THE OPEN MAPPING THEOREM AND RELATED THEOREMS

ANTON. R SCHEP

We start with a lemma, whose proof contains the most ingenious part of Banachs open
mapping theorem. Given a norm k ki we denote by Bi (x, r) the open ball {y X :
ky xki < r}.
Lemma 1. Let X be a vector space with two norms k k1 , k k2 such that (X, k k1 ) is a
Banach space and assume that the identity map I : (X, k k1 ) (X, k k2 ) is continuous.
If B2 (0, 1) B1 (0, r)

kk2

, then B2 (0, 1) B1 (0, 2r) and the two norms are equivalent.

Proof. From the hypothesis we get B2 (0, 1) B1 (0, r) + B2 (0, 21 ), so by scaling we get that
1
B2 (0, 21n ) B1 (0, 2rn ) + B2 (0, 2n+1
) for all n 1. let now kyk2 < 1. Then we can write
y = x1 + y1 , where kx1 k1 < r and ky1 k2 < 12 . Assume we have kyn k2 < 21n we can write
1
yn = xn+1 + yn+1 , where kxn+1 k1 P
< 2rn and kyn+1 k2 < 2n+1
. By completeness of (X, k k1 )

there exists x X such that x = n=1 xn , where the series converges with respect to the
norm k k1 . By continuity of the identity map I : (X, k k1 ) (X, k k2 ) it follows that
the same series also converges to x with respect to k k2 . On the other hand the equation
P
P
y = n+1
n=1 xn converges to y with respect to k k2 .
k=1 xk + yn+1 shows that the series
P
Hence y = x and thus kyk1 = kxk1 n=1 kxn k1 < 2r. It follows thatB2 (0, 1) B1 (0, 2r)
and thus kyk1 2rkyk2 for all y X. As the continuity of I gives that there exists C such
that kyk2 Ckyk1 for all y X, we get that the two norms are equivalent.

Theorem 2. Let X be a vector space with two norms k k1 , k k2 such that (X, k k1 ) and
(X, k k2 ) are Banach spaces. Assume that the identity map I : (X, k k1 ) (X, k k2 ) is
continuous. Then the norms k k1 and k k2 are equivalent.
Proof. Applying the Baire Category theorem in (X, k k2 ) to X =
n=1 B1 (0, n) we can
kk2

find n0 , x0 and r0 > 0 such that B2 (x0 , r0 ) B1 (0, n0 )

kk2

that B2 (0, r0 ) B1 (x0 , n0 )

B1 (0, n0 + kx0 k1 )

kk2

. Translating over x0 we get

. Now by the triangle inequality we get that B2 (0, r0 )

. By the above lemma the two norms are equivalent.

Theorem 3 (Bounded Inverse Theorem). Let X, Y be Banach spaces and assume T : X

Y is an one-to-one, onto continuous linear operator. Then T 1 : Y X is continuous.
Proof. Define kykT = kT 1 yk. Then k kT is a norm on Y and
P kyk kT kkykT for
y Y , so I : (Y, k kT ) (Y, k k) is continuous. Moreover, if
n=1 kyn kT < , then
P
PN
P
1
1 y k 0 as
x0 =
yn exists in X and kT x0 n=1 yn kT = kx0 N
n
n=1 T
n=1 T
N . Hence (Y, k kT ) is also a Banach space. By the above Theorem the two norms
1

ANTON. R SCHEP

on Y are equivalent, so there exists C such that kT 1 (y)k Ckyk for all y Y , i.e. T 1
is continuous.

We recall now that a linear map T : X Y is called open if T (O) is open for all open
O X. It is easy to see that an open linear map is surjective. The Open Mapping theorem
gives a converse to that statement. Before stating and proving that theorem, we recall a few
basic facts about quotient maps. Let X be a Banach space and M X a closed subspace.
Then X/M is a Banach space with respect to the quotient norm k[x]k = inf{kyk; y [x]}.
Denote by Q the quotient map Q(x) = [x]. Then Q is open. In fact, it is easy to see from
the definition of the quotient norm that Q({x : kxk < 1}) = {[x] : k[x]k < 1}.
Theorem 4 (Open Mapping Theorem). Let X, Y be Banach spaces and assume T : X
Y is an onto continuous linear operator. Then T is an open map.
Proof. Let Q : X X/ker(T ) be the quotient map. Then by the above remarks Q is an
open mapping. Let T : X/ker(T ) Y be the induced map such that T = T Q. Then T
is one to one and onto, so by the above Theorem T1 is continuous, so T is open and thus
T is open.

Let now A : D(A) Y be a linear operator, where D(A) is a (not necessarily closed)
linear subspace of the Banach space X. The subspace D(A) is called the domain of A.
Given a linear operator A : D(A) Y we define the graph
(A) = (x, Ax) : x D(A),
It is clear that (A) is linear subspace of X Y . We can equip X Y with the product
norm k(x, y)k = kxk+kyk. Then we say that A has a closed graph (or is a closed operator),
if (A) is a closed subspace of X Y .
Example 5. Let X = Y = C[0, 1] with the supremum norm. Let D(A) = C 0 [0, 1] the
subspace of X consisting of continuously differentiable functions and define A : D(A) Y
by Af = f 0 . One can can see that A is not bounded, by taking fn (t) = tn , and noting
that kfn k = 1 and kAfn k = n. On the other hand A has a closed graph. To see that A
has a closed graph, let (fn , fn0 ) (f, g) in X Y . Then by the Fundamental Theorem of
Rt
Rs
Rt
Calculus fn (t) fn (0) = 0 fn0 (s) ds 0 g(s) ds. It follows that f (t) = f (0) + 0 g(s) ds.
Hence f D(A) and f 0 = g, i.e., (f, g) (A).
The following proposition is immediate from the definition.
Proposition 6. Let X and Y be Banach spaces and assume A : D(A) Y is a linear
operator, where D(A) is a subspace of X. Then the following are equivalent.
(1) A has a closed graph.
(2) If xn D(A), xn x X, and Axn y Y , then x D(A) and Ax = y.
(3) D(A) is a Banach space with respect to the graph norm kxkA = kxk + kAxk.
Theorem 7 (Closed Graph Theorem). et X and Y be Banach spaces and assume A :
X Y is a closed linear operator. Then A is bounded.

THE OPEN MAPPING THEOREM AND RELATED THEOREMS

Proof. Define P : (A) X by P (x, Ax) = x. Then P is clearly a bounded, one-to-one,

onto linear operator, so by the Bounded Inverse Theorem the inverse operator P 1 : X
(A) is bounded. Hence there exists a constant C such that kxk + kAxk Ckxk, i.e.,
kAxk (C 1)kxk for all x X.

We now present a proof of the Uniform Boundedness Principle, based on the Closed
Graph Theorem.
Theorem 8 (Banach-Steinhaus). Let X and Y be Banach spaces and assume A
L(X, Y ) ( F) is a pointwise bounded family of bounded operators, i.e., for all x X
there exists a constant Cx such that kA xk Cx for all F. Then there exists a
constant C such that kA || C for all F.
Proof. Define the space Y = {(y ) : y Y, sup ky k < } with norm k(y )k =
sup ky k. It is straightforward to verify that Y is also a Banach space. Now define
T : X Y by T x = (A x). Note T x Y , since the collection A is pointwise
bounded. Clearly T is linear and we claim that T is closed. To see this, let xn 0
and T xn (y ). Then A xn y for all F, but also A xn 0 for all . Hence
(y ) = (0) and T is closed. By the Closed Graph Theorem T is bounded, i.e., there exists
a constant C such that for all kxk 1 we have sup kA xk C. Hence kA k C for all
F.