Erd oss proof of Bertrands postulate

David Galvin

April 21, 2006

Abstract
In 1845 Bertrand postulated that there is always a prime between n and 2n, and he
veried this for n < 3, 000, 000. Tchebychev gave an analytic proof of the postulate
in 1850. In 1932, in his rst paper, Erd os gave a beautiful elementary proof using
nothing more than a feweasily veried facts about the middle binomial coefcient. We
describe Erd oss proof and make a few additional comments, including a discussion of
how the two main lemmas used in the proof very quickly give an approximate prime
number theorem. We also describe a result of Greeneld and Greeneld that links
Bertrands postulate to the statement that {1, . . . , 2n} can always be decomposed into
n pairs such that the sum of each pair is a prime.
1 Introduction
Write (x) for the number of primes less than or equal to x. The Prime Number Theorem
(PNT), rst proved by Hadamard [4] and de la Vall ee-Poussin [7] in 1896, is the statement
that
(x)
x
ln x
as x . (1)
A consequence of the PNT is that
> 0 n() > 0 : n > n() p prime, n < p (1 + )n. (2)
Indeed, by (1) we have
((1 + )n) (n)
(1 + )n
ln(1 + )n

n
ln n
as n .
Using a more rened version of the PNT with an error estimate, we may prove the following
theorem.
Theorem 1.1 For all n > 0 there is a prime p such that n < p 2n.

Dept of Mathematics, University of Pennsylvania, Philadelphia, PA 19104; [email protected].

1
This is Bertrands postulate, conjectured in the 1845, veried by Bertrand for all N <
3 000 000, and rst proved by Tchebychev in 1850. (See [5, p. 25] for a discussion of the
original references).
In his rst paper Erd os [2] gave a beautiful elementary proof of Bertrands postulate
which uses nothing more than some easily veried facts about the middle binomial coef-
cient
_
2n
n
_
. We describe this proof in Section 2 and present some comments, conjectures and
a consequence in Section 3. One consequence is the following lovely theorem of Greeneld
and Greeneld [3].
Theorem 1.2 For n > 0, the set {1, . . . , 2n} can be partitioned into pairs
{a
1
, b
1
}, . . . , {a
n
, b
n
}
such that for each 1 i n, a
i
+ b
i
is a prime.
Another is an approximate version of (1).
Theorem 1.3 There are constants c, C > 0 such that for all x
c ln x
x
(x)
C ln x
x
.
2 Erd oss proof
We consider the middle binomial coefcient
_
2n
n
_
= (2n)!/(n!)
2
. An easy lower bound is
_
2n
n
_

4
n
2n + 1
. (3)
Indeed,
_
2n
n
_
is the largest term in the 2n+1-term sum

2n
i=0
_
2n
n
_
= (1+1)
2n
= 4
n
. Erd oss
proof proceeds by showing that if there is no prime p with n < p 2n then we can put
an upper bound on
_
2n
n
_
that is smaller than 4
n
/(2n + 1) unless n is small. This veries
Bertrands postulate for all sufciently large n, and we deal with small n by hand.
For a prime p and an integer n we dene o
p
(n) to be the largest exponent of p that
divides n. Note that o
p
(ab) = o
p
(a) + o
p
(b) and o
p
(a/b) = o
p
(a) o
p
(n). The heart of the
whole proof is the following simple observation.
If
2
3
n < p n then o
p
__
2n
n
__
= 0 (i.e., p |
_
2n
n
_
). (4)
Indeed, for such a p
o
p
__
2n
n
__
= o
p
((2n)!) 2o
p
(n!) = 2 2.1 = 0.
So if n is such that there is no prime p with n < p 2n, then all of the prime factors of
_
2n
n
_
lie between 2 and 2n/3. We will show that each of these factors appears only to a small
exponent, forcing
_
2n
n
_
to be small. The following is the claim we need in this direction.
2
Claim 2.1 If p|
_
2n
n
_
then
p
o
p((
2n
n
))
2n.
Proof: Let r(p) be such that p
r(p)
2n < p
r(p)+1
. We have
o
p
__
2n
n
__
= o
p
((2n)!) 2o
p
(n!)
=
r(p)

i=1
_
2n
p
i
_
2
r(p)

i=1
_
n
p
i
_
=
r(p)

i=1
__
2n
p
i
_
2
_
n
p
i
__
r(p), (5)
and so
p
o
p((
2n
n
))
p
r(p)
2n.
In (5) we use the easily veried fact that for integers a and b, 0 [2a/b] 2[a/b] 1. 2
Before writing down the estimates that upper bound
_
2n
n
_
, we need one more simple
result.
Claim 2.2 n

pn
p 4
n
(where the product is over primes).
Proof: We proceed by induction on n. For small values of n, the claim is easily veried.
For larger even n, we have

pn
p =

pn1
p 4
n1
4
n
,
the equality following from the fact that n is even an so not a prime and the rst inequality
following from the inductive hypothesis. For larger odd n, say n = 2m + 1, we have

pn
p =

pm+1
p

m+2p2m+1
p
4
m+1
_
2m + 1
m
_
(6)
4
m+1
2
2m
(7)
= 4
2m+1
= 4
n
.
In (6) we use the induction hypothesis to bound

pm+1
p and we bound

m+2p2m+1
p
by observing that all primes between m + 2 and 2m + 1 divide
_
2m+1
m
_
. In (7) we bound
3
_
2m+1
m
_
2
2m
by noting that

2m+1
i=0
_
2m+1
i
_
= 2
2m+1
and
_
2m+1
m
_
=
_
2m+1
m+1
_
and so the
contribution to the sum from
_
2m+1
m
_
is at most 2
2m
. 2
We are now ready to prove Bertrands postulate. Let n be such that there is no prime p
with n < p 2n. Then we have
_
2n
n
_
(2n)

2n

2n<p2n/3
p (8)
(2n)

2n

p2n/3
p
(2n)

2n
4
2n/3
. (9)
The main point is (8). We have rst used the simple fact that
_
2n
n
_
has at most

2n prime
factors that are smaller than

2n, and, by Claim 2.1, each of these prime factors contributes

at most 2n to
_
2n
n
_
; this accounts for the factor (2n)

2n
. Next, we have used that by hy-
pothesis and by (4) all of the prime factors p of
_
2n
n
_
satisfy p 2n/3, and the fact that
each such p with p >

2n appears in
_
2n
n
_
with exponent 1 (this is again by Claim 2.1);
these two observations together account for the factor

2n<p2n/3
p. In (9) we have used
Claim 2.2.
Combining (9) with (3) we obtain the inequality
4
n
2n + 1
(2n)

2n
4
2n/3
. (10)
This inequality can hold only for small values of n. Indeed, for any > 0 the left-hand
side of (10) grows faster than (4 )
n
whereas the right-hand side grows more slowly than
(4
2/3
+ )
n
. We may check that in fact (10) fails for all n 468 (Maple calculation),
verifying Bertrands postulate for all n in this range. To verify Bertrands postulate for all
n < 468, it sufces to check that
2, 3, 5, 7, 13, 23, 43, 83, 163, 317, 631. (11)
is a sequence of primes, each term of which is less than twice the term preceding it; it
follows that every interval {n + 1, . . . , 2n} with n < 486 contains one of these 11 primes.
This concludes the proof of Theorem 1.1.
(If a Maple calculation is not satisfactory, it is easy to check that (10) reduces to n/3
log
2
(2n + 1) +

2nlog
2
2n. The left hand side of this inequality is increasing faster than
the right, and the inequality is easily seen to fail for n = 2
10
= 1024, so to complete the
proof in this case we need only add the prime 1259 to the list in (11)).
3 Comments, conjectures and consequences
A stronger result than (2) is known (due to Lou and Yao [6]):
> 0 n() > 0 : n > n() p prime, n < p n + n
1
2
+
1
22
+
.
4
The Riemann hypothesis would imply
> 0 n() > 0 : n > n() p prime, n < p n + n
1
2
+
.
There is a very strong conjecture of Cram er [1] that would imply
> 0 n
0
> 0 : n > n
0
p prime, n < p n + (1 + ) ln
2
n.
And here is a very lovely open question much in the spirit of Bertrands postulate.
Question 3.1 Is it true that for all n > 1, there is always a prime p with n
2
< p < (n+1)
2
?
As mentioned in the introduction, a consequence of Bertrands postulate is the appeal-
ing Theorem 1.2. We give the proof here.
Proof of Theorem 1.2: We proceed by induction on n. For n = 1 the result is trivial. For
n > 1, let p be a prime satisfying 2n < p 4n. Since 4n is not prime we have p = 2n+m
for 1 m < 2k. Pair 2n with m, 2n1 with m+1, and continue up to n+k with n+k
(this last a valid pair since m is odd). This deals with all of the numbers in {m, . . . , 2n};
the inductive hypothesis deals with {1, . . . , m 1} (again since m is odd). 2
Finally, we turn to the proof of Theorem 1.3. The upper bound will follow from Claim
2.2 while the lower bound will follow from Claim 2.1.
Proof of Theorem 1.3: For the lower bound on (x) choose n such that
_
2n
n
_
x <
_
2n + 2
n + 1
_
.
For sufciently large n we have ln
_
2n
n
_
> n (from (3)) and for all n we have
_
2n
n
_
/
_
2n+2
n+1
_

1/4 and so
(x) ln x
x


__
2n
n
__
ln
_
2n
n
_
_
2n+2
n+1
_
n
__
2n
n
__
4
_
2n
n
_ (12)
We lower bound the number of primes at most
_
2n
n
_
by counting those which divide
_
2n
n
_
.
By Claim 2.1 each such prime contributes at most 2n to
_
2n
n
_
and so
__
2n
n
__

_
2n
n
_
/2n.
Combining this with (12) we obtain (for sufciently large x)
(x)
x
8 ln x
.
For the upper bound we use Claim 2.2 to get (for x 4)
4
x

px
p

x
(x)(x/2)
and so
(x)
4x ln 2
log x
+ (x/2).
5
Repeating this procedure log
2
x times we reach (for sufciently large x)
(x)
8x ln 2
log x
+ (2)

9x ln 2
log x
.
2
References
[1] Cram er, H., On the Order of Magnitude of the Difference Between Consecutive Prime
Numbers, Acta Arith. 2 (1936) 2346. 1936.
[2] Erd os, P., Beweis eines Satzes von Tschebyschef, Acta Sci. Math. (Szeged) 5 (1930
1932), 194198.
[3] Greeneld, L. and Greeneld, S., Some problems of combinatorial number theory
related to Bertrands postulate, J. Integer Seq. 1 (1998), Article 98.1.2.
[4] Hadamard, J., Sur la distribution des z eros de la fonction (s) et ses cons equences
arithm etiques, Bull. Soc. math. France 24 (1896), 199220.
[5] Havil, J., Gamma: Exploring Eulers Constant, Princeton University Press (2003).
[6] Lou, S. and Yau, Q., A Chebyshevs Type of Prime Number Theorem in a Short
Interval (II), Hardy-Ramanujan J. 15 (1992), 133.
[7] Poussin, C. de la Vall ee, Recherces analytiques sur la th eorie des nombres premiers,
Ann. Soc. Sci. Bruxells (1897).
6