Académique Documents
Professionnel Documents
Culture Documents
M E R O M O E P H IC FU N C TIO N S
By w.
k. haxm an.
1963
T H E T H E O R Y OF L A M I N A R B O U N D A R Y L A Y E R S
IN C O M P R E SSIB L E F L U ID S
By K. SIEWASTSON. 1964
C LA SSIC A L H A R M O N IC A N A L Y S I S A N D
L O C A L L Y COMPACT GROUPS
By H. EBITER. 1968
Q U A N T U M - S T A T I S T I C A L F O U N D A T I O N S OF
C H E M IC A L K IN E T IC S
By s.
golden.
1969
CO M P LE M E N TA R Y V A R IA T IO N A L PR IN C IP LE S
By
a . m. a e t h u b s .
1970
V A R IA T IO N A L P R IN C IP L E S IN H E A T T R A N SF E R
By MAURICE A. BIOT. 1970
P A R T IA L W A V E A M P L IT U D E S AND R ESO N AN CE POLES
By J. H a m i l t o n and
b. tromborg.
1972
THE DISTRIBUTION OF
PRIME NUMBERS
Large sieves and zero-density theorems
BY
M. N . H U X L E Y
OXFORD
AT TH E CLAREN D O N PRESS
1972
NEW YORK
IB A D A N
BOM BAY
TORONTO
N A IR O B I
CALCU TTA
K U A L A LU M PU R
M E LB O U R N E
D A K ES SALAAM
MADRAS
SIN G A PO R E
W E LLIN G TO N
LUSAKA
KARACHI
HONG K O N G
A D D IS A B A llA
LA H O R E
TO K Y O
DACCA
TO
T H E M E M O R Y OF
P R O F E S S O R H. D A V E N P O R T
PREFACE
T h i s book has grown out o f lectures given at Oxford in 1970 and at
CONTENTS
PART
I.
INTRODUCTORY
RESULTS
1. Arithmetical functions
3. Characters
10
4. P olyas theorem
14
5. Dirichlet series
18
6. Schinzels hypothesis
23
28
32
9. Franels theorem
36
P A R T II.
THE
THEOREM
40
45
50
55
58
61
66
70
PART
III.
THE
NECESSARY
TOOLS
73
79
84
89
93
CONTENTS
PAET IV.
ZEROS AND
PRIM E N U M B E R S
98
103
107
110
114
118
N O T A TIO N
123
B IB LIO G R A P H Y
124
IN D E X
127
PART
Introductory Results
A R ITH M E TIC A L
FU N CTION S
I. 110
T h i s chapter serves as a brief resume o f the elementary theory o f prime
m _
( 1,1)
where t h e ^ are primes in increasing order o f size, and the ai are positive
integers. W e shall reserve the letter p for prime numbers, and write a
sum over prime numbers as 2 ancl a product as JT- The p roof o f
p
i>
unique factorization rests on E uclids algorithm that the highest com
mon factor (m, n) o f two integers (not both zero) can be written as
(m,n) = m u + n v,
( 1.2)
where u, v are integers. W e use (m, n) for the highest common factor
and \m, n\ for the lowest common multiple o f two integers where these
are defined.
Let # be a positive integer. Then the statement that m is congruent
to n (m od#), written m = n (m od#), means that m n is a multiple o f q.
Congruence m od q is an equivalence relation, dividing the integers into
q classes, called residue classes m od#. A convenient set o f representa
tives o f the residue classes mod q is 0, 1, 2,..., q I. The residue classes
m od# form a cyclic group under addition, and the exponential maps
m -> eQ(am),
where a is a fixed integer, and
(1.3)
IN T R O D U C T O R Y RESULTS
e(a) = exp(27ria),
eg(a) = exp(27ria/g),
1.1
(1.4)
( i -5>
where the summation is over a complete set o f representatives o f the
residue classes m od# (referred to briefly as a complete set o f residues
m od#). I f on the left-hand side o f eqn (1.5) we replace to b y t o + 1 , the
sum is still over a complete set o f residues, but it has been multiplied
b y ea(a), which is not unity unless a = 0 (m od#). The sum is therefore
zero unless a = 0 (m odg), when every term is unity. Interchange o f a
and to leads to a corresponding identity for the sum o f the images o f m
under a complete set o f maps (a 0, 1,..., q 1). These identities arise
because the images lie in a multiplicative not an additive group.
From E uclid s algorithm comes the Chinese remainder theorem: if
to, n are positive integers and (m, n) = 1, then any pair o f residue classes
a (mod to) and b (mod?i) (which are themselves unions o f residue classes
modwm) intersect in exactly one class c (modtow), given by
c = bmu-\-anv (mod tow)
( 1,6)
(1.7)
whenever (m ,n ) = 1.
(1.8)
(1.9)
for all positive integers m and n seems more natural; if eqn (1.9) holds
as well as (1.8) th e n /(to ) is said to be totally multiplicative, but (1.8) is
the property fundamental in the theory.
1.1
A R IT H M E T IC A L FU N C T IO N S
2 *
a mo dm
I-
(i-io )
2 *
a mods
eq(am)
( 1 . 11)
where
a = c^q^+c^q^,
( 1.12)
Si'^2+ 0'2% ~ 1;
(1.13)
note that % (m od#]) and uz (m od#2) are reduced residue classes, that
< W m) =
2 *
2 *
aiinod<?i agTnodfo
e9s( 2 2)eai(a i% ),
(1.14)
/(to ) = ms,
(1-15)
s CT+it,
(1.16)
(L17)
These include the trivial map for which x(m ) 1 f r each reduced class m.
W e turn these maps into arithmetical functions by defining
x(m) = 0
if (to, q) > 1.
(1-18)
W ith this definition, eqn (1.17) still holds. W e have now assigned a
complex number to each residue class mod q. Hence we have constructed
1.1
IN T R O D U C T O R Y RESULTS
and
(1.19)
H>) = 2 M ,
d\m
( 1-20)
h(m) = '2,f(d)g(mjd).
(1-21)
d\m
W e shall consider eqn (1.21), since (1.20) is a special case, and (1.19) is
evident. W hen (m,n) 1, the divisor d o f mn can be written uniquely
as d = ctb, where a |to and b \n, and (a, b) = 1. Hence
h(mn) 2 f(^ )s imnld)
d\mn
= H 2 f ( ab)g(mnlab)
a\m b\n
= 2
( i - 22)
a\m b\n
(1.23)
d\n
(1.24)
d\m
for primes p
for prime powers p a with a > ]
}.
(1.25)
1.1
A R IT H M E T IC A L F U N C TIO N S
')}
= n a - i ) = o,
(1.26)
% m =1>
t f m > 1.
(1.27)
K
'
(1.28)
? n \n
m\n
ni\n
= 2 m
d\n
d\{nfm)
m\(nld)
f (d)
K m)
(i-29)
when (in, n) 1.
(1.30)
The simplest examples are log to and the number o f prime factors o f m.
There are useful arithmetical functions that are neither multiplicative
nor additive. W e shall make much use o f A(m), given by
/[im \
|0
i f m is a prime power p a, a ^ 1,
if m is not a prime power.
(131)
(1.32)
W e could have used eqn (1.32) to define A(m) and recovered the defini
tion (1.31) b y M obius s inversion formula (1.28).
2
SOME
SUM
FU N CTIO N S
2 A(m).
(2.1)
Our object is to express the sum function as a smooth main term (a power
o f x or o f log x, for example) plus an error term. In place o f the cumber SOme
l/(OI = 0{cj(x)),
f(x )
g(x),
(2.2)
(2.3)
]T 111 =
= lo g (M + l)+ y + 0 (M ~ i).
(2.5)
(2.6)
( )
1.2
SOME SUM F U N C T IO N S
Hence
and 80
9 (to) = 2 2
r=1 d|m
fZ|r
m
(2.8)
2 2 2 mW
m<# r=X d|m
d\r
/ (^)
m
2
m<a
?=!
2 ^(d)
d^x
2
m<$
ml d
m = 0( m o d d )
= 2 /*(<*)n^xjd
2 n>
(2-9)
( 2 . 10 )
in2
d
d<aj
= i 2 ^
^ | r + 0() + O (a;loga:)
1
= cx 2+ 0 (xlogx),
(2.11)
2 d(m),
(2.12)
m ^x
we find that
D (x ) =
2 2 1
m*^x r\m
r^ x '
(2.13)
by (2.5). The error term here is much larger in proportion to the main
853518S
IN T R O D U C T O R Y RESULTS
1.2
2 1
QV^X
= 2
2 i + 2
2 i- 2
2 i
q^y r^xla
r^yq^x/r
q^y >'<?/
= 2 2
{x !q + 0 (l))~ ~ y*
a^u
(2.14)
(2.15)
for each t > 0, and so the error in (2.14) must often be very much greater
than any individual step in the value o f D(x).
To illustrate more difficult examples, we consider
2 d2(m)/m.
(2.16)
m ^x
2 dr - M
uv^x
= ~
+ 0(x\ogr-*x)>
(2.17)
=
m
D i(x ) i p Aog3^'
m ( m + 1)
I x
= <&log4a ;+ 0 (log3a:).
(2.18)
1.2
S O M E SU M F U N C T I O N S
(2.19)
d2(m) =
ch{u)b{v)'
(2.20)
uv=m
Since
l ) a ( a + l ) = (a + 1 )2,
(2.21)
6(m) = H n)
(0
(2.22)
__.
t2
Z-4
n
uK;Xjt2
co
have
70, ,
2
. ,
=
i11^0 this, we
^ 2+ o (l))lo g % .
\4t72
(2.24)
W e require (2.24) and upper estimates for similar sums in the later work.
Any
estimate for a sum involving divisor functions that we quote
will be a corollary o f (2.14) or o f (2.24), possibly using partial summation.
The method we employ in this chapter can be summarized as follows.
To work out a general sum function
2
F(x) =
we try to write
f(m ) =
/(
m),
(2.25)
a,(u)b(v),
(2.26)
uv=m
where we have an asymptotic formula for the sum function o f the a(m),
and b(m) is in some sense smaller than a(m ). A necessary condition is that
b(p) = o{\a(p) |).
(2.27)
uii(v),
(2.28)
uv=m
which is o f the form (2.26); and eqn (2.20) is clearly o f this form.
3
CHARACTERS
T h e reduced residue classes m od # form a group under multiplication,
and the characters m od# correspond to the maps to->- x(m ) from this
group to the group o f complex numbers o f unit modulus under multi
plication. W e define a group operation on the set o f the cp(q) characters
m od#: the product X1 X2 f two characters xi and Xz m od# is the map
m -* Xi(m)X2.{m)- The unit o f this group is the trivial character mod#.
W hen the group o f reduced residues and the group o f characters are
each expressed as a direct sum o f cyclic groups, we can see that they
are isomorphic and that the homomorphisms o f the group o f characters
to the complex numbers are given b y % -> x(m)> where m runs through
the reduced residues m od#. Two finite Abelian groups related in this
way are said to be dual.
W e have already seen that
2
mmodg
()= (
\0
11 Cl
0 ( HIOQ^),
(3.D
for the maps to h* eg(am) from the group o f residue classes to (mod #)
under addition to the complex numbers o f unit modulus. W e have
similar results for the characters xim) m od#:
2
x M = l 9(q)
mmodg
(0
^
istriv ia l>
i f ^ is non-trivial.
(3.2)
Here x(m) is non-zero when the residue class to (mod#) is reduced, and
thus when m is a member o f the group o f reduced residue classes under
multiplication. N ow eqn (3.1) has a dual interpretation, in terms o f
residue classes a (m od#) and maps a -> ea(am). This corresponds to
T vim) = (9{q)
X'mod2
(o
if m = 1 (m o d 2)>
i f to ^ 1 (mod #).
/o o\
CHARACTERS
1.3
11
(3.5)
(3.6)
mmo clg
x(m)eg(m).
(3.7)
IN T R O D U C T O R Y RESULTS
12
1.3
x(aK {am )
(3.8)
flm odg
m = tn,
In this case,
a m od a
q = tr,
xia)eq(am) =
am o d g
t > 1.
(3.9)
xia)er{an)-
(3.10)
x ()x (6)ea(m M -
(3.11)
x(a)x(b) 2
(3.12)
a m o d s 6 m od s
EE0IIC6
2
w m od ff
\x(m)?\T( x ) ? =
a m o d g &mod<z
m m odg
ea(ambm).
mmodtz
\x{)\2 = q
am od g
lx ( ) l2-
( 3 . i3 )
CHARACTERS
1.3
13
By
(3.15)
(3.16)
2 *
amodtf
x ( a ) ea (a )
= bm2*
2* X{cfu+bgv)ea(cu)ef{bv)
od f cnioclf/
=
2 *
b mod/
x ( b )ef( b v )
2 *
cmodtf
e f, M
= Xi(v)r(xiK(u)}
(3.17)
c(?0 =
=
amod{/
eu(au) X ^ ( d)
d\a
mu
2 P'W bmo&gld
2 %ld{bu),
d\g
(3.18)
where we have written a = bd. From eqn (3.1) the inner sum is zero
unless u is a multiple o f gjd. Writing h = g/d, we have
o(u ) = 2 ( # )
(3' 19)
|r(X)|2 = / .
(3.20)
4
PO LY A S THEOREM
x (v) = 2 xim)
(4.1)
Then
where the term o (1) is to be interpreted as f -> oo.
(4.3)
|a|| = min|ma|,
(4.4)
where the maximum in (3.3) and the minimum in (3.4) are over all
integers m. W e now state a lemma.
Lemma. The Fourier series
(4.5)
oo
m=0
1.4
P O L Y A S
converges to
15
t h e o r e m
a __[a] _ i
if a
no%an integer,
i f a is an integer,
(4.6)
(4.7)
2 ttw i
m 7^=0
and thus
(4.8)
* M
_-M
nr
in # 0
{2nim)-1e(ma.)-\-^ a =
J _ i\j
e(mt) di
(4.9)
which by one o f the mean-value theorems for integrals does not exceed
in modulus
sin(2Jf-(-l)77i
sin 77a
At
1
2a (2M + \ ) t t
(4.10)
2 ttM cx
(4.11)
rq + l
and so we can replace the sum X (x) by a sum o f the same form with
0 ^ x < q. I f now x > \q, we can replace m by q-\-m and obtain a
sum with 0 ^ x ^ \q, multiplied by x ( 1). Thus we can suppose
that 0 < x ^ \cq in (4.1). W e now use H (a) to construct 0(a), where
n
0() = I \
\0
(4.12)
16
IN T R O D U C T O R Y RESULTS
1.4
(4.13)
W e now have
X(l) + x(2) + ...+ l* (a 0 =
x(m)0(mlq).
(4.14)
m 1
W hen we use eqn (4.7) to truncate the sums for H (a xjq) and H(a) in
(4.14), the total error in modulus is at most
2J
[ ,/ )-! <
1 dh?
m=l
(4.15)
0(a) = 2 a(m)e(ma),
(4.16)
CO
we have to consider
q
2 X(r)
J .= l
es(TOr)-
(4.17)
2 a (m) 2
-M
x0')e9(') =
>=1
2 a(m)x(m)r(x).
(4.18)
Since x is non-trivial a(0)%(0) is zero, and by eqns (4.5) and (4.13), for
\a(m)\ < \TTin\-1.
(4.19)
< ? i 2 | (nm)-1
m=l
W e choose
so that (4.20) is
(4.20)
M = qi+s,
(4.21)
(4.22)
(4.23)
Finally the omitted term \xix ) is 0(1), and we have P olyas theorem
when x is proper mod q.
P 6 L Y A S
1.4
IT
t h e o r e m
x ( ) =
=
<
m^cc
X iM
d\q
d\m
M d)
^(d)Xi(d)
/ } l o g '/ { ^ + o ( l ) } ,
d\q
m ^xfd
Xi()
(4.24)
d\q
(d, f) = 1
m < p i\ogp
(4.25)
u2 = m (m od#)
(4.26)
< #s
(4.28)
for each 8 > 0 holds in place o f (4.25) has not yet been proved or
confounded.
5
D IR IC H LE T
SERIES
(5.1)
4 ( )= 2 a (m )
(5.2)
f(s ) = j s x - ^ A i x ) da;.
(5.3)
Formula (5.3) can be inverted: fro m /(s) we can recover the sum function
A(x) o f the coefficients. Let a and u be positive real numbers. Then
ar-f- i oo
23
(0
a- ic0
\1
if
0 <C 11 <C 1 ,
if =
1,
(5.4)
if U > 1.
1.5
D IR IC H L E T SER IE S
19
The
( 5 gx
Rj logu
Of.
1' 1
F ig .
f
J
a ico
x ^ f i s ) ds = 2 a(m)+a(x),
m<x
(5.6)
where the last term occurs only if x is an integer. There are many
integral transforms from Dirichlet series to their coefficient sums, all
proved b y the same method. The simplest one after (5.6) itself is
a-f ico
h J
2 ( - )<>
a ico
Lis>x) = 2
(5.8)
m 1
2 m ~s>
m 1
(5.9)
IN TR O D U C TO R Y RESULTS
20
1.5
-| r (
a(p )
a ( p 2)
m=1
holds in a half-plane a ^ a in which one side o f eqn (5.10) converges
absolutely. I f the product in (5.10) converges, f(s ) can be zero only
when one o f the factors on the right-hand side o f (5.10) is zero. The
convergence o f the left-hand side o f (5.10) alone does not im ply that o f
the product; L(s, x) with x non-trivial has a series (5.8) converging for
a > 0, but the function itself has zeros in a >
preventing the product
from converging in 0 < a < J-.
The second defining property is that f(s ) should have a functional
equation
f{s)G (s) = f * ( r - s ) G * ( r - s ) ,
(5.11)
1 ( m + 8) - s,
m=l
(5.12)
(5.13)
D IR IC H L E T SE R IE S
1.5
21
X ( l ) m-1TO~~s = (1 21-S)(s)
m1
(5.14)
M (x) =
2 A{m)
(5.15)
2 Mm)>
(5.16)
m <x
m < (G
which are the coefficient sums o f '(s)/(s) and o f l/(s) (we shall prove
this below). The function '(s)/(s) also has a pole o f residue 1 at
s = 1, but l/(s) does not. I f the corresponding contour integrals were
negligible, we should have
t(x ) = x + o ( x ) ,
(5.17)
M (x) = o(x).
(5.18)
CO
co
2 m~s 2 ( / W W / ) = ( 2 ( / ) / *) ( 2 H g ) r s)-
m= 1
f\m
'/= 1
'
(5 19)
(5) 2 K m)m~a =
m =l
CO
2 w s 2 M /) = 1
m =l
/|m
(5'2)
from eqn (1.27). Since eqns (5.3) and (5.6) im ply that expansions in
Dirichlet series are unique, we have shown that the series on the left-hand
side o f (5.20) represents l/(s) wherever it converges. Similarly, using
eqns (5.19) and (1.32) we can check that '(s)/(s) has a Dirichlet series
with coefficients A(m).
For fifty years (1898-1948) the only proofs known o f eqns (5.17) and
(5.18) used contour integration and other complex-variable techniques.
In 1948, Selberg and Erdos gave a real p roof o f (5.18) (see Hardy and
22
IN TR O D U CTO R Y RESULTS
1.5
2 MN/e),
(5.21)
e < iV
where weJiave written m = tie in the first sum. On the other hand, by
expression (2.5),
= N logN ~ N -}-0(logN )
(5.22)
which agrees with the result o f substituting (5.17) into (5.21). In this
way, Gauss was led to conjecture that
(5.23)
2
6
SC H IN ZE L S H Y PO TH E SIS
And all the good things which an animal lilies
Have the wrong sort of swallow or too many spikes.
II. 30
M a n y problems in prime-number theory follow a similar pattern.
2 XJ {1
853518X
l ) " 2} I (log ) 2
-f-error term,
(6.2)
24
IN T R O D U C T O R Y RESU LTS
1.6
a/ \
S (a )=
v
2,
p^N
n.
m o\
(6>3)
i f m 0,
jj
(6-4)
(6.5)
1.
(6.6)
P<N
p = b (modg)
we have
S(a/q) =
ea(ab)ir(N;q,b).
(6.7)
fcniodg
(6.8)
where cq(a) is Ram anujans sum (1.11). I f (a, q) = 1 the Ram anujans
sum is just /x(g), from eqn (3.19). This argument suggests that S(a) has
a spike at a/q o f height proportional to ii(q)l<p(q), that is, that |$(a)|
has a local maximum close to ajq. Now the area under the graph o f
|$(a)|2 near 0 (which certainly is the site o f a spike) may plausibly be
written
2V'"17r2(iV*)-{-error term,
(6.9)
and this can be proved to be true. I f we assume further that all the
spikes at rational points are the same shape, the spikes at rational
points ajq contribute
N^1tt2(N) 2 lJ'2(q){<?(q))~z 2 *
amodg
ea(2a)+error term,
(6.10)
(6. n )
1.6
SC H IN ZE L S H YP O TH E SIS
25
For small q the argument above can be made rigorous; but then part
o f the range o f integration in (6.5) does not support spikes. Away from
a spike we cannot estimate S(a) except b y replacing it b y its absolute
l
value; and the spikes with small q contribute very little to J |$(a) |2 da.
o
In the integral o f |$(a)|3 the spikes do dominate, and b y this method
I. M. Vinogradov was able to prove that every large odd number is the
sum o f three primes.
The approach to Schinzels hypothesis through exponential sums does
lead to an upper bound for the number o f sets x 1}..., xn o f integers not
exceeding N for w h ic h /l v .., f m are all prime. To explain the method
we shall take n = 1, so we are considering integers * in the range
1 ^ x
N for w liich /1(a;),...,/m(a;) are all prime. W e now work modulo
a prime # . Apart from the finite number o f x for which one o f f t(x),...,
f m(x) is p, x must be such that none o f fx(x),...,fm(x) ~ 0 (m odp). This
means that x must be confined to certain residue classes m od#. W e
therefore divide the residue classes m od # into a set H (#) o f /(# ) forbidden
classes and a set K (p ) o f g(p) = # / ( # ) permitted classes; h m od # is
forbidden if and only if one o f the p o ly n o m ia ls /^ ) is a multiple o f # .
I f x falls into a forbidden class for any prime # smaller than each o f the
fi(x), then one at least o f t h e / i(x) cannot be prime.
The values o f x that make f x(x),..., f m(x) primes greater than some
bound Q form a sifted sequence, in the following sense. The increasing
sequence J f o f positive integers % , n 2>... is sifted b y the primes # < Q
if for each prime # < Q there is a set H(p) (possibly empty) o f /( # )
residue classes m od # into which no member o f J r falls. W e shall show
in Chapter 8 that, i f J/' satisfies the above condition, the number o f
members o f J f in any interval o f N consecutive integers is
N
v ... >,...wi, w ..+ error t e r m >
I /**(?)/(?)/?(?)
( 6 - 12)
a<Q
where
f{q) = q H f ( p ) l p ,
(6-!3)
g(i) = ? IT { l - f i v ) h ? } v\a
(6-14)
26
IN T R O D U C T O R Y RESULTS
1.6
eK )
!< N
(6-15)
k eK (q )
ea(ak)l9(q)-
(6' 16)
2 e(ma),
m=l
(6.17)
(6.18)
2*
q^ Q omodg
K (a ,q )F (a -cilq ).
(6.19)
gCQ
= g2 :Q
k eK (q )
(6 .2 0 )
using the definition (1.11) o f Ram anujans sum. W e know the sum
over 1c must be zero if q has a repeated prime factor; i f q is square-free
then
i \
e K(q
n)
k eK
2d\n
d\q
d\(k-m)
M 'I
g\g(g)
d\q
meK(d)
w M )
= pte)9te)M ftei)lgtei),
(6.21)
(6.21) becomes
A<lMlM<l)lf(b)-
(6.22)
1.6
SC H IN ZE L S H Y PO TH ESIS
N ow
27
(6.23)
d=C_Q
meH(d)
ii{d)d
fid )
Z-,
'
^{q)f{q)
g = o(m ocld )
(6>24)
glq)
s<Q
jo W
nm^N 'a^QT
and
'
<6'26>
J |T(/3)|2 dj3,
0
0
(6.26)
f | (a )| 2 d a =
da
then m e H(d)
f j S ( a ) /S ( -a ) d a =
( 6 '2 7 )
0J
f |T (a)|2 da = N ^
o
sQ
+ error term.
(6.28)
W e can prove eqn (6.28) either directly from (6.19) or b y writing ^ ( a )!2
as T(a)T( a) and using (6.4) again and the expression (6.24) for the
coefficients. Selberg used the second method, which can be expressed
entirely in terms o f the manipulation o f the coefficients (6.24), with the
integrations (6.25) and (6.28) occurring only implicitly. The first proof
uses eqn (6.18) to show that the spikes o f T(a) provide the main contribu
tion to the integral in (6.28), and the observation that
\K(a,qW = ^(q)f(q)lg(q).
(6.29)
am odg
7
THE
LARGE
SIEVE
(or for some related expression) is called a large sieve for residue classes.
Other large sieves will appear in Chapter 18. In this chapter we prove
what is probably the simplest o f the upper bounds for the sum (7.2), and
in the next chapter we shall use it to prove an upper bound o f the form
(6.12) for the number o f elements o f a sifted sequence in a bounded
interval.
The p roof does not require that am in eqn (7.1) takes only the values
0 or 1, and it treats (7.2) as a special case o f the sum
f |S(*,)la
(7.3)
}.=i
where 0 ^ x x < x 2 < ... < xR ^ 1. Since ajq occurs in the sum (7.2) only
i f it is in its lowest terms, the points x r are distinct in our proposed
application. W e write
S = min{*2a^Xgx z,...,xx+ l x ^
(7.4)
and suppose that S > 0. Before proving an upper bound for (7.3), we
consider what form it might possibly take. Certainly there will exist
sequences o f coefficients am and points xr for which
b
1
2 |S( ,)|>8- i f |S()|d
= !
J
= 8-1 K P 1
(7-5)
1.7
29
On the other hand, any one term in the sum (7.3) may be as large as
( 2 K n!)2> ari(l ^ ail the
(7.6)
\S(xr) \ * ^ ( N + 8 ~ i ) f \ a m\*
)=i
(7.7)
always holds. Surprisingly, the right-hand side o f (7.7) has the correct
order o f magnitude: we shall prove that
f
\S(x,.)\^ ( i V + p - W 3 + 0 ( l ) ) i
(7.8)
ml2;
r=l
(g, h) = % g j i ns
x
(7.9)
||g|| = ((g,g))*.
(7.10)
f c ,( u ,f W )
r 1
'
|(fM,fM)|)*.
1,...,R 8=1
(7.11)
'
(u , | c,fW)
2 M W
(7.12)
f crcs(w,f) < 1 f
j = l s = l
r=l
r=l
| (Icri+Kia)i(fw,fw)i
s==l
K l2 2
s=1
l(f(,,)>f(8))l
(7.13)
30
IN T R O D U C T O R Y RESULTS
1.7
(7.14)
C o ro lla ry .
r=1
|(f(,,),f (s))|.
(7.15)
1<9'<B s = l
W e write
M
S(u.) = e(M<xJr a) 2 bmc(ffla),
(7,16)
-M
(7.17)
um = b j k m,
(7.18)
for \m\ ^ M
(7.19)
k.m = 0
2 b me(mx,) = \8(xr)\.
l ( , f (,,))l =
Moreover
M
!ll!2 =
and
-M
(7 .2 0 )
N
IKWK, = 2 K l 2
i
( f fW) = K (x s~ x r),
(7.21)
(7.22)
M + L
where
K(a) =
e(ma)
-M -L
(7.23)
Hence we have
J*
S= 1
l-^r(s' '*'') I
2-ZI4"-f--Zv
}2
i(B -l)
Z -1 coseo27TW
CO
< 2 i f + Z + ( 3 i > 2) - 1+ 0 ( i v - i ) .
(7,24)
1.7
31
(7.25)
S-1(l + 270Ar3S3) f
\am\2,
(7.27)
and
2max(AT, S^1) 2
(7.28)
Of these, (7.2 6) is the result o f this chapter, appropriate when N > f 8 _1V3,
and (7.27) and (7.28) are results o f Bombieri and Davenport (1969,1968).
(7.27) is appropriate when S_1 > 3(10)i/V, and (7.28) for the inter
mediate range.
Note added in proof. H. M ontgomery and R . C. Vaughan have now
proved the conjecture (7.7). This supersedes (7.26) and (7.28) but not
(7.27).
8
THE
U P P E R -B O U N D
SIEVE
(8.1)
The g(q) classes o f K(q) are those that for each # dividing q are congruent
m o d # to some member o f J r \their union contains all members o f the
sequence J f.
W e work with the exponential sum o f eqn (6.15):
#( ) =
( ).
(8 '2 )
S(alq)ea{ - a h ) =
am odg
c ^ h)
2
U; < N
= p{q)M,
(8.3)
2* 2 8{alq)eq(-ah).
(8.4)
a m o d g heH (q)
2 *
' am odg
| S ( / ? ) l a) (
heH ( q)
hef-T(n\
'
>
( 8 -5 )
TH E U P P E R -B O U N D SIE V E
1.8
33
The second sum over a on the right-hand side o f (8.5) can be rearranged
as follows.
2*
amodg heH(q)
ea( ah) 2
geH(q)
%iacJ)
geH(q) JisHiq)
05H(q) heH(q)
c fl( 0 A)
2 M v l d)
d\(g-h)
= 'ZMqld)P(q)lf(d).
(8.6)
d\q
In eqn (8.6) we have a sum that is a Mobius inverse (in the sense o f
(1.27)) o f that in
' /*2(})fl,(r) _
rlm
f (r )
(8.7)
m y
an equation that expresses the fact that for a prime modulus p, every
residue class is either in H(p) or in K{p). The terms involving d in
(8.6) therefore come to p?(q)g(q)lf(q), and we can put (8.5) into the form
'V '*
j^
/
J
2
""
q lq )
Jw
Z -i
amodg
Sa
( 8 .8)
W e apply the large sieve (7.8) with the rationals ajq with q ^ Q and
(a ,q ) = 1 as the points x x,..., xR, so that
= (Q (Q -l))-1
(8.9)
2*|-S(/?)|2 < ( ^ + 0 ( e 2) ) ^ .
q<,Q amodg
(8-10)
(8.11)
q^Q
or
M ^
^ + ^ ( ^ 2)
M ^ W
i i i '
q<Q
fo i9\
( '
which is the relation (6.12) with an explicit error term. W e have stated
our result for the interval 1 ;;C n $ ' N o f the sequence J /\ but (8.12)
holds as an upper bound for the number o f members o f J r in any interval
LA-1 < % <c L-\-N) we have to consider merely the integers n L,
where n is in J f , themselves a sifted sequence with the same values o f
f(q ) and g(q) but the sets H(q) and K (q) translated b y L. The deduction
o f (8.12) from (7.8) is due to Montgomery (1968). His argument differs
from that given here; there are several alternative proofs.
34
IN T R O D U C T O R Y RESU LTS
1.8
g( 2) = 2
f ( p ) = U p - 1)>
9(P) = U P + !)
fo r # > 3
(< H -o (i))Q ,
(s.i4 )
v(p
g(q)
- 1 =
( 8 -1 5 )
Z-t <p(q)
a
Tu'
= (l+ o (l))lo g Q ,
(8.16)
2 ^ = 2 fir n K + ? + ~ )
q^Q r
q^Q
P\q '
= 2 iK
t8-1?)
the sum being over all m whose prime factors are at most Q. All m ^ Q
are included in this sum, and thus
<8'I8>
we see
(8.19)
The right-hand side o f (8.19) is just double the true value (5.23). M ore
over, (8.19) is also an upper bound for the number o f primes in any
interval o f length N.
1.8
TH E U P P E R -B O U N D SIE V E
35
lx K (q )
e(cl&)
(8'2)
and this in its turn measures how evenly J f is distributed among the
g(q) residue classes m od# into which it is allowed to fall. W e could add
an explicit term on the right o f (8.8) to measure the unevenness (what
statisticians might term a variance). The inequality (7.8) gives a strong
upper bound for this variance as well as for the main term. W hen we
use (7.8) to prove Bom bieris theorem, it is the variance bound that is
important, not the bound for the main term.
9
FR A N E L S THEOREM
Its just a thing you discover, said Christopher Robin
carelessly, not being quite sure himself.
I. 109
(9.1)
from eqn (2.11). The properties o f the Farey sequence are discussed by
H ardy and W right (1960, Chapter 3).
W e shall sketch a p roof that
fr+l
fr =
( M m ) - 1-
(9.2 )
1- a F ,
(9.3)
so that E(a) is the excess number o f Farey fractions in (0, a] beyond the
expected number aF. Franel considered the sum
S Im fr) I2
r= l
(9.4)
>
F R A N E L S TH EO REM
1.9
37
and showed that it is o (Ql) i f and only i f eqn (5.18) holds; and an upper
bound
for (9.4) with 3 < A < 4 is valid if and only if
\M{x)\
(9.5)
(9-6)
In fact, the ratio o f (9.4) and (9.6) lies between bounded multiples o f F ;
but this fact requires proof, as the Riemann sum corresponding to the
integral (9.6) and the p o in t s /i,...,/F is
f
)=1
(fr+ l-fr) i m W ,
(9.7)
and from eqn (9.2) we see that the difference f r+1f r varies from Q -1
almost down to Q~2. Franel (1924) produced a curious identity for the
sum (9.4). W e shall show in (9.20) that (9.4) is less than a bounded
multiple o f Franels expression involving M (x), and deduce the only i f
clause o f Franels result b y a method o f Landau (1927, Vol. II, pp. 16977).
W e use the function H(a) o f eqns (4.5) and (4.6):
tt, \
# ( ) =
M \
n
(0
i f a is not an integer,
+
it-fa is an integer.
9-8
W e have
_ (F(a.)
i f a is in the Farey sequence,
\ J ? ( ) - i if not,
. .
(J' J)
for 0 < a -<C I. To verify eqn (9.9), we observe that the left-hand side
is a,F plus a step function that is zero at a = 0 and has discontinuities
\ on each side o f the Farey p o in ts /r Taking (4.7) in the form
5 (a ) +
-
J]
m# 0
e(mot)
277-1TO
(9.10)
Q2INI
t=i
V, rt# 0
38
IN TRO D U C TO RY RESULTS
1.9
B y (9.2), the consecutive Farey fractions are at least Q~2 apart, and so
the sum of the error terms in (9.11) is
< Q~z X Q2lt < log -P < log Q,
(9.12)
t= i
where we have used expressions (2.5) and (9.1). We can now replace
the term t = r (since H ( 0) = 0) and rearrange the first term on the
right o f (9.11) as
F
Q2
t~ 1
Q2
{27nm)-1e{mfrmfi)
2
Q2
Q1
m# 0
to #
3 <Q
X*
e (m /^ )e ( a m )
amodg
2 (2wim)-1e(TO/1.) X cq(-).
S o
a<1
(9.13)
We apply the large sieve (7.8) with (9.13) as the exponential sum and
/x,..., fff as the points. By eqn (9.2), S-1 is here Q(Q 1). We have
F
Q2
Q
X (2 7r i m ) - 1e (m /).) X c a( m )
2 cg(m)
X
r= 1 -Q2
to#0
TO
TO##0
Q2 2
2<z
ra=l
2 = 0 (m o d d )
(9.14)
where we have used eqn (3.19) for Ramanujans sum. The coefficient
of Q 2 in (9.14) is now
<
2
d^Q
dM(Qld) X f M (Qlf)
f^Q
m = 0 (m o d tf)
w = 0 (m o d /)
m~2>
(9 J 5 )
where we have taken the sum over to from m = 1 to infinity, since the
coefficient o f to~2 in (9.14) is non-negative. The inner sum in (9.15) is
772
_ tr2 ( d J Y
6 \ d ,ff ~
6 dT
1 '
u(d) = to2.
(9.17)
d\m
(9.18)
tf
(0 (
\ 4 f r S o a ,)
d -'M (Q ld )y.
1
(9.19)
1.9
FR A N E LS THEOREM
39
'
t^ Q
c i-m (Q id )Y + F iog *Q
< Q 2l H
'
d<Q
(Z= 0 (m od t)
KO v d^Q
d -i M ( Q l d ) Y + Q * l o g * Q .
'
d = 0(m od t)
(9.20)
The leading term on the right of (9.20) is essentially the sum involving
Mobius functions o f Franels identity.
The inequality from (9.8) to M (x ) is less involved. Since, by (3.19),
2 * ea(a) = c9(l) = rfq),
(9.21)
am odg
we have
Now
M (Q ) = 2 e(fr)-
(9-22)
r1
\ e (f^ e {r lF ) \^ 2 n \ f - r l F \
< 2 ^ 1 E ( f r)\.
Since
from (3.1), we have
2 ^(rjF) = 0
(9.23)
(9.24)
r=l
|jf(<3)|< 2 ^ 2 TO O!
r=l
< Q - i( z m
\r=i
r) \ f ,
(9-25)
and so a bound for (9.8) of the form o(Q 4) implies the prime-number
theorem in the form (5.18).
853518
t)
P A R T II
T h e P r im e -N u m b e r T h e o r e m
10
A MODULAR
RELATION
I. 46
I n this second part we study (s) and L(s, x) as functions o f the complex
Lem m a.
>
ki l
X
m = -co
(10 .1 )
l,e + 1
f(x )e (m x )d x -y (k )-...-y (k + l)
M
^ lA M -H o g M .
( 10 . 2 )
2.10
A M ODULAR RE LA T IO N
41
y
M
= H
Z-t 27rim
10
mco
v
m^O
if o < * < i,
if x ~ 0 or I,
7)1#0
Hence
/
V2
H ( x ) Jr 2 (277im)_1e(m3;) da;
2 m -1 d a ;+
1'
da;
1I'M
M -1 log M .
(10.5)
We now have
jif, }
2
- m
I f(x)e(mx) da;
J
m#0 0
=
[/(%)
(277-iTO)_1e(?wa;)j J jV'O) 2
m^0
(27rim)_1e(OTa;) d*.
(10.6)
The first term on the right of eqn (10.6) is zero, since the terms for m
and m cancel, and by (10.5) the second is
1
J f ( x ) H ( x ) d x - { - 0 ( A M ^ 1log M ) .
0
(10.7)
j / '( * ) ( * - * ) da = i / ( 0 ) + i / ( l ) ~ J /(* ) d.
o
o
(10.8)
2 f f{x)e(mx)
dx = | /( 0 ) + i /( l ) + 0 (^ ilf-1log Jf),
(10.9)
42
THE P R IM E -N U M B E R TH EOREM
2.10
oo
exp{(to-|-8)27t2:-1} = x i
m = co
m=
exj)( i7i2Trx)e(m8),
(10.10)
co
where x is real and 0 s ' S <1 1. Rather than apply (10.2) directly, we
vise eqn (10.9), so that
M
| exp{(r + i+ S )2^ - 1} di
-M
| e x p { -(r + 8 )2} + i e x p { - ( r + l + ) 2} +
+ 0[ne_1exp{ ( r - ^ S y n x ^ j M ^ l o g Jf],
(10.11)
OO
m = co
e x i ^ { t + d f n x ^ } e { m t ) At
(10.12)
r = co ^
2
y =
e x p { - ( r + 8 ) 277a;-1}.
(10.13)
CO
(10.14)
CO
where we have put t-\-8 = xz. The integral in (10.14) can naturally be
regarded as the contour integral of e~z2 along the line Ixnz = to. To
evaluate the integral, we can move the line o f integration to the real
axis, when (10.14) becomes
CO
a exp(
rrxm2 2mm)
J exp(irxt2) At
CO
= C77~sa;=exp(nxm2 2irim),
(10.15)
co
where
c=
J e x p ( t2) At.
(10.16)
A M ODULAR R E LA T IO N
2.10
43
CO
2
?=
j
7)1
C7r~%*exp(7rm2x)e(mS).
(10.17)
co
8 ( w ),
(1 0 .2 0 )
(10.21)
(10.22)
44
THE P R IM E -N U M B E R TH EOREM
w ith
and
/(o j
2,
OJjJ
2.10
co
= w2/wi(10.23)
162(co2/co1)
= / K . w8)
/(oij, a)2+2co1) cox 102(co2/a)1-)-2) = /(aij, co2).
(10.24)
(10.25)
11
THE
FUNCTIONAL
EQUATIONS
II. 21
r () = / e - v y ^ d y ,
(11.1)
0
valid for cr > 0 (we recall s cr+itf, cr and t real), becomes
A i s)
CO
e~~m7rxx is~1
da;
(H -2)
7 t - } sr ( \ s ) i ( s )
>
q - 2itxx Is - 1
da;.
( 11 .3 )
m= 1 0
Since the sum and integral in eqn (11.3) each converge absolutely, we
can rearrange the right-hand side of (11.3) as
co
CO
J 2
dx =
J 4 (0 (i )-l)* * -1 da;,
(11.4)
>
(11.5)
co
f ^ ( ^ ( i t J -lJ H 8- 1 dt =
f I0(it ) t ~ ^ i dt - K 2 S - 1)
i
CO
= J 4 ( 0 ( i i ) - l ) H - * d i - s - 1- ( l - a ) - 1.
(11.6)
46
THE P R IM E -N U M B E R TH EOREM
2.11
J (0(ia;) l ) ( J - i + a r + ' - * ) d { ( i a) } - i
(1 1 .7 )
The expression (11.7) was obtained under the assumption a > 1, but,
Since
0 (i )-l< e -
(1 1 .8 )
for x > 1, the integral in (11.7) converges for all complex s. Since
r(|s) is a known function, and ( m s ) ) - 1 is integral (single-valued and
regular over the whole s-plane), we can take (11.3) with (11.7) as the
CO
when the series converges. We have now continued l(s) over the whole
plane. Further, (11.7) is unchanged when we replace s by 1s, so that
^ r ( i s ) U s ) = t ^ - ^ T ( ! - 1 S) ( 1 - S),
(11.9)
= 21^7r^T(s)cosis7r,
(11.10)
2 )
(11-11)
We now list some properties of F(s) (see for example Jeffreys and
Jeffreys 1962, Chapter 15). The product
1
r(s+ i)
where y is tbe constant of (2.5), converges for all s, and defines F(s) as
a function tliat is never zero and has simple poles at 0, 1, 2,....
Using this information in (11.7) we see that the pole of (11.7) at 1 comes
from (s), the pole at 0 from -T(^s), and that (s) must have zeros at
s = 2 , 4,..., to cancel the other poles of F(^s). From eqn (11.12),
T (s + 1 ) = sF(s),
and
Jrr( 1-)s)_Z^(1s) =
its cosec
(11.13)
77s,
(11.14)
where we have used the product formula for sin 7rs. We can verify
eqn ( 11. 10) by showing that the ratio of the two sides is a constant.
Equation (11.1) is obtained by evaluation of the limit of
N
C *10+1( 1t/N )N dt
(11.15)
2.11
47
in two ways as N tends to infinity. We can also obtain from (11.12) the
asymptotic formulae
log ics) = (s -P o g s -s + llo g 2 7 r + 0 (l/| s | ),
and
! / /
= lo g s+0(l/|s|)
(11.16)
(11.17)
which hold as |s| co uniformly in any angle 7i+S < args < tt S
for any 8 > 0.
Next we consider an L-function L(s, x) with x a proper character
mod#. There are two cases. I f x(~~l) is 1> we argue as above up to
00
Tr-*sqissr(^ s)L(s, x) =
x(m)e~m*n!>:lQ
a;is-1 2
m=1
00
=
(11.18)
x i ^ '? (x ,x )d x ,
0
co
where
2 x ( m ) e ~m*nx^a'
( p ( x ,x ) =
(11.19)
CO
CO
2 e -( n + 8 )M * =
xh 2 e ~ mS7r-r e ( m S ) .
co
( 1 1 .2 0 )
co
2*
(H.21)
x ( a )ea (a m )>
am o d g
so that
r(xMx>x)
=* 2* r n 2
co
= 2* x{a)(<llxf 2 e^m+a^v2/a!, (11 .22)
am o d g
m = co
am o d g
m ~ co
co
2*
= (?/*)*
co a m od g
?= oo
x ( r ) ^ ,m
(q lx M l/ x ,x ).
(H .23)
This will play the part of the modular relation (10.10). As before, we
split up the range of integration in (11.18) and find that
1
CO
(11.24)
48
THE P R IM E -N U M B E R TH E O R E M
2.11
co
TT~lsqisr(^s)L{s, x ) = i J
x)
x~^-^(x, x) d a .
(11.25)
mmodg
X(m)eq(m ) =
mmodg
x ( - ) e g{ m)
(11.26)
= r(x),
m0dq
(11.27)
We now see that the right-hand side of (11.25) is r(x)g,_i times the
corresponding expression with s replaced by 1-s and x by x> which
gives the functional equation
ir-bqi*r($s)L(8,x) = T(x)q~lTT-l+*sq ^ sr ( ~ s ) L ( l - s , jf).
(11.28)
We now consider characters xim) proper modg with %( 1) = 1.
Since we want to consider a sum from co to co, we use mx(m) in place
o f x(m). Writing s + 1 for s in (11.2), we have
7T-Ks+I)^i(s+I)p (i(s-|-I))i(s, x) =
OD
co
where
p(x, x)
= iJ p (a ? ,X)**-*da!,
0
(11.29)
(11.30)
co
= i ^ x - i p ( l f x , x).
(11.31)
= i | />(*, x ) ^ d x + ^ ) f
CO
P(X>x)*-** d.
(11.32)
2 .II
49
Again, the integrals on the right of eqn (11.32) converge for all s, so that
L (s ,x ) has an analytic continuation; it must have zeros at 1, 3,
5,... to cancel the poles of r (| (s + l)), and satisfies the functional
equation
is
lq i
i s r ( l % 8)L(l s,x)
=
(11.33)
(11.35)
when we write m = dr. The sum over r in (11.35) is L(s, ^i), which has
an analytic continuation since Xi m o d / is proper, and the sum over d
is defined for all /. The corresponding functional equation for L(s, x)
contains the sum over d explicitly. We shall not need this case again.
A number of proofs of the functional equation can be found in
Chapter 2 of Titchmarsh (1951).
12
HAD AM ARD S PRODUCT
FORMULA
plane, with
(12.1)
for some constant A , as |s| -s*oo. The order o f/(s ) is the lower bound
o f those A for which an inequality o f the form (12.1) holds. Hadamard
showed that an integral function o f finite order can be written as an
infinite product containing a factor s p corresponding to each zero p
o f the function. This generalizes the theorem that a polynomial can be
written as a product of linear factors. Weierstrass s definition (11.12) of
the gamma function is an example. The product is especially simple
w h en /(s) has order at most unity. The order of l/r(s-\-l) is unity,
from (11.16). We shall obtain the product formulae for (s) an(l (s>x)
g iv e n b y
and
= 8(l-8)n-*r(i8)Z(a)
(,*) = ( q M ^ r { U s + a ) ) L ( s , x ),
( 12.2)
(12.3)
\(S,X)\-
(12.4)
First we show that |(s, x) has order one. By eqn (5.3), if a > 0,
L(s, x ) =
f
j
2 x (m) dx.
(12.5)
2.12
H A D A M A R D S PRODUCT FORMULA
61
(12.7)
and we have
(1 -2 1 -) (S) = 2 { - v r - h n , - ,
(12.9)
m=1
valid for a > 0. As in the derivation of (12.7) from (12.5), we deduce
that the right-hand side of (12.9) is
|s| for cr 5;
Now when a ^
w e lm v e
( 1 s ) / ( l 2 i- s ) <
and so
|a |,
(1 2 .1 0 )
log|(ls)(s)| <^log|s| + l,
(12.11)
( 12.12)
uniformly in cr Js
Since (ls) = (s), (12.12) is also valid for cr
i.
We have now shown that (s) and (s, %) have order at most one. I f s
is real and positive then (12.7) and (12.12) are best possible, by Stirlings
formula, so that the order of (s) and (s , x)
exactly unity.
We now discuss the zeros p of the functions (s) and |(s, %). First we
prove a lemma.
Lem m a.
(Jensens formula.)
(12.13)
where N(r) is the number of zeros o f f(s ) inside the circle G(r): |s| = r.
Proof. We can write the left-hand side of (12.13) as
27T R
27T R
0 i-=0
J R e cis 6 drdB
0 r=0
R
27T
27T
= Re f i f W
J
'=o
2niJ f(s)
6=0
(12.14)
>
THE P R IM E -N U M B E R TH EO R EM
52
2.12
(12.17)
T ^ ys^ T + 1 ,
(12.18)
^log(g(|y|-fe)),
(12.19)
< l o g ( m + e).
(12.20)
( 12 .21 )
/( ) = ( - * , x)/(* 0.X).
( 12.22)
By (12.7) and Stirlings formula (11.16) the left-hand side o f eqn (12.13) is
< l o g g+log(|T|+e).
(12.23)
2.13
H A D A M A R D S PRODUCT FORMULA
53
The circle radius 3 and centre s0 clears the box B by a distance at least \,
so that, if N is the number of zeros required, the right-hand side of
eqn (12.13) is
> t flo g 6 /5 .
(12.24)
(12.25)
IT ( ! s/p)exP(//>)>
(12.26)
P# o
where if 0 is a zero we add a factor s at the beginning. P(s) is a regular
function with the same zeros as f(s ) . We should like f(s)jP (s) to be a
constant or some other simple function. Certainly
p (s) =
(12.27)
(12.29)
for each zero p. W e want a lower bound for log P(s) on the circle |s] = B .
Now
log|(l//>)exp(/p)| < B
0<\p\<iB
0<|p\<iR
a)exp(s/p)|
log|(1 s/p)exp(s/
p) | <
(1+logIogU )
B log B loglog B ,
(12.31)
by (12.29). Finally
-
lpl>2B
log'l(1*(p)exp(s//>))| < B 2
lp|>2B
l/>l~2 < R ^ o g B .
(12.32)
(12.33)
T H E P R IM E -N U M B E R T H E O R E M
54
2.12
so that
B,eg(Rcmd) =
and
ra(n)Rn =
(12.34)
(12.35)
2tt
f cos(?i0)Reg{R cis 6) |dd.
(12.36)
Hence
*7/
J |Re<7(i?cis 0) d0
\a(n)\Bn
o
27T
1+
< 1 -f RlogZR.
(12.37)
(12.38)
p
C(x)eB{x)s I I ( ! slp)exp(slp),
(12.39)
p
with the modification mentioned above if p = 0 occurs as a zero. These
i(s, x) =
(12.40)
and
(S>X) _ R/,,\
110" ^
fl^ 41)
where p runs over all zeros of ((s) or L(s, x) that do not coincide with
the gamma-function poles. We can substitute the equation
- l m
= ir +
(12'42)
which follows from the corresponding product formula (11.13) for F(s),
and obtain a sum over all zeros (except p 0) of (s) or L(s, x). It can
be shown that
= I0g 2 + i l o g w - l - i y ,
(12.43)
13
ZEROS
O F f ( s)
I. 131
W e saw in the last chapter that i(s) and g(s, x) have 110 zeros p /3+iy
with /3 > 1. The prime-number theorem corresponds to the fact that
no zeros of (5) (these include the zeros o f (s)) have j8 = 1. All the
direct proofs that j8 < 1 at a zero are based on the following argument.
The function '(s)/(s) has poles of residue 1 at the poles o f (s) and
poles of residue 1 at the zeros. Now
2 log p i l p - * ) - 1
(13.1)
where A(m) is the function o f eqn (1.31). At s = 1, the series diverges
to + 00, corresponding to the pole of residue -(-1. Now, if 1 + iy were
a zero of (s), we should expect the series in (13.1) to diverge to co,
and the partial sums to be as large as those for the case s = 1, but
negative. To achieve these, the numbers to-1? must be predominantly
near 1. The values of to2i>/ are therefore predominantly near + 1 , and
there is a pole at l + 2iy with residue -(-1) aild so a simple pole o f (s)
at s = l + 2iy, which we know does not occur.
To make this argument rigorous, we use (13.1) with s = a + ii where
a > 1, so that the series converges. For all real 9,
Since
(13.3)
(13.4)
we have
853518 X
(13.2)
56
TH E P R IM E -N U M B E R TH EO R EM
2.13
We now make cx+if tend to a zero j8+iy. Since (s) has a pole at 1, there
is a circle centre 1 and some radius r, within which (s) is non-zero.
(Calculation shows that r = 3 has this property.) I f we suppose that
j3 > 1 - f r ,
(13.5)
t'W
m
P,
W l )
V / 1 ,1)
= ~ B + i " ll0 8 2 - i T ( i i + i ) - 2 , 1 ^ + W
(13'6)
we shall assume 1 < a < 2, \t\ ^ fr > 0. Here the sum is over zeros p
o f (s), not over all zeros o f (s), and, since s p and p have positive
real part, we have
/ -,
^
Re _ i _ + > 0
(13.7)
\S~~-P
PI
(13.8)
At s = cr-j-iy we omit all terms in (13.6) except those from the particular
zero with which we are concerned; by (12.7) this gives us the inequality
_ '(cr+ iy)/(ff+ i y) < - {a- p ) - i + 0 ( l o g ( \ y \ + e ) ) .
Similarly,
(13.9)
(13.10)
(13.11)
valid as a -> 1 from the right. By giving a a suitable value, we see that
j3 < 1,
log(|y|+e)
(13. 12)
P < 1
k O T + ^ T
no
ion
(13J3)
2.13
ZE R O S 03?
(s)
57
14
ZEROS
O F |(s, x )
In this chapter and the next we prove results like (13.12) for the zeros
p = /3+iy o f L(s, x), with uniform constants in the upper bounds. We
actually work with (s, %), where % is proper mod#; the zeros of (s, x)
are those zeros of L(s, x) that are not at negative integers and so are
not cancelled by gamma-function poles. We use the product formula
in the differentiated form (12.41),
L ( s ,x )
TT
Z-4\s p
2 i (!( + ))
pj
where the sum is over zeros p of (s, x)> the term 1/p being omitted if
p = 0. We shall assume throughout the chapter that s = a-\-it with
1 < (7 < f.
The first complication is the elimination o f B(x) from eqn (14.1). We
subtract from (14.1) its value at s 2, noting that
CO
2
(14.2)
m=1
which is bounded independently o f x Estimating the gamma-function
term from (11.17), we have
tM
5 5
2 R6( ^ , - ^ ) +0(log(|,1+e))
( 1 4 '3 )
^ ^ - f ) - 1 = 2 ( 2 ^ ) \ 2 ~ p \ -2 < lo g ?,
p
p
l(t) = log{q(\t\+e)},
by (12.19). Writing
we have
Be ^
JjyS} ^ j
(14.4)
(14.5)
(14.6)
2.14
ZE R O S OF
{{s,
59
x )
L ( a Jr it, x )
p,
( 14.7)
L (a -\ -2it,x)
m=2?
p\q
p \Xf
( u -8)
Z jL _ R e V _ i |- 0 (l (t))
(14.9)
Is 1|
for the trivial character Xo mod q.
I f xa is non-trivial, substitution of (14.6) and (14.9) into (14.7) gives
L(s,Xo)
implying that
(14-10)
/? < 1 -cJ H y )
(14.11)
(14.13)
for some positive c2, then by choice of a in (14.12) we can show that
0 < 1- c 8/Z(y),
(14.14)
with a smaller absolute constant c3. We have now shown that either
(14.14) is true or
(14.15)
60
2.14
y ^ 0 we can still deduce an upper bound for f3, but it is very close to
unity, and tends to 1 as y
0.
The third and greatest difficulty is to deal with zeros close to unity
when x2 is trivial. First we show that there is at most one. We have
_ R e ^ 4
L(a, x)
= Re y ^
A*
- )
co
^ 2 A(m )m ~a
i
> {a 1)-X+ 0 (1 ).
(14.16)
L(a,x)
^ (cr-jSi)*+yf ~
p2
+ 0(l0g q)
(14J7)
cJ l(Yi)-
(14.18)
Clearly we can choose c4 ^ c3, and so (14.18) is true for every zero
fti+iyi f
x) except (possibly) p2, and the possible exception p2
occurs only if ^(m) is always real, so that x %is trivial. Since p2 is also
a zero when L(s, x) has real coefficients, if p2 fails to satisfy (14.18) we
conclude that p2 is real. We devote the next chapter to the case of an
exceptional zero ft on the real axis.
15
THE
EXCEPTIONAL
ZERO
I. 57
Ik this chapter we consider real characters, that is, characters for which
x(m) is always real and thus
is trivial. As far as we know, the corre
sponding -L-functions may have real zeros /3 with \ ^ /3
1. Just as
before we saw that L (s ,x ) cannot have two zeros both close to 1, we
shall now see that two functions L(s, x) corresponding to different proper
characters cannot both have zeros close to I. Suppose Xi is proper
m od?!, x% is proper mod<j2, and the corresponding i-functions vanish
at j8j and jS2. In place of (13.2) we use
(1+ Xi(m)) (1+ X2()) > 0,
(15.1)
L'(o, Xi)
L ' ( a , x 2)
L'{a,
Xss) ^
(ct)
L{a,X\)
L (a>X2)
L(a>X 1 X2)
^15 2j
(15-3)
(15.4)
and if
{a
I)-1
O ilogq ^),
(15.5)
(15.6)
THE P R IM E -N U M B E R THEOREM
62
2.15
1ca/log{Q(|y|+e)}
(15.7)
except possibly at a point J81 on the real axis, where L(s, x) has at worst
a simple zero. All ^modr? with q ^ Q for which L (pv X ) 0 are
induced by the same real character.
To prove the prime-number theorem for an arithmetic progression
with common difference q, we need to know that neither (s) nor any
.L-function formed with a character x m o d q has a zero p = fi-j-iy with
P = 1. The proof is simpler if we have p explicitly bounded away from
unity. We have to deal only with the case x real, p real. One method
is to interpret L( 1, x) as the density of ideals in a quadratic number field.
This gives a very weak bound. We shall prove Siegels theorem.
Theorem. For each e > 0 there is a constant c(e) such that if
L Wi> X) = >
ivhere x is a character mod q then
^ < 1c(e)<r.
(15.8)
The constant c(e) in Siegel s theorem is ineffective; that is, the proof does
not enable us to calculate it. All previous constants in upper bounds,
such as c2 in (15.7), have been ones we could calculate, given a table of
values o f (s) for |s|
3 and standard inequalities such as Stirling s
formula.
Following Estermanns account (1948) of Siegels theorem, we con
sider the function
F(s) =
(15-9)
where Xi and Xz are real characters proper m o d ^ and mod <72 respec
tively. By (15.1), the Dirichlet series
CO
log F{s) =
7)11
(15-10)
F(s) =
2
m=1
(15.11)
(15-12)
2.15
THE E X C E P T IO N A L ZERO
63
(15.13)
F(s) = 2 b ( r ) ( 2 - 8y ,
where
b(r) =
(-l)'\ F (2 )/r!
(15.14)
(15.15)
which was proved from eqn (5.3) and Polyas theorem, and is thus true
for any non-trivial x mod?, proper or not. We have
(15.17)
2
r =--R
R+1
2?-}"1
(15.18)
64
T H E P R IM E -N U M B E R T H E O R E M
2.15
and thus (taking only the first term in the series for F(s))
F ( a ) ----- > 1 - A f ( 2 - Cr) - 0 ( g 1g2log3g1g'2(ii)iJ)
a 1
this is
(15.20)
> i -A ( 2 -a ) * /( l-( T )
A ( l a)^ 1e x p { i? ( l cr)}
> i - A ( l - a ) - i ( ?1?a)^-">,
(15-21)
1;
(15.22)
^<
<Aq f ^ \
(15.23)
Let
f sa;-*-1 2
x M d:K-
r = gHogg,
(15.24)
(15.25)
|L(,X)| ^
oo
da; +
J \s\x~a-h' da;.
(15.26)
| i(,x )l< l* | lo g 2 .
(15.27)
(15.28)
the bound
(15.29)
2.15
T H E E X C E P T I O N A L ZE R O
65
< l o g q 1{ l - P 1)L,{a,Xl)
(15-30)
for some a in /3X < a < 1, and so by (15.23) and (15.29) we have
1 < ( l - / W (1-^ lo g 3<Zi < ( 1 - A )qb
(15.31)
16
THE
THEOREM
WE can now prove the prime-number theorem in the form (5.17). Com
bining (5.4) and (5.5), we have
277-1
if u
'0 ( u a{T\\ogu\)-x)
a + iT
-A s
cciT
i + 0 ( a /T )
<
1,
if u = 1,
(16.1)
( 16.2 )
A(m)m-S
m= l
J_
27rt J
a_ iT
Z-,
rn=l
ms s
m<a;
A(m) +
o k
\T
'
no
J -V
log x m\ m a
Z-i
m=1 1 a '
(16.3)
To estimate the series in the error term of (16.3), we first note that
2 A (m )< 2/,
m <y
(16.4)
(16.5)
ix < m < 2 x
2x
(16.6)
2.16
T H E P R IM E -N U M B E R T H E O R E M
67
x lo g 2x\
T (a 1) 1
( 16. 7)
-mm*
\T
-i r
F ig . 2. The contour G
We must now find an approximate value for the integral on the left
of eqn (16.3). By (12.20), there is a value of T in each interval
t < T < t+ 1
for which
^ n __m\~i
\v\ T\ > (log T)-
(16.8)
for each zero p = /3+iy of (s)- With such a value of T we move the
integral to the contour 0 consisting of
C^: the line segment [ai T,
Gz: the line segment [
-in
iT , | + i T \
In eqn (12.40),
^ 1
= 1
THE P R IM E -N U M B E R TH EOREM
68
2,16
1 r'(tts + i))
1 r (f+ ijQ
2 -r d + ii^ )
^ -^'(i('s+ l ) )
/ 1
^
Vs p
.1
2+ i i p )
2a
(a + iip)(2-(-itp)
(16.10)
where we have used (12.20) and (16.8) on the sum over zeros, and
(11.17) on the gamma-function terms.
We can now estimate the integral round C. First
-1 f S ^ d .
27ri
m
a
iog2y
x a do-
x a log2T j (T log x ),
(16.11)
: x ~ -l o g zT
ds
s
o2
_ ilog ST.
(16.12)
l-f(log.'K)-1,
(16.13)
whereupon we have
1
r
277-i , () 5
0
. *log 2T
j-ai^log3! 1.
T log x
(16.14)
The contour has been moved past several poles of the integrand. The
residues of '(s)/(s) are + 1 at zeros of (s), 1 at poles, and so the
integral in (16.3) differs from that in (16.14) by
xP
2
Iy \ < T
'(0)
(16.15)
(0)
where the sum is over zeros p = /3+i y of (s) with )3 > 0. These are all
zeros o f (s), and by inequality (13.12) each satisfies the relation
(16.16)
2.16
THE P R IM E -N U M B E R TH EOREM
69
rsSalogr
< a ;1- slog2T )
where
w lyl<2r
21
8 = cJlogiT-^e).
(16.17)
(16.18)
(16.19)
(16.20)
where the 0(1) is to allow us to find a T for which (16.8) holds, the
error terms in (16.19) are bounded by expressions of the form
* e x p {c^loga:)4}
(16.21)
= a;_|_o|-;Cexp(c2(loga:)*)}.
(16.22)
(16.23)
p<x
A x) =
2 1 = f + 0 { e x p ( - c 4(loga;)i )},
p<x
(16.24)
17
THE
AN ARITHM ETIC
THEOREM
FOR
PROGRESSION
P(x>x) = mX,
A ('m)x(m)<x
L et
C17 1)
(17.3)
IIHi?| X l o g g ) - 1
(17.4)
for each zero p = jS+iy of |(s, x). By (12.19), we can suppose that
A < R < J. The contour C consists of
Gx: the line segment [aiT, f i? 1],
C2: the line segment [
\T, | + i T],
The circle Gi avoids the pole of x*/s and a possible Z-function zero at
s = 0.
As in (16.10), the inequality
(17.5)
\L'(S, x )IL(S, x ) \ ^ \ o g \ T
(17.6)
ft ^
(17.7)
1 c J lo g q T
< c2(log i# ,
(17.8)
2.17
A R IT H M E T IC P R O G R E S S IO N
71
(17.9)
/'(a, x)
f 3^7/?!+ 0 { x exp(c3(log x )i )}
,0 {* e x p (c3(loga:)*)}
if there is an exceptional
zero Ji>
(17.10)
if not.
(17.11)
and
(17.12)
(17.13)
(17.14)
THE P R IM E -N U M B E R TH EO REM
72
2.17
Xi(Pr)>
(17.15)
log ,t\
< log a; lo g ?.
k>g'W
(17.16)
The estimate (17.16) also holds if x is trivial mod#, <jj(x, Xi) being tfi(x).
This error term absorbs easily into that o f (17.14), which thus holds
whenever % is non-trivial mod q. For the trivial character Xo mo(i q>
(17.16) gives
1#^ Xo)
and again the first error term will absorb the second.
Let
? , )=
2
A(m).
m<a?
(17.17)
(17.18)
w = a(mod<2)
There are two cases. I f (a, q) is not unity, only the powers o f primes
that divide q can occur in the sum (17.18). If, however, (a,q) = 1, we
have from eqn (3.3)
if>(x;q,a) =
2
Xiad9
X) =
0 { * e x p ( - c 7(loga:)*)},
(17.19)
PART
III
18
A SURVEY
OF SIE VE S
to into
(i) Apart from the trivial function a for which a(m) is always 1, the
sum (or integral as appropriate) of ct(to) over a complete set of classes
is zero.
(ii) I f to and n are in different classes there is at least one function a
for which a (to) ^ a(n).
Then a sequence m i of integers is uniformly distributed among the
classes if and only if
| t% ) = o ( X |a(TOf)|)
(18.2)
for each a except the trivial one. This is known as W eyls criterion for
uniform distribution. As an example, the prime numbers form a sequence
that is uniformly distributed among those residue classes a mod q for
which (a,q) = 1, in the sense that a proportion (l/<p(<z)+o(l)) of the
primes up to some bound x falls into each reduced residue class. We
74
T H E N E C E S S A R Y TOOLS
3.18
deduced this from the fact that j/r(x, %) is of smaller order than x when
X is a non-trivial character mod?.
Franels theorem o f Chapter 9 fits this scheme. We were concerned
there with the distribution of the F Farey fractions in the interval [0,1].
The appropriate functions a(m) are e(bfm), where b is an integer, and a
calculation with Ramanujan sums shows that
F
(18.3)
e(bfm) = 2 d M ( Q l d ) ,
d\b
$ ( ) =
m=M + l
(18.5)
u(m)e(mtx);
(18.6)
3.18
A S U R V E Y OF S I E V E S
75
b moclg'
(18.7)
m.M + 1
m = 6(modg)
which follows from eqn (3.3). The stun on the left of (18.6) can be bounded
by the large sieve for exponential sums (7.8), and we have
M+iV
I \8X\ < r W s H x + o i r 1)) I
M)l2(18-8)
X mod q
m M + 1
and we have
I*
am odg
(18.9)
x ( a ) e a( a m ) ,
M +N
r(x)>s x =
2
2 * X ( a ) u { m ) e q{ a m )
m=M + 1 amodg
(18.10)
* qn * >
-t-t-i/-/!
''-L'
a2
mod
By (3.3) again we have
^modg
x{a)S
X mod a
< ? (?) 2 ;l
(18.11)
am odg
jl y
ap(q) Z-*
g< 0
am odg
M +N
< {N + o m )
m=M+l
k * )is
(18.12)
2 2
a=SQ x mocla
p(<z)
< { N + 0 ( Q 2) ) X |m(to)|2
1
< (A ^+0(Q 2))( tx(iY) + 0(Q )).
(18.13)
76
3.18
T H E N E C E S S A R Y TOOLS
f^Q
2
xmocl/
(18.14)
j s O fm o d /)
<p(<7)
s=s Q
(18.15)
Z . + i S < xr < Z a- J S
for r = 1,..., R.
(18.10)
Then
it
Z
I l/( r )la < g J
r=1
!/(*,) I2 = \f(x)\2~
Ax
(18.18)
j/(a;)|2 da;
S|/(a;,)|2 <
9,v+*8 V
\f(x)\2 da; -f-
!'f- id
Xr-iS
j - ^ \ f ( x ) \ 2 dx
dy
X,
x r+ tS
\f(x)\2 da
da; -f-
Xr
da;
da;.
(18.19)
i r i S
2 I /M 2<
(18.20)
3.18
A S U R V E Y OF S I E V E S
77
s la
< (Q2+
Q amodg 1
ttN)
K()!2-
m =M +l
(18-21)
2'i
Tt
T2
V i ^ l L 2 f dt + V
Z-t
m 2a
rp
Z-t
m ana
J \mj
11
rp
d*.
(18-23)
22
m
2 2
u ( m) u ( n ) (n j m ) iT2 (,/w)iTi
ma?ia
n
m^n
ilo &(nlm)
|2
\(n m)lm\ \2
?w=l ]nK)K2m x
, V / V
1 V
m2a
\
+ mZ=1 n>2m
Z + ?KZ?n. 7
1 1-1/7<T)\12'
2
n2a
1 A l()l2 , 1 M l 2'!
no t>/n
(18'24)
'
where we have used the geometric-mean inequality. The coefficient of
\u{m)\hn~iCL in (18.24) is 0 { N \ o g N ) , and thus we can write the righthand side of (18.23) as
(T2 T y -}-0 (N lo g N )) 2 \u{m) |2m_2 .
m=i
(18.25)
Since
f'(t) =
2 iiog m u (m )m ~ a~il,
(18.26)
and log to ^ logiV, the upper bound for the integral o f \f'{t)\2 is at
78
T H E N E C E S S A R Y TOOLS
3.18
most log2lY times that for (18.23). We have now proved that if
tR
satisfy the relations
,
,
2
/,., /, >
(18.27)
for r = 1,..., R 1 and
T y + l 8 < tr < T2- 18
(18.28)
(18.29)
19
THE
HYBRID
SIEVE
Kanga saicl to Roo, Drink up your milk first, dear, and talk
afterwards. So Roo, who was drinking his milk, tried to say
that he could do both at once . . . and had to be patted on the
back and dried for quite a long time afterwards.
I. 151
f(t)=
\ f(x)e(xt) dx.
(19.1)
CO
J / W e ( - ^ ) di.
/( ) =
(19.2)
CO
00
CO
oo
co
co
co
co
CO
CO
CO
=j
co
l/(*)lada,
(19,3)
80
T H E N E C E S S A R Y TOO LS
3.19
Plancherels identity. We shall use eqns (19.1) and (19.2) w ith /a linear
combination o f the values of
* H r &!>$
Here
F{t) = 2 ^ ,
(19.5)
S(t) = 2 o(v)e(vt),
Let
(19.6)
o{v).
(19.7)
oo
Then
(19.8)
co
(19.9)
and
(19.10)
By (19.3),
co
co
co
co
*27
(19.11)
and
\F(t)\^2/TT
(19.12)
for |f|
T , which gives (19.8).
We apply the lemma to a Dirichlet polynomial, that is, a finite Dirichlet
series
N
$ (s>x) = X, a(m)x(m)m~a~it>
(19.13)
m=l
where
Ae*
2 a(m)x(m)>
A-ie*
A = exp((2T )-1).
(19.14)
(19.15)
3.19
T H E H Y B R I D S IE V E
81
.2
XinodQ a~1gx
()x()
< (( A -A -1)e *+ ? ) f
\a{m)\
A_1ez
< ( 2 1- 1ea:+ ? ) X
|(m)|2.
(19.16)
A ~'e*
i (U>
xyl
* - ( l o f > ' xY)-1
--
11
1*
1 + (lo g A r)_1
i(log A ) - 1
F ig . 4. T h e contour C
By the lemma,
T
f
J T
logfiiA
xm oda
m=l
( T - V + g ) dx
log(m./A)
X {m-\-qT)\a(m)\2m ~Za.
=1
(19.17)
T s ^ t s ^ T ,
(19.18)
(19.19)
82
T H E N E C E S S A R Y TOOLS
3.19
and hence
du,
(19.20)
(19.21)
uniformly in cr. We sum over r and apply (18.17) under the integral sign.
Next we sum over x mod?. One of the terms on the right of (18.17) is
a geometric mean, but Cauchys inequality allows us to sum over X hi
each factor before taking the geometric mean. We have now to estimate
two sums over x of integrals, which are of the form (19.17). In each,
a and T have been replaced by A and f + f S , where %i = A+ir, A and r
real; in the first a(m) has been replaced by a(m)m~iT, and in the second
by ia(m)logmro~iT. It is simpler to describe this inequality in words
than to write it out! After applying (19.17) we have
R(x)
vmodtf i==l
m 1
(19.22)
J
a -(lo g iV )-1
m - 2 '' dA
(19.23)
Hence
x 2 "
(19.24)
where for each X the points sr s(r,x) satisfy (19.18) and (19.19).
We can use (18.12) in place of (18.8) and show that
<19'26
and
R(X)
2
T'
2* 2
Xmocla r=l
-'* * '* )1
N
^ J ^ l !.
m= 1
&v
7
where the points $(?*,%) satisfy (19.18) and (19.19) for each
(19.26)
3.19
T H E H Y B R I D S IE V E
83
The point of these hybrid sieve results is that we have N-\-Q 2T (in
fact, m-\-Q2T) in (19.25) etc. in place o f (N-\-Q2) T or (N -\ -T ) Q 2 which
we would have from sieving one parameter and summing the other.
A useful mnemonic for large-sieve results is that N 2 \a(m)\2m ~2a corre
sponds to the square o f the trivial term t = 0, % trivial, and that there
are 0 ( Q 2T) non-trivial terms, whose mean square is 2 |(to)|2to 2ct.
We have certainly shown that eqn (18.2) holds for almost all pairs s,
20
AN A P P R O X I M A T E F U N C T I O N A L
E Q U A T I O N (I)
In a corner of the room, the tablecloth began to wriggle.
Then it wrapped itself into a ball and rolled across the room.
Then it jumped up and down once or twice, and put out two
ears. It rolled across the room again, and unwound itself.
II. 135
(20.1)
(20.3)
(ferM/TT-)2*-1.
(20.4)
We use the ideas of Chapter 5 to get a partial sum for the Dirichlet
series of L 2(s, x). In the fundamental equation (5.4) the convergence of
3.20
A N A P P R O X I M A T E F U N C T I O N A L E Q U A T I O N (I)
85
and
(20.5)
K(w) = Z ^ w J + ^ - o * ) = ^ c o s h w f l (<(r
-l
( 2 0 . 6)
1 e
'r '
u>
2 a)
/ w 3 2 ( 2 ? ) ! ( 1 - ) - ? ') ! ai (?'
(20.7)
2rri
'tt i
m w
co
\x j
2 ico
where
c(v) = 0
if v > e
if e -1 < v ^ e
J. (20.9)
L%s+aj,x)daj.
(20.10)
(20. 12)
T H E N E C E S S A R Y TOOLS
86
3.20
ox
|+ico
1
2771
(20.13)
s + tt 1
f i c o
a\21,-1
'
we have
*\2\A1
G(%)xl- Uy x~n
47r2x y
?ki
provided that
j|-'.
I f we choose y by
4772x y = q2t2
(20.15)
(20.16)
(20.17)
d(m)x(m)m-u N
d (m )x{m )m -,
d(m)x (m )m -v
m<ev
m <e~ iy
(Pz(u )
(20.18)
m >ey
? i(u )
d{m )x{m)m-u
3.20
A N A P P R O X I M A T E F U N C T I O N A L E Q U A T I O N (I)
87
l-.s
l<Pa( )l <
m >oy
d{m )m ~x
2 m ~ \ D (m ) D { m 1)}
m>ey
< {ey^ log y,
(20.19)
by partial summation and the estimate (2.13) for the sum function D(m)
of the divisor function. By (20.19) and (20.4),
T .'B1- s~X 1(s+'M 1) j g^2" - 1 (
G(u)<ps(u) du,
2m J
s-\-u 1
1
<J
Ci
,.1 -a -A
Jxyir y ~ *
< ?-
y^logy dr
{l+\t+r\f\2rr j
< x - at~2
853518 X
/^|t |\2A-1
4qH2)
_1iog y-
q\t\logy
( 20 . 20 )
88
T H E N E C E S S A R Y TOOLS
3.20
The same estimate holds for the integral along Cs. On the semicircle C2,
(20.4) and (20.15) are valid, and we have the upper bound
< x - a(l+ \ t\ i) - 5q\t\logy(ir\t\i)
<^.qx~a\t\-1\ogy.
(20.21)
21
AN A P P R O X I M A T E F U N C T I O N A L
E Q U A T I O N (II)
Wliy, whats happened to your tail ? he said in surprise.
Wliat lias happened to it ? said Eeyore. It isnt there!
I. 43
ioo, J it i||*],
i, Ji],
The contour D for t < 10 is the reflection in the real axis of that
described above. The indentation about the origin avoids a possible
double pole of G(u) at u = 0. The analogue of (20.19) is that, uniformly
l<Pi( )l < ( e y ^ l o g y .
1
(21 . 1)
C x 1~s~uK 1(s-}-u 1)
!7ii J
D
(21.3)
90
T H E N E C E S S A R Y TOOLS
3.21
w =
d(m)x(m) ,lm
c
m 1s
(21.4)
where
c'(v) = 0
if v ^ e
+2
1
9
32(1+J*)!(2?)! \n2vxyi
X ( ? ( l 5 + ( r l)7ii)
(21.5)
if e_1 ^ v < e
12s
c'{v)
G ( l s)
If
is
then Q{u ) is regular at 0. For \t\ ^ 10 we take the
contour D to be the line A = f . The bound for Jt and / 2 on this contour
3.21
A N A P P R O X I M A T E F U N C T I O N A L E Q U A T I O N (II)
91
is the same, but without the factor ||-1 and with a different implied
constant. I f x( 1) is + 1> however, there is a double pole. I f a is
bounded away from 1 we can run the contour D between the origin
and the poles of /^ (s+ m l)/(s-\-u 1). I f a < 1 we take D as
D x\ the line segment (
ioo,
i],
[1 i, 1(1 cr)i],
[1(1 cr)i , l ( l cr)+i],
[1(1 cr)+i, 1 + i],
[ 1 + i, 1+ioo).
qx~(l a)~zlo g y.
I f all else fails, we can take I) as the vertical line X = \ and get an
explicit but very complicated extra term in the approximate functional
equation from the residue at n 0 of the multiple pole. When || < 10,
(20.14) is interpreted as
q 2 < ^ x y < g 2.
(21.7)
We have now shown that an equation of the form
7jV jx) _ 2
- M
A ) + 2
,(5 )+
H
( 2 1 .8 )
( 2 1 .9 )
/ 4 is
(1 o ) - 2qx~ulo g y,
(21.10)
[to^ K iiv))
(21.11)
(21.12)
92
T H E N E C E S S A R Y TOOLS
3.21
of (s) about the pole at s = 1. We must add to eqn (21.8) the additional
terms (21.11); they are
^
(21.13)
uniformly i n | / | ^ 1 0 , j ^ c 7 ^ f .
There are many approximate functional equations in the literature.
The one we have proved is easy to sieve, but has more complicated
coefficients. Approximate functional equations for (s) and for 2(s) are
given by Titchmarsh (1951, Chapter 4), and generalized by Chandra
sekhar an and Narasimhan (1963). A. F. Lavrik has a number of papers
on the subject in the Doklady and Izvestia of the USSR Academy of
Sciences. Fogels (1969) has a form rather like that of Montgomerys
given here.
F O U R T H P O W E R S OF L - F U N C T I O N S
They were out of the snow now, but it was very cold, and to
keep themselves warm they sang Poohs song right through
six times, Piglet doing the tiddley-poms and Pooh doing the
rest of it, and both of them thumping on top of the gate with
pieces of stick at the proper places.
II. 7
( 2 2 .2 )
(22.3)
3.22
T H E N E C E S S A R Y TOOLS
94
and we have a similar result for (s), with 7 instead of 6 and an extra
term 0 ( |/|~10). The integers x and y in (22.4) are connected by (20.14):
y = [qH*l( 4 A ) ] .
(22.5)
When we fix x and average over x and t, y is varying. For a good upper
bound, x and y must be of the same order of magnitude. We restrict
ourselves for the moment to P < q < 2P and U < \t\ < 2 U , and
average the right-hand side of (22.4) over all integer values of a; between
\ P U j n and \ P U jit . For each fixed x and t ^he corresponding values
of y given bjr (22.5) are distinct and lie between ^PUj-rr 1 and 1 6 P U / tt.
The average of the square of the terms involving y taken over all integers
y in this range is at least
times the corresponding average over the
values o f y that actualfy occur in the sum. This device allows us to sum
a Dirichlet series of fixed length ey over varying x and t, provided that
x does not occur in the coefficients (and vice versa).
For each value of x in the above range, (22.2) gives
R(X)
<p(q)
d ( m )x (m )
hn'
TO^-X)
cx
r 1
(S_1+loga;) ^
(m +P ^U )
m1
cZ2(to) log x
mlog (m +e)
(22.6)
8> 1
(22 7)
I-
/ .2/^A Y n i
(i)
(22'8)
3.22
F O U R T H P O W E R S O F L -F U N C T I O N S
95
''>
Ti
71
'' '' D
'
(22.9)
and similarly for integrals along the contour G. The first factor here is
We shall find an estimate for the average of Iv Here,
F(u) = x1- s- u(sJr u l ) - lK 1(sJr u \)(qlir)'in- 1G(u,x)qii{w)( 2 2 . 10 )
We wish to average this over % with the length of the sum cp^w) fixed,
that is, with y fixed. By (22.4), x is approximately varying as q2, and
is bounded on the contour D by a function o f t and y alone, by
(20.14). Similarly, (20.4) permits us to take out (q'/tt)2 1 G?(^<-) at its
maximum. B y (18.12),
xmods
msgey
< ( y + P 2)(e/)i-2Mog!ty,
(22.11)
(22.12)
which can be deduced from (2.24). The sum of the integrals involving
F(u) on the right-hand side of (22.9) is now
yy,\ 22 u/^y\4u 2
>
(max&)1_2l7|i|~*(?/-j-P2)log3?/
< (P 2+P|i|)|i|-Jlog3(P(|<|+e)),
(22.13)
(22.14)
and similarly the same upper bound holds for the corresponding sum
involving / 2.
96
T H E N E C E S S A R Y TO OLS
3.22
(22.15)
Gallaghers lemma (18.17) gives
277
i
it
kj+
T"2
,/ ^
2d'~
V/~a
+i
x\ ,1 /, 2U + i
J i/ii2^+ J i-^rdiW J
U < t r^ W
U~i
'u - t
'
'77-i
i j
dt 1
. jl
\t
dtj
(22.16)
We sum (22.16) over % (using Cauchys inequality on the second term)
to find
,
_ mx)
2
Jo ) Z
2
IAI8 < ( - P + ^ ^ - 8log4(P (?7+ e)) (22.17)
P<g<2P
x m od2 r = 1
when we substitute in (22.14) and (22.15). The same upper bound holds
for the sum with Ia.
The integrals I 3 and / 4 present a further complication in that <p3 and
cp4 are not finite sums. To treat 93 we write
__
2
19
d (m )x (m )m ~ u\
>i>ey
/ J"?
\/ ^
( 2 ?l_2)( 2 ,l2
V=1
' 'n=l
,_i
2
leM
y<m<eM
+1j/
\
d (m )x (m )m ~ u 2)
'
(22.
By (18.11) we have
g
cp(<7)
d(m)x (m)
'V *
. ,
in
(e +1 /+ P 2)(e,li/) 1_2;'(ii+log?/)3,
(22.19)
and the sum of terms of the form (22.19) from n = 1 to infinity converges
uniformly in A to
( 22.20)
When we use (22.20) in place of (22 . 11) we show that the sums over I 3
and J4 also satisfy the bound (22.17).
3.22
F O U R T H P O W E R S OF ^ -F U N C T I O N S
97
(22.21)
1 ,...,
R(x) we have
(22.23)
and for r = 1,..., R ( x ) ~ 1 we have
t(r + l,x )-t{r ,x ) >
1,
(22.24)
(22.25)
and still have the upper bound (22.22), so that the root mean fourth
power of L(s, %) is
log Q T , Without the sieving over T this method
proves that the root mean fourth power at afixedsis <^log(Q(|s|+e)).
A simpler proof of a fourth-power moment is in Gallagher (1967).
Another result is given by Linnik (1964, section 41). These relate to
summing over x- The situation is much easier if we merely integrate
over f; there are examples in Titchmarsh (1951, Chapter 7). The hybrid
moment we have proved here is sketched in Montgomerys paper (19696),
but with a larger logarithm power.
P A R T IV
23
INGHAMS THEOREM
He tried Counting Slieep, which is sometimes a good way of
getting to sleep, and, as tliat was no good, he tried counting
Heffalumps.
I. 62
< 0 < 1,
T < y < T
(23.1)
(23.2)
m<:X
(23.3)
a(m)x{'in)m~s,
(23.4)
m >X
a(m) =
2 M<Z).
(23-5)
d^X
The series (23.4) converges without any hypothesis for a > 1. Until
A. I. Vinogradovs work (1965), the number of zeros was customarily
estimated by integration o f the logarithm of 1+/(> x) round the
boundary of the rectangle (23.1). As we shall see below, f(s , x ) is less
than unity in root mean square, whether averaged over t or over x r
4.23
IN G H A M S THEOREM
99
I f
2t t i
JY\W
r ( w ) ^ j W dIV = e - m'r
(23.6)
2 ico
2 + ico
277i
L i p + t v , x ) M ( p Jr w, x ) Y wr (w ) div.
(23.7)
2 ico
1
27ri
(23.8)
(23.9)
I log Q T
lo g !7
(23.10)
10Z.
(23.11)
The terms on the left of (23.7) with to > 100IY contribute less than ,
if I is sufficiently large. We recall Stirlings formula in the form
(20.3), valid if A ^ |t |j :
|r(A+ir)| = e-^W|A+ir|;-i{(277^ + 0(|T|^)}.
(23.12)
The term (23.9) is now seen to be at most ^ when |y| ^ 100Z, again
provided I is sufficiently large. Apart from the zeros o f (s) with
|y|^ 100Z, all zeros fall into one or both o f the following classes.
Class (i). Zeros p with
y
a(m)x('>n)m-Pe~m!Y > .
(23.13)
100
4.23
(23.14)
i-fl-ic o
(23.15)
nisi?
JL
V * V
<1<Q
xmoda
Z-,
p
V
/-i
m elr
a(m)x(m ) e -,ir 2
mP
^ Z 2 ( +Q 2y)^2()~2a:exp( 2')
melr
(23.16)
X{<922,(2,T ) 1- 2 + (2 ,T ) 2- 2}
(23.17)
(23.18)
There are several ways of treating zeros of class (ii). The ingenious
work of Montgomery (19696) makes much use of this flexibility. To
4.23
IN G H A M S THEOREM
101
V JL V* V
(pftf) jL-4 Z-4
q ^ Q T SJ-' xmod q
) 1 CO
i +ico
2 ^ 2 * 2
f
a<Q9 U ;xmoda P s-ico
4+1C
2 ^
' e=so
2 *
\2
xmas p
| d| r x
(23.19)
<
(23.20)
When we apply the hybrid sieve (19.26), the second integral on the right
of (23.19) becomes
x ( m + Q 2T)
lo g Z
m
log(m +e)
m=1
<^i(a- i ) - n x + Q m )
(23.21)
<
<
and thus
^
a<Q
We now choose
2 *
xmoda
^ {ot l ) - * Q iTY*a-*<Mii.
(23.23)
(23.24)
(23.25)
102
Z E R O S A N D P R IM E N U M B E R S
4.23
a > $ + fZ -ilo g Z.
(23.26)
N (x) < T l ,
so that
^
a^Q
T
iV(/Y) ^ Q2Tl
xvaoaq
(23.27)
(23.28)
2m2
(23.29)
Ingham proved his theorem for the zeta function only, with no sum
over
his result (1940) was
N < ysa-oo/ta-oO^
(23.30)
where N refers to the zeros of the zeta function satisfying (23.1). (23.25) is
stronger than (23.30) for a near but weaker for a > f . We have lost
a logarithm factor by taking representative zeros instead of counting
according to multiplicity.
24
BOMBIERFS THEOREM
I see, I see, said Pooh, nodding his head. Talking about
large somethings, he went on dreamily, I generally have a
small something about now about this time in the morning.
I. 48
(log x)N,
(24.1)
. am odg y ^ x
Hy> <?>)-
q<Q
9 (g)
^ (log)^
Q = x*Q.ogx)-B
where
(24'2)
(24.3)
q ^ Q TV- "
863518 X
xm odg
X non-trivial
"
</<Q
H
, 2 4 -4 )
ZEROS A N D P R I M E N U M B E R S
104
4.24
1 / Q
x m o c l/
V<X
s s O t m o d /)
a<Q
(24.5)
A = log*.
(24.6)
v^o;
<?(q)
ffl(ff)
Z -!
a<Q
ro(f) Z -, m(rn)
Tw '
'
cp(/)
(24.7)
rw '
Relations (16.22) and (24,3) assure us that the second and third terms
on the right of (24.5) are
^ x^_A
^24 ^
for x sufficiently large. Since / -<C XiB is of the form (24.1), the SiegelWalfisch inequality (17.14) implies that (24.8) is an upper bound for the
terms in (24.5) w it h / < A43; we are assuming x is sufficiently large.
For q > A4B we combine (17.2) and (17.6) into-
4>{y, x)
= ~ 2 ^+
i+ ^l)log2? Ty\
(24.9)
where the sum is over zeros o f (s, %) with |p| > R and |y| < T , R and
T being chosen for each % so that (17.3) and (17.4) hold; in particular
the O-constant in (24.9) is independent of
T, or y and uniform in
^ R
J. When we choose T to be a little smaller than x we have
Ip{y>x)\<
Y^+O{x^logzx)
(24.10)
4.24
B O M BIB RI S THEOREM
105
(24.12)
_A_ <
\P \ <?(q) ^
JL
P U <p(q)
(24.13)
K
/ 3 ( l - a ) / ( 2 - a ) 1
^ ^
(24.14)
when a > |- and the sum is over powers of two not exceeding x.
In the sum over P we distinguish two cases. The exponent o f P in
(24.12) is greater than unity for a < f and less than unity for a. > f.
The values of P are the powers of two between JA4B and Q; they are
0(A) in number. Hence for a ^ f we have
2 p8(l-o0/(2-a)-l
^^4B)6(l-a)/(2-a)-l
^ ^ l+ 4 B(0(l- a )- l)
(24.15)
AQ6a-)/(2-a)-l
P
a. 3 ( l - a ) / ( 2 - a ) - ^ l - J ?
(24.16)
where we have substituted Q = a^A~B from (24.3). The terms from zeros
o f L-functions formed with characters whose conductors exceed A43 are
now seen to be
^ aA10--5
(24.17)
(24.18)
We have thus proved (24.2) with B given by (24.18), which clearly can
be improved slightly, since (24.15) and (24.16) can be improved.
There are two ways o f proving Bombieri s theorem. The shorter, due
to Gallagher (1968), is to perform the sieving directly on L'(s, x)jL(s, x)
106
4.24
>
25
I. M. V I N O G R A D O V S E S T I M A T E
I f we are to capture Baby Roo, we must get a Long Start,
because Kanga runs faster than any of Us, even Me.
I. 93
T h e sum
8{oc) = S(y, a) =
X logpe(pa)
(25.1)
(25.3)
2 A(m)e(ma) + 0 ( y n o g y ) ,
(25.5)
the error term arising from squares and higher powers of primes. When
ajq is in its lowest terms,
the error term arising from powers of the O(logg) primes that divide q;
these are not congruent to reduced residues b mod#. The sum of the
terms involving b in (25.6) is x(a)T(x)> and by ecln (3.20) the modulus
108
4.25
1 /5 1
lyl<
whenever y + x, the O-constant being absolute. The stun is over zeros
o f (s, Xj), where Xi proper mod/ induces x modg. Since every zero of
i ( s >Xi) i,s a zero
x)> we shall take the sum in (25.7) to be over
zeros of (s, x). From (25.5), (25.6), and (25.7), we have
<*)
(l ~
p \
<j
(25.9)
\
'
p/ f
2*
R(x)
(25.10)
Xmod/ r=l
and hence
2
2
xmoda >=!
m + it(r ,
x))l4 <
T lo g tq T
Y <p(/) T T 1 + -*
II \
p\
PXf
T l o g 5q T J ! (p l + l+^J- *)
p\a
qT\og5qT,
(25.11)
A = log qT.
(25.12)
(25.13)
4.25
I. M. V I N O G R A D O V S E S T I M A T E
109
(25.18)
(25.19)
26
I. M. V I N O G R A D O V S T H R E E - P R I M E S
THEOREM
W i t h S ( y } a) given
x
f S 3(x, ot)e(-xot) da = 2 2 2 lo g ^ jlo g ^ lo g ^ ,
(j
(26.1)
Pi P 2 Pa
(26.2)
Q = [xl~2S],
(26.3)
where I = log a:. Since S(x,u) is periodic, we can take the range of
integration in (26.1) to be the unit interval [1/(Q +1), (Q-t-2)/(Q-|-l)].
We divide this interval into arcs
- i
Ir
A r + ir -l
a,+a,.+1
(26.4)
5V+2V + 1.
(26.5)
The intervals I r are known as Farey arcs (if we think o f e(a) as a com
plex variable, the integration is round the unit circle). We divide the Ir
into major arcs, on which we work out the integral over the spike at
arlqr explicitly, as promised in Chapter 6, and minor arcs, on which we
4.20
I. M. V I N O G R A D O V S T H R E E - P R I M E S T H E O R E M
111
use an upper estimate for the integrand. We call Ir a minor are if qr > Z26,
when (25.2) with B 26 gives
\S(x, ar/qr+ P ) | < (1 +Z-aeQ -1)a:Z-8 < x
(26.6)
log
?(?
p itv ,,
3?= &(modg)
(26.7)
xl~
< 1+
y
log -
., 8
?(?)
35=vDk
(mvodg)
2 *
bm odq
(26.9)
cp(q)xl~32.
<
The simplest sum with a spike is
F(a) =
(26.10)
^ ( 4
m^x
<g; (l+|/3|)max
s^
y^x
) m
l2Sq - 1cp(q)xl-3i
(26.11)
xh*.
We now have
/A\?) F ( m ,
S ( a l q + P ) - V M F 3(j3) < x l ^ [ \ S ( a l q + P ) \ * + ^
? 3(q)
? a(?)
(26.12)
and the integral in eqn (26.1) is
,
2
(Q+2)/(Q+D
Ir majore Ir
|*%)|2da) +
I K Q + l)
+ {
\ I r m ajor
I,
'( -? J
dot .
(26.13)
2J^x
(26.14)
112
4.26
both
<a:Z~3.
(26.16)
1 ^
||a||
we have
*
eqr( - a rx)
J JP8(i8)c(-j8)dj8-Ji,8|a-|:Je(-a[)da
%x{x l),
< >
Q^V20 amodg
U1 am odg
s(q)
/-<
22
g
d\x
d\x
d\q
<f(q)
dix{qld)iJ?{q)
'
* *'
g^O(modfZ) r
\ n l-i
n + ( p - wny
(p - 1)3
( p - i ) 3 ( ^ - i ) 3+ i
4.26
I. M. V I N O G R A D O V S T H R E E - P R I M E S T H E O R E M
113
(26.22)
(26.23)
(26.24)
ail expression which again is seen to be non-zero for x odd and sufficiently
large. The formula (26.24) is the famous theorem of I. M. Vinogradov.
27
H A L A S Z S METHOD
And I know it seems easy, said Piglet to himself, but it
isnt everyone who could do it.
II. 18
W e have been concerned with estimating how often various sums S are
large. The large-sieve results that we have had gave an upper bound
for the sum o f |$|2 over different values of a parameter. The upper
bounds we obtained contained two terms; the first was the maximum
o f \S\2 and the second the mean square o f |S| multiplied by the number
of values of the parameter. Thus if
N
S(t) =
2
m=1
(27.1)
r= l
(27-2)
where the points tr are at least (log N )^ 1 apart and lie between
and 2\,
The second term in (27.2) corresponds to the mean value o f |$()'|2 and
the first to its maximum. Halasz s method sometimes enables us to
count the number o f values o f r for which \S(tr) \ is large, without the
second terms being present on the right of (27.2).
Halaszs method is based on the lemma of Chapter 7,
f c,(u,fW) < ||u||( % |c,|2W max f |(fW,f<fl>)|a)*.
(27.3)
4.27
H A L A S Z S M ETH OD
115
so that the second factor on the right of (27.3) is R*. I f each term on
the left o f (27.3) is at least V we can square (27.3) to give
i?2T/2 < P||u||2 max f |(fW,ffe>)|
< -B||u||2(max||f(,')||2+ ( P l)max|(fW,ffe))|],
\ r
}
r,g
(27.4)
r^q
u,
for r = 1, 2,,.., R .
(27.5)
(27.6)
F2 ^ 2||u||2max|(fw,f (9))|.
(27.7)
Then
provided that
<l=r
<YYi\
um = emlNa(m)
um = 0
iiff 1
1
<T" N'
AH
^ rm.
m^
if m > N
fm = e - mlNm - ar<--iir
Here,
(27.8)
m^N
||u||2 < e2 f
(27.9)
for all m ^ 1.
(27.10)
|a(m)|2
(27.11)
CO
and
(27.12)
0
2//iV^crit
m 1
2 + ico
f
2771
J
2 ico
r(w)(%N)w(w-{-<j-\-it) dw
(27.13)
116
ZE RO S A N D PRIME N U M B E R S
4.27
A residue
(27.14)
r ( l cr~ i t ) ( ^ N ) 1~'J- it
valid when A
(27.15)
|r|-. Hence if
\t\ > logA7
(27.16)
m<0(|r|)
d { m ) m x~1+
-H K A + iT + ^ i)}
+ max
r
{i(l-A
- i- r - f 77i)}
-uv'
M<2
m^o('|r|)
< M Alog2M,
(27.17)
where we have used (20.4) and (2.14), and partial summation. The
functional equation and the estimate (20.4) now give
|(A+ir)|2 < M ^ lo g ^ r l,
(27.18)
(27.19)
uniformly i n O ^ A ^ l , |t | ^ 10.
We have now shown that
|(f,ffa>)|
< 1 + J | r^ )| | ?(ty + A + ir)| | P 7|,u |dw|
C
CO
dr
< |i|*logaJV||,
(27.20)
4.27
HALASZS
m ethod
117
where we have used (27.15) for the gamma function and (27.19) for the
zeta function. We can now restate the condition (27.8) as
T < T0)
(27.21)
|a(m)|a2 * lo g O T .
(27.22)
Clearly when T > Ta we must divide up the range for T into intervals
of length at most Tg. Repeated application o f the inequality (27.7)
gives us
IT
\
R < (m +A y-*
\a{m)\m.
v*0
/
m= 1
When we substitute for T0 we have the result which follows:
T h e o r e m . Let
If
(27.23)
< 3 = 2 \a(m)\2.
m=1
(27.24)
(27.25)
^ F
m 1
for s =
\ K - t a\ > lo g iV
(27.26)
for q ^ r, then
R <^ G N V -*+G P N T V -*lo ffN T ,
(27.27)
The form of the second term in (27.27) arises from our choice o f
functions f^ ; it is larger than the first term unless (27.22) holds with
T in place o f T0. A plausible conjecture is that
R < G N V -2
whenever
F2
GTS
(27.28)
(27.29)
28
GAPS B E T W E E N P R I M E N U M BERS
I shall do i t , said Pooh, after waiting a little longer, by
means of a trap. And it must be a Cunning Trap, so you
will have to help me, Piglet.
I. 56
(28.1)
but the representatives are in number ;> l ~2times the zeros in that class.
We suppose a > f , since the result (28.19) which we obtain below
improves on Inghams theorem only for a > f . The parameters X and
Y will satisfy
x ^ y2;
im Y ^ T 2
(28.2)
(28.3)
ito
the parts of the integrand with |Imt| > 100? give less than ^ (if I is
sufficiently large). The integral of |.T(|-+ii)| converges rapidly so, for
(28.3) to hold, there must be some t with \t y\ sj 100? for which
m + i t ) M ( i + i t ) \ > c Y P -i,
(28.4)
(28.5)
(28.6)
4.28
GAPS B E T W E E N P R I M E N U M B E R S
119
Otherwise we have
|i(i+ii)| > V = c U - lY - ! ,
(28.7)
We choose
jj
< X V - H + X T V - 9? .
(28.8)
__ ^Y-i/ioys(2a-i)/io;
(28.9)
V = eX1'10^ 2*-1)'5,
(28.10)
and on adding (28.6) and (28.8) and multiplying by I2we see that class (ii)
zeros number
^ jf2/5y-e(2a~i)/5^9^_jy4/5y-2(2a-i)/5^3;
(28.11)
the second term in (28.11) being less than the first provided
Z 2r 2- < TH30.
(28.12)
(28.13)
(28.14)
(28.16)
y 5 o;-3)|{aa+a:-l))
(28.17)
(28.18)
Then whenever a: is
(28.20)
120
4.28
Such a result was first proved by Hoheisel (1930) with c a little less
than one. Ingham (1937) obtained the result with c > f and indicated
how to replace f with a smaller number by improving an upper bound
for |(--f-i?)|. Several authors achieved this by means of intricate
arguments. Recently Montgomery (19696) obtained the result for c >
by the method given here, but with a less efficient use o f the Halasz
lemma; the improvement to
was seen by the author in preparing
the present exposition. As we have seen, Montgomery s method rests
on the Halasz lemma, and thus on bounds for |( 1+i?) |. As with Inghams
result c > , improvements at -|+i? improve the constant, in that a good
estimate for the mean of a higher power than |(-J+it) |4 would decrease
the estimate for class (ii) zeros, both in (28.19) and in Ingham s theorem.
However, even if we knew Lindel5fs hypothesis, we should only be able
to deduce (28.20) for c >
It has long been conjectured (Cramer 1936)
that for large x there is always a prime p with
x < p ^ a;+ 0(log2a;),
(28.21)
(28.22)
(28.23)
with c < 1. No better way to prove bounds for |(l-f~i?) |is known than
to replace exp(iilogm ) by exp{itP(m)}, where P(m) is a polynomial
arising from the first few terms in the expansion of the logarithmic series.
Since m runs through integer values, the resulting sum depends only on
the fractional part of ^ / tt. More direct arguments fail, because t is much
larger than any other parameter involved. After this transformation we
must use the intricate methods of I. M. Vinogradov; W eyls simpler
approach gives only
l(+i*)l ^ io g '^ l+ e y io g lo g ^ l+ e 2),
(28.24)
where any higher power o f loglog t than the first would suffice for the
application. The proof of (28.22) and (28.23) occupies one and a half
chapters of Titchmarsh (1951).
4.28
GAPS B E T W E E N P R IM E N U M B E R S
121
-+ o K ^ ),
Iy \ < T
'
(28.25)
'
where T is chosen less than x with the property that each zero p = /3+iy
of l(s ) has
\y T\ > l o g 2 \
(28.26)
the sum in (28.25) being over all zeros p of (s, x) with \y\ < T . Hence
ftx+li)-t(x)=h+
xp~ { x + h ) p + o ( XP j ,
Iy\ < T
'
A = logo:.
(28.27)
'
(28.28)
H x + B h f-1
xP-1
(28.29)
Iy \ < T
\y\<T
for some 9 in 0 < 9 < 1. To estimate the sum in (28.29) we divide the
interval [0,1] into ranges [0, i], [|, i+ A ^ 1],..., [| + rA -\ \ + { r + 1)A-1] ,....
The number of zeros p = /3+i y with a ^ j3 ^ a+A -1 is
y*.*<i-<*>A27,
(28.30)
(28.31)
\y\< T
(28.32)
< 7iA28expfSA^A-*)},
(28.33)
Iy\<T
which is o (1) as x tends to infinity; here we have used (28.22). The error
term in (28.27) is also less than h when
h > a:7'12+8A2.
(28.34)
(28.35)
122
ZE R O S A N D P R IM E N U M B E R S
4.28
x <p^x+7i
loEP >
(28.37)
tt )'
(28.38)
NOTATION
2 .n
V
(m , n )
m od q
2*
amodg
9(m)
cq(m)
s = cr+ ii
x (m )
d(m)
A(m)
>fi(x)
fi x ) < g(x)
fix)
t(x )
X o (m )
M
IN I
H(a)
L (s ,X)
()
S(a)
7t ( x )
a/q
r(s)
(s)> (s X)
p
= /S+iy
< A ( . x)
ip(x;q,a)
Sx
X*
Xiiiodg
u = A+ ir
0(u)
M( 8 , x )
N ( X )
(Chapter 1)
sum over a set of representatives of reduced residue classes mod q
(Chapter 1)
Eulers function (1.10)
Ramanujans sum (1.11)
complex variable (1.10)
a Dirichlets character (Chapter 1)
the number of divisors of m (1.23)
Mobiuss function (1.25)
logp if m is a prime power p a, otherwise 0 (1.31)
sum function of A(m) (2.1)
\f(x)\ = 0(g(x)) (2.3)
the conductor of a character (Chapter 3)
Gausss sum (3.7)
a trivial character (Chapter 3)
largest integer not exceeding a (4.3)
distance of a from the nearest integer (4.4)
the saw-tooth Fourier series (4.5)
Dirichlets -function (5.8)
Riemanns zeta function (5.9)
an exponential sum (Chapter 6)
number of primes up to x (Chapter 6)
a rational number in its lowest terms (Chapter 6)
Eulers gamma function (11.1), (11.12)
functions occurring in function equations for (s) and L{s, x)
(12.2), (12.3)
a zero of (s) or of (s,x) (Chapter 12)
sum function of A(m)x(m) (17.1)
sum function of A(m) in the arithmetic progression a (modg)
(17.18)
a character sum corresponding to the exponential sum S(a) (18.4)
a sum over proper characters modg (Chapter 18)
auxiliary complex variable (Chapter 20)
all the junk in the functional equation (20.2)
partial stun for the inverse of L(s, x) (23.2)
number of zeros of L (s ,x ) m a rectangle (Chapter 23)
BIBLIOGRAPHY
The epigraphs are from
I. Winnie the Pooh
II. The House at Pooli Corner
B o m b i e r i , E. (1965). On the large sieve. Matliematika 12, 201-25.
------- (1972). A note on the large sieve. To appear.
------- and D a v e n p o r t , H . ( 1 9 6 8 ) . On the large sieve method. Abhandlungen aus
Zahlentheorie und Analysis zur Erinnerung an Edmund Landau. Berlin.
----------------(1969). Some inequalities involving trigonometric polynomials. Annali
Scu. norm, sup., Pisa 23, part 2, 223-41.
C h a n d r a s e k h a r a n , K . and N a r a s i m h a n , R. (1963). The approximate functional
equation for a class of zeta-functions. Math. Annaln 152, 30-64.
C r a m e r , H . (1936). On the order of magnitude of the difference between con
secutive prune numbers. Acta arith. 2, 23-46.
D a v e n p o r t , H. (1967). Multiplicative number theory. Markham, Chicago.
E s t e r m a n n , T. (1948). On Dirichlets ^-functions. J. Lond. math. Soc. 23, 275-9.
F o g e l s , E . (1969). Approximate functional equation for Heckes ^-functions of
quadratic field. Acta arith. 16, 161-78.
F r a n e l , J. (1924). Les suites de Farey et les problemes des nombres premiers.
Nachr. Qes. Wiss. Gottingen, 198-201.
G a l l a g h e r , P. X . (1967). The large sieve. Matliematika 14, 14-20.
------- (1968). Bombieris mean value theorem. Ibid. 15, 1-6.
H a l A s z , G. and T u r A n , P. (1969). On the distribution of the roots of Riemann
Zeta and allied functions I. J. Number Theory 1, 121-37.
H a l b e r s t a m , H . and R o t h , K . F. (1966). Sequences, vol. 1. Oxford.
H a p i d y , G. H. and W r i g h t , E. M. (1960). A n introduction to the theory of numbers.
4th edn. Oxford.
H o h e i s e l , G. (1930). Primzahlprobleme in der Analysis. Sber. berl. math. Qes.
580-8.
I n g h a m , A. E. (1937). On the difference between consecutive primes. Q. Jl Math.
8, 255-66.
------- (1940). On the estimation of N(a, T). Ibid. 11, 291-2.
J e f f r e y s , H . and J e f f r e y s , B. (1962). Methods of mathematical physios.
Cambridge.
L a n d a u , E. (1927). Vorlesungen iiber Zahlentheorie. Leipzig.
L i n n i k , Y u . V . (1945). On the possibility of a unique method in certain prob
lems of additive and multiplicative number theory. Doklady Alcad. Nauh
SSSB,, ser. mat. 49, 3-7.
------- (1964). All large numbers are sums of a prime and two squares (a problem
of Hardy and Littlewood) II. Am. math. Soe. Transl. (2) 37, 197-240.
--------- and R e n y i , A. (1947). On some hypotheses in the theory of Dirichlet
charactexs. Izv. Alcad. Nauk SSSB, ser. mat. 11, 539-46.
M o n t g o m e r y , H . L . (1968). A note on the large sieve. J. Lond. math. Soc. 43,
93-8.
------- (1969a). Mean and large values of Dirichlet jjolyxiomials. Invent, math. 8,
334-45.
------- (19696). Zeros of i-functions. Ibid. 8, 346-54.
------- (1971). Lectures on multiplicative number theory. Springer.
B IBLIO G RAPH Y
P o ly a ,
125
INDEX
abscissa o f covergence, 18
additive function, 5
approxim ate functional equation, 84-93,
106, 116
arithm etical functions, 2
additive, 5
m ultiplicative, 2
totally m ultiplicative, 2
B om bieris theorem , 35, 103-9
ch aracters:
con du ctor of, 11
defined, 4
induced, 11
propriety of, 11
sieved, 74-6, 79-82
trivial, 3, 13
classification o f zeros, 99-100
congruence 1
D irich le t:
characters, see characters
divisor problem , 8
polynom ials, 80, 84
series, defined, 18
divisor fu n ctio n :
averaged, 7 -9
defined, 4
elem entary proofs, 69, 107
E ulers :
function defined, 3
product., 20, 52
exceptional zero, 60-5, 7072, 76
exponential maps, 1
exponential s u m :
defined, 24
sieved, 28-31, 77
spikes of, 24-6, 110-11
F arey:
arcs, 110
sequence, 36, 74, 110
F o u rie r:
series, 14, 37, 40-1
transform, 79
Franels theorem, 36-9, 74
functional equation, 20, 41, 45-50, 69,
84, 116
Gallaghers :
first lemma, 76-8, 812
second lemma, 80
Gauss, 11, 22
Gausss sum, 1113
H adam ards product, 50-4, 69
H alaszs m ethod, 30, 114-18
ineffective constants, 62, 72, 110
Ingham s theorem , 98-102, 105,
118-19, 121
integral functions, 50, 85
108,
128
IN DEX