First off, this probably isn’t going to be an easy read for you.

I wasn’t able to put this proof

together on my own, and even then, we used stuff from the rough equivalent of Precalculus,
Calculus, and even maybe a dash of first year number theory to get the job done. That said, I
think the final proof is readable for someone at your level. The problem which you wanted me to
explain was this:

inf φ( n)
“Find
nϵ N n ( ) ”

At this point, this problem is likely totally unintelligible to you. We need to break it down a little
bit, and understand what it’s asking. Although our end result will be a simple unassuming
number, it is truly the journey that is more important. This simple equation tells us something
very interesting about φ (n), and more generally, the prime numbers.

You hopefully at least have heard of the “prime numbers”: numbers that are divisible only by
one and themselves. Prime numbers fall with a regular irregularity that almost infuriates
mathematicians: 2, 3, 5, 7, 11, 13, etc. There is no known way to efficiently find, say, the 56789th
prime without getting down to the basics and checking numbers. It’s worth noting at this point
(as it will be important later) that there are infinitely many primes, but for the sake of brevity, I’ll
omit that proof. If you want to see it, just ask.

There’s a simple way to extend this a little bit, and it turns out to be quite useful. Let’s pick two
numbers (for use in equations, let’s call them m and n). We call these numbers “relatively prime”
(or “coprime”) if there is no number but one that evenly divides both of them. For an example,
take the numbers 6 and 35. These two numbers are relatively prime because 6=2∙ 3 and 35=5 ∙7.
No matter which number you pick, at least one will divide unevenly, with a fraction left over. 6
and 14, on the other hand, are not relatively prime since they are both divisible evenly by 2.

With this definition, we can look at φ ( n ). This function is called “Euler’s Totient Function”
(pronounced “Oiler’s toy-she-ent function”) and is defined as the number of numbers less than n
that are relatively prime to n. It should be noted that 1 is generally counted as being coprime to
every other number. To get a feel for this, let’s calculate φ ( 12 ). To do this, we might write out a
list of numbers, 1 to 11.

1
 1 2 3 4 5 6 7 8 9 10 11

Note that 12=22 ∙3 . This means that any number not divisible by both 2 and 3 is coprime to 12.
Cross these ones off.

1 2 3 4 5 6 7 8 9 10 11

Now count up what’s left: four numbers (1, 5, 7, and 11). This means that φ ( 12 )=4. While this
method will always work, it’s a little tedious for big numbers. It’d be nice to have a more general
formula for φ. We can create this formula from the method above by considering the following
series of questions: what fraction of numbers are divisible by a given number p? Well, we’d

1
expect to see one such number every p numbers, so the fraction should be . Now we might ask
p
the opposite question: what fraction of numbers are not divisible by p? Clearly, this ought to be

1
(1− ).
p

Now let me ask a more interesting question. Take two numbers, p1 and p2. What fraction of
numbers are divisible by neither p1 or p2? Perhaps you can see that this number is simply

1 1
(1− )(1− ), the product of the two independent fractions. This is true in many cases, but
p1 p2
there is a catch: p1 and p2 must be relatively prime to each other. If they are not, then we will be
discounting some numbers twice, and we’ll get a result smaller than the actual one we want.

You may have noticed that to find φ ( 12 ), we first factored it into primes, then looked to see what
numbers were not divisible by any of its prime factors. This will always work, and better yet, it
can easily be extended to a formula!

If n= pa1 ∙ pa2 ∙ pa3 ∙ …∙ pak (In other words, we have all its prime factors)
1 2 3 k

k
1 1 1 1
(
φ ( n )=n ∙ 1−
p1 )(
∙ 1−
p2 ) (
∙ … ∙ 1−
pk )
=n ∙ ∏ 1−
i=1 p (
i
)

2
Don’t let the capital pi on the right intimidate you. It’s just a way to abbreviate multiplication so
that it’s easy to write formulas like this down. There’s a similar notation (using capital sigma) for
repeated addition which I’ll introduce a little later on. In the mean time, let’s recompute φ ( 12 )
with the new method:

Since 12=22 ∙3

( 12 ) ∙(1− 13 )=12∙ 12 ∙ 23 = 123 =4

φ ( 12 )=12 ∙ 1−

As you can see, we get the correct answer, which is good. Using the formula I gave above, we

φ ( n)
now have a feel for : it’s the fraction of numbers less than n relatively prime to n. Using the
n
formula above, we can express this mathematically as

φ ( n) k 1
n
=∏ 1−
i=1
( )
pi

φ (n) inf
So we’ve worked out a basic feel for what means. What about the rest? What does ( …)
n nϵ N
mean? This operation is called “the infinitum of a set”, and the idea behind it is quite simple.
First though, what is a set?

In the mathematical sense, a set is simply a collection of mathematical objects. There might be
finitely many objects (say S={1,2,3,4,5 }) or infinitely many objects, say all integers greater than
zero (this is a special set. It’s called the natural numbers, and denoted N={1,2,3,4,5 , …} ). You
may already have spotted N above. It’s written in as “nϵ N”. While this may seem cryptic, it’s
really quite simple: you read it as “for each n in the set N…”. In other words, this notation is
asking us to build a new set out of the old one: for each element n in the natural numbers, this

φ (n)
new set has an element . This new set is still infinite in length, but it’s very different in
n
substance. Instead of progressively larger integers, it seems we get a bunch of seemingly random
values that seem to be between zero and one. I say seem because we haven’t shown any of that
yet. It just looks that way.

3
Now we can talk about what inf S (The “infimum of set S”) means. The idea is simple: we want
the smallest element of the set. There is one catch though. If the set is infinite, there might not be

1 1 1 1 1
a smallest element! To see why, take the set S={1 , , , , , … , ,… }. No matter which
2 3 4 5 n
element we pick, we can find one that’s smaller, so no element can be the absolute smallest.
What then is the infimum of this set? As it turns out it’s zero. The idea is that we can extend the
idea to infinite sets by looking for the largest possible lower bound. To see why it’s zero in this
particular case, consider the following.

1
First, for no positive n is < 0, so zero is a lower bound for the set. Now we have to show that
n
it’s the best possible bound. Let’s imagine that we have some better, very tiny bound. Call it ℇ.

1 1
By our definitions, we know that for any positive integer n, E ≤ . This implies that n ≤ , which
n ℇ

1
is clearly untrue, since n can get as large as we like. Pick any ℇ, we can still find n so that n> .
ℇ
Thus, zero is the best lower bound and the infimum of the set.

Now we can understand what the problem is asking: It wants us to find the largest possible lower
bound for the fraction of numbers relatively prime to a given number. It’s pretty clear that this is
at “worst” zero, since both n and φ (n) are positive integers. So how to find this? One approach
would be to try to prove that this value is actually zero. We would do this by proving we can

φ ( n)
pick n such that is smaller than any given positive ℇ. In order to do this though, we’ll need
n
to develop a few other tools.

You’ve in all likelihood seen powers: stuff like 23=8 or 25=32. As you may know these are
easily extended to the idea of an exponential function like y=2x . A plot of this function looks
like the figure below.

4
This should be easy enough. I’m going to throw in one more minor complication: a constant (like
π) called e. This number shows up quite a bit in math and has some interesting and important
properties, but for the sake of this discussion, it’s enough to think of it as just a number, one
about equal to 2.71828… (like π, e is transcendental, so it goes on forever and isn’t a fraction) .

y=e x has some interesting properties all to itself, but it looks like a perfectly normal exponential
in all respects.

However, the important thing here isn’t actually e x . It’s the inverse of e x , called ln ⁡( x) (a French
abbreviation for “natural logarithm”). ln ⁡( x) has the following two defining properties:

ln ( e x ) =x

ln (x)⁡
e =x

A plot of y=ln ⁡(x ) looks like e x flipped over the line y=x.

5
There are a few important properties to note about ln ⁡(x). First, it has an infinite tail near zero.
Informally, we might write ln ( x ) →−∞ as x → 0. Second, because of the properties of
exponentials, ln ⁡(x) turns multiplication into addition. To (somewhat informally) see why,
consider the following:

e ln ⁡( a ∙ b)=a ∙ b=e ln ⁡( a) ∙ eln ⁡( b)=e ln (a )+ln ⁡(b)

So ln ( a ∙ b ) =ln ( a ) + ln ⁡(b).

In general, ln ( a1 ∙ a2 ∙ a3 ∙ … ∙ ak ) =ln ( a1 ) + ln ( a2 ) +…+ ln ⁡(ak )

Or, using Sigma/Pi notation that I mentioned earlier,

k k
ln (∏ ) ∑
i=1
ai =
i=1
ln ⁡( ¿ ai )¿

Again, the notation is just a way to abbreviate repeated multiplication and addition in formulas
like this. In any case, this turns out to be a very useful property of logarithms because sums tend
to be easier to work with than products. Again, like everything else in this letter, the ln ( x) many
other beautiful properties (at least one more of which I’ll use later on) but for now, if you want to
think of it as just a way to turn multiplication into addition, that will suffice. Let’s leave ln (x) for
a while and discuss something even more unusual.

Imagine that we have a series of numbers (one that goes on forever), all generated by some kind

1
of rule. For instance, imagine the nth term of the sequence is . One question we might ask is
2n
what happens when we add more and more terms, starting from n=1?

6
1 1
=
2 2

1 1 3
+ =
2 4 4

1 1 1 7
+ + =
2 4 8 8

This seems simple enough. Let me take this one step further by proposing a formula for the sum
after n terms.

1 1 1 1 2n−1
+ + +…+ n = n
2 4 8 2 2

While I haven’t provided a good justification for this formula, it is indeed true. Stare at it a while
and you may be able to see why, and where I got it from. A rigorous proof it is correct can be
made using induction, but I won’t do that here.

Now that we have a formula for the sum of n terms, we can ask our original question. What
happens to this sum when n get very large, even infinite? In mathematical notation, we would
phrase this question in terms of something called a limit, written as follows:

2n−1
lim
n→∞ ( )
2n

Limits are the building blocks of calculus, but this particular one turns out to be easy to
manipulate:

2n−1 2n 1 1 1
lim
n→∞ ( ) (
2 n
=lim
n→∞ 2
n
2 ) ( )
− n =lim 1− n =1−lim n
n→∞ 2 n →∞ 2

1
You may be able to see that as n gets very large, approaches zero rather quickly. For this
2n

1
reason, lim n
=0. Using this result, we discover that as n get very large, the sum of our series
n→∞ 2

approaches one! In other words, even though we have added together an infinite number of

7
 1
terms, we have gotten a finite answer. In math speak we would say that the series
2n
“converges”.

Of course, not all series add up to a finite value this way. One simple example would be to use 1
for every n . In other words, the first three terms add to 3, the first hundred to 100, and the first n
terms add up to n. Now adding an infinite number of terms gives us an infinite answer. The
series never approaches any finite value, so it “diverges”.

While in the previous two examples it was easy to tell if the series converged or diverged, it can
be very difficult to tell with some series. For this reason, there are a number of tests that are used
to check whether a series diverges or converges. One such test, which I’ll use later on, is the
comparison test. There’s actually a bit more to this test, but the part we’ll use says the following:
Say we have two series, a n and b n, both of which are positive for every n or negative for every n.
The comparison test says that

If |b n| ≥|an| for every n and ∑ an diverges

Then ∑ bn diverges.

This should seem a little intuitive. Basically, we have two lists of numbers, and we can match
them up so that the element from one list is always “bigger” than its corresponding element from
the other. We then know that the sum of the first list is “bigger” than the sum of the second. If
the second list adds up to be infinitely large, so must the first.

As with a lot of other stuff in this letter, while there is a lot more interesting stuff about infinite
series, this is all we need to know to formulate our proof. With this information, we can now
solve the problem. Let’s look at it again.

inf φ( n)
Find
nϵ N n( )
If we think the answer is the absolute minimum that makes sense (zero) we already have a
strategy to prove it: show that no matter what positive ℇ we pick, we can find an n so that

8
φ(n)
<ℇ. With this in mind, let’s try to find a way to make ns like this. From the formulas, we
n
can see that we want big n with small φ (n) (lots of factors, so not much is relatively prime). In
this spirit, let’s define

n k = p1 ∙ p2 ∙ p3 ∙ … ∙ p k

Where pi is the ith prime.

i.e. n k =2∙ 3 ∙5 ∙ … ∙ p k

φ ( n5 ) 16
For example, n5 =2∙ 3 ∙5 ∙ 7 ∙11=2310, but φ ( n 5 )=480, so = ≈.21. More generally, we
n5 77

φ ( nk )
can use our formula for φ (n) to write a formula for .
nk

k
φ ( nk ) 1
nk
=∏ 1−
i=1
( pi )
From this formula, it might be clear that we want look at what happens as k → ∞. If not, examine
the following:

φ ( nk ) k 1 1
k−1
1 1 φ ( nk−1 )
nk (i=1 pi)(
=∏ 1− = 1−
pk ) ( )(
∏
i=1
1−
pi
= 1−
)
pk n k−1

In other words,

φ ( nk ) φ ( n k−1 )
<
nk nk−1

φ ( nk )
So if gets close to zero, it will be as k → ∞. However, this corresponds to an infinite
nk
product, which we don’t have the tools to deal with. We do, however, have a way to turn
multiplication into addition: the ln ( x) (Note: I’m relaxing notation a little bit. To be perfectly
rigorous, I would need to use limits to express the following, but that would clutter stuff up).

9
 ∞ ∞
φ ( n∞ ) 1 1
ln ( ) ( ∏ ( ))
n∞
=ln
i=1
1−
pi i=1
( )
=∑ ln 1−
pi

Now we have an infinite series, but it’s less clear what we want it to do. The key is to remember
the infinite tail of ln (x) near zero. In other words, as x → 0+¿ ¿, ln x →−∞. Since we want

φ ( nk )
→0 as k → ∞, we want the infinite series on the right of the equation to diverge towards
+¿¿
nk
−∞ as we add up an infinite number of terms. Unfortunately, it is not immediately clear whether
this is actually the case. The series might converge to some finite value, which would mean we
have to go back to the drawing board. We need to compare this series to some other series to tell.

( p1 )<1, a quick look at the graph of ln x reveals that ln (1− p1 ) is always

Since pi >0, 1−
i i

negative. In order to use the comparison test, we need some function that will always give us
negative values for x <1, but that is still greater than or equal to ln x. The first function we might
try would be the simple line y=x −1, and in this case it turns out to work. Without calculus, it is
difficult for me to prove to you that ln x ≤ x−1, but this fact is quite apparent from a simple plot
of the two functions.

Using this result, we get that

1 1 −1
( )(
ln 1−
pi
≤ 1− −1=
pi )pi

So by the comparison test.

10
∞ ∞

∑ ln 1− p1 ≤−∑ 1p
( )
i=1 i i=1 i

∞
1
That ∑ diverges to + ∞ is a well known fact that was first proved by Euler in the 18th century.
i=1 pi

Because of this result, we can finish our proof. ∑ ln

i=1
(1− p1 ) must diverge to −∞, and as a
i

φ ( nk ) inf φ( n) =0.
consequence we can pick k so that
nk
is as close to zero as we want. Thus, ( )
nϵ N n
We have found the answer to our original question.

So why do I find this problem and my proof beautiful? It’s difficult to explain. It’s not exactly
like I can point to a waterfall or a flower to show you the beauty here. What makes this so pretty
are really the primes themselves. They are such a simple sequence, one that seems natural at first
glance, yet they’ve defied simple explanation and shaped mathematics since its beginning. One
of the first clear instances of math that we have is an ancient bone with the first few primes
carved out in simple notches. The primes are an unimaginably hard rough gem, and though even
the best modern mathematicians can’t yet cut it, we can still get a glimpse of how it might
sparkle through this simple proof.

∞
1
That ∑ diverges somehow implies that the primes are kind of close to each other in some
i=1 pi
sense. Too far apart, and like the powers of two, their reciprocals might add to some finite
number. Yet, what we’ve proved here using this fact says almost the opposite. If the primes were
too close to each other, all numbers would be relatively prime to a large number of numbers

φ (n)
smaller than themselves, and might never fall all the way to zero.
n

We’ve shown that the primes must be kind of far apart using the fact that they must be kind of
close together! It’s kind of like (to use the analogy of a famous mathematician) we’ve been
wandering around blindly in a dark mansion, and by accident we’ve hit a chord on a piano in the

11
corner. The notes themselves might be a bit discordant to a true pianist, but now we can at least
imagine an entire piece being played. That’s the beauty.

Sincerely,

-Will Oursler

12