Properties of Context-Free

Languages
Two Pumping Lemmas

1
Topics

• Two Pumping Lemmas

• A Pumping Lemma for Context-Free Languages
• A Pumping Lemma for Linear Languages
• Closure Properties and Decision Algorithms for Context-Free Languages

2
Pumping Lemmas

• The pumping lemma for regular languages was given in Theorem 4.8.
• It is a tool for showing certain languages are not regular.
• There exist similar pumping lemmas for other languages.
• We will learn pumping lemmas for
• context-free languages, and
• linear languages, which are more restricted form of context-free languages.
• We can use them to show that certain languages are not context-free.
3
A Pumping Lemma for Context-Free Languages
Theorem 8.1

• Let L be an in nite context-free language.

• Then there exists some positive integer m s.t. any w ∈ L with | w | ≥ m can be
decomposed as

• w = uvxyz, (8.1) with

• | vxy | ≤ m, (8.2) and | vy | ≥ 1, (8.3) such that
i i
• uv xy z ∈ L, (8.4)
• for all i = 0,1,2,⋯.
• This is the pumping lemma for context-free languages.
4
fi
A Pumping Lemma for Context-Free Languages
Theorem 8.1

i i
• w = uvxyz (8.1) | vxy | ≤ m (8.2), | vy | ≥ 1 (8.3) uv xy z ∈ L (8.4) for all i = 0,1,2,⋯.
• If a given language L is a context-free language,
• every string w ∈ L satis es the above condition for some positive integer m.
• We don't know which m and uvxyz will satisfy the condition for w.
i i
• However, if m and uvxyz are satisfying the condition, then uv xy z for all i must be in L.
i i
• It means that for every m and uvxyz, if we show that uv xy z is not in L for a
certain i, then L is not context-free.

5
fi
A Pumping Lemma for Context-Free Languages
Theorem 8.1

• This can be also viewed as a game - to apply the pumping lemma.

• Of course, we have to play both sides.
1. The opponent picks m.

2. Given m, we pick a string w, | w | ≥ m.

3. The opponent chooses a decomposition uvxyz, satisfying (8.2) and (8.3).
i i
• We need to assume that the opponent makes it hardest to us to nd an i, which makes wi = uv xy z
is not in L.

• wi ∉ L for a certain decomposition does not mean that there's no decomposition for wi ∈ L.
4. We try to pick i which makes wi ∉ L. If we can, we win the game, and L is not context-free.

6 fi
A Pumping Lemma for Context-Free Languages
Example 8.1

n n n
• Show that the language L = {a b c : n ≥ 0} is not context-free.
m m m
• Once the opponent X picks m, we pick w = a b c .
• Now X has several choices for decompositions.
• Make vxy to contain only a's: wi will obviously not in L.
• v and y contain an equal number of a's and b's:
k k m
• then pumped strings in the form a b c , k ≠ m, which is not in L.
• The only way that X can win is to select a decomposition with vxy contain an equal number of a's, b's, and
c's, which is not possible due to | vxy | ≤ m.
• Therefore, L is not context-free.
7
A Pumping Lemma for Context-Free Languages
Example 8.1

n n
• The same argument will be resulted in failure for L = {a b }.
m m k k
• If we pick a b , the opponent can always choose v = a and y = b .
• Then regardless of i we pick, wi will always belong to L.
• Note that this doesn't prove that L is context-free.
• It's not like to show that L is not context-free.
• We only showed that there is an example satisfying the pumping lemma.
• To prove that L is context-free, then we should use other methods such as nd a context-
free grammar for L.

fi
A Pumping Lemma for Context-Free Languages
Example 8.1

̂ n n n 2n n n n
• We claimed that L = {a b } ∪ {a b } ∪ {a b c } is not context-free, in
Example 7.11.
m m m
• We already showed that we can win the game using a b c , which is also in
L .̂
• L
If ̂ is a context-free language, then the pumping lemma must hold for any
w ∈ L .̂
• Hence the claim was proved by our previous argument.

9
A Pumping Lemma for Context-Free Languages
Example 8.2

• Consider the language L = {ww : w ∈ {a, b}*}. Figure 8.2

R
• This looks very similar to ww , but it's not context-free.
m m m m
• Take a b a b .
• For any decomposition the opponent picks, we have a winning countermove.
• For instance, for the choice in Figure 8.2, we can use i = 0,
k j m m
• a b a b , k < m or j < m, which is not in L.
• For other choices of decompositions, we can argue similarly, hence L is not context-free.

10
A Pumping Lemma for Context-Free Languages
Example 8.3

n!
• Show that the language L = {a : n ≥ 0} is not context free.
m!
• For given m, we pick w = a .
k l
• Any decomposition for it always of the form v = a , y = a .
• Then w0 = uxz has length m! − (k + l). Then w0 ∈ L only if
• m! − (k + l) = j! for some j.
• This is impossible since k + l ≤ m, m! − (k + l) > (m − 1)!.
• Therefore, the language is not context free.
11
A Pumping Lemma for Context-Free Languages
Example 8.4

n j 2
• Show that the language L = {a b : n = j } is not context free.
2
m m
• For given m, we pick a b .
• Now the opponent has several choices.
• All a's, a's and b's, all b's. Figure 8.3

• We need to think about a's and b's with certain numbers, like Figure 8.3.
2
• Pumping i times, wi will be with m + (i − 1)k1 a's and m + (i − 1)k2 b's.

12
A Pumping Lemma for Context-Free Languages
Example 8.4

2
• Pumping i times, wi will be with m + (i − 1)k1 a's and m + (i − 1)k2 b's.
m 2−k1 m−k2
• If the opponent picks k1 ≠ 0, k2 ≠ 0, we can pick w0 = a b .
2 2 2 2
• (m − k2) ≤ (m − 1) = m − 2m + 1 < m − k1,
• the result is not in L.
• If the opponent picks k1 = 0, k2 ≠ 0 or k1 ≠ 0, k2 = 0,
• then with i = 0, the pumped string is not in L.
• Hence we can conclude that L is not a context-free language.
13
A Pumping Lemma for Context-Free Languages
Exercise 7 (d)

n j k
• Show that L = {a b c : k > n, k > j} on Σ = {a, b, c} is not context-free.
m m m+1
• For given m, we pick a b c which is in L.
k l
• The only possibility that the opponent wins the game is choosing v = b , y = c
for k ≠ 0, l ≠ 0.
m m+(i−1)k m+1+(i−1)l
• Then the pumped string wi = a b c .
m m−k m+1−l
• We choose i = 0, then w0 = a b c .

• Since l ≥ 1, m + 1 − l ≤ m, which means w0 ∉ L, and L is not context-free.

14
A Pumping Lemma for Linear
Languages

15
 Linear Languages
De nition 8.1

• A context-free language L is said to be linear if there exists a linear context-

free grammar G such that L = L(G).

• It means that every linear language is context free.

• We don't know yet if there exists some languages which are
• context free languages, but not linear languages.

16
fi
Linear Languages
Example 8.5

n n
• The language L = {a b : n ≥ 0} is a linear language.
• G = ({S}, {a, b}, S, P) with productions S → aSb | λ.
• A linear grammar is a grammar in which at most one variable can occur on the right side of any
production, without restriction on the position of the variable (Example 3.14).

• L1 = {w : na(w) = nb(w)}.
• We know that the grammar for L1 is S → SS | aSb | bSa | λ, which is not linear.
• Hence L1 is not necessarily linear, but giving one non-linear grammar for L1 is not enough to prove
that.

• It is possible that there exists a linear grammar for L1 which we have not found yet.
17
A Pumping Lemma for Linear Languages
Theorem 8.2

• Let L be an in nite linear language.

• Then there exists some positive integer m s.t. any w ∈ L with | w | ≥ m can be
decomposed as w = uvxyz with

• | uvyz | ≤ m, (8.5) and | vy | ≥ 1, (8.6) such that

i i
• uv xy z ∈ L, (8.7)
• for any i = 0,1,2,⋯.
• Note that (8.5) is di erent from (8.2). Now strings v and y must be located within m
symbols of the left or right ends of w.
18
fi
ff
A Pumping Lemma for Linear Languages
Visualization with a game

1. The opponent picks m.

2. Given m, we pick a string w, | w | ≥ m, to make the opponent di cult to choose a

decomposition.

• So far we've seen the cases that the choice of w is quite obvious.
3. The opponent chooses a decomposition uvxyz, satisfying (8.5) and (8.6).

• We need to assume that the opponent makes it hardest to us to nd an i, which

i i
makes wi = uv xy z is not in L.

4. We try to pick i which makes wi ∉ L. If we can, we win the game, and L is not linear.
19

ffi
fi
A Pumping Lemma for Linear Languages
Example 8.6

• The language L = {w : na(w) = nb(w)} is not linear.

• To show this, assume that L is linear and apply Theorem 8.2 to
m 2m m
• w=a b a .
• Inequality (8.5) shows that strings u, v, y, z must all consist of a's only in this case.
k l
• For v = a , y = a , k ≥ 1 or l ≥ 1,
m+k 2m m+l
• w2 = a b a is not in L.

• Hence L is not linear.

20
A Pumping Lemma for Linear Languages
Exercise 13

• Show that L = {w ∈ {a, b, c}* : na(w) + nb(w) = nc(w)} is context-free, but not linear.
m 2m m
• Given m, we pick w = a c b .
k l
• Whatever the decomposition is picked, v = a , y = b .
m+(i−1)k 2m m+(i−1)l
• Then wi = a c b ,1 ≤ k + l ≤ m.
m−k 2m m−l
• We pick i = 0, then w0 = a c b .

• Since k + l ≥ 1, m − k + m − l = 2m − (k + l) ≤ 2m.
• Hence w0 ∉ L, so L is not linear.
21
A Pumping Lemma for Linear Languages
Exercise 13

• L = {w ∈ {a, b, c}* : na(w) + nb(w) = nc(w)}.

• How can we prove that L is context-free?
• It seems to be complicated to nd a context-free grammar.
• We can consider an NPDA for this.
• Every time an input symbol is a or b,
• either push 0 to the stack (if the stack top is not 1), or pop 1 from the stack.
• Every time an input symbol is c,
• either push 1 to the stack (if the stack top is not 0), or pop 0 from the stack.
22
fi
A Pumping Lemma for Linear Languages
Exercise 14

n j
• Determine whether or not L = {a b : j ≤ n ≤ 2j − 1} is linear.
• If it looks linear, we can come up with a linear grammar.
• If it doesn't look like linear, we can apply the pumping lemma.
• We can add more a's, but that number cannot exceed the twice of b's.
• Make it twice: S → aaSb.
• We can also make the same number: S → aSb.
• In the end, we cannot have n = 2j, hence S → ab.
• S → aaSb | aSb | ab is a linear grammar for L, hence L is linear.
23
Closure Properties and Decision
Algorithms for Context-Free
Languages

24
Closure of Context-Free Languages
Theorem 8.3

• The family of context-free languages is closed under union, concatenation, and star-closure.
• Proof: Let L1 and L2 be two context-free languages generated by the context-free grammars
G1 = (V1, T1, S1, P1) and G2 = (V2, T2, S2, P2), respectively.

• We can assume that V1 and V2 are disjoint, w/o loss of generality.

• Consider L(G3) generated by G3 = (V1 ∪ V2 ∪ {S3}, T1 ∪ T2, S3, P3), where S3 ∉ V1 ∪ V2.
• P3, the production of G3 is P3 = P1 ∪ P2 ∪ {S3 → S1 | S2}.
• Then P3 is a context-free grammar, so that L(G3) is a context-free language.
• Since L(G3) = L1 ∪ L2, we can conclude that the family of context-free language is closed under
union.

25
Closure of Context-Free Languages
Theorem 8.3

• Now consider G4 = (V1 ∪ V2 ∪ {S4}, T1 ∪ T2, S4, P4).

• P4 = P1 ∪ P2 ∪ {S4 → S1S2}.
• Then L(G4) = L(G1)L(G2).
• Finally, L(G5) with G5 = (V1 ∪ {S5}, T1, S5, P5), and
• P5 = P1 ∪ {S5 → S1S5 | λ}.
• Then L(G5) = L(G1)*.
26
Closure of Context-Free Languages
Theorem 8.4

• The family of context-free languages is not closed under intersection and

complementation.

• Proof: Consider the two languages

n n m
• L1 = {a b c : n ≥ 0, m ≥ 0},
n m m
• L2 = {a b c : n ≥ 0, m ≥ 0}.
• L1 and L2 are context-free languages.

27
Closure of Context-Free Languages
Theorem 8.4

n n m n m m
• L1 = {a b c : n ≥ 0, m ≥ 0}, L2 = {a b c : n ≥ 0, m ≥ 0}.
• A context-free grammar for L1 is
• S → S1S2, S1 → aS1b | λ, S2 → cS2 | λ.
• Or, we can use the fact that L1 is a concatenation of two context-free
n n m
languages, say {a b } and {c }.

• Then due to Theorem 8.3, L1 is also context-free.

28
Closure of Context-Free Languages
Theorem 8.4

n n n
• Now, consider the intersection L1 ∩ L2 = {a b c : n ≥ 0}.
• We have already shown that this is not context-free.
• Thus the family of context-free languages is not closed under intersection.
• For complementation, we consider
C C C
• L1 ∩ L2 = (L1 ∪ L2 ) .
• If context-free languages are closed under complementation, the right side of the expression
would be a context-free language.

• This contradicts that L1 ∩ L2 is not context-free, hence the family of context-free languages are
not closed under complementation.

29
Closure of Context-Free Languages
Theorem 8.5

• Let L1 be a context-free language and L2 be a regular language.

• Then L1 ∩ L2 is context-free.
• Proof: Let M1 = (Q, Σ, Γ, δ1, q0, z, F1) be an NPDA that accepts L1 and
M2 = (P, Σ, δ2, p0, F2) be a DFA that accepts L2.

• We construct a PDA M ̂ = ( Q ,̂ Σ, Γ, δ ,̂ q0 ,̂ z, F )̂ that simulates the parallel action of

M1 and M2:

• Whenever a symbol is read from the input string, M ̂ simultaneously executes the
moves of M1 and M2.

30
Closure of Context-Free Languages
Theorem 8.5

• We let Q ̂ = Q × P, q0 ̂ = (q0, p0), F ̂ = F1 × F2,

• and de ne δ ̂ such that

• ((q ,
k lp ), x) ∈ δ ̂
((q ,
i jp ), a, b) , i .

• (qk, x) ∈ δ1(qi, a, b) and δ2(pj, a) = pl.

• We also require that if a = λ, then pj = pl, for the DFA.

31
fi
ff
Closure of Context-Free Languages
Theorem 8.5

• The states of M ̂ are labeled with pairs (qi, pj), representing the respective states in which M1 and M2 can
be after reading a certain input string.

• Using induction argument,

• ((q0, p0), w, z) ⊢*M ̂ ((qr, ps), λ, x),

• with qr ∈ F1, and ps ∈ F2 if and only if

• (q0, w, z) ⊢*M1 (qr, λ, x), and δ*(p0, w) = ps.

• Therefore, a string is accepted by M ̂ i . it is accepted by both M1 and M2, which means that
L(M1) ∩ L(M2) = L1 ∩ L2.
• This property is called closure under regular intersection.

32
ff
Closure of Context-Free Languages
Example 8.7

n n
• Show that the language L = {a b : n ≥ 0,n ≠ 100} is context-free.
• We can prove it more neatly and concisely with Theorem 8.5, compared to
construct a PDA or a context-free grammar.
100 100
• Let L1 = {a b }, which is nite, and thus it is also regular.
n n C
• It is easy to see that L = {a b : n ≥ 0} ∩ L1 .
C n n
• We know L1 is regular since L1 is regular, and {a b } is context-free.

• Therefore, due to closure under regular intersection, L is context-free.

33
fi
Closure of Context-Free Languages
Example 8.8

• Show that the language

• L = {w ∈ {a, b, c}* : na(w) = nb(w) = nc(w)} is not context-free.
• Suppose that L were context free.
n n n
• Then L ∩ L(a*b*c*) = {a b c : n ≥ 0} would be also context-free.
• We already know that this is not true.
• Therefore, we can conclude that L is not context-free.

34
Decidable Properties of Context-Free Languages
Theorem 8.6

• Given a context-free grammar G = (V, T, S, P), there exists an algorithm for

deciding whether or not L(G) is empty.

• Proof: Assume that λ ∉ L(G) for simplicity.

• We use the algorithm for removing useless variables and productions.
• If S is found to be useless, then L(G) is empty;
• Otherwise, L(G) contains at least one element.

35
Decidable Properties of Context-Free Languages
Theorem 8.7

• Given a context-free grammar G = (V, T, S, P), there exists an algorithm for

determining whether or not L(G) is in nite.

• Proof: Assume that G contains no λ-productions, no unit-productions, and no

useless symbols.

• Suppose the grammar has a repeating variable like this.

• There exists some A ∈ V for which there is a derivation A ⇒* xAy.
• Since G is assumed to have no λ- and unit-productions, x and y cannot be
simultaneously empty.
36
fi
Decidable Properties of Context-Free Languages
Theorem 8.7

• Since A is neither nullable nor a useless symbol, we have

• S ⇒* uAv ⇒ w and A ⇒* z,
• where u, v, and z are in T*.
• But then with A ⇒* xAy,
n n n n
• S ⇒* uAv ⇒* ux Ay v ⇒* ux zy v
• is possible for all n, so that L(G) is in nite.
• If no variable can ever repeat, then the length of any derivation is bounded by | V | .
• In this case, L(G) is nite.
37
fi
fi
 Decidable Properties of Context-Free Languages
Theorem 8.7

• For the algorithm, we only need to determine whether the grammar has some
repeating variables.

• We can do this by drawing dependency graph for variables, in a way that

• there is an edge (A, B) whenever there is a corresponding production
• A → xBy.
• Then any variable that is at the base of the cycle is a repeating one.
• Using this, we can have an algorithm for determining whether or not L(G) is
in nite.
38
fi
Decidable Properties of Context-Free Languages
Exercise 23

• Show that there exists an algorithm to determine if a context-free language contains

any even-length strings.

• Let L is the given language on Σ, and Le = {w ∈ Σ* : | w | is even}.

• Le is regular: we can easily think of an NFA.
• Using closure under regular intersection, L ∩ Le is context-free.
• Hence, by Theorem 8.6, there is an algorithm to decide whether or not L ∩ Le is empty.
• It means that we can decide whether L contains any even-length strings.

39
Summary

• You must understand

• what are the pumping lemmas for context-free/linear languages, and how to
apply them for a given language.

• Practice to apply the pumping lemmas for a given language.

• Also you have to consider construct a linear grammar to determine whether a
given language is linear or not.

• You also need to know closure properties of context-free languages.

• Practice to use closure properties and decision algorithms for other arguments.
40