CS 208: Automata Theory and Logic

DFA Equivalence and Minimization

Ashutosh Trivedi

b a a

start A ∀x(La (x) → ∃y.(x < y) ∧ Lb (y))

B

b
Department of Computer Science and Engineering,
Indian Institute of Technology Bombay.

Ashutosh Trivedi – 1 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
DFA Equivalence and Minimization

1. Let A = (Q, Σ, δ, q0 , F) be a DFA.

2. Recall the definition of extended transition function δ̂.
3. Let L(A, q) be the languages {w : δ̂(q, w) ∈ F}
4. Recall the language of A is defined as L(A) = L(A, q0 ).
5. Two states q1 , q2 ∈ Q are equivalent if L(A, q1 ) = L(A, q2 ).
6. We say that two DFAs A1 and A2 are equivalent iff L(A1 ) = L(A2 ).

Theorem (DFA Equivalence)

For every DFA there exists a unique (up to state naming) minimal DFA.

Ashutosh Trivedi – 2 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
DFA Equivalence and Minimization

1. Let A = (Q, Σ, δ, q0 , F) be a DFA.

2. Recall the definition of extended transition function δ̂.
3. Let L(A, q) be the languages {w : δ̂(q, w) ∈ F}
4. Recall the language of A is defined as L(A) = L(A, q0 ).
5. Two states q1 , q2 ∈ Q are equivalent if L(A, q1 ) = L(A, q2 ).
6. We say that two DFAs A1 and A2 are equivalent iff L(A1 ) = L(A2 ).

Theorem (DFA Equivalence)

For every DFA there exists a unique (up to state naming) minimal DFA.

How to minimize DFAs?

Ashutosh Trivedi – 2 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
How to minimize a DFA?

Two observations:
– Removing unreachable states: removing states unreachable from the
start state does not change the language accepted by a DFA.

Ashutosh Trivedi – 3 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
How to minimize a DFA?

Two observations:
– Removing unreachable states: removing states unreachable from the
start state does not change the language accepted by a DFA.
– Merging equivalent states: merging equivalent states does not change
the language accepted by a DFA.

Ashutosh Trivedi – 3 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
How to minimize a DFA?

Two observations:
– Removing unreachable states: removing states unreachable from the
start state does not change the language accepted by a DFA.
– Merging equivalent states: merging equivalent states does not change
the language accepted by a DFA.
Algorithms:
1. Breadth-first search or depth-first search (to identify reachable states)
2. table-filling algorithm (by E. F. Moore) (other algorithms exist due to
Hopcroft and Brzozowski)

Ashutosh Trivedi – 3 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Table-filling algorithm
– Two states are distinguishable if they are not equivalent.
– Formally, two states q1 , q2 are distinguishable, if there exists a string
w ∈ Σ∗ such that exactly one of δ̂(q1 , w) and δ(q2 , w) is an accepting
state.
– Table-filling algorithm is recursive discovery of distinguishable pairs.

Ashutosh Trivedi – 4 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Table-filling algorithm
– Two states are distinguishable if they are not equivalent.
– Formally, two states q1 , q2 are distinguishable, if there exists a string
w ∈ Σ∗ such that exactly one of δ̂(q1 , w) and δ(q2 , w) is an accepting
state.
– Table-filling algorithm is recursive discovery of distinguishable pairs.
– Basis: Pair (p, q) is distinguishable if p ∈ F and q 6∈ F.

Ashutosh Trivedi – 4 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Table-filling algorithm
– Two states are distinguishable if they are not equivalent.
– Formally, two states q1 , q2 are distinguishable, if there exists a string
w ∈ Σ∗ such that exactly one of δ̂(q1 , w) and δ(q2 , w) is an accepting
state.
– Table-filling algorithm is recursive discovery of distinguishable pairs.
– Basis: Pair (p, q) is distinguishable if p ∈ F and q 6∈ F. why?

Ashutosh Trivedi – 4 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Table-filling algorithm
– Two states are distinguishable if they are not equivalent.
– Formally, two states q1 , q2 are distinguishable, if there exists a string
w ∈ Σ∗ such that exactly one of δ̂(q1 , w) and δ(q2 , w) is an accepting
state.
– Table-filling algorithm is recursive discovery of distinguishable pairs.
– Basis: Pair (p, q) is distinguishable if p ∈ F and q 6∈ F. why?
– Induction: Pair (p, q) is distinguishable if states δ(p, a) and δ(q, a) are
distinguishable for some a ∈ Σ.

Ashutosh Trivedi – 4 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Table-filling algorithm
– Two states are distinguishable if they are not equivalent.
– Formally, two states q1 , q2 are distinguishable, if there exists a string
w ∈ Σ∗ such that exactly one of δ̂(q1 , w) and δ(q2 , w) is an accepting
state.
– Table-filling algorithm is recursive discovery of distinguishable pairs.
– Basis: Pair (p, q) is distinguishable if p ∈ F and q 6∈ F. why?
– Induction: Pair (p, q) is distinguishable if states δ(p, a) and δ(q, a) are
distinguishable for some a ∈ Σ. why?

Ashutosh Trivedi – 4 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Table-filling algorithm
– Two states are distinguishable if they are not equivalent.
– Formally, two states q1 , q2 are distinguishable, if there exists a string
w ∈ Σ∗ such that exactly one of δ̂(q1 , w) and δ(q2 , w) is an accepting
state.
– Table-filling algorithm is recursive discovery of distinguishable pairs.
– Basis: Pair (p, q) is distinguishable if p ∈ F and q 6∈ F. why?
– Induction: Pair (p, q) is distinguishable if states δ(p, a) and δ(q, a) are
distinguishable for some a ∈ Σ. why?

TABLE -F ILLING A LGORITHM:

1. D ISTINGUISHABLE = {(p, q) : p ∈ F and q 6∈ F}.
2. Repeat while no new pair is added
2.1 for every a ∈ Σ

add (p, q) to D ISTINGUISHABLE if (δ(p, a), δ(q, a)) ∈ D ISTINGUISHABLE.

3. Return D ISTINGUISHABLE.
Ashutosh Trivedi – 4 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Correctness of Table-Filling Algorithm
Theorem
If two states are not distinguished by table-filling algorithm, then they are
equivalent.

Ashutosh Trivedi – 5 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Correctness of Table-Filling Algorithm
Theorem
If two states are not distinguished by table-filling algorithm, then they are
equivalent.

– The proof is by contradiction.

Ashutosh Trivedi – 5 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Correctness of Table-Filling Algorithm
Theorem
If two states are not distinguished by table-filling algorithm, then they are
equivalent.

– The proof is by contradiction.

– Assume that there is a pair (p, q) that is not distinguished by the
algorithm, but they are not equivalent, i.e. they are indeed
distinguishable (it is just that algorithm did not find them).
– Let us call such pair (p, q) a bad pair.

Ashutosh Trivedi – 5 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Correctness of Table-Filling Algorithm
Theorem
If two states are not distinguished by table-filling algorithm, then they are
equivalent.

– The proof is by contradiction.

– Assume that there is a pair (p, q) that is not distinguished by the
algorithm, but they are not equivalent, i.e. they are indeed
distinguishable (it is just that algorithm did not find them).
– Let us call such pair (p, q) a bad pair.
– There must be a string w ∈ Σ∗ that distinguishes a bad pair (p, q). Let
us take shortest such distinguishing string w among any bad pair, and
consider corresponding bad pair (p, q).

Ashutosh Trivedi – 5 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Correctness of Table-Filling Algorithm
Theorem
If two states are not distinguished by table-filling algorithm, then they are
equivalent.

– The proof is by contradiction.

– Assume that there is a pair (p, q) that is not distinguished by the
algorithm, but they are not equivalent, i.e. they are indeed
distinguishable (it is just that algorithm did not find them).
– Let us call such pair (p, q) a bad pair.
– There must be a string w ∈ Σ∗ that distinguishes a bad pair (p, q). Let
us take shortest such distinguishing string w among any bad pair, and
consider corresponding bad pair (p, q).
– Notice that w can not be  (Why?)

Ashutosh Trivedi – 5 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Correctness of Table-Filling Algorithm
Theorem
If two states are not distinguished by table-filling algorithm, then they are
equivalent.

– The proof is by contradiction.

– Assume that there is a pair (p, q) that is not distinguished by the
algorithm, but they are not equivalent, i.e. they are indeed
distinguishable (it is just that algorithm did not find them).
– Let us call such pair (p, q) a bad pair.
– There must be a string w ∈ Σ∗ that distinguishes a bad pair (p, q). Let
us take shortest such distinguishing string w among any bad pair, and
consider corresponding bad pair (p, q).
– Notice that w can not be  (Why?)
– Let w be of the form ax. Since p and q are distinguishable, we know that
exactly one of δ̂(p, ax) and δ̂(q, ax) is accepting.
– Then p0 = δ(p, a) and q0 = δ(q, a) are also distinguished by string x.

Ashutosh Trivedi – 5 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Correctness of Table-Filling Algorithm
Theorem
If two states are not distinguished by table-filling algorithm, then they are
equivalent.

– The proof is by contradiction.

– Assume that there is a pair (p, q) that is not distinguished by the
algorithm, but they are not equivalent, i.e. they are indeed
distinguishable (it is just that algorithm did not find them).
– Let us call such pair (p, q) a bad pair.
– There must be a string w ∈ Σ∗ that distinguishes a bad pair (p, q). Let
us take shortest such distinguishing string w among any bad pair, and
consider corresponding bad pair (p, q).
– Notice that w can not be  (Why?)
– Let w be of the form ax. Since p and q are distinguishable, we know that
exactly one of δ̂(p, ax) and δ̂(q, ax) is accepting.
– Then p0 = δ(p, a) and q0 = δ(q, a) are also distinguished by string x.
– if (p0 , q0 ) were discovered by table-filling algorithm and (p, q) must have
been discovered as well.

Ashutosh Trivedi – 5 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Correctness of Table-Filling Algorithm
Theorem
If two states are not distinguished by table-filling algorithm, then they are
equivalent.

– The proof is by contradiction.

– Assume that there is a pair (p, q) that is not distinguished by the
algorithm, but they are not equivalent, i.e. they are indeed
distinguishable (it is just that algorithm did not find them).
– Let us call such pair (p, q) a bad pair.
– There must be a string w ∈ Σ∗ that distinguishes a bad pair (p, q). Let
us take shortest such distinguishing string w among any bad pair, and
consider corresponding bad pair (p, q).
– Notice that w can not be  (Why?)
– Let w be of the form ax. Since p and q are distinguishable, we know that
exactly one of δ̂(p, ax) and δ̂(q, ax) is accepting.
– Then p0 = δ(p, a) and q0 = δ(q, a) are also distinguished by string x.
– if (p0 , q0 ) were discovered by table-filling algorithm and (p, q) must have
been discovered as well.
– If (p0 , q0 ) were not discovered by table-filling algorithm, then (p0 , q0 ) is a
bad pair with a shorter distinguishing string. Ashutosh Trivedi – 5 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Minimization of DFAs

– Let A be a DFA with no unreachable state.

– Let ≡A ⊆ Q × Q be the state equivalence relation (computed by, say
table-filling algorithm).
– Note that ≡A is an equivalence relation.
– Let us write [q] for the equivalence class of the state q.

Ashutosh Trivedi – 6 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Minimization of DFAs

– Let A be a DFA with no unreachable state.

– Let ≡A ⊆ Q × Q be the state equivalence relation (computed by, say
table-filling algorithm).
– Note that ≡A is an equivalence relation.
– Let us write [q] for the equivalence class of the state q.
– Given a DFA A and ≡A we can minimize A to the DFA
A≡ = (Q0 , Σ0 , δ 0 , q00 , F0 ), called Quotient Automata, where
– Q0 = {[q] : q ∈ Q},
– Σ0 = Σ,
– δ 0 ([q], a) = δ(q, a) for all a ∈ Σ,
– q00 = [q0 ], and
– F0 = {[q] : q ∈ F}.

Ashutosh Trivedi – 6 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Minimization of DFAs

– Let A be a DFA with no unreachable state.

– Let ≡A ⊆ Q × Q be the state equivalence relation (computed by, say
table-filling algorithm).
– Note that ≡A is an equivalence relation.
– Let us write [q] for the equivalence class of the state q.
– Given a DFA A and ≡A we can minimize A to the DFA
A≡ = (Q0 , Σ0 , δ 0 , q00 , F0 ), called Quotient Automata, where
– Q0 = {[q] : q ∈ Q},
– Σ0 = Σ,
– δ 0 ([q], a) = δ(q, a) for all a ∈ Σ,
– q00 = [q0 ], and
– F0 = {[q] : q ∈ F}.

Theorem
A≡ is the minimum and unique (up to state renaming) DFA equivalent to A.

Ashutosh Trivedi – 6 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Proof of Minimality
Theorem
A≡ is the minimum and unique (up to state renaming) DFA equivalent to A.

Proof.
– The proof is by contradiction.
– Assume that there is a DFA B whose size is smaller than A≡ and
accepts that same language.
– Compute equivalent states of A≡ and B using table-filling algorithm.
– The initial states of both DFAs must be equivalent. (Why?)
– After reading any string w from their initial states, both DFAs will go
to states that are equivalent. (Why?)
– For every state of A≡ there is an equivalent state in B.
– Since the number of states of B are less than that of A≡ , there must be
at least two states p, q of A≡ that are equivalent to some state of B.
– Hence p and q must be equivalent, a contradiction.

Ashutosh Trivedi – 7 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
DFA Equivalence and Minimization

Myhill-Nerode Theorem

Pumping Lemma

Ashutosh Trivedi – 8 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Myhill-Nerode Theorem

– Given a languages L, two strings u, v ∈ L are equivalent if for all

strings w ∈ Σ we have that u.w ∈ L iff v.w ∈ L.
– Let ≡L ⊆ Σ∗ × Σ∗ be such string-equivalence relation.
– Note that ≡L is an equivalence relation.
– Consider the equivalence classes of ≡L .
– When there are only finitely mane classes?

Ashutosh Trivedi – 9 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Myhill-Nerode Theorem

John Myhill Anil Nerode

Theorem (Myhill-Nerode Theorem)

A language L is regular if and only if there exists a string-equivalence relation ≡L
with finitely many classes.

Ashutosh Trivedi – 10 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Myhill-Nerode Theorem

John Myhill Anil Nerode

Theorem (Myhill-Nerode Theorem)

A language L is regular if and only if there exists a string-equivalence relation ≡L
with finitely many classes.
Moreover, the number of states in the minimum DFA accepting L is equal to the
number of equivalence classes in ≡L .
Ashutosh Trivedi – 10 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Myhill-Nerode Theorem

Theorem (Myhill-Nerode Theorem)

A language L is regular if and only if there exists a string-equivalence relation ≡L
with finitely many classes.

Proof.
The proof is in two parts.
– If L is regular, then a string-equivalence relation ≡L with finitely
many classes can be given by states of DFA accepting L.

Ashutosh Trivedi – 11 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Myhill-Nerode Theorem

Theorem (Myhill-Nerode Theorem)

A language L is regular if and only if there exists a string-equivalence relation ≡L
with finitely many classes.

Proof.
The proof is in two parts.
– If L is regular, then a string-equivalence relation ≡L with finitely
many classes can be given by states of DFA accepting L. How?

Ashutosh Trivedi – 11 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Myhill-Nerode Theorem

Theorem (Myhill-Nerode Theorem)

A language L is regular if and only if there exists a string-equivalence relation ≡L
with finitely many classes.

Proof.
The proof is in two parts.
– If L is regular, then a string-equivalence relation ≡L with finitely
many classes can be given by states of DFA accepting L. How?
– If there is a string-equivalence relation ≡L with finitely many classes,
one can find a DFA accepting L.

Ashutosh Trivedi – 11 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Myhill-Nerode Theorem

Theorem (Myhill-Nerode Theorem)

A language L is regular if and only if there exists a string-equivalence relation ≡L
with finitely many classes.

Proof.
The proof is in two parts.
– If L is regular, then a string-equivalence relation ≡L with finitely
many classes can be given by states of DFA accepting L. How?
– If there is a string-equivalence relation ≡L with finitely many classes,
one can find a DFA accepting L. How?

Ashutosh Trivedi – 11 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Applying Myhill-Nerode Theorem

Theorem (Myhill-Nerode Theorem)

A language L is regular if and only if there exists a string-equivalence relation ≡L
with finitely many classes.

Ashutosh Trivedi – 12 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Applying Myhill-Nerode Theorem

Theorem (Myhill-Nerode Theorem)

A language L is regular if and only if there exists a string-equivalence relation ≡L
with finitely many classes.

Equivalently,

A language L is nonregular if and only if there exists an infinite subset M of Σ∗

where any two elements of M are distinguishable with respect to L.

Ashutosh Trivedi – 12 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Applying Myhill-Nerode Theorem

Theorem
The language L = {0n 1n : n ≥ 0} is not regular.

Ashutosh Trivedi – 13 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Applying Myhill-Nerode Theorem

Theorem
The language L = {0n 1n : n ≥ 0} is not regular.

Proof.
1. The proof is by contradiction.
2. Assume that L is regular.

Ashutosh Trivedi – 13 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Applying Myhill-Nerode Theorem

Theorem
The language L = {0n 1n : n ≥ 0} is not regular.

Proof.
1. The proof is by contradiction.
2. Assume that L is regular.
3. By Myhill-Nerode theorem, there is a string-equivalence relation ≡L
over L with finitely equivalence classes.

Ashutosh Trivedi – 13 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Applying Myhill-Nerode Theorem

Theorem
The language L = {0n 1n : n ≥ 0} is not regular.

Proof.
1. The proof is by contradiction.
2. Assume that L is regular.
3. By Myhill-Nerode theorem, there is a string-equivalence relation ≡L
over L with finitely equivalence classes.
4. Let us consider the set of strings {0, 00, 000, . . . , 0i , . . .}.

Ashutosh Trivedi – 13 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Applying Myhill-Nerode Theorem

Theorem
The language L = {0n 1n : n ≥ 0} is not regular.

Proof.
1. The proof is by contradiction.
2. Assume that L is regular.
3. By Myhill-Nerode theorem, there is a string-equivalence relation ≡L
over L with finitely equivalence classes.
4. Let us consider the set of strings {0, 00, 000, . . . , 0i , . . .}.
5. It must be the case that some two string 0m and 0n , with m 6= n are
mapped to same equivalence class.

Ashutosh Trivedi – 13 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Applying Myhill-Nerode Theorem

Theorem
The language L = {0n 1n : n ≥ 0} is not regular.

Proof.
1. The proof is by contradiction.
2. Assume that L is regular.
3. By Myhill-Nerode theorem, there is a string-equivalence relation ≡L
over L with finitely equivalence classes.
4. Let us consider the set of strings {0, 00, 000, . . . , 0i , . . .}.
5. It must be the case that some two string 0m and 0n , with m 6= n are
mapped to same equivalence class.
6. It implies that for all strings w ∈ Σ∗ we have that 0m .w ∈ L iff 0n .w ∈ L.

Ashutosh Trivedi – 13 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Applying Myhill-Nerode Theorem

Theorem
The language L = {0n 1n : n ≥ 0} is not regular.

Proof.
1. The proof is by contradiction.
2. Assume that L is regular.
3. By Myhill-Nerode theorem, there is a string-equivalence relation ≡L
over L with finitely equivalence classes.
4. Let us consider the set of strings {0, 00, 000, . . . , 0i , . . .}.
5. It must be the case that some two string 0m and 0n , with m 6= n are
mapped to same equivalence class.
6. It implies that for all strings w ∈ Σ∗ we have that 0m .w ∈ L iff 0n .w ∈ L.
7. However, 0m 1m ∈ L but 0n 1m 6∈ L, a contradiction.
8. Hence L is not regular.

Ashutosh Trivedi – 13 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Applying Myhill-Nerode Theorem

Theorem
The language L = {0n 1n : n ≥ 0} is not regular.

Proof.

Ashutosh Trivedi – 14 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Applying Myhill-Nerode Theorem

Theorem
The language L = {0n 1n : n ≥ 0} is not regular.

Proof.
1. From Myhill-Nerode theorem, a language L is nonregular if and only
if there exists an infinite subset M of Σ∗ where any two elements of M
are distinguishable with respect to L.

Ashutosh Trivedi – 14 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Applying Myhill-Nerode Theorem

Theorem
The language L = {0n 1n : n ≥ 0} is not regular.

Proof.
1. From Myhill-Nerode theorem, a language L is nonregular if and only
if there exists an infinite subset M of Σ∗ where any two elements of M
are distinguishable with respect to L.
2. Consider the set M = {0i : i ≥ 0}.

Ashutosh Trivedi – 14 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Applying Myhill-Nerode Theorem

Theorem
The language L = {0n 1n : n ≥ 0} is not regular.

Proof.
1. From Myhill-Nerode theorem, a language L is nonregular if and only
if there exists an infinite subset M of Σ∗ where any two elements of M
are distinguishable with respect to L.
2. Consider the set M = {0i : i ≥ 0}.
3. Since any two string in M are distinguishable with respect to L (i.e.
0m 0n ∈ L but 0n 1m 6∈ L for n 6= m), it follows from Myhill-Nerode
theorem that L is a non-regular language.

Ashutosh Trivedi – 14 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Some languages are not regular!

The following languages are regular or non-regular?

– The language {0n 1n : n ≥ 0}
– The set of strings having an equal number of 0’s and 1’s
– The set of strings with an equal number of occurrences of 01 and 10.
– The language {ww : w ∈ {0, 1}∗ }
– The language {ww : w ∈ {0, 1}∗ }
– The language {0i 1j : i > j}
– The language {0i 1j : i ≤ j}
– The language of palindromes of {0, 1}

Ashutosh Trivedi – 15 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Some languages are not regular!

The following languages are regular or non-regular?

– The language {0n 1n : n ≥ 0}
– The set of strings having an equal number of 0’s and 1’s
– The set of strings with an equal number of occurrences of 01 and 10.
– The language {ww : w ∈ {0, 1}∗ }
– The language {ww : w ∈ {0, 1}∗ }
– The language {0i 1j : i > j}
– The language {0i 1j : i ≤ j}
– The language of palindromes of {0, 1}

Ashutosh Trivedi – 15 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
DFA Equivalence and Minimization

Myhill-Nerode Theorem

Pumping Lemma

Ashutosh Trivedi – 16 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Pumping Lemma

Theorem (Pumping Lemma for Regular Languages)

For every regular language L there exists a constant p (that depends on L)

Ashutosh Trivedi – 17 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Pumping Lemma

Theorem (Pumping Lemma for Regular Languages)

For every regular language L there exists a constant p (that depends on L)
such that
for every string w ∈ L of length greater than p,
there exists an infinite family of strings belonging to L.

Ashutosh Trivedi – 17 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Pumping Lemma

Theorem (Pumping Lemma for Regular Languages)

For every regular language L there exists a constant p (that depends on L)
such that
for every string w ∈ L of length greater than p,
there exists an infinite family of strings belonging to L.

Why?

Ashutosh Trivedi – 17 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Pumping Lemma

Theorem (Pumping Lemma for Regular Languages)

For every regular language L there exists a constant p (that depends on L)
such that
for every string w ∈ L of length greater than p,
there exists an infinite family of strings belonging to L.

Why? Think: Regular expressions, DFAs

Ashutosh Trivedi – 17 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Pumping Lemma

Theorem (Pumping Lemma for Regular Languages)

For every regular language L there exists a constant p (that depends on L)
such that
for every string w ∈ L of length greater than p,
there exists an infinite family of strings belonging to L.

Why? Think: Regular expressions, DFAs Formalize our intuition!

Ashutosh Trivedi – 17 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Pumping Lemma

Theorem (Pumping Lemma for Regular Languages)

For every regular language L there exists a constant p (that depends on L)
such that
for every string w ∈ L of length greater than p,
there exists an infinite family of strings belonging to L.

Why? Think: Regular expressions, DFAs Formalize our intuition!

If L is a regular language, then

there exists a constant (pumping length) p such that
for every string w ∈ L s.t. |w| ≥ p
there exists a division of w in strings x, y, and z s.t. w = xyz such that
1. |y| > 0,
2. |xy| ≤ p, and
3. for all i ≥ 0 we have that xyi z ∈ L.

Ashutosh Trivedi – 17 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
A simple observation about DFA
0 1
1

start E O

computation string computation string

start E . start E .

0 0

E . E .

1 0

O . E .

0 1

E . O .
Ashutosh Trivedi – 18 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
A simple observation about DFA

Image source: Wikipedia

– Let A = (S, Σ, δ, s0 , F) be a DFA.

– For every string w ∈ Σ∗ of the length greater than or equal to the
number of states of A, i.e. |w| ≥ |S|, we have that
– the unique computation of A on w re-visits at least one state after
reading first |S| letters !
Ashutosh Trivedi – 19 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Pumping Lemma
Theorem (Pumping Lemma for Regular Languages)
If L is a regular language, then there exists a constant (pumping length) p such
that for every string w ∈ L s.t. |w| ≥ p there exists a division of w in strings x, y,
and z s.t. w = xyz such that
1. |y| > 0,
2. |xy| ≤ p, and
3. for all i ≥ 0 we have that xyi z ∈ L.

Ashutosh Trivedi – 20 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Pumping Lemma
Theorem (Pumping Lemma for Regular Languages)
If L is a regular language, then there exists a constant (pumping length) p such
that for every string w ∈ L s.t. |w| ≥ p there exists a division of w in strings x, y,
and z s.t. w = xyz such that
1. |y| > 0,
2. |xy| ≤ p, and
3. for all i ≥ 0 we have that xyi z ∈ L.

– Let A be the DFA accepting L and p be the set of states in A.

– Let w = (a1 a2 . . . ak ) ∈ L be any string of length ≥ p.
– Let s0 a1 s1 a2 s2 . . . ak sk be the run of w on A.
– Consider first n + 1 states—at least one state must occur twice.
– Let i be the index of first state that the run revisits and let j be the
index of second occurrence of that state, i.e. si = sj ,
– Let x = a1 a2 . . . ai−1 and y = ai ai+1 . . . aj−1 , and z = aj aj+1 . . . ak .
– notice that |y| > 0 and |xy| ≤ n
– Also, notice that for all i ≥ 0 the string xyi z is also in L.
Ashutosh Trivedi – 20 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Applying Pumping Lemma

Theorem (Pumping Lemma for Regular Languages)

L ∈ Σ∗ is a regular language
=⇒
there exists p ≥ 1 such that
for all strings w ∈ L with |w| ≥ p we have that
there exists x, y, z ∈ Σ∗ with w = xyz, |y| > 0, |xy| ≤ p such that
for all i ≥ 0 we have that
xyi z ∈ L.

Ashutosh Trivedi – 21 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Applying Pumping Lemma

Theorem (Pumping Lemma for Regular Languages)

L ∈ Σ∗ is a regular language
=⇒
there exists p ≥ 1 such that
for all strings w ∈ L with |w| ≥ p we have that
there exists x, y, z ∈ Σ∗ with w = xyz, |y| > 0, |xy| ≤ p such that
for all i ≥ 0 we have that
xyi z ∈ L.

Pumping Lemma (Contrapositive)

For all p ≥ 1 we have that
there exists a string w ∈ L with |w| ≥ p such that
for all x, y, z ∈ Σ∗ with w = xyz, |y| > 0, |xy| ≤ p we have that
there exists i ≥ 0 such that
xyi z 6∈ L
=⇒
L ∈ Σ∗ is not a regular language.
Ashutosh Trivedi – 21 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Applying Pumping Lemma

How to show that a language L is non-regular.

1. Let p be an arbitrary number (pumping length).
2. (Cleverly) Find a representative string w of L of size ≥ p.
3. Try out all ways to break the string into xyz triplet satisfying that
|y| > 0 and |xy| ≤ n. If the step 3 was clever enough, there will be
finitely many cases to consider.
4. For every triplet show that for some i the string xyi z is not in L, and
hence it yields contradiction with pumping lemma.

Ashutosh Trivedi – 22 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Applying Pumping Lemma

Theorem
Prove that the language L = {0n 1n } is not regular.

Ashutosh Trivedi – 23 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Applying Pumping Lemma

Theorem
Prove that the language L = {0n 1n } is not regular.

Proof.
1. State the contrapositive of Pumping lemma.
2. Let p be an arbitrary number.
3. Consider the string 0p 1p ∈ L. Notice that |0p 1p | ≥ p.

Ashutosh Trivedi – 23 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Applying Pumping Lemma

Theorem
Prove that the language L = {0n 1n } is not regular.

Proof.
1. State the contrapositive of Pumping lemma.
2. Let p be an arbitrary number.
3. Consider the string 0p 1p ∈ L. Notice that |0p 1p | ≥ p.
4. Only way to break this string in xyz triplets such that |xy| ≤ p and
y 6= ε is to choose y = 0k for some 1 ≤ k ≤ p.

Ashutosh Trivedi – 23 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Applying Pumping Lemma

Theorem
Prove that the language L = {0n 1n } is not regular.

Proof.
1. State the contrapositive of Pumping lemma.
2. Let p be an arbitrary number.
3. Consider the string 0p 1p ∈ L. Notice that |0p 1p | ≥ p.
4. Only way to break this string in xyz triplets such that |xy| ≤ p and
y 6= ε is to choose y = 0k for some 1 ≤ k ≤ p.
5. For each such triplet, there exists an i (say i = 0) such that xyi z 6∈ L.
6. Hence L is non-regular.

Ashutosh Trivedi – 23 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Proving a language Regular

Proving Regularity
Pumping Lemma is necessary but not sufficient condition for regularity.

Ashutosh Trivedi – 24 of 24
Ashutosh Trivedi DFA Equivalence and Minimization
Proving a language Regular

Proving Regularity
Pumping Lemma is necessary but not sufficient condition for regularity.

Consider the language

L = {#an bn : n ≥ 1} ∪ {#k w : k 6= 1, w ∈ {a, b}∗ }.

Verify that this language satisfies the pumping condition, but is not
regular!

Ashutosh Trivedi – 24 of 24
Ashutosh Trivedi DFA Equivalence and Minimization