I Integral Equations and Operator Theory
I Integral Equations and Operator Theory
Eugen J. Ionascu
I. PRELIMINARIES
We let 7-( be a separable, infinite dimensional, complex Hilbert space, and let
s denote the algebra of all bounded linear operators on 7/. If u, v C 7/, we shall write
u | v for the operator of rank one defined by
(uOv)x=<x,v>u, x~7/,
where <, > denotes the inner product of the Hilbert space 7/. The class Af of operators
T in s which can be written in the form T = N + (u | v), where N is a normal
operator and (u | v) ~ 0 is still not very well understood. Indeed, even the smaller class
of operators of the above form, where N is a diagonalizable normal operator, is not in a
much better situation, despite the structural simplicity of diagonalizable operators. In this
paper we are interested in this second class of operators which will be denoted simply by 79.
Some spectral properties, examples, applications and the equivalence between two known
results about the eigenvalues of rank-one perturbations and one-codimensional compressions
of selfadjoint compact operators are discussed in Section 2. We characterize those operators
in 7:) which are normal and prove that under mild assumptions they have the single value
extension property in Section 3. In Section 4 we give a characterization for an operator in a
relatively natural subclass of 7) to be a contraction. Finally in the last section, we combine
our previous results with known reductions that one would naturally make in dealing with
422 Ionascu
the invariant subspaee problem for this class of operators and give a sufficientcondition for
the existence of a non-trivial invariant subspace based on the Lomonosov's theorem [21].
Similar problems concerning operators in the class Af, or rank-one perturbations
of different classes of operators such as isometries, selfadjointcompact operators, selfadjoint
Toeplitz operators, shift restriction operators, cyclic operators, differentialoperators, (or
Volterra operator) have been studied in a series of papers of which we cite only a few of
them: [3], [4], [16]-[19], [22]-[25], [28]-[32]. It is worth mentioning that the class of rank-
one perturbations of bounded (or unbounded) selfadjoint operators has been extensively
studied and many interesting spectral properties have been established in various works
(see for instance [8]-[Ii], [14], [15], [29], [30]).
We let {en}n~176 denote an orthonormal basis for 7~ which will remain fixed through-
out the paper. We also let { A -}n=l
oo be an arbitrary bounded sequence of complex numbers
and throughout the remainder of the paper we shall write Diag({A~}) for the unique op-
erator D satisfying Den = A~e~, n E IN. We shall denote henceforth by 7:)o the subset of
L(~) consisting of all operators T which can be written in the form
(1) iF = Diag({)~n}) + u | v, u # O, v # O.
Throughout the paper we shall suppose t h a t u and v are nonzero vectors in ?t and their
expansions with respect to the (ordered, orthonormal) basis {en} are
Note that up to unitary equivalence, 7)o consists exactly of a11 sums N + 7~, where N is a
normal operator whose eigenvectors span ~ and R is an operator of rank one. Note also
that the inclusion T) C A f is a strict one. One way to see this is to make use of Kato and
Rosenblum's result (cf. [20]) stating that the absolutely continuous parts of a selfadjoint
operator and its selfadjointtrace class perturbation are unitarfiy equivalent.
Observe that the expression for T in (I) is not necessarily unique. If we restrict
our study, though, to the class 7)1 of those operators in :D o which admit a representation as
in (i) with u and v having nonzero components an and ~n for all n E IN, we have uniqueness
in the following sense.
PROPOSITION 1.1. I f T E 2)1 then the representation (1) for T is unique in the
sense that if T = Diag({A~}) § (u | v) = Diag({A$}) + (u' | v'), then Diag({An}) =
Diag({AS} ) and (u | v) = (u' | v').
PROOF. We may assume T = D i ~ g ( { ~ } ) + ( ~ | = D i ~ g ( { ~ ' } ) + (~'| where
all the Fourier coefficients of u and v in (2) are not zero. This means t h a t Diag({A~}) -
essentially different situations. If the range of S is (0) we are done. If the range of S is one-
dimensional--say, spanned by e~1, then since (u'| - (uev) would have a two-dimensional
range if {u, u'} and {v, v ~} are linearly independent sets of vectors, we get that either u
and u ~ are linearly dependent or v and v ~ are. Let us suppose that u and u I are linearly
dependent. Then u = ~nlen~ and u' -- fl=le=1. But this cannot happen since we have
assumed that < u, ek >~ 0 for all k E ]IN-. Similarly the case in which v and v ~ are linearly
dependent is ruled out. If the range of S were two-dimensional, then V{u, u ~} = V{en~, e~2},
and again we would have a contradiction. []
2. SPECTRAL PROPERTIES
The next two propositions show that when looking for nontrivial invariant subspaces for
operators in D0, one can then restrict his attention to the subset 7) 2 of :DI consisting of
those operators T = D + (u @ v) in T) I such that D has uniform multiplicity one (i.e., if
D = Diag({A~}), then all of the numbers An, n E IN, are pairwise distinct).
PROPOSITION 2.1. Suppose T = Diag((An}) + (u @ v) E Do is not a normal
operator, and for some no E ]N, a~o = 0 or fl~o = O. Then T* [resp. T] has point spectrum
and T and T* have nontrivial hyperinvariant subspaces (n.h.s).
PROOF. In case ~no =< u, | > = 0 , we have
T* eno = -A~oeno + (v | u)e~ o = -A~oeno+ < e~o, u > v = -~noeno,
which shows that ap(T*), the point spectrum ofT*, is nonempty, and since T* is non-normal,
the eigenspace associated with ~ o is a n.h.s, for T*. Its orthogonal complement is thus
hyperinvariant for T. The case j3~o = 0 is handled similarly. []
For a diagonal operator D -- Diag(~n}) we denote by A(D) the set of all its
eigenvahies A=.
PROPOSITION 2.2. If T = D + (u | v) E D1 and at least one A E A(D) has
multiplicity larger than 1, then T has A in its point spectrum.
PROOF. Suppose A -- A,,o -- A,u, no ~ nl. Then ( T - A)e),~o - - < e;~o ,v >
u = ~ o u, and (T - A ) e ~ = < e ~ , v > u = ~lu. Hence, if fl~o ~ 0 and/?n~ ~ 0 then
(T - # ) ( & ~ e ~ ~ - & o e ~ ) -- 0. In any case T - A is not injective, and then A E ~rp(T). []
For an operator T C ~1 given by (1), an interesting phenomenon happens with the
isolated eigenvalues of Diag(A~): they are not in the spectrum of T. The following theorem
gives necessary and sufficient conditions for a point # in or(D) (T -- D + (u | v) E Do) to
be in ~(T) (resolvent set).
THEOREM 2.3. Suppose we have T ---- D + (u | v) E :Do and # E or(D). Then
# E ~(T) if and only if the following two conditions are satisfied:
(i) # is an isolated eigenvalue of D, A~o, of multiplicity one,
(ii) ~ o - - < v, e~o > ~ 0 and ~ o --< u, e~o > ~ O.
424 Ionascu
PROOF. For the necessity part of this theorem, let us assume first that (i) is not
satisfied. We have three cases: (a) # is not an eigenvalue; (b) # is an eigenvaiue but is not
isolated, and (c) # is an isolated eigenvalue but has multiplicity larger than I. In the cases
(a) and (b), there exists a sequence of distinct eigenvalues {An~}k>1 such that Ank --~ #.
Then, since (T - # ) e ~ = (~nk - # ) e ; ~ + < e;~k , v > u we have
(iii) ~k r A ( D ) ,
(iv) z ~ ~ < oo, a,~
(v) E;,.eA(D) :,_~,.. = 1.
PROOF. For the necessity part, let A E 9 be an eigenvalue for T and x ~ ~ \ {0},
such t h a t T x = )~x. T h e n < x, v > u = (A - D ) x . We cannot have < x, v > = 0 because
we o b t a i n then )~ = A~~ x = ~ % , ~ E ~ \ {0}, and then -~io
-
=< %,v > = ~1 < x , v > = 0
which is not possible since T E 19~. Hence, if we write ~ = - < ~,~>x, t h e n u = (D - A)~
and < ~ , v > + l = 0 .
For the sufficiency part, we can assume t h a t there exists x ~ ?f such t h a t u =
(D-zk)xand<x,v>+l=0. Thenx#0andTx=Dx+<x,v>u=u+)~x-u= )~x.
Ionascu 425
(3) s (z) = z
n=l
Also we need the following formula which can be proved easily by a change of variables to
polar coordinates:
f/ dx dy { ~---2-2
~-a ' if Iz-al < r ,
(5) D(a,r) Z - - ( x + y i ) = 7r(~--~), if Iz - a I_<r,
for every a E C and r > 0. Then if z ~ ID, by (3) and (5), we have
fT(z) =
k~ rk21~/~ d x d y k ~
z _U)~k -= -~ z -- (X + yi)
_ ~//dxdyl_ z -- (x + yi) z'
D( k, k) ~)
Hence, by Proposition 2.4, T does not have any eigenvahes z E 9 Let us suppose that
z C D \ U~e~ D~. In this case if z were an eigenvalue for T then by Proposition 2.4, the
sum ~ ._5L would be absolutely convergent and it would be equal to 1. But using again
kE~lz--.k~
(5), we have
2
z -- ~k
1
~r
~J'.r dxdy
z - (x + yi)
_ ffdxdy z - (x + yi) - ~"
D( ~, k) ID
This implies that the only possible point which may be an eigenvalue is z -- 1. In fact, under
our hypothesis, z -- 1 is indeed an eigenvalue because v" _ 5 ~ < ~ e ~ r~ < oc. Suppose
L_, 11-),~1
kE1N
z E Dn\{A,~} for some n E IN and let us assume that z is an eigenvalue for T. Then using
(5) again we can compute
2 1 ff dx dy r~ 2 2
z Z-~k ~rke ,
~
7" n D ( ) ~ , r ~ )
z -- (x + yi) -k .
z -- k~. . . z - ,~,~ k~ +--.z - k~
This shows that z = k~ + 1 - ~ is the only possible eigenvalue for T in this case. In fact, it
is easy to see that these values are indeed eigenvalues
.
for T. Hence, %(T) = {An + ~
1 --k=
:
n E ~N} U {1}.
A natural question which arises at this point is whether or not there exist op-
erators T E 9 with empty point spectrum, An example of such an operator was first
constructed by J. G. Stampfli in [31], for the case when the spectrum of T is a square.
Given an arbitrary nonempty compact subset of the plane/(, it is interesting to know if
there are examples of operators T E :Do with empty point spectrum and such that o-(T) = K.
Next, we put together some information about the resolvent of operators T in 2P2 around
points which are isolated in A(D).
LEMMA 2.7. Let A E L ( ~ ) be an invertible operator, and let S = A + (u
v). @
Then S is invertible if and only if < A-iu, v > +1 ~ 0, and its inverse is given by the
formula
(6) S_ ~ = A_ ~ _ 1 (d_,u | (d,)_~v).
< A-lu, v > +1
Ionascu 427
(9)
OZ -- " --I O~k-~k
k#no
where D : ~kr -- r k | ek.
PROOF. Formula (8) can be easily derived from (7). Each r e A ( D ) \ A ( D ) ' is an
isolated eigenvalue of multiplicity one for D, and hence by Theorem 2.3, T - r is invertible.
We have r - D = r - T + (u | v) and then by Lemma 2.7, < (r - T)-~u, v > +1 : 0, which
proves t h a t FT(r = --1. To compute dFr {,~ we differentiate (8) at a point z different of
dz ~w
and take the limit as z --+ r
(z-OVa(z) =_
-~z = (1 - fT(z)) 2 [z -- r -- (z -- r 2
The equality (9) follows from (7) by a similar argument of passing to the limit as z ---* r []
As an application to formula (9) we will show the equivalence of two interesting
facts from the theory of selfadjoint compact operators. The first result appears in [32] (see
also [16]) and the second result was proved independently by several authors (cf. [7], [13]
and [26]).
428 Ionascu
THEOREM 2.9. (i) Let {~k}ke~ and {Pk}ke~ be two distinct monotone in-
creasing sequences of real numbers, each having zero as the limit point. Further assume
that (pk} belongs to (Pk, ~k+l) for each k E IN. Then if A is a selfadjoint compact op-
erator on a separable Hilbert space H having the sequence ~ (k ~ IN) as its eigenvalues
(with multiplicity one), there exists a vector x ~ H such that A + x | x has precisely
the eigenvaIues {#~}ke~.
(ii) Let { ~ } ~ and { # ~ } k ~ be two distinct monotone decreasing sequences of real num-
bers, each having zero as the limit point and such that {p~} belongs to (~+1, ~'~) for
each k ~ IN. Then if A is a selfadjoint compact operator on a Hilbert space H having
the eigenvalues ~k (k ~ IN) (with multiplicity one), there exists a vector y ~ H such
that if P denotes the orthogonal projection on the one-dimensional space spanned by the
vector y, the compact operator (I - P ) A ( I - P)I(Z-P)(H) has exactly as its eigenvalues
the. sequence {#~}~e~.
PROOP. For the implication (i)=>(ii)we assume that {~'k}k~, {#~}k~ and A
are as in (ii)and let us take the diagonal operator D on H whose eigenvalues are {,~k}k~
where %; = - i , Ak+l = (l+#k) - 1 - 1 f o r k E ]hi. Then by (i) we can findx suchthat
T = D + x | x has exactly the eigenvalues {(1 + ~k) -1 - 1 } k ~ . We take ~ = 51 and
apply formula (9) for D, u = v = z and no = 1. Let Q be the orthogonal projection
on el. We see that (I - Q ) ( T - ( i ) - i ( i _ Q)I(• is a diagonal whose eigenvalues are
precisely {~-;~_;}k>2
i = { 1 + #k}~e~. Hence, by spectral mapping theorem the operator
S = (T - r _ I is compact and has the eigenvalues {L'k}ke~. Thus, we can find an
unitary operator U such that U*SU = A. To finish the proof we take y = U*ei and observe
that (I - P ) A ( I - P) = U*(I - Q ) S ( I - Q)U, where P is the orthogonal projection on the
one-dimensional space spanned by y.
For the implication (ii)~(i), let {~k}k~m, {#k}ke~ and d be as in (i). Without loss
of generality, we can assume that A is a diagonal operator with respect to the basis {ek}ke~
and ~i = - I . Let B be an arbitrary compact operator on H which has {(#k + 1) - i - 1}ke~
has its only eigenvalues (multiplicity one). Using (ii) we can find y :-- Yi E H such that
( I - P ) A ( I - P ) I ( f _ p ) ( H ) has precisely { ( u k + i + l ) - i - 1}ke~ as its eigenvalues. Let {Yk+i}k~
be an orthonormal basis in (I - P ) ( H ) with respect to which (I - P ) A ( I - P)[(Z-P)(H)
diagonalizes. Then the matrix of B + I with respect to the basis {Yk}kc~ looks exactly as
the right hand side of (9) (for D = A, Ak = ~k (k E IN), 4 = ~'1, u = v and e,~0 = 1).
We shall show that we can determine the coefficients of u such that these two matrices
coincide (which will give a unitarily equivalence between the operators which admit this
same representation matrix in different orthonormal basis). Let us write the representation
Ionascu 429
of B as follows
(10) 1 = bl - ~ ( ~ k + 1)lbkl=.
15~12 k>2
This will allow us to solve for 51 if the right hand side of (10) is not zero. Suppose by way
of contradiction t h a t this is not true. Then a simple c o m p u t a t i o n shows t h a t (B + I ) z = 0
where z = Yl - ~k>2(uk + 1)b-~kand so B + I admits the value 0 as one of its eigenvalues
but by our assumption the only eigenvalues of B + I are the elements o f the sequence
{(#k + 1)-l}ke~. This proves t h a t we have a solution for u e H and so b y spectral theorem
A + u | u has precisely the eigenvalues { # k } k ~ . 9
(ii) u, v are linearly independent vectors and there exist 5, fl E 9 such that
01)
where ~ ( ~ ) = - 1 / 2 .
PROOF. We observe t h a t the equation T * T = T T * is equivalent to
N*u | v + v | N * u + Ilull2v | v =
(12)
N~|174 I1~11~ | ~.
It is a simple computation to check t h a t (12) is satisfied if (i) or (ii) is true.
Let us assume t h a t T is a normal operator. We distinguish two distinct cases.
CASE I. We assume t h a t u, v are linearly dependent. Thus, there exists 5 E ~ such t h a t
u = 5 v (5 = < u , v > / l l v l l 2 ) . Since Ilvll2u | u = lal2llvll2v | v = Hul]2v | V, if we write
w = (aN* - N Y ) v ( = 2.~(hN*)v), (12) becomes w | v = - v | w. This last equality holds
if and only if w = itv for some t E ]R and (i) is proved.
CASE II. We assume t h a t u, v are linearly independent vectors. From (12) we get t h a t
Hence < iV*u, x > = < N v , x > = 0 for every x E (V{u, v}) • which means that
for some a/j E C. Substituting in (12) we obtain that the aij satisfy the following relations:
(b) there exist a E 9 and t E IR such that A(D) lies on the circle {z e (~ : Iz - a] = t},
t E ]R, and
tulllu[I = e ~ ( D - oI)(v/llvll),
where ~ E [0, 7r) is determined by the equation ~(te~O/llull Ilvll) = -1/2.
PROOF. Suppose that (a) or (b) holds. Then either .~(aD*) = t I or ID-~Zl = tz.
If (a) holds then (i) in Proposition 3.1 holds and hence T is a normal operator. If (b) holds
then an easy computation shows that (11) holds for/3 = teiO/l[ul[ Ilvll. The two relations in
(11) alone imply that (12) holds and so T is normal.
On the other hand if T is normal then, by Proposition 3.1, (i) or (ii) holds. In case
(i) is true then ~ ( a D * ) u = tu for some t E IR. Thus ~'(o~-~oe~) = tan for all n E IN and since
a~ r 0 for every n in IN we obtain that A(D) is a subset of the line {z E (l:J : .~(~g) = t} and
(a) follows. If (ii) holds, we get from (11) that ( D - c ~ I ) * ( D - a I ) v = IMl~llvll21~l%, and by
a similar argument as above, we get that A(D) is a subset of the circle {z E e : I z - a I = t},
where t = Ilull Ilvlll~l. Then, the other part of (b) follows easily from (11). []
It is worth mentioning that actually if A(D) is a subset of a line or of a circle
then T = D + u | v is a decomposable operator (el. Theorem 5.2, [5]). Moreover, T has the
property (Triang0) (of. Theorem 6.16, [5]), i.e., for any pair S~ C & of invariant subspaces
for T such that dim($1/S2) > 1 there exists another invariant subspace $3 of T verifying
Another interesting question about the class :Do is whether we have the decompos-
ability property for operators in :Do whose spectrum is not necessarily an arc of an analytic
curve. It is known ([5]) that every decomposable operator has the following property.
Ionascu 431
DEFINITION 3.3. We say that an operator T E s has the single valued ex-
tension property (notation: SVEP) if the only vector-valued analytic function f : O --+ ~ ,
where G is an arbitrary open connected subset of (B, which satisfies the equality
( T - z I ) f ( z ) : O, z E G,
is the function identically equal to zero.
PROPOSITION 3.4. Every operator T : D + ( u | E lPi for which the set(B\A(D)
is connected has the SVEP.
PROOF. Let f : G ~ 7-I be an analytic function such t h a t (T - z I ) f ( z ) : 0
for every z E G. If G F] ((I] \ A(D)) # ~ then by Corollary 2 . 5 , T - z I is invertible for
all z E ( G \ A ( D ) ) \ { z E (B\A(D);fr(z) = 1} and so f ( z ) = 0 for all z E ( G \ A ( D ) ) \ { z E
~]\A(D); fT(z) = 1}. The function fT cannot be identically equal to 1 on the connected set
9C \ A(D) because limlzl_~co fT(z) = 0. Hence the set {z; fT(z) = 1} is discrete arid since G
is connected it follows t h a t f is identically zero.
We m a y assume t h a t actually G C A(D). If we expand f in the basis {en} as
f = ~,~~176
1 fne,~, where fn : G --+ C are scalar-valued analytic functions, we get
(14) (A~ - z ) f ~ ( z ) + < f ( z ) , v > ~n = o, z e a, ~ 9 ~.
If we take z = , ~ 9 G fl A(D) in the above equation, we obtain t h a t < f(A~), v >= 0 for
all A~ 9 G F1 A(D). Since the set A(D) is dense in A(D) and G C A(D), the set G F1 A(D)
is clearly dense in G. Hence < f(z), v >= 0 for all z 9 G. Thus (14) implies t h a t for every
integer n 9 ]iN, f~(z) = 0 for all z 9 G\A(D). Since each fn is a continuous function and
G \ A ( D ) is dense in G, it follows t h a t f,, is identically equal to zero on G for every n 9 ]N
and so is f . []
4. CONTRACTIONS IN ~)0(]D)
In this section we consider the class 7)0(]D) of the operators T--- D+u| 9 2Po for
which A(D) C ~. In this section we will characterize the contraction operators in T)0(ID).
The following proposition provides one such characterization and leads us to Corrolary 4.3
which gives a simple sufficient condition for an operator T 9 :Do(ID)MT~2 to be a contraction.
PROPOSITION 4.1. T = D + u | 9 :Do(]]:)) is a contraction operator if and only
g
1 ,~=~
~ (1 - s ak ~) > s (k__~l(1 - slakl~)) (k~__~l(], _ slAkl=)), s 9 (0, ].).
432 Ionascu
We apply Lemma 2.7 for the case A = D*D - tI, a = D*u + l]u]12v, b = c = v,
and d = D'u, where t E JR, t > 1. Hence, T*T is a contraction if and only if the determinant
of the matrix
i+ ti)-l(D. + > < (D'D- tI)- v, > ]
< (D D - t I ) - l ( D * u + II~IP~), D*u > i + < (D*D - tI)-=v, D*u > J
Ionascu 433
equals zero for no t e (I, oo). If we multiply the second column of this matrix by INI 2 and
subtract it from the first column, the determinant is the same as the determinant of the
resulting matrix
[ l+ < (D*D-tf)-lD*u,v > < (D*D-tI)-iv, v > ]
< (D*D - t I ) - l D * u , D*u > -IluI[ 2 1+ < (D*D - t I ) - % , D*u > "
The (2, 1) entry can be written differently as follows:
< (D*D - t I ) - l D * u , D*u > -Ilu]] 2 = < (D*D - t I ) - l D * D u , u > - I [ u [ [ 2 =
< (D*D - t I ) - l ( D * D - tI)u, u > -Ilu]l 2 + t < (D*D - t I ) - l u , u >
= t < (D*D - tI)-lu, u >.
If we observe that the (1, 1) entry is the complex conjugate of the (2, 2) entry, we obtain
that T * T is a contraction operator if and only if the equation (in t)
]1+ < (D*D - t f ) - l D * u , v > 12 - t < (D*D - t I ) - l u , u > < (D*D - t I ) - l v , v > = 0
has no solution in the interval (1, oo). Finally, if we change variables by setting s = 1/t,
s E (0, 1), the above equation becomes
[ 1 - s < ( I - s D * D ) - l D * u , v > [2
< (I- sD*D)-lu, u >< (I- sD*D)-lv, v >
which implies (15) since both members of the above equMity are continuous functions of s
and the sign of the inequMity is determined when s = 0. The inequality (16) follows form
(15) taking into account the explicit form of the operator D. []
COROLLARY 4.3, Assume that for T = D 4- (u | v) C 2Po(]D) rh'D2 the coordinates
of u and v satisfy the inequality
First we choose a~ = (maxl<k<. [suP.ex [pk(z)12)]) -1 for each n E IN, and let then
= + ~ = 2 ~a,~_16~. Clearly, al _> a2 >_ ... _> a~, > 0. It is easy to observe that (a) is
satisfied, and in order to check the second property we take r = a~-i if z = ),~, n > 2,
l if z = A1 and zero anywhere else. Hence, 0 <_ r < max{a1, 1} = a0 for every z ~ X.
To check the third property, we want to show that p~ converges in L2(X, u) to the function
z ---+-2:
s Is-p~(z)[2S.(z) = s Is-p~(z)?S.(z) <_2 s jsl'd.(z)+
by our assumption. This proves that q~(M~)r V2 converges to f in L2(X, 7) which finishes
the proof. []
Let us observe that if T = D + (u | v) G D2 we have ~r(T) = ~r(D), and hence,
since ~r(D) is normal in the Calkin algebra, we have that ate(D) = ~r,e(D) and consequently
~ . ( T ) = ~..(T) = ~.(T) = a . ( D ) = A(D)'. Hence, well-known reductions of the invariant
subspace problem (see [5] for part (iii)) applied to our particular case and together with
what we have proved so far give the following proposition.
PROPOSITION 5.3. If T = D + (u | v) E D2, and
(i) cr(T) # A(D)' (equivalently 1 e fTI(~\-A)), or
(ii) A(D)' is not connected, or
(iii) A(D)' is a singleton, or
(iv) ~[~esp. ~] is not cyclic fo~ 9[~esp. D*],
then T has a n.h.s.
436 Ionascu
When one searches for invariant subspaces for an operator T it is useful to have
a description of its commutant {T}' := {A E Z:(7-L) : AT = TA}.
PROPOSITION 5.4. Let T = D + (u | v) C 1)2, and A E s Then A E { T y
if and only if there exist a sequence of complex numbers {t,~}ne~ and a positive constant C
such that
(i) for every square-summable sequence {~k}k_>l we have
(21) E f J- E <_c E
n k>_l,k~n n
(23) = - E k e
n>l,nek
is a bounded sequence.
PROOF. The equality A T = T A can be written equivalently as
which implies that we can express Aek as in (22). Taking the inner product of both sides of
(22) with v, we obtain that sk is given by (23). To obtain the inequality (21) we first need
to observe that sk = < Aek, ek > (by (22) and so {sk} is a bounded sequence. Thus, the
inequality (21) follows easily from the boundedness of the operator A - D, where D is the
diagonal operator defined by Dek = s~ek, k G IN.
For Sufficiency, we observe that the linear operator A defined by (22) is bounded
because of (21) and the hypothesis that {sk} is bounded. Then from (22) and (23) we get
that < ek, A*v > = tkfl--k and Au = ~k~176tkakek. Using these two relations and (22), we
obtain (25) which is equivalent to (24). []
Next we would like to combine Proposition 5.4 with Lomonosov's theorem (cf.
[21]) to obtain sufficient conditions for existence of n.i.s, for operators in /)2. For this
Ionascu 437
purpose we introduce some more notation. Let 7-L(U) be the set of analytic functions on
the open set U(C ~). For a fixed w E U we define a linear transformations on ~(U),
r -. r(r by
r162 -r ~ ~ .
(26) r(r =
i
l r if z = ~,
z e u, eer(u).
For T E 7)2 given by (I), and U such that A(D) C U we define another linear transformation
on ~(U) by
where C = c2]lu][2Hvil2. This proves that inequality (21) is satisfied. Also, the sequence
defined by (23) is bounded since {tn} is clearly bounded and for every k E IN
Then, by Proposition 5.4, A O commutes with T. From (23), for every k C IN we have
s~ = r - ~ ~Z~--~,. = r - B~(r + ~Zr
n>l,n~:
438 Ionascu
The assumptions on r allows one to do the computations in (28) and (29). Moreover, if
rr = %b, comparing (28) with (29) we have BT(r = %b(z) for z e ]D, and so we can
apply Theorem 5.5 to conclude the corollary. []
Ionascu 439
REFERENCES
[1] N. Benamara and N. Nikolski, Resolvent tests for similarity to normal operators, Proc. London Math.
Soc. 78(1999), 585-626.
[2] L. Brown, A. Shields and K. Zeller, On absolutely convergent exponential sums, Trans. Amer. Math.
Soc., 96(1960), 162-183.
[3] K. Clancey, Seminormal operators, Lecture Notes in Math., vol. 742, Springer-Verlag, New York, 1979.
[4] D. Clark, One-dimensional perturbations of restricted shifts, J. Analyse Math. 25(1972), 169-191.
[5] I. Colojoara and C. Foias, Theory of generalized spectral operators, Science Publishers, New York, 1968.
[6] J. B. Conway, A course in functional analysis, Springer-Verlag, New York, 1985.
[7] C. Davis, Eigenvalues of compressions, Bull. Math. de la Soc. Sci. Math. Phys. de la R.P.Roumaine
(N.S.), 3(51) (1959), 3-5.
[8] R. Del Rio, S. Jitomirskaya, Y. Lasta, and B. Simon, Operators with singular continuous spectrum. IV.
Hausdorff dimensions, rank one perturbations, and localization, J. Anal. Math. 69(1996), 153-200.
[9] R. Del Rio, N. Makarov and B. Simon, Operators with singular continuous spectrum. II. Rank one
operators, Comm. Math. Phys. 165(1994), 59-67.
[10] R. Del Rio and B. Simon, Point spectrum and mixed spectral types for rank one perturbations, Proc.
Amer. Math. Soc., 125(1997), 3593-3599.
[11] W. F. Donoghue, On the perturbation of spectra, Comm: Pure, Appl. Math. 18(1965), 559-579.
[12] J. Eschmeier and M. Putinar, Bishop's condition (~) and rich eztensions of linear operators, Indiana
Univ. Math. J., 37 (1988), 325-348.
[13] K. Fan and G. Pall, Imbedding conditions for Hermitian and normal matrices, Canad. J. Math. 9(1957),
298-304.
[14] S. Hassi and H. de Snoo, On rank one perturbations of selfadjoint operators, Integral Equations
Operator Theory, 29(1997), 288-300.
[15] S. Hassi, H. de Shoo, and A. Willemsma, Smooth rank one perturbations of selfadjoint operators, Proc.
Amer. Math. Soc. 126(1998), 2663-2675.
[16] H. Hochstadt, One dimensionaIperturbations of compact operators, Proc. Amer. Math. Soc. 37(1973),
465-467.
[17] G. Islamov, Properties of one-rank perturbations, Izv. Vyssh. Uchebn. Zaved. Mat. 4(1989), 29-35.
[18] W. Johnston, A condition for absence of singular spectrum with an application to perturbations of
selfadjoint Toeplitz operators, Amer. J. Math. 113 (1991), 243-267.
[19] V. Kapustin, One-dimensional perturbations of singular unitary operators, Zap. Nanchn. Sem. S.-
Peterburg. Otdel. Mat. Inst. Steklov. (POMI) 232 (1996), 118-122, 216.
[20] T. Kato, Perturbation theory for linear operators, Springer-Verlag, New York, 1976.
[21] V. Lomonosov, On invariant subspaces of families of operators commuting with a completely continuous
operator, Funk. Anal. i Prilozen 7(1973), 55-56 (Russian).
[22] N. G. IvIakarov, One-dimensional perturbations of singular unitary operators, Acta Sci. Matt~. 42(1988),
459-463.
[23] N. G. Makarov, Perturbations of normal operators and the stability of the continuous spectrum, Izv.
Akad. Nauk SSSR Ser. Mat. 50 (1986), 1178-1203, 1343.
[24] M. Malamud, Remarks on the spectrum of one-dimensional perturbations of Volterra, Mat. Fiz. No. 32
(1982), 99-105.
[25] Y. Mikityuk, The singular spectrum of selfadjoint operators, (Russian) Dokl. Akad. Nauk SSSt~ 303
(1988), 33-36; translation in Soviet Math. Dokl. 38 (t989), 472-475.
[26] L. Mirsky, Matrices with prescribed characteristic roots and diagonal elements, J. London Math. Soc.
33(1958), 14-21.
[27] B. Sz.-Nagy and C. Foias, Harmonic analysis of operators on Hilbert space, American Elsevier, New
York, 1970.
[28] Nakamura, Yoshihiro, One-dimensional perturbations of the shift, Integral Equations Operator Theory
3(1993), 337-403.
[29] B. Simon, Spectral analysis of rank one perturbations and applications. Mathematical quantum theory.
IL Schrdinger operators, CRM Proc. Lecture Notes, 8 Amer. Math. Soc., Providence, RI, (1995),
109-149.
440 Ionascu
[30] B. Simon, Operators with singular continuous spectrum. VII. Examples with borderline time decay,
Comm. Math. Phys. 176(1996), 713-722.
[31] J. G~ Stampfli, One-dimensional perturbations of operators, Pacific J. Math., 115(1984), 481-491.
[32] H. Vasudeva, One dimensional perturbations of compact operators, Proc. Amer. Math. Soc. 57 (1976),
58-60.
[33] J. Wolff, Sur les sgries ~ A k / ( z - ak), C. R. Acad. Sci. Paris, 173 (1921), 1057-158, 1327-1328.