Smartworld.asia Specworld.

in

smartworlD.asia
Formal Languages & Automata Theory Complete Notes

Smartzworld.com 1 jntuworldupdates.org
Smartworld.asia Specworld.in

Contents

1 Mathematical Preliminaries 3

2 Formal Languages 4
2.1 Strings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Languages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.3 Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.4 Finite Representation . . . . . . . . . . . . . . . . . . . . . . . 13
2.4.1 Regular Expressions . . . . . . . . . . . . . . . . . . . 13

smartworlD.asia
3 Grammars 18
3.1 Context-Free Grammars . . . . . . . . . . . . . . . . . . . . . 19
3.2 Derivation Trees . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.2.1 Ambiguity . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.3 Regular Grammars . . . . . . . . . . . . . . . . . . . . . . . . 32
3.4 Digraph Representation . . . . . . . . . . . . . . . . . . . . . 36

4 Finite Automata 38
4.1 Deterministic Finite Automata . . . . . . . . . . . . . . . . . 39
4.2 Nondeterministic Finite Automata . . . . . . . . . . . . . . . 49
4.3 Equivalence of NFA and DFA . . . . . . . . . . . . . . . . . . 54
4.3.1 Heuristics to Convert NFA to DFA . . . . . . . . . . . 58
4.4 Minimization of DFA . . . . . . . . . . . . . . . . . . . . . . . 61
4.4.1 Myhill-Nerode Theorem . . . . . . . . . . . . . . . . . 61
4.4.2 Algorithmic Procedure for Minimization . . . . . . . . 65
4.5 Regular Languages . . . . . . . . . . . . . . . . . . . . . . . . 72
4.5.1 Equivalence of Finite Automata and Regular Languages 72
4.5.2 Equivalence of Finite Automata and Regular Grammars 84
4.6 Variants of Finite Automata . . . . . . . . . . . . . . . . . . . 89
4.6.1 Two-way Finite Automaton . . . . . . . . . . . . . . . 89
4.6.2 Mealy Machines . . . . . . . . . . . . . . . . . . . . . . 91

Smartzworld.com 2 jntuworldupdates.org
Smartworld.asia Specworld.in

5 Properties of Regular Languages 94

5.1 Closure Properties . . . . . . . . . . . . . . . . . . . . . . . . 94
5.1.1 Set Theoretic Properties . . . . . . . . . . . . . . . . . 94
5.1.2 Other Properties . . . . . . . . . . . . . . . . . . . . . 97
5.2 Pumping Lemma . . . . . . . . . . . . . . . . . . . . . . . . . 104

smartworlD.asia

Smartzworld.com 3 jntuworldupdates.org
Smartworld.asia Specworld.in

Chapter 1

Mathematical Preliminaries

smartworlD.asia

Smartzworld.com 4 jntuworldupdates.org
Smartworld.asia Specworld.in

Chapter 2

Formal Languages

A language can be seen as a system suitable for expression of certain ideas,

facts and concepts. For formalizing the notion of a language one must cover
all the varieties of languages such as natural (human) languages and program-
ming languages. Let us look at some common features across the languages.
One may broadly see that a language is a collection of sentences; a sentence
is a sequence of words; and a word is a combination of syllables. If one con-

smartworlD.asia
siders a language that has a script, then it can be observed that a word is a
sequence of symbols of its underlying alphabet. It is observed that a formal
learning of a language has the following three steps.

1. Learning its alphabet - the symbols that are used in the language.

2. Its words - as various sequences of symbols of its alphabet.

3. Formation of sentences - sequence of various words that follow certain

rules of the language.

In this learning, step 3 is the most difficult part. Let us postpone to discuss
construction of sentences and concentrate on steps 1 and 2. For the time
being instead of completely ignoring about sentences one may look at the
common features of a word and a sentence to agree upon both are just se-
quence of some symbols of the underlying alphabet. For example, the English
sentence
"The English articles - a, an and the - are
categorized into two types: indefinite and definite."
may be treated as a sequence of symbols from the Roman alphabet along
with enough punctuation marks such as comma, full-stop, colon and further
one more special symbol, namely blank-space which is used to separate two
words. Thus, abstractly, a sentence or a word may be interchangeably used

Smartzworld.com 5 jntuworldupdates.org
Smartworld.asia Specworld.in

for a sequence of symbols from an alphabet. With this discussion we start

with the basic definitions of alphabets and strings and then we introduce the
notion of language formally.
Further, in this chapter, we introduce some of the operations on languages
and discuss algebraic properties of languages with respect to those operations.
We end the chapter with an introduction to finite representation of languages
via regular expressions.

2.1 Strings
We formally define an alphabet as a non-empty finite set. We normally use
the symbols a, b, c, . . . with or without subscripts or 0, 1, 2, . . ., etc. for the
elements of an alphabet.
A string over an alphabet Σ is a finite sequence of symbols of Σ. Although
one writes a sequence as (a1 , a2 , . . . , an ), in the present context, we prefer to
write it as a1 a2 · · · an , i.e. by juxtaposing the symbols in that order. Thus,
a string is also known as a word or a sentence. Normally, we use lower case
letters towards the end of English alphabet, namely z, y, x, w, etc., to denote

smartworlD.asia
strings.

Example 2.1.1. Let Σ = {a, b} be an alphabet; then aa, ab, bba, baaba, . . .
are some examples of strings over Σ.

Since the empty sequence is a finite sequence, it is also a string. Which is

( ) in earlier notation; but with the notation adapted for the present context
we require a special symbol. We use ε, to denote the empty string.
The set of all strings over an alphabet Σ is denoted by Σ∗ . For example,
if Σ = {0, 1}, then

Σ∗ = {ε, 0, 1, 00, 01, 10, 11, 000, 001, . . .}.

Although the set Σ∗ is infinite, it is a countable set. In fact, Σ∗ is countably

infinite for any alphabet Σ. In order to understand some such fundamen-
tal facts we introduce some string operations, which in turn are useful to
manipulate and generate strings.
One of the most fundamental operations used for string manipulation is
concatenation. Let x = a1 a2 · · · an and y = b1 b2 · · · bm be two strings. The
concatenation of the pair x, y denoted by xy is the string

a1 a2 · · · an b1 b2 · · · bm .

Smartzworld.com 6 jntuworldupdates.org
Smartworld.asia Specworld.in

Clearly, the binary operation concatenation on Σ∗ is associative, i.e., for all

x, y, z ∈ Σ∗ ,
x(yz) = (xy)z.
Thus, x(yz) may simply be written as xyz. Also, since ε is the empty string,
it satisfies the property
εx = xε = x
for any sting x ∈ Σ∗ . Hence, Σ∗ is a monoid with respect to concatenation.
The operation concatenation is not commutative on Σ∗ .
For a string x and an integer n ≥ 0, we write
xn+1 = xn x with the base condition x0 = ε.
That is, xn is obtained by concatenating n copies of x. Also, whenever n = 0,
the string x1 · · · xn represents the empty string ε.
Let x be a string over an alphabet Σ. For a ∈ Σ, the number of occur-
rences of a in x shall be denoted by |x|a . The length of a string x denoted
by |x| is defined as X
|x| = |x|a .

smartworlD.asia
a∈Σ

Essentially, the length of a string is obtained by counting the number of

symbols in the string. For example, |aab| = 3, |a| = 1. Note that |ε| = 0.
If we denote An to be the set of all strings of length n over Σ, then one
can easily ascertain that [
Σ∗ = An .
n≥0

And hence, being An a finite set, Σ∗ is a countably infinite set.

We say that x is a substring of y if x occurs in y, that is y = uxv for
some strings u and v. The substring x is said to be a prefix of y if u = ε.
Similarly, x is a suffix of y if v = ε.
Generalizing the notation used for number of occurrences of symbol a in a
string x, we adopt the notation |y|x as the number of occurrences of a string
x in y.

2.2 Languages
We have got acquainted with the formal notion of strings that are basic
elements of a language. In order to define the notion of a language in a
broad spectrum, it is felt that it can be any collection of strings over an
alphabet.
Thus we define a language over an alphabet Σ as a subset of Σ∗ .

Smartzworld.com 7 jntuworldupdates.org
Smartworld.asia Specworld.in

Example 2.2.1.

1. The emptyset ∅ is a language over any alphabet. Similarly, {ε} is also

a language over any alphabet.

2. The set of all strings over {0, 1} that start with 0.

3. The set of all strings over {a, b, c} having ac as a substring.

Remark 2.2.2. Note that ∅ 6= {ε}, because the language ∅ does not contain
any string but {ε} contains a string, namely ε. Also it is evident that |∅| = 0;
whereas, |{ε}| = 1.
Since languages are sets, we can apply various well known set operations
such as union, intersection, complement, difference on languages. The notion
of concatenation of strings can be extended to languages as follows.
The concatenation of a pair of languages L1 , L2 is

L1 L2 = {xy | x ∈ L1 ∧ y ∈ L2 }.

Example 2.2.3.

smartworlD.asia
1. If L1 = {0, 1, 01} and L2 = {1, 00}, then
L1 L2 = {01, 11, 011, 000, 100, 0100}.

2. For L1 = {b, ba, bab} and L2 = {ε, b, bb, abb}, we have

L1 L2 = {b, ba, bb, bab, bbb, babb, baabb, babbb, bababb}.

Remark 2.2.4.

1. Since concatenation of strings is associative, so is the concatenation of

languages. That is, for all languages L1 , L2 and L3 ,

(L1 L2 )L3 = L1 (L2 L3 ).

Hence, (L1 L2 )L3 may simply be written as L1 L2 L3 .

2. The number of strings in L1 L2 is always less than or equal to the

product of individual numbers, i.e.

|L1 L2 | ≤ |L1 ||L2 |.

3. L1 ⊆ L1 L2 if and only if ε ∈ L2 .

Smartzworld.com 8 jntuworldupdates.org
Smartworld.asia Specworld.in

Proof. The “if part” is straightforward; for instance, if ε ∈ L2 , then

for any x ∈ L1 , we have x = xε ∈ L1 L2 . On the other hand, suppose
ε∈/ L2 . Now, note that a string x ∈ L1 of shortest length in L1 cannot
be in L1 L2 . This is because, if x = yz for some y ∈ L1 and a nonempty
string z ∈ L2 , then |y| < |x|. A contradiction to our assumption that
x is of shortest length in L1 . Hence L1 6⊆ L1 L2 .

4. Similarly, ε ∈ L1 if and only if L2 ⊆ L1 L2 .

We write Ln to denote the language which is obtained by concatenating
n copies of L. More formally,
L0 = {ε} and
Ln = Ln−1 L, for n ≥ 1.
In the context of formal languages, another important operation is Kleene
star. Kleene star or Kleene closure of a language L, denoted by L∗ , is defined
as [
L∗ = Ln .
n≥0

Example 2.2.5.

smartworlD.asia
1. Kleene star of the language {01} is
{ε, 01, 0101, 010101, . . .} = {(01)n | n ≥ 0}.

2. If L = {0, 10}, then L∗ = {ε, 0, 10, 00, 010, 100, 1010, 000, . . .}
n
Since
[ an arbitrary string in L is of the form x1 x2 · · · xn , for xi ∈ L and
L∗ = Ln , one can easily observe that
n≥0

L∗ = {x1 x2 · · · xn | n ≥ 0 and xi ∈ L, for 1 ≤ i ≤ n}

Thus, a typical string in L∗ is a concatenation of finitely many strings of L.

Remark 2.2.6. Note that, the Kleene star of the language L = {0, 1} over
the alphabet Σ = {0, 1} is

L∗ = L0 ∪ L ∪ L2 ∪ · · ·
= {ε} ∪ {0, 1} ∪ {00, 01, 10, 11} ∪ · · ·
= {ε, 0, 1, 00, 01, 10, 11, · · · }
= the set of all strings over Σ.

Thus, the earlier introduced notation Σ∗ is consistent with the notation of

Kleene star by considering Σ as a language over Σ.

Smartzworld.com 9 jntuworldupdates.org
Smartworld.asia Specworld.in

The positive closure of a language L is denoted by L+ is defined as

[
L+ = Ln .
n≥1

Thus, L∗ = L+ ∪ {ε}.
We often can easily describe various formal languages in English by stat-
ing the property that is to be satisfied by the strings in the respective lan-
guages. It is not only for elegant representation but also to understand the
properties of languages better, describing the languages in set builder form
is desired.
Consider the set of all strings over {0, 1} that start with 0. Note that
each such string can be seen as 0x for some x ∈ {0, 1}∗ . Thus the language
can be represented by
{0x | x ∈ {0, 1}∗ }.

Examples
1. The set of all strings over {a, b, c} that have ac as substring can be
written as

smartworlD.asia
{xacy | x, y ∈ {a, b, c}∗ }.
This can also be written as
{x ∈ {a, b, c}∗ | |x|ac ≥ 1},
stating that the set of all strings over {a, b, c} in which the number of
occurrences of substring ac is at least 1.
2. The set of all strings over some alphabet Σ with even number of a0 s is
{x ∈ Σ∗ | |x|a = 2n, for some n ∈ N}.
Equivalently,
{x ∈ Σ∗ | |x|a ≡ 0 mod 2}.

3. The set of all strings over some alphabet Σ with equal number of a0 s
and b0 s can be written as
{x ∈ Σ∗ | |x|a = |x|b }.

4. The set of all palindromes over an alphabet Σ can be written as

{x ∈ Σ∗ | x = xR },
where xR is the string obtained by reversing x.

Smartzworld.com 10 jntuworldupdates.org
Smartworld.asia Specworld.in

5. The set of all strings over some alphabet Σ that have an a in the 5th
position from the right can be written as

{xay | x, y ∈ Σ∗ and |y| = 4}.

6. The set of all strings over some alphabet Σ with no consecutive a0 s can
be written as
{x ∈ Σ∗ | |x|aa = 0}.

7. The set of all strings over {a, b} in which every occurrence of b is not
before an occurrence of a can be written as

{am bn | m, n ≥ 0}.

Note that, this is the set of all strings over {a, b} which do not contain
ba as a substring.

2.3 Properties

smartworlD.asia
The usual set theoretic properties with respect to union, intersection, comple-
ment, difference, etc. hold even in the context of languages. Now we observe
certain properties of languages with respect to the newly introduced oper-
ations concatenation, Kleene closure, and positive closure. In what follows,
L, L1 , L2 , L3 and L4 are languages.

P1 Recall that concatenation of languages is associative.

P2 Since concatenation of strings is not commutative, we have L1 L2 6= L2 L1 ,

in general.

P3 L{ε} = {ε}L = L.

P4 L∅ = ∅L = ∅.

Proof. Let x ∈ L∅; then x = x1 x2 for some x1 ∈ L and x2 ∈ ∅. But ∅

being emptyset cannot hold any element. Hence there cannot be any
element x ∈ L∅ so that L∅ = ∅. Similarly, ∅L = ∅ as well.

P5 Distributive Properties:

1. (L1 ∪ L2 )L3 = L1 L3 ∪ L2 L3 .

10

Smartzworld.com 11 jntuworldupdates.org
Smartworld.asia Specworld.in

Proof. Suppose x ∈ (L1 ∪ L2 )L3

=⇒ x = x1 x2 , for some x1 ∈ L1 ∪ L2 ), and some x2 ∈ L3

=⇒ x = x1 x2 , for some x1 ∈ L1 or x1 ∈ L2 , and x2 ∈ L3
=⇒ x = x1 x2 , for some x1 ∈ L1 and x2 ∈ L3 ,
or x1 ∈ L2 and x2 ∈ L3
=⇒ x ∈ L1 L3 or x ∈ L2 L3
=⇒ x ∈ L1 L3 ∪ L2 L3 .

Conversely, suppose x ∈ L1 L3 ∪ L2 L3 =⇒ x ∈ L1 L3 or x ∈ L2 L3 .
Without loos of generality, assume x 6∈ L1 L3 . Then x ∈ L2 L3 .

=⇒ x = x3 x4 , for some x3 ∈ L2 and x4 ∈ L3

=⇒ x = x3 x4 , for some x3 ∈ L1 ∪ L2 , and some x4 ∈ L3
=⇒ x ∈ (L1 ∪ L2 )L3 .

Hence, (L1 ∪ L2 )L3 = L1 L3 ∪ L2 L3 .

smartworlD.asia
2. L1 (L2 ∪ L3 ) = L1 L2 ∪ L1 L3 .

Proof. Similar to the above.

From these properties it is clear that the concatenation is distributive

over finite unions. Moreover, we can observe that concatenation is also
distributive over countably infinite unions. That is,
à !
[ [
L Li = LLi and
i≥1 i≥1
à !
[ [
Li L = Li L
i≥1 i≥1

P6 If L1 ⊆ L2 and L3 ⊆ L4 , then L1 L3 ⊆ L2 L4 .

P7 ∅∗ = {ε}.

P8 {ε}∗ = {ε}.

P9 If ε ∈ L, then L∗ = L+ .

P10 L∗ L = LL∗ = L+ .

11

Smartzworld.com 12 jntuworldupdates.org
Smartworld.asia Specworld.in

Proof. Suppose x ∈ L∗ L. Then x = yz for some y ∈ L∗ and z ∈ L.

But y ∈ L∗ implies y = y1 · · · yn with yi ∈ L for all i. Hence,

x = yz = (y1 · · · yn )z = y1 (y2 · · · yn z) ∈ LL∗ .

Converse is similar. Hence, L∗ L = LL∗ .

Further, when x ∈ L∗ L, as above, we have x = y1 · · · yn z is clearly in
L+ . On the other hand, x ∈ L+ implies x = x1 · · · xm with m ≥ 1
and xi ∈ L for all i. Now write x0 = x1 · · · xm−1 so that x = x0 xm .
Here, note that x0 ∈ L∗ ; particularly, when m = 1 then x0 = ε. Thus,
x ∈ L∗ L. Hence, L+ = L∗ L.

P11 (L∗ )∗ = L∗ .

P12 L∗ L∗ = L∗ .

P13 (L1 L2 )∗ L1 = L1 (L2 L1 )∗ .

Proof. Let x ∈ (L1 L2 )∗ L1 . Then x = yz, where z ∈ L1 and y =

smartworlD.asia
y1 · · · yn ∈ (L1 L2 )∗ with yi ∈ L1 L2 . Now each yi = ui vi , for ui ∈ L1 and
vi ∈ L2 . Note that vi ui+1 ∈ L2 L1 , for all i with 1 ≤ i ≤ n − 1. Hence,
x = yz = (y1 · · · yn )z = (u1 v1 · · · un vn )z = u1 (v1 u2 · · · vn−1 un vn z) ∈
L1 (L2 L1 )∗ . Converse is similar. Hence, (L1 L2 )∗ L1 = L1 (L2 L1 )∗ .

P14 (L1 ∪ L2 )∗ = (L∗1 L∗2 )∗ .

Proof. Observe that L1 ⊆ L∗1 and {ε} ⊆ L∗2 . Hence, by properties P3

and P6, we have L1 = L1 {ε} ⊆ L∗1 L∗2 . Similarly, L2 ⊆ L∗1 L∗2 . Hence,
L1 ∪ L2 ⊆ L∗1 L∗2 . Consequently, (L1 ∪ L2 )∗ ⊆ (L∗1 L∗2 )∗ .
For converse, observe that L∗1 ⊆ (L1 ∪ L2 )∗ . Similarly, L∗2 ⊆ (L1 ∪ L2 )∗ .
Thus,
L∗1 L∗2 ⊆ (L1 ∪ L2 )∗ (L1 ∪ L2 )∗ .
But, by property P12, we have (L1 ∪ L2 )∗ (L1 ∪ L2 )∗ = (L1 ∪ L2 )∗ so
that L∗1 L∗2 ⊆ (L1 ∪ L2 )∗ . Hence,

(L∗1 L∗2 )∗ ⊆ ((L1 ∪ L2 )∗ )∗ = (L1 ∪ L2 )∗ .

12

Smartzworld.com 13 jntuworldupdates.org
Smartworld.asia Specworld.in

2.4 Finite Representation

Proficiency in a language does not expect one to know all the sentences
of the language; rather with some limited information one should be able
to come up with all possible sentences of the language. Even in case of
programming languages, a compiler validates a program - a sentence in the
programming language - with a finite set of instructions incorporated in it.
Thus, we are interested in a finite representation of a language - that is, by
giving a finite amount of information, all the strings of a language shall be
enumerated/validated.
Now, we look at the languages for which finite representation is possible.
Given an alphabet Σ, to start with, the languages with single string {x}
and ∅ can have finite representation, say x and ∅, respectively. In this way,
finite languages can also be given a finite representation; say, by enumerating
all the strings. Thus, giving finite representation for infinite languages is a
nontrivial interesting problem. In this context, the operations on languages
may be helpful.
For example, the infinite language {ε, ab, abab, ababab, . . .} can be con-
sidered as the Kleene star of the language {ab}, that is {ab}∗ . Thus, using

smartworlD.asia
Kleene star operation we can have finite representation for some infinite lan-
guages.
While operations are under consideration, to give finite representation for
languages one may first look at the indivisible languages, namely ∅, {ε}, and
{a}, for all a ∈ Σ, as basis elements.
To construct {x}, for x ∈ Σ∗ , we can use the operation concatenation
over the basis elements. For example, if x = aba then choose {a} and {b};
and concatenate {a}{b}{a} to get {aba}. Any finite language over Σ, say
{x1 , . . . , xn } can be obtained by considering the union {x1 } ∪ · · · ∪ {xn }.
In this section, we look at the aspects of considering operations over
basis elements to represent a language. This is one aspect of representing a
language. There are many other aspects to give finite representations; some
such aspects will be considered in the later chapters.

2.4.1 Regular Expressions

We now consider the class of languages obtained by applying union, con-
catenation, and Kleene star for finitely many times on the basis elements.
These languages are known as regular languages and the corresponding finite
representations are known as regular expressions.
Definition 2.4.1 (Regular Expression). We define a regular expression over
an alphabet Σ recursively as follows.

13

Smartzworld.com 14 jntuworldupdates.org
Smartworld.asia Specworld.in

1. ∅, ε, and a, for each a ∈ Σ, are regular expressions representing the

languages ∅, {ε}, and {a}, respectively.
2. If r and s are regular expressions representing the languages R and S,
respectively, then so are
(a) (r + s) representing the language R ∪ S,
(b) (rs) representing the language RS, and
(c) (r∗ ) representing the language R∗ .
In a regular expression we keep a minimum number of parenthesis which
are required to avoid ambiguity in the expression. For example, we may
simply write r + st in case of (r + (st)). Similarly, r + s + t for ((r + s) + t).
Definition 2.4.2. If r is a regular expression, then the language represented
by r is denoted by L(r). Further, a language L is said to be regular if there
is a regular expression r such that L = L(r).
Remark 2.4.3.
1. A regular language over an alphabet Σ is the one that can be obtained

smartworlD.asia
from the emptyset, {ε}, and {a}, for a ∈ Σ, by finitely many applica-
tions of union, concatenation and Kleene star.
2. The smallest class of languages over an alphabet Σ which contains
∅, {ε}, and {a} and is closed with respect to union, concatenation,
and Kleene star is the class of all regular languages over Σ.
Example 2.4.4. As we observed earlier that the languages ∅, {ε}, {a}, and
all finite sets are regular.
Example 2.4.5. {an | n ≥ 0} is regular as it can be represented by the
expression a∗ .
Example 2.4.6. Σ∗ , the set of all strings over an alphabet Σ, is regular. For
instance, if Σ = {a1 , a2 , . . . , an }, then Σ∗ can be represented as (a1 + a2 +
· · · + an )∗ .
Example 2.4.7. The set of all strings over {a, b} which contain ab as a
substring is regular. For instance, the set can be written as
{x ∈ {a, b}∗ | ab is a substring of x}
= {yabz | y, z ∈ {a, b}∗ }
= {a, b}∗ {ab}{a, b}∗
Hence, the corresponding regular expression is (a + b)∗ ab(a + b)∗ .

14

Smartzworld.com 15 jntuworldupdates.org
Smartworld.asia Specworld.in

Example 2.4.8. The language L over {0, 1} that contains 01 or 10 as sub-

string is regular.

L = {x | 01 is a substring of x} ∪ {x | 10 is a substring of x}
= {y01z | y, z ∈ Σ∗ } ∪ {u10v | u, v ∈ Σ∗ }
= Σ∗ {01}Σ∗ ∪ Σ∗ {10}Σ∗

Since Σ∗ , {01}, and {10} are regular we have L to be regular. In fact, at this
point, one can easily notice that

(0 + 1)∗ 01(0 + 1)∗ + (0 + 1)∗ 10(0 + 1)∗

is a regular expression representing L.

Example 2.4.9. The set of all strings over {a, b} which do not contain ab
as a substring. By analyzing the language one can observe that precisely the
language is as follows.
{bn am | m, n ≥ 0}
Thus, a regular expression of the language is b∗ a∗ and hence the language is

smartworlD.asia
regular.

Example 2.4.10. The set of strings over {a, b} which contain odd number
of a0 s is regular. Although the set can be represented in set builder form as

{x ∈ {a, b}∗ | |x|a = 2n + 1, for some n},

writing a regular expression for the language is little tricky job. Hence, we
postpone the argument to Chapter 3 (see Example 3.3.6), where we construct
a regular grammar for the language. Regular grammar is a tool to generate
regular languages.

Example 2.4.11. The set of strings over {a, b} which contain odd number
of a0 s and even number of b0 s is regular. As above, a set builder form of the
set is:

{x ∈ {a, b}∗ | |x|a = 2n + 1, for some n and |x|b = 2m, for some m}.

Writing a regular expression for the language is even more trickier than the
earlier example. This will be handled in Chapter 4 using finite automata,
yet another tool to represent regular languages.

Definition 2.4.12. Two regular expressions r1 and r2 are said to be equiv-

alent if they represent the same language; in which case, we write r1 ≈ r2 .

15

Smartzworld.com 16 jntuworldupdates.org
Smartworld.asia Specworld.in

Example 2.4.13. The regular expressions (10+1)∗ and ((10)∗ 1∗ )∗ are equiv-
alent.
Since L((10)∗ ) = {(10)n | n ≥ 0} and L(1∗ ) = {1m | m ≥ 0}, we have
L((10)∗ 1∗ ) = {(10)n 1m | m, n ≥ 0}. This implies

L(((10)∗ 1∗ )∗ ) = {(10)n1 1m1 (10)n2 1m2 · · · (10)nl 1ml | mi , ni ≥ 0 and 0 ≤ i ≤ l}

Xl
= {x1 x2 · · · xk | xi = 10 or 1}, where k = (mi + ni )
i=0
⊆ L((10 + 1)∗ ).

Conversely, suppose x ∈ L((10 + 1)∗ ). Then,

x = x1 x2 · · · xp where xi = 10 or 1
=⇒ x = (10)p1 1q1 (10)p2 1q2 · · · (10)pr 1qr for pi , qj ≥ 0
=⇒ x ∈ L(((10)∗ 1∗ )∗ ).

Hence, L((10+1)∗ ) = L(((10)∗ 1∗ )∗ ) and consequently, (10+ 1)∗ ≈ ((10)∗ 1∗ )∗ .

smartworlD.asia
From property P14, by choosing L1 = {10} and L2 = {1}, one may notice
that
({10} ∪ {1})∗ = ({10}∗ {1}∗ )∗ .
Since 10 and 1 represent the regular languages {10} and {1}, respectively,
from the above equation we get

(10 + 1)∗ ≈ ((10)∗ 1∗ )∗ .

Since those properties hold good for all languages, by specializing those prop-
erties to regular languages and in turn replacing by the corresponding regular
expressions we get the following identities for regular expressions.
Let r, r1 , r2 , and r3 be any regular expressions

1. rε ≈ εr ≈ r.

2. r1 r2 6≈ r2 r1 , in general.

3. r1 (r2 r3 ) ≈ (r1 r2 )r3 .

4. r∅ ≈ ∅r ≈ ∅.

5. ∅∗ ≈ ε.

6. ε∗ ≈ ε.

16

Smartzworld.com 17 jntuworldupdates.org
Smartworld.asia Specworld.in

7. If ε ∈ L(r), then r∗ ≈ r+ .

8. rr∗ ≈ r∗ r ≈ r+ .

9. (r1 + r2 )r3 ≈ r1 r3 + r2 r3 .

10. r1 (r2 + r3 ) ≈ r1 r2 + r1 r3 .

11. (r∗ )∗ ≈ r∗ .

12. (r1 r2 )∗ r1 ≈ r1 (r2 r1 )∗ .

13. (r1 + r2 )∗ ≈ (r1∗ r2∗ )∗ .

Example 2.4.14.

1. b+ (a∗ b∗ + ε)b ≈ b(b∗ a∗ + ε)b+ ≈ b+ a∗ b+ .

Proof.

b+ (a∗ b∗ + ε)b ≈ (b+ a∗ b∗ + b+ ε)b

smartworlD.asia
≈ b+ a∗ b∗ b + b+ b
≈ b+ a∗ b+ + b+ b
≈ b+ a∗ b+ , since L(b+ b) ⊆ L(b+ a∗ b+ ).

Similarly, one can observe that b(b∗ a∗ + ε)b+ ≈ b+ a∗ b+ .

2. (0+ (01)∗ 0 + 0∗ (10)∗ )∗ ≈ (0 + 10)∗ .

Proof.

(0+ (01)∗ 0 + 0∗ (10)∗ )∗ ≈ (0+ 0(10)∗ + 0∗ (10)∗ )∗

≈ ((0+ 0 + 0∗ )(10)∗ )∗
≈ (0∗ (10)∗ )∗ , since L(0+ 0) ⊆ L(0∗ )
≈ (0 + 10)∗ .

Notation 2.4.15. If L is represented by a regular expression r, i.e. L(r) = L,

then we may simply use r instead of L(r) to indicated the language L. As
a consequence, for two regular expressions r and r0 , r ≈ r0 and r = r0 are
equivalent.

17

Smartzworld.com 18 jntuworldupdates.org
Smartworld.asia Specworld.in

Chapter 3

Grammars

In this chapter, we introduce the notion of grammar called context-free gram-

mar (CFG) as a language generator. The notion of derivation is instrumental
in understanding how the strings are generated in a grammar. We explain
the various properties of derivations using a graphical representation called
derivation trees. A special case of CFG, viz. regular grammar, is discussed
as tool to generate to regular languages. A more general notion of grammars

smartworlD.asia
is presented in Chapter 7.
In the context of natural languages, the grammar of a language is a set of
rules which are used to construct/validate sentences of the language. It has
been pointed out, in the introduction of Chapter 2, that this is the third step
in a formal learning of a language. Now we draw the attention of a reader
to look into the general features of the grammars (of natural languages) to
formalize the notion in the present context which facilitate for better under-
standing of formal languages. Consider the English sentence

The students study automata theory.

In order to observe that the sentence is grammatically correct, one may

attribute certain rules of the English grammar to the sentence and validate
it. For instance, the Article the followed by the Noun students form a
Noun-phrase and similarly the Noun automata theory form a Noun-phrase.
Further, study is a Verb. Now, choose the Sentential form “Subject Verb
Object” of the English grammar. As Subject or Object can be a Noun-phrase
by plugging in the above words one may conclude that the given sentence
is a grammatically correct English sentence. This verification/derivation is
depicted in Figure 3.1. The derivation can also be represented by a tree
structure as shown in Figure 3.2.

18

Smartzworld.com 19 jntuworldupdates.org
Smartworld.asia Specworld.in

Sentence ⇒ Subject Verb Object

⇒ Noun-phrase Verb Object
⇒ Article Noun Verb Object
⇒ The Noun Verb Object
⇒ The students Verb Object
⇒ The students study Object
⇒ The students study Noun-phrase
⇒ The students study Noun
⇒ The students study automata theory
Figure 3.1: Derivation of an English Sentence

In this process, we observe that two types of words are in the discussion.
1. The words like the, study, students.

2. The words like Article, Noun, Verb.

The main difference is, if you arrive at a stage where type (1) words are

smartworlD.asia
appearing, then you need not say anything more about them. In case you
arrive at a stage where you find a word of type (2), then you are assumed
to say some more about the word. For example, if the word Article comes,
then one should say which article need to be chosen among a, an and the.
Let us call the type (1) and type (2) words as terminals and nonterminals,
respectively, as per their features.
Thus, a grammar should include terminals and nonterminals along with a
set of rules which attribute some information regarding nonterminal symbols.

3.1 Context-Free Grammars

We now understand that a grammar should have the following components.
• A set of nonterminal symbols.

• A set of terminal symbols.

• A set of rules.

• As the grammar is to construct/validate sentences of a language, we

distinguish a symbol in the set of nonterminals to represent a sen-
tence – from which various sentences of the language can be gener-
ated/validated.

19

Smartzworld.com 20 jntuworldupdates.org
Smartworld.asia Specworld.in

Sentence

Subject Verb Object

Noun−phrase Noun−phrase

Article Noun Noun

The students study automata theory

Figure 3.2: Derivation Tree of an English Sentence

smartworlD.asia
With this, we formally define the notion of grammar as below.
Definition 3.1.1. A grammar is a quadruple
G = (N, Σ, P, S)
where
1. N is a finite set of nonterminals,
2. Σ is a finite set of terminals,
3. S ∈ N is the start symbol, and
4. P is a finite subset of N × V ∗ called the set of production rules. Here,
V = N ∪ Σ.
It is convenient to write A → α, for the production rule (A, α) ∈ P .
To define a formal notion of validating or deriving a sentence using a
grammar, we require the following concepts.
Definition 3.1.2. Let G = (N, Σ, P, S) be a grammar with V = N ∪ Σ.
1. We define a binary relation ⇒
G
on V ∗ by
α⇒
G
β if and only if α = α1 Aα2 , β = α1 γα2 and A → γ ∈ P,
for all α, β ∈ V ∗ .

20

Smartzworld.com 21 jntuworldupdates.org
Smartworld.asia Specworld.in

2. The relation ⇒
G
is called one step relation on G. If α ⇒
G
β, then we call
α yields β in one step in G.

3. The reflexive-transitive closure of ⇒ is denoted by ⇒ ∗ . That is, for

G G
α, β ∈ V ∗ ,
½
∗ ∃n ≥ 0 and α0 , α1 , . . . , αn ∈ V ∗ such that
α⇒ β if and only if
G α = α0 ⇒ G
α1 ⇒G
··· ⇒ G
αn−1 ⇒G
αn = β.
∗ β, then we say β is derived from α or α derives
4. For α, β ∈ V ∗ , if α ⇒
G
β. Further, α ⇒ ∗ β is called as a derivation in G.
G

5. If α = α0 ⇒
G
α1 ⇒
G
··· ⇒
G
αn−1 ⇒G
αn = β is a derivation, then the length
n
of the derivation is n and it may be written as α ⇒G
β.

6. In a given context, if we deal with only one grammar G, then we may

simply write ⇒, in stead of ⇒G
.
∗ β is a derivation, then we say β is the yield of the derivation.
7. If α ⇒

8. A string α ∈ V ∗ is said to be a sentential form in G, if α can be derived

smartworlD.asia
from the start symbol S of G. That is, S ⇒ ∗ α.

9. In particular, if α ∈ Σ∗ , then the sentential form α is known as a

sentence. In which case, we say α is generated by G.

10. The language generated by G, denoted by L(G), is the set of all sen-
tences generated by G. That is,
∗ x}.
L(G) = {x ∈ Σ∗ | S ⇒

Note that a production rule of a grammar is of the form A → α, where

A is a nonterminal symbol. If the nonterminal A appears in a sentential
form X1 · · · Xk AXk+1 · · · Xn , then the sentential form X1 · · · Xk αXk+1 · · · Xn
can be obtained in one step by replacing A with α. This replacement is
independent of the neighboring symbols of A in X1 · · · Xk AXk+1 · · · Xn . That
is, Xi0 s will not play any role in the replacement. One may call A is within
the context of Xi0 s and hence the rules A → α are said to be of context-free
type. Thus, the type of grammar that is defined here is known as context-
free grammar, simply CFG. In the later chapters, we relax the constraint and
discuss more general types of grammars.

21

Smartzworld.com 22 jntuworldupdates.org
Smartworld.asia Specworld.in

Example 3.1.3. Let P = {S → ab, S → bb, S → aba, S → aab} with

Σ = {a, b} and N = {S}. Then G = (N, Σ, P, S) is a context-free grammar.
Since left hand side of each production rule is the start symbol S and their
right hand sides are terminal strings, every derivation in G is of length one.
In fact, we precisely have the following derivation in G.

1. S ⇒ ab

2. S ⇒ bb

3. S ⇒ aba

4. S ⇒ aab

Hence, the language generated by G,

L(G) = {ab, bb, aba, aab}.

Notation 3.1.4.

1. A → α1 , A → α2 can be written as A → α1 | α2 .

smartworlD.asia
2. Normally we use S as the start symbol of a grammar, unless otherwise
specified.

3. To give a grammar G = (N, Σ, P, S), it is sufficient to give the produc-

tion rules only since one may easily find the other components N and
Σ of the grammar G by looking at the rules.

Example 3.1.5. Suppose L is a finite language over an alphabet Σ, say L =

{x1 , x2 , . . . , xn }. Then consider the finite set P = {S → x1 | x2 | · · · | xn }.
Now, as discussed in Example 3.1.3, one can easily observe that the CFG
({S}, Σ, P, S) generates the language L.

Example 3.1.6. Let Σ = {0, 1} and N = {S}. Consider the CFG G =

(N, Σ, P, S), where P has precisely the following production rules.

1. S → 0S

2. S → 1S

3. S → ε

22

Smartzworld.com 23 jntuworldupdates.org
Smartworld.asia Specworld.in

We now observe that the CFG G generates Σ∗ . As every string generated

by G is a sequence of 00 s and 10 s, clearly we have L(G) ⊆ Σ∗ . On the other
hand, let x ∈ Σ∗ . If x = ε, then x can be generated in one step using the
rule S → ε. Otherwise, let x = a1 a2 · · · an , where ai = 0 or 1 for all i. Now,
as shown below, x can be derived in G.

S ⇒ a1 S (if a1 = 0, then by rule 1; else by rule 2)

⇒ a1 a2 S (as above)
..
.
⇒ a1 a2 · · · an S (as above)
⇒ a1 a2 · · · an ε (by rule 3)
= x.

Hence, x ∈ L(G). Thus, L(G) = Σ∗ .

Example 3.1.7. The language ∅ (over any alphabet Σ) can be generated

by a CFG.

Method-I. If the set of productions P is empty, then clearly the CFG G =

smartworlD.asia
({S}, Σ, P, S) does not generate any string and hence L(G) = ∅.

Method-II. Consider a CFG G in which each production rule has some non-
terminal symbol on its right hand side. Clearly, no terminal string can be
generated in G so that L(G) = ∅.

Example 3.1.8. Consider G = ({S}, {a}, P, S) where P has the following

rules

1. S → aS

2. S → ε

Let us look at the strings that can be generated by G. Clearly, ε can be

generated in one step by using the rule S → ε. Further, if we choose rule (1)
and then rule (2) we get the string a in two steps as follows:

S ⇒ aS ⇒ aε = a

If we use the rule (1), then one may notice that there will always be the
nonterminal S in the resulting sentential form. A derivation can only be
terminated by using rule (2). Thus, for any derivation of length k, that

23

Smartzworld.com 24 jntuworldupdates.org
Smartworld.asia Specworld.in

derives some string x, we would have used rule (1) for k − 1 times and rule
(2) once at the end. Precisely, the derivation will be of the form
S ⇒ aS
⇒ aaS
..
.
k−1
z }| {
⇒ aa · · · a S
k−1
z }| {
⇒ aa · · · a ε = ak−1
Hence, it is clear that L(G) = {ak | k ≥ 0}
In the following, we give some more examples of typical CFGs.
Example 3.1.9. Consider the grammar having the following production
rules:
S → aSb | ε
One may notice that the rule S → aSb should be used to derive strings other
than ε, and the derivation shall always be terminated by S → ε. Thus, a

smartworlD.asia
typical derivation is of the form
S ⇒ aSb
⇒ aaSbb
..
.
⇒ an Sbn
⇒ an εbn = an bn
Hence, L(G) = {an bn | n ≥ 0}.
Example 3.1.10. The grammar
S → aSa | bSb | a | b | ε
generates the set of all palindromes over {a, b}. For instance, the rules S →
aSa and S → bSb will produce same terminals at the same positions towards
left and right sides. While terminating the derivation the rules S → a | b
or S → ε will produce odd or even length palindromes, respectively. For
example, the palindrome abbabba can be derived as follows.
S ⇒ aSa
⇒ abSba
⇒ abbSbba
⇒ abbabba

24

Smartzworld.com 25 jntuworldupdates.org
Smartworld.asia Specworld.in

Example 3.1.11. Consider the language L = {am bn cm+n | m, n ≥ 0}. We

now give production rules of a CFG which generates L. As given in Example
3.1.9, the production rule
S → aSc
can be used to produce equal number of a0 s and c0 s, respectively, in left and
right extremes of a string. In case, there is no b in the string, the production
rule
S→ε
can be used to terminate the derivation and produce a string of the form
am cm . Note that, b0 s in a string may have leading a0 s; but, there will not be
any a after a b. Thus, a nonterminal symbol that may be used to produce
b0 s must be different from S. Hence, we choose a new nonterminal symbol,
say A, and define the rule
S→A
to handover the job of producing b0 s to A. Again, since the number of b0 s
and c0 s are to be equal, choose the rule

A → bAc

smartworlD.asia
to produce b0 s and c0 s on either sides. Eventually, the rule

A→ε

can be introduced, which terminate the derivation. Thus, we have the fol-
lowing production rules of a CFG, say G.

S → aSc | A | ε

A → bAc | ε
Now, one can easily observe that L(G) = L.

Example 3.1.12. For the language {am bm+n cn | m, n ≥ 0}, one may think
in the similar lines of Example 3.1.11 and produce a CFG as given below.

S → AB
A → aAb | ε
B → bBc | ε

25

Smartzworld.com 26 jntuworldupdates.org
Smartworld.asia Specworld.in

S ⇒ S∗S S ⇒ S∗S S ⇒ S+S

⇒ S∗a ⇒ S+S∗S ⇒ a+S
⇒ S+S∗a ⇒ a+S∗S ⇒ a+S∗S
⇒ a+S∗a ⇒ a+b∗S ⇒ a+b∗S
⇒ a+b∗a ⇒ a+b∗a ⇒ a+b∗a

(1) (2) (3)

Figure 3.3: Derivations for the string a + b ∗ a

3.2 Derivation Trees

Let us consider the CFG whose production rules are:
S → S ∗ S | S + S | (S) | a | b

Figure 3.3 gives three derivations for the string a + b ∗ a. Note that the
three derivations are different, because of application of different sequences of

smartworlD.asia
rules. Nevertheless, the derivations (1) and (2) share the following feature.
A nonterminal that appears at a particular common position in both the
derivations derives the same substring of a + b ∗ a in both the derivations.
In contrast to that, the derivations (2) and (3) are not sharing such feature.
For example, the second S in step 2 of derivations (1) and (2) derives the
substring a; whereas, the second S in step 2 of derivation (3) derives the
substring b ∗ a. In order to distinguish this feature between the derivations
of a string, we introduce a graphical representation of a derivation called
derivation tree, which will be a useful tool for several other purposes also.
Definition 3.2.1. Let G = (N, Σ, P, S) be a CFG and V = N ∪ Σ. For
A ∈ N and α ∈ V ∗ , suppose
A⇒ ∗ α

is a derivation in G. A derivation tree or a parse tree of the derivation is

iteratively defined as follows.
1. A is the root of the tree.
2. If the rule B → X1 X2 · · · Xk is applied in the derivation, then new
nodes with labels X1 , X2 , . . . , Xk are created and made children to the
node B from left to right in that order.
The construction of the derivation tree of a derivation is illustrated with
the following examples.

26

Smartzworld.com 27 jntuworldupdates.org
Smartworld.asia Specworld.in

Example 3.2.2. Consider the following derivation in the CFG given in Ex-
ample 3.1.9.
S ⇒ aSb
⇒ aaSbb
⇒ aaεbb
= aabb
∗ aabb is shown below.
The derivation tree of the above derivation S ⇒

S=
¢¢ ==
¢¢ =
a S= b
¢¢ ==
¢¢ =
a S b
ε
Example 3.2.3. Consider the following derivation in the CFG given in Ex-
ample 3.1.12.

smartworlD.asia
S ⇒ AB
⇒ aAbB
⇒ aaAbbB
⇒ aaAbbB
⇒ aaAbbbBc
⇒ aaAbbbεc
= aaAbbbc
∗ aaAbbbc is shown below.
The derivation tree of the above derivation S ⇒

p S NNNN
ppppp NNN
pp N
A= B>
¡¡ == ¡¡ >>
¡¡ = ¡¡ >
a A= b b B c
¡¡ ==
¡¡ =
a A b ε

Note that the yield α of the derivation A ⇒ ∗ α can be identified in the

derivation tree by juxtaposing the labels of the leaf nodes from left to right.
Example 3.2.4. Recall the derivations given in Figure 3.3. The derivation
trees corresponding to these derivations are in Figure 3.4.

27

Smartzworld.com 28 jntuworldupdates.org
Smartworld.asia Specworld.in

S
{ CCCC
S
{ CCCC {
S CC
{{ {{ {{ CC
{{ C {{ C {{ C
S CC ∗ S S CC ∗ S S + S
{ ???
ÄÄÄ CC
C ÄÄÄ CC
C {{{ ?
Ä Ä {
S + S a S + S a a S ∗ S

a b a b b a

(1) (2) (3)

Figure 3.4: Derivation Trees for Figure 3.3

Now we are in a position to formalize the notion of equivalence between

derivations, which precisely captures the feature proposed in the beginning of
the section. Two derivations are said to be equivalent if their derivation trees
are same. Thus, equivalent derivations are precisely differed by the order of
application of same production rules. That is, the application of production

smartworlD.asia
rules in a derivation can be permuted to get an equivalent derivation. For
example, as derivation trees (1) and (2) of Figure 3.4 are same, the derivations
(1) and (2) of Figure 3.3 are equivalent. Thus, a derivation tree may represent
several equivalent derivations. However, for a given derivation tree, whose
yield is a terminal string, there is a unique special type of derivation, viz.
leftmost derivation (or rightmost derivation).

Definition 3.2.5. For A ∈ N and x ∈ Σ∗ , the derivation A ⇒ ∗ x is said

to be a leftmost derivation if the production rule applied at each step is on
the leftmost nonterminal symbol of the sentential form. In which case, the
derivation is denoted by A ⇒∗ x.
L

Similarly, for A ∈ N and x ∈ Σ∗ , the derivation A ⇒ ∗ x is said to be a

rightmost derivation if the production rule applied at each step is on the
rightmost nonterminal symbol of the sentential form. In which case, the
∗ x.
derivation is denoted by A ⇒R

Because of the similarity between leftmost derivation and rightmost deriva-

tion, the properties of the leftmost derivations can be imitated to get similar
properties for rightmost derivations. Now, we establish some properties of
leftmost derivations.

Theorem 3.2.6. For A ∈ N and x ∈ Σ∗ , A ∗⇒ x if and only if A ∗⇒

L
x.

28

Smartzworld.com 29 jntuworldupdates.org
Smartworld.asia Specworld.in

Proof. “Only if” part is straightforward, as every leftmost derivation is, any-
way, a derivation.
For “if” part, let
A = α0 ⇒ α1 ⇒ α2 ⇒ · · · ⇒ αn−1 ⇒ αn = x
be a derivation. If αi ⇒L αi+1 , for all 0 ≤ i < n, then we are through.
Otherwise, there is an i such that αi 6⇒L αi+1 . Let k be the least such that
αk 6⇒L αk+1 . Then, we have αi ⇒L αi+1 , for all i < k, i.e. we have leftmost
substitutions in the first k steps. We now demonstrate how to extend the
leftmost substitution to (k + 1)th step. That is, we show how to convert the
derivation
A = α0 ⇒L α1 ⇒L · · · ⇒L αk ⇒ αk+1 ⇒ · · · ⇒ αn−1 ⇒ αn = x
in to a derivation
0 0 0
A = α0 ⇒L α1 ⇒L · · · ⇒L αk ⇒L αk+1 ⇒ αk+2 ⇒ · · · ⇒ αn−1 ⇒ αn0 = x
in which there are leftmost substitutions in the first (k + 1) steps and the
derivation is of same length of the original. Hence, by induction one can
extend the given derivation to a leftmost derivation A ⇒∗ x.
L

smartworlD.asia
Since αk ⇒ αk+1 but αk 6⇒L αk+1 , we have
αk = yA1 β1 A2 β2 ,
for some y ∈ Σ∗ , A1 , A2 ∈ N and β1 , β2 ∈ V ∗ , and A2 → γ2 ∈ P such that
αk+1 = yA1 β1 γ2 β2 .
But, since the derivation eventually yields the terminal string x, at a later
step, say pth step (for p > k), A1 would have been substituted by some string
γ1 ∈ V ∗ using the rule A1 → γ1 ∈ P . Thus the original derivation looks as
follows.
A = α0 ⇒L α1 ⇒L · · · ⇒L αk = yA1 β1 A2 β2
⇒ αk+1 = yA1 β1 γ2 β2
= yA1 ξ1 , with ξ1 = β1 γ2 β2
⇒ αk+2 = yA1 ξ2 , ( here ξ1 ⇒ ξ2 )
..
.
⇒ αp−1 = yA1 ξp−k−1
⇒ αp = yγ1 ξp−k−1
⇒ αp+1
..
.
⇒ αn = x

29

Smartzworld.com 30 jntuworldupdates.org
Smartworld.asia Specworld.in

We now demonstrate a mechanism of using the rule A1 → γ1 in (k + 1)th

step so that we get a desired derivation. Set
0
αk+1 = yγ1 β1 A2 β2
0
αk+2 = yγ1 β1 γ2 β2 = yγ1 ξ1
0
αk+3 = yγ1 ξ2
..
.
0
αp−1 = yγ1 ξp−k−2
αp0 = yγ1 ξp−k−1 = αp
0
αp+1 = αp+1
..
.
αn0 = αn
Now we have the following derivation in which (k + 1)th step has leftmost
substitution, as desired.
A = α0 ⇒L α1 ⇒L · · · ⇒L αk = yA1 β1 A2 β2
0
⇒L αk+1 = yγ1 β1 A2 β2
0
⇒ αk+2 = yγ1 β1 γ2 β2 = yγ1 ξ1
0
⇒ αk+3 = yγ1 ξ2

smartworlD.asia ⇒
..
.
0
αp−1
0
= yγ1 ξp−k−2
⇒ αp = yγ1 ξp−k−1 = αp
0
⇒ αp+1 = αp+1
..
.
⇒ αn0 = αn = x
As stated earlier, we have the theorem by induction.
Proposition 3.2.7. Two equivalent leftmost derivations are identical.
Proof. Let T be the derivation tree of two leftmost derivations D1 and D2 .
Note that the production rules applied at each nonterminal symbol is pre-
cisely represented by its children in the derivation tree. Since the derivation
tree is same for D1 and D2 , the production rules applied in both the deriva-
tions are same. Moreover, as D1 and D2 are leftmost derivations, the order
of application of production rules are also same; that is, each production is
applied to the leftmost nonterminal symbol. Hence, the derivations D1 and
D2 are identical.
Now we are ready to establish the correspondence between derivation
trees and leftmost derivations.

30

Smartzworld.com 31 jntuworldupdates.org
Smartworld.asia Specworld.in

Theorem 3.2.8. Every derivation tree, whose yield is a terminal string,

represents a unique leftmost derivation.

Proof. For A ∈ N and x ∈ Σ∗ , suppose T is the derivation tree of a derivation

A⇒∗ x. By Theorem 3.2.6, we can find an equivalent leftmost derivation
∗
A ⇒L x. Hence, by Proposition 3.2.7, we have a unique leftmost derivation
that is represented by T .

Theorem 3.2.9. Let G = (N, Σ, P, S) be a CFG and

κ = max{|α| | A → α ∈ P }.

For x ∈ L(G), if T is a derivation tree for x, then

|x| ≤ κh

where h is the height of the tree T .

Proof. Since κ is the maximum length of righthand side of each production

rule of G, each internal node of T is a parent of at most κ number of children.

smartworlD.asia
Hence, T is a κ-ary tree. Now, in the similar lines of Theorem 1.2.3 that is
given for binary trees, one can easily prove that T has at most κh leaf nodes.
Hence, |x| ≤ κh .

3.2.1 Ambiguity
Let G be a context-free grammar. It may be a case that the CFG G gives
two or more inequivalent derivations for a string x ∈ L(G). This can be
identified by their different derivation trees. While deriving the string, if
there are multiple possibilities of application of production rules on the same
symbol, one may have a difficulty in choosing a correct rule. In the context
of compiler which is constructed based on a grammar, this difficulty will lead
to an ambiguity in parsing. Thus, a grammar with such a property is said
to be ambiguous.

Definition 3.2.10. Formally, a CFG G is said to be ambiguous, if G has

two different leftmost derivations for some string in L(G). Otherwise, the
grammar is said to be unambiguous.

Remark 3.2.11. One can equivalently say that a CFG G is ambiguous, if G has
two different rightmost derivations or derivation trees for a string in L(G).

31

Smartzworld.com 32 jntuworldupdates.org
Smartworld.asia Specworld.in

Example 3.2.12. Recall the CFG which generates arithmetic expressions

with the production rules

S → S ∗ S | S + S | (S) | a | b

As shown in Figure 3.3, the arithmetic expression a + b ∗ a has two different

leftmost derivations, viz. (2) and (3). Hence, the grammar is ambiguous.
Example 3.2.13. An unambiguous grammar which generates the arithmetic
expressions (the language generated by the CFG given in Example 3.2.12) is
as follows.

S → S+T | T
T → T ∗R | R
R → (S) | a | b

In the context of constructing a compiler or similar such applications, it

is desirable that the underlying grammar be unambiguous. Unfortunately,
there are some CFLs for which there is no CFG which is unambiguous. Such
a CFL is known as inherently ambiguous language.

smartworlD.asia
Example 3.2.14. The context-free language

{am bm cn dn | m, n ≥ 1} ∪ {am bn cn dm | m, n ≥ 1}

is inherently ambiguous. For proof, one may refer to the Hopcroft and Ullman
[1979].

3.3 Regular Grammars

Definition 3.3.1. A CFG G = (N, Σ, P, S) is said to be linear if every
production rule of G has at most one nonterminal symbol in its righthand
side. That is,

A → α ∈ P =⇒ α ∈ Σ∗ or α = xBy, for some x, y ∈ Σ∗ and B ∈ N.

Example 3.3.2. The CFGs given in Examples 3.1.8, 3.1.9 and 3.1.11 are
clearly linear. Whereas, the CFG given in Example 3.1.12 is not linear.
Remark 3.3.3. If G is a linear grammar, then every derivation in G is a
leftmost derivation as well as rightmost derivation. This is because there is
exactly one nonterminal symbol in the sentential form of each internal step
of the derivation.

32

Smartzworld.com 33 jntuworldupdates.org
Smartworld.asia Specworld.in

Definition 3.3.4. A linear grammar G = (N, Σ, P, S) is said to be right

linear if the nonterminal symbol in the righthand side of each production rule,
if any, occurs at the right end. That is, righthand side of each production
rule is of the form – a terminal string followed by at most one nonterminal
symbol – as shown below.

A → x or A → xB

for some x ∈ Σ∗ and B ∈ N .

Similarly, a left linear grammar is defined by considering the production
rules in the following form.

A → x or A → Bx

for some x ∈ Σ∗ and B ∈ N .

Because of the similarity in the definitions of left linear grammar and right
linear grammar, every result which is true for one can be imitated to obtain
a parallel result for the other. In fact, the notion of left linear grammar is
equivalent right linear grammar (see Exercise ??). Here, by equivalence we

smartworlD.asia
mean, if L is generated by a right linear grammar, then there exists a left
linear grammar that generates L; and vice-versa.

Example 3.3.5. The CFG given in Example 3.1.8 is a right linear grammar.
For the language {ak | k ≥ 0}, an equivalent left linear grammar is given
by the following production rules.

S → Sa | ε

Example 3.3.6. We give a right linear grammar for the language

L = {x ∈ {a, b}∗ | |x|a = 2n + 1, for some n}.

We understand that the nonterminal symbols in a grammar maintain the

properties of the strings that they generate or whenever they appear in a
derivation they determine the properties of the partial string that the deriva-
tion so far generated.
Since we are looking for a right linear grammar for L, each production
rule will have at most one nonterminal symbol on its righthand side and
hence at each internal step of a desired derivation, we will have exactly one
nonterminal symbol. While, we are generating a sequence of a0 s and b0 s for
the language, we need not keep track of the number of b0 s that are generated,
so far. Whereas, we need to keep track the number of a0 s; here, it need not be

33

Smartzworld.com 34 jntuworldupdates.org
Smartworld.asia Specworld.in

the actual number, rather the parity (even or odd) of the number of a0 s that
are generated so far. So to keep track of the parity information, we require
two nonterminal symbols, say O and E, respectively for odd and even. In the
beginning of any derivation, we would have generated zero number of symbols
– in general, even number of a0 s. Thus, it is expected that the nonterminal
symbol E to be the start symbol of the desired grammar. While E generates
the terminal symbol b, the derivation can continue to be with nonterminal
symbol E. Whereas, if one a is generated then we change to the symbol O,
as the number of a0 s generated so far will be odd from even. Similarly, we
switch to E on generating an a with the symbol O and continue to be with
O on generating any number of b0 s. Precisely, we have obtained the following
production rules.
E → bE | aO
O → bO | aE
Since, our criteria is to generate a string with odd number of a0 s, we can
terminate a derivation while continuing in O. That is, we introduce the
production rule
O→ε

smartworlD.asia
Hence, we have the right linear grammar G = ({E, O}, {a, b}, P, E), where P
has the above defined three productions. Now, one can easily observe that
L(G) = L.

Recall that the language under discussion is stated as a regular language

in Chapter 1 (refer Example 2.4.10). Whereas, we did not supply any proof
in support. Here, we could identify a right linear grammar that generates
the language. In fact, we have the following result.

Theorem 3.3.7. The language generated by a right linear grammar is reg-

ular. Moreover, for every regular language L, there exists a right linear
grammar that generates L.

An elegant proof of this theorem would require some more concepts and
hence postponed to later chapters. For proof of the theorem, one may refer
to Chapter 4. In view of the theorem, we have the following definition.

Definition 3.3.8. Right linear grammars are also called as regular gram-
mars.

Remark 3.3.9. Since left linear grammars and right linear grammars are
equivalent, left linear grammars also precisely generate regular languages.
Hence, left linear grammars are also called as regular grammars.

34

Smartzworld.com 35 jntuworldupdates.org
Smartworld.asia Specworld.in

Example 3.3.10. The language {x ∈ {a, b}∗ | ab is a substring of x} can be

generated by the following regular grammar.

S → aS | bS | abA

A → aA | bA | ε

Example 3.3.11. Consider the regular grammar with the following produc-
tion rules.
S → aA | bS | ε
A → aS | bA
Note that the grammar generates the set of all strings over {a, b} having
even number of a0 s.

Example 3.3.12. Consider the language represented by the regular expres-

sion a∗ b+ over {a, b}, i.e.

{am bn ∈ {a, b}∗ | m ≥ 0, n ≥ 1}

smartworlD.asia
It can be easily observe that the following regular grammar generates the
language.

S → aS | B
B → bB | b

Example 3.3.13. The grammar with the following production rules is clearly
regular.

S → bS | aA
A → bA | aB
B → bB | aS | ε

It can be observed that the language {x ∈ {a, b}∗ | |x|a ≡ 2 mod 3} is

generated by the grammar.

Example 3.3.14. Consider the language

n ¯ o
¯
L = x ∈ (0 + 1)∗ ¯ |x|0 is even ⇔ |x|1 is odd .

It is little tedious to construct a regular grammar to show that L is regular.

However, using a better tool, we show that L is regular later (See Example
5.1.3).

35

Smartzworld.com 36 jntuworldupdates.org
Smartworld.asia Specworld.in

3.4 Digraph Representation

We could achieve in writing a right linear grammar for some languages. How-
ever, we face difficulties in constructing a right linear grammar for some lan-
guages that are known to be regular. We are now going to represent a right
linear grammar by a digraph which shall give a better approach in writ-
ing/constructing right linear grammar for languages. In fact, this digraph
representation motivates one to think about the notion of finite automaton
which will be shown, in the next chapter, as an ultimate tool in understanding
regular languages.
Definition 3.4.1. Given a right linear grammar G = (N, T, P, S), define the
digraph (V, E), where the vertex set V = N ∪ {$} with a new symbol $ and
the edge set E is formed as follows.
1. (A, B) ∈ E ⇐⇒ A → xB ∈ P , for some x ∈ T ∗
2. (A, $) ∈ E ⇐⇒ A → x ∈ P , for some x ∈ T ∗
In which case, the arc from A to B is labeled by x.

smartworlD.asia
Example 3.4.2. The digraph for the grammar presented in Example 3.3.10
is as follows.
a,b a,b
ª ª
G@ABC
/ FED ab / G@ABC
FED ε / ?89:;
>=<
S A $
Remark 3.4.3. From a digraph one can easily give the corresponding right
linear grammar.
Example 3.4.4. The digraph for the grammar presented in Example 3.3.12
is as follows.
a b
ª ª
G@ABC
/ FED ε / G@ABC
FED b / ?89:;
>=<
S B $
Remark 3.4.5. A derivation in a right linear grammar can be represented, in
its digraph, by a path from the starting node S to the special node $; and
conversely. We illustrate this through the following.
Consider the following derivation for the string aab in the grammar given
in Example 3.3.12.
S ⇒ aS ⇒ aaS ⇒ aaB ⇒ aab
The derivation can be traced by a path from S to $ in the corresponding
digraph (refer to Example 3.4.4) as shown below.
a /S a /S ε /B b /$
S

36

Smartzworld.com 37 jntuworldupdates.org
Smartworld.asia Specworld.in

One may notice that the concatenation of the labels on the path, called label
of the path, gives the desired string. Conversely, it is easy to see that the
label of any path from S to $ can be derived in G.

smartworlD.asia

37

Smartzworld.com 38 jntuworldupdates.org
Smartworld.asia Specworld.in

Chapter 4

Finite Automata

Regular grammars, as language generating devices, are intended to generate

regular languages - the class of languages that are represented by regular
expressions. Finite automata, as language accepting devices, are important
tools to understand the regular languages better. Let us consider the reg-
ular language - the set of all strings over {a, b} having odd number of a0 s.
Recall the grammar for this language as given in Example 3.3.6. A digraph

smartworlD.asia
representation of the grammar is given below:
b b
ª ª
/ G@ABC
FED a / G@ABC
FED ε / ?89:;
>=<
E f O $
a

Let us traverse the digraph via a sequence of a0 s and b0 s starting at the node
E. We notice that, at a given point of time, if we are at the node E, then
so far we have encountered even number of a0 s. Whereas, if we are at the
node O, then so far we have traversed through odd number of a0 s. Of course,
being at the node $ has the same effect as that of node O, regarding number
of a0 s; rather once we reach to $, then we will not have any further move.
Thus, in a digraph that models a system which understands a language,
nodes holds some information about the traversal. As each node is holding
some information it can be considered as a state of the system and hence a
state can be considered as a memory creating unit. As we are interested in the
languages having finite representation, we restrict ourselves to those systems
with finite number of states only. In such a system we have transitions
between the states on symbols of the alphabet. Thus, we may call them as
finite state transition systems. As the transitions are predefined in a finite
state transition system, it automatically changes states based on the symbols
given as input. Thus a finite state transition system can also be called as a

38

Smartzworld.com 39 jntuworldupdates.org
Smartworld.asia Specworld.in

finite state automaton or simply a finite automaton – a device that works

automatically. The plural form of automaton is automata.
In this chapter, we introduce the notion of finite automata and show that
they model the class of regular languages. In fact, we observe that finite
automata, regular grammars and regular expressions are equivalent; each of
them are to represent regular languages.

4.1 Deterministic Finite Automata

Deterministic finite automaton is a type of finite automaton in which the
transitions are deterministic, in the sense that there will be exactly one tran-
sition from a state on an input symbol. Formally,
a deterministic finite automaton (DFA) is a quintuple A = (Q, Σ, δ, q0 , F ),
where
Q is a finite set called the set of states,
Σ is a finite set called the input alphabet,
q0 ∈ Q, called the initial/start state,
F ⊆ Q, called the set of final/accept states, and

smartworlD.asia
δ : Q × Σ −→ Q is a function called the transition function or next-state
function.
Note that, for every state and an input symbol, the transition function δ
assigns a unique next state.

Example 4.1.1. Let Q = {p, q, r}, Σ = {a, b}, F = {r} and δ is given by
the following table:
δ a b
p q p
q r p
r r r
Clearly, A = (Q, Σ, δ, p, F ) is a DFA.

We normally use symbols p, q, r, . . . with or without subscripts to denote

states of a DFA.

Transition Table
Instead of explicitly giving all the components of the quintuple of a DFA, we
may simply point out the initial state and the final states of the DFA in the
table of transition function, called transition table. For instance, we use an
arrow to point the initial state and we encircle all the final states. Thus, we
can have an alternative representation of a DFA, as all the components of

39

Smartzworld.com 40 jntuworldupdates.org
Smartworld.asia Specworld.in

the DFA now can be interpreted from this representation. For example, the
DFA in Example 4.1.1 can be denoted by the following transition table.

δ a b
→p q p
q r p
°
r r r

Transition Diagram
Normally, we associate some graphical representation to understand abstract
concepts better. In the present context also we have a digraph representa-
tion for a DFA, (Q, Σ, δ, q0 , F ), called a state transition diagram or simply a
transition diagram which can be constructed as follows:

1. Every state in Q is represented by a node.

2. If δ(p, a) = q, then there is an arc from p to q labeled a.

3. If there are multiple arcs from labeled a1 , . . . ak−1 , and ak , one state to

smartworlD.asia
another state, then we simply put only one arc labeled a1 , . . . , ak−1 , ak .

4. There is an arrow with no source into the initial state q0 .

5. Final states are indicated by double circle.

The transition diagram for the DFA given in Example 4.1.1 is as below:

b a, b
­ ®
a a
p
/ ?89:;
>=< q
/ ?89:;
>=< / ?89:;
>=<
/.-,
()*+
r
b
b

Note that there are two transitions from the state r to itself on symbols
a and b. As indicated in the point 3 above, these are indicated by a single
arc from r to r labeled a, b.

Extended Transition Function

Recall that the transition function δ assigns a state for each state and an input
symbol. This naturally can be extended to all strings in Σ∗ , i.e. assigning a
state for each state and an input string.

40

Smartzworld.com 41 jntuworldupdates.org
Smartworld.asia Specworld.in

The extended transition function δ̂ : Q × Σ∗ −→ Q is defined recursively

as follows: For all q ∈ Q, x ∈ Σ∗ and a ∈ Σ,
δ̂(q, ε) = q and
δ̂(q, xa) = δ(δ̂(q, x), a).
For example, in the DFA given in Example 4.1.1, δ̂(p, aba) is q because
δ̂(p, aba) = δ(δ̂(p, ab), a)
= δ(δ(δ̂(p, a), b), a)
= δ(δ(δ(δ̂(p, ε), a), b), a)
= δ(δ(δ(p, a), b), a)
= δ(δ(q, b), a)
= δ(p, a) = q
Given p ∈ Q and x = a1 a2 · · · ak ∈ Σ∗ , δ̂(p, x) can be evaluated easily
using the transition diagram by identifying the state that can be reached by
traversing from p via the sequence of arcs labeled a1 , a2 , . . . , ak .
For instance, the above case can easily be seen by the traversing through

smartworlD.asia
the path labeled aba from p to reach to q as shown below:

a b a
p
?>=<
89:; q
/ ?89:;
>=< p
/ ?89:;
>=< q
/ ?89:;
>=<

Language of a DFA
Now, we are in a position to define the notion of acceptance of a string, and
consequently acceptance of a language, by a DFA.
A string x ∈ Σ∗ is said to be accepted by a DFA A = (Q, Σ, δ, q0 , F ) if
δ̂(q0 , x) ∈ F . That is, when you apply the string x in the initial state the
DFA will reach to a final state.
The set of all strings accepted by the DFA A is said to be the language
accepted by A and is denoted by L(A ). That is,
L(A ) = {x ∈ Σ∗ | δ̂(q0 , x) ∈ F }.
Example 4.1.2. Consider the following DFA

q0 SSS a
/ G@ABC
FED q1 ?
/ G@ABC
FED b
q2
/ GFED
@ABC
?>=<
89:; b
q
/ GFED
kk 3
@ABC
?>=<
89:;
SSS ??? a Ä kkk
SSS a ÄÄ
Ä kk
SSS kk
SSS ??? Ä kkk
b SSS ?? ÄÄÄ kkkkk a,b
SSS Â ÄÄ kkk
)
?>=<
89:; ku
t
T
a,b

41

Smartzworld.com 42 jntuworldupdates.org
Smartworld.asia Specworld.in

The only way to reach from the initial state q0 to the final state q2 is through
the string ab and it is through abb to reach another final state q3 . Thus, the
language accepted by the DFA is

{ab, abb}.

Example 4.1.3. As shown below, let us recall the transition diagram of the
DFA given in Example 4.1.1.

b a, b
­ ®
a a
p
/ ?89:;
>=< q
/ ?89:;
>=< / ?89:;
>=<
/.-,
()*+
r
b
b

1. One may notice that if there is aa in the input, then the DFA leads us
from the initial state p to the final state r. Since r is also a trap state,
after reaching to r we continue to be at r on any subsequent input.

2. If the input does not contain aa, then we will be shuffling between p
and q but never reach the final state r.

smartworlD.asia
Hence, the language accepted by the DFA is the set of all strings over {a, b}
having aa as substring, i.e.

{xaay | x, y ∈ {a, b}∗ }.

Description of a DFA
Note that a DFA is an abstract (computing) device. The depiction given in
Figure 4.1 shall facilitate one to understand its behavior. As shown in the
figure, there are mainly three components namely input tape, reading head,
and finite control. It is assumed that a DFA has a left-justified infinite tape
to accommodate an input of any length. The input tape is divided into cells
such that each cell accommodate a single input symbol. The reading head is
connected to the input tape from finite control, which can read one symbol at
a time. The finite control has the states and the information of the transition
function along with a pointer that points to exactly one state.
At a given point of time, the DFA will be in some internal state, say p,
called the current state, pointed by the pointer and the reading head will be
reading a symbol, say a, from the input tape called the current symbol. If
δ(p, a) = q, then at the next point of time the DFA will change its internal
state from p to q (now the pointer will point to q) and the reading head will
move one cell to the right.

42

Smartzworld.com 43 jntuworldupdates.org
Smartworld.asia Specworld.in

Figure 4.1: Depiction of Finite Automaton

Initializing a DFA with an input string x ∈ Σ∗ we mean x be placed on

the input tape from the left most (first) cell of the tape with the reading head
placed on the first cell and by setting the initial state as the current state.
By the time the input is exhausted, if the current state of the DFA is a final

smartworlD.asia
state, then the input x is accepted by the DFA. Otherwise, x is rejected by
the DFA.

Configurations
A configuration or an instantaneous description of a DFA gives the informa-
tion about the current state and the portion of the input string that is right
from and on to the reading head, i.e. the portion yet to be read. Formally,
a configuration is an element of Q × Σ∗ .
Observe that for a given input string x the initial configuration is (q0 , x)
and a final configuration of a DFA is of the form (p, ε).
The notion of computation in a DFA A can be described through con-
figurations. For which, first we define one step relation as follows.

Definition 4.1.4. Let C = (p, x) and C 0 = (q, y) be two configurations. If

δ(p, a) = q and x = ay, then we say that the DFA A moves from C to C 0 in
one step and is denoted as C |−−A C 0 .

Clearly, |−−A is a binary relation on the set of configurations of A .

In a given context, if there is only one DFA under discussion, then we may
simply use |−− instead of |−−A . The reflexive transitive closure of |−− may
be denoted by −| *−. That is, C −
| *− C 0 if and only if there exist configurations

43

Smartzworld.com 44 jntuworldupdates.org
Smartworld.asia Specworld.in

C0 , C1 , . . . , Cn such that
C = C0 |−− C1 |−− C2 |−− · · · |−− Cn = C 0
Definition 4.1.5. A the computation of A on the input x is of the form
C |−*− C 0
where C = (q0 , x) and C 0 = (p, ε), for some p.
Remark 4.1.6. Given a DFA A = (Q, Σ, δ, q0 , F ), x ∈ L(A ) if and only if
(q0 , x) |−*− (p, ε) for some p ∈ F .
Example 4.1.7. Consider the following DFA
b a b a a, b
­ ­ ­ ­ ­
a b a b
q0
/ G@ABC
FED q1
/ G@ABC
FED q2
/ G@ABC
FED
?>=<
89:; q3
/ G@ABC
FED
?>=<
89:; q4
/ G@ABC
FED

As states of a DFA are memory creating units, we demonstrate the language

of the DFA under consideration by explaining the roles of each of its states.
1. It is clear that, if the input contains only b0 s, then the DFA remains in

smartworlD.asia
the initial state q0 only.
2. On the other hand, if the input has an a, then the DFA transits from
q0 to q1 . On any subsequent a0 s, it remains in q1 only. Thus, the role
of q1 in the DFA is to understand that the input has at least one a.
3. Further, the DFA goes from q1 to q2 via a b and remains at q2 on
subsequent b0 s. Thus, q2 recognizes that the input has an occurrence
of an ab.
Since q2 is a final state, the DFA accepts all those strings which have
one occurrence of ab.
4. Subsequently, if we have a number of a0 s, the DFA will reach to q3 ,
which is a final state, and remains there; so that all such strings will
also be accepted.
5. But, from then, via b the DFA goes to the trap state q4 and since it is
not a final state, all those strings will not be accepted. Here, note that
role of q4 is to remember the second occurrence of ab in the input.
Thus, the language accepted by the DFA is the set of all strings over {a, b}
which have exactly one occurrence of ab. That is,
n ¯ o
¯
x ∈ {a, b}∗ ¯ |x|ab = 1 .

44

Smartzworld.com 45 jntuworldupdates.org
Smartworld.asia Specworld.in

Example 4.1.8. Consider the following DFA

c c
­ ­
a,b
q0 A`
/ G@ABC
FED
?>=<
89:; q1
/ G@ABC
FED
AA }}
AA }}
AA }
a,b AA }} a,b
}~ }
q2
GFED
@ABC
T
c

1. First, observe that the number of occurrences of c0 s in the input at any

state is simply ignored by the state by remaining in the same state.

2. Whereas, on input a or b, every state will lead the DFA to its next
state as q0 to q1 , q1 to q2 and q2 to q0 .

3. It can be clearly visualized that, if the total number of a0 s and b0 s in

the input is a multiple of 3, then the DFA will be brought back to the
initial state q0 . Since q0 is the final state those strings will be accepted.

4. On the other hand, if any string violates the above stated condition,

smartworlD.asia
then the DFA will either be in q1 or be in q2 and hence they will not be
accepted. More precisely, if the total number of a0 s and b0 s in the input
leaves the remainder 1 or 2, when it is divided by 3, then the DFA will
be in the state q1 or q2 , respectively.

Hence the language of the DFA is

n ¯ o
¯
x ∈ {a, b, c}∗ ¯ |x|a + |x|b ≡ 0 mod 3 .

From the above discussion, further we ascertain the following:

1. Instead of q0 , if q1 is only the final state (as shown in (i), below), then
the language will be
n ¯ o
¯
x ∈ {a, b, c}∗ ¯ |x|a + |x|b ≡ 1 mod 3

2. Similarly, instead of q0 , if q2 is only the final state (as shown in (ii),

below), then the language will be
n ¯ o
∗ ¯
x ∈ {a, b, c} ¯ |x|a + |x|b ≡ 2 mod 3

45

Smartzworld.com 46 jntuworldupdates.org
Smartworld.asia Specworld.in

c c c c
­ ­ ­ ­
a,b a,b
q0 A`
/ G@ABC
FED q1
/ G@ABC
FED
?>=<
89:; q0 A`
/ G@ABC
FED q1
/ G@ABC
FED
AA }} AA }}
AA }} AA }}
AA } AA }
a,b AA }} a,b a,b AA }} a,b
}~ } }~ }
q2
GFED
@ABC q2
GFED
@ABC
?>=<
89:;
T T
c c

(i) (ii)
3. In the similar lines, one may observe that the language
n ¯ o
∗ ¯
x ∈ {a, b, c} ¯ |x|a + |x|b 6≡ 1 mod 3
n ¯ o
¯
= x ∈ {a, b, c}∗ ¯ |x|a + |x|b ≡ 0 or 2 mod 3
can be accepted by the DFA shown below, by making both q0 and q2
final states.
c c
­ ­
a,b
q0 A`
/ G@ABC
FED
?>=<
89:; q1
/ G@ABC
FED

smartworlD.asia
AA }}
AA }}
AA }
a,b AA }} a,b
}~ }
q2
GFED
@ABC
?>=<
89:;
T
c

4. Likewise, other combinations of the states, viz. q0 and q1 , or q1 and q2 ,

can be made final states and the languages be observed appropriately.
Example 4.1.9. Consider the language L over {a, b} which contains the
strings whose lengths are from the arithmetic progression
P = {2, 5, 8, 11, . . .} = {2 + 3n | n ≥ 0}.
That is, n ¯ o
¯ ∗
L = x ∈ {a, b} ¯ |x| ∈ P .
We construct a DFA accepting L. Here, we need to consider states whose
role is to count the number of symbols of the input.
1. First, we need to recognize the length 2. That is, first two symbols; for
which we may consider the following state transitions.
a,b a,b
q0
/ G@ABC
FED q1
/ G@ABC
FED q2
/ G@ABC
FED

Clearly, on any input over {a, b}, we are in the state q2 means that, so
far, we have read the portion of length 2.

46

Smartzworld.com 47 jntuworldupdates.org
Smartworld.asia Specworld.in

2. Then, the length of rest of the string need to be a multiple of 3. Here,

we borrow the idea from the Example 4.1.8 and consider the following
depicted state transitions, further.

q3
GFED
@ABC
}}>
a,b }}
}
}}
a,b a,b }}
q0
/ G@ABC
FED q1
/ G@ABC
FED q2 A`
/ G@ABC
FED
?>=<
89:; a,b
AA
AA
A
a,b AAA ²
q4
@ABC
GFED

Clearly, this DFA accepts L.

In general, for any arithmetic progression P = {k 0 + kn | n ≥ 0} with
k, k 0 ∈ N, if we consider the language
n ¯ o
¯
L = x ∈ Σ∗ ¯ |x| ∈ P

over an alphabet Σ, then the following DFA can be suggested for L.

smartworlD.asia Σ
0 ONML
qk0 +1
HIJK \ U
K
>
3
,
q0
/ G@ABC
FED Σ
q1
/ G@ABC
FED Σ ./ . . _ _ _ Σ
qk 0
/ GFED
@ABC
?>=<
89:; '»
S
¶
¯
¢
q_^]\ t
Σ XYZ[
k0 +k−1
pc j

Example 4.1.10. Consider the language over {a, b} consisting of those

strings that end with a. That is,

{xa | x ∈ {a, b}∗ }.

We observe the following to construct a DFA for the language.

1. If we assume a state q0 as the initial state, then an occurrence of a in
the input need to be distinguished. This can be done by changing the
state to some other, say q1 . Whereas, we continue to be in q0 on b0 s.

2. On subsequent a0 s at q1 let us remain at q1 . If a string ends on reaching

q1 , clearly it is ending with an a. By making q1 a final state, all such
strings can be accepted.

47

Smartzworld.com 48 jntuworldupdates.org
Smartworld.asia Specworld.in

3. If we encounter a b at q1 , then we should go out of q1 , as it is a final

state.

4. Since again the possibilities of further occurrences of a0 s need to cross-

checked and q0 is already taking care of that, we may assign the tran-
sition out of q1 on b to q0 .

Thus, we have the following DFA which accepts the given language.
b a
· ­
a
q0
/ G@ABC
FED q1
/ G@ABC
FED
?>=<
89:;
d
b

Example 4.1.11. Let us consider the following DFA

a

.-,w
//()*+ b //()*+
.-,e a /.-,
/()*+
ÂÁÀ¿
»¼½¾
Q I
a
b

smartworlD.asia
b

² a
/.-,
()*+
ÂÁÀ¿
»¼½¾
Q
b

Unlike the above examples, it is little tricky to ascertain the language ac-
cepted by the DFA. By spending some amount of time, one may possibly
report that the language is the set of all strings over {a, b} with last but one
symbol as b.
But for this language, if we consider the following type of finite automa-
ton one can easily be convinced (with an appropriate notion of language
acceptance) that the language is so.

a, b
­
b a, b
p
/ ?89:;
>=< q
/ ?89:;
>=< / ?89:;
>=<
/.-,
()*+
r

Note that, in this type of finite automaton we are considering multiple (pos-
sibly zero) number of transitions for an input symbol in a state. Thus, if
a string is given as input, then one may observe that there can be multiple
next states for the string. For example, in the above finite automaton, if
abbaba is given as input then the following two traces can be identified.
a b b a b a
p
/ ?89:;
>=< p
/ ?89:;
>=< p
/ ?89:;
>=< p
/ ?89:;
>=< p
/ ?89:;
>=< p
/ ?89:;
>=< p
/ ?89:;
>=<

48

Smartzworld.com 49 jntuworldupdates.org
Smartworld.asia Specworld.in

a b b a b a
p
/ ?89:;
>=< p
/ ?89:;
>=< p
/ ?89:;
>=< p
/ ?89:;
>=< p
/ ?89:;
>=< q
/ ?89:;
>=< / ?89:;
>=<
r
Clearly, p and r are the next states after processing the string abbaba. Since
it is reaching to a final state, viz. r, in one of the possibilities, we may
say that the string is accepted. So, by considering this notion of language
acceptance, the language accepted by the finite automaton can be quickly
reported as the set of all strings with the last but one symbol as b, i.e.
n ¯ o
¯ ∗
xb(a + b) ¯ x ∈ (a + b) .
Thus, the corresponding regular expression is (a + b)∗ b(a + b).
This type of automaton with some additional features is known as non-
deterministic finite automaton. This concept will formally be introduced in
the following section.

4.2 Nondeterministic Finite Automata

In contrast to a DFA, where we have a unique next state for a transition from

smartworlD.asia
a state on an input symbol, now we consider a finite automaton with nonde-
terministic transitions. A transition is nondeterministic if there are several
(possibly zero) next states from a state on an input symbol or without any in-
put. A transition without input is called as ε-transition. A nondeterministic
finite automaton is defined in the similar lines of a DFA in which transitions
may be nondeterministic.
Formally, a nondeterministic finite automaton (NFA) is a quintuple N =
(Q, Σ, δ, q0 , F ), where Q, Σ, q0 and F are as in a DFA; whereas, the transition
function δ is as below:
δ : Q × (Σ ∪ {ε}) −→ ℘(Q)
is a function so that, for a given state and an input symbol (possibly ε), δ
assigns a set of next states, possibly empty set.
Remark 4.2.1. Clearly, every DFA can be treated as an NFA.
Example 4.2.2. Let Q = {q0 , q1 , q2 , q3 , q4 }, Σ = {a, b}, F = {q1 , q3 } and δ
be given by the following transition table.
δ a b ε
q0 {q1 } ∅ {q4 }
q1 ∅ {q1 } {q2 }
q2 {q2 , q3 } {q3 } ∅
q3 ∅ ∅ ∅
q4 {q4 } {q3 } ∅

49

Smartzworld.com 50 jntuworldupdates.org
Smartworld.asia Specworld.in

The quintuple N = (Q, Σ, δ, q0 , F ) is an NFA. In the similar lines of a DFA,

an NFA can be represented by a state transition diagram. For instance, the
present NFA can be represented as follows:
b a
­ ­
a ε a,b
q0 L
/ G@ABC
FED q1
/ G@ABC
FED
?>=<
89:; q2
/ G@ABC
FED
hh4 q3
/ G@ABC
FED
?>=<
89:;
LLL
LLL h hhhhhh
LLL hhhh
LLL h hhhhhh
h
ε LL hhhh b
LLL
h hhhhhh
LLL hh
& hhhh
q4 hh
GFED
@ABC
T
a

Note the following few nondeterministic transitions in this NFA.

1. There is no transition from q0 on input symbol b.
2. There are multiple (two) transitions from q2 on input symbol a.

3. There is a transition from q0 to q4 without any input, i.e. ε-transition.

smartworlD.asia
Consider the traces for the string ab from the state q0 . Clearly, the following
four are the possible traces.
a b
q0
(i) GFED
@ABC q1
/ G@ABC
FED q1
/ G@ABC
FED

ε a b
q0
(ii) GFED
@ABC q4
/ G@ABC
FED q4
/ G@ABC
FED q3
/ G@ABC
FED

a ε b
q0
(iii) @ABC
GFED q1
/ G@ABC
FED q2
/ G@ABC
FED q3
/ G@ABC
FED

a b ε
q0
(iv) GFED
@ABC q1
/ G@ABC
FED q1
/ G@ABC
FED q2
/ G@ABC
FED

Note that three distinct states, viz. q1 , q2 and q3 are reachable from q0 via
the string ab. That means, while tracing a path from q0 for ab we consider
possible insertion of ε in ab, wherever ε-transitions are defined. For example,
in trace (ii) we have included an ε-transition from q0 to q4 , considering ab as
εab, as it is defined. Whereas, in trace (iii) we consider ab as aεb. It is clear
that, if we process the input string ab at the state q0 , then the set of next
states is {q1 , q2 , q3 }.
Definition 4.2.3. Let N = (Q, Σ, δ, q0 , F ) be an NFA. Given an input string
x = a1 a2 · · · ak and a state p of N , the set of next states δ̂(p, x) can be easily
computed using a tree structure, called a computation tree of δ̂(p, x), which
is defined in the following way:

50

Smartzworld.com 51 jntuworldupdates.org
Smartworld.asia Specworld.in

1. p is the root node

2. Children of the root are precisely those nodes which are having transi-
tions from p via ε or a1 .

3. For any node, whose branch (from the root) is labeled a1 a2 · · · ai (as
a resultant string by possible insertions of ε), its children are precisely
those nodes having transitions via ε or ai+1 .

4. If there is a final state whose branch from the root is labeled x (as a
resultant string), then mark the node by a tick mark X.

5. If the label of the branch of a leaf node is not the full string x, i.e.
some proper prefix of x, then it is marked by a cross X – indicating
that the branch has reached to a dead-end before completely processing
the string x.

Example 4.2.4. The computation tree of δ̂(q0 , ab) in the NFA given in
Example 4.2.2 is shown in Figure 4.2

smartworlD.asia
q
0
a ε

q1 q4

b ε
a

q1 q2 q4
✓
ε b b

q q q3
2 3 ✓ ✓

Figure 4.2: Computation Tree of δ̂(q0 , ab)

Example 4.2.5. The computation tree of δ̂(q0 , abb) in the NFA given in
Example 4.2.2 is shown in Figure 4.3. In which, notice that the branch
q0 − q4 − q4 − q3 has the label ab, as a resultant string, and as there are no
further transitions at q3 , the branch has got terminated without completely
processing the string abb. Thus, it is indicated by marking a cross X at the
end of the branch.

51

Smartzworld.com 52 jntuworldupdates.org
Smartworld.asia Specworld.in

q
0
a ε

q1 q4

b ε
a

q1 q2 q4

b ε b
b

q1 q2 q q3
3
✓ ✕ ✕
ε b

q q
2 3 ✓

Figure 4.3: Computation Tree of δ̂(q0 , abb)

smartworlD.asia
As the automaton given in Example 4.2.2 is nondeterministic, if a string
is processed at a state, then there may be multiple next states (unlike DFA),
possibly empty. For example, if we apply the string bba at the state q0 , then
the only possible way to process the first b is going via ε-transition from q0 to
q4 and then from q4 to q3 via b. As there are no transitions from q3 , the string
cannot be processed further. Hence, the set of next states for the string bba
at q0 is empty.
Thus, given a string x = a1 a2 · · · an and a state p, by treating x as
εa1 εa2 ε · · · εan ε and by looking at the possible complete branches starting at
p, we find the set of next states for p via x. To introduce the notion of δ̂ in
an NFA, we first introduce the notion called ε-closure of a state.

Definition 4.2.6. An ε-closure of a state p, denoted by E(p) is defined as

the set of all states that are reachable from p via zero or more ε-transitions.

Example 4.2.7. In the following we enlist the ε-closures of all the states of
the NFA given in Example 4.2.2.
E(q0 ) = {q0 , q4 }; E(q1 ) = {q1 , q2 }; E(q2 ) = {q2 }; E(q3 ) = {q3 }; E(q4 ) = {q4 }.

Further, for a set A of states, ε-closure of A, denoted by E(A) is defined

as [
E(A) = E(p).
p∈A

52

Smartzworld.com 53 jntuworldupdates.org
Smartworld.asia Specworld.in

Now we are ready to formally define the set of next states for a state via
a string.

Definition 4.2.8. Define the function

δ̂ : Q × Σ∗ −→ ℘(Q)

by

1. δ̂(q, ε) = E(q) and

³ [ ´
2. δ̂(q, xa) = E δ(p, a)
p∈δ̂(q,x)

Definition 4.2.9. A string x ∈ Σ∗ is said to be accepted by an NFA N =

(Q, Σ, δ, q0 , F ), if
δ̂(q0 , x) ∩ F 6= ∅.
That is, in the computation tree of (q0 , x) there should be final state among
the nodes marked with X. Thus, the language accepted by N is
n ¯ o

smartworlD.asia
¯
L(N ) = x ∈ Σ∗ ¯ δ̂(q0 , x) ∩ F 6= ∅ .

Example 4.2.10. Note that q1 and q3 are the final states for the NFA given
in Example 4.2.2.

1. The possible ways of reaching q1 from the initial state q0 is via the set
of strings represented by the regular expression ab∗ .

2. The are two possible ways of reaching q3 from q0 as discussed below.

(a) Via the state q1 : With the strings of ab∗ we can clearly reach q1
from q0 , then using the ε-transition we can reach q2 . Then the
strings of a∗ (a + b) will lead us from state q2 to q3 . Thus, the set
of string in this case can be represented by ab∗ a∗ (a + b).
(b) Via the state q4 : Initially, we use ε-transition to reach q4 from
q0 , then clearly the strings of a∗ b will precisely be useful to reach
from q4 to q3 . And hence ab∗ itself represent the set of strings in
this case.

Thus, the language accepted by the NFA can be represented by the

following regular expression:

ab∗ + a∗ b + ab∗ a∗ (a + b)

53

Smartzworld.com 54 jntuworldupdates.org
Smartworld.asia Specworld.in

Example 4.2.11. Consider the following NFA.

a b
­ ­
a ε
q0
/ G@ABC
FED
f
q1
/ G@ABC
FED q2
/ G@ABC
FED
?>=<
89:;

1. From the initial state q0 one can reach back to q0 via strings from a∗ or
from aεb∗ b, i.e. ab+ , or via a string which is a mixture of strings from
the above two sets. That is, the strings of (a + ab+ )∗ will lead us from
q0 to q0 .

2. Also, note that the strings of ab∗ will lead us from the initial state q0
to the final state q2 .

3. Thus, any string accepted by the NFA can be of the form – a string
from the set (a + ab+ )∗ followed by a string from the set ab∗ .

Hence, the language accepted by the NFA can be represented by

smartworlD.asia
(a + ab+ )∗ ab∗

4.3 Equivalence of NFA and DFA

Two finite automata A and A 0 are said to be equivalent if they accept the
same language, i.e. L(A ) = L(A 0 ). In the present context, although NFA
appears more general with a lot of flexibility, we prove that NFA and DFA
accept the same class of languages.
Since every DFA can be treated as an NFA, one side is obvious. The
converse, given an NFA there exists an equivalent DFA, is being proved
through the following two lemmas.

Lemma 4.3.1. Given an NFA in which there are some ε-transitions, there
exists an equivalent NFA without ε-transitions.

Proof. Let N = (Q, Σ, δ, q0 , F ) be a finite automaton in which there are

some ε-transitions. Set N 0 = (Q, Σ, δ 0 , q0 , F 0 ), where δ 0 : Q × Σ −→ ℘(Q) is
given by
δ 0 (q, a) = δ̂(q, a) for all a ∈ Σ, q ∈ Q
and F 0 = F ∪ {q ∈ Q | E(q) ∩ F 6= ∅}. It is clear that N 0
is a finite
automaton without ε-transitions.

54

Smartzworld.com 55 jntuworldupdates.org
Smartworld.asia Specworld.in

We claim that L(N ) = L(N 0 ). First note that

ε ∈ L(N ) ⇐⇒ δ̂(q0 , ε) ∩ F 6= ∅
⇐⇒ E(q0 ) ∩ F 6= ∅
⇐⇒ q0 ∈ F 0
⇐⇒ ε ∈ L(N 0 ).
Now, for any x ∈ Σ+ , we prove that δ̂ 0 (q0 , x) = δ̂(q0 , x) so that L(N ) =
L(N 0 ). This is because, if δ̂ 0 (q0 , x) = δ̂(q0 , x), then
x ∈ L(N ) ⇐⇒ δ̂(q0 , x) ∩ F 6= ∅
⇐⇒ δ̂ 0 (q0 , x) ∩ F 6= ∅
=⇒ δ̂ 0 (q0 , x) ∩ F 0 6= ∅
⇐⇒ x ∈ L(N 0 ).
Thus, L(N ) ⊆ L(N 0 ).
Conversely, let x ∈ L(N 0 ), i.e. δ̂ 0 (q0 , x)∩F 0 6= ∅. If δ̂ 0 (q0 , x)∩F 6= ∅ then
we are through. Otherwise, there exists q ∈ δ̂ 0 (q0 , x) such that E(q) ∩ F 6= ∅.
This implies that q ∈ δ̂(q0 , x) and E(q) ∩ F 6= ∅. Which in turn implies that

smartworlD.asia
δ̂(q0 , x) ∩ F 6= ∅. That is, x ∈ L(N ). Hence L(N ) = L(N 0 ).
So, it is enough to prove that δ̂ 0 (q0 , x) = δ̂(q0 , x) for each x ∈ Σ+ . We
prove this by induction on |x|. If x ∈ Σ, then by definition of δ 0 we have
δ̂ 0 (q0 , x) = δ 0 (q0 , x) = δ̂(q0 , x).
Assume the result for all strings of length less than or equal to m. For a ∈ Σ,
let |xa| = m + 1. Since |x| = m, by inductive hypothesis,
δ̂ 0 (q0 , x) = δ̂(q0 , x).
Now
δ̂ 0 (q0 , xa) = δ 0 (δ̂ 0 (q0 , x), a)
[
= δ 0 (p, a)
p∈δ̂ 0 (q0 ,x)
[
= δ̂(p, a)
p∈δ̂(q0 ,x)

= δ̂(q0 , xa).
Hence by induction we have δ 0 (q0 , x) = δ̂(q0 , x) ∀x ∈ Σ+ . This completes
the proof.

55

Smartzworld.com 56 jntuworldupdates.org
Smartworld.asia Specworld.in

Lemma 4.3.2. For every NFA N 0 without ε-transitions, there exists a DFA
A such that L(N 0 ) = L(A ).
Proof. Let N 0 = (Q, Σ, δ 0 , q0 , F 0 ) be an NFA without ε-transitions. Con-
struct A = (P, Σ, µ, p0 , E), where
n o
P = p{i1 ,...,ik } | {qi1 , . . . , qik } ⊆ Q ,

p0 = p{0} ,
n o
E = p{i1 ,...,ik } ∈ P | {qi1 , . . . , qik } ∩ F 0 6= ∅ , and

µ : P × Σ −→ P defined by
[
µ(p{i1 ,...,ik } , a) = p{j1 ,...,jm } if and only if δ 0 (qil , a) = {qj1 , . . . , qjm }.
il ∈{i1 ,...,ik }

Clearly, A is a DFA. To show L(N 0 ) = L(A ), we prove that, for all x ∈ Σ∗ ,

µ̂(p0 , x) = p{i1 ,...,ik } if and only if δ̂ 0 (q0 , x) = {qi1 , . . . , qik }. (?)

smartworlD.asia
This suffices the result, because

x ∈ L(N 0 ) ⇐⇒ δ̂ 0 (q0 , x) ∩ F 0 6= ∅
⇐⇒ µ̂(p0 , x) ∈ E
⇐⇒ x ∈ L(A ).

Now, we prove the statement in (?) by induction on the length of x.

If |x| = 0 then δ̂ 0 (q0 , x) = {q0 }, also µ̂(p0 , x) = p0 = p{0} so that the
statement in (?) holds in this case. (In case x = 1 also, one may notice that
the statement directly follows from the definition of µ.)
Assume that the statement is true for the strings whose length is less
than or equal to n. Let x ∈ Σ∗ with |x| = n and a ∈ Σ. Then

µ̂(p0 , xa) = µ(µ̂(p0 , x), a).

By inductive hypothesis,

µ̂0 (p0 , x) = p{i1 ,...,ik } if and only if δ̂ 0 (q0 , x) = {qi1 , . . . , qik }.

Now by definition of µ,
[
µ(p{i1 ,...,ik } , a) = p{j1 ,...,jm } if and only if δ 0 (qil , a) = {qj1 , . . . , qjm }.
il ∈{i1 ,...,ik }

56

Smartzworld.com 57 jntuworldupdates.org
Smartworld.asia Specworld.in

Thus,

µ̂(p0 , xa) = p{j1 ,...,jm } if and only if δ̂ 0 (q0 , xa) = {qj1 , . . . , qjm }.

Hence, by induction, the statement in (?) is true.

Thus, we have the following theorem.

Theorem 4.3.3. For every NFA there exists an equivalent DFA.

We illustrate the constructions for Lemma 4.3.1 and Lemma 4.3.2 through
the following example.

Example 4.3.4. Consider the following NFA

b a, b
­ ­
ε, b
q0 @
/ G@ABC
FED q1
/ G@ABC
FED
@@ ~~
@@ ~~
@@ ~
@ ~~
ε @@@ ~~~ b
@@ ~
@Ã ~~~~

smartworlD.asia
q2
GFED
@ABC
?>=<
89:;

That is, the transition function, say δ, can be displayed as the following table

δ a b ε
q0 ∅ {q0 , q1 } {q1 , q2 }
q1 {q1 } {q1 , q2 } ∅
q2 ∅ ∅ ∅

In order to remove ε-transitions and obtain an equivalent finite automaton,

we need to calculate δ̂(q, a), for all states q and for all input symbols a. That
is given in the following table.

δ 0 = δ̂ a b
q0 {q1 } {q0 , q1 , q2 }
q1 {q1 } {q1 , q2 }
q2 ∅ ∅

Thus, (Q = {q0 , q1 , q2 }, {a, b}, δ 0 , q0 , {q0 , q2 }) is an equivalent NFA in which

there are no ε-transitions. Now, to convert this to an equivalent DFA, we
consider each subset of Q as a state in the DFA and its transition function
µ assigns the union of respective transitions at each state of the above NFA

57

Smartzworld.com 58 jntuworldupdates.org
Smartworld.asia Specworld.in

which are in the subset under consideration. That is, for any input symbol
a and a subset X of {0, 1, 2} (the index set of the states)
[
µ(pX , a) = δ 0 (qi , a).
i∈X

Thus the resultant DFA is

(P, {a, b}, µ, p{0} , E)

where n ¯ o
¯
P = pX ¯ X ⊆ {0, 1, 2} ,
n ¯ o
¯
E = pX ¯ X ∩ {0, 2} 6= ∅ ; in fact there are six states in E, and
the transition map µ is given in the following table:

µ a b
p∅ p∅ p∅
p{0} p{1} p{0,1,2}
p{1} p{1} p{1,2}

smartworlD.asia
p{2} p∅ p∅
p{0,1} p{1} p{0,1,2}
p{1,2} p{1} p{1,2}
p{0,2} p{1} p{0,1,2}
p{0,1,2} p{1} p{0,1,2}

It is observed that while converting an NFA N to an equivalent DFA A ,

the number of states of A has an exponential growth compared to that of
N . In fact, if N has n states, A will have 2n states. It may also be noted
that, among these 2n states, some of the states are dead – the states which
are either inaccessible from the initial state or no final state is accessible from
them. If there is a procedure to remove some of the dead states which do
not alter the language, then we may get a DFA with lesser number of states.
In the following we provide certain heuristics to convert a given NFA with
n states to an equivalent DFA with number of states less than 2n .

4.3.1 Heuristics to Convert NFA to DFA

We demonstrate certain heuristics to convert an NFA without ε-transitions
to its equivalent DFA. Nondeterminism in a finite automaton without ε-
transitions is clearly because of multiple transitions at a state for an input
symbol. That is, for a state q and an input symbol a, if δ(q, a) = P with
|P | = 0 or |P | > 1, then such a situation need to be handled to avoid

58

Smartzworld.com 59 jntuworldupdates.org
Smartworld.asia Specworld.in

nondeterminism at q. To do this, we propose the following techniques and

illustrate them through examples.
Case 1: |P | = 0. In this case, there is no transition from q on a. We
create a new (trap) state t and give transition from q to t via a. For all
input symbols, the transitions from t will be assigned to itself. For all such
nondeterministic transitions in the finite automaton, creating only one new
trap state will be sufficient.
Case 2: |P | > 1. Let P = {p1 , p2 , . . . , pk }.
If q ∈
/ P , then choose a new state p and assign a transition from q to
p via a. Now all the transitions from p1 , p2 , . . . , pk are assigned to p. This
avoids nondeterminism at q; but, there may be an occurrence of nondeter-
minism on the new state p. One may successively apply this heuristic to
avoid nondeterminism further.
If q ∈ P , then the heuristic mentioned above can be applied for all the
transitions except for the one reaching to q, i.e. first remove q from P and
work as mentioned above. When there is no other loop at q, the resultant
scenario with the transition from q to q is depicted as under:
a
­

smartworlD.asia
a
q
?>=<
89:; p
/ ?>=<
89:;

This may be replaced by the following type of transition:

a
­
a
q
?>=<
89:; p
/ ?89:;
>=<

In case there is another loop at q, say with the input symbol b, then scenario
will be as under:
a,b
­
a
q
?>=<
89:; p
/ ?>=<
89:;

And in which case, to avoid the nondeterminism at q, we suggest the equiv-

alent transitions as shown below:
b a
· ­
a
q
?>=<
89:; p
/ ?89:;
>=<
b
b

We illustrate these heuristics in the following examples.

Example 4.3.5. For the language over {a, b} which contains all those strings
starting with ba, it is easy to construct an NFA as given below.
a,b
®
b a
p
/ ?89:;
>=< q
/ ?89:;
>=< / ?89:;
>=<
/.-,
()*+
r

59

Smartzworld.com 60 jntuworldupdates.org
Smartworld.asia Specworld.in

Clearly, the language accepted by the above NFA is {bax | x ∈ a, b∗ }. Now,

by applying the heuristics one may propose the following DFA for the same
language.
a,b
®
b a
p=
/ ?89:;
>=< q
/ ?89:;
>=< / ?89:;
>=<
/.-,
()*+
r
== ¢¢
== ¢¢
=
a == ¢¢b
=Á ¡¢¢¢
?>=<
89:;
t
T
a,b

Example 4.3.6. One can construct the following NFA for the language
{an xbm | x = baa, n ≥ 1 and m ≥ 0}.
a b
´ a b a a °
//()*+
.-, //()*+
.-, /()*+
/ .-, //()*+
.-, //()*+
.-,
ÂÁÀ¿
»¼½¾

By applying the heuristics, the following DFA can be constructed easily.

a b
a ´ b a ° a
/()*+
/ .-,UUU /()*+
/ .-, //()*+
.-, /.-,
/()*+ ii//()*+
.-,
ÂÁÀ¿
»¼½¾
UUUU rr r ii
UUUU iiii

smartworlD.asia
UUUU b rrr iiii
UUUUU b rr i iii
UUU* ² ry tiriririiii
b a
/.-,
()*+
Q
a,b

Example 4.3.7. We consider the following NFA, which accepts the language
{x ∈ {a, b}∗ | bb is a substring of x}.
a,b a,b
· ®
b b
p
/ ?89:;
>=< q
/ ?89:;
>=< / ?89:;
>=<
/.-,
()*+
r
By applying a heuristic discussed in Case 2, as shown in the following finite
automaton, we can remove nondeterminism at p, whereas we get nondeter-
minism at q, as a result.
a b a,b
· ­ ®
b b
p
/ ?89:;
>=< q
/ ?89:;
>=< / ?89:;
>=<
/.-,
()*+
r
b
a

To eliminate the nondeterminism at q, we further apply the same technique

and get the following DFA.
a a,b
· ®
b b
p
/ ?89:;
>=< q
/ ?89:;
>=< / ?89:;
>=<
/.-,
()*+
r
b
a

60

Smartzworld.com 61 jntuworldupdates.org
Smartworld.asia Specworld.in

Example 4.3.8. Consider the language L over {a, b} containing those strings
x with the property that either x starts with aa and ends with b, or x starts
with ab and ends with a. That is,

L = {aaxb | x ∈ {a, b}∗ } ∪ {abya | y ∈ {a, b}∗ }.

An NFA shown below can easily be constructed for the language L.

a,b
a ° b
/.-,
()*+
9 //()*+
.-, //()*+
.-,
ÂÁÀ¿
»¼½¾
a rrr
rrr
rr
rrrr
//()*+
.-,L
LLL
LLL
L
a LLL
L%
/.-,
()*+ /()*+
/ .-, //()*+
.-,
ÂÁÀ¿
»¼½¾
b Q a
a,b

By applying the heuristics on this NFA, the following DFA can be obtained,
which accepts L.

smartworlD.asia a rrr
rrr9 _
°
/.-,
()*+
a
b °
//()*+
.-,
ÂÁÀ¿
»¼½¾
b

rrr a
rr
//()*+
.-, a / rL
/.-,
()*+
LLL
LLL b
b L
² b LLLL Ä
/.-,
()*+ %
/.-,
()*+ //()*+
.-,
ÂÁÀ¿
»¼½¾
Q Q a Q
a,b b a

4.4 Minimization of DFA

In this section we provide an algorithmic procedure to convert a given DFA
to an equivalent DFA with minimum number of states. In fact, this asser-
tion is obtained through well-known Myhill-Nerode theorem which gives a
characterization of the languages accepted by DFA.

4.4.1 Myhill-Nerode Theorem

We start with the following basic concepts.

61

Smartzworld.com 62 jntuworldupdates.org
Smartworld.asia Specworld.in

Definition 4.4.1. An equivalence relation ∼ on Σ∗ is said to be right in-

variant if, for x, y ∈ Σ∗ ,

x ∼ y =⇒ ∀z(xz ∼ yz).

Example 4.4.2. Let L be a language over Σ. Define the relation ∼L on Σ∗

by
x ∼L y if and only if ∀z(xz ∈ L ⇐⇒ yz ∈ L).
It is straightforward to observe that ∼L is an equivalence relation on Σ∗ .
For x, y ∈ Σ∗ assume x ∼L y. Let ³ z ∈ Σ∗ be arbitrary. Claim:
´ xz ∼L yz.
That is, we have to show that (∀w) xzw ∈ L ⇐⇒ yzw ∈ L . For w ∈ Σ∗ ,
write u = zw. Now, since x ∼L y, we have xu ∈ L ⇐⇒ yu ∈ L, i.e.
xzw ∈ L ⇐⇒ yzw ∈ L. Hence ∼L is a right invariant equivalence on Σ∗ .

Example 4.4.3. Let A = (Q, Σ, δ, q0 , F ) be a DFA. Define the relation ∼A

on Σ∗ by
x ∼A y if and only if δ̂(q0 , x) = δ̂(q0 , y).
Clearly, ∼A is an equivalence relation on Σ∗ . Suppose x ∼A y, i.e. δ̂(q0 , x) =

smartworlD.asia
δ̂(q0 , y). Now, for any z ∈ Σ∗ ,

δ̂(q0 , xz) = δ̂(δ̂(q0 , x), z)

= δ̂(δ̂(q0 , y), z)
= δ̂(q0 , yz)

so that xz ∼A yz. Hence, ∼A is a right invariant equivalence relation.

Theorem 4.4.4 (Myhill-Nerode Theorem). Let L be a language over Σ. The

following three statements regarding L are equivalent:

1. L is accepted by a DFA.

2. There exists a right invariant equivalence relation ∼ of finite index on

Σ∗ such that L is the union of some the equivalence classes of ∼.

3. The equivalence relation ∼L is of finite index.

Proof. (1) ⇒ (2): Assume L is accepted by the DFA A = (Q, Σ, δ, q0 , F ).

First, we observe that the right invariant equivalence relation ∼A is of finite
index. For x ∈ Σ∗ , if δ̂(q0 , x) = p, then the equivalence class containing x

[x]∼A = {y ∈ Σ∗ | δ̂(q0 , y) = p}.

62

Smartzworld.com 63 jntuworldupdates.org
Smartworld.asia Specworld.in

That is, given q ∈ Q, the set

Cq = {x ∈ Σ∗ | δ̂(q0 , x) = q}

is an equivalence class (possibly empty, whenever q is not reachable from q0 )

of ∼A . Thus, the equivalence classes of ∼A are completely determined by
the states of A ; moreover, the number of equivalence classes of ∼A is less
than or equal to the number of states of A . Hence, ∼A is of finite index.
Now,

L = {x ∈ Σ∗ | δ̂(q0 , x) ∈ F }
[
= {x ∈ Σ∗ | δ̂(q0 , x) = p}
p∈F
[
= Cp .
p∈F

as desired.
(2) ⇒ (3): Suppose ∼ is an equivalence with the criterion given in (2).
We show that ∼ is a refinement of ∼L so that the number of equivalence

smartworlD.asia
classes of ∼L is less than or equal to the number of equivalence classes of
∼. For x, y ∈ Σ∗ , suppose x ∼ y. To show x ∼L y, we have to show that
∀z(xz ∈ L ⇐⇒ yz ∈ L). Since ∼ is right invariant, we have xz ∼ yz, for all
z, i.e. xz and yz are in same equivalence class of ∼. As L is the union of
some of the equivalence classes of ∼, we have

∀z(xz ∈ L ⇐⇒ yz ∈ L).

Thus, ∼L is of finite index.

(3) ⇒ (1): Assume ∼L is of finite index. Construct

AL = (Q, Σ, δ, q0 , F )

by setting
n ¯ o
¯
Q = Σ /∼L = [x] ¯ x ∈ Σ , the set of all equivalence classes of Σ∗ with
∗ ∗

respect to ∼L ,

q0 = [ε],
n ¯ o
¯
F = [x] ∈ Q ¯ x ∈ L and

δ : Q × Σ −→ Q is defined by δ([x], a) = [xa], for all [x] ∈ Q and a ∈ Σ.

63

Smartzworld.com 64 jntuworldupdates.org
Smartworld.asia Specworld.in

Since ∼L is of finite index, Q is a finite set; further, for [x], [y] ∈ Q and a ∈ Σ,

[x] = [y] =⇒ x ∼L y =⇒ xa ∼L ya =⇒ [xa] = [ya]

so that δ is well-defined. Hence AL is a DFA. We claim that L(AL ) = L. In

fact, we show that
δ̂(q0 , w) = [w], ∀w ∈ Σ∗ ;
this serves the purpose because w ∈ L ⇐⇒ [w] ∈ F . We prove this by
induction on |w|. Induction basis is clear, because δ̂(q0 , ε) = q0 = [ε]. Also,
by definition of δ, we have δ̂(q0 , a) = δ([ε], a) = [εa] = [a], for all a ∈ Σ.
For inductive step, consider x ∈ Σ∗ and a ∈ Σ. Now

δ̂(q0 , xa) = δ(δ̂(q0 , x), a)

= δ([x], a) (by inductive hypothesis)
= [xa].

This completes the proof.

Remark 4.4.5. Note that, the proof of (2) ⇒ (3) shows that the number of

smartworlD.asia
states of any DFA accepting L is greater than or equal to the index of ∼L
and the proof of (3) ⇒ (1) provides us the DFA AL with number of states
equal to the index of ∼L . Hence, AL is a minimum state DFA accepting L.
Example 4.4.6. Consider the language L = {x ∈ {a, b}∗ | ab is a substring of x}.
We calculate the equivalence classes of ∼L . First observe that the strings ε,
a and ab are not equivalent to each other, because of the following.
1. The string b distinguishes the pair ε and a: εb = b ∈
/ L, whereas ab ∈ L.

2. Any string which does not contain ab distinguishes ε and ab. For in-
stance, εb = b ∈
/ L, whereas abb ∈ L.

3. Also, a and ab are not equivalent because, aa ∈

/ L, whereas aba ∈ L.
Thus, the strings ε, a and ab will be in three different equivalence classes,
say [ε], [a] and [ab], respectively. Now, we show that any other x ∈ {a, b}∗
will be in one of these equivalence classes.
– In case ab is a substring of x, clearly x will be in [ab].

– If ab is not a substring of x, then we discuss in the following subcases.

– For n ≥ 1, x = an : In this case, x will be in [a].

– In case x = bn , for some n ≥ 1, x will be in [ε].

64

Smartzworld.com 65 jntuworldupdates.org
Smartworld.asia Specworld.in

– If x has some a’s and some b’s, then x must be of the form bn am ,
for n, m ≥ 1. In this case, x will be in [a].

Thus, ∼L has exactly three equivalence classes and hence it is of finite index.
By Myhill-Nerode theorem, there exists a DFA accepting L.
Example 4.4.7. Consider the language L = {an bn | n ≥ 1} over the alphabet
{a, b}. We show that the index of ∼L is not finite. Hence, by Myhill-Nerode
theorem, there exists no DFA accepting L. For instance, consider an , am ∈
{a, b}∗ , for m 6= n. They are not equivalent with respect to ∼L , because, for
bn ∈ {a, b}∗ , we have

an bn ∈ L, whereas am bn ∈
/ L.

Thus, for each n, there has to be one equivalence classes to accommodate an .

Hence, the index of ∼L is not finite.

4.4.2 Algorithmic Procedure for Minimization

Given a DFA, there may be certain states which are redundant in the sense

smartworlD.asia
that their roles in the language acceptance are same. Here, two states are
said to have same role, if it will lead us to either both final states or both non-
final states for every input string; so that they contribute in same manner
in language acceptance. Among such group of states, whose roles are same,
only one state can be considered and others can be discarded to reduce the
number states without affecting the language. Now, we formulate this idea
and present an algorithmic procedure to minimize the number of states of
a DFA. In fact, we obtain an equivalent DFA whose number of states is
minimum.
Definition 4.4.8. Two states p and q of a DFA A are said to be equivalent,
denoted by p ≡ q, if for all x ∈ Σ∗ both the states δ(p, x) and δ(q, x) are
either final states or non-final states.
Clearly, ≡ is an equivalence relation on the set of states of A . Given
two states p and q, to test whether p ≡ q we need to check the condition for
all strings in Σ∗ . This is practically difficult, since Σ∗ is an infinite set. So,
in the following, we introduce a notion called k-equivalence and build up a
technique to test the equivalence of states via k-equivalence.
Definition 4.4.9. For k ≥ 0, two states p and q of a DFA A are said to be
k
k-equivalent, denoted by p ≡ q, if for all x ∈ Σ∗ with |x| ≤ k both the states
δ(p, x) and δ(q, x) are either final states or non-final states.

65

Smartzworld.com 66 jntuworldupdates.org
Smartworld.asia Specworld.in

k
Clearly, ≡ is also an equivalence relation on the set of states of A . Since
there are only finitely man strings of length up to k over an alphabet, one may
easily test the k-equivalence between the states. Let us denote the partition
Q Q k
of Q under the relation ≡ by , whereas it is k under the relation ≡.
Remark 4.4.10. For any p, q ∈ Q,
k
p ≡ q if and only if p ≡ q for all k ≥ 0.
k k−1
Also, for any k ≥ 1 and p, q ∈ Q, if p ≡ q, then p ≡ q.
Given a k-equivalence relation over the set of states, the following theorem
provides us a criterion to calculate the (k + 1)-equivalence relation.

Theorem 4.4.11. For any k ≥ 0 and p, q ∈ Q,

k+1 0 k
p ≡ q if and only if p ≡ q and δ(p, a) ≡ δ(q, a) ∀a ∈ Σ.
k+1 k
Proof. If p ≡ q then p ≡ q holds (cf. Remark 4.4.10). Let x ∈ Σ∗ with |x| ≤
k+1

smartworlD.asia
k and a ∈ Σ be arbitrary. Then since p ≡ q, δ(δ(p, a), x) and δ(δ(q, a), x)
k
both are either final states or non-final states, so that δ(p, a) ≡ δ(q, a).
0 k
Conversely, for k ≥ 0, suppose p ≡ q and δ(p, a) ≡ δ(q, a) ∀a ∈ Σ. Note
that for x ∈ Σ∗ with |x| ≤ k and a ∈ Σ, δ(δ(p, a), x) and δ(δ(q, a), x) both are
either final states or non-final states, i.e. for all y ∈ Σ∗ and 1 ≤ |y| ≤ k + 1,
both the states δ(p, y) and δ(q, y) are final or non-final states. But since
0 k+1
p ≡ q we have p ≡ q.
Remark 4.4.12. Two k-equivalent states p and q will further be (k + 1)-
k Q Q
equivalent if δ(p, a) ≡ δ(q, a) ∀a ∈ Σ, i.e. from k we can obtain k+1 .
Q Q Q Q
Theorem 4.4.13. If k = k+1 , for some k ≥ 0, then k = .
Q Q
Proof. Suppose k = k+1 . To prove that, for p, q ∈ Q,

k+1 n
p ≡ q =⇒ p ≡ q ∀n ≥ 0

it is enough to prove that

k+1 k+2
p ≡ q =⇒ p ≡ q.

66

Smartzworld.com 67 jntuworldupdates.org
Smartworld.asia Specworld.in

Now,
k+1 k
p ≡ q =⇒ δ(p, a) ≡ δ(q, a) ∀a ∈ Σ
k+1
=⇒ δ(p, a) ≡ δ(q, a) ∀a ∈ Σ
k+2
=⇒ p ≡ q

Hence the result follows by induction.

Q
Using the above results, we illustrate calculating the partition for the
following DFA.

Example 4.4.14. Consider the DFA given in following transition table.

δ a b
→ q0 q3 q2
q1 q6 q2
q2 q8 q5
07q
16523 43 q0 q1
q2 q5

smartworlD.asia
07q
16524 43
q5 q4 q3
07q
16526 43 q1 q0
q7 q4 q6
07q
16528 43 q2 q7
q9 q7 q10
q10 q5 q9

From the definition, two states p and q are 0-equivalent if both δ̂(p, x) and
δ̂(q, x) are either final states or non-final states, for all |x| = 0. That is, both
δ̂(p, ε) and δ̂(q, ε) are either final states or non-final states. That is, both p
and q are either final states or non-final states.
0
Thus, under the equivalence relation ≡, all final states are equivalent and
all non-final states are equivalent.
Q Hence, there are precisely two equivalence
classes in the partition 0 as given below:
Q n o
0 = {q0 , q1 , q2 , q5 , q7 , q9 , q10 }, {q3 , q4 , q6 , q8 }

From the Theorem 4.4.11, we know that any two 0-equivalent states, p
and q, will further be 1-equivalent if
0
δ(p, a) ≡ δ(q, a) ∀a ∈ Σ.

67

Smartzworld.com 68 jntuworldupdates.org
Smartworld.asia Specworld.in

Q
Using this condition we can evaluate 1 by checking every two 0-equivalent
states whether they are further 1-equivalent or not. If they are 1-equivalent
they continue to be in the same equivalence class. Otherwise, they will be put
in different
Q equivalence classes. The following shall illustrate the computation
of 1 :

1. Consider the 0-equivalent states q0 and q1 and observe that

(i) δ(q
Q 0 , a) = q3 and δ(q1 , a) = q6 are in the same equivalence class of
0 ; and also
(ii) Q
δ(q0 , b) = q2 and δ(q1 , b) = q2 are in the same equivalence class of
0.

1
Thus,
Q q0 ≡ q1 so that they will continue to be in same equivalence class
in 1 also.

2. In contrast to the above, consider the 0-equivalent states q2 and q5 and

observe that

(i) δ(q2 , a) and δ(q5 , a) are, respectively, q8 and q4 ; which are in the

smartworlD.asia
Q
same equivalence class of 0 .
(ii) Whereas, δ(q2 , b)Q= q5 and δ(q5 , b) = q3 are in different equiva-
lences classes of 0 .

Hence, q2 and q5 are Qnot 1-equivalent. So, they will be in different

equivalence classes of 1 .
1
3. Further, as illustrated above, one may verify whether are not q2 ≡ q7
and realize that q2 and q7 are not 1-equivalent. While putting q7 in a
1
different class that of q2 , we check for q5 ≡ q7 to decide to put q5 and
q7 in the same equivalence class or not. As Q q5 and q7 are 1-equivalent,
they will be put in the same equivalence of 1 .

Since, any two states which are not k-equivalent cannot be (k Q+1)-equivalent,
we check for those pairs belonging to same equivalence of 0 whether they
are further 1-equivalent. Thus, we obtain
Q n o
1 = {q0 , q1 , q2 }, {q5 , q7 }, {q9 , q10 }, {q3 , q4 , q6 , q8 }
Q Q
Similarly, we continue to compute 2 , 3 , etc.
Q n o
2 = {q0 , q1 , q2 }, {q5 , q7 }, {q9 , q10 }, {q3 , q6 }, {q4 , q8 }

68

Smartzworld.com 69 jntuworldupdates.org
Smartworld.asia Specworld.in

Q n o
3 = {q ,
0 1q }, {q2 }, {q ,
5 7 q }, {q ,
9 10q }, {q ,
3 6 q }, {q ,
4 8q }
Q n o
4 = {q0 , q1 }, {q2 }, {q5 , q7 }, {q9 , q10 }, {q3 , q6 }, {q4 , q8 }

Note Qthat theQ process for a DFA will always terminate at a finite stage
and get k = k+1 , for some k. This is because, there are finite number
of states and in the worst case equivalences may end up with singletons.
Thereafter no further refinementQ isQpossible. Q
In the present context, 3 = 4 . Thus it is partition corresponding
to ≡. Now we construct a DFA with these equivalence classes as states (by
renaming them, for simplicity) and with the induced transitions. Thus we
have an equivalent DFA with fewer number of states that of the given DFA.
The DFA with the equivalences classes as states is constructed below:
Let P = {p1 , . . . , p6 }, where p1 = {q0 , q1 }, p2 = {q2 }, p3 = {q5 , q7 },
p4 = {q9 , q10 }, p5 = {q3 , q6 }, p6 = {q4 , q8 }. As p1 contains the initial state
q0 of the given DFA, p1 will be the initial state. Since p5 and p6 contain the
final states of the given DFA, these two will form the set of final states. The
induced transition function δ 0 is given in the following table.
δ0 a b

smartworlD.asia
→ p1 p5 p2
p2 p6 p3
p3 p6 p5
p4 p3 p4
07p1652543 p1 p1
07p
6152643 p2 p3

Here, note that the state p4 is inaccessible from the initial state p1 . This will
also be removed and the following further simplified DFA can be produced
with minimum number of states.
­
b
p1
/ GFED
@ABC p2
/ GFED
@ABC
E } } Y
}}}
b }}
a,b a
}} a a
}}
}
²
b
}~ } a ²
p5 o
GFED
@ABC
?>=<
89:; p3
GFED
@ABC p6
/ GFED
@ABC
?>=<
89:;
f
b

The following theorem confirms that the DFA obtained in this procedure,
in fact, is having minimum number of states.
Theorem 4.4.15. For every DFA A , there is an equivalent minimum state
DFA A 0 .

69

Smartzworld.com 70 jntuworldupdates.org
Smartworld.asia Specworld.in

Proof. Let A = (Q, Σ, δ, q0 , F ) be a DFA and ≡ be the state equivalence as

defined in the Definition 4.4.8. Construct

A 0 = (Q0 , Σ, δ 0 , q00 , F 0 )

where
Q0 = {[q] | q is accessible from q0 }, the set of equivalence classes of Q
with respect to ≡ that contain the states accessible from q0 ,
q00 = [q0 ],
F 0 = {[q] ∈ Q0 | q ∈ F } and
δ 0 : Q0 × Σ −→ Q0 is defined by δ 0 ([q], a) = [δ(q, a)].
For [p], [q] ∈ Q0 , suppose [p] = [q], i.e. p ≡ q. Now given a ∈ Σ, for each
x ∈ Σ∗ , both δ̂(δ(p, a), x) = δ̂(p, ax) and δ̂(δ(q, a), x) = δ̂(q, ax) are final or
non-final states, as p ≡ q. Thus, δ(p, a) and δ(q, a) are equivalent. Hence, δ 0
is well-defined and A 0 is a DFA.
Claim 1: L(A ) = L(A 0 ).
Proof of Claim 1: In fact, we show that δ̂ 0 ([q0 ], x) = [δ̂(q0 , x)], for all

smartworlD.asia
x ∈ Σ∗ . This suffices because

δ̂ 0 ([q0 ], x) ∈ F 0 ⇐⇒ δ̂(q0 , x) ∈ F.

We prove our assertion by induction on |x|. Basis for induction is clear,

because δ̂ 0 ([q0 ], ε) = [q0 ] = [δ̂(q0 , ε)]. For inductive step, consider x ∈ Σ∗ and
a ∈ Σ. Now,

δ̂ 0 ([q0 ], xa) = δ 0 (δ̂ 0 ([q0 ], x), a)

= δ 0 ([δ̂(q0 , x)], a) (by inductive hypothesis)
= [δ(δ̂(q0 , x), a)] (by definition of δ 0 )
= [δ̂(q0 , xa)]

as desired.
Claim 2: A 0 is a minimal state DFA accepting L.
Proof of Claim 2: We prove that the number of states of A 0 is equal to
the number of states of AL , a minimal DFA accepting L. Since the states of
AL are the equivalence classes of ∼L , it is sufficient to prove that

the index of ∼L = the index of ∼A 0 .

Recall the proof (2) ⇒ (3) of Myhill Nerode theorem and observe that

the index of ∼L ≤ the index of ∼A 0 .

70

Smartzworld.com 71 jntuworldupdates.org
Smartworld.asia Specworld.in

On the other hand, suppose there are more number of equivalence classes for
A 0 then that of ∼L does. That is, there exist x, y ∈ Σ∗ such that x ∼L y,
but not x ∼A 0 y.
That is, δ̂ 0 ([q0 ], x) 6= δ̂ 0 ([q0 ], y).
That is, [δ̂(q0 , x)] 6= [δ̂(q0 , y)].
That is, one among δ̂(q0 , x) and δ̂(q0 , y) is a final state and the other is
a non-final state.
That is, x ∈ L ⇐⇒ y ∈
/ L. But this contradicts the hypothesis that
x ∼L y.
Hence, index of ∼L = index of ∼A 0 .
Example 4.4.16. Consider the DFA given in following transition table.
δ a b
→ 07q16520 43 q1 q0
q1 q0 q3
07q 16522 43 q4 q5
07q 16523 43 q4 q1

smartworlD.asia 07q 16524 43

q5
q6
q2
q0
q3
q6
q2
q4
Q
We compute
n the partition through o the following:
Q
0 = {q0 , q2 , q3 , q4 }, {q1 , q5 , q6 }
Q n o
1 = {q0 }, {q 2 , q 3 , q 4 }, {q 1 , q 5 , q 6 }
Q n o
2 = {q0 }, {q2 , q3 , q4 }, {q1 , q5 }, {q6 }
Q n o
3 = {q0 }, {q ,
2 3 q }, {q 4 }, {q ,
1 5 q }, {q 6 }
Q n o
4 = {q0 }, {q 2 , q 3 }, {q 4 }, {q 1 , q 5 }, {q 6 }
Q Q Q Q
Since 3 = 4, we have 4 = . By renaming the equivalence classes
as
p0 = {q0 }; p1 = {q1 , q5 }; p2 = {q2 , q3 }; p3 = {q4 }; p4 = {q6 }
we construct an equivalent minimal DFA as shown in the following.
a

a b v a b
p0
/ GFED
@ABC
?>=<
89:; p1
/ GFED
@ABC p2
/ GFED
@ABC
?>=<
89:; p3
/ GFED
@ABC
?>=<
89:; p4
/ GFED
@ABC
d d d d
a b a b

71

Smartzworld.com 72 jntuworldupdates.org
Smartworld.asia Specworld.in

4.5 Regular Languages

Language represented by a regular expression is defined as a regular language.
Now, we are in a position to provide alternative definitions for regular lan-
guages via finite automata (DFA or NFA) and also via regular grammars.
That is the class of regular languages is precisely the class of languages ac-
cepted by finite automata and also it is the class of languages generated
by regular grammars. These results are given in the following subsections.
We also provide an algebraic method for converting a DFA to an equivalent
regular expression.

4.5.1 Equivalence of Finite Automata and Regular Lan-

guages
A regular expressions r is said to be equivalent to a finite automaton A , if
the language represented by r is precisely accepted by the finite automaton
A , i.e. L(r) = L(A ). In order to establish the equivalence, we prove the
following.

smartworlD.asia
1. Given a regular expression, we construct an equivalent finite automa-
ton.

2. Given a DFA A , we show that L(A ) is regular.

To prove (1), we need the following. Let r1 and r2 be two regular ex-
pressions. Suppose there exist finite automata A1 = (Q1 , Σ, δ1 , q1 , F1 ) and
A2 = (Q2 , Σ, δ2 , q2 , F2 ) which accept L(r1 ) and L(r2 ), respectively. Under
this hypothesis, we prove the following three lemmas.

Lemma 4.5.1. There exists a finite automaton accepting L(r1 + r2 ).

Proof. Construct A = (Q, Σ, δ, q0 , F ), where

Q = Q1 ∪ Q2 ∪ {q0 } with a new state q0 ,
F = F1 ∪ F2 and
δ : Q × (Σ ∪ {ε}) −→ P(Q) defined by

 δ1 (q, a), if q ∈ Q1 , a ∈ Σ ∪ {ε};
δ(q, a) = δ2 (q, a), if q ∈ Q2 , a ∈ Σ ∪ {ε};

{q1 , q2 }, if q = q0 , a = ε.

Then clearly, A is a finite automaton as depicted in Figure 4.4.

72

Smartzworld.com 73 jntuworldupdates.org
Smartworld.asia Specworld.in

A
q1
ε A1 F1

q0

ε A2 F2
q2

Figure 4.4: Parallel Composition of Finite Automata

Now, for x ∈ Σ∗ ,

x ∈ L(A ) ⇐⇒ (δ̂(q0 , x) ∩ F ) 6= ∅

smartworlD.asia
⇐⇒ (δ̂(q0 , εx) ∩ F ) 6= ∅
⇐⇒ (δ̂(δ̂(q0 , ε), x) ∩ F ) 6= ∅
⇐⇒ (δ̂({q1 , q2 }, x) ∩ F ) 6= ∅
⇐⇒ ((δ̂(q1 , x) ∪ δ̂(q2 , x)) ∩ F ) 6= ∅
⇐⇒ ((δ̂(q1 , x) ∩ F ) ∪ (δ̂(q2 , x) ∩ F )) 6= ∅
⇐⇒ (δ̂(q1 , x) ∩ F1 ) 6= ∅ or (δ̂(q2 , x) ∩ F2 ) 6= ∅
⇐⇒ x ∈ L(A1 ) or x ∈ L(A2 )
⇐⇒ x ∈ L(A1 ) ∪ L(A2 ).

Hence, L(A ) = L(A1 ) ∪ L(A2 ).

Lemma 4.5.2. There exists a finite automaton accepting L(r1 r2 ).
Proof. Construct
A = (Q, Σ, δ, q1 , F2 ),
where Q = Q1 ∪ Q2 and δ is defined by

 δ1 (q, a), if q ∈ Q1 , a ∈ Σ ∪ {ε};
δ(q, a) = δ2 (q, a), if q ∈ Q2 , a ∈ Σ ∪ {ε};

{q2 }, if q ∈ F1 , a = ε.

Then A is a finite automaton as shown in Figure 4.5.

73

Smartzworld.com 74 jntuworldupdates.org
Smartworld.asia Specworld.in

A
ε
q1
A1 F1 A2 F2
q2

Figure 4.5: Series Composition of Finite Automata

We claim that L(A ) = L(A1 )L(A2 ). Suppose x = a1 a2 . . . an ∈ L(A ),

i.e. δ̂(q1 , x) ∩ F2 6= ∅. It is clear from the construction of A that the only
way to reach from q1 to any state of F2 is via q2 . Further, we have only
ε-transitions from the states of F1 to q2 . Thus, while traversing through x
from q1 to some state of F2 , there exist p ∈ F1 and a number k ≤ n such that

p ∈ δ̂(q1 , a1 a2 . . . ak ) and δ̂(q2 , ak+1 ak+2 . . . an ) ∩ F2 6= ∅.

Then
smartworlD.asia
x1 = a1 a2 . . . ak ∈ L(A1 ) and x2 = ak+1 ak+2 . . . an ∈ L(A2 )

so that x = x1 x2 ∈ L(A1 )L(A2 ).

Conversely, suppose x ∈ L(A1 )L(A2 ). Then x = x1 x2 , for some x1 ∈
L(A1 ) and some x2 ∈ L(A2 ). That is,

δ̂1 (q1 , x1 ) ∩ F1 6= ∅ and δ̂2 (q2 , x2 ) ∩ F2 6= ∅.

Now,

(q1 , x) = (q1 , x1 x2 )
|−*− (p, x2 ), for some p ∈ F1
= (p, εx2 )
|−− (q2 , x2 ), since δ(p, ε) = {q2 }
|−*− (p0 , ε), for some p0 ∈ F2

Hence, δ̂(q1 , x) ∩ F2 6= ∅ so that x ∈ L(A ).

Lemma 4.5.3. There exists a finite automaton accepting L(r1 )∗ .

74

Smartzworld.com 75 jntuworldupdates.org
Smartworld.asia Specworld.in

Proof. Construct
A = (Q, Σ, δ, q0 , F ),
where Q = Q1 ∪ {q0 , p} with new states q0 and p, F = {p} and define δ by
½
{q1 , p}, if q ∈ F1 ∪ {q0 } and a = ε
δ(q, a) =
δ1 (q, a), if q ∈ Q1 and a ∈ Σ ∪ {ε}

Then A is a finite automaton as given in Figure 4.6.

A ε

q1
A1 F1
ε p
q0 ε

smartworlD.asia
Figure 4.6: Kleene Star of a Finite Automaton

We prove that L(A ) = L(A1 )∗ . For x ∈ Σ∗ ,

x ∈ L(A ) =⇒ δ̂(q0 , x) = {p}

=⇒ either x = ε or δ̂(q0 , x) ∩ F1 6= ∅.

If x = ε then trivially x ∈ L(A1 )∗ . Otherwise, there exist

p1 , p2 , . . . , pk ∈ F1 and x1 , x2 , . . . , xk

such that
p1 ∈ δ̂(q1 , x1 ), p2 ∈ δ̂(q1 , x2 ), . . . , pk ∈ δ̂(q1 , xk )
with x = x1 x2 . . . xk . Thus, for all 1 ≤ i ≤ k, xi ∈ L(A1 ) so that x ∈ L(A1 )∗ .
Conversely, suppose x ∈ L(A1 )∗ . Then x = x1 x2 · · · xl with xi ∈ L(A1 )
for all i and for some l ≥ 0. If l = 0, then x = ε and clearly, x ∈ L(A ).
Otherwise, we have

δ̂1 (q1 , xi ) ∩ F1 6= ∅ for 1 ≤ i ≤ l.

75

Smartzworld.com 76 jntuworldupdates.org
Smartworld.asia Specworld.in

Now, consider the following computation of A on x:

(q0 , x) = (q0 , εx)

−
|− (q1 , x)
= (q1 , x1 x2 · · · xl )
|−*− (p01 , x2 x3 · · · xl ), for some p01 ∈ F1
= (p01 , εx2 x3 . . . xl )
−
|− (q1 , x2 x3 . . . xl ), since q1 ∈ δ(p01 , ε)
..
.
|−*− (p0l , ε), for some p0l ∈ F1
−
|− (p, ε), since p ∈ δ(p0l , ε).

As δ̂(q0 , x) ∩ F 6= ∅ we have x ∈ L(A ).

Theorem 4.5.4. The language denoted by a regular expression can be ac-
cepted by a finite automaton.
Proof. We prove the result by induction on the number of operators of a

smartworlD.asia
regular expression r. Suppose r has zero operators, then r must be ε, ∅ or a
for some a ∈ Σ.
If r = ε, then the finite automaton as depicted below serves the pur-
pose.
q
/ ?89:;
>=<
7654
0123

If r = ∅, then (i) any finite automaton with no final state will do;
or (ii) one may consider a finite automaton in which final states are
not accessible from the initial state. For instance, the following two
automata are given for each one of the two types indicated above and
serve the purpose.
∀a∈Σ ∀a∈Σ
­ ·
∀a∈Σ
p
/ ?89:;
>=< p o
/ ?89:;
>=< q
?>=<
89:;
7654
0123

(i) (ii)

In case r = a, for some a ∈ Σ,

a
p
/ ?89:;
>=< q
/ ?89:;
>=<
7654
0123

is a finite automaton which accepts r.

76

Smartzworld.com 77 jntuworldupdates.org
Smartworld.asia Specworld.in

Suppose the result is true for regular expressions with k or fewer operators
and assume r has k + 1 operators. There are three cases according to the
operators involved. (1) r = r1 +r2 , (2) r = r1 r2 , or (3) r = r1∗ , for some regular
expressions r1 and r2 . In any case, note that both the regular expressions r1
and r2 must have k or fewer operators. Thus by inductive hypothesis, there
exist finite automata A1 and A2 which accept L(r1 ) and L(r2 ), respectively.
Then, for each case we have a finite automaton accepting L(r), by Lemmas
4.5.1, 4.5.2, or 4.5.3, case wise.

Example 4.5.5. Here, we demonstrate the construction of an NFA for the

regular expression a∗ b + a. First, we list the corresponding NFA for each
subexpression of a∗ b + a.
//()*+
.-, a //()*+
.-,
ÂÁÀ¿
»¼½¾
a :
ε

ε Ä a ε
a∗ : //()*+
.-, /.-,
/()*+ //()*+
.-, /.-,
7/()*+
ÂÁÀ¿
»¼½¾

ε
//()*+
.-, b //()*+
.-,
ÂÁÀ¿
»¼½¾
b :

smartworlD.asia
ε

ε Ä a ε ε b
a∗ b : //()*+
.-, /.-,
/()*+ //()*+
.-, /.-,
7/()*+ //()*+
.-, //()*+
.-,
ÂÁÀ¿
»¼½¾

ε
Now, finally an NFA for a∗ b + a is:
ε

ε Ä a ε ε b
/.-,
()*+ /.-,
()*+
/ /()*+
/ .-, /.-,
7/()*+ /()*+
/ .-, /()*+
/ .-,
ÂÁÀ¿
»¼½¾
ÄÄ?
ε ÄÄ
ÄÄÄ
.-,Ä?
ε
//()*+
??
??
ε ???
Â
/.-,
()*+ //()*+
.-,
ÂÁÀ¿
»¼½¾
a

Theorem 4.5.6. If A is a DFA, then L(A ) is regular.

Proof. We prove the result by induction on the number of states of DFA. For
base case, let A = (Q, Σ, δ, q0 , F ) be a DFA with only one state. Then there
are two possibilities for the set of final states F .

F = ∅: In this case L(A ) = ∅ and is regular.

F = Q: In this case L(A ) = Σ∗ which is already shown to be regular.

77

Smartzworld.com 78 jntuworldupdates.org
Smartworld.asia Specworld.in

L1
L3

q0

q 0 does not appear on these paths

Figure 4.7: Depiction of the Language of a DFA

Assume that the result is true for all those DFA whose number of states is
less than n. Now, let A = (Q, Σ, δ, q0 , F ) be a DFA with |Q| = n. First note
that the language L = L(A ) can be written as
L = L∗1 L2
where
L1 is the set of strings that start and end in the initial state q0

smartworlD.asia
L2 is the set of strings that start in q0 and end in some final state. We
include ε in L2 if q0 is also a final state. Further, we add a restriction
that q0 will not be revisited while traversing those strings. This is
justified because, the portion of a string from the initial position q0 till
that revisits q0 at the last time will be part of L1 and the rest of the
portion will be in L2 .
Using the inductive hypothesis, we prove that both L1 and L2 are regular.
Since regular languages are closed with respect to Kleene star and concate-
nation it follows that L is regular.
The following notation shall be useful in defining the languages L1 and
L2 , formally, and show that they are regular. For q ∈ Q and x ∈ Σ∗ , let us
denote the set of states on the path of x from q that come after q by P(q,x) .
That is, if x = a1 · · · an ,
n
[
P(q,x) = {δ̂(q, a1 · · · ai )}.
i=1

Define
L1 = {x ∈ Σ∗ | δ̂(q0 , x) = q0 }; and
½
L3 , if q0 ∈
/ F;
L2 =
L3 ∪ {ε}, if q0 ∈ F,

78

Smartzworld.com 79 jntuworldupdates.org
Smartworld.asia Specworld.in

where L3 = {x ∈ Σ∗ | δ̂(q0 , x) ∈ F, q0 ∈ / P(q0 ,x) }. Clearly, L = L∗1 L2 .

Claim 1: L1 is regular.
Proof of Claim 1:
Consider the following set
 ¯ 
 ¯ δ̂(q0 , axb) = q0 , for some x ∈ Σ∗ ; 
¯
A = (a, b) ∈ Σ × Σ ¯¯ δ(q0 , a) 6= q0 ;
 ¯ q0 ∈ 
/ P(qa ,x) , where qa = δ(q0 , a).

and for (a, b) ∈ A, define

L(a,b) = {x ∈ Σ∗ | δ̂(q0 , axb) = q0 and q0 ∈

/ P(qa ,x) , where qa = δ(q0 , a)}.

Note that L(a,b) is the language accepted by the DFA

A(a,b) = (Q0 , Σ, δ 0 , qa , F 0 )

where Q0 = Q \ {q0 }, qa = δ(q0 , a), F 0 = {q ¯ ∈ Q | δ(q, b) = q0 } \ {q0 }, and δ

0
¯
is the restriction of δ to Q0 × Σ, i.e. δ 0 = δ ¯ . For instance, if x ∈ L(A(a,b) )
Q0 ×Σ

smartworlD.asia
0
then, since q0 ∈ / P(qa ,x) and δ̂ (qa , x) ∈ F 0 . This implies,
/ Q , q0 ∈ 0

δ̂(q0 , axb) = δ̂(δ(q0 , a), xb)

= δ̂(qa , xb)
= δ(δ̂(qa , x), b)
¯
¯
= δ(p, b), where p ∈ F 0 , as δ 0 = δ ¯
Q0 ×Σ
= q0 ,

so that x ∈ L(a,b) . Converse is similar. Thus L(a,b) = L(A(a,b) ). Hence, as

|Q0 | = n − 1, by inductive hypothesis, L(a,b) is regular. Now if we write

B = {a ∈ Σ | δ(q0 , a) = q0 } ∪ {ε}

then clearly, [
L1 = B ∪ aL(a,b) b.
(a,b)∈A

Hence L1 is regular.
Claim 2: L3 is regular.
Proof of Claim 2: Consider the following set

C = {a ∈ Σ | δ(q0 , a) 6= q0 }

79

Smartzworld.com 80 jntuworldupdates.org
Smartworld.asia Specworld.in

and for a ∈ C, define

La = {x ∈ Σ∗ | δ̂(q0 , ax) ∈ F and q0 ∈
/ P(qa ,x) , where qa = δ(q0 , a)}.
For a ∈ C, we set
Aa = (Q0 , Σ, δ 0 , qa , F 00 )
where Q0 = Q \ {q0 }, qa¯ = δ(q0 , a), F 00 = F \ {q0 }, and δ 0 is the restriction of
¯
δ to Q0 × Σ, i.e. δ 0 = δ ¯ . It easy to observe that L(Aa ) = La . First note
Q0 ×Σ
that q0 does not appear in the context of La and L(Aa ). Now,
x ∈ L(Aa ) ⇐⇒ δ̂ 0 (qa , x) ∈ F 00 (as q0 does not appear)
⇐⇒ δ̂ 0 (qa , x) ∈ F
⇐⇒ δ̂(qa , x) ∈ F
⇐⇒ δ̂(δ(q0 , a), x) ∈ F
⇐⇒ δ̂(q0 , ax) ∈ F (as q0 does not appear)
⇐⇒ ax ∈ La .
Again, since |Q0 | = n − 1, by inductive hypothesis, we have La is regular.

smartworlD.asia
But, clearly [
L3 = aLa
a∈C

so that L3 is regular. This completes the proof of the theorem.

Example 4.5.7. Consider the following DFA
a
a b
· z ­
b b
q0 f
/ G@ABC
FED q1
/ G@ABC
FED q2
/ G@ABC
FED
?>=<
89:;

1. Note that the following strings bring the DFA from q0 back to q0 .
(a) a (via the path hq0 , q0 i)
(b) ba (via the path hq0 , q1 , q0 i)
(c) For n ≥ 0, bbbn a (via the path hq0 , q1 , q2 , q2 , . . . , q2 , q0 i)
Thus L1 = {a, ba, bbbn a | n ≥ 0} = a + ba + bbb∗ a.
2. Again, since q0 is not a final state, L2 – the set of strings which take
the DFA from q0 to the final state q2 – is
{bbbn | n ≥ 0} = bbb∗ .

80

Smartzworld.com 81 jntuworldupdates.org
Smartworld.asia Specworld.in

Thus, as per the construction of Theorem 4.5.6, the language accepted by

the given DFA is L∗1 L2 and it can be represented by the regular expression

(a + ba + bbb∗ a)∗ bbb∗

and hence it is regular.

Brzozowski Algebraic Method

Brzozowski algebraic method gives a conversion of DFA to its equivalent reg-
ular expression. Let A = (Q, Σ, δ, q0 , F ) be a DFA with Q = {q0 , q2 , . . . , qn }
and F = {qf1 , . . . , qfk }. For each qi ∈ Q, write

Ri = {x ∈ Σ∗ | δ̂(q0 , x) = qi }

and note that

k
[
L(A ) = R fi
i=1

In order to construct a regular expression for L(A ), here we propose an

smartworlD.asia
unknown expression for each Ri , say ri . We are indented to observe that ri ,
for all i, is a regular expression so that

rf1 + rf2 + · · · + rfk

is a regular expression for L(A ), as desired.

Suppose Σ(i,j) is the set of those symbols of Σ which take A from qi to
qj , i.e.
Σ(i,j) = {a ∈ Σ | δ(qi , a) = qj }.
Clearly, as it is a finite set, Σ(i,j) is regular with the regular expression as
sum of its symbols. Let s(i,j) be the expression for Σ(i,j) . Now, for 1 ≤ j ≤ n,
since the strings of Rj are precisely taking the DFA from q0 to any state qi
then reaching qj with the symbols of Σ(i,j) , we have

Rj = R0 Σ(0,j) ∪ · · · ∪ Ri Σ(i,j) ∪ · · · ∪ Rn Σ(n,j) .

And in case of R0 , it is

R0 = R0 Σ(0,0) ∪ · · · ∪ Ri Σ(i,0) ∪ · · · ∪ Rn Σ(n,0) ∪ {ε},

as ε takes the DFA from q0 to itself. Thus, for each j, we have an equation

81

Smartzworld.com 82 jntuworldupdates.org
Smartworld.asia Specworld.in

q
0

R0
Rn

R1 Rj
Ri
q q q q q
0 1 i j n
Σ
(1, j ) Σ Σ
Σ
( i, j ) ( j, j ) Σ
( n ,j )
(0, j )
q
j

Figure 4.8: Depiction of the Set Rj

for rj which is depending on all ri 0 s, called characteristic equation of rj . The

system of characteristic equations of A is:

r0 = r0 s(0,0) + r1 s(1,0) + · · · + ri s(i,0) + · · · + rn s(n,0) + ε

smartworlD.asia
r1 = r0 s(0,1) + r1 s(1,1) + · · · + ri s(i,1) + · · · + rn s(n,1)
..
.
rj = r0 s(0,j) + r1 s(1,j) + · · · + ri s(i,j) + · · · + rn s(n,j)
..
.
rn = r0 s(0,n) + r1 s(1,n) + · · · + ri s(i,n) + · · · + rn s(n,n)

The system can be solved for rfi 0 s, expressions for final states, via straight-
forward substitution, except the same unknown appears on the both the left
and right hand sides of a equation. This situation can be handled using
Arden’s principle (see Exercise ??) which states that
Let s and t be regular expressions and r is an unknown. A equation
of the form r = t + rs, where ε ∈
/ L(s), has a unique solution given by
∗
r = ts .
By successive substitutions and application of Arden’s principle we evaluate
the expressions for final states purely in terms of symbols from Σ. Since the
operations involved here are admissible for regular expression, we eventually
obtain regular expressions for each rfi .
We demonstrate the method through the following examples.

82

Smartzworld.com 83 jntuworldupdates.org
Smartworld.asia Specworld.in

Example 4.5.8. Consider the following DFA

b b
­ ­
a
q0
/ G@ABC
FED q1
/ G@ABC
FED
?>=<
89:;
d
a

The characteristic equations for the states q0 and q1 , respectively, are:

r0 = r0 b + r1 a + ε
r1 = r0 a + r1 b
Since q1 is the final state, r1 represents the language of the DFA. Hence,
we need to solve for r1 explicitly in terms of a0 s and b0 s. Now, by Arden’s
principle, r0 yields
r0 = (ε + r1 a)b∗ .
Then, by substituting r0 in r1 , we get
r1 = (ε + r1 a)b∗ a + r1 b = b∗ a + r1 (ab∗ a + b)
Then, by applying Arden’s principle on r1 , we get r1 = b∗ a(ab∗ a + b)∗ which
is the desired regular expression representing the language of the given DFA.

smartworlD.asia
Example 4.5.9. Consider the following DFA

­
b
a a
­
a, b

q1
/ G@ABC
FED
?>=<
89:; q2
/ G@ABC
FED
?>=<
89:; q3
/ G@ABC
FED
d
b

The characteristic equations for the states q1 , q2 and q3 , respectively, are:

r1 = r1 b + r2 b + ε
r2 = r1 a
r3 = r2 a + r3 (a + b)
Since q1 and q2 are final states, r1 + r2 represents the language of the DFA.
Hence, we need to solve for r1 and r2 explicitly in terms of a0 s and b0 s. Now,
by substituting r2 in r1 , we get
r1 = r1 b + r1 ab + ε = r1 (b + ab) + ε.
Then, by Arden’s principle, we have r1 = ε(b + ab)∗ = (b + ab)∗ . Thus,
r1 + r2 = (b + ab)∗ + (b + ab)∗ a.
Hence, the regular expressions represented by the given DFA is
(b + ab)∗ (ε + a).

83

Smartzworld.com 84 jntuworldupdates.org
Smartworld.asia Specworld.in

4.5.2 Equivalence of Finite Automata and Regular Gram-

mars
A finite automaton A is said to be equivalent to a regular grammar G, if
the language accepted by A is precisely generated by the grammar G, i.e.
L(A ) = L(G). Now, we prove that finite automata and regular grammars
are equivalent. In order to establish this equivalence, we first prove that
given a DFA we can construct an equivalent regular grammar. Then for
converse, given a regular grammar, we construct an equivalent generalized
finite automaton (GFA), a notion which we introduce here as an equivalent
notion for DFA.

Theorem 4.5.10. If A is a DFA, then L(A ) can be generated by a regular

grammar.

Proof. Suppose A = (Q, Σ, δ, q0 , F ) is a DFA. Construct a grammar G =

(N , Σ, P, S) by setting N = Q and S = q0 and

P = {A → aB | B ∈ δ(A, a)} ∪ {A → a | δ(A, a) ∈ F }.

smartworlD.asia
In addition, if the initial state q0 ∈ F , then we include S → ε in P. Clearly
G is a regular grammar. We claim that L(G) = L(A ).
From the construction of G, it is clear that ε ∈ L(A ) if and only if
ε ∈ L(G). Now, for n ≥ 1, let x = a1 a2 . . . an ∈ L(A ) be arbitrary. That is
δ̂(q0 , a1 a2 . . . an ) ∈ F . This implies, there exists a sequence of states

q1 , q 2 , . . . , q n

such that
δ(qi−1 , ai ) = qi , for 1 ≤ i ≤ n, and qn ∈ F.
As per construction of G, we have

qi−1 → ai qi ∈ P, for 1 ≤ i ≤ n − 1, and qn → an ∈ P.

Using these production rules we can derive x in G as follows:

S = q0 ⇒ a1 q1
⇒ a1 a2 q2
..
.
⇒ a1 a2 · · · an−1 qn−1
⇒ a1 a2 · · · an = x.

84

Smartzworld.com 85 jntuworldupdates.org
Smartworld.asia Specworld.in

Thus x ∈ L(G).
∗
Conversely, suppose y = b1 · · · bm ∈ L(G), for m ≥ 1, i.e. S ⇒ y in G.
Since every production rule of G is form A → aB or A → a, the derivation
∗
S ⇒ y has exactly m steps and first m − 1 steps are because of production
rules of the type A → aB and the last mth step is because of the rule of the
form A → a. Thus, in every step of the deviation one bi of y can be produced
in the sequence. Precisely, the derivation can be written as
S ⇒ b1 B1
⇒ b1 b2 B2
..
.
⇒ b1 b2 · · · bm−1 Bm−1
⇒ b1 b2 · · · bm = y.
From the construction of G, it can be observed that
δ(Bi−1 , bi ) = Bi , for 1 ≤ i ≤ m − 1, and B0 = S
in A . Moreover, δ(Bm−1 , bm ) ∈ F . Thus,

smartworlD.asia
δ̂(q0 , y) = δ̂(S, b1 · · · bm ) = δ̂(δ(S, b1 ), b2 · · · bm )
= δ̂(B1 , b2 · · · bm )
..
.
= δ̂(Bm−1 , bm )
= δ(Bm−1 , bm ) ∈ F
so that y ∈ L(A ). Hence L(A ) = L(G).
Example 4.5.11. Consider the DFA given in Example 4.5.8. Set N =
{q0 , q1 }, Σ = {a, b}, S = q0 and P has the following production rules:
q0 → aq1 | bq0 | a
q1 → aq0 | bq1 | b
Now G = (N , Σ, P, S) is a regular grammar that is equivalent to the given
DFA.
Example 4.5.12. Consider the DFA given in Example 4.5.9. The regular
grammar G = (N , Σ, P, S), where N = {q1 , q2 , q3 }, Σ = {a, b}, S = q1 and
P has the following rules
q1 → aq2 | bq1 | a | b | ε
q2 → aq3 | bq1 | b
q3 → aq3 | bq3

85

Smartzworld.com 86 jntuworldupdates.org
Smartworld.asia Specworld.in

is equivalent to the given DFA. Here, note that q3 is a trap state. So, the
production rules in which q3 is involved can safely be removed to get a simpler
but equivalent regular grammar with the following production rules.

q1 → aq2 | bq1 | a | b | ε
q2 → bq1 | b

Definition 4.5.13. A generalized finite automaton (GFA) is nondetermin-

istic finite automaton in which the transitions may be given via stings from
a finite set, instead of just via symbols. That is, formally, GFA is a sextuple
(Q, Σ, X, δ, q0 , F ), where Q, Σ, q0 , F are as usual in an NFA. Whereas, the
transition function
δ : Q × X −→ ℘(Q)
with a finite subset X of Σ∗ .

One can easily prove that the GFA is no more powerful than an NFA.
That is, the language accepted by a GFA is regular. This can be done
by converting each transition of a GFA into a transition of an NFA. For
instance, suppose there is a transition from a state p to a state q via s string

smartworlD.asia
x = a1 · · · ak , for k ≥ 2, in a GFA. Choose k − 1 new state that not already
there in the GFA, say p1 , . . . , pk−1 and replace the transition
x
p −→ q

by the following new transitions via the symbols ai ’s

a
1 2a k a
p −→ p1 , p1 −→ p2 , · · · , pk−1 −→ q

In a similar way, all the transitions via strings, of length at least 2, can be
replaced in a GFA to convert that as an NFA without disturbing its language.

Theorem 4.5.14. If L is generated by a regular grammar, then L is regular.

Proof. Let L be generated by the regular grammar G = (N , Σ, P, S). Note

that every production rule of G is either of the form A → xB or of the form
A → x, for some x ∈ Σ∗ and A, B ∈ N . We construct a GFA that accepts
L, so that L is regular.
Let X be the set of all terminal strings that are on the righthand side of
productions of G. That is,
n ¯ o
¯
X = x ¯ A → xB ∈ P or A → x ∈ P .

86

Smartzworld.com 87 jntuworldupdates.org
Smartworld.asia Specworld.in

Since P is a finite set, we have X is a finite subset of Σ∗ . Now, construct a

GFA
A = (Q, Σ, X, δ, q0 , F )
by setting Q = N ∪ {$}, where $ is a new symbol, q0 = S, F = {$} and the
transition function δ is defied by

B ∈ δ(A, x) ⇐⇒ A → xB ∈ P
and
$ ∈ δ(A, x) ⇐⇒ A → x ∈ P
We claim that L(A ) = L.
Let w ∈ L, i.e. there is a derivation for w in G. Assume the derivation
has k steps, which is obtained by the following k − 1 production rules in the
first k − 1 steps
Ai−1 → xi Ai , for 1 ≤ i ≤ k − 1, with S = A0 ,
and at the end, in the kth step, the production rule

Ak−1 → xk .

smartworlD.asia
Thus, w = x1 x2 · · · xk and the derivation is as shown below:

S = A0 ⇒ x1 A1
⇒ x1 x2 A2
..
.
⇒ x1 x2 · · · xk−1 Ak−1
⇒ x1 x2 · · · xk = w.

From the construction of A , it is clear that A has the following transitions:

(Ai−1 , xi ) |— (Ai , ε) for 1 ≤ i ≤ k,

where A0 = S and Ak = $. Using these transitions, it can be observed that

w ∈ L(A ). For instance,
(q0 , w) = (S, x1 · · · xk ) = (A0 , x1 x2 · · · xk )
|— (A1 , x2 · · · xk )
..
.
|— (Ak−1 , xk )
|— ($, ε).

87

Smartzworld.com 88 jntuworldupdates.org
Smartworld.asia Specworld.in

Thus, L ⊆ L(A ). Converse can be shown using a similar argument as in the

previous theorem. Hence L(A ) = L.

Example 4.5.15. Consider the regular grammar given in the Example 3.3.10.
Let Q = {S, A, $}, Σ = {a, b}, X = {ε, a, b, ab} and F = {$}. Set A =
(Q, Σ, X, δ, S, F ), where δ : Q × X −→ ℘(Q) is defined by the following
table.
δ ε a b ab
S ∅ {S} {S} {A}
A {$} {A} {A} ∅
$ ∅ ∅ ∅ ∅
Clearly, A is a GFA. Now, we convert the GFA A to an equivalent NFA.
Consider a new symbol B and split the production rule

S → abA

of the grammar into the following two production rules

S → aB and B → bA

smartworlD.asia
and replace them in place of the earlier one. Note that, in the resultant
grammar, the terminal strings that is occurring in the righthand sides of
production rules are of lengths at most one. Hence, in a straightforward
manner, we have the following NFA A 0 that is equivalent to the above GFA
and also equivalent to the given regular grammar. A 0 = (Q0 , Σ, δ 0 , S, F ),
where Q0 = {S, A, B, $} and δ 0 : Q0 × Σ −→ ℘(Q0 ) is defined by the following
table.
δ ε a b
S ∅ {S, B} {S}
A {$} {A} {A}
B ∅ ∅ {A}
$ ∅ ∅ ∅
Example 4.5.16. The following an equivalent NFA for the regular grammar
given in Example 3.3.13.

δ ε a b
S ∅ {A} {S}
A ∅ {B} {A}
B {$} {S} {B}
$ ∅ ∅ ∅

88

Smartzworld.com 89 jntuworldupdates.org
Smartworld.asia Specworld.in

4.6 Variants of Finite Automata

Finite state transducers, two-way finite automata and two-tape finite au-
tomata are some variants of finite automata. Finite state transducers are
introduced as devices which give output for a given input; whereas, others
are language acceptors. The well-known Moore and Mealy machines are ex-
amples of finite state transducers All these variants are no more powerful than
DFA; in fact, they are equivalent to DFA. In the following subsections, we
introduce two-way finite automata and Mealy machines. Some other variants
are described in Exercises.

4.6.1 Two-way Finite Automaton

In contrast to a DFA, the reading head of a two-way finite automata is
allowed to move both left and right directions on the input tape. In each
transition, the reading head can move one cell to its right or one cell to its
left. Formally, a two-way DFA is defined as follows.

Definition 4.6.1. A two-way DFA (2DFA) is a quintuple A = (Q, Σ, δ, q0 , F ),

smartworlD.asia
where Q, Σ, q0 and F are as in DFA, but the transition function is

δ : Q × Σ −→ Q × {L, R},

where L is indicating a left move and R is indicating a right move.

A configuration (or ID) of a 2DFA is an element of Q × Σ∗ × Σ × Σ∗

or Q × Σ∗ . A configuration (q, x, a, y) ∈ Q × Σ∗ × Σ × Σ∗ indicates that
the current state of the 2DFA is q and while the input is xay, the reading
head is scanning the symbol a. For convenience, we write (q, xay) instead of
(q, x, a, y) by denoting the position of reading head with an underscore. If
the reading head goes beyond the input, i.e. it has just finished reading the
input, then there will not be any indication of reading head in a configuration.
In which case, a configuration is of the form just (q, x) which is an element
of Q × Σ∗ ; this type of configurations will be at the end of a computation of
a 2DFA with input x.
As usual for an automaton, a computation of a 2DFA is also given by
relating two configurations C and C 0 with |−*−, i.e. a computation is of the
form C − | *− C 0 , where −
| − is the one-step relation in a 2DFA which is defined
as follows:
If δ(p, a) = (q, X), then the configuration C = (p, xay) gives the configu-
ration C 0 in one-step, denoted by C |−− C 0 , where C 0 is given by

89

Smartzworld.com 90 jntuworldupdates.org
Smartworld.asia Specworld.in

Case-I: X = R.
½
0 (q, xa), if y = ε;
C =
(q, xaby 0 ), whenever y = by 0 , for some b ∈ Σ.

Case-II: X = L.
½
0 undefined, if x = ε;
C =
(q, x0 bay 0 ), whenever x = x0 b, for some b ∈ Σ.

A string x = a1 a2 · · · an ∈ Σ∗ is said to be accepted by a 2DFA A if

(q0 , a1 a2 · · · an ) |−*− (p, a1 a2 · · · an ), for some p ∈ F . Further, the language
accepted by A is
L(A ) = {x ∈ Σ∗ | x is accepted by A }.
Example 4.6.2. Consider the language L over {0, 1} that contain all those
strings in which every occurrence of 01 is preceded by a 0, i.e. before every 01
there should be a 0. For example, 1001100010010 is in L; whereas, 101100111
is not in L. The following 2DFA given in transition table representation
accepts the language L.

smartworlD.asia
δ 0 1
→ 07q16520 43 (q1 , R) (q0 , R)
07q16251 34 (q1 , R) (q2 , L)
q2 (q3 , L) (q3 , L)
q3 (q4 , R) (q5 , R)
q4 (q6 , R) (q6 , R)
q5 (q5 , R) (q5 , R)
q6 (q0 , R) (q0 , R)
The following computation on 10011 shows that the string is accepted by the
2DFA.
(q0 , 10011) |−− (q0 , 10011) |−− (q1 , 10011) |−− (q1 , 10011)
|−− (q2 , 10011) |−− (q3 , 10011) |−− (q4 , 10011)
|−− (q6 , 10011) |−− (q0 , 10011) |−− (q0 , 10011).
Given 1101 as input to the 2DFA, as the state component the final con-
figuration of the computation
(q0 , 1101) |−− (q0 , 1101) |−− (q0 , 1101) |−− (q1 , 1101)
|−− (q2 , 1101) |−− (q3 , 1101) |−− (q5 , 1101)
|−− (q5 , 1101) |−− (q5 , 1101).
is a non-final state, we observe that the string 1101 is not accepted by the
2DFA.

90

Smartzworld.com 91 jntuworldupdates.org
Smartworld.asia Specworld.in

Example 4.6.3. Consider the language over {0, 1} that contains all those
strings with no consecutive 00 s. That is, any occurrence of two 10 s have to
be separated by at least one 0. In the following we design a 2DFA which
checks this parameter and accepts the desired language. We show the 2DFA
using a transition diagram, where the left or right move of the reading head
is indicated over the transitions.
(1,R) (1,R)
­ (0,R)
­
q0 A`
/ G@ABC
FED
?>=<
89:; q1
/ G@ABC
FED
?>=<
89:;
AA }
AA }}}
AA }
(1,R) AA }} (0,L)
}~ }
q2
GFED
@ABC
?>=<
89:;
T
(0,L)

4.6.2 Mealy Machines

As mentioned earlier, Mealy machine is a finite state transducer. In order
to give output, in addition to the components of a DFA, we have a set of
output symbols from which the transducer produces the output through an

smartworlD.asia
output function. In case of Mealy machine, the output is associated to each
transition, i.e. given an input symbol in a state, while changing to the next
state, the machine emits an output symbol. Thus, formally, a Mealy machine
is defined as follows.

Definition 4.6.4. A Mealy machine is a sextuple M = (Q, Σ, ∆, δ, λ, q0 ),

where Q, Σ, δ and q0 are as in a DFA, whereas, ∆ is a finite set called output
alphabet and
λ : Q × Σ −→ ∆
is a function called output function.

In the depiction of a DFA, in addition to its components, viz. input tape,

reading head and finite control, a Mealy machine has a writing head and an
output tape. This is shown in the Figure 4.9.
Note that the output function λ assigns an output symbol for each state
transition on an input symbol. Through the natural extension of λ to strings,
we determine the output string corresponding to each input string. The
extension can be formally given by the function

λ̂ : Q × Σ∗ −→ ∆∗

that is defined by

91

Smartzworld.com 92 jntuworldupdates.org
Smartworld.asia Specworld.in

Figure 4.9: Depiction of Mealy Machine

smartworlD.asia
1. λ̂(q, ε) = ε, and

2. λ̂(q, xa) = λ̂(q, x)λ(δ̂(q, x), a)

for all q ∈ Q, x ∈ Σ∗ and a ∈ Σ. That is, if the input string x = a1 a2 · · · an

is applied in a state q of a Mealy machine, then the output sequence

λ̂(q, x) = λ(q1 , a1 )λ(q2 , a2 ) · · · λ(qn , an )

where q1 = q and qi+1 = δ(qi , ai ), for 1 ≤ i < n. Clearly, |x| = |λ̂(q, x)|.

Definition 4.6.5. Let M = (Q, Σ, ∆, δ, λ, q0 ) be a Mealy machine. We say

λ̂(q0 , x) is the output of M for an input string x ∈ Σ∗ .

Example 4.6.6. Let Q = {q0 , q1 , q2 }, Σ = {a, b}, ∆ = {0, 1} and define the
transition function δ and the output function λ through the following tables.

δ a b λ a b
q0 q1 q0 q0 0 0
q1 q1 q2 q1 0 1
q2 q1 q0 q2 0 0

Clearly, M = (Q, Σ, ∆, δ, λ, q0 ) is a Mealy machine. By incorporating the

information of output, following the representation of DFA, M can be rep-

92

Smartzworld.com 93 jntuworldupdates.org
Smartworld.asia Specworld.in

resented by a digraph as shown below.

b/0
b/0
· x a/0 b/1
q0
/ G@ABC
FED q1
/ G@ABC
FED q2
/ G@ABC
FED
J d
a/0
a/0

For instance, output of M for the input string baababa is 0001010. In fact,
this Mealy machine prints 1 for each occurrence of ab in the input; otherwise,
it prints 0.
Example 4.6.7. In the following we construct a Mealy machine that per-
forms binary addition. Given two binary numbers a1 · · · an and b1 · · · bn (if
they are different length then we put some leading 00 s to the shorter one),
the input sequence will be considered as we consider in the manual addition,
as shown below. µ ¶µ ¶ µ ¶
an an−1 a1
···
bn bn−1 b1
Here, we reserve a1 = b1 = 0 so as to accommodate the extra bit, if any,
during addition. The expected output

smartworlD.asia
is such that
cn cn−1 · · · c1

a1 · · · an + b1 · · · bn = c1 · · · cn .
Note that there are four input symbols, viz.
µ ¶ µ ¶ µ ¶ µ ¶
0 0 1 1
, , and .
0 1 0 1
For notational convenience, let us denote the above symbols by a, b, c and d,
respectively. Now, the desired Mealy machine, while it is in the initial state,
say q0 , if the input is d, i.e. while adding 1 + 1, it emits the output 0 and
remembers the carry 1 through a new state, say q1 . For other input symbols,
viz. a, b and c, as there is no carry, it will continue in q0 and performs the
addition. Similarly, while the machine continues in q1 , for the input a, i.e.
while adding 0 + 0, it changes to the state q0 , indicating that the carry is
0 and emits 1 as output. Following this mechanism, the following Mealy
machine is designed to perform binary addition.

b/1 a/0 b/0 c/0

­ d/0
·
( GFED
q0 `@i
@ABC q1 v
/ G@ABC
FED
J @@ T
@@ a/1
c/1 @@ d/1
@

93

Smartzworld.com 94 jntuworldupdates.org
Smartworld.asia Specworld.in

Chapter 5

Properties of Regular
Languages

In the previous chapters we have introduced various tools, viz. grammars,

automata, to understand regular languages. Also, we have noted that the
class of regular languages is closed with respect to certain operations like
union, concatenation, Kleene closure. Now, with this information, can we

smartworlD.asia
determine whether a given language is regular or not? If a given language
is regular, then to prove the same we need to use regular expression, regular
grammar, finite automata or Myhill-Nerode theorem. Is there any other way
to prove that a language is regular? The answer is “Yes”. If a given lan-
guage can be obtained from some known regular languages by applying those
operations which preserve regularity, then one can ascertain that the given
language is regular. If a language is not regular, although we have Myhill-
Nerode theorem, a better and more practical tool viz. pumping lemma will
be introduced to ascertain that the language is not regular. If we were some-
how know that some languages are not regular, then again closure properties
might be helpful to establish some more languages that are not regular. Thus,
closure properties play important role not only in proving certain languages
are regular, but also in establishing non-regularity of languages. Hence, we
are indented to explore further closure properties of regular languages.

5.1 Closure Properties

5.1.1 Set Theoretic Properties
Theorem 5.1.1. The class of regular languages is closed with respect to
complement.

94

Smartzworld.com 95 jntuworldupdates.org
Smartworld.asia Specworld.in

Proof. Let L be a regular language accepted by a DFA A = (Q, Σ, δ, q0 , F ).

Construct the DFA A 0 = (Q, Σ, δ, q0 , Q − F ), that is, by interchanging the
roles of final and nonfinal states of A . We claim that L(A 0 ) = Lc so that
Lc is regular. For x ∈ Σ∗ ,
x ∈ Lc ⇐⇒ x 6∈ L
⇐⇒ δ̂(q0 , x) 6∈ F
⇐⇒ δ̂(q0 , x) ∈ Q − F
⇐⇒ x ∈ L(A 0 ).

Corollary 5.1.2. The class of regular languages is closed with respect to

intersection.
Proof. If L1 and L2 are regular, then so are Lc1 and Lc2 . Then their union
Lc1 ∪ Lc2 is also regular. Hence, (Lc1 ∪ Lc2 )c is regular. But, by De Morgan’s
law
L1 ∩ L2 = (Lc1 ∪ Lc2 )c
so that L1 ∩ L2 is regular.

smartworlD.asia
Alternative Proof by Construction. For i = 1, 2, let Ai = (Qi , Σ, δi , qi , Fi ) be
two DFA accepting Li . That is, L(A1 ) = L1 and L(A2 ) = L2 . Set the DFA
A = (Q1 × Q2 , Σ, δ, (q1 , q2 ), F1 × F2 ),
where δ is defined point-wise by
δ((p, q), a) = (δ1 (p, a), δ2 (q, a)),
for all (p, q) ∈ Q1 × Q2 and a ∈ Σ. We ³ claim´that³L(A ) = L1 ∩ L´2 . Using
induction on |x|, first observe that δ̂ (p, q), x = δ̂1 (p, x), δ̂2 (q, x) , for all
x ∈ Σ∗ .
Now it clearly follows that
³ ´
x ∈ L(A ) ⇐⇒ δ̂ (q1 , q2 ), x ∈ F1 × F2
³ ´
⇐⇒ δ̂1 (q1 , x), δ̂2 (q2 , x) ∈ F1 × F2
⇐⇒ δ̂1 (q1 , x) ∈ F1 and δ̂2 (q2 , x) ∈ F2
⇐⇒ x ∈ L1 and x ∈ L2
⇐⇒ x ∈ L1 ∩ L2 .

95

Smartzworld.com 96 jntuworldupdates.org
Smartworld.asia Specworld.in

Example 5.1.3. Using the construction given in the above proof, we design
a DFA that accepts the language

L = {x ∈ (0 + 1)∗ | |x|0 is even and |x|1 is odd}

so that L is regular. Note that the following DFA accepts the language
L1 = {x ∈ (0 + 1)∗ | |x|0 is even}.
1 1
­ ­
0
q1
/ G@ABC
FED
?>=<
89:;
g q2
/ G@ABC
FED

Also, the following DFA accepts the language L2 = {x ∈ (0+1)∗ | |x|1 is odd}.
0 0
­ ­
1
p1
/ GFED
@ABC
h
p2
/ GFED
@ABC
?>=<
89:;

Now, let s1 = (q1 , p1 ), s2 = (q1 , p2 ), s3 = (q2 , p1 ) and s4 = (q2 , p2 ) and

construct the automaton that accepts the intersection of L1 and L2 as shown

smartworlD.asia
below.
1
v
/ GFED
@ABC
s1 / GFED
@ABC
?>=<
89:;
s2
E 1 Y
0 0 0 0
² ²
GFED
@ABC 1 / GFED
@ABC
s3 h s4
1

Further, we note the following regarding the automaton. If an input x takes

the automaton (from the initial state) to the state
1. s1 , that means, x has even number of 00 s and even number of 10 s.
2. s2 , that means, x has even number of 00 s and odd number of 10 s (as
desired in the current example).
3. s3 , that means, x has odd number of 00 s and even number of 10 s.
4. s4 , that means, x has odd number of 00 s and odd number of 10 s.
By choosing any combination of states among s1 , s2 , s3 and s4 , appropri-
ately, as final states we would get DFA which accept input with appropriate
combination of 00 s and 10 s. For example, to show that the language
n ¯ o
0 ∗ ¯
L = x ∈ (0 + 1) ¯ |x|0 is even ⇔ |x|1 is odd .

96

Smartzworld.com 97 jntuworldupdates.org
Smartworld.asia Specworld.in

is regular, we choose s2 and s3 as final states and obtain the following DFA
which accepts L0 .
1
v
/ GFED
@ABC
s1 / GFED
@ABC
?>=<
89:;
s2
E 1 Y
0 0 0 0
² ²
GFED
@ABC
?>=<
89:; 1 / GFED
@ABC
s3 h s4
1

Similarly, any other combination can be considered.

Corollary 5.1.4. The class of regular languages is closed under set differ-
ence.

Proof. Since L1 − L2 = L1 ∩ Lc2 , the result follows.

Example 5.1.5. The language L = {an | n ≥ 5} is regular. We apply

Corollary 5.1.4 with L1 = L(a∗ ) and L2 = {ε, a, a2 , a3 , a4 }. Since L1 and L2
are regular, L = L1 − L2 is regular.

smartworlD.asia
Remark 5.1.6. In general, one may conclude that the removal of finitely many
strings from a regular language leaves a regular language.

5.1.2 Other Properties

Theorem 5.1.7. If L is regular, then so is LR = {xR | x ∈ L}.

To prove this we use the following lemma.

Lemma 5.1.8. For every regular language L, there exists a finite automaton
A with a single final state such that L(A ) = L.

Proof. Let A = (Q, Σ, δ, q0 , F ) be a DFA accepting L. Construct B =

(Q ∪ {p}, Σ, δ 0 , q0 , {p}), where p 6∈ Q is a new state and δ 0 is given by
½
0 δ(q, a), if q ∈ Q, a ∈ Σ
δ (q, a) =
p, if q ∈ F, a = ε.

Note that B is an NFA, which is obtained by adding a new state p to A that

is connected from all the final states of A via ε-transitions. Here p is the
only final state in B and all the final states of A are made nonfinal states.
It is easy to prove that L(A ) = L(B).

97

Smartzworld.com 98 jntuworldupdates.org
Smartworld.asia Specworld.in

Proof of the Theorem 5.1.7. Let A be a finite automaton with the initial
state q0 and single final state qf that accepts L. Construct a finite automaton
A R by reversing the arcs in A with the same labels and by interchanging
the roles of initial and final states. If x ∈ Σ∗ is accepted by A , then there is
a path q0 to qf labeled x in A . Therefore, there will be a path from qf to q0
in A R labeled xR so that xR ∈ L(A R ). Conversely, if x is accepted by A R ,
then using the similar argument one may notice that its reversal xR ∈ L(A ).
Thus, L(A R ) = LR so that LR is regular.

Example
 0  5.1.9. Consider the alphabet Σ = {a0 , a1 , . . . , a7 }, where ai =
bi
 b00i  and b0i b00i b000
i is the binary representation of decimal number i, for 0 ≤
000
bi        
0 0 0 1
i ≤ 7. That is, a0 =  0 , a1 =  0 , a2 =  1 , . . ., a7 =  1 .
0 1 0 1
Now a string x = ai1 ai2 · · · ain over Σ is said to represent correct binary
addition if
b0i1 b0i2 · · · b0in + b00i1 b00i2 · · · b00in = b000 000 000
i1 bi2 · · · bin .

smartworlD.asia
For example, the string a5 a1 a6 a5 represents correct addition, because 1011 +
0010 = 1101. Whereas, a5 a0 a6 a5 does not represent a correct addition, be-
cause 1011 + 0010 6= 1001.
We observe that the language L over Σ which contain all strings that
represent correct addition, i.e.

L = {ai1 ai2 · · · ain ∈ Σ∗ | b0i1 b0i2 · · · b0in + b00i1 b00i2 · · · b00in = b000 000 000
i1 bi2 · · · bin },

is regular. Consider the NFA shown in the following.

011 000 010 100
­ ·
* ?>=< 110
89:;
7654
0123
0J _>h 1T t
/ ?89:;
>=<
>>
101
>> 001 111
>>
>

Note that the NFA accepts LR ∪ {ε}. Hence, by Remark 5.1.6, LR is regular.
Now, by Theorem 5.1.7, L is regular, as desired.

Definition 5.1.10. Let L1 and L2 be two languages over Σ. Right quotient

of L1 by L2 , denoted by L1 /L2 , is the language

{x ∈ Σ∗ | ∃y ∈ L2 such that xy ∈ L1 }.

98

Smartzworld.com 99 jntuworldupdates.org
Smartworld.asia Specworld.in

Example 5.1.11. 1. Let L1 = {a, ab, bab, baba} and L2 = {a, ab}; then
L1 /L2 = {ε, bab, b}.

2. For L3 = 10∗ 1 and L4 = 1, we have L3 /L4 = 10∗ .

3. Let L5 = 0∗ 10∗ .

(a) L5 /0∗ = L5 .
(b) L5 /10∗ = 0∗ .
(c) L5 /1 = 0∗ .

4. If L6 = a∗ b∗ and L7 = {(an bn )a∗ | n ≥ 0}, then L6 /L7 = a∗ .

Theorem 5.1.12. If L is a regular language, then so is L/L0 , for any lan-
guage L0 .
Proof. Let L be a regular language and L0 be an arbitrary language. Suppose
A = (Q, Σ, δ, q0 , F ) is a DFA which accepts L. Set A 0 = (Q, Σ, δ, q0 , F 0 ),
where
F 0 = {q ∈ Q | δ(q, x) ∈ F, for some x ∈ L0 },

smartworlD.asia
so that A 0 is a DFA. We claim that L(A 0 ) = L/L0 . For w ∈ Σ∗ ,

w ∈ L(A 0 ) ⇐⇒ δ̂(q0 , w) ∈ F 0
⇐⇒ δ̂(q0 , wx) ∈ F, for some x ∈ L0
⇐⇒ w ∈ L/L0 .

Hence L/L0 is regular.

Note that Σ∗ is a monoid with respect to the binary operation concate-
nation. Thus, for two alphabets Σ1 and Σ2 , a mapping

h : Σ∗1 −→ Σ∗2

is a homomorphism if, for all x, y ∈ Σ∗1 ,

h(xy) = h(x)h(y).

One may notice that to give a homomorphism from Σ∗1 to Σ∗2 , it is enough to
give images for the elements of Σ1 . This is because as we are looking for a
homomorphism one can give the image of h(x) for any x = a1 a2 · · · an ∈ Σ∗1
by
h(a1 )h(a2 ) · · · h(an ).
Therefore, a homomorphism from Σ∗1 to Σ∗2 is a mapping from Σ1 to Σ∗2 .

99

Smartzworld.com 100 jntuworldupdates.org

Smartworld.asia Specworld.in

Example 5.1.13. Let Σ1 = {a, b} and Σ2 = {0, 1}. Define h : Σ1 −→ Σ∗2 by

h(a) = 10 and h(b) = 010.

Then, h is a homomorphism from Σ∗1 to Σ∗2 , which for example assigns the
image 10010010 for the string abb.
We can generalize the concept of homomorphism by substituting a lan-
guage instead of a string for symbols of the domain. Formally, a substitution
is a mapping from Σ1 to P(Σ∗2 ).
Example 5.1.14. Let Σ1 = {a, b} and Σ2 = {0, 1}. Define h : Σ1 −→
P(Σ∗2 ) by
h(a) = {0n | n ≥ 0}, say L1 ;
h(b) = {1n | n ≥ 0}, say L2 .
Then, h is a substitution. Now, for any string a1 a2 · · · an ∈ Σ∗1 , its image
under the above substitution h is

h(a1 a2 · · · an ) = h(a1 )h(a2 ) · · · h(an ),

smartworlD.asia
the concatenation of languages. For example, h(ab) is the language

L1 L2 = {0m 1n | m, n ≥ 0} = L(0∗ 1∗ ).

Given a substitution h from Σ1 to Σ2 one may naturally define h(L) for

a language L over Σ1 by [
h(L) = h(x).
x∈L

Example 5.1.15. Consider the substitution h given in Example 5.1.14 and

let L = {an bn | n ≥ 0}. For which,
[
h(L) = h(x)
x∈L
[
= h(an bn )
n≥0
n times n times
[z }| {z }| {
= h(a) · · · h(a)h(b) · · · h(b)
n≥0
n times n times
[ z }| {z }| {
= 0∗ · · · 0∗ 1∗ · · · 1∗
n≥0
= {0m 1n | m, n ≥ 0} = 0∗ 1∗ .

100

Smartzworld.com 101 jntuworldupdates.org

Smartworld.asia Specworld.in

Example 5.1.16. Define the substitution h : {a, b} −→ P({0, 1}∗ ) by

h(a) = the set of strings over {0, 1} ending with 1;

h(b) = the set of strings over {0, 1} starting with 0.

For the language L = {an bm | n, m ≥ 1}, we compute h(L) through regular

expressions described below.
Note that the regular expression for L is a+ b+ and that are for h(a) and
h(b) are (0 + 1)∗ 1 and 0(0 + 1)∗ . Now, write the regular expression that is
obtained from the expression of L by replacing each occurrence of a by the
expression of h(a) and by replacing each occurrence of b by the expression of
h(b). That is, from a+ b+ , we obtain the regular expression

((0 + 1)∗ 1)+ (0(0 + 1)∗ )+ .

This can be simplified as follows.

((0 + 1)∗ 1)+ (0(0 + 1)∗ )+ = ((0 + 1)∗ 1)∗ ((0 + 1)∗ 1)(0(0 + 1)∗ )(0(0 + 1)∗ )∗
= ((0 + 1)∗ 1)∗ (0 + 1)∗ 10(0 + 1)∗ (0(0 + 1)∗ )∗

smartworlD.asia
= (0 + 1)∗ 10(0 + 1)∗ .

The following Theorem 5.1.17 confirms that the expression obtained in this
process represents h(L). Thus, from the expression of h(L), we can conclude
that the language h(L) is the set of all strings over {0, 1} that have 10 as
substring.

Theorem 5.1.17. The class of regular languages is closed under substitu-

tions by regular languages. That is, if h is a substitution on Σ such that h(a)
is regular for each a ∈ Σ, then h(L) is regular for each regular language L
over Σ.

Proof. Let r be a regular expression for language L over Σ, i.e. L(r) = L,

and, for each a ∈ Σ, let ra be a regular expression for h(a). Suppose r0 is the
expression obtained by replacing ra for each occurrence of a (of Σ) in r so
that r0 is a regular expression. We claim that L(r0 ) = h(L(r)) so that h(L)
is regular. We prove our claim by induction on the number of operations
involved in r.
Assume that the number of operations involved in r is zero. Then there
are three possibilities for r, viz. 1. ∅, 2. ε and 3. a for some a ∈ Σ.

1. If r = ∅, then r0 = ∅ and h(L(∅)) = ∅ so that the result is straight-

forward.

101

Smartzworld.com 102 jntuworldupdates.org

Smartworld.asia Specworld.in

2. In case r = ε, r0 = ε and hence we have

h(L(r)) = h({ε}) = {ε} = L(r0 ).

3. For other case, let r = a, for some a ∈ Σ. Then, r0 = ra so that

h(L(r)) = h({a}) = L(ra ) = L(r0 ).

Hence basis of the induction is satisfied.

For inductive hypothesis, assume L(r0 ) = h(L(r)) for all those regular
expressions r which have k or fewer operations. Now consider a regular
expression r with k + 1 operations. Then

r = r1 + r2 or r = r1 r2 or r = r1∗

for some regular expressions r1 and r2 . Note that both r1 and r2 have k or
fewer operations. Hence, by inductive hypothesis, we have

L(r10 ) = h(L(r1 )) and L(r20 ) = h(L(r2 )),

smartworlD.asia
where r10 and r20 are the regular expressions which are obtained from r1 and
r2 by replacing ra for each a in r1 and r2 , respectively.
Consider the case where r = r1 + r2 . The expression r0 (that is obtained
from r) is nothing else but replacing each ra in the individual r1 and r2 , we
have
r0 = r10 + r20 .
Hence,

L(r0 ) = L(r10 + r20 )

= L(r10 ) ∪ L(r20 )
= h(L(r1 )) ∪ h(L(r2 ))
= h(L(r1 ) ∪ L(r2 ))
= h(L(r1 + r2 ))
= h(L(r))

as desired, in this case. Similarly, other two cases, viz. r = r1 r2 and r = r1∗ ,
can be handled.
Hence, the class of regular languages is closed under substitutions by
regular languages.
Corollary 5.1.18. The class of regular languages is closed under homomor-
phisms.

102

Smartzworld.com 103 jntuworldupdates.org

Smartworld.asia Specworld.in

Theorem 5.1.19. Let h : Σ1 −→ Σ∗2 be a homomorphism and L ⊆ Σ∗2 be

regular. Then, inverse homomorphic image of L,

h−1 (L) = {x ∈ Σ∗1 | h(x) ∈ L}

is regular.

Proof. Let A = (Q, Σ2 , δ, q0 , F ) be a DFA accepting L. We construct a DFA

A 0 which accepts h−1 (L). Note that, in A 0 , we have to define transitions for
the symbols of Σ1 such that A 0 accepts h−1 (L). We compose the homomor-
phism h with the transition map δ of A to define the transition map of A 0 .
Precisely, we set A 0 = (Q, Σ1 , δ 0 , q0 , F ), where the transition map

δ 0 : Q × Σ1 −→ Q

is defined by
δ 0 (q, a) = δ̂(q, h(a))
for all q ∈ Q and a ∈ Σ1 . Note that A 0 is a DFA. Now, for all x ∈ Σ∗1 , we
prove that

smartworlD.asia
δ̂ 0 (q0 , x) = δ̂(q0 , h(x)).
This gives us L(A 0 ) = h−1 (L), because, for x ∈ Σ∗1 ,

x ∈ h−1 (L) ⇐⇒ h(x) ∈ L

⇐⇒ δ̂(q0 , h(x)) ∈ F
⇐⇒ δ̂ 0 (q0 , x) ∈ F
⇐⇒ x ∈ L(A 0 ).

We prove our assertion by induction on |x|. For basis, suppose |x| = 0. That
is x = ε. Then clearly,

δ̂ 0 (q0 , x) = q0 = δ̂(q0 , ε) = δ̂(q0 , h(x)).

Here h(ε) = ε, because h is a homomorphism. Further, by definition of δ 0 ,

we have
δ̂ 0 (q0 , a) = δ 0 (q0 , a) = δ̂(q0 , h(a)),
for all a ∈ Σ∗1 , so that the assertion is true for |x| = 1 also. For inductive
hypothesis, assume
δ̂ 0 (q0 , x) = δ̂(q0 , h(x))

103

Smartzworld.com 104 jntuworldupdates.org

Smartworld.asia Specworld.in

for all x ∈ Σ∗1 with |x| = k. Let x ∈ Σ∗1 with |x| = k and a ∈ Σ1 . Now,

δ̂ 0 (q0 , xa) = δ̂ 0 (δ̂ 0 (q0 , x), a)

= δ̂ 0 (δ̂(q0 , h(x)), a) (by inductive hypothesis)
= δ 0 (δ̂(q0 , h(x)), a)
= δ̂(δ̂(q0 , h(x)), h(a)) (by definition of δ 0 )
= δ̂(q0 , h(x)h(a))
= δ̂(q0 , h(xa)). (by the property of homomorphism)

Hence the result.

Example 5.1.20. Suppose L ⊆ (0 + 1)∗ is a regular language. Now we

observe that the language

L0 = {a1 b1 · · · an bn ∈ (0 + 1)∗ | a1 · · · an ∈ L and ai = 0 iff bi = 1}

is regular. For instance, define

f : {0, 1} −→ {0, 1}∗

smartworlD.asia
by f (0) = 01 and f (1) = 10 so that f is a homomorphism. Now note that

f (L) = {f (x) | x ∈ L}
= {f (a1 · · · an ) | a1 · · · an ∈ L}
= {f (a1 ) · · · f (an ) | a1 · · · an ∈ L}
= {a1 b1 · · · an bn | a1 · · · an ∈ L and ai = 0 iff bi = 1}
= L0

Being homomorphic image of a regular language, L0 is regular.

5.2 Pumping Lemma

Theorem 5.2.1 (Pumping Lemma). If L is an infinite regular language,
then there exists a number κ (associated to L) such that for all x ∈ L with
|x| ≥ κ, x can be written as uvw satisfying the following:

1. v 6= ε, and

2. uv i w ∈ L, for all i ≥ 0.

104

Smartzworld.com 105 jntuworldupdates.org

Smartworld.asia Specworld.in

Proof. Let A = (Q, Σ, δ, q0 , F ) be a DFA accepting L. Let |Q| = κ. Since L

is infinite, there exists a string x = a1 a2 · · · an in L with n ≥ κ, where ai ∈ Σ.
Consider the accepting sequence of x, say

q0 , q 1 , . . . , q n

where, for 0 ≤ i ≤ n − 1, δ(qi , ai+1 ) = qi+1 and qn ∈ F . As there are only

κ states in Q, by pigeon-hole principle, at least one state must be repeated
in the accepting sequence of x. Let qr = qs , for 0 ≤ r < s ≤ n. Write
u = a1 · · · ar , v = ar+1 · · · as and w = as+1 · · · an ; so that x = uvw. Note
that, as r < s, we have v 6= ε. Now we will prove that uv i w ∈ L, for all
i ≥ 0.

q q
r+1 s−1
a r+1
as
a1
q q1 qr−1 qr q q an
0 ar a s+1 s+1 n−1
qn

smartworlD.asiaFigure 5.1: Pumping Sequence

For a given i ≥ 0, uv i w = a1 · · · ar (ar+1 · · · as )i as+1 · · · an . Since qr = qs ,

we see that there is a computation for uv i w in A , as given below:

(q0 , uv i w) |−*− (qr , v i w)

|−*− (qs , v i−1w )
= (qr , v i−1 w)
|−*− (qs , v i−2 w)
..
.
|−*− (qs , w)
|−*− qn

Thus, for i ≥ 0, uv i w ∈ L.
Remark 5.2.2. If L is finite, then by choosing κ = 1 + max{|x| | x ∈ L}
one may notice that L vacuously satisfies the pumping lemma, as there is no
string of length greater than or equal to κ in L. Thus, the pumping lemma
holds good for all regular languages.

105

Smartzworld.com 106 jntuworldupdates.org

Smartworld.asia Specworld.in

A Logical Formula. If we write

P (L) : L satisfies pumping lemma and
R(L) : L is regular,
then the pumping lemma for regular languages is R(L) =⇒ P (L). The state-
ment P (L) can be elaborated by the following logical formula:
h
(∀L)(∃κ)(∀x) x ∈ L and |x| ≥ κ =⇒
³ ´i
i
(∃u, v, w) x = uvw, v 6= ε =⇒ (∀i)(uv w ∈ L) .

A Tool to Ascertain Non-regularity. If a language fails to satisfy the

pumping lemma, then it cannot be regular. That is, ¬P (L) =⇒ ¬R(L) – the
contrapositive form of the pumping lemma. The ¬P (L) can be formulated
as below:
h
(∃L)(∀κ)(∃x) x ∈ L and |x| ≥ κ and
³ ´i
i
(∀u, v, w) x = uvw, v 6= ε and (∃i)(uv w ∈
/ L) .

smartworlD.asia
This statement can be better explained via the following adversarial game.
Given a language L, if we want to show that L is not regular, then we play
as given in the following steps.
1. An opponent will give us an arbitrary number κ.

2. Given κ, we pickup a string x ∈ L with |x| ≥ κ.

3. Opponent will divide x into u, v and w, arbitrarily, with v 6= ε. (Here
x = uvw.)

4. We pickup an i such that uv i w 6∈ L.

Example 5.2.3. We wish to show that L = {an bn | n ≥ 0} is not regular. An
opponent will give us a number κ. Then, we choose a string x = am bm ∈ L
with |x| ≥ κ. The opponent will give us u, v and w, where x = uvw and
v 6= ε. Now, there are three possibilities for v, viz. (1) ap , for p ≥ 1 (2) bq ,
for q ≥ 1 (3) ap bq , for p, q ≥ 1. In each case, we give an i such that uv i w 6∈ L,
as shown below.
Case-1 (v = ap , p ≥ 1). Choose i = 2. Then uv i w = am+k bm which is
clearly not in L.
(In fact, except for i = 1, for all i, uv i w 6∈ L)

106

Smartzworld.com 107 jntuworldupdates.org

Smartworld.asia Specworld.in

Case-2 (v = bq , q ≥ 1). Similar to the Case-1, one may easily observe

that except for i = 1, for all i, uv i w 6∈ L. As above, say i = 2, then
uv 2 w 6∈ L.
Case-3 (v = ap bq , p, q ≥ 1). Now, again choose i = 2. Note that

uv i w = am−p (ap bq )2 bm−q = am bq ap bm .

Since p, q ≥ 1, we observe that there is an a after a b in the resultant

string. Hence uv 2 w 6∈ L.
Hence, we conclude that L is not regular.
Example 5.2.4. We prove that L = {ww | w ∈ (0 + 1)∗ } is not regular. On
contrary, assume that L is regular. Let κ be the pumping lemma constant
associated to L. Now, choose x = 0n 1n 0n 1n from L with |x| ≥ κ. If x is
written as uvw with v 6= ε, then there are ten possibilities for v. In each
case, we observe that pumping v will result a string that is not in L. This
contradicts the pumping lemma so that L is not regular.
Case-1 (For p ≥ 1, v = 0p with x = 0k1 v0k2 1n 0n 1n ).

smartworlD.asia
Through the following points, in this case, we demonstrate a contra-
diction to pumping lemma.
1. In this case, v is in the first block of 00 s and p < n.
2. Suppose v is pumped for i = 2.
3. If the resultant string uv i w is of odd length, then clearly it is not
in L.
4. Otherwise, suppose uv i w = yz with |y| = |z|.
4n+p
5. Then, clearly, |y| = |z| = 2
= 2n + p2 .
6. Since z is the suffix of the resultant string and |z| > 2n, we have
p
z = 1 2 0n 1n .
p
7. Hence, clearly, y = 0n+p 1n− 2 6= z so that yz 6∈ L.
Using a similar argument as given in Case-1 and the arguments shown
in Example 5.2.3, one can demonstrate contradictions to pumping lemma
in each of the following remaining cases.
Case-2 (For q ≥ 1, v = 1q with x = 0n 1k1 v1k2 0n 1n ). That is, v is in the first
block of 10 s.
Case-3 (For p ≥ 1, v = 0p with x = 0n 1n 0k1 v0k2 1n ). That is, v is in the
second block of 00 s.

107

Smartzworld.com 108 jntuworldupdates.org

Smartworld.asia Specworld.in

Case-4 (For q ≥ 1, v = 1q with x = 0n 1n 0n 1k1 v1k2 ). That is, v is in the

second block of 10 s.

Case-5 (For p, q ≥ 1, v = 0p 1q with x = 0k1 v1k2 0n 1n ). That is, v is in the

first block of 0n 1n .

Case-6 (For p, q ≥ 1, v = 1p 0q with x = 0n 1k1 v0k2 1n ). That is, v is in the

block of 1n 0n .

Case-7 (For p, q ≥ 1, v = 0p 1q with x = 0n 1n 0k1 v1k2 ). That is, v is in the

second block of 0n 1n .

Case-8 (For p, q ≥ 1, v = 0p 1n 0q with x = 0k1 v0k2 1n ). That is, v is in the

block of 0n 1n 0n .

Case-9 (For p, q ≥ 1, v = 1p 0n 1q with x = 0n 1k1 v1k2 ). That is, v is in the

block of 1n 0n 1n .

Case-10 (For p, q ≥ 1, v = 0p 1n 0n 1q with x = 0k1 v1k2 ). That is, v extended

over all the blocks of 0’s and 10 s.

smartworlD.asia
Hence, L is not regular.
Remark 5.2.5. Although it is sufficient to choose a particular string to counter
the pumping lemma, it is often observed that depending on the string chosen
there can be several possibilities of partitions as uvw that are to be considered
as we have to check for all possibilities. For instance, in Example 5.2.3 we
have discussed three cases. On the other hand, in Example 5.2.4 instead of
choosing a typical string we have chosen a string which reduces the number
of possibilities to discuss. Even then, there are ten possibilities to discuss.
In the following, we show how the number of possibilities, to be con-
sidered, can be reduced further. In fact, we observe that it is sufficient to
consider the occurrence of v within the first κ symbols of the string under
consideration. More precisely, we state the assertion through the following
theorem, a restricted version of pumping lemma.
Theorem 5.2.6 (Pumping Lemma – A Restricted Version). If L is an infi-
nite regular language, then there exists a number κ (associated to L) such that
for all x ∈ L with |x| ≥ κ, x can be written as uvw satisfying the following:
1. v 6= ε,

2. |uv| ≤ κ, and

3. uv i w ∈ L, for all i ≥ 0.

108

Smartzworld.com 109 jntuworldupdates.org

Smartworld.asia Specworld.in

Proof. The proof of the theorem is exactly same as that of Theorem 5.2.1,
except that, when the pigeon-hole principle is applied to find the repetition
of states in the accepting sequence, we find the repetition of states within
the first κ + 1 states of the sequence. As |x| ≥ κ, there will be at least κ + 1
states in the sequence and since |Q| = κ, there will be repetition in the first
κ + 1 states. Hence, we have the desired extra condition |uv| ≤ κ.

Example 5.2.7. Consider that language L = {ww | w ∈ (0 + 1)∗ } given

in Example 5.2.4. Using Theorem 5.2.6, we supply an elegant proof for L
is not regular. If L is regular, then let κ be the pumping lemma constant
associated to L. Now choose the string x = 0κ 1κ 0κ 1κ from L. Using the
Theorem 5.2.6, if x is written as uvw with |uv| ≤ κ and v 6= ε then there is
only one possibility for v in x, that is a substring of first block of 00 s. Hence,
clearly, by pumping v we get a string that is not in the language so that we
can conclude L is not regular.

Example 5.2.8. We show that the language L = {x ∈ (a + b)∗ | x = xR } is

not regular. Consider the string x = 0κ 1κ 1κ 0κ from L, where κ is pumping
lemma constant associated to L. If x is written as uvw with |uv| ≤ κ and

smartworlD.asia
v 6= ε then there is only one possibility for v in x, as in the previous example.
Now, pumping v will result a string that is not a palindrome so that L is not
regular.

Example 5.2.9. We show that the language

L = {xcy | x, y ∈ {a, b}∗ and |x| = |y|}

is not regular. On contrary, assume that L is regular. Then, since regular

languages are closed with respect to intersection, L0 = L ∩ a∗ cb∗ is regular.
But, note that
L0 = {an cbn | n ≥ 0},
because xcy ∈ a∗ cb∗ and |x| = |y| implies xcy = an cbn . Now, using the
homomorphism h that is defined by

h(a) = 0, h(b) = 1, and h(c) = ε

we have
h(L0 ) = {0n 1n | n ≥ 0}.
Since regular languages are closed under homomorphisms, h(L0 ) is also reg-
ular. But we know that {0n 1n | n ≥ 0} is not regular. Thus, we arrived at a
contradiction. Hence, L is not regular.

109

Smartzworld.com 110 jntuworldupdates.org