MATHEMA TICS

CLOSED LINEAR OPERATORS ON BANACH SPACES

BY

M. A. KAASHOEK

(Communicated by Prof. A. C. ZAANEN at the meeting of January 30, 1965)

o. INTRODUCTION. In the present paper we summarize a number of

results and examples about closed linear operators from one Banach
space into another. Some of these results are well-known; for full proofs
of the theorems quoted without proof we refer to [7] and [8]. Theorem 2.3
is probably new, but it is a simple consequence of a theorem due to
A. BROWN [3]. Section 4 contains necessary and sufficient conditions in
order that a closed linear operator has a closed range. Some theorems
in sections 2 and 4 can be generalized to the case of closed linear operators
on locally convex linear topological spaces (cf. F. E. BROWDER [2]).
Section 6 is devoted to stability problems for closed linear operators
for which the range is a Banach space such that the range of the
perturbation is contained in the same space, and here full proofs are
supplied. This section (and also sections 3 and 5) serves as an introduction
to an investigation of more general stability problems to be discussed
in a forthcoming paper. The author is indebted to Professor A. C. Zaanen
for suggestions and guidance.

1. CLOSED EXTENSIONS. Let X and Y be Banach spaces (over the

complex numbers). By a subspace of X or Y we shall always mean a
linear subspace, not necessarily closed. Let T be a linear operator (either
bounded or not) with domain D(T) C X and range R(T) C Y. The sub-
space G(T) = {(x, Tx): x E D(T)} of the product space X x Y is called
the graph of T. Given the positive constants iX and (3, the product space
X x Y is a Banach space with respect to the norm II(x, y) II = iXllxll + (3llyll.
The following definition is well-known.

Definition 1.1. The linear operator T is said to be closed whenever

G(T) is a closed subspace 01 X x Y. Equivalently, T is closed whenever it
follows from Xn E D(T) for n= 1, 2, ... , Xn -+ x and TXn -+ y that x E D(T)
and Tx=y.

The linear operator Tl with domain D(T1) C X and range R(T1) C Y

is called an extension of T if D(T) C D(T1) and T1x=Tx for all x E D(T).
If, in addition, Tl is closed, then Tl is called a closed extension of T.
It is well-known that T has a closed extension if and only if it follows
 406
°
from Xn E D(T) for n= 1,2, ... , Xn ~ and TXn ~ y that y=O, and in
this case there exists a linear operator To such that G(To)=G(T), where
the bar denotes closure in the norm topology of X x Y. Obviously, To
is the smallest closed extension of T.
If D(T) is dense in X and T has a closed extension T I , then it is
well-known (see, e.g. [12], section 4.2) that TI is bounded if and only
if D(TI ) = X, and in this case TI is uniquely determined. If T is unbounded,
i.e., if D(T I ) =1= X, then T has no maximal closed extension. This is a
consequence of the following theorem.

Theorem 1.2. Let T be an unbounded, closed linear operator. Then

there exists a sequence {Tn; n= 1,2, ... } of unbounded closed linear operators
such that TI is a proper extension of T and each T n+l is a proper extension
of Tn.

2. A LINEAR OPERATOR AND ITS ADJOINT. The Banach dual space

of the Banach space X will be denoted by X*. Given the non-empty

°
subset S of X, the annihilator Sl. of S is set of all f E X* such that
f(x) = for all XES. Similarly, given now the non-empty subset V of X*,
the inverse annihilator l. V is the set of all x E X such that f(x) = 0 for
all f E V. We shall make use of the theorem that, for any subspace
M of X*, the weak* closure of M in X* is (l.M)l..

Let T be a linear operator with domain D(T) dense in X and range

R(T) in Y; it is not assumed that T is closed. For such an operator T,
there is an important connection between the graphs G(T) and G(T*),
where T* denotes the adjoint operator of T. In order to describe this
connection we introduce the mapping E: y* x x* ~ (X x Y)* defined
by the formula {E(f, g)}(x, y) = f(y) + g(x). It is not difficult to prove that
E is a one-one linear mapping onto. Moreover, E and its inverse are
continuous with respect to the norm topologies in y* x X* and (X x Y)*
respectively. The following theorems are well-known.
Theorem 2.1. Let T be a linear operator with domain dense in X. Then
(i) E{G(T*)}= {G( -T)}l., and hence l.[E{G(T*)}] =G( -T).
(ii) l.[E{G(T*)}]=G( -T) if and only if T is closed.
Theorem 2.2. Let D(T) be dense in X. Then T has a closed extension
if and only if the weak* closure of D(T*) is Y*.
Theorem 2.3. Let D(T) be dense in X, and let T have a closed extension.
Then T is bounded if and only if T* is bounded.
Proof. Let T be bounded. Then the smallest closed extension To of T
is a bounded linear operator with D(To) =X, and hence D(To*) = Y*.
Since Y*=D(To*)CD(T*)CY*, we have D(T*)=Y*. The last fact
implies, by the closed graph theorem, that T* is bounded.
 407
Conversely, let T* be bounded and let To be the smallest closed
extension of T. Then T* is a closed extension of To* (actually, T* =To*),
and so To* is bounded. This implies, by a theorem of A. BROWN [3],
that D(To*)=Y*, and hence D(T*)=Y*. Then a simple computation
shows that T is bounded.
In the particular case that X and Yare reflexive Banach spaces, we
have the following theorem.
Theorem 2.4. Let T be a linear operator with domain dense in X.
Then T has a closed extension il and only il D(T*) is dense in Y*, and
in this case T** is the smallest closed extension 01 T.
We note that Theorem 2.2 is also true if X and Yare locally convex
linear Hausdorff spaces (see F. E. BROWDER [2]). In the case that
X = Y is a Hilbert space, the result in Theorem 2.4 is due to J. von
Neumann (cf. F. RIESZ and B. Sz.-NAGY [9], section 117).
The question may be raised whether D(T*) is always dense in Y*,
since then the second adjoint T** exists. We will show by means of an
example that D( T*) is not necessarily dense in Y*, not even if T is closed.
Also it will be shown that without the hypothesis that T has a closed
extension Theorem 2.3 does not necessarily hold.
We recall that lp( 1 <:, P < =) is the linear vector space consisting of all
sequences X=(Xl, X2, ... ) of complex numbers such that !~Ixnl p<=.
It is well-known that lp is a Banach space with respect to the norm
Ilxlip = (!~ Ixnl p)l/p. Similarly, loo is the space of all bounded sequences,
and loo is a Banach space with respect to the norm Ilxll oo = supn Ixnl. It
is also well-known that any bounded linear functional I(x) on lp(1 <:,p<=)
can be represented uniquely in the form I(x) = !~ tXnX n , where tX =
(tXl, tX 2 , ••. ) E lq (p-l + q-l = 1) and 11/11 = IltXllq. Conversely, every tX E lq defines
in this wayan element IE lp*.
Example 2.5. Given X=(Xl, X2, ... ) E It, we define the vector y by
y=(xl, X2, X2, x 3 , x 3 , x 3 , x 3 , X4, ... ); generally, yn=x k for n=2 k - l +j,
j=0,1, ... ,2 k - l _1. Let D(T)={X:XElt,YElt}. Then D(T) is a proper
subspace oflt. For any X= (xl, X2, ... ) E It we have Xk= (Xl, ... , Xk, 0, 0, ... ) E
D(T) for k= 1,2, ... , and so D(T) is dense in It.
Defining now Tx=y for all x E D(T), it is not difficult to verify that
T is a closed linear operator with domain D(T) dense in It and range in It.
Hence, T* exists with domain and range in loo. Given tX E D(T*), the
corresponding I E It * satisfies I(x) = !~ tX n x n , so
I(Tx) = !~(tX2n-l + ... + tX2n-l )xn
is a bounded linear functional on D(T). It follows that tX E D(T*) if
and only if

Writing tXo= (1, 1, ... ), it is evident that the neighbourhood {tx: IltX-tXolloo<H
 408

contains no points of D(T*). Hence D(T*) is not dense in lo,,, so T**

does not exist.

Example 2.6. Let D(T)= {x: x E l2' 2~lxkl <oo}, and define Tx=y=
(yl, y2, ... ) by setting yn=n-1. 2~Xk for all x ED(T). Then T is a linear
operator with domain D(T) dense in l2 and range in l2. Hence T* exists
with domain and range in l2. Given 1X E D(T*), the corresponding IE l2*
satisfies I(x) = 2~ 1XkXk, and so I(Tx) = 2:~1 (2~~lk-11Xk)xn is a bounded
linear functional on D(T). It follows that 1X E D(T*) if and only if
00

and in this case T* 1X = o. Hence T* is bounded.

Define for k=l, 2, ... the sequence Xk=(Xk 1,Xk2, ... ) by xkn=k-1 for
n= 1,2, ... , k and Xk n = 0 for n>k. Then Xk E D(T) and TXk= (1, t, 1, ... )
for k = 1, 2, .... From IIxkl122 = k- 1 , it follows that Xk ---+ 0 in l2, and so
T is not bounded.

3. NULL SPACES AND RANGES. The null space {x: xED(T),Tx=O}

of a linear operator T will be denoted by N(T). If T is closed, then N(T)
is closed. We recall the following theorem, relating null spaces and ranges.

Theorem 3.1. Let T be a linear operator with D(T) dense in X. Then

(i) R(T)l.=N(T*), and hence R(T)=l.N(T*).
(ii) II, in addition, T is closed, then l.R(T*)=N(T), and hence
R(T*) C N(T)l..

The dimension of any subspace L of the linear space X (denoted by

dim L) is the maximal number of linearly independent elements in L.
Hence, the value of dim L can be zero, any natural number or + 00.
For T a linear operator with domain D(T) in the Banach space X and
range R(T) in the Banach space Y, we set

1X(T) = dim N(T), f3(T) = dim R(T)l.,

and if, in addition, at least one of 1X(T) and f3(T) is finite, then we define

v(T) = 1X(T) - f3(T).

The number 1X(T) is called the nullity, f3(T) the deficiency and v(T) the
index of T.
If D(T) is dense in X, then T* exists and R(T)l.= N(T*) by the
preceding theorem. In this case, therefore, we have f3(T) = dim N(T*) =
1X(T*). Another important case will be investigated in the next section.
We will find that for T densely defined and closed with closed range,
we have also that 1X(T) = f3(T*).
 409
4. CLOSED LINEAR OPERATORS WITH CLOSED RANGE. Let T be a closed
linear operator with domain D(T) C X and range R(T) C Y. The null
space N(T) is a closed subspace of X, and hence the quotient space
XjN(T) = [X] is a Banach space with respect to the norm

II[x]11 =inf {llx-zll: z E N(T)}=d(x, N(T)),

where d(x, N(T)) denotes the distance from x to N(T). Since N(T) C D(T),
the quotient space D(T)/N(T) =D is contained in [X]. Defining [T] [x] =Tx
for every [x] ED, it follows that [T] is a well defined closed linear operator
with domain D([T])=D C [X] and range R([T])=R(T) C Y. Since [T]
is one-one, the inverse [T]-l exists on R(T). Note that [T]-l is closed.
Hence [T]-l is bounded if and only if its domain R(T) is closed. We
now define the number y(T) for the given operator T by

0 if R(T) is not closed,

y(T) = ( II[T]-lll-l ifR(T) is closed and T 9'=0,
+= if T=O,

where ° is the null operator. Equivalently, y(T) is the supremum of

all real y such that IITxll;;;.yll[x]11 holds for all x E D(T).
If the domain D(T) is dense in X, then T* exists. Since T* is closed,
the number y(T*) exists. The numbers y(T) and y(T*) are introduced
by T. KATO [8], who proved that y(T*) =y(T). This implies that R(T*)
is closed if and only if R( T) is closed. The first proof of the last fact is
probably due to G. C. ROTA [10]. Summarizing, we have the following
theorem.

Theorem 4.1. Let T be a closed linear operator with domain D(T)

dense in X. Then R(T) is closed it and only it R(T*) is clo8ed, and then
y(T*) =y(T) > o.
It is well-known that R(T) being closed implies that R(T*) is weak*
closed, and hence R(T*)=N(T)1.. Combining this fact with the results
in section 3 it is not difficult to prove that the following theorem holds.

Theorem 4.2. Let T be a closed linear operator with domain dense in

X and closed range in Y. Then rx(T)={3(T*) and (3(T)=rx(T*).
If T is closed and densely defined, then it follows from Theorem 3.1 (ii)
that rx(T) <(3(T*). The question may be raised whether rx(T) equals always
(3(T*). We will show by means of an example that rx(T) is not necessarily
equal to (3(T*).
Let X=Y=h, and let Tx be defined for x=(x1,x2, ... ) Eh by
y=TX=(yl, y2, ... ) with yn=n-1xn . Evidently, rx(T)=O. For any t E h*,
represented by rx=(rxl,rx 2, ... ) Eloo, we have T*t=({31,{32, ... ) with
(3n=n-lrxn, and so the range of T* is included in the subspace (co) of loo.
Hence (3(T*) = dim loo/R(T*);;;. dim looj(co) = + =. Summarizing, we have
 410
that T is a bounded linear operator on h such that IX(T) = 0 and
fJ(T*) = + 00, so these numbers are as far apart as they can be.
We proceed to present some sufficient conditions in order that a closed
linear operator has a closed range.

Theorem 4.3. Let T be a closed linear operator with domain in X and

range in Y such that tor every bounded closed set F C X the image ot
F (I D(T) is a closed subset ot Y. Then, tor any closed subspace L ot X,
the image ot L (I D(T) is a closed subspace ot Y; in particular, R(T)
is closed.
This theorem is a generalization of a well-known theorem about bounded
linear" operators (cf. N. DUNFORD and J. T. SCHWARTZ [5], p. 489). The
theorem has a partial converse, as follows.

Theorem 4.4. Let T be a closed linear operator such that IX(T) < 00.
Then the range R(T) is closed it and only it tor every bounded closed set
Fe X the image ot F (I D(T) is a closed subset ot Y.

Without the hypothesis that IX(T) < 00 the theorem does not hold,
not even if there exists a closed subspace Xo C X such that X =XoffiN(T).
By way of example, let X be a separable Hilbert space with a complete
orthonormal system {b o, bI, b2 , ••• }, and let T be the orthogonal projection
on the subspace Xo spanned by boo Then N(T) is the closed subspace
spanned by bI, b2 , ... , and X =XoffiN(T). Obviously, T is closed and
has a closed range. Let F be the sequence {bn+n-1bo; n=l, 2, ... }. Then
F is closed, since F consists of isolated points of mutual distance greater
than a positive constant, but the image of F is the sequence {n-1bo},
and this is not a closed set.
Let T be a closed linear operator with domain in X and range R(T)
in Y. Even if R(T) fails to be closed we can consider, algebraically, the
quotient space Y jR(T). Given this situation, the following theorem holds.

Theorem 4.5. Let T be a closed linear operator. It the dimension ot

Y jR(T) is finite, then R(T) is closed.
The result in this theorem is due to T. KATO ([8], Lemma 332). The
theorem derives its significance from the observation that if the Banach
space X satisfies X =XoffiZ, where Z is of finite dimension, then Xo
is not necessarily closed.
Finally, we present a theorem about closed linear operators with
non-closed range.

Theorem 4.6. It R(T) is not closed, and s> 0 is given, there exists
a sequence {Xn} C D(T) and also a sequence {tn} C N(T)J. such that
l=lltnll<lIxnll<l+s, /k(xn)=O tor k=l, 2, ... ,n-1,tn(xn)=1 and
IITxnll<2-n tor n=l, 2, ....
 411
5. BOUNDEDNESS WITH RESPECT TO A CLOSED LINEAR OPERATOR. Let
T be a closed linear operator with domain D(T) in X and range in Y.
The following definition is well-known.

Definition 5.1. The linear operator S with D(T) C D(S) C X and

R(S) C Y i8 8aid to be bounded with re8pect to T if there exi8t p08itive
con8tant8 x and fJ 8uch that IISxl1 <: xllxll + fJlITxll for all x EO D(T).
In this case, it is well-known that the linear operator T + AS with
domain D(T+AS)=D(T) is closed for IAI <fJ-1 (cf. B. SZ.-NAGY [11]).
It was also proved by B. Sz.-Nagy that, if the linear operator S with
D(T) C D(S) C X and R(S) C Y has a closed extension, then S is bounded
with respect to T.
Let S be bounded with respect to T. In the following sections we are
mainly interested in the properties of the null space and range of
T + AS for variable complex A. The following theorem shows the existence
of two bounded linear operators A and B from G(T) into Y such that
A + AB has, in many respects, the same behaviour as T + AS.

Theorem 5.2. Let S be bounded with re8pect to T, i.e., IISxll <:

xllxll + fJlITxll for all x EO D(T), and let X x Y be normed by lI(x, y)1I =
xllxll +fJllyli. Then the linear operator8 A and B from G(T) into Y, defined
by A(x, Tx)=Tx and B(x, Tx)=Sx, have the following propertie8.
(i) A and Bare continuou8; IIAII <:fJ-1 and IIBII <: 1.
(ii) R(A + AB) = R(T + AS) for all A.
(iii) x(A + AB) = x(T + AS) and fJ(A + AB) = fJ(T + AS) for all A.
(iv) y(A)=y(T)j{x+fJy(T)}.

6. STABILITY THEOREMS FOR A CLOSED LINEAR OPERATOR ONTO. Let

M and N be subspaces of the Banach space X. The number <5(M, N) is
now defined by
<5(M N) = { sup {d(x, N): x EO M, IIxll = I} for M =1= (0),
, 0 for M = (0).
Evidently, <5(M, N) <: 1. The opening ()(M, N) of M and N is defined by
()(M, N) =max {<5(M, N), <5(N, M)}.
In this section we need the following lemma.

Lemma 6.1. Let M and N be cl08ed 8ub8pace8 of the Banach 8pace X.

Then <5( M, N) < 1 implie8 that dim M <: dim N, and hence ()( M, N) < 1
implie8 that dim M = dim N.
All known proofs of this lemma depend on a theorem of K. BORSUK [1]
about antipode preserving continuous mappings. For proofs of the lemma
we refer to [4] and [6].
 412
Theorem 6.2. Let A and B be bounded linear operators from the Banach
space X into the Banach space Y, and assume that IIBII <; l. 11 R(A) = Y
and IA.I <y(A), then
(i) <x(A+AB)=<x(A),
(ii) R(A + AB) = R(A) = Y.

Proof. Without loss of generality, we may assume that X and Y

do not consist only of the null element.
(ii) By Theorem 3.1 (i) we have N(A *) =R(A)J.= YJ.= (0), and so
A * is a one-one bounded linear operator from y* into X*. In addition,
Theorem 4.1 shows that A * has a closed range (since A has a closed
range), and y(A *) = y(A) > o. From

II(A * + AB*) III;;;. IIA */11-1 AI·IIB*/II

;;;.y(A*) ·11/11-IAI·II/II=(y(A)-IAI) ·11111

it follows then that, for IAI <y(A), the operator A*+AB* is one-one
and has a closed range. Hence, once more by Theorem 4.1, A + AB has
a closed range, and by Theorem 3.1 (i) this range satisfies
R(A + AB) =J.{N(A * + AB*)} = J.{(O)} = Y.
Furthermore
y(A + AB) =y(A * + AB*) ;;;.y(A) -IAI.
(i) It will be proved first that, for IAI <iy(A), the opening of N(A)
and N (A + AB) does not exceed l. Indeed, if x E N (A), then

IAI·llxll;;;. IIABxl1 = II(A + AB)xll

;;;.y(A+AB)·d(x, N(A+AB))
;;;.(y(A)-IAI).d(x, N(A+AB)),

and hence
(1) b(N(A), N(A + AB)) <; IAI/(y(A) -IAI) < 1

for IAI<iY(A). Similarly, if x EN(A+AB), then

IAI·llxll;;;. IIABxll= IIAxll;;;.y(A) ·d(x, N(A)),
and hence
(2) b(N(A + AB), N(A)) <; IAl/y(A) < i
for IAI <iY(A). From (1) and (2) we obtain the desired result that
O(N(A),N(A+AB))<l for IAI<iy(A). By Lemma 6.1 this implies that
<x(A+AB)=<x(A) for IAI<iy(A).
Choose a fixed AO satisfying IAol=iy(A). Then R(A+AOB)=Y and
y(A+AOB);;;.y(A)-IAol=iy(A) by (ii). Hence the argument can be
repeated for A + AoB instead of A, and it follows that for all A satisfying
 413
IA-Aol<i(iy(A)) we have <x(A + AB)=<x(A + AoB). This is done for every
Ao satisfying IAol = iy(A), and we thus obtain that <x(A + AB) = <x(A) for
IAI<y(A).{i+(l-i)}. Proceeding by induction, we obtain a sequence
{Sn} with the properties that sl=i, Sn=sn-l+i(l-s n- l ) for n=2, 3, ... ,
and such that for all A satisfying IAI<Sn·y(A) we have <x(A+AB)=<x(A).
Since lims n =l, it follows that <x(A+AB)=<x(A) for IAI<y(A).
We note that the preceding theorem could be proved with a weak
form of Lemma 6.1. But then the proof requires more refined means
(cf. [7], section 13).
By means of Theorem 5.2, it is now easy to deduce from Theorem 6.2
the following stability theorem for closed linear operators onto.

Theorem 6.3. Let T be a closed linear operator with domain D(T) in

X and range R(T) in Y, and let S be bounded with respect to T, i.e ..
IISxl1 <<xllxll +PIITxll tor all x ED(T). It R(T)=Y and
(3) IAI <y(T)J{<x+py(T)},
then T + AS is a closed linear operator such that R(T + AS) =R(T) = Y and
<x(T + AS) = <x(T).
Note that if S is a bounded linear operator from X into Y, then S
is bounded with respect to T, and we may take <X = IISII and p arbitrarily
small positive (we cannot immediately take p = 0, since this would
invalidate Theorem 5.2 where <xllxll + Pllyll has to be a norm on X x Y).
Obviously, however, the final result in Theorem 6.3 holds now for all

IAI < y(T)J<x = y(T)JIISII·

It will be shown now by an example that without further hypothesis,
the bound for IAI in (3) cannot be improved.

Example 6.4. Let X = Y =h, and let (<Xl, <X2, ... ) be an enumerable
set of complex numbers dense in {A: IAI = I} and such that <Xi'F <Xj for
i=l=j. For X=(Xl, X2, ... ) E h, let Tx=y=(yl, y2, ... ) with yn=<Xnxn for
n= 1,2, .... Evidently, T is a one-one bounded linear operator onto,
and IITII =y(T) = l. Let I be the identity operator. If y= (T + AI)x, then
yn=(<Xn+A)Xn for n=l, 2, .... It is not difficult to verify that

(4) <x(T+AI)= {O for A~<Xi,~:l, 2, ... ,

1 for A-<Xi, ~-1, 2, ... .

It follows from Theorem 6.3 (cf. also the remark to this theorem) that
T + AI is a bounded linear operator from h onto h with <x(T) = <x(T + AI) = °
for all A satisfying IAI <y(T)JIIIII = l. This is in accordance with (4), and
it follows also that the constant 1 cannot be enlarged, since in every open
neighbourhood of Ao with IAol = 1 the function <x(T + AI) fails to have a
constant value. In addition, Theorem 6.3 shows that R(T + AI) =1= Y for
IAI = l. It is not difficult to prove that R(T + AI) is not closed for IAI = l.
27 Series A
 414
If in Theorem 6.3 the condition R(T) = Y is replaced by the weaker
condition that R(T) is closed, then it is not true in general that
cX(T+AS)=cX(T) for alllAI <y(T)/{cX+{3y(T)}, not even for alllAI <e where
e is an arbitrarily small positive constant.
Example 6.5. Let X = Y =h, and, for x= (xl, X2, ... ) E h, let Tx=
(yl, y2, ... ) with yl = 0, yn = x n+2 for n = 2, 3, .... Evidently, T is a bounded
linear operator with closed range R(T), which is a proper subspace of Y = h.
LetSxbedefinedfor x E h by y=Sx with yl=X 2_X 1 and yn=xn+2_xn+l
for n= 2, 3, .... It follows easily that S is a bounded linear operator
from X into Y. If now y = (T + AS)X, then
yl=A(X 2_X 1 ), yn=(1+A)xn+2-kn+1 for n=2, 3, ....
For A0;60 and A0;6 -1 we have x EN(T+AS) if and only if Xl=x2 and
x n+1 = A( 1 + A)-lXn for n = 3, 4, .... Hence cX(T + AS) = 1 for A0;6 - 1 and
IA(l+A)-ll;;;d, and cX(T+AS)=2 for A0;60, -1 and IA(l+A)-ll<l. It is
not difficult to verify now that
1 for Re A< -t,
(5) cX(T+AS)= ( 2 for Re A> -t, A0;6 0,
3 for A=O.
From (5) it IS evident that cX(T) = 3 > cX(T + AS) for all A0;6 0.
Leiden State University

REFERENCES
1. BORSUK, K., Drei Satze tiber die n-dimensionale euklidische Sphare, Fund.
Math. 20, 177-190 (1933).
2. BROWDER, F. E., Functional Analysis and partial differential equations I,
Math. Ann. 138, 55-79 (1959).
3. BROWN, A., On the adjoint of a closed transformation, Proc. Amer. Math.
Soc. 15, 239-240 (1964).
4. DAY, M. M., On the basis problem in normed spaces, Proc. Amer. Math.
Soc. 13, 655-658 (1962).
5. DUNFORD, N. and J. T. SCHWARTZ, Linear operators I, New York, 1958.
6. GOHBERG, I. C. and M. G. KREIN, The basic propositions on defect numbers,
root numbers and indices of linear operators, Uspekhi Math. Nauk
12, 2 (74), 43-118 (1957), (Russian) = Amer. Math. Soc. Transl.
(2) 13, 185-265 (1960).
7. KAASHOEK, M. A., Closed linear operators on Banach spaces, Thesis University
of Leiden, Amsterdam, 1964.
8. RATO, T., Perturbation theory for nullity, deficiency and other quantities of
linear operators, J. Analyse Math. 6, 261-322 (1958).
9. Rmsz, F. and B. SZ.-NAGY, LE'gons d'analyse fonctionnelle, 2 ed., Budapest,
1953.
10. ROTA, G. C., Extension theory of differential operators I, Comm. Pure Appl.
Math. 11, 23-65 (1958).
11. SZ.-NAGY, B., Perturbations des transformations lineaires fermees, Acta Sci.
Math. Szeged 14, 125-137 (1951).
12. TAYLOR, A. E., Introduction to functional analysis, New York, 1958.