Definition 1.

A PDA A = (Q, Σ, Γ, ∆, q0 , F ) is said to be simple if

((p, x, α), (q, β)) ∈ ∆ =⇒ |α| ≤ 1, (1)

((p, x, ε), (q, β)) ∈ ∆ =⇒ ∀A ∈ Γ, ((p, x, A), (q, βA)) ∈ ∆. (2)

Remark 2. For every PDA there exists an equivalent simple PDA.

Theorem 3. For every PDA there exists an equivalent CFG.

Proof (Construction Part). With out loss of generality, let A = (Q, Σ, Γ, ∆, q0 , F ) be a

simple PDA. Construct a CFG G = (N, T = Σ, P, S) described as follows:
Set N = {S} ∪ {hp, A, qi | p, q ∈ Q; A ∈ Γ ∪ {ε}} and include the following four types of
production rules in P .

(a) For each qf ∈ F , the rule S −→ hq0 , ε, qf i.

(b) For each ((p, x, A), (q, X1 X2 · · · Xn )) ∈ ∆ with n > 0 and for all r, q1 , . . . , qn−1 ∈ Q,
the rule
hp, A, ri −→ xhq, X1 , q1 ihq1 , X2 , q2 i · · · hqn−1 , Xn , ri.

(c) For each ((p, x, A), (q, ε)) ∈ ∆ and for all r ∈ Q, the rule

hp, A, ri −→ xhq, ε, ri.

(d) For each q ∈ Q, the rule hq, ε, qi −→ ε.

A sample construction. Consider the PDA A = (Q, Σ, Γ, ∆, p, F ) which accepts the

language {0n 1n | n ≥ 0}, where Q = {p, q}, Σ = Γ = {0, 1}, F = {q} and ∆ has the
following transitions:

((p, 0, ε), (p, 0)) (3)

((p, ε, ε), (q, ε)) (4)
((q, 1, 0), (q, ε)) (5)

Step-I: Conversion to Simple PDA. All the transitions are satisfying the condition in (1)
but not of (2). So we include the following transitions in ∆.

((p, 0, 0), (p, 00)) (6)

((p, 0, 1), (p, 01)) (7)
((p, ε, 0), (q, 0)) (8)
((p, ε, 1), (q, 1)) (9)

Thus we have seven transitions (3) through (9) in ∆; and A is simple.

Step-II: ∗ Construction of CFG from A . We have the following production rules in the
required CFG.
Type (a):
[as q is only final state] S −→ hp, ε, qi (10)

Type(b):
[for (3) with r = p] hp, ε, pi −→ 0hp, 0, pi (11)

∗
One may remove 1 from the Γ as we are not using it in stack, i.e. Γ = {0}. That minimizes the number
of production rules of CFG. In fact, the transitions (7) and (9) can be dropped from ∆. Accordingly, we
will not have the production rules: (17), (18), (19), (20), (23) and (24)
 [for (3) with r = q] hp, ε, qi −→ 0hp, 0, qi (12)

[for (6) with r = p and 0p0] hp, 0, pi −→ 0hp, 0, pihp, 0, pi (13)

[for (6) with r = p and 0q0] hp, 0, pi −→ 0hp, 0, qihq, 0, pi (14)

[for (6) with r = q and 0p0] hp, 0, qi −→ 0hp, 0, pihp, 0, qi (15)

[for (6) with r = q and 0q0] hp, 0, qi −→ 0hp, 0, qihq, 0, qi (16)

[for (7) with r = p and 0p1] hp, 1, pi −→ 0hp, 0, pihp, 1, pi (17)

[for (7) with r = p and 0q1] hp, 1, pi −→ 0hp, 0, qihq, 1, pi (18)

[for (7) with r = q and 0p1] hp, 1, qi −→ 0hp, 0, pihp, 1, qi (19)

[for (7) with r = q and 0q1] hp, 1, qi −→ 0hp, 0, qihq, 1, qi (20)

[for (8) with r = p] hp, 0, pi −→ hq, 0, pi (21)

[for (8) with r = q] hp, 0, qi −→ hq, 0, qi (22)

[for (9) with r = p] hp, 1, pi −→ hq, 1, pi (23)

[for (9) with r = q] hp, 1, qi −→ hq, 1, qi (24)

Type(c):
[for (4) with r = p] hp, ε, pi −→ hq, ε, pi (25)

[for (4) with r = q] hp, ε, qi −→ hq, ε, qi (26)

[for (5) with r = p] hq, 0, pi −→ 1hq, ε, pi (27)

[for (5) with r = q] hq, 0, qi −→ 1hq, ε, qi (28)

Type(d):
[for state p] hp, ε, pi −→ ε (29)

[for state q] hq, ε, qi −→ ε (30)

Corresponding to the computation

(p, 0011, ε) ` (p, 011, 0) ` (p, 11, 00) ` (q, 11, 00) ` (q, 1, 0) ` (q, ε, ε)

for 0011 in A , we observe the following derivation for 0011 in G.

S =⇒ hp, ε, qi (by (10))

=⇒ 0hp, 0, qi (by (12))
=⇒ 00hp, 0, qihq, 0, qi (by (16))
=⇒ 00hq, 0, qihq, 0, qi (by (22))
=⇒ 001hq, ε, qihq, 0, qi (by (28))
=⇒ 001hq, 0, qi (by (30))
=⇒ 0011hq, ε, qi (by (28))
=⇒ 0011 (by (30))

2
Proof Continued. For p, q ∈ Q, A ∈ Γ ∪ {ε} and x ∈ Σ∗ , we claim that
∗ ∗
hp, A, qi =⇒ x if and only if (p, x, A) ` (q, ε, ε).
∗
Since, x ∈ L(G) ⇐⇒ x ∈ L(A ) is equivalent to hq0 , A, qf i =⇒ x if and only if
∗
(q0 , x, A) ` (qf , ε, ε), for some qf ∈ F , the above claim suffices.

∗ ∗
⇒: Assume hp, A, qi =⇒ x. We prove that (p, x, A) ` (q, ε, ε) by induction on the length of
the derivation.
∗
Basis: If the derivation hp, A, qi =⇒ x has only one step then hp, A, qi −→ x is a rule of
∗
type (d). Then A = x = ε and p = q so that (p, x, A) = (q, ε, ε). Hence, (p, x, A) ` (q, ε, ε).

∗ ∗
Inductive hypothesis: If a derivation hp, A, qi =⇒ x has k or fewer steps then (p, x, A) `
(q, ε, ε).
k+1 k
Induction: Assume hp, A, qi =⇒ x and write hp, A, qi =⇒ α =⇒ x. Then the rule
hp, A, qi −→ α is of type (b) or (c) (refer construction part).

Case-I: hp, A, qi −→ α is of type (b). Then, for n ≥ 1, write

α = yhp0 , X1 , p1 i · · · hpn−1 , Xn , pn = qi
∗
so that ((p, y, A), (p0 , X1 · · · Xn )) ∈ ∆. Let x = yy1 · · · yn such that hpi−1 , Xi , pi i =⇒ yi , for
1 ≤ i ≤ n. Thus for each i these derivations are of k or fewer steps so that
∗
(pi−1 , yi , Xi ) ` (pi , ε, ε).

Combining these computations

(p, x, A) = (p, yy1 · · · yn , A) ` (p0 , y1 · · · yn , X1 · · · Xn )

∗
` (p1 , y2 · · · yn , X2 · · · Xn )
..
.
∗ ∗
` (pn−1 , yn , Xn ) ` (pn = q, ε, ε).

Case-II: hp, A, qi −→ α is of type (c). Then, α = yhr, ε, qi and ((p, y, A), (r, ε)) ∈ ∆,
where x = yz for some z ∈ Σ∗ , so that
∗
(p, x, A) = (p, yz, A) ` (r, z, ε) ` (q, ε, ε).
∗ ∗
⇐: Assume (p, x, A) ` (q, ε, ε). We prove that hp, A, qi =⇒ x by induction on the length of
the computation.
∗
Basis: If a computation (p, x, A) ` (q, ε, ε) has zero steps, then p = q and x = ε = A. Then
∗
the conclusion hp, A, qi =⇒ x holds trivially as hp, ε, pi −→ ε is a type (d) production rule.
∗ ∗
Inductive hypothesis: If a computation (p, x, A) ` (q, ε, ε) has k or fewer steps then hp, A, qi =⇒
x.
k+1
Induction: Suppose (p, x, A) ` (q, ε, ε) and write, for n ≥ 0,
k
(p, x, A) ` (r, z, X1 · · · Xn ) ` (q, ε, ε).

3
Then there is y ∈ Σ∗ such that x = yz and (p, y, A) ` (r, ε, X1 · · · Xn ). This implies that

((p, y, A), (r, X1 · · · Xn )) ∈ ∆

or
n ≥ 1, A = Bn and ((p, y, ε), (r, X1 · · · Xn−1 )) ∈ ∆.
Since A is a simple the latter case implies the former so that ((p, y, A), (r, X1 · · · Xn )) ∈ ∆,
in any case.
Now, if n ≥ 1, each Xi is eventually removed from the stack via a sequence of states
say r = p0 , p1 , . . . , pn−1 , pn = q by consuming the corresponding input say z1 , . . . , zn , where
z = z1 · · · zn . That is,

(p, x, A) = (p, yz, A) ` (r, z, X1 · · · Xn )

= (r, z1 · · · zn , X1 · · · Xn )
∗
` (q1 , z2 · · · zn , X2 · · · Xn )
..
.
∗
` (qi , zi+1 · · · zn , Xi+1 · · · Xn )
∗
` (qi+1 , zi+2 · · · zn , Xi+2 · · · Xn )
..
.
∗
` (q, ε, ε).

Notice each computation (in each iteration) has k or fewer steps. Then, for 0 ≤ i ≤ n, by
∗
inductive hypothesis hpi , Xi+1 , pi+1 i =⇒ zi+1 . Also, since ((p, y, A), (r, X1 · · · Xn )) ∈ ∆, we
have
hp, A, qi −→ yhr = p0 , X1 , p1 ihp1 , X2 , p2 i · · · hpn−1 , Xn , pn = qi
is a production rule of type (b). Thus by combining these we have

hp, A, qi =⇒ yhr = p0 , X1 , p1 ihp1 , X2 , p2 i · · · hpn−1 , Xn , pn = qi

∗
=⇒ yz1 hp1 , X2 , p2 i · · · hpn−1 , Xn , pn = qi
..
.
∗
=⇒ yz1 · · · zn = yz = x.
k ∗
If n = 0, (r, z, ε) ` (q, ε, ε). By inductive hypothesis hr, ε, qi =⇒ z. And since
hp, A, qi −→ yhr, ε, qi is a rule (of type (c)), we have
∗
hp, A, qi =⇒ yhr, ε, qi =⇒ yz = x.