Sunteți pe pagina 1din 38

CHARACTERISTICS OF COUNTER EXAMPLE LOOPS IN THE COLLATZ

CONJECTURE

by

Thomas W. Allen II

An Abstract
presented in partial fulfillment
of the requirements for the degree of
Master of Science
in the Department of Mathematics and Computer Science
University of Central Missouri

May, 2012
ABSTRACT

by

Thomas W. Allen II

This thesis was undertaken in order to obtain new results about the Collatz Conjecture:
that (4, 2, 1) is the only loop created by the Collatz function. We look at some past results
and generalize one of them. We discover a few general characteristics about loops. After
odd elements of a loop are discussed, a general formula for them is discovered. Using this
formula, the Collatz Conjecture is restated in terms of a family of linear Diophantine systems
which have unique solutions. In particular, we show that (4, 2, 1) is the only loop created
by the Collatz function if and only if every solution set to each of these systems contains an
element x such that x ∈ R\N or x = 2k ∈ N for some k ∈ N.
CHARACTERISTICS OF COUNTER EXAMPLE LOOPS IN THE COLLATZ
CONJECTURE

by

Thomas W. Allen II

A Thesis
presented in partial fulfillment
of the requirements for the degree of
Master of Science
in the Department of Mathematics and Computer Science
University of Central Missouri

May, 2012
CHARACTERISTICSOF COUNTER EXAMPLE LOOPSIN THE COLLATZ
CONJECTURE

by

ThomasW. Allen II

APPROVED:

ittee Member

ommittee

ACCEPTED:

epartment of Mathematics

and Computer Science

Dean, Graduate School

UNIVERSITY OF CEhITRAL MISSOURI


WARRENSBURG.MISSOURI
ACKNOWLEDGMENTS

This thesis would not have been possible without the support of many people. I would like
to thank my supervisor, Dr. Dale Bachman, for the tutelage that was provided throughout
this process. Thanks are due to the members of my thesis committee, Dr. Nicholas Baeth
and Dr. David Ewing, for taking time out of their busy schedules and assisting with edits of
this thesis. I would also like to thank the Mathematics and Computer Science Department
at the University of Central Missouri. Without them taking a chance on me, this thesis
would have never been completed. Last but not least, I would like to express gratitude to
my lovely wife Laura. She is a strong reason I continue my pursuit of Mathematics further.
Contents

List of Tables vii

1 Origins 1
1.1 History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 What is the Collatz Conjecture? . . . . . . . . . . . . . . . . . . . . . 1
1.2 Previous Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.1 About the Collatz Conjecture: Stefan Andrei and Cristian Masalagiu,
1997 [AM97] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.2 On the minimum cycle lengths of the Collatz Sequences: Matti Sin-
isalo, 2003 [Sin03] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2 Results/Methods 8
2.1 Plan of Attack . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2 Elementary Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.3 Advanced Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

3 Future work 26

Bibliography 31

vi
List of Tables

1.1 Cycle Lengths given a Computational Bound R. . . . . . . . . . . . . . . . . . . 7

2.1 Matrix Sizes given a Computational Bound R . . . . . . . . . . . . . . . . . . . 25

vii
Chapter 1

Origins

1.1 History

1.1.1 What is the Collatz Conjecture?

The Collatz Conjecture (CC) is a classic conjecture in number theory that has plagued1
mathematicians for over seventy years. It is fairly simple to state yet its proof or disproof
has so far eluded mathematicians. The Collatz function f : N → N is defined by





 3x + 1 ; if x is odd

f (x) =


 x
; if x is even


2
and a Collatz Sequence (CS) is defined as {x, f (x), f (f (x)), . . . , f n (x) = 1} where x is some
starting value, n is a nonnegative integer, and f i (x) 6= f j (x) for all i, j such that 0 ≤ i < j ≤
n. Throughout this entire paper f will refer to the Collatz function and N will denote the set
of positive integers. A Collatz Sequence is a very specific type of sequence generated by the
1
“plagued” is used intentionally here. According to Joshua Cooper, “[the Collatz Problem] is a spectac-
ular way to waste days, weeks, or years of your life.”[Coo09]

1
CHAPTER 1. ORIGINS 2

Collatz function. At times, it will be necessary to discuss a Generalized Collatz Sequence.


A Generalized Collatz Sequence (GCS) is defined as a sequence of natural numbers {ak }k∈I
where I = {1, 2, . . . , n} or I = N satisfying ak+1 = f (ak ) for k ∈ I, k not a maximum
element of I.

Example 1.1.1. A few examples of Collatz Sequences are 3, 10, 5, 16, 8, 4, 2, f 7 (3) = 1 ,


21, 64, 32, 16, 8, 4, 2, f 7 (21) = 1 , and f 0 (1) = 1 . It is evident that if c is an element
 

of a CS then there exists a CS with c as the starting value. As an example, the CS


3, 10, 5, 16, 8, 4, 2, f 7 (3) = 1 also gives the CS 10, 5, 16, 8, 4, 2, f 6 (10) = 1 when we con-
 

sider 10 to be our starting value.

Example 1.1.2. Every Collatz Sequence is a Generalized Collatz Sequence so every sequence
seen in Example 1.1.1 is a Generalized Collatz Sequence. Other examples of GCS’s include
{3, 10, 5}, {21, 64, 32, 16, 8, 4, 2, 1, 4}, and {4, 2, 1, 4, 2, 1, 4, 2, 1, . . .}. These three GCS’s show
that GCS’s do not have to include 1, elements may repeat, and elements may repeat multiple
times. In fact, GCS’s may be infinite.

Now that we have discussed what Collatz Sequences and Generalized Collatz Sequences
are, we are ready for our first conjecture.

Conjecture 1.1.3 (Collatz Conjecture). If c ∈ N, then c is an element of a Collatz Sequence.

The conjecture originated with mathematician Lothar Collatz in 1937 when he stated
it during a lecture given at Syracuse University. Although the original problem was stated
differently, it has evolved into the more general form stated here.
Since this problem has been around for so long, it is only natural to question why it hasn’t
been solved yet. It’s not for a lack of trying. There have been numerous mathematicians
who have studied the Collatz Conjecture and numerous papers have been published on the
topic. Christopher Jones says it best when he states, “It is in a sense, a testament to the
CHAPTER 1. ORIGINS 3

beauty and power of mathematics that, for such a simple set of instructions, an intricate
and complex relationship between numbers can be formed.” [Jon98]
There is no clear practical reason for studying the Collatz Conjecture. However, that
doesn’t mean the conjecture shouldn’t be studied. An article written by Peter Rowlett
discusses how the study of abstract conjectures has led to practical applications in fields as
varied as video game design, pandemics, and nuclear power. [Row11]
Before we explore the Collatz Conjecture, a few definitions must be stated.

Definition 1.1.4. A natural number c is said to satisfy the Collatz Condition if f n (c) = 1
for some n ∈ N. The smallest positive n such that f n (c) = 1 is called the Collatz Length of
c. If there is no Collatz Sequence containing c, we say the Collatz Length of c is infinite.

Example 1.1.5. The Collatz Length of 3 is 7 because f 7 (3) = 1 and f i (3) 6= 1 for any
natural number i < 7. The Collatz Length of 1 is 3 because f 3 (1) = 1 and f i (1) 6= 1 for any
natural number i < 3. The Collatz Length of 8 is 3 because f 3 (8) = 1 and f i (8) 6= 1 for any
natural number i < 3. Note that two distinct positive integers may have the same Collatz
Length. We have already seen this with 8 and 1. Both of these integers have Collatz Length
3.
CHAPTER 1. ORIGINS 4

1.2 Previous Results

We summarize results from two papers that will be similar to results in our research. Subsec-
tion 1.2.1 has results dealing with natural numbers that satisfy the Collatz Condition as well
as their Collatz Lengths. One of these results is generalized in Theorem 2.3.9. Subsection
1.2.2 deals with cycles and cycle lengths, the definitions of which are similar to loops as
defined in Section 2.2.

1.2.1 About the Collatz Conjecture: Stefan Andrei and Cristian

Masalagiu, 1997 [AM97]

This paper gives some basic results on the Collatz Conjecture as well as a few formulas
for finding natural numbers that always satisfy the Collatz Condition. Each result will be
referenced with a prefix of AM for Andrei and Masalagiu.

AM 1. The function f is surjective, but not injective.

This is quite easy to verify.


2c
Proof. Let c ∈ N. Since c ∈ N, 2c ∈ N. Since 2c is even, f (2c) = = c. Therefore f is
2
surjective. Also, f (3) = 10 = f (20). Therefore f is not injective.

The next result characterizes a subset of the natural numbers that always satisfy the
Collatz Condition. It also gives the Collatz Length for each element of this subset.
22t+2 − 1
AM 2. If t ∈ N ∪ {0}, then satisfies the Collatz Condition with Collatz Length
3
2t + 3. In particular,
22t+2 − 1
 
(2t+3)
f = 1.
3
 2t+2 
(2t+3) 2 −1
Example 1.2.1. When t = 0, f = f 3 (1). Thus we see that 1 satisfies the
3
Collatz Condition with Collatz Length 3 as we saw in Example 1.1.5.
CHAPTER 1. ORIGINS 5

22t+2 − 1
 
(2t+3)
Example 1.2.2. When t = 1, f = f 5 (5). Therefore 5 satisfies the Collatz
3
Condition with Collatz Length 5.

Andrei and Masalagiu have another formula for characterizing a subset of the natural
numbers that always satisfy the Collatz Condition. This formula also gives the Collatz
Length of these numbers.
m
2m+1 (23 n + 1)
AM 3. If m, n ∈ N and n is congruent to 1 or 5 modulo 6, then − 1 satisfies
3m+1
the Collatz Condition with Collatz Length 3m n + 2m + 2. In particular,

m
2m+1 (23 n + 1)
 
(3m n+2m+2)
f − 1 = 1.
3m+1

Example 1.2.3. Let m = 1 and n = 7. Then

m
2m+1 (23 n + 1)
 
(3m n+2m+2)
f − 1 = f 25 (932, 067).
3m+1

Therefore 932, 067 satisfies the Collatz Condition with Collatz Length 25.

Example 1.2.4. Let m = 1 and n = 11. Then

m
2m+1 (23 n + 1)
 
(3m n+2m+2)
f − 1 = f 37 (3, 817, 748, 707).
3m+1

Therefore 3, 817, 748, 707 satisfies the Collatz Condition with Collatz Length 37.

Result AM 3 will be generalized in Chapter 2 with Theorem 2.3.9.


CHAPTER 1. ORIGINS 6

1.2.2 On the minimum cycle lengths of the Collatz Sequences:

Matti Sinisalo, 2003 [Sin03]

Definition 1.2.5. Let n ∈ N. A cycle (c1 , c2 , . . . , cn ) is a finite sequence of distinct natural


numbers such that for 1 ≤ i ≤ n − 1 f (ci ) = ci+1 and f (cn ) = c1 . The cycle (4, 2, 1) is called
the trivial cycle. The length of a cycle is the number of distinct elements in the cycle.

In this paper we use different terminology; see Definition 2.2.3.


The results in this section will be referenced with a prefix of S for Sinisalo.

S 1. There are no nontrivial cycles with length less than 275,000.

That is, if a cycle other than the trivial cycle exists, then it must be quite large. The
next result gives a bound for cycle lengths. One can manually check any c ∈ N by repeatedly
applying f . If 1 is eventually reached, then c satisfies the Collatz Condition. In fact, this
has been done for all natural numbers up to R where R = 1.25208 · 1018 and they were all
found to satisfy the Collatz Condition. We call R the current Computational Bound. Using
R, a minimum bound on cycle lengths can be obtained. By letting n be the number of even
elements in a given cycle and k be the number of odd elements in the same cycle we have
the following:

n
S 2. Let the Collatz conjecture be verified up to some bound R > 1. Let be the rational
1
k
ln 3 n ln 3 + R
number with least possible denominator k such that < ≤ . Then the least
ln 2 k ln 2
possible cycle length for a nontrivial cycle is n + k.

Sinisalo uses S 2 and the current computational bound of 1.25208 · 1018 to obtain a
minimum bound for the lengths of nontrivial cycles. The result includes a small table to
illustrate if the Collatz Conjecture is verified to a new computational bound R, then there
is a new minimum bound for lengths of nontrivial cycles. The particular values of R chosen
CHAPTER 1. ORIGINS 7

ln 3
here are related to specific rational approximations to in the following way. Suppose
ln 2
n ln 3 n ln 3 + R1
we have a rational approximation > .
The inequality in S 2 gives us ≤
k ln 2 k ln 2
1
which can be manipulated to R ≤ n/k .
Thus if it were shown that all numbers less
2 −3
1
than n/k satisfy the Collatz Condition, then the minimum cycle length is at least n + k.
2 −3
1
The following table gives the values of R = n/k and the corrsponding cycle lengths.
2 −3

S 3. The length of any nontrivial cycle in a Collatz Sequence is at least 1,027,712,276.

Table 1.1: Cycle Lengths given a Computational Bound R.

Computational Bound Cycle Length


3.80765 · 1023 2, 302, 268, 119, 908
5.10126 · 1022 355, 504, 839, 929
4.35849 · 1021 186, 265, 759, 595
2.16891 · 1020 17, 026, 679, 261
1.25208 · 1018 1, 027, 712, 276
Chapter 2

Results/Methods

2.1 Plan of Attack

In order to prove the Collatz Conjecture we must show that for every natural number c there
exists a natural number n such that f n (c) = 1. However, in order to disprove the Collatz
Conjecture we must find a natural number c such that f n (c) 6= 1 for all n ∈ N. Note that
this requires the existence of a natural number with infinite Collatz Length thus creating an
infinite subset of the natural numbers each of which does not satisfy the Collatz Condition.
Let us take a closer look at the Collatz Sequence {4, 2, 1}. This CS is a portion of
the Generalized Collatz Sequence {4, 2, 1, 4, 2, 1, . . . , 4, 2, 1, . . .}. This GCS will motivate our
definition of a loop. By assuming that the Collatz function produces another loop (a sequence
that repeats) that is distinct from the (4, 2, 1) loop, general characteristics about loops can
be found. These characteristics shall lead to restrictions in order for another loop to coexist
with the (4, 2, 1) loop.

8
CHAPTER 2. RESULTS/METHODS 9

2.2 Elementary Results

In order to consider more advanced topics associated with the Collatz Conjecture, a strong
understanding of basic results is required. Every result and proof in this section is fairly
straightforward, but are necessary for use in later sections. A few of these results are results
that can be seen in almost every research paper on the Collatz Conjecture, though they are
often left without proof. However, each of the results in this section pertaining to loops is
original.
We first consider the range of the function f . In particular, if c is an odd or even number,
what can we say about f (c)? Our first pair of results categorizes f (c) based on the parity of
c.

Lemma 2.2.1. If c is odd, then f (c) is even.

Proof. Let c ∈ N be odd. Then c = 2k + 1 for some integer k ≥ 0. Now f (c) = f (2k + 1) =
3(2k + 1) + 1 = 2(3k + 2), and thus f (c) is even.

Lemma 2.2.2. The value of f (c) is odd if and only if c ≡ 2 mod 4.

Proof. Let c = 4k + i for some integer k ≥ 0 and i ∈ {0, 1, 2, 3} . There are three cases to
check. If i is 1 or 3, then c is odd and thus f (c) = 3c + 1 is even by Lemma 2.2.1. If i
is 0, then c is even and thus f (c) = 2k which is even. If i is 2, then c is even and thus
f (c) = 2k + 1 which is odd.

As promised in Section 2.1, we now define a loop.

Definition 2.2.3. Let n ∈ N. A loop (c1 , c2 , . . . , cn ) is a finite sequence of distinct natural


numbers such that for 1 ≤ i ≤ n − 1 f (ci ) = ci+1 and f (cn ) = c1 . The loop (4, 2, 1) is
called the trivial loop. The order of a loop is the number of distinct elements in the loop
CHAPTER 2. RESULTS/METHODS 10

and is denoted by |L| = n. Note that loops A and B are equal if and only if A is a cyclic
permutation of B.

These definitions are identical to the definitions of cycles and cycle lengths for Sinisalo.
A survey of literature reveals that both the words cycle and loop are used for this concept.
We have chosen to use loop throughout this paper. Since we have chosen to use loop instead
of cycle, we feel the word order is a reasonable substitute for cycle length.

Notation 2.2.4. An arbitrary loop shall be designated with the letter L and a loop other
than (4, 2, 1) with the letter C.

Since a loop is a finite sequence of distinct numbers it is only natural to discuss the order
of a loop.

Lemma 2.2.5. If L is a loop, then |L| ≥ 3.

Proof. Suppose L = (c). By definition of a loop, f (c) = c. We now consider two cases. If c
c
is even, then f (c) = = c, contradicting that c ∈ N. If c is odd, then f (c) = 3c + 1 which
2
is even, contradicting f (c) = c. Therefore |L| ≥ 2. Now suppose L = (c1 , c2 ). Since L is a
loop, f (c1 ) = c2 and f (c2 ) = c1 . This leaves three cases to check.

1. Assume c1 and c2 are both odd. This isn’t possible since f (c1 ) is even and c2 is odd.

c1 c2
2. Assume c1 and c2 are both even. Since f (c1 ) = = c2 and f (c2 ) = = c1 , we get
2 2
c1 = 2c2 and c2 = 2c1 . These two equations imply c1 = 0 = c2 which is a contradiction.

3. Assume now that either c1 or c2 is odd. Without loss of generality we may assume
c2
c1 is odd and c2 is even. Therefore f (c1 ) = 3c1 + 1 = c2 and f (c2 ) = = c1 . Thus,
2
3c1 + 1 = 2c1 which implies c1 = −1, contradicting c1 ∈ N.

Therefore |L| =
6 2 and thus |L| ≥ 3.
CHAPTER 2. RESULTS/METHODS 11

Since |L| ≥ 3, (4, 2, 1) is one of the smallest loops possible. Later we will show that
(4, 2, 1) is the only loop of order 3 thus making it the smallest loop.

Definition 2.2.6. The set of odd elements in a loop L is denoted by O(L) whereas the set
of even elements in a loop L is denoted by E(L). Note that |E(L)| + |O(L)| = |L|.

Lemma 2.2.7. If L is a loop, then |O(L)| =


6 0 and |E(L)| =
6 0.

Proof. Let L = (c1 , c2 , . . . , cn ) be a loop of order n. By Lemma 2.2.5, n ≥ 3. Assume


|E(L)| = 0 and choose i < n. Since ci is odd, f (ci ) is even. This contradicts our assumption,
ci
so |E(L)| 6= 0. Now assume ci is even for all i ≤ n. If each ci is even, then = ci+1 or
2
ci = 2ci+1 for all i, 1 < i < n. By induction we have c1 = 2n−1 cn , so c1 = 2n c1 . This implies
c1 = 0, contradicting c1 ∈ N. Therefore |O(L)| =
6 0.

Lemma 2.2.8. Let L be a loop with |L| = n.

n
1. If n is even, then |O(L)| ≤ .
2
n−1
2. If n is odd, then |O(L)| ≤ .
2

Proof. By Lemma 2.2.7, |O(L)| 6= 0 and |E(L)| 6= 0. If c is odd, then f (c) is even. That
n
is, for every ci ∈ O(L) we have f (ci ) ∈ E(L). Therefore 2|O(L)| ≤ n, so |O(L)| ≤ . Now
2
n n−1
assume n is odd. In this case ∈ / N, and since |O(L)| ∈ N, we have |O(L)| ≤ .
2 2

This lemma leads to an obvious corollary whose proof we omit.

Corollary 2.2.9. Let L be a loop with |L| = n.

n
1. If n is even, then |E(L)| ≥ .
2
n+1
2. If n is odd, then |E(L)| ≥ .
2
CHAPTER 2. RESULTS/METHODS 12

Lemma 2.2.10. Let C be a loop with |C| = n. Then |O(C)| ≥ 2 if and only if the loop is
not (4, 2, 1).

Proof. (⇐) Let C be a loop that is not (4, 2, 1). Suppose |O(C)| = 1. Without loss of
ck−1
generality we may assume c1 is the odd element of C. We note that c2 = 3c1 +1 and ck =
  2
3c1 + 1 1 3c1 + 1 3c1 + 1
for 2 < k ≤ n since ck−1 is even. If ck−1 = k−3 , then ck = = k−2 . Thus
2 2 2k−3 2
3c1 + 1 3c1 + 1
by induction ck = for 2 ≤ k ≤ n. Now f (cn ) = c1 and cn is even, so = c1 .
2k−2 2n−1
1
Therefore 3c1 +1 = c1 (2n−1 ) or c1 = n−1 . Because c1 ∈ N, c1 = 1. This is a contradiction
2 −3
and thus |O(C)| ≥ 2.
(⇒) If |O(C)| ≥ 2, then C 6= (4, 2, 1) .

Lemma 2.2.11. If C is a loop other than (4, 2, 1), then |C| ≥ 4.

Proof. Let C be a loop other than (4, 2, 1) with order n. By Lemma 2.2.10, |O(C)| ≥ 2.
Now using the results from Lemma 2.2.8, |C| ≥ 4.

Continuing our examination of the elements of loops, the next two definitions are natural
ones.

Definition 2.2.12. Let L = (c1 , c2 , . . . , cn ) be a loop. The maximum element in L, denoted


by M, is the element such that ci ≤ M for all i. The minimum element in L, denoted by m,
is the element such that ci ≥ m for all i.

The first few results for maximum and minimum elements are trivial results. They are
listed below because they will be used extensively throughout the rest of the paper.

Lemma 2.2.13. If m is the minimum element of a loop L, then m is odd.

m
Proof. If m is even, then f (m) = < m, contradicting the minimumity of m. Therefore m
2
is odd.
CHAPTER 2. RESULTS/METHODS 13

Lemma 2.2.14. If M is the maximum element of a loop L, then 4|M .

Proof. If M is odd, then f (M ) = 3M + 1. Since M < 3M + 1, we get a contradiction that


M is the maximum element of L. Therefore M is even. Suppose M is not divisible by 4.
Since M is even and not divisible by 4, M = 2k for some odd integer k ≥ 1. Therefore
2k
f (M ) = = k and thus f 2 (M ) = 3k + 1. Since k ≥ 1, we know 2k < 3k + 1 and thus
2
M < 3k + 1 which contradicts the maximumity of M . Therefore 4|M .

n
Corollary 2.2.15. Let L = (c1 , c2 , . . . , cn ) be a loop with |L| = n. Then |E(L)| > . In
2
particular, |E(L)| > |O(L)|.

Proof. Due to Lemma 2.2.1, for all c ∈ O(L) there exists f (c) ∈ E(L). We know L has a
M M
maximum element M such that f (M ) = . However, by Lemma 2.2.14, ∈ E(L). That
2 2
M n
is, there is no c ∈ O(L) such that f (c) = . Therefore |E(L)| > .
2 2

Lemma 2.2.16. Let L = (c1 , c2 , . . . , cn ) and let m be the minimum element of L. Then
4|(3m + 1) if and only if L = (4, 2, 1).

Proof. Assume L = (4, 2, 1). It is clear that m = 1 and 4|(3m+1). Now assume L 6= (4, 2, 1).
Since the minimum element m is odd, f (m) = 3m + 1 is even by Lemma 2.2.1. Let’s assume
3m + 1
4|(3m + 1). Then f 2 (3m + 1) = . In order to maintain the minimumity of m,
4
3m + 1
m≤ , or m ≤ 1. This implies that m = 1 which contradicts L 6= (4, 2, 1). Therefore
4
4 does not divide (3m + 1).

Lemma 2.2.14 and Lemma 2.2.16 lead to an obvious Corollary whose proof we omit.

Corollary 2.2.17. Let L = (c1 , c2 , . . . , cn ) be a loop with maximum element M and minimum
element m. Then M = 3m + 1 if and only if L = (4, 2, 1).

We now discuss the congruence classes of M and m in order to discover more about loop
structure.
CHAPTER 2. RESULTS/METHODS 14

Lemma 2.2.18. If M is the maximum element of a loop L, then M ≡ 1 mod 3.

Proof. Since M is even and maximum, M = f (x) for some odd natural number x. That is,
f (x) = 3x + 1 and thus M ≡ 1 mod 3.

Lemma 2.2.19. Let C be a loop with minimum element m. Then m ≡ 3 mod 4 if and only
if C 6= (4, 2, 1).

Proof. (⇐) Let C 6= (4, 2, 1). Then m = 4x + a for some integer x ≥ 0 and a ∈ {0, 1, 2, 3}.
3m + 1
Since m is odd, a ∈
/ {0, 2}. Suppose a = 1. Then m = 4x+1. We know is an element
2
3m + 1 3(4x + 1) + 1
of C because |C| ≥ 4 and 3m + 1 is even. Therefore = which implies
2 2
3m + 1 12x + 4 3m + 1 3m + 1
= or = 6x + 2. It is clear 6x + 2 is even. However, is odd
2 2 2 2
because 3m + 1 is not divisible by 4 due to Lemma 2.2.16. Therefore a 6= 1 and thus m ≡ 3
mod 4. (⇒) Let C be a loop and suppose m ≡ 3 mod 4. Then clearly C 6= (4, 2, 1).

Definition 2.2.20. If c is an element of a loop L and c has a preimage y ∈ / L, then the tail
of c is Tc = t ∈ N | ∃ n ∈ N with f n (t) = c and f k (t) ∈

/ L ∀ k, 0 ≤ k < n . In this case,
c is called a root of the loop L or the root of the tail Tc .

As an example, {5, 8, 16} ⊂ T4 , which is the tail of the root 4 in the loop (4, 2, 1). This
example shows that the (4, 2, 1) loop has a tail, but what about loops other than (4, 2, 1)?

Lemma 2.2.21. If L is a loop, then L has a tail.

Proof. Let L be a loop with maximum element M . We see that 2M is a preimage of M


because f (2M ) = M . Since M is maximum, 2M ∈
/ L. Therefore 2M ∈ TM .

Lemma 2.2.22. If c is a root of the loop L, then c has two preimages opposite in parity.

Proof. Since c is a root, there exists a ∈ Tc such that f (a) = b. Likewise, there exists b ∈ L
such that f (b) = c. We can see that f (a) = f (b) = c but a 6= b. Therefore by definition of
the function f , a and b are opposite in parity.
CHAPTER 2. RESULTS/METHODS 15

2.3 Advanced Results

Theorem 2.3.1. A loop L has a root c if and only if 2|c and c ≡ 1 mod 3.

Proof. (⇒) Since c is a root, c has two preimages opposite in parity. In particular, f (2j+1) =
c for some integer j ≥ 0. Therefore c = f (2j + 1) = 6j + 4 ≡ 1 mod 3. It is also clear that
2|c.
(⇐) Let c ∈ L such that 2|c and c ≡ 1 mod 3. Clearly 2c, which is even, is a preimage of c.
Since 2|c, c = 2k for some k ∈ N. Also, c ≡ 1 mod 3 so 2k = 3x + 1 for some x ∈ N. This
2k − 1
implies x = which is odd. In particular, x 6= 2c and thus c has a preimage not in L.
3
Therefore c is a root of L.

Let c and d be roots of disjoint loops. Is it possible for Tc ∩ Td 6= ∅? Is it possible if c


and d are roots of the same loop? These questions will be addressed in Lemma 2.3.2 and
Corollary 2.3.3.

Lemma 2.3.2. If X and Y are disjoint loops, then their tails do not intersect.

Proof. Since X and Y are disjoint loops, X ∩ Y = ∅. Assume X and Y have tails that share
an element b. Then f n (b) ∈ X and f m (b) ∈ Y for some n, m ∈ N. Without loss of generality
assume m ≥ n. Since f n (b) ∈ X implies f n+i (b) ∈ X for all i ∈ N, we see f m (b) ∈ X ∩ Y .
This contradicts X and Y being disjoint and thus their tails do not intersect.

Corollary 2.3.3. If a loop L has more than one tail, then the tails do not intersect.

Proof. Let L be a loop with two distinct tails having roots x and y. Assume Tx ∩ Ty 6= ∅
and let c ∈ Tx ∩ Ty . Since c ∈ Tx , f n (c) = x for some n ∈ N and f k (c) ∈
/ L for all k < n.
Likewise, since c ∈ Ty , f m (c) = y for some m ∈ N and f j (c) ∈
/ L for all j < m. Without
loss of generality we may assume n ≤ m. If n = m, then x = f n (c) = y, and the tails are
CHAPTER 2. RESULTS/METHODS 16

not distinct. Therefore n < m and since f m (c) = y and f j (c) ∈


/ L for all j < m we have
x = f n (c) ∈
/ L, which contradicts x being a root. Thus Tx ∩ Ty = ∅.

This is an interesting find. Every element x of Tc has a nonempty preimage f −1 (x) ⊆ Tc .


Thus the nonempty set f −j (x) = a ∈ N|f j (a) = x ⊆ Tc for all j ≥ 1. Note that if i 6= j,


then f −i (x) ∩ f −j (x) = ∅, so Tc has an infinite number of elements. That is, if a loop other
than (4, 2, 1) exists, then there exist at least two infinite disjoint subsets of N defined by the
function f .

Definition 2.3.4. The set of roots in a loop L is denoted by R(L).

Theorem 2.3.5. Let L be a loop with |L| = n and |O(L)| = k. Then |O(L)| ≤ |R(L)| <
|E(L)|.

Proof. Since |O(L)| = k, |E(L)| = n − k. Because a root is even, |R(L)| ≤ |E(L)|. Let
x ∈ O(L). Then f (x) = 3x + 1 is even and f (x) ≡ 1 mod 3. Therefore f (x) is a root,
and thus |O(L)| ≤ |R(L)|. Therefore |O(L)| ≤ |R(L)| ≤ |E(L)|. Since L has a maximum
M M
element M ∈ E(L) and M ≡ 1 mod 3, M is a root. Also, 4|M so ∈ L. However, is
2 2
M M
even and ≡ 2 mod 3. Thus, is not a root. Therefore there exist even elements in L
2 2
that are not roots and so |R(L)| < |E(L)|.

Theorem 2.3.5 leads to three immediate corollaries. The proofs of Corollary 2.3.6 and
Corollary 2.3.7 are omitted.

Corollary 2.3.6. Let L be a loop. Then |R(L)| < |E(L)|.

Corollary 2.3.7. Let L = (4, 2, 1). Then |R(L)| = 1.


CHAPTER 2. RESULTS/METHODS 17

Corollary 2.3.8. A loop C has multiple roots if and only if C 6= (4, 2, 1).

Proof. Let C be a loop with multiple roots. Then C 6= (4, 2, 1) because (4, 2, 1) has only
one root. Now assume C 6= (4, 2, 1). By Lemma 2.2.10, |O(C)| ≥ 2. From Theorem 2.3.5,
|R(C)| ≥ 2.

Assuming a loop other than (4, 2, 1) exists implies that there are at least three tails of
infinite length that all lie within the natural numbers yet never intersect at all.
Now let us take a second to recall the paper by Andrei and Masalagiu. They discovered
different classes of natural numbers that always satisfy the Collatz Condition. We derive a
generalization of AM 3 here.

/ {1, 2, 4}, and 3Z |P for


Theorem 2.3.9. If P − 1 satisfies the Collatz Condition, P − 1 ∈
2Z P
some Z ∈ N, then Z − 1 satisfies the Collatz Condition. Also, if the Collatz Length for
3
2Z P
P − 1 is k, then the Collatz Length for Z − 1 is k + 2Z.
3

/ {1, 2, 4}, and 3Z |P for some Z ∈ N.


Proof. Let P − 1 satisfy the Collatz Condition, P − 1 ∈
2Z P
Since 3Z |P , we see Z − 1 is a natural number and thus is in the domain of f . Since
3
2Z P
− 1 is odd,
3Z

2Z P
   Z 
2 P
f −1 = 3 −1 +1
3Z 3Z
2Z P
= Z−1 − 3 + 1
3
2Z P
= Z−1 − 2.
3

2Z P 2Z−1 P
 
This number is even, so applying f yields f − 2 = −1. The number obtained
3Z−1 3Z−1
after applying f twice is another number that appears in the same form as the original
CHAPTER 2. RESULTS/METHODS 18

number. Since 3Z |P , we see 3Z−1 |P . In general,

2Z P 2Z−j P
 
2j
f − 1 = −1
3Z 3Z−j

for 0 ≤ j ≤ Z.
In particular,

2Z P 2Z−Z P
 
2Z
f − 1 = −1
3Z 3Z−Z
20 P
= −1
30
= P − 1.

2Z P
Since P − 1 satisfies the Collatz Condition, Z − 1 satisfies the Collatz Condition.
 Z 3  Z 
n 2 P n 2 P
We also need to verify that f − 1 6= 1 for 0 ≤ n < 2Z. Assume f −1 =
3Z  3 Z

2Z P
  Z 
n n+i 2 P
1 for some n, 0 ≤ n < 2Z. Since f Z
− 1 = 1, f − 1 ∈ {1, 2, 4} for all
 Z 3 3Z
2 P
i ∈ N. We have just seen f 2Z − 1 = P − 1 which contradicts P − 1 ∈/ {1, 2, 4}. Now
3Z
2Z P
since k is the Collatz Length for P − 1, Z − 1 has Collatz Length k + 2Z.
3

The loop (4, 2, 1) can be rewritten as (1, 4, 2). Disregarding the exact numbers in the loop
and considering only the parity of each element, the loop can be written as (odd, even, even).
By Definition 2.2.3, a loop L with order n has the characteristic f n (c) = c. There-
fore the loop (1, 4, 2) written as a Generalized Collatz Sequence is {1, 4, 2, 1, 4, 2, . . .} or
{odd, even, even, odd, even, even, . . .}. Does any other loop follow a pattern like this?

Definition 2.3.10. A loop L with |L| = n is said to be parity periodic if the elements of L fol-
 
low the pattern of d1 , e1 , . . . , ek , d2 , ek+1 , . . . , e2k , . . . , d k+1
n , e kn
−(k−1) , e kn
−(k−2) , . . . , e kn
k+1 k+1 k+1

where di ∈ O(L), ei ∈ E(L), for all i ∈ {1, . . . , k}.


CHAPTER 2. RESULTS/METHODS 19

Theorem 2.3.11. A loop L is parity periodic if and only if L = (4, 2, 1).

Proof. (⇐) Let L = (4, 2, 1). Clearly (4, 2, 1) = (1, 4, 2) is parity periodic.
(⇒) Let L be a loop other than (4, 2, 1) that is parity periodic with |L| = n. It is impossible

for L = (d1 , d2 , . . . , dn ) by Lemma 2.2.1. It is impossible for L = d1 , e1 , d2 , e2 , . . . , d n2 , e n2
by Lemma 2.2.14. Now assume
 
n , e kn
L = d1 , e1 , . . . , ek , d2 , ek+1 , . . . , e2k , . . . , d k+1 −(k−1) , e kn
−(k−2) . . . , e kn with k ≥ 2.
k+1 k+1 k+1

Without loss of generality, suppose d1 is the minimum element of L. Therefore f 2 (d1 ) =


3d1 + 1
f (e1 ) = e2 , which is even since k ≥ 2. Thus = e2 = 2l for some integer l, so
2
4|(3d1 + 1) which contradicts Lemma 2.2.16. Thus a loop L is parity periodic if and only if
L = (4, 2, 1).

Definition 2.3.12. Let O(L) = {d0 , d1 , . . . , dk−1 } and write



L = d0 , e0,1 , . . . , e0,j0 , d1 , e1,1 , . . . , e1,j1 , . . . , dk−1 , ek−1,1 , . . . , ek−1,jk−1 where el,i ∈ E(L) for
all l ∈ {0, . . . , k − 1} and all i ∈ {1, . . . , jl }. Then for each l ∈ {0, . . . , k − 1}, define a
subsequence of L to be (dl , el,1 , . . . , el,jl ). The order of a subsequence is the number of
elements in the subsequence. If (dl , el,1 , . . . , el,jl ) is a subsequence of L, we denote the pair
[dl , jl ] as the index of the subsequence. Note that for l ∈ {0, . . . , k − 1} f jl (3dl + 1) =
3dl + 1
d(l+1 mod k) or = d(l+1 mod k) . We say L is written as a juxtaposition of indices if
2jl
L = [d0 , j0 ] [d1 , j1 ] . . . [dk−1 , jk−1 ] where d0 is the minimum element of L.

By Theorem 2.3.11, (4, 2, 1) is the only parity periodic loop. That is, if C 6= (4, 2, 1),
then there exist two indices of C [di , ji ] and [dk , jk ] such that ji 6= jk . Also, it should be
noted that j0 = 1 whenever L 6= (4, 2, 1) due to Lemma 2.2.16.
CHAPTER 2. RESULTS/METHODS 20

Theorem 2.3.13. Let L be a loop. Then if da ∈ O(L) with 0 ≤ a ≤ k − 1, then

P k−1  P i−1 
3k−1−i · 2 l=0 j(l+a mod k)
i=0
da = P k−1 .
2 i=0 ji − 3k

Proof. Let L be a loop and write L = [d0 , j0 ] [d1 , j1 ] . . . [dk−1 , jk−1 ] with dk = d0 . Since
3di + 1
di+1 = for each i ∈ {0, . . . , k − 1}, we see
2ji

3d0 + 1
d1 =
2j0

and

3d1 + 1 32 d0 + 3 1
d2 = = + .
2j1 2j0 +j1 2j1

If we assume

3i d0 + 3i−1 3i−2 1
di = j +···+j
+ j +···+j
+ · · · + j
20 i−1 21 i−1 2 i−1

for some i, then

3di + 1
di+1 =
2ji
3 d0 + 3 i
i+1
3i−1 1
= j +···+j
+ j +···+j
+ · · · + .
20 i 21 i 2ji
CHAPTER 2. RESULTS/METHODS 21

Thus
3k d0 + 3k−1 3k−2 1
d0 = dk = j +···+j
+ j +···+j
+ ··· + j .
2 0 k−1 2 1 k−1 2 k−1

Finding a common denominator and simplifying,

P k−1 X  P i−1 
i=0 ji k−1
d0 · 2 = d0 · 3k + i=0 3k−1−i · 2 l=0 jl

or P k−1  P i−1 
i=0 3k−1−i · 2 l=0 jl

d0 = P k−1 .
2 i=0 ji − 3k

Since d0 was arbitrary, any di could have been chosen to play the role of d0 . That is, for
any integer a ∈ {0, . . . , k − 1}

P k−1  P i−1 
3k−1−i · 2 l=0 j(l+a mod k)
i=0
da = P k−1 .
2 i=0 ji − 3k

We can now set up a system of linear Diophantine equations. For notational ease, let
P k−1
B=2 i=0 ji − 3k be the denominator of the expression in Theorem 2.3.13. If

X  P i−1 
k−1
i=0 3k−1−i · 2 l=0 j(l+p mod k) =B

for some p ∈ {0, . . . , k − 1}, then dp = 1. Therefore L = (4, 2, 1). If

X  P i−1 
k−1
i=0 3k−1−i · 2 l=0 j(l+p mod k) <B

for some p ∈ {0, . . . , k − 1}, then dp < 1. Therefore

X  P i−1 
k−1
i=0 3k−1−i · 2 l=0 j(l+a mod k)
CHAPTER 2. RESULTS/METHODS 22

is a positive multiple of B for every p ∈ {0, . . . , k − 1}.


3di + 1
Alternatively we can use di+1 = for i = 0, . . . , k − 1 to create the following matrix
2ji
equation.

    
j0
 −3 2 0 ··· ··· 0   d0   1 
 ..    
 0
 −3 2j1 0 ··· .  d
 1
 
  1  
 .. ... ... ..   .. .. 
    
 . 0 −3 .  . . 
 
 
  = .
 .. .. .. ..   ..   .. 
 .
 . 0 . . 0   .
  
  . 

 .    
 0
 · · · .. · · · −3 2jk−2 
  dk−2
  
  1  
    
2jk−1 0 · · · ··· 0 −3 dk−1 1

For notational ease, let


     
j0
 −3 2 0 ··· ··· 0   d0   1 
 ..    
 0
 −3 2j1 0 ··· .

 d
 1



 1  
 .. .. .. ..  .. .. 
     
 . 0 −3 . . .  . . 
  
 , d~ =   , and ~1 = 
  
A=
 .. ..  ..
.
.. 
.. ..   
 .
 . 0 . . 0  
 .




 . 

 .     
 0
 · · · .. · · · −3 2jk−2 

 d
 k−2



 1  
     
2jk−1 0 · · · ··· 0 −3 dk−1 1

By using cofactor expansion on the first column of A,

Pk−1
det(A) = −3((−3)k−1 ) + (−1)k−1 (2 i=0 ji
)
Pk−1
= (−1)k−1 (2 i=0 ji
− 3k )

which is not 0. Therefore A−1 exists and the system has the unique solution d~ = A−1~1. It is
also evident that det(A) = ±B.
How do we solve a linear Diophantine system?
CHAPTER 2. RESULTS/METHODS 23

Theorem 2.3.14. [GP90] To solve the system of Linear Diophantine equations AX = B,


unimodular row reduce [At |I] to [R|T ], where R is in row-echelon form. Then the system
AX = B has integer solutions if and only if the system Rt K = B has integer solutions for
K, and all the solutions of AX = B are of the form X = T t K.

Note: Row-echelon form allows the pivots of the matrix to be any integer, not necessarily
1. Also, an elementary unimodular row operation on a matrix consists one of the following
three types of operations.

1. Add an integer multiple of one row of the matrix to another row.

2. Interchange two rows of the matrix.

3. Multiple one row of the matrix by −1.

Solving Ad~ = ~1 would show whether every di , with i ∈ {0, . . . , k − 1}, is an odd natural
number or not. If there exists some di ∈
/ N or di 6= 2j + 1 for some integer j ≥ 0, then
(4, 2, 1) is the only loop possible by Theorem 2.3.13.
With the aid of Mathematica, a few elementary examples were explored.

Example 2.3.15. Let L be a loop with |O(L)| = 1.


This leads to the 1 × 1 matrix
A = [2j0 − 3].

It is evident that j0 = 2 and thus L = (4, 2, 1). This is not surprising since we already know
(4, 2, 1) is the only loop such that |O(L)| = 1.
CHAPTER 2. RESULTS/METHODS 24

Example 2.3.16. Let L be a loop with |O(L)| = 2.


This leads to the 2 × 2 matrix
 
 −3 2 
A= .
j1
2 −3

~
As you can see, j0 = 1 because L 6= (4, 2, 1). Now solving for d,

5
   
 d0   −9 + 2j1 +1 
  =  3 + 2 j1  .
d1
−9 + 2j1 +1

5
Since each entry must be an odd positive integer, is an odd positive integer which
−9 + 2j1 +1
is impossible. Therefore |O(L)| 6= 2. In particular, there are no loops with exactly two odd
elements.

Example 2.3.17. Let L be a loop with |O(L)| = 3.


This leads to the 3 × 3 matrix
 
 −3 2 0 
 
A= j1 .
 0 −3 2 
 
2j2 0 −3

~
As before, j0 = 1. Now solving for d,

15 + 2j1 +1
   

 d0   −27 + 2j1 +j2 +1 


  9 + 32j1 + 2j1 +j2
 
 d = .

 1   −27 + 2j1 +j2 +1
  
9 + 52j2
  
d2
−27 + 2j1 +j2 +1
CHAPTER 2. RESULTS/METHODS 25

It is evident that |E(L)| = j0 + j1 + j2 , but is |E(L)| ∈ N? Using the result S 2, |O(L)| = 3,


18
and the current computational bound of R = 1.25208·10 we can see that |E(L)| ∈
/ N. That
1

ln 3 j0 + j1 + j2 ln 3 + R
is: the inequality < ≤ does not hold if |E(L)| ∈ N. Therefore
ln 2 3 ln 2
|O(L)| =6 3. In particular, there are no loops with exactly three odd elements.

A similar analysis can be done for larger sized matrices. Since |E(L)| + |O(L)| = |L| and
|E(L)| ln (3 + R1 )
≤ (cf S 2) we obtain the inequality
|O(L)| ln 2

|L| ln 2
|O(L)| ≥ .
ln ( 6R+2
R
)

We know from S 3 that |L| ≥ 1, 027, 712, 276 and R = 1.25208 · 1018 . Therefore |O(L)| ≥
397, 573, 379. That is, the smallest potential matrix A has size 397, 573, 379 × 397, 573, 379.
Table 1.1 can be used to generate the size of a potential matrix once a new computational
bound R is obtained. A new table is listed which gives the sizes of such matrices.

Table 2.1: Matrix Sizes given a Computational Bound R

Computational Bound Loop Order Matrix Size


3.80765 · 1023 2, 302, 268, 119, 908 890, 638, 885, 193 × 890, 638, 885, 193
5.10126 · 1022 355, 504, 839, 929 137, 528, 045, 312 × 137, 528, 045, 312
4.35849 · 1021 186, 265, 759, 595 72, 057, 431, 991 × 72, 057, 431, 991
2.16891 · 1020 17, 026, 679, 261 6, 586, 818, 670 × 6, 586, 818, 670
1.25208 · 1018 1, 027, 712, 276 397, 573, 379 × 397, 573, 379
Chapter 3

Future work

Some potential future explorations of the Collatz Conjecture include:

1. Find the unique solution of Ad~ = ~1 for every matrix A.

Knowing the value for every di is crucial in order to see if the Collatz function produces
more than one loop or not.

2. Discuss how tails can coexist in N without intersecting.

How can multiple tails exist in N? A loop C 6= (4, 2, 1) has |R(C)| ≥ 2 by Corollary
2.3.8. Therefore, if there exists a loop C 6= (4, 2, 1), then there are at least three tails
in N. In-depth research on tails could lead to a result about tails not being able to
coexist in N.

3. Extend the Collatz function to a more general function.

In order to prove the Collatz Conjecture it may be easier to look at a more general
function.

26
CHAPTER 3. FUTURE WORK 27

The General Collatz function fn (x) : N → N with integer n ≥ 0 is defined by





 (2n − 1)x + 1 ; if x is odd

fn (x) = .


 x
; if x is even


2

The first three General Collatz functions are:





 1 ; if x is odd

f0 (x) = ,

 x ; if x is even



2




 x + 1 ; if x is odd

f1 (x) = ,


 x
; if x is even


2
and





 3x + 1 ; if x is odd

f2 (x) = .


 x
; if x is even


2

The f2 function is the Collatz function f that has been discussed throughout this
paper. It is clear that the f0 function produces the loop (1) since f0 (1) = 1. The proof
that this is the only loop is quite simple and will be given later. It is also clear that
the f1 function produces the loop (2, 1) because f1 (2) = 1 and f1 (1) = 2. The proof
that this is the only loop is also quite simple and shall be given later.
CHAPTER 3. FUTURE WORK 28

Each of the functions in the family {fn : n ∈ N ∪ {0}} satisfies several of the properties
we have discussed earlier. These properties will be stated here without proof.

Theorem 3.0.1. Let fk ∈ {fn : n ∈ N ∪ {0}}. Suppose L is a loop generated by the


function fk with maximum element M and minimum element m. Then

a) If c ∈ O(L), then fk (c) ∈ E(L).

b) If c ∈ E(L), then either fk (c) ∈ E(L) or fk (c) ∈ O(L).

c) M ∈ E(L).

d) m ∈ O(L).

e) M ≡ 1 mod 2k − 1.

A few less elementary results on this family of functions follow.

Theorem 3.0.2. If g is a function in the {fn : n ∈ N ∪ {0}} family, then g has a loop.

Proof. Let g be a function in the {fn : n ∈ N ∪ {0}} family. Then g = fk for some
k ≤ n.

g(1) = (2k − 1)(1) + 1

= 2k − 1 + 1

= 2k .

Since g k (2k ) = 1, g has a loop.

As previously stated, the function f0 and the function f1 only have one loop. The
proofs of these follow but first we need a definition.
CHAPTER 3. FUTURE WORK 29

Definition 3.0.3. We say c ∈ N satifies the General Collatz Condition for the function
fk , if fkn (c) = 1 for some integer n ≥ 0.

Theorem 3.0.4. If c ∈ N, then c satisfies the General Collatz Condition for the
function f0 . In particular, (1) is the only loop in f0 .

Proof. Let c ∈ N. By the fundamental theorem of arithmetic c = 2k1 pk22 pk33 · · · pknn
where ki ≥ 0 for all i ∈ {1, . . . , n} and each pi is an odd prime. We can see that
f0k1 (c) = pk22 pk33 · · · pknn which is odd. Therefore f0k1 +1 (c) = 1 and thus c satisfies the
General Collatz Condition for the function f0 .

Theorem 3.0.5. If c ∈ N, then c satisfies the General Collatz Condition for the
function f1 . In particular, (2, 1) is the only loop in f1 .

Proof. This proof is by strong induction. We see that f1 (1) = 2, f1 (2) = 1, f1 (3) = 4,
and f1 (4) = 2. Therefore 1,2,3 and 4 satisfy the General Collatz Condition for the
function f1 . Assume every natural number up to k for some k > 4 satisfies the General
Collatz Condition for the function f1 . Assume k + 1 is even. It is clear that k + 1 < 2k
k+1 k+1
as long as k > 1. Therefore < k and thus f1 (k+1) = < k. By the induction
2 2
k+1
hypothesis, for all c ∈ N, c < k, there exists n ∈ N such that f1n (c) = 1. Hence,
2
satisfies the General Collatz Condition for the function f1 . Therefore k + 1 satisfies
the General Collatz Condition for the function f1 . Now assume k + 1 is odd. It is clear
k+2
that k + 2 < 2k as long as k > 2 and thus < k. Since f1 (k + 1) = k + 2 is even,
2
k+2
f1 (k + 2) = < k. By the induction hypothesis, for all c ∈ N, c < k, there exists
2
k+2
n ∈ N such that f1n (c) = 1. Hence, satisfies the General Collatz Condition for
2
the function f1 . Therefore k +2 satisfies the General Collatz Condition for the function
f1 and thus k + 1 satisfies the General Collatz Condition for the function f1 .
CHAPTER 3. FUTURE WORK 30

Why is it so easy to verify that f0 and f1 have a unique loop yet it is so difficult in the
f2 case?

Looking at the family of functions {fn : n ∈ N ∪ {0}} leads to a new more General
Collatz Conjecture.

Conjecture 3.0.6 (General Collatz Conjecture). If c ∈ N, then c satisfies the General


Collatz Condition for {fn : n ∈ N ∪ {0}}. In particular, (2n , 2n−1 , . . . , 2, 1) is the only
loop in {fn : n ∈ N ∪ {0}}.

In this paper we reviewed some past results on the Collatz Conjecture. Andrei, Masalagiu,
and Sinisalo gave us some nice results about classes of numbers that satisfy the Collatz
Condition as well as potential orders of future loops.
We discovered a linear Diophantine system which, if solved, would definitively show
whether (4, 2, 1) is the only loop or not. We also conjectured about {fn : n ∈ N ∪ {0}}, a
family of functions to which f belongs.
The purpose of this thesis was to explore the Collatz Conjecture in a new way. To
the best of my knowledge, there are no papers in which the author assumes the existence
of another loop and then discovers characteristics about said loop. These characteristics
include maximum/minimum elements, tails, and roots. As previously stated in Section 2.1,
these characteristics shall lead to restrictions in order for another loop to coexist with the
(4, 2, 1) loop. We hope that these restrictions will lead to a contradiction, thus showing that
(4, 2, 1) is the only loop.
Bibliography

[AM97] Stefan Andrei and Cristian Masalagiu. About the Collatz conjecture. Acta Infor-
matica, 35:167–179, 1997.

[Coo09] Josh Cooper, Nov. 2009. http://www.math.sc.edu/ cooper/combprob.html.

[GP90] William J. Gilbert and Anu Pathria. Linear Diophantine Equations. Master’s
thesis, University of California at Berkeley, 1990.

[Jon98] Christopher Jones. The Collatz Problem: An Interactive Algorithm. International


Baccalaureate Extended Essay, Mathematics, pages 1–11, 1998.

[Row11] Peter Rowlett. The unplanned impact of mathematics. Nature, 475:166–169, 2011.

[Sin03] Matti K. Sinisalo. On the Minimal Cycle Lengths of the Collatz Sequence. Master’s
thesis, University of Oulu, Finland, 2003.

31

S-ar putea să vă placă și