An introduction to Cellular Automata

and their applications

DIMACS Reconnect 2000
Group #2 Sam Northshield (contact), Mike Daven, Alex Kheyts, Melkamu Zeleke
June 12, 2001
1 Introduction for the student
A cellular automaton or CA is a mathematical machine or organism that lends itself to some
very remarkable and beautiful ideas. One of the most exciting aspects of this area of mathematics is
that cellular automata arise from very basic mathematical principles. Though they are remarkably
simple at the start, CAs have a variety of applications and can show up in places you would never
expect! From simple games of chance to computer simulations of biology to art, we will see that
these simple ideas are evidence that mathematics is very real and is all around us.
In this module, we will show you how cellular automata are built with basic mathematical
building blocks. Where necessary, denitions you may know will be reviewed and new concepts
may be introduced. After we have seen a few examples of cellular automata, we take a look at the
most famous example, the Game of Life. We will then examine CAs from a dierent perspective,
applying these ideas to several problems of probability. We will use a dierent sort of machine
to solve several problems that involve games of chance. The nal section will equip you with the
material necessary to pursue this topic in an advanced class or a semester of independent study.
2 Introduction for the teacher
This module was designed to be part of a class for rst- or second-year liberal arts students. Our
intention is to introduce the theory of nite-state automata in an interesting way to those whose
studies are in non-technical elds.
In Section 3 we begin with Pascals triangle, a concept with which students may already be
familiar. In Sections 4 and 5 we use modular arithmetic to create Pascals triangle module 2, which
we soon see can be viewed as a one-dimensional cellular automata. In Section 7 we take a brief look
at the most famous example in cellular automata, the Game of Life. In Section 8 we give a few
interesting applications of simple nite-state machines, also introducing some basic probability
and graph theory. Finally in Section 9 we formalize the material presented in the previous sections
in such a way that students will be prepared to study cellular automata at a more advanced level.
This module is self-contained, so there are no prerequisites. We expect that this material could
be completed in (2,3 weeks), with time for review of the exercises contained herein.
This module was produced as part of the Reconnect Program 2000, organized by the DIMACS
Institute at Rutgers University. Our thanks to Lawrence Gray of the University of Minnesota for
introducing us to this topic.
1
3 Pascals Triangle
Pascals triangle is an array of numbers which has found uses in many areas of mathematics.
1
1 1
1 2 1
1 3 3 1
1 4 6 ? 1
Do you see the pattern?
Exercise 3.1
Can you nd the missing number?
Exercise 3.2
Can you nd the next row?
Answer: The missing number is 4 and is found by adding the two numbers above the empty space.
Answer: The next row is formed by the addition rule for nding missing spaces: 1, 5, 10, 10, 5, 1.
The 1s on either end also follow this rule if there are implicit zeros. So the triangle really looks
like
. . . 0 1 0 . . .
. . . 0 1 1 0 . . .
. . . 0 1 2 1 0 . . .
. . . 0 1 3 3 1 0 . . .
. . . 0 1 4 6 4 1 0 . . .
One of the most common uses of numbers from Pascals triangle is in the binomial formulas,
such as:
(a + b)
2
= a
2
+ 2ab + b
2
,
(a + b)
3
= a
3
+ 3a
2
b + 3ab
2
+ b
3
,
and (a + b)
4
= a
4
+ 4a
3
b + 6a
2
b
2
+ 4ab
3
+ b
4
.
The coecients of (a + b)
n
come from row n of Pascals triangle.
Exercise 3.3
Use Pascals triangle to nd (a + b)
5
and (a + b)
7
.
Perhaps a better picture is:
0 0 0 0 0 1 0 0 0 0 0 0
0 0 0 0 1 0 1 0 0 0 0 0
0 0 0 1 0 2 0 1 0 0 0 0
0 0 1 0 3 0 3 0 1 0 0 0
0 1 0 4 0 6 0 4 0 1 0 0
2
One can think of each successive row as a new state resulting from application of a transition
rule to the state before.
Exercise 3.4
(a) What are the states that we are talking about? Give 3 examples of new states.
(b) What is the transition rule?
(c) Apply the transition rule to the following state:
(. . . 2 4 2 4 1 2 1 2 0 1 1 0 . . . )
(d) Compare the two states:
(. . . 0 0 0 1 2 3 0 0 0 0 0 0 . . . )
(. . . 0 0 0 0 1 2 3 0 0 0 0 0 . . . )
Are they dierent?
(e) Apply the transition rule to the two states in (d); are the results dierent?
It is perhaps unsatisfactory to dene a state to be an innite (in both directions!) row of
numbers since two dierent such rows are indistinguishable unless placed next to each other (see
Exercise 3.4(d) above). To get around this diculty, we can dene a state to be an assignment of
numbers to each integer: for example, A(2) = 1, A(3) = 2, A(4) = 3, and A(n) = 0 for all n not
equal to 2, 3, or 4.
Exercise 3.5
If A is the state corresponding to the rst row of Exercise 3.4(d), what is the state corresponding
to the second row?
The transition rule could then be described by the equation
B(n) = A(n 1) + A(n + 1) .
4 Addition Modulo 2
We are all familiar with ordinary addition of whole numbers. Here we present addition which only
works for a very small set of whole numbers, namely {0, 1}. The addition rule is
0 + 0 = 0 , 0 + 1 = 1 , 1 + 0 = 1 , and 1 + 1 = 0 .
We write, for a, b equal to 0 or 1, a + b mod 2.
Essentially, we are throwing out all properties of numbers except their even-ness or odd-ness
and adding those.
Exercise 4.1
Find a + b + c mod 2 for all eight possible congurations of a, b, and c.
Exercise 4.2
Explain why there are exactly eight congurations possible in the exercise above.
3
5 Pascals Triangle Modulo 2
Suppose we now replace addition in the transition rule in Section 3 by addition modulo 2. That is,
given a state A (being an assignment of 0 or 1 to each of the integers), the transition rule changes
it to the state B dened by
B(n) = A(n 1) + A(n + 1) mod 2 .
Exercise 5.1
Let A be dened by
A(n) =
_
1 if n is odd
0 if n is even
To what does the transition rule take A? Apply the transition rule to the result. Repeat. Do you
see a pattern?
Let us start with A(n) =
_
1 if n = 0 ; 0 otherwise
_
. This is our initial state. If we draw that
state as a row with an asterisk for a one and a blank for a zero, we get
*
By applying the transition rule repeatedly and drawing consecutive states vertically downward,
we get
0 0 0 0 0 1 0 0 0 0 0
0 0 0 0 1 0 1 0 0 0 0
0 0 0 1 0 0 0 1 0 0 0
0 0 1 0 1 0 1 0 1 0 0
0 1 0 0 0 0 0 0 0 1 0
Here are a few further diagrams:

This is where we had a few cutouts, including one like this . . .

Figure 5.1: One of several ways to picture Pascals triangle.
The gure above is sometimes called Pascals triangle mod 2 (for, I hope, obvious reasons).
4
Exercise 5.2
Interestingly, we may sometimes run a cellular automaton backwards. The question is, if B is the
state following A by applying a transition rule, is it possible to nd the reverse transition rule
which, when applied to B, gives A? Hint: The only way that one can reverse a transition rule is if
the rule is one-to-one that is, there are no two distinct states that can precede a given state.
Is there a state other than (. . . 0001000 . . . ) preceding (. . . 000101000 . . . )?
A curious sequence is obtained by counting the number of 1s in consecutive rows of Pascals
triangle mod 2:
1, 2, 2, 4, 2, 4, 4, 8, 2, 4, 4, 8, 4, 8, 8, 16, 2, 4, 4, 8, . . .
Exercise 5.3
What are the next 50 terms of this sequence?
To do the above exercise, one must have noticed some patterns (e.g., every number in the
sequence is a power of 2). Is there a way to predict the n
th
term of this sequence without calculating
all the previous terms? For example, how many 1s are there in the 1000
th
row of Pascals triangle
mod 2?
Exercise 5.4
(A challenging exercise)
Show that the number of ones in the n
th
row of Pascals triangle mod 2 is 2 to the power j,
where j is the number of ones in the binary expansion of n. Recall that the binary expansions of
1,2,3,4,5,6,7,8 . . . are 1,10,11,100,101,110,111,1000, . . . .
Exercise 5.5
Use the result of Exercise 5.4 (even if you didnt do it) to nd the number of 1s in the n
th
row of
Pascals triangle mod 2 if n = 10, 25, 31, 66, 134, and 123456.
Exercise 5.6
(A challenging exercise based on Exercise 5.4)
Recall that n! (n-factorial) is the product of the rst n natural numbers. It is even, for n greater
than or equal to 2. Furthermore, the larger n is, the more 2 divides n!. Show that 2 to the n
th
power is the largest power of 2 which divides n!x, where x is the number of 1s in the n
th
row of
Pascals triangle mod 2.
6 Cellular Automata
We are interested in generalizing the previous example. There are two ways to do this: one is to
change the initial state, the other to change the transition rule. Before we proceed, lets formalize
what we have so far.
We have a initial state (say A) and, implicitly, a set of possible states. In the case above, the
set of possible states is simply the set of sequences of 0s and 1s more explicitly, it is the set of
all possible assignments of 0s and 1s to the integers. Furthermore, we have a transition rule which
assigns to each possible state, another state.
The rest of this section is devoted to examples.
5
Exercise 6.1
What happens if we start with a dierent starting state? Trace out the rst 5 states following each
of the following initial states:
(a) ( . . . 000001001001000000 . . . ) (3 ones, 3 apart)
(b) ( . . . 000011001100000000 . . . ) (two pairs of ones)
(c) ( . . . 111000111000111000 . . . ) (alternating triples of ones and zeros)
(d) ( . . . 010001011010001101 . . . ) (random)
Another possible change is obtained by changing the transition rule. Such a rule can be thought
of as a machine or computer which takes as input a string of 0s and 1s and gives a 0 or 1 as
output.
For example, we may dene B(n) = A(n 1) + A(n) + A(n + 1) mod 2. It is similar to the
rule above in that it is local; the new value at n depends only on the old values at and/or near
n. We shall call this rule Rule 111 (the rst rule, the one for Pascals triangle, is denoted Rule
101).
Exercise 6.2
Repeat Exercise 6.1 but with Rule 111 instead of Rule 101.
Exercise 6.3
How would you interpret Rule 10101? Repeat Exercise 6.1 with this rule.
It has been proposed that there are 4 types of transition rules for cellular automata resulting in
dierent qualitative behavior. The possibilities are that the evolution of the automaton leads to a
homogeneous state, leads to a set of separated simple stable or periodic structures, leads to a chaotic
pattern, or leads to complex localized structures, sometimes long-lived. Here are some examples
(these all follow binary rules of the form B(n) = F
_
A(n 2), A(n 1), A(n), A(n + 1), A(n + 2)
_
,
where binary means that cells are either 0 or 1.)

Insert gures here

Figure 6.1: The 4 transition rules for cellular automata.
Exercise 6.4
Use a computer implementation of cellular automata (see references) to draw pictures similar to
those above. Part of this exercise is to gure out the transition rules.
6
7 The Game of Life
John H. Conway developed the Game of Life in the 1960s. It is an example of a two-dimensional
CA. It is called two-dimensional since a possible state looks like an innite checkerboard with sites
either occupied (with a checker, say) or not. For theoretical purposes, we usually assume that the
checkerboard is innite in all directions. For practical purposes, such as running the game on a
computer, we must assume that the checkerboard is nite. First we dene neighbors: two cells
(i.e., squares) are neighbors if they dier by one space (left/right/up/down/diagonal) that is, one
chess kings move apart. Thus, every cell has eight neighboring cells.
1 2 3
4
5 6 7
8
Figure 7.1: The neighbors of a cell are one chess kings move away.
Here, of course, the innite case is preferable; how does one dene neighbors for cells on the
edge of a nite board? There are two common ways to do this. In the rst case, one simply counts
neighbors regardless of whether the cell is on the edge or not. For example, a corner square of
a checkerboard has only 3 neighbors. Often we wish to preserve the fact that each cell has eight
neighbors in spite of being on the edge. The way to work this out is to wrap the board around
so that the cell to the left of a cell on the left edge of row k is the rightmost cell of row k, and so
on.
1 2 3
4
5 6 7
8 1 2 3
4
5 6 7
8
Figure 7.2: There are always eight neighbors on a torus.
Exercise 7.1
Explain why all cells in the arrangements above have exactly 8 neighbors. This setup is sometimes
said to be on a torus. Why?
A state is the checkerboard plus an assignment, for each cell, whether that cell is occupied
or not.
Exercise 7.2
What is the state with the fewest number of occupied sites? How many states are there with exactly
one occupied site if the board is 3 by 3? 8 by 8? Innite?
7
As above, the transition rule is local. Here it is: an occupant of a cell with one or fewer neighbors
will, sadly, die of loneliness. On the other hand, with 4 or more neighbors, an occupant will die
of over-excitement. With two neighbors, the occupant will continue into the next generation. The
apparently tri-sexual occupants will reproduce occasionally: if a cell, empty or not, has three
occupied neighbors, then the cell will be occupied in the next generation. These simple rules allow
for virtually unlimited complexity.
Exercise 7.3
Apply the transition rule to each of the following congurations (assume that cells outside the
visible region are all empty).
Block
Ship
..
Blinker
r-pentomino
Doing Exercise 7.3, you saw that a couple of congurations were stable. In particular, the block
and the ship remained the same after applying the transition rule. These terms were coined by
Conway; see these and more in Winning Ways for Your Mathematical Plays [2] (volume 2) with
Elwyn Berlekamp and Richard K. Guy.
We are interested in the evolution of this system. That is, given an initial state, what happens
after several generations? Several thousand generations? For example, we saw above that the state
consisting of one block or one ship will remain the same generation after generation. Thus the
evolution of this system is completely known! Also from Exercise 7.3, you saw that a row of three
becomes a column of three and vice versa. Thus, starting with row of three, we get back to the
original state in two generations. This conguration is called a blinker (take a look at it on a
computer).
Exercise 7.4
(a) Why do we use the word cyclic when some state occurs more than once?
(b) Give four examples of initial states that become cyclic.
(c) Give a convincing argument for why, in either of the two nite setups, the evolution always
becomes cyclic.
We say a state is periodic if it occurs more than once following some starting state.
Exercise 7.5
(a) Show that a block and a ship are each periodic.
(b) Show that if a state is periodic, then the subsequent state is also periodic.
(c) Show that every state following a periodic state is periodic.
8
If we start with a periodic state, then that state reappears for the rst time after a certain
number of generations; the number of such generations is called the period of that state. For
example, the block has period 1.
Exercise 7.6
(a) Find the period of the ship.
(b) Find the period of the blinker.
(c) Begin with a row of 10 consecutive occupied cells; the state is eventually periodic. What is
the period of the rst periodic state? This exercise all but requires a computer.
(d*) Find a periodic state with period 3.
Although every state is eventually periodic on a nite board (by an exercise above), this is not
true on an innite board. Consider the initial state in the gure below, called a glider:
Figure 7.3: A glider.
Exercise 7.7
(a) Iterate Life for 10 generations starting with a glider.
(b) Can you explain why, on an innite board, the glider is not eventually periodic?
(c) What happens to the glider on the torus? That is, why does this not violate the conclusion
of Exercise 7.7(b)?
An interesting phenomenon is that a nite state can result in unbounded growth. For example,
consider the r-pentomino.
Figure 7.4: The r-pentomino.
Exercise 7.8
(a) How many gliders are produced by an r-pentomino? (Use a computer!)
9
(b) What is the total biomass at the end? That is, how many occupied sites are there after things
settle down?
If one iterates this (hopefully with the aid of a computer), one sees that the mess eventually
produces a couple of gliders going o in distinct directions. Hence the diameter of the occupied
sites grows without bound (on an innite board).
Figure 7.5: A glider gun.
The objects like the blinker, the glider, and the block are indeed objects; we can com-
prehend them as entities that persist through time (e.g., they blink, glide, or just sit). They are
not states or even, in the case of the glider, parts of states.
Exercise 7.9
(A problem for philosophers):
If blinkers, gliders, and blocks are objects, what is an object? For the glider, what is it that
persists through time?
Here are two more objects:
10
Figure 7.6: On the left, a clock; on the right, a horizontal glider.
Exercise 7.10
(a) What is the period of the clock?
(b) Verify that the hozirontal glider earns its name.
One of the most interesting results due to J.H. Conway is that the Game of Life is capable of
universal computation that is, it can do anything that any computer can do. For example,
there is a nite initial state such that any paragraph of English prose, when properly coded as
a sequence of gliders, will result in a spell-checked paragraph of English prose (again coded as
a sequence of gliders). More details on this and other topics on the Game of Life can be found in [2].
8 Applications of CAs and probability
Until now, we have considered a very particular sort of cellular automaton. It was important to
know the transition rule from one state to another state. In this section, we will discuss elementary
probability and a few applications of the ideas of CAs. The states and transition rules will appear
dierent, as will the gures we draw, but we will still rely on the fundamental notions we have seen
in the previous sections.
Exercise 8.1
Flip a coin 3 times, and draw a tree diagram that represents the possible outcomes.
What are the chances that you get 3 heads in a row? What are the chances that you get two
heads and one tails? What are the chances that you get two heads and one tails, not necessarily in
that order? These are simple questions, but the answers are not so simple. It will help if we clarify
the terminology we will use.
Call the result of each coin toss an outcome. Here, an outcome can be either heads or tails.
An event is a sequence of several outcomes, such as three heads in a row or heads, tails, heads,
11
tails. We call the collection of all possible events the sample space. The chances that a certain
event will occur is called the probability of that event.
For each of the questions above, What are the chances . . . ?, the answer will be between 0 and
1. Whenever we ask a probability question, this will always be the case. We will usually assume
that the coin we use is a fair coin.
Exercise 8.2
(a) The chances of getting heads on a single toss of a fair coin is 1/2. For the same coin, what
are the chances of getting three heads in a row in 3 tosses? What are the chances of getting
heads, tails, heads, tails in 4 tosses?
(b) With a fair coin, what are the chances of getting three heads in a row in 4 tosses? In 5
tosses? In each case, consider what three heads in a row really means.
(c) Suppose that we have a biased coin, where the chances of getting heads on a single toss is
1/3. Repeat parts (a) and (b) for this unfair coin.
Example 8.3
A bold gamble
You have $1 and you need $5. You can reach your goal by a fair gamble. You decide on the bold
strategy: at each round you stake so much of your current fortune that you come as close to your
goal as possible if you win. For example, if you have $2, you bet it all; if you have $4, you would
only need to bet $1. You are bold enough to risk whatever it takes to get $5, but you can never
bet more than you have at the moment. The bold gamble can be translated into Figure 8.1.
0
1
2
3
4
5
1/2
1/2
1/2
1/2
1/2
1/2
1/2
1/2
Figure 8.1: If you had $1, how could you get $5 by several bold gambles?
An event here is your current fortune, and each event is represented by a vertex or state. If the
event B can be obtained from event A by a single coin-toss, we draw a directed line or arc from
A to B and we call B a neighbor of A. We call the resulting diagram a directed graph or digraph.
So we use a digraph to picture the transition from one state to another; this is a dierent way of
looking at what we have seen in the previous sections.
There are two kinds of states we will want to identify. If a state has both incoming and
outgoing arcs, we say this state is internal. If a state has only incoming arcs, we say this state
is a boundary or absorbing state. From each internal vertex in our digraph there are two outgoing
arcs, since we can either win or lose from this position. We have marked each arc with a 1/2 to
indicate that a win or loss is equally likely for each individual bet you make.
12
The direction of each arc is important. For instance, the vertex labelled 1 indicates the start
of the game, where you hold $1. At this point, you could bet just $1 and either win (go to state
2) or lose (go to state 0). Suppose you win, and are now in state 2. From here, you will bet
your entire $2. If you win this next bet, you will go to state 4; if you lose, you will go to state 0.
Take a moment to conrm that each of the directed edges corresponds to a bet won or lost.
With this digraph and some new terminology, we can return to our original puzzle. What are
the chances of successfully winning the $5 you need? That is, what are the chances of getting from
$1 to $5?
describe weighted average, duration, mean value
To answer this question, we will need to know two important properties that are used in prob-
ability. First, the probability of going from an internal state to one of the boundary states is the
weighted average of the probabilities of the neighboring states. Second, the expected duration going
from an internal state to the one of the boundaries is the average of the duration of its neighbors,
plus 1.
describe why it doesnt matter which boundary
redraw digraph in several steps, describe

Exercise 8.4
List the internal states and the boundary states in the digraph above.
Exercise 8.5
Why do the states 0 and 5 have no outgoing arcs? Why does each internal state have a single
incoming arc?
Exercise 8.6
Suppose you start with $1, but the object is to win $6. Draw a digraph with 7 vertices and all of the
appropriate edges. In this new digraph, which of the states are internal, and which are absorbing?
Exercise 8.7
For each of the following, give an example of a digraph:
(a) a digraph with several boundary states
(b) a digraph with at least one vertex which has only outgoing arcs
(c) a digraph with no boundary
(d) a digraph with no internal states
These digraphs dont necessarily have to come from the bold strategy.
Of course, as in Exercise 8.7, digraphs may represent a variety of scenarios. Here is another
game, and we will again use digraphs to map out all the possibilities.
Example 8.8
A simple coin game?
Aristophanes and Zeilberger have chosen the numbers 011 and 111, respectively. They play the
13
following game: toss a fair coin with faces 0 and 1 until one of their numbers has appeared. What
are the chances that Zeilberger wins this game?
Lets draw the appropriate digraph:
start
1
0
11
01
111
011
Figure 8.2: What are the chances that 111 will come up before 011? Use this digraph.
At the very start, we toss the die once and get either a 0 or 1; lets say we have rolled a 1.
Consider the next roll; either this is a 1 (so we have 11) or a 0. Recall that the sequences we are
looking for are 011 and 111. Since 10 does not occur in either nal sequence and order is
important!, the rst roll is discarded and we only take note of the 0. Take a moment to convince
yourself that the vertices and arcs in the digraph above describe all the possible events in this game.

nish this game! discuss counting paths, (1/2)

n
for paths of length n, etc.

Exercise 8.9
What are the chances that Aristophanes will win?
Exercise 8.10
Suppose this game is played with an unfair coin. Lets say the chances of a 0 is 1/3, while the
chances of a 1 is 2/3. In this case, the arcs should be labelled with either 1/3 or 2/3. What are
the chances that Aristophanes will win using this coin?
Exercise 8.11
Alice chooses the number 010 and Bob chooses the number 101. Sketch an appropriate digraph
and repeat the exercises above.
Of course, the game above could be played with a normal coin; just let 1 corresponds to
heads and 0 correspond to tails, for instance. By doing something as simple as playing a
coin game, we have had the opportunity to ask some very challenging questions and to experience
some rather interesting mathematics. The theory of probability is a rich and interesting eld which
arises from a few familiar rst concepts. Here is another game which is easy to state.
Example 8.12
Suppose we roll a fair 6-sided die repeatedly. Here is the question: what are the chances that both
a 1 and a 3 both occur before an even number occurs?
14
In this game, an outcome is a single roll of the die and can be any whole number from 1 to
6. An event is a sequence of 5 consecutive rolls of the die. The event 1,5,3,2,3 works, since
both 1 and 3 occur before 2. Another successful event is 4,3,5,1,1. The 1 and 3 neednt occur
back-to-back or in order, as long as they occur before any even number. On the other hand,
the event 1,2,3,3,2 does not match the question, since 2 is an even number and occurs before the
3. Can you think of other events that work or do not work? There are many possibilities, and this
is why a carefully-drawn digraph is so important. Here is a diagram to describe how the game is
played:
start
a {1, 3}
even
4 a
1/3
1/2
1/6
1/2
1/6
1/3
Figure 8.3: Toss a fair 6-sided die, looking for a 1 and a 3 before an even number.
In order to answer our question, we have created a digraph that we will call a chip-ring
machine. We will place markers or chips onto a vertex. Since a regular die has six faces, we allow
ring when a vertex holds six or more chips. By ring, we mean that we distribute the chips on
a particular vertex to the neighboring vertices, which we will describe more below. In Figure 8.4,
we see how the rst ring takes place.
1/3
1/2
1/6
1/2
1/6
1/3
1/3
1/2
1/6
1/2
1/6
1/3
Figure 8.4: At left we see there are 6 chips at the start, so this vertex is ready to re. After ring,
we see how the chips were distributed at right. Several chips must be placed before ring again.
Notice that there is only one internal state in this digraph. Since nothing will happen unless
there are at least 6 chips on at least one vertex, we sometimes pre-load a vertex with one or more
15
chips. Without loss of generality, load the internal state in the diagram with ve chips, and begin
the game by feeding 6 chips into the start.
We are ready to play the game. The game will be nished when the initial loading of the
internal state recurs (with 5 chips in this case). The probabilities we have assigned to the edges
determine the distribution of the chips during each ring.
1/3
1/2
1/6
1/2
1/6
1/3
1/3
1/2
1/6
1/2
1/6
1/3
1/3
1/2
1/6
1/2
1/6
1/3
1/3
1/2
1/6
1/2
1/6
1/3
1/3
1/2
1/6
1/2
1/6
1/3
Figure 8.5: The game, start to nish (left-to-right, then down).
16

say more here

From the chain of diagrams in Figure 8.5, one can see that the chances of getting 1 and 3 before
an even number is 1/10. The expected duration of the game (that is, total number of steps made,
divided by the number of chips absorbed) is 18/10 = 1.8.
Exercise 8.13
Why is an event a sequence of ve consecutive rolls of the die? Why pre-load we label the internal
vertex with 5 chips?
Exercise 8.14
Suppose we want to know the chances that 1,3 occurs before an even number. That is, we toss
a fair 6-sided die repeatedly and look for a 1 then a 3, both before an even number. For example,
1,5,3,2,3 and 1,1,3,6,5 are successful events, while 1,2,3,3,2 and 5,3,3,1,4 are not successful.
Sketch an appropriate digraph and label the arcs.
more exercises, especially on duration and mean value?
The way we moved chips around the chip-ring machine, and some of the other digraphs we
have seen, might remind you of an abacus, a primitive sort of calculating machine. In fact, several
of the problems in this section were drawn from a paper entitled The probabilistic abacus (Engel,
[4]), and that paper contains several more examples. For a more theoretical discussion, see also [5].
For more on the mathematics of gambling-type games, see [3].
9 Formal denitions
In preceding sections we have considered a few problems that can be described in terms of the same
mathematical concept, called a Cellular Automaton. An automaton is a (real or imaginable) device
capable to perform certain actions according to a specic program, without human interference. We
do not discuss here as to who has created the program. A mathematical (not necessarily cellular)
automaton is a model describing this behavior. This model involves ve mathematical objects,
three sets and two mappings
1
. The cellular automata can be also dened in this way, and this is
our goal in this section. While reading this section, the reader should keep in mind the exercises
in Section 6.
Recall that in the preceding examples, we have actually dealt with sequences of zeros and ones
and changed one such sequence to another, subject to some specied rules. We can imagine an
innite tape separated into equal squares, called cells, with the digits 0 and 1 written in these
squares, exactly one digit in each cell. This is why the model is called a cellular automaton.
Thus, we can assume each row of the Pascal triangles in the preceding sections to be written on
such tape. When we go downward in the triangle, from one line to another, we change the contents
of its cells according to the transition rule described. We assume this tape is innite in both
directions with all empty cells initially occupied by zeros.
The set of symbols used by the automaton, in our case the set Z
2
= {0, 1}, is called the (internal)
alphabet of the automaton. It is possible to use any set of dierent characters, not just two digits.
Moreover, rather than a one-dimensional tape, we can consider a plane or even a many-dimensional
1
Namely, a mathematical automaton is a quintuple {X, Y, S, f, g}. The sets X, Y , and S are called the input
alphabet, the output alphabet, and the set of (internal) states. The map f : X S Y is called the output function,
and the map G : X S S is called the transition function.
17
space separated into cells, or (which is, of course, the same) we can place these characters at the
points with integer Cartesian coordinates (also called cells or sites) of the n-dimensional Euclidean
space. However, for our introductory exposition, it is enough to consider only two symbols and
one tape
2
. We call these objects 1-Dimensional Binary Cellular Automata (1-D BCA) and while
referring to CA in this section, we shall always mean 1-D BCA.
Now, given a CA, at every time instance we have an innite sequence of zeros and ones written
on the tape. So we can forget about the tape and consider only this sequence of zeros and ones.
Such innite sequences are called states of a cellular automaton. For example, s

= (. . . , 0, 0, 0, . . . ),
s

= (. . . , 1, 1, 1, . . . ), and s

= (. . . , 1, 0, 1, 0, . . . ) are three dierent states. The set of all states,

that is, the set of all double-innite binary sequences is denoted by S. Sometimes we need also
to specify the zero term of this sequence, that is, the zero cell of a state. Hereafter, the initial
term (cell) is denoted by a hat like s = (. . . , 0,

0, 0, . . . ). If we denote a state as a sequence

s = (. . . , c
2
, c
1
, c
0
, c
1
, c
2
, . . . ), we assume the initial cell to be c
0
.
A CA operates in discrete time. This means that at an initial moment in time, let us denote it
by t
0
, the CA is in its initial state s
0
= s(t
0
). Within a certain time unit, all the cells synchronously
change their holdings and at the next time moment t
1
, the CA transfers to the next state s
1
. In
the following time moments, these steps repeat. We often denote these moments by 0,1,2, . . . .
When we need to show the dependence of the cells holdings on the time moment t, we write
s(t) = (. . . , c
2
(t), c
1
(t), c
0
(t), c
1
(t), c
2
(t), . . . ), etc.
The transition rule of a CA is a function f which, given a state s
t
at the moment t, calculates
the state s
t+1
at the next moment t +1. That is, f : S S, or more specically, s(t +1) = f(s(t)).
For instance, in Exercise 3.4(b) on page 3, the transition rule for Pascals triangle can be written
as c
k
(t +1) = c
k1
(t) +c
k+1
(t), where k and t are integers and t 0. The transition rule in Section
4 is given by the formula c
k
(t + 1) = c
k1
(t) + c
k+1
(t) mod 2. To apply these transition rules, we
must specify the initial state s(0). In both examples above, s = (. . . , 0,

1, 0, . . . ); that is, c
0
(0) = 1
and c
k
(0) = 0 if k 1. To dene a transition rule, it is often convenient to assign its action on
specic cells as in the examples above. In terms of abstract automata theory, CAs are autonomous
automata without output.
Thus, we arrive at the following basic denition.
Denition. A cellular automaton is a triple U =
_
S, f, s(0)
_
, where S is a set of double-innite
0-1 sequences, called the states of the automaton, a map f : S S is called its transition rule, and
s(0) S is its initial state.
The collection of all consecutive states s
0
, s
1
, s
2
, . . . of an automaton is called its evolution.
Example 9.1
Let S consist of sequences containing exactly one 1, the initial state be s(0) = (. . . , 0,

1, 0, . . . ),
and the transition rule be given by the following equations:
c
t+1
(t + 1) =

k=
c
k
(t) and c
k
(t + 1) = 0 for k = t + 1 .
Exercise 9.2
(a) Verify that this rule moves the 1 consecutively one step to the right.
(b) Construct a CA that moves the 1 consecutively one step to the left.
2
From a mathematical point of view, these extensions do not provide a more general theory.
18
Example 9.3
We sometimes consider periodic states. This means that there exists a nite group of the digits 0
and 1 which repeats endlessly in both directions. Thus, in the state s

above this group consists

of one symbol 0, the length of this period is 1. In s

above, the period which is (1, 0) has

the length of 2. In s

and s

, the initial cell is irrelevant, but it should be noted that the state
s

= (. . . , 1, 0, 1, 0, 1, . . . ) is dierent from both s

and s

. It also has the period of the length

2, however now the period is (0, 1). Certainly, the groups (1, 0, 1, 0), (1, 0, 1, 0, 1, 0), etc., are also
periods of s

, but we always look for the shortest period. Periodic states of length N are denoted by
s = (c
0
, c
1
, . . . , c
N1
), where N is assumed to be the length of the shortest period and c
0
occupies
the initial cell we do not hat it now. Due to the periodicity, c
1
= c
N1
, c
0
= c
N
, and so
forth. We call these equations periodic boundary conditions. So we can write down the same state
as s = (. . . , c
N1
, c
0
, c
1
, . . . , c
N1
, c
0
, c
1
, . . . ).
Exercise 9.4
Determine periods of the following states:
(a) (1, 1, 0, 1, 1, 1, 0, 1, 1, 1, 0, 1)
(b) (. . . , 1, 0, 1, 1, 0, 1, 1, 1, 0, 1, 1, 1, 1, 0, . . . )
(c) What about Part (c) of Exercise 6.1?
Answer: For (a), the period is (1, 1, 0, 1), its length is 4, so the state can be written as (1, 1, 0, 1).
In both (b) and (c), these states have no period.
For further examples, review all relevant examples from the preceding sections.
Given a time instant t and the state s(t) = (. . . , c
2
(t), c
1
(t), c
0
(t), c
1
(t), c
2
(t), . . . ), a transition
rule must calculate the contents of all cells at time t +1. This means that the transition rule must
involve, explicitly or implicitly, functions c
k
(t +1) = F
k
(. . . , c
2
(t), c
1
(t), c
0
(t), c
1
(t), c
2
(t), . . . ) for
some integer k. The functions F
k
may themselves depend on t; that is, they may depend on the
preceding states s(t 1), s(t 2), etc. However, we consider only a simple case when the functions
F
k
(that is, the transition rule f) does not depend on t. This means that
c
k
(1) = F
k
_
. . . , c
2
(0), c
1
(0), c
0
(0), c
1
(0), c
2
(0), . . .
_
,
c
k
(2) = F
k
_
. . . , c
2
(1), c
1
(1), c
0
(1), c
1
(1), c
2
(1), . . .
_
,
c
k
(3) = F
k
_
. . . , c
2
(2), c
1
(2), c
0
(2), c
1
(2), c
2
(2), . . .
_
, etc.
Moreover, we assume that the functions F
k
depend only on the cell itself and its 2r closest neighbors
but do not depend on k. That is, the transition rule is given by a xed function F depending on
2r + 1 arguments:
c
k
(t + 1) = F
_
c
kr
(t), c
kr+1
(t), . . . , c
k1
(t), c
k
(t), c
k+1
(t), . . . , c
k+r1
(t), c
k+r
(t)
_
for some integer k. We call such automata isotropic. The number r 1 is called the range of a
CA. Thus, in the example with Pascals triangle, r = 1.
19
Exercise 9.5
In Example 9.1, what is r?
Example 9.6
Let r = 2 and F(a, b, c, d, e) = (a b + c d + e) mod 2. Assume s
0
= s(0) = (. . . , 0,

1, 0, . . . ),
calculate s(1) = (. . . , c
1
(1), c
0
(1), c
1
(1), . . . ).
We have the following:
c
0
(1) = F
_
c
2
(0), c
1
(0), c
0
(0), c
1
(0), c
2
(0)
_
= (0 0 + 1 0 + 0) mod 2 = 1 mod 2 = 1 ;
c
1
(1) = F
_
c
1
(0), c
0
(0), c
1
(0), c
2
(0), c
3
(0)
_
= (0 1 + 0 0 + 0) mod 2 = 1 mod 2 = 1 ;
c
1
(1) = F
_
c
3
(0), c
2
(0), c
1
(0), c
0
(0), c
1
(0)
_
= (0 0 + 0 1 + 0) mod 2 = 1 mod 2 = 1 ;
c
2
(1) = F
_
c
0
(0), c
1
(0), c
2
(0), c
3
(0), c
4
(0)
_
= (1 0 + 0 0 + 0) mod 2 = 1 mod 2 = 1 ;
c
2
(1) = F
_
c
4
(0), c
3
(0), c
2
(0), c
1
(0), c
0
(0)
_
= (0 0 + 0 0 + 1) mod 2 = 1 mod 2 = 1 .
Clearly, c
k
(1) = 0 if |k| 3, since the evaluation of these c
k
(1) do not involve c
0
(0) = 1. Therefore
s(1) = (. . . , 0, 1, 1,

1, 1, 1, 0, . . . ).
Exercise 9.7
Find s(2), s(4), s(4), and s(5) for this CA.
Exercise 9.8
* Can you use the Principle of Mathematical Induction to nd all states of this CA?
The transition rule in this example can be written as a linear function,
c
k
(t + 1) = c
k2
(t) + c
k1
(t) + c
k
(t) + c
k+1
(t) + c
k+2
(t) ,
where k is an integer. This is an example of a linear rule. Generally, a transition rule is called
linear if F(a, b, . . . , m) = a + b + + m, where , , . . . , are constants.
Exercise 9.9
Design a linear transition rule with r = 1 and calculate the evolution of this CA.
Now consider the following transition rule with r = 1: F(a, b, c) = (a
2
+b
2
+c
2
) mod (2). This
rule is obviously non-linear.
Exercise 9.10
Find the evolution of the CA with this transition rule and the initial state s(0) = (. . . , 0,

1, 0 . . . ).
Find a linear transition rule that generates an equivalent CA, that is, an automaton with the same
evolution.
20
Exercise 9.11
* Is it always possible to nd an equivalent linear transition rule for a general transition rule
F
_
c
kr
(t), c
kr+1
(t), . . . , c
k1
(t), c
k
(t), c
k+1
(t), . . . , c
k+r1
(t), c
k+r
(t)
_
?
Exercise 9.12
* For a general transition rule F
_
c
kr
(t), c
kr+1
(t), . . . , c
k1
(t), c
k
(t), c
k+1
(t), . . . , c
k+r1
(t), c
k+r
(t)
_
,
is it always possible to nd an equivalent transition rule of kind f(a, b, . . . , m) = f(a +b + +
m)? That is, a rule depending only on a linear combination of its arguments.
Next, we introduce several important classes of CA. If the function F does not actually depend
on c
k
(t), that is, it is a function of only 2r arguments, then the automaton is said to be without
memory. An automaton or a corresponding transition rule is said to be symmetric if
F
_
c
kr
(t), c
kr+1
(t), . . . , c
k1
(t), c
k
(t), c
k+1
(t), . . . , c
k+r1
(t), c
k+r
(t)
_
= F
_
c
k+r
(t), c
k+r1
(t), . . . , c
k+1
(t), c
k
(t), c
k1
(t), . . . , c
kr+1
(t), c
kr
(t)
_
.
An automaton or a corresponding transition rule is called legal if F(0, 0, . . . , 0) = 0 (an empty state
can generate only itself).
Examples.
Representations of cellular automata. Digraphs (diagrams of states), Tables of transfers,
Vertices of the unit cube.
To materialize an automaton, we can use dierent ways, depending on our task. For instance,
given a transition rule, the corresponding CA can be represented by a directed graph, as we have
seen in preceding sections.
More examples.
To store a CA in computer memory, it is convenient to use transfer tables. We again consider
a 1-D BCA with the range r, that is, its transfer function (the rule) depends on 2r + 1 arguments
and the internal alphabet is Z
2
= {0, 1}. We can assign such a transfer rule as a chart with 2r + 1
columns for the arguments and an additional column for the values of the transfer function. Since
the entry of each cell in this chart may be either 0 or 1, there are 2
2r+1
possibilities to ll in these
rows for the arguments; that is, the chart must contain 2
2r+1
rows:
c
kr
(t) c
kr+1
(t) c
k
(t) c
k+r1
(t) c
k+r
(t) c
k
(t + 1)
0 0 0 0 0
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
1 1 1 1 1
where c
k
(t + 1) = F
_
c
kr
(t), . . . , c
k
(t), . . . , c
k+r
(t)
_
.
Since there are 2
2r+1
cells in the last column and two possible entries, there are 2
2
2r+1
ways to
complete the chart. So there are 2
2
2r+1
one-dimensional binary cellular automata. For example,
the following chart exhibits the trivial CA, which transforms any state to an empty one, with all
21
cells containing only zeros:
c
kr
(t) c
kr+1
(t) c
k
(t) c
k+r1
(t) c
k+r
(t) c
k
(t + 1)
0 0 0 0 0 0
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
1 1 1 1 1 0
Exercise 9.13
How many 1-D BCA are there without memory? Legal CA? Symmetric CA?
Answer: 2
2
2r
; 2
2
2r+1
1
; 2
2
2r
.
Exercise 9.14
How many 1-D CA are there with the alphabet containing k symbols?
Answer: k
k
2r+1
.
We again consider binary CA, so let k = 2. The 2
2
2r+1
dierent right-most columns, representing
2
2
2r+1
various automata or transition rules, can be viewed as the binary expansions of the whole
numbers 0, 1, 2, . . . , 2
2
2r+1
1 . These decimal numbers are often considered as labels of dierent
transition rules. For example, since the decimal number 17 has the binary representation 00010001,
the rule 17 can be written as
c
k1
(t) c
k
(t) c
k+1
(t) c
k
(t + 1)
0 0 0 1
0 0 1 0
0 1 0 0
0 1 1 0
1 0 0 1
1 0 1 0
1 1 0 0
1 1 1 0
When k = 2, the charts above represent the truth tables of Boolean functions, that is, mappings
f : Z
2
Z
2
Z
2
Z
2
, where Z
2
Z
2
Z
2
is the Cartesian product of 2
2
2r+1
copies of Z
2
. In
this sense, the theory of 1-D BCA is equivalent to the theory of Boolean functions. In particular, we
can dene Boolean operations: negation, conjunction, disjunction, conditional implication on BCA.
For example, the negation of the rule 17 above is the rule 238, since (000100001) = (11101110),
which is the binary representation of 238.
We call a CA elementary (or an atom) if the binary representation of its rule contains exactly
one 1. Therefore, there are 2
2r+1
atoms (8 for r = 1).
Theorem 9.15 Each CA can be represented by the disjunction of the atoms.
??? Define dual, monotone, self-dual CA.
Examples.
??? Representation by Boolean polynomials.
??? Completeness.
22
References
[1] P. Bak, How Nature Works: the science of self-organized criticality, New York: Springer-
Verlag, 1996.
In this book, the author describes dierent models for evolution and natural phenomena. These
models are based in the area of cellular automata.
[2] E.R. Berlekamp, J.H. Conway, and R.K. Guy, Winning Ways for Your Mathematical Plays,
London: Academic Press, 1982.
This books contains mathematical formulations of games, and more. In particular, the Game
of Life (Chapter 25) is the most famous of all cellular automata.
[3] L.E. Dubins and L.J. Savage, How to Gamble If You Must: inequalities for stochastic processes,
New York: McGraw-Hill, 1965.
[4] A. Engel, The probabilistic abacus, Ed. Stud. Math. 6 (1975), 1-22.
[5] A. Engel, Why does the probabilistic abacus work?, Ed. Stud. Math. 7 (1976), 59-69.
The ideas in these two papers were the inspiration for the chip-ring games and other games
in Section 8, and they contain many more useful examples.
[6] S. Wolfram, Computer software in Science and Mathematics, Scientic American 251 (Septem-
ber 1984), 188-203.
[7] S. Wolfram, Geometry of binomial coecients, Amer. Math. Monthly 91 (1984), 566-570.
[8] S. Wolfram, Theory and Applications of Cellular Automata, New York: World Scientic, 1986.
Steven Wolfram is one of the pioneers in the area of cellular automata, and is the creator of the
computer algebra system Mathematica. Wolfram described the future of computer simulation,
and this article is still relevant today.
[9] http://psoup.math.wisc.edu The Primordial Soup Kitchen.
This is one of the premiere websites on cellular automata. Links, articles and software available.
See especially the following freeware programs:
MCell (Mireks Cellebration)
Automalab
Cellab
[10] http://www.dimacs.rutgers.edu The DIMACS Institute.
DIMACS is the Center for Discrete Mathematics and Theoretical Computer Science. Reconnect
is one of many programs organized by DIMACS, which is located at Rutgers University.
23