Biçimsel Diller
ve
Otomata
Teorisi
Sunu X
Düzenli Olmayan
Diller
The Pumping Lemma
• Some languages can be described in English but they cannot
be defined by an FA
• Such as the language PALINDROME or PRIME (of all words ap, where p
is a prime number)
• A language that cannot be defined by a regular expression is
called a nonregular language
• By Kleene’s theorem, a nonregular language can also not be
accepted by any FA or TG
• All languages are either regular or nonregular; none are both
The Pumping Lemma
• Let us define the language L
• L = {Λ ab aabb aaabbb aaaabbbb aaaaabbbbb …}
• We could also define this language by the formula
• L = {anbn for n = 0 1 2 3 4 5 …}
• Or for short L = {anbn}
• It is a subset of many regular languages, such as a*b*
• Note that anbn does not include aab or bb
The Pumping Lemma
• Suppose on the contrary that this language were regular
• Then there would have to exist some FA that accepts it
• Let us picture one of these FAs (there might be several)
• This FA might have many states
• Let us say that it has 95 states
• Yet, we know it accepts the word a96b96
• The first 96 letters of this input string are all a’s and they trace a
path through this machine
The Pumping Lemma
• The path cannot visit a new state with each input letter
read
• Because there are only 95 states
• Therefore, at some point the path returns to a state that it has
already visited
• The first time it was in that state it left by the a-road
• The second time it is in that state it leaves by the a-road again
• Even if it only returns once, we say that the path contains a
circuit (A circuit is a loop that can contain several edges)
The Pumping Lemma
• First , the path wanders up to the circuit and then it starts
to loop around the circuit maybe many times
• It cannot leave the circuit until a b is read from the input
• Then the path can take a different turn
• For example, the path could make 30 loops around a three-state
circuit before the first b is read
• After the first b is read, the path goes off and does some
other stuff following b-edges and eventually winds up at a
final state where the word a96b96 is accepted
The Pumping Lemma
• Let us say that the circuit that the a-edge path loops
around has seven states in it
• The path enters the circuit
• Loops around it
• Then goes off on the b-line to a final state
• What would happen to the input string a96+7b96?
• The path loops around this circuit one more time (precisely one
extra time)
• That string is not in the language L = {anbn}
• This is a contradiction, in other words, L is nonregular
The Pumping Lemma
• Let us review what happened
• We choose a word in L that was so large (had so many letters)
that its path through the FA had to contain a circuit
• Once we found that some path with a circuit could reach a
final state, we ask..
• What happens to a path that is just like the first one, but that loops
around the circuit one extra time and then proceeds identically
through the machine
• The new path also leads to the same final state
• But it is generated by a different input string – an input string not in the
language L
The Pumping Lemma
• Let the path a9b9 be • The path for a13b9
The Pumping Lemma – Theorem
• Let L be any regular language that has infinitely many
words
• There exist some three strings x, y, and z (where y is not the
null string)
• xynz for n = 1 2 3 …
are words in L
The Pumping Lemma – Proof
• If L is a regular language, then there is an FA that accepts
exactly word in L
• The machine has only finitely many states
• L has infinitely many words in it (there are arbitrarily long words in L)
• Let w be some word in L that has more letters in it that there
are states in the machine
• When this word generates a path through the machine, the
path cannot visit a new state for each letter
• Because there are more letters than states
• It must at some point revisit a state that it has been to before
The Pumping Lemma – Proof
• Let us break the word w up into three parts
• Part 1 (x) All the letters of w starting at the beginning that
lead up to the first state that is revisited. x may be the null string if
the path for w revisits the start state as its first revisit
• Part 2 (y) Starting at the letter after x, y travels around the
circuit coming back to the same state the circuit began.
Because there must be a circuit, y cannot be null
• Part 3 (z) The rest of w. z could be null
• w = xyz
The Pumping Lemma
• What is the path through this machine of input string
• xyyz?
• xyyyz?
• xyyyyyyyyyyz?
• All these must be accepted by the mahine and therefore
are all in L
• L must contain all strings of the form
• xynz for n = 1 2 3 …
Example
• The machine below accepts an infinite language and has only six
states
• Any word with six or more letters must correspond to a path that
includes a circuit (also some words with fewer than six letters such as
baaa)
• We will consider in detail: w = bbbababa
Example
• w = bbbababa has more than six letters (includes a
circuit)
•w=b bba baba
x y z
What would happen to xyyz?
x y y z = b bba bba baba
Example
• The pumping lemma says that there must be strings x, y, and z
such that all words in the form xynz are in L
• Is this possible?
• A typical word of L
• aaa … aaaabbbb … bbb
• How do we break this into three pieces as x, y, and z?
• y is made entirely of a’s (xyyz, more a’s than b’s)
• y is made entirely of b’s (xyyz, more b’s than a’s)
• y contains some a’s and some b’s (xyyz would have two copies of
substring ab)
• xyyz cannot be a word in L, L is not regular
Example
• The language EQUAL, of all words with the same total number
of a’s and b’s is also nonregular
• EQUAL = {Λ ab ba aabb abab abba baab baba bbaa
aaabbb …}
• The language anbn is the intersection of all words defined by
the RE a*b* and the language EQUAL
• {anbn = a*b* ∩ EQUAL}
• If EQUAL were a regular language, then anbn would be the
intersection of two regular languages. Because anbn is not
regular, EQUAL cannot be
Example
• Consider the language anban = {b aba aabaa …}
• If this language were regular, there would exist three
strings x, y, and z such that
• xyx and xyyz were both words in this language
• We can show that this is impossible
Example
• Consider the language anban = {b aba aabaa …}
• Observations
1. If the y string contained b, then xyyz would contain two b’s (no word
in this language can have)
2. If the y string is all a’s then the b in the middle of the word xyz is in the
x-side or z-side. In either case, xyyz has increased the number of a’s
either in front of the b or after the b (but not both)
• Conclusions
1. xyyz does not have its b in the middle and is not in the form anban
2. This language cannot be pumped and is therefore not regular
Example
• Consider the language anbnabn+1 for n = 1, 2, 3, …
• Show that if xyz is in this language, then xyyz is not
• Observations
1. If we know the total number of a’s we can calculate the
number of b’s and vice versa. No two different words have the
same number of a’s or b’s
2. All words have exactly two substrings equal to ab and one
equal to ba
3. If xyz and xyyz are both in this language, then y cannot contain
either ab or ba because then xyyz would have too many
Example
• Consider the language anbnabn+1 for n = 1, 2, 3, …
• Conclusions
1. Because y cannot be Λ, it must contain either only a’s or b’s,
any mixture contains the forbidden substrings (observation 3)
2. If y is solid a’s, then xyz and xyyz are different words with the
same total b’s, violating observation 1. If y is solid b’s, then xyz
and xyyz are different words with the same number of a’s
violating observation 1
3. It is impossible for both xyz and xyyz to be in this language for
any strings x, y, and z. The language is unpumpable and not
regular
Theorem
• Let L be an infinite language accepted by a finite
automaton with N states. Then for all words w in L that
have more than N letters, there are strings x, y, and z,
where y is not null and length(x) + length(y) does not
exceed N such that
w = xyz
And all strings in the form xynz (for n = 1 2 3 …) are in L
Example
• We shall show that the language PALINDROME is nonregular
• We cannot use the first version of the pumping lemma because the
strings x=a, y=b, z=a satisfy the lemma and do not contradict the
language
• All words of the form xynz = abna are in PALINDROME
• Let us consider one of the FAs that might accept this language
• The machine has 77 states
• w = a80ba80
• Because it has more letters than the machine has states, we can break
w into the three parts: x, y, and z
Example
• Because the length of x and y must be in total 77 or less
• They must both be made of solid a’s (because the first 77 letters of w
are all a’s)
• When we form the word xyyz, we are adding more a’s to the front of w
(but we are not adding more a’s to the back of w because all the rear
a’s are in the z-part, which stays fixed at 80 a’s)
• The string xyyz is not a palindrome because it will be of the form amore
than 80ba80
• But the second version of the pumping lemma says that
PALINDROME has to include this string
• PALINDROME is nonregular
Example
• Let us consider the language PRIME = {ap where p is a prime}
• Is PRIME a regular language?
• If it is, then there is some FA that accepts exactly these words
• Let us suppose that it has 345 states
• Let us choose a prime number bigger than 345 – for example, 347
• a347 can be broken into parts x, y, and z such that xynz is in
PRIME for any value of n
• x, y, z are all just a’s
• Let us take n = 348, the word xy348z must be in PRIME
Example
• xy348z = xyzy347 (all a’s, order doesn’t matter)
• xyzy347 = a347y347 (x, y, and z came originally from a347
• We also know that y is some (nonempty) string of a’s
• Let us say that y = am
• a347y347 = a347(am)347
= a347+347m
= a347(m+1)
• Because m ≠ 0, we know that 347(m+1) is not a prime number
• This is a contradiction
• Therefore, PRIME is nonregular
