§3. Equations with Compact Operators

Equations of the second kind

Let A be a compact operator de…ned on the normed space E into itself,
the operator T = I A where I denotes the identity operator de…nes an
operator equation called equation of the second kind, given as

T ' = (I A)' = f;

or merely

' A' = f;
where f is a given function of E and ' is the unknown function of E:

Theorem 1
The null-space N (T ) of the operator T de…ned by

N (T ) = ker T = f' 2 E; T ' = (I A)' = 0g;

is a closed and …nite dimensional subspace.

Proof
Indeed, it is known that the kernel N (T ) of a bounded operator T is a
closed subspace of E; since, for all sequence 'n in N (T ) converges to ' in
E; we obtain ' in N (T ): Really, due to the boundedness of T we have

T 'n = 0 ) lim T 'n = 0;

n!1

or still
T lim ' = 0 ) T (') = 0:
n!1 n

Hence, the null-space N (T ) is closed.

On the other hand, all functions ' 2 N (T ) must satisfy the equation

T' = ' A' = 0;

or still
A' = ':
Noting that, the restriction of the operator A to the subspace N (T )
coincides with the identity operator on N (T ): The operator A is compact

1
 

 
from E to E and therefore also compact from N (T ) to N (T ) since N (T ) is
closed. Evidently, for all bounded sequence 'n in E in particular in N (T )
the sequence A'n = 'n contains a convergent subsequence A'nk = 'nk in
the closed N (T ): Hence, the compact operator A represents the identity
operator on the subspace N (T ) and therefore the subspace N (T ) is of …nite
dimensional.

Remark 1
The null-space N (T n ) of the operator T n for all n 2 N; is a closed and
…nite dimensional subspace. Indeed, the operators T n can be written in the
form
T n = (I A)n = I An ;
where An is a compact operator as combination of compact operators given
by
Xn
n
An = ( 1)i+1 Ai
i
i=1

Theorem 2
The sequence of null-spaces sets

N (T ) ; N (T 2 ); :::; N (T n ); :::

is increasing and stationary sequence. In other words, the sequence contains

uniquely a …nite number of distinct sets, so there exists a nonnegative integer
p 2 N; such that

f0g N (T ) N T2 ::: N (T p ) = N T p+1 = :::;

the number p is called the Riesz number of the compact operator A for the
null-spaces sets N (T n ):

Proof
Indeed, the inclusion is evident, since

' 2 N (T n ) ) T n ' = 0;

and therefore

T (T n ') = T n+1 ' = 0 ) ' 2 N (T n+1 ):

Hence, the inclusion of sets

N (T n ) N (T n+1 ); for all n 2 N: (1)

2
 

 
Suppose that there is no integer p 2 N; such that the sequence N (T n ) is
stationary, that is to say

N (T m ) 6= N (T n ); for all m; n 2 N; with m < n:

In other words, we write

f0g N (T ) ::: N (T m ) N (T m+1 ) ::: N (T n 1

) N (T n ) :::

In particular, taking N (T n 1 ) 6= N (T n ); the relation N (T n 1 ) N (T n )

between a closed subspaces involves the existence of an element 'n in N (T n );
with unit norm k'n k = 1; such that
1
'n 'n 1 > ; for all 'n 1 2 N (T n 1
):
2
Generally, for all sequence 'n 2 N (T n ) and for all m; n such that m < n;
we have the following relation

kA'n A'm k = k(I T )'n (I T )'m k

= k'n T 'n 'm + T 'm k
1
= k'n ('m T 'm + T 'n )k > : (2)
2
For, the elements of the sequence ('m T 'm + T 'n ) belong to the
subspace N (T n 1 ): Indeed, due to the relation

N (T m ) N (T n 1
) N (T n );

it comes

'm 2 N (T m ) ) 'm 2 N (T n 1
) and 'm 2 N (T n );

or still

'm 2 N (T m ) ) T m 'm = 0; T n 1
'm = 0 and T n 'm = 0:

Noting that, for 'n 2 N (T n ); we get

Tn 1
('m T 'm + T 'n ) = T n 1
'm T n 'm + T n 'n = 0:

Hence
('m T 'm + T 'n ) 2 N (T n 1
):

3
 

 
The sequence 'n is bounded, so by virtue of the compactness of the
operator A; we can extract a convergent subsequence from the sequence
A'n : Contradiction with the relation (2): Hence

N (T n 1
) = N (T n ):

It remains to demonstrate now the relation

N (T n ) = N (T n+1 ):

Indeed, for ' 2 N (T n+1 ) we get

' 2 N (T n+1 ) ) T n+1 ' = T n (T ') = 0;

it gives
T ' 2 N (T n ) = N (T n 1
);
that means

T ' 2 N (T n 1
) ) Tn 1
(T ') = T n ' = 0 ) ' 2 N (T n );
and therefore
N (T n+1 ) N (T n ):
Hence, there exists a nonnegative integer p 2 N; such that

f0g N (T ) N (T 2 ) ::: N (T p ) = N (T p+1 ) = N (T p+2 ) = :::;

where p is given by

p = minfk 2 N; such that N (T k ) = N (T k+1 )g:

Theorem 3
The range space R(T ) of the operator T de…ned by

R(T ) = Im T = T (E) = f ; 9' 2 E; T ' = g;

is a closed subspace.

Proof
It is known that, the range R(T ) of a linear operator T is a linear
subspace. Let f be an element of the closure T (E); then there exists a
sequence fn of the set T (E) such that

lim fn = f:
n!1

4
 

In other words, fn 2 T (E) there exists a sequence 'n 2 E such that

T 'n = fn ;

with the relation of convergence

lim T 'n = lim fn = f:

n!1 n!1

First case 'n bounded

Suppose that, the sequence 'n is bounded then, due to the compactness
of the operator A there exists a subsequence A'n(k) from the sequence A'n
such that, A'n(k) converges to : Hence, the convergence of the subsequence
'n(k) to an element ' in E: Say

lim 'n(k) = lim T 'n(k) + A'n(k)

k!1 k!1
= lim T 'n(k) + lim A'n(k)
k!1 k!1
= f+ = ' 2 E:

Due to the boundedness of the operator T and the convergence of the

sequence T 'n ; we get

f = lim fn = lim T 'n

n!1 n!1

= lim T 'n(k) = T lim 'n(k) = T ':

k!1 k!1

Hence f = T ' 2 T (E) = T (E):

Second case 'n unbounded

Suppose that, the sequence 'n is not bounded, then we get

1. If 'n 2 N (T )

For the sequence 'n inthe null space N (T ); we have T 'n = fn = 0

T 'n = 0 ) lim T 'n = 0 ) f = 0 2 T (E) = T (E);

n!1

as a linear subspace contains the null element.

5
 

 
2. If 'n 2
= N (T )

Taking the subspace G of E spanned by 'n and N (T ) de…ned as

G = span f'n + N (T )g:

The subspace N (T ) is closed in G: Hence, there exists an element n 2G

with a unit norm k n k = 1 such that
1
k n nk > ; 8 n 2 N (T );
2
with the following relation

n = an 'n + n ; an 2 R; n 2 N (T ):
Noting that, there is no subsequence an(k) of the sequence an converges
to the null element. For, if we suppose there exists a such subsequence, say
lim an(k) = 0; we obtain
k!1

lim T n(k) = lim (an(k) T 'n(k) ) + lim T n(k)

k!1 k!1 k!1
= lim an(k) : lim T 'n(k) + lim T n(k)
k!1 k!1 k!1
= 0f + 0 = 0:

In other words, there exists a subsequence n(j) of the subsequence n(k)

of the bounded sequence n such that A n(j) converges to an element of E:
This implies the convergence of the subsequence n(j) to the same element
of E; for, we have

lim n(j) = lim T n(j) + lim A n(j) = :

j!1 j!1 j!1

It is clear that, T = 0: Hence 2 N (T ): Contradiction with the fact

that
1
k n nk
> ; 8 n 2 N (T ):
2
We can therefore conclude that an 1 is bounded. Say

an 1 n = 'n + an 1 n:

Then, it comes

lim T (an 1 n) = lim T 'n + lim T (an 1 n)

n!1 n!1 n!1
= lim T 'n + 0 = f:
n!1

6
 

 
The sequence an 1 n is bounded as product of two bounded sequences
1 1 1
an and n : Hence there exists a subsequence an(k) n(k) such that A(an(k) n(k) )
converges to an element a 1 of E: This implies the convergence of the sub-
1 1 of E; for, we have
sequence an(k) n(k) to the same element a

lim a 1 n(k)
1
= lim T (an(k) n(k) )
1
+ lim A(an(k) n(k) ) =a 1
2 E:
k!1 n(k) k!1 k!1

The operator T is continuous, then we write

1 1
lim T (an(k) n(k) ) = T ( lim (an(k) n(k) ))
k!1 k!1
1
= T (a ) = f 2 T (E) = T (E):

Hence, the result

T (E) = T (E):

Theorem 4
The sequence of range spaces sets

R(T ); R(T 2 ); :::; R(T n ); :::

is decreasing and stationary sequence. In other words, the sequence contains

uniquely a …nite number of distinct sets, so there exists a nonnegative integer
q 2 N; such that

::::: = R(T q+1 ) = R(T q ) :::; : R(T 2 ) R(T ) E

The number q is called the Riesz number of the compact operator A for
the range spaces sets R(T n ):

Proof
Indeed, the inclusion is evident, since

2 R(T n+1 ) ) = T n+1 (') = T n (T ') = T n '1 2 R(T n );

and therefore
= T n+1 ' ) = T n '1 :
Hence, the inclusion of sets

R(T n+1 ) R(T n ): (3)

7
 

 
Suppose that there is no integer q 2 N; such that the sequence R(T n)

is stationary, that is to say

R(T m ) 6= R(T n ); for all m; n 2 N; with n < m:

In other words, we have

::: R(T m )::: R(T n+1 ) R(T n ) ::: R(T ) E

In particular, taking R(T n+1 ) 6= R(T n ); the relation R(T n+1 ) R(T n )
between a closed subspaces involves the existence of an element n in R(T n );
with unit norm k n k = 1; such that
1
k n n+1 k> ; for all n+1 2 R(T n+1 ):
2
Generally, for all sequence n 2 R(T n ) and for all m; n such that n < m;
we have the following relation

kA n A mk = k(I T) n (I T) mk
= k n T n m +T mk
1
= k n ( m T m +T n )k > : (4)
2
For, the elements of the sequence ( m T m +T n) belong to the
subspace N (T n+1 ): Indeed, due to the relation

R(T m+1 ) R(T m ) R(T n+1 ) R(T n );

it comes
m 2 R(T m ) ) m 2 R(T n+1 );
also, we have

T m 2 R(T m+1 ) ) T m 2 R(T n+1 ):

Noting that, for n 2 R(T n ); we get

n 2 R(T n ) ) T n 2 R(T n+1 )

Hence
( m T m +T n) 2 R(T n+1 ):

8
 

 
The sequence n is bounded, so by virtue of the compactness of the
operator A; we can extract a convergent subsequence from the sequence
A n : Contradiction with the relation (4): Hence

R(T n+1 ) = R(T n ):

It remains to demonstrate now the relation

R(T n+2 ) = R(T n+1 ):

Indeed, the …rst inclusion R(T n+2 ) R(T n+1 ) is always true following
(3); for the second one, we get

2 R(T n+1 ) ) = T n+1 ' = T (T n ')

= T (T n+1 '1 ) = T n+2 '1 2 R(T n+2 );

or still
R(T n+1 ) R(T n+2 ):
Hence, there exists a nonnegative integer q 2 N; such that

:::: = R(T q+2 ) = R(T q+1 ) = R(T q ) :::R(T 2 ) R(T ) E

where q is given by

q = minfk 2 N; such that R(T k ) = R(T k+1 )g:

Lemma 1
The Riesz number p of the null-spaces sets N (T n ) and the Riesz number
q of the ranges spaces R(T n ) are equal. Say

p=q

Proof
Suppose that, the Riesz numbers p and q are di¤erent, say p 6= q:

1. First case p < q;

f0g N (T ) ::: N (T p ) = N (T p+1 ) = ::: = N (T q 1

) = N (T q ) = ::; (5)

9
 

 
and also

::: = R(T q+1 ) = R(T q ) R(T q 1

) ::: R(T p ) ::: R(T ) E: (6)

We can see that, there exists a function 2 R(T q 1) such that 2

=
R(T q ); that is to say

= Tq 1
' 2 R(T q 1
);
the composition by the operator T of both sides, gives us

T = T q ' 2 R(T q ) = R(T q+1 );

this relation shows that, there exists a function '1 such that

T = T q ' = T q+1 '1 ;

or still
T q+1 '1 T q ' = 0:
Hence, we obtain

T q (T '1 ') = 0 ) T '1 ' 2 N (T q ) = N (T q 1

): (7)

It is to remark that the relation (7) gives us

T ('1 ) ' 2 N (T q 1
) ) Tq 1
(T '1 ') = 0 , T q '1 = T q 1
'= ;

this implies that = T q '1 2 R(T q ); contradiction with the fact that 2
=
R(T q ):

2. Second case q < p;

f0g N (T ) ::: N (T q ) ::: N (T p 1

) N (T p ) = N (T p+1 ) = ::: (8)

and also

::: = R(T p ) = R(T p 1

) = ::: = R(T q ) R(T q 1
) ::: R(T ) E (9)

We can see that, there exists a function ' 2 N (T p ) such that ' 2
=
N (T p 1 ); that is to say

10
 

Tp 1
' 2 R(T p 1
) = R(T p ) = R(T q );
this relation shows that, there exists functions '1 and '2 such that

TP 1
' = T p '1 = T q '2 ; (10)

the composition by the operator T of both sides and the relation ' 2 N (T p )
give us
T p ' = T p+1 '1 = T q+1 '2 = 0;
this implies that
'1 2 N (T p+1 ) = N (T p ):
Hence, it comes
T p+1 '1 = T p '1 = 0:
It is to remark that the relation (10) gives us T p '1 = T p 1 ' = 0; this
implies that ' 2 N (T p 1 ); contradiction with the fact that ' 2
= N (T p 1 ):

Theorem 5
The subspaces N (T r ) and R(T r ) are supplementary. That is to say

E = ker T r Im T r = N (T r ) R(T r );

where r = p = q is the Riesz number.

Proof
For all element 2 E; we get

2 E ) Tr 2 R(T r ) = ::: = R(T 2r ):

This relation implies the existence of a function '; such that

Tr = T 2r ' ) T r ( T r ') = 0;

or still,
( T r ') = 2 N (T r ):
Therefore, we have

= + T r '; with 2 N (T r ) and T r ' 2 R(T r ):

For all element 2 N (T r ) \ R(T r ); we get

11
 

2 R(T r ) and 2 N (T r );
this relation implies T r = 0 and the existence of a function '; such that
= T r '; it comes

= T r' ) Tr = 0 = T 2r ';

or still,
' 2 N (T 2r ) = :::; ::: = N (T r ):
Therefore, we have
= T r ' = 0:
Lemma 2
The operator T = I A is injective if and only if, the operator T r is
injective for all r 2 N:

Proof
Supposing that, The operator T is injective then, for all r 2 N; we have

T r '1 = T r '2 ) T (T r 1
'1 ) = T (T r 1
'2 ) ) T r 1
'1 = T r 1
'2
r 2 r 2 r 2 r 2
) T (T '1 ) = T (T '2 ) ) T '1 = T '2
) :::T (T '1 ) = T (T '2 ) ) T '1 = T '2
) '1 = '2 :

Hence, the operator T r is injective.

Supposing that, The operator T r is injective for all r 2 N; then we have

T '1 = T '2 ) T r 1
(T '1 ) = T r 1
(T '2 ) ) T r '1 = T r '2
) '1 = '2 :

Hence, the operator T is injective. That is to say

f0g = N (T ) = N (T 2 ) = ::: = N (T r ) = :::; :::

Lemma 3
The operator T = I A is surjective if and only if, the operator T r is
surjective for all r 2 N:

Proof

12
 

 
Suppose that, the operator T is surjective then, for all r 2 N; say

8 2 E; 9'1 2 E; = T '1 ) 9'2 2 E; '1 = T '2

) = T '1 = T (T '2 ) = T 2 '2
) ::9'r 2 E; 'r 1 = T 'r
) = T '1 = T (T '2 ) = ::: = T (T r 1
'r ) = T r 'r :

Finally, we obtain

8 2 E; 9'r 2 E; = T r ':

Hence, the operator T r is surjective.

Suppose that, the operator T r is surjective for all r 2 N; say

8 2 E; 9'1 2 E; = T r '1 ;

we can also write

T r '1 = T (T r 1
'1 ) = T ';
where the functîon ' = T r 1'
1 2 E: Finally, we obtain

8 2 E; 9' = T r 1
'1 2 E; = T ':

Hence, the operator T is surjective. That is to say

E = R(T ) = R(T 2 ) = ::: = R(T r ) = :::; :::

Theorem 6
Let A be a compact operator de…ned from a Banach space E into itself
then, the operator T = I A is injective if and only if, T = I A is
surjective. Besides the inverse operator T 1 = (I A) 1 de…ned from E
into E is bounded.

Proof
It is known that, for all Riesz number r = p = q; The subspaces N (T r )
and R(T r ) are supplementary. Say

E = N (T r ) R(T r ):

13
 

 
The injection of the operator T implies the one of T r : Hence the sur-
jection of the operator T r which it assures us the surjection of the
operator T:

The surjection of the operator T implies the one of T r : Hence the

injection of the operator T r which it assures us the injection of the
operator T:

The injection of the operator T or its surjection implies the bijection

of this operator T = (I A): Hence the boundedness of its inverse
T 1 = (I A) 1 :

Theorem 7
Let A be a compact operator from a Banach space E into itself then, the
nonhomogeneous equation

T' = ' A' = f (11)

admits a unique solution ' 2 E; for all f 2 E; if and only if, the homoge-
neous equation
T ' = ' A' = 0 (12)
admits uniquely a trivial solution ' = 0:

Proof
Indeed, suppose that the equation (11) admits a solution for all f 2 E;
it wants to say that the operator T is surjective and the Riesz number r is
null. Hence the operator T is injective. In other words, the equation (12)
admits the trivial solution ' = 0:
Reciprocally, suppose that the equation (12) admits uniquely the trivial
solution ' = 0; it wants to say that the operator T is injective and the
Riesz number r is null. Hence, the operator T is surjective and therefore
this operator is bijective. In other words, the existence and the uniqueness
of the solution of the equation (11):

14
 

 
Bibliography

[1] M. NADIR. Cours d’analyse fonctionnelle, université de Msila 2004.

[2] H. WIDOM. Lectures on integral equations, university of california
1969.

Address. Prof. Dr. Mostefa NADIR

Department of Mathematics
Faculty of Mathematics and Informatics
University of Msila
28000 ALGERIA
E-mail: [email protected]

15