Mathematisches Institut, Seminars, (Y. Tschinkel, ed.), p.

81–92
Universität Göttingen, 2004-05

UNIVERSALITY OF L-FUNCTIONS
J. Steuding
Departamento de Matemáticas, Universidad Autónoma de Madrid, C.
Universitaria de Cantoblanco, 28049 Madrid, Spain
E-mail : jorn.steuding@uam.es

Abstract . We survey recent results on the value-distribution of L-functions with

emphasis on aspects of universality.

1. Voronin’s theorem
In 1975 Voronin [Vor75b] discovered a remarkable analytical property of
the Riemann zeta-function ζ(s). Roughly speaking, he proved that any non-
vanishing analytic function can be approximated uniformly by certain purely
imaginary shifts of the zeta-function in the right half of the critical strip. Af-
ter significant improvements due to Reich [Rei77] and Bagchi [Bag81] the
strongest version of Voronin’s theorem has the form (see [Lau96]):

Theorem 1. Suppose that K is a compact subset of the strip D := {s ∈

1
C : 2 < Re s < 1} with connected complement, and let g(s) be a non-vanishing
continuous function on K which is analytic in the interior of K . Then, for
any ε > 0,
 
1
lim inf meas τ ∈ [0, T ] : max |ζ(s + iτ ) − g(s)| < ε > 0.
T →∞ T s∈K

October 26, 2004.

82 Mathematisches Institut, Seminars, 2004-05

We may interpret the absolute value of an analytic function as an analytic

landscape over the complex plane. Then the universality theorem states that
any (finite) analytic landscape can be found (up to an arbitrarily small error) in
the analytic landscape of ζ(s). This remarkable property has several interesting
consequences on the value-distribution of the zeta-function. For instance, for
any n ∈ N and any fixed s with Re s ∈ ( 12 , 1) the set
{(ζ(s + iτ ), ζ 0 (s + iτ ), . . . , ζ (n−1) (s + iτ )) : τ ∈ R}
n
is dense in C . Moreover, it follows that the zeta-function does not satisfy any
algebraic differential equation.
In the first half of the twentieth century, Harald Bohr applied probabilistic
methods in order to study the value distribution of ζ(s). Voronin’s proof of his
universality theorem relies heavily on Bohr’s ideas. Here we sketch a variant
of the proof in the language of weakly convergent probability measures due to
Bagchi [Bag81].
Denote by γ = {s ∈ C : |s| = 1} the unit circle in the complex plane and put
Y
Ω= γp ,
p

where γp = γ for each prime number p. With product topology and point-
wise multiplication this infinite dimensional torus Ω is a compact topological
abelian group, and hence the normalized Haar measure m on the metric space
(Ω, B(Ω)) exists; here B(Ω) denotes the class of Borel sets of Ω. This induces
a probability space (Ω, B(Ω), m). Let ω(p) denote the projection of ω ∈ Ω on
the coordinate space γp . Since the Haar measure m on Ω is the product of the
Haar measures mp on the coordinate spaces,{ω(p) : p prime} is a sequence of
independent complex-valued random variables defined on (Ω, B(Ω), m). Now
denote by H (D) the set of analytic functions defined on the strip D, equipped
with the topology of uniform convergence on compacta. For Re s > 21 and
ω ∈ Ω, let
Y −1
ω(p)
ζ(s, ω) = 1− s .
p
p
This defines an H (D)-valued random element on the probability space
(Ω, B(Ω), m). It can be shown that for almost all ω ∈ Ω the infinite product
defining ζ(s, ω) converges uniformly on compact subsets of D.
The first step in the proof of Theorem 1 is to establish the weak convergence
of the probability measure PT , defined by
1
PT (A) = meas {τ ∈ [0, T ] : ζ(s + iτ ) ∈ A} for A ∈ B(H (D)),
T
 J. Steuding: Universality of L-functions 83

to the distribution P of the random element ζ(s, ω), given by

P(A) = m{ω ∈ Ω : ζ(s + iτ ) ∈ A} for A ∈ B(H (D)),
as T → ∞. For short,
(1) PT ⇒ P (T → ∞).
The proof of this limit theorem relies on fundamental results from probabilty
theory, e.g. Prokhorov’s theorems, ergodic theory for random processes, and
the simple but important fact that the logarithms of different prime numbers
are linearly independent.
The support of the random element
 −1
X ω(p)
log ζ(s, ω) = log 1 − s
p
p

is the closure of the set of all convergent series

 −1
X a(p)
log 1 − s with a(p) ∈ γ.
p
p

The second step is to prove that the set of all these convergent series is dense
in H (DM ), where DM := {s ∈ : 12 < Re s < 1, |Im s| < M } and M is an
arbitrary positive constant. This involves the theory of entire functions of
exponential type, a rearrangement theorem in Hilbert spaces, and the prime
number theorem. The map f 7→ exp f sends H (DM ) to
 
1
g ∈ H (DM ) : g(s) 6= 0 for < Re s < 1 .
2
Now, roughly speaking, the limit theorem (1) ties both ends together. Since
K is a compact subset of D, there exists some M for which K ⊂ DM . It follows
that any g which is contained in the support of the random element ζ(s, ω)
and has a non-vanishing analytic continuation to DM can be approximated
uniformly by some shift ζ(s + iτ ) for s ∈ K : if Φ denotes the set of ϕ ∈ H
such that
max |ϕ(s) − g(s)| < ε,
s∈K
then
 
1
lim inf meas τ ∈ [0, T ] : max |ζ(s + iτ ) − g(s)| < ε > P(Φ) > 0.
T →∞ T s∈K

The case of functions g with zeros in DM \ K follows from an application of

Mergelyan’s approximation theorem). This yields the statement of Theorem 1.
84 Mathematisches Institut, Seminars, 2004-05

All known proofs of universality results for Dirichlet series, like the one for
the zeta-function above or those covered by Theorem 2 below, depend on some
arithmetical conditions. But is universality really an arithmetic phenomenon
or not? Reviewing the proof one might understand universality as a kind of
ergodicity on function spaces. It seems reasonable that the universality of
Dirichlet series is a common phenomenon in analysis, that it is related to Julia
rays in value-distribution theory and to ergodical dynamical systems as well.

2. Zeros and the Riemann hypothesis

Since any non-vanishing analytic function possesses an analytic logarithm,
it follows that log ζ(s) is strongly universal, i.e., log ζ(s) can uniformly approx-
imate functions having zeros; in fact, we have almost given a proof of this fact
in the previous section. It is natural to ask whether ζ(s) is also strongly uni-
versal. The answer is negative. We give a heuristic argument which can be
made waterproof with a bit more effort by the techniques of Section 5.
Assume that g(s) is an analytic function on |s| 6 r, where 0 < r < 41 , which
has a zero ξ with |ξ| < r but which is non-vanishing on the boundary. Then,
whenever the inequality
   
 3 
max ζ s + + iτ − g(s) < ε < min |g(s)|;
|s|6r 4 |s|6r
3


holds, ζ s + 4 + iτ has to have a zero inside |s| 6 r. This can be seen as
follows. By the maximum principle the maximum on the left hand side of the
inequality above is taken on the boundary. The second inequality holds for
sufficiently small ε (since the zeros of an analytic function form a discrete set
or the function vanishes identically). Consequently, an application of Rouché’s
theorem yields the existence of a zero of ζ(s + 34 + iτ ) inside |s| 6 r. If now for
any ε > 0
     
1  3 
lim inf meas τ ∈ [0, T ] : max ζ s + + iτ − g(s) < ε > 0,
 
T →∞ T |s|6r 4
then we expect  T many complex zeros of ζ(s) in the strip 43 −r < Re s < 34 +r
up to height T . This contradicts classical density estimates: for any σ > 21 , the
number of zeros ρ = β + iγ satisfying β > σ, 0 < γ 6 T is known to be o(T )
as T → ∞. Thus, uniform approximation of a function g(s) with a zero by the
zeta-function is impossible.
Bohr [Boh22] discovered an interesting relation between the Riemann hy-
pothesis and almost periodicity. He showed that if χ is non-principal character,
then the Riemann hypothesis for the associated Dirichlet L-function L(s, χ)
 J. Steuding: Universality of L-functions 85

(i.e., the non-vanishing of L(s, χ) for Re s > 21 ) is equivalent to the almost

periodicity of L(s, χ) in the half-plane Re s > 12 . Because of the restriction on
non-principal characters χ this result does not cover the case of ζ(s). More
than half a century later Bagchi [Bag81] proved that the same criterion holds
also for ζ(s), namely that Riemann’s hypothesis is true if and only if for any
compact subset K of the strip 12 < Re s < 1 with connected complement and
for any ε > 0
 
1
lim inf meas τ ∈ [0, T ] : max |ζ(s + iτ ) − ζ(s)| < ε > 0.
T →∞ T s∈K

The crucial implication of Bagchi’s proof relies essentially on Voronin’s univer-

sality theorem, which, of course, was unknown to Bohr.

3. The Selberg class

Meanwhile, it is known that there exists a rich zoo of universal Dirichlet
series; for a list we refer to [Lau96], [Mat04], [Ste04]. It was conjectured by
Linnik and Ibragimov that all functions given by Dirichlet series and analyt-
ically continuable to the left of the half plane of absolute convergence, which
satisfy some natural growth conditions, are universal. In this section we are
interested in the universality of L-functions.
In 1989 Selberg [Sel92] defined a general class S of Dirichlet series having
an Euler product, analytic continuation and a functional equation of Riemann-
type, and formulated some fundamental conjectures concerning them. His aim
was to study the value-distribution of linear combinations of L-functions. In the
meantime this so-called Selberg class became an important object of research.
All known examples of functions in the Selberg class are automorphic (or at
least conjecturally automorphic) L-functions, and for all of them it turns out
that the related Euler factors are the inverse of a polynomial in p−s . This spe-
cial shape of the Euler product is related to Langlands’ reciprocity conjecture.
In the sequel we will consider a subclass of polynomial Euler products.
The class S˜ consists of Dirichlet series
∞
X a(n)
L (s) :=
n=1
ns

satisfying the following axioms:

86 Mathematisches Institut, Seminars, 2004-05

– Polynomial Euler product: for 1 6 j 6 m and each prime p there exist

complex numbers αj (p) with |αj (p)| 6 1 such that
m  −1
YY αj (p)
L (s) = 1− ;
p j=1
ps

– Mean-square: there exists a positive constant κ such that

1 X
lim |a(p)|2 = κ;
x→∞ π(x)
p6x

– Analytic continuation: there exists a non-negative integer k such that

(s − 1)k L (s) is an entire function of finite order;
– Functional equation: there are positive real numbers Q, λj , and there
are complex numbers µj , ω with Re µj > 0 and |ω| = 1, such that
ΛL (s) = ωΛL (1 − s),
where
f
Y
ΛL (s) := L (s)Q s
Γ(λj s + µj ).
j=1
It should be noted that the axiom on the mean-square is intimately related to
Selberg’s conjectures (see [Sel92]). We expect that S˜ contains all non-constant
functions from the Selberg class: S˜ = S \ {1}.
The degree of any non-constant function L ∈ S (and so in S˜) is defined
by
Xf
dL = 2 λj .
j=1
This quantity is well-defined. If NL (T ) counts the number of zeros of L ∈ S
in the rectangle 0 6 Re s 6 1, |Im s| 6 T (counting multiplicities) one can show
by standard contour integration
dL
NL (T ) ∼ T log T,
π
in analogy to the classical Riemann-von Mangoldt formula for ζ(s). It is con-
jectured that the degree is always a positive integer (provided that L is not
constant one).
The functions of degree one in S˜ are the Riemann zeta-function and shifts
of Dirichlet L-functions L(s + iθ, χ) attached to primitive characters χ with
θ ∈ R. Examples of degree two are normalized L-functions associated with
holomorphic newforms; normalized L-functions attached to non-holomorphic
 J. Steuding: Universality of L-functions 87

newforms are expected to lie in S˜. The Rankin-Selberg L-function of any two
holomorphic newforms is an element of the Selberg class of degree 4. Further
examples are Dedekind zeta-functions to number fields K; their degree is equal
to the degree of the field extension K/Q.
In [Ste04] the following generalization of Voronin’s universality theorem was
proved.
Theorem 2. Let L ∈ S˜ and K be a compact subset of the strip
   
1 1
DL := s ∈ C : max ,1 − < Re s < 1
2 dL
with connected complement, and let g(s) be a non-vanishing continuous function
on K which is analytic in the interior of K . Then, for any ε > 0,
 
1
lim inf meas τ ∈ [0, T ] : max |L (s + iτ ) − g(s)| < ε > 0.
T →∞ T s∈K

This theorem may be regarded as a verification of the Linnik-Ibragimov con-

jecture for Dirichlet series in the Selberg class.

4. The strip of universality

Besides the arithmetic axioms on the polynomial Euler product and on the
mean-square, a further important ingredient in the proof of Theorem 2 is the
second moment bound
1 T
Z
lim sup |L (σ + it)|2 dt < ∞.
T →∞ T 1

In [Ste04] an asymptotic formula for L ∈ S˜ in the range DL was proved. If

d L > 2, the strip DL does not cover the right half of the critical strip; any
extension of this strip to the left would imply universality in this extended strip.
However, the critical line is a natural boundary for universality of L ∈ S˜ (at
least in the sense of Theorem 2).
In particular cases of functions L ∈ S˜ with degree d L > 2 the existence of
the mean-square covering the strip DL is known. For instance, let L(s, χ) be an
arbitrary Dirichlet L-function to a primitive character χ. Then ζ(s)2 L(s, χ) is
an element of S˜ of degree 3, so Theorem 2 gives universality for 23 < Re s < 1.
Using Montgomery’s estimates for the fourth moment of Dirichlet L-functions
and Ivić’s eigth-moment estimate for ζ(s), the Cauchy-Schwarz inequality yields
Z T
11−8σ 1
|ζ(σ + it)2 L(σ + it, χ)|2 dt  T 12 + 2 +ε  T
1
88 Mathematisches Institut, Seminars, 2004-05

for any σ > 58 . Thus ζ(s)2 L(s, χ) is universal in the strip 5

8 < Re s < 1. If the
generalized Lindelöf hypothesis is true for L ∈ S˜, i.e.,
 
1
L + it  tε
2
for any ε > 0 as t → ∞, then the strip of universality can be extended to the
full open right half of the critical strip.

5. Effectivity
The known proofs of universality theorems are ineffective, giving neither an
estimate for the first approximating shift τ nor bounds for the positive lower
density. There are some remarkable attempts due to Garunkštis [Gar03], Good
[Goo81], and Laurinčikas [Lau00], however, their results are either restricted
to rather small classes of functions or conditional subject to certain unproved
hypotheses. Following [Ste03] we now consider the problem of effective upper
bounds for the upper density of universality.
Denote by Br the closed disc of radius r > 0 with center in the origin. We
define for a meromorphic function L(s), an analytic function g : Br → C with
fixed r ∈ 0, 41 , and positive ε the densities


     
1  3 
d (ε, g, L) = lim inf meas τ ∈ [0, T ] : max L s + + iτ − g(s) < ε ,
T →∞ T |s|6r 4
and
     
1  3 
d (ε, g, L) = lim sup meas τ ∈ [0, T ] : max L s + + iτ − g(s) < ε .
 
T →∞ T |s|6r 4
We consider analytic isomorphisms g : Br → B1 , i.e., the inverse g −1 exists
and is analytic. Obviously, such a function g has exactly one simple zero ξ in
the interior of Br . By the Schwarz lemma any such g has a representation
ξ−s
g(s) = r exp(iϕ) with ϕ ∈ R and |ξ| < r.
r2 − ξs
Denote by Ar the class of analytic isomorphisms from Br (with fixed 0 < r < 14 )
to the unit disc. Further, let NL (σ1 , σ2 , T ) count the number of zeros of L(s)
in 21 < σ1 < Re s < σ2 < 1, 0 6 t < T (counting multiplicities).

Theorem 3. Suppose that g ∈ Ar . Assume that L(s) is analytic in Re s >

3
4 − r except for at most o(T ) many singularities inside Re s > 34 − r, 0 6
 J. Steuding: Universality of L-functions 89

Im s 6 T , as T → ∞,

and that d (ε, g, L) > 0 for all ε > 0. Then, for any
1 1
ε ∈ 0, 2r 4 + Re |ξ| ,
8r3 ε
 
1 3 3
d (ε, g, L) 6 2 lim sup N L + Re ξ − 2rε, + Re ξ + 2rε, T .
r − |ξ|2 T →∞ T 4 4
We sketch the proof (which is a bit in the spirit of Section 2). The zero ξ of g is
related to some zeros of L(s) in 12 < Res s < 1. Since g maps the boundary of
Br onto the unit circle, Rouché’s theorem implies the existence of one simple
zero λ of L(z) in  
3
Kτ := z = s + + iτ : s ∈ Br ,
4
whenever
   
 3 
max L s + + iτ − g(s) < ε < 1 = min |g(s)|.

s∈Br  4 s∈Br

We may say that the zero λ of L(s) is generated by the zero ξ of g(s). Uni-
versality is a phenomenon that happens in intervalls. Suppose that a zero λ
of L(s), generated by ξ, lies in two different sets Kτ1 and Kτ2 . Then one can
show that
8r4 ε
|τ1 − τ2 | < 2 .
r − |ξ|2
Now denote by S Ij (T ) the disjoint intervalls in [0, T ] such that (1) is valid
exactly for τ ∈ j Ij (T ) =: I (T ). By the latter estimate, in every intervall
Ij (T ), there lie at least
 2
r − |ξ|2 r2 − |ξ|2

1+ 3
meas I j (T ) > meas Ij (T )
8r ε 8r3 ε
zeros λ of L(s) in the strip 21 < Re s < 1. Therefore, the number N (T ) of such
zeros λ satisfies the estimate
8r3 ε
(2) N (T ) > meas I (T ).
r − |ξ|2
2

The value distribution of L(z) in Kτ is ruled by that of g(s) in Br . This

 on the real parts of the zeros λ. One can show that
gives a 3restriction
Re λ − − Re ξ  < 2rε. This yields
4
 
3 3
N (T ) 6 NL + Re ξ − 2rε, + Re ξ + 2rε, T .
4 4
Since d (ε, g, L) > 0, this leads via (2) to the estimate of the theorem. Note that
the set of singularities of L(s) in σ > 34 − r has zero density but d (ε, g, L) > 0.
So the singularities do not affect the above observations.
90 Mathematisches Institut, Seminars, 2004-05

Theorem 3 relates the density of universality to the value-distribution of L.

In the case of the Riemann zeta-function we can be more explicit. In view of
classical density theorems the set of singularities of log ζ(s) has density zero.
Hence, we may apply Theorem to L(s) = log ζ(s). By Bohr and Jessen [BJ32],
Hilfssatz 6, the limit
 
1 3 3
lim Nlog ζ + Re ξ − 2rε, + Re ξ + 2rε, T
T →∞ T 4 4
exists, and tends to zero as ε → 0. Hence, under the above assumptions
d (ε, exp g, ζ(s)) = o(ε).
Thus, the decay of d (ε, exp g, ζ) with ε → 0 is more than linear in ε.

6. Joint universality
We conclude with another interesting problem concerning universality of
L-functions.
Voronin [Vor75a] also obtained joint universality for Dirichlet L-functions,
that is simultaneous uniform approximation by a family of L-functions associ-
ated with non-equivalent characters; the non-equivalence of the characters as-
sures a certain independence of the related L-functions, and this independence
is necessary for joint universality. Recently, Laurinčikas & Matsumoto [LM04]
proved a joint universality theorem for L-functions associated with newforms
twisted by characters. It is natural to ask for joint universality in the Selberg
class. However, all known jointly universal families are given by (multiplicative
or additive) twists of a single universal Dirichlet series by characters. In some
sense, Selberg’s Conjecture B (see [Sel92]) states that primitive functions form
an orthonormal system in the Selberg class. As proved by Bombieri & Hejhal
[BH95], this implies the statistical independence of primitive functions. There
is some hope that this can be used as substitute for the independence induced
by non-equivalent characters in order to prove joint universality for distinct
primitive L-functions from the Selberg class.
For 1 6 j 6 m, assume that the L-functions
∞
X aLj (n)
Lj (s) =
n=1
ns

from S˜ satisfy the orthogonality condition

X aLj (p)aLk (p)
(3) = δjk κj log log x + O(1),
p
p6x
 J. Steuding: Universality of L-functions 91

where κj is a positive constant depending on Lj , and δjk = 1 if j = k and

δjk = 0 otherwise. This condition is known to hold for several families of L-
functions in S˜, for example for Dirichlet L-functions associated with pairwise
non-equivalent characters (in which case it is nothing else than the orthogonal-
ity relation for characters); it is expected to hold for any two distinct primitive
L-functions from the Selberg class (Selberg’s Conjecture B). Moreover, (3) may
be regarded as an extension of the axiom on the mean square in the definition
of S˜.
Conjecture. Suppose that L1 , . . . , Lm are elements of S˜ satisfying condition
(3). For 1 6 j 6 m let gj (s) be a continuous function on Kj which is non-
vanishing in the interior, where Kj is a compact subset of the strip
   
1 1
D := s : max ,1 − < Re s < 1
2 d
with connected complement, and d is the maximum of the degrees of the Lj (a
quantity determined by the functional equation for Lj ). Then, for any ε > 0,
 
1
lim inf meas τ ∈ [0, T ] : max max |Lj (s + iτ ) − gj (s)| < ε > 0.
T →∞ T 16j6m s∈Kj

Acknowledgements. The author is very grateful to Prof. Yuri Tschinkel for

the invitation to Göttingen and the possibility of having a look into Riemann’s
original manuscripts on the zeta-function.

References
[Bag81] B. Bagchi – The statistical behaviour and universality properties of the
Riemann zeta-function and other allied Dirichlet series, 1981, Ph.D.Thesis,
Calcutta, Indian Statistical Institute.
[BH95] E. Bombieri & D. A. Hejhal – On the distribution of zeros of linear
combinations of Euler products, Duke Math. J. 80 (1995), no. 3, p. 821–
862.
[BJ32] H. Bohr & B. Jessen – Über die Werteverteilung der Riemannschen Zeta-
funktion, zweite Mitteilung, Acta Math. 58 (1932), p. 1–55.
[Boh22] H. Bohr – Über eine quasi-periodische Eigenschaft Dirichletscher Reihen
mit einer Anwendung auf die Dirichletschen l-Funktionen, Math. Annalen
85 (1922), p. 115–122.
[Gar03] R. Garunkštis – The effective universality theorem for the Riemann zeta
function, in Proceedings of the Session in Analytic Number Theory and Dio-
phantine Equations (Bonn), Bonner Math. Schriften, vol. 360, Univ. Bonn,
2003, p. 21.
92 Mathematisches Institut, Seminars, 2004-05

[Goo81] A. Good – On the distribution of the values of Riemann’s zeta function,

Acta Arith. 38 (1980/81), no. 4, p. 347–388.
[Lau96] A. Laurinčikas – Limit theorems for the Riemann zeta-function, Mathe-
matics and its Applications, vol. 352, Kluwer Academic Publishers Group,
Dordrecht, 1996.
[Lau00] A. Laurinchikas – Effectivization of the universality theorem for the Lerch
zeta function, Liet. Mat. Rink. 40 (2000), no. 2, p. 172–178.
[LM04] A. Laurinčikas & K. Matsumoto – The joint universality of twisted
automorphic L-functions, J. Math. Soc. Japan 56 (2004), no. 3, p. 923–939.
[Mat04] K. Matsumoto – Probabilistic value-distribution theory of zeta-functions,
Sugaku Expositions 17 (2004), no. 1, p. 51–71.
[Rei77] A. Reich – Universelle Werteverteilung von Eulerprodukten, Nachr. Akad.
Wiss. Göttingen Math.-Phys. Kl. II (1977), no. 1, p. 1–17.
[Sel92] A. Selberg – Old and new conjectures and results about a class of Dirichlet
series, in Proceedings of the Amalfi Conference on Analytic Number Theory
(Maiori, 1989) (Salerno), Univ. Salerno, 1992, p. 367–385.
[Ste03] J. Steuding – Upper bounds for the density of universality, Acta Arith.
107 (2003), no. 2, p. 195–202.
[Ste04] J. Steuding – Value-distribution of L-functions, 2004, habilitation thesis,
Frankfurt.
[Vor75a] S. M. Voronin – The functional independence of Dirichlet L-functions,
Acta Arith. 27 (1975), p. 493–503, Collection of articles in memory of Juriı̆
Vladimirovič Linnik.
[Vor75b] , A theorem on the “universality” of the Riemann zeta-function, Izv.
Akad. Nauk SSSR Ser. Mat. 39 (1975), no. 3, p. 475–486, 703.