Sie sind auf Seite 1von 21

Ramanujan J (2008) 17: 123–143

DOI 10.1007/s11139-007-9046-4

On the zeros of approximations of the Ramanujan


-function

Haseo Ki

Received: 29 March 2006 / Accepted: 23 July 2007 / Published online: 10 October 2007
© Springer Science+Business Media, LLC 2007

Abstract The analogue of the Riemann hypothesis for the Ramanujan zeta function
states that all zeros of the Ramanujan -function have real zeros only. We study the
zeros of approximations of the Ramanujan -function.

Keywords Ramanujan zeta function · Riemann hypothesis · Zeros of


approximations of the Ramanujan -function

Mathematics Subject Classification (2000) 11M26 · 30D10

1 Introduction

We let τ (n) be the Ramanujan τ -function which satisfies




y[(1 − y)(1 − y 2 ) · · ·]24 = τ (n)y n .
n=1

We consider the Ramanujan zeta function L(s) defined by



 τ (n)
L(s) = ,
ns
n=1

This work was supported by the Korea Research Foundation Grant funded by the Korean
Government (MOEHRD, Basic Research Promotion Fund) (KRF-2006-312-C00021).
H. Ki ()
Department of Mathematics, Yonsei University, Seoul 120-749, Korea
e-mail: haseo@yonsei.ac.kr
124 H. Ki

13
where the series and the product are absolutely convergent for Re(s) > 2 . Using the
fact that for any prime number p
τ (p n+1 ) = τ (p)τ (p n ) − p 11 τ (p n−1 ) (n = 1, 2, 3, . . .),
we have
  
−s 11−2s −1 13
L(s) = (1 − τ (p)p +p ) Re(s) > .
2
p: prime

The Ramanujan zeta function satisfies the following identity:


 ∞
R (is) = (2π)−s−6 L(s + 6)(s + 6) = x s+5 g(e−2πx )dx,
0

where


g(y) = τ (n)y n .
n=1
We call R (z) the Ramanujan -function. Using the functional equation
 1/2
x 6 g(e−2πx ) = g(e−2πx )g(e−2π/x ) ,
we have
 ∞
R (z) = φ(t)eizt dt,
−∞
where

  −t 12
φ(t) = e−2π cosh t (1 − e−2πke )(1 − e−2πke ) .
t

k=1
We note that a Riemann hypothesis for the Ramanujan zeta function asserts that all
zeros of R (z) are real.
In the sequel, we assume that F is a finite sequence of complex or real numbers
a0 , a1 , . . . , an such that at least one of them is nonzero. We define
 ∞
F (z) = φF (t)eizt dt,
−∞

where


n 
n
−2π cosh t −2πmet −2πme−t
φF (t) = e am e am e .
m=0 m=0
We immediately have
F (z) = F (z)
for any complex number z. One can see that for some sequences Fk , the function
Fk (z) converges uniformly to R (z) on any compact subset of the complex plane.
In this article, we study the zeros of F (z).
On the zeros of approximations of the Ramanujan -function 125

We define zero-counting functions of a given function. Let G(z) be a func-


tion. For G(z), we define N (T ; G), Nk (T ; G), Noff (T ; G) and N (α, β, a, b; G)
(k = 1, 2, . . . , α, β, a, b ∈ R):
N (T ; G) = #{z : G(z) = 0, 1 ≤ Re(z) ≤ T },
Nk (T ; G) = #{z : z is a zero of multiplicity k for G(z), 1 ≤ Re(z) ≤ T , Im(z) = 0},
Noff (T ; G) = #{z : G(z) = 0, 1 ≤ Re(z) ≤ T , Im(z) = 0},
N (α, β, a, b; G) = #{z : G(z) = 0, α ≤ Re(z) ≤ β, a ≤ Im(z) ≤ b}.
We have the following

Theorem 1 For F (z), we have


T T
N (T ; F ) = log + O(log T ),
π eπ
N (T ; F ) − N1 (T ; F ) = O(T ).
In the left half-plane, the same holds for zero-counting functions. In particular, almost
all zeros of F (z) are real and simple.

We note that we have the same results as in Theorem 1 for a more general function
 ∞ n
n

  −t
e−2a cosh t am e−bme am e−bme
t
eizt dt (a, b > 0).
−∞ m=0 m=0

We omit the proof for this general case, because the proof is similar to the proof of
Theorem 1. Besides, the function F (z) is the important one related to the Ramanu-
jan -function.
Concerning Theorem 1, we refer to the following result. Conrey and Ghosh [2]
proved that R (z) has at least AT 1/6−ε simple zeros in the region 0 < Re(z) < T ,
where the positive constant A depends on ε > 0.
Define R,n (z) by
 ∞ 
n
 −t 12
e−2π cosh t (1 − e−2πme )(1 − e−2πme ) eizt dt.
t
R,n (z) =
−∞ m=1

De Bruijn [1] pointed out that R,n (z) may have infinitely many nonreal zeros. In
fact he showed that the function
 ∞
t −t
f1 (z) = e−2a cosh t 1 − e−be 1 − e−be eizt dt (a, b > 0)
−∞

has infinitely many nonreal zeros. More precisely, from a certain integral value of |ν|
onwards, f1 (z) has just one zero in the neighborhood of any point ±i + 2νπ/ log(1 +
b/a). He also showed that apart from these, for any δ > 0, f1 (z) has but a finite
number of zeros for |Im(z)| > δ. For this, see [1, p. 225]. Thus Theorem 1 cannot be
improved in general. In this article, we complement and strengthen de Bruijn’s result
for f1 (z).
126 H. Ki

For F , we define ψF (s) by


n
ψF (s) = π −s am (2m + 1)−s .
m=0

Let k be such that

P (1) = P  (1) = · · · = P (k−1) (1) = 0, P (k) (1) = 0,


n
where P (y) = m=0 am y
m. Then we have the following

Theorem 2 Let ∗ and ∗∗ be positive. Suppose that ψF (iz − k) has finitely many
zeros in − ∗ < Im(z) < ∗∗ . Let δ be such that 0 < δ < ∗ . Then all but finitely
many zeros of F (z) which lie in |Im(z)| ≤ δ are real. In particular, all but finitely
many zeros of F (z) are real if ψF (iz − k) has finitely many zeros in Im(z) < ∗∗ .

We conjecture that the second case for ψF (iz − k) in Theorem 2 is not possible.
However, the first one can be possible, because for example, a −s −(a +b)−s for f1 (z)
fulfills this case. We want to introduce a way to approach the analogue of the Riemann
hypothesis for the Ramanujan zeta function. Let Rn be such that Rn (z) = R,n (z).
We note that Rn (z) converges uniformly to R (z) in any compact subset of the
complex plane. It is worth remarking that k = 12n for ψRn (iz − k). Therefore the
first case in Theorem 2 should be valid for ∗ > 1/2 in a stronger sense. Namely
we may expect that ψRn (iz − k) has a uniformly bounded number of zeros in − ∗ <
Im(z) < ∗∗ ( ∗ > 1/2 and ∗∗ > 0). Then R (z) has a uniformly bounded number
of nonreal zeros in −1/2 ≤ Im(z) ≤ 1/2. Using the Euler product formula for the
Ramanujan zeta function, we can show that there are no zeros in Im(z) > 1/2. Thus
we conclude that R (z) has a uniformly bounded number of nonreal zeros. Hence all
but finitely many zeros of the Ramanujan zeta function are real. Further, if Rn (z)
has no zeros in −1/2 ≤ Im(z) < 0 for sufficiently large n, the Riemann hypothesis
for the Ramanujan zeta function follows.
It is interesting to know the behavior of the zeros of f1 (z) near the real axis. Let δ
be such that 0 < δ < 1. Are all zeros of f1 (z) in |Im(z)| ≤ δ real and simple, except
for the finite number of nonreal zeros? Certainly, de Bruijn’s result [1, p. 225] cannot
say that in |Im(z)| ≤ δ, all but finitely many zeros of the function f1 (z) are real. Using
the same method as in the proof of Theorem 2 and the behavior of the argument
change of functions on the real axis, we answer the above question affirmatively.

Theorem 3 With at most finitely many exceptions, the function f1 (z) has either
real and simple zeros or it has just one zero in the neighborhood of any point
±i + 2νπ/ log(1 + b/a) from a certain integral value of |ν| onwards.

The paper is organized as follows. In Sect. 2 we deal with some facts about F (z).
In Sect. 3 we shall prove Theorem 1. The proof of Theorem 2 is given in Sect. 4.
We shall supply two proofs for the second part of Theorem 2. In Sect. 5 we show
Theorem 3. In Sect. 6 we introduce de Bruijn’s results [1]. We then discuss various
On the zeros of approximations of the Ramanujan -function 127

things about them. In particular, we give different proofs for the results by using these
methods that the author has developed.
Our method in the paper was invoked in de Bruijn’s paper [1] and the author’s
papers [4–7]. In particular, the author has been inspired by de Bruijn’s remarkable
work.

2 Preliminaries

Let F , k and ψF (s) be as in the introduction. We denote WF (z) by


 ∞
WF (z) = F (t)eizt dt,
φ
−∞

where

e−2π cosh t 
n 
n
F (t) = −2πmet −2πme−t
φ am e am e .
et/2 + e−t/2
m=0 m=0

Using the methods in de Bruijn [1, pp. 210–213, 224–225], we can prove the
following

Theorem 2.1 (1) For any  > 0,



  
F (z) = bm ψF (iz − m)(iz − m) + O (iz − k)z−2
m=k

holds uniformly in the half-plane Im(z) ≤ −.


(2) For any  > 0,


WF (z) = bm ψF (iz − m − 1/2)(iz − m − 1/2) + O (iz − k − 1/2)z−2
m=k

holds uniformly in the half-plane Im(z) ≤ −.


(3) For |y| ≤ and |x| ≥ 1,

F (x + iy)/ (i(x + iy) − k) = O(1)|x|μ(y) .

(4) For |y| ≤ and |x| ≥ 1,

WF (x + iy)/ (i(x + iy) − k − 1/2) = O(1)|x|μ(y) .

Here bk and μ(y) are such that


n ∞

−πe−t −2πme−t
e am e = bm e−mt (bk = 0),
m=0 m=k
128 H. Ki

the real number is sufficiently large and



2y if y > 1/2,
μ(y) = 1/2 + y if −1/2 ≤ y ≤ 1/2,
0 otherwise.

Proof For (1) and (2), we mimic the methods in [1, pp. 210–213 and pp. 224–225].
In particular, for (2), we should use
 ∞
F (2t)eizt dt,
φ
−∞

where
 ∞
WF (z) = 2 F (2t)ei2zt dt.
φ
−∞

Using (1), the identity F (z) = F (z) and Stirling’s formula, we can see that

F (x + iy)/ (i(x + iy) − k) = O(1)|x|2y

for ≥ y ≥ 1/2. From (1) we immediately obtain μ(y) = 0 for − < y < −1/2.
For the remaining case, we follow the Phragmén–Lindelöf theorem [3, p. 47]. With
the same method we can demonstrate (4). 

Concerning Theorem 2.1(1), we have a more precise formula as follows:



 ∞

F (z) = bm ψF (iz − m)(iz − m) + bm ψF (−iz − m)(−iz − m).
m=k m=k

For this, see [1, p. 225].

Lemma 2.2 We let f (z) be a holomorphic and almost-periodic function in the strip
a < Im(z) < b. Let δ > 0. Then we have N (x, x + 1, a + δ, b − δ; f ) = O(1).

For the proof of this lemma, we refer to [8, p. 267, Lemma 2].

Proposition 2.3 (1) For sufficiently large > 0, we have

F (z) = bk π −iz+k (iz − k) (1 + O(1/|z|))

uniformly in the half-plane Im(z) ≤ − . In particular, for sufficiently large > 0,


F (z) does not have any zeros in |Im(z)| ≥ .
(2) For sufficiently large > 0, we have

WF (z) = bk π −iz+k+1/2 (iz − k − 1/2) (1 + O(1/|z|))

uniformly in the half-plane Im(z) ≤ − . In particular, for sufficiently large > 0,


WF (z) does not have any zeros in |Im(z)| ≥ .
On the zeros of approximations of the Ramanujan -function 129

(3) Let T > 1. For F (z), we have


T T
N (T ; F ) = log + O(log T ).
π eπ
(4) Let T > 1. For WF (z), we have
T T
N (T ; WF ) = log + O(log T ).
π eπ
(5) For WF (z), there exist c > 0 and H > 0 such that for sufficiently large |x| > 0,
we have
N (x, x + H, −1/4, ; WF ) > c log |x|,
where satisfies (2).
(6) Let δ > 0. For WF (z), we have

N (x, x + 1, − , −δ; WF ) = O(1),

where satisfies (2) and the implied constant depends on δ.

Proof Parts (1) and (2) are immediate consequences of Theorem 2.1 (1), (2). Let CT
be the polygonal path [−i , −i + T , i + T , i ]. Using (1), the identity F (z) =
F (z), Theorem 2.1(3) and Stirling’s formula, we can see that for F (z), we have

1 F (z)
N(T ; F ) = dz
2πi CT F (z)
 −i +T   T
1 F (z) 1 F (z)
= dz + dz + O(1)
πi −i F (z) πi −i +T F (z)
1
= (−T log π + the argument change of (ix + − k) from 0 to T )
π
+ O(log T )
T T
= log + O(log T ).
π eπ
Similarly (4) follows. It suffices to show (5) for sufficiently large x > 0. Using The-
orem 2.1(2), we can see that

WF (z) = (iz − k − 1/2)(bk ψ(iz − k − 1/2) + O(|z|−1/2 )) (2.1)

uniformly in the half-plane Im(z) ≤ −1/4. From Lemma 2.2, we immediately obtain
(z) = ψ(iz − k − 1/2), there exists a κ1 > 0 such that
that for ψ
) ≤ κ1
N (x, x + 1, − , −1/8; ψ (2.2)

for any x. Applying Rouché’s theorem to (2.1), we immediately prove from (2.2) that
there exists a κ2 > κ1 such that for WF (z),

N (x, x + 1, − , −1/6; WF ) ≤ κ2 (2.3)


130 H. Ki

for any x. Using (2.3) and (4), (5) follows. Using Theorem 2.1(2), Lemma 2.2 and
Rouché’s theorem, (6) follows. 

We need the classical theorem of Hermite-Biehler. The following modified form


is due to de Bruijn.

Lemma 2.4 If U (z) and V (z) are real polynomials such that W (z) = U (z) + iV (z)
has n roots in the lower half-plane, then U (z) has at most n pairs of conjugate com-
plex roots.

Proof See [1, p. 215]. 

For the proof of Lemma 2.4, we appeal to the argument change of the function
W (z). Then Lemma 2.4 immediately follows. This means that Lemma 2.4 follows
from the argument change method.
This simple lemma is very useful for investigating the precise behavior of real
zeros of the real or imaginary part of an entire function which can be uniformly ap-
proximated by polynomials in any compact subset of the complex plane. Namely we
can exactly count the number of real zeros of the real or imaginary part of polynomial
approximations of the entire function with the uniformly bounded number of possible
nonreal zeros and then we show that all but finitely many zeros of the real or imag-
inary part of the entire function are real. Moreover, we can sometimes remove ‘but
finitely many’ so that we are able to demonstrate that all zeros of the real or imaginary
part of the function are real. Usually it is very hard to know the exact location of zeros
of the real or imaginary part of a given entire function by only using the argument
change of the entire function. However, the author has successfully used Lemma 2.4
in order to obtain the precise location of zeros of some functions (see e.g., [4] and
[7]). We note that there is a certain limitation to the application of Lemma 2.4 to its
application to a more general situation. We will discuss this in Remark 4.1.
In this article we shall use this lemma in giving another proof of the second part
of Theorem 2.

3 Proof of Theorem 1

For the proof of Theorem 1, we shall exactly follow the argument in [6].
It suffices to show the theorem in the right half-plane, using the same argument,
one can prove it in the left half-plane. We note that

F (z) = WF (z − i/2) + WF (z + i/2). (3.1)

Thus it is easy to see that for real number x, F (x) is zero when
π
arg WF (x − i/2) ≡ mod π.
2
On the zeros of approximations of the Ramanujan -function 131

For F (z), we define ND (T ; F ) by




ND (T ; F ) = Nk (T ; F ).
k=1

Then, we get
1
ND (T ; F ) ≥ · (the argument change of WF (x − i/2) from 1 to T ).
π
Clearly we obtain
1
(N (T ; F ) − N1 (T ; F )) ≤ N (T ; F ) − ND (T ; F ) ≤ N (T ; F ) − N1 (T ; F ),
2
because


N (T ; F ) − N1 (T ; F ) = Noff (T ; F ) + kNk (T ; F ) − N1 (T ; F )
k=1


= Noff (T ; F ) + kNk (T ; F )
k=2

and


N(T ; F ) − ND (T ; F ) = Noff (T ; F ) + (k − 1)Nk (T ; F ).
k=2
This immediately implies that for proving Theorem 1, it suffices to show that for
F (z),
N (T ; F ) − ND (T ; F ) = O(T ).
Now we follow the standard method developed in [9] or [6]. We precisely mimic the
method in [6]. For a function h, we define

h(a + ) = lim h(x) and h(a − ) = lim h(x).


x→a,x>a x→a,x<a

We set
WF (z − i/2)
f (z) = .
(iz − k − 1/2)
Let x1 , . . . , xM be the zeros of f (x) on 0 < x < T . Let

H0 = arg (ix1− − k − 1/2)f (x1− ) − arg (−k − 1/2)f (0),


Hj = arg (ixj−+1 − k − 1/2)f (xj−+1 ) − arg (ixj+ − k − 1/2)f (xj+ )

for 1 ≤ j ≤ M − 1 and
+
HM = arg (iT − k − 1/2)f (xM ).
132 H. Ki

Fix a satisfying Proposition 2.3(2). From Proposition 2.3(6), for WF (z), we recall

N (x, x + 1, − , −δ; WF ) = O(1),

for any δ > 0. Thus, we get

N (0, T , − , 0; f ) = O(T ) (3.2)

for f (z). We apply the principle of the argument to f (z). Then we have

T log T − T + O(log T )

− Hj − π(N(0, T , − , 0; f ) − N (0, T , − , 0+ ; f ))
0≤j ≤M

≤ 2πN(0, T , − , 0+ ; f )

or

Hj ≥ T log T − T + O(log T )
0≤j ≤M

− 2πN(0, T , − , 0+ ; f ) + π(N (0, T , − , 0; f )


− N (0, T , − , 0+ ; f )).

For this, we refer to the argument in [9, pp. 503–504]. Thus we obtain

T T
ND (T ; F ) ≥ log T − + O(log T ) − 2N (0, T , − , 0; f ).
π π
From this and (3.2), we get

T
ND (T ; F ) ≥ log T + O(T ). (3.3)
π

We also recall that for F (z), N (T ; F ) = T


π
T
log eπ + O(log T ). By this and (3.3),
we obtain
N (T ; F ) − ND (T ; F ) = O(T ).
Thus Theorem 1 follows.

4 Proof of Theorem 2

We shall follow the method in [5]. It suffices to show the theorem in the right half-
plane, using the same argument, one can prove it in the left half-plane.
We first recall that we have

F (z) = WF (z − i/2) + WF (z + i/2) .


On the zeros of approximations of the Ramanujan -function 133

We investigate the properties of WF (z). Using WF (z) = WF (z), by Hadamard’s prod-


uct theorem we have
∞  
z z
WF (z) = czm1 eαz 1− e αm +βm i ,
αm + βm i
m=1

where c, α are real numbers, m1 is a nonnegative integer, αm + βm i runs through all


zeros of WF (z) and

 1
< ∞. (4.1)
α2 + βm
2
m=1 m
Here
n 
 
z z
czm1 eαz 1− e αm +βm i
αm + βm i
m=1
converges uniformly to WF (z) in any compact subset of the complex plane. We also
note that from our assumption, we have zeros from the effect of ψF (iz − k − 1/2).
Namely Rouché’s theorem implies that from Theorem 2.1(2), for any arbitrar-
ily small  > 0, there may exist zeros of WF (z) in |Im(z)| ≤ 1/2 − ∗∗ +  or
|Im(z)| ≥ 1/2 + ∗ −  which are arbitrarily close to zeros of ψF (iz − k − 1/2),
if |Re(z)| → ∞. Using this fact with our assumption of ψF (iz − k), WF (z) = WF (z)
and Theorem 2.1(2), we can write

WF (z) = czm1 eαz 1 (z)2 (z)3 (z)4 (z), (4.2)

where
  z

z  (z − αm )2 + βm
2 2zαm
1 (z) = 1−
2 +β 2
αm
e αm 2 + β2
e m,
αm αm m
βm =0 0<βm ≤δ0

 (z − αm )2 + βm
2 2zαm
2 (z) =
2 2
e αm +βm ,
αm + βm
2 2
βm ≥ 0 +1/2

 (z − αm )2 + βm
2 2zαm
3 (z) =
2 +β 2
αm
2 + β2
e m
αm m
δ0 <βm <δ1

and
 (z − αm )2 + βm
2 2zαm
4 (z) =
2 2
e αm +βm
αm + βm
2 2
δ1 ≤βm < 0 +1/2


n
(z − am )2 + bm
2 2zam
=
2 +b2
am
2 + b2
e m
am m
m=1

for some n, a1 , a2 , . . . , an , b1 , b2 , . . . , bn with δ1 ≤ b1 , b2 , . . . , bn < 0 + 1/2.


Here we choose 0 such that δ < 0 < ∗ , we set δ1 = 1/2 − ∗∗ /2 and we let
134 H. Ki

δ0 = max(1/2 − 0 , min(δ/2, 1/4)). Note that 1/2 − 0 can be negative. Possibly,


3 (z) does not appear, if δ1 is less than δ0 . We also have 0 < δ0 < 1/2. In order to
estimate Q1 , we need this condition. From Proposition 2.3(6) and the symmetry of
the zeros of WF (z) to the real axis, it is not hard to see that the main source of zeros
of WF (z) are from 1 (z). Now we calculate
 
 WF (z + i/2) 
 
 W (z − i/2) 
F

for −δ < Im(z) < 0. We shall show that this quantity is strictly less than 1 in the
region −δ < Im(z) < 0 and Re(z) > x0 , where x0 is sufficiently large. In fact, the
zeros of 1 (z) play a major role for estimating the above quantity. On the other
hand, the zeros of 2 (z) and 4 (z) prevent the validity of the required inequality.
However, 1 (z) supplies a chunk of zeros in any given (x, x + H ) for an appropriate
H , so we overcome any difficulty from the zeros of 2 (z) and 4 (z). For x, y, β ∈ R,
we obtain
 
 (x − iy + i/2)2 + β 2 2
 = log (x + β − (y − 1/2) ) + 4x (y − 1/2)
2 2 2 2 2 2

log 
(x − iy − i/2)2 + β 2  (x 2 + β 2 − (y + 1/2)2 )2 + 4x 2 (y + 1/2)2
 
4y(x 2 − β 2 + y 2 + 1/4)
= log 1 − 2
(x + β 2 − (y + 1/2)2 )2 + 4x 2 (y + 1/2)2
4y(x 2 − β 2 + y 2 + 1/4)
≤− . (4.3)
(x 2 + β 2 − (y + 1/2)2 )2 + 4x 2 (y + 1/2)2
We also obtain
 (z+i/2)α 2
e 
log (z−i/2)α  = 0 (4.4)
e
for α ∈ R and z ∈ C. We let z = x − iy such that x > 0 and 0 < y ≤ δ. From (4.2),
(4.4) and the fact that
 
 z + i/2 

log   ≤ 0,
z − i/2 
we have
 
 WF (z + i/2) 2

log   ≤ Q1 + Q2 + Q3 + Q4 , (4.5)
W (z − i/2) 
F
where

Q1 = log |1 (z + i/2)/1 (z − i/2)|2 ,


Q2 = log |2 (z + i/2)/2 (z − i/2)|2 ,
Q3 = log |3 (z + i/2)/3 (z − i/2)|2

and
Q4 = log |4 (z + i/2)/4 (z − i/2)|2 .
On the zeros of approximations of the Ramanujan -function 135

We are now ready to estimate Q1 , Q2 , Q3 and Q4 . By (4.3), (4.4) and the fact that
δ1 < 1/2, we have

 4y((x − αm )2 − βm2 + y 2 + 1/4)


Q3 ≤ −
((x − αm )2 + βm
2 − (y + 1/2)2 )2 + 4x 2 (y + 1/2)2
δ0 <βm <δ1

 4y(−βm2 + 1/4)
≤−
((x − αm )2 + βm
2 − (y + 1/2)2 )2 + 4x 2 (y + 1/2)2
δ0 <βm <δ1

< 0. (4.6)

We calculate Q4 and Q2 . By (4.3) and (4.4), we have


n
4y((x − am )2 − bm
2 + y 2 + 1/4)
Q4 ≤ −
((x − am ) + bm − (y + 1/2)2 )2 + 4x 2 (y + 1/2)2
2 2
m=1
≤ 4ny. (4.7)

for sufficiently large x > 0. We choose 1 > 0 such that 1 > 0 satisfies Proposi-
tion 2.3(2). Then we note that

− 1 ≤ βm ≤ 1 (4.8)

for any positive integer m. Proposition 2.3(6) implies that for any positive number x,

1 = O(1). (4.9)
0 +1/2<βm ≤ 1 , x≤αm ≤x+1

Using (4.3), (4.4) and (4.8–4.9), we conclude that there exists an absolute constant
κ4 > 0 such that

 4y((x − αm )2 − βm2 + y 2 + 1/4)


Q2 ≤ −
((x − αm )2 + βm
2 − (y + 1/2)2 )2 + 4x 2 (y + 1/2)2
0 +1/2<βm ≤ 1

 4yβm2

((x − αm )2 + βm
2 − (y + 1/2)2 )2 + 4x 2 (y + 1/2)2
0 +1/2<βm ≤ 1

κ4  1
≤ y
4 k4
k=1
≤ κ4 y (4.10)

holds. For Q1 we need to understand the behavior of zeros of 1 (z) near the real
axis. There are many zeros of 1 (z) in a given interval which are applied to get
the required inequality. For this purpose, we choose H > 0 such that H satisfies
Proposition 2.3(5). Set H0 = H + 3 + δ. Proposition 2.3(5), (6) implies that there
136 H. Ki

exists an absolute constant κ5 > 0 such that



1 ≥ κ5 log x (4.11)
0≤βm ≤δ0 , x≤αm ≤x+H

for sufficiently large x > 0. Using (4.3), (4.4), (4.11) and the fact that δ0 < 1/2, it
follows that there exists an absolute constant κ6 > 0 such that for sufficiently large
x > 0,
 4y((x − αm )2 − βm2 + y 2 + 1/4)
Q1 ≤ −
((x − αm )2 + βm
2 − (y + 1/2)2 )2 + 4x 2 (y + 1/2)2
0<βm ≤δ0
 2y

(x − αm )2 + (y + 1/2)2
βm =0

 4y(−δ02 + 1/4)
≤−
0<βm ≤δ0 ,δ+2<αm −x≤H0
H04 + 4H02 (y + 1/2)2
 2y

H2
0<βm ≤δ0 ,δ+2<αm −x≤H0 0
+ (y + 1/2)2

≤ −κ6 y log x. (4.12)

From (4.5–4.7), (4.10) and (4.12), we see that


 
 WF (z + i/2) 
  −y(κ6 log x−4n−κ4 )/2
 W (z − i/2)  ≤ e
F
<1 (4.13)

for sufficiently large x > 0 and 0 < y < δ. Now suppose that

F (x − yi) = 0

for sufficiently large x > 0 and 0 < y < δ. Note that WF (x − yi − i/2) = 0. Then we
get
 
 WF (x − yi + i/2) 
 
 W (x − yi − i/2)  = 1.
F
By (4.13) this is a contradiction. The first part of Theorem 2 follows.
The second part of Theorem 2 is an immediate consequence of the first part and
Proposition 2.3(1).

Remark 4.1 In Theorem 2, the condition for ψF (iz − k) is more natural than the
similar condition for ψF (iz − k − 1/2). For example, see the precise formula of
F (z) after Theorem 2.1.

We give another proof for the second part of Theorem 2. We have two reasons for
supplying it. One is the simplicity of the method. Furthermore, we can observe that
On the zeros of approximations of the Ramanujan -function 137

from this proof, all but finitely many zeros of the function are simple. However, we
need an accurate argument change method for this, (see e.g., the proof of Theorem 3).
We shall use Lemma 2.4 for this alternative proof. We recall
∞ 
 
z z
WF (z) = czm1 eαz 1− e αm +βm i ,
αm + βm i
m=1

where c, α are real numbers, m is a nonnegative integer and βm ≥ 0 for each positive
integer m. We denote Vn (z) and Wn (z) by
n 
 
z − i/2 z−i/2
Vn (z) = c(z − i/2)m1 eα(z−i/2) 1− e αm +βm i
αm + βm i
m=1

and
 
α(z − i/2) n
Wn (z) = c(z − i/2)m1 1 +
n
n    n
z − i/2 z−i/2 z − i/2
× 1− e α m +βm i 1+ .
αm + βm i n(αm + βm i)
m=1

Recall that Vn (z) converges uniformly to WF (z − i/2) in any compact subset of the
complex plane.

Claim 4.1 The function Wn (z) converges uniformly to WF (z − i/2) in any compact
subset of the complex plane.

Proof We note that


 
z n z2
e−z 1 + = e n (1+O(|z|/n)) (4.14)
n
for sufficiently large n and |z| n. We write

|Vn (z) − Wn (z)| = |Vn (z)En (z)|,

where
  n  n
α(z − i/2) n  −(z−i/2) z − i/2
En (z) = 1 − e−α(z−i/2) 1 + e αm +βm i 1 + 2 + β2 )
.
n n(αm m
m=1

Thus it suffices to show that En (z) converges uniformly to 0 in any compact subset
of the complex plane. Using (4.14), we easily get

En (z) = 1 − egn (z) , (4.15)


138 H. Ki

where
α 2 (z − i/2)2
gn (z) = (1 + O(|z − i/2|/n))
n
(z − i/2)2 
n
1
+ (1 + O(|z − i/2|/n)).
n (αm + βm i)2
m=1

By (4.1) we have
∞ 
   ∞
 1  1
  ≤ < ∞.
 (αm + βm i) 
2 αm + βm
2 2
m=1 m=1

Thus, we have
gn (z) = O((z − i/2)2 /n) (4.16)
for |z| n. From (4.15–4.16), we obtain

En (z) = O(|z − i/2|2 /n)

for |z| n. Therefore En (z) converges uniformly to 0 in any compact subset of the
complex plane. Claim 4.1 follows. 

We note that there exists a positive integer n0 such that the number of zeros of
Wn (z) which are lying in the lower half-plane is uniformly bounded by n0 . Thus, by
Lemma 2.4, the function Wn (z) + Wn (z) has at most 2n0 many nonreal zeros for any
n. Since by Claim 4.1 the function Wn (z) + Wn (z) converges uniformly to F (z)
in any compact subset of the complex plane, F (z) has at most 2n0 many nonreal
zeros. The second part of Theorem 2 follows.

Remark 4.2 Lemma 2.4 provides another proof of the second part of Theorem 2,
but we cannot prove the first part of Theorem 2 by using Lemma 2.4. The reason is
that in the lower half-plane, we may have infinitely many nonreal zeros of WF (z). In
this case, Lemma 2.4 cannot be applied to prove the first part of Theorem 2. As an
example, this is the case for f1 (z). Furthermore we comment on a crucial weakness
of Lemma 2.4 which is the limitation of the argument change method. Suppose that
F (z) = W (z) + W (z), where W (z) has only finitely many zeros in the lower half-
plane and the order of W (z) is less than 2. In general, even with this strong condition
for W (z), we cannot use Lemma 2.4 in demonstrating that all but finitely many zeros
of F (z) are real. In order to overcome the limitation of Lemma 2.4 for the general
case, we need the method as in the proof of the first part of Theorem 2. The author
has successfully developed and used this method in [4] and [5].

5 Proof of Theorem 3

For f1 (z), we have that for |Re(z)| > 1, Im(z) < −δ ∗ and 0 < δ ∗ < 1/2,

f1 (z) = W (z − i/2) + W (z + i/2),


On the zeros of approximations of the Ramanujan -function 139

where

W (z) = bψ1 (iz − 3/2)(iz − 3/2) + O(|z|−2δ (iz − 3/2))
and
ψ1 (s) = a −s − (a + b)−s .
Then we follow the same method as in the proof of the first part of Theorem 2 and we
can demonstrate that for any 0 < δ < 1, all but finitely many zeros of f1 (z) which lie
in |Im(z)| ≤ δ are real. In order to show that all but finitely many zeros of f1 (z) are
simple, we calculate the argument change of W (x − i/2) along the real axis. Clearly
we have
W (x − i/2) = bψ1 (ix − 2)(ix − 2)(1 + o(1))
as x → ∞. Then we have

arg(W (x − i/2)) = Im(log(a + b)2−ix + log (ix − 2))

for sufficiently large x > 0. The derivative of this function is


d arg(W (x − i/2))
= − log(a + b) + Re(  (ix − 2)/ (ix − 2))
dx
= − log(a + b) + log x + O(1).

Thus the argument of W (x − i/2) strictly increases for sufficiently large x. Therefore
all but finitely many zeros of f1 (z) on the real line are simple. Thus all but finitely
many zeros of f1 (z) which lie in |Im(z)| ≤ δ are real and simple. For finishing the
proof of Theorem 3, we recall de Bruijn’s result. Namely, from a certain integral
value of |ν| onwards, f1 (z) has just one zero in the neighborhood of any point ±i +
2νπ/ log(1 + b/a) and apart from these, for any δ̃ > 0, f1 (z) has but a finite number
of zeros for |Im(z)| > δ̃. Since δ is arbitrary in (0, 1), using de Bruijn’s result and our
above conclusion, Theorem 3 immediately follows.

6 Remarks

It is interesting to note that de Bruijn [1] proved the following. We briefly introduce
de Bruijn’s results.
(1) Let f (t) be an integral function of t such that its derivative f  (t) is the limit
(uniformly in any bounded domain of the t-plane) of a sequence of polynomials,
all whose roots lie on the imaginary axis. Suppose furthermore that f (t) is not a
constant, and
∞that f (t) = f (−t), f (t) ≥ 0 for real values of t. Then all roots of
the integral −∞ e−f (t) eizt dt are real.
(2) Let N be a positive integer and put


N
P (t) = pn ent (Re(pn ) > 0; pn = p−n , n = 0, 1, 2, . . .).
−N
140 H. Ki

Let the function q(x) be regular in the sector −π/N − N −1 arg pN < arg x <
π/2N − N −1 arg pN and on its boundary, with the possible exception of x = 0
and x = ∞ which may be poles (of arbitrary finite order) for q(x). Furthermore
suppose
q(x) = q(1/x)
in this sector (in other words, q(x) is real for |x| = 1). Then all but a finite number
of the roots of the function
 ∞
(z) = e−P (t) Q(t)eizt dt (Q(t) = q(et ))
−∞

are real.
(3) The special functions
 ∞ 
N
e−λ cosh t cn ent eizt dt (λ > 0, cn = c−n )
−∞ n=−N

have N pairs of non-real zeros at most.


Statement (3) can be proved by (2) and Lemma 2.4 after some observations on the
Fourier transform of e−λ cosh t Nn=−N nc e nt . We note that (2) is the main theorem

in [1].
We are able to approximate the Riemann -function using the functions in (3),
where (z) is defined by
 ∞
(z) = φ(t)eizt dt
−∞
and


(2n4 π 2 e9t/2 − 3n2 πe5t/2 )e−n πe .
2 2t
φ(t) = 2
n=1

The Riemann hypothesis states that all zeros of (z) are real. For example, Pólya
[11] used the following approximations to the Riemann -function:
 ∞  ∞
1 (z) = φ1 (t)eizt dt and 2 (z) = φ2 (t)eizt dt,
−∞ −∞

where
φ1 (t) = 8π 2 cosh(9t/2)e−2π cosh(2t)
and

φ2 (t) = 8π 2 cosh(9t/2) + (8π 3 − 12π) cosh(5t/2) + 8π 2 cosh(t/2) e−2π cosh(2t) .

We note that φ(t) ∼ φ1 (t) and φ(t) = φ2 (t) + O(et/2 e−2π cosh(2t) ). In fact, Pólya [11]
showed that all zeros of 1 (z) and 2 (z) are real. Actually we can show that all zeros
On the zeros of approximations of the Ramanujan -function 141

of these functions are simple. For a general function see [6]. The discussion for (z)
and 1 (z) [13, p. 256] is interesting. As de Bruijn [1, p. 221] commented, it seems
very difficult to approach the Riemann hypothesis using approximations in (3).
We introduce some more results related to (3). Suppose 0 < A1 < · · · < AN and
c1 , . . . , cN are real. Let


N  ∞
F (z) = cn e−2An cosh t eizt dt
n=1 −∞

and

N
H (z) = cn A−iz
n .
n=1

Here we assume that cn = 0 for some n. Then we have the following.


(A) If lim inf|z|→∞ |H (z)| = 0, then for some δ > 0, all but finitely many zeros of
F (z) which lie in |Im(z)| ≤ δ are real.
(B) If lim inf|z|→∞ |H (z)| = 0 and for H (z), N (−∞, ∞, a1 , a2 ; H ) = O(1)
(a1 > 0), then for any δ1 and any δ2 with a1 < δ1 < δ2 < a2 , we have
N(−∞, ∞, δ1 , δ2 ; F ) = O(1).
(C) In any case, we have N (T ; F ) − N1 (T ; F ) = O(T ) for F (z).
In (A) the condition lim inf|z|→∞ |H (z)| = 0 cannot be substituted for the condi-
tion that H (z) has no zeros on the real axis. For this, see [10]. The proof of (B) is easy.
The proof of (A) essentially follows from the method of the proof of the first part of
Theorem 2. By using the method as in the proof of Theorem 1, we can demonstrate
(C). We omit the proofs of (A), (B) and (C). For further studies, see [6] and [12].
The Ramanujan -function can also be approximated by the functions in (3), be-
cause
t −t
1 − e−2πme 1 − e−2πme


= 4π 2 m2 e−2πm cosh t (1 − m2 k −2 )2 + 4m2 k −2 cosh2 t .
k=1

This shows that the functions in (3) are indeed interesting.


De Bruijn [1, pp. 199–200] supplied a nice method for the proof of (2) before he
proved the asymptotic expansion [1, Theorem 18]. Define U (z) by
 ∞
Q(t)
U (z) = e−P (t) t eizt dt.
−∞ e + 2 + e−t

De Bruijn first derives the asymptotic expansion of U (z). The resulting asymptotic
expansion says that the zeros of this function are arbitrarily close to the real axis. Then
he appeals to a nice property of et + 2 + e−t , namely that the function et + 2 + e−t
pushes the zeros of the Fourier transform of e−P (t) Q(t)/(et + 2 + e−t ) to the real
axis, so in this way, (2) follows.
142 H. Ki

We can avoid his method by demonstrating (2) via the methods of the proof of
Theorem 2 in this paper. We write

(z) = W (z − i/2) + W (z + i/2) and W (z) = U (z − i/2) + U (z + i/2).

Since the zeros of U (z) are arbitrarily close to the real axis, our method as in the
proof of Theorem 2 immediately implies that for each > 0, there exists an x0 > 0
such that
 
 U (x − iy + i/2) 
 
 U (x − iy − i/2)  < 1
for x > x0 and 0 < y < . Here U (z) has only finitely many zeros in Im(z) ≥ .
Thus all but finitely many zeros of W (z) are real. We repeat the argument with W (z)
and then (2) follows.
One more method for (2) comes from Lemma 2.4. We apply Hadamard’s product
theorem to U (z) and then we have

 (z − αm )2 + βm
2 2zαm
U (z) = czm1 eαz
2 2
e αm +βm ,
αm + βm
2 2
m=1

where c, α are real numbers, m1 is a nonnegative integer and βm ≥ 0 for each positive
integer m. Here the important thing is that βm → 0 as m → ∞. In order to show that
all but finitely many zeros of W (z) are real, we mimic the second proof of the second
part of Theorem 2 where we used Lemma 2.4. We repeat the same argument with
W (z) and (2) follows.
While we pursue these two arguments, we can observe that all but finitely many
zeros of the function in (2) are simple.
For the proof of (2), the author believes that de Bruijn’s method is the best among
the three methods. However, our two methods have wider applications related to the
Riemann zeta function and the Epstein zeta function; see [4, 5] and [7].
All but finitely many zeros of the functions in (2) are simple. This result immedi-
ately follows from the method in Sect. 5.

Acknowledgement I sincerely thank Professor Masahiko Taniguchi for his hospitality while I was visit-
ing the department of mathematics at Kyoto University via a JSPS Fellowship. I also wish to thank Andreas
O. Bender for valuable comments on this paper. I deeply thank the referee for his or her many valuable
suggestions and comments on the paper.

References

1. de Bruijn, N.G.: The roots of trigonometric integrals. Duke Math. J. 17, 197–226 (1950)
2. Conrey, J.B., Ghosh, A.: Simple zeros of the Ramanujan τ -Dirichlet series. Invent. Math. 94(2), 403–
419 (1988)
3. Erdélyi, A., et al.: Higher Transcendental Functions, vol. 1. McGraw-Hill, New York (1953)
4. Ki, H.: On a theorem of Levinson. J. Number Theory 107, 287–297 (2004)
5. Ki, H.: All but finitely many nontrivial zeros of the approximations of the Epstein zeta function are
simple and on the critical line. Proc. Lond. Math. Soc. 90, 321–344 (2005)
6. Ki, H.: On the distribution of zeros of linear combinations of K-Bessel functions and the Riemann
zeta function. Preprint
On the zeros of approximations of the Ramanujan -function 143

7. Ki, H.: Zeros of the constant term in the Chowla-Selberg formula. Acta Arith. 124, 197–204 (2006)
8. Levin, B.J.: Distribution of Zeros of Entire Functions. Transl. Math. Monographs, vol. 5. Am. Math.
Soc., Providence (1964)
9. Levinson, N.: On theorems of Berlowitz and Berndt. J. Number Theory 3, 502–504 (1971)
10. Moreno, C.J.: The zeros of exponential polynomials (I). Compos. Math. 26, 69–78 (1973)
11. Pólya, G.: Über trigonometrische Integrale mit nur reellen Nullstellen. J. Reine Angew. Math. 158,
6–18 (1927)
12. Strömberg, F.: On the zeros of linear combinations of K-Bessel functions. Uppsala universitet licentiat
thesis (2000-06-02)
13. Titchmarsh, E.C.: The Theory of the Riemann Zeta-function, 2d edn. Oxford University Press, Oxford
(1986), revised by D.R. Heath-Brown

Das könnte Ihnen auch gefallen