Part9 Vol18 No9 2008

Tutorials and Reviews
International Journal of Bifurcation and Chaos, Vol. 18, No. 9 (2008) 2487–2642
c World Scientific Publishing Company
A NONLINEAR DYNAMICS PERSPECTIVE

OF WOLFRAM’S NEW KIND OF SCIENCE.
PART IX: QUASI-ERGODICITY
LEON O. CHUA, GIOVANNI EGIDIO PAZIENZA∗ , LASZLO ORZO†,

VALERY I. SBITNEV‡ and JINWOOK SHIN
Department of Electrical Engineering and Computer Sciences,
University of California at Berkeley,
Berkeley, CA 94720, USA
∗Grup de Recerca en Sistemes Intel·ligents,
Enginyeria i Arquitectura La Salle, Universitat Ramon Llull,

Barcelona, Spain
†Cellular Sensory Wave Computing Laboratory,
Computer and Automation Research Institute,

SZTAKI, Hungarian Academy of Sciences,
H-111 Budapest, Kende-u. 13, Hungary
‡B.P. Konstantinov Saint-Petersburg Nuclear Physics Institute,
RAN, Russia
Received December 12, 2007; Revised May 15, 2008
Our scientific odyssey through the theory of 1-D cellular automata is enriched by the definition
of quasi-ergodicity, a new empirical property discovered by analyzing the time-1 return maps of
local rules. Quasi-ergodicity plays a key role in the classification of rules into six groups: in fact,
it is an exclusive characteristic of complex and hyper Bernoulli-shift rules. Besides introducing
quasi-ergodicity, this paper answers several questions posed in the previous chapters of our quest.
To start with, we offer a rigorous explanation of the fractal behavior of the time-1 characteristic
functions, finding the equations that describe this phenomenon. Then, we propose a classification
of rules according to the presence of Isles of Eden, and prove that only 28 local rules out of 256
do not have any of them; this result sheds light on the importance of Isles of Eden. A section of
this paper is devoted to the characterization of Bernoulli basin-tree diagrams through modular
arithmetic; the formulas obtained allow us to shorten drastically the number of cases to take
into consideration during numerical simulations. Last but not least, we present some theorems
about additive rules, including an analytical explanation of their scale-free property.
Keywords: Cellular automata; quasi-ergodicity; ergodicity; nonlinear dynamics; attractors; Isles

of Eden; Bernoulli shift; shift maps; basin tree diagram; Bernoulli velocity; Bernoulli return time;
complex Bernoulli shifts; hyper Bernoulli shifts; Binomial series; scale-free phenomena; Rule 45;
Rule 60; Rule 90; Rule 105; Rule 150; Rule 154; additive rules; permutive rules; dissipative rules;
conservative rules; fractals; basin-tree generation formula.
Contents
1. Remembrance of Things Past . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2488
1.1. Boolean cube representation . . . . . . . . . . . . . . . . . . . . . . . . . . 2489
1.2. Index of complexity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2489
2487
2488 L. O. Chua et al.
1.3.One formula specifies all 256 rules . . . . . . . . . . . . . . . . . . . . . . . . 2497

1.4.Space-time pattern and time-τ return maps . . . . . . . . . . . . . . . . . . . 2498
1.5.We only need to study 88 rules! . . . . . . . . . . . . . . . . . . . . . . . . . 2515
1.6.The “Magic” rule spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2515
1.7.Symmetries among Boolean cubes . . . . . . . . . . . . . . . . . . . . . . . . 2518
1.7.1. Local complementation T c . . . . . . . . . . . . . . . . . . . . . . . . 2518
1.7.2. Three equivalence transformations T † , T , and T ∗ . . . . . . . . . . . . . 2518
1.7.3. Perfect complementary rules . . . . . . . . . . . . . . . . . . . . . . . 2522
1.7.4. Permutive rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2522
1.7.5. Superposition of local rules . . . . . . . . . . . . . . . . . . . . . . . . 2523
1.7.6. Rules with explicit period-1 and/or period-2 orbits . . . . . . . . . . . . 2523
1.7.7. Most rules harbor at least one Isle of Eden . . . . . . . . . . . . . . . . 2524
2. Quasi-Ergodicity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2545
2.1. Only complex and hyper Bernoulli- shift rules are quasi-ergodic . . . . . . . . . 2545
2.2. Quasi-ergodicity and Gardens of Eden . . . . . . . . . . . . . . . . . . . . . . 2562
3. Fractals in 1D Cellular Automata . . . . . . . . . . . . . . . . . . . . . . . . . . 2565
3.1. All time-1 characteristic functions are fractals . . . . . . . . . . . . . . . . . . 2565
3.2. Fractals in CA additive rules . . . . . . . . . . . . . . . . . . . . . . . . . . 2569
3.2.1. Rule 60 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2570
3.2.2. Rule 90 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2572
3.2.3. Rule 105 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2574
3.2.4. Rule 150 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2576
3.3. From the time-1 characteristic function to the rule number . . . . . . . . . . . . 2579
3.4. Number of fractal patterns . . . . . . . . . . . . . . . . . . . . . . . . . . . 2581
4. New Results about Isles of Eden . . . . . . . . . . . . . . . . . . . . . . . . . . . 2583
4.1. Definitions and basic lemmas . . . . . . . . . . . . . . . . . . . . . . . . . . 2584
4.2. Alternate proof that rules 45 and 154 are conservative for odd lengths . . . . . 2584
4.3. There are exactly 28 strictly-dissipative local rules . . . . . . . . . . . . . . . . 2585
4.4. Isles of Eden for rules of group 1 . . . . . . . . . . . . . . . . . . . . . . . . 2591
5. How to Find Analytically the Basin-Tree Diagrams for Bernoulli Attractors . . . . . . 2592
5.1. Bernoulli-στ basin-tree generation formula . . . . . . . . . . . . . . . . . . . . 2592
5.2. A practical application of the Bernoulli στ -shift basin tree generation formula . . . 2595
6. Old Theorems and New Results for Additive Cellular Automata . . . . . . . . . . . . 2597
6.1. Theorems on the maximum period of attractors and Isles of Eden . . . . . . . . 2599
6.2. Scale-free property for additive rules . . . . . . . . . . . . . . . . . . . . . . . 2602
7. Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2638
Appendix A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2639
Appendix B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2640
Appendix C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2640
1. Remembrance of Things Past [Chua et al., 2005a], Part V [Chua et al., 2005b],
Part VI [Chua et al., 2006], Part VII [Chua et al.,
This exposition continues our saga on a nonlinear
2007a], and Part VIII [Chua et al., 2007b].1 In this
dynamics perspective of the 256 elementary cellular
automata rules, as featured in eight tutorial-review paper, we examine the 18 yet untamed rules listed
papers: Part I [Chua et al., 2002], Part II [Chua in Tables 11 and 12 of [Chua et al., 2007a];
namely,
et al., 2003], Part III [Chua et al., 2004], Part IV the ten complex Bernoulli-shift rules 18 , 22 , 54 ,
1
Parts I to VI have been collected into two recent edited volumes [Chua, 2006] and [Chua, 2007], respectively. Part VII and
VIII will appear in a future edited volume III.
A Nonlinear Dynamics Perspective of Wolfram’s New Kind of Science. Part IX 2489

73 , 90 , 105 , 122 , 126 , 146 , 150 and the the “cube” shown in Fig. 1(b), henceforth called a
Boolean cube. The rationale for identifying which
eight hyper Bernoulli-shift rules 26 , 30 , 41 , vertex corresponds to which pattern was presented
45 , 60 , 106 , 110 , 154 . in [Chua et al., 2002], in order to provide the genesis
Remarkably, we have observed empirically that of the truth tables from a nonlinear physical sys-
all complex and hyper Bernoulli-shift rules exhibit tem perspective, namely, cellular neural networks
an Ergodic-like dynamics, which we christened (CNN) [Chua, 1996], thereby providing a bridge
quasi-ergodicity. Our main goal of this paper is to between nonlinear dynamics and cellular automata.
describe and characterize this unifying empirical For readers who have not been exposed to the age-
phenomenon. 1 episodes alluded to above, it is not necessary to
We will revisit our fabled Isles of Eden from read the cited literature. Simply map the output of
Parts VII and VIII and offer an alternate perspec- each prescribed Boolean function (i.e. the 8-bit, yet
tive of such rare gems. We will show that all local unspecified, binary string in Fig. 1(d)) onto the cor-
rules harbored a few precious Isles of Eden, except responding colors (red for 1, blue for 0) at the ver-
for 28 rules, which we will prove analytically to be tices of the Boolean cube. Since there are 28 = 256
devoid of Isles of Eden; these are the God forsaken distinct combinations of eight bits, there are exactly
rules! 256 Boolean cubes with distinct vertex color combi-
We will also revisit the scale-free phenomenon nations, one for each Boolean function, as displayed
reported in Parts VII and VIII for additive rules, in Table 1.
and prove these empirical observations are in fact
fundamental properties possessed by such rules. In 1.2. Index of complexity
particular, we will present and prove several analyt-
ical theorems for rules 60 , 90 , 105 and 150 . A careful examination of these 256 Boolean cubes
For the reader’s convenience, let us review shows that it is possible to separate, and segregate,
briefly some highlights from our earlier adven- all red vertices of each Boolean cube from the blue
tures, henceforth referred to collectively as Age-1 vertices by κ = 1, 2, or 3 parallel planes. An exam-
Episodes, a la Tolkien’s “The Lord of the ple illustrating this separation is shown in Fig. 2 for
Rings”. rules 170 , 110 and 184 , respectively. The integer
We are concerned exclusively with our tiny κ is called the index of complexity of rule N . We
universe of 256 one-dimensional binary cellular will always use the color red for κ = 1, blue for
automata, with a periodic boundary condition, as κ = 2, and green for κ = 3 to code the rule number
depicted in Fig. 1(a). Each “ring” has L I + 1 N of each of the 256 Boolean cubes, as printed at
cells, labeled consecutively from i = 0 to i = I. Each the bottom of each Boolean cube in Table 1.
cell “i” has two states xi ∈ {0, 1}, where we usu- It is natural to associate the 8-bit pattern of
ally code the states “0” and “1” by the color “blue” each Boolean function with a decimal number N
and “red ”, respectively. A clock sets the pace in dis- representing the corresponding 8-bit word; namely,
crete times, dubbed “iterations” by the mathemat- N = β7 • 27 + β6 • 26 + β5 • 25 + β4 • 24 + β3 • 23
ical community, or “generations” by the life science
+ β2 • 22 + β1 • 21 + β0 • 20 , β ∈ {0, 1}.
community. The state xt+1 i of all “i” at time t + 1
(i.e. the next generation) is determined by the state Observe that since βi = 0 for each blue vertex
of its nearest neighbors xti−1 , and xti+1 , and itself in Fig. 1(b), N is simply obtained by adding the
xti , at time t [Fig. 1(c)], in accordance with a pre- weights (indicated next to each pattern in Fig. 1(b))
scribed Boolean truth table of eight distinct 3-input associated with all red vertices. For example, for the
patterns [Fig. 1(d)].2 Boolean cube shown in Fig. 3(b), we have
N = 0 • 27 + 1 • 26 + 1 • 25 + 0 • 24 + 1 • 23
1.1. Boolean cube representation + 1 • 22 + 1 • 21 + 1 • 20
We have found it extremely useful to map these = 26 + 25 + 23 + 22 + 21
eight 3-input patterns into the eight vertices of = 110
2
Throughout the paper we intentionally use both t and n to indicate the time (or iterations) to stress the equivalence between
a discrete Cellular Automaton and a continuous nonlinear system.
2490
Fig. 1. Notations, symbols, and universal formula for local rule N .
Table 1. Boolean cubes defining 256 CA rules. Each vertex kmcodes for a 3-bit input pattern.
0 0 0 0 1 0 0 1 2 0 1 0 3 0 1 1
4 1 0 0 5 1 0 1 6 1 1 0 7 1 1 1
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
0 1 2 3 4 5 6 7
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
8 9 10 11 12 13 14 15
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
16 17 18 19 20 21 22 23
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
24 25 26 27 28 29 30 31
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
32 33 34 35 36 37 38 39
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
40 41 42 43 44 45 46 47
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
48 49 50 51 52 53 54 55
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
56 57 58 59 60 61 62 63
Table 1. (Continued )
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
64 65 66 67 68 69 70 71
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
72 73 74 75 76 77 78 79
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6` 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
80 81 82 83 84 85 86 87
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
88 89 90 91 92 93 94 95
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
96 97 98 99 100 101 102 103

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
104 105 106 107 108 109 110 111

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
112 113 114 115 116 117 118 119

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
120 121 122 123 124 125 126 127

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
128 129 130 131 132 133 134 135

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
136 137 138 139 140 141 142 143

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
144 145 146 147 148 149 150 151

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
152 153 154 155 156 157 158 159

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
160 161 162 163 164 165 166 167

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
168 169 170 171 172 173 174 175

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
176 177 178 179 180 181 182 183

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
184 185 186 187 188 189 190 191

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
192 193 194 195 196 197 198 199

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
200 201 202 203 204 205 206 207

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
208 209 210 211 212 213 214 215

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
216 217 218 219 220 221 222 223

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
224 225 226 227 228 229 230 231

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
232 233 234 235 236 237 238 239

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
240 241 242 243 244 245 246 247

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
248 249 250 251 252 253 254 255

Fig. 2. The number κ of parallel planes which separate all vertices having one color from these having a different color on
the other side is called the index of complexity of rule N .
2496
` ´
Fig. 3. An example Rule 110 illustrating the notations, symbols and universal formula from Fig. 1.
Consequently, the Boolean function defined by For example, for rule 110 , we read off the fol-
the Boolean cube in Fig. 3(b) is identified as lowing values from page 1363 of [Chua et al., 2006]:
N = 110 .
1
c1 = −3, c2 = 2, c3 = 1, c4 = − ,
2 (2)
1.3. One formula specifies all
c5 = 1, c6 = −1, c7 = 0, c8 = 0
256 rules
While the usual procedure for specifying a Boolean Substituting the eight real numbers from Eq. (2)
function is to give the truth table, as God-given into Eq. (1), we obtain the following difference equa-
laws, we have discovered the following nonlin- tion for executing rule 110 :

ear difference equation, with eight parameters t 1
t+1 t
xi = −1 + xi−1 + 2xi − 3xi+1 − t
(3)
{c1 , c2 , . . . , c8 }, which is capable of generating 2
any of the 256 Boolean cubes in Table 1, by
To verify that Eq. (3) can indeed generate the truth
merely assigning eight real numbers to these eight
table for rule 110 , let us substitute the eight input
parameters:
patterns listed on the left side of Fig. 1(d):
xt+1
i = {c8 + c7 |[c6 + c5 |(c4 + c3 xti−1 Input code 0m: (xti−1 , xti , xti+1 ) = (0, 0, 0)
+ c2 xti + c1 xti+1 )|]|} (1)

xt+1 = −1 + 0+2•0−3•0− 1
The difference equation (1) is extremely robust in i 2
the sense that a very large set of real numbers can be
chosen to generate each Boolean cube, as depicted = [−1 + 0.5]
in the parameter space R8 in Fig. 4. One such set =0 (4a)
of numbers is given for each of the 256 rules in
Table A-3 of [Chua et al., 2006]. Input code 1m: (xti−1 , xti , xti+1 ) = (0, 0, 1)

1
t+1
xi = −1 + 0 + 2 • 0 − 3 • 1 −
2
= [−1 + 3.5]
=1 (4b)
Input code 2m: (xti−1 , xti , xti+1 )
= (0, 1, 0)

1
xt+1 =
−1 + 0 + 2 • 1 − 3 • 0 −
i
2
= [−1 + 1.5]
=1 (4c)
Input code 3m: (xti−1 , xti , xti+1 ) = (0, 1, 1)

1
t+1
xi =
−1 + 0 + 2 • 1 − 3 • 1 −
2
= [−1 + 1.5]
=1 (4d)
Input code 4m: (xti−1 , xti , xti+1 ) = (1, 0, 0)

Fig. 4. An abstraction showing a curve meandering through 1
t+1
xi =
−1 + 1 + 2 • 0 − 3 • 0 −
8-dimensional parameter space, showing all 256 local rules, 2
not necessary in consecutive order. The robustness of the uni-
versal
` formula is depicted by an open set of parameter´points
= [−1 + 0.5]
surrounding a typical parameter vector for rule N2 all of
which would generate the same truth table as N2 . =0 (4e)
Input code 5m: (xti−1 , xti , xti+1 ) = (1, 0, 1) We end this subsection by emphasizing that
the significance of the universal difference equa-
1 tion enshrined in Fig. 1(e) should not be construed
t+1
xi =
−1 + 1 + 2 • 0 − 3 • 1 −
2 merely as an elegant mathematical formula, but
rather as a mathematical bridge essential for deriv-
= [−1 + 4.5] ing and proving analytical results and theorems,
=1 (4f) as demonstrated in the rigorous derivation of the
Bernoulli shift formulas for rules 170 , 240 , 15 ,
Input code 6m: (xti−1 , xti , xti+1 ) = (1, 1, 0) 85 , and 184 for finite L in [Chua et al., 2005a].
Such a feat would not have been possible without
1 exploiting this universal formula in an essential way.
t+1
xi =
−1 + 1 + 2 • 1 − 3 • 0 −
2
= [−1 + 2.5] 1.4. Space-time pattern and time-τ

return maps
=1 (4g)
Given any initial binary bit-string configuration at
Input code 7m: (xti−1 , xti , xti+1 ) = (1, 1, 1) time t = 0, the local rule N is used to update the
state xt+1
i of each cell “i” at time t + 1, using the
1 states xti−1 , xti , and xti+1 of the three neighboring
xt+1 =
−1 + 1 + 2 • 1 − 3 • 1 −
i
2 cells i − 1, i, and i + 1, centered at location “i”,
respectively. The space-time pattern for the initial
= [−1 + 0.5] state shown in Fig. 3(a) is shown in Fig. 3(c) for
=0 (4h) t = 0, 1, 2, . . . , 11. For simplicity, space-time pat-
terns are generally plotted by displaying a line of
Mapping each vertex kmred if xt+1 i = 1, and L cells, with the implicit understanding that the
blue if xt+1
i = 0, onto the “blank ” vertices in the leftmost bit is “glued ” to the rightmost bit. Such
Boolean cube in Fig. 1(b), we obtain the Boolean space-time patterns are useful if they are T-periodic
cube for 110 in Fig. 3(b), which indeed defines rule with a small period T and a relatively short tran-
110 , as expected. sient regime. For large period T , such as the space-
For ease of future reference, we have enshrined time patterns generated from the 18 complex and
the “shot-gun” Eq. (1) in Fig. 1(e). Observe that hyper Bernoulli rules to be studied in this paper,
in the most general case, this universal formula it is much more revealing to recast space-time pat-
has eight non-zero parameters and two “nested ” terns into a time-τ return map [Chua et al., 2005a]
absolute value functions. The analytical derivation
ρτ [N ]: φn−τ → φn (5)
of this formula in [Chua et al., 2002] shows that
the number of absolute value functions required for where
each rule N is precisely equal to κ − 1, where κ I
is the index of complexity of N , which in turn is
φ 2−(i+1) xi (6)
defined to be the number of parallel planes needed i=0
to separate the “red” vertices from the “blue” ver-
tices. In the case of rule 110 , we recall from Fig. 3 is the decimal equivalent of the binary bit string
that κ 110 = 2 because we only need two parallel
xn = (xn0 xn1 xn2 ··· xnL−1 ) (7)
planes. Consequently, we expect the two parameters
c7 = c8 = 0 so that only one absolute value function and τ is an integer. In most cases, such as the
is needed for 110 .3 For the readers convenience, quasi-ergodic space-time patterns to be presented
Table 2 gives the explicit formula for each of the 256 in Sec. 2, we choose τ = 1 and will be concerned
local rules defined by the Boolean cubes displayed usually with the time-1 map ρ[N ] ρ1 [N ], where
in Table 1. we drop the subscript “1” to avoid clutter. Observe
3
We caution the reader that to avoid ambiguity, Eq. (1) should actually be written as three separate “telescoping” equations,
as in Eq. (A.3) of [Chua et al., 2006]. Hence, c6 = c7 = 0 in Eq. (1) should be interpreted to mean the deletion of the outer
absolute value function associated with c6 and c7 .
Table 2. Formula defining the truth table of all 256 local rules.
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
that 0 ≤ φ < 1 for finite I, and when I → ∞, we though there may exist several Bernoulli attrac-
have a time-1 map over the unit interval tors with different σ and τ , each with its basin of
attraction.
ρ: [0, 1) → [0, 1) (8)
It is important to remember that each time-1 map Group 5 and Group 6 Rules
is uniquely associated with one space-time pattern,
The space-time patterns typically have very long
or “orbit ”, from one initial bit-string configuration.
transients and converge to a period-T attractor with
a very large period T . Moreover, the asymptotic
1.5. We only need to study 88 rules! behavior depends not only on the initial configura-
Although there are 256 local rules, only 88 rules tion, but also on the length L of the bit string. One
are globally independent [Chua et al., 2004] from difference between a group 5 rule and a group 6 rule
each other. All other rules are equivalent to one is that the former is bilateral (and hence has only
of the 88 rules listed in Table 4 of [Chua et al., one globally-equivalent rule), whereas the latter is
2007a]. These 88 rules are listed4 in Table 3 along non-bilateral (and hence has three other globally-
with an integer code M ∈ {1, 2, 3, 4, 5, 6}, where M equivalent rules).
denotes one of the following six distinct qualitative The classification of each of the 256 local rules
dynamics exhibited by a particular local rule N is given in Tables 7–9, 11, and 12 in [Chua et al.,
[Chua et al., 2007a], corresponding to random ini- 2007a].
tial configurations: Given any rule not among those listed in the
88 globally-equivalence classes in Table 3, one can
Group 1 Rules easily look up Table 4 from [Chua et al., 2007a],
Almost all space-time patterns converge to a or Table 3 from [Chua et al., 2007b], to identify
period-1 orbit. The time-1 map corresponding to its equivalent rule, and then look up its complexity
each period-1 orbit would consist of a single point index κ (red, blue or green), and group M (1, 2, . . .
attractor, or an Isle of Eden,5 on the main diagonal or 6) from Table 3. For future reference, the com-
line, after deleting points belonging to the transient plexity index κ and class M of all 256 rules are
regime. listed in Table 4. Counting the number of globally
equivalent rules from each class from Tables 3 and
Group 2 Rules 4, respectively, we summarize their distributions in
Almost all space-time patterns converge to a Figs. 5 and 6, respectively.
period-2 orbit. The time-1 map of each period-2
attractor, or Isle of Eden, consists of two points, 1.6. The “Magic” rule spaces
symmetrical with respect to the main diagonal line.
In [Cattaneo & Quaranta Vogliotti, 1997], a subset
Group 3 Rules of 104, among 256, local rules have been derived
Almost all space-time patterns converge to a and shown to exhibit “neural-like” behaviors. The
period-3 orbit. The time-1 map of the period-3 authors’ approach is based on an exhaustive mathe-
attractor, or Isle of Eden, consists of three points. matical analysis on a bi-infinite sequence space, con-
suming more than 20 printed pages. The authors
Group 4 Rules were so perplexed by their discovery that they
Almost all space-time patterns converge to a dubbed these rules “magic ”.
Bernoulli στ -shift attractor, or Isle of Eden, where A cursory inspection of the 256 Boolean cubes
|σ| ∈ {1, 2, 3} and |τ | ∈ {1, 2, 3, 4, 5}. listed in Table 1 would extract, in a few minutes,
We stress that the above qualitative behaviors 104 local rules with a complexity index κ = 1,
do not depend on the length L of the bit strings, and namely, those Boolean cubes whose red vertices can
do not depend on the initial configurations, even be separated from the blue vertices by no more than
4
This list is not unique in the sense that one can pick many other groups containing 88 independent rules. Our choice is
obtained by scanning the 256 rules from N = 0 to N = 255 , and deleting any rule that is equivalent to a previously listed
rule.
5
Robust Isles of Eden can be observed only for those rules endowed with dense Isles of Eden orbits [Chua et al., 2007a,
2007b].
Table 3. List of 88 globally independent rules. Color surrounding rule number N corresponds to complexity index
κ = 1 (red), 2 (blue) or 3 (green). The integer on the lower right corner identifies the characteristic property of the
rule, as specified in the color legend.
0 1 2 3 4 5 6 7
1 2 4 4 1 2 4 4
8 9 10 11 12 13 14 15
1 4 4 4 1 1 4 4
18 19 22 23 24 25 26 27
5 2 5 2 4 4 6 4
28 29 30 32 33 34 35 36
2 2 6 1 2 4 4 1
37 38 40 41 42 43 44 45
2 4 1 6 4 4 1 6
46 50 51 54 56 57 58 60
4 2 2 5 4 4 4 6
62 72 73 74 76 77 78 90
3 1 5 4 1 1 1 5
94 104 105 106 108 110 122 126

1 1 5 6 2 6 5 5
128 130 132 134 136 138 140 142

1 4 1 4 1 4 1 4
146 150 152 154 156 160 162 164

5 5 4 6 2 1 4 1
168 170 172 178 184 200 204 232

1 4 1 2 4 1 1 1
Color Legend
N N N N N N
1 2 3 4 5 6
Period-1 Period-2 Period-3 Bernoulli Complex Hyper
Table 4. List of 256 local rules. Color surrounding each rule number N corresponds to the complexity index κ = 1 (red), 2 (blue) or 3 (green) of N .
The integer in the lower right corner identifies the characteristic property of the rule, as specified in Table 3.
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
1 2 4 4 1 2 4 4 1 4 4 4 1 1 4 4
16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
4 4 5 2 4 4 5 2 4 4 6 4 2 2 6 4
32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47
1 2 4 4 1 2 4 4 1 6 4 4 1 6 4 4
48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63
4 4 2 2 4 4 5 2 4 4 4 4 6 4 3 4
64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79
1 4 4 4 1 1 2 2 1 5 4 6 1 1 1 1
80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95
4 4 6 4 4 4 6 4 4 6 5 2 1 1 1 2
96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111
1 6 4 4 1 6 6 4 1 5 6 6 2 5 6 4
112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127
4 4 4 4 4 4 3 4 6 6 5 5 4 4 5 2
2517
128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143
1 5 4 3 1 1 4 6 1 6 4 4 1 1 4 4
144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159
4 3 5 5 4 6 5 5 4 6 6 4 2 2 4 4
160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175
1 5 4 4 1 5 6 6 1 6 4 4 1 4 4 4
176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191
4 4 2 2 6 6 5 5 4 4 4 4 4 4 4 4
192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207
1 6 4 6 1 1 2 2 1 2 1 1 1 1 1 1
208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223
4 4 6 4 4 4 4 4 1 1 1 1 1 1 1 1
224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239
1 6 4 4 1 4 4 4 1 1 1 1 1 1 1 1
240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255
4 4 4 4 4 4 4 4 1 1 1 1 1 1 1 1
κ = 1 (Red) 104 rules, κ = 2 (Blue) 126 rules, κ = 3 (Green) 26 rules

1.7. Symmetries among Boolean cubes

Many of the 256 Boolean cubes in Table 1 share
interesting symmetrical features which give rise to
important predictable dynamics and applications.
We will briefly recall some of these symmetries and
present new interpretations.
1.7.1. Local complementation T c

We define the local complementation
Tc
N −→ N c (9)
of a Boolean cube N to be the Boolean cube N c
obtained by complementing the color of each ver-
Fig. 5. Partitioning of the 88 globally-independent rules into tex, i.e. 0 → 1 and 1 → 0 . Table 6 shows ten
6 classes. Boolean cubes and their local complements. This
transformation is called local to emphasize that the
space-time patterns of N and N c , with the same
initial configuration, is not the complement of each
other, because the complementation is valid only for
one iteration. This observation is demonstrated in
Fig. 5 of [Chua et al., 2004], where
Tc
110 −→ 145 (10)
Observe that the space-time patterns of 110 and
145 (for the same initial configuration) are not
the complement of each other, except for the first
iteration.8
1.7.2. Three equivalence transformations

T † , T , and T ∗
There are exactly three global transformations that
hold for all iterations, and for all initial config-
urations. They are the Left-Right Transformation
Fig. 6. Partitioning of the 256 local rules into 6 classes. T † , the Global Complementation T , and the Left-
Right Complementation T ∗ [Chua et al., 2004].
These three transformations, along with the identity
one plane.6 These 104 rules are listed in Table 5 transformation, have been shown in [Chua et al.,
along with their classification number M , extracted 2004] to form an Abelian group known as Klein’s
from Table 4. A comparison of the 104 κ = 1 rules Vierergruppe.
in Table 5 with those derived in [Cattaneo & Quar- Given any local rule N , the transformed rules
anta Vogliotti, 1997] shows that they are identical.7 N † T † (N ), N T (N ), and N ∗ T ∗ (N ) can
From our Boolean cube perspective, the “magic” be derived by inspection via the simple geometrical
connotation is perhaps a bit of an anti-climax. operations illustrated in Fig. 7.
6
Since no plane is needed for rules 0 and 255 , these two rules may be reclassified with a complexity index κ = 0.
7
Except for rule “36” listed in Tables 4 and 8 of [Cattaneo & Quaranta Vogliotti, 1997], which we believe is a typo that should
be rectified to rule 136, as correctly reported in Fig. 12 of the same paper.
8
There are some rules, however, where these local complementation T c (N ) coincides with the global complementation T (N )
to be defined below. In such cases, the space-time patterns of N and N c are also complements of each other for all t.
Table 5. A gallery of 104 linearly-separable local rules. The red color engulfing each rule number N implies a complexity
index κ = 1 for all 104 rules.
: List of 104 Linearly-Separable Boolean Function Rules.

0 1 2 3 4 5 7 8 10 11 12 13 14
1 2 4 4 1 2 4 1 4 4 1 1 4
15 16 17 19 21 23 31 32 34 35 42 43 47
4 4 4 2 4 2 4 1 4 4 4 4 4
48 49 50 51 55 59 63 64 68 69 76 77 79
4 4 2 2 2 4 4 1 1 1 1 1 1
80 81 84 85 87 93 95 112 113 115 117 119 127

4 4 4 4 4 1 2 4 4 4 4 4 2
128 136 138 140 142 143 160 162 168 170 171 174 175
1 1 4 1 4 4 1 4 1 4 4 4 4
176 178 179 186 187 191 192 196 200 204 205 206 207
4 2 2 4 4 4 1 1 1 1 1 1 1
208 212 213 220 221 223 224 232 234 236 238 239 240
4 4 4 1 1 1 1 1 1 1 1 1 4
241 242 243 244 245 247 248 250 251 252 253 254 255
4 4 4 4 4 4 1 1 1 1 1 1 1
Color Period-1 N Period-2 N Bernoulli N

Legend Rule 1 Rule 2 Rule 4
To derive the left-right transformation N † pairs, respectively,

we
obtain its global comple-
T † (N ) of any local rule N , simply obtain the mir- ment
137

= T 110 , and conversely, 110 =
ror image of the Boolean cube N about the main T 137 .
diagonal plane (shown shaded in Fig. 7(a)) pass- To derive the left-right complementation N ∗
∗
ing through the vertex 0m . Note that this opera- T (N ) of any local rule N , simply take the global
tion can be implemented by identifying each pair of complementation first, followed by the left-right
symmetrically located vertices with opposite colors, transformation, or vice-versa, i.e.
and then
complementing the colors.

Hence 124 =
†
T 110 , and 110 = T 124 . † N ∗ T ∗ (N ) = T † (T (N )) = T (T † (N )) (11)
To derive the global complementation N
T (N ) of any local rule N , simply identify each For example, consider taking first the global com-
pair of diagonally opposite vertices that have the plementation of 110 to obtain 137 = T 110
same color (either both red or both blue), and in Fig. 7(b). If we follow this operation by apply-
then change the color. For the example illustrated ing the left-right transformation

to 137 , we would
in Fig. 7(b), we have N = 110 . Among the obtain 193 = T † 137 by reflecting the Boolean
four pairs of diagonally opposite vertices of 110 , cube 137 in Fig. 7(b) about the main diagonal, as
we find that only three
pairs of vertices 0m, shown in Fig. 7(c), to obtain
m
7 , m m
2 , 5 , and m
1, 6 m have the same

colors. Changing only the color of these three 193 = T † 137 = T † T 110 (12)
Table 6. Some Boolean cubes and their local complements.
c c
N N N N
2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5
30 225 110 145
2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5
45 210 150 105
2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5
60 195 154 101
2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5
90 165 170 85
2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5
105 150 184 71

Fig. 7. Geometrical constructions for deriving (a) N † T † (N ), (b) N T (N ), and (c) N ∗ T ∗ (N ).

Table 7. A gallery of 16 centrally-symmetric local rules.
2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5
0 24 36 60
2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5
66 90 102 126
2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5
129 153 165 189
2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5
195 219 231 255
1.7.3. Perfect complementary rules in Table 7. Since these Boolean cubes exhibit per-
fect symmetry in color with respect to the origin
It follows from the geometrical construction of the
located at the center of the cube, we will henceforth
global complementary transformation T (N ) → N
call them centrally-symmetric local rules. Clearly,
in Fig. 7(b) that
if all four

pairs of
diagonally

oppo-
the local complementary space-time patterns of all
site
vertices 0m, 7m, 2m , 5m, 3m , 4m, 1m ,
m centrally-symmetric local rules hold for all times.
6 of N have identical colors, respectively, then
N T (N ) = T C (N ) N C (13) 1.7.4. Permutive rules9

We will henceforth call such perfectly symmetri- There are 28 local rules whose Boolean cubes
cal transformations (relative to the complementa- exhibit an anti-symmetry with respect to some
tion transformation) perfect complementations. The vertical plane through the center of the cube, as
set of all perfect complementary rules are listed illustrated in Fig. 8.
9
Permutive rules are originally defined by [Hedlund, 1969] on a formal topological setting. We have opted for an equivalent
but geometrical definition via the Boolean cubes for pedagogical reasons. These two representations are equivalent, as shown
in Appendix C.
(a) (b) (c)
Fig. 8. Geometrical illustrations of permutive rules. (a) Rule 30 is Left-Permutive because the colors of the vertices
˘ ¯ ˘ ¯
0m, 1m , 2m , 3m in the back face are the complement of the colors of the vertices 4m, 5m , 6m, 7m in the front
˘ ¯
m m m m in the right face are the com-
face. (b) Rule 154 is Right-Permutive ¯ of the vertices 1 , 3 , 5 , 7
˘ because the colors
m m m
plement of the colors of the vertices 0 , 2 , 4 , 6 m on the left face. (c) Rule 150 is Bi-Permutive because it is both
Left and Right Permutive.
A local rule N is said to be Left-Permutive, depicted in Table 13. It is easy to prove that each
iff the vertical symmetry plane is parallel to the rule N in Table 13 has a unique decomposition via
paper, as depicted by the “green” plane in Fig. 8(a). the eight Boolean cubes basis functions in Table 12.
It is said to be Right-Permutive, iff the vertical
symmetry plane is perpendicular to the paper, as
depicted by the “pink ” plane in Fig. 8(b). It is said 1.7.6. Rules with explicit period-1 and/or
to be Bi-Permutive, iff it is both Left- and Right- period-2 orbits
Permutive as depicted by the “green” and “pink ” Recall from Fig. 6 that among the 256 local rules, 69
vertical symmetry planes, respectively. A local rule are endowed with robust period-1 (attractor or Isle-
N is said to be Permutive iff N is Left and/or of-Eden) orbits, and another 25 rules are endowed
Right-Permutive. with period-2 orbits. It is generally impossible to
An examination of the 256 Boolean cubes in predict the bit-string pattern of such period-1 or
Table 1 shows that there are only 16 Left-Permutive period-2 orbits without actually evolving the rule
rules, 16 Right-Permutive rules, and 4 Bi-Permutive from some initial state.
rules, as displayed in Tables 8–10, respectively. The The purpose of this subsection is to prove a sur-
union of all these local rules gives only 28 distinct prising and quite remarkable result asserting that
Permutive rules, as exhibited in Table 11. We will the period-1 and period-2 bit strings of a large num-
show in the following sections that Permutive rules ber of local rules can be predicted without carrying
possess some remarkable properties. out any simulations. Such period-k bit string (k = 1
or k = 2) patterns are endowed upon those
Boolean
1.7.5. Superposition of local rules cubes whose main-diagonal vertices 0m , 2m
, 5m ,

m
7 exhibit certain color combinations.
The eight Boolean cubes exhibited in Table 12 are
independent in the sense that it is impossible to
Explicit period-(1, 2) pattern theorem
decompose any of them into the “union” of two
or more simpler Boolean cubes by taking the logic There are ten distinct
color combinations
among
“OR” operation between the colors of correspond- m m m m
the four vertices 0 , 2 , 5 , 7 on the main-
ing vertices, where “red ” is coded “1” and “blue” diagonal plane of the Boolean cubes, labeled Type
is coded “0”, respectively. Since each of these eight A, B, . . . , J in Tables 14(A), (B), . . . ,(J) for which
Boolean cubes contains one, and only one, red ver- the corresponding local rules have an explicit
tex, together they constitute a basis function where period-1 and/or period-2 bit-string pattern, regard-
the “union” of two or more such rules can generate less of the colors of the remaining nondiagonal ver-
any of the remaining 256 − 8 = 248 local rules, as tices 1m , 3m, 4m
, 6m.
Table 8. Sixteen left-permutive rules.
Proof. It follows directly from the evolution of the 3. The Boolean cube inset in Tables 14(G)–14(J)
Boolean cube inset on top of each table. Here, the has four color vertices on the main diagonal.
“white” vertices denote irrelevant vertices.
Remarks 1.7.7. Most rules harbor at least one

1. The Boolean cube inset in Tables 14(A) and Isle of Eden
14(B) have only one color vertex along the main Among the 256 local rules, 228 harbored at least
diagonal. one Isle of Eden. In particular, there are 79 (out of
2. The Boolean cube inset in Tables 14(C)–14(F) 88) topologically distinct rules with at least one Isle
has two color vertices (with opposite colors) on of Eden, and Table 15 displays the initial and final
the main diagonal. bit-string configurations of a typical Isle of Eden
Table 9. Sixteen right-permutive rules.
orbit for each of them, so that readers can verify τ = 480 return map in Fig. 10(b), we see the φn —
that it is indeed periodic. versus — φn−τ time-τ map is virtually identical to
We end this recap section by exhibiting in Fig. 9 the Bernoulli map
a period-3240 Isle of Eden of rule 30 . with L = 27,
φn = 2φn−τ mod 1 (14)
a real gem that should trigger a rush to uncover
Isles of Eden with even longer periods. associated with the time-1 return map of rule 170 .
The φn — versus — φn−1 return map of this Since it is well known that the Bernoulli map
period-3240 Isle of Eden is shown in Fig. 10(a). Note Eq. (14) is ergodic, Fig. 10(b) strongly suggests
the very messy plot gives very little information. empirically that rule 30 is quasi-ergodic in the
However, if we plot this Isle of Eden orbit as a sense of Definition 2.1 of Sec. 2.1.
Fig. 9. Orbit of a period-3240 (with σ = 1, τ = 480) Isle of Eden of 30 with L = 27. The number inscribed inside each
“capsule” is the decimal representation of a 27-bit string.
Fig. 9. (Continued )
Table 10. Four bi-permutive rules.
2528
Table 11. Twenty-eight permutive rules.

Table 12. Eight Boolean cube basis functions.
2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5
1 2 4 8
2530
2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5
16 32 64 128
Table 13. Decomposition of χ1 into superposition of eight basis characteristic functions.
N
1 2 4 8 16 32 64 128 1 2 4 8 16 32 64 128 1 2 4 8 16 32 64 128

0 22 2 4 16 44 4 8 32
1 1 23 1 2 4 16 45 1 4 8 32
2 2 24 8 16 46 2 4 8 32
3 1 2 25 1 8 16 47 1 2 4 8 32
4 4 26 2 8 16 48 16 32
5 1 4 27 1 2 8 16 49 1 16 32
6 2 4 28 4 8 16 50 2 16 32
7 1 2 4 29 1 4 8 16 51 1 2 16 32
8 18 30 2 4 8 16 52 4 16 32
9 1 18 31 1 2 4 8 16 53 1 4 16 32
2531
10 2 18 32 32 54 2 4 16 32
11 1 2 18 33 1 32 55 1 2 4 16 32
12 4 18 34 2 32 56 8 16 32
13 1 4 18 35 1 2 32 57 1 8 16 32
14 2 4 18 36 4 32 58 2 8 16 32
15 1 2 4 18 37 1 4 32 59 1 2 8 16 32
16 16 38 2 4 32 60 4 8 16 32
17 1 16 39 1 2 4 32 61 1 4 8 16 32
18 2 16 40 8 32 62 2 4 8 16 32
19 1 2 16 41 1 8 32 63 1 2 4 8 16 32
20 4 16 42 2 8 32 64 64
21 1 4 16 43 1 2 8 32 65 1 64
1 2 4 8 16 32 64 128 1 2 4 8 16 32 64 128 1 2 4 8 16 32 64 128

66 2 64 88 8 16 64 110 2 4 8 32 64
67 1 2 64 89 1 8 16 64 111 1 2 4 8 32 64
68 4 64 90 2 8 16 64 112 16 32 64
69 1 4 64 91 1 2 8 16 64 113 1 16 32 64
70 2 4 64 92 4 8 16 64 114 2 16 32 64
71 1 2 4 64 93 1 4 8 16 64 115 1 2 16 32 64
72 8 64 94 2 4 8 16 64 116 4 16 32 64
73 1 8 64 95 1 2 4 8 16 64 117 1 4 16 32 64
74 2 8 64 96 32 64 118 2 4 16 32 64
75 1 2 8 64 97 1 32 64 119 1 2 4 16 32 64
2532
76 4 8 64 98 2 32 64 120 8 16 32 64
77 1 4 8 64 99 1 2 32 64 121 1 8 16 32 64
78 2 4 8 64 100 4 32 64 122 2 8 16 32 64
79 1 2 4 8 64 101 1 4 32 64 123 1 2 8 16 32 64
80 16 64 102 2 4 32 64 124 4 8 16 32 64
81 1 16 64 103 1 2 4 32 64 125 1 4 8 16 32 64
82 2 16 64 104 8 32 64 126 2 4 8 16 32 64
83 1 2 16 64 105 1 8 32 64 127 1 2 4 8 16 32 64
84 4 16 64 106 2 8 32 64 128 128
85 1 4 16 64 107 1 2 8 32 64 129 1 128
86 2 4 16 64 108 4 8 32 64 130 2 128
87 1 2 4 16 64 109 1 4 8 32 64 131 1 2 128
1 2 4 8 16 32 64 128 1 2 4 8 16 32 64 128 1 2 4 8 16 32 64 128

132 4 128 154 2 8 16 128 176 16 32 128
133 1 4 128 155 1 2 8 16 128 177 1 16 32 128
134 2 4 128 156 4 8 16 128 178 2 16 32 128

135 1 2 4 128 157 1 4 8 16 128 179 1 2 16 32 128
136 8 128 158 2 4 8 16 128 180 4 16 32 128
137 1 8 128 159 1 2 4 8 16 128 181 1 4 16 32 128
138 2 8 128 160 32 128 182 2 4 16 32 128
139 1 2 8 128 161 1 32 128 183 1 2 4 16 32 128

140 4 8 128 162 2 32 128 184 8 16 32 128
141 1 4 8 128 163 1 2 32 128 185 1 8 16 32 128
2533
142 2 4 8 128 164 4 32 128 186 2 8 16 32 128
143 1 2 4 8 128 165 1 4 32 128 187 1 2 8 16 32 128
144 16 128 166 2 4 32 128 188 4 8 16 32 128

145 1 16 128 167 1 2 4 32 128 189 1 4 8 16 32 128
146 2 16 128 168 8 32 128 190 2 4 8 16 32 128

147 1 2 16 128 169 1 8 32 128 191 1 2 4 8 16 32 128
148 4 16 128 170 2 8 32 128 192 64 128
149 1 4 16 128 171 1 2 8 32 128 193 1 64 128
150 2 4 16 128 172 4 8 32 128 194 2 64 128
151 1 2 4 16 128 173 1 4 8 32 128 195 1 2 64 128
152 8 16 128 174 2 4 8 32 128 196 4 64 128

153 1 8 16 128 175 1 2 4 8 32 128 197 1 4 64 128
1 2 4 8 16 32 64 128 1 2 4 8 16 32 64 128 1 2 4 8 16 32 64 128

198 2 4 64 128 220 4 8 16 64 128 242 2 16 32 64 128
199 1 2 4 64 128 221 1 4 8 16 64 128 243 1 2 16 32 64 128
200 8 64 128 222 2 4 8 16 64 128 244 4 16 32 64 128
201 1 8 64 128 223 1 2 4 8 16 64 128 245 1 4 16 32 64 128
202 2 8 64 128 224 32 64 128 246 2 4 16 32 64 128
203 1 2 8 64 128 225 1 32 64 128 247 1 2 4 16 32 64 128
204 4 8 64 128 226 2 32 64 128 248 8 16 32 64 128
205 1 4 8 64 128 227 1 2 32 64 128 249 1 8 16 32 64 128
206 2 4 8 64 128 228 4 32 64 128 250 2 8 16 32 64 128
207 1 2 4 8 64 128 229 1 4 32 64 128 251 1 2 8 16 32 64 128
2534
208 16 64 128 230 2 4 32 64 128 252 4 8 16 32 64 128
209 1 16 64 128 231 1 2 4 32 64 128 253 1 4 8 16 32 64 128
210 2 16 64 128 232 8 32 64 128 254 2 4 8 16 32 64 128
211 1 2 16 64 128 233 1 8 32 64 128 255 1 2 4 8 16 32 64 128
212 4 16 64 128 234 2 8 32 64 128
213 1 4 16 64 128 235 1 2 8 32 64 128
214 2 4 16 64 128 236 4 8 32 64 128
215 1 2 4 16 64 128 237 1 4 8 32 64 128
216 8 16 64 128 238 2 4 8 32 64 128
217 1 8 16 64 128 239 1 2 4 8 32 64 128
218 2 8 16 64 128 240 16 32 64 128

219 1 2 8 16 64 128 241 1 16 32 64 128
Table 14(A). There are 128 local rules endowed with a period-1 spatially-homogeneous blue orbit · · · for all
L = 3, 4, 5, . . . , n.
Type A
2 3 There are 128 Local Rules with a type A boolean cube defined by a blue color
6 7 at vertex 0 ( ).
0 1
4 5
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
0 2 4 6 8 10 12 14 16 18 20
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
22 24 26 28 30 32 34 36 38 40 42
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
44 46 48 50 52 54 56 58 60 62 64
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
66 68 70 72 74 76 78 80 82 84 86
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
88 90 92 94 96 98 100 102 104 106 108

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
110 112 114 116 118 120 122 124 126 128 130
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
132 134 136 138 140 142 144 146 148 150 152
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
154 156 158 160 162 164 166 168 170 172 174
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
176 178 180 182 184 186 188 190 192 194 196
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
198 200 202 204 206 208 210 212 214 216 218
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
220 222 224 226 228 230 232 234 236 238 240
2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5
242 244 246 248 250 252 254

Table 14(B). There are 128 local rules endowed with a period-1 spatially-homogeneous red orbit · · · for all
L = 3, 4, 5, . . . , n.
Type B
2 3 There are 128 Local Rules with a type B boolean cube defined by a red color
6 7 at vertex 7 ( ).
0 1
4 5
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
128 129 130 131 132 133 134 135 136 137 138
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
139 140 141 142 143 144 145 146 147 148 149
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
150 151 152 153 154 155 156 157 158 159 160
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
161 162 163 164 165 166 167 168 169 170 171
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
172 173 174 175 176 177 178 179 180 181 182
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
183 184 185 186 187 188 189 190 191 192 193
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
194 195 196 197 198 199 200 201 202 203 204
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
205 206 207 208 209 210 211 212 213 214 215
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
216 217 218 219 220 221 222 223 224 225 226
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
227 228 229 230 231 232 233 234 235 236 237
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
238 239 240 241 242 243 244 245 246 247 248
2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5
249 250 251 252 253 254 255

Table 14(C). There are 64 local rules endowed with two period-1 spatially-homogeneous orbits · · · and · · · for
all L = 3, 4, 5, . . . , n.
Type C
2 3 There are 64 Local Rules whose Boolean Cube has a
6 7
0 1 type C color combination at vertex ( ) and
4 5 vertex ( )
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
128 130 132 134 136 138 140 142

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
144 146 148 150 152 154 156 158

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
160 162 164 166 168 170 172 174

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
176 178 180 182 184 186 188 190

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
192 194 196 198 200 202 204 206

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
208 210 212 214 216 218 220 222

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
224 226 228 230 232 234 236 238

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
240 242 244 246 248 250 252 254

Table 14(D). There are 64 local rules endowed with two period-1 spatially-alternating orbits · · · and · · ·
for all even L = 4, 6, 8, . . . , 2n.
Type D
6 7
0 1 type D color combination at vertex ( ) and
4 5 vertex ( )
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
4 5 6 7 12 13 14 15
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
20 21 22 23 28 29 30 31
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
68 69 70 71 76 77 78 79
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
84 85 86 87 92 93 94 95
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
132 133 134 135 140 141 142 143

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
148 149 150 151 156 157 158 159

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
196 197 198 199 204 205 206 207

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
212 213 214 215 220 221 222 223

Table 14(E). There are 64 local rules endowed with a period-2 spatially-homogeneous orbit · · · ↔ · · · for all
L = 3, 4, 5, . . . , n.
Type E
6 7
0 1 type E color combination at vertex ( ) and
4 5 vertex ( )
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
1 3 5 7 9 11 13 15
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
17 19 21 23 25 27 29 31
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
33 35 37 39 41 43 45 47
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
49 51 53 55 57 59 61 63
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
65 67 69 71 73 75 77 79
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
81 83 85 87 89 91 93 95
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
97 99 101 103 105 107 109 111

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
113 115 117 119 121 123 125 127

Table 14(F). There are 64 local rules endowed with a period-2 spatially alternating orbit · · · ↔ · · · for
all even L = 4, 6, 8, . . . , 2n.
Type F
6 7
0 1 type F color combination at vertex ( ) and
4 5 vertex ( )
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
32 33 34 35 40 41 42 43
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
48 49 50 51 56 57 58 59
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
96 97 98 99 104 105 106 107

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
112 113 114 115 120 121 122 123

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
160 161 162 163 168 169 170 171

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
176 177 178 179 184 185 186 187

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
224 225 226 227 232 233 234 235

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
240 241 242 243 248 249 250 251

Table 14(G). There are 16 local rules endowed with two period-1 spatially-homogeneous orbits · · · and · · ·
for all L = 3, 4, 5, . . . , n, as well as two period-1 spatially-alternating orbits · · · and · · · for all even
L = 4, 6, 8, . . . , 2n.
Type G
6
0
7
1
type G color combination at vertex ( ),
4 5 vertex ( ), vertex ( ), and
vertex ( )
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
132 134 140 142 148 150 156 158

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
196 198 204 206 212 214 220 222
Table 14(H). There are 16 local rules endowed with two period-2 spatially-homogeneous orbits · · · and · · ·
for all L = 3, 4, 5, . . . , n, as well as two period-2 spatially-alternating orbits · · · and · · · for all even
L = 4, 6, 8, . . . , 2n.
Type H
6
0
7
1
type H color combination at vertex ( ),
vertex ( )
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
33 35 41 43 49 51 57 59
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
97 99 105 107 113 115 121 123

Table 14(I). There are 16 local rules endowed with two period-1 spatially-homogeneous orbits · · · and · · · for all
L = 3, 4, 5, . . . , n, as well as a period-2 spatially-alternating orbit · · · ↔ · · · for all even L = 4, 6, 8, . . . , 2n.
Type I
6
0
7
1
type I color combination at vertex ( ),
vertex ( )
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
160 162 168 170 176 178 184 186

2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
224 226 232 234 240 242 248 250
Table 14(J). There are 16 local rules endowed with two period-1 spatially-alternating orbits · · · and · · ·
for all even L = 4, 6, 8, . . . , 2n, as well as a period-2 spatially-homogeneous orbit · · · ↔ · · · for all
L = 3, 4, 5, . . . , n.
Type J
6
0
7
1
type J color combination at vertex ( ),
vertex ( )
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
5 7 13 15 21 23 29 31
2 3 2 3 2 3 2 3 2 3 2 3 2 3 2 3
6 7 6 7 6 7 6 7 6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5 4 5 4 5 4 5 4 5
69 71 77 79 85 87 93 95
Table 15. Gallery exhibiting the initial (t = 0) and final bit string (t = T − 1) for an Isle of Eden of the 79
topologically-independent rules having at least one of them.
Initial bit string x0 and Initial bit string x0 and

N L T final bit string xT-1 N L T final bit string xT-1
x0 x0
1 9 2 x1 27 9 18 x17
x0 x0
2 9 3 x2 28 6 2 x1
x0 x0
3 9 18 x17 29 9 2 x1
x0 x0
4 9 1 x0 30 8 1 x0
x0 x0
5 9 2 x1 32 8 2 x1
x0 x0
6 8 8 x7 33 8 2 x1
x0 x0
7 8 1 x0 34 8 2 x1
x0 x0
9 9 3 x2 35 9 18 x17
x0 x0
10 8 4 x3 37 7 7 x6
x0 x0
11 9 2 x1 38 9 18 x17
x0 x0
12 8 1 x0 40 9 9 x8
x0 x0
13 9 2 x1 41 9 9 x8
x0 x0
14 9 18 x17 42 9 9 x8
x0 x0
15 9 18 x17 43 9 18 x17
x0 x0
18 8 2 x1 44 9 3 x2
x0 x0
19 8 2 x1 45 9 504 x503
x0 x0
22 8 2 x1 50 8 2 x1
x0 x0
23 9 2 x1 51 9 2 x1
x0 x0
25 9 2 x1 54 8 8 x7
x0 x0
26 8 16 x15 56 9 9 x8
Initial bit string x0 and Initial bit string x0 and

N L T final bit string xT-1 N L T final bit string xT-1
x0 x0
57 9 3 x2 138 9 9 x8
x0 x0
58 9 18 x17 140 9 1 x0
x0 x0
62 9 3 x2 142 9 18 x17
x0 x0
72 9 1 x0 146 9 2 x1
x0 x0
73 9 2 x1 150 8 4 x3
x0 x0
74 9 1 x0 152 9 1 x0
x0 x0
76 9 1 x0 154 9 72 x71
x0 x0
77 9 2 x1 156 9 2 x1
x0 x0
94 8 2 x1 160 9 1 x0
x0 x0
104 9 1 x0 162 9 1 x0
x0 x0
105 8 4 x3 164 9 1 x0
x0 x0
106 9 9 x8 168 9 9 x8
x0 x0
108 9 2 x1 170 9 9 x8
x0 x0
110 8 8 x7 172 9 3 x2
x0 x0
122 4 2 x1 178 9 1 x0
x0 x0
128 9 1 x0 184 9 9 x8
x0 x0
130 9 3 x2 200 9 1 x0
x0 x0
132 9 1 x0 204 9 1 x0
x0 x0
134 9 1 x0 232 9 1 x0
x0
136 9 1 x0
(a) (b)
Fig. 10. (a) Time-1 return map of period-3240 Isle of Eden. (b) Time-τ return map of the same orbit with τ = 480 resembles
the left-shift Bernoulli-map.
2. Quasi-Ergodicity Remarkably, “quasi-ergodicity” captures a

peculiarity unique among the complex and hyper
In the previous parts of our saga about Cellular
Bernoulli rules, thereby adding a further raison
Automata we introduced two kinds of graphical rep-
d’être for the classification summarized in Sec. 1.5,
resentations for the local rules: the time-1 return
besides those based on the period, or the Bernoulli
map and the time-1 characteristic function [Chua
στ -shift, of the attractors defined in [Chua et al.,
et al., 2005a]. The first one illustrates the evolu-
2007a], which differ from the complex (group 5) and
tion from a specific initial bit string as a Lamerey
hyper (group 6) Bernoulli-shift rules in the fact that
(cobweb) diagram for all times; the second one shows
they are independent from the initial configuration
the output bit string corresponding to any pos-
and the length L of the bit strings.
sible input for one iteration. In general, a local
rule has a unique time-1 characteristic function 2.1. Only complex and hyper Bernoulli-
but many time-1 return maps, one for each initial
shift rules are quasi-ergodic
condition.
Therefore, a question arises: Can the time-1 By visual inspection, we noticed that for certain
return map and the time-1 characteristic function rules the time-1 return maps corresponding to dif-
coincide? In other words, is it possible to extract ferent initial conditions are indistinguishable, and
information about the temporal behavior — which they tend to be very similar to the time-1 charac-
is the time-1 return map — of a Cellular Automa- teristic function too. As we mentioned before, this
ton by looking at its spatial representation, like the behavior is somehow related to “ergodicity” but,
time-1 characteristic function? in order to emphasize that this phenomenon has
The correspondence between the temporal and been observed empirically, we preferred to introduce
the spatial behaviors of an important class of the property of “quasi-ergodicity” via the following
dynamical systems is a well-known concept dubbed definition.
“ergodicity”, and it is fundamental for a num- Definition 2.1. Quasi-ergodicity. A rule is said to
ber of branches of mathematics and physics. Since be quasi-ergodic iff given an arbitrary point P
a thorough treatment of ergodicity would go belonging to an attractor ΛP , it is possible to find
beyond the purpose of this introductory section, a point Q arbitrarily close to it which belongs to
we will opt for a pedagogically more transpar- another attractor ΛQ which is visually indistin-
ent approach by introducing the empirical con- guishable from ΛP .
cept “quasi-ergodicity”10 to stress its difference
with “ergodicity”, which would require an in-depth A graphical illustration of the quasi-ergodic
measure-theoretic analysis. property is given in Fig. 11, for a generic rule N .
10
“Quasi-ergodicity” is an expression widely used also in statistical physics; however, in this context we employ it in a different,
albeit related, way.
sub-intervals made of Gardens of Eden, and isolated

1
Isles of Eden which are not visible due to printing
Φn resolution.
Furthermore, the vignettes in Table 17 show
0.75 point P belonging to that for these 18 rules the time-1 return map for
attractor ΛP through P an arbitrary initial condition tends to be very simi-
lar (except for rule 73 ) to the time-1 characteristic
Q function, as required by quasi-ergodicity.
0.5 An accurate analysis reveals that rule 73 is
-ε P ε an exception to this general behavior, because dif-
ferent initial conditions generate different time-
point Q arbitrarily
arbitrarly close
close 1 return maps. Nevertheless, we noticed that
0.25 (|P-Q ε ) to P belongs to
(|P-Q|< rule 73 has only two kinds of possible time-1
another attractor ΛQ return maps corresponding to two sets of mutu-
ally exclusive initial conditions (except for, as
usual, a finite number of isolated periodic points
0 or Isles of Eden) exhibited in Table 17. More-
0 0.25 0.5 0.75 Φ n-1 1 over, all initial conditions belonging to the same set
give indistinguishable time-1 return maps. Hence,
Fig. 11. For a quasi-ergodic rule, given an arbitrary point P instead of having quasi-ergodicity all over the
(in red) belonging to the attractor ΛP in the time-1 return axis φn−1 , there are two regions, each exhibiting
map, there is a nearby point Q (in blue) belonging to another
quasi-ergodicity. Therefore, we say that rule 73
attractor ΛQ which is empirically indistinguishable from ΛP .
is weakly quasi-ergodic, according to the following
definition.
Figure 12 shows one of the consequences of quasi- Definition 2.2. Weak quasi-ergodicity. A rule is
ergodicity of rule 110 : Since the time-1 return maps said to be weakly quasi-ergodic iff it is quasi-ergodic
corresponding to three different initial conditions in a finite number of disjoint regions, whose union
cling arbitrarily close to each other, the Lamerey gives the whole axis φn−1 .
(cobweb) diagrams (Figs. 12(a)–12(c)) obtained
from their superposition in Fig. 12(d) are visually Among the 256 local rules, 73 and its global
indistinguishable. topologically-equivalent rule 109 are the only
We cannot overemphasize that Definition 2.1 ones to exhibit weak quasi-ergodicity. The prop-
is only an empirical definition for the new phe- erty of weak quasi-ergodicity of rule 73 is evident
nomenon to be described in this section. from Fig. 13, in which the two different time-1
In particular, we have found that only the com- return maps are depicted in Figs. 13(a) and 13(b)
plex and hyper Bernoulli-shift rules, corresponding and then their union in Fig. 13(c) is compared
to the fifth and sixth group in Sec. 1, are quasi- with the time-1 characteristic function χ173 in
ergodic. This observation is remarkable because it Fig. 13(d).
introduces a particular feature possessed by only Finally, we can identify a further type of rules
rules belonging to these two groups, within the more within the quasi-ergodic ones thanks to the follow-
general class of Bernoulli-shift rules. So far, our dis- ing observation. While it is true that the time-1
tinction between the Bernoulli στ -shift rules from return map of most quasi-ergodic rules tends to
group 4 and the complex and hyper Bernoulli-shift cover almost everywhere the unit interval φn−1 ∈
rules from groups 5 and 6 is that the two Bernoulli [0, 1], sometimes relatively large compact regions
parameters σ and τ from group 4 do not depend on are excluded (notice the big gap for rule 18 in
the initial configuration, or the length L. Table 16 when φ > 0.8); in general, this behavior
Table 16 shows the Lamerey (cobweb) diagrams is observable by increasing the length L of the bit
for the 18 complex and hyper Bernoulli-shift rules. strings. Nonetheless, for some rules we cannot see
Observe that the map (starting from a random any gap in a generic time-1 return map indepen-
initial condition) tends to cover the whole hori- dently from the length L of the bit strings. These
zontal axis — except for some gaps formed by rules — namely 30 , 45 , 60 , 90 , 105 , 106 ,
(a) (b)
(c) (d)
Fig. 12. Lamerey (cobweb) diagrams corresponding to three different initial conditions for rule 110 : Because of quasi-
ergodicity, when the diagrams are superimposed they are practically indistinguishable.
Table 16. Lamerey (cobweb) diagrams for 18 complex and hyper Bernoulli-shift rules: Since these rules are quasi-ergodic, the
qualitative features of their Lamerey (cobweb) diagrams do not depend on the initial condition.
18 L=203 22 L=101
26 L=201 30 L=101
41 L=601 45 L=101
54 L=201 60 L=101
73 L=80 73 L=94
90 L=101 105 L=201

106 L=101 110 L=601
122 L=201 126 L=201

146 L=301 150 L=201
154 L=601
Table 17. Quasi-ergodic vignettes of 18 complex and hyper Bernoulli-shift rules. For each rule it is depicted: (a) the time-1 characteristic function, where χ(φ) is in
red (resp. blue) if the last bit of the string is 0 (resp. 1), and purple when both points coincide due to printer resolution; (b) the time-1 return map corresponding to
the initial condition indicated in (c); (c) the space-time pattern corresponding to a fixed initial condition.
2553
2554
2555
2556
2557
2558
2559
2560
2561
150 , 154 — will henceforth be called strongly second, the Isles of Eden; and third, the Gardens
quasi-ergodic.11 of Eden.
Definition 2.3. Strong quasi-ergodicity. A quasi-

2.2. Quasi-ergodicity and Gardens
ergodic rule is said to be strongly quasi-ergodic iff
its time-1 return maps tend to cover the whole unit of Eden
interval φ ∈ [0, 1], except for isolated points, inde- From Definition 2.3, it is clear that a necessary con-
pendently from the bit string length L. dition for strong quasi-ergodicity is the absence of
The gaps are created by three different mech- non-isolated compact sets of Gardens of Eden since,
anisms: First, the transients, because we do not as we mentioned previously, they generate gaps on
display them in the Lamerey (cobweb) diagrams; the horizontal axis. As an illustration, Table 18
11
It is also possible to prove that strongly quasi-ergodic rules, along with 15 and 170 , are actually ergodic, according to a
formal mathematical definition [Shirvani et al., 1991; Shereshevsky, 1992].
(a) (b)
(c) (d)
Fig. 13. Rule 73 exhibits a weaker form of quasi-ergodicity with 2 distinct time-1 maps (a) and (b), whose union in (c)
resembles the time-1 characteristic function in (d). Points in (d) not present in (c) are transient points.
Table 18. Non-strongly quasi-ergodic rules have compact sets of Gardens of Eden.
Table 19. Strongly quasi-ergodic rules do not have compact sets of Gardens of Eden.
gives examples of non-isolated compact set of Gar- functions of one-dimensional Cellular Automata,
dens of Eden for those quasi-ergodic rules that are and some partial explanations for this phenomenon
not strongly quasi-ergodic. In contrast, Table 19 were given. In this section we prove a general the-
shows that strongly quasi-ergodic rules can have orem that gives more thorough results and pro-
only isolated Gardens of Eden. vides an analytical formula for finding the fractal
Finally, we note that all strongly quasi-ergodic patterns.
rules are also permutive (see Tables 8 and 9). This
follows from [Hedlund, 1969], who proved that in
one-dimensional CA only permutive rules are sur- 3.1. All time-1 characteristic
jective, in the sense that they have no Gardens functions are fractals
of Eden when L → ∞.12 Hence, only permutive In order to specify explicitly the number of bits
rules can satisfy our necessary condition for strong contained in a generic string x, we introduce the
quasi-ergodicity. notation
. .
3. Fractals in 1D Cellular Automata xI+1 = (x0 . . . xk−1 .. ..xI ) (15)
In [Chua et al., 2005b] it was shown that frac- to indicate the set of bit strings composed of I + 1
tals arise naturally in the time-1 characteristic elements in which the first k elements and the last
12
When L is finite, injectivity implies surjectivity and vice versa, as proved in [Moore, 1962] and [Myhill, 1963].
. .
one are known. For example, x6 = (01.. ..1) is
equivalent to the set {(010001), (010011), (010101),
(010111), (011001), (011011), (011101), (011111)}.
Since any bit string x can be uniquely associ-
ated with a real number φ(x) ∈ [0, 1), the set of bit
strings defined by Eq. (15) corresponds to a subset
of points of the unit interval [0, 1], with the formula
. .
φI+1 (xI+1 ) = φI+1 (x0 . . . xk−1 .. ..xI )
I

= xi 2−(i+1) (16)
i=0
Similarly, the time-1 characteristic function for the

local rule N corresponding to the set of bit strings (a)
φI+1 (xI+1 ) is defined by
. .
χ1N ,I+1
(xI+1 ) = χ1N ,I+1
(x0 . . . xk−1 .. ..xI )
I

= T N (xi−1 xi xi+1 )2−(i+1)
i=0
(17)
where T N (xi−1 xi xi+1 ) denotes the application of
the local rule N to the triplet pattern (xi−1 xi xi+1 ).
To avoid clutter, we will usually delete the commas
separating bits, and abbreviate χ1 (xI+1 ) by
N ,I+1
χI+1 (xI+1 ).
Theorem 3.1. Fractality of the time-1 characteris-
tic function χ1 . (b)
N
The time-1 characteristic function χ1 of any
N Fig. 14. The four fractal patterns χα , χβ , χγ and χδ and
rule N exhibits a fractal behavior in any subinter- their position in the time-1 characteristic function.
val of φ ∈ [0, 1).
Proof. Given an arbitrary natural number I > 1, . .

.. .. and χβ = χI+1 (1.. ..0) in correspondence to φβ [see
we define the two bit strings xα I+1 = (0. .0) and
.. .. Fig. 14(a)].
xβ
I+1 = (1. .0). According to the notation intro- An analogous procedure can be followed when
duced previously, xαI+1 is the set of all bit strings xI = 1, leading to the definition of the two
composed by I + 1 elements with x0 = 0 and . . . .
strings xγI+1 = (0.. ..1) and xδI+1 = (1.. ..1),
xI = 0, whereas xβ I+1 is the set of all bit strings and their equivalent decimal representations φγ =
composed by I + 1 elements with x0 = 1 and . . . .
xI = 0. Their equivalent decimal representations φI+1 (0.. ..1) and φδ = φI+1 (1.. ..1), where 0 <
. . . . φγ < 1/2 and 1/2 ≤ φδ < 1. Also in this case the
are φα = φI+1 (0.. ..0) and φβ = φI+1 (1.. ..0), time-1 characteristic function can be divided into
and it is easy to prove that 0 ≤ φα < 1/2 and . .
two parts: χγ = χI+1 (0.. ..1) in correspondence to
1/2 ≤ φβ < 1. Since φα ∪ φβ = [0, 1) as I → ∞, we . .
can divide the time-1 characteristic function χ1 φγ and χδ = χI+1 (1.. ..1) in correspondence to φδ
N
[see Fig. 14(b)].
for a generic rule N into two parts, one for each Let us consider a generic set of bit strings
. .
interval: χα = χI+1 (0.. ..0) in correspondence to φα xk+I+1 , each composed of k +I +1 elements, having
in common the first k + 1 and the last element k+1

= T N (xi−1 xi xi+1 )2−(i+1)
k bits I+1 bits
i=0
.. ..
xk+I+1 = (x0 x1 . . . xk−1 xk . .xk+I ) (18) k+I+2

+ T N (xi−1 xi xi+1 )2−(i+1) (21)
According to the formula (16), we can associate to i=k+2
the set xk+I+1 a unique interval of values in [0, 1)
called φk+I+1 , in which a certain χk+I+1 can be Let us analyze separately the two terms of this last
defined. equation.
Our purpose is to explore the behavior of the By using the formula (20), we find that the first
function χk+I+1 in subintervals of φk+I+1 , thereby term is equivalent to
showing the emergence of fractal patterns. In gen-
eral, splitting φk+I+1 into n parts corresponds to k+1

introducing log2 n bits after xk in the set of bit T N (xi−1 xi xi+1 )2−(i+1)
strings xk+I+1 . For example, we can subdivide i=0
φk+I+1 into four regions just by adding two extra 1

bits — called xleft and xright — after xk in (18), and = T N (xk+I−1 x0 x1 )
2
creating the new set of bit strings xk+I+3
k−1

k+2 bits I+1 bits + T N (xi−1 xi xi+1 )2−(i+1)
i=1
.. ..
xk+I+3 = (x0 x1 . . . xk xleft xright . .xk+I ) 1
+ T N (xk−1 xk xleft )
k+I+3 bits 2k+1

1
= (x0 x1 . . . xk+I+2 ) (19) + T (xk xleft xright ) (22)
2k+2 N
The correspondence between xk+I+1 and xk+I+3
Since for any local rule N the value
can be obtained by comparing (18) and (19)
T N (xi−1 xi xi+1 ) is either 1 or −1, from Eq. (22)
 k+1 −(i+1) is a
 x0 = x0 it follows that i=0 T N (xi−1 xi xi+1 )2



 .. rational number, with

 .




xk = xk

 k+1
k+2


x
k+1 = xleft 0≤ T N
(xi−1 xi xi+1 )2−(i+1) ≤ 2−j < 1
(20) i=0 j=1
xk+2 = xright




 (23)
xk+3 = xk+1




 .. The second term of the formula (21) can be

 .
 written as
xk+I+2 = xk+I
k+I+2

The first of the four subregions corresponds to
T N (xi−1 xi xi+1 )2−(i+1)
the case (xleft xright ) = (00), the second subre-
i=k+2
gion to (xleft xright ) = (01), the third subregion to
(xleft xright ) = (10), and finally the fourth subre- I

gion to (xleft xright ) = (11). The expression for the = T N (xj+k+1 xj+k+2 xj+k+3 )2−(j+k+3)
time-1 characteristic function χk+I+3 (xk+I+3 ) can j=0
be found by using the formula (17) I
1
χk+I+3 (xk+I+3 ) = T N (xj+k+1 xj+k+2 xj+k+3 )2−(j+1)
2k+2
j=0
k+I+2

= T N (xi−1 xi xi+1 )2−(i+1) =
1 1
T (xleft xright xk+1 )
i=0 2k+2 2 N
1 In particular, in the first subregion χI+1

+ T N (xright xk+1 xk+2 ) . . . .
4 (xright .. ..xleft ) = χI+1 (0.. ..0) = χα , in the second
. . . .
I−1 subregion χI+1 (xright .. ..xleft ) = χI+1 (1.. ..0) =
. .
+ T N (xj+k−1 xj+k xj+k+1 )2−(j+1) χβ , in the third subregion χI+1 (xright .. ..xleft ) =
. .
j=2 χI+1 (0.. ..1) = χγ , and in the fourth subregion
. . . .
χI+1 (xright .. ..xleft ) = χI+1 (1.. ..1) = χδ . Since
1 k and I are arbitrary, it follows that the time-
+ T (x x x
k+I−1 k+I 0 )
2I+1 N 1 characteristic function of a generic rule N is
1 . . composed of the four fractal patterns χα , χβ ,
χI+1 (xright .. ..xleft ) (24) χγ and χδ (in this order) in any subinterval
2k+2
of φ ∈ [0, 1).
because the actual outcome of the term
(1/2I+1 )T N (xk+I−1 xk+I x0 ) is negligible for I 1,
and (xk+1 . . . xk+I ) are arbitrary bits, as stated As an example, let us consider rule 110 : Its
in (18). fractal patterns for the case xI = 0 are shown in
. .
Therefore, using (22) and (24), we can rewrite Fig. 15(a), where χα = χI+1 (0.. ..0) is plotted in
. .
the expression (21) as follows: red and χβ = χI+1 (1.. ..0) is in purple; the frac-
tal patterns for the case xI = 1 are in Fig. 15(b),
k+1 . .
where χγ = χI+1 (0.. ..1) is plotted in blue and
χk+I+3 = T N (xi−1 xi xi+1 )2−(i+1) . .
χδ = χI+1 (1.. ..1) is in cyan.
i=0 According to Theorem 3.1, these four patterns
vertical shift appear naturally in any subinterval of φ ∈ [0, 1). For
1 . .. instance, let us consider φ ∈ [0.25, 0.5), which cor-
+ χI+1 (xright .. .xleft ) (25) . .
2k+2 responds to the set of bit strings xI+2 = (01.. ..0);
the restriction of χ to φ ∈ [0.25, 0.5) is depicted in
scale factor
Fig. 15(a). Let us divide this interval into four parts
This means that in each subregion there is a by adding two additional bits: the first subregion
. . . .
scaled and shifted copy of χI+1 (xright .. ..xleft ). corresponds to the case xI+4 = (0100.. ..0) and
χ xI = 0 χ xI = 1
110 110
1 1
0.75 0.75
0.5 0.5
0.25 0.25
0 0
0 0.25 0.5 0.75 1
φ 0 0.25 0.5 0.75 1
φ
(a) (b)
Fig. 15. Time-1 characteristic function of rule 110 : (a) case xI = 0, (b) case xI = 1. The fractal pattern χα , χβ , χγ and χδ
are red, purple, blue, and cyan, respectively.
χ x I = 0 and 1 χ x I = 0 and 1
110 110
1 0.875
0.9375 0.84375
0.875 0.8125
0.8125 0.78125
0.75 0.75
0.25 0.3125 0.375 0.4375 0.5
φ 0.375 0.40625 0.4375 0.46875 0.5
φ
(a) (b)
Fig. 16. Fractality of χ1 : The four fractal patterns appear in two arbitrary subintervals of φ ∈ [0, 1).
110
0.25 ≤ φI+4 (xk+I+4 ) < 0.3125, the second sub- 1 1 1 1 γ

. . = + + + χ
region to xI+4 = (0101.. ..0) and 0.3125 ≤ 2 4 8 8
φI+4 (xI+4 ) < 0.375, the third subregion to xI+4 = 7 1
. . = + χγ
(0110.. ..0) and 0.375 ≤ φI+4 (xI+4 ) < 0.4375, and 8 8
. .
the fourth subregion to xI+4 = (0111.. ..0) and . .. 1 1
χI+4 (0111.. .0) T 110 (001) + T 110 (011)
0.4375 ≤ φI+4 (xI+4 ) < 0.5. 2 4
The analytical expressions for χI+4 (xI+4 ) in 1 1 . .
the four subregions can be derived from the for- + T 110 (111) + χI+1 (1.. ..1)
8 8
mulas (22) and (25), and the results are illustrated
1 1 1
in Fig. 16(a). In particular, = + + χδ
2 4 8
. .. 1 1
χI+4 (0100.. .0) = T 110 (001) + T 110 (010) 3 1
2 4 = + χδ
4 8
1 1 . .
+ T 110 (100) + χI+1 (0.. ..0) As expected, the time-1 characteristic function χI+4
8 8
in the four subregions is a scaled and shifted version
1 1 1 3 1 of χα , χβ , χγ and χδ , respectively. In Fig. 16(b) a
= + + χα = + χα
2 4 8 4 8 further subdivision of the horizontal axis 0.4375 ≤
. .. 1 1 φI+6 (xk+I+6 ) < 0.5 is depicted, and also in this
χI+4 (0101.. .0) = T 110 (001) + T 110 (010) case the four fractal patterns emerge naturally.
2 4
1 1 . .
+ T 110 (101) + χI+1 (1.. ..0) 3.2. Fractals in CA additive rules
8 8
1 1 1 1 7 1 Further analytical results can be given for nontrivial
= + + + χβ = + χβ additive rules — namely 60 , 90 , 105 and 150 —
2 4 8 8 8 8
which will be extensively analyzed in Sec. 6. These
. .. 1 1
χI+4 (0110.. .0) T 110 (001) + T 110 (011) rules can be either described in terms of Boolean
2 4 operations (see Sec. 6), or by using absolute values
1 1 . . and standard arithmetic operations. The relation-
+ T 110 (110) + χI+1 (0.. ..1)
8 8 ship between these two kinds of representations can
be derived straightforwardly; for example
60 : xn+1
i = xni−1 ⊕ xni → T 60 (xi−1 xi xi+1 ) = |xi−1 − xi | (26)
90 : xn+1
i = xni−1 ⊕ xni+1 → T 90 (xi−1 xi xi+1 ) = |xi−1 − xi+1 | (27)
105 : xn+1
i = xni−1 ⊕ xni ⊕ xni+1 → T 105 (xi−1 xi xi+1 ) = 1 − ||xi−1 + xi + xi+1 − 1| − 1| (28)
150 : xn+1
i = xni−1 ⊕ xni ⊕ xni+1 → T 150 (xi−1 xi xi+1 ) = ||xi−1 + xi + xi+1 − 1| − 1| (29)
In the following we consider individually these four

rules, finding for each of them an analytical formula patterns are presented in red, purple, blue and cyan,
for the lines delimiting the fractal patterns. respectively.
The function χ 60 can be bounded by using
the following formula, whose proof is given in
3.2.1. Rule 60
Appendix A
The time-1 characteristic function χ 60 can be I
I

obtained from Eq. (26): 1− xi−1 2−(i+1) + xi 2−(i+1) − 1

I i=0 i=0

−(i+1)
χ 60 = T 60 (xi−1 xi xi+1 )2 ≥ χ 60
i=0 I
I

I
−(i+1) −(i+1)
≥ xi−1 2 − xi 2
= |xi−1 − xi |2−(i+1) (30)
i=0 i=0
i=0
⇒ sup 60 ≥ χ 60 ≥ inf 60 (31)
χ 60 is plotted in Fig. 17(a) for xI = 0, and
in Fig. 18(a) for xI = 1. The four fractal where sup 60 (inf 60 ) is the supremum (infimum)
of χ 60 .
χ xI = 0 χ xI = 0
60 60
1 1
0.75 0.75
0.5 0.5
0.25 0.25
0 0
0 0.25 0.5 0.75 1
φ 0 0.25 0.5 0.75 1
φ
(a) (b)
Fig. 17. Time-1 characteristic function of rule 60 , case xI = 0: The fractal pattern χα is in red and the fractal pattern χβ
is in purple.
χ xI = 1 χ xI = 1
60 60
1 1
0.75 0.75
0.5 0.5
0.25 0.25
0 0
0 0.25 0.5 0.75 1
φ 0 0.25 0.5 0.75 1
φ
(a) (b)
Fig. 18. Time-1 characteristic function of rule 60 , case xI = 1: The fractal pattern χγ is in blue and the fractal pattern χδ
is in cyan.
I −(i+1)
Moreover, the term i=0 xi−1 2 appearing in (31) can be simplified as follows
I
I−1
I−1
−(i+1) −(j+2) 1 1 1 1
xi−1 2 = xj 2 = xI + xj 2−(j+1) xI + φ (32)
2 2 2 2
i=0 j=−1 j=0
In the last step we have discarded a term proportional to 2−(I+1) , which is extremely small for I 1.
The analytical expression for sup 60 and inf 60 can be found by combining (31) and (32)
 I
 I

 −(i+1) −(i+1) 1

 sup : 1 − xi−1 2 + x i 2 − 1 = 1 − |3φ + xI − 2|,
 60 2
i=0 i=0
60
 I

 I 1

inf 60 :
 xi−1 2−(i+1) − xi 2−(i+1) = |φ − xI |.
2
i=0 i=0
Therefore, the equations for the upper and lower bounds are

 3
(a): sup 60 = 1 − 2 φ − 1 ,
 φ ∈ (0, 1)
for xI = 0,

 1
(b): inf
60 = 2 φ, φ ∈ (0, 1)

 3 1
(c): sup 60 = 1 − 2 φ − 2 ,
 φ ∈ (0, 1)
for xI = 1,

 1 1
(d): inf
60 = 2 − 2 φ, φ ∈ (0, 1)
and the letters correspond to the lines in Figs. 17(b) and 18(b), respectively.
Furthermore, in Fig. 19 we show the emergence of the fractal behavior, since the four subpatterns
appear when the function χ 60 is restricted to arbitrary subintervals of the axis φ.
χ xI = 0 χ xI = 0
60 60
1 1
0.9375 0.984375
0.875 0.96875
0.8125 0.953125
φ φ
0.75 0.9375
0.5 0.5625 0.625 0.6875 0.75 0.625 0.640625 0.65625 0.671875 0.6875
(a) (b)
60
3.2.2. Rule 90
The time-1 characteristic function χ 90 can be obtained from Eq. (27):
I
I

−(i+1)
χ 90 = T 90 (xi−1 xi xi+1 )2 = |xi−1 − xi+1 |2−(i+1) (33)
i=0 i=0
It is plotted in Fig. 20(a) for xI = 0, and in Fig. 21(a) for xI = 1. As usual, the four fractal patterns are
depicted in different colors.
χ xI = 0 χ xI = 0
90 90
1 1
0.75 0.75
0.5 0.5
0.25 0.25
0 0
0 0.25 0.5 0.75 1
φ 0 0.25 0.5 0.75 1
φ
(a) (b)
is in purple.
χ xI = 1 χ xI = 1
90 90
1 1
0.75 0.75
0.5 0.5
0.25 0.25
0 0
0 0.25 0.5 0.75 1
φ 0 0.25 0.5 0.75 1
φ
(a) (b)
Fig. 21. Time-1 characteristic function of rule 90 , case xI = 1: The fractal pattern χγ is in blue and the fractal pattern χδ
is in cyan.

Similarly as for rule 60 , the time-1 characteris- The term Ii=0 xi+1 2−(i+1) appearing in (34) can
tic function for rule 90 can be bounded as follows: be simplified as follows:

I I I I+1

1− xi−1 2−(i+1) + xi+1 2−(i+1) − 1 xi+1 2−(i+1) = xj 2−j + x0 − x0 2φ − x0

i=0 i=0
i=0 j=1

I I
(35)

≥ χ 90 ≥ xi−1 2−(i+1) − xi+1 2−(i+1)
Here, we have discarded a term proportional to 2−I ,
i=0 i=0
which is extremely small for I 1.
⇒ sup 90 ≥ χ 90 ≥ inf 90 (34) The analytical expression for sup 90 and inf 90
can be found by combining (34), (32) and (35)
 I

 I

 1

 sup 90 : 1 − xi−1 2−(i+1) + xi+1 2−(i+1) − 1 = 1 − |5φ − 2x0 + xI − 2|,

 2
i=0 i=0
90 I

 I 1

 −(i+1) −(i+1)

 inf : x i−1 2 − xi+1 2 = |3φ − 2x0 − xI |.


90 2
i=0 i=0
Therefore, the equations for the upper and lower bounds when xI = 0 are

 5 1

(a): sup 90 = 1 − 2 φ − 1 ,
 φ∈ 0,
2
for (x0 , xI ) = (0, 0)

 3 1

(b): inf 90 = φ, φ∈ 0,
2 2

 5 1
(c): sup

 90 = 1 − 2 φ − 2 , φ∈
2
,1
for (x0 , xI ) = (1, 0)

 3 1

(d): inf 90 = φ − 1 ,
φ∈ ,1
2 2
where the letters correspond to the lines in Fig. 20(b), whereas when xI = 1 the equations for the upper
and lower bounds are

 5 1 1

(e): sup 90 = 1 − 2 φ − 2 ,
 φ∈ 0,
2
for (x0 , xI ) = (0, 1)

 3 1 1

(f): inf 90 = φ − , φ ∈ 0,
2 2 2

 5 3 1

(g): sup 90 = 1 − 2 φ − 2 ,
 φ∈
2
,1
for (x0 , xI ) = (1, 1)

 3 3 1

(h): inf 90 = − φ + , φ∈ ,1
2 2 2
where the letters correspond to the lines in Fig. 21(b).
The fractality of χ 90 is evident from Fig. 22, in which we show that the four subpatterns appear in
two arbitrary subintervals of the axis φ.
3.2.3. Rule 105

The time-1 characteristic function χ 105 can be obtained from Eq. (28):
I
I

−(i+1)
χ 105 = T 105 (xi−1 xi xi+1 )2 = (1 − ||xi−1 + xi + xi+1 − 1| − 1|)2−(i+1) (36)
i=0 i=0
It is shown in Fig. 23(a) for xI = 0, and in Fig. 24(a) for xI = 1. Also in this case, the four fractal patterns
are plotted in different colors.
As usual, the time-1 characteristic function can be bounded; namely,
sup 105 ≥ χ 105 ≥ inf 105 (37)
The analytical calculation of the upper and lower bounds for the four patterns is tedious due to
the complicated expression for χ 105 . Therefore, here we give the explicit formulas for sup 105 and
χ xI = 1 χ xI = 1
90 90
0.5 0.125
0.375 0.09375
0.25 0.0625
0.125 0.03125
0 0
0 0.125 0.25 0.375 0.5
φ 0.25 0.28125 0.3125 0.34375
φ
0.375
(a) (b)
90
χ xI = 0 χ xI = 0
105 105
1 1
0.75 0.75
0.5 0.5
0.25 0.25
0 0
0 0.25 0.5 0.75 1
φ 0 0.25 0.5 0.75 1
φ
(a) (b)
is in purple.
χ xI = 1 χ xI = 1
105 105
1 1
0.75 0.75
0.5 0.5
0.25 0.25
0 0
0 0.25 0.5 0.75 1 φ 0 0.25 0.5 0.75 1 φ
(a) (b)
Fig. 24. Time-1 characteristic function of rule 105 , case xI = 1: The fractal pattern χγ is in blue and the fractal pattern
χδ is in cyan.
inf 105 : When xI = 0 they are



 7 1

(a): sup 105 = − φ + 1, φ∈ 0,

 6 2




7 2
for (x0 , xI ) = (0, 0) (b): inf 105 = − φ + 1, φ∈ 0,

 2 7





 7 1 2 1
(c): inf 105 = 6 φ − 3 ,
 φ∈ ,
7 2


 7 1 6

(d): sup 105 = φ, φ∈ ,

 6 2 7





 7 6
(e): sup 105 = − 2 φ + 4, φ ∈ 7 , 1

for (x0 , xI ) = (1, 0)

 7 1 4

(f): inf 105 = − φ + 2, φ ∈ ,

 2 2 7





 7 2 4

(g): inf = φ − , φ ∈ , 1
105 6 3 7
where the letters correspond to the lines in Fig. 23(b); when xI = 1 the formulas are



7 1 3
(h): sup 105 = φ + ,
 φ ∈ 0,

 6 2 7





 7 5 3 1

(i): sup 105 = − 2 φ + 2 ,
 φ∈ ,
7 2
for (x0 , xI ) = (0, 1)

 7 1 1

(j): inf 105 = − φ + , φ ∈ 0,



 2 2 7





 7 1 1 1
(k): inf 105 = φ − , φ∈ ,
6 6 7 2


 7 1 1 5

(l): sup 105 = φ + , φ∈ ,

 6 6 2 7




7 7 5
for (x0 , xI ) = (1, 1) (m): sup 105 = − φ + , φ∈ ,1

 2 2 7





 7 7 1

(n): inf 105 = − φ + , φ∈ ,1
6 6 2
where the letters correspond to the lines in I

Fig. 24(b). χ 150 = T 150 (xi−1 xi xi+1 ) 2−(i+1)
As we did for 60 and 90 , we show that i=0
also the time-1 characteristic function for 105 is I
a fractal, showing in Fig. 25 that the four subpat- = ||xi−1 + xi + xi+1 − 1| − 1| 2−(i+1)
terns appear in two arbitrary subintervals of the i=0
(38)
axis φ.
It is plotted in Fig. 26(a) for xI = 0, and in
Fig. 27(a) for xI = 1. We can easily draw analogies
3.2.4. Rule 150 between 150 and 105 because of the alternating
The time-1 characteristic function χ 150 can be symmetry duality between the two rules. Follow-
obtained from Eq. (29): ing the analogy with Eq. (37), we found that χ 150
χ xI = 1 χ xI = 1
105 105
1 0.625
0.875 0.59375
0.75 0.5625
0.625 0.53125
0.5 0.5
0 0.125 0.25 0.375 0.5 φ 0.25 0.28125 0.3125 0.34375 0.375φ
(a) (b)
105
χ xI = 0 χ xI = 0
150 150
1 1
0.75 0.75
0.5 0.5
0.25 0.25
0 0
0 0.25 0.5 0.75 1 φ 0 0.25 0.5 0.75 1 φ
(a) (b)
is in purple.
χ xI = 1 χ xI = 1
150 150
1 1
0.75 0.75
0.5 0.5
0.25 0.25
0 0
0 0.25 0.5 0.75 1
φ 0 0.25 0.5 0.75 1
φ
(a) (b)
γ
Fig. 27. Time-1 characteristic function of rule 150 , case xI = 1: The fractal pattern χ is in blue and the fractal pattern
χδ is in cyan.
is bounded; namely,
sup 150 ≥ χ 150 ≥ inf 150 (39)
The explicit formulas for sup 150 and inf 150 for the case xI = 0 are


 7 2

(a): sup 150 = φ, φ∈ 0,




2 7


 7 4 2 1
for (x0 , xI ) = (0, 0) (b): sup 150 =− φ+ , φ∈ ,

 6 3 7 2





 7 1

 φ, φ ∈ 0,
(c): inf 150 =
6 2


(d): 7 1 4

 sup 150 = φ − 1,
φ∈ ,

 2 2 7





 7 5 4

 sup 150 = − φ + , φ ∈ ,1
(e): 6 3 7
for (x0 , xI ) = (1, 0)




7
inf 150 = − φ + 1, φ∈
1
,1

(f):

 6 2





 7 6
(g): inf 150 = φ − 3, φ∈ ,1
2 7
where the letters correspond to the lines in Fig. 26(b); when xI = 1 the formulas are

 7 1 1

(h): sup 150 = φ + , φ ∈ 0,

 2 2 7





 7 7 1 1

(i): sup 150 = − 6 φ + 6 ,
 φ∈ ,
7 2
for (x0 , xI ) = (0, 1)

 7 1 3

(j): inf 150 = − φ + , φ ∈ 0,




6 2 7



 7 3 3 1

(k): inf 150 = φ − , φ∈ ,
2 2 7 2



 7 1 1

(l): sup 150 = φ − , φ∈ ,1

 6 6 2




7 5 1 5
for (x0 , xI ) = (1, 1) (m): inf 150 = − φ + , φ∈ ,

 6 6 2 7





 7 5 5

(n): inf 150 = φ − , φ∈ ,1
 2 2 7
and the letters correspond to the lines in Fig. 27(b).

Finally, also for 150 the four subpatterns 3.3. From the time-1 characteristic
appear in two arbitrary subintervals of the axis φ, function to the rule number
as shown in Fig. 28.
The analytical characterization of the fractality of
Note that the time-1 characteristic function for
χ1 for a generic rule N makes it possible to
150 (and consequently, the four fractal patterns N
and their bounds) can be obtained from that of obtain the rule number directly from the time-
105 by means of a reflection through the axis 1 characteristic function diagram. Here we con-
χ 105 = 1/2. sider strings composed by I + 1 bits, and their
decimal representation φI+1 defined as in (16). Let
χ xI = 0 χ xI = 0
150 150
0.5 0.125
0.375 0.09375
0.25 0.0625
0.125 0.03125
0 0
0 0.125 0.25 0.375 0.5
φ 0 0.03125 0.0625 0.09375 0.125
φ
(a) (b)
150
us differentiate between xI = 0 and xI = 1: For the . .

and φ1,IV = φI+1 (11.. ..1) corresponding to 3/4 <
first case, we divide the axis φ into four intervals:
φ1,IV ≤ 1.
φ0,I = φI+1 (00 . . . 0) corresponding to 0 ≤ φ0,I <
The expression for the time-1 characteristic
1/4, φ0,II = φI+1 (01 . . . 0) corresponding to 1/4 ≤
function in the first interval can be found through
φ0,II < 1/2, φ0,III = φI+1 (10 . . . 0) corresponding to
Eq. (25)
1/2 ≤ φ0,III < 3/4, and φ0,IV = φI+1 (11 . . . 0) cor-
responding to 3/4 < φ0,IV < 1. In each of these .
χ1,I = χI (00..
..
.1)
intervals, the time-1 characteristic function for a
generic rule N and its representation through frac- 1 . ..
(T N (100) + χI−1 (0.. .0))
tal patterns can be defined by means of the formula 2
(25). For example, in the first subregion 1 1
= T N (100) + χα
1 1 2 2
χ0,I = χI+1 (00 . . . 0) = T N (000) + χα
2 2 therefore
therefore 1
χ1,I > ⇔ T N (100) = 1
1 2
χ0,I > ⇔ T N (000) = 1
2 In this case the value of T N (100) is completely
This means that the value of T N (000) is com- determined by the position of the characteristic
pletely determined by the position of the charac- function in the first quarter of the diagram for
teristic function in the first quarter of the diagram xI = 1: If the points of χ1,I are below 1/2 then
for xI = 0: If the function χ0,I is below 1/2 then T N (100) = 0; if the points of χ1,I are above 1/2
T N (000) = 0; if it is above 1/2 then T N (000) = 1. then T N (100) = 1.
Following an analogous procedure we find that We can extend the same result to the other
. .. three subregions obtaining
χ0,II = χI (01.. .0)
. ..
1 1 1 χ1,II = χI (01.. .1)
= T N (001) + χβ → χ0,II >
2 2 2 1 1 1
T N (101) + χγ → χ1,II >
⇔ T N (001) = 1 2 2 2
⇔ T N (101) = 1
. ..
χ0,III = χI (10.. .0)
. ..
1 1 1 χ1,III = χI (10.. .1)
T N (010) + χγ → χ0,III >
2 2 2 1 1 1
= T N (110) + χγ → χ1,III >
⇔ T N (010) = 1 2 2 2
⇔ T N (110) = 1
. ..
χ 0,IV
= χI (11.. .0)
. ..
1 1 1 χ1,IV = χI (11.. .1)
T N (011) + χδ → χ0,IV >
2 2 2 1 1 1
= T N (111) + χδ → χ1,IV >
⇔ T N (011) = 1 2 2 2
⇔ T N (111) = 1
and then the values for T N (001), T N (010) and
T N (011) depend on the position of the time-1 char- and then the values for T N (101), T N (110) and
acteristic function in the second, third and fourth T N (111) depend on the position of the time-1 char-
regions, respectively. acteristic function in the second, third and fourth
As for the case xI = 1, we can also divide the regions, respectively.
. .
axis φ into four intervals: φ1,I = φI+1 (00.. ..1) cor- In conclusion, the number of the local rule
. .
responding to 0 < φ1,I < 1/4, φ1,II = φI+1 (01.. ..1) can be found straightforwardly by visual inspection
corresponding to 1/4 < φ1,II < 1/2, φ1,III = of the time-1 characteristic function diagrams, as
. .
φI+1 (10.. ..1) corresponding to 1/2 < φ1,III < 3/4, summarized in Fig. 29. Observe that this procedure
χ xI = 0 χ xI = 1
1 1
1 1
0 0
0 0
0 1 φ 0 1 φ
{
000 001 010 011 100 101 110 111
(a) (b)
Fig. 29. Correspondence between time-1 characteristic function and local rule number.
generalizes the concept of Stratification of char- the same; in fact, this rule has only three distinct
acteristic functions illustrated in [Chua et al., patterns.
2005b]. In order to identify the number of different frac-
tal patterns for each local rule, we divide χα , χβ ,
χγ and χδ for a generic rule N into two subpat-
3.4. Number of fractal patterns terns, indicating the first one with the superscript
I and the second one with the superscript II: For
Often a rule does not exhibit four distinct fractal
example, χα is composed by χα,I and χα,II , χβ by
patterns because some of them can coincide. For
χβ,I and χβ,II etc.
example, in rule 110 , whose time-1 characteristic
The analytical expressions for the subdivisions
function is drawn in Fig. 15, χα and χγ are exactly
can be found by using Eq. (25):


 α,I .. .. 1 1 . .. 1 α
χ = χI+1 (00.
 .0) = T N (000) + χI (0..
2 2
.0) = (T N (000) + χ )
2
χα : (40)



χα,II = χI+1 (01... .. 1 1 . .. 1 β
.0) = T N (001) + χI (1.. .0) = (T N (001) + χ )
2 2 2


 β,I .. .. 1 1 . .. 1 γ
χ = χI+1 (10.
 .0) = T N (010) + χI (0..
2 2
.1) = (T N (010) + χ )
2
χβ : (41)



χβ,II = χI+1 (11... .. 1 1 . .. 1 δ
.0) = T N (011) + χI (1.. .1) = (T N (011) + χ )
2 2 2


 γ,I .. .. 1 1 . .. 1 α
χ = χI+1 (00.
 .1) = T N (100) + χI (0..
2 2
.0) = (T N (100) + χ )
2
χγ : (42)



χγ,II = χI+1 (01... .. 1 1 . .. 1 β
.1) = T N (101) + χI (1.. .0) = (T N (101) + χ )
2 2 2


 . .. 1 1 . . 1

χδ,I = χI+1 (10.. γ
.1) = T N (110) + χI (0.. ..1) = (T N (110) + χ )
2 2 2
χδ : (43)

 . . 1 1 . . 1

χ
δ,II = χI+1 (11.. ..1) = T N (111) + χI (1.. ..1) = (T N (111) + χδ )
2 2 2
Therefore, if T N (000) = T N (100) and T N (000) = T N (100) = 1 and T N (001) =

T N (001) = T N (101) it follows that χα = χγ . T N (101) = 0, for which χα = χγ − (1/2).
This kind of representation allows us to find Thanks to this procedure, it is possible to clas-
easily all the cases in which two (or more pat- sify all the rules according to the number of fractal
terns) coincide. For instance, from the formu- patterns they exhibit. Representing each rule via an
las (40) and (42) we can see that χα = χγ 8-bit code N ↔ (β7 β6 β5 β4 β3 β2 β1 β0 ), βj ∈ {0, 1},
when T N (000) = T N (100) and T N (001) = namely,
T N (101). Moreover, one of the two patterns can N = β7 · 27 + β6 · 26 + β5 · 25 + β4 · 24 + β3 · 23
be a shifted copy of another one: This happens
+ β2 · 22 + β1 · 21 + β0 · 20 ,
when T N (000) = T N (100) = 0 and T N (001) =
T N (101) = 1, for which χγ = χα − (1/2), or when where β0 ≡ T N (000) . . . β7 ≡ T N (111), it is pos-
sible to prove that rule N contains:

 (β β β β ) = (0011)
 7 6 3 2


 (β7 β6 β3 β2 ) = (1100)



(β5 β4 β1 β0 ) = (0011)
4 patterns ⇔

 (β5 β4 β1 β0 ) = (1100)



 (β3 β2 ) = (β7 β6 )


(β β ) = (β β )
1 0 5 4
  
 (β β β β ) = (0011) (β β β β ) = (0011)  (β β β β ) = (1100)
 5 4 1 0
  7 6 3 2
  7 6 3 2

(β5 β4 β1 β0 ) = (1100) (β5 β4 β1 β0 ) = (0011) (β5 β4 β1 β0 ) = (0011)
3 patterns ⇔ or or

 (β3 β2 ) = (β7 β6 ) 
(β5 β4 β1 β0 ) = (1100) 
 (β5 β4 β1 β0 ) = (1100)

(β β ) = (β β ) 
(β β ) = (β β ) 
(β β ) = (β β )
1 0 5 4 1 0 5 4 1 0 5 4
  
(β7 β6 β3 β2 ) = (0011)  (β7 β6 β3 β2 ) = (0011)  (β7 β6 β3 β2 ) = (0011)

 
 

(β7 β6 β3 β2 ) = (1100) (β7 β6 β3 β2 ) = (1100) (β7 β6 β3 β2 ) = (1100)
or or or

(β3 β2 ) = (β7 β6 ) 
 (β5 β4 β1 β0 ) = (0011) 
 (β5 β4 β1 β0 ) = (1100)

(β β ) = (β β ) 
(β β ) = (β β ) 
(β β ) = (β β )
1 0 5 4 3 2 7 6 3 2 7 6
  

 (β7 β6 β5 β4 ) = (0011) 
β6 = β7 
(β5 β4 ) = (β7 β6 )

 
 

  
(β7 β6 β5 β4 ) = (1100)
 (β5 β4 ) = (β7 β6 )
 β4 = β5

2 patterns ⇔ (β5 β4 ) = (β7 β6 ) or β3 = |1 − β7 | or (β3 β2 ) = (β7 β6 )

 
 

(β3 β2 ) = (β7 β6 )
 β2 = β3
 β1 = |1 − β5 |


 
 

(β1 β0 ) = (β5 β4 ) (β1 β0 ) = (β5 β4 ) β1 = β0
  
(β7 β6 ) = (00)  β7 = β6 β7 = β6

 
 

β4 = β5 (β5 β4 ) = |1 − (β7 β6 )| (β5 β4 ) = (β7 β6 )
or or or

(β3 β2 ) = (00) 
 (β3 β2 ) = (β5 β4 ) 
(β3 β2 ) = |1 − (β7 β6 )|

(β β ) = |1 − (β β )| 
(β β ) = (β β ) 
(β β ) = (β β )
1 0 5 4 1 0 7 6 1 0 7 6
 
  β6 = β7  β6 = β7
 
(β5 β4 ) = (β7 β6 )
 
(β5 β4 ) = (β7 β6 ) 
 (β5 β4 ) = |1 − (β7 β6 )|
1 pattern ⇔ (β3 β2 ) = (β7 β6 ) or or

(β β ) = (β β ) 
 (β3 β2 ) = |1 − (β7 β6 )| 
 (β3 β2 ) = (β7 β6 )
1 0 3 2 
(β β ) = (β β ) 
 (β β ) = |1 − (β β )|
1 0 3 2 3 2 1 0
The 39 globally independent rules (equivalent to Table 22. List of eight globally independent local rules
100 local rules) with four distinct fractal patterns with two distinct one fractal patterns.
are listed in Table 20; the 36 globally independent
rules (equivalent to 120 local rules) with three dis-
tinct fractal patterns are listed in Table 21; the 3 12 29 34
eight globally independent rules (listed equivalent
to 28 local rules) with two distinct fractal patterns
are listed in Table 22; the five globally independent
46 60 136 184
rules (listed equivalent to eight local rules) with
only one fractal pattern are listed in Table 23. Note Table 23. List of five globally independent local rules with
that this last class coincides with the trivial additive only one fractal pattern.
rules.
0 15 51 170 204
Table 20. List of 39 globally independent local rules with
four distinct fractal patterns.
5 6 9 10 22 23 4. New Results about Isles of Eden

24 26 27 36 37 40 The notion of “Isle of Eden” was first introduced in
[Chua et al., 2007a], and it was clear from the begin-
41 43 54 57 58 73 ning that it is of crucial importance in the analysis
74 77 78 90 94 104 of one-dimensional Cellular Automata. In this sec-
tion we give some new definitions that allow us to
105 108 122 126 130 134 classify the local rules according to the presence of
Isles of Eden.
142 146 150 156 160 164 Moreover, we also prove that 45 and 154 have
Isles of Eden if, and only if, L is odd.13
172 178 232 This result is not totally new, since it was pre-
sented in [Chua et al., 2007b]. However, the proof
was based on a graph-theoretical tool called Isles
Table 21. List of 36 globally independent local rules with of Eden digraph, which is powerful but complex,
three distinct fractal patterns.
because it requires the knowledge of further con-
cepts like the De Bruijn graphs. Here we give an
1 2 4 7 8 11 alternate proof for this theorem, based exclusively
13 14 18 19 25 28 on induction, and then simpler than the one known
so far.
30 32 33 35 38 42 Finally, we collect all the rules with no Isle of
Eden for any length L, giving a rigorous proof of
44 45 50 56 62 72 this property for all of them. This result is remark-
able, because for the first time we find all the rules
76 106 110 128 132 138 that have no everywhere invertible orbit in the state
140 152 154 162 168 200 transition graph.
13
Obviously, this result holds also for the local rules that are global equivalent to 45 and 154 , which are 75 , 89 , 101
and 210 , 180 , 166 , respectively.
4.1. Definitions and basic lemmas Proof. If the local rule N is left-permutive, then
The following definitions allows us to make a dis- At+1 t
n−1 defines xn−2 ; if N is right-permutive then
tinction among rules based on the presence of Isles At+1
n defines xt0 . In both cases, the same procedure
of Eden. These names are chosen via an analogy can be used to determine, step by step, all remain-
with the harmonic oscillator. ing unknown bits xti .
Definition 4.1. Strictly-Dissipative local rules. A Now, we are ready to enunciate the following the-
local rule N is said to be strictly-Dissipative iff orem, equivalent to Theorem 4.1 in [Chua et al.,
it has no Isle of Eden for any L. 2007b], giving an alternate proof based exclusively
on mathematical induction.
Definition 4.2. Conservative local rules. A local
rule N is said to be Conservative or non- Theorem 4.1. Every orbit of local rules 45 and
Dissipative iff every orbit is an Isle of Eden. 154 is an Isle of Eden, if, and only if, L is an odd
Definition 4.3. Semi-Dissipative Local Rules. A integer.
local rule N is said to be Semi-Dissipative iff it
Proof. Necessity: If L is even then all orbits of 45
has at least one Isle of Eden for some length L.
or 154 are not Isles of Eden.
Let us recall next two basic lemmas, whose formal From Lemma 4.2, it follows that we can prove
proofs can be found in [Chua et al., 2007b]. this statement by showing that there are at least
two different bit strings having the same output.
Lemma 4.1. A bit string x = (x0 x1 . . . xL−1 ) is a We can find a counterexample intuitively, since
period-n Isle of Eden of local rule N if, and only any string containing k times the pattern (01) has
if, x has a unique preimage under T n . the same output as a string containing k times the
N
pattern (00): In fact, in both cases the output is a
Lemma 4.2. A local rule N is conservative if, and string containing k times the pattern (11), because
only if, every bit string x = (x0 x1 . . . xL−1 ) has a T 45 (010) = T 45 (101) = T 45 (000) = 1.
unique preimage under T N , for any L ∈ N. However, the same result can be used through
an inductive procedure, introducing the notation
(x)L to indicate that a string x is composed by L
4.2. Alternate proof that rules 45
elements, (x)L = (x0 x1 . . . xL−1 ).
and 154 are conservative for Let us consider the two following strings with
odd lengths length L = 2k:
From Table 11, we can see that both 45 and 154 (xα)tL = (xα0 xα1 . . . xαL−1 ), with
belong to the class of permutive rules. We recall
that for left-permutive rules α 0, if i = 2n − 1
xi =
T N ( xti−1 xti xti+1 ) = xt+1 1, if i = 2n
i
and
⇒ T N ( x ti−1 xti xti+1 ) = x t+1
i
(xβ )tL = (xβ0 xβ1 . . . xβL ), with xβi = 0,
and for right-permutive rules ∀ i ∈ {0, L − 1}.
T N ( xti−1 xti xti+1 ) = xt+1
i For k = 1, L = 2 we have (xα)t2 = (01) and
⇒ T N ( xti−1 xtix ti+1 ) = x t+1 (xβ )t2 = (00), and (xα)t+12 = (xβ )t+1
2 = (11).
i
Let us assume that for L = 2n we have
It is easy to notice that for a permutive rule, given (xα)t+1 β t+1 α t
2n = (x )2n with (x )2n = (0101 . . . 01) and
xti , xti+1 , xt+1
i and the local rule N , we can uni- (xβ )t2n = (0000 . . . 00) (induction hypothesis), and
vocally identify xti−1 . This result was generalized let us analyze the case L = 2(n + 1) = 2n + 2. The
in [Wuensche et al., 1992] through the following strings (xα)t2n+2 and (xβ )t2n+2 are
lemma.
(xα)t2n+2 = (0101 . . . 01) = ((xα)t2n 0 1)
Lemma 4.3. If N is permutive, then knowing xt+1 and
and any two adjacent bits xti and xti+1 of xt , the bit
string xt is unambiguously determined. (xβ )t2n+2 = (0000 . . . 00) = ((xβ )t2n 0 0)
Noticing that T 45 (010) = T 45 (101) =

Note that this method cannot be used to find
T 45 (000) = 1, it follows that (xα)t+1
2n+2 = Isles of Eden, but only to confirm that all orbits
(xβ )t+1
2n+2 .
are actually Isles of Eden. This result about 45
The proof for rule 154 is very similar, except and 154 is similar to that obtained in [Chua et al.,
for the fact that T 154 (010) = T 154 (101) = 2007a] for 105 and 150 , which are conservative for
any L that is not divisible by 3. There are also some
T 154 (111) = 1, and where the string xβ should
rules that are conservative tout court, like 15 , 170 ,
be defined as follows 204 and 51 (and their global equivalent rules).
(xβ )L = (xβ0 xβ1 . . . xβL ), with xβi = 1, This follows directly from the nature of these rules,
∀ i ∈ {0, L − 1}. since they consist in a mere shift 15 and 170

or in transformation

depending only on the central
Sufficiency: When L is odd, all orbits of rules 45 pixel 204 and 51 .
and 154 are Isles of Eden.
According to Lemma 4.2, it is sufficient to prove 4.3. There are exactly 28
that in this case there is no Garden of Eden. strictly-dissipative local rules
We apply the mathematical induction as fol-
lows: Let L = 2n + 1 be the length of the string: While some rules — like 45 and 154 — are conser-
vative, others – like 60 [Chua et al., 2007b] and 90
Our assumption holds for n = 1 ⇒ L = 3; we [Chua et al., 2007a] — are strictly-dissipative, which
assume it is true for a generic n = k ⇒ L = 2k + 1 means that they have no Isle of Eden. In addition
(induction hypothesis), and then, we will confirm to the already mentioned 60 and 90 , we found
that it is true for n = k + 1 ⇒ L = 2k + 3 too. that all the centrally-symmetric (also called perfect-
Then, for all L = 2k + 1 (xα)t2k+1 = (xα0 xα1 . . . complementary) local rules are strictly-dissipative
(see Table 24) as well as rules 8 , 46 , and 78 and
xα2k ) and (xβ )t2k+1 = (xβ0 xβ1 . . . xβ2k ) such that
their equivalent rules (see Table 25). These results
(xα)t+1 β t+1 α t
2k+1 = (x )2k+1 ⇔ (x )2k+1 = (x )2k+1 .
β t
are proved in Theorems 4.2 and 4.3, respectively,
Now, we analyze the case L = 2k +3, supposing and all the 28 strictly-dissipative rules (nine topo-
that there is a Garden of Eden. This implies that logically independent) are listed in Table 26.
there are two different strings with the same image:
(xα)t2k+3 = (xα0 xα1 . . . xα2k+2 ) and (xβ )t2k+3 = Theorem 4.2. All centrally-symmetric local rules
(xβ0 xβ1 . . . xβ2k+2 ), with (xα)t2k+3 = (xβ )t2k+3 and are strictly-dissipative.
(xα)t+1 β t+1
2k+3 = (x )2k+3 . Proof. By definition, any triplet of a centrally-
The two strings can be decomposed into
symmetric rule produces the same output as its
(x )t2k+3
α = ((xα)t2l+1 xα2l+1 . . . xα2k+2 ) and
complement; therefore, any bit string xt has the
(xβ )t2k+3 = ((xβ )t2l+1 xβ2l+1 . . . xβ2k+2 ). If (xα)t2l+1 same output as its complement x t . Then, the proof
and (xα)t2l+1 have periodic boundaries, that is follows from Lemma 4.1.
xα2l = xα2k+2 , xα0 = xα2l+1 and xβ2l = xβ2k+2 ,
xβ0 = xβ2l+1 , then according to the induction For example, we can show how this theorem applies
hypotheses, (xα)t2l+1 = (xβ )t2l+1 . But for Lemma to rule 24 , which is centrally-symmetric. Its firing
patterns are and , and they are the com-
4.3, since we know the string (xα)t+1 2k+3 and two plement of the other; as for the quenching patterns,
adjacent pixels of (xα)t2k+3 , the remaining pix- they are also complement pairs: and ,
els xα2l+1 . . . xα2k+2 are fixed; the same happens and , and . Therefore, any
with (xβ )2k+3 , xβ2l+1 . . . xβ2k+2 . Since (xα)t+1 2k+3 = bit string has the same output as its complement,
(xβ )t+1
2k+3 and (x α)t
2k+1 = (xβ )t
2k+1 , it follows that and hence rule 24 has no Isles of Eden.
α β α β
j = l . . . k : x2j+1 = x2j+1 , x2j+2 = x2j+2 . But this Theorem 4.3. Rules 8 , 46 and 78 are the
means that (xα)t2k+3 = (xβ )t2k+3 , contrary to the only non centrally-symmetric rules to be strictly-
hypothesis.14 dissipative.
14
To be precise, we can always choose a (xα )t2k+1 according to the previous conditions when L > 17; for the values of L ≤ 17
we checked the theorem experimentally.
Table 24. Sixteen centrally-symmetric strictly-dissipative local rules.
N T †[ N ] T [N ] T *[ N ]
0 2 3 255 2 3
6 7 6 7
0 1 0 1
4 5 4 5
24 2 3 66 2 3 231 2 3 189 2 3
6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5
36 2 3 219 2 3
6 7 6 7
0 1 0 1
4 5 4 5
60 2 3 102 2 3 195 2 3 153 2 3
6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5
90 2 3 165 2 3
6 7 6 7
0 1 0 1
4 5 4 5
126 2 3 129 2 3
6 7 6 7
0 1 0 1
4 5 4 5
Table 25. Twelve non-centrally-symmetric strictly-dissipative local rules.
N T †[ N ] T [N ] T *[ N ]
8 2 3 64 2 3 239 2 3 253 2 3
6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5
46 2 3 116 2 3 139 2 3 209 2 3
6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5
78 2 3 92 2 3 141 2 3 197 2 3
6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5
Proof. We found, through computer simulations,

at least one Isle of Eden for each non centrally- Lemma 4.5. Rule 46 has no Isle of Eden.
symmetric rule, as shown in Table 15. The only
exceptions are rules 8 , 46 , 78 , for which we pro- Proof. We recall that the firing patterns of rule
vide an analytical proof of their strict dissipativity 46 are , , , and . First of all,
in the next three lemmas, respectively. we notice that strings with only either red or blue
pixels have the same successor, which is a string
Lemma 4.4. Rule 8 has no Isle of Eden. containing only blue pixels, because T 46 (000) =
T 46 (111) = 0. From Lemma 4.1, it follows that a
Proof. We recall that the only firing pattern of rule necessary condition for a string to belong to an Isle
8 is . Then, from Fig. 30 we see that all the of Eden contains the pattern .
strings containing more than one consecutive red Nevertheless, strings including the pattern
pixel — pattern — are Gardens of Eden. More- are Gardens of Eden, as proved in Fig. 31,
over, strings belonging to an Isle of Eden cannot and this implies that any string belonging to an
contain an isolated red pixel — pattern — Isle of Eden has to include at least two consecutive
because, for the uniqueness of the firing pattern, red pixels. However, not even the pattern
their predecessor would necessarily have the pattern can be part of an Isle of Eden, because it has the
. same successor as a different string, as shown in
Consequently, the only possible Isle of Eden Fig. 32. Therefore, the last case to analyze is the
would include exclusively the pattern ; case whose predecessors either give rise to
but since T 8 (000) = T 8 (111) = 0, the contradictory conditions or are Gardens of Eden, as
string (00 . . . 0) has two predecessors, which are detailed is Fig. 33.
(00 . . . 0) and (11 . . . 1), and this would contradict
Lemma 4.1. Lemma 4.6. Rule 78 has no Isles of Eden.
Table 26. Twenty-eight no Isle of Eden rules.
0 2 3 8 2 3 24 2 3 36 2 3
6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5
46 2 3 60 2 3 64 2 3 66 2 3
6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5
78 2 3 90 2 3 92 2 3 102 2 3
6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5
116 2 3 126 2 3 129 2 3 139 2 3
6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5
141 2 3 153 2 3 165 2 3 189 2 3
6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5
195 2 3 197 2 3 209 2 3 219 2 3
6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5
231 2 3 239 2 3 253 2 3 255 2 3
6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5
We start proving that strings containing more

than three consecutive red pixels are Gardens of
Eden (see Fig. 34).
The following step consists in showing that
strings containing exactly three adjacent red
pixels — pattern — do not belong to an Isle
of Eden because there is always another string with
the same successor, as evident from Fig. 35; a simi-
lar situation happens for strings containing exactly
two adjacent red pixels because their predecessor is
not unique, as implied by Fig. 36.
Fig. 30. Rule 8 : Strings containing the pattern are Finally, we need to see what happens for strings
Gardens of Eden, because their predecessor would give rise including an isolated red pixel, which is the pattern
to contradictory conditions on the pixel .
. In order to perform a thorough analysis,
we divide this case into four subcases, according
Proof. We recall that the firing patterns of 78 are
to what the neighbors for the pattern are:
, , , and . the case , analyzed in Fig. 37; the case
Fig. 31. Rule 46 : Strings containing the pattern are Gardens of Eden, because all the possible predecessors would
give rise to contradictory conditions on the pixel .
Fig. 32. Rule 46 : The pattern has the same successor as a different string, hence it cannot belong to an Isle of Eden.
Fig. 33. Rule 46 : Strings with more than two consecutive red pixels (pattern ) cannot belong to an Isle of Eden
because their predecessor either gives rise to contradictory conditions (cases (a)–(d)), or contains an isolated red pixel (cases
(e) and (f)) and then it is a Garden of Eden (see Fig. 31).
Fig. 34. Rule 78 : Strings containing more than three consecutive red pixels are Gardens of Eden, because all of their possible
predecessors would give rise to contradictory conditions.
Fig. 35. Rule 78 : Strings containing exactly three consecutive red pixels cannot belong to an Isle of Eden, because it is
always possible to find another bit string with the same successor.
Fig. 36. Rule 78 : Strings containing exactly two consecutive red pixels have at least two different predecessors, and hence
they cannot belong to an Isle of Eden.
Fig. 37. Rule 78 : The pattern cannot belong to an Isle of Eden because its predecessor either gives rise to con-
tradictory conditions (cases (a)–(c)), or contains more than three consecutive red pixels (case (d)) and then it is a Garden of
Eden (see Fig. 34).
the case , analyzed in Fig. 37; the case The last possibility would be having an Isle
, analyzed in Fig. 38; the case , of Eden composed by a single string, which can
analyzed in Fig. 39; and finally the case , be either (00 . . . 0) or (11 . . . 1); however, this pos-
analyzed in Fig. 40. As shown in the respective fig- sibility is excluded by the fact that T 78 (000) =
ures, none of these patterns can be in a string that T 78 (111) = 0, as it happened for 8 and 46 .
is part of an Isle of Eden.
Fig. 38. Rule 78 : The pattern cannot belong to an Isle of Eden because its predecessor either gives rise to con-
tradictory conditions (cases (a)–(c)), or contains more than three consecutive red pixels (case (d)) and then it is a Garden of
Eden (see Fig. 34).
4.4. Isles of Eden for rules of principle, rules of group 1 have a robust period-1
group 1 global attractor, which means that most of the 2L
bit strings belong to a unique basin of attraction.
We mentioned in Sec. 1 that our classification
Here we want to show that, at least for some
of local rules into six groups is based mainly on
rules of group 1, the only possible attractor is the
the properties of attractors (number, period etc.),
robust one, and all strings not belonging to it form
although it is also supported by other features,
Isles of Eden.
like quasi-ergodicity. According to our classification
Let us take into consideration rule 40 , whose
firing patterns are and . Rule 40 has a
robust attractor in 0; in other words, for any L most
of bit strings will fall into the basin of attraction of
L elements

x = ( 00 . . . 0 ). Which strings do not belong to this
attractor, and what happens with them? In [Ohi,
2007] the following theorem is proved.15
Fig. 39. Rule 78 : The pattern cannot belong
Theorem 4.4. For rule 40 , x = (00 . . . 0) is a global
to an Isle of Eden because it has at least two different pre-
decessors. attractor, except for the strings with at most one con-
secutive 0 and at most two consecutive 1 s, which
form a Bernoulli Isle of Eden with σ = 1 and τ = 1.
The following proof is far simpler than the one
in [Ohi, 2007], mainly because the original paper
also includes other results requiring a more com-
plex notation and treatment.
Proof. The first part of the proof consists in show-

ing that any string containing more than one con-
secutive 0 or more than two consecutive 1’s is in
the basin of attractor of x = (00 . . . 0). As for the
first case, we consider, without loss of generality, the
string (. . . 100 . . .), which is the “border” between a
“1” and the run of 0’s; from Fig. 41, it is evident
that the “border” tends to move towards left, and
then all the 1’s will be substituted, step by step,
Fig. 40. Rule 78 : The pattern cannot belong to
an Isle of Eden because, after a finite number of steps, it by 0’s. As for the second case, we consider, without
evolves into the pattern which cannot belong to an loss of generality, the string (. . . 0111 . . .), which is
Isle of Eden, as shown in Fig. 39. the “border” between a “0” and the run of 1’s; from
15
This theorem is a summary of the results presented in several lemmas and theorems in [Ohi, 2007] using a different
terminology and notation.
Theorem 4.5. For rule 168 , x = (00 . . . 0) is a

global attractor, except for the strings with more
than one consecutive 0, which form a Bernoulli Isle
of Eden with σ = 1 and τ = 1.16
Proof. The first part of the proof is very similar

to that of Theorem 4.4, but we need to refer only
Fig. 41. Rule 40 : All strings containing the pattern to Fig. 41, discarding the results in Fig. 42. Also
are in the basin of attraction of · · · . the second part is very similar, but in this case
the hypothesis excludes the triplets (000), (001)
and (100), and then we need to analyze only the
triplets (010), (011), (101), (110) and (111), for
which T 168 (010) = 0 = T 170 (010), T 168 (011) =
1 = T 170 (011), T 168 (101) = 1 = T 170 (101),
T 168 (110) = 0 = T 170 (110) and T 168 (111) = 1 =
T 170 (111). The rest of the proof is identical to the
Fig. 42. Rule 40 : The pattern evolves into the pat- one for the previous theorem.
tern after one iteration, and then it is in the basin of
attraction of · · · (see Fig. 41).
5. How to Find Analytically the
Fig. 42, we can see that after one step we always Basin-Tree Diagrams for
obtain a string with a run of at least two 0’s which, Bernoulli Attractors
as proved previously, will eventually evolve into the Our previous papers [Chua et al., 2007a] and [Chua
string x = (00 . . . 0). et al., 2007b] were in large part devoted to show-
The second part of the theorem states that ing the basin-tree diagrams, all of them found by
strings containing no more than one consecutive brute force, of the complex and hyper Bernoulli-shift
0 and at most two consecutive 1’s belong to a rules. In general, these kinds of rules have many
Bernoulli attractor. By hypothesis, the triplets Bernoulli orbits, either attractors or Isles of Eden,
(000), (001), (100), (111) will never be present into whose number increases with the length L of the
these kinds of strings, and hence we need to ana- bit strings. Therefore, it is particularly important
lyze only the triplets (010), (011), (101) and (110). to characterize them analytically, in order to avoid
It is straightforward to notice that with such kind long and tedious simulations, which make it prac-
of restriction rule 40 is equivalent to rule 170 tically impossible to find the basins of attraction
because T 40 (010) = 0 = T 170 (010), T 40 (011) = when L is greater than 30.
1 = T 170 (011), T 40 (101) = 1 = T 170 (101) and Here we present a method that allows us
T 40 (110) = 0 = T 170 (110). Since rule 170 is to reconstruct the whole basin-tree diagram of a
Bernoulli-shift attractor by analyzing only a frac-
a right-copycat rule (or equivalently, a left shift),
tion of all bit strings with a fixed L.
for these initial conditions the dynamics of rule 40
corresponds exactly to the dynamics of rule 170 ,
5.1. Bernoulli-στ basin-tree
which is a Bernoulli orbit with σ = 1 and τ = 1.
Moreover, the basin of attraction of this orbit is generation formula
empty, because either a string fulfills the conditions By definition, given any string belonging to a
of the Theorem, and then it belongs to the orbit, or Bernoulli στ -shift orbit, after τ iterations we find
it does not, and then it converges to 0. the same string shifted σ positions towards left, if σ
A very similar theorem holds for rule 168 , is positive.17 For example, Fig. 43 depicts a period
belonging to group 1 too. In this case the firing pat- T = 14 Bernoulli στ -shift orbit for rule 110 , in
terns are , , and . which σ = 1 and τ = 4.
16
Note that this statement is very similar to the one for 40 , but the limitation on the number of consecutive 1’s has been
removed.
17
In the following we consider only the case σ > 0, because any attractor with σ < 0 is equivalent to another one with
σ = L − |σ| > 0 and same τ .
iterations, nτ = T τ (n0 ), it follows that

N
nτ = n0 · 2σ (mod 2L − 1) (44)
Furthermore, this formula can be modified to find

not only nτ , but also all other elements of the
orbit generated by n0 . A Bernoulli στ -shift orbit
with period T harbors exactly lcm(τ, T ) different
subgroups, where lcm indicates the least common
multiple, each with length T /lcm(τ, T ), henceforth
called order of the subgroup. This means that for-
mula (44) can be iterated (T /lcm(τ, T )) − 1 times
to obtain a corresponding number of strings, using
at each iteration the result nτ as new value for n0 .
As a consequence, Eq. (44) can be generalized as
follows
np·τ (mod T ) = n0 · 2p·σ (mod2L − 1),

T (45)
0≤p< , p∈N
lcm(τ, T )
where the parameter p indicates the current iter-

ation. Obviously, iteration T /lcm(τ, T ) would give
n0 itself.
Fig. 43. Example of Bernoulli στ -shift attractor with σ = 1,
τ = 4, for rule 110 and L = 7. For example, this procedure can be applied to
the Bernoulli στ -shift attractor (with τ = 2 and
In general, a left shift by σ positions in the T = 14) of rule 110 when L = 7, as depicted in
binary representation x = (x0 x1 . . . xL−1 ) is equiv- Fig. 44. From what was explained previously, there
alent to a multiplication by 2σ in the decimal rep- will be lcm(τ, T ) = 2 different subgroups, each with
L−1 length T /lcm(τ, T ) = 7. Starting from an arbitrary
resentation n, where n = i=0 xL−1−i 2i .
Therefore, if n0 is a (decimal) element of element of a period-14 Bernoulli στ -shift orbit (cyan
a Bernoulli-στ orbit, and nτ the result after τ circles in Fig. 44), like n0 = 31, and using the for-
mula (45), we obtain the values:
p = 1: n1·4(mod 14) = n0 · 21·1 mod (27 − 1) ⇒ n4 = 31 · 2 (mod127) = 62

p = 2: n2·4(mod 14) = n0 · 22·1 mod (27 − 1) ⇒ n8 = 31 · 4 (mod127) = 124
p = 3: n3·4(mod 14) = n0 · 23·1 mod (27 − 1) ⇒ n12 = 31 · 8 (mod127) = 121
p = 4: n4·4(mod 14) = n0 · 24·1 mod (27 − 1) ⇒ n2 = 31 · 16 (mod127) = 115
p = 5: n5·4(mod 14) = n0 · 25·1 mod (27 − 1) ⇒ n6 = 31 · 32 (mod127) = 103
p = 6: n6·4(mod 14) = n0 · 26·1 mod (27 − 1) ⇒ n10 = 31 · 64 (mod127) = 79
Finally, for p = T /lcm(τ, T ) = 7, n7·4( mod 14) = n0 = 31, as expected.
p = 7: n7·4( mod 14) = n0 · 27·1 mod (27 − 1) ⇒ n0 = 31 · 128 (mod127) = 31
The seven elements belonging to the second subgroup, which can be identified in the period-14 Bernoulli
orbit in Fig. 44, can be found by using a different starting point. Since T 110 (31) = 49, n0 = 49 is a natural
candidate; then, it follows that
p = 1: n1·4( mod 14) = n0 · 21·1 mod (27 − 1) ⇒ n4 = 49 · 2 (mod127) = 98
p = 2: n2·4( mod 14) = n0 · 22·1 mod (27 − 1) ⇒ n8 = 49 · 4 (mod127) = 69
L=7 Bernoulli (σ=1,τ=4) Period-14 Attractor

110
Fig. 44. Basin-tree diagram of the Bernoulli στ -shift attractor with σ = 1, τ = 4, for rule 110 and L = 7.
p = 3: n3·4( mod 14) = n0 · 23·1 mod (27 − 1) ⇒ n12 = 49 · 8 (mod127) = 11

p = 4: n4·4( mod 14) = n0 · 24·1 mod (27 − 1) ⇒ n2 = 49 · 16 (mod127) = 22
p = 5: n5·4( mod 14) = n0 · 25·1 mod (27 − 1) ⇒ n6 = 49 · 32 (mod127) = 44
p = 6: n6·4( mod 14) = n0 · 26·1 mod (27 − 1) ⇒ n10 = 49 · 64 (mod127) = 88
p = 7: n7·4( mod 14) = n0 · 27·1 mod (27 − 1) ⇒ n0 = 49 · 128 (mod127) = 49
Also in this case, when p = T /lcm(τ, T ) = 7 we find the basin tree. This is because in general
the initial value n0 = 49, as expected. T T
N N
Thanks to the formula (45), which generalizes xn −−−→ xn+1 ⇒ σ(xn ) −−−→ σ(xn+1 ) (46)
the analogue presented in [Chua et al., 2006] about
the στ -Bernoulli rules of group four, all 14 elements where σ(xn ) is the string obtained by shifting
of the Bernoulli στ -shift orbit were found starting σ times the bit string xn . For example, since
just from n0 = 31 and n0 = 49. T 110 (13) = 31, we can expect that a shifted ver-
Remarkably, this formula can be also used to sion of the bit string representing 31 is obtained
find elements that do not belong to the orbit but to by transforming under the rule N the shifted bit
string representing 13.

p = 1: n1·4( mod 14) = n0 · 21·1 mod (27 − 1) ⇒ n4 = 13 · 2 (mod127) = 26
and indeed
T T
110 110
13 −−−−→ 31 ⇒ σ(13) = 26 −−−−→ σ(31) = 62
The values for n0 = 13 are
p = 2: n2·4( mod 14) = n0 · 22·1 mod (27 − 1) ⇒ n8 = 13 · 4 (mod127) = 52
p = 3: n3·4( mod 14) = n0 · 2 3·1
mod (2 − 1) ⇒ n12 = 13 · 8 (mod127) = 104
7
p = 4: n4·4( mod 14) = n0 · 2 4·1

mod (27 − 1) ⇒ n2 = 13 · 16 (mod127) = 81
p = 5: n5·4( mod 14) = n0 · 25·1 mod (27 − 1) ⇒ n6 = 13 · 32 (mod127) = 35
p = 6: n6·4( mod 14) = n0 · 26·1 mod (27 − 1) ⇒ n10 = 13 · 64 (mod127) = 70
and, as usual, for p = T /lcm(τ, T ) = 7 we obtain the initial value n0 = 13, as expected.
p = 7: n7·4( mod 14) = n0 · 27·1 mod (27 − 1) ⇒ n0 = 13 · 128 (mod127) = 13
To sum up, all the information about a Bernoulli [Chua et al., 2007b]), and suppose that no infor-
attractor can be retrieved from only one element of mation about the form of the basin-tree diagrams
each subgroup of the orbit, like 31 and 49, and their is known. We can use x0 = (0000001) = 1 as the
basins of attraction (in this example the basins are
formed by 15 bit strings altogether, as depicted in
Fig. 45). Detail of Bernoulli (σ=1, τ=4)
Then, using only these 17 elements we are able
to draw the whole basin-tree diagram, which is com- Period-14 Attractor, L=7
posed of 119 bit strings!
For obvious reasons, we christen Eq. (45) as the
“Bernoulli στ -shift basin-tree generation formula”,
and it can be applied to any rule having Bernoulli- 110
shift attractors.18
5.2. A practical application of the

Bernoulli στ -shift basin tree
generation formula
The procedure described previously allows us to
find all elements of a Bernoulli basin tree start-
ing from a limited subset of them. In fact, in
some cases, this method is even more powerful,
because it allows us to obtain a basin tree with-
out knowing any of its elements! This happens
when all of the possible shifts of a bit string x do
not belong to the same basin tree as x, but form
a different, though topologically equivalent, basin Fig. 45. Detail of the basin-tree diagram of the Bernoulli
tree. στ -shift attractor with σ = 1, τ = 4, for rule 110 and L = 7.
Let us illustrate this case through an example, These 17-bit strings allow us to reconstruct the whole basin-
considering rule 110 with L = 6 (see Table 11 in tree diagram, thanks to formula (45).
18
Rules belonging to groups 4–6 have robust Bernoulli στ -shift attractors. In addition, rules belonging to groups 1–3 may also
harbor isolated Bernoulli στ -shift Isles of Eden and attractors.
starting point and find its successors iteratively: Since x5 is a shifted version of x2 — in particular
x5 = σ 4 (x2 ) — both strings belong to a Bernoulli
x1 = T 110 (x0 ) = (000011) = 3,
στ -shift orbit with τ = 5−2 = 3 and σ = 4. By using
x2 = T 110 (x1 ) = (000111) = 7, the formula (45) with n0 = 7, it is possible to notice
that n9 = n0 = 7, which means that the period of
x3 = T 110 (x2 ) = (001101) = 13, the attractor is the minimum T ≡ 0 (mod 9) ⇒
T = 9. Therefore, in the Bernoulli στ -shift orbit
x4 = T 110 (x3 ) = (111111) = 31,
there are lcm(τ, T ) = 3 different subgroups, each
x5 = T 110 (x4 ) = (110001) = 49. with length (order) T /lcm(τ, T ) = 3.
The elements of the first subgroup, found using
n0 = 7, are as follow:
p = 1: n1·3( mod 9) = n0 · 21·4 mod (26 − 1) ⇒ n3 = 7 · 16 (mod 63) = 49

p = 2: n2·3( mod 9) = n0 · 22·4 mod (26 − 1) ⇒ n6 = 7 · 256 (mod 63) = 28
p = 3: n3·3( mod 9) = n0 · 23·4 mod (26 − 1) ⇒ n0 = 7 · 4096 (mod 63) = 7
As for the second subgroup, we use n0 = 13 because T 110 (7) = 13, and obtain
p = 1: n1·3( mod 9) = n0 · 21·4 mod (26 − 1) ⇒ n3 = 13 · 16 (mod 63) = 19

p = 2: n2·3( mod 9) = n0 · 22·4 mod (26 − 1) ⇒ n6 = 13 · 256 (mod 63) = 52
p = 3: n3·3( mod 9) = n0 · 23·4 mod (26 − 1) ⇒ n0 = 13 · 4096 (mod 63) = 13
Finally, for the third subgroup n0 = 31 because T 110 (13) = 31, and obtain
p = 1: n1·3( mod 9) = n0 · 21·4 mod (26 − 1) ⇒ n3 = 31 · 16 (mod 63) = 55

p = 2: n2·3( mod 9) = n0 · 22·4 mod (26 − 1) ⇒ n6 = 31 · 256 (mod 63) = 61
p = 2: n2·3( mod 9) = n0 · 22·4 mod (26 − 1) ⇒ n6 = 31 · 256 (mod 63) = 31
Therefore, the basin tree corresponding to strings

7, 13 and 31 is sufficient to reconstruct the whole For instance, the Bernoulli στ -shift orbit of the
basin-tree diagram exhibited in Table 11 of [Chua first basin tree is 7 → 13 → 31 → 49 → 19 → 55 →
et al., 2007b]. 28 → 52 → 61 → 7, or equivalently
The binary representation of 7, which we used → →
as the “seed” for the first subgroup of bit strings
in the orbit, is , and it has six possi- → → →
ble shifts: ↔ 14, ↔ 28, → → →
↔ 56, ↔ 49, ↔
→
35, and ↔ 7 itself.
Surprisingly, only three of them form part of the Applying the operator σ 1 , we are able to find the
orbit, namely 7, 28 and 49. What about the others? Bernoulli στ -shift orbit of the “twin” basin tree
Obviously, they are included in a different attractor,
→ →
which, for the formula (46), must be strictly related
to the one we already know. → → →
In particular, the “missing” strings are 14, 56 → → →
and 35, and observe that 14 = σ 1 (7), 56 = σ 1 (28),
and 35 = σ 1 (49). Therefore, this “twin” basin tree →
can be found by applying the operator σ 1 (one posi- or, using the decimal representation,
tion left shift) to every string of the first basin tree,
and keeping the same topology, as evident from 14 → 26 → 62 → 35 → 38 → 47 → 56
Table 11 of [Chua et al., 2007b]. → 41 → 59 → 14.
Through a similar procedure, we can draw the whole x has only two possible shifts, because σ 2 (x) = x.
“twin” basin tree starting from the first one. But the string x obtained by changing only the
In general, let us consider a string x with length last pixel from 1 to 0, x = (0101 . . . 00), will have
L which can be shifted s times, corresponding to s clearly s = L shifts, where L can be very large.
different bit strings, with s ≤ L. If x belongs to However, the parameter s for a string x can be
a subgroup with order o under the transformation obtained experimentally as s = mint≤L {σ t (x) = x}.
T N (x), then o must divide s and there will be s/o
“twin” orbits. In the example just presented, the
string x = (000111) corresponding to n = 7 could 6. Old Theorems and New Results
be shifted six times, s = 6, and the order of the sub- for Additive Cellular Automata
group, under the transformation T 110 (x), is o = 3; The behavior of additive cellular automata can be
therefore, there will be s/o = 6/3 = 2 “twin” orbits, described through the expression
which we actually found.
Unfortunately, calculating a priori the number xn+1
i = T N (xni−1 xni xni+1 )
of shifts s of a string x may not be easy, since it
involves certain deep concepts from coding theory. = axni−1 + bxni + cxni+1 (mod 2) (47)
Sometimes, modifying only one bit in a string can
change dramatically the number of possible shifts where a, b, c ∈ {0, 1}.
s. For example, let us consider a bit string x with Since Eq. (47) contains three free parameters,
length L made of L/2 repetitions of the pattern 01; there are 23 = 8 distinct additive rules, which can
namely, string x = (0101 . . . 01). It is evident that be represented through simple Boolean expressions
as follows
(abc) = (000): xn+1

i =0 (mod 2) ⇒ xn+1
i = 0 ↔ Rule 0
(abc) = (001): xn+1
i = xni+1 (mod 2) ⇒ xn+1
i = xni+1 ↔ Rule 170
(abc) = (010): xn+1
i = xni (mod 2) ⇒ xn+1
i = xni ↔ Rule 204
(abc) = (011): xn+1
i = xni + xni+1 (mod 2) ⇒ xn+1
i = xni ⊕ xni+1 ↔ Rule 102
(abc) = (100): xn+1
i = xni−1 (mod 2) ⇒ xn+1
i = xni−1 ↔ Rule 240
(abc) = (101): xn+1
i = xni−1 + xni+1 (mod 2) ⇒ xn+1
i = xni−1 ⊕ xni+1 ↔ Rule 90
(abc) = (110): xn+1
i = xni−1 + xni (mod 2) ⇒ xn+1
i = xni−1 ⊕ xni ↔ Rule 60
(abc) = (111): xn+1
i = xni−1 + xni + xni+1 (mod 2) ⇒ xn+1
i = xni−1 ⊕ xni ⊕ xni+1 ↔ Rule 150
Each additive rule N has a complementary anti-

additive rule N c = 255 − N . The eight additive axis a passing through the vertices containing rules
and the eight anti-additive rules are represented 0 and 150 , all the other rules are included into two
in Table 27, along with their Boolean cubes and planes, A and B, perpendicular to a. Thanks to this
Boolean expressions. representation, it is possible to separate the rules
Table 28 shows that four additive rules are glob- into four different subsets, according to the plane
ally equivalent to their corresponding anti-additive they belong to: The first subset contains rule 0 ;
rules, whereas in the remaining four cases addi- the second subset contains rules belonging to plane
tive and anti-additive rules exhibit the so-called A, namely 170 , 204 and 240 ; the third subset
alternating symmetry duality [Chua et al., 2007a] contains rules belonging to plane B, namely 60 ,
described in Sec. 6.2. 90 and 102 ; finally, the fourth subset contains
rule 150 . Rules belonging to the same subset have
Additive rules can be displayed on the vertices the same complexity index κ (κ = 1, κ = 2, and
of a cube19 as illustrated in Fig. 46(a). Drawing the κ = 3, respectively) and the same number of inputs
19
This cube shows the connection among additive rules, and it should not be confused with the 256 Boolean cubes in Table 1.
Table 27. Eight additive and eight anti-additive rules, their Boolean cubes, and their
analytical formulas.
c
8 Additive Rules N 8 Anti-Additive Rules N
2 3 2 3
6 7 6 7
0 0 1 255 0 1
4 5 4 5
n +1 n +1
x i =0 x i =1
2 3 2 3
6 7 6 7
60 0 1 195 0 1
4 5 4 5
n +1
x i = x ⊕x n
i −1
n
i x n +1
i = x ⊕ xin n
i −1
2 3 2 3
6 7 6 7
90 0 1 165 0 1
4 5 4 5
n +1
x i = x ⊕x n
i −1
n
i +1 x n +1
i = x ⊕ xin+1n
i −1
2 3 2 3
6 7 6 7
102 0 1 153 0 1
4 5 4 5
n +1
x i =x ⊕x n
i
n
i +1 xn +1
i = x ⊕ xin+1
n
i
2 3 2 3
6 7 6 7
150 0 1 105 0 1
4 5 4 5
n +1
x
i =x ⊕x ⊕x
n
i −1
n
i
n
i +1 xn +1
i = x ⊕ x ⊕ xin+1
n
i −1
n
i
2 3 2 3
6 7 6 7
170 0 1 85 0 1
4 5 4 5
n +1
x i =x n
i +1 x n +1
i =x n
i +1
2 3 2 3
6 7 6 7
204 0 1 51 0 1
4 5 4 5
n +1
xi =x n
i x n +1
i =x n
i
2 3 2 3
6 7 6 7
240 0 1 15 0 1
4 5 4 5
n +1
x i =x n
i −1 x n +1
i =x n
i −1
Table 28. Relationship between additive and anti-additive which it has been possible to find several remarkable
rules.
results.
Anti-additive Relationship between the
Additive Rule
rule two rules 6.1. Theorems on the maximum
0 255 Global equivalence period of attractors and Isles
of Eden
60 195 Global equivalence Some partial results about additive rules were given
in [Chua et al., 2007a] and [Chua et al., 2007b]. For
90 165 Global equivalence example, the periods of the attractors of rule 90
found by brute force for 1 ≤ L ≤ 100 were listed
102 153 Global equivalence in Table 25 of [Chua et al., 2007a]. Unfortunately,
this table is incomplete, because the period of the
150 105 Alternating symmetry attractors corresponding to certain L is so large that
exceeds the simulation time.
170 85 Alternating symmetry Nevertheless, [Martin et al., 1984] contains sev-
eral valuable results about additive rules, which
204 51 Alternating symmetry allow us to find all of the missing values of the table
under consideration. For the reader’s convenience,
240 15 Alternating symmetry we summarized such results in two theorems using
a plain style, our notation and our nomenclature.
First of all, we need to introduce three concepts
from number theory that will be extensively used
n (n = 0, n = 1, n = 2, and n = 3, respectively);
in the following.
moreover, the Boolean cube corresponding to one
element of a subset can be transformed into that Definition 6.1 (Euler totient function). Given a
corresponding to another element of the same sub- positive integer n, the Euler totient φ(n) is the num-
set through a rotation. A similar procedure can ber of positive integers less than n that are coprime
be followed for anti-additive rules, as illustrated in to n.
Fig. 46(b). Remarkably, except for rule 0 and 255 , Definition 6.2 (Multiplicative order function).
the number of inputs of any additive or anti-additive Given a positive integer n, the multiplicative order
rule coincides with its complexity index.20 of 2 (mod n) is the minimum positive integer o(n)
There are only nine globally-independent addi- for which
tive rules: 0 , 15 , 51 , 60 , 90 , 105 , 150 , 170 ,
204 . Five of them — namely 0 , 15 , 51 , 170 2o(n) = 1 (mod n)
and 204 — have complexity index κ = 1, and their Definition 6.3 (Multiplicative suborder function).
behavior can be easily analyzed; for this reason, we Given a positive integer n, the multiplicative subor-
dubbed them “trivial additive rules”. In contrast, der of 2 (mod n) is the minimum positive integer
the remaining four local rules — 60 , 90 , 105 s(n) for which
and 150 —
have complexity index κ = 2 rules
2s(n) = ±1 (mod n)
60 and 90 or κ = 3 rules 105 and 150 .
According to the classification introduced in [Chua Remark 6.1. It is possible to prove that φ(n) ≤ n−1,
et al., 2007a], rule 60 belongs to group 6 (hyper s(n) ≤ (n − 1)/2, and s(n)|o(n)|φ(n), where the bar
Bernoulli-shift rules), and rules 90 , 105 and 150 “|” denotes “divides”.
belong to group 5 (complex Bernoulli-shift rules); The values of φ(n), o(n) and s(n) for n odd and
therefore, they exhibit complex behaviors, and we n < 100 are listed in Table 29.21
dubbed them “nontrivial additive rules”. Now, we are ready to present the two afore-
In the following, we will take into considera- mentioned theorems, whose proofs can be found in
tion only these four nontrivial additive rules, for [Martin et al., 1984].
20
We could assign complexity index κ = 0 to 0 and 255 , but this would mean introducing a fourth value for κ exclusive to
only two rules, and not being consistent with the notation used in our previous works.
21
Note that the multiplicative order and suborder functions of 2 (mod n) are not defined for n even.
(a)
(b)
Fig. 46. Geometrical interpretation of additive (a) and anti-additive (b) rules.
Table 29. Values of the Euler’s totient function φ(n), multiplicative order function o(n), and multiplicative suborder
function s(n), for n < 100, n odd.
φ (n) : Euler's o(n) : Multiplicative s(n) : Multiplicative φ (n) : Euler's o(n) : Multiplicative s(n) : Multiplicative
n n
totient function order function suborder function totient function order function suborder function
1 1 1 1 51 32 8 8
3 2 2 1 53 52 52 26
5 4 4 2 55 40 20 20
7 6 3 3 57 36 18 9
9 6 6 3 59 58 58 29
11 10 10 5 61 60 60 30
13 12 12 6 63 36 6 6
15 8 4 4 65 48 12 6
17 16 8 4 67 66 66 33
19 18 18 9 69 44 22 22
21 12 6 6 71 70 35 35
23 22 11 11 73 72 9 9
25 20 20 10 75 40 20 20
27 18 18 9 77 60 30 30
29 28 28 14 79 78 39 39
31 30 5 5 81 54 54 27
33 20 10 5 83 82 82 41
35 24 12 12 85 64 8 8
37 36 36 18 87 56 28 28
39 24 12 12 89 88 11 11
41 40 20 10 91 72 12 12
43 42 14 7 93 60 10 10
45 24 12 12 95 72 36 36
47 46 23 23 97 96 48 24
49 42 21 21 99 60 30 15
Theorem 6.1. (Maximum period of the orbits of (c) If L is even but not of the form L = 2n , then
rules 90 and 150 ). Let L be the length of the bit TL = 2 · TL/2 .
strings and let TL be the maximum period of the
Although these theorems do not include any state-
orbit (attractor or Isle of Eden) of rules 90 and
ment about rule 105 , in [Chua et al., 2007a] it was
150 , then
proved that rules 105 and 150 exhibit an alter-
nating symmetry duality, which is a kind of rela-
(a) If L is odd, then TL divides the quantity T ∗ =
tionship weaker than the topologically conjugated
2s(L) − 1, where s(L) is the multiplicative sub-
transformations from the Vierergruppe defined in
order of 2 (mod n);
[Chua et al., 2004] but still sufficient to relate the
(b) If L = 2n , then x = (00 . . . 0) is a global
orbits of the two rules.
attractor ;
The alternating symmetry duality can be
(c) If L is even but not of the form L = 2n , then
described as follows: Given two bit strings xn =
TL = 2 · TL/2 .
(xn0 xn1 . . . xnL−1 ) and xn = (xn n n
0 x1 . . . xL−1 ), gen-
Theorem 6.2. (Maximum period of the attractors erated respectively by rules 105 and 150 from
of rule 60 ). Let L be the length of the bit strings the same initial state, x and x obey the following
and let TL be the maximum period of the attractor relations
of rule 60 , then xn
1 n n n
i = [1 − (−1) ] + (−1) xi (48)
2
(a) If L is odd, then TL divides the quantity T ∗ = 1
2o(L) − 1, where o(L) is the multiplicative order xni = [1 − (−1)n ] + (−1)n xn
i (49)
2
of 2 (mod n);
(b) If L = 2n , then x = (00 . . . 0) is a global Therefore, xn = xn for n even and xn = xn
attractor; for n odd; in other words, χ 105 = χ 150 , where
2 2
χ2 is the time-2 characteristic function of the local to rule 150 — containing the actual value of TL
N
along with a bit string belonging to the orbit and its
rule N .
result after TL − 1 iterations, so that the reader can
For L odd, Theorem 6.1 states that TL is also
check the correctness of our simulations. In some
odd for rule 150 . By applying the alternating sym-
cases the value of TL is so large that no simulation
metry duality, it is straightforward to see that TL
was possible. However, in these cases the orbits are
has to be even for 105 , otherwise we would obtain
so robust that practically every bit string has the
a contradiction. In particular, when L is odd then
indicated TL .
TL for 105 has to be twice as that of TL for 150 .
Finally, a graphical representation of the rela-
For L even, Theorem 6.1 states that TL is also
tionship between the length L of the bit string and
even for rule 150 . Therefore, starting from any ele-
the maximum period TL of the orbit for nontrivial
ment of the orbit and going through the whole orbit
additive rules is depicted in Figs. 47–50.
and taking every other element, we will eventually
go back to the first element after TL /2 steps. As
a consequence of the alternating symmetry dual- 6.2. Scale-free property for additive
ity, the orbit for rule 105 contains all these TL /2 rules
elements of the orbit for rule 150 , replacing the All nontrivial additive rules are either complex or
remaining TL /2 with new ones. In conclusion, for L hyper Bernoulli-shift rules; hence, their Bernoulli
even TL for rule 105 is equal to TL for rule 150 . parameters σ and τ , and consequently the periods
We summarize these results in the following of their orbits, depend crucially on L.
theorem: In principle, given L and the corresponding max-
Theorem 6.3. (Maximum period of the orbits of imum period TL of the orbits of a nontrivial addi-
rule 105 ). Let L be the length of the bit strings and tive rule, it is not possible to extract any information
let TL be the maximum period of the orbit (attractor about the maximum period TL of the orbits for L =
or Isle of Eden) of rule 105 , then L; in other words, TL /L is unrelated to TL /L .
Nevertheless, in [Chua et al., 2007a] it was
(a) If L is odd, then TL divides the quantity TL∗ = empirically noticed that for rules 90 , 105 and
2 · (2s(L) − 1), where s(L) is the multiplicative 150 there exist certain values of L and L for which
suborder of 2 (mod n); (log(TL )−log(TL ))/(log(L )−log(L)) = 1, or equiv-
(b) If L is even, then TL divides the quantity TL∗ = alently TL /L = TL /L . This means that such rules
2 · (2s(L/2) − 1). exhibit a scale-free property as L → ∞, and hence
there are some L = L for which it is possible to
It is important to notice that the parameter TL∗ does know TL using only the information about TL ; the
not give necessarily the maximum period TL of the presence of a scale-free property also for rule 60
orbits, because in general TL |TL∗ ; however, we found is confirmed in [Chua et al., 2007b]. The scale-free
experimentally that TL = TL∗ for most L. property can be easily noticed looking at Figs. 51–54
The practical implications of Theorems 6.1–6.3, and noticing that most of the points lie on diagonal
and Remark 6.1 are noteworthy. First of all, they lines with slope 1 (here we draw just a few of them).
state that TL can assume very few values because it A question naturally arises: For what values of
has to divide TL∗ , and then resort to brute force, and L and L does the scale-free property hold? In order
numerical simulations is extremely minimal; second, to characterize quantitatively this phenomenon, we
we discover that the maximum period of the orbits give the following definition:
for additive rules is not 2L , as hypothesized in our
L−1
previous papers, but 2 2 −1 for rules 90 and 150 , Definition 6.4 (Scale-free order). Let L be the
L+1 length of the bit strings and let TL∗ be the maximum
2L−1 for rule 60 , and 2 2 − 2 for rule 105 .
period of the orbits (attractors or Isles of Eden) of
The values of TL∗ for the nontrivial additive rules
a nontrivial additive rule, then its scale-free order
for L odd and L < 100 are listed in Table 30.
ξL is defined as22
In addition to this, we also give four tables —
Table 31 refers to rule 60 , Table 32 refers to rule TL∗ 2s(L) − 1
ξL = =
90 , Table 33 refers to rule 105 , and Table 34 refers L L
22
We have chosen the symbol ξL to denote the scale-free order to recall the parallel straight lines visually representing the
scale-free property.
Table 30. Maximum period-T of attractors and/or Isles of Eden of local rules 60 , 90 , 105 and 150 , for L < 100, L odd.
Rules 90 and 150: Rule 105: Rule 60: Rules 90 and 150: Rule 105: Rule 60:
L * s( L) * * o( L) * s( L) * *
TL = 2 − 1 TL = 2 · (2s ( L ) − 1) T =2
L −1 L TL = 2 − 1 TL = 2 · (2s ( L ) − 1) T = 2o ( L ) − 1
L
1 1 1 1 51 255 510 255

3 1 2 3 53 67,108,863 134,217,726 252-1
5 3 6 15 55 1,048,575 2,097,150 1,048,575
7 7 14 7 57 511 1,022 262,143
9 7 14 63 59 536,870,911 1,073,741,822 258-1
11 31 62 1,023 61 1,073,741,823 2,147,483,646 260-1
13 63 126 4,095 63 63 126 63
15 15 30 15 65 63 126 4,095
17 15 30 255 67 8,589,934,591 17,179,869,182 266-1
19 511 1,022 262,143 69 4,194,303 8,388,606 4,194,303
35 35 35
21 63 126 63 71 2 -1 2·(2 -1) 2 -1
2603
23 2,047 4,094 2,047 73 511 1,022 511
25 1,023 2,046 1,048,575 75 1,048,575 2,097,150 1,048,575
27 511 1,022 262,143 77 1,073,741,823 2,147,483,646 1,073,741,823
39
29 16,383 32,766 268,435,455 79 2 -1 2·(239-1) 239-1
54
31 31 62 31 81 134,217,727 268,435,454 2 -1
41 41 82
33 31 62 1,023 83 2 -1 2·(2 -1) 2 -1
35 4,095 8,190 4,095 85 255 510 255
37 262,143 524,286 236-1 87 268,435,455 536,870,910 268,435,455
39 4,095 8,190 4,095 89 2,047 4,094 2,047
41 1,023 2,046 1,048,575 91 4,095 8,190 4,095
43 127 254 16,383 93 1,023 2,046 1,023
36 36
45 4,095 8,190 4,095 95 236-1 2·(2 -1) 2 -1
47 8,388,607 16,777,214 8,388,607 97 16,777,215 33,554,430 248-1
49 2,097,151 4,194,302 2,097,151 99 32,767 65,534 1,073,741,823
Table 31. Data for generating and verifying the period of period-T orbits of 60 .
2604
2605
2606
2607
2608
2609
2610
2611
2612
2613
2614
2615
2616
2617
2618
2619
2620
2621
2622
2623
2624
2625
2626
2627
Fig. 47. Relationship between the length L of the bit strings and the maximum period T of the attractors (symbol ) for
rule 60 . In some cases (symbol ) the maximum theoretical value is too large to be verified experimentally, and we hypnotize
that it is a multiple of the actual maximum period T .
Fig. 48. Relationship between the length L of the bit strings and the maximum period T of the attractors (symbol ) for
rule 90 .
Fig. 49. Relationship between the length L of the bit strings and the maximum period T of the attractors (symbol ) or
Isles of Eden (symbol ) for rule 105 .
Fig. 50. Relationship between the length L of the bit strings and the maximum period T of the attractors (symbol ) or
Isles of Eden (symbol ) for rule 150 .
Fig. 51. Scale-free property for rule 60 .




Therefore, we need to find all L and L , with L = L, is TL∗ = 15 (see Table 30); therefore, ξL = TL∗ /L = 1
for which ξL = ξL . From now on, we will refer and o(ξL ) = 1.
exclusively to rules 90 and 150 , for which TL∗ = From Eq. (50) n = (o(L)/o(ξ)) − 1 = (4/1)
2s(L) − 1; however, the conclusions for rules 60 and − 1 = 3, and ((21 − 1)/1)1 + 21·1 + 21·2 + 21·3 ) =
105 are very similar and can be achieved through 15, as expected.
a similar procedure. In this case, we consider all values n > 1, i.e.
The following theorem represents an important n = 1, 2, 4, etc. to be sure of finding all L = L such
advance towards our goal. that ξL = ξL = 1. For example,
Theorem 6.4 (Scale-free decomposition). Let L 21 − 1

n : 1 L = (1 + 21 ) = 3
be the length of the bit strings, L odd, and let ξL 1
be its scale-free order. If o(L) = s(L),23 then L can 21 − 1
be uniquely expressed as n : 2 L = (1 + 21 + 21·2 ) = 7
1
i
2o(ξL ) − 1 i·o(ξL ) 21 − 1
L= 2 (50) n : 4 L = (1 + 21 + 21·2 + 21·2 + 21·3 + 21·4 )
ξL 1
i=0
= 31
where n = (o(L)/o(ξL )) − 1. ..
.
This theorem is extremely powerful, because given
any L for which o(L) = s(L) and its scale-free order Note that in this example L = m i
i=0 2 = 2
(m+1) − 1,
ξL , we can easily find analytically all the L for and since ξL = 1 it follows that TL∗ = 2(m+1) − 1, as
which ξL = ξL by just varying the parameter n. already stated in Theorem 4 of [Chua et al., 2007b].
Its proof requires only basic concepts from number Therefore, the scale-free decomposition gives an
theory, but we present it in Appendix B in order alternate proof for such theorem.
to illustrate here the application of the theorem
Example 6.3. For L = 73, the values of the multi-
through some examples.
plicative order and suborder are o(73) = s(73) = 9
Example 6.1. For L = 21, the values of the multi- (see Table 29), and the maximum period of the
plicative order and suborder are o(21) = s(21) = 6 orbit is TL∗ = 511 (see Table 30); therefore,
(see Table 29), and the maximum period of the orbit ξL = TL∗ /L = 7 and o(ξL ) = 3.
is TL∗ = 63 (see Table 30); therefore, ξL = TL∗ /L = 3
and o(ξL ) = 2. From Eq. (50) n = (o(L)/o(ξ))−1 = Table 35. Values of L < 105 having the same scale-free
(6/2) − 1 = 2, and ((22 − 1)/3)(1 + 21·2 + 22·2 ) = order ξL ≤ 15 for rule 90 , ξL = (2s(L) − 1)/L.
21, as expected.
Considering the following values for the param- Scale-free order ξL Length L of the bit strings
eter n, i.e. n = 3, 4, 5, etc., we can find an infinite
number of L = L such that ξL = ξL . For example, 3, 7, 15, 31, 63, 127, 255, 511,
1 1023, 2047, 4095, 8191, 16383,
22 − 1 32767, 65535, …
n : 3 L = (1 + 21·2 + 22·2 ) = 21 21, 85, 341, 1365, 5461, 21845,
3 3 87381, …
22 − 1
n : 4 L = (1 + 21·2 + 22·2 + 23·2 ) = 85 5 51, 819, 13107, …
3
22 − 1 7 73, 585, 4681, 37449, …

n : 5L = (1 + 21·2 + 22·2 + 23·2 + 24·2 ) = 341
3 9 455, 29127, …
..
. 11 93, 95325, …
Example 6.2. For L = 15, the values of the multi- 13 315, …

plicative order and suborder are o(15) = s(15) = 4
15 273, 4369, 69905, …
(see Table 29), and the maximum period of the orbit
23
We assume that o(1) = 1, as illustrated in Example 6.2.
Fig. 55. For rule 90 , lines corresponding to a scale-free order ξL ∈

/ N (blue lines) include attractors of only one L odd; lines
corresponding to scale-free orders ξL ∈ N (red lines) include attractors of infinitely many different odd values of L.
From Eq. (50) n = (o(L)/o(ξ)) − 1 = (9/3) − but Table 30 indicates that T9∗ = 31, and
1 = 2, and ((23 − 1)/7)(1 + 21·3 + 22·3 ) = 73, as then ξ9 = 3.44 = ξ73 . However, this happens
expected. because o(9) = 6 and s(9) = 3, and since
If we “go backward” using n = 1, as already o(9) = s(9), Theorem 6.4 cannot be applied
done as in Example 6.3, we apparently find a con- to L = 9.
tradiction, because
27 − 1 The values of L corresponding to ξL ≤ 15 are
n = 1 : L = (1 + 21·3 ) = 9 presented in Table 35. The results of this section
7
can be summarized in the following theorem: behavior of the corresponding local rule; this aspect
will be further analyzed in our next papers.
Theorem 6.5 (Scale-free property of additive A whole section has been devoted to introduce
rules). Let L be the length of the bit strings and let a new nomenclature for rules which harbor, or do
TL∗ be the maximum period of the orbit of a nontriv- not harbor, Isles of Eden. For the first time we were
ial additive rule, then for any L ∈ N the scale-free able to find that Isles of Eden are everywhere in 1-D
property TL∗ /L = TL∗ /L holds iff cellular automata: in fact, only 28 rules out of 256
(a) both L and L satisfy the conditions of Theo- do not have any Isle of Eden! Furthermore, we show
rem 6.4, if L is odd; that at least for two rules of group 1, a string either
(b) L = 2n · L, with n ∈ N, if L is even and belongs to the basin of a global attractor or is an
L, L = 2i ; Isle of Eden; hence, we can completely characterize
these rules by analyzing their Isles of Eden.
Figure 55 gives a graphical interpretation of this The Bernoulli στ -shift basin tree generation for-
theorem: red lines correspond to values of ξL ∈ N, mula presented in Sec. 4 has a practical importance,
which will include all odd L obtained from The- because thanks to it we only need to analyze a frac-
orem 6.4 (as stated in the first point of Theorem tion of the 2L possible bit strings to draw all of the
6.5) plus their even multiples (as stated in the sec- basin-tree diagrams of a given local rule. This for-
ond point of Theorem 6.5); blue lines correspond to mula will be also used in our next papers, in which
values of ξL ∈/ N, and then they include only one we will relate it with the “traveling waves” in cel-
odd L (because L does not satisfy the conditions of lular automata.
Theorem 6.4) plus its even multiples (as stated in Finally, the last section completes the work that
the second point of Theorem 6.5). began in [Chua et al., 2007a] and [Chua et al.,
2007b] about additive rules 60 , 90 , 105 and 150 .
Although some of the results were known, they were
7. Concluding Remarks not easily accessible and we have therefore recast
The first and most important result of this them using our notation for the reader’s conve-
paper is the characterization of the quasi-ergodic nience. Furthermore, we have included many tables,
phenomenon. Although it has been introduced so that all experiments reported can be accurately
empirically, quasi-ergodicity is tremendously useful reproduced and verified.
to support our classification of local rules into six In our opinion, the results presented in this
groups, since only complex and hyper Bernoulli-shift paper shed new light on a number of phenom-
rules appear to be quasi-ergodic. For the first time, ena, especially on the role of Isles of Eden and the
we have found an additional feature distinguishing behavior of Bernoulli στ -orbits. Most of our theo-
the Bernoulli στ -shift rules (group 4) from the com- rems and conjectures can be easily generalized to
plex and hyper Bernoulli-shift rules (groups 5 and bi-infinite bit strings by taking advantage of the
6), besides the dependence of the two Bernoulli analogy between non-spatially periodic finite strings
parameters σ and τ on the initial point and on and spatially periodic bi-infinite strings. Further-
the length L. We also included numerous figures more, many rules behave like a left- or right-shift
and tables in order to show the qualitative meaning for certain subsets of strings, and this property can
of quasi-ergodicity, and encourage readers to give a be used to define a completely new class of problems
quantitative interpretation of it. still unexplored.
In addition to quasi-ergodicity, the paper
presents further results about a variety of topics. References
First of all, a complete explanation for the
Cattaneo, G. & Quaranta Vogliotti, C. [1997] “The
emergence of fractals in the time-1 characteristic
‘magic’ rule spaces of neural-like elementary cellular
functions was given. These results, which general-
automata,” Theoret. Comput. Sci. 178, 77–102.
ize those presented in [Chua et al., 2005b], allow Chua, L. O. [1996] CNN: A Paradigm for Complexity
us to classify each rule according to the number (World Scientific, Singapore).
of fractal patterns it exhibits, and retrieve the rule Chua, L. O., Yoon, S. & Dogaru, R. [2002] “A nonlinear
number from the time-1 characteristic function. In dynamics perspective of Wolfram’s new kind of sci-
our opinion, the number and the form of the frac- ence. Part I: Threshold of complexity,” Int. J. Bifur-
tal patterns may give information on the dynamical cation and Chaos 13, 2377–2491.
Chua, L. O., Sbitnev, V. I. & Yoon, S. [2003] “A non- Shereshevsky, M. A. [1992] “Ergodic properties of cer-
linear dynamics perspective of Wolfram’s new kind of tain surjective cellular automata,” Mh. Math. 114,
science. Part II: Universal neuron,” Int. J. Bifurcation 305–316.
and Chaos 13, 2377–2491. Shirvani, M. & Rogers, T. D. [1991] “On ergodic
Chua, L. O., Sbitnev, V. I. & Yoon, S. [2004] “A non- one-dimensional cellular automata,” Commun. Math.
linear dynamics perspective of Wolfram’s new kind of Phys. 136, 599–605.
science. Part III: Predicting the unpredictable,” Int. Wuensche, A. & Lesser, M. [1992] The Global Dynamics
J. Bifurcation and Chaos 14, 3689–3820. of Cellular Automata: An Atlas of Basin of Attrac-
Chua, L. O., Sbitnev, V. I. & Yoon, S. [2005a] “A non- tion Fields of One-Dimensional Cellular Automata
linear dynamics perspective of Wolfram’s new kind of (Addison-Wesley, Reading, MA).
science. Part IV: From Bernoulli shift to 1/f spec-
trum,” Int. J. Bifurcation and Chaos 15, 1045–1183.
Chua, L. O., Sbitnev, V. I. & Yoon, S. [2005b] “A non- Appendix A
linear dynamics perspective of Wolfram’s new kind of
science. Part V: Fractals everywhere,” Int. J. Bifur- Theorem. Inequality (31) is given by
cation and Chaos 15, 3701–3849.

I I

Chua, L. O., Sbitnev, V. I. & Yoon, S. [2006] “A non- 1− xi−1 2−(i+1) + xi 2−(i+1) − 1
linear dynamics perspective of Wolfram’s new kind of
i=0 i=0
science. Part VI: From time-reversible attractors to
the arrow of time,” Int. J. Bifurcation and Chaos 16, I I

1097–1373. ≥ χ 60 ≥ xi−1 2−(i+1) − xi 2−(i+1)

Chua, L. O., Guan, J., Sbitnev, V. I. & Jinwook, S. i=0 i=0
[2007a] “A nonlinear dynamics perspective of Wol- where
fram’s new kind of science. Part VII: Isles of Eden,” I
Int. J. Bifurcation and Chaos 17, 2839–3012.
χ 60 = |xi−1 − xi |2−(i+1) (A.1)
Chua, L. O., Karacs, K., Sbitnev, V. I., Guan, J. & Jin-
i=0
wook, S. [2007b] “A nonlinear dynamics perspective of
Wolfram’s new kind of science. Part VIII: More Isles of Proof. The above formula can be split into two
Eden,” Int. J. Bifurcation and Chaos 17, 3741–3894.
parts, one for the upper bound
Courbage, M., Mercier, D. & Yasmineh, S. [1999]
“Travelling waves and chaotic properties in cellular I

automata,” Chaos 9, 893–901. |xi−1 − xi |2−(i+1)
Courbage, M. & Yasmineh, S. [2001] “Wavelengths dis- i=0
tribution of chaotic travelling waves in some cellular I
I

automata,” Physica D 150, 63–83. −(i+1) −(i+1)
≥ xi−1 2 − xi 2 (A.2)
Garay, B. M. & Chua, L. O. [2008] “Isles of Eden and the
i=0 i=0
ZUK theorem in Rd ,” Int. J. Bifurcation and Chaos
18(10), in press. and one for the lower bound
I
Hedlund, G. A. [1969] “Endomorphisms and automor- I

−(i+1) −(i+1)
phisms of the shift dynamical system,” Th. Comput. 1− xi−1 2 + xi 2 − 1
Syst. 3, 320–375.
i=0 i=0
Ireland, K. & Rosen, M. [1990] A Classical Introduction
to Modern Number Theory (Springer-Verlag). I

Martin, O., Odlyzko, A. M. & Wolfram, S. [1984] ≥ |xi−1 − xi |2−(i+1) (A.3)
“Algebraic properties of cellular automata,” Com- i=0
mun. Math. Phys. 93, 219–258. which can be analyzed separately. However, in both
Moore, E. F. [1962] “Machine models of self- cases we will make use of the well-known triangle
reproduction,” Proc. Symp. Appl. Math. 14, 17–33. inequality
Myhill, J. [1963] “The converse of Moore’s Garden-of-
Eden theorem,” Proc. Amer. Math. Soc. 14, 685–686. |a1 | + |a2 | ≥ |a1 + a2 | ⇒ |a1 | + |a2 | + · · · + |aN |
Ohi, F. [2007] “Chaotic properties of the elementary cel- N

lular automaton rule 40 in Wolfram’s class I,” Compl.
≥ ai (A.4)
Syst. 17, 295–308.
i=1
Pivato, M. & Yassami, R. [2008] “The spatial structure
of odometers in certain cellular automata,” Journ’ees The proof for inequality (A.2) follows directly
Automates Cellulaires 1, 119–126. from (A.4).
As for (A.3), it is equivalent to scale-free order. If o(L) = s(L), then L can be

I uniquely expressed as
I

1≥ xi−1 2−(i+1) + xi 2−(i+1) − 1 n
2o(ξL ) − 1 i·0(ξL )
L= 2 (B.3)
i=0 i=0
ξL
i=0
I

+ |xi−1 − xi |2−(i+1) (A.5) where n = (o(L)/o(ξL )) − 1.
i=0
Proof. Since o(L) = s(L), the scale-factor order
Applying (A.4) to the second part of inequality
becomes
(A.5), we find
I TL 2s(L) − 1 2o(L) − 1
I
ξL = = = (B.4)
L L L
xi−1 2−(i+1) + xi 2−(i+1) − 1
and, by the definition of multiplicative suborder, it
i=0 i=0
I follows that
I

−(i+1) −i
+ |xi−1 − xi |2 ≥ xi 2 − 1 o(ξL L) = o(2o(L) − 1) = o(L) (B.5)

i=0 i=0
(A.6) But, from Lemma B.2,
o(ξL )o(L)
therefore the inequality (A.5) is equivalent to o(ξL L) = lcm(o(ξL )o(L)) = (B.6)
I p

−i where p|o(L), which means that
1≥ xi 2 − 1 (A.7)

i=0 o(L) = p · q (B.7)

But xi ∈ {0, 1} and max( Ii=0 xi 2−i ) = 1; this Then, combining Eqs. (B.5) and (B.6), we obtain
means that (A.7) holds, and consequently the
inequality (A.3) holds. o(ξL )o(L)
o(L) = ⇒ o(ξL ) = p (B.8)
p
and then we also find an expression for q, because
Appendix B o(L) o(L)
q= = (B.9)
In order to prove Theorem 6.4 (scale-free decom- p o(ξL )
position), presented in Sec. 6 and recalled in the
By using Lemma B.1, we can easily find that
following, we need to introduce two lemmas from
number theory, whose proof can be found in any 2o(L) − 1 = (2p − 1)(1 + 2p + · · · + 2(q−1) · p )
good book on number theory.
= (2o(ξ) − 1)(1 + 2o(ξ) + · · · + 2(q−1) · o(ξ) )
Lemma B.1. For n ∈ N, n = ab
(B.10)
2n − 1 = 2ab − 1
where we used Eq. (B.8)
a a
= (2 − 1)(1 + 2 + 2 2a
+ ··· + 2
(b−1)a
)
2o(ξL ) − 1
(B.1) 1= (1 + 2p + · · · + 2(q−1) · p ) ⇒
2o(L) − 1

Lemma B.2. Let o(n) be the multiplicative order of 2o(ξL ) − 1 p ( o(L) −1) · p
L= 1 + 2 + ··· + 2 o(ξ) (B.11)
2 (mod n), n = ab, then ξL
o(n) = lcm(o(a), o(b)) (B.2) or, in other words,
Now, we are ready to prove the following theorem. 2o(ξL ) − 1

L= 1 + 2o(ξL ) + · · · + 2n · o(ξL ) (B.12)
ξL
Theorem (Scale-free decomposition). Let L be the
length of the bit strings, L odd, and let ξL be its where n = (o(L)/o(ξL )) − 1.
Appendix C {0, 1} [Hedlund, 1969; Courbage, 1999; Courbage &

The Classical Definitions and Yasmineh, 2001; Pivato & Yassami, 2008].
Definitions Based on Boolean Cubes
Theorem C.2. The classical definition and defini-
of Permutive Rules are Equivalent tion based on Boolean cubes of right-permutive rules
are equivalent.
Definition C.1. A local rule N is said to be Left-
Permutive, iff for every x̃0 , x̃1 ∈ {0, 1} the map
Proof. Following the same steps as in the previous
f N (x−1 , x̃0 , x̃1 ) is a permuting map of the alphabet
proof, a right-permutive rule can be written in the
{0, 1} [Hedlund, 1969; Courbage, 1999; Courbage & form [Wuensche & Lesser, 1992, p. 30]
Yasmineh, 2001; Pivato & Yassami, 2008].
(0) (1) (2) (3) (4) (5) (6) (7)
Theorem C.1. The classical definition and defini-
000 001 010 011 100 101 110 111
tion based on Boolean cubes of left-permutive rules
↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓
are equivalent.
X0 X0 X2 X2 X4 X4 X6 X6
(C.2)
Proof. A left-permutive rule can be written in the
form [Wuensche & Lesser, 1992, p. 30] Numbers within brackets refer to the correspond-
ing vertices of Boolean cubes in Sec. 1. It is evident
(0) (1) (2) (3) (4) (5) (6) (7)
that it is necessary to define only four outputs —
000 001 010 011 100 101 110 111
namely X0 , X2 , X4 and X6 — and the remain-
↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓
ing four outputs will be automatically defined, since
X0 X1 X2 X3 X0 X1 X2 X3
X1 = X 0 = 1 − X0 , X3 = X 2 = 1 − X2 , X5 = X 4 =
(C.1) 1 − X4 , X7 = X 6 = 1 − X6 , where Xi ∈ {0, 1}.
Straightforwardly, we can conclude that the
Numbers within brackets refer to the correspond-
definition of left-permutive rule given in Sec. 1,
ing vertices of Boolean cubes in Sec. 1. It is evident
based on the opposite colors of the vertex pairs
that it is necessary to define only four outputs —
{(0), (1)}, {(2), (3)}, {(4), (5)}, and {(6), (7)} of the
namely X0 , X1 , X2 and X3 — and the remain-
Boolean cubes, is equivalent to the classical defini-
ing four outputs will be automatically defined, since
tion of expression (C.2).
X4 = X 0 = 1 − X0 , X5 = X 1 = 1 − X1 , X6 = X 2 =
1 − X2 , X7 = X 3 = 1 − X3 , where Xi ∈ {0, 1}.
Straightforwardly, we can conclude that the Example. Rule 86 = T † 30 has the following fir-
definition of left-permutive rule given in Sec. 1, ing/quenching patterns
based on the opposite colors of the vertex pairs
{(0), (4)}, {(1), (5)}, {(2), (6)}, and {(3), (7)} of the (0) (1) (2) (3) (4) (5) (6) (7)
Boolean cubes, is equivalent to the classical defini- 000 001 010 011 100 101 110 111
86 :
tion of expression (C.1). ↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓
0 1 1 0 1 0 1 0
Example. Rule 30 , whose Boolean cube is shown
in Fig. 8(a), has the following firing/quenching then, according to (C.2), it is a right-permutive rule.
patterns
Definition C.3. A local rule N is said to be
(0) (1) (2) (3) (4) (5) (6) (7) Bi-Permutive, iff both maps f N (x−1 , x̃0 , x̃1 ) and
000 001 010 011 100 101 110 111 f N (x̃−1 , x̃0 , x1 ) are permuting maps of the alpha-
30 :
↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓
bet {0, 1} for every x̃−1 , x̃0 ∈ {0, 1} and x̃0 , x̃1 ∈
0 1 1 1 1 0 0 0
{0, 1} [Courbage, 1999; Courbage & Yasmineh,
then, according to (C.1), it is a left-permutive rule. 2001; Pivato & Yassami, 2008].
Definition C.2. A local rule N is said to be Right- Theorem C.3. The classical definition and the defi-
Permutive, iff for every x̃−1 , x̃0 ∈ {0, 1} the map nition based on Boolean cubes of bi-permutive rules
f N (x̃−1 , x̃0 , x1 ) is a permuting map of the alphabet are equivalent.
Proof. From expressions (C.1) and (C.2), we find that bi-permutive rules have the form
(0) (1) (2) (3) (4) (5) (6) (7)
000 001 010 011 100 101 110 111
(C.3)
↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓
X0 X0 X2 X2 X0 X0 X2 X2
Numbers within brackets refer to the corresponding vertices of Boolean cubes in Sec. 1. It is evident that
it is necessary to define only two outputs — namely X0 and X2 — and the remaining six outputs will be
automatically defined, since X1 = X4 = X 0 = 1 − X0 , X3 = X6 = X 2 = 1 − X2 , X5 = X0 , X7 = X2 ,
where Xi ∈ {0, 1}.
Straightforwardly, we can conclude that the definition of bi-permutive rule given in Sec. 1, based on the
fact that vertices of the Boolean cubes can be divided into two subsets with opposite colors {(0), (3), (5), (6)}
and {(1), (2), (4), (7)}, is equivalent to the classical definition of expression (C.3).
There are only four bi-permutive rules: 90 , 105 , 150 , and 105 .
(0) (1) (2) (3) (4) (5) (6) (7)
000 001 010 011 100 101 110 111
90 :
↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓
0 1 0 1 1 0 1 0
(0) (1) (2) (3) (4) (5) (6) (7)
000 001 010 011 100 101 110 111
105 :
↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓
1 0 0 1 0 1 1 0
(0) (1) (2) (3) (4) (5) (6) (7)
000 001 010 011 100 101 110 111
150 :
↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓
0 1 1 0 1 0 0 1
(0) (1) (2) (3) (4) (5) (6) (7)
165 : 000 001 010 011 100 101 110 111
↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓
1 0 1 0 0 1 0 1

Part9 Vol18 No9 2008

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Part9 Vol18 No9 2008

Uploaded by

Copyright:

Available Formats

Tutorials and Reviews

A NONLINEAR DYNAMICS PERSPECTIVE

LEON O. CHUA, GIOVANNI EGIDIO PAZIENZA∗ , LASZLO ORZO†,

Enginyeria i Arquitectura La Salle, Universitat Ramon Llull,

Computer and Automation Research Institute,

Received December 12, 2007; Revised May 15, 2008

Keywords: Cellular automata; quasi-ergodicity; ergodicity; nonlinear dynamics; attractors; Isles

1.3.One formula speciﬁes all 256 rules . . . . . . . . . . . . . . . . . . . . . . . . 2497

96 97 98 99 100 101 102 103

104 105 106 107 108 109 110 111

112 113 114 115 116 117 118 119

120 121 122 123 124 125 126 127

128 129 130 131 132 133 134 135

136 137 138 139 140 141 142 143

144 145 146 147 148 149 150 151

152 153 154 155 156 157 158 159

160 161 162 163 164 165 166 167

168 169 170 171 172 173 174 175

176 177 178 179 180 181 182 183

184 185 186 187 188 189 190 191

192 193 194 195 196 197 198 199

200 201 202 203 204 205 206 207

208 209 210 211 212 213 214 215

216 217 218 219 220 221 222 223

224 225 226 227 228 229 230 231

232 233 234 235 236 237 238 239

240 241 242 243 244 245 246 247

248 249 250 251 252 253 254 255

= [−1 + 2.5] 1.4. Space-time pattern and time-τ

94 104 105 106 108 110 122 126

128 130 132 134 136 138 140 142

146 150 152 154 156 160 162 164

168 170 172 178 184 200 204 232

κ = 1 (Red) 104 rules, κ = 2 (Blue) 126 rules, κ = 3 (Green) 26 rules

1.7. Symmetries among Boolean cubes

1.7.1. Local complementation T c

1.7.2. Three equivalence transformations

: List of 104 Linearly-Separable Boolean Function Rules.

80 81 84 85 87 93 95 112 113 115 117 119 127

Color Period-1 N Period-2 N Bernoulli N

To derive the left-right transformation N † pairs, respectively,

Table 6. Some Boolean cubes and their local complements.

30 225 110 145

45 210 150 105

60 195 154 101

105 150 184 71

Fig. 7. Geometrical constructions for deriving (a) N † T † (N ), (b) N T (N ), and (c) N ∗ T ∗ (N ).

Table 7. A gallery of 16 centrally-symmetric local rules.

129 153 165 189

195 219 231 255

N T (N ) = T C (N ) N C (13) 1.7.4. Permutive rules9

(a) (b) (c)

Table 8. Sixteen left-permutive rules.

Remarks 1.7.7. Most rules harbor at least one

Table 9. Sixteen right-permutive rules.

Table 11. Twenty-eight permutive rules.

1 2 4 8 16 32 64 128 1 2 4 8 16 32 64 128 1 2 4 8 16 32 64 128

1 2 4 8 16 32 64 128 1 2 4 8 16 32 64 128 1 2 4 8 16 32 64 128

1 2 4 8 16 32 64 128 1 2 4 8 16 32 64 128 1 2 4 8 16 32 64 128

134 2 4 128 156 4 8 16 128 178 2 16 32 128

138 2 8 128 160 32 128 182 2 4 16 32 128

139 1 2 8 128 161 1 32 128 183 1 2 4 16 32 128