You are on page 1of 4

Topics in analytic number theory, Lent 2013.

Lecture 16: Zero-detector II


Bob Hough
April 19, 2013
Reference for this lecture: Bombieri Section 6.
Recall that in last lecture we proved the following lemma, which guarantees
that near a non-exceptional zero of L(s, ), some derivative of F 0 /F (s, ) is
large.
Lemma (Bombieri Lemma A). Let mod q primitive, q T , |w| T , and
also 2 |w| in the case that = 1. Let log1 T r < 1012 . There exists fixed
C > 0 such that if K > Cr log T , and if L(s, ) has non-exceptional zero 0
with |0 (1 + iw)| < r then for some k [K + 1, 2K],
  
 
k+1
k
1
d
F0
1


(1 + iw + r, )
.

k! ds

F
400r
Our goal in this lecture is to turn the lower bound from Lemma A into a
large sum over primes. We prove the following statement.
Lemma 16.1 (Bombieri Lemma B). Keep the conditions on , q, T, w, 0 , r
from Lemma A. There exist constants A, B, C, D > 1 such that, if x > T B then

2

Z xA X

dy
a
(p)
p


D(log x)3 xCr .

1+iw y
p
x
xp<y

Here ap = log p if there is no exceptional zero, and ap = log p(1 +
is an exceptional zero for L(s, 1 ).

1 (p)
p11

) if 1

We make several comments.


We are most interested in the case where x = T B is a polynomial in T , and
r is of size about 1/ log T , so that we are very near the 1 line. In this case,
xCr = eBC is just a constant, so that we may think of (log x)3 xCr as
large.
It will be important to the argument to follow that x > T , so that we
are making a statement about large primes. This implies that for all of
conductor < T we have (p) 6= 0 for all primes in the sum, which permits
a strengthening of the large sieve inequality (next lecture).
1

The statement of Lemma A could be translated directly to a statement


about primes because the logarithmic derivative is evaluated to the right of
the 1 line. However, note that in Lemma A, the derivative k depends upon
both w and . The chief advantage of Lemma B is that this dependence
is removed, so that the new quantity is suitable for averaging in w and .
We need a pair of preliminary estimates.
Lemma 16.2. We have the bounds
X log p
= O(log x)
pn
pn <x
and

X X log p
log x
= O( ).
n
p
x
pn >x

n2

Proof. The first of these is just partial summation from the prime number theorem.

P
log
p
X log p Z x d
n
p <y
=
n
p
y
2
pn <x
P
Z xP
Z x
dy
pn <y log p
pn <x log p
+
dy
=
O(1)
+
O(
) = O(log x)
=
2
x
y
y
2
2
For the second (here we write log2 x to mean logarithm to the base 2 of x)
X X log p
X log p
1
1
=
(1 + + 2 + ...) +
n
2
p
p
p p

pn >x

n2

p x

2<nlog2 x

x n p<x n1

1
1
log p
(1 + + 2 + ...).
n
p
p p

Bounding the sums over primes with sums over integers, this is


X log m
+
m2

2<nlog2 x

m> x

X
1

log m
log x
 +
mn
x

2<nlog2 x

x n <mx n1

For k = 0, 1, 2, ... we introduce the function Hk (u) =

log x
1
x1 n

log x
= O( )
x

uk u
.
k! e

Lemma 16.3. Assume k > 1010 . If u < 108 k then Hk (u) < 103k . If
u > 1010 k then Hk (u) 103k eu/2 . Also, Hk0 (u) = Hk1 (u) Hk (u) so that
|Hk0 (u)| 1.
Proof. The first two statements follow easily from Stirlings approximation, and
the latter is elementary.
Proof of Lemma 16.1. (See Bombieri pp. 46-48.) For <(s) > 1,
(1)k
k!

d
ds

k 


F0
1 X (pn ) log p(n log p)k 1 (pn ) log p(n log p)k
(s, ) =
+
F
k! p,n
pns
pn(s+(11 ))

Here the second term is to be included only in case of an exceptional zero (we will
assume this case throughout, although it makes no difference to the argument).
Specialize to s = 1 + iw + r, and let K > cr log T , so that by Lemma A, there
exists k, K < k 2K with
  

k
1
d
F0
1


(s, )
.

k! ds
(400r)k+1
F
Then




X (pn ) 
1 (pn )
(nr log p)k 1 1

1 + n(1 ) log p
.
(1)


1

k!pnr r 400k+1
pn(1+iw)
p
p,n


n
1 (p )
Well write apn = 1 + pn(1
log p (this matches ap above) and notice that
1)
(nr log p)k
k!pnr

= Hk (nr log p).


We now trim off the small and large primes, and the prime powers. We have




n
X
X


n
2 log p
ap (p )
103k

H
(nr
log
p)
k


n(1+iw)
pn
p


n
8
8
nr log p<10

p exp(10

k/r)

k
1
1
= O(103k ) <
r
4 r400k+1
if k is larger than a sufficiently large constant. Similarly,




n
X
X


n
a
(p
)
2 log p
p

Hk (nr log p) 103k

n(1+iw)
n(1+r/2)
p
nr log p>1010 k p

pn >exp(1010 k/r)
103k (4/r + O(1))

1
1
4 r(400)k+1

P
2 log p
since the sum over primes is bounded by p,n pn(1+r/2)
= 0 /(1 + r/2).
The remaining sum over prime powers is bounded by
X

n2 exp(108 k/r)pn

k
k
2 log p
= O( exp(
))
pn
r
2 108 r

<exp(1010 k/r)

and this is less than 14 1r 4001k+1 once r is less than a sufficiently small constant.
We deduce from (1) and the above bounds that




X


a
(p)
1 1
1
1
p

Hk (r log p) (1 3/4)

.

1+iw
k+1
r 400
r 1000k+1
108 krn log p1010 k p

Now we remove the weight Hk . Set
S(y) =

X
exp(108 k/r)<p<y

ap (p)
.
p1+iw

Notice S(y) = O(log y). The above sum over primes is the integral
Z

exp(1010 k/r)

Hk (r log y)dS(y)
exp(108 k/r)

= Hk (1010 k)S(exp(1010 k/r))

exp(1010 k/r)

exp(108 k/r)

S(y)Hk0 (r log y)r

dy
.
y

The evaluation term is


k
O(103k exp(1010 k/2) ),
r
which is negligible. We deduce that
Z
Z
exp(1010 k/r)
exp(1010 k/r)
1
1
dy
dy

0

S(y)H
(r
log
y)r
r|S(y)|


k
exp(108 k/r)
r 2000k+1
y
y
8
exp(10 k/r)
Let x = exp(108 k/r). This gives that
18

1 C 0 r
x

r2

x10

|S(y)|
x

dy
.
y

Recall that we assume that r > 1/ log T > 1/ log x. Thus


0
0
1
1 C 0 r log x
eC r log x (log x)2 e(C +2)r log x .
e
= (log x)2 2
2
2
r
r (log x)

Thus, by Cauchy-Schwarz,
18

4 2(C 0 +2)r log x

(log x) e

x10

|S(y)|
x

and therefore

18

2 dy

x10

dy
,
y

18

x10

|S(y)|2

dy
D(log x)3 xCr
y

as claimed.
The above lemma puts us in position to obtain an estimate for the number
of zeros by bounding
2



X X Z T Z xA X ap (p) dy


dw.

p1+iw y
T
x

qT mod q
xp<y
on average over and w. To do so, we will require an approximate orthogonality
property of the primitive characters to different moduli. This is the subject of
the large sieve, which we introduce next.

You might also like