A Conjecture on the Collatz-Kakutani Path Length for the Mersenne Primes

Toru Ohira and Hiroshi Watanabe

We present here a new conjecture for the nature of the Mersenne prime numbers by connecting it with the Collatz-Kakutani problem. By introducing a natural path length on the basis of the Collatz-Kakutani tree, we conjecture that this path length of a Mersenne prime from the root of the Collatz-Kakutani tree is approximately proportional to the index of the Mersenne prime. We also discuss difference of behaviors between Mersenne numbers and Mersenne primes.

1 Introduction

Prime numbers have attracted mathematically oriented minds. They are fountains of the interesting problems which are left unresolved to the present. As well as prime numbers, natural numbers themselves sometimes show unexpected behaviors in spite of their simple appeal. In this note, we would like to present a simple property arising from the combination of two unsolved problems in number theory; the Mersenne prime numbers and the Collatz-Kakutani conjecture. Namely, we show that there exists an approximate linear relation between an index of the Mersenne prime and its Collatz-Kakutani path length, both of which are defined in the next section.

2 Mersenne primes and Collatz-Kakutani Conjecuture

Let us start with a brief descriptions of the Mersenne primes and the Collatz-Kakutani conjecture[1]. A Mersenne number is a positive integer given by


where nn is a positive integer, called “index”. A Mersenne prime is a Mersenne number that is prime. It has been shown that if MnM_{n} is a Mersenne prime, nn must be prime. However, the converse is not true. In fact, only forty-seven Mersenne primes have been found up to date with the largest one given when n=43,112,609n=43,112,609. This is also the largest known prime number. There are many fundamental questions left unresolved such as whether there are infinitely many Mersenne primes. The behaviors of Mersenne primes are irregular and unpredictable as seen in other types of prime numbers. Therefore, the recent discoveries of the Mersenne primes were achieved with the help of massively distributed computers, and the effort for finding new Mersenne primes has been made continuously [2]. The Mersenne primes are known for their relationships with the perfect numbers, and they are also applied to create pseudorandom number generators [3].

The Collatz-Kakutani conjecture is also one of the unsolved problems in number theory. This conjecture is a halting problem that the following operations will stop for an arbitrary positive number. Consider a positive integer XX.

If XX is odd, 3X+13X+1 is the next integer.

If XX is even, X/2X/2 is the next integer.

If we repeat this process, it will eventually reach the halting state X=1X=1 for all positive integers. For example, if we start from X=7X=7, we have a sequence as


This process can be visualized as the Collatz-Kakutani tree as shown in Fig. 1A. The conjecture states that this tree covers all the positive integers. Even though this conjecture is unsolved, it is generally believed as true from arguments through probability theories and through computational verifications.

Let us introduce the “Collatz-Kakutani path length” D(X)D(X) for a number of steps for XX to reach X=1X=1 through the above operations, i.e., a number of operational steps needed to reach 11 (“the root”) on the Collatz-Kakutani tree. For example, D(7)=16D(7)=16. The relationship between XX and D(X)D(X) is highly irregular and no simple law for the path length is found11In fact, it may not be defined for some XX if the Collatz-Kakutani conjecture is false. As a result, the plot of D(X)D(X) versus XX produces a irregular graph as shown in Fig. 1B. While the general relationship between XX and D(X)D(X) is unknown, there are some trivial relations such as D(2n)=nD(2^{n})=n. Note that, the number 2n2^{n} is just one larger than the Mersenne number Mn=2n-1M_{n}=2^{n}-1.

Figure 1: A. Schematic View of Collatz-Kakutani tree. B. Plot of the path length D(X)D(X) on the Collatz-Kakutani tree to the root of tree X=1X=1 for integers up to X=2000.

3 Main Results

Our main finding to report is the fact that a path length of a Mersenne prime is approximately proportional to its index for large n, namely,

D(Mn)13.45n.D(M_{n})\approx 13.45n. (1)

This is shown in Fig. 2, which we were able to compute up to the 38th Mersenne prime, Mp(38)=M6972593{M_{p}}(38)=M_{6972593}. The behavior of the path length DD is non monotonic for small indices, e.g., D(M89)>D(M107)D(M_{89})>D(M_{107}). We expect, however, beyond n=107n=107, the path length increases monotonically with nn for the Mersenne primes. Also, there seems no regular path to reach the root of the tree, X=1X=1. For example, we have examined the path for two nearby Mersenne primes, Mp(16)=M2203{M_{p}}(16)=M_{2203} and Mp(17)=M2281{M_{p}}(17)=M_{2281}. While they have the path length DD of approximately 30,00030,000, they share the last 16 steps, namely,

22,11,34,17,52,26,13,40,20,10,5,16,8,4,2,22,11,34,17,52,26,13,40,20,10,5,16,8,4,2, and 11.

The relations between indices of the Mersenne primes and their path lengths are summarized in Table 1.

Number Indices D(X)D(X) D(X)/n{D(X)/n}
1 2 7 3.5
2 3 16 5.33333
3 5 106 21.2
4 7 46 6.57143
5 13 158 12.1538
6 17 224 13.1765
7 19 177 9.31579
8 31 450 14.5161
9 61 860 14.0984
10 89 1454 16.3371
11 107 1441 13.4673
12 127 1660 13.0709
13 521 6769 12.9923
14 607 8494 13.9934
15 1279 17094 13.3651
16 2203 29821 13.5365
17 2281 30734 13.4739
18 3217 43478 13.5151
19 4253 55906 13.1451
20 4423 60716 13.7273
21 9689 129608 13.3768
22 9941 134345 13.5142
23 11213 153505 13.6899
24 19937 265860 13.335
25 21701 293161 13.5091
26 23209 312164 13.4501
27 44497 598067 13.4406
28 86243 1158876 13.4373
29 110503 1482529 13.4162
30 132049 1771117 13.4126
31 216091 2906179 13.4489
32 756839 10197081 13.4732
33 859433 11568589 13.4607
34 1257787 16927967 13.4585
35 1398269 18807193 13.4503
36 2976221 40055567 13.4585
37 3021377 40663017 13.4584
38 6972593 93778449 13.4496
Table 1: Path lengths of the Mersenne primes.

This linear behavior can be understood by the following heuristic arguments. Suppose NN is a large random number. The standard heuristic arguments for the path length for NN gives

D(N)(3/(ln4/3))lnN.D(N)\approx(3/(\ln 4/3))\ln N. (2)

A Mersenne number 2n-12^{n}-1 becomes 3×2n-1-13\times{2^{n-1}}-1 after two steps in Collatz-Katutani tree. Therefore, A Mersenne Number 2n-12^{n}-1 becomes 3n-13^{n}-1 after 2n2n steps. If we consider the number N=3n-13nN=3^{n}-1\sim 3^{n} to be a large random number for large nn, then the path length of the number is given by

D(3n-1)(3/(ln4/3))nln3.D(3^{n}-1)\approx(3/(\ln 4/3))n\ln 3. (3)

Finally, we obtain the heuristic estimation of the path length of the Mersenne number to be

D(2n-1)2n+(3/(ln4/3))nln313.45652n,D(2^{n}-1)\approx 2n+(3/(\ln 4/3))n\ln 3\approx 13.45652n, (4)

which is consistent with our numerical results [4, 5, 6]. This arguments implies that our finding is true not only for the Mersenne primes, but also for the Mersenne numbers. Our main conjecture, however, is that this linear relationship for the Mersenne primes are better than that for the general Mersenne numbers, or other general sequence of integers.

In order to show the difference between the Mersenne numbers in general and the Mersenne primes, the ratio of the path length to the index, D(Mn)/nD(M_{n})/n, is plotted against nn in Fig. 2B. If the distance increases linearly to the index, then D(Mn)/nD(M_{n})/n becomes a constant. Figure 2B shows that the Mersenne primes show better linearity than the general Mersenne numbers.

Figure 2: A. Plot of D(Mn)D(M_{n}) for Mersenne primes with index nn up to around Mp(38){M_{p}(38)}. B. Plot of D(Mn)/nD(M_{n})/n for the Mersenne primes (Blue) and the Mersenne numbers with randomly chosen index (Red) up to around Mp(31){M_{p}(31)}.

To be more quantitative, we have performed the following statistical analyses. We have chosen thirteen Mersenne primes from the 26th26^{\text{th}} (Mp(26)=M23,209{M_{p}}(26)=M_{23,209}) to the 38th38^{\text{th}} (Mp(38)=M6,972,593{M_{p}}(38)=M_{6,972,593}), and computed the mean and the variance of D(Mn)/nD(M_{n})/n. Let us call the prime indices of the Mersenne primes as the Mersenne prime indices. We compare the result with that from the following set of thirteen points approximately in the same interval.

A) Thirteen prime indices which is the next smallest to thirteen Mersenne prime indices. We note the Mersenne numbers associated with this set is not prime numbers themselves.

B) Thirteen indices which are near the mid-point of two successive Mersenne prime indices.

C) Thirteen indices which are twice the Mersenne prime indices. These are all even numbers. We have selected some which are close to the Mersenne prime indices.

D) Thirteen indices which is on the least squre best-fit line based on the heuristics that, given the kk-th Mersenne prime Mp(k){M_{p}}(k), the plot of log2(log2(Mp(k))\log_{2}(\log_{2}({M_{p}}(k)) versus kk lies approximately on the straight line (Figure 3). We have selected some which are close to the Mersenne prime indices [7].

Figure 3: Plot of log2(log2(Mp(k))\log_{2}(\log_{2}({M_{p}}(k)) versus kk for all the known Mersenne prime up to Mp(47){M_{p}(47)}. The straight line is obtained by the least square error fit and 0.92757+0.55715k\sim 0.92757+0.55715k.

The results of these comparison are shown in Table 2.

Set Indices D(X)D(X) E[D(X)/n][D(X)/n] Var[D(X)/nD(X)/n]
Mersenne Prime 23209, 44497, 86243, 110503, 132049, 216091, 756839, 859433, 1257787, 1398269, 2976221, 3021377, 6972593 312164, 598067, 1158876, 1482529, 1771117, 906179, 10197081, 11568589, 16927967, 18807193, 40055567, 40663017, 93778449 13.4473 0.0002977
A 23227, 44501, 86249, 110527, 132059, 216103, 756853, 859447, 1257827, 1398281, 2976229, 3021407, 6972607 312182, 598071, 1158882, 1482553, 1771127, 2906191, 10197095, 11568603, 16928007, 18807205, 40055575, 40663047, 93778463 13.4460 0.0003194
B 22455, 33853, 65370, 98373, 121276, 174070, 486465, 808136, 1058610, 1328028, 2187245, 2998799, 4996985 299801, 457438, 875438, 1327329, 1633743, 2344640, 6524449, 10868120, 14246657, 17876449, 29428265, 40364153, 67195624 13.4485 0.0017853
C 22426, 43402, 46418, 88994, 172486, 221006, 264098, 432182, 1513678, 1718866, 2515574, 2796538, 5952442 299772, 584422, 627877, 1201650, 2320161, 2974984, 3556035, 5828307, 20384499, 23124964, 33827530, 37632788, 80085173 13.4618 0.00132591
D 20160, 43644, 64216, 94484, 139021, 204550, 442830, 651562, 958682, 1410567, 2075452, 3053739, 4493150 270868, 587280, 866299, 1265873, 1871202, 2748585, 5947053, 8769774, 12911249, 18991590, 27939124, 41095221, 60441877 13.4515 0.000502943
Table 2: Statistical analyses on path lengths of Mersenne primes and Mersenne numbers. E[xx] denotes the arithmetic average of xx and Var[xx] denotes the variance of xx, respectively.

The data summarized in Table. 2 show that the path lengths of the Mersenne primes have the smaller variance than other sets. However, this difference is not enough to single out a Mersenne prime index.

In spite of the above, we look more detail into this property. We have examined the plot of D(X=2n-1)D(X=2^{n}-1) versus nn more closely and noted that they are composed of a collection of “flat” regions with jumps like a staircase. Correspondingly, the plot of D(X)/n{D(X)/n} versus nn becomes collection of stripes, which interestingly appears near parallel. They are plotted in Figure 4. These plots are made for the sampled prime indices near three Mersenne prime indices, 23209,110503,21609123209,110503,216091 which give 26th26^{\text{th}}, 29th29^{\text{th}}, and 31th31^{\text{th}} Mersenne primes respectively. It is also interesting that there are occasional “overlaps” of “stair steps”.

Figure 4: Plot of D(Mn)D(M_{n}) (top) and D(Mn)/nD(M_{n})/n (bottom) versus nn. The Mersenne primes are marked by arrows, and horizontal line is at D(Mn)/n=13.45D(M_{n})/n=13.45. A. Prime indices near n=23209n=23209 (Mp(26)){M_{p}(26)}), B. Every fifth prime indices near n=110503n=110503 (Mp(29){M_{p}(29)}), C. Every fifth prime indices near n=216091n=216091 (Mp(31)){M_{p}(31)}).

The nature and mechanism of these properties appeared in Figure 4 need to be studied further. We, nevertheless, conjecture that these qualitative behavior will continue to larger Mersenne numbers, and the Mersenne prime appears with its D(X)/n13.45{D(X)/n}\approx 13.45. Unfortunately, again, this is clearly not sufficient to select out the Mersenne primes. If this conjecture is true, however, we can heuristically rule out points that are further away from the horizontal line of D(X)/n13.45{D(X)/n}\approx 13.45 from candidates of the Mersenne primes.

4 Discussion

We have used the Collatz-Kakutani tree as a provider of “length” for natural numbers, and obtained a conjecture for the Mersenne primes. This may be yet another example where prime numbers can give rise to an emergence of unexpected order, such as Ulam Spiral [8, 9].

Though the reason behind such simple behavior is left unclear, this approach of using the Collatz-Kakutani tree can be extended to other types of set of numbers known for irregularities, possibly leading to interesting insights.


Authors would like to thank Dr. Masaru Suzuki, Dr. Atsushi Kamimura, and Shigenori Matsumoto for useful discussions. T.O. is also affiliated with Sony Computer Science Laboratories in Tokyo, and would like to thank Emeritus Prof. Yuji Ito of Keio University, for his account on his advisor, late Prof. Shizuo Kakutani of Yale University.


  • 1 The Math Forum Internet Mathematics Library, available at http://mathforum.org/library/.
  • 2 Great Internet Mersenne Prime Search:GIMPS, available at http://www.mersenne.org/default.php/.
  • 3 M. Matsumoto, T. Nishimura, Mersenne twister: A 623-dimensionally equidistributed uniform pseudorandom number generator, ACM Trans. on Modeling and Computer Simulations 8 (1998) 3.
  • 4 This heuristic arguments of this paragraph is provided by one of the anonymous reviewers.
  • 5 J. C. Lagarias, The 3x+13x+1 Problem and Its Generalizations, Amer. Math. Monthly 92 (1985) 3
  • 6 A. V. Kontorovich, J. C. Lagarias, Stochastic Models for the 3x+13x+1 and 5x+15x+1 Problems, arXive:09101944 [math.NT] (2009)
  • 7 M. R. Schroeder, Where Is the Next Mersenne Prime Hiding? Mathematical Intelligencer 5 (1983) 31
  • 8 M. L. Stein, S. M. Ulam, M. B. Wells, A Visual Display of Some Properties of the Distribution of Primes, Amer. Math. Monthly 71 (1964) 516
  • 9 M. L. Stein, S. M. Ulam, An Observation on the Distribution of Primes, Amer. Math. Monthly 74 (1967) 43.

Shirokanedai Institute, Tokyo, Japan.

The Institute of Solid State Physics, The University of Tokyo, Kashiwa, Japan.