The 3x + 1 Problem: An Annotated Bibliography, II

(2000-2009)
J. C. Lagarias
Department of Mathematics
University of Michigan
Ann Arbor, MI 48109–1109
arXiv:math/0608208v6 [math.NT] 12 Feb 2012

[email protected]

(Jan. 10, 2012)

ABSTRACT. The 3x + 1 problem concerns iteration of the map T : Z → Z given by

3x + 1

 if x ≡ 1 (mod 2) .
2


T (x) =
 x
if x ≡ 0 (mod 2) .


2
The 3x + 1 Conjecture asserts that each m ≥ 1 has some iterate T (k) (m) = 1. This is the
second installment of an annotated bibliography of work done on the 3x+1 problem and related
problems, mainly covering the period 2000 through 2009, with some related later papers (which
were preprints by 2009). At present the 3x + 1 Conjecture remains unsolved.

1. Introduction

The 3x + 1 problem is most simply stated in terms of the Collatz function C(x) defined
on integers as “multiply by three and add one” for odd integers and “divide by two” for even
integers. That is, 
 3x + 1 if x ≡ 1 (mod 2) ,

C(x) =
x
if x ≡ 0 (mod 2) ,


2
The 3x + 1 problem (or Collatz problem) is to prove that starting from any positive integer,
some iterate of this function takes the value 1. The problem other names: it has also been
called Kakutani’s problem, the Syracuse problem, and Ulam’s problem.
Much work on the problem is stated in terms of the 3x + 1 function
3x + 1

 if x ≡ 1 (mod 2)
2


T (x) =
 x
if x ≡ 0 (mod 2) .


2
The 3x + 1 Conjecture states that every m ≥ 1 has some iterate T (k) (m) = 1.
The 3x + 1 Conjecture has now been verified up to 17 × 258 > 4.899 × 1018 (as of Feb.
21, 2008) by an ongoing computation run by T. Oliveira e Silva (2004+). An independent
computation of Roosendaal(2004+) verifies it to 612 × 250 > 6.89 × 1017 .

1
 At present the 3x + 1 conjecture remains unsolved. The proofs claimed in Yamada (1981),
Cadogan (2006) and Bruckman (2008) are incomplete.
Surveys on results on the 3x + 1 problem can be found in Lagarias (1985), Müller (1991),
and the first chapter of Wirsching (1998a), described in the first installment of the annotated
bibliography, Lagarias (2003+). A more recent survey appears in Chamberland (2003).

2. Terminology

We use the following definitions. The trajectory or foward orbit of an integer m is the set

O+ (m) := {m, T (m) , T (2) (m), . . .} .

The stopping time σ(m) of m is the least k such that T (k) (m) < m, and is ∞ if no such k exists.
The total stopping time σ∞ (m) is the least k such that m iterates to 1 under k applications
of the function T i.e.
σ∞ (m) := inf {k : T (k) (m) = 1}.
The scaled total stopping time or gamma value γ(m) is given by

σ∞ (m)
γ(m) := .
log m

The height h(m) is the least k for which the Collatz function C(x) has C (k) (m) = 1. It is also
given by
h(m) := σ∞ (m) + d(m),
where d(m) counts the number of iterates T (k) (m) ≡ 1 (mod 2) for 0 ≤ k < σ∞ (m). Finally,
the function πa (x) counts the number of n with |n| ≤ x that contain a in their forward orbit
under T .

3. Bibliography

This bibliography covers research articles, survey articles and PhD theses on the 3x + 1
problem and related problems from 2000 to the present. The first installment of the annotated
bibliography is Lagarias(2003+), which covers the period 1963–1999. Articles in Chinese have
the authors surname listed first.

1. Ethan Akin (2004), Why is the 3x + 1 Problem Hard?, In: Chapel Hill Ergodic Theory
Workshops (I. Assani, Ed.), Contemp. Math. vol 356, Amer. Math. Soc. 2004, pp.
1–20. (MR 2005f:37031).

This paper analyzes the 3x + 1 problem by viewing the map T as acting on the
domain Z2 of 2-adic integers. The map T is topologically conjugate over Z2 to the 2-adic
shift map
x−1


 2
 if x ≡ 1 (mod 2) ,
S(x) =
 x
if x ≡ 0 (mod 2) ,


2

2
 by a conjugacy map Q3 : Z2 → Z2 , i.e. Q3 ◦ T = S ◦ Q3 . (The map Q3 equals the map
denoted Q∞ in Lagarias (1985), and is the inverse of the map Φ in Bernstein (1994).)
The 3x + 1 Conjecture can be reformulated in terms of the behavior of Q3 acting on
integers, namely that Q3 maps Z+ into 13 Z. Consider more generally for any odd rational
a the map Ta (x) which sends x 7→ ax+1 2 or x2 , according as x is an odd or even 2-adic
integer. The author observes there is an associated conjugacy map Qa : Z2 → Z2 with
the same property as above, and formulates the Rationality Conjecture for Qa , which
asserts that Qa maps the rationals with odd denominators to rationals. He shows that
the Rationality conjecture is true for a = ±1 and is false for any odd rational a that is not
an integer. For the remaining cases of odd integer a, where the Rationality Conjecture
remains unsolved, he presents a heuristic argument suggesting that it should be true for
a = ±3 and false for all odd integers |a| ≥ 5.
2. João F. Alves, Mário M. Graca, M. E. Sousa Dias, and José Sousa Ramos (2005), A lin-
ear algebra approach to the conjecture of Collatz, Lin. Alg. Appl. 394 (2005), 277–289.
(MR2100588)

This paper studies the conjecture that the only periodic orbit of the Collatz map on
the positive integers goes through n = 1. They form an n × n zero-one matrix An whose
entries are
Ai,j = 1 if T (i) = j, 1 ≤ i, j ≤ n.
where T (n) is the 3x + 1 function, and Ai,j = 0 otherwise. The assertion that {1, 2}
is the only periodic orbit of T on the positive integers is shown to be equivalent to
det(I − xAn ) = 1 − x2 for all n ≥ 1. They prove that det(I − xAn ) = det(I − xAn−1 ) for
all n 6= 8 (mod 18). They deduce that if there is another periodic orbit on the positive
integers then there exists m ≡ 8 (mod 18) such that n = m 2 is in a periodic orbit.
Various further conditions are deduced in the case n ≡ 8(mod 18), e.g. det(I − xAn ) =
det(I − xAn−1 ) if n ≡ 8 (mod 54).
3. Tewodrus Amdeberhan, Dante Manna and Victor H. Moll (2008), The 2-adic valuation
of a sequence arising from a rational integral, J. Combinatorial Theory, Series A, 115
(2008), no. 8, 1474–1486. (MR 2009k:11194)

This paper studies certain integer sequences {Ak,m : k ≥ 0} arising from evalua-
tion of the integral Z ∞
dx
N0,4 (a; m) =
0 (x + 4ax2 + 1)m+1
4

expanded in Taylor series in the parameter a, as

∞
π X ak
N0,4 (a; m) = √ m+1 Ak,m .
2m!(4(2a + 1)) 2 k!
k=0

These sequences are given by

m    
k!m! X j 2m − 2j m+j j
Ak,m = m−k 2 .
2 m−j m k
j=k

In section 6 a relation is shown between divisibility of A1,m by powers of 2 and the 3x + 1

problem. Namely am := ord2 (A1,m ) − 1 gives the number of iterations of the 3x + 1 map

3
 T (x) starting from x0 = m, in which the parity of the iterates does not change, i.e
m ≡ T (m) ≡ · · · ≡ T am −1 (m) 6≡ T am (m) (mod 2).
This is given as Theorem 6.1 of the paper.
4. Paul Andaloro (2000), On total stopping times under 3X + 1 iteration, Fibonacci Quar-
terly 38 (2000), 73–78. (MR 2000m:11024).

This paper shows various results on the minimal elements having a given stop-
ping time, where the “stopping time” is defined to be the number of odd elements in the
trajectory up to and including 1. It also obtains a new congruential “sufficient set” cri-
terion to verify the 3x + 1 Conjecture. It shows that knowing that the 3x + 1 Conjecture
is true for all n ≡ 1 (mod 16) implies that it is true in general.
5. Paul Andaloro (2002), The 3X + 1 problem and directed graphs, Fibonacci Quarterly 40
(2002), 43–54. (MR 2003a:11018).

This paper considers various “compressed” versions of the 3x + 1 graph, in which

only a subset of the vertices are retained with certain directed paths in original 3x + 1
graph iterates of T (·) replaced by single directed edges. The initial “compressed” graph
corresponds to odd integers, and the paper introduces two further “compressed” graphs
with fewer allowed vertices. In each case, the 3x+ 1 Conjecture is equivalent to the graph
being weakly connected, i.e. being connected when viewed as an undirected graph. The
paper shows that certain kinds of vertex pairs in such graphs are weakly connected,
typically for allowed vertices in certain congruence classes (mod 2k ) for small k.
6. Stefan Andrei, Manfred Kudlek, Radu Stefan Niculescu (2000), Some results on the Col-
latz problem, Acta Informatica 37 (2000), 145–160. (MR 2002c:11022).

This paper studies the internal structure of 3x + 1-trees. Among other things, it
observes the that integers of the form 2m k − 1 eventually iterate to the integers 3m k − 1,
respectively. and that integers of the form 23m k − 5 iterate to the integers 32m k − 5, and
integers of the form 211m k − 17 iterate to the integers 37m k − 17. These facts are related
to the cycles associated to −1, −5 and −17, respectively.
7. Stefan Andrei, Wei-Ngan Chin, and Huu Hai Nguyen (2007+), A Functional View over
the Collatz’s Problem, preprint.

This paper describes chains in the 3x + 1 tree using a phrase structure grammar.
8. David Applegate and Jeffrey C. Lagarias (2003), Lower bounds for the total stopping time
of 3x + 1 iterates, Math. Comp. 72 (2003), 1035–1049. (MR 2004a:11016).

This paper proves there are infinitely many positive n which have a finite total
stopping time σ∞ (n) > 6.14316 log n. It also shows that there is a positive c such that at
least cx1/60 of all integers 1 < n ≤ x have a finite total stopping time σ∞ (n) > 5.9 log n.
The proofs are computer-intensive, and produce a “certificate” encoding a proof, which is
based on a search of 3x + 1 trees to depth 60. The “certificates” are quite large, involving
about 350 million trees for the lower bound 6.14316 log n, which corresponds to a density
of odd integers in a trajectory (the” ones-ratio”) of 14
29 ≈ 0.483.

4
 This rigorous bound is below the bound σ∞ (n) ≈ 6.95212 log n that one expects to hold
for almost all integers, which corresponds to a ones-ratio of 21 . The paper gives heuristic
arguments suggesting that the method of this paper might prove σ∞ (n) ≈ 6.95212 log n
holds for infinitely many n, but that it would likely require a search of 3x + 1 trees to
depth at least 76. This would require a very large computation.

9. David Applegate and Jeffrey C. Lagarias (2006), The 3x + 1 semigroup, J. Number The-
ory 177 (2006), 146–159. (MR 2006k:11037).

This paper considers a weak version of the 3x + 1 problem proposed by Farkas

(2005). It considers the multiplicative semigroup R of positive rational numbers gener-
2n+1
ated by { 3n+2 : n ≥ 0} together with {2}. The weak 3x + 1 conjecture asserts that this
semigroup contains all positive integers. The relation to the 3x + 1 problem is that the
semigroup contains 1 and its generators encode the action of the inverse 3x + 1 function.
It follows that the truth of the 3x + 1 conjecture implies the truth of the weak 3x + 1
conjecture. This paper proves the conjecture. Its main result shows that the semigroup
R consists of all positive rational numbers ab such that 3 does not divide b. The proof is
an induction motivated by certain results established in Lagarias (2006).

10. Edward Belaga and Maurice Mignotte (2000), Cyclic Structure of Dynamical Systems
Associated with 3x + d Extensions of Collatz Problem, U. Strasbourg report 2000-18, 57
pages. (http://hal.archives-ouvertes.fr/IRMA-ACF, file hal-00129656)

This paper is a theoretical and experimental study of the distribution of number

and lengths of finite cycles to the 3x + d map, for d ≡ ±1 ( mod 6). It contains much data
and a wealth of interesting results and conjectures.

11. Edward Belaga (2003), Effective polynomial upper bounds to perigees and numbers of
(3x + d)-cycles of a given oddlength, Acta Arithmetica 106, No. 2, (2003), 197–206.
(MR 2003m:11120).

Let d be a positive odd integer, and consider the 3x + d map Td (x) = 3x+d2 if x
x
is odd; Td (x) = 2 if x is even, acting on the domain of positive integers. This paper
shows that for any cycle C of the 3x + d map of length l containing k odd elements, the
smallest element prg(C) in the cycle satisfies
d
prg(C) ≤ .
2l/k−3
From this follows
length(C)
log2 3 < ≤ log2 (d + 3)
oddlength(C)
The author shows that the upper bound is sharp, and gives evidence that the lower
bound is probably asymptotically approachable. Using bounds from transcendence the-
ory (linear forms in logarithms) the author gives an upper bound for the total number
of cycles Ud,k of odd-length k. This upper bound is dkc0 , for a constant c0 , and states
that one may take c0 = 32. He also shows that the largest element S of any such cycle
is bounded above by
3
S < dk c0 ( )k .
2

5
12. Edward Belaga and Maurice Mignotte (2006a), Walking Cautiously into the Collatz
Wilderness: Algorithmically, Number Theoretically, and Randomly, Fourth Colloquium
on Mathematics and Computer Science, DMTS (Discrete Mathematics and Computer
Science) Proceedings AG, 2006, 249–260.

This paper discusses many open questions about the 3x + 1 map and related maps,
and recent new numerical evidence supporting them. It formulates some new conjectures.

13. Edward Belaga and Maurice Mignotte (2006b), The Collatz problem and its generaliza-
tions: Experimental Data. Table 1. Primitive cycles of 3x + d mappings, Univ. of
Strasbourg preprint 2006-015, 9 pages+400+ page table .

This paper gives detailed tables of primitive cycles on the positive integers for 3x + d
maps for 1 < d < 20000. The authors conjecture they obtain the complete list of such
cycles, for these values of d. The text prior to the table formulates some interesting new
conjectures.

14. Edward Belaga and Maurice Mignotte (2006c), The Collatz problem and its generaliza-
tions: Experimental Data. Table 2. Factorization of Collatz numbers 2l − 3k , Univ. of
Strasbourg preprint 2006-018, 6 pages text+156 page table.

These tables give known divisors of numbers of form D = 2l − 3k for l ≤ 114.

Numbers D of this form give 3x + D problems having many primitive cycles.

15. Vitaly Bergelson, Michal Misiurewicz and Samuel Senti (2006), Affine Actions of a Free
Semigroup on the Real Line, Ergodic Theory and Dynamical Systems 26 (2006), 1285–
1305. MR2266362 (2008f:37019).

This extends the analysis of Misiurewicz and Rodrigues (2005) to semigroups gen-
eralizing those associated to the 3x + 1 map. The current paper considers the orbits
of a semigroup generated by T0 (x) = ax, T1 (x) = bx + 1, in which 0 < a < 1 < b,
viewed as acting on the positive real numbers R≥0 . It defines various notions of “uniform
distribution” on the positive real axis, and derives various results concerning uniform dis-
tribution of orbits of such semigroups. In particular, if the iterates of the transformation
are decreasing on average, and if one assumes the symbolic dynamics of T0 , T1 are drawn
from an ergodic invariant measure ν on the shift space, then it proves the existence and
uniqueness of an invariant measure µ on R≥0 , which depends on ν. See Theorem F of
the paper for a precise statement.

16. Konstantin Borovkov and Dietmar Pfeifer (2000), Estimates for the Syracuse problem
via a probabilistic model, Theory of Probability and its Applications 45, No. 2 (2000),
300–310. (MR 1 967 765).

This paper studies a multiplicative random walk imitating the 3x + 1 iteration. Let
X0 be given and set Yj = X0 X1 · · · Xj where each Xi for i ≥ 1 are i.i.d. random variables
assuming the value 12 or 32 with probability 21 each. Let σ∞ (X0 , ω) be a random variable
equal to the smallest J such that Yj < 1. Then E[σ∞ (X0 , ω)] = ( 12 log 43 )−1 log X0 and

6
 the normalized variable
σ∞ (X0 , ω) − c1 log X0
σ̂∞ (X0 , ω) =
c2 (log X0 )1/2
3/2
with c1 = ( 21 log 34 )−1 = 6.95212 and c2 = c1 ( 21 log 3), has a distribution converging to a
standard normal distribution as X0 → ∞ (Theorem 5). Various refinements of this result
are given. Comparisons are made to empirical 3x + 1 data for values around n ≈ 106 ,
which show good agreement.

17. Barry Brent (2002+), 3X + 1 dynamics on rationals with fixed denominator, eprint:
arXiv math.DS/0204170.

This paper reports on computer experiments looking for cycles with greatest com-
mon divior 1 for the 3x + k problem, for various k ≡ ±1(mod 6). It suggests that for
k = 7, k = 19 and k = 31 there is only one such cycle on the positive integers. Various
other statistics are reported on.

18. Thomas Brox (2000), Collatz cycles with few descents, Acta Arithmetica 92 (2000),
181–188. (MR 2001a:11032)

This paper considers the variant of the 3x + 1 map, call it T1 , that divides out
all powers of 2 at each step. A cycle is written {x1 , x2 , ...xm } where each xi is an odd
integer. A descent means |T1 (x)| < |x|. Let |C| denote the number of (odd) elements in
a 3x + 1 cycle C, and let d(C) denote the number of descents in C. The author proves
that the the number of 3x + 1 cycles C that have d(C) < 2log|C| is finite. In particular
the number of Collatz cycles with d(C) < r is finite for any fixed r. The proof uses a
bound of Baker and Feldman and is in principle effective. Thus for each r there exists
an algorithm to determine all cycles with d(C) < r. This greatly strengthens the result
of R. Steiner (1978) who detemined all cycles with one descent.

19. Paul S. Bruckman (2008), A proof of the Collatz conjecture, International Journal of
Mathematical Education in Science and Technology, 39, No. 3 (2008), 403–407. [Erra-
tum: 39, No. 4 (2008), 567.] (MR 2009d:11043b)

This paper asserts a proof of the Collatz conjecture. However the argument given
has a gap which leaves the proof incomplete. The erratum points out this gap and with-
draws the proof.

The gap is, suppose N0 is the starting value, and that Nk is the k-th odd iterate to occur.
Let Ek denote the number of divisions by 2 that occur in reaching Nk , then 2Ek Nk −
Pk−1
k
3 N0 = Sk , where Sk is the positive integer Sk = Ej k−1−j . Next determine
j=0 2 3
(Ak , Bk ) by requiring 2Ek Bk − 3k Ak = 1, with 0 ≤ Bk < 3k . The author notes that there
is an integer Tk such that

N0 = Ak + Tk 2Ek , Nk = Bk + Tk 3k .

Here Tk depends on k and may be positive or negative. The author then argues by
contradiction, asserting in Section 2 the claim that the minimal counterexample N0

7
 must have 2Ek < 3k for all k ≥ 1, which would imply that the sequence of iterates of N0
diverges. The argument justifying this claim has a gap, it supposes 2Ek > 3k , and asserts
the contradiction that Nk < N0 . But Nk ≥ N0 may hold if Tk is sufficiently negative,
and in fact for every starting value n0 ≥ 1 the values Tk → −∞ as k → ∞. This can
be seen from the equation (11) saying that n0 = Ak Sk + Tk 2Nk , noting that n0 is fixed,
Sk → ∞ by its recurrence Sk+1 = 3Sk + 2Ek and Ak ≥ 1. (The author’s argument as
given would prove the cycle starting at 1 does not exist.)

20. Charles C. Cadogan (2000), The 3x + 1 problem: towards a solution, Caribbean J. Math.
Comput. Sci. 10 (2000), paper 2, 11pp. (MR 2005g:11032)

The paper studies trajectories of the 3x + 1 problem. calling two integers n1 and
n2 equivalent, written n1 ∼ n2 , if their trajectories eventually coalesce. Various results
are obtained giving sufficient conditions for equivalence. The author conjectures that for
each positive odd integer n, 9n + 4 ∼ 3n + 1. His main result is that this conjecture
implies the truth of the 3x + 1 Conjecture.

21. Charles C. Cadogan (2003), Trajectories in the 3x+1 problem, J. of Combinatorial Math-
ematics and Combinatorial Computing, 44 (2003), 177–187. (MR 2004a:11017)

This paper describes various pairs of trajectories that coalesce under the 3x + 1
iteration. For example the trajectories of 3n + 1 and 4n + 1 coalesce, and the trajectory
of 16k + 13 coalesces with that of 3k + 4. The main result (Theorem 3.9) gives a certain
infinite family of coalescences.

22. Charles C. Cadogan (2006), A Solution to the 3x+1 Problem, Caribbean J. Math. Comp.
Sci. 13 (2006), 1–11.

This paper asserts a proof of the 3x + 1 conjecture. However the argument given has
a gap which leaves the proof incomplete. Namely, on the line just before equation (2.6)
the expression 1 + 2ti,j ∼ 1 + 3ni+1,j should instead read 1 + 2ti,j ∼ 1 + 3ni,j , as given by
equation (2.5). Hence instead of obtaining equation (2.6) in the form ti,j ∼ ti+1,j ∼ ti+2,j ,
one only obtains ti,j ∼ ti+1,j . This renders the proof of Theorem 2.15 incomplete, as it
depends on equation (2.6). Then the induction step in Lemma 3.1 cannot be completed,
as it depends on Theorem 2.15. Finally the main result Theorem 3.3 has a gap since it
depends on Lemma 3.1.

23. Mónica del Pilar Canales Chacón and Michael Vielhaber (2004), Structural and Compu-
tational Complexity of Isometries and their Shift Commutators, Electronic Colloquium
on Computational Complexity, Report No. 57 (2004), 24 pp. (electronic).

The paper considers functions on f : {0, 1}∞ → {0, 1}∞ computable by invert-
ible transducers. They give several formulations for computing such maps and consider
several measures of computational complexity of such functions, including tree complex-
ity T (f, h), which measures the local branching of a tree computation, where h is the tree
height of a vertex. They also study the bit complexity B(f, n) which is the complexity
of computing the first n input/ output symbols. Tree complexity is introduced in H.
Niederreiter and M. Vielhaber, J. Complexity 12 (1996), 187–198 (MR 97g:94025).

8
 The 3x + 1 function is considered as an example showing that some of the general com-
plexity bounds they obtain are sharp. Interpreting the domain {0, 1}∞ as the 2-adic
integers, the map Q∞ associated to the 3x + 1 map given in Lagarias (1985) [Theorem L]
is a function of this kind. It is invertible and the inverse map Q−1
∞ is studied in Bernstein
(1994) and Bernstein and Lagarias (1996). In Theorem 33 the authors give a 5-state
shift automaton that computes the “shift commutator” of the 3x + 1 function, which
they show takes a 2-adic integer a to a if a is even, and to 3a + 2 if a is odd. In Theorem
34 they deduce that the tree complexity of Q∞ is bounded by a constant. Here Q∞
corresponds to their function T (c, ·).

24. Mark Chamberland (2003), Una actualizachio del problema 3x+1, Butletti de la Societat
Catalana, 22 (2003) 19–45. (MR 2004i:11019).

This is a survey paper (in Catalan) describing recent results on the 3x + 1 problem,
classified by area.
Note. An English version of this paper: “An Update on the 3x + 1 Problem” is posted
on the author’s webpage: http://www.math.grin.edu/∼chamberl/

25. Dean Clark, Periodic solutions of arbitrary length in a simple integer iteration, Advances
in Difference Equations, 2006 Article 35847, pp. 1–9.

This paper studies the second order nonlinear recurrence

yn+1 = ⌈ayn ⌉ − yn−1

in which the initial conditions (y0 , y1 ) are integers, and a is a constant with {a ∈ R :
|a| < 2, }. These generate an infinite sequence. For a = 32 , presuming integer initial
conditions, the recurrence can be rewritten as
3yn + 1


 − yn−1 if yn ≡ 1 (mod 2)
2


yn+1 :=
 3yy − y


 n−1 if yn ≡ 0 (mod 2) .
2
which the author views as a second-order analogue of the 3x + 1 iteration. The main
result of the paper shows that for fixed |a| < 2, all orbits of the recurrence are purely
periodic.
In addition, the author shows when a = pq is a non-integer rational with |a| < 2, then
there are orbits of arbitrarily large period.
The author√presents some computer plots of values T (yn−1 , yn ) = (yn , yn+1 ) in the plane,
For a = 1+2 5 and other values, the periodic orbits plotted this way can appear complex,
approximating fractal-like shapes.

26. Lisbeth De Mol (2008), Tag Systems and Collatz-like functions, Theoretical Computer
Science 390 (2008), 92–101.

A Post tag stystem T with alphabet size µ and shift ν consists of an alphabet
A = {a1 , a2 , · · · , aµ } and rules ai 7→ Ei , where Ei is a finite word with symbols drawn
from this alphabet. Given an input word w with leftmost symbol ai , the tag system

9
 iteration produces an output word w′ which chops off the ν leftmost symbols and tags on
to the right end of w the word Ei , thus the new word w′ has length |w′ | = |w| + |Ei | − ν.
Given an initial word w0 , the tag system produces subsequent words w1 , w2 , .... It is said
to halt at step n if wn is the empty word. The halting problem for a tag system T is
to determine for each possible input word w0 whether or not the tag system eventually
halts on that input. The reachability problem for T is, given a word x, to determine for
each input w0 whether or not the word x is eventually reached under repeated iteration.
Let T S(µ, ν) denote the collection of all Tag systems with parameters (µ, ν). In 1921
Emil Post showed the halting and reachability problems are decidable for all tag systems
in T S(2, 2); his proofs were not published. He could not resolve the case of T S(2, 3).
In 1961 Minsky showed that the halting problem, and hence the reachability problem
are undecidable for tag systems in general. It was later shown that a universal Turing
machine is encodable as a tag system in T S(576, 2), so both the halting and reachability
problems are unsolvable in this class.
In this paper the author shows the 3x + 1 problem is encodable as a reachability problem
in T S(3, 2). The tag system takes A = {a, b, c} with rules a 7→ bc, b 7→ a, c 7→ aaa. The
reachability question concerns reaching x = a, Starting from w0 = an (word repeating the
letter a n times) the next word that is a power of a that the tag system reaches is aT (n) ,
where T (n) is the 3x + 1 function. To answer this reachability problem for all iterates
requires solving the 3x+1 problem. It follows that in should be a difficult problem to find
a decision procedure for the reachability problem on the class T S(3, 2). The author also
shows that the iteration of a generalized Collatz function which is affine on congruence
classes (mod d) is encodable in a tag system in some T S(µ, d) with µ ≤ 2d + 3.

27. Diego Domenici (2009) A few observations on the Collatz problem, Inter. J. Appl. Math.
Stat. 14 (2009), No. J09, 97–107. (MR 2524878)
The paper proves results encoding the 3x + 1 conjecture as a problem of representing
integers in particular forms. It shows a necessary and sufficient condition for the 3x + 1
conjecture to hold is that each positive integer n has a representation
1
2m X 2bk
n= −
3l 3k
k=1

with 1 ≤ l ≤ m − 3 and 0 ≤ b1 < b2 < · · · < bl ≤ m − 4.

28. Jean-Guillaume Dumas (2008), Caractérisation des quenines et leur représentation spi-
rale Mathématiques et Sciences Humaines [Mathematics and Social Science], 184 No. 4,
9–23.

This paper solves the problem raised by Queneau (1963) on allowable spiral rhyme
patterns generalizing the ‘sestina” rhyme pattern of Arnaut Daniel, a 12-th century
troubadour. One considers iterations of the (3x + 1)-like function
x

 2 if x is even
σn (x) :=
 2n+1−x
2 if x is odd

on the domain {1, 2, ..., n}. This function gives a permutation in the symmetric group
Sn of this domain, which is called by the author a quenine. We also recall the inverse

10
 permutation δ2,n (x) = σn−1 (x) given by
n

 2x if 1 ≤ x ≤ 2
δ2,n (x) :=
n
2n + 1 − 2x if < x ≤ n.

2

The Raymond Queneau numbers n (or 2-admissible numbers) are those numbers for
which this permutation is a cyclic permutation.
The author first recalls (Theorem 1) the results of Berger (1969), which used the inverse
permutation δ2,n (x) above to put restrictions on Queneau numbers. The author then
proves (Theorem 2) that an integer n is a Queneau number if and only if either p = 2n+1
is prime, and either (i) p ≡ 3 or 5 ( mod 8) and 2 is a primitive root of p, i.e. ordp (2) = 2n,
or (ii) p ≡ 7 ( mod 8) and ordp (2) = n. This establishes a corrected form of a Conjecture
of Roubaud (1993).
The author goes on to consider generalizations of quenines introduced by Roubaud (1993).
He gives pictures showing the various spiral patterns associated with such permutations.
To start, he sets

if 1 ≤ x ≤ n3


 3x



δ3,n (x) := 2n + 1 − 3x if n3 < x ≤ 2n
3




3x − (2n + 1) if 2n3 < x ≤ n.


More generally, one can replace 3 with any g ≤ n, to get a g-spiral permutation, whose
definition can be worked out from a picture of the spiral. Call such a permutation
g-admissible if it is a cyclic permutation. Theorem 3 gives a necessary and sufficient
condition for n to be g-admissible, which is that p = 2n + 1 is prime and either (i) g
is a primitive root (mod p), or (ii) n is odd, and ordp (g) = n. The paper includes a
table giving for n ≤ 1000 such that p = 2n + 1 is prime the minimal g for which n is
g-admissible.
Various other types of permutations are considered: a pérecquine is one associated to the
permutation
if 1 ≤ x ≤ n2

 2x
πn (x) :=
2x − (n + 1) if n2 < x ≤ n.


This pattern is named after Georges Perec, an Oulipo member, cf. Queneau (1963).
Theorem 5 shows that this permutation is cycle if and only if p = n + 1 is prime and 2
is a primitive root (mod p).

29. Jeffrey P. Dumont and Clifford A. Reiter (2001), Visualizing Generalized 3x+1 Function
Dynamics, Computers and Graphics 25 (2001), 883–898.

This paper describes numerical and graphical experiments iterating generalizations

of the 3x + 1 function. It plots basins of attraction and false color pictures of escape
times for various generalizations of the 3x + 1 function to the real line, as in Chamber-
land (1996), and to the complex plane, as in Letherman, Schleicher and Wood (1999). It

11
 introduces a new generalization to the complex plane , the winding 3x+1 function,
1  mod2 (z) 
W (z) := 3 z + mod2 (z) ,
2
in which
1 πz 2 i
mod2 (z) := (1 − eπiz ) = (sin ) − sin πz,
2 2 2
Plots of complex basins of attraction for Chamberland’s function appear to have a struc-
ture resembling the Mandlebrot set, while the basins of attraction of the winding 3x + 1
function seems to have a rather different structure. The programs were written in the
computer language J.

30. Jeffrey P. Dumont and Clifford A. Reiter (2003), Real dynamics of a 3-power extension
of the 3x + 1 function, Dynamics of Continuous, Discrete and Impulsive Systems, Series
A: Mathematical Analysis 10 (2003), 875–893. (MR2005e:37099).

This paper studies the real dynamics of the function

1 mod2 (x)
T (x) := (3 x + mod2 (x)),
2
in which the function
1
mod2 (x) := (sin πx)2 .
2
This function agrees with the 3x+ 1-function on the integers. The authors show that this
function has negative Schwartzian derivative on the region x > 0. They study its periodic
orbits and critical points, and show that any cycle of positive integers is attractive. They
define an extension of the notion of total stopping time to all real numbers x that are
attracted to the periodic orbit {1, 2}, representing the number of steps till the orbit enters
the immediate basin of attraction of this attracting periodic orbit. They formulate the odd
critical point conjecture, which asserts for an odd positive integer n ≥ 3 with associated
nearby critical point cn , that the critical point cn is attracted to the periodic orbit {1, 2},
and that cn and n have the same total stopping time.

31. Hershel M. Farkas (2005), Variants of the 3N + 1 problem and multiplicative semigroups,
In: Geometry, Spectral Theory, Groups and Dynamics: Proceedings in Memory of Robert
Brooks, Contemporary Math., Volume 387, Amer. Math. Soc., Providence, 2005, pp.
121–127. (MR 2006g:11052)

This paper formulates some weakenings of the 3x + 1 problem where stronger results
can be proved. It first shows that iteration of the map
 n
 3 if n ≡ 0 (mod 3));
3n+1
F (n) = 2 if n ≡ 7 or 11 (mod 12);
 n+1
2 if n ≡ 1 or 5 (mod 12);

on the positive integers has all trajectories get to 1. The trajectories of the iterates
of these functions can have arbitrarily long subsequences on which the iterates increase.
The author then asks questions of the type: “Which integers can be represented in a mul-
tiplicative semigroup whose generators are a specified infinite set of rational numbers?”

12
 He proves that the integers represented by the multiplicative semigroup generated by
{ d(n)
n : n ≥ 1}, where d(n) is the divisor function, represents exactly the set of positive
odd integers. The analysis involves the function F (n) above. Finally the author pro-
poses as an open problem a weakened version of the 3x + 1 problem, which asks: “Which
integers are represented by the multiplicative semigroup generated by { 2n+1
3n+2 : n ≥ 1}
together with {2}?” The truth of the 3x + 1 Conjecture implies that all positive integers
are so represented.
Note. Applegate and Lagarias (2006) prove that all positive integers are represented in
the semigroup above.
Of the author’s three results, Theorem 1 holds for T̃ (x). The arguments of Theorems 2
and 3 presented appear to apply to T̃ (x) as well; line 4 of Theorem 3 needs to be modified
to T̃ (L+1) (n) ≡ T̃ (L) (n) + 1 ( mod 2). The “proof” of Theorem 2 seems not to adequately
relate the mathematics and the metamathematics.

32. David Gluck and Brian D. Taylor (2002), A new statistic for the 3x + 1 problem, Proc.
Amer. Math. Soc. 130 (2002), 1293–1301. (MR 2002k:11031).

This paper considers iterations of the Collatz function C(x). If a = (a1 , a2 , ..., an ) is
a finite Collatz trajectory starting from a1 , with an = 1 being the first time 1 is reached,
they assign the statistic
a1 a2 + a2 a3 + ... + an−1 an + an a1
C(a) = .
a21 + a22 + ... + a2n
9
They prove that 13 < C(a) < 75 . They find sequences of starting values that approach
the upper and lower bounds, given that the starting values terminate.

33. Jeffrey R. Goodwin (2003), Results on the Collatz conjecture, Annalele Stiintifice ale
Universitatii “Al. I. Cuza” din lasi serie noua. Informatica (Romanian), XIII (2003)
pp. 1–16. MR2067520 (2005b:11025). [Scientific Annals of the “Al. I. Cuza” University
of Iasi, Computer Science Section, Tome XIII, 2003, 1–16]

This paper partitions the inverse iterates of 1 under the 3x + 1 map into various
subsets, and studies their internal recursive structure.

34. He, Sheng Wang (2003), 3n+1 problem’s simplifying and structure property of {T (n)}(n ∈
N) (Chinese), Acta Scieniarum Naturalium Universitatis Neimonggol [Nei Menggu da xue
xue bao. Zi ran ke xue] (2003), No. 2.

[I have not seen this paper.]

35. Kenneth Hicks, Gary L. Mullen, Joseph L. Yucas and Ryan Zavislak (2008), A Polynomial
Analogue of the 3N + 1 Problem?, American Math. Monthly 115 (2008), No. 7, 615–622.

This paper studies an iteration on polynomials resembling the 3x + 1 problem.

It considers the function on the polynomial ring GF2 [x], where GF2 is the finite field
with 2 elements, given by

(x + 1)f (x) + 1 if f (0) 6= 0


C1 (f (x)) = f (x) (1)
x if f (0) = 0.

13
 They show that the iteration always converges to a constant; for an n-th degree starting
polynomial in at most n2 + 2n steps. They also show there exist starting polynomials
of degree n which take at least 3n steps to get to a constant. They observe that their
result applies more generally to the map CF acting on monic polynomials f (x) = xn +
an−1 xn−1 + · · · + a1 x + a0 ∈ F [x], where F is an arbitrary field, by:
a2
(
(x − aa0j )f (x) + a0j if f (0) 6= 0
CF (f (x)) = f (x)
(2)
x if f (0) = 0,

where aj is the smallest value 1 ≤ j ≤ n such that aj 6= 0. Here a degree n polynomial

takes at most n2 + 2n steps to get to a constant polynomial.
Note. Iterations of a polynomial ring over a finite field were first introduced in Matthews
and Leigh (1987). They exhibited a mapping having a provably divergent trajectory.

36. Wernt Hotzel (2003), Beitäge zum 3n + 1-Problem, Dissertation: Univsität Hamburg
2003, 62pp. [Zbl 1066.11501]

This thesis consists of 8 short chapters.

Chapter 2 describes symbolic dynamics of the 3x+1 map T , calling it the Collatzfunction
d, and noting it extends to the domain of 2-adic integers Z2 . He encodes a symbolic
dynamics of forward iterates, using binary labels (called I and O, described here as 1
and 0). He introduces the 3x + 1 conjugacy map, denoting it T .
Chapter 3 studies the set of rational periodic points, letting Q2 denote the set of rationals
with odd denominators. He observes that there are none in the open interval (−1, 0).
Chapter 4 studies periodic cycles using Farey sequences, following Halbeisen and Hungerbühler
(1997).
Chapter 5 studies unbounded orbits, observing that for rational inputs r ∈ Q2 has an
unbounded orbit if and only if r 6∈ Q2 .
Chapter 6 studies rational cycles with a fixed denominator N ,
1
Chapter 7 studies periodic points of the 3x + 1 conjugacy map. The fixed point 3 is
known. An algorithm which searches for periodic points is described.
Chapter 8 describes an encryption algorithms based on the 3x + 1 function.

37. Huang, Guo Lin and Wu, Jia Bang (2000), One-to-one correspondence between the natu-
ral numbers and the parity vectors in the Collatz problem (Chinese), J. of South-Central
University for the Nationalities, Natural Science Ed. [Zhong nan min zu da xue xue bao.
Zi ran ke xue ban] 19 (2000), No. 3, 59–61.

English summary: ”It is shown in this paper that MN ◦ = {0, 1, 2, · · · , 2N − 1} and

VN denotes the set of all vn of truncations up to the N -th term, viz. {x0 , x1 , ..., xN −1 }
◦ and m → v (m) then the mapping
of the parity vector v = {x0 , x1 , · · · }, if m ∈ MN N
σ : MN ◦ → V
N is one-to-one. The following lemmas are from this. Lemma 1. Let
MN = {1, 2, · · · , 2N }, if m ∈ MN and m → vN (m) then the correspondence is one-to-
one. Lemma 2. Let v denote the set of all parity vectors v = {x0 , x1 , · · · }, if m ∈ N and
m → v(m) then the correspondence is one-to-one. Thus the investigation on any natural
number can be converted to the investigation of its parity vector.”

14
 Note. This result is implicit in Terras (1976), and properties of the parity vector are
described in Lagarias (1985).

38. Yasuaki Ito and Koji Nakano (2009), A Hardware-Software Cooperative Approach to the
Exhaustive verification of the Collatz conjecture, International Symp. on Parallel and
Distributed Processing with Applications, 2009, pp. 63–70.

[I have not seen this paper.]

39. Ke, Wei (2001), Generalization Concerning the 3x + 1 Problem (Chinese), Journal of
Jinan University (Science and Technology) [Jinan da xue xue bao. Zi ran ke xue ban]
(2001), No. 4.

[I have not seen this paper.]

40. Ke, Yong-Sheng (2005), The proof of the hypothesis about ”3x + 1” (Chinese), Journal
of Tianjin Vocational Institute [Teacher’s College] [Tianjin zhi ye ji shu shi fax xue yuan
xue bao] (2005), No. 2.

[I have not seen this paper.]

41. Immo O. Kerner (2000), Die Collatz-Ulam-Kombination CUK, Rostocker Informatik-

Berichte No. 24 (2000), preprint.

This paper attibutes the Collatz problem to Collatz in 1937. Let c(n) be a function
counting the number of iterations of the Collatz function C(n) to get to 1, with c(1) = 0
and c(3) = 7, and so on. It reviews some basic results on the Collatz function, and
introduces an auxiliary two-variable function B(n, t) to describe the iteration, and plots
some graphs of its behavior.

42. Stefan Kohl (2005), Restklassenweise affine Gruppen, Universität Stuttgart, Ph. D.
Dissertation, 2005. (eprint: http://deposit.ddb.de/dokserv?idn=977164071)

In this thesis, the author studies the semigroup Rcwa(Z)consisting of all functions f :
Z → Z for which there exists a modulus m = m(f ) such that the restriction of f to each
residue class ( mod m) is an affine map. It also considers the group RCW A(Z) consisting
of the set of invertible elements of Rcwa(Z). The group RCW A(Z) is a subgroup of the
infinite permutation group of the the integers. Both the 3x + 1 function and the Collatz
function belong to the semigroup Rcwa(Z). The original Collatz map (see Klamkin
(1963)) given by f (3n) = 2n, f (3n − 1) = 4n − 1, f (3n − 2) = 4n − 3 is a permutation
belonging to RCW A(Z).
Some of the results of the thesis are as follows. The group RCW A(Z) is not finitely
generated (Theorem 2.1.1). It has finite subgroups of any isomorphism type (Theorem
2.1.2). It has a trivial center (Theorem 2.1.3). It acts highly transitively on Z (Theorem
2.1.5). All nontrivial normal subgroups also act highly transitively on Z, so that it has
no notrivial solvable normal subgroup (Corollary 2.1.6). It has an epimorphism sgn
onto {1, −1}, so has a normal subgroup of index 2 (Theorem 2.12.8). Given any two
subgroups, it has another subgroup isomorphic to their direct product(Corollary 2.3.3).

15
 It has only finitely many conjugacy classes of elements having a given odd order, but it
has infinitely many conjugacy classes having any given even order (Conclusion 2.7.2).
The author notes that the 3x+1 function can be embedded as a permutation in RCW A(Z×
Z), as (x, y) 7→ ( 3x+1 x x
2 , 2y) if x ≡ 1(mod 2); 7→ ( 2 , y) if x ≡ 0, 2(mod 6); 7→ ( 2 , 2y + 1)if
x ≡ 4(mod 6), where it represents the iteration projected onto the x-coordinate.
The author develops an algorithm for efficiently computing periodically linear functions,
whether they are permutations or not. Periodically linear functions are functions which
are defined as affine functions on each congruence class j (mod M ) for a fixed modulus
M , as in Lagarias (1985). The 3x + 1 function is an example of such a function. The
author has written a corresponding package RCWA (Residue Class-Wise Affine Groups)
for the computational algebra and group theory system GAP (groups, Algorithms, Pro-
gramming). This package is available for download at:
http://www.gap-system.org/Packages/rcwa.html.
A manual for RCWA can be found at:
http://www.gap-system.org/Manuals/pkg/rcwa/doc/manual.pdf

43. Stefan Kohl (2007), Wildness of iteration of certain residue class-wise affine mappings,
Advances in Applied Math. 39 (2007), 322–328. (MR 2008g:11041)

A mapping f : Z → Z is called residue class-wise affine (abbreviated RCWA) if

it is affine on residue classes ( mod m) for some fixed m ≥ 1. (This class of functions was
termed periodically linear in Lagarias (1985).) The smallest such m is called the modulus
of f . This class of functions is closed under pointwise addition and under composition.
A function f is called tame if the modulus of its k-th iterate remains bounded as k → ∞;
it is wild otherwise. The author shows that if f : Z → Z is an RCWA -function, which
is surjective, but not injective, then f is necessarily wild. The paper also presents coun-
terexamples showing that each of the three other possible combinations of hypotheses of
(non-)surjectivitiy or of (non)-injectivity of f permits no conclusion whether it is tame
or wild.

44. Stefan Kohl (2008a), On conjugates of Collatz-type mappings, Int. J. Number Theory 4,
No. 1 (2008), 117–120. (MR 2008m: 11055)

A map f : Z → Z is said to be almost contracting if there is a finite set S such

that every trajectory of f visits this finite set. This property holds if and only if there is
a permutation σ of the integers such that g = σ −1 ◦ f ◦ σ decreases absolute value off a
finite set, a property that is called
em monotonizable. Suppose that f : Z → Z is a surjective, but not injective, RCWA
mapping (see Kohl(2006+) for a definition) having the property that the preimage set of
any integer under f is finite. The main result asserts that if f is almost contracting and k
such that the k-th iterate f (k) decreases almost all integers, then any permutation σ that
establishes the almost contracting property of f cannot itself be an RCWA mapping.
The 3x + 1 function T (x) is believed to be a function satisfying the hypotheses of the
author’s main result: It is surjective but not injective, and is believed to be almost con-
tracting. The almost contacting property for the 3x + 1 map is equivalent to establishing
that its iteration on Z has only finitely many cycles and no divergent trajectories.

45. Stefan Kohl (2008b), Algorithms for a class of infinite permutation groups, J. of Symbolic

16
 Computation 43, No. 8 (2008), 545–581. (MR 2009e:20003)

A mapping f : Z → Z is called residue-class-wise afffine (RCWA) if there is a

positive integer m such that it is an affine mapping when restricted to each residue class
(mod m). The 3x + 1 mapping T is of this kind, as is a permutation constructed by
Collatz. This paper describes a collection of algorithms and methods for computing in
permutation groups and monoids whose members are all RCWA mappings.

46. Stefan Kohl (2010), A simple group generated by involutions interchanging residue classes
of the integers, Math. Z. 264 (2010), no. 4, 927–938.

The author defines a simple group of permutations of the integers, generated by

compositions of certain RCWA functions.

47. Pavlos B. Konstadinidis (2006), The real 3x + 1 problem, Acta Arithmetica 122 (2006),
35–44. (MR 2007c:11029)

The author extends the 3x + 1 function to the real line as:

3x + 1

 if ⌊x⌋ ≡ 1 (mod 2) .
2


U (x) =
 x
if ⌊x⌋ ≡ 0 (mod 2) .


2
The paper shows that the only periodic orbits of the function U (x) on the positive
real numbers are those on the positive integers. The paper also considers some related
functions.

48. Alex V. Kontorovich and Steven J. Miller (2005), Benford’s law, values of L-functions,
and the 3x + 1 problem, Acta Arithmetica 120 (2005), 269–297. (MR 2007c:11085).

Benford’s law says that the leading digit of decimal expansions of certain sequences
are not uniformly distributed, but have the probability of digit j being log10 (1 + 1j ). This
paper gives a general method for verifying Benford’s law for certain sequences. These
include special values of L-functions and ratios of certain 3x + 1 iterates, the latter case
being covered in Theorem 5.3. It considers the 3x+1 iteration in the form of Sinai(2003a)
and Kontorovich and Sinai (2002), which for an odd integer x has m(x) being the next
odd integer occurring in the 3x + 1 iteration. For a given real base B > 1 it looks at
the distribution of the quantities logB (xm /( 34 )m x0 )(mod 1) as x0 varies over odd inte-
gers in [1, X] and X → ∞. It then takes a second limit as m → ∞ and concludes that
the uniform distribution is approached, provided B is such that log2 B is an irrational
number of finite Diophantine type. The case B = 10 corresponds to Benford’s law. The
theorem applies when B = 10 because log2 10 is known to be of finite Diophantine type
by A. Baker’s results on linear forms in logarithms. A main result used in the proof
is the Structure Theorem in Kontorovich and Sinai (2002). Note that the assertion of
Theorem 5.3 concerns a double limit: first X → ∞ and then m → ∞. See Lagarias and
Soundararajan (2006) for related results.

49. Alex V. Kontorovich and Yakov G. Sinai (2002), Structure Theorem for (d, g, h)-maps,
Bull. Braz. Math. Soc. (N.S.) 33 (2002), 213–224. (MR 2003k:11034).

17
 This paper studies (d, g, h)-maps, in which g > d ≥ 2, with g relatively prime to d,
and h(n) is a periodic integer-valued function with period d, with h(n) ≡ −n(mod d)
and 0 < |h(n)| < g. The (d, g, h)-map is defined on Z\dZ by

gx + h(gx)
T (x) := , with dk ||gx + h(gx).
dk
A path of m iterates can be specified by the values (k1 , k2 , ..., km ) and a residue class
ǫ (mod dg), and set k = k1 + k2 + ... + km . The structure theorem states that exactly
(d − 1)m triples (q, r, δ) with 0 ≤ q < dk , 0 < r < g m and δ ∈ E = {j : 1 ≤ j < dg, d ∤
j, g ∤ j} produce a given path, and for such a triple (q, r, δ) and all x ∈ Z,

T (m) (gd(dk x + q) + ǫ) = gd(g m x + r) + δ.

They deduce that, as m → ∞, a properly logarithmically scaled version of iterates

d
converges to a Brownian motion with drift log g − d−1 log d. More precisely, fix m, and
choose points 0 = t0 < t1 < ... < tr = 1 and set mi = ⌊ti m⌋ and yi = log T (mi ) (x). Then
the values yi − yi−1 converge to a Brownian path. These results imply that when the
drift is negative, almost all trajectories have a finite stopping time with |T (m) (x)| < |x|.

50. Benjamin Kraft and Kennan Monks (2010), On conjugacies for the 3x + 1 map induced
by continuous endomorphisms of the shift dynamical system, Discrete Math. 310 (2010),
no. 13-14, 1875–1883. (MR 2011e:37027)

The continuous endomorphisms of the shift dynamical system were classified by

Maria Monks (2009). A conjugacy Φ between the 3x + 1 map T extended to the 2-adic
integers to the one-sided shift S was shown in Lagarias (1985), i.e. Φ ◦ S ◦ Φ−1 = T .
This paper studies maps Hf = Φ ◦ f ◦ Φ−1 where f is a continuous endomorphism of
the 2-adic integers Z2 (a.k. a. binary shift dynamical system). It generalizes results of
Maria Monks (2009).

51. Ilia Krasikov and J. C. Lagarias (2003), Bounds for the 3x + 1 problem using difference
inequalities, Acta Arithmetica 109 (2003), no. 3, 237–258. (MR 2004i:11020)

This paper deals with the problem of obtaining lower bounds for πa (x), the counting
function for the number of integers n ≤ x that have some 3x + 1 iterate T (k) (n) = a. It
improves the nonlinear programming method given in Applegate and Lagarias (1995b)
for extracting lower bounds from the inequalities of Krasikov (1989). It derives a nonlin-
ear program family directly from the Krasikov inequalities (mod 3k ) whose associated
lower bounds are expected to be the best possible derivable by this approach. The non-
linear program for k = 11 gives the improved lower bound: If a 6≡ 0 (mod 3), then
πa (x) > x.841 for all sufficiently large x. The interest of the new nonlinear program
family is the (not yet realized) hope of proving πa (x) > x1−ǫ by this approach, taking a
sufficiently large k.

52. Stuart A. Kurtz and Janos Simon (2007), The undecidability of the generalized Collatz
problem, In: J-Y. Cai et al, Ed., Theory and Applications of Models of Computation, 4-
th International Conference, TAMC 2007, Shanghai, China, May 22-25, 2007. Lecture
Notes in Computer Science No. 4484, Springer-Verlag: New York (2007), pp. 542–553.

18
 (MR 2374341)

A generalized Collatz function g is a function mapping the nonnegative integers to

itself such that there is a modulus m such that for each congruence class x ≡ i ( mod m)
it is given by g(x) = ai x + bi when where (ai , bi ) are nonnegative rational numbers such
that (i + km)ai + bi is an integer for all integer k. Generalized Collatz functions can
easily be listed with a Gödel numbering ke .
The authors build on the work of Conway (1972), who considered the subclass of such
functions with all bi = 0, which we term multiplicative generalized Collatz functions.
Conway proved that the following problem is undecidable: Given as input (g, n) where
g is a multiplicative generalized Collatz function and n a nonnegative integer, can it
be decided whether there is some i ≥ 1 such that the i-th iterate g (i) (2n ) is a power
of 2? Conway obtained the result by showing that the partially defined function M (n)
determined by such a function g (with 2M (n) being the first power of 2 encountered as
an iterate g (i) (2n ) if one exists, and M (n) is undefined otherwise) could be any unary
partial recursive function. Here the authors formulate:
Generalized Collatz Problem (GCP ). Given a representation of a generalized Collatz
function g, can it be decided for all x ≥ 0 whether there exists some i ≥ 1 such that
g (i) (x) = 1?
They first formulate (Theorem 1): Let Me be an acceptable Gödel numbering of partial
recursive functions. Then from the index e one can compute a representation of a gen-
eralized Collatz function ge such that Me is total if and only if for every x ≥ 1 there
(i)
exist a value ige1 such that ge (x) = 1. From this result they formulate (Theorem 2):
The problem GCP is Π20 -complete. This locates precisely in the degrees of unsolvability
hierarchy the difficulty of this undecidable problem.
This is a conference paper and does not supply detailed proofs.

53. Jeffrey C. Lagarias (2003+), The 3x + 1 Problem: An Annotated Bibliography (1963–

1999),
eprint: arxiv:math.NT/0309224 Sept. 13, 2003, v11.

This is the initial installment of the annotated bibliography. It contains over 180
items.

54. Jeffrey C. Lagarias (2006), Wild and Wooley Numbers, American Mathematical Monthly,
113 (2006), 97–108. (MR 2007g:11029).

This paper considers some problems about multiplicative semigroups of positive

rationals motivated by work of Farkas (2005) on variants of the 3x + 1 problem. The wild
semigroup S is the semigroup of positive rational numbers generated by { 3n+1
2n+1 : n ≥ 0}
1
together with 2 , and the Wooley semigroup is the sub-semigroup generated by { 3n+1 2n+1 :
n ≥ 0} without 12 , This paper considers the question of which integers occur in these
semigroups. The wild integer semigroup is the set of all integers in S, and generators
of the wild integer semigroup are termed wild numbers. The paper develops evidence in
favor of the conjecture that the wild numbers consist of the set of all primes, excluding

19
 3. It shows that 3 is not a wild number, that all other primes below 50 are wild num-
bers, and that there are infinitely many wild numbers. The term “wild numbers” was
suggested by the novel “The Wild Numbers” by Philibert Schogt. The conjecture above
was proved subsequently in Applegate and Lagarias (2006).

55. Jeffrey C. Lagarias and Neil J. A. Sloane (2004), Approximate squaring, Experimental
Math. 13 (2004), 113–128. (MR 2005c:11098).

This paper studies iteration of the “approximate squaring” map f (x) = x⌈x⌉, and
asks the question whether for a rational starting value x0 = r > 1 some iterate is an
integer. It conjectures that the answer is always “yes”, and proves it for rationals r with
denominator 2. It shows that this holds for most rationals having a fixed denominator
d ≥ 3 with an exceptional set of integers below x of size at most O(xαd ) for certain
constants 0 < αd < 1. It then considers a variant of this problem on the p-adic numbers,
where an exceptional set exists and is shown to have Hausdorff dimension equal to αp .
The paper also studies the iteration of “approximate multiplication” maps fr (x) = r⌈x⌉,
where r is a fixed rational number. It conjectures that for r > 1 all but a finite number
of integer starting values have some subsequent iterate that is an integer, and proves this
for rationals r with denominator 2. It shows for rationals r with denominator d that
the size of the exceptional set of integers below x that have no integer in their forward
orbit under fr has cardinality at most O(xβd ) with βd = log(d−1)
log d . It suggests that this
conjecture is likely to be hard in the general case, by noting an analogy with iteration of
the map appearing in Mahler’s Z-number problem, see Mahler (1968).

56. Jeffrey C. Lagarias and K. Soundararajan (2006), Benford’s Law for the 3x + 1 Function,
J. London Math. Soc. 74 (2006), 289–303. (MR 2007h:37007)

Kontorovich and Miller (2005) proved results concerning Benford’s law for initial
3x + 1 iterates, in a double limit as the number of iterates N → ∞. This paper proves
a quantitative version of Benford’s law valid for finite N . Benford’s law (to base B)
for an infinite sequence {xk : k ≥ 1} of positive quantities xk is the assertion that
{logB xk : k ≥ 1} is uniformly distributed (mod 1). This paper studies the initial iter-
ates xk = T (k) (x0 ) for 1 ≤ k ≤ N of the 3x + 1 function, where N is fixed. It shows
that for most initial values x0 , such sequences approximately satisfy Benford’s law, in
the sense that the discrepancy of the finite sequence {logB xk : 1 ≤ k ≤ N } is small. The
precise result treats the uniform distribution of initial values 1 ≤ x0 ≤ X, with x ≥ 2N ,
1
and shows that for any (real) base B > 1 the discrepancy is smaller than 2N − 36 for all
1
but an exceptional set |E(X, B)| of cardinality |E(X, B)| ≤ c(B)N − 36 X, where c(B) is
independent of N and X.

57. Eero Lehtonen (2008), Two undecidable versions of Collatz’s problems, Theor. Comp.
Sci. 407 (2008), 596–600. (MR 2009k:68092)

The author gives two constructions. He first constructs a function


 3x + t if n ∈ At , for t = −9, −8, ..., 8, 9,

f (x) =
 x

if x ≡ 0 (mod 2) ,
2

20
 in which {At : − 9 ≤ t ≤ 9} are recursive sets of nonnegative integers that partition the
class of odd positive integers, for which the problem of deciding if a given integer iterates
to 1 is undecidable. It encodes the halting problem for Turing machines, when halting
occurs if and only if the function iterations eventually arrive at 1. His second construction
is a variant of Collatz’s original function f (3n) = 2n, f (3n+1) = 4n+1, f (3n+2) = 4n+3,
which is a permutation of the positive integers N. He constructs a recursive function
f : N → N which is a bijection of N, such that the question whether a given input n is in
a finite cycle under iteration of f is undecidable.

58. Dan Levy (2004), Injectivity and Surjectivity of Collatz Functions, Discrete Math. 285
(2004), 190–199. (MR 2005f:11036).

This paper gives necessary and sufficient conditions on members of a class of gener-
alized Collatz maps of the form T (x) = mi x−r
d
i
for x ≡ i (mod d) to be injective maps,
resp. surjective maps, on the integers. These give as a corollary a criterion of Venturini
(1997) for such a map to be a permutation of the integers.
The author frames some of his results in terms of concepts involving integer matrices.
He introduces a notion of gcd matrix if its elements can be written Mij = gcd(mi , mj )
and a difference matrix if its elements can be written Mij = mi − mj . Then he considers
a relation that M is a total non-divisor of N if Mij ∤ Nij for all i, j. Then the author’s
condition for injectivity of a generalized Collatz map above is that the d × d gcd matrix
Mij = gcd(mi , mj ) is a total non-divisor of the d × d difference matrix Nij = qi − qj , with
r −jm
qj = j d j .
A very interesting result of the author is an explicit example of an injective function T (·)
in the class above which has a (provably) divergent trajectory, and which has iterates
both increasing and decreasing in size. This particular map T is not surjective.

59. Li, Xiao Chun (2002), Compressive Iteration for the 3N + 1 conjecture (Chinese), J.
Huazhong Univ. of Science and Technology (Natural Science) [Hua zhong gong xue
yuan] 30 (2002), no. 2.

[I have not seen this paper.]

60. Li, Xiao Chun (2003), Contractible iteration for the 3n + 1 conjecture (Chinese, English
summary), J. Huazhong Univ. Sci. Technol. Nat. Sci. [Hua zhong gong xue yuan] 31
(2003), no. 7, 115–116. (MR 2025640).

English summary: ”The concept of contractible iteration for the 3N + 1 conjec-

ture was presented. The results of contractible iteration were gien as follows: some
equivalence propositions for the 3N + 1 conjecture; sequence of Syracuse order; proof of
term formula of n; some theories about ta (n) = tc (n).”
Note. The adequate stopping time ta (n) and coefficient stopping time tc (n) are those
given in Wu and Hao (2003). Terras (1976) made a conjecture equivalent to asserting
the equality ta (n) = tc (n) always holds.

61. Li, Xiao Chun (2004), Same-flowing numbers and super contraction iteration in 3N + 1
conjecture (Chinese, English summary), J. Huazhong Univ. Sci. Technol. Nat. Sci. [Hua
zhong gong xue yuan] 32 (2004) no. 10, 30–32.

21
 English summary: ”The concept of same-flowing numbers in 3N + 1 conjecture
was given. Some definitions and theorems were set up. Research into contraction itera-
tion of odd numbers 2a + 1(a ∈ Nd ) could be converted to research into Syracuse order
of odd number a and contraction iteration simplified. Infinite sequence same-flowing
to 3a(a ∈ Nd ) was constructed. The concept of super contraction iterations and term
formula of x was given.”
Note. The super-contraction iteration keeps track of the successive odd integers in the
Collatz iteration.

62. Li, Xiao Chun (2005), Necessary condition for periodic numbers in the 3N + 1 conjecture
(Chinese, English summary), J. Huazhong Univ. Sci. Technol. Nat. Sci. [Hua zhong
gong xue yuan] 33 (2005), no. 11, 102–103. (MR 2209315).

English summary: ”After the introduction of the conception of periodic numbers

in 3N + 1 conjecture, numerical theory function was defined, the definition of sum-line
number in binary bit described and the Gauss function given. At the same time, some
properties of number theory function and calculation formula of were discussed. By the
use of number theory function, a necessary condition of periodic numbers in 3N + 1 con-
jecture was proposed. The presentation of this necessary condition could play a certain
role in further solution to periodic numbers question in 3N + 1 conjecture.”

63. Li, Xiao Chun (2006), Some properties of super contraction iteration in the 3N + 1 con-
jecture (Chinese, English summary), J. Huazhong Univ. Sci. Technol. Nat. Sci. [Hua
zhong gong xue yuan] 34 (2006), no. 8, 115–117. (MR 2287431).

English summary: ”In researching of 3N +1 conjecture, by using operator of dividing

even factor, this paper provided the concept of super contraction and series of iteration
super contraction iteration, which greatly increased iteration velocity in comparison with
former contraction iteration. At the same time, this paper first provided the concept of
omission numbers and obtained the cyclic relation between contraction iteration and su-
per contraction iteration, and obtained no unique property about precursor numbers of
monadic-order to odd number in super contraction iteration, and to the odd number y
of 4k + 3, obtained no unique property about precursor numbers of monadic-order to
odd number in super contraction iteration. At last, this paper first provided necessary
condition of existence of cyclic numbers in super contraction iteration, which can play
active role in cyclic numbers researching of 3N + 1 conjecture. All presented definitions
and theories can simplify in researching world well-known problem 3N + 1 conjecture in
number theory. This paper also provides new method in 3N + 1 conjecture continuous
researching.”
Note. As in his earlier paper Li (2004), the author studies odd numbers appearing in the
3x + 1 iteration.

64. Li, Xiao Chun and Liang, You Min (2002) Research on the Syracuse Operator and its
properties (Chinese), Journal of Air Force Radar Academy (2002), No. 3.

[I have not seen this paper.]

22
65. Li, Xiao Chun and Liu, Jun (2006), Equivalence of the 3N +1 and 3N +3k conjecture and
some related properties (Chinese, English summary), J. Huazhong Univ. Sci. Technol.
Nat. Sci. [Hua zhong gong xue yuan] 34 (2006), no. 7, 120-121. (MR 2287654)

English summary: ”3N + 1 conjecture in number theory was extended to 3N + 3k

one. It was pointed out that 3N + 3 conjecture is equivalent to 3N + 3k one. Some prop-
erties related to 3N + 3 were obtained. The generalization of 3N + 1 conjecture and these
newly obtained properties not only simplify the operation about 4K + 3-odd numbers in
3N + 3 conjecture but also provide new way in researching of 3N + 1 conjecture.”

66. Li, Xiao Chun and Wu, Jia Bang (2004), Study of periodic numbers in the 3N + 1 con-
jecture (Chinese, English summary), J. Huazhong Univ. Sci. Technol. Nat. Sci. [Hua
zhong gong xue yuan] 32 (2004), no. 10, 100–101. (MR 2121229)

English summary: ”A necessary condition about the existence of periodic num-

bers in the 3N + 1 conjecture: S l (mxi ) = nxi , and its generalized necessary condition:
S l (σb xi ) = nb+1 xi were presented. A formula about periodic numbers was given, where
x1 = 1−3rl1/2k . It was proved that the length of periodic numbers being 2 or 3 could not
exist.”
Note. The numbers xi are the successive odd numbers appearing in the 3x + 1 iteration.
The authors exclude periodic orbits on the odd itntegers, where there are exactly 2 or 3
odd numbers.

67. Josefine López and Peter Stoll (2009), The 3x + 1 conjugacy map over a Sturmian word,
Integers 9 (2009), A13, 141–162. (MR 2506145).

This paper studies the 3x + 1 conjugacy map Φ : Z2 → Z2 on the 2-adic integers,

which conjugates the 3x+1 map T (x) to the 2-adic shift map, in the sense Φ−1 ◦T ◦Φ = S,
and which was studied by Bernstein and Lagarias (1996). It is unknown whether there is
any aperiodic x ∈ Z2 such that Φ(x) is periodic; this is conjectured not to happen. This
paper studies this function for x whose 2-adic expansion is a Sturmian word, correspond-
ing to a line with irrational slope. This paper finds a generalized continued fraction
−1
expansion for Φ(x) in this case convergent in the metric on the 2-adic integers Z2 . It
explicitly computes a number of examples, suggesting that the images Φ(x) then have
2-adic expansions of full complexity.

68. Florian Luca (2005), On the nontrivial cycles in Collatz’s problem, SUT Journal of Math-
ematics 41 (2005), no. 1, 31–41 (MR 2006e:11034).

This paper establishes conditions on non-trivial cycles on the positive integers of

the 3x + 1 function. Let n denote the cycle length, and let {x1 , ..., xk } denote the set
of odd integers that appear in such a cycle, so that 1 ≤ k < n, and let lj denote the
number of iterates between xi and xi+1 , so that xi+1 = 3x2il+1 , and n = ki=1 li . Let
P
i
1 ≤ J ≤ k denote the number of blocks of consecutive lj taking the same value, say Lj ,
with the block length Nj for 1 ≤ j ≤ J, so that li = li+1 = ... = li+Nj −1 = Lj with
li−1 6= Lj , li+Nj 6= Lj . Then L1 N1 + ... + Lj Nj = n. Call an ascent a value Lj = 1, i.e.
a consecutive string of increasing values xi < xi+1 < ... < xi+Nj −1 with xi−1 > xi and
xi+Nj < xi+Nj −1 . The author’s main result (Theorem 1) states that there is an absolute

23
 constant C1 such that there are at least C1 log n ascents in any cycle. In particular for any
fixed c, there are only finitely many nontrivial cycles of positive integers having at most
c ascents. This result improves on that of Mimuro (2001). The proof uses transcendence
results coming from linear forms in logarithms.
This result complements a result of Brox(2000), who showed that there are only finitely
many integer cycles having at most 2 log k descents, where a descent is a value i such
that xi+1 < xi .
Note. A misprint occurs in the the definition of li on page 32, in condition (ii) ”largest”
should read ”smallest”.

69. Maurice Margenstern (2000), Frontier between decidability and undecidability: a survey,
Theor. Comput. Sci. 231 (2000), 217–251. (MR 2001g:03079).

This paper surveys results concerning the gap between decidability and undecid-
ability, as measured by the number of states or symbols used in a Turing machine,
Diophantine equations, the word problem, molecular computations. Post systems, regis-
ter machines, neural networks and cellular automata. The 3x + 1 function (and related
functions) have been studied as a test for possible undecidability of behavior for small
state Turing machines, which are so small they are not known to simulate a universal
Turing machine. The author summarizes known results in Figure 12, summarizing how
large a Turing machine has to be (number of states, number of symbols) to encode the
3x+1 problem. For related results, see Margenstern and Matiyasevich (1999) and Michel
(2004).
Note. This paper is the journal version of a conference paper in: M. Margenstern, Ed.,
International Colloquium on Universal Machines and Computations, MCU ’98 , Metz,
France, March 23-27,1998, Proceedings, Volume I, IUT: Metz 1998.

70. Keith R. Matthews (2005+), The generalized 3x+1 mapping, preprint, 23pp., dated Oct.
31, 2005, downloadable as pdf file from: http://www.maths.uq.edu.au/∼krm/interests.html

This paper discusses the behavior of 3x + 1-like mappings and surveys many ex-
ample functions as considered in Matthews and Watts (1984, 1985), Leigh (1986), Leigh
and Matthews (1987) and Matthews (1992), see also Venturiri (1992). A particularly
tantalizing example is


 7x + 3 if x ≡ 0 (mod 3)




 7x + 2

U (x) = if x ≡ 1 (mod 3)
 3


 x − 2 if x ≡ 2 (mod 3) .




3
Almost all trajectories contain an element n ≡ 0 (mod 3) and once a trajectory enters
the set {n : n ≡ 0 (mod 3)} it stays there. Matthews offers $100 to show that if a
trajectory has all iterates U (k) (x) ≡ ±1 (mod 3) then it must eventually enter one of
the cycles {1, −1} or {−2, −4, −2}. The paper also considers some maps on the rings of

24
 √ √
integers of an algebraic number field, for example U : Z[ 2] → Z[ 2] given by
 √
(1 − 2)α √
√ if α ≡ 0 (mod ( 2)) ,



2


U (α) =

 3α + 1 √
√ if α ≡ 1 (mod( 2)) .



2
√
The author√conjectures that if U (k) (α) = xk + yk 2 is a divergent trajectory, then
xk /yk → − 2 as k → ∞.

71. Karl Heinz Metzger (2000), Untersuchungen zum (3n+1)-Algorithmus. Teil II: Die Kon-
struktion des Zahlenbaums, PM (Praxis der Mathematik in der Schule) 42 (2000), 27–32.

The author treats the Collatz function C(x), and studies the directed graph (”number-
forest”) on the positive integers formed with edges iterating the Collatz function. This
graph might have many connected components. The 3x + 1 Conjecture asserts this graph
forms a single tree, with one extra directed edge added making the trivial cycle at the
bottom. The author constructs the graph as follows. To each odd number the author
forms a directed path containing all numbers {2k u : k ≥ 0} (a ”Spross”), with directed
edges from 2k+1 u → 2k u. He then arranges an infinite table of all positive integers, in
which the odd numbers u are placed successively in a bottom horizontal row and the
”Sprossen” are then put in vertical columns over them. To the ”Sprossen” edges he now
adds directed edges starting from the odd nodes u at the bottom row taking u → 3u + 1.
If u ≡ 1 ( mod 4) these edges move horizontally to the left and upward at least two rows,
(”leftmovers”) while if u ≡ 3 (mod 4) they move horizontally to the right and upward
exactly one row (”rightmovers”). He then considers the successive odd integers reached
during an iteration. He shows that any chain of successive rightmoving odd numbers
reached under iteration must be finite (Satz 9), i.e. a number of form 4n + 1 is eventually
reached in the iteration. Satz 10 then asserts that positive integers can all be accounted
for in a rearranged table having only odd numbers of form 4n + 1 along its bottom row.
The results through Satz 9 are rigorous; maybe Satz 10 too.
The final Satz 11 asserts the truth of the 3x + 1 Conjecture. However its proof on
page 32 is incomplete. The proof seems to implicitly assume the decrease in size during
leftmoving steps overcomes the increase in size of iterates during rightmoving steps, but
this is not rigorously shown.

72. Karl Heinz Metzger (2003), Untersuchungen zum (3n + 1)-Algorithmus, Teil III:
Gesetzmässigkeiten der Ablauffolgen, PM (Praxis der Mathematik in der Schule) 45
(2003), No. 1, 25–32.

This paper is independent of parts I and II, which had gaps in some proofs, and
does not reference any of the papers Metzger (1995) (1999) (2000).
It studies iterates of the Collatz function C(n), viewing the iterates ( mod 6). It observes
a kind of self-similar structure in the tree of inverse iterates, and introduces some symbolic
dynamics to describe it. It gives formulas for elements on certain branches of the tree.
If one writes n = 6(k − 1) + i with 0 ≤ i ≤ 5 and C(n) = 6(k̄ − 1) + j then it describes
how to update k to k̄, over a sequence of iterates.

25
73. Pascal Michel (2004), Small Turing machines and generalized busy beaver competition,
Theor. Comp. Sci. 326 (2004), 45–56. [MR 2005e:68049, Zbl 1071.68025]

Let T M (k, l) denote the set of one-tape Turing machines with k states and lsymbols;
the tape is two-way infinite. Much work has been done on how large k and l must be to en-
code an undecidable halting problem. It is known that the halting problem is decidable if
k = 1 or l = 1 and also for classes T (2, 3), T (3, 2) (hence T (2, 2). It is known that univer-
sal Turing machines exist in the classes T (2, 18), T (3, 9), T (4, 6), T (5, 5), T (7, 4), T (10, 3)
and T (19, 2), hence the halting problem is undecidable in these classes. (See Y.Rogozhin,
Small universal Turing machines, Theor. Comp. Sci. 168 (1996), 215–240.) It is
also known that the 3x + 1 problem can be encoded in a Turing machine in classes
T (2, 8), T (3, 5), T (4, 4), T (5, 3) and T (2, 10). It follows that there is no method currently
known for solving the halting problem for Turing machines in these classes. (See Mar-
genstein (2000) for a survey, and also Michel (1993).) Here the author shows there are
Turing machines computing 3x + 1-like functions for which halting is not known in the
small Turing classes T (2, 4), T (3, 3), T (5, 2). Thus the only remaining class for which
decidability of the halting problem is likely to be possible is T (4, 2).
For the class T (2, 4) the author constructs a machine that can encode iteration of the
function g : Z≥0 × {0, 1} → Z≥0 × {0, 1} given by

g(3k, 0) = (5k + 1, 1)
g(3k + 1, 0) = halt
g(3k + 2, 0) = (5k + 4, 0)
g(3k, 1) = halt
g(3k + 1, 1) = (5k + 5, 0)
g(3k + 2, 1) = (5k + 7, 1).

It is not known whether or not every input value Z≥0 × {0, 1} iteration of the function g
eventually halts.
The paper also constructs machines in small Turing classes that achieve new records on
the maximal number of steps before halting on empty input (busy beaver function S(k, l))
and for the maximal number of symbols printed before halting on empty input (Σ(k, l).
These show S(2, 3) ≥ 38, S(2, 4) ≥ 7195, S(3, 3) ≥ 40737 and Σ(2, 3) ≥ 9, Σ(2, 4) ≥ 90
and Σ(3, 3) ≥ 208. The author conjectures the values for (2, 3) and (2, 4) are best possible.

74. Tomoaki Mimuro (2001), On certain simple cycles of the Collatz conjecture, SUT Journal
of Mathematics, 37, No. 2 (2001), 79–89. (MR 2002j:11018).

The paper shows there are only finitely many positive integer cycles of the 3x + 1
function whose symbol sequence has the form 1i (10j )k , where i, j, k vary over nonnega-
tive integers. (The symbol seqence is read left to right.) This result includes the trivial
cycle starting from n = 1, whose symbol sequence is (10), where (i, j, k) = (0, 1, 1). Sup-
pose that the periodic orbit has period p = i + k(j + 1) terms, of which d = i + k are
odd. The author shows by elementary arguments that there are no integer orbits of the
d d
above type with 34 > 23p , and the trivial cycle is the unique solution with 34 = 23p . Using
bounds from transcendence theory (linear forms in logarithms) he shows that there are
d
finitely many values of (i, j, k) giving an integer orbit with 1 > 32p > 34 , with an effective

26
 3d
bound on their size. Any orbit on the positive integers necessarily has 1 > 2p , so the
result follows.
For other papers using transcendence theory to classify some types of periodic orbits, see
Steiner (1978), Belaga and Mignotte (1999), Brox (2000), Simons (2005), Simons and de
Weger (2005). A further improvement on this result is given in Luca (2005).
Note. There are two known integer orbits on the negative integers of the author’s form.
d
They are n = −1 with symbol seqence (1), where (i, j, k) = (1, ∗, 0), and 32p = 32 , and
d
n = −5 with symbol sequence (110) where (i, j, k) = (1, 1, 1), and 32p = 98 . The author’s
d
finiteness result might conceivably be extended to the range 32 ≥ 23p ≥ 1 and so cover
them.

75. Michal Misiurewicz and Ana Rodriguez (2005), Real 3X + 1, Proc. Amer. Math. Soc.,
133 (2005), 1109–1118. (MR 2005j:37011).

The authors consider the semigroup generated by the two maps T1 (x) = x2 and
T2 (x) = 3x+1
2 . They show this semigroup is a free semigroup on two generators. The
forward orbit of a positive input x0 under this semigroup is

O + (x0 ) := {Ti1 ◦ ... ◦ Tin (x0 ) : n ≥ 1, each ik ∈ {0, 1}}.

They show that each orbit O + (x0 ) is dense on (0, ∞). Furthermore they show that
starting from x0 one can get an iterate T (n0 ) (x0 ) within a given error ǫ of a given value
y while remaining in the bounded region

min(x, y − ǫ ≤ Ti1 ◦ · · · Tij (x0 ) ≤ max(11x + 4, 4y − x). 1 ≤ j ≤ n0 .

They show that orbits having a periodic point are dense in (0, ∞). Finally they show
that the group of homeomorphisms of the line generated by T1 , T2 consists of all maps
x 7→ 2k 3l x + 2m
i 3j , in which k, l, m are integers and i, j are nonnegative. It is not a free
group. These results concern topological dynamics; for results concerning measurable
dynamics see Bergelson, Misiurewicz and Senti (2006).

76. Kenneth G. Monks (2002), 3X + 1 Minus the +, Discrete Math. Theor. Comput. Sci.
5 (2002), 47–53. (MR 2203f:11030).

This paper formulates a FRACTRAN program (see Conway(1987)) of the form

Ri (n) ≡ ri n if n ≡ i(mod d), such that the 3x + 1 Conjecture is true if and only if the
R-orbit of 2m contains 2, for all positive integers m. The author determines information
on the behavior under iteration of the function R(n) for all positive integers n, not just
powers of 2. He then deduces information on the possible structure of an integer 3x + 1
cycle (for the function T (·)), namely that the sum of its even elements must equal the
sum of its odd elements added to the number of its odd elements. Finally he notes that
this fact can be deduced directly without using the FRACTRAN encoding.

77. Kenneth G. Monks and Jonathan Yazinski (2004), The Autoconjugacy of the 3x + 1 func-
tion, Discrete Mathematics 275 (2004), No. 1, 219–236. MR2026287 (2004m:11030).

This paper studies the iteration of the 3x + 1 map T (x) on the 2-adic integers
Z2 . It shows that the set of Aut(T ) = {U ∈ Aut(Z2 ) : U T U −1 = T } consists of the

27
 identity map and a map Ω = Φ ◦ V ◦ Φ−1 where V (x) = −1 − x is the map reversing
the bits in a 2-adic integer and Φ is the 3x + 1 Conjugacy map studied in Bernstein
and Lagarias (1996). It formulates the Autoconjugacy Conjecture that Ω(Qodd ) ⊆ Qodd ,
and proves this conjecture is equivalent to no rational number with odd denominator
having a divergent T -orbit. It defines a notion of self-conjugate cycle under the 3x + 1
map, which is a periodic orbit C such that Ω(C) = C. It proves that {1, 2} is the only
self-conjugate cycle of integers. It shows that all self-conjugate cycles consist of positive
rational numbers.

78. Kenneth M. Monks (2006), The sufficiency of arithmetic progressions for the 3x + 1
conjecture, Proc. Amer. Math. Soc. 134 (2006), No. 10, 2861–2872. MR2231609
(2007c:11030).

This paper shows that the 3x + 1 conjecture is true if it is true for all the integers in
any arithmetic progression {A + Bn : n ≥ 0}, provided A ≥ 0, B ≥ 1. It gives analogous
reductions for the divergent orbits conjecture and the nontrivial cycles conjecture.

79. Maria Monks (2009), Endomorphisms of the shift dynamical systems, discrete deriva-
tives, and applications, Discrete Math. 309 (2009), 5196–5205. (MR 2010i:37217)

The continuous endomorphisms of the one-sided shift dynamical system S on the

2-adic integers are induced by block maps from binary sequences of length n to those of
length 1. Let D denote the endomorphism associated to 00 7→ 0, 01 7→ 1, 10 7→ 1, 11 7→ 0.
Also let V (x) = 1 − x. A map h : Z2 → Z2 is solenoidal if for each k ≥ 1 the first k digits
of the input vector x uniquely determine the first k digits of the output vector h(x). The
main result is the only continuous endomorphisms of S whose parity vector is solenoidal
are D, V ◦ D, S, and V ◦ S.
The map D is interpreted as a “discrete derivative,” and its dynamics under iteration are
studied. The 2-adic integers which under iteration reach a fixed point are characterized.
It is shown that each integer N ∈ Z is eventually periodic under iteration of D, with
minimal period a power of 2.
Finally, a necessary and sufficient condition for the truth of the 3x + 1 conjecture is given
in terms of these concepts. There is a unique conjugacy R taking D to the 3x + 1 map
T , i.e R ◦ D ◦ R−1 = T . In Theorem 5.2 the 3x + 1 conjecture is shown equivalent to the
action of R−1 on positive integers m being eventually periodic with preperiod of length
2 and period of length a power of 2.

80. Helmut Müller (2009), Uber Periodenlängen and die Vermutungen von Collatz und Cran-
dall, [On period lengths and the conjectures of Collatz and Crandall], Mitt. Math. Ges.
Hamburg, 28 (2009), 121–130.

[I have not seen this paper.]

81. Tomas Oliveira e Silva (2004+), Computational verification of 3x + 1 conjecture, web

document at http://www.ieeta.pt/ ~ tos/; email: [email protected].

In Oliveira e Silva (1999) the author reported on computations verifying the 3x + 1

conjecture for n < 3 · 253 = 2.702 × 1016 . In 2004 he implemented an improved version of

28
 this algorithm. As of February 2008 his computation verified the 3x+1 conjecture up to
17·258 > 4.899× 1018 . This is the current record value for verifying the 3x+ 1 conjecture.
Compare Roosendaal (2004+).

82. Reiko Ohira and Michinori Yamashita (2004), A Generalization of the Collatz problem
(Japanese), PC Literacy [Pasocon Literacy] (Personal Computer Users Application Tech-
nology Association] 31 (2006), No. 4, pp. 16–21.

The paper presents the 3x = 1 function and then describes some speed-ups of
the iteration, by only stopping at steps that are odd numbers, with the previous step
dividing by a power 2e with e > 1. It then defines the p-Collatz function for an odd p by

px + (p − 2)


 if x ≡ 1 (mod 2)
 2
fp (x) =
 x
if x ≡ 0 (mod 2) .


2

The authors study cycles of these functions. They note the identity fp (p−2) = (p−2)(p+1)
2 .
This yields for p = 2m − 1 that fp (p − 2) = (p − 2)2m−1 , so that p − 2 is in a periodic
orbit; the case p = 3 gives the trivial cycle of the 3x = 1 function. It notes that
2
fp (p2 − 4) = (p−2)(p−1)
2 . This implies for p = 2m − 1 then fp (p2 − 4) = (p − 2)22m−1 , so
this enters the periodic orbit given by p − 2. The paper gives a table of periodic orbits
found for various smallp including p = 5, 7, 9, 15, 17, 25, 27, 29. (Note that for p ≥ 5 most
trajectories of the map fp are expected to be divergent.)

83. Reiko Ohira and Michinori Yamashita(2004), On the p-Collatz problem (Japanese), PC
Literacy [Pasocon Literacy] (Personal Computer Users Application Technology Associa-
tion] 31 (2006), No. 4, pp. 61–64.

[I have not seen this paper.]

84. Pan, Hong-liang (2000), Notes on the 3x+1 problem (Chinese), Journal of Suzhou Univer-
sity, Natural Science [Suzhou da xue xue bao. Zi ran ke xue ban] 16 (2000), No. 4, 13–16.

The author considers the 3x + 1 function T (n), and first shows the following state-
ment is equivalent to the 3x + 1 Conjecture: The smallest element Pn reached in the
trajectory of n ≥ 1 always satisfies Pn ≡ 1 (mod 4).
Next, given n ≥ 1, let yk (n) = 1 (resp. −1) according as the k-th iterate T (k) (n) is odd
(resp. even), and define

T (k) (n)
Dk (n) := .
2y0 (n)+y1 (n)+···+yk−1 (n)
The author proves that Dk+1 (n) ≤ Dk (n) for all k ≥ 1. Since Dk (n) is nonnegative it
follows that the limit Dn := limk→∞ Dk (n) exists. For initial values that enter the trivial
cycle, one has Dn > 0, since eventually the yk (n) alternate between +1 and −1. The
author shows that Dn = 0 for all initial values n ≥ 1 that either enter a non-trivial cycle
or else have a divergent trajectory. It follows that the 3x + 1 conjecture is equivalent to
the assertion that Dn > 0 for all n ≥ 1.

29
85. Joseph L. Pe (2004), The 3x + 1 Fractal, Computers & Graphics 28 (2004), 431–435.

This paper considers iteration of the following extension of the Collatz function
to complex numbers z, which he terms the complex Collatz function. Define C(z) = 2z
if ⌈|z|⌉ is an even integer, and C(z) = 3z + 1 otherwise. A complex number has the
tri-convergence property if its iterates contain three subsequences which converge to 1, 4
anbd 2, respectively. The 3x + 1 conjecture now asserts that all positive integers have the
tri-convergence property. He gives a sufficient condition for a complex number to have
this property, and uses it to show that z = 1 + i has the tri-convergence property. He
states that it is unlikely that z = 3+5i has this property. The 3x+1 problem now asserts
that all positive integers have the tri-convergence property. He gives some density plots
of iterates exhibiting where they are large or small; self-similarity patterns are evident
in some of them. The author makes conjectures about some of these patterns, close to
the negative real axis.

86. Yuval Peres, Károly Simon and Boris Solomak (2006), Absolute continuity for random
iterated function systems with overlaps, J. London Math. Soc. 74, No. 2 (2006), 739–
756. MR2286443 (2007m:37053).

This paper contains results which apply to a question of Ya. G. Sinal which was
motivated by the 3x + 1 iteration. It is stated in Section 3, as follows. Let 0 < a < 1 be
constant let Zi (i ≥ 1) be independent identically distributed discrete random variables
taking values 1 + a or 1 − a, each with probability 21 . Form the random variable

X := 1 + Z1 + Z1 Z2 + ... + Z1 Z2 · · · Zn + ...

With probability one this sum converges, and the distribution of X is given by a measure
ν a supported on the interval Ia = [ a1 , +∞). Sinai asked for which values of a is the mea-
sure
√ ν a absolutely continuous with respect to Lebesgue measure. It is known that when
3 a
2 < a < 1, the measure ν is singular with respect to Lebesgue √
measure. Conjecturally
3
ν a is absolutely continuous for almost all a in the interval (0, 2 ), and this is unsolved.

The results of this paper imply the truth of a ”randomly perturbed” version of this
conjecture. In Corollary 3.1 the authors consider
′ ′ ′ ′ ′ ′
X = 1 + Z1 + Z1 Z2 + ... + Z1 Z2 · · · Zn + ...
′
in which Zi := Zi Yi , in which each Yi is drawn from a fixed absolutely continuous
distribution Y supported on (1 − ǫ1 , 1 + ǫ2 ) for small positive ǫ1 , ǫ2 , having a bounded
density and expectation E[log Y ] = 0. The Yi and Zi are drawn independently of all other
′
Yi and all other Zi . Let νya be the conditional distribution of Zi given a fixed sequence
of draws y = (y1 , y2 , y3 , ...) for all the Yi . The conclusions are:
√
(a) If 0 < a < 23 , then almost every choice of draws y = (y1 , y2 , y3 , ....) yields an
absolutely continuous conditional distribution νya .
√
(b) If 23 ≤ a < 1, then νya is always singular with respect to Lebesgue measure. For
1
almost all y the Hausdorff dimension of the support of νya is (2 log 2)/ log( 1−a 2 ).

The paper contains many other results.

30
87. Qu, Jing-hua (2002), The 3x + 1 problem and its necessary and sufficient condition (Chi-
nese), Journal of Sangqiu Teacher’s College (2002), No. 5.

[I have not seen this paper.]

88. Eric Roosendaal (2004+), On the 3x + 1 problem, web document, available at:
http://www.ericr.nl/wondrous/index.html

The author maintains an ongoing distributed search program for verifying the 3x + 1
Conjecture to new records and for searching for extremal values for various quantities as-
sociated to the 3x + 1 function. These include quantities termed the glide, delay, residue,
completeness, and gamma. Many people are contributing time on their computers to
this project.
As of February 2008 the 3x + 1 Conjecture is verified up to 612 × 250 ≈ 6.89 × 1017 . The
largest value γ(n) found so far is 36.716918 at n = 7, 219, 136, 416, 377, 236, 271, 195 ≈
7.2 × 1021 .
[The current record for verification of the 3x+1 conjecture published in archival literature
is that of Oliveira e Silva (1999). Note that Oliveira e Silva has extended his computations
to 4.899 × 1018 , the current record.]
89. Jean-Louis Rouet and Marc R. Feix (2002), A generalization of the Collatz problem.
Building cycles and a stochastic approach, J. Stat. Phys. 107, No. 5/6 (2002), 1283–
1298.
(MR 2003i:11035).

The paper studies the class of functions U (x) = (li x + mi )/n if x ≡ i (mod n),
with ili + mi ≡ 0 ( mod n). These functions include the 3x + 1 function as a special case.
They show that there is a bijection between the symbolic dynamics of the first k iterations
and the last k digits of the input x written in base n if and only if n is relatively prime
to the product of the li . They show that for fixed n and any given {mi : 1 ≤ j ≤ k} and
can find a set of coefficients {li : 0 ≤ i ≤ n − 1} and {mi : 0 ≤ i ≤ n − 1} with n relatively
prime to the product of the li which give these values as a k-cycle, U (mi ) = mi+1 and
U (mk ) = m0 . They give numerical experiments indicating that for maps of this kind on k
digit inputs (written in base n) “stochasticity” persists beyond the first k iterations. For
the 3x + 1 problem itself (with base n = 2), it is believed that “stochasticity” persists for
2
about c0 k iterations, with c0 = log (4/3) = 4.8187, as described in Borovkov and Pfeifer
2
(2000), who also present supporting numerical data.
90. Giuseppe Scollo (2005), ω-rewriting the Collatz problem, Fundamenta Informaticae 64
(2005), 401–412. [MR 2008g:68060, Zbl 1102.68051]

This paper reformulates the Collatz iteration dynamics as a term rewriting sys-
tem. ω-rewriting allows infinite input sequences and infinite rewriting. In effect the
dynamics is extended to a larger domain, allowing infinitary inputs. The author shows
the inputs extend to the the 3x + 1 problem on rationals with odd denominator. The
infinitary extension seems analogous to extending the 3x + 1 map to the 2-adic integers.
91. Douglas J. Shaw (2006), The pure numbers generated by the Collatz sequence, Fibonacci
Quarterly, 44 , No. 3, (2006), 194-201.

31
 A positive integer is called pure if its entire tree of preimages under the Collatz
map C(x) contains no integer that is smaller than it is; otherwise it is called impure.
Equivalently, an integer n is impure if there is some r < n with C (k) (r) = n for some
k ≥ 1. Thus n = 4 is impure since C (5) (3) = 4. This paper develops congruence
conditions characterizing pure and impure numbers, e.g. all n ≡ 0 (mod 18) are im-
pure, while all n ≡ 9 (mod 18) are pure. It proves that the set of pure numbers and
impure numbers each have a natural density. The density of impure numbers satisfies
91 2
162 < d < 3 . The subtlety in the structure of the set of pure numbers concerns which
numbers n 6≡ 0 (mod 3) are pure.
92. Shi, Qian Li (2005a), Properties of cycle sets (Chinese), J. Yangzte Univ. Natural Sci-
ence. [Changjiang daxue xuebao. Zi ke ban] 2 (2005), No. 7, 191–192. (MR 2241924)

[I have not seen this paper.]

93. Shi, Qian Li (2005b), Properties of the 3x+1 problem (Chinese), J. Yangzte Univ. Natural
Science. [Changjiang daxue xuebao. Zi ke ban] 2 (2005), No. 10, 287–289. (MR 2241263)

[I have not seen this paper.]

94. John L. Simons (2005), On the non-existence of 2-cycles for the 3x + 1 problem, Math.
Comp. 74 (2005), 1565–1572. MR2137019(2005k:11050).

The author defines an m-cycle of the 3x + 1 problem to be an periodic orbit of

the 3x + 1 function that contains m local minima, i.e. contains m blocks of consecutive
odd integers. This is equal to the number of descents in the terminology of Brox (2000),
which correspond to local maxima. The result of Steiner (1978) shows there are no non-
trivial 1-cycles of the 3x + 1 problem on the positive integers. The author proves there
are no non-trivial 2-cycles for the 3x + 1 function. See Simons and de Weger (2005) for
an impressive generalization of this result.
95. John L. Simons (2007), A simple (inductive) proof for the non-existence of 2-cycles for
the 3x + 1 problem, J. Number Theory 123 (2007), 10-17.

This paper gives complements Simons (2005) by giving another proof of the non-
existence of 2-cycles (periodic orbits containing exactly two blocks of consecutive odd
elements) for the 3x + 1 function on the positive integers. The last section sketches a
proof that the 3x − 1 function on the positive integers has a single 2-cycle with minimal
element n0 = 17; this is equivalent to showing that the 3x + 1 function on the negative
integers has a single 2-cycle, starting from n0 = −17. The author’s method applies to
the 3x± q problem, with fixed q with gcd(6, q) = 1 to find a finite list of 2-cycles; however
it does not extend to classify m-cycles with m ≥ 3. For results on m-cycles, see Simons
and de Weger(2005).
96. John L. Simons (2008), On the (non)-existence of m-cycles for generalized Syracuse se-
quences, Acta Arithmetica 131 (2008), No. 3, 217–254. (MR 2008m:11056)

The author defines an m-cycle of the 3x + 1 problem to be a periodic orbit of

the 3x + 1 function that contains m local minima, i.e. contains m blocks of consecutive

32
 odd integers. This paper genralizes a proof of Simons and de Weger (2005) to give criteria
for the non-existence of m-cycles for the 3x + 1 problem of generalized Collatz sequences
such as the 3x + q problem, the px + 1 problem and the inverse Collatz problem. (Note
that such problems may have m-cycles for various small values of m.)

97. John L. Simons (2008), Exotic Collatz Cycles, Acta Arithmetica 134 (2008), 201–209.
(MR 2438845)

The author considers the px + q problem, with gcd(p.q) = 1, p ≥ 5 odd. He

studies primitive m-cycles. He shows, modulo a conjecture, that for each p there exist
a q with the number of cycles being arbitrarily large. He calls such examples “exotic”
because empirically such problems appear to have most orbits diverging to infinity.

98. John L. Simons (2008a+), Post-transcendence conditions for the existence of m-cycles
for the 3x + 1 problem, preprint.

The author defines an m-cycle of the 3x + 1 problem to be a periodic orbit of

the 3x + 1 function that contains m local minima, i.e. contains m blocks of consecutive
odd integers. This paper extends the bounds of Simons and de Weger for non-existence
of m-cycles for the Collatz problem to larger values of m, including all m ≤ 73.

99. John L. Simons (2008b+), On isomorphism between Farkas sequences and Collatz se-
quences, preprint.

The author considers generalizations of sequences studied by Farkas (2005). These

are sequences taking odd integers to odd integers, of two types, F1 (a, b, c, d(x) and
F2 (a, b, c, d)(x) where a, b, c, d are odd integers. They are given by
(
ax+b
2 if x ≡ 1(mod 4);
F1 (x) = cx+d
2 if x ≡ 3(mod 4).
and
x

3
 if x ≡ 3, 9 (mod 12);
F2 (x) = ax+b
2 if x ≡ 1, 5 (mod 12);
 cx+d

2 if x ≡ 7, 11 (mod 12).
He introduces a notion of isomorphism between orbits {xn } and {yn } of two recurrence
sequences, namely that there are nonzero integers α, β and an integer γ such that

αxk + βyn = γ, for all n ≥ n0 .

This notion takes place on the orbit level and is weaker than that of conjugacy of the
maps. For example, he observes that each 3x + 1 orbit is isomorphic to some 3x + q orbit.
The author shows that the some sequences F1 (a, b, c, d)(x) are isomorphic to some px + q
orbits. He shows that no F2 (a, b, c, d(x) orbit is isomorphic to a px + 1-map orbit.

100. John L. Simons and Benne M. M. de Weger (2004), Mersenne en het Syracuseprobleem
[Mersenne and the Syracuse problem] (Dutch), Nieuw Arch. Wiskd. 5 (2004), no. 3,
218–220. (MR2090398).

33
 This paper is a brief survey of the work of the authors on cycles for the 3x + 1
problem, given in Simons(2005) and Simons and de Weger (2005). It also considers cy-
cles for the 3x + q problem, with q > 0 an odd number. It defines the invariant S(q) to
be the sums of the lengths of the cycles of the 3x + 1 function on the positive integers.
The invariant S(q) is not known to be finite for even a single value of q, though the
3x + 1 conjecture implies that S(1) = 2. The paper observes that S(3t ) = S(1), for each
t ≥ 1. It also considers the case that q = Mk := 2k − 1 is a Mersenne number. It gives
computational evidence suggesting that the minimum of S(Mk2 ) − S(Mk ), taken over all
k ≥ 3, occurs for k = 3, with S(M3 ) = 6 and S(M32 ) = 44.

101. John L. Simons and Benne M. M. de Weger (2005), Theoretical and computational
bounds for m-cycles of the 3n + 1 problem, Acta Arithmetica, 117 (2005), 51–70. (MR
2005h:11049).

The authors define an m-cycle of the 3x + 1 problem to be an orbit of the 3x + 1

function that contains m local minima, i.e. contains m blocks of cosecutive odd inte-
gers. This is equal to the number of descents in the terminology of Brox (2000), which
correspond to local maxima. These methods rule out infinite classes of possible symbol
sequences for cycles. The author’s main result is that for each fixed m there are only
finitely many m-cycles, there are no non-trivial such cycles on the positive integers for
1 ≤ m ≤ 68, and strong constraints are put on any such cycle of length at most 72. The
finiteness result on m-cycles was established earlier by Brox (2000). To obtain their sharp
computational results they use a transcendence result of G. Rhin [Progress in Math., Vol
71 (1987), pp. 155-164] as well as other methods, and extensive computations. [Sub-
sequent to publication, the authors showed there are also no notrivial such cycles for
69 ≤ m ≤ 74, see http://www.win.tue.nl/∼bdeweger/3n+1 v1.41.pdf]

102. Yakov G. Sinai (2003a), Statistical (3X +1)-Problem, Dedicated to the memory of Jürgen
K. Moser. Comm. Pure Appl. Math. 56 No. 7 (2003), 1016–1028. (MR 2004d:37007).

This paper analyzes iterations of the variant of the 3x + 1 map that removes all
powers of 2 at each step, so takes odd integers to odd integers; the author restricts the
iteration to the set Π of positive integers congruent to ±1( mod 6), which is closed under
the iteration. It gives a Structure Theorem for the form of the iterates having a given
symbolic dynamics. The discussion in section 5 can be roughly stated as asserting: There
is √an absolute constant c > 0 such that “most” 3x + 1 trees (mod 3m ) contain at most
ec m log m 22m /3m nodes whose path to the root node has length at most 2m and which
has exactly m odd iterates. Here one puts a probability density on such trees which for
a given tree counts the number of such nodes divided by the total number of such nodes
summed over all trees, and “most” means that the set of such trees having the property
contains 1 − O(1/m) of the total probability, as m → ∞. (The total number of such
1
nodes is 22m , and the total number of trees is 2 · 3m−1 .) Furthermore at least m of the
probability is distributed among trees having at most M c 22m /3m such nodes. From this
latter result follows an entropy inequality (Theorem 5.1) which is the author’s main re-
sult: The entropy Hm of this probability distribution satisfies Hm ≥ m log 3 − O(log m).
For comparison the uniform distribution on [1, 3m ] has the maximal possible entropy
H = m log 3. He conjectures that the entropy satisfies Hm ≥ m log 3 − O(1).
See Kontorovich and Sinai (2002) for related results on the paths of iterates of 3x + 1-like

34
 maps.

103. Yakov G. Sinai (2003b), Uniform distribution in the (3x + 1) problem, Moscow Math.
Journal 3 (2003), No. 4, 1429–1440. (S. P. Novikov 65-th birthday issue). MR2058805
(2005a:11026).

Define the map U (x) taking the set Π of postive integers congruent to ±1 (mod 6)
into itself, given by U (x) = 3x+1
2k
where k = k(x) is the largest power of 2 dividing 3x + 1.
Then consider all the preimages at depth m under U (·) of a given integer y = 6r + δ
with integer 0 ≤ r < 3m and δ = ±1. This consists of the (infinite) set of all integers x
such that U (m) (x) = y. Let such a preimage x have associated data (k1 , k2 , ..., km ) with
kj = k(U (j−1) (x)), and assign to x the weight 2−(k1 +...+km) multiplied by 31 if δ = 1 and
by 32 if δ = −1. Define the mass assigned to y to be the sum of the weights of all its
preimages at depth m. The sum of these masses over 0 ≤ r < 3m and δ = ±1 adds up to
1 by the Structure Theorem proved in Sinai (2003). Let the scaled size of y be the ra-
tional number ρ = 3ym . This now defines a probability distribution P (m) on these values
ρ viewed as a discrete set inside [0, 1]. The main theorem states that as m → ∞ these
probability distributions P (m) weakly converge to the uniform distribution on [0, 1].

104. Yakov G. Sinai (2004), A theorem about uniform distribution, Commun. Math. Phys.
252 (2004), 581–588. (F. Dyson birthday issue) MR2104890 (2005g:37009).

This paper presents a simplified and stronger version of the uniform distribution
theorem given in Sinai (2003b).

105. Matti K. Sinisalo (2003+), On the minimal cycle lengths of the Collatz sequences, preprint.,
Univ. of Oulu, Finland.

This paper shows that the minimal length of a nontrivial cycle of the (3x + 1)-
function on the positive integers is at least 630,138,897. It uses a method similar to that
of Eliahou (1993), and takes advantage of the verification of the 3x + 1 conjecture below
the bound 2.70 × 1016 of Oliveira e Silva (1999). It also considers bounds for cycles of
the (3x − 1)-function.

106. Alain Slakmon and Luc Macot (2006), On the almost convergence of Syracuse sequences,
Statistics and Probability Letters 76, No. 15 (2006), 1625–1630.

The paper shows that the ”random ”Syracuse conjecture is true in the sense that ran-
dom Syracuse sequences get smaller than some specified bound B ≥ 1 almost surely. Con-
sider identical independent 0 − 1 random variables Xn having probability P [Xn = 1] = p,
P [Xn = 0] = q = 1−p. The authors consider the random Syracuse model Sn+1 = 23 Sn + 12 ,
if Xn+1 = 1 and Sn+1 = 21 Sn if Xn+1 = 0, starting from a given S0 . For the actual 3x + 1
problem one would take p = q = 12 . They then consider an auxiliary sequence Yn with
Y0 = S0 , formed by the rule Yn+1 = Yn (1 + σγ) if Xn+1 = 1 and Yn+1 = Yn (1 − σ) if
Xn = 0, where σ and γ are positive constants. They formulate results showing that if
pγ − q > 0 then there is a positive threshold value c such that Yn → ∞ almost surely if
0 < σ < c and Yn → 0 almost surely if σ > c. Finally they show that the parameters
(σ, γ) can be chosen so that σ > c such that for all starting values S0 ≥ B, one has
Tn ≥ Sn holding at every step, as long as Tn ≥ B, while Tn → 0 with probability one.

35
 The conclude that some Sn ≤ B with probability one.
107. Bart Snapp and Matt Tracy (2008), The Collatz problem and Analogues, J. Integer Se-
quences 11 (2008), Article 08.4.7, 10pp. (MR 2009i:11144)

The authors study a polynomial analogue of the Collatz problem, as considered

in Hicks et al (2008), and then give an application to the original 3x + 1 problem on Z.
They consider for polynomials in Z/nZ[X] the map

(x + 1)f (x) − f (0)

 if f (x) is not divisible by x
x


Tp (f ) =

 f (x)
if f (x) is divisible by x.


x
They show that for n = 2 that all polynomials eventually iterate to 1 under this map,
but for n ≥ 3 there always exists some polynomial f (x) that never iterates to 1 under
this map. In Section 3 they apply this result for n = 2 to define an enlarged version
of the Collatz graph on (all) integers which they call the combined Collatz graph. This
graph glues together positive and negative integer iterates, adding solid and dotted edges,
representing moves of the 3x + 1 and 3x − 1 iterations., respectively. They observe that
the 3x + 1 conjecture is equivalent to the claim that all positive integers appear in this
graph connected by undotted edges.
108. Tang, Ren Xian (2006), About recursion relation of 3x + 1 Conjecture (Chinese), J. of
Hunan Univ. Science Engineering [Hunan ki ji da xue xue bao] (2006), No. 5

[I have not seen this paper.]

109. Toshio Urata (2002), Some holomorphic functions connected with the Collatz problem,
Bulletin Aichi Univ. Education (Natural Science) [Aichi Kyoiku Daigaku Kenkyu hokoku.
Shizen kagaku.] 51 (2002), 13–16.

The author constructs an entire function F (z) which interpolates the values of the
speeded up Collatz function φ(n) which takes odd integers to odd integers by dividing
out all powers of 2, i.e. for an odd integer n, φ(n) = 2e3n+1
(3n+1) , where e(m) = ord2 (m).
Furthermore the entire function F (z) is constructed so that F ′ (z) = 0 at all positive odd
integers. The author analyzes the holomorphic dynamics of iterating F (z). He observes
that z = 1 is a superattacting fixed point, that z = 0 is a repelling fixed point, and
1
that there is an attracting fixed point z on the negative real axis with − 20 < z < 0. He
proves that all positive odd integers are in the Fatou set of F (z), and that all negative
odd integers are in the Julia set of F (z). He observes that the intersection of the Julia
set with the negative real axis coincides with the inverse iterates of z = 0. Finally he
notes that every component of the Fatou set is simply connected.
110. Toshio Urata (2003), The Collatz problem over 2-adic integers, Bulletin Aichi Univ. Ed-
ucation (Natural Science) [Aichi Kyoiku Daigaku Kenkyu hokoku. Shizen kagaku.] 52
(2003), 5–11.

This is an English version of Urata (1999). The author studies a 2-adic interpo-
lation of the speeded-up Collatz function φ(n) defined on odd integers n by dividing out

36
 all powers of 2, i.e. for an odd integer n, φ(n) = 2p3n+1
(3n+1) , where p(m) = ord2 (m). Let
Z2 = {x ∈ Z2 : x ≡ 1 (mod 2)} denote the 2-adic units. The author sets OQ := Q ∩ Z∗2 ,
∗

and one has Z∗2 is the closure OQ of OQ ⊂ Z2 . The author shows that the map φ uniquely
1
extends to a continuous function φ : Z∗2 \{ −1 ∗
3 } → Z2 . He shows that if f (x) = 2x + 3 then
f (x) leaves φ invariant, in the sense that φ(f (x)) = φ(x) for all x ∈ Z∗2 \{ −1
3 }. It follows
that f (f (x)) = 4x + 1 also leaves φ invariant.
To each x ∈ Z∗2 \{ −1
3 } he associates the sequence of 2-exponents (p1 , p2 , ...) produced
by iterating φ. He proves that an element x ∈ Z∗2 \{ −1 3 } uniquely determine x; and
that every possible sequence corresponds to some value x ∈ Z∗2 \{ −1 3 } He shows that
all periodic points of φ on Z∗2 are rational numbers x = pq ∈ OQ,and that there is a
unique such periodic point for any finite sequence (p1 , p2 , · · · , pm
P) m−1
of positive integers,
representing 2-exponents, having period m. If C(p1 , p2 , ..., pm ) = j=0 2p1 +···+pj 3m−1−j
then this periodic pont is

C(p1 , ..., pm )
x = R(p1 , p2 ..., pm ) :=
2p1 +···+pm − 3m
He shows that an orbit is periodic if and only if its sequence of 2-exponents is periodic.
Examples are given.

111. Toshio Urata (2007), Collatz’s problem (Japanese), Aichi University of Education Book-
let, Math. Sciences Select 1, 60 pages.

This short book considers many aspects of the 3x + 1 iteration, as discussed in

his papers Urata (2002) (2003), (2005). In particular, it deals with holomorphic dynam-
ics of iterating a function that is a speeded-up version of the Collatz function. It includes
many pictures of parts of the Fatou and Julia set for these dynamics.

112. Toshio Urata and Kazuhiro Hamada (2005), Positive values of a holomorphic function
connected with the Collatz problem, Bulletin Aichi Univ. Education (Natural Science),
[Aichi Kyoiku Daigaku Kenkyu hokoku. Shizen kagaku.] 54 (2005), 1–10

The authors study the entire function F (z) introduced in Urata (2002) which interpo-
3n+1
lates the speeded-up Collatz function φ(n) defined on the odd integers by φ(n) = 2p(3n+1) ,
where p(m) := ord2 (m). They show that F (x) > 0 for all real x > 0. This holds even
though F (z) wildly oscillates on the positive real axis.

113. Tanguy Urvoy (2000), Regularity of congruential graphs, in: Mathematical Foundations
of Computer Science 2000 (Bratislava), Lecture Notes in Computer Science Vol. 1893,
Springer: Berlin, 2000, pp. 680–689. (MR 2002d:68083).

The Collatz problem is viewed as a reachability problem on a directed graph with

vertices labelled by positive integers n, and edges 2n 7→ n, 2n+1 7→ 3n+2, This is an infi-
nite, possibly disconneted graph, call it the Collatz graph. The 3x + 1 Conjecture for this
graph can be formulated as a closed formula in monadic second order logic. The author
considers more generally congruential systems which give infinite graphs with vertices
labelled by positive integers and edges specified by a finite set of rules pn + r 7→ qn + s,
with 0 ≤ r < p, 0 ≤ s < q.

37
 In section 1.4 the author shows that a congruential system without remainders (all rules
have r = s = 0) can be encoded as a Petri net with one free place. The reachability
problem for a Petri net with initial position n given is known to be a decidable problem.
A large class of infinite graphs called regular graphs have a decidable monadic second
order theory, see Courcelle [Handbook of Theoretical Computer Science, Volume B, El-
sevier: New York 1990, pp. 193-242.] In section 3 the author shows that every regular
graph with bounded vertex degrees can be obtained as a graph of some congruential sys-
tem. In section 4 the author shows that the Collatz graph, first viewed as an undirected
graph, and then directed with any choice of orientations of the edges, is never a regular
graph.

114. Jean Paul Van Bendegem (2005), The Collatz Conjecture: A Case Study in Mathemati-
cal Problem Solving, Logic and Logical Philosophy 14, No. 1 (2005), 7–23. (MR 2163301)

This philosophical essay concerns the issues of what mathematicans do beyond prov-
ing theorems. The work on the 3x + 1 problem is discussed from this viewpoint. Such
work includes: computer experiments, heuristic arguments concerning the truth of the
conjecture, metamathemical heuristics concerning the likelihood of finding a proof, etc.

115. Stanislav Volkov (2006), A probabilistic model for the 5k + 1 problem and related prob-
lems, Stochastic Processes and Applications 116 (2006), 662–674.

This paper presents a stochastic model for maps like the 5x + 1 problem, in which
most trajectories are expected to diverge. For the 5x + 1 problem it is empirically ob-
served that the number of values of n ≤ x that have some iterate equal to 1 appears to
grow like xα , where α ≈ 0.68. The author develops stochastic models which supply a
heuristic to estimate the value of α.
The stochastic model studied is a randomly labelled (rooted) binary tree model. At
each vertex the left branching edge of the tree gets a label randomly drawn from a
(discrete) real distribution X and the right branching edge gets a label randomly drawn
from a (discrete) real distribution Y. Each vertex is labelled with the sum of the edge
labels from the root; the root gets label 0. The rigorous results of the paper concern
such stochastic models. The author assumes that both X and Y have positive expected
values µx , µy , but that at least one random variable assumes some negative values. He
also assumes that the moment generating functions of both variables are finite for all
parameter values.
The author first considers for each real α > 0 the total P number of vertices Rn (α) at
depth n in the tree having label ≤ nα, and sets R(α) = ∞ n=1 Rn (α). He defines a large
deviations rate function γ(α) associated to the random variable W that draws from −X
or −Y with equal probability. He derives a large deviations criterion (Theorem 1) which
states that if γ(−α) > log 2 then R(α) is finite almost surely, while if γ(−α) < log 2 then
R(α) is infinite almost surely. The author next studies the quantity Q(x) counting the
number of vertices with labels smaller than x. This is a refinement of the case α = 0
above. He supposes that γ(0) > log 2 holds, and shows (Theorem 2) that this implies
that Q(x) is finite almost surely for each x. He then shows (Theorem 3) that
1
β := lim log Q(x)
x→∞ x

38
 exists almost surely and is given by
1
β := max (log 2 − γ(−a)).
a∈(0, 12 (µx +µy )] a

He constructs a particular stochastic model of this kind that approximates the 5x + 1

problem. For this model he shows that Theorem 3 applies and computes that β ≈ 0.678.
The author observes that his stochastic model has similarities to the branching random
walk stochastic model for the 3x+1 problem studied in Lagarias and Weiss (1992), whose
analysis also used the theory of large deviations.
Note. The exponential branching of the 5x + 1 tree above 1 allows one to prove that the
number of such n ≤ x that have some 5x + 1 iterate equal to 1 is at least xβ for some
small positive β.

116. Wang, Nian-liang (2002), The way to prove Collatz problem (Chinese), Journal of Shangluo
University (2002), No. 2.

[I have not seen this paper.]

117. Wang, Xing-Yuan; Wang, Qiao Long; Fen, Yue Ping; and Xu, Zhi Wen (2003), The
distribution of the fixed points on the real axis of a generalized 3x + 1 function and some
related fractal images (Chinese), Journal of Image and Graphics [Zhongguo tu xiang tu
xing xue bao] 8(?) (2003), No. 1.

[I have not seen this paper.]

118. Xing-yuan Wang and Xue-jing Yu (2007), Visualizing generalized 3x+ 1 function dynam-
ics based on fractal, Applied Mathematics and Computation 188 (2007), no. 1, 234–243.
(MR2327110).

This paper studies two complex-valued generalizations of the Collatz function. It

replaces the function mod2 (x) defined on integers x by the function (sin pix 2
2 ) , defined
for complex x. These are then substituted in the definitions
x
C(x) = (1 − mod2 (x)) + (3x + 1)mod2 (x)
2
and
3mod2 (x) + mod2 (x)
T (x) = .
21−mod2 (x)
Both these functions agree with the Collatz function C(n) on the positive integers. The
first function simplifies to
1
C(x) = (7x + 2 − (5x + 2) cos πx).
4
The paper studies iteration of the entire functions C(x) and T (x), from the viewpoint of
escape time, stopping time and total stopping time. It presents graphics illustrating the
results of the algorithms. The total stopping time plots exhibit vaguely Mandlebrot-like
sets of various sizes located around the positive integers.

39
119. Wang, Xing-yuan and Yu, Xue-jing (2008), Dynamics of the generalized 3x + 1 function
determined by its fractal images, Progress Natural Science (English Edition) 18 (2008),
217–223. (MR 2009i:37020).

From the abstract: “Two different complex maps are obtained by generalizing the
3x + 1 function to the complex plane, and fractal images for these two complex maps are
constructed by using escape-time, stopping time and total-stopping-time arithmetic.”
120. Günther J. Wirsching (2000), Über das 3n+1 Problem, Elem. Math. 55 (2000), 142–155.
(MR 2002h:11022).

This is a survey paper, which discusses the origin of the 3n + 1 problem and re-
sults on the dynamics of the 3x + 1 function.
121. Günther J. Wirsching (2001), A functional differential equation and 3n + 1 dynamics,
in: Topics in Functional Differential and Functional Difference Equations (Lisbon 1999),
(T. Faria, E. Frietas, Eds.), Fields Institute Communications No. 29, Amer. Math. Soc.
2001, pp. 369–378. (MR 2002b:11035).

This paper explains how a functional differential equation arises in trying to un-
derstand 3n + 1 dynamics. as given in Wirsching (1998a). It analyzes some properties
of its solutions.
122. Günther J. Wirsching (2003) On the problem of positive predecessor density in 3N + 1
dynamics, Disc. Cont. Dynam. Syst. 9 (2003), no. 3, 771–787. (MR 2004f:39028).

This paper discusses an approach to prove positive predecessor density, which for-
mulates three conjectures which, if proved, would establish the result. This approach
presents in more detail aspects of the approach taken in the author’s Springer Lecture
Notes volume, Wirsching (1998a).
123. Wu, Jia Bang and Huang, Guo Lin (2001), On the traditional iteration and the elongate
iteration of the 3N + 1 conjecture (Chinese), Journal of South-Central University for
Nationalities. Natural Sciences Ed. [Zhong nan min zu da xue xue bao. Zi ran ke xue
ban] 21 (2001), No. 4

[I have not seen this paper.]

124. Wu, Jia Bang and Hao, Shen Wang (2003), On the equality of the adequate stopping
time and the coefficient stopping time of n in the 3N + 1 conjecture (Chinese, English
summary), J. Huazhong Univ. Sci. Technol. Nat. Sci. [Hua zhong gong xue yuan] 31
(2003), no. 5, 114–116. (MR 2000420)

English summary: ”In the paper the equality of adequate stopping time ta (n) and
the coefficient stopping time tc (n) in the 3N + 1 conjecture was discussed. It was proved
Pk−1
that ta (n) = tc (n) on condition that d = i=1 xi (n) is not very large. Therefore it was
conjectured that when the bound condition for d was unnecessary, or was automatically
satisfied, tc (n) = ta (n) in all cases.”
Note: The notion of coefficient stopping time was introduced in Terras (1976) for the
3x+1 function; here tc (n) is its analogue for the Collatz function. The adequate stopping

40
 time ta (n) is the analogue of the stopping time for the Collatz function. Terras (1976)
made a conjecture equivalent to asserting the equality ta (n) = tc (n) always holds.

125. Wu, Jia Bang and Huang, Guo Lin (2001a), Family of consecutive integer pairs of the
same height in the Collatz conjecture (Chinese, English summary), Mathematica Ap-
plicata (Wuhan) [Ying yung shu hsüeh] 14 (2001), suppl. 21–25. (MR 1885838, Zbl
1002.11023)

English summary: ”In the Collatz conjecture, certain runs of consecutive integers
have the same height. In particularly in this paper, pairs of successive integers of the
same height are investigated. It is found that families of consecutive integer pairs of the
same height occur infinitely often and in different patterns.”
Note. The authors show coalescence of iteration of the 3x + 1 function for consecutive
pairs in arithmetic progessions of form {(2k m + r, 2k m + r + 1) : m ≥ 0}, for example
(32m + 5, 32m + 6), (64m + 45, 64m + 46), (128m + 29, 128m + 30). They exhibit infinitely
many such arithmetic progressions, specified by the pattern of residues (modulo 2) of the
successive iterates of the 3x + 1 function, e.g. for (128m + 29, 128m + 30) the patterns
are (1001101), (0111100), respectively.

126. Wu, Jia Bang and Huang, Guo Lin (2001b), Elongate iteration for the 3N + 1 Conjecture
(Chinese, English summary), J. Huazhong Univ. Sci. Tech. [Hua zhong gong xue yuan]
29 (2001), no.2 , 112–114. (MR 1887558)

English summary: ”The concept of Elongate Iteration for the 3N + 1 conjecture

is presented and discussed. Some results of Elongate Iteration are given: a. The corre-
spondence of numbers with parity vectors. b. Invariability of l-tuple. c. Proof of term
formula of n. d. Some equivalence propositions for the 3N + 1 conjecture. e. A property
of coefficient stopping time tc (n), etc.”

127. Wu, Wen Quan (2003), An equivalent proposition of Collatz conjecture (Chinese), Journal
of Aba Teacher’s College [Aba shi fan gao deng zhuan ke xue xiao xue bao] (2003), No. 3.

[I have not seen this paper.]

128. Xia, Ye (2003), Some thoughts about 3X + 1 Conjecture (Chinese), [Zhongxue Shuxue Za
Zhi (Chu zhong ban)] (2003), No. 6.

[I have not seen this paper.]

129. Xie, Jiu Guo (2006), The interesting 3x + 1 problem (Chinese), J. of Hunan Univ. Sci-
ence Technology [Hunan ki ji da xue xue bao] (2006), No. 11

[I have not seen this paper.]

130. Michinori Yamashita (2002), 3x + 1 problem from the (e, k)-perspective (Japanese), PC
Literacy [Pasocon Literacy] (Personal Computer Users Application Technology Associa-
tion] 27 (2002), No. 10, 22–27.

[I have not seen this paper.]

41
131. Michinori Yamashita (2006), Note on the 3x± 1 Problem (Japanese), PC Literacy [Paso-
con Literacy] (Personal Computer Users Application Technology Association] (2006).

This paper studies the 3x + 1 function f (x) = T (x) and the 3x − 1 function
g(x) = −T (−x). It lets (e, k) = 2e k − 1, [e, k] = 2e k + 1, where k is an odd inte-
ger. It proves a number of identities for these functions, such as 3(e, k) = (1, (e − 1, 3k)),
given a table of what happens when one multiplies by a small number l. It studies
patterns of successive iterations.
132. Roger E. Zarnowski (2001), Generalized inverses and the total stopping time of Collatz
sequences, Linear and Multilinear Algebra 49 (2001), 115–130. (MR 2003b:15011).

The 3x + 1 iteration is formulated in terms of a denumberable Markov chain with

transition matrix P . The 3x + 1 Conjecture is reformulated in terms of the limiting
behavior of P k . The group inverse A♯ to an n × n matrix A is defined by the proper-
ties AA♯ = A♯ A, AA♯ A = A and A♯ AA♯ = A♯ , and is unique when it exists. Now set
A = I − P , an infinite matrix. Assuming there are no nontrivial cycles, the group inverse
A♯ exists, and satisfies limk→∞ P k = I − AA♯ . An explicit formula is given for A♯ .
133. Roger E. Zarnowski (2008), The congruence structure of the 3x+ 1 map, Fibonacci Quar-
terly 46/47 (2008/09), no. 2, 115–125. (MR 2010c:11037)

This paper studies the structure of forward iteration of the 3x + 1 map taking con-
gruence classes ( mod 2n ) to congruence classes ( mod 3k ) where k depends on the parity
sequence of the iterates. He describes relations of image classes (which may overlap),
using a concept termed “congruence class triangle.”
This paper observes that the set A of nonnegative integers relatively prime to 3 is an
invariant set under forward iteration of the 3x+1 map. It then shows that under iteration
the set of positive integers in any fixed congruence class (mod 2a 3b ) has forward orbits
visiting every integer in A infinitely often. Thus to prove the 3x + 1 conjecture it suffices
to verify it on integers in any one of these congruence classes.
Note. There are quite a few prior results known on sufficiency of 3x + 1 conjecture on a
“thin” congruence implying it is true in general, e.g. Korec and Znam (1987).
134. Zhou, Yun Cai and Zhou, Bao Lan (2007), On the Collatz Conjecture (Chinese), J.
Yangtze University Natural Science. [Changjiang daxue xuebao. Zi ke ban] (2007), No.
1, 24–26.

English summary: “The Collatz wave set (CWS) and the 3-1 mapping trim ap-
plied on it are defined, and the properties of CWS under trim mapping are discussed.
Collatz’s conjecture, namely, the 3x + 1 problem, is also discussed from the viewpoint
of CWS and 3-1 mapping, by which a new possibility is provided for proving Collatz’s
conjecture.”

Acknowledgements. I thank M. Chamberland, J. Goodwin, C. Hewish, S. Kohl, Wang

Liang, C. Reiter, J. L. Simons, T. Urata, and S. Volkov and G. J. Wirsching for supplying
references. I thank W.-C. Winnie Li, Xinyun Sun and Yang Wang for help with Chinese ref-
erences. I thank T. Ohsawa for help with Japanese references.

42