TRIGONOMETRIA
TRIGONOMETRIA
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
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
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.
v
Contents
vii
viii Contents
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
On a Category of Cotangent Sums
Related to the Nyman-Beurling Criterion
for the Riemann Hypothesis
1 Introduction
Definition 1.1
b−1
r m π mr
c0 := − cot , (1)
b b b
m=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
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.
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
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”.
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
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 −∞
ζ (s)
Fλ (s) = (λs − λ) , s ∈ ,
s
ζ (s)
Gl (s) = (l −s − l −1 ) , s ∈ .
s
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].
• cn = O( n1 ),
then
∞
cn = a.
n=0
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
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.
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
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
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
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
r̄
× cos E 1 + α − s, , α
2 b
πα r̄
− cos π s − E 1 + α − s, − , α , (13)
2 b
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
(−1)α+1 Bα+1
E(0, 1, α) = ,
2(α + 1)
(−1)m+1 2(2m)!
B2m = ζ (2m), f or m ≥ 1.
(2π )2m
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
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
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
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
Then the difficulty lies in obtaining a good approximation of the sum S(L; b)
defined by
1a
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
2N +1
(2N +1) (2N +1) 2N + 1
(u−u+B) = ((u−u)+B) := (u−u)j B2N +1−j ,
j
j =0
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
12 N. Derevyanko et al.
b−1
r π mr rm
Q = cot .
b b b
m=1
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
+∞
1 − 2{lx}
g(x) := ,
l
l=1
and
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
(b)
(A1 − A0 )Hk/2 , for even k
x dμ =
k
0, otherwise.
1 1 r
lim f c0 = f dμ,
b→+∞ φ(b) b b
r:(r,b)=1,
A0 b≤r≤A1 b
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.
:= (b, C) = b−C .
We define
M(b, C, A0 ) := max c0 r .
A0 b≤r<(A0 +)b 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
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.
1
D+ε < − C.
2
Set
:= b−(D+ε) .
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 .
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
Then we have
in contradiction to vs ≤ b.
In the same manner we obtain from vlj ≥ ε log b, j = 1, 2 :
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
and
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
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
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.
There is an interesting connection between the cotangent sums c0 and the function
+∞
1 − 2{lx}
g(x) := , (36)
l
l=1
b k
1 r
c0 = Hk bk + Oε (bk−1+ε (Ak log b)2k ), (37)
φ(b) b
r=1
(r,b)=1
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
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
1
γk (x) = βk−1 (x) log , where k ≥ 0,
αk (x)
converges.
Wilton’s function W is defined by
W= (−1)k γk (x)
k≥0
where
1
l(x) = log
x
and the operator T is defined by
Tf (x) = xf (α(x)).
22 N. Derevyanko et al.
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
n
L(x, n) = (−1)k (T k l)(x). (43)
k=0
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
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
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
eγ
|g(x)|K dx = (K + 1)(1 + O(exp(−CK)))
π
0
5 Dedekind Sums
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
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
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
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
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.
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]
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
where
2π q 1/q
1 it
Mq (Q) := Q(e ) dt , q > 0.
2π 0
n
M0 (Q) = |c| max{1, |zk |}
k=1
n
Q(z) = c (z − zk ) , c, zk ∈ C .
k=1
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
n
Q(z) = aj zj , aj ∈ C ,
j =0
n
∗
Q (z) := z Q(1/z) :=
n
a n−j zj .
j =0
By Parseval’s formula,
2π 2
Pn (eit ) dt = 2π(n + 1)
0
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
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 )| .
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
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
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
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
2π 2
n
ak,n − a n−k,n 2 = 1 (Pn − Pn∗ )(eit ) dt ≥
1
+ hn n ,
2π 0 3
k=0
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
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
2
n
2 1 2π
2n3
k 2 ak,n − a n−k,n = (Pn − Pn∗ )(eit ) dt ∼ .
2π 0 3
k=0
n
ak,n an−k,n = o(n) .
k=0
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.
M1 (P ) ≤ (1 − ε) 1/3 n3/2
M∞ (P ) ≥ (1 + ε) 1/3 n3/2
M∞ (P ) ≥ (1 + ε) 4/3 n1/2
and
M∞ (P ) ≥ (1 + ε) 1/3 n3/2
M∞ (P ) ≥ (1 + ε) 1/3 n3/2
or at least
M∞ (P ) ≥ (1 + ε) 1/3 n3/2
or at least
M∞ (P ) ≥ (1 + ε)n1/2
or at least
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
1 Mq (f ) q 1/q
lim 1/2
= 1+
n→∞ 2n+1 n 2
f ∈Ln
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
5 Rudin-Shapiro Polynomials
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 .
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
Hence
and hence
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
M2 (f )4 n2
MF(f ) = = .
M4 (f )4 − M2 (f )4 M4 (f )4 − n2
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
!k := 2−(k+1)/2 Pk
P and !k := 2−(k+1)/2 Qk .
Q
|P !k (z)|2 = 1 ,
!k (z)|2 + |Q z ∈ ∂D .
2π
Iq (P !k ) q := 1
!k ) := Mq (P !k (eit )|q dt .
|P
2π 0
Consequently
!k ) ≤ L
Im (P , k = 1, 2, . . . , m = 1, 2, . . . .
log(m + 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
and
2πj
zj := eitj , tj := , j ∈ Z,
n
then
0 < τ1 ≤ τ2 ≤ · · · ≤ τm ≤ 2π , τ0 := τm − 2π , τm+1 := τ1 + 2π ,
δ := max{τ1 − τ0 , τ2 − τ1 , . . . , τm − τm−1 } .
m
τj +1 − τj −1 2π
log |P (e iτj
)| ≤ log |P (eiτ )| dτ + B
2 0
j =1
(2n)1/2
Mq (Pk ) = Mq (Qk ) ∼
(q/2 + 1)1/q
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
n−1
|Pk (z)|2 = Pk (z)Pk (1/z) = aj zj , z ∈ ∂D ,
j =−n+1
then
where
48 T. Erdélyi
√ 1/3 √ 1/3
44 + 3 177 + 44 − 3 177 −1
λ := − = −1.658967081916 · · ·
3
n
f (t) = am cos(mt + αm ) , am , αm ∈ R ,
m=0
n/
h 2
am
sn/ h = ≤ 2−9 c6
μ2
m=1
|v| ≤ V = 2−5 c3 ,
then
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
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
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
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?
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
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
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
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
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
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
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
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)
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
n
T (t) := P (eit )e−int = an + 2|aj +n | cos(j t + θj )
j =1
aj = an−j , j = 0, 1, . . . , n . (13)
n
T (t) := P (eit )e−int = an + 2aj +n cos(j t) .
j =1
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
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
n
Tn (t) = aj cos(j t) ,
j =0
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
n
Tn (t) = aj,n cos(j t) ,
j =0
is finite and
n
P (z) = aj zj , aj ∈ C ,
j =0
let
and
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
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→∞
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
then
∗ c log log log |P (1)|
NZ (Tn ) ≥ −1
1 + log M log log log log |P (1)|
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)|
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∈→∞
n−1
k
Pn (z) = z2 , n = 1, 2, . . . ,
k=0
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
min M4 (P )4 = 2n(n − 1) + n ,
P ∈An
1 (Mp (f ))p
lim = (1 + p/2) ,
n→∞ 2n+1 np/2
f ∈Ln
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, . . . ,
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
= 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→∞
1
and hence S0 ≥ √ .
2 2
64 T. Erdélyi
4π (log n)3/2
≤ ≤ β − α ≤ 2π , (16)
n n1/2
while
√
M1 (Qn , I ) ≤ c2 (ε) n , n = 1, 2, . . . ,
(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
Reference
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
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.
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]
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
then
n
0< ak sin(kx) (n ≥ 1; 0 < x < π ).
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
n−1
n k
0≤ − sin(kx) (n ≥ 1; 0 ≤ x ≤ π ). (5)
k n
k=1
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
m−1
1
sin(kx) + sin(mx) ≥ 0.
2
k=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
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),
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
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
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,
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
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,
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
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
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
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
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
with z1 (β), z2 (β), z3 (β) as defined in (16). Then we have the representation
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).
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
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
∂ 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
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
1
4 β 1 4 β
lim δn,β = 1·1· ·1= . (22)
n→∞ 2 e 2 e
Since
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
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
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.
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
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]
(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
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) = ,
jπ
j =1 1 − cos n+1
n
wj
Cn∗ (w) = , (3)
jπ
j =1 1 + cos 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
n
wj
Sn (w) = ,
jπ
j =1 1 − sin n+1
n
wj
Sn∗ (w) = , (4)
jπ
j =1 1 + sin n+1
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
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
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
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
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
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
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
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
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
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
(Bn (w))a ≤ Bn (w) ≤ Sn (w) Sn∗ (w) ≤ (Bn (w))2 ≤ (Bn (w))b .
S2 (w)S2∗ (w)
1≤ = 41−a .
(B2 (w))a
Thus, a ≤ 1.
Inequalities for Weighted Trigonometric Sums 93
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
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
gives
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
n
wj
Dn (w) = .
2 2j π
j =1 sin n+1
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.
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]
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
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
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
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
∞
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
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
Tf p,ω c1 c2
T = sup ≤ ( )1/p ( )1−(1/p) . (14)
f =0 f p λ1 λ2
then
c3
T ≥ . (16)
λ2
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
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
where
c0
c = T = (19)
λ0
3 Proof of Theorem 3
1 1 λ p λ
+ + (1 − ) = 1, + p(1 − ) = 1.
p1 q1 n q1 n
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
Tf p,ω c1 c2
T = sup ≤ ( )1/p ( )1−(1/p) . (25)
f =0 f p λ 1 λ 2
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))
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
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
(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
∞
(Γ (1/α))n n
−1 c3
T ≥ n−1 K(u)u p1 λ2 du = . (34)
α Γ (n/α)λ2 0 λ2
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
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
c0
T5 ≤ 1/p 1/q
.
λ1 λ2
If 0 < λ1 ≤ λ2 , then
c0
T5 ≥ .
λ2
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)
∞
(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
(Γ (1/α))n n
T7 = n−1
Γ( ). (51)
λ0 βα Γ (n/α) λ0 βq
π n/2 n
T7 = n−1
Γ ( ). (53)
2 Γ (n/2) q
∞
(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
(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
)
(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)
π(Γ (1/α))n
T10 = e−(β1 β2 ) . (61)
2λ0 β2 α n−1 Γ (n/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
)
Thus, we have
(π/2)1/2
T11 = .
Γ ( p1 ) cos( 2q
π
)
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
)
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
(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
(Γ (1/α))n π
c0 = × .
βα n−1 Γ (n/α) sin( λnπ
2 βq
)
(Γ (1/α))n π
T13 = × . (69)
λ0 βα n−1 Γ (n/α) sin( λnπ
0 βq
)
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
(Γ (1/α))n n n
c0 = B( ,β − ).
α n−1 Γ (n/α) λ2 q λ2 q
(Γ (1/α))n n n
T14 = n−1
B( ,β − ). (71)
λ0 α Γ (n/α) λ0 q λ0 q
1 1
T3 = B( , β − ),
q q
(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
(Γ (1/α))n n n
c0 = × {B ,1 − β + B β − ,1 − β }
α n−1 Γ (n/α) qλ2 qλ2
(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.
1 Introduction
D. S. Lubinsky ()
School of Mathematics, Georgia Institute of Technology, Atlanta, GA, USA
e-mail: [email protected]
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
Ap 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, Ap 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 = Ap 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 = A2 = 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
α,β n+α
and is normalized by Pn (1) = n . Let
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
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
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
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
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
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
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
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
is the Hermite weight, and {pn } are the orthonormal Hermite polynomials, so that
∞
pn pm W 2 = δmn . (16)
−∞
−∞ < 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
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)
−∞
1 1
r <1− ; r ≤ R; R>− . (20)
p 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
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
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
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
∞
∞
|f (ak )|p 6
≤A 2 |f (t)|p dt. (27)
Ai (ak ) 2 π −∞
k=1
xn ∼ yn
C1 ≤ xn /yn ≤ C2
We also let
n−1
Kn (x, y) = pj (x) pj (y)
j =0
pn (x)
j n (x) = .
pn xj n x − xj n
Lemma 2.1
(a)
-
γn−1 n
= . (32)
γn 2
we have
(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→∞
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
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
(f) We note the following estimates [10, p. 465–467]: uniformly for n ≥ 1 and
x ∈ R,
W 2 (x)
λn (x) ∼ √ ψn (x)−1/2 , (42)
n
W 2 (x)
λn (x) ≥ C √ ψn (x)1/2 . (43)
n
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
1/2 −1/2
= λj n W −2 xj n λn (x) W −2 (x)
−1/4
≤ Cψn xj n ψn (x)1/4 .
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 .
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 .
Here since [32, p. 106, (5.5.10)], Hn (x) = 2nHn−1 (x) so from (17),
√
pn (x) = 2npn−1 (x) .
(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)
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
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 ,
where
√
Dn = 2n 1 + Bn−2/3 .
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
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 −∞
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
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)
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
C
≤ L
1 + m2 Dxn − 1
1
≤ Cn−2L/3 L
n−2/3 + axn − 1
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
and
we have
|Rn W | (x)p dx ≤ o n−1/6 + Cn−1/6 R −1 .
[−Dn ,Dn ]\I1
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 ⎭
Now let ε → 0+ :
M ∞
−p |P (ak )|p −2
(1 + η) ≤ 6π A |P (t)|p dt.
k=1
Ai (ak )2 −∞
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
−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|
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
∞
≤C |SN [f ] − SM [f ]|2 (t) dt
−∞
→ 0 as M, N → ∞,
∞ π 2 |P (ak )|p
M
|P (t)|p dt ≤ B . (74)
−∞ 6
j =1
Ai (ak )2
Let
M
Rn (x) = P (ak ) kn (x) W −1 (xkn ) .
k=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
. /−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
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
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
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]
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
{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
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
r b−1
mr π mr
V := cot ,
b b b
m=1
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
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
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
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
:= (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
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
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.
log(2π x) − γ
ψ(x) = − + O(log x), (x → 0) .
πx
and
χ2 (u, v) := −1 χ1 (u, v) dv .
0
and
−1
χ4 (u) := χ3 (u, v)dv .
0
:= q −(D+) .
Then
N(q, , ) > 0
+∞
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
and obtain
E(m, n, q) q 31/32 .
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
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 .
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
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 Introduction
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).
sin x
4 √ 2 − sin x2 1
2− 2 < < , (2)
π2 x2 4
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
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
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) ,
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),
3 Main Results
∞
(−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
|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
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
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]
and
∞
1
g2 (x) = x 4k+2 ,
24k+2 (4k + 4)!
k=0
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).
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
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
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.
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]
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)
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
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 ,
+∞
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
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
t t t π
θ (t) = ln − −
2 2π 2 8
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].
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
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
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.
2π
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
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
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 ,
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
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
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 .
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
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
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
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
−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
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
π i/4+2π iP 2 (2 ln P −1)
− 2 Re e Σ ,
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)
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
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.
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]
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
pπ
Sn (q; ξ ) = cotn and Sn∗ (q) = cotn ,
q q
p=0 p=1
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
pπ
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
kπ
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
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
kπ
(−1)k cos2m ,
2n + 2
k=1
π 2m
n−1
kπ
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.
∞
∞
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
H = {z : Im (z) > 0} .
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
Observe that
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)
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
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
∞
1 (−2π i)k k−1 2π inz
= n e
(z + m)k (k − 1)!
m∈Z 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
∞
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
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)
2 (2π i)k
G (z, k) = 2ζ (k) + A (z, k) .
(k − 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 [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
k−1
n nh
sp (h, k) = Bp ,
k k
n=0
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
B n (x) = Bn (x)
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
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
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
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)
p−1
π ma1 π ma2 π maj
d(p; a1 , a2 , . . . , aj ) = (−1) j/2
cot cot · · · cot .
p p p
m=1
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
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)
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
2π nφ k πφ
a sin = − cot ,
k 2 k
a mod k
p!
b−1 π na n
sp (a, b) = i cot ζ p, . (13)
(2π i)p b b
n=1
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
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
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
k−1
π(2m − 1)hj k π h(2m − 1)
j sin = − cot .
k 2 2k
j =1
k−1 πm
π mj
(−1)j sin = − tan .
k 2k
j =1
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
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
∞
1 π h(2n − 1)
s1 (h, k) = − cot ; (17)
π 2n − 1 2k
n=1
2n − 1 ≡ 0 (mod k)
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
2
∞
1 π h(2n − 1)
s5 (h, k) = tan . (21)
π 2n − 1 2k
n=1
2n − 1 ≡ 0 (mod k)
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
k
1 π h(2m − 1) π(2m − 1)
s1 (h, k) = − cot cot ;
2k 2k 2k
m=1
m = (k + 1)/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
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.
E n (x) = En (x)
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
∞
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
∞
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
∞
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).
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
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
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
∞
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
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
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
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
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 alternating zeta function (also called Dirichlet’s eta function η(s))
∞
(−1)n−1
Φ(−1, s, 1) = ζ ∗ (s) = ,
ns
n=1
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)
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
2(m!) −π ix
E m (x) = (−1) m+1
χ m+1 (e ) + χ m+1 (eπ ix
) .
(π i)m+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
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
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,
1 1
s5 (h, k) + s5 (k, h) = − . (38)
2 2hk
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
tan π hz tan π kz
Fm (z) =
z2m+1
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
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
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
and
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
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
kπ
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
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
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)
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
(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
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
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
π m 2
I3/2 (m) = − .
4m (m − 1)(m + 2) m/2
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
√ ⎧
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
⎧
⎪ 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
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
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
2π
2π 2n
f (x) dx = f (xk ) + R2n+1 [f ], (51)
0 2n + 1
k=0
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
and
* +
2n+1 (f ) = h f1/2 + f3/2 + · · · + f2n + f2n+1/2 ,
QM
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
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
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]
1 Introduction
∞
∞
p q
0< am < ∞, 0 < bn < ∞,
m=1 n=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
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
∞ 1 ∞ 1
p q
p(1− λ2 )−1 p q(1− λ2 )−1 q
< kλ |x| f (x)dx |y| g (y)dy , (4)
−∞ −∞
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.
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
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)
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
Since
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 ).
∞
2 csc2 β σ
−1
≥ . /γ uγ sech(u)du
γρ σ/γ ρ 1+cos β
(|x|+x cos α)δ
We deduce that
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
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
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
3 Main Results
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
⎧
⎨∞
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
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)
−∞
we derive (18).
On a Half-Discrete Hilbert-Type Inequality in the Whole Plane with the Kernel. . . 241
⎡ ⎤1
∞
q
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
|n|=1
⎡ ⎤1− 1
∞
q
|n|=1
∞ 1
p
< Kα,β (σ ) (|x| + x cos α) p(1+δσ )−1 p
f (x)dx ,
−∞
∞
($ (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
⎡ ⎤1
1 ∞
q
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
|n|=1
∞ 1− 1
p
J2 = (|x| + x cos α) p(1+δσ )−1 p
f (x)dx
−∞
⎡ ⎤1
∞
q
|n|=1
and
ε
! (σ − q )−1
bn := (|n| + n cos β) (|n| ∈ N).
244 M. T. Rassias et al.
|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
Φ(x) := (|x| + x cos α)p(1+δσ )−1 and Ψ (n) := (|n| + n cos β)q(1−σ )−1 ,
wherefrom
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
satisfying
||T (1) f ||p,Ψ 1−p = ||H (1) ||p,Ψ 1−p ≤ Kα,β ||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
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)
∞
H (2) (x) := h(x, n)bn (x ∈ R\{0}),
|n|=1
satisfying
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 ,
By the fact that the constant factor Kα,β (σ ) in (30) is the best possible, we have
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
(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
The constant factors in the above inequalities are the best possible.
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
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)
−∞
Then we get
On a Half-Discrete Hilbert-Type Inequality in the Whole Plane with the Kernel. . . 251
⎡ ⎤1
∞
p
|n|=1
∞ 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 ,
−∞
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
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)
−∞
and
ε
! (σ − q )−1
bn := (|n| + n cos β) (|n| ∈ N).
|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
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.
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
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
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 ,
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 ,
∞ 1
p
I≥ (|x| + x cos α) p(1+δσ )−1 p
f (x)dx J2 . (49)
−∞
and
ε
! (σ − q )−1
bn := (|n| + n cos β) (|n| ∈ N).
|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
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 ).
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
1 ˆ
f (x) = f (γ )ep (xγ ). (1)
p
γ ∈Fp
|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
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
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)
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
1 ; 2 ;
E+ (A, B) = |A(ξ )| |B(ξ )|2 . (12)
N
ξ
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
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
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| .
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
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
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
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
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
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
4 On the Quantity M+
;
− (S) = M− (S) = max(−1S (x)) ,
MG
x=0
and
;
+ (S) = M+ (S) = max 1S (x) .
MG
x=0
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
From this formula it follows that M− (S) > 0 because otherwise by inequality (26)
of Lemma 13, we obtain
; 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
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
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
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 .
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 .
1 Introduction
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]
π
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
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
where
n−1
Sn−1 (f ; x) = fˆ(k)eikx ;
k=−n+1
We set
ψ
En (Lβ,p )s = sup ρn (f ; ·)s , 1 ≤ p, s ≤ ∞, (4)
ψ
f ∈Lβ,p
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)
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
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∞ ).
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
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
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
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
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
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)
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
1
m
2m−1
k
Vm (t) := + cos kt + 2 1− cos kt, m ∈ N. (23)
2 2m
k=1 k=m+1
⊥ ∗ 1
e2n (fp,n )∞ ≥ I1 . (26)
3π
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)
π
∗ ∗
(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
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)
⊥ 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
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
∞
⊥ 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
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
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)|
ψ(t) η(t) − t 1
= → 0, as t → ∞ for ψ ∈ M+
∞.
|ψ (t)| t μ(t)
284 T. A. Stepanyuk
⊥ 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)
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
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),
3π
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
⊥ ⊥ 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
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
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
B. Yang ()
Department of Mathematics, Guangdong University of Education, Guangdong,
Guangzhou, P. R. China
e-mail: [email protected]; [email protected]
∞ ∞ ∞ 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
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
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 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
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
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
1
(k(σ − δ0 ) + k(σ ))
σ1 − σ
∞ ∞
≥ I2 = (cot h(xy) − 1)fn (x)gn (y)dxdy ≥ MJ2 = Mn. (15)
0 0
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
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
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
3 Main Results
∞ ∞ 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
(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
(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
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
∞ ∞ 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
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
1 1 1
k( ) = 21/q Γ ( )ζ ( )
p p p
∞ ∞ 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
∞ ∞ 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
(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
∞ ∞ 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
1 1 1
k( ) = 21/q Γ ( )ζ ( )
p p p
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.
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
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