Probability Surveys

Vol. 11 (2014) 121–160

ISSN: 1549-5787
DOI: 10.1214/13-PS214

Statistical properties of zeta

functions’ zeros
Vladislav Kargin
282 Mosher Way, Palo Alto, CA 94304
e-mail: [email protected]

Abstract: The paper reviews existing results about the statistical distri-
bution of zeros for three main types of zeta functions: number-theoretical,
geometrical, and dynamical. It provides necessary background and some
details about the proofs of the main results.

AMS 2000 subject classifications: 11M26, 11M50, 62E20.

Keywords and phrases: Riemann’s zeta, Selberg’s zeta, Ruelle’s zeta,
Montgomery’s conjecture, distribution of zeros.

Received February 2013.

Contents

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
2 Number-theoretical zetas . . . . . . . . . . . . . . . . . . . . . . . . . 123
2.1 Riemann’s zeta . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
2.1.1 Statistics of zeros on global scale . . . . . . . . . . . . . . 127
2.1.2 Statistics of zeros on local scale . . . . . . . . . . . . . . . 129
2.2 Dirichlet’s L-functions . . . . . . . . . . . . . . . . . . . . . . . . 132
2.2.1 Global scale . . . . . . . . . . . . . . . . . . . . . . . . . . 133
2.2.2 Local scale . . . . . . . . . . . . . . . . . . . . . . . . . . 134
2.3 L-functions for modular forms (The Hecke L-functions) . . . . . . 136
2.3.1 L-functions from modular forms . . . . . . . . . . . . . . 137
2.3.2 L-functions from Maass forms . . . . . . . . . . . . . . . . 140
2.3.3 Statistical properties of zeros: Global scale . . . . . . . . . 141
2.3.4 Local scale . . . . . . . . . . . . . . . . . . . . . . . . . . 142
2.4 Elliptic curve zeta functions . . . . . . . . . . . . . . . . . . . . 143
3 Selberg’s zeta functions for compact and non-compact manifolds . . . 145
3.1 Selberg’s zeta function and trace formula . . . . . . . . . . . . . 145
3.2 Statistics of zeros . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
3.3 Comparison with the circle problem . . . . . . . . . . . . . . . . 149
4 Zeta functions of dynamical systems . . . . . . . . . . . . . . . . . . . 150
4.1 Zetas for maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
4.1.1 Permutations . . . . . . . . . . . . . . . . . . . . . . . . . 150
4.1.2 Smooth mappings of compact manifolds . . . . . . . . . . 150
4.1.3 Subshifts . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
4.1.4 Ihara’s zeta function . . . . . . . . . . . . . . . . . . . . . 152
121
122 V. Kargin

4.1.5 Frobenius maps . . . . . . . . . . . . . . . . . . . . . . . . 154

4.1.6 Maps of an interval . . . . . . . . . . . . . . . . . . . . . . 155
4.2 Zetas for flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156

1. Introduction

The distribution of zeros of Riemann’s zeta function is one of the central prob-
lems in modern mathematics. The famous Riemann conjecture states that all
of these zeros are on the critical line Res = 1/2, and the accumulated numeri-
cal evidence supports this conjecture as well as a more precise statement that
these zeros behave like eigenvalues of large random Hermitian matrices. While
these statements are still conjectural, a great deal is known about the statistical
properties of Riemann’s zeros and the zeros of closely related functions. In this
report we aim to summarize findings in this research area.
We give necessary background information, and we cover the three main
types of zeta functions: number-theoretical, Selberg-type, and dynamical zeta
functions.
Some interesting and important topics are left outside of the scope of this re-
port. For example, we do not discuss quantum arithmetic chaos or characteristic
polynomials of random matrices.
The paper is divided in three main sections according to the type of the zeta
function we discuss. Inside each section we tried to separate the discussion of
the properties of zeros at the global and local scales.
Let us briefly describe these types of zeta functions and their relationships.
First, the number-theoretical zeta functions come from integers in number fields
and their generalizations. Due to the additive and multiplicative structures of
the integers, and in particular due to the unique decomposition
Q in prime factors,
the zeta functions have the Euler product formula ζ(s) = p (1 − (N p)s )−1 and
a functional equation, ζ(1 − s) = c(s)ζ(s), with the multiplier c(s) equal to a
ratio of Gamma functions.
It is an important discovery of Hecke that one can define number-theoretical
zeta functions in a different, and potentially more general way if one starts with
modular forms, which are functions on the space of two-dimensional lattices
that are invariant relative to the change of scale. If they are considered as
functions of the basis (z, 1) then they become functions of z invariant relative
to an action of the group SL2 (Z). They are periodic and hence can be written as
P
n≥0 cn exp(2πinz), where z is theP ratio of the periods of the lattice. If c0 = 0,
then one can define a zeta function cn n−s . It turns out that this zeta function
satisfies a functional equation. In addition, if the original modular form is an
eigenvector for certain operators (the Hecke operators), then the zeta function
will have the Euler product property.
It appears that all number-theoretical zeta functions can be obtained by using
this construction. However, the class of modular zeta functions is wider. Indeed,
 Statistics of zetas’ zeros 123

zeta functions with an arithmetic flavor can come not only from number fields
but also from other algebraic objects, for example, from elliptic curves. A signif-
icant development occurred recently when it was proved that all zeta functions
associated with elliptic curves come from modular zeta functions. Among other
applications, this discovery was a key to the proof of Fermat’s last theorem.
(The existence of a non-trivial solution for xn + y n = z n , n > 2, would imply
the existence of a non-modular zeta function for an elliptic curve.)
An important representative of the second class of zetas is Selberg’s zeta
function, which is essentially a generating function for the lengths of closed
geodesics on a surface with constant negative curvature. In more detail, let H be
the upper half-plane with the hyperbolic metric and Γ be a discrete subgroup
of SL2 (Z). Selberg showed that certain sums over eigenvalues of the Laplace
operator on Riemann surface Γ\H can be related to sums over closed geodesics
of Γ\H. This relation is called Selberg’s trace formula. Selberg’s zeta function
is constructed in such a way that it has the same relation to Selberg’s trace
formula as Riemann’s zeta function has to the so-called “explicit formula” for
sums over Riemann’s zeros.
While Selberg’s zeta function resembles Riemann’s zeta in some features,
there are significant differences. In particular, the statistical behavior of its zeros
depends on the group Γ and often it is significantly different from the behavior
of Riemann’s zeros.
The third class of zetas, the dynamical zeta functions, are generating func-
tions for the lengths of closed orbits of a map f that sends a set M to itself. The
most spectacular example of these functions is Weil’s zeta functions of algebraic
varieties over finite fields, which can be described as follows.
Let M be the algebraic closure of an algebraic variety embedded in Fnq (where
Fq is a finite field), and let f be the Frobenius map: (x1 , . . . , xn ) → (xq1 , . . . , xqn ).
Then, the dynamic zeta associated to f is called Weil’s zeta function. These
functions are remarkable since the Riemann conjecture is proved for them: It is
known that their zeros are located on a circle that corresponds to the line Rez =
1/2. Moreover, much is known about the statistical distribution of these zeros.
A particular case of these zeta functions, Weil’s zeta functions for curves,
can be understood as number-theoretic zeta functions for finite extensions of
the field Fq (x). Hence, Weil’s zeta functions work as a bridge between dynamic
and number-theoretic zeta functions. In addition, Selberg’s zeta function can be
understood as the dynamic zeta function for the geodesic flow on the surface
Γ\H. This shows that the three classes of zeta functions are intimately related
to each other.
With this overview in mind, we now come to a more detailed description of
available results about the statistics of zeta function zeros.

2. Number-theoretical zetas
2.1. Riemann’s zeta
There are several excellent sources on Riemann’s zeta and Dirichlet L-functions,
for example the books by Davenport [9] and Titchmarsh [59]. In addition, a very
124 V. Kargin

good reference for all topics in this report is provided by Iwaniec and Kowalski’s
book [26].
By definition Riemann’s zeta function is given by the series
X∞
1
ζ (s) = s
, (1)
n=1
n

for Res > 1. It can be analytically continued to a meromorphic function in the

entire complex plane and it satisfies the functional equation
   
− 2s 1 − 1−s 1
π Γ s ζ (s) = π 2 Γ (1 − s) ζ (1 − s) . (2)
2 2
P
Indeed, we can relate ζ(s) to the series θ(x) = ∞ n=1 e
−n2 πx
:

∞ Z Z
Γ 12 s ζ (s) X ∞ s/2−1 −n2 πx ∞
= x e dx = xs/2−1 θ (x) dx.
π s/2 n=1 0 0

(As an aside remark, this representation can be used as a starting point for some
surprising connections of the Riemann zeta function with the Brownian motion
and Bessel processes, see [3].) From the identities for the Jacobi theta-functions,
implied by the Poisson summation formula, it follows that
   
1 1
2θ (x) + 1 = √ 2θ √ +1 . (3)
x x
Writing
Z ∞ Z 1 Z ∞
xs/2−1 θ (x) dx = xs/2−1 θ (x) dx + xs/2−1 θ (x) dx,
0 0 1

and applying the identity to the integral from 0 to 1, we find that

Z ∞ 
Γ 21 s ζ (s) 1 −s/2−1/2 s/2−1
= + x + x θ (x) dx, (4)
π s/2 s (s − 1) 1

which is symmetric relative to the change s → 1 − s.

In addition, formula (4) implies that the function s(s − 1)π −s/2 Γ( 21 s)ζ(s) is
entire. Hence, the only pole of ζ(s) is at s = 1, and ζ(s) has zeros at −2, −4,
. . . , that correspond to poles of Γ( 12 s). These zeros are called trivial, while all
others are called non-trivial. We will order the non-trivial zeros according to
their imaginary part and denote them by ρk . By the functional equation, ρk are
located symmetrically relative to the line Res = 1/2, which is called critical,
and it is known that 0 < Reρk < 1. The Riemann’s hypothesis asserts that all
non-trivial zeros are on the critical line.
The second fundamental property of Riemann’s zeta is the Euler product
formula:
Y 1
−1
ζ (s) = 1− s (5)
p
p
 Statistics of zetas’ zeros 125

valid for Res > 1. It follows from the existence and uniqueness of prime factor-
ization. This formula is a starting point for a very important idea which relates
the sums over prime numbers and sums over zeta function zeros. For example,
the Riemann-von Mangoldt formula says that
X X xρ X x−2n ζ′
Λ (n) = x − + − (0) , (6)
ρ
ρ n
2n ζ
n≤x

where Λ(n) = log p, if n is a prime p or a power of p, and otherwise Λ(n) = 0.

(The order of summation
P ρ over zeros can be important here, so it is assumed
that in computing ρ xρ , one takes all zeros with imaginary part between −T
and T , and then let T → ∞. In addition, if x is a prime power the formula has
to be modified by subtracting 21 Λ(x) on the left-hand side.)
The idea of the proof of (6) is to take the logarithmic derivative of (5):
X∞
ζ′ Λ (n)
(s) = − , (7)
ζ n=2
ns
and then to use the following formula with y = x/n and c > 0:

Z c+i∞ 0 if 0 < y < 1,
1 ds  1
ys = if y = 1,
2πi c−i∞ s  2
1 if y > 1.
One integrates (7) against the test function xs /s, in order to pick out the terms
in the series in (7) with n ≤ x.
From (7), one gets
X Z c+i∞  ′ 
1 ζ ds
Λ (n) = − (s) xs .
2πi c−i∞ ζ s
n≤x

Moving the line of integration away to infinity on the left and collecting the
residues at the poles one finds formula (6). (See Chapter 17 in Davenport [9] for
a detailed proof.)
By a similar method one can obtain:
X Λ (n) x1−s X xρ−s X x−2n−s ζ′
s
= − + − (s) . (8)
n 1−s ρ
ρ−s n
2n + s ζ
n≤x

While formula (6) is useful to study the distribution of primes if something is

known about the distribution of zeros, formula (8) can be used in the reverse
′
direction to study the behavior of ζζ (s) if some information is known about
primes.
Selberg discovered a variant of this formula that avoids the problem of con-
ditional convergence in the sum over the zeta zeros. Define


  Λ (n) ,  for 1 ≤ n ≤ x,

 3 2
log2 xn −2 log2 xn
Λx (n) = Λ (n) 2 log2 x
, for x ≤ n ≤ x2 ,



 log2 x
3

Λ (n) 2 log2nx , for x2 ≤ n ≤ x3 .

126 V. Kargin

Then, (Lemma 10 in [51] )

X Λx (n) 2
ζ′ 1 X xρ−s (1 − xρ−s )
(s) = − + (9)
ζ
n≤x3
ns log2 x ρ (s − ρ)
3

2 ∞
2
x1−s 1 − x1−s 1 X x−2q−s 1 − x−2q−s
+ 3 + .
log2 x · (1 − s) log2 x q=1 (2q + s)
3

Another useful variant is sometimes called Weil’s and sometimes Delsarte’s

explicit formula (see [8] and [41]). Suppose that H(s) is an analytic function in
the strip −c ≤ Im s ≤ 1 + c (for c > 0) and that |H(σ + it)| ≤ A(1 + |t|)−(1+δ)
uniformly in σ in the strip. Let h(t) = H( 12 + it) and define
Z
b
h (x) = h (t) e−2πitx dt.
R

(Note that analyticity of H(s) implies that bh(x) has finite support.) Then,
X ∞     
1 X Λ (n) b log n b log n
H (ρ) = H (0) + H (1) − √ h +h −
ρ
2π n=1 n 2π 2π
Z ∞
1
− h (t) Ψ (t) dt, (10)
2π −∞
where    
Γ′ 1 Γ′ 1
Ψ (t) = + it + − it .
Γ 2 Γ 2
The idea of the proof is similar to the proof of the Riemann-von Mangoldt
formula. One starts with the formula
Z 2+i∞  ′  X∞ Z ∞
1 ζ Λ (n) 1
− (s) H (s) ds = − 1/2 2π
h (t) e−i(log n)t dt
2πi 2−i∞ ζ n=1
n −∞

X∞  
1 Λ (n) b log n
= − h .
2π n=1 n1/2 2π

Next, one can move the line of integration to (−1 − i∞, −1 + i∞) and use
the calculus of residues and the functional equation to obtain formula (10).
A wonderful illustration for the significance of explicit formulas is given by
Landau’s formula:
X T √

xρ = − Λ x + O (log T ) .
2π
0<Imρ<T
1
Assuming Riemann’s hypothesis, we write ρ = 2 + iγ, and then
X T Λ (x)
cos (γ log x) = − + O (log T ) .
2π x
0<γ<T

Note that the right hand side has a spike when x is a prime power. This is
illustrated in Figure 1.
 Statistics of zetas’ zeros 127

Fig 1. Sums of cos(γ log(x)) over the first 100,000 zeros (T ≈ 75 × 103 ). The horizontal axis
shows x.

The most general explicit formula was derived by Weil (see [61]). We will not
present it here since it involves adelic language and this would take us too far
afield.

2.1.1. Statistics of zeros on global scale

Let N (T ) denote the number of zeros with the imaginary part strictly between
0 and T . If there is a zero with imaginary part equal to T , then we count this
zero as 1/2. Define  
1 1
S(T ) := Im log ζ + iT ,
π 2
where the logarithm is calculated by continuous variation along the contour
σ + iT, with σ changing from +∞ to 1/2.
By applying the argument principle to ζ and utilizing the functional equation,
it is possible to show (see Chapter 15 in [9]) that
 
T T 7 1
N (T ) = log + + S (T ) + O .
2π 2πe 8 1+T
Let √
2πS(t)
X (t) := √ .
log log t
Then, we have the following theorem by Selberg. (See Theorem 6 in [51].)
128 V. Kargin

Theorem 2.1 (Selberg) Let T a ≤ H ≤ T 2 , where 21 < a ≤ 1. Then for every

k≥1 Z
1 T +H 2k 2k! −1/2
|X (t)| dt = + O((log log T ) ),
H T k!2k
with the constant in the remainder term depending on k and a only.
In other words, if t is chosen randomly in the interval [0, T ] and T → ∞,
then X(t) converges in distribution to a Gaussian random variable. Note that
the Riemann Hypothesis is not assumed in this result. Under the assumption
of the Riemann Hypothesis, Selberg proved an analogous result with a better
error term assuming only a > 0 (Theorem 3 in [52]).
Recently, Selberg’s result was generalized by Bourgade, who determined the
correlation structure of X(t) on relatively small scales. The basis for this de-
velopment is the following result (cf. Theorem 1.1 in [7]). Take a large t > 0
and small εt ≥ 0 and look at l points on the line Rez = 21 + εt with imaginary
(i)
parts ωt + ft , 1 ≤ i ≤ l, where ω is a random variable uniformly distributed
on [0, 1]. Note that the randomness ω is the same for all points so the distance
(i)
between the points is measured by offsets ft . We assume that εt → 0, so the
points approach the critical line as t → ∞. The first case is when εt ≫ 1/ log t
(j) (i) c
and we look at the scale |ft − ft | ≈ εt ij , with cij ≥ 0. The second case
is when εt ≪ 1/ log t, for example, εt = 0. In this case, we look at the scale
(j) (i)
|ft − ft | ≈ (1/ log t)cij . It turns out that after a proper normalization the
values of the logarithm of the Riemann zeta function at this cluster of points
converge to a non-trivial multivariate Gaussian distribution.
Theorem 2.2 (Bourgade) Let ω be uniform on (0, 1), ǫt → 0, t → ∞, ǫt ≫
(1) (l)
1/ log t, and functions 0 ≤ ft < · · · < ft < c < ∞. Suppose that for all i 6= j,
(j) (i)
log ft − ft
→ ci,j ∈ [0, ∞] . (11)
log ǫt
Then the vector
  
1 1 (1) 1 (l)
√ log ζ + εt + ift + iωt, . . . , + εt + ift + iωt (12)
− log ǫt 2 2
converges in law to a complex Gaussian vector (Y1 , . . . , Yl ) with the zero mean
and covariance function

1 if i = j,
Cov (Yi , Yj ) =
1 ∧ ci,j if i =
6 j.
Moreover, the above result remains true if ǫt ≪ 1/ log t, replacing the normal-
ization − log εt with log log t in (11) and (12).
This theorem implies the following result for the zeros of the Riemann zeta
function (cf. Corollary 1.3 in [7]). Let
   
t2 t2 t1 t1
∆ (t1 , t2 ) = N (t2 ) − log − N (t1 ) − log ,
2π 2πe 2π 2πe
 Statistics of zetas’ zeros 129

which represents the number of zeros with the imaginary part between t1 and t2
minus its deterministic prediction. Then the claim is that this excess number of
zeros in an interval with the length of order (log t)−δ (0 < δ < 1) is a Gaussian
variable with the variance proportional to (1−δ) log log t. Moreover, the limiting
Gaussian process has an interesting covariance structure.
Corollary 2.3 (Bourgade) Let Kt be such that, for some ε > 0 and all t,
Kt > ε. Suppose log Kt / log log t → δ ∈ [0, 1), as t → ∞. Then the finite-
dimensional distributions of the process

∆ (ωt + α/Kt , ωt + β/Kt )

1
p , 0 ≤ α < β < ∞,
π (1 − δ) log log t

e
converge to those of a centered Gaussian process (∆(α, β), 0 ≤ α < β < ∞)
with the covariance structure


 1 if α = α′ , and β = β′,


   1/2 if α = α′ , and β 6= β ′ ,
e e ′ ′
E ∆ (α, β) ∆ (α , β ) = 1/2 if α 6= α′ , and β = β′,


 −1/2
 if β = α′ ,

0 elsewhere.

Note that since the average spacing between zeros is 1/ log t, hence the num-
ber of zeros in the interval (ωt + α/Kt , ωt + β/Kt ) is of order (log t)1−δ .
This result perfectly corresponds to a result of Diaconis and Evans about
eigenvalue fluctuations of random unitary matrices (cf. Theorem 6.3 in [10]).
The key to both Selberg and Bourgade’s results is Selberg’s approximation
for the function S(t) (cf. Theorem 4 in [51]).
Proposition 2.4 (Selberg) Suppose k ∈ Z+ , 0 < a < 1. Then there exists
ca,k > 0 such that for any 1/2 ≤ σ ≤ 1 and ta/k ≤ x ≤ t1/k , it is true that
2k
Z t X
1 1
log ζ (σ + is) − ds ≤ ca,k .
t 1 pσ+is
p≤x3

The proof of this statement, in turn, depends on Selberg’s formula (9).

2.1.2. Statistics of zeros on local scale

Now, assume the Riemann Hypothesis and suppose that we are interested in
calculating
P local statistics for the pairs of the zeta zeroes, for example, in
′ log T
0<γ,γ ′ ≤T r((γ − γ ) 2π ), where r(x) is a test function. By representing r(x)
as the Fourier transform, we can write this statistic differently,
X   Z ∞
′ log T ′ T
r (γ − γ ) w (γ − γ ) = log T F (α) rb (α) dα, (13)
2π 2π −∞
0<γ,γ ≤T
′
130 V. Kargin

Rwhere we added a convenient weighting function w(u) = 4/(4+u2 ). Here rb(α) =

∞ −2πiαu
−∞
r(u)e du, and

2π X
T iα(γ−γ ) w (γ − γ ′ ) .
′
F (α) = F (α, T ) :=
T log T
0<γ,γ ′ <T

Consequently, all information about local statistics is encoded in the function

F (α). Montgomery proved the following result [40]. (In fact, the estimate holds
uniformly throughout 0 ≤ α ≤ 1, as was later proved by Goldston in [15].)
Theorem 2.5 (Montgomery) For real α, the function F (α) is real and
F (α) = F (−α). If T > T0 (ε), then F (α) ≥ −ε for all α. For fixed α satis-
fying 0 ≤ α < 1, we have

F (α) = (1 + o (1)) T −2α log T + α + o (1) ,

as T tends to infinity; this holds uniformly for 0 ≤ α ≤ 1 − ε.

This result allows us to calculate the local statistics for smooth test functions
that have their Fourier transforms compactly supported on the interval [−1, 1].
Montgomery conjectures that

F (α) = 1 + o (1)

for α ≥ 1, uniformly in bounded intervals. If it is true, then it can be used

to calculate the local statistics for a much larger class of test functions. In
particular, Montgomery’s conjecture can also be formulated as follows.
Conjecture 2.6 (Montgomery) For fixed α < β,

X Z β"  2 # !
sin πu T
1∼ 1− du + δ (α, β) log T,
′ α πu 2π
0<γ,γ <T
2πα/ log T <|γ−γ ′ |<2πβ/ log T

as T goes to infinity. Here δ(α, β) = 1 if 0 ∈ [α, β], δ(α, β) = 0 otherwise.

The proof of Montgomery’s theorem is based on the analysis of the following
variant of the explicit formula.
Lemma 2.7 (Montgomery) If 1 < σ < 2 and x ≥ 1 then
 
X (2σ − 1)xiγ X  x 1−σ+it X  x σ+it
−1/2  
1

2 = −x Λ (n)
n
+ Λ (n)
n
γ σ − 2 + (t − γ) n≤x n>x
 
1/2−σ+it
+x (log τ + Oσ (1)) + Oσ x1/2 τ −1 ,

where τ = |t| + 2. The implicit constants depend only on σ.

 Statistics of zetas’ zeros 131

3
Set σ = 2 and x = T α . One computes the integral of the left-hand side:
Z T X
2
 
xiγ
2 dt = 2πF (α, T ) T log T + O (log T )3 .
0 γ 1 + (t − γ)2
For the corresponding integral of the right-hand side, one uses the Montgomery-
Vaughan formula,
Z T X∞ 2 ∞
X
an 2
it
dt = |an | (T + O (n))
0 n=1
n n=1

and finds that the integral of the right-hand side equals

T h 2
i
T log x + O (x log x) + 2 (log T ) + O (log T ) ,
x
which gives the claim of the theorem when one substitutes x = T α .
For more information about Montgomery conjecture, see [14].
Montgomery’s result allows us to compute the statistic (13) for pairs of Rie-
mann’s zeta zeros, provided that the Fourier transform of the test function r is
supported on the interval [−1, 1]. What about statistics of other k-tuples of the
zeros? First of all, the case of linear statistics (k = 1) is similar to the questions
considered by Selberg and Bourgade except that now we allow for more general
test functions. The main interest here is to see how far we can go in localizing
this functions.
In this directions Hughes and Rudnick in [22] studied the distribution of
X  log T 
Nf (t, T ) := f (γj − t) ,
γ
2π
j

where γj = 1i (ρi − 12 ) and γj are not assumed real. The function f is a real-
valued even function with the smooth compactly-supported Fourier transform.
(If f is the indicator function of an interval [−a, a] and if all γj are real, then
2π 2π
Nf (t, T ) counts number of zeros in the interval [t − a log T , t + a log T ]. However,
the Fourier transform of the indicator function does notR have compact support.)
Choose a weight function w(x), such that w ≥ 0, w(x)dx = 1, and w(x) b is
compactly supported, and define an averaging operator
Z  
t − T dt
hF iT,H := F (t) w .
R H H
Theorem 2.8 (Hughes-Rudnick) Let the averaging window H = T a for 0 <
R
a ≤ 1, and let be such that fb(u) = f (x)e−2πixu dx ∈ Cc∞ (R) and Supp f ⊂
(−2a/m, 2a/m) with integer m ≥ 1. Then, as T → ∞, the first m moments of
Nf , hNfm iT,H converge to those of a Gaussian random variable with expectation
R
f (x)dx and variance
Z
σf = min (|u| , 1) fb(u)2 du.
2
132 V. Kargin

Hence, if the frequency of the test function oscillations is bounded (and there-
fore the function is very smooth and well delocalized in the x-space), then the
first moments of the linear statistic converge to those of the Gaussian variable.
What about higher moments? Hughes and Rudnick show (Theorem 6.5 in [23])
that a similar result holds in the random matrix theory for eigenvalues of a
unitary random matrix. For random matrices, higher moments do not converge
to Gaussian values (Theorem 7.4 in [23]) Based on this analogy, they conjecture
that it is the same for the linear statistics of the zeta function zeros.
We will describe the ideas of the proof of Theorem 2.8 below in the case when
they are applied to the zeros of Dirichlet’s L-functions.
What about statistics of k-tuples of zeros when k > 2? This case was consid-
ered by Rudnick and Sarnak in [47]. Their results hold for a quite large class of
L−functions, and we will postpone their discussion to a later section. Briefly,
they are similar to Montgomery’s results since they show that the behavior of
the zeta zeros is very similar to the behavior of eigenvalues of random unitary
matrices. Another similarity is that the results are proven under some restrictive
conditions on the Fourier transform of the test function. It is an outstanding
problem to prove that all results about correlations of zeros hold without these
restrictive hypotheses.

2.2. Dirichlet’s L-functions

In order to understand the behavior of the Riemann zeta zeros, it is worthwhile

to check for which functions their zeros have similar behavior. The simplest
example of such a family of functions is provided by Dirichlet’s L-functions.
Let χ(n) denote a multiplicative character modulo a positive integer q. That
is, the function χ maps integers to the unit circle; it is multiplicative, χ(nm) =
χ(n)χ(m), and χ(n) = 0 if n and q are not relatively prime. The character which
sends every integer relatively prime to q to 1 is called the principal character
modulo q. The conductor of the character is the minimal integer N such that the
character is periodic modulo N. For simplicity, let q be a prime in the following.
In this case the conductor equals q. A character is odd if χ(−1) = −1, and even
if χ(−1) = 1.
The Dirichlet L-function corresponding to the character χ is defined by the
series
X∞
χ (n)
L (s, χ) = .
n=1
ns
This function has the Euler product representation:
Y χ (p)
−1
ζ (s) = 1− s (14)
p
p

because of the multiplicativity of χ(n). Moreover, the argument behind the

relation to theta functions (4) can be repeated and as a consequence, one finds
that series (2.2)can be continued to a function which is meromorphic in the
 Statistics of zetas’ zeros 133

whole complex plane and satisfies a functional equation. Namely, let µ = 0 if χ

is even and µ = 1 if χ is odd. Define
 
1
(s+µ) − 21 (s+µ) 1
Φ (s, χ) = q 2 π Γ (s + µ) L (s, χ) ,
2
then the functional equation has the form
√
iµ q
Φ (1 − s, χ) = Φ (s, χ) ,
τ (χ)
where τ (χ) is the Gauss sum:
q
X
τ (χ) = χ (m) e2πim/q .
m=1

The reason for appearance of τ (χ) is that the modularity relation (3) becomes
more complicated in this case. For proofs, see Chapter 9 in Davenport [9].
Many other properties of the Dirichlet L-functions is similar to that of the
Riemann zeta functions. In particular, one can establish similar explicit formu-
las.
Two notable differences from the Riemann zeta is that (i) if χ is not principal,
then L(s, χ) is entire; and (ii) if χ is not even, then a complex conjugate of a
zero is not necessarily a zero.

2.2.1. Global scale

For T > 0, let N (T, χ) denote the number of zeros of L(s, χ) with 0 < σ < 1
and 0 ≤ t ≤ T, counting possible zeros with t = 0 or t = T as one half only. Let
 
1 1
S (t, χ) = Im log L + it, χ .
π 2
Then it can be shown that
 
T Tq χ (−1) 1
N (T, χ) = log − + S (T, χ) − S (0, χ) + O .
2π 2πe 8 1+T
See formula (1.3) in Selberg’s paper [50] and Chapter 16 in Davenport [9].
If the character is fixed and T is large, then the results for N (T, χ) are quite
similar to results for N (T ).
A different situation arises when the interval [0, T ] is fixed and the character
χ varies (in particular, if χ is random). This situation was studied by Selberg,
who proved the following result (cf. Theorem 9 in [50]).
Theorem 2.9 (Selberg) For |t| ≤ q 1/4−ε , we have
1 X (2r)!
|S (t, χ)|2r = (log log q)r + O((log log q)r−1 ),
q−2 χ r! (2π)2r

where the summation if over all non-principal characters over the base q.
134 V. Kargin

In other words, if t is fixed and q grows to infinity, then the distribution

of S(t, χ) approaches the distribution of a Gaussian random variable with the
variance 2π1 2 log log q.
One is naturally let to the question of correlations between S(t, χ) for different
χ. One result in this direction is stated by Fujii (see p. 233 in [13]). Namely,
Z T  
δχ ,χ T
S (t, χ1 ) S (t, χ2 ) dt = 1 2 T log T + A (χ1 , χ2 ) T + O √ ,
0 2π log T
where A (χ1 , χ2 ) is a constant that depends on χ1 and χ2 , which basically says
that S(t, χ1 ) and S(t, χ2 ) are uncorrelated as functions of a random t if χ1 6= χ2 .
(See, however, a critique of Fujii’s proof on p. 4 in [31].)
Apparently, the question of correlations between S(t1 , χ) and S(t2 , χ) as func-
tions of a random χ has not yet been investigated.

2.2.2. Local scale

In [23], Hughes and Rudnick study the linear statistics of low-lying zeros of L-
functions on the local scale. Hughes and Rudnick order the zeros ρi,χ = 12 + iγi,χ
as follows:

· · · ≤ Reγ−2,χ ≤ Reγ−1,χ < 0 ≤ Reγ1,χ ≤ Reγ2,χ ≤ · · ·

and define
log q
xi,χ = γi,χ .
2π
Then they define
∞
X
Wf (χ) = f (xi,χ )
i=−∞

where f is a rapidly decaying test function.

The question is to understand the behavior of the averages
1 X
Wfm = Wf (χ) .
q−2
χ6=χ0

The basis for their analysis is a variant of the explicit formula (10) that
relates a sum over zeros of L(s, χ) to a sum over prime powers. This formula is
a particular version of the formula from Rudnick and Sarnak [47], which is valid
for a more general class of zeta functions. Let h(r) be any even analytic function
in the strip −c ≤ Imr ≤ 1 + c (for c > 0) such that |h(r)| ≤ A(1 + |r|)−(1+δ)
(for r ∈ R, A > 0, δ > 0). Then (cf. formula (2.1) in [23]),
X Z ∞
1
h (γj,x ) = h (r) (log q + Gχ (r)) dr (15)
j
2π −∞
X Λ (n)
− √ b h (log n) (χ (n) + χ (n)) ,
n
n
 Statistics of zetas’ zeros 135

R
where b
h(u) = 1
h(r)e−iru dr, and
2π
   
Γ′ 1 Γ′ 1 1
Gχ (r) = + µ (χ) + ir + + µ (χ) − ir − log π.
Γ 2 Γ 2 2
(Recall that µ(χ) = 0, if χ is even, and = 1, if χ is odd.)
Take h(r) = f ( log q b 1 b u
2π r), so that h(u) = log q f ( log q ). We say that f is admissible,
R
if it is a real, even function whose Fourier transform fb(u) := f (r)e−2πiru dr is
compactly supported, and such that |f (r)| ≤ A(1 + |r|)−(1+δ) . Then, from (15)
we get the following decomposition:
Wf (χ) = Wf (χ) + Wfosc (χ) ,
where Z ∞  
log q
Wf (χ) := f r (log q + Gχ (r)) dr,
−∞ 2π
and  
1 X Λ (n) b log n
Wfosc (χ) := − √ f (χ (n) + χ (n)) .
log q n n log q
For large q, Z  
∞
1
Wf (χ) := f (x) dx + O ,
−∞ log q
which is asymptotically independent of χ.
For the oscillating part one has the following result (cf. Theorem 5.1 in [23]).
Theorem 2.10 (Hughes and Rudnick) Let f be an admissible function and
assume that  
Supp fb ⊆ [−α, α] , α > 0.
If m < 2/α, then the m-th moment of Wfosc is
 m! m

m σ (f ) , if m is even,
lim Wfosc q
= 2m/2 (m/2)!
q→∞ 0, if m is odd,
where Z 1
2
σ (f ) = |u| fb(u)2 du.
−1
In other words, the first several moments of the statistic Wf converge to the
corresponding moments of a Gaussian random variable. A similar situation holds
for an eigenvalue statistic of random unitary matrices. Hughes and Rudnick
show (Theorem 7.4) that higher moments for this eigenvalue statistic are not
Gaussian (using results from the work of Diaconis and Shahshahani [11]). They
conjecture that the same result should hold for the statistic Wf .
As an application, Hughes and Rudnick derived some results for the small-
est zero of L(s, χ). In particular, they showed that for infinitely many q there
are characters χ such that the imaginary part of the zero is between 0 and
1/4 (Corollary 8.2 in [23]). They conjecture that 1/4 can be substituted with
arbitrary positive constant.
136 V. Kargin

Moreover, if β > 6.333, then a proportion of characters χ whose L function

has a zero with imaginary part between 0 and β is greater than c(β) > 0 for all
sufficiently large q. (Theorem 8.3 in [23]). The conjecture is that this is in fact
true for every β > 0.

2.3. L-functions for modular forms (The Hecke L-functions)

Hecke generalized the Riemann zeta function by using ideals in an imaginary

quadratic field K instead of integers:
X −s
LK (s) = (N (a)) .
a

This function has a Euler product formula and a functional equation, although
the latter is somewhat different: let
p !s
|D|
ΛK (s) = Γ (s) LK (s) ,
2π
where D is the discriminant of the imaginary quadratic field K. Then
ΛK (1 − s) = ΛK (s) .
(Compare this with (2).) The proof of this functional equation is similar to
Riemann’s proof and relies on a modularity property of a certain complex-
analytic function, that is, on its behavior relative to the change of variable
z → 1/z.
Motivated by this example, Hecke had a fruitful idea of obtaining L-functions
from complex-analytic functions that transform well under the action of the
modular group, and then checking which additional conditions are needed to
ensure that a functional equation and a Euler product formula holds. The objects
constructed in this way are called Hecke L-functions.
In order to illustrate, let
X
f (z) = cn e2πinz , and
n>0
X
L (f, s) = cn n−s ,
n>0

Define √ !s
N
Λ (f, s) = Γ (s) L (f, s) .
2π
Since √ !s Z ∞ √
N dy
Γ (s) n−s = e−2πny/ N y s ,
2π 0 y
we have the representation
Z ∞  
iy dy
Λ (f, s) = f √ ys ,
0 N y
 Statistics of zetas’ zeros 137

which implies that

Z ∞   Z ∞  
iy dy iy dy
Λ (f, s) = f √ ys + ik Wf √ y k−s , (16)
1 N y 1 N y

where √ −k  −1 
W f (z) = Nz f . (17)
Nz
Consequently, any eigenfunction of the operator W will have a functional equa-
tion similar to the functional equation for the Riemann zeta function. The
idea is to find a suitable finite-dimensional space of functions f, which is in-
variant under the action of W and diagonalize W in this finite-dimensional
space.
Below, we give a brief outline of these ideas. A good source for this material
is Chapter 14 in [26] and Chapter V in [38].

2.3.1. L-functions from modular forms

Let Γ is a subgroup of finite index in SL2 (Z) and let H denote the upper half-
plane {z|Imz > 0}. The set H∗ = H∪{∞} ∪ Q can be made into a Hausdorff
topological space and one can define a continuous action of SL2 (Z) on H∗ as an
extension of the action of SL2 (Z) on H :
 
a b az + b
if γ = , then γz = .
c d cz + d

The cusps are points in H∗ \H. One can also show that Γ\H∗ is a compact
Hausdorff space (that is, a compact space in which every two points have disjoint
open neighbourhoods), which is a Riemann surface (that is, it admits a complex
structure).
Next, we define an action of SL2 (Z) on functions f : H∗ → C :
 
a b −k
if γ = , then f ◦ [γ]k := (cz + d) f (γz) .
c d

In fact if γ ∈ GL2 (Z), then one can define its action by

h i
−1/2
f ◦ [γ]k := f ◦ det (γ) γ .
k

It can be checked that this is indeed a group action of GL2 (Z) on functions.
While modular forms can be defined for any discrete subgroup Γ, the most
studied are subgroups Γ0 (N ):
  
a b
Γ0 (N ) := | c ≡ 0 (modN ) .
c d

For these subgroups, we have the following definition.

138 V. Kargin

Definition 2.11 Let χ be a Dirichlet character modulo N. A modular form for

Γ0 (N ) of weight k ≥ 1 and character χ is a function H∗ → C such that
(a) f is holomorphic on H;
(b) for any γ ∈ Γ0 (N ), f ◦ [γ]k = χ(d)f ;
(c) f is holomorphic at the cusps.
A cusp form is a modular form which is zero at all the cusps.
In particular, the complex analyticity at infinity implies that a modular form
for Γ0 (N ) can be written as
X
f (z) = cn e2πinz .
n≥0

For a cusp form, c0 = 0.

For simplicity, we will only consider the forms with the principal character χ
and we will write Mk (N ) and Sk (N ) to denote the linear spaces of the modular
and cusp forms. One can show that these spaces are finite dimensional for each k.
Note that if k is odd then Mk (N ) is zero since f ◦ [−I]k = (−1)k f, hence we
should have f = −f.
Definition 2.12 Let f be a cusp form of weight 2k for Γ0 (N ),
X
f (z) = cn e2πinz .
n>0

The L-series of the cusp form f is the Dirichlet series

X
L (f, s) = cn n−s .
n>0

It is possible to estimate that |cn | ≤ Cnk , and therefore this series is conver-
gent for Res > k + 1.
The crucial fact is that the space of modular forms is invariant under
the
action of
operator W from (17). Indeed, W f = f ◦ [ω]k , where ω = N0 −1 0 . If
γ = ac db ∈ Γ0 (N ). Then
 
d −c/N
ωγω −1 = ∈ Γ0 (N ) .
−N b a

Hence  
W f ◦ [γ]k = f ◦ ωγω −1 k ◦ [ω]k = f ◦ [ω]k = W f.
where the second step is by modularity of f. One can also check that W preserves
S2k (N ). Since W 2 = 1, hence the only eigenvalues of wN are ±1, and S2k (N )
+1 −1
is a direct sum of the corresponding eigenspaces, S2k = S2k + S2k .
By the argument in the beginning of this section (see formula (16)), one can
infer the following result.
 Statistics of zetas’ zeros 139

Theorem 2.13 (Hecke) Let f ∈ S2k (Γ0 (N )) be a cusp form in the ε-eigenspace,
ε = 1 or −1. Then the function Λ(s, f ) := N s/2 (2π)−s Γ(s)L(f, s) extends ana-
lytically to a holomorphic function on the whole complex plane, and satisfies the
functional equation
k
Λ (s, f ) = ε (−1) Λ (k − s, f ) .
The natural question is what about the Euler product formula?
By studying the properties of the modular forms that arise from the L-
functions of the quadratic imaginary fields (and that have the product formula
almost by definition), Hecke was able to formulate a list of properties which
should be imposed on the modular form f, so that L(f, z) had a product for-
mula.
Namely, define the Hecke operators (cf. formula (14.46) in [26])
 
1 X k X az + b
[T (n) f ] (z) := a f .
n d
ad=n 0≤b<d

It can be checked that these are linear operators on S2k (Γ0 (N )). (See Section
IX.6 in Knapp [30] or Section V.4 of Milne [38] or Section VII.5 of Serre [53] for
details). They have the following properties:
Theorem 2.14 (Hecke) The maps T (n) have the following properties:
(a) T (mn) = T (m)T (n) if gcd(m, n) = 1;
(b) T (p)T (pr ) = T (pr+1 ) + p2k−1 T (pr−1 ) if p does not divide N ;
(c) T (pr ) = T (p)r , r ≥ 1, if p|N ;
(d) all T (n) commute.
Moreover, by acting on the Fourier expansion, one finds that the first Fourier
coefficient in the expansion of T (n)f is cn . Hence, if f is an eigenfunction of f
with eigenvalue λn , then cn = λn c1 .
Since the Hecke operators commute we can look for the modular functions f
which are eigenfunctions for all of them. Then, the multiplicativity properties
of T (n) imply the corresponding properties for coefficients cn , which leads to a
Euler product formula for L(f, s). This is formalized in the following result.
Theorem 2.15 (Hecke) Let f be a cusp form of weight 2k for Γ0 (N ) that is
simultaneously an eigenvector for all T (n), say T (n)f = λn f, and let
X
f= cn q n , q = e2πiz .
n≥1

Let c1 = 1. Then, (i) coefficients of the series are eigenvalues of the Hecke
operators,
cn = λn
and (ii)
Y 1 Y 1
L (f, s) = −s
.
1 − cp p 1 − cp p + p2k−1−s
−s
p|N p∤N
140 V. Kargin

For example, S12 (Γ0 (1)) has dimension 1, and therefore it is generated by a
single function, which is called the ∆-function:
∞
Y X
∆ (q) = q (1 − q n )24 = τ (n) q n ,
1

where τ (n) is the Ramanujan τ -function. It follows that the L-function associ-
ated to the ∆-function has both a functional equation and the Euler product
property.
In a more general situation, if we wish to find forms that have both a func-
tional equation and a Euler product, then we must overcome the obstacle that
in some exceptional cases operators W and T (n) do not commute. However,
this obstacle can be circumvented and it can be proved that such good modular
forms do exist. They are called primitive forms or newforms.
In summary, the L-functions of primitive forms have both a functional equa-
tion and the Euler product property. As a consequence, one can write explicit
formulas for these L-functions.

2.3.2. L-functions from Maass forms

A nice source for the material in this section is the lecture notes by Liu [33].
A lot of additional information about Maass forms can be found in the book by
Iwaniec [25].
Modular forms are holomorphic and they are not easy to construct or com-
pute. One can try to use Hecke ideas for a different class of functions that satisfy
a modularity condition. In this way one comes to the concept of a Maass form.
Definition 2.16 A smooth function f 6= 0 is called a Maass form for group Γ,
if
(i) for all g ∈ Γ and all z ∈ H, f (gz) = f (z);

(ii) f is an eigenfunction of the non-Euclidean Laplace operator:

 2 
2 ∂ ∂2
−y + 2 f = λf,
∂x2 ∂y
and
(iii) there exists a positive integer N, such that
f (z) ≪ y N , y → ∞.
A Maass form f is said to be a cusp form if the equality
Z 1
f (z + b) db = 0
0

holds for all z ∈ H.

A Maass form f is call odd if f (−x+iy) = −f (x+iy), and even if f (−x+iy) =
f (x + iy).
 Statistics of zetas’ zeros 141

Note that it is relatively easy to generate Maass forms as eigenfunctions of

the Laplace operator on a fundamental domain of the group Γ. By expanding a
Maass form in Fourier series and taking the Fourier coefficients as the coefficients
of a Dirichlet series, one can construct new L-functions. Precisely, let f be either
an even or an odd cusp Maass form with eigenvalue 1/4+r2 . Then, one can write:
√ X
f (x + iy) = y cn Kir (2π |n| y) e2πinx , (18)
n6=0

where Kir are Bessel functions, and we define

X
L (f, s) = cn n−s . (19)
n>0

The key idea here is the fact that the Laplace operator commutes with Hecke op-
erators, and therefore all these operators can be simultaneously diagonalized. By
a computation, the first Fourier coefficient of T (n)f is cn c1 . As a consequence,
L-functions corresponding to Maass forms have a product formula.
What about the functional equation? It holds. However, instead of the stan-
dard formula for the Gamma function one needs the following integral:
Z ∞    
dy s + ir s − ir
Kir (y) y s =Γ Γ .
0 y 2 2
Theorem 2.17 Let f be a Maass form with eigenvalue 1/4 + r2 . Let ε = 0 or
1 depending on whether f is even or odd. Let
   
s + ε + ir s + ε − ir
Λ (f, s) = π −s Γ Γ L (f, s) .
2 2
Then Λ(s, f ) is an entire function that satisfies
ε
Λ (f, s) = (−1) Λ (f, 1 − s) .

2.3.3. Statistical properties of zeros: Global scale

Let  
1 1
S (f, t) := arg L f, + it ,
π 2
where f is the Maass form with an eigenvalue λ and L is the corresponding
L-function. S(f, t) is related to the number of zeros of L in the critical strip in
the same way as the usual S(t) function is related to the number of zeros of
Riemann’s zeta function.
We are interested here in the distribution of S(f, t) with respect to the ran-
dom choice of f .
Of course one need to explain what is meant by the random choice of f.
Let Sj (t) := S(fj , t) where fj has an eigenvalue λj = 14 + rj2 . Define νj (n) :=
p
cj (n)/ cosh πrj , where cj (n) are coefficients in the expansion (18) for the Maass
form fj . The numbers νj (1) will be used as weights in the limiting procedure.
(We assume that fj are normalized to have a unit norm as L2 -functions. There-
142 V. Kargin

√
fore, νj (1) are not necessarily equal to 1.) One knows that νj (1) are O( rj )
and  
1 X 2 1 log T
|ν j (1)| = + O .
T2 π2 T
rj ≤T

Since by the Weyl law, the number of rj below T is proportionate to T 2 , these

weights can be thought as having bounded magnitude and not too sparse. Here
is one of the results about the randomness of Sj (t) (Theorem 3 in [21]).
Theorem 2.18 (Hejhal-Luo) Let h > 0 and t > 0 be fixed. Then we have
X 2 n
1 π 2 |νj (1)| (Sj (t))
lim n/2
= Cn ,
T →∞ (2HT ) 2 (log log T )
|rj −T |≤H

where Cn are moments of the Gaussian distribution.

2.3.4. Local scale

Rudnick and Sarnak [47] extended the results of Montgomery to zeta functions
that arise from modular and Maass forms. In fact, they work in greater gen-
erality and study the zeta functions that arise from the automorphic cuspidal
representations of GLm .The Hecke modular L-functions correspond to the case
m = 2. Their main tool is the following explicit formula, which we formulate for
the case of the Hecke L-functions. Let
Y 1 Y 1
L (s, f ) = −s
1 − c (p) p 1 − c (p) p + p2k−1−s
−s
p|N p∤N
Y
= Lp (s, f ) .

where
1
Lp (s, f ) = ,
(1 − α1 (p)p−s ) (1 − α2 (p)p−s )
with the convention that for p|N one of αi (p) is zero. Let a(pk ) = α1 (p)k +
α2 (p)k , and define b(n) = Λ(n)a(n). Then
X∞
L′ b (n)
=− .
L n=1
ns

Theorem 2.19 (Rudnick and Sarnak) Let b h ∈ Cc∞ (R) be a smooth com-
R
b
pactly supported function, and let h(r) = h(u)eiru du. Then
X Z
log Q ∞
h(γ) = h(r)dr
2π −∞
Z ∞ X Γ′  1   !
1 Γ′ 1
+ h(r) + µj + ir + + µj − ir dr
2π −∞ j
Γ 2 Γ 2
 Statistics of zetas’ zeros 143

X∞  
b(n) b b(n)
− √ h(log n) + √ b h(− log n) , (20)
n=1
n n

where µj are some parameters that depend on the form f, and Q is the conductor
of the form.
By using this result and estimates on the size of coefficients b(n), Rudnick and
Sarnak proved a generalization of the Montgomery theorem. Their result is valid
not only for the Riemann zeta function, but also for Dirichlet L-functions, for
Hecke modular L-functions and for L-functions that correspond to automorphic
cuspidal representations of GL3 . We formulate it for Hecke modular L-functions.
Consider the class of smooth test functions F (x1 , . . . , xn ) that satisfy the
following conditions:
TF 1 F (x1 , . . . , xn ) is symmetric.
TF 2 F (x + t(1, . . . , 1)) = F (x) for all t ∈ R. P
TF 3 F (x) → 0 rapidly as |x| → ∞ in the hyperplane xj = 0.
If BN is a set of N numbers x1 , . . . , xN , then the n-level correlation sum is
defined by
n! X
Rn (BN , F ) = F (S) .
N
S⊂BN
|S|=n

Define the n-level correlation density by

sin πx
Wn (x1 , . . . , xn ) = det (K (xi − xj )) , K (x) = .
πx
Then the following result holds (cf. Theorem 1.2 in [47]).
Theorem 2.20 Assume the Riemann hypothesis for the zeros of L(s, f ). Let
F (x b
1 , . . . , xN ) satisfy TF 1, 2, 3 and in addition assume that F (ξ) is supported
P
in j |ξj | < 1. Then,
Z  
x1 + · · · + xn
Rn (BN , F ) → F (x) Wn (x) δ dx1 . . . dxn
n
as N → ∞.
Rudnick and Sarnak mention that the result can
P probably be proven for func-
tions F with the Fourier transform supported in j |ξj | < 2 by an improvement
of their method, and conjecture that it holds without any assumption on the
support of Fb(ξ).

2.4. Elliptic curve zeta functions

The main source for this section is the book [38] by Milne. Consider an elliptic
curve
E : Y 2 Z = X 3 + aXZ 2 + bZ 3 ,
144 V. Kargin

where a and b are integer, and assume that |∆| = |4a3 + 27b2 | cannot be made
smaller by a change of variable X → X/c2 , Y → Y /c3 . This equation is called
minimal. The equation
E : Y 2 Z = X 3 + aXZ 2 + bZ 3 ,
with a and b the images of a and b in Fp (the finite field with p elements) is
called the reduction of E modulo p. (It is assumed here that p 6= 2, 3. In the
case when p is 2 or 3, a somewhat different notion of the minimal equation is
needed.) Let Np is the number of solutions of this equation in Fp .
There are four possible cases:
(a) Good reduction. E is an elliptic curve. (That is, the determinant does
not vanish and therefore the curve is smooth.) This happens if p 6= 2 and
p does not divide ∆.
(b) Cuspidal, or additive, reduction. This is the case in which the reduced
curve has a cusp. For p 6= 2, 3, this case occurs exactly when p|4a3 + 27b2
and p| − 2ab.
(c) Nodal, or multiplicative, reduction. The reduced curve has a node.
For p 6= 2, 3, it occurs exactly when p|4a3 + 27b2 , p ∤ −2ab.
(c1) Split case. The tangents at the node are rational over Fp . This
happens when −2ab is a square in Fp .
(c2) Non-split case. The tangents at the node are not rational over
Fp .This occurs when −2ab is not a square in Fp .
The names additive and multiplicative refer to the group of points on the
reduced curve, which in these cases is isomorphic either to (Fp , +), or (F∗p , ×).
We define the L function associated with the elliptic curve E as follows.
Y 1
L (s, E) := −s )
.
p
L p (p

Here, the local factors Lp (T ) are defined as follows:



 1 − ap T + pT 2 , if p is good, with ap = p + 1 − Np ,

1−T if E has split multiplicative reduction,
Lp (T ) =

 1+T if E has non-split multiplicative reduction,

1 if E has additive reduction.
Let S be the (finite) set of primes with bad reduction. Then we can also write
Y 1 Y 1
L (s, E) := −s
Lp (p ) 1 + (Np − p − 1) p−s + p1−2s
p∈S p∈S
/
Y 1 Y 1
= −s
.
Lp (p ) (1 − αp p−s ) (1 − βp p−s )
p∈S p∈S
/

The Hasse-Weil conjecture says that L(s, E) can be analytically continued to

a meromorphic function on the whole of C and satisfies a functional equation.
 Statistics of zetas’ zeros 145

A recent work by Wiles and others confirmed this conjecture by showing that all
elliptic curves are “modular”, in particular, their L-functions arise from modular
forms. To a certain extent, this result reduces the study of the elliptic L-functions
to the study of the Hecke modular L-functions.
√
It is known that the numbers ap do not exceed 2 p in absolute value. For
a fixed elliptic curve and different primes p, these numbers are believed to be
√ √
distributed on the interval [−2 p, 2 p] according to the semicircle distribution
but this is not proven. In fact, this conjecture is related to the Birch-Swinnerton
conjecture that states that
r
L (s, E) ∼ C (s − 1) as s → 1,
where r is the rank of the group of rational points on E, and C is a certain
predicted constant. (see Chapter 10 in [38] for more information.)
It is possible to construct zeta functions for other nonsingular projective va-
rieties and the conjecture by Hasse and Weil states that these zeta functions
satisfy a functional equation (and the Riemann hypothesis). However, appar-
ently not much is known beyond the cases of projective spaces and elliptic
curves.

3. Selberg’s zeta functions for compact and non-compact manifolds

It is useful to keep in mind that we will now talk about a new type of zeta
functions, which is significantly different from number-theoretical zeta functions.
While there is an explicit formula, it relates Laplace eigenvalues and geodesics,
not zeta zeros and primes. The possibility of a relation between these two types
of zetas is only conjectural.
The main source for this section is Hejhal’s book [19].

3.1. Selberg’s zeta function and trace formula

Let M be a compact Riemann surface of genus g ≥ 2. Then, M can be identified

with a quotient space Γ\H, where H is the upper half-plane and Γ is a discrete
subgroup of SL2 (R). We assume that H has the Poincare metric |dz|/y with
the area element dxdy/y 2 , and therefore it has the constant negative curvature.
This metric is naturally projected on the surface M.
This is not the most general situation of interest since most of the quotient
spaces Γ\H occurring in arithmetic applications have cusps and therefore are
non-compact. However, the theory is most clear and transparent for the compact
surfaces.
The Laplace operator on M can be defined by the following formula.
−∆ : u → −y 2 (uxx + uyy ) .
It can be shown that this operator has a discrete set of non-positive eigenvalues:
0 = λ0 < λ1 ≤ λ2 ≤ . . . ,
and the only point of accumulation of these eigenvalues is ∞.
146 V. Kargin

Let us define  q

 λn − 41 , if λn ≥ 14 ,
rn = q

 i −λn + 1 , if λn < 1/4,
4

so that λn = 41 + rn2 .
Also let m = max{k : λk < 1/4}.
Let G(M ) be the set of all closed geodesics on M , and let P(M ) be the
subset of all prime closed geodesics (that is, the closed geodesics that cannot
be represented as a non-trivial multiple of another closed geodesic). It is known
that G(M ) is a countable set, which we can order by the lengths of its elements.
Closed geodesics correspond to hyperbolic elements of the group Γ (that is,
the elements of Γ with the trace outside of [−2, 2]) up to conjugacy of these
elements. If P ∈ Γ corresponds to a geodesic γ, then γ is prime if and only if
there is no P0 ∈ Γ such that P = P0k for an integer k > 1.
If l(γ) denotes the length of the geodesic γ, corresponding to P ∈ Γ, then we
set
|γ| := el(γ) ,
and note that
|γ|1/2 + |γ|−1/2 = |TrP | .
We will also write |γ| = N [P ] (meaning norm of P ).
The Selberg trace formula relates sums over eigenvalues λk to sums over
hyperbolic elements (geodesics) [P ]. Let h(u) be a function which (i) is analytic
in the strip |Imu| ≤ 1/2 + δ, (ii) is even: h(u) = h(−u), and (iii) declines
sufficiently fast in the strip: |h(u)| = O((1 + |Reu|)−2−δ ).
R
Let bh(t) = 2π1
h(u)e−itu du. Then the Selberg trace formula holds (cf. The-
orem I.7.5 in Hejhal [19]),
∞
X Z
µ (F )
h (rn ) = rh (r) tanh (πr) dr (21)
n=0
2π R
X ln N [T0 ]
+ b
h (ln N [T ]) ,
1/2 −1/2
[T ] N [T ] − N [T ]

where the sum is over all distinct conjugacy classes of hyperbolic elements [T ],
[T0 ] is the primitive element for T, T = T0k , and µ(F ) is the area of the funda-
mental region of the group Γ.
It is instructive to compare this formula with formula (20). Since (21) resem-
bles the explicit formulas from number theory, it is natural to define Selberg’s
zeta function as follows (cf. Definition II.4.1 in [19]):

Y ∞ 
Y 
−s−k
Z (s) = 1 − |γ| , Res > 1. (22)
γ∈P(M) k=0
 Statistics of zetas’ zeros 147

It turns out that Selberg’s zeta function is closely related to the eigenvalues
of the Laplace operator on M (cf. Theorem II.4.10 and II.4.11 in [19]).
Theorem 3.1 (Hejhal-Selberg) (a) Z(s) is an entire function;
(b) Let β be a real number ≥ 2. For all s, the following identity holds:

∞
" #
1 Z ′ (s) 1 Z ′ 12 + β X 1 1
=
+
− 2
2s − 1 Z (s) 2β Z 21 + β 2
n=0 rn + s − 2
1 2 rn + β 2
∞  
µ (F ) X 1 1
+ 1 − .
2π β+ 2 +k s+k
k=0

(c) Z(s) has “trivial” zeros s = −k, k ≥ 1, with multiplicity (2g − 2)(2k + 1);
(d) s = 0 is a zero of multiplicity 2g − 1;
(e) s = 1 is a zero of multiplicity 1;
(f ) the nontrivial zeros of Z(s) are located at 12 ± irn .
Since all but a finite number of eigenvalues are greater than 1/4 hence all
but a finite number of rn is real and therefore the claim (f) implies that all but
a finite number of zeros of Z(s) are located on the line Imz = 1/2.
The formula in claim (b) of this theorem follows from Selberg’s trace formula
and it can be thought as a functional equation for the logarithmic derivative
of Z(s). In particular, it implies the functional equation for the zeta functions
itself (cf. Theorem 4.12 in [19]).
Theorem 3.2 (Hejhal-Selberg) Selberg’s zeta function satisfies the following
functional equation:
" Z s− 12 #
Z (s) = Z (1 − s) exp µ (F ) v tan (πv) dv .
0

3.2. Statistics of zeros

The number of zeta zeros in a long interval has the following asymptotic expres-
sion:
µ (F ) 2
N [k : 0 ≤ rk ≤ T ] = T + S (T ) + E (T ) ,
4π
where  
1 1
S (T ) = arg Z + iT ,
π 2
and Z T
E (T ) = O (1) = 2c t [tanh (πt) − 1] dt − (m + 1) .
0

In other words, the number of zeros in a unit interval is ∼ cT. In comparison,

for Riemann’s zeta function we have ∼ c log T zeros in the unit interval.
148 V. Kargin

It is known (cf. Theorems 8.1 and 17.1 in [19]) that

  " 1/2 #
T log T
S (T ) = O , and S (T ) = Ω± .
log T log log T

(Recall that the notation f (x) = Ω+ [g(x)] means that lim sup( fg(x)
(x)
) > 0, and
f (x) = Ω− [g(x)] means that lim inf( fg(x)
(x)
) < 0.)
It was found that it is difficult to generalize the results concerning the mo-
ments of Riemann’s S(x) function to the case of Selberg’s zeta. Since these
results are essential for the study of statistical properties of zeta zeros, there is
a stumbling block here.
Selberg managed to resolve this problem partially for a particular choice of
the group Γ.
Let p ≥ 3 be a prime and A be a quadratic non-residue modulo p. Define
 √ √ √  
y0 + y1 A y2 p + y3√ Ap
Γ = Γ (A, p) = √ √ ; y0 , y1 , y2 , y3 are integer.
y2 p − y3 Ap y0 − y1 A

and call it a quaternion group.

Let S(t) = S + (t) − S − (t), where S + (t) = max{0, S(t)} and S − (t) = max{0,
−S(t)}. Then the following theorem holds (cf. Theorem 18.8 in [19]).
Theorem 3.3 (Hejhal-Selberg) Let Γ = Γ(A, p) with p ≡ 1 (mod4). Then
(for large T ):
Z
1 qT + 2 T
S (t) dt ≥ c1 2
T T (log T )
where c1 is a positive constant that depends only on Γ. A similar inequality holds
for S − (t).
In order to appreciate this result note that it suggests that √ the average de-
viation of S(T ) from its mean is of the order larger than T /(log T ) which
should be compared with the number of zeros in the interval [0, T ], that is, cT 2 .
In other words, the deviation is larger than (N (T ))1/4−ε . To put it in prospec-
tive note that the average deviation of the zeros of S(T ) for Riemann’s zeta
function is of the order (log log T )1/2 which is smaller than log log N (T ), where
N (T ) ∼ cT log T is the number of Riemann’s zeros in [0, T ]. These situations
appear to be quite different.
Moreover, recently there was some numeric and theoretical work on the eigen-
values of the Laplace operator on manifolds Γ\H for arithmetic groups Γ. First,
numeric and heuristic analysis showed that the spacings between eigenvalues re-
semble spacings between points from a Poisson point process rather than spac-
ings between eigenvalues of a random matrix ensemble (see Bogomolny et al. [5]
and references wherein). Next some rigorous explanations of this finding have
been given that relate it to large multiplicities of closed geodesics with the same
length. See Luo and Sarnak ([35] and [36]) and Bogomolny et al. [6].
 Statistics of zetas’ zeros 149

There is also some work on correlations of closed geodesics – see Pollicott

and Sharp [43].

3.3. Comparison with the circle problem

The Selberg zeta function is closely related to counting geodesics on a space Γ\H,
in the same way as the Riemann zeta function is related to counting primes.
It is natural to look at Laplace eigenvalues and geodesic counting problem in a
simpler situation, such as a compact Riemann surface of genus 1. Such a surface
can be represented as a quotient space Λ\C, where Λ is a lattice. Consider,
for concreteness, Λ = [1, i]. Then the eigenvalues of the Laplace operator are
4π 2 (m2 + n2 ), where m and n are integer, and the number of the eigenvalues
below t equals the number of integer points in the circle t/π. Let

r (n) = N (a, b) ∈ Z × Z : a2 + b2 = n ,

and X
A (x) = r (n) = πx + R (x) .
0≤n≤x

The function A(x) can be thought as the counting function both for eigen-
values of the Laplace operator and for closed geodesics of bounded length.
Then by using the Poisson summation formula it is possible to derive the
following result (cf. Theorem 4.1 in [20] and Theorem 559 in [32]).
X X Z ∞
√

r (n) f (n) = π r (n) f (x) J0 2π nx dx.
0

Informally, if one uses this identity with the indicator function for f (x) (which
is, in fact, not allowed under the conditions of the theorem), then one obtains
the following formula (cf. formula (4.10) in [20] )
∞
√ X r (n) √

R (x) = x √ J1 2π nx .
n=1
n

Rigorous variants of this formula lead to various estimates on R(x), in particular

it is known (cf. Theorems 509, 542, and 548 in [32]) that
   
R = O x1/3 and R = Ω± x1/4 ,

and that Z x h i
1
R (t)2 dt = cx1/2 + O (log x)3 .
x 0

This suggest that the “standard deviation” of R(t) is x1/4 . Similar to the case
with Laplacian eigenvalues on Γ\H, the statistical behavior of eigenvalues does
not resemble the behavior of random matrix eigenvalues or Riemann’s zeta zeros.
150 V. Kargin

Some more details about this problem can be found in [29], which considers
the question about the number of points inside a random circle. More recent
research can be found in [18], where it is shown that the distribution of the
error term R(x) converges to a non-Gaussian distribution as x → ∞, and in [4],
where this result is extended to circles with the center at a point (α, 0), and it
is shown that the nature of the resulting distribution depends strongly on α.

4. Zeta functions of dynamical systems

Dynamical zeta functions are closely related to Selberg’s zeta function which
can be thought as a dynamical zeta function for the geodesic flow on a Riemann
surface. At the same time, there is a connection to number-theoretical zeta func-
tions, namely, to the zeta functions of curves over finite fields. The main sources
for this section are reviews by Ruelle ([48] and [49]) and Pollicott ([44] and [45]).

4.1. Zetas for maps

Let f be a map of a set M to itself, let the periodic orbits of f be denoted by

P, and let |P | denote the period of the orbit P. Then, we can define the zeta of
f by the following formula:
Y −1
ζ (z) = 1 − z |P | (23)
P
X∞
zm
= exp |Fix f m | ,
m=1
m
where |Fix f m | denote the number of fixed points of f m .

4.1.1. Permutations

Let M be a finite set, and let f be given by a permutation matrix A. Then the
number of fixed points of f m is given by TrAm . Hence, we have
X∞ m
(zA)
ζ (z) = exp Tr
m=1
m
= exp (−Tr log (1 − zA))
= 1/ det (1 − zA) ,
which is closely related to the characteristic polynomial of matrix A. In partic-
ular, all poles of the zeta are on the unit circle.

4.1.2. Smooth mappings of compact manifolds

Let f be a differentiable mapping of a compact orientable smooth manifold X

to itself. Assume that that f is non-singular at all fixed points. Recall that the
degree of f at a fixed point x equals to +1 if the map preserves orientation
 Statistics of zetas’ zeros 151

at the fixed point, and to −1 if it reverses the orientation, that is, degx (f ) :=
sign det(dfx − I). We define the Lefschetz zeta function as
X∞ X
zm
ζL (z) = exp dx (f m ) .
m=1
m
x∈Fix(f m )

In this case one can use the Lefschetz fixed point formula that says:
X dim
XM i
dx (f m ) = (−1) Tr ((f m )∗i : Hi → Hi ) ,
x∈Fix(f m ) i=0

where Hi is the i-th homology group of the compact manifold M with real
coefficients, and (f m )∗i is the map induced by f m on Hi .
In particular, if λij are eigenvalues of f∗i , then we get

X∞ dim M dim
XHi
1 X
ζL (z) = exp (−1)i (zλij )m
m=1
m i=0 j=1
 (−1)i
dim
YM dim
YHi
 −1 
= (1 − zλij ) ,
i=0 j=1
dim
YM (−1)i+1
= det (1 − zf∗i )
i=0

which is a rational function. If the map f is a complex-analytic map of two

complex compact manifold then this calculation can be refined by using the
holomorphic Lefschetz formula that relates a sum over the fixed points of such
a map to a sum over its Dolbeault cohomology groups. This often leads to
additional information about λij .
As an example, let M be a torus R2 /Z2 and let f be induced by a linear
transformation A ∈ SL2 (Z). Assume that the eigenvalues
P of A are positive and
not on the unit circle: λ1 > 1 > λ2 > 0. Then x degx (f m ) = det(Am − I).
Hence, we have
X∞
zm
ζL (z) = exp det (Am − I)
m=1
m
(1 − zλ1 ) (1 − zλ2 )
=
(1 − z) (1 − zλ1 λ2 )
(1 − zλ1 ) (1 − zλ2 )
= 2 .
(1 − z)
The original dynamical zeta of continuous maps (in which fixed points are
counted without taking into account the degree of f m ) is often called the Artin-
Mazur zeta function (see Artin-Mazur [1]). If this map is a diffeomorphism (a bi-
jection smooth in both directions) and if it satisfies some additional conditions
152 V. Kargin

(hyperbolicity or Axiom A), then it is known that this function is rational.

(This was conjectured by Smale [55], and proved by Guckenheimer [16] and
Manning [37].)

4.1.3. Subshifts

Suppose next that A is an N -by-N matrix of zeros and ones, and that the set M
consists of doubly infinite sequences {xi } of symbols 1, . . . , N, which satisfy the
following criterion. A sequence {xi } belongs to M if and only if Axi xi+1 = 1 for
every i. In other words, the symbol xi determines which of the other symbols
are possible candidates for xi+1 . The map f is simply a shift on this set M :
{xi } → {xi+1 }. In this case, the number of fixed points of f m is Tr(Am ), and
we have
ξ (z) = 1/ det (1 − zA) . (24)

4.1.4. Ihara’s zeta function

A basic reference for this section is a book by Terras [58].

Let G be a finite graph. Ihara’s zeta function of G is a dynamical zeta function
for a subshift associated with this graph. Namely, orient edges of G arbitrarily.
Let the 2|E| oriented edges be denoted e1 , e2 , . . . , en , en+1 = e−1 −1
1 , . . . , e2n = en .
The subshift matrix is a 2n-by-2n edge adjacency matrix WG which is defined
as follows.
Definition 4.1 The edge adjacency matrix WG is a 2n-by-2n matrix with the
rows and columns corresponding to oriented edges such that its (i, j) entry equals
1 if the terminal vertex of edge i equals the initial vertex of edge j and edge j is
not the inverse of edge i.
In particular, from (24) we have a determinantal formula:
−1
ζG (u) = det (I − uWG ) .
It is possible to define Ihara’s zeta function more directly. The finite points
of f m in this example are closed non-backtracking tailless paths of length m,
where a path is a sequence of oriented edges such that the end of one edge equals
the beginning of the next edge. A path (e1 , e2 , . . . , em ) is closed if the end of em
corresponds to the beginning of e1 . It is non-backtracking if ei+1 6= e−1 i for any
i, and it is tailless if em 6= e−1
1 .
Hence, according to (23),
Y −1
ζG (u) = 1 − ul(P ) , (25)
[P ]

where the product is over all primes, that is, all equivalence classes of primitive
closed non-backtracking tailless paths. (A closed path P is primitive if P 6= Dm
for m ≥ 2 and any other path D; and two paths are called equivalent if they
can be obtained from each other by a cyclic permutation of edges.)
 Statistics of zetas’ zeros 153

For arbitrary regular finite graphs, this definition of Ihara’s zeta function was
introduced by Sunada ([56] and [57])) following a suggestion in the book “Trees”
by Serre.
Ihara’s zeta function has another representation as a determinant which
was first discovered by Ihara for regular graphs [24] and then by Bass [2] and
Hashimoto [17] for arbitrary finite graphs:
−1
|E|−|V |

ζG (u) = 1 − u2 det I − AG u + QG u2 ,
where AG is the adjacency matrix of G, and QG is the diagonal matrix whose
j-th diagonal entry is (−1 + degree of j-th vertex).
If the graph is q + 1 regular, that is, if every vertex has degree q + 1, then
QG is scalar and we can see that poles of ζG (u) are related to the zeros of the
characteristic polynomial of AG , that is to the eigenvalues of the matrix AG.
Precisely, the poles ui are related to the eigenvalues λi by the formula:
p
λi ± λ2i − 4d
ui = .
2d
Since the√eigenvalues are always real, we find that a pole ui is on the circle
|u| = 1/ √ d if and only if the corresponding eigenvalue is sufficiently small,
|λi | ≤ 2 d.
There are two trivial poles at 1 and 1/q corresponding to the largest eigen-
value λ = d + 1. The Riemann hypothesis for regular graphs says that all non-
trivial poles are on this circle. This is not always true, and it holds if and only if
√
|λ1 | ≤ 2 q, where λ1 is the second largest in magnitude eigenvalue of AG . The
graphs that satisfy this condition are often called Ramanujan graphs following
a paper by Lubotsky, Phillips, and Sarnak [34], which constructed an infinite
family of such graphs by using the Ramanujan conjecture from the theory of
modular forms.
A random regular graph is approximately Ramanujan with high probability.
√
This means that for arbitrary ε > 0, the probability that |λ1 | ≥ 2 q +ε becomes
arbitrarily small as the size of the graph grows. (This is known as Alon’s con-
jecture, and was proved in a lengthy paper by Friedman [12]). The distribution
properties of the largest eigenvalue are still unknown. It is also unknown what
√
proportion of the eigenvalues exceed the threshold 2 q.
In his Ph.D. thesis [42], Derek Newland studied spacings of eigenvalues of
random regular graphs and spacings of the Ihara zeta zeros and found numer-
ically that they resemble spacings in the Gaussian Orthogonal Ensemble. See
also an earlier paper by Jacobson, Miller, Rivin and Rudnick [27].
For Ihara’z zeta function, there is an analog of Selberg’s trace formula which
we formulate for the case of a q + 1 regular graph.
First, note that formula (25) implies
ζ′ X  
u G (u) = l (P ) ul(P ) + u2l(P ) + · · ·
ζG
[P ]
X
= Nm u m ,
154 V. Kargin

where Nm be the number of all non-backtracking tailless closed paths of length m.

Then the following explicit formula holds (cf. Proposition 25.1 in [58]).
Theorem 4.2 (Terras) Suppose 0 < a < 1/q. Assume that h(u) is mero-
morphic in the plane and holomorphic for |u| > a − ε, ε > 0 Assume that
R
h(u) = O(|u|−1−α ), α > 0. Let b 1
h(n) := 2πi n
|u|=a u h(u)du and assume that
b
h(n) decays rapidly enough. Then,
X X
ρh (ρ) = Nn bh (n) ,
ρ n≥1

where the sum on the left is over the poles of ζG (u).

The book [58] mentions that this formula can be used to derive the limit law
for eigenvalues of a large random regular graph (the McKay-Kesten law) and
give references to this and some other applications.

4.1.5. Frobenius maps

Let M be the set of solutions of a system of algebraic equations in r variables over

the algebraic closure of the finite field Fq and let f be the Frobenius map Frob:
(x1 , . . . , xr ) → (xq1 , . . . , xqr ). In the case of curves, the dynamic zeta function
defined by (23) is equivalent to a number-theoretic zeta function introduced by
Artin.
Namely, let an affine curve C be given by the equation f (X, Y ) = 0 over the
finite field Fq . Let p denote a prime ideal of the field Fq [X, Y ]/f (X, Y ) and let
the order of Fq [X, Y ]/p be denoted by N p. Then, by analogy with Riemann’s
zeta function we can define
Y 1
ζC (s) = .
p
1 − N p−s

It turns out that if one uses the change of variable u = q −s , then this function
is equivalent to the dynamical zeta function for the Frobenius map acting on
the algebraic closure of the curve C.
Here is an example. Let C be an elliptic curve, then
1 − u (q + 1 − N1 ) + u2 q
ζC (u) = ,
(1 − u) (1 − uq)
where N1 is the number of the points of C in Fq . One proof of this fact is based
on the Riemann-Roch theorem that allows to count the ideals with a given
norm more or less explicitly. (See [38].) Another proof uses an analogy with
the smooth maps of compact manifolds. It proceeds by constructing a theory
of cohomologies for algebraic varieties over finite fields, which has a suitable
Lefschetz fixed point formula. See the book by Silverman [54] for more details
about this proof.
The Riemann hypothesis in this example is equivalent to the statement that
√
|N1 − q − 1| ≤ 2 q,
 Statistics of zetas’ zeros 155

since this implies that the roots of the polynomial in the numerator are on the
circle with radius q −1/2 . For elliptic curves this was proved by Hasse.
In [60] Weil conjectured that the zeta function of every algebraic variety over
finite fields is rational, that it has a functional equation, and that it satisfies the
Riemann hypothesis. The proof of this general conjecture led to an introduction
of many new ideas in algebraic geometry by Dwork, Grothendick, and Deligne.
The statistical properties of the zeta’s zeros were investigated in the case
of curves over finite fields by Katz and Sarnak [28]. They have studied zeros’
distribution when the genus of the corresponding curve grows to infinity and
found that for “most” of the curves the local statistics of the zeros approach
those of the eigenvalues of random matrix ensembles.

4.1.6. Maps of an interval

For yet another example consider the map x → 1 − µx2 of the interval [−1, 1]
to itself. For a special value of µ ≈ 1.401155 . . . (the Feingenbaum value), this
map has one periodic orbit of period 2n for every integer n ≥ 0. Therefore, for
the dynamic zeta function we have:
∞ 
Y −1 ∞ 
Y n+1
n n
ζ (z) = 1 − z2 = 1 + z2 .
n=0 n=0

This ζ satisfies the functional equation ζ(z 2 ) = (1 − z)ζ(z). More generally,

the piecewise monotone maps of the interval [−1, 1] to itself correspond to the
Milnor-Thurston zeta functions. See [39]. Apparently, so far there have been no
systematic study of the statistical properties of their poles and zeros.

4.2. Zetas for flows

If f is a flow on M, that is, a map M × R+ → M, then we can define the zeta

function of this flow as
Y −1
ζ (s) = 1 − e−sl(ω) ,
ω

where ω denotes a periodic orbit of f, and l(ω) is its length. It is a more general
case than the case of maps since there is a construction (“suspension”) that
allows us to realize maps as flows (see Example 5 on p. 70 in [46] or Section 4.3
in [45]) but not vice versa. Unsurprisingly, it turns out that zeta functions for
flows are more difficult to investigate than zeta functions for maps.
If we imagine that prime numbers correspond to periodic orbits of a flow and
that the length of the orbit indexed by p is given by log p, then the zeta function
of the flow will coincide with Riemann’s zeta function.
One particularly important example of a flow is the geodesic flow on a smooth
manifold M. In the case when M has a constant negative curvature, the corre-
sponding dynamical zeta function is closely related to the Selberg zeta function.
156 V. Kargin

Namely,
Y∞
Z (s + 1)
ζ (s) = and Z (s) = ζ (s + n)−1 ,
Z (s) n=0
where Z(s) is as defined in (22) (see Remark 2.5 in [45]). Another important
example is the geodesic flow for billiards on polygons.
I am not aware about the functional equation for zeta functions that comes
from general geodesic flows (other than flows on manifolds of a constant nega-
tive curvature). Also, it appears that not much is known about the statistical
properties of the distribution of zeros and poles of these zeta functions.
On the other hand one can study the location of the pole with the maxi-
mal real part and the results of this study give valuable information about the
distribution of closed geodesics. In this way, one can study the distribution of
closed geodesics on spaces of variable curvature (see Corollary 6.11 in [45]).

5. Conclusion

We considered the statistical properties of various zeta functions. For the zeros
of number-theoretical zeta functions, the main observation is that they satisfies
many properties which are true for eigenvalues of random matrices. The main
outstanding problem (besides the Riemann hypothesis) is to push this similarity
to its natural limits and, in particular, show that the Montgomery conjecture
about the correlations of the Riemann zeros is true.
Next, we observed that for some of groups Γ ⊂ SL2 (Z), the statistical prop-
erties of Selberg’s zeta zeros are different from those of the random matrix
eigenvalues. The exact description of these properties is not known.
The statistical properties of the zeros of dynamical zeta functions are not
investigated in many cases. Two notable exceptions are the zeta functions of
curves over finite fields and Ihara’s zeta functions. However, even in this case
there are many unsolved problems. For example, it is not known whether the
distribution of local statistics for the zeros of Ihara’s zeta functions coincide
with the corresponding distribution for the random matrix eigenvalues.

References

[1] Artin, M. and Mazur, B., On periodic points. Annals of Mathematics,

81:82–99, 1965. MR0176482
[2] Bass, H., The Ihara-Selberg zeta function of a tree lattice. International
Journal of Mathematics, 3:717–797, 1992. MR1194071
[3] Biane, P., Pitman, J., and Yor, M., Probability laws related to the Ja-
cobi theta and Riemann zeta functions, and Brownian excursions. Bulletin
of American Mathematical Society, 38:435–465, 2001. MR1848256
[4] Bleher, P. M., Chang, Z., Dyson, F. J., and Lebowitz, J. L.,
Distribution of the error term for the number of lattice points inside a
shifted circle. Communications in Mathematical Physics, 154:433–469,
1993. MR1224087
 Statistics of zetas’ zeros 157

[5] Bogomolny, E., Georgeot, B., Giannoni, M.-J., and Schmit, C.,
Chaotic billiards generated by arithmetic groups. Physical Review Letters,
69:1477–1480, 1992. MR1178884
[6] Bogomolny, E., Leyvraz, F., and Schmit, C., Distribution of eigen-
values for the modular group. Communications in Mathematical Physics,
176:577–617, 1996. MR1376433
[7] Bourgade, P., Mesoscopic fluctuations of the zeta zeros. Probability
Theory and Related Fields, 148:479–500, 2010. MR2678896
[8] Conrey, J. B., The Riemann hypothesis. Notices of American Mathe-
matical Society, 50:341–353, 2003. MR1954010
[9] Davenport, H., Multiplicative Number Theory. Markham Publishing Co.,
1967. MR0217022
[10] Diaconis, P. and Evans, S. N., Linear functionals of eigenvalues
of random matrices. Transactions of American Mathematical Society,
353(7):2615–2633, 2001. MR1828463
[11] Diaconis, P. and Shahshahani, M., On eigenvalues of random matrices.
Journal of Applied Probability, 31:49–62, 1994. MR1274717
[12] Friedman, J., A Proof of Alon’s Second Eigenvalue Problem, volume 195
of Memoirs of American Mathematical Society. AMS, Providence, RI, 2008.
MR2437174
[13] Fujii, A., Explicit formulas and oscillations. In Emerging Applications of
Number Theory, volume 109 of The IMA Volume in Mathematics and Its
Applications, pages 219–267. Springer-Verlag, New York, 1999. MR1691535
[14] Goldston, D. A., Notes on pair correlations of zeros and prime numbers.
In F. Mezzadri and N. C. Snaith, editors, Recent Perspectives in Random
Matrix Theory and Number Theory, volume 322 of London Mathematical
Society Lecture Notes, pages 79–109. Cambridge University Press, 2005.
MR2166459
[15] Goldston, D. A., Large differences between consequitive prime numbers.
PhD thesis, University of California, Berkeley, 1981. available at http://
www.math.sjsu.edu/∼goldston/thesis81.pdf. MR2631517
[16] Guckenheimer, J., Axiom A + no cycles → ζf (t) is rational. Bulletin of
American Mathematical Society, 76:592–594, 1970. MR0254871
[17] Hashimoto, K., Zeta functions of finite graphs and representations of
p-adic groups. In Advanced Studies in Pure Mathematics, volume 15, pages
211–280. Academic Press, N.Y., 1989. MR1040609
[18] Heath-Brown, D. R., The distribution and moments of the error term
in the Dirichlet divisor problem. Acta Arithmetica, 60:389–414, 1992.
MR1159354
[19] Hejhal, D. A., The Selberg Trace Formula for P SL(2, R), volume 548 of
Lecture Notes in Mathematics. Springer, 1976. MR0711197
[20] Hejhal, D. A., The Selberg trace formula and the Riemann zeta function.
Duke Mathematical Journal, 43:441–482, 1976. MR0414490
[21] Hejhal, D. A. and Luo, W., On a spectral analogue of Selberg’s result
on s(t). International Mathematics Research Notices, 1997:135–151, 1997.
MR1434906
158 V. Kargin

[22] Hughes, C. and Rudnick, Z., Linear statistics for zeros of Riemann’s
zeta function. C. R. Acad. Sci. Paris, Ser. I 335:667–670, 2002. available
at . MR1941646
[23] Hughes, C. and Rudnick, Z., Linear statistics of low-lying zeros of
l-functions. Quaterly Journal of Mathematics, 54:309–333, 2003. available
at . MR2013141
[24] Ihara, Y., On discrete subgroups of the two by two projective linear group
over p-adic fields. Journal of Mathematical Society of Japan, 18:219–235,
1966. MR0223463
[25] Iwaniec, H., Spectral Methods of Automorphic Forms, volume 53 of Grad-
uate Studies in Mathematics. American Mathematical Society, second edi-
tion, 2002. MR1942691
[26] Iwaniec, H. and Kowalsky, E., Analytic Number Theory, volume 53
of Colloquium Publications. American Mathematical Society, first edition,
2004. MR2061214
[27] Jacobson, D., Miller, S. D., Rivin, I., and Rudnick, Z., Eigenvalue
spacings for regular graphs. IMA Volumes in Mathematics and Its Ap-
plications, 109:317–327, 1999. Available at http://www.math.rutgers.edu/
∼sdmiller/ima.pdf. MR1691538
[28] Katz, N. M. and Sarnak, P., Random Matrices, Frobenius Eigenval-
ues, and Monodromy, volume 45 of Colloquium Publications. American
Mathematical Society, 1999. MR1659828
[29] Kendall, D. G., On the number of lattice points inside a random oval.
Quarterly Journal of Mathematics, 19:1–26, 1948. MR0024929
[30] Knapp, A. W., Elliptic Curves. Princeton University Press, Princeton,
NJ, first edition, 1992. MR1193029
[31] Korolev, M. A., On the Gram’s law in the theory of Riemann zeta
function. arxiv:1011.3997, 2010.
[32] Landau, E., Vorlesungen über Zahlentheorie, volume 2. Chelsea Publish-
ing Company, 1947. Reprint of 1927 edition.
[33] Liu, J., Lectures on Maass forms. available at http://www.prime.sdu.edu.
cn/lectures/LiuMaassforms.pdf, 2007.
[34] Lubotzky, A., Phillips, R., and Sarnak, P., Ramanujan graphs. Com-
binatorica, 8:261–277, 1988. MR0963118
[35] Luo, W. and Sarnak, P., The behavior of eigenstates for arithmetic
hyperbolic surfaces. Communications in Mathematical Physics, 161:195–
213, 1994. MR1266491
[36] Luo, W. and Sarnak, P., Number variance for arithmetic hyperbolic
surfaces. Communications in Mathematical Physics, 161:419–432, 1994.
MR1266491
[37] Manning, A., Axiom A diffeomorphisms have rational zeta functions.
Bulletin of London Mathematical Society, 3:215–220, 1971. MR0288786
[38] Milne, J. S., Elliptic Curves. BookSurge Publishers, 2006. MR2267743
[39] Milnor, J. and Thurston, W., On Iterated Maps on the Interval, volume
1342 of Lecture Notes in Mathematics. Springer, 1988. MR0970571
 Statistics of zetas’ zeros 159

[40] Montgomery, H. L., The pair correlation of the zeta function. Proc.
Symp. Pure Math, 24:181–193, 1973. MR0337821
[41] Moreno, C. J., Advanced Analytic Number Theory: L-functions, volume
115 of Mathematical Surveys and Monographs. American Mathematical
Society, 2005. MR2135107
[42] Newland, D. B., Kernels in the Selberg trace formula on the k-regular tree
and zeros of the Ihara zeta function. PhD thesis, University of California,
San Diego, 2005.
[43] Pollicott, M. and Sharp, R., Correlations for pairs of closed geodesics.
Inventiones Mathematicae, 163:1–24, 2006. MR2208416
[44] Pollicott, M., Dynamical zeta functions. Available at http://
homepages.warwick.ac.uk/∼masdbl/leshouches.pdf, 2003.
[45] Pollicott, M., Dynamical zeta functions and closed orbits for geodesic
and hyperbolic flows. Available at http://homepages.warwick.ac.uk/
∼masdbl/leiden-lectures.pdf, 2010. MR3054430
[46] Pollicott, M. and Yuri, M., Dynamical Systems and Ergodic Theory,
volume 40 of London Mathematical Society student texts. Cambridge Uni-
versity Press, 1998. MR1627681
[47] Rudnick, Z. and Sarnak, P., Zeros of principal L-functions and random
matrix theory. Duke Mathematical Journal, 81:269–322, 1996. MR1395406
[48] Ruelle, D., Dynamical zeta functions: where do they come from and
what are they good for? In K. Schmadgen, editor, Mathematical Physics
X, pages 43–51. Springer, Berlin, 1992. Available at http://www.ihes.fr/
∼ruelle/Publications.html. MR1386396
[49] Ruelle, D., Dynamic zeta functions and transfer operators. Notices of
American Mathematical Society, 49:887–895, 2002. Available at http://
www.ihes.fr/∼ruelle/Publications.html. MR1920859
[50] Selberg, A., Contributions to the theory of Dirichlet’s L-functions. In
Collected Papers of A. Selberg, volume 1, pages 281–340, 1989. The year of
the original publication is 1946.
[51] Selberg, A., Contributions to the theory of the Riemann zeta-function.
In Collected Papers of A. Selberg, volume 1, pages 214–280, 1989. The year
of the original publication is 1946.
[52] Selberg, A., On the remainder in the formula for N (T ), the number
of zeros of ζ(s) in the strip 0 < t < T . In Collected Papers of A. Selberg,
volume 1, pages 179–203, 1989. The year of the original publication is 1944.
MR0015426
[53] Serre, J.-P., A Course in Arithmetic, volume 7 of Graduate Texts in
Mathematics. Springer-Verlag, 1973. MR0344216
[54] Silverman, J. H., The Arithmetic of Elliptic Curves, volume 106 of
Graduate Texts in Mathematics. Springer, 1986. MR0817210
[55] Smale, S., Differentiable dynamical systems. In Collected Papers of
Stephen Smale, volume 2, pages 664–734, 2000. The year of the original
publication is 1967. MR0228014
[56] Sunada, T., L-functions in geometry and some applications. In Proc.
Taniguchi Symp. 1985, Curvature and Topology of Riemannian Manifolds,
160 V. Kargin

volume 1201 of Lecture Notes in Mathematics, pages 266–284. Springer-

Verlag, New York, 1986. MR0859591
[57] Sunada, T., Fundamental groups and laplacians. In Proc. Taniguchi
Symp. 1987, Geometry and Analysis on Manifolds, volume 1339 of Lecture
Notes in Mathematics, pages 248–277. Springer-Verlag, New York, 1988.
MR0961485
[58] Terras, A., Zeta Functions of Graphs: A Stroll Through the Garden,
volume 128 of Cambridge Studies in Advanced Mathematics. Cambridge
University Press, 2011. MR2768284
[59] Titchmarsh, E. C., The Theory of the Riemann Zeta-Function. Claren-
don Press, Oxford, second edition, 1986. revised by D. R. Heath-Brown.
MR0882550
[60] Weil, A., Number of solutions of equations in finite fields. In Collected
papers of A. Weil, volume 1, pages 497–508, 1979. The year of the original
publication is 1949. MR0029393
[61] Weil, A., Sur les “formules explicites” de la théorie de nombres premiers.
In Collected papers of A. Weil, volume 2, pages 48–62, 1979. The year of
the original publication is 1952.