I posted some rushed Python calculations in another thread earlier, but now I’ve taken the time to think about the best way to test the difference method (the cycle elimination strategy from this thread which has been a central focus on the forum), and ran optimized code in C written by Claude to run as many calculations as I could.
I’m using the terminology and variable choice from the linked thread, which is essentially:
v - the “parity vector”, which is a binary string where 1 represents a \cfrac{3n+1}{2} step and 0 represents an \cfrac{n}{2} step.
\beta(v) - the parity summation applied to v
x - the number of odd steps in v
k - the total length of v
The program generates the irreducible rotationally unique vectors for each value of x, where k=ceil(x*log_2(3)), and for each v, calculates the magnitude of the difference between each pair of rotations, divides out all factors of 2 and 3 from these results, then finds the smallest one. Once it has found the smallest one for each v of that length, it then finds the largest of all of those. This is what I will refer to as the MMRBD (maximum minimum reduced beta difference). This represents the absolute worst case scenario for the difference method for each vector length, which makes it the limit for a generalized bound for ruling out cycles derived from this particular method (combining two beta values each with a coefficient of 1).
Here is the program output. Each five line chunk is the results for one vector length. The vector shown is the one that resulted in the MMRBD, rotated to its minimal position for uniformity (the minimal position being the rotation corresponding to the smallest beta). Two lines are placed in each vector to denote the two rotations that give the minimum reduced beta difference for that vector. For example, 11|011|00 means one of those rotations is 0110011 and the other rotation is 0011011. The last line, “greater than denominator”, counts how many vectors have a minimum reduced beta difference greater than 2^k-3^x and cannot therefore be ruled out as an integer cycle, and after the slash is the total number of irreducible rotationally unique vectors of that length.
x = 1, k = 2
MMRBD: 1
|1|0
denominator: 1
greater than denominator: 0/1
x = 2, k = 4
MMRBD: 1
1|1|00
denominator: 7
greater than denominator: 0/1
x = 3, k = 5
MMRBD: 1
11|1|00
denominator: 5
greater than denominator: 0/2
x = 4, k = 7
MMRBD: 5
11|011|00
denominator: 47
greater than denominator: 0/5
x = 5, k = 8
MMRBD: 7
1111|1|000
denominator: 13
greater than denominator: 0/7
x = 6, k = 10
MMRBD: 67
1|1110100|10
denominator: 295
greater than denominator: 0/20
x = 7, k = 12
MMRBD: 131
11|011110100|0
denominator: 1909
greater than denominator: 0/66
x = 8, k = 13
MMRBD: 413
|11011|11010100
denominator: 1631
greater than denominator: 0/99
x = 9, k = 15
MMRBD: 1447
11011|111011|0000
denominator: 13085
greater than denominator: 0/333
vectors tied for largest: 1
x = 10, k = 16
MMRBD: 3113
1|11101110100|1010
denominator: 6487
greater than denominator: 0/497
x = 11, k = 18
MMRBD: 14437
110|1|11110101110000
denominator: 84997
greater than denominator: 0/1768
x = 12, k = 20
MMRBD: 47239
1101|11110101100|10100
denominator: 517135
greater than denominator: 0/6288
x = 13, k = 21
MMRBD: 93161
111|01111101101010100|0
denominator: 502829
greater than denominator: 0/9690
x = 14, k = 23
MMRBD: 438943
|110111|11011010111000100
denominator: 3605639
greater than denominator: 0/35530
x = 15, k = 24
MMRBD: 924751
111110111101001010|01101|0
denominator: 2428309
greater than denominator: 0/54477
x = 16, k = 26
MMRBD: 3314077
1|1110|111010111111010001000
denominator: 24062143
greater than denominator: 0/204248
x = 17, k = 27
MMRBD: 7902995
111|1011111101100011010010|10
denominator: 5077565
greater than denominator: 334/312455
x = 18, k = 29
MMRBD: 26525755
111|0111110101111011010001100|0
denominator: 149450423
greater than denominator: 0/1193010
x = 19, k = 31
MMRBD: 118535341
|110111|1101101011110101101000100
denominator: 985222181
greater than denominator: 0/4552275
x = 20, k = 32
MMRBD: 243501565
11111011111101000101011010|01101|0
denominator: 808182895
greater than denominator: 0/7055776
x = 21, k = 34
MMRBD: 873236051
1111|01111110101110100101010110100|0
denominator: 6719515981
greater than denominator: 0/27293640
x = 22, k = 35
MMRBD: 1735810211
|1|1111011110101101110110101110000010
denominator: 2978678759
greater than denominator: 0/42181080
x = 23, k = 37
MMRBD: 7459818685
1101|1111010|11110110111010101000110100
denominator: 43295774645
greater than denominator: 0/165056400
x = 24, k = 39
MMRBD: 30769015709
1|10111110110|101111110101100110100010100
denominator: 267326277407
greater than denominator: 0/644636907
x = 25, k = 40
MMRBD: 66958939751
11|101111010111111011001101010101101000|10
denominator: 252223018333
greater than denominator: 0/1005633625
x = 26, k = 42
MMRBD: 230064011941
1|1110|1110101111110101101110110101010001000
denominator: 1856180682775
greater than denominator: 0/3964512336
x = 27, k = 43
MMRBD: 558697669033
1111|1011111110100011011101101010110000110|10
denominator: 1170495537221
greater than denominator: 0/6167026726
x = 28, k = 45
MMRBD: 1898355558433
1111|0111111011001011101011111010100010110100|0
denominator: 12307579633871
greater than denominator: 0/24512635642
x = 29, k = 46
MMRBD: 4945653403507
111|10111111010101011101101111010001011010010|10
denominator: 1738366812781
greater than denominator: 33504124/38036848410
I had the program check if a given vector length had more than one vector that shared the MMRBD, and this was only the case for x = 3, 4, and 5, each of which had a two-way tie. I also ran the program a second time just for a couple minutes and had it record the top eight MMRBD and their vectors. They seem to typically be all more than 0.9 times the highest one. I can share this output if anyone’s interested.
Here is my quick analysis on the MMRBD trend. This is where the more advanced math around determining a general bound comes in and I would like to leave that for someone else, but I made a plot which shows that the ratio between MMRBD and 3^x seems to hover between 0.05 and 0.1 for this range with a slightly increasing trend. This explains why two of the vector lengths (at x = 17 and x = 29) have an MMRBD greater than the denominator. Those two denominators are only 3.9% and 2.5% of 3^x respectively.
Edit: Note that I excluded x = 1, 2, and 3 from the plot since they seem to be outliers from being so small, and including them zooms the plot out unnecessarily.
Here are the raw MMRBD values if anyone wants to do their own analysis on them:
1, 1, 1, 5, 7, 67, 131, 413, 1447, 3113, 14437, 47239, 93161, 438943, 924751, 3314077, 7902995, 26525755, 118535341, 243501565, 873236051, 1735810211, 7459818685, 30769015709, 66958939751, 230064011941, 558697669033, 1898355558433, 4945653403507
The program runtime was around 7 hours. I would run it again if there are suggestions for changes to make, or I can paste the code. Each new x increases the runtime by 2-3 times. Shortcuts could definitely be made but I wanted to start as exhaustive as possible. I did take one big shortcut already which is k=ceil(x*log_2(3)) and I believe it’s a completely reasonable restriction but it’s worth emphasizing.






