April 5, 2018 10:17 WSPC/S0218-1274 1830008

International Journal of Bifurcation and Chaos, Vol. 28, No. 3 (2018) 1830008 (14 pages)

c World Scientific Publishing Company
DOI: 10.1142/S0218127418300082

A Study of Chaos in Cellular Automata

Supreeti Kamilya∗ and Sukanta Das†

Department of Information Technology,
Indian Institute of Engineering Science and Technology,
Shibpur, Howrah, West Bengal 711103, India
∗kamilyasupreeti779@gmail.com
†sukanta@it.iiests.ac.in

Received February 16, 2018

Int. J. Bifurcation Chaos 2018.28. Downloaded from www.worldscientific.com

This paper presents a study of chaos in one-dimensional cellular automata (CAs). The commu-
by UTRECHT UNIVERSITY on 04/12/18. For personal use only.

nication of information from one part of the system to another has been taken into consideration
in this study. This communication is formalized as a binary relation over the set of cells. It is
shown that this relation is an equivalence relation and all the cells form a single equivalence
class when the cellular automaton (CA) is chaotic. However, the communication between two
cells is sometimes blocked in some CAs by a subconfiguration which appears in between the cells
during evolution. This blocking of communication by a subconfiguration has been analyzed in
this paper with the help of de Bruijn graph. We identify two types of blocking — full and par-
tial. Finally a parameter has been developed for the CAs. We show that the proposed parameter
performs better than the existing parameters.

Keywords: Cellular Automata (CAs); chaos; Rule Min Term (RMT); parametrization.

1. Introduction study, he considered a special type of binary CAs,

A Cellular Automaton (CA) consists of an infinite called Elementary CAs (ECAs). Based on his obser-
number of identical cells that are organized as a reg- vation, the ECAs are classified into four types —
ular network over a D-dimensional cellular space. simple, periodic, chaotic and complex.
It is a discrete time dynamical system [Kari, 2005], In the second research direction, the CAs are
and can be studied in terms of chaos theory. A good studied in terms of topological dynamics. Since
number of works have already been reported deal- there are different definitions of chaos, such as
ing with the chaotic behavior of cellular automata Devaney chaos [Devaney, 1992], Knudsen chaos
(CAs) [Wolfram, 1984; Langton, 1990; Martinez, [Knudsen, 1994], Li–Yorke chaos [Arai, 2017], etc.,
2013; Wuensche, 1998; Kůrka, 2008; Li & Packard, the CAs are studied considering a standard defini-
1990; Cattaneo et al., 1999]. However, most of the tion; see for example [Kůrka, 2008; Schüle & Stoop,
works deal with one-dimensional CAs. This work 2012; Cattaneo et al., 1999; Blanchard et al., 1997].
also focuses on one-dimensional cases. In many of these studies, the CAs are also classified
There are three research directions that are like those by Wolfram. However, all the classifica-
found in the literature of chaos and CAs. The tions do not always converge. It is pointed out in
first direction is due to Wolfram, who studied the [Cattaneo et al., 1999] that, some CAs which are
chaotic behavior in CAs by observing the evolution marked chaotic according to some of the definitions
of the systems [Wolfram, 1984]. In this experimental are not actually chaotic.

1830008-1
 April 5, 2018 10:17 WSPC/S0218-1274 1830008

S. Kamilya & S. Das

It has also been shown that given an arbi- Finally, exploiting the idea of information prop-
trary CA, in one or more dimension, it is not pos- agation and information cooking, we propose a
sible to decide whether the CA is chaotic or not parameter to predict about the chaotic nature of
[Durand et al., 2003]. In spite of this fact, some a given CA (see Sec. 6). This parameter can cap-
researchers have developed some parameters to pre- ture the intuitive understanding of chaos in CAs.
dict the behavior of CAs. Some well known param- We claim that this parameter is better than all
eters are λ-parameter [Langton, 1990], mean field other existing parameters. We have examined all
curve [McIntosh, 1990], Z-parameter [Wuensche, the ECAs to establish our claim.
1998], etc. In many cases the parameters succeed To proceed further, we next present some useful
to predict the dynamics of CAs, and in some cases definitions and terminologies.
they fail. This is the third research direction, which
has attracted many researchers.
Under this scenario, we undertake this research 2. Definitions and Terminologies
to study the chaotic behavior of one-dimensional 2.1. Cellular automata
CAs. Like the third research direction, we finally
target to develop a parameter that will predict The cellular automata, we consider here, are one-
Int. J. Bifurcation Chaos 2018.28. Downloaded from www.worldscientific.com

the behavior of CAs. As we perceive about chaotic dimensional, and consist of infinitely many cells
which are placed over Z. A cell can assume a state
by UTRECHT UNIVERSITY on 04/12/18. For personal use only.

systems, they are unpredictable. In fact, based on

this intuitive understanding of chaos, the authors from a set S = {0, 1, . . . , d − 1} and it moves to its
of [Cattaneo et al., 1999] have criticized some next state depending on the present states of its left
CAs which have been declared as chaotic following and right neighbors and itself. Therefore, we have
Devaney’s definition. We also explore this unpre- the next state (NS) function f : S 3 → S for a cell.
dictability behavior of chaotic systems in our study. The function f induces the global transition func-
In a chaotic system, information can flow from tion F : S Z → S Z which can be defined as, for any
a part of the system to another. However, only i ∈ Z, F (x)i = f (xi−1 , xi , xi+1 ). A configuration (c)
the information flow does not always make a CA is a collection of states of the cells of a CA at a given
chaotic. As shown in some works (see for example time, that is, c = (si )i∈Z where si is the state of cell
[Wolfram, 1984; Cattaneo et al., 1999]), informa- i. During evolution, therefore, a CA hops from one
tion flows from one side to another in few CAs, configuration to another.
but they are not unpredictable, hence not consid- The function f can be represented in tabu-
ered as chaotic. To address this issue, we divide lar form. Table 1 shows three such functions with
the information flow into two parts — information S = {0, 1}. These CAs are commonly called as ele-
propagation and information cooking. Information mentary cellular automata (ECAs). The function
propagation notes how a change in a cell affects its (f ) which is represented as d-ary (binary in case of
neighbors, whereas information cooking points out ECAs) or the equivalent decimal numbers, is called
how the cell itself is being changed in response to “rule”.
the updates of its neighbors. Both these together
Definition 1 (Rule Min Term (RMT)) [Bhattachar-
make the system unpredictable. We discuss this
jee & Das, 2016]. Let f : S 3 → S be the next state
issue in Sec. 3.
function of a CA. The tuple x1 , x2 , x3  ∈ S 3 is
Based on the information propagation and
called a Rule Min Term or RMT, which is gen-
information cooking, we define a binary relation
erally represented by its decimal equivalent r =
communicate, over the set of cells of a CA (see
x1 · d2 + x2 · d + x3 .
Sec. 5). This relation is an equivalence relation, and
all the cells of a CA form a single equivalence class
when the CA is chaotic. However, it is found that Table 1. The local transitions of ECAs rules 60, 90 and 150.
in some CAs, information flow between two cells is PS 111 110 101 100 011 010 001 000
blocked (fully or partially) if a blocking word exists in RMT (7) (6) (5) (4) (3) (2) (1) (0) Rule
between these two cells. Existence of blocking words
is a good reason for the CA to be nonchaotic. We f 0 0 1 1 1 1 0 0 60
f 0 1 0 1 1 0 1 0 90
have identified these blocking words in Sec. 4 by
f 1 0 0 1 0 1 1 0 150
exploring de Bruijn graph.

1830008-2
 April 5, 2018 10:17 WSPC/S0218-1274 1830008

A Study of Chaos in Cellular Automata

The first row of Table 1 shows the combinations rule are said to be sibling to each other if  sd1  =
of the present states (PS) of the neighbors of a cell.  sd2  = · · · =  sdd  (mod d3 ).
Each such combination is the RMT. For a d-state
CA, total number of RMTs is d3 . Hence, number In this paper, the sets of equivalent and sibling
of RMTs, in case of an ECA, is 23 = 8 and num- RMTs are presented as — Equik and Siblj respec-
ber of rules is 28 (= 256). We often omit commas tively. Equik = {k, d2 +k, 2d2 +k, . . . , (d−1)d2 +k},
within an RMT, if it does not lead to any confusion. (0 ≤ k ≤ d2 − 1). That is, the set Equik contains
That is, an RMT x1 , x2 , x3  is generally presented RMT k and all of its equivalent RMTs. Whereas,
as x1 x2 x3 . Siblj = {d · j, d · j + 1, . . . , d · j + (d − 1)} (0 ≤ j ≤
A CA, however, can also be represented by a de d2 −1) [Bhattacharjee & Das, 2016]. For a two-state
Bruijn graph. CA, Equi0 = {000, 100} and Sibl0 = {000, 001}. In
case of a three-state CA, Equi0 = {000, 100, 200}
Definition 2 (de Bruijn Graph). Let Σ be a set of and Sibl0 = {000, 001, 002}.
symbols, and s ≥ 1 be a number. Then, the de Note that, in the de Bruijn graph B(2, {0, 1})
Bruijn graph is B(s, Σ) = (V, E), where V = Σs (Fig. 1), all the RMTs associated with the outgoing
is the set of vertices, and E = {(ax, bx) | a, b ∈ Σ, edges of a node are sibling to each other and the
Int. J. Bifurcation Chaos 2018.28. Downloaded from www.worldscientific.com

x ∈ σ s−1 } is the set of edges. RMTs associated with the incoming edges of a node
by UTRECHT UNIVERSITY on 04/12/18. For personal use only.

are the equivalent RMTs of a CA.

A de Bruijn graph consists of ms vertices, where Apart from the above classes of RMTs (equiv-
m is the number of symbols in Σ. The graph is bal- alent and sibling), we design other sets of RMTs.
anced in the sense that each vertex has both in- These sets are designed with respect to given
degree and out-degree s. For our CAs, s = 2 and RMTs. We name the sets R-set(r) and L-set(r) for
Σ = S. The edges (ax, xb) of B(2, S) show the over- a given RMT r.
lap of nodes and (axb) ∈ S 3 . To represent a CA by
de Bruijn graph, we have marked each edge (ax, xb) Definition 5 (L-set). Let r be the decimal equiv-
of B(2, {0, 1}) by axb/f (axb) ∈ S, where f is the alent of RMT xyz where x, y, z ∈ S. Then, L-
local rule. set(r) = {RMT xy  z  | y  = y and z  = z, y  , z  ∈ S}.
Figure 1 is the de Bruijn graph of ECA rule 90.
This graph shows that if the left, self and right Definition 6 (R-set). Let r be the decimal equiv-
neighbors of a cell are all 0s, then the next state alent of RMT xyz where x, y, z ∈ S. Then, R-
of the cell (that is, f (0, 0, 0)) is 0, if the neighbors set(r) = {RMT x y  z | x = x and y  = y, x , y  ∈ S}.
are 0, 0 and 1 respectively, the next state is 1, and
We get L-set and R-set for every RMT of a CA.
so on.
For example, In case of a three-state CA, the L-set
Definition 3 (Equivalent RMT) [Bhattacharjee & for RMT 0 is L-set(0) = {011, 012, 021, 022} and the
Das, 2016]. A set of d RMTs r1 , r2 , . . . , rd of a d- R-set for RMT 0 is R-set(0) = {110, 120, 210, 220}.
state CA rule are said to be equivalent to each other
Definition 7. An RMT xyz (x, y, z ∈ S) of a CA
if r1 d ≡ r2 d ≡ · · · ≡ rd d (mod d3 ).
rule is called passive if f (xyz) = y. On the other
Definition 4 (Sibling RMT) [Bhattacharjee & Das, hand, if f (xyz) = y  (y = y  , y  ∈ S), it is an active
2016]. A set of d RMTs s1 , s2 , . . . , sd of a d-state CA RMT.

For example, in ECA rule 60, the RMTs 0

(000), 1 (001), 2 (010) and 3 (011) are passive (see
01 Table 1), whereas the rest of the RMTs are active.
001/1 011/1

010/0 101/0
000/0 00 11 111/0 2.2. Cellular automata and chaos
The global transition function F of a CA is a con-
100/1 110/1 tinuous map on the compact metric space S Z . This
10 implies, cellular automata are discrete time dynam-
ical systems, and can be studied in terms of chaos
Fig. 1. de Bruijn graph for two-state CA with rule 90. theory.

1830008-3
 April 5, 2018 10:17 WSPC/S0218-1274 1830008

S. Kamilya & S. Das

Many definitions of chaos, however, use the 3. The Source of Unpredictability

notion of sensitivity to initial condition. In a sensi-
Chaotic systems are unpredictable in the long run.
tive CA, a (small) change in initial condition would
Knowing the current configuration of a chaotic sys-
greatly affect the CA in future. Sensitivity to ini-
tem, it is not practically possible to determine how
tial condition is recognized as a key notion in chaos
the system will behave after a good amount of time.
theory, which makes a system unpredictable. On the
In this section we explore the properties of a CA
other hand, a CA is called transitive if and only if for
which will lead to it becoming unpredictable.
any two open subsets U and V of S Z , there exists an
In a CA, the cells change their states in the next
n ∈ N so that F n (U ) ∩ V = ∅. For CAs, however,
time step depending on the present states of their
transitivity implies sensitivity to initial condition.
neighbors. When some of its neighbors change their
A CA has dense periodic points if and only if the
states, the cell may or may not be affected by that
set {x ∈ S Z | ∃ k ∈ N : F k (x) = x} of all periodic
change. That is, a cell may be sensitive or insensi-
points is a dense subset of S Z . Now we define chaos
tive to its neighbors. Let us consider a CA where all
according to Devaney [Banks et al., 1992].
of its cells are insensitive to their neighbors. That
is, any change in the states of the neighbors of a cell
Definition 8. A CA is chaotic if (1) it is transitive,
Int. J. Bifurcation Chaos 2018.28. Downloaded from www.worldscientific.com

does not affect the cell. This CA is stable, and so

(2) its temporally periodic points are dense, and (3)
predictable. Moreover, it is not sensitive to initial
by UTRECHT UNIVERSITY on 04/12/18. For personal use only.

it is sensitive to initial condition.

condition, because a small change in initial config-
Let us now define leftmost and rightmost per- uration of the CA, which implies change of states
mutive CAs, which are shown as Devaney chaotic of some cells, will not remarkably affect the future
[Cattaneo et al., 1999; Kůrka, 2008]. configurations of the CA. On the other hand, if the
cells of a CA are sensitive to their neighbors, change
Definition 9. A CA with local rule f is left- of states of some cells would affect their neighbors,
most (resp., rightmost) permutive if and only if for and eventually all the cells will be affected. In this
any two equivalent (resp., sibling) RMTs r and s, case, a small change in initial configuration will
f (r) = f (s). matter in the course of evolution of the CA. There-
fore, whether a CA is unpredictable is dependent
In other words, in a leftmost (resp., rightmost) on whether the cells are sensitive to their neighbors
permutive CA, all the RMTs of Equik (resp., Siblj ) or not.
have different next state values. However, if a cell is sensitive to its neighbors,
Apart from the Devaney’s definition, there are then there is a question about degree of sensitiv-
restrictive definitions of chaos, such as Knudsen ity — out of total possible changes of the neigh-
chaos, positively expansive chaos, etc. A CA is bors, how many times does the cell get affected.
chaotic according to the definition of Knudsen if and If the cells are more sensitive to their neighbors,
only if it has a dense orbit and is sensitive to initial the CA would be more sensitive to initial condi-
conditions. On the other hand, a CA is positively tion. Please relook at the existing works [Lang-
expansive chaotic if and only if it is transitive, has ton, 1990; Wuensche, 1998; Blanchard et al., 1997;
dense periodic points, and is positively expansive. Kůrka, 2008; Schüle & Stoop, 2012] on the chaotic
Positive expansivity is a stronger form of sensitivity. behavior of classical CAs, where one can get sup-
However, these definitions of chaos are related. port for this idea. When the parameters, such as
In an interesting discussion, Cattaneo et al. λ, Z, are developed to analytically understand the
[1999] have pointed out that many of the definitions chaotic behavior of CAs, how the cells are sensitive
of chaos, including Devaney’s definition, declare to their neighbors can be measured.
some CAs as chaotic, which is against our intuitive
understanding of chaos. The shift CAs (ECA 170,
for example) exhibit predictable behavior in some 3.1. Information propagation
sense, but are claimed as chaotic by these defini- Let us assume that the state of a cell has been
tions, whereas, a chaotic system is unpredictable, changed. Since this cell is also a neighbor of other
and the system, iteration by iteration, will need to cells, we can now examine how this update affects
destroy all the information contained in the initial its neighbors at the next time steps. This can eas-
configuration. ily be understood by looking at the sibling RMTs

1830008-4
 April 5, 2018 10:17 WSPC/S0218-1274 1830008

A Study of Chaos in Cellular Automata

(see Definition 4) and equivalent RMTs (see Defi- We claim that Λp is a measure of information
nition 3) of the given rule. If it is leftmost (resp., propagation to the left neighbors of cell i. In case
rightmost) permutive CA, any change of a cell will of rightmost permutive CA, Λp attains the highest
affect the right (resp., left) neighbor of the cell. value as 1.
One can interpret this event as the propagation of Let us now take the following examples to illus-
information. trate the calculation of Λp .
Consider that the present state of cell i ∈ Z is
Example 1. Consider a three-state CA, rule of
si , which is updated to si . If cell i − 1 reacts then it (−1)
can be said that the information about the update which is shown in Table 2. Let us calculate δ 0
will propagate to the left. Now if f (si−2 , si−1 , si ) where Sibl0 = {0, 1, 2}. For this rule, RMT 0
and f (si−2 , si−1 , si ) are same, then it is concluded and RMT 1 have different next state values. So,
(−1) (−1)
that the change of state of cell i from si to si has no δ 0 (0, 1) = 1. Similarly, δ0 (1, 2) = 1. Hence,
effect on cell i − 1. But if the cell i moves to state (−1) 1  (−1) 4
si and f (si−2 , si−1 , si ) = f (si−2 , si−1 , si ), then this A0 = 3
3 −3
δ0 (r, s) = .
6
change of cell i has an effect on its left neighbor. r,s∈{0,1,2}
r=s
Observe that the RMTs si−2 si−1 si , si−2 si−1 si and
Int. J. Bifurcation Chaos 2018.28. Downloaded from www.worldscientific.com

si−2 si−1 si are sibling to each other. Therefore, to In this way, we can find out other Aj
(−1)
s; Hence,

by UTRECHT UNIVERSITY on 04/12/18. For personal use only.

measure the effect of cell i on its left neighbor i.e. (−1)

Λp = 312 8j=0 Aj = 0.74.
to measure the flow of information to the left, we
need to observe in how many cases a set of sibling
RMTs of R give different next state values. Let Table 2. A rule of a three-state three-
(−1) neighborhood CA.
si−2 si−1 si ∈ Siblj and δ j be a binary function
over Siblj which notes the differences for two sibling PS PS
(−1) RMT NS RMT NS
RMTs. The function δj : Siblj × Siblj → {0, 1} is
222 0 110 0
defined below, where Siblj is a set of sibling RMTs.
(26) (12)
(−1) 221 2 102 1
δj (r, s) (25) (11)

220 0 101 2
1 if fi−1 (r) = fi−1 (s), r, s ∈ Siblj (24) (10)
= (1)
0 otherwise. 212 1 100 0
(23) (9)
Using this function δ(−1) , therefore, one can 211 2 022 1
understand the number of times cell i − 1 changes (22) (8)
its state when the state of cell i moves to any other 210 0 021 1
(21) (7)
d − 1 possible states. Let us use the following para-
meter to capture this fact. 202 1 020 0
(20) (6)
(−1) 1  (−1) 201 1 012 1
Aj = δj (r, s). (2)
d2 − d (19) (5)
r,s∈Siblj
200 2 011 2
r=s
(18) (4)
Hence, when the states of cells i − 2 and i − 1 122 1 010 2
are stable at si−2 and si−1 , but the state of cell i (17) (3)
(−1) 121 2 002 0
changes, then Aj measures the possibility of cell
(16) (2)
i − 1 getting affected by cell i. Now, to find out the
120 2 001 2
total possibility of information propagation to the (15) (1)
left of cell i, we use the following parameter. 112 0 000 0
(14) (0)
d2 −1
1  (−1) 111 1
Λp = Aj . (3)
d2 (13)
j=0

1830008-5
 April 5, 2018 10:17 WSPC/S0218-1274 1830008

S. Kamilya & S. Das

Example 2. Consider the ECA rule 102. To calcu- the update affects the cell itself. It is obvious that
(−1) when left neighbor or right neighbor is updated in
late the value δ0 we take the sibling RMT set
(−1) a reaction to the update of cell i, the cell i will be
Sibl0 = {0, 1}. In that case, δ0 (0, 1) = 1 as the
next state values of both the RMTs are different. further updated in the next time step in response
(−1) to the updates of its neighbors. This kind of depen-
Similarly, we can calculate the values of δ j for
all the sibling RMT sets. From these values, we get dence on neighbors is required for a CA to be unpre-
(−1) (−1) (−1) (−1) dictable. We call the updates of cell i in subsequent
A0 = A1 = A2 = A3 = 1. Finally, the
time steps in reaction to the update of cell i as Infor-
value of Λp = 1.
mation Cooking.
Now using the same rationale, however, one can To guess the impact of the update on the
find another measure, say ηp which indicates the cell itself from the rule, the RMTs of L-set and
possibility of information propagation to the right R-set of the rule are to be considered. Let us
cell i + 1 when state of cell i is changed. In this consider that the states si−1 and si are updated
case, we need to use other functions for information to si−1 and si respectively. Now if f (si−1 , si , si+1 ) =
propagation, similar to δ(−1) [Eq. (1)]. Let us use f (si−1 , si , si+1 ), then it indicates that updates of
(+1) itself and its left neighbor have an impact on cell i.
: Equik × Equik → {0, 1} for this purpose.
Int. J. Bifurcation Chaos 2018.28. Downloaded from www.worldscientific.com

δk
That is, here we consider equivalent RMT sets for One may consider this case as information flow from
by UTRECHT UNIVERSITY on 04/12/18. For personal use only.

(+1) left neighbor to the cell. Note that si−1 si si+1 and
the right cell of cell i. Hence, δ k can be calculated
si−1 si si+1 are the two RMTs of an R-set. To mea-
as,
sure the amount of impact, we define a function

(+2)
(+1) 1 if f (r) = f (s), r, s ∈ Equik δr : R-set(r) → {0, 1} by,
δ k (r, s) =

0 otherwise.
(+2)
1 if f (r) = f (s), s ∈ R-set(r)
(4) δr (s) = (7)
0 otherwise.
The number of occurrences that the cell i + 1 (+2)
Using the parameter δ r , one can now deter-
changes its state with the change of cell i is denoted
(+1) mine the number of times cell i changes with the
by B k and it can be defined as, change of cell i − 1 and cell i as follows,
(+1) 1  (+1) 1 
Bk = 2
d −d
δk (r, s). (5) B (+2) = δ (+2) (s). (8)
(d − 1)2 + 1 s∈R-set(r)
r r
r,s∈Equik
r=s
Hence, the total possibility that cell i will
Finally, the effect of cell i on cell i + 1, when only
change with the change of cells i − 1 and i can be
the state of cell i is changed, can be calculated as,
defined as,
2 −1
d
d −13
Bk
(+1) 1  (+2)
ηc = 3 Br . (9)
k=0 d
ηp = . (6) r=0
d2
Example 4. Let us illustrate the idea by ECA rule
Now ηp = 1 if the CA is leftmost permutive. (+2)
102. To know the values of δ r (s) we need to know
Example 3. Consider the ECA rule 102. With the next state values of the RMTs of R-set(r). Here,
(+2)
the help of the above equations one can calculate R-set(0) = 3. The value of δ 0 (3) = 1 because
the information flow in the right direction (ηp ). f (0) = f (3) for this rule. Similarly, we get other
For the rule 102, the value of ηp = 0. This means (+2)
values of δ r (s). Hence, we obtain the values of
the rule does not have any information flow in the (+2) (+2) (+2) (+2)
B r . Here, B 0 = B1 = · · · = B7 = 1.
right direction of the system.
Finally, we get the value of ηc as 1. That is, the
change of cells i − 1 and i affect the cell i with
3.2. Information cooking probability 1.
We have examined the effect of the update of cell Using the same rationale, one can now find the
i on cells i − 1 and i + 1. Now we examine how measure, Λc , where Λc denotes the information flow

1830008-6
 April 5, 2018 10:17 WSPC/S0218-1274 1830008

A Study of Chaos in Cellular Automata

to the left when the cells i + 1 and i are updated. Definition 10. A subconfiguration of length k ∈
Like before, we need to use another function for N of a configuration c = (si )i∈Z is a string
(−2) (si )i∈[j,j+k−1] where j ∈ Z.
information cooking, δ r : L-set(r) → {0, 1} to
measure in how many cases an L-set’s RMTs give
(−2) For example, 0110 is a subconfiguration of
different next state values. Hence, δr can be
length 4 of a configuration · · · 0011001110 · · · of a
defined as,
binary CA.


1 if f (r) = f (s), s ∈ L-set(r) Definition 11. Let X = {· · · sj sj+1 · · · sj+k−1 · · ·}
δ (−2)
r (s) =
0 otherwise. be a set of configurations of a CA. The subconfig-
(10) uration s = (si )i∈[j,j+k−1] is called a blocking word
if for any c ∈ X and any t ≥ 1, F t (c)|i = si ,
The number of times cell i changes its state ∀ i ∈ [j, j + k − 1].
with the change of cells i + 1 and i, is denoted by
(−2) In this kind of blocking, the blocking word con-
the parameter B r and it can be defined as,
tinues to stay in the configurations during evolution

Int. J. Bifurcation Chaos 2018.28. Downloaded from www.worldscientific.com

1 of the CA. The concept of blocking word is very

B (−2)
r = 2 δ (−2)
r (s). (11) close to Kurka’s work [Kůrka, 2008]. Packard et al.
(d − 1) + 1 s∈L-set(r)
by UTRECHT UNIVERSITY on 04/12/18. For personal use only.

have also used this concept under the name domain

Finally, the possibility that the left neighbor wall [Li & Packard, 1990].
of cell i gets affected by cells i and i + 1 can be Example 6. Let us consider the ECA rule 73. The
calculated as, space-time diagram of the rule (Fig. 2) shows that,
3 −1
d a subconfiguration 0110 at the initial configuration
B (−2)
r
remains the same at any time step. This means any
l=0
cell on the left (right) side of the subconfiguration
Λc = . (12) has no effect on any cell on the right (left) side of
d3
the subconfiguration. That is, 0110 blocks the infor-
Example 5. Consider the ECA rule 102. We can mation propagation from one side to another. Thus
now calculate the information cooking due to the we can say there exists a blocking word 0110 in the
left (Λc ) with the help of Eq. (12). For the rule 102, dynamics of ECA rule 73.
the value of Λc = 0. This means the change of states
of a cell and its right neighbor cannot affect the left To find out whether a blocking word exists in a
neighbor of the cell. given CA, we have used de Bruijn graph for the
given CA. Since the blocking word (si )i∈[j,j+k−1]
In case of a CA, however, all the cells are remains unchanged during evolution, the RMTs
updated in parallel. Even then these parameters acting on cells j, j + 1, . . . , j + k − 1 are passive.
give measures of the effect on the neighbors of cells. Further, any change in cell j − 1 (resp., j + k) does
If all the cells affect their neighbors, that is, the val- not affect cell j (resp., cell j + k − 1). So, a set of
ues of Λp , ηc , Λc and ηp are nonzero for each i ∈ Z, equivalent (resp., sibling) RMTs are to be passive.
then any change in any cell would make the system
unpredictable.

4. Blocking of Information
If the values of Λp , ηc , Λc and ηp are nonzero but
less than 1, then there is a chance that the infor-
mation does not flow from one side of the system
to another. In that case a blocking word may exist
in the system that separates the cells on the left
of the blocking word from the cells on its right. To
understand the blocking word, let us first define the
subconfiguration as follows. Fig. 2. Space-time diagram of rule 73.

1830008-7
 April 5, 2018 10:17 WSPC/S0218-1274 1830008

S. Kamilya & S. Das

In particular, the equivalent set is {xsj sj+1 | x ∈ S} states in the left neighbor of a cell cannot affect the
and the sibling set is {sj+k−1 sj+k x | x ∈ S}. cell but change of states of right neighbor can affect
In a de Bruijn graph, the incoming edges to the cell, the information flows to the left only and
a node are the equivalent RMTs (where only the the flow in right cells is blocked. Such a blocking
leftmost digit is changed) and the outgoing edges word can be called left blocking word. On the other
from a node are the sibling RMTs (where only the hand, if any change in right neighbor cannot affect
rightmost digits are changed). We can find out the a cell but any change in left neighbor can affect the
blocking word in the following way. cell, then it can be called right blocking word.
(1) Keep the edges of a de Bruijn graph, which rep- Definition 12. Let
resent only passive RMTs
(2) Now, if there exists a node having d incom- X = {· · · sj−1 sj sj+1 · · · sj+k−1sj+k · · ·}
ing edges and another node having d outgoing
edges, and there is a path from first node to the be a set of configurations. The subconfiguration s =
second, then the CA has a blocking word (si )i∈[j,j+k−1] is called a left blocking word if for
(3) The subconfigurations, represented by the path, each state of cell j − 1 there exist at least a state
Int. J. Bifurcation Chaos 2018.28. Downloaded from www.worldscientific.com

is the blocking word. in cell j + k for which F t (c)|i = si , where c ∈ X

and t ≥ 1, ∀ i ∈ [j, j + k − 1]. The blocking word is
by UTRECHT UNIVERSITY on 04/12/18. For personal use only.

However, a CA may have more than one blocking

a right blocking word if for each state of cell j + k
word.
there exists cell j − 1 for which F t (c)|i = si .
Example 7. Figure 3 shows a de Bruijn graph of
ECA rule 73. Here only those edges are shown that The left blocking or right blocking word can
correspond to passive RMTs. In this figure it is be called a partial blocking word. With the help of
shown that the node 01 has two incoming edges de Bruijn graph, one can also find out whether a
(here, d = 2). Node 10 has two outgoing edges. CA consists of a partial blocking word or not. Here
There exists a path from the node 01 to the node 10. also, we keep only those edges that represent passive
The path is shown by the dashed line in the figure. RMTs. To find a partial blocking word, we need to
Since a path exists between node 01 and node 10 find a node that has a number of incoming edges
through node 11, there exists a blocking word 0110 < d (resp., = d) and a node that has a number
in the ECA 73. Figure 2 also shows the existence of of outgoing edges = d (resp., < d). If there exists
blocking word 0110. In fact, (0110)k for any k ∈ N a path between these two nodes, we can obtain a
is the blocking word. partial blocking word.

Hence, we can observe that if a blocking word Example 8. Let us consider ECA rule 74. The de
exists in a system, it blocks the information to com- Bruijn graph for the same is given in Fig. 4. The
municate from one side of the system to another. node 00 has two incoming edges. This means, what-
Now we extend the idea of blocking to partial ever be the state of the left neighbor of 00, the left
blocking where a subconfiguration can only block neighbor cannot affect its right cell. Again, the end
information based on some particular neighboring is also the node 00 that has only one outgoing edge.
states of the subconfiguration. If any change of Hence, only if the state of the right neighbor of 00 is

01 01
001/0 011/1 011/1

00 101/0 11 000/0 00 101/0 11

100/0 110/1 100/0 110/1

10 10
Fig. 3. de Bruijn graph of ECA rule 73 where edges corre- Fig. 4. de Bruijn graph for rule 74 where edges correspond-
sponding to passive RMTs are given. ing to passive RMTs are given.

1830008-8
 April 5, 2018 10:17 WSPC/S0218-1274 1830008

A Study of Chaos in Cellular Automata

Therefore, the information may be communicated

from one part of the system to another for almost
all the initial configurations. Depending on the com-
munication, the next section introduces a relation
among the cells of the CA.

5. The Communication Class

and Chaos
From the previous discussions, we understand that
Fig. 5. Space-time diagram of rule 74.
any change in any cell should influence all the cells
of the CA if the probability of affecting its neighbor
is greater than 0, and there is no blocking word
0, the right neighbor cannot affect its left cell. Once
in the system. This behavior was also noted by
the state of the right neighbor of 00 is changed, it
Wolfram in his book A New Kind of Science: “any
can affect its left cell. Hence we can say the infor-
Int. J. Bifurcation Chaos 2018.28. Downloaded from www.worldscientific.com

change made anywhere in the system will almost

mation flow is in the left direction. Figure 5 shows
always eventually be communicated even to the
the space-time diagram of ECA 74. Here also we
by UTRECHT UNIVERSITY on 04/12/18. For personal use only.

most distant part of the system” [Wolfram, 2002].

can observe that the direction of flow is towards
Therefore, a change of state in a cell has to influence
the left.
all the cells to be a chaotic CA.
Thus a communication gets interrupted if a Let us now define a relation communicate over
blocking word is easily reached during its evolu- a set of cells of a CA. We represent this relation by
tion from almost all the initial configurations. Now ‘
’. That is, “i
j” means “i communicates with
the question is, can we determine the probability j”. Intuitively, this relation reflects the fact that a
with which a subconfiguration is reachable from any state change in cell i has an influence on the cell j
configuration? If the probability is high, the sys- in updating its state.
tem can easily block the communication and thus
Definition 13. A communication relation (
) is a
the behavior of the system becomes simple. Other-
relation from Z to Z where i
j holds if and only
wise, the behavior can be chaotic. However, this is a
if
well studied problem in cellular automata, known as
reachability problem — given two configurations of (1) ηp > 0 or ηc > 0 when j > i
any CA, decide whether one configuration is reach- (2) Λp > 0 or Λc > 0 when i > j
able from another. In the existing works on the (3) both are true when i = j.
reachability problems of cellular automata [Sutner,
1989; Clementi & Impagliazzo, 1995], it has been The relation
is transitive by definition: if
shown that such reachability problem is P-SPACE i
j and j
k then i
k. Whether
is reflexive
complete problem. If it would have been possible or symmetric, however, depends on the given CA. If
to solve the problem, one could decide whether a i
j but j
i, then it signifies that, any change
blocking word could be reached from almost all the in cell i can influence the state of cell j but any
configurations or not. perturbation in cell j has no effect on cell i. To be a
In space-time diagrams of CAs, it is seen that chaotic system, a tiny perturbation in initial condi-
rule 74 has a partial blocking word 00 and for tion should largely affect the future configurations
almost all the initial configurations it reaches the of the system. Therefore, for each pair of i, j ∈ Z,
word and blocks the information to flow to the right “if i
j, then j
i” has to hold in a chaotic CA.
side. Again, in case of rule 30, the partial block- This implies, the relation ‘
’ has to be an equiva-
ing word is 01 when the state of left neighbor is 1 lence relation, and all the cells have to form a single
and state of right neighbor is 0 or 1. That is, if the equivalence class. Let us call this class as commu-
state of left neighbor changes its state from 1 to 0, nication class.
the system cannot block the communication. But, Figure 6 shows the space-time diagram of ECA
for any initial configuration the probability of the rule 150. Here each cell communicates with other
subconfiguration 101 to remain unchanged is low. cells and thus all the cells are in communication

1830008-9
 April 5, 2018 10:17 WSPC/S0218-1274 1830008

S. Kamilya & S. Das

F t (U ) ∩ V = ∅. But according to the definition of

the set V , it is not possible for the set V to over-
lap with the set F t (U ) i.e. F t (U ) ∩ V = ∅. This
contradicts our assumption. Therefore we need to
consider only a single communication class to sat-
isfy the transitivity property. Hence proved. 

It is, therefore, obvious that for an arbi-

trary CA, all the cells may not form a single
communication class. Under this circumstance, let
Fig. 6. ECA rule 150 that forms a communication class. us consider that the relation
is an equivalence
relation. Then,
partitions the set of all cells into
relation among themselves. Thus this rule forms a some disjoint sets. Let us consider that Ci is a par-
communication class. tition of Z, the set of cells. If there are K such par-
K−1
To be a chaotic CA, all of its cells have to com- titions then Z = i=0 Ci .
municate among themselves. Thus communication
Int. J. Bifurcation Chaos 2018.28. Downloaded from www.worldscientific.com

However, the cells of two partitions do not

cannot be blocked by a blocking word. Hence, one communicate among themselves (as blocking word
by UTRECHT UNIVERSITY on 04/12/18. For personal use only.

of the properties of a chaotic CA can be given as exists), but each cell of a partition communicates
follows. with every other cell of the same partition. There-
fore, the dynamics of the cells of Ci are confined
Property 1. All the cells of a chaotic CA form a
by boundary walls. However, this dynamics within
single communication class.
walls can be chaotic. We call this system as locally
A CA, which is a dynamical system under its chaotic, and Ci as chaotic subsystem.
global transition function F , is topologically tran- Definition 14. A CA is called locally chaotic if
sitive if for any two nonempty open subsets U and there exists at least one communication class Ci .
V of S Z , there exists a natural number t such that
F t (U ) ∩ V = φ. If a CA is not transitive, it can- For example, consider the ECA rule 73. It cre-
not be chaotic. And, if the CA does not form a ates a block 0110 which blocks the communication
single communication class, it cannot be in general from one part to another. However, there exists
transitive. some communication classes within the boundary
walls (see Fig. 2). Hence, the CA can be said as
Proposition 1. If a CA is transitive, and
is an locally chaotic.
equivalence relation, then all of its cells form a sin- If there is exactly one communication class cov-
gle communication class. ering all cells of the CA, the system is globally
chaotic or simply chaotic which we have discussed
Proof. Let us consider that the CA is transitive. before. A locally chaotic system is not chaotic in
That is, there exist U ⊆ S Z and V ⊆ S Z such that true sense because a tiny perturbation in initial con-
F t (U ) ∩ V = ∅, ∀ t > 0. We prove this proposition dition may not greatly affect the system.
by the method of contradiction.
Without loss of generality, let us consider that
there are only two communication classes — C0 and 6. Parametrization
C1 and there exists a blocking word w that blocks Section 3 describes that if the state of a cell i is
the communication between C0 and C1 . Also con- updated, this update can affect its left or right
sider that q ∈ S is the quiescent state of the given neighbor. Again, at the next time step, the change
CA. Let us now construct two sets U and V in the may also affect cell i back. Thus a disturbance may
following way. U = {xwqqq · · · | x = (si )i∈C0 } and be created within the system if any change in any
V = {qqq · · · wy | y = (si )i∈C1 }. Here, we denote x, cell can affect its left (right) neighbor with a high
y as two configurations. Now, say u is an arbitrary probability at the next time step and the change of
element of U i.e. u ∈ U . So, for a certain value state of cells has an impact on the right (left) cells.
of t, F t (u) ∈ V because of transitivity. That is, Therefore, in order to determine the possibility of

1830008-10
 April 5, 2018 10:17 WSPC/S0218-1274 1830008

A Study of Chaos in Cellular Automata

creating disturbances to the left and right direc-

tions of the system, we have considered two param-
eters — L and R respectively.
L ≡ (Λp , max(ηp , ηc )), (13)
R ≡ (ηp , max(Λp , Λc )) (14)
where, L ≥ R holds if Λp > ηp . If Λp = ηp , then
L ≥ R holds if max(ηp , ηc ) ≥ max(Λp , Λc ).
Thus if the disturbance has been created in
either direction, the dynamics of the system will Fig. 8. ECA rule 15 where p = (1, 0).
become chaotic. Therefore, the probability that the
system will become chaotic can be defined by, p is 0, the disturbance cannot be created within the
p = max(L, R). (15) system and hence the CA is a simple CA. In fact,
the cells do not form a single communication class.
Let us say, the parameter p obtained for a CA is Figure 8 also shows that the dynamics of the CA is
Int. J. Bifurcation Chaos 2018.28. Downloaded from www.worldscientific.com

derived as (a, b). It is observed that the dynamics of simple.

CAs change from simple to chaotic if the parameter
by UTRECHT UNIVERSITY on 04/12/18. For personal use only.

value p changes from low to high. If the value of a is In the above two examples, we have seen two
0, then there is no information flow and the system extreme cases when the CAs are chaotic and simple
is simple. However, once we obtain the value of a, respectively. A number of experiments have been
then we will consider the value of b. If the value of done to observe the efficacy of the proposed param-
a is high (i.e. tends to 1) and the value of b is also eter. From a large number of experiments it is seen
high, the CA can be said to be a chaotic CA. that, if the value of a ≥ 0.75 and the value of
b ≥ 0.5, the behavior becomes chaotic for most
Example 9. In Sec. 3, we have already seen that
of the cases. Hence, if two parameter values for
the value of Λp = ηc = 1 and ηp = Λc = 0 for ECA
two CAs are given such that p1 = (1, 0.25) and
102. From Eqs. (13) and (14), we obtain L ≡ (1, 1)
p2 = (0.75, 0.75), then even though the value of a
and R ≡ (0, 1). Here, L ≥ R. Thus the probability
for p1 is large, the CA with parameter p2 is more
of disturbance in the system is, p = L. The value
chaotic than that of p1 . Some more examples show-
of p is (1, 1). Therefore, the CA is a chaotic CA.
ing the parameter values along with the dynamics
Figure 7 shows the chaotic behavior of the ECA
of the CA are given in Fig. 9.
rule 102.
The Z-parameter is probably the best parame-
Example 10. Let us consider another example of ter in determining the behavior of CAs so far [Mar-
ECA rule 15. Using the equations given in Sec. 3, tinez et al., 2012], but we claim that the proposed
one can find out the values of Λp , Λc , ηp and ηc . In parameter gives better result than the existing ones.
this case, Λp = Λc = 0 and ηp = ηc = 1. Following Table 3 shows a comparison among our proposed
Eqs. (13) and (14), one can get L ≡ (0, 1) and R ≡ parameter, λ-parameter and Z-parameter for the
(1, 0). Thus, p = (1, 0). As the second argument of minimal ECA rules. The column “Intuitively” in
the table mentions whether a CA is chaotic or not,
following the notion of chaos given in [Wolfram,
2002; Li & Packard, 1990]. Such chaotic behav-
iors are determined according to what human brain
perceives by observing the dynamics in space-time
diagrams. The bold rules in the table shows the
cases where the Z-parameter fails but our proposed
parameter p succeeds. It can be derived from the
table that, the efficiencies of λ-parameter and Z-
parameter, in this case, are 57% and 68% respec-
tively whereas the efficiency of our parameter p in
Fig. 7. ECA rule 102 where p = (1, 1). determining the behavior of a CA is 82%.

1830008-11
 April 5, 2018 10:17 WSPC/S0218-1274 1830008

S. Kamilya & S. Das

(a) Rule 1, (b) Rule 7, (c) Rule 22, (d) Rule 30, (e) Rule 35,
p = (0.25, 0.25) p = (0.75, 0.25) p = (0.75, 0.75) p = (1.0, 0.5) p = (0.25, 0.75)
Int. J. Bifurcation Chaos 2018.28. Downloaded from www.worldscientific.com
by UTRECHT UNIVERSITY on 04/12/18. For personal use only.

(f) Rule 45, (g) Rule 51, (h) Rule 54, (i) Rule 56, (j) Rule 60,
p = (1.0, 0.5) p = (0.0, 1.0) p = (0.5, 0.5) p = (0.75, 0.75) p = (1.0, 1.0)

(k) Rule 62, (l) Rule 75, (m) Rule 90, (n) Rule 107, (o) Rule 120,
p = (0.75, 0.75) p = (1.0, 0.5) p = (1.0, 1.0) p = (0.75, 0.75) p = (1.0, 0.5)

(p) Rule 131, (q) Rule 154, (r) Rule 181, (s) Rule 240, (t) Rule 251,
p = (0.75, 0.75) p = (1.0, 0.5) p = (0.75, 0.75) p = (1.0, 0.0) p = (0.25, 0.25)
Fig. 9. The values of p for some ECAs.

1830008-12
 Int. J. Bifurcation Chaos 2018.28. Downloaded from www.worldscientific.com
by UTRECHT UNIVERSITY on 04/12/18. For personal use only.

Table 3. Comparison of some parameter values for minimal ECA rules.

Minimal ECA Rules from 0 to 54 Minimal ECA Rules from 56 to 232

Is Chaotic? Is Chaotic?
Rule p λ Z p-Wise λ-Wise Z-Wise Intuitively Rule p λ Z p-Wise λ-Wise Z-Wise Intuitively

0 (0.0, 0.0) 0.0 0.0 No No No No 56 (0.75, 0.75) 0.375 0.75 Yes No Yes No
1 (0.25, 0.25) 0.125 0.25 No No No No 57 (0.5, 0.5) 0.5 0.5 No Yes No No
2 (0.25, 0.25) 0.125 0.25 No No No No 58 (0.5, 1.0) 0.5 0.5 No Yes No No
3 (0.5, 0.5) 0.25 0.5 No No No No 60 (1.0, 1.0) 0.5 1.0 Yes Yes Yes Yes
4 (0.25, 0.25) 0.125 0.25 No No No No 72 (0.5, 0.5) 0.25 0.5 No No No No
5 (0.5, 0.5) 0.25 0.5 No No No No 73 (0.75, 0.25) 0.375 0.75 No No Yes No
6 (0.5, 0.5) 0.25 0.5 No No No No 74 (0.75, 0.75) 0.375 0.75 Yes No Yes No
7 (0.75, 0.25) 0.375 0.75 No No Yes No 76 (0.25, 0.75) 0.375 0.25 No No No No
8 (0.25, 0.25) 0.125 0.25 No No No No 77 (0.5, 0.5) 0.5 0.5 No Yes No No
9 (0.5, 0.5) 0.25 0.5 No No No No 78 (0.5, 1.0) 0.5 0.5 No Yes No No
April 5, 2018

10 (0.5, 0.5) 0.25 0.5 No No No No 90 (1.0, 1.0) 0.5 1.0 Yes Yes Yes Yes
11 (0.75, 0.25) 0.375 0.75 No No Yes No 104 (0.75, 0.25) 0.375 0.75 No No Yes No
12 (0.5, 0.5) 0.25 0.5 No No No No 105 (1.0, 1.0) 0.5 1.0 Yes Yes Yes Yes
13 (0.75, 0.25) 0.375 0.75 No No Yes No 106 (1.0, 0.5) 0.5 1.0 Yes Yes Yes No
14 (0.75, 0.25) 0.375 0.75 No No Yes No 108 (0.5, 0.5) 0.5 0.5 No Yes No No
10:17

15 (1.0, 0.0) 0.5 1.0 No Yes Yes No 128 (0.25, 0.25) 0.125 0.25 No No No No
18 (0.5, 0.5) 0.25 0.5 No No No Yes 129 (0.5, 0.5) 0.25 0.5 No No No Yes
19 (0.25, 0.75) 0.375 0.25 No No No No 130 (0.5, 0.5) 0.25 0.5 No No No No
22 (0.75, 0.75) 0.375 0.75 Yes No Yes Yes 131 (0.75, 0.75) 0.375 0.75 Yes No Yes Yes
23 (0.5, 0.5) 0.5 0.5 No Yes No No 132 (0.5, 0.5) 0.25 0.5 No No No No
24 (0.5, 0.5) 0.25 0.5 No No No No 133 (0.75, 0.75) 0.375 0.75 Yes No Yes No

1830008-13
25 (0.75, 0.75) 0.375 0.75 Yes No Yes No 134 (0.75, 0.75) 0.375 0.75 Yes No Yes No
26 (0.75, 0.75) 0.375 0.75 Yes No Yes Yes 136 (0.5, 0.25) 0.25 0.5 No No No No
27 (0.5, 0.5) 0.5 0.5 No Yes No No 137 (0.75, 0.75) 0.375 0.75 Yes No Yes Yes
28 (0.75, 0.75) 0.375 0.75 Yes No Yes No 138 (0.75, 0.25) 0.375 0.75 No No Yes No
29 (0.5, 0.5) 0.5 0.5 No Yes No No 140 (0.25, 0.75) 0.375 0.25 No No No No
WSPC/S0218-1274

30 (1.0, 0.5) 0.5 1.0 Yes Yes Yes No 142 (0.5, 0.5) 0.5 0.5 No Yes No No
32 (0.25, 0.25) 0.125 0.25 No No No No 146 (0.75, 0.75) 0.375 0.75 Yes No Yes Yes
33 (0.5, 0.5) 0.25 0.5 No No No No 150 (1.0, 1.0) 0.5 1.0 Yes Yes Yes Yes
34 (0.5, 0.5) 0.25 0.5 No No No No 152 (0.75, 0.75) 0.375 0.75 Yes No Yes No
35 (0.25, 0.75) 0.375 0.25 No No No No 154 (1.0, 0.5) 0.5 1.0 Yes Yes Yes Yes
36 (0.5, 0.5) 0.25 0.5 No No No No 156 (0.5, 0.5) 0.5 0.5 No Yes No No
1830008

37 (0.75, 0.75) 0.375 0.75 Yes No Yes No 160 (0.5, 0.5) 0.25 0.5 No No No No
38 (0.75, 0.75) 0.375 0.75 Yes No Yes No 161 (0.75, 0.75) 0.375 0.75 Yes No Yes Yes
40 (0.5, 0.5) 0.25 0.5 No No No No 162 (0.75, 0.25) 0.375 0.75 No No Yes No
41 (0.75, 0.75) 0.375 0.75 Yes No Yes No 164 (0.75, 0.75) 0.375 0.75 Yes No Yes No
42 (0.75, 0.25) 0.375 0.75 No No Yes No 168 (0.75, 0.25) 0.375 0.75 No No Yes No
43 (0.5, 0.5) 0.5 0.5 No Yes No No 170 (1.0, 0.0) 0.5 1.0 No Yes Yes No
44 (0.75, 0.75) 0.375 0.75 Yes No Yes No 172 (0.5, 1.0) 0.5 0.5 No Yes No No
45 (1.0, 0.5) 0.5 1.0 Yes Yes Yes Yes 178 (0.5, 0.5) 0.5 0.5 No Yes No No
46 (0.5, 0.5) 0.5 0.5 No Yes No No 184 (0.5, 0.5) 0.5 0.5 No Yes No No
50 (0.25, 0.75) 0.375 0.25 No No No No 200 (0.5, 0.75) 0.375 0.5 No No No No
51 (0.0, 1.0) 0.5 0.0 No Yes No No 204 (0.0, 1.0) 0.5 0.0 No Yes No No
54 (0.5, 0.5) 0.5 0.5 No Yes No No 232 (0.5, 0.5) 0.5 0.5 No Yes No No
 April 5, 2018 10:17 WSPC/S0218-1274 1830008

S. Kamilya & S. Das

7. Conclusion Cattaneo, G., Formenti, E., Margara, L. & Mauri, G.

[1999] “On the dynamical behavior of chaotic cellular
A study of chaos in one-dimensional cellular automata,” Theoret. Comput. Sci. 217, 31–51.
automata (CAs) has been presented in this paper. Clementi, A. & Impagliazzo, R. [1995] “The reachability
The probability to which one cell communicates problem for finite cellular automata,” Inform. Pro-
with another has been analyzed in order to deter- cess. Lett. 53, 27–31.
mine the behavior of a CA. Based on the com- Devaney, R. L. [1992] A First Course in Chaotic Dynam-
munication among cells, a binary relation over the ical Systems (Westview Press).
set of cells has been formalized. The relation is an Durand, B., Formenti, E. & Varouchas, G. [2003]
equivalence relation and all the cells form a single “On undecidability of equicontinuity classification for
equivalence class when the cellular automaton (CA) cellular automata,” DMTCS Proc. (DMCS ) 3, 117–
is chaotic. Sometimes the communication between 128.
Kari, J. [2005] “Theory of cellular automata: A survey,”
two cells is blocked in some CAs by a blocking word
Theoret. Comput. Sci. 334, 3–33.
which appears in between the cells during evolution.
Knudsen, C. [1994] “Aspects of noninvertible dynamics
With the help of de Bruijn graph, one can find out and chaos,” PhD thesis.
a blocking word. We have identified two types of Kůrka, P. [2008] “Topological dynamics of one-
Int. J. Bifurcation Chaos 2018.28. Downloaded from www.worldscientific.com

blocking — full and partial. Finally to determine dimensional cellular automata,” Mathematical Basis
the behavior of the CAs, a parameter has been of Cellular Automata. Encyclopedia of Complexity
by UTRECHT UNIVERSITY on 04/12/18. For personal use only.

developed for the CAs. We have shown that the and System Science (Springer, Heidelberg).
proposed parameter performs better than the exist- Langton, C. G. [1990] “Computation at the edge of
ing parameters. However, there are still some excep- chaos: Phase transitions and emergent computation,”
tions in this new parameter. As a future work, we Physica D 42, 12–37.
would like to further improve the parametrization Li, W. & Packard, N. [1990] “The structure of the ele-
technique. mentary cellular automata rule space,” Compl. Syst.
4, 281–297.
Martinez, G. J., Adamatzky, A. & Alonso-Sanz, R. [2012]
Acknowledgment “Complex dynamics of elementary cellular automata
emerging from chaotic rules,” Int. J. Bifurcation and
This work is supported by Visvesvaraya PhD
Chaos 22, 1250023-1–13.
Scheme, Department of Electronics and Informa- Martinez, G. J. [2013] “A note on elementary cellular
tion Technology, Ministry of Communication and automata classification,” J. Cell. Autom. 8, 233–259.
IT, Govt. of India. McIntosh, H. V. [1990] “Wolfram’s class IV automata
and a good life,” Physica D 45, 105–121.
Schüle, M. & Stoop, R. [2012] “A full computation-
References
relevant topological dynamics classification of elemen-
Arai, T. [2017] “Devaney’s and Li–Yorke’s chaos in uni- tary cellular automata,” Chaos 22, 043143.
form spaces,” J. Dyn. Contr. Syst. 24, 93–100. Sutner, K. [1989] “A note on Culik–Yu classes,” Compl.
Banks, J., Brooks, J., Cairns, G., Davis, G. & Stacey, Syst. 3, 107–115.
P. [1992] “On Devaney’s definition of chaos,” Amer. Wolfram, S. [1984] “Universality and complexity in
Math. Monthly 99, 332–334. cellular automata,” Physica D 10, 1–35.
Bhattacharjee, K. & Das, S. [2016] “Reversibility of d- Wolfram, S. [2002] A New Kind of Science (Wolfram
state finite cellular automata,” J. Cell. Autom. 11, Media).
213–245. Wuensche, A. [1998] “Discrete dynamical networks and
Blanchard, F., Kůrka, P. & Maass, A. [1997] “Topo- their attractor basins,” Compl. Int. 6, 3–21.
logical and measure-theoretic properties of one-dim-
ensional cellular automata,” Physica D 103, 86–99.

1830008-14