POLONICI MATHEMATICI LXVI (1997)
On the intertwinings of regular dilations
by Dumitru Gas¸par and Nicolae Suciu (Timi¸soara)
W lodzimierz Mlak in memoriam
Abstract. The aim of this paper is to find conditions that assure the existence of the commutant lifting theorem for commuting pairs of contractions (briefly, bicontractions) having (∗-)regular dilations. It is known that in such generality, a commutant lifting theorem fails to be true. A positive answer is given for contractive intertwinings which doubly intertwine one of the components. We also show that it is possible to drop the doubly intertwining property for one of the components in some special cases, for instance for semi-subnormal bicontractions. As an application, a result regarding the existence of a unitary (isometric) dilation for three commuting contractions is obtained.
0. Introduction. It is well known that the theorem of B. Sz.-Nagy and C. Foia¸s regarding the lifting of the commutant of a pair of contractions plays an important role in the applications of dilation theory in operator interpolation problems, optimization and control, in geology and geophysics.
This is excellently illustrated in the book [5] of C. Foia¸s and A. E. Frazho.
Lately, the dilation theory method was extended to the study of com- muting multioperators by many authors (W. Mlak, M. S loci´ nski, M. Kosiek, M. Ptak, E. Albrecht, V. M¨ uller, R. E. Curto, F. H. Vasilescu, A. Octavio, B. Chevreau and others). In 1993, at the B. Sz.-Nagy Anniversary Inter- national Conference in Szeged, C. Foia¸s raised the problem of obtaining a commutant lifting theorem for a pair of bicontractions having regular uni- tary dilations. In 1994, at the XV-th International Conference on Operator Theory in Timi¸soara, V. M¨ uller proved that in such a generality, the com- mutant lifting theorem fails.
In the present work it is our aim to find conditions that assure the existence of such a lifting. In this frame a structure for regular (or ∗-regular) dilations is needed.
1991 Mathematics Subject Classification: Primary 47A20; Secondary 47A13.
Key words and phrases: commuting multioperator, ∗-regular dilation, contractive intertwining, (semi-)subnormal pair.
[105]
1. Preliminaries. For a complex separable Hilbert space H, B(H) means the C
∗-algebra of all bounded linear operators on H (with Hilbert adjoint as involution). The elements of the (closed) unit ball in B(H) are called contractions on H. An n-tuple of operators will be called a multiop- erator . If the members of the n-tuple commute, then we have a commuting multioperator . A commuting multioperator consisting of contractions will be called a multicontraction (bicontraction if n = 2) on H.
For a multicontraction T := (T
1, . . . , T
n) we define T
∗:= (T
1∗, . . . , T
n∗).
We shall also use the multiindex notation
T
m:= T
1m1. . . T
nmn, m = (m
1, . . . , m
n) ∈ Z
n+,
where Z (resp. Z
+) is the set of all (resp. positive) integers. A multicontrac- tion T on H will be briefly denoted by [H, T ].
An isometric (resp. unitary) dilation of a multicontraction [H, T ] is a multicontraction [K, U ] consisting of isometric (resp. unitary) operators, such that K contains H as a closed subspace and
(1) T
m= P
HU
m|H (m ∈ Z
n+), where P
H= P
K,His the orthogonal projection of K on H.
It is known (see [1]) that each bicontraction has an isometric (and uni- tary) dilation, and generally speaking, an n-tuple consisting of more than three commuting contractions has no isometric dilation (see [15]). An iso- metric, respectively unitary, dilation [K, U ] of [H, T ] is called minimal if
(2) K = _
m∈Zn+
U
mH, or respectively,
(2
0) K = _
m∈Zn
U
mH.
Let us first note that if [K, V ] is an isometric minimal dilation of the mul- ticontraction [H, T ], then by (2), H is invariant with respect to V
∗and V
i∗|H = T
i∗(i = 1, . . . , n). Let us also mention that if [K, V ] is an isometric minimal dilation of [H, T ], and [ e K, U ] is the minimal unitary extension (see [18]) of [K, V ], then it is the unitary minimal dilation of [H, T ].
On the other hand, it is known that in case of a single contraction, the minimality condition (2) or (2
0) implies that the isometric (resp. unitary) dilation is uniquely determined up to a unitary equivalence which fixes H.
But for n > 1 this is not true ([1], [18]).
An isometric (resp. unitary) minimal dilation [K, U ] of the multicontrac- tion [H, T ] is called regular (respectively ∗-regular) if it satisfies
(3) T
∗m−T
m+= P
HU
∗m−U
m+|H (m ∈ Z
n),
or respectively,
(3
∗) T
m+T
∗m−= P
HU
∗m−U
m+|H (m ∈ Z
n),
where m
+:= (m
+1, . . . , m
+n), m
−= (m
−1, . . . , m
−n) and m
+i:= max{m
i, 0}, m
−i:= max{−m
i, 0}.
Regular dilations were studied in [3], [10], [18] and recently in [4] and [8]. Their existence is not assured for any multicontraction, not even for n = 2. However, if such a dilation exists, then by the minimality condition (2) or (2
0) it is uniquely determined up to unitary equivalence (see [18]). It is easy to see from (3) and (3*) that [K, U ] is a regular (resp. ∗-regular) unitary dilation of [H, T ] iff [K, U
∗] is a ∗-regular (resp. regular) unitary dilation of [H, T
∗]. We also note that if [K, V ] is a regular (resp. ∗-regular) isometric dilation of [H, T ] then the minimal unitary extension [ e K, U ] of [K, V ] is a regular (resp. ∗-regular) unitary dilation. On the other hand, if [K, U ] is a regular unitary dilation, by putting K
+:= W
m∈Zn+
U
mH, K
+∗:= W
m∈Zn+
U
∗mH, V
∗i:= U
i∗|K
+∗and V
i:= U
i|K
+(i = 1, . . . , n), then [K
+, V ] is a regular isometric dilation of [H, T ], whereas [K
∗+, V
∗] is a ∗-regular isometric dilation of [H, T
∗]. Let us also recall that a multicon- traction [H, T ] has a regular isometric dilation iff
∆
T:= X
|m|≤n
(−1)
|m|T
∗m−T
m+≥ 0, where |m| := m
1+ . . . + m
n(see [18], [4]).
A multicontraction is called a polydisc isometry ([4]) when ∆
T= 0. It is easily seen that if T
i(i = 1, . . . , n) are isometries (i.e. T is an n-toral isometry [2]), then T is a polydisc isometry. Now if I − P
ni=1
T
i∗T
i≥ 0, then
∆
T≥ 0. If P
ni=1
T
i∗T
i= I, then [H, T ] is called a spherical isometry ([2]).
When [H, T
∗] is a polydisc or a spherical isometry, we say that [H, T ] is a polydisc or a spherical coisometry, respectively. If the multicontraction is doubly commuting, then obviously ∆
T= (I − T
1∗T
1) . . . (I − T
n∗T
n) ≥ 0.
Furthermore, it is easy to verify
Proposition 0. For a multicontraction [H, T ] the following statements are equivalent :
(i) [H, T ] is doubly commuting;
(ii) [H, T ] has a regular isometric dilation which is doubly commuting ; (iii) [H, T ] has a regular unitary dilation [K, U ] such that [K, U
∗] is a regular unitary dilation for [H, T
∗].
In particular , a multicontraction consisting of coisometries has a regular
isometric (or unitary) dilation iff it is doubly commuting.
Let us observe that Proposition 0(iii) means more than that T has a reg- ular and a ∗-regular dilation. For example, for a bicontraction T = (T
0, T
1) with kT
0k
2+ kT
1k
2≤ 1, we have ∆
T≥ 0 and ∆
T∗≥ 0 but it is possible that T
0T
1∗6= T
1∗T
0.
Finally, also recall that an isometric pair [H, V ] is called a shift n-tuple (see [7], [8]) or a multishift (see [4]) if there exists a wandering (closed) subspace E in H (i.e. V
mE ⊥ V
pE, m 6= p, m, p ∈ Z
n+) such that H = L
m∈Zn+
V
mE. For the sake of simplicity we shall work in the case n = 2.
2. ∗-Regular isometric dilations. The isometric dilations consisting of doubly commuting isometries are in some sense connected with regular dilations. Precisely this is given in
Theorem 1. For a bicontraction [H, T ] with T = (T
0, T
1) the following assertions are equivalent :
(i) T has a doubly commuting minimal isometric dilation;
(ii) T has a minimal isometric dilation of the form [M ⊕ G, W ⊕ V ], where W is a bishift on M and V is a bidisc coisometry on G;
(iii) If [K
0, S
0] is the minimal isometric dilation of T
0, then there exists a contraction S
1on K
0which doubly commutes with S
0, such that P
HS
1= T
1P
H;
(iv) T has a ∗-regular isometric (unitary) dilation.
P r o o f. (i)⇒(ii). Let [K, U ] be a minimal isometric dilation of T with U
0, U
1doubly commuting isometries on K. By the Wold decomposition ([17], [7]) we have K = K
u⊕K
s⊕K
s0⊕K
s1, so that U
0and U
1reduce each subspace and U
0, U
1are unitary on K
u, U is a shift pair on K
sand U
iis unitary (resp. a shift) on K
s1−i(resp. K
si), i = 0, 1. Put G = K
u⊕ K
s0⊕ K
s1, V
i= V
ii⊕ V
i1−iwith V
ii= U
i|K
u⊕ K
s0, V
i1−i= U
i|K
s1(i = 0, 1) and V = (V
0, V
1), W
0= (V
00, V
11), W
00= (V
01, V
10). Because V
11is unitary, W
0∗is a bidisc isometry on K
u⊕ K
s0, and since V
01is unitary, W
00∗is a bidisc isometry on K
s1. Then
∆
V∗= ∆
W0∗+ ∆
W00∗= 0,
so V
∗is a bidisc isometry on G. Therefore since W := (U
0|K
s, U
1|K
s) is a bishift on M = K
sand K = M ⊕ G, W ⊕ V = U , we see that the dilation [K, U ] of T has the form described in (ii).
(ii)⇒(iii). Let [M⊕G, W ⊕V ] be as in (ii). Since W is a bishift on M, the isometries W
0and W
1doubly commute on M ([16]). Also the isometries V
0and V
1doubly commute on G, because V has a ∗-regular dilation. Therefore
the isometries U
0= W
0⊕ V
0and U
1= W
1⊕ V
1doubly commute on M ⊕ G.
Put
K
0= _
m∈Z+
W
0mH, S
0= U
0|K
0, S
1= P
K0U
1|K
0.
Then [K
0, S
0] is the minimal isometric dilation of T
0, S
0S
1= S
1S
0and P
HS
1= T
1P
H. Since U
0∗|H = T
0∗= S
0∗|H, we also have U
0∗|K
0= S
∗0. Furthermore, for k = P
p∈Z+
S
0ph
p∈ K with the sequence {h
p} ⊂ H with finite support, we obtain
S
1S
0∗k = S
1S
0∗h
0+ X
p≥1
S
1S
0p−1h
p= P
K0U
1U
0∗h
0+ X
p≥1
S
0p−1S
1h
p= P
K0U
0∗U
1h
0+ X
p≥1
S
0∗S
0pS
1h
p= S
0∗S
1h
0+ S
0∗X
p≥1
S
1S
0ph
p= S
0∗S
1k,
where we have used the fact that P
K0U
0∗U
1|K
0= S
0∗S
1. Consequently, S
0and S
1doubly commute on K
0.
(iii)⇒(iv). If K
0, S
0and S
1are as in (iii), then S
i∗|H = T
i∗(i = 0, 1) and since S
0and S
1doubly commute, we have
∆
T∗= I − T
0T
0∗− T
1T
1∗+ T
0T
1T
0∗T
1∗= P
H(I − S
0S
0∗− S
1S
1∗+ S
0S
1S
0∗S
1∗)|H
= P
H(I − S
0S
0∗)(I − S
1S
1∗)|H ≥ 0.
Consequently, T has a ∗-regular isometric (or unitary) dilation.
(iv)⇒(i). Suppose ∆
T∗≥ 0. Denote by M = H
2(T
2, H) and Z = (Z
0, Z
1) the shift pair (that is, Z
0and Z
1are the operators of multiplication with the coordinate functions) on M. Using Theorem 3.15 of [4], there are a Hilbert space H
1, a bicontraction N = (N
0, N
1) on H
1with N
0and N
1normal operators and with N
∗a bidisc isometry, and an isometry A of H in M ⊕ H
1such that AH is invariant for (Z
i⊕ N
i)
∗and (Z
i⊕ N
i)
∗A = AT
i∗, i = 0, 1. Then it results that
T
0pT
1q= A
∗(Z
0⊕ N
0)
p(Z
1⊕ N
1)
qA (p, q ∈ Z
+).
Let now [K
1, (M
0, M
1)] be a minimal isometric dilation of N with M
0and M
1doubly commuting isometries on K
1. Put
K = M ⊕ K
1, U
i= Z
i⊕ M
i(i = 0, 1).
Then U
0and U
1are doubly commuting isometries on K. Denoting by J the
embedding of M ⊕ H
1in K, we find that J A is an isometry of H into K.
For m = (p, q) ∈ Z
2+and h ∈ H we obtain
(J A)
∗U
mJ Ah = (J A)
∗U
0pU
1q(P
H0Ah ⊕ P
H1Ah)
= (J A)
∗(Z
mP
H0Ah ⊕ M
mP
H1Ah)
= A
∗(P
H0Z
mP
H0Ah ⊕ P
H1M
mP
H1Ah)
= A
∗(Z
mP
H0Ah ⊕ N
mP
H1Ah)
= A
∗(Z
0⊕ N
0)
p(Z
1⊕ N
1)
qAh = T
mh.
Let us observe that the subspace AH is invariant for U
i∗(i = 0, 1), because for h ∈ H we have
U
i∗Ah = U
i∗(P
H0Ah ⊕ P
H1Ah) = Z
i∗P
H0Ah ⊕ M
i∗P
H1Ah
= Z
i∗P
H0Ah ⊕ N
i∗P
H1Ah = (Z
i⊕ N
i)
∗Ah = AT
i∗h.
Now define
K
+= _
m∈Z2+
U
mAh, V
i= U
i|K
+(i = 0, 1)
and B = J
+A, where J
+is the embedding of M ⊕ H
1in K
+. Then B is an isometry of H into K
+and we have
T
m= B
∗V
mB (m ∈ Z
2+).
Identifying H with BH in K
+, we deduce that [K
+, V ] is a minimal isometric dilation of T . It remains to prove that V
0V
1∗= V
1∗V
0. First, since U
0and U
1doubly commute on K, it results that K
+is invariant for U
i∗, i = 0, 1.
Indeed, for k = P
m∈Z2+
U
mAh
mwith the sequence {h
m} ⊂ H with finite support, we obtain
U
0∗k = X
q≥0
U
0∗U
1qAh
0q+ X
p≥1 q≥0
U
0∗U
0pU
1qAh
pq= X
q≥0
U
1qU
0∗Ah
0q+ X
p≥1 q≥0
U
0p−1U
1qAh
pq= X
q≥0
U
1qAT
0∗h
0q+ X
p≥1 q≥0
U
0p−1U
1qAh
pq.
Therefore U
0∗K
+⊂ K
+and analogously U
1∗K
+⊂ K
+. Then V
i∗= U
i∗|K
+(i = 0, 1), and consequently, V
0V
1∗= V
1∗V
0. Hence [K
+, (V
0, V
1)] is a doubly commuting minimal isometric dilation of T .
Corollary 2. A bicontraction T on H has a regular isometric (uni-
tary) dilation if and only if T has a doubly commuting minimal coisometric
extension.
P r o o f. If ∆
T≥ 0, then T
∗has a doubly commuting minimal isometric dilation [K, (W
0, W
1)]. Hence (W
0∗, W
1∗) is a minimal coisometric extension of T and W
0∗, W
1∗doubly commute on K. The converse is obvious.
R e m a r k. If K = K
iu⊕ K
isis the Wold decomposition of K relative to W
i(in the previous proof), then W
1−ireduces K
iuand K
is, i = 0, 1. Thus, the matrix of W
1−i∗relative to the decomposition K = K
iu⊕ K
sihas diagonal form for i = 0, 1, that is, T is diagonally extendable (see [11]).
Now we can give the following characterization of the double commuta- tivity of an isometric dilation of [H, T ].
Proposition 3. Suppose T = (T
0, T
1) is a bicontraction on H and [K, (U
0, U
1)] a minimal isometric dilation of T . Then the isometries U
0and U
1doubly commute on K iff [K, U ] is a ∗-regular isometric dilation of [H, T ].
In particular , the doubly commuting minimal isometric dilation of T (if it exists) is unique up to unitary equivalence.
P r o o f. It is not difficult to see that the condition (3
∗) is equivalent to (4) T
ipT
1−i∗q= P
HU
1−i∗qU
ip|H (p, q ∈ Z
+; i = 0, 1).
Suppose that the dilation U = (U
0, U
1) satisfies (4) and let [ e K, (f U
0, f U
1)] be the minimal unitary extension of U . Then for p, q ∈ Z
+and i = 0, 1 we have
T
ipT
1−i∗q= P
He U
1−i∗qe U
ip|H = P
HU e
piU e
∗q1−i|H
and we deduce that [ e K, ( e U
∗0, e U
1∗)] is a regular minimal unitary dilation for T
∗. By Theorem 1, T has a doubly commuting minimal isometric dilation [M, (V
0, V
1)]. Obviously, V
isatisfies (4) (in place of U
i) and consequently (∗) P
HU
i∗qU
1−ip|H = P
HV
i∗qV
1−ip|H (p, q ∈ Z
+; i = 0, 1).
Let us prove that the dilations U = (U
0, U
1) and V = (V
0, V
1) are unitar- ily equivalent. Let {h
n}
n∈Z2+
⊂ H be a sequence with finite support. Since U and V are dilations of T and satisfy (∗), by defining m := (i, j) ∈ Z
2+and n := (p, q) ∈ Z
2+we obtain
X
n∈Z2+
U
nh
n2
= X
m,n∈Z2+
(U
nh
n, U
mh
m)
= X
j<q
(U
0∗iU
0pU
1q−jh
n, h
m) + X
j≥q
(U
0∗iU
1∗(j−q)U
0ph
n, h
m)
= X
i<p j<q
(U
0p−iU
1q−jh
n, h
m) + X
i≥p j<q
(U
0∗(i−p)U
1q−jh
n, h
m)
+ X
i<p j≥q
(U
1∗(j−q)U
0p−ih
n, h
m) + X
i≥p j≥q
(U
0∗(i−p)U
1∗(j−q)h
n, h
m)
= X
i<p j<q
(V
0p−iV
1q−jh
n, h
m) + X
i≥p j<q
(V
0∗(i−p)V
1q−jh
n, h
m)
+ X
i<p j≥q
(V
1∗(j−q)V
0p−ih
n, h
m) + X
i≥p j≥q
(V
0∗(i−p)V
1∗(j−q)h
n, h
m)
= X
m,n∈Z2+
(V
∗mV
nh
n, h
m) =
X
n∈Z2+
V
nh
n2
.
Using the minimality conditions of the spaces K and M and the norm equalities above, we deduce that there exists a unitary operator W from K to M satisfying
W X
n∈Z2+
U
nh
n= X
n∈Z2+
V
nh
nfor {h
n} ⊂ H with finite support. Consequently, W |H = I and W U
i= V
iW , i = 0, 1, and in particular, it results that U
0and U
1doubly commute on K.
Since the other assertions were also implicitly proved, the proof is finished.
Now, having in mind the condition (iii) of Theorem 1, we obtain Corollary 4. Let T = (T
0, T
1) be a bicontraction on H and [K
0, S
0] (re- spectively [K
∗0, S
∗0]) the minimal isometric dilation of T
0(resp. T
0∗). Then T has a ∗-regular (resp. regular ) isometric dilation if and only if T
1∗(resp.
T
1) has a contractive extension on K
0(resp. K
∗0) which doubly commutes with S
0(resp. S
∗0).
It is obvious (by the proof of Proposition 3) that if [K, V ] is a ∗-regular (resp. regular) isometric dilation of [H, T ] and if [ e K, U ] is the minimal unitary extension of V then the regular (resp. ∗-regular) isometric dilation of T
∗is [K
∗, (V
∗0, V
∗1)], where
(5) K
∗= _
m,n∈Z+
U
0∗mU
1∗nH, V
∗i= U
i∗|K
∗(i = 0, 1).
Furthermore, with the notations of Theorem 1(iii), the ∗-regular isometric dilation [K, V ] of T is the regular and ∗-regular isometric dilation of the doubly commuting bicontraction S = (S
0, S
1) (see Proposition 0), and in fact, [K, V
1] is the minimal isometric dilation of S
1.
3. Intertwinings of regular dilations. Let H and H
0be two Hilbert
spaces and T = (T
0, T
1) and T
0= (T
00, T
10) two bicontractions on H and
H
0respectively. A bounded linear operator A : H → H
0intertwines T and
T
0if AT
i= T
i0A, i = 0, 1. The operator A doubly intertwines T and T
0if AT
i= T
i0A and AT
i∗= T
i0∗A, i = 0, 1.
V. M¨ uller has shown in [14] that if A intertwines two bicontractions which have regular dilations, then in general, A cannot be “lifted” in the sense of [5], [18] to an operator which intertwines these dilations. In order to give conditions under which this is possible, we will first prove
Theorem 5. Let [H, T ] and [H
0, T
0] be two bicontractions having ∗- regular isometric dilations [K, U ] and [K
0, U
0] respectively. Let A be a con- traction from H in H
0such that AT
i= T
i0A (i = 0, 1) and AT
0∗= T
00∗A.
Then there is a contraction B from K in K
0with BU
i= U
i0B (i = 0, 1), BU
0∗= U
00∗and P
H0B = AP
H.
P r o o f. Let A : H → H
0be a contraction which satisfies AT
i= T
i0A (i = 0, 1) and AT
0∗= T
00∗A. Let [K
0, S
0] and [K
00, S
00] be the minimal iso- metric dilations of T
0and T
00respectively. By Theorem 1(iii) there are con- tractions S
1on K
0and S
10on K
00such that S
1doubly commutes with S
0and P
HS
1= T
1P
H, while S
10doubly commutes with S
00and P
H0S
10= T
10P
H0. Since
K
0= H ⊕ M
p∈Z+
S
0p(S
0− T
0)H, K
00= H
0⊕ M
p∈Z+
S
00p(S
00− T
00)H
0(see [5], [18]), and A doubly intertwines T
0and T
00, we can define a contrac- tion A
0: K
0→ K
00by setting
A
0k
0:= Ah + X
p≥0
S
00p(S
00− T
00)Ah
pfor k
0= h + P
p∈Z+
S
0p(S
0− T
0)h
p, where h, h
p∈ H. We have A
0|H = A, and for k
0∈ K
0as above,
A
0S
0k
0= A
0h
T
0h + (S
0− T
0)h + X
p≥0
S
0p+1(S
0− T
0)h
pi
= AT
0h + (S
00− T
00)Ah + X
p≥0
S
0p+10(S
00− T
00)Ah
p= S
00h
Ah + X
p≥0
S
00p(S
00− T
00)Ah
pi
= S
00A
0k
0.
Therefore A
0S
0= S
00A
0. Also, A
0S
0∗= S
00∗A
0and A
∗0|H
0= A
∗, because for k = P
p≥0
S
0ph
p(finite sum) with h
p∈ H, we have A
0S
0∗k = AT
0∗h
0+ X
p≥1
A
0S
0p−1h
p= T
00∗Ah
0+ X
p≥1
S
00p−1Ah
p= S
00∗Ah
0+ S
00∗X
p≥1
S
00pAh
p= S
00∗X
p≥0
S
00pAh
p= S
00∗A
0k,
and for h
0∈ H
0, (A
∗0h
0, k) = X
p
(h
0, S
00pAh
p) = X
p
(T
00∗ph
0, Ah
p) = X
p
(A
∗T
00∗ph
0, h
p)
= X
p
(T
0∗pA
∗h
0, h
p) = X
p
(A
∗h
0, S
0ph
p) = (A
∗h
0, k).
Next, we also have A
0S
1= S
10A
0, because for {h
0p}
p≥0⊂ H
0with finite support,
A
∗0S
10∗X
p
S
00ph
0p= X
p
A
∗0S
00pS
10∗h
0p= X
p
S
0pA
∗0T
10∗h
0p= X
p
S
0pA
∗T
10∗h
0p= X
p
S
0pT
1∗A
∗h
0p= X
p
S
0pS
1∗A
∗0h
0p= S
1∗A
∗0X
p
S
00ph
0p, and consequently, A
∗0S
10∗= S
1∗A
∗0, whence A
0S
1= S
10A
0. We conclude that A intertwines S
1and S
10and doubly intertwines S
0and S
00, and A
0is an extension for A, while A
∗0is an extension for A
∗. Hence P
H0A
0= AP
H.
Now let [K, U
1], [K
0, U
10] be the minimal isometric dilations of S
1, S
10respectively, and let U
0, U
00be the ∗-extensions of S
0(on K) and of S
00(on K
0), respectively, such that U
0doubly commutes with U
1and U
00doubly commutes with U
10. Using the sequences of n-step dilations for S
1and S
10and the corresponding n-step intertwining liftings of A, we can define a contraction B : K → K
0by
Bk = lim
n
A
nP
Knk (k ∈ K),
where {K
n} and {A
n} are inductively defined with K
1= K
0⊕ D
S1and A
1: K
0⊕ D
S1→ K
00⊕ D
S01
of the form A
1=
A
00
X
1D
A0Y
1,
D
Cbeing the defect space of the operator C. Here the operator (X
1, Y
1) : D
A0⊕ D
S1→ D
S01
is (X
1, Y
1) = Γ
0P
0, where P
0is the orthogonal projection of D
A0⊕D
S1on the subspace {D
A0S
1k ⊕D
S1k : k ∈ K
0}
−and Γ
0(D
A0S
1k ⊕ D
S1k) = D
S01A
0k, k ∈ K
0. Then B satisfies BU
1= U
10B, BU
0= U
00B, BU
0∗= U
00∗B and P
K00B = A
0P
K0(see [9] for details). Hence P
H0B = BP
Hand since [K, (U
0, U
1)] and [K
0, (U
00, U
10)] are the ∗-regular isometric dilations for T and T
0respectively, B is the desired operator. The proof is finished.
Corollary 6. Let [H, T ] and [H
0, T
0] be two bicontractions which have
∗-regular (or regular ) isometric dilations with the minimal unitary exten-
sions [ e K, e U ] and [ e K
0, e U
0] respectively. If A is a contraction from H to H
0which satisfies AT
i= T
i0A (i = 0, 1) and AT
0∗= T
00∗A, then there ex-
ists a contraction e A from e K to e K
0such that e A e U
i= e U
0iA (i = 0, 1) and e
P
H0A|H = A. e
P r o o f. Suppose that T and T
0have the ∗-regular isometric dilations [K, U ] and [K
0, U
0] and let [ e K, e U ] and [ e K
0, e U
0] be the minimal unitary exten- sions of U and U
0respectively. If A is an intertwining contraction of T and T
0and B is an intertwining contraction of U and U
0with P
H0B = AP
Hgiven by Theorem 5, then there exists (see [12]) a contraction e B from e K into e K
0which intertwines e U and e U
0, such that e B|K = B. It results that P
H0B|H = A, whence P e
HB e
∗|H
0= A
∗and e B
∗intertwines e U
0∗and e U
∗. Ob- viously, [ e K, e U
∗] and [ e K
0, e U
0∗] are the regular unitary dilations of T
∗and T
0∗respectively.
Theorem 7. Let [H, T ] and [H
0, T
0] be two bicontractions having regular isometric dilations [K, U ] and [K
0, U
0] respectively, such that T
1∗or T
10is an isometry. If A is a contraction of H into H
0such that AT
i= T
i0A (i = 0, 1) and AT
0∗= T
00∗A, then there exists a contraction B from K to K
0with BU
i= U
i0B (i = 0, 1), and P
H0B = AP
H.
P r o o f. Suppose that the bicontractions T = (T
0, T
1) and T
0= (T
00, T
10) have regular isometric dilations. Then T
∗= (T
0∗, T
1∗) has a ∗-regular isomet- ric dilation and therefore if [K
0∗, S
0∗] is the minimal isometric dilation of T
0∗, then there is a contraction S
1∗on K
0∗which doubly commutes with S
0∗, such that P
HS
1∗= T
1∗P
H. Let [ e K
0, e S
0] be the minimal isometric dilation of the coisometry S
0∗∗and let e S
1be the ∗-extension of S
1∗∗to e K
0which doubly commutes with e S
0. But e S
0is a unitary operator on e K
0and [K
0, S
0] given by
K
0= _
n∈Z+
S e
0nH, S
0= e S
0|K
0,
is the minimal isometric dilation of T
0. We have S
1∗∗|H = T
1and therefore S e
1|H = T
1. Hence K
0is an invariant subspace for e S
1and S
1= e S
1|K
0is a contraction on K
0which satisfies S
0S
1= S
1S
0and S
1|H = T
1.
Analogously, if [K
00, S
00] is the minimal isometric dilation of T
00, then there is a contraction S
10on K
00which satisfies S
00S
10= S
10S
00and S
10|H
0= T
10.
Now let A : H → H
0be a contraction which intertwines T
1and T
10and doubly intertwines T
0and T
00. As in the proof of Theorem 5 there is a contraction A
0: K
0→ K
00which doubly intertwines S
0and S
00, such that A
0|H = A. Then for any sequence {h
n} ∈ H with finite support we have
A
0S
1X
n
S
0nh
n= X
n
S
00nA
0S
1h
n= X
n
S
00nAT
1h
n= X
n
S
00nT
10Ah
n= X
n
S
00nS
10A
0h
n= S
10A
0X
n
S
0nh
n,
therefore A
0S
1= S
10A
0. Let us remark that if T
1∗or T
10is an isometry, then
so is S
1∗(respectively S
10). In this case it is known (see [5]) that A
0has
a unique contractive intertwining lifting of the minimal isometric dilations
of S
1and S
10. Now as in the proof of Theorem 5 (see [9]) we can obtain a contraction B : K → K
0, where [K, U
1] and [K
0, U
10] are the minimal isometric dilations of S
1and S
10respectively, such that P
K00B = A
0P
K0and BU
1= U
10B, BU
0= U
00B, U
0and U
00being the isometric extensions of S
0and S
00to K and K
0which commute with U
1and U
10respectively. Finally, it is easy to see that [K, (U
0, U
1)] and [K
0, (U
00, U
10)] are the regular isometric dilations of T and T
0respectively. The proof is finished.
Now we can obtain the versions of Theorems 5 and 7 for double inter- twinings which complete those obtained in [14].
Proposition 8. Let [H, T ] and [H
0, T
0] be two bicontractions having regular (or ∗-regular ) isometric dilations [K, V ] and [K
0, V
0] respectively. If A is a contraction from H into H
0which doubly intertwines T and T
0, then there exists a (unique) ∗-extension of A from K to K
0which preserves the norm of A and doubly intertwines V and V
0.
P r o o f. Suppose first that T and T
0have ∗-regular isometric dilations.
Let [K
0, (S
0, S
1)] and [K
00, (S
00, S
10)] be as in the proof the Theorem 5. Con- sider A : H → H
0a contractive double intertwining of T and T
0and A
0: K
0→ K
00with A
0S
i= S
i0A
0(i = 0, 1), A
0S
0∗= S
00∗A
0, A
0|H = A, A
∗0|H
0= A
∗and kA
0k = kAk. Then for {h
p}
p≥0⊂ H with finite support, we have
A
0S
1∗X
p
S
0ph
p= X
p
S
00pA
0S
1∗h
p= X
p
S
00pAT
1∗h
p= X
p
S
00pT
10∗Ah
p= X
p
S
00pS
10∗A
0h
p= S
10∗A
0X
p
S
0ph
p, and so A
0S
1∗= S
10∗A
0.
Now let [K, V
1] and [K
0, V
10] be the minimal isometric dilations of S
1and S
10and let V
0, V
00be the extensions of S
0, S
00to K, K
0which doubly commute with V
1, V
10respectively. As above, there exists a contraction B : K → K
0with BV
i= V
i0B, BV
i∗= V
i0∗B, (i = 0, 1), B|K
0= A
0, B
∗|K
00= A
∗0, whence B|H = A, B
∗|H
0= A
∗and kBk = kA
0k = kAk. So the conclusion holds for the ∗-regular isometric dilations [K, V ] and [K
0, V
0] of T and T
0.
Next let [ e K, e U ], [ e K
0, e U
0] be the minimal unitary extensions of V , V
0and
[K
∗, V
∗], [K
0∗, V
∗0] be the regular isometric dilations of T
∗, T
0∗respectively
(as in (5)). Then there exists ([12], [7]) a contraction e A : e K → e K
0such that
A e e U
i= e U
0iA (i = 0, 1), e e A|K = B, e A
∗|K
0= B
∗and k e Ak = kBk. Because
A|H = A and e e A e U
∗i= e U
0∗iA (i = 0, 1), we have e e AK
∗⊂ K
0∗. But e A
∗|H
0= A
∗and e A
∗U e
0∗i= e U
∗iA
∗(i = 0, 1) imply e A
∗K
0∗⊂ K
∗. So we can define the
operator C : K
∗→ K
0∗by C = e A|K
∗. Then C
∗= e A
∗|K
0∗and C|H = A,
C
∗|H
0= A
∗and since V
∗i= e U
∗i|K
∗, V
∗i0= e U
0∗i|K
0∗(i = 0, 1), it results that
CV
∗i= V
∗i0C and CV
∗i∗= V
∗i0∗C (i = 0, 1). Finally, kAk ≤ kCk ≤ k e Ak = kBk = kAk and so kCk = kAk. Thus the conclusion holds for the regular isometric dilations of T
∗and T
0∗, and consequently, in the case when T and T
0have regular isometric dilations.
Under certain conditions we can drop the doubly intertwining property on a component. The first fact in this context is contained in
Proposition 9. Let [H, T ] be a bicontraction with the first component T
0a coisometry, and [H
0, T
0] be another bicontraction which has a ∗-regular isometric dilation. Let [K, U ] and [K
0, U
0] be the ∗-regular isometric dilations of T and T
0respectively. If A is a contraction from H to H
0with AT
i= T
i0A (i = 0, 1), then there exists a contraction B from K to K
0such that BU
i= U
i0B (i = 0, 1), and P
H0B = AP
H.
P r o o f. Let A : H → H
0be a contractive intertwining of T and T
0. Preserving the notations of the proof of Theorem 5, we find (by the lifting theorem) that there exists a contraction A
0from K
0to K
00which satisfies A
0S
0= S
00A
0and P
H0A
0= AP
H. Because T
0is a coisometry, its minimal isometric dilation S
0is a unitary operator on K
0. Then from Theorem B of [6] (which can be extended to operators acting on different spaces) it results that A
0S
0∗= S
00∗A
0. So A
0doubly intertwines S
0and S
00. Furthermore, A
0intertwines S
1and S
10, the doubly commuting commutants of S
0and S
00which lift T
1and T
10respectively, given by Theorem 1(iii). By Theorem 5, A
0has a contractive lift, which intertwines the ∗-regular isometric dilations of S = (S
0, S
1) (of T ) and S
0= (S
00, S
10) (of T
0), whence the conclusion follows.
The dual version of Proposition 9 is in fact an extension of Proposi- tion 5.2 from [12] (for bicontractions) and of Proposition 10 from [2].
Corollary 10. Let [H, T ] be a bicontraction which has a regular isomet- ric dilation, and [H
0, T
0] be a bicontraction with T
00an isometry. We also suppose that T
1∗or T
10is an isometry. If A is a contraction of H into H
0which intertwines T and T
0, then there exists a contraction B which intertwines the regular isometric dilations of T and T
0and satisfies P
H0B = AP
H.
P r o o f. If [K
0, S
0] is the minimal isometric dilation of T
0and S
1is a contraction on K
0with S
0S
1= S
1S
0and S
1|H = T
1, then A
0= AP
His a contraction from K
0into H
0and satisfies A
0S
0= T
00A
0, A
0S
1= T
10A
0, and A
0is a lifting for A. Since S
0and T
00and respectively S
1∗or T
10are isometries, there is a lifting for A
0which intertwines the regular isometric dilations for (S
0, S
1) and (T
00, T
10).
Recall ([13]) that a bounded linear operator S on H is subnormal if
there exists a normal operator N on a Hilbert space K ⊃ H such that H is
invariant for N and S = N |H. If, furthermore, K = W
p≥0
N
∗pH, then N is said to be the minimal normal extension of S. In this case, N is unique (up to unitary equivalence) and kN k = kSk.
A bicontraction T = (T
0, T
1) will be called semi-subnormal if one of the contractions is subnormal and the other one has an extension which commutes (therefore doubly commutes) with the minimal normal extension of the subnormal one. Such a bicontraction T has a regular isometric dilation because, if T
0is subnormal and N
1is an extension of T
1commuting with the minimal normal extension N
0of T
0, then we have ∆
T= P
H∆
(N0,N1)|H ≥ 0.
It is easy to see that every subnormal bicontraction is semi-subnormal.
Now we have the following completion of Corollary 10.
Proposition 11. Let T = (T
0, T
1) and T
0= (T
00, T
10) be two semi- subnormal bicontractions on H and H
0respectively, such that T
0and T
00are subnormal and T
10is an isometry. If A is a contraction from H to H
0which intertwines T and T
0and A has an extension which intertwines the minimal normal extensions of T
0and T
00, then A has an extension which intertwines the regular isometric dilations of T and T
0.
P r o o f. Let T, T
0and A be as in the hypothesis and let [ e H, N ] and [ e H
0, N
0], where N = (N
0, N
1) and N
0= (N
00, N
10), be such that N
0(resp. N
00) is the minimal normal extension on e H (resp. e H
0) of T
0(resp.
T
00) and N
1(resp. N
10) is a contraction on e H (resp. e H
0) which extends T
1(resp. T
10) and doubly commutes with N
0(resp. N
00). From the hypothesis and the Fuglede–Putnam Theorem, there is an operator e A : e H → e H
0which doubly intertwines N
0and N
00, and e A|H = A. Then for q ≥ 0 and h ∈ H we have
AN e
1N
0∗qh = ∗ e AN
0∗qN
1h = N
00∗qAT e
1h = N
00∗qAT
1h
= N
00∗qT
10Ah = N
00∗qN
10Ah = N e
10AN e
0∗qh.
Using the structure of the space e H, it results that e AN
1= N
10A. Let us e remark that because T
10is an isometry on H
0, N
10is also an isometry on H e
0, hence the minimal unitary extension [K
0, V
10] of N
10is just the minimal coisometry extension of N
10. Let [K, V
1] be the minimal coisometry extension of N
1and let V
0, V
00be the ∗-extensions of N
0, N
00which doubly commute with V
1, V
10respectively. Since e A intertwines N
1and N
10, there exists a con- traction e A
1from K into K
0which satisfies e A
1V
1= V
10A e
1and e A
1| e H = e A.
In fact, e A
1doubly intertwines V
1and V
10(see [6]). Moreover, e A
1doubly
intertwines the normal operators V
0and V
00, because for {h
n}
n≥0⊂ e H with
finite support we have
A e
1V
0X
n
V
1∗nh
n= X
n
V
10∗nA e
1V
0h
n= X
n
V
10∗nAN e
0h
n= X
n
V
10∗nN
00Ah e
n= X
n
V
10∗nV
00A e
1h
n= V
00A e
1X
n
V
1∗nh
n. Hence e A
1doubly intertwines the bicontractions V = (V
0, V
1) and V
0= (V
00, V
10). Then by Proposition 8, e A
1has an ∗-extension e B which inter- twines the regular isometric dilations M = (M
0, M
1) on M of V and M
0= (M
00, M
10) on M
0of V
0respectively. Setting
K = _
n∈Z2+
M
nH, K
0= _
n∈Z2+