How Not to Prove Collatz

IDK Math?
youtube: @MathIguess
bluesky: @mathiguess.bsky.social

December 4, 2024
 Abstract

The aim of this paper is to fail to prove the Collatz conjecture, but in a way that might
show others more clearly why this conjecture is so difficult to tackle. Particularly
from the perspective of perhaps a "clever undergraduate" or a gifted layperson.
Contents

1 Introduction 3

2 An Obvious Path 4
2.1 Made-up Terminology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 Made-up Notation and the Set D . . . . . . . . . . . . . . . . . . . . . . 5
2.2.1 Constructing D0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2.2 Constructing Dn . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2.3 Constructing D . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

3 An Alternate Route 7
3.1 Disproving Other Loops . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.1.1 Odd Factors Looping Directly . . . . . . . . . . . . . . . . . . . . 7
3.2 Other Ways the Conjecture Could Fail . . . . . . . . . . . . . . . . . . . 7

4 Conclusion 9
4.1 Are We Done? Collatz THEOREM???... No. . . . . . . . . . . . . . . . . 9
4.2 Did We Waste Our Time? . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2
Chapter 1

Introduction

The Collatz conjecture, a.k.a. the "3n + 1 problem", is extremely well known and it
involves the function f : N → N, such that

n
if n is even


f (n) =  .
2

3n + 1 if n is odd



The conjecture is that the output of this function always eventually ends up in the
loop (..., 4, 2, 1, 4, 2, 1, ...) if we apply it iteratively to any given positive integer.

As a bright-eyed, bushy-tailed undergraduate, I believed that I was special. I had

been pretty good at mathematics thus far and was widely regarded as "smart". Look-
ing at this problem, I thought "hey, that looks doable. I should wrestle with it as
independently as possible and try to come up with an approach that has not been
done before!", which turned out to be an excellent idea. I learned plenty from my
work on the conjecture and had great fun doing so. In order to facilitate my "never-
done-before" approach, I made up some notation to help me navigate the Collatz
world. This brings us to the next chapter.
Note: It turns out everything I thought of was very obvious and it has all been done
before. I just did it independently, which is still fun.

3
Chapter 2

An Obvious Path

2.1 Made-up Terminology

Definition 2.1.1 The odd factor of a positive integer m is an odd number b such that
m = 2k b, where 2k is the full (greatest) power of 2 which divides m and k ∈ {0, 1, 2, ...}.

Definition 2.1.2 A given positive integer m is said to terminate if there exists some k ∈
{0, 1, 2, ...} such that f (k) (m) = 1, where f (k) (x) = f ( f ( f (... f (x)...))). In other words, this is
how we denote the composition of f with itself, k times, applied to x. (So the kth iteration, one
could say. A number x terminates if there exists a k such that the kth iteration of the Collatz
function applied to x outputs 1.)
The least such k ∈ {0, 1, 2, ...} is then called the termination number of x. We note that a
number terminates if and only if its termination number is finite.

Lemma 2.1.3 Let m ∈ {1, 2, 3, ...}. Then n is the termination number of m if and only if n + k
is the termination number of 2k m.

Proof Obvious.
Okay but seriously, suppose that the termination number of m is n. Then f (n) (m) = 1.
Then consider 2k m. This number is even, and would stay even so long as k > 0, so
applying f to it would only divide by 2 for at least k iterations. Hence, f (k) (2k m) = m.
This implies that f (n) ( f (k) (2k m)) = f (n) (m) = 1. In other words, f (n+k) (2k m) = 1.
Next, suppose that the termination number of 2k m is n + k. Then f (n+k) (2k m) = 1. As
noted before, so long as k > 0, the input of this function is even, so for at least k
iterations, we would only divide by 2, implying that f (n) (m) = 1, so the termination
number of m is n. ⊣

Theorem 2.1.4 A number m terminates if and only if its odd factor terminates.

Proof Let m = 2k b such that the odd factor of m is b. Assume that m terminates and
has termination number n + k. By the lemma above, we then have that the termination
number of b is n. Note that we can assume that the termination number of m is n + k
since any non-negative integer can be expressed as the sum of non-negative integers,

4
2.2. Made-up Notation and the Set D Chapter 2. An Obvious Path

especially since k = 0 is allowed (and n + k = 0 is also allowed).

Next, assume that b has a termination number n. By the lemma above, m has a
termination number n + k, as desired. ⊣

Definition 2.1.5 Let m ∈ N. If m is even, we say that f (m) = m2 is an even iteration of

the Collatz function of m. If m is odd, we say that f (m) = 3m + 1 is an odd iteration of the
Collatz function of m.
We will often omit the "of the Collatz function of input" part when using this terminology.

2.2 Made-up Notation and the Set D

The galaxy-brain move that I wanted to make is to flip the conjecture on its head
somehow. What if instead of showing that every natural number eventually outputs
1 under f , I try to show that we can generate all of N using some sort of inverse of f ?
That’s crazy! Who would think of that!?
[spoiler: Literally everyone]
My goal was to find D = {n ∈ N : n terminates}. Then I could hopefully show, perhaps
with induction, that the pre-image of f is such that f −1 (1) = D = N.

2.2.1 Constructing D0
We begin by considering numbers that clearly terminate. Numbers whose termination
number we can see by inspection. To this end, let

D0 = {n ∈ N : the odd factor of n is b and f (b) = 2k for some k ∈ {0, 1, 2, ...}}.

Thus, D0 consists of all the numbers that map to a power of 2 under f , so for n =
2 j b ∈ D0 , the termination number of b is k + 1, where k is the power of 2 to which b
maps under f . Thus, the termination number of n is j + k + 1. We can actually say
more about members of D0 and make the properties of its elements more precise. Let
m ∈ N be some odd number. Suppose that m ∈ D0 . Then, by definition of D0 ,

f (m) = 2k ,

which implies that

3m + 1 = 2k ,
from which we can deduce that

2k ≡ 1 (mod 3).

Suppose that k is odd. Then k = 2j + 1 for some j ∈ {0, 1, 2, ...}. Then

2k = 2 · 22j = 2 · 4 j .
j
From the definition of 2k , we have that 2k = 3m + 1, so 2 · 4 j = 3m + 1, giving 2·43−1 = m,
contradicting the fact that m is supposed to be an (odd) integer, since this expression

5
2.2. Made-up Notation and the Set D Chapter 2. An Obvious Path

always results in a non-integer for all choices of j (this can be shown via modulo
arithmetic, 2 · 4 j − 1 . 0( mod 3)).
Thus, k cannot be odd.
In other words, we have that m ∈ D0 implies that f (m) = 22j = 4 j for some j ∈ {0, 1, 2, ...}.
So we can refine our definition to be

D0 = {n ∈ N : the odd factor of n is b and f (b) = 4k for some k ∈ {0, 1, 2, ...}}.

This produces the set {n ∈ N : the odd factor of n is 1, 5, 21, 85, ...}. Additionally,
notice that f (b) = 4k is equivalent to 3b + 1 = 4k in the above definition of D0 , since b
k
is odd. This means that b = 4 3−1 .

2.2.2 Constructing Dn
Next, we construct D1 = {n ∈ N : the odd factor of n is b and f (b) ∈ D0 }. We think of
this set as the collection of numbers that are two odd iterations away from terminating,
since members of D0 are (at most) one odd iteration away from terminating.
We can reason as we did before to find out more properties of members of D1 .
Particularly, we can let m ∈ N and let the odd factor of m be b. Then we set
k
!
j 4 −1
3b + 1 = 2 ,
3
 k 
since members of D0 are all of the form n = 2 j 4 3−1 .
From this, we could reason as we did in the subsection above to discover more
properties of members of D1 . This is not necessary, however, since we can discard all
non-integers that are generated by the above expression. We therefore deduce that
!
2 j 4k − 1 1
 
D1 = n ∈ N : the odd factor of n is b and b = − .
3 3 3
We can continue this process to find D2 , D3 , ..., eventually obtaining that

Dk = {n ∈ N : the odd factor of n is b and f (b) ∈ Dk−1 }.

It will turn out that

 ak  ak−1   a1  a0
2 2 2 4 −1 1 1 1
     
Dk = n ∈ N : the odd factor of n is b and b = ... k − ... − − .
3 32 3 3 3 3 3

2.2.3 Constructing D
∞
We now let D =
S
Dk . This is the set of all positive integers that eventually terminate.
k=1
Thus, the conjecture is now equivalent to the claim that D = N. To see this, suppose
that D = N. Then all natural numbers eventually terminate, so the conjecture is true.
Suppose that the conjecture is true. Then we can use the Collatz function to find a
termination number of any given natural number, which allows us to find some Dk
to which the given number belongs. Hence, D = N.

6
Chapter 3

An Alternate Route

3.1 Disproving Other Loops

The conjecture says that (...4, 2, 1, 4, 2, 1, ...) is the only loop. What if we find some other
loop? Can we disprove the conjecture, or arrive at a contradiction or something?

3.1.1 Odd Factors Looping Directly

The odd factor of the known loop stays the same. That is, the loop is such that m = 2k ·1
where 1 is the odd factor of members of the loop. What if there is some other loop
with this property? Suppose b is the odd factor of all members of a loop. We would
have
3b + 1 = 2k b,
implying that 1 = (2k − 3)b, so
1
b= .
2k − 3
Now, b ∈ N, so we can see that the only option is k = 2, implying that b = 1. This
means that there is no loop other than the known loop with the property that the odd
factor of all members of the loop are the same.

3.2 Other Ways the Conjecture Could Fail

Earlier, we investigated 3b+1 = 2k b. Here, we’ll try to see whether the Collatz function
could have some unbounded output sequence. What if b1 < b2 where 3b1 + 1 = 2k b2 ?
We then have
3b1 + 1 < 3b2 + 1,
so
2k b2 < 3b2 + 1.
From This,
(2k − 3)b2 < 1.

7
3.2. Other Ways the Conjecture Could Fail Chapter 3. An Alternate Route

Now, b2 ∈ N, so the only way this could happen is if k < 2. It can be show that the
output of f for a given odd number b is always even, so k ≥ 1, allowing us to deduce
that k = 1. Thus, if the odd factor of the output of the Collatz function is larger than
the odd factor of its input, the step where the odd factor changed would result in
the output being twice the new odd factor. Unfortunately, this does not place any
restrictions on the odd factors themselves and it does not rule out an unbounded
sequence existing.
We could reason similarly to see that the odd factor shrinks when the "even factor" is
4 or more. Again, this unfortunately does not shed light on whether other loops are
possible, so we cannot rule them out from this.

8
Chapter 4

Conclusion

4.1 Are We Done? Collatz THEOREM???... No.

Intuition tells us that D = N is true. It seems clear. Take any number, find its odd
factor and then find the family of numbers generated by this odd factor to see which
Dk the number belongs to. Alternatively, acknowledge that D contains families of
recurrence relations (this can be proven using induction, yay!) an = 4an−1 + 1 and we
vary the starting value a0 over all of N, essentially. This should produce all of the
natural numbers, just ordered differently. The problem is that we used the Collatz
function to construct D. Our task when trying to show D = N becomes proving
(thanks to Theorem 2.1.4) that for all odd numbers m = 2n − 1, m ∈ N, there exists
some k ∈ {0, 1, 2, ...} such that
 ak  ak−1   a1  a0
2 2 2 4 −1 1 1 1
    
2n − 1 = b = ... k − ... − − ,
3 32 3 3 3 3 3
Where ai ∈ {0, 1, 2, ...} are some constants.
I want to emphasise this: It is not enough to produce a set that’s generated by some
rule that should produce all the natural numbers eventually. Our set must be such that
we can be given a natural number and then we explicitly show that this number is in
the set since the number meets a condition for being in the set. In this case, that means
checking whether the number terminates. Saying that D = N, so the conjecture is
true, is circular because of this.
Additionally, the claim that one has proven the conjecture is the same as the claim
that one has found a method for proving existence (and uniqueness) of a solution of
the above equation, for any given n ∈ N.
This sheds light on why this conjecture is so difficult. This is called a Diophantine
equation, which is a type of equation that commonly arises in conjectures. This is where
we have more unknown variables than equations, but restrictions on the variables
can sometimes make it possible to find solutions anyway (some of these even have
unique solutions, as we saw in the previous chapter). It is less common for there to
be so many variables, though, and our goal is simply to show that they exist. The
problem is that we don’t know how many we’ll need to express m as a member of D.

9
4.2. Did We Waste Our Time? Chapter 4. Conclusion

In fact, if there is some counter-example to the conjecture, that number would never
terminate and it shouldn’t belong to D. We intuitively deduce that there are either
no counter-examples, or infinitely many (since any number that is k odd iterations
away from terminating has a family of numbers given by the recurrence relation we
described earlier, that are the same number of odd iterations away from terminating),
but this is again a mistake. There is no guarantee that the counter-example, if it exists,
plays by the same rules as numbers that do terminate.

4.2 Did We Waste Our Time?

No. The journey with the above (which I did in far greater detail as an undergraduate)
was invaluable to me as a mathematician. I was able to independently arrive at and
apply several concepts that I later learned about in courses such as Discrete Structures
and Algebra. I also had several opportunities to practise applying skills they would
teach us in those courses, making me a more experienced student when it came to
problems such as proving explicit formulas for recurrence relations (using induction),
or reasoning based on modulo arithmetic. It also gave me a feel for some number
theory concepts that would be helpful later on. Perhaps most importantly, it showed
me that intuition can be unreliable. Intuitively, the problem should be easy. The set
D should equal the set N.

10