Operator Theory: Advances and Applications Vol. 175
Editor: I. Gohberg
Editorial Office: S. T. Kuroda (Tokyo) School of Mathematical P. Lancaster (Calgary) Sciences L. E. Lerer (Haifa) Tel Aviv University B. Mityagin (Columbus) Ramat Aviv, Israel V. Olshevsky (Storrs) M. Putinar (Santa Barbara) Editorial Board: L. Rodman (Williamsburg) D. Alpay (Beer-Sheva) J. Rovnyak (Charlottesville) J. Arazy (Haifa) D. E. Sarason (Berkeley) A. Atzmon (Tel Aviv) I. M. Spitkovsky (Williamsburg) J. A. Ball (Blacksburg) S. Treil (Providence) A. Ben-Artzi (Tel Aviv) H. Upmeier (Marburg) H. Bercovici (Bloomington) S. M. Verduyn Lunel (Leiden) A. Böttcher (Chemnitz) D. Voiculescu (Berkeley) K. Clancey (Athens, USA) D. Xia (Nashville) L. A. Coburn (Buffalo) D. Yafaev (Rennes) R. E. Curto (Iowa City) K. R. Davidson (Waterloo, Ontario) Honorary and Advisory R. G. Douglas (College Station) Editorial Board: A. Dijksma (Groningen) C. Foias (Bloomington) H. Dym (Rehovot) P. R. Halmos (Santa Clara) P. A. Fuhrmann (Beer Sheva) T. Kailath (Stanford) B. Gramsch (Mainz) H. Langer (Vienna) J. A. Helton (La Jolla) P. D. Lax (New York) M. A. Kaashoek (Amsterdam) M. S. Livsic (Beer Sheva) H. G. Kaper (Argonne) H. Widom (Santa Cruz)
Subseries: Advances in Michael Demuth Partial Differential Equations Technische Universität Clausthal Germany Subseries editors: Bert-Wolfgang Schulze Jerome A. Goldstein Universität Potsdam The University of Memphis, TN Germany USA Sergio Albeverio Nobuyuki Tose Universität Bonn Keio University, Yokohama Germany Japan Operator Theory in Inner Product Spaces
Karl-Heinz Förster Peter Jonas Heinz Langer Carsten Trunk Editors
Advances in Partial Differential Equations
Birkhäuser Basel . Boston . Berlin Editors:
Karl-Heinz Förster Heinz Langer Peter Jonas Mathematik Carsten Trunk Technische Universität Wien Institut für Mathematik, MA 6-4 Wiedner Hauptstrasse 8–10/1411 Technische Universität Berlin A-1040 Wien Straße des 17. Juni 136 e-mail: [email protected] D-10623 Berlin e-mail: [email protected] [email protected] [email protected]
0DWKHPDWLFV6XEMHFW&ODVVL¿FDWLRQ'$$%%& $[[%%(
/LEUDU\RI&RQJUHVV&RQWURO1XPEHU
Bibliographic information published by Die Deutsche Bibliothek 'LH'HXWVFKH%LEOLRWKHNOLVWVWKLVSXEOLFDWLRQLQWKH'HXWVFKH1DWLRQDOELEOLRJUD¿HGHWDLOHG bibliographic data is available in the Internet at
ISBN 3978-3-7643-8268elston 978-3-7643-8269-8 – BirkhäuserBerlin Verlag, Basel – Boston – Berlin
This work is subject to copyright. All rights are reserved, whether the whole or part of the PDWHULDOLVFRQFHUQHGVSHFL¿FDOO\WKHULJKWVRIWUDQVODWLRQUHSULQWLQJUHXVHRI LOOXVWUDWLRQVUHFLWDWLRQEURDGFDVWLQJUHSURGXFWLRQRQPLFUR¿OPVRULQRWKHUZD\VDQG storage in data banks. For any kind of use permission of the copyright owner must be obtained.
© 2007 Birkhäuser Verlag AG, P.O. Box 133, CH-4010 Basel, Switzerland Part of Springer Science+Business Media Printed on acid-free paper produced from chlorine-free pulp. TCF f Cover design: Heinz Hiltbrunner, Basel Printed in Germany ISBN-10: 3-7643-8269-3 e-ISBN-10: 3-7643-8270-8 ISBN-13: 978-3-7643-8269-8 e-ISBN-13: 978-3-7643-8270-4
ZZZELUNKDXVHUFK Contents
Preface ...... vii
T.Ya. Azizov and L.I. Soukhotcheva LinearOperatorsinAlmostKreinSpaces ...... 1 J. Behrndt, A. Luger and C. Trunk Generalized Resolvents of a Class of Symmetric Operators inKreinSpaces ...... 13 J. Behrndt, H. Neidhardt and J. Rehberg Block Operator Matrices, Optical Potentials, Trace Class PerturbationsandScattering ...... 33 V. Derkach, S. Hassi and H. de Snoo Asymptotic Expansions of Generalized Nevanlinna Functions andtheirSpectralProperties ...... 51 A. Fleige A Necessary Aspect of the Generalized Beals Condition for the Riesz Basis Property of Indefinite Sturm-Liouville Problems ...... 89 K.-H. F¨orster and B. Nagy On Reducible Nonmonic Matrix Polynomials with General andNonnegativeCoefficients ...... 95 S. Hassi, H. de Snoo and H. Winkler On Exceptional Extensions Close to the Generalized Friedrichs ExtensionofSymmetricOperators ...... 111 P. Jonas and H. Langer On the Spectrum of the Self-adjoint Extensions of a Nonnegative LinearRelationofDefectOneina KreinSpace ...... 121 M. Kaltenb¨ack and H. Woracek Canonical Differential Equations of Hilbert-Schmidt Type ...... 159 I. Karabash and A. Kostenko Spectral Analysis of Differential Operators with Indefinite Weightsanda LocalPointInteraction ...... 169 vi Contents
C. Mehl and C. Trunk Normal Matrices in Degenerate Indefinite Inner Product Spaces ...... 193 V. Pivovarchik SymmetricHermite-BiehlerPolynomialswithDefect ...... 211 L. Rodman A Note on Indefinite Douglas’ Lemma ...... 225 A. Sandovici Some Basic Properties of Polynomials in a Linear Relation inLinearSpaces ...... 231 Preface
This volume contains papers written by the participants of the 4th Workshop on Operator Theory in Krein Spaces and Applications, which was held at the Technische Universit¨at Berlin, Germany, December 17 to 19, 2004. The workshop covered topics from spectral, perturbation and extension the- ory of linear operators and relations in inner product spaces. They included spec- tral analysis of differential operators, the theory of generalized Nevanlinna func- tions and related classes of functions, spectral theory of matrix polynomials and problems from scattering theory. All these topics are reflected in the present vol- ume. The workshop was attended by 58 participants from 12 countries. It is a pleasure to acknowledge the substantial financial support received from the – Deutsche Forschungsgemeinschaft (DFG), – DFG-Forschungszentrum MATHEON “Mathematik f¨ur Schl¨ussel- technologien”, – Institute of Mathematics of the Technische Universit¨at Berlin. We would also like to thank Petra Grimberger for her great help. Last but not least, special thanks are due to Jussi Behrndt and Christian Mehl for their excellent work in the organisation of the workshop and the preparation of this volume. Without their assistance the workshop might not have taken place.
The Editors Operator Theory: Advances and Applications, Vol. 175, 1–11 c 2007 Birkh¨auser Verlag Basel/Switzerland
Linear Operators in Almost Krein Spaces
Tomas Ya. Azizov and Lioudmila I. Soukhotcheva
Abstract. The aim of this paper is to study the completeness and basicity problems for selfadjoint operators of the class K(H)inalmostKreinspaces and prove criteria for the basicity and completeness of root vectors of linear pencils. Mathematics Subject Classification (2000). Primary 47B50; Secondary 46C50. Keywords. Krein space, operator pencil, completeness and basicity problem.
1. Introduction Let H be a Hilbert space, let A and B be compact operators and let A be addi- tionally a positive operator. Consider the linear operator pencil L(λ)=A−1 − λ(I + B). (1)
Such a pencil appears, for instance, if the following spectral problem in L2(0,π) is considered: ⎧ ⎨ d2f π − + q(t)f(t)=λ(f(t)+ K(t, s)f(s)ds) dt2 ⎩ 0 (2) f(0) = f(π)=0. Assume q is a continuous real function with q(t) > −1, the kernel K(t, s)issym- metric and continuous on [0,π]×[0,π]. It remains to define A and B in the following way: d2f A−1f = − + qf, dt2 −1 dom A = f ∈ L2(0,π) | f,f absolutely continuous on (0,π), d2f − + qf ∈ L2(0,π),f(0) = f(π)=0 dt2
This research is supported partially by the grant RFBR 05-01-00203. 2 T.Ya. Azizov and L.I. Soukhotcheva
π (Bf)(t)= K(t, s)f(s)ds. 0 The completeness and basicity problems for the pencil (1) and for the operator
H = A(I + B)(3) are closely related. The completeness problem for operators (3) under the assump- tion λ = 0 is not an eigenvalue of H (0 ∈/ σp(H)) was considered for the first time by M.V. Keldysh (see, for instance, [6, Theorem 8.1]). The case when B is also selfadjoint and ker H = {0} was studied by I.Ts. Gokhberg and M.G. Krein in [6, p. 322]. There it was shown that there is a Riesz basis in H which consists of root vectors of H. The proof is based on the selfadjointness of the operator H with respect to the indefinite inner product
[·, ·]=((I + B)·, ·), (4) which turns {H, [·, ·]} into a Pontryagin space. General criteria for the completeness and basicity of root vectors of selfadjoint operators in Pontryagin spaces can be found in [2], and, more generally, in [3, Theorem IV.2.12].
If ker(I + B) = 0, the inner product (4) is degenerate and, due to the com- pactness of the operator B,wehave (a) the isotropic part
H0 := {x ∈H|[x, y]=0, for all y ∈H}
of H is finite-dimensional, and (b) the factor-space H = H/H0 is a Pontryagin space. Indefinite inner product spaces H with the properties (a) and (b) are called almost Pontryagin spaces. The operator H in (3) is selfadjoint in the degenerate almost Pontryagin space {H, [·, ·]}. First results about the completeness and ba- sicity problems for compact selfadjoint operators in almost Pontryagin space were obtained in [1]. Recently, spectral properties of operators acting in almost Pontryagin and almost Krein spaces (for a definition of almost Krein spaces we refer to Definition 1 below) and their applications were studied in, for example, [4], [5], [7]–[13].1 The main aim of this paper is to study the completeness and basicity problems for selfadjoint operators of the class K(H) in almost Krein spaces (a definition of the class K(H) see below on p. 3). Moreover, we will prove criteria for the basicity and completeness of root vectors of the pencil (1).
1The authors thank Chr. Mehl (TU Berlin) for his help with the bibliography. Linear Operators in Almost Krein Spaces 3
2. Main definitions We shortly recall some definitions and notions related to Krein spaces. For more details we refer to [3]. A linear space K equipped with an indefinite inner product [·, ·] is called a Krein space if it admits a canonical decomposition
K = K+[+]K−, where K+ is [·, ·]-orthogonal to K−,andK± is a Hilbert space with respect to the scalar product ± [·, ·], respectively.
Definition 1. AspaceK with an indefinite inner product [·, ·] is called an almost Krein space if its isotropic part K0 is finite-dimensional (we do not exclude the case K0 = {0}) and the factor-space
K = K/K0, [ˆx, yˆ]=[x, y], x,ˆ yˆ ∈ Kˆ,x∈ x,ˆ y ∈ y,ˆ (5) is a Krein space with respect to the naturally reduced indefinite inner product [·, ·].
Let us note that each almost Pontryagin space is an almost Krein space. As in the Krein space case we say that a nonnegative/nonpositive subspace ± L± belongs to the class h , if it is represented as a sum of a finite-dimensional neutral and a uniformly positive/negative subspace.
Remark 2. From the definitions it follows immediately that nonnegative/nonpositi- ± ve subspace L± in an almost Pontryagin space belongs to the class h .
All operators below are considered to be linear, everywhere defined and bounded. An operator A in an almost Krein space is called selfadjoint, if [Ax, y]= [x, Ay] for all x, y ∈K. By definition, a selfadjoint operator A belongs to the class H,ifithasa maximal nonpositive and a maximal nonnegative invariant subspaces L± and each such a subspace belongs to h±, respectively. We say that a selfadjoint operator A belongs to K(H), if there is a selfadjoint operator B ∈ H, which commutes with A.
Remark 3. Let K be an almost Pontryagin space. From Remark 2 and the definition of the class H it follows that I ∈ H. Hence each selfadjoint operator in an almost Pontryagin space belongs to the class K(H). Moreover, from Theorem 4 below we have that every selfadjoint operator in an almost Pontryagin space belongs to the class H. 4 T.Ya. Azizov and L.I. Soukhotcheva
3. Invariant dual pairs
{L+, L−} is said to be a dual pair, if L± is a nonnegative/nonpositive subspace and [x+,x−] = 0 for all x± ∈L±. We prove an analog of a known result about invariant subspaces of operators of the class K(H) in a Krein space (see [3, § III.5]) for the case of almost Krein spaces. Theorem 4. Let A ∈ K(H) be a selfadjoint operator in an almost Krein space K and let B be a selfadjoint operator in H which commutes with A. Assume {L+, L−} is an A-invariant dual pair (it is not excluded that L+ = L− = {0}). + − ± Then there exists an A-invariant maximal dual pair {L , L } such that L± ⊂L . If {L+, L−} is also B-invariant, then we can choose a maximal dual pair which is simultaneously A-andB-invariant. In particular, there exists an A-invariant maximal dual pair {L+, L−} such that L± ∈ h±. Proof. Since the isotropic part K0 of K is a part of each maximal semidefinite 0 0 subspace and {L+ + K , L− + K } is A-invariant, without loss of generality we 0 assume that K ⊂L+ ∩L−. Consider the factor-space (5). Let us note that the isotropic part K0 of K is invariant under all selfadjoint operators in K, in particular, it is A-andB- invariant. Therefore the selfadjoint operators A and B in K, generated by A and B, are well defined. Since AB = BA and B ∈ H,wehaveA ∈ K(H). 0 Denote L± = L±/K . The dual pair {L+, L−} is A-invariant. By [3, Corol- lary III.5.13] and [3, Theorem III.1.13] there is an A-invariant maximal dual pair {L+, L−} in K.Then{L+, L−},where L± = {x ∈K|xˆ ∈ L±}, is a desired A-invariant dual pair. If {L+, L−} is an A-andB-invariant dual pair then by [3, Theorem III.5.12], [3, Theorem III.1.13] and the same arguments as above there is an A-andB- invariant maximal dual pair which is an extension of {L+, L−}. In particular, if L+ = L− = {0}, then, by the above statement, there exists an A-andB-invariant maximal dual pair {L+, L−}. B ∈ H implies L± ∈ h±. From Definition 1 it follows that a space K with an indefinite inner product [·, ·] is an almost Krein space if and only if it admits a decomposition in a direct sum 0 K = K + K1, (6) 0 with finite-dimensional isotropic part K and a subspace K1 which is a Krein space with respect to [·, ·]. + − ± 0 ± Let {L , L } be an A-invariant maximal dual pair in K.ThenL = K +L1 , ± ± + − where L1 = L ∩K1. The dual pair {L1 , L1 } is maximal in the Krein space K1. 0 + − 0 0 Let L1 = L1 ∩L1 and let M1 ⊂K1 be a subspace skew-linked with L1. Then: 0 0 + − 0 K = K + L1 + L1 + L1 + M1. (7) Linear Operators in Almost Krein Spaces 5
Introduce in K a Hilbert scalar product (·, ·) such that all subspaces in (7) are ± ± orthogonal to each others with respect to (·, ·)and(·, ·)|L1 = ± [·, ·]|L1 . Then with respect to (7) the operator A has a triangular representation ⎡ ⎤ A00 A01 A02 A03 A04 ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ 0 A11 A12 A13 A14⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ A = ⎢ 00A22 0 A24⎥ , (8) ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ 000A33 A34⎥ ⎣ ⎦
0000A44 ∗ where A0k, k =0, 1, 2, 3, 4, are bounded operators, A11 and A44 are similar, and A22 and A33 are selfadjoint with respect to the introduced scalar product. We omit information about other components of (8) since it is not essential in this paper. Corollary 5. The spectrum of a selfadjoint operator A ∈ K(H) in an almost Krein space is real, except for at most finite number of non-real normal eigenvalues. There are not more than a finite number of real eigenvalues λ of A with degenerate ker(A − λ). In particular, the set of eigenvalues with nontrivial Jordan chains is finite. 4 Proof. The triangular representation (8) implies σ(A)= k=0 σ(Akk). Since A22 and A33 are Hilbert space selfadjoint operators, the non-real spectrum of A coin- cides with the non-real spectra of A00, A11 and A44. Hence the statement about the non-real spectrum of A follows immediately from the finite dimensionality of 0 0 0 K + L1 + M1. Assume that A has an infinite number of real eigenvalues λ such that ker(A− λ) is degenerate. Let L be the closed linear span of the isotropic parts of these kernels. Let B ∈ H be an operator which commutes with A.ThenBL⊂L.This contradicts the assumption B ∈ H. The last statement of the corollary follows from the fact that kernels with non-trivial Jordan chains are degenerate.
4. Criteria for the completeness and basicity In the sequel we will need the following lemma. Lemma 6. Assume K is an almost Krein space and A ∈ K(H) is a selfadjoint operator in K with σ(A)={0}.ThenA is a nilpotent operator of finite rank. Proof. We consider the decomposition (7) of K and the corresponding decompo- sition (8) of A.Notethatσ(A)={0} implies
σ(A00)=σ(A11)=σ(A22)=σ(A33)=σ(A44)={0} and the selfadjointness of A22 and A33 implies A22 =0andA33 =0. 6 T.Ya. Azizov and L.I. Soukhotcheva
Below we use the following notations:
– Lλ(A) is the root subspace of A related to the eigenvalue λ : λ ∈ σp(A); – E(A) is the closed linear span (c.l.s. ) of the root vectors of A: E(A)= c.l.s. {Lλ(A) | λ ∈ σp(A)};
– E0(A)=c.l.s. {ker(A − λ) | λ ∈ σp(A)};
– ER\0(A)=c.l.s. {Lλ(A) | λ ∈ R \{0}}.
A vector system {ek} is called almost orthonormal if it is a union of two { } { }∪{ } { } systems ek = ek ek ,where ek is orthonormal: [ek,ek] =0, [ek,ej]= { } δkj sign [ek,ek], and ek is a finite system with [ek,ej ]=0. Corollary 7. Let A ∈ K(H) be a compact selfadjoint operator in a Krein space K. Then E(A)=K if and only if L0(A) ∩ ER\0(A)={0}. Moreover, if E(A)=K, there exists in K an almost orthonormal Riesz basis composed of root vectors of A.
Proof. Indeed, let L = L0(A)∩ER\0(A). Then L is orthogonal to all root subspaces of A.IfE(A)=K, we have the orthogonality of L to the whole Krein space which is non-degenerate. Hence L = {0}. Assume L = {0}.LetB ∈ H and AB = BA.SinceER\0(A)isA-andB- 0 invariant, the isotropic part ER\0(A) of ER\0(A) is also A-andB-invariant. Hence, 0 taking into account that B ∈ H and ER\0(A) is B-invariant neutral subspace, we 0 obtain from Theorem 4 that dim ER\0(A) < ∞. Since the root subspaces Lλ(A), 0 0 λ = 0 are non-degenerate, we have ER\0(A) ⊂L0(A), and hence ER\0(A) ⊂L, 0 that is, ER\0(A) = {0}. Therefore ER\0(A) is a Krein space. The spectrum of the [⊥] restriction of A to the orthogonal complement ER\0(A) to ER\0(A) consists of [⊥] a unique point λ = 0. Hence, it follows from Lemma 6 that L0(A)=ER\0(A) . The latter implies E(A)=K. The existence of an almost orthonormal basisisprovedin[3,Thm.IV.2.12]. Below we give necessary and sufficient conditions for the basicity and com- pleteness of the set of root vectors of a compact selfadjoint operator in an almost Krein space. We show that in almost Krein spaces, in contrast to the Pontryagin or Krein space case, the completeness of the root vector system is not sufficient for its basicity. Theorem 8. Let A ∈ K(H) be a compact selfadjoint operator in an almost Krein space K and E(A)=K. Then the following assertions are equivalent: 0 (i) K ∩L0(A) ∩ ER\0(A)={0}; (ii) L0(A) ∩ ER\0(A)={0}; (iii) there exists in K an almost orthonormal Riesz basis consisting of root vectors of A. Proof. Corollary 5 implies that the non-real spectrum of A and the real nonzero eigenvalues λ ∈ σp(A) with degenerate kernels ker(A − λ) is a finite subset of the normal eigenvalues. Let P be the Riesz projector related to this part of σ(A). Since Linear Operators in Almost Krein Spaces 7 dim P K < ∞ and P K is both A-andB-invariant, where B ∈ H commutes with A,wehavethatA|(I − P )K satisfies all conditions of the theorem and the basicity property for the root systems of A and A|(I − P )K is equivalent. Moreover, 0 0 K ∩L0(A) ∩ ER\0(A)=(I − P )K ∩L0(A|(I − P )K) ∩ ER\0(A|(I − P )K) and
L0(A) ∩ ER\0(A)={0}⇐⇒L0(A|(I − P )K) ∩ ER\0(A|(I − P )K)={0}. Hence without loss of generality we can assume P = 0, that is, both the set of the non-real eigenvalues and the set of nonzero real eigenvalues with degenerate kernels are empty. 0 (i) ⇒ (ii). Let K ∩L0(A) ∩ ER\0(A)={0}.ThenER\0(A) is a nondegenerate subspace. Indeed, let us suppose the opposite, i.e., that this subspace is degener- 0 0 ate and its isotropic part ER\0(A) is nonzero. The subspace ER\0(A) is finite- dimensional and A-invariant. Therefore, there exists an eigenvalue λ0 and a cor- 0 responding eigenvector x0 ∈ ER\0(A) . Since all kernels ker(A − λ)withλ =0are nondegenerate, we obtain λ0 =0.So,x0 ∈ ker A, it is orthogonal to L0(A)and 0 thus is orthogonal to E(A). Hence, x0 ∈K ∩L0(A) ∩ ER\0(A), and x0 =0.This contradicts the assumption x =0.Wehavethat ER\0(A) is nondegenerate and therefore L0(A) ∩ ER\0(A)={0}. (ii) ⇒ (i) is trivial. (ii) ⇒ (iii). According to the assumption all eigenvalues of A are real and all ker(A − λ), λ = 0, are nondegenerate. Hence the subspace ER\0(A) is nondegen- erate. Let us prove that this subspace is a Krein space. Really, it is B-invariant, where B ∈ H is an operator commuting with A. In each subspace ker(A − λ) {L+ L−} {L+ L−} the operator B has an invariant maximal dual pair λ , λ ,and , with L± {L± | ∈ } E =c.l.s. λ λ σ(A) ,isaB-invariant maximal dual pair in R\0(A). The assumption B ∈ H implies L± ∈ h±. Hence L+ + L− is an almost Krein space. By construction, the defect of this subspace in ER\0(A) is finite. Hence ER\0(A)is also an almost Krein space and, by assumption, it is nondegenerate. So, ER\0(A) is a Krein space. Therefore the orthogonal decomposition [⊥] K = ER\0(A)[+]ER\0(A) holds. Since B ∈ H has in ER\0(A) an invariant maximal dual pair, we have [⊥] B|ER\0(A) ∈ H and B|ER\0(A) ∈ H. Hence, [⊥] A|ER\0(A) ∈ K(H)andA|ER\0(A) ∈ K(H). [⊥] [⊥] By Lemma 6, the subspace ER\0(A) ⊃L0(A) coincides with L0(A|ER\0(A) ) ⊂ [⊥] L0(A). This implies ER\0(A) )=L0(A)and
K = ER\0(A)[+]L0(A). (9)
Since ER\0(A) is a Krein space and A|ER\0(A) ∈ K(H) it follows from [3, Theorem IV.2.12] that there exists in ER\0(A) an orthonormal Riesz basis constructed from 8 T.Ya. Azizov and L.I. Soukhotcheva eigenvectors of A. If we add to this basis an arbitrary almost orthonormal Riesz basis composed of vectors of L0(A), we obtain desired basis in K. (iii) ⇒ (ii). Assume there exists an almost orthonormal Riesz basis in K composed of root vectors of A.Let{ek} be the part of this basis contained in ER\0(A). Using the same arguments as above we can assume, without loss of generality, that all ∈L ∩ E vectors ek are definite and orthogonal to each other. If x0 0(A) R\0(A), then x0 ∈ c.l.s. {ek} : x0 = αkek. On the other hand it is orthogonal to all these vectors. Hence, αk =[x0,ek]sign[ek,ek] = 0, that is, x0 =0. Remark 9. In the proof of Theorem 8 we have used the assumption that the root vector system of A is complete only in the proof of the implication (i) ⇒ (ii). 0 We can rewrite the assumptions (i) and (ii) as 0 ∈/ σp(A|K ∩ ER\0)and 0 ∈/ σp(A|L0(A) ∩ ER\0), respectively. Theorem 10. Let A ∈ K(H) be a compact selfadjoint operator in an almost Krein space K.Then 0 {E(A)=K} ⇐⇒ {L0(A) ∩ ER\0(A) ⊂K }. (10) 0 Proof. {E(A)=K} =⇒{L0(A) ∩ ER\0(A) ⊂K }.
Let x0 ∈L0(A) ∩ ER\0(A). Since L0(A) is orthogonal to c.l.s. {Lλ(A) | λ =0 },the 0 vector x0 is orthogonal to E(A), that is, x0 ∈K . 0 {E(A)=K} ⇐= {L0(A) ∩ ER\0(A) ⊂K }. With respect to decomposition (6) the operator A admits the following matrix representation: ⎡ ⎤ A00 A01 A = ⎣ ⎦ . (11) 0 A11 Consider also the operator A induced by the operator A in the factor space K = 0 0 K/K . It follows from L0(A) ∩ ER\0(A) ⊂K that L0(A) ∩ ER\0(A)={0}.By Theorem 8, there is in K a Riesz basis composed of root vectors of A. Since the operators A and A11 are similar, there is in K1 a Riesz basis composed of root 0 vectors of A11. Hence, E(A11)=K1. It remains to use that E(A)=K [+]E(A11)= K (see (11)).
5. Keldysh type operators Let H be a Hilbert space with the scalar product (·, ·), let A and B be compact selfadjoint operators acting in this space and A>0. Consider an operator H as in (3) and introduce in H an inner product [·, ·]givenby [·, ·]=((I + B)·, ·). Denote K = {H, [·, ·]}.ThenK is an almost Krein space, moreover, K is an almost Pontryagin space. The operator H is selfadjoint in K. Thus, we can apply to H Theorems 8 and 10. Linear Operators in Almost Krein Spaces 9
Theorem 11. Let an operator H as in (3) satisfy the above-mentioned properties. Then (a) E(H)=K; (b) There is an almost orthonormal Riesz basis in the almost Krein space K composed of root vectors of H if and only if one of the following equivalent assumptions holds. (b1) dim ker A(I + B)=dimker(I + B)A; (b2) ker(I + B) ⊂ ran A. Proof. (a) The isotropic part K0 of the almost Krein space K coincides with ker(I+ B). Since A>0, we have L0(H)=kerH =ker(I +B). Hence, L0(H)∩ER\0(H) ⊂ ker(I + B)=K0.Now(a) follows directly from Theorem 10. (b) Let us note that the assumptions (b1) and (b2) are equivalent since the operator A has a trivial kernel. Suppose that there exists an almost orthonormal Riesz basis in K composed of root vectors of H. By Theorem 8(ii) we have
K =kerH[+]ER\0(H). (12)
Since the operator H is diagonal with respect to (12) and the operator H|ER\0(H) is a Pontryagin space selfadjoint, H|ER\0(H) is similar to its Hilbert space adjoint. Hence H = A(I + B) and its Hilbert space adjoint H∗ =(I + B)A are similar too. This implies (b1). Assume (b2). Consider the linear pencil L(λ)=A−1 − λ(I + B). (13) Since the set of root vectors of this pencil and of the operator H are the same, it is sufficient to check that there is in K an almost orthonormal Riesz basis composed of root vectors of the pencil (13). Consider the orthogonal decomposition of H: H =ker(I + B) ⊕ ran (I + B). (14) Let us rewrite the operator A−1 and the pencil (13) in the matrix form with respect to the decomposition (14): ⎡ ⎤ C00 C01 A−1 = ⎣ ⎦ , C10 C11 ⎡ ⎤ C00 C01 L(λ)=⎣ ⎦ . (15) C10 C11 − λ(I + B11)
From (b2) it follows that C00 is a finite-dimensional operator and positive with respect to the Hilbert space scalar product, that C01 and C10 are bounded operators of finite ranks and that the operator C11 is positive with respect to the x0 Hilbert space scalar product and has a compact inverse. Hence, x = is an x1 10 T.Ya. Azizov and L.I. Soukhotcheva
−1 eigenvector of (15), related to λ = 0, if and only if, x0 = −C00 C01x1,wherex1 is an eigenvector of the pencil −1 L1(λ)=C11 − C10C00 C01 − λ(I + B11), (16) corresponding to the same eigenvalue. As in the proof of Theorem 8 we assume without loss of generality that the root subspaces and kernels corresponding to −1 the same eigenvalue coincides. Since C11 − C10C00 C01 is a Hilbert space positive operator in ran (I + B), its inverse is compact, and ker(I + B11)={0},thesetsof eigenvectors of (16) coincide with the sets of eigenvectors of the operator −1 −1 H1 =(C11 − C10C00 C01) (I + B11). (17)
Consider ran (I + B) as a Pontryagin space K1 with indefinite inner product [·, ·]=((I + B11)·, ·). Because the compact operator H1 is positive in K1,the assumption (ii) of Theorem 8 is fulfilled. Hence, there is an almost orthonormal Riesz basis in K1 composed of eigenvectors of H1. Taking into account the relation between eigenvectors of H1 and H and the equality ker(I + B)=kerH,weobtain the existence of an almost orthonormal Riesz basis in K composed of eigenvectors vectors of H. The latter is true if there are no associated vectors. In the general case, there is an almost orthonormal Riesz basis in K composed of root vectors of H.
References [1] Azizov T.Ya. On completely continuous operators that are selfadjoint with respect to a degenerate indefinite metric, Matem. issled., 7 (1972), 4, 237–240. (Russ.) [2] Azizov T.Ya., Iokhvidov I.S. A criterion of the completeness and basicity of root vectors of a completely continuous J-selfadjoint operator in a Pontryagin space Πκ. Matem. issled., 6 (1971), 1, 158–161. (Russ.) [3] T.Ya. Azizov, I.S. Iokhvidov, Foundation of the theory of linear operators in spaces with an indefinite metric, Nauka, Moscow, 1986 (Russ.); English transl.: Linear op- erators in spaces with an indefinite metric, Wiley, New York, 1989. [4] P. Binding and R. Hryniv. Full and partial range completeness. Oper. Theory Adv. Appl. 130:121–133, 2002. [5] Vladimir Bolotnikov, Chi-Kwong Li, Patrick Meade, Christian Mehl, Leiba Rodman. Shells of matrices in indefinite inner product spaces. Electron. J. Linear Algebra, 9: 67–92, 2002. [6] I.Ts. Gokhberg, M.G. Krein. Introduction to the theory of linear non-selfadjoint operators in a Hilbert space. Nauka, Moscow, 1965 (Russian). [7] M. Kaltenb¨ack and H. Woracek. Selfadjoint extensions of symmetric operators in degenerated inner product spaces. Integral Equations Operator Theory, 28 (1997), 289–320. [8] P. Lancaster, A.S. Markus, and P. Zizler. The order of neutrality for linear operators on inner product spaces. LAA 259 (1997), 25–29. Linear Operators in Almost Krein Spaces 11
[9] H. Langer, R. Mennicken, and C. Tretter. A self-adjoint linear pencil Q − λP of ordinary differential operators. Methods Funct. anal. Topology, 2 (1996), 38–54. [10] A. Luger. A factorization of regular generalized Nevanlinna functions.IntegralEqua- tions Operator Theory 43: 326–345, 2002. [11] Christian Mehl, Leiba Rodman. Symmetric matrices with respect to sesquilinear forms. Linear Algebra Appl., 349: 55–75, 2002. [12] Christian Mehl, Andre C.M. Ran, Leiba Rodman. Semidefinite invariant subspaces: degenerate inner products, to appear in Oper. Theory Adv. Appl. [13] H. Woracek. Resolvent matrices in degenerate inner product spaces.Math.Nachr. 213 (2000), 155–175.
Tomas Ya. Azizov and Lioudmila I. Soukhotcheva Department of Mathematics Voronezh State University Universitetskaya pl., 1 394022, Voronezh, Russia e-mail: [email protected] e-mail: [email protected] Operator Theory: Advances and Applications, Vol. 175, 13–32 c 2007 Birkh¨auser Verlag Basel/Switzerland
Generalized Resolvents of a Class of Symmetric Operators in Krein Spaces
Jussi Behrndt, Annemarie Luger and Carsten Trunk
Abstract. Let A be a closed symmetric operator of defect one in a Krein space K and assume that A possesses a self-adjoint extension in K which locally has the same spectral properties as a definitizable operator. We show that the Krein-Naimark formula establishes a bijective correspondence between the compressed resolvents of locally definitizable self-adjoint extensions A of A acting in Krein spaces K×H and a special subclass of meromorphic functions. Mathematics Subject Classification (2000). Primary: 47B50; Secondary: 47B25. Keywords. Generalized resolvents, Krein-Naimark formula, self-adjoint exten- sions, locally definitizable operators, locally definitizable functions, boundary value spaces, Weyl functions, Krein spaces.
1. Introduction Let A be a densely defined closed symmetric operator with defect one in a Hilbert ∗ space K and let {C, Γ0, Γ1} be a boundary value space for the adjoint operator A . ∗ Let A0 be the self-adjoint extension A ker Γ0 of A in K and denote the γ-field and Weyl function corresponding to the boundary value space {C, Γ0, Γ1} by γ and M, respectively. Here M is a scalar Nevanlinna function, that is, it maps the upper half-plane C+ holomorphically into C+ ∪ R and is symmetric with respect to the real axis. It is well known that in this case the Krein-Naimark formula −1 −1 −1 ∗ PK(A − λ) |K =(A0 − λ) − γ(λ) M(λ)+τ(λ) γ(λ) (1.1) establishes a bijective correspondence between the class of Nevanlinna functions τ (including the constant ∞) and the compressed resolvents of self-adjoint extensions A of A in K×H,whereH is a Hilbert space, cf. [22, 32]. The compressed resolvent on the left-hand side of (1.1) is said to be a generalized resolvent of A.Wenote that if A has equal deficiency indices > 1 the generalized resolvents of A can still be described with formula (1.1), where the parameters τ are so-called Nevanlinna families, cf. [11, 23, 30, 31]. 14 J. Behrndt, A. Luger and C. Trunk
Various generalizations of the Krein-Naimark formula in an indefinite setting have been proved in the last decades. The case that A is a symmetric operator in a Pontryagin space K and H is a Hilbert space was investigated by M.G. Krein and H. Langer in [24]. Later V. Derkach considered both K and H to be Pontryagin or even Krein spaces, cf. [10]. In the general situation of Krein spaces K and H one obtains a correspondence between locally holomorphic relation-valued functions τ and self-adjoint extensions A of A with a non-empty resolvent set. Under additional assumptions other variants of (1.1) were proved in [6, 7, 8, 9, 10, 14, 27]. If, e.g., H is a Pontryagin space, then the parameters τ belong to the class of Nκ-families, a class of relation-valued functions which includes the generalized Nevanlinna functions. If H is a Krein space and the hermitian forms [A·, ·]and [A·, ·] both have finitely many negative squares, then τ belongs to a special subclass of the definitizable functions, cf. [7, 19]. It is the aim of this paper to prove a new variant of formula (1.1). Here we allow both K and H to be Krein spaces and we assume that A is of defect one and possesses a self-adjoint extension A0 in K which locally has the same spectral properties as a definitizable operator or relation, cf. [20, 28]. Under the assumption that A is also locally definitizable and that its sign types coincide “in essence” (i.e., with the exception of a discrete set, see Definition 2.6) with the sign types of A0 we prove in Theorem 3.2 that there exists a so-called locally definitizable function τ such that (1.1) holds. The proof is based on a coupling method developed in [11, §5] and a recent perturbation result from [4]. One of the main difficulties here is to show that the symmetric relation A∩H2 possesses a self-adjoint extension in the Krein space H with a non-empty resolvent set and to choose a boundary value space for the adjoint of A ∩H2 in H such that (1.1) holds with the corresponding Weyl function τ. In connection with a class of abstract λ-dependent boundary value problems the converse direction was already proved in [3], i.e., for a given locally definitizable function τ a self-adjoint extension A of A in K×Hsuch that (1.1) holds was constructed. The paper is organized as follows: In Section 2 we recall the definitions and basic properties of locally defini- tizable self-adjoint operators and relations and the class of locally definitizable functions introduced and studied by P. Jonas, see, e.g., [20, 21]. The notion of d- compatibility of sign types of locally definitizable relations and functions is defined in the end of Section 2.3. In the beginning of Section 3 we recall some basics on boundary value spaces and associated Weyl functions. Section 3.2 contains our main result. We prove in Theorem 3.2 that formula (1.1) establishes a bijective correspondence between an appropriate subclass of the locally definitizable functions and the compressed resolvents of locally definitizable K-minimal self-adjoint exit space extensions A of A in a Krein space K×Hwith spectral sign types d-compatible to those of A0. Finally, in the end of Section 3.2, we formulate a variant of the Krein-Naimark formula for self-adjoint extensions A0 and A of A in K and K×H, respectively, which locally have the same spectral Generalized Resolvents of Symmetric Operators in Krein Spaces 15 properties as self-adjoint operators or relations in Pontryagin spaces and functions τ from the local generalized Nevanlinna class.
2. Locally definitizable self-adjoint relations and locally definitizable functions 2.1. Notations and definitions Let (K, [·, ·]) be a separable Krein space with a corresponding fundamental sym- metry J. The linear space of bounded linear operators defined on a Krein space K1 with values in a Krein space K2 is denoted by L(K1, K2). If K := K1 = K2 we simply write L(K). We study linear relations in K, that is, linear subspaces of K2. The set of all closed linear relations in K is denoted by C(K). Linear operators in K are viewed as linear relations via their graphs. For the usual definitions of the linear operations with relations, the inverse etc., we refer to [15] and [16]. . The sum and the direct sum of subspaces in K2 are denoted by and . We define an indefinite inner product on K2 by f g f,ˆ gˆ := i [f,g ] − [f ,g] , fˆ = , gˆ = ∈K2. f g K2 · · J 0 −iJ ∈LK2 Then ( , [[ , ]]) is a Krein space and = iJ 0 ( ) is a correspond- ing fundamental symmetry. If necessary we will indicate the underlying space by subscripts, e.g., [[·, ·]] K2 . Let A be a linear relation in K. The adjoint relation A+ ∈ C(K) is defined as A+ := A[[ ⊥]] = hˆ ∈K2 | h,ˆ fˆ =0forallfˆ ∈ A , where A[[ ⊥]] denotes the orthogonal companion of A with respect to [[·, ·]] . A is said to be symmetric (self-adjoint)ifA ⊂ A+ (resp. A = A+). Let S be a closed linear relation in K. The resolvent set ρ(S)ofS ∈ C(K) is the set of all λ ∈ C such that (S − λ)−1 ∈L(K), the spectrum σ(S)ofS is the complement of ρ(S)inC. The extended spectrum σ(S)ofS is defined by σ(S)=σ(S)ifS ∈L(K)andσ(S)=σ(S) ∪{∞} otherwise. A point λ ∈ C is called a point of regular type of S, λ ∈ r(S), if (S − λ)−1 is a bounded operator. We say that λ ∈ C belongs to the approximate point spectrum of S, denoted by xn σap(S), if there exists a sequence yn ∈ S, n =1, 2,..., such that xn =1and limn→∞ yn − λxn =0.Theextended approximate point spectrum σap(S) of S is defined by ∪{∞} ∈ −1 σap(S) if 0 σap(S ) σap(S):= −1 . σap(S)if0∈ σap(S )
We remark, that the boundary points of σ(S)inC belong to σap(S). Next we recall the definitions of the spectra of positive and negative type of self-adjoint relations, cf. [20] (for bounded self-adjoint operators see [29]). For 16 J. Behrndt, A. Luger and C. Trunk equivalent descriptions of the spectra of positive and negative type we refer to [20, Theorem 3.18].
Definition 2.1. Let A0 be a self-adjoint relation in K.Apointλ ∈ σap(A0)is said to be of positive type (negative type) with respect to A0, if for every sequence xn yn ∈ A0, n =1, 2,..., with xn = 1, limn→∞ yn − λxn =0wehave lim inf [xn,xn] > 0 resp. lim sup [xn,xn] < 0 . n→∞ n→∞
If ∞∈σap(A0), then ∞ is said to be of positive type (negative type) with respect −1 to A0 if 0 is of positive type (resp. negative type) with respect to A0 .Wedenote the set of all points of σap(A0) of positive type (negative type) by σ++(A0)(resp. σ−−(A0)).
We remark that the self-adjointness of the relation A0 yields that the points of positive and negative type introduced in Definition 2.1 belong to R. An open subset ∆ of R is said to be of positive type (negative type)with respect to A0 if each point λ ∈ ∆ ∩ σ(A0) is of positive type (resp. negative type) with respect to A0.Anopensubset∆ofR is called of definite type with respect to A0 if it is either of positive or of negative type with respect to A0. For each λ ∈ σ++(A0)(σ−−(A0)) there exists an open neighborhood Uλ in C such that (Uλ ∩ σ(A0) ∩ R) ⊂ σ++(A0)(resp.(Uλ ∩ σ(A0) ∩ R) ⊂ σ−−(A0)), Uλ\R ⊂ ρ(A0)and −1 −1 (A0 − λ) ≤M|Im λ| holds for some M>0andallλ ∈Uλ\R, cf. [1], [20] (and [29] for bounded operators).
2.2. Locally definitizable self-adjoint relations In this section we briefly recall the notion of locally definitizable self-adjoint rela- tions and intervals of type π+ and type π− from [20]. Let Ω be some domain in C symmetric with respect to the real axis such that Ω ∩ R = ∅ and the intersections of Ω with the upper and lower open half-planes are simply connected.
Definition 2.2. Let A0 be a self-adjoint relation in the Krein space K such that σ(A0) ∩ (Ω\R) consists of isolated points which are poles of the resolvent of A0, and no point of Ω ∩ R is an accumulation point of the non-real spectrum of A0 in Ω. The relation A0 is said to be definitizable over Ω, if the following holds.
(i) Every point µ ∈ Ω ∩ R has an open connected neighborhood Iµ in R such that both components of Iµ\{µ} are of definite type with respect to A0. (ii) For every finite union ∆ of open connected subsets of R, ∆ ⊂ Ω ∩ R,there exists m ≥ 1, M>0 and an open neighborhood U of ∆inC such that −1 2m−2 −m (A0 − λ) ≤M(1 + |λ|) |Im λ| holds for all λ ∈U\R. Generalized Resolvents of Symmetric Operators in Krein Spaces 17
By [20, Theorem 4.7] a self-adjoint relation A0 in K is definitizable over C if and only if A0 is definitizable, that is, the resolvent set of A0 is non-empty and there exists a rational function r =0withpolesonlyin ρ(A0) such that r(A0) ∈L(K) is a nonnegative operator in K,thatis
[r(A0)x, x] ≥ 0 holds for all x ∈K(see [28] and [16, §4and§5]). + Let A0 = A0 be definitizable over Ω and let δ → E(δ) be the local spectral function of A0 on Ω ∩ R. Recall that E(δ) is defined for all finite unions δ of connected subsets of Ω ∩ R, δ ⊂ Ω ∩ R, the endpoints of which belong to Ω ∩ R and are of definite type with respect to A0 (see [20, Section 3.4 and Remark 4.9]). With the help of the local spectral function E(·) the open subsets of definite type in Ω ∩ R can be characterized in the following way. An open subset ∆, ∆ ⊂ Ω ∩ R, is of positive type (negative type) with respect to A0 if and only if for every finite union δ of open connected subsets of ∆, δ ⊂ ∆, such that the boundary points of δ in R are of definite type with respect to A0, the spectral subspace (E(δ)K, [·, ·]) (resp. (E(δ)K, −[·, ·])) is a Hilbert space (cf. [20, Theorem 3.18]). We say that an open subset ∆, ∆ ⊂ Ω ∩ R,isoftype π+ (type π−)with respect to A0 if for every finite union δ of open connected subsets of ∆, δ ⊂ ∆, such that the boundary points of δ in R are of definite type with respect to A0 the spectral subspace (E(δ)K, [·, ·]) is a Pontryagin space with finite rank of negativity (resp. positivity). We shall say that A0 is of type π+ over Ω(type π− over Ω) if Ω ∩ R is of type π+ (resp. type π−) with respect to A0 and σ(A0) ∩ Ω\R consists of eigenvalues with finite algebraic multiplicity. We remark, that spectral points in sets of type π+ and type π− can also be characterized with the help of approximative eigensequences (see [1, 2]).
2.3. Matrix-valued locally definitizable functions In this section we recall the definition of matrix-valued locally definitizable func- tions from [21]. Although in the formulation of the main theorem in Section 3.2 below only scalar locally definitizable functions appear, matrix-valued functions will be used within the proof. Let Ω be a domain as in the beginning of Section 2.2 and let τ be an L(Cn)- valued piecewise meromorphic function in Ω\R which is symmetric with respect to the real axis, that is τ(λ)=τ(λ)∗ for all points λ of holomorphy of τ.If,in addition, no point of Ω∩ R is an accumulation point of nonreal poles of τ we write τ ∈ M n×n(Ω). The set of the points of holomorphy of τ in Ω\R and all points µ ∈ Ω ∩ R such that τ can be analytically continued to µ and the continuations from Ω ∩ C+ and Ω ∩ C− coincide, is denoted by h(τ). The following definition of sets of positive and negative type with respect to matrix functions and Definition 2.4 below of locally definitizable matrix functions can be found in [21]. 18 J. Behrndt, A. Luger and C. Trunk
Definition 2.3. Let τ ∈ M n×n(Ω). An open subset ∆ ⊂ Ω ∩ R is said to be of n positive type with respect to τ if for every x ∈ C and every sequence (µk)of points in Ω ∩ C+ ∩ h(τ) which converges in C to a point of ∆ we have lim inf Im τ(µk)x, x ≥ 0. k→∞ An open subset ∆ ⊂ Ω ∩ R is said to be of negative type with respect to τ if ∆ is of positive type with respect to −τ. ∆ is said to be of definite type with respect to τ if ∆ is of positive or of negative type with respect to τ. Definition 2.4. A function τ ∈ M n×n(Ω) is called definitizable in Ω if the following holds.
(i) Every point µ ∈ Ω ∩ R has an open connected neighborhood Iµ in R such that both components of Iµ\{µ} are of definite type with respect to τ. (ii) For every finite union ∆ of open connected subsets in R, ∆ ⊂ Ω ∩ R,there exist m ≥ 1, M>0 and an open neighborhood U of ∆inC such that τ(λ) ≤M(1 + |λ|)2m |Im λ|−m holds for all λ ∈U\R. The class of L(Cn)-valued definitizable functions in Ω will be denoted by Dn×n(Ω). In the case n =1wewriteD(Ω) instead of D1×1(Ω) and we set D(Ω) := D(Ω) ∪{d∞}, 0 | ∈ C ∈ C C where d∞ denotes the relation c c ( ). A function τ ∈ M n×n(C) which is definitizable in C is called definitizable,see [20]. We note that τ ∈ M n×n(C) is definitizable if and only if there exists a rational function g symmetric with respect to the real axis such that the poles of g belong to h(τ) ∪{∞} and gτ is the sum of a Nevanlinna function and a meromorphic function in C (cf. [20]). For a comprehensive study of definitizable functions we refer to the papers [18, 19] of P. Jonas. We mention only that the generalized Nevanlinna class is a subclass of the definitizable functions. Recall that a function n×n τ ∈ M (C)issaidtobeageneralized Nevanlinna function if the kernel Kτ , τ(λ) − τ(µ) K (λ, µ)= , τ λ − µ has finitely many negative squares (see [25] and [26]). In [21] it is shown that a function τ ∈ M n×n(Ω) is definitizable in Ω if and only if for every finite union ∆ of open connected subsets of R such that ∆ ⊂ Ω∩R, τ can be written as the sum τ = τ0 + τ(0) (2.1) n n of an L(C )-valued definitizable function τ0 and an L(C )-valued function τ(0) which is locally holomorphic on ∆. We say that a locally definitizable function n×n τ ∈D (Ω) is a generalized Nevanlinna function in Ω if the function τ0 in (2.1) can be chosen as a generalized Nevanlinna function. Generalized Resolvents of Symmetric Operators in Krein Spaces 19
The class of L(Cn)-valued generalized Nevanlinna functions in Ω will be de- noted by N n×n(Ω). In the case n =1wewriteN (Ω) instead of N 1×1(Ω) and we set N (Ω) := N (Ω) ∪{d∞}, 0 | ∈ C ∈ C C where d∞ denotes the relation c c ( ). The following theorem is a consequence of [21, Propositions 2.8 and 3.4]. It establishes a connection between self-adjoint relations which are locally definiti- n zable (locally of type π+)andL(C )-valued locally definitizable functions (resp. local generalized Nevanlinna functions).
Theorem 2.5. Let Ω be a domain as above and let A0 be a self-adjoint relation in the Krein space K which is definitizable over Ω.Letγ ∈L(Cn, K) and S = S∗ ∈ n L(C ), fix some point λ0 ∈ ρ(A0) ∩ Ω and define + −1 τ(λ):=S + γ (λ − Re λ0)+(λ − λ0)(λ − λ0)(A0 − λ) γ for all λ ∈ ρ(A0) ∩ Ω. Then the following holds. (i) The function τ is definitizable in Ω, τ ∈Dn×n(Ω). n×n (ii) If A0 is of type π+ over Ω,thenτ belongs to N (Ω). (iii) An open subset ∆ of Ω ∩ R which is of positive type (negative type) with respect to A0 is of positive type (resp. negative type) with respect to τ. In the sequel we shall often assume that the sign types of self-adjoint relations which are definitizable over Ω, and definitizable functions in Ω coincide outside of a discrete set in Ω ∩ R. A notion for this concept is introduced in the next definition, cf. [3, Definition 2.8].
Definition 2.6. Let A0 and A1 be self-adjoint relations which are definitizable over Ω and let τ and τ be L(Cn)-valued definitizable functions in Ω. We shall say that the sign types of A0 and A1 (A0 and τ, τ and τ) are d-compatible in Ω if for every µ ∈ Ω ∩ R there exists an open connected neighborhood Iµ ⊂ Ω ∩ R of µ such that each component of Iµ\{µ} is either of positive type with respect to A0 and A1 (resp. A0 and τ, τ and τ) or of negative type with respect to A0 and A1 (resp. A0 and τ, τ and τ). n×n If A0 is definitizable over Ω and the function τ ∈D (Ω) is defined as in Theorem 2.5, then obviously the sign types of A0 and τ are d-compatible in Ω. A typical nontrivial situation where d-compatibility of sign types of locally definitizable self-adjoint relations appears is shown in Theorem 2.7 below. For a proof we refer to [4, Theorem 3.2].
Theorem 2.7. Let A0 and A1 be self-adjoint relations in K, assume that the set ρ(A0) ∩ ρ(A1) ∩ Ω is non-empty and that A0 is definitizable over Ω.If −1 −1 (A1 − µ) − (A0 − µ) is a finite rank operator for some (and hence for all) µ ∈ ρ(A0) ∩ ρ(A1) ∩ Ω,then A1 is definitizable over Ω and the sign types of A0 and A1 are d-compatible in Ω. 20 J. Behrndt, A. Luger and C. Trunk
3. Generalized resolvents of a class of symmetric operators 3.1. Boundary value spaces and Weyl functions associated with symmetric relations in Krein spaces Let (K, [·, ·]) be a separable Krein space, let J be a corresponding fundamental symmetry and let A be a closed symmetric relation in K.WesaythatA has defect m ∈ N ∪{∞}, if both deficiency indices ∗ ± n±(JA) = dim ker((JA) − λ),λ∈ C , of the symmetric relation JA in the Hilbert space (K, [J·, ·]) are equal to m. Here ∗ denotes the Hilbert space adjoint. We remark, that this is equivalent to the fact that there exists a self-adjoint extension of A in K and that each self- adjoint extension Aˆ of A in K satisfies dim A/Aˆ = m. We shall use the so-called boundary value spaces for the description of the self-adjoint extensions of closed symmetric relations in Krein spaces. The following definition is taken from [10]. Definition 3.1. Let A be a closed symmetric relation in the Krein space K.We + say that {G, Γ0, Γ1} is a boundary value space for A if G is a Hilbert space and + Γ0 + 2 0 1 →G →G Γ , Γ : A are mappings such that Γ := Γ1 : A is surjective, and the relation ˆ ˆ Γf,Γˆg G2 = f,gˆ K2 holds for all f,ˆ gˆ ∈ A+. In the following we recall some basic facts on boundary value spaces which can be found in, e.g., [8] and [10]. For the Hilbert space case we refer to [17], [12] and [13]. Let A be a closed symmetric relation in K.Then + [⊥] Nλ,A+ := ker(A − λ)=ran(A − λ) denotes the defect subspace of A at the point λ ∈ r(A)andweset fλ Nˆ + ∈N + λ,A := λfλ fλ λ,A . ˆ ˆ When no confusion can arise we write Nλ and Nλ instead of Nλ,A+ and Nλ,A+ . If there exists a self-adjoint extension A of A such that ρ(A ) = ∅ then we have + . A = A Nˆλ for all λ ∈ ρ(A ). + In this case there exists a boundary value space {G, Γ0, Γ1} for A such that ker Γ0 = A (cf. [10]). Let in the following A, {G, Γ0, Γ1} and Γ be as in Definition 3.1. It follows that the mappings Γ0 and Γ1 are continuous. The self-adjoint extensions
A0 := ker Γ0 and A1 := ker Γ1 + of A are transversal, i.e., A0 ∩A1 = A and A0 A1 = A . The mapping Γ induces, via −1 + AΘ := Γ Θ= fˆ ∈ A | Γfˆ ∈ Θ , Θ ∈ C(G), (3.1) Generalized Resolvents of Symmetric Operators in Krein Spaces 21 a bijective correspondence Θ → AΘ between the set of all closed linear relations + C(G)inG and the set of closed extensions AΘ ⊂ A of A. In particular (3.1) gives a one-to-one correspondence between the closed symmetric (self-adjoint) extensions of A and the closed symmetric (resp. self-adjoint) relations in G. If Θ is a closed operator in G, then the corresponding extension AΘ of A is determined by AΘ =ker Γ1 − ΘΓ0 . (3.2)
Assume that ρ(A0) = ∅ and denote by π1 the orthogonal projection onto the first component of K×K. For every λ ∈ ρ(A0) we define the operators −1 −1 γ(λ):=π1(Γ0|Nˆλ) ∈L(G, K)andM(λ):=Γ1(Γ0|Nˆλ) ∈L(G). The functions λ → γ(λ)andλ → M(λ) are called the γ-field and Weyl function corresponding to {G, Γ0, Γ1}. They are holomorphic on ρ(A0) and the relations −1 γ(ζ)=(1+(ζ − λ)(A0 − ζ) )γ(λ) (3.3) and M(λ) − M(ζ)∗ =(λ − ζ)γ(ζ)+γ(λ) (3.4) hold for all λ, ζ ∈ ρ(A0) (cf. [10]). It follows that + M(λ)=ReM(λ0)+γ(λ0) (λ − Re λ0) −1 (3.5) +(λ − λ0)(λ − λ0)(A0 − λ) γ(λ0) holds for a fixed λ0 ∈ ρ(A0)andallλ ∈ ρ(A0). If, in addition, the condition K =clsp{Nλ | λ ∈ ρ(A0)} is fulfilled, then it follows from (3.3) and (3.4) that the function M is strict,thatis M(λ) − M(µ)∗ ker = {0} (3.6) λ − µ λ∈h(M) holds for some (and hence for all) µ ∈ h(τ). If Θ ∈ C(G)andAΘ is the corresponding extension of A (see (3.1)), then for every point λ ∈ ρ(A0)wehave
λ ∈ ρ(AΘ) if and only if 0 ∈ ρ(Θ − M(λ)). (3.7)
For λ ∈ ρ(AΘ) ∩ ρ(A0) the well-known resolvent formula −1 −1 −1 + (AΘ − λ) =(A0 − λ) + γ(λ) Θ − M(λ) γ(λ) (3.8) holds (for a proof see, e.g., [10]).
3.2. A variant of the Krein-Naimark formula We choose a domain Ω as in the beginning of Section 2.2. Let A be a (not necessar- ily densely defined) closed symmetric operator in the Krein space K,let{G, Γ0, Γ1} be a boundary value space for A+ and let H be a further Krein space. A self-adjoint extension A of A in K×H is said to be an exit space extension of A and H is 22 J. Behrndt, A. Luger and C. Trunk called the exit space. The exit space extension A of A is said to be K-minimal if ρ(A) ∩ Ω is non-empty and K×H=clsp K, (A − λ)−1K|λ ∈ ρ(A) ∩ Ω holds. Note, that the definition of K-minimality depends on the domain Ω. The elements of K×H will be written in the form {k, h}, k ∈K, h ∈H.LetPK : K×H→H, {k, h} → k, be the projection onto the first component of K×H. Then the compression −1 PK(A − λ) |K,λ∈ ρ(A), of the resolvent of A to K is called a generalized resolvent of A.
Theorem 3.2. Let A be a closed symmetric operator of defect one in the Krein + space K and let {C, Γ0, Γ1}, A0 =kerΓ0, be a boundary value space for A with corresponding γ-field γ and Weyl function M. Assume that A0 is definitizable over Ω and that the condition K =clsp{Nλ,A+ | λ ∈ ρ(A0) ∩ Ω} is fulfilled. Then the following assertions hold. (i) Let A be a K-minimal self-adjoint exit space extension of A in K×Hwhich is definitizable over Ω and assume that the sign types of A and A0 are d- compatible in Ω. Then there exists a locally definitizable function τ ∈ D(Ω) such that the sign types of τ, A and A0 are d-compatible in Ω, ρ(A) ∩ ρ(A0) ∩ h(τ) ∩ Ω is a subset of h((M + τ)−1) and the formula −1 −1 −1 + PK(A − λ) |K =(A0 − λ) − γ(λ) M(λ)+τ(λ) γ(λ) (3.9) holds for all λ ∈ ρ(A) ∩ ρ(A0) ∩ h(τ) ∩ Ω. (ii) Let τ ∈ D(Ω) be a locally definitizable function such that M(µ)+τ(µ) =0 for some µ ∈ Ω, assume that the sign types of τ and A0 are d-compatible in Ω and let Ω be a domain with the same properties as Ω, Ω ⊂ Ω. Then there exists a Krein space H and a K-minimal self-adjoint exit space extension A of A in K×H which is definitizable over Ω , such that the sign types of A, τ and A0 are d-compatible in Ω , −1 ρ(A0) ∩ h(τ) ∩ h (M + τ) ∩ Ω
is a subset of ρ(A) and formula (3.9) holds for all points λ belonging to −1 ρ(A0) ∩ h(τ) ∩ h((M + τ) ) ∩ Ω .
Proof. The proof of assertion (i) consists of four steps. Let A be a K-minimal self- adjoint exit space extension of A in K×Hwhich is definitizable over Ω such that the sign types of A and A0 are d-compatible in Ω. Generalized Resolvents of Symmetric Operators in Krein Spaces 23
1. In this first step we prove assertion (i) for the case H = {0}. Here A is a canonical extension of A and therefore, by (3.1), there exists a self-adjoint constant τ ∈ R ∪{d∞} such that −1 −1 −1 + (A − λ) =(A0 − λ) − γ(λ) M(λ)+τ γ(λ) holds (cf. (3.8)), that is, A coincides with the canonical self-adjoint extension A−τ −1 of A and by (3.7) we have ρ(A−τ ) ∩ ρ(A0) ⊂ h((M + τ) ). Here each point in Ω∩R is of positive as well as of negative type with respect to τ and hence assertion (i) follows. 2. In the following we assume H = {0}. Following the lines of [11, §5] we define in this step a symmetric relation T in H and a special boundary value space for the adjoint T +. Below we will deal with direct products of linear relations. The following notation will be used. If U is a relation in K and V is a relation in H we shall write U × V for the direct product of U and V which is a relation in K×H, {f1,f2} f1 f2 U × V = ∈ U, ∈ V . {f1,f2} f1 f2 {f1,f2} f1 f2 For the pair we shall also write {fˆ1, fˆ2},wherefˆ1 = and fˆ2 = . {f1,f2} f1 f2 The linear relations k {k, 0} S := A ∩K2 = ∈ A k {k , 0} and h {0,h} T := A ∩H2 = ∈ A h {0,h } are closed and symmetric in K and H, respectively, and we have A ⊂ S.LetJK and JH be fundamental symmetries in the Krein spaces K and H, respectively, and choose JK 0 J := ∈L(K×H) 0 JH 2 as a fundamental symmetry in the Krein space K×H.ThenJKS = JA ∩K and 2 JHT = JA ∩H are symmetric relations in the Hilbert spaces (K, [JK·, ·]) and (H, [JH·, ·]), respectively. It follows from [11, §5] that the deficiency indices of JKS and −JHT coincide. As JKS is a symmetric extension of the symmetric operator JKA in the Hilbert space (K, [JK·, ·]) the deficiency indices n±(JKS)ofJKS are (1, 1) or (0, 0). The case n±(JKS) = 0 is impossible here as otherwise also the relation JHT would be self-adjoint in (H, [JH·, ·]) and therefore JA would coincide with JKS × JHT .ButasA = S × T is by assumption a K-minimal exit space extension of A we would obtain H = {0}, a contradiction. 24 J. Behrndt, A. Luger and C. Trunk
Therefore, it remains to consider the case n±(JKS) = 1. Then the operators A and S coincide. Let us show that A+ coincides with PK{k, h} {k, h} k {k, h} R = ∈ A = ∈ A . PK{k ,h} {k ,h} k {k ,h}
In fact, as A is self-adjoint we have g,PK{k ,h} − g ,PK{k, h} = {g,0}, {k ,h} − {g , 0}, {k, h} =0 g + ∈ {ˆ ˆ}∈ ˆ k ˆ h ⊂ for all g A and k, h A, k = k , h = h . Hence A R . Similarly it follows that R+ ⊂ A holds. Therefore A+ coincides with the closure of R and as + A has finite defect and R is an extension of A we conclude A = R. Replacing PK by the projection PH onto the second component of K×Hthesamearguments show + PH{k, h} {k, h} h {k, h} T = ∈ A = ∈ A . PH{k ,h} {k ,h} h {k ,h} We define the mappings PK and PH by + {k, h} k PK : A → A , → {k ,h } k and + {k, h} h PH : A → T , → . {k ,h } h + + In the sequel we denote the elements in A and T by fˆ1 and fˆ2, respectively. −1 −1 As the multivalued part of PH coincides with A it follows that Γ0PKPH and −1 + Γ1PKPH are operators. We define Γ0, Γ1 : T → C by ˆ −1 ˆ ˆ −1 ˆ ˆ + Γ0f2 := −Γ0PKPH f2, Γ1f2 := Γ1PKPH f2, f2 ∈ T , cf. [11]. Taking into account that A is self-adjoint, one verifies that {C, Γ0, Γ1} is a boundary value space for T +.Weset T0 := ker Γ0. (3.10) ˆ ˆ + + ˆ −1 ˆ An element {f1, f2}∈A × T belongs to A if and only if f1 − PKPH f2 is contained in A. Therefore A is the canonical self-adjoint extension of the symmetric relation A × T in K×Hgiven by + + A = {fˆ1, fˆ2}∈A × T | Γ0fˆ1 +Γ0fˆ2 =Γ1fˆ1 − Γ1fˆ2 =0 . (3.11)
3. In order to show that T0 has a non-empty resolvent set we construct in this step an auxiliary self-adjoint extension Tα of T in H such that ρ(Tα) ∩ Ωisnon-empty and with the help of Theorem 2.7 we will show that Tα is definitizable over Ω and that the sign types of Tα are d-compatible with the sign types of A and A0 in Ω. Generalized Resolvents of Symmetric Operators in Krein Spaces 25
2 It is easy to see that {C , Γ0 , Γ1 },where ˆ ˆ ˆ ˆ Γ0f1 ˆ ˆ Γ1f1 ˆ + ˆ + Γ0 {f1, f2} := and Γ1 {f1, f2} := , f1 ∈ A , f2 ∈ T , (3.12) Γ0fˆ2 Γ1fˆ2 is a boundary value space for A+ × T +. Setting ⎛ ⎞ 00−11 ⎜11 0 0⎟ W := ⎜ ⎟ ∈L(C4) ⎝10 0 0⎠ 00 0 1 and & ' Γ0 Γ0 := W (3.13) Γ1 Γ1 2 + + we obtain a boundary value space {C , Γ0, Γ1} for A × T . This follows, e.g., 4 from the fact that W is unitary in the Krein space (C , [[ ·, ·]] 4 ), where C 0 −iIC2 [[ ·, ·]] C4 := J·, · , J = , (3.14) iIC2 0 (see Section 2.1). Here we have A =kerΓ0. (3.15) As A is by assumption definitizable over Ω it follows from Theorem 2.5 and (3.5) ( 2 that the Weyl function M corresponding to {C , Γ0, Γ1} is a definitizable function in Ω, M( ∈D2×2(Ω), and the sign types of M( and A are d-compatible in Ω. It is not difficult to verify that −1 + ran PH(A − λ) |K =ker(T − λ)=Nλ,T + ,λ∈ ρ(A), holds. Since A is an K-minimal exit space extension of A we have −1 H =clsp ran PH(A − λ) |K | λ ∈ ρ(A) ∩ Ω (3.16) =clsp Nλ,T + | λ ∈ ρ(A) ∩ Ω and from the assumption K =clsp{Nλ,A+ | λ ∈ ρ(A0) ∩ Ω} we obtain K×H=clsp Nλ,A+×T + | λ ∈ ρ(A0) ∩ ρ(A) ∩ Ω . This implies that the function M( ∈D2×2(Ω) is strict (see (3.6)). We claim that ∈ R there exists α such that the function 00 λ → M((λ) − 0 α ∈ ∩ ( 2 is invertible for some λ ρ(A) Ω. Indeed, let M(λ)=(mij (λ))i,j=1 and suppose that for all λ ∈ ρ(A) ∩ Ω and every α ∈ R we have ( 00 det M(λ) − = m11(λ)(m22(λ) − α) − m21(λ)m12(λ)=0. 0 α 26 J. Behrndt, A. Luger and C. Trunk
This implies m11(λ)=m12(λ)m21(λ) = 0 and since m12 and m21 are piecewise meromorphic functions in Ω\R and M( is symmetric with respect to the real axis we conclude m12(λ)=m21(λ)=0,λ ∈ ρ(A) ∩ Ω, which contradicts the strictness of M(. It is straightforward to check that the matrix ⎛ ⎞ 0010 ⎜ 0 −αα1⎟ V := ⎜ ⎟ ∈L(C4) (3.17) ⎝−1001⎠ 0 −110
4 2 is unitary in (C , [[ ·, ·]] C4 ), cf. (3.14). Let {C , Γ0, Γ1} be the boundary value space for A+ × T + defined by & ' & ' Γ0 Γ0 Γ0 := V = VW , (3.18) Γ1 Γ1 Γ1
(see (3.13)). From ⎛ ⎞ 1000 ⎜0 −α 01⎟ VW = ⎜ ⎟ ⎝0010⎠ 0 −100 we obtain ˆ ˆ ˆ Γ0f1 ˆ + ˆ + Γ0{f1, f2} = , f1 ∈ A , f2 ∈ T , Γ1fˆ2 − αΓ0fˆ2 and ˆ ˆ ˆ Γ1f1 ˆ + ˆ + Γ1{f1, f2} = , f1 ∈ A , f2 ∈ T . −Γ0fˆ2 We denote the self-adjoint extension ker(Γ1 − αΓ0) ∈ C(H)ofT in H by Tα.Then the self-adjoint extension ker Γ0 of A × T in K×Hcoincides with A0 × Tα. Since (3.18) and (3.17) imply 00 10 A0 × T =kerΓ0 =ker Γ0 + Γ1 α 0 −α α 1 00 =ker Γ1 − Γ0 , 0 α we find from (3.15), (3.2) and (3.7) that a point λ ∈ ρ(A) belongs to the set ρ(A0 × Tα) if and only if 0 belongs to the resolvent set of 00 M((λ) − . 0 α Generalized Resolvents of Symmetric Operators in Krein Spaces 27
But we have chosen α such that this function is invertible for some λ ∈ ρ(A) ∩ Ω, therefore λ belongs to ρ(A0 × Tα). In particular λ ∈ ρ(A0) ∩ ρ(Tα)and ρ(Tα) ∩ ρ(A0) ∩ ρ(A) ∩ Ω = ∅. As A × T is a symmetric relation of defect two and A and A0 × Tα are self-adjoint extensions of A × T in K×Hwe have −1 −1 dim ran (A − λ) − ((A0 × Tα) − λ) ≤ 2 for all λ ∈ ρ(A) ∩ ρ(A0) ∩ ρ(Tα) ∩ Ω. Since A is definitizable over Ω we obtain from Theorem 2.7 that also the self-adjoint relation A0 × Tα is definitizable over Ω and that the sign types of A and the sign types of A0 × Tα are d-compatible in Ω. It is a simple consequence from Definition 2.1 that σ++(A0 × Tα) ∩ σap(Tα) ⊂ σ++(Tα) and σ−−(A0 × Tα) ∩ σap(Tα) ⊂ σ−−(Tα) holds. Hence, real points from σ++(A0 ×Tα)(σ−−(A0 ×Tα)) belong to ρ(Tα)orto σ++(Tα)(resp.σ−−(Tα)). Therefore Tα is definitizable over Ω and the sign types of Tα in Ω are d-compatible with the sign types of A0 × Tα and, hence, with the sign types of A and A0 in Ω.
4. In this step we show that also T0 in (3.10) has a non-empty resolvent set and that formula (3.9) holds with the Weyl function τ corresponding to the boundary value space {C, Γ0, Γ1}. Moreover, we show that τ is locally definitizable and that its sign types are d-compatible with the sign types of A0 and A in Ω. It is straightforward to verify that {C, Γ1 − αΓ0, −Γ0} is a boundary value + space for T and we have Tα =ker(Γ1 − αΓ0)andT0 = ker(−Γ0). The cor- responding Weyl function τα is defined for all λ ∈ ρ(Tα). As Tα is definitizable over Ω the function τα belongs to the class D(Ω) and the sign types of τα are d-compatible with the sign types of Tα, A and A0 in Ω (cf. Theorem 2.5 and (3.5) or [3, Proposition 3.2]). Relation (3.16) implies that τα is strict and in particular τα is not identically equal to zero. ∈ ∩ h −1 Then, by (3.8), for λ ρ(Tα) (τα )wehave 1 − −1 − −1 − + (T0 λ) =(Tα λ) γα(λ) γα(λ) , τα(λ) {C − − } where γα is the γ-field of the boundary value space , Γ1 αΓ0, Γ0 . Therefore the set ρ(Tα)∩ρ(T0)∩Ω is non-empty and by Theorem 2.7 the self-adjoint relation T0 is definitizable over Ω and the sign types of T0 and Tα are d-compatible in Ω. The Weyl function τ corresponding to the boundary value space {C, Γ0, Γ1} satisfies − −1 ∈ h −1 ∩ τ(λ)= τα(λ) + α, λ (τα ) ρ(Tα). and is holomorphic on ρ(T0). It follows from Theorem 2.5 and (3.5) that τ belongs to the class D(Ω) and that its sign types are d-compatible with the sign types of T0 and Tα and hence 28 J. Behrndt, A. Luger and C. Trunk
also with the sign types of A and A0.Theγ-field corresponding to {C, Γ0, Γ1} will be denoted by γ . Since A0 and T0 are both definitizable over Ω the set ρ(A0)∩ρ(T0)∩Ωisnon- empty. The γ-field γ and the Weyl function M corresponding to the boundary value space {C2, Γ , Γ } defined in (3.12) are given by 0 1 γ(λ)0 λ → γ (λ)= ,λ∈ ρ(A0) ∩ ρ(T0) ∩ Ω, (3.19) 0 γ (λ) and M(λ)0 λ → M (λ)= ,λ∈ ρ(A0) ∩ ρ(T0) ∩ Ω, (3.20) 0 τ(λ) respectively. The relation {u, −u} Θ:= u, v ∈ C ∈ C(C2) (3.21) {v, v} is self-adjoint and the corresponding self-adjoint extension of A × T is given by −1 Γ0 + + Θ= {fˆ1, fˆ2}∈A × T | Γ0fˆ1 +Γ0fˆ2 =Γ1fˆ1 − Γ1fˆ2 =0 (3.22) Γ1 and coincides with A (see (3.11)). By (3.7) a point λ ∈ ρ(A0 × T0) belongs to ρ(A) if and only if 0 ∈ ρ(Θ − M (λ)). Hence, for λ ∈ ρ(A0 × T0) ∩ ρ(A) ∩ Ω=ρ(A0) ∩ h(τ) ∩ ρ(A) ∩ Ω −1 {v − M(λ)u, v + τ(λ)u} Θ − M (λ) = u, v ∈ C {u, −u} is an operator. Therefore (M(λ)+τ(λ))u = 0 implies u = 0 and we conclude that −1 the set ρ(A)∩ρ(A0)∩h(τ)∩Ω is a subset of h((M + τ) ). Setting x = v − M(λ)u and y = v + τ(λ)u we obtain −1 −1 u = − M(λ)+τ(λ) x + M(λ)+τ(λ) y for λ ∈ ρ(A0 × T0) ∩ ρ(A) ∩ Ω. This implies −1 −1 −1 −(M(λ)+τ(λ)) (M(λ)+τ(λ)) Θ − M (λ) = . (3.23) (M(λ)+τ(λ))−1 −(M(λ)+τ(λ))−1 For all λ ∈ ρ(A0 × T0) ∩ ρ(A) ∩ Ωtherelation −1 −1 −1 + (A − λ) = (A0 × T0) − λ + γ (λ) Θ − M (λ) γ (λ) (3.24) holds (cf. (3.8)) and it follows from (3.24), (3.19) and (3.23) that the formula −1 −1 −1 + PK(A − λ) |K =(A0 − λ) − γ(λ) M(λ)+τ(λ) γ(λ) holds. This completes the proof of assertion (i). 5. Assertion (ii) was already proved in [3] in a slightly different form. For the convenience of the reader we sketch the proof. Generalized Resolvents of Symmetric Operators in Krein Spaces 29
If τ is identically equal to a real constant, then A−τ := ker(Γ1 + τΓ0)isa canonical self-adjoint extension of A. As the Weyl function M corresponding to + A and {C, Γ0, Γ1} is strict we obtain ρ(A−τ ) ∩ Ω = ∅ and Theorem 2.7 implies that A−τ is definitizable over Ω and that the sign types of A0, A−τ and τ ∈ R are d-compatible. By (3.8) −1 −1 −1 + (A−τ − λ) =(A0 − λ) − γ(λ) M(λ)+τ γ(λ) ∈ ∩ −1 0 | ∈ C holds for all λ ρ(A0) ((M + τ) ). In the case τ = d∞ = c c we have A−τ = A0. Assume now that τ ∈D(Ω) is not equal to a constant and let Ω be a domain with the same properties as Ω, Ω ⊂ Ω. With the help of [21, Theorem 3.8] it was shown in [3, Theorem 3.3] that there exists a Krein space H, a closed symmetric + operator T of defect one in H and a boundary value space {C, Γ0, Γ1} for T such that T0 := ker Γ0 is definitizable over Ω ,thesigntypesofτ and T0 are d-compatible and τ coincides with the Weyl function corresponding to {C, Γ0, Γ1} on Ω ∩ ρ(T0). Moreover the condition H =clsp γ (λ) | λ ∈ ρ(T0) ∩ Ω (3.25)
2 + + is fulfilled. We choose the boundary value space {C , Γ0 , Γ1 } for A × T as in (3.12) with γ-field and Weyl function given by (3.19) and (3.20), respectively. The self-adjoint extension corresponding to Θ in (3.21) via (3.1) is denoted by A.Then A has the form (3.22) and the relation (3.24) holds for all λ ∈ Ω which belong to ρ(A0 × T0)andfulfil0∈ ρ(Θ − M (λ)). From (3.23) we conclude −1 ρ(A0) ∩ h(τ) ∩ h (M + τ) ∩ Ω ⊂ ρ(A) and (3.24) implies that the formula (3.9) holds. Since the minimality condition (3.25) is fulfilled it follows from (3.24) that A is a K-minimal exit space extension of A.AsA0 × T0 is definitizable over Ω the relation (3.24) and Theorem 2.7 imply that A is also definitizable over Ω and the sign types of A, A0 and τ are d-compatible.
The next theorem is a variant of the Krein-Naimark formula for the case that A0 and A are locally of type π+ and τ is a local generalized Nevanlinna function. The proof of Theorem 3.3 below is essentially the same as the proof of Theo- rem 3.2. Instead of the result on finite rank perturbations of locally definitizable self-adjoint relations from [4], cf. Theorem 2.7, one has to use [5, Theorem 2.4] on the stability of self-adjoint operators and relations locally of type π+ under compact perturbations in resolvent sense. We leave the details to the reader.
Theorem 3.3. Let A be a closed symmetric operator of defect one in the Krein + space K and let {C, Γ0, Γ1} be a boundary value space for A with corresponding γ-field γ and Weyl function M. Assume that A0 =kerΓ0 is of type π+ over Ω and that the condition K =clsp{Nλ,A+ | λ ∈ ρ(A0) ∩ Ω} is fulfilled. 30 J. Behrndt, A. Luger and C. Trunk
Then the following assertions hold: (i) For every K-minimal self-adjoint exit space extension A of A in K×H which is of type π+ over Ω there exists a function τ ∈ N (Ω) such that ρ(A) ∩ ρ(A0) ∩ h(τ) ∩ Ω is a subset of h((M + τ)−1) and the formula −1 −1 −1 + PK(A − λ) |K =(A0 − λ) − γ(λ) M(λ)+τ(λ) γ(λ) (3.26) holds for all λ ∈ ρ(A) ∩ ρ(A0) ∩ h(τ) ∩ Ω. (ii) Let τ ∈ N(Ω) be a local generalized Nevanlinna function such that M(µ)+ τ(µ) =0 for some µ ∈ Ω and let Ω be a domain with the same properties as Ω, Ω ⊂ Ω. Then there exists a Krein space H and a K-minimal self-adjoint exit space extension A of A in K×Hwhich is of type π+ over Ω , such that −1 ρ(A0) ∩ h(τ) ∩ h (M + τ) ∩ Ω
is a subset of ρ(A) and formula (3.26) holds for all points λ belonging to −1 ρ(A0) ∩ h(τ) ∩ h((M + τ) ) ∩ Ω.
References
[1] T.Ya. Azizov, J. Behrndt, P. Jonas, C. Trunk: Spectral Points of Type π+ and Type π− for Closed Linear Relations in Krein Spaces, submitted.
[2] T.Ya. Azizov, P. Jonas, C. Trunk: Spectral Points of Type π+ and Type π− of Selfadjoint Operators in Krein Spaces, J. Funct. Anal. 226 (2005), 114–137. [3] J. Behrndt: A Class of Abstract Boundary Value Problems with Locally Definitizable Functions in the Boundary Condition, Operator Theory: Advances and Applications 163 (2005), 55–73. [4] J. Behrndt: Finite Rank Perturbations of Locally Definitizable Operators in Krein Spaces, to appear in J. Operator Theory. [5] J. Behrndt, P. Jonas: On Compact Perturbations of Locally Definitizable Selfadjoint Relations in Krein Spaces, Integral Equations Operator Theory 52 (2005), 17–44. [6] J. Behrndt, H.C. Kreusler: Boundary Relations and Generalized Resolvents of Sym- metric Relations in Krein Spaces, submitted. [7] J. Behrndt, C. Trunk: On Generalized Resolvents of Symmetric Operators of Defect One with Finitely many Negative Squares, Proceedings of the Algorithmic Informa- tion Theory Conference, Vaasan Yliop. Julk. Selvityksi¨a Rap., 124,
[10] V.A. Derkach: On Generalized Resolvents of Hermitian Relations in Krein Spaces, J.Math.Sci.(NewYork)97 (1999), 4420–4460. [11] V.A. Derkach, S. Hassi, M.M. Malamud, H.S.V. de Snoo: Generalized Resolvents of Symmetric Operators and Admissibility, Methods Funct. Anal. Topology 6 (2000), 24–53. [12] V.A. Derkach, M.M. Malamud: Generalized Resolvents and the Boundary Value Problems for Hermitian Operators with Gaps, J. Funct. Anal. 95 (1991), 1–95. [13] V.A. Derkach, M.M. Malamud: The Extension Theory of Hermitian Operators and the Moment Problem, J. Math. Sci. (New York) 73 (1995), 141–242. [14] V.A. Derkach, M.M. Malamud: On some Classes of Holomorphic Operator Functions with Nonnegative Imaginary Part, Operator Theory, Operator Algebras and related Topics (Timi¸soara, 1996), Theta Found., Bucharest (1997), 113–147. [15] A. Dijksma, H.S.V. de Snoo: Symmetric and Selfadjoint Relations in Krein Spaces I, Operator Theory: Advances and Applications 24,Birkh¨auser Verlag Basel (1987), 145–166. [16] A. Dijksma, H.S.V. de Snoo: Symmetric and Selfadjoint Relations in Krein Spaces II, Ann. Acad. Sci. Fenn. Math. 12, (1987), 199–216. [17] V.I. Gorbachuk, M.L. Gorbachuk: Boundary Value Problems for Operator Differen- tial Equations, Kluwer Academic Publishers, Dordrecht (1991). [18] P. Jonas: A Class of Operator-valued Meromorphic Functions on the Unit Disc, Ann. Acad. Sci. Fenn. Math. 17 (1992), 257–284. [19] P. Jonas: Operator Representations of Definitizable Functions, Ann. Acad. Sci. Fenn. Math. 25 (2000), 41–72. [20] P. Jonas: On Locally Definite Operators in Krein Spaces, in: Spectral Theory and Applications, Theta Foundation 2003, 95–127. [21] P. Jonas: On Operator Representations of Locally Definitizable Functions, Operator Theory: Advances and Applications 162,Birkh¨auser Verlag Basel (2005), 165–190. [22] M.G. Krein: On Hermitian Operators with Defect-indices equal to Unity, Dokl. Akad. Nauk SSSR, 43 (1944), 339–342. [23] M.G. Krein: On the Resolvents of an Hermitian Operator with Defect-index (m, m), Dokl. Akad. Nauk SSSR, 52 (1946), 657–660. [24] M.G. Krein, H. Langer: On Defect Subspaces and Generalized Resolvents of Her- mitian Operators in Pontryagin Spaces, Funktsional. Anal. i Prilozhen. 5 No. 2 (1971) 59-71; 5 No. 3 (1971) 54-69 (Russian); English transl.: Funct. Anal. Appl. 5 (1971/1972), 139–146, 217–228. [25] M.G. Krein, H. Langer: Uber¨ einige Fortsetzungsprobleme, die eng mit der Theorie hermitescher Operatoren im Raume Πκ zusammenh¨angen. I. Einige Funktionen- klassen und ihre Darstellungen, Math. Nachr. 77 (1977), 187–236. [26] M.G. Krein, H. Langer: Some Propositions on Analytic Matrix Functions Related to the Theory of Operators in the Space Πκ, Acta Sci. Math. (Szeged), 43 (1981), 181–205. [27] H. Langer: Verallgemeinerte Resolventen eines J-nichtnegativen Operators mit endlichem Defekt, J. Funct. Anal. 8 (1971), 287–320. 32 J. Behrndt, A. Luger and C. Trunk
[28] H. Langer: Spectral Functions of Definitizable Operators in Krein Spaces, Functional Analysis Proceedings of a Conference held at Dubrovnik, Yugoslavia, November 2–14 (1981), Lecture Notes in Mathematics 948, Springer Verlag Berlin-Heidelberg-New York (1982), 1–46. [29] H. Langer, A. Markus, V. Matsaev: Locally Definite Operators in Indefinite Inner Product Spaces, Math. Ann. 308 (1997), 405–424. [30] H. Langer, B. Textorius: On Generalized Resolvents and Q-functions of Symmetric Linear Relations (Subspaces) in Hilbert Space, Pacific J. Math. 72 (1977), 135–165. [31] M.M. Malamud: On a Formula for the Generalized Resolvents of a Non-densely de- fined Hermitian Operator (Russian), Ukrain. Mat. Zh. 44 (1992), 1658–1688; trans- lation in Ukrainian Math. J. 44 (1993), 1522–1547. [32] M.A. Naimark: On Spectral Functions of a Symmetric Operator, Izv. Akad. Nauk SSSR, Ser. Matem. 7 (1943), 373–375.
Jussi Behrndt Institut f¨ur Mathematik, MA 6-4 Technische Universit¨at Berlin Straße des 17. Juni 136 D-10623 Berlin, Germany e-mail: [email protected] Annemarie Luger Institut f¨ur Analysis und Scientific Computing Technische Universit¨at Wien Wiedner Hauptstraße 8-10 A-1040 Wien, Austria e-mail: [email protected] Carsten Trunk Institut f¨ur Mathematik, MA 6-3 Technische Universit¨at Berlin Straße des 17. Juni 136 D-10623 Berlin, Germany e-mail: [email protected] Operator Theory: Advances and Applications, Vol. 175, 33–49 c 2007 Birkh¨auser Verlag Basel/Switzerland
Block Operator Matrices, Optical Potentials, Trace Class Perturbations and Scattering
Jussi Behrndt, Hagen Neidhardt and Joachim Rehberg
Abstract. For an operator-valued block-matrix model, which is called in quan- tum physics a Feshbach decomposition, a scattering theory is considered. Un- der trace class perturbations the channel scattering matrices are calculated. Using Feshbach’s optical potential it is shown that for a given spectral param- eter the channel scattering matrices can be recovered either from a dissipative or from a Lax-Phillips scattering theory. Mathematics Subject Classification (2000). Primary 47A40; Secondary 47A55, 47B44. Keywords. Feshbach decomposition, optical potential, Lax-Phillips scattering theory, dissipative scattering theory, scattering matrix, characteristic func- tion, dissipative operators.
1. Introduction
Let L and L0 be self-adjoint operators in a separable Hilbert space L and denote ac by P (L0) the orthogonal projection onto the absolutely continuous subspace ac L (L0)ofL0. The pair {L, L0} of self-adjoint operators is said to perform a scattering system if the wave operators W±(L, L0),
itL −itL0 ac W±(L, L0):=s − lim e e P (L0), (1.1) t→±∞ exist, cf. [5]. If the wave operators exist, then they are isometries from the ab- ac solutely continuous subspace L (L0) into the absolutely continuous subspace ac ac L (L), i.e., ran (W±(L, L0)) ⊆ L (L). The scattering system {L, L0} is called ac complete if the ranges of the wave operators W±(L, L0) coincide with L (L), cf. [5]. The operator ∗ S(L, L0):=W+(L, L0) W−(L, L0) is called the scattering operator of the scattering system {L, L0}. The scattering ac ac operator regarded as an operator in L (L0)commuteswithL0 . If the scattering 34 J. Behrndt, H. Neidhardt and J. Rehberg
ac system {L, L0} is complete, then S(L, L0) is a unitary operator in L (L0). In physical applications L0 is usually called the unperturbed or free Hamiltonian while L is called the perturbed or full Hamiltonian. Since S(L, L0)commutes with the free Hamiltonian L0 the scattering operator is unitarily equivalent to a multiplication operator induced by a family {S(λ)}λ∈R of unitary operators in the spectral representation of L0. This family is called the scattering matrix of the complete scattering system {L, L0} and is the most important quantity in the analysis of scattering processes. In this paper we investigate the special case that the Hilbert space L splits into two subspaces H1 and H2,
H1 L = ⊕ , H2 and the unperturbed Hamiltonian L0 is of the form
H1 H1 H1 0 L0 = : ⊕ −→ ⊕ . (1.2) 0 H2 H2 H2
In physics the subspaces Hj and the self-adjoint operators Hj, j =1, 2, are often called scattering channels and channel Hamiltonians, respectively. With respect to the decomposition (1.2) one introduces the channel wave operators
itL −itHj ac W±(L, Hj):=s − lim e Jje P (Hj) t→±∞ where Jj : Hj −→ L is the natural embedding operator. Introducing the channel scattering operators ∗ Sij = W+(L, Hi) W−(L, Hj):Hj −→ Hi,i,j=1, 2, one obtains a channel decomposition of the scattering operator S11 S12 S(L, L0)= . (1.3) S21 S22 In physics the decomposition (1.2) is often motivated either by the exclusive in- terest to scattering data in a certain channel or by the limited measuring process which allows to measure the scattering data only of a certain channel, say H1. ac ac Thus, let us assume that only the channel scattering operator S11 : H1 −→ H1 in the scattering channel H1 is known. This gives rise to the following problem: Is it possible to replace the full Hamiltonian L by an effective one H acting only in H1 such that the scattering operator of the scattering system {H, H1} coincides with S11?SinceS11 is a contraction, in general, this implies that either the scattering system {H, H1} cannot be complete or H is not self-adjoint. The problem has a solution within the scope of dissipative scattering systems developed in [18, 19, 20] for pairs {H, H1} of dissipative and self-adjoint operators Block Operator Matrices and Scattering 35
D in some separable Hilbert space. For such pairs the wave operators W± (H, H1) are defined by ∗ D itH −itH1 ac W+ (H, H1):=s − lim e e P (H1) t→+∞ and D −itH itH1 ac W− (H, H1):=s − lim e e P (H1), t→+∞ and the notion of completeness is generalized, cf. [18, 19, 20]. The scattering op- erator of a dissipative scattering system {H, H1} is defined by D ∗ D SD := W+ (H, H1) W− (H, H1).
It turns out that SD is a contraction acting on the absolutely continuous sub- ac space H1 of H1 which commutes with H1.In[17,18]itwasshownthatforany self-adjoint operator H1 in H1 and any contraction SD acting on the absolutely ac continuous subspace H1 and commuting with H1 there is a maximal dissipa- tive operator H on H1 such that {H, H1} performs a complete scattering system with scattering operator given by SD. In particular, this holds for the self-adjoint operator H1 and the channel scattering operator S11. That means, there is a max- imal dissipative operator H on H1 such that the channel scattering operator S11 is the scattering operator of the complete dissipative scattering system {H, H1}. Hence, roughly speaking, the scattering operator S11 can be always viewed as the scattering operator of a suitable chosen dissipative scattering system on H1.The disadvantage of this fact is that H is not known explicitly. Another approach to this problem was suggested by Feshbach in [10, 11], see also [6, 9]. He proposes a concrete dissipative perturbation V1 of the channel Hamiltonian H1, called “optical potential”, such that the scattering operator S1 of the dissipative scattering system {H1 + V1,H1} approximates S11 with a certain accuracy. To explain this approach in more detail let us assume that the full Hamiltonian L is obtained from L0 by an additive perturbation, L = L0 + V , where V is given by H H 0 G 1 1 V = : ⊕ −→ ⊕ . (1.4) G∗ 0 H2 H2 Introducing the “optical potential” −1 ∗ V1(λ):=−G(H2 − λ − i0) G ,λ∈ R, (1.5) it was shown in [8, Theorem 4.4.4] that under strong assumptions indeed the scat- tering operator S1[λ] of the (in general dissipative) scattering system {H1(λ),H1},
H1(λ):=H1 + V1(λ),λ∈ R, (1.6) coincides with the scattering operator S11 with an error of second order in the coupling constant. We show that Feshbach’s proposal can be made precise in another sense. Note first that the decomposition (1.2) leads not only to the decomposition (1.3) 36 J. Behrndt, H. Neidhardt and J. Rehberg of the scattering operator S but also to a decomposition of the scattering matrix {S(µ)}µ∈R, S11(µ) S12(µ) S(µ):= , S21(µ) S22(µ) where {Sij (µ)}µ∈R are called the channel scattering matrices. Denoting the scat- tering matrix of the dissipative scattering system {H1(λ),H1} by {S1[λ](µ)}µ∈R we prove that
S11(λ)=S1[λ](λ) (1.7) holds for a.e. λ ∈ R. This shows, that Feshbach’s proposal gives in fact a good approximation of the channel scattering matrix {S11(µ)}µ∈R in a neighborhood of the chosen spectral parameter λ of the optical potential V1(λ). Moreover, Feshbach’s proposal implies a second problem. Similarly to the optical potential V1(λ) in the first channel H1 one can introduce an optical potential V2(λ) in the second channel,
∗ −1 V2(λ):=−G (H1 − λ − i0) G, λ ∈ R, (1.8) and define a perturbed operator H2(λ),
H2(λ):=H2 + V2(λ),λ∈ R, (1.9) in H2. We show below that the characteristic function Θ2[λ](ξ), ξ ∈ C−,ofthe dissipative operator H2(λ) and the scattering matrix {S11(λ)}λ∈R are related by
∗ S11(λ)=Θ2[λ](λ) (1.10) for a.e. λ ∈ R. By [1]–[4] the last relation also yields that the scattering matrix {S11(λ)}λ∈R can be regarded as the scattering matrix SLP [λ](µ) of a Lax-Phillips scattering system at the point λ. Below we restrict ourself to a complete scattering system {L,L0}, L=L0 +V , where the perturbation V is a self-adjoint trace class operator. The assumption that V is a trace class operator is made for simplicity. Indeed, it would be sufficient −p −p to assume that the resolvent difference (L − z) − (L0 − z) is nuclear for a certain p ∈ N or, more generally, that the conditions of the so-called “stationary” scattering theory are satisfied, cf. [5, Section 14]. However, we emphasize that in contrast to [8] the smallness of the perturbation V is not assumed. Following the lines of [5] we show in Section 2 how the scattering matrix of the scattering system {L, L0} can be calculated. Under the additional assumptions (1.2) and (1.4) we find in Section 3 the channel scattering matrices {Sij (λ)}λ∈R. In Section 4 we prove relation (1.7). Section 5 is devoted to the proof of (1.10). Moreover, the ∗ Lax-Phillips scattering theory for which {Θ2[λ](µ) }µ∈R is the scattering matrix is indicated. Block Operator Matrices and Scattering 37
2. Scattering matrix
In this section we briefly recall the notion of the scattering matrix {S(λ)}λ∈R of a scattering system {L, L0}, where it is assumed that the unperturbed operator L0 is self-adjoint in the separable Hilbert space L and the perturbed operator L differs from L0 by a self-adjoint trace class operator V ∈B1(L), ∗ L = L0 + V, V = V ∈B1(L). (2.1)
Let E0(·) be the spectral measure of L0 and denote by B(R)thesetofallBorel subsets of the real axis R. Without loss of generality we assume throughout the paper that the condition
L =clospan{E0(∆)ran (|V |):∆∈ B(R)} (2.2) is satisfied, where |V | := (V ∗V )1/2. By Theorem X.4.4 of [13] the scattering system {L, L0} is complete, that is, the ranges of the wave operators W±(L, L0) in (1.1) coincide with the absolutely continuous subspace Lac(L)ofL. The operator V admits the representation V = |V |1/2C|V |1/2, |V | =(V ∗V )1/2,C=sgn(V ), (2.3) 1/2 where |V | belongs to the Hilbert-Schmidt class B2(L) and sgn(·) is the signum function. By Proposition 3.14 of [5] the limits |V |1/2(L − λ ± i0)−1|V |1/2 = lim |V |1/2(L − λ ± i )−1|V |1/2 (2.4) →+0 exist in B2(L) for a.e. λ ∈ R. The same holds for the limits 1/2 −1 1/2 |V | (L0 − λ ± i0) |V | . Moreover by Proposition 3.13 of [5] the derivative 1/2 1/2 |V | E0(dλ)|V | M0(λ):= ≥ 0 (2.5) dλ exists in B1(L) for a.e. λ ∈ R.Weset Qλ := clo ran (M0(λ)) ⊆ L.
By {Q(λ)}λ∈R we denote the family of orthogonal projections from L onto Qλ. One verifies that {Q(λ)}λ∈R is measurable. Let us consider the standard Hilbert space L2(R,dλ,L). On L2(R,dλ,L) we introduce the projection Q (Qf)(λ):=Q(λ)f(λ),λ∈ R,f∈ L2(R,dλ,L), and set Q =ran(Q). Further, in L2(R,dλ,L) we define the multiplication operator ML by (MLf)(λ):=λf(λ),λ∈ R, 2 2 dom (ML):= f ∈ L (R,dλ,L):λf(λ) ∈ L (R,dλ,L) .
Obviously, the multiplication operator ML and the projection Q commute. We set
MQ := ML dom (ML) ∩ Q. 38 J. Behrndt, H. Neidhardt and J. Rehberg
From Section 4.5 of [5] one gets that the absolutely continuous part Lac of the per- turbed operator L and the operator MQ are unitarily equivalent. In the following 2 we denote the subspace Q by L (R,dλ,Qλ) which can be regarded as the direct integral of the family of subspaces {Qλ}λ∈R with respect to the Lebesgue measure dλ on R, cf. [5]. ∗ ac Since the scattering operator S = W+(L, L0) W−(L, L0)actsonL (L0)and ac commutes with L0 there is a measurable family {S(λ)}λ∈R of operators
S(λ):Qλ −→ Qλ such that S is unitarily equivalent to the multiplication operator
(MQ(S)f)(λ):= S(λ)f(λ), 2 dom (MQ(S)) := L (R,dλ,Qλ).
The family {S(λ)}λ∈R is called the scattering matrix of the scattering system {L, L0}. Since the scattering system {L, L0} is complete the operator S(λ)is unitary on Qλ for a.e. λ ∈ R. The following representation theorem of the scattering matrix is a conse- quence of Corollary 18.9 of [5], see also [5, Section 18.2.2].
Theorem 2.1. Let L, L0 and V be self-adjoint operators in L as in (2.1).Then {L, L0} is a complete scattering system and the corresponding scattering matrix matrix {S(λ)}λ∈R admits the representation S λ I − πi M1/2 λ C − C|V |1/2 L − λ − i −1|V |1/2C M 1/2 λ ( )= Qλ 2 0 ( ) ( 0) 0 ( ) for a.e. λ ∈ R.
3. Channel scattering matrices Let us now assume that the Hilbert space L is the orthogonal sum of two subspaces H1 and H2, L = H1 ⊕ H2,thatL0 is a diagonal block operator matrix of the form
H1 H1 H1 0 L0 = : ⊕ −→ ⊕ , (3.1) 0 H2 H2 H2 cf. (1.2), where H1 and H2 are self-adjoint operators in H1 and H2 and that V ∈B1(L) is a self-adjoint trace class operator of the form H H 0 G 1 1 V = : ⊕ −→ ⊕ , (3.2) G∗ 0 H2 H2 see (1.4). The operator G : H2 −→ H1 describes the interaction between the channels. Since V is a trace class operator we have
G ∈B1(H2, H1). Block Operator Matrices and Scattering 39
The perturbed or full Hamiltonian L has the form
H1 H1 H1 G L := L0 + V = ∗ : ⊕ −→ ⊕ . (3.3) G H2 H2 H2 The following lemma is known as the Feshbach decomposition in physics, cf. [10, 11]. We use the notation
H1(z)=H1 + V1(z)andH2(z)=H2 + V2(z),z∈ C\R, (3.4) where −1 ∗ ∗ −1 V1(z)=−G(H2 − z) G and V2(z)=−G (H1 − z) G, (3.5) see (1.6), (1.9), (1.5) and (1.8).
Lemma 3.1. Let L, H1(z) and H2(z), z ∈ C\R, be given by (3.3) and (3.4), respectively. Then we have z ∈ res(Hi(z)), i =1, 2, for all z ∈ C\R and & ' −1 −1 −1 (H1(z) − z) −(H1 − z) G(H2(z) − z) L − z −1 . ( ) = −1 ∗ −1 −1 (3.6) −(H2(z) − z) G (H1 − z) (H2(z) − z) Proof. From 2 −1 ∗ 2 Im (H1(z) − z)h1,h1 =Imz h1 +Imz (H2 − z) G h1 , −1 z ∈ C\R, h1 ∈ H1, we conclude that (H1(z)−z) is a bounded everywhere defined −1 operator for all z ∈ C\R. Analogously one verifies that (H2(z)−z) is a bounded everywhere defined operator for all z ∈ C\R. A straightforward computation shows (L−z)−1 & ' −1 −1 −1 ∗ −1 −1 −1 (H1 −z) +(H1 −z) G(H2(z)−z) G (H1 −z) −(H1 −z) G(H2(z)−z) = −1 ∗ −1 −1 −(H2(z)−z) G (H1 −z) (H2(z)−z) for z ∈ C\R.Fromtheidentity ∗ −1 −1 −1 ∗ ∗ −1 −1 ∗ −1 I − G (H1 − z) G(H2 − z) G = G I − (H1 − z) G(H2 − z) G we obtain −1 −1 −1 ∗ −1 (H1 −z) +(H1 −z) G(H2(z)−z) G (H1 −z) −1 −1 ∗ −1 −1 −1 ∗ −1 =(H1 −z) I +G(H2 −z) I −G (H1 −z) G(H2 −z) G (H1 −z) −1 −1 ∗ −1 −1 =(H1 −z) I +G(H2 −z) G (H1(z)−z) =(H1(z)−z) for all z ∈ C\R, which proves (3.6). In the next lemma we calculate the limit |V |1/2(L − λ − i0)−1|V |1/2, λ ∈ R, cf. (2.4). Here and in the following it is convenient to use the functions ∗ 1/2 −1 ∗ 1/2 N1(z):=|G | (H1 − z) |G | , z ∈ C\R, (3.7) 1/2 −1 1/2 N2(z):=|G| (H2 − z) |G| , 40 J. Behrndt, H. Neidhardt and J. Rehberg and ∗ 1/2 −1 ∗ 1/2 F1(z):=|G | (H1(z) − z) |G | , z ∈ C\R. (3.8) 1/2 −1 1/2 F2(z):=|G| (H2(z) − z) |G| ,
Lemma 3.2. Let V ∈B1(L) be given by (3.2) with G ∈B1(H2, H1) and let U be a partial isometry such that G = U|G|. Then the limits
Ni(λ) := lim Ni(λ + i ) and Fi(λ) := lim Fi(λ + i ),i=1, 2, (3.9) →+0 →+0 exist in B2(H ) for a.e. λ ∈ R and the representation i 1/2 −1 1/2 F1(λ) −N1(λ)UF2(λ) |V | (L − λ − i0) |V | = ∗ (3.10) −F2(λ)U N1(λ) F2(λ) holds for a.e. λ ∈ R.
1/2 ∗ 1/2 Proof. By |G| ∈B2(H2)and|G | ∈B2(H1) the existence of the limits Ni(λ) in (3.9) for a.e. λ ∈ R follows from Proposition 3.13 of [5]. Using the representations ∗ 1/2 −1 ∗ 1/2 1/2 −1 1/2 F1(z)=|G | P1(L − z) H1 |G | and F2(z)=|G| P2(L − z) H2 |G| , z ∈ C\R, which follow from (3.6), and taking into account [5, Proposition 3.13] we again obtain the existence of Fi(λ), i =1, 2, for a.e. λ ∈ R. It is easy to see that |G∗|1/2 0 |V |1/2 =(V ∗V )1/4 = (3.11) 0 |G|1/2 holds. Let U be a partial isometry from H2 into H1 such that G = U|G|.Making use of the factorizations G = |G∗|1/2U|G|1/2 and G∗ = |G|1/2U ∗|G∗|1/2, (3.12) the block matrix representation of (L − z)−1 in Lemma 3.1 and relation (3.11) one verifies (3.10). We note that if U is a partial isometry such that G = U|G| holds and C is defined by 0 U C := , (3.13) U ∗ 0 then the operator V in (3.2) can be written in the form |V |1/2C|V |1/2, cf. (2.3). Let E1(·)andE2(·) be the spectral measures of H1 and H2, respectively. The operator function M0(·) from (2.5) here admits the representation M1(λ)0 M0(λ)= (3.14) 0 M2(λ) for a.e λ ∈ R,wherethederivatives ∗ 1/2 ∗ 1/2 1/2 1/2 |G | E1(dλ)|G | |G| E2(dλ)|G| M1(λ)= and M2(λ)= (3.15) dλ dλ exist in B1(H1)andB1(H2) for a.e. λ ∈ R, respectively. Setting Qj,λ := clo ran (Mj(λ)) ,j=1, 2, Block Operator Matrices and Scattering 41 and Qλ := Q1,λ ⊕ Q2,λ (3.16) for a.e. λ ∈ R we obtain the decomposition 2 2 2 L (R,dλ,Qλ)=L (R,dλ,Q1,λ) ⊕ L (R,dλ,Q2,λ), cf. Section 2. From (2.2) the conditions ∗ H1 =clospan{E1(∆)ran (|G |):∆∈ B(R)}, (3.17) H2 =clospan{E2(∆)ran (|G|):∆∈ B(R)} follow. Moreover, the converse is also true, that is, condition (3.17) implies (2.2). Hence, without loss of generality we assume that condition (3.17) is satisfied.
Therefore the reduced multiplication operators MQj , ∩ 2 R Q MQj := MHj dom (MHj ) L ( ,dλ, j,λ), where (MH f)(λ):=λf(λ),λ∈ R, j ∈ 2 R H ∈ 2 R H dom (MHj ):= f L ( ,dλ, j ):λf(λ) L ( ,dλ, j) . ac are unitary equivalent to the absolutely continuous parts Hj of the operators Hj, j =1, 2. With respect to the decomposition (3.16) the scattering matrix {S(λ)}λ∈R admits the decomposition Q1,λ Q1,λ S11(λ) S12(λ) S(λ)= : ⊕ −→ ⊕ (3.18) S21(λ) S22(λ) Q2,λ Q2,λ for a.e. λ ∈ R.Theentries{Sij (λ)}λ∈R, i, j =1, 2, are called channel scattering ma- trices. We note that the multiplication operators induced by the channel scattering matrices are unitary equivalent to the channel scattering operators Sij = PiSPj, i, j =1, 2, where Pi is the orthogonal projection in L onto the subspace Hj and S is the scattering operator of the complete scattering system {L, L0}. In the next proposition we give a more explicit description of the channel scattering matrices Sij (λ). The proof is an immediate consequence of Theorem 2.1, Lemma 3.2 and relations (3.14), (3.15) and (3.13).
Proposition 3.3. Let L0, V and L be given in accordance with (3.1), (3.2) and (3.3), respectively. Then the scattering matrix {S(λ)}λ∈R of the complete scattering system {L, L0} admits the representation (3.18) with entries Sij (λ) given by 1/2 ∗ 1/2 S11(λ)=IQ1,λ +2πiM1(λ) UF2(λ)U M1(λ) , 1/2 ∗ 1/2 S12(λ)=−2πiM1(λ) {U + UF2(λ)U N1(λ)U}M2(λ) , (3.19) 1/2 ∗ ∗ ∗ 1/2 S21(λ)=−2πiM2(λ) {U + U N1(λ)UF1(λ)U }M1(λ) , 1/2 ∗ 1/2 S22(λ)=IQ2,λ +2πiM2(λ) U F1(λ)UM2(λ) . for a.e. λ ∈ R. 42 J. Behrndt, H. Neidhardt and J. Rehberg
4. Dissipative channel scattering
In this section we consider the (dissipative) scattering system {H1(λ),H1} for a.e. λ ∈ R,where H1(λ)=H1 + V1(λ), (4.1) is defined for a.e. λ ∈ R,andH1 is the self-adjoint operator in H1 from (3.1). The limit V1(λ) = lim→+0 V1(λ + i ) (see Lemma 4.1) is called the optical potential of the channel H1. We recall that a linear operator T in a Hilbert space is said to be dissipative if Im (Tf,f) ≤ 0, f ∈ dom (T ), and T is called maximal dissipative if T is dissipative and does not admit a proper dissipative extension. In Theorem 4.4 below we establish a connection between the scattering matrices corresponding to the scattering systems {H1(λ),H1} and the channel scattering matrix S11(λ)from (3.18) and (3.19). −1 ∗ Lemma 4.1. Let V1(z)=−G(H2 − z) G , z ∈ C\R, be defined by (3.5) with G ∈B1(H2, H1). Then the limit V1(λ) = lim→+0 V1(λ + i ) exists in B1(H1) and V1(λ) is dissipative for a.e. λ ∈ R. Proof. Using the factorizations (3.12) of G and G∗ we find ∗ 1/2 ∗ ∗ 1/2 V1(z)=−|G | UN2(z)U |G | ,z∈ C\R, (4.2) where N2(z) is given by (3.7). According to Lemma 3.2 the limit lim→+0 N2(λ+i ) ∗ 1/2 exists in B2(H1) and since |G | ∈B2(H1) we conclude that the limit ∗ 1/2 ∗ ∗ 1/2 V1(λ) = lim V1(λ + i ) = lim −|G | UN2(λ + i )U |G | →+0 →+0 exists in B1(H1) for a.e. λ ∈ R. It is not difficult to see that Im V1(z) ≤ 0for + z ∈ C and therefore also the limit V1(λ) is dissipative for a.e. λ ∈ R.
It follows from Lemma 4.1 that for a.e. λ ∈ R the operator H1(λ)=H1+V1(λ) is maximal dissipative and therefore {H1(λ),H1} is a dissipative scattering system in the sense of [19, 20]. By Theorem 4.3 of [20] the corresponding wave operators ∗ D itH1(λ) −itH1 ac W+ (H1(λ),H1)=s − lim e e P (H1) t→+∞ and D −itH1(λ) itH1 ac W− (H1(λ),H1)=s − lim e e P (H1) t→+∞ exist and are complete which yields that {H1(λ),H1} performs a complete dissipa- tive scattering system for a.e. λ ∈ R, see [19] for details. The associated scattering operators are defined by D ∗ D SD[λ]:=W+ (H1(λ),H1) W− (H1(λ),H1) ac and act on the absolutely continuous subspaces H1 (H1). Since SD[λ]commutes with H1 the scattering operator is unitary equivalent to a multiplication operator 2 in the spectral representation L (R,dλ,Q1,µ)ofH1 induced by a family of con- tractions {SD[λ](µ)}µ∈R. The family {SD[λ](µ)}µ∈R is called the scattering matrix of the complete dissipative scattering system {H1(λ),H1}. Block Operator Matrices and Scattering 43
Using the fact that every maximal dissipative operator admits a self-adjoint dilation, i.e., there exists a self-adjoint operator in a (in general) larger Hilbert space such that its compressed resolvent coincides with the resolvents of the maxi- mal dissipative operator for all z ∈ C+, cf. [7, Section 7], see also [12], one concludes from Proposition 3.14 of [5] that the limit
F1[λ](µ) = lim F1[λ](µ + i ) →+0 exist in B1(H1) for a.e. µ ∈ R,where ∗ 1/2 −1 ∗ 1/2 F1[λ](z):=|G | (H1(λ) − z) |G | ,z∈ C+, is defined for a.e. λ ∈ R. The next proposition is a direct consequence of Theorem 2.2 of [16], see also [15].
Proposition 4.2. Let G ∈B1(H2, H1) and H1(λ) be given by (4.1). Then for a.e. λ ∈ R the scattering matrix {SD[λ](µ)}µ∈R of the complete dissipative scattering system {H1(λ),H1} admits the representation 1/2 ∗ ∗ 1/2 SD[λ](µ)=IQ1,µ +2πiM1(µ) U N2(λ)+N2(λ)U F1[λ](µ)UN2(λ) U M1(µ) for a.e. (µ, λ) ∈ R2 with respect to the Lebesgue measure in R2.
In the next lemma we show that the limit F1[λ](λ), ∗ 1/2 −1 ∗ 1/2 F1[λ](λ) = lim F1[λ](λ + i ) = lim |G | (H1(λ) − λ − i ) |G | , (4.3) →+0 →+0 exist in B2(H1) for a.e. λ ∈ R.
Lemma 4.3. Let L0, V and L be given by (3.1), (3.2) and (3.3), respectively, with G ∈B1(H2, H1).Further,letF1(λ) be as in Lemma 3.2 and let H1(λ) be defined by (4.1). Then the limit F1[λ](λ) in (4.3) exists in B2(H1) for a.e. λ ∈ R and the relation F1[λ](λ)=F1(λ) (4.4) holds for a.e. λ ∈ R. Proof. We have ∗ 1/2 −1 −1 ∗ 1/2 F1(z) − F1[λ](z)=|G | (H1(z) − z) − (H1(λ) − z) |G | (4.5) ∗ 1/2 −1 −1 ∗ 1/2 = |G | (H1(z) − z) (V1(λ) − V1(z))(H1(λ) − z) |G | . From (4.2) we obtain ∗ 1/2 ∗ ∗ 1/2 V1(λ) − V1(z)=|G | U N2(z) − N2(λ) U |G | and inserting this expression into (4.5) and using the definitions of F1(z) in (3.8) and F1[λ](z) yields ∗ F1(z) − F1[λ](z)=F1(z)U(N2(z) − N2(λ))U F1[λ](z). Hence ∗ F1(z)={IH1 + F1(z)U(N2(z) − N2(λ))U } F1[λ](z) 44 J. Behrndt, H. Neidhardt and J. Rehberg and for z = λ + i , >0 sufficiently small, the operator ∗ {IH1 + F1(z)U(N2(z) − N2(λ))U } is invertible. Therefore we conclude ∗ −1 {IH1 + F1(λ + i )U(N2(λ + i ) − N2(λ))U } F1(λ + i )=F1[λ](λ + i ).
From this representation we get the existence of F1[λ](λ)inB2(H1) and the equality (4.4) for a.e. λ ∈ R.
The next theorem is the main result of this section. We show how the channel scattering matrix S11(λ) of the scattering system {L, L0} is connected with the scattering matrices SD[λ](µ) of the dissipative scattering systems {H1(λ),H1}.
Theorem 4.4. Let {L, L0} be the scattering system from Section 3,whereL0, V and L are given by (3.1), (3.2) and (3.3), respectively, and G ∈B1(H2, H1).Fur- ther, let {Sij (λ)}, i, j =1, 2, be the corresponding scattering matrix from (3.18) and let SD[λ](µ) be the scattering matrices of the dissipative scattering systems {H1(λ),H1}. Then the scattering matrix SD[λ](λ) exists for a.e λ ∈ R and satis- fies the relation
SD[λ](λ)=S11(λ) for a.e. λ ∈ R.
Proof. From Proposition 4.2 and Lemma 4.3 we obtain that SD[λ](λ)existsfor a.e λ ∈ R and has the form 1/2 ∗ ∗ 1/2 SD[λ](µ)=IQ1,µ +2πiM1(µ) U N2(λ)+N2(λ)U F1(λ)UN2(λ) U M1(µ) (4.6) A similar calculation as in the proof of Lemma 3.1 shows ∗ F2(z)=N2(z)+N2(z)U F1(z)UN2(z),z∈ C+. If z tends to λ ∈ R,thenweget ∗ F2(λ)=N2(λ)+N2(λ)U F1(λ)UN2(λ) (4.7) for a.e. λ ∈ R. Inserting (4.7) into (4.6) we obtain
1/2 ∗ 1/2 SD[λ](λ)=IQ1,λ +2πiM1(λ) UF2(λ)U M1(λ) and by Proposition 3.3 this coincides with S11(λ) for a.e. λ ∈ R.
5. Lax-Phillips channel scattering
Similarly to Lemma 4.1 one verifies that V2(λ) = lim→+0 V2(λ + i )existsin B1(H2) for a.e. λ ∈ R. The limit V2(λ), which is called the optical potential of the channel H2, is dissipative for a.e. λ ∈ R. The optical potential defines the maximal dissipative operator H2(λ):=H2 + V2(λ) Block Operator Matrices and Scattering 45 for a.e. λ ∈ R. The operator H2(λ) decomposes for a.e. λ ∈ R into a self-adjoint part and a completely non-self-adjoint part. Let Θ2[λ](ξ), ξ ∈ C−, be the charac- teristic function, cf. [12], of the completely non-self-adjoint part of H2(λ). We are ∗ going to verify S11(λ)=Θ2[λ](λ) for a.e. λ ∈ R which shows that for a.e. λ ∈ R the scattering matrix S11(λ) can be regarded as the result of a certain Lax-Phillips scatteringtheory,cf.[1,2,3,4,14]. There is an orthogonal decomposition H Hcns ⊕ Hself 2 = 2,λ 2,λ ∈ R Hcns Hself for a.e. λ such that 2,λ and 2,λ reduce H2(λ) into a completely non-self- cns self adjoint operator H2 (λ) and a self-adjoint operator H2 (λ), cns self H2(λ)=H2 (λ) ⊕ H2 (λ). Taking into account Proposition 3.14 of [5] we get that 1/2 ∗ 1/2 m(V2(λ)) = −π|G| U M1(λ)U|G| for a.e. λ ∈ R. Let us introduce the operator ) 1/2 α(λ):= 2πM1(λ) U|G| . (5.1) Notice that clo{ran (α(λ))} = Q1,λ for a.e. λ ∈ R. With the completely non-self-adjoint part Hcns(λ) one associates the characteristic function Θ2[λ](·):Q1,λ −→ Q1,λ defined by − ∗ − −1 ∗ Θ2[λ](ξ):=IQ1,λ iα(λ)(H2(λ) ξ) α(λ) ,
ξ ∈ C−. The characteristic function is a contraction-valued holomorphic function in C−. From [12, Section V.2] we get that the boundary values
Θ2[λ](µ):=s − lim Θ2[λ](µ − i ) →+0 exist for a.e. µ ∈ R.
Theorem 5.1. Let L0, V and L be given by (3.1), (3.2) and (3.3). If the condition G ∈B1(H2, H1) is satisfied, then the limit Θ2[λ](λ),
Θ2[λ](λ):=s − lim Θ2[λ](λ − i ) →+0 exists for a.e. λ ∈ R and the relation ∗ S11(λ)=Θ2[λ](λ) holds for a.e. λ ∈ R. Proof. We set ∗ ∗ − −1 ∗ Θ2[λ](ξ):=Θ2[λ](ξ) = IQ1,λ + iα(λ)(H2(λ) ξ) α(λ)
ξ ∈ C+.Using(5.1)weget ) ) ∗ F ∗ Θ2[λ](ξ)=IQ1,λ +2πi M1(λ) U 2[λ](ξ)U M1(λ) 46 J. Behrndt, H. Neidhardt and J. Rehberg for a.e. λ ∈ R,where 1/2 −1 1/2 F2[λ](ξ):=|G| (H2(λ) − ξ) |G| ,ξ∈ C+.
Similar to the proof of Lemma 4.3 one verifies that the limit F2[λ](λ)
F2[λ](λ) = lim F2[λ](λ + i ) →+0 exist in B2(H2) for a.e. λ ∈ R and satisfies the relation F2[λ](λ)=F2(λ). Hence the ∗ ∗ limit Θ2[λ](λ)=s − lim→+0 Θ2[λ](λ − i ) exists for a.e. λ ∈ R and the relation ) ) ∗ F ∗ Θ2[λ](λ)=IQ1,λ +2πi M1(λ) U 2[λ](λ)U M1(λ) ∗ holds for a.e λ ∈ R. From (3.19) we obtain that S11(λ)=Θ2[λ](λ) for a.e. λ ∈ R. ∗ Since the limit Θ2[λ](λ) exists for a.e. λ ∈ R one concludes that
Θ2[λ](λ):=s − lim Θ2[λ](λ − i ) →+0 ∗ ∗ exists for a.e. λ ∈ R and Θ2[λ](λ) =Θ2[λ](λ) is valid. This completes the proof Theorem 5.1.
The last theorem admits an interpretation of the scattering matrix S11(λ) as the result of a Lax-Phillips scattering. Indeed, let us introduce the minimal self-adjoint dilation K2(λ) of the maximal dissipative operator H2(λ). We set
K2,λ = D−,λ ⊕ H2 ⊕ D+,λ, where 2 D±,λ := L (R±,dx,Q1,λ). Further, we define ⎛ ⎞ ⎛ ⎞ − d f− i dx f− ⎝ ⎠ ⎝ 1 ∗ ⎠ K2(λ) f := e(H2(λ))f − 2 α(λ) [f+(0) + f−(0)] d + − f i dx f+ for elements of the domain⎧⎛ ⎞ ⎫ ⎨ f− f ∈ dom (H2(λ)) ⎬ ⎝ ⎠ ∈ 1,2 R Q dom (K2(λ)) := ⎩ f : f± W ( ±,dx, 1,λ) ⎭ . f+ f+(0) − f−(0) = −iα(λ)f
The operator K2(λ) is self-adjoint and is a minimal self-adjoint dilation of the maximal dissipative operator H2(λ), that is,
−1 K2,λ −1 2 − 2 − H2 (H (λ) z) = PH2 (K (λ) z) for z ∈ C+ and K H ∈B R 2,λ =clospan EK2(λ)(∆) 2 :∆ ( ) , · D where EK2(λ)( ) is the spectral measure of K2(λ). It turns out that ±,λ are incoming and outgoing subspaces with respect to K2(λ), i.e.,
−itK2(λ) e D+,λ ⊆ D+,λ,t≥ 0, Block Operator Matrices and Scattering 47 and −itK2(λ) e D−,λ ⊆ D−,λ,t≤ 0. However, we remark that the completeness condition −itK2(λ) K2,λ =clospan e D±,λ : t ∈ R (5.2) is in general not satisfied. Condition (5.2) holds if and only if the maximal dissipa- tive operator H2(λ) is completely non-selfadjoint and H2 is singular, that means, ac the absolutely continuous part H2 of H2 is trivial. On the subspace Dλ, 2 Dλ = D−,λ ⊕ D+,λ = L (R,dx,Q1,λ) ⊆ K2,λ, let us define the operator K0(λ),
d 1,2 (K0(λ)g)(x):=−i g(x), dom (K0(λ)) := W (R,dx,Q1 ). dx ,λ The self-adjoint operator K0(λ) generates the shift group, i.e,
−itK0(λ) (e g)(x)=g(x − t),g∈ Dλ. 2 2 Using the Fourier transform F : L (R,dx,Q1,λ) −→ L (R,dµ,Q1,λ), 1 (Ff)(µ)=√ dx e−iµxf(x), 2π R the operator K0(λ) transforms into the multiplication operator on the Hilbert 2 space L (R,dµ,Q1,λ). Furthermore, one has
−itK2(λ) −itK0(λ) e D+,λ = e D+,λ,t≥ 0, and −itK2(λ) −itK0(λ) e D−,λ = e D−,λ,t≤ 0. The last properties yield the existence of the Lax-Phillips wave operators
LP itK2(λ) −itK0(λ) W± [λ]:=s − lim e J±(λ)e , t→±∞ cf. [5, 14] where J±(λ):D±,λ −→ K2,λ is the natural embedding operator. The Lax-Phillips scattering operator SLP (λ) is defined by LP ∗ LP SLP [λ]:=W+ [λ] W− [λ], 2 cf. [5, 14]. With respect to the spectral representation L (R,dµ,Q1,λ)theLax- ∗ Phillips scattering matrix {SLP [λ](µ)}µ∈R coincides with {Θ2[λ](µ) }µ∈R,see[1, 2, 3, 4]. Hence the scattering matrix {S11(λ)}λ∈R can be regarded as the result of a Lax-Phillips scattering for a.e. λ ∈ R.
Acknowledgement The support of the work by DFG, Grant 1480/2, is gratefully acknowledged. The authors thank Prof. P. Exner for discussion and literature hints. 48 J. Behrndt, H. Neidhardt and J. Rehberg
References [1] Adamjan, V.M., Arov, D.Z.: Unitary couplings of semi-unitary operators,Akad. Nauk Armjan. SSR Dokl. 43, no. 5, 257–263 (1966). [2] Adamjan, V.M., Arov, D.Z.: Unitary couplings of semi-unitary operators, Mat. Issled. 1, vyp. 2, 3–64 (1966). [3] Adamjan, V.M., Arov, D.Z.: On scattering operators and contraction semigroups in Hilbert space, Dokl. Akad. Nauk SSSR 165, 9–12 (1965). [4] Adamjan, V.M., Arov, D.Z.: On a class of scattering operators and characteristic operator-functions of contractions. Dokl. Akad. Nauk SSSR 160, 9–12 (1965). [5] Baumg¨artel, H., Wollenberg, M.: Mathematical scattering theory, Akademie-Verlag, Berlin, 1983. [6] Davies, E.B.: Two-channel Hamiltonians and the optical model of nuclear scattering, Ann. Inst. H. Poincar´e Sect. A (N.S.) 29, no. 4, 395–413 (1979). [7] Davies, E.B.: Quantum theory of open systems, Academic Press, London-New York, 1976. [8] Exner, P.: Open quantum systems and Feynman integrals, D. Reidel Publishing Co., Dordrecht, 1985. [9] Exner, P., Ulehla,´ I.: On the optical approximation in two-channel systems,J.Math. Phys. 24, no. 6, 1542–1547 (1983). [10] Feshbach, H.: A unified theory of nuclear reactions. II, Ann. Physics 19, 287–313 (1962). [11] Feshbach, H.: Unified theory of nuclear reactions, Ann. Physics 5, 357–390 (1958). [12] Foias, C., Sz.-Nagy, B.: Harmonic analysis of operators on Hilbert space,North- Holland Publishing Co., Amsterdam-London; American Elsevier Publishing Co., Inc., New York; Akad´emiai Kiad´o, Budapest 1970. [13] Kato, T.: Perturbation theory for linear operators, Die Grundlehren der mathema- tischen Wissenschaften, Band 132 Springer-Verlag New York, Inc., New York. [14] Lax, P.D., Phillips, R.S.: Scattering theory, Academic Press, New York-London, 1967. [15] Neidhardt, H.: Scattering matrix and spectral shift of the nuclear dissipative scatter- ing theory. II, J. Operator Theory 19, no. 1, 43–62 (1988). [16] Neidhardt, H.: Scattering matrix and spectral shift of the nuclear dissipative scat- tering theory, In “Operators in indefinite metric spaces, scattering theory and other topics” (Bucharest, 1985), 237–250, Oper. Theory Adv. Appl., 24, Birkh¨auser, Basel, 1987. [17] Neidhardt, H.: On the inverse problem of a dissipative scattering theory. I,In“Ad- vances in invariant subspaces and other results of operator theory” (Timi¸soara and Herculane, 1984), 223–238, Oper. Theory Adv. Appl., 17, Birkh¨auser, Basel, 1986. [18] Neidhardt, H.: Eine mathematische Streutheorie f¨ur maximal dissipative Operatoren, Report MATH, 86-3. Akademie der Wissenschaften der DDR, Institut f¨ur Mathe- matik, Berlin, 1986. [19] Neidhardt, H.: A nuclear dissipative scattering theory, J. Operator Theory 14, no. 1, 57–66 (1985). Block Operator Matrices and Scattering 49
[20] Neidhardt, H.: A dissipative scattering theory, In “Spectral theory of linear operators and related topics” (Timi¸soara/Herculane, 1983), 197–212, Oper. Theory Adv. Appl., 14, Birkh¨auser, Basel, 1984.
Jussi Behrndt Technische Universit¨at Berlin Institut f¨ur Mathematik Straße des 17. Juni 136 D–10623 Berlin, Germany e-mail: [email protected] Hagen Neidhardt Weierstraß-Institut f¨ur Angewandte Analysis und Stochastik Mohrenstr. 39 D–10117 Berlin, Germany e-mail: [email protected] Joachim Rehberg Weierstraß-Institut f¨ur Angewandte Analysis und Stochastik Mohrenstr. 39 D–10117 Berlin, Germany e-mail: [email protected] Operator Theory: Advances and Applications, Vol. 175, 51–88 c 2007 Birkh¨auser Verlag Basel/Switzerland
Asymptotic Expansions of Generalized Nevanlinna Functions and their Spectral Properties
Vladimir Derkach, Seppo Hassi and Henk de Snoo
Abstract. Asymptotic expansions of generalized Nevanlinna functions Q are investigated by means of a factorization model involving a part of the gener- alized zeros and poles of nonpositive type of the function Q. The main results in this paper arise from the explicit construction of maximal Jordan chains in the root subspace R∞(SF ) of the so-called generalized Friedrichs extension. A classification of maximal Jordan chains is introduced and studied in ana- lytical terms by establishing the connections to the appropriate asymptotic expansions. This approach results in various new analytic characterizations of the spectral properties of selfadjoint relations in Pontryagin spaces and, con- versely, translates analytic and asymptotic properties of generalized Nevan- linna functions into the spectral theoretical properties of self-adjoint relations in Pontryagin spaces. Mathematics Subject Classification (2000). Primary 46C20, 47A06, 47B50; Sec- ondary 47A10, 47A11, 47B25. Keywords. Generalized Nevanlinna function, asymptotic expansion, Pontrya- gin space, symmetric operator, selfadjoint extension, operator model, factor- ization, generalized Friedrichs extension.
1. Introduction
Let Nκ be the class of generalized Nevanlinna functions, i.e., meromorphic func- tions on C \ R with Q(¯z)=Q(z) and such that the kernel Q(z) − Q(λ) N (z,λ)= ,z,λ∈ ρ(Q),z= λ,¯ Q z − λ¯
The research was supported by the Academy of Finland (project 212150) and the Research Institute for Technology at the University of Vaasa. 52 V.A. Derkach, S. Hassi and H.S.V. de Snoo has κ negative squares on the domain of holomorphy ρ(Q)ofQ,see[20].Ifthe function Q ∈ Nκ belongs to the subclass Nκ,−2n, n ∈ N, (see [6]) then it admits the following asymptotic expansion 2-n+1 s −1 1 Q(z)=γ − j + o ,z.→∞, (1.1) zj z2n+1 j=1 where γ,sj ∈ R and z.→∞ means that z tends to ∞ nontangentially (0 <ε< arg z<π− ε). Asymptotic expansions for Q ∈ Nκ of the form (1.1) (with γ =0) were introduced in [21]. They naturally appear, for instance, in the indefinite mo- ment problem considered in [22]. The expansion (1.1) is equivalent to the following operator representation of the function Q ∈ Nκ,−2n: Q(z)=γ +[(A − z)−1ω,ω], (1.2) where ω ∈ dom An and A is a selfadjoint operator in a Pontryagin space H;see [21, Satz 1.10] and Corollary 3.4 below. The representation (1.2) can be taken to be minimal in the sense that ω is a cyclic vector for A, i.e., H = span { (A − z)−1ω : z ∈ ρ(A) }, in which case the negative index sq−(H)ofH is equal to κ. The representation (1.2) shows that ∞ is a generalized zero of the function Q(z) − γ,orequivalently, that ∞ is a generalized pole of the function Q∞(z)=−1/(Q(z) − γ). This means that the underlying symmetric operator S is nondensely defined in H with dom S = { f ∈ dom A :[f,ω]=0} (1.3) and that SF = S+({0}×span {ω}) (1.4) is a selfadjoint extensions of S in H with ∞∈σp(SF ). Here + stands for the componentwise sum in the Cartesian product H×H. In other words, the extension SF is multivalued and, in fact, can be interpreted as the generalized Friedrichs extension of S, see [5] and the references therein. It follows from (1.1) and (1.2) that s0 =[ω,ω] ∈ R.
If κ>0 then it is possible that s0 ≤ 0, in which case ∞ is a generalized pole of nonpositive type (GPNT) of the function Q∞, cf. [23]. More precisely, if ∞ is a GPNT of Q∞ with multiplicity κ∞ := κ∞(Q∞) (see (2.2) below for the definition), then in (1.1) one automatically has
s0 = ···= sj =0, for every j<2κ∞ − 2.
Furthermore, if m is the first nonnegative index in (1.1) such that sm =0(ifit exists), then, equivalently, the function Q∞ admits an asymptotic expansion of the form m+1 2+1 2+1 Q∞(z)=pm+1z + ···+ p2+1z + o z ,z.→∞, (1.5) Asymptotic expansions 53
where pm+1 =1/sm, pi ∈ R, i = m+1,...,2+1, and the integers m, n,and are connected by = m − n with m ≥ 2; see Theorem 5.4 below for further details. It turns out that (1.5) holds for some ≤ 0 if and only if ∞ is a regular critical point of SF , or equivalently, if and only if the corresponding root subspace R { ∈ H { }∈ k ∈ N } ∞(SF )= h : 0,h SF for some k of the generalized Friedrichs extensions SF in (1.4) is nondegenerate. In this case the GPNT ∞ of Q∞ as well as the corresponding root subspace R∞(SF )are shortly called regular. On the other hand, if ∞ is a singular critical point of SF , then in (1.5) >0 and, moreover, the minimal integer such that the expansion 0 (1.5) exists coincides with the dimension κ∞ of the isotropic subspace of the root subspace R∞(SF ), see Theorem 5.6. In this case the GPNT ∞ of Q∞ and the corresponding root subspace R∞(SF ) are shortly called singular with the index of 0 singularity κ∞. The above-mentioned results reflect the close connections between the asymptotic expansions (1.1), (1.5), and the root subspace R∞(SF )ofSF .The given assertions are examples of the results in the present paper which have been derived by means of the factorization model of the function Q∞ recently con- structed by the authors in [9]. This model is based on the following “proper” factorization of the function Q∞ ∈ Nκ: Q∞(z)=q(z)q (z)Q0(z), (1.6)
where q is a (monic) polynomial, q (z)=q(¯z), and Q0 ∈ Nκ such that κ∞(Q0)=0 and κ = κ − deg q, see Lemma 4.3 below. Such a factorization for Q∞ is in general not unique, but the factorization model based on such a factorization carries the complete information about the root subspace R∞(SF )ofSF . A major part of the results presented in this paper is associated with the structure of the root subspace R∞(SF )ofSF in a model space and the various connections to the asymptotic expansions (1.1) and (1.5). By using the factoriza- tion model based on a proper factorization (1.6) of Q∞ maximal Jordan chains in R∞(SF ) are constructed in explicit terms. Their construction leads to three dif- ferent types of maximal Jordan chains in R∞(SF ). Each of these three types of maximal Jordan chains admits its own characteristic features, reflecting various properties of the root subspace R∞(SF ). The construction shows explicitly, for instance, when the root subspace R∞(SF ) is regular and when it is singular. The length of the maximal Jordan chain as well as the signature of the root subspace R∞(SF ) can be easily read off from their construction. In the case that the root subspace R∞(SF ) is regular, the three types of maximal Jordan chain can be char- acterized by their length. The first type of maximal Jordan chain is of length 2k+1, where k =degq = κ∞(Q∞), and the second and third type of maximal Jordan chains are of length 2k and 2k−1, respectively. The classification of these maximal Jordan chains remains the same in the case when the root subspace R∞(SF )is 54 V.A. Derkach, S. Hassi and H.S.V. de Snoo
0 singular. In that case the index of singularity κ∞ as introduced above enters to the formulas, while the difference κ−(R∞(SF )) − κ+(R∞(SF )) of the negative and the positive index of R∞(SF ) remains unaltered, see Theorem 4.12. All of these facts can be translated into the analytical properties of the functions Q∞ and Q = γ − 1/Q∞ via the asymptotic expansions (1.1) and (1.5), and conversely. The classification of maximal Jordan chains in R∞(SF ) motivates an analo- gous classification of generalized zeros and poles of nonpositive type of the function Q ∈ Nκ, which turns out to be connected with the characterization of the mul- tiplicities of GZNT and GPNT of the function Q due to H. Langer in [24]; see Subsection 3.2 for the definitions of generalized zeros and poles of types (T1)– (T3). This induces a classification for the asymptotic expansions for the functions Q and Q∞; see Theorems 5.3 and 5.4. Some further characterizations of the three different types of generalized zeros and poles are obtained by means of the fac- torized integral representations of the functions Q and Q∞, which are based on their canonical factorizations, see [11]; for definitions, see Subsection 2.1, cf. also [5]. In particular, Theorem 6.1 and Theorem 6.3 extend some earlier results by the authors in [6] (where κ = 1) and in [8], from the regular case to the singular case 0 in an explicit manner involving the index of singularity κ∞, which is characterized in Theorem 5.6 below. The construction of the maximal Jordan chains in R∞(SF )usingthefac- torization model for Q∞ in (1.6) is carried out in Section 4. The most careful treatment of the model is required in the construction of maximal Jordan chains which are of the third type (T3). The reason is that the factorization of Q∞ does not produce a minimal model for the function Q∞ directly. In the minimal fac- torization model the maximal Jordan chains of type (T3) are roughly speaking the shortest ones, cf. (4.21), (4.24), (4.28); see also Theorem 5.3 and Theorem 6.3. The results in Lemma 4.11 and part (iii) of Theorem 4.12 characterize maximal Jordan chains of type (T3). In this case the underlying symmetric relation S(Q)is multivalued (before the auxiliary part of the space is factored out). This statement is true more generally: for an arbitrary Nκ-function Q the occurrence of general- ized zeros and poles of type (T3) in R ∪{∞}is an indication that point spectrum σp(S(Q)) of S(Q) is nonempty, see Lemma 6.4 below, which by part (i) of Theo- rem 4.6 is equivalent to S(Q) being not simple. In fact, the existence of maximal Jordan chains of type (T3) or, equivalently, the existence of GZNT and GPNT of type (T3) can be used to give criteria for minimality of various factorization models for Nκ-functions, see Propositions 6.6 and 6.7 below. The topics considered in this paper have connections to some other recent studies involving asymptotic expansions of Nκ-functions, see in particular [6], [8], [9], [12], [13], [15], and their canonical factorization, see, e.g., [3], [5], [7], [11], [14]. For instance, in [13] the authors investigate the subclass of Nκ-functions with κ = κ∞(Q) and extend some results, e.g., from [6], [8]. General operator models based on the canonical factorization of Nκ-functions have been introduced in [3]; for another model not using the canonical factorization of Q, see [18]. The construction of a minimal canonical factorization model by using reproducing kernel Pontryagin Asymptotic expansions 55 space methods has been recently worked out in [12], cf. also [3, Theorem 4.1]. Some of the results in the present paper can be naturally augmented by the results which can be found from [15], where characteristic properties of the generalized zeros and poles of Nκ-functions have been studied with the aid of their operator representations. The present paper forms a continuation of the paper [9], where the details concerning the construction of the announced factorization model can be found. Some basic definitions and concepts which will be used throughout the paper are given in Section 2. In Section 3 some additions concerning the subclasses Nκ,− as introduced in [6] are given, including a proof for [6, Proposition 6.2] as announced in that paper, cf. Theorem 3.3 below; see also Theorem 5.4 for an extension of these results. Asymptotic expansions are introduced in Section 3 and a classification of generalized zeros and poles is given. In Section 4 the main ingredients concerning the factorization model are given and the construction of maximal Jordan chains in R∞(SF ) is carried out. The connection between the properties of the root subspace R∞(SF ) and the asymptotic expansions of the form (1.1) and (1.5) is investigated in Section 5. Finally, in Section 6 the classification of GZNT and GPNT is connected with factorized integral representations of the functions Q and Q∞(z). In this section also the generalized zeros and poles of nonpositive type of Nκ-functions which belong to R are briefly treated and some consequences as announced above are established.
2. Preliminaries
2.1. Canonical factorization of Q ∈ Nκ The notions of generalized poles and generalized zeros of nonpositive type were introduced in [23]. The following definitions are based on [24]. A point α ∈ R is called a generalized pole of nonpositive type (GPNT) of the function Q ∈ Nκ with multiplicity κα (= κα(Q)) if −∞ < lim (z − α)2κα+1Q(z) ≤ 0, 0 < lim (z − α)2κα−1Q(z) ≤∞. (2.1) z.→α z.→α Similarly, the point ∞ is called a generalized pole of nonpositive type (GPNT) of Q with multiplicity κ∞ (= κ∞(Q)) if
≤ Q(z) ∞ −∞ ≤ Q(z) 0 lim 2 +1 < , lim 2 −1 < 0. (2.2) z.→∞ z κ∞ z.→∞ z κ∞ Apointβ ∈ R is called a generalized zero of nonpositive type (GZNT) of the function Q ∈ Nκ if β is a generalized pole of nonpositive type of the function −1/Q. The multiplicity πβ (= πβ(Q)) of the GZNT β of Q can be characterized by the inequalities:
Q(z) ≤∞ −∞ Q(z) ≤ 0 < lim 2 +1 , < lim 2 −1 0. (2.3) z.→β (z − β) πβ z.→β (z − β) πβ 56 V.A. Derkach, S. Hassi and H.S.V. de Snoo
Similarly, the point ∞ is called a generalized zero of nonpositive type (GZNT) of Q with multiplicity π∞ (= π∞(Q)) if −∞ ≤ lim z2π∞+1Q(z) < 0, 0 ≤ lim z2π∞−1Q(z) < ∞. (2.4) z.→∞ z.→∞
It was shown in [23] that for Q ∈ Nκ the total number (counting multiplic- ities) of poles (zeros) in C+ and generalized poles (zeros) of nonpositive type in R ∪{∞} is equal to κ.Letα1,...,αl (β1,...,βm) be all the generalized poles (zeros) of nonpositive type in R and the poles (zeros) in C+ with multiplicities κ1,...,κl (π1,...,πm). Then the function Q admits a canonical factorization of the form p Q(z)=r(z)r (z)Q00(z),Q00 ∈ N0,r= , (2.5) q / / m − πj l − κj where p(z)= j=1(z βj) and q(z)= j=1(z αj ) are relatively prime polynomials of degree κ−π∞(Q)andκ−κ∞(Q), respectively; see [11], [5]. It follows from (2.5) that the function Q admits the (factorized) integral representation 1 t p − Q(z)=r(z)r (z) a + bz + 2 dρ(t) ,r= , (2.6) R t − z 1+t q where a ∈ R, b ≥ 0, and ρ(t) is a nondecreasing function satisfying the integrability condition dρ(t) ∞ 2 < . (2.7) R t +1
2.2. The subclasses Nκ,1 and Nκ,0
A function Q ∈ Nκ is said to belong to the subclass Nκ,1,if Q (z) ∞ |Im Q (iy) | lim =0and dy < ∞, .→∞ z z η y with η>0 large enough. Similarly Q ∈ Nκ is said to belong to the subclass Nκ,0,if Q(z) lim = 0 and lim sup |z Im Q(z)| < ∞, z.→∞ z z.→∞ see [5]. In the following theorems the subclasses Nκ,1 and Nκ,0 are characterized both in terms of the integral representation (2.6) and in terms of operator repre- sentations of the form (1.2). Let Et be a spectral function of a selfadjoint operator A in a Pontryagin space H0, see [1]. Denote by H := H(A), ∈ N,thesetofall ∈ H | | ∞ ±∞ elements h such that ∆ t d[Eth, h] < for some neighborhood ∆ of . Moreover, let H−(A), ∈ N, be the corresponding dual spaces. Here, for instance, H−1(A) can be identified as the set of all generalized0 elements obtained by com- −1 pleting H with respect to the inner product ∆(1 + |t|) d[Eth, h] < ∞ with some neighborhood ∆ of ±∞. The operator A admits a natural continuation A from H into H−1, see [5] for further details. The classes Nκ,1 and Nκ,0 are characterized in the following two theorems, see [5]. Asymptotic expansions 57
Theorem 2.1. ([5]) For Q ∈ Nκ the following statements are equivalent:
(i) Q belongs to Nκ,1; (ii) Q(z)=γ +[(A − z)−1ω,ω], z ∈ ρ(A), for some selfadjoint operator A in a Pontryagin space H, a cyclic vector ω ∈ H−1,andγ ∈ R; (iii) Q has the integral representation (2.6) with deg q − deg p = π∞(Q) > 0,or with deg p =degq (π∞(Q)=0), b =0,and (1 + |t|)−1dρ(t) < ∞. (2.8) R
Theorem 2.2. ([5]) For Q ∈ Nκ the following statements are equivalent:
(i) Q belongs to Nκ,0; (ii) Q (z)=γ + O (1/z), z.→∞; (iii) Q(z)=γ +[(A − z)−1ω,ω], z ∈ ρ(A), for some selfadjoint operator A in a Pontryagin space H, a cyclic vector ω ∈ H,andγ ∈ R; (iv) Q has the integral representation (2.6) with deg q − deg p = π∞(Q) > 0,or with deg p =degq (π∞(Q)=0), b =0,and dρ(t) < ∞. (2.9) R
Remark 2.3. If Q ∈ Nκ,0, then the operator representation of Q in part (iii) of Theorem 2.2 implies that lim −z(Q(z) − γ)=[ω,ω]. z.→∞ Hence, the statement (ii) in Theorem 2.2 can be strengthened in the sense that for every function Q ∈ N 0 there are real numbers γ and s0, such that κ, s0 1 Q (z)=γ − + o ,z.→∞. (2.10) z z
3. Asymptotic expansions of generalized Nevanlinna functions Asymptotic expansions of generalized Nevanlinna functions (as in (2.10)) can be used for studying operator and spectral theoretical properties of selfadjoint exten- sions of symmetric operators in Pontryagin and Hilbert spaces, see [20], [16]. In this section a subdivision of the class Nκ of generalized Nevanlinna functions is given along the lines of [16], [6]. Moreover, a classification for generalized zeros of nonpositive type is introduced and interpreted via asymptotic expansions.
3.1. The subclasses Nκ,− of generalized Nevanlinna functions
Definition 3.1. A function Q ∈ Nκ is said to belong to the subclass Nκ,−2n, n ∈ N, if there are real numbers γ and s0,...,s2 −1 such that the function ⎛ n ⎞ -2n s −1 Q(z)=z2n ⎝Q(z) − γ + j ⎠ (3.1) zj j=1 58 V.A. Derkach, S. Hassi and H.S.V. de Snoo
is O(1/z)asz.→∞.Moreover,Q ∈ Nκ is said to belong to the subclass Nκ,−2n+1 if the function Q in (3.1) belongs to Nκ ,1 for some κ ∈ N.
The next lemma clarifies the above definition of the subclasses Nκ,−, ∈ N.
Lemma 3.2. If the function Q belongs to the subclass Nκ,−2n (Nκ,−2n+1) for some n ∈ N, then the function Q in (3.1) belongs to the subclass Nκ ,0 (resp. Nκ ,1) with κ ≤ κ. Moreover, the following inclusions are satisfied
···⊂Nκ,−2n−1 ⊂ Nκ,−2n ⊂ Nκ,−2n+1 ⊂···⊂Nκ,0 ⊂ Nκ,1. (3.2)
Proof. Rewrite the expression for the function Q in (3.1) in the form Q(z)= z2nQ(z). Then Q(z) as a sum of two generalized Nevanlinna functions is also a generalized Nevanlinna function, and therefore, in view of (2.5), Q is a generalized Nevanlinna function, too. Next it is shown that the inequality κ(Q) ≤ κ(Q)is satisfied. First observe that the condition Q(z)=O(1) and hence, in particular, the condition Q(z)=O(1/z)asz.→∞ implies that κ∞(Q)=0, (3.3) cf. (2.2), (2.4). Clearly, κα(Q)=κα(Q) for every α =0 , ∞, while for α =0one derives from (2.1) the estimate κ0(Q) ≤ κ0(Q). (3.4)
Therefore, one can conclude from (3.3) and (3.4) that κ(Q) ≤ κ(Q). Now by Theorem 2.2 the condition Q(z)=O(1/z), z.→∞, is equivalent to Q ∈ Nκ ,0 with κ ≤ κ, which proves the first statement for the subclasses Nκ,−2n.IfQ ∈ Nκ,−2n+1,thenQ(z)=O(1) and since κ(Q) ≤ κ(Q), one actually has Q ∈ Nκ ,1 for κ ≤ κ. Since Nκ,0 ⊂ Nκ,1 the inclusions Nκ,−2n ⊂ Nκ,−2n+1, n ∈ N, follow from thefirstpartofthelemma.NowletQ ∈ Nκ,−2n−1. Then by definition 2 z Q(z)+zs2n + s2n+1 ∈ Nκ ,1, (3.5) where Q is as in (3.1) and κ ≤ κ. It is clear from (3.5) (see Theorem 2.1) that Q(z)=O(1/z)asz.→∞. Hence, Q ∈ Nκ,−2n and this proves the remaining inclusions in (3.2).
The subclasses Nκ,−, ∈ N, are now characterized by means of the operator and the integral representation of Q in (1.2) and (2.6), respectively.
Theorem 3.3. For Q ∈ Nκ the following statements are equivalent:
(i) Q ∈ Nκ,−, ∈ N; (ii) Q(z)=γ +[(A − z)−1ω,ω], z ∈ ρ(A), for some selfadjoint operator A in a Pontryagin space H, a cyclic vector ω ∈ H,andγ ∈ R; Asymptotic expansions 59
(iii) Q has an integral representation (2.6) with π∞(Q)=degq− deg p ≥ 0 (and b =0if π∞(Q)=0), such that (1 + |t|)−2π∞ dρ(t) < ∞. (3.6) R
Proof. (i) ⇒ (iii) Let Q ∈ Nκ,−,where is either 2n or 2n − 1, n ∈ N.Inviewof (3.2) and Theorems 2.1, 2.2 one has π∞(Q)=degq− deg p ≥ 0, and if π∞(Q)=0 then b = 0 and (2.8) or (2.9) is satisfied. By Lemma 3.2 the function Q in (3.1) belongs to Nκ ,2n− with κ ≤ κ. Hence, Q admits the factorization 1 t p2 − Q = r(z)r (z) a + bz + 2 dρ(t), r = , (3.7) R t − z 1+t q2 where p2 and q2 are the polynomials associated to Q, cf. (2.5). Moreover, the inequality π∞(Q)=degq2 − deg p2 ≥ 0 holds by Theorems 2.1 and 2.2. On the other hand, it follows from (2.6) and (3.1) that Q admits also the representation 1 t 2n − Q(z)=z r(z)r (z) a + bz + 2 dρ(t)+p1(z), (3.8) R t − z 1+t where p1 is a polynomial with deg p1 ≤ 2n. An application of the generalized Stieltjes inversion formula (see [19]) shows that the measures dρ(t) in (3.7) and dρ(t) in (3.8) are connected by |r(t)|2dρ(t)=t2n|r(t)|2 dρ(t). (3.9) Therefore, if Q ∈ Nκ ,1 \ Nκ ,0 so that =2n − 1, then deg p2 =degq2 and dρ(t) satisfies the condition (2.8) in Theorem 2.1. The condition (3.6) follows now from (3.9). If Q ∈ Nκ ,0 so that =2n, then either deg p2 =degq2 in which case dρ(t) satisfies the condition (2.9) in Theorem 2.2, or π∞(Q)=degq2 − deg p2 > 0in which case dρ(t) satisfies the condition (2.7). In both cases |r(t)|2dρ(t) < ∞ for M>0 large enough. |t|>M Hence, again the condition (3.6) follows from (3.9).
(ii) ⇔ (iii) Let Et be the spectral function of a selfadjoint operator A in the minimal representation (1.2) of Q. It follows from (1.2), (2.6), and the generalized Stieltjes inversion formula that 2 d[Etω,ω]=|r(t)| dρ(t),t∈ ∆, in some neighborhood ∆ of ±∞. This implies that
−2π∞ (1 + |t|) dρ(t) < ∞ if and only if (1 + |t|) d[Etω,ω] < ∞, R ∆ i.e., ω ∈ H, which proves the equivalence of (ii) and (iii). 60 V.A. Derkach, S. Hassi and H.S.V. de Snoo
n (ii) ⇒ (i) First consider the case =2n.Thenω ∈ H means that ω ∈ dom A . Define the function Q in (3.1) by setting j j n sj =[A ω,ω],sn+j =[A ω,A ω],j=0,...,n. (3.10) Then a straightforward calculation shows that Q admits the operator representa- tion Q(z)=[(A − z)−1ω ,ω ],ω = Anω ∈ H. (3.11) Therefore, Q(z)=O(1/z)andQ ∈ Nκ,−. n Now let =2n − 1. Then ω ∈ H means that ω := A ω ∈ H−1. Hence s2n−1 := [Aω ,ω ] is well defined. Moreover, by defining s0,...,s2n−2 as in (3.10) it follows that the function Q in (3.1) admits the operator representation −1 n Q(z)=[(A − z) ω ,ω ],ω= A ω ∈ H−1. (3.12) Hence, by Theorem 2.1 Q ∈ Nκ ,1 for some κ ∈ N and thus Q ∈ Nκ,−.This completes the proof. Inthecaseofevenindices =2n the equivalence of (i) and (ii) in Theorem 3.3 coincides with the following result of M.G. Kre˘ın and H. Langer, see [21, Satz 1.10].
Corollary 3.4. ([21]) The function Q ∈ Nκ admits an operator representation −1 n Q(z)=γ +[(A − z) ω,ω] with γ ∈ R and ω ∈ H2n (= dom A ) if and only if there are real numbers γ and s0,...,s2n, such that 2-n+1 s −1 1 Q(z)=γ − j + o ,z.→∞. (3.13) zj z2n+1 j=1
In this case the numbers s0,...,s2n are given by (3.10). Proof. The proof of Theorem 3.3 shows that the condition ω ∈ dom An is equiv- alent to the operator representation (3.11) of the function Q(z) in (3.1). Now by applying (2.10) in Remark 2.3 to the function Q(z) in (3.11) and taking into account (3.1) the equivalence to the expansion (3.13) follows. The criterion of M.G. Kre˘ın and H. Langer formulated in Corollary 3.4 does not hold in the case of an odd index =2n − 1. However, it is clear that if ω ∈ H2n−1 then the analog of the expansion (3.13) exists.
Corollary 3.5. If the function Q ∈ Nκ admits an operator representation Q(z)= −1 n−1/2 γ +[(A − z) ω,ω] with γ ∈ R and ω ∈ H2n−1 (= dom |A| ) then there are real numbers γ and s0,...,s2n, such that -2n s −1 1 Q(z)=γ − j + o ,z.→∞. (3.14) zj z2n j=1
Proof. Since ω ∈ H2n−1 the operator representation (3.12) in the proof of Theo- rem 3.3 shows that Q(z)=o(1). The expansion (3.14) for the function Q follows now from the definition of Q in (3.1). Asymptotic expansions 61
It is emphasized that the existence of the expansion (3.14) does not imply n that ω ∈ H2n−1 or, equivalently, that ω := A ω belongs to H−1.Inthiscase [Aω ,ω ] need not be defined and hence it cannot coincide with the coefficient s2n−1 in (3.14). 3.2. A classification of generalized zeros of nonpositive type For what follows it will be useful to give a classification for generalized zeros and poles of nonpositive type of a function Q ∈ Nκ. Let ∞ be a GZNT of Q with multiplicity π∞ > 0. It follows from (2.4) that precisely one of the following three cases can occur: − 2π∞+1 2π∞−1 (T1) s2π∞ := limz.→∞ z Q(z) < 0, limz.→∞ z Q(z)=0; 2π∞+1 2π∞−1 (T2) limz.→∞ z Q(z)=∞, limz.→∞ z Q(z)=0; 2π∞+1 ∞ − 2π∞−1 (T3) limz.→∞ z Q(z)= , s2π∞−2 := limz.→∞ z Q(z) > 0. In these cases ∞ is said to be a generalized zero of type (T1), (T2), or (T3), respec- tively; the shorter notations GZNT1, GZNT2, and GZNT3 are used accordingly. The corresponding classification for a finite generalized zero β ∈ R of Q is defined analogously: Q(z) Q(z) (T1) limz.→β > 0, limz.→β =0; (z − β)2πβ +1 (z − β)2πβ −1 Q(z) Q(z) (T2) limz.→β = ∞, limz.→β =0; (z − β)2πβ +1 (z − β)2πβ −1 Q(z) Q(z) (T3) limz.→β = ∞, limz.→β < 0. (z − β)2πβ +1 (z − β)2πβ −1 A generalized pole of nonpositive type β ∈ R ∪{∞}of Q is said to be of type (T1), (T2), or (T3), if β is a generalized zero of nonpositive type of the function −1/Q which is of type (T1), (T2), or (T3), respectively. To give some immediate implications of the above classification consider the generalized zero ∞ of Q. If it is of the first type, then it follows from (T1) that ∈ N Q κ,−2π∞ .Moreover,Q has the following asymptotic expansion: s2 1 − π∞ .→∞ Q(z)= + o ,z ,s2π∞ > 0. (3.15) z2π∞+1 z2π∞+1 ∞ ∈ N If the generalized zero of Q is of type (T3), then Q κ,−2π∞−2 and Q has the following asymptotic expansion s2 −2 1 − π∞ .→∞ Q(z)= + o ,z ,s2π∞−2 < 0. (3.16) z2π∞−1 z2π∞−1 In the case that the generalized zero ∞ is of type (T2) there are two possibilities: N either Q belongs to κ,−2π∞ , in which case both of the moments s2π∞−1 and s2π∞ are finite and Q has the asymptotic expansion s2 −1 s2 1 − π∞ − π∞ .→∞ Q(z)= + o ,z ,s2π∞−1 =0, (3.17) z2π∞ z2π∞+1 z2π∞+1 62 V.A. Derkach, S. Hassi and H.S.V. de Snoo
N \ N or Q belongs to κ,−2(π∞−1) κ,−2π∞ and it has the asymptotic expansion s2 −1 1 Q(z)=− π∞ + o ,z.→∞, (3.18) z2π∞ z2π∞ or 1 Q(z)=o ,z.→∞. (3.19) z2π∞−1 Observe, that the expansions (3.17) and (3.18) are also special cases of the ex- pansion (3.19). Hence, if ∞ is a generalized zero of type (T2), then Q has an expansion of the form (3.16), but now with s2π∞−2 = 0; however, Q does not have an expansion of the form (3.15). Similar observations remain true for generalized zeros β ∈ R and poles α ∈ R ∪{∞}. For instance, to get the analogous expansions for a generalized zero β ∈ R apply the transform −Q(1/(z − β)) to the expansions in (3.15)–(3.19); cf. also [15]. The role of the above classification for generalized zeros and poles of nonpositive type will be described in detail in Sections 4–6.
4. An operator model for the generalized Friedrichs extension 4.1. Boundary triplets and Weyl functions The construction of the model uses the notion of a boundary triplet in a Pontryagin space setting. Let H be a Pontryagin space with negative index κ,letS be a closed symmetric relation in H with defect numbers (n, n), and let S∗ be the adjoint of S. n ∗ AtripletΠ={C , Γ0, Γ1 } is said to be a boundary triplet for S , if the following two conditions are satisfied: ∗ n n (i) the mapping Γ : f →{Γ0f,Γ1f} from S to C ⊕ C is surjective; (ii) the abstract Green’s identity [f ,g] − [f,g ]=(Γ1f,Γ0g) − (Γ0f,Γ1g) (4.1) holds for all f = {f,f }, g = {g,g }∈S∗, see, e.g., [2], [10]. It is easily seen that A0 =kerΓ0 and A1 =kerΓ1 are selfadjoint extensions of S. Associated to every boundary triplet there is the Weyl function Q defined by Q(z)Γ0fz =Γ1fz,z∈ ρ(A0), ∗ where fz := {fz,zfz}∈Nz,andNz =ker(S − z) denotes the defect subspace of S at z ∈ C. It follows from (4.1) that the Weyl function Q is also a Q-function of the pair {S, A0} in the sense of Kre˘ın and Langer, see [20]. If S is simple,sothat
H = span {Nz : z ∈ ρ(A0)}, then the Weyl function Q belongs to the class Nκ,otherwiseQ ∈ Nκ with κ ≤ κ. Moreover, if S is simple and H is a selfadjoint extension of S in H, then the point spectrum of H is also simple, that is, every eigenspace of H is one-dimensional, and if α ∈ R ∪{∞}, then the root subspace at α is at most 2κ + 1-dimensional. Asymptotic expansions 63
In the case where S is given by (1.3) one can define a boundary triplet for S∗ as follows. Proposition 4.1. (cf. [5]) Let A be a selfadjoint operator in a Pontryagin space H and let the restriction S of A be defined by (1.3) with ω ∈ H. Then the adjoint S∗ of S in H is of the form S∗ = {{f,Af + cω} : f ∈ dom A, c ∈ C } ∞ ∞ ∞ ∗ and a boundary triplet Π = {C, Γ0 , Γ1 } for S is determined by ∞ ∞ ∗ Γ0 f =[f,ω], Γ1 f = c, f = {f,Af + cω}∈S .
The corresponding Weyl function Q∞ is given by 1 Q∞(z)=− ,z∈ ρ(A). (4.2) [(A − z)−1ω,ω]
4.2. The model operator S(Q∞) corresponding to a proper factorization Operator models for generalized Nevanlinna functions whose only generalized pole of nonpositive type is at ∞ have been constructed in [6] and [14]. Such functions admit a canonical factorization of the form Q∞(z)=q(z)q (z)Q0(z), (4.3) k k−1 where Q0 ∈ N0, q(z)=z + qk−1z + ···+ q0 is a polynomial, and q (z)= q(¯z). In general, models which are based on the canonical factorization of Q ∈ Nκ are not necessarily minimal, i.e., the underlying model operator S(Q∞) need not be simple and it can even be a symmetric relation (multivalued operator). However, with the canonical factorization the nonsimple part of S(Q∞)canbe easily identified and factored out to produce a simple symmetric operator from S(Q∞), cf. [3]. The model constructed for S(Q∞) in [6] uses an orthogonal coupling of two symmetric operators. In [9] this model was adapted to the case where the function Q0 is allowed to be a generalized Nevanlinna function, too. In this case the situation becomes more involved and, in general, one cannot represent S(Q∞) as an orthogonal sum of a simple symmetric operator and a selfadjoint relation. However, such a simple orthogonal decomposition for S(Q∞) can still be obtained if the factorization (4.3) of Q∞ is proper. This concept is defined as follows.
Definition 4.2. ([9]) The factorization Q∞(z)=q(z)q (z)Q0(z)issaidtobeproper if q is a divisor of degree κ∞(Q∞) > 0 of the polynomial q in the canonical factorization of the function Q∞,cf.(2.5).
Clearly, proper factorizations of Q∞ ∈ Nκ always exist, but they are not unique if q has more than one zero and κ∞(Q∞) <κ. Proper factorizations Q∞ = qq Q0 can be characterized also without using the canonical factorization of Q∞.
Lemma 4.3. Let Q∞ ∈ Nκ have a factorization of the form Q∞(z)=q(z)q (z)Q0(z), deg q = k ≥ 1, (4.4) 64 V.A. Derkach, S. Hassi and H.S.V. de Snoo where q(z) is a monic polynomial, and let α ∈ σ(q) beazeroofq with multiplicity kα. Then the following statements are equivalent:
(i) the factorization (4.4) of Q∞ is proper; (ii) the multiplicities κ∞(Q∞) and πα(Q∞) satisfy the following relations:
κ∞(Q∞)=degq and πα(Q∞) ≥ kα for all α ∈ σ(q); (4.5)
(iii) κ∞(Q0) and κ(Q∞)=κ satisfy the following identities
κ∞(Q0)=0and κ(Q∞)=degq + κ(Q0). (4.6)
Proof. (i) ⇔ (ii) In a proper factorization (4.4) κ∞(Q∞)=degq and clearly the inequalities in (4.5) just mean that q divides the polynomial q in the canonical factorization of Q∞. (i) ⇒ (iii) If the factorization (4.4) is proper, then in the canonical factor- ization of the function Q0 the numerator q0 and denominator p0 (= p)ofthe corresponding rational factor r0 are of the same degree κ(Q0), and this implies (4.6). (iii) ⇒ (i) It follows from the second equality in (4.6) that q and the poly- nomial p0 in the canonical factorization of the function Q0 are relatively prime and, therefore, q is a factor of the polynomial q in the canonical factorization of Q∞.Moreover,π∞(Q0) = 0. Now the assumption κ∞(Q0) = 0 implies that κ∞(Q∞)=degq.
The construction of factorization models is now briefly described. Let q be a polynomial as in (4.4) of degree k =degq. Define the k × k matrices Bq and Cq by ⎛ ⎞ ⎛ ⎞ q1 ... qk−1 1 01... 0 ⎜ ⎟ ⎜ . . ⎟ ⎜ . . . ⎟ ⎜ . .. 10⎟ ⎜ . .. .. 0 ⎟ Bq = ⎜ ⎟ , Cq = ⎜ ⎟ , ⎝ . . .⎠ ⎝ 00... 1 ⎠ q −1 .. .. . k − − − 10... 0 q0 q1 ... qk−1 so that σ(Cq)=σ(q). Moreover, let Hq be a 2k-dimensional Pontryagin space defined by 0 B (Ck ⊕ Ck, B·, ·), B = q . Bq 0
A general factorization model for functions Q∞ of the form (4.4) was constructed in [9] and can be applied, in particular, for proper factorizations of Q∞.
Theorem 4.4. (cf. [9]) Let Q∞ ∈ Nκ be a generalized Nevanlinna function and let Q∞(λ)=q(λ)q (λ)Q0(λ), (4.7) be a proper factorization of Q∞,whereq is a monic polynomial of degree k = deg q ≥ 1.LetS0 be a closed symmetric relation in a Pontryagin space H0 with 0 0 0 the boundary triplet Π = {C, Γ0, Γ1} whose Weyl function is Q0. Asymptotic expansions 65
Then:
(i) the function Q0 in (4.7) belongs to the class Nκ−k; (ii) the linear relation ⎧ ⎧⎛ ⎞ ⎛ ⎞⎫ ⎫ ⎪ ∗ ⎪ ⎨ ⎨ f0 f0 ⎬ f0 = {f0,f0}∈S0 , ⎬ 0 ⎝ f ⎠ ⎝ C f ⎠ S(Q∞)=⎪ ⎩ , q ⎭ : f1 =Γ1f0, ⎪ (4.8) ⎩ 0 ⎭ f Cqf +Γ0f0ek f1 =0
is closed and symmetric in H := H0 ⊕ Hq and has defect numbers (1, 1); ∗ (iii) the adjoint S(Q∞) of S(Q∞) is given by ⎧ ⎧⎛ ⎞ ⎛ ⎞⎫ ⎫ ∗ ⎨ ⎨ f0 f0 ⎬ f0 = {f0,f0}∈S0 , ⎬ ∗ ⎝ ⎠ ⎝ ⎠ 0 S(Q∞) = f , Cq f + ϕek : f1 =Γ1f0, ; ⎩ ⎩ 0 ⎭ ⎭ f Cqf +Γ0f0ek ϕ ∈ C ∗ (iv) a boundary triplet Π={C, Γ0, Γ1} for S(Q∞) is determined by ∗ Γ0(f0 ⊕ F )=f1, Γ1(f0 ⊕ F )=ϕ, f0 ⊕ F ∈ S(Q∞) ; (4.9)
(v) the corresponding Weyl function coincides with Q∞. Proof. Since the factorization (4.7) is proper the statement (i) is immediate from the equality (4.6) in Lemma 4.3. All the other statements are contained in [9].
In fact, the statement (iv) in Theorem 4.4 can be obtained directly also from Proposition 4.1, since S(Q∞) in (4.8) is a restriction of the selfadjoint relation ⎧ ⎧⎛ ⎞ ⎛ ⎞⎫ ⎫ ⎨ ⎨ f0 f0 ⎬ ∗ ⎬ f0 = {f0,f0}∈S0 , ⎝ ⎠ ⎝ ⎠ A(Q∞)= f , Cq f : 0 (4.10) ⎩ ⎩ 0 ⎭ f1 =Γ1f0 ⎭ f Cqf +Γ0f0ek to the subspace H ω0,whereω0 =col(0,ek, 0); compare (1.3). The generalized Friedrichs extension of S(Q∞)isgivenby ⎧ ⎧⎛ ⎞ ⎛ ⎞⎫ ⎫ ⎪ ∗ ⎪ ⎨ ⎨ f0 f0 ⎬ f0 = {f0,f0}∈S0 , ⎬ 0 ⎝ f ⎠ ⎝ C f + ϕe ⎠ SF (Q∞)=⎪ ⎩ , q k ⎭ : f1 =Γ1f0, ⎪ . (4.11) ⎩ 0 ⎭ f Cqf +Γ0f0ek f1 =0, ϕ ∈ C
According to (4.2) in Proposition 4.1 the Weyl function Q∞(z) corresponding to the boundary triplet (4.9) is of the form
− 1 Q∞(z)= −1 . [(A(Q∞) − z) ω0,ω0]
Thus the function Q = −1/Q∞ has the representation −1 Q(z)=[(A(Q∞) − z) ω,ω], (4.12) which is, however, not necessarily minimal, since mul S and ker (S − α), α ∈ σ(q), can be nontrivial. The following lemma describes these subspaces. 66 V.A. Derkach, S. Hassi and H.S.V. de Snoo
Lemma 4.5. ([9]) Under the assumptions of Theorem 4.4 let S0 be a simple closed H 0 0 ⊃ symmetric operator in the Pontryagin space 0 and let Ai =kerΓi ( S0), i =0, 1. Then: 0 (i) mul S(Q∞) is nontrivial if and only if mul A1 is nontrivial and in this case
0 0 mul S(Q∞)={ (g,0, Γ0ge k) : g = {0,g}∈A1 }; (4.13)
(ii) if mul S(Q∞) is nontrivial, then it is spanned by a positive vector; 0 (iii) if mul A0 is nontrivial, then it is spanned by a positive vector; 0 0 (iv) σp(S(Q∞)) = σp(A0) ∩ σ(q ) and for α ∈ σp(A0) ∩ σ(q ) one has
0 0 ker (S(Q∞) − α)={ (g0, Γ1g0Λ|λ=α, 0) : g0 ∈ ker (A0 − α) }, (4.14)
where Λ=(1,λ,...,λk−1), λ ∈ C; 0 (v) if ker (S(Q∞) − α) or, equivalently, ker (A0 − α) is nontrivial, then it is spanned by a positive vector.
It follows from (ii) and (iv) that the linear relation S(Q∞) can be decomposed into a direct sum of an operator S with an empty point spectrum and a selfadjoint part in a Hilbert space which is the sum of mul S(Q∞)andker(S(Q∞) − α), 0 α ∈ σp(A0) ∩ σ(q ). The next theorem shows that the reduced operator S is simple.
Theorem 4.6. Let the assumptions of Theorem 4.4 and Lemma 4.5 be satisfied and let S(Q∞), A(Q∞),andSF (Q∞) be given by (4.8), (4.10),and(4.11), respectively. Then:
(i) S(Q∞) is simple if and only if σp(S(Q∞)) = ∅. In this case the linear rela- tions S = S(Q∞), A = A(Q∞),andSF = SF (Q∞) satisfy the equalities (1.3) and (1.4) with ω = ω0 and the operator representation (4.12) of Q = −1/Q∞ is minimal. (ii) If S(Q∞) is not simple, then the subspace