Bertrand’s Postulate

Michael Tang

October 2, 2015

1 Introduction
Bertrand’s Postulate is a theorem in number theory on the existence of prime numbers:

Theorem 1 (Bertrand’s Postulate). For all positive integers n ≥ 2, there is a prime p such
that n < p < 2n.

In this article, we follow the proof of the theorem that uses the beautiful, elementary
method found by Paul Erdős1 , and reflect on the implications of the ideas brought forth.

2 Prerequisites
For a clear reading, the reader should also be familiar with the following:

• Binomial coefficients,

• Factorials, vp (x) notation, and Legendre’s formula2

• Product (Capital pi) notation,

• Proof by induction.

3 Proof
3.1 Main idea
 
2n
The idea of the proof is the following: we consider the (central) binomial coefficient ,
n
and prove that it must have a prime factor n < p < 2n in order to be large enough; that
is, the central binomial coefficients are such that they cannot consist solely of powers of
“small” prime
  numbers. Therefore, first, we analyze the prime factors and prime powers
2n
dividing , and second, we give bounds that force the existence of a prime p in the
n
range n < p < 2n.
1
Erdős (1913-1996) discovered this proof in 1932, at the age of nineteen!
2
Legendre’s formula states that the highest power of a prime p that divides n! is
∞        
X n n n n
vp (n!) = i
= + 2 + 3 + ...
i=1
p p p p

1
3.2 Prime factors
For the sake of contradiction, suppose that Theorem 1 is false: that for some n ≥ 2, there
is no prime p in the range n < p < 2n. Consider the prime factors of the integer
 
2n (2n)!
= .
n (n!)2
 
2n
This integer has no prime factors greater than 2n, since if a prime divides , it must
n
  (2n)!. Also, since n ≥ 2, the integer 2n is not a prime, so all of
also divide its numerator,
2n
the prime factors of are strictly less than 2n. By our assumption, there are no primes
n  
2n
n < p < 2n, so all prime factors of are now at most n.
n

In fact, we can increase this restriction even further. It may be helpful to consider the
following example, taken from the (first ever!) AIME contest:
 
200
Problem (AIME, 1983). Find the greatest two-digit prime factor of .
100
 
200
Solution. We try the largest two-digit prime, 97. Sadly, 97 is not a factor of =
100
200!
, because the numerator has two factors of 97 (one each from 97 and 2 · 97 = 194),
(100!)2
as does the denominator (from two 97s). The same reasoning holds for the next prime, 89,
as well as the next primes 83, 79, . . . However, when we get to the prime 61, we note that
200! has three factors of 61(from 61, 2 · 61, and 3 · 61), and (100!)2 has two factors of 61.
200
Therefore, 61 divides , so it is our answer.
100
Using the same methods as in this problem, we establish the following result.
 
2n
Lemma 2 (No primes greater than 2n/3). The integer has no prime factors p in
n
2n
the range < p ≤ n.
3
2n
Proof. If a prime lies in the range < p ≤ n, then the numerator (2n)! has two factors
3
of p (from p and2p), 2
 as does the denominator (n!) (from two factors of p). Therefore, p
2n
does not divide .
n

 
2n 2n
Hence, all of the prime factors of are at most . We now try to bound the sizes
n 3
of these powers of “smaller” primes:
 
k 2n
Lemma 3 (Prime power bounding). If a prime power p divides , then pk ≤ 2n.
n

2
Proof. It suffices to prove that k, the power of p, is at most logp (2n). For this, we apply
Legendre’s formula:    
2n (2n)!
vp = vp
n (n!)2
= vp ((2n)!) − 2vp (n!)
∞   ∞  
X 2n X n
= i
−2
p pi
i=1 i=1
∞    
X 2n n
= − 2 .
pi pi
i=1

Consider the summand here. If we let x = n/pi , then the summand is b2xc − 2bxc. This
quantity is either 0 or 1, depending on the fractional part of x: if {x} ≥ 1/2, then it equals
1, and otherwise it equals 0.

In addition, when x < 1/2, both terms equal zero, so the summand equals zero. Now,
x < 1/2 when n/pi < 1/2, or pi > 2n, which gives i < logp (2n). Therefore, the only terms
of the sum that can be nonzero (i.e. equal to 1) are when i ≤ logp (2n); Hence,
∞    
X 2n n
−2 i ≤ logp (2n),
pi p
i=1

which is exactly what we wanted to prove.

In addition, this result (k ≤ logp (2n)) gives the following:

√
Lemma  4 (Larger
 primes only occur once). If a prime p is
  larger than 2n, then it is a
2n 2n
factor of at most once. That is, p2 does not divide .
n n
 
2 2n
Proof. This is a direct result of the previous Lemma: if p divides , then by Lemma 3,
√ n
p2 ≤ 2n, so p ≤ 2n.

This improves
  our result about prime powers: Lemma 3 gives us that all prime powers
2n
dividing are at most 2n. However, Lemma 4 gives that prime powers of a prime
√ n
p > 2n are at most p, which is in turn at most 2n/3.
 
2n
Let us present the “big picture” now. The binomial coefficient has all prime
n
2n
factors at most . Furthermore, each prime power that divides it is at most 2n, and each
3 √
prime factor greater than 2n (which is a relatively small number compared to n and  2n)
2n
can only occur once. Having developed these strict bounds on the prime factors of ,
n
we seek now to derive some contradictory inequality on n (or an inequality true only for
very certain, small values of n).

3
3.3 Bounding
 
2n
We begin with a lower bound on :
n
4n
 
2n
Lemma 5 (Binomial coefficient lower bound). For all positive integers n, ≤ .
2n n
Proof. For this proof, we view
4n = 22n = (1 + 1)2n
and use the Binomial Theorem:
     
n 2n 2n 2n
4 =1+ + + ... + + 1.
1 2 2n − 1

Adding the two 1sseparately,

 we obtain exactly
 2n terms, one of which is 2 and the others
2n 2n
being of the form , for some r. Since is the largest combination in its row (and
r n
is always at least 2), we have  
n 2n
4 ≤ 2n
n
which is what we wanted to prove.

Now we may employ our results from  the

 first half of the proof. For a prime p, let f (p)
2n
be the largest power of p that divides . Then
n
 
2n Y
= f (p).
n
1≤p≤2n/3
√
We know that f (p) ≤ 2n for all p, but for p > 2n, we have an even better bound:
f (p) ≤ p1 = p. Therefore, we write
  
Y Y Y
f (p) =  f (p)  f (p) .
√ √
1≤p≤2n/3 1≤p≤ 2n 2n<p≤2n/3
√ √
There areY
at most 2n primes
√ from 1 to 2n, inclusive, and each prime power is at most
2n
√
2n, so f (p) ≤ (2n) . We have f (p) ≤ p for p > 2n as well, which gives the
√
1≤p≤ 2n
inequality
  
Y Y √ Y
2n
 f (p)  f (p) ≤ (2n) · p.
√ √ √
1≤p≤ 2n 2n<p≤2n/3 2n<p≤2n/3

At this point, we introduce a new function to give a bound on the remaining product.

4
Definition 6. For a positive real number x, the primorial3 of x, written x#, is the product
of all primes less than or equal to x. (If x < 2, x# is the empty product, which by convention
equals 1.)

Y (2n/3)#
Then p = √
. We bound the primorial with the following lemma:
√ 2n #
2n<p≤2n/3

Lemma 7 (Primorial bound). For all x ≥ 1, the primorial x# ≤ 4x .

Proof. First note that it suffices to prove the lemma when x is an integer, since if x is not
an integer, then
x# = (bxc) # ≤ 4bxc ≤ 4x .
Also, isolate the base cases x = 1, 2, 3, which are easy to check by hand.
 
2x − 1 (2x − 1)!
The critical idea is as follows. Note that = is divisible by every
x (x − 1)!x!
prime p in the range x + 1 ≤ p ≤ 2x − 1, since the numerator is divisible by p but not the
denominator. Since (2x − 1)#/x# is the product of all primes x + 1 ≤ p ≤ 2x − 1, we have
 
(2x − 1)# 2x − 1
≤ .
x# x
We also have that
     
2x−1 2x−1 2x − 1 2x − 1 2x − 1
2 = (1 + 1) ≥ + =2 ,
x−1 x x
 
2x − 1
so ≤ 22x−2 = 4x−1 . Thus,
x

(2x − 1)# ≤ 4x−1 · x#

for all x. This allows us to induct on the even and odd integers.

If x is odd, say x = 2y − 1, then

x# = (2y − 1)# ≤ 4y−1 · y# ≤ 4y−1 · 4y = 42y−1 ,

concluding the induction step. If x is even, then since x ≥ 4 is not prime, we have

x# = (x − 1)# ≤ 4x−1 ≤ 4x

by applying the previous inequality to the odd integer x − 1. This completes the induction.

 
Y (2n/3)# 2n
Returning to the problem, we have p= √
≤ # ≤ 42n/3 . Hence,
√ 2n # 3
2n<p≤2n/3

√ √
 
2n 2n
Y
2n
≤ (2n) · p ≤ (2n) · 42n/3
n √
2n<p≤2n/3

3
primorial, not primordial!

5
so we have the inequality
4n √
≤ (2n) 2n · 42n/3 .
2n
√
Rearranging, we have 4n/3 ≤ (2n)1+ 2n , or, taking logs,
n √
ln 4 ≤ (1 + 2n) ln(2n).
3
Rewriting, we get √
(1 + 2n) ln(2n) ln 4
0≤ − .
n 3
It turns out that (as a function over the real numbers), the right-hand side decreases steadily
for n ≥ 3, which can be seen by differentiating; n = 467.4 is the approximate location of
a root of the right-hand side, so for n ≥ 468, this inequality is false, and we have our
contradiction.

For n ≤ 467, it is easy to check manually. In fact, consider the sequence of primes

3, 5, 7, 13, 23, 43, 83, 163, 317, 631.

Each prime is less than twice the previous prime, so for n ≤ 467, there must be at least
one prime p in the range n < p < 2n. This gives a contradiction as well, so Theorem 1 is
proven.

4 Reflections
Here are a few discussion points about the content and nature of this proof.

• The function we ended up with is something approximately

√
n ln n ln n
= √
n n

(eliminating some small factors). The natural logarithm grows slower than any power
√
of n, so as n → ∞, ln n/ n → 0. Formally, by l’Hôpital’s rule,

ln n n−1
lim √ = lim = lim 2n−1/2 = 0.
n→∞ n n→∞ 1/2 · n−1/2 n→∞

This is why we expected the inequality to hold only for small n.

• Why is the theorem called a “postulate,” a term reserved for basic assumptions and
axioms only? The explanation is a historical relic: French mathematician Joseph
Bertrand (1822-1900), who worked in group theory, found that the statement of The-
orem 1 would be helpful to him in solving a group theory problem. He could not
prove the theorem conclusively, but it seemed incredibly obvious, so he took it as a
“postulate” for his purposes.

6
 • Note that some of the√bounds we placed were√incredibly weak. We wrote, for example,
“There are at most 2n primes from 1 to 2n, inclusive,” or, equivalently, that all
prime numbers are positive integers. For the lower bound proved in Lemma 5, we
used only the fact that the central binomial coefficient is the largest of the entries in
its row (and a similar idea to prove Lemma 7), which becomes very, very weak as n
grows. There are ways to improve the bounds, and as a result, decrease the number
467 for a smaller manual check (see [1] for an example), but the proofs are all the
same in spirit.

• Even though some bounds were incredibly weak, it is quite subtle what was really
necessary to establish a contradictory result. For instance, one might think, “why
is Lemma 2 necessary, if it √
only eliminates one-third of the prime factors possible?
Lemma 4 eliminates all but 2n prime factors, which is much smaller in comparison.
However, if we replace 2n/3 with n, we get the inequality

4n √
≤ (2n) 2n · 4n
2n
which is always true! So one has to be careful with the exact details of bounding.

• The proof really reveals how little we know about primes (how elusive prime numbers
can be). Theorem 1 seems incredibly simple, almost trivial - we know from intuition
that there ought to be not just one, but many primes between n and 2n - yet it takes
very detailed and precise methods to prove the theorem via elementary means. The
idea of considering the binomial coefficient on its own is quite groundbreaking, and
not by any means easy to discover. Other proofs of Theorem 1 exist (a notable one
by Ramanujan (1887-1920) in [3] is short but quite advanced), but Erdös’ proof, as
was so much of his other work, is truly a classic.

References
[1] Shigenori Tochiori Stronger proof of Bertrand-Chebyshev’s theorem, available at http:
//www.chart.co.jp/subject/sugaku/suken_tsushin/76/76-8.pdf

[2] Wikipedia Proof of Bertrand’s postulate, available at https://en.wikipedia.org/

wiki/Proof_of_Bertrand\%27s_postulate

[3] Jaban Meher, M. Ram Murty Ramanujan’s Proof of Bertrand’s Postulate, available at
http://www.mast.queensu.ca/~murty/Meher-Murty-Monthly.pdf