a

r
X
i
v
:
1
3
0
6
.
6
6
7
8
v
1


[
m
a
t
h
.
F
A
]


2
7

J
u
n

2
0
1
3
Invertible extensions of symmetric operators and
the corresponding generalized resolvents.
S.M. Zagorodnyuk
1 Introduction.
Let A be a closed symmetric invertible operator in a Hilbert space H. The
domain of A is not supposed to be necessarily dense in H. Let

A be a self-
adjoint extension of A, acting in a Hilbert space

H H. Recall that an
operator-valued function R

, given by the following relation:

R

= R

(A) = R
s;
(A) = P

H
H
_

A E

H
_
1
|
H
, R
e
,
is said to be a generalized resolvent of the symmetric operator A (corre-
sponding to the extension

A). Fix an arbitrary point
0
R
e
. An arbitrary
generalized resolvent R
s;
of the operator A is given by Shtrauss formula:
R
s;
=
_
_
_
_
A
F()
E
H
_
1
,

0
_
A
F

()
E
H
_
1
,

0
, (1)
where F() is a function from S
a;
0
(

0
; N

0
(A), N

0
(A)). Conversely, an
arbitrary function F() S
a;
0
(

0
; N

0
(A), N

0
(A)) denes by relation (1)
a generalized resolvent R
s;
of the operator A. Moreover, for dierent
functions from S
a;
0
(

0
; N

0
(A), N

0
(A)) there correspond dierent gen-
eralized resolvents of the operator A. By S
a;
0
(

0
; N

0
(A), N

0
(A)) =
S
a
(

0
; N

0
(A), N

0
(A)) we denote a set of all operator-valued functions
F() S(

0
; N

0
(A), N

0
(A)), which are
0
-admissible with respect to
the operator A. Throughout this paper we shall use the notations from the
survey paper [1].
Our main aim here is to characterize those generalized resolvents R
s;
,
which are generated by an (at least one) invertible self-adjoint extension.
Such generalized resolvents we shall call generalized I-resolvents. A gen-
eralized I-resolvent will be described by Shtrauss formula (1), where the
parameter F() satises a boundary condition at 0. As a by-product, using
the generalized Neumanns formulas we describe all invertible extensions of
a symmetric operator inside H.
Notations. As it was already mentioned, in this paper we use notations
from [1].
1
2 Invertible extensions.
Let A be a closed symmetric invertible operator in a Hilbert space H. Let z
from C

(C
+
) be a xed number. Recall that the following formulas (see [1,
Theorem 3.13])
D(B) = D(A) (T E
H
)D(T), (2)
B(f +T ) = Af +zT z, f D(A), D(T), (3)
establish a one-to-one correspondence between all admissible with respect to
A isometric operators T, D(T) N
z
(A), R(T) N
z
(A), and all symmetric
extensions B of the operator A. We have
D(T) = N
z
(A) R(B zE
H
), (4)
T (B zE
H
)(B zE
H
)
1
. (5)
Formulas (2),(3) dene a one-to-one correspondence between all admissible
with respect to A non-expanding operators T, D(T) N
z
(A), R(T)
N
z
(A), and all dissipative (respectively accumulative) extensions B of the
operator A. Relations (4),(5) hold in this case, as well.
Consider the Cayley transformation of the operator A:
U
z
= U
z
(A) = (AzE
H
)(A zE
H
)
1
= E
H
+ (z z)(A zE
H
)
1
. (6)
The operator B may be also determined by the following relations:
W
z
= (B zE
H
)(B zE
H
)
1
= E
H
+ (z z)(B zE
H
)
1
, (7)
B = (zW
z
zE
H
)(W
z
E
H
)
1
= zE
H
+ (z z)(W
z
E
H
)
1
. (8)
W
z
= U
z
(A) T. (9)
It is readily checked that
M

(A) = M1

(A
1
), N

(A) = N1

(A
1
), R
e
; (10)
U

(A) =

U1

(A
1
), R
e
. (11)
Theorem 1 Let A be a closed symmetric invertible operator in a Hilbert
space H, and z R
e
be a xed point. Let T be an admissible with respect
to A non-expanding operator with D(T) N
z
(A), R(T) N
z
(A). The
following two conditions are equivalent:
2
(i) The operator B, dened by (2) and (3), is invertible;
(ii) The operator
z
z
T is
1
z
-admissible with respect to A
1
.
Proof. From relation (8) it follows that B is invertible if and only if the
operator
z
z
W
z
has no non-zero xed elements. By (9),(11) we see that
z
z
W
z
=
z
z
U
z
(A)
z
z
T = U1
z
(A
1
)
z
z
T.
It remains to apply Theorem 3.6 from [1].
Remark 1 By the denition of an admissible operator condition (ii) of the
last theorem is equivalent to the following condition:
(iii) the operator T
z
z
X1
z
(A
1
) is invertible,
where X

(A
1
) is the forbidden operator for A
1
.
Theorem 2 Let A be a closed symmetric invertible operator in a Hilbert
space H. Suppose that A has nite defect numbers. Then there exists an
invertible self-adjoint operator

A A in a Hilbert space

H H.
Proof. Let the operator A from the formulation of the theorem have the
deciency index (n, m), n, m Z
+
, n + m = 0. Set H = H H and
A = A(A). The closed symmetric operator A has equal defect numbers.
Fix an arbitrary number z R
e
. Let {f
k
}
n+m
k=1
be an orthonormal basis in
N
z
(A). Set
T(f
1
) = h, C,
where h N
z
(A) is an arbitrary element such that h
H
= 1, and h =
z
z
X1
z
(A
1
)f
1
if f
1
D(X1
z
(A
1
)); h = X
z
(A)f
1
if f
1
D(X
z
(A)). The
operator T with the domain span{f
1
} is z-admissible with respect to A
and condition (iii) holds for A. By Theorem 1 the symmetric extension B,
corresponding to T, is invertible. The defect numbers of B are equal to
n +m1. If B is not self-adjoint, we can take B instead of A in the above
construction to obtain a closed symmetric extension with the deciency
index (n + m 2, n + m2). Repeating this procedure we shall construct
a self-adjoint invertible extension of A.
3 Generalized resolvents.
Consider a closed symmetric invertible operator A in a Hilbert space H.
Choose and x an arbitrary point
0
R
e
. A function F() S
a;
0
(

0
; N

0
(A), N

0
(A))
3
is said to be
0
-I-admissible with respect to the operator A, if the
validity of
lim

0
, 0
F() =

0

0
X 1

0
(A
1
), (12)
lim

0
, 0
_
1
||
(
H
F()
H
)
_
< +, (13)
for some : 0 < <

2
, implies = 0.
A set of all operator-valued functions F() S
a;
0
(

0
; N

0
(A), N

0
(A))
which are
0
-I-admissible with respect to the operator A, we shall denote
by
S
I;
0
(

0
; N

0
(A), N

0
(A)) = S
I
(

0
; N

0
(A), N

0
(A)).
In the case R(A) = H, we have D(X 1

0
(A
1
)) = {0} and therefore
S
I;
0
(

0
; N

0
(A), N

0
(A)) = S
a;
0
(

0
; N

0
(A), N

0
(A)).
Theorem 3 Let A be a closed symmetric invertible operator in a Hilbert
space H, and
0
R
e
be an arbitrary point. An arbitrary generalized I-
resolvent R

of the operator A has the following form:

R

=
_
_
_
_
A
F()
E
H
_
1
,

0
_
A
F

()
E
H
_
1
,

0
, (14)
where F() is a function from S
I;
0
(

0
; N

0
(A), N

0
(A)). Conversely,
an arbitrary function F() S
I;
0
(

0
; N

0
(A), N

0
(A)) denes by rela-
tion (14) a generalized I-resolvent R

of the operator A. Moreover, for dif-

ferent functions from S
I;
0
(

0
; N

0
(A), N

0
(A)) there correspond dierent
generalized I-resolvents of the operator A.
Proof. Let A be a closed symmetric invertible operator in a Hilbert space
H,
0
R
e
.
Let us prove the rst statement of the theorem. Let R

be an arbitrary
generalized I-resolvent of the operator A. The generalized I-resolvent R

is
generated by a self-adjoint invertible operator

A A in a Hilbert space

H
H. Repeating the arguments from the proof of Theorem 3.34 in [1] we obtain
a representation (14), where F() is a function fromS
a;
0
(

0
; N

0
(A), N

0
(A)).
It remains to check that F() S
I;
0
(

0
; N

0
(A), N

0
(A)).
The self-adjoint extension

A
1
of the operator A
1
in a Hilbert space

H H generates the following objects, see Subsection 3.6 in [1]: the

operator-valued function B

(A
1
,

A
1
), R
e
; the operator B

(A
1
,

A
1
);
4
the operator-valued function F(;
1

0
, A
1
,

A
1
), 1

0
; the operator

(
1

0
; A
1
,

A
1
).
On the other hand, the operator A can be identied with the operator
A = A A
e
, (15)
with A
e
= o
H
, in the Hilbert space

H = H H
e
, (16)
where H
e
:=

H H, see Subsection 3.5 in [1]. Then
A
1
= A
1
A
1
e
= A
1
A
e
. (17)
By the generalized Neumanns formulas, for

A A there corresponds an iso-
metric operator T, D(T) = N

0
(A), R(T) = N

0
(A), which is
0
-admissible
with respect to A. Moreover, since the operator

A is invertible, then by The-
orem 1 we conclude that the operator

0

0
T is
1

0
-admissible with respect to
A
1
.
Applying Theorem 3.16 in [1] for the operator A
1
, with z =
1

0
, we
obtain that the operator (
1

0
; A
1
,

0

0
T) is
1

0
-admissible with respect to
A
1
.
By the generalized Neumanns formulas, for

A
1
A
1
there corre-
sponds an isometric operator V , D(V ) = N 1

0
(A
1
) = N

0
(A), R(V ) =
N 1

0
(A
1
) = N

0
(A), which is
1

0
-admissible with respect to A
1
. Consider
the following operator B(
1

0
; A
1
, V ), see (3.38) in [1]:
B(
1

0
; A
1
, V )h = P

H
H
(A
1
)
V
h = P

H
H

A
1
h, h D((A
1
)
V
)H = D(

A
1
)H.
Comparing this denition with the denition of the operator B

(A
1
,

A
1
)
we conclude that
B(
1

0
; A
1
, V ) = B

(A
1
,

A
1
). (18)
Applying Theorem 3.20 in [1] for the operator

A
1
, with z =
1

0
, we get:
B(
1

0
; A
1
, V ) = (A
1
)
(
1

0
;A
1
,V ),
1

0
.
5
On the other hand, by the denition of

(
1

0
; A
1
,

A
1
) we have:
B

(A
1
,

A
1
) = (A
1
)
(
1

0
;A
1
,

A
1
),
1

0
.
Then

(
1

0
; A
1
,

A
1
) = (
1

0
; A
1
, V ). (19)
By the generalized Neumanns formulas (4),(5) we get:
D(T) = N

0
(A) R(

A
0
E

H
),
T (

A
0
E

H
)(

A
0
E

H
)
1
;
D(V ) = N 1

0
(A
1
) R(

A
1

0
E

H
) = N

0
(A) R(

A
0
E

H
),
V (

A
1

0
E

H
)(

A
1

0
E

H
)
1
=

0

0
(

A
0
E

H
)(

A
0
E

H
)
1
.
Therefore
V =

0

0
T.
By (19) we get:

(
1

0
; A
1
,

A
1
) = (
1

0
; A
1
,

0

0
T). (20)
Therefore the operator

(
1

0
; A
1
,

A
1
) is
1

0
-admissible with respect to
A
1
.
The self-adjoint operator

A A in a Hilbert space

H H generates
the operator-valued function B

(A,

A), R
e
, and the operator-valued
function F(;
0
, A,

A),

0
(see [1, Subsection 3.6]). Denote (see [1, p.
222])

(A,

A) =
_
h D(

A) : (

A E

H
)h H
_
, C; (21)
L

(A,

A) = P

H
H

L

, C.
We shall also need sets

L

(A
1
,

A
1
), L

(A
1
,

A
1
), C, constructed
by (21) with the operators A
1
,

A
1
instead of A,

A.
6
Choose an arbitrary element h

L

(A,

A), R
e
. Then

Ah R(

A),
and
(

A
1

H
)

Ah =
1

A E

H
)h H.
Therefore

Ah

L1

(A
1
,

A
1
), and

(A,

A)

L1

(A
1
,

A
1
), R
e
.
In order to obtain the equality:

(A,

A) =

L1

(A
1
,

A
1
), R
e
, (22)
it remains to apply the proved inclusion for the operators A
1
,

A
1
instead
of A,

A, and with
1

instead of .
The operator P

H
H
|

(A,

A)
, R
e
, is invertible, see considerations be-
low (3.48) in [1]. By the denition of B

(A,

A) we may write:
B

(A,

A) = P

H
H

A
_
P

H
H
|

(A,

A)
_
1
, R
e
. (23)
Applying this representation for the operators A
1
,

A
1
instead of A,

A,
and with
1

instead of we get:
B1

(A
1
,

A
1
) = P

H
H

A
1
_
P

H
H
|

L
1

(A
1
,

A
1
)
_
1
, R
e
.
Observe that
R(B

(A,

A)) = P

H
H

A

(A,

A) = L1

(A
1
,

A
1
), R
e
,
and
B1

(A
1
,

A
1
)B

(A,

A)g = g, g L

(A,

A), R
e
.
Therefore
B

(A,

A)
1
= B1

(A
1
,

A
1
), R
e
. (24)
By the denition of the operator-valued function F() and by (24) we may
write:
F(;
0
, A,

A) = (B

(A,

A)
0
E
H
)(B

(A,

A)
0
E
H
)
1

0
(A)
,

0
;
7
and, also for

0
, we have:
F(
1

;
1

0
, A
1
,

A
1
)
= (B1

(A
1
,

A
1
)
1

0
E
H
)(B1

(A
1
,

A
1
)
1

0
E
H
)
1

0
(A)
=

0

0
(B

(A,

A)
0
E
H
)(B

(A,

A)
0
E
H
)
1

0
(A)
.
Therefore
F(
1

;
1

0
, A
1
,

A
1
) =

0

0
F(;
0
, A,

A),

0
. (25)
Suppose that relations (12),(13) with F() = F(;
0
, A,

A) hold for some
: 0 < <

2
. Then using the change of a variable y =
1

we get:
lim
y

0
, y
F(
1
y
;
0
, A,

A) =

0

0
X 1

0
(A
1
), (26)
lim
y

0
, y
_
|y|(
H
F(
1
y
;
0
, A,

A)
H
)
_
< +. (27)
By (25) we obtain that
F(
1
y
;
0
, A,

A) =

0

0
F(y;
1

0
, A
1
,

A
1
), y 1

0
.
Substituting this expression in relations (26),(27) we get:
lim
y

0
, y
F(y;
1

0
, A
1
,

A
1
) = X 1

0
(A
1
), (28)
lim
y

0
, y
_
|y|(
H
F(y;
1

0
, A
1
,

A
1
)
H
)
_
< +. (29)
By Theorem 3.32 in [1] we conclude that D(

(
1

0
; A
1
,

A
1
)), and

(
1

0
; A
1
,

A
1
) = X 1

0
(A
1
).
8
Since

(
1

0
; A
1
,

A
1
) is
1

0
-admissible with respect to A
1
, we obtain that
= 0. Consequently, F() S
I;
0
(

0
; N

0
(A), N

0
(A)).
Let us check the second statement of the theorem. Let F() be an arbi-
trary function from S
I;
0
(

0
; N

0
(A), N

0
(A)). We shall use the following
lemma.
Lemma 1 Let n, m Z
+
{}: n + m = 0. There exists a closed sym-
metric invertible operator A in a Hilbert space H, D(A) = H, R(A) = H,
which has the deciency index (n, m).
Proof. Let H
0
be an arbitrary Hilbert space, and A = {f
k
}

k=0
be an
orthonormal basis in H
0
. Consider the following operator (unilateral shift):
V
0
h =

k=0

k
f
k+1
, h =

k=0

k
f
k
H
0
,
k
C,
with D(V
0
) = H
0
. The operator V
0
is closed, isometric, and its deciency
index is (0, 1). The condition V
0
g = g, for an element g H
0
, implies
g = 0. Consequently, the inverse Cayley transformation:
A
0
= i(V
0
+E
H
0
)(V
0
E
H
0
)
1
,
is a closed symmetric invertible operator in H
0
, with the deciency index
(0, 1). If h H
0
and h (V
0
E
H
0
)H
0
, then V

0
h = h. The condition
V

0
h = V
1
0
P
H
0
R(V
0
)
h = h, for an element h H
0
, implies h = 0. Therefore,
D(A
0
) = H
0
, R(A
0
) = H
0
. Set
H
l
=
l

j=0
H
0
, W
l
=
l

j=0
V
0
, l Z
+
{}.
W
l
is a closed isometric operator in H
l
. The deciency index of W
l
is equal
to (0, l + 1). If W
l
h = h, or W

l
h = h, then h = 0. Then
A
l
= i(W
l
+E
H
l )(W
l
E
H
l )
1
,
is a closed symmetric invertible operator in H
l
, with the deciency index
(0, l+1), l Z
+
. Moreover, we have D(A
l
) = H
l
, R(A
l
) = H
l
. Observe that
the operator A
l
has the deciency index (l +1, 0), l Z
+
. If m > 0, n > 0,
we set
H = H
m1
H
n1
, A = (A
m1
) A
n1
.
9

Let us return to the proof of the theorem. If the operator A is self-adjoint,

then the set S
I;
0
(

0
; N

0
(A), N

0
(A)) consists of a unique function F() =
o
H
. Of course, this function generates the resolvent of A by (14). Thus, we
can assume that A is not self-adjoint.
We shall use the scheme of the proof of the corresponding statement
in Theorem 3.34 in [1]. By Lemma 1 there exists a closed symmetric in-
vertible operator A
1
in a Hilbert space H
1
, D(A
1
) = H
1
, R(A
1
) = H
1
,
which has the same defect numbers as A. Let U and W be arbitrary iso-
metric operators, which map respectively N

0
(A
1
) on N

0
(A), and N

0
(A
1
)
on N

0
(A). Set F
1
() = W
1
F()U,

0
. Since F
1
() belongs to
S(

0
; N

0
(A
1
), N

0
(A
1
)), by Shtrauss formula it generates a generalized
resolvent R

(A
1
) of A
1
.
Let us check that R

(A
1
) is generated by a self-adjoint invertible opera-
tor

A
1
A
1
in a Hilbert space

H
1
H
1
. Suppose that R

(A
1
) is generated
by a self-adjoint operator

A
1
A
1
in a Hilbert space

H
1
H
1
. Suppose
that U
i
(

A
1
)h = h, for an element h

H
1
. Then
(h, U
i
(

A
1
)g)

H
1
= (U
i
(

A
1
)h, U
i
(

A
1
)g)

H
1
= (h, g)

H
1
, g

H
1
,
0 = (h, (U
i
(

A
1
) +E

H
1
)g)

H
1
, g

H
1
.
In particular, h is orthogonal to (U
i
(A
1
) + E
H
1
)D(U
i
(A
1
)) = R(A
1
). Then
h

H
1
H
1
. Set

H
0
= {h

H
1
: U
i
(

A
1
)h = h}.
Observe that

H
0
is a subspace of

H
1
H
1
. Then

H
1
= H
1
(

H
1
H
1
) = H
1

_
(

H
1
H
1
)

H
0
_

H
0
=

H
1

H
0
,
where

H
1
:= H
1

_
(

H
1
H
1
)

H
0
_
.
Notice that U
i
(

A
1
)

H
0
=

H
0
and U
i
(

A
1
)

H
1
=

H
1
. Set W
1
= U
i
(

A
1
)|

H
1
.
There are no non-zero elements g

H
1
such that W
1
g = g. Then the
inverse Cayley transformation

A
1
:= i(W
1
+E

H
1
)(W
1
E

H
1
)
1
,
is invertible. Since

A
1


A
1
, then
(

A
1
E

H
1
)
1
(

A
1
E

H
1
)
1
.
10
Therefore

A
1
generates R

(A
1
).
Set
H
e
:=

H
1
H
1
.
Starting from the same formula, we repeat the rest of the arguments in
the proof of Theorem 3.34 in [1]. In what follows, we shall use notations and
constructions from this proof without additional references. We shall obtain
a self-adjoint operator

A A in a Hilbert space

H H, which generates a
generalized resolvent R

of A. This generalized resolvent is related to F()

by (14). It remains to check that the operator

A is invertible.
Since the operator

A
1
is invertible, then by Theorem 1 we obtain that the
operator

0

0
T is
1

0
-admissible with respect to A
1
1
. By Theorem 3.16 in [1]
we conclude that the operators (
1

0
; A
1
1
,

0

0
T) and

0

0
T
22
are
1

0
-admissible
with respect to A
1
1
and A
1
e
= A
e
, respectively.
Comparing the domains of (
1

0
; A
1
1
,

0

0
T) and (
1

0
; A
1
,

0

0
V ) we con-
clude that
D((
1

0
; A
1
,

0

0
V )) = UD((
1

0
; A
1
1
,

0

0
T)). (30)
Using Remark 3.15 and formula (3.28) in [1, p. 218] for (
1

0
; A
1
,

0

0
V )
and (
1

0
; A
1
1
,

0

0
T), we get:
(
1

0
; A
1
,

0

0
V ) = W(
1

0
; A
1
1
,

0

0
T)U
1
. (31)
We can apply the arguments in the proof of the already proved rst
statement of the theorem for the operator A := A
1
; the point
0
; the gener-
alized I-resolvent R

:= R

(A
1
) of A
1
, which is generated by the self-adjoint
invertible operator

A :=

A
1
in

H
1
H
1
.

(
1

0
; A
1
1
,

A
1
1
) = (
1

0
; A
1
1
,

0

0
T). (32)
F(
1

;
1

0
, A
1
1
,

A
1
1
) =

0

0
F(;
0
, A
1
,

A
1
),

0
. (33)
Then
F(y;
1

0
, A
1
1
,

A
1
1
) =

0

0
F(
1
y
;
0
, A
1
,

A
1
), y 1

0
. (34)
11
By Theorem 3.32 in [1] we have:
D(

(
1

0
; A
1
1
,

A
1
1
)) = { N 1

0
(A
1
1
) :
lim

0
,
_
||(
H
1
F(;
1

0
, A
1
1
,

A
1
1
)
H
1
)
_
< +
_
, (35)

(
1

0
; A
1
1
,

A
1
1
) = lim

0
,
F(;
1

0
, A
1
1
,

A
1
1
), D(

(
1

0
; A
1
1
,

A
1
1
)),
(36)
where 0 < <

2
.
Using (30),(32),(35),(36),(34) and the change of a variable: y =
1

, we
obtain that
D((
1

0
; A
1
,

0

0
V )) = { N

0
(A) :
lim
y

0
,y0
_
1
|y|
(
H
F(y)
H
)
_
< +
_
; (37)
(
1

0
; A
1
,

0

0
V ) =

0

0
lim
y

0
,y0
F(y), D((
1

0
; A
1
,

0

0
V )).
(38)
Suppose that there exists an element D((
1

0
; A
1
,

0

0
V ))X 1

0
(A
1
)
such that the following equality holds:
(
1

0
; A
1
,

0

0
V ) = X 1

0
(A
1
).
By (37),(38) this means that N

0
(A) and
lim
y

0
,y0
_
1
|y|
(
H
F(y)
H
)
_
< +,

0
lim
y

0
,y0
F(y) = X 1

0
(A
1
).
Since F() is
0
-I-admissible with respect to the operator A, we get = 0.
This means that (
1

0
; A
1
,

0

0
V ) is
1

0
-admissible with respect to A
1
.
Since

0

0
T
22
is
1

0
-admissible with respect to A
e
, then by Theorem 3.16 in [1]
we obtain that the operator

0

0
V is
1

0
-admissible with respect to A
1
. By
Theorem 1 we conclude that the operator A
V
=

A is invertible.
The last statement of the theorem follows directly from Shtrauss for-
mula.
12
References
[1] S.M. Zagorodnyuk, Generalized resolvents of symmetric and isometric
operators: the Shtraus approach. Annals of Functional Analysis, 4,
No. 1 (2013), 175-285.
Invertible extensions of symmetric operators and the
corresponding generalized resolvents.
S.M. Zagorodnyuk
In this paper we study invertible extensions of a symmetric operator in
a Hilbert space H. All such extensions are characterized by a parameter in
the generalized Neumanns formulas. Generalized resolvents, which are gen-
erated by the invertible extensions, are extracted by a boundary condition
among all generalized resolvents in the Shtraus formula.
13