Académique Documents
Professionnel Documents
Culture Documents
sin ~F(s)
The First 50 Million I - s ;
1
Pn+1 = [I - log2( ~ +
n (-1)r
)]
r~1 ISil<'~'<irSn 2 pi1"''pir - I
x ~(x) x/~(x)
10 4 2.5
1OO 25 4.0
1OOO 168 6.0
10,OO0 1,229 8.1
1OO,OOO 9,592 10.4
1,OOO,OOO 78,498 12.7
10,0OO,0OO 664,579 15.O
100,0OO,OOO 5,761,455 17.4
1,O00,0OO,OOO 50,847,534 19.7
10,OOO,OOO,OOO 455,052,512 22.0
~
smoothness of the latter:
4000
~t~(x
3000
6000
2000
5000
1000
&O00
I X
10 000 20 000 30 000 40 000 501000 .
3000 log x
2000
There is one more a p p r o x i m a t i o n which
1000 I would like to mention. Riemann's re-
search on prime numbers suggests that
I i I I I X
v
the p r o b a b i l i t y for a large number x to
10 000 20 000 30 000 40 000 50 000 be prime should be even closer to 1/log x
if one c o u n t e d not only the prime numbers
but also the powers of primes, counting
Therefore it is natural to ask for better the square of a prime as half a prime,
approximations. If we take another look the cube of a prime as a third, etc. This
at our table of the ratios of x to ~(x), leads to the a p p r o x i m a t i o n
we find that this ratio is almost exact-
ly log x - I. W i t h a more careful calcu- (x) + 89 ~(/~) + ~ ~(3~s + ... = Li(x)
l a t i o n and with m o r e detailed data on
~(x), Legendre (5) found in 1808 that a or, equivalently,
p a r t i c u l a r l y good a p p r o x i m a t i o n is ob-
tained if in place of I we subtract (x) = Li(x) - 89 ~ i ( ~ ) - 89 ~i(3/~) _ . . .
Correspondingly, the probability that As you can see, the agreement with the
two random numbers near x are both prime theory is extremely good. This is espe-
is approximately I/(log x) 2. Thus if one cially surprising in the case of the
asks how many prime twins (i.e. pairs of prime pairs, since it has not yet even
primes differing by 2, like 11 and 13 or been proved that there are infinitely
59 and 61) there are in the interval from many such pairs, let alone that they are
x to x+a then we m i g h t expect approxi- distributed according to the conjectured
m a t e l y a/(log x) 2. Actually, we should law.
expect a bit more, since the fact that n
is already prime slightly changes the I want to give one last illustration
chance that n+2 is prime (for example of the predictability of primes, namely
n+2 is then certainly odd). An easy heu- the problem of the gaps between primes.
ristic argument (9) gives C.a/(log x) 2 If one looks at tables of primes, one
as the expected number of twin primes in sometimes finds unusually large inter-
the interval [x, x+a] where C is a .con- vals, e.g. between 113 and 127, which
stant with value about 1.3 (more exactly: don't contain any primes at all. Let g(x)
C = 1.3203236316...). Thus between 1OO be the length of the largest prime-free
m i l l i o n and 1OO m i l l i o n plus 150 thousand interval or "gap" up to x. For example,
there should be about the largest gap below 200 is the inter-
val from 113 to 127 just mentioned, so
g(2OO) = 14. Naturally, the number g(x)
grows very erratically, but a heuristic
150~000 584 argument suggests the asymptotic formu-
(1.32''')(18.427)2= la (II)
g(x) ~ (log x) 2
pairs of prime twins. Here are data com-
puted by Jones, Lal and Blundon (io) giv--
ing the exact number of primes and prime In the following picture, you can see
twins in this interval, as well as in how well even the wildly irregular func-
several equally long intervals around tion g(x) holds to the expected behav-
larger powers of 10: iour.
12
700 This picture, I think, shows w h a t the
person who decides to study number the-
600 ory has let himself in for. As you can
see, for small x (up to a p p r o x i m a t e l y I
million) Legendre's a p p r o x i m a t i o n
500 x/(log x - 1.O8366) is c o n s i d e r a b l y bet-
ter than Gauss's Li(x), but after 5 mil-
&O0 lion Li(x) is better, and it can be shown
that Li(x) stays better as x grows.
300
But u p to 10 m i l l i o n there are only
200 some 600 thousand prime numbers; to s h o w
you the promised 50 m i l l i o n primes, I
have to go not to 10 m i l l i o n but all the
100 way out to a b i l l i o n (American style:
10 ). In this range, the g r a p h of
0 R(x) - ~(x) looks like this(13) :
10 102 10 3 10 4 10 5 106 107 108 109 1010 1011 1012
600~ R(x)-X(x) /~ AA
.A,.,, ,A , fL
.;o
-300| -- O~< x ~< 1,000,000,000
Up to now, I have s u b s t a n t i a t e d m y
claim about the o r d e r l i n e s s of the primes
m u c h m o r e t h o r o u g h l y than m y c l a i m about The o s c i l l a t i o n of the function K ( x ) - ~ ( x )
their disorderliness. Also, I have not become larger and larger,;but even for
yet fulfilled the promise of m y title to these a l m o s t i n c o n c e i v a b l y large values
s h o w you the first 50 m i l l i o n primes, but of x they never go beyond a few hundred.
have only shown you a few thousand. So In c o n n e c t i o n with these data I can
here is a graph of ~(x) compared w i t h the m e n t i o n yet another fact about the num-
a p p r o x i m a t i o n s of Legendre, Gauss, and ber of prime numbers ~(x). In the pic-
Riemann up to 10 m i l l i o n (12) . Since these ture up to 10 million, G a u s s ' s approxima-
four functions lie so close together that tion was always bigger than ~(x). That
their graphs are i n d i s t i n g u i s h a b l e to the remains so until a billion, as you can
naked eye - as we already saw in the pic- see in the picture on the following page
ture up to 50,000 - I have plotted only (in w h i c h the above data are plotted log-
the differences b e t w e e n them: arithmically).
X(X)+3OO-
z(x}+200-
~(x)+l1~176 Legendre
V~ Riemann /
i,fV Vm
3~'(x]1 " R,emann , M -- Rlemann " V ~ / ~ ~ ' " - ~N
~(x)-1ool
x (m rmlilons)
13
200C verges, i.e. e v e n t u a l l y exceeds any pre-
v i o u s l y given number. His proof, which
is also very simple, used the function
100C
~(s) = I + i_ + I__ + ... ,
2s 3s
Lt(x)-R(x)
500 "~/// whose importance for the study of ~(x)
/ t
was fully r e c o g n i z e d only later, with
the work of Riemann. It is amusing to re-
m a r k that, a l t h o u g h the sum of the recip-
200 rocals of all primes is divergent, this
sum over all the known primes (let's say
the first 50 million) is smaller than
four(15).
10C I I i I I 1 J I
10 20 30 50 70 100 200 500 1000 The first major result in the direc-
x (in millions) tion of the prime number t h e o r e m was
proved by C h e b y s h e v in 1850(16). He
showed that for s u f f i c i e n t l y large x
S u r e l y this graph gives us the impres-
sion that with i n c r e a s i n g x the differ- x
x x < ~(x)
0.89 log < 1.11 log x
ence Li(x) - ~(x) grows steadily to in-
finity, that is, that the logarithmic in-
tegral Li(x) c o n s i s t e n t l y o v e r e s t i m a t e s i.e., the prime number t h e o r e m is cor-
the number of primes up to x (this w o u l d rect with a relative error of at most
agree with the o b s e r v a t i o n that R(x) is 11%. His proof uses binomial c o e f f i c i e n t s
a better a p p r o x i m a t i o n than Li(x), since and is so p r e t t y that I c a n n o t r e s i s t at
R(x) is always smaller than Li(x)). But least sketching a simplified v e r s i o n of
this is false: it can be shown that there it (with somewhat worse constants).
are points where the o s c i l l a t i o n s of In the one direction, we will prove
R(x) - ~(x) are so big that ~(x) a c t u a l -
ly becomes larger than Li(x). Up to now (x) < 1.7 ~
x .
no such numbers have been found and per-
haps none ever will be found, but Little-
wood proved that they exist and Skewes(14) This i n e q u a l i t y is valid for x < 12OO.
proved that there is one that is smaller Assume i n d u c t i v e l y that it has been
than proved for x < n and consider the m i d d l e
binomial c o e f f i c i e n t
10101034
n I
~(2n+I) < ~(2n) + I < 3.09 ~ o g n + I Pl = ~ + 1 4 . 1 3 4 7 2 5 i,
I
2n+I (n > 1200) P2 = ~ + 2 1 . 0 2 2 0 4 0 i,
< 1.7 l o g (2n+i")
I
P3 = ~ + 2 5 . 0 1 0 8 5 6 i,
it is a l s o v a l i d for 2n+I, c o m p l e t i n g the
induction. I
P4 = ~ + 3 0 . 4 2 4 8 7 8 i,
For t h e b o u n d in the o t h e r d i r e c t i o n ,
w e need a s i m p l e l e m m a w h i c h c a n b e I
P5 = ~ + 3 2 . 9 3 5 0 5 7 i,
p r o v e d e a s i l y u s i n g the w e l l - k n o w n for-
m u l a for the p o w e r of p w h i c h d i v i d e s n!
(17) :
Pl ~ 89 - 14.134725 i,
Lemma: Let p be a prime. If pVp is the lar-
I
P2 = ~ - 2 1 . 0 2 2 0 4 0 i,
gest power of p dividing (~), 1
then P3 = ~ - 2 5 . 0 1 0 8 5 6 i,
pVP < n.
54 = 89 - 3 0 . 4 2 4 8 7 8 i,
Corollary: Evergbinemial coefficient"(~)
55 = 89 - 3 2 . 9 3 5 0 5 7 i.
satisfies
0.3
0.2 i
o~ N
It follows that Rk(x) is also a smooth
function for each k. As k grows, these
= T1 (x)
~ 0 functions approach ~(x). Here, for exam-
-0.1 ple, are the graphs of the 10th and 29th
approximations,
-0.2
-0.3
0.2
N
i/ ~ = T2 (X)
0.1
~
15--
O.
N
T3 (x)
-0"II
-0.2
0.2
0.1
0
-0.1
-0.2
~/ ~~o
N
Tz. (x)
I
50
I
100
X
..-
16
Xtx)
.............. R10(x)
R29(x)
R29(x)
I I
50 t00 I I X
50 100
I h o p e t h a t w i t h this a n d the o t h e r
p i c t u r e s I h a v e shown, I h a v e c o m m u n i c a t -
ed a c e r t a i n i m p r e s s i o n of the i m m e n s e
a n d if w e c o m p a r e t h e s e c u r v e s w i t h the ibeauty o f the p r i m e n u m b e r s and of the
g r a p h of ~(x) u p to 100 (p. 9) we g e t iendless s u r p r i s e s w h i c h t h e y h a v e in
the following picture: istore for us.
Remarks (3) T h e r e is a g o o d r e a s o n w h y so m a n y
of t h e n u m b e r s in t h i s l i s t h a v e
the f o r m M k = 2k - 1: A t h e o r e m of
(I) J.M. G a n d h i , F o r m u l a e for the n t h L u c a s s t a t e s that M k (k>2) is p r i m e
prime, Proc. W a s h i n g t o n S t a t e Univ. if a n d o n l y if M k d i v i d e s Lk-1,
Conf. o n N u m b e r Theory, W a s h i n g t o n w h e r e the n u m b e r s L n a r e d e f i n e d in-
S t a t e Univ., P u l l m a n , Wash., 1971 d u c t i v e l y b y L I = 4 a n d L n + I = L~ - 2
96 - 106 (so L 2 = 14, L 3 = 194, L 4 = 37634,
...) a n d h e n c e it is m u c h e a s i e r to
t e s t w h e t h e r M k is p r i m e t h a n it is
(2) J.P. Jones, D i o p h a n t i n e r e p r e s e n t a - to t e s t a n o t h e r n u m b e r of the same
t i o n of the s e t o f p r i m e n u m b e r s , order of magnitude.
N o t i c e s of t h e A M S 22 (1975) A - T h e p r i m e n u m b e r s o f the f o r m
326. 2K - I (for w h i c h k i t s e l f m u s t n e c -
17
e s s a r i l y be prime) are called Mer- (8) R a m a n u j a n has given the following
senne primes (after the French m a t h - a l t e r n a t i v e forms for this function:
ematlcian M e r s e n n e who in 1644 gave
a list 9 of such p r i m e s up to 10 79 , (lOg X) "t dt
c o r r e c t up to 1018 ) and play a role R(x) = 7 t r(t+i)~(t+l)
o
in connection w i t h a c o m p l e t e l y dif-
ferent p r o b l e m of number theory.
Euclid d i s c o v e r e d that when 2P - I (~(s) = the Riemann zeta function
is prime then the number 2P -I (2P-I) and F(s) = the gamma function) and
is "perfect", i.e. it equals the
sum of its proper d i v i s o r s (e.g.
6 = I + 2 + 3 , 2 8 = I + 2 + 4 + 7 +
R ( e 2 ~ X ) ' = ~2 (~
2 x + 3--~4x
4 3 + ~ x6+ . .
5 1
%4, 496 = I + 2 + 4 § 8 + 16 + 31 +
62 + 124 + 248) and Euler showed
that every even p e r f e c t number has
_ ~2 ( 12x 2525
+ 40x 3 + --~--x +... )
this form. It is u n k n o w n whether
there are any odd perfect numbers (Bk = k th Bernoulli number; the sym-
at all; if they exist, they must be bol ~ m e a n s that the d i f f e r e n c e of
at least 10 IQ0 . There are exactly the two sides tends to O as x grows
24 values of p < 20,000 for w h i c h to infinity). See G.H. Hardy, Rama-
2P - 1 is prime. nujan: Twelve Lectures on Subjects
S u g g e s t e d by His Life and Work, Cam-
bridge U n i v e r s i t y Press, 1940, Chap-
(4) C.F. Gauss, Werke II (1892) 444 - ter 2.
447. For a d i s c u s s i o n of the histo-
ry of the various a p p r o x i m a t i o n s to
~(x), in which an E n g l i s h transla- (9) Namely: The p r o b a b i l i t y that for a
tion of this letter also appears, r a n d o m l y chosen pair (m, n) of num-
see L.J. Goldstein: A history of bers both m and n are ~ 0 (mod p)
the prime number theorem, Amer. is o b v i o u s l y [ (p-1)/p] 2 , while for
Math. Monthly, 80 (1973) 599 - 615. a r a n d o m l y chosen number n, the.
p r o b a b i l i t y that n and n+2 are both
O (mod p) is I/2 for p = 2 and
(5) A.M. Legendre, E s s a i sur la theorie (p-2)/p for p ~ 2. Thus the proba-
de Nombres, 2nd edition, Paris, bility for n and n+2 m o d u l o p to be
1808, p. 394. prime twins differs by a factor
p-2 2
(6) M o r e precisely p
Ls(x) - 1.5 < Li(x) < Ls(x), for p r 2 and by 2 for p = 2 from
the c o r r e s p o n d i n g p r o b a b i l i t y for
i.e. the d i f f e r e n c e b e t w e e n Li(x) two i n d e p e n d e n t numbers m and n. Al-
and Ls(x) is bounded. We should also together, we have therefore improved
m e n t i o n that the l o g a r i t h m i c inte- our c h a n c e s by a factor
gral is often d e f i n e d as the C a u c h y
principal value
p2 _ 2p
C = 2 p>2 p2 _ 2p + I
Xdt = lim /i-~ dt
Li(x) = o§ log t p prime
+
I-5
s oTG~q ) = 1.32032. For a somewhat more
careful p r e s e n t a t i o n of this argu-
ment, see Hardy and Wright, An In-
t r o d u c t i o n to the Theory of Numbers,
but this d e f i n i t i o n d i f f e r s from C l a r e n d o n Press, Oxford, 1960,
that given in the text only by a w 22.20 (p. 371 - 373)
constant.
(10) M.F. Jones, M. Lal, and W.J. Blundon,
S t a t i s t i c s on certain large primes,
(7) The c o e f f i c i e n t s are formed as fol-
Math. Comp. 21 (1967) 103 - 107.
lows: the c o e f f i c i e n t of Li(n/x) is
+ 1/n if n is the p r o d u c t of an e v e n
number of d i s t i n c t primes, -I/n if (11) D. Shanks, On maximal gaps b e t w e e n
n is the p r o d u c t of an odd number successive primes, Math. Comp. 18
of d i s t i n c t primes, and 0 if n con- (1964) 646 - 651. The g r a p h of g(x)
tains m u l t i p l e prime factors. was m a d e from the tables found in
18
the f o l l o w i n g papers: L.J. L a n d e r w h e r e ~(x) § 0 as x t e n d s to infin-
and T.R. P a r k i n , On f i r s t a p p e a r - ity and C ~ 0 . 2 6 1 4 9 7 is a constant.
ance of p r i m e d i f f e r e n c e s , M a t h . T h i s e x p r e s s i o n is s m a l l e r than 3.3
Comp. 21 (1967) 483 - 488, R.P. w h e n x = 109 , and e v e n w h e n x = 1018
Brent, The f i r s t o c c u r r e n c e o f large it s t i l l lies b e l o w 4.
gaps b e t w e e n s u c c e s s i v e p r i m e s ,
Math. Comp. 27 (1973) 959 - 963.
(16) P.L. C h e b y s h e v , R e c h e r c h e s n o u v e l l e s
sur les n o m b r e s p r e m i e r s , Paris,
(12) The d a t a for this g r a p h are t a k e n 1851, CR P a r i s 29 (1849) 397 - 401,
f r o m L e h m e r ' s table of p r i m e n u m b e r s 738 - 739. For a m o d e r n p r e s e n t a t i o n
(D.N. L e h m e r , L i s t of P r i m e N u m b e r s (in German) of C h e b y s h e v ' s proof,
f r o m I to 1 0 , O O 6 , 7 2 1 , H a f n e r P u b - see W. Schwarz, E i n f U h r u n g in M e t h o -
l i s h i n g Co., N e w York, 1956). den und Ergebnisse der Primzahltheo-
rie B I - H o c h s c h u l t a s c h e n b u c h 278/278a,
(13) T h i s and the f o l l o w i n g g r a p h w e r e M a n n h e i m 1969, Chapt. II.4, P. 42 -
m a d e u s i n g t h e v a l u e s of ~(x) found 48.
in D.C. M a p e s , F a s t m e t h o d for com-
p u t i n g the n u m b e r of p r i m e s l e s s
than a g i v e n limit, Math. Comp. 17 (17) T h e l a r g e s t p o w e r of p d i v i d i n g p!
(1963) 179 - 185. In c o n t r a s t to is p [ n / P ] + [ n T p 2 ] + . . . , w h e r e [x] is
L e h m e r ' s d a t a u s e d in the p r e v i o u s the l a r g e s t i n t e g e r S x. Thus in the
graph, t h e s e v a l u e s w e r e c a l c u l a t e d n o t a t i o n of the l e m m a w e h a v e
from a f o r m u l a for ~(x) and n o t by
c o u n t i n g the p r i m e s u p to x.
Vp = r~17.( I V ] - [~ - [i)
(14) S. Skewes, O n the d i f f e r e n c e ~(x) - E v e r y s u m m a n d in this sum is e i t h e r
li(x) (I), J. L o n d o n Math. Soc, 8 O o r I and is c e r t a i n l y O for
(1933) 277 - 283. Skewes' p r o o f of r > (log n / l o g p) (since then [n/p r]
this b o u n d a s s u m e s the v a l i d i t y of = O). T h e r e f o r e 9p S (log n / l o g p),
the R i e m a n n h y p o t h e s i s w h i c h w e dis- f r o m w h i c h the c l ~ i m f o l l o w s .
cuss later. T w e n t y - t w o y e a r s l a t e r
(On the d i f f e r e n c e ~(x) - li(x) (II),
Proc. Lond. Math. Soc. (3) 5 (1955) (18) T h e d e f i n i t i o n of ~(s) as I + I/2 s
48 - 70) he p r o v e d w i t h o u t u s i n g the + I/3 s + ... g i v e n a b o v e m a k e s s e n s e
R i e m a n n h y p o t h e s i s that t h e r e e x i s t s o n l y w h e n s is a c o m p l e x n u m b e r
an x less t h a n the (yet m u c h larger) w h o s e r e a l p a r t is l a r g e r than I
bound
101010964 (since the series c o n v e r g e s for
t h e s e v a l u e s of s only) and in this
d o m a i n ~(s) has no zeroes. B u t the
for w h i c h ~(x) > Li(x). This b o u n d f u n c t i o n ~(s) can b e e x t e n d e d to a
has b e e n l o w e r e d to f u n c t i o n for all c o m p l e x n u m b e r s s,
1010529.7 so t h a t it m a k e s s e n s e to s p e a k of
its r o o t s in the w h o l e c o m p l e x
plane. The s i m p l e s t w a y to e x t e n d
by C o h e n a n d M a y h e w and to 1 . 6 5 x i O 1165 the d o m a i n of d e f i n i t i o n of ~(s) at
by L e h m a n (On the d i f f e r e n c e l e a s t to the h a l f , p l a n e Re(s) > O
~(x) - li(x), A c t a A r i t h m . 11 (1966) is to u s e the i d e n t i t y
397 - 410). L e h m a n e v e n s h o w e d t h a t
t h e r e is a n i n t e r v a l of at l e a s t (1-21-s)~(s) = I + 1 + I_~ + ...
10500 n u m b e r s b e t w e e n 1.53•
2s 3s
and 1 . 6 5 x i O I165 w h e r e ~(x) is larger
than Li(x). A s a c o n s e q u e n c e o f his
i n v e s t i g a t i o n , it a p p e a r s l i k e l y - 2 + 4-~ 6s ...
t h a t t h e r e is a n u m b e r n e a r 6 . 6 6 3
x10370 w i t h ~(x) > Li(x) a n d t h a t = (_I)n-1
there is no n u m b e r less t h a n 1020
n=l ns '
w i t h this p r o p e r t y .
w h i c h is v a l i d for Re(s) > I, and
(15) N a m e l y (as c o n j e c t u r e d by G a u s s in to o b s e r v e that the s e r i e s o n the
1796 and p r o v e d by M e r t e n s in 1874) r i g h t c o n v e r g e s for a l l s w i t h p o s -
i t i v e r e a l part. W i t h this, the "in-
t e r e s t i n g " r o o t s o f the zeta func-
7 1 = log log x + C + r tion, i.e. the r o o t s p = B + iy w i t h
p<x p O < 8 < I, can be c h a r a c t e r i z e d in
19
an elementary way by the two equa- The 1977 Mathematics Calendar has found a
tions worthy successor:
n~ I (-I)n-I c o s ( y l o g n) = O,
Psychology
n8
n~ 1 (-I)n-I sin(ylog n) = O.
n8
%/
P.S. No cause for concern. The next Mathematics
Calendar will appear in 1979.