# Legend

Language, Automata: , \alpha : usually denotes a string in this course. , \beta : usually denotes a string in this course. , \delta : usually denotes a transition function in this course. , \sigma : usually denotes a symbol in an alphabet in this course. , \Delta : usually denotes a blank space in this course. , \Gamma : usually denotes a set of stack symbols in this course. , \Lambda : usually denotes an empty string in this course. , \Pi : usually denotes a partition in this course. , \Sigma : usually denotes an alphabet in this course. , \goto : usually denotes a (one step) transition in this course. Logic: , ~ : logical not , ^ : logical and , V : logical or , -> : logical imply , <-> : logical if and only if (equivalent) , => : logical tautologically imply , <=> : logical tautologically equivalent , \A : logical for all , \E : logical for some (there exists)

Sets: , \in : belongs to , \not\in : does not belong to , @ : empty set U, : universal set , \subset : proper subset , \not\subset : not a proper subset , \subseteq : subset , \not\subseteq : not a subset , \cup : set union Ai , \cup(i=1 to n) A_i : union of n sets , \cap : set intersection Ai , \cap(i=1 to n) A_i : intersection of n sets , \bar A : complement of set A (A) , P(A) : power set of set A , X : Cartesian product Ai , X(i=1 to n) A_i : cartesian product of n sets Relation: < a, b > : ordered pair < a1, a2, ..., an > : ordered n-tuple , <= : precedes (partial order) Functions: xi , Sum(i=1 to n) x_i : sum of n xi's O(f) , O(f) : of order smaller than or equal to f

o(f) , o(f) : of order smaller than f (f) , Omega : of order greater than or equal to f (f) , omega : of order greater than f (f) , Theta : of the same order as f f(x) , lim(x -> inf) f(x) : limit of f as x goes to infinity

Introduction to Theoretical Computer Science
Today computers are used everywhere: banks, hospitals, schools, airline companies, gas stations, grocery stores, in our cars, in home appliances, PCs, etc., etc. Some are used to crunch numbers, some are used to process images, some are used to process other nonnumeric data and some are used to control operations of various devices. They can reason, they can prove many mathematical theorems, they can beat chess experts in their game, they can solve some very complex problems, they can understand our languages, they can answer our questions and of course they can crunch numbers much much faster than us. Let us for a moment call what computers do computation for convenience, though some of the things computers do such as controling appliances, answering our questions etc. don't fall into our traditional sense of computation. Then these computers seem to be able to compute an awfully lot of things if not everything. But are they capable of computing anything ? Are there things computers can not do ? If there are things computers can not do, what are they ? And why ? If there aren't things computers can not do, then how can we tell ? What do we exactly mean by computation ? Unfortunately there are many things computers can not do. Computers can not solve certain types of problems. For example no computer can tell in general whether or not a given computer program stops after a finite amount of time on a given input. They can not solve some other types of problems fast enough even though they can solve them in some finite amount of time. For example take the traveling salesman problem: a salesman is given a road map with distances between cities and wants to find a shortest round trip route that visits all the cities on the map exactly once. At the moment the so called traveling salesman problem requires an extremely large amount of time to solve. No one has been able to find a reasonably fast algorithm to solve it and the consensus is that it is not likely that anyone can find such an algorithm. I have just given you an example of the problems that computers could not solve. How do we know that that is the case ? Are there other problems like that ? How can we tell whther or not a given problem can be solved and solved fast enough ?

It turns out that solving a problem can be viewed as recognizing a language. After briefly studying context-free languges. Those four languages are together called formal languages. The languages Turing machines recognize are called Type 0 (or phrase structure) languages (regular languages are Type 3) and they are more complex than regular languages. that is to study limitations of computers and computation. Though it has not been proven. Thus they are a powerful tool to design and study those systems with. There we learn how computers can be simulated by Turing machines and what it means that a Turing machine recognizes (decides) a language. every task modern computers perform can also be accomplished by Turing machines. Then with Turing machines we investigate limitations of computers and computations. The other two are context-free languages and context-sensitive languages. ways to describe them and how to use them to model many of the real life systems. We are going to learn their properties.The main objective of this course is to answer those questions. it is generally believed (Church's thesis) that any "computation" humans do can be done by Turing machines and that "computation" is the computation performed by Turing machines. which are also heavily used in practice. We are going to investigate limitations of computers and computations by studying the essence of compuers and computations rather than all the variations of computer and computation. we study a simpler type of computing device called finite automata. Thus when a finite automaton is processing strings. Finite automata process strings. The languages that are recognized by finite automata are called regular languages. These two type of languages belong to a hierarchy of four languages called Homsky hierarchy. which is the key to the unsolvability of problem by computers. This essence is a device called Turing machine. We call a set of strings (of symbols) a language. In particular we are going to see a few problems that can not be solved by Turing machines hence by computers and how we can tell that they are unsolvable. Finite automata are very similar to Turing machines but a few restrictions are imposed on them. We say finite automata recognize languages. Turing machines also recognize languages. Consequently they are less capable than Turing machines but then their operations are simpler. In addition finite automata can model a large number of systems used in practice. we are also going to study regular languages in detail as well as finite automata. Thus by studying Turing machines we can learn capabilities hence limitatgions of computers. we go to Turing machines. . Since so many systems in practice can be described by regular languages. Before proceeding to the study of Turing machines and their computations in this course. it can actually be solving a problem. More specifically they answer the question whether or not a given string belongs to a language. It is a very simple device but remarkably. So they provide a good introduction to our study of Turing machines. It was first conceived of by Alan Turing in early 20-th century.

We are going to see some of those which take a large amount of time.000 times it can handle only ten or so more larger problem sizes. The time needed to solve such a problem increases at least exponentially with the size of the problem as long as we use Turing machines (hence computers). Increasing the processor speed does not help much for such problems. Among the solvable problems there are problems that can be solved within a reasonable amount of time and there are problems that are known to require a finite but very large amount of time to solve. • • 2 is a odd number. Let us start with review of mathematics. • • • • Logic Sets Relations Functions Logic Proposition and Logical Connectives "Proposition" can be defined as a declarative statement having a specific truth-value. 4 is a perfect square. The time complexity issues are investigated using Turing machines so that the results apply to all computers. then even if the processor speed increased 1. then with the increase in the processor speed of 1. Example: The following statements are propositions as they have precise truth values. If the computation time is 2n where n is the size of the problem. Their truth values are false and true.Our last topic is time complexities of various problems. respectively. Basic Mathematical Objects Back to Table of Contents The following are the contents of this introductory chapter.000 times 110 cities would already be too many. Unfortunately there is nothing we can do to speed them up. true or false. For example for the traveling salesman problem if 100 cities were too many to solve fast enough. "Connective": Two or more propositions can be combined together to make compound propositions with the help of logical connectives. .

Conjunction The logical conjunction is understood in the same way as commonly used ôandö. • • 2 is an odd number AND 4 is a perfect square. . Truth table is given below p T T F F q T F T F pVq T T T F c. For the first compound proposition to be true both the propositions have to be true as the connective is AND and as OR is the connective for the second one if either of the propositions is true the truth value of the compound proposition is true. Negation This is the logical "negation" and it is expressed by Truth table is given below as p for "not p". It is represented as " ^ ".Example: Above two propositions can be used to make a compound proposition using any of the logical connectives. Disjunction This is logical "or" read as either true value of the individual propositions. Their truth vales are false and true respectively. The compound proposition truth-value is true iff all the constituent propositions hold true. Truth table for two individual propositions p and q with conjunction is given below p T T F F q T F T F p^q T F F F b. 2 is an odd number OR 4 is a perfect square. The following are the logical connectives used commonly: a.

"p is sufficient for q" .: p V p g. then p is said to logically imply q. E. It is represented as p => q.p T F p F T d. Tautology A compound proposition. "q is necessary for p".g. which is false in every case.e. e. For example. "p only if q" . Biconditional A proposition (p q) ^ (q p) can be abbreviated using biconditional conjunction as p q and is read as "if p then q. Contradiction This is the opposite of tautology. then q" Truth table is given below p T T F F q T F T F p T F T T q p -> q is also expressed in a number of different (but equivalent) ways in English. "q is a necessity/consequence of p" and "q whenever p" are all differnt ways of saying "if p then q". If p and q have the same truth-value in every case then they are said to be logically equivalent and it is represented as p <=> q. f. which is true in every case.g. p q is read as "if p. E.: p ^ p Logical implication and equivalence If the value of p -> q is true in every case. Following are some of the useful identities and implications from propositional logic: Identities . and if q then p". Conditional This is used to define as "a proposition holds true if another proposition is true" i. "if not q then not p" .

(P 4.contrapositive For explanations.implication Q) R] [P (Q R)] ----. The phrase "is blue" is a predicate and it describes the property of being blue. . For example. (P 2. A predicate is a template involving a verb that describes a property of objects. and "The cover of this book is blue" come from the template "is blue" by placing an appropriate noun/noun phrase in front of it. examples and proofs of these identities go to Identities Implications 1. examples and proofs of these implications go to Implications Predicate and Predicate Logic The propositional logic is not powerful enough to represent all types of assertions that are used in computer science and mathematics.DeMorgan's Law Q) ( P Q) ----. or a relationship among objects represented by the variables.exportation Q) ( Q P) ----. [(P 2. "The sky is blue". [(P Q) Q] P ----. [(P 3. (P 3.DeMorgan's Law Q) ( P Q) ----. or to express certain types of relationship between propositions such as equivalence ( for more detail click here for example for example ). the sentences "The car Tom is driving is blue". The predicate logic is one of the extensions of propositional logic and it is fundamental to most other types of logic. Central to the predicate logic are the concepts of predicate and quantifier. [(P 5. (P Q) ( P Q) ----. For more complex reasoning we need more powerful logic capable of expressing complicated propositions and reasoning.modus tollens Q) (R S)] [(P R) (Q S)] Q) (Q R)] (P R) For explanations.1.

For example. for example. The universal quantifier turns. For example any of "is_blue". a quantification is performed on formulas of predicate logic (called wff ). This new statement is true or false in the universe of discourse. and it becomes a true statement. B(x) reads as "x is blue". . quantify the variable using a quantifier (see below). the statement x > 1 to "for some object x in the universe. Similarly the existential quantifier turns. x > 1". assign a value to the variable 2. the statemen t x > 1 to "for every object x in the universe. also called universe . A predicate with variables. The propositions in the predicate logic are statements on objects of a universe. where x represents an arbitrary object. for example. hence a proposition. There are two types of quantifiers: universal quantifier and existential quantifier. by using quantifiers on variables . Hence it is a proposition once the universe is specified. it is true or false in the universe of discourse. sentences that assert an object is blue can be represented as "B(x)". which is expressed as " x x > 1. which is expressed as " x x > 1". x > 1". can be made a proposition by applying one of the following two operations to each of its variables: 1. is the set of objects of interest. In general. If we adopt B as the name for the predicate "is_blue". and hence it is a proposition once the universe is specified. x > 1 becomes 3 > 1 if 3 is assigned to x. Universe of Discourse The universe of discourse. The universe is thus the domain of the (individual) variables. such as x > 1 or P(x)." Again. called atomic formula. It can be the set of real numbers. "Blue" or "B" can be used to represent the predicate "is blue" among others.Predicates are often given a name.

see Reasoning(with predicate logic) and Quantifiers and Connectives in Discrete Structures course. x [ P(x) Q(x) ] [ x P(x) x Q(x) ] For more discussions and examples on these rules and others. But it should be obvious from the context. It allows one to reason about properties and relationships of individual objects. Predicate logic is more powerful than propositional logic. as well as those for propositional logic such as the equivalences. some of which are given below. Important Inference Rules of Predicate Logic: First there is the following rule concerning the negation of quantified statement which is very useful: x P(x) x P(x) Next there is the following set of rules on quantifiers and connvectives: 1. [ x P(x) x Q(x) ] x [ P(x) Q(x) ] 3. x [ P(x) Q(x) ] [ x P(x) x Q(x) ] 2. one can use some additional inference rules. the set of all cars on a parking lot. x [ P(x) Q(x) ] [ x P(x) x Q(x) ] 4. Also for proof and proof techniques see Mathematical Reasoning.the set of integers. the set of all students in a classroom etc. In predicate logic. The universe is often left implicit in practice. implications and inference rules. Sets .

having a property that characterizes those elements.5} Alternate way is to give the properties that characterize the elements of the set. Also. All the elements belonging to the set are explicitly given. Using this notation we can specify the set {0. Note: If A is a subset of B and B is a subset of A then A=B.4. A is a subset of B. Universal Set The set U of all the elements we might ever consider in the discourse is called the universal set.2.What is a set? Set is a group of elements. Example: A = {1.2. How to specify a Set? One way is to enumerate the elements completely. Set terminology Belongs To x B means that x is an element of set B. if A is a subset of.3.3. A is a subset of B is represented as A B. if every element of A is an element of B. It is read as "the set of natural numbers that are less than or equal to 5". Subset Let A and B be two sets. Example: B = {x | x is a positive integer less than or equal to 5} Some sets can also be defined recursively. . but not equal to B represented as A B.4} call it Z by writing Z = {x | x N | x 5} where N represents the set of natural numbers.1.

2} Note that in general A . where means " is not an element of ".8.2.2.5} then A B = {1. then the difference of A from B is the set that consists of the elements of A that are not in B. For A and B of the above example B . then A' = { x | x U ^ x > 3}. It is denoted by A .3} B = {3.8}.Complement If A is a set.A = {4.8} B = {3. Example: If U is the set of natural numbers and A = { 1. Example: If A = {1. then the intersection of A and B is the set that consists of the elements in both A and B . It is denoted by A B.2. and B = { 6. Thus A' = { x | x U ^ x A } .5} 2.B B . Disjoint sets A and B are said to be disjoint if they contain no elements in common i.2.e.5 } .B = {1.2. Example: If A = {1. A B= ø.4. Example: If A = {1.3 } . Difference If A and B are two sets.2. then the complement of A is the set consisting of all elements of the universal set that are not in A.4. then the union of A and B is the set that contains all the elements that are in A and B including the ones in both A and B. where ø is the Empty set.3.4.A .5} then A .4. Set Operations The operations that can be performed on sets are: 1. Intersection If A and B are two sets. It is denoted by A' or .4. 3.3. It is denoted by A B.. Example: A = { 1.8} then A B = {3.5.B.3} and B = {3. Union If A and B are two sets.9 } are disjoint.5} .3.

The Commutative laws: A B=B A A B=B A The Associative laws: A (B C) = (A A (B C) = (A The Distributive laws: A (B C) = (A A (B C) = (A The Idempotent laws: A A=A A A=A The Absorptive laws: A (A B) = A A (A B) = A The De Morgan laws: (A B)' = A' B' (A B)' = A' B' Other laws involving Complements: ( A' )' = A A A A' = ø A' = U B) B) B) B) C C (A (A C) C) Other laws involving the empty set A A ø=A ø=ø Other laws involving the Universal Set: A U=U A U=A Venn Diagrams A common technique in working with Set Operations is to illustrate them by drawing Venn Diagrams. It is a very good tool to get a general idea.Following is a list of some standard Set Identities A. C represent arbitrary sets and ø is the empty set and U is the Universal Set. B. .

that Venn Diagrams must NOT be used for rigorous discussions.3.4 } and B = { 6.Note.8.2. The idea of Venn Diagram is to draw a region representing the universe and within that to draw the regions representing the component sets we are starting with so that the resulting diagram describes their interrelationships.2. however. For example sets A = { 1.4 } can be represented as shown below using Venn Diagrams: Set A U represents the Universal set in which A is one of the Set. because they can represent only very limited situations and miss many other possibilities. .

Set B The following Venn Diagram is used to illustrate A B .

A B .

The following Venn Diagram is used to illustrate A U B A A B B is the set consisting of all the different elements in A and B. .

5.8 } .2.6.(A B) = { 5.6.(A B)' is the yellow region in the Venn diagram given below.8 } A = { 1.2. For example: U = { 1.3. 7 } B = { 2.7.4.4 } A B = { 1.8 } (A B)' = U .4.6.3.3.2.4.

For example expressions we often use A holds. A2 . This can be generalized for the union of any finite number of sets as A1 An .. intersection and Cartesian product of sets are associative.B is the yellow shaded region and B .A .A is the blue shaded region in the Venn Diagram shown below Generalized Set Operations Union... To denote either of these B C. which we write as Ai This generalized union of sets can be rigorously defined as follows: Definition ( Ai) : .

The basis clause (or simply basis) of the definition establishes that certain objects are in the set. Based on these definitions. Inductive Clause: Ai = ( Ai) An+1 Ai and generalized Cartesian product Similarly the generalized intersection Ai can be defined. .Basis Clause: For n = 1 . This part of the definition specifies the "seeds" of the set from which the elements of the set are generated using the methods given in the inductive clause. Ai = A1. De Morgan's law on set union and intersection can also be generalized as follows: Theorem (Generalized De Morgan) = . and = Recursive Definition Recursive Definition Subjects to be Learned • • • • • recursive/inductive definition basis clause basis inductive clause extremal clause A recursive definition of a set always consists of three distinct clauses: 1. The set of elements specified here is called basis of the set being defined.

1. Note that if we don't have (3). Following this definition. then they can be combined in certain specified ways to create other objects. The Set of Nonnegative Even Numbers Basis Clause: Inductive Clause: For any element x in . Extremal Clause: Nothing is in unless it is obtained from the Basis and Inductive Clauses. The x + 1 in the Inductive Clause is the parent of x. 3.5. Let us call the objects used to create a new object the parents of the new object. Examples of Recursive Definition of Set Example 1. The Set of Even Integers Basis Clause: . 1 is the parent of 2. Extremal Clause: Nothing is in unless it is obtained from the Basis and Inductive Clauses. which is not what we want as the set of natural numbers. The Set of Natural Numbers Basis Clause: Inductive Clause: For any element x in . since 0 is in N. 0. The inductive clause (or simply induction) of the definition establishes the ways in which elements of the set can be combined to produce new elements of the set. 0 is put into N. Example 2. The inductive clause always asserts that if objects are elements of the set. Then by (2). Proceeding in this manner all the natural numbers are put into N.. the object is not a member of the set.2. Example 3. 0 is the parent of 1. The extremal clause asserts that unless an object can be shown to be a member of the set by applying the basis and inductive clauses a finite number of times. can be included in N. and 1 is the child of 0. 1 + 1 (= 2) is in N. x + 2 is in . . and 2 is the child of 1. The basis for this set N is { 0 } .. x + 1 is in . 2. and the new object is their child . and x is the child of x + 1.5.5. the set of natural numbers N can be obtained as follows: First by (1). 0 + 1 (= 1) is in N. Then by (2) again.

.Inductive Clause: For any element x in . and . or 1). bbabaa. They are all on functions from integer to integer except the last one. Here ax means the concatenation of a with x. and x . say x. Basis Clause: . . simplest expressions. Recursive Definition of Function Some functions can also be defined recursively. The set of propositions (propositional forms) can also be defined recursively. Inductive Clause: For any element x in . The Set of Strings over the alphabet excepting empty string This is the set of strings consisting of a's and b's such as abbab. try simplest elements in the set such as smallest numbers (0. and generalize that generation process for the "Inductive Clause". A few examples are given below. Then the value of the function at an element. Extremal Clause: Nothing is in unless it is obtained from the Basis and Inductive Clauses. Extremal Clause: Nothing is in unless it is obtained from the Basis and Inductive Clauses. and . Then see how other elements can be obtained from them. etc. x + 2. Click Yes or No . There are two sets of questions. To see how it is defined click here. How to define function recursively: First the values of the function for the basis elements of the domain are specified. or shortest strings. of the domain is defined using its value at the parent(s) of the element x. Tips for recursively defining a set: For the "Basis Clause".2 are in . Example 4. Condition: The domain of the function you wish to define recursively must be a set defined recursively. then Submit. Test Your Understanding of Recursive Definition Indicate which of the following statements are correct and which are not.

See Example 5 for the extremal clause. L(xy) = L(x) + L(y) . Example 6: The function f(n) = 2n + 1 for natural numbers n can be defined recursively as follows: Basis Clause: f(0) = 1 Inductive Clause: For all natural number n. Example 7: The function f(n) = 2n for natural numbers n can be defined recursively as follows: Basis Clause: f(0) = 1 Inductive Clause: For all natural number n. because the set of natural numbers can be defined recursively and that has the extremal clause in it. Inductive Clause: For any string x and y of S. This function L gives the number of a's and b's L(x) for a string x is also often denoted by | x |. See above for the extremal clause. Hence 2 ! = 2 * 1 ! = 2 * 1 = 2 . L(a) = 1 and L(b) = 1. b} to the set of natural numbers that gives the length of a string can be defined recursively as follows: Basis Clause: For symbols a and b of the alphabet.Example 5: The function f(n) = n! for natural numbers n can be defined recursively as follows: Basis Clause: f(0) = 0! = 1 Inductive Clause: For all natural number n. 1 ! = 1 * 0 ! = 1 * 1 = 1 . 3! can be found as follows: Since 0 ! = 1. Using this definition. So there is no chance of other elements to come into the function being defined. See Example 5 for the extremal clause. Induction Mathematical Induction . where xy is the concatenation of strings x and y. f(n+1) = (n+1) f(n). Note that here Extremal Clause is not necessary. Example 8: The function L from the set S of strings over {a. f(n+1) = f(n) + 2 . Hence 3 ! = 3 * 2 ! = 3 * 2 * 1 = 6 . f(n+1) = 2 f(n) .

and the process of generating an element from a known element in the set. and its elements can be generated one by one starting with 0 by adding 1. When these two are proven.Subjects to be Learned • • • • • first principle of mathematical induction basis step induction hypothesis induction second principle of mathematical induction Contents First Priciple of Mathematical Induction As we have seen in recursion . has the property in question. the element next to it. + n = n( n + 1 )/2 can be proven by mathematical induction by the same token. has the same property again by the inductive step. then P(k+1) is true. the element next to it. To prove that a statement P(n) is true for all natural number natural number. that is n + 1. The first principle of mathematical induction states that if the basis step and the inductive step are proven. . has the property in question (basis step). which is 2. natural numbers can be proven to have certain properties as follows: First it is proven that the basis element. that is 0. This process is somewhat analogous to the knocking over a row of dominos with knocking over the first domino corresponding to the basis step. Then it is proven that if an arbitrary natural number. if P(k) is true (called induction hypothesis).. it is often a good idea to restate P(k+1) in terms of . Then since 1 has the property. As a first step for proof by induction. then the next element. then it follows that all the natural numbers have that property.. has the same property by the inductive step. More generally mathematical statements involving a natural number n such as 1 + 2 + . Proceeding likewise. any natural number can be shown to have the property. the set of natural numbers can be defined recursively. Thus the set of natural numbers can be described completely by specifying the basis element (0). then P(n) is true for all natural number . we proceed as follows: Basis Step: Prove that P( ) is true. where is a Induction: Prove that for any integer . For since 0 has the property by the basis step. which is 1. Taking advantage of this. denote it by n. has that property (inductive step).

Example 1: Let us prove the following equality using the second principle: For any natural number n . then LHS = 0. Hence LHS = RHS.. Here k [ k < n P(k) ] is the induction hypothesis. we get (n + 1)(n + 2) / 2 . and in the inductive step P(n) is proved assuming P(k) holds for all k < n . then n P(n) can be concluded. + ( 2n + 1 ) = ( n + 1 )2. which is equal to the RHS for n+1. first try to express LHS for n+1 in terms of LHS for n. + n + (n + 1) = (0 + 1 + . and somehow use the induction hypothesis. The reason that this principle holds is going to be explained later after a few examples of proof.. + n) + (n + 1) . Example: Prove that for any natural number n. and RHS = 0 * (0 + 1) = 0 . Formally the second principle of induction states that if n [ k [ k < n P(k) ] P(n) ] . 0 + 1 + . Using the induction hypothesis... Certain problems can be proven more easily by using the second principle than the first principle because P(k) for all k < n can be used rather than just P(n . End of Proof. Second Priciple of Mathematical Induction There is another form of induction over the natural numbers based on the second principle of induction to prove assertions of the form x P(x) . + n = n( n + 1 )/2 .. Thus LHS = RHS for n+1. Proof: Basis Step: If n = 0. Here let us try LHS for n + 1 = 0 + 1 + .1) to prove P(n). -------.. 0 + 1 + . .P(k) so that P(k).. This form of induction does not require the basis step. Factoring (n + 1) out. + n = n( n + 1 )/2 . can be used. the last expression can be rewritten as n( n + 1 )/2 + (n + 1) . which is assumed to be true.Induction Hypothesis To prove this for n+1... Induction: Assume that for an arbitrary natural number n.. 1 + 3 + .

. + ( 2n . Proof: Assume that for all positive integers k. If n is a prime number... let A = {blue shirt. Since n is an integer.1 for all k. say p and q.1 )! + n * n! = n! . More precisely..1 ) * ( n . which is a prime number. Therefore the statement holds true. + k * k! = ( k + 1 )! . A binary relation from A into B is any subset of the Cartesian product A x B. n > k > 1. the individual may wish to restrict .. = ( n + 1 )n! . Then 1 * 1! + 2 * 2! + . Relations Definition Relation Let A and B be sets. can be written as the product of prime numbers. then it is a product of two positive integers. tan slacks}.. + ( 2n + 1 ) = ( n + 1 )2 holds for all natural numbers. k < n. If n is not a prime number. + ( 2k + 1 ) = ( k + 1 )2 holds for all k.. mint green shirt} and B = {gray slacks. Example 2: Prove that for all positive integer n.1 Proof: Assume that 1 * 1! + 2 * 2! + . Since both p and q are smaller than n.1 Hence by the second principle of induction positive integers.. k < n. and itself..1 ) ) + ( 2n + 1 ) = n2 + ( 2n + 1 ) = ( n + 1 )2 by the induction hypothesis.Proof: Assume that 1 + 3 + . the individual can wear. However. Then 1 + 3 + . Let's assume that a person owns three shirts and two pairs of slacks. Hence by the second principle of induction 1 + 3 + . + ( n . Example1: nbsp..1 + n * n! by the induction hypothesis. We are going to prove that n can be written as the product of prime numbers.1 holds for all Example 3: Prove that any positive integer n > 1. Hence n can also be written as the product of prime numbers. i ( i! ) = ( n + 1 )! . by the induction hypothesis they can be written as the product of prime numbers (Note that this is not possible if the First Principle is being used).. it is either a prime number or not a prime number. k can be written as the product of prime numbers. + ( 2n + 1 ) = ( 1 + 3 + . Then certainly A x B is the set of all possible combinations (six) of shirts and slacks that nbsp. i ( i! ) = ( n + 1 )! . then it is the product of 1..

is the set of pairs of the form(a. 2). Then P is a relation on A which we might call a parent-child relation. as in the previous example. where (a. gray slack). Composition Let R be a relation from a set A into set B. 3). 3. written as RS. Consider the following relation on real numbers. A typical element in R is an ordered pair (x. 6). b) R if and only if a divides evenly into b. y) | y is the square of x} and S = { (x. (6. y) | x <= y}. (black shirt. where P is the parent-child relation given above. c) A x C. c) RS if and only if there exists b B such that (a. Let A be a set of people and let P = {(a. This may not be all possible pairs in A x B but will certainly be a subset of A x B. The composition of R and S. R = { (x. Relation on a Set A relation from a set A into itself is called a relation on A. b) Rand (b. R = {(2. or "related". 3. Example2: Let A = {2. 6}. 5. 5). y). R and S of Example 2 above are relations on A = {2. one such subset may be { (blue shirt. PropertiesOf Relations . Other notations are used depending on the past practice. is the composition of P with itself and it is a relation which we know as grandparentgrandchild relation. (mint green shirt. For example PP. c) S. (3. (2. or R(x) =y where y = x2 . This may not be convenient if R is relatively large. tan slacks) }. 6)}. In some cases R can be described by actually listing the pairs which are in R. (5. tan slacks). 5. and S be a relation from set B into set C. So.6). 6) and define a relation R from A into A by (a. (3. b) | a A ^ b A ^ a is a child of b } . R could be more naturally expressed as R(x) = x2 . For example.himself to combinations which are color coordinated.

there is an element b in B such that <a.Assume R is a relation on set A. a R a. Function Functions Subjects to be Reviewed • • • • • • • • • • • • function domain. The set A in the above definition is called the domain of the function and B its codomain. Thus. then bRa. 3. denote it by f. Reflexive: R is reflexive if for every a A. b> is in the relation. 4. R R b to denote (a. b and c in A. Transitive: R is transitive if for every a. b) R . if aRb. from a set A to a set B is a relation from A to B that satisfies 1. c> are in the relation. . in other words. if aRb and bRc. 1. Let us write a Symmetric: R is symmetric if for every a and b in A. Equivalence: R is an equivalence relation on A if R is reflexive. A x A. for each element a in A. 2. b> and <a. and 2. symmetric and transitive. then b = c . codomain image image of set range sum of functions product of functions one-to-one function (injection) onto function (surjection) one-to-one onto function (bijection) inverse function composite function Definition (function): A function. if <a. f is a function if it covers the domain (maps every element of the domain) and it is single valued. then aRc.

Definition (bijection): A function is called a bijection . ( f*g )(x) = f(x)*g(x) . 4. Then the domain and codomain of this f are N. if and only if whenever f(x) = f(y) . ( f + g )(x) = f(x) + g(x) . under this function is 9. say 3. Every bijection has a function called the inverse function. where f(x)*g(x) is the product of two real numbers f(x) and g(x). f(S) = { f(a) | a S }. Example: Let f(x) = 3x + 1 and g(x) = x2 .. However. x = y . Example: The function f(x) = 2x from the set of natural numbers N to the set of nonnegative even numbers E is an onto function. Then the sum and the product of f and g are defined as follows: For all x. 16. 1. that is. Definition (onto): A function f from a set A to a set B is said to be onto(surjective) . . Example: The function f(x) = 2x from the set of natural numbers N to the set of nonnegative even numbers E is one-to-one and onto. because. there is an element x in A such that f(x) = y . i.e. { 0. nothing in N can be mapped to 3 by this function.The relation given by f between a and b represented by the ordered pair <a. . that is. because for example f(1) = f(-1) = 1 . f is onto if and only if f( A ) = B . f(x) = 2x from the set of natural numbers N to N is not onto. The set of images of the elements of a set S under a function f is called the image of the set S under f. Then ( f + g )(x) = x2 + 3x + 1 . Example: Let f be the function from the set of natural numbers N to N that maps each natural number x to x2 . Note that f(x) = x2 is not one-to-one if it is from the set of integers(negative as well as non-negative) to N . Thus it is a bijection.. if it is onto and one-to-one.} . The image of the domain under f is called the range of f . and ( f*g )(x) = 3x3 + x2 Definition (one-to-one): A function f is said to be one-to-one (injective) . where S is a subset of the domain A of f . and is denoted by f(S) . 9. Definition (sum and product): Let f and g be functions from a set A to the set of real numbers R. and for all x.. and b is called the image of a under f . Example: The function f(x) = x2 from the set of natural numbers N to N is a one-to-one function. the image of. b> is denoted as f(a) = b . if and only if for every element y of B . and its range is the set of squares. for example.

and it is denoted by f -1 . Therefore one can also talk about composition of functions. and let f be a function from B to a set C . Then the function g is called the inverse function of f. Introduction to Language A language is. g(y) = x . the rightmost function in the above figure is a bijection and its inverse is obtained by reversing the direction of each arrow. denoted by fg . A function is a relation.These concepts are illustrated in the figure below. if for every element y of B. in this course. Note that such an x is unique for each y because f is a bijection. It is also a bijection. the points on the left are in the domain and the ones on the right are in the codomain. f(x) > relation. are all languages in that sense. Then f( g(x) ) = ( x + 1 )2 . Definition (composite function): Let g be a function from a set A to a set B . In each figure below. languages of mathematics. Definition (inverse): Let f be a bijection from a set A to a set B. where f(x) = y . For example. Others such as languages of logics. and g(x) = x + 1 . a set of strings of symbols. . Programming langauges we use are a language in that sense. Example: Let f(x) = x2 . natural languages etc. Example: The inverse function of f(x) = 2x from the set of natural numbers N to the set of non-negative even numbers E is f -1(x) = 1/2 x from E to N . is the function from A to C that satisfies fg(x) = f( g(x) ) for all x in A . and arrows show < x. Then the composition of functions f and g .

However. These formal languages and automata capture the essense of various computing devices and computation in a very simple way. solving them can be seen as recognizing languages i. Definitions on Language Subjects to be Learned .). The four classes are regular (or type 3) languages. These formal languages are characterized by grammars which are essentially a set of rewrite rules for generating strings belonging to a language as we see later. It can be rigorously shown that some problems can not be solved by computers in any finite amount of time and that some others are practically unsolvable because of the time it takes to solve them. nondeterministic finite automata (NFA) and nondeterministic finite automata with transitions (NFA. Then we study regular languages. Then after seeing yet another way of representing regular laguages. in general there are more than one NFAs and DFAs that reconize one language. regular and context-free languages. Also for some important classes of problems. Then we investigate various kinds of finite automata: deterministic finite automata (DFA). NFA and NFAare conceptually simpler and easier to use when modeling a system because there are no restrictions on transitions for them unlike for DFA. They are devices that recognize regular languages. Type 3 is a subset of type 2 which is a subset of type 1 and type 0 is the most general including the other three as a subset. In asddition two of the formal languages. Our last topic on regular language is testing of languages for non-regularity. We are going to see an algorithm for converting NFAto NFA which recognizes the same language and another for NFA to DFA conversion. regular grammars. Also there are various kinds of computing devices called automata which process these types of languages Thus formal languages can also be characterized by the computing devices which process them. context-sensitive (or type 1) languages and phrase structure (or type 0) languages. are quite useful for modeling systems used in practice such as co9mputer network communication protocols. On the other hand DFAs are suited for writing a simulator program because there is no nondeterminism such as going to two or more states from a state upon reading one input symbol. Using automata and formal languages we can study limitations of computer and computation. if the number of states of DFA is minimized. together with regular expressions which are a method of representing regular languages. the simplest of the four formal languages. As we are going to learn next. then the resulting DFA is unique up to the state names for a given regular language. In the following chapters we first learn about languages. context-free (or type 2) languages.e.What we are going to study on languages in this course are four classes of languages called (Chomsky) formal languages and their properties. we are going to learn modeling of systems finite automata. checking whether or not a string is in a language. lexical analyzers and parser for compilers for programming languages.

b. it has no symbols. For example if u = aab and v = bbab. That is. uv is the string obtained by appending the sequence of symbols of v to that of u. b} and 0. Let u and v be strings. intersetion and difference of two languages over an alphabet are languages over . For example {0. 1. So a string is a substring of itself. v is called a suffix of y. Thus a language over alphabet is a subset of . 1} is an alphabet with two symbols. So it is not empty. Note that u and v may be an empty string. Thus {a. 1}. The empty string is denoted by (capital lambda). The set { } is a language which has one string. ab. union. this set has an object in it. operations on languages and some of their properties. Then uv denotes the string obtained by concatenating u with v. A string x is called a substring of another string y if there are strings u and v such that y = uxv. For any alphabet . a and aabab are examples of string over alphabet {a. intersection. Some special languages The empty set is a language which has no strings. b} is another alphabet with two symbols and English alphabet is also an alphabet. The number of symbols in a string is called the length of the string.• • • • alphabet string (word) language operations on languages: concatenation of strings. 10 and 001 are examples of string over alphabet {0. an alphabet is a finite set of symbols. 2. Though has no symbols. Operations on languages Since languages are sets. For a string w its length is represented by |w|. 3. Note that vu = bbabaab uv.1}). then uv = aabbbab. A string x is a prefix of another string y if there is a string v such that y = xv. the set of all strings over (including the empty string) is denoted by . Thus | | = 0. all the set operations can be applied to languages. {a.b}) and {0. Thus the union. Basic concepts First. Kleene star Contents Here we are going to learn the concept of language in very abstract and general sense. 111} is a language (over alphabet {0. The empty string (also called null string) is the string with length 0. It can be defined more formally by recursive definition. namely . that is. A language is a set of strings over an alphabet. A string (also called a word) is a finite sequence of symbols of an alphabet. baa} is a language (over alphabert {a. We are going to use first few symbols of English alphabet such as a and b to denote symbols of an alphabet and those toward the end such as u and v for strings.

xw L*. ababb. the extremal clause is not necessary. bb. Since Lk is defined for natural numbers k. Recursive definition of Lk: Basis Clause: L0 = { } Inductive Clause: L(k+1) = Lk L. This * is called Kleene star. For a string u and a natural number k. . bbaba. These powers can be formally defined recursively. is . abb. baaba}. That is L1L2 is the set of strings obtained by concatenating strings of L1 with those of L2. The complement of a language L over an alphabet language. L* is the set of strings obtained by concatenating zero or more strings of L as we are going to see in Theorem 1. Then the concatenation of L1 with L2 is denoted as L1L2 and it is defined as L1L2 = { uv | u L1 and v L2 }. ak and uk can be defined similarly. Lk means the concatenation of k L's.. Powers : For a symbol a and a natural number k. abaaba. aba. Let L1 and L2 be languages. Here a0 = and u0 = . } The * in * is also the same Kleene star defined above. bbbb.L and it is also a Another operation onlanguages is concatenation. baaa. Hence Lk is the set of strings that can be obtained by concatenating k strings of L. Extremal Clause: Nothing is in L* unless it is obtained from the above two clauses. For example if L = { aba. Similarly for a language L. * The following two types of languages are generalizations of them quite often in this course.. uk denotes the concatenation of k u's. abaaba. For example {ab. bb }. aaba} = {abaaa.. b} {aaa. Recursive definition of L+: Basis Clause: L L+ . ak represents the concatenation of k a's. For example Lk can be defined recursively as follows. Recursive definition of L*: Basis Clause: L* and we are going to see Inductive Clause: For any x L* and any w L. ababb. babb. then L* = { .

. w1w2. Proof: Because we can see that L* (L*)*. L* and L* have a number of interesting properties.. } Let us also define natural number k } . Extremal Clause: Nothing is in L+ unless it is obtained from the above two clauses. Theorems 1 and 2 are proven in "General Induction" which you study in the next unit. any nonempty string in L* or L+ can be expresssed as the concatenation of strings of L.w2m2. w2.... Theorem 5: L* = (L*)*.w1m1w21..e. Let us list one of them as a theorem and prove it. for each wi there are strings wi1.wk for some k. bbaba. bb.. then L+ = { aba.Inductive Clause: For any x L+ and any w L... wk in L* such that x = w1w2.e. where wi's are strings of L.. . ) as ={x|x Lk for some Then the following relationships hold on L* and L+.wk ....wm1. Other proofs are omitted. L* can be proven as follows: by Theorem 2.. . abaaba. wk are strings of L*. Then there are nonempty strings w1. Since w1.wimi Hence x = w11 . . bbbb. wimi in L such that wi = wi1wi2. Theorem 1: Ln Theorem 2: Theorem 3: Theorem 4: L+ = L L* = L*L Note: According to Theorems 2 and 3.. For example if L = { aba. by applying Theorem 2 to the language L* L* Conversely ( L* )* Let x be an arbitrary nonempty string of ( L* )*.. . i.. Hence x is in L* .. . ababb. w2. bb }.. (i. L0 L L2 ... wi2... xw L+.. ... Thus L+ is the set of strings obtained by concatenating one or more strings of L.wmmk .

If x is an empty string, then it is obviously in L* . Hence ( L* )* Since L* L* . L* , L* = ( L* )* .

(L*)* and ( L* )*

Problem Solving as Language Recognition
Subjects to be Learned
• • •

problem instance problem as language problem solving as language recognition

Contents
In the previous section the concept of language was introduced and its properties have been briefly studied. You might be wondering why we study language. The main reason for studying language is that solving problems can be viewed as a language recognition problem as explained below, that is, the problem of checking whether or not a string belongs to a language. Thus instead of studying what kind of problems can be solved by what kind of computational devices and how, we can study languages and devices to recognize them which are simpler to deal with uncluttered with variations in actual devices, programming languages etc. Below an example is given to illustrate how solving a problem can be viewed as recognizing a language. Consider the following problem: Is the longest of the distances between two nodes(i.e. the diameter) of a given graph less than a given integer k ? Here the distance is the smallest number of edges (or hops) between the nodes. Some of the instances of this problem are as shown below:

Instance 1 asks whether or not the diameter of the given graph with one edge and two nodes is less than 1. Instance 2 asks whether or not the diameter of the given graph with four edges and four nodes is less than 2. Simiarlyt for Instance 3. These problem instances can be represented by a string as follows: Instance 1: 1,2;(1,2);1 Instance 2: 1,2,3,4;(1,2)(1,3)(1,4)(3,4);2 Instance 3: 1,2,3,4;(1,2)(1,3)(1,4)(2,3)(2,4)(3,4);3 Here the set of nodes, the set of edges and k are separated by ; in that order in the strings. The solutions to these instances are: Instance 1: No Instance 2: No Instance 3: Yes There are infinitely many 'Yes' instances and 'No' instances for this problem. The set of 'Yes' instances is a language and so is the set of 'No' instances as well as the set of all

instances and many others for this problem. We can thus see that solving the problem for a given instance is equivalent to checking whether or not the string representing the given instance belongs to the language of 'Yes' instances of the problem. That is, the problem solving is the same as the language recognition. A problem can be solved if and only if the language of its 'Yes' instances is recognizable or decidable by a Turing machine. It is not solvable if the language is merely accecptable but not recognizable, or even worse if it is not even acceptable.

(a.k.a Structural Induction)
Mathematical statements involving an element of a recursively defined set can be proven by induction. To prove by induction that a statement P(x) is true for all the elements x of a recursively defined set S, proceed as follows: Basis Step: Prove that P(x) is true for all the elements x in the basis of S. Induction: Prove that for any element(s) x of S if P(x) is true, then P(y) is true for any element y obtained from x by the induction step of the recursive definition of S. Note 1 : In the Induction we try to prove that if a parent has the property then all of its children also have that property. In the process we need the relationship between the parent and the children. That relationship is found in the Inductive Clause of the recursive definition of the set in question. Note 2 : As a first step for general induction proof, it is often a good idea to express y in terms of x so that P(x) can be used. Example 1 (Theorem 1 in "Language") : Prove that Ln L* for any natural number n and any language L. Let us first review the definitions. Recursive definition of Lk: Basis Clause: L0 = { } Inductive Clause: L(k+1) = LkL. Since Lk is defined for natural numbers k, the extremal clause is not necessary. Recursive definition of L*:

. Then there exist strings x and y that satisfy x L and w = xy by the definition of Lk+1. number k. L0 L* . x L* . Lk Next let us prove L* . --. L* by theInduction Hypothesis. xy Hence w Thus Lk+1 L* .Induction L* . Note in the proof below that Basis and Inductive Steps mirror the Basis and Inductive Clauses of the definition of Ln . L* since y L. Proof: Let us first prove Suppose that x Lk for some natural L* . x Example 2 (Theorem 2 in "Language") Let us prove L* = Note that ={x|x Lk for some natural number k } .Basis Clause: L* Inductive Clause: For any string x L* and any string w L. Basis Step: Since by the definitions L0 = { Inductive Step: Assume that Lk Hypothesis We are going to show that Lk+1 }. Now let us prove that Ln L* by induction on Ln. By Example 1 above . y Let w be an arbitrary string in Lk+1 . and L* . L* . xw L*. L* . Since Lk L* . Lk . Hence . . Then by the definition of L*. Hence x . Then by the definition of L* . L* for an arbitrary natural numer k. Extremal Clause: Nothing is in L* unless it is obtained from the above two clauses.

Hence xy End of Inductive Step and Proof Hence we have proven Example 3 . by concatenating y) a child of x in is obtained. then its children xy. . It produces the reversal of a given string x (i. So we show that the property for x is inherited by its children xy. . of L* has the property. and any symbol . . x Lk . Basis Clause: REV( )= . xy holds.Note that L* is defined recursively and that below we are trying to prove that the elements of L* have the property that they also belong to . where y is an arbitrary elememt of L.e. Let us prove the inheritance. The function REV(x) on strings x over the alphabet is defined as follows. Hence xy Lk+1 by the definition of Ln . So we first prove that * the element of the basis of L has the propertyy. Hence by the definition of Inductive Step: Assume that for an arbitrary x in L*. then for some natural number k . Prove that for arbitrary strings x and y of . by Example 1 above. Note that each step mirror the recursive definition of . REV(xa) = Inductive Clause: For any string aREV(x). also have the property. Then we show that if any element. x We are going to show that for an arbitrary element y L .e. holds. Note here that x is a parent and by applying an operation (i. x spelled backward). If x . say x. REV(xy) = REV(y) REV(x) holds. Basis Step: L0 since L0 = { }.

REV(xy) = REV(y) REV(x) holds. Definition of Set of Regular Languages : Basis Clause: . Induction: Assume that for an arbitrary string y of . . REV(xya) = REV((xy)a) = a REV(xy). REV(xya) = REV(ya)REV(x). But by induction hypothesis a REV(xy) = a REV(y)REV(x). Basis Step: REV(x ) = REV( x ) = REV( )REV( x ) . xa is also in . REV(xy) = REV(y) REV(x) holds. then Lr Ls . Thus the statement to be proven is for an arbitrary fixed string x. Regular language The set of regular languages over an alphabet is defined recursively as below. * Basis Clause: where is an empty string. Definitions of Regular Language and Regular Expression Subjects to be Learned • • regular language regular expression 1. Inductive Clause: If Lr and Ls are regular languages.Induction Hypothesis Then for an arbitrary symbol a of . Since a REV(y) = REV(ya).Proof First let us note that * can be defined recursively as follows: . and an arbitrary string y of . End of Proof. Any language belonging to this set is a regular language over . Omitted. -. which is what we needed. The proof of the equality in question is going to be proven for an arbitrary fixed x by induction on y. The proof mirrors the recursive definition of . Extremal Clause: Nothing is a regular language unless it is obtained from the above two clauses. Inductive Clause: For arbitrary strings x of and a of ExtremalClause: As usual. LrLs and Lr* are regular languages. { } and {a} for any symbol a are regular languages.

is a regular language because {a.For example. Inductive Clause: If r and s are regular expressions corresponding to languages Lr and Ls . ( rs ) and ( r*) are regular expressions corresponding to languages Lr Ls . Note also that *. (4) We use ( r+) as a regular expression to represent Lr+ . let = {a. (2) The operation * has precedence over concatenation. respectively. For a recursive definition of Lrk click here. Also since {a} is regular. LrLs and Lr* . Regular expression Regular expressions are used to denote regular languages. Thus for example rr = r2 . The language corresponding to rk is Lrk. So for example. which has precedence over union ( + ). where a is an element of . respectively. (3) The concatenation of k r's . then ( r + s ) . Basis Clause: . {a. Conventions on regular expressions (1) When there is no danger of confusion. The set of regular expressions over an alphabet is defined recursively as below. . a. where Lr is the language corresponding to the regular expression r. is written as rk. { } and {a}. Extremal Clause: Nothing is a regular expression unless it is obtained from the above two clauses. b} is regular. 2. b}. b} ( = {a} {b} ) and {ab} ( = {a}{b} ) are regular languages. which is the set of strings consisting of a's and b's. Any element of that set is a regular expression. They can represent regular languages and operations on them succinctly. aa. where r is a regular expression. Thus the regular expression ( a + ( b( c*) ) ) is written as a + bc*. {a}* is a regular language which is the set of strings consisting of a's such as . ( r + s ) is used in stead of ( r + s ). aaaa etc. and a are regular expressions corresponding to languages . aaa. Then since {a} and {b} are regular languages. bold face may not be used for regular expressions.

That is. ba. that is the set of strings of length 2 over the alphabet {a.. it is not easy to see by inspection whether or not two regular expressions are equal. bb}. In general. Definition of Equality of Regular Expressions Regular expressions are equal if and only if they correspond to the same language. . b}. ab. ( ab )+ corresponds to the language {ab. a*b+a* corresponds to the set of strings consisting of zero or more a's followed by one or more b's followed by zero or more a's. a*b* corresponds to the set of strings consisting of zero or more a's followed by zero or more b's.. abab. b}. b}. the set of strings of repeated ab's. because they both represent the language of all strings over the alphabet {a. b}. }. in general. ababab. corresponds to more than one regular expressions.Examples of regular expression and regular languages corresponding to them • • • • ( a + b )2 corresponds to the language {aa. . Note:A regular expression is not unique for a language. Thus for example ( a + b )* = ( a*b* )* . a regular language. b}. For example ( a + b )* and ( a*b* )* correspond to the set of all strings over the alphabet {a. ( a + b )* corresponds to the set of all strings over the alphabet {a. In general ( a + b )k corresponds to the set of strings of length k over the alphabet {a. that is.

r1r2 + r2r1 in the given regular expression is redundant. a. the strings of (r1(r1 + r2)*) start with a string of r1 followed by any number of strings taken arbitrarily from r1 and/or r2. (a) find a string corresponding to r2 but not to r1 and (b) find a string corresponding to both r1 and r2. Ex. a+. b and the strings consiting of only b's (from (a*b)*). r1 = a* + b* r2 = ab* + ba* + b*a + (a*b)* Solution: (a) Any string consisting of only a's or only b's and the empty string are in r1. 3: Let r1 and r2 be arbitrary regular expressions over some alphabet. Thus (r1 + r2 + r1r2 + r2r1)* is reduced to (r1 + r2)*. (a) (r1 + r2 + r1r2 + r2r1)* (b) (r1(r1 + r2)*)+ Solution: One general strategy to approach this type of question is to try to see whether or not they are equal to simple regular expressions that are familiar to us such as a. However. (a) Since (r1 + r2)* represents all strings consisting of strings of r1 and/or r2 . The only strings corresponding to r2 which consist of only a's or b's are a. 1: Find the shortest string that is not in the language represented by the regular expression a*(ab)*b*. Thus anything that comes after the first r1 in (r1(r1 + r2)*)+ is represented by (r1 + r2)*. (a + b)+ etc. Hence (r1(r1 + r2)*) . 2: For the two regular expressions given below. Of the strings wiht length 2 aa. (b) A string corresponding to r1 consists of only a's or only b's or the empty string. bb and ab are in the language. a*. Solution: It can easily be seen that . ba is not in it. b. they do not produce any strings that are not represented by (r1 + r2)*. (a + b)*. Thus the answer is ba. (b) (r1(r1 + r2)*)+ means that all the strings represented by it must consist of one or more strings of (r1(r1 + r2)*). that is.Exercise Questions on Regular Language and Regular Expression Ex. So we need to find strings of r2 which contain at least one a and at least one b. which are strings in the language with length 1 or less. Find a simple (the shortest and with the smallest nesting of * and +) regular expression which is equal to each of the following regular expressions. For example ab and ba are such strings. However. Ex.

Ex. Thus (aab)*(bb)* is a regular expression for L. b*a b*a b* is a regular expression for this language. Then starting with or a. Ex. and since an arbitrasry string of b's can be represented by the regular expression b*. 7: Find a regular expression corresponding to the language of all strings over the . Since any string of b's can be placed in front of the first a. Solution: Let us see what kind of strings are in L. behind the second a and between the two a's. Extremal Clause: Nothing is in L unless it can be obtained from the above two clauses. strings of L are generated one by one by prepending aab or bb to any of the already generated strings. 5: Find a regular expression corresponding to the language L defined recursively as follows: Basis Clause: L and a L. and conversely (r1(r1 + r2)*)+ represents the strings represented by (r1(r1 + r2)*).also represents the strings of (r1(r1 + r2)*)+. strings of L are generated one by one by prepending aab or appending bb to any of the already generated strings. 6: Find a regular expression corresponding to the language of all strings over the alphabet { a. 4: Find a regular expression corresponding to the language L over the alphabet { a . Hence (r1(r1 + r2)*)+ is reduced to (r1(r1 + r2)*). Hence a string of L has zero or more of aab's and bb's in front possibly followed by a at the end. Solution: Let us see what kind of strings are in L. Thus (aab + bb)*(a + ) is a regular expression for L. Solution: A string in this language must have at least two a's. Inductive Clause: If x L . Then starting with . b } that contain exactly two a's. Hence a string of L consists of zero or more aab's in front and zero or more bb's following them. then aabx L and bbx L . b } defined recursively as follows: Basis Clause: L Inductive Clause: If x L . First of all L . First of all and a are in L . Ex. then aabx L and xbb L . Extremal Clause: Nothing is in L unless it can be obtained from the above two clauses. Ex.

b } that contain an odd number of b's Ex. Ex. Solution: Since any string of even length can be expressed as the concatenation of strings of length 2 and since the strings of length 2 are aa. On the other hand if an a precedes the aa. ( b + ab )*a( b + ba )* is obtained as a regular expression corresponding to such strings. then applying the same argument as for aa to a. b }. Hence a string preceding the aa can be represented by ( b + ab )*. 11: Describe as simply as possible in English the language corresponding to the .alphabet { a. Hence the string is in this language. Hence any string that follows aa is represented by ( b + ba )*. Altogether ( b + ab )*( + a + aa )( b + ba )* is a regular expression for the language. Hence if a string does not end with ab then it ends with a or if it ends with b the last b must be preceded by a symbol b. b } that contain no more than one occurence of the string aa. then applying the same argument as for aa to . bb. Solution: Any string in a language over { a . Solution: If there is one substring aa in a string of the language. ab. Since it can have any string in front of the last a or bb. Any numbe of a's can appear any place in the string. then that aa can be followed by any number of b. If there is no aa but at least one a exists in a string of the language. Ex. ( a + b )*( a + bb ) is a regular expression for the language. If there may not be any a in a string of the language. it is the set of strings over the alphabet { a. Solution: A string in the language can start and end with a or b. 9: Find a regular expression corresponding to the language of strings of even lengths over the alphabet of { a. Hence if a string of the language contains aa then it corresponds to the regular expression ( b + ab )*aa( b + ba )* . If an a comes after that aa. b } that do not end with ab. it has at least one b. Ex. a regular expression corresponding to the language is ( aa + ab + ba + bb )*. then it must be followed by b. 10: Describe as simply as possible in English the language corresponding to the regular expression a*b(a*ba*b)*a* . b } must end in a or b. Note that 0 is an even number. and after the first b all the b's in the string appear in pairs. ba. Thus simply put. then that a must be preceded by b because otherwise there are two occurences of aa. ( b + ab )*( b + ba )* is obtained as a regular expression corresponding to such strings. 8: Find a regular expression corresponding to the language of all strings over the alphabet { a.

LrLs and Lr* are regular languages and they are obviously over the alphabet . 12: Describe as simply as possible in English the language corresponding to the regular expression ( b + ab )*( a + ab )*. Proof: Let Lr and Ls be regular languages over an alphabet . Lr Ls . Note 2: The union of infinitely many regular languages is not necessarily regular. Since (( a + b )3)*( a + b ) represents the strings of length 3n + 1. Hence altogether it represents any string consisting of a substring with no aa followed by one b followed by a substring with no bb. a finite language is a set of n . Then by the definition of the set of regular languages . where n is a natural number. concatenation and Kleene star. Note 1: Later we shall see that the complement of a regular language and the intersection of regular laguages are also regular. For example while { akbk } is regular for any natural number k . Solution: (( a + b )3) represents the strings of length 3. the given regular expression represents the strings of length 3n and 3n + 1. that is. concatenation and Kleene star operations. Regularity of finite languages Theorem 1: The set of regular languages over an alphabet is closed under operations union. Ex. and ( a + ab )* represents strings which do not contain any substring bb. Properties of Regular Language Subjects to be Learned • • Closure of the set of regular languages under union. Solution: ( b + ab )* represents strings which do not contain any substring aa and which end in b. where n is a natural number. is not regular as we shall see later. The following theorem shows that any finite language is regular. Hence (( a + b )3)* represents the strings of length a multiple of 3.regular expression (( a + b )3)*( +a+b). { anbn | n is a natural number } which is the union of all the languages { akbk } . We say a language is finite if it consists of a finite number of strings. Thus the set of regular languages is closed under those operations.

Hence by the Inductive Clause of the definition of regular language { a }{ w } is regular. Theorem 2: A finite language is regular. Claim 2: Let w be a string over an alphabet . End of proof of Theorem 2. Hence { aw } is regular. then we have proven the theorem. Then for any symbol a of . } and { a } are Inductive Step: Assume that { w } is a regular language for an arbitrary string w over . { a } is a regular language from the Basis Step. Basis Step: By the Basis Clause of the definition of regular language. a finite language is regular) if { w } is regular for any string w. We then prove that a language consisting of a single string is regular. Inductive Step: Assume that a language L consisting of n strings is a regular language (induction hypothesis). { regular languages for any arbitrary symbol a of . End of proof of Claim 1 Thus if we can show that { w } is a regular language for any string w. Proof: Let us first assume that a language consisting of a single string is regular and prove the theorem by induction.strings for some natural number n. Then { w } is a regular language. Then since { w } is a regular language as proven below. Proof of Claim 2: Proof by induction on strings. Proof of the Claim 1: Proof by induction on the number of strings. Claim 1: A language consisting of n strings is regular for any natural number n (that is. . Basis Step: (corresponding to n = 0) is a regular language by the Basis Clause of the definition of regular language. End of proof for Claim 2 Note that Claim 2 can also be proven by induction on the length of string. L { w } is a regular language by the definition of regular language.

We are going to learn languages which are not regular and ways to test languages for non-regularity. they are much more powerful computing devices than finite automata. and equivalence and conversions between them. In this example you as a vending machine have gone through (transitions between) a number of states responding to the inputs from the customer (coins in this case). Unfortunately not all languages and systems are simple like regular languages or finite automata. We are then going to study how finite automata can be used to simulate operations of systems we see in practice. Later we are going to learn an extension of finite automata called Turing machines. Click "NICKEL" or "DIME" in the figure and see how it operates (see how arrows turn red). A vending machine looked at this way is an example of finite automaton. Their operations can be simulated by a very simple computer program. If the customer puts in a nickel. In the figure. A kind of systems finite automnata can model and a computer program to simulate their operations are discussed later. therefore. It is assumed that the machine terminates its operation when it receives 15 cents or more. circles represent states and arrows state transitions. When a customer comes and puts in the first coin. Let us consider the operation of a soft drink vending machine which charges 15 cents for a can. can not be recognized by finite automata. Pretend that you are the machine. you are no longer in the waiting-for-customer state. In fact Turing machines are as . say a dime. When the customer selects a soft drink.e. So we might say you are in the 10-cents state. you must give the customer a can of soft drink. You have received 10 cents and are waiting for more coins to come. There are languages which are not regular and which. Let us assume that only nickels and dimes are used for simplicity. say 15-cents state.Introduction to Finite Automata In this chapter we are going to study a class of machines called finite automata. We have learned that regular languages are represented by regular expressions and conversely. In the next few chapters first we are going to learn different kinds of finite automata. Though Turing machines are simple modification of finite automata. you are in the waiting-for-customer state. Ds on arrows represent a dime and Ns a nickel. Finite automata are computing devices that accept/recognize regular languages and are used to model operations of many systems we find in practice. that is. tell whether or not a given string belongs to the regular language). After that you stay in that state until another coin is put in to start the process anew or you may terminate the operation and start all over from the initial state. The states and the transitions between them of this vending machine can be represented with the diagram below. then you have now received 15 cents and you wait for the customer to select a soft drink. Then we are going to see that for every regular language a unique finite automaton can be constructed which can recognize the language (i. Initially you are waiting for a customer to come and put some coins. So you are in another state.

however. the states of the machine such as "waiting for a customer to put a coin in". though not proven. "Waiting for a customer to put a coin in" can be considered the initial state of this automaton and the state in which the machine gives out a soda can can be considered the accepting state. We call the elements of Q a state. q0 . Then a deterministic finite automaton is a 5-tuple < Q . the sequence of input symbols given to the finite automaton is "accepted". Its elements can. Also let be a function from Q to Q . be interpreted as a state that the system (automaton) is in. For example. q0 the initial state and A the set of accepting states. . Thus in the example of vending machine. that any computation human beings do (with or without computers) can be performed by Turing machines. let q0 be a state in Q and let A be a subset of Q. Definition of Deterministic Finite Automata Subjects to be Learned • • • Finite automata State transition diagram State transition table Definition of deterministic finite automaton Let Q be a finite set and let be a finite set of symbols. in the Example 1 below. for example. "have received 5 cents" etc. The accepting states are used to distinguish sequences of inputs given to the finite automaton. . Thus in the example of vending machine. A > Notes on the definition 1. the transition function. are the elements of Q. if q is the initial state and a nickel is put in. 4. . a) must be specified.powerful as computers and it is generally believed. a) if it receives the input symbol a while in state q. The set Q in the above definition is simply a set with a finite number of elements. Thus for each state q of Q and for each symbol a of . a) is equal to "have received 5 cents". then (q. 2. The transition function is also called a next state function meaning that the automaton moves into the state (q. 3. Otherwise it is not accepted. If the finite automaton is in an accepting state when the input ceases to come. (q. Note that is a function.

A = { 1 }. The accepting states are indicated by double circles. Examples of finite automaton Example 1: Q = { 0. aaa. the initial state is 0 and is as shown in State (q) Input (a) Next State ( 0 a 1 1 a 2 2 a 2 (q. An arc ( p . are not accepted. It is a little more complex DFA. They are called transition table. If the alphabet of the Example 1 is changed to { a. The vertices (denoted by single circles) of a transition diagram represent the states of the DFA and the arcs labeled with an input symbol correspond to the transitions. ) = q . a) ) A state transition diagram for this DFA is given below. Transition functions can also be represented by tables as seen below. q ) from vertex p to vertex q with label represents the transition (p. DFAs are often represented by digraphs called (state) transition diagram. 2 }. .the string a is accepted by the finite automaton. A deterministic finite automaton is also called simply a "finite automaton". Abbreviations such as FA and DFA are used to denote deterministic finite automaton. 1. But any other strings such as aa. etc. = { a }. then we need a DFA such as shown in the following examle to accept the same string a. the following table. 5. b } in stead of { a }.

A = { 1 }. the initial state is 0 and is as shown in State (q) Input (a) Next State ( 0 a 0 0 b 1 1 a 1 1 b 1 (q. = { a. 1 }. A DFA that accepts all strings consisting of only symbol a over the alphabet { a. b }. a) ) Note that for each state there are two rows in the table for corresponding to the symbols a and b. while in the Example 1 there is only one row for each state. 1. a) ) . the initial state is 0 and is as shown State (q) Input (a) Next State ( 0 a 1 0 b 2 1 a 2 1 b 2 2 a 2 2 b 2 (q. b } is the next example. = { a. in the following table. Example 3: Q = { 0.Example 2: Q = { 0. 2 }. b }. the following table. A state transition diagram for this DFA is given below. A = { 0 }.

20 }. If we make it a DFA. a) ) . 10. A = { 15. N }. Example 4: For the example of vending machine of the previous section. its transition function is as shown in the following table. Q = { 0. 15. the initial state q0 = 0. 5. = { D.A state transition diagram for this DFA is given below. 20 }. State (q) Input (a) Next State ( 0 N 5 0 D 10 5 N 10 5 D 15 10 N 15 10 D 20 15 N 5 15 D 10 20 N 5 20 D 10 (q.

The tape has the left end and extends to the right without an end. It never moves to the left. The tape is divide into squares in each of which a symbol can be written prior to the start of the operation of the automaton. it stops and the automaton terminates its operation. 3. The head is always at the leftmost square at the beginning of the operation. When it sees no symbol. There is a finite control which determines the state of the automaton and also controls the movement of the head. 6. . 4. The tape has a read only head. 2.A finite automaton as a machine A finite automaton can also be thought of as the device shown below consisting of a tape and a control circuit which satisfy the following conditions: 1. The head moves to the right one square every time it reads a symbol. 5.

If b is read while it is in state 0 (initially or after reading some a's). For example strings aaa. when all the a's on the tape are read. it goes into state 1 and the input string is not accepted by the DFA.Operation of finite automata Let us see how an automaton operates when it is given some inputs. Once it gets to state 1. Hence when b appears anywhere in the input. are not accepted by this automaton. When zero or more a's are given as an input to it. Since the state 0 is also the accepting state. Initially it is in state 0. then no matter what symbol is read. it moves to state 1. As an example let us consider the DFA of Example 3 above. aaaaaa etc. . Thus this automaton accepts any string of a's. b etc. are accepted but strings such as aaba. it stays in state 0 while it reads all the a's (without breaks) on the tape. the DFA is in the accepting state. this DFA never leaves state 1.

*

of DFA and its Properties

Subjects to be Learned
• •
*

Language accepted by DFA

Contents
Here we are going to formally describe what is meant by applying a transition repeatedly, that is the concept of * For a state q and string w, *( q , w ) is the state the DFA goes into when it reads the string w starting at the state q. In general a DFA goes through a number of states from the state q responding to the symbols in the string w. Thus for a DFA < Q , , q0 , , A > , the function
* :Q -> Q is defined recursively as follows: *

Definition of

*

:
*

Basis Clause: For any state q of Q ,

(q,

) = q , where
*

denotes the empty string. and any symbol a ,

Inducitve Clause: For any state q of Q, any string y * ( q , ya ) = ( *( q , y ) , a ) .

In the definition, the Basis Clause says that a DFA stays in state q when it reads an empty string at state q and the Inductive Clause says that the state DFA reaches after reading string ya starting at state q is the state it reaches by reading symbol a after reading string y from state q. Example For example suppose that a DFA contains the transitions shown below.

Then
*

*

( q , DNR ) can be calculated as follows:

( q , DNR ) = ( *( q , DN ) , R ) by the Inductive Clause. = ( ( *( q , D ) , N ) , R ) by applying the Inductive Clause to *( q , DN ). = ( ( *( q , D ) , N ) , R ) since D = D . = ( ( ( *( q , ) , D ) , N ) , R ) by applying the Inductive Clause to *( q , D ). = ( ( ( q , D ) , N ) , R ) , since ( q , ) = q . = ( ( q1 , N ) , R ) , since ( q , D ) = q1 as seen from the diagram. = ( q2 , R ) , since ( q1 , N ) = q2 as seen from the diagram. = q3 since ( q2 , R ) = q3 as seen from the diagram. Properties of
*

We can see the following two properties of

*

. for a DFA < Q , , q0 , ,A

Theorem 1: For any state q of Q and any symbol a of >,
*

(q,a)=

(q,a)

Proof : Since a = a , * ( q , a ) = *( q , a ) . By the definition of * , * ( q , a ) = ( *( q , ) , a ) But *( q , ) = q by the definition of Hence ( *( q , ) , a ) = ( q , a ) .

*

.

The next theorem states that the state reached from any state, say q , by reading a string, say w , is the same as the state reached by first reading a prefix of w, call it x, and then by reading the rest of the w, call it y. Theorem 2: For any state q of Q and any strings x and y over q0 , , A > , for a DFA < Q , ,

*

( q , xy ) =

*

(

*

(q,x),y).

Proof : This is going to be proven by induction on string y. That is the statement to be proven is the following: * For an arbitrary fixed string x, ( q , xy ) = *( *( q , x ) , y ) holds for any arbitrary string y. First let us review the recursive definition of *. Recursive definition of Basis Clause:
* *

:

.

* * Inductive Clause: If x and a , then xa . * Extremal Clause: Nothing is in unless it is obtained from the above two clauses.

Now the proof of the theorem. Basis Step: If y = , then *( q , xy ) = *( q , x ) = *( q , x ) . Also *( *( q , x ) , y ) = *( *( q , x ) , ) = *( q , x ) by the definition of * . Hence the theorem holds for y = . Inductive Step: Assume that *( q , xy ) = *( *( q , x ) , y ) holds for an arbitrary string y. This is the induction hypothesis. We are going to prove that *( q , xya ) = *( *( q , x ) , ya ) for any arbitrary symbol a of . ( q , xya ) = ( *( q , xy ) , a ) by the definition of * = ( * ( *( q , x ) , y ) , a ) by the induction hypothesis. = *( *( q , x ) , ya ) by the definition of * . Thus the theorem has been proven.
*

For the following DFA answer the questions given below.

if and only if L = { w | *( q0 . q0 . A language L is accepted by a DFA < Q . q0 . Example 1 : . . That is. w ) A . if and only if ( q0 . . A > . the language accepted by a DFA is the set of strings accepted by the DFA. * . A > . .The following notations are used in the questions: : \delta * : \delta^* : \Lambda Language Accepted by DFA Subjects to be Learned • Language accepted by DFA A string w is accepted by a DFA < Q . That is a string is accepted by a DFA if and only if the DFA starting at the initial state ends in an accepting state after reading the string. w ) A } .

1 any number of times by reading substring ab any number of times to come back to state 1.2 . which is not an accepting state. This DFA has a cycle: 1 .2 . first from the initial state go to state 1 by reading one a.1 and it can go through this cycle any number of times by reading substring ab repeatedly. by reading an empty string .e.This DFA accepts { } because it can go from the initial state to the accepting state (also the initial state) without reading any symbol of the alphabet i. Then from state 1 go to state 2 and then to state 3 by reading aa. and it stays there. This is represented by (ab)*. . Example 2 : This DFA does not accept any string because it has no accepting state. Then from state 1 go through the cycle 1 . Thus a string that is accepted by this DFA can be represented by a(ab)*aa . It accepts nothing else because any non-empty symbol would take it to state 1. To find the language it accepts. Thus the language it accepts is the empty set Example 3 : DFA with one cycle .

2 . Example 5 : DFA with two interleaved cycles This DFA has two cycles: 1 .3 . Thus a string that is accepted by this DFA can be represented by ( ab + bb )*.0.1 .2 .1. To find the language accepted by this DFA.1 . first from state 0 go to state 1 by .Example 4 : DFA with two independent cycles This DFA has two independent cycles: 0 .0 .0 .0 .1 and 1 .0 and it can move through these cycles any number of times in any order to reach the accepting state from the initial state such as 0 .2 .2 .0 and 0 .2 .

There is a systematic way of finding the language accepted by a DFA and we are going to learn it later. At this point a substring a( baa + bba )* will have been read. To find the language accepted at state 1. Thus altogether a( baa + bba )*bb will have been read when state 3 is reached from state 0.1 and 1 .0 . Then from state 1 go through the two cycles 1 . At this point (b*a) will have been read.2 . The language accepted at state 0 is b* . Thus the language that is accepted by this DFA is the union of the language accepted at state 0 and the one accepted at state 1. respectively.1 any number of times by reading substring ba repeatedly. So we are not going to go any further on this problem here. At state 1 go through the cycle 1 .3 .reading a ( any other state which is common to these cycles such as state 2 can also be used instead of state 1 ). Then go to state 1 by reading one a. Then go from state 1 to state 2 and then to state 3 by reading bb.2 .1 any number of times in any order by reading substrings baa and bba. first at state 0 read any number of b's.2 . Thus the language accepted at state 1 is b*a(ba)* . Definition of Nondeterministic Finite Automata . Example 6 : This DFA has two accepting states: 0 and 1.

Thus for each state q of Q and for each symbol a of (q. in which case the NFA aborts its operation. Note that any DFA is also a NFA. 1 }. Unlike DFAs an NFA moves into one of the states given by (q. We call the elements of Q a state. a) if it receives the input symbol a while in state q. But it can be the empty set. A > Notes on the definition 1. 4. the transition function. the sequence of input symbols given to the finite automaton is "accepted". let q0 be a state in Q and let A be a subset of Q.Subjects to be Learned • • • Nondeterministic finite automata State transition diagram State transition table Definition of nondeterministic finite automaton Let Q be a finite set and let be a finite set of symbols. 5. Its elements can be interpreted as a state that the system (automaton) is in. the following table. . the initial state is 0 and is as shown in State (q) Input (a) Next State ( 0 a {1} 1 a (q. a) must be specified. = { a }. Then a nondeterministic finite automaton is a 5-tuple < Q . 2. Note that is a function. 3. As in the case of DFA the accepting states are used to distinguish sequences of inputs given to the finite automaton. As in the case of DFA the set Q in the above definition is simply a set with a finite number of elements. ceases to come. Also let be a function from Q to 2Q . a) to select is determined nondeterministically. q0 the initial state and A the set of accepting states. q0 . . a) ) . If the finite automaton is in an accepting state when the input ends i. Which one of the states in (q. The transition function is also called a next state function .e. A = { 1 }. Examples of NFA Example 1: Q = { 0. Otherwise it is not accepted.

2 }. while in the Example 1 there is only one row for each state. . in the following table. b } in stead of { a }. the initial state is 0 and is as shown State (q) Input (a) Next State ( 0 a {1. b }.A state transition diagram for this finite automaton is given below.2} 0 1 1 2 2 b a b a b {2} (q. A state transition diagram for this finite automaton is given below. this is still an NFA that accepts { Example 2: Q = { 0. If the alphabet a}. = { a. A = { 2 }. a) ) Note that for each state there are two rows in the table for corresponding to the symbols a and b. is changed to { a. 1.

Since the state 2 is the accepting state. We say that this automaton accepts the string a. if it moves to state 2 and no more inputs are given. When it reads the symbol a. Thus the string ab is also accepted by this NFA. it moves to either state 1 or state 2. if the next input is b and if no more inputs are given. Let us now define the function strings and languages by NFA. it does not accept any of them. * and then formalize the concepts of acceptance of . Initially it is in state 0. As an example let us consider the automaton of Example 2 above.Operation of NFA Let us see how an automaton operates when some inputs are applied to it. If on the other hand it moves to state 1 after reading a. then it stays in the accepting state. If any other strings are given to this NFA. then it goes to state 2 and remains there.

a) ) For example consider the NFA with the following transition table: . Thus for an NFA < Q . Example State (q) Input (a) Next State ( 0 a {0. In general an NFA nondeterministically goes through a number of states from the state q as it reads the symbols in the string w.3} 0 b {2} 1 1 2 2 3 3 a b a b a b {1} {3} {3} (q. . string. where * denotes the empty . . any string y * and any symbol a ( q . w ) is the set of states that the NFA can reach when it reads the string w starting at the state q. ) = { q }. the function * :Q -> 2Q is defined recursively as follows: * Definition of *: Basis Clause: For any state q of Q. the Basis Clause says that an NFA stays in state q when it reads an empty string at state q and the Inductive Clause says that the set of states NFA can reach after reading string ya starting at state q is the set of states it can reach by reading symbol a after reading string y starting at state q. q0 .1. A > . Inducitve Clause: For any state q of Q. * (q. *( q . ya ) = In the definition.Language Accepted by NFA Subjects to be Learned • • • for NFA Language accepted by NFA Properties of * * Definition of * For a state q and string w.

ab ) = ( 0 . a ) is the union of ( p.b)={2} {3} {1}={1. ab ) can be calculated as follows: ( p. a ) for all p ( 0 . a ) = ( 0 . *( 0 . Suppose that the state 3 is an accepting state of this NFA. b ) for all p * ( 0 . 3 } . ) again by the Inductive * Clause of the definition of . ( 0 . Hence *( 0 . Then * * ( 0 . Hence *( 0 .The transition diagram for this NFA is as given below. By the Basis Clause of the definition of *. a ) = { 0 . . b ) (1. 3}. 1 . a ) by the Inductive Clause of * Now *( 0 . ) = { 0 } . ab ) is the union of the definition of * .b) (3.2.

* . .a) for an NFA < Q . ab. . that is. The language accepted by an NFA < Q. and for NFA has properties similar to that for DFA.* We say that a string x is accepted by an NFA < Q. . A > if and only if * ( q0 . A > is the set of strings that are accepted by the NFA. q0 . . . for an NFA < Q . * ( q . . . q0. if and only if it can reach an accepting state by reading x starting at the initial state.A Theorem 1: For any state q of Q and any symbol a of >. q0. aaa. . xy ) = These theorems can be proven in a manner similar to those for Theorems 1 and 2 for DFA. Theorem 2: For any state q of Q and any strings x and y over q0 . . a. abbbb etc. Some of the strings accepted by the NFA given above are the language it accepts is a*( ab + a + ba )(bb)* . A > . x ) A is not empty. * (q.a)= (q.

q0 . 2. These operations on FAs can be described conveniently if -Transitions are used. let q0 be a state in Q and let A be a subset of Q. Note that any NFA is also a NFA.) and see some examples. Thus the tape head does not move when is read. We call the elements of Q a state. Basically an NFA with -Transitions is an NFA but can respond to an empty string and move to the next state. . . the transition function. A transition on reading means that the NFA. . for any NFA. As we are going to see later. Here we are going to formally define NFA with -Transitions (abbreviated as NFA. Definition of nondeterministic finite automaton with Let Q be a finite set and let -Transitions be a function from Q be a finite set of symbols.there is a NFA (hence DFA) which accepts the same language and vice versa.A> Notes on the definition 1.Definition of Nondeterministic Finite Automata with Transitions Subjects to be Learned • • • - Nondeterministic finite automata with State transition diagram State transition table -Transitions Contents One of the objectives of this chapter is to show that there is a one-to-one correspondence between regular languages and finite automata. q0 the initial state and A the set of accepting states. Then a nondeterministic finite automaton with -Transitions is a 5-tuple < Q .makes the transition without reading any symbol in the input.. Also let { } to 2Q . concatenation and Kleene star operations. We are going to do that by showing that a finite automaton can be constructed from a given regular expression by combining simpler FAs using union.

4 and 5 without reading any symbol on the tape. for example. it can move to any of the states other than 0. = { a. For once you are in state 1. 3. For though you go to states 1. 5 }. a) ) Here the transitions to are omitted from the table. the initial state is 0 and is as shown in the State (q) Input (a) Next State ( 0 a {1} 0 {4} 1 {2} 2 { 3. 2. 4 and 5 by reading a. for example. then you come to state 4. A = . If you read string ab. b }. 1. A state transition diagram for this finite automaton is given below. When a symbol a is read at the initial state 0. following table. 3. 2. there are no transitions on reading b except from state 3. 3. Thus 4 is the only state you can go to from the initial state . you can go to state 2. 4.Example of NFA- Q = { 0. 4 } 3 {5} 3 b {4} 4 a {5} (q.

Language Accepted by NFASubjects to be Learned
• • • •

-closure for NFALanguage accepted by NFAProperties of *
*

Contents
To formally define * for NFA- , we start with the concept of -closure for a state which is the set of states reachable from the state without reading any symbol. Using that concept we define * and then strings and languqges accepted by NFA- . Definition of -closure

Let < Q , , q0 , , A > be an NFA- . Let us denote the -closure of a set S of states of Q by ( S ). Then ( S ) is defined recursively as follows: Basis Clause: S (S)

Inductive Clause: For any state q of Q, if q ( S ) , then ( q , ) (S). Extremal Clause: Nothing is in ( S ) unless it is obtained by the above two clauses.

For the NFAFirst { 2 }

of the above figure, ( { 2 } ) , that is, 2 (2, )

( { 2 } ) is obtained as follows: ( { 2 } ) . Then since 2 ( { 2 } ) , by the

Inductive Clause, Since (2,

({2}). ({2}).

) = { 3 , 4 }, we now have { 2 , 3 , 4 }

Since 3 and 4 have been added to

({2}),

(3,

) = { 5 } and

(4,

)=

must

be included in ( { 2 } ) . Thus now { 2 , 3 , 4 , 5 } ({2}). Though 5 has become a memeber of the closure, since ( 5 , ) is empty, no new members are added to ( { 2 } ) . Since ( q , ) has been examined for all the states currently in ( { 2 } ) and no more elements are added to it, this process of generating the closure terminates and ( { 2 } ) = { 2 , 3 , 4 , 5 } is obtained. As we can see from the example, ( S ) is the set of states that can be reached from the states of S by traversing any number of arcs. That is, it is the set of states that can be reached from the states of S without reading any symbols in . Now with this -closure, we can define
*

recursively as follows:

As in the cases of DFA and NFA, * gives the result of applying the transition function repeatedly as dictated by the given string. Definition of
* *

is going to be defined recursively. Let < Q , , q0 , , A > be an NFA-

.

Basis Clause: For any state q of Q,
*

(q,

)=

({q}).
*

Inductive Clause: For any state q, a string y in

and a symbol a in

,

*

( q , ya ) =

(

).

What the Inductive Clause means is that *( q , ya ) is obtained by first finding the states that can be reached from q by reading y ( *( q , y ) ), then from each of those states p by reading a (i.e. by finding ( p , a ) ), and then by reading 's ( i.e. by taking the closure of the ( p , a )'s ) . Example : For the NFAbelow: of the following figure,
*

( 0 , ab ) can be obtained as

First let us compute *( 0 , a ) . For that we need ( { 0 } ). Since it is the set of states reached by traversing the arcs from state 0, ( { 0 } ) = {0,3,4}. Next from each of the states in ( { 0 } ) we read symbol a and move to another state (i.e. apply ). They are ( 0 , a ) = { 1 } , ( 3 , a ) = ( 4 , a ) = { 5 }. Hence We then traverse the } ) = { 1 , 2 , 3 } and = { 1 , 5 } for q = 0 . arcs from { 1 , 5 } to get to the states in *( 0 , a ) . Since * ({5})={5}, (0,a)={1,2,3,5}.

({1

q2. NFAs and NFAto NFA 1 Conversion of NFA- Let M1 = < Q1 . b ) = { 4 } . Thus Since ( { 4 } ) = { 3 . NFAs and NFA. . 4 } .0 . and ( 2 . NFA M2 = < Q2. a . ( 3 . *( 0 . The language accepted by an NFA. . ab ) = { 3 . b ) and ( 5 . . A1 > be an NFA. A2 > that satisfies the following conditions recognizes L: . For example the NFAof the figure given above accepts the language { . 4 } . .s Subjects to be Learned • • • Conversion of NFA.to (equivalent) NFA Conversion of NFA to (equivalent) DFA Equivalence of DFAs. A > is the set of strings accepted by the NFA. ab } . A > if and only if *( q0 . ab ) read b from each of the states in *( 0 . b ) . Then the 2 . Now ( 1 .< Q . x ) contains at least one accepting state. a ) and then take the arcs from there.Then to find *( 0 . . A string x is accepted by an NFA. . q1. q0 . q0 .that recognizes a language L.0 .. b ) are empty sets. Equivalence of DFAs.< Q . .

Q2 = Q1.0 . Then collect all the states that can be reached from each state of ( {q} ) by traversing one arc labeled with the symbol a. .0 through arcs in M1 . 2.0 are the accepting states of M2 . then all the accepting states of M1 plus state q1. that is if an accepting state can be reached from the initial state q1. a ) . q1. . 1. the initial state is 0 and the accepting states are 1 and 0.M1 = < Q1 . .0. The set of accepting states A2 is the same as A1 if no accepting states can be reached from the initial state q1.a)= ( ) A2 = A1 { q1. Thus to obtain an NFA M2 = < Q2. 1 } .0 . 3 ). A1 > does.0 } ) A1 = A1 otherwise . a ) = 1 * (q. a ): First ( { 0 } ) = { 0 . Otherwise. a ) as follows: Find ( {q} ). 2 first copy the states of Q1 into Q2. since 1 is in ( { 0 } ) . The set of states Q2 of NFA is { 0. Example 1: Let us convert the following NFAto NFA. as the given NFA.0 = q1. that is all the states that can be reached from q by traversing arcs.0 } if ( { q1. Then for each state q of Q2 and each symbol a of find 2 ( q . Then from the transition function of the NFA- .0 through arcs in M1 . The transition function 2 is obtained as follows: 2( 0 . q2. 2 ( q. A2 > which accepts the same language 1 . q2. The closure of the set of those states is 2( q .

(q.b)= 1 ( 1. a ) = 1 ( 1 . a ) = { 1 . )(= ( {1. 2 ( 0 .3} {1.2} {1.b)= Similarly 2 can be obtained for other states and symbols.2} The NFA thus obtained is shown below.3} {3} {1.2}. ({1. .1} {0.2} )) ({q}) {0. a ) = .2} {1. b ) . since ( { 0 } ) = { 0 .1} {1} {1} {2} {2} {1. They are given in the table ( { q } ) and 2 below together with State q Input 0 0 1 1 2 2 3 3 a b a b a b a b . 1 For .2} {1. 2 (0. and Hence 2( 0 .3} {1. b ) = .2} {1.( 0 . 2 }. 1 } and 1 (0.2})={1.

) and State q Input 0 0 1 1 2 2 3 3 4 a b a b a b a b a ({q}) {0.3} {3} {3} {1.3} {2.4} {1. The transition function 2 is obtained as for Example 1.3} {1.2. )(= ( {1.2.3} )) {1.2} {1. 2 is given in the table below together with ( { q } ) .2} {4} {4} {4} {1.1} {1} {1} {2.1} {0. The set of states Q2 of NFA is { 0.4} {1.3} .2. 1 ( p . 3.Example 2: Let us convert the following NFA- to NFA.2} 2 (q.4} {1. 2.4} {1. since 1 is in ( { 0 } ) . 1. the initial state is 0 and the accepting states are 1 and 0. 4 ).

4 b {1. . Proof of Equivalence of NFA- and NFA We are going to prove that the NFA obtained from NFAby the conversion algorithm accepts the same language as the NFA.4} The NFA thus obtained is shown below.

a)= 2 * (q. q1. Thus we first prove that it is true for any arbitrary symbol.w)= 2 * ( q . then * 1 ( q . --. and the conversion of NFAto NFA click here.a)= 2 * 2 (q. = = 2 * 2 (q .(2) .a). a ) . A2 > (q. . = 1 * Then we show that ( q . (q.0 . (q. wa ) = 2 * ( q .a). and NFA- When it is proven. A1 > and 2 NFA obtained by the conversion is denoted by M2 = < Q2. the induction hypothesis and the construction of NFA from NFA. 1 * (q.w). Proof: This is going to be proven by induction on w. First we are going to prove that To review the definition of * 1 * . Basis Step: We need to show that for any symbol a in 1 * (q. . )={q}. a ) = 1*( q . Hence (q.a)= 2 ( q . Secondly 2 * (q. 1 * (q. then assuming it holds for any arbitrary string w we prove it for any of the children of w. q2. for NFA. wa ) --. w ) for any non-empty string w. w ) for an arbitrary string w (Induction Hypothesis). We are going to prove it by showing that both of them are equal to Firstly 2( q . Inductive Step: We need to show that if 1 * (q. it implies that NFAM1 and NFA M2 accept the same non-empty strings. Recall that the set of strings is defined recursively (click here for a quick review). wa ) = --..NFA- that recognizes a language L is denoted by M1 = < Q1 .(1) .0 . 1 . that is wa for any symbol a in the alphabet. a ) --. a)= by the definition of Since * 2 * for NFA (click here for a review) . Claim 1: For any non-empty string w and for any state q.a). .(1) because of the way NFA is constructed from NFA(click here for a review) . .a)= 2 2 * (q .a). .(2) Hence from (1) and (2). wa ) holds for any arbitrary symbol a in First we show that using the definition of 2 * 2 * ( q .w)= 2 * (q. The case when w is an empty string is going to be proven separately.w)= 2 * ( q .

Let us next prove (2). . . by the definition of 1 * . ( q . 1 * Thus from (1) and (2) ( q . wa ) . Hence we have proven (2). w ) by the induction hypothesis. The right hand side of this equality is equal to ( the first and have been swapped to get this) . This can be shown to be equal to . = 1*( q . as proven below in Claim 3. = Since 2 (q. a ) by the way NFA is constructed from NFA= .w)= 1 * ( q . for NFA- (p. wa ) = Since 2 * (q.basically using the definition of Then from (1) and (2) we can see that 1 * 1 * . wa ) = 2 * ( q . that is (1) has been proven. .a)= Substituting this into the left hand side of (2) produces = . Hence On the other hand = Hence 1 * = ( q . because = . wa ) . that is By the definition of 1 * = 1 * 1 * ( q . Hence 2 * ( q . wa ) . wa ) = 2 * ( q . wa ) = . By the definition of 2 * 2 * ( q . wa ) .a)= 1 * ( q . Let us first prove (1). To see an explanation for this click here. wa ) .

. Hence is accepted by NFA. We are going to prove this in two parts: (S T) (S) (S) (T) ( T ) and (S T). and the corresponding NFA accept the same language. ) (X). Hence is accepted by NFA. By the way NFA is constructed from . . Let X be the set of states of an NFA. if it is accepted by an NFA. Part 1 : (S T) (S) (T) This is going to be proven by induction on (S T). then q20 this means that Thus NFA( { q10 } ) A1 A2 . let us prove the following claim. Claim 2: (S T)= (S) (T). then ( q . Inductive Clause: If q ( X ) . Then the -closure of X is defined recursively as Basis Clause: X (X).End of Induction With this Claim 1 we can see that any non-empty string w is accepted by NFA if and only if it is accepted by the corresponding NFA. Extremal Clause: Nothng is in ( X ) unless it is obtained by the Basis and Inductive Clauses. Hence by the way A2 is constructed. As a preparation for the proof of commutativity of union and -closure operations. Let us review the definition of the -closure of the set of states of an NFA. A2 . then its childen also have it. Since (S T ) is defined recursively. in the Basis Step of our proof we prove the property for the elements of the basis of (S T ) and in the Inductive Step we prove that if an arbitrary element of (S T ) has that property. q20 Conversely if NFAis accepted by NFA. What Part 1 states is that all the elements of (S T ) have the property of being in the set (S) (T). then ( { q10 } ) A1 . As for the empty string . For that let us restate the statement so that the induction becomes clearer.

) . (T). Hence ) Similarly if q Hence if q is an arbitrary element of (S) (T). then (q. Basis Step: We need to show that S Since S (S T ) . (S) (T).S (S T). ( T ) . (q. then (q. ) (S) (T).q ( S ) or q ( S ) . and ( S T) (S T). S and T are subsets of (S T) Inductive Step: We need to prove that if q is an arbitrary element of the property of being in (S) (S ( T ) . (T) Thus all the elements of (S T ) have the property of being in (S T) which is to say that (S) (T). if q is in . Since q If q ( S ) . ) (S) (T). ( T ) . T ) with (T). T). then (q. Inductive Step: We need to prove that for an arbitrary element q in ( S ) . then ( q . ) (S) (S) (T). Hence (T).Proof of Part 1: Basis Step: We need to prove that ( S Since S ( S ) and T T) (S) (S) (S (S) (S) (T). T T). (S T ) with the property of being in (S) ( T ) . That would imply that Proof of (S) (S By induction on (S T): (S). ) Let q be an arbitrary element of T ) with the property of being in ( S ) by the definition of ( T ) . End of Proof for Part 1 Part 2 : Proof of Part 2: We are going to prove (S) (S (S) T ) and (S) (T) (T) (T) (S (S (S T).

( Sn+1 ) by the induction hypothesis. ) (S T). Similarly Hence (S) (T) (T) (S T ) holds. ( Si ) = ( Si ) = ( S1 ) and ( S1 ) . Si is a set as well as Sn+1. End of Proof for Claim 3 Sn+1 ) by Claim 2 above. since = ( Equivalence of NFA and DFA We are going to prove that the DFA obtained from NFA by the conversion algorithm . Si ) by the definition of union.Inducion Hypothesis ( Si ) = ( = = (( Si ) ( ( Si ) ) Si ) ( Sn+1 ) by the definition of union. Proof : Proof by induction on n. ( Si ) holds for n. Thus (S) (S T ) has been proven. (S T ) holds. --. - T ) is a -closure. End of Proof of Part 2 End of Proof of Claim 2 Claim 3: ( Si ) = ( Si ) . then Hence Inductive Step: Assume that ( ( Si ) = Si ) = ( Si ) holds for n = 1. If n = 1. Basis Step: n = 1. by the definition of (q.(S Since q is in (S T ) and since (S T ) . ) (S closure T). then (q.

w ) . Proof: This is going to be proven by induction on w. --- For the string w and an arbitrry symbol a in 1 * . q2.0 . = * 1 ( q1. 1 * ( q1. Basis Step: For w = 2 * .0 . w ) for an arbitrary string w. w ) . w ) holds. w ) for any string w. ) = q2. w ) . A2 > First we are going to prove by induction on strings that 1*( q1.0 .0 . w ) = 2 * ( q2. Inductive Step: Assume that Induction Hypothesis 1 * ( q1. 2 1 .0 .0 .0 . .0 . When it is proven. NFA that recognizes a language L is denoted by M1 = < Q1 . a ) 2( * 2 ( q2. w ) = 2*( q2.0 .0 . . wa ) = = = = 2 ( 1 2 * * ( q1. ) by the definition of 1 * . q1.0 .0 } by the construction of DFA M2 .0 . ( q2. wa ) * 1 Thus for any string w ( q1. Theorem: For any string w.0 .Part 1 Subjects to be Learned • • • • Union of FAs Concatenation of FAs Kleene Star of FAs Acceptance of regular languages by FAs Contents . w ) = 2 * ( q2.0 . Kleene's Theorem --.0 .0 . w ) = 2 * ( q2. obtained by the conversion is denoted by M2 = < Q2.0 by the definition of 2* . a ) ( q2. ( q1.accepts the same language as the NFA. = { q1. it obviously implies that NFA M1 and DFA M2 accept the same strings. A1 > and DFA .

. A2 > . We assume that Q1 Q2 = without loss of generality since states can be renamed if necessary. Ak > . concatenation and Kleene star operations. Since any regular language is obtained from { } and { a } for any symbol a in by using union. . q2. Basis Step: As shown below the languages . It states that any regular language is accepted by an FA and conversely that any language accepted by an FA is regular. Mc . 2 .0 . respectively. Qu = Q1 Q2 . u Mu = < Qu .0 is a state which is neither in Q1 nor in Q2 . A1 > and M2 = < Q2 .0 . Theorem 1 (Part 1 of Kleene's theorem): Any regular language is accepted by a finite automaton. L1L2 and L1* are accepted by the FAs Mu = < Qu . qu. .0 . 1 . Then L1 = < Qc . . Ac > and Mk = < Q2 . c . Au > . L1L2 and L1* are accepted by FAs. then L1 L2 . respectively. which are given below.0 . qc.{ } and { a } for any symbol a in are accepted by an FA.Kleene's theorem. qk.0 } . k . . that together with the Basis Step would prove the theorem. .0 . Au > : { qu.0 . qu. Suppose that L1 and L2 are accepted by FAs M1 = < Q1 . u L2 . Proof: This is going to be proven by (general) induction following the recursive definition of regular language. where qu. . Inductive Step: We are going to show that for any languages L1 and L2 if they are accepted by FAs. q1.

qk. q2. where qk.u = 1 2 { (qu.0 } ) | q A1 } Ak = { qk. u ) = { q1. . q2. . These NFA- s are illustrated below. { q1. .0.0 } ) } { (q. a ) = for all a in .0.0 } = 1 { (qk. Au = A1 A2 Mc = < Qc .0 } .0 } ) } . .0 } . Ak > : { qk. qc.0 is a state which is not in Q1 . { q2.0 = q1.0. k . . c .0. { q1.0 . .0 . Note that (qu.0 } ) | q A1 } Ac = A2 Mk = < Qk . Ac > : Qc = Q1 Q2 qc. that is u (qu.0 c = 1 2 { (q.0 .0 . { qk. Qk = Q1 k .

in fact accept L1 L2 . respectively. that these NFA. .s .It can be proven. Mc and Mk . though we omit proofs. L1L2 and L1*. Mc and Mk: Example 1: An NFAthat accepts the language represented by the regular expression (aa + b)* can be constructed as follows using the operations given above. Mu. End of Proof Examples of Mu .

Example 2: An NFAthat accepts the language represented by the regular expression ((a + b)a*)* can be constructed as follows using the operations given above. .

Part 2 .Kleene's Theorem -.

k+1. Then the following lemmas hold. then from k+1 to k+1 any number of times. L(p. 2. k) : The set of strings going first from p to k+1. k+1. k+1. k) : The set of strings representing paths from p to q passing through states labeled wiht k or lower numbers. q. Given a finite automaton. k)*L(k+1. Before proceeding to a proof outline for the converse. k) L(p. See the figure below for the illustration. then from k+1 to q. q. Next denote by L(p.Subjects to be Learned • Languages accepted by FAs are regular Contents The converse of the part 1 of Kleene Theorem also holds true. k+1) = L(p. k+1. q. k) . where n is the number of states of the finite automaton. Lemma 1: L(p. Note that paths may go through arcs and vertices any number of times. k)L(k+1. It states that any language accepted by a finite automaton is regular. . let us study a method to compute the set of strings accepted by a finite automaton. k)*L(k+1. q. What this lemma says is that the set of strings representing paths from p to q passing through states labeled with k+1 or lower numbers consists of the following two sets: 1. q. k) the set of strings representing paths from state p to state q that go through only states numbered no higher than k. L(p. k)L(k+1. first relabel its states with the integers 1 through n. q. all without passing through states labeled higher than k.

Lemma 3: L(p. 0) is regular. Proof: L(p. Hence if p and q are different. Theorem 2 (Part 2 of Kleene's Theorem): Any language accepted by a finite automaton is regular. q. q. 0) is the set of strings representing paths from p to q without passing any states in between. k) is regular for any states p and q and any natural number k. L(p. then it consists of single symbols representing arcs from p to q. n) over all accepting states q. we have the following converse of the part 1 of Kleene Theorem. Since the language accepted by a finite automaton is the union of L(q0. then is in it as well as the strings representing any loops at p (they are all single symbols). where n is the number of states of the finite automaton. q. 0) is regular.Lemma 2: L(p. If p = q. >From Lemmas 1 and 2 by induction the following lemma holds. Since the number of symbols is finite and since any finite language is regular. Example : Let us find the language accepted by the following finite automaton using the . q. q.

2): r(1. 3.0) = b.1.0) = and r(2. r(1. 1. 1.0) = ba+ . 1.0)r(1.1. 1. 3. 2.1.0)r(1. 2) = r(3. r(1.0) = a* .0) = and r(3.0) + r(2. 1)r(2.lemmas. 1) r(1. 2)*r(3.0)r(1. since r(3. 1) = r(3.1.0)*r(1. 2) . 2) = r(1. 1)*r(2. 1) + r(3. 1) = r(1. 1)*r(2. 3. 1) = r(1.0)*r(1. 1) = r(2. 1. q. Then the language accepted by this NFA is r(1. 1) + r(1.0) + r(1. 2. 3). )*a r(3.0)*r(1.0) = a+ . 2.2.1. 1. 2.0)r(1. 1. 2) = a* + a+(b a+)*b a* . 1) = r(3.0)*r(1. 3. q. 2) = r(1.2. 1) r(1. r(2. 2)r(3. 2) + r(1. 1)*r(2.1.2.1. 3. k). 1. r(1. 1.0) = a + . 2.0)*r(1. r(1. 3. 3.0) = ba+ + . since r(1.1. 1) r(3. Hence r(1. 1. 2): r(3. since r(2. 1. 1) = r(2. 1)r(2.0) = a . 2. 2.1. k) the regular expression for the set of strings L(p. 2): r(1.2.1.0) = ba* . .0) + r(3.0) + r(1. 2) = a+(b a+ + = a+(b a+ )*a .2.1.2. 3.0) + r(2. 3.0) = b . 3.1. 3.1.2. since r(1. 3. 2. 1) = r(2. 3.1.1. By Lemma 1. 1) + r(1.2. 2. Let us denote by r(p.1. 3) = r(1.2. 1) = a Hence r(1. 3.1. 1)r(2.0)r(1. r(2.

Hence r(3. q0 .1. 3. n)'s must be added together to get the regular expression for the language accepted by the automaton. 1) = r(3.L. . 2) = = + ( ba+)+a + ba+( ba+ + )*a r(3.1. 1) r(3. 2. q. n) must be found for each accepting state q. i. . 1.e. Q . can be obtained by swapping its accepting states with its non-accepting states. q.0)r(1.0) + r(3. that is Mc = < Q . 2) = r(3.A > is a DFA that accepts * .0)<SUP*< SUP>r(1. 2. 1. 2): r(3. and all the r(p. 3) = a* + a+(b a+)*ba* + ( a+( ba+ )*a )( This can be further simplified to (a + ab + abb)*. A > be a DFA that accepts a language L. 1)*r(2.L . If there are more accepting states. 1.0) = ba* Hence r(3. . 1)r(2. The detail is left as an exercise though it would be quite challenging. where p is the initial state and n is the number of states in the given finite automaton. then r(p. 1. then to (a + ab)*. .1. In this example there is only one accepting state. Comlement and Intersection of Regular Language Subjects to be Learned • • • Complement of Regular Language Complement of DFA Intersection of Regular Languages Contents Complement Let M = < Q . )*ba* + ( ba+)+a )*( ba+ )*ba*. .1. 1) + r(3. 1. Then a DFA that accepts the complement of L. q0 . * . 2) = ba* + ba+( ba+ + = ( ba+ )*ba* . 1. Hence r(1. 1.

the complement of a regular language is also regular.For example the following DFA accepts the language a+ over = { a . . Remark 2: Since a language is regular if and only if it is accepted by some NFA. A DFA that accepts its complement is obtained from the above DFA by changing all single circles to double circles and vice versa as shown below. Remark 1: If we have NFA rather than DFA. b }. we must first convert it to DFA before swapping states to get its complement.

if L1 and L2 are regular languages. concatenation and Kleene star operations. In particular De Morgan's law also applies to languages. Since L1 L2 = by De Morgan's law. then their complements are regular languages. L1 L2 is regular. Therefore all the properties of sets are inherited by languages.Intersection of Regular Languages Langauges are sets. By Remark 2 above. then Submit. Context-Sensitive and Phrase Structure Grammars . difference. Thus summing all this up we can say that the set of regular languages over an alphabet is closed with respect to union.Regular Grammar Back to Study Schedule Back to Table of Contents Finite Automata Regular Grammar Subjects to be Learned • • • Production and Grammar Regular Grammar Context-Free. intersection. Next -. Test Your Understanding of Complemnent and Intersection of FAs Indicate which of the following statements are correct and which are not. Click True or Fals .

The following theorem holds for regular grammars. These rules mean that S is rewritten as a or as aS.e. a ) = Z. Add another state as the accepting state Z. Then apply the first rule to aS to rewrite S as a. i. aS. If we are not interested in the intermediate steps. finite automata and construction from simple languages using simple operations. aa. which is { a. . } .{ by a regular grammar. that is by something called grammar. If there is no is derived from ambiguity about the grammar G that is referred to. . a ) = Y and for every production X -> a add the transition ( X. where X and Y are nonterminals and a is a terminal. we write =>*G is obtained from a string and say that by applying productions of . to obtain aS. In general if a string a grammar G. . add the transition ( X. start with S and apply the second rule to replace S with the right hand side of the rule. V = { S } and P = { S -> aS. and a set of rewrite rules (productions) P.Contents We have learned three ways of characterising regular languages: regular expressions. A grammar is regular if and only if is a single nonterminal and is a single terminal or a single terminal followed by a single nonterminal. = {a.e. One can generate the strings of this language by the following procedure: Let S be a symbol to start the process with. a set of terminals (the alphabet of the language). Rewrite S using one of the following two rules: S -> a . For example. regular. which ia a nonterminal. b}. A grammar is a set of rewrite rules which are used to generarte strings by successively rewriting symbols. We write S => aS to express that aS is obtained from S by applying a single production. Then for every production X -> aY. } can be generated This can be proven by constructing an FA for the given grammar as follows: For each nonterminal create a state. and S -> aS . then we simply write =>* Formally a grammar consists of a set of nonterminals (or variables) V. Theorem 3: A language L is accepted by an FA i. Thus the process of obtaining aa from S is written as S => aS => aa . For example consider the language represented by a+. That gives us aa. if L . S corresponds to the initial state. S -> bS. aaa. a start symbol S. To generate the string aa for example. There is yet another way of characterizing them. where is a string of terminals and nonterminals with at least one nonterminal in it and is a string of terminals and nonterminals. A production has in general the form -> . that is a production is of the form X -> a or X -> aY. . S -> } is a regular grammar and it generates all the strings consisting of a's and b's including the empty string. the fact that aa is obtained from S is written as S =>* aa .

P. and adding transitions ( S.{ regular grammar. . Z } and ( S. then since { } is regular . V = { S } and P = { S -> aS. . Theorem 4 : If L is regular i. Thus L . b }. An NFA that recognizes this language can be obtained by creating two states S and Z. X. a) = Y . The NFA thus obtained is shown below. Y -> bS. If L contains ( L -{ } ) { } is also regular. Z } . L = Conversely from any NFA < Q. a ) = { S. } is generated by a For example. a regular grammar corresponding to the NFA given below is < Q. where Q = { S. Thus the following converse of Theorem 3 is obtained. P = { S -> aS. as its member. q0. and for any a in and any nonterminal X. S > . . . X -> a is in P if and only if (X. X -> aY. S -> aX. q0 > is obtained as follows: for any a in .e. S -> a. and nonterminals X and Y. accepted by an NFA. S -> b } form a regular grammar which generates the language ( a + b )+. X -> bS. Y } . where S is the initial state and Z is the accepting state of the NFA. b ) = { S. X -> aY is in P if and only if (X.For example = {a.{ } is regular. S -> bS. A > a regular grammar < Q. then L . a) = Y for some accepting state Y. b}. { a. P. S -> a }.

ZX -> XZ. S1 -> XYZS1. bZ -> bc. Z. YX -> XY. For example P = { S -> aSb. possibly empty except Thus the nonterminal X can be rewritten as only in the context of 1X 2 . aY -> ab. where X is a nonterminal and . for every production -> . b } and V = { S } is a contex-free grammar and it generates the language { anbn | n is a positive integer } . that is. ZY -> YZ. context-sensitive languages and phrase structure languages. where is a string of terminals and nonterminals. 1 . These grammars are distinguished by the kind of productions they have but they also form a hierarchy. As we shall see later this is an example of context-free language which is not regular. that is the set of regular languages is a subset of the set of context-free languages which is in turn a subset of the set of context-sensitive languages and the set of context-sensitive languages is a subset of the set of phrase structure languages. aX -> aa. They are characterized by context-free grammars. BY -> bb. S1 -> XYZ. S -> XYZ. A grammar is a context-free grammar if and only if its production is of the form X -> . Context-sensitive grammars are also characterized by productions whose left hand side is not longer than the right hand side. context-sensitive grammars and phrase structure grammars. . 2 and are strings of terminals and nonterminals. S. cZ -> cc } with = { a. possibly the empty string. S1 } is a context-sensitive grammar and it generates the language { anbncn | n is a positive integer } .| | | |. A grammar is a context-sensitive grammar if and only if its production is of the form 1 X 2 -> 1 2 . X -> a. b. respectively.In addition to regular languages there are three other types of languages in Chomsky hierarchy : context-free languages. Y. It is an example of context-sensitive language which is not context-free. For example P = { S -> XYZS1. c } and V = { X. S -> ab } with = { a.

Let M = < Q . there is no restriction on the form of production. There are two sets of questions. Minimization Algorithm for DFA Construct a partition = { A. Then the following algorithm produces the DFA. then Submit. is that for any regular language there is a unique DFA having the smallest number of states that accepts it. that has the smallest number of states amomg the DFAs that accept L. Click True or Fals .Minimization of DFA Back to Study Schedule Back to Table of Contents Finite Automata Minimization of DFA One important result on finite automata. . Q . A > be a DFA that accepts a language L. Next -. new := new_partition( . that is a production of a phrase structure grammar can take the form can be any string. q0 . }. -> . both theoretically and practically. denote it by M1.A } of the set of states Q . . where and Test Your Understanding of Regular Grammar Indicate which of the following statements are correct and which are not.For a phrase structure grammar.

The accepting states of M1 are representatives that are in A. new := new_partition( := . If a transition from s to t on symbol a exists in M. Let us also denote by p and q the sets of states of the original DFA M represented by p and q. S remains in the output partition. These representatives are states of minimum DFA M1. end Minimum DFA M1 is constructed from • • final as follows: • • Select one state in each set of the partition final as the representative for the set. Let s be a state in p and t a state in q. Remove from M1 the dead states and the states not reachable from the start state. states of minimum DFA M1. Note that the sets of final are either a subset of A or disjoint from A.e. The subsets thus formed are sets of the output partition in place of S. Any transitions to a dead state become undefined. if there are any. then the minimum DFA M1 has a transition from p to q on symbol a. A state is a dead state if it is not an accepting state and has no out-going transitions except to itself. final function new_partition( ) for each set S of do partition S into subsets such that two states p and q of S are in the same subset of S if and only if for each input symbol. The start state of M1 is the representative which contains the start state of M.while ( := new new ) ) . If S is not partitioned in this process. Example 1 : Let us try to minimize the number of states of the following DFA. p and q make a transition to (states of) the same set of . Let p and q be representatives i. . respectively.

4 } }. since on b 2 goes to 1. Note here that state 4 is a dead state because the only transitionout of it is to itself. 2 . 5 } . Since the rest of the states are singletons. { 4 ] }. Thus final = { { 1 . { 2 } . Since the rest are singletons. they have the obvious representatives. 5 }. { 4 ] }. since 1 and 5 do the same transitions. and to 2 on b in the original DFA. Further. states 2 and 3 are going to be separated from each other in new . in the When new_partition is applied to this new . . states 3 and 4 are going to be separated from each other in new. So they are not going to be split. 2 and 4 are separated from each other in new. and 1 to 2 on b. all transitions between them are inherited for the minimized DFA. state 3 goes to state 5 and 4 and 5 are in different sets in . Also since on a sate 4 goes to sate 4.Initially = { { 1 . This becomes the second iteration. Thus the new partition is { { 1 . and 3 goes to 1 on a in the original DFA. new_partition is applied to . 5 } . Since on b state 2 goes to state 1. remains unchanged. in the minimized DFA transitions are added from 1 to 3 on a. state 3 goes to state 4 and 1 and 4 are in different sets in . { 3 } . On the other hand 1 and 5 make the same transitions. 5 } . and from 3 to 1 on a. Select 1 as the representative for { 1 . since 1 goes to 3 on a. 4 goes to 4 and 1 and 4 are in different sets in . For the transitions. 3 . Thus the set of states for the minimized DFA is { 1 . { 2 } . { 3 } . { 2 . 3 }. in the minimized DFA transitions are added from 2 to 1 on b. Also since 2 goes to 1 on b.

Thus the minimized DFA is as given in the following figure: Example 2 : Let us try to minimize the number of states of the following DFA. .

By applying new_partition to this . 4 . then Submit. { 3 } . Test Your Understanding of Minimization of DFA Indicate which of the following statements are correct and which are not. 6 } } is obtained. { 6 } } is obtained. { 2 } . { 2 } .Application of FA Back to Study Schedule Back to Table of Contents Finite Automata Application of Finite Automata Subjects to be Learned • Reactive system . Click True or Fals . Thus the number of states of the given DFA is already minimum and it can not be reduced any further. { 6 } } is obtained. Applyting new_partition to this . Applyting new_partition again. new = { { 1 } . new = { { 3 } . new = { { 3 } . 6 } }. Next -. { 1 . { 5 } . 2 . 4 . { 2 .Initially = { { 3 } . { 5 } . 4 } . 5 } . { 1 . { 1 . { 4 } . 5 .

If it is valid. then it accepts the user and starts a session. We could make it go to a different state and count the number of login attempts for security purpose. If the second password fails. Again what we have seen is a model for one level of abstraction. it goes to the initial state and starts all over again. which is another state. different states would be identified and transitions would have to be selected accrdingly. different states and transitions would have to be chosen. it checks whether or not the name is valid. When the session terminates. Upon completion of the RFNM. then it asks for and then waits for the password. Again depending on the level of abstraction. it goes into another wait state waiting for an ACK to come from the other computer. it sends "Acknowledgement" (ACK) to the other computer. Thus a finite automaton that models this protocol has the following five states: initial state (wait for RFNM). which is another state. it goes into the state of receiving it (Our interpretation is that the computer is in a state of receiving an RFNM and it is taking the action of receiving the RFNM) . If the user name typed in is not valid. When it is complete. The next example is a protocol for a computer to follow in communicating with another computer. Again it is a very simplified version. receiving RFNM. If a negative ACK is received. That is another state though it could further be broken down into a number of more states. When a RFNM starts coming. it goes back to the initial state and waits for another RFNM to come.Initially the computer waits for a user name to be typed in. When a password is typed in and it is correct. sending ACK. If a positive ACK is received. This is one state of the system. That is a fourth state. goes back to the initial state and waits for another login. it gets a signal. it goes back to the initial state. Depending on how much detail we are interested in. If the password typed in is incorrect. After sending the ACK. sending message and waiting for ACK. Initially the computer is in wait state waiting for "Request for Next Message" (RFNM) to come from another computer. . it starts sending the requested message to the other party. But let us make it simple. then it informs the user of that and waits for the next try. When a name is typed in. it resends the message.

holds the next state for each input symbol for each state. called TOKEN.Next -. It uses four arrays. called ACTION. indicates the actions taken at each state and a fourth. One array. while ( TOKEN [index] . keeps the index of the first symbol in the TOKEN array for each state. One such simulation algorithm is given below. called STATEX. input := read_input( ) . called NEXT_STATE. stores for each state the input symbols that trigger transitions from the state. Algorithm FA Simulator state := INITIAL_STATE. while ( state NO_of_STATES and not End of Input ) index := STATEX [state] . we can use a general purpose program to simulate its operation. 0 and TOKEN [index] input ) index := index + 1.Simulation of FA Back to Study Schedule Back to Table of Contents Finite Automata Simulation of Finite Automata Subjects to be Learned • Simulation of FA Contents Once a finite automaton is constructed. A third array. Another array. Those indices are used to access the contents of the other arrays.

S is the initial state and B and H are accepting states. B.e. C and H.if ( TOKEN [index] 0) perform the action specified by ACTION [index]. The corresponding transitions are going to be shown by red arrows in the transition diagram. . The numbers below NEXT_STATE array show the correspondence between the indices of the STATEX array and the states A. else error input := read_input( ) . Then type 3 Tab . Tab 4 Tab 5. end Here 0 in the TOKEN array is a marker between states. Then every time you click "SHOW" the number is processed digit by digit. S corresponds to 1. So no action is taken as a number is processed.45. to input 3. You must hit the "Tab" key to move to the next box. For example. first click the box pointed by the red arrow in the figure below. The ACTION array would contain pointers to actions to be taken corresponding to arcs traversed such as converting a digit in BCD form to the corresponding binary number. state := NEXT_STATE [index]. In the DFA below all the transitions to the empty state (i. At the moment it is empty. Then type in a number you want the FA to recognize. first click the box under the red arrow. empty transitions) are omitted. To see how this algorithm works.

So be patient. click here It is extremely slow. .If you are also interested in how code is executed.

The main idea behind these test methods is that finite automata have only finite amount of memory in the form of states and that they can not distinguish infinitely many strings. because aak and aaak are in the language an for any . There are. Thus there is no way for a finite automaton to remember how many a's it has read for all possible strings anbn . For example to recognize the language { anbn | n is a natural number} . For example.Non-Regular Languages Back to Study Schedule Back to Table of Contents Regular Languages Non-Regular Languages Subjects to be Learned • • • Existence of non-regular languages Myhill . languages that are not regular and therefore require devices other than finite automata to recognize them. Non-regularity test based on Myhill-Nerode's theorem Indistinguishability of strings: Strings x and y in * are indistinguishable with respect to a language L if and only if for every string z in *. a and aa are indistinguishable with respect to the language an over alphabet { a }. we can conclude that { anbn | n is a natural number} is not regular. where n is a positive integer. Thus it must be in different states when it has read different number of a's and starts reading the first b. That is the main limitation of finite automata.Nerode Theorem for non-regularity test Pumping Lemma Contents We have learned regular languages.Next -. their properties and their usefulness for describing various systems. either xz and yz are both in L or they are both not in L. a finite automaton must remember how many a's it has read when it starts reading b's. however. In this section we are going to study some of the methods for testing given languages for regularity and see some of the languages that are not regular. But any finite automaton has only finite number of states. This is the basis of two of the regularity test methods we are going to study below: Myhill-Nerode Theorem and Pumping Lemma. Since a regular language must be recognized by a finite automaton.

that is. Example 3: Let L3 be the set of algebraic expressions involving identifiers x and y. Theorem : A language L over alphabet is nonregular if and only if there is an infinite subset of * . Select bm as a string to be appended to ak and am . Hence ak and am are distinguishable with respect to L1 . Then akbakb is in L2 while ambakb is not in L2 . Using this concept of indistinguishability. ( ( x + y ) * x ) and (( (x*y) + x ) + (y*y) ) are algebraic expressions. S1 is over alphabet { a . Then akbm is not in L1 while ambm is in L1 . b }* } is nonregular. operations + and * and left and right parentheses. with respect to the language anbn . Consider the set of strings S3 = { (k x | k is a positive integer } . x . Hence L2 is nonregular. whose strings are pairwise distinguishable with respect to L. S2 satisfies the conditions of Myhill-Nerode theorem. (x*y) . It is stated without a proof. For example. This set is infinite . Let ak and am be arbitrary two different members of the set S1. However. then ( + ) and ( * ) are in L3 . Hence L1 is nonregular. Let ak and am be arbitrary two different members of the set. where k and m are positive integers and k m . Consider the set of strings S2 which is the same as S1 of Example 1 above. L3 can be defined recursively as follows: Basis Clause: x and y are in L3 . Select bakb as a string to be appended to ak and am . the set of strings consisting of one or more right parentheses followed by identifier x. Example 1: L1 = { anbn | n is a positive integer } over alphabet { a . S1 satisfies the conditions of Myhill-Nerode theorem. Inductive Clause: If and are in L3 . Since ak and am are arbitrary strings of S2. b } can be shown to be nonregular using Myhill-Nerode as follows: Consider the set of strings S1 = { an | n is a positive integer } . We are going to show that its strings are pairwise distinguishable with respect to L1. the following theorem by Myhill and Nerod gives a criterion for (non)regularity of a language. For more on Myhill-Nerode theorem click here. It can be shown to be pairwise distinguishable with respect to L2 as follows. a and aa are not indistinguishable (hence distinguishable). because ab is in the language anbn while aab is not in the language. Extremal Clause: Nothing is in L3 unless it is obtained from the above two clauses. Since ak and am are arbitrary strings of S1.positive integer k. b } and it is infinite. Example 2: L2 = { ww | w {a. where k and m are positive integers and k m . Hence ak and am are distinguishable with respect to L2 .

where k and m are positive integers and k m . abbabbabbb. then there must be a cycle in the NFA along some path from the initial state to some accepting state (such as the cycle 2-3-4-2 in the above example). Hence S3 is pairwise distinguishable with respect to L3 . The following theorem which is called Pumping Lemma is based on this observation. abbabbabbabbb etc. In general if a string w (such as abbabbb in the example above) is accepted by an NFA with n states and if its length is longer than n. the resultant strings such as abbb (bba repeated 0 times). Those strings which are accepted by this NFA and whose length is greater than 5 have a substring which can be repeated any number of times without being rejected by the NFA.and it can be shown to be pairwise distinguishable with respect to L3 as follows: Let (k x and (m x be arbitrary two strings of S3 . Hence L3 is not regular. For example [ + x ) ]3 is +x) +x)+x) . It . Select [ + x ) ]k as a string to be appended to (k and (m . abbabbabbb etc. Then the substring representing that cycle (bba in the example) can be repeated any number of times within the string w without being rejected by the NFA. Then (k x + [ + x ) ]k is in L3 but (m x + [ + x ) ]k is not in L3 because the number of ('s is not equal to the number of )'s in the latter string. Pumping Lemma Let us consider the NFA given below. This NFA accepts among others some strings of length greater than 5 such as abbabbb. For example the string abbabbb is accepted by the NFA and if one of its substrings bba is repeated any number of times in abbabbb. are also accepted by the NFA.

Then there must be strings u. Example 4: As an example to illustrate how Pumping Lemma might be used to prove that a language is nonregular. n. v has at least one symbol. and w such that x = uvw. uvmw L. Since p > 0 . Note that Pumping Lemma gives a necessity for regular languages and that it is not a sufficiency. let us prove that the language L = akbk is nonregular. Then there is an FA that accepts L. It is stated without a proof here. Thus Pumping Lemma can not be used to prove the regularity of a language. Then for any string x in L with |x| n. Since |v| > 0 . This violates the condition that for every m language. the language is not necessarily regular. where k is a natural number. Also since |uv| Let us now consider the string uvmw for m = 2.states that if a language is regular. uvmw L. then any long enough string of the language has a substring which can be repeated any number of times with the resultant strings still in the language. and for every m 0. for some p > 0 . that is. even if there is an integer n that satisfies the conditions of Pumping Lemma. Then uv2w = an-pa2pbn = an+pbn . n + p language L represented by akbk . Hence an+pbn can not be in the L. Pumping Lemma : Suppose that a language L is regular. n . |uv| n |v| > 0 . Hence L is not a regular 0. v and w which satisfy the following relationships: x = uvw |uv| n |v| > 0 and for every integer m 0. Suppose that L is regular and let n be the number of states of an FA that accepts L. there are strings u. It can only show that a language is nonregular. v. Let n be the number of states of that FA. uvmw Test Your Understanding of Non-regularity . v = ap. Consider a string x = anbn for that n.

if and only if they are indistinguishable with respect to . This means that if x and y are read by an DFA that recognizes L. Also it is a corollary to Myhill-Nerode theorem: Let { be the followijg relation on }={ : For strings and of . If there are three strings that are distinguished with respect .Context-Free Grammar Back to Study Schedule Back to Table of Contents Myhill-Nerode Theorem The non-regularity test for languages by Myhill-Nerode is based on the following theorem which is in the contrapositive form of the theorem used for nonregularity test. the DFA reaches different states. Then there is a string z such that xz is in L and yz is not in L (or xz is not in L and yz is in L). Proof of Theorem Necessity Suppose that a language L is regular and two strings.Indicate which of the following statements are correct and which are not. that is. Click True or Fals . then Submit. Then the theorem is is staed as follows: Theorem: A language L over alphabet is regular if and only if the set of equivalence classes of is finite. Next -. say x and y. if and only if }. are distinguishable with respect to L.

DFA ( = {[ =[ = {[ ( . To prove this... are in different classes for and are distinguishable with respect to L. ]. is in exactly one equivalence class.. the index) of and let .. then the language L is regular. . ]. For. then the language is not regular. . [ ]=[ } ].. namely ] and ].. [ .e.. making them belong to different [ Hence is a function. be the number of distinct equivalence classes (i. then the DFA must have infinitely many states. let [x] denote a class of strings that are indistinguishable from a string x with respect to L. . Let .. be representatives of those distinct equivalence classes. Hence if there is an infinite set of strings which are pairwise distinguishable with respect to a language.to L. Then we construct a . if the number of classes of strings that are pairwise indistinguishable with respect to a language L is finite. Let us now show that this machine is in fact a DFA and it accepts the language First. . where [ ] =[ ] for all . Note that "indistinguishable with respect to L" is an equivalence relation over the set of strings (denote it by ) and [x]'s are equivalence classes. then ]'s. note that for every string [ ]. if and in [ ].. We will show that a DFA that accepts L can be constructed using these equivalence classes. then the DFA reaches three different states after reading those three strings. Sufficiency Conversely. Hence if there are infinitely many strings to be distinguished with respect to L. which it can not because a DFA must have a finite number of states. . ) as follows: ]} ].

With these terminology. = [ ]. where ] is the equivalence class that belongs to. if then for every . . . if the set of its equivalence classes is finite. ]. by the definition of = . ] by the induction hypothesis. first note that if . . then every We then show that for every string . Also an equivalence relation is said to be of finite index. Then by the definition of . this means that the DFA accepts Myhill-Nerode Theorem Let us here state Myhill-Nerode Theorem. For that. First some terminology. Inductive Step: Assume Then for every But =[ Hence =[ ]. Hence =[ = . Hence Hence we have shown that for every string ]. for DFA.Next. Our proof is by structural induction on string Basis Step: = =[ ]. =[ ]. . let us show that this DFA accepts string in [ ] is also in L. if a string in [ in ] is in . Since . Myhill-Nerode Theorem can now be stated as follows: The following three statements are equivalent: (1) A language is regular. by the definition of = . =[ ]. . where ]. . where . An equivalence relation on is said to be right invariant if for every .

S -> ab }. Most programming languages can be approximated by context-free grammar and compilers for them have been developed based on properties of context-free languages. (3) is of finite index. They are grammars whose productions have the form X -> . S . grammar (CFG) if V and . b } and P1 = { S -> aSb . Let us define context-free grammars and context-free languages here. Definition (Context-Free Grammar) : A 4-tuple G = < V . P > is a context-free V is V. and P is a finite set of productions of the form X -> and (V )* . are finite sets sharing no elements between them. where X is a nonterminal and is a nonempty string of terminals and nonterminals. For the following context-free grammar G1 = < V1 . A language is a context-free language (CFL) if all of its strings are generated by a context-free grammar. P1 > generates L1 : V1 = { S } . Example 1: L1 = { anbn | n is a positive integer } is a context-free language.(2) L is the union of some of the equivalence classes of a right invariant equivalent relation of finite index. The set of strings generated by a context-free grammar is called a context-free language and context-free languages can describe many practically important systems. . S . where X the start symbol. Context-Free Languages Context-Free Grammar Subjects to be Learned • • • Context-Free Grammar Context-Free Languages Push Down Automata Contents Earlier in the discussion of grammars we saw context-free grammars. Proofs are omitted. . S . = { a .

} . w is spelled backward to obtain wr . Example 3: Let L3 be the set of algebraic expressions involving identifiers x and y. Pu > is a context-free grammar that generates the language L1 L2 . For the following context-free grammar G2 = < V2 . respectively. b } and P2 = { S -> aSa . . Then L1 context-free languages. that is. Then let Sc be a symbol which is not in V1 V2 . . S -> aa . Next define Vc = V1 V2 { Sc } and Pc = P1 P2 { Sc -> S1S2 } . S -> x . = { a . b }+ } is a context-free language . S1 . . Su . . Sc . < if-statement > -> if ( < expression > ) < statement > . < expression > -> < logicalexpression > . P3 > generates L3 : V3 = { S } . S2 .Example 2: L2 = { wwr| w {a. ( . . P1 > and G2 = < V2 . . < statement > -> < for-statement > . * } and P3 = { S -> ( S + S ) . Then let Su be a symbol which is not in V1 V2 . ) . S . Then it can be easily seen that Gc = < Vc . . let Ss be a symbol which is not in V1 . < forstatement > -> for ( < expression > . operations + and * and left and right parentheses. . first relabel symbols of V2 . L1L2 . Then it can be easily seen that Gu = < Vu . so that V1 and V2 don't share any symbols. < expression > . < expression > -> < algebraic-expression > . P2 > generates L2 : V2 = { S } . . first relabel symbols of V2 . y . S -> S*S . Ss -> L2 . For L1* . if necessary. S -> bb }. . if necessary. S -> bSb . Similarly for L1L2 . Properties of Context-Free Language Theorem 1: Let L1 and L2 be context-free languages. . P2 > be context-free grammars generating L1 and L2 . + . S . 3 = { x . < statement > -> < assignment > . . so that V1 and V2 don't share any symbols. Su -> S2 } . where w is a nonempty string and wr denotes the reversal of string w. For the following context-free grammar G3 = < V3 . < expression > ) < statement > . concatenation and Kleene star of context-free grammars as follows: Let G1 = < V1 . Then for L1 L2 . . Pc > is a context-free grammar that generates the language L1L2 . Next define Vu = V1 V2 { Su } and Pu = P1 P2 { Su -> S1 . 3. Outline of Proof This theorem can be verified by constructing context-free grammars for union. . Then L3 is a context-free language. Then let Ps = P1 { Ss -> SsS1 . and L1* are . . For example { < statement > -> < if-statement > . S -> y }. Example 4: Portions of the syntaxes of programming languages can be described by context-free grammars. .

It can be seen that the grammar Gs = < Vs . check the stack. generates the language L1* . as a's are read by the finite automaton. q0 . . If another a (or anything other than b) is read after the first b. reject the string. Z0 . accept the string. > . respectively ). If it is empty. This automaton behaves like a finite automaton except the following two points: First. ) means the following: The automaton moves from the current state of p to the next state q when it sees an input symbol a at the input and X at the top of the stack. Let us define this new type of automaton formally. A is the set of accepting states is the transition function and :Q ( ( } -> 2 Q * . A pushdown automaton ( or PDA for short ) is a 7-tuple M = < Q . and are finite sets ( the input and stack alphabet. q0 . Let us consider a context-free language anbn . They need a little more complex automata called pushdown automata. the contents of the stack can also be changed every time an input symbol is read. When all the symbols of the input string are read. X ) = ( q . For example. Any string of this language can be tested for the membership for the language by a finite automaton if there is a memory such as a pushdown stack that can store a's of a given input string. Z0 . a . Otherwise reject it. and it replaces X with the string at the top of the stack. Pushdown Automata . . its next state is determined not only by the input symbol being read. . Thus its transition function specifies the new top of the stack contents as well as the next state. but also by the symbol at the top of the stack. . . A . Thus ( p . Second. A . context-free languages are also accepted by automata but not finite automata. As soon as the symbol b appears stop storing a's and start popping a's one by one every time a b is read.} . where Q = { q0 . Z0 is the initial stack symbol and it is a member of . where Q is a finite set of states. Ss . Ps > is a context-free grammar that Like regular languages which are accepted by finite automata. Example 1 : Let us consider the pushdown automaton < Q . push them into the stack. > . q0 is the initial state.

aaZ0 ) ( q1 . Z0 ). A = { q2 } and let be as given in the State Input Top of Stack Move q0 a Z0 ( q0 . b . ) ( q1 . aZ0 ). When the second b is read. its configuration is ( q0 . ) in a single move (a single application of the transition function) we write (p. Z0 ). y . aZ0 ) ( q0 . ) by a sequence of zero or more moves. ( q0 . (p. it is ( q0 . following table: = { a . x is the unread portion of the input string and is the current stack contents. Thus the configuration is ( q1 . where the input is read from left to right and the top of the stack corresponds to the leftmost symbol of . . = { a . . Z0 ) This pushdown automaton accepts the language anbn . abb . aabb . q0 . ) ) (q. we can also write ( q0 .x. b . b . To express that the PDA moves from configuration ( p . After reading the first a. for example. x . . another a is popped from the top of the stack and the PDA stays in state q1 . To describe the operation of a PDA we are going to use a configuration of PDA. After reading the second a. x . aabb . . we write ) is reached from ( p . This entire process can be expressed using the configurations as ( q0 . Next it moves to the state q2 which is the accepting state. . Z0 } . aa ) q0 q1 q1 b b a a Z0 ( q1 . abb . > is a triple ( q . aabb . aZ0 ) ( q1 . If ( q . . Z0 ) * ( q2 . ). Z0 ). * (q. Z0 ) q2 . where q is the state the PDA is currently in. ) to configuration ( q . bb . it moves to state q1 and pops a from the top of the stack. Initially its configuration is ( q0 . bb . ) . aZ 0 ) q0 a a ( q0 . Let us now see how the PDA of Example 1 operates when it is given the string aabb . q2 } .q1 . ) ( q2 . Z0 ) ( If we are not interested in the intermediate steps. . x . A .y. Z0 ) . Z0 .x. Then when the first b is read. A configuration of a PDA M = < Q .y. aZ0 ). Thus the configuration is ( q1 . y . aaZ0 ). b } . Thus aabb is accepted by this PDA. ).

X / ) indicating that X at the top of the stack is replaced by upon reading a from the input. . Example 2 : Let us consider the pushdown automaton < Q . Z0 } .a. Z0 . X ) = ( p . Z0) . . . arcs are labeled differently than FAs. for some in *. > . b . ) ) ) ) ( q1 . = { a . b ( q1 . however. then an arc from state p to state q is added to the diagram and it is labeled with ( a . Like FAs. q0 .k. b .A string x is accepted by a PDA (a. = { a . a . For example the transition diagram of the PDA of Example 1 is as shown below. where Q = { q0 . q2 } . Z 0 ) In this table represents either a or b. bZ 0 ) q0 q0 q0 q1 q1 q1 a b c a b a b Z0 ( q0 . acceptance by final state) if (q0. PDAs can also be represented by transition diagrams. c } . For PDAs. ) . x. If ( q . A . A = { q2 } and let be as given in the following table: State Input Top of Stack Move q0 a Z0 ( q0 . ). * (q. ( q1 . aZ 0 ) q0 b Z0 ( q0 . a ( q0 . ) ( q2 . q1 . and an accepting state q.

1 Further topics on CFL • PDA and Context-Free Language There is a procedure to construct a PDA that accepts the language generated by a given context-free grammar and conversely. In the figure and 2 represent a or b. Z 0 ) . Z0 ) ( q0 . bcbba .This pushdown automaton accepts the language { wcwr | w { a . When c is detected. it accepts the input string. Otherwise it rejects the input string. abbcbba . . aZ0 ) ( q1 . Z 0 ) ( q2 . bbaZ0 ) . The transition diagram of the PDA of Example 2 is as shown below. This PDA pushes all the a's and b's in the input into stack until c is encountered. For example for the input abbcbba. baZ0 ) ( q1 . ba . it pops the stack. it ignores c and from that point on if the top of the stack matches the input symbol. . b }* } . ( q0 . a . bba . bbcbba . ( q0 . That means that a language is . When there are no more unread input symbols and Z0 is at the top of the stack. aZ0 ) ( q0 . cbba . baZ0 ) ( q1 . which is the set of palindromes with c in the middle. bbaZ0 ) ( q1 . it goes through the following configurations and accepts it.

then multiply the result by z. Though we are accustomed to interpreting this as x + (yz) i. it could also be interpreted as ( x + y )z meaning that first compute x + y.e. then add the result to x. Similar things happen when English sentences are processed by computers (or people as well for that matter). Parsing is the process of interpreting given input strings according to predetermined rules i. Contect-free grammars are powerful grammars. y and z which satisfy u = vwxyz |wy| > 0 |wxy| n 0 . we are not going to study parsing here. productions of grammars. The parsing for context-free languages and regular languages have been extensively studied. However. They can describe much of programming languages and basic structures of natural languages.e. a verb phrase usually follow the noun phrase and the first word in the verb phrase is the verb and it is followed by noun phrases reprtesenting object(s) of the verb. Thus if a computer is given the string x + yz. Thus they are widely used for compilers for high level programming languages and natural language processing systems. w. compute yz first. Then there is a positive integer n such that for any string u in L with |u| n . a computer like non-English speaking people must be told how to interpret sentences such as the first noun phrase (" A dog") is usually the subject of a sentence. "A dog" is the subject. • Pumping Lemma for Context-Free Language Let L be a CFL. Interested readers are referred to the textbook and other sources.context-free if and only if there is a PDA that accepts it. vwmxymz L for every integer m • Parsing and Parsers for CFL Consider the algebraic expression x + yz. it does not know which interpretation to use unless it is explicitly instructed to follow one or the other. "bites" is the verb and "a man" is the object of the verb. ???? references on Parsing ???? . x. However. By parsing sentences we identify the parts of the sentences and determine the strutures of the sentences so that their meanings can be understood correctly. For example in the sentence "A man bites a dog". there are strings v. Those procedures are omitted here. native English speakers know that it is the dog that bites and not the other way round.

Turing Machines Back to Schedule Back to Table of Contents Turing Machines Turing Machines Subjects to be Learned • • • Definition of Turing Machine Configuration Operation of Turing Machine Contents Introduction We have studied two types of languages from the Chomsky hierarchy: regular languages and context-free languages. then Submit. They are. and the machines that can process them: Turing machines. Next -. Click True or Fals . These languages can describe many practically important systems and so they are heavily used in practice. the phrase structure languages (also called Type 0 languages).Test Your Understanding of Contect-Free Language Indicate which of the following statements are correct and which are not. In this chapter we are going to study the most general of the languages in Chomsky hierarchy. however. of limited capability and there are many languages that they can not process. Turing machines were conceived of by the English mathematician Alan .

At any state it reads the symbol under the head. Computers we use today are as powerful as Turing machines except that computers have finite memory while Turing machines have infinite memory. At any time it is in one of the finite number of states. The tape has the left end but it extends infinitely to the right. consists of a finite control and a tape. right or stay at the same square after a read or write. it stops its operation. Definition Conceptually a Turing machine. . a Turing machine starts at the initial state. It is also divided into squares and a symbol can be written in each square. its head is a read-write head and it can move left. This conjecture is known as Church's thesis and today it is generally accepted as true. unlike finite automata. We are going to study Turing machines here and through that limitations of computers and computation as we know today. Later Alonzo Church conjectured that any computation done by humans or computers can be carried out by some Turing machine. either erases it or replaces it with a symbol (possibly the same symbol). One of its states is the halt state and when the Turing machine goes into the halt state. However. It then moves the head to left or right or does not move it and goes to the next state which may be the same as the current state.Turing as a model of human "computation". like finite automata. Given a string of symbols on the tape.

. the symbol X currently being read is changed to Y and the tape head is moved as directed by D. . an arc from q to r is drawn with label ( X/Y . S}. L or S . q0 is the initial state. respectively. > . q3 } . . is a finite set of symbols containing as its subset and it is the set of tape symbols.Formally a Turing machine is a 5-tuple T = < Q. Y.S) A transition diagram of this Turing machine is given below. which is assumed not to contain the symbol h. Example 1 : The following Turing machine < Q1 . q0 . = { a . The states are represented by vertices and for a transition ( q. State (q) Input (X) Move ( q0 q1 q2 q3 q3 a b a (q. q2.L. left and do not move it. It is a mapping from Q ( { } ) to ( Q { h } ) ( { }) {R. where Q1 = { q0. is the transition function but its value may not be defined for certain points. D ) . R ) ( q3 . D ) indicating that the state is changed from q to r. q0. is a finite set of symbols and it is the input alphabet. L and S denote move the head right. b } . R ) ( q3 . X ) = ( r. X) ) ( q1 . It is assumed that the tape has at the left end and the head is initially at the left end of the tape. A transition diagram can also be drawn for a Turing machine. q1. = { a . > accepts the language aba* . where Q is a finite set of states. where D represents R. a . b } and is as given by the table below. R ) ( q2 . R ) (h. Here denotes the blank and R. b . a . . . The symbol h is used to denote the halt state. . .

( q0 . For example ( q . aba ) ( q3 . aababb ) shows that the Turing machine is currently in state q. Note that the Turing machine does not stop if a string is not in the language. aba ) ( q1 . xay ) * ( q . . aba ) ( q2 . aba ) (h. yaz ) for some symbol a { } and some strings y and z in ( * { } ) . In this case we also say that the Turing machine halts on input x. A Turing machine T is said to decide a language L if and only if T writes "yes" and halts if a string is in L and T writes "no" and halts if a string is not in L. For example the Turing machine of Example 1 above goes through the following sequence of configurations to accept the string aba: ( q0 . > if x ) * ( h. xay ) ( q . The set of strings accepted by a Turing machine is the language accepted by the Turing machine.Turing Machine that accepts aba* To describe the operation of Turing machine we use configuration. and ( p . If the Turing machine needs to be explicitly indicated T or T* is used. We write ( p . aba ) . q0 . A string x is said to be accepted by a Turing machine* T = < Q . the taper contents are the string aababb and the head is reading the last a of the string. A configuration for a Turing machine is an ordered pair of the current state and the tape contents with the symbol currently under the head marked with underscore. zbw ) if the Turing machine goes from the first configuration to the second in one move. . zbw ) if the Turing machine goes from the first configuration to the second in zero or more moves.

where = { a }.The first of the following figures shows a Turing machine that accepts but does not decide the language { a }. the second is a Turing machine that accepts { a } but goes into a loop if a string is not in the language (hence it accepts but doe not decide { a }) and the third decides { a }. .

.

. to the right of Example 3 : The following Turing machine erases the string on the tape and moves the head to the left end.Example 2 : The following Turing machine moves the head to the first the current position. It is assumed that initially the tape has at the left end. This Turing machine is denoted by TE. It is denoted by TR .

(2) no transition is specified for the current configuration and (3) the head is at the left end and it is instructed to move left. but it goes into an infinite loop for any strings that are not in the language. In cases (2) and (3). . the operation of the Turing machine is aborted. that is when a Turing machine does not halt on a string.Strings not Accepted by Turing Machines When a string is not accepted by a Turing machine. For example the following Turing machine accepts the language a+. one of the following three things happens: (1) The Turing machine goes into an infinite loop.

a string is accepted by a Turing machine if given the string. T does not halt on x. A Turing machine thus may accept a string and halt. in the Turing machines those books define. there is no difference between these two definitions of "accept". Then we say T computes f or f is computable if for every x ( q0 . reject a string and halt.Turing machine accepting a+ Computabler Function Let S * and let f be a function f : S -> S. * and for every x that is not in S. there are two halt states: "accept halt" and "reject halt". the Turing machine eventually goes into the accept halt state. or loop. * Note on "Turing-acceptable": Some books define "acceptance by Turing machine" slightly differently. With this definition. f(x) ) * . A language is a phrase structure (type 0) langauage if and only if it is Turing-acceptable in either sense and it has no effects on decidablility. . As far as the material discussed in this class note. x) * ( h. That is.

There are two sets of questions. We have already seen TR .Combination of Turing Machines Back to Study Schedule Back to Table of Contents Turing Machines Combination of Turing Machines Subjects to be Learned • Combination of Turing Machines Contents Combination of Turing Machines We have seen the definition of Turing machine and a few simple examples. Here we are going to study how complex Turing machines can be constructed using simple Turing machines and how computers can be simulated by Turing machines. Next -. It moves the head to the first symbol (which may be ) . Let us start with some basic Turing machines. Click True or Fals . any "computation" done by human beings or machines can be done by some Turing machine. Furthermore according to the Church's thesis. In fact Turing machines that simulate computers and Turing machines that perform computations done by any algorithm can be constructed. One can construct many more Turing machines that perform various functions. then Submit.Test Your Understanding of Turing Machines Indicate which of the following statements are correct and which are not.

then T2 is started as in the case of T1T2 . Below is assumed to be at the left end of the tape initially. T1T2 and T1 -> T2 denote the Turing machine that behaves initially like T1 and when T1 halts T2 takes over inheriting the head position and the tape contents of T1 . Then by T we denote a Turing machine that writes symbol at the current position and does not move the head (stays). . Otherwise it crashes. T1 -> T2 denote the Turing machine that first executes T1. let us construct a little more complex Turing machines. The halt state of T1 becomes the initial state of T2 . To combine Turing machines we use the following conventions: Let T1 and T2 represent arbitrary Turing machines. Example 4: The following machine shifts the tape contents to the left one position. Then if T1 halts and if the symbol currently under the head is . Using these basic machines and the convention.to the right of the current position. Similarly by TL we denote a Turing machine that moves the head to the first symbol (which may be ) to the left of the current position. Also by TR and TL we denote Turing machines that move the head to right and left one position. respectively. takes the head to the right end of the string and halts.

natural numbers are represented on a Turing machine using symbol I. it goes through the following -> aab -> a -> a -> abb -> ab -> ab -> ab Example 5: The left-shift machine of Example 4 can be used to construct an adder for natural numbers. In general to represent a natural number k. For example the number 3 is represented by three consecutive I's on the tape and 5 by five I's. x x).For example with the initial tape contents of sequence of tape contents and ends with ab ab -> ab : ab . First. So the initial configuration for adding 2 and 3 is ( q0 . Example 6: The following Turing machine copies the tape contents at the left end to their right separated by a blank . After the addition the configuration becomes ( h . . An adder can be constructed for example as TR -> TSL TL . IIIII ) . k consecutive I's are put on the tape. that is ( q0 . x) * b b (h. m I's and n I's with a blank between them are placed on the tape. II III ) . To add two numbers m and n. After adding two numbers placed on the tape it moves the head to the left end and halts.

it is not difficult to construct a Turing machine that performs subtraction using the same representation of numbers as for the addition. all of those instructions can be realized using combinations of a small number of basic instructions. Click True or Fals . The following notations are used in the questions: . On the other hand as we have seen above. Since any complex computer instructions can be realized using those basic instructions. branching. and store and load operations can be taken care of by a Turing machine that copies tape contents. then Submit. Test Your Understanding of Combination of Turing Machines Indicate which of the following statements are correct and which are not. there is a Turing machine that performs addition. Furthermore if the subtraction operation is necessary. Thus by combining appropriate Turing machines a computer with a minimal instruction set can be constructed. However. one can say that computers can be simulated by Turing machines. the branch operation is already in Turing machines because next configurations are determined based on the current state and tape symbol being looked at. store and load operations.Today's computers are very complex machines and their instruction sets contain complicated operations. A bare minimum instruction set would contain addition. In fact many of the earlier computers had a much smaller instruction set but still could do everything today's computers can do albeit much more slowly. All the other operations can be realized by using those basic operations.

what one type can compute any other can also compute. may vary. that is. That is. However. the efficiency of computation. first we map the squares of the two dimensional tape to those of the one dimensional tape diagonally as shown in . that is. nondeterministic Turing machines etc. Turing Machines with Two Dimensional Tapes This is a kind of Turing machines that have one finite control. the former can be simulated by the latter.Types of Turing Machines Back to Study Schedule Back to Table of Contents Turing Machines Types of Turing Machines Subjects to be Learned • Variation of Turing Machine Contents There are a number of other types of Turing machines in addition to the one we have seen such as Turing machines with multiple tapes.T_a : Ta T_R : TR ->^b : ->b Next -. To simulate a two dimensional tape with a one dimensional tape. For any Turing machine of this type there is a Turing machine with a one dimensional tape that is equally powerful. one read-write head and one two dimensional tape. ones having one tape but with multiple heads. ones with two dimensional tapes. The tape has the top end and the left end but extends indefinitely to the right and down. how fast they can compute. It is divided into rows of small squares. It turns out that computationally all these Turing machines are equally powerful.

It is denoted by a 5-tuple < Q . . > . Since the converses are obviously true. It can be proven that any language accepted by an n-tape Turing machine can be accepted by a one tape Turing machine and that any function computed by an n-tape Turing machine can be computed by a one tape Turing machine. {h}) ( { } {R.S It can be easily seen that this type of Turing machines are as powerful as one tape Turing machines. right or left on the two dimensional tape. . . A configuration for this kind of Turing machine must show the current state the machine is in and the state of each tape. H2 . It turns out that this type of Turing machines are only as powerful as one tape Turing machines whose tape has a left end. Hn } ( { } ) -> ( Q }. In each state only one of the heads is allowed to read and write.Similarly formulas can be found for the head position on the one dimensional tape corresponding to move up. Its transition function is a partial function :Q ( { } )n -> ( Q { h } ) ( { } )n { R . Hn denote the tape heads.. Turing Machines with Multiple Tapes : This is a kind of Turing machines that have one finite control and more than one tapes each with its own read-write head. . Turing Machines with Multiple Heads : This is a kind of Turing machines that have one finite control and one tape but more than one read-write heads. q0.. where H1 . >.. Since Turing machines with a two dimensional tape obviously can simulate Turing machines with a one dimensional tape. . q0.L. it can be said that they are equally powerful. L . The transition function is a partial function : Q { H1 . Nondeterministic Turing Machines . Turing Machines with Infinite Tape : This is a kind of Turing machines that have one finite control and one tape which extends infinitely in both directions. . Hence they are at least as powerful as Turing machines with a two dimensional tape. Thus some Turing machines with a one dimensional tape can simulate every move of a Turing machine with one two dimensional tape. Details are omitted. one can say that one tape Turing machines are as powerful as n-tape Turing machines. S } n . H2 .. It is denoted by a 5-tuple < Q .

For example consider the following nondeterministic Turing machine that accepts a+ . . Then it would compare the first half of x with the second half by comparing the i-th symbol of the first half with the i-th symbol of the second half for i = 1. It must find the midpoint by for example pairing off symbols from either end of x. Given a string x. TN starts at the initial configuration and goes through a sequence of configurations until it reaches a halt configuration . It can be shown that a nondeterministic Turing machine is only as powerful as a deterministic Turing machine. that is the place where the second half of x starts. Formally a nondeterministic Turing machine is a Turing machine whose transition function takes values that are subsets of ( Q {h}) ( { } {R. The root of the tree is the initial configuration and it is the only vertex of level 0. The children of all the vertices of level i form level i+1. can take any action selecting from a set of specified actions rather than taking one definite predetermined action. A deterministic Turing machine.L. At any point in the process TN is in some configuration and has a finite set of configurations to choose from for its next configuration. In general for each vertex of level i all possible configurations that are reachable by applying the transition function of TN are its children. moving the tape head and going to a next state.. . 2. Theorem Any language accepted by a nondeterministic Turing machine is also accepted by some deterministic Turing machine. like nondeterministic finite automata. it is understood that a nondeterministic Turing machine at any configuration selects one combination of next state. Given a string x . on the other hand.A nondeterministic Turing machine is a Turing machine which. . tape symbol and head movement out of the set of triples without following any specific predetermined rule. a nondeterministic Turing machine that accepts this language L would first guess the midpoint of x. As in the case of NFA. The set of all possible computations that TN can perform for a given string x can be represented by a rooted tree as follows. Note that the number of children for a vertex in this tree is finite because the number of states is finite and there are a finite number of tape symbols.. They form level 1. b }* } . All possible configurations that are reachable by applying the transition function of TN once form the children of the initial configuration. For example let us consider the language L = { ww : w { a . Even in the same situation it may take different actions at different times. Here an action means the combination of writing a symbol on the tape. at any state it is in and for the tape symbol it is reading.S}. can not guess the midpoint of the string x. goes into an infinite loop or aborts. Proof : Let TN denote a nondeterministic Turing machine.

Turing machine accepting a+ Given the string aa. aa ) and ( q2 . it has two candidates for the next configuration: ( q1 . aa ) . aa ) for the third. aa ) ( q1 . The tree for this case would be as follows: . it would proceed as follows to accept it: ( q0 . aa ) for the second. aa ) and ( q2 . At the second and third configurations in the above sequence. aa ) ( q1 . aa ) ( q2 . and ( q1 . aa ) ( h .

is to traverse this tree breadth-first way from the root until the halt state is reached. Test Your Understanding of Different Types of Turing Machines Indicate which of the following statements are correct and which are not. These children are the configurations of the next level and they are stored on the tape (if necessary a second tape may be used). call it T1. then Submit. At each level of the tree. Next -.Unsolvable Problems Back to Study Schedule Back to Table of Contents Turing Machines Unsolvable Problems . then T2 accepts the string and halts. Many other variations of Turing machine are possible. If there is the halting state among these children.One way to simulate a nondeterministic Turing machine. Click True or Fals . call it T2. T2 applies the transition function of T1 to each configuration at that level and computes its children. with a deterministic one. In fact the Church's thesis conjectures that any so called computation done by humans or computers can be performed by a basic deterministic Turing machine. it has been shown that none of them exceed the capability of basic deterministic Turing machine as far as accepting languages is concerned. However. Thus any language accepted by a nondeterministic Turing machine is also accepted by a deterministic Turing machine. It can be easily seen that T2 accepts a string if and only if T1 accepts it. though a deterministic Turing machine might take much more time than a nondeterministic Turing machine to accept a string.

In this chapter we are going to learn that there are problems that can not be solved by Turing machines hence by computers. that is computationally they are equally powerful. Then given a string a Turing machine that accept the language starts the computation. Proof (by M. Theorem 1 : The halting problem is undecidable. Minsky): This is going to be proven by "proof by contradiction". So we are going to look at the unsolvability in terms of language recognition. L. . First recall that solving a problem can be viewed as recognizing a language (see Problem Solving as Language Recognition). Thus if a language is not decidable.Subjects to be Learned • • • Halting Problem Languages not Accepted by Turing Machines Other Unsolvable Problems Contents We have learned that deterministic Turing machines are capable of doing any computation that computers can do. the question is unanswerable that is the problem is unsolvable. Suppose that a language is acceptable but not decidable. and then T halts. b } . if the Turing machine is running. T writes "yes" if M halts on w and "no" if M does not halt on w. and that any of their variations do not exceed the computational power of deterministic Turing machines. hence unsolvable. It asks the following question: Given an arbitrary Turing machine M over alphabet = { a . given a description of a Turing machine M (over the alphabet ) and a string w. It is also conjectured that any "computation" human beings perform can be done by Turing machines (Church's thesis). the question of whether or not a string is in the language may not be answered in any finite amount of time. Halting Problem One of well known unsolvable problems is the halting problem. That is. Below we are going to see some well known unsolvable problems and see why we can say they are unsolvable. At any point in time. and an arbitrary string w over . Then there is a Turing machine T that solves the halting problem. there is no way of telling whether it is in an infinite loop or along the way to a solution and it needs more time. Since we can not wait forever for an answer. Suppose that the halting problem is decidable. does M halt when it is given w as an input ? It can be shown that the halting problem is not decidable. Here "unsolvability" is in the following sense.

where * is a symbol that separates the two copies of d(M) and then supplies d(M)*d(M) to the Turing machine Tm . denoted by d(M). then Tm goes into an infinite loop (Tm halts if the original T rejects a string and halts). . copies it to obtain the string d(M)*d(M). First we construct a Turing machine Tm by modifying T so that if T accepts a string and halts. Next using Tm we are going to construct another Turing machine Tc as follows: Tc takes as input a description of a Turing machine M.We are now going to construct the following new Turing machine Tc.

Thus Tc goes into an infinite loop if Tc halts on d(Tc) and it halts if Tc does not halt on d(Tc). This contradiction has been deduced from our assumption that there is a Turing machine that solves the halting problem. Thus one implication of the halting problem is that there can be no computer programs (Turing machines) that check whether or not any arbitrary computer program stops for a given input. When Tc gets the input d(Tc) . The way T was modified the modified T is going to go into an infinite loop if Tc halts on d(Tc) and halts if Tc does not halt on d(Tc).Let us now see what Tc does when a string describing Tc itself is given to it. . Thus the question of whether or not a program halts for a given input is nothing but the halting problem. Thus the modified T is given a description of Turing machine Tc and the string d(Tc). Program correctness and Halting Problem Note that for any computer program a Turing machine can be constructed that performs the task of the program. This is a contradiction. Hence there is no Turing machine that solves the halting problem. constructs the string d(Tc)*d(Tc) and gives it to the modified T. it makes a copy. Hence that assumption must be wrong.

More Unsolvable Preoblems Back to Study Schedule Back to Table of Contents Turing Machines More Unsolvable Problems Subjects to be Learned • • Languages not Accepted by Turing Machines Other Unsolvable Problems Contents The next unsolvable problem is in a sense more difficult than the halting problem. Language NonSelfAccepting Let us first define two languages NSA1 and NSA2 as follows: . Next -. then Submit. Click True or Fals . It is presented as a language and it can be shown that there are no Turing machines that accept the language.Test Your Understanding of Unsolvable Problems Indicate which of the following statements are correct and which are not.

Problem Accepts( ) The problem Accepts( ) asks whetehr or not a given Turing machine T accepts . Thus there can not be Turing machine T0 that accepts the language SelfAccepting . that is w0 is a description of the Turing machine T0 . then w0 is not in NonSelfAccepting because T0 accepts NonSelfAccepting. let w = a. Since NonSelfAccepting is a language. Certainly more symbols than a single a are needed to describe even the simplest Turing machine. Then there is no Turing machine that can be described by the string a. Suppose there is a Turing machine. NSA1 is the set of strings that describe a Turing machine but that are not accepted by the Turing machine they describe. However. b }*. Hence it is in SelfAccepting . call it T0. w0 is in NSA1 . It . For NSA2. Neither NSA1 nor NSA2 is empty. Hence T0 can not accept w0 . NSA2 is the set of strings that do not describe any Turing machine. then w0 NonSelfAccepting because T0 accepts NonSelfAccepting. Hence by the definition of NSA1 . This is again a contradiction. w d(T) for any Turing machine T } . This is a contradiction. For let T be a Turing machine that accepts { a } and let w = d(T). Knowing the unsolvability of the halting problem some other problems can be shown to be unsolvable. (1) If T0 accepts w0. Thus neither NSA1 nor NSA2 is empty. This means that there can not be any Turing machine that accepts the language NonSelfAccepting. where d(T) is a description of the Turing machine T. that accepts NonSelfAccepting. Theorem 2 There are no Turing machines that accept the language NonSelfAccepting. Let w0 = d( T0 ). w0 is in neither NSA1 nor NSA2 . Hence it is not accepted by T. (2) If T0 does not accept w0 . Also T0 does not accept w0 . which is absurd. Let us define the language NonSelfAccepting as NonSelfAccepting = NSA1 Then we can prove the following theorem: NSA2 . Hence a is in NSA2 . Hence w is in NSA1 . We are going to see that T0 neither accepts w0 nor rejects it. by the definitions of NSA1 and NSA2. w = d(T) for a Turing machine T and T does not accept NSA2 = { w | w { a. Hence w0 is not in NonSelfAccepting . Then this w is a description of a Turing machine but it must be longer than one symbol. either w0 is in NonSelfAccepting or it isn't. But w0 = d( T0 ) because that is how we selected w0 . b }*. Hence either T0 accepts w0 or rejects it. Proof: This is going to be proven by contradiction.NSA1 = { w | w w} { a.

M solves the halting problem. which is an instance of the halting problem. where Terase is a Turing machine that erases the input on the tape and halts. call it M. Then there is a Turing machine that solves it. This T halts on every string over if and only if T1 halts on . Since the halting problem is unsolvable. Using this T. Problem AcceptsEverything The problem AcceptsEverything asks whether or not a given Turing machine T halts on every string over a given alphabet . this means that Accepts( ) is unsolvable. We are going to show that the halting problem becomes solvable using this A. Let A be a Turing machine that solves Accepts( ). Then consider the Turing machine T = TeraseT' .can be shown to be unsolvable. Consider a Turing machine T = TwT'. call it M. Then M halts on d(T') and w if and only if T' halts on w. the halting problem can be solved. That is. that solves Accepts( ) can be constructed as . Suppose that AcceptsEverything is solvable. This T halts on if and only if T' halts on w. Let a Turing machine T' and a string w be an instance of the halting problem. Let T' be an instance of Accepts( ). Thus if Accepts( ) is solvable. that solves the halting problem can be constructed as follows: Given a description d(T') of a Turing machine T' and a string w as inputs. a Turing machine. We are going to show that Accepts( ) can be solved using the solution to it. Using a similar idea the following problem can also be shown to be unsolvable. Using this T. where machine Tw is a Turing machine that writes w. Suppose that Accepts( ) is solvable. Let A be a Turing machine that solves AcceptsEverything. M writes the string d( T ) on the tape and let A take over. a Turing machine.

By similar arguments the following problems can be shown to be unsolvable. Equivalence This problem asks whether or not two Turing machines accept the same language. AcceptsNothing This problem asks whether or not a Turing machine accepts nothing. Then the following problems are all unsolvable. It can be shown to be unsolvable using AcceptsEverything. it means that AcceptsEverything is unsolvable. Other Unsolvable Problems Let G1 and G2 be context-free grammars and let L(G) denote the language generated by grammar G.shown below. Is L( G1 ) L( G2 ) ? Is L( G1 ) L( G2 ) = ? finite ? infinite ? context-free ? Is L( G1 ) = L( G2 ) ? Is L( G1 ) = * ? Is the complement of L( G1 ) context-free ? . It can be shown to be unsolvable using Accepts( ) . Since Accepts( ) is unsolvable.

No one can write computer programs that solve those problems and halt after a finite amount of time. A problem is solvable if some Turing .Test Your Understanding of Unsolvable Problems Indicate which of the following statements are correct and which are not. then Submit. Next -. Click True or Fals .Time Complexity of Problem Back to Study Schedule Back to Table of Contents Complexity Time Complexity Subjects to be Learned • • • • • Time Complexity of Problems Decision Tree Class NP Polynomial Time Transformation NP-Complete Problems Contents In the previous sections we have learned that some problems are unsolvable by Turing machines hence by computers.

machine can solve it in finite time. They take much more time to execute than polynomial time algorithms. There are also problems that must be solved at best by exponential time algorithms in the worst case. as we are going to see below. in practice if it takes a million years. Similarly ( P V Q ) is also satisfiable. Then there are problems that require double exponential ( e. S). problems that need k-exponential time algorithms. the satisfiability problem for the propositional dynamic logic is proven to take exponential time to solve in the worst case. etc. X [ if ~( X = 1 ). So it is certainly satisfiable. after(A. For example the satisfiability problem for Presburger arithmetic is double-exponential (2-fold exponentail). They are all polynomial time algorithms. where A is an algorithm and S is a statement. Presburger arithmetic is a logic that allows statements involving positive integers. For example "after( if P then Q else ~Q. In logic there is a well known problem of "satisfiability". For many problems a day or even an hour would be too long. a quick sort needs O(n2) time. The satisfiability problem becomes even harder when logic becomes more complex. It is estimated by counting the largest possible number of key operations to be performed in terms of the input size. where P is a propositional variable. The problems that can not be solved with any polynomial time algorithm are called intractable problems . For example the formula ( P V ~P ) is always true. This is the problem of asking whether or not a given formula can take the value true for some values of its variables. The satisfiability problem for PDL is known to take at least exponential time to solve in the worst case. ~Q )" and "if P then after( if P then Q else ~Q. that is it requires at least O( aan ) time to solve in the worst case. the addition operation +. it is still solvable. Among the solvable problems there are problems that can be solved by algorithms with the worst case time which is a polynomial in the problem size (polynomial time algorithms). For example. etc. or etc. Even if it takes a million years to solve a problem.g. Before proceeding to predicate logic let us consider the following logic called propositional dynamic logic (PDL for short). For example a binary search takes O(lg n) time. O(f(x)) (big-oh) and other related subjects click here. then Y Z [ X = Y + Z ] ] is a proposition of Presburger arithmetic. 22n ) time algorithms. variables taking positive integers as their values. For example. as well as all the connectives such as and. the equality symbol = and quantifiers and . S) says that S is true after executing A. it is as good (or bad) as unsolvable. Q )" . For more detailed review of this. So it is not satisfiable. where P and Q are propositions. In this section we are going to study solvable problems and learn a hierarchy of solvable problems based on the computation time required to solve them. They are both satisfiable. a heap sort needs O(n lg n) time. and there are problems that require algorithms with the worst case time worse than k-exponential time for any natural number k. The measure for computation time we use is the worst case time. second order logic. However. . One can ask the same question for formulas of first order predicate logic. Let us see some of those intractable problems. are propositions of PDL. But (P ^ ~P ) is always false. This is a propositional logic with an extra construct (proposition) after(A. where k is a natural number.

" Is it possible to assign colors to vertices of a given graph using a given number of colors or less so that no two vertices connected directly by an edge have the same color assigned ? " etc. Some of these decision problems are NP-complete problems. there are problems that are solved by answering with yes or no. This problem is called "Graph Coloring" problem or more precisely "Vertex Color" problem. These problems are called decision problems. if. For example. First. Let us here review nondeterministic Turing machines. in addition. sets of integers and the predicate "belongs to" (an element X belongs to a set S) are allowed. At the moment. the consensus is that they ca not be solved with polynomial time algorithms. Let us try to solve the following instances of this graph coloring problem: Given the following graph. For the problems of this class there are no known polynomial time algorithms for solving them nor are they known to be unsolvable with polynomial time algorithms. Below we are going to characterize this class of problems. For the satisfiability problem of WS1S. This problem belongs to a peculiar class of problems called NP-Complete problems.In Presburger arithmetic (minus addition operation). however. "Is it possible to schedule committee meetings without conflicts into a given number of time slots ? " . there are no K-fold exponential time algorithms to solve it for any number K. "Is a string w in the language a*b ? ". is it possible to color its vertices with three or less colors ? . Consider the problem of coloring vertices of a graph with a given number of colors or less so that no two vertices connected directly by an edge have the same color assigned. the logic is called WS1S (Weak Second-order theory of 1 Successor). Now let us go back to the satisfiability problem of propositional logic. Such a problem (having no K-fold exponential time algorithms) is called nonelementary.

For the graphs of (a) and (b), you could find a solution very easily by inspection. You could see a right coloring as soon as you saw the graphs. However, you can most likely not tell how you arrived at your solutions. You probably don't have any algorithms you could use to solve them. You could somehow see the solutions. This is basically the idea of nondeterministic (Turing) machine. There is no fixed procedure which you can use repeatedly to solve instance after instance of this problem. But you can somehow solve them. Let us move on to a slightly more complex example of (c). For this graph to find a right coloring you could start with vertex 1 and assign color a. Then move on to vertex 2 and assign color b(it has to be something other than a ). Then go to vertex 3 and assign a third color, say c. Then at vertex 4 select color b and for vertex 5 use color a. In this process we make a decision as to what color to use for each vertex and when a decision is made for all the vertices we have a solution to the problem. This process applies to any decision problem. That is to solve a decision problem a number of smaller decisions are made one after another and as a result a solution to the problem is obtained. This process can be represented by a tree called decision tree. For example, for the graph coloring problem let us first decide on the order of vertices we color in, say 1, 2, 3, 4 and 5 for the graph of (c) above. Then the root of its decision tree corresponds to the vertex we assign a color to first (vertex 1 in this example). Then for each possible color for the first vertex, a child is created for the first vertex of the tree. So the second level of the

decision tree corresponds to the second vertex to be colored. Then in general, for each possible color for each vertex of level i of the decision tree, a child is created. Those children form level i+1 of the decision tree. The decision tree for the graph of (c) is given below. Since any color can be assigned to vertex 1 without loss of generality, it has just one child in the actual decision tree. Also since in this case the i-th and (i+1)-th vertices are connected by an edge for i = 1, 2, 3, 4, they can not have the same color. So each vertex after vertex 1 has two colors to choose from. So they each have two children in the decision tree.

Thus during the process of solving the problem a decision is made at each level and when all levels are covered, the problem is solved. A path from the root to a leaf corresponds to a coloring of the vertices of the given graph. A decision tree, however, does not tell us how to make decisions. Also a decision tree does not tell how to order the vertices for coloring, that is which vertex to color first, second etc. A deterministic machine (or algorithm) has a specific fixed set of rules for making a decision at each level of the decision tree. Although it knows what to do at every stage of problem solving, the decisions it makes are not necessarily the right ones. When it makes wrong decisions, it must retract earlier decisions and try different paths, which is called backtracking. For the graph coloring problem a deterministic algorithm might first order the vertices of the graph in decreasing order of their degree and also order colors. Then, following the

order of the vertices, assign to each vertex the highest order color available for the vertex. Since that kind of algorithm is not guaranteed to use the minimum number of colors, it may produce a wrong answer unless there is some provision for backtracking. A nondeterministic (Turing) machine, on the other hand, is a fictitious machine and somehow knows which branch (child) to select at each step. It always makes a right selection. A decision problem is said to belong to class NP if each vertex in its decision tree has a finite number of children and if it can be solved by a nondeterministic (Turing) machine in polynomial time. The graph coloring problem is in class NP, so are the satisfiability problem for propositional logic and most of the scheduling problems just to name a few. Also there are other characterizations of class NP. Interested readers click here. At this moment it is not known whether or not problems in class NP can be solved with a polynomial time algorithm in the worst case. The consensus is that there is no polynomial time algorithm to solve them. It would take at least exponential time. Among the problems in class NP, there are problems which all problems of class NP can be transformed to in polynomial time. Those problems are called NP-complete problems. If a polynomial time algorithm is found for any one of the NP-complete problems, all the problems in NP can be solved in polynomial time. Below we are going to study NPcomplete problems. We start our discussion with the concept of polynomial time transformation (reduction). Basically we say a decision problem Q1 is polynomially reducible to a decision problem Q2 if and only if there is a transformation that transforms any arbitrary instance of Q1 into an instance of Q2 in polynomial time such that the answer to Q1 is yes if and only if the answer to Q2 is yes. A little more formally we define this in terms of languages. Note that a decision problem can be viewed as a language of its instances and that solving it can be considered as recognizing the language as we have seen earlier. Let L1 and L2 be languages over alphabets 1 and 2, respectively. We say that L1 is polynomial-time reducible to L2 if and only if there is a function f from 1* to 2* such that for any string x in polynomial time.
1 *

,x

L1 if and only if f(x)

L2 and f can be computed

For example let us consider the following two problems: graph coloring and scheduling of committee meetings. The graph coloring problem is as given above. In the scheduling of committee meetings problem, committees with their members and a positive integer k are given. The problem is whether or not the meetings of the committees can be scheduled in k or less time slots so that everyone can attend one's meetings. Note that some people may be in more than one committee. Let us try to show that this scheduling problem is polynomial time reducible to the graph coloring problem.

and 1 and 4 share members. 2. Proceeding similarly the following graph is obtained corresponding to the committee memberships. Then since committees 1 and 2 share a. { b. 3 and 4 with the memberships { a. {a. 2. that is construct a graph and give the number of colors to be used to color its vertices so that the meetings can be scheduled if and only if graph can be colored. edges are added between 1 and 3. Suppose also that k = 3. connect with an edge the vertices corresponding to the committees. add vertices 1. Let us consider the following transformation: For each committee add a vertex to the graph. For example suppose that we are given the committees 1. d }. Thus the scheduling problem asks whether or not the meetings of the given committees can be scheduled in 3 time slots without any conflicts. b }. Then the meetings can be scheduled in k or less time slots if and only if the graph can be colored with k or less colors. an edge is inserted between vertices 1 and 2. The corresponding graph for the graph coloring problem can be constructed as follows: Corresponding to the committees 1. where p k. respectively.What we need to do is given an instance of the scheduling problem construct an instance of the graph coloring problem. c. Then the committees can be grouped into p groups so that the committees in the same group can meet at the same time. 2. c }. and 1 and 4. Corresponding to this grouping assign colors to the vertices of the graph so that the vertices in the same group are given the same color and those in . Suppose that the meetings can be scheduled in p time slots. Similarly since committees 1 and 3. 3 and 4. and if and only if two committee have some members in common. d } and { a. 3 and 4 to the graph.

4. a traveling salesman wants to know a shortest route to visit all cities exactly once and come back to where he/she started. NP-complete Problems 1. Some of them are listed below. It is also easily seen that the transformation.e. all the problems can be solved with polynomial time algorithms. then the consensus today is that it is most likely that no polynomial time algorithms i. fast algorithms exist to solve it. This coloring uses p colors which does not exceed k. Graph Color Problem 3. Partition Problem Given a set of integers. Thus these two vertices must get different colors. 7. 5. Conversely if the graph can be colored with k or less colors. then that means that the corresponding committees share some members and that they are scheduled to meet in different time slots. find out whether or not the objects can be put into the bins. 6. For if any two vertices are connected with an edge. then P is also NP-complete. Cook that the problems of class NP can be polynomial time reducible to the satisfiability problem of propositional logic. their sizes and a number of bins of the same size. It can be easily seen that if a problem P at hand is NP-hard and if a problem known to be NP-complete can be polynomial time reducible to P. Today hundreds of problems are known to be NP-complete. Satisfiability Problem for Propositional Logic 2. This group of problems are called NP-complete problems. then it can be easily seen that the committees can meet in k or less time slots. Committee Meeting Schedule Problem In fact most scheduling problems are NPcomplete. Subsequently the satisfiability problem was found to be polynomial time reducible to many other problems. A problem is NP-complete if it is in class NP and NP-hard. For all the problems in class NP can be reduced to P through the known NP-complete problem in polynomial time. Subgraph Isomorphism Problem Given two graphs. group them into two groups so that the sum of the numbers of one group is equal to that of the other group. As a consequence if a polynomial time algorithm is found for any one of those problems. that is the construction of graph for a given set of committees. Bin Packing Problem Given a set of objects. Formally a problem is NP-hard if every problem in class NP can be polynomial time reducible to it. We are now ready to discuss NP-completeness. which in this case can be taken as the number of committees. . If a problem is NP-complete. It was first proven by S. find out whether or not one is a subgraph of the other. can be done in time polynomial in the size of the problem. Traveling Salesman Problem Given cities and traveling times between cities. and vertices connected with an edge have different colors.different groups are given different colors.

b. Is there a subset T. and C = {x.y)}. 3-Dimensional Matching Given three sets A. their sizes. called a matching.2}. B.y). and S = {(1. (2.y). .a. a set of objects. Set Cover Problem Given a set S. B = {a.b. Note that {(1.(1.y}.x). (2.b}.b. B and C of the same size. Knapsack Problem Given a knapsack of size S.x).(2. their values and an integer V. of S such that every element of A.b.x)} is a desired set satisfying all the requirements. find out whether or not there are k or less subsets in the collection whose union is S. and C appears exactly once in T ? For example. is it possible to select objects so that the sum of their sizes does not exceed S and the sum of their values is V or larger ? 10.x). 9. and a subset S of the Cartesian product A X B X C. a collection of subsets of S and an integer k.8. Then T = {(1.a.b. (2.a.x)} is not a matching. let A = {1.