A note on the grand theorems of functional analysis

S. Kesavan
Department of Mathematics,
Indian Institute of Technology, Madras,
Chennai - 600 036
email: kesh@iitm.ac.in

Abstract

Three grand theorems of functional analysis are the uniform boundedness

(or, Banach-Steinhaus) theorem, the open mapping theorem and the closed
graph theorem. All these are consequences of a topological result known as
Baire’s (category) theorem. The mathematical literature has several results
which show the interconnections between these theorems, some well known
and well documented, while some others are less known. The aim of this
note is to present all these interdependences in a cogent manner.

1
1 Introduction
The high point in any first course on functional analysis is the proof of
three grand theorems, known as the uniform boundedness theorem, the open
mapping theorem and the closed graph theorem. All three depend on the
completeness of some or all of the spaces involved and their proofs depend
on a topological result known as Baire’s thoerem (or, the Baire category
theorem). It is well recorded that the open mapping theorem and the closed
graph theorem are equivalent in the sense that each can be deduced from
the other, though in most text books one of them is proved, starting from
Baire’s theorem, and the other is deduced as a consequence. The uniform
boundedness theorem is proved independently. Looking at the mathematical
literature, we also see works wherein it is shown how to deduce the uniform
boundedness theorem from the closed graph theorem. The converse is also
true with some restrictions. These results seem to be less known. The aim
of this note is to present all the possible equivalences and deductions in a
cogent manner. We will also show that all the three results are ‘equivalent’
to each other in the case of Hilbert spaces.
It needs to be stressed again that all these results are scattered in the
literature and no claim to originality in the proof techniques is made.
The proof of the uniform boundedness theorem from Baire’s theorem is
probably the simplest of all these proofs. However, there also exist several
‘elementary’ proofs of this result, in the sense that these proofs do not use
Baire’s theorem. We will present here a really simple proof, which does not
involve Baire’s theorem, of the uniform boundedness theorem. We do not
present any direct proofs of the open mapping or closed graph theorems as
these could be found in any text book on functional analysis.

2 The main result

Theorem 2.1 Each of the following statements imply the others.

(i) The closed graph theorem

Let V and W be Banach spaces and let T : V → W be a linear map. If the
graph of T defined by

G(T ) = {(x, T x) | x ∈ V } ⊂ V × W

2
is closed in V × W , then T is continuous.

(ii) The open mapping theorem

Let V and W be Banach spaces and let T : V → W be a continuous linear
map which is surjective. Then T is an open map, i.e. T maps open sets of
V onto open sets of W .

(iii) The bounded inverse theorem

Let V and W be Banach spaces and let T : V → W be a continuous linear
bijection. Then T is an isomorphism, i.e. T −1 is also continuous.

(iv) The two norms theorem

Let V be a vector space and let k · k1 and k · k2 be two norms on V . If V
is a Banach space with respect to either norm and if there exists a constant
C > 0 such that
kxk1 ≤ Ckxk2 (2.1)
for every x ∈ V , then the two norms are equivalent.

Each of the above statements implies the following:

(v) Uniform boundedness theorem

Let V be a Banach space and W be a normed linear space. Let, Ti : V → W
be a continuous linear map for each i ∈ I. If

sup kTi xkW < ∞ (2.2)

i∈I

for each x ∈ V , then there exists a constant C > 0 such that kTi k ≤ C for
each i ∈ I.

Proof: (i) ⇒ (ii)

Assume that the closed graph theorem is true. Let V and W be Banach
spaces and let T : V → W be a continuous linear map which is surjective.
Let N = Ker(T ), which is a closed subspace. Let y ∈ W . Since T is onto,
there always exists x ∈ V such that T x = y. Define Sy = x + N ∈ V /N . If
T x1 = T x2 = y, then x1 − x2 ∈ N and so x1 + N = x2 + N and so the map
S is well-defined, and clearly linear. We claim that the graph of S is closed.
To see this, let yn → y in W and let Syn → z + N in V /N . Let T xn = yn so

3
that Syn = xn + N . Then it follows that

inf kxn − z − wkV → 0

w∈N

as n → ∞. Thus, there exists a sequence {wn } in N such that kxn − z −

wn kV → 0, or in other words, xn −wn → z. Then yn = T (xn ) = T (xn −wn ) →
T z. Thus y = T z. Then it follows that Sy = z + N , which establishes our
claim.
Consequently, by the closed graph theorem, S is continuous, i.e. there
exists K > 0 such that for all y ∈ W , we have kSykV /N ≤ KkykW . Thus
kykW < K1 implies that kSykV /N < 1, or, in other words, if kykW < K1 ,
for y ∈ W , there exists x ∈ V and z ∈ N such that kx − zkV < 1 and
y = T x = T (x − z). Thus it follows that
 
1
BW 0; ⊂ T (BV (0; 1)
K

(where BX (x; r) denotes the open ball centered at a point x and of radius
r > 0 in a normed linear space X). From this we can easily conclude that T
maps open sets of V onto open sets of W .

(ii) ⇒ (iii)
Assume that the open mapping theorem is true. Then if T : V → W is a bi-
jective and continuous linear map between the Banach spaces V and W , then
it maps open sets onto open sets. This immediately implies that T −1 : W →
V is continuous since, if U ⊂ V is an open set, then (T −1 )−1 (U ) = T (U )
which is open in W .

(iii) ⇒ (iv)
Let V be a vector space which is complete with respect to both the norms
k·k1 and k·k2 . If I : V → V is the identity mapping, then the given inequality
(2.1) implies that this map is a continuous bijection from the Banach space
V (with the norm k · k2 ) to the Banach space V (with the norm k · k1 ) and
hence, by (iii), it is an isomorphism, which establishes the equivalence of the
two norms.

(iv) ⇒ (i)
Let V and W be Banach spaces and let T : V → W be a linear map whose

4
graph is closed. Define the norm

kxk1 = kxkV + kT xkW

for x ∈ V . If {xn } is a Cauchy sequence in V with respect to this norm, then

it is also a Cauchy sequence with respect to the orignal norm k · kV . Further
{T xn } is a Cauchy sequence in W . Since the spaces are complete with the
original norms, it follows that xn → x in V and T xn → y in W . Since the
graph is closed this implies that T x = y. Thus xn → x in V with respect to
the norm k · k1 . Thus we deduce that V is a Banach space with respect to
the norm k · k1 as well. Clearly, for each x ∈ V , we have

kxkV ≤ kxk1 .

Thus the two norms are equivalent by virtue of (iv) and so there exists a
constant C > 0 such that kxk1 ≤ CkxkV for each x ∈ V . In particular, we
also have kT xkW ≤ CkxkV , which proves the continuity of T .

We now deduce (v) assuming the validity of the preceding statements.

Let {Ti }i∈I be a collection of continuous linear maps from a Banach space V
to a normed linear space W such that (2.2) holds for all x ∈ V . Now consider
the norm on V defined by

kxk2 = kxkV + sup kTi xkW , x ∈ V

i∈I

which is well-defined in view of (2.2). If {xn } is a Cauchy sequence in V

with respect to this norm, then it is also a Cauchy sequence with the original
norm and so, xn → x in V (with respect to the original norm). From the
definition of k · k2 , it follows that the sequences {Ti xn } are uniformly Cauchy
from which we immediately see that
n→∞
sup kTi xn − Ti xkW −→ 0.
i∈I

Thus {xn } converges to x with respect to the norm k · k2 as well and so the
space V is complete with respect to both norms. Since we have that kxkV ≤
kxk2 for all x ∈ V , it follows from (iv) that these norms are equivalent. Thus
there exists C > 0 such that for all x ∈ V , we have kxk2 ≤ CkxkV . In
particular, we have
sup kTi xkW ≤ CkxkV
i∈I

5
for all x ∈ V which implies that kTi k ≤ C for every i ∈ I. This completes
the proof. 

Remark 2.1 In many books on functional analysis, the open mapping theo-
rem is proved using Baire’s theorem and the closed graph theorem is deduced
in the same way we have proved the implications (ii)⇒(iii)⇒(iv)⇒(i). The
uniform boundedness theorem is proved independently, and in a fairly sim-
ple manner, from Baire’s theorem. The proof of the implication (iv)⇒(v)
appears in a note by Ramaswamy and Ramasamy [2]. 

The proof of the uniform boundedness theorem starting from Baire’s the-
orem is, as already remarked, fairly simple and appears in all texts on func-
tional analysis. We now present a very simple proof (due to Sokal [4]) of this
theorem without appealing to Baire’s theorem.

Proof of the uniform boundedness theorem:

Step 1. Let x, y ∈ V . Since 2y = (x + y) − (x − y), the triangle inequality

gives
1
kT ykW ≤ [kT (x+y)kW +kT (x−y)kW ] ≤ max{kT (x+y)kW , kT (x−y)kW }
2
where T is a continuous linear map from V to W . If we now take the
supremum as y varies over the open ball centred at the origin and of radius
r > 0, we deduce that
sup kT x0 kW ≥ rkT k. (2.3)
x0 ∈BV (x;r)

Step 2. Assume that {kTi k}i∈I is unbounded. Then choose a sequence

{Tn }∞ n
n=1 from this family such that kTn k ≥ 4 for each n. Set x0 = 0. Then,
by Step 1, we can inductively find {xn }∞ that kxn −xn−1 kV < 3−n
n=1 in V suchP
2 −n
and kTn xn kW > 3 3 kTn k. By construction, since 3−n is convergent, the
sequence {xn } is Cauchy and since V is complete, we have xn → x in V .
Now, if m > n, we have
kxn − xm kV ≤ 3−(n+1) + 3−(n+2) + · · · + 3−m .
Keeping n fixed and letting m → ∞, we deduce that
1 −n
kxn − xkV ≤ 3 .
2
6
Then, by the triangle inequality, we get that
 n
1 1 4
kTn xkW ≥ kTn xn kW − kTn (xn − x)kW ≥ 3−n kTn k ≥ .
6 6 3

Thus {kTn xkW } is unbounded which contradicts (2.2). This completes the
proof. 

3 Operators on Hilbert spaces

All the statements (i)-(v) of Theorem 2.1 are equivalent to each other in the
context of Hilbert spaces. More precisely, we have the following result.

Theorem 3.1 Each of the following statements imply the others.

(i) The closed graph theorem

Let V and W be Hilbert spaces and let T : V → W be a linear map. If the
graph of T is closed in V × W , then T is continuous.

(ii) The open mapping theorem

Let V and W be Hilbert spaces and let T : V → W be a continuous linear
map which is surjective. Then T is an open map.

(iii) The bounded inverse theorem

Let V and W be Hilbert spaces and let T : V → W be a continuous linear
bijection. Then T is an isomorphism, i.e. T −1 is also continuous.

(iv) The two norms theorem

Let V be a vector space and let (·, ·)1 and (·, ·)2 be two inner-products on V
yielding the norms k · k1 and k · k2 on V , respetively. If V is a Hilbert space
with respect to either inner-product and if there exists a constant C > 0 such
that
kxk1 ≤ Ckxk2 (3.1)
for every x ∈ V , then the two norms are equivalent.

(v) Uniform boundedness theorem

Let V and W be Hilbert spaces. Let, Ti : V → W be a continuous linear map

7
for each i ∈ I. If
sup kTi xkW < ∞ (3.2)
i∈I

for each x ∈ V , then there exists a constant C > 0 such that kTi k ≤ C for
each i ∈ I.
Proof: The equivalence of statements (i)-(iv) has already been established
in Theorem 2.1. The implication (iv)⇒(v) has also been proved there. So we
just need to show that the statement (v) implies the others. We will show
that (v)⇒(i). Let T : V → W be a linear map. Then we can define its
adjoint T ∗ : D(T ∗ ) ⊂ W → V as follows. We define
D(T ∗ ) = {y ∈ W | |(T x, y)W | ≤ CkxkV for all x ∈ V }.
If y ∈ D(T ∗ ), then x 7→ (T x, y)W defines a continuous linear functional on V
and so, by the Riesz representation theorem, there exists a unique element,
that we denote as T ∗ y, in V such that
(x, T ∗ y)V = (T x, y)W . (3.3)
The map T ∗ : D(T ∗ ) ⊂ W → V is clearly linear. We claim that if the graph
of T is closed, then D(T ∗ ) = W . We show this in two steps.

Step 1: D(T ∗ ) is dense in W . If not, by the Hahn-Banach theorem, there

exists y ∈ W such that y 6= 0 and (z, y)W = 0 for all z ∈ D(T ∗ ). Now (0, y)
does not belong to the graph of T and since the graph of T is closed, there
exists (v, w) ∈ V × W such that (y, w)W 6= 0 and (x, v)V + (T x, w)W = 0 for
all x ∈ V , again by the Hahn-Banach theorem. Thus, for all x ∈ V ,
|(T x, w)W | ≤ kxkV kvkV
and so w ∈ D(T ∗ ) which then implies that (w, y)W = 0, which is a contra-
diction. This establishes the density of D(T ∗ ) in W .

Step 2: D(T ∗ ) = W . Let y ∈ W . Then, there exists a sequence {yn } in

D(T ∗ ) such that yn → y. Thus, for all x ∈ V ,
(x, T ∗ yn )V = (T x, yn )W . (3.4)
Since the sequence {yn } is convergent, it is bounded, and we deduce from
(3.4) that
|(x, T ∗ yn )V | ≤ CkT xkW for all x ∈ V.

8
It then follows from the uniform boundedness theorem that {T ∗ yn } is bounded
as well. Let kT ∗ yn kV ≤ K for all n. Thus (3.4) yields
|(T x, yn )W | ≤ KkxkV for all x ∈ V
and passing to the limit, we get
|(T x, y)W | ≤ KkxkV for all x ∈ V.
Thus we deduce that y ∈ D(T ∗ ) which establishes our claim since y was
arbitrarily chosen in W .

Thus, from (3.3), it follows that for all x ∈ V such that kxk ≤ 1, and for
all y ∈ W , we have
|(T x, y)W | ≤ kT ∗ ykV .
Consequently, it follows, once again from the uniform boundedness theorem,
that
sup kT xkW < ∞.
kxkV ≤1

In other words, T is continuous. This completes the proof. 

Remark 3.1 Similar arguments appear in the literature. Ramaswamy and

Ramasamy [2] show that the uniform boundedness theorem implies the closed
graph theorem provided we assume that the target space W is reflexive,
while it is enough to assume that V is Banach. Ramm [3] also proves this
implication when V = W is a Hilbert space.
The main point to note here is that for the implication (iv)⇒(v), it was
enough that V be complete while nothing need be assumed on W . For the
reverse implication, we need the reflexivity (and hence the completeness) of
W . The validity of the implication (v)⇒(iv), in the case of an arbitrary
normed linear space W , seems to be still open. 

Remark 3.2 In view of the simple and ‘elementary’ proof of the uniform
boundedness theorem given in Section 2, we now have a complete proof of
all the statements in Theorem 3.1 without appealing to Baire’s theorem, in
the framework of Hilbert spaces. 

Remark 3.3 Halmos [1] also gives a proof of the uniform boundedness theo-
rem in the framework of Hilbert spaces without appealing to Baire’s theorem.


9
References
[1] Halmos, P. R. A Hilbert Space Problem Book, Springer-Verlag, New
York, 1974.

[2] Ramaswamy, S. and Ramasamy, CT. Deduction of one from another of

three major theorems in functional analysis, personal communication.

[3] Ramm, A. G. A simple proof of the closed graph theorem, Global J.

Math. Anal., 4(1), 2016, page 1.

[4] Sokal, A. D. A really simple elementary proof of the uniform bound-

edness theorem, Amer. Math. Monthly, 118, No. 5, 2011, pp. 450-452.

10