Professional Documents
Culture Documents
REVISITED
Abstract. This paper describes a method to compute lower bounds for moments of ζ and
L-functions. The method is illustrated in the case of moments of |ζ( 21 + it)|, where the
results are new for small moments 0 < k < 1.
1. Introduction
This paper reexamines the problem of obtaining lower bounds of the correct order of mag-
nitude for moments of the Riemann zeta function on the critical line, and related problems
for central values in families of L-functions. Our work is motivated by recent work on the
complementary problem of obtaining upper bounds for such moments. For example, [8]
enunciates the principle that an upper bound for a particular moment (with a little flexibil-
ity) may be used to establish upper bounds of the correct order of magnitude for all smaller
moments. Recent work of the authors with Radziwill [3] provides such upper bounds for all
moments of the Riemann zeta-function below the fourth moment. In those papers, one key
idea is to approximate Euler products that mimic suitable powers of the zeta-function using
Dirichlet series of small length. The aim of this paper is to demonstrate how that idea may
also be used to establish lower bounds of the right order of magnitude for all moments of
the Riemann zeta-function.
1 1
Theorem 1. Let T be large. Uniformly for (log T )− 2 ≤ k ≤ (log T ) 2 −δ (for any fixed δ > 0)
we have Z 2T
2
|ζ( 21 + it)|2k dt ≥ Ck T (log T )k ,
T
2
where we may take Ck = C1 k in the range k ≤ 1, and Ck = (C2 k 2 log(ek))−k for some
absolute positive constants C1 and C2 .
There is a long history concerning such lower bounds for ζ and L-functions. To place our
result briefly in context, we recall that in the range k ≥ 1 such a lower bound was established
by [7], although our quantification of Ck is better and the proof arguably simpler. Theorem
1 is new in the range 0 < k ≤ 1. Previous work of Heath-Brown [4] had established such a
bound for rational k in this range, and for real k such a bound was known to hold conditional
1
on the Riemann Hypothesis (see [4, 10, 11]). In the range c(log log T )− 2 ≤ k = o(1),
1
2 WINSTON HEAP AND K. SOUNDARARAJAN
2
Laurinchikas [5] has shown that the 2k-th moment is ∼ T (log T )k . The constant Ck in our
result tends to zero as k → 0; with more effort, our argument could be made to yield Ck ≫ 1
for all k ≤ 1, but we have not done so in the interest of keeping the exposition simple.
Combining the upper bound result of Heap, Radziwill, and Soundararajan [3] with the
lower bound of Theorem 1, we obtain the following corollary.
1
Corollary 1. For T large, uniformly for (log T )− 2 ≤ k ≤ 2 we have
Z 2T
k2 2
T (log T ) ≫ |ζ( 12 + it)|2k dt ≫ kT (log T )k .
T
The moments of ζ( 21 +it) encode information on the distribution of large values of |ζ( 12 +it)|.
In [12] it was observed that the 2k-th moment of |ζ( 21 + it)| should be dominated by values
2
of size (log T )k , which should occur on a set of measure about T /(log T )k . On RH, it was
2
shown in [12] that the measure of {t ∈ [T, 2T ] : |ζ( 21 + it)| ≥ (log T )k } is T (log T )−k +o(1)
for any fixed positive k. From Corollary 1, we may obtain a sharper form of such a result
unconditionally in the limited range 0 < k < 2.
Corollary 2. Uniformly in the range
p p
log log T log log log T ≤ V ≤ 2 log log T − 2 log log T log log log T
we have
V
V2 V log log log T
meas{t ∈ [T, 2T ] : |ζ( 21 + it)| ≥ e } = T exp − +O √ .
log log T log log T
Recall that Selberg’s central limit theorem (see [9] for a proof related to ideas of this
paper) states that for t chosen uniformly from [T, 2T ], log |ζ( 21 + it)| has an approximately
normal distribution with mean 0 and variance ∼ 21 log log T . Radziwill [6] has established a
3
uniform version of this result showing that for V ≤ (log log T ) 5 −ǫ one has
Z ∞
T
q
2
1 1
meas {t ∈ [T, 2T ] : log |ζ( 2 + it)| ≥ V = ∆ 2 log log T } ∼ √ e−x /2 dx.
2π ∆
Corollary 2 gives a crude version of such a result but in a wider range for V .
Acknowledgments. The second author is partially supported by grants from the NSF
(including the FRG grant DMS1854398), and through a Simons Investigator grant from the
Simons Foundation. We are grateful to Maksym Radziwill for many valuable discussions on
these themes.
establish Theorem 1 for k ≥ 1, and also discuss lower bounds for moments of central values
of L-functions in families.
Throughout, logj will denote the j-fold iterated logarithm. Let T be large and assume
√
that 1/ log T ≤ k ≤ 1. Let ℓ denote the largest integer such that logℓ T ≥ 104. Define a
sequence Tj by setting T1 = e2 , and for 2 6 j 6 ℓ by
k log T
Tj := exp .
(logj T )2
Note that T2 is already large. Further, the sequence Tj is in ascending order, and lastly
k log T ≪ log Tℓ ≤ 10−8k log T .
For each 2 6 j 6 ℓ, set
X 1 X 1
Pj (s) := , and P j = Pj (1) = .
T 6p<T
ps T 6p<T
p
j−1 j j−1 j
Note that
log Tj 1 log T
j−1
Pj = log +O ∼ 2 log = 2 logj T − 2 logj+1 T,
log Tj−1 log Tj−1 logj T
so that Pℓ > 104 , Pℓ−1 > exp(104 ), and so on.
Let N denote the set of integers n = n2 · · · nℓ where each nj is divisible only by primes in
the interval Tj−1 to Tj and such that Ω(nj ) ≤ Kj := 500Pj for all 2 ≤ j ≤ ℓ. If n ∈ N then
(1) n = n2 · · · nℓ ≤ T2500P2 T3500P3 · · · Tℓ500Pℓ ≤ T k/9 .
Let g(n) denote the multiplicative function given on prime powers by g(pr ) = 1/r!. Define,
for any real number α and 2 ≤ j ≤ ℓ
Kj
X 1 X αΩ(n) g(n)
(2) Nj (s, α) = (αPj (s, α))r = ,
r=0
r! ns
p|n =⇒ Tj−1 ≤p≤Tj
Ω(n)≤Kj
and put
X αΩ(n) g(n) Y ℓ
(3) N (s, α) := s
= Nj (s, α).
n∈N
n j=2
In view of (1), N (s, α) is a short Dirichlet polynomial. The idea is that N (s, α) behaves
in many ways like ζ(s)α, but with the advantage that since N (s, α) is a short Dirichlet
polynomial, one can compute mean-values involving it and ζ(s). The proof of our theorem
rests on the following three propositions dealing with such mean values involving ζ(s) and
N (s, α) for suitable values of α.
4 WINSTON HEAP AND K. SOUNDARARAJAN
√
Proposition 1. Let T be large. Uniformly in the range 1 ≥ k ≥ 1/ log T we have
Z 2T
2
ζ( 21 + it)N ( 12 + it, k − 1)N ( 12 − it, k)dt ≥ C1 T (log T )k ,
T
for some positive constant C1 .
√
Proposition 2. Let T be large. Uniformly in the range 1 ≥ k ≥ 1/ log T we have
Z 2T
2
|ζ( 21 + it)N ( 21 + it, k − 1)|2 dt ≤ C2 k −1 T (log T )k ,
T
for some positive constant C2 .
√
Proposition 3. Let T be large. Uniformly in the range 1 ≥ k ≥ 1/ log T we have
Z 2T
2 2
|N ( 21 + it, k)| k |N ( 12 + it, k − 1)|2dt ≤ C3 T (log T )k ,
T
for some positive constant C3 .
Two applications of Hölder’s inequality give
Z 2T
1 1 1
ζ( 2 + it)N ( 2 + it, k − 1)N ( 2 − it, k)dt
T
Z 2T 12 Z 2T 1−k
2
2k
≤ 1
|ζ( 2 + it)| dt × |ζ( 12 + it)N ( 21 + it, k − 1)|2 dt
T T
Z 2T 2
k2
× |N ( 12 + it, k)| k |N ( 12 + it, k − 1)|2 dt ,
T
so that the lower bound of the theorem follows at once from the three propositions.
Deducing Corollary 2 from Corollary
√ 1. Let V be in the range of the corollary, and put k =
V / log log T and δ = log3 T / log log T so that k + 2δ ≤ 2. The upper bound implicit in the
corollary follows (in a stronger form) upon noting that
Z 2T
V −2kV
V2
1
meas{t ∈ [T, 2T ] : |ζ( 2 + it)| ≥ e } ≤ e |ζ( 21 + it)|2k dt ≪ T exp − .
T log log T
To prove the lower bound, consider
Z 2T
2 T 2
(4) |ζ( 21 + it)|2(k+δ) dt ≫ (k + δ)T (log T )(k+δ) ≫ √ (log T )(k+δ) .
T log log T
The contribution to the integral from t with |ζ( 12 + it)| ≤ eV is
Z 2T
2δV 2
T 2
≤e |ζ( 21 + it)|2k dt ≪ T (log T )k +2kδ = o √ (log T )(k+δ) .
T log log T
LOWER BOUNDS FOR MOMENTS OF ZETA AND L-FUNCTIONS REVISITED 5
Similarly, the contribution to the integral from t with |ζ( 12 + it)| ≥ eV (log T )2δ is
Z 2T
−2δ(k+2δ) 2
T 2
≤ (log T ) |ζ( 21 + it)|2(k+2δ) dt ≪ T (log T )k +2δk = o √ (log T )(k+δ) .
T log log T
Thus the left side of (4) is dominated by values of |ζ( 21 + it)| lying between eV = (log T )k
and (log T )k+2δ and it follows that the measure of the set of such t is
Z
−2(k+δ)(k+2δ)
≫ (log T ) t∈[T,2T ] |ζ( 12 + it)|2(k+δ) dt
1
(log T )k+2δ ≥|ζ( +it)|≥(log T )k
2
T 2 2
≫√ (log T )−k −4kδ−3δ .
log log T
The corollary follows.
3. Proof of Proposition 1
Expanding out, we have
Z 2T
ζ( 12 + it)N ( 12 + it, k − 1)N ( 12 − it, k)dt
T
X (k − 1)Ω(n) k Ω(m) g(n)g(m) Z 2T m it
1
(5) = √ ζ( 2 + it) dt.
n,n∈N
mn T n
Using the simple approximation
X 1
ζ(1/2 + it) = + O(T −1/2 ), t ∈ [T, 2T ]
r6T
r 1/2+it
we find that
Z 2T m it δ(rn = m) 1 X 1
ζ( 21 + it) dt = T √ +O T2 + √ .
T n r r≤T
r| log(rn/m)|
rn6=m
Here δ(rn = m) equals 1 if n|m and r = m/n, and there is no main term if n ∤ m. If
rn 6= m, we may estimate 1/| log(rn/m)| trivially by ≪ m, and so the remainder term above
1
is O(mT 2 ). From these remarks, it follows that the right side of (5) equals
X (k − 1)Ω(n) k Ω(m) g(n)g(m) X 1 1
(6) T +O √ mT .2
m,n∈N
m m,n∈N
mn
n|m
Since the elements of N are all bounded by T 1/9 , the error term above is seen to be O(T 7/9 ),
which is negligible.
6 WINSTON HEAP AND K. SOUNDARARAJAN
If we drop the condition that Ω(mj ) ≤ Kj , then the sums over nj , mj above may be replaced
with (thinking of a as the power of p dividing mj and b the power dividing nj )
Y X k a (k − 1)b Y k2
a b
1+ a
g(p )g(p ) ≥ 1 + .
T ≤p≤T a≥1
p T ≤p≤T
p
j−1 j j−1 j
a≥b≥0
2
for an absolute positive constant C. Since log Tℓ ≫ k log T , and k k ≫ 1 for 0 < k ≤ 1, this
proves Proposition 1.
4. Proof of Proposition 2
It is a simple matter to compute the mean square of the zeta function multiplied by a short
Dirichlet polynomial. For example, from [1], we obtain
Z 2T X (k − 1)Ω(m)+Ω(n) g(m)g(n) BT (m, n)2
|ζ( 21 + it)N ( 21 + it, k − 1)|2 dt = T log
T m,n∈N
[m, n] mn
(8) + o(T ),
LOWER BOUNDS FOR MOMENTS OF ZETA AND L-FUNCTIONS REVISITED 7
for a constant B. We must now bound the main term above. While one can work out an
asymptotic for this main term, we give a quick proof of an upper bound, which is all that is
needed in Proposition 2.
Write BT (m, n)2 BT (m, n)2 z dz
1
Z
log = ,
mn 2πi |z|=1/ log T mn z2
so that the main term in (8) becomes
T
Z X (k − 1)Ω(m)+Ω(n) g(m)g(n) BT (m, n)2 z dz
.
2πi |z|=1/ log T m,n∈N [m, n] mn z2
By the triangle inequality, we may estimate the above by
X (k − 1)Ω(m)+Ω(n) g(m)g(n) (m, n)2 z
(9) ≤ 3T log T max .
|z|=1/ log T
m,n∈N
[m, n] mn
We can now analyze the sum over m and n in (9) by adapting the argument of the previous
section. Thus decompose m = m2 · · · mℓ and n = n2 · · · nℓ where mj and nj are composed
only of the primes in (Tj−1 , Tj ) and Ω(mj ) and Ω(nj ) are both ≤ Kj . By multiplicativity,
the sum in (9) factors as
ℓ
Y X (k − 1)Ω(mj )+Ω(nj ) g(mj )g(nj ) (m, n)2 z
(10) .
j=2 mj ,n j
[m j , nj ] mn
Ω(mj ),Ω(nj )≤Kj
As before, we handle these terms by first dropping the condition on Ω(mj ) and Ω(nj ), and
then bounding the error in doing so. If we drop the conditions on Ω(mj ) and Ω(nj ) the sums
over mj and nj become
∞
Y X (k − 1)a+b −|b−a|z Y (k − 1)2 + 2(k − 1)p−z 1
p = 1 + + O
T ≤p≤T a,b=0
a!b!pmax(a,b) T ≤p≤T
p p2
j−1 j j−1 j
Y k2 − 1 log p 1
= 1+ +O + 2 .
Tj−1 ≤p≤Tj
p p log T p
The error incurred in dropping the conditions on Ω(mj ) and Ω(nj ) is bounded in magnitude
by
∞ X
−Kj
X g(m)g(n)
Ω(mj )+Ω(nj )
mn 1/ log T
−Kj
Y X ea+b a/ log T
≤e e ≤ e 1 + 2 p
m ,n
[m, n] (m, n)2 T ≤p≤T a=1 0≤b≤a
a!b!p a
j j j−1 j
∞ a
Y X e X 35e
≤ e−500Pj 1 + 35 a
≤ e−500Pj
exp ≤ e−400Pj .
Tj−1 ≤p≤Tj a=1
a!p T ≤p≤T
p
j−1 j
8 WINSTON HEAP AND K. SOUNDARARAJAN
5. Proof of Proposition 3
Recall from (2) and (3) the definitions of Nj (s, α) and N (s, α). The following simple lemma
is the key to establishing Proposition 3.
Lemma 1. For 2 ≤ j ≤ ℓ
1
2 −Kj
|Nj ( 12 + it, k − 1)Nj ( 12 + it, k) | ≤
k |Nj ( 12 , +it, k)|2 (1 + O(e 2/k
/k)) + O 2 Qj (t) ,
so that
K
X zr
(12) = ez 1 + O(e−K ) .
r=0
r!
Consider first the case |Pj ( 21 + it)| ≤ Kj /10, where three applications of (12) show that
2
|Nj ( 21 + it, k − 1)|2 |Nj ( 12 + it, k)| k = exp(2kRePj ( 21 + it)) 1 + O(e−Kj /k)
1 2 −Kj
= |Nj ( 2 + it, k)| 1 + O(e /k) .
K (1+1/k)
is a short Dirichlet polynomial (since Tj j ≤ T 1/10 ), and so by (14)
Z 2T X (Kj + r)!2 g(n)2
|Pj ( 21 + it)|2(Kj +r) dt = (T + O(T 1/2 ))
T n
Ω(n)=Kj +r
p|n =⇒ Tj−1 ≤p≤Tj
Kj +r
≤ (Kj + r)!Pj (T + O(T 1/2 )),
where the last bound follows upon noting that g(n)2 ≤ g(n). Using this bound in the
definition of Qj (t), we find
Z 2T 12 2Kj KXj /k
2e 2r K +r
(15) Qj (t)dt ≪ T (Kj + r)!Pj j .
T Kj r=0
r+1
Stirling’s formula and a little calculus shows that the terms above attain a maximum
p for r
around the solution to r 2 = 4Pj (Kj + r), and since Kj = 500Pj , such r satisfies 2 Pj Kj ≤
p
r ≤ 2.1 Pj Kj . It follows that the right side of (15) is
12 2Kj K 2P K Kj √
j j j
≪T e2.1 Pj Kj ≪ T e−Kj .
Kj k e
We need one more observation for the proof of the proposition. Suppose we are given R
Dirichlet polynomials
X
Aj (s) = aj (n)n−s ,
n∈Sj
where the sets Sj satisfy the following two properties: (i) If j1 6= j2 then the elements of Sj1
are all coprime to the elements of Sj2 , and (ii) R
Q
j=1 nj ≤ N for all nj ∈ Sj . The coprimality
condition implies that there is at most one way to write n = R
Q
j=1 nj with nj ∈ Sj . Thus
applications of (14) give
R R
1 2T Y
Z X X Y 2
|Aj (it)|2 dt = (1 + O(NT −1 log N)) a (n )
j j
T T j=1
n≤N n=n1 ···nR j=1
nj ∈Sj
R X
Y
−1 2
= (1 + O(NT log N)) |aj (nj )|
j=1 nj ∈Sj
R 2T
1
Y Z
−1 2
(16) = (1 + O(NT log N)) |Aj (it)| dt .
j=1
T T
LOWER BOUNDS FOR MOMENTS OF ZETA AND L-FUNCTIONS REVISITED 11
We are now ready to combine the above observations to prove Proposition 3. Applying
Lemma 1 we find
Z 2T
2
|N ( 21 + it, k − 1)|2 |N ( 12 + it, k)| k dt
T
Z 2T ℓ
Y
≤ |Nj ( 12 + it, k)|2 (1 + O(e−Kj /k)) + O(22/k Qj (t)) dt.
T j=2
Applying the mean-value theorem for Dirichlet polynomials (14), we see that
Z 2T X k 2Ω(n) g(n)2
|Nj ( 12 + it, k)|2 dt = (T + O(T 1/2 ))
T n
p|n =⇒ Tj−1 ≤p≤Tj
Ω(n)≤Kj
1/2
Y k2 1
≤ (T + O(T )) 1+ +O 2 .
Tj−1 ≤p≤Tj
p p
By modifying the argument of Proposition 3 (indeed the details are even a little simpler) the
second term above may be bounded by
Y k2 k 4 2k−1
2k
Y k 2 2k−1
2k
≪ T 1+ +O 2 ≪ T 1+ .
T ≤p≤T
p p T ≤p≤T
p
1 ℓ 1 ℓ
where q is a large prime, and k > 0. Or, that for k > 0 and large X
♭
X k(k+1)
(19) |L( 21 , χd )|k ≫k X(log X) 2 ,
|d|≤X
where the sum is over fundamental discriminants d. Previously, (18) and (19) were accessible
for all k ≥ 1 by [7], and (18) was known for rational 0 ≤ k ≤ 1 by the work of Chandee and
Li [2]. A third example is the family of quadratic twists of a newform f , where the second
moment of the central L-values is not known. Here one can establish
♭
X k(k−1)
(20) L( 21 , f × χd )k ≫k X(log X) 2 ,
|d|≤X
for all k ≥ 1. Such a result would be accessible also to the method of [7], but the problem
of obtaining satisfactory lower bounds for the small moments k < 1 (which is connected to
the delicate question of non-vanishing of L-values) remains open.
References
[1] R. Balasubramanian, J. B. Conrey, D. R. Heath-Brown, Asymptotic mean square of the product of the
Riemann zeta-function and a Dirichlet polynomial, J. Reine Angew. Math. 357 (1985), 161–181.
LOWER BOUNDS FOR MOMENTS OF ZETA AND L-FUNCTIONS REVISITED 13
[2] V. Chandee, X. Li, Lower bounds for small fractional moments of Dirichlet L-functions, Int. Math. Res.
Not. 2013, no. 19, 4349–4381.
[3] W. Heap, M. Radziwill, K. Soundararajan, Sharp upper bounds for fractional moments of the Riemann
zeta function, Quarterly J. Math. 70 no. 4 (2019), 1387–1396.
[4] D. R. Heath-Brown, Fractional moments of the Riemann zeta function, J. London Math. Soc., 24, no. 1
(1981), 65–78.
[5] A. P. Laurinchikas, Moments of the Riemann zeta-function on the critical line. Math. Notes of the Acad.
of Sciences of the USSR, 39 (1986), 267–272.
[6] M. Radziwill, Large deviations in Selberg’s central limit theorem, preprint arXiv:1108.509v1 (2011).
[7] M. Radziwill, K. Soundararajan, Continuous lower bounds for moments of zeta and L-functions, Math-
ematika, 59 no. 1 (2013), 119–128.
[8] M. Radziwill, K. Soundararajan, Moments and distribution of central L-values of quadratic twists of
elliptic curves, Invent. Math. 202 no. 3 (2015), 1029–1068.
[9] M. Radziwill, K. Soundararajan, Selberg’s central limit theorem for log |ζ( 12 + it)|. Enseign. Math. 63
(2017) 1–19.
[10] K. Ramachandra, Some remarks on the mean value of the Riemann zeta function and other Dirichlet
series, Ann. Acad. Sci. Fennicae, 5 (1980), 145–158.
[11] K. Ramachandra, Some remarks on the mean value of the Riemann zeta function and other Dirichlet
series. II Hardy-Ramanujan J., 3 (1980), 1–24.
[12] K. Soundararajan, Moments of the Riemann zeta function, Ann. Math. 170 (2009) 981–993.