Structure to the Collatz Conjecture?

Category:

By

/

4–6 minutes

read

This is going to be a bit different than my other posts.

A few weeks ago, I stumbled across a video about the Collatz Conjecture.  It was from one of the YouTube channels I follow, TippingPointMath.  The video very clearly explains the Conjecture and gives a little background on the history of the problem.

Out of pure curiosity I began to play with the (as I now knew it) “3N+1 Conjecture” myself.  I did not look much into other people’s methods of solving (though I did a quick search to ensure it was still an “unproven” conjecture), but attacked it myself from two specific angles.

The first attack was to evaluate what, in my mind, was a similar proposal, but one that is much easier to prove and observe.  I looked at the “N+1 Conjecture”.  That is, I formulated an algorithm similar to Collatz’s that goes:

{  N is a positive, whole integer

If N is even, N’=N/2;  if N is odd, N’=N+1.}

I did not write out a formal proof, but I think it is easy to see that this algorithm always collapses to N=1, and then perpetuates a cycle of {1,2}.  If you start out with an even N, N’ is obviously less than N;  If N is odd, N’ will be greater than N by a value of 1, but this guarantees that N’ will be even, so then N”=N’/2=(N+1)/2.  It is easy to see that N” will be less than N for all integers N>2.

A few examples of number chains this algorithm would produce:

Starting with N=64: {64,32,16,8,4,2,1,2,1…}

Starting with N=99: {99,100,50,25,26,13,14,7,8,4,2,1,2,1…}

Starting with N=214527: {214527, 214528, 107264, 53632, 26816, 13408, 6704, 3352, 1676, 838, 419, 420, 210, 105, 106, 53, 54, 27, 28, 14, 7, 8, 4, 2, 1, 2, 1…}

Even starting with obscure primes and odd numbers, it is easy to see the decline of the chains to the {1,2} cycle.

Another way to look at both the N+1 and the 3N+1 algorithms, is in reverse.  In reverse the N+1 algorithm becomes:

{If N’ is odd, N=2*N’

If N’ is even, N=2*N’    AND   N’-1}

By looking at them this way, a tree like structure emerges, where N’ can branch off into more than one Ns.  For example, if N’ is 4, it could be that N is 8 and was divided by 2, or that N is 3 and 1 was added to N.  Both situations will result in N’=4.  This is best shown in plot format, where it clears displays a branching tree structure:

18 generations of the N+1 tree
18 generations of the N+1 tree
15 generations of the N+1 tree
15 generations of the N+1 tree

Though these plots show the tree structure, it does look very chaotic and as if there is no pattern to the way it all reduces to 1.  After noticing a pattern in the “evenness” of the numbers in the chains I thought of an idea.  I plotted the tree again, but on different axes.  I plotted the points such that the value p of each number in the tree can be found by the equation:

p=x*2y

In order to still maintain the visualization of generations, I made it into a gif as well, with each step showing a separate generation.

each point can be given by p=x*2^y
each point can be given by p=x*2^y

Then I made a few 20-generation plots to show how this structure looks on a grand scale:

20 generations of the N+1 tree
20 generations of the N+1 tree
20 generations of the N+1 tree, zoomed out to show all data
20 generations of the N+1 tree, zoomed out to show all data
20 generations zoomed in to show structure
20 generations zoomed in to show structure
20 generations zoomed in to show structure
20 generations zoomed in to show structure

 

 

 

Now I began to attack the 3N+1 problem using similar techniques.  First I figured out an algorithm that generates the reverse tree:

{If N’-1 is thirdable* and not even

N=13(N’-1)   AND     2*N’

Otherwise,

N=2*N’                                                }

(*Sidenote: I use the word “thirdable” to describe the quality of a number to be divided by 3 evenly, with no remainder.  Another way of saying this is that if a number is “thirdable” it has 3 as one of its factors)

From this I got a tree that looks like:

Generation tree for the 3N+1 tree
Generation tree for the 3N+1 tree
20 generations of the 3N+1 tree
20 generations of the 3N+1 tree

Again, you can see the branching tree-like organization, but it looks like chaos.

It did not do me any good to plot it on the same axes as I did for the N+1 algorithm, but after examining the ways the numbers branch off, I thought of a similar way to illustrate an analogous structure.  I plotted the tree in a 3-Dimensional plot, such that the value of each point p can be given by the equation:

p=x*2y*3z

Again, I made a gif of how this tree “grows” such that each step shows the next generation of branches.  This goes through 30 generations:

30 generations of the 3N+1 tree
30 generations of the 3N+1 tree

It was taking my computer a very long time to compute the higher generations so I wrote a different function which took N’ to be all integers less than a set value A.  Here is a gif which inputs A to be various high values, the largest being N=10000.

Reverse Collatz "tree" in web form, largest N=10000
Reverse Collatz “tree” in web form, largest N=10000

And here is a zoomed out perspective with the largest N being 500:

500

Though the appearance of the branches growing has dissolved, it is my opinion that this is an illustration of structure.  If this structure can be defined, it could lead to a proof of the Collatz Conjecture.

What do you all think, is there structure here?  Am I on to something?  I used Octave to create all of these plots, and then Photoshop to make them into gifs.  If anyone is interested in seeing the functions I wrote to create them, please contact me.

3 responses to “Structure to the Collatz Conjecture?”

  1. Gayle Avatar
    Gayle

    Nice to see you using your math degree. Love the pics the rest is over my head

  2. Stephanie Rivera Avatar
    Stephanie Rivera

    I’m in town. Maybe we should have lunch to discuss this?

  3. Jerome Steen Avatar
    Jerome Steen

    Great minds think alike. I am working on a proof using the set limits and frequencies in this very structure.
    Namely that the intersecting variables occur at the same rate as the source variables ie multiples of 3 and that odd numbers occur every 2 meaning that the base proceeds one third faster than it branches off. If you use the proof against Euclidean thirds it will prove that no loop exists other than 1421, and the pace proves no series will jump off into infinity.

Leave a reply to Stephanie Rivera Cancel reply