The University of Maine

DigitalCommons@UMaine

Honors College

5-1969

An Elementary Proof of the Prime Number Theorem

James G. Huard
[email protected]

Follow this and additional works at: https://digitalcommons.library.umaine.edu/honors

Part of the Number Theory Commons

Recommended Citation
Huard, James G., "An Elementary Proof of the Prime Number Theorem" (1969). Honors College. 20.
https://digitalcommons.library.umaine.edu/honors/20

This Honors Thesis is brought to you for free and open access by DigitalCommons@UMaine. It has been accepted
for inclusion in Honors College by an authorized administrator of DigitalCommons@UMaine. For more information,
please contact [email protected].
AN ELEMENTARY PROOF OF THE PRIME NUMBER THEOREM

James G. Huard

A Thesis Submitted in
Partial Fulfillment of
the Senior Honors Program

··· .. May 5, 1969

 INTRODUCTION

Among the oldest and most intriguing problems of number

theory is that of the distribution of prime numbers. The most
important result yet obtained in this area is the Prime Number
Theorem. In order to discuss it, we shall find the following
introductory material particularly useful.
Defini,tion. An integer p > 1 that is· not the product of

two other positive integers, both smaller than p, is called

a prime number: an integer that is not prime is called a
composite.
Thus, for example, the numbers 2,3,5,7,11, and 13 are
: - ,.... _,_,'}.r'<'-i(,

' prime, whereas 4,6,8,9,10, and 12 are composite.

It has been known since antiquity (Euclid, about 300 B.C.)
tha·,:. the number ·of primes is infinite. The proof of this fact
is qQite short: Let 2,3,5, ••• , p be the series of primes up to
p. Now form the number q=(2·3•5• •••• p)+l. Since q is clearly
not divisible by any of the numbers 2,3,5, ••. p, it is therefore
eit'her itself prime or is divisible by a :prime between· p and q.
Il'• . ~ther case there is a prime greater than p.
t:Iowever, there are also arbitraril:Y.{large gaps in the series
o. imes': Let k be any positive integer. Now consider the
ir.·, ers (k+l) !+2, (k+l) !+3, ••• , '(k+l) !+k, (k+l) !+k+l, where n!
is o.efined to be n(n-1) (n-2) ••• 3•2•1. .Each of these is composite
since j divides (k+l) !+j if 2·~j~k+L ThuEO given an~ positive
integer k, there e~ist k consecutive composite integers.
 2

It is obviously important to know something about the

occurrence of prime numbers among the. natural numbers. Let

Tf(x) denote .<the number.of primes that do not exceed x.

Because of the irregular occurrence of the primes, we cannot

expect a simple formula for lT(x).

In a notebook published posthumously, Gauss (1777-1855)

conjectures .that

1(.-4"" Lj X
= ?

where Li x= S' ~ dt/log;.:t. He arrived at this conjecture by

observing that Jthi;!.,~primes seem to have an asymptotic density

which is 1/log x at x . Legendre (1752-1833) conjectured that

Tf<.'X.) = f. ·, +hzl- /S , lTcx.) "" X

.,.Je<YJ x ·. ld X
This conjecture is today known as the ·Prime Number Theorem

(henceforth denoted PNT). Gauss's conjecture is the· more

profound since it has since been shown that

+ ..... + x o+l [I +
PxJ' ee~(:il.. '
X)
D •,•
.
( ~ . .
where C(X)~ 0 as X becomes infinite. In an attempt to

prove the PNT, Tchebycheff (1821-1894). showed that

(o.92-···)_'i_ :f 1··~
''d)(
 3

He also proved that if the limit L in Legendre's conjecture

exists, then L=l.
In 1859, Riemann (1826-1866} approached the problem
indirectly and connected it with the zeta-function
a>

'! (S) = L=
'Y1 I
1'1
-s

He was not completely successful; his pr9.of had severa.l serious

gaps. The most important of the~~ co~ld not be. :filled until the
properties of entire functions had been established. In 1876,
using Hadamard's theory. of entire functions, J. Hadamard (1865-
1963} and de la Vall~e Poussin (1866-1962) succeeded in proving

the PNT. Several of the gaps which still remained have s·ince
been taken care of. However, the so-called Riemann Hypothesis
(if '! (x+iy) =0, then x=r>,. which is most imp<:>rtant to a more nearly
'
precise formulation of the PNT, has so far defied all attempts atl
a proof or refutation.
In 1948, Atle Selberg' and P. Erdos succeeded in finding an
"elementary" proof of the .PNT. Here "elementary" is used in the
sense of avoiding the use of complex variables, Fourrier analysis,
and similar non-elementary methods employed previously. 1hat it
should not be construed to mean "easy" will soon become· apparent.
The rest of this paper is devoted to such a proof of the PNT.*

*Specifically, the following proof follows closely .that given

by J.G. van der Corput (see Bibliography) which is based on
notes from the 'conference Erdos gave at Amsterdam for the
"Wiskundig Genootschap" ·in October, 1948.
 4

PROOF OF THE PRIME NUMBER THEOREM

'' .

In the theory of numbers we usually consider, not the

function TI\x), but the function

'1.9-cx) = E /09u p ,
p~ l(.

which is much easier to work with. The above sum is extended

over all prime numbers p~ x; It is sufficient to show that
~(X) approaches l'n.as x becomes infinite.
X.
Now
== lTcl<) /d X' •
For 1 ~ y~ x, one has

lTcl()- ~ ~ mx)~7T~d) -«

Thus
(1)
+ I'6 l( • 15!-o<) .
l?(l()
tcr~ )(
Let us choose y=x , where (> (x) < 1 and E' (x) ---» 1 slowly
enough so that

loj X _--..;. 0 CiS )( --4- co • ·

V-l(C-1.)
X
, Such a f> (x) exists: Let !(CX) - 1 -
. .. I
Then · /iw. .( I - I•'i /<Xf )( ) = l .
X~co U 0

Now show

.•
 5

I
?<

0
(1. /o~)/) C/"Q /o~){- t<_)
3

Cl"o 1-q. )( - 1)3 )( /o1tlq ll y ~~X

:: 0 , .:,'nee. f,rn, =0·

I!~=
 6

-r-~''' Equation
I
(l) now becomes

-0/-C X) +
)(

Thus 't9cll) ~ implies :that TT<Ji~ lo x --? {i.e., PNT) •

;;:r 0
Our· proof is divided into two distinct parts. The first is
devoted almost exclusively to the proof of Selberg's formula:

rifcx) +
)(

In the second part we will deduce the PNT from this formula. ·

I Proof of Selberg's Formula.

"
Let j1(m) .denote the Mobius function, which is defined as
follows:
i) f1- (1) =1;
2
ii) ,P- (m) =0, it. p divides m for some p;
'iii) p. (m)={-1) r .... Pris the decompo-
sition of m into prime factors.

; ·.·

. we have defined f'- (m) •

Lemma 1. . For each integer h).> 0, the function
.(ph (n.) " V pcd.)/bqhd
. . .. Jim , . ·a .
(where the sum is extended over ;;~11 •diviso:J:>s of m, ·including 1 and
m) equals zero, if thenatp.ral number m.contains more than h
different prime factors.-.
 7

Remark: We will use this lemma only for h=O,l,2.

Proof: For h=O the function becomes
(2) ,fu;. jJ.Cd.) = 0
for all integers m > 1 (the sum equals 1 for m=l).
This formula is evident. Let ,m=p~' Pt''"••• p~r lethe decomposition
Since ,)l(d)=O if Pi 2 divides d for some'
of minto prime factors.
Pi• only I +( n + (:) +- • • • + (
of m have to be considered in the sum.
n 2 I"
Now the divisor
divisors
~=1

contributes 1, the(r)divisors p 1 ,p 2 , ... ;preach contribute -1,

the(~)divisors P1P2•P1P3•···•~r-1Pr each contribute+ 1, etc.

Therefore the total contribution is 1- (~) + (~)- ••• ± (~)
Thus the lemma is shown for h=O.
We now use finite induction on h. Assume the lemma has been
D(
shown for h~k-1. Let h=k. Also let m=p b, where o( ~ 1 and where
the integer b is not divisible by the prime number p. We then have

--
 8
Since m contains more than k different prime factors,
b contains more than k-l different prime factors. Therefore
by our inductive hypothesis ~(b)=O for n=l,2, ••• ,k-l. The
remaining term ~(b) Cf, (po!)=O since ~ (poi)=O, as we have
already shown. Therefore the lemma is true for all 'h.

Lemma 2 • Let x > 0 and set

and tcm) =Ld /'>I ACd)
(these are clearly also functions of x). Then we have
2
f(l)=log x:
o( ' 2
f(p )=-log p + 2 (log x) (log p), p prime, o( ~ l:
f(p"qP)=2(log p) (log q), p,q different primes, r:/.,13 ~ l:
f(m)=O, if m contains more than two different prime factors.
,Proof. t (I)= L
t/./1
~ cd) ·= A(l) = ILCI)
,-
k"10 2 )( = /.,"1: "'-x.
6

>
 9

/o~ 7.)( • 0 .o + 0
0

·by the previous lermna.

'/!)-C)()
Lemma 3. For x ~2 the quotient X is situated between
the two fixed positive bounds. (Tchebycheff's Theorem, 1851-1852)
·Remark: For the proof of Selberg's formula it suffices to
'k now that th e upper . .t of
lJ.m~ "13-c.>~)
x ·
~s f ~n~
· · te. We prove here
that the lower limit is positive because the proof is quite
similar and because we shall need this result in applying Selberg's
formula to the proof of the PNT.•
-c9cx)
Proof:l. Let us first show that the greatest limit of X
(.2,.)!
is finite. Consider the natural number P= • Since the
prime factors p '> n and ~ 2n appear in the numerator and not in the
denominator of P, the product of these primes is a divisor of P,
and thus is·at most equal to I+ ( .1 1.>1 ) . .
+···+.,+-····. ,+-I(::1.") + !.= (:2")
(l+l) 2 n= 2 2n. The logarithm of this product is equal to
"'"P"'2"
/o9. p
G
> =
'19- (2n)- 1.9-(n). It follows that ,9'(2n)'" -{f(n) ~log 2 2n= 2n log 2.·
Fo.r x=2m (m an integer ·:;::,: 1) , we obtain
1.9-c x) - { 1.3-(.2"') -1.9-ca"'-'A + [-J-c:;,"'-') -V{.21?1-'')J + ···
.f.. • • • + a..) /"(j 2..
 10

For 2m-l< x 0::: 2m 1 we therefore have

tt9-cx> ~ '\Juz"") < :2.n.+l lo~ ~

< 4~< /o~ :;;.. ?

so that we find that for each number x > 1

n,!)-('.1)
:X
< ' I ho /J '-f,q""•
2. Now let us show that the least limit of is

positive. The series 1 1 21 ••• 1 n contains f;] multiples of the prime

number p (where [x] designates the greatest integer -!5 x) 1 { J

;~
multiples of p2 1 etc. There·fore n: contains precisely

[~J +·[ ;.,.J + ... factors of p. Consequently, the number of

factors p appearing in P is exactly equal to

o = L f C ~ J- a [ ;.,JL
where the sum is extended over all natural numbers o( such that "'
ol < . < [ /o~ .Z~ J
p , 2n. The number of terms of Q is - 1<>cr p J.
[2y]-2[y] is a function of y with period one which equals
l
zero in the interval 0 "- y < ;i and equals one in the interval
I .
:£:!y<:l. Therefore (2y] -2(y] is always ~ l,from whic~ it follows

that Q ~
r ,D~~~] , so that P divides the integer U defined by.
,.
[ 1'1 r . /o~ 211 ....... ·,

'"~~ p
Since

we have

)') r"J a, ~ 'CJ?

We also have
lc(j .2v. J
[
''a p ~ p

a01d [ /;f~]
 11

Thus

We therefore find that ~(2n)~~ log 2)- (log 2n) J2n. But by

. l' Uopi tal's Rule, we know / iYM cJ:4 :Z11) 1f2i

)J-;>CO I 1
. -,;:-{n-•) ~~~

= ./9_ ICR
0
j_ {J 'v){.
h-7=

Therefore for c "'1, 3 (there exists) N > 0 3 (such that) n> N ::>
(implies)

( lo;1.h)~ :::. (J~2vt)0 .{; e - 1

1
~ ( 1._,) I~ 2. :&.en- 1) l'f! 2..
;> (Ia~ 2"')& ~ (Yi;') l'tf :z. · = "f~ ;;;_ -l C:n +J_) fd :2. >fur- ~If h >N.
Thus -8'(2n) ~ i(n+l)log 2 for sufficiently large n. If x is

sufficiently large, and if we let 2n ;;;' x < 2n+2, we obtain ·

"l9-c)() ~ 19- (2n) ~ fi (h +)) /oJ ;J.. > ~X fty ~

from which it follows that for x > 2, ~ has a positive
lower bound.

Remark: The above reasoning leads us to another formula

which we shall not need for the proof of Selberg's formula, but·

which we shall use in applying this formula. We have stated

that n ~ contains the prime factor p erac,tly

[; J + [T] + ··~
 12

times, thus log n: = ):: ( [ ~J + [ ;_,_J + ·· ·) /""l p.

. f'\!iYI (.J I

2
Since this sum <. ~([~J + ;~ +···)lo<tp = 2:._[~] 1 p + l1 ~-p--<J<-p.L...~-,1 )?
we find that log n: ~s approximately equal to
p~YI a f~}ll 1'1
[f> /oq p
J P

p!!11 . (J
and their difference is at most of the order of n since
-c- '"I! F converges. ·Now is
Yl L- pcp-r)
p
approximately equal t o P~"
L
"" - 111
p and their difference ~

is at most equal to L lo'l p p~h u

, which by
. Lemma 3 (l) is at mos~t the same order a's n. In this manner we
obtain· that -¥·
Lpto 'Vi .
is equal to
~

~ 1~ 'Yi ! " {;" t;:>. /ccr h ,

to within a bounded term. We have that for every integer h ~ 2,

l·oh = lod ( (h:~~~l . (hhl r-1)

= hloah -ch-1)t -ch-i) 1ca (1 + -i-J-).
6 ck-1)
1

'

But for any natural number n we' know from the Taylor series
expansion that log(l+~)=!,- G;, wher~ 0 <e.< l .
. ., 11 .
Thus n ' e .
log(l+~)=l- ~is bounded. Consequently,
~ .
-?- I•Qh =
~ 0 h~ Q u
t=. {
h IO'Jh- Ch-i)h..'ICh-1) -1- OCI)L
.
(where "f (x) is 0 (g(x))" denotes that. jf (x)j < M•g (x), for some
positive constant M and for sufficiently large x.)
- 'r1 fod V1 -0 + .0C'V)).
Thus·

gl~h
l<l?l p
L
'f'~YI
p = ~
lrl "+ 0(1) =
l·~f"' + o( 1),
' '
 13

Lemma 4: ~(l~p)(lod ~)__,.o .:>s X-;, co.

Proof. Let E.> 0.

~ I
We have log p <log£ for p >
I
c x, so that
the expression in question is at most equal to

'x' i,x ( L
p~€X
(/o~\l p )(lex;0 )i) + "Lp~X
(loqo.p )( /o~ T))
q

'l!J-cc:x) + -&ex). lo(l Ye < I !.;}

x >< lc. x , . c [
0 c ..j. ~.)'5~
Where cis the upper bound of Lemma 3(1). When£~ 0 slowly
enough so that

Cf:Jx)= satisfies there conditions), we obtain the desired

result.

Lemma 5:*

the sum extending over all primes p and integers ~ such that
)(
p<;J.~
r".
Proof. + ... +--..9'( kR ).
. ;,vi/ '
X
where k is the greatest integer such that 2k ~ ~ ; >.e.,
X
6
k '" '"~ TIS .
"ci 2
*This lemma is not as it originally appears in van der Corput.• s
proof. I have altered it slightly so 1;:hat it more nearly
directly applies to the proof of Lemma 6 •..

--
 14

Thus p~.l!. Id p ::; JC ;~)

r/ ~ v-cr:)
)(

We know that is bounded above Y (for all) y?: 2 and·

equals 0 for l~ y <. 2.

Thus !VI
}o(/

"'it
1·.:s.
0 ~ f,;,.,
><~co '"6 .7- ~cJ~h) ==
' ;, w.,
ll-= '"6 ~z/3. _:?c J7:-)
~h.
·Y foQ- x
b. . lo?f 2 /;I' ·ff Iocr-<

(But the given. limit

The lemma is thus proved.

Lemma 6: £. fcm)
m"'X
"' lo51 >< ·1k>O .J-a L'l9-C~) log p +
\J F~fi CJ .

o (x log x). (In general, o (g:(x)) designates a function of x

such that o (~C>O) __..;;. o as )( -:l> co. )

. 3 ,!()
. '
Proof. From Lemma 2, we have
(3)
+ .2 lo~ >1 b~ p)+ .:z L loH !on
\J 1'~%1',;~

p< ~
'
·r-·......, . ,,
~.

'. __
! ~·-· Let :us· first' consider !±he second term on .. the right ,hand:side
(RHS). ·.The contribution of' the terms with c1. >,;:;2;:'.iscT·ati'most
 15

where k designates the greatest integer such that 2k~ x. So

2 . ::IYx 3
that the contribution is at most 2,[X k log x ~ '(f ::z. log x

equals o(x log XY) 1 since

~·
,,~
)/..;;.([)
JX (Thx ::: /.YJ{
I(_;,~
/o
w 2 )/ ::. {,%
l(...>ro
(2
X
x' )"a Jtivx
= 4 X-"'=
/,.,;.,
¥ k 0 .,

The contribution of the terms with ~ ~1 is equal to

L {-lott2. p + Q
p~ X 0
/oql(
J .
f<>t
\}
pJ "' JO<J X
0
Lp,.; X
1°9· F
U
+ LC/<>3 p)
p;; X IJ
loy
V
~ ~

=(log x) '\J'(x) + o(x log :it), Jnr Lemma 4.

Let's consider finally the third term of the RHS. The
contribution of terms with·; 0:. 2 and ol. ~1 is by Lemma 5 equal
to :2 2::.(1 ..9. p)(lo ~ <f-) . :: :2 I:. /og ~ ( ~ '"~ r)
p"'~~~X u a V '(/ox UIJ f'~ -'S, CJ
f' "' 'I> JB "',2, . o

We show

~~cq b o(f /~ ~)) I

= 0.

~ (l'(t ~ ~~,' /~ x)
Now
"'"'"' by
.
definition :::.ve>o
 16

=! N 9 Y\ ""N > \ o ( ;" lo~ 11)\ < e \ ;tl I~.,!

lh l(S vE ::> 0 ::l N ;) )II >N =>
. '

~ e 2: (Jo~ ~ • ;P /~n).
;(~Vi .

'orJil))
:;:; 0 ,
L
0,8~ )(
/orr 6 ( {,.. /~x)
'

that is : J;JI~ b o{ ;" 11-x)) "' D {~'I /'1 'b ;t< !"(><).
Now lim
!(_.;,co

'
' .
 17

since N is finite,

~ ctll /~x)
2-X c1rct 'D l)!o,B rcaJ
~.13 6
)
- .~ {

:: 6 , si11ce. v E:>o ;1 N >o ,; l1 >N

. I
< e ( loq 1> f.s 1[-lf J) V 0 pr'"»tr:.

~ e 3{;;_ ( /'a- 3 'l'D ~ laJ I(). .

Now we have that the third term with o{'}. 1, jd"':- 2

= o( ~ (hf6 ~7/ foax))."' 6 (~,_ /'1o "18 Joax)

Zp/.2. . . 3 r~'~"" .
=o(x log x) ~ lqrp~ , where this sum is extended over all
~
primes q and all integers f ~ 2 and is convergent since
.fh- ..c _E:.
foz I
?? - _ I
[)Co- I) ~
0 .a Consequeptly the contri-
bution of terms with jJ 3- 2 (and· equally that of terms with
d ~ 2) is equal to o(x log x), so that the last term of the
RIH S of (3) is equal to
.....
~(--~,.. 2 ~·· /oar'~ 6 . + o ex /1 -x).
The first term is equal to
"'"'o

L ( fo91 f')( I~ 0) - ~ /u:;1, ·-. p

I Po ~ x u. (j p.:fl v
•

I
 18

~
= rL~ f1< 1"9.P
u
9
o :r'JX. I~ p b~ 6 - p-·"'-
+-L ~ lo'" p /o~
~
L lo;r'·p
o P"..r;;. u
(f
9_ -

\>{).; )( . h ~11. %~"" .

,. P~ lo~ f 19-L~) +- ~ /, 6119-c;) - 19 ~lfi) + 0 ( YY: Iof)()

= 2 L fo~ p '19-C~) + oc x illl'X) >

. po"\he IT

vcvxJ
because we know that 1/i. is bounded as x -7>-CD • Therefore
. . rc)/l"(J=j;) ~ 0 (I) '=? {) 2( J7r.) "' 0 (>() -::::;> /1 ..,;_, J}- 7. ( fi J
X __,..co 'X. l G- )I'

;::o.,

The term 0 (.[X log 2x) clearly is o (x log x).

Thus we find that the last two terms in the RHS of (3) are

equal respectively to (logx) V(x) + o (xlogx) and

2 U-& ~)
p~ll'l
log PJ + o (x log x) , from which we have the lemma.

Lemma 7: For each natural number x, ~ I•

..\~

Proof. Because of (2), we obtain

1 = t. ~f-Cd) = ~ J-Cd) 4- 1 ,where h designates

the positive multiples ~ x of d. Thus from which

~ . )(
1 = ,L__ )" Cd) [ T].
d=.l
 19

Consequently,

1 '!( t J;-:d) - 11 =I k ;M:d)fi - rrJ~l

I(

-: I:"lj
d=
-r1Jl I
~ x-1
since each term of the last sum is ~ 1 and the last term
equals zero. Thus we find x J {
d=l
A:J I ~ 1 + (x;-;1) = x:,~
from which the lemma follows.

Lemma 8: If g(t) is a monotonically decreasing function,

g(t) )' 0 for all t '> 0, then
~
1-VI"Yll
g_()1) ~ s~ 0qCf) ci+ + c + ocC}(.xJ).
0
Here n runs through integers only; X can be any real number,
X ~1: and Cis a constant depending only on the function g(t).*
Proof. Since g(t) is decreasing in the interval [n,n+l],
511.J.I
we have g(n+l) ~
~::) dt ~ g (n) •

-s
. 11

and thus 0 ~ dn =gn >1 act) eLf ~ g(n) - g(n+l) •

Therefore we have for any positive integers M -< N,

·. 1Mdl1 ~ 1~ i 8c»)- iJc~+oJ "' 'del\!) -(feN,.,) ~0 ckl)

I
This shows that the series ~ dn converges. In particular,
we have L
~-tt:l
dn ~ g<CM). If;;' put C = L\1:)
dn, we have
I
I

*Hans Rademacher, Lectures on Elementary Number Theory

(New York, 1964), pp. 98-99.
The material found in van der Corput's Lemma 8 is. given
below in Corollaries 8.1 and 8.2~
 20

jL SN;-1
It follows that 2:__ g(n) = 1 g(t)dt + C + O(g(N+l)).
I'\= I
For N=[X] this becomes
\ L'XJ+ I .
.r £:Y o
1.; 11
qcl'\)
.
= .J I .
.
q UJJ+.
a
r c -1- 0 ( q(fYJ +I))
a

·~ .··.sY + c + 0 CgCXJ)
since
s~+l. . .
sc+Jd.f. ~
. I

(\(.lO.
d(t) cit
o< (]([X] -1- i) ~
y avd (f CX.J • .
This proves the lemma,·

Corollary 8.1
,,
L. :h = log. y + C tO('~~).
I

ll~~ . . " )-
Furthermore if we let · t:(y) = L__ 'h - log y - C , then e (y) --.> o
. ~~~

and 8 (y) log y --l> 0 as y _... o:..

Proof. The first part follows directly from the lemma

if we simply note that S~ idt = r lo0 tJ~ =- lo6 ~-

For the second part, since C. (y) = 0 ( t ), the two limits follow.

I' . Corollary 8 • 2 ·~ ·~ - -b lo~ 2 ~ r C ~ 0 ;ts !f~w. ·

i
' Proof. Since s; ¥dr . = [£Ooa+)':J r = Jj_ hoY· .
we have L !of' - -1 }~ ?.~ 4- C = 0 C;1) ~ () <lS
0--?>co.
"'" ~
Lemma 9. If ~(n) designates the number of divisors of n,
. ~ ?'CV\) 2
then L-- ~. = log y + c3log y + c4 + o(l), where c3 and
"~~ . '
C4 are su~.tably chosen constants.
Proof. Since ~ (n) is equal to the number of pairs of
·~ '1-GVI)
natural numbers a and b such 'j:ha t ab = n, the s urn .L.___:_
, "'1 ."Vl

is equal to the sum ~ cJ; extended over all natural humbe~s

a and b such that .ab ~ y.

 21

First evaluate the conttribution of pairs a and b with a ,::;: fj.

This contribution is increased by the equal contribution of pairs

a and b with b ~JY, and the result must be diminished by the

contribution of pairs a and b such that a,b~JY.

Now apply the preceding corollaries. The contribution of

pairs a and b such that a ~{Y can be written in the form

L *- :L t = ~ ± f /o~u Ira- +C 1 + 0 { lo~ 't ~

~~ ~~ ~~ . vo

_:r /otct,
a. <";j

= c)o(J~)(I,d~ 1-c, fo(r~ 21 )~- [~'"a. ~ + cs + oci)S

+ c, { f"c ~ +- c, +· oc,)S · +· .o ( .~"~ ~)- { I'd~· +-c, + oc•))
3 I z I
-= 8 "6 t + c& 0 J J + C 7 + oc r),
in this .reasoning cs.• ·c6i c7, designate constante conveniently
chosen •.
Finally the contribu;ti6n.:1of terms with a,b { {Y is equal to
( ~ ~ )Q, . . .
a~~ .
Clo:~ ,fi;
u d
+- c, i-. ecvl,)
. a
y
" ~Icc/(] +- c, 100 J + c, 2 +- I?CJJJ{.21~f _+ :Zc, +ec1f~

··x,-z
:- 4 Od d +-c,
I"J d +C, ::1..· f-OCI).
Thus we have shown that the lemma is true since -}& + j/11" - Y.-;

''
 22

Lemma 10

Proof. From Coi:ollary 8.1, the left-side member of the

asser;tion is equal to ~. g.Jcd)

L- f ~·
L- .Y, + c, +- e c.1J5
<:l<d. ·. 1\" j-
p,w .
+-c.> ds X
-d-
· .
+ L A:}J
d6')(;
ecJ)

(we have set dn=m). 'I'he first term is by (2) equal to 1 and

the second is from lemma 7 in a,bsolute v:=tlue ~ )ell•

so that it suffices to demonstrate

L ~ fecfJ\ = o(~Y).
~<;~) d.:)[
Let y =x '· where ~ (x) < 1 and \) (x) _,. 1 slowly enough so
1-('01)
that x - 1b . as x __.;,a;. • (Such a \) (x) exists since we have
seen on pagesc;4-5 that EJO.'J = I- 1, los, X · satisfies these
conditions.) For the numbers d t: y we have
CJ u
;:rx ~ x:.::1-e so that
X
t( ;r) ~ 0 and the contribution of these numbers d equals

0 ( L
d-6-,t.
t) = 0 (/o'f)():
'J
To prove this last statement, we show that

= C)

2:JJ
d~xe

Now lim o ( 1) =0 means that V e> 0 :1 N 3 d ~ N -::? \o ( 1) l <'g.

~""

""Th QS ve "> 0 g [\) § d ~N §> 2: cr. 0 c 1) < ~ .t .£ < c ~ Ycl.

N {oJ{ x'? . tJ~d! liP d ~ >1f
 23

.;rI ·ocl) ·J_

d • DO) . ::: 0.
~,t 2_~
.h )(f'
. . c&¥e
Therefore we have shown the given limit equals 0 as x~ co •
Sotha~. d~2_;]:/e:cfJ=--T
)(I' ·
~-oGI) dt!){'C
·. 6(z--.~)
<1~ l('i' .
= o(2_ -11)
<iO(

= o( t. (x) + log x +C) = o(log x) since 8(x) ____,.. 0 as x -Q:I

The contribution of numbers d such that y < d { x possesses at

~ j_ . X J
most the order of magnitude of L-d ,sJ.nce £(d") =::JO(i()
d'd< X .
= 0(1). This sum is,according·to corollary 8.l,(applied twice,
' '
the second time with x instead of y) approximately equal to
log x +.<,;!logy= (1- ~)log x in such a way that

.C !J - c,_ e) x
~<.J" X
---;. o • Thus we obtain the desired
result since 1 - ecxJ -o.

Lemma 11: For each natural number k,

LfcUJCL~) = I .
dlk .
Proof.
. '
=·~ I · ') the
Since 'rcVl<J
f.TY;i
left mernber.of the
assertion can be written in the form

L. ~ra) L_
d\k r: d,/:

"·
 24

The contribution of divisors dl<k is zero according to {2) 1 and

the contribution of the divisor k is equal to 1.

j.L{.d) !2 X
Lennna 12:
.
Ld!>< d bq
0
;I ==a/'dx +oC/c(lx).
Proof: From lemma 9 {applied with y = d:< ) 1 the left

member of the formula can be written in the form

a L Af) { L ?:~) - c lo~ 1 - cJ + o ( lo~x ), 3
d "' )( . "" • >Yd
since ~ /y~J.> ox '1 ) ) = o{log x)~ where g{j) = o {1)
' . J!i ")( ,- d 1
means that lim g { ; ) = 0 • Apply corollary 8.1
~ -i"'Q,?
X
The contribution of values of d ~ '".ct )( is equal to

o{ {..,. ~ ) = o{log x) 1 since d f XfJ"'a Y :;;.- ~·

Thus X--?>c;:, =:> /~ x ~ = =Y

X
-;;r- ----;. cD 9 o ~ c: !.) · ~ c:> •
so that ox
;r {1) is
. in this case 6{1).

Thu..s {,w:, L .t
~-"""' cl•')(;f"dx
Ox.
d
c I) "
·
/,>i<., LcT·OCi)
x~ J~'A·;t>~
= 0 (
.
L
J,;<Ji'Gx
YJ)
·

Z Ycl . ) = o {log x) •. The contribution of the other values

= o { d"'-)to

of d is at most of the orderoof

to U 'I' o{log x) 1

where

u =a~~
-·jz=.t f::.:.x
II

I
I .

I
 25

After these preliminary consideration, we can now prove

Selberg's formula:
'l5k.)() + .2 - ~
----r- ~ -&c r ) lo(l p --}-a a1t x-?>cO.
From Lemma 6, the left side has, to a term which ----r 0 as
the value xI, X

We must now show that this expression approaches

£=>< tcru).
2. From the
definition of the function fc(m},
L
\11~ X
fL'M) = 2_ L_ >.cJ)
111!1£ Jj))1
=
where h designated .the multiples ~ x of d. consequently L.1
h
·
l.S -.-X, to a term .wh'hh
equa 1 to ·iT'• l.C as absolute value ~/ 1, The
~ ;

sum L
WI" X
f(m) is therefore approximately equal to
)..cd) - X ·
x ~ --;r
d;}z
= X 2_
d cS X
TI1(J) lac,CJ 2
d = 2Y I~)( l-o
'()
1
(J( Jo~ x)
\J
from Lemma 12, and the error is at most of the order of
magnitude of L..
d.,;)(
l xc.n \ = z=
o.,;x
lo,."'
\J
J , so it suffices to
show that this sum is equal to o(x log x).
X X
,;;,~
The contribution of the terms such. that .:zk+i -<... .,

. 2 ~
where k designates an integer .,0,. is at most (k+l) ;-r< ?

r-"'~''iiince the number of terms is. at most ;I< and ~ach ·~arm . ~ (k+l) 2 •
Consequently the sum is ·less
~ d~+J)' )(
\<"'O :zk
thus at most of the order of x and therefore o(x log x).
 26

II Proof of PNT Using Selberg's Formula.

In this part we will employ the three following facts:

(1) '\9-\t.) + Q ~ '1.9-C ~) Jo ~ :2 a'S Y ---;> ro

X )( Ia:! ')( j>o!i{i[ dp
'
(Selberg's Formula);
'l!kY)
(2) lim inf -':>(- 0, demonstrated in Lemma 3;
l(...,>c:o
(3)
I
L lo~p 'X~co
' shown in
-')-
foG -x p6'll "'""
the remark following Lemma. 3.
fl!kl!)
From these we shall prove that x ----.. as x~ro ,

which, as we have shown previously, implies PNT.

'l9<;y)
The formula (1) shows that x- is bounded, so that the

limits

exist. We have 0 <fa~ A, and means that a = 1 = A.

Lemma 13 : A + a = 2 •

Proof.· It. is possible to make x become infinite in such

a way that
'\J{.x)
~ · ~ A. If
. S designates a fixed positive number,
_() X
'LT ( p )> (a - S lpX for each x sufficiently large and for each

number p ~ -.Jx, thus

'1. - X :2Ca-S) )
)()o<>X
(i
L
P"R
'\.9-C-p) lo't p
<l
'::;: I
o(l
It
P'" n<:
The last term approaches a -S in view of (3), so that Selberg's.

formula shows that A + a - ~ ~ 2. This is true for each positive

. fixed S Thus A + a ~2.

If in the above reasoning we replace· A,a, 0, and> by a,A,-[;,

and< respectively, we find that A + a , 2. Thus A +a = 2.

 27

In the rest of this argument x ~a:> in such a way that

~.,..) h r.1.s positive and fixed.
)( . __,..A, and t e number <i

Lei!Ulla 14: For each fixed A> a, - I

L_ I .5Le. ~ 0
p
' )I ~)(
if is extended over numbers p :::=:x such that '19-(p) ~ p •
Remark: This lemma is a theorem of compensation: . if 0'(x)

is large, v_<l-(;) is small for "almost" ,each number p~x.

1

Proof: ~)( &C ~) I'd p . ~)( /cap '"crt

. = ~~d)lo0 p }~~ct),,r - (~ Jod p-{

"' .2 L
p~f"x.
-3' [~ J lo:~u p. •~..( {) CVk.))'.J, •
'
where p and q are prime • . The last term. is .at most of the order.·

of X in view of (1) -. applied with {X instead Of X.'

We can therefore write the formula of Selberg also in the

form
'19-cx) + I \ x
1< xl ;x ~yc-p)l~p ~ :Z.
)( 1
If p is greater than the value u conveniently chosen and ·
X X
dependent on £, t9 Cp-) > C d - ~) -p There exists a

positive humber b(u) such that {)-{~ ) '> Ca-, S) ~ - b

for all p' s such that ~ *'u. Thus the inequality is valid for

all p~x.
 28

If we divide the sum ~>x into L 1 and L; we obtain

~ l~f'
~ (a-.r)x ~- p + C A- J)'x' ~ ~
L- 1 p - bc9-c0.
If we substitute this result in Selberg's formula we obtain:
- ~ 1~e
A -1--J - b + (A-G) rIYJt, St-<p
r
/d I( L_ 1 p ~ :2.
which im~lies, since A+a = 2,
(A-a) lmvvL<f I~>< ~ 1 ¥ ~~,
from .which follows the lemma since ). -a > 0 and Sis an

arbitrary positive number.

Lemma 15: For every fixed jlu < A,

I
"d~X """ ~, h_j_
L-z P ~-
~ 0 .
1

if ~ 2 is taken over pairs of numbers p and q such that

. )( \ p_.Y.
p "' fi<- , D~ J1: )19- ( P'f> J ~ p~ •
Remark: Lemma 15 is. a theorem o_f double compensation.
l
II If tt9'(x) ·:is :iarge, 'l9- ( ~) is small for "almost" every number
X
p~x. and '19-(?'%) is large for "almost" each pair of numbers·

p ~ rx and q ~rr
\(
Proof: If we substitute p for x in Selberg's formula,

we obtain
::z.x
p
 29 -- - -

Substituting this result in Selberg's formula gives

..9-oo " :Zx +-o(X)- t\v . I'~L {If

{ a +- om1 ~
fl
where Y~ b
\'--IX, ~-Yp
loo'Yfp rlJ' ( ~) /"" p lo~
a
CJ •
u b . ·'
By virtue of (3) we find
~ ~ 1 4V
/ p :: ;;~. lo(] X + o C lu';!ll) ~ Thu.s V(X) = '"'SI )( +- ooo.
pn'X U - ;· U_.L 1 B
In each term of V, p ~ [X and q,;; ~ ~ • Thus pq '= p "- (pq 2 ) a: ~ x 71 ,

so that. '!3' (;~) < (A+ S) ;'? , i~ ,

x is sufficiently large.
I
If we divide V into two sums L:;. and! 2::.:2. ,we have
I

v {:_ p-x ~
La.
( lo IlYp -.· ~
p
• ~)-
'I
'
+ (A+ Ox ~~ (_ l • ~
L-.z\.''0 lifp p
·m)
0 0
= (A + ~ }x W-. (A + ~ - JA) x L 2 tr~ lYp • ¥ ·£ 1
#),

where W = "'?; "1I ( }o~

P- ·~-v-r
Xfp • ¥ ·¥ ) o
·
=
L '"o 'lp I b£
P.
p:5ff

becau.se. ot (3)

We thus have
(I~~)
~'~ ~LX) -6
.
(A+- n, X - iodX
4- -
(4 + 6 -p)x L2UoCJt. G> • D +-ooi),
from which

' .,

~ A+ s- ~ -1- OCI) •
 30

Consequently,
-..:::-- (~ • ~) < ('
lot)( L 2. f' '6 . - 0 '
from which results the lemma, since A-jJ- >0 and Sis an
arbitrary positive number.

End of the Proof: Let ([ be any positive number such that ua <A
and let the numoer Sbe so small that
( 4) A- o.IJ >-- So + a S •.
consider the sum
S -- \.Ls ~.
p
!:u
~
\.
.L- '+
~.
P •
~3 is taken over pairs of prime numoers p and q such that

p ~ JX ) ~ ~ J~ ' Pi -;;- N , -if ( ;~ ) ~ ( A - g) :'8 ,

where N designates any fixed natural numoer. ;· L 4
is taken over
prime numbers r such that JJ <: r '!': crpq. If cr~ 1, the sum
)
is naturally equal to zero.
For each term of L 4 ,
Y. !t ~Y-2 3;4
r ~ up~ = cr-p 2
( p{Yl. :::0 a- X x· = a-X
•
if x is sufficiently large •
. If xis sufficiently large, all the terms of~~ satisfy
the inequality
(5)

This inequality is evident for the terms with r ~ pq, because the'ii.
~o- (L ) '>
a.(.!) ':>--.·u X X > X
rut- p 0 ~ (A-.Op'(, ~(A-SJa-r --(a+O""F
because of (4).
 31
)(
Consider now the terms with r > pq. If we let r = u and
)(.
Pb = v, we obtain u< v "cru. If we replace x by v and by u' in

. Selberg's formula,

~ -z9c~) lo~ p = :Z'IIC><tx ·-~- o ex h Y),

(loqo x)-3-Gx) + .Q p~n u o ·-cr
we obtain by subtraction:

thus
. ~ ( l~v -lcai.L)
{}-(. u.) ~ loa u '\9-(v) - 2. (v- u) - 2.v · lo .u.. +- o c. J.L) •
0
In the second member the first term is at least ~(v) );;. (A- S) v
and the third term is o(u), thus

"SCtv)~ CA-g)v -2cv-tv) +-ocu....) =2.u....-(2-A+a)v +oci.J..,)

: : : . 1u.. ,.- (a+ £)0'"i.i:, + ocu..)
~ (a .;.· Q~)U.... + OGU,) 7

because of (4). and since a+ A= 2. Thus. we find, if x is large,

enough, that
X . · . . . X
'1.9--(f) ='19-(v..) ~(CJ.;.S)u = (a-d')T
so that (5) is valid for each term of ):__'I • Consequently,
(6) S-:> L ¥
~'"'"X
Ls¥·~ '
'19-(f) ~ (;H ~)~
where Ls is taken over pairs p and q such that

~·P% <err •.
 32

We thus obtain
CT
L.s loa r reo~ ~ F ~ I~ p L. rcl~ ~
p.:<H; 'b< ~
. (J
,... k 16~p~(o;) - 'I

.t: c, f'~ ¥ f c.:z lo(l X ,

where c1, c2 are positive constants conveniently chosen.

·consequently,
,_bj:_
S { C
2
foo,
G
X
f-"')(
f

-&cf)'>c<>+~) f:
from Lemma 14. We now introduce the sum
~

T=L::-:;-,::--::~c=---=-=--;-;-;­
P ,;r"' , 't ~ P''r , P'P *IJ
which is at least equal to
loG p
p •

Each of these two factors has, from (3), at most the order of
magnitude of log x, thus .T > c3 log 2x, where c3 is a positive
number independent of x. If we let .. ,
~~--· T =L 3
¥. '1 t. + ¥ .~ ,
L.~
where each term of the latter sum satisfies

pt H , ~ ~ J~ , '13- ( p;) < (A - S) p\

Lemma 15 shows that this sum equals o(log 2x).
' Thus
"
.L-a
.i:af.
P
. ~
o >:2
I
c3
ICO)(.,
:1_

if x is large enough.
For a fixed value of x, we· cons·ider the prime pairs p and q
in
~ o;:;;:::-
L- i< for wh~ch £.__ 1-f
!:a.::.
r takes the minimum value }J. ,
where p depends on x alone.
 33

Thus S ~ f- ) 3
Comparing this result with (6), we obtain 1 I)

. . ~
(7) f- = Z 1f r" ----,)> o.
Consequently to each positive c and each natural number N there
I

corresponds a number t = pq ~ N such ·:that

thus
L
t ·O·;;
ff'""
crt + < € ~

I
vi: L log r < €
1-J,-<i"" ~!T/: '
from which 73-'(rrt) -1.3-(+!a-) < eut.
If N, thus also t,is large enough, we have

Thus (a-£)cr- A;c < £[)'"".

..
This inequality holds for each positive number 8 ,so that
~
aiT - A 0.

Thus each positive number (J such that a cr <A satisfies also

the inequality a!J'"'"'.l.::; A. This result gives nothing new if a = 0,
'\9(¥)
but let us apply the fact that lim inf -x- is positive; that is·,
A
a is.positive. Each number rr< E possesses the ;ppoperty

{)
:z.
.c:
A
d • If Q'"'~
A/d 1 We Obta~n
. (Aa );2.. L_
A.
;I _...,
.._y

A 1 Thus A = a =1 since a~ A and a + A = 2.

a L •
Thus = 1 and :therefore

::. 1 .
Q.E.D.
 34

It has latelY::been possible to improve the original proof

given by Selberg and Erdas. The older versionsgave just the
asymptotic equality
l(
lTCx) ""' ) od X

without an estimate of the error term

)(
Ec:x) = 1Tc.x) - 1o
3x
S.A. Amitsur, E. Bombieri, W.B. J~rkat, R. Breusch, and E.
Wirsing have since given estimates for E(x) which approach the
accuracy obtained by transcendental arguments. However, these
recent elementary proofs become extremely sophisticated and the
. .
. ' . .
sharpest results' are still obtainable by transcendental methods.

\.
 35

BIBLIOGRAPHY

Rademacher, Hans. Lectures on Elementary Number Theory. New York:

Blaisdell Publishing co., 1964.

vim der Corput, J .G. D6monstration ~l~mentaire du th~oreme sur

la distribution des nombres premiers. Scriptum I.

Amsterdam: Mathematisch Centrum., ~948.

., ·,·(