Topics in analytic number theory, Lent 2013.
Lecture 12: Introduction to Linniks Theorem
Bob Hough
February 25, 2013
Reference for this lecture: Davenport Chaps 20, 22. Bombieri, Le grand
crible dans la theorie analytique des nombres.
We continue to fix > 0 the small constant from Theorem 10.2, such that
for all Q, among all primitive Dirichlet characters to modulus q Q, at most
one function L(s, ) has a zero in the region
{s : <(s) > 1
, |=(s)| < Q}.
log Q
Theorem 12.1 (Prime number theorem in arithmetic
progressions). There
exists a constant c > 0 such that, for all q < exp(c log x) and a with (a, q) = 1,
X
p
(a)x
x
+ O(x exp(c log x)).
log p =
(q)
(q)
px
pa mod q
Here the ed term is to be included only if there exists primitive real character
mod q 0 with q 0 |q and such that (s, ) has real zero > 1 log q .
Let A > 0. If q < (log x)A then there exists cA > 0 such that
X
p
x
+ O(x exp(cA log x)).
log p =
(q)
px
pa mod q
Proof. By orthogonality of characters,
X
X
X
1
log p =
(a)
(p) log p.
(q)
px
pa mod q
mod q
px
Let 0 denote the primitive character inducing (we take 1 inducing the principal character). Then the above is
X
X
1
(a)
0 (p) log p + O(log q)
(q)
mod q
px
Let q < T < x. Inserting the explicit formula, we obtain
X x
X
x(log x)2
x
1
(a)
+ O(
) + O( x log x).
T
(q) (q)
0
mod q
(, )=0
||> 2 log
x
||<T
Note that in the case = 0 is principal, (s, 0 ) = (s), and we have pulled
out the contribution of the pole at 1. In the case that there is an exceptional
zero for one of the real characters (there is at most one) pull it out from the
0
sum over zeros. The remaining zeros 0 = 0 + i 0 satisfy 0 1 logc T . Thus,
excluding the exceptional zero, each sum over zeros is bounded by
c0
O(x1 log T
(,0 )=0
||> 2 log
x
||<T
c0
1
) = O(x1 log T (log x)2 ).
||
Choosing log T = c00 log x as in the proof of the prime number theorem optimizes the error and completes the proof of the first statement.
To prove the second, recall that Siegels theorem guarantees (ineffectively)
1
that the exceptional zero satisfies 1 c()q . Choose <
2A to obtain
c
A
1 log x , so that the exceptional zero term is O(x exp(cA log x)), which
completes the proof.
Introduction to Linniks Theorem
The Generalized Riemann Hypothesis asserts that for all , (, ) = 0 implies
<() = 21 (also this statement for ). Plugging this bound in
for the sum over
zeros in the proof of PNT in AP above, and choosing T = x, we find that
GRH implies the asymptotic formula
X
x
log p =
+ O( x(log x)2 ),
(q)
px
pa mod q
which is a genuine asymptotic formula when q < (log xx)2 . Thus GRH will tell us
that there exists a prime p a mod q, p < x if x q 2 (log q)4 . (The notation
A B means there exists a constant C such that A > CB). Unfortunately, we
do not know that the GRH holds, and even assuming that there does not exist
a Siegel zero, the best that we can say based on the PNT in AP is that there
exists p a mod q with p = O(exp(C(log q)2 )). Linnik proved the following
theorem which nearly recovers the GRH bound.
Theorem 12.2 (Linniks Theorem). There exist effectively computable constants c1 and c2 such that, whenever (a, q) = 1, there exists a prime p a mod q
with p < c1 q c2 .
The current record for c2 is about 5.2.
Evidently PNT in AP does not give a good result regarding the least prime
in an arithmetic progression because we have put in an L bound for the
sum over zeros (that is, we have just used the zero-free region). Linnik wins
by showing that the zero sum is much smaller on average. In particular, the
following quantity N (, T ; ) is a natural object of study.
Definition 13. Let q T , let mod q be a primitive character, and let
< 1. Define
N (, T ; ) = #{ = + i : (, ) = 0, > , || < T }.
2
1
2
We follow Bombieris treatment of Linniks Theorem, for which he proved
the following bound on N (, T ; ).
Theorem 13.1 (Bombieri). There exists c > 0 such that, for all T > 2, and
all 1/2 < 1,
X X
N (, T ; ) = O(T c(1) ).
qT mod q
Here the star on the sum over characters indicates that only primitive characters
are to be included.
If there exists a real primitive character 0 mod q 0 with q 0 < T , such that
(s, 0 ) has zero > 1 log T , then we have the improved estimate
X X
N 0 (, T ; ) = O(T c(1) (1 ) log T ).
qT mod q
Here the
cluded.
on N 0 (, T ; ) indicates that the single exceptional zero is to be ex-
Theorems of this type are called zero density theorems, since they show
that there are relatively few zeros in some region to the right of the half line
(there should be none).
There are two essential remarks concerning Bombieris theorem. The first is
that relatively straightforward methods will yield a bound of shape
X X
N (, T ; ) = O(T c(1) log T )
qT mod q
and, indeed, with a better constant c than will be available here. We may
discuss how to prove bounds of this type later in the course. The crucial feature
for Linniks theorem is that the factor of log T is absent, so that the bound is
very good if 1. This feature was already used by Linnik in his proof.
The second remark is that in the presence of an exceptional zero, the available bound for the remaining zeros is improved. This effect is known as the zero
repulsion phenomenon or Deuring-Heilbronn phenomenon, and it is what permits the method to produce effective constants. The second bound of the theorem is new in Bombieris work, although repulsion ideas had also been used in
earlier proofs of Linniks theorem. We record one consequence of the repulsion.
Corollary 1 (Bombieri, Theorem 16). Let T > 2. If there exists primitive real
0 mod q 0 with q 0 < T , such that L(s, 0 ) has exceptional zero 0 > 1 log T ,
then all other zeros = + i, || < T , of all primitive L-functions L(s, )
with mod q, q < T satisfy
1 c1
log
c2
(1 0 ) log T
log T
Notice, in particular, that if 1 0 < T , then we have a pseudo-Riemman
Hypothesis for the remaining zeros, namely that all other zeros of all L-functions
of conductor at most T satisfy a zero-free region up to height T that is bounded
away from the half-line by a constant depending only on .
Proof. By Bombieris Theorem, if
log(c0 (1 0 ) log T ) + c(1 ) log T < 0.
then
X X
N 0 (, T ; ) < 1
qT mod q
Unwrapping this condition, we find that it is sufficient that
1
log
c00
(1 0 ) log T
c log T
