See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/341451219

ON THE TRAJECTORIES OF THE 3x + 1 PROBLEM

Article · May 2020

CITATIONS READS
0 446

1 author:

Roy Burson
California State University, Northridge
9 PUBLICATIONS   0 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

The incompressible Naiver-Stokes Equations View project

Abstract Algebra Book View project

All content following this page was uploaded by Roy Burson on 16 October 2020.

The user has requested enhancement of the downloaded file.

 1

ON THE TRAJECTORIES OF THE 3x + 1 PROBLEM

ROY BURSON

Abstract. This paper studies certain trajectories of the Collatz function.

The main purpose of this paper is to show that if n ∼ 3n + 2 for each odd
number n then every positive integer n ∈ N \ 2N has the representation
k
!
X
ak+1 ai k−i
n= 2 − 2 3 /3k+1
i=0
where 0 ≤ a0 ≤ a1 ≤ · · · ≤ ak+1 . As a consequence, in order to prove
Collatz Conjecture it is sufficient to prove n ∼ 3n + 2 for each n ∈ N \ 2N .
Afterwards, this work illustrates that the average distribution of the first point
of coalescence C(n, 3n + 2) is well approximated by the linear function 45 n for
any n in the interval [1, 100000], and the first point of coalescence C(n, 3n + 2)
appears to be related to solutions of the Diophantine equation
 k
3
N = ξ, k∈Z
2
which gives rise to a possible connection between the uniformity conjecture
 k
3
regarding of the set { ξ : k ∈ Z+ } mod (1) and
2
 
1
: n ∈ N \ 2N mod (1).
C(2n, 2n + 1)
Lastly, it is shown that the Collatz conjecture holds if
( )
α2r−1 + β3s 3 u
N \ 2N ⊆ : (r, s, t, u) ∈ Z , α = Rk , β = 2 + 1, k ∈ N
2t − 3r
were Rk a some recursive sequence related to 3 smooth integers of some level
and the conjecture is false otherwise.

Contents
1. Introduction 2
2. Terminology 3
3. Preliminaries 4
4. Backwards Iterations and Integer Representations 5
5. The first point of coalescence of 2n and 2n + 1 11
6. Dispersion Testing of Spatial Data 13
7. Figures and Tables 13
8. Appendix 19
References 21
1
In this work the natural numbers N are strictly positive, so that 0 6∈ N. The set 2N will be
used to denote the set {2k : k ∈ N} so that N \ 2N = {n ∈ N : n 6∈ 2N }. Summation will always be
given by using the Sigma notation Σ instead
Q of the usual addition sign +. Similiary all products
will be written using the ”pie” notation . The symbol ◦ will always be taken to mean the usual
composition symbol. Specific or arbitrary Relations will be denoted by using the symbols ∼ or ≡.
Date: October 16, 2020.
Key words and phrases. Collatz function, Trajectories, Representation.
1
2 ROY BURSON

1. Introduction
Some problems in mathematics are easy to state but take very complex tools to
prove and often it takes new tools to be developed. The Collatz Conjecture is either
one of this type or else it might be a conspiracy theory to slow down the field of
mathematics (a joke that spread across Yale University). The Collatz Conjecture is
a well known unsolved mathematical problem that concerns the recursive behavior
of the function
(
n
if n ≡ 0(mod2)
T (n) = 2
3n + 1 if n ≡ 1(mod2)
over the set of integers Z. This paper specifically focuses on the Collatz
function over the positive whole numbers N. The problem is most commonly
referred to as the ”3x + 1” problem. The history and exact origin of the prob-
lem is somewhat vague. Some early history on the problem is discussed by
Jeffrey C. Lagarias at http://www.cecm.sfu.ca/organics/papers/lagarias/
paper/html/node1.html. Lagarias gives 197 different documentations on the
topic in his annotated bibliographies [16] and [13]. L. Collatz is credited for
the discovery of the problem during his career as a student. Collatz even
asserts himself that he was the first to study this problem in his letter http:
//www.cecm.sfu.ca/organics/papers/lagarias/paper/html/letter.html.
The problem is stated as follows.

Conjecture 1. For all n ∈ N there is a value k ∈ N such that T k (n) = 1 were T k

is the kth application of the map T . That is T k is the map T k : N × N → N defined
by the rule
(n, k) 7→ (T T · · · ◦ T}) ◦ (n)
| ◦ T ◦{z
k−times

Example 1. For the value n = 11 we can view the iterations with arrows (to
indicate direction) as followed
11→34→17→52→26→13→40→20→10→5→16→ · · · →1
One can also visualize this backwards by reversing the operations of the map and
traces our steps in the reverse direction. Doing so for this example we have the
following
24 − 1 25 − 2 26 − 22 27 − 23 27 − 23 − 3
1→ · · · → → → → →
3 3 3 3 32
2 − 2 − 2 · 3 2 − 2 − 2 · 3 2 − 2 − 2 · 3 − 32
8 4 9 5 2 9 5 2
→ → →
32 32 33
1 6 3 2
2 0−2 −2 ·3−2·3 2 − 2 − 2 · 3 − 2 · 32 − 33
10 6 3
→ → = 11
33 34
As of the authors knowledge the problem has been verified for all natural
numbers n < 87 · 260 by [5]. Many other authors such as [19], and [3] discov-
ered other bounds which are slightly less than achieved by [5] but nonetheless
are worth mentioning. A neat discussion about the empirical results and the
record-holders are discussed by Tomás Oliveria e Silvia at his home page http:
//sweet.ua.pt/tos/3x+1.html, and by Eric Roosendaal http://www.ericr.nl/
 ON THE TRAJECTORIES OF THE 3x + 1 PROBLEM 3

wondrous/index.html. A complete list of the record holders can be found here

http://www.ericr.nl/wondrous/pathrecs.html which was accomplished by the
yoyo@home project in 2017. In [16] and [13] Lagarias also mentions the empirical
evidence of the problem given by Oliveirá e Silva. An interesting study on proper-
ties of divergent trajectories of the Collatz function (if one exist) can be found at
[20] ( or http://www.csun.edu/~vcmth02i/Collatz.pdf). A similar result of this
paper is given by C. Cadogan in his works [8] and [7], which is also discussed in
Lagarias’s bibliography [13]. Some more interesting discussions and findings that
relate to this work can be found in [9, 11, 14, 15, 17].
This work serves as both a numerical evaluation of the Collatz function and a
theoretical study. The main purpose is to illustrate that in order to prove the Col-
latz conjecture it suffices to prove that set O+ (2n) ∪ O+ (2n + 1) is non-empty for
all n ∈ N \ 2N . As a result it becomes apparent that the points in the sets O+ (2n),
O+ (2n + 1) which coalescence are highly important in determining the validity of
the conjecture. Driven by these results this work then provides an algorithmic
study on the first point of coalescence for these specific trajectories and studies
some theoretical connections between them and some Diophantine equations. The
framework of the paper is as follows. Section 2 serves to outline all of the terminol-
ogy and notation needed to grasp the material presented, whereas Section 3 serves
to underline the preliminaries which are needed throughout the work. Section 4
contains majority of the mathematical details of the proof to the main theorem.
Section 5 and 6 serve to discuss the results obtained from the algorithm developed
for this work (which is left in the Appendix).

2. Terminology
This section contains all of the definitions that are needed throughout the paper.
Instead of displaying the definitions immediately before its usage I have gathered
them in a single section to encourage the reader to visit this section when they need
to revisit a definition. This I believe this allows the reader an easier way to find
them, and can avoid them elsewhere in the reading. The reader is encouraged to
come back to these when needed. In order to prove the main result of the paper
the following definitions are needed.
Definition 1. let m ∈ N. Write m ≡ R if and only if the number m has the
representation !
k
X
ak+1 ai k−i
m= 2 − 2 3 /3k+1
i=0
where (ai ) is a monotonically increasing sequence of positive integers for i ≤ k + 1.
Definition 2. The Trajectory or Forward Orbit of a positive integer n is the set
O+ (n) = {n, T (n), T 2 (n), · · ·}
were T : N → N is the Collatz function defined by n → n2 if n is even and n → 3n+1
if n is odd, and T k is the function T : N × N → N defined by
(n, k) 7→ (T ◦ T ◦ · · · ◦ T ) ◦(n)
| {z }
k−times

Definition 3. Given two integers n1 and n2 define the relation n1 ∼ n2 if and only
if the two trajectories O+ (n1 ) and O+ (n2 ) coalesce, i.e. O+ (n1 ) ∩ O+ (n2 ) 6= ∅.
4 ROY BURSON

Definition 4. Given two integers n, m such that n ∼ m, the Merging Height of n

or m, denoted ϕ(n, m) , is defined by the rule

ϕ(n, m) = min{k1 , k2 ∈ N : T k1 (n) = T k2 (m)}

If ϕ(n, m) does not exist then we decree that ϕ(n, m) = ∞.

Definition 5. Given two integers n, m such that n ∼ m, the Coalescence Point

C : N × N → N of the integers n and m is defined by the rule (n, m) 7→ T ϕ(n,m) (n)
or (n, m) 7→ T ϕ(n,m) (m) since they are equivalent.

Definition 6. Let k ∈ N and f : N → N be a function. The average coalescence

point ς : N → N with respect to f is defined as the mapping
1X 1 X ϕ(n,f (n))
k 7−→ C(n, f (n)) = T (n)
k k
n≤k n≤k

Definition 7. A number n ∈ N is said to be 3-smooth special of level k if it can

written as
k
X
(2.1) n= 2i 3k−i
i=1

with 1 ≤ 1 ≤ 2 ≤ 3 ≤ · · · ≤ k .

3. Preliminaries
Proposition 1. The relation ∼ is an equivalence relation.

Proof. Reflexive: choose a ∈ N arbitrary. Since a ∼ a since O+ (a) = O+ (a) it

follows that O+ (a) ∩ O+ (a) 6= ∅.

Symmetric: If a ∼ b then O+ (a) ∩ O+ (b) 6= ∅. Since O+ (a) ∩ O+ (b) =

O (b) ∩ O+ (a) it follows that O+ (b) ∩ O+ (a) 6= ∅ which implies b ∼ a.
+

Transitive: Suppose that a ∼ b and b ∼ c. Since ∼ is symmetric it follows that

b ∼ a and b ∼ c. Then by defintion O+ (b) ∩ O+ (a) 6= ∅ and O+ (b) ∩ O+ (c) 6= ∅.
This means that there is a value k1 and k2 such T k1 (b) = a and T k2 (b) = c. If
k1 < k2 and choose  as k2 − k1 =  and write

c = T k2 (b)
= T k1 + (b)
(3.1)
= T  T k1 (b)

= T  (a)

The case k2 < k1 follows by analogy. Lastly if k1 = k2 then we find that

a = T k1 (b) = T k2 (b) = c

and again a ∼ c. This completes all three possible cases. 

 ON THE TRAJECTORIES OF THE 3x + 1 PROBLEM 5

4. Backwards Iterations and Integer Representations

As shown by [1, 2, 6, 4, 18], and discussed in [10] we have that

k
!
X
+ ak+1 ai k−i
N
∀m ∈ N \ 2 , 1 ∈ O (m) ⇐⇒ m = 2 − 2 3 /3k+1
i=0

for some sequence (aj )k+1

where 0 ≤ a0 ≤ a1 ≤ a2 ≤ · · ·ak+1 . If m has
j=1
this representation given above then m cannot be a power a 2 because any ele-
ment of the set 2N = {2, 22 , 23 , · · ·} cannot be represented in this form. This can
easily
 be verified by observing
 that the numbers that can be written in the form
ak+1
Pk ai k−i k+1
2 − i=0 2 3 /3 are exactly those numbers obtained by iterating the
functions n 7→ 2n and n 7→ n−1 3 in succession, beginning at the number n = 1,
were as each function is applied at least once and the function n 7→ n−1
3 is never
applied twice in succession. The numbers in the set 2N can only be obtained by
iterating the function n → 2n, beginning at n = 1, so that the function n → n−1 3
is
 never applied. This shows that no integer in the set 2N has the representation
Pk
2ak+1 − i=0 2ai 3k−i /3k+1 . The next lemma shows that if a number n has this
2n−1
representation then the number 2n and 3 must also have the same representa-
tion.
2n−1
Lemma 1. (Closure of R) If n ≡ R then 2n ≡ R and if n ≡ R then 3 ≡ R.
k+1
Proof. (Direct proof) First suppose n ≡ R. Then there is a sequence (aj )j=1 so
that
k
!
X
ak+1 ai k−i
n= 2 − 2 3 /3k+1
i=0
where 0 ≤ a0 ≤ a1 ≤ a2 ≤ · · ·ak+1 . Write

k
! ! k
!
X X
ak+1 ai k−i k+1 ak+1 +1 ai +1 k−i
2n = 2 2 − 2 3 /3 = 2 − 2 3 /3k+1
i=0 i=0
k+1
Define the sequence (bj )j=1 by bj = aj+1 + 1 for each 0 ≤ j ≤ k + 1 so that (bj )
is also positive and increasing. Then we have
k
!
X
bk +1 bj k−i
2n = 2 − 2 3 /3k+1 ≡ R
i=0

Therefore 2n ≡ R. Now suppose that n ≡ R then by the above 2n ≡ R. Write

k
!
X
ak+1 ai k−i
2n = 2 − 2 3 /3k+1
i=0

were as 0 ≤ a0 ≤ a1 ≤ · · ·ak+1 . Then

 Pk 
ak+1 ai k−i
2n − 1 2 − i=0 2 3 /3k+1 − 1
=
3 3
6 ROY BURSON

k+1
Define the sequence (bj )j=1 by
(
0 if j = 0
bj =
aj−1 if 1 ≤ j ≤ k + 1
Then !
k+1
2n − 1 bk+1
X
bi k+1−i
= 2 − 2 3 /3k+2
3 i=0
k+1
and the sequence (bj )j=1 remains monotonically increasing. Therefore, it follows
that 2n−1
3 ≡R 

Lemma 2. Let a ∈ N. Then

a
3 2 +1 + 2 < 2a + 1
for all a ≥ 8.

Proof. (By induction) The proof follows by induction. Let

a
(4.1) S = {1, 2, 3, · · ·, 7} ∪ {a ∈ N : 3 2 +1 + 2 ≤ 2a + 1}
The value a = 8 is in S since 245 = 35 + 2 < 28 + 1 = 257 Now if a ≥ 8 ∈ S then
a 1 a
3 2 +1 + 2 ≤ 2a + 1 ⇒ 3 2 (3 2 +1 + 2) ≤ 2(2a + 1)
1 1 (a+2)
⇒ 3 2 (3 2 + 2) ≤ 2(2a + 1)
1 (a+3) 1
⇒ 32 + 23 2 ≤ 2a+1 + 2
(4.2) 1 (a+3)
⇒ 32 + 3 ≤ 2a+1 + 2
1 (a+3)
⇒ 32 + 2 ≤ 2a+1 + 1
a+1
⇒3 2 +1 + 2 ≤ 2a+1 + 1
Therefore a + 1 ∈ S so that S = N. 

Lemma 3. Let n ∈ 2N = {2n : n ∈ N} and write n is its conical form

k
Y
(4.3) n = 2 pαi
βi
i=1

Then there exist a value k ∈ N so that

 Q 
3 2 k αi
i=1 βi + 1
p if  ≡ 0(mod2)
(4.4) T k (n + 1) =  
3b 2 c+1 Qk pαi + 2

if  ≡ 1(mod2)
i=1 βi

Qk
Proof. Let n ∈ 2N and write n = 2 pα
βi for some  ∈ N. First,
i
i=1 assume  ≡
Q 
 k  k αi
0(mod2) and take k = 3 2 . Then the claim is that T (n + 1) = 3 2 i=1 pβi + 1.
Qk
Write n + 1 = 2 i=1 pα
βi + 1. Define the sequence (ai ) by the rule
i

(
ai + 1 if i ≡ 0(mod2)
(4.5) ai+1 =
ai + 2 if i ≡ 1(mod2)
 ON THE TRAJECTORIES OF THE 3x + 1 PROBLEM 7

were as a1 = 1. Now since n + 1 must be odd by computing T k consecutively we

find
k
Y k
Y
n + 1 = 2 pβαii + 1 ⇒ T (n + 1) = 32 pα
βi + 2
i 2

i=1 i=1
k
Y
⇒ T 3 (n + 1) = 32 2−2 pα
βi + 1
i

i=1
k
Y
⇒ T 4 (n + 1) = 33 2−2 pα
βi + 2
i 2

(4.6) i=1
k
Y
6
⇒ T (n + 1) = 3 2 4 −4
pα
βi + 1
i

i=1
..
.
k
Y
⇒ T k (n + 1) = 3i pα
βi + 1
i

i=1

for some i and k because  is even. Actually we know i = 2 and k = ai (were (ai )
was defined above) so
k

Y
T k (n + 1) = 3 2 pα
βi + 1
i

i=1
as desired. Now if  ≡ 1(mod2) then we may write  = 0 + 1 were  ≡ 0(mod2).
Hence, we see that
k
Y
T k (n + 1) = T k (2 pα
βi + 1)
i

i=1
k
0 Y
= T k (2 +1 pα
βi + 1)
i

i=1
(4.7)
k
0
Y
=T k+1
(2 pα
βi + 1)
i

i=1
k
!
0
Y
=T T (2k
pαi
βi + 1)
i=1

Now we may apply the first part of this proof since 0 is even. We can successfully
compute T k (n + 1) as we did above. Thus
k
! k
! k
!
0
k 0 
Y Y Y
αi α b c+1 α
T T (2 pβi + 1) = T 3 2 pβi + 1 = 3 2
i
pβi + 2
i

i=1 i=1 i=1

Lemma 4. Assume that a ∈ N. If a is even, then there exist a value k such

a
that T k (2a + 1) = 3 2 + 1. If a is a odd, then there exist a value k such that
a
T k (2a + 1) = 3b 2 c+1 + 2.
8 ROY BURSON

Qk
Proof. Let n = 2a for a ∈ N. In regards to Lemma 3 we have i=1 pα
βi = 1.
i

Therefore there is a value k ∈ N so that

( a
k 32 + 1 if a ≡ 0(mod2)
T (n + 1) = ba c+1
3 2 + 2 if a ≡ 1(mod2)

Theorem 1. If for all n ∈ N \ 2N n ∼ 3n + 2, then for all n ∈ N \ 2N it follows
that n ≡ R
Proof. (By induction) First suppose for all n ∈ N \ 2N that n ∼ 3n + 2. Let
X = {x1 , x2 , · · ·} = N \ 2N were as x1 < x2 < x3 < · · · and let S be the set
defined by S = {n ∈ N : xn ≡ R}. Now 1 ∈ S since x1 = min(N \ 2N ) = 3 and
5 2
3 = 2 −232
−1
≡ R. Now Suppose xk ∈ S for 1 ≤ k ≤ n. The proof that xn+1 ∈ S is
broken into three cases.

In the first case suppose xn is not one less than a power of 2 and xn is even.
Then it follows that xn+1 is odd. Since xn+1 is odd there exist a value t ≥ 2 ∈ N
such that xn+1 = 2t + 1. Write

2(3t + 2) − 1
xn+1 = 2t + 1 =
3
By the inductive hypothesis we know xn ≡ R. Also under the assumption
t ∼ 3t + 2 for each value t it follows that 3t + 2 ≡ R or else a power of 2. If
3t + 2 ≡ R then by direct application of Lemma 1 it follows that 2(3t+2)−1
3 ≡ R.
a 2(3t+2)−1 2a+1 −1
Otherwise, if 3t + 2 = 2 for some positive integer a then 3 = 3 ≡ R.
Therefore in either case xn+1 ≡ R.

In the second case, suppose xn is not one less than a power of 2 and that
xn is odd. Then it follows that xn+1 is even. Since xn+1 is even there is
a value t such that xn+1 = 2t. Notice that t cannot be a power of 2 since
xn+1 = 2t = xn + 1 6= (2a − 1) + 1 = 2a . From this it follows that t cannot be a
power of 2. Therefore, since t = xn+1
2 = xn2+1 < xn it follows that t is an element
of the inductive set, and hence t ≡ R. By Lemma 1 it follows that 2t ≡ R.
Therefore, xn+1 ≡ R.

In the third and final case, suppose that xn is exactly one less than a power of
2. That is, suppose xn = 2a − 1 for some positive integer a. Moreover, for reasons
that will become clear later assume that a ≥ 8. Then it follows that xn+1 = 2a + 1.
Now if a is even then by direct application of Lemma 4 there exist a value k such
a
that T k (2a + 1) = 3 2 + 1. However by Lemma 2 we have the inequality
a a
3 2 + 1 < 3 2 +1 + 2 < 2a + 1 = xn+1
a
whenever a ≥ 8. Therefore xn+1 = 2a + 1 iterates to the number 3 2 + 1 and this
number is either a power of 2 or or it is an element of the inductive set S. In any
case we have xn+1 ≡ R. Now if a is odd then by direct application of Lemma 4
a
there exist a value k such that T k (2a + 1) = 3b 2 c + 2. By Lemma 2 we have the
inequality
a a
3b 2 c+1 + 2 < 3 2 +1 + 2 < 2a + 1 = xn+1
 ON THE TRAJECTORIES OF THE 3x + 1 PROBLEM 9

a
whenever a ≥ 8. Therefore xn+1 = 2a + 1 iterates to the number 3b 2 c + 2 and this
number is either a power of 2 or it is an element of the inductive set S, in either
case we have xn+1 ≡ R. The separate cases a < 8 can be checked and verified by
strait forward computation.

Now in all three we found that xn+1 ≡ R. Since there are no more cases it
follows that S = N. This competes the proof. 

Proposition 2.
n ∼ 3n + 2 ⇐⇒ 2n ∼ 2n + 1 ∀n ∈ N \ 2N

Proof. First suppose that n ∼ 3n + 2. It is easy to see that T (2n) = n ∼ 3n + 2.

Therefore 2n ∼ 3n + 2. Similarly, note that T (2n + 1) = 3(2n + 1) + 1 = 6n + 4 and
so T 2 (2n + 1) = 3n + 2. This means that 2n + 1 ∼ 3n + 2. So we have found that
2n ∼ n ∼ 3n + 2 ∼ 2n + 1. By Proposition 1 we know ∼ is an equivalence relation
and thus 2n ∼ 2n + 1. Conversely, if 2n ∼ 2n + 1 then again we easily see that
n ∼ 2n ∼ 2n + 1 ∼ 3n + 2 and thus n ∼ 3n + 2. This completes the proof. 
k
Lemma 5. Let n ∈ N \ 2N . If there is a sequence (j )j=1 so that
k−1
3k (2n + 1) + 3k−1 X 3k−i−1
2n = Pk + Pk
2( j=1 j ) 2( j=1 j ) i=1

then 2n ∼ 2n + 1.

Proof. The proof is simple and instructive. We first assume that the hypothesis
holds valid. Write
k−1
3k (2n + 1) + 3k−1 X 3k−i−1
(4.8) 2n = Pk + Pk
2( j=1 j ) 2( j=1 j ) i=1

Define k ∗ as
k
X
∗
(4.9) k = j + k
i=1

Then compute
∗
T k (2n + 1) = (T
| ◦ T ◦{z
T · · · ◦ T}) ◦ (2n + 1)
k? −times
Pk
j +k
=T i=1 (2n + 1)
ak a2
=T (· · · T (T (T a1 (2n + 1))))
(4.10) k−1
3k (2n + 1) + 3k−1 X 3k−i−1
= Pk + Pk
2( j=1 aj ) 2( j=1 aj ) i=1
k k−1 k−1
3 (2n + 1) + 3 X 3k−i−1
= Pk + Pk
2( j=1 j ) i=1 2( j=1 j )
k ∗
for any sequence (ai )i=1 . Therefore it follows that 2n = T k (2n + 1) and so 2n ∼
2n + 1. 
10 ROY BURSON

Theorem 2. Let n ∈ N \ 2N . If there exist a value k > 2 and a positive sequence

k
of integers (j )j=1 such that
Pi
3k−1 + i=1 3k−i−1 2( j=1 j )
Pk−1
(4.11) n= Pk ,
2( j=1 j )+1 − 2 · 3k
then n ∼ 3n + 2.

Proof. Suppose that the hypothesis stated is true. Then compute n as follows
Pi
3k−1 + i=1 3k−i−1 2( j=1 j )
Pk−1
n= Pk
2( j=1 j )+1 − 2 · 3k
3k−1 ( kj=1 j )−1 · Pk 3k−i−1
P
2 +2 i=1 (Pk
2 j=i+1 j )

= Pk

2 j=1 j − 3k
 
3k−1
Pk−1 3k−i−1
(4.12) + i=1 ( k
 2( j=1 j )
Pk
2 j=i+1 j )  ( k
P
  P  )−1
= Pk
2
j=1 j

2( j=1 j ) − 3k


3k−1 3k−i−1
Pk−1
+ i=1
2( j=1 j ) 2( j=1 j )
Pk Pk
 
= 3k
2−
2( )
Pk

j=1 j

Thus by Eq.(4.12) it follows that

k−1
3k 3k−1 3k−i−1
  X
(4.13) n 2− Pk = Pk + Pk
2( j=1 j )
2( j=1 j )
i=1 2( j=1 j )
Isolating for the value 2n in Eq.(4.13) yields the result
k−1
3k (2n + 1) + 3k−1 X 3k−i−1
(4.14) 2n = Pk + Pk
2( j=1 j ) 2( j=1 j ) i=1

By Lemma 5 it then follows that 2n = T k (2n + 1). As a consequence of Proposition

2 we have 2n ∼ 2n + 1 ⇒ n ∼ 3n + 2. 

Proposition 3. Every 3-smooth number α of level k > 2 can be written as α = Rk

k
were (Rn )n=1 is the sequence defined by

(4.15) Rn = 2an (Rn−1 + 3n )

with R1 = 2a1 + 1 for some sequence (an )ki=1 were ak ≥ 1 for each k ∈ {1, 2, · · · , k}.

Proof. Let α be a 3-smooth number of level k > 2. By definition α can be written

as
k
X
(4.16) α= 2i 3k−i
i=1
 ON THE TRAJECTORIES OF THE 3x + 1 PROBLEM 11

with 1 ≤ 1 ≤ 2 ≤ · · · ≤ k . Write
k k
!
X X
i k−i 1 i −1 k−i
2 3 =2 2 3 + 3k−1
i=1 i=2
1 k−1
+ 22 −1 3k−2 + · · · + 2k −1


=2 3
(4.17)
= 21 3k−1 + 21 −2 3k−2 + · · · + 2k −1 −2

.
= ..
= Rk
Pk−1
were we let a1 = 2k −( j=1 j )
+ 1. This completes the proof. 
Theorem 3. If the Collatz conjecture is true, then every number n ∈ N can be
written as
α3k−1 + β2y
(4.18) n= ζ
2 − 2 · 3k
for some α, β, k, y, ζ ∈ N.
Proof. Suppose the conjecture holds. Then we must have 2n ∼ 2n + 1 for each
n ∈ N \ 2N . So the reverse condition of Theorem 2 must then have that n can be
written as
Pi
3k−1 + i=1 3k−i−1 2( j=1 j )
Pk−1
3k−1 + Rk
n= = ζ
2 − 2 · 3k
Pk
2( j=1 j )+1 − 2 · 3k


3k−1 + 2an Rk−1 + 3k
=
2ζ − 3ω
k−1
(4.19) 3 + 2ak 3k + 2ak Rk−1
=
2ζ − 2 · 3k
k−1
3 (2 + 1) + 2ak Rk−1
ak
=
2ζ − 2 · 3k
α3k−1 + β2y
= ζ
2 − 2 · 3k
P 
i y
with ζ = j=1 j + 1, y = ak , α = 2 + 1, and β = Rk−1 . 

Theorem 4. If every number n ∈ N can be written as

α3k−1 + β2y
(4.20) n=
2ζ − 2 · 3k
k
with α = 2 + 1 and β = Rk then the Collatz conjecture is true.

5. The first point of coalescence of 2n and 2n + 1

In this section I discuss the empirical results that have been obtained from the
algorithm designed for this work. After discussing the empirical results I elaborate
on the dispersion algorithm that utilizes the spatial data in order to identify the
equations which localize the behavior of the first point of coalescence of these specific
trajectories. I will demonstrate that the first coalescence point of 2n and 2n + 1 can
be approximated near specific regions or lines in R2 and then after make a couple
observations from the data obtained. First, as discussed prior, the algorithm used
12 ROY BURSON

in this work has been developed to handle a broader scenario than actually need
be. That is if a, b ∈ N then the algorithm is designed to compute the first point of
coalescence of the to integers n and an + b up to any value that the user specifies.
By letting a = 3 and b = 2 then one obtains the important trajectory relationship
described earlier. The first test (for these settings) deployed accounted for all of
the integers k ≤ 100. The scatter plot of these results are depicted in Figure 1.
Even for such small values of k the spatial data exhibits some linear behaviors.
Figure 2 illustrates the first employment of the behavior of the average value of the
first point of coalescence for the trajectories O+ (2k) and O+ (2k + 1). For these
specific trajectories the average value of the first point of coalescence is defined by
the summation
1X 1 X ϕ(2k,2k+1)
ς(k) = C(2k, 2k + 1) = T (2k)
k k
n≤k n≤k

As seen in figure 2 it appears that the average value can be estimated by a simple
linear approximation. Figure 3 and 4 illustrate the behavior of T ϕ(2k,2k+1) (2k) and
ς(k) for values k ∈ [1, 1000]. By extending the domain from k ≤ 100 to k ≤ 1000
we see dramatic changes and observe even sharper linear properties are observed.
The linear properties of the functions T ϕ(2k,2k+1) (2k) and ς(k) become clearer as
we enlarge the domain. Figures 5-8 show the behavior of T ϕ(2k,2k+1) (2k) and ς(k)
in the interval ∈ [1, 10000] and [1, 100000]. One can easily see that majority of the
data seem to be located near the solutions to specific diophantine equations. The
surprising result is the algorithm indicates that ς(k) ≈ 45 k for large values of k.
From here it seems reasonable to be tempted to immediately try to find a direct
formulation that guarantees this type of behavior. It is thus of most interest to find
the set of functions
{f : f (n) = an + b = T ϕ(2k,2k+1) (n), a, b ∈ Z}
that resemble the linear behavior of the first point of coalescence of the trajectories
that are being studied herein. This means solving for the corresponding coefficients
a and b so that the equality holds f (n) = an + b = T ϕ(2k,2k+1) (2k) holds valid.
In order to find the values of a and b for each localized trajectory this works
developed a dispersion algorithm that measures the disbursement and runs a
Nearest Neighbors Algorithm (NNA) and finds the nearest points to a specific
trajectory to form the best linear approximations possible. The precise results of
this algorithm will be discussed in the next section.

The time complexity of the algorithm it quite interesting. The algorithm took
5.49ms to process all of the integers up 100, 7.381ms for integers up to 1000,
22.297ms for integers up to 10000, and a massive 338.165ms for all the integers
up to 100000. These values were produced by timing the algorithm with the built
in timing mechanism and functions available. These results have been documented
and are given in the Table 1. The left column gives the input interval and the
right column denotes how much time has elapsed from the beginning to the need
of the algorithm during each test. Figure 9 illustrates the behavior of the time
complexity for these specific trajectories that have been analyzed. Amazingly, it
seems reminiscent off that of the average value of the first point of coal sense. That
is because the time complexity is a linear function similarly to the approximation.
A small cusp is also observed near the lower proportional of the time complexity
 ON THE TRAJECTORIES OF THE 3x + 1 PROBLEM 13

curve. The graphs indicate that average value is roughly ten times larger than the
value of the time complexity on all test runs. Using this time complexity curve we
can estimate the time it will take to run any test.

6. Dispersion Testing of Spatial Data

In this section I discuss the dispersion test on the spatial data that has been ob-
tained from the algorithm developed to depict the first point of coalescence ϕ(n, m)
of n and m. The dispersion test consist of finding the suitable linear approxima-
tions to the trajectories that are being analyzed. The algorithm indicates that the
distribution of the coalescence points are closely related to the integer solutions of
the equations
 k
3
N= ξ, k ∈ Z
2
Majority of the coalescence points seem to reside on these curves but not all of
them are. As we can see from the figures provided there are some isolated points
of interest. The purpose of this dispersion algorithm was to find these specific
coefficients and intercepts in order to investigate the properties of the first point of
coalescence further.
The experimental evidence is crude and does not explain a lot. Henceforth, in
future investigations it would be of particular interest to study the relationship
between the exponent y and the x component of these two functions which are
being modeled here (that is the f and the g functions). in order to analyze these
trajectories more thoroughly one needs to try to relate the value y and x through
the value n. Given a particular value n ≤ 100000 the algorithm proves that the
trajectories first merge near the diophantine solutions to the equations
 k
3
N= ξ, k ∈ Z
2
These equations have integral solutions whenever 2y |ξ or 3y |ξ. These are precisely
the integers divisible by a power of 2 or 3. In this regard these empirical results
indicate that the Collatz function is shuffling through the powers of 2 and 3. This
”shuffling” that occurs between powers of 2 and 3 seems to be the next most
important observation to investigate.

7. Figures and Tables

Interval Time elapsed

[1, 100] 5.493256600 ms
[1, 1000] 7.3807948 ms
[1, 10000] 22.9695312 ms
[1, 100000] 338.1653204 ms
Table 1. Time complexity for certain Intervals
14 ROY BURSON
ON THE TRAJECTORIES OF THE 3x + 1 PROBLEM 15
16 ROY BURSON
ON THE TRAJECTORIES OF THE 3x + 1 PROBLEM 17
18 ROY BURSON
 ON THE TRAJECTORIES OF THE 3x + 1 PROBLEM 19

8. Appendix

1 import numpy as np
2 from numpy import savetxt
3 import sys
4 import fileinput
5
6 __author_ = ’ Roy Burson ’
7 __copyright_ = ’ - - - - - - - - - @copyright 2019 - - - - - - - - - ’
8
9 def collatz ( number ) :
10 if number % 2 == 0:
11 return number // 2
12
13 elif number % 2 == 1:
14 result = 3 * number + 1
15 return result
16
17 def a v e r a g e _ v a l u e _ o f _ a r r a y ( args ) :
18 length = len ( args )
19 k = sum ( args )
20 quoteint = k / length
21 return quoteint
22
23 x_list = []
24 y_list = []
25 t r a j e c t o r y _ 0 mo d 3 = []
26 t r a j e c t o r y _ 2 mo d 3 = []
27 a v e r a g e _ v a l u e s _ o f _ m o d u l u s e s = []
28
29 y_range = int ( input ( " Enter max value of domain : " ) )
30 for n in range (1 , y_range ) :
31 x_list . append ( n )
32 k = 2* n +1
33 t r a j e c t o r y _ 0 m od 3 = []
34 t r a j e c t o r y _ 2 m od 3 = []
35 c o a l e s c e n c e _ l is t = []
36 t r a j e c t o r y _ 0 m od 3 . append ( n )
37 t r a j e c t o r y _ 2 m od 3 . append ( k )
38
39 while n != 1:
40 n = collatz ( int ( n ) )
41 t r a j e c t o r y _ 0 m o d3 . append ( n )
42
43 while k != 1:
44 k = collatz ( int ( k ) )
45 t r a j e c t o r y _ 2 m o d3 . append ( k )
46
47 if len ( t r a j e c t o r y _ 0 m o d3 ) >= len ( t r a j e c t o r y _ 2 m o d 3 ) :
48 for i in range ( len ( t r a j e c t o r y _ 2 m o d 3 ) ) :
49 for j in range ( len ( t r a j e c t o r y _ 0 m o d 3 ) ) :
50 if ( t r a j e c t o r y _ 0 m o d 3 [ j ] == t r a j e c t o r y _2 m o d 3 [ i ]) :
51 c o a l e s c e n c e _ p o i n t = t r a je c t o r y _ 0 m o d 3 [ j ]
 20 ROY BURSON

52 c o a l e s c e n c e _ l is t . append ( c o a l e s c e n c e _ p o i n t )
53 break
54
55 if len ( t r a j e c t o r y _ 2 m o d3 ) > len ( t r a j e c t o r y _ 0 m o d 3 ) :
56 for i in range ( len ( t r a j e c t o r y _ 0 m o d 3 ) ) :
57 for j in range ( len ( t r a j e c t o r y _ 2 m o d 3 ) ) :
58 if ( t r a j e c t o r y _ 2 m o d 3 [ j ] == t r a j e c t o r y _0 m o d 3 [ i ]) :
59 c o a l e s c e n c e _ p o i n t = t r a je c t o r y _ 2 m o d 3 [ j ]
60 c o a l e s c e n c e _ l is t . append ( c o a l e s c e n c e _ p o i n t )
61 break
62
63 if len ( c o a l e s c e n c e _ l i st ) == 0:
64 print ( " sorry there is no intersection point " )
65 y_list . append (0)
66 y_list . append ( c o a l e s c e n c e_ l i s t [0])
67 a v e r a g e _ v a l u e s _ o f _ m o d u l u s e s . append ( a v e r a g e _ v a l u e _ o f _ a r r a y ( y_list ) )
68 xvals = np . array ( x_list )
69
70 yvals = np . array ( y_list )
71 Z = [( xvals [ i ] , yvals [ i ]) for i in range ( len ( xvals ) ) ]
72 savetxt ( ’ data . csv ’ , Z , delimiter = ’ , ’)
73 yvals2 = np . array ( a v e r a g e _ v a l u e s _ o f _ m o d u l u s e s )
74 Z2 = [( xvals [ i ] , yvals2 [ i ]) for i in range ( len ( xvals ) ) ]
75 savetxt ( ’ a v e r a g e _ v a l u e _ d a t a . csv ’ , Z2 , delimiter = ’ , ’)
76 print ( __author_ )
77 print ( __copyright_ )
 ON THE TRAJECTORIES OF THE 3x + 1 PROBLEM 21

References
1. J. Amigo, Representing the integers with powers of 2 and 3, Acta Inform., 43 (2006) 293?306.
2. S. Andrei, M. Kudlek, and R. S. Niculescu, Some results on the Collatz Problem, Acta Inform.,
37 (2000) 145?160
3. e Silva, T. Oliveira. ”Empirical verification of the 3x+ 1 and related conjectures.” The Ultimate
Challenge: The 3x+ 1 Problem (2010): 189-207.
4. Crandall, Richard E. ”On the ”3x+1” problem.” Mathematics of Computation 32.144 (1978):
1281-1292.
5. Barina, David. ”Convergence verification of the Collatz problem.” The Journal of Supercom-
puting (2020): 1-8.
6. Böhm, Corrado, and Giovanna Sontacchi. ”On the existence of cycles of given length in integer
sequences like xn+1 = xn /2 if xn even, and xn+1 = 3xn + 1.” Atti della Accademia Nazionale
dei Lincei. Classe di Scienze Fisiche, Matematiche e Naturali. Rendiconti 64.3 (1978): 260-264.
7. Charles C. Cadogan (2000), The 3x+1 problem: towards a solution, Caribbean J. Math.
Comput. Sci. 10 (2000), paper 2, 11pp. (MR 2005g:11032)
8. Charles C. Cadogan (2003), Trajectories in the 3x+1 problem, J. of Combinatorial Mathe-
matics and Combinatorial Computing, 44 (2003), 177?187. (MR 2004a:11017)
9. L. Collatz, On the motivation and origin of the (3n + 1)?Problem, J. Qufu Normal University,
Natural Science Edition, 12(3) (1986) 9?11
10. Goodwin, J. (2015). The 3x 1 Problem and Integer Representations.
11. Klamkin, M. S. ”An infinite permutation.” SIAM Review 5.3 (1963): 275.
12. J. C. Lagarias, The 3x + 1 problem: An annotated bibliography (1963?
1999), web document arXiv:math/0309224v13 [math.NT], available at arXiv.org:
http://arxiv.org/abs/math/0309224v13
13. J. C. Lagarias, The 3x + 1 Problem: An Annotated Bibliography, II
(2000?2009), web document arXiv:math/0608208v6 [math.NT], available at arXiv.org:
http://arxiv.org/abs/math/0608208v6
14. lagarius, J. C ”The 3x+1 problem and its generalization ” Amer Math. Monthly 3-23, 1985.
15. lagarius, J. C ” The 3x+1 problem: An Annotated Bibliography” Amer Math. Monthly 3-23,
2012.
16. J. C. Lagarias, The 3x + 1 problem: An annotated bibliography (1963?
1999), web document arXiv:math/0309224v13 [math.NT], available at arXiv.org:
http://arxiv.org/abs/math/0309224v13
17. J. C. Lagarias, The Ultimate Challenge: The 3x+1 Problem, American Mathematical Society,
2010
18. G. Wirsching, On the combinatorial structure of 3n + 1 predecessor sets, Discrete Math., 148
(1996) 265-286.
19. Chamberland, Marc. ”A3x + 1 survey: number theory and dynamical systems.” The Ultimate
Challenge: The3x + 1 Problem (2010): 57-78.
20. Motta, Francis Charles, et al. ”An Analysis of the Collatz Conjecture.”
21. Salazar, Jorge. ”On the Behavior of Unbounded Collatz Sequences.” arXiv preprint
arXiv:2003.04615 (2020).
Email address: [email protected].

View publication stats