Arxiv:1903.01772V4 [Math.FA] 4 Mar 2021 Dition Ibr Pc Operator Space Hilbert Smti Itnsaduiaydilations

Home , Unitary operator

arXiv:1903.01772v4 [math.FA] 4 Mar 2021 dition ibr pc operator space Hilbert smti itnsaduiaydilations. s unitary direct tha and a and liftings as theorem) isometric written decomposition Wold be (the can operator isometry unitary every a that cons mind further in For keep [39]. to and liftin [17] isometric monographs by celebrated played the role in Theory central Operator the in by study instance, of for objects basic been (opera have contractions one) than space Hilbert of dilations unitary and hoyo -smtishsbe eeoe nrcn er.A years. recent in developed been has 2-isometries of theory O 1A. PRELIMINARIES AND INTRODUCTION 1. k T 2 x .OEAO THEORY OPERATOR J. k IBR PC PRTR IHTWO-ISOMETRIC WITH OPERATORS SPACE HILBERT scaatrzda h etito fabcwr hf naH a on shift operators appli (i.e., are backward tors results a These of functions. ad restriction analytic which the vector-valued operator as an of characterized adjoint is The 2-isometries. to lations smtyi h operator the if isometry steetnin nue yte,aecntutdadisomo 2-is and to constructed are liftings discussed. them, of are by liftings types induced Two extensions the isometries. as or contractions to trigwt eiso he aes[–]b ge n Stan and Agler by [2–4] papers three of series a with Starting e shift let K S 21) 70,4A5 72,47A63 47A20, 47A15, 47A05, (2010): MSC A I 0000) 101–130 00:0(0000), 2 VERVIEW k EYWORDS BSTRACT = − Tx .I hsppr esuyHletsaeoeaoshvn lif having operators space Hilbert study we paper, this In 0. 2 k k 2 Tx k − k otnoslna ibr pc operator space Hilbert linear continuous A . einn ihteS.Ng iainterm smti l isometric theorem, dilation Sz.-Nagy the with Beginning . : 2 x iain,2ioerclfig ocv prtr A-cont operator, concave lifting, 2-isometric dilations, + k 2 k = C x k AT S ˘ aifidb smtis h iihe hf sa example an is shift Dirichlet The isometries. by satisfied 0 2 T uhthat such LNBDAadLUINSUCIU LAURIAN and BADEA ALIN = ˘ hc aifistescn re ifrnecondition difference order second the satisfies which omnctdb aiBercovici Hari by Communicated S o every for 0 n t adjoint its and DILATIONS S ∗ 2 S 2 − x 2 nta ftecascl rtodrcon- order first classical, the of instead , S S ∗ ∗ S aif h relation the satisfy + I ≤ )adt prtr similar operators to and 0) dt ocv opera- concave to ed saduiaydilations unitary and gs oso omn greater no norm of tors oyih yT by Copyright © S mtis swell as ometries, ssi ob 2- a be to said is issc liftings such mits drtosi suseful is it iderations S hsi witnessed, is This . otatoshave contractions t letsaeof space ilbert ato,Dirich- raction, pi minimal rphic ∗ mo hf and shift a of um 2 S ig rdi- or tings 2 2- − isometry 2 u,arich a kus, S ∗ HETA S + iftings 0000 , sa is 102 C.BADEA AND L.SUCIU of a 2-isometry which is not an isometry. Operators arising from a certain class of nonstationary stochastic processes related to Brownian motion (Brownian unitaries) play an essential role in the theory of 2-isometries of Agler and Stankus, the same that unitary operators play for isometries. The fact that a 2-isometry has an extension to a Brownian unitary has been proved in [3, Theorem 5.80]. As an analogue of the Wold decomposition theorem for isometries, it has been proved in [29] (see also [31]) that a pure 2-isometry is unitarily equivalent to a shift operator (multiplication by the independent variable) on a Dirichlet space D(µ) corresponding to a positive operator measure µ on the unit circle. The aim of this paper is to undertake a systematic study of operators pos- sessing 2-isometric liftings. This class of operators, denoted C2, can be viewed as the class of “2-contractions”. To give a flavor of the results obtained in this paper we mention now several sample results. In Theorem 2.1, we give a characterization of adjoints of operators in C2 as restrictions of a backward shift operator on some Hilbert spaces of vector-valued analytic functions. We also prove that operators in C2 are compressions to semi-invariant subspaces of analytic Brownian unitaries. An analogue of the von Neumann inequality is obtained in Theorem 2.1, (iii), with the shift operator on a Dirichlet space D(µ) as an extremal operator. For a bounded linear concave operator, T, acting on a complex Hilbert space , that is one which satisfies the condition H T2x 2 2 Tx 2 + x 2 0 forevery x , k k − k k k k ≤ ∈ H it is proved in Theorem 3.10 below that T has a 2-isometric lifting S, acting on a larger Hilbert space , such that S is an isometry on and S∗S I = K |K⊖H K ⊖ H k − k T T I . Moreover, the 2-isometric lifting S is minimal in the usual sense. k ∗ − k 1B. BASIC DEFINITIONS AND NOTATION. Throughout this paper ( , ) de- B H H′ notes the Banach space of all bounded linear operators acting from a complex Hilbert space into another one, , and ( ) is a short for ( , ). For an H H′ B H B H H operator T ( , ) its adjoint operator in ( , ) is denoted by T , while ∈ B H H′ B H′ H ∗ (T) and (T) stand for the range, respectively the kernel of T. An operator T is R N a contraction if T∗T I, where I = I is the identity operator on . The operator ≤ H H T is an isometry if T T = I and T is unitary if it is an isometry with (T) = . ∗ R H′ For a closed subspace of , the orthogonal projection in ( ) with the range M H B H is denoted by P . The subspace is invariant (reducing) for T ( ) if M M M ∈ B H T (respectively, T and T ). An isometry T on is M⊂M M⊂M ∗M⊂M H pure (or a shift operator) if there is no subspace = 0 in that reduces T to M 6 { } H a unitary operator. An operator T on is said to be positive (in notation T 0) if Th, h 0 H ≥ h i ≥ for every h . Here , denotes the inner product. When T is positive, 1/2 ∈ H h· ·i T stands for the positive square root of T. For a contraction T ( , ′), the 1/2 ∈ B H H operator DT = (I T∗T) ( ) and its closed range T = (DT) are H − ∈ B H D R called the defect operator, respectively the defect space, of T. OPERATORS WITH TWO-ISOMETRIC DILATIONS 103

An operator Z ( , ) is invertible whenever (Z) = 0 and (Z) = ∈B H H′ N { } R . An operator Z ( , ) intertwines T ( ) with T ( ) if ZT = H′ ∈ B H H′ ∈ B H ′ ∈ B H′ T′Z. The operators T and T′ are similar, respectively unitarily equivalent, whenever an intertwining relation ZT = T′Z holds with an invertible, respectively unitary operator Z. Whenever ZT = T′Z with Z an isometry, one says that T′ is an extension of T (to ), or that T is a restriction of T (on ). If this holds one has T Z = Z T H′ ′ H ∗ ∗ ∗ ′∗ and one says that T′∗ is a lifting of T∗. In these cases T is unitarily equivalent to T0′ := T′ (Z), while T∗ is unitarily equivalent to T0′∗. Thus, if we identify |R H with Z into , then T becomes an extension of T from the subspace of H H′ ′ H H′ to ′, while Z is the embedding mapping J , of into ′. In this way we can H H H′ H H consider only extensions and liftings of operators on to another Hilbert spaces H containing . Then the intertwining relations become J , T = T′ J , , respec- H H H′ H H′ tively P , T′∗ = T∗P , , where P , = J∗ ( ′, ) is the projection of H′ H H′ H H′ H , ′ ∈ B H H onto , while the inclusion meansH H that is a closed subspace of . H′ H H ⊂ H′ H H′ According to [17], two liftings S on and S on of an oper- 1 K1 ⊃ H 2 K2 ⊃ H ator T on are called isomorphic if there exists a unitary operator Z ( , ) H ∈B K1 K2 such that ZS1 = S2Z and Z = I, i.e. S1 and S2 are unitarily equivalent by Z |H and Z fixes the elements of . H n For a lifting S on of T one considers the subspace 0 = n 0 S , i.e. K K ≥ H the smallest invariant subspace for S in that contains . Let S0 W= S be the K H |K0 restriction of S to . Clearly S is also a lifting of T on , and it is called a K0 0 K0 minimal lifting of T. Let be a closed subspace of a Hilbert space . The compression of R H K ∈ ( ) to is defined as T = P , RJ , (= P R ). Recall the following useful B K H K H H K H |H result due to Sarason (see [39]). The operator R is a (power) dilation of T, that is, Tn = P Rn for all positive integers n, if and only if the subspace is semi- H | H H invariant for R, that is = for two invariant subspaces and of H H1 ⊖ H2 H1 H2 R. According to [33], if T ( ) is a fixed left invertible operator (i.e. injective ∈B H 1 and with closed range), then T′ = T(T∗T)− is called the Cauchy dual operator of T. Clearly, T′ is also left invertible and T∗T′ = T′∗T = I. So T is the Cauchy dual operator of T and (T ) = (T ) =: . Obviously, is a wandering subspace ′ N ∗ N ′∗ E E for T, that is Tn for each n 1. E ⊥ E ≥ 1C. TWO-ISOMETRIES. In this paper we are interested in 2-isometric liftings for operators in ( ). We refer to [2–7, 11, 19, 23, 24, 26, 29–33, 35, 37] for different B H aspects of 2-isometries. Recall that an operator T on is a 2-isometry if it satisfies the condition H T ∆ T = ∆ , where ∆ = T T I. In this case ∆ is called the operator of ∗ T T T ∗ − T covariance and cov(T) = ∆ 1/2 is the covariance of T. Clearly, cov(T) = 0 if and k Tk only if T is an isometry. If T is a 2-isometry, then ∆T 0, which means that T is an ≥ n < ∞ expansive operator. A 2-isometry T that is power bounded, i.e. supn N T , ∈ k k 104 C.BADEA AND L.SUCIU is necesarily an isometry. Also, (∆ ) is an invariant subspace for a 2-isometry N T T and V = T (∆ ) is an isometry, so the canonical matrix representation of T on |N T = (∆ ) (∆ ) has the form H N T ⊕ R T V E (1.1) T = , 0 Y with V E = 0. This yields ∆ = 0 (E E + ∆ ), with ∆ = E E + ∆ = ∗ T ⊕ ∗ Y 0 ∗ Y ∆T (∆ ) an injective operator. Moreover, using that T∗∆TT = ∆T, one obtains |R T Y∗∆0Y = ∆0. Some special 2-isometries introduced by Agler and Stankus, called Brown- ian isometries and unitaries, play a special role in our investigations. Namely, T is called a Brownian isometry if in the block matrix (1.1) one has E = δE0, where E is an injective contraction from (∆ ) into (V ), δ = cov(T), and Y is a 0 R T N ∗ unitary operator which commutes with E∗E (cf. [3, Proposition 5.37]). Also, T is called a Brownian unitary if E is an isometry with (E ) = (V ) and Y is 0 R 0 N ∗ unitary (cf. [3, Proposition 5.12]). Using the above relationships between T and ∆T, respectively Y and ∆0, we see that 2-isometries are closely related to A-contractions, defined as follows. When T, A ( ) with A 0, the operator T is said to be an A-contraction ∈ B H ≥ if T AT A and T is an A-isometry if T AT = A. Thus, 2-isometries are ∗ ≤ ∗ ∆T-isometries and the concave operators are ∆T-contractions, but many other classes of operators can be viewed as A-contractions for suitable operators A. A- contractions and A-isometries frequently appear in operatorial interpolation and robust control problems, based on the commutant lifting theory (see [13,17,18]), or in other topics of operator theory [10–14, 16, 20–27, 32, 34–38, 40].

1D. ORGANIZATION OF THE PAPER. Following [33], we give in the next section a characterization of the adjoints of operators in the class C2, as restrictions of a backward shift on some spaces of vector-valued analytic functions. This implies that operators in C2 always have analytic 2-isometric liftings. An analogue of the von Neumann inequality for operators in the class C2 is also obtained. Further- more, in the case of analytic Brownian unitaries, we reﬁne the Richter-Olofsson model by taking into account additional spectral information. In Section 3 we describe those operators having 2-isometric liftings among the operators in the class of A-isometries T with A ∆ . Special features of ≥ T such liftings, called of type I, are investigated. We apply our results to a class of expansive operators containing concave operators, and to operators similar to isometries. In particular, for these expansive operators we give a generalization of the extension theorem of Agler and Stankus [3, Theorem 5.80]. The use of the Treil-Volberg generalization of the commutant lifting theorem is to be mentioned here. In Section 4 we study operators T C in the context of A-contractions with ∈ 2 A ∆ . Such liftings, called of type II, are more general than those from Section ≥ T OPERATORS WITH TWO-ISOMETRIC DILATIONS 105

3. We show that these liftings, like those from Section 3, can be always chosen to be minimal. The results can be applied to operators similar to contractions. We also discuss some conditions for two minimal 2-isometric liftings to be isomorphic. Two examples are given in order to show that some operators similar to contractions may or may not have 2-isometric liftings of type I.

1E. AN OPEN PROBLEM. We end this Introduction by mentioning the following problem. Let M > 0 be a real number. Consider the class C2(M) of Hilbert space operators which have 2-isometric liftings of covariance less than or equal to M. Then C2(M) is a family in the sense of Agler’s abstract approach to model theory (see [1]). It is an interesting problem to ﬁnd the boundary and the extremal elements of the family C2(M) (see [1] for the undeﬁned terms). We hope to return to this problem in the future.

2. CHARACTERIZATIONS OF C2 AND ANALYTIC 2-ISOMETRIC LIFTINGS

Assume that the left invertible operator S ( ) is analytic, that is ∈B H n ∞ = S = 0 . H H { } n\1 ≥ 1 Let S′ = S(S∗S)− be the Cauchy dual operator of S. Then, as in [33], one can associate to a Hilbert space of -valued analytic functions Θh, for h , H E ∈ H where = (S ) and E N ∗ n n 1 (Θh)(z) = ∑ (P S′∗ h)z , z < r(S′)− , n 0 E | | ≥ r(S ) being the spectral radius of S . Consider := Θh : h as a Hilbert ′ ′ D { ∈ H} space with the norm induced by . In this case, the operator Θ from onto H H is an isometry such that ΘS = MΘ, where M is the forward shift on , i.e. D D (Md)(z) = zd(z) for d . Also, one has ΘS = BΘ, where B is the backward ∈ D ′∗ shift on , i.e. (Bd)(z) = 1 (d(z) d(0)), d . D z − ∈D It was proved in [33, Theorem 3.6] that a concave operator S, not necessary analytic, admits a Wold-type decomposition, in the sense that n = ∞ S . H H ⊕ n_0 E ≥ In this case the operators S and S′ are simultaneously analytic. When this happens, one can associate to another Hilbert space of -valued analytic func- H D′ E tions Θ′h, where n n (Θ′h)(z) = ∑ (P S∗ h)z , h , z < 1. n 0 E ∈ H | | ≥ The operator Θ isometrically maps onto such that Θ S = M Θ and Θ S = ′ H D′ ′ ′ ′ ′ ′ ∗ B Θ , where M (B ) is the forward (backward) shift on . The space is the ′ ′ ′ ′ D′ D′ 106 C.BADEA AND L.SUCIU dual of with respect to the Cauchy pairing (see also [33]). In this way S can D ∗ be identiﬁed (by Θ) with the adjoint of the forward shift M on , or with the D backward shift B on (by Θ ). It is also known that the space is a Bergman ′ D′ ′ D′ space if and only if ∆ = I S S (see [29] and [19, Theorem 3.1]). S − ′ ′∗ In addition, if S is a 2-isometry, then, by [29], the associated space be- D comes a Dirichlet type space (µ) of -valued analytic functions on the open D E unit disk D obtained with respect to some positive ( )-valued operator mea- B E sure µ on the unit circle T. We refer to [29, (3.1)] for a description of the norm in (µ) induced by µ. D In general, if M is the forward shift on a Dirichlet type space (µ), then M is D an analytic 2-isometry (see [29, Theorem 3.1]), while the operator M′ can be seen as the forward shift on the associated Hilbert space of (M )-valued analytic D′ N ∗ functions on D, D being the dual of (µ) with respect to the Cauchy pairing. As ′ D above, the backward shift B in is unitarily equivalent to M in (µ). ′ D′ ∗ D The following result characterizes membership into the class C2.

THEOREM 2.1. For T ( ) the following statements are equivalent: ∈B H (i) T is an element of the class C2, that is T has a 2-isometric lifting; (ii) T has an analytic 2-isometric lifting; (iii) there is an operator-valued positive measure µ on T such that the inequality

pij(T) pij(Mz) n( ( )) ≤ n( ( (µ))) M B H M B D

holds true for all finite matrices of polynomials pij , where Mz is the multiplication by the variable z on (µ) and ( ( )), theset of n n matrices D Mn B H × with entries in ( ), is identified with the set of bounded linear operators act- B H ing on (n), the ℓ -sum of n copies of . H 2 H (iv) T has a Brownian unitary (power) dilation; (v) T has an analytic Brownian unitary dilation; (vi) T∗ is unitarily equivalent to the restriction to an invariant subspace of the backward shift B on a Hilbert space of vector-valued analytic functions on D, ′ D′ being in Cauchy pairing to a Dirichlet type space (µ). D′ D Moreover, if these conditions hold, then the lifting in (ii) for T can be chosen minimal. Also, the liftings and Brownian unitary dilations in (i), (ii), (iv) and (v) can be chosen to have the same covariance. Proof. The implication (i) (iv) follows from [3, Theorem 5.80], using the ⇒ fact that an extension of a lifting for T gives a (power) dilation for T. Let us now assume (iv), that is T has a Brownian unitary dilation S on K ⊃ . Because S is not necessarily analytic we can write S as a direct sum S = U0 H n ⊕ S1 on = ∞ 1, where ∞ = n 1 S and U0 = S ∞ is unitary, while K K ⊕K K ≥ K |K S1 = S is an analytic Brownian unitary.T But U0 can be lifted to a Brownian |K1 unitary S0 on a Hilbert space 2 := ℓ ( (E)) ∞, K0 + R ⊕K OPERATORS WITH TWO-ISOMETRIC DILATIONS 107 where E ( ∞) is an isometry. Then S has the following block matrix form ∈B K 0 S E S = + , 0 0 U e0 2 with S+ the forward shift on = ℓ ( (E)) and E = δJE, where δ = cov(S) and K0′ + R J is the embedding of (E) into . Clearly, S has not any isometric summand R K0′ 0 e (S+ being a shift with R(E) = (S )), hence S is analytic. N +∗ 0 Consider the operator S = S S on := . To see that S is 0 ⊕ 1 K K0 ⊕K1 a Brownian unitary operator we use the matrix representation of S on = e e 1 K1e (∆ ) (∆ ) =: (similar to that of S from above) given by a for- N S1 ⊕ R S1 K2 ⊕K3 0 ward shift S on , a unitary operator U on and an operator E = δE , with +′ K2 1 K3 1 1 δ as above and E an isometry from onto (S ). Thus we can represent S on 1 K3 N +′∗ e = ∞ =( ) ( ∞ ), respectively, in the form K K0′ ⊕K2 ⊕K ⊕K3 K0′ ⊕K2 ⊕ K ⊕K3 e e S+ 0 E 0  0 S 0 E  S G (2.1) S = +′ e 1 = ′ . 0 0 U 0 0 U  0 e  e e  0 0 0 U   1 Here S = S+ S is a forward shift, U = U U is a unitary and G = δG with ′ ⊕ +′ 0 ⊕ 1 G an isometry from ∞ onto (S ) in , δ being as above. Hence K ⊕K3 N ′∗ K0′ ⊕K2 e S is an analytic Brownian unitary with cov(S) = δ = cov(S). Since S is a power dilation for T we infer from the representation (2.1) that S is also a power dilation e e for T. This concludes the proof that (iv) implies (v). e It is a known fact that from a power dilation S for T one can obtain a lifting S for T, as a restriction of S to an invariant subspace (see [17]). Furthermore, if S e S S S is Brownian unitary, then e is a 2-isometry, and is analytic if is, because thee analyticity is preserved on invariant subspaces. Hence (v) implies (ii). e Next, we assume (ii). Let S be an analytic 2-isometric lifting on for T. K ⊃ H Then, by [29, Theorem 4.1], and using the same notation as in the Introduction, we obtain a ( )-valued positive measure µ on the unit circle such that S is unitarily B E equivalent to the forward shift M = M on a Dirichlet space D(µ) of -valued z E analytic functions on D, where = (S ). It follows that M is an extremal E N ∗ operator in the sense of (iii). Conversely, if (iii) holds, then, using the terminology of [8], T is completely polynomially dominated by M. Using [8, Theorem 2.1.3], we get that T is unitarily equivalent to the compression of an operator R to a semi-invariant subspace, R being the image of M by a unital C∗-representation. It follows that R is a 2-isometry, and thus T is in the class C2, that is T satisfies (i). So the equivalences (i)-(v) are proved. Next, we assume (ii). Then, again, S is an analytic 2-isometric lifting on for T, unitarily equivalent to the forward shift M on a Dirichlet space K ⊃ H D(µ) of -valued analytic functions on D. Also, S is unitarily equivalent to the E ∗ backward shift B on the Hilbert space = Θ k : k of -valued analytic ′ D′ { ′ ∈ K} E 108 C.BADEA AND L.SUCIU functions of the form

n n (Θ′k)(z) = ∑ (P S∗ k)z , z < 1. n 1 E | | ≥

Since S∗ = T∗, it follows that B′Θ′h = Θ′T∗h for h . Hence, the closed |H ∈ H subspace = Θ h : h is invariant for B . Because the operator Θ from D0′ { ′ ∈ H} ′ ′ K onto ′ is an isometry and Θ′S∗ = B′Θ′, we infer that Θ0′ = Θ′ is an isometry D |H from onto 0′ and Θ0′ T∗ =(B′ )Θ0′ . Hence T∗ is unitarily equivalent to B′ , H D |D0′ |D0′ and we conclude that (ii) implies (vi). Conversely, let us assume that T∗ is unitarily equivalent to B0′ = B′ , |D0′ where B is as in (vi) and is an invariant subspace of B . The Cauchy pairing ′ D0′ ′ between the space , where B acts, and the Dirichlet type space (µ) ensures D′ ′ D that B is unitarily equivalent to M , the adjoint of the forward shift on (µ). ′ ∗ D Since M is a 2-isometry on (µ) by [29, Theorem 3.1], it induces a 2-isometry B D ′∗ on which is a lifting for B (B being an extension of B ). In the same time, D′ 0′∗ ′ 0′ T is unitarily equivalent to B0′∗ by our assumption. We deduce that T has B′∗ as a analytic 2-isometric lifting on . We have thus shown that (vi) implies (ii). In D′ conclusion, all equivalences (i)-(vi) are now proved. Remark that, by [3, Theorem 5.80], a 2-isometry S has a Brownian unitary extension S which preserves the covariance of S, and, as we already have seen, cov(S) can be also preserved for an analytic Brownian unitary extension of S. On e the other hand, if S is a 2-isometric lifting for T, then, as in the proof of implication (iv) (v), S can be lifted to an analytic 2-isometry of the same covariance as S. ⇒ Thus in all assertions (i), (ii), (iv) and (v) we can obtain 2-isometric liftings and dilations for T of the same covariance. Finally, if S is an analytic 2-isometric lifting for T on (as in (ii)), n K ⊃ H then 0 = n 0 S is an invariant subspace for S and S 0 is a minimal ana- K ≥ H |K lytic 2-isometricW lifting for T with δ0 = cov(S ) cov(S). However, by the |K0 ≤ above discussion, we can get an analytic Brownian unitary dilation S for T with (S) = δ cov 0. This ends the proof. e e COROLLARY 2.2. If T ( ) is an operator similar to a contraction, then T ∈ B H and T∗ are restrictions to invariant subspaces of backward shifts on Hilbert spaces of vector-valued analytic functions on D. The following theorem reﬁnes the model of A. Olofsson [29] for analytic Brownian unitaries. By Theorem 2.1, (v), every operator in C2 has an analytic Brownian unitary dilation.

THEOREM 2.3. Let S ( ) be an analytic Brownian unitary acting on . Let ∈ B K K = (S ) and let µ be the positive ( )-valued measure deﬁned on the σ-algebra E N ∗ B E Bor(T) of all Borel subsets of T by T (2.2) µ(σ) = P F(σ)∆S , σ Bor( ), E |E ∈ OPERATORS WITH TWO-ISOMETRIC DILATIONS 109 where F is the ( (∆ ))-valued spectral measure on T of the unitary operator U with B R S U∗ = S∗ (∆ ). Then S is unitarily equivalent to the forward shift on the Dirichlet space |R S (µ) = Θk : k of -valued analytic functions on D, where D { ∈ K} E n n (2.3) (Θk)(z) = ∑ (P S′∗ k)z , z D, n 0 E ∈ ≥

S being the Cauchy dual operator of S, and the norm on (µ) is induced by µ as in [29, ′ D (3.1)]. 1 Moreover, if E0 = δ− P (∆ )S (∆ ) and δ = cov(S), then N S |R S

1 (2.4) = δ− E U∗d d : d (∆ ) E {− 0 ⊕ ∈ R S } and (1 + δ2)δ 4µ(σ) is an orthogonal projection, for every σ Bor(T). − ∈ Proof. Let U be the unitary operator on (∆S) with U∗ = S∗ (∆ ), and R |R S let F : Bor(T) ( (∆ )) be the spectral measure of U. Since ∆ = 0 δ2 I → B R S S ⊕ on (∆ ) (∆ ), where δ = cov(S), we have ∆1/2Sk = ∆1/2Uk = U∆1/2k N S ⊕ R S S S S for k (∆ ). The second relation ensures that ∆1/2F(σ)k = F(σ)∆1/2k for ∈ R S S S k (∆ ). Thus, as in [29, Lemma 4.1], we can associate to S a positive ( )- ∈ R S B E valued measure on T, which in this case has the form (2.2), where = (S ). E N ∗ Also, corresponding to this measure µ, one can deﬁne the Dirichlet space (µ) D with the norm induced by µ as in [29, (3.1)]. By [29, Theorem 4.1] we obtain that S is unitarily equivalent, via the operator Θ : (µ) from (2.3), with the K→D forward shift on (µ). D In addition, in this case the subspace of can be easily determined and E K we are able to obtain more information about the measure µ. Indeed, let us consider the canonical matrix representation of the form (1.1) of S on = (∆ ) K N S ⊕ (∆ ), where V = S is a shift operator (S being analytic), Y = U is as R S | (∆S) before, and E = δE , withN E an isometry from (∆ ) onto (V ). Then, an 0 0 R S N ∗ element k = k k (∆ ) (∆ ) belongs to = (S ) if and only if 0 ⊕ 1 ∈ N S ⊕ R S E N ∗ V k = 0 and δE k + U k = 0. Since (V ) = (E ), it follows that k = E d ∗ 0 0∗ 0 ∗ 1 N ∗ R 0 0 0 0 for some element d (∆ ), while the previous equality yields k = δUd . 0 ∈ R S 1 − 0 Considering d of the form d = δ 1U d with d (∆ ), we infer that has 0 0 − − ∗ ∈ R S E the form (2.4). Now let e and d (∆ ) be such that e = δ 1E U d d. Then ∈E ∈ R S − − 0 ∗ ⊕ 2 2 2 (1 + δ− )d = e δ− (δE U∗d UU∗d) = e δ− S(U∗d). ⊕ 0 ⊕ ⊕ It follows that

δ2 δ2 1 P d = e = ( E0U∗d d), d (∆S). E 1 + δ2 1 + δ2 − δ ⊕ ∈ R 110 C.BADEA AND L.SUCIU

Thus, for σ Bor(T) and e as above, we obtain ∈ ∈E 2 2 µ(σ)e = P F(σ)∆Se = δ P F(σ)P (∆ )e = δ P F(σ)d E E R S E δ4 1 = ( E U∗F(σ)d F(σ)d) 1 + δ2 − δ 0 ⊕ δ4 1 = ( E F(σ)E∗E U∗d F(σ)d) 1 + δ2 − δ 0 0 0 ⊕ δ4 = (E0F(σ)E0∗P (∆ )e F(σ)P (∆ )e) 1 + δ2 N S ⊕ R S δ4 = (E F(σ)E∗ F(σ))(e e ), 1 + δ2 0 0 ⊕ 0 ⊕ 1 where e0 = P (∆ )e, e1 = P (∆ )e. Finally, keeping in mind that F is a spectral N S R S measure, so F(σ)2 = F(σ), we obtain 2 1 + δ2 µ(σ) e = (E F(σ)E∗ F(σ))(E F(σ)E∗e F(σ)e ) δ4 0 0 ⊕ 0 0 0 ⊕ 1 2 2 = E F(σ) E∗e F(σ) e = E F(σ)E e F(σ)e 0 0 0 ⊕ 1 0 0 0 ⊕ 1 1 + δ2 = µ(σ)e, δ4 for every σ Bor(T) and e . Since (1 + δ2)δ 4µ(σ) is a contraction, we ∈ ∈ E − conclude that (1 + δ2)δ 4µ(σ) is an orthogonal projection for σ Bor(T). This − ∈ ends the proof.

REMARK 2.4. V. Müller proved in [28, Corollary 2.3] that if is a separable H Hilbert space, then, for any operator T ( ), T and T are unitarily equivalent ∈B H ∗ with restrictions to invariant subspaces of a backward weighted shift B ( ) α ∈B K where = ℓ2 ( ) and is a separable Hilbert space. Recall that such an K + H′ H′ operator Bα is deﬁned with respect to a bounded sequence of positive numbers α = αn n 1 by the relation Bα( xn n 0) = αnxn n 1 for all square summable { } ≥ { } ≥ { } ≥ sequences xn n 0 of vectors xn ′. So the operators T and T∗ can be { } ≥ ∈ K ∈ H lifted to the forward weighted shift S = B on , which is an analytic operator, α α∗ K but Sα is not 2-isometric, in general.

3. TWO-ISOMETRIC LIFTINGS OF TYPE I

3A. TYPE I ANDTYPE II LIFTINGS. Let T ( ) and suppose that S ( ) is ∈ B H ∈B K a lifting of T on . Then the canonical representation of S on = K ⊃ H K H⊥ ⊕ H has the form W X (3.1) S = , 0 T OPERATORS WITH TWO-ISOMETRIC DILATIONS 111 with W = S and X = P S . In general, the operators W, X and T can be | ⊥ ⊥ |H arbitrary, butH they have to satisfyH some constraints whenever S belongs to some particular classes of operators. Here we are interested in the case when S is a 2-isometry. In this case, W is also a 2-isometry (as a restriction to an invariant subspace for S), and ∆S has the form

∆W W∗X (3.2) ∆S = . X∗W X∗X + ∆T Since necessarily ∆ 0 and ∆ 0, one also has X X + ∆ 0. Therefore S ≥ W ≥ ∗ T ≥ (see [17], [18]), there exists a contraction Γ : (X X + ∆ ) (∆ ) such that R ∗ T → R W 1/2 1/2 (3.3) W∗X = ∆W Γ(X∗X + ∆T) .

This relation gives X∗W (∆ ) = 0, which by (3.2) implies that ∆S (∆ ) = 0, |N W |N W hence (∆W) (∆S). But (∆W) is invariant for W = S and so for S, N ⊂ N N |H⊥ while W (∆ ) = S (∆ ) is an isometry. When W is an isometry on ⊥, one |N W |N W H has = (∆ ) (∆ ). Therefore W X = 0 in this case. Conversely, if H⊥ N W ⊂ N S ∗ ⊥ (∆S), then, since ⊥ is invariant for S by (3.1), it follows that W = S H ⊂N H | ⊥ is an isometry. H On the other hand, we see from (3.3) that W∗X = 0 if and only if Γ = 0, because (Γ) (∆ ). Notice however that, in general, it is difficult to R ⊂ R W have significant information about the operator Γ. Using (3.2), one has W∗X = 0 if and only if is invariant for S S. In this case, by (3.2), we have (∆ ) = H ∗ N S (∆ ) (X X + ∆ ). In particular, if X X + ∆ = 0, which forces T and N W ⊕N ∗ T ∗ T X to be contractions (as D2 = X X 0), then W X = 0. In this case it is T ∗ ≥ ∗ easy to see that S is an extension of the minimal isometric lifting of T, because (∆ ) = (∆ ) is an invariant subspace for S. N S N W ⊕ H In light of the preceding discussion, we introduce the following definition.

DEFINITION 3.1. Using the previous notation, we say that a 2-isometry S of the form (3.1) on = is a lifting of type I for T whenever (∆ ). K H⊥ ⊕ H H⊥ ⊂ N S We say that S is a lifting of type II for T whenever is an invariant subspace for H S∗S. Observe that S S if and only if W X = 0 (in (3.3)), hence a lifting of ∗ H ⊂ H ∗ type I is also a lifting of type II.

3B. A-ISOMETRIES. We describe now the operators which have 2-isometric liftings of type I. These operators form a special class of A-isometries as follows.

THEOREM 3.2. For T ( ) the following statements are equivalent: ∈B H (i) T has a 2-isometric lifting of type I on a Hilbert space containing ; H (ii) T is either an A-isometry for a positive operator A = 0 on with ∆ A, or 6 H T ≤ T is a strongly stable contraction; 112 C.BADEA AND L.SUCIU

(iii) T has an extension T on a Hilbert space with T of the form M ⊃ H e C δE e (3.4) T = 0 U e on a decomposition = , where C, E are contractions, U is uni- M M0 ⊕M1 tary, and δ is a positive scalar, such that there exist a Hilbert space and M′ isometries J : ,J : satisfying the condition 0 DC →M′ 1 DE →M′ (3.5) DC J0∗ J1DE + C∗E = 0; (iv) T has a 2-isometric lifting S of the form (3.1) such that

(3.6) X∗∆W X + 2Re(X∗W∗XT) = 0. Moreover, if these statements are true, then the lifting S of T in (i) and (iv) can be chosen minimal with cov(S) = A 1/2 for A from (ii), or with cov(S) δ for δ from k k ≤ (3.4). Proof. Let S ( ) be a 2-isometry as in (3.1), with (∆ ). Thus ∈B K K⊖H⊂N S W = S is an isometry. Then ∆W = 0 and, as we noticed before, one has |K⊖H W∗X = 0. Therefore, the condition (3.6) from (iv) is satisﬁed for such a lifting S of T. We obtain that (i) implies (iv). Assume that T has a 2-isometric lifting S on of the form (3.1) with K ⊃ H the operators W and X. Then S∗∆SS = ∆S, and this implies (by (3.1) and (3.2)) the relation

(3.7) X∗∆W X + 2Re(X∗W∗XT) + T∗(X∗X + ∆T)T = X∗X + ∆T. Now, if S veriﬁes the condition (3.6), then one obtains from (3.7) that T is an A- isometry, where A = X X + ∆ 0 (as ∆ 0 in (3.2)), and so A ∆ . This ∗ T ≥ S ≥ ≥ T condition in the case A = 0 forces T to be a contraction, but in this case T is an n n 2 A0-isometry with A0 = s limn ∞ T∗ T ∆T = D . Here the case A0 = 0 − → ≥ − T corresponds to T being strongly stable, i.e. Tnh 0 for h . We have k k → ∈ H proved that (iv) implies (ii). Next, let T be an operator as in (ii), i.e. satisfying T AT = A with 0 = A 0 ∗ 6 ≥ and A ∆ . Then the operator A := A ∆ is positive and one can suppose ≥ T T − T A = 0, or in other words, T is not a 2-isometry. T 6 We deﬁne the lifting S of T on the space := ℓ2 ( (A )) with A,T HA,T + R T ⊕ H the block matrix S A (3.8) S = + T , A,T 0 T e 2 1/2 where S+ is the forward shift on ℓ ( (A )) and A = JA , J being the canon- + R T T T ical injection of (A ) into ℓ2 ( (A )). Then, on the above decomposition of R T + R T e , we have HA,T ∆ = 0 (A + ∆ ) = 0 A. SA,T ⊕ T T ⊕ OPERATORS WITH TWO-ISOMETRIC DILATIONS 113

Therefore, S∗ ∆ S = 0 T∗ AT = 0 A = ∆ . A,T SA,T A,T ⊕ ⊕ SA,T Hence SA,T is a 2-isometry and SA,T = S+ is an isometry. We conclude |HA,T⊖H that SA,T is a 2-isometric lifting of type I for T. In the case when T is a contraction, it has even an isometric lifting (so of type I and of covariance zero as a 2-isometry). In this case, if T AT = A = 0, ∗ 6 then A1/2 D . Therefore (A ) = (R(A ) ). Using this and (3.8), T ≥ T R T DT ⊕ T ⊖DT one infers that SA,T is a 2-isometric lifting of type I for T and for the minimal isometric lifting of T. But when T is a strongly stable contraction, T cannot be an A-isometry with A = 0, so A = D2 and S is even the minimal isometric 6 T T A,T lifting of T. Hence (ii) implies (i), and thus the assertions (i), (ii) and (iv) are equivalent. We also remark that cov(S ) = A 1/2 = 0 and thus cov(S ) = A,T k k 6 A,T 0 = A , when T is a strongly stable contraction. k 0k Now we show that SA,T is a minimal lifting of T, that is, we have A,T = n H∞ n 0 SA,T . To see this, let k = d h A,T with h and d = 0 dj ≥ H ⊕ ∈ H ∈ H ∈ ℓW2 n L +( (AT)) be such that k is orthogonal to n 0 SA,T . As k , it follows that R ≥ H ⊥ H h = 0. So k = d S = A T . WorkingW in the space (A ), we have ⊥ A,TH TH ⊕ H R T d A1/2 , and thus d = 0. By induction, one obtains d = 0 for each j 1; 0 ⊥ T H 0 e j ≥ consequently k = d = 0. Thus the minimality condition is true. The ﬁnal part of the proof consists in showing the equivalence of (i) with (iii). Assume that S is a 2-isometric lifting of T on of covariance δ, with V = K S an isometry. Without loss of generality we can assume δ > 0 (otherwise, |K⊖H S is an isometry and so T is a contraction, hence T has trivially the form (3.4)). Then, by [3, Theorem 5.80], S has a Brownian unitary extension S on of K⊃K covariance δ with the canonical representation on = (∆ ) (∆ ) of the K N S ⊕e R eS form e e e V δE S = . 0 U e e e Here V, E are isometries with (V ) = (E) and U is unitary. N ∗ R Now, S, as an extension of S, has on the decomposition =( ) e e e e K K ⊖ H ⊕ H ⊕ ( ) a block matrix of the form K⊖K e e e V ⋆ ⋆ S =  0 T ⋆ . 0 0 ⋆ e   Since S = V is an isometry and S is a 2-isometry, it follows that ′ := |K⊖H H (∆ ) and V = V. Inserting V into the matrix representation of S K⊖H⊂Ne S |H′ e on = ( (∆ ) ) (∆ ) we get K H′ ⊕ Ne S ⊖e H′ ⊕ R S e e e eV C′ δE′ S =  0 C δE  . 0 0 U e   114 C.BADEA AND L.SUCIU

Here we have represented the isometries V on , where = (∆ ) H′ ⊕M0 M0 N S ⊖ , and E from = (∆ ) into , as H′ M1 R S H′ ⊕Me0 e e e V C E V = ′ , E = ′ , 0 C E e e where C, C′, E and E′ are contractions satisfying

V∗E = V∗E′ (C′∗E′ + C∗E) = 0. ⊕ Also, we have V∗C′ = 0e becausee V and V are isometries. Comparing the two 3 3 matrices of S we infer × e T ⋆ e C δE H = P S = M0 , 0 ⋆ ′ | ′ 0 U K⊖K K⊖H K⊖H M1 e e e the two block matricese being given on two different decompositions of := M . Hence T has an extension on = of the form (3.4). K ⊖ H′ M M0 ⊕M1 It remains now to verify the condition (3.5). Indeed, since V and E are e isometries, we have C∗C + C′∗C′ = I and E∗E + E′∗E′ = I . Therefore, M0 M1e e C C = D2 and E E = D2 . Since V C = 0 and V E = 0, we have (C ) ′∗ ′ C ′∗ ′ E ∗ ′ ∗ ′ R ′ ∪ (E ) (V ) =: . Thus we infer by polar decompositions that C = J D R ′ ⊂ N ∗ M′ ′ 0 C and E = J D , where J , J are isometries from , respectively from , into ′ 1 E 0 1 DC DE . So, J and J are the canonical mappings of = (C ) onto (C ), re- M′ 0 1 DC R ′∗ R ′ spectively of = (E ) onto (E ), in the space . Finally, the condition DE R ′∗ R ′ M′ V∗E = 0 implies C′∗E′ + C∗E = 0, which becomes DC J0∗ J1DE + C∗E = 0, i.e. the condition (3.5). We completed the proof that (i) implies (iii). e e Conversely, let us assume that T has an extension T of the form (3.4) on = , as in (iii). We denote by D the operator D = J D J D M M0 ⊕M1 e 0 C 1 E from into . Consider also the operator J from into := M0 ⊕M1 M′ M′ M2 ℓ2 ( (D)) deﬁned as the canonical injection of (D) into . Let D = JJ D , + R R M2 C 0 C D = JJ D , and let S be the (minimal) lifting of T acting on the space := E 1 E e M ( ) deﬁned by the block matrix eM2 ⊕M0 ⊕M1 b e f V δF S = ′ . 0 U b Here V on and F from into have the representations ′ M2 ⊕M0 M1 M2 ⊕M0 S D D V = + C , F = E , ′ 0 C E e e where S+ is the forward shift on , while U, C, E and δ > 0 are as in (3.4). It is M2 clear that V′ and F are isometries and that

V′∗F = S∗ D (D∗ D + C∗E) = 0 (D J∗ J D + C∗E) = 0, + E ⊕ C E ⊕ C 0 1 E e e e OPERATORS WITH TWO-ISOMETRIC DILATIONS 115 taking into account that (S ) = J (D) (D ) and using the relation (3.5). N +∗ R ⊃ R E Thus, one obtains ∆ = 0 δ2 I , and we conclude that S is a 2-isometry, in fact S ⊕ M1 e even a Brownian isometry with δ 2∆ an orthogonal projection. b − S b Expressing now T in the matrix of its extension T on = ( ) b M H ⊕ M ⊖ H in the form T ⋆ e T = , 0 ⋆ e and then inserting T in the matrix of S (as a lifting of T), we infer that the subspace ′ := 2 is invariant for S and that S′ = S is a 2-isometric lifting for T H M ⊕ H e b |H′e with S′ = S+ an isometry. So T satisﬁes (i). In addition, since S′∗S′ , we |M2 b b H ⊂ H get ∆S = P ∆ and cov(S′) δ. If 1 = 0 , then cov(S′) = δ, while if ′ H S|H ≤ H∩M 6 { } 0, then T is a contraction and S′ = V′ is an isometry with cov(S′) = 0. H⊂M b |H′ Notice that, by construction, S′ is not minimal. One can consider S0 := S′ |K0 where = S n . Since is reducing for S (see Remark 4.10 below), it K0 n 0 ′ H K0 ′ follows that SW is≥ a minimal 2-isometric lifting of type I for T with cov(S ) δ. 0 0 ≤ Hence (iii) implies (i) and all assertions are now proved. It is easy to see that every operator T of the form (3.4) on = is a H H0 ⊕ H1 P -isometry. In addition, one has H1 2 2 DC δC∗E δ P ∆T = −2 2 . H1 − δE C δ D − ∗ E Thus, when the condition (3.5) holds, i.e. C∗E = DC J0∗ J1DE, the above matrix 2 − is positive. Hence δ P ∆T. Applying for such an operator T the arguments H1 ≥ used in the proof of equivalences of (i) with (ii) and (iii), we deduce the following result.

COROLLARY 3.3. For T ( ) the following statements are equivalent: ∈B H (i) T has the form (3.4) and the condition (3.5) holds true; (ii) T is a P-isometry for an orthogonal projection P with δ2P ∆ and some ≥ T scalar δ > 0; 2 (iii) T has a (minimal) Brownian isometric lifting S of type I, with δ− ∆S an orthogonal projection and δ = cov(S). An application of Theorem 3.2 concerns the (A,2)-expansive operators stud- ied in [20], which in fact are the ∆ (T)-contractions, where ∆ (T) = T AT A A A ∗ − and A is a positive operator. We obtain the following consequence.

COROLLARY 3.4. Let T ( ) be an invertible (A,2)-expansive operator such ∈B H 1 that A ∆T or A ∆ 1 . Then T, respectively T− , have minimal 2-isometric liftings ≥ ≥ T− of type I.

Proof. From hypothesis and [20, Theorem 3.10 (ii)], it follows that T∗ AT = 1 1 A, or, equivalently, (T∗)− AT− = A. Thus one can apply Theorem 3.2 to T if 1 A ∆T, respectively to T− if A ∆ 1 , to obtain the conclusion. ≥ ≥ T− 116 C.BADEA AND L.SUCIU

REMARK 3.5. Theorem 3.2 can be seen as a generalization of the well-known theorem of isometric lifting of a contraction (see [17, 39]). The operator SA,T in (3.8) is a minimal 2-isometric lifting of T on the space = ℓ2 ( (A )) , K + R T ⊕ H while in the case when T is a contraction (corresponding to A = 0 in (ii)), the operator S0,T is the minimal isometric lifting of T. But the 2-isometric lifting SA,T with A = 0 is not uniquely determined by the minimality condition, up to unitary 6 equivalence which ﬁxes , as we will see in the next section. This happens even H n n for contractions T with A := s limn ∞ T∗ T = 0 when we consider the corre- − → 6 sponding lifting S . In this case T AT = A ∆ = D2 . Then A = A + D2 A,T b ∗ ≥ T − T T T and S is a minimal 2-isometric lifting of T of covariance A 1/2 > 0, so S is A,T b b b k k b b A,T not isometric. Hence S cannot be unitarily equivalent to S . b A,T b0,T b REMARK 3.6. Theb condition (3.5) does not imply the condition C∗E = 0, in general, but the converse implication holds by choosing J0, J1 such that the subspaces J and J are orthogonal in . Therefore, the operators T of 0DC 1DE M′ the form (3.4) with the condition C∗E = 0 form a special class of operators with Brownian isometric liftings of type I having the covariance operators the scalar multiples of orthogonal projections (by Corollary 3.3). We mention here two special cases. If C is a coisometry in (3.4), then (3.5) implies E = 0, hence the operator C U is a contraction. Also, if C or E are isometries, then the two conditions (3.5) ⊕ and C∗E = 0 are simultaneously satisﬁed.

3C. EXPANSIVE OPERATORS. The case when C, the upper left entry of the matrix (3.4), is an isometry is related to expansive operators. Recall that T is said to be expansive if T T I. ∗ ≥ THEOREM 3.7. For T ( ) the following statements are equivalent : ∈B H (i) T is expansive and has a minimal 2-isometric lifting of type I; (ii) T is an A-contraction for some A ( ) such that A ∆ 0; ∈B H ≥ T ≥ (iii) T has an extension T, deﬁned on a Hilbert space , which on an orthogonal K decomposition = has the form K Ke 0 ⊕K1 V δE (3.9) T = , 0 U e with V an isometry on , U a unitary operator on , E a contraction from K0 K1 into satisfying V E = 0, and δ ∆ 1/2. K1 K0 ∗ ≥ k Tk Moreover, if these statements are true, then one can chose T in (3.9) with δ = A 1/2 for A as in (ii). k k e Proof. We may assume that T is non-isometric, i.e. ∆ = 0. It is clear that T 6 (i) implies (ii) by Theorem 3.2. Conversely, let us assume that T AT A with ∗ ≤ A ∆ 0, as in (ii). Then there exists a contraction T on (A) such that ≥ T ≥ R TA1/2h = A1/2Th for h . In fact, deﬁning the operator A : (A) by ∈ H b 0 H → R A h = A1/2h, h , we have A T = TA . b 0 ∈ H 0 0 b OPERATORS WITH TWO-ISOMETRIC DILATIONS 117

Let V on (A) be the minimal isometric lifting of T. Then, as T T I, H ⊃ R ∗ ≥ by the Treil-Volberg generalization of the commutant lifting theorem (see [13], b b b [18], [40]), there exists an operator A ( , ) with A = A such that ∈B H H k k k 0k b b b P , (A)A = A0, AT = VA. H R b b b b b Deﬁning B ( ) by B = A A, we have T BT = B, i.e. T is a B-isometry with ∈ B H ∗ ∗ B = A and B A A = A ∆ 0. Hence (ii) implies (i) by Theorem 3.2. k k k k ≥ 0∗ 0 b b≥ T ≥ To show that (i) implies (iii) we can suppose (by Theorem 3.2, (i)) that T is a B-isometry with B ∆ 0. Then there exists an isometry W on (B) satisfying ≥ T ≥ R the relation WB1/2 = B1/2T. Let us denote B = B 1B. Since T is expansive and a B -isometry, we 0 k k− 0 have

T∗(I B )T = T∗T B I B , − 0 − 0 ≥ − 0 and then, by Douglas’ lemma [15], one obtains a contraction C on satisfying H the relation 1/2 1/2 (I B ) = T∗(I B ) C. − 0 − 0 This, together with the relation B ∆ , yield the inequality ≥ T 1/2 1/2 T∗(I B ) (I CC∗)(I B ) T = T∗T I B = B B . − 0 − − 0 − ≤ k k 0 Consider V on 0 to be the minimal isometric lifting of C. Therefore K ⊃ H 1/2 P , V = CP , . Deﬁne the linear operator E0 : (B ) 0 by the relation K0 H K0 H R →K 1/2 1/2 E (B h)=(I VV∗)(I B ) Th, h . 0 0 − − 0 ∈ H Since V∗ = C∗, from the above relations one obtains |H 1/2 2 1/2 1/2 E B h = (I VV∗)(I B ) Th, (I B ) Th k 0 0 k h − − 0 − 0 i 1/2 1/2 = (I CC∗)(I B ) Th, (I B ) Th h − − 0 − 0 i = ∆ h, h Bh, h . h T i≤h i Setting E := δ 1E , where δ = B 1/2 ∆ 1/2, the previous inequality be- 1 − 0 k k ≥ k Tk comes E B1/2h B1/2h , h . k 1 k ≤ k k ∈ H So E can be continuously extended to a contraction, also denoted E , from (B) 1 1 R into . In addition, by the deﬁnition of E , one has V E = 0. K0 0 ∗ 1 Let now U be the minimal unitary extension of the above isometry W on

n 1 = U∗ (B), K n_0 R ≥ and let E : be a contractive extension of E to , for example E = E P K1 →K0 1 K1 1 where P = P , (B) is the projection of 1 onto (B). Clearly, one has V∗E = 0. K1 R K R 118 C.BADEA AND L.SUCIU

Consider the Hilbert space = and the operator T ( ) having K K0 ⊕K1 ∈B K the block matrix (3.9) with the above operators V, U, E and number δ. Deﬁne now e the operator Z : with (Z) (B), by the relation H→K R ⊂K0 ⊕ R Zh =(I B )1/2h B1/2h, h . − 0 ⊕ 0 ∈ H

Obviously, Z is an isometry. We now show that ZT = TZ. We have U (B) = W |R and E (B) = E1, so |R e ZTh =(I B )1/2Th B1/2Th =(I B )1/2Th UB1/2h. − 0 ⊕ 0 − 0 ⊕ 0 On the other hand, using (3.9), we have TZh =[V(I B )1/2h E B1/2h] UB1/2h. − 0 ⊕ 0 0 ⊕ 0 But V∗ = C∗, ande from the deﬁnitions of E0 and C we get |H 1/2 1/2 1/2 (I B ) Th = VV∗(I B ) Th (I VV∗)(I B ) Th − 0 − 0 ⊕ − − 0 1/2 1/2 1/2 1/2 = VC∗(I B ) Th E B h = V(I B ) h E B h. − 0 ⊕ 0 0 − 0 ⊕ 0 0 Hence ZT = TZ, which means that T is unitarily equivalent to the operator T′ = T (Z) and T is an extension of T′ of the form (3.9). |R e T T T e Identifyinge with ′, it follows that has the property (iii), and we conclude that (i) implies (iii). In addition, we note that δ = B 1/2. We have seen k k that such an operator B can be induced by an operator A satisfying (ii) with B = A , so we can choose δ = A 1/2 ∆ 1/2 (as ∆ 0) in this case. k k k k k k ≥ k Tk T ≥ To complete the proof we show that (iii) implies (i). Indeed, let us assume that T on = is an extension of T, T having the form (3.9) with M M0 ⊕M1 ⊃ H V∗E = 0. Then , as a subspace of , is invariant for T. But T is expansive e H 2 M e because ∆ = 0 δ E∗E 0, and we infer that ∆T = P ∆ 0. Thus T T ⊕ ≥ eH T|He≥ is also expansive.e Furthermore, as we have seen before in Core ollary 3.3, T is a P -isometry and δ2P ∆ . Now P = 0 , because otherwise one has M1 M1 ≥ T M1 H 6 { } e 0, so T = T = V . Hence is invariant for V and T is an isometry, H⊂M |H |H e H in contradiction with our assumption from the beginning of the proof. Therefore e P = 0, which implies M1 |H 6 2 2 P P δ− P ∆ = δ− ∆T. H M1 |H ≥ H T|H e Thus P P 1 = 0 (as ∆T = 0). Finally, from the relation T∗P 1 T = P 1 , we H M |H 6 6 M 2 M deduce that T∗ A1T = A1, where A1 = P P . Hence T is a δ A1-isometry H M1 |H e e with δ2 A ∆ . Therefore, T satisﬁes (by Theorem 3.2) the requirements of (i). 1 ≥ T This proves that (iii) implies (i). The proof is complete. Let us note that in the proof of implication (i) (iii) we have used an ar- ⇒ gument inspired from [3, Theorem 5.80], which concerns 2-isometries. In the case when T is a 2-isometry one can choose in the above proof B = ∆T. This leads to the isometry E1 and one can consider E an isometric extension of E1 with OPERATORS WITH TWO-ISOMETRIC DILATIONS 119

(E) = (V ). Then, in this case, T is a Brownian unitary extension of T with R N ∗ cov(T) = δ = cov(T), and so one recovers the result of [3]. e Concerning the operator T in (3.9), we remark that we do not assume any e relationship between the operators E and U. However, the operator T satisfy the e equivalent conditions of Theorem 3.7 and they can be described by the special e Brownian isometric liftings from Corollary 3.3, as follows.

PROPOSITION 3.8. For T on the following statements are equivalent: H (i) T has a block matrix decomposition (3.9) on = , with an in- H H0 ⊕ H1 H0 variant subspace for T∗T; (ii) T is an expansive P-isometry for an orthogonal projection P with δ2P ∆ ≥ T and some scalar δ > 0; (iii) T has a (minimal) Brownian isometric lifting S of type I, on a space = K such that (∆ ) is an invariant subspace for S, δ 2∆ is H⊥ ⊕ H N S ⊖ H⊥ − S an orthogonal projection and δ = cov(S). Proof. Suppose that T, acting on = , has the form (3.9) with the H H0 ⊕ H1 block matrix given by the operators V, E and U, with V∗E = 0 and δ > 0. Obvi- ously, the condition V E = 0 in (3.9) means that reduces T T. Proceeding as in ∗ H0 ∗ the proof of Theorem 3.2 (the construction of S), since V = T is an isometry in |H0 (3.9), we ﬁnd a lifting S of T on = ℓ2 ( ) with the representations K + DE ⊕b H0 ⊕ H1 e S+ 0 δDE V δE S+ ⋆ S =  0 V δE  = = , e 0 U 0 T 0 0 U e e   where S+ is the forward shift on ⊥ = , V = S+ V on ⊥ 0, E = tr H K ⊖ H ⊕ H ⊕ H D E : , while D = J D with J the canonical injection E H1 → H⊥ ⊕ H0 E E E e E e of into . Since V and E are isometries and V E = 0, it follows that S is a eDE H⊥ e ∗ Brownian isometric lifting of type I for T, with (∆S) = ⊥ 0 and (∆S) e e N H ⊕2 H N ⊖ ⊥ = 0 an invariant subspace for S. Remark also that δ− ∆S = P , so δ = H H H1 cov(S) and, in addition, that S is a minimal lifting of T. Hence (i) implies (iii). Conversely, let us assume that such an operator S on = is a K H⊥ ⊕ H Brownian isometric lifting for T with (∆ ) (so S is of type I) and that H⊥ ⊂ N S (∆S) ⊥ is an invariant subspace for S. Then the operators V0 = S and N ⊖ H |H⊥ V1 = S (∆ ) are isometries. Hence the isometry V := S (∆ ) can be written |N S ⊖H⊥ |N S as the direct sum V = V V on ( (∆ ) ) (as both subspaces are 0 ⊕ 1 H⊥ ⊕ N S ⊖ He ⊥ invariant for V). Then, from the block matrix of S on = (∆ ) (∆ ) e K N S ⊕ R S with the operators V, E, U (as above) and δ = cov(S), it follows that T has on e =( (∆ ) ) (∆ ) =: the representation H N S ⊖ H⊥ e⊕eR S H0 ⊕ H1 V δF T = P S = 1 , F = P E . H |H 0 U H0 |H1 e 120 C.BADEA AND L.SUCIU

Clearly, δ = cov(S) > 0 because T (like S) can be assumed non-isometric. Since

V∗E = 0 in the block matrix of the 2-isometry S, we infer V1∗F = 0. Hence T has a representation of the form (3.9) on = , where the subspace reduces e e H H0 ⊕ H1 H0 T∗T. Therefore (iii) implies (i). Now, since the operators of the form (3.9) are expansive and they satisfy (3.4) with the condition (3.5), the implication (i) (ii) follows from Corollary 3.3. ⇒ Conversely, if T satisﬁes the assertion (ii), then T has the form (3.4) on = H H0 ⊕ with the condition (3.5), by Corollary 3.3. The operator T being expansive H1 by (ii), it follows that C = T is expansive, hence C is an isometry (C being a |H0 contraction by (3.5)). Then the condition (3.5) reduces to C∗E = 0, where E = P T . We conclude that T has the form (3.9). Hence (ii) implies (i). H0 |H1 REMARK 3.9. The Brownian isometric liftings mentioned in Corollary 3.3 and Proposition 3.8 are not Brownian unitaries, because (V ) = (E) (in the N ∗ 6 R previous proof), in general. While both satisfy the conditions (i), (ii) and (iii) of [3, e e Theorem 5.20], they not satisfy the condition (iv) of the same result [3, Theorem 5.20] in which Brownian unitaries are characterized.

3D. CONCAVE OPERATORS AND OPERATORS SIMILAR TO ISOMETRIES. Inthecase of concave operators we can say more.

THEOREM 3.10. For an operator T ( ) the following statements hold: ∈B H (i) T is concave if and only if T has an extension T on = of 1 M M0 ⊕M1 ⊃ H the form

V δE (3.10) T = 1 , 1 0 W

where V, W are isometries, E is a contraction with V E = 0, W E E W 1 ∗ 1 ∗ 1∗ 1 ≤ E E , and δ = ∆ 1/2. In this case T is also concave with ∆ 1/2 = δ. 1∗ 1 k Tk 1 k T1 k (ii) If T is concave, and one of the sequences ∆1/2Tn or ∆1/2Tn converges { T } { T1 1 } strongly to zero, then the other also strongly converges to zero. In this case the sequences 1 Tn and 1 Tn converge strongly to zero. { √n } { √n 1 } (iii) A concave operator T has a minimal 2-isometric lifting of type I and of covariance ∆ 1/2. k Tk Proof. Assume T concave. By applying Theorem 3.7 with A = ∆T, and using the same notation from its proof, one obtains an extension T1 of T on the space = (B) of the form (3.10). More precisely, V, W are isometries on M K0 ⊕ R , respectively on (B), E is a contraction from (B) into with V E = 0, K0 R 1 R K0 ∗ 1 and δ = ∆ 1/2 = B 1/2. As T is concave, T will be concave, too. Indeed, the k Tk k k 1 representation (3.10) of T1 implies

2 2 ∆ = 0 δ E∗E , T∗∆ T = 0 δ W∗E∗E W, T1 ⊕ 1 1 1 T1 1 ⊕ 1 1 OPERATORS WITH TWO-ISOMETRIC DILATIONS 121 and from the proof of Theorem 3.7 we have, for h , ∈ H 1/2 2 1/2 2 E WB h = E B Th = T∗∆ Th, h k 1 k k 1 k h T i ∆ h, h = E B1/2h 2. ≤ h T i k 1 k Therefore W E E W E E , or equivalently T ∆ T ∆ . Thus T is concave. ∗ 1∗ 1 ≤ 1∗ 1 1∗ T1 1 ≤ T1 1 It is clear that ∆ 1/2 δ. It follows from the proof of Theorem 3.2 that one k T1 k ≤ can consider Z an isometry from into such that ZT = T Z. This implies H M 1 ∆ = Z ∆ Z, so δ = ∆ 1/2 ∆ 1/2. Hence ∆ 1/2 = δ. The direct T ∗ T1 k Tk ≤ k T1 k k T1 k implication of the assertion (i) is proved, while the converse part is easy (that is, if T has an extension T1 of the form (3.10), then T = T1 is concave because is |H H invariant for T1). To show the assertion (ii) we remark that if

1/2 n n ∆ T (k0 k1) = δ E1W k1 0, n ∞, k T1 1 ⊕ k k k→ → for all k , k (B), then, for k = B1/2h with h , we have 0 ∈K0 1 ∈ R 1 ∈ H ∆1/2Tnh = E WnB1/2h 0, n ∞. k T k k 1 k→ → Conversely, if this last convergence holds, then, by the Banach-Steinhaus theorem (E and W being contractions), one has E Wnk 0 for any k (B). In other 1 1 1 → 1 ∈ R words, ∆1/2Tn 0 strongly. So, if one of the sequences ∆1/2Tn or ∆1/2Tn T1 1 → { T } { T1 1 } converges strongly to zero, then the other also strongly converges to zero. If this happens, then, by a similar argument to that used in the proof of [20, Theorem 3.10], one can show that 1 Tn 0 and 1 Tn 0 strongly. So (ii) holds. √n → √n 1 → The assertion (iii) follows immediately. Thus, considering T a B-isometry (by Theorem 3.7) with B ∆ and B = ∆ , one can construct the 2-isometry ≥ T k k k Tk SB,T as in (3.8). This operator is a minimal lifting of T of type I and of covariance ∆ 1/2. k Tk REMARK 3.11. If T is a concave operator, then, by Theorem 3.7, T has an extension T of the form (3.9) which, in fact, can be obtained from T1 in (3.10) by extending W to a unitary operator. Notice that T is not concave, in general. So e Theorem 3.10 shows that the appropriate extensions describing concave opera- e tors are those of the form (3.10). This provides a model for the concave operators T with ∆ δ2, for some ﬁxed δ > 0. A related fact is given by the last as- k Tk ≤ sertion of (i) in Theorem 3.10, which says that a concave T is of class C0 (as a · ∆T-contraction) if and only if the corresponding concave model T1 is of the same class C0 (as a ∆T -contraction). We refer to [39] for details about the class C0 . · 1 · Let us remark that T1 in (3.10) is different from the Brownian extension Tb of a concave T obtained in [26, Theorem B], when T √2 (i.e. ∆ I). Indeed, k k ≤ T ≤ in this case ∆Tb is an orthogonal projection, contrary to ∆T1 (in general). 122 C.BADEA AND L.SUCIU

Note also that the extension T1 from (3.10) of a concave operator T is an improved version of the extensions obtained in [11, Proposition 2.2, Theorem 2.1], where different 2-isometric liftings have been directly constructed. Theorem 3.2 applies in particular to operators similar to isometries (these are A-isometries with A invertible). If T is such an operator satisfying T∗ AT = A with A βI for a scalar β > 0, then A βT T β∆ and T is also an A - ≥ ≥ ∗ ≥ T β isometry, where A = β 1 A. In addition, since A ∆ T T ∆ = I, one β − β − T ≥ ∗ − T has (A ∆ ) = . Hence the corresponding lifting S acts on ℓ2 ( ). R β − T H Aβ,T + H Another interesting case is that of quasi-isometries, i.e. the T∗T-isometries, where A = T∗T is not necessary invertible in this case.

COROLLARY 3.12. If T ( ) is similar to an isometry, or T is a quasi- ∈ B H isometry, then T has a minimal 2-isometric lifting of type I on the space ℓ2 ( ). Such + H a lifting is the operator S (respectively S ), having covariance A 1/2 (respec- Aβ,T T∗T,T k βk tively T ). k k Notice that a quasi-isometry T is similar to an isometry if and only if T is injective with (T) closed (see [25]). Also, a quasi-isometry T is expansive if and R only if V = T ( ) is an isometry, X = P ( )T (T ) is expansive and V∗X = 0. | T T |N ∗ Hence TheoremR 3.7 cannot be applied to quasi-isometries,R in general. A result obtained in [10, Theorem 1.1] shows that an expansive operator is 1 n j j similar to an isometry if and only if the Cesàro means Mn := n+1 ∑j=0 T∗ T are bounded. In this case the sequence M strongly converges to an operator M { n} such that T MT = M and M T T. Thus, the expansive operators T which ∗ ≥ ∗ are similar to isometries have 2-isometric liftings of type I. These liftings can be chosen of covariance equal to M 1/2 by Corollary 3.12. k k

4. TWO-ISOMETRIC LIFTINGS OF TYPE II

4A. A-CONTRACTIONS. As in Deﬁnition 3.1, a 2-isometric lifting S for T ( ) ∈B H is of type II if is invariant for S S. The following result should be compared to H ∗ Theorem 3.2 for the type I case.

THEOREM 4.1. For T ( ) the following statements are equivalent: ∈B H (i) T has a 2-isometric lifting of type II on a Hilbert space containing ; H (ii) T is an A-contraction for a positive operator A = 0 on with ∆ A; 6 H T ≤ (iii) T has a lifting T on which is a B-isometry for some B ( ) with ∗ H∗ ⊃ H ∈B H∗ 0 = B 0, such that ∆T B and B ; 6 ≥ ∗ ≤ H ⊂ H (iv) T has a 2-isometric lifting S of the form (3.1) with operators W and X satisfying the condition

(4.1) X∗∆ X + 2Re(X∗W∗XT) 0. W ≥ OPERATORS WITH TWO-ISOMETRIC DILATIONS 123

Moreover, if these statements are true, then the lifting S in (i) and (iv) can be chosen either minimal, or with ∆S = 2P, where P is an orthogonal projection, with | ⊥ shifts on the main diagonal of S H in its canonical representation and with cov(S) = | ⊥ √2 max 1, A 1/2 for A fromH (ii). · { k k } Proof. If S on is a 2-isometric lifting of type II of T with the block ma- K ⊃ H trix (3.1), then W∗X = 0 and so the condition (4.1) is satisfied. Hence (i) implies (iv). Next, if a 2-isometric lifting S of the form (3.1) satisfies (4.1), then, from the relation (3.7), we infer that T AT A, where A = X X + ∆ ∆ . Clearly, one ∗ ≤ ∗ T ≥ T has A 0 since ∆ 0 in (3.2). Notice that if A = 0, then I T T = X X 0. ≥ S ≥ − ∗ ∗ ≥ Therefore T is a contraction, i.e. an I-contraction with I ∆ . Hence one can ≥ T always consider A = 0, and thus we proved that (iv) implies (ii). 6 Now, let T be non-isometric and let A be as in (ii), i.e. satisfying T AT A ∗ ≤ with A 0 and A ∆ = 0. Then there exists a contraction T on (A) such that ≥ ≥ T 6 R TA1/2 = A1/2T. Let S be the forward shift on ℓ2 ( ) and set D = JD , where + T b T ℓ2 D J is the canonical injection of into +( ). Define the lifting T of T and the b b DT DT b ∗ b ℓ2 extension A of A on the space := +( ) by the block matrices Hb∗ DTb ⊕ H b S DA1/2 b I 0 (4.2) T = , A = . ∗ 0 T 0 A b b 1/2 We have S∗DA = 0. We use this relation to obtain that 1/2 2 1/2 1/2 1/2 T∗ AT b = I (A D A + T∗ AT) = I (A A T∗TA + T∗ AT) ∗ ∗ ⊕ T ⊕ − b = I A = A. b b b ⊕ Therefore, T is an A-isometry.b Moreover, taking into account that ∆T A, one ∗ ≤ has b ∆T = 0 (A T∗ AT + ∆T) 0 (A + ∆T) 2A. ∗ ⊕ − ≤ ⊕ ≤ Setting B = 2A, we obtain that B and T satisfy the conditions fromb (iii). There- ∗ fore (ii) implies (iii). b n We also notice that the lifting T of T is minimal, that is = n 0 T . ∗ H∗ ≥ ∗ H This fact follows easily (as for SA,T in the proof of Theorem 3.2), keepingW in mind that = D A1/2 , T being defined on (A). DT T H R In order to prove that (iii) implies (i) we assume that T on and B b b b ∗ H∗ ⊃ H are as in (iii). Then, using Theorem 3.2, we find a 2-isometric lifting S = SB,T as ∗ in (3.8) for T on a Hilbert space such that ∆S = 0 B on = ⊥ . ∗ K ⊃ H∗ ⊕ K H ⊕ H∗ Clearly, S will be a lifting for T having on = a representation∗ of the K H′ ⊕ H form (3.1) with W = S and X ( , ′). So one obtains ∆S on = ′ |H′ ∈ B H H K H ⊕ H with representation (3.2). Now, since B and B 0, it follows that H ⊂ H ≥ H reduces B, so B = B0 B1 on = ⊥ . Then, as S is an isometry, we ⊕ H∗ H ⊕ H | ⊥ get the representations H∗ ∆ = 0 B =(0 B ) B S ⊕ ⊕ 0 ⊕ 1 124 C.BADEA AND L.SUCIU on the decompositions = ⊥ = ′ , respectively. From the block K H∗ ⊕ H∗ H ⊕ H matrix (3.2) of ∆S we infer that W∗X = 0. This means that S is a 2-isometric lifting of type II for T. Thus we have shown that (iii) implies (i) and all equivalences (i)- (iv) are provided. Let us remark that the above lifting S = SB,T can be chosen in such a man- ∗ ner that it has the form (3.1) with ∆W = 2P, where P is an orthogonal projection. Indeed, if we take B = 2A with A = I A and A from (ii), as in the implication ⊕ (ii) (iii), then S can be expressed in terms of the operator A. Keeping in mind ⇒ b b the form of ∆T , we get ∗ BT := B ∆T = 2A ∆T = 2I (A + T∗ AT ∆T) = 2I B0 ∗ − ∗ − ∗ ⊕ − ⊕ ℓ2 on = +( ) , whereb B0 = A + T∗ AT ∆T 0. So, by (3.8), the H∗ DT ⊕ H − ≥ ℓ2 operator S acts onb = ⊥ , where ⊥ = +( (BT )) and (BT ) = K H∗ ⊕ H∗ H∗ R ∗ R ∗ ℓ2 ( ) (B ). Therefore, S has on = ℓ2 ( ) = ( ) + DT ⊕ R 0 K H⊥ ⊕ + DT ⊕ H K ⊖ H ⊕ H the representations ∗ b b S √2I B0 ∗ W X (4.3) S =  0 S DA1/2 = , e e 0 T 0 0 T  b  1/2 where S is a shift operator on ⊥, while I and J, with B0 = JB0 , are the ∗ H∗ canonical injections of , respectively of (B0), into (S∗). We infer from H∗ ⊖ H e R e Ne ∗ e these matrices that W = S has a special form on ′ = with two shifts |H′ H K ⊖ H on the main diagonal and I an isometry. Therefore ∆W = 0 2I = 2P (∆ ). Also, ⊕ R W since ∆ = 0 B = 0 2(I A), we get cov(S) = √2 max 1, A 1/2 . Thus S ⊕ ⊕ e ⊕ · { k k } we have proved a part of the last assertion of the theorem. It remains to show that the lifting S in (i) can be also chosen minimal, i.e. n n with = n 0 S . Indeed, let us denote by 0 = n 0 S the smallest K ≥ H K ≥ H invariant subspaceW in for S which contains , and let SW0 = S . Then S0 is a K H |K0 2-isometry. Since S∗ is an extension of T∗ and a lifting of S0∗, we have

S0∗ = P S∗ = T∗. |H K0 |H Hence S0 is a lifting of T. So, S0 has the form W X S = 0 0 0 0 T on 0 = 0 , where W0 = S0 = S = W and X0 = P X . Here W K H ⊕ H |H0 |H0 |H0 H0 |H and X are coming from the block matrix (3.1) of S on = and they have K H′ ⊕ H representations of the form W ⋆ X W = 0 , X = 0 0 ⋆ ⋆ on = , respectively from into for the matrix of X. But H′ H0 ⊕ H0⊥ H H0 ⊕ H0⊥ S is a lifting of type II for T, so S∗S . Then S0∗S0 = P 0 S∗S , so H ⊂ H H K H ⊂ H OPERATORS WITH TWO-ISOMETRIC DILATIONS 125

S0 is also a lifting of type II for T. In addition, S0 is a minimal lifting. Since any lifting of T satisfying (i) also satisﬁes (iv), it follows that in (iv) one can also chose a minimal lifting for T. This completes the proof of theorem.

REMARK 4.2. The condition B in the assertion (iii) of Theorem 4.1 is H ⊂ H essential (as we have seen in the proof of (iii) (i)). If T has a lifting which has a ⇒ 2-isometric lifting of type I, then it is not necessarily true that T has a 2-isometric lifting of type II. Obviously, Theorem 4.1 generalizes Theorem 3.2. The assertion (ii) of the latter can be reformulated in terms of A-contractions and Q-expansive operators T, i.e. operators satisfying 0 Q T QT. More precisely, from Theorem 3.2 and ≤ ≤ ∗ [13, Theorem 2.1], we deduce the following result which shows when an operator with 2-isometric liftings of type II has a lifting of type I. This always happens for expansive operators (by Theorems 3.7 and 4.1). In addition, the corollary below reproves the assertion (ii) of Theorem 3.7 in the case when Q is a scalar multiple of the identity (for T expansive).

COROLLARY 4.3. An operator T = 0 has a 2-isometric lifting of type I if and only 6 if there exist two positive operators A and Q satisfying ∆ Aand T AT A Q T ≤ ∗ ≤ ≤ ≤ T∗QT.

Notice that if T is an operator similar to a contraction, then T is an A0- contraction for an invertible operator A , which can be chosen such that A 0 0 ≥ T∗T. As a direct consequence of Theorem 4.1, we obtain the following result.

COROLLARY 4.4. If T ( ) is non-contractive and similar to a contraction, ∈ B H then T has a 2-isometric lifting of type II and of covariance √2 A 1/2, with A · k 0k k 0k ≥ T . k k Among the operators similar to contractions we can consider those having ρ-dilations. For ρ > 0, an operator T ( ) is said to have a ρ-dilation if there ∈ B H n n exists a unitary operator Uρ on some space ρ such that T = ρP Uρ for H ⊃ H H |H n 1 (see [39]). Such a ρ-dilation U of T is not a lifting for T. From Corollary 4.4 ≥ ρ we deduce the following

COROLLARY 4.5. Any operator T having a ρ-dilation has a minimal 2-isometric lifting of type II.

Recall (see [38]) that an operator with a ρ-dilation is a ST-isometry for some positive operator S . Soif S ∆ it follows that T has even a 2-isometric lifting T T ≥ T of type I. Another special class of operators similar to contractions is given by quasi- contractions, that is the T∗T-contractions (see [11, 14]). For such an operator T one obtains from Theorem 4.1 a 2-isometric lifting S of type II with

cov(S) = √2 max 1, T = √2 T , · { k k} k k 126 C.BADEA AND L.SUCIU whenever T is non-contractive. Now we show that there exist operators similar to contractions, even quasi- contractions, which do not have 2-isometric liftings of type I.

EXAMPLE 4.6. Let C be a contraction on such that Cn 0 strongly and H → let δ > 1 be a scalar. Then the operator T on with the matrix representation H ⊕ H C δJ T = , 0 0 where J(0 h) = h 0, for h , is a non-contractive quasi-contraction. Thus ⊕ ⊕ ∈ H T is similar to a contraction. But Tn 0 strongly, because C has this property. → Hence T cannot be an A-isometry with A = 0. Thus, by Theorem 3.2, T does not 6 possess 2-isometric liftings of type I. A simpler example of this form is a nilpotent operator of order 2 (and thus having ρ-dilations for suitable ρ) on C C. ⊕ On the other hand, we show that an operator T similar to a contraction can have 2-isometric liftings of type I (not only of type II), without being similar to an isometry. Therefore, in this case, Corollary 4.3 applies to a non-invertible operator A, so not equal to the operator A0 from Corollary 4.4 (otherwise T will be similar to an isometry by [13, Theorem 2.1]).

EXAMPLE 4.7. Let U be a unitary operator on = ( = 0 ) H H0 ⊕ H1 Hj 6 { } with a block matrix of the form V ⋆ U = , 0 V′∗ where V and V are isometries. Let C = V 0 on and set E = J(0 V ), ′ ⊕ H − ⊕ ′∗ acting from 0 to 0 , where J is as in Example 4.6. Finally, we deﬁne { } ⊕ H H ⊕ { } the operator T on by H ⊕ H C E T = . 0 U

It is clear that C∗E = 0. Therefore T has the form (3.4). Furthermore, if Z = JP 1 (an operator from 0 into 0 ), then it is easy to see that H { } ⊕ H H ⊕ { } CZ ZU = J(0 V′∗) = E. − − ⊕ A known result (see [9]) and the last relation ensure that T is similar to a contraction, more precisely to the diagonal operator C U. As C is not an isometry, ⊕ the operator T is not similar to an isometry. However, T has a minimal Brownian isometric lifting of type I by Corollary 3.3.

4B. ISOMORPHICMINIMAL 2-ISOMETRIC LIFTINGS. Previous results show the existence of minimal 2-isometric liftings, but their uniqueness up to an isomor- phism (i.e. a unitary equivalence which ﬁxes ) is not guaranteed, in this context. H This is in contrast to the classical case of isometric (unitary) dilation theory of contractions (see [17], [39]). However, the following fact about minimal 2-isometric liftings of type II is true. OPERATORS WITH TWO-ISOMETRIC DILATIONS 127

PROPOSITION 4.8. Let T ( ) and let S on and S on be ∈ B H K ⊃ H ′ K′ ⊃ H two minimal 2-isometric liftings of T with S S and S S . ThenS andS ∗ H ⊂ H ′∗ ′H ⊂ H ′ are isomorphic if and only if S∗S = S′∗S′ . If this is the case, then the 2-isometries |H |H S and S′ are unitarily equivalent. |K⊖H |K′⊖H 2 2 Proof. Let T, S and S′ be as above. Since S is a 2-isometry, we have S∗ S = n n 2S∗S I and, for n > 2, one obtains the formula S∗ S = nS∗S (n 1)I.A − n−n − similar relation holds for S′. Since S∗S , we infer that S∗ S for Hn ⊂ H n n H ⊂ H n 1. On the other hand, we have P S = T = P S′ for n 1. Using ≥ H |H H |H ≥ these relations, we obtain, for any ﬁnite system h n , that { j}0 ⊂ H n n j 2 j l 2 ∑ S hj = ∑ S hj, S hl = h0 + k j=0 kK j,l=0h i k k n n l (l j) j j j (j l) l l ∑ h0, T hl + ∑ S∗ − S∗ S hj, hl + ∑ T hj, h0 + ∑ hj, S∗ − S∗ S hl l=1h i l j 1h i j=1h i j>l 1h i ≥ ≥ ≥ n 2 j j j l j j l l l = h0 + 2Re ∑ T hj, h0 + ∑ S∗ S hj, T − hl + ∑ T − hj, S∗ S hl k k j=1h i l j 1h i j>l 1h i ≥ ≥ ≥ n 2 j l j j l = h0 + 2Re ∑ T hj, h0 ∑ (j 1) hj, T − hl ∑ (l 1) T − hj, hl k k j=1h i − l j 2 − h i − j>l 2 − h i ≥ ≥ ≥ l j j l + ∑ j S∗Shj, T − hl + ∑ l T − hj, S∗Shl . l j 1 h i j>l 1 h i ≥ ≥ ≥ Assume now that S∗S = S′∗S′ . Then the last expression in the above |H |H computation can be also written in terms of T and S′. So, proceeding in a reverse way, one obtains n n Sjh = S jh . ∑ j ∑ ′ j ′ k j=0 kK k j=0 kK Then, by a standard argument, it follows that there exists a unitary operator Z from onto ′ satisfying the relations ZS = S′Z and Z = I. Therefore S and K K |H S′ are isomorphic as 2-isometric liftings of T. Conversely, if there is such an operator Z which preserves the elements of and intertwines S with S′, then S∗S = Z∗S′∗S′Z. We get S∗S = Z∗S′∗S′ = H |H |H S′∗S′ , because S′∗S′ and Z = I = Z∗ . Thus the ﬁrst assertion of |H H ⊂ H |H |H Proposition 4.8 is proved. Furthermore, if Z is as above, then Z( ) = K ⊖ H K′ ⊖ H and Z′ := Z is unitary. So one obtains |K⊖H Z′(S )=(ZS) =(S′Z) =(S′ )Z′. |K⊖H |K⊖H |K⊖H |K′⊖H Therefore the 2-isometries S and S′ are unitarily equivalent by Z′. |K⊖H |K′⊖H From the last assertion of the previous proposition we derive the following

COROLLARY 4.9. Suppose that two minimal 2-isometric liftings of type II of an operator are isomorphic. Then one of them is of type I if and only if the other is of type I. 128 C.BADEA AND L.SUCIU

REMARK 4.10. Assume that S on is a 2-isometric lifting of type I for K ⊃ H T ( ). If S does not satisfy the minimality condition, then the minimal lifting ∈B H n S0 = S on 0 = n 0 S is also of type I for T. Indeed, since 0 and |K0 K H K K ⊖ H are invariant for S,W it follows≥ that S ( ) = S( ) , and so 0 K0 ⊖ H K0 ⊖ H ⊂ K0 ⊖ H S0 = S is an isometry. |K0⊖H |K0⊖H Moreover, is a reducing subspace for S. Indeed, since S S , we K0 ∗ H ⊂ H have n n S∗ ( S∗S ) ( ∆ S ) . K0 ⊂H∨ H ⊂K0 ∨ S H ⊂K0 n_2 n_1 ≥ ≥ The last inclusion holds because (∆S), S being of type I and so by (3.1) K⊖H⊂N n n we have ∆S = 0 ∆S on = ( ) , hence ∆SS h = ∆ST h for ⊕ |H K K ⊖ H ⊕ H ∈ H h , n 1. Thus the minimality condition for 2-isometric liftings of type I can ∈ H ≥ be deﬁned with respect to reducing subspaces, or equivalently, to subspaces assumed to be only invariant. This fact is analogous to the notion of minimal Brow- nian unitary (respectively isometry, in the cyclic case) extension for a 2-isometry, which appears in [4, Section 10].

Acknowledgements. The first named author was supported in part by the project FRONT of the French National Research Agency (grant ANR-17-CE40-0021) and by the Labex CEMPI (ANR-11-LABX-0007-01). The second named author was supported by a project financed by Lucian Blaga University of Sibiu and Hasso Plattner Foundation research Grants LBUS-IRG-2019-05. This paper has been finalized when both authors were in residence at CIRM, Luminy, France during a “Recherche en Binôme no. 2198” pro- gramme.

References

[1] J. AGLER, An abstract approach to model theory, in Surveys of Some Recent Results in Operator Theory, vol. II, Pitman Res. Notes Math. Ser., vol. 192, John Wiley and Sons, New York 1988, pp. 1–23. [2] J. AGLER, M. STANKUS, m-isometric transformations of Hilbert spaces, Integral Equa- tions Operator Theory 21, 4 (1995), 383–429. [3] J. AGLER,M. STANKUS, m-isometric transformations of Hilbert spaces II, Integral Equa- tions Operator Theory 23, 10 (1995), 1–48. [4] J. AGLER,M.STANKUS, m-isometric transformations of Hilbert spaces III, Integral Equa- tions Operator Theory 24 (1996), 379–421. [5] A. ALEMAN, The multiplication operator on Hilbert spaces of analytic functions, Habilita- tionsschrift, Fern Universität, Hagen, 1993. [6] A. ALEMAN,W. ROSS, The backward Shift on weighted Bergman spaces, Michigan Math. J. 43, (1996), 291–319. [7] A. ANAND,S. CHAVAN, Z.J. JABLONSKI,J. STOCHEL, A solution to the Cauchy dual subnormality problem for 2-isometries, J. Func. Anal. 277(2019), Issue 12, 108292. OPERATORS WITH TWO-ISOMETRIC DILATIONS 129

[8] C.BADEA, Operators near completely polynomially dominated ones and similarity problems, J. Operator Th. 49 (2003), 3–23. [9] C. BADEA, Perturbations of operators similar to contractions and the commutator equation, Studia Math. 150 (2002), 273–293. [10] C. BADEA, L. SUCIU, Similarity problems, Følner sets and isometric representations of amenable semigroups, Mediterr. J. Math. 16(2019), article no. 5. [11] C. BADEA, L. SUCIU, The Cauchy dual and 2-isometric liftings of concave operators J. Math. Anal. Appl. 472 (2019), 1458–1474. [12] H. BERCOVICI,B.PRUNARU, Quasiafﬁne transforms of polynomially bounded operators, Arch. Math. (Basel) 71 (1998) 384–387. [13] A. BISWAS,A.E.FRAZHO,C.FOIAS, Weighted commutant lifting, Acta Sci. Math. (Szeged), 65, 3–4 (1999), 657–686. [14] G. CASSIER,L. SUCIU, Mapping theorems and similarity to contractions for classes of A- contractions, in: Hot Topics in Operator Theory, Theta Ser. Adv. Math. (2008), 39–58. [15] R.G. DOUGLAS, On majorization, factorization and range inclusion of operators in Hilbert space, Proc. Amer. Math. Soc. 17 (1966) 413–416.

[16] R.G. DOUGLAS, On the operator equation S∗XT = X and related topics. Acta. Sci. Math. (Szeged) 30 (1969), 19–32. [17] C FOIAS,A.E.FRAZHO, The Commutant Lifting Approach to Interpolation Problems, Birkhäuser Verlag, Basel-Boston-Berlin, 1990. [18] C. FOIAS,A.E.FRAZHO,M.A. KAASHOEK, Contractive liftings and the commutator, C. R. Acad. Sci. Paris, Ser. I 335 (2002) 431–436. [19] O. GISELSSON,A. OLOFSSON, On some Bergman shift operators, Complex Anal. Oper. Theory 6 (2012) 829–842. [20] S. JUNG, Y. KIM,E.KO,J.E.LEE, On (A, m)-expansive operators, Studia Math. 213, 1, (2012), 3–23. [21] L. KERCHY, Generalized Toeplitz operators, Acta Sci. Math. (Szeged) 68 (2002), 373–400. [22] C.S. KUBRUSLY, An Introduction to Models and Decompositions in Operator Theory, Birkhäuser, Boston, 1997. [23] W. MAJDAK,M.MBEKHTA,L.SUCIU, Operators intertwining with isometries and Brow- nian parts of 2-isometries, Linear Algebra Appl. 509 (2016), 168–190. [24] W. MAJDAK,L.SUCIU, Brownian isometric parts of concave operators, New York J. Math. 25 (2019) 1067-1090. [25] M. MBEKHTA,L. SUCIU, Classes of operators similar to partial isometries, Integr. Equat. Oper. Th. vol. 63, Number 4, (2009), 571–590. [26] S. MCCULLOUGH, SubBrownian operators, J. Operator Th. 22 (1989), 291–305. [27] S. MCCULLOUGH,B. RUSSO, The 3-Isometric Lifting Theorem, Integr. Equat. Oper. Th. 84, 1 (2016), 69–87. [28] V. MÜLLER, Models for operators using weighted shifts, J. Operator Th. 20, 1 (1988), 3–20. [29] A. OLOFSSON, A von Neumann-Wold decomposition of two-isometries, Acta Sci. Math. (Szeged) 70 (2004), 715–726. 130 C.BADEA AND L.SUCIU

[30] S. RICHTER, Invariant subspaces of the Dirichlet shift, J. Reine Angew. Math. 386 (1988), 205–220. [31] S. RICHTER, A representation theorem for cyclic analytic two-isometries, Trans. Amer. Math. Soc. 328 (1991), 325–349. [32] E. RYDHE, Cyclic m-isometries, and Dirichlet type spaces, J. London Math. Soc. (2) 99 (2019) 733–756. [33] S. SHIMORIN, Wold-type decompositions and wandering subspaces for operators close to isometries, J. Reine Angew. Math. 531 (2001), 147–189. [34] O. A. M. SID AHMED,A. SADDI, A m-Isometric operators in semi-Hilbertian spaces, − Linear Algebra Appl. 436 (10) (2012), 3930–3942. [35] M. STANKUS, m-Isometries, n-Symmetries and other linear transformations which are hered- itary roots, Integr. Equ. Oper. Th. 75 (2013), 301–321. [36] L. SUCIU, Maximum A-isometric part of an A-contraction and applications, Israel J. Math. 174 (2009), 419–442. [37] L. SUCIU, On Operators with Two-Isometric Liftings, Complex Analysis and Operator Theory (2020) 14:5, 1–16. [38] L. SUCIU, N. SUCIU, Asymptotic behaviours and generalized Toeplitz operators, J. Math. Anal. Appl., vol. 349, issue 1, (2009), 280–290. [39] B. SZ.-NAGY,C.FOIAS,H.BERCOVICI,L.KÉRCHY, Harmonic Analysis of Operators on Hilbert Space, Revised and enlarged edition, Universitext, Springer, New York, 2010. [40] S. TREIL, A. VOLBERG, A ﬁxed point approach to Nehari’s problem and its applications, Operator Theory: Advances and Applications 71, Birkhäuser, 1994, 165–186.

C. BADEA, UNIV LILLE, CNRS, UMR 8524 - LABORATOIRE PAUL PAINLEVÉ, FRANCE E-mail address: [email protected]

L. SUCIU, DEPT. OF MATHEMATICS AND INFORMATICS,“LUCIAN BLAGA” UNI- VERSITY OF SIBIU,ROMANIA E-mail address: [email protected]

Received Month dd, yyyy; revised Month dd, yyyy.