Recap so far: In DFA, the only memory available is in the states. Automata can only “remember” finitely far in the past and finitely much information, because they can have only finitely many states. If a computation path of a DFA visits the same state more than once, the machine can’t tell the difference between the first time and future times it visits this state. Thus, if a DFA accepts one long string, then it must accept (infinitely) many similar strings.
Definition A positive integer \(p\) is a pumping length of a language \(L\) over \(\Sigma\) means that, for each string \(s \in \Sigma^*\), if \(|s| \geq p\) and \(s \in L\), then there are strings \(x,y,z\) such that \[s = xyz\] and \[|y| > 0, \qquad \qquad \text{ for each $i \geq 0$, $xy^i z \in L$}, \qquad \text{and} \qquad \qquad |xy| \leq p.\]
Negation: A positive integer \(p\) is not a pumping length of a language \(L\) over \(\Sigma\) iff \[\exists s \left(~ |s| \geq p \wedge s \in L \wedge \forall x \forall y \forall z \left( ~\left( s = xyz \wedge |y| > 0 \wedge |xy| \leq p~ \right) \to \exists i ( i \geq 0 \wedge xy^iz \notin L ) \right) ~\right)\] Informally:
Restating Pumping Lemma: If \(L\) is a regular language, then it has a pumping length.
Contrapositive: If \(L\) has no pumping length, then it is nonregular.
The Pumping Lemma cannot be used to prove that a language is regular.
The Pumping Lemma can be used to prove that a language is not regular.
Extra practice: Exercise 1.49 in the book.
Proof strategy: To prove that a language \(L\) is not regular,
Consider an arbitrary positive integer \(p\)
Prove that \(p\) is not a pumping length for \(L\)
Conclude that \(L\) does not have any pumping length, and therefore it is not regular.
Example: \(\Sigma = \{0,1\}\), \(L = \{ 0^n 1^n \mid n \geq 0\}\).
Fix \(p\) an arbitrary positive integer. List strings that are in \(L\) and have length greater than or equal to \(p\):
Pick \(s =\)
Suppose \(s = xyz\) with \(|xy| \leq p\) and \(|y| > 0\).
Then when \(i = \hspace{1in}\), \(xy^i z = \hspace{1in}\)
Example: \(\Sigma = \{0,1\}\), \(L = \{w w^{\mathcal{R}} \mid w \in \{0,1\}^*\}\).
Fix \(p\) an arbitrary positive integer. List strings that are in \(L\) and have length greater than or equal to \(p\):
Pick \(s =\)
Suppose \(s = xyz\) with \(|xy| \leq p\) and \(|y| > 0\).
Then when \(i = \hspace{1in}\), \(xy^i z = \hspace{1in}\)
Example: \(\Sigma = \{0,1\}\), \(L = \{0^j1^k \mid j \geq k \geq 0\}\).
Fix \(p\) an arbitrary positive integer. List strings that are in \(L\) and have length greater than or equal to \(p\):
Pick \(s =\)
Suppose \(s = xyz\) with \(|xy| \leq p\) and \(|y| > 0\).
Then when \(i = \hspace{1in}\), \(xy^i z = \hspace{1in}\)
Example: \(\Sigma = \{0,1\}\), \(L = \{0^n1^m0^n \mid m,n \geq 0\}\).
Fix \(p\) an arbitrary positive integer. List strings that are in \(L\) and have length greater than or equal to \(p\):
Pick \(s =\)
Suppose \(s = xyz\) with \(|xy| \leq p\) and \(|y| > 0\).
Then when \(i = \hspace{1in}\), \(xy^i z = \hspace{1in}\)
Extra practice:
| Language | \(s \in L\) | \(s \notin L\) | Is the language regular or nonregular? |
|---|---|---|---|
| \(\{a^nb^n \mid 0 \leq n \leq 5 \}\) | |||
| \(\{b^n a^n \mid n \geq 2\}\) | |||
| \(\{a^m b^n \mid 0 \leq m\leq n\}\) | |||
| \(\{a^m b^n \mid m \geq n+3, n \geq 0\}\) | |||
| \(\{b^m a^n \mid m \geq 1, n \geq 3\}\) | |||
| \(\{ w \in \{a,b\}^* \mid w = w^\mathcal{R} \}\) | |||
| \(\{ ww^\mathcal{R} \mid w\in \{a,b\}^* \}\) | |||
Theorem: For an alphabet \(\Sigma\), For each language \(L\) over \(\Sigma\),
\(L\) is recognized by some
DFA
iff
\(L\) is recognized by some NFA
iff
\(L\) is described by some regular
expression
If (any, hence all) these conditions apply, \(L\) is called regular.
Prove or Disprove: There is some alphabet \(\Sigma\) for which there is some language recognized by an NFA but not by any DFA.
Prove or Disprove: There is some alphabet \(\Sigma\) for which there is some finite language not described by any regular expression over \(\Sigma\).
Prove or Disprove: If a language is recognized by an NFA then the complement of this language is not recognized by any DFA.
| Set | Cardinality |
|---|---|
| \(\{0,1\}\) | |
| \(\{0,1\}^*\) | |
| \(\mathcal{P}( \{0,1\})\) | |
| The set of all languages over \(\{0,1\}\) | |
| The set of all regular expressions over \(\{0,1\}\) | |
| The set of all regular languages over \(\{0,1\}\) | |
Pumping Lemma (Sipser Theorem 1.70): If \(A\) is a regular language, then there is a number \(p\) (a pumping length) where, if \(s\) is any string in \(A\) of length at least \(p\), then \(s\) may be divided into three pieces, \(s = xyz\) such that
\(|y| > 0\)
for each \(i \geq 0\), \(xy^i z \in A\)
\(|xy| \leq p\).
True or False: A pumping length for \(A = \{ 0,1 \}^*\) is \(p = 5\).
True or False: A pumping length for \(A = \{1, 01, 001, 0001, 00001 \}\) is \(p = 4\).
True or False: A pumping length for \(A = \{0^j 1 \mid j \geq 0 \}\) is \(p = 3\).
True or False: For any language \(A\), if \(p\) is a pumping length for \(A\) and \(p' > p\), then \(p'\) is also a pumping length for \(A\).