1,
let { uvii }si=1 be the sequence of partial fractions of br̄ . Then there are at most 3 values
of l for which
 
1 vl−1
ψ ≥ log log b,
vl vl

and at most one value of l for which


 
1 vl−1
ψ ≥ ε log b.
vl vl

Proof Let li (i = 1, 2, 3, 4) be such that


 
1 vl−1
ψ ≥ log log b.
vl vl

Then we have

vl1 ≥ log log b, vl2 ≥ exp(vl1 ) ≥ log b,

vl3 ≥ exp(vl2 ) ≥ b, vl4 ≥ exp(vl3 ) ≥ exp(b),

in contradiction to vs ≤ b.
In the same manner we obtain from vlj ≥ ε log b, j = 1, 2 :

vs ≥ exp(exp((log b)ε )) > b.

Assume ε > 0 to be fixed but arbitrarily small, Z > 0 fixed but arbitrarily large.
Definition 3.3 By Proposition 3.3 there is at most one value of l for which
 
1 vl−1
ψ ≥ ε log b.
vl vl

In case of the existence of l, we write

ul−1 (r, b) = ul−1

and

vl−1 (r, b) = vl−1 .

Then for s, t with 1 ≤ s, t = Z, (s, t) = 1, and for fixed θ with 0 < θ < 1,
18 N. Derevyanko et al.

F(s, t) :
   
 r̄ s 
= (b, r, r̄) : B ≤ b < 2B, A0 b ≤ r = (A0 + )b,  −  ≤ θ, r r̄ ≡ 1(mod b) .
b t 

Now we can formulate a proposition.


Proposition 3.4
 
N(b, , ) ≤ |F(s, t)|
B≤b<2B 1≤s,t≤Z

By Dirichlet’s approximation theorem there is (C0 , D0 ) ∈ Z2 with 1 ≤ D0 ≤


B 2, (C0 , D0 ) = 1, such that
 
 −1 C0 
A − ≤ 1 . (30)
 0 D0  D0 B 2

Let us estimate the cardinality of the set F(s, t).


From A0 b ≤ r = (A0 + )b, the definition of r̄, and (30), we obtain
 
C0 u
r r̄ = y r+ , with y ∈ Z, (31)
D0 D0

which after multiplication by C0 D0 becomes

(C0 y − D0 r̄)(C0 r − u) = −D0 (C0 − r̄u). (32)

If C0 − r̄u = 0, one can deduce from the well-known estimate for the number
of divisors of an integer that for a given pair (r̄, u), there are at most O(B ε ) pairs
(r, y) such that (32) holds.
There are at most O(B ε ) pairs (r̄, u) such that C0 − r̄u = 0. Thus we obtain

|F(s, t)| = O(B 2+2ε−C θ ). (33)

From Proposition 3.4 and (33) we obtain for  = θ B



N(b, , ) = O(B 2+2ε−C θ ). (34)
B≤b<2B

We now apply (34) with θ = B − D0 , where

D > D > 2 − C − E.

If we choose ε > 0 sufficiently small, then we conclude from (34) the following:
For all b with B ≤ b < 2B we have, with at most B E exceptions:
On a Category of Cotangent Sums Related to the Nyman-Beurling Criterion for. . . 19

N(b, B 1−C , ) = 0.

Thus,
 
1 vl−1 D
ψ ≤ b log b(1 + o(1)), ∀l ≤ Z. (35)
vl vl π vl−1

The result of Theorem 3.2 follows now from Propositions 3.1 and 3.3.

4 The Function g(x) and Moments of c0

There is an interesting connection between the cotangent sums c0 and the function

+∞
 1 − 2{lx}
g(x) := , (36)
l
l=1

which, as we mentioned above, naturally  appeared in the investigation of the


moments of c0 and of the sum Q br , which is related to c0 by Proposition 2.2.
To be precise, this series is related to c0 by the important Theorem 2.8.
Later S. Bettin in his paper [6] extended the result of Theorem 2.8 and proved
the following:
Theorem 4.1 Let b ≥ 1 and k ≥ 0. Then


b  k
1 r
c0 = Hk bk + Oε (bk−1+ε (Ak log b)2k ), (37)
φ(b) b
r=1
(r,b)=1

for some absolute constant A > 0 and any ε > 0.


Moreover, if 0 ≤ A0 < A1 ≤ 1, then we have

  k
1 r 1
c0 = (A1 − A0 )Hk bk + Oε (bk− 2 +ε (Ak log b)2k ). (38)
φ(b) b
(r,b)=1
A0 < br <A1

The function g(x) is interesting not only in connection to the study of the
cotangent sums c0 , but also in its own right. For example, it is also studied in [10]
by R. Bretèche and G. Tenenbaum.
Theorem 4.2 For each x ∈ Q the series g(x) converges.
For x ∈ R \ Q, the series g(x) converges if and only if the series
20 N. Derevyanko et al.

 log qm+1
(−1)m
qm
m≥1

converges, where (qm )m≥1 denotes the sequence of partial denominators of the
continued fraction expansion of x.
Proof The statement of the theorem is part of Theorem 4.4 of the paper by R.
Bretèche and G. Tenenbaum in [10].
The function g(x) also has the following property:
Theorem 4.3 The series
+∞
 1 − 2{lx}
g(x) =
l
l=1

converges almost everywhere in [0, 1).


The function g(x) was also of interest to L. Báez-Duatre, M. Balazard, B.
Landreau and E. Saias. In [1] they studied the function
 +∞ dt
A(λ) := {t}{λt} .
0 t2

and proved the following theorem.


Theorem
  4.4 Let λ > 0 be such that the series g(λ) converges. Then the series
1
g λ converges too, and we have:
 
1−λ λ+1 1
A(λ) = log λ + (log 2π − γ ) − g(λ) − λg .
2 2 λ

Now let us show an important property of g(x), which was proven in [20].
Theorem 4.5 There are constants c1 , c2 > 0, such that
 1
c1 (2k + 1) ≤ g(x)2k dx ≤ c2 (2k + 1)
0

for all k ∈ N, where (·) stands for the gamma function.


Sketch of a proof Let us consider the continued fraction expansion of x

x = [a0 (x); a1 (x), . . . , ak (x), . . . ].


On a Category of Cotangent Sums Related to the Nyman-Beurling Criterion for. . . 21

The ak (x) are obtained via the Gauss map α, defined by


 
1 1
α(x) = , αk (x) = α(αk−1 (x)), ak (x) = .
x αk−1 (x)

Definition 4.1 Let x ∈ X = (0, 1)\Q. Let also

βk (x) = α0 (x)α1 (x) . . . αk (x), β−1 (x) = 1,

1
γk (x) = βk−1 (x) log , where k ≥ 0,
αk (x)

so that γ0 (x) = log x1 .


The number x is called a Wilton number if the series

(−1)k γk (x)
k≥0

converges.
Wilton’s function W is defined by

W= (−1)k γk (x)
k≥0

for each Wilton number x.


M. Balazard and B. Martin proved in [4] the following proposition:
Proposition 4.1 There is a bounded function H : (0, 1) → R, which is continuous
at every irrational number, such that

g(x) = W(x) + H (x) (39)

almost everywhere. Also a number x ∈ X is a Wilton number if and only if α(x) is


a Wilton number. In this case, we have:

W(x) = l(x) − T W(x), (40)

where
1
l(x) = log
x
and the operator T is defined by

Tf (x) = xf (α(x)).
22 N. Derevyanko et al.

One can express (40) as

l(x) = (1 + T )W(x). (41)

The main idea in the evaluation of


 1
g(x)2k dx
0

is to solve the operator equation (41) for W(x), which is:

W(x) = (1 + T )−1 l(x). (42)

An idea which has long been used in functional analysis for the case when T is
a differential operator is to express the right-hand side of (42) as a Neumann series,
which is obtained by the geometric series identity, i.e.

+∞

(1 + T )−1 = (−1)k T k .
k=0

Thus one can approximate W(x) by


n
L(x, n) = (−1)k (T k l)(x). (43)
k=0

Definition 4.2 The measure m is defined by



1 dx
m(E) = ,
log 2 1+x
E

where E is any measurable subset of (0, 1).


Proposition 4.2 For f ∈ Lp we have

1 1
|T f (x)| dm(x) ≤ g
n p (n−1)p
|f (x)|p dm(x),
0 0

where

5−1
g= < 1.
2
On a Category of Cotangent Sums Related to the Nyman-Beurling Criterion for. . . 23

Proof Marmi, Moussa, and Yoccoz, in their paper [21], consider a generalized
continued fraction algorithm, depending on a parameter α, which becomes the usual
continued fraction algorithm for the choice α = 1. The operator Tv is defined in
(2.5) of [21] and becomes T for α = 1, v = 1. Then, Proposition 4.2 is the content
of formulas (2.14), (2.15) of [21].
Using standard techniques of functional analysis one can prove that

1/2 
 
lim L(x, n)L − W(x)L  dx = 0.
n→∞
0

One can eventually prove that


⎛ ⎞
1/2 1/2
⎜ ⎟
L(x, n)2k dx = ⎝ l(x)2k ⎠ (1 + o(1)) = (2k + 1)(1 + o(1)), (k → +∞).
0 0

The order of magnitude of

1/2
g(x)2k dx
0

now follows from (39), by the binomial theorem, since H (x) is a bounded
function.
Corollary 4.1 The series
 Hk
xk
(2k)!
k≥0

has radius of convergence π 2 .


H. Maier and M. Th. Rassias proved in [14] an improvement of Theorem 4.5,
by establishing an asymptotic result for the corresponding integral. Namely, they
proved the following theorem.
Theorem 4.6 Let
∞
{t}2
A= dt
t2
0

and K ∈ N. There is an absolute constant C > 0 such that


24 N. Derevyanko et al.

1
|g(x)|K dx = 2e−A (K + 1)(1 + O(exp(−CK)))
0

for K → ∞.
In [15], they improved this result settling also the general case of arbitrary
exponents K.
Theorem 4.7 Let K ∈ R, K > 0. There is an absolute constant C > 0 such that

1

|g(x)|K dx = (K + 1)(1 + O(exp(−CK)))
π
0

for K → ∞, where γ is the Euler-Mascheroni constant.

5 Dedekind Sums

Dedekind sums have applications in many fields of mathematics, especially in


number theory. These sums appear in R. Dedekind’s study of the function


π is
η(s) = e 12 (1 − e2π ims ), (44)
m=1

where Im s > 0.
Definition 5.1 Let r, b be integers, (r, b) = 1, k ≥ 1. Then the Dedekind sum s( br )
is defined as follows

r  b 
 rμ   μ 
s := , (45)
b b b
μ=1

where ((·)) is the sawtooth function defined as follows:



x − [x] − 1
if x is not an integer,
((x)) := 2 (46)
0 if x is an integer.

It is a fascinating fact that the Dedekind sum can also be expressed as a sum of
cotangent products:
On a Category of Cotangent Sums Related to the Nyman-Beurling Criterion for. . . 25

Proposition 5.1
     
1 
b−1
r πm π mr
s =− cot cot .
b 4b b b
m=1

It is a well-known fact that Dedekind sums satisfy a reciprocity formula:


Theorem 5.1 (Dedekind sums’ reciprocity formula)
     
r b 1 1 r b
s +s − = + −3 . (47)
b r 12rb 12 b r

We shall not present the proof of Proposition 5.1 and of Theorem 5.1. The
interested reader can find these proofs, as well as more fundamental facts concerning
Dedekind sums in the famous book by Rademacher and Grosswald [12].
It is interesting to study the relation between the cotangent sums c0 and the
Dedekind sums. We’ve already considered ca for arbitrary a ∈ C (see (20)). S.
Bettin in [5] proved the very interesting result that c−1 is a Dedekind sum up to a
constant:
Proposition 5.2 It holds that
   
r 1 r
s = c−1 . (48)
b 2π b

6 Sums Appearing in the Nyman-Beurling Criterion for the


Riemann Hypothesis Containing the Möbius Function

In a recent paper [16] H. Maier and M. Th. Rassias investigated the following
sums (36)
  
n
μ(n)g , (49)
b
n∈I

for a suitable interval I , where μ(n) is the Möbius function and the function g(x) is
as defined in Definition 2.3.
These sums appear in the study of the integral
 +∞ 
  2   2
  
ζ 1 + it  DN 1 + it  dt
(50)
 2   2  1
+ t2
−∞ 4
26 N. Derevyanko et al.

Particularly we could express (50) using formulas (6) and (7) and Proposition 1.1,
as follows:
  
+∞  2   2
  
ζ 1 + it  DN 1 + it  dt
 2   2  + t2 1
−∞ 4
    
log r log b
= μ(r)μ(b) 1 − 1−
log N log N
1≤r,b≤N
       
log 2π − γ 1 1 b−r r π r b
× + + log − V +V
2 r b 2rb b 2rb b r

If we expand the last equation, we will obtain the following sum for fixed b

    
log n 1 n
μ(n) 1 − V ,
log N n b
n∈I

which is equal to

    
log n n
μ(n) 1 − g .
log N b
n∈I

In [16] H. Maier and M. Th. Rassias proved the following result concerning the
sums (49):
Theorem 6.1 Let 0 ≤ δ ≤ D/2, b2δ ≤ B ≤ bD , where b−δ ≤ η ≤ 1. Then there is
a positive constant β depending only on δ and D, such that

  
n
μ(n)g = O((ηBb)1−β ). (51)
b
Bb≤n≤(1+η)Bb

Finally, the above result was recently improved by the same authors in [17], by
proving the following theorem:
Theorem 6.2 Let D ≥ 2. Let C be the number which is uniquely determined by

5+1 1
C≥ , 2C − log C − 1 − 2 log 2 = log 2 .
2 2
Let v0 be determined by
⎛  ⎞
 −1 −1
log 2 4
v0 ⎝1 − 1 + 2 log 2 C + +2+ C⎠ = 2 .
2 log 2
On a Category of Cotangent Sums Related to the Nyman-Beurling Criterion for. . . 27

 
Let z0 := 2 − 2 + 4
log 2 C v0 . Then for all ε > 0 we have

  
n
μ(n)g ε bD−z0 +ε .
b
bD ≤n<2bD

References

1. L. Báez-Duatre, M. Balazard, B. Landreau, E. Saias, Etude de l’autocorrelation multiplicative


de la fonction ‘partie fractionnaire’. Ramanujan J. 9, 215–240 (2005)
2. B. Bagchi, On Nyman, Beurling and Baez-Duarte’s Hilbert space reformulation of the Riemann
hypothesis. Proc. Indian Acad. Sci. Math. Sci. 116(2), 137–146 (2006)
3. R. Balasubramanian, J. Conrey, D. Heath-Brown, Asymptoticmeansquare of the product of
the Riemann zeta-function and a Dirichlet polynomial. J. Reine Angew. Math. 357, 161–181
(1985)
4. M. Balazard, B. Martin, Sur l’autocorrélation multiplicative de la fonction “partie fraction-
naire” et une fonction définie par. J.R. Wilton, arXiv:1305.4395v1
5. S. Bettin, A generalization of Rademacher’s reciprocity law. Acta Arithmetica 159(4), 363–374
(2013)
6. S. Bettin, On the distribution of a cotangent sum. Int. Math. Res. Not. 2015(21), 11419–11432
(2015)
7. S. Bettin, J. Conrey, A reciprocity formula for a cotangent sum. Int. Math. Res. Not. 2013(24),
5709–5726 (2013)
8. S. Bettin, J. Conrey, Period functions and cotangent sums. Algebra Number Theory 7(1), 215–
242 (2013)
9. S. Bettin, J. Conrey, D. Farmer, An optimal choice of Dirichlet polynomials for the Nyman-
Beurling criterion. (in memory of Prof. A. A. Karacuba), arXiv:1211.5191
10. R. de la Bretèche, G. Tenenbaum, Series trigonometriques à coefficients arithmetiques. J. Anal.
Math. 92, 1–79 (2004)
11. A. Fokas, A novel approach to the Lindelöf hypothesis. arXiv:1708.06607v4 [math.CA]
12. E. Grosswald, H. Rademacher, Dedekind sums. Mathematical Association of America (1972).
https://doi.org/10.5948/UPO9781614440161
13. M. Ishibashi, The value of the Estermann zeta function at s = 0. Acta Arith. 73(4), 357–361
(1995)
14. H. Maier, M.Th. Rassias, Asymptotics for moments of certain cotangent sums. Houst. J. Math.
43(1), 207–222 (2017)
15. H. Maier, M.Th. Rassias, Asymptotics for moments of certain cotangent sums for arbitrary
exponents. Houst. J. Math. 43(4), 1235–1249 (2017)
16. H. Maier, M.Th. Rassias, Estimates of sums related to the Nyman-Beurling criterion for the
Riemann Hypothesis. J. Number Theory 188, 96–120 (2018)
17. H. Maier, M.Th. Rassias, Explicit estimates of sums related to the Nyman-Beurling criterion
for the Riemann Hypothesis. J. Funct. Anal. (in press). https://doi.org/10.1016/j.jfa.2018.06.
022
18. H. Maier, M.Th. Rassias, Generalizations of a cotangent sum associated to the Estermann zeta
function. Commun. Contemp. Math. 18(1), 1550078 (2016)
19. H. Maier, M.Th. Rassias, The maximum of cotangent sums related to Estermann’s zeta function
in rational numbers in short intervals. Applicable Anal. Discret. Math. 11, 166–176 (2017)
20. H. Maier, M.Th. Rassias, The order of magnitude for moments for certain cotangent sums. J.
Math. Anal. Appl. 429, 576–590 (2015)
28 N. Derevyanko et al.

21. S. Marmi, P. Moussa, J.-C. Yoccoz, The Brjuno functions and their regularity properties.
Commun. Math. Phys. 186, 265–293 (1997)
22. M.Th. Rassias, On a cotangent sum related to zeros of the Estermann zeta function. Appl.
Math. Comput. 240, 161–167 (2014)
23. M.Th. Rassias, Analytic investigation of cotangent sums related to the Riemann zeta function,
Doctoral Dissertation, ETH-Zürich (2014)
Recent Progress in the Study of
Polynomials with Constrained
Coefficients

Tamás Erdélyi

Abstract This survey gives a taste of the author’s recent work on polynomials
with constrained coefficients. Special attention is paid to unimodular, Littlewood,
Newman, Rudin-Shapiro, and Fekete polynomials, their flatness and ultraflatness
properties, their Lq norms on the unit circle including Mahler’s measure, and bounds
on the number of unimodular zeros of self-reciprocal polynomials with coefficients
from a finite set of real numbers. Some interesting connections are explored, and a
few conjectures are also made.

Keywords Unimodular · Littlewood · Newman · Rudin-Shapiro · Fekete ·


polynomials · Lq norms · Mahler’s measure · Zeros

2010 Mathematics Subject Classifications. 11C08, 41A17, 26C10, 30C15

Notation

Let Pn be the set of all algebraic polynomials of degree at most n with real
coefficients. Let Pnc be the set of all algebraic polynomials of degree at most n
with complex coefficients. Let
 

n
Kn := Qn : Qn (z) = ak z , ak ∈ C , |ak | = 1
k
.
k=0

The class Kn is often called the collection of all (complex) unimodular polynomials
of degree n. Let

T. Erdélyi ()
Department of Mathematics, Texas A&M University, College Station, TX, USA
e-mail: [email protected]

© Springer Nature Switzerland AG 2020 29


A. Raigorodskii, M. T. Rassias (eds.), Trigonometric Sums and Their Applications,
https://doi.org/10.1007/978-3-030-37904-9_2
30 T. Erdélyi

 

n
Ln := Qn : Qn (z) = ak z , ak ∈ {−1, 1}
k
.
k=0

The class Ln is often called the collection of all (real) unimodular polynomials of
degree n. Let D denote the open unit disk of the complex plane. We will denote
the unit circle of the complex plane by ∂D. We define the Mahler measure of Q
(geometric mean of Q on ∂D) by
  2π 
1
M0 (Q) := exp log |Q(eit )| dt
2π 0

for bounded measurable functions Q on ∂D. It is well known, see [76], for instance,
that

M0 (Q) = lim Mq (Q) ,


q→0+

where
  2π  q 1/q
1  it 
Mq (Q) := Q(e ) dt , q > 0.
2π 0

It is also well known that for a function Q continuous on ∂D we have

M∞ (Q) := max |Q(eit )| = max |Q(eit )| = lim Mq (Q) .


t∈[0,2π ] t∈R q→∞

It is a simple consequence of the Jensen formula that


n
M0 (Q) = |c| max{1, |zk |}
k=1

for every polynomial of the form


n
Q(z) = c (z − zk ) , c, zk ∈ C .
k=1

We define the Mahler measure (geometric mean of Q on [α, β])


  β 
1
M0 (Q, [α, β]) := exp log |Q(e )| dt
it
β −α α

for [α, β] ⊂ R and bounded measurable functions Q(eit ) on [α, β]. It is well known,
see [76], for instance, that
Recent Progress in the Study of Polynomials with Constrained Coefficients 31

M0 (Q, [α, β]) = lim Mq (Q, [α, β]) ,


q→0+

where, for q > 0,


  β  q 1/q
1  
Mq (Q, [α, β]) := Q(eit ) dt .
β −α α

If Q ∈ Pnc is of the form


n
Q(z) = aj zj , aj ∈ C ,
j =0

then its conjugate polynomial is defined by


n

Q (z) := z Q(1/z) :=
n
a n−j zj .
j =0

A polynomial Q ∈ Pnc is called conjugate-reciprocal if Q = Q∗ .


The Lebesgue measure of a measurable set A ⊂ R will be denoted by m(A)
throughout the paper.

1 Ultraflat Sequences of Unimodular Polynomials

By Parseval’s formula,
 2π  2
 
Pn (eit ) dt = 2π(n + 1)
0

for all Pn ∈ Kn . Therefore



min |Pn (eit )| ≤ n + 1 ≤ max |Pn (eit )| .
t∈R t∈R

An old problem (or rather an old theme) is the following.


Problem 1.1 (Littlewood’s Flatness Problem) How close can a polynomial Pn ∈
Kn or Pn ∈ Ln come to satisfying

|Pn (eit )| = n + 1, t ∈ R? (1)

Obviously (1) is impossible if n ≥ 1. So one must look for less than (1), but
then there are various ways of seeking such an “approximate situation”. One way
is the following. In his paper [87] Littlewood had suggested that, conceivably, there
32 T. Erdélyi

might exist a sequence (Pn ) of polynomials Pn ∈ Kn (possibly even Pn ∈ Ln )


such that (n + 1)−1/2 |Pn (eit )| converge to 1 uniformly in t ∈ R. We shall call
such sequences of unimodular polynomials “ultraflat”. More precisely, we give the
following definition.
Definition 1.2 Given a positive number ε, we say that a polynomial Pn ∈ Kn is
ε-flat if
√ √
(1 − ε) n + 1 ≤ |Pn (eit )| ≤ (1 + ε) n + 1 , t ∈ R.

Definition 1.3 Let (nk ) be an increasing sequence of positive integers. Given a


sequence (εnk ) of positive numbers tending to 0, we say that a sequence (Pnk ) of
polynomials Pnk ∈ Knk is (εnk )-ultraflat if each Pnk is (εnk )-flat. We simply say that
a sequence (Pnk ) of polynomials Pnk ∈ Knk is ultraflat if it is (εnk )-ultraflat with a
suitable sequence (εnk ) of positive numbers tending to 0.
The existence of an ultraflat sequence of unimodular polynomials seemed very
unlikely, in view of a 1957 conjecture of P. Erdős (Problem 22 in [68]) asserting
that, for all Pn ∈ Kn with n ≥ 1,

max |Pn (eit )| ≥ (1 + ε) n + 1 , (2)
t∈R

where ε > 0 is an absolute constant (independent of n). Yet, refining a method of


Körner [83], Kahane [79] proved that there
 exists
√ a sequence
 (Pn ) with Pn ∈ Kn
which is (εn )-ultraflat, where εn = O n−1/17 log n . (Kahane’s paper contained
though a slight error which was corrected in [100].) Thus the Erdős conjecture (2)
was disproved for the classes Kn . For the more restricted class Ln the analogous
Erdős conjecture is unsettled to this date. It is a common belief that the analogous
Erdős conjecture for Ln is true, and consequently there is no ultraflat sequence
of polynomials Pn ∈ Ln . An interesting result related to Kahane’s breakthrough
is given in [5]. For an account of some of the work done till the mid 1960s, see
Littlewood’s book [88] and [101].
Let (εn ) be a sequence of positive numbers tending to 0. Let the sequence (Pn )
of polynomials Pn ∈ Kn be (εn )-ultraflat. We write

Pn (eit ) = Rn (t)eiαn (t) , Rn (t) = |Pn (eit )| , t ∈ R. (3)

It is simple to show that αn can be chosen to be in C ∞ (R). This is going to be our


understanding throughout the paper. It is easy to find a formula for αn (t) in terms of
Pn . We have
 
eit Pn (eit )
αn (t) = Re , (4)
Pn (eit )
Recent Progress in the Study of Polynomials with Constrained Coefficients 33

see formulas (7.1) and (7.2) on p. 564 and (8.2) on p. 565 in [104]. The angular
function αn∗ and modulus function Rn∗ = Rn associated with the polynomial Pn∗ are
defined by

Pn∗ (eit ) = Rn∗ (t)eiαn (t) , Rn∗ (t) = |Pn∗ (eit )| .

Similarly to αn , the angular function αn∗ can also be chosen to be in C ∞ (R) on R.


By applying formula (4) to Pn∗ , it is easy to see that

αn (t) + αn∗ (t) = n , t ∈ R. (5)

The structure of ultraflat sequences of unimodular polynomials is studied in [45–


47], and [48], where several conjectures of Saffari are proved. These are closely
related to each other.
Conjecture 1.4 (Uniform Distribution Conjecture for the Angular Speed) Let
(Pn ) be a fixed ultraflat sequence of polynomials Pn ∈ Kn . With the notation (3),
in the interval [0, 2π ], the distribution of the normalized angular speed αn (t)/n
converges to the uniform distribution as n → ∞. More precisely, we have

m({t ∈ [0, 2π ] : 0 ≤ αn (t) ≤ nx}) = 2π x + on (x) (6)

for every x ∈ [0, 1], where on (x) converges to 0 uniformly on [0, 1]. As a
consequence, |Pn (eit )|/n3/2 also converges to the uniform distribution as n → ∞.
More precisely, we have

m({t ∈ [0, 2π ] : 0 ≤ |Pn (eit )| ≤ n3/2 x}) = 2π x + on (x)

for every x ∈ [0, 1], where on (x) converges to 0 uniformly on [0, 1].
The basis of this conjecture was that for the special ultraflat sequences of
unimodular polynomials produced by Kahane [79], (6) is indeed true.
In Section 4 of [45] we prove this conjecture in general.
In the general case (6) can, by integration, be reformulated (equivalently) in
terms of the moments of the angular speed αn (t). This was observed and recorded
by Saffari [104]. For completeness the proof of this equivalence is presented in
Section 4 of [45] and we settle Conjecture 1.4 by proving the following result.
Theorem 1.5 (Reformulation of the Uniform Distribution Conjecture) Let (Pn )
be a fixed ultraflat sequence of polynomials Pn ∈ Kn . For any q > 0 we have
 2π
1  q q
α (t) dt = n + on,q nq . (7)
n
2π 0 q +1
with suitable constants on,q converging to 0 for every fixed q > 0.
An immediate consequence of (7) is the remarkable fact that for large values of
n ∈ N, the Lq (∂D) Bernstein factors
34 T. Erdélyi

 2π  it q
P (e ) dt
0 n
 2π  
Pn (eit )q dt
0

of the elements of ultraflat sequences (Pn ) of unimodular polynomials are essen-


tially independent of the polynomials. More precisely Theorem 1.5 implies the
following result.
Theorem 1.6 (The Bernstein Factors) Let q be an arbitrary positive real number.
Let (Pn ) be a fixed ultraflat sequence of polynomials Pn ∈ Kn . We have
 2π  it q
P (e ) , dt nq+1
0 n
 2π  q = + on,q nq+1 ,
Pn (eit ) dt q +1
0

and as a limit case,

max0≤t≤2π |Pn (eit )|


= n + on n .
max0≤t≤2π |Pn (eit )|

with suitable constants on,q and on converging to 0 for every fixed q.


In Section 3 of [45] we prove the following result which turns out to be stronger
than Theorem 1.5.
Theorem 1.7 (Negligibility Theorem for Higher Derivatives) Let (Pn ) be a fixed
ultraflat sequence of polynomials Pn ∈ Kn . For every integer r ≥ 2, we have

max |αn(r) (t)| = on,r nr


0≤t≤2π

with suitable constants on,r converging to 0 for every fixed r = 2, 3, . . ..


We show in Section 4 of [45] how Theorem 1.4 follows from Theorem 1.7.
In Section 4 of [45] we also prove an extension of Saffari’s Uniform Distribution
Conjecture 1.4 to higher derivatives.
Theorem 1.8 (Distribution of the Modulus of Higher Derivatives of Ultraflat
Sequences of Unimodular Polynomials) Let (Pn ) be a fixed ultraflat sequence of
polynomials Pn ∈ Kn . The distribution of
 1/r
|Pn(r) (eit )|
nr+1/2

converges to the uniform distribution as n → ∞. More precisely, we have




m t ∈ [0, 2π ] : 0 ≤ |Pn(r) (eit )| ≤ nr+1/2 x r = 2π x + or,n (x)
Recent Progress in the Study of Polynomials with Constrained Coefficients 35

for every x ∈ [0, 1], where or,n (x) converges to 0 uniformly for every fixed r =
1, 2, . . ..
In [51], based on the results in [45], we proved yet another conjecture of
Queffelec and Saffari, see (1.30) in [101]. Namely we proved asymptotic formulas
for the Lq norms of the real part and the derivative of the real part of ultraflat
unimodular polynomials on the unit circle. A recent paper of Bombieri and Bourgain
[9] is devoted to the construction of ultraflat sequences of unimodular polynomials.
In particular, they obtained a much improved estimate for the error term. A major
part of their paper deals also with the long-standing problem of the effective
construction of ultraflat sequences of unimodular polynomials.
For λ ≥ 0, let
 
n
Kn := Pn : Pn (z) =
λ
ak k z , ak ∈ C , |ak | = 1 .
λ k

k=0

Ultraflat sequences (Pn ) of polynomials Pn ∈ Knλ are defined and studied


thoroughly in [67] where various extensions of Saffari’s conjectures have been
proved. Note that it is not yet known whether or not ultraflat sequences (Pn ) of
polynomials Pn ∈ Knλ exist for any λ > 0, in particular, it is not yet known for
λ = 1.
In [47] we examined how far an ultraflat unimodular polynomial is from being
conjugate-reciprocal, and we proved the following three theorems.
Theorem 1.9 Let (Pn ) be a fixed ultraflat sequence of polynomials Pn ∈ Kn . We
have
 2π  2  
1 it ∗ it 1
|Pn (e )| − |Pn (e )| dt = + γn n3 ,
2π 0 3

with some constants γn converging to 0.


Theorem 1.10 Let (Pn ) be a fixed ultraflat sequence of polynomials Pn ∈ Kn . If

n
Pn (z) = ak,n zk , k = 0, 1, . . . , n, n = 1, 2, . . . ,
k=0

then
   2  
n
 2 1 2π
  1
k 2 ak,n − a n−k,n  = (Pn − Pn∗ )(eit ) dt ≥ + hn n3 ,
2π 0 3
k=0

with some constants hn converging to 0.


Theorem 1.11 Let (Pn ) be a fixed ultraflat sequence of polynomials Pn ∈ Kn .
Using the notation of Theorem 1.10 we have
36 T. Erdélyi

  2π  2  
n
   
ak,n − a n−k,n 2 = 1 (Pn − Pn∗ )(eit ) dt ≥
1
+ hn n ,
2π 0 3
k=0

with some constants hn (the same as in Theorem 1.10) converging to 0.


There are quite a few recent publications on or related to ultraflat sequences of
unimodular polynomials. Some of them (not mentioned before) are [10, 95, 100,
105], and [92].

2 More Recent Results on Ultraflat Sequences of


Unimodular Polynomials

In a recent paper [64] we revisited the topic. Theorems 2.1–2.4 and 2.6 are new in
[64], and Theorems 2.5 and 2.7 recapture old results.
In our results below  denotes the usual gamma function, and the ∼ symbol
means that the ratio of the left and right hand sides converges to 1 as n → ∞.
Theorem 2.1 If (Pn ) is an ultraflat sequence of polynomials Pn ∈ Kn and q ∈
(0, ∞), then
 
 2π  q 2q  q+1
1   2
(Pn − Pn∗ )(eit ) dt ∼  q  √ nq/2 .
2π 0  2 +1 π

Our next theorem is a special case (q = 2) of Theorem 2.1. Compare it with


Theorem 1.11.
Theorem 2.2 Let (Pn ) be an ultraflat sequence of polynomials Pn ∈ Kn . If


n
Pn (z) = ak,n zk , k = 0, 1, . . . , n, n = 1, 2, . . . ,
k=0

then

  2π  2
n
   
ak,n − a n−k,n 2 = 1 (Pn − Pn∗ )(eit ) dt ∼ 2n .
2π 0
k=0

Our next theorem should be compared with Theorem 1.10.


Theorem 2.3 Let (Pn ) be an ultraflat sequence of polynomials Pn ∈ Kn . Using the
notation in Theorem 2.2 we have
Recent Progress in the Study of Polynomials with Constrained Coefficients 37

   2
n
 2 1 2π
  2n3
k 2 ak,n − a n−k,n  = (Pn − Pn∗ )(eit ) dt ∼ .
2π 0 3
k=0

We also proved the following result.


Theorem 2.4 If (Pn ) is an ultraflat sequence of polynomials Pn ∈ Kn and q ∈
(0, ∞), then
 
  q  q+1
1 2π d  2
 |(Pn − P ∗ )(eit )| dt ∼   √ n3q/2 .
2π  dt n  (q + 1) q2 + 1 π
0

As a Corollary of Theorem 2.2 we have recaptured Saffari’s “near orthogonality


conjecture” raised in [104] and proved first in [48].
Theorem 2.5 Let (Pn ) be a fixed ultraflat sequence of polynomials Pn ∈ Kn . Using
the notation in Theorem 2.2 we have


n
ak,n an−k,n = o(n) .
k=0

As a Corollary of Theorem 2.3 we have proved a new “near orthogonality”


formula.
Theorem 2.6 Let (Pn ) be a fixed ultraflat sequence of polynomials Pn ∈ Kn . Using
the notation in Theorem 2.2 we have


n
k 2 ak,n an−k,n = o(n3 ) .
k=0

Finally we have recaptured the asymptotic formulas for the real part and the
derivative of the real part of ultraflat unimodular polynomials proved in [51] first.
Theorem 2.7 If (Pn ) is an ultraflat sequence of unimodular polynomials Pn ∈ Kn ,
and q ∈ (0, ∞), then for fn (t) := Re(Pn (eit )) we have
 
 2π  q+1
1 2
|fn (t)|q dt ∼ q  √ nq/2
2π 0  2 +1 π

and
 
 q+1
1 2π  q 2 
f (t) dt ∼ q √ n3q/2 .
n
2π 0 (q + 1) 2 + 1 π
38 T. Erdélyi

We remark that trivial modifications of the proof of Theorems 2.1–2.7 yield that
the statement of the above theorem remains true if the ultraflat sequence (Pn ) of
polynomials Pn ∈ Kn is replaced by an ultraflat sequence (Pnk ) of polynomials
Pnk ∈ Knk , where (nk ) is an increasing sequence of positive integers.

3 Flatness of Conjugate-Reciprocal Unimodular Polynomials

There is a beautiful short argument to see that

M∞ (P ) ≥ 4/3 n1/2 (8)

for every conjugate-reciprocal unimodular polynomial P ∈ Kn . Namely, Parseval’s


formula gives
 1/2
m(m + 1)(2m + 1)
M∞ (P ) ≥ M2 (P ) = , P ∈ Kn .
3

Combining this with Lax’s Bernstein-type inequality


n
M∞ (P ) ≤ M∞ (P )
2
valid for all conjugate-reciprocal algebraic polynomials with complex coefficients
(see p. 438 in [17], for instance), we obtain
 1/2
2 n(n + 1)(2n + 1)
M∞ (P ) ≥ ≥ 4/3 m1/2
n 3

for all conjugate-reciprocal unimodular polynomials P ∈ Kn . In [56] we prove the


following results.
Theorem 3.1 There is an absolute constant ε > 0 such that

M1 (P ) ≤ (1 − ε) 1/3 n3/2

for every conjugate-reciprocal unimodular polynomial P ∈ Kn .


Theorem 3.2 There is an absolute constant ε > 0 such that

M∞ (P ) ≥ (1 + ε) 1/3 n3/2

for every conjugate-reciprocal unimodular polynomial P ∈ Kn .


Theorem 3.3 There is an absolute constant ε > 0 such that
Recent Progress in the Study of Polynomials with Constrained Coefficients 39

M∞ (P ) ≥ (1 + ε) 4/3 n1/2

for every conjugate-reciprocal unimodular polynomial P ∈ Kn .


Theorem 3.4 There is an absolute constant ε > 0 such that

Mq (P ) ≤ exp(ε(q − 2)/q) 1/3 n3/2 , 1 ≤ q < 2,

and

Mq (P ) ≥ exp(ε(q − 2)/q) 1/3 n3/2 , 2<q,

for every conjugate-reciprocal unimodular polynomial P ∈ Kn .


A polynomial P ∈ Pnc is called skew-reciprocal if P (−z) = P ∗ (z) for all z ∈ C.
A polynomial P ∈ Pnc is called self-reciprocal if P ∗ = P , that is, P (z) = zn P (1/z)
for all z ∈ C \ {0}.
Problem 3.5 Is there an absolute constant ε > 0 such that

M∞ (P ) ≥ (1 + ε) 1/3 n3/2

holds for all self-reciprocal and skew-reciprocal unimodular polynomials P ∈ Kn ?


Problem 3.6 Is there an absolute constant ε > 0 such that

M∞ (P ) ≥ (1 + ε) 1/3 n3/2

or at least

max |P (z)| − min |P (z)| ≥ εn3/2


z∈∂D z∈∂D

holds for all unimodular polynomials P ∈ Kn ?


Our method to prove Theorem 3.2 does not seem to work for all unimodular
polynomials P ∈ Kn . In an e-mail communication several years ago Saffari
speculated that the answer to Problem 3.6 is no. However, we know the answer
to neither Problem 3.6 nor Problem 3.5.
Let Lm be the collection of all polynomials of degree n with each of their
coefficients in {−1, 1}.
Problem 3.7 Is there an absolute constant ε > 0 such that

M∞ (P ) ≥ (1 + ε) 1/3 n3/2

or at least

max |P (z)| − min |P (z)| ≥ εn3/2


z∈∂D z∈∂D
40 T. Erdélyi

holds for all Littlewood polynomials P ∈ Ln ?


The following problem due to Erdős [68] is open for a long time.
Problem 3.8 Is there an absolute constant ε > 0 such that

M∞ (P ) ≥ (1 + ε)n1/2

or at least

max |P (z)| − min |P (z)| ≥ εn1/2


z∈∂D z∈∂D

holds for all Littlewood polynomials P ∈ Ln ?


The same problem may be raised only for all skew-reciprocal Littlewood
polynomials P ∈ Ln , and as far as we know, it is also open.

4 Average Lq Norm of Littlewood Polynomials on the Unit


Circle

P. Borwein and Lockhart [25] investigated the asymptotic behavior of the mean
value of normalized Mq norms of Littlewood polynomials for arbitrary q > 0. They
proved the following result.
Theorem 4.1

1  (Mq (f ))q  q
lim = 1+ .
n→∞ 2n+1 nq/2 2
f ∈Ln

In [32] we showed the following.


Theorem 4.2

1  Mq (f )   q 1/q
lim 1/2
=  1+
n→∞ 2n+1 n 2
f ∈Ln

for every q > 0.


In [32] we also proved the following result on the average Mahler measure of
Littlewood polynomials.
Theorem 4.3 We have
Recent Progress in the Study of Polynomials with Constrained Coefficients 41

1  M0 (f )
lim = e−γ /2 = 0.749306 · · · ,
n→∞ 2n+1 n1/2
f ∈Ln

where
 n 
1
γ := lim − log n = 0.577215 · · ·
n→∞ k
k=1

is the Euler constant.


These last two results are analogues of the results proved earlier by Choi and
Mossinghoff [35] for polynomials in Kn .

5 Rudin-Shapiro Polynomials

Finding polynomials with suitably restricted coefficients and maximal Mahler


measure has interested many authors. The classes Ln and Kn are two of the most
important classes considered. Observe that Ln ⊂ Kn and

M0 (Q) ≤ M2 (Q) = n+1

for every Q ∈ Kn .
It is open whether or not for every ε > 0 there is a sequence (Qn ) of polynomials
Qn ∈ Ln such that

M0 (Qn ) ≥ (1 − ε) n .

Beller and Newman [7] constructed a sequence (Qn ) of unimodular polynomials


Qn ∈ Kn such that
√ c
M0 (Qn ) ≥ n− .
log n

Littlewood asked if there were Qnk ∈ Lnk satisfying

c1 nk + 1 ≤ |Qnk (z)| ≤ c2 nk + 1 , z ∈ ∂D ,

with some absolute constants c1 > 0 and c2 > 0, see [10, p. 27] for a reference to
this problem of Littlewood. No sequence (Qnk ) of Littlewood polynomials Qnk ∈
Lnk is known that satisfies the lower bound. A sequence of Littlewood polynomials
that satisfies just the upper bound is given by the Rudin-Shapiro polynomials. The
Rudin-Shapiro polynomials appear in Harold Shapiro’s 1951 thesis [109] at MIT
and are sometimes called just Shapiro polynomials. They also arise independently
42 T. Erdélyi

in Golay’s paper [72]. They are remarkably simple to construct and are a rich source
of counterexamples to possible conjectures. The Rudin-Shapiro polynomials are
defined recursively as follows:

P0 (z) := 1 , Q0 (z) := 1 ,
k
Pk+1 (z) := Pk (z) + z2 Qk (z) ,
k
Qk+1 (z) := Pk (z) − z2 Qk (z) , k = 0, 1, 2, . . . .

Note that both Pk and Qk are polynomials of degree n − 1 with n := 2k having each
of their coefficients in {−1, 1}. In signal processing, the Rudin-Shapiro polynomials
have good autocorrelation properties and their values on the unit circle are small.
Binary sequences with low autocorrelation coefficients are of interest in radar,
sonar, and communication systems. It is well known and easy to check by using
the parallelogram law that

|Pk+1 (z)|2 + |Qk+1 (z)|2 = 2(|Pk (z)|2 + |Qk (z)|2 ) , z ∈ ∂D .

Hence

|Pk (z)|2 + |Qk (z)|2 = 2k+1 = 2n , z ∈ ∂D .

It is also well known (see Section 4 of [10], for instance), that

Qk (−z) = Pk∗ (z) = zn−1 Pk (1/z) , z ∈ ∂D ,

and hence

|Qk (−z)| = |Pk (z)| , z ∈ ∂D .

P. Borwein’s book [10] presents a few more basic results on the Rudin-Shapiro
polynomials. Various properties of the Rudin-Shapiro polynomials are discussed in
[29] by Brillhart and in [30] by Brillhart, Lemont, and Morton.
As for k ≥ 1 both Pk and Qk have odd degree, both Pk and Qk have at least one
real zero. The fact that for k ≥ 1 both Pk and Qk have exactly one real zero was
proved by Brillhart in [29]. Another interesting observation made in [30] is the fact
that Pk and Qk cannot vanish at root of unity different from −1 and 1.
Obviously

M2 (Pk ) = 2k/2

by the Parseval formula. In 1968 Littlewood [88] evaluated M4 (Pk ) and found that
Recent Progress in the Study of Polynomials with Constrained Coefficients 43

 1/4  1/4
4k+1 4n2
M4 (Pk ) ∼ = . (9)
3 3

The M4 norm of Rudin-Shapiro like polynomials on ∂D are studied in [26].


The merit factor of a Littlewood polynomial f ∈ Ln−1 is defined by

M2 (f )4 n2
MF(f ) = = .
M4 (f )4 − M2 (f )4 M4 (f )4 − n2

Observe that (9) implies that MF(Pk ) ∼ 3 .

6 Mahler Measure and Moments of the Rudin-Shapiro


Polynomials

Despite the simplicity of their definition not much is known about the Rudin-Shapiro
polynomials. It has been shown in [59] fairly recently that the Mahler measure (M0
norm) and the M∞ norm of the Rudin-Shapiro polynomials Pk and Qk of degree
n−1 with n := 2k on the unit circle of the complex plane have the same size, that is,
the Mahler measure of the Rudin-Shapiro polynomials of degree n − 1 with n := 2k
is bounded from below by cn1/2 , where c > 0 is an absolute constant.
Theorem 6.1 Let Pk and Qk be the k-th Rudin-Shapiro polynomials of degree n−1
with n = 2k . There is an absolute constant c1 > 0 such that

M0 (Pk ) = M0 (Qk ) ≥ c1 n , . . . k = 1, 2, . . . .

The following asymptotic formula, conjectured by Saffari in 1985, for the Mahler
measure of the Rudin-Shapiro polynomials has been proved recently in [61].
Theorem 6.2 We have
 1/2
M0 (Pk ) M0 (Qk ) 2
lim = lim = = 0.857763 · · · .
n→∞ n1/2 n→∞ n1/2 e

To formulate our next theorem we define

!k := 2−(k+1)/2 Pk
P and !k := 2−(k+1)/2 Qk .
Q

By using the above normalization, we have

|P !k (z)|2 = 1 ,
!k (z)|2 + |Q z ∈ ∂D .

For q > 0 let


44 T. Erdélyi


  2π
Iq (P !k ) q := 1
!k ) := Mq (P !k (eit )|q dt .
|P
2π 0

The following result is a simple consequence of Theorem 6.1.


Theorem 6.3 There exists a constant L < ∞ independent of k such that

 !k )
Im (P
< L, k = 0, 1, . . . .
m
m=1

Consequently

!k ) ≤ L
Im (P , k = 1, 2, . . . , m = 1, 2, . . . .
log(m + 1)

In [59] we also proved the following result.


Theorem 6.4 There exists an absolute constant c2 > 0 such that

M0 (Pk , [α, β]) ≥ c2 n , k = 1, 2, . . . ,

with n := 2k for all α, β ∈ R such that

12π (log n)3/2


≤ ≤ β − α ≤ 2π .
n n1/2

7 Lemmas for Theorem 6.1

As the proof of Theorem 6.1 is based on interesting new properties of the Rudin-
Shapiro polynomials which have been observed only recently in [59], we list them
in this section.
Lemma 7.1 Let k ≥ 2 be an integer, n := 2k , and let

2πj
zj := eitj , tj := , j ∈ Z.
n
We have

Pk (zj ) = 2Pk−2 (zj )

whenever j is even, and

Pk (zj ) = (−1)(j −1)/2 2i Qk−2 (zj )


Recent Progress in the Study of Polynomials with Constrained Coefficients 45

whenever j is odd, where i is the imaginary unit.


Lemma 7.2 If Pk and Qk are the k-th Rudin-Shapiro polynomials of degree n − 1
with n := 2k ,

ω := sin2 (π/8) = 0.146446 · · · ,

and
2πj
zj := eitj , tj := , j ∈ Z,
n
then

max{|Pk (zj )|2 , |Pk (zj +1 )|2 } ≥ ω2k+1 = 2ωn , j ∈ Z.

Lemma 7.3 Let n, m ≥ 1,

0 < τ1 ≤ τ2 ≤ · · · ≤ τm ≤ 2π , τ0 := τm − 2π , τm+1 := τ1 + 2π ,

δ := max{τ1 − τ0 , τ2 − τ1 , . . . , τm − τm−1 } .

For every A > 0 there is a B > 0 depending only on A such that


m 
τj +1 − τj −1 2π
log |P (e iτj
)| ≤ log |P (eiτ )| dτ + B
2 0
j =1

for all P ∈ Pnc and δ ≤ An−1 .

8 Saffari’s Conjecture on the Shapiro Polynomials

In 1980 Saffari conjectured the following.


Conjecture 8.1 Let Pk and Qk be the Rudin-Shapiro polynomials of degree n − 1
with n := 2k . We have

(2n)1/2
Mq (Pk ) = Mq (Qk ) ∼
(q/2 + 1)1/q

for all real exponents q > 0.


Conjecture 8.1* Equivalently to Conjecture 8.1, we have
46 T. Erdélyi

   
 Pk (eit ) 2
lim m t ∈ [0, 2π ) :  √  ∈ [α, β]

k→∞ 2k+1
   
 Qk (eit ) 2

= lim m t ∈ [0, 2π ) :  √ 
 ∈ [α, β] = 2π(β − α)
k→∞ 2k+1

whenever 0 ≤ α < β ≤ 1.
This conjecture was proved for all even values of q ≤ 52 by Doche [39] and
Doche and Habsieger [40]. Recently B. Rodgers [102] proved Saffari’s Conjec-
ture 8.1 for all q > 0. See also [44].
An extension of Saffari’s conjecture is Montgomery’s conjecture below.
Conjecture 8.2 Let Pk and Qk be the Rudin-Shapiro polynomials of degree n − 1
with n := 2k . We have
 
Pk (eit )
lim m t ∈ [0, 2π ) : √ ∈E
k→∞ 2k+1
 
Qk (eit )
= lim m t ∈ [0, 2π ) : √ ∈E = 2μ(E) ,
k→∞ 2k+1

where μ(E) denotes the Jordan measure of a Jordan measurable set E ⊂ D.


B. Rodgers [102] proved Montgomery’s Conjecture 8.2 as well.

9 Consequences of Saffari’s Conjecture

Let Pk and Qk be the Rudin-Shapiro polynomials of degree n − 1 with n := 2k ,

Rk (t) := |Pk (eit )|2 or Rk (t) := |Qk (eit )|2 ,

ω := sin2 (π/8) = 0.146446 · · · .

In [62] we proved Theorems 9.1–9.5 below.


Theorem 9.1 Pk and Qk have o(n) zeros on the unit circle.
The proof of Theorem 9.1 follows by combining the recently proved Saffari’s
conjecture stated as Conjecture 8.1* and the theorem below.
Theorem 9.2 If the real trigonometric polynomial R of degree n is of the form

R(t) = |P (eit )|2 ,


Recent Progress in the Study of Polynomials with Constrained Coefficients 47

where P ∈ Pnc , and P has at least k zeros in K (counting multiplicities), then



α k
m({t ∈ [0, 2π ) : |R(t)| ≤ αRK }) ≥
e n

for every α ∈ (0, 1).


Theorem 9.3 There exists an absolute constant c > 0 such that each of the
functions Re(Pk ), Re(Qk ), Im(Pk ), and Im(Qk ) has at least cn zeros on the unit
circle.
Theorem 9.4 There exists an absolute constant c > 0 such that the equation
Rk (t) = ηn has at most cη1/2 solutions in [0, 2π ) for every η ∈ (0, 1] and
sufficiently large k ≥ kη , while the equation Rk (t) = ηn has at most c(2 − η)1/2 n
solutions in [0, 2π ) for every η ∈ [1, 2) and sufficiently large k ≥ kη .
Theorem 9.5 The equation Rk (t) = ηn has at least (1−ε)ηn/2 solutions in [0, 2π )
for every η ∈ (0, 2ω), ε > 0, and sufficiently large k ≥ kη,ε . The equation Rk (t) =
ηn has at least (1 − ε)(2 − η)n/2 solutions in [0, 2π ) for every η ∈ (2 − 2ω, 2),
ε > 0, and sufficiently large k ≥ kη,ε .
Theorem 9.6 There exists an absolute constant c > 0 such that the equation
Rk (t) = (1 + η)n has at least cn0.5394282 distinct solutions in [0, 2π ) whenever
η ∈ R with |η| < 2−8 .
In [2] we combined close to sharp upper bounds for the modulus of the
autocorrelation coefficients of the Rudin-Shapiro polynomials with a deep theorem
of Littlewood (see Theorem 1 in [86]) to prove the above Theorem 9.6.
Theorem 9.7 If


n−1
|Pk (z)|2 = Pk (z)Pk (1/z) = aj zj , z ∈ ∂D ,
j =−n+1

then

c1 n0.7302852··· ≤ max |aj | ≤ c2 n0.7302859···


1≤j ≤n−1

with an absolute constants c1 > 0 and c2 > 0.


Theorem 9.7 has been recently improved by Choi in [31] by showing that

(0.27771487 · · · )(1 + o(1)) |λ|k ≤ max |aj | ≤ (3.78207844 · · · ) |λ|k ,


1≤j ≤n−1

where
48 T. Erdélyi

 √ 1/3  √ 1/3
44 + 3 177 + 44 − 3 177 −1
λ := − = −1.658967081916 · · ·
3

is the real root of the equation x 3 − x 2 − 2x + 4 = 0 and |λ|k = n07302852598··· . This


settles a conjecture expected earlier by Saffari.
Theorem 9.8 (Littlewood) If the real trigonometric polynomial of degree at most
n is of the form


n
f (t) = am cos(mt + αm ) , am , αm ∈ R ,
m=0

satisfies M1 (f ) ≥ cμ, μ := M2 (f ), where c > 0 is a constant, a0 = 0,

n/
h 2
am
sn/ h = ≤ 2−9 c6
μ2
m=1

for some constant h > 0, and v ∈ R satisfies

|v| ≤ V = 2−5 c3 ,

then

N (f, v) > Ah−1 c5 n ,

where N (f, v) denotes the number of real zeros of f − vμ in (−π, π ), and A > 0
is an absolute constant.
In [63] we improved Theorem 9.6 by showing the following two results.
Theorem 9.9 The equation Rk (t) = n has at least n/4 + 1 distinct zeros in [0, 2π ).
Moreover, with the notation tj := 2πj/n, there are at least n/2 + 2 values of
j ∈ {0, 1 . . . , n − 1} for which the interval [tj , tj +1 ] has at least one zero of the
equation Rk (t) = n.
Theorem 9.10 The equation Rk (t) = (1 + η)n has at least (1/2 − |η| − ε)n/2
distinct zeros in [0, 2π ) for every η ∈ (−1/2, 1/2), ε > 0, and sufficiently large
k ≥ kη,ε .
In [62] we proved the theorem below.
Theorem 9.11 There exist absolute constants c1 > 0 and c2 > 0 such that both Pk
and Qk have at least c2 n zeros in the annulus

c1 c1
z∈C:1− < |z| < 1 + .
n n
Recent Progress in the Study of Polynomials with Constrained Coefficients 49

A key to the proof of Theorem 9.11 is the result below.


Theorem 9.12 Let t0 ∈ K. There is an absolute constant c3 > 0 depending only
on c > 0 such that Pk has at least one zero in the disk

c3
z ∈ C : |z − eit0 | < ,
n
whenever

Tk (t0 ) ≥ cn2 , Tk (t) = Pk (eit )Pk (e−it ) .

We note that for every c ∈ (0, 1) there is an absolute constant c4 > 0 depending
only on c such that every Un ∈ Pnc of the form


n
Un (z) = aj zj , |a0 | = |an | = 1 , aj ∈ C , |aj | ≤ 1 ,
j =0

has at least cn zeros in the annulus


 
c4 log n c4 log n
z∈C:1− < |z| < 1 + .
n n

See Theorem 2.1 in [49].


On the other hand, there is an absolute constant c4 > 0 such that for every
n ∈ N there is a polynomial Un ∈ Kn having no zeros in the above annulus. See
Theorem 2.3 in [49].
So in Theorem 9.11 some special properties, in addition to being Littlewood
polynomials, of the Rudin-Shapiro polynomials must be exploited.

10 Open Problems Related to the Rudin-Shapiro


Polynomials

Problem 10.1 Is there an absolute constant c > 0 such that the equation Rk (t) =
ηn has at least cηn distinct solutions in K for every η ∈ (0, 1) and sufficiently large
k ≥ kη ? In other words, can Theorem 9.5 be extended to all η ∈ (0, 1)?
Problem 10.2 Is there an absolute constant c > 0 such that Pk has at least cn zeros
in the open unit disk?
Problem 10.3 Is there an absolute constant c > 0 such that Qk has at least cn zeros
in the open unit disk?
Recall that
50 T. Erdélyi

Qk (−z) = Pk∗ (z) = zn−1 Pk (1/z) , z ∈ ∂D .

Problem 10.4 Is it true that both Pk and Qk have asymptotically half of their zeros
in the open unit disk?
Observe that Pk (−1) = Qk (1) = 0 if k is odd.
Problem 10.5 Is it true that if k is odd then Pk has a zero on the unit circle ∂D
only at −1 and Qk has a zero on the unit circle ∂D only at 1, while if k is even then
neither Pk nor Qk has a zero on the unit circle?

11 On the Size of the Fekete Polynomials on the Unit Circle

For a prime p the p-th Fekete polynomial is defined as


p−1
k

fp (z) := zk ,
p
k=1

where

  ⎪

⎨1, if x 2 ≡ k (mod p) for an x = 0 ,
k
= 0, if p divides k ,
p ⎪

⎩−1, otherwise

is the usual Legendre symbol. Note that gp (z) := fp (z)/z is a Littlewood


polynomial of degree p − 2, and has the same Mahler measure as fp .
In 1980 Montgomery [91] proved the following fundamental result.
Theorem 11.1 There are absolute constants c1 > 0 and c2 > 0 such that
√ √
c1 p log log p ≤ max |fp (z)| ≤ c2 p log p .
z∈∂D

It was observed in [55] that Montgomery’s approach can be used to prove that for
every sufficiently large prime p and for every 8πp−1/8 ≤ s ≤ 2π there is a closed
subset E := Ep,s of the unit circle with linear measure |E| = s such that

1
|fp (z)| |dz| ≥ c1 p1/2 log log(1/s)
|E| E

with an absolute constant c1 > 0.


In [12] the L4 norm of the Fekete polynomials are computed.
Theorem 11.2 We have
Recent Progress in the Study of Polynomials with Constrained Coefficients 51

 1/4
5p2 4
M4 (fp ) = − 3p + − 12(h(−p))2 ,
3 3

where h(−q) is the class number of Q( −q).
In [66] we proved the following result.
Theorem 11.3 For every ε > 0 there is a constant cε such that
 
1 √
M0 (fp ) ≥ −ε p
2

for all primes p ≥ cε .


From Jensen’s inequality,

M0 (fp ) ≤ M2 (fp ) = p − 1.

However, as it was observed in [E-07] and [60], a result of Littlewood [86] implies
that 12 − ε in Theorem 11.2 cannot be replaced by 1 − ε.
To prove Theorem 11.3 in [E-07] we needed to combine Theorems 11.4, 11.5
and one of Theorems 11.6 and 11.7 below. For a prime number p let
 
2π i
ζp := exp
p

the first p-th root of unity. Our first lemma formulates a characteristic property of
the Fekete polynomials. A simple proof is given in [10, pp. 37–38].
Theorem 11.4 (Gauss) We have
& 
j −1
fp (ζp ) = p, j = 1, 2, . . . , p − 1 ,
p

and fp (1) = 0.
Theorem 11.5 We have
⎛ ⎞1/p

p−1
⎝ |Q(ζp )|⎠
j
≤ 2M0 (Q)
j =0

for all polynomials Q of degree at most p with complex coefficients.

√ 11.6 There is an absolute constant c > 0 such that every Q ∈ Kn has at


Theorem
most c n real zeros.
52 T. Erdélyi

Theorem 11.7 There is an absolute constant c > 0 such that every Q ∈ Ln has at
c log2 n
most zeros at 1.
log log n
For a proof of Theorem 11.6 see [18]. For a proof of Theorem 11.7 see [28].
In [54] Theorem 11.3 was extended to subarcs of the unit circle.
Theorem 11.8 There exists an absolute constant c1 > 0 such that

M0 (fp , [α, β]) ≥ c1 p1/2

for all primes p and for all α, β ∈ R such that (log p)3/2 p−1/2 ≤ β − α ≤ 2π .
In [55] we gave an upper bound for the average value of |fp (z)|q over any subarc
I of the unit circle, valid for all sufficiently large primes p and all exponents q > 0.
Theorem 11.9 There exists a constant c2 (q, ε) depending only on q > 0 and ε > 0
such that

Mq (fp , [α, β]) ≤ c2 (q, ε)p1/2 ,

for all primes p and for all α, β ∈ R such that 2p−1/2+ε ≤ β − α ≤ 2π .


We remark that a combination of Theorems 11.8 and 11.9 shows that there is an
absolute constant c1 > 0 and a constant c2 (q, ε) > 0 depending only on q > 0 and
ε > 0 such that

c1 p1/2 ≤ Mq (fp , [α, β]) ≤ c2 (q, ε)p1/2

for all primes p and for all α, β ∈ R such that (log p)3/2 p−1/2 ≤ 2p−1/2+ε ≤
β − α ≤ 2π .
The Lq norm of polynomials related to Fekete polynomials were studied in
several recent papers. See [10–12, 14, 73, 74], and [75], for example. An interesting
extremal property of the Fekete polynomials is proved in [16].
Fekete might have been the first one to study analytic properties of the Fekete
polynomials. He had an idea of proving non-existence of Siegel zeros (that is, real
zeros “especially close to 1”) of Dirichlet L-functions from the positivity of Fekete
polynomials on the interval (0, 1), where the positivity of Fekete polynomials is
often referred to as the Fekete Hypothesis.
There were many mathematicians trying to understand the zeros of Fekete
polynomials including Fekete and Pólya [71], Pólya [98], Chowla [36], Bateman,
Purdy, and Wagstaff [4], Heilbronn [77], Montgomery [91], Baker and Montgomery
[3], and Jung and Shen [78].
Baker and Montgomery [3] proved that fp has a large number of zeros in (0, 1)
for almost all primes p, that is, the number of zeros of fp in (0, 1) tends to ∞ as p
tends to ∞, and it seems likely that there are, in fact, about log log p such zeros.
Recent Progress in the Study of Polynomials with Constrained Coefficients 53

Conrey, Granville, Poonen, and Soundararajan [37] showed that fp has asymp-
totically κ0 p zeros on the unit circle, where

0.500668 < κ0 < 0.500813 .

An interesting recent paper [6] studies power series approximations to Fekete


polynomials. In [60] we improved Theorem 11.2 by showing the following result.
Theorem 11.10 There is an absolute constant c > 1/2 such that

M0 (fp ) ≥ c p

for all sufficiently large primes.


However, there is not even a conjecture in the literature about what the asymp-
totically sharp constant c in Theorem 11.10 could be.

12 Unimodular Zeros of Self-Reciprocal Polynomials


with Coefficients in a Finite Set

Research on the distribution of the zeros of algebraic polynomials has a long and
rich history. A few papers closely related this section include [8, 13, 19, 21–24, 42,
43, 50, 52, 53, 57, 70, 84, 85, 94, 96, 97, 99, 107, 108, 110–114]. The number of real
zeros trigonometric polynomials and the number of unimodular zeros (that is, zeros
lying on the unit circle of the complex plane) of algebraic polynomials with various
constraints on their coefficients are the subject of quite a few of these. We do not try
to survey these in this section.
Let S ⊂ C. Let Pn (S) be the set of all algebraic polynomials of degree at most n
with each of their coefficients in S. An algebraic polynomial P of the form


n
P (z) = aj zj , aj ∈ C , (10)
j =0

is called conjugate-reciprocal if

a j = an−j , j = 0, 1, . . . , n . (11)

Functions T of the form


n
T (t) = α0 + (αj cos(j t) + βj sin(j t)) , αj , βj ∈ R ,
j =1
54 T. Erdélyi

are called real trigonometric polynomials of degree at most n. It is easy to see that
any real trigonometric polynomial T of degree at most n can be written as T (t) =
P (eit )e−int , where P is a conjugate-reciprocal algebraic polynomial of the form


2n
P (z) = aj zj , aj ∈ C .i (12)
j =0

Conversely, if P is conjugate-reciprocal algebraic polynomial of the form (12), then


there are θj ∈ R, j = 1, 2, . . . n, such that


n
T (t) := P (eit )e−int = an + 2|aj +n | cos(j t + θj )
j =1

is a real trigonometric polynomial of degree at most n. A polynomial P of the


form (10) is called self-reciprocal if

aj = an−j , j = 0, 1, . . . , n . (13)

If a conjugate-reciprocal algebraic polynomial P has only real coefficients, then it


is obviously self-reciprocal. If the algebraic polynomial P of the form (12) is self-
reciprocal, then


n
T (t) := P (eit )e−int = an + 2aj +n cos(j t) .
j =1

In this section, whenever we write “P ∈ Pn (S) is conjugate-reciprocal” we mean


that P is of the form (10) with each aj ∈ S satisfying (11). Similarly, whenever we
write “P ∈ Pn (S) is self-reciprocal” we mean that P is of the form (10) with each
aj ∈ S satisfying (13). This is going to be our understanding even if the degree of
P ∈ Pn (S) is less than n. It is easy to see that P ∈ Pn (S) is self-reciprocal and n is
odd, then P (−1) = 0. We call any subinterval [a, a + 2π ) of the real number line R
a period. Associated with an algebraic polynomial P of the form (10) we introduce
the number

NC(P ) := |{j ∈ {0, 1, . . . , n} : aj = 0}| .

Here, and in what follows |A| denotes the number of elements of a finite set A. Let
NZ(P ) denote the number of real zeros (by counting multiplicities) of an algebraic
polynomial P on the unit circle. Associated with an even trigonometric polynomial
(cosine polynomial) of the form
Recent Progress in the Study of Polynomials with Constrained Coefficients 55


n
T (t) = aj cos(j t)
j =0

we introduce the number

NC(T ) := |{j ∈ {0, 1, . . . , n} : aj = 0}| .

Let NZ(T ) denote the number of real zeros (by counting multiplicities) of a
trigonometric polynomial T in a period. Let NZ∗ (T ) denote the number of sign
changes of a trigonometric polynomial T in a period. The quotation below is from
[21].
“Let 0 ≤ n 1 < n2 < · · · < nN be integers. A cosine polynomial of the
N
form T (θ ) = j =1 cos(nj θ ) must have at least one real zero in a period. This
is obvious if n1 = 0, since then the integral of the sum on a period is 0. The
above statement is less obvious if n1 = 0, but for sufficiently large N it follows
from Littlewood’s Conjecture simply. Here we mean the Littlewood’s Conjecture
proved by S. Konyagin [81] and independently by McGehee, Pigno, and Smith [89]
in 1981. See also [38, pages 285–288] for a book proof. It is not difficult to prove the
statement in general even in the case n1 = 0 without using Littlewood’s Conjecture.
One possible way is to use the identity


nN  
(2j − 1)π
T = 0.
nN
j =1

See [82], for example. Another way is to use Theorem 2 of [90]. So there is certainly
no shortage of possible approaches to prove the starting observation of this section
even in the case n1 = 0.
It seems likely that the number of zeros of the above sums in a period must tend
to ∞ with N . In a private communication B. Conrey asked how fast the number of
real zeros of the above sums in a period tends to ∞ as a function N . In [37] the
authors observed that for an odd prime p the Fekete polynomial


p−1 
k k
fp (z) = z
p
k=0

(the coefficients are Legendre symbols) has ∼κ0 p zeros on the unit circle, where
0.500813 > κ0 > 0.500668. Conrey’s question in general does not appear to be
easy. Littlewood in his 1968 monograph ‘Some Problems in Real and Complex
Analysis’ [88, problem 22] poses the following research problem, which appears to
still be open: ‘If the nm are
integral and all different, what is the lower bound on
the number of real zeros of N m=1 cos(nm θ )? Possibly N − 1, or not much less.’
Here real zeros are counted in a period. In fact no progress appears to have been
made on this in the last half century. In a recent paper [21] we showed that this
56 T. Erdélyi


is false. There exist cosine polynomials N m=1 cos(nm θ ) with the nm integral and
all different so that the number of its real zeros in a period is O(N 9/10 (log N )1/5 )
(here the frequencies nm = nm (N ) may vary with N ). However, there are reasons
to believe that a cosine polynomial N m=1 cos(nm θ ) always has many zeros in a
period.”
Let, as before,
⎧ ⎫
⎨  n ⎬
Ln := P : P (z) = aj zj , aj ∈ {−1, 1} .
⎩ ⎭
j =0

Elements of Ln are often called Littlewood polynomials of degree n. Let


⎧ ⎫
⎨ 
n ⎬
Hn := P : P (z) = aj zj , aj ∈ C, |a0 | = |an | = 1, |aj | ≤ 1 .
⎩ ⎭
j =0

Observe that Ln ⊂ Hn . In [24] we proved that any polynomial P ∈ Kn has at


least 8n1/2 log n zeros in any open disk centered at a point on the unit circle with
radius 33n−1/2 log n. Thus polynomials in Hn have quite a few zeros near the unit
circle. One may naturally ask how many unimodular roots a polynomial in Hn can
have. Mercer [90] proved that if a Littlewood polynomial P ∈ Ln of the form (10)
is skew-reciprocal, that is, aj = (−1)j an−j for each j = 0, 1, . . . , n, then it has
no zeros on the unit circle. However, by using different elementary methods it was
observed in both [50] and [90] that if a Littlewood polynomial P of the form (10)
is self-reciprocal, then it has at least one zero on the unit circle. Mukunda [93]
improved this result by showing that every self-reciprocal Littlewood polynomial
of odd degree has at least 3 zeros on the unit circle. Drungilas [41] proved that
every self-reciprocal Littlewood polynomial of odd degree n ≥ 7 has at least 5
zeros on the unit circle and every self-reciprocal Littlewood polynomial of even
degree n ≥ 14 has at least 4 zeros on the unit circle. In [21] we proved that the
average number of zeros of self-reciprocal Littlewood polynomials of degree n is at
least n/4. However, it is much harder to give decent lower bounds for the quantities
NZn := minP NZ(P ) , where NZ(P ) denotes the number of zeros of a polynomial
P lying on the unit circle and the minimum is taken for all self-reciprocal Littlewood
polynomials P ∈ Ln . It has been conjectured for a long time that limn→∞ NZn =
∞. In [58] we showed that limn→∞ NZ(Pn ) = ∞ whenever Pn ∈ Ln is self-
reciprocal and limn→∞ |Pn (1)| = ∞. This follows as a consequence of a more
general result, see Corollary 2.3 in [58], stated as Corollary 12.5 here, in which the
coefficients of the self-reciprocal polynomials Pn of degree at most n belong to a
fixed finite set of real numbers. In [20] we proved the following result.
Theorem 12.1 If the set {aj : j ∈ N} ⊂ R is finite, the set {j ∈ N : aj = 0} is
infinite, the sequence (aj ) is not eventually periodic, and
Recent Progress in the Study of Polynomials with Constrained Coefficients 57


n
Tn (t) = aj cos(j t) ,
j =0

then limn→∞ NZ(Tn ) = ∞ .


In [20] Theorem 12.1 is stated without the assumption that the sequence (aj ) is
not eventually periodic. However, as the following example shows, Lemma 3.4 in
[20], dealing with the case of eventually periodic sequences (aj ), is incorrect. Let


n−1
Tn (t) := cos t + cos((4n + 1)t) + (cos((4k + 1)t) − cos((4k + 3)t))
k=0
1 + cos((4n + 2)t)
= + cos t .
2 cos t

It is easy to see that Tn (t) = 0 on [−π, π ] \ {−π/2, π/2} and the zeros of Tn at
−π/2 and π/2 are simple. Hence Tn has only two (simple) zeros in a period. So
the conclusion of Theorem 12.1 above is false for the sequence (aj ) with a0 := 0,
a1 := 2, a3 := −1, a2k := 0, a4k+1 := 1, a4k+3 := −1 for every k = 1, 2, . . ..
Nevertheless, Theorem 12.1 can be saved even in the case of eventually periodic
sequences (aj ) if we assume that aj = 0 for all sufficiently large j . See Lemma 3.11
in [22] where Theorem 1 in [20] is corrected as
Theorem 12.2 If the set {aj : j ∈ N} ⊂ R is finite, aj = 0 for all sufficiently large
j , and


n
Tn (t) = aj cos(j t) ,
j =0

then limn→∞ NZ(Tn ) = ∞ .


It was expected that the conclusion of the above theorem remains true even if the
coefficients of Tn do not come from the same sequence, that is,


n
Tn (t) = aj,n cos(j t) ,
j =0

where the set

S := {aj,n : j ∈ {0, 1, . . . , n}, n ∈ N} ⊂ R

is finite and

lim |{j ∈ {0, 1, . . . , n}, aj,n = 0}| = ∞ .


n→∞
58 T. Erdélyi

Associated with an algebraic polynomial


n
P (z) = aj zj , aj ∈ C ,
j =0

let

NCk (P ) := |{u : 0 ≤ u ≤ n − k + 1, au + au+1 + · · · + au+k−1 = 0}| .

In [58] we proved the following results.


Theorem 12.3 If S ⊂ R is a finite set, P2n ∈ P2n (S) are self-reciprocal
polynomials,

Tn (t) := P2n (eit )e−int ,

and

lim NCk (P2n ) = ∞


n→∞

for every k = 1, 2, . . ., then

lim NZ(P2n ) = lim NZ(Tn ) = ∞ .


n→∞ n→∞

Some of the most important consequences of the above theorem obtained in [58]
are stated below.
Corollary 12.4 If S ⊂ R is a finite set, Pn ∈ Pn (S) are self-reciprocal polynomials,
and

lim |Pn (1)| = ∞ ,


n→∞

then

lim NZ(Pn ) = ∞ .
n→∞

Corollary 12.5 Suppose the finite set S ⊂ R has the property that

s1 + s2 + · · · + sk = 0 , s1 , s2 , . . . , sk ∈ S , implies s1 = s2 = · · · = sk = 0 ,

that is, any sum of nonzero elements of S is different from 0. If Pn ∈ Pn (S) are
self-reciprocal polynomials and

lim NC(Pn ) = ∞ ,
n→∞
Recent Progress in the Study of Polynomials with Constrained Coefficients 59

then

lim NZ(Pn ) = ∞ .
n→∞

J. Sahasrabudhe [106] examined the case when S ⊂ Z is finite. Exploiting the


assumption that the coefficients are integer he proved that for any finite set S ⊂ Z a
self-reciprocal polynomial P ∈ P2n (S) has at least

c (log log log |P (1)|)1/2−ε − 1

zeros on the unit circle of C with a constant c > 0 depending only on M = M(S) :=
max{|z| : z ∈ S} and ε > 0.
Let ϕ(n) denote the Euler’s totient function defined as the number of integers 1 ≤
k ≤ n that are relative prime to n. In an √earlier version of his paper Sahasrabudhe
[106] used the trivial estimate ϕ(n) = n for n ≥ 3 and he proved his result with
the exponent 1/4 − ε rather than 1/2 − ε. Using the nontrivial estimate ϕ(n) ≥
n/8(log log n) [103] for all n > 3 allowed him to prove his result with 1/2 − ε.
In the papers [20, 58], and [106] the already mentioned Littlewood Conjecture,
proved by Konyagin [81] and independently by McGehee, Pigno, and B. Smith [89],
plays a key role, and we rely on it heavily in the proof of the main results of this
paper as well. This states the following.
Theorem 12.6 There is an absolute constant c > 0 such that
 2π m 
 
 aj eiλj t  dt ≥ cδ log m
0 j =1

whenever λ1 , λ2 , . . . , λn are distinct integers and a1 , a2 , . . . , am are complex


numbers of modulus at least δ > 0. Here c = 1/30 is a suitable choice.
This is an obvious consequence of the following result a book proof of which has
been worked out by DeVore and Lorentz in [38, pages 285–288].
Theorem 12.7 If λ1 < λ2 < · · · < λm are integers and a1 , a2 , . . . , am are complex
numbers, then
   1  |aj |
2π m m
 
 aj eiλj t  dt ≥ .
0 30 j
j =1 j =1

In [65] we proved the following results.


Theorem 12.8 If S ⊂ Z is a finite set, M = M(S) := max{|z| : z ∈ S}, P ∈
P2n (S) is a self-reciprocal polynomial,

T (t) := P (eit )e−int ,


60 T. Erdélyi

then
 
∗ c log log log |P (1)|
NZ (Tn ) ≥ −1
1 + log M log log log log |P (1)|

with an absolute constant c > 0, whenever |P (1)| ≥ 16.


Corollary 12.9 If S ⊂ Z is a finite set, M = M(S) := max{|z| : z ∈ S}, P ∈
Pn (S) is a self-reciprocal polynomial, then
 
c log log log |P (1)|
NZ(P ) ≥ −1
1 + log M log log log log |P (1)|

with an absolute constant c > 0, whenever |P (1)| ≥ 16.


This improves the exponent 1/2 − ε to 1 − ε in a recent breakthrough result
[106] by Sahasrabudhe. We note that in both Sahasrabudhe’s paper and this paper
the assumption that the finite set S contains only integers is deeply exploited. Our
next result is an obvious consequence of Corollary 12.9.
Corollary 12.10 If the set S ⊂ Z is finite, M = M(S) := max{|z| : z ∈ S},


n
T (t) = aj cos(j t) , aj ∈ S ,
j =0

then
 
∗ c log log log |T (0)|
NZ (T ) ≥ −1
1 + log M log log log log |T (0)|

with an absolute constant c > 0, whenever |T (0)| ≥ 16.

13 Bourgain’s L1 Problem and Related Results

For n ≥ 1 let
 
n 
An := P : P (z) = z : 0 ≤ k1 < k2 < · · · < kn , kj ∈ Z ,
kj

j =1

that is, An is the collection of all sums of n distinct monomials. For p ≥ 0 we define

Mp (Q)
Sn,p := sup √ and Sp := lim inf Sn,p ≤ p := lim sup Sn,p .
Q∈An n n→∞ n→∞
Recent Progress in the Study of Polynomials with Constrained Coefficients 61

We also define
Mp (Q)
In,p := inf √ and Ip := lim sup In,p ≥ p := lim inf In,p .
Q∈An n n→∞ n∈→∞

Observe that Parseval’s formula gives 2 = 2 = 1. The problem of calculating


1 appears in a paper of Bourgain [27]. Deciding whether 1 < 1 or 1 = 1
would be a major step toward confirming
√ or disproving other important conjectures.
Karatsuba [80] observed that 1 ≥ 1/ 2 ≥ 0.707. Indeed, taking, for instance,


n−1
k
Pn (z) = z2 , n = 1, 2, . . . ,
k=0

it is easy to see that

M4 (Pn )4 = 2n(n − 1) + n , (14)

and as Hölder’s inequality implies

n = M2 (Pn )2 ≤ M1 (Pn )2/3 M4 (Pn )4/3 ,

we conclude

n3/2 n n
M1 (Pn ) ≥ = ≥ √ . (15)
(2n(n − 1) + n)1/2 (2n − 1)1/2 2

Similarly, if Sn := {a1 < a2 < · · · < an } is a Sidon set (that is, Sn is a subset of
integers such that no integer has two essentially distinct representations as the sum
of two elements of Sn ), then the polynomials

Pn (z) = za , n = 1, 2, . . . ,
a∈Sn

satisfy (14) and (15). In fact, it was observed in [15] that

min M4 (P )4 = 2n(n − 1) + n ,
P ∈An

and such minimal polynomials in An are precisely constructed by Sidon sets as


above.
Improving Karatsuba’s
√ result, by using a probabilistic method Aistleitner [1]
proved that 1 ≥ π /2 ≥ 0.886. We note that P. Borwein and Lockhart [25]
investigated the asymptotic behavior of the mean value of normalized Lp norms
of Littlewood polynomials for arbitrary p > 0. Using the Lindeberg Central Limit
Theorem and the Dominated Convergence Theorem, they proved that
62 T. Erdélyi

1  (Mp (f ))p
lim = (1 + p/2) ,
n→∞ 2n+1 np/2
f ∈Ln

where Ln is, as before, the set of Littlewood polynomials of degree n. It√follows


simply from the case p = 1 of the result in [25] quoted above that 1 ≥ π/8 ≥
0.626. Moreover, this can be achieved by taking the sum of approximately half of
the monomials of {x 0 , x 1 , . . . , x 2n } and letting n tend to ∞.
Observe that Parseval’s formula gives 2 = 2 = 1. In [33] we proved the
following results.
Theorem 13.1 Let (kj ) be a strictly increasing sequence of nonnegative integers
satisfying
 
cj
kj +1 > kj 1 + 1/2 , j = 1, 2, . . . ,
j
where limj →∞ cj = ∞. Let


n
Pn (z) = zkj , n = 1, 2, . . . .
j =1

We have
Mp (Pn )
lim √ = (1 + p/2)1/p
n→∞ n
for every p ∈ (0, 2).
Theorem 13.2 Let (kj ) be a strictly increasing sequence of nonnegative integers
satisfying

kj +1 > qkj , j = 1, 2, . . . ,

where q > 1. Let



n
Pn (z) = zkj , n = 1, 2, . . . .
j =1

We have
Mp (Pn )
lim √ = (1 + p/2)1/p
n→∞ n
for every p ∈ [1, ∞).
Corollary 13.3 We have p ≥ Sp ≥ (1 + p/2)1/p for all p ∈ (0, 2).
The special case p = 1 recaptures a recent result of Aistleitner [1], the best
known lower bound for 1 .
Recent Progress in the Study of Polynomials with Constrained Coefficients 63


Corollary 13.4 We have 1 ≥ S1 ≥ π /2.
Corollary 13.5 We have p ≤ Ip ≤ (1 + p/2)1/p for all p ∈ (2, ∞).

We remark here that the same results also hold for the polynomials nj=1 aj zkj
with coefficients aj if a general form of the Salem-Zygmund theorem is used (e.g.
see (2) in [69]).
Our final result in [33] shows that the upper bound (1 + p/2)1/p in Corol-
lary 13.5 is optimal at least for even integers.
Corollary 13.6 For any even integer p = 2m ≥ 2, we have

Mp (P )
lim min √ = (1 + p/2)1/p .
n→∞ P ∈An n

Observe that a standard way to prove a Nikolskii-type inequality for trigonomet-


ric polynomials [2, p. 394] applies to the classes An . Indeed,
 2π  p
1  
Mp (P ) =p
P (eit ) dt
2π 0
  2π  2    p−2
1  it   it 
≤ P (e ) dt max P (e )
2π 0 t∈[0,2π ]

= nnp−2 = np−1

for every P ∈ An and p ≥ 2, and the Dirichlet kernel Dn (z) := 1+z+· · ·+zn shows
the sharpness of this upper bound up to a multiplicative factor constant c > 0. So if
we study the original Bourgain problem in the case of p > 2, we should normalize
by dividing by n1−1/p rather than n1/2 .
In [34] we examined

M0 (Q, I )
Sn,0 (I ) := sup √ and S0 (I ) := lim inf Sn,0 (I )
Q∈An n n→∞

for intervals I = [α, β] with 0 < |I | := β − α ≤ 2π and proved the following


results.
Theorem 13.7 There are polynomials Qn ∈ An ∩ PN with N = 2n + o(n) such
that
 
1 √
M0 (Qn ) ≥ √ + o(1) n, n = 1, 2, . . . ,
2 2

1
and hence S0 ≥ √ .
2 2
64 T. Erdélyi

Theorem 13.8 There are polynomials Qn ∈ An ∩ PN with N = 2n + o(n), an


absolute constant c1 > 0, and a constant c2 (ε) > 0 depending only on ε > 0 such
that

M0 (Qn , I ) ≥ c1 n , n = 1, 2, . . . ,

for every interval I := [α, β] ⊂ R such that

4π (log n)3/2
≤ ≤ β − α ≤ 2π , (16)
n n1/2
while

M1 (Qn , I ) ≤ c2 (ε) n , n = 1, 2, . . . ,

for every interval I := [α, β] ⊂ R such that

(n/2)−1/2+ε ≤ β − α ≤ 2π . (17)

Note that Theorem 13.8 implies that there is an absolute constant c1 > 0 such
that S0 (I ) ≥ c1 for all intervals I := [α, β] ⊂ R satisfying (16).
Theorem 13.9 There are polynomials Qn ∈ Ln such that
 
1 √
M0 (Qn ) ≥ + o(1) n, n = 1, 2, . . . .
2

Theorem 13.10 There are polynomials Qn ∈ Ln , an absolute constant c1 > 0, and


a constant c2 (ε) > 0 depending only on ε > 0 such that

M0 (Qn , I ) ≥ c1 n , n = 1, 2, . . . ,

for every interval I := [α, β] ⊂ R satisfying (16), while



M1 (Qn , I ) ≤ c2 (ε) n , n = 1, 2 . . . ,

for every interval I := [α, β] ⊂ R satisfying (17).

Reference

1. C. Aistleitner, On a problem of Bourgain concerning the L1 -norm of exponential sums. Math.


Z. 275, 681–688 (2013)
2. J.-P. Allouche, K.-K.S. Choi, A. Denise, T. Erdélyi, B. Saffari, Bounds on autocorrelation
coefficients of Rudin-Shapiro polynomials. Anal. Math. 45, 705 (2019). https://doi.org/10.
1007/s10476-019-0003-4
Recent Progress in the Study of Polynomials with Constrained Coefficients 65

3. R.C. Baker, H.L. Montgomery, Oscillations of quadratic L-functions, in Analytic Number


Theory, ed. by B.C. Berndt et al. (Birkhäuser, Boston, 1990), pp. 23–40
4. P.T. Bateman, G.B. Purdy, S.S Wagstaff, Jr., Some numerical results on Fekete polynomials.
Math. Comput. 29, 7–23 (1975)
5. J. Beck, “Flat” polynomials on the unit circle – note on a problem of Littlewood. Bull. Lond.
Math. Soc. 23, 269–277 (1991)
6. J. Bell, I. Shparlinski, Power series approximations to Fekete polynomials. J. Approx. Theory
222, 132–142 (2017)
7. E. Beller, D.J. Newman, An extremal problem for the geometric mean of polynomials. Proc.
Am. Math. Soc. 39, 313–317 (1973)
8. A. Bloch, G. Pólya, On the roots of certain algebraic equations. Proc. Lond. Math. Soc. 33,
102–114 (1932)
9. E. Bombieri, J. Bourgain, On Kahane’s ultraflat polynomials. J. Eur. Math. Soc. 11(3), 627–
703 (2009)
10. P. Borwein, Computational Excursions in Analysis and Number Theory (Springer, New York,
2002)
11. P. Borwein, K.-K.S. Choi, Merit factors of polynomials formed by Jacobi symbols. Can. J.
Math. 53(1), 33–50 (2001)
12. P. Borwein, K.-K.S. Choi, Explicit merit factor formulae for Fekete and Turyn polynomials.
Trans. Am. Math. Soc. 354(1), 219–234 (2002)
13. P. Borwein, K.-K.S. Choi, R. Ferguson, J. Jankauskas, On Littlewood polynomials with
prescribed number of zeros inside the unit disk. Can. J. Math. 67, 507–526 (2015)
14. P. Borwein, K.-K.S. Choi, J. Jedwab, Binary sequences with merit factor greater than 6.34.
IEEE Trans. Inform. Theory 50(12), 3234–3249 (2004)
15. P. Borwein, K.-K.S. Choi, I. Mercer, Expected norms of zero-one polynomials. Can. Math.
Bull. 51, 297–507 (2008)
16. P. Borwein, K.-K.S. Choi, S. Yazdani, An extremal property of Fekete polynomials. Proc.
Am. Math. Soc. 129(1), 19–27 (2001)
17. P. Borwein, T. Erdélyi, Polynomials and Polynomial Inequalities (Springer, New York, 1995)
18. P. Borwein, T. Erdélyi, On the zeros of polynomials with restricted coefficients. Ill. J. Math.
41(4), 667–675 (1997)
19. P. Borwein, T. Erdélyi, Trigonometric polynomials with many real zeros and a Littlewood-
type problem. Proc. Am. Math. Soc. 129(3), 725–730 (2001)
20. P. Borwein, T. Erdélyi, Lower bounds for the number of zeros of cosine polynomials in the
period: a problem of Littlewood. Acta Arith. 128(4), 377–384 (2007)
21. P. Borwein, T. Erdélyi, R. Ferguson, R. Lockhart, On the zeros of cosine polynomials: solution
to a problem of Littlewood. Ann. Math. (2) 167(3), 1109–1117 (2008)
22. P. Borwein, T. Erdélyi, G. Kós, Littlewood-type problems on [0, 1]. Proc. Lond. Math. Soc.
79, 22–46 (1999)
23. P. Borwein, T. Erdélyi, G. Kós, The multiplicity of the zero at 1 of polynomials with
constrained coefficients. Acta Arith. 159(4), 387–395 (2013)
24. P. Borwein, T. Erdélyi, F. Littmann, Zeros of polynomials with finitely many different
coefficients. Trans. Am. Math. Soc. 360, 5145–5154 (2008)
25. P. Borwein, R. Lockhart, The expected Lp norm of random polynomials. Proc. Am. Math.
Soc. 129, 1463–1472 (2001)
26. P. Borwein, M.J. Mossinghoff, Rudin-Shapiro like polynomials in L4 . Math. Comput. 69,
1157–1166 (2000)
27. J. Bourgain, On the sprectal type of Ornstein’s class one transformations. Israel J. Math. 84,
53–63 (1993)
28. D. Boyd, On a problem of Byrnes concerning polynomials with restricted coefficients. Math.
Comput. 66, 1697–1703 (1997)
29. J. Brillhart, On the Rudin-Shapiro polynomials. Duke Math. J. 40(2), 335–353 (1973)
30. J. Brillhart, J.S. Lemont, P. Morton, Cyclotomic properties of the Rudin-Shapiro polynomials.
J. Reine Angew. Math. (Crelle’s J.) 288, 37–65 (1976)
66 T. Erdélyi

31. K.-K.S. Choi, Bounds on autocorrelation coefficients of Rudin-Shapiro polynomials II. J.


Approx. Theory (2019)
32. K.-K.S. Choi, T. Erdélyi, On the average Mahler measures on Littlewood polynomials. Proc.
Am. Math. Soc. Ser. B 1, 105–120 (2015)
33. K.-K.S. Choi, T. Erdélyi, On a problem of Bourgain concerning the Lp norms of exponential
sums. Math. Zeit. 279, 577–584 (2015)
34. K.-K.S. Choi, T. Erdélyi, Sums of monomials with large Mahler measure. J. Approx. Theory
197, 49–61 (2015)
35. K.-K.S. Choi, M.J. Mossinghoff, Average Mahler’s measure and Lp norms of unimodular
polynomials. Pacific J. Math. 252(1), 31–50 (2011)
36. S. Chowla, Note on Dirichlet’s L-functions. Acta Arith. 1, 113–114 (1935)
37. B. Conrey, A. Granville, B. Poonen, K. Soundararajan, Zeros of Fekete polynomials. Ann.
Inst. Fourier (Grenoble) 50, 865–884 (2000)
38. R.A. DeVore, G.G. Lorentz, Constructive Approximation (Springer, Berlin, 1993)
39. Ch. Doche, Even moments of generalized Rudin-Shapiro polynomials. Math. Comput.
74(252), 1923–1935 (2005)
40. Ch. Doche, L. Habsieger, Moments of the Rudin-Shapiro polynomials. J. Fourier Anal. Appl.
10(5), 497–505 (2004)
41. P. Drungilas, Unimodular roots of reciprocal Littlewood polynomials. J. Korean Math. Soc.
45(3), 835–840 (2008)
42. A. Dubickas, On the order of vanishing at 1 of a polynomial. Lith. Math. J. 39, 365–370
(1999)
43. A. Dubickas, Polynomials with multiple roots at 1. Int. J. Number Theory 10(2), 391–400
(2014)
44. S.B. Ekhad, D. Zeilberger, Integrals involving Rudin-Shapiro polynomials and sketch
of a proof of Saffari’s conjecture, in Analytic Number Theory, Modular Forms and q-
Hypergeometric Series. Springer Proceedings in Mathematics and Statistics, vol. 221
(Springer, Cham, 2017), pp. 253–265
45. T. Erdélyi, The phase problem of ultraflat unimodular polynomials: the resolution of the
conjecture of Saffari. Math. Ann. 300, 39–60 (2000)
46. T. Erdélyi, The resolution of Saffari’s Phase Problem. C. R. Acad. Sci. Paris Sér. I Math. 331,
803–808 (2000)
47. T. Erdélyi, How far is a sequence of ultraflat unimodular polynomials from being conjugate
reciprocal. Michigan Math. J. 49, 259–264 (2001)
48. T. Erdélyi, Proof of Saffari’s near-orthogonality conjecture for ultraflat sequences of unimod-
ular polynomials. C. R. Acad. Sci. Paris Sér. I Math. 333, 623–628 (2001)
49. T. Erdélyi, On the zeros of polynomials with Littlewood-type coefficient constraints. Michi-
gan Math. J. 49, 97–111 (2001)
50. T. Erdélyi, Polynomials with Littlewood-type coefficient constraints, in Approximation
Theory X: Abstract and Classical Analysis, ed. by C.K. Chui, L.L. Schumaker, J. Stöckler
(Vanderbilt University Press, Nashville, 2002), pp. 153–196
51. T. Erdélyi, On the real part of ultraflat sequences of unimodular polynomials. Math. Ann. 326,
489–498 (2003)
52. T. Erdélyi, An improvement of the Erdős-Turán theorem on the distribution of zeros of
polynomials. C. R. Acad. Sci. Paris, Ser. I 346(5), 267–270 (2008)
53. T. Erdélyi, Extensions of the Bloch-Pólya theorem on the number of real zeros of polynomials.
J. Théor. Nombres Bordeaux 20(2), 281–287 (2008)
54. T. Erdélyi, Sieve-type lower bounds for the Mahler measure of polynomials on subarcs.
Comput. Methods Funct. Theory 11, 213–228 (2011)
55. T. Erdélyi, Upper bounds for the Lq norm of Fekete polynomials on subarcs. Acta Arith.
153(1), 81–91 (2012)
56. T. Erdélyi, On the flatness of conjugate reciprocal unimodular polynomials. J. Math. Anal.
Appl. 432(2), 699–714 (2015)
Recent Progress in the Study of Polynomials with Constrained Coefficients 67

57. T. Erdélyi, Coppersmith-Rivlin type inequalities and the order of vanishing of polynomials at
1. Acta Arith. 172(3), 271–284 (2016)
58. T. Erdélyi, On the number of unimodular zeros of self-reciprocal polynomials with coeffi-
cients from a finite set. Acta Arith. 176(2), 177–200 (2016)
59. T. Erdélyi, The Mahler measure of the Rudin-Shapiro polynomials. Constr. Approx. 43(3),
357–369 (2016)
60. T. Erdélyi, Improved lower bound for the Mahler measure of the Fekete polynomials. Constr.
Approx. 48(2), 383–399 (2018)
61. T. Erdélyi, The asymptotic value of the Mahler measure of the Rudin-Shapiro polynomials. J.
Anal. Math. (accepted)
62. T. Erdélyi, On the oscillation of the modulus of Rudin-Shapiro polynomials on the unit circle.
Mathematika 66, 144–160 (2020)
63. T. Erdélyi, Improved results on the oscillation of the modulus of Rudin-Shapiro polynomials
on the unit circle. Proc. Am. Math. Soc. (2019, accepted)
64. T. Erdélyi, The asymptotic distance between an ultraflat unimodular polynomial and its
conjugate reciprocal. https://arxiv.org/abs/1810.04287
65. T. Erdélyi, Improved lower bound for the number of unimodular zeros of self-reciprocal
polynomials with coefficients from a finite set. Acta Arith. (2019). https://arxiv.org/pdf/1702.
05823.pdf
66. T. Erdélyi, D. Lubinsky, Large sieve inequalities via subharmonic methods and the Mahler
measure of Fekete polynomials. Can. J. Math. 59, 730–741 (2007)
67. T. Erdélyi, P. Nevai, On the derivatives of unimodular polynomials (Russian). Mat. Sbornik
207(4), 123–142 (2016); translation in Sbornik Math. 207(3–4), 590–609 (2016)
68. P. Erdős, Some unsolved problems. Michigan Math. J. 4, 291–300 (1957)
69. P. Erdős, On trigonometric sums with gaps. Publ. Math. Inst. Hung. Acad. Sci. Ser. A 7, 37–42
(1962)
70. P. Erdős, P. Turán, On the distribution of roots of polynomials. Ann. Math. 51, 105–119 (1950)
71. M. Fekete, G. Pólya, Über ein Problem von Laguerre. Rend. Circ. Mat. Palermo 34, 89–120
(1912)
72. M.J. Golay, Static multislit spectrometry and its application to the panoramic display of
infrared spectra. J. Opt. Soc. Am. 41, 468–472 (1951)
73. C. Günther, K.-U. Schmidt, Lq norms of Fekete and related polynomials. Can. J. Math. 69(4),
807–825 (2017)
74. J. Jedwab, D.J. Katz, K.-U. Schmidt, Advances in the merit factor problem for binary
sequences. J. Combin. Theory Ser. A 120(4), 882–906 (2013)
75. J. Jedwab, D.J. Katz, K.-U. Schmidt, Littlewood polynomials with small L4 norm. Adv. Math.
241, 127–136 (2013)
76. G.H. Hardy, J.E. Littlewood, G. Pólya, Inequalities (Cambridge University Press, London,
1952)
77. H. Heilbronn, On real characters. Acta Arith. 2, 212–213 (1937)
78. J. Jung, S.W. Shin, On the sparsity of positive-definite automorphic forms within a family. J.
Anal. Math. 129(1), 105–138 (2016)
79. J.P. Kahane, Sur les polynomes a coefficient unimodulaires. Bull. Lond. Math. Soc. 12, 321–
342 (1980)
80. A.A. Karatsuba, An estimate for the L1 -norm of an exponential sum. Mat. Zametki 64, 465–
468 (1998)
81. S.V. Konyagin, On a problem of Littlewood. Math. USSR Izvestia 18, 205–225 (1981)
82. S.V. Konyagin, V.F. Lev, Character sums in complex half planes. J. Theor. Nombres Bordeaux
16(3), 587–606 (2004)
83. T. Körner, On a polynomial of J.S. Byrnes. Bull. Lond. Math. Soc. 12, 219–224 (1980)
84. J.E. Littlewood, On the mean values of certain trigonometrical polynomials. J. Lond. Math.
Soc. 36, 307–334 (1961)
85. J.E. Littlewood, On the real roots of real trigonometrical polynomials (II). J. Lond. Math. Soc.
39, 511–552 (1964)
68 T. Erdélyi

86. J.E. Littlewood, The real zeros and value distributions of real trigonometrical polynomials. J.
Lond. Math. Soc. 41, 336–342 (1966) m
87. J.E. Littlewood, On polynomials ±z , exp(αm i)zm , z = eiθ . J. Lond. Math. Soc. 41,
367–376 (1966)
88. J.E. Littlewood, Some Problems in Real and Complex Analysis (Heath Mathematical Mono-
graphs, Lexington, 1968)
89. O.C. McGehee, L. Pigno, B. Smith, Hardy’s inequality and the L1 norm of exponential sums.
Ann. Math. 113, 613–618 (1981)
90. I.D. Mercer, Unimodular roots of special Littlewood polynomials. Can. Math. Bull. 49(3),
438–447 (2006)
91. H.L. Montgomery, An exponential polynomial formed with the Legendre symbol. Acta Arith.
37, 375–380 (1980)
92. H.L. Montgomery, Littlewood polynomials, in Analytic Number Theory, Modular Forms
and q-Hypergeometric Series, ed. by G. Andrews, F. Garvan. Springer Proceedings in
Mathematics and Statistics, vol. 221 (Springer, Cham, 2017), pp. 533–553
93. K. Mukunda, Littlewood Pisot numbers. J. Number Theory 117(1), 106–121 (2006)
94. H. Nguyen, O. Nguyen, V. Vu, On the number of real roots of random polynomials. Commun.
Contemp. Math. 18, 1550052 (2016)
95. A. Odlyzko, Search for ultraflat polynomials with plus and minus one coefficients, in
Connections in Discrete Mathematics, ed. by S. Butler, J. Cooper, G. Hurlbert (Cambridge
University Press, Cambridge, 2018), pp. 39–55
96. A.M. Odlyzko, B. Poonen, Zeros of polynomials with 0, 1 coefficients. L’Enseign. Math. 39,
317–348 (1993)
97. C. Pinner, Double roots of [−1, 1] power series and related matters. Math. Comput. 68, 1149–
1178 (1999)
98. G. Pólya, Verschiedene Bemerkungen zur Zahlentheorie. Jahresber. Dtsch. Math. Ver. 28,
31–40 (1919)
99. I.E. Pritsker, A.A. Sola, Expected discrepancy for zeros of random algebraic polynomials.
Proc. Am. Math. Soc. 142, 4251–4263 (2014)
100. H. Queffelec, B. Saffari, Unimodular polynomials and Bernstein’s inequalities. C. R. Acad.
Sci. Paris Sér. I Math. 321(3), 313–318 (1995)
101. H. Queffelec, B. Saffari, On Bernstein’s inequality and Kahane’s ultraflat polynomials. J.
Fourier Anal. Appl. 2(6), 519–582 (1996)
102. B. Rodgers, On the distribution of Rudin-Shapiro polynomials and lacunary walks on SU (2),
to appear in Adv. Math. arxiv.org/abs/1606.01637
103. J.B. Rosser, L. Schoenfeld, Approximate formulas for some functions of prime numbers. Ill.
J. Math. 6, 64–94 (1962)
104. B. Saffari, The phase behavior of ultraflat unimodular polynomials, in Probabilistic
and Stochastic Methods in Analysis, with Applications, ed. by J.S. Byrnes, Jennifer L.
Byrnes, Kathryn A. Hargreaves, K. Berry (Kluwer Academic Publishers, Dordrecht, 1992),
pp. 555–572.
105. B. Saffari, Some polynomial extremal problems which emerged in the twentieth century, in
Twentieth Century Harmonic Analysis – A Celebration, ed. by J.S. Byrnes (Kluwer Academic
Publishers, Dordrecht, 2001), pp. 201–233
106. J. Sahasrabudhe, Counting zeros of cosine polynomials: on a problem of Littlewood. Adv.
Math. 343, 495–521 (2019)
107. E. Schmidt, Über algebraische Gleichungen vom Pólya-Bloch-Typos. Sitz. Preuss. Akad.
Wiss., Phys.-Math. Kl. 321 (1932)
108. I. Schur, Untersuchungen über algebraische Gleichungen. Sitz. Preuss. Akad. Wiss., Phys.-
Math. Kl. 403–428 (1933)
109. H.S. Shapiro, Extremal problems for polynomials and power series, Master thesis, MIT
(1951)
110. K. Soundararajan, Equidistribution of zeros of polynomials. Am. Math. Mon. 126(3), 226–
236 (2019, accepted)
Recent Progress in the Study of Polynomials with Constrained Coefficients 69

111. G. Szegő, Bemerkungen zu einem Satz von E. Schmidt uber algebraische Gleichungen. Sitz.
Preuss. Akad. Wiss., Phys.-Math. Kl. 86–98 (1934)
112. T. Tao, V. Vu, Local universality of zeros of random polynomials. Int. Math. Res. Not. 13,
5053–5139 (2015)
113. V. Totik, Distribution of simple zeros of polynomials. Acta Math. 170, 1–28 (1993)
114. V. Totik, P. Varjú, Polynomials with prescribed zeros and small norm. Acta Sci. Math.
(Szeged) 73, 593–612 (2007)
Classes of Nonnegative Sine Polynomials

Horst Alzer and Man Kam Kwong

Abstract We present several one-parameter classes of nonnegative sine polynomi-


als. One of our theorems states that the inequality
n 
 1 1
0≤ + (n − k + α) sin(kx) (α ∈ R)
n k
k=1

holds for all n ≥ 1 and x ∈ [0, π ] if and only if α ∈ [0, 3]. This extends a result of
Dimitrov and Merlo (2002), who proved the inequality for α = 1.

Keywords Sine polynomials · Inequalities

2010 Mathematics Subject Classification 26D05

1 Introduction and Statement of Main Results

The classical Fejér-Jackson-Gronwall inequality


n
sin(kx)
0< (n ≥ 1; 0 < x < π ) (1)
k
k=1

is one of the first inequalities for sine polynomials which appeared in the literature.
The validity of (1) was conjectured by Fejér in 1910 and the first proofs were
published by Jackson [14] and Gronwall [13] in 1911 and 1912, respectively. This

H. Alzer ()
Waldbröl, Germany
e-mail: [email protected]
M. K. Kwong
Department of Applied Mathematics, The Hong Kong Polytechnic University, Hong Kong, China
e-mail: [email protected]

© Springer Nature Switzerland AG 2020 71


A. Raigorodskii, M. T. Rassias (eds.), Trigonometric Sums and Their Applications,
https://doi.org/10.1007/978-3-030-37904-9_3
72 H. Alzer and M. K. Kwong

result has attracted the attention of many mathematicians, who discovered new
proofs as well as refinements, counterparts and variants of (1). In 1958, Vietoris
[24] proved a remarkable inequality for a certain class of sine polynomials which
includes (1) as special case.
Vietoris’ theorem If

a1 ≥ a2 ≥ · · · ≥ an > 0 and 2ka2k ≤ (2k − 1)a2k−1 (1 ≤ k ≤ n/2), (2)

then


n
0< ak sin(kx) (n ≥ 1; 0 < x < π ).
k=1

For more information on this theorem we refer to Alzer, Koumandos and


Lamprecht [1], Askey [3], Askey and Steinig [5], Koumandos [15], Kwong
[17, 18]. Numerous related results for further classes of trigonometric sums were
published by Belov [6], Brown and Hewitt [7], Brown and Wilson [8], Mondal and
Swaminathan [22], Turán [23] and others.
A collection of the most important and attractive inequalities for trigonometric
polynomials and its relatives as well as historical comments on this subject can be
found in Askey [2], Askey and Gasper [4], Milovanović, Mitrinović and Rassias [21,
chapter 4]. Information on applications of these inequalities in fields – for instance,
Fourier analysis, approximation theory and geometric function theory are given in
Dimitrov and Merlo [9] and Koumandos [16].
It is the aim of this paper to present new one-parameter classes of nonnegative
trigonometric polynomials. More precisely, we provide necessary and sufficient
conditions for the parameters such that certain sine polynomials are nonnegative
for all n and x ∈ [0, π ]. Our work has been inspired by an interesting research
article published by Dimitrov and Merlo [9] in 2002. One of the inequalities given
in [9] states that
n 
 1 1
0≤ + (n − k + 1) sin(kx) (n ≥ 1; 0 ≤ x ≤ π ). (3)
n k
k=1

Our first result offers an extension of (3).


Theorem 1 The inequality
n 
 1 1
0≤ + (n − k + α) sin(kx) (α ∈ R) (4)
n k
k=1

holds for all integers n ≥ 1 and real numbers x ∈ [0, π ] if and only if α ∈ [0, 3].
Classes of Nonnegative Sine Polynomials 73

Inequality (4) with α = 0 can be written in a more elegant form:

n−1
 n k
0≤ − sin(kx) (n ≥ 1; 0 ≤ x ≤ π ). (5)
k n
k=1

This inequality can be generalized. In fact, an application of Vietoris’ theorem


reveals that (5) remains valid if we replace the factor (n/k − k/n) by (n/k − k/n)β
with β ≥ 1. In view of this result it is natural to ask for all real parameters β such
that the modified inequality holds. The following theorem gives an answer to this
question.
Theorem 2 The inequality

n−1
 n k β
0≤ − sin(kx) (β ∈ R) (6)
k n
k=1

holds for all integers n ≥ 2 and real numbers x ∈ [0, π ] if and only if β ≥
log(2)/ log(16/5) = 0.59592 . . ..
It is well-known that (1) remains valid if the summation runs over only odd k.
Indeed, we have

n
sin(kx)
0< (n ≥ 1; 0 < x < π ). (7)
k=1
k
k odd

For an interesting extension of (7) we refer to Meynieux and Tudor [20]. With regard
to this result we ask for corresponding inequalities for the sums given in (4) and (6).
The following counterpart of (4) is valid.
Theorem 3 The inequality
n 
 1 1
0≤ + (n − k + a) sin(kx) (a ∈ R) (8)
k=1
n k
k odd

holds for all integers n ≥ 1 and real numbers x ∈ [0, π ] if and only if a ≥ 0.
Our fourth theorem offers a companion to inequality (6).
Theorem 4 The inequality
n−1 
 n k b
0≤ − sin(kx) (b ∈ R) (9)
k=1
k n
k odd

holds for all integers n ≥ 2 and real numbers x ∈ [0, π ] if and only if b ≥ 0.
74 H. Alzer and M. K. Kwong

In the next section, we collect several lemmas which we need to prove our
theorems. The proofs of the four theorems are presented in Sections 3–6. We
conclude the paper with a few remarks which are given in Section 7.
The numerical values have been calculated via the computer program
MAPLE 13.

2 Lemmas

The first three lemmas are due to Fejér [10–12]. They present inequalities for certain
sine polynomials.
Lemma 1 If a0 ≥ a1 ≥ · · · ≥ am ≥ 0 and θ ∈ [0, 2π ], then

m
ak sin((k + 1/2)θ ) ≥ 0. (10)
k=0

Lemma 2 If m ≥ 2 and x ∈ [0, π ], then


m−1
1
sin(kx) + sin(mx) ≥ 0.
2
k=1

We recall that a sequence {a1 , . . . , am } is said to be convex, if

ak−1 − 2ak + ak+1 ≥ 0 (k = 2, . . . , m − 1).

β β γ γ
Remark 1 Let γ > β > 0. If {a1 , . . . , am } is convex, then {a1 , . . . , am } is convex.
Lemma 3 If {a1 , . . . , am , 0} is convex and am ≥ 0, then, for x ∈ [0, π ],

m
ak sin(kx) ≥ 0.
k=1

Remark 2 The condition {a1 , . . . , am , 0} is convex is equivalent to {a1 , . . . , am } is


convex and am−1 ≥ 2am .
Lemma 4 If b3 ≥ 0 and

b3 + 2 |b2 − b3 | + 2 |2b4 − b3 | ≤ 1, (11)

then, for x ∈ [0, π ],


4
sin(x) + bk sin(kx) ≥ 0. (12)
k=2
Classes of Nonnegative Sine Polynomials 75

Proof Let R(x) be the expression on the left-hand side of (12). Then,


4
R(x) = Qk (x),
k=1

where
 
Q1 (x) = 1 − b3 − 2 |b2 − b3 | − 2 |2b4 − b3 | sin(x),

3
1 
Q2 (x) = b3 sin(kx) + sin(4x) ,
2
k=1
Q3 (x) = 2 |b2 − b3 | sin(x) + (b2 − b3 ) sin(2x),

Q4 (x) = 2 |2b4 − b3 | sin(x) + (b4 − b3 /2) sin(4x).

From (11) and Lemma 2 we conclude that Q1 and Q2 are nonnegative on [0, π ].
Let p1 = b2 − b3 and p2 = 2b4 − b3 . The representations
   
Q3 (x)=2 sin(x) |p1 |+p1 cos(x) and Q4 (x)=2 sin(x) |p2 |+p2 cos(x) cos(2x)

show that Q3 (x) ≥ 0 and Q4 (x) ≥ 0 for x ∈ [0, π ]. It follows that R is nonnegative
on [0, π ].
Lemma 5 Let
 7 β  2 β  7 β
Jβ (x) = 1 − +2 x+4 x2. (13)
45 5 45
If β ∈ [0.5, 1] and x ∈ R, then Jβ (x) > 0.
Proof Let

1  18 β
τβ = − .
4 7
Then, for β ∈ [0.5, 1] and x ∈ R,
 7 β 1  36 β  7 0.5 9
Jβ (x) ≥ Jβ (τβ ) = 1 − − ≥1− − = 0.34 . . . .
45 4 35 45 35
Lemma 6 Let
 7 β  2 β  3 β
y2 (β) = , y3 (β) = , y4 (β) = . (14)
16 9 32

If β ∈ [0.8, 1], then the sequence {1, y2 (β), y3 (β), y4 (β)} is convex.
76 H. Alzer and M. K. Kwong

Proof We have to show that the functions


 2 β  7 β  7 β  3 β  2 β
Φ(β) = 1 + −2 and Λ(β) = + −2
9 16 16 32 9
are nonnegative on [0.8, 1]. Let 0.8 ≤ r ≤ β ≤ s ≤ 1. Then,

Φ(β) ≥ Φ ∗ (r, s) and Λ(β) ≥ Λ∗ (r, s),

where
 2 s  7 r  7 s  3 s  2 r
Φ ∗ (r, s) = 1 + −2 and Λ∗ (r, s) = + −2 .
9 16 16 32 9
Since
 k k + 1
Φ ∗ (0.8, 1) > 0 and Λ∗ 0.8 + , 0.8 + > 0 (k = 0, 1, . . . , 19),
100 100
we conclude that Φ(β) > 0 and Λ(β) > 0 for β ∈ [0.8, 1].
Lemma 7 Let
 2 β  7 β  3 β
Y (β) = 1 + 3 −2 −4 . (15)
9 16 32

If β ∈ [0.59, 0.81], then Y (β) > 0.


Proof Let 0.59 ≤ r ≤ β ≤ s ≤ 0.81. Then,
 2 s  7 r  3 r
Y (β) ≥ 1 + 3 −2 −4 = Y ∗ (r, s), say.
9 16 32
We have
 k k + 1
Y ∗ 0.59 + , 0.59 + >0 (k = 0, 1, . . . , 54).
250 250
This yields Y (β) > 0 for β ∈ [0.59, 0.81].
Lemma 8 Let
3  1 β  11 β  16 β  1 β
z1 (β) = 1 + −5 , z2 (β) = − ,
2 7 175 35 7 (16)
 9 β  1 β
z3 (β) = − .
35 7
If β ∈ [0.59, 0.85], then the sequence {z1 (β), z2 (β), z3 (β), 0} is convex and
z3 (β) > 0.
Classes of Nonnegative Sine Polynomials 77

Proof We have to show that for β ∈ [0.59, 0.85],

K(β) − L(β) ≥ 0 and B(β) ≥ 0, (17)

where
5  1 β  9 β  11 β  16 β
K(β) = 1 + + , L(β) = 5 +2
2 7 35 175 35
and
 1 β  16 β  9 β
B(β) = + −2 .
7 35 35
Let 0.59 ≤ r ≤ β ≤ s ≤ 0.85. Then,

K(β) − L(β) ≥ K(s) − L(r) = M(r, s), say.

We have
 k k + 1
M 0.59 + , 0.59 + > 0 (k = 0, 1, . . . , 99)
2500 2500
and
 k k + 1
M 0.63 + , 0.63 + > 0 (k = 0, 1, . . . , 28).
130 130
This implies that the first inequality in (17) is valid for β ∈ [0.59, 0.85].
Let
 35 β
W (β) = B(β).
9
Since

W (β) ≥ 0 (β ∈ R), W (0.59) = 0.39 . . ., W (0.59) = 0.11 . . . ,

we conclude that the second inequality in (17) holds for β ≥ 0.59.

3 Proof of Theorem 1

We denote the sum in (4) by Sn,α (x). First, we assume that (4) holds for all n ≥ 1
and x ∈ [0, π ]. Then,

0 ≤ S1,α (x) = 2α sin(x).


78 H. Alzer and M. K. Kwong

This gives α ≥ 0. Moreover, since


3 
0 ≤ S2,α (x) = sin(x) (1 + α) + 2α cos(x) ,
2
we get
3
0≤ (1 + α) + 2α cos(x).
2
We let x → π and obtain
3 1
0≤ (1 + α) − 2α = (3 − α).
2 2
Thus, α ≤ 3.
Now, we assume that n ≥ 1 and α ∈ [0, 3]. We define
1 1
xk = xk (n, α) = + (n − k + α) (k = 1, . . . , n).
n k
Since we have for k ≥ 1,

k 2 + k + n2 + nα
xk − xk+1 = >0
nk(k + 1)
and
4k − 1 − α
(2k − 1)x2k−1 − 2kx2k = ≥ 0,
n
we conclude that (2) holds, so that Vietoris’ theorem gives Sn,α (x) ≥ 0 for x ∈[0, π ].

4 Proof of Theorem 2

Inequality (6) is equivalent to


n−1 2
 n − k 2 β
0≤ sin(kx) = Tn,β (x), say. (18)
(n2 − 1)k
k=1

If (18) is valid for all n ≥ 2 and x ∈ [0, π ], then we obtain for x ∈ (0, π ):
T3,β (x)  5 β
0≤ =1+2 cos(x).
sin(x) 16
We let x → π and get
 5 β log(2)
0≤1−2 or β ≥ β1 = = 0.59592 . . . .
16 log(16/5)
Classes of Nonnegative Sine Polynomials 79

Next, let n ≥ 2 and x ∈ [0, π ]. As pointed out in Section 1, in order to prove (18)
we may assume that β ∈ [β1 , 1]. First, we consider the cases n = 2, 3, 4, 5, 6.
Cases n = 2, 3, 4 We have

T2,β (x) = sin(x) ≥ 0

and
  5 β    5 β 
T3,β (x) = sin(x) 1 + 2 cos(x) ≥ sin(x) 1 − 2 ≥ 0.
16 16
A short calculation yields that

T4,β (x) = sin(x)Jβ (X),

where Jβ is defined in (13) and X = cos(x). Using Lemma 5 reveals that T4,β (x) ≥
0.
Case n = 5 We have


4
T5,β (x) = sin(x) + yk (β) sin(kx)
k=2

where y2 (β), y3 (β), y4 (β) are defined in (14). Let β2 = 0.803 . . . be given by
 2 β2  3 β2
=2 .
9 32

Case 1: β ≥ β2 . Then, y3 (β) ≥ 2y4 (β). We apply Lemmas 3 and 6 and Remark 2
and find that T5,β (x) ≥ 0.
Case 2: β < β2 . We have 2y4 (β) > y3 (β) and y2 (β) > y3 (β). This implies

1 − y3 (β) − 2 |y2 (β) − y3 (β) | − 2 |2y4 (β) − y3 (β)|

= 1 − 2y2 (β) + 3y3 (β) − 4y4 (β) = Y (β),

where Y is defined in (15). Applying Lemmas 4 and 7 yields T5,β (x) ≥ 0.


Case n = 6 We obtain


5
T6,β (x) = sin(x) + dk (β) sin(kx),
k=2
80 H. Alzer and M. K. Kwong

where
 16 β  9 β  1 β  11 β
d2 (β) = , d3 (β) = , d4 (β) = , d5 (β) = .
35 35 7 175
By direct computation we obtain that the sequence {1, d2 (β), d3 (β), d4 (β), d5 (β)}
is convex for β = 1/2, so that Remark 1 reveals that this sequence is convex for
β > 1/2. Let β3 = 0.844 . . . be given by
 1 β3  11 β3
=2 .
7 175
Case 1: β ≥ β3 . Then, {1, d2 (β), d3 (β), d4 (β), d5 (β)} is convex and d4 (β) ≥
2d5 (β). Applying Lemma 3 and Remark 2 leads to T6,β (x) ≥ 0.
Case 2: β < β3 . Let


4
1 
P1,β (x) = d4 (β) sin(kx) + sin(5x) ,
2
k=1
 1  
P2,β (x) = d5 (β) − d4 (β) 5 sin(x) + sin(5x)
2
and

P3,β (x) = z1 (β) sin(x) + z2 (β) sin(2x) + z3 (β) sin(3x)

with z1 (β), z2 (β), z3 (β) as defined in (16). Then we have the representation

T6,β (x) = P1,β (x) + P2,β (x) + P3,β (x).

Using Lemma 2 gives P1,β (x) ≥ 0. Since d5 (β) > d4 (β)/2 and

| sin(mx)| ≤ m sin(x) (m ∈ N; 0 ≤ x ≤ π ),

we obtain P2,β (x) ≥ 0. From Lemmas 3 and 8 we conclude that P3,β (x) ≥ 0. Thus,
T6,β (x) ≥ 0.
Case n ≥ 7 Let

Tn,β (x) = Tn,β (π − x).

The following statements are valid.


∗ is nonnegative on [2.5/n, π ].
Proposition 1 Tn,β
∗ is nonnegative on [0, π/n] if n is even.
Proposition 2 Tn,β
∗ is nonnegative on [0, 2.5/n] if n is odd.
Proposition 3 Tn,β
Classes of Nonnegative Sine Polynomials 81

From Propositions 1 and 2 we conclude that (18) is valid for x ∈ [0, π ] and even
n ≥ 8, and Propositions 1 and 3 reveal that this inequality holds for all x ∈ [0, π ]
and odd n ≥ 7.
Outline of the Proofs The proofs of the three propositions require numerous tech-
nical and lengthy computations which are collected in a separate paper published in
arXiv; see [19].
Proposition 1 is proved by carefully applying the Comparison Principle to Tn,β ∗ ,
m
making use of the polynomials k=1 (−1)k+1 sin(kx) which have known closed
forms.
∗ . The proofs of
Let {a1 , a2 , . . . , an−1 } denote the coefficient sequence of Tn,β
the other two Propositions are based on the observation that there exists an integer
m ≤ n, such that the subsequence {a1 , a2 , . . . , am } is convex while the sequence
∗ can then
{am , a2 , . . . , an−1 , 0} has an odd number of terms and is concave. Tn,β
be decomposed into the sum of two sine polynomials, the first constructed using
the first sub-sequence of coefficients and the second polynomial using the second
sub-sequence. Lower bounds for these polynomials are then derived with careful,
albeit rather technical, analysis. The lower bounds obtained happen to be enough to
ensure that the sum is nonnegative.

5 Proof of Theorem 3

Let Sn,α (x), Hn (x), Un,a (x) be the sums given in (4), (7), (8), respectively. If (8)
holds for all n ≥ 1 and x ∈ [0, π ], then

0 ≤ U1,a (x) = 2a sin(x).

This yields a ≥ 0.
Next, we assume that n ≥ 1 and a ≥ 0. We define


n
sin2 (N x)
Gn (x) = sin(kx) = (N = [(n + 1)/2]). (19)
k=1
sin(x)
k odd

From (7) and (19) we obtain for x ∈ [0, π ],

∂ 1
Un,a (x) = Gn (x) + Hn (x) ≥ 0.
∂a n
Thus,

1 
Un,a (x) ≥ Un,0 (x) = Sn,0 (x) + Sn,0 (π − x) . (20)
2
Using (4) with α = 0 we conclude from (20) that (8) holds.
82 H. Alzer and M. K. Kwong

6 Proof of Theorem 4

We denote the sum in (9) by Vn,b (x). If (9) is valid for all n ≥ 2 and x ∈ [0, π ],
then
1
0 ≤ V4,b (π/2) = (45b − 7b ).
12b
This yields b ≥ 0.
Next, let n ≥ 2, x ∈ [0, π ] and b ≥ 0. We set θ = 2x and m = [(n − 1)/2].
Then, we obtain


m
Vn,b (x) = ck sin((k + 1/2)θ ),
k=0

where
 n 2k + 1 b
ck = ck (n, b) = − .
2k + 1 n

Since x → (n/x − x/n)b is nonnegative and decreasing on (0, n], we obtain c0 ≥


c1 ≥ · · · ≥ cm ≥ 0, so that (10) leads to Vn,b (x) ≥ 0.

7 Remarks

The four theorems given in Section 1 provide lower bounds for the sine polynomials.
We ask: do there exist upper bounds for these polynomials which do not depend on
n and x? The answer is “no”. We consider the sum T̃n,β (x) given in (6). Let β ∈ R.
Applying the arithmetic mean – geometric mean inequality yields for n ≥ 2,

1  n k β
n−1
1
T̃n,β (π/n) = − sin(kπ/n)
n−1 n−1 k n
k=1

 n

n−1 k β 1/(n−1)
≥ − sin(kπ/n) = δn,β , say.
k n
k=1

Since


n−1
n k (2n − 1)! 
n−1
n
− = and sin(kπ/n) = ,
k n (n − 1)!nn 2n−1
k=1 k=1
Classes of Nonnegative Sine Polynomials 83

we get

1
 1 1/(n−1)  σ2n 1/(n−1)  4 n/(n−1) β 1/(n−1)
δn,β = √ · · ·n , (21)
2 2 σn e

where

n!en
σn = √ .
nn n

Using Stirling’s formula



lim σn = 2π
n→∞

we conclude from (21) that

1
4 β 1  4 β
lim δn,β = 1·1· ·1= . (22)
n→∞ 2 e 2 e
Since

T̃n,β (π/n) ≥ (n − 1)δn,β ,

we obtain from (22) that

lim T̃n,β (π/n) = ∞.


n→∞

This reveals that there is no upper bound for T̃n,β (x) which is independent of n and
x. Similarly, we can show that this is also true for the other three sine polynomials.

References

1. H. Alzer, S. Koumandos, M. Lamprecht, A refinement of Vietoris’ inequality for sine


polynomials. Math. Nach. 283, 1549–1557 (2010)
2. R. Askey, Orthogonal Polynomials and Special Functions. Regional Conference Series in
Applied Mathematics, vol. 21 (SIAM, Philadelphia, 1975)
3. R. Askey, Vietoris’s inequalities and hypergeometric series, in Recent Progress in Inequalities,
ed. by G.V. Milovanović (Kluwer, Dordrecht, 1998) pp. 63–76
4. R. Askey, G. Gasper, Inequalities for polynomials, in The Bieberbach Conjecture, ed. by A.
Baernstein II et al. Mathematical Surveys and Monographs, vol. 2 (American Mathematical
Society, Providence, 1986), pp. 7–32
5. R. Askey, J. Steinig, Some positive trigonometric sums. Trans. Am. Math. Soc. 187, 295–307
(1974)
84 H. Alzer and M. K. Kwong

6. A.S. Belov, Examples of trigonometric series with non-negative partial sums. Math. USSR Sb.
186, 21–46 (1995) (Russian); 186, 485–510 (English translation)
7. G. Brown, E. Hewitt, A class of positive trigonometric sums. Math. Ann. 268, 91–122 (1984)
8. G. Brown, D.C. Wilson, A class of positive trigonometric sums, II. Math. Ann. 285, 57–74
(1989)
9. D.K. Dimitrov, C.A. Merlo, Nonnegative trigonometric polynomials. Constr. Approx. 18, 117–
143 (2002)
10. L. Fejér, Über die Positivität von Summen, die nach trigonometrischen oder Legendreschen
Funktionen fortschreiten. Acta Litt. Sci. Szeged 2, 75–86 (1925)
11. L. Fejér, Einige Sätze, die sich auf das Vorzeichen einer ganzen rationalen Funktion
beziehen. . .. Monatsh. Math. Phys. 35, 305–344 (1928)
12. L. Fejér, Trigonometrische Reihen und Potenzreihen mit mehrfach monotoner Koeffizienten-
folge. Trans. Am. Math. Soc. 39, 18–59 (1936)
13. T.H. Gronwall, Über die Gibbssche Erscheinung und die trigonometrischen Summen sin x +
2 sin 2x + · · · + n sin nx. Math. Ann. 72, 228–243 (1912)
1 1

14. D. Jackson, Über eine trigonometrische Summe. Rend. Circ. Mat. Palermo 32, 257–262 (1911)
15. S. Koumandos, An extension of Vietoris’s inequalities. Ramanujan J. 14, 1–38 (2007)
16. S. Koumandos, Inequalities for trigonometric sums, in Nonlinear Analysis, ed. by P.M.
Pardalos et al. Optimization and Its Applications, vol. 68 (Springer, New York, 2012), pp. 387–
416
17. M.K. Kwong, An improved Vietoris sine inequality. J. Approx. Theory 189, 29–42 (2015)
18. M.K. Kwong, Improved Vietoris sine inequalities for non-monotone, non-decaying coeffi-
cients. arXiv:1504.06705 [math.CA] (2015)
n−1  n 
k β
19. M.K. Kwong, Technical details of the proof of the sine inequality k=1 k − n sin(kx) ≥ 0,
arXiv:1702.03387 [math.CA] (2017)
20. R. Meynieux, Gh. Tudor, Compleménts au traité de Mitrinović III: Sur un schéma général pour
obtenir des inégalités. Univ. Beograd. Publ. Elektrotehn. Fak. Mat. Fiz. 412–460, 171–174
(1973)
21. G.V. Milovanović, D.S. Mitrinović, Th.M. Rassias, Topics in Polynomials: Extremal Problems,
Inequalities, Zeros (World Scientific, Singapore, 1994)
22. S.R. Mondal, A. Swaminathan, On the positivity of certain trigonometric sums and their
applications. Comput. Math. Appl. 62, 3871–3883 (2011)
23. P. Turán, On a trigonometrical sum. Ann. Soc. Polon. Math. 25, 155–161 (1952)
24. L. Vietoris, Über das Vorzeichen gewisser trigonometrischer Summen, Sitzungsber. Öst. Akad.
Wiss. 167, 125–135 (1958); Anz. Öst. Akad. Wiss. 159, 192–193
Inequalities for Weighted Trigonometric
Sums

Horst Alzer and Omran Kouba

Abstract We prove that the double-inequality


⎛ ⎞a ⎛ ⎞b

n
wj 
n
wj 
n
wj 
n
wj
⎝ ⎠ ≤ · ≤⎝ ⎠
jπ jπ jπ jπ
j =1 1 − sin2 n+1 j =1 1 − sin n+1 j =1 1 + sin n+1 j =1 1− sin2 n+1

holds for all even integers n ≥ 2 and positive real numbers wj (j = 1, . . . , n) with
w1 + · · · + wn = 1 if and only if a ≤ 1 and b ≥ 2. Moreover, we present a cosine
counterpart of this result.

Keywords Inequalities · Sine sums · Cosine sums · Optimal constants

1 Introduction

In the literature, we can find many beautiful identities involving finite trigonometric
sums. As an example, we mention a reciprocity theorem for the tangent sum


k−1
tan(hj π/k)
E(h, k) =
tan(2j π/k)
j =1

which was published by Eisenstein [5, pp. 108–110] in 1844:

H. Alzer ()
Waldbröl, Germany
e-mail: [email protected]
O. Kouba
Department of Mathematics, Higher Institute for Applied Sciences and Technology, Damascus,
Syria
e-mail: [email protected]

© Springer Nature Switzerland AG 2020 85


A. Raigorodskii, M. T. Rassias (eds.), Trigonometric Sums and Their Applications,
https://doi.org/10.1007/978-3-030-37904-9_4
86 H. Alzer and O. Kouba

(h − k)2
hE(h, k) + kE(k, h) = − .
2
Here, h and k are odd natural numbers which are relatively prime. A proof of
Eisenstein’s identity was given by Stern [16] in 1861.
The theory of finite trigonometric sums has attracted the attention of numerous
researchers, mainly because these sums have remarkable applications in various
mathematical branches, like, for example, geometry, theory of matrices, number
theory, graph theory, and even in physics. Detailed information on this subject can
be found in Berndt and Yeap [3], the recently published papers Fonseca et al. [6, 7],
Kouba [12], Merca [13] and the references cited therein.
A proof for the elementary formula


n
1 n(n + 2)
 = (1)
1 − cos2 jπ 3
j =1 n+1

can be found in Chen [4]. From (1) we obtain


n
1 
n
1 n(n + 2)
 =  = . (2)
1 − cos jπ
1 + cos jπ 3
j =1 n+1 j =1 n+1

In this paper, we present inequalities for the weighted versions of the cosine sums
given in (1) and (2),


n
wj
Cn (w) =  ,

j =1 1 − cos n+1


n
wj
Cn∗ (w) =  , (3)

j =1 1 + cos n+1

Cn (w) + Cn∗ (w) 


n
wj
An (w) = =  ,
2 1 − cos2 jπ
j =1 n+1

where
* +
w ∈ Wn  (w1 , . . . , wn ) ∈ Rn | wj > 0 for j =1, . . . , n, and w1 + · · · +wn =1 .

More precisely, we determine all real parameters α and β such that the
inequalities

(An (w))α ≤ Cn (w) Cn∗ (w) ≤ (An (w))β


Inequalities for Weighted Trigonometric Sums 87

are valid for all n ≥ 1 and w ∈ Wn .


We study also the same problem for the corresponding weighted sine sums


n
wj
Sn (w) =  ,

j =1 1 − sin n+1


n
wj
Sn∗ (w) =  , (4)

j =1 1 + sin n+1

Sn (w) + Sn∗ (w) 


n
wj
Bn (w) = =  ,
2 1 − sin2 jπ
j =1 n+1

where n ≥ 2 is an even integer.


A key role in the proofs of our inequalities plays the double-inequality


n
wj 
n
wj 
n
wj 
n
wj 2
≤ · ≤ (5)
1 − pj2 1 − pj 1 + pj 1 − pj2
j =1 j =1 j =1 j =1

which holds for all w ∈ Wn and pj ∈ (−1, 1), (j = 1, . . . , n). An application


of the classical arithmetric mean – geometric mean inequality written in the form
xy ≤ (x/2 + y/2)2 gives the right-hand side of (5). The left-hand side is due to
Milne [14], who published an integral version in 1925. The identity


n
wj  wj
n 
n
wj 1   wi wj (pi − pj )2
n n
− =
1 − pj 1 + pj 1 − pj2 2 (1 − pi2 )(1 − pj2 )
j =1 j =1 j =1 i=1 j =1

leads to a short proof of Milne’s result which is a special case of the well-known
Chebyshev inequality; see [9, section 2.17]. For more information on Milne’s
inequality we refer to Alzer and Kovačec [1, 2] and Rao [15].
Throughout, we maintain the above notation. In Section 2 we state and prove a
lemma and in Section 3 we present two theorems, a corollary and their proofs.

2 A Technical Lemma

We study the sums in (4) with w1 = · · · = wn = 1. Let n ≥ 2 be an even integer


and let


n
1 
n
1  n
1
Fn =   , Gn =   , Hn =  .
jπ jπ jπ
j =1 1 − sin2 n+1 j =1 1 + sin n+1 j =1 1 − sin n+1
(6)
88 H. Alzer and O. Kouba

The following technical lemma is used in the proof of Theorem 2.


Lemma Let n ≥ 2 be an even integer. Then,

Fn = n(n + 2), (7)


 
2(n + 1) π 1
Gn = −1+ +O , (8)
π 6(n + 1) n3
 
2(n + 1) π 1
Hn = 2(n + 1)2 − −1− +O . (9)
π 6(n + 1) n3

Proof (i) Let n = 2m. We denote by Tk the Chebyshev polynomial of the first kind
of degree k which is defined by Tk (cos(θ )) = cos(kθ ). Let

π jπ
θj = − with j ∈ {−m, . . . , m}.
2 2m + 1

Then,
 jπ 
sin = cos(θj ).
2m + 1

It follows that
  
jπ     
T2m+1 sin = T2m+1 cos(θj ) = cos (2m + 1)θj
2m + 1
π 
= cos − (j − m)π = sin((j − m)π ) = 0
2
which implies that (sin(j π/(2m + 1)))−m≤j ≤m are the zeros of T2m+1 . Thus, there
exists a constant λ such that
m   j π 
 m 
  j π 
T2m+1 (x) = λ x − sin = λx x 2 − sin2 .
2m + 1 2m + 1
j =−m j =1

Taking the logarithmic derivative gives


1 
m
T2m+1 (x) 2x
= +  . (10)
T2m+1 (x) x x − sin2
2 jπ
j =1 2m+1


We set x = 1 and use T2m+1 (1) = 1, T2m+1 (1) = (2m + 1)2 . Then, (10) yields


m
1
(2m + 1) = 1 + 2
2
 jπ 
j =1 1 − sin 2
2m+1
Inequalities for Weighted Trigonometric Sums 89


2m
1
=1+  jπ 
j =1 1 − sin 2
2m+1

= 1 + F2m

which leads to (7).


(ii) We apply the Euler-Maclaurin formula


N  
N f (0) + f (N) f (N ) − f (0) 1 N
f (j ) = f (x)dx+ + − P ({x})f (x)dx,
0 2 12 2 0
j =0

where {x} denotes the fractional part of x, and P (x) = x 2 − x + 1/6; see Knopp
[10, section 64] and Kouba [11, section 8].
Taking N = n + 1,

1
f (x) =  πx  ,
1 + sin n+1

and noting that


N
f (0) + f (N) = 2, f (j ) = 2 + Gn ,
j =0

 N 2(n + 1) 2π
f (x)dx = , f (N ) − f (0) = ,
0 π n+1

we obtain
2(n + 1) π
Gn = −1+ + Rn , (11)
π 6(n + 1)

where
 n+1
1
Rn = − P ({x})f (x)dx.
2 0

We have
n 
 k+1
−2Rn = P (x − k)f (x)dx
k=0 k
90 H. Alzer and O. Kouba

n 
 1
= P (x)f (x + k)dx
k=0 0
 1 
n
= P (x) f (x + k)dx.
0 k=0

In order to go one step further, we consider Q the Bernoulli polynomial of degree


4, that is,

1
Q(x) = x 4 − 2x 3 + x 2 − .
30
Then,

1
Q(0) = Q(1) = − , Q (0) = Q (1) = 0, Q (x) = 12P (x).
30
Two integrations by parts yield
 1 
n

−24Rn = Q (x) f (x + k)dx
0 k=0
 1 
n
=− Q (x) f (x + k)dx
0 k=0
 
n
f (n + 1) − f (0) 1
= + Q(x) f (4) (x + k)dx.
30 0 k=0

So, if g is the function defined by

1
g(x) =
1 + sin(π x)

then,

1  x 
f (k) (x) = g (k)
(0 ≤ k ∈ Z).
(n + 1)k n+1

Thus,
 
n x + k 
1   1 1
−24Rn = g (1) − g (0) + Q(x) g (4) dx.
30(n + 1) 3 (n + 1)4 0 n+1
k=0
Inequalities for Weighted Trigonometric Sums 91

Hence, with M  sup0≤x≤1 |g (4) (x)| we conclude that there exists a real number
M∗ such that

M M 1 M∗
24|Rn | ≤ + |Q(x)|dx ≤ , (12)
30(n + 1)3 (n + 1)3 0 (n + 1)3

and (8) follows from (11) and (12).


(iii) We have Hn = 2Fn − Gn , so that (7) and (8) lead to (9). 

Remark 1 The referee pointed out that (7) follows from the representation
n 
  j π 
Fn = 1 + tan2
n+1
j =1

and the identity


n  jπ 
tan2 = n(n + 1) (13)
n+1
j =1

which is valid for all even integers n ≥ 2; see Hansen [8, p. 646]. It is also worth
mentioning that (13) appears in the work of Stern [16, p. 155].

3 Main Results

We are now in a position to present our main results. First, we offer inequalities
involving the three cosine sums given in (3).
Theorem 1 Let α and β be real numbers. The inequalities

(An (w))α ≤ Cn (w) Cn∗ (w) ≤ (An (w))β (14)

hold for all integers n ≥ 1 and w ∈ Wn if and only if α ≤ 1 and β ≥ 2.


Proof We set
 

pj = cos for j = 1, . . . , n,
n+1

then (5) gives

An (w) ≤ Cn (w)Cn∗ (w) ≤ (An (w))2 . (15)

Since An (w) ≥ 1, we obtain for α ≤ 1 and β ≥ 2,


92 H. Alzer and O. Kouba

(An (w))α ≤ An (w) and (An (w))2 ≤ (An (w))β . (16)

From (15) and (16) we conclude that (14) is valid.


Next, suppose that (14) holds for all n ≥ 1 and w ∈ Wn . In particular, for n = 2
we have
2 4 4 4
C2 (w) = + w1 , C2∗ (w) = 2 − w1 , A2 (w) = ,
3 3 3 3
so (14) leads to
 α  β
4 4 4
≤ (1 + 2w1 )(3 − 2w1 ) ≤ (0 < w1 < 1).
3 9 3

Taking w1 = 1/2 we obtain β ≥ 2 and letting w1 tend to 0+ we obtain α ≤ 1. 



Our second theorem provides a sine counterpart of (14) involving the sums
defined in (4). It turns out that its proof is more difficult than the proof of Theorem 1.
The hard part is to show that the right-hand side of (17) (given below) is in general
not true if b < 2.
Theorem 2 Let a and b be real numbers. The inequalities

(Bn (w))a ≤ Sn (w) Sn∗ (w) ≤ (Bn (w))b (17)

hold for all even integers n ≥ 2 and w ∈ Wn if and only if a ≤ 1 and b ≥ 2.


Proof We apply (5) with
 

pj = sin for j = 1, . . . , n
n+1

and use that Bn (w) ≥ 1. Then, for a ≤ 1 and b ≥ 2,

(Bn (w))a ≤ Bn (w) ≤ Sn (w) Sn∗ (w) ≤ (Bn (w))2 ≤ (Bn (w))b .

It follows that (17) holds if a ≤ 1 and b ≥ 2.


Now assume that (17) is valid for all even n ≥ 2 and w ∈ Wn . Then considering
the case n = 2 we get

S2 (w)S2∗ (w)
1≤ = 41−a .
(B2 (w))a

Thus, a ≤ 1.
Inequalities for Weighted Trigonometric Sums 93

Next, we prove that b ≥ 2. The right-hand side of (17) is equivalent to


 
log Sn (w)Sn∗ (w)
Φn (w) =   ≤ b. (18)
log Bn (w)

Let n ≥ 4, t > 0 and

1 1 t t t 1 t
wt = + , ,..., , + .
1+t 2 n n n 2 n

Then,
   
1 1 Hn 1 1 Gn
Sn (wt )=  π  +t , Sn∗ (wt ) =  π  +t ,
1+t 1− sin n+1 n 1+t 1 + sin n+1 n

and
 
1 1 Fn
Bn (wt ) =  π  +t ,
1+t 1 − sin n+1
2 n

where Hn , Gn and Fn are defined in (6).


We obtain
      π 
log Sn (wt ) Sn∗ (wt ) = log (1+tJn )(1+tKn ) −2 log(1+t)− log cos2 ,
n+1
(19)
    π 
log Bn (wt ) = log(1 + tLn ) − log(1 + t) − log cos2 (20)
n+1

with
  π  H
n
Jn = 1 − sin ,
n+1 n
  π  G
n
Kn = 1 + sin ,
n+1 n
 π F
n
Ln = cos2 .
n+1 n

Using (7), (8), (9) and the asymptotic expansions


 π  π π 1  π  π2 1
sin = − 2 +O 3 , cos2 =1− 2 +O 3
n+1 n n n n+1 n n
94 H. Alzer and O. Kouba

gives

2(π − 1)2 2π 2 − 3π + 2 1


Jn = 2n − − +O 2 ,
π πn n
2 π +2  1 
Kn = + +O 2 , (21)
π πn n
π2 1
Ln = n + 2 − +O 2 .
n n

Now, we set t = π/n2 and apply (19), (20), (21) and

1
log(1 + x) = x − x 2 + O(x 3 ) (x → 0).
2
This yields

  2π 2π − 3π 2 1
log Sn (wπ/n2 ) Sn∗ (wπ/n2 ) = + + O (22)
n n2 n3
and
  π π + π 2 /2 1
log Bn (wπ/n2 ) = + + O . (23)
n n2 n3
From (18), (22) and (23) we conclude that

4π 1
Φn (wπ/n2 ) = 2 − +O 2 .
n n
Thus,
 
lim n 2 − Φn (wπ/n2 ) = 4π.
n→∞

In particular, we obtain

lim Φn (wπ/n2 ) = 2. (24)


n→∞

Using (18) and (24) gives b ≥ 2. 



Remark 2 If the weight w = (w1 , . . . , wn ) is symmetric, that is, wj = wn+1−j for
j = 1, . . . , n, then we have Cn (w) = Cn∗ (w) = An (w), so that the second inequality
in (14) (with β = 2) reduces to an equality. This is not true for our sine sums.
An application of the left-hand sides of (14) and (17) leads to a sharp inequality
involving Cn (w), Cn∗ (w), Sn (w), Sn∗ (w) and the sine sum
Inequalities for Weighted Trigonometric Sums 95


n
wj
Dn (w) =  .
2 2j π
j =1 sin n+1

Corollary For all even integers n ≥ 2 and w ∈ Wn we have

4 Dn (w) ≤ Cn (w) Cn∗ (w) + Sn (w) Sn∗ (w). (25)

The constant factor 4 is best possible.


Proof We have

An (w) + Bn (w) = 4 Dn (w),

so that (14) and (17) with α = a = 1 lead to (25). Moreover, since

C2 (w) C2∗ (w) + S2 (w) S2∗ (w) 4


= 4 + w1 (1 − w1 ),
D2 (w) 3

we conclude that in (25) the factor 4 cannot be replaced by a larger number. 




Acknowledgements We thank the referee for helpful comments.

References

1. H. Alzer, A. Kovačec, The inequality of Milne and its converse. J. Inequal. Appl. 7, 603–611
(2002)
2. H. Alzer, A. Kovačec, The inequality of Milne and its converse, II. J. Inequal. Appl. 2006, 7,
article ID 21572 (2006)
3. B.C. Berndt, B.P. Yeap, Explicit evaluations and reciprocity theorems for finite trigonometric
sums. Adv. Appl. Math. 29, 358–385 (2002)
4. H. Chen, On some trigonometric power sums. Int. J. Math. Math. Sci. 30, 185–191 (2002)
5. G. Eisenstein, Mathematische Werke, Band I (Chelsea, New York, 1975)
6. C.M. da Fonseca, M.L. Glasser, V. Kowalenko, Basic trigonometric power sums with applica-
tions. Ramanujan J. 42, 401–428 (2017)
7. C.M. da Fonseca, M.L. Glasser, V. Kowalenko, Generalized cosecant numbers and trigonomet-
ric sums. Appl. Anal. Disc. Math. 12, 70–109 (2018)
8. E.R. Hansen, A Table of Series and Products (Prentice Hall, Englewood Cliffs, 1975)
9. G.H. Hardy, J.E. Littlewood, G. Pólya, Inequalities (Cambridge University Press, Cambridge,
1952)
10. K. Knopp, Theorie und Anwendung der unendlichen Reihen (Springer, Berlin, 1964)
11. O. Kouba, Lecture notes. Bernoulli polynomials and applications, arXiv:1309.7569v2
[math.CA]
12. O. Kouba, Inequalities for finite trigonometric sums. An interplay: with some series related to
harmonic numbers. J. Inequal. Appl. 2016, 15, paper no. 173 (2016)
13. M. Merca, A note on cosine power sums. J. Integer Seq. 15, 7, article 12.5.3 (2012)
96 H. Alzer and O. Kouba

14. E.A. Milne, Note on Rosseland’s integral for the stellar absorption coefficient. Mon. Not. R.
Astron. Soc. 85, 979–984 (1925)
15. C.R. Rao, Statistical proofs of some matrix inequalities. Linear Algebra Appl. 321, 307–320
(2000)
16. M. Stern, Ueber einige Eigenschaften der Function Ex. J. Reine Angew. Math. 59, 146–162
(1861)
Norm Inequalities for Generalized
Laplace Transforms

J. C. Kuang

Abstract This paper introduced the new generalized Laplace transform. It contains
the generalized Stieltjes transform and the Hankel transform etc. The corresponding
new operator norm inequalities are obtained.The discrete versions of the main
results are also given.As applications,a large number of known and new results have
been obtained by proper choice of kernel. They are significant improvement and
generalizations of many famous results.

Keywords Laplace transform · Integral operator · Norm inequality

Mathematics Subject Classification 47A30

1 Introduction

Given a function f on (0, ∞) such that e−αy |f (y)| is integrable over the interval
(0, ∞) for some real α, we define F (z) as
 ∞
F (z) = e−zy f (y)dy, (1)
0

where we require that Re(z) > α so that the integral in (1) converges.F is called the
(one-sided)Laplace transform of f . We consider only one-sided Laplace transforms
with the real parameter x, that is,
 ∞
F (x) = e−xy f (y)dy, x > α, (2)
0

J. C. Kuang ()
Department of Mathematics, Hunan Normal University Changsha, Hunan, P. R. China
e-mail: [email protected]

© Springer Nature Switzerland AG 2020 97


A. Raigorodskii, M. T. Rassias (eds.), Trigonometric Sums and Their Applications,
https://doi.org/10.1007/978-3-030-37904-9_5
98 J. C. Kuang

as these play the most important role in the solution of initial and boundary
value problems for partial differential equations (see [1–3]). In fact, the tools
we shall use for solving Cauchy and initial and boundary value problems are
integral transforms. Specifically, we shall consider the Fourier transform, the Fourier
sine and cosine transforms, the Hankel transform, and the Laplace transform. In
asymptotic analysis, we often study functions f which have N + 1 continuous
derivatives while f N +2 is piecewise continuous on (0, ∞), then by [4], we have
 ∞ 
N
F (x) = e−xy f (y)dy ∼ x −(k+1) f (k) (0)
0 k=0

represents an asymptotic expansion of F , as x → ∞, to N + 1 terms. But the


following Hardy’s results of being neglected [5]:
p
Theorem 1 If f ∈ Lω (0, ∞), 1 < p < ∞, ω(x) = x p−2 , then F is defined by (2)
satisfies

F p ≤ Γ (1/p)f p,ω , (3)


∞
where f p,ω = ( 0 |f (x)|p ω(x)dx)1/p and Γ (α) is the Gamma function:
 ∞
Γ (α) = x α−1 e−x dx (α > 0).
0

It is important to note that x appear in (2) only through the product xy, this suggests
that, as a generalization, Hardy [5] introduced the generalized Laplace transform
of f :
 ∞
T0 (f, x) = K(xy)f (y)dy, x ∈ (0, ∞), (4)
0

and proved the following


Theorem 2 Let K be a non-negative and measurable function on (0, ∞). If f be a
non-negative and not null,1 < p < ∞, p1 + q1 = 1, then the integral operator T0 is
p p
defined by (4):T0 : Lω (0, ∞) → Lp (0, ∞) and Lp (0, ∞) → Lω (0, ∞) exist as a
bounded operator and
1 1
T0 f p ≤ c( )f p,ω , T0 f p,ω ≤ c( )f p , (5)
p q

where ω(x) = x p−2 and


 ∞
c(s) = K(t)t s−1 dt. (6)
0

The constants are the best possible.


Norm Inequalities for Generalized Laplace Transforms 99

Remark 1 When K(xy) = e−xy , Theorem 2 reduces to Theorem 1.


In 2015, the author [6] introduced the wider class of integral operators

T1 (f, x) = K(xλ1 · yλ2 )f (y)dy, (7)
Rn+

n x ∈2 1/2
where Rn+ = {x = (x1 , x2 , · · · , xn ) : xk ≥ 0, 1 ≤ k ≤ n}, x =
( k=1 |xk | ) , λ1 × λ2 = 0, and obtained the operator norm inequalities for T1
defined by (7) on the multiple weighted Orlicz spaces. In particular, if n = 1, λ1 =
λ2 = 1, then T1 in (7) reduces to T0 in (4). In this paper, we introduce the new
integral operator T defined by

T (f, x) = K(xλα1 · yλα2 )f (y)dy, x ∈ En (α), (8)
En (α)

where


n
En (α) = {x = (x1 , x2 , · · · , xn ) : xk ≥ 0, 1 ≤ k ≤ n, xα = ( |xk |α )1/α , α>0},
k=1

λ1 , λ2 > 0, and the corresponding new operator norm inequalities are obtained. The
discrete versions of the main results are also given. As applications, a large number
of known and new results have been obtained by proper choice of kernel. They are
significant improvements and generalizations of many famous results. We note that
En (α) is a n− dimensional vector space, when 1 ≤ α < ∞, En (α) is a normed
vector space. In particular, En (2) is a n− dimensional Euclidean space Rn+ . Hence,
when α = 2, (8) reduces to (7). When n = 1,(7) reduces to
 ∞
T2 (f, x) = K(x λ1 y λ2 )f (y)dy. (9)
0

If λ1 = λ2 = 1, K(xy) = (1 + xy)−β , then


 ∞ 1
T3 (f, x) = x β f (y)dy (10)
0 (1 + xy)β

is called the generalized Stieltjes transform of f . If K(xy) = Jα (xy)(xy)1/2 , then


 ∞
T4 (f, x) = Jα (xy)(xy)1/2 f (y)dy (11)
0

is called the Hankel transform, where Jα (t) is the Bessel function of the first kind
of order α, that is,

t  (−1)k t
Jα (t) = ( )α ( )2k (see [4]).
2 k!Γ (α + k + 1) 2
k=0
100 J. C. Kuang

We use the standard notations:



f p,ω = ( |f (x)|p ω(x)dx)1/p ,
En (α)

Lp (ω) = {f : f is measurable, andf p,ω < ∞},

where, ω is a non-negative measurable function on En (α). If ω(x) ≡ 1, we will


denote Lp (ω) by Lp (En (α)), and f p,1 by f p .
 1
B(u, v) = x u−1 (1 − x)v−1 dx (u, v > 0)
0

is the Beta function and



 1
ζ (p, β) = , (p > 1, β > 0)
(k + β)p
k=0


is the extended Riemann zeta function. In particular, ζ (p, 1) = ζ (p) = 1
k=1 k p is
the Riemann zeta function.

2 Main Results

Our main results reads as follows.


Theorem 3 Let 1 < p, q < ∞, p1 + 1
q ≥ 1,0 < λ
n = 2 − 1
p − 1
q,
λ
n( λ1 (p−1)−1)
ω(x) = xα 2 , x ∈ En (α), and the radial kernel K(xλα1 · yλα2 ) be
a nonnegative measurable function defined on En (α) × En (α), λ1 , λ2 > 0. (i) If
 ∞
(Γ (1/α))n np
(1− q1 ) n2
(p−1)(1− q1 )−1
c1 = {K(u)} λ u λλ2 du < ∞, (12)
α n−1 Γ (n/α) 0
 ∞
(Γ (1/α))n n n
[1− λn (1− q1 )]−1
c2 = {K(u)} λ u λ2 du < ∞, (13)
α n−1 Γ (n/α) 0

then the integral operator T is defined by (8):T : Lp (En (α)) → Lp (ω) exists as a
bounded operator and
c1 1/p c2 1−(1/p)
Tf p,ω ≤ ( ) ( ) f p ,
λ1 λ2
Norm Inequalities for Generalized Laplace Transforms 101

This implies that

Tf p,ω c1 c2
T  = sup ≤ ( )1/p ( )1−(1/p) . (14)
f =0 f p λ1 λ2

(ii) If 0 < λ1 ≤ λ2 , and


 ∞
(Γ (1/α))n n
(1− p1 )−1
c3 = {K(u)}u λ2 du < ∞, (15)
α n−1 Γ (n/α) 0

then
c3
T  ≥ . (16)
λ2

In the conjugate case: λ = n, p1 + 1


q = 1, then by (12), (13) and (15), we get
 ∞
(Γ (1/α))n n
−1
c1 = c2 = c3 = c0 = K(u)u qλ2 du. (17)
α n−1 Γ (n/α) 0

Thus, by Theorem 3, we get


λ
n( λ1 (p−1)−1)
Corollary 1 Let 1 < p < ∞, p1 + q1 = 1, ω(x) = xα 2
, x ∈ En (α),and
the radial kernel K(xλα1· yλα2 ) be a nonnegative measurable function defined
on En (α) × En (α), λ1 , λ2 > 0. If
 ∞
(Γ (1/α))n n
−1
c0 = K(u)u qλ2 du < ∞, (18)
α n−1 Γ (n/α) 0

then the integral operator T is defined by (8): T : Lp (En (α)) → Lp (ω) exists as a
bounded operator and
c0
Tf p,ω ≤ 1/p 1/q
f p .
λ1 λ2

This implies that

Tf p,ω c0
T  = sup ≤ 1/p 1/q .
f =0 f p λ1 λ2

If 0 < λ1 ≤ λ2 , then
c0
T  ≥ .
λ2
102 J. C. Kuang

Corollary 2 Under the same conditions as those of Corollary 1, if λ1 = λ2 = λ0 ,


then

Tf p,ω ≤ cf p

where
c0
c = T  = (19)
λ0

is the best possible.


In particular, for λ0 = n = 1, we get the following
Corollary 3 Under the same conditions as those of Corollary 2, if λ0 = n = 1,
then the integral operator T0 is defined by (4):T0 : Lp (0, ∞) → Lp (ω) exists as a
bounded operator and

T0 f p,ω ≤ cf p ,

where ω(x) = x p−2 and


 ∞  ∞
c = T0  = K(u)u(1/q)−1 du = K(u)u−(1/p) du
0 0

is the best possible.


Hence, Corollary 3 reduces to Theorem 2.

3 Proof of Theorem 3

We require the following Lemmas to prove our results:


Lemma 1 ([7]) If ak , bk , pk > 0, 1 ≤ k ≤ n,f is a measurable function on (0, ∞),
then

n
xk  p −1 p −1
f ( )bk x1 1 · · · xn n dx1 · · · dxn
B(r1 ,r2 ) ak
k=1
,n ,n pk  r
pk
k=1 ak k=1 Γ ( bk ) 2 pk
( n −1)
= ,n × n pk f (t)t k=1 bk dt,
k=1 bk Γ ( k=1 bk ) r1

where B(r1 , r2 ) = {x ∈ En (α) : 0 ≤ r1 ≤ xα < r2 }.


Norm Inequalities for Generalized Laplace Transforms 103

We get the following Lemma 2 by taking ak = 1, bk = α > 0, pk = 1, 1 ≤ k ≤ n,


r1 = 0, r2 = ∞, in Lemma 1.
Lemma 2 Let f be a measurable function on (0, ∞), then
  ∞
(Γ (1/α))n
f (xαα )dx = f (t)t (n/α)−1 dt. (20)
En (α) α n Γ (n/α) 0

Lemma 3 Let f ∈ Lp (ω), g ∈ Lq (E),1 < p < ∞, p1 + q1 = 1, ω be a nonnegative


measurable function on E, then

f p,ω = sup{| f gω1/p dμ| : gq ≤ 1}. (21)
E

Proof of Lemma 3 This is an immediate consequence of the Hölder inequality with


weight (see [8]).
p q
Proof of Theorem 3 Let p1 = p−1 , q1 = q−1 , it follows that

1 1 λ p λ
+ + (1 − ) = 1, + p(1 − ) = 1.
p1 q1 n q1 n

By Hölder’s inequality, we get



T (f, x) = K(xλα1 · yλα2 )f (y)dy
En (α)
 2
( pn λ ) n
= {yα 1
[K(xλα1 · yλα2 )] λ f p (y)}1/q1
En (α)
n 2
−( λq ) n λ
×{yα 1 [K(xλα1 · yα )λ2 ] λ }1/p1 {f (y)}p(1− n ) dy
 2
( pn λ ) n
≤{ yα 1
[K(xλα1 · yλα2 )] λ |f (y)|p dy}1/q1
En (α)
 n 2
−( λq ) n p(1− λn )
×{ yα 1
[K(xλα1 · yλα2 )] λ dy}1/p1 f p
En (α)

1/q1 1/p1 p(1− λn )


= I1 × I2 × f p . (22)

In I2 , by using Lemma 2 and letting u = xλα1 t λ2 /α , and using (13) we get


104 J. C. Kuang

 n 2
−( λq ) n
I2 = yα 1
[K(xλα1 · yλα2 )] λ dy
En (α)
 ∞
(Γ (1/α))n 2
−( q nλα ) λ2 n n
= t 1 [K(xλα1 t α ] λ × t ( α )−1 dt
α n Γ (n/α) 0
λ1 n n  ∞
(Γ (1/α))n λ ( q1 λ −1) n n
[1− λn (1− q1 )]−1
= xα 2 {K(u)} λ u λ2 du
λ2 α n−1 Γ (n/α) 0
λ n
c2 1
λ ( q nλ −1)
= xα 2 1
. (23)
λ2
p
Note that p1 + q1 ≥ 1 implies that q1 ≥ 1, thus, by (22), (23) and the Minkowski’s
inequality for integrals:
   
{ ( |f (x, y)|dy)p ω(x)dx}1/p ≤ { |f (x, y)|p ω(x)dx}1/p dy, 1 ≤ p < ∞,
X Y Y X

λ1
and letting v = yλα2 t ( α ) , we conclude that
 
p/q1 p/p1 p2 (1− λn )
Tf p,ω = ( |T (f, x)| ω(x)dx) p 1/p
≤( I1 I2 f p ω(x)dx)1/p
En (α) En (α)
 λ1 np n λ1
c2 1/p1 p(1− λ ) λ2 p1 ( q1 λ −1)+n( λ2 (p−1)−1)
=( ) f p n { xα
λ2 En (α)
 n2
 p1 λ n p
× yα [K(xλα1 · yλα2 )] λ |f (y)|p dy q1 dx}1/p
En (α)
 n2
c2 1/p1 p(1− λ ) p λ
≤( ) f p n { yα 1 |f (y)|p
λ2 En (α)

 λ1 np n λ
( λ p )( q λ −1)+n( λ1 (p−1)−1)
× xα 2 1 1 2

En (α)
np  q1
×[K(xλα1 · yλα2 )] q1 λ dx p dy}1/q1
 n2
c2 p(1− λ ) p λ
= ( )1/p1 f p n { yα 1 |f (y)|p
λ2 En (α)
 ∞ λ np n
 (Γ (1/α)) n
( 1 )(
λ
−1)+ αn ( λ1 (p−1)−1)
× n t p1 αλ2 q1 λ 2
α Γ (n/α) 0
np n  q1
×[K(t (λ1 /α) yλα2 )] q1 λ t α −1 dt p dy}1/q1
c2 1/p1 p(1− λ ) c1 p/q c2 c1
=( ) f p n × ( )1/p f p 1 = ( )1/p1 ( )1/p f p .
λ2 λ1 λ2 λ1
Norm Inequalities for Generalized Laplace Transforms 105

Thus,
c1 1/p c2 1−(1/p)
Tf p,ω ≤ ( ) ( ) f p . (24)
λ1 λ2

This implies that

Tf p,ω c1 c2
T  = sup ≤ ( )1/p ( )1−(1/p) . (25)
f =0 f p λ 1 λ 2

To prove the reversed inequality (16), setting fε and gε as follows:

fε (x) = x−(n/p)+ε
α ϕB (x), (26)
λ
α n−1 Γ (n/α) 1/p1 −( Pn )+( λ1 −p)ε
gε (x) = (pε)1/p1 { n
} xα 1 2
ϕB c (x), (27)
(Γ (1/α))

where 0 < ε < pα , B = B(0, 1) = {x ∈ En (α) : xα < 1}, ϕB c is the


characteristic function of the set B c = {x ∈ En (α) : xα ≥ 1}, that is,

1, x ∈ B c
ϕ (x) =
Bc
0, x ∈ B.

Thus, we get

 (Γ (1/α))n 1/p
fε p = , (28)
pεα n−1 Γ (n/α)

p p−1
gε p11 = λ1
≤ 1. (29)
p− λ2

Using Lemma 3 and the Fubini Theorem, we get



Tfε p,ω ≥ T (fε , x)gε (x){ω(x)}1/p dx
En (α)
  λ
n( λ1 (p−1)−1)
= K(xλα1 · yλα2 )fε (y)gε (x)(xα 2
)1/p dydx
En (α) En (α)

α n−1 Γ (n/α) 1/p1


= (pε)1/p1 { }
(Γ (1/α))n
  λ1 n
( p +ε)−pε−n
· yλα2 )y−(n/p)+ε
λ
× { K(xλα1 α dy}xα2 1
dx. (30)
Bc B
106 J. C. Kuang

λ2
Letting u = t α xλα1 , and using (20), we have

K(xλα1 · yλα2 )y−(n/p)+ε
α dy
B
 1
(Γ (1/α))n n
−( pα )+ αε + αn −1
= K(t λ2 /α xλα1 )t dt
α n Γ (n/α) 0
λ  λ
xα1
(Γ (1/α))n − λ1 ( pn +ε) 1
( pn +ε)−1
= n−1 xα 2 1 K(u)u λ2 1 du. (31)
α Γ (n/α)λ2 0

We insert (31) into (30) and use Fubini’s theorem to obtain

(Γ (1/α))n 1/p 1
Tfε p,ω ≥ (pε)1/p1 { } ×
α n−1 Γ (n/α) λ2
  xλα1 1
−pε−n ( n +ε)−1
× xα ( K(u)u λ2 p1 du)dx
Bc 0
(Γ (1/α))n 1/p 1
≥ (pε)1/p1 { } ×
α n−1 Γ (n/α) λ2
 ∞ 1
 ∞
( n +ε)−1
× K(u)u λ2 p1 ( xα−pε−n dx)du
0 β(u)

(Γ (1/α))n (1/p)+1 1
= (pε)1/p1 { n−1 } ×
α Γ (n/α) αλ2
 ∞ 1
 ∞
( n +ε)−1
× K(u)u λ2 p1 ( t −(pε)/α−1 dt)du
0 β(u)

(Γ (1/α))n 1
= (pε)−(1/p) { }(1/p)+1 ×
α n−1 Γ (n/α) λ2
 ∞ 1
( pn +ε)−1
× K(u)u λ2 1 (β(u))−(pε)/α du, (32)
0

where β(u) = max{1, u1/λ1 }. Thus, we get

Tf p,ω Tfε p,ω


T  = sup ≥
f =0 f p fε p
 ∞
(Γ (1/α))n 1
( n +ε)−1
≥ n−1 K(u)u λ2 p1 (β(u))−(pε)/α du. (33)
α Γ (n/α)λ2 0

By letting ε → 0+ in (33) and using the Fatou lemma, we get


Norm Inequalities for Generalized Laplace Transforms 107

 ∞
(Γ (1/α))n n
−1 c3
T  ≥ n−1 K(u)u p1 λ2 du = . (34)
α Γ (n/α)λ2 0 λ2

The proof is complete.

4 The Discrete Versions of the Main Results

Let a = {am } be a sequence of real numbers, we define




ap,ω = { |am |p ω(m)}1/p , l p (ω) = {a = {am } : ap,ω < ∞}.
m=1

For m − 1 ≤ x < m, n − 1 ≤ y < n, let

f (x) = am , K(x λ1 y λ2 ) = K(mλ1 nλ2 ),

then the corresponding series form of (9) is




T5 (a, m) = K(mλ1 nλ2 )an . (35)
n=1

By Theorem 3, we get
Theorem 4 Let 1 < p, q < ∞, p1 + 1
q ≥ 1,0 < λ = 2 − 1
p − 1
q,
λ1
(p−1)−1
ω(m) = m λ2 ,K(u) be a nonnegative measurable function defined on
(0, ∞),λ1 , λ2 > 0. If
 ∞ 1
p
(1− q1 ) (p−1)(1− q1 )−1
c1 = {K(u)} λ u λλ2 du < ∞, (36)
0
 ∞ 1
1 [1− λ1 (1− q1 )]−1
c2 = {K(u)} λ u λ2 du < ∞, (37)
0

then the integral operator T5 is defined by (35): T5 : l p → l p (ω) exists as a bounded


operator and
c1 1/p c2 1−(1/p)
T5 ap,ω ≤ ( ) ( ) ap .
λ1 λ2

This implies that

T5 ap,ω c1 c2
T5  = sup ≤ ( )1/p ( )1−(1/p) . (38)
a =0 a p λ 1 λ 2
108 J. C. Kuang

If 0 < λ1 ≤ λ2 , and
 ∞ 1
(1− p1 )−1
c3 = K(u)u λ2 du < ∞, (39)
0

then
c3
T5  ≥ . (40)
λ2

Corollary 4 Under the same conditions as those of Theorem 4, if λ = 1, p1 + q1 = 1,


∞ 1
−1
and c0 = 0 K(u)u qλ2 du < ∞, then

c0
T5  ≤ 1/p 1/q
.
λ1 λ2

If 0 < λ1 ≤ λ2 , then
c0
T5  ≥ .
λ2

Corollary 5 Under the same conditions as those of Corollary 4, if λ1 = λ2 = λ0 ,


then

T5 ap,ω ≤ c3 ap ,

where ω(m) = m(p−2) , and


c0
c3 = T5  = (41)
λ0

is the best possible.


In particular, for λ0 = 1, the corresponding series form of (35) is


T6 (a, m) = K(mn)an . (42)
n=1

We get the following


Corollary 6 Under the same conditions as those of Corollary 5, if λ0 = 1, then
the integral operator T6 is defined by (42):T6 : l p → l p (ω) exists as a bounded
operator and

T6 ap,ω ≤ cap , (43)


Norm Inequalities for Generalized Laplace Transforms 109

where ω(m) = mp−2 and


 ∞
c = T6  = K(u)u(1/q)−1 du
0

is the best possible.

5 Some Applications

As applications,a large number of known and new results have been obtained by
proper choice of kernel K. In this section we present some model and interesting
applications which display the importance of our results. Also these examples are of
fundamental importance in analysis. In what follows, without loss of generality,we
may assume 0 < λ1 ≤ λ2 , thus under the same conditions as those of Theorem 3,
we have
c3 c1 c2
≤ T  ≤ ( )1/p ( )1−(1/p) . (44)
λ2 λ1 λ2
In the conjugate case (λ = n), we get
c0 c0
≤ T  ≤ 1/q
, (45)
λ2 1/p
λ1 λ2

If λ1 = λ2 = λ0 , then
c0
T  = ,
λ0
where the constants c1 , c2 , c3 and c0 are defined by (12), (13), (15) and (18),
respectively.
Example 1 If K(xλα1 · yλα2 ) = exp{−(|xλα1 · yλα2 )β }, β > 0 in Theorem 3,
then the operator T7 is defined by

T7 (f, x) = exp{−(|xλα1 · yλα2 )β }f (y)dy. (46)
En (α)

Thus, in (44) and (45),


 ∞
(Γ (1/α))n np
(1− q1 ) n2
(p−1)(1− q1 )−1
c1 = n−1 {exp(−uβ )} λ u λλ2 du
α Γ (n/α) 0
(Γ (1/α))n  qλ  n2 (p−1)(1− 1 )
= × βλλ2 q
βα n−1 Γ (n/α) np(q − 1)
 n2 1 
×Γ (p − 1)(1 − ) . (47)
βλλ2 q
110 J. C. Kuang

 ∞
(Γ (1/α))n n n
[1− λn (1− q1 )]−1
c2 = n−1 {exp(−uβ )} λ u λ2 du
α Γ (n/α) 0
(Γ (1/α))n λ βλn (1− λn (1− q1 ))  n n 1 
= n−1
( ) 2 Γ (1 − (1 − )) . (48)
βα Γ (n/α) n βλ2 λ q

 ∞
(Γ (1/α))n n
(1− p1 )−1
c3 = {exp(−uβ )}u λ2 du
α n−1 Γ (n/α) 0
(Γ (1/α))n n 1
= n−1
Γ( (1 − )). (49)
βα Γ (n/α) βλ 2 p

(Γ (1/α))n n
c0 = n−1
Γ( ). (50)
βα Γ (n/α) βλ2 q

In particular, if λ1 = λ2 = λ0 , then by Corollary 2, we get

(Γ (1/α))n n
T7  = n−1
Γ( ). (51)
λ0 βα Γ (n/α) λ0 βq

If λ0 = β = 1, α = 2, then T7 reduces to the n− dimensional Laplace transform of


f , that is,

T7 (f, x) = exp{−(x · y)}f (y)dy, (52)
Rn+

thus by (51), we get

π n/2 n
T7  = n−1
Γ ( ). (53)
2 Γ (n/2) q

If n = λ0 = β = 1,then T7 reduces to the one-sided Laplace transform of f , thus


by (53), we have
 ∞
1 1
−1 −u
T7  = Γ ( ) = uq e du.
q 0

That is, it reduces to Theorem 2.


λ λ
| log(xα1 ·yα2 )|β1
Example 2 If K(xλα1 · yλα2 ) = λ λ (β1 > 0, β2 > n
qλ2 ) in
|(xα1 ·yα2 )β2 −1|
Theorem 3, then the operator T8 is defined by

| log(xλα1 · yλα2 )|β1
T8 (f, x) = f (y)dy. (54)
En (α) |(xλα1 · yλα2 )β2 − 1|
Norm Inequalities for Generalized Laplace Transforms 111

Thus, in (44) and (45),


 ∞
(Γ (1/α))n | log u|β1 npλ (1− q1 ) λλn2 (p−1)(1− q1 )−1
c1 = { } u 2 du;
α n−1 Γ (n/α) 0 |uβ2 − 1|
 ∞
(Γ (1/α))n | log u|β1 n λn [1− λn (1− q1 )]−1
c2 = n−1 { }λ u 2 du;
α Γ (n/α) 0 |uβ2 − 1|
 ∞
(Γ (1/α))n | log u|β1 λn (1− p1 )−1
c3 = { }u 2 du.
α n−1 Γ (n/α) 0 |uβ2 − 1|

 ∞
(Γ (1/α))n | log u|β1 qλn −1
c0 = n−1 { }u 2 du
α Γ (n/α) 0 |uβ2 − 1|
(Γ (1/α))n Γ (β1 + 1) n n
= β +1
{ζ (β1 + 1, ) + ζ (β1 + 1, 1 − )}.
α n−1 β2 1 Γ (n/α) qλ2 β2 qλ2 β2

In particular, if λ1 = λ2 = λ0 , then by Corollary 2, we get

(Γ (1/α))n Γ (β1 + 1) n n
T8  = β +1
{ζ (β1 +1, )+ζ (β1 +1, 1− )}. (55)
λ0 α n−1 β2 1 Γ (n/α) qλ0 β2 qλ0 β2

λ λ
log(xα1 ·yα2 )
Example 3 If K(xλα1 · yλα2 ) = λ λ (β > 0) in Theorem 3, then the
(xα1 ·yα2 )β −1
operator T9 is defined by

log(xλα1 · yλα2 )
T9 (f, x) = f (y)dy. (56)
En (α) (xλα1 · yλα2 )β − 1

By (18), we have

(Γ (1/α))n  π 2
c0 = . (57)
β 2 α n−1 Γ (n/α) sin( λnπ
2 βq
)

In particular, if λ1 = λ2 = λ0 , then by Corollary 2, we get

(Γ (1/α))n  π 2
T9  = . (58)
λ0 β 2 α n−1 Γ (n/α) sin( λnπ
0 βq
)

λ λ
cos(β1 xα1 ·yα2 ) (1− qλn )
Example 4 If K(xλα1 ·yλα2 ) = λ λ ×(xλα1 ·yλα2 ) 2 β1 , β2 >
(xα1 ·yα2 )2 +β22
0 in Theorem 3, then the operator T10 is defined by
112 J. C. Kuang


cos(β1 xλα1 · yλα2 ) (1− qλn )
T10 (f, x) = λ
× (xλα1 · yλα2 ) 2 f (y)dy.
En (α) (xα1 · yλα2 )2 + β22
(59)
By (18), we have
 ∞
(Γ (1/α))n cos(β1 u) π {Γ (1/α)}n −(β1 β2 )
c0 = n−1 du = e . (60)
α Γ (n/α) 0 u2 + β22 2β2 α n−1 Γ (n/α)

In particular, if λ1 = λ2 = λ0 , then by Corollary 2, we get

π(Γ (1/α))n
T10  = e−(β1 β2 ) . (61)
2λ0 β2 α n−1 Γ (n/α)

Example 5 If K(xλα1 · yλα2 ) = (2/π )n/α sin(βxλα1 · yλα2 ) (β > 0, λ1 >


0, λ2 > n/q) in Theorem 3, then the operator T11 is defined by

T11 (f, x) = (2/π ) n/α
sin(βxλα1 · yλα2 )f (y)dy. (62)
En (α)

By (18), we have
 ∞
(Γ (1/α))n 2 n/α sin(βu)
c0 = n−1 ( ) 1− qλn
du
α Γ (n/α) π 0 u 2
n
−( )
(2/π )(n/α)−1 β qλ2 (Γ (1/α))n
= n−1 nπ . (63)
α Γ (n/α)Γ (1 − qλn 2 ) cos( 2qλ2
)

In particular, if λ1 = λ2 = λ0 , then by Corollary 2, we get


n
−( )
β qλ0 (2/π )(n/α)−1 (Γ (1/α))n
T11  = nπ . (64)
λ0 α n−1 Γ (n/α)Γ (1 − qλn 0 ) cos( 2qλ0
)

When α = 2, β = λ0 = 1, T11 is called the n− dimensional Fourier sine transform


of f , then by (64), we get
π
T11  = .
2n/2 Γ (n/2)Γ (1 − qn ) cos( nπ
2q )

If n = 1, then T11 reduces to the Fourier sine transform of f in [3]:


 ∞
T11 (f, x) = (2/π ) 1/2
sin(xy)f (y)dy.
0
Norm Inequalities for Generalized Laplace Transforms 113

Thus, we have

(π/2)1/2
T11  = .
Γ ( p1 ) cos( 2q
π
)

Example 6 If K(xλα1 ·yλα2 ) = ( π2 )n/α cos(βxλα1 ·yλα2 ) (β > 0, λ1 > 0, λ2 >


n/q) in Theorem 3, then the operator T12 is defined by

2
T12 (f, x) = ( )n/α cos(βxλα1 · yλα2 )f (y)dy. (65)
π En (α)

By (18), we have
 ∞
(Γ (1/α))n 2 n/α cos(βu)
c0 = ( ) 1− qλn
du
α n−1 Γ (n/α) π 0 u 2
n
−( )
(2/π )(n/α)−1 β qλ2 (Γ (1/α))n
= n−1 nπ . (66)
α Γ (n/α)Γ (1 − qλn 2 ) sin( 2qλ2
)

In particular, if λ1 = λ2 = λ0 , then by Corollary 2, we get


n
−( )
(2/π )(n/α)−1 β qλ0 (Γ (1/α))n
T12  = nπ . (67)
λ0 α n−1 Γ (n/α)Γ (1 − qλn 0 ) sin( 2qλ0
)

When α = 2, β = λ0 = 1, T12 is called the n− dimensional Fourier cosine


transform of f , then by (67), we get
π
T12  = .
2n/2 Γ (n/2)Γ (1 − qn ) sin( nπ
2q )

If n = 1, then T12 reduces to the Fourier cosine transform of f in [3]:


 ∞
2
T12 (f, x) = ( )1/2 cos(xy)f (y)dy.
π 0

Thus, we have

(π/2)1/2
T12  = .
Γ ( p1 ) sin( 2q
π
)

Remark 2 Because the Fourier transform of f can be decomposed into the Fourier
sine and cosine transforms of f , thus, the corresponding operator norm can be
derived from Examples 5 and 6.
114 J. C. Kuang

Example 7 If K(xλα1 · yλα2 ) = λ


1
λ in Theorem 3, then the operator
1+(|xα1 ·yα2 )β
T13 is defined by

1
T13 (f, x) = λ
f (y)dy. (68)
En (α) 1 + (|xα1 · yλα2 )β
q p
If ( βλλ 2 − 1) q−1 < n
βλ2 < p−1 , and λ > n(1 − q1 ), then in (44) and (45),
 ∞
(Γ (1/α))n 1 np n2
(1− q1 ) λλ (p−1)(1− q1 )−1
c1 = { }λ u 2 du
α n−1 Γ (n/α) 0 1+uβ

(Γ (1/α))n
=
βα n−1 Γ (n/α)
 n2 1 n 1 n 
×B (p − 1)(1 − ), (1 − )(p − (p − 1)) .
βλλ2 q λ q βλ2

 ∞
(Γ (1/α))n 1 n n
[1− λn (1− q1 )]−1
c2 = n−1 { } λ u λ2 du
α Γ (n/α) 0 1+uβ

(Γ (1/α))n  n n 1 n n n 1 
= n−1
×B (1 − (1 − )), − (1 − (1 − )) ,
βα Γ (n/α) βλ2 λ q λ βλ2 λ q

 ∞
(Γ (1/α))n 1 n
(1− p1 )−1
c3 = u λ2 du
α n−1 Γ (n/α) 0 1+uβ

(Γ (1/α))n π
= × .
βα Γ (n/α) sin( βλ (1 − p1 ))
n−1 nπ
2

In the conjugate case (λ = n), if β > n


qλ2 , then

(Γ (1/α))n π
c0 = × .
βα n−1 Γ (n/α) sin( λnπ
2 βq
)

In particular,if λ1 = λ2 = λ0 , then by Corollary 2, we get

(Γ (1/α))n π
T13  = × . (69)
λ0 βα n−1 Γ (n/α) sin( λnπ
0 βq
)

Example 8 If K(xλα1 ·yλα2 ) = λ


1
λ in Theorem 3, then the operator
{1+(xα1 ·yα2 )}β
T14 is defined by
Norm Inequalities for Generalized Laplace Transforms 115


1
T14 (f, x) = λ
f (y)dy. (70)
En (α) {1 + (xα1 · yλα2 )}β
q p
If ( βλλ 2 − 1) q−1 < n
βλ2 < p−1 and λ > n(1 − q1 ), then in (44) and (45),
 ∞
(Γ (1/α))n 1 np 1 n2
λ (1− q ) u λλ2
(p−1)(1− q1 )−1
c1 = { } du
α n−1 Γ (n/α) 0 (1 + u) β

(Γ (1/α))n  n2 1 n 1 n 
= ×B (p − 1)(1 − ), (1 − )(pβ − (p − 1)) .
α n−1 Γ (n/α) λλ2 q λ q λ2

 ∞
(Γ (1/α))n 1 n n
[1− λn (1− q1 )]−1
c2 = { } λ u λ2 du
α n−1 Γ (n/α) 0 (1 + u) β

(Γ (1/α))n n n 1 βn n n 1 
= ×B (1 − (1 − )), − (1 − (1 − )) .
α n−1 Γ (n/α) λ2 λ q λ λ2 λ q

 ∞
(Γ (1/α))n 1 n 1
λ2 (1− p )−1 du
c3 = n−1 { }u
α Γ (n/α) 0 (1 + u)β
(Γ (1/α))n  n 1 n 1 
= n−1 B (1 − ), β − (1 − )
α Γ (n/α) λ2 p λ2 p

In the conjugate case (λ = n), if β > n


qλ2 , then

(Γ (1/α))n n n
c0 = B( ,β − ).
α n−1 Γ (n/α) λ2 q λ2 q

In particular, if λ1 = λ2 = λ0 , then by Corollary 2, we get

(Γ (1/α))n n n
T14  = n−1
B( ,β − ). (71)
λ0 α Γ (n/α) λ0 q λ0 q

If n = 1, λ1 = λ2 = 1, then by the generalized Stieltjes transform (10), we have


 ∞ 1
T3 (f, x) = x β
f (y)dy = x β T14 (f, x).
0 (1 + xy)β

If β > 1/q, then by (71), we get

1 1
T3  = B( , β − ),
q q

where ω(x) = x p(β+1)−2 .


116 J. C. Kuang

Example 9 If K(xλα1 · yλα2 ) = λ


1
λ in Theorem 3, then the operator
|1−xα1 ·yα2 |β
T15 is defined by

1
T15 (f, x) = λ
f (y)dy. (72)
En (α) |1 − xα1 · yλα2 |β
q p
If ( βλλ 2 − 1) q−1 < n
βλ2 < p−1 ,and λ > pβn(1 − q1 ), then in (44) and (45),
 ∞
(Γ (1/α))n 1 pn n2
(1− q1 ) λλ (p−1)(1− q1 )−1
c1 = { }λ u 2 du
α n−1 Γ (n/α) 0 |1 − u| β

(Γ (1/α))n  n2 1 pβn 1 
= n−1 {B (p − 1)(1 − ), 1 − (1 − )
α Γ (n/α) λλ2 q λ q
n 1 n pβn 1 
+B (1 − )[pβ − (p − 1)], 1 − (1 − ) }
λ q λ2 λ q

 ∞
(Γ (1/α))n 1 n n
[1− λn (1− q1 )]−1
c2 = n−1 { } λ u λ2 du
α Γ (n/α) 0 |1 − u|β
(Γ (1/α))n n n 1 βn 
= n−1 {B [1 − (1 − )], 1 −
α Γ (n/α) λ2 λ q λ
 βn n n 1 βn 
+B − [1 − (1 − )], 1 − }.
λ λ2 λ q λ

 ∞
(Γ (1/α))n 1 n
(1− p1 )−1
c3 = n−1 { }u λ2 du
α Γ (n/α) 0 |1 − u| β

(Γ (1/α))nn 1   n 1 
= n−1 {B (1 − ), 1 − β + B β − (1 − ), 1 − β }.
α Γ (n/α) λ2 p λ2 p

In the conjugate case (λ = n), if n


qλ2 < β < 1, then

(Γ (1/α))n  n   n 
c0 = × {B ,1 − β + B β − ,1 − β }
α n−1 Γ (n/α) qλ2 qλ2

In particular,if λ1 = λ2 = λ0 , then by Corollary 2, we get

(Γ (1/α))n n n
T15  = n−1
× {B( , 1 − β) + B(β − , 1 − β)}. (73)
λ0 α Γ (n/α) qλ0 qλ0

Remark 3 Defining other forms of the kernel K, we can obtain new results of
interest.
Norm Inequalities for Generalized Laplace Transforms 117

References

1. D.V. Widder, The Laplace Transform (Princeton University Press, Princeton, 1972)
2. K.B. Wolff, Integral Transforms in Science and Engineering (Plenum, New York, 1979)
3. E. Zauderer, Partial Differential Equations of Applied Mathematics (A Wiley-Interscience
Publication, Wiley, 1983)
4. N. Bleistein, R.A. Handelsman, Asymptotic Expansions of Integrals (Dover Publications, Inc.,
New York, 1986)
5. G.H. Hardy, The constants of certain inequalities. J. Lond. Math. Soc. 8, 114–119 (1933)
6. J.C. Kuang, Generalized Laplace transform inequalities in multiple Orlicz spaces, chapter 13, in
Computation, Cryptography, and Network Security ed. by N.J. Daras, M.T. Rassias (Springer,
Berlin, 2015)
7. J.C. Kuang, Real and Functional Analysis (coutinuation), vol. 2 (Higher Education Press,
Beijing, 2015) (in Chinese)
8. J.C. Kuang, Applied Inequalities, 4th ed. (Shandong Science Technology Press, Jinan, 2010) (in
Chinese)
On Marcinkiewicz-Zygmund Inequalities
at Hermite Zeros and Their Airy
Function Cousins

D. S. Lubinsky

Abstract We * establish
+ forward and converse Marcinkiewicz-Zygmund Inequalities
at the zeros aj j ≥1 of the Airy function Ai (x), such as

∞  ∞ ∞
π 2  |f (ak )|p π 2  |f (ak )|p
A ≤ |f (t)| p
dt ≤ B
6
k=1
Ai (ak )2 −∞ 6
k=1
Ai (ak )2

under appropriate conditions on the entire function f and p. The constants A and
B are those appearing in Marcinkiewicz-Zygmund inequalities at zeros of Hermite
polynomials. Scaling limits are used to pass from the latter to the former.

Keywords Marcinkiewicz-Zygmund inequalities · Quadrature sums · Airy


functions · Hermite polynomials

1 Introduction

There is a close relationship between the Plancherel-Polya and Marcinkiewicz-


Zygmund inequalities. The former [9, p. 152] assert that for 1 < p < ∞, and
entire functions f of exponential type at most π ,

  ∞ ∞

Ap |f (k)| ≤
p
|f | ≤ Bp
p
|f (k)|p , (1)
k=−∞ −∞ j =−∞

Research supported by NSF grant DMS1800251.

D. S. Lubinsky ()
School of Mathematics, Georgia Institute of Technology, Atlanta, GA, USA
e-mail: [email protected]

© Springer Nature Switzerland AG 2020 119


A. Raigorodskii, M. T. Rassias (eds.), Trigonometric Sums and Their Applications,
https://doi.org/10.1007/978-3-030-37904-9_6
120 D. S. Lubinsky

provided either the series or integral is finite. For 0 < p ≤ 1, the left-hand inequality
is still true, but the right-hand inequality requires additional restrictions [2]. We
assume that Bp is taken as small as possible, and Ap as large as possible. The
Marcinkiewicz-Zygmund inequalities assert [35, Vol. II, p. 30] that for p > 1, n ≥
1, and polynomials P of degree ≤ n − 1,
n  
A p  p  1   p n  
Bp  p
 2π ik/n     
P e  ≤ P e2π it  dt ≤ P e2π ik/n  . (2)
n 0 n
k=1 k=1

Here too, A p and Bp are independent of n and P , and the left-hand inequality is also
true for 0 < p ≤ 1 [15]. The author [16] proved that the inequalities (1) and (2) are
equivalent, in the sense that each implies the other. Moreover, the sharp constants
are the same:
Theorem A For 0 < p < ∞, Ap = A p and for 1 < p < ∞, Bp = Bp .
These inequalities are useful in studying convergence of Fourier series, Lagrange
interpolation, in number theory, and weighted approximation. They have been
extended to many settings, and there are a great many methods to prove them
[5, 8, 13, 15, 19, 20, 22–26, 30, 31, 33, 34]. The sharp constants in (1) and (2) are
unknown, except for the case p = 2, where of course we have equality rather than
inequality, so that A2 = B2 = A 2 = B2 = 1 [9, p. 150]. It is certainly of interest to
say more about these constants.
In a recent paper, we explored the connections between Marcinkiewicz-Zygmund
inequalities at zeros of Jacobi polynomials, and Polya-Plancherel type inequalities
at zeros of Bessel functions. Let α, β > −1 and

w α,β (x) = (1 − x)α (1 + x)β , x ∈ (−1, 1) .


α,β
For n ≥ 1, let Pn denote the standard Jacobi polynomial of degree n, so that it has
degree n, satisfies the orthogonality conditions
 1
Pnα,β (x) x k w α,β (x) dx = 0, 0 ≤ k < n,
−1

α,β n+α
and is normalized by Pn (1) = n . Let

xnn < xn−1,n < · · · < x1n

α,β
denote the zeros of Pn . Let {λkn } denote the weights in the Gauss quadrature for
w α,β , so that for all polynomials P of degree ≤ 2n − 1,
 1 
n
P w α,β = λkn P (xkn ) .
−1 k=1
On Marcinkiewicz-Zygmund Inequalities at Hermite Zeros and Their Airy. . . 121

There is a classical analogue of (2), established for special α, β by Richard


Askey, and for all α, β > −1 (and for more general “generalized Jacobi weights”)
by P. Nevai, and his collaborators [15, 20, 27, 29], with later work by König and
Nielsen [8], and for doubling weights by Mastroianni and Totik [23]. The following
special case follows from Theorem 5 in [20, eqn. (1.19), p. 534]:
Theorem B Let α, β, τ, σ satisfy α, β, α + σ, β + τ > −1. Let p > 0. For n ≥
α,β
1, let {xkn } denote the zeros of the Jacobi polynomial Pn and {λkn } denote the
corresponding Gauss quadrature weights. There exists A > 0 such that for n ≥ 1,
and polynomials P of degree ≤ n − 1,


n
A λkn |P (xkn )|p (1 − xkn )σ (1 + xkn )τ
k=1
 1
≤ |P (x)|p (1 − x)α+σ (1 + x)β+τ dx. (3)
−1

The converse inequality is much more delicate, and in particular holds only for
p > 1, and even then only for special cases of the parameters. It too was investigated
by P. Nevai, with later work by Yuan Xu [33, 34], König and Nielsen [8]. König and
Nielsen gave the exact range of p for which
 1 
n
|P (x)|p (1 − x)α (1 + x)β dx ≤ B λkn |P (xkn )|p , (4)
−1 k=1

holds with B independent of n and P . Let


 
α+1 β +1
μ (α, β) = max 1, 4 ,4 ;
2α + 5 2β + 5
 
α+1 β +1
m (α, β) = max 1, 4 ,4 ;
2α + 3 2β + 3
m (α, β)
M (α, β) = . (5)
m (α, β) − 1

Then (4) holds for all n and P iff

μ (α, β) < p < M (α, β) . (6)

The most general sufficient condition for a converse quadrature inequality is due
to Yuan Xu [33, pp. 881–882]. When we restrict to Jacobi weights, with the same
weight on both sides, the inequality takes the following form:
p
Theorem C Let α, β, τ, σ satisfy α, β, α + σ, β + τ > −1. Let p > 1, q = p−1 ,
and assume that
122 D. S. Lubinsky

    
p 1 p 1
α+ − (α + 1) < σ < (p − 1) (α + 1) − max 0, α+ . (7)
2 2 2 2
    
p 1 p 1
β+ − (β + 1) < τ < (p − 1) (β + 1) − max 0, β+ . (8)
2 2 2 2

Then there exists B > 0 such that for n ≥ 1, and polynomials P of degree ≤ n − 1,
 1
|P (x)|p (1 − x)α+σ (1 + x)β+τ dx
−1


n
≤B λkn |P (xkn )|p (1 − xkn )σ (1 + xkn )τ . (9)
k=1

Inequalities of the type (9) for doubling weights have been established by
Mastroianni and Totik [23] under the additional condition that one needs to restrict
the degree of P in (9) further, such as deg (P ) ≤ ηn for some η ∈ (0, 1) depending
on the particular doubling weight.
Now let α > −1 and define the Bessel function of order α,

 z α 
∞  z 2k
Jα (z) = (−1) k 2
(10)
2 k!Γ (k + α + 1)
k=0

and

Jα∗ (z) = Jα (z) /zα , (11)

which has the advantage of being an entire function for all α > −1. Jα∗ has real
simple zeros, and we denote the positive zeros by

0 < j1 < j2 < · · ·

while for k ≥ 1,

j−k = −jk .

The connection between Jacobi polynomials and Bessel functions is given by the
classical Mehler-Heine asymptotic, which holds uniformly for z in compact subsets
of C [32, p. 192]:
 
1  z 2
lim n−α Pnα,β 1 −
n→∞ 2 n
 z   z −α
= lim n−α Pnα,β cos = Jα (z) = 2α Jα∗ (z) . (12)
n→∞ n 2
On Marcinkiewicz-Zygmund Inequalities at Hermite Zeros and Their Airy. . . 123

There is an extensive literature dealing with quadrature sums and Lagrange


interpolation at the {jk }. In particular, there is the quadrature formula [6, p. 49]
 ∞ ∞
  
2 1 jk
|x|2α+1 f (x) dx =   f ,
−∞ τ 2α+2 J ∗ (jk )2 τ
k=−∞,k =0 α

valid for all entire functions f of exponential type at most 2τ , for which the integral
on the left-hand side is finite. That same paper contains the following converse
Marcinkiewicz-Zygmund type inequality: let α ≥ − 12 and p > 1; or −1 < α < − 12
and 1 < p < |1+2α| 2
. Then for entire functions f of exponential type ≤ τ for which
1
|x|α+ 2 f (x) ∈ Lp (R\ (−δ, δ)), for some δ > 0, [6, Lemma 14, p. 58; Lemma 13,
p. 57]
 ∞   p
p B∗

 
 α+ 12   1 jk 
|x| f (x) dx ≤  f  . (13)
−∞ τ  τ α+ 2 J ∗ (j )
1
τ 
k=−∞,k =0 α k

Here B ∗ depends on α and p. In the converse direction, since jk+1 − jk is bounded


below by a positive constant for all k, classical inequalities from the theory of entire
functions [9, p. 150] show that

  ∞
|f (jk )| ≤ C
p
|f (x)|p dx
k=−∞,k =0 −∞

for entire functions of finite exponential type for which the right-hand side is finite.
While Grozev and Rahman note the analogous nature of Lagrange interpolation
at zeros of Jacobi polynomials and Bessel functions, and also the Mehler-Heine
formula, their proofs proceed purely from properties of Bessel functions. In [17,
Thms. 1.1, 1.3, pp. 227–228], the author used inequalities like (3) to pass to
analogues for Bessel functions using scaling limitsof the form (12), keeping the
same constants, much as was done in [16]: Let L1 (0, ∞) , t 2α+2σ +1 denote the
p

space of all even entire functions f of exponential type ≤ 1 with


 ∞
|f (t)|p t 2α+2σ +1 dt < ∞.
0

Theorem D Assume that p > 0, α, β, α + σ, β + τ > −1, and


 
α 5
−p + + α + σ + 1 < 0.
2 4
Let A be as in Theorem B. Then

  ∞
2A jk2σ Jα∗ (jk )−2 |f (jk )|p ≤ |f (t)|p t 2α+2σ +1 dt,
k=1 0

p 
for all f ∈ L1 (0, ∞) , t 2α+2σ +1 .
124 D. S. Lubinsky

Theorem E Assume that p > 1, α, β, α + σ, β + τ > −1, and that  (7) and (8)
hold. Let B be as in Theorem C. Then for f ∈ L1 (0, ∞), t 2α+2σ +1 , we have
p

 ∞ ∞

|f (t)|p t 2α+2σ +1 dt ≤ 2B jk2σ Jα∗ (jk )−2 |f (jk )|p . (14)
0 k=1

In particular this holds for σ = τ = 0 if p satisfies (6) with β = α. Moreover,


for any α, β, p, it is possible to choose σ and τ satisfying (7), (8) so that this last
inequality also holds.
A very recent paper of Littmann [13] provides far reaching extensions of the
inequalities of Grozev and Rahman to Hermite-Biehler weights, so that t 2α+2σ +1
is replaced by 1/ |E|p , where E is a Hermite-Biehler function, that is, an entire
function E satisfying |E (z)| > |E (z̄)| for Re z > 0. Moreover, the zeros of Bessel
functions are replaced by the zeros of B (z) = 2i E (z) − E (z̄) . Littmann then
uses these to establish weighted mean convergence of certain interpolation operators
for classes of entire functions.
In this paper, we shall use Marcinkiewicz-Zygmund inequalities at zeros of
Hermite polynomials, to derive Plancherel-Polya type inequalities at zeros of Airy
functions. We begin with our notation. Throughout,
 
1
W (x) = exp − x 2 , x ∈ R, (15)
2

is the Hermite weight, and {pn } are the orthonormal Hermite polynomials, so that
 ∞
pn pm W 2 = δmn . (16)
−∞

The classical Hermite polynomial is of course denoted by Hn . The relationship


between pn and Hn is given by [32, p. 105, (5.5.1)]

pn = π −1/4 2−n/2 (n!)−1/2 Hn . (17)

The leading coefficient of pn is [32, p. 106, (5.5.6)]

γn = π −1/4 2n/2 (n!)−1/2 . (18)


* +
In the sequel, xj n denote the zeros of the Hermite polynomials in decreasing
order:

−∞ < xnn < xn−1,n < · · · < x2n < x1n < ∞,
On Marcinkiewicz-Zygmund Inequalities at Hermite Zeros and Their Airy. . . 125

* +
while λj n denote the weights in the Gauss quadrature formula: for polynomials
P of degree ≤ 2n − 1,
 ∞ 
n
 
PW2 = λj n P xj n .
−∞ j =1

There is an extensive literature on Marcinkiewicz-Zygmund inequalities at zeros


of Hermite polynomials, as well as for orthonormal polynomials for more general
exponential weights [3, 4, 7, 14, 21, 28, 29]. We shall use the following forward and
converse inequalities [14, p. 529], [21, p. 287]:
Theorem F Let 1 ≤ p < ∞. Let r, R ∈ R and S > 0.
(a) Then there exists A > 0 such that for n ≥ 1, and polynomials P of degree at
most n + Sn1/3 ,


n
  p    Rp
λj n P xj n  W p−2 xj n 1 + xj n 
j =1
 ∞  p
 
≤A (P W ) (x) (1 + |x|)R  dx. (19)
−∞

(b) Assume that

1 1
r <1− ; r ≤ R; R>− . (20)
p p

In addition if p = 4, we assume that r < R, while if p > 4, we assume that


   
1 1 4 ≤ 0, if R =
1− 1
r − min R, 1 − + 1− p
. (21)
p 3 p < 0, if R = 1 − 1
p

Then there exists B > 0 such that for n ≥ 1, and polynomials P of degree
≤ n − 1,
 ∞
 
(P W ) (x) (1 + |x|)r p dx
−∞


n
  p    Rp
≤B λj n P xj n  W p−2 xj n 1 + xj n  . (22)
j =1

Recall that the Airy function Ai is given on the real line by [1, 10.4.32, p. 447]
 ∞  
1 1 3
Ai (x) = cos t + xt dt.
π 0 3
126 D. S. Lubinsky

The Airy function Ai is an entire function of order 32 , with only real negative zeros
* +
aj , where

0 > a1 > a2 > a3 > · · · .


* + * +
These are often denoted by ij rather than aj . Ai satisfies the differential
equation

Ai (z) − zAi (z) = 0.

The Airy kernel Ai (·, ·), much used in random matrix theory, is defined [12] by

Ai(a)Ai (b)−Ai (a)Ai(b)
a = b, ,
Ai (a, b) =
a−b
Ai (a) − aAi (a) , a = b.
2 2

Observe that
 
Ai z, aj Ai (z)
Lj (z) =  =   ,
Ai aj , aj
Ai aj z − aj

is the Airy analogue of a fundamental of Lagrange interpolation, satisfying

Lj (ak ) = δj k .

There
* + is an analogue of sampling series and Lagrange interpolation series involving
Lj :
Definition 1.1 Let G be the class of all functions g : C → C with the following
properties:
(a) g is an entire function of order at most 32 ;
(b) There exists L > 0 such that for δ ∈ (0, π ), some Cδ > 0, and all z ∈ C with
|arg z| ≤ π − δ,
  
 2 3 
|g (z)| ≤ Cδ (1 + |z|) exp − z 2  ;
L
3

(c)

∞   2
g aj
 1/2 < ∞. (23)
 
j =1 aj

In [12, Corollary 1.3, p. 429], it was shown that each g ∈ G admits the locally
uniformly convergent expansion
On Marcinkiewicz-Zygmund Inequalities at Hermite Zeros and Their Airy. . . 127

∞   ∞
   Ai z, aj   
g (z) = g aj  = g aj Lj (z) .
j =1
Ai aj , aj j =1

We let


M
 
SM [g] = g aj Lj , M ≥ 1, (24)
j =1

denote the Mth partial sum of this expansion. Moreover, for f, g ∈ G, there is
the quadrature formula [12, Corollary 1.4, p. 429]
 ∞  
∞  (fg) aj
f (x) g (x) dx =  .
−∞ j =1
Ai aj , aj

In particular,
 ∞ ∞   2
g aj
g (x) dx =
2  ,
−∞ j =1
Ai aj , aj

and the series on the right converges because of (23), and the fact that
   2
Ai aj , aj = Ai aj grows like j 1/3 – see Lemma 2.2.
Lagrange interpolation at zeros of Airy functions was considered in [18]. We
shall need a class of functions that are limits in Lp of the partial sums of the Airy
series expansion:
Definition 1.2 Let 0 < p < ∞ and f ∈ Lp (R). We write f ∈ Gp if

lim f − SM [f ]Lp (R) = 0.


M→∞

The relationship between Hermite polynomials and Airy functions lies in the
asymptotic [32, p. 201],

e−x Hn (x) = 31/3 π −3/4 2n/2+1/4 (n!)1/2 n−1/12 {Ai (−t) + o (1)}
2 /2
(25)

as n → ∞, uniformly for

x= 2n(1 − 6−1/3 (2n)−2/3 t), (26)

and t in compact subsets of C. This follows from the formulation in [32] because
of the uniformity. Using this and part (a) of Theorem F with R = r = 0, we shall
prove:
128 D. S. Lubinsky

Theorem 1.3 Let p ≥ 1. Let A be the constant in (19) with R = r = 0 there.


(a) Then for f ∈ Gp , we have


  ∞
|f (ak )|p 6
≤A 2 |f (t)|p dt. (27)

Ai (ak ) 2 π −∞
k=1

(b) In particular, if p ≥ 2, f ∈ G and for some C > 0, β > 14 , we have

|f (x)| ≤ C (1 + |x|)−β , x ∈ R, (28)

then (27) is true.


Remark We expect that (27) also holds for 0 < p < 1, but this would require (19)
for such p, and that does not seem to appear in the literature.
Using part (b) of Theorem F, we shall prove:
Theorem 1.4 Let 1 < p < 4. Let B be the constant in (22) with R = r = 0 there.
(a) For f ∈ Gp , we have
 ∞ ∞

6 |f (ak )|p
|f (t)|p dt ≤ B . (29)
π2 −∞ k=1
Ai (ak )2

(b) In particular, if f ∈ G and



 |f (ak )|p
< ∞, (30)
k 1/3
k=1

then (29) is true.


In the sequel, C, C1 , C2 , . . . denote constants independent of n, z, x, t, and
polynomials of degree ≤ n. The same symbol does not necessarily denote the same
constant in different occurrences. [x] denotes the greatest integer ≤ x. Given two
sequences {xn }, {yn } of non-zeros real numbers, we write

xn ∼ yn

if there exist constants C1 and C2 such that

C1 ≤ xn /yn ≤ C2

for n ≥ 1. Similar notation is used for functions and sequences of functions. We


establish some basic estimates and then prove Theorems 1.3 and 1.4 in Section 2.
On Marcinkiewicz-Zygmund Inequalities at Hermite Zeros and Their Airy. . . 129

2 Proof of Theorems 1.3 and 1.4

We start with properties of Hermite polynomials. Throughout {pn } denote the


orthonormal
* Hermite
+ polynomials satisfying (16), with leading coefficient γn , and
with zeros xj n . In the sequel, we let
 
 |x| 

ψn (x) = 1 − √  + n−2/3 .
2n

We also let


n−1
Kn (x, y) = pj (x) pj (y)
j =0

denote the nth reproducing kernel, and

λn (x) = 1/Kn (x, x)


 
denote the nth Christoffel function. In particular, λj n = λn xj n . The j th
fundamental polynomial at the zeros of pn (x) is

pn (x)
j n (x) =   .
pn xj n x − xj n

It also admits the identity


 
j n (x) = λj n Kn x, xj n . (31)

Lemma 2.1
(a)
-
γn−1 n
= . (32)
γn 2

(b) For each fixed j , as n → ∞,


√ *  +
xj n = 2n(1 − 6−1/3 (2n)2/3 aj  + o (1) ). (33)

(c) Uniformly for t in compact subsets of C, and for


√  
x= 2n 1 − 6−1/3 (2n)−2/3 t , (34)
130 D. S. Lubinsky

we have

(pn W ) (x) = 31/3 π −1 21/4 n−1/12 {Ai (−t) + o (1)} . (35)

(d) For each fixed j , as n → ∞,


   *   +
pn W xj n = 32/3 π −1 23/4 n1/12 Ai aj + o (1) . (36)

(e) For each fixed j , and uniformly for t in compact subsets of C, and x of the
form (34)
   
lim j n W (x) W −1 xj n = Lj (−t) . (37)
n→∞

(f) For all 1 ≤ j ≤ n and all x ∈ R,


 1/4
    ψn (x) 1
j n W  (x) W −1 xj n ≤ C    .
ψn xj n 1 + n1/2 ψn (x)1/2 x − xj n 
(38)
(g) In particular for fixed j , and n ≥ n0 (j ) and all x ∈ R,

    n1/6 ψn (x)1/4
j n W  (x) W −1 xj n ≤ C  √  . (39)

1 + n1/2 ψn (x)1/2 x − 2n

(h) For each fixed j ,


   2
λ−1
j n W xj n = 3
2 4/3 −2 3/2 1/6
π 2 n Ai aj (1 + o (1)) . (40)

Proof
(a) This follows from (18).
(b) See [32, p. 132, (6.32.5)]. We note that Szego uses Ai (−x) as the Airy function,
 
so there zeros are positive there. Moreover there the symbol ij is used for aj .
(c) This follows from (25) and (17).
(d) Because of the uniform convergence, we can differentiate the relation (35):
uniformly for t in compact sets,

* + dx * +
W (x) −xpn (x) + pn (x) = 31/3 π −1 21/4 n−1/12 −Ai (−t) + o (1)
dt
so setting x = xj n and using (34), we obtain (36).
(e) From (33–36),

    (pn W ) (x)
j n W (x) W −1 xj n =      
pn W xj n x − xj n
On Marcinkiewicz-Zygmund Inequalities at Hermite Zeros and Their Airy. . . 131

31/3 π −1 21/4 n−1/12 {Ai (−t) +o (1)}


= *   +    
32/3 π −1 23/4 n1/12 Ai aj +o (1) −6−1/3 (2n+1)−1/6 t− aj  +o (1)
Ai (−t)
=    (1 + o (1)) = Lj (−t) + o (1) .
Ai aj −t − aj

(f) We note the following estimates [10, p. 465–467]: uniformly for n ≥ 1 and
x ∈ R,

n1/4 |pn (x)| W (x) ≤ Cψn (x)−1/4 . (41)

Note that for the Hermite weight, the Mhaskar-Rakhmanov


. √ number is an =
√ √ /
2n. We have uniformly for n ≥ 1 and x ∈ − 2n, 2n ,

W 2 (x)
λn (x) ∼ √ ψn (x)−1/2 , (42)
n

while for all x ∈ (−∞, ∞),

W 2 (x)
λn (x) ≥ C √ ψn (x)1/2 . (43)
n

Also uniformly for 1 ≤ k ≤ n,

|pn−1 W | (xkn ) ∼ n−1/4 ψn (xkn )−1/4 (44)

and
 
p W  (xkn ) ∼ n1/4 ψn (xkn )1/4 . (45)
n

Hence
    |p W | (x)
j n W  (x) W −1 xj n =   n   
p W  xj n x − xj n 
n

n−1/4 ψn (x)−1/4
≤C  1/4  .
n1/4 ψn xj n x − xj n 

Next by Cauchy-Schwarz, and then (42), (43),


        
j n W  (x) W −1 xj n = λj n W −1 xj n W (x) Kn x, xj n 
    1/2
≤ λj n W −1 xj n W (x) Kn (x, x) Kn xj n , xj n
132 D. S. Lubinsky

  1/2  −1/2
= λj n W −2 xj n λn (x) W −2 (x)
 −1/4
≤ Cψn xj n ψn (x)1/4 .

Thus combining the two estimates,


 1/4  
    ψn (x) 1
j n W  (x) W −1 xj n ≤ C   min 1,   ,
ψn xj n n1/2 ψn (x)1/2 x − xj n 

which can be recast as (38).


 x   
(g) First note that as 1 − √j n  ≤ Cn−2/3 , we have ψn xj n ∼ n−2/3 . We have to
2n
show that uniformly in n and for x ∈ R,
   √ 

1 + n1/2 ψn (x)1/2 x − xj n  ∼ 1 + n1/2 ψn (x)1/2 x − 2n . (46)
 √  √

Let L be some large positive number. If firstly x − 2n ≥ L 2nn−2/3 , then
from (33),
 
  x − √2n √
 x − xj n  jn C 2nn−2/3
 √ − 1 =  √   ≤ √
 x − 2n   L 2nn−2/3
x − 2n

so that
 
 x − xj n 
 
 x − √2n  ≤ 1 + C/L,

so that
    √ 

1 + n1/2 ψn (x)1/2 x − xj n  ≤ C 1 + n1/2 ψn (x)1/2 x − 2n .

Also, for some C1 independent of L,


 √ 

1 + n1/2 ψn (x)1/2 x − 2n
  √ 
≤ 1 + n1/2 ψn (x)1/2 x − xj n  + C1 2nn−2/3
   C1  √ 
≤ 1 + n1/2 ψn (x)1/2 x − xj n  + n1/2 ψn (x)1/2 x − 2n
L
    C   √ 

≤ 1 + n1/2 ψn (x)1/2 x − xj n  +
1
1 + n1/2 ψn (x)1/2 x − 2n
L
On Marcinkiewicz-Zygmund Inequalities at Hermite Zeros and Their Airy. . . 133

so that
   

1/2 
√  C1  
1+n 1/2
ψn (x) x − 2n 1 − ≤ 1 + n1/2 ψn (x)1/2 x − xj n  .
L
 √  √

Then we have (46) if L is large enough. Next, if x − 2n < L 2nn−2/3 ,
ψn (x) ∼ n−2/3 and then
 √ 

1 ≤ 1 + n1/2 ψn (x)1/2 x − 2n

≤ 1 + Cn1/2 n−1/3 2nn−2/3
  
≤ C2 ≤ C2 1 + n1/2 ψn (x)1/2 x − xj n  .

Again we have (46).


(h) We use the confluent form of the Christoffel-Darboux formula:
γn−1    
λ−1
jn = pn xj n pn−1 xj n .
γn

Here since [32, p. 106, (5.5.10)], Hn (x) = 2nHn−1 (x) so from (17),

pn (x) = 2npn−1 (x) .

Together with (32) this gives


 2
λ−1
j n = pn xj n .

Then (40) follows from (36).



Next, we record some estimates involving the Airy function:
Lemma 2.2
(a) For x ∈ [0, ∞),
 
−1/4 2 3
|Ai (x)| ≤ C (1 + x) exp − x ;
2 (47)
3

|Ai (−x)| ≤ C (1 + x)−1/4 . (48)


134 D. S. Lubinsky

(b) As x → ∞,
    
2 3 π
Ai (−x) = −π −1/2 x 1/4 cos x2 + + O x −3/2 . (49)
3 4
 1/6   
  3π
Ai aj = (−1)j −1 π −1/2 (4j − 1) 1 + O j −2
8
 1/4
= (−1)j −1 π −1/2 aj  (1 + o (1)) . (50)

(c)
  
1
aj = − [3π (4j − 1) /8]2/3 1 + O
j2
 
3πj 2/3
=− (1 + o (1)) . (51)
2

(d)
     
aj  − aj −1  = π aj −1/2 (1 + o (1)) . (52)

(e) For j ≥ 1 and t ∈ [0, ∞),


 
 
Lj (t) ≤ Cj −5/6 (1 + t)−1/4 exp − 2 t 32 (53)
3

and
  C
Lj (−t) ≤  1/4    . (54)
1/4 
1 + (1 + t) aj  t − aj 

Proof (a) The following asymptotics and estimates for Airy functions are listed on
pages 448–449 of [1]: see (10.4.59–61) there.
 
1 −1/4 2 3
Ai (x) = x exp − x 2 (1 + o (1)) , x → ∞;
2π 1/2 3
    3 
−1/2 −1/4 2 3 π
Ai (−x) = π x sin x2 + + O x − 2 , x → ∞.
3 4

Then (47) and (48) follow


* +as Ai is entire.
(b), (c), (d) The zeros aj of Ai satisfy [1, p. 450, (10.4.94,96)]
    
1 3πj 2/3
aj = − [3π (4j − 1) /8] 2/3
1+O =− (1 + o (1)) .
j2 2
On Marcinkiewicz-Zygmund Inequalities at Hermite Zeros and Their Airy. . . 135

 1/6   
  3π
Ai aj = (−1)j −1 π −1/2 (4j − 1) 1 + O j −2
8
 1/4
= (−1)j −1 π −1/2 aj  (1 + o (1)) .

Then (52) also follows, as was shown in [12, p. 431, eqn. (2.7)].
(e) We first prove (54). For t ∈ [0, ∞),
 
   Ai (−t) 

Lj (−t) =    
 Ai aj −t − aj 

C (1 + t)−1/4
≤   
j 1/6 t − aj 
  
by (48), (50). If (1 + t)1/4 j 1/6 t − aj  ≥ 12 |a1 |, we then obtain (54). In the
contrary case,
   1
(1 + t)1/4 j 1/6 t − aj  < |a1 |
2
   1 1  
⇒ t − aj  < |a1 | ≤ aj  .
2 2
We then have for some ξ between −t and aj , from (49),
   1/4
   Ai (ξ )  |ξ |
Lj (t) =    ≤ C   ≤ C.
 Ai aj  aj 

We again obtain (54). Next, for t ∈ (0, ∞), we have from (48), (50),
 
   Ai (t) 

Lj (t) =    
 Ai aj t − aj 
 
C (1 + t)−1/4 2 3
≤   exp − t 2
j 1/6 aj  3
 
−5/6 −1/4 2 3
≤ Cj (1 + t) exp − t 2 .
3


Next, we record a restricted range inequality:
Lemma 2.3 Let η ∈ (0, 1), 0 < p < ∞. There exists B, n0 such that for n ≥ n0
and polynomials P of degree ≤ n + n1/3 ,

P W Lp (R) ≤ (1 + η) P W Lp [−Dn ,Dn ] , (55)


136 D. S. Lubinsky

where
√  
Dn = 2n 1 + Bn−2/3 .

Proof It suffices to prove that

P W Lp (R\[−Dn ,Dn ]) ≤ η P W Lp [−Dn ,Dn ] . (56)

For p ≥ 1, the triangle inequality then yields (55). For p < 1, we can use the
triangle inequality on the integral inside the norm and then just reduce the size of η
appropriately. Let m = m (n) = n + n1/3 . It follows from Theorem 4.2(b) in [11,
p. 96] that for B ≥ 0, P of degree ≤ m,

P W   . √  √  /
Lp R\ − 2m 1+ 21 Bm−2/3 , 2m 1+ 21 Bm−2/3
 
≤ C1 exp −C2 B 3/2 P W  . √ √ /. (57)
Lp − 2m, 2m

Here C1 and C2 are independent of m, P , B. Choose B ≥ 2 so large that


 
C1 exp −C2 B 3/2 ≤ η. (58)

Now
 
√ 1 −2/3
2m 1 + Bm /Dn
2
-
m 1 + 12 Bm−2/3
=
n 1 + Bn−2/3
1 + 12 Bn−2/3
≤ 1 + n−2/3 ≤1
1 + Bn−2/3

for n ≥ n0 (B) as B ≥ 2. Then also 2m/Dn ≤ 1, and
    
√ 1 √ 1
R\ − 2m 1 + Bm−2/3 , 2m 1 + Bm−2/3 ⊇ R\ [−Dn , Dn ]
2 2
and (56) follows from (57) and (58). 
Following is the main part of the proof of Theorem 1.3:
Lemma 2.4 Fix M ≥ 1 and let


M
P (x) = ck Lk (x) . (59)
k=1
On Marcinkiewicz-Zygmund Inequalities at Hermite Zeros and Their Airy. . . 137

Then

M  ∞
|P (ak )|p 6
≤A |P (t)|p dt. (60)
k=1
Ai (ak )2 π2 −∞

Here A is the constant in (19) with R = r = 0.


Proof Choose η ∈ (0, 1) and Dn , B as in the above lemma. Let


M
Rn (x) = Un (x) ck kn (x) W −1 (xkn ) . (61)
k=1

Here we set
⎛   ⎞L
Tm − Tm (1)
x
Dn
Un (x) = ⎝   ⎠ , (62)
m Dn − 1
2 x

where Tm is the usual


0 Chebyshev
1 polynomial, L is some large enough even positive
integer, and m = Lε n1/3 , while ε ∈ (0, 1). Since Rn has degree ≤ n + n1/3 , we
have by Lemma 2.3, at least for large enough n, that

Rn W Lp (R) ≤ (1 + η) Rn W Lp [−Dn ,Dn ] . (63)

We first estimate the norm on the right by splitting the integral inside the norm into
ranges near 1 and away from 1. First let us deal with the range
.√   /
I1 = 2n 1 − 6−1/3 (2n)−2/3 R , Dn ,

where R is some fixed (large) number. For x ∈ I1 , write for t ∈ [−R, 61/3 22/3 B],
√  
x= 2n 1 + 6−1/3 (2n)−2/3 t . (64)

To find the asymptotics for Un , also write


x s
= cos
Dn m
x s 1  s 2
⇒1− = 2 sin2 = (1 + o (1))
Dn 2m 2 m
&  
x
⇒ s = 2m 1 −
2 + o (1)
Dn
2  
ε
⇒s= 2 B − 6−1/3 2−2/3 t + o (1) .
L
138 D. S. Lubinsky

Then if S (u) = sin u


u is the sinc kernel,
 
Tm x
Dn − Tm (1) cos s − 1 −2 sin2 s
  =  = 2
+ o (1)
− 12 s 2
Dn − 1 m2 Dxn − 1
x
m2
  s 2
= S + o (1)
2
⎛ & ⎞
ε B − 6−1/3 2−2/3 t
= S⎝ ⎠ + o (1) ,
L 2

and uniformly in such x,


⎛ & ⎞L
ε B −6−1/3 2−2/3 t⎠
Un (x) = S ⎝ + o (1) .
L 2

In particular, for each fixed k, as n → ∞, recalling (33), and that ak < 0,


⎛ & ⎞L
ε B +6−1/3 2−2/3 |ak |
Un (xkn ) = S ⎝ ⎠ + o (1) . (65)
L 2

Then uniformly for x in this range, from Lemma 2.1(e) and recalling (59),
 
 M 
 
|Rn W | (x) = Un (x) ck (kn W ) (x) W −1 (xkn )
 
k=1
 ⎛ & ⎞L 
 
 −1/3 −2/3 
 ⎝ε B −6 2 t⎠ 
= S P (−t) + o (1) . (66)
 L 2 
 

Then as |S (u)| ≤ 1,

|Rn W |p (x) dx
I1
 
61/3 22/3 B
−1/3 −1/6
≤6 (2n) |P (−t)| dt + o (1) .
p
(67)
−R
On Marcinkiewicz-Zygmund Inequalities at Hermite Zeros and Their Airy. . . 139

Next, for x ∈ [−Dn , Dn ],


⎧ ⎛ ⎫⎞L
⎨ 2 ⎬
|Un (x)| ≤ ⎝min 1,    ⎠
⎩ m2 x − 1  ⎭
Dn

C
≤  L
 
1 + m2  Dxn − 1

1
≤ Cn−2L/3   L
 
n−2/3 +  axn − 1

by straightforward estimation. Here C depends on ε. Then from Lemma 2.1(g),

1 n1/6 ψn (x)1/4
|Rn (x) W (x)| ≤ Cn−2L/3   L .
 
n−2/3 +  axn − 1 1 + n ψn (x) |x − an |
1/2 1/2

(68)

Of course here C depends on the particular P and ε, but not on n nor R nor x. Then

|Rn W | (x)p dx
[−Dn ,Dn ]\I1
 √  
2n 1−6−1/3 (2n)−2/3 R
−2Lp/3+p/6
≤ Cn
−Dn
⎡ ⎤p
1/4
⎢ 1 n1/6 ψn (x) ⎥
× ⎣   L  √  ⎦ dx
  
n−2/3 +  √ − 1 1 + n ψn (x) x − 2n
x 1/2 1/2
2n
 1−6−1/3 (2n)−2/3 R
≤ Cn−2Lp/3+p/6+1/2
−(1+Bn−2/3 )
9  1/4 :p
1 |1 − |y|| + n−2/3
×  L  1/2 dy
n−2/3 + |y − 1| 1 + n |1 − |y|| + n−2/3 |y − 1|
⎧   1/4 p ⎫

⎪ 0 |1−|y||+n−2/3 ⎪
⎨ −(1+Bn−2/3 ) 1/2 dy ⎪

−2Lp/3+p/6+1/2 1+n(|1−|y||+n−2/3 )
≤ Cn . /p

⎪  −1/3 −2/3 ⎪

⎩ + 1−6 (2n) R 1
L+5/4 dy ⎭
0 n|y−1|
140 D. S. Lubinsky

  9 :p
−2Lp/3+p/6+1/2 −2/3
n2/3 n−1/6 (|s| + 1)1/4
≤ Cn n ds
−B 1 + n2/3 (|s| + 1)1/2
 1−(L+5/4)p 
+ n−p Rn−2/3

  n2/3
−2Lp/3+p/6+1/2 1
≤ Cn n−2/3−5p/6 ds
−B (|s| + 1)p/4
 1−(L+5/4)p 
+ n−p Rn−2/3
  1−(L+5/4)p 
≤ Cn−2Lp/3+p/6+1/2 n−5p/6 + n−p Rn−2/3

≤ Cn−2Lp/3−2p/3+1/2 + Cn−1/6 R 1−(L+5/4)p .

Assuming that L is large enough so that

−2Lp/3 − 2p/3 + 1/2 < −1/6

and

1 − (L + 5/4) p < −1,

we have
  
|Rn W | (x)p dx ≤ o n−1/6 + Cn−1/6 R −1 .
[−Dn ,Dn ]\I1

Then combined with (67) and (63) this gives


 ∞
(1 + η)−p |Rn W |p
−∞
 61/3 22/3 B  
≤ 6−1/3 (2n)−1/6 |P (t)|p dt + o n−1/6 + Cn−1/6 R −1 . (69)
−R

Next from (40), and (65–66), for each fixed k, as P (ak ) = ck ,


. /−1
λkn W −2 (xkn ) |Rn W (xkn )|p = 34/3 π −2 23/2 n1/6 Ai (ak )2
⎧ ⎛ & ⎞Lp ⎫
⎨
⎪ −1/3 2−2/3 |a | 
 ⎪

 ε B + 6 
× S ⎝ ⎠ |P (ak )| + o (1)
k p
⎪  ⎪
⎩ L 2  ⎭
On Marcinkiewicz-Zygmund Inequalities at Hermite Zeros and Their Airy. . . 141

so


M . /−1
λkn W −2 (xkn ) |Rn W (xkn )|p = 34/3 π −2 23/2 n1/6
k=1
⎧  ⎛ & ⎞Lp ⎫

⎨   ⎪

M
|P (ak )|p  −1/3 −2/3 |ak | 
× S ⎝ ε B + 6 2 ⎠ + o (1) . (70)
⎪ Ai (ak )2   ⎪
⎩ k=1  L 2  ⎭

Together with (19) and (69), this gives as n → ∞,


 ⎛ & ⎞Lp
. /−1   
M
|P (ak )|p  ε B + 6−1/3 2−2/3 |a | 
(1 + η)−p 34/3 π −2 23/2 S ⎝ k ⎠
Ai (ak )2  
k=1  L 2 
 61/3 22/3 B
−1/3 −1/6
≤6 2 A |P (t)|p dt + CR −1 .
−R

Here B, ε are independent of R. We let R → ∞ and obtain


 ⎛ & ⎞Lp
  
M
|P (ak )|   −1/3 −2/3 |ak | 
⎝ε B +6
p
2 ⎠
(1 + η)−p S
k=1
Ai (ak )2  L 2 

 61/3 21/6 B
≤ 6π −2 A |P (t)|p dt.
−∞

Now let ε → 0+ :


M  ∞
−p |P (ak )|p −2
(1 + η) ≤ 6π A |P (t)|p dt.
k=1
Ai (ak )2 −∞

Finally we can let η → 0 :


M  ∞
|P (ak )|p −2
≤ 6π A |P (t)|p dt.
k=1
Ai (ak )2 −∞


Proof of Theorem 1.3(a) Recall that SM [f ] is the partial sum defined in (24). As
f ∈ Gp ,
 ∞
lim |f (t) − SM [f ] (t)|p dt = 0.
M→∞ −∞
142 D. S. Lubinsky

Then for a fixed positive integer L, and by Lemma 2.4, and as SM [f ] (ak ) = f (ak )
for k ≤ M,
 1/p  L 1/p

L
|f (ak )|p  |SM [f ] (ak )|p
= lim
k=1
Ai (ak )2 M→∞
k=1
Ai (ak )2
M 1/p
 |SM [f ] (ak )|p
≤ lim sup
M→∞ k=1
Ai (ak )2
 1/p  ∞ 1/p
6
≤ A lim sup |SM [f ] (t)| p
dt
π2 M→∞ −∞
 1/p   ∞ 1/p
6
≤ A lim sup |SM [f ] (t) − f (t)| dt
p
π2 M→∞ −∞
 ∞ 1/p 
+ |f (t)|p dt
−∞
 1/p  ∞ 1/p
6
= A |f (t)| dt
p
.
π2 −∞

Now let L → ∞. 
For Theorem 1.3(b), we need:
Lemma 2.5 Assume that for some β > 14 , we have

|f (x)| ≤ C (1 + |x|)−β , x ∈ (−∞, 0) . (71)

Then for M ≥ 1, and all t ∈ (−∞, 0],

|SM [f ]| (t) ≤ C (1 + |t|)−β log (2 + |t|) . (72)

For t ∈ (0, ∞),


 
−1/4 2 3
|SM [f ]| (t) ≤ C (1 + t) exp − t 2 . (73)
3

Proof From (71) and (54), followed by (52), for t ≥ 0,


 −β

M aj 
|SM [f ]| (−t) ≤ C  1/4   
j =1 1 + (1 + t)1/4 aj  t − aj 
On Marcinkiewicz-Zygmund Inequalities at Hermite Zeros and Their Airy. . . 143

 −β+1/2

M
    aj 
≤C aj  − aj −1   1/4   
j =1 1 + (1 + t)1/4 aj  t − aj 
 ∞ s −β+1/2
≤C ds.
0 1 + (1 + t)1/4 s 1/4 |t − s|

If 0 ≤ t ≤ 1, we can bound this by


 2  ∞
C s −β+1/2 ds + C s −β−3/4 ds ≤ C,
0 2

recall β > 14 . If t ≥ 1, we can bound this by


 ∞ s −β+1/2
C ds
0 1 + t 1/4 s 1/4 |t − s|
 ∞
u−β+1/2
= Ct −β+3/2 du
0 1 + t 3/2 u1/4 |u − 1|
9  1−1/t 3/2 u−β+1/4 du  1+1/t 3/2 :
−β+3/2 t −3/2 0 |u−1| + 1−1/t 3/2 1du
≤ Ct 2 ∞
+t −3/2 1+1/t 3/2 |u−1|
du
+ t −3/2 2 u−β−3/4 du
0 1
≤ Ct −β log (1 + |t|) + 1 + log (1 + |t|) + 1 .

Thus we have the bound (72). Next, if t ≥ 0, we obtain from (53) and (51),
  M
−1/4 2 3   −β −5/6
|SM [f ]| (−t) ≤ C (1 + t) exp − t 2 aj j
3
j =1
  M
−1/4 2 3  −5/6−2β/3
≤ C (1 + t) exp − t 2 j
3
j =1
 
−1/4 2 3
≤ C (1 + t) exp − t 2 ,
3

as 5/6 + 2β/3 > 5/6 + 1/6 > 1. 


Proof of Theorem 1.3(b). Recall that we are assuming p ≥ 2. If N > M, we have
in view of the lemma and our bound on f
 ∞
|SN [f ] − SM [f ]|p (t) dt
−∞
144 D. S. Lubinsky

 ∞
≤C |SN [f ] − SM [f ]|2 (t) dt
−∞

→ 0 as M, N → ∞,

as f ∈ G implies that SM [f ] → f in L2 (R) as M → ∞. It follows that {SM [f ]}


is Cauchy in Lp (R), so has a limit there. This limit must be f , as f ∈ G. Then also
f ∈ Gp and the result follows. 
M
Lemma 2.6 Assume that (22) holds with R = r = 0. Let P = k=1 P (ak ) · Lk
and 1 < p < 4. Then

 ∞ π 2  |P (ak )|p
M
|P (t)|p dt ≤ B . (74)
−∞ 6
j =1
Ai (ak )2

Proof We use (22) with R = r = 0. If Rn is a polynomial of degree ≤ n − 1,


 
∞ n
  p  
|(Rn W ) (x)|p dx ≤ B λj n Rn xj n  W p−2 xj n . (75)
−∞ j =1

Let


M
Rn (x) = P (ak ) kn (x) W −1 (xkn ) .
k=1

Let R > 0 and


.√   √ /
I1 = 2n 1 − 6−1/3 (2n)−2/3 R , 2n(1 + 6−1/3 (2n)−2/3 R) .

From (37) with x of the form (34), we have

|Rn W | (x) = |P (−t)| + o (1) ,

so
  R 
−1/3 −1/6
|Rn W | (x) dx = 6
p
(2n) |P (t)| dt + o (1) .
p
I1 −R

Also, as at (70),


n
  p  
λj n Rn xj n  W p−2 xj n
j =1
On Marcinkiewicz-Zygmund Inequalities at Hermite Zeros and Their Airy. . . 145


M
  p  
= λj n Rn xj n  W p−2 xj n
j =1

. /−1 M
|P (ak )|p
= (1 + o (1)) 34/3 π −2 23/2 n1/6 .
k=1
Ai (ak )2

Then (75) gives


 R 
−1/3 −1/6
6 (2n) |P (t)| dt + o (1)
p
−R

. /−1 M
|P (ak )|p
≤ B (1 + o (1)) 34/3 π −2 23/2 n1/6
k=1
Ai (ak )2

or
 
π 2  |P (ak )|p
R M
|P (t)|p dt + o (1) ≤ B (1 + o (1)) .
−R 6
k=1
Ai (ak )2

Letting R → ∞ gives (74). 


Proof of Theorem 1.4. (a) Lemma 2.6 gives

f Lp (R) ≤ f − SM [f ]Lp (R) + SM [f ]Lp (R)


 1/p
π 2  |f (ak )|p
M
≤ f − SM [f ]Lp (R) + B
6
k=1
Ai (ak )2
∞
1/p
π 2  |f (ak )|p
→0+ B ,
6
k=1
Ai (ak )2

as M → ∞.
(b) Our assumption that f ∈ G ensures that f = limM→∞ SM [f ] uniformly in
compact sets. Next, given N > M, we have from Lemma 2.6,

 ∞ π 2  |f (ak )|p
N
|SN [f ] − SM [f ]|p (t) dt ≤ B
−∞ 6
k=M+1
Ai (ak )2

 |f (ak )|p
≤C → 0,
k 1/3
k=M+1

as k → ∞ – recall (50) and our hypothesis (30). So {SM [f ]} is Cauchy in complete


Lp (R) and as above, its limit in Lp (R) must be f , so that (a) is applicable. 
146 D. S. Lubinsky

References

1. M. Abramowitz, I.A. Stegun, Handbook of Mathematical Functions (Dover, New York, 1965)
2. R.P. Boas, Entire Functions (Academic Press, New York, 1954)
3. S.B. Damelin, D.S. Lubinsky, Necessary and sufficient conditions for mean convergence of
Lagrange interpolation for Erdös weights. Can. Math. J. 40, 710–736 (1996)
4. S.B. Damelin, D.S. Lubinsky, Necessary and sufficient conditions for mean convergence of
Lagrange interpolation for Erdös weights II. Can. Math. J. 40, 737–757 (1996)
5. F. Filbir, H.N. Mhaskar, Marcinkiewicz-Zygmund measures on manifolds. J. Complex. 27,
568–596 (2011)
6. G.R. Grozev, Q.I. Rahman, Lagrange interpolation in the zeros of Bessel functions by entire
functions of exponential type and mean convergence. Methods Appl. Anal. 3, 46–79 (1996)
7. H. König, Vector-valued Lagrange interpolation and mean convergence of Hermite series, in
Functional Analysis (Essen, 1991). Lecture notes in pure and applied mathematics, vol. 150
(Dekker, New York, 1994), pp. 227–247
8. H. König, N.J. Nielsen, Vector-valued Lp convergence of orthogonal series and Lagrange
interpolation. Forum Math. 6, 183–207 (1994)
9. B. Ja Levin, Lectures on Entire Functions, Translations of Mathematical Monographs (Ameri-
can Mathematical Society, Providence, 1996)
10. E. Levin, D.S. Lubinsky, Christoffel functions, orthogonal polynomials, and Nevai’s conjecture
for Freud weights.Constr. Approx. 8, 463–535 (1992)
11. E. Levin, D.S. Lubinsky, Orthogonal Polynomials for Exponential Weights (Springer, New
York, 2001)
12. E. Levin, D.S. Lubinsky, On the Airy reproducing kernel, sampling series, and quadrature
formula. Integr. Equ. Oper. Theory 63, 427–438 (2009)
13. F. Littmann, Marcinkiewicz inequalities for entire functions in spaces with Hermite-Biehler
weights, manuscript
14. D.S. Lubinsky, Converse quadrature sum inequalities for polynomials with Freud weights. Acta
Sci. Math. 60, 527–557 (1995)
15. D.S. Lubinsky, Marcinkiewicz-Zygmund inequalities: methods and results, in Recent Progress
in Inequalities, ed. by G.V. Milovanovic et al. (Kluwer Academic Publishers, Dordrecht, 1998),
pp. 213–240
16. D.S. Lubinsky, On sharp constants in Marcinkiewicz-Zygmund and Plancherel-Polya inequal-
ities. Proc. Am. Math. Soc. 142, 3575–3584 (2014)
17. D.S. Lubinsky, On Marcinkiewicz-Zygmund inequalities at Jacobi zeros and their Bessel
function cousins. Contemp. Math. 669, 223–245 (2017)
18. D.S. Lubinsky, Mean convergence of interpolation at zeros of airy functions, in Contemporary
Computational Mathematics – A Celebration of the 80th Birthday of Ian Sloan, ed. by J. Dick,
F.Y. Kuo, H. Wozniakowski, vol. 2 (Springer International, Switzerland, 2018), pp. 889–909
19. D.S. Lubinsky, G. Mastroianni, Converse quadrature sum inequalities for Freud Weights. II
Acta Math. Hungar. 96, 161–182 (2002)
20. D.S. Lubinsky, A. Maté, P. Nevai, Quadrature sums involving pth powers of polynomials.
SIAM J. Math. Anal. 18, 531–544 (1987)
21. D.S. Lubinsky, D. Matjila, Full quadrature sums for pth powers of polynomials with Freud
weights. J. Comp. Appl. Math. 60, 285–296 (1995)
22. G. Mastorianni, G.V. Milovanovic, Interpolation Processes: Basic Theory and Applications
(Springer, Berlin, 2008)
23. G. Mastroianni, V. Totik, Weighted polynomial inequalities with doubling and A∞ Weights.
Constr. Approx. 16, 37–71 (2000)
24. H.N. Mhaskar, F.J. Narcowich, J.D. Ward, Spherical Marcinkiewicz-Zygmund inequalities and
positive quadrature. Math. Comp. 70(235), 1113–1130 (2001)
On Marcinkiewicz-Zygmund Inequalities at Hermite Zeros and Their Airy. . . 147

25. H.N. Mhaskar, J. Prestin, On Marcinkiewicz-Zygmund-Type Inequalities, in Approximation


Theory: In Memory of A.K. Varma, ed. by N.K. Govil et al. (Marcel Dekker, New York, 1998),
pp. 389–404
26. P. Nevai, Lagrange interpolation at zeros of orthogonal polynomials, in Approximation Theory
II, ed. by G.G. Lorentz et al. (Academic Press, New York, 1976), pp. 163–201
27. P. Nevai, Orthogonal polynomials. Mem. Am. Math. Soc. 18(213), 185 (1979)
28. P. Nevai, Mean convergence of Lagrange interpolation II. J. Approx. Theory 30, 263–276
(1980)
29. P. Nevai, Geza Freud, orthogonal polynomials and Christoffel functions. A case study. J.
Approx. Theory 40, 3–167 (1986)
30. J. Ortega-Cerda, J. Saludes, Marcinkiewicz-Zygmund inequalities. J. Approx. Theory 147,
237–252 (2007)
31. J. Szabados, Weighted Lagrange and Hermite-Fejer interpolation on the real line. J. Inequal.
Appl. 1, 99–123 (1997)
32. G. Szegő, Orthogonal Polynomials (American Mathematical Society, Providence, 1939)
33. Y. Xu, On the Marcinkiewicz-Zygmund inequality, in Progress in Approximation Theory, ed.
by P. Nevai, A. Pinkus (Academic Press, Boston, 1991), pp. 879–891
34. Y. Xu, Mean convergence of generalized Jacobi series and interpolating polynomials II. J.
Approx. Theory 76, 77–92 (1994)
35. A. Zygmund, Trigonometric Series, vols. 1, 2, Second Paperback edition (Cambridge Univer-
sity Press, Cambridge, 1988)
The Maximum of Cotangent Sums
Related to the Nyman-Beurling Criterion
for the Riemann Hypothesis

Helmut Maier, Michael Th. Rassias, and Andrei Raigorodskii

Abstract In a previous paper (see H. Maier, M. Th. Rassias, The maximum of


cotangent sums related to Estermann’s zeta function in rational numbers in short
intervals. Appl. Anal. Discrete Math. 11, 166–176 (2017)) we investigate the
maximum of certain cotangent sums. These cotangent sums can be associated to the
study of the Riemann Hypothesis through its relation with the so-called Vasyunin
sum. Here we continue this research by restricting the rational numbers in short
intervals to rational numbers of special type.

Keywords Cotangent sums · Estermann zeta function · Nyman-Beurling


criterion · Riemann zeta function · Riemann Hypothesis

2000 Mathematics Subject Classification: 26A12; 11L03.

H. Maier
Department of Mathematics, University of Ulm, Ulm, Germany
e-mail: [email protected]
M. Th. Rassias ()
Institute of Mathematics, University of Zurich, Zurich, Switzerland
Moscow Institute of Physics and Technology, Dolgoprudny, Russia
Institute for Advanced Study, Program in Interdisciplinary Studies, Princeton, NJ, USA
e-mail: [email protected]
A. Raigorodskii
Moscow Institute of Physics and Technology, Dolgoprudny, Russia
Moscow State University, Moscow, Russia
Buryat State University, Ulan-Ude, Russia
Caucasus Mathematical Center, Adyghe State University, Maykop, Russia
e-mail: [email protected]

© Springer Nature Switzerland AG 2020 149


A. Raigorodskii, M. T. Rassias (eds.), Trigonometric Sums and Their Applications,
https://doi.org/10.1007/978-3-030-37904-9_7
150 H. Maier et al.

1 Introduction

The authors in various papers [8–12] and the second author in his thesis [13], studied
the distribution of cotangent sums

r  
b−1
m  π mr 
c0 =− cot
b b b
m=1

as r ranges over the set

{r : (r, b) = 1, A0 b ≤ r ≤ A1 b} ,

where A0 , A1 are fixed with 1/2 < A0 < A1 < 1 and b tends to infinity.
Bettin [2] succeeded in replacing the inequality 1/2 < A0 < A1 < 1 by
0 < A0 < A1 ≤ 1.
These sums are related to the values of the Estermann zeta function E(s, r/b, α),
which are defined by the Dirichlet series
 r   σ (n) exp (2π inr/b)
α
E s, , α = ,
b ns
n≥1

where Re s > Re α + 1, b ≥ 1, (r, b) = 1 and



σα (n) = dα .
d|n

Estermann (see [4]) introduced the above function in the special case when α = 0
and Kiuchi (see [7]) for α ∈ (−1, 0]. Ishibashi (see [6]) proved the following
relation regarding the value of E s, br , α at s = 0:
Let b ≥ 2, 1 ≤ r ≤ b, (r, b) = 1, α ∈ N ∪ {0}. Then, for even α, it holds that
 r   i α+1 
b−1
m  π mr  1
E 0, , α = − cot(α) + δα,0 ,
b 2 b b 4
m=1

where δα,0 is the Kronecker delta function.


For α = 0, the sum on the right reduces to c0 (r/b).
The cotangent sum c0 (r/b) can be associated to the study of the Riemann
Hypothesis, also through its relation with the so-called Vasyunin sum. The Vasyunin
sum is defined as follows:

r  b−1

 mr  π mr 
V := cot ,
b b b
m=1

where {u} = u − u, u ∈ R.


The Maximum of Cotangent Sums Related to the Nyman-Beurling Criterion. . . 151

It can be shown that


r   

V = −c0 ,
b b

where r̄ is such that r̄r ≡ 1 (mod b).


The Vasyunin sum is itself associated to the study of the Riemann hypothesis
through the following identity (see [1, 3]):
  
+∞  2  
1 
ζ 1 + it  r dt
it
  (1)
2π(rb)1/2 −∞ 2 b 1
4 + t2
      
log 2π − γ 1 1 b−r r π r b
= + + log − V +V .
2 r b 2rb b 2rb b r

Note that the only non-explicit function in the right hand side of (1) is the Vasyunin
sum. According to this approach, the Riemann Hypothesis is true if and only if

lim dN = 0,
N →+∞

where
 +∞ 
    2
1 
dN2 = inf 1 − ζ 1 + it DN 1 + it  dt
DN 2π  2 2  1
+ t2
−∞ 4

and the infimum is taken over all Dirichlet polynomials


N
an
DN (s) = .
ns
n=1

The authors in several papers [8–12] and the second author in his thesis [13]
investigated moments of the form

1   r 2k 1
c0 , < A0 < A1 < 1
φ(b) b 2
(r,b)=1
A0 b<r≤A1 b

and could show that


1   r 2k
c0 = Hk b2k (1 + o(1)), (b → +∞),
φ(b) b
(r,b)=1
A0 b<r≤A1 b
152 H. Maier et al.

where
 1  g(x) 2k
Hk := dx ,
0 π
 1 − 2{lx}
g(x) := .
l
l≥1

The range 1/2 < A0 < A1 < 1 was later extended to 0 < A0 < A1 < 1 by Bettin
in [2].   
In the paper [12] the authors investigated the maximum of c0 br  for the values
r/b in a short interval. They gave the following definition:
Definition 1.1 (of [12]) Let 0 < A0 < 1, 0 < C < 1/2. For b ∈ N we set

 := (b, C) = b−C .

We set
  r 
 
M(b, C, A0 ) := max c0 .
A0 b≤r<(A0 +)b b

They proved the following results:


Theorem 1.2 (of [12]) With Definition 1.1 of [12] let D satisfy 0 < D < 1
2 − C.
Then we have for sufficiently large b:

D
M(b, C, A0 ) ≥ b log b .
π

Theorem 1.3 (of [12]) Let C be as in Theorem 1.2 of [12] and let D satisfy
D > 2 − C − E, where E ≥ 0 is a fixed constant. Let B be sufficiently large. Then
we have:
D
M(b, C, A0 ) ≤ b log b ,
π

for all b with B ≤ b < 2B with at most B E exceptions.


In this paper we replace b by a prime number q and the numerators r of the rational
numbers r/b by prime numbers. For the statement and proof of the next theorem we
need the following definition.
Definition 1.4 The letter p always denotes a prime number. Let 0 < A0 < 1,
0 < C < 1/2. For q a prime number we set:

 := (q, C) = q −C .
The Maximum of Cotangent Sums Related to the Nyman-Beurling Criterion. . . 153

We set
  
 
L(q, C, A0 ) := max c0 p  .
A0 q≤p≤(A0 +)q  q 

We shall prove the following results:


Theorem 1.5 With Definition 1.4, let D satisfy

1
0<D< −C.
32
Then we have for sufficiently large q:

D
L(q, C, A0 ) ≥ q log q .
π
As in [12] a crucial role is played by the simultaneous localisation of a residue-
class and its multiplicative inverse. In [12] the inverse r̄ (mod b) of r (mod b) is
defined by r̄r ≡ 1 (mod b). The quantities r and r̄ are simultaneously located
by the application of Kloosterman sums. Here we additionally have the primality
condition. We have to localise simultaneously the prime p and its inverse p̄ (mod
q).
This will be achieved by the application of a result of Fouvry and Michel [5],
Lemma 2.1, on exponential sums in finite fields over primes. The analogue of [5]
follows immediately, since the upper bound for the cardinality of a set is always also
an upper bound for all of its subsets. We have the following:
Corollary 1.6 (of Theorem 1.3 of [12]) Let C be as in Theorem 1.5 and let D
satisfy D > 2 − C − E, where E ≥ D is a fixed constant. Let Q be sufficiently
large. Then we have

D
L(q, C, A0 ) ≤ q log q
π

for all q with Q ≤ q ≤ 2Q, q prime, with at most QE exceptions.


We now give the proof of Theorem 1.5.

2 Exponential Sums over Primes in Finite Fields

Lemma 2.1 Let Fr be a finite field with r elements and ψ be a non-trivial additive
character over Fr , f a rational function of the form

P (x)
f (x) = ,
Q(x)
154 H. Maier et al.

P and Q relatively prime monic polynomials



S(f ; r, x) := ψ(f (p))
p≤x

(p denotes the p-fold sum of the element 1 in Fr ).


Then we have

S(f ; r, x) r 3/16+ x 25/32 .

The implied constant depends only on  and the degrees of P and Q.


Proof This is due to Fouvry and Michel [5]. 


3 Other Preliminary Lemmas

Definition 3.1 For x ∈ R, Re(s) > 1 we set


 d(n) sin(2π nx)
Dsin (s, x) := . (2)
ns
n≥1

Lemma 3.2 Let (a0 ; a1 , a2 , . . .) be the continued fraction expansion of x ∈ R.


Moreover, let ur /vr be the r-th partial quotient of x. Then
   
π 2  (−1)l 1 vl−1
Dsin (1, x) = − +ψ , (3)
2 vl π vl vl
l≥1

whenever either of the two series (2) and (3) is convergent.


If x = (a0 ; a1 , a2 , . . . , ar ) is a rational number, then the range of summation of
the series on the right is to be interpreted to be 1 ≤ l ≤ r. Here ψ is an analytic
function satisfying

log(2π x) − γ
ψ(x) = − + O(log x), (x → 0) .
πx

Proof This is Proposition 1 of Bettin [2]. 



Lemma 3.3
r   r  
1 r̄
c0 = Dsin 0, = 2b π −2 Dsin 1, ,
b 2 b b

where r r̄ ≡ 1(mod b).


The Maximum of Cotangent Sums Related to the Nyman-Beurling Criterion. . . 155

Proof This is due to Ishibashi [6]. 



Definition 3.2 Let  be as in Definition 1.4 and  > 0. We set

N (q, , ) := #{p : A0 q ≤ p < (A0 + )q , |p̄| ≤ q} .

Let the functions χ1 , χ2 be defined by



1 , if A0 + v < u ≤ A0 +  − v
χ1 (u, v) :=
0 , otherwise .

and
 
χ2 (u, v) := −1 χ1 (u, v) dv .
0

Lemma 3.5 We have




χ2 (u) = a(n)e(nu),
n=−∞

where a(0) = /2 and



O() , if |n| ≤ −1
a(n) =
O(−1 n−2 ) , if |n| > −1 .

Proof This is Lemma 2.10 of [12]. 



Definition 3.6 Let

1, if −  + v < u < 
χ3 (u, v) :=
0, otherwise .

and
 
−1
χ4 (u) :=  χ3 (u, v)dv .
0

Lemma 3.7 We have


+∞

χ4 (u) = c(n) e(nu),
n=−∞

where c(0) =  and



O() , if |n| ≤ −1
c(n) =
O(−1 n−2 ) , if |n| > −1 .
156 H. Maier et al.

Proof This is Lemma 2.12 of [12]. 



Lemma 3.8 Let  > 0 be such that
1
D+ < − C.
32
Set

 := q −(D+) .

Then

N(q, , ) > 0

for q sufficiently large.


Proof By Definitions 3.2 and 3.6 we have

+∞

N (q, , ) ≥ φ(q)a(0)c(0) + a(m)c(n)E(m, n, q) ,
m,n=−∞
(m,n) =(0,0)

where
  
mp + np̄
E(m, n, q) := e .
q
1≤p≤q−1

For (m, n) = (0, 0) we estimate E(m, n, q) by application of Lemma 2.1, where we


set
 
ml
r = q, P (x) = x 2 + 1, Q(x) = x, ψ(l mod q) = e
q

and obtain

E(m, n, q) q 31/32 .

Lemma 3.8 follows from Lemmas 3.5 and 3.7. 



Lemma 3.9 Let  > 0, q ≥ q(). For 1 ≤ p < q, let
p
= 0; w1 , . . . , ws 
q

be the continued fraction expansion of p/q with partial fractions ui /vi . Then there
are at most 3 values of l for which
The Maximum of Cotangent Sums Related to the Nyman-Beurling Criterion. . . 157

 
1 vl−1
ψ ≥ log log q
vl vl

and at most one value of l, for which


 
1 vl−1
ψ ≥  log q.
vl vl

Proof This is Lemma 2.14 of [12]. 




4 Proof of Theorem 1.5

By Lemma 3.8 there is at least one

p ∈ [A0 q, (A0 + )q],



such that q ∈ (0, ). By Lemmas 3.2 and 3.3 we have:
   (−1)l  1   
p vl−1
c0 = −q +ψ .
q vl π vl vl
l≥1

Let (ui /vi )si=1 be the sequence of partial fractions of p̄/q. From

p̄ 1
≥ ≥ ,
q v1 + 1

we obtain

v1 + 1 ≥ −1 .

By Lemma 3.9 we have

 1 
vl−1

+ψ < 2 log q, for q ≥ q0 ().
π vl vl
l>1

Therefore,
  
 
Dsin 0, p  ≥ 1 log(−1 )(1 + o(1)) (q → +∞).
 q  π

This proves Theorem 1.5. 



158 H. Maier et al.

Acknowledgements M. Th. Rassias: I would like to express my gratitude to the John S. Latsis
Public Benefit Foundation for their financial support provided under the auspices of my current
“Latsis Foundation Senior Fellowship” position.

References

1. S. Bettin, A generalization of Rademacher’s reciprocity law. Acta Arith. 159(4), 363–374


(2013)
2. S. Bettin, On the distribution of a cotangent sum. Int. Math. Res. Not. (2015). https://doi.org/
10.1093/imrn/rnv036
3. S. Bettin, B. Conrey, Period functions and cotangent sums. Algebra Number Theory 7(1), 215–
242 (2013)
4. T. Estermann, On the representation of a number as the sum of two products. Proc. Lond. Math.
Soc. 31(2), 123–133 (1930)
5. E. Fouvry, P. Michel, Sur certaines sommes d’exponentielles sur les nombres premiers. Ann.
Sci. Écope Norm. Sup. (4), 31, 93–130 (1998)
6. M. Ishibashi, The value of the Estermann zeta function at s = 0. Acta Arith. 73(4), 357–361
(1995)
7. I. Kiuchi, On an exponential sum involving the arithmetic function σα (n). Math. J. Okayama
Univ. 29, 193–205 (1987)
8. H. Maier, M.Th. Rassias, Generalizations of a cotangent sum associated to the Ester-
mann zeta function. Commun. Contemp. Math. 18(1) (2016). https://doi.org/10.1142/
S0219199715500789
9. H. Maier, M.Th. Rassias, The order of magnitude for moments for certain cotangent sums. J.
Math. Anal. Appl. 429(1), 576–590 (2015)
10. H. Maier, M.Th. Rassias, The rate of growth of moments of certain cotangent sums.
Aequationes Math. 2015, 90(3), 581–595 (2016)
11. H. Maier, M.Th. Rassias, Asymptotics for moments of certain cotangent sums. Houst. J. Math.
43(1), 207–222 (2017)
12. H. Maier, M.Th. Rassias, The maximum of cotangent sums related to Estermann’s zeta function
in rational numbers in short intervals. Appl. Anal. Discrete Math. 11, 166–176 (2017)
13. M.T. Rassias, Analytic investigation of cotangent sums related to the Riemann zeta function.
Doctoral Dissertation, ETH-Zürich, Switzerland, 2014
Double-Sided Taylor’s Approximations
and Their Applications in Theory of
Trigonometric Inequalities

Branko Malešević, Tatjana Lutovac, Marija Rašajski, and Bojan Banjac

Abstract In this paper the double-sided TAYLOR’s approximations are used to


obtain generalisations and improvements of some trigonometric inequalities.

Keywords Trigonometric inequalities · Double-sided Taylor’s approximations

1 Introduction

Many mathematical and engineering problems cannot be solved without TAYLOR’s


approximations [1–3]. Particularly, their application in proving various analytic
inequalities is of great importance [4–8]. Recently, numerous inequalities have
been generalized and improved by the use of the so-called double-sided TAYLOR’s
approximations [6, 9–14] and [15]. Many topics regarding these approximations are
presented in [15]. Some of the basic concepts and results about the double-sided
TAYLOR’s approximations presented in [15], which will be used in this paper, are
given in the next section.
In this paper, using the double-sided TAYLOR’s approximations, we obtain
generalizations and improvements of some trigonometric inequalities proved by
Sandor [16].
Statement 1 ([16], Theorem 1)
cos x
1−
3 cos x2 4
< < , (1)
8 x2 π2
for any x ∈ (0, π/2).

B. Malešević () · T. Lutovac · M. Rašajski


School of Electrical Engineering, University of Belgrade, Belgrade, Serbia
e-mail: [email protected]; [email protected]; [email protected]
B. Banjac
Faculty of Technical Sciences, University of Novi Sad, Novi Sad, Serbia
e-mail: [email protected]

© Springer Nature Switzerland AG 2020 159


A. Raigorodskii, M. T. Rassias (eds.), Trigonometric Sums and Their Applications,
https://doi.org/10.1007/978-3-030-37904-9_8
160 B. Malešević et al.

Note that D’Aurizio [17] used the infinite products as well as some inequalities
connected with the RIEMANN zeta function ζ to prove the right-hand side inequality
(1).

Statement 2 ([16], Theorem 2)

sin x
4  √  2 − sin x2 1
2− 2 < < , (2)
π2 x2 4

for any x ∈ (0, π/2).

Inequalities (1) and (2) are reducible to mixed trigonometric-polynomial inequal-


ities and can be proved by methods and algorithms that have been developed and
shown in papers [5, 7] and dissertation [18].
In this paper, we propose and prove generalizations of inequality (1) by
determining the sequence of the polynomial approximations. Also, an improvement
of inequality (2) is given for some intervals. The proposed generalizations and
improvements are based on the double-sided TAYLOR’s approximations and the
corresponding results presented in [15].

2 An Overview of the Results Related to Double-Sided


TAYLOR’s Approximations

Let us consider a real function f : (a, b) −→ R, such that there exist finite limits
f (k) (a+) = lim f (k) (x), for k = 0, 1, . . . , n.
x→a+
TAYLOR’s polynomial

f, a+

n
f (k) (a+)
Tn (x) = (x − a)k , n ∈ N0 ,
k!
k=0

and the polynomial




⎨ T f, a+ (x) +
1 f, a+
Rn (b−)(x − a)n , n ≥ 1
f ; a+, b− n−1
(b − a)n
Tn (x) =


f (b−) , n = 0,

are called the first TAYLOR’s approximation for the function f in the right
neighborhood of a, and the second TAYLOR’s approximation for the function f
in the right neighborhood of a, respectively.
Double-Sided Taylor’s Approximations and Their Applications in Theory of. . . 161

Also, the following functions:

f, a+ f, a+
Rn (x) = f (x) − Tn−1 (x), n ∈ N

and
f ; a+, b− f ; a+, b−
Rn (x) = f (x) − T n−1 (x), n ∈ N

are called the remainder of the first TAYLOR’s approximation in the right neighbor-
hood of a, and the remainder of the second TAYLOR’s approximation in the right
neighborhood of a, respectively.
The following Theorem, which has been proved in [19] and whose variants are
considered in [20, 21] and [22], provides an important result regarding TAYLOR’s
approximations.
Theorem 1 ([19], Theorem 2) Suppose that f (x) is a real function on (a, b), and
that n is a positive integer such that f (k) (a+), for k ∈ {0, 1, 2, . . . , n}, exist.
Supposing that f (n) (x) is increasing on (a, b), then for all x ∈ (a, b) the following
inequality also holds :

f, a+ f ; a+, b−
Tn (x) < f (x) < T n (x). (3)

Furthermore, if f (n) (x) is decreasing on (a, b), then the reversed inequality of (3)
holds.
The above theorem is called Theorem on double-sided TAYLOR’s approximations
in [15], i.e. Theorem WD in [10–14].
The proof of the following proposition is given in [15].
Proposition 1 ([15], Proposition 1) Consider a real function f : (a, b) −→ R
such that there exist its first and second TAYLOR’s approximations, for some n ∈ N0 .
Then,
   
f, a+, b− f, a+, b− f, a+
sgn T n (x) − T n+1 (x) = sgn f (b−) − Tn (b) ,

for all x ∈ (a, b).


From the above proposition, as shown in [15], the following theorem directly
follows:
Theorem 2 ([15], Theorem 4) Consider the real analytic functions f : (a, b) −→
R:


f (x) = ck (x − a)k ,
k=0
162 B. Malešević et al.

where ck ∈ R and ck ≥ 0 for all k ∈ N0 . Then,

f, a+ f, a+ f, a+
T0 (x) ≤ . . . ≤ Tn (x) ≤ Tn+1 (x) ≤ . . .
. . . ≤ f (x) ≤ . . .
f ; a+, b− f ; a+, b− f ; a+, b−
. . . ≤ T n+1 (x) ≤ T n (x) ≤ . . . ≤ T 0 (x),

for all x ∈ (a, b).

3 Main Results

3.1 Generalization of Statement 1

Consider the function:





3
, x = 0,


⎨ 8
f (x) = cos x

⎪ 1−

⎪ cos x2
⎩ , x ∈ (0, π ).
x2

First, we prove that f is a real analytic function on [0, π ). Based on the


elementary equality:
cos x x x
1− = 1 + sec − 2 cos ,
cos x2 2 2

and well known power series expansions [23] (formula 1.411):


 (−1)k
cos t = t 2k t ∈ R,
(2k)!
k=0


  
|E2k | π π
sec t = t 2k t ∈ − , ;
(2k)! 2 2
k=0
. 
x π
where Ek are EULER’s numbers [23], for t = ∈ 0, , i.e. for x ∈ [0, π ), we
2 2
have:

 |E2k | − 2(−1)k
f (x) = x 2k−2
22k (2k)!
k=1
Double-Sided Taylor’s Approximations and Their Applications in Theory of. . . 163

i.e.
3 1 2 7 461 16841
f (x) = + x + x4 + x6 + x8 + . . .
8 128 5120 3440640 1238630400

where the power series converges for x ∈ [0, π ).


Further, based on the elementary well-known features of EULER’s numbers Ek ,
we have:

|E2k | − 2(−1)k
c2k−2 = >0 and c2k−1 = 0,
22k (2k)!

for k = 1, 2, . . . .
Finally, from Theorem 2 the following result directly follows.
Theorem 3 For the function



3
, x = 0,

⎨8
f (x) = 1 − cos x

⎪ cos x2
⎪ ⎩ , x ∈ (0, π )
x2

and any c ∈ (0, π ) the following inequalities hold true:

3 f, 0+ f, 0+ f, 0+
= T0 (x) ≤ T2 (x) ≤ . . . ≤ T2n (x) ≤ . . .
8
. . . ≤ f (x) ≤ . . .
 
f ; 0+, c− f ; 0+, c− f ; 0+, c− cos c
≤ T 2m (x) ≤ . . . ≤ T 2 (x) ≤ T 0 (x) = 1 − c
cos
/c2 .
2
(4)
for every x ∈ (0, c), where m, n ∈ N0 .
π
Note that inequalities from Statement 1 can be directly obtained from (4), for c =
2
cos x
1−
3 cos x2 4
< < .
8 x2 π2

Also, Theorem 3 gives a generalization and a sequence of  improvements


 of
π
results from Statement 1. For example, for c = π/2 i.e. for x ∈ 0, we have:
2
 
3 f, 0+ 3 1 2 f ; 0+, π/2− 3 16 3 4
≤ T2 (x) = + x ≤ f (x) ≤ T 2 (x) = + 4 − 2 x2 ≤ .
8 8 128 8 π 2π π2
164 B. Malešević et al.

Using standard numerical methods it is easy to verify:

f,0+
max |R3 (x)| = f (π/2) = 0.01100 . . .
x∈[0,π/2]

and
f ; 0+, π/2−
max |R3 (x)| = f (1.14909 . . .) = 0.00315 . . . .
x∈[0,π/2]

3.2 An Improvement of Statement 2

Let β ∈ (0, π ) be a fixed real number. Consider the function:



1

⎪ , x = 0,


4
g(x) = sin x
⎪ 1 − sin x


⎩ 2
, x ∈ (0, β].
x2
We prove that g is a real analytic function on [0, β].
Notice that

g(x) = g1 (x) − g2 (x) (5)

for x ∈ [0, β], where





1
, x = 0,
⎨ 4
g1 (x) =

⎪ cosh x2 − cos x2
⎩ , x ∈ (0, β]
x2
and


⎨ 0 , x = 0,
g2 (x) =
⎩ cosh 2 + cos 2 − 2 , x ∈ (0, β].
x x

x2
Since the functions g1 and g2 are real analytic functions on [0, β], with the
following power series expansions:

 1
g1 (x) = x 4k
24k (4k + 2)!
k=0
Double-Sided Taylor’s Approximations and Their Applications in Theory of. . . 165

and

 1
g2 (x) = x 4k+2 ,
24k+2 (4k + 4)!
k=0

the function g must also be a real analytic function on [0, β].


Also, from Theorem 2 the following results directly follow.
Theorem 4 For all c ∈ (0, π ) the following inequalities hold true:

1 g , 0+ g , 0+ g1 , 0+
= T0 1 (x) ≤ . . . ≤ T4n1 (x) ≤ T4n+4 (x) ≤ . . .
4
. . . ≤ g1 (x) ≤ . . .
g1 ; 0+, c− g1 ; 0+, c− g ; 0+, c−
... ≤ T 4m+4 (x) ≤ T 4m (x) ≤ . . . ≤ T 01 (x) = g1 (c).

for all x ∈ (0, c), where m, n ∈ N0 .


Theorem 5 For all c ∈ (0, π ) the following inequalities hold true:

1 2 g , 0+ g2 , 0+ g2 , 0+
x = T2 2 (x) ≤ . . . ≤ T4n+2 (x) ≤ T4n+6 (x) ≤ . . .
192
. . . ≤ g2 (x) ≤ . . .
g2 ; 0+, c− g2 ; 0+, c− g ; 0+, c− g2 (c) 2
... ≤ T 4m+6 (x) ≤ T 4m+2 (x) ≤ . . . ≤ T 22 (x) = x
c2

for all x ∈ (0, c), where m, n ∈ N0 .


π
Thus, from (5), Theorems 4 and 5, for c = , an improvement of inequalities
2
from Statement 2 are obtained, as shown bellow.
First, for all x ∈ (0, π/2) the following inequalities hold true:
   
1 4 π π 1 2
− 2 g2 x 2 ≤ g(x) ≤ g1 − x
4 π 2 2 192

i.e.
 √   √ 
1 16 π 2 4 π 2 1 2
− 4 cosh + − 2 x 2 ≤ g(x) ≤ 2 cosh − − x .
4 π 4 2 π 4 2 192

It is easy to check
 √ 
4 π 2 1 2 1
cosh − − x ≤
π2 4 2 192 4

√ π 2
4 3e− 8 π √ π
for all x ∈ [δ2 , π/2], where δ2 = 8 + 8e 2 − (π 2 + 8 2)e 4 =
π
0.22525 . . . .
166 B. Malešević et al.

Also,
 √ 
4 √ 1 16 π 2
(2 − 2) ≤ − cosh + − 2 x2
π2 4 π4 4 2

√ π √
2π e 8 π 2 + 16 2 − 32
for all x ∈ [0, δ1 ], where δ1 = 2√ = 1.55456 . . . .
π π
8 ( 2 − 4)e 4 + e 2 + 1

4 Conclusion

In this paper, we showed a way to prove some trigonometric inequalities using the
double-sided TAYLOR’s approximations. The presented approach enabled general-
izations of inequalities (1) i.e. produced sequences of polynomial approximations
of the given trigonometric function f .
Note that Theorem 2 cannot be applied directly to inequality (2) because the
function g has an alternating series expansion. We overcame this obstacle by
representing this function by a linear combination of two functions whose power
series expansions have nonnegative coefficients.
Our approach makes a good basis for the systematic proving of trigonomet-
ric inequalities. Developing general, automated-oriented methods for proving of
trigonometric inequalities is an area our continuing interest [5–7, 10–15, 24–30]
and [18].

Acknowledgements Research of the first and second and third author was supported in part by
the Serbian Ministry of Education, Science and Technological Development, under Projects ON
174032 & III 44006, ON 174033 and TR 32023, respectively.

References

1. D.S. Mitrinović, Analytic Inequalities (Springer, Berlin, 1970)


2. G. Milovanović, M. Rassias (eds.), Topics in special functions III, in Analytic Number Theory,
Approximation Theory and Special Functions, ed. by G.D. Anderson, M. Vuorinen, X. Zhang
(Springer, New York, 2014), pp. 297–345
3. M.J. Cloud, B.C. Drachman, L.P. Lebedev, Inequalities with Applications to Engineering
(Springer, Cham, 2014)
4. C. Mortici, The natural approach of Wilker-Cusa-Huygens inequalities. Math. Inequal. Appl.
14(3), 535–541 (2011)
5. B. Malešević, M. Makragić, A method for proving some inequalities on mixed trigonometric
polynomial functions. J. Math. Inequal. 10(3), 849–876 (2016)
6. M. Makragić, A method for proving some inequalities on mixed hyperbolic-trigonometric
polynomial functions. J. Math. Inequal. 11(3), 817–829 (2017)
7. T. Lutovac, B. Malešević, C. Mortici, The natural algorithmic approach of mixed
trigonometric-polynomial problems. J. Inequal. Appl. 2017(116), 1–16 (2017)
Double-Sided Taylor’s Approximations and Their Applications in Theory of. . . 167

8. B. Malešević, M. Rašajski, T. Lutovac, Refinements and generalizations of some inequalities


of Shafer-Fink’s type for the inverse sine function. J. Inequal. Appl. 2017(275), 1–9 (2017)
9. H. Alzer, M.K. Kwong, On Jordan’s inequality. Period. Math. Hung. 77(2), 191–200 (2018)
10. B. Malešević, T. Lutovac, M. Rašajski, C. Mortici, Extensions of the natural approach to
refinements and generalizations of some trigonometric inequalities. Adv. Differ. Equ. 2018(90),
1–15 (2018)
11. M. Rašajski, T. Lutovac, B. Malešević, Sharpening and generalizations of Shafer-Fink and
Wilker type inequalities: a new approach. J. Nonlinear Sci. Appl. 11(7), 885–893 (2018)
12. M. Rašajski, T. Lutovac, B. Malešević, About some exponential inequalities related to the sinc
function. J. Inequal. Appl. 2018(150), 1–10 (2018)
13. T. Lutovac, B. Malešević, M. Rašajski, A new method for proving some inequalities related to
several special functions. Results Math. 73(100), 1–15 (2018)
14. M. Nenezić, L. Zhu, Some improvements of Jordan-Steckin and Becker-Stark inequalities.
Appl. Anal. Discrete Math. 12, 244–256 (2018)
15. B. Malešević, M. Rasajski, T. Lutovac, Double-sided Taylor’s approximations and their
applications in theory of analytic inequalities, in Differential and Integral Inequalities, ed. by
D. Andrica, T. Rassias. Springer Optimization and Its Applications, vol. 151 (Springer, 2019),
pp. 569–582. https://doi.org/10.1007/978-3-030-27407-8_20
16. J. Sándor, On D’aurizio’s trigonometric inequality. J. Math. Inequal. 10(3), 885–888 (2016)
17. J. D’Aurizio, Refinements of the Shafer-Fink inequality of arbitrary uniform precision. Math.
Inequal. Appl. 17(4), 1487–1498 (2014)
18. B.D. Banjac, System for automatic proving of some classes of analytic inequalities. Doctoral
dissertation (in Serbian), School of Electrical Engineering, Belgrade, May 2019. Available on:
http://nardus.mpn.gov.rs/
19. S.-H. Wu, L. Debnath, A generalization of L’Hospital-type rules for monotonicity and its
application. Appl. Math. Lett. 22(2), 284–290 (2009)
20. S.-H. Wu, H.M. Srivastva, A further refinement of a Jordan type inequality and its applications.
Appl. Math. Comput. 197, 914–923 (2008)
21. S.-H. Wu, L. Debnath, Jordan-type inequalities for differentiable functions and their applica-
tions. Appl. Math. Lett. 21(8), 803–809 (2008)
22. S.-H. Wu, H.M. Srivastava, A further refinement of Wilker’s inequality. Integral Transforms
Spec. Funct. 19(9–10), 757–765 (2008)
23. I.S. Gradshteyn, I.M Ryzhik, Table of Integrals Series and Products, 8th edn. (Academic Press,
San Diego, 2015)
24. B. Banjac, M. Nenezić, B. Malešević, Some applications of Lambda-method for obtaining
approximations in filter design, in Proceedings of 23-rd TELFOR Conference, Beograd, 2015,
pp. 404–406
25. M. Nenezić, B. Malešević, C. Mortici, New approximations of some expressions involving
trigonometric functions. Appl. Math. Comput. 283, 299–315 (2016)
26. B. Banjac, M. Makragić, B. Malešević, Some notes on a method for proving inequalities by
computer. Results Math. 69(1), 161–176 (2016)
27. B. Malešević, I. Jovović, B. Banjac, A proof of two conjectures of Chao-Ping Chen for inverse
trigonometric functions. J. Math. Inequal. 11(1), 151–162 (2017)
28. B. Malešević, T. Lutovac, B. Banjac, A proof of an open problem of Yusuke Nishizawa for a
power-exponential function. J. Math. Inequal. 12(2), 473–485 (2018)
29. B. Malešević, M. Rašajski, T. Lutovac, Refined estimates and generalizations of inequalities
related to the arctangent function and Shafer’s inequality. Math. Probl. Eng. 2018, Article ID
4178629, 1–8
30. B. Malešević, T. Lutovac, B. Banjac, One method for proving some classes of exponential
analytical inequalities. Filomat 32(20), 6921–6925 (2018)
The Second Moment of the First
Derivative of Hardy’s Z-Function

Maxim A. Korolev and Andrei V. Shubin

Abstract We give a new estimate of the error term in the asymptotic formula for the
second moment of first derivative of Hardy’s function Z(t). This estimate improves
the previous result of R.R. Hall.

Keywords Riemann zeta-function · Hardy’s Z-function · Approximate


functional equation · Second moment · Third derivative test

1 Introduction

We use the following notation. For t >* 0, let ϑ(t)  +be an increment of any
fixed continuous branch of the function arg π − s/2 Γ s/2 along the segment with
endpoints s = 0.5 and s = 0.5 + it. Hardy’s function Z(t) is defined as
 
Z(t) = eiϑ(t) ζ 0.5 + it .

It is known that Z(t) is real for real t and its real zeros coincide with the ordinates
of zeros of ζ (s) lying on the critical line (see, for example, [1, Ch. III, §4]).
One of the significant branches
 in the theory of Riemann zeta function deals with
mean values of the functions ζ 0.5 + it , Z(t) and its derivatives. For example, it is
known that
 T  
T
Z 2 (t)dt = T P1 ln + E(T ),
0 2π

M. A. Korolev ()
Steklov Mathematical Institute of Russian Academy of Sciences, Moscow, Russia
e-mail: [email protected]
A. V. Shubin
Department of Mathematics and Statistics, McGill University, Montreal, QC, Canada
e-mail: [email protected]

© Springer Nature Switzerland AG 2020 169


A. Raigorodskii, M. T. Rassias (eds.), Trigonometric Sums and Their Applications,
https://doi.org/10.1007/978-3-030-37904-9_9
170 M. A. Korolev and A. V. Shubin

where P1 (u) = u+2γ0 −1, γ0 is the Euler constant and E(T ) is the error term which
has a long history of exploration (see, for example, [2, Ch. XV]). Its best present
estimate belongs to J. Bourgain and N. Watt [3] and has the form E(T ) T α+ε ,
4816 = 0.314576 . . . .
α = 1515
Studying the value distribution of Z(t) at the points of local extremum, R.R. Hall
[4] obtained the following asymptotic formula for the second moment of kth
derivative of Hardy’s function:
 T*    
+2 T T
Z (k)
(t) dt = k P2k+1 ln +O T 3/4
(ln T )2k+1/2 , T → +∞,
0 4 (2k + 1) 2π
(1)

where k ≥ 1 is any fixed integer and P2k+1 (u) is a polynomial of degree 2k + 1


whose leading coefficient equals to 1.
One of the main tools for studying the behavior of the Riemann zeta function
ζ (s) on the critical line Re s = 12 is the following Riemann-Siegel approximate
functional equation


m
cos (ϑ(t) − t ln n) 0 1
Z(t) = 2 √ + R(t), m= t/(2π ) . (2)
n
n=1

The error term R(t) in (2) obeys the estimate R(t) t −1/4 ; moreover, it has the
asymptotic expansion of the form

 −1/4   −1/2  −r/2 


t t t
R(t) = (−1)m−1 H0 (t)+ H1 (t)+ . . . + Hr (t) +
2π 2π 2π
 − (r+1)/2 
+ Or t , (3)

where r ≥ 0 is any fixed integer. The functions Hj (t) are expressed as linear
combinations of the values of the function
 
cos π 12 z2 − z − 18
Φ(z) = (4)
cos π z
*√ +
and its derivatives at the point z = 2α = 2 t/(2π ) .
The expansion (2), (3) was discovered by B. Riemann. Its complete proof
was reconstructed by C.L. Siegel, who used Riemann’s drafts from Göttingen
University’s library, and published in [5] in 1932 (for the history of this question,
see [6, Ch. 7]).
First analogues of the approximate functional equation (2) for the derivatives
Z (k) (t), k = 1, 2, . . . were obtained by A.A. Karatsuba [7] and A.A. Lavrik [8]
who studied the gaps between consecutive zeros of Z (k) (t). However, this problem
does not require an explicit form of the error term R(t). These authors obtained
The Second Moment of the First Derivative of Hardy’s Z-Function 171

the estimate R(t) (1.5 ln t)k t − 1/4 , which is uniform over k, and this bound was
sufficient for their purposes.
In [9], the full analogues of the expansion (2), (3) for the derivatives Z (k) (t),
k = 1, 2 of Hardy’s function, were obtained. These expansions allow one to specify
significantly the well-known formulas for the sums
  
Z(tn ), ζ (0.5 + itn ), Z(tn )Z(tn+1 ),
n≤N n≤N n≤N

where tn is the sequence of Gram points (for the definition and basic properties of
Gram points, see, for example, [6, Ch. 6, §5].1 ) Moreover, the following interesting
effect was found in [9]: it appears that first k terms in the expression (3) for Z (k) (t)
are equal to zero in the cases k = 1 and k = 2. It seems that the same fact is true for
any k ≥ 3.
In this paper, we give another application of the approximate formula for Z (t)
which yields a new estimate of the error term in (1). Our main result is the following
Theorem If T → +∞ then
 T*    
+2 T T
Z (t) dt = P3 ln +O T 5/12
(ln T )3 ,
0 12 2π

where P3 (u) = u3 + c2 u2 + c1 u + c0 ,

c0 = 12γ0 + 24γ1 + 24γ2 − 6 = − 7.053561 . . . ,


c1 = 6 − 12γ0 − 24γ1 = 6.820992 . . . ,
c2 = 6γ0 − 3 = 0.463294 . . . ,

γk are Stieltjes constants defined by the expansion

+∞

1 (−1)m
ζ (s) = + γm (s − 1)m .
s−1 m!
m=0

2 Auxiliary Lemmas

In this section, we give some lemmas which are necessary for the proof of the main
theorem.

1 In [6], another notation for Gram points is used: gn instead of tn . The corresponding results of the
first author concerning the above sums will appear soon.
172 M. A. Korolev and A. V. Shubin

Lemma 1 If x → +∞, then


 1  
1 "(x) 1
= ln x + γ0 + √ + O ,
√ n 2 x x
n≤ x
 ln n  
1 "(x) ln x
= ln2 x + γ1 + √ ln x + O ,
√ n 8 2 x x
n≤ x
 
 (ln n)2 1 3 "(x) ln2 x
= ln x + γ2 + √ ln2 x + O ,
√ n 24 4 x x
n≤ x

where "(x) = 0.5 − {x}.


These expansions follow directly from Euler’s summation formula (see, for
example, [1, Appendix, § 1, Th. 2]).
Lemma 2 Let b − a > 1, f ∈ C 3 [a, b] and let λ3 ≤ |f (3) (x)| ≤ hλ3 for some
λ3 > 0, h ≥ 1. Then
 √ 1/6 √ −1/6
exp (2π if (x)) h(b − a)λ3 + b − a λ3 ,
a<x≤b

where the implied constant is absolute.


This is a particular case of van der Corput’s theorem (see [1, Appendix, §11, Th.
2]).
Lemma 3 Let c < b and suppose that real functions f (t), g(t) satisfy the following
conditions on [c, b]: (1) the derivatives g (t) and f (4) (t) are continuous; (2)
f (t) > 0; (3) f (c) = 0. Then the integral
 b
I= g(t)e2π if (t) dt
c

is expressed as follows:
1
I =A−B + (C + D − E + H − K) ,
2π i
where

1+i g(c)e2π if (c) g(c)e2π if (c) +∞ e2π iu
A= · , B= √ du, λ = f (b) − f (c),
2 2f (c) 2f (c) λ u
 
g(b) g(c) 1 g(c) f (3) (c) 2π if (c)
C= − · √ e 2π if (b)
, H = · e ,
f (b) 2f (c) λ 3 (f (c))2

g (c) 2π if (c)
K= e ,
f (c)
The Second Moment of the First Derivative of Hardy’s Z-Function 173

and D, E denote the integrals along the segment [c, b] with the integrands

 
f (t) |f (t)|
g(c) − e2π if (t) ,
(f (t))2 8f (c)(f (t) − f (c))3/2
g (t)f (t) − (g(t) − g(c))f (t) 2π if (t)
e .
(f (t))2

This is a particular case of lemma 7 from [10].


Lemma 4 The following equality holds: ϑ(t) = θ (t) + δ(t), where

t t t π
θ (t) = ln − −
2 2π 2 8

and δ(t) is a smooth function such that


   
1 1 1 1
δ(t) = +O , δ (t) = − 2 + O , as t → +∞.
48t t3 48t t4

These relations follow from the asymptotic expansion for ϑ(t) obtained by
C.L. Siegel [5, formula (43)].
Lemma 5 If t → +∞ then
 1  
Z (t) = −2 √ ϑ (t)− ln n sin (ϑ(t)−t ln n)
n≤m
n
 −3/4  
t Φ(2α)
+(−1) m−1
+O t −5/4 , (5)
2π 2π
0√ 1 *√ +
where m = t/(2π ) , α = t/(2π ) , and the function Φ(z) is defined by (4).
This approximate functional equation for Z (t) is a corollary of theorem 2 from
[9].

3 Proof of the Main Theorem

Let p(t), q(t) and r(t) denote the terms in the right hand side of (5). Then the
integral I (T ) is expressed as follows:

I (T ) = I1 + I2 + I3 + 2 (I4 + I5 + I6 ) ,
174 M. A. Korolev and A. V. Shubin

Here I1 , . . . , I6 denote the integrals of the functions p2 (t), q 2 (t), r 2 (t), q(t)r(t),
p(t)r(t) and p(t)q(t), respectively. Then we have
 T  − 3/2
1 t
I2 = Φ 2 (2α) dt 1,
4π 2 2π 2π
 T  T
− 5/2
I3 t dt 1, I4 t − 3/4 · t − 5/4 dt 1.
2π 2π

Further,

 
T T  
I5 = p(t)r(t)dt = Z (t) − q(t) − r(t) r(t)dt =
2π 2π
 T  T

= Z (t)r(t)dt − I3 − I4 |Z (t)|t − 5/2 dt + 1.
2π 2π

Splitting the interval of integration by points 2k , k = 1, 2, . . . to the segments of the


form τ ≤ t ≤ τ1 , τ1 ≤ 2τ , and using (1) we get:

 τ1  τ1   τ1 1/2
− 5/4 − 5/4 − 5/4
|Z (t)|t dt τ |Z (t)|dt τ τ |Z (t)| dt
2

τ τ τ
 1/2
τ − 5/4 τ 2 (ln τ )3 τ − 1/4 (ln τ ) 3/2 ,

whence
 T  
|Z (t)|t − 5/4 dt τ − 1/4 (ln τ ) 3/2 2− k/4 k 3/2 1, I5 1.
2π τ k≥1

Similarly,

 
T T  
I6 = p(t)q(t)dt = Z (t) − q(t) − r(t) q(t)dt
2π 2π
 T 
|Z (t)|t − 3/4 dt + 1 τ 1/4 (ln τ )3/2 T 1/4 (ln T )3/2 .
2π τ

Thus,
 
I (T ) = I1 + O T 1/4
(ln T )3/2 .
The Second Moment of the First Derivative of Hardy’s Z-Function 175

To calculate I1 , we note that the error arising from the replacement of ϑ (t) by
θ (t) = 12 ln (t/(2π )) in the expression
 1  
p(t) = −2 √ ϑ (t) − ln n sin (ϑ(t) − t ln n) (6)
n≤m
n

is estimated as

 
m 
T 1  
−4 δ (t)p(t) √ sin ϑ(t) − t ln n dt −
2π n
n=1
  m 
T

2  1   2
−4 δ (t) √ sin ϑ(t) − t ln n dt
2π n
n=1
   T
T T 
|p(t)|t − 7/4
dt + t − 7/2
dt Z (t)−q(t)−r(t)t − 7/4 + 1 1.
2π 2π 2π

Further, we replace ϑ(t) by θ (t) in (6). One can check that


       
sin ϑ(t) − t ln n − sin θ (t) − t ln n = 2 sin 0.5δ(t) cos θ (t)−t ln n+0.5δ(t) =
 *        +
= 2 sin 0.5δ(t) cos θ (t) − t ln n cos 0.5δ(t) − sin θ (t) − t ln n sin 0.5δ(t) =
     
= 2 sin 0.5δ(t) cos θ (t) − t ln n + O(δ 2 (t)) = δ(t) cos θ (t) − t ln n + O(δ 2 (t)) =
1    
= cos θ (t) − t ln n + O t −2 .
48t
Setting


m  
ϕn 1
ϕn = ϕn (t) = θ (t) − t ln n, s(t) = −2 √ sin ϕn + cos ϕn ,
n 48t
n=1

we get
 T
I1 = I7 + O (1) , I7 = s 2 (t)dt.

Next, we have

s 2 (t) =
 
ϕn (t)ϕk (t)
m
1
4 √ sin ϕn sin ϕk + (sin ϕn cos ϕk + cos ϕn sin ϕk )
kn 48t
k,n=1

1
+ cos ϕn cos ϕk =
(48t)2
176 M. A. Korolev and A. V. Shubin

  
ϕn (t)ϕk (t)
m
1
2 √ cos(ϕn − ϕk ) − cos(ϕn + ϕk ) + sin(ϕn + ϕk )
kn 24t
k,n=1
 
+ O t − 3/2 ln2 t . (7)

Hence,

I7 = J1 + J2 + J3 + O(1),

where J1 , J2 and J3 denote the contributions arising from the first, second and third
term in right-hand side of (7).
Let J1,1 and J1,2 denote the contributions to J1 coming from the terms with n = k
and n = k, respectively. Then

 m T  2
1 1 t
J1,1 =2 ln − ln n dt =
2π n=1 n 2 2π
 T   m   m 
t 2 1 t  ln n  (ln n)2
m
1
=2 ln − ln + dt.
2π 4 2π n 2π n n
n=1 n=1 n=1

Applying lemma 1 with x = t/(2π ), we conclude that the integrand equals to


       
1 t 3 γ0 t 2 t ln2 t
ln + ln − γ1 ln + γ2 + O .
24 2π 4 2π 2π t

The integration over t yields


 
T T
J1,1 = P3 ln + O(ln3 T ).
12 2π

Next, we have
  
T ϕn (t)ϕk (t) it ln (n/k) jn,k
J1,2 = 4 Re √ e dt = 4 Re √ , (8)
2π 1≤k<n≤m
nk 1≤k<n≤P
nk

where
 T   
1 t 1 t
jn,k = ln − ln n ln − ln k eit ln (n/k) dt, P = T /(2π ).
2π n2 2 2π 2 2π
The Second Moment of the First Derivative of Hardy’s Z-Function 177

By taking t = 2π τ 2 and integrating jn,k by parts twice, we obtain

 
P  
τ τ 2π iτ 2 ln (n/k)
jn,k = 4π τ ln ln e dτ =
n n k
    
n −1 P P 2π iP 2 ln (n/k)
= −i ln ln ln e
k n k
   
1 n −2 P P 2π iP 2 ln (n/k)
+ ln ln + ln e −
4π P 2 k n k
 
1 n −1 2π in2 ln (n/k)
− ln e +
4π n2 k
    
1 n −2 P 2π iτ 2 ln (n/k) τ τ dτ
+ ln e ln + ln − 1 3 =
2π k n n k τ
= −iAn,k + Bn,k − Cn,k + Dn,k ,

where the notation is obvious. The following inequalities hold:


     
1 n −2 P P 1 n −2
|Bn,k | ≤ ln ln + ln ≤ ln ln P ,
4π n2 k n k 2π n2 k
   
1 n −1 1 n −2
|Cn,k | ≤ ln < ln ln P ,
4π n2 k 4π n2 k
      
1 n −2 P τ τ dτ 1 n −2
|Dn,k | ≤ ln ln + ln − 1 3 < ln .
2π k n n k τ 2π n2 k

Consequently, the contribution to the sum on the right hand side of (8) coming from
Bn,k , Cn,k and Dn,k does not exceed

   
1 1 n −2
(ln P ) √ ln
n5/2 k k
1<n≤P 1≤k≤n−1

     2 
1 1 1 n
(ln P ) √ + √ (ln T )2 .
1<n≤P
n5/2
1≤k≤n/2
k 1≤r≤n/2
n − r r

The contribution to (8) from An,k is expressed as

     
1 n −1 P P 2π iP 2 ln (n/k)  
4 Im √ ln ln ln e = 4 Im Σ1 + Σ2 ,
nk k n k
1≤k<n≤P
178 M. A. Korolev and A. V. Shubin

where Σ1 and Σ2 stand for terms with the conditions 1 ≤ k ≤ n/2 and n/2 < k ≤
n − 1, respectively. Changing the order of summation in Σ1 , we have

   
1 P −2π if (k)
Σ1 = √ ln e Sk , Sk = g(n)e2π if (n) ,
k k
1≤k≤P /2 2k<n≤P

where
  
1 P x −1
g(x) = √ ln ln , f (x) = P 2 ln x.
x x k

Splitting the range of summation over n by points 2−s P , s = 1, 2, . . . to the


segments of the form N < n ≤ N1 , where N1 ≤ 2N , we find
 
Sk = Sk (N ), Sk (N ) = g(n)e2π if (n) .
s N <n≤N1

Since g(x) N −1/2 ln P , g (x) N −3/2 ln P for N ≤ x ≤ N1 , Abel summation


formula yields
 
ln P   2π if (n) 
Sk (N ) √  e ,
N N <n≤N
2

where N < N2 ≤ N1 . Obviously, |f (3) (x)|  P 2 N −3 for N ≤ x ≤ N2 . Then the


application of lemma 2 gives

ln P √ 
Sk (N ) √ NP 1/3 + NP −1/3 P 1/3 ln P , Sk P 1/3 (ln P )2 ,
N
Σ1 P 5/6 (ln P )3 T 5/12
(ln T )3 .

Further, setting k = n − r, where 1 ≤ r ≤ n/2, we write the sum Σ2 in the form


 Sr 
Σ2 = , Sr = gr (n)e2π ifr (n) ,
r
1≤r<P /2 2r<n≤P

where
   −1
r P P x x
gr (x) = √ ln ln ln , fr (x) = P 2 ln .
x(x − r) x x−r x−r x−r
The Second Moment of the First Derivative of Hardy’s Z-Function 179

Using the same arguments as above, we find


Sr Sr (N ), N = 2−s P , s = 1, 2, . . . ,
s
 
   
Sr (N ) = gr (n)e2π ifr (n) (ln P )2  e2π ifr (n) , N < N3 ≤ N1 .
N <n≤N1 N <n≤N3

Since |fr (x)|  rP 2 N −4 for N ≤ x ≤ N3 , lemma 2 yields


(3)

  2 1/6  4 1/6 
P r √ N
Sr (N ) (ln P )2 N 4
+ N
N P 2r
 
(N P )1/3 r 1/6 + P −1/3 N 7/6 r − 1/6 (ln P )2 ,
 
Sr P 2/3 r 1/6 + P 5/6 r − 1/6 (ln P )2 ,
  
Σ2 (ln P )2 P 2/3 r − 5/6 + P 5/6 r − 7/6 T 5/12
(ln T )2 .
1≤r≤P /2

Therefore, J1,2 T 5/12 (ln T )2 .


Further, we express the integral
  ϕn (t)ϕk (t)
T m
J2 = 2 √ cos (ϕn (t) + ϕk (t)) dt
2π k,n=1
nk

as a sum J2,1 + J2,2 , where J2,1 and J2,2 denote the contributions coming from the
pairs n, k with the conditions n = k and n = k, respectively. Then
 T
m
(ϕn (t))2  
   j (n) 
− π i/4
J2,1 = 2 cos 2ϕn (t) dt = 2 Re e .
2π n n
n=1 n≤P
(9)
Taking t = 2π τ 2 , we get
    
P x 2 x 1
j (n) = g(τ )e2π if (τ ) dτ, g(x) = x ln , f (x) = 2x 2 ln − .
n n n 2

Since g(n) = g (n) = 0, the application of lemma 3 to j (n) yields: j (n) = (C(n) −
E(n))/(2π i), where
   P
1 P 2π if (P ) 1 dτ
C(n) = ln e , E(n) = e2π if (τ ) .
4 n 4 n τ
180 M. A. Korolev and A. V. Shubin

Therefore,

1 P
|j (n)| ≤ ln , J2,1 (ln T )2 .
4π n
Using the same arguments as above, we write J2,2 as

  
−π i/4 jn,k
4 Re e √ ,
1<n≤P
nk
 T   
1 t 1 t
jn,k = ln − ln n ln − ln k eit (ln (t/(2π ))−ln kn−1) dt =
2π n2 2 2π 2 2π
 P   
τ τ 2π if (τ )
= 4π τ ln ln e dτ,
n n k

where f (τ ) = τ 2 (2 ln τ − ln kn − 1). Integrating by parts twice, we get: jn,k =


−iAn,k + Bn,k − Cn,k − Dn,k , where

 −1   
P2 P P 2π if (P )
An,k = ln ln ln
e ,
kn n k
      
1 P 2 −3 P 2 P 2 2π if (P )
Bn,k = ln ln + ln e ,
4π P 2 kn n k
 
1 n −1 2π if (n)
Cn,k = ln e ,
4π n2 k
 P 2
1 ln (τ/n) + ln2 (τ/k)
Dn,k =  3 +
2π n ln (τ/n) + ln (τ/k)

ln2 (τ/n) + ln2 (τ/k) + ln (τ/n) ln (τ/k) e2π if (τ )
+ 2·  4 dτ.
ln (τ/n) + ln (τ/k) τ3

Next, one can check that


   
1 P P −1 1 n −1
|Bn,k | ≤ ln + ln ≤ ln ,
4π P 2 n k 4π n2 k
      
1 n −1 1 n −1 n −2
|Cn,k | ≤ ln , |D n,k | ≤ ln + ln .
4π n2 k 4π n2 k k

Hence, the contribution arising from the terms Bn,k , Cn,k and Dn,k in the sum over
n in right hand side of (9) does not exceed
The Second Moment of the First Derivative of Hardy’s Z-Function 181

   −1   
1 1 n n −2
√ ln + ln 1.
n5/2 k k k
1<n≤P 1≤k≤n−1

Further, the contribution coming from An,k to (9) is expressed as follows:

 
π i/4+2π iP 2 (2 ln P −1)
− 2 Re e Σ ,

 1 ln (P /k) ln (P /n) 2π iP 2 ln (kn)


Σ = √ e = Σ3 + Σ4 ,
1≤k<n≤P
kn ln (P /k) + ln (P /n)

where Σ3 and Σ4 are responsible for terms with the conditions 1 ≤ k ≤ n/2 and
n/2 < k ≤ n − 1, respectively. Taking k = n − r in Σ4 and changing the order of
summation in both sums we get

  
1 P 2π if (k) 
Σ3 = √ ln e g(n)e2π if (n) ,
k k
1≤k<P /2 2k<n≤P
 
Σ4 = gr (n)e2π ifr (n) ,
1≤r<P /2 2r<n≤P

where

ln (P /x) 1
g(x) = √ , f (x) = P 2 ln x,
ln (P /x) + ln (P /k) x
ln (P /x) ln (P /(x − r)) 1
gr (x) = √ , fr (x) = P 2 (ln x + ln (x − r)).
ln (P /x) + ln (P /(x − r)) x(x − r)

One can check that

ln P ln P
g(x) √ , g (x) , |f (3) (x)|  P 2 x −3 for 2k ≤ x ≤ P ,
x x 3/2
ln P ln P
gr (x) , gr (x) , |fr(3) (x)|  P 2 x −3 for 2r ≤ x ≤ P .
x x2
Estimating Σ3 , Σ4 similarly to Σ1 , Σ2 , we finally get

Σ3 , Σ4 T 5/12
(ln T )3 .

Theorem is proved.
182 M. A. Korolev and A. V. Shubin

Remark Of course, the exponent 5/12 in our theorem is not the best one. It can be
decreased by different ways. First of all, one can apply one-dimensional method of
exponent pairs to the estimation of the sums Sk (N ), Sr (N ) defined above (see [11,
Ch. 3]). Next, it is possible to apply two-dimensional method of exponent pairs to
the sums Σj , 1 ≤ j ≤ 4, to take into account the oscillation over both variables
of summation (see [12–16]). Finally, one can try to derive an explicit formula of
Atkinson type similar to well-known formula for the remainder E(T ) (see [17, Ch.
2]). Each of these approaches requires a long paper. Since our main purpose was
only to demonstrate one more application of the functional equation for Z (t), we
limited ourselves by the simplest estimates.

References

1. A.A. Karatsuba, S.M. Voronin, The Riemann Zeta-Function (Walter de Gruyter, Berlin/New-
York, 1992)
2. A. Ivić, The Riemann Zeta-Function. Theory and Applications, 2nd edn. (Dover Publication,
Mineola/New York, 2003)   
3. J. Bourgain, N. Watt, Decoupling for perturbed cones and mean square of ζ 12 + it .
arXiv:1505.04161 [math.NT]
4. R.R. Hall, The behaviour of the Riemann zeta-function on the critical line. Mathematika 46(2),
281–313 (1999)
5. C.L. Siegel, Über Riemanns Nachlaß zur analytischen Zahlentheorie. Quellen und Studien zur
Geshichte der Mathematik. Astronomie und Physik 2, 45–80 (1932); (see also: C.L. Siegel,
Gesammelte Abhandlungen, B. 1. K. Chandrasekharan, H. Maass (eds.), Springer, Berlin,
1966, 275–310)
6. H.M. Edwards, Riemann’s Zeta Function (Dover Publication, Mineola/New York, 2001)
7. A.A. Karatsuba, On the distance between consecutive zeros of the Riemann zeta function that
lie on the critical line. Trudy Mat. Inst. Steklov. 157, 49–63 (1981), (Russian); Proc. Steklov
Inst. Math. 157, 51–66 (1983) (English)
8. A.A. Lavrik, Uniform approximations and zeros in short intervals of the derivatives of the
Hardy’s Z-function. Anal. Math. 17(4), 257–279 (1991) (Russian)
9. M.A. Korolev, On Rieman-Siegel formula for the derivatives of Hardy’s function. Algebra i
Analiz, 29(4), 53–81 (2017) (Russian); St. Petersburg Math. J. 29(4), 581–601 (2018)
10. M.A. Korolev, On the integral of Hardy’s function Z(t). Izv. RAN. Ser. Mat. 72(3), 19–68
(2008) (Russian); Izv. Math. 72(3), 429–478 (2008) (English)
11. S.W. Graham, G. Kolesnik, Van der Corput’s Method of Exponential Sums. Lecture Notes
Series, vol. 126 (Cambridge University Press, Cambridge, 1991)
12. E.C. Titchmarsh, On Epstein’s zeta-function. Proc. Lond. Math. Soc. 36(2), 485–500 (1934)
13. E.C. Titchmarsh, The lattice-points in a circle. Proc. Lond. Math. Soc. 38(2), 96–115 (1934)
14. B.R. Srinivasan, The lattice point problem of many-dimensional hyperboloids II. Acta Arith.
8(2), 173–204 (1963)
15. B.R. Srinivasan, The lattice point problem of many-dimensional hyperboloids III. Math. Ann.
16, 280–311 (1965)
16. G. Kolesnik, On the method of exponential pairs. Acta Arith. 45(2), 115–143 (1985)
17. A. Ivić, Lectures on Mean Values of the Remann Zeta Function (Tata Institute of Fundamental
Research, Bombay, 1991)
Dedekind and Hardy Type Sums and
Trigonometric Sums Induced by
Quadrature Formulas

Gradimir V. Milovanović and Yilmaz Simsek

Abstract The Dedekind and Hardy sums and several their generalizations, as
well as the trigonometric sums obtained from the quadrature formulas with the
highest (algebraic or trigonometric) degree of exactness are studied. Beside some
typical trigonometric sums mentioned in the introductory section, the Lambert and
Eisenstein series are introduced and some remarks and observations for Eisenstein
series are given. Special attention is dedicated to Dedekind and Hardy sums, as
well as to Dedekind type Daehee-Changhee (DC) sums and their trigonometric
representations and connections with some special functions. Also, the reciprocity
law of the previous mentioned sums is studied. Finally, the trigonometric sums
obtained from Gauss-Chebyshev quadrature formulas, as well as ones obtained from
the so-called trigonometric quadrature rules, are considered.

Keywords Trigonometric sums · Dedekind sums · Hardy sums · Eisenstein


series · Gauss-Chebyshev quadrature sums · Degree of exactness

1 Introduction and Preliminaries

Trigonometric sums play very important role in many various branches of mathe-
matics (number theory, approximation theory, numerical analysis, Fourier analysis,
etc.), physics, as well as in other computational and applied sciences. Inequalities
with trigonometric sums, in particular their positivity and monotonicity are also
important in many subjects (for details see [63, Chap. 4] and [66]).

G. V. Milovanović ()
The Serbian Academy of Sciences and Arts, Belgrade, Serbia
Faculty of Sciences and Mathematics, University of Niš, Niš, Serbia
e-mail: [email protected]
Y. Simsek
Faculty of Arts and Science, Department of Mathematics, Akdeniz University,
Antalya, Turkey
e-mail: [email protected]

© Springer Nature Switzerland AG 2020 183


A. Raigorodskii, M. T. Rassias (eds.), Trigonometric Sums and Their Applications,
https://doi.org/10.1007/978-3-030-37904-9_10
184 G. V. Milovanović and Y. Simsek

There are several trigonometric sums in the well-known books [70, 71], [42,
pp. 36–40] and [47]. The famous Dedekind and Hardy sums and many generalized
sums have also trigonometric representations. In this introduction we mention some
typical trigonometric sums obtained lately.
In 2000 Cvijović and Klinowski [26] gave closed form of the finite cotangent
sums


q−1
(ξ + p)π 
q−1

Sn (q; ξ ) = cotn and Sn∗ (q) = cotn ,
q q
p=0 p=1

where n and q are positive integers (q ≥ 2) and ξ is a non-integer real number.


They obtained Sn (q; ξ ) in a determinant form, as well as the following differential
recurrence relation
q d
Sn+2 (q; ξ ) = −Sn (q; ξ ) − Sn+1 (q; ξ ) (n ≥ 1),
π(n + 1) dξ

where

S1 (q; ξ ) = q cot(π ξ ),
S2 (q; ξ ) = q 2 [cot2 (π ξ ) + 1] − q,
0 1
S3 (q; ξ ) = q 3 cot3 (π ξ ) + cot(π ξ ) − q cot(π ξ ),


etc. Evidently, according to the properties of the cotangent function, S2n+1 (q) = 0,
as well as


q−1
pπ 1 ∗ 
q
pπ 1 ∗
cot 2n
= S2n (2q) an cot2n = S2n (2q + 1).
2q 2 2q + 1 2
p=1 p=1

For example, S2∗ (q) = (q 2 − 3q + 2)/3, S4∗ (q) = (q 4 − 20q 2 + 45q − 26)/45,
S6∗ (q) = (2q 6 − 42q 4 + 483q 2 − 945q + 502)/945, etc. In general, S2n ∗ (q) is a

polynomial of degree 2n with rational coefficients [26] (see also [70, p. 646] for
n = 1 and n = 2).
Using contour integrals and the Cauchy residue theorem, Cvijović and Srivastava
[27] derived formulas for general family of secant sums


q−1    
2rpπ pπ
S2n (q, r) = cos 2n
sec (r = 0, 1, . . . , q − 1),
q q
p=0
q2 (q is even)
p=

when n ∈ N and q ∈ N \ {1}, as well as for various special cases including ones for
r = 0, i.e.,
Dedekind and Hardy Type Sums and Trigonometric Sums Induced by. . . 185


q−1  

S2n (q) = sec2n
.
q
p=0

They also obtained sums which were considered earlier by Chen [17] by using the
method of generating functions. In the Appendix of [17], Chen gave tables of power
sums of secant, cosecant, tangent and cotangent. Among various such trigonometric
summation formulae, we mention only a few of them for tangent function:


n−1  

tan2 = n(n − 1),
n
k=1


n−1  
kπ 1
tan 4
= n(n − 1)(n2 + n − 3),
n 3
k=1


n−1  
kπ 1
tan 6
= n(n − 1)(2n4 + 2n3 − 8n2 − 8n + 15).
n 15
k=1

Using their earlier method, Cvijović and Srivastava [28] obtained closed-form
summation formulas for 12 general families of trigonometric sums of the form


n−1    m
2rkπ kπ
(±1) k−1
f g (n ∈ N \ {1}, r = 1, . . . , n − 1),
n n
k=1

for different combinations of the functions x → f (x) and x → g(x) and different
values (even and odd) of m ∈ N. The first function can be f (x) = sin x or
f (x) = cos x, while the second one can be one of the functions cot x, tan x, sec x,
and csc(x). Such a family of cosecant sums. I.e.,


n−1    
2rkπ kπ
C2m (n, r) = cos csc2m ,
n n
k=1

where m ∈ N, n ∈ N \ {1}, and r = 0, 1, . . . , n − 1, was previously considered


by Dowker [30]. All obtained formulas in [28] involve the higher-order Bernoulli
polynomials (see also [23] and [24]).
In [32] da Fonseca, Glasser, and Kowalenko have considered the trigonometric
sums of the form


n−1   
n−1  
kπ kπ
C2m (n) = cos 2m
and S2m (n) = 2m
sin ,
n n
k=0 k=0
186 G. V. Milovanović and Y. Simsek

and their extensions. In [31] da Fonseca and Kowalenko studied the sums of the
form


n  

(−1)k cos2m ,
2n + 2
k=1

where n and m are arbitrary positive integers.


Recently da Fonseca, Glasser, and Kowalenko [33] have presented an elegant
integral approach for computing the so-called Gardner-Fisher trigonometric inverse
power sum

 π 2m 
n−1  

Sm,2 (n) = sec2m , n, m ∈ N.
2n 2n
k=1

For example,
   
π2 1 π4 5 7
S1,2 (n) = 1− 2 and S2,2 (n) = 1+ 2 − 4 .
6 n 90 2n 2n

By using contour integrals and residues, similar results for secant and cosecant sums
were also obtained by Grabner and Prodinger [41] in terms of Bernoulli numbers
and central factorial numbers.
Recently Chu [21] has used the partial fraction decomposition method to get a
general reciprocal theorem on trigonometric sums. Several interesting trigonometric
reciprocities and summation formulae are derived as consequences.
In this chapter, we mainly give an overview of the Dedekind and Hardy sums
and several their generalizations, as well as the trigonometric sums obtained from
the quadrature formulas with the highest (algebraic or trigonometric) degree of
exactness.
The chapter is organized as follows. In Section 2 we introduce Lambert and
Eisenstein series and give some remarks and observations for Eisenstein series.
Sections 3 and 4 are dedicated to Dedekind and Hardy sums. In Section 5 we
consider the Dedekind type Daehee-Changhee (DC) sums. Their trigonometric
representations and connections with some special functions are presented in
Sections 6 and 7, respectively. The Section 8 is devoted to the reciprocity law of the
previous mentioned sums. Finally, in Sections 9 and 10 we consider trigonometric
sums obtained from Gauss-Chebyshev quadrature formulas, as well as ones obtained
from the so-called trigonometric quadrature rules.

2 Lambert and Eisenstein Series

Lambert series Gp (x) is defined by


Dedekind and Hardy Type Sums and Trigonometric Sums Induced by. . . 187


 ∞

xm
Gp (x) = m−p = m−p x mn ,
1 − xm
m=1 m,n=1

where p ≥ 1. These functions are regular for |x| < 1. The special case p = 1 gives

  
G1 (x) = − log 1 − xm .
m=1

For odd integer values of p, Apostol [2] gave the behavior of these functions in the
neighborhood of singularities, using a technique developed by Rademacher [72] in
treating the case p = 1.
The following series

(m + nz)−s ,
(m,n)∈Z2

for Im z > 0 and Re s > 2, has an analytic continuation to all values of s. In the
paper [56] by Lewittes, it is well known this series has transformation formulae
for the analytic continuation of very large class of the Eisenstein series. These
transformation formulae are related to large class of functions which generalized
the case of the Dedekind eta-function, which is given as follows:
Let z = x + iy and s = σ + it with x, y, σ, t be real. For any complex number
w, branch of log w with −π ≤ arg w < π . Let

az + b
V (z) =
cz + d

be an arbitrary modular transformation. Let H denote the upper half-plane,

H = {z : Im (z) > 0} .

For z ∈ H and σ > 2, the Eisenstein series, G (z, s, r1 , r2 ) is defined by

 e2π i(mh1 +nh2 )


G (z, s, r, h) = , (1)
((m + r1 ) z + n + r2 )s
r =(m,n)∈Z2

where r1 , r2 ∈ R.
Substituting r1 = r2 = 0 into Eq. (1), we have
 1
G (z, s) = (2)
(m + nz)s
r =(m,n)∈Z2
188 G. V. Milovanović and Y. Simsek

(for details see [55, 56]). Let r1 and r2 be arbitrary real numbers. For z ∈ H and
arbitrary s, generalization of Dedekind’s eta-function is given by

 
A (z, s, r1 , r2 ) = k s−1 e2π ikr2 +2π ik(m+r1 )z .
m>−r1 k=1

For a real and σ > 1, Lewittes [55, 56] define ζ (s, a) by



ζ (s, a) = (n + a)−s .
n>−a

Observe that

ζ (s, a) = ζ (s, {a} + χ (a)) ,

where {a} denotes the fractional part of a, and χ (a) denotes the characteristic
function of integers. Since 0 < χ (a) + {a} ≤ 1, ζ (s, {a} + χ (a)) denotes
the classical Hurwitz zeta-function. Lewittes ([56, Eq-(18)]) showed a connection
between G (z, s, r1 , r2 ) and A (z, s, r1 , r2 ) as follows
 
G (z, s, r1 , r2 ) = χ (r1 ) ζ (s, r2 ) + eπ is ζ (s, −r2 )

(−2π i)s  
+ A (z, s, r1 , r2 ) + eπ is A (z, s, −r1 , −r2 ) .
Γ (s)

The above equation was proved by Berndt [10]. He proved transformation


formula under modular substitutions which is derived for very large class of
generalized Eisenstein series. Berndt’s results easily converted into a transformation
formula for a large class of functions that includes and generalizes the Dedekind eta-
function and the Dedekind sums.
A transformation formula of the function A (z, s, r1 , r2 ) is given by Apostol [2]
as follows: Let m (> 0) is an even integer. Then

1  
(cz + d)m A (V (z), −m) = A (z, −m) + ζ (m + 1) 1 − (cz + d)m
2
     
m+2 j id
m+1  
c m+2 B k c B m+2−k c
(2π i) k
+ . (3)
2(m + 2)! (−(cz + d))1−k
j =1 k=0

However, due to a miscalculation of residue, the term 12 ζ (m + 1) (1 − (cz + d)m )


was omitted. The result was also misstated by Carlitz [16]. The proof of this
transformation is also given by Lewittes [56] and after that by Berndt [10]. A special
value of the function A(z, s, r1 , r2 ) is given by
Dedekind and Hardy Type Sums and Trigonometric Sums Induced by. . . 189

π iz
log η(z) = − A(z).
12
Hence, the transformation formula for A(z) is given as follows (cf. [5, 44, 53,
54]):
Theorem 1 For z ∈ H we have
 
1
η − = (−iz) η(z).
z

2.1 Further Remarks and Observations for Eisenstein Series

Now we give some standard results about Eisenstein series.


For 2 ≤ k ∈ N and z ∈ H,

 ∞
1 (−2π i)k  k−1 2π inz
= n e
(z + m)k (k − 1)!
m∈Z n=1

is the known Lipschitz formula.


Apostol-Eisenstein series are given as follows:
If 2 ≤ k ∈ N and z ∈ H, the Eisenstein series G (z, 2k) is defined by
 1
G (z, 2k) = . (4)
(mz + n)2k
0 =(m,n)∈Z2

It converges absolutely and has the Fourier expansion



2 (2π i)2k 
G (z, 2k) = 2ζ (2k) + σ2k−1 (n) e2π inz
(2k − 1)!
n=1

where, as usual, σc (n) = d|n d c and ζ (z) denotes Riemann zeta function.
For k = 1, the series in (4); G (z, 2) is no longer absolutely convergent. G (z, 2)
is an even function,


G (z, 2) = 2ζ (2) + 2 (2π i)2 σ (n) e2π inz (5)
n=1

for z ∈ H.
For x = e2π iz the series in (5) is an absolutely convergent power series for |x| <
1 so that G (z, 2) is analytic in H. The behavior of G (z, 2) under the modular group
is given by (cf. [5])
190 G. V. Milovanović and Y. Simsek

 
1
G − , 2 = z2 G (z, 2) − 2π iz.
z

The well-known Lipschitz formula is given by the following lemma:


Lemma 1 (Lipschitz formula) Let 2 ≤ k ∈ N and z ∈ H. Then

 ∞
1 (−2π i)k  k−1 2π inz
= n e .
(z + m)k (k − 1)!
m∈Z n=1

By using this lemma, the Fourier expansion of the Eisenstein series is given by:
Theorem 2 If k is an integer with k ≥ 2 and z ∈ H, then
∞ ∞
2 (−2π i)k   k−1 2π inmz
G (z, k) = 2ζ (k) + n e .
(k − 1)!
m=1 n=1

Proof We give only brief sketch of the proof since the method is well-known. Now
replacing z by az, where a > 0, substituting in Lemma 1 and summing over all
a ≥ 1, we get
∞ 
 ∞
1 (−2π i)k  k−1 2π inaz
= n e .
(az + m)k (k − 1)!
a=1 m∈Z a,n=1

We rearrange the terms right member of the above equation, we have


∞ ∞
1   1 (−2π i)k  k−1 2π inaz
− 2ζ (k) = n e .
2 (az + m)k (k − 1)!
0 =a∈Z m∈Z a,n=1

After a further little rearrange and use of (2) we obtain the desired result.
Remark 1 Putting r1 = r2 = 0, we have
 
G (z, s, 0, 0) = χ (0) ζ (s, 0) + eπ is ζ (s, 0)

(−2π i)s  
+ A (z, s, 0, 0) + eπ is A(z, s, 0, 0) .
Γ (s)

Replacing s by k (k is an integer with k > 1) in the above, we have

2 (2π i)k
G (z, k) = 2ζ (k) + A (z, k) .
(k − 1)!

After a number of straightforward calculations we arrive at the desired result.


Dedekind and Hardy Type Sums and Trigonometric Sums Induced by. . . 191

The Fourier expansion of the function G (z, k, r, h) is given by:


Corollary 1 Let 2 ≤ k ∈ N, r, h be rational numbers and z ∈ H. Then

2 (−2π i)k  k−1 2π i(n+r)az
G (z, k, r, h) = 2ζ (k, h) + a e .
(k − 1)!
a,n=1

3 Dedekind Sums

The history of the Dedekind sums can be traced back to famous German mathe-
matician Julius Wilhelm Richard Dedekind (1831–1916), who did important work
in abstract algebra in particularly including ring theory, algebraic and analytic
number theory and the foundations of the real numbers. After Dedekind, Hans
Adolph Rademacher (1892–1969), who was one of the most famous German
mathematicians, worked the most deeply the Dedekind sums. Rademacher also
studied important work in mathematical analysis and its applications and analytic
number theory. It is well-known that, the Dedekind sums, named after Dedekind,
are certain finite sums of products of a sawtooth function. The Dedekind sums
are found in the functional equation that emerges from the action of the Dedekind
eta function under modular groups. The Dedekind sums have occurred in analytic
number theory, in some problems of topology and also in the other branches of
Mathematics. Although two-dimensional Dedekind sums have been around since
the nineteenth century and higher-dimensional Dedekind sums have been explored
since the 1950s, it is only recently that such sums have figured flashily in so many
different areas. The Dedekind sums have also many applications in some areas such
as analytic number theory, modular forms, random numbers, the Riemann-Roch
theorem, the Atiyah-Singer index theorem, and the family of zeta functions.
In many applications of elliptic modular functions to analytic number theory,
and theory of elliptic curves, the Dedekind eta function plays a central role. It was
introduced by Dedekind in 1877 by Dedekind. This function is defined on the upper
helf-plane as follows:
∞ 
 
η (τ ) = eπ iτ/12 1 − e2π imτ ,
m=1

,

The infinite product has the form (1 − x n ), where x = e2π iτ . If τ ∈ H,
n=1
then |x| < 1 so the product converges absolutely and it is nonzero. Furthermore,
since the convergence is uniform on compact subsets of H, η(τ ) is analytic on H.
The function η(τ ) is related to analysis, number theory, combinatorics, q-series,
Weierstrass elliptic functions, modular forms, Kronecker limit formula, etc.
192 G. V. Milovanović and Y. Simsek

The behavior of this function under the modular group Γ (1), defined by
   
ab
Γ (1) = A = : ad − bc = 1, a, b, c, d ∈ Z ,
cd

we note that
az + b
Az = .
cz + d

It is well-known that the Dedekind sums s(h, k) first arose in the transformation
formula of the logarithm of the Dedekind-eta function which is given by Apostol,
 
π i(a + d) 1 1
log η(Az) = log η(z) + − π i s(d, c) − + log(cz + d),
12c 4 2

where z ∈ H and s(d, c) denotes the Dedekind sum which defined by


  μ  dμ 
s(d, c) = ,
c c
μ mod c

where (d, c) = 1, c > 0, and



x − [x] − 12 , x ∈
/ Z,
((x)) =
0, otherwise,

where [x] is the largest integer ≤ x. The arithmetical function ((x)) has a period 1
and can thus be expressed by a Fourier series as follows:

1  sin(2π nx)
((x)) = − .
π n
n=1

For basic properties of the Dedekind sums see monograph of Rademacher and
Grosswald [76].
The most important property of Dedekind sums is the reciprocity law. Namely, if
(h, k) = 1 and h and k are positive, then
 
1 h k 1 1
s(h, k) + s(k, h) = + + −
12 k h hk 4

(cf. [2, 100]). This will be discussed in more detail in a separate section.
Dedekind and Hardy Type Sums and Trigonometric Sums Induced by. . . 193

Apostol [2] defined the generalized Dedekind sums sp (h, k) as


k−1  
n nh
sp (h, k) = Bp ,
k k
n=0

where B p (x) is the p-th Bernoulli function defined by

B n (x) = Bn (x − [x]),

where Bn (x) denotes the Bernoulli polynomial. These important polynomials are
defined by the following generating function
 ∞
t tn
ext
= B n (x) .
et − 1 n!
n=0

For x = 0 these polynomials reduce to the well-known Bernoulli numbers Bn =


Bn (0) (cf. [72]). A few first numbers are 1, −1/2, 1/6, 0, −1/30, 0, 1/42, . . . .
The functions B n (x) are 1-periodic, and they satisfy

B n (x) = Bn (x)

for 0 ≤ x < 1, and

B n (x + 1) = B n (x)

for other real x. The Bernoulli function can be expressed by the following Fourier
expansion

n!  1
B n (x) = − e2π imx (6)
(2π i)n mp
0 =m∈Z

Observe that s1 (h, k) = s(h, k). A representation of sp (h, k) as an infinite series


has also given by Apostol [2]. Namely, for odd p ≥ 1, (h, k) = 1 as

  
p! 1 e2π imh/k e2π imh/k
sp (h, k) = − . (7)
(2π i)p mp 1 − e2π imh/k 1 − e2π imh/k
m∈N
m ≡ 0 (mod k)

The relation between Dedekind sums s (h, k) and cot π x are given in the lemma
below. This lemma is a special case of (7). The following well-known result is easily
given:
   mπ 
1 
k−1
mhπ
s (h, k) = cot cot .
4k k k
m=1
194 G. V. Milovanović and Y. Simsek

Recently, many authors proved the above nice formulas by different methods
[4, 11, 12, 14, 16, 29, 90, 101].
Using contour integration and Cauchy Residue Theorem, Berndt [11] proved the
following result:
Lemma 2 Let h, k ∈ N with (h, k) = 1. Then
  
1 1 mhπ
s (h, k) = cot .
2π m k
m∈N
m ≡ 0 (mod k)

The sums sp (h, k) are related to the Lambert series Gp (x) in the same  way 
that s (h, k) is related to η (z), log η (z) being the same as (π iz/12) − G1 e2π iz ,
respectively. The sums sp (h, k) are expressible as infinite series related to certain
Lambert series and, for odd p ≥ 1, sp (h, k) is also seen to be the Abel sum of a
divergent series. This relation is given as follows:
Theorem 3 ([85]) For (h, k) = 1, the Abel sum of the divergent series

  
−p 2π nh
σp (n) n sin
k
n=1

for odd p is given by

(−1 | p) (2π )p (2p!)−1 sp (h, k) ,



where σp (n) = d|n d
p.

Apostol [2] gave a proof of this theorem using a contour integral representation
of the Lambert series Gp (x), but his proof is very different from that given below
(see [85]).
Brief sketch of the proof of Theorem 3 Starting with (6), replacing x by nx (n ∈ N)
and summing over n we get (cf. [85])

 ∞
 
p! 1 2π imnx
B p (nx) = − p
e .
(2π i) mp
n=1 0 =a∈Z n=1

If we rearrange the above equation, we get


⎛ ⎞

 ∞ −1 ∞
p! ⎝  1 2π imnx  1 2π imnx ⎠
B p (nx) = − e + e .
(2π i)p m=−∞
mp mp
n=1 n=1 m,n=1
Dedekind and Hardy Type Sums and Trigonometric Sums Induced by. . . 195

After a little calculation, we easily obtain


⎛ ⎞

 ∞ ∞
p! ⎝  1 2π imnx  1 −2π imnx ⎠
B p (nx) = − e − e . (8)
(2π i)p mp mp
n=1 m,n=1 m,n=1

Because of the identity 2i sin z = eiz − e−iz and putting x = a/b in (8), where
a, b ∈ Z with (a, b) = 1, and writing the Lambert series as a power series Gp (x) =
∞ −p n
n=1 σp (n) n x , we get (cf. [85])


  na  ∞  
p!  −p 2π ma
Bp =− m σp (m) sin . (9)
b (2π i)p b
n=1 m=1
∞ −p n
Using a definition of the Lambert series Gp (x) = n=1 n x / (1 − x )
n and
replacing x by a/b, with (a, b) = 1, in (8), we get

  na    
p! 1 e2π imh/k e2π imh/k
Bp =− − .
b (2π i)p mp 1 − e2π imh/k 1 − e2π imh/k
n=1 m∈N
m ≡ 0 (mod k)

By substituting (7) into the above, we obtain



  na 
Bp = −sp (a, b) . (10)
b
n=1

For odd p we get

(i)1−p = (−1)(1−p)/2 = (−1 | p), (11)

which is known as Jacobi (Legendre) symbol. Finally, combining (9)–(11), we find


the desired result.
Zagier [101] defined the following multiple Dedekind sums


p−1      
π ma1 π ma2 π maj
d(p; a1 , a2 , . . . , aj ) = (−1) j/2
cot cot · · · cot .
p p p
m=1

The sum d(p; a1 , a2 , . . . , aj ) vanishes identically when j is odd. In [90], Simsek,


Kim and Koo gave various formulas for the above sums and finite trigonometric
sums.
196 G. V. Milovanović and Y. Simsek

3.1 Some Others Formulas for the Dedekind Sums

Theorem 4 ([85]) Let a, b ∈ Z with (a, b) = 1 and let p be odd integers. Then
  
m
∞ 
 b−1
2p! 2π mna
sp (a, b) = (−1 | p) sin ζ p, ,
(2π b)p b b
n=1 m=1

where ζ (p, m/b) is the Hurwitz zeta function.


Proof By substituting x = a/b into (8), we easily calculate

  na  ∞  
2p!  1 2π nma
Bp = sin .
b (2π b)p mp b
n=1 m,n=1

Writing m = ub + c, with u = 0, 1, 3, . . . and c = 1, 2, . . . , b − 1 in the above, we


obtain

 na  b−1   
c

 ∞ 

2p! 2π na
Bp =− (−1 | p) ζ p, sin , (12)
b (2π b)p b b
n=1 n=1 c=1

where we assume p > 1 in order to insure that the series involved should be
absolutely convergent and the rearragements valid. Now, after combining Eqs. (10)
and (12), the proof is completed.
Lemma 2, as well as corresponding expression for sp (h, k) in (7), can be

obtained without any knowledge of the function η and the finite sum k−1 n
n=1 nx .
By using the well-known equality
 
e2iπ x e−2iπ x
cot π x = −i − ,
1 − e2iπ x 1 − e−2iπ x

Theorems 3 and 4, a relation between sp (h, k) and cot (anπ /b) can be obtained as
follows:
Theorem 5 ([85]) Let (h, k) = 1. For odd p ≥ 1 we have

  
p! 1 π nh
sp (h, k) = i cot .
(2π i)p np k
n∈N
n ≡ 0 (mod k)

Proof By substituting (6) in to definition of sp (h, k), we have


Dedekind and Hardy Type Sums and Trigonometric Sums Induced by. . . 197


k−1  
n nh
sp (h, k) = Bp
k k
n=1

p! 
k−1  1 2π imnh
=− p n e k
k (2π i) mp
n=1 0 =m∈Z
∞ −1

p! 
k−1  1 2π imnh  1 2π imnh
=− n e k + e k
k (2π i)p mp −∞
mp
n=1 m=1
 
1  2π imnh 

k−1 ∞

p! 2π imnh
=− n e k − e− k .
k (2π i)p mp
n=1 m=1

By applying the well-known identity 2i sin x = eix − e−ix in the above, we obtain


k−1 ∞  
p! 1 2π mnh
sp (h, k) = −2i n sin .
k (2π i)p mp k
n=1 m=1

Now, by using the following well-known identity

    
2π nφ k πφ
a sin = − cot ,
k 2 k
a mod k

where k  φ, φ ∈ Z, then we have the desired result.


By using Theorem 5, we arrive at the following result:
Corollary 2 ([85]) For odd p > 1 we have

p! 
b−1  π na   n 
sp (a, b) = i cot ζ p, . (13)
(2π i)p b b
n=1

The proof this corollary was proved by Apostol [4].


Using a technique developed by Rademacher (Theorem 2, Eq. (5) in [10]) and
Lewittes (Eq. (56) in [4]), we can give the behavior of Lambert series and Dedekind
eta-function. Namely, substituting r1 = r2 = s = 0 into Eq. (3) (and also Eq. (5) in
[10]), we have (cf. [85])
 
πi 1 a+d
A (V (z)) = A (z) + − log (cz + d) + π is (d, c) − π i .
4 2 12c
 
By the definition of G1 eπ iz and A (z), we get (cf. [85])
198 G. V. Milovanović and Y. Simsek

 
 π iV (z)   π iz  π i 1 a+d
G1 e = G1 e + − log (cz + d)+π is (d, c)−π i . (14)
4 2 12c
 
This relation gives modular transformation of G1 eπ iz . Replacing V (z) = az+b
cz+d ,
by W (z) = bz−a
dz−c , where c, d > 0 in (14), we have
 
    πi 1 b−c
G1 eπ iW (z) = G1 eπ iz + − log (dz − c) + π is (−c, d) − π i .
4 2 12d
(15)
Comparing (14) with (15) and using reciprocity law of Dedekind sums

d c 1
12s (d, c) + 12s (c, d) = −3 + + + , (d, c) = 1,
c d dc
we deduce that
    1  
dz − c π i (cb − da − 3dc + 1)
G1 eπ iV (z) − G1 eπ iW (z) = log + .
2 cz + d 12dc

Thus, we arrive at the following results (see [85]).


Theorem 6 Let V (z) = az+b cz+d and W (z) = dz−c be arbitrary modular transforma-
bz−a

tions, with c > 0, and let


 
d
K = z : Re (z) > − , Im (z) > 0 .
c

Then, for z ∈ K we have

 1   1  
dz − c π i (cb − da − 3dc + 1)
e2π inmV (z)
−e2π inmW (z)
= log + .
m 2 cz + d 12dc
m,n=1

Theorem 7 Let m > 0 be an even integer and (a, b) = 1. We have


    
b − ζ s, 1 −
lim ζ s, na na
2 (m + 1)!  s→1 b
sm+1 (a, b) = .
π 2 (4π i)m n1+m
n∈N
n ≡ 0 (mod b)

4 Hardy Sums

Hardy’s sums are derived from theta function. Thus, the well known theta-functions,
ϑn (0, q)(n = 2, 3, 4) related to infinite products are given by
Dedekind and Hardy Type Sums and Trigonometric Sums Induced by. . . 199



ϑ2 (0, q) = 2q 1/2 (1 − q 2n )(1 + q 2n )2 ,
n=1


ϑ3 (0, q) = (1 − q 2n )(1 + q 2n−1 )2 ,
n=1


ϑ4 (0, q) = (1 − q 2n )(1 − q 2n−1 )2 .
n=1

In the sequel we denote ϑ2 (0, q), ϑ3 (0, q) and ϑ4 (0, q) as ϑ2 (z), ϑ3 (z) and ϑ4 (z),
respectively, where q = eπ iz . The relations between theta-functions and Dedekind
eta-function are defined by

η5 (z)
ϑ3 (z) = .
η2 (2z)η2 (z/2)

These relations, as well as others, are studied by Rademacher [72] (see also [84],
as well as the books [99] and [82]).
Let h and k with k > 0 be relatively prime integers (i.e., (h, k) = 1). The Hardy
sums are defined by (see [48])


k−1
S(h, k) = (−1)j +1+[hj /k] ,
j =1


k  j 
s1 (h, k) = (−1)[hj /k] ,
k
j =1


k  j  hj 
s2 (h, k) = (−1)j ,
k k
j =1


k  hj 
s3 (h, k) = (−1)j ,
k
j =1


k−1
s4 (h, k) = (−1)[hj /k] ,
j =1


k  j 
s5 (h, k) = (−1)j +[hj /k] .
k
j =1

By using the following well-known trigonometric formulas, we mention some


relations including the Hardy sums (for details see [40]).
200 G. V. Milovanović and Y. Simsek

If 2m − 1 ≡ 0 (mod k), (h, k) = 1, then


k−1    
π(2m − 1)hj k π h(2m − 1)
j sin = − cot .
k 2 2k
j =1

If m = (2n − 1)h, 2n − 1 ≡ 0 (mod k), and h and k are of opposite parity, an


elementary calculation gives


k−1   πm
π mj
(−1)j sin = − tan .
k 2k
j =1

If 2m ≡ 0 (mod k), and h and k are of opposite parity, an elementary calculation


gives


k−1   πm
π mj k
j
(−1) j sin = tan .
k 2 2k
j =1

If k is odd, then


k−1    
2hπ mj π hm
j
(−1) sin = − tan ;
k k
j =1

if k is even, then


k−1  
2hπ mj
sin = 0.
k
j =1

Also (cf. [15])


k−1  
πj (k − 1)(k − 3)
cot 2
= .
k 3
j =1

The Fourier series of the functionf (x) = (−1)[x] is given by (cf. [40])

1  sin ((2n − 1) xπ )
f (x) = .
2π 2n − 1
n=1

Combining the above finite trigonometric sums and Fourier series of the function
f (x) = (−1)[x] with definitions of the Hardy sums, some relations between Hardy
Dedekind and Hardy Type Sums and Trigonometric Sums Induced by. . . 201

sums and trigonometric functions can be given. Such results were obtained by
Goldberg [40] and Berndt and Goldberg [14]:
Theorem 8 Let h and k denote relatively prime integers with k > 0.
1◦ If h + k is odd, then

4 1  π h(2n − 1) 
S(h, k) = tan ; (16)
π 2n − 1 2k
n=1

2◦ If h is even and k is odd, then

2

 1  π h(2n − 1) 
s1 (h, k) = − cot ; (17)
π 2n − 1 2k
n=1
2n − 1 ≡ 0 (mod k)

3◦ If h is odd and k is even, then

1

 1  π hn 
s2 (h, k) = − tan ; (18)
2π n k
n=1
2n ≡ 0 (mod k)

4◦ If k is odd, then

11  π hn 
s3 (h, k) = tan ; (19)
π n k
n=1

5◦ If h is odd, then

4 1  π h(2n − 1) 
s4 (h, k) = cot ; (20)
π 2n − 1 2k
n=1

6◦ If h and k are odd, then

2

 1  π h(2n − 1) 
s5 (h, k) = tan . (21)
π 2n − 1 2k
n=1
2n − 1 ≡ 0 (mod k)

Using the well-known sum



 1
= π cot πy
n=−∞
n + y
202 G. V. Milovanović and Y. Simsek

in (16) through (21), the relations between Hardy sums and finite trigonometric
sums can be also obtained [14, 40]:
Theorem 9 Let h and k be coprime integers with k > 0.
1◦ If h + k is odd, then
   
1
k
π h(2m − 1) π(2m − 1)
S(h, k) = tan cot ;
k 2k 2k
m=1

2◦ If h is even and k is odd, then


k    
1 π h(2m − 1) π(2m − 1)
s1 (h, k) = − cot cot ;
2k 2k 2k
m=1
m = (k + 1)/2

3◦ If h is odd and k is even, then


  πm
1 
k−1
π hm
s2 (h, k) = − tan cot ;
4k k k
m=1
m = k/2

4◦ If k is odd, then
  πm
1 
k−1
π hm
s3 (h, k) = tan cot ;
2k k k
m=1

5◦ If h is odd, then
   
1
k
π h(2m − 1) π(2m − 1)
s4 (h, k) = cot cot ;
k 2k 2k
m=1

6◦ If h and k are odd, then


k    
1 π h(2m − 1) π(2m − 1)
s5 (h, k) = tan cot .
2k 2k 2k
m=1
m = (k + 1)/2

Using elementary methods, the previous identities were also obtained by Sitara-
machandrarao [91]. Some new higher dimensional generalizations of the Dedekind
sums associated with the Bernoulli functions, as well as ones of Hardy sums, have
Dedekind and Hardy Type Sums and Trigonometric Sums Induced by. . . 203

been recently introduced by Rassias and Tóth [77]. They derived the so-called
Zagier-type identities for these higher dimensional sums, as well as a sequence of
corollaries with interesting particular sums.

5 Dedekind Type Daehee-Changhee (DC) Sums

The first kind n-th Euler function E m (x) is defined by

E n (x) = En (x)

for 0 ≤ x < 1, and by

E n (x + 1) = −E n (x)

for other real x. This function can be expressed by the following Fourier expansion

∞
2m! e(2n−1)π ix
E m (x) = , (22)
(π i)m+1 n=−∞ (2n − 1)m+1

where m ∈ N (for details on Euler polynomials and functions and their Fourier
series see [1, 16, 49–51, 89, 94, 96]). Hoffman [49] studied the Fourier series of
Euler polynomials and expressed the values of Euler polynomials at any rational
argument in terms of tan x and sec x. Suslov [96] considered explicit expansions of
some elementary and q-functions in basic Fourier series of the q-extensions of the
Bernoulli and Euler polynomials and numbers.
Observe that if 0 ≤ x < 1, then (22) reduces to the first kind n-th Euler
polynomials En (x) which are defined by means of the following generating function

 ∞
2etx tn
= En (x) , |t| < π. (23)
e +1
t n!
n=0

Observe that En (0) = En denotes the first kind Euler number which is given by
the following recurrence formula
n  
 n
E0 = 1 and En = − Ek . (24)
k
k=0

Some of them are given by 1, −1/2, 0, 1/4, . . ., En = 2n En (1/2) and E2n = 0


(n ∈ N).
In [50] and [51], by using Fourier transform for the Euler function, Kim derived
some formulae related to infinite series and the first kind Euler numbers. From (22)
it is easy to see that (cf. [1, 49, 51, 89])
204 G. V. Milovanović and Y. Simsek


 1 (−1)m+1 π 2m+2 E2m+1
= . (25)
(2n − 1)2m+2 4(2m + 1)!
n=1

By using the first kind n-th Euler function and above infinite series, we can
construct infinite series representation of the Dedekind type Daehee-Changhee-sum
(DC-sum) and reciprocity law of this sum. We also can give relations between the
Dedekind type DC-sum and some special functions.
The second kind Euler numbers, Em ∗ are defined by means of the following

generating functions (cf. [1, 75, 89])

 ∞
1 2ex xn π
sech x = = 2x = En∗ , |x| < . (26)
cosh x e +1 n! 2
n=0

Kim [51] studied the second kind Euler numbers and polynomials in details. By (23)
and (26), it is easy to see that

m   
 2m  ∗
m−1
∗ m n ∗
Em = 2 En and E2m =− E2n .
n 2n
n=0 n=0

From the above E0∗ = 1, E1∗ = 0, E2∗ = −1, E3∗ = 0, E4∗ = 5, . . ., and E2m+1
∗ =0
(m ∈ N).
The first and the second kind Euler numbers are also related to tan z and sec z,
   
eiz − e−iz e2iz 2 e−2iz 2
tan z = −i iz = − .
e + e−iz 2i e2iz + 1 2i e−2iz + 1

By using (23) and Cauchy product, we have (cf. [89])


∞ ∞ ∞ ∞
1  (2iz)n  (2iz)n 1  (−2iz)n  (−2iz)n
tan z = En − En
2i n! n! 2i n! n!
n=0 n=0 n=0 n=0
∞ 
 n ∞ 
 n
1 (2iz)k (2iz)n−k 1 (−2iz)k (−2iz)n−k
= Ek − Ek
2i k! (n − k)! 2i k! (n − k)!
n=0 k=0 n=0 k=0
∞ n ∞ n
1  Ek 1  Ek
= (2i)n zn − (−2i)n zn
2i k!(n − k)! 2i k!(n − k)!
n=0 k=0 n=0 k=0
⎛ ⎞

 
2j +1 
2j + 1 z2j +1
= (−1)n 22j +1 ⎝ Ek ⎠ .
k (2j + 1)!
j =0 k=0

Finally, using (24) we obtain that (cf. [1, 75, 89])


Dedekind and Hardy Type Sums and Trigonometric Sums Induced by. . . 205


 22n+1 E2n+1 2n+1 π
tan z = (−1)n+1 z , |z| < . (27)
(2n + 1)! 2
n=0

Remark 2 There are several proofs of (27). For example, Kim [51] used

eiz − e−iz 2 4
i tan z = = 1 − 2iz +
eiz + e−iz e − 1 e4iz − 1
to get

 4n (1 − 4n )B2n 2n
z tan z = (−1)n z ,
(2n)!
n=1

i.e., (27). Similarly, Kim [51] proved the following relation for the secant function
(see also [1, 75, 89, 94, 96])

 E∗ π
sec z = (−1)n 2n z2n , |z| < .
(2n)! 2
n=0

Kim [52] defined the Dedekind type Daehee-Changhee (DC) sums as follows:
Definition 1 Let h and k be coprime integers with k > 0. Then


k−1
j  hj 
Tp (h, k) = 2 (−1)j −1 E p , (28)
k k
j =1

where E p (x) denotes the p-th Euler function of the first kind.
The behavior of these sum Tp (h, k) is similar to that of the Dedekind sums.
Several properties and identities of the sum Tp (h, k) and Euler polynomials, as well
as some other interesting results, were derived in [52]. The most fundamental result
in the theory of the Dedekind sums, Hardy-Berndt sums, Dedekind type DC and
the other arithmetical sums is the reciprocity law and it can be used as an aid in
calculating these sums (see Section 8).

6 Trigonometric Representation of the DC-Sums

In this section we can give relations between trigonometric functions and the
sum Tp (h, k). We establish analytic properties of the sum Tp (h, k) and give their
trigonometric representation.
206 G. V. Milovanović and Y. Simsek

Starting from (22) in the form

0 ∞

(π i)m+1 e(2n−1)π ix e(2n−1)π ix
E m (x) = + , (29)
2 m! n=−∞
(2n − 1)m+1 (2n − 1)m+1
n=1

we can get the following auxiliary result (see [89]):


Lemma 3 Let m ∈ N and 0 ≤ x ≤ 1, except for m = 1 when 0 < x < 1. Then we
have

(−1)m 4(2m − 1)!  cos((2n − 1)π x)
E 2m−1 (x) = , (30)
π 2m (2n − 1)2m
n=1

and

(−1)m 4(2m)!  sin((2n − 1)π x)
E 2m (x) = . (31)
π 2m+1 (2n − 1)2m+1
n=1

For 0 ≤ x < 1, E 2m−1 (x) and E 2m (x) reduce to the Euler polynomials, which
are related to Clausen functions (see Section 7).
We now modify the sums Tp (h, k) for odd and even integer p. Thus, by (28), we
define T2m−1 (h, k) and T2m (h, k) sums as follows:
Definition 1 ([89]) Let h and k be coprime integers with k > 0. Then


k−1
j  hj 
T2m−1 (h, k) = 2 (−1)j −1 E 2m−1 (32)
k k
j =1

and


k−1
j  hj 
T2m (h, k) = 2 (−1)j −1 E 2m , (33)
k k
j =1

where E 2m−1 (x) and E 2m (x) denote the Euler functions.


By substituting (30) into (32), we get (cf. [89])


8(−1)m (2m − 1)!  
k−1
cos( (2n−1)π hj
)
T2m−1 (h, k) = − (−1) j
j k
,
kπ 2m (2n − 1) 2m
j =1 n=1

i.e.,
Dedekind and Hardy Type Sums and Trigonometric Sums Induced by. . . 207

∞  
8(−1)m (2m − 1)!   k−1
1 (2n − 1)π hj
T2m−1 (h, k) = − (−1)j j cos .
kπ 2m (2n−1)2m k
n=1 j =1

Since (cf. [14] and [40])



⎪ k

k−1 ⎨ , if 2n − 1 ≡ 0 (mod k),
(2n−1)π ihj/k
je = e(2n−1)π ih/k − 1

⎩1
j =1 − 1), if 2n − 1 ≡ 0 (mod k),
2 k(k

we conclude that


k−1
k
(−1)j j e(2n−1)π ihj/k = ,
e(k+(2n−1)h)π i/k −1
j =1

from which, by some elementary calculations, the following sums follow


k−1  
(2n − 1)π hj k
(−1)j j cos =− (34)
k 2
j =1

and


k−1    
(2n − 1)π hj k (2n − 1)π h
j
(−1) j sin = tan , (35)
k 2 2k
j =1

where 2n − 1 ≡ 0 (mod k).


Using (34) and (25) we obtain the following result:
Theorem 10 Let h and k be coprime positive integers and m ∈ N. Then
 1 
T2m−1 (h, k) = 1 − 2m E2m−1 .
k
Indeed, here we have


4(−1)m (2m − 1)! 1
T2m−1 (h, k) = ,
π 2m (2n − 1)2m
n=1
2n − 1 ≡ 0 (mod k)

i.e.,
208 G. V. Milovanović and Y. Simsek

⎧ ⎫

⎪ ⎪


⎨∞ ∞
 ⎪

4(−1)m (2m − 1)! 1 1
T2m−1 (h, k) = −
π 2m ⎪
⎪ (2n − 1)2m (2n − 1)2m ⎪


⎩n=1 n=1 ⎪

2n − 1 ≡ 0 (mod k)
⎧ ⎫
4(−1)m (2m − 1)! ⎨ ⎬
∞ ∞

1 1
= − ,
π 2m ⎩ (2n − 1)2m k 2m (2j − 1)2m ⎭
n=1 j =1

where we put 2n − 1 = (2j − 1)k in order to calculate the second sum when
2n − 1 ≡ 0 (mod k). It proves the statement.
Similarly, by substituting (31) into (33) we get

∞  π
8(−1)m (2m)!  
k−1
sin (2n−1)hj
T2m (h, k) = j
(−1) j k
,
kπ 2m+1 (2n − 1)2m+1
j =1 n=1

and then, using (35), we arrive at the following theorem.


Theorem 11 ([89]) Let h and k be coprime positive integers and m ∈ N. Then
∞  h
4(−1)m (2m)!  tan (2n−1)π
2k
T2m (h, k) = . (36)
π 2m+1 (2n − 1)2m+1
n=1
2n − 1 ≡ 0 (mod k)

7 DC-Sums Related to Special Functions

In this section, we give relations between DC-sums and some special functions.
In [94], Srivastava and Choi gave many applications of the Riemann zeta
function, Hurwitz zeta function, Lerch zeta function, Dirichlet series for the
polylogarithm function and Dirichlet’s eta function. In [45], Guillera and Sandow
obtained double integral and infinite product representations of many classical
constants, as well as a generalization to Lerch’s transcendent of Hadjicostas’s double
integral formula for the Riemann zeta function, and logarithmic series for the
digamma and Euler beta functions. They also gave many applications. The Lerch
transcendent Φ(z, s, a) (cf. [94, p. 121 et seq.], [45]) is the analytic continuation of
the series

 ∞
1 z z zn
Φ(z, s, a) = + + + ··· = ,
a s (a + 1)s (a + 2)s (n + a)s
n=0
Dedekind and Hardy Type Sums and Trigonometric Sums Induced by. . . 209

which converges for a ∈ C \ Z− 0 , s ∈ C when |z| < 1, and Re (s) > 1 when

|z| = 1, where as usual, Z0 = Z− ∪ {0} and Z− = {−1, −2, . . .}. Φ denotes the
familiar Hurwitz-Lerch zeta function. Here, we mention some relations between this
function Φ and other special functions (cf. [45]).
Special cases include the analytic continuations of the Riemann zeta function

 1
Φ(1, s, 1) = ζ (s) = , Re (s) > 1,
ns
n=1

the Hurwitz zeta function



 1
Φ(1, s, a) = ζ (s, a) = , Re (s) > 1,
(n + a)s
n=0

the alternating zeta function (also called Dirichlet’s eta function η(s))

 (−1)n−1
Φ(−1, s, 1) = ζ ∗ (s) = ,
ns
n=1

the Dirichlet beta function


 1

 (−1)n
2−s Φ −1, s, = β(s) = ,
2 (2n + 1)s
n=0

the Legendre chi function

 1

 z2n+1
2−s z Φ z2 , s, = χs (z) = , |z| ≤ 1, Re (s) > 1,
2 (2n + 1)s
n=0

the polylogarithm

 zk
zΦ(z, n, 1) = Lim (z) =
nm
n=0

and the Lerch zeta function (sometimes called the Hurwitz-Lerch zeta function)

L(λ, α, s) = Φ(e2π iλ , s, α),

which is a generalization of the Hurwitz zeta function and polylogarithm (cf. [3, 9,
19, 20, 22, 25, 45, 92–94]).
By using (29), we can give a relation between the Legendre chi function χs (z)
and the function E m (x):
210 G. V. Milovanović and Y. Simsek

Fig. 1 Graphics of the functions E m (x) for m = 1, 3, 5 (left) and m = 2, 4, 6 (right)

Corollary 3 ([89]) Let m ∈ N. Then we have

2(m!)  −π ix

E m (x) = (−1) m+1
χ m+1 (e ) + χ m+1 (eπ ix
) .
(π i)m+1

The graphics of the functions x → E m (x) on (0, 1) for m = 1, 2, . . . , 6


are presented in Fig. 1. Most of the aforementioned functions are implemented in
Mathematica and Matlab software packages, and the values of these functions can
be calculated with arbitrary precision.
Choi et al. gave relations between the Clausen function, multiple gamma function
and other functions. The higher-order Clausen function Cln (t) is defined for all
n ∈ N \ {1} by (see [94])
⎧ ∞
⎪  sin(kt)



⎨ kn
, if n is even,
k=1
Cln (t) = ∞

⎪  cos(kt)

⎪ , if n is odd.
⎩ kn
k=1

The following functions are related to the higher-order Clausen function (cf. [92],
[22, Eq. (5) and Eq. (6)])

 ∞

sin((2n + 1)x) cos((2n + 1)x)
S(s, x) = and C(s, x) = .
(2n + 1)s (2n + 1)s
n=1 n=1

8 Reciprocity Law

As we mentioned in Section 3, for positive h and k and (h, k) = 1 the reciprocity


law
Dedekind and Hardy Type Sums and Trigonometric Sums Induced by. . . 211

 
1 h k 1 1
s(h, k) + s(k, h) = + + −
12 k h hk 4

holds. For the sums sp (a, b), defined by (13) in Corollary 2, the reciprocity law is
given by

hk n sn (h, k) + khn sn (k, h)

1  n + 1 
n+1
n
= (−1)j Bj hj Bn+1−j k n+1−j + Bn+1 ,
n+1 j n+1
j =0

where (h, k) = 1 and Bn is the nth Bernoulli number (cf. [2, 4, 5]).
In the sequel we mention some reciprocity theorems for Hardy sums:
Theorem 12 Let h and k be coprime positive integers. Then if h + k is odd,

S(h, k) + S(k, h) = 1. (37)

Theorem 13 Let h and k be coprime integers. Then if h and k are odd,

1 1
s5 (h, k) + s5 (k, h) = − . (38)
2 2hk

Theorem 14 Let h and k be coprime integers. If h is even, then


 
1 1 1 h
s1 (h, k) − 2s2 (k, h) = − + . (39)
2 2 hk k

Theorem 15 Let h and k be coprime integers. If k is odd, then

h
2s3 (h, k) − s4 (k, h) = 1 − . (40)
k
These reciprocity theorems appear in Hardy’s list [48], as Eqs. (viii), (vii), (vi),
(vi’) and (ix) on pages 122–123. Berndt [13] deduced (37), (39), and (40), and Gold-
berg [40] deduced (38) from Berndt’s transformation formulae. For other proofs
which do not depend on transformation theory, we refer to Sitaramachandrarao [91].
Otherwise, all reciprocity theorems can be proved by using contour integration and
Cauchy Residue Theorem.
The reciprocity law of the sums Tp (h, k), defined by (28) in Definition 1, is
proved in [52]:
Theorem 16 Let (h, k) = 1 and h, k ∈ N with h ≡ 1 (mod 2) and k ≡ 1 (mod 2).
Then we have
212 G. V. Milovanović and Y. Simsek

k p Tp (h, k) + hp Tp (k, h)
  . hj /p
 j 
k−1
=2 kh E + +k E +h− + (hE + kE)p + (p + 2)Ep ,
k k
j =0
j − [ hj
k ] ≡ 1 mod 2

where
n+1
 n + 1 j
(hE + kE)n+1 = h Ej k n+1−j En+1−j .
j
j =1

The first proof of reciprocity law of the Dedekind sums does not contain the
theory of the Dedekind eta function related to Rademacher [73]. The other proofs of
the reciprocity law of the Dedekind sums were given by Rademacher and Grosswald
[76]. Berndt [11] and Berndt and Goldberg [14] gave various types of Dedekind
sums and their reciprocity laws. Berndt’s methods are of three types. The first
method uses contour integration which was first given by Rademacher [73]. This
method has been used by many authors for example Grosswald [43], Hardy [48],
his method is a different technique in contour integration. The second method is
the Riemann-Stieltjes integral, which was invented by Rademacher [74]. The third
method of Berndt is (periodic) Poisson summation formula. For the method and
technique see also the references cited in each of these earlier works.
The famous property of the all arithmetic sums is the reciprocity law. In the
sequel we prove reciprocity law of (36), by using contour integration.
Theorem 17 Let h, k, m ∈ N with h ≡ 1 (mod 2) and k ≡ 1 (mod 2) and (h, k) =
1. Then we have

hk 2m+1 T2m (h, k) + kh2m+1 T2m (k, h)



m−1
2m 
= 2E2m+1 − E2j +1 E2m−2j −1 h2j +2 k 2m−2j ,
2j + 1
j =0

where En are Euler numbers of the first kind.


Proof Following [11, Theorem 4.2], [14, Theorem 3], [43, 44, 76], we give the
proof of this theorem. We use the contour integration method with the function

tan π hz tan π kz
Fm (z) =
z2m+1

over the contour CN ,



1
IN = Fm (z) dz,
2π i CN
Dedekind and Hardy Type Sums and Trigonometric Sums Induced by. . . 213

where CN is a positively oriented circle of radius RN , with 1 ≤ N < ∞, centered


at the origin. The sequence of radii RN is increasing to ∞ and is chosen so that the
poles of Fm (z) are at a distance from CN greater than some fixed positive number
for each N .
From the above, we have

1 2π    
IN = 2m
e−i2mθ tan π hRN eiθ tan π kRN eiθ dθ.
2π RN 0

 
By the condition on CN , if RN → ∞, then tan RN eiθ is bounded, and therefore
limN →∞ IN = 0 as RN → ∞ for each m ∈ N.
The function Fm (z) has a pole of order 2m−1 at the origin, whose the residue can
be determined from the corresponding Laurent series at z = 0. Using the expansion

 π
tan z = τν z2ν−1 , |z| < ,
2
ν=0

where

4ν (4ν − 1)B2ν 4ν E2ν−1


τν = (−1)ν−1 = (−1)ν (cf. Eq. (27))
(2ν)! 2(2ν − 1)!

we can get the Laurent expansion of Fm (z) at z = 0 in the following form



hkπ 2 
Fm (z) = (−1)n π 2n 4n+1 fn z2n , (41)
z2m−1
n=0

where


n
E2j +1 E2n−2j +1 h2j k 2n−2j
fn = .
(2j + 1)!(2n − 2j + 1)!
j =0

Then, from (41) for n = m − 1, we get the residue of the function Fm (z) at the
pole z = 0 as

Res Fm (z) = hkπ 2 (−1)m−1 π 2m−2 4m fm−1


z=0

(−1)m (2π )2m   2m 


m−1
=− E2j +1 E2m−2j −1 h2j +2 k 2m−2j . (42)
(2m)! hk 2j + 1
j =0
214 G. V. Milovanović and Y. Simsek

The other singularities of the function Fm (z) in the interior of the contour CN are
points of the sets
 
2j − 1
XN = ξj = : |ξj | < RN , j ∈ Z
2h

and
 
2 − 1
YN = η  = : |η | < RN ,  ∈ Z .
2k

Since h and k are odd positive integers, then the sets XN and YN has an intersection
 
2j − 1
ZN = XN ∩ YN = ζj = : |ζj | < RN , j ∈ Z ,
2

with double poles of the function Fm (z) in the interior of CN . Their residues are

d . / (2m + 1)4m+1
Res Fm (z) = lim (z − ζj )2 Fm (z) = − . (43)
z=ζj z→ζj dz (2j − 1)2m+2 hkπ 2

Te residues of the simple poles z = ξj ∈ XN \ ZN and η ∈ YN \ ZN are easily


found to be

22m+1 h2m  (2j − 1)π k 


Res Fm (z) = lim (z − ξj )Fm (z) = − tan (44)
z=ξj z→ξj π(2j − 1)2m+1 2h

and

22m+1 k 2m  (2 − 1)π h 


Res Fm (z) = lim (z − η )Fm (z) = − tan , (45)
z=η z→η π(2 − 1)2m+1 2k

respectively. Now, by the Cauchy residue theorem, we have


  
IN = Res Fm (z) + Res Fm (z) + Res Fm (z) + Res Fm (z).
z=ξj z=η z=0 z=ζj
ξj ∈XN \ZN η ∈YN \ZN ζj ∈ZN
(46)
Using the residues (45), (44), (42), (43), and letting N → ∞ in (46), we have
that IN → 0 and
   
∞ (2−1)π h ∞ (2j −1)π k
4m+1 k 2m  tan 2k 4m+1 h2m  tan 2h
+
π (2 − 1)2m+1 π (2j − 1)2m+1
=1 j =1
2 − 1 ≡ 0 (mod k) 2j − 1 ≡ 0 (mod h)
Dedekind and Hardy Type Sums and Trigonometric Sums Induced by. . . 215

(−1)m (2π )2m   2m 


m−1
=− E2j +1 E2m−2j −1 h2j +2 k 2m−2j
(2m)! hk 2j + 1
j =0

(2m + 1)4m+2  1
− ,
2hkπ 2 (2j − 1)2m+2
j =1

where the last sum is given by (25).


Multiplying this equality by the factor hk(−1)m (2m)!/(2π )2m , we arrive at the
desired result.
Corollary 4 Let For each m ∈ N and each positive odd integer k ≥ 3, we have


m−1
2m 
k 2m+1 T2m (1, k) = 2E2m+1 − E2j +1 E2m−2j −1 h2j +2 k 2m−2j .
2j + 1
j =0

This result can be proved in a similar way as Theorem 17. For the sets of
singularities here ZN = XN , so that the first term on the right hand side in (46)
vanishes.
We now give a relation between Hurwitz zeta function, tan z and the sum
T2m (h, k).
Hence, substituting n = rk + j , 0 ≤ r ≤ ∞, 1 ≤ j ≤ k (j = (k + 1)/2) into
(36), and recalling that tan(π + α) = tan α, then we have
 
∞ π h2(rk+j )+1
4(−1)m (2m)! 
k  tan 2k
T2m (h, k) =
π 2m+1 (2(rk + j ) + 1)2m+1
j =1 r=0
j = (k + 1)/2

4(−1)m (2m)! 
k  (2j − 1)π h  

1
= tan  2m+1 ,
π 2m+1 (2k)2m+1 2k 2j −1
j =1 r=0 r+ 2k
j = (k + 1)/2

where the last sum can be identified as the Hurwitz zeta function ζ (s, x) at s =
2m + 1 and x = (2j − 1)/2. Note that j = (k + 1)/2 provides the condition
2n − 1 ≡ 0 (mod k) in the summation process in (36).
In this way, we now arrive at the following result:
Theorem 18 Let h and k be coprime positive integers and m ∈ N. Then

4(−1)m (2m)! 
k  (2j − 1)π h   2j − 1 
T2m (h, k) = tan ζ 2m + 1, .
(2kπ )2m+1 2k 2k
j =1
j = (k + 1)/2
216 G. V. Milovanović and Y. Simsek

Remark 3 Finally, we mention the sums Y (h, k) defined by Simsek [88] (see also
[86] and [65, p. 211]) as Y (h, k) = 4ks5 (h, k), where h and k are odd with (h, k) =
1. If we integrate the function F (z) = cot(π z) tan(π hz) tan(π kz) over a contour,
obtained from the rectangle with vertices at ±iB, 12 ± iB (B > 0), we see that F (z)
has the poles z = 0 and z = 1/2 on this contour; therefore, reciprocity of the sums
Y (h, k) is given by

hY (h, k) + kY (k, h) = 2hk − 2.

Remark 4 Using the definitions of the q-analogues of some classical arithmetic


functions (Riemann zeta functions, Dirichlet L-functions, Hurwitz zeta functions,
Dedekind sums), Simsek [87] defined q-analogues of these functions and gave some
relations among them.
Remark 5 The higher multiple elliptic Dedekind sums and the reciprocity law have
been introduced and considered by Bayad and Simsek [7] (see also [6, 8, 29, 46, 83]).

9 Sums Obtained from Gauss-Chebyshev Quadratures

It is well-known that Gaussian quadrature formulas with respect to the Chebyshev


weight functions of the first, second, third, and fourth kind,
- -
1 1+t 1−t
w1 (t) = √ , w2 (t) = 1 − t 2, w3 (t) = , w4 (t) = ,
1 − t2 1−t 1+t

respectively (cf. [18, 35], [58, p. 122]) have nodes expressible by trigonometric
functions. In a short note in 1884 Stieltjes [95] gave the explicit expressions for
these quadrature formulas for the weights w1 , w2 , and w4 ,
  
π 
n
1 (2k − 1)π
w1 (t)f (t) dt = f cos + Rn,1 [f ], (47)
−1 n 2n
k=1
  
π  2 kπ
1 n

w2 (t)f (t) dt = sin f cos + Rn,2 [f ], (48)
−1 n+1 n+1 n+1
k=1
  
4π  2 kπ
1 n
2kπ
w4 (t)f (t) dt = sin f cos + Rn,4 [f ], (49)
−1 2n + 1 2n + 1 2n + 1
k=1

where Rn,ν (f ) = 0, ν = 1, 2, 4, for all algebraic polynomials of degree at most


2n − 1. In the class of functions C 2n [−1, 1], the remainder term of these Gaussian
formulas can be done in the form (cf. [58, p. 333])
Dedekind and Hardy Type Sums and Trigonometric Sums Induced by. . . 217

πn,ν 2 (2n)
Rn,ν [f ] = f (ξ ), −1 < ξ < 1,
(2n)!

where the norms of the corresponding orthogonal polynomials πn,ν can be expressed
by the coefficients βk,ν in their three-term recurrence relations

πk+1,ν (t) = (t − αk,ν )πk,ν (t) − βk,ν πk−1,ν (t), k = 1, 2, . . . ,


π0,ν (t) = 1, π−1,ν (t) = 0,

as πn,ν 2 = β0,ν β1,ν · · · βn,ν , with


 1
β0,ν = μ0,ν = wν (t) dt.
−1

The recurrence coefficients for these Chebyshev weights, as well as the cor-
responding values of πn,ν 2 are presented in Table 1 (cf. [35, p. 29]). For
completeness, we also give parameters for the Chebyshev weight of the third kind
w3 . The corresponding quadrature formula for w3 (cf. [61]),
  
4π  2 (2k − 1)π
n
1 (2k − 1)π
w3 (t)f (t) dt = cos f cos + Rn,3 [f ],
−1 2n + 1 2(2n + 1) 2n + 1
k=1
(50)

Table 1 Recurrence coefficients for different kind of Chebyshev polynomials


Weight function Recurrence coefficients πn,ν 2
Chebyshev I αk,1 = 0 (k ≥ 0)

1 1 4n
(ν = 1) β0,1 = π, β1,1 = , βk,1 = (k ≥ 1)
2 4
Chebyshev II αk,2 = 0 (k ≥ 0)
π
π 1 22n+1
(ν = 2) β0,2 = , βk,2 = (k ≥ 1)
2 4
1
Chebyshev III α0,3 = , αk,3 = 0 (k ≥ 1)
2 π
1 4n
(ν = 3) β0,3 = π, βk,3 = (k ≥ 1)
4
1
Chebyshev IV α0,4 = − , αk,4 = 0 (k ≥ 1)
2 π
4n
1
(ν = 4) β0,4 = π, βk,4 = (k ≥ 1)
4
218 G. V. Milovanović and Y. Simsek

can be obtained by changing t := −t and using (49), as well as


 1  1
w3 (t)f (t) dt = w4 (t)f (−t) dt = Qn,4 (f (− · )) + Rn,3 [f ].
−1 −1

Taking certain algebraic polynomials of degree at most 2n and integrating them


by using some of quadrature formulas (47)–(49) and (50), we can obtain some
trigonometric sums. For example, by a polynomial pm (t) of degree m < 2n, using
the quadrature formula (47) we get

n   
(2k − 1)π n 1 pm (t)
pm cos = √ dt,
2n π −1 1 − t 2
k=1

but if p2n (t) = At 2n + terms of lower degree, then



n   
(2k − 1)π n 1 p2n (t) 2nA
p2n cos = √ dt − n .
k=1
2n π −1 1−t 2 4
1
Thus, we need to compute only integrals of the forms −1 pm (t)wν (t) dt for ν =
1, 2, 3, 4. As usual (s)n is the well known Pochhammer symbol defined by
Γ (s + n)
(s)n = s(s + 1) · · · (s + n − 1) = (Γ is the gamma function).
Γ (s)
If we take

pm (t) = Um (c t) = 2m cm t m + terms of lower degree, c ∈ R,

where Um is the Chebyshev polynomial of the second kind and degree m and use
the integral (cf. [71, p. 456])
 1 U2m (cx)
√ dx = π Pm (2c2 − 1),
−1 1 − x2

where Pm is the Legendre polynomial of degree m, we get the following statement:


Theorem 19 Let Um be Chebyshev polynomial of the second kind and degree m
and c ∈ R. We have


⎪ 0, m is odd (m ≥ 1),
  ⎪

n
(2ν − 1)π
= n Pm/2 (2c −1),
2
Um c cos m is even (0≤m<2n),
2n ⎪
⎪  
ν=1 ⎪
⎩n P (2c2 − 1) − 2c2n ,
n m = 2n,

where Pn is the Legendre polynomial of degree n.


Dedekind and Hardy Type Sums and Trigonometric Sums Induced by. . . 219

Similarly, using the Legendre polynomials

(n + 1)n n
Pm (t) = t + terms of lower degree,
2n n!
defined by the generating function (cf. [58, p. 129])

∞
1
√ = Pm (t)x m ,
1 − 2xt + x 2 m=0

for which the following integrals are true (see [71, p. 423])
 1  2
P2m (x) (1/2)m
√ dx = π
−1 1 − x2 m!

and
  2
1 x P2m+1 (x) (1/2)m 2m + 1
√ dx = π ,
−1 1 − x2 m! 2m + 2

we can prove the following statement:


Theorem 20 Let Pm be the Legendre polynomial of degree n. Then


⎪ 0, m is odd (m ≥ 1),


n 
(2ν − 1)π
 ⎨ n  m 2

Pm cos = 4m m/2 , m is even (0 ≤ m < 2n),
2n ⎪

ν=1 ⎪
⎪  2n 2  4n 

⎩ n − 2 , m = 2n,
42n n 2n
and


n  
(2ν − 1)π (2ν − 1)π
cos Pm−1 cos
2n 2n
ν=1


⎪ 0, m is odd (m ≥ 1),


⎨ n m − 1  m − 2 2

= 4m−2 m , m is even (0<m<2n),
⎪ (m − 1)/2

⎪   2n − 2 2  4n − 2 


⎩ 1 2(2n − 1) −n , m = 2n.
42n−1 n−1 2n − 1

Now, we use the following weighted integral for Legendre polynomials,


expressed in terms of the hypergeometric function (see [71, p. 422]),
220 G. V. Milovanović and Y. Simsek

 9 :
1 √ Γ (β) −m, m + 1, β
Iβ (m) = (1−x 2 )β−1 Pm (x) dx = (−1)m π   3 F2 ;1 .
−1 Γ β + 12 2β, 1

For odd m this integral vanishes. We need now its value for β = 3/2 and even
m, i.e.,
9 :
π −m, m + 1, 3/2
I3/2 (m) = 3 F2 ;1
2 3, 1

π π
= ·     m   .
2 2Γ 1 − m
Γ 3
− m
Γ + 1 Γ m2 + 2
2 2 2 2 2

Since Γ (z) = Γ (z + k)/(z)k (k ∈ N), we have


1
1 m Γ m √π  m 
2 m/2 2
− = = !,
1 m
Γ (−1)
2 2 m! 2

2 2 m/2

so that the previous integral becomes

π  m 2
I3/2 (m) = − .
4m (m − 1)(m + 2) m/2

Setting pm (t) = Pm (t) in (48) we get the following result:


Theorem 21 Let Pm be the Legendre polynomial of degree n. Then


n  
kπ kπ
sin2 Pm cos
n+1 n+1
ν=1


⎪ 0, m is odd (m ≥ 1),

⎪  m 2

⎨ n+1
= − , m is even (0 ≤ m < 2n),
⎪ 4m (m + 2)(m − 1) m/2

⎪  1  2n 2  4n 


⎩− 2 + (n + 1) , m = 2n.
42n+1 2n − 1 n 2n

In the sequel we give trigonometric sums obtained by monomials pm (t) = t m in


the quadrature formulas (47)–(49). Because of that, we need the moments
 1
μm,ν = wν (t)t m dt, ν = 1, 2, 3, 4.
−1
Dedekind and Hardy Type Sums and Trigonometric Sums Induced by. . . 221

These moments are


√   ⎧
π ((−1)m + 1) Γ m+1 ⎨ 0, m is odd,
μm,1 =  
2
= π  m 
mΓ m2 ⎩ , m is even;
2m m/2

√   ⎧
π ((−1)m + 1) Γ m+1 ⎨ 0, m is odd,
μm,2 = m 
2
= π  m 
4Γ 2 + 2 ⎩ , m is even;
2m (m + 2) m/2
  2    
2m ((−1)m − 1) Γ m2 + 1 + ((−1)m + 1) Γ m+1
2 Γ m+3
2
μm,4 =
Γ (m + 2)

⎪ π  m+1 

⎨ − m+1 , m is odd,
2 (m + 1)/2
=
⎪ π  m 

⎩ m , m is even.
2 m/2

Note that μm,3 = (−1)m μm,4 .


According to (47)–(49) we conclude that the following results hold.
Theorem 22 We have


⎪ 0, m is odd (m ≥ 1),

⎪  m 
n
(2ν − 1)π ⎨ n
cosm = 2m m/2 , m is even (0 ≤ m < 2n),
2n ⎪
⎪  2n  
ν=1 ⎪

⎩ n − 2 , m = 2n;
4 n n


⎪ 0, m is odd (m ≥ 1),


n ⎪
⎨ n+1  m 
νπ νπ m is even (0 ≤ m < 2n),
sin2 cosm = (m + 2)2m m/2 ,
n+1 n+1 ⎪ ⎪   
ν=1 ⎪



1 2n
− n − 1 , m = 2n;
2 2n+1 n


⎪ 2n + 1  m 

⎪ , m is even (0 ≤ m < 2n),

⎪ 2m+2 m/2

⎨  
n
νπ 2νπ 2n + 1 m+1
sin2 cosm = − m+3 , m is odd (0 ≤ m < 2n),
2n + 1 2n + 1 ⎪
⎪ 2 (m + 1)/2
ν=1 ⎪
⎪   

⎩ 2n + 1 2n − 1 ,

m = 2n.
4n+1 n
222 G. V. Milovanović and Y. Simsek

Trigonometric sums can be also obtained in a similar way using quadrature


formulas of Radau and Lobatto type with respect to Chebyshev weights. We mention
that shortly after Stieltjes’ results [95], Markov [57] (see also [61]) obtained the
explicit expressions for Gauss-Radau and Gauss-Lobatto formulas, with respect to
the Chebyshev weight of the first kind w1 (t) (for both endpoints),
 9  :
1 2π 1  n
(2k − 1)π (−1)
w1 (t)f (t) dt = f (−1) + f cos + Rn+1,1 [f ],
−1 2n + 1 2 2n + 1
k=1

 9  :
1 2π 1
n 2kπ (+1)
w1 (t)f (t) dt = f (1) + f cos + Rn+1,1 [f ],
−1 2n + 1 2 2n + 1
k=1

and
 9   :
1 π 1  n
kπ 1
w1 (t)f (t) dt = f (−1) + f cos + f (1) +Rn+2,1
L
[f ],
−1 n+1 2 n+1 2
k=1

respectively, where

(∓1) πf (2n+1) (ξ )
Rn+1,1 [f ] = ± (f ∈ C 2n+1 [−1, 1])
(2n + 1)!22n

and

πf (2n+2) (ξ )
L
Rn+2,1 [f ] = − , (f ∈ C 2n+2 [−1, 1])
(2n + 2)!22n+1

and ξ ∈ (−1, 1). These Gauss-Radau formulas are exact for all algebraic polyno-
mials of degree at most 2n, while the Gauss-Lobatto is exact for polynomials up to
degree 2n + 1, so that we can obtain trigonometric sums taking monomials x m in
the previous formulas for all m ≤ 2n + 1 and m ≤ 2n + 2, respectively.
There are also similar quadrature formulas for other Chebyshev weights wν (t),
ν = 2, 3, 4. For details see [34, 36, 37, 67, 69, 78]. A general approach in
construction of Gauss-Radau and Gauss-Lobatto formulas can be found in [58,
pp. 328–332]. In some of these cases the nodes of quadratures can be expressed
in terms of trigonometric functions, and such quadratures can be used for getting
trigonometric sums.
Some more complex trigonometric sums can be obtained using quadrature
formulas of Turán type (quadrature with multiple nodes) with respect to the
Chebyshev weight functions. For some details on such quadrature formulas see
[38, 39, 59, 60, 64, 79–81, 97].
Dedekind and Hardy Type Sums and Trigonometric Sums Induced by. . . 223

10 Sums Obtained from Trigonometric Quadrature Rules

Another way for getting trigonometric sums is based on quadrature rules with
maximal trigonometric degree of exactness. With Td we denote the linear space
of all trigonometric polynomials of degree less than or equal to d,

a0 
d

td (x) = + ak cos kx + sin kx (ak , bk ∈ R).
2
k=1

If |ad | + |bd | > 0 the degree of td is strictly d.


We say that a quadrature formula of the form
 2π 
n
f (x)w(x) dx = wν f (xν ) + Rn (f ), 0 ≤ x0 < x1 < · · · < xn < 2π,
0 ν=0

has trigonometric degree of exactness equal to d if the remainder term Rn (f ) = 0


for all f ∈ Td and there exists some g in Td+1 such that Rn (g) = 0.
Quadrature rules with a maximal trigonometric degree of exactness are known
in the literature as trigonometric quadrature rules of Gaussian type. Maximal
trigonometric degree of exactness for quadrature rule with n + 1 nodes is n.
A brief historical survey of available approaches for the construction of quadra-
ture rules with maximal trigonometric degree of exactness has been given in [61]
(see also [62, 68]). Here, we consider only a simple case of the (2n + 1)-point
trigonometric quadrature formula


2π 
2π 2n
f (x) dx = f (xk ) + R2n+1 [f ], (51)
0 2n + 1
k=0

with the nodes


2kπ
xk = θ + , k = 0, 1, . . . , 2n,
2n + 1

where 0 ≤ θ < 2π/(2n + 1). Formula (51) is exact for every trigonometric
polynomial of degree at most 2n (cf. [98]). Such kind of quadratures have
applications in numerical integration of 2π -periodic functions. Two special cases
of the quadrature formula (51) for which θ = 0 and θ = π/(2n + 1) are very
interesting in applications. Their quadrature sums are
 
2π 
2n
2kπ
QT2n+1 (f ) = f (52)
2n + 1 2n + 1
k=0
224 G. V. Milovanović and Y. Simsek

and
 
2π 
2n
(2k + 1)π
QM
2n+1 (f ) = f , (53)
2n + 1 2n + 1
k=0

respectively. Some details on QT2n+1 (f ) and its applications in the trigonometric


approximation can be found in [58, Chap. 3]. The second formula QM 2n+1 (f ) has
been analyzed in [62].
Remark 6 Putting h = 2π/(2n + 1) and fα ≡ f (αh), we can write the formulas
(52) and (53) in the forms
 
1 1
QT2n+1 (f ) =h f0 + f1 + · · · + f2n + f2n+1
2 2

and
* +
2n+1 (f ) = h f1/2 + f3/2 + · · · + f2n + f2n+1/2 ,
QM

where, because of periodicity, we introduced f2n+1 = f (2π ) = f (0) = f0 . These


quadratures (52) and (53) are symmetric with respect to the point x = π , and
they are, in fact, the composite trapezoidal and midpoint rules, respectively. Also,
they are equivalent to the trigonometric version of the Gauss-Radau formulas with
respect to the Chebyshev weight of the first kind on (−1, 1) (cf. [61]).
Taking f (x) = cos2m−2ν x sin2ν x in (51), where 0 ≤ m ≤ n and 0 ≤ ν ≤ m,
we get the following result:
Theorem 23 If n, m, ν ∈ N and 0 ≤ ν ≤ m ≤ n, we have
  

2n     2m m
2kπ 2kπ 2n + 1 m ν
cos2m−2ν θ + sin2ν θ + =  
2n + 1 2n + 1 4m 2m
k=0 2"

for each real θ .


Here, we see that f ∈ T2m , as well as that
 2π  π/2
Im,ν = cos 2m−2ν
x sin2ν
x dx = 4 cos2m−2ν x sin2ν x dx
0 0
 1
=2 t ν−1/2 (1 − t)m−ν−1/2 dt
0

2  1  1
= Γ ν+ Γ m−ν+ ,
m! 2 2
Dedekind and Hardy Type Sums and Trigonometric Sums Induced by. . . 225


after the change of variables t = sin2 x. Since Γ (ν + 1/2) = (2ν − 1)!! π /2ν , by
using (51), we obtain the desired result.
Selecting other functions, such that f ∈ T2n , we can get similar results as in
Section 9.

Acknowledgements The authors have been supported by the Serbian Academy of Sciences and
Arts, Φ-96 (G. V. Milovanović) and by the Scientific Research Project Administration of Akdeniz
University (Y. Simsek).

References

1. M. Abramowitz, I.A. Stegun, Handbook of Mathematical Functions with Formulas, Graphs,


and Mathematical Tables. National Bureau of Standards Applied Mathematics Series, vol. 55
(Dover, New York, 1965)
2. T.M. Apostol, Generalized dedekind sums and transformation formulae of certain Lambert
series. Duke Math. J. 17, 147–157 (1950)
3. T.M. Apostol, On the Lerch zeta function. Pac. J. Math. 1, 161–167 (1951)
4. T.M. Apostol, Theorems on generalized Dedekind sums. Pac. J. Math. 2, 1–9 (1952)
5. T.M. Apostol, Modular Functions and Dirichlet Series in Number Theory (Springer, New York,
1976)
6. A. Bayad, Sommes de Dedekind elliptiques et formes de Jacobi. Ann. Instit. Fourier 51(1),
29–42 (2001)
7. A. Bayad, Y. Simsek, Dedekind sums involving Jacobi modular forms and special values of
Barnes zeta functions. Ann. Inst. Fourier, Grenoble 61(5), 1977–1993 (2011)
8. M. Beck, Dedekind cotangent sums. Acta Arithmetica 109(2), 109–130 (2003)
9. B.C. Berndt, On the Hurwitz zeta-function. Rocky Mt. J. Math. 2(1), 151–157 (1972)
10. B.C. Berndt, Generalized Dedekind eta-Functions and generalized Dedekind sums. Trans. Am.
Math. Soc. 178, 495–508 (1973)
11. B.C. Berndt, Dedekind sums and a paper of G. H. Hardy. J. Lond. Math. Soc. (2). 13(1), 129–
137 (1976)
12. B.C. Berndt, Reciprocity theorems for Dedekind sums and generalizations. Adv. Math. 23(3),
285–316 (1977)
13. B.C. Berndt, Analytic Eisenstein series, theta-functions and series relations in the spirit of
Ramanujan. J. Reine Angew. Math. 303/304, 332–365 (1978)
14. B.C. Berndt, L.A. Goldberg, Analytic properties of arithmetic sums arising in the theory of the
classical theta-functions. SIAM J. Math. Anal. 15, 143–150 (1984)
15. B.C. Berndt, B.P. Yeap, Explicit evaluations and reciprocity theorems for finite trigonometric
sums. Adv. Appl. Math. 29(3), 358–385 (2002)
16. L. Carlitz, Some theorems on generalized Dedekind sums. Pac. J. Math. 3, 513–522 (1953)
17. H. Chen, On some trigonometric power sums. Int. J. Math. Math. Sci. 30(3), 185–191 (2002)
18. T.S. Chihara, An Introduction to Orthogonal Polynomials (Gordon and Breach, New York,
1978)
19. J. Choi, Some Identities involving the Legendre’s chi-function. Commun. Korean Math. Soc.
22(2), 219–225 (2007)
20. J. Choi, D.S. Jang, H.M. Srivastava, A generalization of the Hurwitz-Lerch zeta function.
Integral Transform. Spec. Funct. 19(1–2), 65–79 (2008)
21. W. Chu, Reciprocal relations for trigonometric sums. Rocky Mt. J. Math. 48(1), 121–140
(2018)
226 G. V. Milovanović and Y. Simsek

22. D. Cvijović, Integral representations of the Legendre chi function. J. Math. Anal. Appl. 332,
1056–1062 (2007)
23. D. Cvijović, Summation formulae for finite cotangent sums. Appl. Math. Comput. 215, 1135–
1140 (2009)
24. D. Cvijović, Summation formulae for finite tangent and secant sums. Appl. Math. Comput.
218, 741–745 (2011)
25. D. Cvijović, J. Klinkovski, Values of the Legendre chi and Hurwitz zeta functions at rational
arguments. Math. Comp. 68, 1623–1630 (1999)
26. D. Cvijović, J. Klinkovski, Finite cotangent sums and the Riemann zeta function. Math.
Slovaca 50(2), 149–157 (2000)
27. D. Cvijović, H.S. Srivastava, Summation of a family of finite secant sums. Appl. Math.
Copmput. 190, 590–598 (2007)
28. D. Cvijović, H.S. Srivastava, Closed-form summations of Dowker’s and related trigonometric
sums. J. Phys. A 45(37), 374015 (2012)
29. U. Dieter, Cotangent sums, a further generalization of Dedekind sums. J. Number Theory 18,
289–305 (1984)
30. J.S. Dowker, On Verlinde’s formula for the dimensions of vector bundles on moduli spaces. J.
Phys. A 25(9), 2641–2648 (1992)
31. C M. da Fonseca, V. Kowalenko, On a finite sum with powers of cosines. Appl. Anal. Discrete
Math. 7, 354–377 (2013)
32. C.M. da Fonseca, M.L. Glasser, V. Kowalenko, Basic trigonometric power sums with applica-
tions. Ramanujan J. 42(2), 401–428 (2017)
33. C.M. da Fonseca, M.L. Glasser, V. Kowalenko, Generalized cosecant numbers and trigonomet-
ric inverse power sums. Appl. Anal. Discrete Math. 42(2), 70–109 (2018)
34. W. Gautschi, On the remainder term for analytic functions of Gauss-Lobatto and Gauss-Radau
quadratures. Rocky Mt. J. Math. 21, 209–226 (1991)
35. W. Gautschi, Orthogonal Polynomials: Computation and Approximation (Clarendon Press,
Oxford, 2004)
36. W. Gautschi, S. Li, The remainder term for analytic functions of Gauss-Radau and Gauss-
Lobatto quadrature rules with multiple end points. J. Comput. Appl. Math. 33, 315–329 (1990)
37. W. Gautschi, S. Li, Gauss-Radau and Gauss-Lobatto quadratures with double end points. J.
Comput. Appl. Math. 34, 343–360 (1991)
38. W. Gautschi, G.V. Milovanović, s-orthogonality and construction of Gauss-Turán-type quadra-
ture formulae. J. Comput. Appl. Math. 86, 205–218 (1997)
39. A. Ghizzetti, A. Ossicini, Quadrature Formulae (Akademie Verlag, Berlin, 1970)
40. L.A. Goldberg, Transformation of Theta-Functions and Analogues of Dedekind Sums, Thesis,
University of Illinois Urbana (1981)
41. P.J. Grabner, H. Prodinger, Secant and cosecant sums and Bernoulli-Nörlund polynomials.
Quaest. Math. 30, 159–165 (2007)
42. I.S. Gradshteyn, I.M. Ryzhik, Table of Integrals, Series, and Products, 7th edn. (Else-
vier/Academic Press, Amsterdam, 2007)
43. E. Grosswald, Dedekind-Rademacher sums. Am. Math. Mon. 78, 639–644 (1971)
44. E. Grosswald, Dedekind-Rademacher sums and their reciprocity formula. J. Reine. Angew.
Math. 25(1), 161–173 (1971)
45. J. Guillera, J. Sondow, Double integrals and infinite products for some classical constants via
analytic continuations of Lerch’s transcendent. Ramanujan J. 16(3), 247–270 (2008)
46. R.R. Hall, J.C. Wilson, D. Zagier, Reciprocity formulae for general Dedekind-Rademacher
sums. Acta Arith. 73(4), 389–396 (1995)
47. E.R. Hansen, A Table of Series and Products (Prentice-Hall, Englewood Cliffs, 1975)
48. G.H. Hardy, On certain series of discontinous functions connected with the modular functions.
Quart. J. Math. 36, 93–123 (1905); [Collected Papers, vol. IV (Clarendon Press, Oxford, 1969),
pp. 362–392]
49. M.E. Hoffman, Derivative polynomials and associated integer sequences. Electron. J. Combin.
6, 13 (1999), Research Paper 21
Dedekind and Hardy Type Sums and Trigonometric Sums Induced by. . . 227

50. T. Kim, Note on the Euler numbers and polynomials. Adv. Stud. Contemp. Math. 17, 131–136
(2008)
51. T. Kim, Euler numbers and polynomials associated with zeta functions. Abstr. Appl. Anal.
2008, 11 (2008), Art. ID 581582
52. T. Kim, Note on Dedekind type DC sums. Adv. Stud. Contemp. Math. (Kyungshang) 18(2),
249–260 (2009)
53. M.I. Knoop, Modular Functions in Analytic Number Theory (Markham Publishing Company,
Chicago, 1970)
54. N. Koblitz, Introduction to Elliptic Curves and Modular
Forms (Springer, New York, 1993)
55. J. Lewittes, Analytic continuation of the series (m + nz)−s . Trans. Am. Math. Soc. 159,
505–509 (1971)
56. J. Lewittes, Analytic continuation of Eisenstein series. Trans. Am. Math. Soc. 177, 469–490
(1972)
57. A. Markov, Sur la méthode de Gauss pour le calcul approché des intégrales. Math. Ann. 25,
427–432 (1885)
58. G. Mastroianni, G.V. Milovanović, Interpolation Processes – Basic Theory and Applications.
Springer Monographs in Mathematics (Springer, Berlin/Heidelberg/New York, 2008)
59. G.V. Milovanović, Construction of s-orthogonal polynomials and Turán quadrature formulae,
in Numerical Methods and Approximation Theory III, ed. by G.V. Milovanović (Niš, 1987)
(Univ. Niš, Niš, 1988), pp. 311–328
60. G.V. Milovanović, Quadrature with multiple nodes, power orthogonality, and moment-
preserving spline approximation. J. Comput. Appl. Math. 127, 267–286 (2001)
61. G.V. Milovanović, D. Joksimović, On a connection between some trigonometric quadrature
rules and Gauss-Radau formulas with respect to the Chebyshev weight. Bull. Cl. Sci. Math.
Nat. Sci. Math. 39, 79–88 (2014)
62. G.V. Milovanović, A.S. Cvetković, M.P. Stanić, Trigonometric orthogonal systems and
quadrature formulae. Comput. Math. Appl. 56, 2915–2931 (2008)
63. G.V. Milovanović, D.S. Mitrinović, T.M. Rassias, Topics in Polynomials: Extremal Problems,
Inequalities, Zeros (World Scientific Publishing Co., Inc., River Edge, 1994)
64. G.V. Milovanović, M.S. Pranić, M.M. Spalević, Quadrature with multiple nodes, power
orthogonality, and moment-preserving spline approximation, Part II. Appl. Anal. Discrete
Math. 13, 1–27 (2019)
65. G.V. Milovanović, T.M. Rassias (eds.), Analytic Number Theory, Approximation Theory, and
Special Functions. In Honor of Hari M. Srivastava (Springer, New York, 2014)
66. G.V. Milovanović, T.M. Rassias, Inequalities connected with trigonometric sums, in Constantin
Carathéodory: an International Tribute, vol. II (World Science Publications, Teaneck, 1991),
pp. 875–941
67. G.V. Milovanović, M.M. Spalević, M.S. Pranić, On the remainder term of Gauss-Radau
quadratures for analytic functions. J. Comput. Appl. Math. 218, 281–289 (2008)
68. G.V. Milovanović, M.P. Stanić, Quadrature rules with multiple nodes, in Mathematical
Analysis, Approximation Theory and Their Applications, ed. by T.M. Rassias, V. Gupta
(Springer, Cham, 2016), pp. 435–462
69. S.E. Notaris, The error norm of Gauss-Radau quadrature formulae for Chebyshev weight
functions. BIT Numer. Math. 50, 123–147 (2010)
70. A.P. Prudnikov, Y.A. Brychkov, O.I. Marichev, Integrals and Series, vol. 1. Elementary
functions (Gordon and Breach Science Publishers, New York, 1986)
71. A.P. Prudnikov, Y.A. Brychkov, O.I. Marichev, Integrals and Series, vol. 2 (Gordon and Breach,
New York, 1986)
72. H. Rademacher, Zur Theorie der Modulfunktionen. J. Reine Angew. Math. 167, 312–336
(1932)
73. H. Rademacher, Über eine Reziprozitätsformel aus der Theorie der Modulfunktionen. Mat.
Fiz. Lapok 40, 24–34 (1933) (Hungarian)
74. H. Rademacher, Die reziprozitatsformel für Dedekindsche Summen. Acta Sci. Math. (Szeged)
12(B), 57–60 (1950)
228 G. V. Milovanović and Y. Simsek

75. H. Rademacher, Topics in Analytic Number Theory, Die Grundlehren der Math. Wis-
senschaften, Band 169 (Springer, Berlin, 1973)
76. H. Rademacher, E. Grosswald, Dedekind Sums, The Carus Mathematical Monographs, vol. 16
(The Mathematical Association of America, Washington, DC, 1972)
77. M.T. Rassias, L. Toth, Trigonometric representations of generalized Dedekind and Hardy sums
via the discrete Fourier transform, in Analytic Number Theory: In Honor of Helmut Maier’s
60th Birthday, ed. by C. Pomerance, M.T. Rassias (Springer, Cham, 2015), pp. 329–343
78. T. Schira, The remainder term for analytic functions of Gauss-Lobatto quadratures. J. Comput.
Appl. Math. 76, 171–193 (1996)
79. Y.G. Shi, On Turán quadrature formulas for the Chebyshev weight. J. Approx. Theory 96,
101–110 (1999)
80. Y.G. Shi, On Gaussian quadrature formulas for the Chebyshev weight. J. Approx. Theory 98,
183–195 (1999)
81. Y.G. Shi, Generalized Gaussian quadrature formulas with Chebyshev nodes. J. Comput. Math.
17(2), 171–178 (1999)
82. B. Schoeneberg, Elliptic Modular Functions: An Introduction, Die Grundlehren der mathema-
tischen Wissenschaften, Band 203 (Springer, New York/Heidelberg, 1974)
83. R. Sczech, Dedekind summen mit elliptischen Funktionen. Invent. Math. 76, 523–551 (1984)
84. Y. Simsek, Relations between theta-functions Hardy sums Eisenstein and Lambert series in the
transformation formula of log ηg,h (z). J. Number Theory 99, 338–360 (2003)
85. Y. Simsek, Generalized Dedekind sums associated with the Abel sum and the Eisenstein and
Lambert series. Adv. Stud. Contemp. Math. (Kyungshang) 9(2), 125–137 (2004)
86. Y. Simsek, On generalized Hardy sums S5 (h, k). Ukrainian Math. J. 56(10), 1434–1440 (2004)
87. Y. Simsek, q-Dedekind type sums related to q-zeta function and basic L-series. J. Math. Anal.
Appl. 318(1), 333–351 (2006)
88. Y. Simsek, On analytic properties and character analogs of Hardy sums. Taiwanese J. Math.
13(1), 253–268 (2009)
89. Y. Simsek, Special functions related to Dedekind-type DC-sums and their applications. Russ.
J. Math. Phys. 17(4), 495–508 (2010)
90. Y. Simsek, D. Kim, J.K. Koo, Oon elliptic analogue to the Hardy sums. Bull. Korean Math.
Soc. 46(1), 1–10 (2009)
91. R. Sitaramachandrarao, Dedekind and Hardy sums. Acta Arith. 48, 325–340 (1987)
92. H.M. Srivastava, A note on the closed-form summation of some trigonometric series. Kobe J.
Math. 16(2), 177–182 (1999)
93. H.M. Srivastava, A. Pinter, Remarks on some relationships between the Bernoulli and Euler
polynomials. Appl. Math. Lett. 17(4), 375–380 (2004)
94. H.M. Srivastava, J. Choi, Series Associated with the Zeta and Related Functions (Kluwer
Academic Publishers, Dordrecht/Boston/London, 2001)
95. T.J. Stijeltes, Note sur quelques formules pour l’évaluation de certaines intégrales, Bul. Astr.
Paris 1, 568–569 (1884) [Oeuvres I, 426–427]
96. S.K. Suslov, Some Expansions in basic Fourier series and related topics. J. Approx. Theory
115(2), 289–353 (2002)
97. P. Turán, On the theory of the mechanical quadrature. Acta Sci. Math. Szeged 12, 30–37 (1950)
98. A.H. Turetzkii, On quadrature formulae that are exact for trigonometric polynomials. East
J. Approx. 11(3), 337–335 (2005). [Translation in English from Uchenye Zapiski, Vypusk 1
(149). Seria Math. Theory of Functions, Collection of papers, Izdatel’stvo Belgosuniversiteta
imeni V.I. Lenina, Minsk (1959), pp. 31–54]
99. E.T. Wittaker, G.N. Watson, A Course of Modern Analysis, 4th edn. (Cambridge University
Press, Cambridge, 1962)
100. M. Waldschmidt, P. Moussa, J.M. Luck, C. Itzykson, From Number Theory to Physics
(Springer, New York, 1995)
101. D. Zagier, Higher dimensional Dedekind sums. Math. Ann. 202, 149–172 (1973)
On a Half-Discrete Hilbert-Type
Inequality in the Whole Plane with the
Kernel of Hyperbolic Secant Function
Related to the Hurwitz Zeta Function

Michael Th. Rassias, Bicheng Yang, and Andrei Raigorodskii

Abstract Using weight functions, we obtain a half-discrete Hilbert-type inequality


in the whole plane with the kernel of hyperbolic secant function and multi-
parameters. The constant factor related to the Hurwitz zeta function is proved to
be the best possible. We also consider equivalent forms, two kinds of particular
inequalities, the operator expressions and some reverses.

Keywords Half-discrete Hilbert-type inequality · Weight function · Equivalent


form · Operator expression · Hurwitz zeta function

2000 Mathematics Subject Classification 26D15, 30A10, 47A05

M. T. Rassias ()
Institute of Mathematics, University of Zurich, Zurich, Switzerland
Moscow Institute of Physics and Technology, Dolgoprudny, Russia
Institute for Advanced Study, Program in Interdisciplinary Studies, Princeton, NJ, USA
e-mail: [email protected]
B. Yang
Department of Mathematics, Guangdong University of Education,
Guangzhou, Guangdong, P. R. China
e-mail: [email protected]; [email protected]
A. Raigorodskii
Moscow Institute of Physics and Technology, Dolgoprudny, Russia
Moscow State University, Moscow, Russia
Buryat State University, Ulan-Ude, Russia
Caucasus Mathematical Center, Adyghe State University, Maykop, Russia
e-mail: [email protected]

© Springer Nature Switzerland AG 2020 229


A. Raigorodskii, M. T. Rassias (eds.), Trigonometric Sums and Their Applications,
https://doi.org/10.1007/978-3-030-37904-9_11
230 M. T. Rassias et al.

1 Introduction

Assuming that p > 1, 1


p + 1
q = 1, am , bn > 0,


 ∞

p q
0< am < ∞, 0 < bn < ∞,
m=1 n=1

we have the following discrete Hardy-Hilbert inequality (cf. [1]):


∞ 
 ∞
1  ∞ 1
 am bn π  p
p  q
q
< am bn , (1)
m+n sin(π/p)
n=1 m=1 m=1 n=1

where the constant factor


π
sin(π/p)

is the best possible.


If f (x), g(y) ≥ 0,
 ∞  ∞
0< f p (x)dx < ∞ and 0 < g q (y)dy < ∞,
0 0

then we have the following Hardy-Hilbert integral inequality with the same best
π
possible constant factor sin(π/p) (cf. [2]):

 ∞ ∞  ∞  1  ∞ 1
f (x)g(y) π p
p
q
q
dxdy < f (x)dx g (y)dy .
0 0 x+y sin(π/p) 0 0
(2)
Recently, the following half-discrete Hardy-Hilbert inequality with the same best
possible constant factor was established (cf. [3]):

∞ 
 ∞  ∞  1 

1
q
bn f (x) π p
p
q
dx < f (x)dx bn . (3)
x+n sin(π/p)
n=1 0 0 n=1

Inequalities (1), (2) and (3) are fairly applicable in various domains of mathematical
analysis (cf. [2, 4–6]).
We notice that the inequalities (1)–(3) involve a homogenous kernel of degree
−1. In 2009, a survey of the study of Hilbert-type inequalities with homogeneous
kernels having negative numbers as a degree was presented in [7]. A few inequalities
with homogenous kernels of degree 0 and non-homogenous kernels have been
On a Half-Discrete Hilbert-Type Inequality in the Whole Plane with the Kernel. . . 231

studied in [8–10]. Some other kinds of Hilbert-type inequalities are provided in


[11–21, 24]. It is worth mentioning that all of the above inequalities are built in the
quarter plane of the first quadrant.
In 2007, a Hilbert-type integral inequality in the whole plane was proved by Yang
[22]. Additionally, the following Hilbert-type integral inequality in the whole plane
was established in [23]:
 ∞  ∞1
f (x)g(y)dxdy
−∞ −∞ |1 + xy|
λ

 ∞  1  ∞ 1
p q
p(1− λ2 )−1 p q(1− λ2 )−1 q
< kλ |x| f (x)dx |y| g (y)dy , (4)
−∞ −∞

where the constant factor


   
λ λ λ
kλ = B , + 2B 1 − λ, (0 < λ < 1)
2 2 2

is the best possible.


Furthermore He et al. [25–29] also proved some integral and discrete Hilbert-
type inequalities in the whole plane.
In this chapter, using weight functions, we prove the following new half-discrete
Hilbert-type inequality in the whole plane with the kernel of hyperbolic secant
function
2
sec h(u) := (u ≥ 0)
eu + e−u

and a best possible constant factor:


∞ 
 ∞   γ 
|n|
sech ρ f (x)bn dx
−∞ |x|
|n|=1
 
4 σ σ 1 2 σ 3
< Γ ( ) ζ ( , ) − ζ ( , )
γ (2ρ)σ/γ γ γ 2 2σ/γ γ 4
⎡ ⎤1
 ∞ 1  ∞ q
p
|x|p(1+σ )−1 f p (x)dx ⎣ |n|q(1−σ )−1 bn ⎦ ,
q
× (5)
−∞ |n|=1

where, ρ > 0, 0 < γ < σ ≤ 1, and



 1
ζ (s, a) := (Res > 1; 0 < a ≤ 1)
(k + a)s
k=0
232 M. T. Rassias et al.

is the Hurwitz zeta function. Note that ζ (s) := ζ (s, 1) is the Reimann zeta function
(cf. [30]).
Moreover, an extension of (5) with multi-parameters is given in Theorem 1. The
equivalent forms, two kinds of particular inequalities, the operator expressions and
some reverses are also considered.

2 Weight Functions and Some Lemmas

In what follows, we assume that p ∈ R\{0, 1}, p1 + 1


q = 1, δ ∈ {−1, 1},
α, β ∈ (0, π ), ρ > 0, 0 < γ < σ ≤ 1.
We set
   
|y| + y cos β γ
h(x, y) : = sech ρ
(|x| + x cos α)δ
2
= .
|y|+y cos β
/γ .
|y|+y cos β
/γ (x = 0, y = 0), (6)
ρ −ρ
e (|x|+x cos α)δ +e (|x|+x cos α)δ

wherefrom,
   
y(1 + cos β) γ
h(x, y) = sech ρ (y > 0),
(|x| + x cos α)δ
   
|y| + y cos β γ
h(x, y) = sec h ρ (x > 0),
[x(1 + cos α)]δ

   
|y| + y cos β γ
h(−x, y) = sech ρ (x > 0),
[x(1 − cos α)]δ
   
y(1 − cos β) γ
h(x, −y) = sech ρ (y > 0).
(|x| + x cos α)δ

Lemma 1 We define two weight functions ω(σ, n) and $ (σ, x) as follows:


 ∞ (|n| + n cos β)σ
ω(σ, n) : = h(x, n) dx (|n| ∈ N), (7)
−∞ (|x| + x cos α)1+δσ

 (|x| + x cos α)−δσ
$ (σ, x) : = h(x, n) (x ∈ R\{0}). (8)
(|n| + n cos β)1−σ
|n|=1

Then we have
On a Half-Discrete Hilbert-Type Inequality in the Whole Plane with the Kernel. . . 233

(i)

4 csc2 α σ
ω(σ, n) = kα (σ ) := Γ( )
γ (2ρ)σ/γ γ
 
σ 1 2 σ 3
× (ζ ( , ) − σ/γ ζ ( , )) ∈ R+ (|n| ∈ N); (9)
γ 2 2 γ 4

(ii)

kβ (σ )(1 − θ (σ, x)) < $ (σ, x) < kβ (σ ) (x ∈ R\{0}), (10)

where

2σ/γ
θ (σ, x) :=
2Γ ( σγ )(ζ ( σγ , 12 ) − 2
2σ/γ
ζ ( σγ , 34 ))
. /γ
 ρ 1+cos β
σ
(|x|+x cos α)δ −1
× uγ sech(u)du
0
 
1
= O ∈ (0, 1). (11)
(|x| + x cos α)δσ

Proof We obtain
 0 (|n| + n cos β)σ
ω(σ, n) = h(x, n) dx
−∞ [x(cos α − 1)]1+δσ
 ∞ (|n| + n cos β)σ
+ h(x, n) dx
0 [x(cos α + 1)]1+δσ
 ∞ (|n| + n cos β)σ
= h(−x, n) dx
0 [x(1 − cos α)]1+δσ
 ∞ (|n| + n cos β)σ
+ h(x, n) dx.
0 [x(1 + cos α)]1+δσ

γ 
γ 
|n|+n cos β |n|+n cos β
Setting u = ρ [x(1−cos α)] δ u = ρ [x(1+cos α)] δ in the above first (respec-
tively second) integral, by Lebesgue term by term integrations (cf. [31]), we derive
that
   ∞
1 1 1 σ
−1
ω(σ, n) = + u γ sech(u)du
1 − cos α 1 + cos α γρ σ/γ
0
 σ
−1  ∞ σ
−1
4 csc2 α ∞ e−u u γ du 4 csc2 α ∞  (−1)k u γ
= = du
γρ σ/γ 0 1 + e−2u γρ σ/γ 0 e(2k+1)u
k=0
234 M. T. Rassias et al.

 ∞
∞
4 csc2 α σ
−1
= [e−(4k+1)u − e−(4k+3)u ]u γ du
γρ σ/γ 0 k=0
∞ 
4 csc2 α  ∞ −(4k+1)u σ
−1
= σ/γ
[e − e−(4k+3)u ]u γ du
γρ 0 k=0
∞ 
4 csc2 α  ∞ σ
−1
= σ/γ
(−1)k e−(2k+1)u u γ du (v = (2k + 1)u)
γρ 0
k=0
 ∞ ∞

4 csc2 α σ
−1 (−1)k
= e−v v γ dv
γρ σ/γ 0 (2k + 1)σ/γ
k=0


4 csc2 α σ (−1)k
= Γ( ) ,
γρ σ/γ γ (2k + 1)σ/γ
k=0

where
 ∞
e−v v a−1 dv = Γ (a) (a > 0)
0

is the gamma function (cf. [30]).


Since σγ > 1, we obtain that

 ∞ ∞
(−1)k 1 1
= − 2
(2k + 1)σ/γ (2k + 1)σ/γ (4k + 3)σ/γ
k=0 k=0 k=0
 
1 σ 1 2 σ 3
= σ/γ ζ ( , ) − σ/γ ζ ( , ) ,
2 γ 2 2 γ 4
from which we deduce (9).
We obtain
−∞
 (|x| + x cos α)−δσ
$ (σ, x) = h(x, n)
(|n| + n cos β)1−σ
n=−1

 (|x| + x cos α)−δσ
+ h(x, n)
(|n| + n cos β)1−σ
n=1

(|x| + x cos α)−δσ  h(x, −n)
=
(1 − cos β)1−σ n1−σ
n=1

(|x| + x cos α)−δσ  h(x, n)
+ . (12)
(1 + cos β)1−σ n1−σ
n=1
On a Half-Discrete Hilbert-Type Inequality in the Whole Plane with the Kernel. . . 235

Since

−2γ uγ −1 (eu − e−u )


γ γ
d
sech(uγ ) = < 0 (γ > 0),
(eu + e−u )2
γ γ
du

we observe that for 0 < σ ≤ 1, both

h(x, −y) h(x, y)


1−σ
and
y y 1−σ

are strictly decreasing in y ∈ (0, ∞). By (12) and this decreasing property, we have
 ∞
(|x| + x cos α)−δσ h(x, −y)
$ (σ, x) < dy
(1 − cos β)1−σ 0 y 1−σ
 ∞
(|x| + x cos α)−δσ h(x, y)
+ dy.
(1 + cos β)1−σ 0 y 1−σ
. /γ  . /γ 
y(1−cos β) y(1+cos β)
Setting u = ρ (|x|+x cos α) δ u = ρ (|x|+x cos α) δ in the above first (respec-
tively second) integral, and by carrying out the corresponding simplifications, we
obtain
 
4 csc2 β σ σ 1 2 σ 3
$ (σ, x) < Γ ( ) ζ ( , ) − ζ ( , )
γ (2ρ)σ/γ γ γ 2 2σ/γ γ 4
= kβ (σ ).

By (12) and the decreasing property, we obtain that


 ∞
(|x| + x cos α)−δσ h(x, −y)dy
$ (σ, x) >
(1 − cos β)1−σ 1 y 1−σ
 ∞
(|x| + x cos α)−δσ h(x, y)dy
+ .
(1 + cos β)1−σ 1 y 1−σ
. /γ  . /γ 
y(1−cos β) y(1+cos β)
Still setting u = ρ (|x|+x cos α)δ
u = ρ (|x|+x cos α)δ
in the above first
(respectively second) integral, and by carrying out the corresponding simplifica-
tions, we have
 ∞
1 σ
−1
$ (σ, x) > . /γ uγ sech(u)du
γρ (1 − cos β)
σ/γ
ρ 1−cos β
(|x|+x cos α)δ
 ∞
1 σ
−1
+ σ/γ . /γ uγ sech(u)du
γρ (1 + cos β) ρ 1+cos β
(|x|+x cos α)δ
236 M. T. Rassias et al.

 ∞
2 csc2 β σ
−1
≥ . /γ uγ sech(u)du
γρ σ/γ ρ 1+cos β
(|x|+x cos α)δ

= kβ (σ )(1 − θ (σ, x)) > 0.

We deduce that

lim sech(u) = 1, lim sech(u) = 0


u→0 u→∞

and then 0 < sech(u) ≤ 1 (u ∈ (0, ∞)). Hence, we have

2σ/γ
0 < θ (σ, x) =
2Γ ( σγ )(ζ ( σγ , 12 ) − 2σ/γ
2
ζ ( σγ , 34 ))
. /γ
 ρ 1+cos β
σ
(|x|+x cos α)δ −1
× uγ sech(u)du
0

. /γ
 ρ 1+cos β
2σ/γ (|x|+x cos α)δ σ
−1
≤ uγ du
2Γ ( σγ )(ζ ( σγ , 12 ) − 2σ/γ
2
ζ ( σγ , 34 )) 0
 σ
γ (2ρ)σ/γ 1 + cos β
= ,
2σ Γ ( γ )(ζ ( γ , 2 ) − 2σ/γ ζ ( γ , 4 )) (|x| + x cos α)δ
σ σ 1 2 σ 3

namely, (10) and (11) follow.


Lemma 2 If ε > 0,

 1
Hε (β) := ,
(|n| + n cos β)1+ε
|n|=1

then it holds
1
Hε (β) = (2 csc2 β + o1 (1))(1 + o2 (1)) (ε → 0+ ). (13)
ε

Proof We have

−∞
  ∞
1 1
Hε (β) = +
[n(cos β − 1)] 1+ε [n(cos β + 1)]1+ε
n=−1 n=1
 

1 1 1
= + . (14)
(1 − cos β) 1+ε (1 + cos β) 1+ε n1+ε
n=1
On a Half-Discrete Hilbert-Type Inequality in the Whole Plane with the Kernel. . . 237

By (14) and the decreasing property, we find


  ∞

1 1 1
Hε (β) = + 1+
(1 + cos β)1+ε (1 − cos β)1+ε n1+ε
n=2
   ∞ 
1 1 dy
< + 1+
(1 + cos β)1+ε (1 − cos β)1+ε 1 y 1+ε
1
= (2 csc2 β + o1 (1))(1 + ε),
ε
  ∞
1 1 dy
Hε (β) > +
(1 + cos β)1+ε (1 − cos β)1+ε 1 y 1+ε
1
= (2 csc2 β + o1 (1)).
ε
Hence, we obtain (13).
Lemma 3 For ε > 0, setting

1
Eδ := {x ∈ R\{0}; ≥ 1},
(|x| + x cos α)δ

we have

1 2
Hδ := dx = csc2 α. (15)
Eδ (|x| + x cos α) 1+δε ε

Proof Setting

1
Eδ+ := {x > 0; ≥ 1},
[x(1 + cos α)]δ
1
Eδ− := {x < 0; ≥ 1},
[(−x)(1 − cos α)]δ

it follows that Eδ = Eδ+ ∪ Eδ− and



1 1
Hδ = dx
(1 + cos α)1+δε Eδ+ x 1+δε

1 1
+ dx.
(1 − cos α)1+δε Eδ− (−x)1+δε
238 M. T. Rassias et al.

Setting u = [x(1+cos α)]δ (u = [(−x)(1−cos α)]δ ) in the above first (respectively


second) integral, we obtain
  ∞
1 1 du 2
Hδ = + = csc2 α,
1 + cos α 1 − cos α 1 u1+ε ε

namely, (15) follows.

3 Main Results

Theorem 1 Suppose that p > 1,


1 1
4 σ
Kα,β (σ ) : = kαq (σ )kβp (σ ) = Γ( )
γ (2ρ)σ/γ γ
 
σ 1 2 σ 3 2 2
× ζ ( , ) − σ/γ ζ ( , ) csc q α csc p β. (16)
γ 2 2 γ 4

If f (x), bn ≥ 0, satisfying
 ∞
0< (|x| + x cos α)p(1+δσ )−1 f p (x)dx < ∞,
−∞

 q
0< (|n| + n cos β)q(1−σ )−1 bn < ∞,
|n|=1

then we have the following equivalent inequalities:


∞ 
 ∞    
|n| + n cos β γ
I : = sech ρ f (x)bn dx
−∞ (|x| + x cos α)δ
|n|=1
 ∞ 1
p
< Kα,β (σ ) (|x| + x cos α) p(1+δσ )−1 p
f (x)dx
−∞
⎡ ⎤1


q

×⎣ (|n| + n cos β)q(1−σ )−1 bn ⎦


q
, (17)
|n|=1
On a Half-Discrete Hilbert-Type Inequality in the Whole Plane with the Kernel. . . 239


⎨∞
J1 : = (|n| + n cos β)pσ −1

|n|=1

 ∞    p  p1
|n| + n cos β γ
× sech ρ f (x)dx
−∞ (|x| + x cos α)δ
 ∞ 1
p
< Kα,β (σ ) (|x| + x cos α)p(1+δσ )−1 f p (x)dx , (18)
−∞
 ∞
J2 : = (|x| + x cos α)−qδσ −1
−∞
⎡ ⎤q ⎫ q1

   γ  ⎬
|n| + n cos β
×⎣ sech ρ bn ⎦ dx
(|x| + x cos α)δ ⎭
|n|=1
⎡ ⎤1


q

< Kα,β (σ ) ⎣ (|n| + n cos β)q(1−σ )−1 bn ⎦ .


q
(19)
|n|=1

In particular, for α = β = π
2, we have the following equivalent inequalities:

∞ 
 ∞    
|n| γ
sech ρ f (x)bn dx
−∞ |x|δ
|n|=1
 
4 σ σ 1 2 σ 3
< Γ ( ) ζ ( , ) − σ/γ ζ ( , )
γ (2ρ)σ/γ γ γ 2 2 γ 4
⎛ ⎞1
 ∞ 1  ∞ q
p
× |x| p(1+δσ )−1 p
f (x)dx ⎝ |n| q(1−σ )−1
bn ⎠ ,
q
(20)
−∞ |n|=1
⎧ ⎫1
⎨∞  ∞     p ⎬ p
|n| γ
|n|pσ −1 sech ρ f (x)dx
⎩ −∞ |x|δ ⎭
|n|=1
 
4 σ σ 1 2 σ 3
< Γ ( ) ζ ( , ) − ζ ( , )
γ (2ρ)σ/γ γ γ 2 2σ/γ γ 4
 ∞ 1
p
× |x| p(1+δσ )−1 p
f (x)dx , (21)
−∞
240 M. T. Rassias et al.

⎧ ⎡ ⎤q ⎫ q1
⎨ ∞ ∞
   γ  ⎬
|n| 4 σ
|x|−qδσ −1 ⎣ sech ρ bn ⎦ dx < Γ( )
⎩ −∞ |x|δ ⎭ γ (2ρ)σ/γ γ
|n|=1
⎡ ⎤1
  ∞ q
σ 1 2 σ 3 ⎣  q(1−σ )−1 q ⎦
× ζ ( , ) − σ/γ ζ ( , ) |n| bn . (22)
γ 2 2 γ 4
|n|=1

Proof By Hölder’s inequality with weight (cf. [32]) and (7), we obtain
 ∞ p
h(x, n)f (x)dx
−∞
9 :p
∞ (|x| + x cos α)(1+δσ )/q f (x) (|n| + n cos β)(1−σ )/p
= h(x, n) dx
−∞ (|n| + n cos β)(1−σ )/p (|x| + x cos α)(1+δσ )/q
 ∞ (|x| + x cos α)(1+δσ )(p−1) p
≤ h(x, n) f (x)dx
−∞ (|n| + n cos β)1−σ
9 :p−1
∞ (|n| + n cos β)(1−σ )(q−1)
× h(x, n) dx
−∞ (|x| + x cos α)1+δσ
 ∞
ωp−1 (σ, n) (|x| + x cos α)(1+δσ )(p−1) p
= h(x, n) f (x)dx.
(|n| + n cos β)pσ −1 −∞ (|n| + n cos β)1−σ

Then by (9) and the Lebesgue term by term integration theorem (cf. [31]), in view
of (8), we deduce that
⎡ ⎤1
∞ 
 ∞
p
1
(|x| + x cos α)(1+δσ )(p−1) p
J1 ≤ kα (σ ) ⎣ q
h(x, n) f (x)dx ⎦
−∞ (|n| + n cos β)1−σ
|n|=1
⎡ ⎤1
 ∞ ∞

p
1
⎣ (|x| + x cos α)(1+δσ )(p−1)
= kα (σ )
q
h(x, n) f (x)dx ⎦
p
−∞ |n|=1 (|n| + n cos β)1−σ

1
 ∞ 1
p
= kα (σ )
q
$ (σ, x)(|x| + x cos α) p(1+δσ )−1 p
f (x)dx . (23)
−∞

Hence, by (10), since


1 1
Kα,β (σ ) = kαq (σ )kβp (σ ),

we derive (18).
On a Half-Discrete Hilbert-Type Inequality in the Whole Plane with the Kernel. . . 241

By Hölder’s inequality (cf. [32]), we have


∞ 
  ∞ 
−1 1
I = (|n| + n cos β) p +σ h(x, n)f (x)dx [(|n| + n cos β) p
−σ
bn ]
|n|=1 −∞

⎡ ⎤1


q

≤ J1 ⎣ (|n| + n cos β)q(1−σ )−1 bn ⎦


q
. (24)
|n|=1

Then by (18), we derive (17). On the other hand, assuming that (17) is valid, we set
 ∞ p−1
bn := (|n| + n cos β)pσ −1 h(x, n)f (x)dx (|n| ∈ N).
−∞

Then we obtain
⎡ ⎤1


p

J1 = ⎣ (|n| + n cos β)q(1−σ )−1 bn ⎦


q
.
|n|=1

In view of (23), it follows that J1 < ∞. If J1 = 0, then (18) is trivially valid; if


J1 > 0, then by (17), we have

 q
(|n| + n cos β)q(1−σ )−1 bn
|n|=1
 ∞ 1
p
p
= J1 = I < Kα,β (σ ) (|x| + x cos α) p(1+δσ )−1 p
f (x)dx
−∞
⎡ ⎤1


q

×⎣ (|n| + n cos β)q(1−σ )−1 bn ⎦ ,


q

|n|=1

⎡ ⎤1− 1


q

J1 = ⎣ (|n| + n cos β)q(1−σ )−1 bn ⎦


q

|n|=1
 ∞ 1
p
< Kα,β (σ ) (|x| + x cos α) p(1+δσ )−1 p
f (x)dx ,
−∞

namely, (18) follows, which is equivalent to (17).


242 M. T. Rassias et al.

By Hölder’s inequality with weight, we still obtain


⎡ ⎤q


⎣ h(x, n)bn ⎦
|n|=1
⎡ ⎤q

 (|x| + x cos α)(1+δσ )/q (|n| + n cos β)(1−σ )/p ⎦
=⎣ h(x, n) bn
(|n| + n cos β)(1−σ )/p (|x| + x cos α)(1+δσ )/q
|n|=1
⎡ ⎤q−1

 (|x| + x cos α)(1+δσ )(p−1) ⎦
≤⎣ h(x, n)
(|n| + n cos β)1−σ
|n|=1

 (|n| + n cos β)(1−σ )(q−1) q
× h(x, n) bn
(|x| + x cos α)1+δσ
|n|=1



($ (σ, x))q−1 (|n| + n cos β)(1−σ )(q−1) q
= h(x, n) bn .
(|x| + x cos α)−qδσ −1 (|x| + x cos α)1+δσ
|n|=1

By (10) and the Lebesgue term by term theorem, we have


⎡ ⎤1
 ∞ ∞

q
1
p ⎣ (|n| + n cos β)(1−σ )(q−1) q ⎦
J2 < kα (σ ) h(x, n) bn dx
−∞ |n|=1 (|x| + x cos α)1+δσ

⎡ ⎤1
1 ∞

q

= kα (σ ) ⎣ ω(σ, n)(|n| + n cos β)q(1−σ )−1 bn ⎦ .


p q
(25)
|n|=1

Hence, by (9), we deduce (19).


We have proved that (17) is valid. Setting
⎡ ⎤q−1


f (x) := (|x| + x cos α)−qδσ −1 ⎣ h(x, n)bn ⎦ (x ∈ R\{0}),
|n|=1

it follows that
 ∞ 1
q
J2 = (|x| + x cos α)p(1+δσ )−1 f p (x)dx ,
−∞
On a Half-Discrete Hilbert-Type Inequality in the Whole Plane with the Kernel. . . 243

and in view of (25), we find J2 < ∞. If J2 = 0, then (19) is trivially valid; if


J2 > 0, then by (17), we have
 ∞
(|x| + x cos α)p(1+δσ )−1 f p (x)dx
−∞
 ∞ 1
p
q
= J2 = I < Kα,β (σ ) (|x| + x cos α) p(1+δσ )−1 p
f (x)dx
−∞
⎡ ⎤1


q

×⎣ (|n| + n cos β)q(1−σ )−1 bn ⎦ ,


q

|n|=1

 ∞ 1− 1
p
J2 = (|x| + x cos α) p(1+δσ )−1 p
f (x)dx
−∞
⎡ ⎤1


q

< Kα,β (σ ) ⎣ (|n| + n cos β)q(1−σ )−1 bn ⎦ ,


q

|n|=1

namely, (19) follows.


On the other hand, assuming that (19) is valid, by Hölder’s inequality and
similarly to as we derived (24), we have
 ∞ 1
p
I≤ (|x| + x cos α)p(1+δσ )−1 f p (x)dx J2 . (26)
−∞

Then by (19), we obtain (17), which is equivalent to (19).


Therefore, inequalities (17), (18) and (19) are equivalent.
Theorem 2 With respect to the assumptions of Theorem 1, the constant factor
Kα,β (σ ) in (17), (18) and (19) is the best possible.
Proof For 0 < ε < qσ, we set !
σ =σ− ε
q (∈ (0, 1)),
 1
ε )+1 , x ∈ Eδ ,
f!(x) :=
δ(σ + p
(|x|+x cos α)
0, x ∈ R\Eδ ,

and
ε
! (σ − q )−1
bn := (|n| + n cos β) (|n| ∈ N).
244 M. T. Rassias et al.

By (13) and (15), we obtain that


 ∞ 1
p
I!1 := (|x| + x cos α)p(1+δσ )−1 f!p (x)dx
−∞
⎡ ⎤1


q

×⎣ (|n| + n cos β)q(1−σ )−1!


bn ⎦
q

|n|=1
⎡ ⎤1
 ∞ 1 ∞

q
dx p
⎣ 1 ⎦
=
−∞ (|x| + x cos α)δε+1 (|n| + n cos β)ε+1
|n|=1

1 1 .
p
/1
q
≤ 2 csc2 α (2 csc2 β + o1 (1))(1 + o2 (1)) .
ε
By (10), we have
∞ 
 ∞
I! : = h(x, n)f!(x)!
bn dx
|n|=1 −∞
 ∞
 (|x| + x cos α)−δ(!
σ +ε)−1
= h(x, n) dx
Eδ |n|=1 (|n| + n cos β) σ
1−!

 
$ (!
σ , x) 1 − θ (!
σ , x)
= dx ≥ k β (!
σ ) dx
Eδ (|x| + x cos α) δε+1
Eδ (|x| + x cos α)δε+1
9  :
dx dx
= kβ (!
σ) − δ(σ + pε )+1
Eδ (|x| + x cos α)
δε+1
Eδ O((|x| + x cos α) )
1 ε
= kβ (σ − )(2 csc2 α − εO(1)).
ε q

If the constant factor Kα,β (σ ) in (17) is not the best possible, then, there exists
a positive number k, with Kα,β (σ ) > k, such that (17) is valid when replacing
Kα,β (σ ) by k. Thus, in particular, we have εI˜ < εk I!1 , namely,

ε
kβ (σ −
)(2 csc2 α − εO(1))
q
 1 . /1
p q
< k · 2 csc2 α (2 csc2 β + o1 (1))(1 + o2 (1)) .

It follows that
2 2
2kβ (σ ) csc2 α ≤ 2k csc p α csc q β (ε → 0+ ),
On a Half-Discrete Hilbert-Type Inequality in the Whole Plane with the Kernel. . . 245

namely,
 
4 σ σ 1 2 σ 3 2 2
Kα,β (σ ) = σ/γ
Γ ( ) ζ ( , ) − σ/γ
ζ ( , ) csc q α csc p β
γ (2ρ) γ γ 2 2 γ 4
≤ k.

This is a contradiction. Hence, the constant factor Kα,β (σ ) in (17) is the best
possible.
The constant factor Kα,β (σ ) in (18) (respectively (19)) is still the best possible.
Otherwise, we would reach a contradiction by (24) (or (26)) that the constant factor
Kα,β (σ ) in (17) is not the best possible.

4 Operator Expressions

Suppose that p > 1. We set the following functions:

Φ(x) := (|x| + x cos α)p(1+δσ )−1 and Ψ (n) := (|n| + n cos β)q(1−σ )−1 ,

wherefrom

Φ 1−q (x) = (|x| + x cos α)−qδσ −1 ,

Ψ 1−p (n) = (|n| + n cos β)pσ −1 (x ∈ R\{0}, |n| ∈ N).

Define the following real weight normed linear spaces:


  1 
∞ p
Lp,Φ (R) : = f ; ||f ||p,Φ := Φ(x)|f (x)| dx p
<∞ ,
−∞
  1 
∞ q
Lq,Φ 1−q (R) : = h; ||h||q,Φ 1−q := Φ 1−q q
(x)|h(x)| dx <∞ ,
−∞
⎧ ⎛ ⎞1 ⎫

⎨ ∞ q ⎪


lq,Ψ ∞
: = b = {bn }|n|=1 ; ||b||q,Ψ := ⎝ Ψ (n)|bn | q⎠
<∞

⎩ ⎪

|n|=1
⎧ ⎛ ⎞1 ⎫

⎨ ∞ p ⎪

lp,Ψ 1−p : = c = {cn }∞
|n|=1 ; ||c|| 1−p := ⎝ Ψ 1−p
(n)|cn | p⎠
< ∞ .


p,Ψ


|n|=1
246 M. T. Rassias et al.

(a) In view of Theorem 1, for f ∈ Lp,Φ (R), setting


 ∞
H (1) (n) := h(x, n)|f (x)|dx (|n| ∈ N),
−∞

by (18), we have
⎡ ⎤1


p

||H (1)
||p,Ψ 1−p =⎣ Ψ 1−p
(n)(H (1)
(n)) p⎦
< Kα,β (σ )||f ||p,Φ < ∞,
|n|=1
(27)
namely, H (1) ∈ lp,Ψ 1−p .
Definition 1 Define a Hilbert-type operator in the whole plane

T (1) : Lp,Φ (R) → lp,Ψ 1−p

as follows: For any f ∈ Lp,Φ (R), there exists a unique representation

T (1) f = H (1) ∈ lp,Ψ 1−p ,

satisfying

(T (1) f )(n) = H (1) (n),

for any |n| ∈ N.


In view of (27), it follows that

||T (1) f ||p,Ψ 1−p = ||H (1) ||p,Ψ 1−p ≤ Kα,β ||f ||p,Φ ,

and then the operator T (1) is bounded, satisfying

||T (1) f ||p,Ψ 1−p


||T (1) || = sup ≤ Kα,β (σ ).
f ( =θ)∈Lp,Φ (R) ||f ||p,Φ

In virtue of the fact that the constant factor Kα,β (σ ) in (27) is the best possible, we
have
4 σ
||T (1) || = Kα,β (σ ) = σ/γ
Γ( )
γ (2ρ) γ
 
σ 1 2 σ 3 2 2
× ζ ( , ) − σ/γ ζ ( , ) csc q α csc p β. (28)
γ 2 2 γ 4
On a Half-Discrete Hilbert-Type Inequality in the Whole Plane with the Kernel. . . 247

If we define the formal inner product of T (1) f and b (∈ lq,Ψ ) as follows:


∞ 
 ∞
(T (1)
f, b) := ( h(x, n)f (x)dx)bn
|n|=1 −∞

then we can rewrite the equivalent forms (17) and (18) in the following manner:

(T (1) f, b) < ||T (1) ||·||f ||p,Ψ ||b||q,Φ , ||T (1) f ||p,Ψ 1−p < ||T (1) ||·||f ||p,Φ . (29)

(b) In view of Theorem 1, for b ∈ lq,Ψ , setting



H (2) (x) := h(x, n)bn (x ∈ R\{0}),
|n|=1

by (19) we derive that


 ∞ 1
q
||H (2)
||q,Φ 1−q = Φ 1−q
(x)(H (2) q
(x)) dx < Kα,β (σ )||b||q,Ψ < ∞,
−∞
(30)
namely, H (2) ∈ Lq,Ψ 1−q (R).
Definition 2 Define a Hilbert-type operator in the whole plane

T (2) : lq,Ψ → Lq,Ψ 1−q (R)

as follows: For any b ∈ lq,Ψ , there exists a unique representation

T (2) b = H (2) ∈ Lq,Ψ 1−q (R),

satisfying

(T (2) b)(x) = H (2) (x),

for any x ∈ R.
In view of (30), we have

||T (2) b||q,Φ 1−q = ||H (2) ||q,Φ 1−q ≤ Kα,β (σ )||b||q,Ψ ,

and then the operator T (2) is bounded, satisfying

||T (2) b||q,Φ 1−q


||T (2) || = sup ≤ Kα,β (σ ).
b( =θ)∈lq,Ψ ||b||q,Ψ
248 M. T. Rassias et al.

By the fact that the constant factor Kα,β (σ ) in (30) is the best possible, we have

||T (2) || = Kα,β (σ ) = ||T (1) ||. (31)

If we define the formal inner product of T (2) b and f (∈ Lp,Φ (R)) as follows:
 ∞ ∞

(T (2) b, f ) := h(x, n)bn f (x)dx,
−∞ |n|=1

we can then rewrite the equivalent forms (17) and (19) as follows:

(T (2) b, f ) < ||T (2) ||·||f ||p,Ψ ||b||q,Φ , ||T (2) b||q,Φ 1−q < ||T (2) ||·||b||q,Ψ . (32)

Remark 1 (i) For δ = 1, (20) reduces to (5). If f (−x) = f (x) (x > 0), b−n = bn
(n ∈ N), then (5) reduces to the following half-discrete Hilbert-type inequality
(cf. [6]):
∞ 
 ∞   n γ 
sech ρ f (x)bn dx
x
n=1 0
 
2 σ σ 1 2 σ 3
< Γ ( ) ζ ( , ) − σ/γ ζ ( , )
γ (2ρ)σ/γ γ γ 2 2 γ 4
 ∞ 9
1  ∞
:1
p q
q(1−σ )−1 q
× x p(1+σ )−1 p
f (x)dx n bn . (33)
0 n=1

(ii) For δ = 1, (17) reduces to the following particular inequality with homoge-
neous kernel of degree 0:
∞ 
 ∞    
|n| + n cos β γ
sech ρ f (x)bn dx
−∞ |x| + x cos α
|n|=1
 ∞ 1
p
< Kα,β (σ ) (|x| + x cos α) p(1+σ )−1 p
f (x)dx
−∞
⎡ ⎤1


q

×⎣ (|n| + n cos β)q(1−σ )−1 bn ⎦


q
. (34)
|n|=1

(iii) For δ = −1, (17) reduces to the following particular inequality with non-
homogeneous kernel:
On a Half-Discrete Hilbert-Type Inequality in the Whole Plane with the Kernel. . . 249

∞ 
 ∞  
sech ρ[(|x| + x cos α)(|n| + n cos β)]γ f (x)bn dx
|n|=1 −∞
 ∞ 1
p
< Kα,β (σ ) (|x| + x cos α)p(1−σ )−1 f p (x)dx
−∞
⎡ ⎤1


q

×⎣ (|n| + n cos β)q(1−σ )−1 bn ⎦


q
. (35)
|n|=1

The constant factors in the above inequalities are the best possible.

5 Two Kinds of Equivalent Reverse Inequalities

In the following, for the cases 0 < p < 1 and p < 0, we still use the symbols
||b||q,Φ and ||f ||p,Ψ .
Theorem 3 Suppose that 0 < p < 1,

4 σ
Kα,β (σ ) = σ/γ
Γ( )
γ (2ρ) γ
 
σ 1 2 σ 3 2 2
× ζ ( , ) − σ/γ ζ ( , ) csc q α csc p β.
γ 2 2 γ 4

Let f (x), bn ≥ 0, satisfy 0 < ||f ||p,Ψ < ∞, 0 < ||b||q,Φ < ∞, then we have the
following equivalent inequalities:
∞ 
 ∞    
|n| + n cos β γ
I= sech ρ f (x)bn dx
−∞ (|x| + x cos α)δ
|n|=1

 ∞ 1
p
> Kα,β (σ ) (1 − θ (σ, x))(|x| + x cos α) p(1+δσ )−1 p
f (x)dx ||b||q,Φ ,
−∞
(36)

⎨∞
J1 = (|n| + n cos β)pσ −1

|n|=1

 ∞     p  p1
|n| + n cos β γ
× sech ρ f (x)dx
−∞ (|x| + x cos α)δ
250 M. T. Rassias et al.

 ∞ 1
p
> Kα,β (σ ) (1 − θ (σ, x))(|x| + x cos α) p(1+δσ )−1 p
f (x)dx , (37)
−∞

 ∞ (1 − θ (σ, x))1−q
J!2 :=
−∞ (|x| + x cos α)qδσ +1
⎡ ⎤q ⎫ q1

   γ  ⎬
|n| + n cos β
×⎣ sech ρ bn ⎦ dx
(|x| + x cos α)δ ⎭
|n|=1

> Kα,β (σ )||b||q,Φ , (38)

where the constant factor Kα,β (σ ) is the best possible.


Proof Similarly, by the reverse Hölder inequality (cf. [32]) and (7), we obtain that
 ∞ p
h(x, n)f (x)dx
−∞
 ∞
ωp−1 (σ, n) (|x| + x cos α)(1+δσ )(p−1) p
≥ h(x, n) f (x)dx.
(|n| + n cos β)pσ −1 −∞ (|n| + n cos β)1−σ

In view of (9), by the Lebesgue term by term integration theorem (cf. [31]) and (8),
we deduce that

1
 ∞ 1
p
J1 ≥ kαq (σ ) $ (σ, x)(|x| + x cos α)p(1+δσ )−1 f p (x)dx . (39)
−∞

Hence, by (10), we have (37).


By the reverse Hölder inequality (cf. [32]), we still have
⎡ ⎤1


q

I ≥ J1 ⎣ (|n| + n cos β)q(1−σ )−1 bn ⎦


q
. (40)
|n|=1

In view of (37), we obtain (36).


On the other hand, assuming that (36) is valid, we set
 ∞ p−1
bn := (|n| + n cos β)pσ −1 h(x, n)f (x)dx (|n| ∈ N).
−∞

Then we get
On a Half-Discrete Hilbert-Type Inequality in the Whole Plane with the Kernel. . . 251

⎡ ⎤1


p

J1 = ⎣ (|n| + n cos β)q(1−σ )−1 bn ⎦ .


q

|n|=1

In view of (39), it follows that J1 > 0. If J1 = ∞, then (37) is trivially valid; if


J1 < ∞, then by (36), we have
q p
||b||q,Φ = J1 = I
 ∞ 1
p
> Kα,β (σ ) (1 − θ (σ, x))(|x| + x cos α) p(1+δσ )−1 p
f (x)dx ||b||q,Φ ,
−∞

 ∞ 1
p
q−1
||b||q,Φ = J1 > Kα,β (σ ) (1 − θ (σ, x))(|x| + x cos α) p(1+δσ )−1 p
f (x)dx ,
−∞

namely, (37) holds, which is equivalent to (36).


Similarly to as we obtained (39), we deduce that
⎡ ⎤1
1 ∞

q

J!2 > kα (σ ) ⎣ ω(σ, n)(|n| + n cos β)q(1−σ )−1 bn ⎦


p q
. (41)
|n|=1

Hence, by (9), we have (38). We have proved that (36) is valid. Setting
⎡ ⎤q−1


(1 − θ (σ, x))1−q ⎣
f (x) := h(x, n)bn ⎦ (x ∈ R\{0}),
(|x| + x cos α)qδσ +1
|n|=1

it then follows that


 ∞ 1
q
J!2 = (1 − θ (σ, x))(|x| + x cos α) p(1+δσ )−1 p
f (x)dx ,
−∞

and in view of (41), we get that J!2 > 0. If J!2 = ∞, then (38) is trivially valid; if
J!2 < ∞, then by (36), we have
 ∞
(1 − θ (σ, x))(|x| + x cos α)p(1+δσ )−1 f p (x)dx = J!2 = I
q
−∞
 ∞ 1
p
> Kα,β (σ ) (1 − θ (σ, x))(|x| + x cos α) p(1+δσ )−1 p
f (x)dx ||b||q,Φ ,
−∞
252 M. T. Rassias et al.

 ∞ 1− 1
p
J!2 = (1 − θ (σ, x))(|x| + x cos α) p(1+δσ )−1 p
f (x)dx > Kα,β (σ )||b||q,Φ ,
−∞

namely (38) follows. On the other hand, assuming that (38) is valid, by the reverse
Hölder inequality (cf. [32]), we obtain
 ∞ 1
p
I≥ (1 − θ (σ, x))(|x| + x cos α)p(1+δσ )−1 f p (x)dx J!2 . (42)
−∞

Then by (38), we have (16), which is equivalent to (38).


Therefore, inequalities (36), (37) and (38) are equivalent.
For ε > 0, we set !
σ = σ + pε (> γ ),
 1
ε )+1 , x ∈ Eδ ,
f!(x) :=
δ(σ + p
(|x|+x cos α)
0, x ∈ R\Eδ ,

and
ε
! (σ − q )−1
bn := (|n| + n cos β) (|n| ∈ N).

Then by (13) and (15), we find


 ∞ 1
p
I!1 : = (1 − θ (σ, x))(|x| + x cos α)p(1+δσ )−1 f!p (x)dx
−∞
⎡ ⎤1


q

×⎣ (|n| + n cos β)q(1−σ )−1!


bn ⎦
q

|n|=1
⎡ ⎤1
 ∞ 1 ∞

q
(1 − θ (σ, x))dx p
⎣ 1 ⎦
=
−∞ (|x| + x cos α)δε+1 (|n| + n cos β)ε+1
|n|=1
 ∞  1
2 dx p
= csc2 α − δ(σ +ε)+1 )
ε −∞ O((|x| + x cos α)
⎡ ⎤1
∞ q
1
×⎣ ⎦
(|n| + n cos β)ε+1
|n|=1

1 1 .
p
/1
q
= 2 csc2 α − εO(1) (2 csc2 β + o1 (1))(1 + o2 (1)) .
ε
On a Half-Discrete Hilbert-Type Inequality in the Whole Plane with the Kernel. . . 253

By (10), we still have


∞ 
 ∞
I! : = h(x, n)f!(x)!
bn dx
|n|=1 −∞
∞ 
 (σ − qε )−1
(|n| + n cos β)
= h(x, n) δ(σ + pε )+1
dx
|n|=1 Eδ (|x| + x cos α)
∞ 
 ∞ σ −ε)−1
(|n| + n cos β)(!
≤ h(x, n) σ +1
dx
(|x| + x cos α)δ!
|n|=1 −∞

 ∞

ω(!σ , n) 1
= = k α (!
σ )
(|n| + n cos β)ε+1 (|n| + n cos β)ε+1
|n|=1 |n|=1

1 ε
= kα (σ + )(2 csc2 β + o1 (1))(1 + o2 (1)).
ε p

If the constant factor Kα,β (σ ) in (37) is not the best possible, then there exists
a positive number k, with Kα,β (σ ) < k, such that (37) is valid when replacing
Kα,β (σ ) by k. Thus in particular, we have εI˜ > εk I!1 , namely,

ε
kα (σ + )(2 csc2 β + o1 (1))(1 + o2 (1))
p
 1 . /1
p q
> k · 2 csc2 α − εO(1) (2 csc2 β + o1 (1))(1 + o2 (1)) .

It follows that
2 2
2kα (σ ) csc2 β ≥ 2k csc p α csc q β (ε → 0+ ),

namely

4 σ
Kα,β (σ ) = Γ( )
γ (2ρ)σ/γ γ
 
σ 1 2 σ 3 2 2
× ζ ( , ) − σ/γ ζ ( , ) csc q α csc p β ≥ k.
γ 2 2 γ 4

This is a contradiction. Hence, the constant factor Kα,β (σ ) in (36) is the best
possible.
The constant factor Kα,β (σ ) in (37) (respectively (38)) is still the best possible.
Otherwise, we would reach a contradiction by (40) (or (42)) that the constant factor
Kα,β (σ ) in (36) is not the best possible.
254 M. T. Rassias et al.

Theorem 4 Suppose that p < 0,

4 σ
Kα,β (σ ) = σ/γ
Γ( )
γ (2ρ) γ
 
σ 1 2 σ 3 2 2
× ζ ( , ) − σ/γ ζ ( , ) csc q α csc p β.
γ 2 2 γ 4

Let f (x), bn ≥ 0, satisfy 0 < ||f ||p,Ψ , ||b||q,Φ < ∞, then we have the following
equivalent inequalities:
∞ 
 ∞    
|n| + n cos β γ
I= sech ρ f (x)bn dx > Kα,β (σ )||f ||p,Ψ ||b||q,Φ ,
−∞ (|x| + x cos α)δ
|n|=1
(43)

⎨∞
J1 = (|n| + n cos β)pσ −1

|n|=1

 ∞     p  p1
|n| + n cos β γ
× sech ρ f (x)dx > Kα,β (σ )||f ||p,Ψ , (44)
−∞ (|x| + x cos α)δ

 ∞ 1
J2 =
−∞ (|x| + x cos α)qδσ +1
⎡ ⎤q ⎫ q1

   γ  ⎬
|n| + n cos β
×⎣ sech ρ bn ⎦ dx > Kα,β (σ )||b||q,Φ , (45)
(|x| + x cos α)δ ⎭
|n|=1

where the constant factor Kα,β (σ ) is the best possible.


Proof For p < 0, by the reverse Hölder inequality (cf. [32]) and (7), we obtain that
 ∞ p
h(x, n)f (x)dx
−∞
 ∞
ωp−1 (σ, n) (|x| + x cos α)(1+δσ )(p−1) p
≤ h(x, n) f (x)dx.
(|n| + n cos β)pσ −1 −∞ (|n| + n cos β)1−σ

Then by (9), the Lebesgue term by term integration theorem (cf. [31]) and (8), we
deduce that

1
 ∞ 1
p
J1 ≥ kα (σ )q
$ (σ, x)(|x| + x cos α) p(1+δσ )−1 p
f (x)dx . (46)
−∞
On a Half-Discrete Hilbert-Type Inequality in the Whole Plane with the Kernel. . . 255

Hence, by (10), we have (44).


By the reverse Hölder inequality (cf. [32]), we obtain that
⎡ ⎤1


q

J1 ⎣ (|n| + n cos β)q(1−σ )−1 bn ⎦


q
I≥ . (47)
|n|=1

Then by (44), we have (43).


On the other hand, assuming that (43) is valid, we set
 ∞ p−1
bn := (|n| + n cos β)pσ −1 h(x, n)f (x)dx (|n| ∈ N),
−∞

q
and find J1 = ||b||q,Φ p
. In view of (46), it follows that J1 > 0. If J1 = ∞, then (44)
is trivially valid; if J1 < ∞, then by (43), we have
q p
||b||q,Φ = J1 = I > Kα,β (σ )||f ||p,Ψ ||b||q,Φ ,
q−1
J1 = ||b||q,Φ > Kα,β (σ )||f ||p,Ψ ,

namely, (44) holds, which is equivalent to (43).


We similarly get that
⎡ ⎤1
1 ∞

q

J2 > kβ (σ ) ⎣ ω(σ, n)(|n| + n cos β)q(1−σ )−1 bn ⎦


p q
. (48)
|n|=1

Hence, by (9), we deduce (38). We have proved that (43) is valid. Setting
⎛ ⎞q−1


1 ⎝
f (x) := h(x, n)bn ⎠ (x ∈ R\{0}),
(|x| + x cos α)qδσ +1
|n|=1

p
it follows that J2 = ||f ||p,Ψ
q
and in view of (48), we find J2 > 0. If J2 = ∞, then
(45) is trivially valid; if J2 < ∞, then by (43), we have
p q
||f ||p,Ψ = J2 = I > Kα,β (σ )||f ||p,Ψ ||b||q,Φ ,
p−1
J2 = ||f ||p,Ψ > Kα,β (σ )||b||q,Φ ,

namely (45) follows.


On the other hand, assuming that (45) is valid, by the reverse Hölder’s inequality
(cf. [32]), we obtain
256 M. T. Rassias et al.

 ∞ 1
p
I≥ (|x| + x cos α) p(1+δσ )−1 p
f (x)dx J2 . (49)
−∞

Then by (45), we have (23), which is equivalent to (45).


Therefore, inequalities (43), (44) and (45) are equivalent.
For 0 < ε < |p|(σ − γ ), we set ! σ = σ + pε (> γ ),
 1
ε )+1 , x ∈ Eδ ,
f!(x) :=
δ(σ + p
(|x|+x cos α)
0, x ∈ R\Eδ ,

and
ε
! (σ − q )−1
bn := (|n| + n cos β) (|n| ∈ N).

Then by (13) and (15), we obtain that


 ∞ 1
p
I!1 : = (|x| + x cos α) p(1+δσ )−1 !p
f (x)dx
−∞
⎡ ⎤1


q

×⎣ (|n| + n cos β)q(1−σ )−1!


bn ⎦
q

|n|=1
⎡ ⎤1
 ∞ 1 ∞

q
dx p
⎣ 1 ⎦
=
−∞ (|x| + x cos α)δε+1 (|n| + n cos β)ε+1
|n|=1

1 1 .
p
/1
q
= 2 csc2 α (2 csc2 β + o1 (1))(1 + o2 (1)) .
ε
By (10), we still have
∞ 
 ∞
I! := h(x, n)f!(x)!
bn dx
|n|=1 −∞
∞ 
 (σ − qε )−1
(|n| + n cos β)
= h(x, n) δ(σ + pε )+1
dx
|n|=1 Eδ (|x| + x cos α)
∞ 
 ∞ σ −ε)−1
(|n| + n cos β)(!
≤ h(x, n) σ +1
dx
(|x| + x cos α)δ!
|n|=1 −∞
On a Half-Discrete Hilbert-Type Inequality in the Whole Plane with the Kernel. . . 257


 ∞

ω(!σ , n) 1
= = k α (!
σ )
(|n| + n cos β)ε+1 (|n| + n cos β)ε+1
|n|=1 |n|=1

1 ε
= kα (σ + )(2 csc2 β + o1 (1))(1 + o2 (1))
ε p

If the constant factor Kα,β (σ ) in (43) is not the best possible, then there exists
a positive number k, with Kα,β (σ ) < k, such that (43) is valid when replacing
Kα,β (σ ) by k. Hence in particular, we have εI˜ > εk I!1 , namely

ε
kα (σ + )(2 csc2 β + o1 (1))(1 + o2 (1))
p
 1 . /1
p q
> k · 2 csc2 α (2 csc2 β + o1 (1))(1 + o2 (1)) .

It follows that
2 2
2kα (σ ) csc2 β ≥ 2k csc p α csc q β (ε → 0+ ),

namely
 
4 σ σ 1 2 σ 3 2 2
Kα,β (σ ) = σ/γ
Γ ( ) ζ ( , ) − σ/γ ζ ( , ) csc q α csc p β
γ (2ρ) γ γ 2 2 γ 4
≥ k.

This is a contradiction. Hence, the constant factor Kα,β (σ ) in (43) is the best
possible.
The constant factor Kα,β (σ ) in (44) (respectively (45)) is still the best possible.
Otherwise, we would reach a contradiction by (47) (respectively (49)) that the
constant factor Kα,β (σ ) in (43) is not the best possible.

6 Conclusions

Using weight functions, a half-discrete Hilbert-type inequality in the whole plane


with the kernel of hyperbolic secant function and multi-parameters is established
in Theorem 1. The constant factor related to the Hurwitz zeta function is proved
to be the best possible in Theorem 2. Equivalent forms, two kinds of particular
inequalities, the operator expressions and some reverses are also considered in
Theorem 1, Remark 1, and Theorems 3–4. The lemmas and theorems provide and
extensive account of this type of inequalities.
258 M. T. Rassias et al.

Acknowledgements B. Yang: This work is supported by the National Natural Science Foundation
(No. 61772140), and Science and Technology Planning Project Item of Guangzhou City (No.
201707010229). I would like to express my gratitude for this support.

References

1. G.H. Hardy, Note on a theorem of Hilbert concerning series of positive terms. Proc. Lond.
Math. Soc. 23(2), xlv–xlvi (1925). Records of Proc
2. G.H. Hardy, J.E. Littlewood, G. Pólya, Inequalities (Cambridge University Press, Cambridge,
1934)
3. B.C. Yang, A half-discrete Hilbert’s inequality. J. Guangdong Univ. Edu. 31(3), 1–7 (2011)
4. D.S. Mitrinović, J.E. Pecčarić, A.M. Fink, Inequalities Involving Functions and their Integrals
and Deivatives (Kluwer Academic, Boston, 1991)
5. B.C. Yang, The Norm of Operator and Hilbert-Type Inequalities (Science Press, Beijing, 2009)
6. B.C. Yang, L. Debnath, Half-Discrete Hilbert-Type Inequalitiea (World Scientific Publishing,
Singapore, 2014)
7. B.C. Yang, A survey of the study of Hilbert-type inequalities with parameters. Adv. Math.
38(3), 257–268 (2009)
8. B.C. Yang, On the norm of an integral operator and applications. J. Math. Anal. Appl. 321,
182–192 (2006)
9. J.S. Xu, Hardy-Hilbert’s inequalities with two parameters. Adv. Math. 36(2), 63–76 (2007)
10. D.M. Xin, A Hilbert-type integral inequality with the homogeneous Kernel of zero degree.
Math. Theory Appl. 30(2), 70–74 (2010)
11. G.V. Milovanovic, M.T. Rassias, Some properties of a hypergeometric function which appear
in an approximation problem. J. Glob. Optim. 57, 1173–1192 (2013)
12. M. Krnić, J. Pečarić, General Hilbert’s and Hardy’s inequalities. Math. Inequal. Appl. 8(1),
29–51 (2005)
13. I. Perić, P. Vuković, Multiple Hilbert’s type inequalities with a homogeneous kernel. Banach J.
Math. Anal. 5(2), 33–43 (2011)
14. Q. Huang, A new extension of Hardy-Hilbert-type inequality. J. Inequal. Appl. 2015, 397
(2015)
15. B. He, A multiple Hilbert-type discrete inequality with a new kernel and best possible constant
factor. J. Math. Anal. Appl. 431, 990–902 (2015)
16. V. Adiyasuren, T. Batbold, M. Krnić, Multiple Hilbert-type inequalities involving some
differential operators. Banach J. Math. Anal. 10(2), 320–337 (2016)
17. M.T. Rassias, B.C. Yang, On half-discrete Hilbert’s inequality. Appl. Math. Comput. 220, 75–
93 (2013)
18. M.T. Rassias, B.C. Yang, A multidimensional half – discrete Hilbert – type inequality and the
Riemann zeta function. Appl. Math. Comput. 225, 263–277 (2013)
19. M.T. Rassias, B.C. Yang, A multidimensional Hilbert-type integral inequality related to the
Riemann zeta function, in Applications of Mathematics and Informatics in Science and
Engineering (Springer, New York, 2014), pp. 417–433
20. M.T. Rassias, B.C. Yang, On a multidimensional half – discrete Hilbert – type inequality related
to the hyperbolic cotangent function. Appl. Math. Comput. 242, 800–813 (2014)
21. M.T. Rassias, B.C. Yang, A. Raigorodskii, Two kinds of the reverse Hardy-type integral
inequalities with the equivalent forms related to the extended Riemann zeta function. Appl.
Anal. Discret. Math. 12, 273–296 (2018)
22. B.C. Yang, A new Hilbert-type integral inequality. Soochow J. Math. 33(4), 849–859 (2007)
23. B.C. Yang, A new Hilbert-type integral inequality with some parameters. J. Jilin Univ. (Science
Edition) 46(6), 1085–1090 (2008)
On a Half-Discrete Hilbert-Type Inequality in the Whole Plane with the Kernel. . . 259

24. B.C. Yang, A Hilbert-type integral inequality with a non-homogeneous kernel. J. Xiamen Univ.
(Natural Science) 48(2), 165–169 (2008)
25. Z. Zeng, Z.T. Xie, On a new Hilbert-type integral inequality with the homogeneous kernel of
degree 0 and the integral in whole plane. J. Inequal. Appl. 2010, 9. Article ID 256796 (2010)
26. Q.L. Huang, S.H. Wu, B.C. Yang, Parameterized Hilbert-type integral inequalities in the whole
plane. Sci. World J. 2014, 8. Article ID 169061 (2014)
27. Z. Zeng, K.R.R. Gandhi, Z.T. Xie, A new Hilbert-type inequality with the homogeneous kernel
of degree −2 and with the integral. Bull. Math. Sci. Appl. 3(1), 11–20 (2014)
28. Z.H. Gu, B.C. Yang, A Hilbert-type integral inequality in the whole plane with a non-
homogeneous kernel and a few parameters. J. Inequal. Appl. 2015, 314 (2015)
29. D.M. Xin, B.C. Yang, Q. Chen, A discrete Hilbert-type inequality in the whole plane. J.
Inequal. Appl. 2016, 133 (2016)
30. Z.Q. Wang, D.R. Guo, Inteoduction to Special Functions (Science Press, Beijing, 1979)
31. J.C. Kuang, Real and Functional Analysis(Continuation), vol. 2 (Higher Education Press,
Beijing, 2015)
32. J.C. Kuang, Applied Inequalities (Shangdong Science and Technology Press, Jinan, 2004)
A Remark on Sets with Small Wiener
Norm

I. D. Shkredov

Abstract We show that any set with small Wiener norm has small multiplicative
energy. It gives some new bounds for Wiener norm for sets with small product set.
Also, we prove that any symmetric subset S of an abelian group has a nonzero
Fourier coefficient of size (|S|1/3 ).

Keywords Exponential sums · Wiener norm · Sum-product · Multiplicative


subgroups

1 Introduction

We consider the abelian group Fp = Z/pZ, where p is a prime number. Denote the
Fourier transform of a complex function on Fp to be a new function

fˆ(γ ) = f (x)ep (−xγ ) ,
x∈Fp

where ep (u) = exp(2π iu/p) (we note that ep is correctly defined for u ∈ Fp ). It
is known that the function f can be reconstructed from fˆ by the inverse Fourier
transform

This work is supported by the Russian Science Foundation under grant 19–11–00001.

I. D. Shkredov ()
Steklov Mathematical Institute, Moscow, Russia
IITP RAS, Moscow, Russia
MIPT, Dolgoprudnii, Russia

© Springer Nature Switzerland AG 2020 261


A. Raigorodskii, M. T. Rassias (eds.), Trigonometric Sums and Their Applications,
https://doi.org/10.1007/978-3-030-37904-9_12
262 I. D. Shkredov

1  ˆ
f (x) = f (γ )ep (xγ ). (1)
p
γ ∈Fp

We define the Wiener norm of a function f as



f A(Fp ) = p−1 fˆ1 = p−1 |fˆ(γ )| .
γ ∈Fp

By 1S , S ⊂ Fp denote the characteristic function of a certain set S. The problem of


finding lower bounds for f A(Fp ) for f equals 1S for an arbitrary set S ⊆ Fp , |S| 
p/2 is called the modular Littlewood problem and many results in this direction
were obtained by various authors [1–7, 18] (for the original continues Littlewood
problem see [8, 9]). In this paper we find a new property of sets with small Wiener
norm. For any set S ⊆ Fp consider the multiplicative energy E× (S) of S, namely,
E× (S) := |{ab = cd : a, b, c, d ∈ S}|. We show that if S has small Wiener norm,
then it has small multiplicative energy as well.
Theorem 1 Let S ⊆ Fp be a set, and K := 1S A(Fp ) . Suppose that |S|5  K 2 p3 .
Then

E× (S)  K 4/3 |S|8/3 , (2)

and in any case

|S|4
E× (S)  + S2A(Fp ) |S|p .
p−1

All logarithms in this paper are to base 2. The signs and & are the usual
Vinogradov symbols. If we have a set A, then we will write a  b or b  a if
a = O(b · logc |A|), c > 0.
The proof of Theorem 1 uses some results from the sum–product phenomenon,
see [10] and, e.g., [11, 12] and especially the Balog–Wooley decomposition [13].
Theorem 1 implies an interesting statement which separates the prime field from
general finite fields (also, see comments after Corollary 10 below).
Corollary 2 Let S ⊆ Fp be a set, and E× (S)  |S|3 . If |S|5  1S 2A(Fp ) p3 , then

1S A(Fp )  |S|1/4 . (3)

Corollary above improves some results from [1], see Corollaries 2, 4, 5 from the
last paper, where instead of (3) several inequalities of the form |S|γ , γ < 1/2 were
obtained.
A Remark on Sets with Small Wiener Norm 263

The family of sets with small Wiener norm contains a subfamily of symmetric
sets with small negative Fourier coefficients (see, say, Lemma 13 below). It is a
difficult problem to obtain a good lower bound for maximal modulus of negative
Fourier coefficient, see [14, 15]. For example, in [14] Sanders obtained the following
result.
Theorem 3 Let p be a prime number, and S ⊆ Z/pZ be a symmetric set. Suppose
that |S|  p/2, |S| := δp. Then there are some functions c(δ) > 0, d(δ) > 0 such
that

− min ;
1S (x)  c(δ) · |S|d(δ) .
x =0

Here the functions c(δ) and d(δ) tends to zero as δ → 0 (d(δ) turns out to be a
linear on δ). In particular case δ = 1/2 Sanders calculated the number d(δ) which
is turned out to be 1/3. It is easy to see, that the upper bound here is 1/2. We show
that, in contrary, if one wants to estimate from below the maximal value of positive
nonzero Fourier coefficient, then it can be done rather easily.
Theorem 4 Let S ⊆ Fp be a symmetric set, |S|  p/4. Then

max ;
1S (x) & |S|1/3 . (4)
x =0

A similar result holds for any abelian group. A connection of sets with small nonzero
positive/negative Fourier coefficients and its multiplicative energy is discussed in
Section 4, see Corollary 14.
The author is grateful to the reviewer for valuable suggestions and remarks.

2 Definitions

Let G be an abelian group. If G is finite, then denote by N the cardinality of G. For


example, N = p if G = Fp and N = p − 1 if G = F∗p . It is well–known [16] that
; is isomorphic to G in this case. Let f be a function from G to C.
the dual group G
We denote the Fourier transform of f by f;,

f;(ξ ) = f (x)e(−ξ · x) , (5)
x∈G

; We rely on the following basic identities


where e(x) = e2π ix and ξ ∈ G.
264 I. D. Shkredov

 1   ; 2
|f (x)|2 = f (ξ ) . (6)
N
x∈G ;
ξ ∈G

 1 ; 1
f, g := f (x)g(x) = g (ξ ) = f;, ;
f (ξ ); g . (7)
N N
x∈G ;
ξ ∈G

   2
 1   ; 2  2
 f (x)g(y − x) = f (ξ ) ; g (ξ ) . (8)
N
y∈G x∈G ;
ξ ∈G

and
1 ;
f (x) = f (ξ )e(ξ · x) . (9)
N
;
ξ ∈G

If
 
(f ∗ g)(x) := f (y)g(x − y) and (f ◦ g)(x) := f (y)g(y + x)
y∈G y∈G

then

∗ g = f;;
f g and g=;
◦ g = f;c;
f f;
g, (10)

where for a function f : G → C we put f c (x) := f (−x). Clearly, (f ∗ g)(x) =


(g ∗ f )(x) and (f ◦ g)(x) = (g ◦ f )(−x), x ∈ G. Write E+ (A, B) for the additive
energy of two sets A, B ⊆ G (see, e.g., [10]), that is,

E+ (A, B) = |{a1 + b1 = a2 + b2 : a1 , a2 ∈ A, b1 , b2 ∈ B}| .

If A = B we simply write E+ (A) instead of E+ (A, A). More generally, we write

T+
k (A) := |{(a1 , . . . , ak , a1 , . . . , ak ) ∈ A
2k
: a1 + . . . ak = a1 + . . . ak }| .

In the same way one can define the multiplicative energy of two sets A, B ⊆ Fp as

E× (A, B) = |{a1 b1 = a2 b2 : a1 , a2 ∈ A, b1 , b2 ∈ B}| .

Further, by (10), clearly,


  
E+ (A, B) = (A ∗ B)(x)2 = (A ◦ B)(x)2 = (A ◦ A)(x)(B ◦ B)(x) .
x x x
(11)
and by (8),
A Remark on Sets with Small Wiener Norm 265

1  ; 2 ;
E+ (A, B) = |A(ξ )| |B(ξ )|2 . (12)
N
ξ

We write F∗p for Fp \ {0}.

3 On the Multiplicative Energy of Sets with Small Wiener


Norm

The sum-product decomposition of a set by energy was initiated by Balog and


Wooley in [13]. We need modern form of these results from [17, Proposition 2,
Corollary 1] which develop the method from [12].
Lemma 5 Let S ⊆ Fp be a set, |S|6  p2 E× (S). Then there is a set S1 ⊆ S,
|S1 |2  E× (S)/|S| and

E+ (S1 )2 E× (S)3  |S1 |11 |S|3 . (13)

The same result holds if one swaps + with × and vice versa.
Lemma 6 Let S ⊆ Fp be a set, |S|6  p2 E× (S). Then there is a set S ⊆ S,
|S |3  E× (S) and

E+ (S )2 E× (S)3  |S|14 . (14)

The same result holds if one swaps + with × and vice versa.
We need in a result on the energy of subsets of sets with small Wiener norm (also,
see [3, Lemma 4]).
Lemma 7 Let S1 ⊆ S ⊆ G be sets and K := 1S A(G) . Then

|S1 |3
E+ (S, S1 )  ,
K2
and for any k  2 the following holds

|S1 |2k
T+
k (S1 )  .
|S|K 2k−2

Proof By formula (7), we have



|S1 |N = S1 (ξ )1;S (ξ ) .
1< (15)
ξ

Using the Hölder inequality twice as well as identity (6) and (12), we get
266 I. D. Shkredov

⎛ ⎞2

|S1 |4  ⎝KN −1 |1<
S1 (ξ )| |1;S (ξ )|⎠
2

ξ

 K 2 |S1 |N −1 |1<
S1 (ξ )| |1;S (ξ )| = K |S1 |E (S, S1 ).
2 2 2 +

Similarly, returning to (15) and applying the Hölder inequality again, we obtain
⎛ ⎞2k−1
 
(|S1 |N )2k  |1<
S1 (ξ )|
2k
·⎝ |1;S (ξ )|2k/(2k−1)⎠
ξ ξ
 
 |1<
S1 (ξ )| ·
2k
|1;S (ξ )|2 (N K)2k−2 =
ξ ξ

= N 2k K 2k−2 T+
k (S1 )|S| .

This completes the proof. 



Take k  0 and consider an arbitrary vector s' = (s1 , . . . , sk ) ∈ (F∗p )k . For
any set Q ⊆ Fp and s' put Q× s' (x) := 1Q (x)1Q (s1 x) . . . 1Q (sk x). In particular,
Q× s' ⊆ Q. We can easily bound the multiplicative Fourier coefficients Q̃× s' of the
×
sets Qs' . Recall that the multiplicative Fourier transform of a function f : F∗p → C
is defined as

f˜(χ ) = f (x)χ (x) , (16)
x∈F∗p

where χ is any multiplicative character.


Lemma 8 Let Q ⊆ Fp be a set, k  0 be an integer, and s = (s1 , . . . , sk ) ∈ (F∗p )k
be a vector. Then for any non–principal multiplicative character χ the following
holds

|Q̃×
s' (χ )|  p · 1Q k+1
A(Fp ) . (17)

Proof By the inversion formula, we get



;× (r) =
Q 1Q (x)1Q (s1 x) . . . 1Q (sk x)e(−xr) =
s'
x

1 
= 1< <
Q (z0 )1 <
Q (z1 ) . . . 1Q (zk )e(z0 x + z1 s1 x + · · · + zk sk x − xr) =
pk+1 z0 ,z1 ,...,zk
A Remark on Sets with Small Wiener Norm 267

1 
= Q(z ; 1 ) . . . Q(z
; 0 )Q(z ; k) . (18)
pk z0 ,z1 ,...,zk : z0 +z1 s1 +···+zk sk =r

Further, using the definition of the multiplicative Fourier transform (16) and formula
(7) for the additive Fourier transform, we have for any multiplicative character χ
 1  ;×
Q̃×
s' (χ ) = Q×
s' (x)χ (x) = Qs' (r)G(χ , r) ,
x
p r

where G(χ , r) is the Gauss sum (Q can contain zero or not here because χ (0) =
0 by the definition). Thus, by formula (18) and the well–known estimate for the
absolute value of G(χ , r), we obtain

p  
|Q̃×  ; − z1 s1 + · · · + zk sk )||Q(z
|Q(r ; 1 )| . . . |Q(z
; k )| 
s' (χ )| pk+1 r z1 ,...,zk

√ ; k+1 = √p1Q k+1 .


 p · p−(k+1) Q1 A(Fp )

as required. 

Using Lemmas 5 and 8 we can prove Theorem 1 from the Introduction.
Theorem 9 Let S ⊆ Fp be a set, and K := 1S A(Fp ) . We have

|S|4
E× (S)  + 1S 2A(Fp ) |S|p . (19)
p−1

Further, if |S|5  K 2 p3 , then

E× (S)  K 4/3 |S|8/3 , (20)

and if |S|27  K 6 p17 , then

E× (S)  K 12/17 |S|48/17 . (21)

Proof First of all, notice that by formula (12) in the multiplicative form, combining
with Parseval identity (6) as well as Lemma 8 with k = 0, we see that

1  |S|4 1
E× (S) = |S̃(χ )|4  + · p1S 2A(Fp ) (p − 1)|S|
p−1 χ p−1 p−1

|S|4
= + p1S 2A(Fp ) |S| .
p−1
268 I. D. Shkredov

Further, if (20) takes place, then there is nothing to prove. Otherwise, thanks to
the assumption |S|5  K 2 p3 we see that the condition of Lemma 5 takes place,
namely, |S|6  p2 E× (S). By Lemma 5, we find S1 ⊆ S such that |S1 |2 
E× (S)/|S| and combining this result with Lemma 7, we get

|S1 |8 × 3
E (S)  E+ (S1 )2 E× (S)3  |S1 |11 |S|3 .
K 4 |S|2

and hence

E× (S)3  K 4 |S1 |3 |S|5  K 4 |S|8

as required. If one applies Lemma 6, combining with Lemma 7, then for a set S ⊆
S, |S |3  E× (S) the following holds

E× (S)17/3 |S |8
 4 2 E× (S)3  E+ (S )2 E× (S)3  |S|14 .
K |S|
4 2 K |S|

and it gives us

E× (S)  |S|48/17 K 12/17 .

Again if (21) takes place, then there is nothing to prove and otherwise Lemma 6 can
be applied because of the condition |S|27  K 6 p17 . This completes the proof. 

From the Parseval identity (6) one has 1S A(G)  |S|1/2 for any subset S of an
abelian group G. Theorem 9 gives us an interesting inverse inequality.
Corollary 10 Let S ⊆ Fp be a set, and E× (S)  |S|3 . If |S|5  1S 2A(Fp ) p3 , then

1S A(Fp )  |S|1/4 . (22)

In any case 1S A(Fp ) & |S|p−1/2 , provided that |S|  p/2.
It is interesting to notice that Corollary 10 does not hold in general fields, for
example, if S is a subfield of size p in Fp2 , then 1S A(Fp2 ) = 1 and E× (S) = |S|3 .
Typical sets with E× (S)  |S|3 are large subsets of multiplicative subgroups,
large subsets of the sets of the form {1, g, . . . , g n }, where g ∈ Fp is a primitive
root, see, e.g., [1] or, more generally, sets S with the small product set SS :=
{s1 s2 : s1 , s2 ∈ S}. Let us remark another consequence of Theorem 9 which
coincides with [1, Corollary 6] in the case of multiplicative subgroups (up to some
logarithms).
Corollary 11 Let H ⊆ Fp be a multiplicative subgroup, |H |  p2/3 or H =
{1, g, . . . , g n }, n  p2/3 . Then
A Remark on Sets with Small Wiener Norm 269

1H A(Fp )  |H |1/4 . (23)

In any case 1H A(Fp ) & |H |p−1/2 , provided that |H |  p/2.


Nevertheless, a simple application of Lemma 7, combining with [11, Theorem
3], namely, E+ (H ) |H |49/20 log1/5 |H |, |H |  p1/2 in the case when H is
a multiplicative subgroup, give us (another way to obtain the same is to use [11,
Corollary 7])
Corollary 12 Let H ⊆ Fp be a multiplicative subgroup, |H |  p1/2 . Then

1H A(Fp )  |H |1/4+1/40 . (24)

4 On the Quantity M+

Let S ⊆ G be a symmetric sets. Then Fourier coefficients of ;


1S are real and we put

;
− (S) = M− (S) = max(−1S (x)) ,
MG
x =0

and

;
+ (S) = M+ (S) = max 1S (x) .
MG
x =0

In a similar way we consider MG G


− (α), M+ (α) of a symmetric real function α.
As we noticed in the Introduction there is a connection of the problem of
estimating M+ with the Littlewood conjecture. Let us formulate a simple lemma
about some bounds for 1S A(G) in terms of M+ (S).
Lemma 13 Let α : G → R+ be a symmetric function. Then

|;
α (x)|  2(α1 + N M+ (α)) − N α(0) , (25)
x

and

|;
α (x)|  2N M− (α) + N α(0) . (26)
x

Proof Let us prove (25), another bound follows similarly. By formula (9), we have
  
;
α (x) = ;
α (0) + ;
α (x) + ;
α (x) = N α(0) . (27)
x x =0 : ;
α (x)>0 x =0 : ;
α (x)<0
270 I. D. Shkredov

Thus
   
|;
α (x)| = ;
α (0)+ ;
α (x)− ;
α (x) = 2;
α (0)+2 ;
α (x)−N α(0)
x x =0 : ;
α (x)>0 x =0 : ;
α (x)<0 x =0 : ;
α (x)>0

 2(α1 + N M+ (α)) − N α(0)

and we are done. 



Corollary 14 Let S ⊆ Fp be a symmetric set with E× (S)  |S|3 . Then
M+ (S), M− (S)  |S|1/4 .
Now we are ready to prove our second main result and we use the arguments
from [14]. It gives us an unconditional lower bound for M+ (S).
Theorem 15 Let S ⊆ G be a symmetric set, ζ ∈ (0, 1/2) be a real number, and
|S| = (1/2 − ζ )N. Then

M+ (S)  2−2 min{ζ 1/3 |S|1/3 , ζ |S|1/2 } . (28)

In particular, if |S|  N/4, then

M+ (S) & |S|1/3 . (29)

Proof In view of the Parseval identity, we get


 
|S|N = |S|2 + ;
12S (x) + |;
1S (x)|2 . (30)
x :;
1S (x)>0 x :;
1S (x)<0

From this formula it follows that M− (S) > 0 because otherwise by inequality (26)
of Lemma 13, we obtain

|S|N/2  |S|(N − |S|)  M+ (S)N

and hence M+ (S)  |S|/2. Further, by formula (10), we have

; 1 ; 2|S|;
1S (x) 1 
1S (x) = 1S (x − z);
1S (z) = + 1S (x − z);
; 1S (z) .
N z N N
z =0, z =x

Now take x such that ;


1S (x) = −M− (S). Then

1 
− 2ζ M− (S) = 1S (x − z);
; 1S (z) . (31)
N
z =0, z =x
A Remark on Sets with Small Wiener Norm 271

We see that the negativity of the left–hand side implies that one of the elements in
the product ;
1S (x − z);
1S (z) must be positive. Hence as in Lemma 13, we have

ζ M− (S)  M+ (S) (2M+ (S) + 1) . (32)

Because of |S| = (1/2 − ζ )N, we get in view of (30)



|S|N/2  |S|N(1/2 + ζ )  M2+ (S)N + |;
12S (x)| . (33)
x :;
1S (x)<0

If M2+ (S)  |S|/4, then we are done, otherwise (33) implies that

|S|/4  M− (S) · N −1 |;
1S (x)|  M2− (S) (34)
x :;
1S (x)<0

and, similarly,
⎛ ⎞

|S|/4  M− (S)N −1 ⎝ |;
1S (x)| + |A|⎠  2M− (S)M+ (S) . (35)
x :;
1S (x)>0

Here we have assumed that M+ (S)  1/2. Returning to (32) and using Lemma 13
as well as bound (35), we obtain

ζ |S|  8M2+ (S) (2M+ (S) + 1)  32M3+ (S)

as required. Again, we have assumed that M+ (S)  1/2. If not, then (34) and (32)
give us

M+ (S)  ζ |S|1/2 /4 .

This completes the proof. 



Remark 16 The condition |S| < N/2 of Theorem 15 is quite natural. If |S| =
N/2, then M+ (S) can be zero. Indeed, let G = Fn2 and A be an affine subspace
of codimension 1 such that 0 ∈ / S. Then ; S(0) = |S|, further, there is the only
Fourier coefficient equals (−|S|) and all other Fourier coefficients of S vanish. Thus,
M+ (S) = 0. Moreover, if we delete from S a random subset  of cardinality ζ N ,
then we get with high probability M+ (S ) & (ζ |S |)1/2 for this new set S = S \ .
272 I. D. Shkredov

References

1. V.C. Garcia, The finite Littlewood problem in Fp . 47(85), 1–14 (2018). https://doi.org/10.
1007/s11139-018-0038-3
2. B.J. Green, S.V. Konyagin, On the Littlewood problem modulo a prime. Canad. J. Math. 61(1),
141–164 (2009)
3. S.V. Konyagin, I.D. Shkredov, Quantitative version of Beurling–Helson theorem. Funct. Anal.
Its Appl. 49(2), 110–121 (2015)
4. S.V. Konyagin, I.D. Shkredov, On the Wiener norm of subsets of Z/pZ of medium size. J.
Math. Sci. 218(5), 599–608 (2016)
5. T. Sanders, The Littlewood–Gowers problem. J. Anal. Math. 101, 123–162 (2007)
6. T. Sanders, Boolean functions with small spectral norm, revisited. arXiv:1804.04050v1
[math.CA] 11 Apr 2018
7. T. Schoen, On the Littlewood conjecture in Z/pZ. MJCNT 7(3), 66–72 (2017)
8. S.V. Konyagin, On a problem of Littlewood. Izvestiya Russ. Acad. Sci. 45(2), 243–265 (1981)
9. O.C. McGehee, L. Pigno, B. Smith, Hardy’s inequality and the L1 norm of exponential sums.
Ann. Math. 113, 613–618 (1981)
10. T. Tao, V. Vu, Additive Combinatorics (Cambridge University Press, Cambridge, 2006)
11. B. Murphy, M. Rudnev, I.D. Shkredov, Y.N. Shteinikov, On the few products, many sums
problem. arXiv:1712.00410v1 [math.CO] 1 Dec 2017
12. M. Rudnev, I.D. Shkredov, S. Stevens, On the energy variant of the sum–product conjecture.
arXiv:1607.05053
13. A. Balog, T.D. Wooley, A low-energy decomposition theorem. Q. J. Math. 68(1), 207–226
(2017)
14. T. Sanders, Chowla’s cosine problem. Isr. J. Math. 179(1), 1–28 (2010)
15. I. Ruzsa, Negative values of cosine sums. Acta Arithmetica 111, 179–186 (2004)
16. W. Rudin, Fourier Analysis on Groups (Wiley, New York, 1962)
17. I.D. Shkredov, An application of the sum–product phenomenon to sets avoiding several linear
equations. Sb. Math. 209(4), 580–603 (2018)
18. T. Sanders, The coset and stability rings. arXiv:1810.10461v1 [math.CO] 24 Oct 2018
Order Estimates of Best Orthogonal
Trigonometric Approximations of Classes
of Infinitely Differentiable Functions

Tetiana A. Stepanyuk

Abstract In this paper we establish exact order estimates for the best uniform
orthogonal trigonometric approximations of the classes of 2π -periodic functions,
whose (ψ, β)–derivatives belong to unit balls of spaces Lp , 1 ≤ p < ∞, in the case,
when the sequence ψ(k) tends to zero faster, than any power function, but slower
than geometric progression. Similar estimates are also established in the Ls -metric,
1 < s ≤ ∞, for the classes of differentiable functions, which (ψ, β)–derivatives
belong to unit ball of space L1 .

Keywords Fourier series · Best orthogonal trigonometric approximation ·


Classes of infinitely differentiable functions · (ψ · β)-derivative

1 Introduction

Let C be a space of 2π –periodic continuous functions with the following norm:


f C := max |f (t); L∞ be the space of 2π –periodic functions f , which
t∈[0,2π )
are Lebesgue measurable and essentially bounded with the norm f ∞ :=
ess sup |f (t)| and Lp , 1 ≤ p < ∞, be the space of 2π –periodic functions f
t
 2π
 1
p
summable to the power p on [0, 2π ), with the norm f p := |f (t)|p dt .
0

T. A. Stepanyuk ()
Institute of Analysis and Number Theory, Graz University of Technology, Graz, Austria
Johann Radon Institute for Computational and Applied Mathematics (RICAM), Austrian
Academy of Sciences, Linz, Austria
Institute of Mathematics of NAS of Ukraine, Kyiv, Ukraine
e-mail: [email protected]

© Springer Nature Switzerland AG 2020 273


A. Raigorodskii, M. T. Rassias (eds.), Trigonometric Sums and Their Applications,
https://doi.org/10.1007/978-3-030-37904-9_13
274 T. A. Stepanyuk

Let f : R → R be the function from L1 , whose Fourier series is given by




fˆ(k)eikx ,
k=−∞


where fˆ(k) = 1
2π f (t)e−ikt dt are the Fourier coefficients of the function f ,
−π
ψ(k) is an arbitrary fixed sequence of real numbers and β is a fixed real number.
Then, if the series
 fˆ(k) i(kx+ βπ signk)
e 2
ψ(|k|)
k∈Z\{0}

is the Fourier series of some function ϕ from L1 , then this function is called the
ψ
(ψ, β)–derivative of the function f and is denoted by fβ . A set of functions f ,
ψ
whose (ψ, β)–derivatives exist, is denoted by Lβ (see [16]).
Let
* +
Bp0 := ϕ ∈ Lp : ||ϕ||p ≤ 1, ϕ ⊥ 1 , 1 ≤ p ≤ ∞.
ψ ψ
If f ∈ Lβ , and, at the same time fβ ∈ Bp0 , then we say that the function f belongs
ψ
to the class Lβ,p .
ψ ψ ψ ψ
Denote Cβ = C ∩ Lβ and Cβ,p = C ∩ Lβ,p .
By M we denote the set of all convex (downward) continuous functions ψ(t),
t ≥ 1, such that lim ψ(t) = 0. Assume that the sequence ψ(k), k ∈ N, specifying
t→∞
ψ
the class Lβ,p , 1 ≤ p ≤ ∞, is the restriction of the functions ψ(t) from M to the
set of natural numbers.
Following Stepanets (see, e.g., [16]), by using the characteristic μ(ψ; t) of
functions ψ from ∈ M of the form
t
μ(t) = μ(ψ; t) := , (1)
η(t) − t

where η(t) = η(ψ; t) := ψ −1 (ψ(t)/2), ψ −1 is the function inverse to ψ, we select


the following subsets of the set M:

M+
∞ = {ψ ∈ M : μ(ψ; t) ↑ ∞} .
* +
M∞ = ψ ∈ M+
∞ : ∃K > 0 η(ψ; t) − t ≥ K t ≥ 1 .

The functions ψr,α (t) = exp(−αt r ) are typical representatives of the set M+
∞.
ψ
Moreover, if r ∈ (0, 1], then ψr,α ∈ M∞ . The classes Lβ,p , generated by the
functions ψ = ψr,α are denoted by Lα,r
β,p .
Order Estimates of Best Orthogonal Trigonometric Approximations of Classes. . . 275

If ψ ∈ M+ ∞ then (see, e.g., [15, p. 97]) the function ψ(t) vanishes faster than
any power function, i.e.,

lim t r ψ(t) = 0 ∀r ∈ R.
t→∞

This implies that, under the condition ψ ∈ M+∞ , the Fourier series of any function
ψ
f from Cβ,p , β ∈ R, can be differentiated infinitely many times and, as a result,
we get uniformly convergent series. Hence, the classes Cβ,p with ψ ∈ M+
ψ
∞ consist
of infinitely differentiable functions. On the other hand, as shown in [17, p. 1692],
for any infinitely differentiable 2π –periodic function f , one can indicate a function
from the set M+
ψ
∞ such that f ∈ Cβ for any β ∈ R.
ψ
For functions f from classes Lβ,p we consider: Ls –norms of deviations of the
functions f from their partial Fourier sums of order n − 1, i.e., the quantities

ρn (f ; ·)s = f (·) − Sn−1 (f ; ·)s , 1 ≤ s ≤ ∞, (2)

where


n−1
Sn−1 (f ; x) = fˆ(k)eikx ;
k=−n+1

and the best orthogonal trigonometric approximations of the functions f in metric


of space Ls , i.e., the quantities of the form

em (f )s = inf f (·) − Sγm (f ; ·)s , 1 ≤ s ≤ ∞, (3)
γm

where γm , m ∈ N, is an arbitrary collection of m integer numbers, and



Sγm (f ; x) = fˆ(k)eikx .
k∈γm

We set
ψ
En (Lβ,p )s = sup ρn (f ; ·)s , 1 ≤ p, s ≤ ∞, (4)
ψ
f ∈Lβ,p

en⊥ (Lβ,p )s = sup en⊥ (f )s , 1 ≤ p, s ≤ ∞.


ψ
(5)
ψ
f ∈Lβ,p

It is clear, that if f ∈ C, then


⊥ ⊥
ρn (f ; x)∞ = ρn (f ; x)C , em (f )C = em (f )∞ .
276 T. A. Stepanyuk

That is why

ψ ψ ⊥ ⊥ ψ ψ
En (Cβ,p )C = En (Cβ,p )∞ , em (Cβ,p )C = em (Cβ,p )∞ .

The following inequalities follow from given above definitions (4) and (5)

⊥ ψ ⊥ ψ ψ
e2n (Lβ,p )s ≤ e2n−1 (Lβ,p )s ≤ En (Lβ,p )s , 1 ≤ p, s ≤ ∞. (6)

In the case when ψ(k) = k −r , r > 0, the classes Lβ,p , 1 ≤ p ≤ ∞, β ∈ R,


ψ

are well–known Weyl–Nagy classes Wβ,pr . For these classes, the order estimates of

quantities en⊥ (Lβ,p )s are known for 1 < p, s < ∞ (see [4, 5]), for 1 ≤ p < ∞,
ψ

s = ∞, r > p1 and also for p = 1, 1 < s < ∞, r > s1 , 1s + s1 = 1 (see [5, 6]).
In the case, when ψ(k) tends to zero not faster than some power function, order
estimates for quantities (5) were established in [1, 10, 12–14]. In the case, when
ψ(k) tends to zero not slower than geometric progression, exact order estimates for
en⊥ (Lβ,p )s were found in [11] for all 1 ≤ p, s ≤ ∞.
ψ

Our aim is to establish the exact-order estimates of en⊥ (Lβ,p )∞ , 1 ≤ p < ∞,


ψ

and en⊥ (Lβ,1 )s , 1 < s < ∞, in the case, when ψ decreases faster than any power
ψ

function, but slower than geometric progression (ψ ∈ M∞ ).

2 Best Orthogonal Trigonometric Approximations of the


ψ
Classes Lβ,p , 1 < p < ∞, in the Metric of Space L∞

We write an  bn to mean that there exist positive constants C1 and C2 independent


of n such that C1 an ≤ bn ≤ C2 an for all n.
ψ(t)
Theorem 1 Let 1 < p < ∞, ψ ∈ M∞ and the function |ψ (t)| ↑ ∞ as t → ∞.
Then, for all β ∈ R the following order estimates hold
1
⊥ ψ ⊥ ψ
e2n−1 (Lβ,p )∞  e2n (Lβ,p )∞  ψ(n)(η(n) − n) p . (7)

In Theorem 1 and further we will assume ψ (t) := ψ (t + 0).


Proof According to Theorem 1 from [8] under conditions ψ ∈ M+ ∞ , β ∈ R,
1 ≤ p < ∞, for n ∈ N, such that η(n) − n ≥ a > 2, μ(n) ≥ b > 2 the following
estimate is true

ψ 1− p1 1
En (Lβ,p )∞ ≤ Ka,b (2p) ψ(n)(η(n) − n) p , (8)

where
Order Estimates of Best Orthogonal Trigonometric Approximations of Classes. . . 277

 
1 2b 1
Ka,b = max + , 2π .
π b−2 a

It should be noticed, that condition |ψψ(t)


(t)| → ∞ as t → ∞ implies that always
exists n0 ∈ N, such that for all n > n0 , n ∈ N the following inequalities take place:
η(ψ, n) − n ≥ a > 2 and μ(ψ, n) ≥ b > 2. This fact follows from Remark 3.13.1
[16], which says, that for every ψ ∈ M+ ∞

ψ(t)
K1 (η(t) − t) ≤ ≤ K2 (η(t) − t), K1 , K2 > 0, (9)
|ψ (t)|
and from the definition of quantity μ(ψ, t).
Using inequalities (6) and (8), we obtain

⊥ ψ ⊥ ψ 1− p1 1
e2n (Lβ,p )∞ ≤ e2n−1 (Lβ,p )∞ ≤ Ka,b (2p) ψ(n)(η(n) − n) p . (10)

⊥ (L
Let us find the lower estimate for the quantity e2n
ψ
β,p )∞ . With this purpose we
construct the function

∗ ∗ λp 1
fp,n (t) = fp,n (ψ; t) := 1
ψ(1)ψ(2n)+
2
ψ(n)(η(n) − n) p


n−1 
2n 
1 1
+ ψ(k)ψ(2n − k) cos kt + 2
ψ (k) cos kt , + = 1. (11)
p p
k=1 k=n

∗ ∈L
Let us show that fp,n
ψ
β,p . The definition of (ψ, β)–derivative yields

  βπ 
n−1
∗ ψ λp
(fp,n (t))β = 1
ψ(2n − k) cos kt +
2
ψ(n)(η(n) − n) p k=1

 
 βπ 
2n
+ ψ(k) cos kt + . (12)
2
k=n

Obviously
  
 ∗  λp
n−1 2n
(f (t))ψ  ≤ ψ(2n − k) + ψ(k) <
p,n β 1
ψ(n)(η(n) − n) p k=1 k=n


2n  ∞ 
2λp 2λp
< 1
ψ(k) ≤ 1
ψ(n) + ψ(u)du .
ψ(n)(η(n) − n) p k=n ψ(n)(η(n) − n) p n
(13)
278 T. A. Stepanyuk

To estimate the integral from the right part of formula (13), we use the following
statement [7, p. 500].
Proposition 1 If ψ ∈ M+ ∞ , then for arbitrary m ∈ N, such that μ(ψ, m) > 2 the
following condition holds

∞
2
ψ(u)du ≤ ψ(m)(η(m) − m). (14)
1− 2
μ(m)
m

Formulas (13) and (14) imply that


 
 ∗  2λp 2b
(f (t))ψ  ≤ ψ(n) + ψ(n)(η(n) − n) <
p,n β 1
b−2
ψ(n)(η(n) − n) p

5λp b 1
< (η(n) − n) p . (15)
b−2

We denote

1 βπ   βπ 
k
Dk,β (t) := cos + cos j t + . (16)
2 2 2
j =1

Applying Abel transform, we have


n−1  βπ  
n−2
ψ(2n − k) cos kt + = (ψ(2n − k) − ψ(2n − k + 1))Dk,β (t)
2
k=1 k=1
1 βπ
+ ψ(n + 1)Dn−1,β (t) − ψ(2n − 1) cos (17)
2 2
and


2n  βπ  
2n−1
ψ(k) cos kt + = (ψ(k) − ψ(k + 1))Dk,β (t)
2
k=n k=n

+ ψ(2n)D2n,β (t) − ψ(n)Dn−1,β (t). (18)

Since


N −1  N −1  Nt 1
sin(γ + kt) = sin γ + t sin (19)
2 2 sin 2t
k=0
Order Estimates of Best Orthogonal Trigonometric Approximations of Classes. . . 279

(see, e.g., [2, p.43]), for N = k + 1, γ = (β − 1) π2 , the following inequality holds


 
 cos  kt + βπ  sin k+1 t βπ 
 1
|Dk,β (t)| =  2 2 2
− cos 
 sin 2t 2 2 
  
 sin (k + 1 )t + βπ  − cos t sin βπ  π
 2 
= 2 2 2
 ≤ , 0 < |t| ≤ π. (20)
 2 sin 2t  t

According to (12), (17), (18) and (20), we obtain


 n−2
 ∗ 
(f (t))ψ  ≤ λp π 
p,n |ψ(2n−k)−ψ(2n−k −1)|+ψ(n+1)
β 1
|t|
ψ(n)(η(n)−n) p k=1


2n−1 
+ ψ(2n − 1) + |ψ(k) − ψ(k + 1)| + ψ(2n) + ψ(n)
k=n
λp 2π 4π λp 1
= (ψ(n + 1) + ψ(n)) ≤ . (21)
1
p |t| 1
p |t|
ψ(n)(η(n) − n) (η(n) − n)

So, (15) and (21) imply


= ∗ =
=(f (t))ψ =
p,n β p


5b  
1

dt
1
p
≤ λp max , 4π (η(n)−n)dt +
|t|p
p
b−2 (η(n)−n) p
1 1
|t|≤ η(n)−n η(n)−n ≤|t|≤π


5b  1  p1
5b 1
≤ 2λp max , 4π 1 + = 2λp max , 4π (p ) p .
b−2 p−1 b−2

Hence, for

1
λp = 1

2(p ) max b−2
p 5b
, 4π

∗ ∈L
the embedding fp,n
ψ
β,p is true.
Let us consider the quantity

 π 
 

I1 := inf  (fp,n (t) − Sγ2n (fp,n ; t))V2n (t)dt ,
∗ ∗
(22)
γ2n
−π

where V2n are de la Vallée-Poisson kernels of the form


280 T. A. Stepanyuk

1 
m  
2m−1
k 
Vm (t) := + cos kt + 2 1− cos kt, m ∈ N. (23)
2 2m
k=1 k=m+1

Proposition A1.1 from [3] implies

I1 ≤ inf fp∗ (t) − Sγ2n (fp,n


∗ ⊥
; t)∞ V2n 1 = e2n ∗
(fp,n )∞ V2n 1 . (24)
γ2n

Since (see, e.g., [18, p.247])

Vm 1 ≤ 3π, m ∈ N, (25)

from (24) and (25) we can write down the estimate

⊥ ∗ 1
e2n (fp,n )∞ ≥ I1 . (26)

Notice, that
∗ ∗
fp,n (t) − Sγ2n (fp,n ; t)
   
λp
= 1
ψ(|k|)ψ(2n − |k|)eikt + ψ 2 (|k|)eikt ,
2ψ(n)(η(n) − n) p |k|≤n−1, n≤|k|≤2n,
k ∈γ
/ 2n k ∈γ
/ 2n

(27)

where ψ(0) := ψ(1)


Whereas
π 
0, k + m = 0,
eikt eimt dt = k, m ∈ Z, (28)
2π, k + m = 0,
−π

and taking into account (23), we obtain


∗ ∗
(fp,n (t) − Sγ2n (fp,n ; t))V2n (t)dt (29)
−π

π  
λp
= 1
ψ(k)ψ(2n − k)eikt + ψ(|k|)ψ(2n−|k|)eikt
4ψ(n)(η(n)−n) p 0≤k≤n−1, −n+1≤k≤−1,
−π k ∈γ
/ 2n k ∈γ
/ 2n
Order Estimates of Best Orthogonal Trigonometric Approximations of Classes. . . 281

  
+ 2
ψ (k)e ikt
+ 2
ψ (|k|)e ikt
×
n≤k≤2n, −2n≤k≤−n,
k ∈γ
/ 2n k ∈γ
/ 2n

     |k|  ikt 
× eikt + eikt + 2 1− e dt (30)
4n
0≤k≤2n −2n≤k≤−1 2n+1≤|k|≤4n−1
   
λp π
= 1
ψ(|k|)ψ(2n − |k|) + ψ 2 (|k|) . (31)
2ψ(n)(η(n) − n) p |k|≤n−1, n≤|k|≤2n,
k ∈γ
/ 2n k ∈γ
/ 2n

The function φn (t) := ψ(t)ψ(2n − t) decreases for t ∈ [1, n]. Indeed


 ψ(t) ψ(2n − t) 
φn (t) = |ψ (t)||ψ (2n − t)|
− ≤ 0,
|ψ (t)| |ψ (2n − t)|

because |ψψ(t)
(t)| ↑ ∞ for large n.
Thus, the monotonicity of function φn (t) and (31) imply
  
π λp
I1 = 1
ψ 2 (n) + 2
ψ (|k|)
2ψ(n)(η(n) − n) p n+1≤|k|≤2n


2n 
η(n)
π λp π λp
> 1
ψ (k) ≥
2
1
ψ 2 (t)dt
2ψ(n)(η(n) − n) p k=n 2ψ(n)(η(n) − n) p n
π λp π λp 1
> 1
ψ 2 (η(n))(η(n) − n) = ψ(n)(η(n) − n) p . (32)
p 8
2ψ(n)(η(n) − n)

By considering (26) and (32) we can write

⊥ ψ ⊥ ∗ 1 λp 1
e2n (Lβ,p )∞ ≥ e2n (fp,n )∞ ≥ I1 ≥ ψ(n)(η(n) − n) p . (33)
3π 24
Theorem 1 is proved.
In fact in the proof of Theorem 1 we obtained estimates with constants in explicit
form.
Proposition 2 Let ψ ∈ M+
∞ , β ∈ R, 1 < p < ∞,
1
p + 1
p = 1, and the function
ψ(t)
increases monotonically. Then for n ∈ N, such that μ(ψ, n) ≥ b > 2 and
|ψ (t)|
η(ψ, n) − n ≥ a > 2, the following estimates hold
1 1
⊥ ⊥ ψ ψ
Kb,p ψ(n)(η(n) − n) p ≤ e2n (Lβ,p )∞ ≤ e2n−1 (Lβ,p )∞ ≤ Ka,b,p ψ(n)(η(n) − n) p ,
(34)
282 T. A. Stepanyuk

where
 
1 2b 1 1
Ka,b,p = max + , 2π (2p) p . (35)
π b−2 a

1
Kb,p =
1
. (36)
48 max 5b
b−2 , 4π (p ) p

3 Best Orthogonal Trigonometric Approximations of the


ψ
Classes Lβ,1 in the Metric of Space L∞

Theorem 2 Let ψ ∈ M+
∞ . Then for all β ∈ R order estimates are true

⊥ ψ ⊥ ψ
e2n−1 (Lβ,1 )∞  e2n (Lβ,1 )∞  ψ(n)(η(n) − n). (37)



Proof According to formula (48) from [18] under conditions ψ ∈ M, ψ(k) <
k=1
∞, β ∈ R, for all n ∈ N the following estimate holds

ψ 1
En (Lβ,1 )∞ ≤ ψ(k). (38)
π
k=n

Using Proposition 1, we have


⊥ ψ ⊥ ψ ψ 1
e2n (Lβ,1 )∞ ≤ e2n−1 (Lβ,1 )∞ ≤ En (Lβ,1 )∞ ≤ ψ(k)
π
k=n
 ∞   
1 ψ(n) b
≤ ψ(n) + ψ(u)du ≤ 1+ (η(n) − n) . (39)
π π b−2
n

⊥ (L
Let us find the lower estimate for the quantity e2n
ψ
β,1 )∞ .
We consider the quantity

 π 
 

I2 := inf  (f2n (t) − Sγ2n (f2n ; t))V2n (t)dt ,
∗ ∗
(40)
γ2n
−π

where Vm are de la Vallée-Poisson kernels of the form (23), and


Order Estimates of Best Orthogonal Trigonometric Approximations of Classes. . . 283

1 1   
m 2m
fm∗ (t) = fm∗ (ψ; t) := ψ(1)+ kψ(k) cos kt+ (2m+1−k)ψ(k) cos kt .
5π m 2
k=1 k=m+1
(41)
In [18, p. 263–265] it was shown that (fm∗ )β 1 ≤ 1, i.e., fm∗ belongs to the class
ψ

ψ
Lβ,1 for all m ∈ N.
Using Proposition A1.1 from [3] and inequality (25), we have
∗ ∗ ⊥ ∗
I2 ≤ inf f2n (t) − Sγ2n (f2n ; t)∞ V2n 1 ≤ 3π e2n (f2n )∞ . (42)
γ2n

Assuming again ψ(0) := ψ(1), from (23) and (41), we derive

 π    
1 
I2 = inf  |k|ψ(|k|)eikt + (4n + 1 − |k|)ψ(|k|)eikt ×
20π n γ2n |k|≤2n, 2n+1≤|k|≤4n,
−π k ∈γ
/ 2n k ∈γ
/ 2n

   
 |k|  ikt  
× e ikt
+2 1− e dt 
4n
|k|≤2n 2n+1≤|k|≤4n−1
  
1  |k| 
= inf |k|ψ(|k|) + 1− (4n + 1 − |k|)ψ(|k|)
10n γ2n |k|≤2n, 2n+1≤|k|≤4n,
4n
k ∈γ
/ 2n k ∈γ
/ 2n

  
2n 
1 1
> inf |k|ψ(|k|) = nψ(n) + 2 kψ(k)
10n γ2n |k|≤2n, 10n
k=n+1
k ∈γ
/ 2n


η(n)
1
2n
1 1
> ψ(k) > ψ(t)dt > ψ(n)(η(n) − n), (43)
10 10 20
k=n n

where we have used, that function tψ(t) decreases monotonically from some
number t0 . Indeed,
 ψ(t) 
(tψ(t)) = |ψ (t)|t − 1 ,
|ψ (t)|

and relations (1) and (9) yield

ψ(t) η(t) − t 1

 = → 0, as t → ∞ for ψ ∈ M+
∞.
|ψ (t)| t μ(t)
284 T. A. Stepanyuk

Formulas (42) and (43) imply, that for n > t0

⊥ ψ ⊥ ∗ 1 1
e2n (Lβ,1 )∞ ≥ e2n (f2n )∞ ≥ I2 > ψ(n)(η(n) − n).
3π 60π
Theorem 2 is proved.
Corollary 1 Let r ∈ (0, 1), α > 0, 1 ≤ p < ∞ and β ∈ R. Then for all n ∈ N the
following estimates are true
1−r
en⊥ (Lα,r
β,p )∞  exp(−αn )n
r p . (44)

4 Best Orthogonal Trigonometric Approximations of the


ψ
Classes Lβ,1 in the Metric of Spaces Ls , 1 < s < ∞

ψ(t)
Theorem 3 Let 1 < s < ∞, ψ ∈ M∞ and function |ψ (t)| ↑ ∞ as t → ∞. Then
for all β ∈ R order estimates hold

⊥ ψ ⊥ ψ 1 1 1
e2n−1 (Lβ,1 )s  e2n (Lβ,1 )s  ψ(n)(η(n) − n) s , + = 1. (45)
s s

Proof According to Theorem 2 from [8] under conditions ψ ∈ M+ ∞ , β ∈ R,


1 < s ≤ ∞ for n ∈ N, such that η(n) − n ≥ a > 2, μ(n) ≥ b > 2 the following
estimate holds
1 1
En (Lβ,1 )s ≤ Ka,b (2s ) s ψ(n)(η(n) − n) s .
ψ
(46)

Since, |ψψ(t)
(t)| ↑ ∞, then as it was noticed in the proof of Theorem 1, exists
number n0 , such that for all n > n0 inequalities η(n) − n ≥ a > 2, μ(n) ≥ b > 2
hold.
Using inequalities (6) and (46), we get

⊥ ⊥
 1 1
(Lβ,1 )s ≤ Ka,b,s 2s s ψ(n)(η(n) − n) s .
ψ ψ
e2n (Lβ,1 )s ≤ e2n−1 (47)

⊥ (L
Let us find the lower estimate of the quantity e2n
ψ
β,1 )s .
We consider the quantity

 π 
 
I3 := inf  (f2n
∗∗ ∗∗
(t) − Sγ2n (f2n ; t))fs∗ ,n (t)dt , (48)
γ2n
−π
Order Estimates of Best Orthogonal Trigonometric Approximations of Classes. . . 285

where
1
fm∗∗ (t) = Vm (t),

and fs∗ ,n is defined by formula (11).


On the basis of Proposition A1.1 from [3] we derive
∗∗ ∗∗
I3 ≤ inf f2n (t) − Sγ2n (f2n ; t)s fs∗ s ≤ e2n
⊥ ∗∗
(f2n )s . (49)
γ2n

On other hand, using formulas (28), we write

 π  
λs    |k|  ikt 
I3 = 1
inf  eikt + 2 1− e ×
12π ψ(n)(η(n) − n) s γ2n
|k|≤2n, 2n+1≤|k|≤4n−1,
4n
−π k ∈γ
/ 2n k ∈γ
/ 2n
    

× ψ(|k|)ψ(2n − |k|)e ikt
+ 2
ψ (|k|)e ikt
dt 
|k|≤n−1 n≤|k|≤2n

λs    
= 1
inf ψ(|k|)ψ(2n − |k|) + ψ 2 (|k|)
6ψ(n)(η(n) − n) s γ2n
|k|≤n−1, n≤|k|≤2n,
k ∈γ
/ 2n k ∈γ
/ 2n

λs  
2n  λ 
2n
= 1
2
ψ (n)+2 ψ 2 (k) > 1
ψ 2 (k)
6ψ(n)(η(n)−n) s k=n+1 6π ψ(n)(η(n)−n) s k=n


η(n)
λs λs 1
> 1
ψ 2 (t)dt > ψ(n)(η(n) − n) s . (50)
6ψ(n)(η(n) − n) s 24
n

Hence, formulas (49) and (50) imply

⊥ ⊥ λs 1
(fs∗∗
ψ
e2n (Lβ,1 )s ≥ e2n )s ≥ I3 ≥ ψ(n)(η(n) − n) s . (51)
24
Theorem 3 is proved.
In fact in the proof of Theorem 3 we obtained estimates with constants in explicit
form.
Proposition 3 Let ψ ∈ M+ ψ(t)
∞ , β ∈ R, 1 ≤ p < ∞ and function |ψ (t)| increases
monotonically. Then for all n ∈ N, such that μ(ψ, n) ≥ b > 2 and η(ψ, n) − n ≥
a > 2, the following estimates are true
286 T. A. Stepanyuk

1
⊥ ψ ⊥ ψ
Kb,s ψ(n)(η(n) − n) s ≤ e2n (Lβ,1 )s ≤ e2n−1 (Lβ,1 )s
1
≤ Ka,b,s ψ(n)(η(n) − n) s ,

where Ka,b,s and Kb,s are defined by formulas (35) and (36) respectively.
Corollary 2 Let r ∈ (0, 1), α > 0, 1 < s < ∞ and β ∈ R. Then for all n ∈ N the
following estimates are true

1−r 1 1
en⊥ (Lα,r
β,1 )s  exp(−αn )n
r s , + = 1. (52)
s s
Note, that functions
(1) e−αt t γ , α > 0, r ∈ (0, 1], γ ≤ 0;
r

(2) e−αt lnγ (t + K), α > 0, r ∈ (0, 1], γ ≤ 0, K > e − 1,


r

etc., can be regarded as examples of functions ψ, which satisfy the conditions of


Theorems 1 and 3.
Remark 1 It should be noticed, that from Theorem 1–3 it follows that the orders of
quantities en⊥ (Lβ,p )s for 1 ≤ p < ∞, s = ∞ and p = 1, 1 < s < ∞, coincide
ψ

ψ
with orders of the best approximations En (Lβ,p )s (see [9]).

Acknowledgements The author is supported by the Austrian Science Fund FWF projects F5503
and F5506-N26 (part of the Special Research Program (SFB) “Quasi-Monte Carlo Methods:
Theory and Applications”) and partially is supported by grant of NAS of Ukraine for groups of
young scientists (project No16-10/2018).

References

1. A.S. Fedorenko, On the best m-term trigonometric and orthogonal trigonometric approxima-
ψ
tions of functions from the classes Lβ,p . Ukr. Math. J. 51(12), 1945–1949 (1999)
2. I.S. Gradshtein, I.M. Ryzhik, Tables of Integrals, Sums, Series, and Products [in Russian]
(Fizmatgiz, Moscow, 1963)
3. N.P. Korneichuk, Exact Constants in Approximation Theory, vol. 38 (Cambridge University
Press, Cambridge, New York, 1990)
4. A.S. Romanyuk, Approximation of classes of periodic functions of many variables. Mat.
Zametki 71(1), 109–121 (2002)
5. A.S. Romanyuk, Best trigonometric approximations of the classes of periodic functions of
many variables in a uniform metric. Mat. Zametki 81(2), 247–261 (2007)
6. A.S. Romanyuk, Approximate Characteristics of Classes of Periodic Functions of Many
Variables [in Russian] (Institute of Mathematics, Ukrainian National Academy of Sciences,
Kyiv, 2012)
7. A.S. Serdyuk, Approximation by interpolation trigonometric polynomials on classes of
periodic analytic functions. Ukr. Mat. Zh. 64(5), 698–712 (2012); English translation: Ukr.
Math. J. 64(5), 797–815 (2012)
Order Estimates of Best Orthogonal Trigonometric Approximations of Classes. . . 287

8. A.S. Serdyuk, T.A. Stepaniuk, Order estimates for the best approximation and approximation
by Fourier sums of classes of infinitely differentiable functions. Zb. Pr. Inst. Mat. NAN Ukr.
10(1), 255–282 (2013). [in Ukrainian]
9. A.S. Serdyuk, T.A. Stepanyuk, Estimates for the best approximations of the classes of innately
differentiable functions in uniform and integral metrics. Ukr. Mat. Zh. 66(9), 1244–1256 (2014)
10. A.S. Serdyuk, T.A. Stepaniuk, Order estimates for the best orthogonal trigonometric approxi-
mations of the classes of convolutions of periodic functions of low smoothness. Ukr. Math. J.
67(7), 1–24 (2015)
11. A.S. Serdyuk, T.A. Stepaniuk, Estimates of the best m-term trigonometric approximations of
classes of analytic functions. Dopov. Nats. Akad. Nauk Ukr. Mat. Pryr. Tekh. Nauky No. 2
32–37 (2015). [in Ukrainian]
12. A.S. Serdyuk, T.A. Stepanyuk, Estimates for approximations by Fourier sums, best approxima-
tions and best orthogonal trigonometric approximations of the classes of (ψ, β)–differentiable
functions. Bull. Soc. Sci. Lettres Lodz. Ser. Rech. Deform. 66(2), 35–43 (2016)
13. V.V. Shkapa, Estimates of the best M-term and orthogonal trigonometric approximations of
ψ
functions from the classes Lβ,p in a uniform metric. Differential Equations and Related
Problems [in Ukrainian]. vol. 11, issue 2. (Institute of Mathematics, Ukrainian National
Academy of Sciences, Kyiv, 2014), pp. 305–317
ψ
14. V.V. Shkapa, Best orthogonal trigonometric approximations of functions from the classes Lβ,1 ,
Approximation Theory of Functions and Related Problems [in Ukrainian], vol. 11, issue 3
(Institute of Mathematics, Ukrainian National Academy of Sciences, Kyiv, 2014), pp. 315–329
15. A.I. Stepanets, Classification and Approximation of Periodic Functions [in Russian] (Naukova
Dumka, Kiev, 1987)
16. A.I. Stepanets, Methods of Approximation Theory (VSP: Leiden/Boston, 2005)
17. A.I. Stepanets, A.S. Serdyuk, A.L. Shidlich, Classification of infinitely differentiable periodic
functions. Ukr. Mat. Zh. 60(12), 1686–1708 (2008)
18. T.A. Stepaniuk, Estimates of the best approximations and approximations of Fourier sums of
classes of convolutions of periodic functions of not high smoothness in integral metrics, [in
Ukrainian]. Zb. Pr. Inst. Mat. NAN Ukr. 11(3), 241–269 (2014)
Equivalent Conditions of a Reverse
Hilbert-Type Integral Inequality with the
Kernel of Hyperbolic Cotangent Function
Related to the Riemann Zeta Function

Bicheng Yang

Abstract By the use of techniques of real analysis and weight functions, we


study some equivalent conditions of a reverse Hilbert-type integral inequality
with the non-homogeneous kernel of hyperbolic cotangent function, related to
the Riemann zeta function. Some equivalent conditions of a reverse Hilbert-type
integral inequality with the homogeneous kernel are deduced. We also consider
some particular cases.

Keywords Reverse Hilbert-type integral inequality · Weight function ·


Equivalent form · Homogeneous kernel

2000 Mathematics Subject Classification 26D15

1 Introduction
∞ ∞
If 0 < 0 f 2 (x)dx < ∞ and 0 < 0 g 2 (y)dy < ∞, then we have the following
Hilbert integral inequality (cf. [1]):

 ∞ ∞  ∞  ∞ 1
f (x)g(y) 2 2
2
dxdy < π f (x)dx g (y)dy , (1)
0 0 x+y 0 0

where the constant factor π is the best possible.


In 1925, Hardy [2] gave an extension of (1) as follows:
∞
For p > 1, p1 + q1 = 1, f (x), g(y) ≥ 0, 0 < 0 f p (x)dx < ∞ and 0 <
∞ q
0 g (y)dy < ∞, the following Hardy-Hilbert inequality holds:

B. Yang ()
Department of Mathematics, Guangdong University of Education, Guangdong,
Guangzhou, P. R. China
e-mail: [email protected]; [email protected]

© Springer Nature Switzerland AG 2020 289


A. Raigorodskii, M. T. Rassias (eds.), Trigonometric Sums and Their Applications,
https://doi.org/10.1007/978-3-030-37904-9_14
290 B. Yang

 ∞ ∞  ∞  1  ∞ 1
f (x)g(y) π p
p
q
q
dxdy < f (x)dx g (y)dy ,
0 0 x+y sin(π/p) 0 0
(2)
π
where, the constant factor sin(π/p) is the best possible.
Inequalities (1) and (2) are important in analysis and its applications (cf. [3, 4]).
In 1934, Hardy et al. gave an extension of (2) as follows: If p > 1, p1 + q1 =
1, k1 (x, y) is a non-negative homogeneous function of degree −1,
 ∞ −1
kp = k1 (u, 1)u p du ∈ R+ = (0, ∞),
0

then we have the following Hardy-Hilbert-type integral inequality with the best
possible constant kp :

 ∞ ∞  ∞  1  ∞ 1
p q
k1 (x, y)f (x)g(y)dxdy < kp f p (x)dx g q (y)dy ;
0 0 0 0
(3)
for 0 < p < 1, p1 + q1 = 1, the reverse of (2) follows (cf. [3], Theorem 319, Theorem
336). Also a Hilbert-type integral inequality with the non-homogeneous
∞ kernel is
proved as follows: If p > 1, p1 + q1 = 1, h(u) > 0, φ(σ ) = 0 h(u)uσ −1 du ∈ R+ ,
then
 ∞ ∞
h(xy)f (x)g(y)dxdy
0 0
 ∞  1  ∞ 1
1 p−2 p
p
q
q
< φ( ) x f (x)dx g (y)dy , (4)
p 0 0

where, the constant factor φ( p1 ) is the best possible (cf. [3], Theorem 350).
In 1998, by introducing an independent parameter λ > 0, Yang gave an extension
1
of (1) with the kernel (x+y) λ (cf. [5, 6] ). In 2004, by introducing another pair
conjugate exponents (r, s), Yang [7] gave an extension of (2) as follows: If λ >
∞ λ
0, p, r > 1, p1 + q1 = 1r + 1s = 1, f (x), g(y) ≥ 0, 0 < 0 x p(1− r )−1 f p (x)dx < ∞
∞ λ
and 0 < 0 y q(1− s )−1 g q (y)dy < ∞, then
 ∞ ∞f (x)g(y)
dxdy
0 0 xλ + yλ
 ∞  1  ∞ 1
π λ p λ q
< x p(1− r )−1 f p (x)dx y q(1− s )−1 g q (y)dy , (5)
λ sin(π/r) 0 0

π
where, the constant factor λ sin(π/r) is the best possible. In 2005, [8] also gave an
extension of (2) as follows:
Equivalent Conditions of a Reverse Hilbert-Type Integral Inequality with the. . . 291

 ∞ ∞
f (x)g(y)
dxdy
0 0 (x + y)λ
 ∞  1  ∞ 1
λ λ λ p λ q
< B( , ) x p(1− r )−1 f p (x)dx y q(1− s )−1 g q (y)dy , (6)
r s 0 0

where, the constant factor B( λr , λs )(λ > 0) is the best possible. Krnić et al. [9–16]
provided some extensions and particular cases of (2), (3) and (4) with parameters.
In 2009, Yang gave an extension of (3), (5) and (6) as follows (cf. [17, 18]): If
λ1 + λ2 = λ ∈ R = (−∞, ∞), kλ (x, y) is a non-negative homogeneous function
of degree −λ, satisfying

kλ (ux, uy) = u−λ kλ (x, y)(u, x, y > 0),


∞
k(λ1 ) = 0 kλ (u, 1)uλ1 −1 du ∈ R+ = (0, ∞), then for p > 1, p1 + 1
q = 1, we
have
 ∞ ∞
kλ (x, y)f (x)g(y)dxdy
0 0
 ∞  1  ∞ 1
p q
p(1−λ1 )−1 p q(1−λ2 )−1 q
< k(λ1 ) x f (x)dx y g (y)dy , (7)
0 0

where, the constant factor k(λ1 ) is the best possible; for 0 < p < 1, p1 + q1 = 1,
the reverse of (7) follows. Also an extension of (4) was given as follows: For p >
1, p1 + q1 = 1, we have
 ∞ ∞
h(xy)f (x)g(y)dxdy
0 0
 ∞  1  ∞ 1
p q
< φ(σ ) x p(1−σ )−1 f p (x)dx y q(1−σ )−1 g q (y)dy , (8)
0 0

where, the constant factor φ(σ ) is the best possible; for 0 < p < 1, p1 + q1 = 1, the
reverse of (8) follows (cf. [19]).
Some equivalent inequalities of (7) and (8) are considered by [18]. In 2013,
Yang [19] also studied the equivalency of (7) and (8). In 2017, Hong [20] studied a
equivalent condition between (7) and the related parameters.
In this chapter, by the use of the way of real analysis and the weight functions,
we consider some equivalent conditions of a reverse of (8) in the kernel h(xy) =
coth(xy) − 1 for 0 < p < 1, related to the Riemann zeta function. Some equivalent
conditions of the reverse of (7) in the kernel k0 (x, y) = cot h(x/y) − 1 are deduced.
We also consider some particular cases.
292 B. Yang

2 An Example and Two Lemmas

eu +e−u
Example 1 Setting h(u) = coth(u)−1 = 2
e2u −1
(u > 0), where, coth(u) = eu −e−u
is the hyperbolic cotangent function, then we find coth(xy)−1 = 2
e2xy −1
, cot h( x
y )−
1= 2
e2x/y −1
and for σ > 1,
 ∞
k(σ ) = (cot h(u) − 1)uσ −1 du
0
 ∞  ∞
2uσ −1 2uσ −1 e−2u
= du = du
0 e2u − 1
0 1 − e−2u
 ∞ ∞
 ∞  ∞
=2 uσ −1 e−2(k+1)u du = 2 uσ −1 e−2ku du.
0 k=0 k=1 0

Setting v = 2ku in the above integral, we have


 ∞ ∞
 1
k(σ ) = 2 1−σ
v σ −1 e−v dv = 21−σ Γ (σ )ζ (σ ) ∈ R+ , (9)
0 kσ
k=1
∞
where, Γ (σ ) := 0 v σ −1 e−v dv (σ > 0) is the gamma function, and ζ (σ ) :=
∞ 1
k=1 k σ (σ > 1) is the Riemann zeta function.

Setting δ0 = σ −1 σ −1 σ +1
2 > 0, σ ± δ0 ≥ σ − 2 = 2 > 1, we still have k(σ ± δ0 ) <
∞,
In the following, we make appointment that 0 < p < 1, p1 + q1 = 1, σ > 1, σ1 ∈
R.
For n ∈ N = {1, 2, . . . }, we define the following two expressions:
 ∞  1 1

1
σ + pn −1 σ1 − qn −1
I1 := (coth(xy) − 1)x dx y dy, (10)
1 0

 1  ∞ 1

σ − pn −1 σ + 1 −1
I2 := (coth(xy) − 1)x dx y 1 qn dy. (11)
0 1

Setting u = xy in (10) and (11), we have

 9  σ + 1 −1 :
∞ y u pn 1 σ − 1 −1
I1 = (coth(u) − 1) du y 1 qn dy
1 0 y y
 ∞  y 
1 σ + 1 −1
= y (σ1 −σ )− n −1 (coth(u) − 1)u pn du dy, (12)
1 0
Equivalent Conditions of a Reverse Hilbert-Type Integral Inequality with the. . . 293

 9  σ − 1 −1 :
1 ∞ u pn 1 σ + 1 −1
I2 = (coth(u) − 1) du y 1 qn dy
0 y y y
 1  ∞ 
1
(σ1 −σ )+ n1 −1 σ − pn −1
= y (cot h(u) − 1)u du dy. (13)
0 y

Lemma 1 If there exists a constant M > 0, such that for any non-negative
measurable functions f (x) and g(y) in (0, ∞), the following inequality
 ∞ ∞
I : = (coth(xy) − 1)f (x)g(y)dxdy
0 0
 ∞  1  ∞ 1
p q
≥M x p(1−σ )−1 p
f (x)dx y q(1−σ1 )−1 q
g (y)dy (14)
0 0

holds true, then we have σ1 = σ and M ≤ k(σ ).


Proof If σ1 > σ, then for n > 1
δ0 p (n ∈ N, 0 < p < 1), we set the following two
functions:

0, 0 < x < 1
fn (x) : = 1
σ − pn −1 ,
x ,x ≥ 1
 1
σ1 + qn −1
gn (y) : = y ,0 < y ≤ 1 .
0, y > 1

We find
 ∞  1  ∞ 1
p q
p q
J2 : = x p(1−σ )−1 fn (x)dx y q(1−σ1 )−1 gn (y)dy
0 0
 ∞  1  1  q1
1 p 1
= x − n −1 dx y n −1 dy = n.
1 0

By (13), we have
 1  ∞ 1
(σ1 −σ )+ n1 −1 σ − pn −1
I2 ≤ y dy (coth(u) − 1)u du
0 0
 1
1 1
σ − pn −1
= (cot h(u) − 1)u du
σ1 − σ + 0
1
n
 ∞ 
1
σ − pn −1
+ (cot h(u) − 1)u du
1
294 B. Yang

 1
1
≤ (cot h(u) − 1)u(σ −δ0 )−1 du
σ1 − σ 0
 ∞ 
σ −1
+ (cot h(u) − 1)u du
1
1
≤ (k(σ − δ0 ) + k(σ )) ,
σ1 − σ

and then by (14), it follows that

1
(k(σ − δ0 ) + k(σ ))
σ1 − σ
 ∞ ∞
≥ I2 = (cot h(xy) − 1)fn (x)gn (y)dxdy ≥ MJ2 = Mn. (15)
0 0

By (15), in view of σ1 − σ > 0, 0 ≤ k(σ − δ0 ) + k(σ ) < ∞, for n → ∞, we find

1
∞> (k(σ − δ0 ) + k(σ )) ≥ ∞,
σ1 − σ

which is a contradiction.
If σ1 < σ, then for n ∈ N, n > 1
δ0 p , we set the following two functions:
 1
σ + pn −1
f!n (x) : = x ,0 < x ≤ 1 ,
0, x > 1

0, 0 < y < 1
!
gn (y) : = 1
σ1 − qn −1 .
y ,y ≥ 1

We find
 ∞  1  ∞ 1
p q
J!2 : = x p(1−σ )−1 f!n (x)dx
p q
y q(1−σ1 )−1!
gn (y)dy
0 0
 1  p1  ∞ 1
1 1 q
= x n −1 dx y − n −1 dy = n.
0 1

By (12), we have
 ∞  ∞ 1
1 σ + pn −1
I1 ≤ y (σ1 −σ )− n −1 dy (coth(u) − 1)u du
1 0
 1
1 1
σ + pn −1
= (coth(u) − 1)u du
σ − σ1 + 1
n 0
Equivalent Conditions of a Reverse Hilbert-Type Integral Inequality with the. . . 295

 ∞ 
1
σ + pn −1
+ (cot h(u) − 1)u du
1
 1
1
≤ (cot h(u) − 1)uσ −1 du
σ − σ1 0
 ∞ 
+ (coth(u) − 1)uσ +δ0 −1 du
1
1
≤ (k(σ ) + k(σ + δ0 )) ,
σ − σ1

and then by (14), it follows that


1
(k(σ ) + k(σ + δ0 ))
σ − σ1
 ∞ ∞
≥ I1 = (coth(xy) − 1)f!n (x)!
gn (y)dxdy ≥ M J!2 = Mn. (16)
0 0

By (16), for n → ∞, we still find that


1
∞> (k(σ ) + k(σ + δ0 )) ≥ ∞,
σ − σ1
which is a contradiction.
Hence, we conclude that σ1 = σ.
For σ1 = σ, we have nM = MJ2 ≤ I2 by using (14). Then we reduce (13) as
follows:
1 1
M= MJ2 ≤ I2
n n
 1  ∞ 
1 1 σ − 1 −1
= y n −1 (coth(u) − 1)u pn du dy
n 0 y
 1  1 
1 1
−1
1
σ − pn −1
= y n (cot h(u) − 1)u du dy
n 0 y
 ∞
σ − 1 −1
+ (coth(u) − 1)u pn du
1

1
 1  u 
1 σ − 1 −1
= y n −1 dy (cot h(u) − 1)u pn du
n 0 0
 ∞ 1
σ − pn −1
+ (coth(u) − 1)u du
1
 1  ∞
1
σ + qn −1
≤ (cot h(u) − 1)u du + (coth(u) − 1)uσ −1 du. (17)
0 1
296 B. Yang

Since for n > 1


δ0 |q| (n ∈ N), we have

1
σ + qn −1
(coth(u) − 1)u ≤ (coth(u) − 1)uσ −δ0 −1 (0 < u ≤ 1)

and
 1
(coth(u) − 1)uσ −δ0 −1 du ≤ k(σ − δ0 ) < ∞,
0

then by (17) and Lebesgue control convergence theorem (cf. [21]), we have
 1  ∞ 
1
σ + qn −1 σ −1
M ≤ lim (cot h(u) − 1)u du + (coth(u) − 1)u du
n→∞ 0 1
 1  ∞
1
σ + qn −1
= lim (coth(u) − 1)u du + (cot h(u) − 1)uσ −1 du = k(σ ).
0 n→∞ 1

The lemma is proved.


For σ1 = σ, by Lemma 1, we still have
Lemma 2 If there exists a constant M > 0, such that for any non-negative
measurable functions f (x) and g(y) in (0, ∞), the following inequality
 ∞ ∞
I : = (coth(xy) − 1)f (x)g(y)dxdy
0 0
 ∞  1  ∞ 1
p q
≥M x p(1−σ )−1 p
f (x)dx y q(1−σ )−1 q
g (y)dy (18)
0 0

holds true, then we have M ≤ k(σ ).

3 Main Results

Theorem 1 The following conditions are equivalent:


(i) There exists a constant M > 0, such that for any f (x) ≥ 0,
 ∞
0< x p(1−σ )−1 f p (x)dx < ∞,
0

we have the following inequality:


Equivalent Conditions of a Reverse Hilbert-Type Integral Inequality with the. . . 297

 ∞  ∞ p  p1
pσ1 −1
J : = y (coth(xy) − 1)f (x)dx dy
0 0
 ∞ 1
p
>M x p(1−σ )−1 p
f (x)dx ; (19)
0

(ii) there exist a constant M > 0, such that for any g(y) ≥ 0,
 ∞
0< y q(1−σ1 )−1 g q (y)dy < ∞,
0

we have the following inequality:


 ∞  ∞ q  q1
qσ −1
K : = x (coth(xy) − 1)g(y)dy dx
0 0
 ∞ 1
q
>M y q(1−σ1 )−1 q
g (y)dy ; (20)
0

(iii) there exists a constant M > 0, such that for any f (x), g(y) ≥ 0,
 ∞
0< x p(1−σ )−1 f p (x)dx < ∞,
0

and
 ∞
0< y q(1−σ1 )−1 g q (y)dy < ∞,
0

we have the following inequality:


 ∞ ∞
I = (coth(xy) − 1)f (x)g(y)dxdy
0 0
 ∞  1  ∞ 1
p q
>M x p(1−σ )−1 p
f (x)dx y q(1−σ1 )−1 q
g (y)dy ; (21)
0 0

(iv) σ1 = σ.
Proof "(i) => (iii)". By the reverse Hölder’s inequality (cf. [22]), we have
 ∞  ∞  
σ1 − p1 1
−σ
I = y (coth(xy) − 1)f (x)dx y p 1 g(y) dy
0 0
 ∞ 1
q
≥J y q(1−σ1 )−1 g q (y)dy . (22)
0
298 B. Yang

Then by (19), we have (21).


"(ii) => (iii)". Still by the reverse Hölder’s inequality, we have
 ∞ 1   ∞ 
−σ σ− 1
I = yq f (x) x q (coth(xy) − 1)g(y)dy dx
0 0
 ∞ 1
p
≥ x p(1−σ )−1 f p (x)dx K. (23)
0

Then by (20), we have (21).


"(iii) => (iv)". By Lemma 1, we have σ1 = σ.
"(iv) => (i)". Setting u = xy, we obtain the following weight function: For
y > 0,
 ∞
ω(σ, y) : = y σ
(coth(xy) − 1)x σ −1 dx
0
 ∞
= (coth(u) − 1)uσ −1 du = k(σ ). (24)
0

By the reverse Hölder’s inequality with weight and (24), we have


 ∞ p
(coth(xy) − 1)f (x)dx
0
 9 :9 : p
∞ y (σ −1)/p x (σ −1)/q
= (coth(xy) − 1) (σ −1)/q f (x) dx
0 x y (σ −1)/p
 ∞ y σ −1
≥ (coth(xy) − 1) f p (x)dx
0 x (σ −1)p/q
 ∞ p/q
x σ −1
× (coth(xy) − 1) dx
0 y (σ −1)q/p
. /p−1  ∞ y σ −1
= ω(σ, y)y q(1−σ )−1
(coth(xy) − 1) f p (x)dx
0 x (σ −1)p/q
 ∞
p−1 −pσ +1 y σ −1
= (k(σ )) y (coth(xy) − 1) f p (x)dx (25)
0 x (σ −1)p/q

If (25) takes the form of equality for a y ∈ (0, ∞), then (cf. [22]), there exists
constants A and B, such that they are not all zero, and

y σ −1 x σ −1
A (σ −1)p/q
f p (x) = B a.e. in R+ .
x y (σ −1)q/p
Equivalent Conditions of a Reverse Hilbert-Type Integral Inequality with the. . . 299

We suppose that A = 0 (otherwise B = A = 0). Then it follows that


B
x p(1−σ )−1 f p (x) = y q(1−σ ) a.e. in R+ ,
Ax
∞
which contradicts the fact that 0 < 0 x p(1−σ )−1 f p (x)dx < ∞. Hence, (25)
takes the form of strict inequality.
For σ1 = σ, by (25) and Fubini theorem, we have
 ∞ ∞  p1
1 y σ −1
J > (k(σ )) q (coth(xy) − 1) p
f (x)dxdy
0 0 x (σ −1)p/q
 ∞  ∞   p1
1 y σ −1
= (k(σ )) q (coth(xy) − 1) p
dy f (x)dx
0 x (σ −1)(p−1)
0
 ∞ 1
1 p
= (k(σ )) q ω(σ, x)x p(1−σ )−1 p
f (x)dx
0
 ∞ 1
p
= k(σ ) x p(1−σ )−1 f p (x)dx . (26)
0

Setting 0 < M ≤ k(σ ), then (19) follows.


"(iv) => (ii)". In the same way, we obtain (20).
Therefore, the conditions (i), (ii), (iii) and (iv) are equivalent.
For σ1 = σ , we still have
Theorem 2 The following conditions are equivalent:
∞
(i) For any f (x) ≥ 0, 0 < 0 x p(1−σ )−1 f p (x)dx < ∞, we have the following
inequality:

 ∞  ∞ p  p1
pσ −1
y (coth(xy) − 1)f (x)dx dy
0 0
 ∞ 1
p
> k(σ ) x p(1−σ )−1 p
f (x)dx ; (27)
0
∞
(ii) for any g(y) ≥ 0, 0 < 0 y q(1−σ )−1 g q (y)dy < ∞, we have the following
inequality:

 ∞  ∞ q  q1
qσ −1
x (coth(xy) − 1)g(y)dy dx
0 0
 ∞ 1
q
> k(σ ) y q(1−σ )−1 q
g (y)dy ; (28)
0
300 B. Yang

∞
(iii) for any f (x), g(y) ≥ 0, 0 < 0 x p(1−σ )−1 f p (x)dx < ∞, and
 ∞
0< y q(1−σ )−1 g q (y)dy < ∞,
0

we have the following inequality:


 ∞ ∞
(coth(xy) − 1)f (x)g(y)dxdy
0 0
 ∞  1  ∞ 1
p q
p(1−σ )−1 p q(1−σ )−1 q
> k(σ ) x f (x)dx y g (y)dy . (29)
0 0

Moreover, the constant factor k(σ ) in (27), (28) and (29) is the best possible.
Proof For σ1 = σ in Theorem 1, since 0 < k(σ ) < ∞, setting M = k(σ ) in (19),
(20) and (21), in the same way, we still can prove that the conditions (i), (ii) and (iii)
are equivalent in Theorem 2. If there exists a constant M ≥ k(σ ), such that (29) is
valid, then by Lemma 2, we have M ≤ k(σ ). Hence, the constant factor M = k(σ )
in (29) is the best possible. The constant factor k(σ ) in (27) ((28)) is still the best
possible. Otherwise, by (22) (or (23)) for σ1 = σ, we can conclude that the constant
factor M = k(σ ) in (29) is not the best possible.

4 Some Corollaries

In particular, for σ = 1
p (> 1) in Theorem 2, we have
Corollary 1 The following conditions are equivalent:
∞
(i) For any f (x) ≥ 0, 0 < 0 x p−2 f p (x)dx < ∞, we have the following
inequality:

 ∞  ∞ p  p1  ∞ 1
1 p
(coth(xy) − 1)f (x)dx dy > k( ) x p−2 f p (x)dx ;
0 0 p 0
∞ (30)
(ii) for any g(y) ≥ 0, 0 < 0 g q (y)dy < ∞, we have the following inequality:

 ∞  ∞ q  q1  ∞ 1
1 q
x q−2
(coth(xy) − 1)g(y)dy dx > k( ) q
g (y)dy ;
0 0 p 0
∞ (31)
(iii) for any f (x), g(y) ≥ 0, 0 < 0 x p−2 f p (x)dx < ∞, and 0 <
∞ q
0 g (y)dy < ∞, we have the following inequality:
Equivalent Conditions of a Reverse Hilbert-Type Integral Inequality with the. . . 301

 ∞ ∞
(coth(xy) − 1)f (x)g(y)dxdy
0 0
 ∞  1  ∞ 1
1 p q
> k( ) x p−2 f p (x)dx g q (y)dy ; (32)
p 0 0

Moreover, the constant factor

1 1 1
k( ) = 21/q Γ ( )ζ ( )
p p p

in (30), (31) and (32) is the best possible.


Setting y = Y1 , G(Y ) = g( Y1 ) Y12 in Theorem 1–2, then replacing Y (G(Y )) by y
(g(y)), we have
Corollary 2 The following conditions are equivalent:
(i) There exists a constant M > 0, such that for any f (x) ≥ 0,
 ∞
0< x p(1−σ )−1 f p (x)dx < ∞,
0

we have the following inequality:

 ∞  ∞ p  p1
−pσ1 −1 x
y (coth( ) − 1)f (x)dx dy
0 0 y
 ∞ 1
p
>M x f (x)dx ;
p(1−σ )−1 p
(33)
0

(ii) there exists a constant M > 0, such that for any g(y) ≥ 0,
 ∞
0< y q(1+σ1 )−1 g q (y)dy < ∞,
0

we have the following inequality:

 ∞  ∞ q  q1
qσ −1 x
x (coth( ) − 1)g(y)dy dx
0 0 y
 ∞ 1
q
>M y g (y)dy ;
q(1+σ1 )−1 q
(34)
0

(iii) there exists a constant M > 0, such that for any f (x), g(y) ≥ 0,
302 B. Yang

 ∞
0< x p(1−σ )−1 f p (x)dx < ∞,
0

and
 ∞
0< y q(1+σ1 )−1 g q (y)dy < ∞,
0

we have the following inequality:


 ∞ ∞ x
(coth( ) − 1)f (x)g(y)dxdy
0 0 y
 ∞  1  ∞ 1
p q
>M x p(1−σ )−1 p
f (x)dx y g (y)dy ; (35)
q(1+σ1 )−1 q
0 0

(iv) σ1 = σ.
Corollary 3 The following conditions are equivalent:
∞
(i) For any f (x) ≥ 0, 0 < 0 x p(1−σ )−1 f p (x)dx < ∞, we have the following
inequality:

 ∞  ∞ p  p1
−pσ −1 x
y (coth( ) − 1)f (x)dx dy
0 0 y
 ∞ 1
p
> k(σ ) x p(1−σ )−1 f p (x)dx ; (36)
0
∞
(ii) for any g(y) ≥ 0, 0 < 0 y q(1+σ )−1 g q (y)dy < ∞,we have the following
inequality:

 ∞  ∞ q  q1
qσ −1 x
x (coth( ) − 1)g(y)dy dx
0 0 y
 ∞ 1
q
> k(σ ) y q(1+σ )−1 g q (y)dy ; (37)
0
∞
(iii) for any f (x), g(y) ≥ 0, 0 < 0 x p(1−σ )−1 f p (x)dx < ∞, and
 ∞
0< y q(1+σ )−1 g q (y)dy < ∞,
0

we have the following inequality:


Equivalent Conditions of a Reverse Hilbert-Type Integral Inequality with the. . . 303

 ∞ ∞ x
(coth( ) − 1)f (x)g(y)dxdy
0 0 y
 ∞  1  ∞ 1
p q
> k(σ ) x p(1−σ )−1 f p (x)dx y q(1+σ )−1 g q (y)dy . (38)
0 0

Moreover, the constant factor k(σ ) in (36), (37) and (38) is the best possible.
In particular, for σ = p1 (> 1) in Corollary 3, we have
Corollary 4 The following conditions are equivalent:
∞
(i) For any f (x) ≥ 0, 0 < 0 x p−2 f p (x)dx < ∞, we have the following
inequality:

 ∞  ∞ p  p1
−2 x
y (coth( ) − 1)f (x)dx dy
0 0 y
 ∞ 1
1 p
> k( ) x p−2 p
f (x)dx ; (39)
p 0
∞
(ii) for any g(y) ≥ 0, 0 < 0 y 2(q−1) g q (y)dy < ∞,we have the following
inequality:

 ∞  ∞ q  q1
x
x q−2 (coth( ) − 1)g(y)dy dx
0 0 y
 ∞ 1
1 q
> k( ) y g (y)dy ;
2(q−1) q
(40)
p 0
∞
(iii) for any f (x), g(y) ≥ 0, 0 < 0 x p−2 f p (x)dx < ∞, and 0 <
 ∞ 2(q−1) q
0 y g (y)dy < ∞, we have the following inequality:
 ∞ ∞ x
(coth( ) − 1)f (x)g(y)dxdy
0 0 y
 ∞  1  ∞ 1
1 p q
> k( ) x p−2 f p (x)dx y 2(q−1) g q (y)dy . (41)
p 0 0

Moreover, the constant factor

1 1 1
k( ) = 21/q Γ ( )ζ ( )
p p p

in (39), (40) and (41) is the best possible.


304 B. Yang

5 Conclusions

By the use of the way of real analysis and weight functions, we study some
equivalent conditions of a reverse Hilbert-type integral inequality with the non-
homogeneous kernel of the hyperbolic cotangent function, related to the Riemann
zeta function in Theorems 1–2. Some equivalent conditions of a reverse Hilbert-
type integral inequality with the homogeneous kernel are deduced in Corollary 2.
We also consider some particular cases in Corollarys 1 and 3–4. The lemmas and
theorems provide and extensive account of this type of inequalities.

Acknowledgements This work is supported by the National Natural Science Foundation


(No. 61772140), and Science and Technology Planning Project Item of Guangzhou City (No.
201707010229). We are grateful for this help.

References

1. I. Schur, Bernerkungen sur Theorie der beschrankten Billnearformen mit unendlich vielen
Veranderlichen. J. Math. 140, 1–28 (1911)
2. G.H. Hardy, Note on a theorem of Hilbert concerning series of positive terms. Proc. Lond.
Math. Soc. 23(2), xlv–xlvi (1925). Records of Proc
3. G.H. Hardy, J.E. Littlewood, G. Pólya, Inequalities (Cambridge University Press, Cambridge,
USA)
4. D.S. Mitrinović, J.E. Pecaric, A.M. Fink, Inequalities Involving Functions and their Integrals
and Deivatives (Kluwer Academic, Boston, 1991)
5. B.C. Yang, On Hilbert’s integral inequality. J. Math. Anal. Appl. 220, 778–785 (1998)
6. B.C. Yang, A note on Hilbert’s integral inequality. Chin. Q. J. Math. 13(4), 83–86 (1998)
7. B.C. Yang, On an extension of Hilbert’s integral inequality with some parameters. Aust. J.
Math. Anal. Appl. 1(1), 1–8 (2004), Art.11
8. B.C. Yang, I. Brnetic, M. Krnic, J.E. Pecaric, Generalization of Hilbert and Hardy-Hilbert
integral inequalities. Math. Ineq. Appl. 8(2), 259–272 (2005)
9. M. Krnic, J.E. Pecaric, Hilbert’s inequalities and their reverses. Publ. Math. Debrecen 67(3–4),
315–331 (2005)
10. Y. Hong, On Hardy-Hilbert integral inequalities with some parameters. J. Ineq. Pure Appl.
Math. 6(4), 1–10 (2005), Art. 92
11. B. Arpad, O. Choonghong, Best constant for certain multi linear integral operator. J. Inequal.
Appl. 2006(28582), 1–12 (2006)
12. Y.J. Li, B. He, On inequalities of Hilbert’s type. Bull. Aust. Math. Soc. 76(1), 1–13 (2007)
13. W.Y. Zhong, B.C. Yang, On multiple Hardy-Hilbert’s integral inequality with kernel. J. Inequal.
Appl. 2007, 17, Art ID 27962
14. J.S. Xu, Hardy-Hilbert’s Inequalities with two parameters. Adv. Math. 36(2), 63–76 (2007)
15. M. Krnić, M.Z. Gao, J.E. Pečarić, et al., On the best constant in Hilbert’s inequality. Math.
Inequal. Appl. 8(2), 317–329 (2005)
16. Y. Hong, On Hardy-type integral inequalities with some functions. Acta MathmaticaS inica
49(1), 39–44 (2006)
17. B.C. Yang, The Norm of Operator and Hilbert-Type Inequalities (Science Press, Beijing, 2009)
18. B.C. Yang, Hilbert-Type Integral Inequalities (Bentham Science Publishers Ltd., The United
Emirates, 2009)
Equivalent Conditions of a Reverse Hilbert-Type Integral Inequality with the. . . 305

19. B.C. Yang, On Hilbert-type integral inequalities and their operator expressions. J. Guangaong
Univ. Edu. 33(5), 1–17 (2013)
20. Y. Hong, On the structure character of Hilbert’s type integral inequality with homogeneous
kernal and applications. J. Jilin Univ. (Science Edition) 55(2), 189–194 (2017)
21. J.C. Kuang, Real and Functional Analysis (Continuation)(second volume) (Higher Education
Press, Beijing, 2015)
22. J.C. Kuang, Applied Inequalities (Shangdong Science and Technology Press, Jinan, 2004)
Index

A Contour integration, 212–213


Abel sum, 194 Converse inequality, 121, 125
Additive energy, 263 Converse Marcinkiewicz-Zygmund type
Airy analogue of a fundamental of Lagrange inequality, 123
interpolation, 126 Converse quadrature inequality, 121
Airy function Ai (x), 125–127, 130, 133, 134 Convex continuous functions, 274
Angular speed, 33 Cosecant sums, 185, 186
Apostol-Eisenstein series, 189 Cosine sums, 86
Approximate functional equation, 170, 171, Cotangent function, 184
173 Cotangent sums
Arithmetic mean–geometric mean inequality, distribution, 150
82 Eisenstein series, 8
Atiyah-Singer index theorem, 191 ellipse, 12–14
finite cotangent sums, 184
Hurwitz zeta function, 9
B modularity property, 9
Balog–Wooley decomposition, 262 Nyman-Beurling criterion, 5
Berndt’s methods, 212 reciprocity formula, 9
Berndt’s transformation formulae, 211 Riemann hypothesis, 5
Bernoulli polynomials, 185 Riemann zeta function, 5, 7
Bernstein factors, 34 Vasyunin sum, 6, 150
Bessel function, 99, 120, 122, 123
Bourgain’s L1 problem, 60–64
D
Dedekind eta function, 187, 188, 192, 197, 212
C Dedekind sums, 188
Cauchy residue theorem, 184, 194, 211, 214 analytic number theory, 191–192
Chebyshev inequality, 87 applications, 191
Christoffel-Darboux formula, 133 arithmetical function, 192
Christoffel function, 129 Bernoulli function, 193
Classical arithmetric mean–geometric mean cot function, 193–194, 196–197
inequality, 87 DC sums, 203–205
Classical Hermite polynomial, 124 Dedekind-eta function, 192, 197
Classical Mehler-Heine asymptotic, 122–123 elliptic modular functions, 191–192
Clausen function, 206, 210 Fourier series, 192

© Springer Nature Switzerland AG 2020 307


A. Raigorodskii, M. T. Rassias (eds.), Trigonometric Sums and Their Applications,
https://doi.org/10.1007/978-3-030-37904-9
308 Index

Dedekind sums (cont.) Estermann zeta function, 6, 7


history, 191 cotangent sum, 6
Hurwitz zeta function, 196 functional equation, 7
integral representation, 194 values, 150
Lambert series, 197 Euler beta functions, 208
modular transformation, 198 Euler–Mascheroni constant, 5, 24
multiple, 195 Euler numbers (Em ∗ ), 204, 212

reciprocity formula, 25 Euler phi-function, 14


reciprocity law, 192 Euler’s summation formula, 172
sawtooth function, 24 Euler’s totient function, 59
types, 212 Euler summation formula, 11–12
Dedekind type Daehee-Changhee (DC) sums Extended Riemann zeta function, 100
Clausen function, 210
coprime integers, 205
coprime positive integers, 207–208 F
definition, 205 Fatou lemma, 106–107
Dirichlet beta function, 209 Fejér-Jackson-Gronwall inequality
elementary calculations, 207 nonnegative trigonometric polynomials, 72
Euler beta functions, 208 sine polynomials, 71
Euler polynomials, 203–205 Vietoris theorem, 72, 73
Hadjicostas’s double integral formula, 208 Fekete polynomials, 50–53
Hurwitz-Lerch zeta function, 209 Finite cotangent sums, 184
Hurwitz zeta function, 209 Fourier coefficients, 263, 266, 269, 271, 274
Legendre chi function, 209, 210 Fourier expansion, 189–191, 193, 203
Lerch’s transcendent, 208 Fourier series, 192, 200, 203, 274, 275
Riemann zeta function, 208 Fourier transform, 98, 113, 203, 204, 261, 263,
secant function, 205 266–267
trigonometric representation, 205–208 Fubini theorem, 105, 106, 299
Degree of exactness, 186, 223 Function g(x)
de la Vallée-Poisson kernels, 279–280, absolute constant, 19
282–283 binomial theorem, 23
Dirichlet beta function, 209 continued fraction algorithm, 23
Dirichlet polynomial, 7 continued fraction expansion, 20
Dirichlet’s approximation theorem, 18 Euler–Mascheroni constant, 24
Double-inequality, 87 gamma function, 20
Double-sided Taylor’s approximations Gauss map α, 21
automated-oriented methods, 166 Neumann series, 22
Euler’s number, 162, 163 Wilton’s function, 21
first and second approximations, 160–161
standard numerical methods, 164
Theorem WD, 161 G
trigonometric inequalities, 159–160 Gamma function, 20, 36, 98, 210, 218, 234,
292
Gardner-Fisher trigonometric inverse power
E sum, 186
Eisenstein series, 187 Gauss-Chebyshev quadrature sums
Fourier expansion, 189–191 algebraic polynomials, 218
Lipschitz formula, 189, 190 hypergeometric function, 219–220
Riemann zeta function, 189 Legendre polynomial, 218–221
standard results, 189 quadrature formula, 217
Eisenstein’s identity, 86 recurrence coefficients, 217
Erdős conjecture, 32 trigonometric sums, 222
Index 309

Gauss-Lobatto formulas, 222 Hilbert space, 3


Gauss quadrature, 120, 121, 125 Hilbert-type operator, 246–248
Gauss-Radau formulas, 222, 224 Hölder inequality, 61, 103, 240–243, 250, 252,
Generalized Laplace transform, 98 254, 255, 265–266, 297–298
Generalized Stieltjes transform, 99, 115 Hurwitz-Lerch zeta function, 209
Gram points, 171 Hurwitz zeta function, 188, 196, 208–209,
215–216, 232
Hyperbolic cotangent function, 292–296
H Hyperbolic secant function, 231
Hadjicostas’s double integral formula, 208
Half-discrete Hardy-Hilbert inequality, 230
equivalent inequalities, 238–240, 291 I
equivalent reverse inequalities, 230, Inequalities
248–249 Chebyshev inequality, 87
gamma function, 234 classical arithmetric mean–geometric
Hölder inequality, 240–243, 250, 252–257 mean, 87
homogenous kernel, 230, 248–249 cosine sums, 86–95
Lebesgue term, 233–234, 240, 242–245, double-sided Taylor’s approximations,
250, 254 159–160
operator expressions, 245–249 Vietoris’ theorem, 77–82
weight functions, 232–238 Infinitely differentiable functions, 275
whole plane, 231 Integral operator, 98–102, 107, 108
Hankel transform, 98, 99 Inverse Fourier transform, 261
Hardy-Berndt sums, 205
Hardy-Hilbert inequality, 230, 289
Hardy-Hilbert integral inequality, 230, 290 J
Hardy’s sum Jacobi polynomials, 120–123
Bernoulli functions, 202 Jacobi (Legendre) symbol, 195
elementary methods, 200–202
Fourier series, 199–200
reciprocity theorems, 211
theta-functions vs. Dekind eta-function, K
199 Kronecker delta function, 7, 150, 191
trigonometric formulas, 199–201 Kronecker limit formula, 191
Zagier-type identities, 203
Hardy’s Z-function
Abel summation formula, 178 L
asymptotic expansion, 173 Lagrange interpolation, 120, 123, 126, 127
definition, 169 Lambert series, 186–187, 194, 195
derivatives Z (k) (t), k = 1, 2, 170 Laplace transform
Euler’s summation formula, 172 asymptotic analysis, 98–100
inequalities, 177 one-sided, 97
Riemann-Siegel approximate functional Laurent expansion, 213
equation, 170 Lebesgue control convergence theorem, 296
value distribution, 170 Lebesgue measure, 13
Hermite-Biehler function, 124 Left-hand inequality, 120
Hermite polynomials, 124, 125, 127, 129 Legendre chi function, 209
Hermite weight, 124, 131 Legendre polynomial, 218–221
Higher-order Clausen function, 210 Lemma
Hilbert integral inequality, 289 Chebyshev polynomial, 88
homogeneous kernel, 300–302 Euler-Maclaurin formula, 88–91
non-homogeneous kernel, 290, 296–300 logarithmic derivative, 88
non-negative homogeneous functi, 291 sine polynomials, 74–77
310 Index

Lerch’s transcendent, 208 Plancherel-Polya inequalities, 119, 124


Lipschitz formula, 189, 190 Pochhammer symbol, 218
Littlewood polynomials, 40–41 Poisson summation formula, 212
Littlewood’s flatness problem, 31–32 Polya-Plancherel inequalities, 120
Littlewood’s theorem, 4 Polynomials
algebraic, 29
conjugate-reciprocal, 31
M Jensen formula, 30
Marcinkiewicz-Zygmund inequalities, 119, Lebesgue measure, 31
120, 123–125 Mahler measure, 30
Midpoint rules, 224 p-th Bernoulli function, 193
Milne’s inequality, 87
Minkowski’s inequality for integrals, 104–107
Möbius function, 5 Q
Modular Littlewood problem, 262 Quadrature sums, 123, 223
Montgomery’s conjecture, 46, 50
Multiplicative energy, 262, 264
Multiplicative Fourier transform, 266, 267
R
Random matrix theory, 126
N Rational numbers, maximum of c0
n–dimensional Fourier cosine transform, 113 analytic function, 15
n–dimensional Fourier sine transform, 112 fraction expansion, 15
Negative/positive Fourier coefficients, 263 partial fractions, 16
Negligibility theorem, 34 Real trigonometric polynomials, 54
Newman, D.J., 41 Reciprocity law
Non-negative measurable function, 100, 101, arithmetic sums, property, 212
107 Berndt’s transformation formulae, 211
Norm inequalities, 99 Cauchy residue theorem, 211, 214
Nyman–Beurling–Baez–Duarte–Vasyunin contour integration method, 212–213
approach, 3 Hardy sums, 211
Nyman–Beurling criterion, 6, 25–27 Hurwitz zeta function, 215–216
Laurent expansion, 213
Reciprocity theorem, 85
O Riemann hypothesis, 1, 2
One-sided Laplace transforms, 97 Lindelöf hypothesis, 4
Operator expressions, 245–249 Möbius function, 25–27
Orthogonal trigonometric approximations, 275 Vasyunin sum, 150–151
Abel transform, 278 Riemann-Roch theorem, 191
(ψ, β)–derivative, 274, 277 Riemann-Siegel approximate functional
estimate the integral, 278 equation, 170
metric of space L∞ Riemann-Stieltjes integral, 212
ψ
Classes Lβ,p , 1 < p < ∞, 276–282 Riemann zeta function, 5, 170, 189, 216, 232,
ψ 291, 292
Classes Lβ,1 , 282–284 analytic continuations, 209
metric of space L, 1 < s < ∞ DC-Sums, 208
ψ
Classes Lβ,1 , 284–286 definition, 2
order estimates, 276 extended Riemann zeta function, 100
Hardy’s function Z(t), 169
inequalities, 160
P non-trivial zeros, 2
Parseval identity, 267–268, 270–271 Riemann hypothesis, 3
Parseval’s formula, 31, 61, 62 trivial zeros, 2
Partial fraction decomposition, 186 zeros, 7
Index 311

Rudin-Shapiro polynomials U
binary sequences, 42 Ultraflat sequence, 34–36
Littlewood, 41, 42 Uniform distribution conjecture, 33–34
Mahler measure, 41, 43–44 Unimodular polynomials
moments, 43–44 flatness of conjugate-reciprocal, 38–40
unimodular, 41 Littlewood polynomials, 40–41
Rudin-Shapiro polynomials, 41–44
ultraflat sequences, 31–38
S Unimodular zeros, 53–60
Saffari conjecture
consequences, 46–49
Rudin-Shapiro polynomials, 45–46 V
Salem-Zygmund theorem, 63 Vasyunin sum, 6, 150, 151
Scaling limits, 123
Secant function, 205, 257
Second moment, kth derivative of Hardy’s W
function, 170 Weierstrass elliptic functions, 191
Self-reciprocal polynomials, 53–60 Weight functions, 232–245, 291, 298
Self-reciprocal unimodular polynomials, Weyl–Nagy classes, 276
39 Wiener norm
Sine sums, 87 additive energy, 263
Skew-reciprocal unimodular polynomials, additive Fourier transform, 267
39 Balog–Wooley decomposition, 262
Stieltjes constants, 171 definition, 262
Stirling’s formula, 83 Fourier coefficients, 263
Sum–product phenomenon, 262, 265 Gauss sum, 267
Symmetric real function, 269 inverse inequality, 268
Symmetric set, 269–270 inversion formula, 266–267
multiplicative energy, 262, 264
multiplicative Fourier coefficients, 266–267
T multiplicative Fourier transform, 267
Tangent function, 184–185 multiplicative subgroups, 268–269
Theorem on double-sided Taylor’s Parseval identity, 267–268, 270–271
approximations, 161 sum–product phenomenon, 262
Theta-functions, see Hardy’s sums symmetric set, 269–270
Third derivative test, 169–172 Wiener norm
Trapezoidal rules, 224 estimating M+ , 269–270
Trigonometric polynomials, 63 Wilton number, 21
Trigonometric quadrature rules, 186, 223–225
Trigonometric sums, 183
2π –periodic functions, (ψ, β)–derivative, 273, Z
274, 277 Zagier-type identities, 203

You might also like