This action might not be possible to undo. Are you sure you want to continue?
Advances in Mathematics 226 (2011) 4198–4211
www.elsevier.com/locate/aim
A uniqueness theorem for Dirichlet series satisfying
a Riemann type functional equation
Bao Qin Li
Department of Mathematics, Florida International University, University Park, Miami, FL 33199, USA
Received 13 June 2009; accepted 2 December 2010
Available online 22 December 2010
Communicated by Takahiro Kawai
Abstract
We will prove a uniqueness theorem for Lfunctions in terms of the preimages of two values in the
complex plane.
© 2010 Elsevier Inc. All rights reserved.
MSC: 30D35; 11M06; 30D30; 11M41
Keywords: Lfunction; Dirichlet series; Meromorphic function; Selberg class
1. Introduction and the result
Lfunctions are Dirichlet series with the Riemann zeta function ζ(s) =
¸
∞
n=1
1
n
s
as the pro
totype, which are important objects in number theory and have been studied extensively (cf. the
recent monograph [7] and various references therein). Throughout the paper, an Lfunction al
ways means a Dirichlet series L(s) =
¸
∞
n=1
a(n)
n
s
of a complex variable s =σ +it , satisfying the
following axioms (see e.g. [7, p. 111]):
(i) Ramanujan hypothesis. a(n) n
ε
for every ε >0;
(ii) Analytic continuation. There is a nonnegative integer k such that (s − 1)
k
L(s) is an entire
function of ﬁnite order;
Email address: libaoqin@ﬁu.edu.
00018708/$ – see front matter © 2010 Elsevier Inc. All rights reserved.
doi:10.1016/j.aim.2010.12.001
B.Q. Li / Advances in Mathematics 226 (2011) 4198–4211 4199
(iii) Functional equation. L satisﬁes a functional equation of type Λ
L
(s) =ωΛ
L
(1 − ¯ s), where
Λ
L
(s) =L(s)Q
s
¸
K
j=1
Γ (λ
j
s +μ
j
) with positive real numbers Q, λ
j
, and complex num
bers μ
j
, ω with Re μ
j
0 and ω = 1.
The degree d
L
of an Lfunction L is deﬁned to be d
L
= 2
¸
K
j=1
λ
j
, where K, λ
j
are the
numbers in the axiom (iii).
Note that the above does not assume the Euler product hypothesis: a(1) = 1, and logL(s) =
¸
∞
n=1
b(n)
n
s
, where b(n) = 0 unless n is a positive integer power of a prime and b(n) n
θ
for
some θ <
1
2
. An Lfunction satisfying (i)–(iii) and also the Euler product hypothesis is called
an Lfunction in the Selberg class S (see e.g. [6,7]), which includes the Riemann zetafunction
ζ and essentially those Dirichlet series where one might expect a Riemann hypothesis. At the
same time, there are a whole host of interesting Dirichlet series not possessing Euler product (cf.
[7]). The class S
of Lfunctions satisfying only (ii) and (iii) was introduced in [4], where an
Lfunction satisfying (ii) and (iii) was shown to already satisfy (i) when the degree is between 0
and 1, and the structure of such Lfunctions was given. In the present paper, we will show how an
Lfunctions L satisfying (i)–(iii) are uniquely determined by the zeros of L −c for two distinct
complex numbers c. The result obtained particularly applies to the Selberg class.
By the analytic continuation axiom, an Lfunction can be analytically continued as a mero
morphic function in the complex plane C. The zero set of a meromorphic function f , or more
generally, the zero set f
−1
(c) := {s ∈ C: f (s) = c} of f −c for a complex value c, i.e., the set
of the preimage of c under f or the cvalues of f , is one of the main objects in the value dis
tribution theory of meromorphic functions. It is a famous theorem of Nevanlinna, often referred
to as Nevanlinna’s uniqueness or unicity theorem, that two nonconstant meromorphic functions
f, g in C must be identically equal if f
−1
(c
j
) = g
−1
(c
j
), i.e., if f − c
j
and g − c
j
have the
same zeros (ignoring multiplicities) for ﬁve distinct values c
j
∈ C∪ {∞} (see e.g. [3] or [7]). We
refer the reader to the monograph [7] for a detailed discussion on the topic and related works. In
particular, two Lfunctions in the Selberg class must be identically equal if they have the same
zeros with counting multiplicities, which follows from a result in [5] on difference of multiplic
ities of zeros of two Lfunctions, see also [2] for a related result. In fact, it was shown in [7,
p. 152] that two Lfunctions (not necessarily in the Selberg class) with a(1) = 1 must be identi
cally equal if L
1
−c and L
2
−c have the same zeros with counting multiplicities for a complex
number c. When ignoring multiplicities, the problem becomes more subtle. The simple example
ζ and ζ
2
, which have the same zeros (ignoring multiplicities), shows that the above result is no
longer true when ignoring multiplicities; and the following theorem was proved by Steuding (see
[7, p. 152]):
Theorem A. If two Lfunctions L
1
and L
2
satisfy the same functional equation with a(1) = 1
and L
−1
1
(c
j
) =L
−1
2
(c
j
) for two distinct complex numbers c
1
and c
2
such that
liminf
T →∞
˜
N
c
1
L
j
(T ) +
˜
N
c
2
L
j
(T )
N
c
1
L
j
(T ) +N
c
2
L
j
(T )
>
1
2
+ (1.1)
for some positive with either j = 1 or j = 2, then L
1
≡L
2
.
4200 B.Q. Li / Advances in Mathematics 226 (2011) 4198–4211
On the above, N
c
L
(T ) denotes the number of the zeros of L(σ +it ) −c in the rectangle 0
σ 1, t  T (counting multiplicities), and
˜
N
c
L
(T ) the number of the same zeros but ignoring
multiplicities.
Condition (1.1) reﬂects that more than 50% of the c
1
and c
2
values of L
j
are supposed to be
distinct. It was noted by Steuding that such a condition is very difﬁcult to verify; for instance,
it is known that more than 63% of the zeros of the Riemann zetafunction are distinct; however,
any extension to Lfunctions of large degree seems to be hard to realize, as pointed out in [7,
p. 152].
The purpose of this paper is to completely remove this condition (1.1). That is, we prove the
following
Theorem 1. If two Lfunctions L
1
and L
2
satisfy the same functional equation with a(1) = 1 and
L
−1
1
(c
j
) =L
−1
2
(c
j
) for two distinct complex numbers c
1
and c
2
, then L
1
≡L
2
.
Unlike the proof of Theorem A in [7], we will employ Nevanlinna theory combined with
other analytic tools in our proof, with a careful analysis on the growth and distribution of zeros
of the involved functions. Since Lfunctions are meromorphic functions and Nevanlinna theory is
known as an important tool in studying meromorphic functions, it would be proﬁtable to explore
its further applications to the theory of Lfunctions.
2. Proof of Theorem 1
For convenience of the reader who might not be familiar with Nevanlinna theory, we list here
the notations and results from Nevanlinna theory, which will be used in the proof (see e.g. [3]).
Let f be a meromorphic function in C. Then the Nevanlinna characteristic T (r, f ) is deﬁned as
T (r, f ) =m(r, f ) +N(r, f ),
where
m(r, f ) =
1
2π
2π
0
log
+
f
re
iθ
dθ; log
+
x = max
0, logx
,
and
N(r, f ) =
r
0
n(t, f ) −n(0, f )
t
dt +n(0, f ) logr,
where n(t, f ) denotes the number of poles of f (counting multiplicities) in s < t. Recall the
following known result (see e.g. [3, pp. 5, 18, 40]).
(i) The arithmetic properties of T (r, f ) and m(r, f ):
T (r, fg) T (r, f ) +T (r, g), T (r, f +g) T (r, f ) +T (r, g) +O(1).
The same inequalities hold for m(r, f ).
B.Q. Li / Advances in Mathematics 226 (2011) 4198–4211 4201
(ii) logmax
s=r
{f (s)}
R+r
R−r
T (R, f ) for R >r >0, if f is entire.
(iii) The Nevanlinna ﬁrst fundamental theorem: T (r, f ) =T (r,
1
f
) +O(1).
(iv) The logarithmic derivative lemma: m(r,
f
f
) = O(logr), if the order ρ(f ) :=
limsup
r→∞
logT (r,f )
logr
of f is ﬁnite.
Proof. We ﬁrst look at the simple case that one of L
1
and L
2
, say L
1
, is constant. Then L
1
≡ 1 by
the assumption that a(1) = 1. Since L
2
−c
j
and L
1
−c
j
have the same zeros by the assumption,
it is easy to see that L
2
≡ 1 (when c
1
or c
2
is 1), or L
2
= c
1
, c
2
in C (when c
1
, c
2
= 1). In
the latter case, noting that an Lfunction has at most one pole, L
2
must be constant and thus
L
2
≡ 1 since a(1) = 1, by the classic Picard theorem (see e.g. [1, p. 186]) that a nonconstant
meromorphic function in C assumes each value in C ∪ {∞} inﬁnitely many times with at most
two exceptions (or by the “less known” fact that a nonconstant Lfunction assumes each complex
number, cf. the Riemann–von Mangoldt formula mentioned below). Therefore, L
1
≡L
2
(≡ 1).
We thus assume, in the following, that L
1
and L
2
are nonconstant. Suppose, to the contrary,
that L
1
≡ L
2
. Our aim below is to derive a contradiction. To this end, consider the following
auxiliary function
F(s) =
(s −1)
q
L
1
L
2
(L
1
−L
2
)
2
(L
1
−c
1
)(L
1
−c
2
)(L
2
−c
1
)(L
2
−c
2
)
, (2.1)
where q is an integer chosen so that the function F does not have a pole or zero at s = 1.
Clearly, F is not identically zero by the above assumptions. The proof below is to show that F
is identically zero, which serves our purpose.
We claim that F is an entire function. In fact, the functions L
1
and L
2
both have only one
possible pole at s = 1, which cannot be a pole of F due to the use of the factor (s − 1)
q
. Thus,
possible poles of F may only come from the zeros of L
1
− c
j
or L
2
− c
j
, j = 1, 2. If w is a
zero of L
1
− c
j
of order m 1 and thus a zero of L
2
− c
j
of order n 1 since L
1
− c
j
and
L
2
− c
j
have the same zeros (m might be different from n), it is then a zero of (L
1
− L
2
)
2
of multiplicity at least two. Note also that w is a zero of L
1
of order m − 1 and a zero of L
2
of order n − 1. Thus, the numerator in (2.1) vanishes at w with a multiplicity at least equal to
(m − 1) + (n − 1) + 2 = m + n, which is the order of the zero w of the denominator in (2.1).
Hence, w is not a pole of F. This shows that F does not have any poles and thus is an entire
function in C.
We next control the number of zeros of F in the disc s < r by estimating the counting
function N(r,
1
F
). We deduce from (2.1) that
N
r,
1
L
1
−c
1
+N
r,
1
L
1
−c
2
+N
r,
1
L
2
−c
1
+N
r,
1
L
2
−c
2
=N
r,
1
(L
1
−c
1
)(L
1
−c
2
)(L
2
−c
1
)(L
2
−c
2
)
=N
r,
F
(s −1)
q
L
1
L
2
(L
1
−L
2
)
2
. (2.2)
4202 B.Q. Li / Advances in Mathematics 226 (2011) 4198–4211
Since F is entire, a possible pole of
F
(s−1)
q
L
1
L
2
(L
1
−L
2
)
2
may only come from a zero of (s −
1)
q
L
1
L
2
(L
1
−L
2
)
2
. Recall that F does not have a zero at s = 1, the only possible pole of L
1
and L
2
. Thus, if F(w) = 0 at a point w, then the numerator (s − 1)
q
L
1
L
2
(L
1
−L
2
)
2
in (2.1)
must vanish also at w with an equal or higher order. With this information, we deduce that
N
r,
F
(s −1)
q
L
1
L
2
(L
1
−L
2
)
2
N
r,
1
(s −1)
q
L
1
L
2
(L
1
−L
2
)
2
−N
r,
1
F
N
r,
1
(s −1)
q
+N
r,
1
L
1
+N
r,
1
L
2
+2N
r,
1
L
1
−L
2
−N
r,
1
F
=N
r,
1
L
1
+N
r,
1
L
2
+2N
r,
1
L
1
−L
2
−N
r,
1
F
+O(logr), (2.3)
in view of the fact that N(r,
1
(s−1)
q
) = O(logr). Also, by the ﬁrst fundamental theorem, we
deduce that
N
r,
1
L
1
=T
r,
1
L
1
−m
r,
1
L
1
=m
r, L
1
+N
r, L
1
+O(1) −m
r,
1
L
1
=m
r,
L
1
L
1
L
1
+N
r, L
1
+O(1) −m
r,
1
L
1
m
r,
L
1
L
1
+m(r, L
1
) +N
r, L
1
+O(1) −m
r,
1
L
1
.
Note that N(r, L
1
) =O(logr) since L
1
has at most one pole s = 1, and m(r,
L
1
L
1
) =O(logr) by
the logarithmic derivative lemma. We obtain that
N
r,
1
L
1
m(r, L
1
) −m
r,
1
L
1
+O(logr).
We then use the ﬁrst fundamental theorem and logarithmic derivative lemma again to deduce that
N
r,
1
L
1
T (r, L
1
) −m
r,
1
L
1
+O(logr)
T (r, L
1
−c
j
) −m
r,
1
L
1
+O(logr)
=T
r,
1
L
1
−c
j
−m
r,
1
L
1
+O(logr)
B.Q. Li / Advances in Mathematics 226 (2011) 4198–4211 4203
=N
r,
1
L
1
−c
j
+m
r,
1
L
1
−c
j
−m
r,
1
L
1
+O(logr)
=N
r,
1
L
1
−c
j
+m
r,
L
1
L
1
−c
j
1
L
1
−m
r,
1
L
1
+O(logr)
N
r,
1
L
1
−c
j
+m
r,
L
1
L
1
−c
j
+m
r,
1
L
1
−m
r,
1
L
1
+O(logr)
=N
r,
1
L
1
−c
j
+O(logr).
In the exactly same way, we have that
N
r,
1
L
2
N
r,
1
L
2
−c
j
+O(logr).
These last two inequalities together with (2.3) and (2.2) yield that
N
r,
1
L
1
−c
1
+N
r,
1
L
1
−c
2
+N
r,
1
L
2
−c
1
+N
r,
1
L
2
−c
2
N
r,
1
L
1
−c
2
+N
r,
1
L
2
−c
2
+2N
r,
1
L
1
−L
2
−N
r,
1
F
+O(logr)
or
N
r,
1
F
2N
r,
1
L
1
−L
2
−N
r,
1
L
1
−c
1
−N
r,
1
L
2
−c
1
+O(logr)
=
r
0
¸
2n
t,
1
L
1
−L
2
−n
t,
1
L
1
−c
1
−n
t,
1
L
2
−c
1
−2n
0,
1
L
1
−L
2
+n
0,
1
L
1
−c
1
+n
0,
1
L
2
−c
1
¸
dt
t
+O(logr) (2.4)
by the deﬁnition of the counting function N(r, ·). We then turn to estimate the unintegrated
counting functions n(t, ·) on the righthand side of (2.4). By the assumption of the theorem, L
1
and L
2
satisfy the same functional equation and thus have the same degree. Also, L
1
−L
2
clearly
satisﬁes the axioms (i)–(ii), and also (iii) with the same functional equation and thus the same
degree. For convenience, for an Lfunction L and a complex number c we denote by n
−
(t,
1
L−c
)
the number of zeros (counting multiplicities) of L − c within s < t and on the left halfplane
{σ 0}. It is known that the zeros of L−c on the left halfplane {σ 0} have bounded imaginary
parts and the number of these zeros having real part in [−t, 0] is
1
2
d
L
t +O(1) as t → +∞, where
d
L
is the degree of L (see [7, p. 145]). Thus,
n
−
t,
1
L−c
=
1
2
d
L
t +O(1).
4204 B.Q. Li / Advances in Mathematics 226 (2011) 4198–4211
We then have that
2n
−
t,
1
L
1
−L
2
−n
−
t,
1
L
1
−c
1
−n
−
t,
1
L
2
−c
1
=O(1) (2.5)
as t → ∞. On the other hand, by the Riemann–von Mangoldt formula for Lfunctions (see
[7, pp. 145 and 147]), the number of zeros (counting multiplicities) of L(s) − c in the region
Re s >0,  Ims T , denoted by N
c
L
(T ), is given by
N
c
L
(T ) =
d
L
π
T log
T
e
+
T
π
log
λQ
2
+O(logT ),
where λ =
¸
K
j=1
λ
2λ
j
j
, and λ
j
, K, Q are the numbers in the axiom (iii). (This was proved for
c = 1. We may assume c
1
= 1. Otherwise, replace c
1
by c
2
in (2.4).) Also, notice that L − c
does not have any zeros when σ = Re s is large, say σ α > 0, which follows easily from the
Dirichlet series of L. [In fact, this is clear if c = 1, since L(s) = 1 +O(2
−σ
) → 1 as σ → +∞.
If c = 1, then L−c = L−1
C
1
n
σ
1
>0 for an integer n
1
2 and a constant C
1
>0, cf. (2.12)
below.] Thus, if n
+
(t,
1
L−c
) denotes the number of zeros of L−c within s <t but on the right
halfplane {σ >0}, then
n
+
t,
1
L−c
N
c
L
t
2
−α
2
for large t , which implies that
n
+
t,
1
L
1
−c
1
+n
+
t,
1
L
2
−c
1
2
d
π
t
2
−α
2
log
√
t
2
−α
2
e
+
√
t
2
−α
2
π
log
λQ
2
+O
log
t
2
−α
2
, (2.6)
where d =d
L
1
=d
L
2
is the degree of L
1
and L
2
. Recalling that L
1
−L
2
has the same degree d,
we have that
2n
+
t,
1
L
1
−L
2
2N
0
L
1
−L
2
(t )
= 2
d
π
t log
t
e
+
t
π
log
λQ
2
+O(logt )
. (2.7)
It then follows from (2.6) and (2.7) that
2n
+
t,
1
L
1
−L
2
−n
+
t,
1
L
1
−c
1
−n
+
t,
1
L
2
−c
1
2
d
π
t log
t
e
+
t
π
log
λQ
2
+O(logt )
−2
d
π
t
2
−α
2
log
√
t
2
−α
2
e
+
√
t
2
−α
2
π
log
λQ
2
+O
log
t
2
−α
2
B.Q. Li / Advances in Mathematics 226 (2011) 4198–4211 4205
= 2
d
π
t log
t
e
−2
d
π
t
2
−α
2
log
√
t
2
−α
2
e
+O(logt )
= 2
d
π
t logt −2
d
π
t
2
−α
2
log
t
2
−α
2
+O(logt )
=
2d
π
t logt −
2d
π
t
1 −
α
2
t
2
logt +log
1 −
α
2
t
2
+O(logt )
=
2d
π
t logt −
2d
π
t
1 −
O(1)
t
2
logt +
O(1)
t
2
+O(logt )
=O(logt )
as t → ∞. Combining this with (2.5) yields that
2n
t,
1
L
1
−L
2
−n
t,
1
L
1
−c
1
−n
t,
1
L
2
−c
1
= 2n
+
t,
1
L
1
−L
2
−n
+
t,
1
L
1
−c
1
−n
+
t,
1
L
2
−c
1
+2n
−
t,
1
L
1
−L
2
−n
−
t,
1
L
1
−c
1
−n
−
t,
1
L
2
−c
1
=O(logt ) Clogt
for all t r
0
, where C, r
0
are two positive numbers. We then obtain, from (2.4), the following
control on the zeros of F:
N
r,
1
F
r
r
0
¸
2n
t,
1
L
1
−L
2
−n
t,
1
L
1
−c
1
−n
t,
1
L
2
−c
1
−2n
0,
1
L
1
−L
2
+n
0,
1
L
1
−c
1
+n
0,
1
L
2
−c
1
¸
dt
t
+O(1) +O(logr)
r
r
0
Clogt
t
dt +O(logr) =O
log
2
r
. (2.8)
We next give a tight estimate on the modulus of F, which is needed later in the proof, using the
above estimate (2.8). In order to do this, suppose that the nonzero zeros of F are a
1
, a
2
, a
3
, . . . ,
arranged in the order of increasing moduli and repeated according to multiplicities, and that
s = 0 is a zero of F of order l 0. Then the function
F(s)
s
l
does not vanish at s = 0 and its zeros
are exactly a
1
, a
2
, a
3
, . . . . Then it holds (for any nonzero meromorphic function F, see e.g. [3,
p. 25]) that
∞
¸
k=1
a
k

−1
∞
0
N(t,
s
l
F
)
t
2
dt.
4206 B.Q. Li / Advances in Mathematics 226 (2011) 4198–4211
Note, by (2.8), that
N
t,
s
l
F
N
t, s
l
+N
t,
1
F
O(logt ) +O
log
2
t
=O
log
2
t
. (2.9)
Thus,
¸
∞
k=1
a
k

−1
is convergent. This convergence grantees (see [3, p. 27]) that the canonical
product P(s) :=
¸
∞
k=1
(1−
s
a
n
) is an entire function of order ρ(P) equal to limsup
r→∞
logN(r,
s
l
F
)
logr
,
which is zero by (2.9). Thus, by the deﬁnition of order, T (r, P) =O(r
) for any 0 < <1, which
implies, by the inequality (ii) mentioned in the beginning of Section 2 with R = 2r and r = s,
that
log
P(s)
3T
2s, P
=O
s
. (2.10)
It is known (see [7, p. 150]) that for an Lfunction L,
T (r, L) =
d
L
π
r logr +O(r).
Also,
T
r, L
=m
r, L
+N
r, L
=m
r,
L
L
L
+O(logr)
m
r,
L
L
+m(r, L) +O(logr) T (r, L) +O(logr)
by the logarithmic derivative lemma. Hence, by the arithmetic properties of the characteristic
function and the ﬁrst fundamental theorem, we deduce, in view of (2.1), that
T (r, F) T
r, (s −1)
q
+T
r, L
1
+T
r, L
2
+4T (r, L
1
) +4T (r, L
2
) +O(1)
10
d
π
r logr +O(r).
This clearly implies, by the deﬁnition of order, that F is of order at most 1. We then have, by the
classic Hadamard factorization theorem (see e.g. [1, p. 384]), that F(s) = s
l
P(s)e
As+B
, where
A, B are two complex numbers. We thus obtained, in view of (2.10), the following estimate on
the modulus of F:
F(s) =
s
l
P(s)e
As+B
=O
e
αs
(2.11)
for some α >0 and all large s. (It is worth mentioning here that a rougher estimate on F may
be obtained directly using the above estimate on T (r, F), which is however not good enough for
our use below.)
B.Q. Li / Advances in Mathematics 226 (2011) 4198–4211 4207
Our next step is to examine the end behavior of F(s) as σ → +∞ and σ → −∞, where and
in the sequel s =σ +it . Since L
1
(s) =
¸
∞
n=1
a(n)
n
s
with a(1) = 1, we clearly have that
C
1
n
σ
1
L
1
−1
C
2
n
σ
1
, and L
1
=O
1
n
σ
1
, as σ → +∞, (2.12)
for some positive constants C
1
, C
2
, where n
1
is the smallest integer n 2 such that a(n) = 0.
Similarly, there is an integer n
2
2 such that
C
3
n
σ
2
L
2
−1
C
4
n
σ
2
, and L
2
=O
1
n
σ
2
, as σ → +∞, (2.13)
for some C
3
, C
4
>0. We also have that
L
1
−L
2
=O
1
2
σ
. (2.14)
It is then clear from (2.1) that if c
1
, c
2
= 1,
F(s) =
(s −1)
q
O
1
n
σ
1
O
1
n
σ
2
(O(
1
2
σ
))
2
(1 −c
2
)
2
(1 −c
2
)
2
as σ → +∞ for any ﬁxed t . If one of c
1
, c
2
, say c
1
, is 1, then by (2.12) and (2.13), we have that
F(s)
=
(s −1)
q
O
1
n
σ
1
O
1
n
σ
2
(O(
1
2
σ
))
2
C
1
n
σ
1
C
3
n
σ
2
(1 −c
2
)(1 −c
2
)
as σ → +∞. Thus, in any case, we always have that
F(s)
=O
1
4
σ
=O
1
e
σ
(2.15)
as σ → +∞ (for any ﬁxed t ).
We will prove that (2.15) holds also as σ → −∞. However, in this case, the Dirichlet series
does not give us (2.12) and (2.13) any more. We turn to use the following result for a nonconstant
Lfunction L: For any σ and large t , say t t
0
>1,
0
λQ
2
1
2
−σ
t
(
1
2
−σ)d
L
L(1 −σ +it )
L(σ +it )
C
0
λQ
2
1
2
−σ
t
(
1
2
−σ)d
L
L(1 −σ +it )
for two positive constants
0
, C
0
independent of σ, t , where λ =
¸
K
j=1
λ
2λ
j
j
, and λ
j
, Q, K are
the numbers in the axiom (iii) (see Lemma 6.7 in [7, p. 125]). (Note that Lemma 6.7 in [7,
p. 125] was stated for L functions in the Selberg class, but the proof does not use the Euler
product hypothesis.) As σ → −∞, we have that 1 −σ → +∞. Hence, L(1 −σ +it ) → 1 (cf.
4208 B.Q. Li / Advances in Mathematics 226 (2011) 4198–4211
(2.12) or (2.13)) and then
1
2
L(1 − σ + it ) 2 as σ → −∞, which implies that for t t
0
and as σ → −∞,
λQ
2
1
2
−σ
t
(
1
2
−σ)d
L
L(σ +it )
C
λQ
2
1
2
−σ
t
(
1
2
−σ)d
L
(2.16)
for two positive constants , C. By the Cauchy formula and (2.16), we have for t t
0
and σ →
−∞,
L
(σ +it )
=
1
2πi
w−(σ+it )=1
L(w)
w −(σ +it )
dw
C
1
λQ
2
3
2
−σ
(t +1)
(3/2−σ)d
L
, (2.17)
noting that when w − (σ + it ) = 1, σ − 1 < Re w < σ + 1 and t − 1 < Imw < t + 1, where
C
1
=C if λQ
2
>1 and C
1
=(λQ
2
)
−2
C if λQ
2
<1. The inequalities (2.16) and (2.17) hold for
L
1
and L
2
with d
L
=d. Applying the above stated result to L
1
−L
2
, we also have that for t t
0
and as σ → −∞,
(L
1
−L
2
)(s)
C
0
λQ
2
1
2
−σ
t
(
1
2
−σ)d
L
(L
1
−L
2
)(1 −σ +it )
C
0
λQ
2
1
2
−σ
t
(
1
2
−σ)d
O
1
2
1−σ
. (2.18)
If d = 0, then for ﬁxed t t
0
>1 and as σ → −∞,
λQ
2
1
2
−σ
t
(1/2−σ)d
→ +∞.
This also holds when d = 0 and Q > 1. (When d = 0, the functional equation in the axiom (iii)
does not contain Gamma factors and λ above is 1, cf. (2.20) below.) Thus, in these two cases, we
have, by (2.1), (2.18), (2.16), and (2.17), that for ﬁxed large t t
0
and σ → −∞,
F(s)
s −1
q
{C
1
(λQ
2
)
3
2
−σ
(t +1)
(3/2−σ)d
}
2
{C
0
(λQ
2
)
1
2
−σ
t
(
1
2
−σ)d
O(
1
2
1−σ
)}
2
{(λQ
2
)
1
2
−σ
t
(1/2−σ)d
− c
1
}
2
{(λQ
2
)
1
2
−σ
t
(1/2−σ)d
− c
2
}
2
A
1
s −1
q
(t +1)
2d
1 +
1
t
(1−2σ)d
O
1
4
1−σ
A
1
s −1
q
(t +1)
2d
1 +2(1 −2σ)d
1
t
O
1
4
1−σ
=O
1
e
−σ
=O
1
e
σ
(2.19)
for some constant A
1
> 0, in view of the simple inequality that (1 + x)
m
1 + 2mx for any
positive m and small positive x. Thus (2.15) holds for large t t
0
and as σ → −∞, when d = 0
and when d = 0, Q>1.
B.Q. Li / Advances in Mathematics 226 (2011) 4198–4211 4209
Consider now the cases d = 0, Q1. The ﬁrst inequality in (2.19) may not hold any more.
But, when d = 0, the functional equation in the axiom (iii) does not have any Gamma factors and
becomes
L(s) =ωQ
1−2s
L(1 − ¯ s). (2.20)
When σ → −∞, 1 −σ → +∞. Thus for such σ, L(1 − ¯ s) =L(1 −σ +it ) has Dirichlet series
representation with a(1) = 1. Hence, L is of the form
L(s) =ωQ
1−2s
1 +
a(2)
2
1−s
+
a(3)
3
1−s
+ · · ·
. (2.21)
If Q = 1, then lim
σ→−∞
L(s) = ω for L =L
1
, L
2
, and L
1
−L
2
= O(
1
2
1−σ
). Taking derivative
in (2.21), L
(s) =O(
1
2
1−σ
) for L =L
1
, L
2
. Thus, we deduce by (2.1) that
F(s)
s −1
q
(O(
1
2
1−σ
))
2
(O(
1
2
1−σ
))
2
(ω −c
1
)
2
(ω −c
2
)
2

=O
1
e
−σ
as σ → −∞, provided that c
1
, c
2
=ω. If one of c
1
, c
2
is ω, say c
1
=ω, then by (2.21) and as in
(2.12),
L
1
−c
1
 = L
1
−ω
A
2
n
1−σ
1
, L
1
=O
1
2
1−σ
as σ → −∞ for some A
2
>0, where n
1
2 is the smallest integer n such that a(n) = 0 in (2.21)
with L =L
1
. Similarly,
L
2
−c
1

A
3
n
1−σ
2
, L
2
=O
1
2
1−σ
for some A
3
>0 and an integer n
2
2; and
(L
1
−L
2
)(s) =O
1
2
1−σ
as σ → −∞. Thus,
F(s)
s −1
q
O
1
n
1−σ
1
2
O
1
n
1−σ
2
2
(O(
1
2
1−σ
))
2
A
2
n
1−σ
1
A
3
n
1−σ
2
(ω −c
2
)
2
=O
1
e
σ
.
That is, (15) holds as σ → −∞, when d = 0 and Q= 1.
If d = 0, Q<1, then by (2.21), lim
σ→−∞
L(s) = 0 for L =L
1
, L
2
, and L
1
−L
2
=O(
1
2
1−σ
).
Taking derivative in (2.21), L
(s) =O(Q
1−2σ
). We then have that
4210 B.Q. Li / Advances in Mathematics 226 (2011) 4198–4211
F(s)
s −1
q
(O(Q
1−2σ
))
2
(O(
1
2
1−σ
))
2
c
2
1
c
2
2

=O
1
e
σ
as σ → −∞ provided that c
1
c
2
= 0. If one of c
1
, c
2
is 0, say c
1
= 0, then by (2.21),
L
1
−c
1
=L
1
=ωQ
1−2s
1 +O
1
2
1−σ
for L =L
1
, L
2
. We then have that
F(s)
s −1
q
(O(Q
1−2σ
))
2
(O(
1
2
1−σ
))
2
(ωQ
1−2s
(1 +O(
1
2
1−σ
)))
2
c
2
2

=O
1
e
σ
as σ → −∞. Hence, (2.15) holds also in this case.
We have thus showed that (2.15) always holds for ﬁxed large t t
0
and both σ → +∞ and
σ → −∞. Fix such a large t =t
1
. We have that
F(σ +it
1
)
=O
1
e
σ
(2.22)
as σ → −∞ and as σ → +∞. We are now ready to ﬁnish up the proof by appealing to the
following Carlson theorem (see [8, p. 185]): Let f be holomorphic and of the form O(e
ks
)
for Re(s) > 0 (k is a real number) and let f (s) = O(e
−as
) for a constant a > 0 on the imagi
nary axis. Then f (s) = 0 identically. To apply this result to our situation, we make the variable
transformation:
z =x +iy =is +t
1
=t
1
−t +iσ,
or s =
z−t
1
i
. Set
G(z) =F(s) =F
z −t
1
i
.
Then by (2.11), we have
G(z)
=
F
z −t
1
i
=O
e
α
z−t
1
i

=O
e
αz
for all z. Also, on the imaginary axis of the zplane, x = 0 and z =iy, which correspond to t =t
1
and y =σ, or s =σ +it
1
. Thus we have, by (2.22), that when z =iy,
G(z)
=
F(σ +it
1
)
=O
1
e
σ
=O
1
e
z
.
Therefore, the function G satisﬁes the conditions of the above result and thus G is identically
zero, from which it follows that F is identically zero. This ﬁnally completes the proof of the
theorem. 2
B.Q. Li / Advances in Mathematics 226 (2011) 4198–4211 4211
References
[1] C.A. Berenstein, R. Gay, Complex Variables, SpringerVerlag, New York, 1991.
[2] E. Bombieri, A. Perelli, Distinct zeros of Lfunctions, Acta Arith. 83 (1998) 271–281.
[3] W.K. Hayman, Meromorphic Functions, Oxford Univ. Press, Oxford, 1964.
[4] J. Kaczorowski, A. Perelli, On the structure of the Selberg class. I. 0 d 1, Acta Math. 182 (2) (1999) 207–241.
[5] M.R. Murty, V.K. Murty, Strong multiplicity one for Selberg’s class, C. R. Acad. Sci. Paris Ser. I Math. 319 (1994)
315–320.
[6] A. Selberg, Old and new conjectures and results about a class of Dirichlet series, in: E. Bombieri, et al. (Eds.),
Proceedings of the Amalﬁ Conference on Analytic Number Theory, Maiori, 1989, Univ. Salerno, Salerno, 1992,
pp. 367–385; in: Collected Papers, vol. II, SpringerVerlag, 1991, pp. 47–63.
[7] J. Steuding, Value Distribution of LFunctions, Lecture Notes in Math., vol. 1877, SpringerVerlag, Berlin, 2007.
[8] E.C. Titchmarsh, The Theory of Functions, second edition, Oxford Univ. Press, 1968, reprinted.
152]): Theorem A. The result obtained particularly applies to the Selberg class.Q. At the same time. We refer the reader to the monograph [7] for a detailed discussion on the topic and related works. [6. where b(n) = 0 unless n is a positive integer power of a prime and b(n) 1 some θ < 2 . where ¯ ΛL (s) = L(s)Qs K=1 Γ (λj s + μj ) with positive real numbers Q. [7]). is one of the main objects in the value distribution theory of meromorphic functions. which includes the Riemann zetafunction ζ and essentially those Dirichlet series where one might expect a Riemann hypothesis. p. The zero set of a meromorphic function f . By the analytic continuation axiom.g. and log L(s) = ∞ b(n) nθ for n=1 ns . and the following theorem was proved by Steuding (see [7. L satisﬁes a functional equation of type ΛL (s) = ωΛL (1 − s ). The simple example ζ and ζ 2 . the set of the preimage of c under f or the cvalues of f . ω with Re μj 0 and ω = 1. λj are the j numbers in the axiom (iii).g. which follows from a result in [5] on difference of multiplicities of zeros of two Lfunctions. an Lfunction can be analytically continued as a meromorphic function in the complex plane C. if f − cj and g − cj have the same zeros (ignoring multiplicities) for ﬁve distinct values cj ∈ C ∪ {∞} (see e. If two Lfunctions L1 and L2 satisfy the same functional equation with a(1) = 1 and L−1 (cj ) = L−1 (cj ) for two distinct complex numbers c1 and c2 such that 1 2 ˜c ˜c NL1j (T ) + NL2j (T ) c c NL1j (T ) + NL2j (T ) lim inf T →∞ > 1 + 2 (1. which have the same zeros (ignoring multiplicities).. It is a famous theorem of Nevanlinna. two Lfunctions in the Selberg class must be identically equal if they have the same zeros with counting multiplicities. the problem becomes more subtle.B. .. we will show how an Lfunctions L satisfying (i)–(iii) are uniquely determined by the zeros of L − c for two distinct complex numbers c. g in C must be identically equal if f −1 (cj ) = g −1 (cj ). shows that the above result is no longer true when ignoring multiplicities. 152] that two Lfunctions (not necessarily in the Selberg class) with a(1) = 1 must be identically equal if L1 − c and L2 − c have the same zeros with counting multiplicities for a complex number c. Li / Advances in Mathematics 226 (2011) 4198–4211 4199 (iii) Functional equation. where K. λj .e. Note that the above does not assume the Euler product hypothesis: a(1) = 1. An Lfunction satisfying (i)–(iii) and also the Euler product hypothesis is called an Lfunction in the Selberg class S (see e. then L1 ≡ L2 .1) for some positive with either j = 1 or j = 2. In the present paper. it was shown in [7. and the structure of such Lfunctions was given. [3] or [7]). i. p. In particular. see also [2] for a related result. there are a whole host of interesting Dirichlet series not possessing Euler product (cf. often referred to as Nevanlinna’s uniqueness or unicity theorem. i. or more generally. When ignoring multiplicities. The degree dL of an Lfunction L is deﬁned to be dL = 2 K=1 λj .e.7]). the zero set f −1 (c) := {s ∈ C: f (s) = c} of f − c for a complex value c. where an Lfunction satisfying (ii) and (iii) was shown to already satisfy (i) when the degree is between 0 and 1. In fact. that two nonconstant meromorphic functions f. The class S of Lfunctions satisfying only (ii) and (iii) was introduced in [4]. and complex numj bers μj .
and NL (T ) the number of the same zeros but ignoring multiplicities. t T (counting multiplicities). [3]). it is known that more than 63% of the zeros of the Riemann zetafunction are distinct. f ) + N(r. f ) = 2π 0 log+ f reiθ dθ . NL (T ) denotes the number of the zeros of L(σ + it) − c in the rectangle 0 ˜c σ 1. Recall the following known result (see e. Proof of Theorem 1 For convenience of the reader who might not be familiar with Nevanlinna theory. (i) The arithmetic properties of T (r. 152]. It was noted by Steuding that such a condition is very difﬁcult to verify. p. 40]). .1) reﬂects that more than 50% of the c1 and c2 values of Lj are supposed to be distinct. The purpose of this paper is to completely remove this condition (1. Then the Nevanlinna characteristic T (r. f + g) T (r. f g) T (r. Since Lfunctions are meromorphic functions and Nevanlinna theory is known as an important tool in studying meromorphic functions. we will employ Nevanlinna theory combined with other analytic tools in our proof. where 2π 1 m(r. log x . Li / Advances in Mathematics 226 (2011) 4198–4211 c On the above. then L1 ≡ L2 .Q. Condition (1. Let f be a meromorphic function in C. The same inequalities hold for m(r. f ) = 0 n(t. f ) and m(r. f ). If two Lfunctions L1 and L2 satisfy the same functional equation with a(1) = 1 and L−1 (cj ) = L−1 (cj ) for two distinct complex numbers c1 and c2 . T (r. we list here the notations and results from Nevanlinna theory. it would be proﬁtable to explore its further applications to the theory of Lfunctions.g. f ) is deﬁned as T (r. however. for instance.g. f ) log r. and r N (r. 18. f ) + T (r. f ) − n(0. we prove the following Theorem 1. f ): T (r. [3. 1 2 Unlike the proof of Theorem A in [7]. log+ x = max 0. t where n(t. as pointed out in [7. That is. g) + O(1). which will be used in the proof (see e. with a careful analysis on the growth and distribution of zeros of the involved functions. f ) dt + n(0. 5. g). any extension to Lfunctions of large degree seems to be hard to realize. 2.1). f ).4200 B. f ) denotes the number of poles of f (counting multiplicities) in s < t. pp. f ) + T (r. f ) = m(r.
Clearly. In fact. L2 must be constant and thus L2 ≡ 1 since a(1) = 1. f ) for R > r > 0. if f is entire. the numerator in (2. L1 ≡ L2 (≡ 1). + N r. the functions L1 and L2 both have only one possible pole at s = 1. which is the order of the zero w of the denominator in (2. If w is a zero of L1 − cj of order m 1 and thus a zero of L2 − cj of order n 1 since L1 − cj and L2 − cj have the same zeros (m might be different from n). We deduce from (2. We thus assume. to the contrary. it is easy to see that L2 ≡ 1 (when c1 or c2 is 1). Note also that w is a zero of L1 of order m − 1 and a zero of L2 of order n − 1. + N r. F ). j = 1.1) where q is an integer chosen so that the function F does not have a pole or zero at s = 1. Since L2 − cj and L1 − cj have the same zeros by the assumption. This shows that F does not have any poles and thus is an entire function in C.B. (s − 1)q L1 L2 (L1 − L2 )2 (2. c2 = 1). [1. The proof below is to show that F is identically zero. f ) = O(log r). + N r. = N r. Thus. in the following. Proof. that L1 and L2 are nonconstant. Suppose.1) that 1 L 1 − c1 1 L 1 − c2 1 L 2 − c1 1 L 2 − c2 N r. (L1 − c1 )(L1 − c2 )(L2 − c1 )(L2 − c2 ) F (s) = (2. the Riemann–von Mangoldt formula mentioned below). Thus. is constant. w is not a pole of F . We next control the number of zeros of F in the disc s < r by estimating the counting 1 function N (r. Li / Advances in Mathematics 226 (2011) 4198–4211 4201 (ii) log maxs=r {f (s)} R+r T (R. Hence. 186]) that a nonconstant meromorphic function in C assumes each value in C ∪ {∞} inﬁnitely many times with at most two exceptions (or by the “less known” fact that a nonconstant Lfunction assumes each complex number.1). f ) = T (r. Then L1 ≡ 1 by the assumption that a(1) = 1. say L1 . Therefore.g. F is not identically zero by the above assumptions. cf. which cannot be a pole of F due to the use of the factor (s − 1)q . c2 in C (when c1 . consider the following auxiliary function (s − 1)q L1 L2 (L1 − L2 )2 .f ) log r of f is ﬁnite. noting that an Lfunction has at most one pole. Our aim below is to derive a contradiction. 1 (L1 − c1 )(L1 − c2 )(L2 − c1 )(L2 − c2 ) F . We ﬁrst look at the simple case that one of L1 and L2 . f ) + O(1). possible poles of F may only come from the zeros of L1 − cj or L2 − cj . R−r 1 (iii) The Nevanlinna ﬁrst fundamental theorem: T (r. (iv) The logarithmic derivative lemma: m(r. or L2 = c1 . if the order ρ(f ) := f lim supr→∞ log T (r. which serves our purpose. We claim that F is an entire function. that L1 ≡ L2 . it is then a zero of (L1 − L2 )2 of multiplicity at least two. = N r. In the latter case.2) . p.1) vanishes at w with a multiplicity at least equal to (m − 1) + (n − 1) + 2 = m + n.Q. by the classic Picard theorem (see e. 2. To this end.
With this information. 1 L1 1 L1 = m r. if F (w) = 0 at a point w. then the numerator (s − 1)q L1 L2 (L1 − L2 )2 in (2. L1 L Note that N(r. we deduce that N r. L1 − cj ) − m r. L1 ) = O(log r) by 1 the logarithmic derivative lemma. by the ﬁrst fundamental theorem. a possible pole of may only come from a zero of (s − 1)q L1 L2 (L1 − L2 )2 . We obtain that N r. (s−1)q ) = O(log r). 1 L 1 − cj − m r. 1 L1 − L2 + O(log r). 1 L1 − L2 1 F − N r. and m(r. 1 (s − 1)q L1 L2 (L1 − L2 )2 1 (s − 1)q 1 L1 + N r.1) must vanish also at w with an equal or higher order. L1 ) = O(log r) since L1 has at most one pole s = 1. L1 + O(1) − m r. 1 L1 m(r. 1 in view of the fact that N (r. L1 + O(1) − m r. N r. 1 L1 − m r. 1 F (2.3) (s − 1)q L N r. 1 L2 1 L1 + N r. L1 L1 L1 L1 + m(r. 1 L1 + O(log r). 1 L1 = T r. = N r. = m r. Thus. the only possible pole of L1 and L2 . 1 . Recall that F does not have a zero at s = 1. . Also. 1 L1 + O(log r) 1 L1 1 L1 + O(log r) + O(log r) T (r. + N r. we deduce that N r. We then use the ﬁrst fundamental theorem and logarithmic derivative lemma again to deduce that N r.4202 B. = T r. 1 L2 1 F + 2N r. F 2 1 L2 (L1 − L2 ) − N r. L1 + O(1) − m r. L1 ) + N r. m r. L1 1 L1 + N r. L1 ) − m r.Q. + 2N r. L1 ) − m r. 1 L1 T (r. − N r. L1 + N r. Li / Advances in Mathematics 226 (2011) 4198–4211 F (s−1)q L1 L2 (L1 −L2 )2 Since F is entire.
L1 − L2 clearly satisﬁes the axioms (i)–(ii). r 1 L1 − L2 − N r. − 2n 0. = N r. p. We then turn to estimate the unintegrated counting functions n(t. 1 L1 + O(log r) 1 L1 + O(log r) 1 L1 + O(log r) L1 1 L 1 − cj L 1 L1 L 1 − cj − m r. By the assumption of the theorem. n− t. 1 L1 + m r. 1 L 1 − c1 1 L 1 − c1 1 L 1 − c1 − N r. 1 L 2 − c1 1 L 2 − c1 1 L 2 − c1 + O(log r) = 0 2n t. + m r. ·). 1 L2 N r. − m r.4) + O(log r) by the deﬁnition of the counting function N (r. or N r. 1 L 1 − cj 1 L 1 − cj 1 L 1 − cj 1 L 1 − cj + m r. 0] is 1 dL t + O(1) as t → +∞. + n 0.2) yield that N r. 1 F 1 L 1 − c2 + N r. 1 L 1 − cj − m r. Also. ·) on the righthand side of (2. 1 L 2 − c2 1 F + O(log r) N r. − n t. 1 L 2 − c1 + N r.B. 1 L1 − L2 − n t. where 2 dL is the degree of L (see [7. + m r. 1 L 1 − c2 + N r. 1 L−c 1 = dL t + O(1). 1 L 2 − cj + O(log r).3) and (2. 145]). + n 0. 2 .Q. we have that N r. = N r. For convenience. 1 L 1 − c1 + N r. 1 L1 − L2 dt t (2. These last two inequalities together with (2. Thus. 1 L 2 − c2 + 2N r. 1 L1 − L2 − N r. L1 ) −c the number of zeros (counting multiplicities) of L − c within s < t and on the left halfplane {σ 0}. N r. and also (iii) with the same functional equation and thus the same degree. Li / Advances in Mathematics 226 (2011) 4198–4211 4203 = N r. 2N r.4). In the exactly same way. + O(log r). L1 and L2 satisfy the same functional equation and thus have the same degree. It is known that the zeros of L −c on the left halfplane {σ 0} have bounded imaginary parts and the number of these zeros having real part in [−t. for an Lfunction L and a complex number c we denote by n− (t.
1 L1 − L2 0 2NL1 −L2 (t) =2 It then follows from (2. since L(s) = 1 + O(2−σ ) → 1 as σ → +∞. (This was proved for j c = 1.Q. 1 L 1 − c1 t2 + n+ t. Recalling that L1 − L2 has the same degree d. then L − c = L − 1 Cσ > 0 for an integer n1 2 and a constant C1 > 0. We may assume c1 = 1. t t d t log + log λQ2 + O(log t) . 1 L 1 − c1 − n− t.  Im s T . 1 If c = 1. is given by c NL (T ) = 2λ T T dL T log + log λQ2 + O(log T ).7) that 2n+ t.12) n below. the number of zeros (counting multiplicities) of L(s) − c in the region c Re s > 0. 1 L1 − L2 − n− t. On the other hand. cf. L1 ) denotes the number of zeros of L − c within s < t but on the right −c halfplane {σ > 0}.4). Otherwise. this is clear if c = 1. Li / Advances in Mathematics 226 (2011) 4198–4211 We then have that 2n− t. for large t. K. (2. replace c1 by c2 in (2. which implies that n+ t. denoted by NL (T ).) Also. by the Riemann–von Mangoldt formula for Lfunctions (see [7. 1 L 2 − c1 d t t t log + log λQ2 + O(log t) π e π √ √ d 2 t 2 − α2 t 2 − α2 2 log + log λQ2 + O log t 2 − α 2 t −α −2 π e π . π e π where λ = K=1 λj j . π e π (2. 145 and 147]). (2. pp.7) 1 L 1 − c1 − n+ t. Q are the numbers in the axiom (iii).] Thus. 1 L 2 − c1 = O(1) (2. notice that L − c does not have any zeros when σ = Re s is large. say σ α > 0.5) as t → ∞. we have that 2n+ t. which follows easily from the Dirichlet series of L.4204 B. 2 1 L1 − L2 − n+ t. [In fact.6) and (2. and λj .6) where d = dL1 = dL2 is the degree of L1 and L2 . √ − α 2 log 1 L 2 − c1 t 2 − α2 + e 1 L−c c NL 1 t 2 − α2 d 2 π √ t 2 − α2 log λQ2 + O log t 2 − α 2 π . if n+ (t. then n+ t.
where C. + n 0. Combining this with (2. Then the function Fs(s) does not vanish at s = 0 and its zeros l are exactly a1 . which is needed later in the proof. F r0 2n t. from (2. 1 L 1 − c1 1 L 1 − c1 − n t.Q. 1 L 2 − c1 1 L 2 − c1 = 2n+ t. a2 . 1 L1 − L2 1 L1 − L2 C log t − n+ t. 1 L1 − L2 − n t. for all t r0 . using the above estimate (2.5) yields that 2n t. In order to do this. . . t2 l . . 1 L 2 − c1 1 L 2 − c1 dt + O(1) + O(log r) t (2. . 1 L 1 − c1 1 L 1 − c1 1 L 2 − c1 − n+ t.8) − 2n 0. − n− t. . . see e. p. r 1 L1 − L2 + n 0. 1 L 1 − c1 − n t. 1 L1 − L2 − n t. a3 . suppose that the nonzero zeros of F are a1 . a3 . arranged in the order of increasing moduli and repeated according to multiplicities.g. the following control on the zeros of F : r 1 N r. Then it holds (for any nonzero meromorphic function F . r0 are two positive numbers. . [3. s ) F dt. . C log t dt + O(log r) = O log2 r . a2 . + 2n− t. We then obtain. and that s = 0 is a zero of F of order l 0.4).B. t r0 We next give a tight estimate on the modulus of F . = O(log t) − n− t.8). Li / Advances in Mathematics 226 (2011) 4198–4211 4205 √ t d 2 t 2 − α2 d + O(log t) = 2 t log − 2 t − α 2 log π e π e d d 2 = 2 t log t − 2 t − α 2 log t 2 − α 2 + O(log t) π π = = 2d α2 α2 2d t log t − t 1 − 2 log t + log 1 − 2 π π t t 2d 2d O(1) t log t − t 1− 2 π π t log t + + O(log t) O(1) + O(log t) t2 = O(log t) as t → ∞. 25]) that ∞ k=1 ∞ −1 0 ak  N (t.
by the inequality (ii) mentioned in the beginning of Section 2 with R = 2r and r = s.10). 150]) that for an Lfunction L. s ) ∞ s F . 1 F (2.Q. This convergence grantees (see [3. L) + O(log r) dL r log r + O(r).g. by the deﬁnition of order.4206 B. which is however not good enough for our use below. F ).9).9) O(log t) + O log2 t = O log2 t . L = m r. m r. P = O s . that F (s) = s l P (s)eAs+B . Thus. 384]). p. (2. T (r. B are two complex numbers. π This clearly implies. p.) . k=1 (1− an ) is an entire function of order ρ(P ) equal to lim supr→∞ log r l product P (s) := which is zero by (2. Hence. We thus obtained. (s − 1)q + T r.1). L = m r.8). L2 + 4T (r. (It is worth mentioning here that a rougher estimate on F  may be obtained directly using the above estimate on T (r. in view of (2. s l + N t. L + N r. which implies. by the classic Hadamard factorization theorem (see e. L2 ) + O(1) d 10 r log r + O(r). L1 + T r. that F is of order at most 1. ∞ −1 k=1 ak  is convergent. Thus. π by the logarithmic derivative lemma. T r. L L + O(log r) L L L + m(r. by (2. p. T (r. sl F N t. in view of (2. 27]) that the canonical log N (r. [1. by the deﬁnition of order.10) It is known (see [7. L) = Also. by the arithmetic properties of the characteristic function and the ﬁrst fundamental theorem. we deduce. the following estimate on the modulus of F : F (s) = s l P (s)eAs+B = O eαs (2. that T (r. F ) T r. that log P (s) 3T 2s.11) for some α > 0 and all large s. L1 ) + 4T (r. where A. that N t. L) + O(log r) T (r. We then have. Li / Advances in Mathematics 226 (2011) 4198–4211 Note. P ) = O(r ) for any 0 < < 1.
0 λQ2 1 2 −σ t ( 2 −σ )dL L(1 − σ + it) 1 L(σ + it) C0 λQ2 1 2 −σ t ( 2 −σ )dL L(1 − σ + it) 2λ 1 for two positive constants 0 .12) and (2. then by (2.14) O 1 nσ 2 (O( 21 ))2 σ (1 − c2 )2 (1 − c2 )2 as σ → +∞ for any ﬁxed t. Thus.12) and (2. K are j the numbers in the axiom (iii) (see Lemma 6. say t t0 > 1. but the proof does not use the Euler product hypothesis. in any case. the Dirichlet series does not give us (2. say c1 .7 in [7. If one of c1 .13) for some C3 . However. C2 . where λ = K=1 λj j . t. nσ 1 and L1 = O 1 . 125]). We will prove that (2. 125] was stated for L functions in the Selberg class.B. where n1 is the smallest integer n Similarly. is 1. (2. Li / Advances in Mathematics 226 (2011) 4198–4211 4207 Our next step is to examine the end behavior of F (s) as σ → +∞ and σ → −∞. nσ 2 2 such that a(n) = 0. we have that F (s) = (s − 1)q O 1 nσ 1 O 1 nσ 2 (O( 21 ))2 σ C1 C3 nσ nσ (1 − c2 )(1 − c2 ) 1 2 as σ → +∞.Q. in this case.13).) As σ → −∞. We turn to use the following result for a nonconstant Lfunction L: For any σ and large t. there is an integer n2 2 such that C3 nσ 2 L2 − 1 C4 . C0 independent of σ.13) any more. we clearly have that n=1 ns C1 nσ 1 L1 − 1 C2 . we always have that F (s) = O 1 4σ =O 1 eσ  (2. as σ → +∞. p. nσ 1 as σ → +∞.12) for some positive constants C1 . Q. p. . where and in the sequel s = σ + it.7 in [7. Hence.15) holds also as σ → −∞. c2 . and λj . Since L1 (s) = ∞ a(n) with a(1) = 1. 2σ (2. nσ 2 and L2 = O 1 . (Note that Lemma 6. F (s) = (s − 1)q O 1 nσ 1 1 . c2 = 1. We also have that L1 − L2 = O It is then clear from (2. C4 > 0.1) that if c1 . we have that 1 − σ → +∞.15) as σ → +∞ (for any ﬁxed t). L(1 − σ + it) → 1 (cf. (2.
and (2.16) and (2.13)) and then and as σ → −∞. when d = 0 and when d = 0. This also holds when d = 0 and Q > 1. the functional equation in the axiom (iii) does not contain Gamma factors and λ above is 1. (L1 − L2 )(s) C0 λQ2 C0 λQ2 If d = 0.1). (2. which implies that for t 1 2 −σ t0 1 2 −σ t ( 2 −σ )dL 1 L(σ + it) C λQ2 t ( 2 −σ )dL 1 (2. cf.20) below.16). by (2. σ − 1 < Re w < σ + 1 and t − 1 < Im w < t + 1. . 1 s − 1q {C1 (λQ2 ) 2 −σ (t + 1)(3/2−σ )d }2 {C0 (λQ2 ) 2 −σ t ( 2 −σ )d O( 21−σ )}2 3 1 1 F (s) { (λQ2 ) 2 −σ t (1/2−σ )d − c1 }2 { (λQ2 ) 2 −σ t (1/2−σ )d − c2 }2 A1 s − 1q (t + 1)2d 1 + 1 t (1−2σ )d 1 1 O 1 41−σ A1 s − 1q (t + 1)2d 1 + 2(1 − 2σ )d =O 1 e−σ =O 1 eσ  1 1 O 1−σ t 4 (2. then for ﬁxed t 1 2 −σ 1 2 −σ t ( 2 −σ )dL (L1 − L2 )(1 − σ + it) t ( 2 −σ )d O 1 1 1 .17) noting that when w − (σ + it) = 1.Q. λQ2 1 2 −σ t (1/2−σ )d → +∞.12) or (2.19) for some constant A1 > 0.16) t0 and σ → for two positive constants . Thus (2. Li / Advances in Mathematics 226 (2011) 4198–4211 1 2 (2. that for ﬁxed large t t0 and σ → −∞.17) hold for L1 and L2 with dL = d.16). we also have that for t t0 and as σ → −∞. By the Cauchy formula and (2. 21−σ (2. λQ2 L(1 − σ + it) 2 as σ → −∞.) Thus.18) t0 > 1 and as σ → −∞. we have. where C1 = C if λQ2 > 1 and C1 = (λQ2 )−2 C if λQ2 < 1. in these two cases. C. in view of the simple inequality that (1 + x)m 1 + 2mx for any positive m and small positive x. (When d = 0. The inequalities (2.4208 B.18). we have for t −∞. (2. Applying the above stated result to L1 − L2 .15) holds for large t t0 and as σ → −∞. L (σ + it) = 1 2πi L(w) dw w − (σ + it) w−(σ +it)=1 3 2 −σ C1 λQ2 (t + 1)(3/2−σ )dL . (2. (2.17). Q > 1.
L2 . when d = 0 and Q = 1. Taking derivative in (2. and L1 − L2 = O( 21−σ ). then by (2.21) and as in (2.19) may not hold any more. 1 If d = 0. Taking derivative 1 in (2. L1 = O 1 21−σ 2 is the smallest integer n such that a(n) = 0 in (2. say c1 = ω. 21−s 3 (2. c2 is ω. We then have that . L(1 − s ) = L(1 − σ + it) has Dirichlet series ¯ representation with a(1) = 1. L2 = O 1 21−σ 2. L1 − c1  = L1 − ω as σ → −∞ for some A2 > 0.21) 1 If Q = 1. c2 = ω. Hence. L (s) = O(Q1−2σ ). Thus. Similarly. Thus for such σ . we deduce by (2. But.20) When σ → −∞.21). provided that c1 . Li / Advances in Mathematics 226 (2011) 4198–4211 4209 Consider now the cases d = 0. then by (2. L2 − c1  for some A3 > 0 and an integer n2 A2 1−σ n1 .B. If one of c1 .1) that 1 1 s − 1q (O( 21−σ ))2 (O( 21−σ ))2 F (s) (ω − c1 )2 (ω − c 2 )2  =O 1 e−σ as σ → −∞. That is. L2 .12). where n1 with L = L1 . Q 1. L is of the form L(s) = ωQ1−2s 1 + a(2) a(3) + 1−s + · · · . Q < 1. Thus.21). s − 1q O F (s) 2 2 1 1 1 O 1−σ (O( 21−σ n1−σ n2 1 A2 A3 (ω − c2 )2 n1−σ n1−σ 1 2 ))2 =O 1 eσ  . the functional equation in the axiom (iii) does not have any Gamma factors and becomes L(s) = ωQ1−2s L(1 − s ). and 1 21−σ (L1 − L2 )(s) = O as σ → −∞. The ﬁrst inequality in (2.Q. then limσ →−∞ L(s) = ω for L = L1 . when d = 0. (15) holds as σ → −∞. 1 − σ → +∞. L (s) = O( 21−σ ) for L = L1 . L2 . and L1 − L2 = O( 21−σ ).21). ¯ (2.21) A3 1−σ n2 . limσ →−∞ L(s) = 0 for L = L1 .
c2 is 0. To apply this result to our situation. or s = z−t1 i . by (2. L1 − c1 = L1 = ωQ1−2s 1 + O for L = L1 .22) as σ → −∞ and as σ → +∞. Set G(z) = F (s) = F z − t1 . Then f (s) = 0 identically. G(z) = F (σ + it1 ) = O 1 eσ  =O 1 . ez Therefore. Li / Advances in Mathematics 226 (2011) 4198–4211 1 s − 1q (O(Q1−2σ ))2 (O( 21−σ ))2 2 2 c1 c2  F (s) =O 1 eσ  as σ → −∞ provided that c1 c2 = 0. i Then by (2.21). 2 . on the imaginary axis of the zplane.15) holds also in this case. Thus we have. Hence.Q. This ﬁnally completes the proof of the theorem. (2. x = 0 and z = iy. the function G satisﬁes the conditions of the above result and thus G is identically zero. that when z = iy. If one of c1 . then by (2. or s = σ + it1 . L2 . 185]): Let f be holomorphic and of the form O(eks ) for Re(s) > 0 (k is a real number) and let f (s) = O(e−as ) for a constant a > 0 on the imaginary axis. Also. We then have that F (s) 1 s − 1q (O(Q1−2σ ))2 (O( 21−σ ))2 1 (ωQ1−2s (1 + O( 21−σ 2 )))2 c2  1 21−σ =O 1 eσ  as σ → −∞. p.22). we have G(z) = F z − t1 i = O eα z−t1 i  = O eαz for all z. say c1 = 0. We are now ready to ﬁnish up the proof by appealing to the following Carlson theorem (see [8. We have that F (σ + it1 ) = O 1 eσ  t0 and both σ → +∞ and (2. Fix such a large t = t1 . we make the variable transformation: z = x + iy = is + t1 = t1 − t + iσ. which correspond to t = t1 and y = σ .4210 B. from which it follows that F is identically zero.15) always holds for ﬁxed large t σ → −∞.11). We have thus showed that (2.
V. vol. [8] E. SpringerVerlag. Berlin. Salerno. Press. 1991. Titchmarsh. Salerno. Li / Advances in Mathematics 226 (2011) 4198–4211 4211 References [1] [2] [3] [4] [5] C. 182 (2) (1999) 207–241. W. in: E. Steuding. 1877. I Math. Gay. Maiori. Acad. Perelli. A. 1989. 367–385. 47–63. [6] A. SpringerVerlag. pp. I. Acta Arith. Paris Ser.A. Univ. Bombieri. The Theory of Functions. Selberg. Oxford. R. 319 (1994) 315–320. Kaczorowski. New York. Murty. Hayman. M. second edition. J. Berenstein. Complex Variables. II. 1992. 83 (1998) 271–281. Press. et al. R. 2007. Strong multiplicity one for Selberg’s class. Distinct zeros of Lfunctions.K. pp. 1964. E.R. Perelli. Bombieri.). Oxford Univ. vol. 1991. 1968. reprinted. SpringerVerlag..K. 0 d 1. . Murty.C. Value Distribution of LFunctions. On the structure of the Selberg class. [7] J.B. C. in: Collected Papers. Lecture Notes in Math.Q. A. Proceedings of the Amalﬁ Conference on Analytic Number Theory. (Eds. Old and new conjectures and results about a class of Dirichlet series. Oxford Univ. Acta Math. Sci. Meromorphic Functions.