Lower bounds for moments of L-functions

Z. Rudnick, and K. Soundararajan

PNAS 2005;102;6837-6838; originally published online May 2, 2005;
doi:10.1073/pnas.0501723102
This information is current as of March 2007.

Online Information High-resolution figures, a citation map, links to PubMed and Google Scholar,
& Services etc., can be found at:
www.pnas.org/cgi/content/full/102/19/6837
This article has been cited by other articles:
www.pnas.org/cgi/content/full/102/19/6837#otherarticles
E-mail Alerts Receive free email alerts when new articles cite this article - sign up in the box
at the top right corner of the article or click here.
Rights & Permissions To reproduce this article in part (figures, tables) or in entirety, see:
www.pnas.org/misc/rightperm.shtml
Reprints To order reprints, see:
www.pnas.org/misc/reprints.shtml

Notes:
Lower bounds for moments of L-functions
Z. Rudnick†‡ and K. Soundararajan§
†School of Mathematical Sciences, Tel Aviv University, Tel Aviv 69978, Israel; and §Department of Mathematics, University of Michigan, Ann Arbor, MI 48109

Edited by Peter Sarnak, New York University, New York, NY, and approved March 10, 2005 (received for review March 2, 2005)

The moments of central values of families of L-functions have
recently attracted much attention and, with the work of Keating
and Snaith [(2000) Commun. Math. Phys. 214, 57– 89 and 91–110],
there are now precise conjectures for their limiting values. We
develop a simple method to establish lower bounds of the con-
jectured order of magnitude for several such families of L-func-
tions. As an example we work out the case of the family of all
Dirichlet L-functions to a prime modulus.
a simple method that furnishes lower bounds of the conjectured
order of magnitude for many families of L-functions, including
the three prototypical examples given above. As a rough prin-
ciple, it seems that whenever one can evaluate the first moment
of a family of L-functions (with a little bit to spare) then one can
obtain good lower bounds for all moments. We illustrate our
method by giving a lower bound (of the conjectured order of
magnitude) for the family of Dirichlet characters modulo a
prime.

A classical question in the theory of the Riemann zeta

1
function asks for asymptotics of the moments 兰T1 兩␨(2 ⫹ it)兩2k dt, Theorem. Let k be a fixed natural number. Then for all large primes q

冘
where k is a positive integer. A folklore conjecture states that the
2
2kth moment should be asymptotic to C kT(log T) k for a positive
兩L共 21 , ␹ 兲兩 2k ⬎⬎k q共log q兲 k .
2

constant C k. Only very recently with the work of Keating and ␹ 共mod q 兲
Snaith (1) modeling the moments of ␨ (s) by moments of ␹⫽␹0

characteristic polynomials of random matrices has a conjecture

emerged for the value of C k. This conjecture agrees with classical
results of Hardy and Littlewood and Ingham (see ref. 2) in the
cases k ⫽ 1 and k ⫽ 2, but for k ⱖ 3 very little is known.
1
Ramachandra (3) showed that 兰 T1 兩 ␨ ( 2 ⫹ it)兩 2k dt ⬎⬎ T(log T) k
2
for positive integers 2k, and Heath-Brown (4) extended this for
any positive rational number k. Titchmarsh (see theorem 7.19 of
ref. 2) had previously obtained a smooth version of these lower
Our methods yield corresponding lower bounds for several
other families, as well as other applications, for instance to
fluctuations of matrix elements of Maass wave forms in the
modular domain, which are not presented here. We remark also
that we may take k to be any positive rational number ⱖ1 in the
theorem. If k ⫽ r兾s(ⱖ1) is rational, then we achieve this by
taking A( ␹ ) ⫽ (兺 nⱕq1/(2r) d 1/s(n) ␹ (n)兾 公n) s in the argument

MATHEMATICS
bounds for positive integers k. below.
Analogously, given a family of L-functions an important Proof: Let x :⫽ q1/(2k) be a small power of q, and set A(␹) ⫽
problem is to understand the moments of the central values of 兺 nⱕx ␹ (n)兾 公n. We will evaluate

冘 冘
these L-functions. Modeling the family of L-functions by using
k
the choice of random matrix ensembles suggested by Katz and S1 :⫽ L共 21 , ␹ 兲A共 ␹ 兲 k⫺1A共 ␹ 兲 , and S 2 :⫽ 兩A共 ␹ 兲兩 2k ,
Sarnak (5) in their study of low-lying zeros, Keating and Snaith ␹ 共 mod q 兲 ␹ 共 mod q 兲
␹⫽␹0 ␹⫽␹0
(6) have advanced conjectures for such moments. We illustrate
these conjectures by considering three prototypical examples. 2
and show that S 2 ⬍⬍ q(log q) k ⬍⬍ S 1. The theorem then follows
The family of Dirichlet L-functions L(s, ␹ ) as ␹ varies over from Hölder’s inequality:
primitive characters (mod q) is a unitary family, and it is
conjectured that

冘 * 兩L共 1 , ␹ 兲兩 2k ⬃ C q共log q兲 k2, [1]

冘
␹ 共mod q 兲
兩L共 21 , ␹ 兲兩 2k ⱖ
兩S 1兩 2k
S 2k⫺1
2
⬎⬎ q共log q兲 k .
2

2 1,k
␹ 共mod q 兲
where k 僆 ⺞ and C 1,k is a specified positive constant. The family
of quadratic Dirichlet L-functions L(s, ␹ d), where d is a funda-
mental discriminant and ␹ d is the associated quadratic character,
is a symplectic family and it is conjectured that
冘 L共21, ␹d兲k ⬃ C2,k X共log X兲 k共k⫹1兲/2 ,
␹⫽␹0
If ᐉ 僆 ⺞ then we may write A( ␹ ) ᐉ ⫽ 兺 nⱕxᐉ ␹ (n)d ᐉ(n; x)兾 公n
where d ᐉ(n; x) denotes the number of ways of writing n as
a 1 䡠 䡠 䡠 a ᐉ with each a j ⱕ x. As usual d ᐉ(n) will denote the ᐉth
divisor function, and note that d ᐉ(n; x) ⱕ d ᐉ(n) with equality
holding when n ⱕ x.
We start with S 2. Note that A( ␹ 0) ⬍⬍ 公x and so

冘 冘
[2]
兩d兩ⱕX
兩A共 ␹ 兲兩 2k ⫽ 兩A共 ␹ 兲兩 2k ⫹ O共x k兲
␹ 共mod q 兲
␹ 共 mod q 兲
where k 僆 ⺞ and C 2,k is a specified positive constant. The family ␹⫽␹0

of quadratic twists of a given newform f, L(s, f R ␹ d) (the

L-function is normalized so that the central point is 21); this is an
orthogonal family and it is conjectured that
⫽ 冘
m,nⱕx k
d k共m, x兲d k共n, x兲
冑mn 冘
␹ 共 mod q 兲
␹ 共m兲 ␹៮ 共n兲

冘
兩d兩ⱕX
L共21, f 丢 ␹d兲k ⬃ C3,kX共log X兲 k共k⫺1兲/2 , [3] ⫹ O共x k兲.

Since x k ⫽ 公q ⬍ q the orthogonality relation for characters

where k 僆 ⺞, C 3,k is a specified constant that depends on the
form f.
While asymptotics in Eqs. 1–3 are known for small values of
k, for large k these conjectures appear formidable. Further, the
methods used to obtain lower bounds for moments of ␨ (s) do not
appear to generalize to this situation. In this article we describe
(mod q) gives that only the diagonal terms m ⫽ n survive. Thus
This paper was submitted directly (Track II) to the PNAS office.
‡To whom correspondence should be addressed. E-mail: rudnick@post.tau.ac.il.
© 2005 by The National Academy of Sciences of the USA

www.pnas.org兾cgi兾doi兾10.1073兾pnas.0501723102 PNAS 兩 May 10, 2005 兩 vol. 102 兩 no. 19 兩 6837– 6838
 S2 ⫽ ␾共q兲 冘
nⱕx k
dk共n, x兲2
n
⫹ O共冑q兲. S1 ⫽ ␾共q兲 冘 冘 冘 nⱕX
aⱕxk⫺1 bⱕxk an⬅b 共mod q 兲
d k⫺1共a; x兲d k共b; x兲
冑abn

冉 冊
2
Since d k(n, x) ⱕ d k(n) and 兺 nⱕy d k(n) 2兾n ⬃ c k(log y) k for a S2
2
positive constant c k, we find that S 2 ⬍⬍ q(log q) k , as claimed. ⫹O .
log q
We now turn to S 1. If Re(s) ⬎ 1 then integration by parts gives

冕 冉冘 冊
The main term above will arise from the diagonal terms an ⫽
L共s, ␹兲 ⫽ 冘 ␹共n兲
ns
⫹
⬁
1
ys
d ␹共n兲
b. Let us first estimate the contribution of the off-diagonal terms.
Here we may write an ⫽ b ⫹ qᐉ, where 1 ⱕ ᐉ ⱕ Xx k⫺1兾q ⫽
nⱕX X X⬍nⱕy x k⫺1(log q) 4. The contribution of these off-diagonal terms is

⫽ 冘
nⱕX
␹共n兲
ns
⫹s 冕冘⬁

X
X⬍nⱕy␹ 共n兲
s⫹1
y
dy. ⬍⬍q 冘
bⱕxk
dk共b; x兲
冑b 冘
ᐉⱕxk⫺1共log q 兲 4
1
冘
冑qᐉ an⫽b⫹qᐉ d k⫺1共a; x兲
Since the numerator of the integrand above is ⬍⬍ 公q log q by 1 k⫺1 k q
⬍⬍q 2⫹␧ x 2
⫹
2 ⬍⬍
the Pólya-Vinogradov inequality (see chapter 23 of ref. 7) the log q
above expression furnishes an analytic continuation of L(s, ␹ ) to
Re(s) ⬎ 0. Moreover we obtain since 兺 an⫽b⫹qᐉ d k⫺1(a; x) ⱕ d k(b ⫹ qᐉ) ⬍⬍ (qᐉ) ␧. Therefore,

L共21, ␹兲 ⫽ 冘 ␹共n兲
冑n ⫹O 冉冑 冑 冊q log q
X
.
S1 ⫽ ␾共q兲
bⱕx
冘 k
dk共b; x兲
b aⱕx k⫺1
冘,nⱕX
an⫽b
dk⫺1共a; x兲 ⫹ O
S2
log q
冉 冊
.
nⱕX

We choose here X ⫽ q log4 q and obtain Since

S1 ⫽ 冘 冘 ␹ 共n兲
A共 ␹ 兲 k⫺1 A共 ␹ 兲
k
冘
aⱕxk⫺1,nⱕX
dk⫺1共a; x兲 ⱕ 冘
aⱕxk⫺1,nⱕx
dk⫺1共a; x兲 ⫽ dk共b; x兲,

␹ 共mod q 兲
␹⫽␹0 nⱕX
冑n an⫽b an⫽b

and d k(b; x) ⫽ d k(b) for b ⱕ x, we deduce that

⫹O
冉 1
冘
log q ␹ 共mod q兲
␹⫽␹0
兩A共 ␹ 兲兩 2k⫺1 .
冊 S1 ⱖ ␾共q兲
bⱕx
冘 k
dk共b; x兲2
b
⫹O
S2
log q
冉 冊
Since 兩A(␹)兩2k⫺1 ⱕ 1 ⫹ 兩A(␹)兩2k the error term above is ⬍⬍ (q ⫹ S2)兾
log q. The main term is ⱖ ␾ 共q兲 冘 d k共b兲 2
b
⫹O
S2
log q
冉 冊
⬎⬎ q共log q兲 k .
2

冘 冘
bⱕx
␹ 共n兲
A共 ␹ 兲 k⫺1 A共 ␹ 兲 k ⫹ O共 冑Xx k⫺2兲.
1

␹ 共mod q 兲 nⱕX
冑n This proves the theorem.

This research was partially supported by United States-Israel Binational

Recalling that x ⫽ q 1/(2k) and using the orthogonality relation for Science Foundation Grant 2002088. K.S. was partially supported by the
characters we conclude National Science Foundation.

1. Keating, J. P. & Snaith, N. C. (2000) Commun. Math. Phys. 214, 57–89.
2. Titchmarsh, E. C. (1986) The Theory of Riemann Zeta Function (Oxford Univ.
Press, Oxford).
3. Ramachandra, K. (1980) Hardy-Ramanujan J. 3, 1–25.
4. Heath-Brown, D. R. (1981) J. London Math. Soc. 24, 65–78.
5. Katz, N. & Sarnak, P. (1998) Random Matrices, Frobenius Eigenvalues, and
Monodromy (Am. Soc. Math., Providence, RI).
6. Keating, J. P. & Snaith, N. C. (2000) Commun. Math. Phys. 214,
91–110.
7. Davenport, H. (1980) Multiplicative Number Theory (Springer, New York).

6838 兩 www.pnas.org兾cgi兾doi兾10.1073兾pnas.0501723102 Rudnick and Soundararajan