Professional Documents
Culture Documents
International Journal of Bifurcation and Chaos, Vol. 18, No. 9 (2008) 2487–2642
c World Scientific Publishing Company
RAN, Russia
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.
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. 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 .
A Nonlinear Dynamics Perspective of Wolfram’s New Kind of Science. Part IX 2491
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
2492 L. O. Chua et al.
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
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
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
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
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
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
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
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
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
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
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
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
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
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
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
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
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
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
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
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
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.
A Nonlinear Dynamics Perspective of Wolfram’s New Kind of Science. Part IX 2497
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)
2498 L. O. Chua et al.
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
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
Table 2. (Continued )
2500
Table 2. (Continued )
2501
Table 2. (Continued )
2502
Table 2. (Continued )
2503
Table 2. (Continued )
2504
Table 2. (Continued )
2505
Table 2. (Continued )
2506
Table 2. (Continued )
2507
Table 2. (Continued )
2508
Table 2. (Continued )
2509
Table 2. (Continued )
2510
Table 2. (Continued )
2511
Table 2. (Continued )
2512
Table 2. (Continued )
2513
Table 2. (Continued )
2514
A Nonlinear Dynamics Perspective of Wolfram’s New Kind of Science. Part IX 2515
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].
2516 L. O. Chua et al.
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
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
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.
A Nonlinear Dynamics Perspective of Wolfram’s New Kind of Science. Part IX 2519
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.
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
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
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)
2520 L. O. Chua et al.
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
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
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
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
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
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
1.7.3. 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
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 Nonlinear Dynamics Perspective of Wolfram’s New Kind of Science. Part IX 2523
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.
2524 L. O. Chua et al.
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.
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.
2526 L. O. Chua et al.
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.
A Nonlinear Dynamics Perspective of Wolfram’s New Kind of Science. Part IX 2527
Fig. 9. (Continued )
Table 10. Four bi-permutive rules.
2528
A Nonlinear Dynamics Perspective of Wolfram’s New Kind of Science. Part IX 2529
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
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
Table 13. (Continued )
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
Table 13. (Continued )
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
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
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
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
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
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
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
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
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
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
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
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
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
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
2 3 There are 64 Local Rules whose Boolean Cube has a
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
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
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
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
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
2 3 There are 64 Local Rules whose Boolean Cube has a
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
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
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
2 3 There are 64 Local Rules whose Boolean Cube has a
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
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
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
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
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
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
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
2 3 There are 16 Local Rules whose Boolean Cube has a
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
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
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
2 3 There are 16 Local Rules whose Boolean Cube has a
6
0
7
1
type H 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
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
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
2 3 There are 16 Local Rules whose Boolean Cube has a
6
0
7
1
type I 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
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
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
2 3 There are 16 Local Rules whose Boolean Cube has a
6
0
7
1
type J 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
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
A Nonlinear Dynamics Perspective of Wolfram’s New Kind of Science. Part IX 2543
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.
(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.
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.
2546 L. O. Chua et al.
(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.
2548 L. O. Chua et al.
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
A Nonlinear Dynamics Perspective of Wolfram’s New Kind of Science. Part IX 2549
41 L=601 45 L=101
54 L=201 60 L=101
2550 L. O. Chua et al.
73 L=80 73 L=94
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
Table 17. (Continued )
2554
Table 17. (Continued )
2555
Table 17. (Continued )
2556
Table 17. (Continued )
2557
Table 17. (Continued )
2558
Table 17. (Continued )
2559
Table 17. (Continued )
2560
Table 17. (Continued )
2561
2562 L. O. Chua et al.
150 , 154 — will henceforth be called strongly second, the Isles of Eden; and third, the Gardens
quasi-ergodic.11 of Eden.
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 Nonlinear Dynamics Perspective of Wolfram’s New Kind of Science. Part IX 2563
(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.
2564 L. O. Chua et al.
Table 18. Non-strongly quasi-ergodic rules have compact sets of Gardens of Eden.
A Nonlinear Dynamics Perspective of Wolfram’s New Kind of Science. Part IX 2565
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].
2566 L. O. Chua et al.
. .
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
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).
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.
A Nonlinear Dynamics Perspective of Wolfram’s New Kind of Science. Part IX 2569
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
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)
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.
A Nonlinear Dynamics Perspective of Wolfram’s New Kind of Science. Part IX 2571
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.
2572 L. O. Chua et al.
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)
Fig. 19. Fractality of χ1 : The four fractal patterns appear in two arbitrary subintervals of φ ∈ [0, 1).
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)
Fig. 20. Time-1 characteristic function of rule 90 , case xI = 0: The fractal pattern χα is in red and the fractal pattern χβ
is in purple.
A Nonlinear Dynamics Perspective of Wolfram’s New Kind of Science. Part IX 2573
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
2574 L. O. Chua et al.
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.
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)
Fig. 22. Fractality of χ1 : The four fractal patterns appear in two arbitrary subintervals of φ ∈ [0, 1).
90
A Nonlinear Dynamics Perspective of Wolfram’s New Kind of Science. Part IX 2575
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)
Fig. 23. Time-1 characteristic function of rule 105 , case xI = 0: The fractal pattern χα is in red and the fractal pattern χβ
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.
2576 L. O. Chua et al.
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)
Fig. 25. Fractality of χ1 : The four fractal patterns appear in two arbitrary subintervals of φ ∈ [0, 1).
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)
Fig. 26. Time-1 characteristic function of rule 150 , case xI = 0: The fractal pattern χα is in red and the fractal pattern χβ
is in purple.
2578 L. O. Chua et al.
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,
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
A Nonlinear Dynamics Perspective of Wolfram’s New Kind of Science. Part IX 2579
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
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)
Fig. 28. Fractality of χ1 : The four fractal patterns appear in two arbitrary subintervals of φ ∈ [0, 1).
150
2580 L. O. Chua et al.
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
2582 L. O. Chua et al.
. .. 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
(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
A Nonlinear Dynamics Perspective of Wolfram’s New Kind of Science. Part IX 2583
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.
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.
2584 L. O. Chua et al.
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)
A Nonlinear Dynamics Perspective of Wolfram’s New Kind of Science. Part IX 2585
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.
2586 L. O. Chua et al.
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
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
A Nonlinear Dynamics Perspective of Wolfram’s New Kind of Science. Part IX 2587
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
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
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
6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5
6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5
6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5
6 7 6 7 6 7 6 7
0 1 0 1 0 1 0 1
4 5 4 5 4 5 4 5
A Nonlinear Dynamics Perspective of Wolfram’s New Kind of Science. Part IX 2589
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).
2590 L. O. Chua et al.
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.
A Nonlinear Dynamics Perspective of Wolfram’s New Kind of Science. Part IX 2591
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.
15
This theorem is a summary of the results presented in several lemmas and theorems in [Ohi, 2007] using a different
terminology and notation.
2592 L. O. Chua et al.
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 .
A Nonlinear Dynamics Perspective of Wolfram’s New Kind of Science. Part IX 2593
nτ = n0 · 2σ (mod 2L − 1) (44)
Fig. 44. Basin-tree diagram of the Bernoulli στ -shift attractor with σ = 1, τ = 4, for rule 110 and L = 7.
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
A Nonlinear Dynamics Perspective of Wolfram’s New Kind of Science. Part IX 2595
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
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.
2596 L. O. Chua et al.
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:
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
19
This cube shows the connection among additive rules, and it should not be confused with the 256 Boolean cubes in Table 1.
2598 L. O. Chua et al.
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
A Nonlinear Dynamics Perspective of Wolfram’s New Kind of Science. Part IX 2599
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.
2600 L. O. Chua et al.
(a)
(b)
Fig. 46. Geometrical interpretation of additive (a) and anti-additive (b) rules.
A Nonlinear Dynamics Perspective of Wolfram’s New Kind of Science. Part IX 2601
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
2602 L. O. Chua et al.
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
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
Table 31. (Continued )
2605
Table 31. (Continued )
2606
Table 31. (Continued )
2607
Table 31. (Continued )
2608
Table 31. (Continued )
2609
Table 32. Data for generating and verifying the period of period-T orbits of 90 .
2610
Table 32. (Continued )
2611
Table 32. (Continued )
2612
Table 32. (Continued )
2613
Table 32. (Continued )
2614
Table 32. (Continued )
2615
Table 33. Data for generating and verifying the period of period-T orbits of 105 .
2616
Table 33. (Continued )
2617
Table 33. (Continued )
2618
Table 33. (Continued )
2619
Table 33. (Continued )
2620
Table 33. (Continued )
2621
Table 34. Data for generating and verifying the period of period-T orbits of 150 .
2622
Table 34. (Continued )
2623
Table 34. (Continued )
2624
Table 34. (Continued )
2625
Table 34. (Continued )
2626
Table 34. (Continued )
2627
2628 L. O. Chua et al.
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 .
A Nonlinear Dynamics Perspective of Wolfram’s New Kind of Science. Part IX 2629
Fig. 48. Relationship between the length L of the bit strings and the maximum period T of the attractors (symbol ) for
rule 90 .
2630 L. O. Chua et al.
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 .
A Nonlinear Dynamics Perspective of Wolfram’s New Kind of Science. Part IX 2631
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 .
2632 L. O. Chua et al.
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,
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, …
23
We assume that o(1) = 1, as illustrated in Example 6.2.
A Nonlinear Dynamics Perspective of Wolfram’s New Kind of Science. Part IX 2637
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
2638 L. O. Chua et al.
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.
A Nonlinear Dynamics Perspective of Wolfram’s New Kind of Science. Part IX 2639
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).
2640 L. O. Chua et al.
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.
2642 L. O. Chua et al.
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