Lecture Notes for CIS 341:

Introduction to Logic and Automata

Marvin K. Nakayama
Computer Science Department
New Jersey Institute of Technology
Newark, NJ 07102
August, 2003
c _2003
Marvin K. Nakayama
ALL RIGHTS RESERVED
Contents
1 Introduction 1-1
1.1 Purpose of Course . . . . . . . . . . . . . . . . . . . . . . . . . 1-1
1.2 Mathematical Background . . . . . . . . . . . . . . . . . . . . 1-2
2 Languages 2-1
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-1
2.2 Alphabets, Strings, and Languages . . . . . . . . . . . . . . . 2-1
2.3 Set Relations and Operations . . . . . . . . . . . . . . . . . . 2-6
2.4 Functions and Operations . . . . . . . . . . . . . . . . . . . . 2-10
2.5 Closures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-13
3 Recursive Denitions 3-1
3.1 Denition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-1
3.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-1
4 Regular Expressions 4-1
4.1 Some Denitions . . . . . . . . . . . . . . . . . . . . . . . . . 4-1
4.2 Dening Languages Using Regular Expressions . . . . . . . . . 4-2
4.3 The Language EVEN-EVEN . . . . . . . . . . . . . . . . . . . 4-8
4.4 More Examples and Denitions . . . . . . . . . . . . . . . . . 4-11
0-1
CONTENTS 0-2
5 Finite Automata 5-1
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-1
5.2 Finite Automata . . . . . . . . . . . . . . . . . . . . . . . . . 5-4
5.3 Examples of FA . . . . . . . . . . . . . . . . . . . . . . . . . . 5-8
6 Transition Graphs 6-1
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-1
6.2 Denition of Transition Graph . . . . . . . . . . . . . . . . . . 6-3
6.3 Examples of Transition Graphs . . . . . . . . . . . . . . . . . 6-5
7 Kleenes Theorem 7-1
7.1 Kleenes Theorem . . . . . . . . . . . . . . . . . . . . . . . . . 7-1
7.2 Proof of Part 1: FA TG . . . . . . . . . . . . . . . . . . . . 7-2
7.3 Proof of Part 2: TG RegExp . . . . . . . . . . . . . . . . . 7-2
7.4 Proof of Part 3: RegExp FA . . . . . . . . . . . . . . . . . 7-12
7.5 Nondeterministic Finite Automata . . . . . . . . . . . . . . . 7-27
7.6 Properties of NFA . . . . . . . . . . . . . . . . . . . . . . . . . 7-29
8 Finite Automata with Output 8-1
8.1 Moore Machines . . . . . . . . . . . . . . . . . . . . . . . . . . 8-1
8.2 Mealy Machines . . . . . . . . . . . . . . . . . . . . . . . . . . 8-2
8.3 Properties of Moore and Mealy Machines . . . . . . . . . . . . 8-5
9 Regular Languages 9-1
9.1 Properties of Regular Languages . . . . . . . . . . . . . . . . . 9-1
9.2 Complementation of Regular Languages . . . . . . . . . . . . 9-8
9.3 Intersections of Regular Languages . . . . . . . . . . . . . . . 9-10
CONTENTS 0-3
10 Nonregular Languages 10-1
10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10-1
10.2 Denition of Nonregular Languages . . . . . . . . . . . . . . . 10-3
10.3 First Version of Pumping Lemma . . . . . . . . . . . . . . . . 10-4
10.4 Another Version of Pumping Lemma . . . . . . . . . . . . . . 10-6
10.5 Prex Languages . . . . . . . . . . . . . . . . . . . . . . . . . 10-11
11 Decidability for Regular Languages 11-1
11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11-1
11.2 Decidable Problems . . . . . . . . . . . . . . . . . . . . . . . . 11-1
11.2.1 Is L
1
= L
2
? . . . . . . . . . . . . . . . . . . . . . . . . 11-2
11.2.2 Is L = ? . . . . . . . . . . . . . . . . . . . . . . . . . 11-3
11.2.3 Is L innite? . . . . . . . . . . . . . . . . . . . . . . . 11-8
12 Context-Free Grammars 12-1
12.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12-1
12.2 Context-Free Grammars . . . . . . . . . . . . . . . . . . . . . 12-5
12.3 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12-7
12.4 Trees . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12-18
13 Grammatical Format 13-1
13.1 Regular Grammars . . . . . . . . . . . . . . . . . . . . . . . . 13-1
13.2 Chomsky Normal Form . . . . . . . . . . . . . . . . . . . . . . 13-10
13.2.1 Productions and Nullable Nonterminals . . . . . . . 13-10
13.2.2 Unit Productions . . . . . . . . . . . . . . . . . . . . . 13-15
13.2.3 Chomsky Normal Form . . . . . . . . . . . . . . . . . . 13-20
13.3 Leftmost Nonterminals and Derivations . . . . . . . . . . . . . 13-23
CONTENTS 0-4
14 Pushdown Automata 14-1
14.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-1
14.2 Pushdown Automata . . . . . . . . . . . . . . . . . . . . . . . 14-1
14.3 Determinism and Nondeterminism . . . . . . . . . . . . . . . . 14-7
14.4 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14-8
14.5 Formal Denition of PDA and More Examples . . . . . . . . . 14-12
14.6 Some Properties of PDA . . . . . . . . . . . . . . . . . . . . . 14-15
15 CFG = PDA 15-1
15.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15-1
15.2 CFG PDA . . . . . . . . . . . . . . . . . . . . . . . . . . . 15-1
15.3 PDA CFG . . . . . . . . . . . . . . . . . . . . . . . . . . . 15-12
17 Context-Free Languages 17-1
17.1 Closure Under Unions . . . . . . . . . . . . . . . . . . . . . . 17-1
17.2 Closure Under Concatenations . . . . . . . . . . . . . . . . . . 17-5
17.3 Closure Under Kleene Star . . . . . . . . . . . . . . . . . . . . 17-8
17.4 Intersections . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17-9
17.5 Complementation . . . . . . . . . . . . . . . . . . . . . . . . . 17-10
18 Decidability for CFLs 18-1
18.1 Membership The CYK Algorithm . . . . . . . . . . . . . . . 18-1
19 Turing Machines 19-1
19.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19-1
19.2 Stupid TM Tricks . . . . . . . . . . . . . . . . . . . . . . . . . 19-13
23 TM Languages 23-1
CONTENTS 0-5
23.1 Recursively Enumerable Languages . . . . . . . . . . . . . . . 23-1
23.2 Church-Turing Thesis . . . . . . . . . . . . . . . . . . . . . . . 23-1
23.3 Encoding of Turing Machines . . . . . . . . . . . . . . . . . . 23-2
23.4 Non-Recursively Enumerable Language . . . . . . . . . . . . . 23-3
23.5 Universal Turing Machine . . . . . . . . . . . . . . . . . . . . 23-4
23.6 Halting Problem . . . . . . . . . . . . . . . . . . . . . . . . . 23-5
23.7 Does TM Accept ? . . . . . . . . . . . . . . . . . . . . . . . 23-7
23.8 Does TM Accept Any Words? . . . . . . . . . . . . . . . . . . 23-8
24 Review 24-1
24.1 Topics Covered . . . . . . . . . . . . . . . . . . . . . . . . . . 24-1
Preface
These lecture notes are a revision of what I used to teach my CIS 341 class
(Introduction to Logic and Automata) at NJIT in the spring semester of 1996.
The course textbook is currently Introduction to Computer Theory, Second
Edition (Wiley, 1997) by Daniel I. A. Cohen, and the development of the
material in these notes corresponds to the layout there. My notes are meant
to be a supplement (not a replacement) for the textbook in the course.
My lectures for CIS 341 in spring, 1996, were videotaped, and I tried to follow
these notes as closely as possible. However, there are a number of places
where the material in the notes does not exactly match that which is in the
video tapes. There were several reasons for this. First, students often asked
questions on material not covered in my notes. Second, I frequently made
up examples in the middle of my lectures, and so those are not in the notes.
Third, during some lectures I decided not to cover particular material in my
notes for various reasons (e.g., lack of time).
The following is a rough guideline for how the tapes correspond to the pages
in this set of lecture notes:
Tape 1: Syllabus, pages 1-1 to 2-18
Tape 2: Pages 2-17 to 3-4
Tape 3: Pages 4-1 to 4-6
Tape 4: Pages 4-6 to 4-11
Tape 5: Pages 4-11 to 5-12
Tape 6: Pages 5-12 to 6-5
Tape 7: Pages 6-5 to 7-12
0-6
CONTENTS 0-7
Tape 8: Pages 7-14 to 7-22
Tape 9: Pages 7-24 to 8-1
Tape 10: Pages 8-1 to 9-1
Tape 11: Pages 9-1 to 10-4
Tape 12: Pages 9-4 and 10-4 to 10-11
Tape 13: Pages 10-11 to 11-6
Tape 14: Pages 11-6 to 11-13, and handout on Regular Expressions in the
Real World: egrep from Floyd and Beigel, The Language of Machines.
Tape 15: Pages 11-8 to 12-14
Tape 16: Pages 12-14 to 12-18
Tape 17: Pages 12-18 to 13-5
Tape 18: Pages 13-5 to 13-15
Tape 19: Pages 13-15 to 13-20
Tape 20: Pages 13-20 to 14-8
Tape 21: Pages 14-8 to 14-14
Tape 22: Pages 14-14 to 15-2
Tape 23: Pages 15-2 to 15-22
Tape 24: Pages 15-21 to 15-32
Tape 25: Pages 15-32 to 17-5
Tape 26: Pages 17-5 to 17-9
Supplement 1: Pages 17-9 to 19-14, skipping Chapter 18
Supplement 2: Pages 23-1 to 23-9
CONTENTS 0-8
Finally, as anyone who has written a large document knows, it is virtually
impossible to eliminate all of the errors. I have proofread these notes many
times, but I am sure there are still a number of mistakes in it.
Marvin Nakayama
August, 2003
Chapter 1
Introduction
1.1 Purpose of Course
Course covers the theory of computers:
Not concerned with actual hardware and software.
More interested in abstract questions of the frontiers of capability of
computers.
More specically, what can and what cannot be done by any existing
computer or any computer ever built in the future.
We will study dierent types of theoretical machines that are mathemat-
ical models for actual physical processes.
By considering the possible inputs on which these machines can work,
we can analyze their various strengths and weaknesses.
We can then develop what we may believe to be the most powerful
machine possible.
Surprisingly, it will not be able to perform every task, even some easily
described tasks.
1-1
CHAPTER 1. INTRODUCTION 1-2
1.2 Mathematical Background
In this class, we will be seeing a number of theorems and proofs.
To be able to understand how to prove a theorem, we rst have to
understand how theorems are stated.
Many (but not all) theorems are stated as if p, then q, where p and q
are statements.
Example: If a word w has more es than os, then w has at least one e.
Example: If a word w has m as and n es in it, then the word w has at least
m+n letters in all.
Example: If x
2
= 0, then x = 0.
So what does if p, then q mean?
If a theorem stated in this form is to be true, then it means that if p is
true, then q must also be true.
Note that this does not say that if q is true, then p must also be true.
This may or may not be the case.
Example: The statement, If a word w has at least one e, then w has more
es than os is not true.
For example, consider the word exploration or Exxon.
Example: If a word w has at least m+n letters in all, then the word w has
m as and n es in it.
For example, suppose m = n = 1, and consider the word goof.
Example: If x = 0, then x
2
= 0.
So now how do we prove a result?
We do it by arguing very carefully, where each step in our argument follows
logically from the previous step.
There are several ways of proving that a statement if p, then q holds:
CHAPTER 1. INTRODUCTION 1-3
One way is to use a direct argument:
Example: Prove: If a word w has more es than os, then w has at
least one e.
Proof. Let n
e
be the number of es in w, and let n
o
be the number of
os in w. Since w has more es than os, we must have that n
e
> n
o
, or
in other words n
e
n
o
+1. But since w cannot have fewer than zero os,
we must have that n
o
0. Therefore, n
e
n
o
+1 0 +1 = 1. Thus, w
has at least one e.
Another way of proving results is by contradiction. We do this by as-
suming that p is true and that q is not true, and then showing that an
inconsistency results.
Example: Prove: If x
2
= 0, then x = 0.
Proof. Suppose that x
2
= 0 but x ,= 0. Then either x > 0 or x < 0.
But if x > 0, then x
2
> 0, and if x < 0, then x
2
> 0. In either case,
x
2
> 0. This contradicts the assumption that x
2
= 0.
Example: Prove: If x > 0 with x 1, then x
2
> 0.
Proof. Suppose that x
2
= 0. Then x = 0 so x ,> 0.
There are several equivalent ways of stating if p, then q
if not q, then not p
p only if q
q if p
p implies q
p is sucient for q
q is necessary for p
Example: Let x be a real number. If x > 0, then x
2
> 0.
This is equivalent to stating
CHAPTER 1. INTRODUCTION 1-4
If x
2
> 0 is not true (i.e., x
2
0), then x > 0 is not true (i.e., x 0).
This is also equivalent to stating x > 0 only if x
2
> 0.
This is also equivalent to stating x
2
> 0 if x > 0.
This is also equivalent to stating x > 0 implies x
2
> 0.
Often, the two statements
1. p only if q (i.e., if p, then q) and
2. p if q (i.e., if q, then p)
are combined into p if and only if q (or p is a necessary and sucient
condition for q).
In order for this statement to be true, we need to show that both statements
1 and 2 above are true.
Denition: An integer n is an even number if n = 2k for some k =
0, 1, 2, 3, . . ..
Denition: An integer n is an odd number if n = 2k + 1 for some k =
0, 1, 2, 3, . . ..
Denition: An integer n is a positive even number if n = 2k for some
k = 1, 2, 3, . . ..
Chapter 2
Languages
2.1 Introduction
In English, there are at least three dierent types of entities: letters,
words, sentences.
letters are from a nite alphabet a, b, c, . . . , z
words are made up of certain combinations of letters from the alphabet.
Not all combinations of letters lead to a valid English word.
sentences are made up of certain combinations of words. Not all combi-
nations of words lead to a valid English sentence.
So we see that some basic units are combined to make bigger units.
We want to abstract this to a dierent level.
In particular, we will be studying so-called formal languages.
2.2 Alphabets, Strings, and Languages
Denition: A set is an unordered collection of objects or elements. Sets are
written with curly braces , and the elements in the set are written within
the curly braces.
2-1
CHAPTER 2. LANGUAGES 2-2
Examples:
The set a, b, c has elements a, b, and c.
The sets a, b, c and b, c, b, a, a are the same since order does not
matter in a set and since redundancy does not count.
The set a has element a. Note that a and a are dierent things; a
is a set with one element a.
The set x
n
: n = 1, 2, 3, . . . consists of x, xx, xxx, . . ..
The set of even numbers is 0, 2, 4, 6, 8, 10, 12, . . . = 2n : n = 0, 1, 2, . . ..
In particular, note that 0 is an even number.
The set of positive even numbers is 2, 4, 6, 8, 10, 12, . . . = 2n : n =
1, 2, 3, . . ..
The set of odd numbers is 1, 3, 5, 7, 9, 11, 13, . . . = 2n + 1 : n =
0, 1, 2, . . ..
Denition: An alphabet, denoted by , is a nite set of fundamental units
(called letters) out of which we build structure.
Examples:
The alphabet of lower-case Roman letters is = a, b, c, . . . , z. (There
are 26 lower-case Roman letters.)
The alphabet of upper-case Roman letters is = A, B, C, . . . , Z. (There
are 26 upper-case Roman letters.)
The alphabet of Arabic numerals is = 0, 1, 2, . . . , 9. (There are 10
Arabic numerals.)
Denition: A string over an alphabet is a nite sequence of letters from the
alphabet.
Examples:
cat, food, c, and bbedwxq are strings over the alphabet = a, b, c, . . . , z.
CHAPTER 2. LANGUAGES 2-3
0173 is a string over the alphabet = 0, 1, 2, . . . , 9.
Denition: The empty string or null string, which we shall denote by , is
the string consisting of no letters, no matter what language were considering.
Denition: Given two strings w
1
and w
2
, we dene the concatenation of w
1
and w
2
to be the string w
1
w
2
.
Examples:
If w
1
= xx and w
2
= x, then w
1
w
2
= xxx.
If w
1
= abb and w
2
= ab, then w
1
w
2
= abbab and w
2
w
1
= ababb.
If w
1
= and w
2
= ab, then w
1
w
2
= ab.
If w
1
= bb and w
2
= , then w
1
w
2
= bb.
If w
1
= and w
2
= , then w
1
w
2
= ; i.e., = .
Denition: For any string w, we dene w
n
for n 0 inductively as follows:
w
0
= ;
w
n+1
= w
n
w for any n 0.
Example: If w = cat, then w
0
= , w
1
= cat, w
2
= catcat, w
3
= catcatcat,
and so on.
Denition: Given a string s, a substring of s is any part of the string s; i.e.,
w is a substring of s if there exist strings x and y (either or both possibly null)
such that s = xwy.
Examples:
Take the string 472828. Then , 282, 4, and 472828 are all substrings of
472828.
48 is not a substring of 472828.
CHAPTER 2. LANGUAGES 2-4
Denition: A formal language L is a set of strings over an alphabet for which
there are explicit rules for the strings in the set. Throughout these notes, we
will only consider formal languages, and so we will simplify the discussion by
saying language instead of formal language.
Examples:
Computer languages, e.g., C or C++ or Java, are formal languages with
alphabet = a, b, . . . , z, A, B, . . . , Z, , 0, 1, 2, . . . , 9, >, <, =, +, -, *,
/, (, ), ., ,, &, !, %, ^, , , |, , :, ; . The rules of syntax dene the
rules for the language.
The set of valid variable names in C++ is a formal language. What are
the alphabet and rules dening valid variable names in C++?
Denition: Those strings that are permissable in the language L are called
words of the language L.
Remarks:
A language is just a specic collection of strings.
We will use the words string and word interchangeably.
Thus, for a given string w and a particular language L, we might call w
a word even if it is not in the language L.
Let us consider some simple examples of languages:
Example: Alphabet = x.
Language
L
0
= , x, xx, xxx, xxxx, . . .
= x
n
for n = 0, 1, 2, 3, . . .
= x
n
: n = 0, 1, 2, 3 . . .
where we interpret x
n
to be the string of n xs strung together. In particular,
x
0
= . Note that
L
0
includes as a word.
CHAPTER 2. LANGUAGES 2-5
there are dierent ways we can specify a language.
Example: Alphabet = x.
Language
L
1
= x, xx, xxx, xxxx, . . .
= x
n
for n = 1 2 3 . . .
= x
n
: n = 1, 2, 3 . . .
Note that
L
1
doesnt include as a word.
there are dierent ways we can specify a language.
Example: Alphabet = x.
Language
L
2
= x, xxx, xxxxx, xxxxxxx, . . .
= x
odd

= x
2n+1
: n = 0, 1, 2, 3, . . .
Example: Alphabet = 0, 1, 2, . . . , 9.
Language
L
3
= any string of alphabet letters that does not start with the letter 0
= 1, 2, 3, . . . , 9, 10, 11, . . .
Denition: For any set S, we use the notation w S to denote that w is
an element of the set S. Also, we use the notation y , S to denote that y
is not an element of the set S.
Example: If L
1
= x
n
: n = 1, 2, 3 . . ., then x L
1
and xxx L
1
, but
, L
1
.
Denition: The set , which is called the empty set, is the set consisting of
no elements.
CHAPTER 2. LANGUAGES 2-6
Fact: Note that , since has no elements.
Example: Let = a, b, and we can dene a language L consisting of all
strings that begin with a followed by zero or more bs; i.e.,
L = a, ab, abb, abbb, . . .
= ab
n
: n = 0, 1, 2, . . ..
2.3 Set Relations and Operations
Denition: If A and B are sets, then A B (A is a subset of B) if w A
implies that w B; i.e., each element of A is also an element of B.
Examples:
Suppose A = ab, ba and B = ab, ba, aaa. Then A B, but B , A.
Suppose A = x, xx, xxx, . . . and B = , x, xx, xxx, . . .. Then A
B, but B , A.
Suppose A = ba, ab and B = aa, bb. Then A , B and B , A.
Denition: Let A and B be 2 sets. A = B if A B and B A.
Examples:
Suppose A = ab, ba and B = ab, ba. Then A B and B A, so
A = B.
Suppose A = ab, ba and B = ab, ba, aaa. Then A B, but B , A,
so A ,= B.
Suppose A = x, xx, xxx, . . . and B = x
n
: n 1. Then A B and
B A, so A = B.
Denition: Given two sets of strings S and T, we dene
S +T = w : w S or w T
CHAPTER 2. LANGUAGES 2-7
to be the union of S and T; i.e., S + T consists of all words either in S or in
T (or in both).
Examples:
Suppose S = ab, bb and T = aa, bb, a. Then S +T = ab, bb, aa, a.
Denition: Given two sets S and T of strings, we dene
S T = w : w S and w T,
which is the intersection of S and T; i.e., S T consists of strings that are in
both S and T.
Denition: Sets S and T are disjoint if S T = .
Examples:
Suppose S = ab, bb and T = aa, bb, a. Then S T = bb.
Suppose S = ab, bb and T = ab, bb. Then S T = ab, bb.
Suppose S = ab, bb and T = aa, ba, a. Then S T = , so S and T
are disjoint.
Denition: For any 2 sets S and T of strings, we dene S T = w : w
S, w , T.
Examples:
Suppose S = a, b, bb, bbb and T = a, bb, bab. Then S T = b, bbb.
Suppose S = ab, ba and T = ab, ba. Then S T = .
Denition: For any set S, we dene [S[, which is called the cardinality of S,
to be the number of elements in S.
Examples:
Suppose S = ab, bb and T = a
n
: n 1. Then [S[ = 2 and [T[ = .
CHAPTER 2. LANGUAGES 2-8
If S = , then [S[ = 0.
Denition: If S is any set, we say that S is nite if [S[ < . If S is not
nite, then we say that S is innite.
Examples:
Suppose S = ab, bb. Then S is nite.
Suppose T = a
n
: n 1. Then T is innite.
Fact: If S and T are 2 disjoint sets (i.e., S T = ), then [S +T[ = [S[ +[T[.
Fact: If S and T are any 2 sets such that [S T[ < , then
[S +T[ = [S[ +[T[ [S T[.
In particular, if S T = , then [S +T[ = [S[ +[T[.
Examples:
Suppose S = ab, bb and T = aa, bb, a. Then
S +T = ab, bb, aa, a
S T = bb
[S[ = 2
[T[ = 3
[S T[ = 1
[S +T[ = 4.
Suppose S = ab, bb and T = aa, ba, a. Then
S +T = ab, bb, aa, ba, a
S T =
[S[ = 2
[T[ = 3
[S T[ = 0
[S +T[ = 5.
CHAPTER 2. LANGUAGES 2-9
Denition: The Cartesian product (or direct or cross product) of two sets A
and B is the set A B = (x, y) : x A, y B of ordered pairs.
Examples:
If A = ab, ba, bbb and B = bb, ba, then
A B = (ab, bb), (ab, ba), (ba, bb), (ba, ba), (bbb, bb), (bbb, ba).
Note that (ab, ba) A B.
Also, note that
B A = (bb, ab), (bb, ba), (bb, bbb), (ba, ab), (ba, ba), (ba, bbb).
Note that (bb, ba) B A, but (bb, ba) , A B, so B A ,= A B.
We can also dene the Cartesian product of more than 2 sets.
Denition: The Cartesian product (or direct or cross product) of n sets
A
1
, A
2
, . . . , A
n
is the set
A
1
A
2
A
n
= (x
1
, x
2
, . . . , x
n
) : x
i
A
i
for i = 1, 2, . . . , n
of ordered n-tuples.
Examples:
Suppose
A
1
= ab, ba, bbb,
A
2
= a, bb,
A
3
= ab, b.
Then
A
1
A
2
A
3
= (ab, a, ab), (ab, a, b), (ab, bb, ab), (ab, bb, b), (ba, a, ab), (ba, a, b),
(ba, bb, ab), (ba, bb, b), (bbb, a, ab), (bbb, a, b), (bbb, bb, ab), (bbb, bb, b).
Note that (ab, a, ab) A
1
A
2
A
3
.
CHAPTER 2. LANGUAGES 2-10
Denition: If S and T are sets of strings, we dene the product set (or
concatenation) ST to be
ST = w = w
1
w
2
: w
1
S, w
2
T
Examples:
If S = a, aa, aaa and T = b, bb, then
ST = ab, abb, aab, aabb, aaab, aaabb
If S = a, ab, aba and T = , b, ba, then
ST = a, ab, aba, abb, abba, abab, ababa
If S = , a, aa and T = , bb, bbbb, bbbbbb, . . ., then
ST = , a, aa, bb, abb, aabb, bbbb, abbbb, . . .
Denition: For any set S, dene 2
S
, which is called the power set, to be the
set of all possible subsets of S; i.e., 2
S
= A : A S.
Example: If S = a, bb, ab, then
2
S
= , a, bb, ab, a, bb, a, ab, bb, ab, a, bb, ab.
Fact: If [S[ < , then [2
S
[ = 2
|S|
; i.e., there are 2
|S|
dierent subsets of S.
2.4 Functions and Operations
Denition: For any string s, the length of s is the number of letters in s.
We will sometimes denote the length of a string s by length(s) or by [s[.
Examples:
length(cat) = 3. Also, [cat[ = 3. If we dene a string s such that s = cat,
then [s[ = 3.
CHAPTER 2. LANGUAGES 2-11
[[ = 0.
Denition: A function (or operator, operation, map, or mapping) f maps
each element in a domain D into a single element in a range R. We denote
this by f : D R. Also, we say that the mapping f is dened on the domain
D and that f is an R-valued mapping. In particular, if the range R 1, i.e.,
if the range is a subset of the real numbers, then we say that f is a real-valued
mapping.
Examples:
Let 1 denote the real numbers, and let 1
+
denote the non-negative real
numbers. We can dene a function f : 1 1
+
as f(x) = x
2
.
If we dene f such that f(3) = 4 and f(3) = 8, then f is not a function
since it maps 3 to more than one value.
Let D be any collection of strings, and let R be the non-negative integers.
Then we can dene f : D R to be such that for any string s D,
f(s) = [s[,
which is the length of s.
We can dene a function f : 1 1 1 to be f(x, y) = x +y.
Let L
1
and L
2
be two sets of strings. Then we can dene the concatena-
tion operator as the function f : L
1
L
2
L
1
L
2
such that
f(w
1
, w
2
) = w
1
w
2
Language L
1
= x
n
: n 1 from before.
Can concatenate a = xxx and b = x to get ab = xxxx.
Note that a, b L
1
and that ab L
1
.
Language L
2
= x
2n+1
: n 0 from before.
Can concatenate a = xxx and b = x to get ab = xxxx.
Note that a, b L
2
but that ab , L
2
.
Denition: For a mapping f dened on a domain D, we dene
f(D) = f(x) : x D;
CHAPTER 2. LANGUAGES 2-12
i.e., f(D) is the set of all possible values that the mapping f can take on when
applied to values in D.
Example:
If f(x) = x
2
and D = 1, then f(D) = 1
+
, the set of non-negative real
numbers.
Denition: Suppose f is a mapping dened on a domain D. We say that D
is closed under mapping f if f(D) D; i.e., if x D implies that f(x) D.
In other words, D is closed under f if applying f to any element in D results
in an element in D.
Denition: Suppose f is a mapping dened on a domain D D. We say
that D is closed under mapping f if f(D, D) D; i.e., if (x, y) D D
implies that f(x, y) D.
Examples:
L
1
= x
n
: n = 1, 2, 3, . . . is closed under concatenation.
L
2
= x
2n+1
: n = 0, 1, 2, . . . is not closed under concatenation since x
concatenated with x yields xx , L
2
.
Denition: For any string w, the reverse of w, written as reverse(w) or w
R
,
is the same string of letters written in reverse order. Thus, if w = w
1
w
2
w
n
,
where each w
i
is a letter, then reverse(w) = w
n
w
n1
w
1
.
Examples:
For xxxx L
1
= x
n
: n = 1, 2, 3, . . ., reverse(xxxx) = xxxx L
1
. We
can show that L
1
is closed under reversal.
Recall L
3
is the set of strings over the alphabet = 0, 1, 2, . . . , 9 such
that the rst letter is not 0. For 48 L
3
, reverse(48) = 84 L
3
.
Example: For 90210 L
3
, reverse(90210) = 01209 , L
3
. Thus, L
3
is
not closed under reversal.
CHAPTER 2. LANGUAGES 2-13
Denition: Over the alphabet = a, b, the language PALINDROME
is dened as
PALINDROME = and all strings x such that reverse(x) = x
= , a, b, aa, bb, aaa, aba, . . .
Note that for the language PALINDROME, the words abba, a PALINDROME,
but their concatenation abbaa is not in PALINDROME.
Denition: Suppose f : D 1 and g : D 1 are real-valued mappings
such that f(x) g(x) for all x D. Then f is a bounded above by g, or g
is an upper bound for f. In addition, if there exists some x D such that
f(x) = g(x), then we say that g is a tight upper bound for f.
Examples:
If f(x) = sin(x) and g(x) = 2 for all x 1, then g is an upper bound of
f, but g is not a tight upper bound of f.
If f(x) = sin(x) and g(x) = 1 for all x 1, then g is a tight upper
bound of f.
Suppose f(x) = x and g(x) = x
2
. Then g is an upper bound for f for
all x 1, and g is tight since g(x) = f(x) for x = 1.
Suppose f(x) = x
2
and g(x) = 2
x
. Then g is an upper bound for f for
all x 4. Also, g is a tight upper bound over x 4 since g(x) = f(x)
for x = 4.
2.5 Closures
Denition: Given an alphabet , let

be the closure of the alphabet,

which is dened to be the language in which any string of letters from (with
possible repetition of letters) is a word in

, even the null string . This

notation is also known as the Kleene star. Thus,

= w = w
1
w
2
w
n
: n 0, w
i
for i = 1, 2, . . . , n,
where we dene w
1
w
2
w
n
= when n = 0.
CHAPTER 2. LANGUAGES 2-14
Example: Alphabet = x. Then, the closure of is

= , x, xx, xxx, . . .
Example: Alphabet = 0, 1, 2, . . . , 9. Then, the closure of is

= , 0, 1, 2, . . . , 9, 00, 01, 02, 03, . . .

We can think of the Kleene star as an operation that makes an innite language
(i.e., a language with innitely many words) out of an alphabet.
Denition: Given a set S of strings, we dene S
n
, n 1, to be
S
n
= SS S

n times
= w = w
1
w
2
w
n
: w
i
S, i = 1, 2, . . . , n.
Note that S
1
= S. We also dene S
0
= .
Example: If S = ab, bbb, then S
1
= S, and
S
2
= abab, abbbb, bbbab, bbbbbb
S
3
= ababab, ababbbb, abbbbab, abbbbbbb, bbbabab, bbbabbbb, bbbbbbab, bbbbbbbbb
We can also apply the star-operator to sets of words:
Denition: If S is a set of words, then S

is the set of all nite strings formed

by concatenating words from S, where any word may be used as often as we
like, and where the null string is also included; i.e.,
S

= S
0
+S
1
+S
2
+S
3
+ .
In set notation,
S

= w = w
1
w
2
w
3
w
n
: n 0 and w
i
S for all i = 1, 2, 3, . . . , n,
where we interpret w
1
w
2
w
3
w
n
for n = 0 to be the null string . Thus,
S
0
= for any set S. In particular, if S = , we still have S
0
= .
Example: If S = ba, a, then
S

= plus any word composed of factors of ba and a

= , a, aa, ba, aaa, aba, baa, aaaa, aaba, . . ..
CHAPTER 2. LANGUAGES 2-15
If w S

, can bb ever be a substring of w? No.

Proof.
Suppose xy is a substring of length 2 of w, where x and y are single
letters.
Since w S

, we can write w = w
1
w
2
w
n
, for some n 0, where each
w
i
S, i = 1, 2, . . . , n.
Since S = ba, a, there are ve possibilities for how the 2-letter sub-
string xy could have arisen:
1. xy is the concatenation of two 1-letter words from S; i.e., for some
i = 1, 2, . . . , n1, we have that xy = w
i
w
i+1
, where w
i
and w
i+1
are
words from S having only one letter each. Since the only 1-letter
word from S is a, we must have that w
i
= w
i+1
= a. In this case,
xy = aa, which is not bb.
2. xy is a 2-letter word from S; i.e., for some i = 1, 2, . . . , n, we have
that xy = w
i
, where w
i
is a 2-letter word from S. Since the only
2-letter word from S is ba, we must have that w
i
= ba. In this case,
xy = ba, which is not bb.
3. xy is the concatenation of a 1-letter word from S and the rst letter
of a 2-letter word from S; i.e., for some i = 1, 2, . . . , n 1, we have
that xy = w
i
w
i+1,1
, where
w
i
is a 1-letter word from S.
w
i+1
is a 2-letter word of S with w
i+1
= w
i+1,1
w
i+1,2
and w
i+1,1
and w
i+1,2
are the two letters of w
i+1
.
Since the only 1-letter word from S is a, we must have that w
i
= a.
Since the only 2-letter word from S is ba, we must have that w
i+1
=
ba, whose rst letter is b. In this case, xy = ab, which is not bb.
4. xy is the concatenation of the second letter of a 2-letter word from
S and a 1-letter word from S; i.e., for some i = 1, 2, . . . , n 1, we
have that xy = w
i,2
w
i+1
, where
w
i
is a 2-letter word of S with w
i
= w
i,1
w
i,2
and w
i,1
and w
i,2
are the two letters of w
i
.
w
i+1
is a 1-letter word from S.
Since the only 2-letter word from S is ba, we must have that w
i
= ba,
whose second letter is a. Since the only 1-letter word from S is a,
we must have that w
i+1
= a. In this case, xy = aa, which is not bb.
CHAPTER 2. LANGUAGES 2-16
5. xy is the concatenation of the second letter of a 2-letter word from
S and the rst letter of a 2-letter word from S; i.e., for some i =
1, 2, . . . , n 1, we have that xy = w
i,2
w
i+1,1
, where
w
i
is a 2-letter word of S with w
i
= w
i,1
w
i,2
and w
i,1
and w
i,2
are the two letters of w
i
.
w
i+1
is a 2-letter word of S with w
i+1
= w
i+1,1
w
i+1,2
and w
i+1,1
and w
i+1,2
are the two letters of w
i+1
.
Since the only 2-letter word from S is ba, we must have that w
i
=
w
i+1
= ba, whose rst letter is b and whose second letter is a. In
this case, xy = ab, which is not bb.
This exhausts all of the possibilities for how a 2-letter substring xy can
arise in this example. Since all of them result in xy ,= bb, we have
completed the proof.
Example: If S = xx, xxx, then
S

= and all strings of more than one x

= , xx, xxx, xxxx, . . .
To prove that a certain word is in the closure language S

, we must show how

it can be written as a concatenation of words in S.
Example: If S = ba, a, then aaba S

since we can break aaba into the

factors a S, a S, and ba S; i.e., aaba = (a)(a)(ba).
Note that there is only one way to do the above factorization into words from
S; we then say the factorization is unique.
Example: If S = xx, xxx, then xxxxxx S

since xxxxxx = (xx)(xx)(xx) =

(xxx)(xxx).
Here, the factorization is not unique.
Example: If S = , then S

= .
Example: If S = , then S

= .
Remarks:
CHAPTER 2. LANGUAGES 2-17
Two words are considered the same if all their letters are the same and
in the same order, so there is only one possible word of no letters, .
There is an important dierence between the word that has no letters
and the language that has no words, which we denote by .
It is not true that is a word in the language since doesnt have any
words at all.
If a language L does not contain the word and we wish to add it to L,
we use the union of sets operation denoted by + to form L +.
Note that L ,= L + if , L.
Note that L = L +.
Denition: If S is some set of words, then S
+
= S
1
+ S
2
+ S
3
+ , which
is the set of all nite strings formed by concatenating some positive number
of strings from S.
Example: If = x, then
+
= x, xx, xxx, . . ..
Denition: If A and B are sets, then A B (A is a subset of B) if w A
implies that w B; i.e., each element of A is also an element of B.
Suppose that we have two sets A and B, and we want to prove that A = B.
One way of proving this is to show that
1. A B, and
2. B A.
Example: Suppose A = x, xx and B = x, xx, xxx. Note that A B,
but B , A, and so A ,= B.
Theorem 1 For any set S of strings, we have that S

= S

.
Proof. The way we will prove this is by showing two things:
1. S

CHAPTER 2. LANGUAGES 2-18

2. S

To show part 1, we have to prove that any word w

0
in S

is also in S

.
Note that since w
0
S

, w
0
is made up of factors, say w
1
, w
2
, . . . , w
k
,
k 0, from S

; i.e., w
0
= w
1
w
2
w
k
, with k 0 and w
i
S

for
i = 1, 2, . . . , k.
Also, each factor w
i
, i = 1, 2, . . . , k, is from S

, and so it is made up of
a nonnegative number of factors from S; i.e., w
i
= w
i,1
w
i,2
w
i,n
i
, with
n
i
0 and w
i,j
S for j = 1, 2, . . . , n
i
.
Therefore, we can write
w
0
= w
1
w
2
w
k
= w
1,1
w
1,2
w
1,n
1
w
2,1
w
2,2
w
2,n
2
w
k,1
w
k,2
w
k,n
k
,
where each w
i,j
S, i = 1, 2, . . . , k, j = 1, 2, . . . , n
i
. So the original word
w
0
S

is made up of factors from S.

But S

is just the language made up of the dierent factors in S.

Therefore, w
0
S

.
Since w
0
was arbitrary, we have just shown that every word in S

is also
a word in S

; i.e., S

.
To show part 2, note that in general, for any set A, we know that A A

.
Hence, letting A = S

, we see that S

.
Chapter 3
Recursive Denitions
3.1 Denition
A recursive denition is characteristically a three-step process:
1. First, we specify some basic objects in the set. The number of basic
objects specied must be nite.
2. Second, we give a nite number of rules for constructing more objects in
the set from the ones we already know.
3. Third, we declare that no objects except those constructed in this way
are allowed in the set.
3.2 Examples
Example: Consider the set P-EVEN, which is the set of positive even num-
bers.
We can dene the set P-EVEN in several dierent ways:
We can dene P-EVEN to be the set of all positive integers that are
evenly divisible by 2.
P-EVEN is the set of all 2n, where n = 1, 2, . . ..
3-1
CHAPTER 3. RECURSIVE DEFINITIONS 3-2
P-EVEN is dened by these three rules:
Rule 1 2 is in P-EVEN.
Rule 2 If x is in P-EVEN, then so is x + 2.
Rule 3 The only elements in the set P-EVEN are those that can be
produced from the two rules above.
Note that the rst two denitions of P-EVEN are much easier to apply than
the last.
In particular, to show that 12 is in P-EVEN using the last denition, we would
have to do the following:
1. 2 is in P-EVEN by Rule 1.
2. 2 + 2 = 4 is in P-EVEN by Rule 2.
3. 4 + 2 = 6 is in P-EVEN by Rule 2.
4. 6 + 2 = 8 is in P-EVEN by Rule 2.
5. 8 + 2 = 10 is in P-EVEN by Rule 2.
6. 10 + 2 = 12 is in P-EVEN by Rule 2.
We can make another denition for P-EVEN as follows:
Rule 1 2 is in P-EVEN.
Rule 2 If x and y are both in P-EVEN, then x +y is in P-EVEN.
Rule 3 No number is in P-EVEN unless it can be produced by rules 1 and 2.
Can use the new denition of P-EVEN to show that 12 is in P-EVEN:
1. 2 is in P-EVEN by Rule 1.
2. 2 + 2 = 4 is in P-EVEN by Rule 2.
3. 4 + 4 = 8 is in P-EVEN by Rule 2.
4. 4 + 8 = 12 is in P-EVEN by Rule 2.
CHAPTER 3. RECURSIVE DEFINITIONS 3-3
Example: Let PALINDROME be the set of all strings over the alphabet =
a, b that are the same spelled forward as backwards; i.e., PALINDROME
= w : w = reverse(w) = , a, b, aa, bb, aaa, aba, bab, bbb, aaaa, abba, . . ..
A recursive denition for PALINDROME is as follows:
Rule 1 , a, and b are in PALINDROME.
Rule 2 If w PALINDROME, then so are awa and bwb.
Rule 3 No other string is in PALINDROME unless it can be produced by
rules 1 and 2.
Example: Let us now dene a set AE of certain valid arithmetic expressions.
The set AE will not include all possible arithmetic expressions.
The alphabet of AE is
= 0 1 2 3 4 5 6 7 8 9 + / ( )
We recursively dene AE using the following rules:
Rule 1 Any number (positive, negative, or zero) is in AE.
Rule 2 If x is in AE, then so are (x) and (x).
Rule 3 If x and y are in AE, then so are
(i) x +y (if the rst symbol in y is not )
(ii) x y (if the rst symbol in y is not )
(iii) x y
(iv) x/y
(v) x y (our notation for exponentiation)
Rule 4 AE consists of only those things can be created by the above three
rules.
For example,
(5 (8 + 2))
CHAPTER 3. RECURSIVE DEFINITIONS 3-4
and
5 (8 + 1)/3
are in AE since they can be generated using the above denition.
However,
((6 + 7)/9
and
4(/9 4)
are not since they cannot be generated using the above denition.
Now we can use our recursive denition of AE to show that
8 6 ((4/2) + (3 1) 7)/4
is in AE.
1. Each of the numbers are in AE by Rule 1.
2. 8 6 is in AE by Rule 3(iii).
3. 4/2 is in AE by Rule 3(iv).
4. (4/2) is in AE by Rule 2.
5. 3 1 is in AE by Rule 3(ii).
6. (3 1) is in AE by Rule 2.
7. (3 1) 7 is in AE by Rule 3(iii).
8. (4/2) + (3 1) 7 is in AE by Rule 3(i).
9. ((4/2) + (3 1) 7) is in AE by Rule 2.
10. ((4/2) + (3 1) 7)/4 is in AE by Rule 3(iv).
11. 8 6 + ((4/2) + (3 1) 7)/4 is in AE by Rule 3(i).
Chapter 4
Regular Expressions
4.1 Some Denitions
Denition: If S and T are sets of strings of letters (whether they are nite
or innite sets), we dene the product set of strings of letters to be
ST = w = w
1
w
2
: w
1
S, w
2
T
Example: If S = a, aa, aaa and T = b, bb, then
ST = ab, abb, aab, aabb, aaab, aaabb
Example: If S = a, ab, aba and T = , b, ba, then
ST = a, ab, aba, abb, abba, abab, ababa
Example: If S = , a, aa and T = , bb, bbbb, bbbbbb, . . ., then
ST = , a, aa, bb, abb, aabb, bbbb, abbbb, . . .
Denition: Let s and t be strings. Then s is a substring of t if there exist
strings u and v such that t = usv.
4-1
CHAPTER 4. REGULAR EXPRESSIONS 4-2
Example: Suppose s = aba and t = aababb.
Then s is a substring of t since we can dene u = a and v = bb, and then
t = usv.
Example: Suppose s = abb and t = aaabb.
Then s is a substring of t since we can dene u = aa and v = , and then
t = usv.
Example: Suppose s = bb and t = aababa.
Then s is not a substring of t.
Denition: Over the alphabet = a, b, a string contains a double letter
if it has either aa or bb as a substring.
Example: Over the alphabet = a, b,
1. The string abaabab contains a double letter.
2. The string bb contains a double letter.
3. The string aba does not contain a double letter.
4. The string abbba contains two double letters.
4.2 Dening Languages Using Regular Expres-
sions
Previously, we dened the languages:
L
1
= x
n
for n = 1, 2, 3, . . .
L
2
= x, xxx, xxxxx, . . .
But these are not very precise ways of dening languages.
So we now want to be very precise about how we dene languages, and
we will do this using regular expressions.
CHAPTER 4. REGULAR EXPRESSIONS 4-3
Languages that are associated with these regular expressions are called
regular languages and are also said to be dened by a nite representa-
tion.
Regular expressions are written in bold face letters and are a way of
specifying the language.
Recall that we previously saw that for sets S, T, we dened the operations
S +T = w : w S or w T
ST = w = w
1
w
2
: w
1
S, w
2
T
S

= S
0
+S
1
+S
2
+
S
+
= S
1
+S
2
+
We will precisely dene what a regular expression is later. But for now,
lets work with the following sketchy description of a regular expression.
Loosely speaking, a regular expression is a way of specifying a language
in which the only operations allowed are
union (+),
concatenation (or product),
Kleene- closure,
superscript-+.
The allowable symbols are parentheses, , and , as well as each letter
in written in boldface. No other symbols are allowed in a regular ex-
pression. Also, a regular expression must only consist of a nite number
of symbols.
To introduce regular expressions, think of
x = x;
i.e., x represents the language (i.e., set) consisting of exactly one string,
x. Also, think of
a = a,
b = b,
so a is the language consisting of exactly one string a, and b is the
language consisting of exactly one string b.
CHAPTER 4. REGULAR EXPRESSIONS 4-4
Using this interpretation, we can interpret ab to mean
ab = ab = ab
since the concatenation (or product) of the two languages a and b
is the language ab.
We can also interpret a +b to mean
a +b = a +b = a, b
We can also interpret a

to mean
a

= a

= , a, aa, aaa, . . .
We can also interpret a
+
to mean
a
+
= a
+
= a, aa, aaa, . . .
Also, we have
(ab +a)

b = (ab +a)

b = ab, a

b
Example: Previously, we saw language
L
4
= , x, xx, xxx, . . .
= x

= language(x

)
Example: Language
L
1
= x, xx, xxx, xxxx, . . .
= language(xx

)
= language(x

x)
= language(x
+
)
= language(x

xx

)
= language(x

x
+
)
Note that there are several dierent regular expressions associated with L
1
.
CHAPTER 4. REGULAR EXPRESSIONS 4-5
Example: alphabet = a, b
language L of all words of the form one a followed by some number (possibly
zero) of bs.
L = language(ab

)
Example: alphabet = a, b
language L of all words of the form some positive number of as followed by
exactly one b.
L = language(aa

b)
Example: alphabet = a, b
language
L = language(ab

a),
which is the set of all strings of as and bs that have at least two letters, that
begin and end with one a, and that have nothing but bs inside (if anything at
all).
L = aa, aba, abba, abbba, . . .
Example: alphabet = a, b
The language L consisting of all possible words over the alphabet has the
following regular expression:
(a +b)

Other regular expressions for L include (a

and ( +a +b)

.
Example: alphabet = x
language L with an even number (possibly zero) of xs
L = , xx, xxxx, xxxxxx, . . .
= language((xx)

)
Example: alphabet = x
language L with a positive even number of xs
L = xx, xxxx, xxxxxx, . . .
= language(xx(xx)

)
= language((xx)
+
)
CHAPTER 4. REGULAR EXPRESSIONS 4-6
Example: alphabet = x
language L with an odd number of xs
L = x, xxx, xxxxx, . . .
= language(x(xx)

)
= language((xx)

x)
Is L = language(x

xx

) ?
No, since it includes the word (xx)x(x).
Example: alphabet = a, b
language L of all three-letter words starting with b
L = baa, bab, bba, bbb
= language(b(a +b)(a +b))
= language(baa +bab +bba +bbb)
Example: alphabet = a, b
language L of all words starting with a and ending with b
L = ab, aab, abb, aaab, aabb, abab, abbb, . . .
= language(a(a +b)

b)
Example: alphabet = a, b
language L of all words starting and ending with b
L = b, bb, bab, bbb, baab, babb, bbab, bbbb, . . .
= language(b +b(a +b)

b)
Example: alphabet = a, b
language L of all words with exactly two bs
L = language(a

ba

ba

)
Example: alphabet = a, b
language L of all words with at least two bs
L = language((a +b)

b(a +b)

b(a +b)

)
CHAPTER 4. REGULAR EXPRESSIONS 4-7
Note that bbaaba L since
bbaaba = ()b()b(aaba) = (b)b(aa)b(a)
Example: alphabet = a, b
language L of all words with at least two bs
L = language(a

ba

b(a +b)

)
Note that bbaaba L since bbaaba = b b aaba
Example: alphabet = a, b
language L of all words with at least one a and at least one b
L = language((a +b)

a(a +b)

b(a +b)

+ (a +b)

b(a +b)

a(a +b)

)
= language((a +b)

a(a +b)

b(a +b)

+bb

aa

)
where
the rst regular expression comes from separately considering the two
cases:
1. requiring an a before a b,
2. requiring a b before an a.
the second expression comes from the observation that the rst term
in the rst expression only omits words that are of the form some bs
followed by some as.
Example: alphabet = a, b
language L consists of and all strings that are either all as or b followed by
a nonnegative number of as
L = language(a

+ba

)
= language(( +b)a

)
Theorem 5 If L is a nite language, then L can be dened by a regular
expression.
CHAPTER 4. REGULAR EXPRESSIONS 4-8
Proof. To make a regular expression that denes the language L, turn all
the words in L into boldface type and put pluses between them.
Example: language
L = aba, abba, bbaab
Then a regular expression to dene L is
aba +abba +bbaab
4.3 The Language EVEN-EVEN
Example: Consider the regular expression
E = [aa +bb + (ab +ba)(aa +bb)

(ab +ba)]

.
We now prove that the regular expression E generates the language EVEN-
EVEN, which consists exactly of all strings that have an even number of as
and an even number of bs; i.e.,
EVEN-EVEN = , aa, bb, aabb, abab, abba, baab, baba, bbaa, aaaabb, . . ..
Proof.
Let L
1
be the language generated by the regular expression E.
Let L
2
be the language EVEN-EVEN.
So we need to prove that L
1
= L
2
, which we will do by showing that
L
1
L
2
and L
2
L
1
.
First note that any word generated by E is made up of syllables of
three types:
type
1
= aa
type
2
= bb
type
3
= (ab +ba)(aa +bb)

(ab +ba)
E = [type
1
+ type
2
+ type
3
]

CHAPTER 4. REGULAR EXPRESSIONS 4-9

We rst show that L
1
L
2
:
Consider any string w L
1
; i.e., w can be generated by the regular
expression E.
We need to show that w L
2
.
Note that since w can be generated by the regular expression E,
the string w must be made up of syllables of type 1, 2, or 3.
Each of these types of syllables generate an even number of as and
an even number of bs.
type
1
syllable generates 2 as and 0 bs.
type
2
syllable generates 0 as and 2 bs.
type
3
syllable (ab +ba)(aa +bb)

(ab +ba) generates

exactly 1 a and 1 b at the beginning,
exactly 1 a and 1 b at the end,
and generates either 2 as or 2 bs at a time in the middle.
Thus, the type
3
syllable generates an even number of as
and an even number of bs.
Thus, the total string must have an even number of as and an even
number of bs.
Therefore, w EVEN-EVEN, so we can conclude that L
1
L
2
.
Now we want to show that L
2
L
1
; i.e., we want to show that any word
with an even number of as and an even number of bs can be generated
by E.
Consider any string w = w
1
w
2
w
3
w
n
with an even number of as
and an even number of bs.
If w = , then iterate the outer star of the regular expression E
zero times to generate .
Now assume that w ,= .
Let n = length(w).
Note that n is even since w consists solely of as and bs and since
the number of as is even and the number of bs is even.
Thus, we can read in the string w two letters at a time from left to
right.
CHAPTER 4. REGULAR EXPRESSIONS 4-10
Use the following algorithm to generate w = w
1
w
2
w
3
w
n
using
the regular expression E:
1. Let i = 1.
2. Do the following while i n:
(a) If w
i
= a and w
i+1
= a, then iterate the outer star of E
and use the type
1
syllable aa.
(b) If w
i
= b and w
i+1
= b, then iterate the outer star of E and
use the type
2
syllable bb.
(c) If (w
i
= a and w
i+1
= b) or if (w
i
= b and w
i+1
= a), then
choose the type
3
syllable (ab + ba)(aa + bb)

(ab + ba),
and do the following:
If (w
i
= a and w
i+1
= b), then choose ab in the rst part
of the type
3
syllable.
If (w
i
= b and w
i+1
= a), then choose ba in the rst part
of the type
3
syllable.
Do the following while either (w
i+2
= a and w
i+3
= a) or
(w
i+2
= b and w
i+3
= b):
Let i = i + 2.
If w
i
= a and w
i+1
= a, then iterate the inner star of
the type
3
syllable, and use aa.
If w
i
= b and w
i+1
= b, then iterate the inner star of
the type
3
syllable, and use bb.
Let i = i + 2.
If (w
i
= a and w
i+1
= b), then choose ab in the last part
of the type
3
syllable.
If (w
i
= b and w
i+1
= a), then choose ba in the last part
of the type
3
syllable.
Remarks:
We must eventually read in either ab or ba, which bal-
ances out the previous unbalanced pair. This com-
pletes a syllable of type
3
.
If we never read in the second unbalanced pair, then
either the number of as is odd or the number of bs is
odd, which is a contradiction.
(d) Let i = i + 2.
CHAPTER 4. REGULAR EXPRESSIONS 4-11
This algorithm shows how to use the regular expression E to gen-
erate any string in EVEN-EVEN; i.e., if w EVEN-EVEN, then
we can use the above algorithm to generate w using E.
Thus, L
2
L
1
.
4.4 More Examples and Denitions
Example: b

(abb

( + a) generates the language of all words without a

double a.
Example: What is a regular expression for all valid variable names in C?
Denition: The set of regular expressions is dened by the following:
Rule 1 Every letter of can be made into a regular expression by writing it
in boldface; and are regular expressions.
Rule 2 If r
1
and r
2
are regular expressions, then so are
1. (r
1
)
2. r
1
r
2
3. r
1
+r
2
4. r

1
and r
+
1
Rule 3 Nothing else is a regular expression.
Denition: For a regular expression r, let L(r) denote the language generated
by (or associated with) r; i.e., L(r) is the set of strings that can be generated
by r.
Denition: The following rules dene the language associated with (or gen-
erated by) any regular expression:
Rule 1 (i) If , then L( ) = ; i.e., the language associated with the
regular expression that is just a single letter is that one-letter word
alone.
CHAPTER 4. REGULAR EXPRESSIONS 4-12
(ii) L() = ; i.e., the language associated with is , a one-word
language.
(iii) L( ) = ; i.e., the language associated with is , the language
with no words.
Rule 2 If r
1
is a regular expression associated with the language L
1
and r
2
is
a regular expression associated with the language L
2
, then
(i) The regular expression (r
1
)(r
2
) is associated with the language L
1
concatenated with L
2
:
language(r
1
r
2
) = L
1
L
2
.
We dene L
1
= L
1
= .
(ii) The regular expression r
1
+r
2
is associated with the language formed
by the union of the sets L
1
and L
2
:
language(r
1
+r
2
) = L
1
+L
2
(iii) The language associated with the regular expression (r
1
)

is L

1
, the
Kleene closure of the set L
1
as a set of words:
language(r

1
) = L

1
(iv) The language associated with the regular expression (r
1
)
+
is L
+
1
:
language(r
+
1
) = L
+
1
Chapter 5
Finite Automata
5.1 Introduction
Modern computers are often viewed as having three main components:
1. the central processing unit (CPU)
2. memory
3. input-output devices (IO)
The CPU is the thinker
1. Responsible for such things as individual arithmetic computations
and logical decisions based on particular data items.
2. However, the amount of data the unit can handle at any one time
is xed forever by its design.
3. To deal with more than this predetermined, limited amount of in-
formation, it must ship data back and forth, over time, to and from
the memory and IO devices.
Memory
1. The memory may in practice be of several dierent kinds, such as
magnetic core, semiconductor, disks, and tapes.
2. The common feature is that the information capacity of the mem-
ory is vastly greater than what can be accommodated, at any one
instant of time, in the CPU.
5-1
CHAPTER 5. FINITE AUTOMATA 5-2
3. Therefore, this memory is sometimes called auxilliary, to distinguish
it from the limited storage that is part of the CPU.
4. At least in theory, the memory can be expanded without limit, by
adding more core boxes, more tape drives, etc.
IO devices are the means by which information is communicated back
and forth to the outside world; e.g.,
1. terminals
2. printers
3. tapes
We now will study a severely restricted model of an actual computer called a
nite automaton (FA).
Like a real computer, it has a central processor with xed nite capacity,
depending on its original design.
Unlike a real computer, it has no auxiliary memory at all.
It receives its input as a string of characters.
It delivers no output at all, except an indication of whether the input is
considered acceptable.
It is a language-recognition device.
Why should we study such a simple model of computer with no memory?
Actually, nite automata do have memory, but the amount they have is
xed and cannot be expanded.
Finite automata are applicable to the design of several common types of
computer algorithms and programs.
For example, the lexical analysis phase of a compiler is often based
on the simulation of a nite automaton.
The problem of nding an occurrence of one string within another
for example, a particular word within a large text le can
also be solved eciently by methods originating from the theory of
nite automata.
CHAPTER 5. FINITE AUTOMATA 5-3
To introduce nite automata, consider the following scenario:
Play a board game in which two players move pieces around dierent
squares.
Throw dice to determine where to move.
Players have no choices to make when making their move. The move is
completely determined by the dice.
A player wins if after 10 throws of the dice, his piece ends up on a certain
square.
Note that no skill or choice is involved in the game.
Each possible position of pieces on the board is called a state.
Every time the dice are thrown, the state changes according to what
came up on the dice.
We call the winning square a nal state (also known as a halting state,
terminal state, or accepting state).
There may be more than one nal state.
Lets look at another simple example
Suppose you have a simple computer (machine), as described above.
Your goal is to write a program to compute 3 + 4.
The program is a sequence of instructions that are fed into the computer
one at a time.
Each instruction is executed as soon as it is read, and then the next
instruction is read.
If the program is correct, then the computer outputs the number 7 and
terminates execution.
We can think of taking a snapshot of the internals (i.e., contents of
memory, etc.) of the computer after every instruction is executed.
CHAPTER 5. FINITE AUTOMATA 5-4
Each possible conguration of 0s and 1s in the cells of memory repre-
sents a dierent state of the system.
We say the machine ends in a nal state (also called a halting, terminal,
or accepting state) if when the program nishes executing, it outputs
the number 7.
Two machines are in the same state if their output pages look the same
and their memories look the same cell by cell.
The computer is deterministic, i.e., on reading one particular input in-
struction, the machine converts itself from one given state to some par-
ticular other state (which is possibly the same), where the resultant state
is completely determined by the prior state and the input instruction.
No choice is involved.
The success of the program (i.e., it outputs 7) is completely determined
by the sequence of inputs (i.e., the lines of code).
We can think of the set of all computer instructions as the letters of an
alphabet.
We can then dene a language to be the set of all words over this alphabet
that lead to success.
This is the language with words that are all programs that print a 7.
5.2 Finite Automata
Denition: A nite automaton (FA), also known as a nite acceptor, is a
collection M = (K, , , s, F) where :
1. K is a nite set of states.
Exactly one state s K is designated as the initial state (or start
state).
Some set F K is the set of nal states, where we allow F = or
F = K or F could be any other subset of K.
CHAPTER 5. FINITE AUTOMATA 5-5
2. An alphabet of possible input letters, from which are formed strings,
that are to be read one letter at a time.
3. : K K is the transition function.
In other words, for each state and for each letter of the input al-
phabet, the function tells which (one) state to go to next; i.e., if
x K and , then (x, ) is the state that you go to when you
are in state x and read in .
For each state x and each letter , there is exactly one arc
leaving x labeled with .
Thus, there is no choice in how to process a string, and so the
machine is deterministic.
An FA works as follows:
It is presented with an input string of letters.
It starts in the start state.
It reads the string one letter at a time, starting from the left.
The letters read in determine a sequence of states visited.
Processing ends after the last input letter has been read.
If after reading the entire input string the machine ends up in a nal
state, then the input string is accepted. Otherwise, the input string is
rejected.
Example: Consider an FA with three states (x, y, and z) with input alphabet
= a, b.
Dene the following transition table for the FA:
a b
start x y z
y x z
nal z z z
Input the string aaaa to the FA:
CHAPTER 5. FINITE AUTOMATA 5-6
Start in state x and read in rst a, which takes us to state y.
From state y, read in second a, which takes us to state x.
From state x, read in third a, which takes us to state y.
From state y, read in fourth a, which takes us to state x.
No more letters in input string so stop.
Note that on input aaaa,
We ended up in state x, which is not a nal state.
we say that aaaa is not accepted or rejected by this FA.
Now consider the input string abab:
Start in state x and read in rst a, which takes us to state y.
From state y, read in second letter, which is b, which takes us to state z.
From state z, read in third letter, which is a, which takes us to state z.
From state z, read in fourth letter, which is b, which takes us to state z.
No more letters in input string so stop.
On the input string abab:
We ended up in state z, which is a nal state.
we say that abab is accepted by this FA.
Denition: The set of all strings accepted is the language associated with
or accepted by the FA.
Note that
the above FA accepts all strings that have the letter b in them and no
other strings.
CHAPTER 5. FINITE AUTOMATA 5-7
the language accepted by this FA is the one dened by the regular ex-
pression
(a +b)

b(a +b)

Can also draw transition diagram:

directed graph
directed edge
every state has as many outgoing edges as there are letters in the alpha-
bet.
it is possible for a state to have no incoming edges.
the start state is labeled with a .
nal states are labeled with a +.
some states are neither labeled with or +.
CHAPTER 5. FINITE AUTOMATA 5-8
Example: From before.
z +
x - y
a
a
b
b
b
a
5.3 Examples of FA
Example: regular expression
(a +b)(a +b)

= (a +b)
+
All strings over the alphabet = a, b except .
a, b
a, b
- +
CHAPTER 5. FINITE AUTOMATA 5-9
Example: regular expression
(a +b)

-
+ a, b
This FA accepts all strings over the alphabet = a, b including .
CHAPTER 5. FINITE AUTOMATA 5-10
There are FAs that accept the language having no words:
FA has no nal states
-
a
a, b
a
b
b
Final state cannot be reached from start state because graph discon-
nected.
- + b
a
a, b a, b
b
a
Final state cannot be reached from start state because no path
- + b
a, b a, b a
a, b
CHAPTER 5. FINITE AUTOMATA 5-11
Example: Build FA to accept all words in the language
a(a +b)

+
a, b
+
a, b
or
-
a
b
a, b
+
-
a
b
a, b
a, b
Note that
more than one possible FA for any given language
can have more than one nal state
CHAPTER 5. FINITE AUTOMATA 5-12
Example:
a
b
b a
a
b
a, b
1 - 4 +
2
3
Note that
ababa is not accepted.
baaba is accepted.
FA accepts strings that have a double letter
Regular expression of language
(a +b)

(aa +bb)(a +b)

CHAPTER 5. FINITE AUTOMATA 5-13

Example:
-
a, b a, b b a, b
b a
a, b
a
+
Only accepts words whose third and fourth letters are ab.
Rejects all other words
Regular expressions:
1. (aaab +abab +baab +bbab)(a +b)

2. (a +b)(a +b)ab(a +b)

CHAPTER 5. FINITE AUTOMATA 5-14

Example: Only accepts the word aba.
a, b
- +
a a
a b b a, b
b
Example: Only accepts the words aba and ba.
- +
+
a
b a b
a
a, b
a, b
b a b a, b
CHAPTER 5. FINITE AUTOMATA 5-15
Example: Regular expression:
(a +ba

ba

ba

b)
+
Language with words having at least one letter and the number of bs divisible
by 4.
1 - 2
4
5
3 +
a
b
a
b
a
a a b
b
b
Example: Only accepts the word .
+
-
a, b a, b
CHAPTER 5. FINITE AUTOMATA 5-16
Example: Regular expression:
(a +b)

b
Words that end with b
does not include .
- + a
b
b
a
Example: Regular expression:
+ (a +b)

b
Either or words that end in b; i.e., words that do not end in a.
+
-
b a
a
b
CHAPTER 5. FINITE AUTOMATA 5-17
Example: Regular expression:
(a +b)

aa + (a +b)

bb
Words that end in a double letter.
-
+
+
a
b
a b b
a
b b
a
a
CHAPTER 5. FINITE AUTOMATA 5-18
Example: EVEN-EVEN
b
b
a a
b
b
1
+
-
2
3 4
a a
Note that
Every b moves us either left or right.
Every a moves us either up or down.
Chapter 6
Transition Graphs
6.1 Introduction
Each FA has the following properties (among others):
For each state x and each letter , there is exactly one arc leaving
x labeled with .
Can only read one letter at a time when traversing an arc.
Exactly one start state.
Now we want a dierent kind of machine that relaxes the above requirements:
For each state x and each letter , we do not require that there is
exactly one arc leaving x labeled with .
Able to read any number of letters at a time when traversing an arc.
Specically, each arc is now labeled with a string s

, so the string s
might be or it might be a single letter .
If an arc is labeled with , we traverse the arc without reading any letters
from the input string.
If an arc is labeled with a non-empty string s

, we can traverse the

arc if and only if the next unread letter(s) from the original input string
are the string s.
6-1
CHAPTER 6. TRANSITION GRAPHS 6-2
Suppose that we are in a state and we cannot leave the state because
there is no arc leaving the state labeled with a string that corresponds
to the next unread letters from the input string. Then if there are still
more unread letters from the original input string, the machine crashes.
There may be more than one way to process a string on the machine,
and so the machine may be nondeterministic.
If there is at least one way of processing the string on the machine such
that it ends in a nal state with no unread letters left and without
crashing, then the string is accepted; otherwise, the string is rejected.
There can be more than one start state.
Example: Consider the following machine that processes strings over the
alphabet = a, b:
aa
b
b
a
a, b,
1
2
+ 3
Note that this machine is not a nite automaton:
The arc from state 1 to state 2 is labeled with the string aa, which is
not a single letter.
There are two arcs leaving state 2 labeled with b.
There is no arc leaving state 2 labeled with a.
There is an arc from state 1 to state 3 labeled with , which is not a
letter from .
There is no arc leaving state 3 labeled with b.
CHAPTER 6. TRANSITION GRAPHS 6-3
Example: Only accepts the word aaba
- +
aaba
a, b
or
- +
aaba
a, b a, b
Example: Accepts all words that contain a doubled letter.
- +
a, b
aa, bb
a, b
Note that we must decide how many letters to read from the input string
each time we go back for more.
Depending on how we process the string abb, the machine may or may
not accept it.
Thus, we say that a string is accepted by a machine if there is some way
(called a successful path) to process all of the letters in the string and
end in a nal state without having crashed.
If there is no way to do this, then the string is not accepted.
For example, consider the string baba, which is not accepted.
6.2 Denition of Transition Graph
Denition: A transition graph (TG) is a collection M = (K, , , S, F)
where:
CHAPTER 6. TRANSITION GRAPHS 6-4
1. K is a nite set of states.
S K is a set of start states with S ,= (but possibly with more
than one state), where each start state is designated pictorially by
.
F K is a set of nal states (possibly empty, possibly all of K),
where each nal state is designated pictorially by .
2. An alphabet of possible input letters from which input strings are
formed.
3. K

K is a nite set of transitions, where each transition (arc)

from one state to another state is labeled with a string s

.
If an arc is labeled with , we traverse the arc without reading any
letters from the input string.
If an arc is labeled with an non-empty string s

, we can traverse
the arc if and only if the next unread letter(s) from the original input
string are the string s.
We allow for the possibility that for any state x K and any string
s

, there is more than one arc leaving x labeled with string s.

Also, we allow for the possibility that for any state x K and any
letter , there is no arc leaving state x labeled with .
Remarks:
when an edge is labeled with , we can take that edge without consuming
any letters from the input string.
We can have more than one start state.
Note that every FA is also a TG.
However, not every TG is an FA.
CHAPTER 6. TRANSITION GRAPHS 6-5
6.3 Examples of Transition Graphs
Example: this TG accepts nothing, not even .
-
Example: this TG accepts only the string .
+
-
+
-
or
a, b
CHAPTER 6. TRANSITION GRAPHS 6-6
Example: This TG accepts only the words , aaa and bbbb.
-
-
- +
bbbb
aaa
CHAPTER 6. TRANSITION GRAPHS 6-7
Example: this TG accepts only words that end in aba; i.e., the language
generated by the regular expression
(a +b)

aba
- +
aba
a, b
Example: this TG accepts the language of all words that begin and end with
the same letter and have at least two letters.
-
+
+
a
b
a, b
a, b
b
a
CHAPTER 6. TRANSITION GRAPHS 6-8
Example: this TG accepts the language of all words in which the as occur
in clumps of three and that end in four or more bs.
- +
aaa
b
aaa b
b b b
Example: this is the TG for EVEN-EVEN
+
-
ab, ba
aa, bb
ab, ba
aa, bb
Example: Is the word baaabab accepted by this machine?
Chapter 7
Kleenes Theorem
7.1 Kleenes Theorem
The following theorem is the most important and fundamental result in the
theory of FAs:
Theorem 6 Any language that can be dened by either
regular expression, or
nite automata, or
transition graph
can be dened by all three methods.
Proof. The proof has three parts:
Part 1: (FA TG) Every language that can be dened by an FA can also
be dened by a transition graph.
Part 2: (TG RegExp) Every language that can be dened by a transition
graph can also be dened by a regular expression.
Part 3: (RegExp FA) Every language that can be dened by a regular
expression can also be dened by an FA.
7-1
CHAPTER 7. KLEENES THEOREM 7-2
7.2 Proof of Part 1: FA TG
We previously saw that every FA is also a transition graph.
Hence, any language that has been dened by a FA can also be dened
by a transition graph.
7.3 Proof of Part 2: TG RegExp
We will give a constructive algorithm for proving part 2.
Thus, we will describe an algorithm to take any transition graph T and
form a regular expression corresponding to it.
The algorithm will work for any transition graph T.
The algorithm will nish in nite time.
An overview of the algorithm is as follows:
Start with any transition graph T.
First, transform it into an equivalent transition graph having only one
start state and one nal state.
In each following step, eliminate either some states or some arcs by
transforming the TG into another equivalent one.
We do this by replacing the strings labelling arcs with regular expres-
sions.
We can traverse an arc labelled with a regular expression using any string
that can be generated by the regular expression.
End up with a TG having only two states, start and nal, and one arc
going from start to nal.
The nal TG will have a regular expression on its one arc
Note that in each step we eliminate some states or arcs.
Since the original TG has a nite number of states and arcs, the algo-
rithm will terminate in a nite number of iterations.
CHAPTER 7. KLEENES THEOREM 7-3
Algorithm:
1. If T has more than one start state, add a new state and add arcs labeled
going to each of the original start states.
=>
2
3
4
b
a
a
b
...
1
-
/\
/\
1 - 3
4
b
a
a
b
...
2 -
2. If T has more than one nal state, add a new state and add arcs labeled
going from each of the original nal states to the new state. Need to
make sure the nal state is dierent than the start state.
...
+
+
a
b
a
a, b
a, b
b =>
...
a
b
a
a, b
a, b
b +
/\
/\
CHAPTER 7. KLEENES THEOREM 7-4
3. Now we give an iterative procedure for eliminating states and arcs
(a) If T has some state with n > 1 loops circling back to itself, where the
loops are labeled with regular expressions r
1
, r
2
, . . . , r
n
, then replace
the n loops with a single loop labeled with the regular expression
r
1
+r
2
+ +r
n
.
r
1
r
2
r
3
=> r + r + r
1 2 3
(b) If two states are connected by n > 1 direct arcs in the same di-
rection, where the arcs are labelled with the regular expressions
r
1
, r
2
, . . . , r
n
, then replace the n arcs with a single arc labeled with
the regular expression r
1
+r
2
+ +r
n
.
r
r
r + r
1
2
1
=>
2
CHAPTER 7. KLEENES THEOREM 7-5
(c) Bypass operation:
i. If there are three states x, y, z such that
there is an arc from x to y labelled with the regular expres-
sion r
1
and
an arc from y to z labelled with the regular expression r
2
,
then replace the two arcs and the state y with a single arc from
x to z labelled with the regular expression r
1
r
2
.
x y z x z
r
r
r r
1
2 1 2
=>
r
3
x y z x z
r
r
r r
1
2 1
=>
3 2
*r
CHAPTER 7. KLEENES THEOREM 7-6
ii. If there are
n+2 states x, y, z
1
, z
2
, . . . , z
n
such that there is an arc from
x to y labelled with the regular expression r
0
, and
an arc from y to z
i
, i = 1, 2, . . . , n, labelled with the regular
expression r
i
, and
an arc from y back to itself labelled with regular expression
r
n+1
,
then replace the n + 1 original arcs and the state y with n
arcs from x to z
i
, i = 1, 2, . . . , n, each labelled with the regular
expression r
0
r
n+1
r
i
.
z
z
z
1
2
n
.
.
.
z
z
z
1
2
n
.
.
.
x x y
r
r
r
r
0
1
n+1
2
r
=>
0 n+1 1
0 n+1 2
0 n+1 n
r r* r
r r* r
r r* r
n
iii. If any other arcs led directly to y, divert them directly to the
z
i
s.
CHAPTER 7. KLEENES THEOREM 7-7
iv. Need to make sure that all paths possible in the original TG
are still possible after the bypass operation.
Example
w z
w z
1 3
r (r r )* r
4 2
1 3
r r (r r )* r
5 2 2
1 3
r (r r )* r
4 2 1 3
r r (r r )* r
5 2 2
w
x
y
z
r
r r
r
r
1
2 3
4
5
=>
=>
+
CHAPTER 7. KLEENES THEOREM 7-8
Example:
Suppose we want to get rid of state y.
Need to account for all paths that go through state y.
There are arcs coming from x, w, and z going into y.
There are arcs from y to x and z.
Thus, we need to account for each possible path from a
state having an arc into y (i.e., x, w, z) to each state
having an arc from y (i.e., x, z)
Thus, we need to account for the paths from
x to y to x, which has regular expression r
1
r

2
r
5
x to y to z, which has regular expression r
1
r

2
r
3
w to y to x, which has regular expression r
7
r

2
r
5
w to y to z, which has regular expression r
7
r

2
r
3
z to y to x, which has regular expression r
6
r

2
r
5
z to y to z, which has regular expression r
6
r

2
r
3
Thus, after eliminating state y, we get the following:
v. Never delete the unique start or nal state.
CHAPTER 7. KLEENES THEOREM 7-9
Example:
1 - 2 3
4 5 +
1 -
5 +
1 -
5 +
a
=>
abba
abb
bb
a+b
=>
a*(abba+abb+bb)(a+b)*
/\
/\ bb
abb
a, b
ba ab
a
CHAPTER 7. KLEENES THEOREM 7-10
Example:
=>
1 - 2 -
4 5 +
b b
b
a
a
b a
a
3 +
a, b
3
4
b b
b
a
a
b a
-
5
/\ /\
a
a+b
/\
/\
+
4
b
-
=>
a
a
+
a(a+b)*
/\+b
bb*
bb*a(a+b)*
1 2
4
b
-
5
/\
a
a+b
/\
=>
bb*a
a
a
b
+
bb*
2 2
CHAPTER 7. KLEENES THEOREM 7-11
4
b
-
=>
a
+
a(a+b)*
/\+b
bb*(/\+a(a+b)*)
a(ba)*a(a+b)* + ab(ab)*bb*(/\+a(a+b)*)
=> -
+
=> a(ba)*a(a+b)* + ab(ab)*bb*(/\+a(a+b)*) (/\+b)((ab)*bb*(/\+a(a+b)*) + a(ba)*a(a+b)*) +
a(ba)*a(a+b)* + ab(ab)*bb*(/\+a(a+b)*)
(/\+b)((ab)*bb*(/\+a(a+b)*) + a(ba)*a(a+b)*)
4
b
-
=>
a
a
+
a(a+b)*
/\+b
bb*(/\+a(a+b)*) 2 2
CHAPTER 7. KLEENES THEOREM 7-12
7.4 Proof of Part 3: RegExp FA
To show: every language that can be dened by a regular expression can also
be dened by a FA.
We will do this by using a recursive denition and a constructive algorithm.
Recall
every regular expression can be built up from the letters of the alphabet
and and .
Also, given some existing regular expressions, we can build new regular
expressions by applying the following operations:
1. union (+)
2. concatenation
3. closure (Kleene star)
We will not include r
+
in our discussion here, but this will not be a
problem since r
+
= rr

.
CHAPTER 7. KLEENES THEOREM 7-13
Recall that we had the following recursive denition for regular expressions:
Rule 1: If x , then x is a regular expression. is a regular expression.
is a regular expression.
Rule 2: If r
1
and r
2
are regular expressions, then r
1
+r
2
is a regular expres-
sion.
Rule 3: If r
1
and r
2
are regular expressions, then r
1
r
2
is a regular expression.
Rule 4: If r
1
is a regular expression, then r

1
is a regular expression.
Based on the above recursive denition for regular expressions, we have the
following recursive denition for FAs associated with regular expressions:
Rule 1:
There is an FA that accepts the language L dened by the regular
expression x; i.e., L = x, where x , so language L consists of
only a single word and that word is the single letter x.
There is an FA that accepts the language dened by regular expres-
sion ; i.e., the language .
There is an FA dened by the regular expression ; i.e., the language
with no words, which is .
Rule 2: If there is an FA called FA
1
that accepts the language dened by
the regular expression r
1
and there is an FA called FA
2
that accepts the
language dened by the regular expression r
2
, then there is an FA called
FA
3
that accepts the language dened by the regular expression r
1
+r
2
.
Rule 3: If there is an FA called FA
1
that accepts the language dened by
the regular expression r
1
and there is an FA called FA
2
that accepts the
language dened by the regular expression r
2
, then there is an FA called
FA
3
that accepts the language dened by the regular expression r
1
r
2
,
which is the concatenation.
Rule 4: If there is an FA called FA
1
that accepts the language dened by
the regular expression r
1
, then there is an FA called FA
2
that accepts
the language dened by the regular expression r

1
.
CHAPTER 7. KLEENES THEOREM 7-14
Lets now show that each of the rules hold by construction:
Rule 1: There is an FA that accepts the language L dened by the regular
expression x; i.e., L = x, where x . There is an FA that ac-
cepts language dened by the regular expression . There is an FA that
accepts the language dened by the regular expression .
If x , then the following FA accepts the language x:
- +
- {x}
x
An FA that accepts the language is
+
_
An FA that accepts the language is
-
CHAPTER 7. KLEENES THEOREM 7-15
Rule 2: If there is an FA called FA
1
that accepts the language dened by
the regular expression r
1
and there is an FA called FA
2
that accepts the
language dened by the regular expression r
2
, then there is an FA called
FA
3
that accepts the language dened by the regular expression r
1
+r
2
.
Suppose regular expressions r
1
and r
2
are dened with respect to a
common alphabet .
Let L
1
be the language generated by regular expression r
1
.
L
1
has nite automaton FA
1
.
Let L
2
be the language generated by regular expression r
2
.
L
2
has nite automaton FA
2
.
Regular expression r
1
+r
2
generates the language L
1
+L
2
.
Recall L
1
+L
2
= w

: w L
1
or w L
2
.
Thus, w L
1
+ L
2
if and only if w is accepted by either FA
1
or
FA
2
(or both).
We need FA
3
to accept a string if the string is accepted by FA
1
or
FA
2
or both.
We do this by constructing a new machine FA
3
that simultaneously
keeps track of where the input would be if it were running on FA
1
and where the input would be if it were running on FA
2
.
Suppose FA
1
has states x
1
, x
2
, . . . , x
m
, and FA
2
has states y
1
, y
2
, . . . , y
n
.
Assume that x
1
is the start state of FA
1
and that y
1
is the start
state of FA
2
.
We will create FA
3
with states of the form (x
i
, y
j
).
The number of states in FA
3
is at most mn, where m is the number
of states in FA
1
and n is the number of states in FA
2
.
Each state in FA
3
corresponds to a state in FA
1
and a state in
FA
2
.
FA
3
accepts string w if and only if either FA
1
or FA
2
accepts w.
So nal states of FA
3
are those states (x, y) such that x is a nal
state of FA
1
or y is a nal state of FA
2
.
CHAPTER 7. KLEENES THEOREM 7-16
We use the following algorithm to construct FA
3
from FA
1
and FA
2
.
Suppose that is the alphabet for both FA
1
and FA
2
.
Given FA
1
= (K
1
, ,
1
, s
1
, F
1
) with
Set of states K
1
= x
1
, x
2
, . . . , x
m

s
1
= x
1
is the initial state
F
1
K
1
is the set of nal states of FA
1
.

1
: K
1
K
1
is the transition function for FA
1
.
Given FA
2
= (K
2
, ,
2
, s
2
, F
2
) with
Set of states K
2
= y
1
, y
2
, . . . , y
n

s
2
= y
1
is the initial state
F
2
K
2
is the set of nal states of FA
2
.

2
: K
2
K
2
is the transition function for FA
2
.
We then dene FA
3
= (K
3
, ,
3
, s
3
, F
3
) with
Set of states K
3
= K
1
K
2
= (x, y) : x K
1
, y K
2

The alphabet of FA
3
is .
FA
3
has transition function
3
: K
3
K
3
with

3
((x, y), ) = (
1
(x, ),
2
(y, )).
The initial state s
3
= (s
1
, s
2
).
The set of nal states
F
3
= (x, y) K
1
K
2
: x F
1
or y F
2
.
Since K
3
= K
1
K
2
, the number of states in the new machine FA
3
is [K
3
[ = [K
1
[ [K
2
[.
But we can leave out a state (x, y) K
1
K
2
from K
3
if (x, y)
is not reachable from FA
3
s initial state (s
1
, s
2
).
This would result in fewer states in K
3
, but still we have [K
1
[
[K
2
[ as an upper bound for [K
3
[; i.e., [K
3
[ [K
1
[ [K
2
[.
CHAPTER 7. KLEENES THEOREM 7-17
Example: L
1
= words with b as second letter
with regular expression r
1
= (a +b)b(a +b)

L
2
= words with odd number of as
with regular expression r
2
= b

a(b +ab

a)

x1- x2 x3+
x4
a, b b
a
a, b
a, b
b
a
a
b
y1- y2+
FA2 for L2: FA1 for L1:
x1,y1-
x2,y2+
x4,y1
x3,y2+
x4,y2+
a
b
b
a
b
a a
b
a
a
b
b
a
b
FA3 for L1+L2:
x2,y1
x3,y1+
CHAPTER 7. KLEENES THEOREM 7-18
Rule 3: If there is an FA called FA
1
that accepts the language dened by
the regular expression r
1
and there is an FA called FA
2
that accepts the
language dened by the regular expression r
2
, then there is an FA called
FA
3
that accepts the language dened by the regular expression r
1
r
2
.
For this part,
we need FA
3
to accept a string if the string can be factored into
two substrings, where the rst factor is accepted by FA
1
and the
second factor is accepted by FA
2
.
One problem is we dont know when we reach the end of the rst
factor and the beginning of the second factor.
Example: L
1
= words that end with aa
with regular expression r
1
= (a +b)

aa
L
2
= words with odd length
with regular expression r
2
= (a +b)((a +b)(a +b))

Consider the string baaab.

If we factor it as (baa)(ab), then baa L
1
but ab , L
2
.
However, another factorization, (baaa)(b), shows that baaab
L
1
L
2
since baaa L
1
and b L
2
.
FA2 for L2:
y1- y2+
a, b
a, b
FA1 for L1:
x1- x2
x3+
b
a
b
a
a
b
CHAPTER 7. KLEENES THEOREM 7-19
Basically idea of building FA
3
for L
1
L
2
from FA
1
for L
1
and FA
2
for L
2
:
Recall L
1
L
2
= w = w
1
w
2
: w
1
L
1
, w
2
L
2
.
So a string w is in L
1
L
2
if and only if we can factor w = w
1
w
2
such that w
1
is accepted by FA
1
and w
2
is accepted by FA
2
.
FA
3
initially acts like FA
1
.
When FA
3
hits a

state of FA
1
,
Start a version of FA
2
.
Keep processing on FA
1
and any previous versions of FA
2
.
We need to keep processing on FA
1
because we dont know
where the rst factor w
1
ends and the second factor w
2
begins
Final states of FA
3
are those states that have at least one nal
state from FA
2
.
More formally, we build machine FA
3
in following way:
Suppose that FA
1
and FA
2
have the same alphabet .
Let L
1
be language generated by regular expression r
1
and hav-
ing FA FA
1
= (K
1
, ,
1
, s
1
, F
1
).
Let L
2
be language generated by regular expression r
2
and hav-
ing FA FA
2
= (K
2
, ,
2
, s
2
, F
2
).
Denition: For any set S, dene 2
S
to be the set of all possible
subsets of S.
Example: If S = a, bb, ab, then
2
S
= , a, bb, ab, a, bb, a, ab, bb, ab, a, bb, ab.
Fact: If [S[ < , then [2
S
[ = 2
|S|
; i.e., there are 2
|S|
dierent
subsets of S.
Machine FA
3
= (K
3
, ,
3
, s
3
, F
3
) for L
1
L
2
is as follows:
States
K
3
= x +Y : x K
1
, Y 2
K
2
;
i.e., each state of FA
3
is a set of states, where exactly one
of the states is from FA
1
and the rest (possibly none) are
from FA
2
.
Initial state s
3
= s
1
; i.e., the initial state of FA
3
is the
set consisting of only the initial state of FA
1
.
CHAPTER 7. KLEENES THEOREM 7-20
Transition function
3
: K
3
K
3
is dened as

3
(x, y
1
, . . . , y
n
, )
=

1
(x, ),
2
(y
1
, ), . . . ,
n
(y
2
, ) if
1
(x, ) , F
1
,

1
(x, ),
2
(y
1
, ), . . . ,
n
(y
2
, ), s
2
if
1
(x, ) F
1
,
where x, y
1
, . . . , y
n
K
3
, n 0, x K
1
, y
i
K
2
for
i = 1, . . . , n, and .
Final states
F
3
= x, y
1
, . . . , y
n
: n 1, y
i
F
2
for some i = 1, . . . , n.
The number of states in FA
3
is
[K
3
[ = [K
1
[ [2
K
2
[ = [K
1
[ 2
|K
2
|
.
Actually, we can leave out fromK
3
any states x, y
1
, . . . , y
n

that are not reachable from the initial state s

3
.
In this case, [K
1
[ 2
|K
2
|
still provides an upper bound for
[K
3
[; i.e., [K
3
[ [K
1
[ 2
|K
2
|
.
CHAPTER 7. KLEENES THEOREM 7-21
Example: L
1
= words that end with aa
with regular expression r
1
= (a +b)

aa
L
2
= words with odd length
with regular expression r
2
= (a +b)((a +b)(a +b))

FA2 for L2:

y1- y2+
a, b
a, b
FA1 for L1:
x1- x2
x3+
b
a
b
a
a
b
b
a
b
a
a
b a
b
a
a
b
a
b
a
b
b
b
a
x1- x2
x3,y1 x2,y2+
x1,y1 x1,y2+ x2,y1
x1,y1,y2+
x2,y2,y1+
x3,y2,y1+
FA3 for L1 L2:
b
a
CHAPTER 7. KLEENES THEOREM 7-22
Rule 4: If there is an FA called FA
1
that accepts the language dened by
the regular expression r
1
, then there is an FA called FA
2
that accepts
the language dened by the regular expression r

1
.
Basic idea of how to build machine FA
2
:
Each state of FA
2
corresponds to one or more states of FA
1
.
FA
2
initially acts like FA
1
.
when FA
2
hits a

state of FA
1
, then FA
2
simultaneously keeps
track of how the rest of the string would be processed on FA
1
from
where it left o and how the rest of the string would be processed
on FA
1
starting in the start state.
Whenever FA
2
hits a

state of FA
1
, we have to start a new
process starting in the start state of FA
1
(if no version of FA
1
is
currently in its start state.)
The nal states of FA
2
are those states which have a correspondence
to some nal state of FA
1
.
We need to be careful about making sure that FA
2
accepts .
To have FA
2
accept , we make the start state of FA
2
also a nal
state.
But we need to be careful when there are arcs going into the start
state of FA
1
.
CHAPTER 7. KLEENES THEOREM 7-23
Formally, we build the machine FA
2
for L

1
as follows:
Let L
1
be language generated by regular expression r
1
and having
nite automaton FA
1
= (K
1
, ,
1
, s
1
, F
1
).
For now, assume that FA
1
does not have any arcs entering the
initial state s
1
.
Know that language L

1
is generated by regular expression r

1
.
Dene FA
2
= (K
2
, ,
2
, s
2
, F
2
) for L

1
with
States K
2
= 2
K
1
.
Initial state s
2
= s
1
.
Transition function
2
: K
2
K
2
with

2
(x
1
, . . . , x
n
, )
=

1
(x
1
, ), . . . ,
1
(x
n
, ) if
1
(x
k
, ) , F
1
for all k = 1, . . . , n,

1
(x
1
, ), . . . ,
1
(x
n
, ), s
1
if
1
(x
k
, ) F
1
for some k = 1, . . . , n,
where x
1
, . . . , x
n
K
2
, n 1, x
i
K
1
for all i = 1, . . . , n,
and .
Final states
F
2
= s
1
+x
1
, . . . , x
n
: n 1, x
i
F
1
for some i = 1, . . . , n.
The number of states in FA
2
is
[K
2
[ = [2
K
1
[ = 2
|K
1
|
.
Actually, we can leave out from K
2
any state x
1
, . . . , x
n
that
is not reachable from the initial state s
2
.
In this case, 2
|K
1
|
still provides an upper bound for [K
2
[; i.e.,
[K
3
[ 2
|K
1
|
.
CHAPTER 7. KLEENES THEOREM 7-24
Example: Consider language L having regular expression
r = (a +bb

ab

a)((b +ab

a)b

a)

b
x1+
-
x4
x3
x2,x1+
x4,x2,x1+
x3,x4
x2,x1,x3+ x1,x2,x3,x4+
b a
b
b
a
b
a
a
b
a
a
a
a
b
b
FA for L*:
x1-
x2+
x3
x4
a
b
a
a
b
b
a
b
FA for L:
CHAPTER 7. KLEENES THEOREM 7-25
Example: Consider language L having regular expression
(a +b)

b
Need to be careful since we can return to the start state.
x1- x2+
b
b
FA for L:
a
a
If we blindly applied previous method for constructing FA for L

, we
get the following:
x1+
-
x2,x1+
a
b a
b
Problem:
Note that start state is nal state.
But this FA accepts a , L

, and so this FA is incorrect.

Problem occurs because we can return to start state in original FA,
and since we make the start state a nal state in new FA.
CHAPTER 7. KLEENES THEOREM 7-26
Solution:
Given original FA FA
1
having arcs going into the initial state, cre-
ate an equivalent FA

FA
1
having no arcs going into the initial state
by splitting the original start state x
1
of FA
1
into two states x
1.1
and x
1.2
x
1.1
is the new start state of

FA
1
and is never visited again
after the rst letter of the input string is read.
x
1.2
in

FA
1
corresponds to x
1
after the rst letter of the input
string is read.
Then run algorithm to create FA for L

from the new FA

FA
1
.
x1.2
a
b
a
x1.1+
-
a
b
b
FA for L*:
x2, x1.1+
x2+
x1.1-
x1.2
a
b
b
a
new FA for L:
b
a
CHAPTER 7. KLEENES THEOREM 7-27
7.5 Nondeterministic Finite Automata
Denition: A nondeterministic nite automaton (NFA) is given by M =
(K, , , s, F), where
1. K is a nite set of states.
s K is the initial state, which is denoted pictorially by , and
there is exactly one initial state.
F K is a set of nal states (possibly empty), where each nal
state is denoted pictorially by

.
2. An alphabet of possible input letters.
3. K K is a nite set of transitions, where each transition (arc)
from one state to another state is labeled with a letter . (We do
not allow for to be the label of an arc since is a string and not a
letter of .) We allow for the possibility of more than one edge with the
same label from any state and there may be a state (or states) for which
certain input letters have no edge leaving that state.
CHAPTER 7. KLEENES THEOREM 7-28
Example:
b
a
a
b
b
a
a, b
b -
+
a
Note that
denition of NFA is dierent from that of FA since
a FA must have from each state an arc labeled with each letter of
alphabet, while NFA does not.
a FA is deterministic, while a NFA may be nondetermisic.
An NFA can have repeated labels from any single state.
NFA allows for human choice to become a factor in selecting a way to
process an input string.
The denition of NFA is dierent from that of TG since
a TG can have arcs labeled with substrings of letters while a NFA
has arcs labeled with only letters.
a TG can have arcs labeled with while a NFA cannot.
a TG can have more than one start state while a NFA can only
have one.
CHAPTER 7. KLEENES THEOREM 7-29
Can transform any NFA with repeated labels from any single state to an
equivalent TG with no repeated labels from any single state.
1
3
4
5
b
a
b
b
2 . . .
. . .
. . .
. . .
=>
1
2
3
4
5
b
a
^
b
^
b
7.6 Properties of NFA
Theorem 7 FA = NFA; i.e., any language denable by a NFA is also den-
able by a deterministic FA and vice versa.
Proof. Note that
Every FA is an NFA since we can consider an FA to be an NFA without
the extra possible features.
Every NFA is a TG.
Kleenes theorem states that every TG has an equivalent FA.
NFA useful because
applications in articial intelligence (AI).
CHAPTER 7. KLEENES THEOREM 7-30
given two FAs for two languages with regular expressions r
1
and r
2
, it
is easy to construct an NFA to accept language corresponding to regular
expression r
1
+r
2
.
CHAPTER 7. KLEENES THEOREM 7-31
Example:
b
a, b
a
a, b
b
a, b
a
+ +
- - +
a
b
a
b
b
a
a, b
b
a, b
a
a, b
b
a, b
a
+ +
- +
a
b
a
b
b
a
a, b
FA1: FA2:
NFA
1+2
This works when neither of the original FAs has any arcs going into
their original initial states.
If one or both of the original FAs has an arc going into its original
initial state, the newly constructed FA for the language corresponding
CHAPTER 7. KLEENES THEOREM 7-32
to regular expression r
1
+r
2
may be incorrect. This is because the new
FA may process part of the word on one of the original FAs and then
process the rest of the word on the other FA, and then incorrectly accept
the word.
Chapter 8
Finite Automata with Output
8.1 Moore Machines
Denition: A Moore machine is a collection of ve things:
1. A nite set of states q
0
, q
1
, q
2
, . . . , q
n
, where q
0
is designated as the start
state.
2. A nite alphabet of letters for forming the input string
= a, b, c, . . .
3. A nite alphabet of possible output characters
= x, y, z, . . .
4. A transition table that shows for each state and each input letter what
state is reached next.
5. An output table that shows what character from is printed by each
state that is entered.
8-1
CHAPTER 8. FINITE AUTOMATA WITH OUTPUT 8-2
Example: Input alphabet: = a, b
Output alphabet: = 0, 1
States: q
0
, q
1
, q
2
, q
3
a b Output
q
0
q
3
q
2
0
q
1
q
1
q
0
0
q
2
q
2
q
3
1
q
3
q
0
q
1
0
b
a
a
a
b
a
q / 1
q / 0
q / 0
2
1
3
q / 0
0
b
b
On input string bababbb, the output is 01100100.
8.2 Mealy Machines
Denition: A Mealy machine is a collection of four things:
1. A nite set of states q
0
, q
1
, q
2
, . . . , q
n
, where q
0
is designated as the start
state.
2. A nite alphabet of letters = a, b, . . ..
3. A nite alphabet of output characters = x, y, z, . . ..
CHAPTER 8. FINITE AUTOMATA WITH OUTPUT 8-3
4. A pictorial representation with states reresented by small circles and
directed edges indicating transitions between states. Each edge is labeled
with a compound symbol of the form i/o, where i is an input letter and
o is an output character. Every state must have exactly one outgoing
edge for each possible input letter. The way we travel is determined by
the input letter i. While traveling on the edge, we must print the output
character o.
The key dierence between Moore and Mealy machines:
Moore machines print character when in state.
Mealy machines print character when traversing an arc.
Example: Mealy machine
a / 0
a / 0
b / 1
b / 0
a / 1
b / 0
a / 1
b / 0
q
q
q
q
0
1
2
3
CHAPTER 8. FINITE AUTOMATA WITH OUTPUT 8-4
Example: Mealy machine prints out the 1s complement of an input bit
string.
= = 0, 1.
q
0
0 / 1,
1/ 0
CHAPTER 8. FINITE AUTOMATA WITH OUTPUT 8-5
8.3 Properties of Moore and Mealy Machines
Denition: Given a Mealy machine Me and a Moore machine Mo, which
automatically prints the character x in the start state, we say these two ma-
chines are equivalent if for every input string the output string from Mo is
exactly x concatenated with the output from Me.
Theorem 8 If Mo is a Moore machine, then there is a Mealy machine Me
that is equivalent to it.
Proof.
Consider any state q
i
of Mo.
Suppose Mo prints the charater t upon entering q
i
.
Hence, the label in state q
i
is q
i
/t.
Suppose that there are n arcs entering q
i
, with labels a
1
, a
2
, . . . , a
n
.
We create the machine Me by changing the labels on the incoming arcs
to q
i
to a
m
/t, m = 1, 2, . . . , n.
Change the label of state q
i
to be just q
i
.
a
b
a
a /t
b / t
a / t
q / t q
1 1
=>
Mo: Me:
CHAPTER 8. FINITE AUTOMATA WITH OUTPUT 8-6
CHAPTER 8. FINITE AUTOMATA WITH OUTPUT 8-7
Example: Convert Moore machine into equivalent Mealy machine.
q / 1
q / 0
q / 1
q / 1
q / 0
b
a
b
a
a
a
b
b
b
a
Mo:
Me:
q
q
q q
q
b / 0
q / 0
a / 1
b / 0
b / 1
a / 0
0
1
2 4
0
1
2
3
4
3
a / 0
b / 1
a / 1 b / 0
CHAPTER 8. FINITE AUTOMATA WITH OUTPUT 8-8
Theorem 9 For every Mealy machine Me, there is an equivalent Moore ma-
chine Mo.
Proof.
Consider any state q
i
of Me.
Suppose that there are n arcs entering q
i
, with labels a
1
/t
1
, a
2
/t
2
, . . . , a
n
/t
n
.
So if we enter state q
i
using the kth arc, we just read in a
k
and printed
t
k
.
Suppose that among t
1
, t
2
, . . . , t
n
, there are k dierent characters; call
them c
1
, c
2
, . . . , c
k
.
To create the Moore machine Mo, split the state q
i
into k dierent states;
call them q
1
i
, q
2
i
, . . . , q
k
i
.
State q
l
i
will correspond to the output character c
l
.
For each arc going into q
i
in Me which was labeled with the output
character c
l
, have that arc in Mo go to the state q
l
i
/c
l
. Label that arc
with its input letter.
For any state in Me which has no incoming edges, we arbitrarily assign
it any output character in Mo.
Me:
a / 0
a / 1
b / 0
a / 0
b / 1
b / 0
q
1
=>
Mo:
a
b
b
a
a / 0
b / 1
q / 0 q / 1
1 1
b / 1
a / 0
CHAPTER 8. FINITE AUTOMATA WITH OUTPUT 8-9
CHAPTER 8. FINITE AUTOMATA WITH OUTPUT 8-10
Example: Convert Mealy machine into equivalent Moore machine.
Me:
Mo:
q
q
q q
q
q / 0
q / 1
q / 0
q / 1
q / 0
q / 0
q / 0
b / 0
a / 0
0
1
2
3
4
0
0
1
2
2
3
4
a
a
b
a
a
b
a / 0
b / 1
b / 1
a / 0
b / 0
b / 0
a / 1
a / 0
b
a
b
a
b
b
b
a
Chapter 9
Regular Languages
9.1 Properties of Regular Languages
Denition: A language that can be dened by a regular expression is a
regular language.
Theorem 10 If L
1
and L
2
are regular languages, then L
1
+L
2
, L
1
L
2
, and L

1
are also regular languages.
Proof. (by regular expressions)
If L
1
and L
2
are regular languages, then there are regular expressions r
1
and r
2
that dene these languages.
r
1
+r
2
is a regular expression that denes the language L
1
+L
2
, and so
L
1
+L
2
is a regular language.
r
1
r
2
is a regular expression that denes the language L
1
L
2
, and so L
1
L
2
is a regular language.
r

1
is a regular expression that denes the language L

1
, and so L

1
is a
regular language.
9-1
CHAPTER 9. REGULAR LANGUAGES 9-2
Proof. (by machines)
If L
1
and L
2
are regular languages, then there are transition graphs TG
1
and TG
2
that accept them by Kleenes Theorem.
We may assume that TG
1
has a unique start state and unique nal state,
and the same for TG
2
.
We construct the TG for L
1
+L
2
as follows:
-
- -
TG1 TG2
.
.
.
.
.
.
.
.
new start state
We construct the TG for L
1
L
2
as follows:
new
state
start
- -
TG1 TG2
. . .
. . .
new
final
state
1 + 2 +
^
CHAPTER 9. REGULAR LANGUAGES 9-3
We construct the TG for L

1
as follows:
original
final
state
original
start
state
- - + +
^
^
TG1
^
^
. . .
new start state new final state
CHAPTER 9. REGULAR LANGUAGES 9-4
Remarks:
The technique given in the tapes of lectures 11 and 12 is wrong.
To see why, consider the following FA for the language
L = words having an odd number of bs
+
b
b
-
a a
Note that L

is the language consisting of and all words having

at least one b, which has regular expression + (a +b)

b(a +b)

(which was also wrong in the tape of lecture 12).

If we use the (incorrect) technique to construct a TG for L

given
in the taped lecture, then we get the following:
+
b
b
a a
-
^
^
However, the above TG accepts the string a , L

.
On the other hand, if we use the method presented above to con-
struct a TG for L

, then we get the following correct TG:

b
a a
-
^
^
^
+
b, ^
CHAPTER 9. REGULAR LANGUAGES 9-5
Example: alphabet = a, b
L
1
= all words ending with a
L
2
= all words containing the substring aa.
Regular expressions:
r
1
= (a +b)

a
r
2
= (a +b)

aa(a +b)

FA1:
FA2:
b
a
b
a
b
a
b
a
a, b
-
-
+
+
CHAPTER 9. REGULAR LANGUAGES 9-6
r + r = (a+b)*a + (a+b)*aa(a+b)*
1 2
1+2
b
+ b
a
b
a
a, b
+ a b
a
^
^
-
TG for r + r
1 2
CHAPTER 9. REGULAR LANGUAGES 9-7
b
- b
a
b
a
+
^
a b a, b
a
TG for r r
1 2
b
a
b
a
a, b
^
^
^
^
+
TG for r*
2
-
CHAPTER 9. REGULAR LANGUAGES 9-8
9.2 Complementation of Regular Languages
Denition: If L is a language over the alphabet , we dene L

to be its
complement, which is the language of all strings of letters from that are not
words in L, i.e., L

= w

: w , L.
Example: alphabet = a, b
L = language of all words in

containing the substring abb.

L

= language of all words in

not containing the substring abb.

Note that
(L

= L
Theorem 11 If L is a regular language, then L

is also a regular language.

In other words, the set of regular languages is closed under complementation.
Proof.
If L is a regular language, then there exists some FA that accepts L by
Kleenes Theorem.
Create new nite automaton FA

from FA as follows:
FA

has same states and arcs as FA.

Every nal state of FA becomes a nonnal state in FA

Every nonnal state of FA becomes a nal state in FA

FA

has same start state as FA.

FA

accepts the language L

.
Kleenes Theorem implies that L

is a regular language.
CHAPTER 9. REGULAR LANGUAGES 9-9
Example: = a, b
L = all words with length at least 2 and second letter b
L

= all words with length less than 2 or second letter a

a, b b
a
a, b
a, b
a, b b
a
a, b
a, b
FA:
FA:
- +
+
-
+
+
CHAPTER 9. REGULAR LANGUAGES 9-10
9.3 Intersections of Regular Languages
Theorem 12 If L
1
and L
2
are regular languages, then L
1
L
2
is a regular
language. In other words, the set of regular languages is closed under inter-
section.
Proof.
DeMorgans Law for sets states that
L
1
L
2
= (L

1
+L

2
)

L + L
1 2
L L
1 2
L
2
L
1
(L + L )
1 2
Since L
1
and L
2
are regular languages, Theorem 11 implies that L

1
and
L

2
are regular languages.
Theorem 10 then implies that L

1
+L

2
is a regular language.
Theorem 11 then implies that (L

1
+L

2
)

is a regular language.
CHAPTER 9. REGULAR LANGUAGES 9-11
Example: alphabet = a, b
L
1
= all words with length 2 and second letter b
L
2
= all words containing the substring ab.
CHAPTER 9. REGULAR LANGUAGES 9-12
x3
a, b
a
b
a, b
a, b
y3 b
a
a
a, b
b
x2
x4
a, b
a
b
a, b
a, b
y2 b
a
a
a, b
b
x1- x3+
y1- y3+
x1+
-
x2+
x4+
y1+
-
y2+
FA1 :
FA2 :
FA2 :
FA1 :
r1 = (a+b)b(a+b)* r2 = (a+b)*ab(a+b)*
CHAPTER 9. REGULAR LANGUAGES 9-13
x3, y3
a
b
b
b
b
a, b a
a, b
a
b
b
a
x2, y2
x2, y1 x3, y1
x4, y2 x4, y3
a
b
b
b
b
a, b a
a, b
a
b
b
a
x3, y2
x3, y2+
x1, y1+
-
x2, y1+ x3, y1+
x2, y2+
x4, y2+ x4, y3+
x1, y1-
x3, y3+
FA for (L1+L2):
FA for (L1+L2) :
a
a
a
a
CHAPTER 9. REGULAR LANGUAGES 9-14
As an exercise, we will now derive a regular expression for L
1
L
2
using our
FA for (L

1
+L

2
)

and our algorithm from Kleenes theorem:

CHAPTER 9. REGULAR LANGUAGES 9-15
Proof. (another for Theorem 12)
In proof of Kleenes theorem, we showed how to construct FA
3
that is
the union of FA
1
and FA
2
.
Suppose states of FA
1
are x
1
, x
2
, . . ..
Suppose states of FA
2
are y
1
, y
2
, . . ..
We do the same construction of FA
3
except we now make a state in
FA
3
a nal state only if both the corresponding x and y states are nal
states.
Then FA
3
accepts only words that are accepted by both FA
1
and FA
2
.
Example: alphabet = a, b
L
1
= all words with length 2 and second letter b
L
2
= all words containing the substring ab.
x2, y2
x2, y1 x3, y1
x4, y2 x4, y3
a
b
b
b
b
a, b a
a, b
a
b
b
a
x3, y2
x1, y1-
x3, y3+
a
a
Chapter 10
Nonregular Languages
10.1 Introduction
Example: Consider the following FA having 5 states:
1- 2+ 3
4 5
a
a
a
b
a
b
a
b
b
b
Lets process the string ababbaa on the FA:
1
a
2
b
4
a
3
b
5
b
4
a
3
a
2
Since 2 is a nal state, we accept the string ababbaa.
In general,
We always start in initial state.
After reading rst letter of input string,
10-1
CHAPTER 10. NONREGULAR LANGUAGES 10-2
we end may go to another state or return to initial state.
the maximum number of dierent states that we could have
visited after reading the rst letter is 2.
After reading the rst 2 letters of input string, the maximum num-
ber of dierent states that we could have visited is 3.
In general, after reading the rst m letters of input string, the
maximum number of dierent states that we could have visited is
m+ 1.
In our example above, after reading 5 letters, the maximum number of
dierent states that we could have visited is 5 + 1 = 6. But since the
FA has 5 states, we know that after reading in 5 letters, we must have
visited some state twice.
Consider the string aaabaa.
The string has length 6, which is more than the number of states
in the above FA.
We process the string as follows:
1
a
2
a
1
a
2
b
4
a
3
a
2
and so it is accepted.
Notice that state 1 is the rst state that we visit twice.
In general, if we have an FA with N states and we process a string w
with length(w) N, then there exists at least one state that we visit at
least twice.
Let u be the rst state that we visit twice.
Break up string w as w = xyz, where x, y, and z are 3 strings such
that
string x is the letters at the beginning of w that are read by
the FA until the state u is hit for the rst time.
string y is the letters used by the FA starting from the rst
time we are in state u until we hit state u the second time.
string z is the rest of the letters in w.
For example, for the string w = ababbaa processed on the above FA, we
have u = 2, and x = ab, y = abb, z = aa.
CHAPTER 10. NONREGULAR LANGUAGES 10-3
For example, for the string w = aaabaa processed on the above FA, we
have u = 1, and x = , y = aa, z = abaa.
10.2 Denition of Nonregular Languages
Denition: A language that cannot be dened by a regular expression is
called a nonregular language.
By Kleenes Theorem, a nonregular language cannot be accepted by any FA
or TG.
Consider
L = , ab, aabb, aaabbb, aaaabbbb, . . .
= a
n
b
n
: n = 0, 1, 2, . . . , a
n
b
n

We will show that L is a nonregular language by contradiction.

Suppose that there is some FA that accepts L.
By denition, this FA must have a nite number of states, say 5.
Consider the path the FA takes on the word a
6
b
6
.
The rst 6 letters of the word are as.
When processing the rst 6 letters, the FA must visit some state u at
least twice since there are only 5 states in the FA.
We say that the path has a circuit, which consists of those edges that
are taken from the rst time u is visited to the second time u is visited.
Suppose the circuit consists of 3 edges.
After the rst b is read, the path goes elsewhere and eventually we end
up in a nal state where the word a
6
b
6
is accepted.
Now consider the string a
6+3
b
6
.
When processing the a part of the string, the FA eventually hits state u.
From state u, we can take the circuit and return to u by using up 3 as.
CHAPTER 10. NONREGULAR LANGUAGES 10-4
From then on, we read in the rest of the as exactly as before and go on
to read in the 6 bs in the same way as before.
Thus, when processing a
6+3
b
6
, we end up again in a nal state.
Hence, we are supposed to accept a
9
b
6
.
However, a
9
b
6
is not in L since it does not have an equal number of as
and bs.
Thus, we have a contradiction, and so L must not be regular.
We can use the same argument with any string a
6
(a
3
)
k
b
6
, for k =
0, 1, 2, . . ..
10.3 First Version of Pumping Lemma
Theorem 13 (Pumping Lemma) Let L be any regular language that has
innitely many words. Then there exists some three strings x, y, and z such
that y is not the null string and that all strings of the form
xy
k
z for k = 0, 1, 2, . . .
are words in L.
Proof.
Since L is a regular language, there exists some FA that accepts L by
Kleenes theorem.
FA must have a nite number of states N.
Since L is an innite language and since alphabets are always nite, L
must consist of arbitrarily long words.
Consider any word w accepted by FA with length(w) = m, and assume
that m N.
Since length(w) = m, when processing w on the FA, we visit m + 1
states, not necessarily all unique.
CHAPTER 10. NONREGULAR LANGUAGES 10-5
Since m + 1 N + 1, when processing w on the FA, we visit at least
N + 1 states.
But since the FA has only N states in total, some state must be visited
twice when processing w on the FA.
Let u be the rst state visited twice when processing the string w on the
FA.
Thus, there is a circuit in FA corresponding to state u for this string w.
We break up w into three substrings x, y, z:
1. x consists of the all letters starting at the beginning of w up to those
consumed by the FA when state u is reached for the rst time. Note
that x may be null.
2. y consists of the letters after x that are consumed by the FA as it
travels around the circuit.
3. z consists of the letters after y to the end of w.
Note that the following statements hold:
1. w = xyz.
2. Note that y is not null since at least one letter is consumed by trav-
eling around the circuit. The circuit starts in a particular state, and
ends in the same state. Thus, traveling the circuit requires at least
one transition, which means that at least one letter is consumed.
3. The strings x and y satisfy length(x) + length(y) N, which we
can show as follows. Let v be the string xy except for the last letter
of xy. By the way that we constructed x and y, when we process
v on the FA starting in the initial state, we never visit any state
twice since it is only on reading the last letter of y do we rst visit
some state twice. Thus, processing v on the FA results in visiting at
most N states, which corresponds to reading at most N 1 letters.
Since xy is the same as v with one more letter attached, we must
have that xy has length at most N.
When processing w = xyz,
the FA rst processes substring x and ends in state u.
CHAPTER 10. NONREGULAR LANGUAGES 10-6
then it starts processing substring y starting in state u and ends in
state u.
then it starts processing substring z starting in state u and ends in
some nal state v.
Now process the word xyyz on FA.
For the substring x, the FA follows exactly the same path as when
it processed the x-part of w.
For the rst substring u, the FA starts in state u and returns to
state u.
For the second substring u, the FA starts in state u and returns to
state u.
For the substring z, the FA starts in u and processes exactly as
before for the word w, and so it ends in the nal state v.
Thus, xyyz is accepted by FA.
Similarly, we can show that any string xy
k
z, k = 0, 1, 2, . . ., is accepted
by FA.
10.4 Another Version of Pumping Lemma
Theorem 14 Let L be a language accepted by an FA with N states. Then for
all words w L such that length(w) N, there are strings x, y, and z such
that
P1. w = xyz;
P2. y is not null;
P3. length(x) + length(y) N;
P4. xy
k
z L for all k = 0, 1, 2, . . ..
CHAPTER 10. NONREGULAR LANGUAGES 10-7
Proof. The proof of Theorem 13 actually establishes Theorem 14.
Remarks:
In the textbook Theorem 14 also assumes that L is innite. However,
this additional assumption is not needed.
Example:
1- 2+ 3
4 5
a
a
a
b
a
b
a
b
b
b
w = ababbaa
x = ab
y = abb
z = aa
CHAPTER 10. NONREGULAR LANGUAGES 10-8
Example: Prove L = PALINDROME is nonregular.
We cannot use the rst version of the Pumping Lemma (Theorem 13) since
x = a, y = b, z = a,
satisfy the lemma and do not contradict the language since all words of the
form
xy
k
z = ab
k
a
are in PALINDROME.
We will instead apply Theorem 14 to show that PALINDROME is nonregular.
Proof.
Suppose that PALINDROME is a regular language.
Then by denition, PALINDROME must have a regular expression.
Kleenes Theorem then implies that there is a nite automaton for PALIN-
DROME.
Assume that the FA for PALINDROME has N states, for some N 1.
Consider the string
w = a
N
ba
N
which is in PALINDROME.
Note that length(w) = 2N + 1 N.
Thus, all of the assumptions of Theorem 14 hold, so the conclusions of
Theorem 14 must hold; i.e., there exist strings x, y, and z such that
P1. w = xyz;
P2. y is not null;
P3. length(x) + length(y) N;
P4. xy
k
z L for all k = 0, 1, 2, . . ..
P1 of Theorem 14 says that w = xyz, so
x must be at the beginning of w,
y must be somewhere in the middle of w,
CHAPTER 10. NONREGULAR LANGUAGES 10-9
z must be at the end of w.
P2 of Theorem 14 says that x and y together have at most N letters.
Since w has N as in the beginning and x and y are at the beginning of
w, x and y must consist solely of as.
P1 and P3 of Theorem 14 imply that x and y must consist solely of as.
Since z is the rest of the string after x and y, we must have that z consists
of zero or more as, followed by 1 b and then N as.
In other words,
x = a
i
for some i 0,
y = a
j
for some j 0,
z = a
l
ba
N
for some l 0.
Since y ,= by P2 of Theorem 14, we must have j 1.
Also, since w = xyz by P1 of Theorem 14, note that
w = a
N
ba
N
= xyz = a
i
a
j
a
l
ba
N
= a
i+j+l
ba
N
,
so i +j +l = N.
Now consider the string xyyz, which is supposed to be in L by P4 of
Theorem 14.
Note that
xyyz = a
i
a
j
a
j
a
l
ba
N
= a
i+2j+l
ba
N
= a
N+j
ba
N
since i +j +l = N.
But a
N+j
ba
N
, PALINDROME since reverse(a
N+j
ba
N
) ,= a
N+j
ba
N
.
This is a contradiction, and so PALINDROME must be nonregular.
Can use rst version of Pumping Lemma (Theorem 13) to show that L =
a
n
b
n
: n 0 is not a regular language:
CHAPTER 10. NONREGULAR LANGUAGES 10-10
Suppose L is a regular language.
Pumping Lemma says that there exist strings x, y, and z such that all
words of the form xy
k
z are in L, where y is not null.
All words in L are of the form a
n
b
n
.
How do we break up a
n
b
n
into substrings x, y, z with y nonempty?
If y consists solely of as, then xyyz has more as than bs, and so
it is not in L.
If y consists solely of bs, then xyyz has more bs than as, and so
it is not in L.
If y consists of as and bs, then all of the as in y must come before
all of the bs. However, xyyz then has some bs appearing before
some as, and so xyyz is not in L.
Thus, L is not a regular language.
Example:
Let = a, b.
For any string w

, dene n
a
(w) to be the number of as in w, and
n
b
(w) to be the number of bs in w.
Dene the language L = w

: n
a
(w) n
b
(w); i.e., L consists of
strings w for which the number of as in w is at least as large as the
number of bs in w.
For example, abbaa L since the string has 3 as and 2 bs, and 3 2.
We can prove that L is a nonregular language using the pumping lemma.
What string w L should we use to get a contradiction?
Example: Consider the language EQUAL =, ab, ba, aabb, abab, abba, baba, bbaa, . . .,
which consists of all words having an equal number of as and bs. We now
prove that EQUAL is a non-regular language.
CHAPTER 10. NONREGULAR LANGUAGES 10-11
We will prove this by contradiction, so suppose that EQUAL is a regular
language.
Note that a
n
b
n
: n 0 = a

EQUAL
Recall that the intersection of two regular languages is a regular lan-
guage.
Note that a

is a regular expression, and so its language is regular.

If EQUAL were a regular language, then a
n
b
n
: n 0 would be the
intersection of two regular languages.
This would imply that a
n
b
n
: n 0 is a regular language, which is not
true.
Thus, EQUAL must not be a regular language.
10.5 Prex Languages
Denition: If R and Q are languages, then Pref(Q in R) is the language of
the prexes of Q in R, which is the set of all strings of letters that can be
concatenated to the front of some word in Q to produce some word in R; i.e.,
Pref(Q in R) = strings p : q Q such that pq R
Example: Q = aba, aaabb, baaaba, bbaaaabb, aaaa
R = baabaaba, aaabb, abbabbaaaabb
Pref(Q in R) = baaba, , abbabba, abba
Example: Q = aba, aaabb, baaaba, bbaaaabb, aaaa
R = baab, ababb
Pref(Q in R) =
Example: Q = ab

a
R = (ba)

Pref(Q in R) = (ba)

b
CHAPTER 10. NONREGULAR LANGUAGES 10-12
Theorem 16 If R is a regular language and Q is any language whatsoever,
then the language
P = Pref(Q in R)
is regular.
Proof.
Since R is a regular language, it has some nite automaton FA
1
that
accepts it.
FA
1
has one start state and several (possibly none or one) nal states.
For each state s in FA
1
, do the following:
Using s as the start state, process all words in the language Q on
FA
1
.
When starting s, if some word in Q ends in the nal state of FA
1
,
then paint state s blue.
So for each state s in FA
1
that is painted blue, there exists some word
in Q that can be processed on FA
1
starting from s and end up in a nal
state.
Now construct another machine FA
2
:
FA
2
has the same states and arcs as FA
1
.
The start state of FA
2
is the same as that of FA
1
.
The nal states of FA
2
are the ones that were previously painted
blue (regardless if they were nal states in FA
1
).
We will now show that FA
2
accepts exactly the prex language
P = Pref(Q in R).
To prove this, we have to show two things:
Every word in P is accepted by FA
2
.
Every word accepted by FA
2
is in P.
First, we show that every word accepted by FA
2
is in P.
CHAPTER 10. NONREGULAR LANGUAGES 10-13
Consider any word w accepted by FA
2
.
Starting in the start state of FA
2
, process the word w on FA
2
, and
we end up in a nal state of FA
2
.
Final states of FA
2
were painted blue.
Now we can start from here and process some word from Q and end
up in a nal state of FA
1
.
Thus, the word w P.
Now we prove that every word in P is accepted by FA
2
.
Consider any word p P.
By denition, there exists some word q Q and a word w R such
that pq = w.
This implies that if pq is processed on FA
1
, then we end up in a
nal state of FA
1
.
When processing the string pq on FA
1
, consider the state s we are in
just after nishing processing p and at the beginning of processing
q.
State s must be a blue state since we can start here and process q
and end in a nal state.
Hence, by processing p, we must start in the start state and end in
state s.
Thus, p is accepted by FA
2
.
Chapter 11
Decidability for Regular
Languages
11.1 Introduction
We have three basic questions to answer:
1. How can we tell if two regular expressions dene the same language?
2. How can we tell if two FAs are equivalent?
3. How can we tell if the language dened by an FA has nitely many or
innitely many words in it?
Note that questions 1 and 2 are essentially the same by Kleenes Theorem.
11.2 Decidable Problems
Denition: A problem is eectively solvable if there is an algorithm that
provides the answer in a nite number of steps, no matter what the particular
inputs are (but may depend on the size of the problem).
The maximum number of steps the algorithm will take must be predictable
before we begin executing the procedure.
11-1
CHAPTER 11. DECIDABILITY FOR REGULAR LANGUAGES 11-2
Example: Problem: nd roots of quadratic equation ax
2
+bx +c = 0.
Solution: use quadratic equation
x =
b

b
2
4ac
2a
No matter what the coecients a, b, and c are, we can compute the solution
using the following operations:
four multiplications
two subtractions
one square root
one division
Another solution: keep guessing until we nd a root.
This approach is not guaranteed to nd root in a xed number of steps.
Example: Find the maximum of n numbers. An eective solution for this is
to scan through the list once while updating the maximum observed thus far.
This takes O(n) steps.
Denition: An eective solution to a problem that has a yes or no answer is
called a decision procedure. A problem that has a decision procedure is called
decidable.
11.2.1 Is L
1
= L
2
?
Determine if two languages L
1
and L
2
are the same:
Method 1: Check if the language
L
3
(L
1
L

2
) + (L

1
L
2
)
has any words (even ).
If L
1
= L
2
, then L
3
= .
CHAPTER 11. DECIDABILITY FOR REGULAR LANGUAGES 11-3
If L
1
,= L
2
, then L
3
,= .
Example: Suppose L
1
= a, aa and L
2
= a, aa, aaa. Then L
1
L

2
=
, but L

1
L
2
= aaa. Thus, L
1
,= L
2
.
So now we have reduced the problem of determining if L
1
= L
2
to
determining if L
3
= .
11.2.2 Is L = ?
So we need a method for determining if a regular language is empty.
Since the language is regular, it has a regular expression and a FA.
Given a regular expression, check if there is any part that is not concate-
nated with .
Specically, use the following algorithm to determine if L = given a
regular expression r for L:
Method 1 (for deciding if a language L = given regular expression r
for L):
Write r as
r = r
1
+r
2
+ +r
n
,
where for each i = 1, 2, . . . , n, r
i
= r
i,1
r
i,2
r
i,j
i
for some j
i
1;
i.e., r is written as a sum of other regular expressions r
i
, i =
1, 2, . . . , n, where each r
i
is a concatenation of regular expressions.
It is always possible to write any regular expression r in this form.
If there exists some i = 1, 2, . . . , n such that r
i,j
,= for all 1 j
j
i
, then L ,= . In other words, if one of the summands has none of
its factors being , then the language L is not empty.
If for each i = 1, 2, . . . , n, at least one of r
i,1
, r
i,2
, . . . , r
i,j
i
is , then
L = . In other words, if each of the summands has at least one
factor being , then the language L is empty.
Example: The regular expression
(b +a)

+b
CHAPTER 11. DECIDABILITY FOR REGULAR LANGUAGES 11-4
has the last b not concatenated with so the language is not empty.
Example: The regular expression
(b +a)

+b
has all parts concatenated with so the language is empty.
Remarks: The algorithm in the book for determining if L = given a
regular expression for L is incorrect.
Method 2 (for deciding if a language L = ): Given an FA, we check
if there are any paths from to some + state by using the blue paint
algorithm:
1. Paint the start state blue.
2. From every blue state, follow each edge that leads out of it and paint
the connecting state blue, then delete this edge from the machine.
3. Repeat Step 2 until no new state is painted blue, then stop.
4. When the procedure has stopped, if any of the nal states are
painted blue, then the machine accepts some words, and if not,
the machine accepts no words.
Remarks on Method 2:
The above algorithm will iterate Step 2 at most N times, where N
is the number of states in the machine.
Thus, it is a decision procedure.
CHAPTER 11. DECIDABILITY FOR REGULAR LANGUAGES 11-5
Example:
- +
a
a
b
b
a
a
b
a
b
+
a
a
b
b
a
a
b
a
b
+
b
a
a
b
a
b
+
a
b
a
b
=>
=>
=>
=>
blue
blue
blue
blue blue
b
blue
blue
blue
blue
blue
blue
b
b
b
a
a
a
a
a
b
b
b
b
blue
+
a
b
a
-
-
-
-
CHAPTER 11. DECIDABILITY FOR REGULAR LANGUAGES 11-6
Theorem 17 Let F be an FA with N states. Then if F accepts any
strings at all, it accepts some string with N 1 or fewer letters.
Proof.
Consider any string w that is accepted by F.
Let s = w and DONE = NO.
Do while (DONE == NO)
Trace path of s through F.
If no circuits in path, then set DONE = YES.
If there are circuits in the path, then
Eliminate rst circuit in the path.
Let s be the string resulting from the new path.
Resulting path:
Starts in initial state.
Ends in a nal state.
Has no circuits, so visits at most N states.
This corresponds to a string of at most N 1 letters.
String is accepted by FA.
Method 3 (for deciding if a language L = ): Test all words with N 1
or fewer letters by running them on the FA.
CHAPTER 11. DECIDABILITY FOR REGULAR LANGUAGES 11-7
Example: Consider the languages L
1
and L
2
with FAs:
a
b
b
a
a
a, b
a
b
b
a
b
b
a
a
a, b
a
b
b
x1 +
-
x2 +
x3 +
x4 x5 +
y1+
-
y2+ y3
b a
a, b b a
FA for both
L1 L2 and
L1 L2
FA1: FA2:
x5, y2
x1, y1-
x2, y1
x3, y2
x4, y3
CHAPTER 11. DECIDABILITY FOR REGULAR LANGUAGES 11-8
Theorem 18 There are eective procedures to decide whether:
1. A given FA accepts any words.
2. Two FAs are equivalent; i.e., the two FAs accept the same language.
3. Two regular expressions are equivalent; i.e., the two regular expressions
generate the same language.
Remarks:
We can establish part 3 of Theorem 18 by rst converting the regular
expressions into FAs.
We previously saw an eective procedure for doing this in the proof of
Kleenes Theorem.
Then we just developed an eective procedure to decide whether two
FAs are equivalent.
11.2.3 Is L innite?
Determining if a language L is innite
If we have a regular expression for L, then all we need to do is check if
the is applied to some part of the regular expression that is not nor
.
Note that

= and

= .
Note that a

is innite.
Theorem 19 Let F be an FA with N states. Then
1. If F accepts an input string w such that
N length(w) < 2N
then F accepts an innite language.
2. If F accepts innitely many words, then F accepts some word w such
that
N length(w) < 2N
CHAPTER 11. DECIDABILITY FOR REGULAR LANGUAGES 11-9
Proof.
1. Assume that F accepts an input string w such that
N length(w) < 2N
Since length(w) N, the second version of the pumping lemma
(Theorem 14) implies that there exist substrings x, y, and z such
that y ,= and xy
n
z, n = 0, 1, 2, . . ., are all accepted by F.
Thus, the FA accepts innitely many words.
2. Assume that F accepts innitely many words.
This implies that there exists some word u accepted by F that has
a circuit (possibly more than one). Why?
Each circuit can consist of at most N states since F has only N
states.
Iteratively eliminate the rst circuit in the path until only one cir-
cuit left (as in the proof of Theorem 17).
Let v correspond to the word from this one-circuit path, and note
that v is accepted by F.
We can write v as the concatenation of three strings x, y, and z,
i.e.,
v = xyz,
such that
x consists of the letters read before the circuit.
y consists of the letters read along the circuit.
z consists of the letters read after the circuit
We can show that
0 < length(y) N
as follows:
Since we have eliminated all but the rst circuit, the circuit
starts and ends in the same state and all of the other states are
unique.
Thus, the circuit can visit at most N + 1 states (with at most
one state repeated).
This corresponds to reading at most N letters.
CHAPTER 11. DECIDABILITY FOR REGULAR LANGUAGES 11-10
Also, since a circuit corresponds to at least one transition and
each transition in an FA uses up exactly one letter, we see that
length(y) > 0.
We can show that
length(x) + length(z) < N
as follows:
Since we constructed the string v by eliminating all but the
rst circuit, the paths followed by processing x and z have no
circuits.
Thus, all of the states visited along the paths followed by pro-
cessing x and z are unique.
Hence, the paths followed by processing x and z visit at most
N states.
This means that length(x) + length(z) N 1 < N.
Thus,
length(v) = length(x) + length(y) + length(z) N 1 +N < 2N.
If v has at least N letters, then we are done.
If v has less than N letters, then we can pump up the cycle some
number of times to obtain a word that has the desired characteris-
tics since 0 < length(y) N.
CHAPTER 11. DECIDABILITY FOR REGULAR LANGUAGES 11-11
Example:
1-
2
3
4
5
6+ a b a b b
a
b
a
b
b
a
a
Consider the word w = abaaaababbabb
length(w) = 13 > 2N = 12.
w is accepted by the FA.
Processing w on FA takes the path
1 2 5 3

circuit 1
2 4 5

circuit 2
4 5

circuit 3
4 6 5

circuit 4
4 6
Bypassing all but the rst circuit yields the path
1 2 5 3 2 4 6
which corresponds to the word abaaab, which has length 6.
Thus, Theorem 19 implies that the FA accepts an innite language.
Consider the word w = bbaabb
length(w) = 6 = N
w is accepted by the FA.
CHAPTER 11. DECIDABILITY FOR REGULAR LANGUAGES 11-12
Processing w on FA takes the path
1 3

circuit 1
3 2 4 6

circuit 2
6
Bypassing all but the rst circuit yields the path
1 3

circuit 1
3 2 4 6
which corresponds to the word bbaab, which has length 5.
However, we can go around the circuit one more time, yielding the path
1 3

circuit 1
3

circuit 1
3 2 4 6
which corresponds to the word bbbaab, which has length 6.
CHAPTER 11. DECIDABILITY FOR REGULAR LANGUAGES 11-13
Theorem 20 There is an eective procedure to decide whether a given FA
accepts a nite or an innite language.
Proof.
Suppose that the FA has N states.
Suppose that the alphabet consists of m letters.
Then by Theorem 19, we only need to check all strings w with
N length(w) < 2N
to determine if FA accepts an innite language.
If any of these are accepted, then the FA accepts an innite language.
Otherwise, it accepts a nite language.
The number of strings w satisfying
N length(w) < 2N
is
m
N
+m
N+1
+m
N+2
+ +m
2N1
which is nite.
Thus, checking all of these strings is an eective procedure.
Chapter 12
Context-Free Grammars
12.1 Introduction
English grammar has rules for constructing sentences; e.g.,
1. A sentence can be a subject followed by a predicate .
2. A subject can be a noun-phrase .
3. A noun-phrase can be an adjective followed by a noun-phrase .
4. A noun-phrase can be an article followed by a noun-phrase .
5. A noun-phrase can be a noun .
6. A predicate can be a verb followed by a noun-phrase .
7. A noun can be:
person sh stapler book
8. A verb can be:
buries touches grabs eats
9. An adjective can be:
big small
10. An article can be:
the a an
12-1
CHAPTER 12. CONTEXT-FREE GRAMMARS 12-2
These rules can be used to construct the following sentence:
The small person eats the big sh
sentence subject predicate Rule 1
noun-phrase predicate Rule 2
noun-phrase verb noun-phrase Rule 6
article noun-phrase verb noun-phrase Rule 4
article adjective noun-phrase verb noun-phrase Rule 3
article adjective noun verb noun-phrase Rule 5
article adjective noun verb article noun-phrase Rule 4
article adjective noun verb article adjective noun-phrase Rule 3
article adjective noun verb article adjective noun Rule 5
the adjective noun verb article adjective noun Rule 10
the small noun verb article adjective noun Rule 9
the small person verb article adjective noun Rule 7
the small person eats article adjective noun Rule 8
the small person eats the adjective noun Rule 10
the small person eats the big noun Rule 9
the small person eats the big sh Rule 7
Denition: The things that cannot be replaced by anything are called ter-
minals.
Denition: The things that must be replaced by other things are called
nonterminals.
In the above example,
small and eats are terminals.
noun-phrase and verb are nonterminals.
CHAPTER 12. CONTEXT-FREE GRAMMARS 12-3
Example: restricted class of arithmetic expressions on integers.
start AE
AE AE + AE
AE AE AE
AE AE AE
AE AE / AE
AE AE AE
AE (AE )
AE AE
AE ANY-NUMBER
nonterminals: start , AE
terminals: ANY-NUMBER , +, , , /, , (, )
Can generate the arithmetic expression
ANY-NUMBER +(ANY-NUMBER ANY-NUMBER )/ANY-NUMBER
as follows:
start AE
AE + AE
AE + AE / AE
AE + (AE ) / AE
AE + (AE AE ) / AE
ANY-NUMBER + (AE AE ) / AE
ANY-NUMBER + (ANY-NUMBER AE ) / AE
ANY-NUMBER + (ANY-NUMBER ANY-NUMBER ) / AE
ANY-NUMBER + (ANY-NUMBER ANY-NUMBER ) / ANY-NUMBER
CHAPTER 12. CONTEXT-FREE GRAMMARS 12-4
Could also make ANY-NUMBER a nonterminal:
Rule 1 ANY-NUMBER FIRST-DIGIT
Rule 2 FIRST-DIGIT FIRST-DIGIT OTHER-DIGIT
Rule 3 FIRST-DIGIT 1 2 3 4 5 6 7 8 9
Rule 4 OTHER-DIGIT 0 1 2 3 4 5 6 7 8 9
In this case,
nonterminals: ANY-NUMBER , FIRST-DIGIT , OTHER-DIGIT
terminals: 0, 1, 2, 3, 4, 5, 6, 7, 8, 9
Can produce the number 90210 as follows:
Rule 1 ANY-NUMBER FIRST-DIGIT
Rule 2 FIRST-DIGIT OTHER-DIGIT
Rule 2 FIRST-DIGIT OTHER-DIGIT OTHER-DIGIT
Rule 2 FIRST-DIGIT OTHER-DIGIT OTHER-DIGIT OTHER-DIGIT
Rule 2 FIRST-DIGIT OTHER-DIGIT OTHER-DIGIT OTHER-DIGIT OTHER-DIGIT
Rule 3 9 OTHER-DIGIT OTHER-DIGIT OTHER-DIGIT OTHER-DIGIT
Rule 4 9 0 OTHER-DIGIT OTHER-DIGIT OTHER-DIGIT
Rule 4 9 0 2 OTHER-DIGIT OTHER-DIGIT
Rule 4 9 0 2 1 OTHER-DIGIT
Rule 4 9 0 2 1 0
Note that we had rules of the form:
one nonterminal string of nonterminals
or
one nonterminal choice of terminals
Denition: The sequence of applications of the rules that produces the
nished string of terminals from the starting symbol is called a derivation or
production.
CHAPTER 12. CONTEXT-FREE GRAMMARS 12-5
12.2 Context-Free Grammars
Example: terminals: = a
nonterminal: = S
productions:
S aS
S
Can generate a
4
as follows:
S aS
aaS
aaaS
aaaaS
aaaa = aaaa
Example: terminal: a
nonterminal: S
productions:
S SS
S a
S
Can write this in more compact notation:
S SS [ a [
which is called the Backus Normal Form or Backus-Naur Form (BNF).
CFL is a

Can generate a
2
as follows:
S SS
SSS
SSa
SSSa
SaSa
aSa
aa = aa
CHAPTER 12. CONTEXT-FREE GRAMMARS 12-6
In previous example, unique way to generate any word.
Here, each word in CFL has innitely many derivations.
Denition: Acontext-free grammar (CFG) is a collection G = (, , R, S),
with
1. A (nite) alphabet of letters called terminals from which we make
strings that will be the words of the language.
2. A nite set of symbols called nonterminals, one of which is the
symbol S (i.e., S ), standing for start here.
3. A nite set R of productions, with R (+)

. If a production
(N, |) R with N and | ( + )

, then we write the

production as
N |.
Thus, each production is of the form
one nonterminal nite string of terminals and/or nonterminals
where the strings of terminals, nonterminals can consist of only
terminals or of only nonterminals, or any mixture of terminals and
nonterminals or even the empty string. We require that at least one
production has the nonterminal S as its left side.
Convention:
Terminals will typically be smallcase letters.
Nonterminals will typically be uppercase letters.
Denition: The language generated (dened, derived, produced) by a
CFG G is the set of all strings of terminals that can be produced from
the start symbol S using the productions as substitutions. A language
generated by a CFG G is called a context-free language (CFL) and is
denoted by L(G).
Example: terminals: = a
nonterminal: = S
productions:
S aS
S
CHAPTER 12. CONTEXT-FREE GRAMMARS 12-7
Let L
1
the language generated by this CFG, and let L
2
be the
language generated by regular expression a

.
Claim: L
1
= L
2
.
Proof:
We rst show that L
2
L
1
.
Consider a
n
L
2
for n 1. We can generate a
n
by using
rst production n times, and then second production.
Can generate L
2
by using second production only.
Hence L
2
L
1
.
We now show that L
1
L
2
.
Since a is the only terminal, CFG can only produce strings
having only as.
Thus, L
1
L
2
.
Note that
Two types of arrows:
used in statement of productions
used in derivation of word
in the above derivation of a
4
, there were many unnished stages
that consisted of both terminals and nonterminals. These are called
working strings.
is neither a nonterminal (since it cannot be replaced with some-
thing else) nor a terminal (since it disappears from the string).
12.3 Examples
Example: terminals: a, b
nonterminals: S
productions:
S aS
S bS
S a
S b
CHAPTER 12. CONTEXT-FREE GRAMMARS 12-8
More compact notation:
S aS [ bS [ a [ b
Can produce the word abbab as follows:
S aS
abS
abbS
abbaS
abbab
Let L
1
be the CFL, and let L
2
be the language generated by the
regular expression (a +b)
+
.
Claim: L
1
= L
2
.
Proof:
First we show that L
2
L
1
.
Consider any string w L
2
.
Read letters of w from left to right.
For each letter read in, if it is not the last, then
use the production S aS if the letter is a or
use the production S bS if the letter is b
For the last letter of the word,
use the production S a if the letter is a or
use the production S b if the letter is b
In each stage of the derivation, the working string has the form
(string of terminals)S
Hence, we have shown how to generate w using the CFG, which
means that w L
1
.
Hence, L
2
L
1
.
Now we show that L
1
L
2
.
To show this, we need to show that if w L
1
, then w L
2
.
This is equivalent to showing that if w , L
2
, then w , L
1
.
CHAPTER 12. CONTEXT-FREE GRAMMARS 12-9
Note that the only string w , L
2
is w = .
But note that cannot be generated by the CFG, so , L
1
.
Hence, we have proven that L
1
L
2
.
Example: terminals: a, b
nonterminals: S, X, Y
productions:
S X [ Y
X
Y aY [ bY [ a [ b
Note that if we use rst production (S X), then the only word we
can generate is .
The second production (S Y ) leads to a collection of productions
identical to the previous example.
Thus, the second production produces (a +b)
+
.
CFL is (a +b)

Example: terminals: a, b
nonterminals: S
productions:
S aS [ bS [ a [ b [
CFL is (a +b)

For this CFG, the sequence of productions to generate any word is not
unique.
e.g., can generate bab using
S bS
baS
babS
bab = bab
CHAPTER 12. CONTEXT-FREE GRAMMARS 12-10
or
S bS
baS
bab
Example: terminals: a, b
nonterminals: S, X
productions:
S XaaX
X aX [ bX [
The last set of productions generates any word from

.
CFL is (a +b)

aa(a +b)

Can generate abbaaba as follows:

S XaaX
aXaaX
abXaaX
abbXaaX
abbaaX = abbaaX
abbaabX
abbaabaX
abbaaba = abbaaba
Example: terminals: a, b
nonterminals: S, X, Y
productions:
S XY
X aX [ bX [ a
Y Y a [ Y b [ a
X productions can produce words ending with a.
CHAPTER 12. CONTEXT-FREE GRAMMARS 12-11
Y productions can produce words starting with a.
CFL is (a +b)

aa(a +b)

Can generate abbaaba as follows:

S XY
aXY
abXY
abbXY
abbaY
abbaY a
abbaY ba
abbaaba
Example: Give CFGs for each of the following languages over the alphabet
= a, b:
1. a
n
b
n
: n 0
2. PALINDROME
3. EVEN-PALINDROME
4. ODD-PALINDROME
Example: terminals: a, b
nonterminals: S, B, U
productions:
S SS [ BS [ SB [ [ USU
B aa [ bb
U ab [ ba
Show that this generates EVEN-EVEN
Note that starting from B, we can generate a balanced pair, i.e., either
aa or bb.
CHAPTER 12. CONTEXT-FREE GRAMMARS 12-12
Starting from U, we can generate an unbalanced pair, i.e., either ab or
ba.
First show that every word in EVEN-EVEN can be generated using these
productions.
Recall that EVEN-EVEN has regular expression
[aa +bb + (ab +ba)(aa +bb)

(ab +ba)]

Three types of syllables:

1. aa,
2. bb,
3. (ab +ba)(aa +bb)

(ab +ba)
Consider any word generated from the regular expression for EVEN-
EVEN. Lets examine the way it was generated using the regular
expression, and show how to generate the same word using our
CFG.
Start our derivation using the CFG from S.
Every time we iterate the outer star in the regular expression, we
choose one of the three syllables.
1. If we choose a syllable of type 1, then rst use the production
S BS and then the production B aa. Thus, we end up
with a working string of aaS for this iteration of the outer star.
2. If we choose a syllable of type 2, then rst use the production
S BS and then the production B bb. Thus, we end up
with a working string of bbS for this iteration of the outer star.
3. If we choose a syllable of type 3, then
(a) First use the production S SS.
(b) Then change the rst S using the production S USU,
resulting in USUS.
(c) If the rst (ab+ba) in the syllable (ab+ba)(aa+bb)

(ab+
ba) is used to generate ab, then replace the rst U in USUS
using the production U ab, resulting in abSUS. If the
rst (ab + ba) in (ab + ba)(aa + bb)

(ab + ba) is used

to generate ba, then replace the rst U in USUS using the
production U ba, resulting in baSUS. Do the same for
the second (ab+ba) in (ab+ba)(aa+bb)

(ab+ba). Thus,
CHAPTER 12. CONTEXT-FREE GRAMMARS 12-13
we now have xSyS as a working string for this iteration of
the outer star of the regular expression, where x is either
ab or ba, and y is either ab or ba.
(d) Now suppose the (aa + bb)

is iterated n times, n 0. If
n = 0, then change the rst S in xSyS using the production
S , resulting in xyS = xyS. If n 1, then change
the rst S in xSyS using the production S BS and do
this n times, resulting in xBBB BSyS, where there are
n Bs in the clump of Bs. Then change the rst S using
the production S , resulting in xBBB ByS =
xBBB ByS, where there are n Bs in the clump of Bs.
Finally, if on the kth iteration, k n, of the in (aa +
bb)

we generated aa, then replace the kth B using the

production B aa. If on the kth iteration, k n, of the
in (aa + bb)

we generated bb, then replace the kth B

using the production B bb.
After completing all of the iterations of the outer star in the regular
expression, use the production S .
e.g., for word babbabaa EVEN-EVEN,
S SS
USUS
baSUS
baBSUS
babbSUS
babbUS = babbUS
babbabS
babbabBS
babbabaaS
babbabaa = babbabaa
Now show that all words generated by these productions are in EVEN-
EVEN.
all words derived from S can be decomposed into two-letter sylla-
bles.
unbalanced syllables (ab and ba) come into working string in pairs,
which adds two as and two bs.
CHAPTER 12. CONTEXT-FREE GRAMMARS 12-14
balanced syllables add two of one letter and none of the other
thus, the sum total of as will be even, and the sum total of bs will
be even
Thus, word generated by productions will be in EVEN-EVEN.
Example: terminals: a, b
nonterminals: S, A, B
productions:
S aB [ bA
A a [ aS [ bAA
B b [ bS [ aBB
This generates the language EQUAL, which consists of all strings of positive
length and that have an equal number of as and bs.
Proof. Need to show two things:
1. every word in EQUAL can be generated using our productions.
2. every word generated by our productions is in EQUAL.
First we show 1.
We make three claims:
Claim 1: All words in EQUAL can be generated by some sequence of
productions beginning with the start symbol S.
Claim 2: All words that have one more a than bs can be generated
from these productions by starting with the nonterminal A.
Claim 3: All words that have one more b than as can be generated
from these productions by starting with the nonterminal B.
We will prove that these three claims hold by contradiction.
Assume that one of the three claims does not hold.
Then there is some smallest word w that violates one of the claims.
All words shorter than w must satisfy the three claims.
CHAPTER 12. CONTEXT-FREE GRAMMARS 12-15
First assume that w violates Claim 1.
This means that w is in EQUAL but cannot be generated starting
with S.
Assume that w starts with a and that w = aw
1
.
Since w EQUAL, w
1
must have exactly one more b than as.
However, w
1
is shorter than w.
Thus, we must be able to generate w
1
starting with B; i.e.,
B w
1
But then
S aB aw
1
= w
which is a contradiction.
We similarly reach a contradiction when the rst letter of w is b.
Thus, w cannot violate Claim 1.
Now assume that w violates Claim 2.
This means that w has one more a than bs but cannot be generated
starting with A.
First assume that w starts with a.
Then w = aw
1
, where w
1
EQUAL.
Since w
1
is shorter than w, we must be able to generate w
1
starting with S; i.e.,
S w
1
But then
A aS aw
1
= w
which is a contradiction.
Now assume that w starts with b.
Then if we write w = bw
1
, then w
1
has two more as than bs.
We now split w
1
= w
11
w
12
, where w
11
is the part of w
1
scanning
from left to right until there is exactly one more a than bs, and
let w
12
be the rest of w
1
.
Note that w
12
also has exactly one more a than bs.
CHAPTER 12. CONTEXT-FREE GRAMMARS 12-16
Since w
11
and w
12
are both shorter than w, we must be able to
generate each of them starting with A; i.e.,
A w
11
and
A w
12
But then
A bAA bw
11
w
12
= bw
1
= w
which is a contradiction.
Thus we have shown that Claim 2 must hold.
We can similarly show that Claim 3 must hold.
Thus, all 3 claims hold, and so in particular, Claim 1 holds: all words in
EQUAL can be generated starting from S.
Now we show 2 holds: every word generated by our productions is in EQUAL.
We again make 3 claims
Claim 4 All words generated from S are in EQUAL.
Claim 5 All words generated from A have one more a than bs.
Claim 6 All words generated from B have one more b than as.
We will show that these 3 claims hold by contradiction.
Assume that one of the three claims does not hold.
Then there is some smallest word w generated from S, A, or B that does
not have the required property.
All words shorter than w must satisfy the three claims.
First assume that w violates Claim 4.
We have assumed that w can be generated from S but is not in
EQUAL.
Assume that the rst letter of w is a.
CHAPTER 12. CONTEXT-FREE GRAMMARS 12-17
Then w was generated by rst using the production S aB.
To generate w, this B generates a word w
1
which is shorter than w
and by assumption w
1
has one more b than as.
This implies that w has an equal number of as and bs, which is a
contradiction.
We get a similar contridiction if the rst letter of w is b.
Now assume that w violates Claim 5.
We have assumed that w can be generated from A but does not
have exactly one more a than bs.
w could not have been generated by A a since w = a, which
satises the requirement.
Suppose w was generated by rst using the production A aS.
Then to generate the rest of w, we would have to start from S
to generate w
1
, where w = aw
1
.
However, since w
1
is shorter than w and w
1
is generated starting
with S, we must have that w
1
EQUAL.
This implies that w has exactly one more a than bs, which is
a contradiction.
Suppose w was generated by rst using the production A bAA.
To generate the rest of w, each of the As need to generate
strings w
1
and w
2
which are shorter than w such that w =
bw
1
w
2
.
However, since w
1
and w
2
are shorter than w, we must have
that w
1
and w
2
each have exactly one more a than bs.
Hence, w = bw
1
w
2
must have exactly one more a than bs,
which is a contradiction.
Thus, we have shown that Claim 5 must hold
We can similarly show that Claim 6 must hold.
Thus, all of the claims hold, and in particular, Claim 4: all words gen-
erated from S EQUAL.
CHAPTER 12. CONTEXT-FREE GRAMMARS 12-18
12.4 Trees
Can use a tree to illustrate how a word is derived from a CFG.
Denition: These trees are called syntax trees, parse trees, generation trees,
production trees, or derivation trees.
Example: CFG:
terminals: a, b
nonterminals: S, A
productions:
S AAA [ A
A AA [ aA [ Ab [ a [ b
String abaaba has the following derivation:
S AAA
aAAA
abAA
abAbA
abaAbA
abaabA
abaaba
which corresponds to the following derivation tree:
S
/ | \
/ | \
/ | \
/ | \
A A A
/ | | \ |
/ | | \ |
a A A b a
| / \
CHAPTER 12. CONTEXT-FREE GRAMMARS 12-19
| / \
b a A
|
|
a
Example: CFG for simplied arithmetic expressions.
terminals: +, , 0, 1, 2, . . . , 9
nonterminals: S
productions:
S S +S [ S S [ 0 [ 1 [ 2 [ [ 9
Consider the expression 2 3 + 4.
Ambiguous how to evaluate this:
Does this mean (2 3) + 4 = 10 or 2 (3 + 4) = 14 ?
Can eliminate ambiguity by examining the two possible derivation trees
S S
/ | \ / | \
/ | \ / | \
/ | \ / | \
/ | \ / | \
S + S S * S
/ | \ | | / | \
/ | \ | | / | \
2 * 3 4 2 3 + 4
Eliminate the Ss as follows:
+ *
/ \ / \
/ \ / \
/ \ / \
/ \ / \
* 4 2 +
/ \ / \
/ \ / \
2 3 3 4
CHAPTER 12. CONTEXT-FREE GRAMMARS 12-20
Note that we can construct a new notation for mathematical expressions:
start at top of tree
walk around tree keeping left hand touching tree
rst time hit each terminal, print it out.
This gives us a string which is in operator prex notation or Polish notation.
In above examples,
rst tree yields
+ 2 3 4
second tree yields
2 + 3 4
To evaluate the string:
1. scan string from left to right.
2. the rst time we read a substring of the form operator-operand-operand
(o-o-o), replace the three symbols with the one result of the indicated
arithmetic calculation.
3. go back to step 1
CHAPTER 12. CONTEXT-FREE GRAMMARS 12-21
Example: (from above)
rst tree yields:
string rst o-o-o substring
+ 2 3 4 2 3
+ 6 4 + 6 4
10
second tree yields:
string rst o-o-o substring
2 + 3 4 + 3 4
2 7 2 7
14
Example: Consider the arithmetic expression:
3 + 4 6 + 2 + 8 + 1 5 + 9 7
There are many ways to evaluate this expression, one of which is as
((3 + 4) (6 + 2) + ((8 + 1) 5) + 9) 7
This interpretation has
derivation tree:
*
/ \
+ 7
/ \
+ 9
/ \
/ \
/ \
* *
/ \ / \
/ \ / \
+ + + 5
/ \ / \ / \
3 4 6 2 8 1
CHAPTER 12. CONTEXT-FREE GRAMMARS 12-22
prex notation:
+ + + 3 4 + 6 2 + 8 1 5 9 7
can evaluate prex notation expression:
string rst o-o-o substring
+ + + 3 4 + 6 2 + 8 1 5 9 7 + 3 4
+ + 7 + 6 2 + 8 1 5 9 7 + 6 2
+ + 7 8 + 8 1 5 9 7 7 8
+ + 56 + 8 1 5 9 7 + 8 1
+ + 56 9 5 9 7 9 5
+ + 56 45 9 7 + 56 45
+ 101 9 7 + 101 9
110 7 110 7
770
Example:
terminals: a, b
nonterminals: S, A, B
productions:
S AB
A a
B b
Can produce word ab in two ways:
1. S AB aB ab
2. S AB Ab ab
However, both derivations have the same syntax tree:
S
/ \
A B
| |
a b
CHAPTER 12. CONTEXT-FREE GRAMMARS 12-23
Denition: A CFG is ambiguous if for at least one word in its CFL there are
two possible derivations of the word that correspond to two dierent syntax
trees.
Example: PALINDROME
terminals: a, b
nonterminals: S
productions:
S aSa [ bSb [ a [ b [
Can generate the word babbab as follows:
S bSb
baSab
babSbab
babbab
which has derivation tree:
S
/|\
b S b
/|\
a S a
/|\
b S b
|
^
Can show that this CFG is unambiguous.
CHAPTER 12. CONTEXT-FREE GRAMMARS 12-24
Example:
terminals: a, b
nonterminals: S
productions:
S aS [ Sa [ a
The word aa can be generated by two dierent trees:
S S
/ \ / \
a S S a
| |
a a
Therefore, this CFG is ambiguous.
Example: terminals: a, b
nonterminals: S
productions:
S aS [ a
The CFL for this CFG is the same as above.
The word aa can now be generated by only one tree:
S
/ \
a S
|
a
Therefore, this CFG is unambiguous.
CHAPTER 12. CONTEXT-FREE GRAMMARS 12-25
Example:
terminals: a, b
nonterminals: S, X
productions:
S aS [ aSb [ X
X Xa [ a
The word aa has two dierent derivations that correspond to dierent syntax
trees:
1. S aS aX aa
S
/ \
a S
|
X
|
a
2. S X Xa aa
S
|
X
/ \
X a
|
a
Thus, this CFG is ambiguous.
CHAPTER 12. CONTEXT-FREE GRAMMARS 12-26
Denition: For a given CFG, the total language tree is the tree
with root S,
whose children are all the productions of S,
whose second descendents are all the working strings that can be con-
structed by applying one production to the leftmost nonterminal in each
of the children,
and so on.
Example:
terminals: a, b
nonterminals: S, X
productions:
S aX [ Xa [ aXbXa
X ba [ ab
This CFG has total language tree as follows:
S
/ | \
/ | \
/ | \
/ | \
/ | \
aX Xa aXbXa
/ | / | / \
/ | / | / \
aba aab baa aba ababXa aabbXa
/ \ / \
ababbaa abababa aabbbaa aabbaba
The CFL is nite.
CHAPTER 12. CONTEXT-FREE GRAMMARS 12-27
Example:
terminals: a, b
nonterminals: S, X
productions:
S aSb [ aX
X bX [ a
Total language tree:
S
/ \
/ \
/ \
/ \
/ \
/ \
aSb aX
/ \ / \
/ \ / \
aaSbb aaXb abX aa
/ \ / \ / \
aaaSbbb aaaXbb aabXb aaab abbX aba
. . . .
. . . .
. . . .
CFL is innite.
CHAPTER 12. CONTEXT-FREE GRAMMARS 12-28
Example: terminals: a
nonterminals: S, X
productions:
S X [ a
X aX
Total language tree:
S
/ \
X a
|
aX
|
aaX
.
.
.
Tree is innite, but CFL = a.
Chapter 13
Grammatical Format
13.1 Regular Grammars
We previously saw that
CFGs can generate some regular languages.
CFGs can generate some nonregular languages.
We will see that
all regular languages can be generated by CFGs.
some nonregular languages cannot be generated by CFGs.
Can turn FA into a CFG as follows:
13-1
CHAPTER 13. GRAMMATICAL FORMAT 13-2
Example: L = all words ending in a.
FA:
S-
A+
B
a
b
b a
b
a
Denition: The path development of a word processed on a machine:
Start in starting state S.
For each state visited, print out the input letters used thus far and the
current state.
The word ababba has following path development on the FA:
S
aA
abB
abaA
ababB
ababbB
ababbaA
ababba
Now we dene the following productions:
S aA [ bB
CHAPTER 13. GRAMMATICAL FORMAT 13-3
A aA [ bB [
B aA [ bB
Note that:
The CFG has a production
X cY
if and only if in the FA, there is an arc from state X to state Y labeled
with c.
The CFG has a production
X
if and only if state X in the FA is a nal state.
Derivation of ababba using the CFG:
S aA
abB
abaA
ababB
ababbB
ababbaA
ababba
There is a one-to-one correspondence between path developments on the FA
and derivations in the CFG; i.e., we can use the pigeonhole principle.
The derivation of the word ababba using the CFG is exactly the same as the
path development given above.
Theorem 21 All regular languages are CFLs.
CHAPTER 13. GRAMMATICAL FORMAT 13-4
Example:
FA:
S-
A+
B
C+
a
b
a
a
b
a
b
b
productions:
S aS [ bA
A aC [ bB [
B aB [ bC
C aA [ bB [
Consider a CFG G = (, , R, S), where
is the set of terminals
is the set of nonterminals, and S is the starting nonterminal
R (+)

is the set of productions, where a production (N, |) R

with N and | ( + )

is written as
N |
Denition: For a given CFG G = (, , R, S), W is a semiword if W

;
i.e., W is a string of terminals (maybe none) cancatenated with exactly one
nonterminal (on the right).
Example: aabaN is a semiword if N is a nonterminal and a and b are
terminals.
CHAPTER 13. GRAMMATICAL FORMAT 13-5
Denition: G = (, , R, S) is a regular grammar if (N, |) R implies
| (

) +

; i.e., each production has one of the following two forms:

1. nonterminal semiword
2. nonterminal word
where word

is a string of terminals, possibly .

Theorem 22 If a CFG is a regular grammar, then the language generated by
this CFG is regular.
Proof.
will prove theorem by showing that there is a TG that accepts the lan-
guage generated by the CFG.
Suppose CFG is as follows:
N
1
w
1
M
1
N
2
w
2
M
2
.
.
.
N
n
w
n
M
n
N
n+1
w
n+1
N
n+2
w
n+2
.
.
.
N
n+m
w
n+m
where N
i
and M
i
are nonterminals (not necessarily distinct) and w
i

are strings of terminals.

Thus, w
i
M
i
is a semiword.
At least one of the N
i
= S. Assume that N
1
= S.
Create a state of the TG for each nonterminal N
i
and for each nonter-
minal M
j
.
CHAPTER 13. GRAMMATICAL FORMAT 13-6
Also create a state +.
Make the state for nonterminal S the initial state of the transition graph.
Draw an arc labeled with w
i
from state N
i
to state M
i
if and only if
there is a production N
i
w
i
M
i
.
Draw an arc labeled with w
i
from state N
i
to state + if and only if there
is a production N
i
w
i
.
Thus, we have created a TG.
By considering the path developments of words accepted by the TG,
we can show that there is a one-to-one correspondence between words
accepted by TG and words in CFL.
Thus, these are the same language.
Kleenes Theorem implies that the language has a regular expression.
Thus, language is regular.
Remarks:
all regular languages can be generated by some regular grammars (The-
orem 21)
all regular grammars generate some regular language.
a regular language may have many CFGs that generate it, where some
of the CFGs may not be regular grammars.
Example: CFG
productions:
S aaS [ abS [ baS [ bb
Corresponding TG:
CHAPTER 13. GRAMMATICAL FORMAT 13-7
S-
bb
+
ba
aa
ab
Below is another CFG that is not a regular grammar for the same language:
S AaS [ AbS [ bAS [ bb
A a
CHAPTER 13. GRAMMATICAL FORMAT 13-8
Example: CFG
productions:
S aB [ bA [ abA [ baB [
A abaA [ bb
B baA [ ab
Corresponding TG:
S-
A
B
+
ab
b
a
ba
aba
ba ab
bb
^
CHAPTER 13. GRAMMATICAL FORMAT 13-9
Example: CFG
productions:
S aB [ bA [ abA [ baB
A abaA [ bb
B baA [ ab
Corresponding TG (note that CFG does not generate ):
S-
A
B
+
ab
b
a
ba
aba
ba ab
bb
CHAPTER 13. GRAMMATICAL FORMAT 13-10
Denition: A production (N, |) R is a -production if | = , i.e., the
production is N .
If CFG does not contain a -production, then , CFL.
However, CFG may have -production and , CFL.
Example: productions:
S aX
X
13.2 Chomsky Normal Form
13.2.1 Productions and Nullable Nonterminals
Recall we previously dened -production:
N
where N is some nonterminal.
Note that
If some CFL contains the word , then the CFG must have a -production.
However, if a CFG has a -production, then the CFL does not necessarily
contain ; e.g.,
S aX
X
which denes the CFL a.
Denition: For a given CFG with as its set of nonterminals and as its
set of terminals, a working string W ( + )

is any string of nontermi-

nals and/or terminals that can be generated from the CFG starting from any
nonterminal.
CHAPTER 13. GRAMMATICAL FORMAT 13-11
Example: CFG:
S a [ Xb [ aY a
X Y [
Y X [ a
Then in the derivation
S aY a
aXa
aa
we have that aY a, aXa, and aa are all working strings.
Denition: For a given CFG having a nonterminal X and W a possible
working string, we use the notation
X

W
if there is some derivation in the CFG starting from X that can result in the
working string W.
Example: CFG:
S a [ Xb [ aY a
X Y [
Y X [ a
Since we have the following derivation
S aY a
aXa
aa
we can write S

aY a and S

aXa and S

aa.
Denition: In a given CFG, a nonterminal X is nullable if
1. There is a production X , or
CHAPTER 13. GRAMMATICAL FORMAT 13-12
2. X

; i.e., there is a derivation that starts at X and leads to :
X
Example: CFG:
S a [ Xb [ aY a
X Y [
Y X [ a
has nullable nonterminals X, Y .
Example: CFG:
S X [ XY [ Z
X Z [
Y Wa [ a
Z WX [ aZ [ Zb
W XY Z [ bXa [
has nullable nonterminals S, X, Z, W.
Denition: For any language L, dene the language L
0
as follows:
1. if , L, then L
0
is the entire language L, i.e., L
0
= L.
2. if L, then L
0
is the language L ; i.e., if we let T = , then
L
0
= L T

, so L
0
is all words in L except .
Theorem 23 If L is a CFL generated by a CFG G
1
that includes -productions,
then there is another CFG G
2
with no -productions that generates L
0
.
Basic Idea.
We give constructive algorithm to convert CFG G
1
with -productions
into equivalent CFG G
2
with no -productions:
1. Delete all -productions.
CHAPTER 13. GRAMMATICAL FORMAT 13-13
2. For each production
X something
with at least one nullable nonterminal on the right-hand side, do the
following for each possible nonempty subset of nullable nonterminals
on the RHS:
(a) create a new production
X new something
where the new RHS is the same as the old RHS except with
the entire current subset of nullable nonterminals removed.
(b) do not create the production
X
Example: CFG G
1
S a [ Xb [ aY a
X Y [
Y X [ a
has nullable nonterminals X, Y .
We create new productions:
Original Production New Production
S Xb S b
S aY a S aa
X Y Nothing
Y X Nothing
New CFG G
2
:
S a [ Xb [ aY a [ b [ aa
X Y
Y X [ a
CHAPTER 13. GRAMMATICAL FORMAT 13-14
Example: CFG G
1
S X [ XY [ Z
X Z [
Y Wa [ a
Z WX [ aZ [ Zb
W XY Z [ bXa [
has nullable nonterminals S, X, Z, W.
We create new productions:
Original Production New Production
S X Nothing
S XY S Y
S Z Nothing
X Z Nothing
Y Wa Y a
Z WX Z W and Z X
Z aZ Z a
Z Zb Z b
W XY Z W Y Z, W XY , and W Y
W bXa W ba
New CFG G
2
:
S X [ XY [ Z [ Y
X Z
Y Wa [ a
Z WX [ aZ [ Zb [ W [ X [ a [ b
W XY Z [ bXa [ Y Z [ XY [ Y [ ba
We need to show two things:
1. all non- words generated using original CFG G
1
can be generated
using new CFG G
2
.
2. all words generated using new CFG G
2
can be generated using
original CFG G
1
.
CHAPTER 13. GRAMMATICAL FORMAT 13-15
First we show that all non- words generated using original CFG G
1
can
be generated using new CFG G
2
.
Suppose our CFG G
1
included the productions A bBb and B
.
Suppose we had the following derivation of a word:
S . . .
baAaAa
babBbaAa from A bBb
. . .
babBbaabAa
bbabbaabAa from B
. . .
There would have been no dierence if we had applied the produc-
tion A bb rather than A bBb in the third line.
More generally, we can see that any non- word generated using
original CFG G
1
can be generated using new CFG G
2
.
Now show that all words generated using new CFG G
2
can be generated
using original CFG G
1
.
Note that each new production is just a combination of old produc-
tions (e.g., X aY a and Y ).
Can show that any derivation using G
2
has a corresponding deriva-
tion using G
1
that possibly uses a -production.
Hence, all words generated using new CFG G
2
can be generated
using original CFG G
1
.
13.2.2 Unit Productions
Denition: A production (N, |) R is a unit production if | ; i.e., the
production is of the form
one nonterminal one nonterminal
CHAPTER 13. GRAMMATICAL FORMAT 13-16
Theorem 24 If a language L is generated by a CFG G
1
that has no -
productions, then there is also a CFG G
2
for L with no -productions and
no unit productions.
Basic Idea.
Use the following rules to create new CFG:
For each pair of nonterminals A and B such that there is a production
A B
or a chain of productions (unit derivation)
A

B,
introduce the following new productions:
if the non-unit productions from B are
B s
1
[ s
2
[ . . . [ s
n
where the s
i
( + )

are strings of terminals and nonterminals,

then create the new productions
A s
1
[ s
2
[ . . . [ s
n
Do the same for all such pairs A and B simultaneously.
Remove all unit productions.
Can show that G
1
and G
2
generate the same language.
Example: CFG G
1
:
S X [ Y [ bb
X Z [ aXY
Y Xa [ a
Z Y X [ S [ Zb
CHAPTER 13. GRAMMATICAL FORMAT 13-17
has unit productions and unit derivations
S X
S Y
X Z
Z S
S X Z
X Z S
Z S X
Z S Y
X Z S Y
We create new productions:
Original Unit Production (Derivation) New Productions
S X S aXY
S Y S Xa [ a
S X Z S Y X [ Zb
X Z X Y X [ Zb
X Z S X bb
X Z S Y X Xa [ a
Z S Z bb
Z S X Z aXY
Z S Y Z Xa [ a
New CFG G
2
:
S bb [ aXY [ Xa [ a [ Y X [ Zb
X aXY [ Y X [ Zb [ bb [ Xa [ a
Y Xa [ a
Z Y X [ Zb [ bb [ aXY [ Xa [ a
Theorem 25 Consider a CFG G
1
= (,
1
, R
1
, S
1
), which generates language
L
1
= L(G
1
). Then there exists another CFG G
2
= (,
2
, R
2
, S
2
) such that
L(G
2
) = L(G
1
) and
(N, u) R
2
implies u
+
+ ;
CHAPTER 13. GRAMMATICAL FORMAT 13-18
i.e., G
2
generates all non- strings of L
1
and each production in G
2
is of one
of two basic forms:
1. Nonterminal string of only Nonterminals
2. Nonterminal one terminal
Basic Idea. We will give a constructive proof:
Assume that the nonterminals in the CFG G
1
are S, X
1
, X
2
, . . . , X
n
.
Assume that the terminals in the CFG G
1
are a and b.
Introduce two new nonterminals A and B.
Introduce two new productions:
A a
B b
For each original production involving terminals,
replace each a with the nonterminal A
replace each b with the nonterminal B
Example: Original production in G
1
:
X
5
X
1
abaX
3
bbX
2
becomes new production in G
2
:
X
5
X
1
ABAX
3
BBX
2
which is string of only Nonterminals.
Example: Production in G
1
:
X
2
abaab
becomes new production in G
2
:
X
2
ABAAB
which is string of only nonterminals.
CHAPTER 13. GRAMMATICAL FORMAT 13-19
So now all original productions have been transformed into new produc-
tions that have only nonterminals on the RHS.
Also, we have two new productions for A and B.
Note that any derivation starting from S to produce the word
ababba
will now follow same sequence of (new) productions to derive the string
ABABBA
starting from S.
Then apply A a and B b the proper number of times to get the
word ababba.
Hence, any word generated by the original CFG G
1
can be generated by
the new CFG G
2
.
Need to show that any word generated by the new CFG G
2
can also be
generated by the original CFG G
1
.
Consider new CFG G
2
without the two productions A a and
B b.
Applying the new productions numerous times will result in a string
of As and Bs.
Applying corresponding original productions in same order will re-
sult in the same string with as and bs.
Then change string of As and Bs into as and bs using A a and
B b.
Thus, every word generated using new CFG G
2
can also be gener-
ated using original CFG G
1
.
Example: CFG G
1
:
S abSba [ bX
1
aX
1
[ X
2
[ bb
X
1
aa [ aSX
1
b
X
2
X
1
a [ a
CHAPTER 13. GRAMMATICAL FORMAT 13-20
can be transformed into new CFG G
2
:
S ABSBA [ BX
1
AX
1
[ X
2
[ BB
X
1
AA [ ASX
1
B
X
2
X
1
A [ A
A a
B b
13.2.3 Chomsky Normal Form
Denition: A CFG G = (, , R, S) is in Chomsky Normal Form (CNF) if
(N, |) R implies | () + ; i.e., each of its productions has one of the
two forms:
1. Nonterminal string of exactly two Nonterminals
2. Nonterminal one terminal
Theorem 26 For any CFL L, the non- words of L can be generated by a
CFG in CNF.
Basic Idea. By construction:
Let L
0
= L if , L, and L
0
= L if L.
By Theorem23, we know there is a CFG for L
0
that has no -productions.
By Theorem 24, we know there is a CFG for L
0
that has no unit pro-
ductions.
By Theorem 25, we know there is a CFG for L
0
for which each of its
productions are of one of two forms:
1. Nonterminal string of only nonterminals
2. Nonterminal one terminal
So now assume that our CFG for L
0
has the above three properties.
CHAPTER 13. GRAMMATICAL FORMAT 13-21
Do nothing to the productions of the form
Nonterminal one terminal
For each production of the form
Nonterminal string of Nonterminals
we expand it into a collection of productions as follows:
Suppose we have the production
X
4
X
2
X
5
X
3
X
2
X
1
Replace the production with the new productions
X
4
X
2
R
1
R
1
X
5
R
2
R
2
X
3
R
3
R
3
X
2
X
1
where the R
i
are new nonterminals.
For each transformation of original productions, introduce new non-
terminals R
i
.
This transformation creates a new CFG in CNF.
Now we have to show that the language generated by the new CFG is
the same as that generated by the original CFG.
First show that any word that can be generated by original CFG can
also be generated by new CFG:
In any derivation of a word using the original CFG, we just replace
any production of the form
X
4
X
2
X
5
X
3
X
2
X
1
with the new productions
X
4
X
2
R
1
R
1
X
5
R
2
R
2
X
3
R
3
R
3
X
2
X
1
CHAPTER 13. GRAMMATICAL FORMAT 13-22
This gives us a derivation of the word using the new CFG.
Now show that any word that can be generated by the new CFG can
also be generated by the original CFG:
Note that the nonterminal R
3
is only used in the RHS of the pro-
duction
R
2
X
3
R
3
Thus, that is the only way R
3
would arise.
Similarly, the nonterminal R
2
is only used in the RHS of the pro-
duction
R
1
X
5
R
2
Thus, that is the only way R
2
would arise.
We can similarly show the same for all new nonterminals R
i
Thus, since we use dierent R
i
s in the expansion of each produc-
tion, the new nonterminals R
i
cannot interact to create new words.
Example: CFG
S abSba [ bX
1
aX
2
[ bb
X
1
aa [ aSX
1
b
X
2
X
1
a [ abb
can be transformed into new CFG
S ABSBA [ BX
1
AX
2
[ BB
X
1
AA [ ASX
1
B
X
2
X
1
A [ ABB
A a
B b
which can then be transformed into a CFG in CNF:
S AR
1
R
1
BR
2
CHAPTER 13. GRAMMATICAL FORMAT 13-23
R
2
SR
3
R
3
BA
S BR
4
R
4
X
1
R
5
R
5
AX
2
S BB
X
1
AA
X
1
AR
6
R
6
SR
7
R
7
X
1
B
X
2
X
1
A
X
2
AR
8
R
8
BB
A a
B b
13.3 Leftmost Nonterminals and Derivations
Denition: The leftmost nonterminal (LMN) in a working string is the rst
nonterminal that we encounter when we scan the string from left to right.
Example: In the string bbabXbaY SbXbY , the LMN is X.
Denition: If a word w is generated by a CFG by a certain derivation and
at each step in the derivation, a rule of production is applied to the leftmost
nonterminal in the working string, then this derivation is called a leftmost
derivation (LMD).
CHAPTER 13. GRAMMATICAL FORMAT 13-24
Example: CFG:
S baXaS [ ab
X Xab [ aa
The following is a LMD:
S baXaS
baXabaS
baXababaS
baaaababaS
baaaababaab
CHAPTER 13. GRAMMATICAL FORMAT 13-25
Example: CFG:
S XY
X Y b [ Xa [ aa [ Y Y
Y XbbX [ ab
The word abbaaabbabab has the following derivation tree:
S
/ \
/ \
/ \
/ \
/ \
X _ Y _
/ \ / / \ \
Y b / | | \
/ \ X b b X
a b / \ / \
X a / \
/ \ Y Y
a a / \ / \
a b a b
Note that if we walk around the tree starting down the left branch of the root
with our left hand always touching the tree, then the order in which we rst
visit each nonterminal corresponds to the order in which the nonterminals are
replaced in LMD.
This is true for any derivation in any CFG
Theorem 27 Any word that can be generated by a given CFG by some deriva-
tion also has a LMD.
Chapter 14
Pushdown Automata
14.1 Introduction
Previously, we saw connection between
1. Regular languages
2. Finite automata
We saw that certain languages generaged by CFGs could not be accepted
by FAs.
14.2 Pushdown Automata
Now we will introduce new kind of machine: pushdown automaton
(PDA).
Will see connection between
1. context-free languages
2. pushdown automata
Pushdown automata and FAs share some features, but a PDA can have
one extra key feature: STACK.
innitely long INPUT TAPE on which input is written.
14-1
CHAPTER 14. PUSHDOWN AUTOMATA 14-2
INPUT TAPE is divided into cells, and each cell holds one input
letter or a blank .
a b b . . .
Once blank is encountered on INPUT TAPE, all of the following
cells also contain .
Read TAPE one cell at a time, from left to right. Cannot go back.
START, ACCEPT, and REJECT states.
Once enter either ACCEPT or REJECT state, cannot ever leave.
READ state to read input letter from INPUT TAPE.
Also, have an innitely tall PUSHDOWN STACK, which has last-
in-rst-out (LIFO) discipline.
Always start with STACK empty.
STACK can hold letters of STACK alphabet (which can be same
as input alphabet) and blanks .
Once we encounter a in stack, everything below it is also a .
b
b
a
a
.
.
.
PUSH and POP states alter contents of STACK.
PUSH adds something to the top of the STACK.
POP takes o the thing on the top of the STACK.
CHAPTER 14. PUSHDOWN AUTOMATA 14-3
Example: Convert FA to PDA
-
+
b
a
a b
b
a
FA:
b
a
b
a
a
START
READ READ READ ACCEPT
REJECT REJECT
PDA:
b
For this example, no STACK.
CHAPTER 14. PUSHDOWN AUTOMATA 14-4
Example: Convert FA to PDA
-
+
FA:
+
b
a
a
b
b
a, b a
READ
READ
REJECT
START
READ READ ACCEPT
REJECT
PDA:
a
a
b
a
a, b
b
b
For this example, no STACK.
CHAPTER 14. PUSHDOWN AUTOMATA 14-5
Example: PDA with STACK
START
ACCEPT
REJECT
PUSH a
POP
1
POP
2
READ
READ
2
1
a
b
b
a a, b b,
a
Suppose we had the INPUT TAPE
a a b b
Stack is initially empty:

.
.
.
See what happens when we process it:
CHAPTER 14. PUSHDOWN AUTOMATA 14-6
STATE STACK TAPE
START aabb
READ
1
,aabb
PUSH a a ,aabb
READ
1
a ,a,abb
PUSH a aa ,a,abb
READ
1
aa ,a,a,bb
POP
1
a ,a,a,bb
READ
2
a ,a,a,b,b
POP
1
,a,a,b,b
READ
2
,a,a,b,b,
POP
2
, ,a,a,b,b,
ACCEPT , ,a,a,b,b,
The language accepted by the PDA is
a
n
b
n
: n = 0, 1, 2, . . .
which is a nonregular language.
Proof. see pages 295299 of text.
So, why can PDAs accept certain nonregular languages?
STACK is memory with unlimited capacity.
FAs only had xed amount of memory built in.
CHAPTER 14. PUSHDOWN AUTOMATA 14-7
14.3 Determinism and Nondeterminism
Denition: A PDA is deterministic if each input string can only be processed
by the machine in one way.
Denition: A PDA is nondeterministic if there is some string that can be
processed by the machine in more than one way.
A nondeterministic PDA
may have more than one edge with the same label leading out of a certain
READ state or POP state.
may have more than one arc leaving the START state.
Both deterministic and nondeterministic PDAs
may have no edge with a certain label leading out of a certain READ
state or POP state.
if we are in a READ or POP state and encounter a letter for which there
is no out-edge from this state, the PDA crashes.
Remarks:
For FAs, nondeterminism does not increase power of machines.
For PDAs, nondeterminism does increase power of machines.
CHAPTER 14. PUSHDOWN AUTOMATA 14-8
14.4 Examples
Example: Language PALINDROMEX, which consists of all words of the
form
sXreverse(s)
where s is any string generated by (a +b)

.
PALINDROMEX = X, aXa, bXb, aaXaa, abXba, baXab, bbXbb, . . .
Each word in PALINDROMEX has odd length and X in middle.
When processing word on PDA, rst read letters from TAPE and PUSH
letters onto STACK until read in X.
Then POP letters o STACK, and check if they are the same as rest of
input string on TAPE.
PDA:
Input alphabet = a, b, X
Stack alphabet = a, b
START
READ
2
POP
1
POP
2
POP
3
ACCEPT
PUSH a READ
1
a
PUSH b
b
X
a
b
a
b
CHAPTER 14. PUSHDOWN AUTOMATA 14-9
Example: Language ODDPALINDROME, which consists of all words over
= a, b having odd length that are the same forwards and backwards.
ODDPALINDROME = a, b, aaa, aba, bab, bbb, aaaaa, . . .
Remarks:
For PALINDROMEX, easy to detect when at middle of word when read-
ing TAPE since marked by X.
For ODDPALINDROME, impossible to detect when at middle of word
when reading TAPE.
Need to use nondeterminism.
START
READ
2
POP
1
POP
2
POP
3
ACCEPT
PUSH a READ
1
a
PUSH b
b
a
b
a
b
a, b
CHAPTER 14. PUSHDOWN AUTOMATA 14-10
Example: Language EVENPALINDROME, which consists of all words over
= a, b having even length that are the same forwards as backwords.
EVENPALINDROME = s reverse(s) : s can be generated by (a +b)

= , aa, bb, aaaa, abba, baab, bbbb, aaaaaa, . . .

START
POP
2
POP
1
READ
2
POP
3
ACCEPT
a
PUSH a READ
1
a
PUSH b
b
b
b
a
b
a
CHAPTER 14. PUSHDOWN AUTOMATA 14-11
Suppose we had the INPUT TAPE
b a a b
Stack is initially empty:

.
.
.
See what happens when we process it:
STATE STACK TAPE
START baab
READ
1
,baab
PUSH b b ,baab
READ
1
b ,b,aab
PUSH a ab ,b,aab
READ
1
ab ,b,a,ab
POP
1
b ,b,a,ab
READ
2
b ,b,a,a,b
POP
2
,b,a,a,b
READ
2
,b,a,a,b,
POP
3
,b,a,a,b,
ACCEPT ,b,a,a,b,
Alternatively, we could have processed it as follows:
STATE STACK TAPE
START baab
READ
1
,baab
PUSH b b ,baab
READ
1
b ,b,aab
POP
1
,b ,b,aab
CRASH ,b,aab
This time the PDA crashes.
But since there is at least one way of processing the string baaaab which leads
to an ACCEPT state, the string is accepted by the PDA.
CHAPTER 14. PUSHDOWN AUTOMATA 14-12
14.5 Formal Denition of PDA and More Ex-
amples
Denition: A pushdown automaton (PDA) is a collection of eight things:
1. An alphabet of input letters.
2. An input TAPE (innite in one direction), which initially contains the
input string to be processed followed by an innite number of blanks
3. An alphabet of STACK characters.
4. A pushdown STACK (innite in one direction), which initially contains
all blanks .
5. One START state that has only out-edges, no in-edges. Can have more
than one arc leaving the START state. There are no labels on arcs
leaving the START state.
6. Halt states of two kinds:
(a) zero or more ACCEPT states
(b) zero or more REJECT states
Each of which have in-edges but no out-edges.
7. Finitely many nonbranching PUSH states that introduce characters from
onto the top of the STACK.
8. Finitely many branching states of two kinds:
(a) READ states, which read the next unused letter from TAPE and
may have out-edges labeled with letters from or a blank .
(There is no restriction on duplication of labels and no requirement
that there be a label for each letter of , or .)
(b) POP states, which read the top character of STACK and may have
out-edges labeled with letters of and the blank character , with
no restrictions.
Remarks:
CHAPTER 14. PUSHDOWN AUTOMATA 14-13
The denition for PDA allows for nondeterminism.
If we want to consider a PDA that does not have nondeterminism, then
we will call it a deterministic PDA.
CHAPTER 14. PUSHDOWN AUTOMATA 14-14
Example: CFG:
S S +S [ S S [ 3
terminals: +, , 3
nonterminals: S
(Nondeterministic) PDA:
START READ
1
READ
2
READ
4
ACCEPT
READ
3
PUSH S
1
POP
PUSH S PUSH S
PUSH S PUSH S
PUSH + PUSH *
2
3
4
5
6
7
S
* +
+
*
S S
3
Process 3 3 + 3 on PDA, where we now erase input TAPE as we read in
letters:
CHAPTER 14. PUSHDOWN AUTOMATA 14-15
STATE STACK TAPE
START 3 3 + 3
PUSH
1
S S 3 3 + 3
POP 3 3 + 3
PUSH
5
S 3 3 + 3
PUSH
6
S 3 3 + 3
PUSH
7
S S 3 3 + 3
POP S 3 3 + 3
READ
1
S 3 + 3
POP S 3 + 3
READ
3
S 3 + 3
POP 3 + 3
PUSH
2
S 3 + 3
PUSH
3
+S 3 + 3
PUSH
4
S +S 3 + 3
POP +S 3 + 3
READ
1
+S +3
POP S +3
READ
2
S 3
POP 3
READ
1

POP
READ
4

ACCEPT
14.6 Some Properties of PDA
Theorem 28 For every regular language L, there is some PDA that accepts
it.
CHAPTER 14. PUSHDOWN AUTOMATA 14-16
Note that PDA can reach ACCEPT state and still have non-blank letters on
TAPE and/or STACK.
Example:
READ
1
REJECT
ACCEPT
START PUSH S
PUSH X
b
a
Theorem 29 Given any PDA, there is another PDA that accepts exactly the
same language with the additional property that whenever a path leads to AC-
CEPT, the STACK and the TAPE contain only blanks.
Proof. Can convert above PDA into equivalent one below:
CHAPTER 14. PUSHDOWN AUTOMATA 14-17
READ
1
ACCEPT
REJECT
START PUSH S
PUSH X
b
POP READ
2
a
Chapter 15
CFG = PDA
15.1 Introduction
We will now see that the following are equivalent:
1. the set of all languages accepted by PDAs
2. the set of all languages generated by CFGs.
15.2 CFG PDA
Theorem 30 Given a language L generated by a particular CFG, there is a
PDA that accepts exactly L.
Proof. By construction
By Theorem 26, we can assume that the CFG is in CNF.
15-1
CHAPTER 15. CFG = PDA 15-2
Example: CFG in CNF:
S AS
S BC
B AA
A a
C b
Propose following (nondeterministic) PDA for above CFG:
START READ
1
ACCEPT
READ
POP
PUSH A PUSH B
PUSH A
S
S
PUSH S READ
3
2
PUSH S PUSH C
PUSH A
B
C A
b a
STACK alphabet: = S, A, B, C
Input TAPE alphabet: = a, b
CHAPTER 15. CFG = PDA 15-3
Consider following leftmost derivation of word aaaab:
S AS
aS
aAS
aaS
aaBC
aaAAC
aaaAC
aaaaC
aaaab
Now process string aaaab on PDA:
CHAPTER 15. CFG = PDA 15-4
Leftmost derivation STATE TAPE STACK
START aaaab
S PUSH S aaaab S
POP (S) aaaab
PUSH S aaaab S
AS PUSH A aaaab AS
POP (A) aaaab S
aS READ
2
,aaaab S
POP (S) ,aaaab
PUSH S ,aaaab S
aAS PUSH A ,aaaab AS
POP (A) ,aaaab S
aaS READ
2
,a,aaab S
POP (S) ,a,aaab
PUSH C ,a,aaab C
aaBC PUSH B ,a,aaab BC
POP (B) ,a,aaab C
PUSH A ,a,aaab AC
aaAAC PUSH A ,a,aaab AAC
POP (A) ,a,aaab AC
aaaAC READ
2
,a,a,aab AC
POP (A) ,a,a,aab C
aaaaC READ
2
,a,a,a,ab C
POP (C) ,a,a,a,ab
aaaab READ
1
,a,a,a,a,b
POP () ,a,a,a,a,b
READ
3
,a,a,a,a,b
ACCEPT ,a,a,a,a,b
Note that just before entering the POP state, the current working string
in the LMD is the same as the cancelled letters on the TAPE concate-
nated with current contents of the STACK.
Before the rst time we enter POP,
working string = S
letters cancelled = none
string of nonterminals in STACK = S
Just before entering POP for the last time,
working string = whole word
CHAPTER 15. CFG = PDA 15-5
letters cancelled = all
string of nonterminals in STACK =
CHAPTER 15. CFG = PDA 15-6
Consider the following CFG in CNF:
X
1
X
2
X
3
X
1
X
1
X
3
X
4
X
2
X
5
.
.
.
X
2
a
X
3
a
X
4
b
.
.
.
where start symbol S = X
1
.
Terminals: a, b
Nonterminals: X
1
, X
2
, . . . , X
n
Construction of PDA will correspond to leftmost derivation of words.
PDA will have only one POP and will be nondeterministic.
Begin constructing PDA by starting with
START
POP PUSH X
1
CHAPTER 15. CFG = PDA 15-7
For each production of the form
X
i
X
j
X
k
we include this circuit from the POP back to itself:
POP
PUSH X
PUSH X
k
j
X
i
CHAPTER 15. CFG = PDA 15-8
For all productions of the form
X
i
b
we add the following circuit to the above POP:
i
X
POP
READ
b
Finally, add the following to the above POP:
POP READ ACCEPT
CHAPTER 15. CFG = PDA 15-9
Recall that languages that include the word cannot be put into CNF.
To take care of this, we need to add loop to the above POP when
is in the language:
POP
S
This last loop will kill nonterminal S without replacing it with
anything.
CHAPTER 15. CFG = PDA 15-10
Example: Let L
0
be the language of the following CFG in CNF:
S AB
S SB
A CA
A a
B b
C b
We now want a PDA for the language L = L
0
+.
Propose following (nondeterministic) PDA for above CFG:
START READ
1
READ
2
READ
4
ACCEPT
READ
3
POP
A C B
S
PUSH B
PUSH A
PUSH A
PUSH C
A S
S
a b b
PUSH S
PUSH B
PUSH S
STACK alphabet: = S, A, B, C
CHAPTER 15. CFG = PDA 15-11
Input TAPE alphabet: = a, b
Consider following leftmost derivation of word babb:
S SB
ABB
CABB
bABB
baBB
babB
babb
Now process string babb on PDA:
Leftmost derivation STATE TAPE STACK
START babb
PUSH S babb S
POP (S) babb
PUSH B babb B
S SB PUSH S babb SB
POP (S) babb B
PUSH B babb BB
ABB PUSH A babb ABB
POP (A) babb BB
PUSH A babb ABB
CABB PUSH C babb CABB
POP (C) babb ABB
bABB READ
3
,babb ABB
POP (A) ,babb BB
baBB READ
1
,b,abb BB
POP (B) ,b,abb B
babB READ
2
,b,a,bb B
POP (B) ,b,a,bb
babb READ
2
,b,a,b,b
POP () ,b,a,b,b
READ
4
,b,a,b,b
ACCEPT ,b,a,b,b
CHAPTER 15. CFG = PDA 15-12
15.3 PDA CFG
Theorem 31 Given a language L that is accepted by a certain PDA, there
exists a CFG that generates exactly L.
Proof. Strategy of proof:
1. Start with any PDA
2. Put the PDA into a standardized form, known as conversion form.
3. The purpose of putting a PDA in conversion form is that since the PDA
now has a standardized form, we can easily convert the pictorial rep-
resentation of the PDA into a table. This table will be known as a
summary table. Number the rows in the summary table.
The summary table and the pictorial representation of the PDA will
contain exactly the same amount of information. In other words, if
you are only given a summary table, you could draw the PDA from
it.
The correspondence between the pictorial representation of the PDA
and the summary table is similar to the correspondence between a
drawing of a nite automaton and a tabular representation of the
FA.
4. Processing and accepting a string on the PDA will correspond to a par-
ticular sequence of rows from the summary table. But not every possible
sequence of rows from the summary table will correspond to a processing
of a string on the PDA. So we will come up with a way of determining
if a particular sequence of rows from the summary table corresponds to
a valid processing of a string on the PDA.
5. Then we will construct a CFG that will generate all valid sequences of
rows from the summary table. We call the collection of all valid sequences
of rows the row-language.
6. Convert this CFG for row-language into CFG that generates all words
of as and bs in original language of PDA.
We now begin by showing how to transform a given PDA into conversion form:
CHAPTER 15. CFG = PDA 15-13
rst introduce new state HERE in PDA.
HERE state does not read TAPE nor push or pop the STACK.
HERE is just used as a marker.
Denition: A PDA is in conversion form if it meets all of the following
conditions:
1. there is only one ACCEPT state.
2. there are no REJECT states.
3. Every READ or HERE is followed immediately by a POP.
4. POPs must be separated by READs or HEREs.
5. All branching occurs at READ or HERE states, none at POP states,
and every edge has only one label.
6. The STACK is initially loaded with the symbol $ on top. If the
symbol is ever popped in processing, it must be replaced immedi-
ately. The STACK is never popped beneath this symbol. Right
before entering ACCEPT, this symbol is popped and left out.
7. The PDA must begin with the sequence:
START POP PUSH $
HERE
or
READ
$
8. The entire input string must be read before the machine can accept
a word.
CHAPTER 15. CFG = PDA 15-14
Note that we can convert any PDA into an equivalent PDA in conversion
form as follows:
1. There is only one ACCEPT state:
If there is more than one ACCEPT state, then delete all but one
and have all the edges that formerly went into the others feed into
the remaining one:
ACCEPT ACCEPT
becomes
ACCEPT
CHAPTER 15. CFG = PDA 15-15
2. There are no REJECT states:
If there were previously any REJECT states in the original PDA,
just delete them from the new PDA. This will just lead to a crash,
which is equivalent to going to a REJECT state.
READ REJECT
b
a
becomes
READ
a
CHAPTER 15. CFG = PDA 15-16
3. Every READ or HERE is followed immediately by a POP:
a
READ READ
1 2
b
becomes
READ
1
READ
2
b
a
POP PUSH b
PUSH $
PUSH a
a
b
$
becomes (by property 5)
READ
1
READ
2
POP
POP
POP
PUSH b
PUSH a
PUSH $
a
b
b
b
b
$
a
CHAPTER 15. CFG = PDA 15-17
4. POPs must be separated by READs or HEREs:
POP POP
1 2
b
becomes
POP POP
2 1
b
HERE
CHAPTER 15. CFG = PDA 15-18
5. All branching occurs at READ or HERE states, none at POP states,
and every edge has only one label.
READ
2
b
POP READ
1
READ
3
a
b
becomes
READ
2
READ
3
POP
POP
b
READ
1
a
b
b
CHAPTER 15. CFG = PDA 15-19
6. The STACK is initially loaded with the symbol $ on top. If the
symbol is ever popped in processing, it must be replaced immedi-
ately. The STACK is never popped beneath this symbol. Right
before entering ACCEPT, this symbol is popped and left out.
$
7. The PDA must begin with the sequence:
START POP PUSH $
HERE
or
READ
$
Simple.
8. The entire input string must be read before the machine can accept
a word:
Use algorithm of Theorem 29.
CHAPTER 15. CFG = PDA 15-20
Example: PDA for language a
2n
b
n
: n = 1, 2, 3, . . .:
READ
1
START
POP
1
POP
2
READ
2
ACCEPT POP
3
PUSH a
a
b
b a a
PDA in conversion form:
PUSH a
PUSH a PUSH a
READ
1
POP
1
POP
2
START
READ
2
ACCEPT
HERE
POP
4
POP POP
5 6
POP
3
$
b
b
$
a
a $
PUSH $
a
a a
PUSH $
CHAPTER 15. CFG = PDA 15-21
Example: PDA for language ab:
PUSH a
a
START
1
READ
POP
READ
2
POP
ACCEPT
a
b
CHAPTER 15. CFG = PDA 15-22
PDA in conversion form:
$
a a
$
a
a
$
POP
START
PUSH $
POP POP POP
POP
1
READ
READ
2
ACCEPT
PUSH $
PUSH a PUSH a
PUSH a
b
From To READ POP PUSH Row
where where what what what number
START READ
1
$ $ 1
READ
1
READ
1
a $ a$ 2
READ
1
READ
1
a a aa 3
READ
1
READ
2
b a 4
READ
2
ACCEPT $ 5
CHAPTER 15. CFG = PDA 15-23
Purpose of conversion form is to decompose machine into path segments,
each of the form:
From To Reading Popping Pushing
START READ One or no Exactly Any string
or READ or HERE input letters one STACK onto the
or HERE or ACCEPT character STACK
The states START, READ, HERE, and ACCEPT are called joints.
We can break up any PDA in conversion form into a collection of joint-
to-joint segments.
Each joint-to-joint segment has the following form:
1. It starts with a joint.
2. The rst joint is immediately followed by exactly one POP.
3. The one POP is immediately followed by zero or more PUSHes.
4. The PUSHes are immediately followed by another JOINT.
Summary table describes the entire PDA as list of all joint-to-joint seg-
ments:
From To READ POP PUSH Row
where where what what what number
START READ
1
$ $ 1
READ
1
READ
1
a $ a$ 2
READ
1
READ
1
a a aa 3
READ
1
READ
2
b a 4
READ
2
ACCEPT $ 5
Consider processing string ab on PDA:
CHAPTER 15. CFG = PDA 15-24
STATE Corresponding
Row Number
START 1
POP
PUSH $
READ
1
2
POP
PUSH $
PUSH a
READ
1
4
POP
READ
2
5
POP
ACCEPT
Every path through PDA corresponds to a sequence of rows of the sum-
mary table
Not every sequence of rows of the summary table corresponds to a path
through PDA.
Need to make sure joint consistent; i.e., last STATE of one row is
same as rst STATE of next row in sequence.
Need to make sure STACK consistent; i.e., when a row pops a
character, it should be at the top of the STACK.
Dene row-language of PDA represented by a summary table:
Alphabet letters:
= Row
1
, Row
2
, . . . , Row
5

i.e., terminals
All valid words are sequences of alphabet letters that correspond
to paths from START to ACCEPT that are joint consistent and
STACK consistent.
All valid words begin with Row
1
and end with Row
5
.
The string
Row
1
Row
4
Row
3
Row
3
is not a valid word
CHAPTER 15. CFG = PDA 15-25
Does not end with Row
5
not joint consistent since Row
4
ends in state READ
2
, and Row
3
begins in state READ
1
not STACK consistent since Row
1
ends with $ on the top of the
STACK, and Row
4
tries to pop a from the top of the STACK
We will develop a CFG for row-language and then transform it into
another CFG for the original language accepted by the PDA.
Recall the strategy of our proof:
1. Start with any PDA
2. Redraw PDA in conversion form.
3. Build summary table and number the rows.
4. Dene row-language to be set of all sequences of rows that corre-
spond to paths through PDA. Make sure STACK consistent.
5. Determine a CFG that generates all words in row-language.
6. Convert this CFG for row-language into CFG that generates all
words of as and bs in original language of PDA.
We are now up to Step 5.
So for Step 5, we want to determine a CFG for the row-language.
Dene nonterminal S to be used to start any derivation in row-language
grammar.
Nonterminals in the row-language grammar:
Net(X, Y, Z)
where
X and Y are specic joints (START, READ, HERE, ACCEPT)
Z is any character from stack alphabet .
Interpretation: There is some path going from joint X to joint Y
(possibly going through other joints) that has the net eect on the
STACK of removing the symbol Z from top of STACK.
STACK is never popped below the initial Z on the top, but may be
built up along the path, and eventually ends with the Z popped.
CHAPTER 15. CFG = PDA 15-26
Example:
READ
1
POP PUSH b PUSH a POP POP
READ
2
a
Z a
b
has net eect of popping Z, and is a Net(READ
1
, READ
2
, Z).
Example:
READ
1
READ
2
POP POP PUSH a
Z a a
does not have net eect of popping Z since STACK went below the initial
Z. Hence, this is not a Net(READ
1
, READ
2
, Z).
CHAPTER 15. CFG = PDA 15-27
Productions in the CFG for row-language will typically have
a nonterminal Net(, , ) on the LHS
and on the RHS, there will be a terminal Row
i
followed by zero or
more nonterminals Net(, , ).
The LHS and RHS of each production will have the same net eect on
the STACK.
Recall that the summary table for our example is
From To READ POP PUSH Row
where where what what what number
START READ
1
$ $ 1
READ
1
READ
1
a $ a$ 2
READ
1
READ
1
a a aa 3
READ
1
READ
2
b a 4
READ
2
ACCEPT $ 5
Example: Production:
Net(READ
1
, READ
2
, a) Row
4
Example: Production:
Net(READ
1
, ACCEPT, $)
Row
2
Net(READ
1
, READ
2
, a) Net(READ
2
, ACCEPT, $)
CHAPTER 15. CFG = PDA 15-28
READ
1
READ
2
READ
1
PUSH a
POP POP
ACCEPT
POP
$
PUSH $
$ a
a
Net(READ1, READ2, a)
Row 2 Net(READ2, ACCEPT, $)
CHAPTER 15. CFG = PDA 15-29
In last example, note that
Row
2
POPs the $ o the stack, then PUSHes $ and then a, and
ends in state READ
1
.
Then, Net(READ
1
, READ
2
, a) starts in state READ
1
, has the net
eect of POPping the a o the top of the STACK, and ends in state
READ
2
.
Then, Net(READ
2
, ACCEPT, $) starts in state READ
2
, has the
net eect of POPping the $ o the top of the STACK, and ends in
state ACCEPT.
The above three steps can be summarized by Net(READ
1
, ACCEPT, $).
More generally, use following rules to create productions:
Rule 1: Create production
S Net(START, ACCEPT, $)
Rule 2: For every row of summary table that has no PUSH entry, such
as
FROM TO READ POP PUSH ROW
X Y anything Z i
we include the production:
Net(X, Y, Z) Row
i
CHAPTER 15. CFG = PDA 15-30
Rule 3: For every row that pushes n 1 characters onto the STACK,
such as
FROM TO READ POP PUSH ROW
X Y anything Z m
1
m
2
m
n
j
for all sets of n READ, HERE, or ACCEPT states S
1
, S
2
, . . . , S
n
,
we create the productions:
Net(X, S
n
, Z) Row
j
Net(Y, S
1
, m
1
) Net(S
1
, S
2
, m
2
) Net(S
n1
, S
n
, m
n
)
.
.
.
POP
Y
m
1
POP
m
S
1
2
POP POP
X
Z
PUSH m
n
PUSH m
n-1
PUSH m
n-2
PUSH m
1
m
S
2
3
. . .
S
n
Row j
Net(Y, S1, m1)
Net(S1, S2, m2)
CHAPTER 15. CFG = PDA 15-31
Some productions generated may never be used in a derivation of a word.
This is analogous to the following:
Example: CFG:
S X [ Y
X aX
Y ab
Production S X doesnt lead to a word.
Applying Rule 1 gives
PROD 1 S Net(START, ACCEPT, $)
Applying Rule 2 to Rows 4 and 5 gives
PROD 2 Net(READ
1
, READ
2
, a) Row
4
PROD 3 Net(READ
2
, ACCEPT, $) Row
5
Applying Rule 3 to Row 1 gives
Net(START, S
1
, $) Row
1
Net(READ
1
, S
1
, $)
where S
1
can take on values READ
1
, READ
2
, ACCEPT.
PROD 4 Net(START, READ
1
, $) Row
1
Net(READ
1
, READ
1
, $)
PROD 5 Net(START, READ
2
, $) Row
1
Net(READ
1
, READ
2
, $)
PROD 6 Net(START, ACCEPT, $) Row
1
Net(READ
1
, ACCEPT, $)
CHAPTER 15. CFG = PDA 15-32
Applying Rule 3 to Row 2 gives
Net(READ
1
, S
2
, $) Row
2
Net(READ
1
, S
1
, a) Net(S
1
, S
2
, $)
where S
2
can be any joint except START
and S
1
can be any joint except START or ACCEPT.
PROD 7 Net(READ
1
, READ
1
, $)
Row
2
Net(READ
1
, READ
1
, a) Net(READ
1
, READ
1
, $)
PROD 8 Net(READ
1
, READ
1
, $)
Row
2
Net(READ
1
, READ
2
, a) Net(READ
2
, READ
1
, $)
PROD 9 Net(READ
1
, READ
2
, $)
Row
2
Net(READ
1
, READ
1
, a) Net(READ
1
, READ
2
, $)
PROD 10 Net(READ
1
, READ
2
, $)
Row
2
Net(READ
1
, READ
2
, a) Net(READ
2
, READ
2
, $)
PROD 11 Net(READ
1
, ACCEPT, $)
Row
2
Net(READ
1
, READ
1
, a) Net(READ
1
, ACCEPT, $)
PROD 12 Net(READ
1
, ACCEPT, $)
Row
2
Net(READ
1
, READ
2
, a) Net(READ
2
, ACCEPT, $)
CHAPTER 15. CFG = PDA 15-33
Applying Rule 3 to Row 3 gives
Net(READ
1
, S
2
, a) Row
3
Net(READ
1
, S
1
, a) Net(S
1
, S
2
, a)
where S
2
can be any joint except START
and S
1
can be any joint except START or ACCEPT.
PROD 13 Net(READ
1
, READ
1
, a)
Row
3
Net(READ
1
, READ
1
, a) Net(READ
1
, READ
1
, a)
PROD 14 Net(READ
1
, READ
1
, a)
Row
3
Net(READ
1
, READ
2
, a) Net(READ
2
, READ
1
, a)
PROD 15 Net(READ
1
, READ
2
, a)
Row
3
Net(READ
1
, READ
1
, a) Net(READ
1
, READ
2
, a)
PROD 16 Net(READ
1
, READ
2
, a)
Row
3
Net(READ
1
, READ
2
, a) Net(READ
2
, READ
2
, a)
PROD 17 Net(READ
1
, ACCEPT, a)
Row
3
Net(READ
1
, READ
1
, a) Net(READ
1
, ACCEPT, a)
PROD 18 Net(READ
1
, ACCEPT, a)
Row
3
Net(READ
1
, READ
2
, a) Net(READ
2
, ACCEPT, a)
Our CFG for the row-language has
5 terminals:
Row
1
, Row
2
, . . . , Row
5
16 nonterminals:
S, 9 of the form Net(, , $), 6 of the form Net(, , a).
18 productions:
PROD 1, . . ., PROD 18
CHAPTER 15. CFG = PDA 15-34
Can derive word in row-language using left-most derivation:
S Net(START, ACCEPT, $) PROD 1
Row
1
Net(READ
1
, ACCEPT, $) PROD 6
Row
1
Row
2
Net(READ
1
, READ
2
, a) Net(READ
2
, ACCEPT, $) PROD 12
Row
1
Row
2
Row
4
Net(READ
2
, ACCEPT, $) PROD 2
Row
1
Row
2
Row
4
Row
5
PROD 3
Not all productions in CFG will be used in derivations of actual words.
Our CFG doesnt generate words having as and bs. It generates words
using terminals Row
1
, Row
2
, . . . , Row
5
.
Need to transform this CFG into another CFG that has terminals a and
b.
CHAPTER 15. CFG = PDA 15-35
To convert previous CFG for row-language into CFG for original lan-
guage of as and bs,
Change the terminals Row
i
into nonterminals
Add new terminals a, b.
Also use
Create more productions as below:
Rule 4: For every row
FROM TO READ POP PUSH ROW
A B C D EFGH i
create the production
Row
i
C
Applying Rule 4 gives
PROD 19 Row
1

PROD 20 Row
2
a
PROD 21 Row
3
a
PROD 22 Row
4
b
PROD 23 Row
5

We can continue with the previous derivation in the row-language gram-
mar to get a word in the original language:
S Net(START, ACCEPT, $) PROD 1

Row
1
Row
2
Row
4
Row
5
PROD 3
Row
2
Row
4
Row
5
PROD 19
a Row
4
Row
5
PROD 20
a b Row
5
PROD 22
a b PROD 23
giving us the word ab.
The word ab can be accepted by the PDA in conversion form by following
the path:
Row
1
Row
2
Row
4
Row
5
Chapter 17
Context-Free Languages
17.1 Closure Under Unions
We will now prove some properties of CFLs.
Theorem 36 If L
1
and L
2
are CFLs, then their union L
1
+L
2
is a CFL.
Proof. By grammars.
L
1
CFL implies that L
1
has a CFG, CFG
1
, that generates it.
Assume that the nonterminals in CFG
1
are S, A, B, C, . . ..
Change the nonterminals in CFG
1
to S
1
, A
1
, B
1
, C
1
, . . ..
Do not change the terminals in the CFG
1
.
L
2
CFL implies that L
2
has a CFG, CFG
2
, that generates it.
Assume that the nonterminals in CFG
2
are S, A, B, C, . . ..
Change the nonterminals in CFG
2
to S
2
, A
2
, B
2
, C
2
, . . ..
Do not change the terminals in the CFG
2
.
Now CFG
1
and CFG
2
have nonintersecting sets of nonterminals.
We create a CFG for L
1
+L
2
as follows:
17-1
CHAPTER 17. CONTEXT-FREE LANGUAGES 17-2
Include all of the nonterminals S
1
, A
1
, B
1
, C
1
, . . . and S
2
, A
2
, B
2
, C
2
, . . ..
Include all of the productions from CFG
1
and CFG
2
.
Create a new nonterminal S and a production
S S
1
[ S
2
To see that this new CFG generates L
1
+L
2
,
note that any word in language L
i
, i = 1, 2, can be generated by
rst using the production S S
i
also, since there is no overlap in the use of nonterminals in CFG
1
and CFG
2
, once we start a derivation with the production S S
1
,
we can only use the productions originally in CFG
1
and cannot use
any of the productions from CFG
2
, and so we can only produce
words in L
1
.
Similar situation occurs when we start a derivation with the pro-
duction S S
2
.
Example:
CFG
1
for L
1
S SS [ AaAb [ BBB [
A SaS [ bBb [ abba
B SSS [ baab
CFG
2
for L
2
S aS [ aAba [ BbB [
A aSa [ abab
B BabaB [ bb
To construct CFG for L
1
+L
2
transform CFG
1
S
1
S
1
S
1
[ A
1
aA
1
b [ B
1
B
1
B
1
[
A
1
S
1
aS
1
[ bB
1
b [ abba
B
1
S
1
S
1
S
1
[ baab
CHAPTER 17. CONTEXT-FREE LANGUAGES 17-3
transform CFG
2
S
2
aS
2
[ aA
2
ba [ Bb
2
B
2
[
A
2
aS
2
a [ abab
B
2
B
2
abaB
2
[ bb
construct CFG for L
1
+L
2
:
S S
1
[ S
2
S
1
S
1
S
1
[ A
1
aA
1
b [ B
1
B
1
B
1
[
A
1
S
1
aS
1
[ bB
1
b [ abba
B
1
S
1
S
1
S
1
[ baab
S
2
aS
2
[ aA
2
ba [ Bb
2
B
2
[
A
2
aS
2
a [ abab
B
2
B
2
abaB
2
[ bb
Proof. (of Theorem 36 by machines)
Since L
1
is CFL, Theorem 30 implies that there exists some PDA, PDA
1
,
that accepts L
1
.
Since L
2
is CFL, Theorem 30 implies that there exists some PDA, PDA
2
,
that accepts L
2
.
Construct new PDA
3
to accept L
1
+L
2
by combining PDA
1
and PDA
2
into one machine by coalescing START states of PDA
1
and PDA
2
into
a single START state.
Note that once we leave the START state of PDA
3
, we can never come
back to the START state.
Also, there is no way to cross over from PDA
1
to PDA
2
.
Hence, any word accepted by PDA
3
must also be accepted by either
PDA
1
or PDA
2
.
Also, it is obvious that any word accepted by either PDA
1
or PDA
2
will
be accepted by PDA
3
.
CHAPTER 17. CONTEXT-FREE LANGUAGES 17-4
Example:
PDA
1
for L
1
:
READ READ ACCEPT
START
b a a
1
PDA
2
for L
2
:
ACCEPT READ
START
PUSH a
a
a
b
POP
2
CHAPTER 17. CONTEXT-FREE LANGUAGES 17-5
PDA
3
for L
1
+L
2
:
READ READ ACCEPT READ
START PUSH a
a
a
b a a
b
POP
17.2 Closure Under Concatenations
Theorem 37 If L
1
and L
2
are CFLs, then L
1
L
2
is a CFL.
Proof. By grammars.
L
1
CFL implies that L
1
has a CFG, CFG
1
, that generates it.
Assume that the nonterminals in CFG
1
are S, A, B, C, . . ..
Change the nonterminals in CFG
1
to S
1
, A
1
, B
1
, C
1
, . . ..
Do not change the terminals in the CFG
1
.
L
2
CFL implies that L
2
has a CFG, CFG
2
, that generates it.
Assume that the nonterminals in CFG
2
are S, A, B, C, . . ..
Change the nonterminals in CFG
2
to S
2
, A
2
, B
2
, C
2
, . . ..
CHAPTER 17. CONTEXT-FREE LANGUAGES 17-6
Do not change the terminals in the CFG
2
.
Now CFG
1
and CFG
2
have nonintersecting sets of nonterminals.
We create a CFG for L
1
L
2
as follows:
Include all of the nonterminals S
1
, A
1
, B
1
, C
1
, . . . and S
2
, A
2
, B
2
, C
2
, . . ..
Include all of the productions from CFG
1
and CFG
2
.
Create a new nonterminal S and a production
S S
1
S
2
To see that this new CFG generates L
1
L
2
,
Obviously, we can generated any word in L
1
L
2
using our new CFG.
also, since there is no overlap in the use of nonterminals in CFG
1
and
CFG
2
, once we start a derivation with the production S S
1
S
2
, the
S
1
part will generate a word from L
1
and the S
2
part will generate
a word from L
2
.
hence, any word generated by the new CFG will be in L
1
L
2
.
Example:
CFG
1
for L
1
S SS [ AaAb [ BBB [
A SaS [ bBb [ abba
B SSS [ baab
CFG
2
for L
2
S aS [ aAba [ BbB [
A aSa [ abab
B BabaB [ bb
To construct CFG for L
1
L
2
CHAPTER 17. CONTEXT-FREE LANGUAGES 17-7
transform CFG
1
S
1
S
1
S
1
[ A
1
aA
1
b [ B
1
B
1
B
1
[
A
1
S
1
aS
1
[ bB
1
b [ abba
B
1
S
1
S
1
S
1
[ baab
transform CFG
2
S
2
aS
2
[ aA
2
ba [ Bb
2
B
2
[
A
2
aS
2
a [ abab
B
2
B
2
abaB
2
[ bb
construct CFG for L
1
L
2
:
S S
1
S
2
S
1
S
1
S
1
[ A
1
aA
1
b [ B
1
B
1
B
1
[
A
1
S
1
aS
1
[ bB
1
b [ abba
B
1
S
1
S
1
S
1
[ baab
S
2
aS
2
[ aA
2
ba [ Bb
2
B
2
[
A
2
aS
2
a [ abab
B
2
B
2
abaB
2
[ bb
Remarks:
Dicult to prove Theorem 37 by machines.
Cannot just combine PDA
1
and PDA
2
by removing the ACCEPT state
of PDA
1
and replacing it with the START state of PDA
2
.
Problem is we can reach the ACCEPT state of PDA
1
while there are
still unread characters on the input TAPE and there are still characters
on the STACK.
Thus, when we go to PDA
2
, we may process the last part of the word in
L
1
and the entire word in L
2
and incorrectly accept or reject the entire
word.
CHAPTER 17. CONTEXT-FREE LANGUAGES 17-8
17.3 Closure Under Kleene Star
Theorem 38 If L is a CFL, then L

is a CFL.
Proof.
Since L is a CFL, by denition there is some CFG that generates L.
Suppose CFG for L has nonterminals S, A, B, C, . . ..
Change the nonterminal S to S
1
.
We create a new CFG for L

as follows:
Include all the nonterminals S
1
, A, B, C, . . . from the CFG for L.
Include all of the productions from the CFG for L.
Add the new nonterminal S and the new production
S S
1
S [
We can repeat last production
S S
1
S S
1
S
1
S S
1
S
1
S
1
S S
1
S
1
S
1
S
1
S S
1
S
1
S
1
S
1
S
1
S
1
S
1
S
1
Note that any word in L

can be generated by the new CFG.

To show that any word generated by the new CFG is in L

, note that
each of the S
1
above generates a word in L.
Also, there is no interaction between the dierent S
1
s.
Example: CFG for L:
S AaAb [ BBB [
A SaS [ bBb [ abba
B SSS [ baab
CHAPTER 17. CONTEXT-FREE LANGUAGES 17-9
Convert CFG for L:
S
1
AaAb [ BBB [
A S
1
aS
1
[ bBb [ abba
B S
1
S
1
S
1
[ baab
New CFG for L

:
S S
1
S [
S
1
AaAb [ BBB [
A S
1
aS
1
[ bBb [ abba
B S
1
S
1
S
1
[ baab
17.4 Intersections
We now will give an example showing that the intersection of two CFLs
may not be a CFL.
To show this, we will need to assume that the language L
3
= a
n
b
n
a
n
:
n = 0, 1, 2, . . . is a non-context-free language. This is shown in the
textbook in Chapter 16. L
3
is the set of words with some number of as,
followed by an equal number of bs, and ending with the same number
of as.
Example:
Let L
1
be generated by the following CFG:
S XY
X aXb [
Y aY [
Thus, L
1
= a
n
b
n
a
m
: n, m 0, which is the set of words that have
a clump of as, followed by a clump of bs, and ending with another
clump of as, where the number of as at the beginning is the same as
the number of bs in the middle. The number of as at the end of the
word is arbitrary, and does not have to equal the number of as and bs
that come before it.
CHAPTER 17. CONTEXT-FREE LANGUAGES 17-10
Let L
2
be generated by the following CFG:
S WZ
W aW [
Z bZa [
Thus, L
2
= a
i
b
k
a
k
: i, k 0, which is the set of words that have a
clump of as, followed by a clump of bs, and ending with another clump
of as, where the number of bs in the middle is the same as the number
of as at the end. The number of as at the beginning of the word is
arbitrary, and does not have to equal the number of bs and as that
come after it.
Note that L
1
L
2
= L
3
, where L
3
= a
n
b
n
a
n
: n = 0, 1, 2, . . ., which is
a non-context-free language.
However, sometimes the intersection of two CFLs is a CFL.
For example, suppose that L
1
and L
2
are regular languages. Then The-
orem 21 implies that L
1
and L
2
are CFLs. Also, Theorem 12 implies
that L
1
L
2
is a regular language, and so L
1
L
2
is also a CFL by
Theorem 21. Thus, here is an example of 2 CFLs whose intersection is
a CFL.
Thus, in general, we cannot say if the intersection of two CFLs is a CFL.
17.5 Complementation
If L is a CFL, then L

may or may not be a CFL.

We rst show that the complement of a CFL may be a CFL:
If L is regular, then L

is also regular by Theorem 11.

Also, Theorem 21 implies that both L and L

are CFLs.
We now show that the complement of a CFL may not be a CFL by
contradiction:
CHAPTER 17. CONTEXT-FREE LANGUAGES 17-11
Suppose that it is always true that if L is a CFL, then L

is a CFL.
Suppose that L
1
and L
2
are CFLs.
Then by our assumption, we must have that L

1
and L

2
are CFLs.
Theorem 36 implies that L

1
+L

2
is a CFL.
Then by our assumption, we must have that (L

1
+L

2
)

is a CFL.
But we know that (L

1
+L

2
)

= L
1
L
2
by DeMorgans Law.
However, we previously showed that the intersection of two CFLs
is not always a CFL, which contradicts the previous two steps.
So our assumption that CFLs are always closed under complemen-
tation must not be true.
Thus, in general, we cannot say if the complement of a CFL is a CFL.
Chapter 18
Decidability for CFLs
18.1 Membership The CYK Algorithm
We want to determine if a given string x can be generated from a particular
CFG G.
Theorem 45 Let L be a language generated by a CFG G with alphabet .
Given a string s

, we can decide whether or not s L.

Proof. We will use a constructive algorithm known as the CYK algorithm,
developed by Cocke, Younger and Kasami.
First suppose s = .
The proof of Theorem 23 gives an algorithm to nd all of the nul-
lable nonterminals in a CFG.
If the starting nonterminal S is a nullable nonterminal, then L.
Now suppose s ,= .
The following algorithm is taken from Floyd and Beigel, 1994. The
Language of Machines: An Introduction to Computability and For-
mal Languages. W. H. Freeman and Company, New York.
Assume that , L, so we can transform CFG G into another CFG
G
1
in Chomsky Normal Form by Theorem 26.
18-1
CHAPTER 18. DECIDABILITY FOR CFLS 18-2
Let s = s
1
s
2
s
n
be a string of length n 1, so s
i
is the ith letter
of s.
Let s
ik
= s
i
s
i+1
s
k
, the substring of s from the ith letter to the
kth letter.
The algorithm will determine for each i and k with 0 < i k n
and each nonterminal X whether X

s
ik
.
We denote the answer to this question by T[i, k, X].
First consider the case when i = k so s
ik
= s
ii
= s
i
, a one-character
string.
Then T[i, k, X] is true if and only if the CFG G
1
includes the
production
X s
i
Now suppose i < k, so that length(s
ik
) 2.
Then T[i, k, X] is true if and only if
G
1
includes a production
X Y Z
s
ik
= uv, i.e., can split s
ik
into substrings u and v such that
their concatenation gives s
ik
.
Y

u
Z

v
Formally, T[i, k, Z] is true if and only if
G
1
includes a production
X Y Z
there exists j with i j < k such that
Y

s
ij
Z

s
j+1,k
Thus, we get the following recurrence:
T[i, k, X] =

true if i = k and G
1
has production X s
ii
true if i < k and G
1
has production X Y Z such that
j with i j < k and T[i, j, Y ] and T[j + 1, k, Z]
false otherwise
CHAPTER 18. DECIDABILITY FOR CFLS 18-3
Can solve recursion using dynamic programming.
Store the values of T in an array that is initialized to false
everywhere.
Need to go through the array in such an order that T[i, j, Y ]
and T[j +1, k, Z] are evaluated before T[i, k, X] for i j < k.
Can do this by going through the array for increasing values of
k and, subject to that, decreasing the values of i.
CYK Algorithm: to determine if s L, where L is generated by
CFG G
1
in Chomsky normal form.
/* initialization */
n = length(s);
for every nonterminal X, do begin
for i = 1 to n do
for k = i to n do
T[i, k, X] = false;
for i = 1 to n do
if G
1
has production X s
ii
, then
T[i, i, X] = true;
end;
for k = 2 to n do
for i = k 1 down to 1 do
for all productions in G
1
of the form X Y Z do
for j = i to k 1 do
if T[i, j, Y ] and T[j + 1, k, Z] then
T[i, k, X] = true;
s L i T[1, n, S] = true;
Chapter 19
Turing Machines
19.1 Introduction
Turing machines will be our ultimate model for computers, so they need
output capabilities.
But computers without output statements can tell us something.
Consider the following program
1. READ X
2. IF X=1 THEN END
3. IF X=2 THEN DIVIDE X BY 0
4. IF X>2 THEN GOTO STATEMENT 4
If we assume that the input is always a positive integer, then
if program terminates naturally, then we know X was 1.
if program terminates with error message saying there is an overow
(i.e., crashes), then we know X was 2.
if the program does not terminate, then we know X was greater
than 2.
Denition: A Turing machine (TM) T = (, , , , K, s, H, ), where
1. An alphabet of input letters, and assume that the blank , .
19-1
CHAPTER 19. TURING MACHINES 19-2
2. A Tape divided into a sequence of numbered cells, each containing one
character or a blank.
The input word is presented to the machine on the tape with one
letter per cell beginning in the leftmost cell, called cell i.
The rest of the Tape is initially lled with blanks .
The Tape is innitely long in one direction.
cell ii cell iii cell v cell i
Tape Head
. . .
cell iv
3. A Tape Head that can in one step read the contents of a cell on the
Tape, replace it with some other character, and reposition itself to the
next cell to the right or to the left of the one it has just read.
At the start of the processing, the Tape Head always begins by
reading the input in cell i.
The Tape Head can never move left from cell i. If it is given orders
to do so, the machine crashes.
The location of the Tape Head is indicated as in the above picture.
4. An alphabet of characters that can be printed on the Tape by the
Tape Head .
Assume that , , and we may have that .
The Tape Head may erase a cell, which corresponds to writing
in the cell.
5. A nite set K of states including
Exactly one START state s K from which we begin execution
(and which we may reenter during execution).
CHAPTER 19. TURING MACHINES 19-3
H K is a set of HALT states, which cause execution to terminate
when we enter any of them. There are zero or more HALT states.
The other states have no function, only names such as q
1
, q
2
, q
3
, . . .
or 1, 2, 3, . . ..
6. A program , which is a nite set of rules that, on the basis of the state
we are in and the letter the Tape Head has just read, tells us
(a) how to change states,
(b) what to print on the Tape,
(c) where to move the Tape Head.
The program
K K ( + +) ( +) L, R,
with the restriction that
if (q
1
, q
2
, , c, d) and (q

1
, q

2
,

, c

, d

) with q
1
= q

1
and =

,
then q
2
= q

2
, c = c

and d = d

;
i.e., for any state q
1
and any character + + , there is
only one arc leaving state q
1
corresponding to reading character
from the Tape.
This restriction means that TMs are deterministic.
We depict the program as a collection of directed edges connecting the
states. Each edge is labeled with the triplet of information:
(character, character, direction) (+ +) ( +) L, R
where
The rst character (either or from or ) is the character the
Tape Head reads from the cell to which it is pointing.
From any state, there can be at most one arc leaving that state
corresponding to or any given letter of + ;
i.e., there cannot be two arcs leaving a state both with the same
rst letter (i.e., a Turing machine is deterministic).
The second character (either or from ) is what the Tape Head
prints in the cell before it leaves.
CHAPTER 19. TURING MACHINES 19-4
The third component, the direction, tells the Tape Head whether
to move one cell to the right, R, or one cell to the left, L.
Remarks:
The above denition does not require that every state has an edge leaving
it corresponding to each letter of + .
If we are in a state and read a letter for which there is no arc leaving
that state corresponding to that letter, then the machine crashes. In this
case, the machine terminates execution unsuccessfully.
To terminate execution successfully, machine must be led to a HALT
state. In this case, we say that the word on the input tape is accepted
by the TM.
If Tape Head is currently in cell i and the program tells the Tape Head
to move left, then the machine crashes.
Our denition of TMs requires them to be deterministic. There are also
non-deterministic TMs. When we say just TM, then we mean our
above denition, which means it is deterministic.
Denition: A string w

is accepted by a Turing machine if the following

occurs: when w is loaded onto the Tape and the machine is run, the TM ends
in a Halt state.
Denition: The language accepted by a Turing machine is the set of accepted
strings w

.
CHAPTER 19. TURING MACHINES 19-5
Example: Consider the following TM with input alphabet = a, b and
tape alphabet = a, b:
START 1 2 3 HALT 4
(a,a,R)
(b,b,R)
(b,b,R)
(b,b,R)
(a,a,R)
( , ,R)
and input tape containing input aba
i
a b a
ii iv v vi iii
We start in state START 1 with the Tape Head reading cell i, and we
denote this by
1
aba
The number on top denotes the state we are currently in. The things
below represent the current contents of the tape, with the letter about
to be read underlined.
After reading in a in state 1, the TM then takes the top arc from state 1
to state 2, and so it prints a into the contents of cell i and the Tape Head
moves to the right to cell ii. We record this action by writing
1
aba

2
aba
CHAPTER 19. TURING MACHINES 19-6
The tape now looks like
i
a b a
ii iv v vi iii
Now we are in state 2, and the Tape Head is pointing to cell ii. Since
cell ii contains b, we will take the arc from state 2 to state 3, print b in
cell ii, and move the Tape Head to the right to cell iii. We record this
action by writing
1
aba

2
aba

3
aba
The tape now looks like
i
a b a
ii iv v vi iii
Now we are in state 3, and the Tape Head is pointing to cell iii. Since
cell iii contains a, we will take the arc labeled (a, a, R) from state 3 back
to state 3, print a in cell iii, and move the Tape Head to the right to
cell iv, which contains a blank . We record this action by writing
1
aba

2
aba

3
aba

3
aba
The tape now looks like
CHAPTER 19. TURING MACHINES 19-7
i
a b a
ii iv v vi iii
Now we are in state 3, and the Tape Head is pointing to cell iv. Since
cell iv contains , we will take the arc labeled (, , R) from state 3 to
state HALT 4, print in cell iv, and move the Tape Head to the right
to cell v, which contains a blank . We record this action by writing
1
aba

2
aba

3
aba

3
aba
HALT
Since we reached a HALT state, the string on the input tape is accepted.
Note that if an input string has a as its second letter, then the TM
crashes, and so the string is not accepted.
This TM accepts the language of all strings over the alphabet = a, b
whose second letter is b.
CHAPTER 19. TURING MACHINES 19-8
Example: Consider the following TM with input alphabet = a, b and
tape alphabet = a, b:
START 1 2 HALT 3
(a,a,R)
(b,b,R)
(a,a,R)
(b,b,R)
( , , R)
Consider processing the word baab on the TM
Note that the rst cell on the TAPE contains b, and so upon reading
this, the TM writes b in cell i, moves the tape head to the right to
cell ii, and then the TM loops back to state 1,
The second cell on the TAPE contains a, and so upon reading this,
the TM moves to state 2, writes a in cell ii, and moves the tape
head to the right to cell iii.
The third cell on the TAPE contains a, and so upon reading this,
the TM writes a in cell iii, moves the tape head to the right to
cell iv, and moves to state 3, which is a HALT state
The TM now halts, and so the string is accepted. Note that the
input tape still has a letter b that has not been read.
Consider processing on the TM the word bba.
Note that each of the rst two bs results in the TM looping back
to state 1 and moving the tape head to the right one cell.
The third letter a makes the TM go to state 2 and moves the tape
head to the right one cell.
CHAPTER 19. TURING MACHINES 19-9
The fourth cell of the TAPE has a blank, and so the TM then
crashes. Thus, bba is not accepted.
Consider processing on the TM the word bab.
Note that the rst letter b results in the TM looping back to state 1
and moving the tape head to the right one cell.
The tape head then reads the a in the second cell, which causes the
TM to move to state 2 and moves the tape head to the right one
cell.
The tape head then reads the b in the third cell, which causes the
TM to move back to state 1 and moves the tape head to the right
one cell.
The fourth cell of the TAPE has a blank, and so the TM returns
to state 1, and the tape head moves one cell to the right.
All of the other cells on the TAPE are blank, and so the TM will
keep looping back to state 1 forever.
Since the TM never reaches a HALT state, the string bab is not
accepted.
In general, we can divide the set of all possible strings into three sets:
1. Strings that contain the substring aa, which are accepted by the
TM since the TM will reach a HALT state.
2. Strings that do not contain substring aa and that end in a. For
these strings, the TM crashes, and so they are not accepted.
3. Strings that do not contain substring aa and that do not end in
a. For these strings, the TM loops forever, and so they are not
accepted.
Note: The videotaped lecture contains an error about this point.
Let S
1
be the set of strings that do not contain the substring
aa and that do not end in a.
Let S
2
be the set of strings that do not contain the substring
aa and that end in b.
In the videotaped lecture, I said that S
2
is the set of strings
for which the TM loops forever, but actually, S
1
is the set of
strings for which the TM loops forever.
CHAPTER 19. TURING MACHINES 19-10
Note that S
1
,= S
2
since S
1
but , S
2
.
This TM accepts the language having regular expression (a+b)

aa(a+
b)

.
Denition: Every Turing machine T over the alphabet divides the set of
input strings into three classes:
1. ACCEPT(T) is the set of all strings w

such that if the Tape

initially contains w and T is then run, T ends in a HALT state. This is
the language accepted by T.
2. REJECT(T) is the set of all strings w

such that if the Tape

initially contains w and T is then run, T crashes (by either moving left
from cell i or by being in a state that has no exit edge labeled with the
letter that the Tape Head is currently pointing to).
3. LOOP(T) is the set of all strings w

such that if the Tape initially

contains w and T is then run, T loops forever.
So for our last example,
ACCEPT(T) = set of strings generated by the regular expression (a +
b)

aa(a +b)

.
REJECT(T) = strings in

that do not contain the substring aa and

that end in a, where = a, b.
LOOP(T) = strings in

that do not contain the substring aa and that

do not end in a, where = a, b.
CHAPTER 19. TURING MACHINES 19-11
Example: Below is a TM for the language L = a
n
b
n
a
n
: n = 0, 1, 2, . . .:
= a, b, = a, b,
START 1
( , ,L)
(a, ,L) (a, ,L)
( , ,R)
HALT
2
(a,a,R)
(a,a,L) (b,b,R)
(b,a,R)
(b,b,L)
(a,a,L)
(a,a,R) (b,b,R)
3 4
5
6 7 8
(a,*,R)
(*,*,R)
We now examine why this TM accepts the language L dened above.
Step 1. Presume that we are in state 1, and we are reading the rst letter of
what remains on the input.
So initially, we are reading the rst letter on the input tape, but as
we progress, we may nd ourselves back in state 1 reading the rst
letter of what remains on the tape.
If we read a blank, then we go to HALT.
If what we read is a, then change it to , and move the tape head
to the right.
If we read anything else, we crash.
CHAPTER 19. TURING MACHINES 19-12
Step 2. In state 2, we skip over the rest of the as in the initial clump of as,
looking for the rst b.
When we nd the rst b, we move to state 3.
As long as we keep reading bs, we keep returning to state 3.
When we nd the rst a in the second clump of as, we then go to
state 4, and move the tape head to the left back to the last b in the
clump of bs.
We then change the last b to an a, move the tape head to the right,
and go to state 5. So now the number of bs has reduced by 1, and
the number of as in the second clump of as has increased by 1.
We did all of these last few steps to nd the last b in the clump of
bs.
Step 3. Now we are in state 5 with the tape head pointing to the rst a in
the second clump of as, and we want to nd the last a in the second
clump of as.
Each a that we now read makes us return back to state 5, and move
the tape head to the right.
If we read b, then the machine crashes.
When we nally encounter , then the tape contains no more char-
acters to the right, and the TM goes to state 6.
We then move to state 7 and then to state 8, and change the last
two as to s.
Thus, the number of as in the second clump has decreased by 2.
But in Step 2, we increased the number of as in the second clump by
1, and so now the number of as in the second clump has decreased
by 1 since we started in Step 1.
Recall that Step 2, we also reduced the number of bs by 1.
Recall that in Step 1, we changed the rst a in the rst clump of
as to .
Step 4. Now we are in state 8 with the tape head pointing to the last a
currently in the second clump of as, and we want to get back to the rst
a that is currently in the rst clump of as.
CHAPTER 19. TURING MACHINES 19-13
In state 8, as long as we keep reading as and bs, we move the tape
head to the left, and return back to state 8.
Recall that in Step 1, we changed the rst a in the rst clump of
as into .
So when the tape head nally reaches the rightmost by moving
left, the TM goes to state 1, and we move the tape head to the
right, and we repeat our 4 steps.
19.2 Stupid TM Tricks
There are several tricks that one can do with Turing Machines:
1. Storing information in states; e.g., check if rst letter of input string
appears later in the string.
2. Multiple tracks on tape.
3. Checking o symbols on tape.
4. Inserting a character anywhere onto the input tape and shifting over the
rest of the contents of the tape.
Example: TM to
insert c at the beginning of Tape,
shift entire original contents of Tape one cell to the right, and
nish with Tape Head pointing at cell i.
= a, b, = a, b, A, B, c.
CHAPTER 19. TURING MACHINES 19-14
The language of this TM is all strings over = a, b since starting the
TM with any string in

loaded on the Tape and running the TM will

lead to the Halt state.
5. Can design TM to delete contents of specic cell and shift contents of
all cells to the right one cell to the left.
6. Subroutines.
7. Can convert any FA into a TM.
8. Can convert any PDA into a TM.
Theorem 46 If L is a regular language, then there exists a TM for L.
Chapter 23
TM Languages
23.1 Recursively Enumerable Languages
Denition: A language L over an alphabet is called recursively enumerable
if there is a TM that accepts every word in L and either rejects (crashes) or
loops forever for every word in L

; i.e.,
accept(T) = L,
reject(T) + loop(T) = L

.
In other words, the class of languages that are accepted by a TM is exactly
those languages that are recursively enumerable.
Denition: A language L over an alphabet is called recursive if there is a
TM that accepts every word in L and rejects every word in L

; i.e.,
accept(T) = L,
reject(T) = L

,
loop(T) =
23.2 Church-Turing Thesis
There is an eective procedure to solve a decision problem if and only if there
is a Turing machine that halts for all input strings and solves the problem.
23-1
CHAPTER 23. TM LANGUAGES 23-2
23.3 Encoding of Turing Machines
Can take any pictorial representation of a TM and represent it as two tables
of information.
Example: For the following TM
START 1 2 HALT 3
(a,a,R)
(b,b,R)
(a,a,R)
(b,b,R)
( , , R)
we can represent it as the following tables:
State Start? Halt?
1 1 0
2 0 0
3 0 1
From To Read Write Move
1 1 R
1 1 b b R
1 2 a a R
2 1 b b R
2 3 a a R
Remarks:
We can do this encoding for any TM. We call this an encoded Turing
machine.
CHAPTER 23. TM LANGUAGES 23-3
The encoding can be written as just a string of characters.
For example, we can write the above encoding as
110200301%11R11bbR12aaR21bbR23aaR
where we use the % to denote where the rst table ends and the second
one begins.
The textbook converts the above string into a string of as and bs, which
we wont do.
Thus, we can represent any TM as a string of characters, which we can
think of as a program.
We can use the encoded TM as an input string to another TM, just as a
C++ program is an input string to a C++ compiler, which itself is just
a program.
In particular, a copy of a program may be passed to itself as input.
For our above example, the string
110200301%11R11bbR12aaR21bbR23aaR
is rejected by the TM since it crashes on the rst letter.
23.4 Non-Recursively Enumerable Language
Theorem 64 Not all languages are recursively enumerable.
Proof.
Let L
N
be the set of strings w that are encoded TMs for which w is not
accepted by its own TM.
For example, the string 110200301%11R11bbR12aaR21bbR23aaR is
in L
N
since it was not accepted by its own TM.
We will prove by contradiction that L
N
is not recursively enumerable.
Suppose that L
N
is recursively enumerable.
CHAPTER 23. TM LANGUAGES 23-4
Then there exists a TM T
N
for L
N
.
Let P be the encoded TM of TM T
N
.
There are 2 possibilities: either TM T
N
accepts P or TM T
N
doesnt
accept P.
If T
N
accepts P,
then P , L
N
since L
N
consists of strings w that are encoded TMs
such that w is not accepted by its own TM.
But this is a contradiction since the TM T
N
is only supposed to
accept those strings in L
N
.
If T
N
doesnt accept P,
then P L
N
.
But this is a contradiction since T
N
should accept P since P L
N
.
Therefore, L
N
is not recursively enumerable.
23.5 Universal Turing Machine
Denition: A universal Turing machine (UTM) is a TM that can be fed as
input a string composed of 2 parts:
1. The rst is any encoded TM P, followed by a marker, say $.
2. The second part is a string w called the data.
The UTM reads the input, and then simulates P with input w.
Theorem 65 UTMs exist.
Remarks about UTMs
CHAPTER 23. TM LANGUAGES 23-5
The reason that UTMs are important is that they allow one to write
programs; i.e., UTMs are programmable, just like real computers.
We dont have to build a new Turing machine for each problem.
For a proof of Theorem 65, see pp. 554557 of Cohen.
23.6 Halting Problem
Theorem 69 There is no TM that can accept any encoded TM P and any
input string w for P and always decide correctly whether P halts on w; i.e.,
the halting problem cannot be decided by a TM.
Basic Idea:
Dene halting function H(P, w), where
P is encoding of program (i.e., encoded Turing machine)
w is intended input for P.
Let H(P, w) = yes if P halts on input w.
Let H(P, w) = no if P does not halt on input w.
Assume that a program computing H(P, w) exists.
Construct a program Q(P) with input P:
1. x = H(P, P)
2. While x = yes, goto step 2.
Now run program Q with input P = Q.
Suppose Q(Q) halts. Then H(Q, Q) = yes, but Q is stuck in innite
loop and so it doesnt halt.
Suppose that Q(Q) doesnt halt. Then H(Q, Q) = no, while in fact
Q(Q) halts.
Therefore H(P, w) cannot exist.
CHAPTER 23. TM LANGUAGES 23-6
Proof. (of Theorem 69)
Suppose there is a TM, call it H, to solve the halting problem; i.e., H
works as follows:
Recall that all TMs take a TAPE loaded with an input string.
Our TM H takes as its input an encoded TM P and an input string
w to be used with P.
So we have to specify how P and w can be specied as an input
string to H.
We do this by taking P and rst concatenating it with a special
character, say #, and then concatenating this with the input string
w. We use the # to mark the end of the encoded TM and the
beginning of the input string.
Thus, we now have a single long string P#w.
If we feed the string P#w into H, then
if P halts on w, then H prints yes somewhere on the TAPE.
if P does not halt on w, then H prints no somewhere on the
TAPE.
See p. 449 of the textbook to see how to print characters on
the TAPE.
Now suppose that we create another encoded TM Q that takes an en-
coded TM P as input and uses H as a subroutine as follows:
Since P is the input, the TAPE initially contains P.
First modify the TAPE so that it now contains P#P. (See p. 449
of the textbook to see how this can be done.)
Then run H using input P#P.
if TM H prints yes on input P#P, then loop forever;
if TM H prints no on input P#P, then halt.
Now run Q with input P = Q.
Suppose Q halts on input Q.
CHAPTER 23. TM LANGUAGES 23-7
This means that H prints no on input Q#Q.
But this means that the encoded TM Q does not halt on input Q,
which is a contradiction.
Suppose Q does not halt on input Q.
This means that H prints yes on input Q#Q.
But this means that the encoded TM Q halts on input Q, which
again is a contradiction.
Therefore, H cannot exist.
23.7 Does TM Accept ?
Theorem 70 There is no TM that can decide for every encoded TM T whether
or not T accepts the word ; i.e., the blank-tape problem for TMs is undecid-
able.
Proof.
We will prove this by contradiction.
Suppose that there is a TM, call it B, that can decide for every encoded
TM T whether or not T accepts the word ; i.e., whether T halts when
it starts with a blank tape.
Function B(T), where T is encoding of program (i.e., encoded Tur-
ing machine)
B(T) = yes if T halts on input .
B(T) = no if T does not halt on input .
Dene a new program M(P, w), with input P and w, where P is any
encoded Turing machine and w is any input string:
CHAPTER 23. TM LANGUAGES 23-8
First construct new program P
w
that starts with blank input tape
and works as follows:
First P
w
writes w on the input tape.
Then P
w
positions the tape head back to the beginning of the
tape.
Finally P
w
simulates program P with w on the input tape.
Call B(P
w
), and return M(P, w) = B(P
w
).
Since we started P
w
with a blank tape, we can apply program B to P
w
to see if it halts.
Clearly, P
w
will halt on a blank tape if and only if P halts on w.
If P
w
halts on blank tape (i.e., if B(P
w
) = yes), then P halts on w.
If P
w
does not halt on blank tape (i.e., if B(P
w
) = no), then P does
not halt on w.
Note that M(P, w) solves the halting problem.
But the halting problem is undecidable, and so B cannot exist.
23.8 Does TM Accept Any Words?
Theorem 71 There is no TM that can decide for every encoded TM T whether
or not T accepts any words at all; i.e., the emptiness problem for TMs is un-
decidable.
Proof.
We will prove this by contradiction.
Suppose that there is a TM, call it N, that can decide for every encoded
TM T whether or not N accepts any words at all; i.e., whether the
language L of T is L ,= .
CHAPTER 23. TM LANGUAGES 23-9
Function N(T), where T is encoding of program (i.e., encoded Tur-
ing machine)
N(T) = yes if T accepts language L ,= .
N(T) = no if T accepts language L = .
Dene a new program E(P), with input P, which is any encoded Turing
machine:
First construct new program P

that works as follows:

First P

erases contents of input tape.

Then P

positions the tape head back to the beginning of the

tape.
Finally P

simulates program P with blank input tape.

Call N(P

), and return E(P) = N(P

)
Suppose E(P) = yes. Then
N(P

) = yes.
This implies P

accepts at least one word.

But since P

always erases whatever is on tape and then simulates

P, this means that P accepts .
Suppose E(P) = no. Then
N(P

) = no.
This implies P

does not accept any words; i.e., the language of P

is .
But since P

always erases whatever is on tape and then simulates

P, this means that P does not accept .
Therefore, E(P) solves the blank-tape problem.
But Theorem 70 says this is impossible, and so N cannot exist.
Chapter 24
Review
24.1 Topics Covered
1. Languages
is alphabet with nite number of symbols.
Languages are sets of strings over
For sets S
1
and S
2
, can dene
union S
1
+S
2
= w : w S
1
or w S
2

intersection S
1
S
2
= w : w S
1
and w S
2

product S
1
S
2
= w = w
1
w
2
: w
1
S
1
, w
2
S
2

subtraction S
1
S
2
= w : w S
1
, w , S
2

For any set S of strings over an alphabet , can dene

Kleene closure S

= w = w
1
w
2
w
n
: n 0, w
i
S i =
1, 2, . . . , n
complement S

= w

: w , S
S

= S

2. Regular expressions
3. FA = (K, , , s, F), where
K is nite set of states
is the alphabet
24-1
CHAPTER 24. REVIEW 24-2
: K K is the transition function
s is the initial state
F is the set of nal states.
4. TG = (K, , , S, F),
where
K is nite set of states
is the alphabet
K K is the transition relation
S is the set of initial states
F is the set of nal states.
5. Kleenes Theorem
Any language that can be dened by a
regular expression
FA
TG
can be dened by all three methods.
Given FA for L
1
and L
2
, can construct FA for
L
1
+L
2
L
1
L
2
L

1
Algorithm for generating regular expression from FA.
Nondeterminism
6. FA with output
Moore machine
Mealy machine
These are equivalent.
7. Regular languages
If L
1
and L
2
are RL, then so are
L
1
+L
2
CHAPTER 24. REVIEW 24-3
L
1
L
2
L

1
L

1
L
1
L
2
8. Nonregular languages
Pumping lemma.
9. Decidability
Can tell if two FAs, FA
1
and FA
2
, generate the same language by
checking if either of the following accepts any words:
FA

1
FA
2
FA
1
FA

2
There are eective procedures to decide if
an FA accepts a nite or innite language
a regular expression generates an innite language
an FA has language
a regular expression generates language
10. CFG
CFG G = (, , R, S), where
is the nite set of terminals, i.e., the alphabet
is nite set of nonterminals
R (+)

is nite set of productions, where (N, U) R

is written N U.
S is the starting nonterminal.
CFG used to generate languages.
If L is regular, then L is CFL.
Some, but not all, nonregular languages are CFL.
Trees
Can eliminate ambiguity in meaning by using tree to show how
word was derived using CFG.
11. Grammatical Format
CHAPTER 24. REVIEW 24-4
Denition: A CFG G is a regular grammar if every production
N U in G has U

.
If a CFG is a regular grammar, then the CFL is a regular language.
(Can do this by converting regular grammar into TG.)
Chomsky Normal Form: A CFG G is in CNF if every production
N U in G has U + .
12. PDA
Every regular language L is accepted by some PDA.
13. CFG = PDA
14. CFLs
If L
1
and L
2
are CFLs, then so are
L
1
+L
2
L
1
L
2
L

1
However, CFLs are not closed under intersection or complements; i.e.,
there are examples of CFLs L
1
and L
2
such that L
1
L
2
is not context-
free, and there are examples of CFLs L such that L

is not context-free.
15. Decidability for CFLs
Membership is decidable for CFLs; i.e., for any CFG G and string
w, can decide if G generates w (using CYK algorithm).
16. Turing Machines
Following problems are undecidable:
Halting problem
Whether arbitrary TM halts on a blank tape
Whether arbitrary TM accepts any words
Whether arbitrary TM accepts nite or innite language.