Next Article in Journal
Study of Indoor Ventilation Based on Large-Scale DNS by a Domain Decomposition Method
Next Article in Special Issue
Nonlinear Dynamic Modelling of Two-Point and Symmetrically Supported Pipeline Brackets with Elastic-Porous Metal Rubber Damper
Previous Article in Journal
Dynamic Soft Sensor Development for Time-Varying and Multirate Data Processes Based on Discount and Weighted ARMA Models
Previous Article in Special Issue
About the Orbit Structure of Sequences of Maps of Integers
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

An E-Sequence Approach to the 3x + 1 Problem

Faculty of Science, Zhejiang Sci-Tech University, Hangzhou 310018, China
Symmetry 2019, 11(11), 1415; https://doi.org/10.3390/sym11111415
Submission received: 17 October 2019 / Revised: 6 November 2019 / Accepted: 12 November 2019 / Published: 15 November 2019
(This article belongs to the Special Issue Symmetry and Dynamical Systems)

Abstract

:
For any odd positive integer x, define ( x n ) n 0 and ( a n ) n 1 by setting x 0 = x ,   x n = 3 x n 1 + 1 2 a n such that all x n are odd. The 3 x + 1 problem asserts that there is an x n = 1 for all x. Usually, ( x n ) n 0 is called the trajectory of x. In this paper, we concentrate on ( a n ) n 1 and call it the E-sequence of x. The idea is that we generalize E-sequences to all infinite sequences ( a n ) n 1 of positive integers and consider all these generalized E-sequences. We then define ( a n ) n 1 to be Ω -convergent to x if it is the E-sequence of x and to be Ω -divergent if it is not the E-sequence of any odd positive integer. We prove a remarkable fact that the Ω -divergence of all non-periodic E-sequences implies the periodicity of ( x n ) n 0 for all x 0 . The principal results of this paper are to prove the Ω -divergence of several classes of non-periodic E-sequences. Especially, we prove that all non-periodic E-sequences ( a n ) n 1 with lim ¯ n b n n > log 2 3 are Ω -divergent by using Wendel’s inequality and the Matthews and Watts’ formula x n = 3 n x 0 2 b n k = 0 n 1 ( 1 + 1 3 x k ) , where b n = k = 1 n a k . These results present a possible way to prove the periodicity of trajectories of all positive integers in the 3 x + 1 problem, and we call it the E-sequence approach.

1. Introduction

For any odd positive integer x, define two infinite sequences ( x n ) n 0 and ( a n ) n 1 of positive integers by setting:
x 0 = x ,   x n = 3 x n 1 + 1 2 a n
such that x n is odd for all n N = { 1 , 2 , } . The 3 x + 1 problem asserts that there is n N such that x n = 1 for all odd positive integers x. For a survey, see [1]. For recent developments, see [2,3,4,5,6,7].
Usually, ( x n ) n 0 is called the trajectory of x. In this paper, we concentrate on ( a n ) n 1 and call it the E-sequence of x. The idea is that we generalize E-sequences to all infinite sequences ( a n ) n 1 of positive integers. Given any generalized E-sequence ( a n ) n 1 , if it is the E-sequence of the odd positive integer x, it is called Ω -convergent to x and denoted by Ω lim a n = x ; if ( a n ) n 1 is not the E-sequence of any odd positive integer, it is called Ω -divergent and denoted by Ω lim a n = . Subsequently, these generalized E-sequences are also called E-sequences for simplicity.
The 3 x + 1 problem in the form (1.1) should be owed to Crandall and Sander et al., see [8,9]. E-sequences are some variants of Everett’s parity sequences [10] and Terras’ encoding representations [11]. Everett and Terras focused on finite E-sequences resulting from (1.1). What we are concerned with is the Ω -convergence and Ω -divergence of any infinite sequence of positive integers, i.e., the generalized E-sequences.
A possible way to prove the 3 x + 1 problem was devised by Möller as follows (see [12]):
Conjecture 1.
(i) ( x n ) n 0 is periodic for all odd positive integers x 0 ;
(ii) ( 1 , 1 , ) is the unique pure periodic trajectory.
Usually, we can convert one claim about trajectories into the one about E-sequences. As for E-sequences, we have the following conjecture.
Conjecture 2.
Let b n = i = 1 n a i . Then,
(i) all non-periodic E-sequences are Ω -divergent;
(ii) every E-sequence ( a n ) n 1 satisfying 3 n > 2 b n for all n N is Ω -divergent.
Note that Conjecture 2(i) does not hold for some generalizations of the 3 x + 1 problem studied by Möller, Matthews, and Watts in [12,13]; Conjecture 2(ii) implies that there is some n such that 2 b n > 3 n in the E-sequence ( a n ) n 1 of every odd positive integer x, which is a conjecture posed by Terras in [11] about his τ -stopping time.
A remarkable fact is that Conjecture 1(i) is a corollary of Conjecture 2(i) by Theorem 3. This means that the Ω -divergence of all non-periodic E-sequences implies the periodicity of ( x n ) n 1 for all positive integers x. Then, Conjecture 2(i) is of significance to the study of the 3 x + 1 problem. The principal results of this paper are to prove that several classes of non-periodic E-sequences are Ω -divergent. In particular, we prove that:
(i)
All non-periodic E-sequences ( a n ) n 1 with lim ¯ n b n n > log 2 3 are Ω -divergent.
(ii)
If ( a n ) n 0 is 12121112 , where a n = 2 if n { 2 1 , 2 2 , 2 3 , } and a n = 1 , otherwise, then Ω lim a n = ;
(iii)
Let θ 1 be an irrational number, and define a n = [ n θ ] [ ( n 1 ) θ ] , then Ω lim a n = , where [ a ] denotes the integral part of a for any real a.
Note that we prove the above claim (i) by using Wendel’s inequality and the Matthews and Watts’ formula x n = 3 n x 0 2 b n k = 0 n 1 ( 1 + 1 3 x k ) . In addition, it seems that our approach cannot help to prove the conjecture 1(ii) of the unique cycle. For such a topic, see [14].

2. Preliminaries

Let ( a n ) n 1 be an E-sequence. In most cases, there is no odd positive integer x such that ( a n ) n 1 is the E-sequence of x, i.e., Ω lim a n = . However, there always exists x N such that the first n terms of the E-sequence of x are ( a 1 a n ) . Furthermore, for any 1 u v n , there always exists x N such that the first v u + 1 terms of the E-sequence of x are the designated block ( a u a v ) of ( a 1 a n ) , which is illustrated as ( a 1 a u 1 ) ( a u a v ) ( a v + 1 a n ) .
Definition 1.
Define b 0 = 0 , b n = i = 1 n a i , B n = i = 0 n 1   3 n 1 i 2 b i .
Clearly, B 1 = 1 , B n = 3 B n 1 + 2 b n 1 , 2 B n , 3 B n .
Proposition 1.
Let ( x n ) n 1 and ( a n ) n 1 be defined as in (1.1). Then x n = 3 n x + B n 2 b n .
Proof. 
The proof is by a procedure similar to that of Theorem 1.1 in [11] and omitted. □
Proposition 2.
Given any positive integer n, there exist two integers x n and x 0 such that 2 b n x n 3 n x 0 = B n , 1 x n < 3 n , and 1 x 0 < 2 b n .
Proof. 
By gcd ( 2 b n , 3 n ) = 1 , there exist two integers x n and x 0 such that 2 b n x n 3 n x 0 = B n and 1 x n 3 n . Then, x n < 3 n by 3 B n . By B n 1 , we have x 0 = 2 b n x n B n 3 n < 2 b n x n 3 n < 2 b n . Thus, x 0 < 2 b n .
By 2 b n x n 3 n x 0 = B n , we have 2 b n x n B n ( mod   3 n ) . Then 2 b n 1 ( 2 a n x n 1 ) 3 B n 1 ( mod   3 n ) by B n = 2 b n 1 + 3 B n 1 . Thus, 3 | 2 a n x n 1 . Define x n 1 = 2 a n x n 1 3 . Then, x n 1 Z , x n = 3 x n 1 + 1 2 a n and 2 b n 1 x n 1 B n 1 ( mod   3 n 1 ) . Sequentially, define x n 2 , , x 1 such that x n 1 = 3 x n 2 + 1 2 a n 1 , …, x 1 = 3 x 0 + 1 2 a 1 . Then, x i Z for all 0 i n .
Suppose that x 0 < 0 . We then sequentially have x 1 < 0 , , x n < 0 , which contradicts x n 1 . Thus, x 0 1 . □
Note that the validity of Proposition 2 is dependent on the structure of B n . We formulate the middle part of the above proof as the following proposition.
Proposition 3.
Assume that x n , x 0 Z and 2 b n x n 3 n x 0 = B n . Define x 1 = 3 x 0 + 1 2 a 1 ,…, x n 1 = 3 x n 2 + 1 2 a n 1 . Then, x n = 3 x n 1 + 1 2 a n and x i Z for all 0 i n .
Definition 2.
For any 1 u v , define b u u 1 = 0 , b u v = i = u v a i , B u u 2 = 0 , B u u 1 = 1 , B u v = 3 v u + 1 + 3 v u 2 b u u + + 3 1 2 b u v 1 + 2 b u v = i = 0 v u + 1 3 v u + 1 i 2 b u u 1 + i .
Then, b u u = a u , b u u + 1 = a u + a u + 1 , B u u = 3 + 2 a u , B u u + 1 = 3 2 + 3 · 2 a u + 2 a u + a u + 1 , B u v = 3 B u v 1 + 2 b u v = i = u 1 v 3 v i 2 b u i . Clearly, b 1 n and B 1 n 1 are the same as b n and B n , respectively.
Proposition 4.
B n = 3 n u + 1 B 1 u 2 + 3 n 1 v 2 b u 1 B u v + 2 b v + 1 B v + 2 n 1 .
Proof. 
By B 1 u 2 = i = 0 u 2   3 u 2 i 2 b i and B v + 2 n 1 = i = v + 1 n 1 3 n 1 i 2 b v + 2 i , we have:
B n = B 1 n 1 = i = 0 n 1   3 n 1 i 2 b i = i = 0 u 2   3 n 1 i 2 b i + i = u 1 v   3 n 1 i 2 b i + i = v + 1 n 1   3 n 1 i 2 b i = 3 n u + 1 i = 0 u 2   3 u 2 i 2 b i + 3 n 1 v 2 b u 1 i = u 1 v   3 v i 2 b u i + 2 b v + 1 i = v + 1 n 1   3 n 1 i 2 b v + 2 i = 3 n u + 1 B 1 u 2 + 3 n 1 v 2 b u 1 B u v + 2 b v + 1 B v + 2 n 1 .
 □
Definition 3.
For any 1 u v , define two integers x 0 u , v and x v u + 1 u , v such that 2 b u v x v u + 1 u , v 3 v u + 1 x 0 u , v = B u v 1 , 1 x 0 u , v < 2 b u v , and 1 x v u + 1 u , v < 3 v u + 1 . Further, define x 1 u , v = 3 x 0 u , v + 1 2 a u , x 2 u , v = 3 x 1 u , v + 1 2 a u + 1 ,…, x v u u , v = 3 x v u 1 u , v + 1 2 a v 1 .
Clearly, x 0 1 , n and x n 1 , n are the same as x 0 and x n in Proposition 2, respectively.
Proposition 5.
(i)
x v u + 1 u , v = 3 x v u u , v + 1 2 a v ;
(ii)
For any 0 k v u , x k u , v = 3 k x 0 u , v + B u u + k 2 2 b u u + k 1 , and
x v u + 1 u , v = 3 v u + 1 k x k u , v + B u + k v 1 2 b u + k v ;
(iii)
x 0 u , v x 0 u , v + 1 ;
(iv)
Ω lim a n = x if and only if lim n x 0 1 , n = x ;
(v)
Ω lim a n = if and only if lim n x 0 1 , n = .
Proof. 
(i) is from Proposition 3(ii), which is from (i) and Proposition 1.
(iii) By Definition 3, 2 b u v x v u + 1 u , v 3 v u + 1 x 0 u , v = B u v 1 , 2 b u v + 1 x v u + 2 u , v + 1 3 v u + 2 x 0 u , v + 1 = B u v . Then, 3 v u + 1 x 0 u , v + B u v 1 0   ( mod   2 b u v ) , 3 v u + 2 x 0 u , v + 1 + B u v 0   ( mod   2 b u v + 1 ) . Thus, 3 v u + 1 x 0 u , v + 1 + B u v 1 0   ( mod   2 b u v ) by B u v = 3 B u v 1 + 2 b u v . Hence, x 0 u , v x 0 u , v + 1   ( mod   2 b u v ) . Therefore, x 0 u , v x 0 u , v + 1 by 1 x 0 u , v < 2 b u v and 1 x 0 u , v + 1 < 2 b u v + 1 .
By (iii), ( x 0 1 , n ) n 1 is increasing, then (iv) and (v) hold trivially. □
Proposition 5(iv) shows that if Ω lim a n = x , then x 0 1 , n = x for all sufficiently large n. Proposition 5(v) shows the reasonableness of Ω lim a n = .

3. Periodic E-Sequences

Definition 4.
(i)
( a n ) n 1 is periodic if there exist two integers
l 0 , r 1 such that a n = a n + r for all n > l ;
(ii)
r is called the period of ( a n ) n 1 ;
(iii)
( a 1 a l ) and ( a l + 1 a l + r ) are called the non-periodic part and periodic part of ( a n ) n 1 , respectively;
(iv)
( a n ) n 1 is called purely periodic if l = 0 and eventually periodic if l > 0 ;
(v)
The E-sequence is denoted by a 1 a l a l + 1 a l + r ¯ .
Throughout the remainder of this section, define s = b l + 1 l + r , B r = B l + 1 l + r 1 , and let k 0 be an integer.
Proposition 6.
Let a 1 a l a l + 1 a l + r ¯ be a periodic E-sequence. Then, B r k + l = 3 r k B l + 2 b l B r 3 r k 2 s k 3 r 2 s .
Proof. 
By Proposition 4, B r k + l = B 1 r k + l 1 = 3 r k B 1 l 1 + 3 r k r 2 b l B l + 1 l + r 1 +
3 r k 2 r 2 b l + r B l + r + 1 l + 2 r 1 + + 2 b l + r k r B l + 1 + r ( k 1 ) l + r k 1 . By b l + r = b l + s , b l + 2 r = b l + 2 s , ,
b l + r k r = b l + ( k 1 ) s , B 1 l 1 = B l , B l + r + 1 l + 2 r 1 = = B l + 1 + r ( k 1 ) l + r k 1 = B r , we have:
B r k + l = 3 r k B l + 3 r k r 2 b l B r + 3 r k 2 r 2 b l 2 s B r + + 2 b l 2 ( k 1 ) s B r = 3 r k B l + 2 b l B r ( 3 r k r 2 0 + 3 r k 2 r 2 s + + 3 0 2 ( k 1 ) s ) = 3 r k B l + 2 b l B r 3 r k 2 s k 3 r 2 s .
 □
Proposition 7.
Let a 1 a l a l + 1 a l + r ¯ be a periodic E-sequence. By Proposition 2, define two integers x 0 and x r k + l such that 2 s k + b l x r k + l 3 r k + l x 0 = B r k + l , 1 x 0 < 2 s k + b l and 1 x r k + l < 3 r k + l . Then, there is a constant K N , depending on a 1 , , a l + r such that when k > K and,
(i)
if 2 s > 3 r , there is u r k + l Z , 0 u r k + l < ( 2 s 3 r ) 3 l such that
x 0 = 2 s k + b l u r k + l B l ( 2 s 3 r ) + 2 b l B r ( 2 s 3 r ) 3 l , x r k + l = 3 r k u r k + l + B r 2 s 3 r ;
(ii)
if 3 r > 2 s , there is u r k + l N , 1 u r k + l ( 3 r 2 s ) 3 l such that
x 0 = 2 s k + b l u r k + l B l ( 3 r 2 s ) 2 b l B r ( 3 r 2 s ) 3 l , x r k + l = 3 r k u r k + l B r 3 r 2 s .
Proof. 
(i)
2 s > 3 r . By x r k + l = 3 r k + l x 0 + B r k + l 2 s k + b l , we have 2 s k + b l x r k + l B r k + l   ( m o d   3 r k + l ) . Then, 2 s k + b l x r k + l 3 r k B l + 2 b l B r 2 s k 3 r k 2 s 3 r   ( m o d   3 r k + l ) by Proposition 6. Thus, ( 2 s 3 r ) 2 s k + b l x r k + l ( 2 s 3 r ) 3 r k B l + ( 2 s k 3 r k ) 2 b l B r   ( m o d   ( 2 s 3 r ) 3 r k + l ) . Hence, 2 s k + b l ( ( 2 s 3 r ) x r k + l B r ) 3 r k ( ( 2 s 3 r ) B l 2 b l B r )   ( m o d   ( 2 s 3 r ) 3 r k + l ) . Define u r k + l = ( 2 s 3 r ) x r k + l B r 3 r k . Then, u r k + l Z and 2 s k + b l u r k + l ( 2 s 3 r ) B l 2 b l B r   ( m o d   ( 2 s 3 r ) 3 l ) . Hence x r k + l = 3 r k u r k + l + B r 2 s 3 r and:
x 0 = 2 s k + b l x r k + l B r k + l 3 r k + l = 2 s k + b l 3 r k u r k + l + B r 2 s 3 r 3 r k B l 2 b l B r 2 s k 3 r k 2 s 3 r 3 r k + l = 3 r k 2 s k + b l u r k + l + 2 s k + b l B r 3 r k B l ( 2 s 3 r ) + 3 r k 2 b l B r 2 s k + b l B r ( 2 s 3 r ) 3 r k + l = 2 s k + b l u r k + l B l ( 2 s 3 r ) + 2 b l B r ( 2 s 3 r ) 3 l .
By x r k + l = 3 r k u r k + l + B r 2 s 3 r < 3 r k + l , we have u r k + l < 3 r k + l ( 2 s 3 r ) B r 3 r k = 3 l ( 2 s 3 r ) B r 3 r k < 3 l ( 2 s 3 r ) . By x r k + l = 3 r k u r k + l + B r 2 s 3 r > 0 , we have u r k + l > B r 3 r k . Since lim k B r 3 r k = 0 and u r k + l Z , there is a constant K N , depending on a 1 , , a l + r such that u r k + l 0 when k > K .
(ii)
3 r > 2 s . By x r k + l = 3 r k + l x 0 + B r k + l 2 s k + b l , we have:
2 s k + b l ( ( 3 r 2 s ) x r k + l + B r ) 3 r k ( ( 3 r 2 s ) B l + 2 b l B r )   ( m o d ( 3 r 2 s ) 3 r k + l ) .
Define u r k + l = ( 3 r 2 s ) x r k + l + B r 3 r k . Then:
u r k + l Z , 2 s k + b l u r k + l ( 3 r 2 s ) B l + 2 b l B r   ( m o d ( 3 r 2 s ) 3 l ) .
Thus, x r k + l = 3 r k u r k + l B r 3 r 2 s , x 0 = 2 s k + b l u r k + l B l ( 3 r 2 s ) 2 b l B r ( 3 r 2 s ) 3 l . Since x r k + l = 3 r k u r k + l B r 3 r 2 s > 0 , then u r k + l > B r 3 r k , and thus, 1 u r k + l .
By x 0 = 2 s k + b l u r k + l B l ( 3 r 2 s ) 2 b l B r ( 3 r 2 s ) 3 l < 2 s k + b l , we have u r k + l < ( 3 r 2 s ) 3 l + B l ( 3 r 2 s ) + 2 b l B r 2 s k + b l . Since lim k B l ( 3 r 2 s ) + 2 b l B r 2 s k + b l = 0 and u r k + l Z , there is a K N such that u r k + l ( 3 r 2 s ) 3 l when k > K .
 □
Theorem 1.
If 3 r > 2 s , then a 1 a l a l + 1 a l + 2 a l + r 1 a l + r ¯ is Ω-divergent.
Proof. 
By Proposition 7(ii), x 0 = 2 s k + b l u r k + l B l ( 3 r 2 s ) 2 b l B r ( 3 r 2 s ) 3 l and u r k + l 1 . Then, x 0 + as k . Thus, the E-sequence is Ω -divergent. □
Theorem 2.
If a 1 a l a l + 1 a l + 2 a l + r 1 a l + r ¯ is Ω-convergent to x, then ( x n ) n 0 is periodic.
Proof. 
By Theorem 1, 2 s > 3 r . By Proposition 7(i),
x 0 = 2 s k + b l u r k + l B l ( 2 s 3 r ) + 2 b l B r ( 2 s 3 r ) 3 l
and u r k + l 0 for all k > K . Since x 0 = x < for all sufficiently large k, by Proposition 5(iv), then u r k + l = 0 . Thus, x 0 = 2 b l B r B l ( 2 s 3 r ) ( 2 s 3 r ) 3 l and x r k + l = B r 2 s 3 r for all k 0 . Hence, ( x n ) n 0 is periodic, and its non-periodic part and periodic part are ( x 0 x 1 x l ) and x l + 1 x l + r ¯ , respectively. □
Theorem 3.
Assume that all non-periodic E-sequence are Ω-divergent. Then, the trajectory of every odd positive integer is periodic.
Proof. 
Suppose that x is an odd positive integer, ( x n ) n 0 and ( a n ) n 1 are its trajectory and E-sequence, respectively. Then, Ω lim a n = x . Thus, ( a n ) n 1 is periodic by the assumption. Hence, ( x n ) n 0 is periodic by Theorem 2. □

4. Non-Periodic E-Sequences

For any real number α , { α } denotes its fractional part. The following lemma is due to Matthews and Watts (see Lemma 2(b) in [13]). We present its proof for the reader’s convenience.
Lemma 1.
Let ( a n ) n 1 be an E-sequence such that Ω lim   a n = x 0 and ( x n ) n 0 is unbounded. Then, lim ¯ n b n n log 2 3 .
Proof. 
From x k = 3 x k 1 + 1 2 a k , we have 2 a k = 3 x k 1 + 1 x k . Then:
2 b n = k = 1 n 2 a k = k = 1 n 3 x k 1 + 1 x k = x 0 x n k = 1 n 3 x k 1 + 1 x k 1 = 3 n x 0 x n k = 1 n ( 1 + 1 3 x k 1 ) .
Thus:
x n = 3 n x 0 2 b n k = 1 n ( 1 + 1 3 x k 1 )
which we call the Matthews and Watts’ formula (see Lemma 1(b) in [13]).
Since ( x n ) n 1 is unbounded, all x n are distinct. Then:
1 x n 3 n x 0 2 b n k = 1 n ( 1 + 1 3 k ) .
Thus:
0 log 3 n 2 b n + log x 0 + k = 1 n log ( 1 + 1 3 k ) log 3 n log 2 b n + log x 0 + k = 1 n 1 3 k .
Hence:
log 2 b n log 3 n + log x 0 + 1 3 k = 1 n 1 k .
Therefore:
b n n log 2 3 + log 2 x 0 n + 1 n log 8 k = 1 n 1 k .
Then:
lim ¯ n b n n log 2 3 .
 □
Theorem 4.
Let ( a n ) n 1 be a non-periodic E-sequence such that lim ¯ n b n n > log 2 3 . Then, Ω lim a n = .
Proof. 
Suppose that Ω lim a n = x 0 for some positive integer x 0 . It follows from Lemma 1 and lim ¯ n b n n > log 2 3 that ( x n ) n 0 is bounded. Then, ( x n ) n 0 is periodic. Thus, ( a n ) n 1 is periodic, which contradicts the non-periodicity of ( a n ) n 1 . Hence, Ω lim a n = . □
The following lemma is the well known Wendel’s inequality (see [15]). Lemma 3 is a consequence of an easy calculation.
Lemma 2.
Let x be a positive real number, and let s ( 0 , 1 ) . Then, Γ ( x + s ) Γ ( x ) x s .
Lemma 3.
Let a and b be two integers with a 1 and a b . Then, k = 0 n ( 1 + z a k + b ) = Γ ( b a ) Γ ( b + z a + n + 1 ) Γ ( b + z a ) Γ ( b a + n + 1 ) .
Lemma 4.
1 k < 3 n ,   k 1 , 5 ( mod   6 ) ( 1 + 1 3 k ) < 1 . 5 n 1 9 for all n 1 .
Proof. 
Let 2 | n . Then:
k = 0 n 2 1 ( 1 + 1 3 ( 6 k + 1 ) ) = Γ ( 1 6 ) Γ ( n 2 + 2 9 ) Γ ( 2 9 ) Γ ( n 2 + 1 6 ) Γ ( 1 6 ) Γ ( 2 9 ) ( n 2 + 1 6 ) 1 18
and:
k = 0 n 2 1 ( 1 + 1 3 ( 6 k + 5 ) ) = Γ ( 5 6 ) Γ ( n 2 + 8 9 ) Γ ( 8 9 ) Γ ( n 2 + 5 6 ) Γ ( 5 6 ) Γ ( 8 9 ) ( n 2 + 5 6 ) 1 18
by Wendel’s inequality. Thus,
1 k < 3 n ,   k 1 , 5 ( mod   6 ) ( 1 + 1 3 k ) = k = 0 n 2 1 ( 1 + 1 3 ( 6 k + 1 ) ) k = 0 n 2 1 ( 1 + 1 3 ( 6 k + 5 ) ) Γ ( 1 6 ) Γ ( 5 6 ) Γ ( 2 9 ) Γ ( 8 9 ) ( n 2 + 5 6 ) 1 18 ( n 2 + 1 6 ) 1 18 1.4196 ( n 2 3 ) 1 18 < 1.5 n 1 9 .
Let 2 n . Then:
k = 0 n + 1 2 1 ( 1 + 1 3 ( 6 k + 1 ) ) = Γ ( 1 6 ) Γ ( n 2 + 13 18 ) Γ ( 2 9 ) Γ ( n 2 + 2 3 ) Γ ( 1 6 ) Γ ( 2 9 ) ( n 2 + 2 3 ) 1 18
and
k = 0 n + 1 2 2 ( 1 + 1 3 ( 6 k + 5 ) ) = Γ ( 5 6 ) Γ ( n 2 + 7 18 ) Γ ( 8 9 ) Γ ( n 2 + 1 3 ) Γ ( 5 6 ) Γ ( 8 9 ) ( n 2 + 1 3 ) 1 18
by Wendel’s inequality. Thus,
1 k < 3 n ,   k 1 , 5 ( mod   6 ) ( 1 + 1 3 k ) = k = 0 n + 1 2 1 ( 1 + 1 3 ( 6 k + 1 ) ) k = 0 n + 1 2 2 ( 1 + 1 3 ( 6 k + 5 ) ) Γ ( 1 6 ) Γ ( 5 6 ) Γ ( 2 9 ) Γ ( 8 9 ) ( n 2 + 2 3 ) 1 18 ( n 2 + 1 3 ) 1 18 < 1.5 n 1 9 .
 □
Theorem 5.
Let 2 b n x n 3 n x 0 = B n such that 1 x 0 < 2 b n , 1 x n < 3 n , 3 x 0 , and x 0 , , x n 1 are distinct integers. Then, x 0 > B n 3 n ( 1.5 n 1 9 1 ) .
Proof. 
From the Matthews and Watts’ formula and Lemma 4, we have:
2 b n x n 3 n x 0 = k = 1 n ( 1 + 1 3 x k 1 ) 1 k < 3 n ,   k 1 , 5 ( mod   6 ) ( 1 + 1 3 k ) < 1.5 n 1 9 .
Then, 3 n x 0 + B n 3 n x 0 < 1.5 n 1 9 . Thus, x 0 > B n 3 n ( 1.5 n 1 9 1 ) . □
Corollary 1.
Let θ l o g 2 3 be an irrational number. Define a n = [ n θ ] [ ( n 1 ) θ ] . Then, Ω lim   a n = .
Proof. 
Let θ = l o g 2 3 . Then, B n 3 n = k = 1 n 2 [ ( k 1 ) log 2 3 ] 3 k > n 8 by 2 [ ( k 1 ) log 2 3 ] 3 k > 1 8 . Thus, B n 3 n ( 1.5 n 1 9 1 ) > n 8 ( 1.5 n 1 9 1 ) , as n . Hence, Ω lim   a n = by Theorem 5.
Let θ > log 2 3 . Then, lim n   b n n = lim n   [ n θ ] n = θ > log 2 3 . Since θ is an irrational number, ( a n ) n 1 is non-periodic. Thus, Ω lim   a n = by Theorem 4. □
Lemma 5.
Let x and n be two positive integers. Then, (i) k = 0 n 1 ( 1 + 1 3 ( x + k ) ) 1 + n 3 x ; (ii) k = 0 n 1 ( 1 + 1 3 ( x k ) ) 1 + n 3 x for x n ; (iii) k = 0 n 1 ( 1 + 1 3 ( x k ) ) > 3 x 3 x n for x n 2 .
Proof. 
(i) The proof is by induction on n. For the base step, let n = 1 , then k = 0 n 1 ( 1 + 1 3 ( x + k ) ) = 1 + 1 3 x = 1 + n 3 x . For the induction step, assume that k = 0 n 1 ( 1 + 1 3 ( x + k ) ) 1 + n 3 x . Then, k = 0 n ( 1 + 1 3 ( x + k ) ) ( 1 + n 3 x ) ( 1 + 1 3 ( x + n ) ) = 1 + n 3 x + 1 3 ( x + n ) + n 9 x ( x + n ) 1 + n + 1 3 x . Thus, the inequality holds for all n 1 . The proof of (ii) is similar to that of (i) and omitted.
(iii) Let n = 2 . Since 3 x · 3 x 2 · 3 x 3 x + 2 > 3 x · 3 x 3 · 3 x , then 3 x 1 3 x · 3 ( x 1 ) > 1 3 x 2 . Thus, 1 + 1 3 x + 1 3 ( x 1 ) + 1 3 x · 3 ( x 1 ) > 3 x 2 + 2 3 x 2 = 1 + 2 3 x 2 . Hence, ( 1 + 1 3 x ) ( 1 + 1 3 ( x 1 ) ) > 3 x 3 x 2 . Therefore, k = 0 n 1 ( 1 + 1 3 ( x k ) ) > 3 x 3 x n .
Assume that k = 0 n 1 ( 1 + 1 3 ( x k ) ) > 3 x 3 x n . Since ( 3 x 3 n + 1 ) ( 3 x n 1 ) > ( 3 x n ) ( 3 x 3 n ) , then 3 x ( 3 x 3 n ) + 3 x ( 3 x n ) ( 3 x 3 n ) > 3 x 3 x n 1 . Thus, k = 0 n ( 1 + 1 3 ( x k ) ) > 3 x 3 x n ( 1 + 1 3 ( x n ) ) = 3 x 3 x n + 3 x ( 3 x n ) ( 3 x 3 n ) > 3 x 3 x n 1 . □
Lemma 6.
Let 2 b n x n 3 n x 0 = B n such that 1 x 0 < 2 b n , 1 x n < 3 n , x i x j for all 0 i < j n 1 . Then, (i) B n 3 n n 3 if x k > x 0 for all 1 k n 1 ; (ii) B n 2 b n < n 3 if x n < x k for all 0 k n 1 ; (iii) B n 2 b n > n 3 if x n > x i for all 0 i n 1 ; (iv) B n 3 n n 3 if x 0 > x k for all 1 k n .
Proof. 
(i) From 2 b n x n 3 n x 0 = k = 0 n 1 ( 1 + 1 3 x k ) , we have:
1 + B n 3 n x 0 = k = 0 n 1 ( 1 + 1 3 x k ) k = 0 n 1 ( 1 + 1 3 ( x 0 + k ) ) .
Then, 1 + B n 3 n x 0 1 + n 3 x 0 by Lemma 5(i). Thus, B n 3 n n 3 .
(ii) From 2 b n x n 3 n x 0 = k = 0 n 1 ( 1 + 1 3 x k ) , we have:
2 b n x n B n 2 b n x n = k = 0 n 1 ( 1 + 1 3 x k ) 1 k = 0 n 1 ( 1 + 1 3 ( x n + k ) ) 1 .
Then, 1 B n 2 b n x n k = 0 n 1 ( 1 + 1 3 ( x n + k ) ) 1 ( 1 + n 3 x n ) 1 by Lemma 5(i). Thus:
B n 2 b n x n 1 ( 1 + n 3 x n ) 1 = n 3 x n + n .
Hence, B n 2 b n n x n 3 x n + n < n 3 .
(iii) Let n = 1 . Then, x 1 = 3 x + 1 2 a 1 > x . Thus, ( 3 2 a 1 ) x + 1 > 0 . Hence, a 1 = 1 . Therefore, B n 2 b n = B 1 2 b 1 = 1 2 > 1 3 = n 3 .
Let x n n 2 . By Lemma 5(iii), we have 2 b n x n 3 n x 0 = k = 0 n 1 ( 1 + 1 3 x k ) k = 0 n 1 ( 1 + 1 3 ( x n k ) ) > 3 x n 3 x n n . Then, 2 b n x n 2 b n x n B n > 3 x n 3 x n n . Thus, 2 b n x n B n 2 b n x n < 3 x n n 3 x n . Hence, B n 2 b n > n 3 .
(iv) By Lemma 5(ii), we have:
1 + B n 3 n x 0 = 2 b n x n 3 n x 0 = k = 0 n 1 ( 1 + 1 3 x k ) k = 0 n 1 ( 1 + 1 3 ( x 0 k ) ) 1 + n 3 x 0 .
Then, B n 3 n n 3 . □
A direct consequence of Lemma 6 is the following theorem, which may imply something unknown.
Theorem 6.
Let 2 b n x n 3 n x 0 = B n such that 1 x 0 < 2 b n , 1 x n < 3 n , x i x j for all 0 i < j n 1 . Then:
(i) B n 3 n > n 3 implies x k x 0 for some 1 k n 1 ;
(ii) B n 3 n < n 3 implies x 0 x k for some 1 k n ;
(iii) B n 2 b n n 3 implies x n x i for some 0 i n 1 ;
(iv) B n 2 b n n 3 implies x n x k for some 0 k n 1 .
Theorem 7.
Let ( a n ) n 1 be an E-sequence such that (i) 3 n > 2 b n for all n N ; (ii) there is a constant c > log 2 3 such that there are infinitely many distinct pairs ( k , l ) of positive integers such that l > k c , a k + 1 = = a l = 1 . Then, Ω lim a n = .
Proof. 
It follows from (i) that B n < 3 n n for all n N by induction on n. B k + 1 l 1 = 3 l k 2 l k by Proposition 6.
Let x l 1 , l = 3 l x 0 1 , l + B 1 l 1 2 b l , 1 x 0 1 , l < 2 b l , 1 x l 1 , l < 3 l . Then, x k 1 , l = 3 k x 0 1 , l + B 1 k 1 2 b k , x l 1 , l = 3 l k x k 1 , l + B k + 1 l 1 2 b k + 1 l by Proposition 5(ii). By B k + 1 l 1 = 3 l k 2 l k , 2 b k + 1 l = 2 l k , we have 2 l k ( x l 1 , l + 1 ) = 3 l k ( x k 1 , l + 1 ) . Thus, x k 1 , l = 2 l k w 1 for some 1 w . Hence, x k 1 , l = 3 k x 0 1 , l + B 1 k 1 2 b k = 2 l k w 1 . Therefore, x 0 1 , l = 2 l k 2 b k w 2 b k B 1 k 1 3 k 2 l 3 k 2 b k k 1 k ( 2 c 3 ) k 2 b k k 1 k . If there are only finitely many distinct k in all pairs ( k , l ) , x 0 1 , l 2 l 3 k 2 b k k 1 k , as l ; otherwise, x 0 1 , l ( 2 c 3 ) k 2 b k k 1 k , as k . Then, Ω lim a n = . □
Corollary 2.
Let ( a n ) n 1 be the E-sequence 12121112 , where a n = 2 if n { 2 1 , 2 2 , 2 3 , } and a n = 1 otherwise. Then, Ω lim a n = .
Proof. 
Take c = 7 4 > log 2 3 , k = 2 m , and l = 2 m + 1 1 . Then, a k + 1 = = a l = 1 , l > k c for all m 3 . Thus, Ω lim a n = by Theorem 7. □
Theorem 8.
Let ( a n ) n 1 be an E-sequence such that (i) 3 n > 2 b n for all n N ; (ii) there is a constant c > log 2 3 such that there are infinitely many distinct pairs ( r , l ) of positive integers such that l > r , b l + r > l c , a l + k = a k for all 1 k r , i.e., ( a 1 a r ) a r + 1 a l ( a l + 1 a l + r ) is contained in ( a n ) n 1 . Then, Ω lim a n = .
Proof. 
Let x l + r 1 , l + r = 3 l + r x 0 1 , l + r + B 1 l + r 1 2 b 1 l + r , 1 x 0 1 , l + r < 2 b 1 l + r , 1 x l + r 1 , l + r < 3 l + r . Then, x l 1 , l + r = 3 l x 0 1 , l + r + B 1 l 1 2 b 1 l , x l + r 1 , l + r = 3 r x l 1 , l + r + B l + 1 l + r 1 2 b l + 1 l + r = 3 r x l 1 , l + r + B 1 r 1 2 b 1 r by Proposition 5(ii). By 3 l > 2 b 1 l , we have x l 1 , l + r > x 0 1 , l + r .
Let x r 1 , r = 3 r x 0 1 , r + B 1 r 1 2 b 1 r , 1 x 0 1 , r < 2 b 1 r , 1 x r 1 , r < 3 r . Then, x 0 1 , r x l 1 , l + r ( mod   2 b 1 r ) . By Proposition 5(iii), we have x 0 1 , l + r x 0 1 , r . Let x l 1 , l + r = 2 b 1 r u + x 0 1 , r . Then, u 1 by x l 1 , l + r > x 0 1 , l + r x 0 1 , r . Thus:
x 0 1 , l + r = 2 b 1 l 2 b 1 r u + 2 b 1 l x 0 1 , r B 1 l 1 3 l 2 b 1 l + r 3 l l ( 2 c 3 ) l l ,   a s   l .
Hence, Ω lim a n = . □
Theorem 9.
Let 1 θ < log 2 3 , and define a n = [ n θ ] [ ( n 1 ) θ ] . Then, Ω lim a n = .
Proof. 
If θ is a rational number, then ( a n ) n 1 is purely periodic, and the result follows from Theorem 1. Let θ be an irrational number in the following. By the Hurwitz theorem, there are infinite convergents s r of θ such that | θ s r | < 1 5 r 2 . There are two cases to be considered.
Case 1. There are infinite convergents s r of θ such that 0 < θ s r < 1 5 r 2 . We prove that [ θ n ] = [ s r n ] for all 1 n [ 5 r ] . By 1 n 5 r , we have 0 < θ n s r n < n 5 r 2 < 5 r 5 r 2 = 1 r . Then, 0 { s r n } < θ n [ s r n ] < 1 r + { s r n } 1 . Thus, 0 < θ n [ s r n ] < 1 . Hence, [ θ n ] = [ s r n ] . Then, we have the following periodic table for ( a n ) 1 n 5 r .
a 1 a 2 a 5 r 2 r a r
a r + 1 a 2 + r a 5 r r a 2 r
a 2 r + 1 a 2 + 2 r a 5 r
By Proposition 7(ii), x 0 1 , 2 r = 2 2 [ r θ ] u 2 r B r 3 r 2 [ r θ ] for some u 2 r 1 .
By B r = i = 0 r 1 3 r 1 i 2 b i = 3 r 1 i = 0 r 1 2 b i 3 i 3 r 1 i = 0 r 1 2 [ i θ ] 3 i 3 r 1 i = 0 r 1 2 i θ 3 i = 3 r 3 1 ( 2 θ 3 ) r 1 2 θ 3 = 3 r 2 r θ 3 2 θ 3 r 3 2 θ , we have:
x 0 1 , 2 r 2 2 [ r θ ] B r 3 r 2 [ r θ ] 4 r θ 1 3 r 3 2 θ 3 r 2 r θ 1 = 1 4 ( 4 θ 3 ) r 1 3 2 θ 1 1 2 ( 2 θ 3 ) r .
Thus, x 0 1 , 2 r , as r . Hence, Ω lim a n = .
Case 2. There are infinite convergents s r of θ such that 0 < s r θ < 1 5 r 2 .
Firstly, we prove [ θ n ] = [ s r n ] for all 1 n [ 5 r ] , n { r , 2 r } . By 0 < s r θ < 1 5 r 2 , we have s r 1 5 r 2 < θ < s r . Then, s r n [ s r n ] n 5 r 2 < θ n [ s r n ] < s r n [ s r n ] < 1 . By 1 n [ 5 r ] , n { r , 2 r } , we have 0 < 1 r n 5 r 2 s r n [ s r n ] n 5 r 2 . Then, 0 < θ n [ s r n ] < 1 . Thus, [ θ n ] = [ s r n ] .
Secondly, we prove [ r θ ] = s 1 , [ 2 r θ ] = 2 s 1 . By 1 n , 0 < s r θ < 1 5 r 2 , we have n 5 r 2 + s r n < n θ < s r n . By n < 5 r , we have 1 < 1 r < n 5 r 2 . Then, 1 + s r n < n 5 r 2 + s r n < n θ < s r n . By taking n = r , 2 r , we have [ r θ ] = s 1 , [ 2 r θ ] = 2 s 1 .
Let 2 j r 1 , then r + 2 r + j 2 r 1 and r + 1 r + j 1 2 r 2 . Thus, a r + j = [ θ ( r + j ) ] [ θ ( r + j 1 ) ] = [ s r ( r + j ) ] [ s r ( r + j 1 ) ] = [ s + s r j ] [ s + s r ( j 1 ) ] = [ s r j ] [ s r ( j 1 ) ] = a j .
Let 2 j [ 5 r ] 2 r . Then, 2 r + 2 2 r + j [ 5 r ] and 2 r + 1 2 r + j 1 [ 5 r ] 1 . Thus, a 2 r + j = [ θ ( 2 r + j ) ] [ θ ( 2 r + j 1 ) ] = [ s r ( 2 r + j ) ] [ s r ( 2 r + j 1 ) ] = [ s r j ] [ s r ( j 1 ) ] = a j .
By easy calculation, we have a r = a 2 r = 1 , a r + 1 = a 2 r + 1 = 2 .
Then, we have the following periodic table for ( a n ) 1 n 5 r .
a 1 a 2 a 3 a 5 r 2 r a r a r + 1
a 2 + r a 3 + r a 5 r r a 2 r a 2 r + 1
a 2 + 2 r a 3 + 2 r a 5 r
Since θ < log 2 3 , we then take all convergents s r of θ such that s r < log 2 3 , and thus, 2 s < 3 r . By a 1 = 1 , b 2 r + 1 = [ r θ ] + 1 = s and Proposition 7(ii), we have:
x 0 1 , 2 r + 1 = 2 2 s + 1 u 2 r + 1 ( 3 r 2 s ) 2 B 2 r 3 ( 3 r 2 s )
for some u 2 r + 1 1 . By B 2 r = i = 0 r 1 3 r 1 i 2 b 2 i + 1 = 3 r 1 i = 0 r 1 2 [ i θ + θ ] 1 3 i 3 r 1 2 θ 1 i = 0 r 1 2 i θ 3 i = 2 θ 1 3 r 3 1 ( 2 θ 3 ) r 1 2 θ 3 = 2 θ 1 3 r 2 r θ 3 2 θ C 3 r , where C = 2 θ 1 3 2 θ , we have:
x 0 1 , 2 r + 1 2 3 4 [ r θ ] + 1 C 3 r 3 r 2 s 1 3 2 3 4 r θ C 3 r 3 r 2 s 1 3 2 3 4 r θ C 3 r 3 r 1 3 = 2 3 ( 4 θ 3 ) r 2 3 C 1 3 .
Thus, lim r x 0 1 , 2 r + 1 = . Hence, Ω lim a n = . □

5. Concluding Remarks and Open Problems

The results on non-periodic E-sequences in Section 4 were based on the theory of periodic E-sequences in Section 3 and the Matthews and Watts’ formula. Currently, we have no other way to tackle non-periodic E-sequences. We can obtain various generalizations and analogues of Theorems 4–8. However, we need good problems to make some progress.
One seemingly simple problem that we are not able to prove is whether ( a n ) n 1 is divergent, where a n = 2 if n { 2 2 , 3 2 , 4 2 , } and a n = 1 otherwise, i.e., ( a n ) n 1 is 111211112 .
Another interesting problem is whether ( a n ) n 1 with infinitely many n satisfying b n > n log 2 3 is Ω -divergent. By virtue of Theorem 4, we only need to consider the case of lim ¯ n b n n = log 2 3 . Theorem 5 answers the problem if B n 3 n ( 1 . 5 n 1 9 1 ) , as n . Currently, we do not know how to tackle the other cases of the problem.
Conjecture 2(ii) is also important in some sense.

Funding

This work is supported by the National Foundation of Natural Sciences of China (Grant Nos. 61379018, 61662044, 11571013, and 11671358).

Acknowledgments

I am greatly indebted to Junde Miao for his constant encouragement while I was working on the problem.

Conflicts of Interest

The author declares no conflict of interest.

References

  1. Lagarias, J.C. The 3x + 1 problem and its generalizations. Am. Math. Mon. 1985, 92, 3–23. [Google Scholar] [CrossRef]
  2. Kadic, S.; Tomovic, S. Computer-Based Validation of 3n + 1 Hypothesis for Numbers 3n − 1. Teh. Vjesn. 2019, 26, 289–293. [Google Scholar] [CrossRef]
  3. Thomas, A. A non-uniform distribution property of most orbits, in case the 3x + 1 conjecture is true. Acta Arith. 2017, 178, 125–134. [Google Scholar] [CrossRef]
  4. Colussi, L. The convergence classes of Collatz function. Theor. Comput. Sci. 2011, 412, 5409–5419. [Google Scholar] [CrossRef]
  5. Tao, T. Almost all orbits of the Collatz map attain almost bounded values. arXiv 2019, arXiv:1909.03562. [Google Scholar]
  6. Kari, J. Universal pattern generation by cellular automata. Theor. Comput. Sci. 2012, 429, 180–184. [Google Scholar] [CrossRef]
  7. Javaheri, M. A Collatz-type conjecture on the set of rational numbers. J. Number Theory 2012, 132, 2582–2594. [Google Scholar] [CrossRef]
  8. Crandall, R.E. On the 3x + 1 Problem. Math. Com. 1978, 32, 1281–1292. [Google Scholar] [CrossRef]
  9. Sander, J.W. On the (3N + 1)-conjecture. Acta Arith. 1990, 55, 241–248. [Google Scholar] [CrossRef]
  10. Everett, C.J. Iteration of the number-theoretic function f(2n) = n, f(2n + 1) = 3n + 2. Adv. Math. 1977, 25, 42–45. [Google Scholar] [CrossRef]
  11. Terras, R. A stopping time problem on the positive integers. Acta Arith. 1976, 30, 241–252. [Google Scholar] [CrossRef]
  12. Möller, H. Über Hasses Verallgemeinerung des Syracuse -Algorithmus (Kakutanis Problem). Acta Arith. 1978, 34, 219–226. [Google Scholar] [CrossRef]
  13. Matthews, K.R.; Watts, A.M. A generalization of Hasse’s generalization of the Syracuse algorithm. Acta Arith. 1984, 43, 167–175. [Google Scholar] [CrossRef]
  14. Simons, J.L. On the (non-)existence of m-cycles for generalized Syracuse sequences. Acta Arith. 2008, 131, 217–254. [Google Scholar] [CrossRef]
  15. Wendel, J.G. Note on the Gamma Function. Am. Math. Mon. 1948, 55, 563–564. [Google Scholar] [CrossRef]

Share and Cite

MDPI and ACS Style

Wang, S. An E-Sequence Approach to the 3x + 1 Problem. Symmetry 2019, 11, 1415. https://doi.org/10.3390/sym11111415

AMA Style

Wang S. An E-Sequence Approach to the 3x + 1 Problem. Symmetry. 2019; 11(11):1415. https://doi.org/10.3390/sym11111415

Chicago/Turabian Style

Wang, Sanmin. 2019. "An E-Sequence Approach to the 3x + 1 Problem" Symmetry 11, no. 11: 1415. https://doi.org/10.3390/sym11111415

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop