Closed range theorem explained

In the mathematical theory of Banach spaces, the closed range theorem gives necessary and sufficient conditions for a closed densely defined operator to have closed range.

The theorem was proved by Stefan Banach in his 1932 Théorie des opérations linéaires.

Statement

Let

X

and

Y

be Banach spaces,

T:D(T)\toY

a closed linear operator whose domain

D(T)

is dense in

X,

and

T'

the transpose of

T

. The theorem asserts that the following conditions are equivalent:

R(T),

the range of

T,

is closed in

Y.

R(T'),

the range of

T',

is closed in

X',

the dual of

X.

R(T)=N(T')\perp=\left\{y\inY:\langlex*,y\rangle=0{forall

}\quad x^* \in N(T') \right\}.

R(T')=N(T)\perp=\left\{x*\inX':\langlex*,y\rangle=0{forall

}\quad y \in N(T) \right\}.

Where

N(T)

and

N(T')

are the null space of

T

and

T'

, respectively.

Note that there is always an inclusion

R(T)\subseteqN(T')\perp

, because if

y=Tx

and

x*\inN(T')

, then

\langlex*,y\rangle=\langleT'x*,x\rangle=0

. Likewise, there is an inclusion

R(T')\subseteqN(T)\perp

. So the non-trivial part of the above theorem is the opposite inclusion in the final two bullets.

Corollaries

Several corollaries are immediate from the theorem. For instance, a densely defined closed operator

T

as above has

R(T)=Y

if and only if the transpose

T'

has a continuous inverse. Similarly,

R(T')=X'

if and only if

T

has a continuous inverse.

Sketch of proof

Since the graph of T is closed, the proof reduces to the case when

T:X\toY

is a bounded operator between Banach spaces. Now,

T

factors as

X\overset{p}\toX/\operatorname{ker}T\overset{T0}\to\operatorname{im}T\overset{i}\hookrightarrowY

. Dually,

T'

is

Y'\to(\operatorname{im}T)'\overset{T0'}\to(X/\operatorname{ker}T)'\toX'.

Now, if

\operatorname{im}T

is closed, then it is Banach and so by the open mapping theorem,

T0

is a topological isomorphism. It follows that

T0'

is an isomorphism and then

\operatorname{im}(T')=\operatorname{ker}(T)\bot

. (More work is needed for the other implications.)

\square

References