Spectral Methods for Operators of Mathematical Physics-Birkhäuser B - (Operator Theory_ Advances and Applications 154) Vadim Adamyan, Heinz Langer (Auth.), Jan Janas, Pavel Kurasov, Sergei Naboko (Eds.)

Operator Theory: Advances and
Applications
Vol. 154
Editor:
I. Gohberg
Editorial Office:
School of Mathematical H. G. Kaper (Argonne)
Sciences S. T. Kuroda (Tokyo)
Tel Aviv University P. Lancaster (Calgary)
Ramat Aviv, Israel L. E. Lerer (Haifa)
B. Mityagin (Columbus)
Editorial Board: V. V. Peller (Manhattan, Kansas)
D. Alpay (Beer-Sheva) L. Rodman (Williamsburg)
J. Arazy (Haifa) J. Rovnyak (Charlottesville)
A. Atzmon (Tel Aviv) D. E. Sarason (Berkeley)
J. A. Ball (Blacksburg) I. M. Spitkovsky (Williamsburg)
A. Ben-Artzi (Tel Aviv) S. Treil (Providence)
H. Bercovici (Bloomington) H. Upmeier (Marburg)
A. Bottcher (Chemnitz) S. M. Verduyn Lunel (Lei den)
K. Clancey (Athens, USA) D. Voiculescu (Berkeley)
L. A. Coburn (Buffalo) H. Widom (Santa Cruz)
K. R. Davidson (Waterloo, Ontario) D. Xia (Nashville)
R. G. Douglas (College Station) D. Yafaev (Rennes)
A. Dijksma (Groningen)
H. Dym (Rehovot) Honorary and Advisory
P. A. Fuhrmann (Beer Sheva) Editorial Board:
S. Goldberg (College Park) C. Foias (Bloomington)
B. Gramsch (Mainz) P. R. Halmos (Santa Clara)
G. Heinig (Chemnitz) T. Kailath (Stanford)
J. A. Helton (La Jolla) P. D. Lax (New York)
M. A. Kaashoek (Amsterdam) M. S. Livsic (Beer Sheva)
Spectral Methods
for Operators of
Mathematical Physics
J.Janas
P. Kurasov
S.Naboko
Editors
Springer Basel AG
Editors:
Jan Janas Pavel Kurasov

Institute of Mathematics P AN Department of Mathematics
ul. Sw. Tomasz 30 Lund Institute of Technology
31-027 Cracow Box 118
Poland 22100 Lund
e-mail: najanas@cyf-kr.edu.pl Sweden
e-mail: kurasov@maths.lth.se
Sergei Naboko
Department of Mathematical Physics
Institute of Physics
St. Petersburg University
Ulianovskaia I
198904 St. Petersburg
Russia
2000 Mathematics Subject Classification 47-06; 47AIO, 47A45, 47A75, 47B25, 47B39
A CIP catalogue record for this book is available from the

Library of Congress, Washington D.C., USA
Bibliographic information published by Die Deutsche Bibliothek

Die Deutsche Bibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data is available
in the Internet at <http://dnb.ddb.de>.
ISBN 978-3-0348-9632-0 ISBN 978-3-0348-7947-7 (eBook)

DOI 10.1007/978-3-0348-7947-7
This work is subject to copyright. All rights are reserved, whether the whole or part of the material is concerned,
specifically the rights of translation, reprinting, re-use of illustrations, recitation, broadcasting, reproduction on microfilms
or in other ways, and storage in data banks. For any kind of use permis sion of the copyright owner must be obtained.
2004 Springer Basel AG

Originally published by Birkhiiuser Verlag, Basel, Switzerland in 2004
Softcover reprint of the hardcover lst edition 2004
Printed on acid-free paper produced from chlorine-free pulp. TCF 00
Cover design: Heinz Hiltbrunner, Basel
ISBN 978-3-0348-9632-0
987654321 www.birkhauser-science.com
Contents
Introduction ............................................................. vii
Vadim Adamyan and Heinz Langer

Partial Non-stationary Perturbation Determinants 1
Dariusz Cichon, Jan Stochel and Franciszek Hugon SzaJraniec

Noncommuting Domination......................................... 19
Petru Cojuhari
The Problem of the Finiteness of the Point Spectrum for
Self-adjoint Operators. Perturbations of Wiener-Hopf
Operators and Applications to Jacobi Matrices. . . . . . . . . . . . . . . . . . . . . . 35
Vladimir Derkach, Seppo Hassi and Henk de Snoo
Singular Perturbations as Range Perturbations
in a Pontryagin Space. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
Aad Dijksma, Heinz Langer, Annemarie Luger and Yuri Shondin
Minimal Realizations of Scalar Generalized Nevanlinna
Functions Related to Their Basic Factorization 69
Joanne Dombrowski
Eigenvalues and Spectral Gaps Related to
Periodic Perturbations of Jacobi Matrices ........................... 91
Jeffrey S. Geronimo, Oscar Bruno and Walter Van Assche
WKB and Turning Point Theory for
Second-order Difference Equations .................................. 101
Daphne J. Gilbert, Bernard J. Harris and Suzanne M. Riehl
Bounds for the Points of Spectral Concentration of
One-dimensional Schrodinger Operators ............................. 139
Johanna Michor and Gerald Teschl
Reconstructing Jacobi Matrices from Three Spectra 151
Andrey S. Osipov
On Some Asymptotic Properties of Solutions for
a Particular Class of Finite Difference Equations 155
VI Contents
Boris Pavlov
A Remark on Spectral Meaning of the
Symmetric Functional Model........................................ 163
Roman Romanov
A Remark on Equivalence of Weak and Strong Definitions of the
Absolutely Continuous Subspace for Nonself-adjoint Operators ...... 179
Alexei Rybkin
On a Transformation of the Sturm-Liouville Equation with Slowly
Decaying Potentials and the Titchmarsh-Weyl m-function ........... 185
Luis O. Silva
Uniform Levinson Type Theorems for Discrete Linear Systems 203
Alexey Tikhonov
Free Functional Model Related to Simply-connected Domains 219
Marcin J. Zygmunt
Jacobi Block Matrices with Constant Matrix Terms 233
List of Lecture Titles ..................................................... 239

Introduction
This volume contains proceedings of the International Conference: Operator The-

ory and its Applications in Mathematical Physics - OTAMP 2002 held at Mathe-
matical Research and Conference Center in Bedlewo near Poznan.
The aim of the conference was to gather researches working in close areas
of operator theory, analysis and mathematical physics, which is reflected by the
titles of scientific sessions
Random and quasiperiodic Schrodinger operators
(P. Stollmann and G. Stolz);
Jacobi matrices and orthogonal polynomials
(J.S. Geronimo and W. Van Assche);
Singular perturbations of self-adjoint operators
(W. Karwowski and P. Kurasov).
The current volume contains in addition to materials of the lectures given at the
conference original research articles, several ones initiated during the conference.
Two main entirely connected themes dominate the volume
spectral properties of l-dimensional Schrodinger operators
and infinite Jacobi matrices,
theory of self-adjoint and dissipative operators.
Contributions devoted to the first theme contain in particular results on the ex-
istence and finiteness of the point spectrum of Jacobi matrices, bounds for the
points of spectral concentration of one-dimensional Schrodinger operators, WKB
and turning points for the second order difference equations. The second theme
is represented by the articles devoted to partial non-stationary perturbation de-
terminants, self-adjointness by domination of commutators, symmetric functional
models etc.
The Organizing Committee of the conference takes this opportunity to thank
all session organizers for helping in putting together the scientific programm and
all participants for coming to Bedlewo and making this conference into a useful
scientific event. Special thanks go to M. Moszynski for helping in organization. We
would like to thank
Stefan Banach International Mathematical Center
viii Introduction
for generous financial support. It is impossible to imagine the conference without

warm hospitality from everybody working at
Mathematical Research and Conference Center in Bedlewo.
The Editors would like to thank all the referees assisting in preparation of this
volume and coming with numerous suggestions helping to keep the high standard
of this volume.
Finally we are indebted to The Editorial Board and in particular to Professor
1. Gohberg for including these Proceedings in the series Operator Theory: Advances
and Applications and to Birkhauser Verlag for patience and help in preparation of
the volume.
Krakow-Lund-St. Petersburg, March 2004

The Editors
Operator Theory:
Advances and Applications, Vol. 154, 1-18
2004 Birkhiiuser Verlag Basel/Switzerland
Partial Non-stationary
Perturbation Determinants
Vadim Adamyan and Heinz Langer
Abstract. A partial non-stationary perturbation determinant ~I(t) is defined

as follows:
~I(t):= det (eitAPre-itHI1-tJ, t > 0;
here A is a self-adjoint operator in some Hilbert space 'lt1, H is a self-adjoint
operator in a larger Hilbert space 'It :) 'lt1, Pr is the orthogonal projection in
'It onto 'ltl and Pr(H - A)I1-tl is a trace class operator in 'lti. If the operator
Pr H (I - Pr) is finite-dimensional, ~ I (t) is expressed by the resolvent kernel
of a system of Fredholm integral equations on (0, t) of second kind. Moreover,
in a particular situation the asymtotic behavior of ~1 (t) for t --> 00 is studied.
Mathematics Subject Classification (2000). 47D06, 47A55, 47BlO, 47A99;
45B05, 47A40.
Keywords. Evolution equation, groups of unitary operators, perturbation de-
terminant, block operator matrix, spectral shift function.
1. Introduction
By a non-stationary perturbation determinant we mean a function of the form
t > 0, (1.1)
where Ho and H are self-adjoint operators in some Hilbert space 1i such that the
difference H - Ho is a trace class operator. A partial non-stationary perturbation
determinant is defined by the relation
~l(t) := det (eitAPle-itHI1-tJ, t > 0, (1.2)
where A is a self-adjoint operator in some Hilbert space 1i l , H is a self-adjoint

operator in a larger Hilbert space 1i J 1i l , Pl is the orthogonal projection in 1i
V. Adamyan was supported by grant CRDF UMI 2090 of the U.S. Civilian Research and Devel-
opment Foundation. He also thanks the Vienna University of Technology for its hospitality.
2 V. Adamyan and H. Langer
onto Ji l , and H(H - A)IJ-ll is a trace class operator in Ji l . We are interested in

more explicit expressions for ~(t) and ~l(t) and, in particular, in the asymptotic
behavior of ~l (t) for t -----; 00.
These studies are motivated by the paper [4] which, in turn, was inspired by
results of P.W. Anderson, cf. [5]. In [4] it was shown that the function f(w), which
describes the intensity of the spectral lines in the x-ray photoemission spectrum
of metals plotted against the radiation frequency w, within a certain degree of
approximation can be calculated by the formula
f(w) = ~rg lR 1
00
eiwte-ctdet (f + E~F (eitHoe-itH - 1)) dt. (1.3)
Here Ho and H are the effective self-adjoint one-particle energy operators for
the electron in the metal before and after the photo emission, E~ is the spectral
function of H o , and CF is the Fermi energy level.
On the right-hand side of (1.3) there appears a partial non-stationary per-
turbation determinant of the form
(1.4)
with Jil = ran E~ and A = P1HoPl . In this case the operators Hand Ho in (1.4)
have the following additional property: if we represent them by block operator
matrices with respect to the decomposition Ji = Ji l EB Jit, Ji l := P1Ji = E~Ji,
of the space Ji, then Hand Ho have the same diagonal blocks, say A and D and
the spectra of these diagonal blocks are weakly separated. The latter means that
there exists some real a such that
max(J(A) :::; a :::; min (J(D) (1.5)
and that a is not an eigenvalue of A and of D. Therefore the results from [1], [3]
can be applied.
This note is organized as follows. In Section 2 we consider non-stationary
perturbation determinants (1.2), actually in the slightly more general setting with
the unitary groups replaced by families of solutions of two evolution equations.
For the particular case of self-adjoint operators Ho and H, H = Ho + V with a
trace class operator V, it turns out that the well-known formula for the finite-
dimensional case generalizes to
t > O.
In Section 3 we consider partial non-stationary perturbation determinants
under the assumption that the diagonal blocks of H have weakly separated spec-
trum. This allows to apply the angular operator representations from [2], [3] of
the spectral subs paces of H, corresponding to (-00, a) and (a, +(0) . As the main
result of this note we show in Theorem 3.4 that if the spectrum of the perturbed
operator is absolutely continuous in at least one of the intervals (-00, a), (a, +(0)
then
~l(t) = e- b - iat (1 + 0(1))
with non-negative numbers a and b. Moreover, under an additional assumption

these numbers are zero if and only if the off-diagonal blocks of H are zero.
Finally, in Section 4 for the case of a finite-dimensional off-diagonal per-
turbation we express the partial non-stationary perturbation determinant by the
Fredholm kernel of a system of second kind Fredholm integral equations.
2. Non-stationary perturbation determinants

Let H be a Hilbert space. In the sequel we often deal with integrals J:
S( T) dT for
operator-valued functions S(t) which are continuous in the nuclear norm II . III of
the ideal Sl of trace class operators in H. These integrals are to be understood as
the limits of the corresponding Riemann sums with respect to this norm. Evidently,
they define operators of the trace class Sl.
The following theorem is formulated in a more general form than needed
below. We use the notion of an evolution system as defined in [10, Definition 5.3],
which is a two-parameter family of bounded linear operators W(t, s), 0 :::; s :::; t <
00, in H such that the following two conditions are satisfied for 0 :::; s :::; r :::; t < 00:
(i) W(s, s) = I, W(t, s) = W(t, r)W(r, s),
(ii) the mapping (t,s) -----+ W(t,s) is strongly continuous.
Sufficient conditions for the fact that the operator functions Ho(t) and H(t) gen-
erate evolution systems can be found in [10, Chapter 5J.
Theorem 2.1. For t E [0,00), let Ho(t) be self-adjoint and H(t) be densely defined
operators, such that
H(t) - Ho(t) =: V(t) E Sl,
and that the operator function V(t) is continuous on [0,00) with respect to the
norm of Sl. Suppose further that there exist evolution systems
Wo(s, t), W(s, t), 0:::; s :::; t < 00,
and a dense subset V c H, such that for 0:::; s < t, x E V, it holds
. aWo(t, s)x _ II, ( )W, ( ) . aW(t, s)x = H( )W( )
1 at - 0 t o t , s x, 1 at t t, s x,
with the derivatives on the left-hand sides to be understood with respect to the norm
ofH. Then
Wo(t, s)*W(t, s) = I - i it WO(T, S)*V(T)W(T, s) dT, (2.1)
where the integral on the right-hand side exists with respect to the norm of Sl and
belongs to Sl, and
i.l(t,s) := det (Wo(t,s)*W(t,s)) = e

-ii s
t
tr [V(T)J dT
(2.2)
4 v. Adamyan and H. Langer
Proof For x, y E V and 0 S; s < t < 00 we consider the following relation where'
denotes the derivative with respect to t:
(W(t, s)x, Wo(t, s)y)' = (W'(t, s)x, Wo(t, s)y) + (W(t, s)x, W~(t, s)y)
= (-iH(t)W(t, s)x, Wo(t, s)y) + (W(t, s)x, -iHo(t)Wo(t, s)y)
= -i ((H(t) - Ho(t))W(t, s)x, Wo(t, s)y)
= -i (V(t)W(t, s)x, Wo(t, s)y).

Since the expression on the right-hand side is a continuous function of t for s S; t
and the set V is dense in 1{ the relation (2.1) follows. From the special choice
H(t) = Ho(t) we find that
Wo(t, s)*Wo(t, s) = I, 0 S; s S; t < 00.
Further,
Wo(t + 8, s)*W(t + 8, s) = Wo(t, s)*Wo(t + 8, t)*W(t + 8, t)W(t, s)
= [Wo(t, s)*Wo(t + 8, t)*W(t + 8, t)Wo(t, s)J [Wo(t, s)*W(t, s)J
and hence
det [Wo(t + 8, s)*W(t + 8, s)J = det [Wo(t + 8, t)*W(t + 8, t)J det [Wo(t, s)*W(t, s)J
or
~(t + 8, s) = ~(t + 8, t)~(t, s). (2.3)
It follows that
~(t+8,s) -~(t,s) = A( )~(t+8,t) -1
(2.4)
8 ~ t, s 8 .
We denote 8(8) := fttH WO(T, t)*V(T)W(T, t) dT. Then 118(8)111 < 1 if 181 is suffi-
ciently small. The relation (2.1) implies for these values of 8
~(t + 8, t) det (Wo(t + 8, t)*W(t + 8, t))
= det (I - i8(8))
exp (tr [In(1 - i 8(8)])
exp (-i tr [8(8)J + O(118(8)IID) ,
and we find for 8 ---+ 0:
~(t+8,t) -1 exp (-itr[8(8)J + O(1I8(8)11I)) -1
8 8
(-i tr[8(8)J + O(118(8)11i))
8
~ -i t, [ j [H Wo(t; s )'V(s )W(t; s) ds 1+ O(IISi 0) 111)
---+ -i tr V(t),
and, finally,
d~d:'s) = -itr[V(t))]~(t,s).
Observing that ~(s, s) = 1 the relation (2.2) follows.
D
If, in Theorem 2.1, Ho(t) == Ho and H(t) == H are constant self-adjoint

operators, then with the unitary groups generated by Ho and H we have, e.g.,
Wo(t, s) = e-i(t - s)Ho, 0 ::; s ::; t,
and the relation (2.2) implies the following corollary.
Corollary 2.2. If Ho and H are self-adjoint operators and H - Ho E S1, then
det (eitHoe-itH) = e- it tr(H - Ho). (2.5)
A discrete analog is as follows. Let Ho and H be as in Corollary 2.2. For

z -I- z we consider theCayley transforms
Uo(z) := (Ho - z)(Ho - Z)-1, U(z):= (H - z)(H - Z)-1, (2.6)
which are unitary operators. Since V = H -Ho is a trace class operator the unitary
operator UO(z)-1U(z) has the property
UO(z)-1U(z) - I UO(z)-1 [U(z) - Uo(z)]
2i <Sz Uo(z)-1(H - Z)-1 V(Ho - Z)-1 E S1.
Hence for n E N we have
n-1
Uo(z)-nU(zt - I = L UO(z)-k (UO(Z)-1U(Z) - 1) U(z)k E S1.
k=O
By DH/Ho(Z) we denote the perturbation determinant of the ordered pair H,Ho,
cf. [7]:
DH/HO(Z) = det (I + (Ho - Z)-1V) = det ((Ho - z)-1(H - z)).
Theorem 2.3. If Ho and H are self-adjoint operators, H - Ho E S1 and Uo(z),
U(z) are the Caley transforms given by (2.6) then
det [Uo(z)-nU(z)n] = e2niarg DH/Ho (z). (2.7)
Proof. The properties of determinants (cf. [7]) imply

dn : det [Uo(z)-nU(z)n] = det [Uo(z)-(n-1)U(zt-1U(z)Uo(z)-1]
dn- 1 det [U(z)UO(Z)-1] = dn- 1det [UO(Z)-1U(Z)] = ...
(det [Uo(z)-1U(z)]r.
Further,
det [UO(Z)-IU(Z)] = det [(Ho - z)(Ho - z)-I(H - z)(H - Z)-I]
= det [(H - z)-I(Ho - z)(Ho - z)-I(H - z)]

= DHo/H(Z)DH/Ho(Z).
Taking into account that for a pair of self-adjoint operators H o, H with H - Ho E SI
we have DHo/H(Z) = (DH/Ho(Z)) -1 the claim follows. D
3. Asymptotic behavior of ~l(t)

Let again A be a self-adjoint operator in the Hilbert space HI, and consider in the
Hilbert space H = HI EEl H2 the operator
H = ( A;* V ~) (3.1)
which is supposed to be self-adjoint. We assume that V, B E SI and, additionally,

that the spectra of A + V and D are weakly separated, that is, for some 0 E lR,
maxa(A + V) ::; 0 ::; mina(D) (3.2)
and 0 is neither an eigenvalue of A + V nor of D. By L_ and L+ we denote
the spectral subspaces of H corresponding to the intervals (-00,0) and [0,00),
respectively.
According to the results of [1], [3] there exists a contraction X, IIXII ::; 1, from
HI into H2 such that L_ and L+ admit graph representations with an angular
operator X:
Moreover, X is a trace class operator: X E SI, ranX C domD, ranX* C domA,

and X is the unique contractive solution of the Riccati equation
XBX + X(A + V) - DX - B* = o. (3.4)
The operator Hie is isomorphic to the self-adjoint operator A + V + BX in the
Hilbert space (HI, ((1 + X* X), . )), and the operator Hlc+ is isomorphic to the
self-adjoint operator D - B* X* in the Hilbert space (H2' ((1 + XX*), . )). The
operator T in H, given by the matrix
(3.5)
is invertible,
(1 + X* X)-1
T- l = ( -(1 +XX*)-IX (3.6)
and it diagonalizes H:
H=T( A+VO+BX o (3.7)

D - B*X*
It follows that also the operator e- itH admits the representation

.
e- ltH = T
( e-it(A+V+BX) o (3.8)
o e-it(D-B* x*)
If P l denotes again the projection onto Hl in H, then

Ple-iHtIHI = e-it(A+V+BX) (I+X* X)-l+X*e-it(D-B* x*)(I+XX*)-l X. (3.9)
Lemma 3.1. Under the assumptions at the beginning of this section we have
e-ittr(V+BX)
~l(t) == dete-itAPle-itHIHI = det(I +X*X) .6.(t), (3.10)
where
.6.(t) := det (I + eit(A+V+BX) X*e-it(D-B* x*) X) .
Proof. Using the identity

(I + XX*)-l X = X(I + X* X)-l,
it follows from (3.9) that
e itA P 1 e- itH IHI
= eitAe-it(A+V+BX) (I + eit(A+V+BX) X*e-it(D-B* x*) X) (I + X* X)-l
and hence
det (e itA P e-itHI ) = det (eitAe-it(A+V+BX)) .6.(t)

1 HI det(I + X* X)
By Theorem 2.1
(3.11)
and (3.10) is proved. o

Lemma 3.2. Let (U(t))-oo<t<+oo be a group of unitary operators in the Hilbert
space H l , let Y be a trace class operator from Hl into the Hilbert space H 2 , and
let W(t), t > 0, be a function whose values are bounded operators in H2 and which
is bounded in the operator norm: IIW(t)11 ::; c, t > o. If the infinitesimal generator
of the unitary group U(t) has absolutely continuous spectrum then
lim det (I
t---+oo
+ U(t)Y*W(t)Y) = l.
Proof. First we observe that

det (1 + U(t)Y*W(t)Y) = det (I + YU(t)Y*W(t)) .
The trace class operator Y can be represented as
(3.12)
v
where ('l/Jv) and (IPv) are orthonormal systems of elements in HI and H 2 , respec-
tively, and Tv > 0, l: Tv < 00. It follows that
v
and
(3.13)
I",V
Since the spectrum of the infinitesimal generator of the unitary group (U(t)) is
absolutely continuous we have
J
00
(U(t)'l/Jv, 'l/JI") = ei'\t fl"v(>")d>", -00 <t< +00,

-00
where fl"v E 1L 1 (-00, 00). The Riemann-Lebesgue lemma yields

(U(t)'l/Jv, 'l/JI") ----+ 0, t ----+ 00.
Observing that
we get from (3.13)

lim IIYU(t)Y*W(t)ll l = 0,
t--->oo
and hence
lim det (I
t--->oo
+ YU(t)Y*W(t)) = 1, t ----+ 00. o
For the proof of the main theorem of this section we need one more result
which may be of independent interest. Here ~H/ Ho (>..) denotes the spectral shift
function of the pair of self-adjoint operators H, H o, cf. [6].
Theorem 3.3. For the pair of self-adjoint operators
Ho := (~ ~), H := (:* ~) (3.14)
with B E SI the relation

~H/Ho(>")=O a.e.onIR
is equivalent to B = 0.
Proof. If B = 0 then, evidently, E;,H/Ho(>\) = 0 a.e.

To prove the converse, recall that by [9] with Q := H - Ho we have
In DH/Ho(z) = (1 + (Ho - z)-lQ)
lndet
= tr In (1 + (Ho - Z)-lQ)
= JCXJ
E;,H/Ho(>\) d)'"
)...-z '
8'z -I- 0,
-CXJ
where In denotes the continuous branch of the logarithm with In 1 = O. Therefore

E;,H / Ho ()...) = 0 a.e. implies
In DH/Ho(z) = trln (1 + (Ho - Z)-lQ) = 0, 8'z -I- O. (3.15)
For the self-adjoint operator Ho and Q E 51 we have
y> 0, (3.16)
and
and the special form of the operators H o, H in (3.14) implies for non-real z
0 (A-Z)-lB) __ O
tr ((Ho - Z)-lQ) = tr ( (3.18)
(D - Z)-l B* 0 '
From (3.15), (3.16), (3.18) we obtain

0= In DH/Ho(iy) = tr ((Ho - iy)-lQ)
- ~tr ((Ho - 0(:3 )

iy)-lQ(Ho - iy)-lQ) +
= -~tr ((Ho - iy)-lQ(Ho - iy)-lQ) + 0(:3) , y i 00,
and, using (3.17), tr Q2 = 0, which because of the self-adjointness of Q implies

Q=O. 0
Note that in general for a pair of self-adjoint operators H, Ho with H - Ho E

51 the relation E;,H/ Ho ()...) = 0 a.e. does not imply that H = Ho. For example, let
Ho be a bounded self-adjoint operator and let 1 + T with T E 51 be a unitary
operator:
such that
H := (I + T*)Ho(1 + T) -I- Ho
Then for z with 'Sz -I- 0

DH/Ho(Z) = det (I + (Ho - z)-l(H - Ho))
det ((I + T)-l(Ho - z)(I + T)(Ho - Z)-l) = 1,
and hence eH/Ho(>\) = 0 a.e.
Theorem 3.4. Suppose that, in addition to the assumptions in the first paragraph
of this section, at least one of the operators Hlc_' Hlc+ has absolutely continuous
spectrum. Then
~l(t) = e- b - iat (1 +0(1)), t - t 00, (3.19)
where
a = tr (V + BX) ~ 0, b = In det(I + X* X) ~ o.
The relation b = 0 is equivalent to B = 0; if V ~ 0 then a ~ 0, and a = 0 is
equivalent to V = B = o.
Proof. We consider the case that Hlc_ has absolutely continuous spectrum. Then
the same is true for the isomorphic self-adjoint operator A + V + BX in the Hilbert
space (HI, ((I + X* X), . )). Because of (3.10) the relation (3.19) will follow if we
show that
Li(t) = det (I + eit(A+V+BX) X*e-it(D-B* x*) X) = 1 + 0(1), t -t 00. (3.20)
The self-adjointness of the operator A + V + BX in (HI, ((I + X* X), . )) means

that
(I + X* X)(A + V + BX) = (A + V + BX)*(I + X* X),
or that the operator
A := (I + X* X)1/2(A + V + BX)(I + X* X)-1/2
is self-adjoint in HI. Then
eitA = (I +X*X)1/2eit (A+V+BX)(I +X*X)-1/2, -00 <t < 00,
is a group of unitary operators in HI with generator A which has absolutely

continuous spectrum. Properties of determinants allow to represent Li(t) in the
form
Li(t) = det (I + eitA(I + X* X)1/2X*e- it (D-B* x*) X(I + X* X)-1/2) ,
and (3.20) follows from Lemma 3.2 if we observe that X E 8 1 and that the group
e-it(D-B* X*), -00 < t < 00, is bounded in HI since also the operator D - B* X*
is similar to a self-adjoint operator in HI.
If B = 0 then X = 0, det(I + X* X) = 1, and b = Indet(I + X* X) = o.
Conversely, b = 0 implies X = 0, and the Riccati equation (3.4) yields B = o.
In order to prove the last claim of the theorem we introduce the operator
Ho := ( A ~ V ~).
i:
Then, because of [1, Theorem 3.3],
a = tr (V + BX) = ~H/iIo (A) dA + tr V. (3.21)
In [1, Theorem 2.1 1 it was shown that ~H/iIo(A) :::; 0 a.e. on (-oo,a) and
~H/iIo(A) ~ 0 a.e. on (a,oo). In particular, the integral in (3.21) is:::; 0, and
since V :::; 0 by assumption, also the expression on the right-hand side of (3.21) is
:::; 0 and hence a :::; O.
Suppose now that a = O. Then (3.21), V :::; 0 and ~H/iIo (A) < 0 a.e. on
(-00, a) imply that
and hence
V=O, ~H/iIo(A) = 0 a.e. on (-00, a).
Since V = 0 we have Ho = Ho, and also
J~H/Ho(A)dA = J~H/Ho(A)dA,
00 00
0= tr(H - Ho) =
-00 a
which, because of ~H/Ho(A) ~ 0 a.e. on (a,oo), yields ~H/Ho(A) = 0 a.e. Finally,

Theorem 3.3 implies B = O. 0
In the following for the self-adjoint operator
H=(:* ~) (3.22)
in 1i = 1i1 EB 1i2 the invertibility of the operator

W(t) := PIe- itH /
1-1
plays some role.
Lemma 3.5. Suppose that the spectra of the self-adjoint operators A and D in
(3.22) are separated:
maxO"(A) :::; a :::; minO"(D)
with some a E ~ which is neither an eigenvalue of A nor of D, and suppose that
B is compact. Then for all t E ~ the operator P I e- itH I1-1 is invertible.
Proof. We consider the representation (3.3) of the spectral subspaces L_ and L+

of H corresponding to the intervals (-oo,a) and [a, 00), respectively. Since a is
not an eigenvalue of A and of D we have IIXxl1 < IIxll for all x E 'lt 1, X =I- 0,
and the compactness of X implies that IIXII < 1. According to (3.9), the operator
W (t) can be written as
W(t) = e-it(A+BX)W(t)(J + X* X)-I,
where
W(t) := J + eit(A+BX) X*e-it(D-B* x*) X.
The operator A + BX is similar to a self-adjoint operator, therefore the group
(e-it(A+BX))tEIR is similar to a group of unitary operators and hence invertible.
Now the invertibility of W(t) follows if we show the invertibility of W(t).
If W(t) would not be invertible, because of the compactness of X there would
exist an element x E 'lt1, X =I- 0, such that
x = _eit(A+BX)X*e-it(D-B*X*)Xx.
Applying the operator (J + X* X)I/2, using the identities

(J +XX*)1/2X = X(J +X*X)I/2
and taking into account that the operators (J +X* X)I/2e- it (A+BX)(J +X* X)-1/2
and (J + XX*)1/2e it (D-B* x*)(J + XX*)-1/2 are unitary, we find
II(I + X* X)I/ 2x l
= II(I +X*X)I/2e it (A+BX)(J +X*X)-1/2(1 +X*X)I/2X*e- it (D-B*X*) Xx ll

= IIX*(J + XX*)1/2e- it (D-B* x*) Xxii
< 11(1 +XX*)1/2e- it (D-B*X*)(J +XX*)-1/2(J +XX*)1/2Xxll
= IIX(J + X* X)I/2XII < 11(1 + X* X)I/ 2XII,

a contradiction. o
4. Partial perturbation determinants and Fredholm resolvents

In this section we consider the partial perturbation determinant
~I(t) = det (eitAPIe-itHI1tJ.

for the operator
in the Hilbert space H = Hl EB H2, where A and D are self-adjoint operators in

Hl and H 2 , respectively, and the operator B is finite-dimensional. We represent
Bas
n
B = I>v(- ,'ljIv)CPv,
v=l
with orthonormal systems (CPv)l and ('ljIv)'l in Hl and H2, respectively, and positive
numbers Tv, l/ = 1,2, ... , n.
First we recall the following formulas which hold for any self-adjoint opera-
tor T:
J
00
1 (T - >. - ic:)-le-L\t d>', t > 0, c: > 0, (4.1)

27ri
-00
i J
00
e-it(T->-.-iE:)dt, c: > 0; (4.2)

o
the integrals are defined in the strong operator topology. It follows that
e -Etple -itHI
H,
J
-00
00 ( 4.3)
1 W(>. + ic:)-le-iAtd>., t> 0, c: > 0,
27ri
-00
where
W(z) :=A-z-B(D-z)-lB*, 'Sz i= 0.
On the resolvent sets p(A) and p(D) we introduce the holomorphic n x n-matrix
functions
A straightforward calculation leads to the representation

n
W(Z)-l = (A - Z)-l + 'L gl"v(z) ( . ,(A - Z)-lcpl") (A - z)-l CPv, (4.4)
I",v=l
where the functions gl"v(z) are the entries of the n x n-matrix function
Recall that a Nevanlinna function is a function which is analytic in the upper

half-plane and has a non-negative imaginary part there.
Lemma4.1. The functions G1(z), G2(Z) and G(z) are Nevanlinnafunctions. More-
over, G(z) admits the representation
J A~Z d~(A),
00
G(z) = SSz =I- 0, (4.5)

-00
with a non-decreasing bounded n x n-matrix function ~(A) on JR.. Consequently,

with the positive definite n x n-matrix function
J
00
Q(t):= e- iAt d~(A), t E JR., (4.6)

-00
the representation
J
00
G(z) = i eiztQ(t) dt, SSz > 0, (4.7)

o
holds.
Proof. That G1(z) and G2(z) are Nevanlinna functions is clear, for G(z) this fol-
lows from the relation
-G(Z)-l = G1(z) + (-G 2(Z)-1)
and the fact that both summands on the right-hand side are Nevanlinna functions.
Since
y i 00,
and
supyIIG 2 (iy)11 < 00,
y>O
the function G(z) admits the representation (4.5), cf. [8, Theorem 1.4.1]. 0
Inserting (4.4) with z = A + if: into the right-hand side of (4.3), using (4.1),
(4.2), (4.7) and the convolution theorem, and setting Q(t) =: (qjll/(t)):,I/=l we find
= e- itA _ R(t)e- itA (4.8)

with
(4.9)
and hence
eitAP1e-itHIHl = 1 - eitAR(t)e-itA, (4.10)
~l(t) := det (eitAP1e-itHIHJ = det (1 - R(t)). (4.11)
Observe that R(t) and R'(t) = r::,V=l ( . ,'Pv(t)) 'l/Jv(t) are trace class operators.
For 1/ = 1,2, ... ,n we set
n
'l/Jv(",) := L k/1-V("')'
/1-=1
Then 'Pv (",) , 'l/Jv (",) , p,,1/ = 1,2, ... ,n, are norm-continuous 1{-valued functions and
R( t) becomes
R(t) = ~ fat ( . ,'Pv("')) ~ k/1-v("') d", = ~ fat ( . ,'Pv("')) 'l/Jv(",) d",. (4.12)
To find a more explicit expression for ~l(t) we use the following lemma,
which is a slight extension of [7, (IV.1.14)].
Lemma 4.2. Let R(t) be a function which is defined in a real neighborhood of to
and with values in Sl, which is differentiable in to with respect to the nuclear norm
and such that (1 - R(tO))-l exists. Then the function e(t) := det (1 - R(t)) is
differentiable in to and the following relation holds:
e'e(to)()
- = d In e(t) I
-d = -tr ((1 - R(to))-1 ') .
R (to) (4.13)
- to t t=to
Proof. We have for t ----- to
e(t) det (1 - R(to) - R'(to)(t - to) to)) + oCt -
= det(1 - R(to)) det (1 - (1 - R(tO))-l R'(to)(t - to) + oCt - to))
= e(to) (I - tr ((1 - R(tO))-l R'(to))(t - to) + oCt - to))),
where the symbol oCt - to) is to be understood with respect to the nuclear norm.
It follows that
e(t) - e(to) = -tr

(t - to)e(to)
(1 _ R(tO))-l R'(to) + oCt - to)),
t - to
t _____ to,
which implies (4.13). 0
In order to calculate the trace on the right-hand side of (4.13) for the operator
R(t) from (4.9), with
k~v(e,,,,) := ('l/Jv (",), 'P/1-(e)) , p,,1/ = 1,2, ... , n, 0 ~ e,,,, ~ t,
we denote by Kt(e,,,,) the n x n-matrix kernel
Kt(e,,,,) := (k~v(e,,,,)) :,V=l '

and by rt(~,ry) := (r~,v(~,ry)):,V=l the corresponding Fredholm resolvent kernel.

We recall that this means that
Lemma 4.3. Let the operator R( t) be given by (4.9) or (4.12). If the inverse (I -
R(t))-l exists then the relation
tr ((1 - R(t))-l R'(t)) = tr rt(t, t),
holds, where on the right-hand side tr denotes the matrix trace.
Proof. Consider for f E 1i the equation
x - R(t)x = f,
that is
(4.15)
For an element y E 1i we denote
Yv(ry) := (y, <Pv(ry)) , 0:::; ry :::; t, l/ = 1,2, ... , n,
and y(ry) := (Yl (ry), Y2(ry),'" ,Yn(ry)t Taking the L2(O, t)-inner product of equa-
tion (4.15) with <pp(~) for p = 1,2, ... , n we obtain
Hence
x(ry) = f(ry) + fat rt(ry,w)f(w)dw, 0:::; ry:::; t,
and we get from (4.15)
x = (1 - R(t))-l f = f + fat (f(ry) + fat rt(ry,w)f(w)dw, 'l/J(ry)) I(n dry.
We have to apply this formula to an element of the form

n n
f = R'(t)g = L(g, <Pv(t))'l/Jvk(t) = Lgv(t)'l/Jv(t).
v=l v=l
It follows that
(I - R(t))-I R'(t)g ~gv(t) (v(t) + fat (~(v(t), I{Jp (ry))p (ry)

+~ l~ 'Y;"(q,w)(","(t), 1'" (w))d""p,(q) )dq )
~gv(t) (v(t) + ~ fat (k~)ry, t)

+ fat ~ 'Y~J1-(ry,w)ktv(w, t)dw )p(ry)dry)
~(g, I{Jv(t)) (v(t) + ~ fat 'Y~v(ry, t)p(ry)dry)

and hence
v=1
=tr (rt(t,t)). o
Combing Lemma 4.2 and Lemma 4.3 we obtain the following result.
Theorem 4.4. Let the self-adjoint operator H in the Hilbert space H = HI EE1H2
be given by the matrix
H=(;* ~)
with self-adjoint operators A and D in the Hilbert spaces HI and H 2 , respectively,
and a finite-dimensional operator B:
n
B = LTv(-,v)I{Jv,
v=1
where (I{Jv)1" and (v)r are orthonormal systems in HI and H2, respectively, and
Tv, V = 1,2, ... , n, are positive numbers. If, for 0 ~ s ~ t, the operator I - R(s)
is invertible and rt(~, ry), 0 ~ ~, ry ~ t, denotes the Fredholm resolvent kernel as
defined by (4.14) then jar the partial perturbation determinant ~l(t) it holds
<'>,(t)=det (ei<Ap,'-;'''I",) ~ exp ( I t,(r'(','ll d') .

Remark. A sufficient condition for the invertiblity of the operators (I - R(t))-l is
the separation of the spectra of the diagonal operators A and D, see Lemma 3.5.
References
[1] V. Adamyan and H. Langer: Spectral properties of a class of rational operator-valued
functions. 1. Operator Theory 33 (1995), 259-277.
[2] V. Adamyan and H. Langer: The spectral shift function for certain block operator
matrices. Math. Nachr. 21 (2000), 5-24.
[3] V. Adamyan, H. Langer, R Mennicken, and R Saurer: Spectral components of self-
adjoint block operator matrices with unbounded entries. Math. Nachr. 178 (1996),
43-80.
[4] V. Adamyan, J. Ortner, A. Salistra, and I. Tkachenko: X-Ray absorption problem
in metals within one-electron approximation. Phys. Rev. B 52 (1995), 13827-13837.
[5] P.W. Anderson: Infrared catastrophe in Fermi gases with local scattering potentials.
Phys. Rev. Lett. 18 (1967), 1049-1051.
[6] M.S. Birman and D.R Yafajev: The spectral shift function. The papers of M.G.
Krein and their further development. Algebra i Analiz 4 (1992), 1-44 (Russian).
[7] I.e. Gohberg and M.G.Krein: Introduction to the Theory of Linear Non-self-adjoint
Operators. Translations of Mathematical Monographs 18, AMS, Providence, RI.,
1969.
[8] I.S. Kac and M.G.Krein: R-functions - analytic functions mapping the upper half-
plane into itself. American Mathematical Society Translations, Series 2, 103, 1974.
[9] M.G. Krein: On the trace formula in perturbation theory. Mat. Sbornik N.S. 33 (75)
(1953),597-626 (Russian).
[10] A. Pazy: Semigroups of linear operators and applications to partial differential equa-
tions. Applied Mathematical Sciences, 44, Springer-Verlag, New York, 1983.
Vadim Adamyan
Department of Theoretical Physics
Odessa National University
65026 Odessa, Ukraine
e-mail: vadamyan@paco.net
Heinz Langer
Institute for Analysis and Scientific Computing
Vienna University of Technology
Wiedner Hauptstr. 8-10
A-1040 Vienna, Austria
e-mail: hlanger@email.tuwien.ac.at
Operator Theory:
N oncommuting Domination
Dariusz Cichon, Jan Stochel and Franciszek Hugon Szafraniec
Abstract. We extend and refine the approach of [1] concerning essential self-
adjointness (normality) of a densely defined operator subject to some domi-
nation condition involving the first or the second commutator.
Mathematics Subject Classification (2000). Primary 47B25; Secondary 47B47.

Keywords. Symmetric operator, self-adjoint operator, domination, commu-
tator.
Introduction
Essential self-adjointness of symmetric operators is an important issue both from
the theoretical point of view as well as for applications. There are different methods
of achieving this. One of them is to use an auxiliary operator, already essentially
self-adjoint, which interacts with the operator in question. The way the essentially
self-adjoint operator interacts with the candidate operator is twofold. The first is
a kind of domination, the other splits in two different approaches depending on
if the operators commute in a sense (from the pioneering work of [2] to recent [5]
with some intermediate references like [3]) or not (here [2] again, the treatise [4]
as well as [1] with further references therein). The latter consists in replacing the
zero commutators by those which are controlled somehow. For instance, in [1] the
commutators are relatively bounded. In the present paper we refine the technique
of [1] working under more subtle growth conditions for the first or the second
commutator. Our approach (avoiding the form language of [1]), besides providing
substantially more general results, turns out to be better suited for making the
arguments more precise and simpler. In particular, so simple a case of bounded
perturbation of a self-adjoint operator fits in our results but fails to satisfy the
requirements of [1] (cf. Example 17).
This work was supported by the KBN grant 2 P03A 037 024.
20 D. Cichon, J. Stochel and F.H. Szafraniec
1. Preparatory facts
Throughout the paper 1i stands for a complex Hilbert space, B(1i) for the C*-
algebra of all bounded linear operators on 1i and I for the identity operator on 1i.
Given a linear operator A in 1i, we denote by V(A), N(A), A*, A and a(A) the
domain, the kernel, the adjoint, the closure and the spectrum of A, respectively.
Set VOO(A) = n::'=l
V(An). The graph norm of A is denoted by II . IIA' i.e., Ilfll~ =
IIfl12 + IIAfl12 for f E V(A). Recall that a linear subspace of V(A) is said to
be a core of A if the graph of A is contained in the closure of the graph of Ale.
We say that a symmetric operator A in 1i is essentially self-adjoint on if is
a dense linear subspace of V(A) and (Ale)* = Ale. By maximality of self-adjoint
operators, a symmetric operator A is essentially self-adjoint on if and only if
A is self-adjoint and is a core of A. If A and B are linear operator in 1i, then
[A, B] stands for the commutator of A and B, i.e., the operator [A, B] ~ AB - BA
with the domain V(AB) n V(BA). For a given integer k ;;,: 0, we define the k-th
commutator (adA)k(B) via
(adA)O(B) = Band (adA)k+l(B) = [A, (adA)k(B)] for k ;;,: O.
An induction argument based on C~l) + e) = (kjl) shows that l
k
(adAl(B) 2 ~ G)Aj B( _A)k- j , k;;': O. (1)
We say that B dominates A on if is a linear subspace of V(A) n V(B) and

there exits c ;;,: 0 such that IIAfl1 ~ c(llfll + IIBfll) for all f E (cf. [5] for more
details).
Suppose that H is a closed densely defined linear operator in 1i. Throughout
the whole paper R(z) stands for the resolvent of H, i.e., R(z) = (Z-H)-l E B(1i)
for z E C \ a(H). We show by induction on m that
R(z)V(Hm) ~ V(Hm+l) for all integers m;;': O. (2)
Indeed, if (2) holds for a fixed m ;;,: 0 (the case m = 0 is trivial), then
R(z)V(Hm+l) = R(z)V((z - H)m+l) = R(z)(R(z)m+l1i)
= R(z)m+21i = V((z - H)m+2) = V(Hm+2).
Condition (2) implies R(z)VOO(H) ~ VOO(H). In consequence, if A is a linear
operator in 1i such that VOO(H) ~ V(A), then VOO(H) ~ V([A, R(z)k]) and
VOO(H) ~ V((adR(z))k(A)) for all integers k ;;,: O. Likewise, if V(Hm) ~ V(A),
then V(Hm) ~ V([A,R(z)k]) and V(Hm) ~ V((adR(z))k(A)) for all integers
k,m;;':O.
Proposition 1. Let A be a closed operator in 1i and H be a self-adjoint operator
in 1i. Then VOO(H) ~ V(A) if and only if there exists an integer m ;;,: 0 such that
1 Notice that if V(A) = 'It, then the inclusion (1) becomes an equality.
Noncommuting Domination 21
(i) D(Hm) <:;;; D(A).

Suppose that m ~ 0 is an integer, E is a core2 of Hm and F <:;;; D(AR(z)m) is
a linear subspace such that D(AR(z)m) <:;;; F. Then (i) is equivalent to any of the
following conditions:
(ii) E <:;;; D(A) and there exists c ~ 0 such that
IIAfl1 ~ c(llfll + IIHm fll), fEE, (3)
(iii) the operator AR(z)mIF is bounded and densely defined in H for some z E
C \ a(H) (equivalently: for every z E C \ a(H)).
Furthermore, if (ii) holds, then D(Hm) <:;;; D((adR(z))k(A)) and
II(adR(z))k(A)fll ::::: c2kIIR(z)llk(llfll + IIHmfll), f E D(Hm), (4)
for all integers k ~ 0 and z E C \ a(H); what is more, AlE = AID(H=).
If (iii) holds, then D(Hm) <:;;; D(A) and AR(z)m E B(H) for every z E C \ a(H).
Proof. First we show that if DCXJ(H) <:;;; D(A), then there exists an integer m ~ 0
such that (ii) holds for E = DCXJ(H). For this, observe that DCXJ(H) is a Frechet
space with the topology given by the system of graph norms {IIIIHJ }~o (because
each operator Hj, j ~ 0, is closed) and (D(A), II . IIA) is a Banach space. By the
closed graph theorem the inclusion DCXJ(H) 3 f f-----+ f E D(A) is a continuous
linear operator, and hence there exist c' ~ 0 and an integer m ~ 0 such that
m
IIAfl12 ~ c'L IIHj f11 2, f E DCXJ(H).

j=O
Since there exists d m ~ 0 such that 1+x 2 + .. +x 2m ~ d m (1+x2m) for all x E IR,
we get (by the spectral theorem)
IIAfl12 ~ c'dm(llfl1 2+ IIHm fI1 2), f E DCXJ(H),
which proves our claim.
If (ii) holds, then E being a core of Hm and the closedness of A imply
D(Hm) <:;;; D(A), IIAfl1 ~ c(llfll + IIHm fll) for all f E D(Hm) (5)
and AlE = AIv(H=). Combining (5) with the previous paragraph, we see that
VCXJ(H) <:;;; D(A) if and only if V(Hm) <:;;; D(A) for some m ~ o.
Applying the closed graph theorem, we see that (i) implies (ii) with E =
V(Hm), which by the previous paragraph justifies also the equivalence (i){:}(ii).
In the next step we show the equivalence (ii){:}(iii) for a fixed z E C \ a(H).
(ii)=}(iii) Fix 9 E H. Since (ii)=}(5), we can substitute f ~ R(z)mg into (5)
and apply the identity H m R(z)mg = (H R(z))mg = (zR(z)-I)m g. In consequence,
we get AR(z)m E B(H).
2 Recall that TJOO(H) is a core of H m for every integer m ? O.

(iii)::::}(ii) Since A is closed and R(z)rn E B(H), the operator AR(z)rn is

closed as well. By (iii), AR(z)rn E B(H). This means that V(Hrn) ~ V(A) and
there exists c' ~ 0 such that
IIAR(z)rn III ~ e'll/ll, I E H. (6)
Take 9 E V(Hrn). Plugging 1= (z-H)rng into (6), we have IIAgl1 ~ e'll(z-H)rngll.
Since there exists d:n ~ 0 such that Iz - xl 2rn ~ d:n(1 + x2rn) for all x E JR, the
spectral theorem implies
IIAgl12 ~ (e,)2d:"(llgI1 2 + IIHrngI12), 9 E V(Hrn),
which manifestly yields (ii).
Suppose now that (ii) holds. Then, by (5), inequality (4) is valid for k = o.
Assuming (4) for a fixed k ~ 0, we can proceed as follows (use (2))
II (adR(z))k+l (A)/II = IHR(z), (adR(z))k(A)]/11
~ IIR(z)(adR(z))k(A)/11 + II(adR(z))k(A)R(z)/11
~ e2 k IIR(z)ll k + 1 (11/11 + IIH rn III) + e2kIIR(z)llk(IIR(z)/11 + IIR(z)Hrn III)
~ e2k+ 1 1IR(z)ll k +1(11/11 + IIHrn III), IE V(Hrn).
This completes the proof. D
The set flc,d(H) defined below plays a pivotal role in our paper. Given a
closed densely defined operator H in H, we put
flc,d(H) = {z E C: dist(z,o-(H)) ~ clzl and Izl > d}, c> O,d ~ 0,
where dist(z, a(H)) = inf{lz-wl : w E a(H)}. It is clear that flc,d(H) ~ C\a(H).

Suppose that H is self-adjoint. Then, due to IIR(z)11 = dist(z,a(H))-l, we have
1
flc ,d(H) = {z E C \ a(H) : IlzR(z)11 ~ -e and Izl > d}, e> 0, d ~ O. (7)
One can show that if H is self-adjoint, then flc,d(H) is unbounded if and only if
e ~ 1; if this is the case, then {i t : t E JR, It I > d} ~ flc,d(H). If H is self-adjoint
and bounded below (resp. above), e E (0,1) and d ~ 0, then (-00, a) ~ flc,d(H)
(resp. (a,oo) ~ flc,d(H)) for some a E JR. Remark 6 contains further properties of
the set flc,d(H). The reader should be aware that dist(zn,a(H)) ~ 00 may not
imply the boundedness of {llznR(zn)II}~=l' e.g., if H is self-adjoint, a(H) = {n 2 :
n = 1,2,3, ... } and Zn = ~(n2 + (n+ 1)2). This shows that the set flc,d(H) is
optimal taking into account uniform boundedness of the function zR(z).
The following result collects some properties of resolvents and commutators.
Proposition 2. Let A be a linear operator in H, H be a closed densely defined
linear operator in Hand m, k ~ 0 be integers.
IIV(Hrn) ~ V(A), then
(i) [A, R(z)k]1 = L7:~ R(z)j[A, R(z)]R(z)k-l- j I lor all I E V(Hrn) and z E
C \ a(H),
(ii) the operator (adR(z))k(A)IV(H"') is bounded for every z E C\a(H), provided

(adR(w))k(A)IV(H"') is bounded for some w E C \ a(H).
If H is self-adjoint, then for all c E (O,ll and d ~ 0,
(iii) limnC,d(H)3z--->oo(zR(z))m f = f for all f E H,
(iv) SUPZEnc,d(H) IlzR(z)11 : :; ~,
(v) sUPZEnc,d+o(H) I (ad R(z))k(A)lv(H"') II < 00, provided c: > 0, V(Hm) ~ V(A)
and (adR(w))k(A)b(H"') is bounded for some w E C \ a(H).
If A is closable, H is self-adjoint and V(Hm) ~ V(A), then
(vi) the operator (adR(z))k(A)b(H"') is bounded, provided (adR(z))k(A) is
bounded on a core of Hm, z E C \ a(H).
If A is symmetric, H is self-adjoint and V(Hm) ~ V(A), then
(vii) the operators R(z)mA and AR(z)m are bounded for every z E C \ a(H),
(viii) the operator (adR(z))k(A) is closable for every z E C \ a(H),
(ix) ik(adR(x))k(A) is symmetric for every x E ~ \ a(H).
Proof. (i) Fixing k ~ 1 and f E V(Hm), we compute

k-l
L R(z)j[A, R(z)lR(z)k-l- j f
j=O
k-l k-l
= L R(z)j AR(z)k- j f - L R(z)j+l AR(z)k-l- j f
j=O j=O
k-l k
= L R(z)j AR(z)k- j f - L R(z)j AR(z)k- j f = [A, R(zllf.
j=O j=1
= (w-H)R(z) = I+(w-z)R(z) E B(H). Then

(ii) Let z,w E C\a(H). Set G
G- 1 =(z - H)R(w) = 1+ (z - w)R(w). By (2) we have V(Hm) ~ V([R(~), AD
for every ~ E C \ a(H), which together with the resolvent identity gives us
[R(w),Alf - [R(z),Alf = (z - w)([R(w), AlR(z) + R(w)[R(z),AJ)f
for all f E V(Hm). This implies
[R(w), Al(I + (w - z)R(z))f = (I + (z - w)R(w))[R(z), Alf, f E V(H m ),
which leads to
[R(z), AlJ = G[R(w), AlGf, f E V(Hm). (8)
We show by induction that for all integers k ~ 0,
Indeed, suppose (9) holds for a fixed k ~ (the case k =

is trivial).
Then, applying (8) to the operator (adR(w))k(A) in place of A and (9), we get
(adR(z))k+l(A)f = [R(z), (adR(z))k(A)lJ
= [R(z),Ck(adR(w))k(A)CklJ
= Ck[R(z), (adR(w))k(A)]C k f
= Ck+l[R(w), (adR(w))k(A)]C k+l f
= Ck+l(ad R(w))k+l (A)C k+l f, f E V(Hffi).
It is clear that (9) implies (ii).
Conditions (iii) and (iv) can be deduced from the spectral theorem (cf. (7)).
(v) By (9), we have
(adR(z))k(A)f = (I + (w - z)R(z))k(adR(w))k(A)(I + (w - z)R(z))kf
for all f E V(Hffi). This and (iv) implies (v) (because sUPzEf]c,d+c(H) IIR(z)11 ~ el ,,}
(vi) If f E V(Hffi), then there exists a sequence {fn}~=o <;;;; [; such that
limn--->oo Ilf - fnllH= = O. By (4), (adR(z))k(A)fn ----> (adR(z))k(A)f as n ----> 00,
which implies the boundedness of (adR(z))k(A)lv(H=)'
(vii) By Proposition 1, AR(z)ffi = AR(z)ffi E B(H) for every z E C \ (J(H).
Taking adjoints, we get R(Z)ffi A <;;;; R(Z)ffi A* <;;;; (AR(z)ffi)* for every z E C\(J(H),
which completes the proof of (vii).
(viii) By the von Neumann theorem it suffices to show that V(Hffi) is con-
tained in the domain of the adjoint of (adR(z))k(A). The case k = 0 is obvious.
Suppose that our claim is true for a fixed k ~ O. Since the operator (ad R(z))k+ l (A)
is densely defined (because its domain contains V(Hffi)), we can calculate
(adR(z))k+l(A))* :2 (adR(z))k(A))* R(z) - R(z)(adR(z))k(A))*,
which together with (2) completes the proof of (viii).
An induction argument similar to that in (viii) enables us to prove (ix). D
2. Symmetric operators dominated by self-adjoint ones

We now formulate one of the main results of the paper, which generalizes Theorem
1 of [1] (see Section 5).
Theorem 3. Let A be a symmetric operator in H, H be a self-adjoint operator in
Hand n be an unbounded subset of ne,d(H), where c E (0,1] and d ~ O.
(i) If V(Hffi) <;;;; V(A) for an integer m ) 0 and SUPzES] Izlll[R(z),A]11 < 00,
then A is essentially self-adjoint on any core of Hffi.
(ii) IfVOO(H) <;;;; V(A), SUPzES] Izlll[R(z),A]11 < 00 and [; is a core 3 of Hffi for
every integer m ) 0, then A is essentially self-adjoint on [;.
3 For instance, [; = DOC (H) or [; = the set of all bounded vectors of H, d. [5].
Remark 4. Since the operator [R(z),Al is closable (cf. Proposition 2 (viii)), it

is bounded if and only if so is its restriction to a dense linear subspace :F of
V([R(z), A]); moreover, the following equality holds II[R(z),Alll = IHR(z),AlIFII.
Proof of Theorem 3. There is no loss of generality in assuming that the operator
A is closed.
(i) By the closed graph theorem, we can assume that m ~ 1. According to our
assumption a ~ SUPzE!? II[A, zR(z)lll < 00. Let be a core of Hm. It follows from
Proposition 1 that AI = Alv(H"'l' Hence N((AI)* + z) = N((Alv(H"'l)* + z)
which is equivalent to
(10)
We show that
((A + i u)).l = {O} for every u > amc1- m . (11)
To prove (11), take f E ((A+iu)).l and zEn, and put 4 fz = zmR(z)mf E

V(Hm). Since by (2) R(z)m fz E V(H 2m ), (10) gives us
(f, (A + i u)R(z)mfz) = 0,
which can be rewritten as (f, (R(z)m(A + i u) + [A, R(z)m])fz) = O. Hence, we

have
i u(f, R(z)m fz) = (f, (R(z)m A + [A, R(z)m])fz). (12)
Since zm i u(f, R(z)m fz) = i ullfzl1 2 and zm(f, R(z)m Afz) = (fz, Afz), (12) yields
i ullfzl1 2 = (fz, Afz) + zm(f, [A, R(z)mlfz)' (13)
Comparing imaginary parts of both sides of (13), using (fz, Afz) E ~ and employ-
ing conditions (i) and (iv) of Proposition 2, we get
ullfzl1 2 = 3m (zm(f, [A, R(z)mlfz)) ~ l(f, [A, (zR(z))mlfz) I
m-l
~ L l(f, (zR(z))j[A, zR(z)](zR(z))m-l- j fz)1
j=O
~ mllfllllfzllllzR(z)llm-111[A, zR(z)liv(H"'l II
:::; amci-mllfllllfzll, zEn. (14)
Since the set n is unbounded, there exists a sequence {zn}~=l ~ n such that
lim n- HXl IZnl = 00. As zn E nc,d(H) for all n ~ 1, we infer from part (iii) of
Proposition 2 that limn->oo fZn = f. This, when combined with (14), gives us
ullfl1 2 ~ amc 1- m llfI1 2 . Since u > amc 1- m , it must be f = 0, which proves (11).
Applying (11) to -A, we see that ((A-i u)).l = {O} for every u > amc 1- m,
which means that both deficiency indices of AI are equal to zero. Hence A is
essentially self-adjoint on .
4 Notice that Z E ilc,d(H) <:;; C \ u(H).

(ii) By Proposition 1, D(Hm) ~ D(A) for some integer m ;? 0. Applying (i)

completes the proof. D
Theorem 3 (as well as Theorems 8 and 11) can be combined with commuting
domination results of [5]. What follows is a sample application of Corollary 19 of
[5] and Theorem 3.
Corollary 5. Let A be a symmetric operator in H, H be a self-adjoint operator

in Hand p be a real polynomial in one variable of degree n ;? 1. Assume that
D(Hm) ~ D(An) for some integer m ;? and sUPzES? Izlll[R(z),p(A)]11 < 00,
where Jl is an unbounded subset of Jlc,d(H) for some c E (0,1] and dE [0,(0).
Then for every real polynomial q in one variable with deg q ~ n, the operator q(A)
is essentially self-adjoint, D(An) is a core of q(A) and q(A) = q(A).
Remark 6. The set Jl appearing in Theorem 3 (as well as in Theorems 8 and 11)
can be specified in many ways not excluding the choice of unbounded sequences
in Jlc,d(H). For example one can always take Jl of the form Jl = { it: t > d} ~
Jlc,d(H). In the case when H is semibounded, Jl can be chosen as an unbounded
interval on the real line, which is disjoint from the spectrum of H. Let us discuss

two examples of semibounded self-adjoint operators. Consider first H with a(H) =
{O, 1, 2, ... }. Then Jlc,d(H) n (a, (0) is bounded for all c E (0,1]' d;? and a E R

Indeed, otherwise there exists {Xn}~=l ~ Jlc,d(H) n (a, (0) such that < Xn /00.
Then CXn ~ dist(xn, a(H)) ~ ~ for all n ;? 1, which is impossible. This shows that,
for this particular H, there is no way of choosing a subset Jl of (a, (0) satisfying the
assumptions of Theorem 3 (though for each c E (0,1) there exists a E IR such that
(-00, a) ~ Jlc,d(H)). However, the other example will show the opposite. Let now
{a 1,a2 ,a3 , ... } be the spectrum of H, where a > 1. Set Xn = ~(an +a n + 1) for
n ;? 1. Then Jl ~ {Xl, X2, X3, ... } ~ Jl =:...! 1 (H) n (0,00). Finally, the self-adjoint
0+1 '
operator H with a(H) = {... , _a 3 , -a 2 , -aI, aI, a 2 , a 3 , ... } is not semibounded,

whereas the set Jl a - I I (H) n IR is unbounded.
0:+1 '
Remark 7. A careful reader can ensure himself that Theorem 3 is valid if the self-
adjoint operator H is replaced by a normal one. However in this case, contrary to
the self-adjoint one, it may happen that the set Jlc,d(H) is bounded or even empty
regardless of the choice of c, d.
3. Semibounded symmetric operators dominated

by self-adjoint ones
In this section we formulate a criterion for essential self-adjointness of semibounded
operators, which generalizes Theorem 4 of [1] (see Section 5).
Theorem 8. Let A be a semibounded symmetric operator in H, H be a self-adjoint
operator in Hand Jl be an unbounded subset of Jlc,d(H) n IR, where c E (0,1] and
d;? 0.
(i) IfV(Hm) ~ V(A) for an integerm ~ 0 and sUPxEf.? x 2 11(adR(x))2(A)11 < 00,
then A is essentially self-adjoint on any core of Hm.
(ii) If VCXJ(H) ~ V(A), sUPxEf.? x 211(adR(x))2(A)11 < 00 and is a core of Hm
for every integer m ~ 0, then A is essentially self-adjoint on .
Remark 9. By Proposition 2 (viii), the operator (adR(x))2(A) is bounded if and
only if so is its restriction to a dense linear subspace F of V((ad R(x) )2(A)); if this
is the case, then II(adR(x))2(A)11 = II(adR(x))2(A)I.ll
Proof of Theorem 8. Assuming A is closed involves no loss of generality.
(i) The case m = 0 forces A E B(H). Suppose m ~ 1. Without loss of general-
ity we can assume that A :s; O. Let be a core of Hm. As in the proof of Theorem
3, one can show that (10) holds. By our assumption f3 ~ sUPxES? IIGxl1 < 00, where
Gx ~ ((ad(xR(x)))2(A)) - E B(H) (cf. Remark 9). We show that
((u - A))-L = {O} for every u > ~f3m2c2(1-m). (15)

2
Take f E ((u - A))-L and x E il, and put fx = xmR(x)m f E V(Hm). Then by
(2) we have R(x)m fx E V(H2m). This and (10) give us
0= (j, (u - A)R(x)m fx) = u(j, R(x)m fx) - (j, R(x)m Afx) - (j, [A, R(x)mlfx).
Since xmu(j, R(x)m fx) = ullfxl1 2 and xm(j, R(x)m Afx) = (jx, Afx) :s; 0, we get
ullfxl1 2 = (jx, Afx) + (j, [A, (xR(x))mlfx).
Comparing real parts of both sides of the above equality, we obtain
ullfxl1 2 = (jx, Afx) + '.R.e(j, [A, (xR(x))m]fx)
(16)
::; xm '.R.e(j, [A, (xR(x))m]R(x)m 1).
!
Set Dx = I:;:~l I:;;'=-;1(xR(x))j+kGx (xR(x))2(m-l)-(j+k) E B(H). Applying
formula (i) of Proposition 2 to [A, R(x)m] and [[A, R(x)], R(x)m], we conclude
xm '.R.e(g, [A, (xR(x))m]R(x)mg)
= ~x2m( (g, [A, R(x)m]R(x)mg) - (g, R(x)m[A, R(x)m]g))

m-l
= ~x2m L (g, R(x)j[[A, R(x)], R(x)m]R(x)m-l- j g)
j=O
m-l m-l
=~ L L (g, (xR(x))i+k (ad(xR(x)))2 (A)(xR(x))2(m-l)-(j+k)g)
j=O k=O
m-l m-l
=~ L L (g, (xR(x))j+kGx (xR(x))2(m-l)-(j+k)g)
j=O k=O
(17)
However by part (vii) of Proposition 2, [A, (xR(x))m]R(x)m E B(H), which to-

gether with (17) leads to
xm 9le(h, [A, (xR(x))m]R(x)m h) = (h, Dxh), h E H. (18)
Combining (16), (18) and part (iv) of Proposition 2, we have
ullfxl1 2 ~ xm 9le(1, [A, (xR(x))m]R(x)m 1) = (1, Dx1) ~ IIfl1211Dxli
(19)
~ ~m2I1fI121IxR(X)112(m-l)IICxll ::::; ~,Bm2c2(1-m)llfI12, x E n.
Since, by part (iii) of Proposition 2, fx ~ f as x E [l and Ixl ~ 00, we infer from
(19) that ullfl1 2 ~ ~,Bm2c2(1-m)llfI12, which implies (15). As the operator AlE is
semi bounded , its deficiency indices are both equal to 0, which means that A is
essentially self-adjoint on [;.
(ii) is a direct consequence of (i) and Proposition 1. 0
The next result is an application of Theorem 25 of [5] and Theorem 8.
Corollary 10. Let A l , ... ,AI< be formally normal operators in H and let H be a
self-adjoint operator in H such that the domain of Hm is contained in the domain
of A ;g Ai Al + ... + A:AI< for some integer m ~ 0. Assume that [; is a core of
Hm and
(i) [; ~ V(AiAj) n V(AjAi) and AiAjf = AjA;!, f E [;, for 1 ::::; i < j ::::; "',
(ii) (AiA;!,Ajg) = (AYf,AiAig), f,g E [;, fori,j = 1, ... ,"',
(iii) sUPxEn x 211(adR(x))2(A)11 < 00 for an unbounded subset [l of ftc,d(H) n JR.,
where c E (0,1] and d ~ 0.
Then Al , ... ,AI< are spectrally commuting normal operators and [; is a joint core
of any subsystem of {A l , ... , AI<}'
4. More on symmetric operators dominated by self-adjoint ones

Given a densely defined operator A in H, we set JmA = 2\ (A-A*). If the operator
Jm A is densely defined, then it is symmetric and consequently closable.
Theorem 11. Let A be a symmetric operator in H, H be a self-adjoint operator in
Hand [l be an unbounded subset of [lc,d(H) n JR., where c E (0,1] and d ~ 0. If

V(Hm) ~ V(A) for an integer m ~ and sUPxEn Ilxm+l Jm ([R(x), A]R(x)m) II <
00, then A is essentially self-adjoint on any core of Hm.
Remark 12. By (2) and part (vii) of Proposition 2, [R(x), A]R(x)m E B(H) and
consequently Jm([R(x),A]R(xr) E B(H). Since
-R(x)m[R(x), A] ~ ([R(x), A]R(x)m)*,
we get iT ([R(x), A]R(x)m + R(x)m[R(x), AD ~ Jm([R(x), A]R(x)m).

Proof of Theorem 11. Let be a core of Hm. As in the proof of Theorem 3 we

can assume that A is closed, m ;?: 1 and (10) holds. According to Remark 12, for
every xED, Gx ~ x m + 1 Jm([R(x), A]R(x)m) E B(Ji) and
xm+l([R(x),A]R(x)m +R(x)m[R(x),A]) <::;; 2iGx.
By our assumption "/ ~ sUPxED IIGxll < 00. We show that
((A + i u)) 1- = {O} for every u > "/ m c1 - m . (20)
Take f E ((A + i u))1- and xED, and put fx = xm R(x)m f E V(Hffi). Then by
(2) we have R(x)m fx E V(H2m). This and (10) give us
0= (1, (A + i u)R(x)mfx)
= - i u(1, R(xr fx) + (1, R(x)m Afx) + (1, [A, R(x)m]fx),
which leads to
Comparing imaginary parts of both sides of the above equality, we obtain

(21)
Set Dx = L.~-~/(xR(x))jGx(xR(x))m-l-j E B(Ji). Applying formula (i) of

Proposition 2, we get
X2m Jm(g, [A, R(x)m]R(x)m g)
1
= 2 i x2m( (g, [A, R(x)m]R(x)m g) + (g, R(x)m[A, R(x)m]g))
2\ x2m L (g, R(x)j ([A, R(x)]R(x)m + R(X)ffi[A, R(x)])R(x)m-l- j g)

m-l
=
J=O
(22)
However by part (vii) of Proposition 2, [A, R(x)m]R(x)m E B(Ji), which together

with (22) leads to
x2m Jm(h, [A, R(x)m]R(x)mh) = (h, Dxh), hE Ji. (23)
Combining (21), (23) and part (iv) of Proposition 2, we have
ullfxl1 2 = (1,Dxf) ~ mllfI1 2 1IxR(x)llm- 1 1IGxll :::; ,,/mc 1 - m llfI1 2 , x E D. (24)
Since, by part (iii) of Proposition 2, fx --+ f as xED and Ixl --+ 00, we infer from
(24) that ullfl1 2 ~ ,,/mc 1 - m llfI1 2 , which implies (20). Applying (20) to -A, we
conclude that A is essentially self-adjoint on . 0
Note that Theorem 11 implies the version of Theorem 3 in which the set D
is additionally assumed to be contained in R
30 D. Cichon, J. Stochel and F.R. Szafraniec
5. Comments and examples

The first two remarks emphasize the role played by domination in our paper.
Remark 13. The inclusion D(Hm) ~ D(A) is one of the main assumptions of
Theorems 3, 8 and 11. In virtue of Proposition 1, D(Hm) ~ D(A) if and only if A
is dominated by H m on a core of H m (provided A is closed). The choice of this
core does not determine in any way that appearing in the respective conclusions.
Remark 14. Conclusions of Theorems 3, 8 and 11 can be supplied with a joint
core assertion. Namely any core of Hm is always a joint core of (A, Hm), which
means that the joint graph 5 Q(A,Hm) of the pair (A,Hm) is contained in the
closure of Q(AI, Hml). This can be deduced directly from domination (use (5)).
Remark 15. Typical assumption in Theorems 3 and 8 is of the form
sup Izlkll(adR(z))k(A)11 < 00
zED
with k = 1 or 2. Consequently
lim II(adR(z))k(A)11 = 0
D3z-+oo
and this carries further information about the behavior of the commutator
(adR(z))k(A) at infinity (compare with part (v) of Proposition 2).
We now concentrate on relating Theorems 1 and 4 of [1] to our Theorems 3
and 8.
Remark 16. We begin with proving that Theorem 1 of [1] follows from Theorem 3.
For this, let us assume that A and H satisfy the assumptions of Theorem 1 in [1],
i.e., A is a closed symmetric operator in 1{ and H is a self-adjoint operator in 1{
such that for some w E If:- \ (J(H) , IIAR(w)mlv=(H) II < 00 for some integer m;? 0
(the latter is equivalent to D(Hm) ~ D(A), cf. Proposition 1) and the sesquilinear
form rp associated with the expression (adH)(A)R(w) is bounded on DOO(H). Set
C z = I +(w-z)R(z). Then the equalities R(w)Cz = R(z) and HR(~) = ~R(~)-I,
~ E If:- \ (J(H), imply
rp(Czj, R(z)g) = (AR(w)Czj, HR(z)g) - (AH R(w)Czj, R(z)g)

= (AR(z)j,HR(z)g) - (AHR(z)j,R(z)g) (25)
= ([R(z),A]j,g), j,g E DOO(H).
By Remark 4 and part (iv) of Proposition 2, we have
sup Izlll[R(z),A]II:::; Ilrpll sup IlzR(z)IIIII + (w - z)R(z)11 < 00,
zEDc,d(H) zEDc,d(H)
for all c E (0,1] and d > O. By Theorem 3, A is essentially self-adjoint on any core
of Hm.
5 Cf. [5],
Theorem 4 in [1] can be deduced from Theorem 8 by proving that if A is a

closed semibounded symmetric operator in 1i and H is a semibounded self-adjoint
operator in 1i such that for some wEe \ a(H), IIAR(w)mIVOO(H) II < 00 for
some integer m ~ 0 and the sesquilinear form ' associated with the expression
R(w)(adH)2(A)R(w) is bounded on VOO(H), then
sup IzI 211(adR(z))2(A)1I < 00, c E (0,1), d> O.
zEDc,d(H)
Reasoning similar to that in (25) enables us to show that

'(CzR(z)j,C;R(z)g) = (adR(Z))2(A)j,g), j,g E VOO(H),
which in view of Remark 9 and part (iv) of Proposition 2 gives us
sup IzI 211(adR(z))2(A)II:::; II'II sup IIzR(z)1I2111 + (w - z)R(z)1I 2 < 00.
zEDc,d(H) zEDc,d(H)
Hence, by Theorem 8, we get the essential self-adjointness of A on any core of Hm.

We conclude this paper with an example of a pair (A, H) which is covered
by our theorems but fails to satisfy assumptions of theorems in [1].
Example 17. Fix an integer m ~ 1. Let H be an unbounded self-adjoint operator
in 1i such that H ~ I. Write E for the spectral measure of H. First we show that
there exists a vector e E 1i of norm 1 such that
e tt U V(Ht). (26)
tE(O,oo)
Indeed, since the operator log H ~ It

log xE( dx) is unbounded (because the
closed support of E is unbounded), we can find a unit vector e which is not in
V(logH). This means that IlO(log x)2(E(dx)e,e) = 00. Noticing that for every
t E (0,00) there exits c(t) E (0,00) such that 6 logx ~ c(t)x t for all x E [1,00), we
see that e tt UtE(O,oo) V(Ht).
Define e e E B(1i) by
(ee)(f) = (j,e)e, j E 1i.
It is clear that the operator Ae ~ Hm + e e is positive and self-adjoint, and
V(Hm) = V(Ae). Moreover, [R(z), Ae] ~ [R(z), e e] and, in consequence,
(adR(z))k(Ae) ~ (adR(z))k(e e)
for all z E C \ a(H). Hence, by part (iv) of Proposition 2, SUPzED Ilz[R(z), Ae]11 <
00, SUPxED IIx 2(adR(x))2(Ae)1I < 00 and SUPxED Ilxm+l Jm ([R(x), Ae]R(x)m) II <
00 for [l = [lc,d(H), which means that the pair (Ae, H) satisfies all the assumptions
of our Theorems 3, 8 and 11.
It turns out that the pair (Ae, H) does not satisfy the assumption (b) of
Theorem 1 even in a weakened version proposed in the remark following Theorem 1
6E.g., expanding the series defining e cxt one can calculate c(t) = (n(t)!)l/n(t), where net) is the
least integer greater than or equal to lit.
32 D. Cichon, J. Stochel and F.R. Szafraniec
in [1]. We first show that for all 0:,,6 E [0,(0) and z E C\a(H), the sesquilinearform
rp associated with the expression IR(z)I"[H, Ae]IR(z)l,B is unbounded on Voo(H)
if and only if either 0: < 1 or,6 < 1. For f,g E Voo(H), we compute 7
rp(j,g) = (AeIR(z)l,B f, HIR(z)I"g) - (AeHIR(z)l,B f, IR(z)I"'g)
= (e e)IR(z)l,B f, HIR(z)I"'g) - (e e)HIR(z)l,B f, IR(z)I"'g)
= (IR(z)l,B f, e)(e, HIR(z)I"'g) - (HIR(z)l i3 f, e)(e, IR(z)I"'g) (27)
i1
Notice that if t E [0,(0), then HIR(z)li = 00 Iz-"'xl,E(dx) and consequently the
operator HIR(z)li is bounded if and only if t ~ 1. This implies that if 0:,,6 ~ 1,
then by (27) the form rp is bounded. Consider now the case 0: E [0,1) and suppose
contrary to our claim that rp is bounded. Since IR(z)l,B is injective, we may choose
f E Voo(H) such that (IR(z)l,B f, e) -=I- 0. It follows from (27) that the linear
functional g f---7 (HIR(z)I"g, e) is bounded on Voo(H). As Voo(H) is a core8
of HIR(z)I'" we see that e E V(HIR(z)I") = V(H 1 -,,), which contradicts (26).
Interchanging the roles of 0: and,6 (as well as f and g), we settle the case,6 E [0,1).
Similar argument shows that the sesquilinear forms associated with [H, Ae]R(z)k
and R(z)k[H, Ae] are both unbounded on Voo(H) for every integer k ~ 0. In
particular, the pair (Ae, H) does not satisfy the assumption (c) of Theorem 3
of [1].
Our next aim is to show that the pair (Ae, H) does not satisfy the assump-
tion (b) of Theorems 3 and 4 of [1] (as well as Remarks following them). More
precisely, we prove that for all z E C \ a(H) and 0:,,6 E [0,(0), the sesquilinear
form 1/; associated with the expression IR(z)I"(ad H)2(Ae)IR(z)l,B is unbounded on
Voo(H) if and only if either 0: < 2 or,6 < 2. For f,g E Voo(H), we can calculate
1/;(j, g) = (AeIR(z) 1,13 f, H2IR(z) I"g) - 2(Ae H IR(z)l,B f, HIR(zW'g)
+ (Ae H2 IR(z)l,B f, IR(z)I"g)
= (e e)IR(z)l,B f, H2IR(z)I"g) - 2(e e)HIR(z)li3 f, HIR(zW'g)
+ (e e)H2IR(z)l,B f, IR(z)I"g)
= (IR(z)l,B f, e)(e, H2IR(z)I"g) - 2(HIR(z)l,B f, e)(e, HIR(z)I"g) (28)
+ (H2IR(z)l,B f, e)(e, IR(z)I"g)
Since HIR(z)li, H2IR(zW E B(Ji) for t E [2, (0), we infer from (28) that the
form 1/; is bounded for all 0:,,6 E [2,(0). Consider now the case 0: E [0,2) and
suppose contrary to our claim that 1/; is bounded. Fix f E Voo(H) such that
a ~ (e, IR(z)l,B I) -=I- 0. Putting b = -2(e, HIR(zW I), we deduce from (28) that
the linear functional g f---7 (aH2 + bH)IR(z)I"g, e) is bounded on Voo(H). Since
Voo(H) is a core of (aH2 + bH)IR(z)la (cf. footnote 8), we obtain
e E VaH2 + bH)IR(z)la) = V(H 2-,,),
7 Notice that the spectral theorem yields IR(z)ltV=(H) <;;; V=(H) for t E [0,00).
8 Because for 9 E V(HIR(z)I"'), gn ~ E([l,n])g E V=(H) and limn~= Ilg - gnIlHIR(z)la = o.
which again contradicts (26). The proof of the remaining case f3 E [0,2) goes
through as for a E [0,2). Likewise, one can verify that the sesquilinear form as-
sociated with the expression R(z)(adH)2(A e )R(z) is unbounded on VOO(H) for
every z E C \ a(H).
Summarizing, we have argued that our Theorems 3 and 8 essentially gener-
alize Theorems 1 and 4 in [1] and, consequently, extend their applicability; the
subsequent paper will be devoted to that.
References
[1] W. Driessler, S.J. Summers, On commutators and self-adjointness, Lett. Math. Phys.
7 (1983), 319-326.
[2] E. Nelson, Analytic vectors, Ann. Math. 70 (1959), 572-615.
[3] A.E. Nussbaum, A commutativity theorem for unbounded operators in Hilbert space,
Trans. Amer. Math. Soc. 140 (1969), 485-491
[4] M. Reed, B. Simon, Methods of Modern Mathematical Physics; II, Fourier analysis,
self-adjointness, Academic Press, New York, 1975.
[5] J. Stochel, F.H. Szafraniec, Domination of unbounded operators and commutativity,
J. Math. Soc. Japan 55 (2003), 405-437.
Dariusz Cichon
Instytut Matematyki
U niwersytet J agiellonski
ul. Reymonta 4
PL-30059 Krakow
e-mail: cichon<!lim.uj.edu.pl
Jan Stochel
Instytut Matematyki
ul. Reymonta 4
PL-30059 Krakow
e-mail: stochel<!lim.uj.edu.pl
Franciszek Hugon Szafraniec
Instytut Matematyki
ul. Reymonta 4
PL-30059 Krakow
e-mail: fhszafra<!lim.uj.edu.pl
Operator Theory:
The Problem of the Finiteness of the

Point Spectrum for Self-adjoint Operators.
Perturbations of Wiener-Hopf Operators
and Applications to Jacobi Matrices
Petru Cojuhari
Abstract. The purpose of this paper is to study the point spectrum for the
case of self-adjoint operators. Based on the direct methods of the perturbation
theory, sufficient conditions for the finiteness of the point spectrum of some
self-adjoint operators are established. The problem is treated for the general
case of abstract operators and then, as applications, some concrete classes of
operators are investigated. In particular, operators associated to Jacobi type
matrices are also considered.
Mathematics Subject Classification (2000). Primary 47A55; Secondary 47A75.
Keywords. Spectral theory, compact perturbation, Wiener-Hopf operators,
Jacobi matrices.
1. Introduction
The paper is devoted to the problem of the finiteness of the point spectrum for
the case of self-adjoint operators. Within the framework of abstract operators the
problem can be formulated as follows. Let A and B denote symmetric operators
on a Hilbert space Jt, and let A be an interval of the real axis which is free of
eigenvalues of A. Let us consider the operator B as an additive perturbation of the
operator A. Our purpose is to find conditions under which the set of perturbed
eigenvalues (counted according to their multiplicities) from A is at most finite.
This problem as well as any other one connected with the study of the structure
of the spectrum of an operator represents one of the most important problems of
spectral analysis and its applications. Such a problem frequently appears in various
principle situations from such domains as theoretical and mathematical physics
(especially quantum mechanics), theory of differential (and pseudo-differential)
operators, eigenfunction expansions theory and, in the general, spectral operator
theory itself, etc. A good deal of background material on the development and
36 P. Cojuhari
perspectives of the problem can be found in [25], [15] (see also the references
quoted there). It should be pointed out the works [21], [22] (see also [23],[26])
dedicated to scattering theory, from which the necessity to study the structure of
the point spectrum in some basic questions becomes clear. The main results of
this paper are obtained from the same point of view based on the perturbation
theory of operators. First, we state an abstract result on the finiteness of the point
spectrum for some perturbations of symmetric operators. This abstract result is
given by Theorem 2.1 (see also Corollary 2.2) from Section 2. Then, combining the
theory of Wiener-Hopf type operators developed as in [18] with the general results
from Section 2, in Section 3 we study the point spectrum of perturbed Wiener-
Hopf abstract operators. It is worth to emphasize the role of the abstract Hardy
type inequalities from [13] used in the proof of the main results of Section 3. The
corresponding results concerned perturbations of Wiener-Hopf discrete operators
are presented in Section 3. Finally, in Section 4 the spectra of operators generated
by band Jacobi matrices are investigated. The main results of this section are
derived from those of Section 3 as direct applications.
2. The abstract results

In this section we cite an abstract result on the finiteness of the point spectrum
(i.e., the set of eigenvalues, including those contained in the continuous spectrum)
for self-adjoint operators. The detailed prove of it together with some applications
can be found in our earlier work [3] (see also [8] for applications to Dirac operators).
On the basis of this abstract approach there are proved the main results considered
in the next sections.
In the sequel, H will denote a complex Hilbert space. We denote by IB(H) the
space of all bounded operators on H and by IB(X) (H) the subspace of IB(H)consisting
of all compact operators in H. The domain and the range of an operator A are
denoted by Dom(A) and Ran(A), respectively. The resolvent set of A is denoted
by p(A) and the spectrum by a(A). ap(A) stands for the point spectrum of A. The
resolvent operator (A - zI)-l, z E p(A), will be denoted briefly by R(z; A).
Theorem 2.1. Let A and B be symmetric operators in a space H and let the op-
erator A has no eigenvalues on a closed interval A of the real axis. Suppose that
there exists a operator-valued function T()..) defined on the interval A having the
properties that
(i) T()") E IB(X)(H) ().. E A),
(ii) T()..) is continuous on A in the uniform norm topology, and
(iii) for each)" E A and for each u E Dom(B) such that Bu E Ran(A -)..I) there
holds the following inequality
II (A - )..I)-1 Bu 11:::;11 T()..)u II . (2.1)
Then the point spectrum of the perturbed operator A + B on the interval A consists
only of finite number of eigenvalues of finite multiplicity.
Finiteness of Point Spectrum 37
As we already mentioned this theorem (in a slightly different form) is con-

tained in our work [3] (see also [8]).
In particular, an immediate consequence of Theorem 2.1 is the following result
often useful for concrete applications.
Corollary 2.2. Let A and B be symmetric operators in H, and let A denote a
closed interval of the real axis such that the set O"p(A) n A is empty. Suppose that
the operator B can be represented as B = ST, where T is a compact operator
from H to another Hilbert space HI and S is a closed operator from HI into H,
respectively. In addition, if the following estimate holds
II (A - )..I)-ISu II:::; c II u II (c = const;).. E A),
whenever u E Dom(S) and Su E Ran(A - )..I), then O"p(A + B) n A is a finite set
and each possible eigenvalues of the perturbed operator A + B from A has a finite
multiplicity.
We will apply Theorem 2.1 just as Corollary 2.2 to the study of the problem
of the finiteness of the perturbed eigenvalues for the concrete classes of operators
considered in the forthcoming sections. We mainly study the eigenvalues contained
in the continuous spectrum of a given operator. However, Theorem 2.1 and Corol-
lary 2.2 can be also applied to the study of the discrete part of the spectrum. For
instance, in view of this remark, Theorem 2.1 (more exactly, Corollary 2.2) implies
the following perturbation theorem.
Theorem 2.3. Let A be a self-adjoint operator in H, A = (a, b) C p(A), a ~
O"p(A) and B be a symmetric operator with Dom(B) J Dom(A). If the operator
(A - aI)-I B is densely defined and compact in H, then the perturbed operator
H = A + B with Dom(H) = Dom(A) is self-adjoint in H, the spectrum of H on
A is only discrete and a is not an accumulation point for the set O"(H) n A.
In order to show how this theorem can be obtained from Corollary 2.2 denote
by T the complete extension (which evidently is unique) of the operator (A -
aI)-I B. Then T* is an extension of B(A - aI)-I, and it is easy to verify that
BR()"; A) = T* + ().. - a)T* R()..; A)()" E p(A)).
Consequently, since T* is compact, also BR()"; A) is compact for each)" E p(A),
and so a result from [16] can now be applied to conclude that the operator H is
self-adjoint and its spectrum on A is only discrete.
Next, let S = A - aI. Then the operator B can be expressed as B = ST and,
since
II R()"; A)Su 11=11 R()"; A)(A - aI)u II:::; c II u II (c = const)
holds for each u E Dom(A) and for each)" nearly of a from the right, it follows
that all of hypotheses of Corollary 2.2 are fulfilled.
The above theorem is given in [2]. Among other applications, in [2] it has
been studied the spectrum for an equation of radiation energy transfer. In this
context we note the works [4], [7] (see also [5] and [10]) for further results and
38 P. Cojuhari
other applications. We also note the discussion undertaken in [4] (see also [7]) on
the connection with related results from the works [1], [20].
3. Perturbations of Wiener-Hopf abstract operators

In this section we study the point spectrum of the operators obtained by perturba-
tions of Wiener-Hopf abstract operators. The concept of the Wiener-Hopf abstract
operators has been given by I. C. Gohberg and I. A. Feldman [18], see also S. Pross-
dorf [24]. According to the theory developed in [18], Wiener-Hopf type operators
can be regarded in a certain sense as functions of one-sided invertible operators.
As our case is that of Hilbert's space, the considered operators will be presented as
functions of an isometric operator. Moreover, these functions (so-called symbols of
the corresponding operators) will be assumed to be continuous on the unit circle
of the complex plane.
In order to specify the notion connected with the foregoing discussion, let 1i
be an arbitrary Hilbert space and consider on it an isometric operator denoted
by V. In what follows, it will be always assumed that the operator V* has no
eigenvalues on the unit circle 'JI' = {z E ([/Izl = I}. Then, it is clear that the point
spectrum of V is empty and, hence, there exist the closed and unbounded operators
(V - z1)-l and (V* - z1)-l for all z E 'JI'. We denote, as in [18], by ~(V) the closed
hull of all operators vn(n = 0, 1,), where vn = (v*)-n(n = -1, -2, ... )
and the closure is taken with respect to the operator norm of JR(1i). In line with
approaches from [18], it follows that to each A E ~(V) corresponds a complex-
valued function A(z) continuous on the unit circle 'JI'. This function A(z), z E 'JI',
is called the symbol of the operator A. We allow ourselves to write (formally for
the moment at least) A = A(V). The set of symbols which correspond to all
operators from ~(V) coincide with the set C('JI') of all complex-valued functions
continuous on 'JI' (see [18], p. 34). If AU E C('JI') is a real-valued function, then
the corresponding operator A E ~(V) is self-adjoint and the spectrum of A is
the set of all values attained by A(z), z E 'JI'. In spite of the fact that the most
results presented below can be adapted effortlessly for the general case, we will be
concerned exclusively with the case of operators with rational symbols. However,
this will be entirely enough for our applications from the next sections.
Thus, in what follows, we consider an operator A E ~(V) of the following
form
n
A -- '~
" aVj
] , (3.1)
j=-n
where aj(j = 0, 1, , n) are fixed complex numbers. As it was already men-
tioned the symbol of A is the polynomial A(z) = L:~=-n akzk, and if A(z) rep-
resents a real-valued function on 'JI' or, equivalently, aj = 7Lj(j = 0,1, ... , n),
then A is a self-adjoint operator on 1i. Henceforth this property will be always
assumed. Note that the spectrum of A is a closed interval, namely a(A) = [a, b],
where a = minA(z) and b = maxA(z) on 'JI'.
Before formulating the main result of this section it will be convenient to

recall some notions and definitions.
Let C be a symmetric operator in 'H. The operator C is called semi-bounded
from below, if there exists a number ry,ry E ffi., such that (Cu,u) 2: ry I U 112 for
each u E Dom(C). For the later, we denote by ry(C) the greatest lower bound of
C, i.e.,
ry(C) = inf{(Cu, u)/u E Dom(C), II u 11= I}.
The operator C is said to be nonnegative (C 2: 0) if 1'(C) 2: o. If ry(C) 2: 0, but
(Cu, u) > 0 for each u E Dom(C), u -I- 0, the operator C is said to be positive.
Finally, C is said to be positive definite (C > > 0) if ry( C) > O. In a similar manner
can be understood the notion of an operator semi-bounded from above and all the
corresponding notions related with it.
Next, let us consider an operator J on the space 'H for which the following
properties are assumed.
(i) J is a bounded and positive operator;
(ii) V*(Ran(J)) c Ran(J) and there is a definite operator C (so that either
C 0 or C 0 ) such that C = J- 1 - V J- 1 V* on the set Ran(J).
Remark 3.1. The existence of an operator J for which the properties (i) and (ii) are
satisfied assures applicability of obtained results in [13] (see also [14]) on abstract
Hardy type inequalities. It turns out that, the following inequality holds
ell Ju 11:: ;11 (V - zI)u I (3.2)
for all u E H and all z E ee, I z 12: 1 ,
where c = ~ I z I ry(C). In addition, we
note that from the inequality (3.2) it can be extracted some useful information.
For instance, it follows that the operator V has no eigenvalue on the unit circle 11'.
Indeed, if Vu = zu for some z E 11', then (3.2) implies Ju = 0 and, since J > 0,
we get u = O. Moreover, the unbounded operator (V - zI)-l, where z E 1I',becomes
to be a bounded one by multiplying it with J from left. More precisely, for every
z E 11' the operator J(V - zI)-l, which is densely defined, has a bound extension
on H. This fact will be systematically used in the proof of the main results given
below (see Theorems 3.2 and 3.3).
Now, let us consider an operator A E ~(V) of the form (3.1), and denote
by A(z), z E 11', its symbol, and let, as before, a = minA(z) and b = maxA(z)
on 11'. It was also noted that a(A) = [a, b]. It is well known, but this fact will be
also clear from the proof of Theorem 3.2 given below, that the operator A has no
eigenvalues.
Next, let n(A) be the number of all zeros of the polynomial A(z) - A(a :::; A:::;
b), which belong to the unit circle 11' (and counted according to their multiplicities),
and let m(A) be the maximal multiplicity of them. It is clear that n(A) is a piecewise
constant function on the interval [a, b], and the set N of its discontinuous points
is finite.
Let B be another self-adjoint operator on 'H and let us consider it as a
perturbation of the operator A. Let A be an interval contained in the continuous
40 P. Cojuhari
spectrum of A. Our purpose is to find conditions under which the set of perturbed
eigenvalues (counted according to their multiplicities) from A is only finite. We
start by considering the situation in which m(A) = 1 on A.
Theorem 3.2. Let A be an operator of the form (3.1), let AO E (a, b) \ Nand
m(Ao) = 1. In addition, let J be an operator on Ji for which the above properties
(i) and (ii) are satisfied. If the operator B E B(Ji) is self-adjoint and the operator
J- 1 B is compact in Ji, then AO is not an accumulation point for the set of eigen-
values of the perturbed operator A + B. Each possible eigenvalue of A + B in a
neighborhood of AO has a finite multiplicity.
Proof. It is clear that, since m(Ao) = 1, also m(A) = 1 and n(A) = n(Ao) for A
belonging to a closed neighborhood A of AO. Then the symbol A(z) - A can be
represented as follows
no
A(z) - A = Z)z-1 - aj(A))A>.(z)(A E A), (3.3)
j=1
where no = n(Ao), aj(A)(laj(A)1 = l;j = 1, ... , no) are continuous functions on
A, aj(A) -I ak(A)(A E A; j -I k, j, k = 1, ... ,no), A>.(z) is a polynomial in Z and
depending continuously on A, and such that A>.(z) -I 0 for 1Z 1= 1 and A E A.
Clearly, for each fixed A E A, A>.(z) can be regarded as a symbol of an operator, let
it be denoted by A>.(V), from ~(V). From the above factorization (3.3) of A(Z)-A
it follows that
no
A - AI = II (V* - aj(A)I)A>.(V)(A E A). (3.4)
j=1
According to the restrictions V, the operators V* - aj(A)I (j = 1, ... ,no) are one-
to-one, and their corresponding inverses Rj(A) = (V* - aj(A)I)-1 (j = 1, ... ,no)
are closed and unbounded operators in the space Ji. Henceforth, for convenience,
we let R(A) = II;!1 Rj(A) (A E A). Also, it is easy to see that the operator
A>. = A>.(V) is invertible on the left, and let A~-1) (E B(Ji)) denote its left inverse.
Since in addition the operators V* - aj (A)I (j = 1, ... , no) are commutative it
follows from (3.4) that
(A - AI)-1U = A~-1) R(A)U(A E A) (3.5)

for every u E Ran(A - AI).
Next, it will be shown that all of hypotheses of Theorem 2.1 are satisfied. To
this end, note that since the values aj(A) (j = 1, ... , no) are pairwise distinct for
every A E A, there exist some functions aj (A) (j = 1, ... , no), which evidently can
be supposed to be continuous, such that
no no
II (z - aj(A))-1 =L aj(A)(z - aj(A))-1(A E A). (3.6)
j=1 j=1
It turns out that the relation (3.6) implies that
Dom(R(A)) = nDom(Rj(A))(A
no
j=l
E A), (3.7)
and, respectively,
no
R(A)U = L aj(A)Rj(A)U(A E A) (3.8)
j=l
for every U E Dom(R(A)).
In order to show that, we rewrite the relation (3.6) as follows
no
L aj(A) II (z - OOk(A)) = l(A E A)
j=l k#j
from which, we readily get
no
L aj(A) II (V* - OOk(A)I) = J(A E A). (3.9)
j=l k#j
Now,let U E n7~1 Dom(Rj(A)), i.e., for each j = 1, ... , no there exists Uj E H
such that U = (V* - OOj(A)I)Uj. Then, by (3.9),the element U can be expressed in
the form
no
U= II (V* -
OOj(A)I)V,
j=l
where v = L7~1 aj(A)Uj. Hence U E Dom(R(A)), and so n7~1 Dom(Rj(A)) C
Dom(R(A)). The opposite inclusion is evident, and thus the relation (3.7) is estab-
lished. Now, it is clear that the relation (3.8) is a simple consequence of (3.9).
As has been mentioned above (see Remark 3.1), the operator J is related to
V in such way that for arbitrary z E 1I' the densely defined operator J(V - zI)-l
has a bound extension on H. Moreover, the norm of each such extension is less
then 21'-1, where l' = 1'( C) (cf. the property (ii) for J). It then follows by duality
that the operator (V* - zI)-l J, for each fixed z E 1I', is bounded on its domain.
Namely, there holds the following estimate
I (V* - zI)- l Ju II::; 21'-1 I U II (z E 1I'), (3.10)
for all U E H such that Ju E Ran(V* - zI).
Next, let U be an arbitrary element in H such that Ju E Ran(A - AI). Then,
as is clear from (3.4), also Ju E Dom(R(A)), and further from (3.8), by virtue of
(3.10), one obtains
no
I R(A)JU II::; 21'-1(L I aj(A) I) II U I (A E A). (3.11)
j=l
But aj(A) (j = 1, ... , no), as continuous functions on a closed interval A, are
uniformly bounded on A. The uniformly bounded ness of operators A~-l) as A
42 P. Cojuhari
ranges over A is also clear. Consequently, by (3.5) and (3.11), we find that there
exists a finite positive constant c independent of A and u such that
I (A - AI)- l Ju II::; c II u I (A E A).
Thus, taking S = J and T = J- 1
B, all hypotheses of Theorem 2.1 (or Corollary
2.2) are satisfied, and therefore, the proof of our theorem is complete. D
Let us now consider the case m(Ao) > 1, where AD E (a, b)\N as above. Let
D:k (k = 1, ... ,p) be all distinct roots of the polynomial A( z) - AD contained on the
unit circle 11', and let mk (k = 1, ... ,p) designate their corresponding multiplicities.
For simplicity, we let p = 2. The arguments for the general case are similar. In
this case a representation as was given by (3.3) also occurred. Obviously, it can
meaningfully asserted that for any A, A i=- AD, from a certain closed neighborhood A
of AD the roots D:j (A) (j = 1, ... , no) in that representation are simple. Moreover,
D:j(A) (j = 1, ... , no) can be enumerated in such way that the first m1 of them to
converge to D:1 as A ----+ AD and, respectively, the others m2 (= no - md to D:2.
Next, it will be shown that the abstract results from the previous section can
also be applied. To this end, it will be sufficient for the perturbation operator B to
indicate a factorization of the form B = ST, in which S E lB(H), T E lBoo (H) and,
in addition, the operator S to be chosen in such way that the following estimate
I R(A)SU II::; c I u II (A E A) (3.12)
holds for each u E H such that Su E Dom(R(A)). Here c is a constant independent
of A and u, and it is used the notation R(A) = TI7!1 Rj(A) and Rj(A) = (V* -
D:j(A)I)-l (j = 1, ... , no) as before. It turns out that an operator S suitable for
our purposes can be useful S = Jm o , where ma = m( AD), and the operator J is
chosen that the following estimates hold
cT II r u 11::;11 r- 1 (V - zI)u II (7 = 1, ... , mo) (3.13)
for all u E Hand Z E 11', where CT are some positive constants. In order to see this,
we choose polynomials Pj(z, A)(j = 1,2) in Z and depending continuously on A for
A E A such that (cf. the decomposition (3.6) )
no m,
II (Z-D:j(A))-l = P1(Z, A) II (Z-D:j(A))-1+ p2 (z, A) II
no
(Z-D:j(A))-l(A E A).
j=l j=l
An argument like that leading to (3.8) now implies that
(3.14)
j=l j=m,
whenever u E Dom(R(A)).
Further, by means of (3.13), it is easily verified that
I (V - zlI) ..... (V - zmI)u 112 C1 . . . . . Cm I Jmu II (3.15)
for u E H, Zl, ... , Zm E 11', m ::; ma and C1, ... , Cm determined as in (3.13).
Using this fact, it can be obtained by some simple manipulations estimates

for the operator-valued functions
no
II R j (A) II
ml
F1 (A) = and F2 (A) = Rj(A)S for A E A.

j=l
Namely, for every u E H such that Su E Dom(R(A)), one can be obtained

II Fj(A)U II:::; K j II u II (j = 1,2; A E A)
with constantsKj(j = 1,2) independent of A and u. Taking this into account and
the fact that the operator-functions Pj(V*, A) (j = 1,2) are uniformly bound on
A, it follows from (3.14) the desired estimate (3.12).
Now, assuming that the operator J is one-to-one and the operator T =
J-moB is compact in the space H, we get a factorization B = ST, where all
hypotheses of Corollary 2.2 are fulfilled.
The above discussion it is summarized by Theorem 3.4 from below. Before
formulating this theorem we make the following remark.
Remark 3.3. If an operator J has the following properties
(i) J is a bounded and positive operator in H,
(ii)r V*(Ran(J2r-1)) C Ran(J2r-1)(r = 1, ... ,mo),
and
(iii)r for each r = 1, ... , mo there is a definite operator Sr,r-1 {let Sr,r-1 0,
for instance} such that
Sr,r-1 = J- 1 - J r - 1V J- 2r +1V* J r - 1
on the set Ran(F),
then the estimates (3.13) hold for all u E H and for all z E C such that Iz I~ 1,
where Cr = 2- 1 I z I ')'(Sr,r-1) [13].
Theorem 3.4. Let A be an operator of the form (3.1), let AO E (a,b) \ Nand
mo = m(Ao) > 1. In addition, let J be an operator on H satisfying either (3.13)
or (i), (ii)r' (iii)r from Remark 3.3. If the operator B E Jffi(H) is self-adjoint and
the operator J-mo B is compact in H, then AO is not an accumulation point for
the set of eigenvalues of the perturbed operator A + B. Each possible eigenvalues
of A + B in a neighborhood of AO has a finite multiplicity.
Remark 3.5. An argument similar to that used in proving Theorem 3.4 can be
applied to prove that if AO E Nand n(A) = n(Ao) for A E [AO - c, AO] (A E
[AO, AO + cD,where c > 0, and the operator A and B satisfy the conditions of
Theorem 3.4, then the point spectrum of the perturbed operator A + B on the
interval [AO - c, AO] ([AO, AO + cD consists only of finite set of eigenvalues of finite
multiplicities.
As a consequence of Theorems 3.2 and 3.4 there holds the following result.
44 P. Cojuhari
Corollary 3.6. Let A be a self-adjoint operator of the form (3.1) and let J be defined
as in Theorem 3.1. If the operator B E Jffi(1t) is self-adjoint and the operator J- 1 B
is compact in 1t, then the set of all eigenvalues of the perturbed operator A + B
has at most a finite number of accumulation points. Each eigenvalue of A + B can
be only of a finite multiplicity.
The assertion follows at once from the fact that the function m(A) has at
most a finite number of discontinuous points.
The following result can be also regarded as an immediate consequence of
Theorems 3.2 and 3.4 which however is important for the relevant results about
Jacobi matrices (cf. Section 5 below).
Corollary 3.7. Let A = V + V*, let B be a bounded self-adjoint operator on 1t,

and let J denote an operator defined as in Theorem 3.1.
(i) If operator J- 1 B is compact in 1t, then the set of all eigenvalues of the
perturbed operator A + B has only the end points of the essential spectrum of
A + B, i.e., 2, as accumulation points.
(ii) If however the operator J- 2 B is compact in 1t, then the point spectrum of
A + B from the essential spectrum [-2,2] is only finite.
In both cases each eigenvalue of the perturbed operator A + B has a finite multi-
plicity.
Proof. The point is that the polynomial A(z) = z + z-l, z E T, is the symbol of
the unperturbed operator A. Therefore a(A) = [-2,2]' n(A) = 2 (-2 ~ A ~ 2),
m(A) = 1 (-2 < A < 2), N :....: {-2, 2}, m(2) = 2. There remains then to use
Theorem 3.2 for proving (i) and Theorem 3.4 or, more convenient, the assertions
made in remark 3.5 for (ii). 0
4. Perturbations of Wiener-Hopf discrete operators

In this section the results obtained in the previous section are adapted to study
the discrete case of perturbed Wiener-Hopf operators.
1. In the sequel, it is considered instead of 1t the Hilbert space b (N) of all square
summable sequence ~ = (~n)' ~n E C (n = 1,2, ... ), and V denote the elementary
shift in b(N), i.e., (V~)n = ~n-l (n = 1,2, ... ; ~o = 0). V is an isometric operator
on the space l2(N), Le., V*V = I, and we recall that the operator V* (and V as
well) has no eigenvalue on unit circle T. The operators of ~(V) set up the class of
discrete Wiener-Hopf operators (see [18], for instance). We remark that an operator
A from the class ~(V) is generated in the space l2(N) by a matrix [aj-k]f, where
aj(j = 0, 1,) stand for the Fourier coefficients of the symbol A(z) of A. In
what follows, as in previous section, it is assumed only the case in which a finite
number of Fourier coefficients aj are different of zero. In other terms this means
that the symbol A(z) of A represents a polynomiaIA(z) = E7=-n ajz j , z E T.
Thus, as in Sections 3, we consider an operator A E ~(V) of the form (3.1), where

aj (j = 0, 1, , n) are complex number such that aj = (Lj (j = 0,1, ... , n).
Let B denote a bounded self-adjoint operator on b(N) defined by B = [bjkli'"
and let us consider it as a perturbation of the operator A. We will preserve the
notations introduced in Section 3.
Theorem 4.1. Let A and B be defined as above, let Ao E ( a, b) \N and mo = m( AO)'

If Bo = [jmObjkli'" E lffi oo (l2(N)), then AO is not an accumulation point for the set
of eigenvalues of the perturbed operator A + B. Each possible eigenvalue of A + B
in a neighborhood of Ao has a finite multiplicity.
Proof. Consider the operator J defined on l2(N) by

(J~)n = n-I~n' (n = 1,2, ... ; ~ = (~n) E b(N)).
In view of the Hardy inequality ([19], Theorem 326; also, for more general results,
see [13]), one gets
II JT(V - zI)-T~ II::::: 2T II ~ II, (~ E Ran(V - zIr; z E 1I', T > 0).
Therefore, the operators F (V - zI) -T (z E '][', T > 0) have bounded extensions
on l2 (N), hence the operators (V* - zI) -T JT (z E '][', T > 0) are also bounded on
l2(N), and so, taking into account that J-moB = Bo, for the case of elementary
shift V in l2(N) the operator J together with B = [bjkJi'" E lffi(l2(N)) satisfy all the
conditions of Theorem 3.4. 0
The next theorem stands out as an alternate version of Corollary 3.6 for the
considered discrete case of operators.
Theorem 4.2. Let A be a self-adjoint discrete Wiener-Hopf operator of the form

(3.1) and let B denote a bounded self-adjoint operator on l2(N) defined by B =
[bjkJi"'. If BI = [jbjkli'" E lffioo(b(N)), then the point spectrum of the perturbed
operator A+B has at most a finite number of accumulation points. Each eigenvalue
of A + B can be only of a finite multiplicity.
It is also clear the reformulation of the results given by Corollary 3.7 concern-
ing the concrete case in which the unperturbed operator A has the special form
A = V + V*. Unconditionally, the correspond results can be derived straightly
from above Theorems 4.1 and 4.2.
Corollary 4.3. Let B denote a bounded self-adjoint operator on l2(N) defined by

B = [bjkli"'.
(i) If BI = [jbjkli'" E lffioo(l2(N)), then the point spectrum of the perturbed oper-
ator Al = V + V* + B has only the points 2 as accumulation points.
(ii) If however B2 = [Pbjkli'" E lffi oo (l2(N)), then the set O"p(AI) n[-2, 2l is only
finite.
In both cases each eigenvalue of the perturbed operator Al has a finite multiplicity.
46 P. Cojuhari
Remark 4.4. In our earlier work [12], we proved results on the absence of eigenval-
ues for perturbed Wiener-Hopf discrete operators (see also [11] for related results
concerning operators generated by Jacobi matrices) . We note that the conditions
on the perturbation are in a sense close to the ones stated above (see Theorems
4.1 and 4.2). However, it should be noted that they depend not only on maximal
multiplicity of the zeros of the symbol, but also on their number, the position of
roots of the symbol in the complex plane, etc.
2. The methods of Section 3 can be applied to obtain similar results by considering
other type of isometries. The next consideration can be regarded as typical in this
respect. Let us consider the case when 1-l is the space L 2(IR.+) and the operator
V is the translation operator in L 2(IR.+) defined by (Vu)(x) = u(x - h) for x > h
(h> 0) (taking u(x - h) = 0 for x < h).
It is easy to see that an operator A E ~(V) of the form (3.1) is given in
L2(IR.+) by
n
(Au)(x) = L aju(x - jh),
j=-n
where aj(j = 0, 1, ,n) are complex numbers. As above, the conditions for
self-adjointness of A, i.e., aj = (Lj(j = 0,1, ... , n) are always assumed. So, let A
be a self-adjoint operator defined as above and let us consider a perturbation of
A given by an integral operator
(Bu)(x) = 100
b(x, y)u(y)dy,
where b(x, y) is a measurable function with respect to both variables x, y E IR.+.

Suppose that the operator B is also self-adjoint and bounded on L2(IR.+).
Let us represent only a version of Theorem 3.4 in which the following situation
AO tJ. N, mo = m(AO) > 1 is considered.
Theorem 4.5. If the integral operator with a kernel (1 + x)mob(x, y) is compact in
the space L2 (IR.+), then AO is not an accumulation point for the set of eigenvalues of
the perturbed operator A + B. Each possible eigenvalue of A + B in a neighborhood
of AO has a finite multiplicity.
We note that the operator J 1 defined in L 2(IR.+) by
(J1)(x) = (1 + X)-lU(X)(U E L 2(IR.+))
plays the role of J in the arguments of Section 3.
Remark 4.6. It is worth to note that the methods developed in Section 3 can be
also applied to study the point spectrum for perturbation Wiener-Hopf integral op-
erators. This is the case in which the operator V defining the unperturbed operator
is nothing but the canonical shift with respect to the system of Laguerre functions
(see [18], p. 69). We cite our earlier work [6] in which results on the finiteness of
the point spectrum for perturbed Wiener-Hopf integral operators are presented.
5. Jacobi matrices
In this section we study the point spectrum of the operators generated by Jacobi
matrices. The main results are obtained by applying directly the results from the
previous section concerning the perturbations of Wiener-Hopf discrete operators.
1. We first focus our attention on the problem of the finiteness of the point spec-
trum for so-called band Jacobi matrices (or, in other terms, for generalized Jacobi

matrices). We recall that a matrix A = [ajk]f with ajk E C (j, k = 1,2, ... ) is
said to be a band Jacobi matrix of order 2n if ajk = for 1j - k I> n, where n
is a positive integer number. In the following, it is as ever considered the case of
self-adjoint operator A on the space l2(N), that is, ajk = akj (j, k = 1,2, ... ).
Furthermore, we assume that
aj+T,j ----+ ar(j ----+ 00; r = 0, 1,'" ,n).
Then the operator A can be written as a sum A = Ao + B, where

n
Ao= L a j V j (aj=a_j,j=O,l, ... ,n) (5.1)

j=-n

and B = [bjkJr with bjk = ajk - aj-k for 1j - k I~ nand bjk = for 1j - k I> n.
It is clear from (5.1) that Ao is a Wiener-Hopf discrete operator with a
rational symbol Ao(z) = 'L,]=-n ajz j (z E 'IT'). Thus, the spectrum of the operator
Ao is the interval a(A) = [a, b], where a = min Ao(z) and b = maxAo(z) on'IT'.
In the next theorem there are used the notations similar those introduced
in Section 3 before Theorem 3.2. n(A) designates for the number of all zeros of
Ao(z) - A(a ~ A ~ b) belonging to 'IT' (counted according to their multiplicities),
m(A) stands for the maximal multiplicity of them, and N designates for the set of
discontinuous points of the function n(A) on [a, b].
Theorem 5.1. Let Ao E (a, b)\N and mo = m(Ao). If
lim imobj+T,j = 0, (j = 0, 1,'" ,n),

J---+(X)
then Ao is not an accumulation point for the set of eigenvalues of the operator A.
Each possible eigenvalue of A in a neighborhood of Ao has a finite multiplicity.
The assertion is an immediate consequence of Theorem 4.1.

For a band Jacobi matrix A of order 2n there holds the following result.
Corollary 5.2. If Ao rt N and if
lim j 2n bj+T,j = 0, (r = 0, 1,'" ,n),

J---+(X)
then the point spectrum of A in a neighborhood of Ao is only a finite set.
Theorem 4.2 in turn implies the following result.

48 P. Cojuhari
Theorem 5.3. If
.lim jbj+r,j = 0, (r = 0, 1, ... ,n),
J-->OC
then the point spectrum of A has at most a finite set of accumulation points.
Moreover, each eigenvalue of A can be only of a finite multiplicity.
2. In the case of an ordinary Jacobi matrix, i.e., when n = 1, the results given by
Theorem 5.1 and Corollaries 5.2 and 5.3 can be refined. There will be no loss of
generality in supposing that the operator A is defined in the space h(N) by the
following Jacobi matrix
0 al 0 )
al 0 a2 .. .
A= (
o a2 0 ... '
. . . . ..
where aj E lR. (j = 1,2, ... ), aj --+ 1 (j --+ (0).
In what follows we denote
bj = aj - 1, Cj =1 bj 1+ 1bj+l 1(j = 1,2, ... ).

Clearly A = Ao + B, where Ao = V + V* and B = A - Ao E lBoc(h(N)).
Thus,
in accord with our notations, Ao(z) = z + z-l, z E 1[', O"(Ao) = [-2,2]' n(>') =
2( -2 :::; >. :::; 2), m(>') = 1 (-2 < >. < 2), m(2) = 2 and N = {-2, 2}.
)
Theorem 5.4. If the operator defined by
Cl
...
...
c= ( 0
o .. , '
is compact in the space h(N), then on the continuous spectrum [-2,2] is contained
at most a finite set of eigenvalues of the operator A. Moreover, each possible eigen-
value of A has a finite multiplicity.
The proof of Theorem 5.4 is given in our work [9]. Also, in this work it is
proved the following result.
)
Theorem 5.5. If the operator defined by
...
...
... '
is compact in the space 12(N), then the accumulation points for the set of all eigen-
values of A can be only the end points of the continuous spectrum of A, i. e., >. = -2
or>. = 2.
Remark 5.6. The methods developed in the present paper can be applied to more
general operators involving matrix Wiener-Hopf operators (in this respect we note
the work [17] where the theory of Wiener-Hopf equations for the matrix case is
developed). By this approach, it is possible to study from the same point of view
the point spectrum for operators generated by periodic Jacobi matrices. We note the
work [4] for some results on the problem of the finiteness of the discrete spectrum
for perturbed matrix Wiener-Hopf operators and applications to periodic Jacobi
matrices.
References
[1] M.S. BIRMAN: On the spectrum of singular boundary-value problems Mat. Sb. 55,
(1961),125-174 (Russian).
[2] P.A. COJUHARI, LA. FELDMAN: On the finiteness of the discrete spectrum of a
self-adjoint operator Mat. Issled., 47, (1978), 35-40 (Russian).
[3] P.A. COJUHARI: On the discrete spectrum of a perturbed Wiener-Hopf operator
Mat. Issled., 54, (1980), 50-55 (Russian).
[4] P .A. COJUHARI: On the spectrum of a perturbed matrix Wiener-Hopf operator Mat.
Issled., 61, (1981), 29-39 (Russian).
[5] P.A. COJUHARI: On the spectrum of perturbed self-adjoint operators Izv. Akad.
Nauk MSSR Mat. 2, (1982), 51-53 (Russian).
[6] P.A. COJUHARI: On the discrete spectrum of a perturbed Wiener-Hopf integral
operator Issled. Funkts. Anal. DifJerents. Uravn., Sht. Kishinev, (1984), 69-82.
[7] P .A. COJUHARI: On the finiteness of the discrete spectrum of some pseudodifferential
operators Mat. Issled., 80, (1985), 85-97 (Russian).
[8] P.A. COJUHARI: On the point spectrum of the Dirac operator Dif. Urav., Minsk,
Dep. VINITI 24.06.87., N4611-B87, (1987), 1-17 (Russian).
[9] P.A. COJUHARI: Finiteness of the discrete spectrum of Jacobi matrices Issled. Dif-
ferents. Uravn. i Mat. Anal., 173, Sht. Kishinev, (1988).
[10] P .A. COJUHARI: On the finiteness of the discrete spectrum of some matrix pseudo-
differential operators Izv. Vyssh. Uchebn. Zaved. Mat. 1, (1989), 42-50 (Russian).
[11] P.A. COJUHARI: The absence of eigenvalues of operators close to operators generated
by infinite Jacobi matrices Izv. Akad. Nauk SSRM, Ser. Mat., 2, (1990), 15-2l.
[12] P.A. COJUHARI: The absence of eigenvalues of the perturbed discrete Wiener-Hopf
operator Izv. Akad. Nauk SSRM, Ser. Mat. 3, (1990), 26-35.
[13] P .A. COJUHARI: Generalized Hardy type inequalities and some applications to spec-
tral theory, Operator theory, operator algebras and related topics in Proc. of OT 16
Conf. Bucharest, (1996), 79-99.
[14] P.A. COJUHARI: Hardy type inequalities for abstract operators Bul. Acad. de Sht. a
Repub. Moldova Mat., 2(33), (2000), 79-84.
[15] H.L. CYCON, R.G. FROESE, W. KIRSCH, B. SIMON: Schri:idinger operators with
application to quantum mechanics and global geometry Springer- Verlag, Berlin-
Heidelberg-New York, (1987).
50 P. Cojuhari
[16] I.C. GOHBERG, M.G. KREIN: The basic propositions on defect numbers, root num-
bers and indices of linear operators Uspehi Mat. Nauk (N. S.) 12, 2(74), (1957),
43-118 (Russian).
[17] I.C. GOHBERG, M.G. KREIN: Systems of integral equations on a half line with kernels
depending on the difference of arguments Uspehi Mat. Nauk. 13, 2(80), (1958),3-72.
[18] I.C. GOHBERG, LA. FELDMAN: Equations in Convolutions and Projective Methods
for their Solution Nauka, Moscow, (1971), (Russian).
[19] G.H. HARDY, J.E. LITTELWOOD, G. POLYA: Inequalities Cambridge Univ. Press,
Cambridge, (1952).
[20] R. KONNO, S.T. KURODA: On the finiteness of perturbed eigenvalues J. Fac. Sci.
Univ. Tokyo Sect. IA Math. 13, (1966), 55-63.
[21] S.T. KURODA: An abstract stationary approach to perturbation of continuous spec-
tra and scattering theory J. d'Analyse Math. 20, (1967), 57-117.
[22] S.T. KURODA: Scattering theory for differential operators I, J. Math. Soc. Japan,
25, no. 1, (1973), 75-103.
[23] S.T. KURODA: An introduction to scattering theory Aarhus Universitet, Lecture
Note Series, no. 51, (1978).
[24] S. PROSSDORF: Einige Klassen singuliirer Gleichungen Akademie- Verlag-Berlin,
(1974).
[25] M. REED, B. SIMON: Methods of modern mathematical physics IV. Analysis of
operators., Academic Press, New York-London, (1978).
[26] D.R. YAFAEV: Mathematical scattering theory Vol. 105. Transl. Math. Monographs,
AMS, (1992).
Petru Cojuhari
Department of Applied Mathematics
AGH University of Science and Technology,
AI. Mickievicza 30
30-059 Cracow
Poland
e-mail: cojuhari@uci.agh.edu.pl
Operator Theory:
Advances and Applications, Vol. 154, 51--68
Singular Perturbations as Range Perturbations

in a Pontryagin Space
Vladimir Derkach, Seppo Hassi and Henk de Snoo
Abstract. When the singular finite rank perturbations of an unbounded self-

adjoint operator Ao in a Hilbert space 5)0, formally defined by A(a) = Ao +
GaG-, are lifted to an exit Pontryagin space SJ by means of an operator
model, they become ordinary range perturbations of a self-adjoint operator
Hoc in 5) ::J 5)0: HT = Hoc - nT-In*. Here G is a mapping from <cd into some
scale space SJ-k(Ao), kEN, of generalized elements associated with Ao, while
n is a mapping from <cd into the extended space 5), where HT is defined. The
connection between these two perturbation formulas is studied.
Mathematics Subject Classification (2000). Primary 47A55, 47B25, 47B50j Sec-
ondary 34L40, 81QlO, 81Q15.
Keywords. Singular finite rank perturbation, extension theory, KreIn's for-
mula, boundary triplet, Weyl function, generalized Nevanlinna function, op-
erator model.
1. Introduction
Let Ao be an unbounded self-adjoint operator in a Hilbert space flo and let
fl-k(A o), kEN, be the dual space of generalized elements corresponding to the
space fl+k(Ao) = dom IAolk/2 equipped with the graph norm, cf. [6]. Singular finite
rank perturbations of an unbounded self-adjoint operator Ao in a Hilbert space
flo are defined formally as
A(a) = Ao + GaG*, (1.1)
where G is a linear mapping from H = Cd into fl-k(Ao) and a is a self-adjoint
operator in H. In [15], [23] an operator model for one-dimensional singular per-
turbations of the form (1.1) was constructed by extending the space flo with a
finite-dimensional exit space flQ; see also [2], [3] for the case of finite rank fl-2-
perturbations. Several further references on this topic can be found in [3]. The
model given in [9], [11] uses a coupling method for identifying the singular finite
52 v. Derkach, S. Hassi and H. de Snoo
rank perturbations A(Ct) in (1.1) with the self-adjoint extensions HT of a symmet-

ric operator in 5) = 5)0 EB 5)Q with some natural additional assumptions on G (see
Theorem 3.1 below). It turns out that the extensions HT are in fact ordinary range
perturbations of one of the extensions, namely of the self-adjoint operator Hoo in
5) :J 5)0:
(1.2)
where n is a mapping from H into 5) and T is a self-adjoint parameter in H. The
perturbations HT in (1.2) induce a symmetric restriction S of Hoo in 5) via
domS = {F E domHoo : n* F = O},
which, due to the assumption rann c 5), is maximally nondensely defined in
5). Therefore, among the self-adjoint extensions of S there are linear relations
which are not operators. In particular, the generalized Friedrichs extension (see
[17], [18]) of S is not an operator. A classification of the perturbations HT by
decomposing the self-adjoint parameter T into its operator and multi-valued parts
leads to intermediate symmetric extensions of S and their generalized Friedrichs
extensions. These extensions of S turn out to be precisely those which are given by
the so-called extremal boundary conditions and whose compressed resolvents to
the original space 5)0 are canonical, i.e., coincide with a resolvent of a self-adjoint
relation in 5)0.
The contents of this paper are now briefly described. Section 2 contains the
necessary facts concerning boundary triplets and Weyl functions in a Pontryagin
space. A concise introduction to finite rank singular perturbations of a self-adjoint
operator in a Hilbert space is given in Section 3. Such finite rank singular pertur-
bations are identified with self-adjoint relations in a larger Pontryagin space. They
are interpreted as range perturbations in Section 4. A connection with so-called
extremal boundary conditions can be found in Section 5.
2. Boundary triplets and abstract Weyl functions

Let 5) be a Pontryagin space with negative index /"\', cf. [5]. The set of all bounded
everywhere defined linear operators acting on 5) is denoted by [5)]. If T is a linear
relation in 5), then dom T, ker T, ran T, and mul T indicate the domain, kernel,
range, and multi-valued part of T, respectively; moreover, p(T) denotes the set of
regular points of the linear relation T. Let S be a not necessarily densely defined
closed symmetric relation in 5) with equal defect numbers d+ (S) = d_ (S) < 00
and let S* be the adjoint linear relation of S, so that S c S*. Recall (see [16], [7])
that a triplet II = {H, r o, r 1 } of a Hilbert space H with dim H = n(S) and two
linear mappings r j, j = 0, 1, from S* to H is called a boundary triplet for S* , if the
mapping r = (ro, r 1)T : i---+
(roi, r1i) T from S* into HEBH is surjective and the
i
following abstract Green's identity holds for every = {j, f'}, 9 = {g, g'} E S*:
(f',g) - (f,g') = (r1i,roghi - (rOi,r1ghi = i(rg)* J(ri);

Singular Perturbations and Range Rerturbations 53
here J stands for the block operator
J = C~H -iIH )
o .
The adjoint 8* of every closed symmetric relation 8 with equal defect numbers has
a boundary triplet II = {H, r o, r 1}. All other boundary triplets IT = {H, 1\, r\}
are related to II via a J-unitary transformation W: f = wr. In particular, the
transposed boundary triplet lIT = {H, rJ, rD, is defined by rT = iJr. When 8
is densely defined, 8* can be identified with its domain dom 8* and the boundary
mappings can be interpreted as mappings from dom 8* onto H.
Let II = {H,rO,r1} be a boundary triplet for 8*. The mapping rT : 1--;
{r11, -rol} from 8* onto HffiH establishes a one-to-one correspondence between
the set of all self-adjoint extensions of 8 and the set of all self-adjoint linear
relations 7 in H via
A T := ker(ro +7rd = {1 E 8*: {rd;-ro!} E 7} = {1 E 8*: r T l E 7}.
(2.1)
When the parameter 7 is an operator in H the equation (2.1) takes the form
(2.2)
The identity 7 = 00 is to be interpreted as 7- = 0 or, more precisely, by using
1
graph notation as 7 = {O,IH}; in this case the equation in (2.2) takes the form
r 11 = o. More generally, there is a similar interpretation, when 7 is decomposed
orthogonally in terms of its operator part and multi-valued part. To each boundary
triplet II one may naturally associate two self-adjoint extensions of 8 by Ao =
ker r 0, A 1 (= Aoo) = ker r 1, corresponding to the linear relations 7 = 0 and
7 = 00 via (2.1).
Let IJh(8*) = ker (8* - ,\), ,\ E p(8), be the defect subspace of 8 and let
1}t>.(8*) := { {f>., ,\J>.} : J>. E 1)1>.(8*)}; here the notations 1)1>. and I}t>. are used
when the context is clear. Every boundary triplet II gives rise to two operator
functions defined for ,\ E p( Ao) (# 0) by the formulas
,('\) = P1 (ro r1}t>.)-1 (E [H,I)1>.]), M('\) = r 1(ro r1}t>.)-1 (E [H]). (2.3)
Here P1 denotes the orthogonal projection onto the first component of H ffi H. The
functions, and M in (2.3) are holomorphic on p(Ao) and they are called the ,-field
and the Weyl function of 8 corresponding to the boundary triplet II, respectively;
cf. [7], [13]. The function M is also the Q-function of the pair (8, Ao) in the sense
of [21]). The ,-field, T and the abstract Weyl function MT corresponding to the
transposed boundary triplet lIT are related to , and M via
,T (,\) = ,('\)M(,\)-l, M('\) T = _M(,\)-l, ,\ E P(A1) (# 0).
When S) is a Hilbert space, a Weyl function M of 8 belongs to the class
of Nevanlinna functions, that is, M is holomorphic in the upper half-plane C+,
ImM('\) 2 0 for all ,\ E C+, and M satisfies the symmetry condition M('\)* =
M(X) for ,\ E C+ U C_. In the case where S) is a Pontryagin space of negative
index "', the Weyl function M of S belongs to the class N k, k :"::: "', of generalized
Nevanlinna functions which are meromorphic on C+ UC_, satisfy M(>')* = M(5-),
and for which the kernel
NM(>',Il) = M(>'l =:(fl) , NM(>', X) = d~ M(>'), >',11 E C+,
has k negative squares [21]. If S is simple, that is,

fl = span {SJt;.(S*) : >. E p(Ao) (# 0) },
then S is an operator without eigenvalues. In this case the Weyl function M belongs
to the class NI<, Le., k = "', and the domain of holomorphy p(M) of M coincides
with the resolvent set p(Ao).
The resolvent of the extension AT and its spectrum O"(AT) can be expressed
in terms of 7 and the Weyl function M via KreIn's formula. In the terminology of
boundary triplets the result can be formulated as follows, see [12], [13], [7].
Proposition 2.1. Let S be a closed symmetric relation in the Pontryagin space fl
with equal defect numbers (d,d), d < 00, let II = {1i,ro,r 1 } be a boundary triplet
for S* with the Weyl function M, let 7 be a linear relation in 1i connected with
AT via (2.1). Then the resolvent of AT is given by
(AT - >.)-1 = (Ao - >.)-1 - /,(>')(7- 1 + M(>.))-l/,(X)*, >. E p(AT) n p(Ao).
Moreover, for every >. E p(Ao) the following equivalences hold:
(i) >. E p(AT) if and only if 7- 1 + M(>.) is invertible;
(ii) >. E O"p(AT) if and only ifker(7- 1 + M(>')) is nontrivial.
Similarly, for a (generalized) Nevanlinna family 1"(>') the function
(Ao - >.)-1 - /,(>')(1"(>') + M(>.))-l/,(X)*,
is the compressed resolvent of an exit space extension of S in a Hilbert (or a
Pontryagin) space, cf. [21], [24], [12], [7].
3. A model for singular perturbations

In a number of papers singular rank one perturbations of Ao generated by w E
fl-2n-2 have been studied by means of exit space extensions of a symmetric op-
erator S connected with Ao, see [23], [14], [15], [22]. In this section the main
ingredients for constructing a model for finite rank singular perturbations of Ao
generated by G with ranG c fl-2n-j, n > 0, j = 1,2, are given. This model
was established in [9] and further used in [11], see also [10] for a special case. The
model uses an orthogonal coupling of two symmetric operators and it is motivated
by a perturbation result concerning the extending inner product space fl ::J flo:
the resolvents associated with the perturbations of Ao should be finite rank per-
turbations of the resolvent generated in fl by (Ao - >.)-1 (see Theorem 3.1 below).
In a more general setting a similar model to give realizations for arbitrary scalar
NI<-functions has been established recently in [8].
3.1. Some operators associated with matrix polynomials

Let q be a monic d x d matrix polynomial of the form
q(A) = h{An + qn_lAn-1 + ... + qlA + qo, (3.1)
and let r be a self-adjoint d x d matrix polynomial of the form
r ( /\') = r2n-l/\,2n-l + r2n-2/\,2n-2 + ... + rl/\, + ro, rj = rj, j = 0, ... , 2n - l.
(3.2)
Observe, that the function Q given by
Q(A) = (q~~A) q(A))

r(A) , (3.3)
is a 2dx 2d matrix polynomial whose leading coefficient is, in general, noninvertible.

In fact, Q is a strict generalized matrix Nevanlinna function whose Nevanlinna
kernel has dn negative (and dn positive) squares.
Associated with the matrix polynomial q there are n x n block matrices Bq
and Cq defined by
ql q2 qn-l hi
q2 qn-l J1-{ 0
Bq = 0 0
qn-l J1-{
J1-{ 0 0 0
and
0 J1-{ 0 0
0 0 hi
Cq =
0
0 0 0 J1-{
-qo -ql -qn-2 -qn-l
Moreover, the following block matrices are needed
~q2), B,. = (rj+k+dj,k~o'

(3.4)
where
rn+l
rn ) 0
(J1-{)
D= ( : (qO,ql, ... ,qn-d- : (ro,rl, ... ,rn-d
r2n-l 0
In addition, the following vectors depending on A E C are used:
A = (I1-{, AJ1-{, ... , An - 1J1-{),
56 V. Derkach, S. Hassi and H. de Snoo
o
The main objective here is the matrix polynomial Q defined in (3.3). It determines
the structure of the exit space flQ = H n EB H n (= C2dn) used for constructing the
model for singular perturbations. The inner product in flQ is defined by the block
matrix B via (-, ')SjQ = (B,,) in which case the companion type operator C in (3.4)
becomes self-adjoint in flQ. The restriction of C to the subspace
dom SQ = { F = (}) E flQ : II = h = 0} (3.5)
defines a closed simple symmetric operator SQ in flQ with defect numbers (2d,2d).
It is maximally nondensely defined and a straightforward calculation shows that
its adjoint SQ (a linear relation in flQ) is given by
SQ = { F = { F, CF + B- 1 (~::~) } : F E flQ, <p, (j5 E H}.
It is possible to associate a boundary triplet Irq = {H EB H, r~, r~} with SQ by

defining the boundary mappings on SQ via
FE SQ.
In this case the Weyl function of SQ associated with the boundary triplet Irq
coincides with the matrix polynomial Q, cf. [9J.
3.2. A perturbation result for the resolvents

Let C be an injective linear mapping from H = Cd into the scale space fl-2n-1
generated by the self-adjoint operator Ao and let Ao be the [fl-2n+1,fl-2n-1J-
continuation of Ao. The adjoint operator C* maps fl2n+1 into H. The case where
C is a mapping into fl-2n is similar to the present case; it can be found in [l1J.
Observe, that ifran C n fl-2 = {O}, then the restriction of Ao to
dom So = dom Ao n ker C*
gives rise to an essentially self-adjoint operator So whose closure coincides with
Ao. In this case the vector RACh = (Ao - >.)-lCh, h E H\{O}, >. E p(Ao), does
not belong to the space flo, since C is injective. However, one can give a sense to
the vector RACh by extending the space flo suitably. For instance, if 0 E p(Ao),
then the vector
'Y(>')h := RACh = Aa1Ch + ... + >.n-1 Aa nCh + >.n RAAa nCh
can be considered as a vector from an extended inner product space 5) satisfying

the condition
5) ::J span {5)0, Aojran G : j = 1, ... ,n}. (3.6)
In this space the continuation Ao of Ao generates an operator, say H o , for which
the operator function ')'(>.) , >. E p(Ao), can be interpreted to form its ,),-field in the
sense that
')'(>'l- ')'(J-l) = (Ho - >.)-I,),(J-l), >., J-l E p(Ao).
-J-l
This identity implies that
d~ ')'(>.) = (Ho - >.)-1,),(>,), >. E p(Ao).
The inner product (u, 'P)Sj in 5) should coincide with the form (u, 'P) generated
by the inner product in 5)0 if the vectors u, 'P are in duality, say, u E 5)2(n-j)+I,
'P E AojranG. Now, for the other vectors in (3.6) it will be supposed that the
conditions
(AojGh, Ar/Gf) Sj = (tHk-lh, fht., j, k = 1, ... , nj h, f E H, (3.7)
are satisfied for some operators tj = t;
E [HJ, j = 1, ... ,2n - 1. The next result
shows that under such mild conditions on the extending space the structure of
perturbed resolvents becomes completely fixed under some basic assumptions on
Ho. This fact gives rise to the model presented in [9] for singular finite rank
perturbations of Ao.
Theorem 3.1. ([11, Theorem 4.8]) Let 0 E p(Ao), let ranG\{O} C 5)-2n-l \5)-2n
with kerG = {O}, and let Go = AonG. Moreover, assume that 5) ::J 5)0 is (an
isometric image of) an inner product space satisfying (3.6), (3.7), and let Hand
Ho be self-adjoint linear relations in 5) such that
(i) p(Ho) = p(Ao);
(ii) ')'(>')' = (HO->.)-I')'(>.) holds for (an isometric image of) the function ')'(>.) =
(Ao - >.)-IG, >. E p(Ao);
(iii) (H - >.)-1 - (Ho - >.)-1 = -')'(>.)a(>.)')'(>.) * , >. E p(H) n p(Ho);
for some matrix function a(>.) holomorphic and invertible for>. E p(Ho) n p(H).
Then a(>.)-1 can be represented in the form
(3.8)
where 13 = 13* E [HJ, t(>.) = h>' + ... + t2n_l>.2n-l, and Mo(>') = GoR)..G o is a
Nevanlinna function in H.
In Theorem 3.1 the function a-I can be seen as a Weyl function (or a Q-
function) of an underlying symmetric operator S. The formula for a-I in (3.8)
shows that it is a generalized Nevanlinna function and therefore in general the
operator S cannot be symmetric in some Hilbert space. The model constructed
in [9] for S uses a coupling method resulting in a Pontryagin space 5) such that
5 becomes symmetric in Sj. The construction of the model space via the coupling
method is briefly recalled in the next subsection.
Note that the condition 0 E p(Ao), which was assumed for simplicity, leads
to the particular form of 0-(,X)-1 in (3.8). Other invertibility conditions on Ao lead
to the more general form of 0-(,X)-1 in (3.9).
3.3. The model

Let So be a closed symmetric operator in a Hilbert space Sjo with defect num-
bers (d, d) and the Weyl function Mo. Let 5 Q be the symmetric operator in the
Pontryagin space SjQ = C dn EEl C dn defined as the restriction of C to (3.5). The
next theorem (cf. [9]) gives a symmetric linear relation 5 in the Pontryagin space
Sj = Sjo EEl SjQ = Sj EEl (C dn EEl cdn) as a coupling of the operators So and 5 Q , such
that the following function is a Weyl function for 5:
M('x) = r('x) + q~('x)Mo(,X)q(,X). (3.9)
Here the matrix polynomials q and r are as in (3.1) and (3.2).
Theorem 3.2. ([9, Theorem 4.2]) Let So be a closed symmetric operator in the
Hilbert space Sjo and let no = {H, rg, rn be a boundary triplet for So with the
Weyl function Mo and the ,,(-field "(0. Let q, r, and Q be as in (3.1), (3.2), and
(3.3), respectively. Then:
(i) The linear relation
io = {fo, f~} E So }
f,l E C dn
h = r~io,h = 0
is closed and symmetric in Sjo EEl SjQ and has defect numbers (d, d).
(ii) The adjoint 5* is given by
io = {fo,f~} E So }
f,l E Cdn,;P E Cd .
h = r~io
(iii) A bounda'ry triplet n = {H,ro,rd for 5* is determined by
ro(io EEl F) = h, r 1 (io EEl F) =;p, io EEl FE 5*.
(iv) The corresponding Weyl function M is of the form (3.9) and the ,,(-field "( is
given by
"(('x)h = "(o(,X)q('x)h EEl ((AT Mo(,X)q(,X) + Ai)h +A Th), hE H. (3.10)
If the operator So is densely defined in 5)0, then 5 is an operator. When r = 0
the formulas for 5 and 5* in Theorem 3.2 can be simplified and the Weyl function
takes the factorized form
3.4. Self-adjoint extensions of the model operator

The self-adjoint extensions of the model operator 8 can be parametrized by the
self-adjoint relations r in the parameter space 'It via Hr = ker(ro + rr 1 ). From
Theorem 3.2 one obtains the following explicit expressions for HT) cf. [11].
Proposition 3.3. Let the assumptions be as in Theorem 3.2, and let'Y and M be
given by (3.10) and (3.9), respectively. Then:
(i) The self-adjoint extensions Hr of 8 in fl = flo EB flQ are in a one-to-one
correspondence with the self-adjoint relations r in 'It via
f~ 10 = {fo, fndnE 8 0 }
Hr ={{ ( t,
fO)
f
c(t)
f
(
+ (rifo B-1
) }
Q9 e 1 ) :
'(J Q9 e1
f,l ~C _
h = ruo, h + r'{J = 0
.
(ii) For every A E p(Hr) n p(Ho) the resolvent (Hr - A)-1 satisfies the relation
(Hr - A)-1 = (Ho - A)-1 - 'Y(A)(r- 1 + M(A))-1'Y(.\)*. (3.11)
(iii) For every A E p(Ho) the following equivalences hold:
A E (jp(Hr) {:} 0 E (jp(r- 1 + M(A)),
A E p(Hr) {:} 0 E p(r- 1 + M(A)).
Proof. (i) The condition 10

EB F E ker(r o + rrd means that {cp,];.} E -r, or
equivalently, that];. + rcp = 0, see (iii) of Theorem 3.2. The representation of Hr
is now obtained from the formula for 8* in Theorem 3.2.
(ii) The form of the resolvent of Hr is obtained by applying Proposition 2.1
to the data in Theorem 3.2.
(iii) This statement is immediate from Proposition 2.1. o
The operator 8 0 in Theorem 3.2 is allowed to be nondensely defined in the
original Hilbert space flo. If 8 0 is densely defined in flo then 8 is an operator in the
model Pontryagin space flo EB flQ. However, even in this case the model operator
8 is not densely defined in flo EB flQ. Therefore, among the self-adjoint extensions
of 8 there are linear relations which are not operators. In fact, the following result
holds.
Proposition 3.4. The multi-valued parts of 8 and Hr are given by
(3.12)
(3.13)
and the equalities

dimmu18 = dimmulA I , dimmulHr = dim mul Al + dim ker T (3.14)
hold. In particular, Hr is an operator in 5) if and only if Al = ker r~ is an operator

in 5)0 and kerT = {O}. Moreover, Ho is the unique self-adjoint extension Hr of 8
for which the equality mul Hr = mu18* holds, and it has the representation
Ho = 8 +({O} E6 mu18*), (3.15)
where + stands for the componentwise sum of the graphs.

Proof. The form of mul Hr is straightforward to check by using the formulas for
the self-adjoint extensions Hr in Proposition 3.3. By letting 'P = 0 in (3.13) one
obtains the description (3.12) for mu18. The equalities (3.14) follow from (3.12)
and (3.13).
Moreover, by comparing mulHr with the multi-valued part of the adjoint
relation 8* (see [9]) one concludes that the condition mul Hr = mu18* is equivalent
to dim ker T = d. This means that T = 0, i.e., the only self-adjoint extension with
the maximal multi-valued part mu18* is Ho.
The representation (3.15) of Ho is now obvious. D
If the self-adjoint extension Al = ker r~ of 8 0 is an operator in 5)0, then

mu18 = {O} and Hr is an operator in 5) if and only if ker T = {O}. In view of (3.15)
the extension Ho is always multi-valued, since 8 is nondensely defined in 5). In
fact, Ho has a natural interpretation as a generalized Friedrichs extension of 8, see
[17], [18]. The representation (3.15) shows that, together with 8, Ho is maximally
nondensely defined in 5). In fact, Ho has a nontrivial root subspace ..c at 00 and,
moreover, the following results shows that the finite spectrum of Ho coincides with
the spectrum of the self-adjoint extension Ao of 8 0 in the original Hilbert space
5)0. Hence, in particular the assumption (i) in Theorem 3.1 is satisfied.
Proposition 3.5. ([11, Proposition 3.4]) Let the assumptions be as in Theorem 3.2
and let Ho = kerr o be as in Proposition 3.3 (with T = 0). Then:
(i) p(Ho) = p(Ao);
(ii) the compression of the resolvent of Ho to the subspace 5)0 is given by
PS)o (Ho - A)-Ii 5)0 = (Ao - A)-I, A E p(Ho);
(iii) the subspace = {O} E61t n E6 {O} of 5) = 5)0 E6 5)Q is maximal neutral and
invariant under the resolvent (Ho - A)-I. It satisfies (Ho - A)-n = {O},
A E p(Ho).
This result will be extended in Section 5 to a certain subclass of self-adjoint

extensions of the model operator 8 in 5) (i.e., for certain singular perturbations
of Ao).
4. Singular perturbations as range perturbations

Let Hoc = ker r 1 be the self-adjoint extension of 8 corresponding to 7- 1 = 0 in
Proposition 3.3. The self-adjoint extensions Hr of 8 in Proposition 3.3 can be seen
as "range perturbations" of Hoc in the Pontryagin space fl = flo EEl flQ, cf. [19],
[20] for the Hilbert space case. For simplicity the results in this section are stated
when Al = ker r~ is an operator in flo, which is always the case when 8 0 is densely
defined in flo. In this case Hoc is also an operator by Proposition 3.4. Introduce
o : 1{ --7 mul8* C flo EEl flQ by
nh~ (h~e"), hE 'Ii. (4.1)
In the rest of this paper the following notations will be used
F~m'G~mEMnQ
Proposition 4.1. Let the assumptions be as in Theorem 3.2 and assume that Al =
ker r~ is an operator. Then 8 is a domain restriction of Hoc given by
dom8 = {F E domHoc : O*F = O}, (4.2)
and the self-adjoint extensions Hr and Hoc of 8 in Proposition 3.3 are connected by
Hr = Hoc - 07- 1 0*, (4.3)
where the difference is understood in the sense of relations.
Proof. Since Al is assumed to be an operator, Proposition 3.4 shows that Hoc is
an operator. The adjoint 0* : flo EEl flQ --7 1{ of 0 in (4.1) is given by
(4.4)
The equality (4.2) is now clear from the formulas for Hoc in Proposition 3.3 and
for 8 in Theorem 3.2.
Let F = (fo,f,j)T,G = (gO,g,g-yT E fl. By definition, {F,G} E 07- 1 0* if
and only if {O*F, cp} = tA, cp} E 7- 1 and G = Ocp for some cp E 1i. Consequently,
{F, G} E Hoc - 07- 1 0* if and only if
{F,G}={F,HocF+Ocp}, FEdomHoc , {cp,h}E-7. (4.5)
Now using (4.1) and comparing (4.5) with the expression for H y in Proposition
3.3 the equality (4.3) follows. D
The above result depends on the fact that the model operator 8 in Theorem
3.2 is maximally nondensely defined in fl = flo EElflQ. In the case of defect numbers
(1,1) the extension Ho is the only self-adjoint extension of 8 which is not an
operator and the other extensions Hr, 7 =I- 0, are given by (4.3), cf. [10]. In the
special case of defect numbers (1,1) a result similar to Proposition 4.1 has been
obtained in [20, Theorem 3.2] for the model concerning perturbations in Sj-2.
The perturbation formula (4.3) in Proposition 4.1 gives an explicit expression
for the self-adjoint extensions HT of S. Moreover, the resolvent formula (3.11) can
be obtained by a straightforward calculation from (4.3), cf., e.g., [17]. It is also
clear from (4.3) that HT is an operator if and only if the inverse T- 1 is an operator
in H, in which case HT) kerT = {O}, is an ordinary range perturbation of the
operator Hoo in the Pontryagin space Sj. An opposite extreme case is T = O. Then
the condition {0, h} E -T in (4.5) is equivalent to F E ker 0* which together
with F E domHoo implies that F E domS, while muIOT- 1 0* = ranO. Hence,
the perturbation (4.3) for T = 0 coincides with the form of H o given in (3.15), i.e.,
with the generalized Friedrichs extension of S in Sj. A more specific classification
associated with the perturbation formula (4.3) is obtained by decomposing the
self-adjoint parameter T into its operator and multi-valued parts,
T=TsEBToo, Ts={{h,k}ET: k~muIT}, Too = {O}EBmuIT. (4.6)
Here Ts is a self-adjoint operator in Hs = dom T , Too is a self-adjoint relation in
Hoo = mul T, and H = Hs EB Hoo
Proposition 4.2. Let the assumptions be as in Theorem 3.2 and assume that Al =
ker r? is an operator. Let P be an orthogonal projection in H and define Op =
Orran P, where 0 is given by (4.1). Then:
(i) The domain restriction Sp of H oo , given by
domSp = {F E domHoo : O;'F = O},
is a closed symmetric operator in Sj whose defect numbers are given by (s, s),
s = dimranP.
(ii) The adjoint of Sp is given by
S;, = { {F, G} E S* : (I - P)0 = O}.
(iii) The self-adjoint extensions He of S with the property HT n Hoo = Sp are
in one-to-one correspondence with the parameters T for which mul T = ker P
and they are given by
(4.7)
where T = Ts EBToo is decomposed as in (4.6) and the difference is understood
in the sense of relations.
(iv) The generalized Friedrichs extension of Sp corresponds to Ts = 0 in (4.7)
and is given by
Sp +({O} EB muIS;') = { {F, G} E S* : O;'F (= ph) = 0, (I - P)0 = O}.
Proof. Let F = (fo, f,j) T, G = (gO, g, g) T E Sj. In view of (4.4), O;'F = PO*F =
Ph. Hence, Sp = { {F, G} E S* : O;'F = ph = 0,0 = O} from which the
statements (i) and (ii) easily follow.
(iii) Clearly, {F, G} E HoonHr if and only if {F, G} E S*, and the conditions
cp = 0 and {cp,,h} E -T are satisfied. Equivalently, F E domHoo and,h E mulT.
Comparing this with the condition OJ,F = P,h = 0 for Sp in (i), one concludes
that mul T = ker P. It is easy to check that for such T, the equality OT- 10* =
OPT;10j, holds (cf. the proof of Proposition 4.1). Hence, (4.7) follows from (4.3).
(iv) The discussion concerning Ho above shows that the generalized Friedrichs
extensions of Sp corresponds to Ts = 0 in (4.7), in which case {cp,,h} E -T is
equivalent to (I - P)cp = 0 and P,h = O. 0
Proposition 4.2 shows that Sp is maximally nondensely defined: dim mul Sp =
s. Clearly, the perturbation formula (4.7) is an analog of (4.3). The characterization
of operator extensions in (4.7) agrees with the one in (4.3), since kerTs = kerT.
5. The class of self-adjoint extensions with extremal boundary

conditions
According to Proposition 3.5 the compressed resolvent of Ho from fl to the Hilbert
space flo C fl coincides with the resolvent of the (unperturbed) operator Ao in flo.
In this section the corresponding property will be proved for a certain subclass of
self-adjoint extensions Hr of the model operator S. A compressed resolvent of S
in flo is said to be canonical if it coincides with the resolvent of some self-adjoint
extension A of So in the Hilbert space flo.
Proposition 4.2 shows that the generalized Friedrichs extension of the inter-
mediate symmetric extension Sp c H 00 is determined by the (abstract) boundary
conditions
PfoF = (I - P)f1F = 0, FE S*, P = P* = p2, (5.1)
where {H, fo, fd is the boundary triplet associated with S* in Theorem 3.2. In
what follows boundary conditions of the form (5.1) are called extremal boundary
conditions associated with S*, since they have an interpretation as extreme points
in the parameter space, see, e.g., [4]. When in the model of Theorem 3.2 the matrix
polynomial r = 0 and the matrix polynomial q is of the form q = ht 0 q, where q
is a monic scalar polynomial, a different description for this class of extensions of
S can be obtained by means of the compressed resolvents in flo. In fact, for these
extensions of S the following analog of Proposition 3.5 can be proved.
Theorem 5.1. Let the assumptions be as in Theorem 3.2. Let r = 0 in (3.2), let
q = ht0q, where q is a monic scalar polynomial, and letC be given by (3.4). Then
the compressed resolvent of Hr to flo is canonical if and only if Hr is given by the
extremal boundary conditions of the form (5.1). In this case T = {{Ph, (I - P)h} :
h E H} and for the corresponding Hr the following assertions hold:
(i) p(Hr) = p(Ar) n p(C) (= p(Ar)\O"(q~q)), where Ar = ker(fg + Tf~) and
T #0.
(ii) PS)o (Hr - A)-1 f flo = (Ar - A)-1, A E p(Hr).
(iii) The subspace r = 1 +2 with

1 = {a} EEl (ranPt EEl {a}, 2 = {a} EEl {a} EEl (ker p)n
is maximal neutral and invariant under the resolvent (Hr - >.)-1. Moreover,
(Hr - >.)-n1 = {a}, (Hr - >.)-I(O,O,g)T = (0,0, (Cq - >.)-Ig)T, (5.2)
where (O,O,g)T E 2 and>' E p(Hr).
(iv) The Weyl junctions Mr of (S,Hr ) and Mo,r of (So,A r ) are given by
Mr = (~ q~ 1) Mo,r (~ q~ #)
(Mll -
and
M _ MI2M:;} M21 - MI2 M :;})
O,r - _ M:;} M21 - M:;} ,
where iiI = IHs 0q, (12 = IHoc 0q, and the decomposition of the Weyl function
Mo = (Mij)~,j=1 of (So, Ao) is according to 11. = 11.s EEl 11.00 = ranP EEl ker P.
Proof It follows from Proposition 3.3 that the compressed resolvent of Hr is given
by
PSjo(Hr - >.)-1 f 5)0 = (Ao - >.)-1 -1'0(>')(7'(>.)-1 + M O(>.))-I1'O(X)*, (5.3)
where 7'(>') = q(>.)rq#(>.) due to assumption r = 0. The formula (5.3) coincides
with a canonical resolvent of So if and only if the function 7' does not depend on

>.. Clearly, this condition is satisfied if and only if rs = in (4.6), i.e., Hr is given
by the extremal boundary conditions (5.1) for some P = P* = p2.
To prove (i)-(iv) introduce the following boundary mappings
~o = pro - (I - p)r1,
{ (5.4)
r 1 = (I - P)ro + prl,
so that Hr = ker1' o and Ar = ker1'g.
(i) Let G = (gO,g,g)T E 5),10
= {fo,j~} E So, and let>. E p(Ar) n p(C).
Then by Proposition 3.3 the relation G E ran (Hr - >.) can be rewritten as a
system of equalities
(5.5)
(I - P)cp = 0.
Now, one can solve prg10
from the third equality in (5.5) by means of the com-
panion operator Cq (cf. [11]). Since>. E p(Cq~) the second equality in (5.5) gives
(I - P) 0 f = (Cq~ - >.)-1 (I - P) 0 g. In particular, (I - p)r~1o = (I - P)!I and
consequently 1'g1o has been solved. Let ;:Yr be the 1'-field for (So, Ar) associated
with the boundary triplet {1i, fg, f~} in (5.4). Then one can write fa and f~ in
the form
-1
+ 1AA)rof0, = Afo + go
~ ~o~
fa = (AT - A) go f~
1
Now can be solved from the third equation in (5.5) and, since II = r~lo, the
vectors (12, ... , fn) and <p can be solved from the second equality in (5.5). This
proves p(AT) n p(C) c P(HT).
To prove the reverse inclusion it is first shown that O"(C) C O"p(HT) holds for
every 7 i= 0. In view of
(C q - A)ATh = (0, ... ,0, -q(A)h)T, A E C, hE 1i,
the eigenspace of Cq at A is given by
ker (C q - A) = {AT h: h E ker q(A) }. (5.6)

Assume that A E O"(Cq). Since 7 i= 0, one has P i= I and hence in view of (5.6) and
1
the assumption q = IH 0q one can find i= such that ph = and (C q - A)1 = 0.
It is easy to check that (0,0,1) T E ker (HT - A). Hence, O"(Cq) C O"p(HT) and by
the symmetry of spectra O"(Cq~) C O"p(HT), so that O"(C) C O"p(HT).
Now, let A E P(HT) and let g =?i = 0. Then A E p(C) and it follows from the
second and the third equalities in (5.5) that
pr810 = (I - p)r~lo = 0.
Therefore, 10 E AT and the first equality in (5.5) means that

{fa, go} E AT - A.
By assumption A E P(HT) and since go E 5)0 is arbitrary it follows that A E p(AT).
Therefore, P(HT) C p(AT) n p(C).
(ii) The statement follows from the identity (with A E P(HT))
(iii) Clearly, .c is a neutral subspace of 5)0 EEl 5)Q with dimension dn, so that
it is maximal neutral, cf. [5]. Moreover,
where Xn stands for
This implies (5.2).

(iv) The transform of the boundary mappings in (5.4) corresponds to the

following transform of the Weyl function M = q~ Moq (cf. [13]):
Mr = [(I - P) + PM][P - (I - p)Mtl
= (QU Mnq QU M 12 ii\ ( --1 I_I _ 0 )

o I) -q M22 M 21 q _q-l M:;/q-~
= (QU(Mn_:-_lMl_2Ml:;/ 1!21)q -QUMI2M:;/q-~)
-q M22 M 21 q _q-l Mi}q-U '
from which the statement follows. D
Recall that p(Ar) C p(Mo,r) and p(Hr) C p(Mr ), and that the inclusions
are equalities if 8 0 and 8 are simple. These properties are also reflected in (i) and
(iv) of Theorem 5.l.
The proof of Theorem 5.1 gives also the following result, which shows the
difference between the cases T = 0 and T =F 0, cf. Proposition 3.5.
Corollary 5.2. Let the assumptions be as in Theorem 5.1 and let T be given by
T = {{Ph, (I - P)h} : hE 1i} for some orthogonal projection P in 1i. If T =F 0
(i.e., P =F I) then a( qU q) C ap(Hr) and a(Hr) = a(Ar) U a(q~q).
Acknowledgment
This work has been supported by the Research Institute for Technology at the
University of Vaasa, the Academy of Finland (project 79774), and the Dutch As-
sociation for Mathematical Physics (MFOO/34).
References
[1] S. Albeverio, F. Gesztesy, R. H!2Segh-Krohn, and H. Holden, Solvable models in quan-
tum mechanics, Springer-Verlag, New York, 1988.
[2] S. Albeverio and P. Kurasov, "Finite rank perturbations and distribution theory",
Proc. Amer. Math. Soc., 127 (1999), 1151-116l.
[3] S. Albeverio and P. Kurasov, Singular perturbations of differential operators, London
Math. Soc., Lecture Notes Series 271, Cambridge University Press, 1999.
[4] Yu.M. Arlinskil, S. Hassi, Z. Sebestyen, and H.S.V. de Snoo, "On the class of extremal
extensions of a nonnegative operator", Bela Szokefalvi-Nagy memorial volume, Oper.
Theory Adv. Appl., 127 (2001), 41-8l.
[5] T.Ya. Azizov and I.S. Iokhvidov, Foundations of the theory of linear operators in
spaces with an indefinite metric, Nauka, Moscow, 1986 (Russian) (English transla-
tion: Wiley, New York, 1989).
[6] Yu.M. Berezanskil, Expansions in eigenfunctions of self-adjoint operators, Naukova
Dumka, Kiev, 1965 (Russian) (English translation: Translations of Mathematical
Monographs, Volume 17, American Mathematical Society, 1968).
[7] V.A. Derkach, "On generalized resolvents of Hermitian relations", J. Math. Sciences,
97 (1999), 4420-4460.
[8] V.A. Derkach and S. Hassi, "A reproducing kernel space model for N",-funetions",
Proc. Amer. Math. Soc., 131 (2003), 3795-3806.
[9] V.A. Derkach, S. Hassi, and H.S.V. de Snoo, "Operator models associated with
singular perturbations", Methods of Funetional Analysis and Topology, 7 (2001),
1-2l.
[10] V.A. Derkach, S. Hassi, and H.S.V. de Snoo, "Rank one perturbations in a Pontryagin
space with one negative square", J. Funet. Anal., 188 (2002), 317-349.
[11] V.A. Derkach, S. Hassi, and H.S.V. de Snoo, "Singular perturbations of self-adjoint
operators", Mathematical Physics, Analysis and Geometry, 6 (2003), 349-384.
[12] V.A. Derkach and M.M. Malamud, "Generalized resolvents and the boundary value
problems for hermitian operators with gaps", J. Funet. Anal., 95 (1991), 1-95.
[13] V.A. Derkach and M.M. Malamud, "The extension theory of hermitian operators
and the moment problem", J. Math. Sciences, 73 (1995), 141-242.
[14] J.F. van Diejen and A. Tip, "Scattering from generalized point interaction using self-
adjoint extensions in Pontryagin spaces", J. Math. Phys., 32 (3), (1991), 631-64l.
[15] A. Dijksma, H. Langer, Yu.G. Shondin, and C. Zeinstra, "Self-adjoint operators with
inner singularities and Pontryagin spaces", Oper. Theory Adv. Appl., 118 (2000),
105-175.
[16] V.I. Gorbachuk and M.L. Gorbachuk, Boundary value problems for operator differ-
ential equations, Naukova Dumka, Kiev, 1984 (Russian) (English translation: Kluwer
Academic Publishers, Dordrecht, Boston, London, 1990).
[17] S. Hassi, M. Kaltenbiick, and H.S.V. de Snoo, "Triplets of Hilbert spaces and
Friedrichs extensions associated with the subclass Nl of Nevanlinna functions", J.
Operator Theory, 37 (1997), 155-18l.
[18] S. Hassi, H. Langer, and H.S.V. de Snoo, "Self-adjoint extensions for a class of
symmetric operators with defect numbers (1, 1)", 15th OT Conference Proceedings,
(1995), 115-145.
[19] S. Hassi and H.S.V. de Snoo, "One-dimensional graph perturbations of self-adjoint
relations", Ann. Acad. Sci. Fenn. A.I. Math., 22 (1997), 123-164.
[20] S. Hassi and H.S.V. de Snoo, "Nevanlinna functions, perturbation formulas and
triplets of Hilbert spaces", Math. Nachr., 195 (1998), 115-138.
[21] M.G. KreIn and H. Langer, "Uber die Q-Funktion eines 7r-hermiteschen Operators
im Raume II",", Acta Sci. Math. (Szeged), 34 (1973), 191-230.
[22] P. Kurasov, "5)-n-perturbations of self-adjoint operators and KreIn's resolvent for-
mula", Integr. Equ. Oper. Theory, 45 (2003), 437-460.
[23] Yu.G. Shondin, "Quantum-mechanical models in lR n associated with extensions of
the energy operator in Pontryagin space", Teor. Mat. Fiz., 74 (1988), 331-344) (Rus-
sian) (English translation: Theor. Math. Phys., 74 (1988), 220-230).
[24] A.V. Straus, "Extensions and generalized resolvents of a symmetric operator which is
not densely defined", Izv. Akad. Nauk SSSR, Ser. Mat., 34 (1970), 175-202 (Russian)
(English translation: Math. USSR-Izvestija, 4 (1970), 179-208).
[25] E.R. Tsekanovskil and Yu. L. Shmulyan, "The theory of bi-extensions of operators on
rigged Hilbert spaces. Unbounded operator colligations and characteristic functions",
Uspekhi Mat. Nauk, 32:5 (1977), 69-124 (Russian) (English translation: Russian
Math. Surveys, 32:5 (1977), 73-131).
Vladimir Derkach
Department of Mathematics
Donetsk National University
U niversitetskaya str. 24
83055 Donetsk
Ukraine
e-mail: derkach@univ.donetsk.ua
Seppo Hassi
Department of Mathematics and Statistics
University of Vaasa
P.O. Box 700, 65101 Vaasa
Finland
e-mail: sha@uwasa.fi
Henk de Snoo
Department of Mathematics and Computing Science
University of Groningen
P.O. Box 800, 9700 AV Groningen
Nederland
e-mail: desnoo@math.rug.nl
Operator Theory:
2004 Birkhauser Verlag Basel/Switzerland
Minimal Realizations of
Scalar Generalized N evanlinna Functions
Related to Their Basic Factorization
Aad Dijksma, Heinz Langer, Annemarie Luger and Yuri Shondin
Abstract. In this paper we present a minimal realization of a scalar general-

ized Nevanlinna function q which corresponds to the basic factorization of q
as a product of a Nevanlinna function qo and of a rational function r#r, which
collects the generalized poles and generalized zeros of q that are not of positive
type. The key tool are reproducing kernel Pontryagin spaces.
Mathematics Subject Classification (2000). Primary 47 A06, 47 A50, 47B32,
47B50.
Keywords. Generalized Nevanlinna function, generalized pole, generalized
zero, reproducing kernel Pontryagin space, symmetric operator, self-adjoint
extension.
1. Introduction
In [DLLSh] (see also [DeHS1]) it was shown that a generalized Nevanlinna function
q E N", admits a unique basic factorization
q(z) = r#(z)qo(z)r(z) (1.1)
with a Nevanlinna function qo E No and a rational function r, whose zeros (poles,
respectively) are the generalized zeros (poles, respectively) of q in C+ UlFt (C- UlFt,
respectively) which are not of positive type. Here and in the following, for a vector
function f, by f# we denote the function f#(z) := f(z*)*; for the definition of r
and of generalized poles and zeros we refer to Section 3.
The research for this paper was supported by the Netherlands Organization for Scientific Research
NWO (grant 047-008-008), the Russian Foundation for Basic Research RFBR (grant N 0001-
00544), the Research Training Network HPRN-CT-2000-00116 of the European Union, and the
"Fond zur Fi:irderung der wissenschaftlichen Forschung" (FWF, Austria) grant number P15540-
N05.
70 A. Dijksma, H. Langer, A. Luger and Yu. Shondin
It is well known that every generalized Nevanlinna function q E Nt< admits a

minimal realization in some Pontryagin space of negative index K,. In the present
paper we construct a minimal realization of q E Nt<, K, > 0, which corresponds to
the basic factorization (1.1) of q. This means, we construct a realization of q in
terms of realizations of the function qo and of the matrix generalized Nevanlinna
function
r#(z))
o .
Recall that a realization (A, r z) for an n x n matrix generalized Nevanlinna
function Q is given by a self-adjoint relation A in some Pontryagin space P and
a corresponding r-jield, that is, a family of mappings rz E .C(C n , P), z E p(A),
which satisfy
z, w E p(A), (1.2)
and
Q-,--(z-,----)-_Q--=--('*----w-'----)* = r*w r z, Z,W E P(A) ,z -I-
--=- r W *. (1.3)
z-w
If a point Zo E p(A) is fixed this implies the following representation of Q:
Q(z) = Q(zo)* + (z - za)r;o (I + (z - zo)(A - z)-l)rzo, z E V(Q).
Here V(Q) denotes the domain of holomorphy of the function Q. Note that the
function Q is determined by A and the r -field r z up to an additive constant
self-adjoint n x n matrix. The space P is called the state space of the realization
(A, r z)' The realization (A, r z) can always be chosen minimal which means that
P = span {rzc I z E e(A),c E cn}.
In this case two minimal realizations of Q are unitarily equivalent and, moreover,
V(Q) = p(A); see [DLS2, Theorem 1.1]. In a minimal realization of Q the negative
index of the state space P is equal to the number of negative squares of the kernel
._ Q(z) - Q(w)*
K Q ( z,w ) .- , z,w E V(Q),
z-w*
where the expression on the right-hand side for w = z* is to be understood as
Q'(z).
For K, = 0 and n = 1 (and under some additional growth condition on Q) the
self-adjoint operator of a realization of Q is given by the operator of multiplication
in the space L;, where a is the measure in an integral representation of Q, see,
for example, [LLu].
With a realization (A, r z) often a symmetric restriction S of the relation A
is defined by
S:= {{f,g} E A Ir;o(g - zaf) = a}.
This definition is independent of Zo E V( Q) and r z maps C n onto the defect
subspace (ran (S - z*)) 1. of S. The triplet (A, r z, S) is sometimes called a model
for the realization of Q or, for short, a model for the function Q.
Minimal Realizations of Generalized Nevanlinna Functions 71
In Section 2, Theorem 2.1, we construct the canonical model for a realization

of a generalized Nevanlinna function Q E N;:xn. Here by canonical we mean that
the model acts in the reproducing kernel Pontryagin space (Q). To emphasize
the dependence of the canonical model on the function Q we sometimes write
A Q, SQ etc. instead of A, S, etc. As a byproduct of this construction, in (Q) a
simple characterization of the Jordan chains of the operator AQ is given. Finally,
in Theorem 2.4 we describe all self-adjoint extensions in ( Q) of the symmetric
operator SQ.
In Section 3 with the factorization (1.1) the matrix function
qO(Z) o
Q(z) = ( ~ o
r(z)
is defined and it is shown that a realization of q can be derived from a realiza-
tion of Q. The latter is evidently the orthogonal sum of realizations of qo E No
and of Mr(z) = (r~z) r#~z)). Then the self-adjoint relation A of a minimal
realization of q can, for example, be obtained as the compression of a certain
self-adjoint extension of SQ to the orthogonal complement of a finite-dimensional
Hilbert subspace Ho of (Q) = (qo) EB (MT). Moreover, in Theorem 3.4 the
finite-dimensional space (MT) is described explicitly.
In Section 4 the Hilbert subspace Ho of (qo) EB (MT) is characterized in
terms of eigenfunctions of Aqo corresponding to those generalized zeros of q which
belong to the point spectrum O"p(Aqo) of Aqo and of eigenfunctions of A_ 1 / qo corre-
sponding to those generalized poles of q which belong to O"p(A_ 1 / qo ). In particular,
the dimension of the Hilbert space Ho is determined.
Recently, V. Derkach and S. Hassi in [DeH] constructed a model for q by a
certain coupling of the self-adjoint relations AqO and AMr in terms of boundary
triples. Earlier, jointly with H. de Snoo in [DeHS2, DeHS3] they also used the
coupling method to construct models for matrix-valued generalized Nevanlinna
functions q with a matrix polynomial r. In all three papers necessary and sufficient
conditions for minimality are formulated. As is well known minimal models are
unitarily equivalent, which implies that the models are essentially the same. In
the present paper we obtain a minimal model in a more direct way. Minimality
is described analytically in Theorem 4.4. The fact that we use canonical models
for the factors qo and r makes the model in this paper rather explicit and in fact
unique from the point of view of reproducing kernel Pontryagin spaces.
We assume that the reader is familiar with operator theory in spaces with
an indefinite metric such as in [AzI] and [IKL] and with the theory of reproducing
kernel spaces such as in [ADy] and [ADRS]. For linear relations in Krein spaces
we refer to, for example, [DS] and for canonical models to [dB]' [ADRS], [DLS2],
and [DeHS 1].
2. The canonical model for a matrix generalized

Nevanlinna function
1. The generalized Nevanlinna class N:::xn (Ny< := N~Xl) consists of all n x n
matrix functions Q which are meromorphic in e \ JR., satisfy Q(z*)* = Q(z) for
z E V(Q), and for which the kernel
K Q( z,w ) -- Q(z) - Q(w)* , z,w E V(Q)..J.

, z r w ,
*
z-w*
has Ii negative squares. We denote by (Q) the corresponding reproducing kernel
Pontryagin space: This is the completion of the linear span of all functions of the
form KQ(. ,w)c, c E en, wE V(Q), with respect to the norm generated by the
inner product
(KQ( ,w)c, KQ(, z)d).C(Q) := d* KQ(z,w)c,
which has Ii negative squares. The elements of (Q) are holomorphic n-vector
functions J on V(Q). Sometimes the function J is also denoted by J(). The
reproducing property of the kernel KQ(z, w) is expressed by the inner product
formula
(1(- ),KQ( ,w)c)C(Q) = (1(w),c)cn = c* J(w).
The following theorem describes a realization of the function Q in the space ( Q).
The origins of this theorem go back to M.G. KreIn for the case Ii = 0 and to
[KL1, KL2, KL3] for the general case, see also [DLS1]. In these papers the so-
called c-method or formal sum method was used to obtain a realization of Q.
Alpay showed in [Aj that the space (Q) is invariant under the difference-quotient
operator
Rw(1)(z) = J(z) - J(w) (2.1)
z-w
and that this operator satisfies the resolvent identity, see also [dB]. In turn this
result was used in [DLS2, Section 3], see also [ABDS], to obtain the realization
below. Here we make use of reproducing kernel space methods only. To keep this
paper reasonably self-contained and for the convenience of the reader, we prove
most of the statements.
The definition of a matrix generalized Nevanlinna function can easily be
extended to functions whose values are bounded linear operators in a Hilbert space.
As can be seen from its proof, the following theorem (except for the formula for
the defect indices in statement) also holds in that case. In the sequel we use the
theorem for the cases n = 1,2, and 3 only.
Theorem 2.1. Let Q E N:::xn be given. Then:
(i) A:= {{j, g} E ( Q)213 c E en: g( () - (f( () == c} is a self-adjoint linear
relation in (Q) with p(A) i- 0, and
(r zC)() = KQ(, z*)c = Q(~ =~(z) c, cE en,
is a corresponding r-field. The pair (A,r z ) is a minimal realization ofQ.
(ii) The resolvent of A is the difference-quotient operator in (Q):

(A - w)-l = R w , w E p(A).
(iii) S := {{f,g} E (Q)21 g() - (f() == o} is a symmetric operator in (Q)
with equal defect indices n - d, where d = dimkerr z . Moreover, ap(S) = 0
and the adjoint of S is given by
S*=span{{rzh,zrzh} 1 hECn,ZEV(Q)}
= {{f, g} E (Q)21:l c, dE C n : g() - (f() == c - Q()d}.

Recall that a nonself-adjoint symmetric operator is called simple if it has no
non-real eigenvalues and the defect elements to the nonreal spectral points form a
total subset of the whole space. Hence the symmetric operator S in the theorem
is simple. If n = 1 the function
r.p(. ,z) := rz1 = (1 + (z - w)(A - z)-l)r.p(. ,w),
called a defect function for S and A, spans the defect subspace of S at z and the
representation of Q takes the form
Q(z) = Q(w)* + (z - w*)(r.p( ,z),r.p( ,w)).cCQ)'
Proof of Theorem 2.1. Consider the linear relations Ao and Uo in (Q) defined by
Ao:= {{ LKQ(- ,w)cw , Lw*KQ(- ,w)cw } ICw E cn, LCw = o}

w w w
and
Uo := { {L(w* -p,)KQ(" w)c w , L(w* -p,*)KQ( " w)c w } Icw E C n , L Cw = o}.
w w w
Here 2:w stands for sums over finite sets of points w E V(Q) and p, is a fixed
non-real point in V(Q). Then Ao C Ail and Uo is isometric. The equalities
domUo = span {KQ(. ,w)c Iw =I p,*, c E C n },
ranUo = span {KQ(' ,w)c I w =I p" c E cn}
imply that the domain and the range of Uo are dense in (Q). Hence (see [ADRS,
Theorem 1.4.2]) the closure U of Uo is a (bounded) unitary operator on (Q) and
the closure A of Ao is a self-adjoint relation in (Q). Moreover, we have
I
A = {{(U - l)u, (p,U - p,*)u} u E (Q)}.
From the relation
(A - p,*)-1 = _l_(U -1)
p,* - p,
it follows that p,* E p(A), that is, V(Q) n (C \ JR) c p(A). Note that for arbitrary
c E C n and )..,W E V(Q)
{KQ(" )..)c - KQ(. ,w)c, )..* KQ(-, )..)c - w* KQ(-, w)c} E Ao c A
and hence
(A - W*)-l KQ((, '\)c = KQ((, ~~ =~~((, w) c.
Taking the (Q) inner product of f E (Q) with the elements on both sides of
this equality we find
(A - w)-l f('\) = f('\) - f(w) = Rwf('\), (2.2)

'\-w
or, in words, the difference-quotient operator is a bounded operator on (Q) and
coincides with the resolvent operator of A. This implies that
A = {{Rwh, h + wRwh} I h E (Q)}
{{f,g} E (Q)213c E en: g(() - (f(() == c}.
From formula (2.2) it follows that the family r z given in (i) is a r -field for A.
We find that r~ =: Ew' is the operator of point evaluation at w* and hence the
representation for Q holds. This realization is minimal, since (Q), by definition a
reproducing kernel space, is generated by the functions K Q (, w). This completes
the proofs of (i) and (ii).
From 8 C A = A * it follows that 8 is a symmetric operator with equal defect
indices. The point spectrum O'p(8) of 8 is empty and by the reproducing property
of the kernel KQ(z, w) the first equality for 8* in (iii) holds true. Now we use that
8 can be written in the form
where the right-hand side does not depend on the choice of the non-real point
/1 E V(Q). On the one hand this implies
8* = A + {{rJtd,/1rJtd} IdE en}
= {{f,g} E (Q)213c,d E en: g(() - (f(() == c - Q(()d}.
On the other hand it also implies that ran (8 - /1*) = ker r; = (ran r Jt).1 and
hence
dim ker (8* - /1) = dim (ran (8 - /1*)).1 = dim ran r Jt = n - dim ker r Jt'
that is, the defect indices of 8 are n - d with d as in the theorem. D
We call the realization of Q in Theorem 2.1 the canonical realization for the
function Q and the corresponding model (A, r z, 8) the canonical model. As men-
tioned in the introduction, if need be, to denote the dependence of the canonical
model on Q we write AQ, rQz, and 8Q.
Recall that a point a E e U {(X)} is called a generalized pole of Q E NI< if
it is an eigenvalue of the relation A Q . A generalized zero of Q is by definition a
generalized pole of the function _Q(Z)-l. As a direct consequence of Theorem 2.1
we find the following description of the algebraic eigenspaces of the relation A Q .
Corollary 2.2. Let Q E N;:xn be given. Then:

(i) The point 0: E e is a generalized pole of Q if and only if for some non-zero
vector C1 E en
C1
(_o:E(Q).
In this case, the Jordan chains of length k at the eigenvalue 0: of AQ are the
sequences in (Q) of the form
C1 C1 C2 C1 C2 Ck
(-0: ' (-0:)2 + (-0: , ... , (-o:)k + (_0:)k-1 + ... + (-0:
with some vectors Ci E en, i = 1,2, ... , k.
(ii) The point 00 is a generalized pole of Q if and only if for some non-zero vector
C1 E en the constant function C1 belongs to (Q). In this case, the Jordan
chains of length k at the eigenvalue 00 of AQ are the sequences in (Q) of
the form
C1, (C1 + C2, ... , (k-1 C1 + (k-2 C2 + ... + Ck
with some vectors Ci E en, i = 1,2, ... ,k.
The following corollary gives a connection between the spaces (Q) and
(_Q-1) and a description of the generalized zeros of Q.
Corollary 2.3. Let Q E N;:xn and assume that Q(p,) is invertible for some non-real
point p, E V(Q). Then:
(i) The mapping f() 1-----+ Q()f() from (_Q-1) onto (Q) is unitary, and
under this mapping the self-adjoint relation A_Q-l in (_Q-1) 2S 2somor-
phic to the self-adjoint relation
AQ := {{f,g} E (Q? i3d E en: g() - (f() == Q()d}
in (Q). In particular, p(AQ) =I- 0.
(ii) The point 13 E e is a generalized zero of Q if and only if for some non-zero
vector d 1 E en,
~(~~1 E (Q).
In this case, the Jordan chains of length k at the eigenvalue 13 of AQ are the
sequences in (Q) of the form
Q( ()d1 Q( ()d1 + Q( ()d2 Q( ()d1 + Q( ()d2 + ... + Q(()dk
(-13 '(-13)2 (-13' ... , (_f3)k (_f3)k-1 (-13
with some vectors d i E en, i = 1,2, ... , k.
(iii) The point 00 is a generalized zero of Q if and only if for some non-zero vector
d 1 E en the function Q()d 1 belongs to the space E (Q). In this case, the
Jordan chains of length k at the eigenvalue 00 of AQ are the sequences in
(Q) of the form

Q()d 1 , (Q()d 1 +Q()d2, ... , (k-lQ()d 1 +(k-2Q()d 2+ +Q()d k
with some vectors d i E en, i = 1,2, ... , k.
Proof. The first statement is implied by the following relation for the kernels:
K_Q-l (, z) = Q()-l KQ(, z)Q(z*)-l
and [ADRS, Theorem 1.5.7]. The rest is an immediate consequence of the previous
corollary. 0
2. A function Q E N;::xn is called strict if for some non-real point IL E V(Q) it

holds
n
zEV(Q)
ker KQ(z, IL) = {O}.
For a minimal realization (A, r z) this means that the mappings r z are for some and
hence for every z E p(AQ) injective. In this case the following theorem describes
all canonical self-adjoint extensions of S. This theorem can be derived from [De,
Corollary 2.4], but we give a direct proof using the tools developed so far. The
formula (2.3) can be seen as KreIn's resolvent formula.
Theorem 2.4. Suppose that the generalized Nevanlinna function Q E N;::xn is strict
and let (A, r z, S) be the canonical model for Q. Then:
(i) A linear relation is a canonical self-adjoint extension of S if and only if it is
of the form
AA,B := {{f, g} E (Q)213 hE en: g() - (f() == (A + Q()8)h}
with n x n matrices A and 8 satisfying the relations
rank (:) = n, A*8-8*A=O.
If AA,B and AA',B' are two such canonical self-adjoint extensions of S then
AA' ,13' = AA,B if and only if A' = A C and 8' = 8 C for some invertible n x n
matrix C.
(ii) p (AA,B) =1= 0 if and only if for some non-real point W E V(Q) the matrices
A+Q(w)8 andA+Q(w*)8 are invertible. In this case for z E P(AA,B)np(A):
(AA,B - Z)-l= (A-z)-1-r z 8 (A + Q(z)8)-1 r:.. (2.3)
Proof. (i) With IL E V(Q) n e+ we have the direct decomposition

S* = A + {{r Jlx, ILr Jlx} 1 x E en} .
Hence each element {f, g} E S* can be written as
{f,g} = {fA + r JlX,gA + ILr Jlx}
with {fA, gAl E A and a vector x E en, which is uniquely determined since f J1 is
by assumption injective. Define the mapping b : S* ----+ e 2n by
b ({f }) .= (BO({f,9})).= (f~(9A-J.L*fA)+Q(J.L)X)

,g. B 1({f,g})' x .
Since A is self-adjoint and J.L E D(Q) the elements gA - J.L* fA run through C(Q)
when {fA, gAl runs through A. From the injectivity of f J1 it follows that f~ and
hence also b are surjective. Also, evidently, S c ker b. Moreover, Greens identity
holds: If {f, g} = {fA + f J1 X,gA + J.Lf J1x} and {h, k} = {hA + f J1Y' kA + J.Lf J1Y}
are arbitrary elements of S*, then
(k,f)C(Q) - (h,g)c(Q)
= (kA - J.L* hA, f J1x) C(Q) - (f J1Y' gA - J.L* fA) C(Q) + (J.L - J.L*) (f J1Y' f J1x) C(Q)
= (Bo{h,k},B1{f,g})cn - (B 1{h,k},B o{f,g})cn '
Hence b is a boundary mapping as in [M] and [De], where boundary triples are
considered, and as in [DLS1, Section 3] and [AzCD].
From Greens formula it follows that the self-adjoint extensions of S can be
parametrized as restrictions of S*:
AT = {{f,g} E S*I {B1({f,g}), Bo({f, g})} E T},

where the relation
is self-adjoint. Choose Zo E e+ and define A and B as the matrix representations

of the operators - (I + Zo (T - ZO) -1) and (T - zo) -1, respectively. They satisfy the
relations in (i) and we have
T = {{ - Bh, Ah} 1 h E en} .
Since S* in C( Q) is given by
S* = A + {{f J1d, J.Lf J1d} d
1en}E
= {{f,g} E C(Q)21 :lc,d E en: g(() - (I(() == c - Q(()d},
the boundary mapping b = (~~) can be written as
Bo ({f,g}) = c, B1 ({f,g}) = d.
Hence the relation AT =: AA,B is described by the formula
With the above descriptions of AT and AA,B Krein's formula reads as

or, equivalently,
The proofs of the other statements in (i) are left to the reader.
(ii) We first assume that p (AA,B) -=I- 0. Then, since AA,B is self-adjoint and r~* =
Ew is surjective, we find
Therefore, by the definition of AA,B, the matrix A + Q(w)B is invertible.

To prove the converse denote the operator on the right-hand side of (2.3) by
K(z). Using
r z - rw = (z - w)(A - z)-lr w, (A - w)-lr z = (A - z)-lr w,
and
B(A + Q(Z)B)-l = (A* + B*Q(Z))-lB*
we find that K (z) has all the properties of a resolvent operator of a canonical
self-adjoint extension of S with a nonempty resolvent set: D(K), the domain of
holomorphy of K, is symmetric with respect to the real axis and
K(z*)* = K(z), K(z) - K(w) = (z - w)K(z)K(w), K(z)(S - z) c I.
If we denote this self-adjoint extension by A K , then
AK = {{K(z)h, h + zK(z)h} IhE C(Q)}.
Using again r;* = Ez we see that for h E (Q),
h() + (z - ()K(z)h() == (A + Q()B)c, c = (A + Q(z)B)-lh(z).

Hence AK C AA,B. Since both relations are self-adjoint, equality prevails. D
Remark 2.5. Note that according to (2.3) the resolvent of AQ in Corollary 2.3 is
an at most n-dimensional perturbation of the resolvent of A Q .
3. A minimal model related to the basic factorization

1. Recall that a generalized pole a of q E NK is by definition an eigenvalue of the
relation A q . A generalized pole a is called of positive type if the corresponding
eigenspace of Aq is positive. If the generalized pole a is not of positive type, its
degree of non-positivity is by definition the dimension of a maximal non-positive
invariant subspace of the algebraic eigenspace of Aq at a. A generalized zero f3 of
q, which is by definition a generalized pole of -q(z)-l, is called of positive type if
it is a pole of positive type of -q( z) -1, and if f3 is not of positive type its degree of
non-positivity is the degree of non-positivity of f3 as a generalized pole of _q(Z)-l.
If q E NI<' its basic factorization (1.1) is defined as follows. Let the points
ai, i = 1,2, ... ,, ((3j, j = 1,2, ... ,k, respectively) be the generalized poles (ze-
ros, respectively) of q in C+ U lR that are not of positive type, denote by Vi ("'j,
respectively) the degrees of non-positivity of ai ((3j, respectively), and define
r(z) .= (z - (31)1<1 ... (z - (3k)l<k (3.1)
. (z - ai)V1 ... (z - a;)" .
Then there exists a function qo E No such that
q(z) = r#(z)qo(z)r(z), (3.2)
see [DLLSh], [DeHS1]. Note that, if
T := "'I + ... + "'k - (VI + ... + Vi) (3.3)
is positive (negative, respectively), then 00 is a generalized pole (zero, respectively)
of q which is not of positive type and with degree of non-positivity ITI. Since 00
cannot be a generalized zero and a generalized pole at the same time, we have
'" = max {"'I + ... + "'k, VI + ... + vt}.
By Mr(z) and Q(z) we denote the following matrix functions:
0
( r(z) r#(z)) Q(z) = (
qO(Z)
0
0 0)
0 r#(z) ,
Mr(z) = 0 '
o r(z) 0
which belong to the classes ~X2 and N!X3, respectively (see Theorem 3.4 below).
Note that the elements of (qo) are scalar functions, the elements of (Mr) are 2-
vector functions, and the elements of (Q) = (qo)EJ1(Mr) are 3-vector functions.
By v( () we denote the vector function
v(() := ( r~() ),
r(()qo(()
which is not necessarily an element of (Q). The first result concerns the state
space (q), which will be described in terms of a unitary mapping from a subspace
of (Q) to (q).
Lemma 3.1.
(i) The bounded linear operator T : (Q) f--t (q) defined by
(Ti)(() := v#(()i(()
is a partial isometry with ran T = (q).
(ii) The linear manifold
'Ito := kerT = {i E (Q) Iv#(()i(() == O}

is a finite-dimensional Hilbert subspace of ( Q).
(iii) With P := 'Itt the restriction T := T Ip is a unitary mapping from P
onto (q).
Below in Corollary 4.3 we give a formula for the dimension of the space lto.
Proof of Lemma 3.1. A straightforward calculation shows that the two kernels
Kq(z, w) and KQ(z, w) are related by the formula
Kq((, z) = v#(()KQ((, z)v(z*).
Since the numbers of negative squares of these kernels coincide, according to
[ADRS, Theorem 1.5.7.] the mapping T is a surjective partial isometry and its
kernel lto is a Hilbert space. Write
T = (0 T): Ci,o) --7 (q).
Then T : P --7 (q) satisfies (iii) of the theorem. From the defining equation for
lto we find dim lto ::; dim (Mr). It is well known (or follows from Theorem 3.4
below) that dim(Mr) = 21>:. D
Theorem 3.2. Let q E N" be given. Then:

(i) The self-adjoint relation
A:= {{1,:q} E p21:3c E c: v#(()(:q(() - (J(()) == c}
and the corresponding f -field f z 1 with
fz1(():= ",((,z):= KQ((,z*)v(z)
= Q(() - Q(z) v(z) = (fQzv(z))((), z E V(q),
(-z
form a m1:nimal realization of q in the state space P.
(ii) Under the mapping T this realization is unitarily equivalent to the canonical
model for q.
(iii) The corresponding symmetric operator S is given by
S:= {{1,:q} E p2J v#(()(:q(() - (J(()) == o}
and its adjoint by
S* = { {1,:q} E p21 :3 c, dEC: v#(()(:q(() - (J(()) == c- q(()d} .
Proof Theorem 2.1 and the relation
A = {{1,:q} I {T1,T:q} E Aq}
imply that A is self-adjoint. Likewise, with S = T-1SqT and
S* = { {1,:q} I {T 1, T:q} E s; }
the statement (iii) follows. The reproducing property of K Q( (, z) implies for 1E
lto and z E V(q)
( J( . ), K Q( . , z*)v(z)) = v(z)* 1(z*) = v#(z*)J(z*) = 0,

L:(Q)
and hence <p( . , z) E P. From the relation

T<p(, z) = v#( )KQ(, z*)v(z) = K q(, z*) = <pq(., z)
we see that <p( ., z) is a r-field corresponding to A and, moreover, since T is unitary,
we have the representation
q(z) - q(zo)
z _ z* = (<pq( ., z), <pq( ., zo)) C(q) = (<p(., z), <p( ., zo))p.
o
The equality
/ J(. ), KQ(, z)v(z*) \ = v#(z)J(z) = (TJ)(z), J E .c(Q), z E V(q),

\ / C(Q)
where T is the partial isometry defined in Lemma 3.1, and the fact that Ho = kerT
imply
P = span{<p( ,z) I z E V(q)},
that is, the realization is minimal. D
2. In the following we give another description of the self-adjoint relation A in P.

The symmetric operator SQ in the canonical model for Q can be decomposed as
SQ = SqO EEl SMr and it has defect indices (3,3). The self-adjoint linear relation
A Q = AqO EEl AMr is a canonical self-adjoint extension of SQ. By Theorem 2.4,
another canonical self-adjoint extension of SQ is given by
A:= {{J,g} E (.c(Q))213h E (:3: g(() - (J(() == (I + Q(()B)h},

where I is the 3 x 3 unit matrix and
0 0 1)
B=- ( 0 0 0 .
100
It has a nonempty resolvent set since the matrix
1 0 -qo(Z))
I + Q(z)B = ( -r#(z) 1 0
o 0 1
is invertible. Furthermore,
(A - z)-l = (A Q - z)-l - rQzB (I + Q(Z)B)-l r Qz . (3.4)
The next theorem implies in particular that in the state space .c( Q) also the
pair (A, <p(., z)) is a (not necessarily minimal) realization of q.
Theorem 3.3. With respect to the decomposition .c( Q) = PEEl Ho we have

(A - Z)-l = (A - z)-l EEl (Ao - z)-l, (3.5)
where A is the self-adjoint relation in P as in Theorem 3.2 and Ao is a self-adjoint
relation in Ho.
It follows that A = An p2, Ao = An H6, and the relation (3.5) can be

written as
A={{J+h,g+k} I {J,g}EA,{h,k}EAo}=AEBA o.
Proof of Theorem 3.3. We show first that for z, w E p(A),
(z - w)(A - Z)-l<p(., w) = <p( ., z) - <p(., w). (3.6)
For this we apply (3.4) with
B(I + Q(z)B)-l = - (~~ ~)

1 0 qo(z)
to the r-field rQw. Using the defining relations (1.2) and (1.3) for a r-field, we
obtain
(z - w)(A- - z) -1 rQw = rQz - rQw
0 0 01 ) (qo(Z) -0 qo(w) 0 0)
+rQz ( 0 0 0 r#(z) -0 r#(w)
1 0 qo(z) 0 r(z) - r(w)
r Qz - r Qw + r Qz (
0
0
r(z) - r(w)
0
0)
00 .
qo(z) - qo(w) qo(z)(r(z) - r(w))
Multiplying both sides of this equality from the right by the column vector v(w)
and observing that rQwv(w) = <p(. ,w) we find
(z - w)(A - Z)-l<p(., w)
rQzv(w) _ <p(. ,w) + rQz ( ~

r(z) r(w) )
qo(z)r(z) - qo(w)r(w)
= rQzv(w) - <p(. ,w) + rQz{v(z) - v(w))

= <p(., z) - <p(- , w).
This implies (3.6). From Theorem 3.2 we have that
(z - w)(A - Z)-l<p(. ,w) = <p(. ,z) - <p(. ,w),
and comparing this with (3.6) we obtain
(A - Z)-l<p( . , w) = (A - z)-l<p( . , w).
Since the elements <p( ., w), wE p(A), form a total set in P we have in fact that
(A - z)-lh = (A - z)-lh, hE P.
It follows that P is invariant under (A - Z)-l for all z E p(A), and by taking
adjoints we see that also Ho is invariant under this resolvent operator. Thus the
decomposition of (1 - z)-l with respect to the decomposition (Q) = P E9 Jio is

a diagonal block operator matrix with entries (A - Z)-l and an operator Ro(z),
say, on Ji o. Since Ro(z) satisfies the resolvent identity and Ro(z)* = Ro(z*), it is
the resolvent of a self-adjoint relation Ao in Ji o. D
3. Now we give a description of (Mr). This space is also considered in [DeH],

but here we emphasize the structure of the space via the spectral decomposition
of the self-adjoint relation AMr in the model for the function Mr. Recall that the
function r is given by (3.1). We write
r(z) = ro(z) +rl(z) + ... +rc(z),

with
ri(z):= L
Vi
-lJij
( z - a* )J ' i = 1,2, ... ,f,
j=l '
where all the coefficients are complex numbers, IJVa -I- 0, and lJivi -I- 0 for i
1, 2, ... J Set M i (z) := Mri (z ), Ai := AMri ' i = 0, 1, 2, ... , f.
Theorem 3.4. We have
(Mr) = (Mo) E9 (Md E9 ... E9 (Mc) (3.7)
and this orthogonal decomposition is the spectral decomposition for the self-adjoint
relation AMr = Ao E9 Al E9 ... E9 Ac in the model for M r :
(a) For i = 1,2, ... , f, the space (Mi) is spanned by two Jordan chains, each
of length Vi, corresponding to the eigenvalues ai and ai of Ai in (Mi) :
fr(-~i)2) fr(-~i)Vi) (~) , (~\

"0 "0
((":ai) , , ... ,
"0 and
(-a; ((-a:Y:)
, ... , ( 01
((-an"i
)
The Gram matrix Qi for this basis is given by

-1
lJil
T
lJi2 )
Qi) (
Qi := .
o '
lJiVi o
in particular the linear spans of the two chains are skewly-linked.
(b) If the polynomial ro is not a constant, the space (Mo) is spanned by two
chains of length Vo, corresponding to the eigenvalue 00 of Ao in (Mo):
The Gram matrix go for this basis is given by
I
-1
a1
- (0
go:= go go)
0 ' go :=
( a2
:
)
avo o
in particular the spans of the chains are skewly-linked.
Proof We write K i ( ,w) for KMi(' w), i = 0,1,2, ... ,f. From Mr(z) = Mo(z)+
+ ... + Me(z), we obtain
M 1 (z)
K Mr ( ,w) = Ko( ,w) + K 1( ,w) + ... + Ke( ,w),
whence (3.7), where, by [ADRS, Section 1.5], the sum is orthogonal because it is
direct. The latter follows from the description of these spaces which we give now;
in particular, elements from different summands have different singularities.
(a) For (:) , (~) E C 2 we have
(Ki(.,W) (:) ,Ki(-,V) (~)) CCMi) = (c* d*) Ki(v,w) (:)
* rf(v) - rf(w*)b d* ri(v) - ri(w*)

=c
v - w*
+ v - w*
a.
Choosing in particular a = c = 0 or b = d = 0 we find that the subspaces
of (Mi) are neutral subspaces. This accounts for the zero entries on the main
diagonal of the Gram matrix gi.
For a = d = 1 and b = c = 0, the above formula
reads as
(( 0 ) (rfC()-rfCv*)))
riC()-ri~w')' (-v*
ri(v) - ri(w*)
(3.8)
(-w 0 CCMi)
v-w*
Now we use that

v' j-l
ri(() - ri(w*) ~" aij
(- w* - ~ ~ (( - a*)j-s(w* - a*)s+1
J=1 s = o '
and
rf(() - rf(v*) _ ~~ aij
(- v* - ~ ~ (( - ai)j-s(V* - ai)s+1 .
J=ls=O
We multiply both sides of (3.8) by (w* - ai)n(v - ai)m, integrate over a small
circle around at with respect to w* and likewise with respect to v, and obtain for
n, m = 0, 1, ... , Vi - 1,
( ( '\'V
L.."j=n+1
0
((-a*)j
(Jij
n
) , (L~=m+1 0(( -:;lj m)) = O'i,n+m+1,
, (Mi)
where we set O'i,j = 0 if j 2:: Vi + 1. These equalities can be written in matrix form
~r;lHig;l = gi- 1 where Hi = (hZ,m)~'m=l with
hz
,m
= (( 01
((-an m
)
'
(_1 ))
(( -ai)l
0 (Md
From this relation we see that Hi = gi, and the formula for the Gram matrix gi
now readily follows. The rest of the proof of part (a) is left to the reader.
(b) The proof is similar to the previous one. The analog of (3.8) is
If we multiply with w*( -n-1)v- m - 1 and integrate with respect to w* and v over
small circles around the origin, we obtain for n, m = 0,1, ... ,Vo - 1,
( ( ,\,Va 0 ) (L~~m+l O'j(j-m-1))

_(j-n-1, 0 = O'n+m+1,
L.."j=n+1 O'J (Ma)
~here O'j = 0 if j 2:: Va + 1. This readily leads to the formula for the Gram matrix
go in this case. 0
4. The space Ho
It may happen that in the factorization (3.2) a real generalized zero f3 (generalized
pole 0:, respectively) of the function q is a generalized pole (generalized zero,
respectively) of qo as well.
Example. The function qo(z) = 1- ~ E No has a generalized pole at f3 = 0, which

is of positive type. If q(z) = 1 + Z~l' then q E N1 and q(z) = r#(z)qo(z)r(z) with
r(z) = z':l'
In this example the factor z - f3 cancels and the function q E N1 does not
have a pole at z = O. In the more general situation this "cancellation" corresponds
to the fact that the model (A,<p(' ,z)) is not minimal in the state space (Q). In
this section we describe the space H o, which is nontrivial in exactly this case.
Lemma 4.1. Let qo E No be given.

(i) Assume either {3 E C and n > 1 or n = 1 and the point {3 E C is not a
generalized pole of qo If
then every f E (qo) has the same behavior at {3 as qo, that is,
lim(( - (3tf(()
(~(3
= o.
(ii) Assume n > lorn = 1 and 00 is not a generalized pole of qo. If z"":;'oo
lim qO~) = 0
z
then also l~m f5~)

(--->00 .,
=0 for every f E ( qo).
Proof. We only prove statement (i), the other case can be proven in a similar
way. With the abbreviation x(,(*) := ((* -(3*)nKqo('() we find under the
assumptions of the lemma and since qo(z) E No that for wE D(qo)
t21(x(,e),Kqo (-,w))C(qo) = t21x(w,e) =0
and
t21 (x( ., e), x(, e)) C(qo) = t21(( - {3tx((, (*) = o.
That is, x( . , (*) converges for (~{3 weakly to the zero element. Hence we have
o
Let the rational function r(z) and the matrix function Mr(z) again be given
as in Section 3. For simplicity we use the following notations: ao := 00, {30 := 00,
and
1+ = {o : :; i :::; C I ai E O"p(Aqo)}, J+ = {o : :; j :::; k I {3j E O"p (Aqo) },
where Aqo is the self-adjoint relation in (qo) which is the isomorphic copy of the
self-adjoint relation A_ 1/ qo in ( -ljqo) in the canonical model for the function
-ljqo under the operator of multiplication by qo(() mapping ( -ljqo) onto (qo);
see Corollary 2.3. Moreover, by
{ qo() for i E 1+ \ {O},
Xai ( ( ) := (q~~i
for i = 0 E 1+
and
1
xp,() ,~ { ( - {3j
for j E J+ \ {O},
1 for j = 0 E J+
we denote those eigenelements of Aqo and Aqo' respectively, which correspond to
the eigenvalues that are cancelled. Note that the ai's and the {3j's are real numbers.
Now we can give an explicit description of the space 1io.
Theorem 4.2. The space H o, defined in Lemma 3.1, is given by
Ho = span { ( ~~~ ), (-::~~j) liE 1+, j EJ+}. (4.1)

-qo X"'i 0
(f)
Proof. Note that the span on the right-hand side of (4.1) is contained in Ho.
Conve".,ly, by definition >ill element f ~ with f E L (q,) and (~) E L(M,.)

belongs to the space Ho if and only if
g(()
f(() + r#(() + qo(()h(() = O. (4.2)
The function g(()jr#(() is a linear combination of the functions

((_/3j)-m, m=l, ... ,K:j, j=l, ... ,k,
and (recall T from (3.3)) if T > 0, of the functions (m-l, m = 1, ... , T . The
function h(() in (4.2) is holomorphic at the points ( = /3j for j = 1, ... , k and,
if T > 0, also at ( = 00. Observe that limz":'~j(z - /3j)n qo (z) = 0 for n > 1 and
j = 1, ... , k, and, if j ~ J+, also for n = 1. Hence multiplying both sides of the
relation (4.2) by (( - /3j)n for n > 1 and by ( - /3; for j ~ J+ and then taking
the limit (~/3; we find - step by step - that g( () j r# (() reduces to a sum of the
form I: gj X~j with gj E C. Dividing both sides of (4.2) by qo(() gives
jEJ+
1 g(()
u(() + qo(() r#(() + h(() = 0
with u( () := f (() j qo (() E ( -1 j qo). In the same way as above it follows that
qo(()h(() = I: hi X"'i (() with hi E <C and hence
iEI+
jEJ* iEI*
which completes the proof. o
Corollary 4.3. The dimension of Ho is the sum of the number of generalized poles
of q that are generalized zeros of qo and the number of generalized zeros of q that
are generalized poles of qo.
The following example illustrates the foregoing theorem.
Example. For qo E No and r(z) = ~=~ there appear four cases. If we identify
(Mr) with <c 2 according to its basis given in Theorem 3.4 we obtain:
(i) If neither 0: is a generalized zero nor /3 is a generalized pole of qo, then
Ho = {O}.
(ii) If a is not a generalized zero, but f3 is a generalized pole of qo, then
(iii) If a is a generalized zero of qo, but f3 is not a generalized pole, then
(iv) If both a is a generalized zero and f3 is a generalized pole of qo, then
Whether a generalized zero (generalized pole) of q is also a generalized pole

(generalized zero, respectively) of qo can also be characterized in terms of limits.
Theorem 4.4. Let q E NK be given. Then

. q(z)
:3 /'l,[3 EN: hPl (z - f3)2 K(3 -1 < 0,
z--+[3
and
a E O"p(Aq) n O"p(Aqo) 3vo: EN: liPl(z - a) 2Va - 1 q(z) >
z--->o:
o.
Proof. According to [LJ, f3 E O"p(Aq) has degree of non-positivity /'l,[3 if and only if
lim q(z) ~0 and lim q(z) > 0 or 00.

z~[3 (z - f3)2Kr1 z~[3 (z - f3)2K(3+1
From the factorization (3.2) of q we see that the first limit above is =f. 0 if and
only if liPl(z - f3)qo(z) =f. 0, which happens exactly if f3 E O"p(Aqo). The second
z--+[3
equivalence can be proved in a similar way. D
References
[AJ D. Alpay, Reproducing kernel Kre?:n spaces of analytic functions and inverse
scattering, Doctoral dissertation, The Weizmann Institue of Science, Rehovot,
Israel, 1985.
[ABDSJ D. Alpay, P. Bruinsma, A. Dijksma, and H.S.V. de Snoo, A Hilbert space as-
sociated with a Nevanlinna /unction, in: M.A. Kaashoek, J.H. van Schuppen,
A.C.M. Ran (editors) Signal processing, scattering and operator theory, and nu-
merical methods. Proceedings of the international symposium MTNS 89, Volume
III, Birkha.user Verlag, Basel, 1990, 115-122.
[ADRS] D. Alpay, A. Dijksma, J. Rovnyak, and H. de Snoo, Schur functions, operator

colligations, and reproducing kernel Pontryagin spaces, Operator Theory: Adv.
Appl. 96, Birkhauser Verlag, Basel, 1997.
[ADy] D. Alpay and H. Dym, On applications of reproducing kernel spaces to the
Schur algorithm and rational J-unitary factorization, Operator Theory: Adv.
Appl. 18, Birkhauser Verlag, Basel, 1986, 89-159.
[AzCD] T.Ya. Azizov, B. Curgus, and A. Dijksma, Standard symmetric operators in
Pontryagin spaces: A generalized von Neumann formula and minimality of
boundary coefficients, J. Functional Anal. 198 (2) (2003), 361-412.
[AzI] T.Ya. Azizov and 1.S. Iokhvidov, Foundations of the theory of linear operators
in spaces with an indefinite metric, Nauka, Moscow, 1986 (Russian); English
transl. Linear operators in spaces with an indefinite metric, Wiley, New York,
1989.
[dB] L. de Branges, Hilbert spaces of entire functions, Prentice Hall, Englewood
Cliffs, New Jersey, 1968.
[De] V.A. Derkach, On generalized resolvents of Hermitian relations in Krein Spaces,
J. Math. Sciences, 97 (1999), 4420-4460.
[DeH] V.A. Derkach and S. Hassi, A reproducing kernel space model for N" -functions,
Proc. Amer. Math. Soc. 131 (12) (2003), 3795-3806.
[DeHS1] V. Derkach, S. Hassi, and H. de Snoo, Operator models associated with Kac
subclasses of generalized Nevanlinna functions, Methods of Functional Analysis
and Topology 5 (1) (1999), 65-87.
[DeHS2] V. Derkach, S. Hassi, and H. de Snoo, Operator models associated with singular
perturbations, Methods of Functional Analysis and Topology 7 (3) (2001), 1-2l.
[DeHS3] V. Derkach, S. Hassi, and H. de Snoo, Rank one perturbations in a Pontryagin
space with one negative square, J. Funet. Anal. 188 (2002), 317-349.
[DLLSh] A. Dijksma, H. Langer, A. Luger, and Yu. 8hondin, A factorization result for
generalized Nevanlinna functions of the class N", Integral Equations Operator
Theory 36 (2000), 121-125.
[DLS1] A. Dijksma, H. Langer, and H.S.V. de Snoo, Hamiltonian systems with eigen-
value depending boundary conditions, Operator Theory: Adv. Appl. 35 (1988),
37-83.
[DL82] A. Dijksma, H. Langer, and H.8.V. de 8noo, Eigenvalues and pole functions
of Hamiltonian systems with eigenvalue depending boundary conditions, Math.
Nachr. 161 (1993), 107-154.
[D8] A. Dijksma and H.8.V. de 8noo, Symmetric and self-adjoint relations in Krdn
spaces I, Operator Theory: Adv. Appl. 24 (1987), 145-166.
[IKL] 1.8. Iokhvidov, M.G. KreIn, and H. Langer, Introduction to the spectral theory
of operators in spaces with indefinite metric, Akademie-Verlag, Berlin, 1982.
[KL1] M.G. KreIn and H. Langer, Uber die Q-Funktion eines 7r-hermiteschen Opera-
tors im Raume II", Acta Sci. Math. (8zeged) 34 (1973), 191-230.
[KL2] M.G. KreIn and H. Langer, Uber einige Fortsetzungsprobleme, die eng mit der
Theorie hermitescher Operatoren im Raume II" zusammenhiingen. 1. Einige
Funktionenklassen und ihre Darstellungen, Math. Nachr. 77 (1977), 187-236.
[KL3] M.G. KreIn and H. Langer, Some propositions on analytic matrix functions
related to the theory of operators on the space IT"" Acta Sci. Math. (Szeged) 43
(1981), 181-205.
[L] H. Langer, A characterization of generalized zeros of negative type of functions
of the class N"" Operator Theory: Adv. Appl. 17, Birkhauser Verlag, Basel,
1986, 201-212.
[LLu] H. Langer and A. Luger, A class of 2 x 2-matrix functions, Glas. Mat. Ser. III
35 (55) (2000), 149-160.
[M] M.M. Malamud, On a formula for the generalized resolvent of a nondensely
defined Hermitian operator, Ukrain. Mat. Zh. 44 (1992), 1658-1688.(Russian)
Aad Dijksma
University of Groningen
P.O. Box 800
NL-9700 AV Groningen, The Netherlands
~mail: a.dijksma@math.rug.nl
Heinz Langer
Wiedner Hauptstrasse 8-10
A-1040 Vienna, Austria
~mail: hlanger@mail.zserv.tuwien.ac.at
Annemarie Luger
Wiedner Hauptstrasse 8-10
A-I040 Vienna, Austria
~mail: aluger@mail.zserv.tuwien.ac.at
Yuri Shondin
Department of Theoretical Physics
Pedagogical State University
Str. Uly'anova 1, GSP 37
Nizhny Novgorod 603950, Russia
~mail: shondinlDshmath.nnov.ru
Operator Theory:
Advances and Applications, Vol. 154,91-100
Eigenvalues and Spectral Gaps Related to

Periodic Perturbations of Jacobi Matrices
Joanne Dombrowski
Abstract. This paper investigates the existence of eigenvalues in the spectral

gaps of a class of Jacobi matrices resulting from periodic perturbations of
Jacobi operators with smooth coefficients.
Mathematics Subject Classification (2000). 47B36, 47B15, 47B39 .

Keywords. Jacobi matrices, eigenvalues, spectral gaps.
1. Introduction
This paper considers infinite Jacobi matrices determined by a real sequence {an},
an =I- 0, of the following form:
al 0 0
al 0 a2 0
[ 00
C~ a2 0 a3 (1.1)
0 a3 0
Dc = {x = {x n } E 2: Ox E 2}.
In general it will be assumed that an = jn + Cn where {in} is monotone increas-
ing with limn->co jn = 00, {in - jn-I} is bounded, and Cn+2 = Cn. In this case,
co
Carleman's condition '" 6 an = 00 implies that 0 is self-adjoint on the indicated
~
n=1
maximal domain. If C is self-adjoint, then the spectral theorem asserts that it is
unitarily equivalent to a multiplication operator Mx : D ---+ L 2 (J.L) defined on a
dense subset D of L 2 (J.L) by Mx : f(x) ---+ xf(x). If 0 = fAdE).., then the Borel
measure J.L is defined by J.L(f3) = \\E(,B)<PI\\2, where <PI is the first standard basis
vector. The vector <PI is a cyclic vector since the subdiagonal entries of the matrix
are non-zero.
92 J. Dombrowski
The given sequence {an} uniquely determines a set of polynomials {Pn(X)}

defined as follows:
P1(X) = 1,P2(x) = x/a 1
Pn+1(x) = [XPn(x) - an-1Pn-1(X)j lan, n;::: 2. (1.2)
When (1.1) defines a self-adjoint operator, these polynomials form an or-
thonormal basis for the corresponding Hilbert space 2(J.L). They are useful in
establishing the results that follow.
The results in this paper are closely related to those in [DJMPj, which dis-
cusses the existence of spectral gaps under similar assumptions on the sequence
{an}. Partial results are presented on the existence of eigenvalues in these gaps.
This paper looks specifically at the eigenvalue problem, using an approach which
can simplify some of the arguments in [DJMPj. Theorem 3.3 and Theorem 3.5 be-
low extend Corollary 3.10 in [DJMPj to the case ~ < p < 1. Related ideas appear
in [DP02bj.
2. Preliminary results
This section considers the action of the operator C, and its spectral projections,
on a particular class of vectors, which includes the eigenvectors. The next lemma
generalizes some results established in [DP02aj and [DP02bj.
Lemma 2.1. Let C be a self-adjoint Jacobi matrix generated by some positive
sequence {an}. Then the spectrum of C is symmetric about the origin. If C =
J 'xdE,>.and 'ljJ is any vector in Dc such that E(O,oo)'ljJ = 'ljJ(orE( -00, O)'ljJ = 'ljJ)
00
and 'ljJ = E ('ljJ, <(In) <{In, then

n=l
L L
00 00
1('ljJ, J2n-l) 12 = 1('ljJ, J2n) 12

n=l n=l
Proof. Let {<{In} be the standard basis for 2. Define U : 2 -+ 2 by U <{In =
(-l)n+l<{Jn' Then UCU- 1 = -C and it follows that the spectrum of C is symmet-
ric about the origin. Viewed as an operator on 2(J.L), UPn(x) = (_l)n+lPn(x) =
Pn ( -x). It follows that U'ljJ(x) = 'ljJ( -x). If E(O,oo)'ljJ = 'ljJ(orE( -00, O)'ljJ = 'ljJ),
then ('ljJ, U'ljJ) = O. Hence
00 00
0= ('ljJ,C'ljJ) = L 1('ljJ,<{J2n-l)1 2 - L 1('ljJ,<{J2n)1 2

n=l n=l
3. Eigenvalues
The results in this section address the existence of eigenvalues. In some cases it is
also possible to obtain results on absolute continuity. In the first theorem x = y
Eigenvalues and Spectral Gaps 93
and so there is no gap. It will be shown that there are non-zero eigenvalues, since
the spectral measure is absolutely continuous on (-00,0) U (0,00).
Theorem 3.1. Let {jn} be a monotone increasing sequence of real numbers with
difference sequence {d n } ,dn = jn - jn-l, n 2: 2. Assume {d n } is bounded, ~d3 S; d 2
and ~(dn+1 + dn- 1) S; dn , n > 2. Let a2n-l = j2n-l + x, a2n = hn + x with x
chosen so that al - ~d2 > 0. Then C defined by (1.1) is self-adjoint and the spectral
measure of C is absolutely continuous on (-00,0) U (0,00).
Proof. Since {d n } is bounded, Carleman's condition implies that C is self-adjoint.

Choose a bounded interval~, such that ~ C (0,00). Let 'I/J = E(~)<Pl,'l/Jn =
('I/J, <Pn). Let S be the unilateral shift defined on p2 by Sn = n+1, n = 1,2, .... If
J = (S - S*) / i and K is defined by the commutator equation C J - J C = - 2iK
it follows that
L jn-I) l'l/Jnl 2+ L
00 00
(K'I/J,'I/J) = all'I/J11 2+ (jn - (jn - jn-l)'l/Jn-l'I/Jn+l

n=2 n=2
n=2
1 2 1 2~ 1 1 2
2: (al - "2d2) l'l/Jll + (d2 - "2d 3 ) 1'l/J21 + ~ (dn - "2dn+1 - "2dn-1) I'l/Jnl
n=3
1 2 1 4
2: (al - "2d2) l'l/Jll = (al - "2d2) IIE(~)<pIil .
On the other hand, if .>.. is the midpoint of the interval ~, then CJ - JC
(C - '>"I)J - J(C - '>"I), and if I~I denotes the Lebesgue measure of the interval
~ it follows that I(KE(~)l' E(~)l) I S; 21IJIIIIE(~)(C - '>"I)IilIIE(~)<Plll S;
IIJIII~IIIE(~)Iil2.
The two inequalities can be combined to show that if f.L(~) = IIE(~)1112 =I
then f.L(~) :::; C,I~~ldJ I~I so that the measure is absolutely continuous on
(-oo,O)U(O,oo). D
Example 3.2. Choose jn = n, x > - ~. The conditions of Theorem 3.1 are satisfied
since dn = 1 and al - ~d2 = 1 + x - ~.
Theorem 3.3. For 0< 0: < 1, let jk = kC>. For k 2: 1, let a2k-l = j2k-l + x, a2k =
j2k + y. Assume x > y 2: -(2 - 2C. Then (-Ix - yl, Ix - yl) is a gap in the
essential spectrum containing no non-zero eigenvalues.
94 J. Dombrowski
Proof. It follows from the results in [DJMP] that (- Ix - yl , Ix - yl) is a gap in

the essential spectrum. Thus it remains to consider the issue of eigenvalues. The
argument below uses the restriction of an eigenvector to the subspace spanned by
the odd subscripted basis vectors to show that any non-zero eigenvalue must be
outside the gap.
It follows from Lemma 2.1 above that for any vector 'I/J in Dc and>' =I=- 0 such
that C'I/J = >''I/J,
~ (C 2'I/J,'I/J) = ~ (C'I/J,C'I/J)
L
00
= [((2k - 1)" + X)'l/J2k-l + ((2k)" + y)'l/J2k+l]2

k=l
= L [((2k - 1)" + y)'l/J2k-l + ((2k)" + y)'l/J2k+1]2

00
k=l
+L
00
2(x - y) [((2k - 1)" + y)'l/J2k-l + ((2k)" + y)'l/J2k+l] 'l/J2k-l

k=l
L 'I/J~k-l
00
+ (x - y)2
k=l
L 2(x -
00
~ y)((2k - 1)" + y)'I/J~k-l

k=l
- 2(x - y) f= {~((2k)a + y)'I/J~k+l + ~((2k)a + Y)'I/J~k-l}
k=l
+L
00
(x - y?'I/J~k-l
k=l
~ 2(x - y) ~ [(2k _1)a - ~(2k)a + ~Y]'I/J~k-l

- 2(x - y) f
k=2
~((2k - 2)" + y)'I/J~k-l + f
k=l
(x - y)2'I/J~k_1
~ 2(x - y) ~ [(2k - 1)a - ~(2k)a - ~(2k - 2t] 'I/J~k-l

1 a
+ (; (X -
00
2 2
y) 'l/J2k-l + 2(x - +"21 Y
1 - "2(2)]
y) [
2
'l/Jl
L (x -
00
~ y)2'I/J~k_1 .
k=l
D
Corollary 3.4. If x > y > -(2 - 2"') then (C21/J,1/J) > (x - y)2111/J112 and so the
endpoints of the gap Ix - yl cannot be eigenvalues.
Theorem 3.5. For 0 < a < 1, let jk = kO:. For k?:. 1,let a2k-1 = j2k-1 + x, a2k =
12k + y. Assume y > x ?:. _(20:+1 - 3"'). Then (-Ix - yl , Ix - yl) is a gap in the
essential spectrum containing no non-zero eigenvalues.
Proof. As in the proof above, it follows from the results in [DJMP] that
(- Ix - yl , Ix - yl) is a gap in the essential spectrum. It remains then to con-
sider the issue of non-zero eigenvalues. The argument below uses the restriction of
the eigenvector to the subspace spanned by the even subscripted basis vectors. It
will again be shown that the corresponding eigenvalue must be outside the gap.
It follows from Lemma 2.1 that for any vector 1/J in Dc and A#-O such that
C1/J= A1/J,
~ (C 21/J,1/J) = ~ (C1/J,C1/J)
11/J212 + L
CXJ
= ai [a2k1/J2k + a2kH 1/J2k+2]2

k=1
CXJ
CXJ
k=1
L
CXJ
+ 2(y - x) [((2k)'" + X)1/J2k + ((2k + 1)'" + X)1/J2k+2]1/J2k

k=l
L
CXJ
+ (y - x)2 11/J2k1 2
k=l
CXJ
k=1
CXJ
k=1
- 2(y - x)~ f
k=2
((2k - I)'" + x) 11/J2k1 2+ f
K=l
(y - x)211/J2kI 2
96 J. Dombrowski
:::: (1 + x)21'lhI 2 + 2(y - x) L ((2k)<> -

00 1 1
2(2k + 1)<> - 2(2k - 1)< 1'2k1 2
k=2
+ 2(y - x)((2)<> - ~(3t + ~x) 1'21 2 + f

k=l
(y - x)21'2kI 2 .
o
Corollary 3.6. Ify > x > _(2<>+1 - 3< then (C2',') > (y - x)211,112 and thus
the endpoints of the gap Ix - yl cannot be eigenvalues.
It remains to consider the case a = 1. The next theorem summarizes results
established in [DJMPj.
Theorem 3.7. Choose x> -1,y > -2. Let a2n-1 = (2n -1) + x,a2n = (2n) + y.
Then (-Ix - yl , Ix - yl) is a gap in the essential spectrum of C containing at most
2 eigenvalues. Furthermore, 0 is an eigenvalue if and only if y > x. If y > x then
o is the only eigenvalue in the spectral gap.
With relatively minor changes in the proof, this result can be improved to
allow the possibility of negative entries in (1.1) so that the sequence {an} is still
monotone increasing with limn--->oo an = 00. An outline of the proof will be given
indicating the needed changes.
Theorem 3.S. Let a2n-1 = (2n - 1) + x, a2n = (2n) + y. Assume an =I- OVn. Then
(-Ix - yl , Ix - yl) is a gap in the essential spectrum of C. Let N be the smallest
positive integer so that N + eX ;l) :::: 0, N + ~ :::: 0 Then the spectral gap contains
at most 2N eigenvalues. Furthermore, 0 is an eigenvalue if and only if y > x.
Proof Carleman's condition implies that C defined by (1.1) is self-adjoint. Let
A + denote the restriction of C 2 to the subspace spanned by the basis vectors
{ <P2k- d %"=1 Then A + is tridiagonal with subdiagonal sequence a;'; = a2n-1 a2n
and diagonal sequence b;'; = a~n-2 + a~n-1. Obtain A- from A+ by negating the
diagonal entries. Define the diagonal operator D so that A - = .Ii - + D where
A has row sums equal to O. Then D<Pn = [-(a 2n - 2 + a2n-1) + (a2n-3a2n-2 +
-- 2 2
a2n-1a2n)j<Pn if ao = O.
(A-,,,) = (.Ii-,,'l + (D',')
L a~ l'n+1 -
00
=- 'n1 2 + (D', ')

n=l
=- L (2n - 1 + x)(2n + y) l'n+1 - 'n1 2 + (D', ')

n=l
Let N be the smallest positive integer so that N + (x;l) ::::: 0, N + ~ : : : O. Then for
any 'I/J = {'l/Jn} , 'l/Jn = 0, n = 1, ... , N,
( A-'l/J,'I/J ) ::; -4 ~
L....- (n +N +
x-I
-2-)(n
y
+ N + 2) I'l/Jn+l - 'l/Jnl
2
+ (D'I/J,'I/J)
n=l
::; (-4)~ f: l'l/Jnl 2 + (D'I/J,'I/J)

n=l
CXJ
n=l
CXJ
Hence (A+'I/J, 'I/J) ::::: (x - y)2 L l'l/Jnl 2 and it follows that (-Ix - yl , Ix - yl) is a
n=l
gap in the essential spectrum of C containing at most 2N eigenvalues. 0
The next result gives sufficient conditions for the existence of non-zero eigen-
values in the gap.
Theorem 3.9. Let a2n-l = (2n - 1) + x, a2n = (2n) + y. Assume an i=- OVn. If
x ::::: y and 11 + xl ::; Ix - yl then C defined by (1.1) has a non-zero eigenvalue in
the gap (-Ix - yl, Ix - YI)
Proof. It was shown in [DP02b] that if an+l - an ::::: 0 with J = a2n - a2n-l, d =
a2n+1 -an, then 0 is an eigenvalue if and only if J > d. In this case, it follows that 0
is an eigenvalue if and only if y > x. Also, replacing an by lanlleads to a unitarily
equivalent operator. Note that A = 11 + xl is a root for P 3 (A). Let S(A) = Pn(A).
Then the modified Wronskian Wn(S(A),S(-A)) = (-1)n2a n Pn (A)Pn+1 (A) has at
least one node. Therefore by the results in [T96] the interval (-A, A)contains at
least one eigenvalue, which in this case must be non-zero. 0
Remark 3.10. If x> -1, and x::::: y, then 11 + xl ::; Ix - yl implies that y ::; -1.
If x::; -1 and x::::: y, then 11 + xl ::; Ix - yl implies that y ::; 2x + 1.
Example 3.11. Let x = - ~, y = -1. Then the spectral gap is ( - ~, ~) and it follows
from the above result that there is at least one non-zero eigenvalue in the spectral
gap. By symmetry there must be at least two eigenvalues in the spectral gap, and by
Theorem 3.9, with N = 2, there are exactly two non-zero eigenvalues in this gap.
4. Special case: j2n-l = j2n
In this section it will generally be assumed that a2n-l = n'" + x, a2n = n'" + y, 0 <
a::; 1, with x and y chosen so that an i=- OVn.
Theorem 4.1. For x::::: y ::::: 0, let a2n-l = n + x, a2n = n + y. Then for C defined
by (1.1)
(-I ~ + x - YI ' 1~ + x - yl)
is a gap in the essential spectrum no non-zero eigenvalues.
98 J. Dombrowski
Proof. It was shown in [DJMP] that +x - (-I! YI ,I!

+ x - yl) is a gap in the
essential spectrum. Thus it remains to consider the eigenvalues. The argument
below considers the restriction of a possible eigenvector to the subspace spanned
by the odd subscripted basis vectors. For any vector 'l/J in Dc and>' -I 0 such that
C'l/J = >''l/J.
"21 (2
C'l/J,'l/J ) ="21 (C'l/J,C'l/J)
L
00
= [(k + X)'l/J2k-l + (k + y)'l/J2k+1]2

k=l
= L [(k + y)'l/J2k-l + (k + y)'l/J2k+l + (x - y)'l/J2k_l]2

00
k=l
00
k=l
00 00
k=l k=l
00
~ L k 2 1'l/J2k-l + 'l/J2k+11 2
k=l
+ 2(x - y) ~[ 2
~ (k + y)'l/J2k-l 1 + y)'l/J2k+l
- "2(k 2 1 + y)'l/J2k-l
- "2(k 2]
00
k=l
00
~ L k 2 1'l/J2k-l + 'l/J2k+11 2
k=l
+ 2(x - y) f=
k=l
D(k + y)'l/J~k-l - ~(k + Y)'l/J~k+l] + (x - y)2 f=
k=l
1'l/J2k-11 2
~ ~ k 2 1'l/J2k_l + 'l/J2k+11 2 + (x - y) [(1 + y)'l/Jr + ~ 'l/J~k+ll

00
k=l
D
Corollary 4.2. If x > Y > 0, then the endpoints of the gap I~ + x - YI cannot be
eigenvalues.
Proof. In this case the above proof shows that for any vector 7/J in Dc and A =I- 0
00 2
suchthatC7/J=A7/J,(C27/J,7/J(~+x-y)2L:I7/Jkl. 0
k=l
Remark 4.3. It should be noted that it follows from Theorem 2.1 in [DP02] that if
x> -1,0 :S y-x :S I, y+~(y-x) > 0 then there are no non-zero eigenvalues in the
I I' I
spectral gap (- ~ + x + Y ~ + x + Y I),
since the spectral measure is absolutely
continuous. Furthermore, a careful look at the proof indicates that the conclusion
holds if x =I- -1 and 0 :S y - x :S l,y + ~(y - x) > O. Thus, for example, the
conclusion holds if x = -~, y = -~. In this case a1 = 1 + x = -~.
Theorem 4.4. For x > y > -I, let a2n-1 = nO: + x, a2n = nO: + y, 0 < ex < 1. Then
for C defined by (1.1), (-Ix - yl , Ix - yl) is a gap in the essential spectrum, and
the closed interval [-Ix - yl , Ix - yl] contains no non-zero eigenvalues.
Proof. It was shown in [DJMP] that (- Ix - yl , Ix - yl) is a gap in the essential
spectrum. Thus it only remains to consider the eigenvalues. The argument below
uses the restriction of a possible eigenvector to the subspace spanned by the odd
subscripted basis vectors. For any vector 7/J in Dc and A =I- 0 such that C7/J = A7/J.
1 2 1
2'C 7/J,7/J) = 2,C7/J,C7/J)
00
k=l
00
00
k=l
00
k=l
00
k=l
00
k=l
00 00 00
k=l k=l k=l

100 J. Dombrowski
= =
2: (x - y)(l + y)'lj;i + (x - y) L [((k + 1)<> - k<'lj;~k+l] + (x - y)2 L 'lj;~k-l
k=l k=l
=
> (x - y)2 L 'lj;~k-l.
k=l
o
References
[B68] J. Berezanskii, Expansions in Eigenfunctions of Self-adjoint Operators, Trans.
Math. Monographs. 17 Amer. Math. Soc. Providence, RI 1968.
[D87] J. Dombrowski, Cyclic operators, commutators and absolutely continuous mea-
sures, 100 (1987), 457-463.
[DP95] J. Dombrowski and S. Pedersen, Orthogonal polynomials, spectral measures and
absolute continuity, J. Comput. Appl. Math 65 (1995), 115-124.
[DP97] _ _ , Spectral measures and Jacobi matrices related to Laguerre type systems
of orthogonal polynomials, Constr. Approx. 13 (1997), 421-433.
[DP02a] _ _ , Absolute continuity for Jacobi matrices with constant row sums, J. Math.
Anal. Appl. 277 (2002), 695-713.
[DP02b] _ _ , Spectral Transition parameters for a class of Jacobi matrices, Studia
Math. 152 (2002), 217-229.
[DJMP] J. Dombrowski, J. Janas. M. Moszynski and S. Pedersen, Spectral gaps resulting
from periodic perturbations of a class of Jacobi operators, Constr. Approx. (to
appear.)
[RS78] M. Reed and B. Simon, Methods of Modern Mathematical Physics IV: Analysis
of Operators, Academic Press, New York, 1978.
[T96] G. Teschl, Oscillation theory and renormalized oscillation theory for Jacobi op-
erators, Jour. Differential Equations 129 (1996), 532-558.
Joanne Dombrowski
Wright State University
Dayton, OH 45435, USA
e-mail: joanne. dombrowski<owright. edu
Operator Theory:
WKB and Turning Point Theory for

Second-order Difference Equations
Jeffrey S. Geronimo, Oscar Bruno and Walter Van Assche
Abstract. A turning point method for difference equations is developed. This

method is coupled with the LG-WKB method via matching to provide ap-
proximate solutions to the initial value problem. The techniques developed
are used to provide strong asymptotics for Hermite polynomials.
Mathematics Subject Classification (2000). Primary 39AlO; Secondary 41A60,

65Q05.
Keywords. Thrning point, difference equations, WKB, orthogonal polynomi-
als, hermite polynomials.
1. Introduction
In this article we will be interested developing a turning point theory for difference
equations of the form
a + 1 1(x) +(b-x)u(x)+au-(x) = 0 (1.1 )
with an > 0, and bn real and matching the solutions obtained from this theory to
those obtained from the LG-WKB method. Various asymptotics for the solutions
of (1.1) with 'l/J-l = 0, 'l/Jo = 1 when the coefficients tend to infinity in magnitude
such as nth root asymptotics, ratio asymptotics, strong asymptotics, and weak
asymptotics, have been investigated by a large number of authors (Braun [B],
Geronimo-Smith [GS], Maejima-Van Assche [MV], Nevai-Dehesa [ND], Schulten-
Gordon [SG], Van Assche [V], Van Assche-Geronimo [VG], etc.). However with
the exception of weak asymptotics these studies required x to be exterior to the
region where the solutions of the above difference equation oscillate. In order to
control the growth of the coefficients it was suggested in Nevai-Dehesa and further
developed in Van Assche, and Van Assche et al. that the preliminary change of
J.S.G. was supported by NSF grants. J.S.G. and W.V.A. were supported by a NATO grant.
102 J.S. Geronimo, O. Bruno and W. Van Assche
variables x = A/y be performed so that a(n, l) = ~7 and b(n, l) = ~7 for n :S l are

bounded functions of nand l. This produces the difference equation
a(n + 1, l)-J;(n + 1, y, l) + (b(n, l) - y)-J;(n, y, l) + a(n, l)-J;(n - 1, y, l) = O. (1.2)
To study the solutions of (1.2) in the oscillatory region and near the turning
points a WKB method based upon an epsilon difference equation (see equation
2.6) was proposed by Deift and McLaughlin [DM] and further developed by Costin
and Costin [Ce]o While satisfactory away from the turning points, i.e., where
y:t~~)l) ~ 2, this technique requires the matching of various solution of (1.2) in
neighborhoods of the turning points, that shrink as l-1/2 as l tends to infinity. As
noted by Wang and Wong [WW] in their study of the difference equation satisfied
by Bessel functions the above theory does not lead to a satisfactory uniform asymp-
totic expansion for Bessel functions. Other methods to study (1.2) have been
proposed by Dingle and Morgan [DM] and Schulten and Gordon [SG] , and we will
use these methods to modify the technique of Deift and McLaughlin. We proceed
as follows: in Section 2 we motivate the turning point technique to be discussed.
In particular a Langer transformation for difference equations is proposed. In the
next section (Section 3) this technique is put on firm ground and it is extended
to the complex plane in Section 4. One of the main drawbacks of the techniques
of the preceeding sections is that they cannot be used to solve the initial value
problem for coefficients in (1.2) that tend to zero as l tends to infinity, for instance
when the coefficients in (1.1) tend to infinity. In order to overcome this we recall,
in Section 5 the WKB method proposed by Deift and McLaughlin away from the
turning points and following Geronimo and Smith modify it to be applicable to
the initial value problem. In Section 6 these techniques are used to obtain uniform
asymptotics for Hermite polynomials. In a sequel to this article [G] the theory of
external fields will be used to help obtain uniform asymptotics for other sets of
orthogonal polynomials including those associated with discrete measures.
2. Motivation
a. Differential equations - the Langer transformation
Consider the differential equation,
(:2 - f- 2k(t)2) (t) = 0 k: R --+ C (2.1)
where k 2 : R --+ R is assumed to be a smooth monotonically increasing function

of t in [tin, tfi] tin = tinitial, tfi = tfinal with one simple zero, to, in the region.
Away from to the LG-WKB method ([0, p. 191]) gives two solutions
(t) = 1 eE-" l' k(u)du(1 + O(f)).
(lk2(t)l)t
While adequate in this region the asymptotics provided by these solutions breaks
down in a neighborhood of to. The standard procedure introduced by Langer([L],
WKB and Turning Point Theory 103
[0, p. 398]) to overcome this difficulty is to make the change of variables
2
_p3/2(t)
3
= it
to
k(u)du (2.2)
which yields the equation,

d2
dp2 = (E- 2p + 8(p)),
where = (p')1/2, 8 = k- I / 2!?k l / 2, with k2 = k;.

Langer showed that under
suitable conditions there are two solutions of (2.1) which provide uniform asymp-
toties in a neighborhood of to, namely
I = k- I / 2Ai(E- 2/ 3 p) + O(E)!(E- 2/ 3 p),
and
2 = k- I / 2Bi(E- 2/ 3 p) + O(E)!(E- 2/ 3 p)-1,
where the Airy functions Ai( C 2/ 3 p) and Bi( C 2/ 3 p) satisfy the differential equa-
tion,
(2.3)
and ! is a function that decays at the same rate as Ai for large values of its
argument.
b. A model equation
In order to find an analog of the Langer transformation for difference equations as
well as the corresponding solutions we begin by considering the model equation
'IjJ(t + E) + 'IjJ(t - E) - 2 cosh k(t)'IjJ = O.
Following Schulten and Gordon we search for approximate solutions by making the
ansatz 'IjJ(t) = g(t)Ai(C 2/ 3p(t)) where the functions g and p will be chosen so that
'IjJ satisfies the difference equation up to order E2. Roughly speaking this will allow
us to show that there are actual solutions to the above difference equation that
are within E of these approximate solutions after c i steps. Using the asymptotic
formula for Ai (Olver[O] p. 392),
A( -2/3 P) -_ 2.j1rc
1 E
e-2/3e-lp3/2
l / 6pl/4
(1 _~)
~p3/2 +
O( 2)
E,
we find
With f = ~p3/2 write,
e C - ' f(t)e- c-' f(tc) = ef'(t) (1 _Ef~(t)) + O(E2).
Thus,
which implies
'ljJ(t + E) + 'ljJ(t - E) - 2coshk(t)'ljJ(t)
= (e-f'(t) + ef'(t) _ 2coshk(t))'ljJ(t) + E (f'(t) _ coshf'(t) f"(t) _ 29'(t))
3f( t) sinh f'(t) g(t)
e- c -' f(t) ( EJ.L)
X g(t) sinhf'(t) 2J1fc 1/ 6 p(t)1/4 1- f(t\ + O(E2).
In order to have only the O(E2) term remain choose f and 9 so that
!,(t) = k(t),
and
g'(t) 1 coshk(t)k(t)' p'(t)
g(t) = -2 sinhk(t) + 4p(t)
J
Thus
~p3/2 = k(t)dt,
and
t _ ( p(t) ) 1/4
g( ) - sinh2 k(t)
With this choice of p and g, 'ljJ satisfies the second order difference equation up to
order E2.
c. The Langer transformation for difference equations

For difference equations of the form
an+1'ljJn+1(X) + (bn - X)'ljJn(X) + an'ljJn-1(X) = 0, (2.4)
with an > 0, and bn real, we assume that an and bn are discretizations of the at
least three times differentiable functions a( u) and b( u) respectively. With u = ~
set
a( ~) b(~)
a(t,E) = ~ and b(t,E) =~' (2.5)
then (2.4) becomes the epsilon difference equation
a(t n+1, E);jJ(t n+1, y, E) + (b(t n , E) - y);jJ(t n , y, E) + a(t n , E);jJ(t n-1, y, E) = 0, (2.6)
where tn = nE, x = AY, .(j;(tn' y, E) = 'lj;n(AY), and A is chosen so that a(t, E)

and b(t, E) are bounded functions of E. In order to find the modifications to the
9 and k above write a(t + E, E) = a(t + E/2, E) + a1(t~/2,) + 0(E2) and a(t, E) =
a(t + E/2, E) - a1(t~/2,) + 0(E2) and set
Y - b(t,E)
coshk(t,y, E) = ( ) for t ~ to, (2.7)
2a t + 2' E
Y - b(t, E)
cosk(t,y,E) = ( ) for t ~ to.
2a t + 2' E
Here to is such that ~;;-(~~~';~ = 1. We now define the Langer transformation for
the above difference equation as
~p3/2(t,y,E)= y-b(U~E) )du

3 Jtotcosh-l( 2a(u+ 2 ,E)
for t ~ to, (2.8)
2
_(_p)3/2(t,y,E) =
ito cos- y - b(u E) du1 ~ for t ~ to.
3 t 2a (u + 2' E)
The function 9 can be written as
(2.9)
t < to,
and the corresponding approximate solutions are given by

'!/}(t, y, E) = g(t, y, E)Ai(C~ p(t, y, E)), (2.10)
and
'lj;2(t,y, E) = g(t,y,E)Bi(E-~p(t,y,E)). (2.11)
From (2.7) we can rewrite k as
y - b(t,E)
k(t, y, E) = In ( 2 ( ) + (2.12)
at + 2' E
where the branch of the square root is chosen so that In(z + v'z2-=-1) rv In z for z
large. Then (2.8) can be written in the compact form
"32 p"2(t,y,E) =
3 it k(U,y,E)du. (2.13)
to
We will suppose that for each fixed y E [Yl, Y2], there is an EO (y) > 0 so that for
each E E [O,EO(Y)]' k 2(t,y,E) is a monotonically increasing function oft with one
zero, to, in the interval [tin, tfi].
Also
t:i k2(t, y, E) E CO([tin' tf;] X [0, EO]) i = 0, ... ,3,

1.
k 2 (t,y,E) E C O([ tin,

a i t-to(y,E)
n. ati tfi ] X [0 ,EO ]).
,Z -- , 1 , 2 WIt
. h k 2 (t,y,E) strictly positive
t-to(y,E)
in [tin, tfi] X [0, EO] and,
sinh2(k(t,y,E)) 1 0 C
lll. I t-tO(y,E) > lor (t),E E [t in, t]
fi X
[0
,EO]
In order to obtain solutions that are continuous in Y we will suppose,
tt', k2(t, y, E) E CO([tin' tf;] [0, EO]), i = 0, ... ,3

ia. X [YI, Y2] X
iia. ai k 2 (t,y,E)
at' t-tO(y,E) E tin, t]
fi X [YI, Y2 ] X [0 ,EO ]) ,Z. -- , 1 , 2 WI'th
C O([ k 2 (t,y,E) t' I
t-to(y,E) S nct y
positive in [tin, tfi] X [Yl, Y2] X [0, EO], and
iiia. ISin;~~:~~',~iE))1 > for (y,t,E) E [tin,tfi] X [Yl,Y2] X [O,EO]'
Finally we will suppose that
ai
ati a(t, E) E CO([tin' tfi] X [0, EO]), i = 0, ... ,3. (2.14)
The turning points are located at (2~(t~~:~))2 = 1 which will be either at

Y= E)+b(t, E) = 1'+(t, E) or y = b(t, E)-2a(t+~, E) = 1'-(t, E). Conditions ii
2a(t+~,
and iii impose that for each fixed Y there is a unique simple zero of k 2, to E [tin' tf;]
such that Y = 1'+ (to, E) and the second turning point is not encountered.
If
(2.15)
and
B(E) = sup 1'+(t, E), (2.16)
tE [tin ,t fd
then we see from above that for each fixed E in order for Y to be a turning point
it must be in the interval [A(E), B(E)].
The next section is devoted to making the arguments leading to (2.10) and
(2.11) rigorous.
3. Error analysis
We begin by considering the Langer transformation (2.13). The lemma below which
follows from a lemma of Olver ([0] p. 399) shows that smoothness in k is transferred
to smoothness in p.
Lemma 3.1. Sl1,ppose that for (y,E) E [Yl,Y2] X [O,EO], k 2(t,y,E) is montonically
increasing for t in [tin' tfi] and satisfies i and ii. Then equation (2.13) gives a one
to one map between t and p such that p satisfies i and ii. If it is assumed that k 2
satisfies ia and iia then so does p.
We now examine the solution to the Airy differential equation. From (2.3) it
can be seen using Taylor series that X(C 2/ 3p(t + E)) and X(C 2/ 3p(t - E)) can be
written in terms of X(C 2/ 3p(t)) and X'(C 2/ 3p(t)). With
P = P(t E) - p(t) , (3.1)
E
the methods of Wang and Wong [WW, Lemma 1] can be slightly modified to
obtain
Lemma 3.2. Suppose X is any solution of (2.3) with p and p continuous then
X(E- 2/ 3p(t E)) = X(E- 2/ 3p(t) EI / 3p) (3.2)
= X(E- 2/ 3p(t))XI (p, P, E) EI/ 3X'(E- 2/ 3p(t))X2(p, P, E),
with 00
Xi(p,P,E) = ~)E)nXi,n(P'P) i = 1,2, (3.3)

n=O
where for p > 0,
sinh ..jPP
Xl ,o(p, p) = cosh V
'PP,
P 2 o(p, p) =
X, ..jP
P , (3.4)
and
Xi,k(p,p) = ~1P sXi,k_l(p,s)sinh..jP(p-s)ds. (3.5)
Above if p < 0 then # is replaced by i N .

Also using induction one finds (see Wang and Wong [WW])
IX1,k(p,p)1 ~ 3k (~) k ~!r;! el~(v'Pp)l, (3.6)

and
IX 2 k(p p-)I < 3k (~) Ip13k+1 el~(v'Pp)l. (3.7)
, , - 3 k (3k + I)!
By taking the partial of (3.2) with respect to P it is not difficult to see that
X'(E- 2/ 3p(t) EI/ 3p) = E-I/3X(E-2/3p(t))X3(p, P, E) (3.8)
+ X'(E- 2/ 3p(t))X4 (p, P, E),
where
(3.9)
and
8
X 4 = 8pX2. (3.10)
Equations (3.4) and (3.5) imply that
_ pi. _ P _ 1. _
XII(P,P)
, = -4# smh..jPp - -4p cosh..jPp + ~/2
4p
smh..jPP,
and
-2 -
-)
X 2,1 ( p, P P h r;;- P. h r;;-
= 4yIP cos V PP - 4 p3/2 sm v PP
If p can be expanded as
_
p=pt
( )' Ep(t)"
--+r,
A
(3.11)
2
with
(3.12)
then
ylPP"
X 1 0 - cosh ypp' =f E -2- sinh ypp' (3.13)
= f sinh ypp' + f!..

2
lp'
p
cosh yps(p - s )ds,
X1 1 (p') 2 . h r;; I p' h r;; I 1 . h r;; I (3.14)

- -- sm v pp + - cos v pp - ~/2 sm v pp
, 4yIP 4p 4p
= lp'
p S
( - sinh yps + - cosh yps) ds,
4yIP
s2
sinh ylPp' p" I I

X 20 Ipp
=f E2- cosh VI' Ipp
r cosh VI' (3.15)
A
, -
yIP =
+ -yIP
2
l p'
p
sinh yps(fJ - s)ds,
and
(p')2 p'
X 21 - -- cosh ypp' + ~/2 sinh ypp' (3.16)
, 4p 4p
= lpI
p S2 1
((- - -)sinhyps+ -
4 4p 4 yIP
s
coshyps)ds.
If p < 0 then replace yIP by iA in the above formulas.

The theorems below makes rigorous the argument of Shulten and Gorden and
show that the Langer transformation plays a role in difference equations similar
to that in differential equations.
Set
and
W = clos(W).
Theorem 3.3. Suppose that for each fixed Y E [Y1, Y2], k 2 is monotonically increas-
ing in t with a single zero, to E [tin, tfi], and satisfies i, ii, and iii. Also suppose
a(t, f) is strictly positive and satisfies (2.14). Let 1/i(t, y, f) i = 1,2 be given by
equations (2.10) and (2.11) respectively. Then for all (t, f) E W,
a(t + f, f)1/}(t + f, y, f) + a(t, f)'ljJi(t - f, y, f) (3.17)
- 2a(t + f/2, f) coshk(t, y, f)'ljJi(t, y, f)
= j3(i)(t,y,f) i = 1,2
where for each y E [Y1, Y2], 'ljJi E CO([tin, tfi] X (0, fO]), and j3(i) E CO(W). Further-
more,
\j3(l)(t, y, f)\ ~ C(Y)f2 sup [\Ai(f- 2/ 3p(u, y, f))\ + f1/3\Ai'(f-2/3 p(U, y, f))\],
UE(t-E,t+E)
while
\j3(2)(t,y,f)\ ~ C(Y)f2 sup [\Bi(f- 2/ 3p(u,y,f))\ + f 1/ 3 \Bi'(f- 2/ 3p(u,y,f))\].
UE(t-E,HE)
If it is assumed that ia-iiia. hold then'ljJi E CO ([tin' tfi] X [Y1, Y2] x (0, fO]), and
j3i E CO(W X [Y1, Y2]).
Proof. We will only consider the case i = 1 as the i = 2 case is proved in a similar
manner. From the Lemma 3.1 it follows that k:~~)t) satisfies i. and is nonzero for
y E [Yl, Y2] and (t, f) E [tin' tfi] X [0, fO]. From the hypotheses on k we see also that
(sinkhk)2 satisfies i in the same region. This implies g(t) satisfies i and is nonzero.
Lemma 3.2 says that
Ai(f- 2 / 3 p(t f)) = Ai(f- 2 / 3 P(t))Xl (p, P, f)
f1/3 Ai' (f- 2 / 3 p(t))X2(p, P, f)
(3.18)
where P has the expansions given in (3.11). Equations (3.13) through (3.16),
conditions i and ii and equations (3.6) and (3.7) imply that
yIP"
\X1,0 - coshypp' =f fsinhyp/ ~p \ < C(Y)f2, (3.19)
\X1 1 - (p')2.
-4- sm h y r;;'
pp p'
+ -4 cos h y r;;pp, - ~/2
1 . h y r;;pp'\ < c (Y) f,
sm (3.20)
, yIP p 4p
sinh ypp' p"

\X2,0 - yIP =f f cosh YPP'2\ < C(Y)f2, (3.21)
(p'? p" p'

\X2,1 - ~ cosh ypp' - 2 cosh ypp" + 4p3/2 sinh ypp'\ < C(y)f, (3.22)
and
where c(y) is independent of (t, E) E W. Thus,
Ai(E-2/3 p(tE)) = (COShfpP' (3.23)
E [(p')2 sinhfpp' + ypp" sinhfpp'

4yp 2
-!!.-
4p
coshfpp' +
4p
~/2 sinhfpp'])Ai
Sinh ypp ' [(p')2 p" p' ]) Ai'

E1 / 3 ( E - - cosh fpp' + - cosh fpp' - ~/2 sinh fpp'
yp 4p 2 4p
+ e(t, y, E).
From the above estimates and the convergence of the series (3.3) we obtain the
bound
le(t, y, E)I :::; E2Cl(Y) sup [lAi(E- 2/ 3p(u, y, E))I + El/3IAi'(E-2/3p(U, y, E))ll
uE(t-,t+)
Suppose for now that p "I- O. If g(tE) and a(t~) are expanded out to second order
a'(t+2) cosh(
in E we find using the equation' g[L = 14 E-
1/2 ') -1/2
_ 1 _ _ _ . p 1 2p l (P
2 P +pl 2p")
1 1 12 /
P 2 a(t+~) 2smh(p / p )
that
a(t + E)(gAi)(t + E) + a(t) (gAi) (t - E) - (y - b(t, E))(gAi)(t) (3.24)
= a (t + ~) g(t) [Ai(E- 2/ 3p(t + E)) + Ai(E- 2/ 3p(t - E))
- 2cosh(pl/2P')Ai(C 2/ 3p(t))
1 p' COSh(pl/2p') (p-l/2 p,2 1 2 "))

+E ( - - - +p/ P
4 p 2 sinh(pl/2 p') 2
x (Ai(E- 2/ 3p(t + E)) - Ai(E- 2/ 3p(t - E))) 1

+el(t,y,E),
where
lel(t,y,E)I:::; 2E2 sup [la'g'l + lag"l + Iga"l +E(la'g"l + la"g'l)

uE(t-,t+)
+ E21a" g"llIAi( E- 2/ 3p( U, y, E)) I

:::; E2C2(Y) sup IAi(E- 2/ 3p(u,y,E))I.
uE(t-,t+)
Here,
C2(y) = sup [la'g'l + lag"l + Iga"l + E(la'g"l + la"g'l) + E2Ia"g"ll

(t,)EW
WKB and Thrning Point Theory 111
This as well as (3.23) show that there is cancellation on the right-hand side of
(3.24) out to order E2 so that
a(t + E) (gAi)(t + E) + a(t) (gAi) (t - E) - 2a (t + E2) COsh(pI/2P') (gAi) (t)
= (3(1) (t, y, E), (3.25)
where
1(3(1)(t, y, E)I :s E2C(Y) sup [lAi(E- 2/ 3p(u,y, E))I + EI/3IAi/(E-2/3p(U,y, E))I]
uECt-,t+)
Since
(a(t+~) 1/2)
g(t)
I
= a (t +.:) g(t) (~pl _ C~Sh(pI/2pl) ) (p-I/2pI22 + pI/2pll) ,

2 4 p 2 smh(pl/2 p')
the restriction p =I- 0 can now be removed. An analogous bound for (3C2) (t, y, E)
holds with Ai replaced by Bi. The continuity properties follow from conditions
i-iii, (3.2) and the fact that the Airy functions are continuous (in fact entire)
functions of their arguments. The uniformity of the error when ia-iiia are imposed
follow by taking the supremum over y in the constants given and the continuity
properties follow as above. 0
To show that the above functions stay close to real solutions of the difference
equation we introduce some auxiliary functions.
Let u Cj ) (x) = g(x)u Cj ) (x) with u(i)(x) = (-:n
I / 2 e C- I )i+li7l"/6 Hi%(~( _x)3/2).
where Hi~;, Hi~; are the first and second kind Hankel functions of order one third.
Here we define ln z so that it is positive for large enough positive z. From the series
representation for Hi/3 we see that UCi) (x) is continuous and infinitely differentiable
for x real and each satisfies the Airy differential equation. The following give the
relations between u(j), Ai and Bi [0, pp. 250 and 392]. For x > 0,
Ai(x) 1 (X)I/2.
= -"2 u(1)(x) ="2"3 1 CI)
e 22 71"/3HI/3 (23x3/2e271"/2
.) ,
and
Bi(x) = (i)I/2Re (e-i7l"/6Hi~~ (~x3/2ei7l"/2)),

while for x < 0
Ai(x) = Re(u(1) (x)),
and
Bi(x) = -Im(u CI ) (x)).
112 J.S. Geronimo, 0. Bruno and W. Van Assche
For x < 0, u (2 )(x) = u(1)(x) (Olver [0] p. 238). Note that for large negative x we
find that
U
-Cj)( )
x
= _1_(_
fi x
)-1/4 C1)(j+1)(i~C-x)3/2_':f)
e
(1 + _1_)
Ix1 2/ 3 '
(3.26)
and
-Cj)( )'
U x
= _1_ (_
fi x
)1/4 (1)(J+l)Ci~C-x)3/2_T)
e
(1 + _1_)
Ix1 2/ 3 .
(3.27)
An important property of uCi ), is that they do not vanish for all x real [W]. Some
useful auxiliary functions we will use below are the modulus function, M, and
weight function, E, associated with the Airy functions. If Co < 0, Co rv -.36 is the
largest point on the x axis where Ai(x) = Bi(x) we find from Olver [0, p. 395]
that
Bi(X)) 1/2
M(x) = (2Ai(x)Bi(x))1/2, E(x) = ( Ai(x) , x 2: Co, (3.28)
and
M(x) = (Ai(x)2 + Bi(x)2)1/2, E(x) = 1, x :S Co
The Airy functions Ai and Bi can be expressed in terms of these functions as

E(x)Ai(x) = M(x) sinB(x), E-1(x)Bi(x) = M(x) sinB(x), (3.29)
where B is called the phase function. The modulus and weight functions have the
following asymptotic properties
2
V2e 3x2 ,
;J.
E(x) rv
for large positive x while, M(x) rv filxl- 1/ 4, for large x. The weight function is
a nondecreasing function of its argument while the modulus function is increasing
for x :S Co ([0, Lemma 5.1]). Since M is continuous there is a positive constant C3
such that for all x,
(3.30)
We also have [0, p. 395] that
7rlxI 1 / 2M(X)2 :S Al < 2. (3.31 )
Formulas similar to (3.29) also hold for the derivatives of the Airy functions. In
this case the functions M and B are replaced by Nand w respectively. For large
x, N(x) rv filxl 1 / 4 so from the continuity of N(x) we find that
N(x) :S c4(1 + Ixll/4) (3.32)
Furthermore it also follows from the above asymptotic formulas that there is a
constant, say C5, such that
E2(X)UCl) (x)
1 u(2)(X) I, (3.33)
That u(j) solve the same differential equations as the Airy functions implies
using the above arguments,
Lemma 3.4. Suppose that for each fixed y E [Yl, Y2], k 2 is monotonically increasing
in t with a single zero, to E [tin' t Ji], and satisfies i, ii, and iii. Also suppose a( t, E)
is strictly positive and satisfies (2.14). Then for all (t, E) E W,
a(t + E, E)U(j) (t + E, y, E) + a(t, E)U(j) (t - E, y, E) (3.34)

- 2a(t + E/2) coshk(t, y, E)U(j) (t, y, E) = f3i j ) (t, y, E), j = 1,2
where for each y E [Yl, Y2], u(j) E CO ([tin , tJi] X (0, EO]), and f3i j ) E CO(W).
Furthermore,
lf3ij )(t,y,E)1 ~ C(y)E2 sup (M(E- 2/ 3 p(U,y, E))

uE(t-e,t+e)
+ El / 3 N(E- 2/ 3 p(u, y, E)) )E( -1)1 (E- 2/ 3 p(u, y, E)).

If it is assumed that ia.-iiia. hold then u(j) E CO([tin, tJi] X [Yl, Y2] x (0, EO]), and
f3(j) E CO(W x [Yl,Y2]).
Since u(j) satisfy the Airy equation it is not difficult to compute their
Wronskian
With this we can obtain,
Lemma 3.5. Suppose that for each fixed Y E [Yl, Y2], k 2 is monotonically increasing
in t with a single zero, to E [tin' t Ji], and satisfies i, ii, and iii. Also suppose a( t, E)
is strictly positive and satisfies (2.14). Then for all (t, E) E w,
u(1) (t + E, y, E)U(2)(t, y, E) - u(l) (t, y, E)U(2) (t + E, y, E) (3.35)

1/3 2i _
-g(t + E, y, E)g(t, y, E)X2(p, p+, E).
_
- E
7r
For EO sufficiently small there is a constant c6 (y) such that
l3.7r g(t+ 1,y,E)g(t,y,E)X2(p,p+,E)I- l < C6(Y) (3.36)
If it is assumed that ia.-iiia. hold then C6 can be choosen uniform in y.
Proof. Equation (3.35) follows from the Wronskian identity above and (3.2) with
X replaced by u(i), i = 1,2. Conditions ii and iii imply that for EO sufficiently
small X 2(p,p+) ((3.4)) is nonzero. The convergence of the series for X 2 which
follows from (3.7) shows that X 2 is nonzero for EO sufficiently small which leads to
(3.36). 0
In the rest of the paper we will assume that EO is chosen sufficiently small so
that (3.36) holds.
We now apply the above results to show how close 'ljJj (tn' y, E) stays to a
solution of the difference equation
al((n + l)E, E)f(n + 1) + al(nE, E)f(n - 1) - (y - bl(nE, E))f(n) = O. (3.37)
h
W it - j - fn)-1j;j(n) . - 1 2 .7,j( ) _ 1j;j(n) j3'(j) - {3(j)(n) d
tn - En, (J"i - u(j)(n) Z- ... ,'P n - u(j)(n) ' - u(j)(n) ' an
~(w(n)) == w(n) - wI(n),

we find using (3.17) and (3.37),
u(j) (n - 1) ((J"j (n) - (J"j (n - 1)) - u(j) (n + 1) a( (n + 1)E, E) ((J"j (n + 1) - (J"j (n))
I I a(nE, E) I I
(3.38)
where
and
qi(n) = ~ (y -/(nE)E))
a
u(j)(n)~j(n)
nE,E
_~(a((n+1)E'E))u(j)(n+1)~j(n+1)+~(j)(nE) u(j)(n).
a(nE, E) a(nE, E)
Selecting a solution of II of (3.37) such that (J"{(n2) = 0 = (J"I(n2 -1) yields
where
j. i-I a(iE, E)u(j)(i - l)u(j)(i)
C I (n, z) = - ~ a((k + l)E, E)U(j) (k)u(j) (k + 1) . (3.39)
The above formula for (J" can be recast as
j( ) _ n2:
2
-
I
C- j ( .) qI(i)
+ '"
n2
(3.40)
(J"j
.n - I n,z u ()C
J z-l
)
~ KI(n,ikI(i),
i=n+1 i=n+1
where
Kj(n,i) = Cj(n,i)
1 1
(~(y - b(iE,E))
a(iE,E)
+ Sij)(iE))
a(iE,E)
u(j)(i)
u(J)(i-1)
-j. (a((i)E,E)) u(j)(i)
+G1(n,z-1)~ a((i-1)E,E) u(j)(i-2)
Here C{ (n, i) = G{ (n, i) for i :::; n2 - 1 and zero otherwise. Note that G{ (n, i) =0
for i :::; n.
Applying a similar argument to that above on the equation
u(j)(n + 1)(o1(n + 1) - o1(n)) - (t(nE,;) ) u(j)(n -1) (0"4(n) - 0"4(n - 1))

a n + 1 E,E
(3.41)
= h~(n) + ~(n),
where
and
~ (n) = -~ ( y - b(nE, E) ) u(j) (n){j;j (n)
2 a1((n+1)E,E)
+~ ( a(nE,E) ) u(j)(n -1){j;j(n -1) _ S(j)(nE)u(j)(n)
a((n + 1)E, E) a((n + 1)E, E) ,
and selecting a solution h of (3.37) such that o1(n1) = 0 = 0"4(n1 + 1) gives
n-1 ~ . n-1
0"2j( n ) -- 'W" Gj( .) ( .)'l2(Z)
2 n, z (. ) + 'w
" Kj( .) j(.)
2 n, Z 0"2 Z (3.42)
U J Z+1
i=nl +1 i=nl
Here C~(n, i) = G~(n, i) for i ~ n1 + 1 and zero otherwise,

.
GJ(n z)
. nI:-1 a((i + 1)E,E)U(j)(i + 1)u(j) (i)
= (3.43)
2' k=i a((k+1)E,E)U(j)(k)u(j)(k+1)'
and
j . _ -j . ( (Y-b(iE,E)) Si j )(iE) ) u(j)(i)
K2(n,z)--G 2(n,z) ~ a((i+1)E,E) + a((i+1)E,E) u(j)(i+1)
+Cj(n,i+1)~(a((i+1)E'E)) u(j)(i) .
2 a(i + 2)E,E) u(J)(i + 2)
The lemma below gives a bound on the above Green's functions.
Lemma 3.6. Set z(i) = C 2 / 3 p(iE,y,E) and suppose that the hypotheses of Lem-
ma 3.5 hold. Then for fixed y such that [tn, til C [tin + E, tfi - EJ, E E (0, EO] the
following inequalities hold
IG;(n, i)1 :::; G(i) = C(Y)E- 1/ 3M(i)M(i - 1), j = 1,2 (3.44)

If ia-iiia hold then the above constant can be chosen uniform in y.
Proof. Set
C7(y)= sup Ig(t+E,y,E)I,

(t,E)EW
cs(y) = sup E-l(X(t))E(x(t + E)), x = E- 2 / 3 p(t, y, E).

(t,E)EW
Since g is strictly bounded (from ii and iii) we find that C7 is finite. The continuity
and asymptotic properties of E show that Cs is also finite. If conditions ia-iiia hold
then the above constants can be made uniform in y. The hypothesis that k 2 is a
3
monotonically increasing function of t implies the e - iE p 2" (t) will be exponentially
decreasing fort> to and of magnitude one for t < to. Thus it is appropriate to
set j = 1 when considering the solution to (3.38) and j = 2 when considering the
solutions (3.41).
We now take up the case i = 1 and we will temporarily suppress the depen-
dence on all variables except k. Observe that the denominators in each term of Gt
can be recast as
1
(3.45)
a(k + l)u(1)(k)u(1)(k + 1)
1 1 (u(2)(k + 1) U(2)(k))
= a(k + 1) u(2)(k + l)u(1)(k) - u(2)(k)u(1)(k + 1) u(1)(k + 1) - u(1)(k) .
Since
1 1
(3.46)
a(k + 1) u(2)(k + l)u(l)(k) - u(2)(k)u(1)(k + 1)
1 1
a(k + 2) u(2)(k + 2)u(1)(k + 1) - u(2)(k + l)u(1)(k + 2)
=-u(2)(k + 1) a(k + 2)u(1)(k + 2) + a(k + l)U(l)(k))
Dk
(l)(k ) a(k + 2)u(2) (k + 2) + a(k + 1)u(2) (k)))
+u +1 Dk '
where Dk is the common denominator of the left-hand side of the above equation,
equation (3.34) can be use to recast (3.46) as
f3i 2)(k + l)u(l)(k + 1) - f3i 1 )(k + 1)u(2)(k + 1)

Dk
With the above equations, summation by parts yields

i-I 1
~ a(k + l)u(1)(k)u(1)(k + 1) (3.47)
From equation (3.33) we find using the monotonicity of E that
for m ~ i. Also the definitions of (3 and (31, equations (3.32) and (3.30) yield,
and
Finally from Lemma 3.5 we find that if 0 is chosen sufficiently small so that
equation (3.36) holds for all (t, ) E W, then
Ia(k + 1)(U(2) (k + l)u(1) 1(k) - < 1

U(2) (k )U(I) (k + 1)) I -
-1/3
C6(Y) sup a(t, r
Since there are at most C 1 terms in the sum on the right-hand side of (3.47) we
find using the above bounds that it is bounded by a constant times 1/3 E2(i). The
above inequalities also show that the boundary term in (3.47) is bounded by a
constant times C 1 / 3 E2(i). From the relations between u(1) and the Airy functions
we find the inequality
which used with C8 yields
1 . <~ a(i)u(1)(i - l)u(1)(i) -1/3..

IG1 (n,z)l_ ~la(k+l)u(1)(k)u(1)(k+l)1 ~c(Y) M(z)M(z-I),
which gives the result for i = 1. The result for i = 2 follows in an analogous
manner with the roles of U(I) and U(2) interchanged. 0
With the above lemma we can now show
Lemma 3.7. Suppose that the hypotheses of Lemma 3.5 hold and [tNl -1, tN2+1] C
[tin,tJi], E E (O,EO]. Then,
N2
L G(i) < C(Y)E-1.
j=N 1
If ia-iiia hold then the above constant can be chosen uniform in y.
Proof. From Lemma 3.6 we find

N2 N2
L G(i) :S C(Y)E- 1/ 3 L M(i - l)M(i).
The sum on the right-hand side of the above equation can be bounded by
N2
L M(i - l)M(i) :S c~ + L M(i -l)M(i) + L M(i - l)M(i).
(i:p(i-10) (i:p(i)<O)
If (3.31) is now used we find
L M(i - .x2
l)M(i) :S _1_ E-2/3
ltN2 t
dt
t 1/2:S C(Y)E- 2/ 3 ,
(i:p(i-10) Cg(y) to ( - 0)
where
() . fl p(t,y,E) 11
Cg y =m
t,E
(t - to (y,)E) 2.
Condition ii assures that cg(y) > 0. Applying the same reasoning to the remaining
sum yields the result. 0
With the above we will now show that there is an actual solution of the
difference equation (3.37) that is close to the approximate solution. First set
u(j)(t+1,y,E) u(j)(t,y, E) )
ClO(y) = max ( sup
(t,E)EW
I (')(
U J t, y, E
) I, sup I (')(
(t,E)EW U J t + 1, y, E
)I j = 1,2
Theorem 3.8. Suppose that for each fixed y E (Y1, Y2), k 2 is monotonically in-
creasing in t with a single zero to E [tin, t Ji] and satisfies i, ii, and iii,. Also sup-
pose a(t, E) is strictly positive and satisfies (2.14), and that a1(t,E) and b1(t, E) E
CO([tin,tJi]) x CO ([0, EO]), are uniformly bounded on [tin,tJi] X [0, EO], and a1(t,E)
is strictly positive. Let [tNl-1,tN2+1] c [tin,tJi], E E (0, EO], and set
. . ( ( b(iE)) )
K(z) = ClO(y)G(Z)
a(iE))
I~ a((i _ l)E) I + I~ (y -a(iE) I, j = 1,2.
Then there exists solutions hand h of (3.37) such that
I0"11( nE )1 = Ih(m) -7jJ1(nE) I

u(l)(nE) (3.48)
N2
:::; c(y) 2)IK(i) + G(i)I~(1)(i)l)eC(Y)L~~(K(il+G(i)Ii3;')(i))I),
N,
and
I 2( )1 = Ih(nE) _7jJ2(m) I (3.49)

0"2 m u(2)(nE)
N2
:::; c(y) 2)K(i) + G(i)I~(2)(i)l)ec(Y)L~~(K(i)+G(i)IW)(i))I).
N,
Thus, if
L K(i) = o(I),j = 1,2, (3.50)
then
j = 1,2, (3.51 )
uniformly in W. If
sup la(t,E) - a1(t,E)1 = 0(E2) = sup Ib(t, E) - b1(t,E)I, (3.52)
(t.)EW t,
then the 0(1) in equation (3.51) is in fact O(E). If it is assumed that ia.-iiia. hold
then the above convergence is uniform on W x [Y1, Y2].
Proof. The interval [tN" tN2l C [tin' tfi], E E (0, Eol so that Theorem 3.3 and
Lemma 3.4 can be used. Take the magnitude of both sides of equation (3.40) with
j = 1. Replace IGUn,i)1 by G(i) and note that l?,bjl is uniformly bounded as are
la(nE, E) I, la(m, E) 1-1, la1 (m, E)I, and la1 (nE, E) 1- 1. Inequality (3.48) now follows
from the Picard iteration. The inequality (3.49) follows from equation (3.42) (with
j = 2) in a similar manner. Suppose now that (3.50) holds. From Theorem 3.3 and
Lemma 3.4 we see that 1~(1)(i)1 = C(Y)E2 = l~i1)(i)1 for all i E [N1 + I,N2 -1].
Thus
where Lemma 3.7 has been used. Since the same inequality holds with ~(1) replaced
by ~i1) we see that there is a solution h satisfying (3.51). If (3.52) holds then
E K1(i) :::; CE 2 E G(i) = C(Y)E so that (3.51) is true for j = 1 with 0(1) replaced
by O( E). That the convergence is uniform on W or uniform on W x [Y1, Y2] follows
from the fact that the constants in the error terms may be chosen uniform on W
or W x [Y1, Y2] respectively. Similar reasoning can be used for h. D
4. Extension to complex values

The above considerations can be extended to complex values of the parameter
y. We first extend Theorem 3.3. We remind the reader that (p~ p')(t, y, E)
-k(t, y, E), so 9 can be written as,
(t )_ ( p(t,y,E) )1/4
9 ,y,E - a2(t+~,E)sinh2(pl/2p')(t,y,E) ,
where the above differentiation is with respect to t. Let 0 be a region in the
complex plane and H(O) the set of functions holomorphic in O. We will assume
lb.
a and lfi'P
9 and p E C 00 ([tin, tJi] X [0, EO]) X H(O), and lfi'g a E C 0 ([tin, tJi] X
i i
[0, EO] X 0), i = 0, ... ,3,

iib. there is a point to(Y, E) such that 4
E C 2([tin' tJi]) X CO ([0, EO] X 0) and
1 t-to~y,e) 1 is uniformly bounded away from zero for (t, y, E) E [tin, tJd X 0 X
[0, EO],
iiib. for fixed (y, E) E 0 X [0, EO], Rep(t, y, E)3/2 is a nonincreasing function of t.
Remark. The choice of iiib. is dictated by the examples discussed below.
Theorem 4.1. Let 0 be a region in the complex plane. Suppose ib holds and
a E Coo ([tin, t Ji] X [0, EO]) is strictly positive for (t, E) E [tin, t Ji] X [0, EO]. Let f
be any entire function solution of the Airy differential equation and 'IjJ(t, y, E) =
g(p(t, y, E))f(c 2/ 3p(t, y, E)). Then'IjJ E COO ([tin , tJi] X (0, EO]) X H(O) and for all
(t, E) E W satisfies
a(t + E, E)'IjJ(t + E, y, E) + a(t, E)'IjJ(t - E, y, E) (4.1)
- 2a(t + E/2, E) cOsh(p-l/2 p' (t, y, E))'IjJ(t, y, E) = (3U) (t, y, E)
where (3(f) E C(W) X H(O), and (3U) E COO(W X 0). Furthermore,
I(3U)(t, y, E)I :::; d(Y)E2 sup [If(E- 2/ 3p(U,y, E))I + E1/ 31f'(E- 2/ 3p(u,y, E))I].
uE(t-e,t+e)
In the above formulas the constant d(y) may be chosen uniform on compact subsets
ofO.
Proof. Let K be a compact subset of 0 and set di(K) = sUPK Ci, i = 1,2, and
d = sUPK c. The prooffollows as in Theorem 3.3 replacing Ci and c by di and d. 0
In the extension of the Airy differential equation to the complex plane it is

customary to introduce the solutions Aio = Ai, and
AiI(Z) = Ai(ze=F2i7r/3), (4.2)
and the regions So = {z : 1 arg z 1 :::; J} and
S1 = e27r/3So.
Since the above function and Bi all satisfy the same differential equation there is
a relation among them given by ([0, p. 414)),
Ai = ~e'f-(Ai(Z) =fiBi(z)). (4.3)
Thus it follows from the asymptotic expansions of Ai and Bi [0, p. 413] that Aij
is recessive in 8 j and dominant in 81+1, and 8 j - 1 where the suffix j is enumerated
mod 3. Furthermore since the zeros of Ai are all real and negative [0, p. 418], Ah
is nonzero in 8 0 U 8 1 . Two other solutions of the Airy equation that will be useful
are iiii) which for complex values of z are defined by
(2 . !.!L)
_(j)( ) _
W Y - 2"1 (Y)!
"3 2 e (-l)j+l i -"'H(j)
6 1/3"3 Y 3 e 2 , j = 1,2 (4.4)
where we take the branch of the square root so that Re(z3/2) ~ 0 for z E 8 0 and
Re(z3/2) :::; 0 for z E 8 1 which is the principal branch of z3/2. It follows from the
asymptotic expansions for Hankel functions ([0, p. 238]) that uP) is recessive in
8 0 while W(2) dominant in 8 0 . The equations ([0, p. 239))
H~(zem7ri) = -[sin{(m - 1)//11" }H~(z) + e- v7ri sin{m//1I" }H~(z)l/ sin //11", (4.5)
and
H~(zem7ri) = _[e v7ri sin{ m//1I" }H~(z) + sin{(m + 1)//11" }H~(z)]/ sin//1I", (4.6)
with // = 1/3 and m = 1 show that w(1) is dominant in 8 1 as is W(2). Since K 1/ 3(Z)
is nonzero for I arg(z)I :::; 11" (Watson [Wj, Olver [0, p. 254]) and ;ie- Hi~~(ze-'f) =
K 1/ 3(Z), we find that w1(z) is nonzero for I arg(z) I :::; l Lommel's method [0, p.
414] applied to W(l) and h = H~)!e-ii Hi~~(~y~e--'f) can now be used to show
that W(l) has no zeros in 81 U 80 U 8-1. Another application of Lommel's method
with W(2) in place of w(!) above and a corresponding change to h shows that W(2)
is also nonzero in 8 1 U 8 0 U 8_ 1.
Set 8 = 8 0 U 8 1, and let n a region in the complex y plane. We will restrict
n so that
(4.7)
Extensions to 8 0 U 8_ 1 can be accomplished using the symmetry properties of the
solutions. Because w(2) is dominant in 8 1 U 8 0 it is not a numerically satisfactory
auxillary function in this region thus we set
u(1) = W(l),
u(2) = Ail,
u(1)(z) = e2/3Z3/2u(1)(z), U(2)(Z) = e-2/3z3/2u(2)(z),Aio(z) = e2/3z3/2Aio(z),
and
It is not difficult to see from the continuity and asymptotic expansions for the
above functions ([0, p. 392, 413]) that the following constants are finite,
d3 = max (sup lu(1) (z)l, sup lAir (z)l, sup IAio(Z)I) ,

s s S
VI = max (s~p Iz 1/ 2u(1)(z)l, s~p IZ 1/ 2U(2)(Z)I) ,

_ ( le2/3z3/2 u(I)(z)'1 le-2/3z3/2 u(2)(Z)'I)
d4 - max sup
S l+z I 11/4 ,sup
s l+z I 11/4 '
and
Set
'ljJl = gAio, (4.8)
and
(4.9)
With k(t,y,E) = p(t,y,E)I/2 p'(t,y,E), we now obtain an analog of Lemma 3.6.
Because of iib. the appropriate functions to use are Gi and G in 3.39 and 3.43
respectively, also the role of the weight function will be played by
E(t) = le 2 / 3P (t,y,E)3/2 1. (4.10)
Lemma 4.2. Suppose that (4.7) holds, ib-iiib are satisfied, and
a(t,E) E COO([tin,tJi] X [O,EO])
is strictly positive. Let K be a compact subset of n. Then there is a constant d(K)
such that for [m, iE] C [tin + E, t Ji - E], EE (0, EO] the following inequality holds
IG~(n, i)1 :::; 6 1 (i) = d(K)E- 1 / 3 Iu(I)(i)u(I)(i - 1)1, (4.11)
and
Pmoj. The Wronskian of Al and u(l) is

21Ti
W(AI u(1)) = _ e- 3
, 2V311"
which is less in magnitude than 2/11'.
Thus let di(K) = sUPK Ci(y) i = 6,7 where C6 is given in (3.36) and C7 in
Lemma 3.6. Let ds(K) = sup(t,y,E)EWxK E(x(t, Y, E))E-l(x(t+E, y, E)), x(t, y, E) =
C 2 / 3 p(t, y, E). Conditions ib and iib insure that these constants are finite. We begin
with Gi and observe that for y fixed the steps in Lemma 3.6 can be followed to
equation (3.47) with the roles of u(1) and u(2) interchanged. From iiib we find the
bound
u(1)(m) - 1 2
1u(2)(m) I:::; d5 (K)E- (i) ,
for m :::; i. The rest of the steps to (4.11) now follow as in Lemma 3.6. A similar
analysis gives (4.12). D
With the above lemma we can now show

Lemma 4.3. Suppose the hypotheses of Lemma 4.2. Let K be a compact subset ofn.
Then there is a constant d(K) such that such that for [tNl-1, tNd1] c [tin, t ji], E E
(0, EO],
N2
L cj (i) < d(K)E-l, j = 1,2.
j=N1
Proof. Set d g = inf [tin,ifd X S1 X [O,EO] 1t!to 11/2 which by iiib is nonzero. From Lem-
rna 4.2
N2 N2
L C1 (i) :::; d(K)E- 1 / 3 L lil(l)(i - l)il(l)(i)l,
which can be bounded by

N2
L lil(l)(i - l)il(l)(i)1 :::; d~ +
(dR(p(i)3/2)<0)
+
(i:~(p( i-1)3/2) >0)
If 1/1 is now used we find
L lil(l)(i - l)il(l)(i)1 :::; d k-

1/ 2
E- 2/ 3
jtN2 dt
t t 1/2 :::; d(K)E- 2 / 3 .
(i:~(p(i-1)3/20) g() tNl 1 - 01
A similar argument bounds the remaining sum which gives the result for j = l.
The result for j = 2 follows in a similar manner. D
With the above lemmas we can now prove the main result of this section
Theorem 4.4. Suppose that (4.7) holds, ib-iiib are satisfied, and
a(t,E) E COO([tin,tji] X [0, EO])
is strictly positive. Furthermore suppose that a1 (t, E) and b1(t, E) are continuous
on [tin,tji] X [O,EO] and a1(t,E) is strictly positive. Let [(N1 + 1)E,(N2 -l)E] C
[tin' t ji], E E (0, EO], and set
j = 1,2.
Then there exist solutions hand h of (3.37) such that
1 2l( )1 = Ih(nE) -1j;l(nE)1 (4.13)

a nE u(1) (nE)
s:: d( K) 2) IKI (i) + C l (i) 1,8(1) (i)l)ed(K) L ~i (K' (i)+C ' (i) l)Ji 'l (i)) I) ,
n2
and
(4.14)
Here 1j;j, j = 1,2 are defined by (4.8) and (4.9) respectively. Thus if
j = 1,2, (4.15)
then
(4.16)
If
sup la(t, E) - al(t,E)1 = 0(E2) = suplb(t,E) - bl(t,E)I, (4.17)
t,E t,E
with the sup be taken over [tin, tJi] X [0, EO] then the 0(1) in equation (4.16) is in
fact O(E).
Proof. The proof follows like the one given in Theorem 3.8 with dlO = SUPyEK ClO.
o
In the next section we consider the initial value problem.
5. Initial value problem singular case

If tin can be taken to be equal to zero then the above results can be used to
solve the initial value problem and thus give uniform asymptotics for orthogonal
polynomials. Unfortunately in the case when a(n) and Ib(n)1 tend to infinity, in
general k(t, y, E)2 will not satisfy condition i and aCt, E) will not satisfy (2.14) for t
in a neighborhood of zero. This is because if aCt, E) and bet, E) are to be bounded
then AE must increase to infinity (see Section 6 for an example). Another case
where condition i or (2.14) can be violated is in the case of varying recurrence
coefficients [DM], [KV]. Finally it may not by evident where p is a single-valued
analytic function so a LG-WKB approach is useful. We begin by switching to the
equation satisfied by the polynomials Pn = 2npn/kn,
Pn+l(X) + 2(bn - X)Pn(X) + 4a~Pn-l(X) = O.
If the scaling indicated in the introduction is performed we arrive at the difference

equation
Pn+l(Y) + 2(b(m, E) - Y)Pn(Y) + 4a 2(nE, E)Pn-l(Y) = O. (5.1)
We now consider the E-difference equation
'ljJ(t + E, y, E) + 2(b(t, E) - y)'ljJ(t, y, E) + 4a 2(t, E)'ljJ(t - E, y, E) = O. (5.2)
Following Deift and McLaughlin [DM, Appendix I] we look for solutions of (5.2)
of the form
(5.3)
Substitute the expressions for f in (5.2) then expand f(t E, E) (here again we
suppress the dependence upon y) in powers of E. The coefficient of EO is
eSO(t,E)' + 4a(t, E)2 e -so (t,E)' = 2(y - b(t, E)), (5.4)
while the coefficient of El is
eSO(t,E)' + 4a2(t E)e-SO(t,E)'
Sl(t,E)' = - 2(e SO (t,E)' _ 4a2(~,E)e-So(t'E)') SO(t, E)". (5.5)
With the use of
and the derivative of (5.4) we find

, (eSO(t,E)' _ 4a(t, E)2 e-so(t,E)')'
Sl(t, E) = - 2(eSO (t,E)' _ 4a 2 (t, E)e-SO(t,E)')
b(t E)'
+ ' + S0 (t , E)"
eSO(t,E)' - 4a 2(t, E)e-sO(t,E)' 2
Solving the relevant equations yields
S~(t,E) = jt In (y - b(u, E) J(y - b(U,E))2 - 4a(u,E)2) du, (5.6)
and
1
St(t,E) = - 4In22((y - b(t,E))2 - 4a(t,E)2) (5.7)
1
+ "2ln(y - b(t, E) J(y - b(t, E))2 - 4a(t, E)2)
~ jt b'(U,E)du
2 J(y-b(U,E))2-4a(u,E)2'
This gives us two approximate solutions to the difference equation given by
f (tn" Y E) -- e~st(tn,y,E)+St(tn,y,E) ,
with s~ and st given above and tn = nE.

Set
j
vti,tj(g) = lt Ig(t,E)ldt, (5.8)
VI(t, E) = vt,t+E(b"()(t + E- .)) + vt,HE(a 2C)"(t + E- .)) (5.9)

+ vt,HE(b'C)2(t + E - .))
+ vt,HE((4a'a())2(t + E - .))),
V2(t, E) = (vt-E,t(b"C)( - (t - E)) + vt_f,t(la 2(')"IC - (t - E))) (5.10)

+ vt_E,t(b'C)2(. - (t - E))) + vt_E,t((4a'aC))2(. - (t - E)))),
and
V3 (t, E) = vt,t+E(aa')2 + vt,t+f(b')2. (5.11)
We will impose the following assumptions on a(t, E) and b(t, E)
ic. a(t, E), b(t, E) E GO([o, tfi] x [0, EO]),
iic. tti,a(t, E), ttiib(t,E) E GO((O,tfi] X [O,EO]) i = 1,2,
iiic.
vic.
vc.

sup J~Ii Ib'(t, E)I < 00, sup J~Ii la'a(t, E)I < 00
a(O,O) = but a(t,E) > 0, (t,E) E (O,tf] X [0, EO],
VI (t, E) E GO ([0, t fi - E] X [0, Eo]),v2 (t, E) E GO([E, t fi] X [0, EO]), andV3 (t, E) E
GO ([0, t fi] x [0, EO])
Let
A(E) = inf (b(t, E) - 2a(t, E)), B(E) = sup (b(t, E) + 2a(t, E)),
[O,t fd [O,t fd
I(E) = [A(E), B(E)],

Aa,b(E) = inf (b(t,E) - 2a(t,E)), Ba,b(E) = sup (b(t,E) + 2a(t,E)),
[tin ,t Iil [tin ,t til
and
Ia,b(E) = [Aa,b(E), Ba,b(E)].
We now examine the analytic properties of the above approximate solution.
Set
h(t, y, E) = J(y - b(t, E))2 - 4a(t, E)2
and,
ft(t, y, E) = y - b(t, E) J(y - b(t, E))2 - 4a(t, E)2.
Lemma 5.1. Suppose conditions ie. holds. Then h, and fi are nonvanishing in
[0, tfi] x G\I(E), and h, fi, and lnfi E GO ([0, tfi]) x H(G\I(E)). Likewise f; is
nonvanishing in [tin, tfi] X G and lnf; E GO([tin, tfi]) x H(G\Ia,b(E)) for tin> 0.
For any compact set K such that 1(0) C int(K) there is an EK such that hand
fi are non-zero in [0, t fi] X [0, EK] x G\K, and h, fi, and In fi E GO ([0, t fi] x
[O,EK]) x H(G\K). If K is a compact set in G \ K then h, fi and lnfi E

GO ([0, tfi] x [0, EK] X K). Finally if tin > and K a compact set such that Ia,b(O) C
int(K), then there is an EK such that ft and lnft E G 2([tin' tfi] X [0, EK]) x H(G\
K), f:; is nonzero in [tin,tji] X [O,EK] x C\K, and fi:, lnfi: E C 2([tin,tji] X
[0, EK] X K) where K is a compact subset of C\K.
Proof. From the definition of 1(E) we see that (y - b(t,E))2 - 4a(t,E)2 is nonzero
for all (t,y) E [O,tji] x C\I(E) hence a branch of the square root function can
be choosen so that h(t,y,E) E H(C \ 1(E)) for each t E [O,tji] and such that
for large y, ft '" 2y. That h(t,y,E) E CO([O,tji]) x H(C\I(E)) follows from
the continuity properties of a, b, and the square root function. The above argu-
ment also shows that ft E CO([O, tji]) x H(C\I(E)). Since x + Jx 2 - 1 maps
C\[-1,1] to the exterior of the unit circle we see that ft is nonvanishing in
[0, tji] x C\I(E). This implies that lnft E CO([O, tji]) x H(C\I(E)). For t > tin iic
and the differentiability of In and square root functions show that %:i h, %tii fi: and
%tii In It
E CO([tin, tji]) x H(C \ 1(E)) i = 1. .. 2. Since a(t, E) is strictly greater
than zero in [tin' tji] we find
j _ _ 4a(E, t)2
2 - It '
which does not vanish for each
The properties of In I:; now follow. Let K be a compact set such that 1(0) C
int(K). The continuity of a(t, E) and b(t, E) imply that there exists an EK ::; EO such
that 1(E) C int(K) for all E::; EK. This plus the continuity properties of the In and
square root functions give that h, It, and In It E CO ([0, tJi] x [0, EK]) x H(C\K)
and h, It and In It E CO ([0, tji] x [0, EK] X K) for any compact set K C C\K.
For t E [tin' tjd the properties of ft and lnft follow in a similar manner. 0
With this we now discuss the approximate solutions and their relation to the
initial value problem. Define
H~,Sl (t) = E(So(t)" /2 s~ (t)),
J~,Sl (t) =
1
2E
I t
t
s~'(u)(t E - u)2du
+ I t
t
s~(u)(t E - u)du,
and
With the above we have
Lemma 5.2. Snppose ic-ivc hold and set
f +( ) - (y - bet, E) + J(y - bet, E))2 - 4a(t, E)2)1/2

(5.12)
t, y, E - ((y _ bet, E))2 _ 4a(t, E)2)l/4
x exp (i t b' (n, E )dn

2J(y - ben, E))2 - 4a(n, E)2
)
x exp(l/E lot In(y - ben, E) + J(y - ben, E))2 - 4a(n, E)2)dn),
and
f -( ) - (y - bet, E) - J(y - bet, E))2 - 4a(t, E)2)1/2
(5.13)
t, y, E- ((y _ bet, E))2 _ 4a(t, E)2)l/4
t!i b'(n E)dn
x exp( - i ' )
t 2J(y - ben, E))2 - 4a(n, E)2
x exp(l/E it!i In(y - ben, E) - J(y - ben, E))2 - 4a(n, E)2)dn).

Then for mE [0, tfi], < E :::; EO, f+ E CO ([0, tfi]) x H(C\I(E)), is nonzero and
satisfies the difference eqnation
f+((n + l)E) + 2(b(m, E) - y)f+(m) (5.14)
2
+ 4a (nE, E)f+((n - l)E = 7]+(m)f+(m), E:::; m < tfi - E
where
(5.15)

Likewise for mE [tin, tfi], tin> and < E:::; EO, f- E C[tin, tfi] x H(C\Ia,b(E))
is nonzero and satisfies the above difference eqnation for tin + E < nE < t fi - E with
7]+ replaced by
7]-(t)=es~(t)/(J~ ~(t)+R+~ ~(t))+e-s~(t)/(J--=- ~(t)+R-~ ~(t)). (5.16)
So ,8 1 So ,8 1 So ,8 1 So ,8 1
Let K be compact snbset of C snch that 1(0) C int(K) and K be a compact snbset
of C\K. Then there is an EK snch that f+ E CO ([0, tfi] x (0, EK]) x H(C\K),
f+ E CO([O, tfi] x (0, EK] x K), 7]+ E CO([E, tfi - E] X [0, EK] x K), and 7]+ E
CO([E, tfi - E] X [0, EK]) x H(C \ K). If tin> 0, K and K are compact sets snch
that Ia,b(O) C int(K), and K C C\K, then there exists an EK snch that f- E
CO([tin,tfi] X (O,EK]) xH(C\K), andf- E CO([tin,tfi]X(O,EK]xK) where tin > 0.
Also 7]- E CO ([tin , tfi] X [0, EK]) x H(C\K) and 7]- E CO ([tin , tfi] X [0, EK] X K).
Proof. It follows from Morera's Theorem, Lemma 5.1 and the integrability of b'
that the integrals in (5.12) yield functions analytic in C\I(E) for each t E [0, tfJ
The fact that j+ E COlO, tfi] x H(C\I(E)) for E > now follows from Lemma 5.1
and the integrability of b'. The nonvanishing of II and j~ in [0, t fi] x C\I (E) shows
that j+ is nonvanishing in this set also. Let K be a compact set in C such that
1(0) C int(K) then Lemma 5.1 and condition ic imply that there is an EK so that
the integral
h(t, y, E) = fat in (y - b(u, E) + J(y - b(u, t))2 - 4a 2(u, t)) du,
is CO ([0, tfi] x [0, EK]) x H(C\K). Also for any compact set K E C\K, h E
CO ([0, t fi] x [0, EK] x K). From the integral representation,
1
J(y - b)2 - 4a 2 =;:
11 b+ 2a dx
b-2a J(x - b)2 - 4a 2 '
(5.17)
we find that 7;
is uniformly bounded for (t, E, y) E [0, tfi] X [0, EK] x C\K. The
continuity conditions ic and iic, and the uniform integrability of b' imply that
tfi b'(u E)
h(t,y,E) = Jo J(y - b(U'E))'2 - 4a(u,E)2'
is in CO([O, tfi] x [0, EK]) x H(C\K) and h E CO ([0, tfi] x [0, EK] X K) for K as
above. This plus Lemma 5.1 and the fact that 1E is continuous for E > gives the
continuity properties of j+. A similar argument gives the continuity properties

of j-.
The equations for ry are obtained by substituting (5.12) or (5.13) into (5.14)
expanding to first order using Taylor's Theorem with the remainder and equations
(5.4) and (5.5). From the definition of Hand J we find
(5.18)
s(t)" = b'(t, E) 4a'(t, E)a(t, E) (5.19)

=f II =f j~(t)1I .
Lemma 5.1, iic, vc, and (5.18) show that R\
So ,SI
+ E CO ([0, tfi-E] x [0, EO]) xH(C\K),
and R\ So 'SI
+ E CO([O, tfi - E] X [0, EO] x K). In a similar manner it follows that
R-+ + E CO([E, tfi] X [0, EO]) x H(C\K), and R-+ + E CO([E, tf;] x [0, EO] x K).
So ,SI So 'SI
By differentiating st(t)" with respect to t and likewise si(t)' it follows from vc.
after a tedious computation that J~ + E CO([E,tfi - E] x [0, EO]) x H(C\K) and
So ,8 1
J~ + E CO([E, tfi - E] X [0, EO] x K) while J~ + E CO([E, tfi] X [0, EO]) x H(C\K)
So ,8 1 So ,SI
and J~+ E CO([E, t fi] X[0, EO] x K). Thus TJ+ has the continuity properties claimed
So ,8 1
in the Lemma. A similar analysis follows for ry- . 0

The following Lemma will be useful in controlling the error between the
approximate solutions and bona fide solutions of the initial value problem.
Lemma 5.3. Suppose ic-vc hold and let K be a compact subset of 0\1(0). Then
there exists an Ek such that for all E E [0, Ek],
n-l n-l
L 1R++ So ,8
1
+ (')1
ZE ~ L vti,tHl (So+") + vti,tHl (SI+'))2 e d(k) (5.20)
i=1 i=1
n
~ d(K) L V (ti, E),3
i=1
n-l
~ d(K) L V3(ti, E)
i=O
n-l n-l
L IJ+(st,St)(ti)1 ~ L(~vti,tHl(St"'(ti+1 - .)) + vt i ,ti+l(St"(ti+l - .))
i=1 i=1
(5.22)
n-l
~ d(K) L VI (ti' E),
i=1
and
n-l n-2
L IJ- (st, st)(ti) 1 ~ L ~ vt i ,tHl (stili (. - ti)) + vti,tH' (st" (. - ti)) (5.23)
i=1 i=O
n-2
~ d(K) L V (ti, E).
2
i=1
Here d(K) is a constant depending only upon K and ti = iE.

Proof. From Lemma 5.1 there exists a constant M such that
1 1
1!I I, 1fi 1< M
for all (t, E, y) E [0, tfil x [0, Ekl x K. Thus from Lemma 5.1, equation (5.19) and
equation (5.5) we find, Ist(t)"1 ~ Ib'1W'E) I + 4Ia/(t~:(t'E)1 and Ist(t)'1 <
J(K)(Ib'~E)1 + 4Ia/(tA1:(t,E)I) , where d(K) depends only upon K. Thus
IH~,st(t) + J~,st(t)1 ~ d(K) I

t
t+E
Ib'(u,E)1 + la'(u,E)a(u,E)ldu,
and
IH:C;,st(t) + J:C;,st(t)l:::; d(K) l~E Ib'(U,E)1 + la'(u,E)a(u,E)ldu,
Therefore,
IR So++
,8 1
+(ti)l:::; (1-vt
E
+" (ti+l - .))
i ,ti+l(SO (5.24)
ti E [10, tf].
(5.25)
Summing (5.24) from 1 to n - 1 then using standard inequalities gives the first
part of (5.20). Substituting the above inequalities for Ist(t)"1 and Isi(t)'1 in (5.24)
and using standard inequalities gives the second part of (5.20). Equation (5.21) is
obtained in the same way. Lemma 5.1 and elementary calculus gives the bound
Ist(t)1II1 :::; d(K)(lb(t)"1 + 4(la(t)a"(t)1 + la'(t)21) + (lb'(t)1 + 4Ia(t)a(t)'1)2),
and likewise for Isi(t)l. These bounds give the second inequalities in (5.22) and
(5.23). D
Lemma 5.4. Suppose that %tiia(t,E), %;ib(t,E) E CO([tin,tfi] X [O,Eo]),i = 0 ... 2

[~J
then L:i=[t~n J 117 (ti)1 < CEo
Proof. The boundedness of a, b and the first and second partial time derivatives
imply that
[~J [t~iJ
L 117(ti)1 < L CE 2 < CE.
i=[t~nJ i=[t~nJ
D
We now obtain error bounds on how close f+ stays to a solution of the initial
value problem. Similar results were obtained (without error bounds) by [GS] in
the case when the coefficients in the recurrence formula were in certain classes of
regularly varying or slowly varying functions. For the case when a(t, E) are strictly
bounded away from zero and a(t, E), b(t, E) E Coo similar results were obtained by
Costin and Costin [Ce] and Deift and McLaughlin [DM].
Theorem 5.5. Suppose ic-vc. hold, and suppose that al (t, E) and b1(t, E) are in
C[O,tf;] x C[O, EO]) and al(t,E) > 0, for (t,E) E (O,tf] x [0, EO]. Let K be a compact
set in C\1(O) and Pn(Y, E) be a solution of
Pn+l (y, E) + 2(b 1(nc, E) -
Y)Pn (y, E) + 4al (nc, E)2Pn_l (y, E) = 0, (5.26)
PO(Y, E) =1, Pl(y,E)=2(y-b 1 (0,E)).
Then there exists an EK such that for (n., E) E [0, tjl x (0, EKl
Pn(Y, E) = f+(nE, y, E)/ f+(O, y, E)(l + (nE, y, E)), (5.27)

where
1(nE,y,E)1 < d(K)(~(b(O'E)) + 4a 2(0, E) (5.28)
+ ed(K)(VO,E(b'HVo,E(aa')) - 1 + L(l, n - 1)),

where
Proof. The proof is very close to that given in the earlier sections and we will only
indicate the differences. Substituting (5.27) into (5.26) and following the steps that
led to (3.41) it is not difficult to see that satisfies the equation,
f+(n - 1)
(n + 1) - (n) - 4a(nE)2 f+(n + 1) ((n) - (n - 1)) (5.29)
= (~(b(n.)) - ",+(n)) f!(:(; 1) (1 + (n))

2) f+(n - 1)
+ 4~(a(n.) f+(n + 1) (1 + (n - 1)), lS;n
where we have suppressed the dependence on all variables but n. Solve the above
equation with the initial conditions (O) = 0, (1) = f+(O)Pl(y, E)/ f+(l) - 1 to
obtain,
f+(O)P(l))
(n) = ( f+(l) - 1 G(n,O) + 4~ a(E)
(2) f+(O)
f+(2) G(n, 1) (5.30)
n-l f+C)
+ ~ G(n, i)(~(b(iE)) - ",+(i)) f+(i: 1) (1 + (i))
n-2 f+(')
~ G(n, i + 1)4~(a((i + 1)E)2) f+(i: 2) (1 + (i))
where
, n-l f+(i + l)f+(i)
L IT
j 2
G(n, z) = f+( , l)f+( ') 4a(kE, E) .
j=i J+ J k=Hl
With
f+(i + l)f+(i) = lIT'+! f+(k - 2)
f+(j + l)f+(j) H2 f+(k) ,
so that
f+(i + l)f+(i)
f+(" + l)f+(")
II j
2 _ f+(j - 1 ) .
4a(kE, E) - f+(" + 1) 4a((z+I)E, E)
2
II
j 4a(kE, E)2 g(kE,E)
f+(k)2 e ,
J J k=i+1 J k=i+2
where
1 1(k-2)E 8
g(kE, E) = - ~ log fi(u, E)((k - 2)E - u)du + st((k - 2)E, E) - st(kE, E).
2E kE uU
From Lemma 5.1, the above hypothesis, and the mapping properties of x +
";x 2 -1 mentioned above we find that there is an EK, such that for an (t,E,y) E
[0, tfil x [0, EKl x K, In 1ft /2a(t, E)I > In r with r > 1. Also from the nonvanishing
of ft we find
I f+(k) 1< eUtt:+1 Ii n Ui(u))ldu+d(K)(Vtk ,tk+l (b')+Vtk ,tk+l (aa') < d(K).
f+(k + 1) -
Thus,
1 -2(n-i)
IG(n,i)l::; d(K)ed(K)(VO,tji (b')+VO,tji (aa') ~~r-I
From (5.4) the inequality
IP(/!(:jO) - 11 < d(K)(~(b(O, E)) + 4a 2(0, E) + ed(K)(Vo,(b')+Vo,(aa')) - 1)

can be obtained, and (5.15) and Lemma 5.3 can be used to obtain
~ le-sci(iE)' 1/(iE)1 ::; d(K) (~VI(ti, E) + V2(ti, E) + V (ti-l, E)), 3 n> 1.
Thus the Picard iteration applied to (5.30) gives

1(n)1 ::; d(K)(~(b(O, E)) + 4a2(0, E) + ed(K)(Vo,(b')+Vo,(aa')) - 1 (5.31)
+ S(n - l))e d (K)S(n-I),
where S(n - 1) is given above. This leads to the required bound for Ip(n)l. D
A formula that will be useful for the matching of various solutions is
l tn
lny - b(u) + J(y - b(U))2 - 4a 2(u + ~)du (5.32)
= l tn
lny - b(u) + J(y - b(U))2 - 4a 2(u)du
E tn 4a(u)a'(u) I
- 2" io f(u)J(y - b(U))2 - 4a(u)2 + ,
where I has properties similar to above.
6. Applications
The previous results can be combined to obtain uniform asymptotics for solutions
the difference equation of the form (2.6) including the initial value problem. This
is accomplished by matching solutions in various overlapping regions. Here we will
study the case of Hermite polynomials whose asymptotics can be found in [S]
and more recently [DKMVZ]. More general problems will be considered at a later
time [G].
The orthonormal Hermite polynomials {Hn(x)} (Olver [0, p. 403], Szego
[S]) have recurrence coefficients bn = 0, n ~ 0, an = In/2, n ~ 1 and we choose
f. = l/l, tn = n/l and).l = Ae = v'2l + 1 which yields a (t, t) = ~J 27~1' Sterling's
formula implies
e1fJln2a(u+-h,t)du e 1/ 4 1
rr ~ = 1fl/4 (1 + O( y)),
1 A1/l
so that for y E C\ [- 21~:~il, 21~{~il], tfi ~ 1 equation (5.27) and (5.32) give
1
Pn(Y,y)
= e 1/ 4 ( y2 )l/\lf~nln( v';;::+1+VC';;::+lr-l)dU(I+0(~))
1fl/4 y2 _ 4a 2(t n + 1..
21' 1)
I l '
(
A1Y
v'2lu + 1
A )2 ) u -
1 d
(6.1)
The convergence in the above formula is uniform on compact subsets of C\ [-1, 1].
We now analyze the behavior of the Hermite polynomials in the oscillatory region.
With to = ()..lYdz2 - 1 , the Langer transformation (2.2) gives for y > J2lt fi + 1
_2 p3/2(t, y, l) = lto In (
3 t
~zY
y'2lu + 1
+ (
AZY
A
y'2lu + 1
)2 )
_ 1 du
.
(6.2)
The change of variables v = y'2lu + 1 in the above integral gives
2 l2
3 / 1 r)..LY
(t,y,l) = T Jo In
(~ZY F-~Zy)2 )
--:;- + ~ \ --:;-) -1 vdv
- T1 Jrv'2lm
o In (~y
-;- + ~F-~y)2
\ -;-) -1 ) vdv
The last integral in the above formula gives an analytic extension to p3/2 which
shows that Rep3/2 is a decreasing function of t and limImy---+O,Imy>O p3/2 is a con-
tinuous function which for y real and positive has only one zero at to. Also for
S = {y E C : ~(y) > 0,
;>
: ;
large enough y, p3/2 is nonzero. Thus Theorems 3.8 and 4.4 can be used. Let
y ::; <5} where <5 is chosen sufficiently small so that
p2 ]! is nonzero in S. For fixed yES there exists an interval [ta, tb] C (0, tfi)
(t-to(y,E 2
tfi > 1 and an E(y) such that for all E < E(y), to(y, E) tf- [ta, tb]. From equation
(6.1) we find
Pr:(y,t) = 1 ( 1 )1/4e-1Zp3/2(tn,y't)(1+0(~)). (6.3)

e(ALY)2/2 ~7fl/4 y2-4a 2(tn+.Jz) l
Since the function on the left-hand side of the above equation satisfies the same
difference equation as Pn(y, t)
Theorem 4.4 implies that it can be written as c1'l/Jl +
c 'l/J2 with c and c independent of tn. Equation 4.3, the Wronskian of Ai and Bi,
2 1 2
W[Ai, Bi] = .!.7f

and equation 3.4 gives
The asymptotic expansions of the Airy functions,

1 1 2Z 3/2 ( ') 1
Ai(l3 p(t, y, -)) = tn,y" (1 + O( - )),
2
1 1 1 e- 3 p
l 2Jr2l6p(tn,y,t);;; l
1 1 2Z 3/2 ( ') 1
,e 3P tn,y" (1+0(-)),
2
Bi(l3p(t,y,-))= 1 1
l Jr2l6p(tn,y,t);;; l
coupled with the previous equation yields
1 Jr1/4V2Kli 1
e = '1/2 (1+0(-T))'
AZ
and
Ie21< e(y) Jr1/4J2 e-32Re ( P3/2( tn,y,E )+p3/2( tn+E,y,E )) 0(-)
1
,X,1/12 l .
Z
Thus for ta ~ in,
le2?jJ21 < e(y)e-~Re(p3/2(ta+E,y'E))0( ~),
where the fact that Re(p3/2) is a decreasing function of t has been used. This leads
to
:~A~~;2~~ = Jr1/~V:;; li g (tn' y, ~) Ai (,X,Z/3 P (tn' y, ~) ) (1 +0 (~) ) ,

The above formula is valid for tb < tn < tfi with tfi > 1 so that for tn = 1,
, , J2 / 1/4
e~~~~~2 = (2l ~~~1~12 CY:~)1)2
) Ai((2l + 1)1/3 p(y)) (1 + 0 ) , (~)
2/3
where p= ~4/3 p. For 1 < y we find
Al
l2/3 p (l,y, ~) = (2l + 1)2/3 (~y~ _ ~ COSh- 1(y)) 2/3,

while for 0 < y < 1 we find
l2/3 p (1, y, ~) = -(2l + 1)2/3 (~COS-1(y) _ ~ yJ1=Y2) 2/3

The above formula is similar to the one found in Olver [0, p. 403]. Examination
of the errors show that these formulas convergence uniformly on compact subsets
of S. Extensions to other regions of the complex plane may be accomplished by
using the symmetries Hn(-z) = (-l)nH n (z) and HnCz) = Hn(z).
Acknowledgment
J.S.G. would like to thank K. McLaughlin, and P. Miller for discussion pertaining
to the matching problem S. Garoufalidis for comments regarding the introduc-
tion, A. Martinez Finkelstein for pointing out reference [WW], and F. Olver for
discussion pertaining to Airy and Hankel functions in the complex plane.
References
[B] P.A. Braun, WKB method for three-term recurrence relations and quasienergies
of an anharmonic oscillator, Theoret. and Math. Phys. 37 (1978) 1070-108l.
[CC] O. Costin, R. Costin, Rigorous WKB method for finite-order linear recurrence
relations with smooth coefficients, SIAM J. Math. Anal. 27 (1996) 110-134.
[DM] P. Deift, K. T.-R. McLaughlin, A Continuum Limit of the Toda Lattice, Memoirs
of AMS #624 (1998) Providence, RI.
[DMo] R.B. Dingle, G.J. Morgan, WKB methods for difference equations I, Appl. Sci.
Res. 18 (1967) 221-237.
[DKMVZ] P. Deift, T, Kriecherbauer, K. T.-R. McLaughlin, S. Venakides, X. Zhou,
Strong Asymptotics of Orthogonal Polynomials with respect to exponential
weights, Comm. Pure. Appl. Math.52 (1999) 1491-1552.
[G] J.S. Geronimo, WKB and Turning point Theory for second order difference equa-
tions: External Fields and Strong Asymptotics for Orthogonal Polynomials, In
prep.
[GS] J.S. Geronimo, D. Smith, WKB (Liouville-Green) Analysis of Second Order Dif-
ference Equations and Applications, JAT 69 (1992) 269--30l.
[L] R.E. Langer On the asymptotic solutions of ordinary differential equations, with
an application to the Bessel function of large order, Trans. Amer. Math. Soc. 33
(1931) 23-64.
[MV] M. Maejina, W. Van Assche, Probabilistic proofs of asymptotic formulas for some
classical polynomials, Math. Proc. Cambridge Philos. Soc. 97 (1985) 499-510.
[ND] P. Nevai, J.S. Dehisa, On asymptotic properties of zeros of orthogonal polynomi-
als, SIAM J. Math. Anal. 10 (1979) 1184-1192.
[0] F.W.J. Olver, Asymptotics and Special Functions (1974) Academic Press London.
[SG] K. Schulten, R.G. Gordon, Semiclassical analysis of 31' and 61' coefficients for
quantum mechanical of angular momentum, JMP 16 (1975) 1971-1988.
[S] G. Szego, Orthogonal Polynomials, AMS Colloq. Pub. Vol. 23, 4th ed, Providence,
RI (1978).
[V] W. Van Assche, Asymptotics for Orthogonal Polynomial, Lectures Notes in Math-
ematics 1265 Springer-Verlag (1967) London.
[VG] W. Van Assche, J.S. Geronimo, Asymptotics for orthogonal polynomials with
regularly varying recurrence coefficients, Rocky Mtn. J. Math. 19 (1989) 39-49.
[WW] Z. Wang, R. Wong, Uniform asymptotic expansion of Jv(va) via a difference
equation, Numer. Math. 91 (2002) 147-193.
[W] G.N. Watson, A treatise on the theory of Bessel functions (2nd ed), Cambridge
Univ. Press (1944) London.
Jeffrey S. Geronimo
School of Mathematics
Georgia Institute of Technology
Atlanta, GA 30:n2-0160, USA
e-mail: geronimo@math.gatech. edu
Oscar Bruno
Applied & Computational Mathematics
California Institute of Technology
Pasadena, CA 91125, USA
Walter Van Assche
Department Wiskunde
Celestijnenlaan 200B
B-3030 Leuven, Belgium
Operator Theory:
Bounds for the Points of

Spectral Concentration of
One-dimensional Schrodinger Operators
Daphne J. Gilbert, Bernard J. Harris and Suzanne M. Riehl
Abstract. We investigate the phenomenon of spectral concentration for one-

dimensional Schrodinger operators with decaying potentials on the half-line.
For suitable classes of short range and long range potentials, we outline sys-
tematic procedures which enable numerical estimates of upper bounds for
points of spectral concentration to be obtained. Our approach involves use of
the Riccati equation to construct appropriate convergent series for a gener-
alised Dirichlet m-function, from which the existence and properties of deriva-
tives of the corresponding spectral functions can be established. An incidental
outcome in the case of long range potentials is that upper bounds for embed-
ded singular spectrum can also be obtained.
Mathematics Subject Classification (2000). 34L05, 47E05.

Keywords. Spectral concentration, Sturm-Liouville operators.
1. Introduction
In the context of this paper, we use the term spectral concentration to refer to
a significant localised intensification of the spectrum occuring within an interval
of purely absolutely continuous spectrum. This terminology and usage in con-
nection with the one-dimensional Schrodinger operator goes back to Titchmarsh,
who provided a detailed mathematical analysis in a number of well-known cases
[22]. Titchmarsh's investigations were themselves developments of earlier work by
Schrodinger, Oppenheimer and others on such problems as ionisation of the hy-
drogen atom in a weak electric field (the Stark effect), and the behavior of the hy-
drogen atom in a uniform magnetic field (the Zeeman effect) [16], [19]. A common
feature of these problems is that a discrete spectrum is replaced by a continuous
spectrum after the introduction of a 'perturbing' field while, in the vicinity of
the eigenvalues of the original problem, there are poles of the meromorphic con-
tinuation of the perturbed Green's function located just below the real axis, on
140 D.J. Gilbert, B.J. Harris and S.M. Riehl
the so-called unphysical sheet. These poles, also known as resonances or pseudo-
eigenvalues, are of significant interest to physicists in connection with impedance
theory, resonance scattering and spectral stability, and are correlated with the
existence of scattering states which remain localised for a long time [5], [12].
In Titchmarsh's examples, the influence of the eigenvalues of the unperturbed
problem is also reflected in the occurence nearby of sharp peaks in the perturbed
spectral density function, and away from these peaks the perturbed spectral den-
sity is small. Thus close to the eigenvalues of the original operator, there are
so-called points of spectral concentration of the perturbed operator, so that in
some sense the spectrum of the perturbed operator is 'close' to that of the original
[11]. Moreover, the fact that both resonance poles and points of spectral concen-
tration are located in the vicinity of the eigenvalues of the unperturbed problems
suggests that a correlation between spectral concentration and resonances may
hold more widely, and indeed there is considerable evidence to support this idea,
whether or no the phenomena arise in connection with the perturbation of a dis-
crete spectrum. For example, in those cases for which the Kodaira formula for the
spectral density holds ([13], p.940), it is evident that if a resonance pole in the
meromorphic continuation of the Green's function into the unphysical sheet lies
sufficiently close to the real axis, then it is liable to increase the spectral density
on part of the real axis closest to the pole. These related phenomena have given
rise to a considerable literature over the years and useful references may be found
in [12], [17].
Independently of the connection with resonance poles, the study of spectral
concentration in its own right introduces a valuable additional dimension into
analysis of the absolutely continuous part of the spectrum. In recent years, there
has been a significant focus on the development of analytical and numerical meth-
ods for investgating the existence and location of this phenomenon (see, e.g., [3],
[6]). The present paper provides an overview of some recent contributions of the
authors in this direction.
2. Mathematical background
We consider the one-dimensional Schrodinger operator He> associated with the
system
Ly(x) := -y"(x) + q(x)y(x) = zy(x), x E [0,(0), z E C,
y(O) coso: + y'(O) sino: = 0, 0: E [0,7r),
where the potential q(x) is real-valued, locally square integrable and decays to zero
as x ~ 00. In this situation the self-adjoint operator He> acting on 1i = L 2 [0, (0)
is defined by
where
D(He = {f E 1i : Lf E 1i; f, I'l0cally a.c.; cos o:f(O) + sin 0:1' (0) = O}.
Bounds for the Points of Spectral Concentration 141
It is well known that L is in Weyl's limit point case at infinity, so that for each
z E C\R there exists precisely one linearly independent solution of Lu = zu in
L 2 [0, (0), and that the essential spectrum of Ha fills the semi-axis [0, (0).
Associated with Ha is a non-decreasing spectral function, Pa(>'), and we
define the spectrum, O'(Ha), to be the complement in R of the set of points in a
neighbourhood of which Pa(>') is constant (which is consistent with the more usual
definition in terms of the resolvent operator). Subject to the normalisation Pa(O) =
0, the spectral function can be decomposed uniquely into absolutely continuous,
singular continuous and pure point parts, in terms of which the corresponding
components of the spectrum, O'a.c.(Ha ), O's.c.(Ha) and O'p.p.(Ha ), are defined in a
similar way.
The spectrum of Ha may also be studied through properties of the related
Titchmarsh-Weyl m-function, ma(z), which is defined and analytic on C\R and is
a Herglotz function in the upper half-plane. Formulae connecting Pa(>') and ma(z)
include
which holds for Zl, Z2 E C+, and
(2.1)
which holds for all >. E R for which the respective limits exist. The m-function is
closely related to the Green's function for Ha and hence reflects the analyticity
properties of the resolvent operator. It is also related to solutions of the differential
equation in various ways. For example, for Z E C+, let 'ljJ(x, z) denote the so-called
Weyl solution of Lu = ZU, which is in L 2 [0, (0) and satisfies 'ljJ(x, z) '" expikx as
x -> 00, where k 2 = z and the principal branch is chosen. Then the logarithmic
derivative of 'ljJ(x, z) with respect to x, evaluated at x = 0, is equal to the value of
the Dirichlet m-function, mo(z), at z, i.e.,
'ljJ'(O, z)
'ljJ(O,z) =mo(z)
We remark that for z E C+, 'ljJ(x, z) cannot vanish at x = 0, since to do so

would imply that Ho had a non-real eigenvalue at z, and thus contradict the self-
adjointness of Ho. In a similar way, 'ljJ(xo, z) #- for any Xo > 0, z E C+. We may
therefore set
'ljJ'(x, z)
m(x, z) := 'ljJ(x, z) (2.2)
for x ~ 0, z E C+, and it is straightforward to check that for each Xo ~ 0,

m(xo, z) is the Dirichlet m-function associated with the system: Lu = zu, x ~ Xo,
u(xo, z) = 0. We will refer to m(x, z) as the generalised Dirichlet m-function,
and note that in terms of our earlier notation, m(O, z) == mo(z). Since 'ljJ(x, z) is a
solution of Lu = zu, it follows from (2.2) that m(x, z) satisfies the Riccati equation
a
ax m(x, z) = -z + q(x) - (m(x, z))
2
(2.3)
for x E [0,(0), z E C+.
In the present context, we formally define spectral concentration as follows
(cf. [3]).
Definition 2.1. The point Ac E R is said to be a point of spectral concentration of
Ha if
(i) p~(A) exists finitely and is continuous in a neighbourhood of Ac, and
(ii) p~(A) has a local maximum at Ac.
: :;
We note that since Pa(A) is non-decreasing, the definition implies that p~(A) ex-
ists and satisfies p~(A) < p~(Ac) < 00 in a deleted neighbourhood of Ac, from
which it follows by the local continuity of p~(A) that p~(A) > in a neighbour-
hood of Ac. Thus the definition effectively restricts attention to points of spectral
concentration which occur in subintervals of the essential spectrum in which the
spectral function is purely absolutely continuous and strictly increasing. A further
consequence of the definition is that if p~(A) exists and has one sign for A > M,
then p~(A) exists, is absolutely continuous, but has no local maxima in (M, (0),
so that M is an upper bound for points of spectral concentration of Ha.
3. Short range potentials

In the case where q E L 1 [0,(0), it was shown by Titchmarsh [21J that for A> 0,
ma(A) := lim m",(A + if)
dO
exists and satisfies
(A) _ /11 (A) + iVl(A)
ma - /1(A) + iV(A) ,
where J.Ll (A), VI (A), J.L(A), V(A) are continuous functions of A, and J.L(A) , V(A) do not
vanish simultaneously. It follows from the properties of these functions and (2.1)
above that
, 1 ( 1
Pa(A) = ;CSma A) = 1fv0.(J.L2(A) + V2(A)) ,
so that p~(A) is continuous with
< p~(A) < 00 for A > 0, and hence a(Ha)
is purely absolutely continuous on (0,00) [9J. It is not hard to show that the
generalised Dirichlet m-function and corresponding Dirichlet spectral functions on
L 2 [x, (0) have similar properties, so that m(x, z) may be continuously extended
on to the non-negative real axis, z = A E R+, for x 2: 0. We then have (cf. (2.2))
that m(x, A) is well defined, continuous and non-real for x 2: 0, A > 0, and satisfies
'Ij;'(x, A)
m(x, A) = 'Ij;(x, A) ,
where 'IjJ(x, A) <I- LdO, (0) is the pointwise limit as z 1 A of the Weyl solution, and
is itself a solution of Lu = AU. lt follows that the Riccati equation (2.3) is also
satisfied for z = A > 0, so that
8 2
8xm(X,A)=-A+q(x)-(m(x,A)) , (3.1)
where m(x, A) is the finite non-real limit as z 1 A of the generalised Dirichlet
m-function, m(x, z).
We now show that we can investigate the behavior of the spectral density,
p~(A), using the Riccati equation. For x ?: 0, A > 0, we have
m(x, A) := ?Rm(x, A) + i<Zrm(x, A),

from which by (2.1),
m(O, A) := mo(A) = ?Rmo(A) + i1lp~(A). (3.2)

Hence, in principle, p~(A) can be obtained for A > by finding the appropriate
solution of the Riccati equation and evaluating at x = 0. If in addition it can be
shown that m(x, A) is differentiable with respect to A for sufficiently large A, we
can also seek conditions under which p~(A) exists and satisfies
p~(A) = ~ [<Zr :A m(x,A)L=o (3.3)
Equations (3.1) and (3.3) will form the basis for our investigation into the existence
of upper bounds for points of spectral concentration of H o.
lt is rarely possible to solve the Riccati equation explicitly, so we proceed by
postulating a series representation for m(x, A), which is substituted into (3.1). We
then choose the terms of the series, establish sufficient conditions for the validity
of the representation, and investigate the existence of p~(A). Based on the known
asymptotic behavior of m(x, A) as A ---+ 00 (cf. [10]; [18], Theorem 5.1), we seek a
series representation in the form
L mn(x, A)
00
m(x, A) = iV).. + g(x, A) where g(x, A) := (3.4)

n=O
is in L1([0, (0); dx), and satisfies g(X,A) ---+ as x ---+ 00. lt may be shown that
the Riccati equation has at most one solution of this form (see [7]), from which it
follows that if the series representation in (3.4) is a valid solution of (3.1), then it
does indeed represent the extension of the generalised Dirichlet m-function onto
the real axis, as sought.
Substituting for m(x, A) from (3.4) into (3.1) and rearranging yields
00
n=3
= q- mi - L m;_l + 2m n - l n-2
00 (
L mk )
n=3 k=l
where' denotes differentiation with respect to x. Choosing the {m n } to satisfy
m~ + 2iv'>:ml q
m~ + 2iv'>:m2 -mi
+ 2mn-l L
n-2 )
- ( m~_1 mk , n 2: 3,
k=1
yields
_e- 2i v'Xx 1 00
e2i v'Xtq(t)dt,
e- 2i v'Xx 1 00
e 2i v'Xt m i(t, )")dt, (3.5)
e- 2i v'Xx 1 x
00
e 2i v'Xt (m~_1 + 2mn-l I:
k=1
mk) dt,
In order to determine under what circumstances p~()..) exists for sufficiently large
).., we now define
wn(x,)..) :=
o
0).. mn(x, )..), n = 1,2,3, ....
for those).. for which the derivatives exist. The following lemma establishes some
key properties of {mn (x, A)} and {wn (x, A)}, and is proved in [7].
> such that

and)" > Al
Lemma 3.1. Let q(x) E L 1[0, 00) and suppose that there exists Al
for x 2:
11 00
e2iv'Xtq(t)dtl :::; a(x)r]()..),
where a(x) E L 1 [0, 00) is decreasing, ry()..) ---+ as)..

1. Then for x 2: 0, ).. > Al and n = 1,2,3, ...
---+ 00, and 32ry()..) 10 00
a(t)dt :::;
a(x)ry()..)
1
2n - 1
00
I wn(x,)..) I < ry()..) a(t)dt,
2n - 1 v0: x
so that the series 2::=1 mn(x,)..) and 2:::1 wn(x,)..) are uniformly absolutely con-
vergent in x and )...
We remark that the conditions of Lemma 3.1 are satisfied, for example, if (1 +
x )q( x) E Ll [0, 00) or if q( x) E Ll [0,00) is monotonic [10]. The convergence and
continuity properties of the series in Lemma 3.1 imply that
(i) m(x, .>..), as defined in (3.4) and (3.5), is a valid series representation of the
generalised Dirichlet m-function for x ~ 0, .>.. > AI, and
(ii) p~ (.>..) exists for .>.. > A1 and is given by
p~(.>..) = ~ (2~ + ~8'wn(O,'>")).

This leads to the following result.
Theorem 3.2. Let q(x) E L1[0,(0) and suppose that the hypothesis of Lemma 1 is
satisfied. Then for all .>. > AI, p~(.>..) exists and satisfies
"I
I p~(.>..) _ 21fY'>"
1 s 4,,1]('>") rX) a(t)dt,
1fY'>" io
so that p~ (.>..) > 0 for all .>.. > AI. In particular, there are no points of spectral
concentration of Ho for.>.. > AI.
Theorem 3.2 enables explicit upper bounds for points of spectral concentration to
be calculated, as illustrated in the first example in Section 5. Details of the proof
of this theorem are given in [7].
4. Long range potentials

In the case where q(x) ---+ 0 as x ---+ 00, but q(x) tf. L1[0, (0), the situation is
more delicate. It is no longer true in general that p~(.>..) exists, is continuous and
satisfies p~(.>..) > 0 for all .>.. > O. If q(x) decays more slowly than the Coulomb
potential, examples can be constructed where po('>") is discontinuous on a dense
set of eigenvalues in [0, (0) [15]. If q(x) fails to be in L2[0, (0), then the absolutely
continuous spectrum may be empty, in which case there is a dense set of points
in [0, (0) on which p~(.>..) does not exist as a finite limit [4], [9], [20]. The situation
is more fully understood in the case of von Neumann-Wigner type potentials,
where the spectrum is purely absolutely continuous on (0,00) apart from an at
most countable set of isolated eigenvalues, known as resonances [1], [2]. Moreover,
under fairly minimal smoothness conditions, classes of decaying, but non-integrable
potentials do exist for which the spectrum is purely absolutely continuous on
(0, (0) or on (M, (0) for sufficiently large M [2], [6]. In such circumstances, we can
investigate whether it is possible to extend m(x, z) continuously onto part of the
non-negative real axis in such a way that p~(.>..) exists and (3.1), (3.2) and (3.3) are
satisfied for sufficiently large .>... If this can be achieved, then estimates of upper
bounds for both embedded singular spectrum and points of spectral concentration
can normally be obtained.
We proceed as before by postulating the existence of a series representation
for the generalised Dirichlet m-function, m(x, z), in this case for 3tz > 0, 8'z ~ 0,
and I z I sufficiently large. It is necessary to consider the m-function in the upper
half-plane as well as on the real axis, so as to ensure that if m(x,.>..) exists and is
continuous in x and >. for 1>. I> M, then it is the unique continuous extension of
m(x, z), z E C+, as z ! >. E R. Also, in order to construct a series with the desired
convergence properties, it is helpful to introduce an additional term, R(x, z), into
the expression for m(x, z), so we now seek a representation in the form
m(x, z) = iy'z + R(x, z) + g(x, z) (4.1)
where
L mn(x, z)
00
g(x, z) :=
n=O
is in L 1([0,00);dx), and satisfies g(x,z) ---+
as x ---+ 00. The introduction of
R(x, z) enables the terms of the series to be generated iteratively with Q(x, z) :=
q(x) - R' - R2 - 2iv'AR E L 1([0, (0); dx) having an analogous role to that of q(x)
in the short range case. We observe that in general the choice of R(x, z) is not
unique.
Proceeding as before, we substitute for m(x, z) from (4.1) into (2.3), and after
rearrangement the {m n } are chosen to satisfy:
m~ + (2iy'z + 2R)ml Q
m~ + (2iy'z + 2R)m2 -mi
m~ + (2iy'z + 2R)mn - (m;_1
k=1
mk) , + 2mn-l
n 3, I:: ~
The solutions {m n } of these equations, and their derivatives, form the basis
of our analysis of the long range case. This leads to the following theorem which
is proved in [8].
Theorem 4.1. Let q(x) be continuously differentiable and satisfy q(x) ---+
00, q(x) <I. L 1[0, (0). Define

as x ---+
Q(x, z) := q(x) - R' - R2 - 2iy'zR

for ~z > 0, 'Sz ~ 0, where R = R(x,z) is chosen so that Q(',z) E L 1[0,(0),

, denotes differentiation with respect to x, and Q, R, ~, ~~ are continuous in x
and z. Suppose that there exists M > so that
(a) for ~z > 0, 'Sz 0, 1z I> M,
:s
~
(i) there exists K E R so that for x < t,
~ (2iy'z(t - x) + 21t R(s, Z)dS) :S K,

(ii) for :S x < t,
e2i y'z(t-x)+2J: R(s,z)dsQ(t, z)dtl :S a(x)1](z),
00
11

where a(x), 1](z), are real-valued functions with a(x) E L 1[0, (0) and decreas-
Jo
ing, 1](z) ---+ as 1 z 1---+ 00 and 321]e K oo a(t)dt :S 1,
(iii) 1:z it R(s, z)dsl :::; const(t - x) for 0:::; x < t < 00,
(b) for.>.. = Rz > M, 8'z = 0, there exists a decreasing function b(x) such that for
1= ~~
x ~ 0,
eK 1 1+ 1 ~ + 2 ~~ 1a(t)7](.>..)dt :::; j1 b(x).

Then p~(.>..) exists for.>.. > M, and satisfies
I p~(.>..) - 1" _
.>.
21fY
~8'R(O, .>..)1:::;
1f
3,,7](.>..)b(0).
1fY .>.
The following corollary may be inferred from the proof of Theorem 4.1 (see [8]).

Corollary 4.2. Let q(x), Q(x, z) and R(x, z) be as in Theorem 4.1 and suppose that
Ao > exists such that for Rz > 0, <Sz ~ 0, 1z I> Ao, conditions (a)(i}and (ii)
of Theorem 4.1 are satisfied. Then for.>.. = Rz > Ao, p~(.>..) exists as a finite limit,
and hence the spectrum of Ho is purely absolutely continuous on (Ao, (0).
5. Applications
Example. Let q(x) = (1 + x)-r sin(l + x), 'Y > 1. By expressing sin(l + x) in
exponential form and integrating by parts we obtain
11 x
00
e2iv'Xtq(t)dtl <
-
2
(2v1A-1)(1+x)r
for.>.. > ~,from which we may choose 7](.>..) = 2(2vIA -1)-1 and a(x) = (1 +x)-r.
It is then straightforward to show from Theorem 3.2 that
A1 = (~+~)2
2 'Y - 1
is an upper bound for points of spectral concentration of Ho.
Example. Let q(x) = sin(l + x)~(l + x)-~ . In this case q(x), q'(X) and (q(X))2
are not in L 1 [0,(0), so we take
q q'
R(x, z) = 2iy'z - (2iy'z)2 (2iy'z)3
to give Q(x, z) = 0(1 + x)-~ E L 1 [0, (0). This leads to the choice
a(x) - 1 '1l(z) _ _1_ b(x) _ 47
-(l+x)~' -5Izl' -5yT+X'

'I
from which it follows by Theorem 4.1 that p~(.>..) > if.>.. > 30, so that A1 = 30
is an upper bound for points of spectral concentration of Ho. Note that O"(Ho)
is known to be purely absolutely continuous on (0,00) [2], so that the issue of
embedded singular spectrum does not arise.
Example. We consider the von Neumann Wigner type potential
L
M
q(x) = hk(x)e2ickX,

k=-M
where for each k = -M, ... , M, Ck E R\{O}, hk(x) ----t as x ----t 00, hk(x) E
CL[O, 00) and hf+1(x) E AC[O, 00). We suppose also that there exists a real-valued
non-negative function p(x) such that x(p(x))L+2 is decreasing with (p(x))L+2,
x(p(x))L+2 E L 1[0, 00), and that for j = 0, ... , L + 1, k = -M, ... , M,
1h~)(x) I:::; (p(X))H1.
Then R(x, z) may be chosen so that, after successive integrations by parts,
1 Q(x,z) I: :; Cz Ir(:~LcJ L+2,

C
where c and c* are real constants which are computable for given q(x). We may
1
then take
roo c
ix (p(t))L+ 2dt, ".,(z) = b(x) =
00
a(x) = (I z I! -2Lc*)L+2' a(t)dt,
from which the existence of computable upper bounds, Ao and A1 , for resonances
(embedded eigenvalues) and points of spectral concentration follows from Corollary
4.2 and Theorem 4.1 respectively. Further details may be found in [8].
References
[1] H. Behncke, Absolute continuity of Hamiltonians with von Neumann Wigner poten-
tials, Pmc. AMS 111 (1991), 373-384.
[2] H. Behncke, Absolute continuity of Hamiltonians with von Neumann Wigner poten-
tials II, Manuscripta Math. 71 (1991), 163-181.
[3] B.M. Brown, M.S.P. Eastham and D.K.R. McCormack, Spectral concentration and
perturbed discrete spectra, J. Compo Appl. Math. 86 (1997), 415-425.
[4] P. Deift and R. Killip, On the absolutely continuous spectrum of one-dimensional
Schrodinger operators with square summable potentials, Comm. Math. Phys. 203
(1999), 341-347.
[5] C.L. Dolph, J.B. Mcleod and D. Thoe, The analytic continuation of the resolvent
kernel and scattering operator associated with the Schroedinger operator, J. Math.
Anal. Appl. 16 (1966), 311-332.
[6] M.S.P. Eastham, On the convexity of the Sturm-Liouville spectral functions with
slowly decaying potential, Indian J. Math. 42 (2000), 9-20.
[7] D.J. Gilbert and B.J. Harris, Bounds for the points of spectral concentration of
Sturm-Liouville problems, Mathematika 47 (2000), 327-337 (2002).
[8] D.J. Gilbert, B.J. Harris and S. Riehl, The spectral function for Sturm-Liouville
problems where the potential is of von Neumann-Wigner type or slowly decaying,
J. Differential Equations 201 (2004), 139-159.
[9] D.J. Gilbert and D.B. Pearson, On subordinacy and analysis of the spectrum of
one-dimensional Schrodinger operators, J. Math. Anal. Appl. 128 (1987), 30-56.
[10] B.J. Harris, The form of the spectral functions associated with Sturm- Liouville
problems with continuous spectrum, Mathematika 44 (1997), 149-161.
[11] D.B. Hinton and J.K. Shaw, Spectral concentration for Hamiltonian systems, Helv.
Phys. Acta 58 (1985), 982-994.
[12] P.D. Hislop and I.M. Segal, Introduction to Spectral Theory, with Applications to
Schrodinger Operators, Springer-Verlag, New York, 1996.
[13] K. Kodaira, The eigenvalue problem for ordinary differential equations of the second
order, and Heisenberg's theory of S-matrices, Amer. J. Math. 71 (1949), 921-945.
[14] J. von Neumann and E. Wigner, Uber merkwiirdige diskrete Eigenwerte, Phys. Z.
30 (1929), 465-467.
[15] S. Naboko, On the dense point spectrum of Schrodinger and Dirac operators, Teoret.
Mat. Fiz. 68 (1986), 18-28.
[16] J.R. Oppenheimer, Three notes on the quantum theory of aperiodic effects, Phys.
Rev. 31 (1928), 66-8l.
[17] M. Reed and B. Simon, Methods of Modem Mathematical Physics, IV: Analysis of
Operators, Academic Press, 1978.
[18] A. Rybkin, Some new and old asymptotic representations of the Jost solution and
the Weyl m-function for Schrodinger operators on the line, Bull. London Math. Soc.
34 (2002), 61-72.
[19] E. Schrodinger, Quantisierung als Eigenwertproblem III, Ann. der Physik 80 (1926),
457-490.
[20] B. Simon, Operators with singular continuous spectrum, I. General operators, Annals
Math. 141 (1995), 131-145.
[21] E.C. Titchmarsh, Eigenfunction Expansions I, 2nd edition, Oxford University Press,
1962.
[22] E.C. Titchmarsh, Eigenfunction Expansions II, Oxford University Press, 1958.
Daphne J. Gilbert
School of Mathematical Sciences
Dublin Institute of Technology
Ireland
Bernard J. Harris
Department of Mathematical Sciences
Northern Illinois University, USA
Suzanne M. Riehl
University of Northern Iowa, USA
Operator Theory:
Reconstructing Jacobi Matrices

from Three Spectra
Abstract. Cut a Jacobi matrix into two pieces by removing the n-th column
and n-th row. We give necessary and sufficient conditions for the spectra ofthe
original matrix plus the spectra of the two submatrices to uniquely determine
the original matrix. Our result contains Hochstadt's theorem as a special case.
Mathematics Subject Classification (2000). Primary 36AlO, 39A70; Secondary

34B24, 34L05.
Keywords. Jacobi matrices, spectral theory, trace formulas, Hochstadt's the-
orem.
1. Introduction
The topic of this paper is inverse spectral theory for Jacobi matrices, that is,
matrices of the form
H= (1.1)
aN-2 bN - 1 aN-l
aN-l bN
This is an old problem closely related to the moment problem (see [9] and the
references therein), which has attracted considerable interest recently (see, e.g.,
[1] and the references therein, [3], [4], [8]). For analogous results in the case of
Sturm-Liouville operators see [2], [6], and [7]. In this note we want to investigate
the following question: Remove the n-th row and the n-th column from Hand
denote the resulting submatrices by H_ (from b1 to bn - 1 ) respectively H+ (from
bn + 1 to bN ). When do the spectra of these three matrices determine the original
matrix H? We will show that this is the case if and only if H _ and H + have no
eigenvalues in common.
152 J. Michor and G. Teschl
From a physical point of view such a model describes a chain of N particles

coupled via springs and fixed at both end points (see [11], Section 1.5). Determining
the eigenfrequencies of this system and the one obtained by keeping one particle
fixed, one can uniquely reconstruct the masses and spring constants. Moreover,
these results can be applied to completely integrable systems, in particular the
Toda lattice (see, e.g., [11]).
2. Main result
To set the stage let us introduce some further notation. We denote the spectra of
the matrices introduced in the previous section by
O"(H) = {Aj}f=l' O"(H_) = {tlk}~~i, O"(H+) = {tln~-;-n (2.1)
Moreover, we denote by (tlj)f=-/ the ordered eigenvalues of H_ and H+ (listing
common eigenvalues twice) and recall the well-known formula (see [1], Theorem
2.4 and Theorem 2.8)
TI;=-;-l(Z - tlj) -1
g(z,n)=- N
TI j =l (z - Aj)
= b
z - n + anm+ z,n
2 ( )
+ a n2 _ 1 m- (z,n
)' (2.2)
where g( z, n) are the diagonal entries of the resolvent (H - z) -1 and m (z, n) are
the Weyl m-functions corresponding to H_ and H+. The Weyl functions m(z, n)
are Herglotz and hence have a representation of the following form
n-l _ 71,-1
L~'
k=l tlk - Z k=l
(2.3)
N-n + N-n
"
L... +
_a_
l _
' aT > 0, L aT = 1. (2.4)
1=1 tll - Z 1=1
With this notation our main result reads as follows

Theorem 2.1. To each Jacobi matrix H we can associate spectral data
{Aj}f=l, (tlj,O"j)f=-:/, (2.5)
where O"j = +1 if tlj E O"(H+)\O"(H_), O"j = -1 if tlj E O"(H_)\O"(H+), and
2 +
anal - a n2 - 1 a k-
O"j = 2 + 2 - (2.6)
anal +a n - 1 a k
if tlj = tlk = tlT-
Then these spectral data satisfy
(i) Al < tll :::; A2 :::; tl2 :::; ... < AN,
(ii) O"j = O"j+l E (-1,1) if tlj = tlj+l and O"j E {1} if tlj -I- J-li for i -I- j
and uniquely determine H. Conversely, for every given set of spectral data satis-
fying (i) and (ii), there is a corresponding Jacobi matrix H.
Reconstructing Jacobi Matrices from Three Spectra 153
Proof. We first consider the case where H_ and H+ have no eigenvalues in com-
mon. The interlacing property (i) is equivalent to the Herglotz property of g(z, n).
Furthermore, the residues ai can be computed from (2.2)
N-n +
z-bn+a n2 ~
' " al
--+-
TI nk=1 J-Lk- ) TI 1=1- J-LI+)
- 1( N n(
Z - Z - 1=1 z - J-LI
(2.7)
and are given by ai = a;;~I(3i, where
(3-:- = _ TI)=1
N ( -:- - A)
J-L
n-l
a~-1 = L(3i (2.8)

2 )
2 TI l#i (-
J-Li - J-LI-)TI 1=1- ( J-Li-
N n
- J-LI+)' i=1
1ar 1y, a l+ = an-2(3+

SIml h
I ' were
N-n
a~ = L (3t. (2.9)
1=1
Hence m(z, n) are uniquely determined and thus H by standard results from the
moment problem. The only remaining coefficient bn follows from the well-known
trace formula
N n-l N-n
bn = tr(H) - tr(H_) - tr(H+) = L Aj - L J-Lk - L J-Lf. (2.10)
j=1 k=1 1=1
Conversely, suppose we have the spectral data given. Then we can define an,
an-I, bn , a k , at as above. By (i), a k and at are positive and hence give rise
to H. Together with an, an-I, bn we have thus defined a Jacobi matrix H. By
construction, the eigenvalues J-Lk' J-Lt are the right ones and also (2.2) holds for H.
Thus Aj are the eigenvalues of H, since they are the poles of g(z, n).
Next we come to the general case where J-Ljo = J-Lko = J-L~ (= Ajo) at least for
one jo. Now some factors in the left-hand side of (2.7) will cancel and we can no
longer compute (3ko' (3~, but only "fjo = (3ko + (3~. However, by definition of ajo
we have
- 1 - a jo (3+ _ 1 + a jo .
(3k o = 2 "fjo' 10 - 2 "f)o (2.11)
Now we can proceed as before to see that H is uniquely determined by the spectral
data.
Conversely, we can also construct a matrix H from given spectral data, but
it is no longer clear that Aj is an eigenvalue of H unless it is a pole of g(z, n).
However, in the case Ajo = J-Lko = J-L~ we can glue the eigenvectors u_ of H_ and
u+ of H+ to give an eigenvector (u_, 0, u+) corresponding to Ajo of H. 0
154 J. Michor and G. Teschl
The special case where we remove the first row and the first column (in which
case H_ is not present) corresponds to Hochstadt's theorem [5]. Similar results for
(quasi-)periodic Jacobi operators can be found in [10].
Acknowledgments
We thank the referee for useful suggestions with respect to the literature.
References
[1] F. Gesztesy and B. Simon, m-functions and inverse spectral analysis for finite and
semi-infinite Jacobi matrices, J. Anal. Math. 73, 267-297 (1997).
[2] F. Gesztesy and B. Simon, On the determination of a potential from three spectra,
in Differential operators and spectral theory, 85-92, Amer. Math. Soc. Transl. Ser.
2, 189, Amer. Math. Soc., Providence, RI, 1999.
[3] G.M.L. Gladwell, On isospectral spring-mass systems, Inverse Probl. 11, 591-602
(1995).
[4] P. C. Gibson, Inverse spectral theory of finite Jacobi matrices, to appear in Memoires
of the Amer. Math. Soc.
[5] H. Hochstadt, On the construction of a Jacobi matrix from spectral data, Lin. Algebra
Appl. 8, 435-446, 1974.
[6] R. Hryniv and Ya. Mykytyuk, Inverse spectral problems for Sturm-Liouville operators
with singular potentials. Part III: Reconstruction by three spectra, J. Math. Anal.
Appl. 248, 626-646, (2003)
[7] V. Pivovarchik, An inverse Sturm-Liouville problem by three spectra, Integral Equa-
tions Operator Theory 34 no. 2, 234-243, (1999).
[8] C.-T. Shieh, Reconstruction of a Jacobi matrix with mixed data, preprint.
[9] B. Simon, The classical moment problem as a self-adjoint finite difference operator,
Advances in Math. 137,82-203 (1998).
[10] G. Teschl, Trace Formulas and Inverse Spectral Theory for Jacobi Operators, Comm.
Math. Phys. 196, 175-202 (1998).
[11] G. Teschl, Jacobi Operators and Completely Integrable Nonlinear Lattices, Math.
Surv. and Mon. 72, Amer. Math. Soc., Rhode Island, 2000.

Institut fur Mathematik
Strudlhofgasse 4
A-1090 Wien, Austria
and
International Erwin Schrodinger Institute for Mathematical Physics
Boltzmanngasse 9
A-1090 Wien, Austria
e-mail: Johanna.Michoresi.ac.at
e-mail: Gerald.Teschlunivie.ac.at
URL: http://www.mat.univie.ac.at/rvgerald/
Operator Theory:
On Some Asymptotic Properties of Solutions

for a Particular Class of
Finite Difference Equations
Andrey S. Osipov
Abstract. For one class of linear difference equations of arbitrary order we

obtain some results about the asymptotic behavior of its solutions, applying
a method used in the spectral analysis of discrete Sturm-Liouville operators.
We apply this asymptotic method to show, that the analog of the Hellinger-
Wall theorem of invariance in z2 is not valid in lP for p > 2.
Mathematics Subject Classification (2000). 47B36;39All.
Keywords. Band operators; asymptotic methods; lP-properly; Hellinger-Wall
theorem.
One of the main methods of spectral analysis of nonsymmetric difference opera-

tors generated by infinite band matrices (band operators) is based on the study of
asymptotical behavior of polynomials defined by systems of orthogonality relations
[1], [2] . Such polynomials satisfy the difference equation, associated with the stud-
ied operator. Although some asymptotic methods for such sort of linear difference
equations are known [3], [4] either they are not giving the explicit formula, or
require some special conditions to be fulfilled (e.g., dichotomy-type condition). In
the present paper, we consider the asymptotic method, which was used for the
analysis of absolutely continuous spectrum of second order difference operators in
[5]. This method is based on direct analysis of the transfer matrices, corresponding
to the difference equation, and under some simple assumptions it gives an explicit
asymptotic formula. First for the studied system of finite-difference equations of an
arbitrary order we obtain some asymptotic formulas similar to [5]. Then we apply
the method to study an analog of the Hellinger-Wall theorem for the difference
operator of a second order. Note that the continuous analog of the Hellinger-Wall
theorem is a known Weyl theorem, which states that for the Sturm-Liouville oper-
ator on a semi-axis, the property of all solutions of the corresponding differential
This work was supported by the Academy of Finland and RFBR: projects 02-01-00790 and
00-15-96100.
156 A.S.Osipov
equation to belong to the space L2, is invariant with respect to the spectral pa-
rameter [6]. We show that the claim of the Hellinger-Wall theorem is not true for
the spaces [P, when p > 2.
1. The asymptotic method

We consider the following infinite system of finite difference equations
J.Ln-qUn- q + an,n-q+lUn-q+l + ... + an,n+q-lUn+q-l + Anun+q = ZUn , (1)
n > q, J.Ln =I- 0, An =I- 0,
(for some fixed index q 2: 1 and parameter Z E C) with the complex coefficients,
satisfying the following conditions:
J.Ln = n'Y(n) (1 + ~n), An = na(n) (1 + 6n ); {6 n }, {~n} E [2, (2)
for alln > q, o:(n),')'(n) E (1/2,00),
co
2: lan,n_iln-a(n) < 00, i = -q + 1, ... , q - l. (3)

n=q+l
First consider the case when for all n > q 0:( n) = ')'( n) = 0: and assume for
simplicity that all the coefficients an,n-i, (i = -q+ 1, ... , q-1) for n > q are equal
to zero. Denote by h(al,"" a2q-k) (or Lk(al, ... , a2q-k)) the matrix of order
2q where the only nontrivial diagonal, consisting of the elements al, ... , a2q-k
is shifted k steps to the right (or k steps to the left) with respect to the main
diagonal; h == h(l, ... , 1). Then system (1) admits the following representation:
Un +1 = B(n)un , n > q, where Un =

Cn_, )
: ' and (4)
Un+q-l
0 1 0 0 0 0 0
0 0 1 0 0 0 0
0 0 0 1 0 0 0
B(n) = (B(n)i,j)7,;=1 =
0 0 0 0 0 0 1
-b n 0 02q,q Pn 02q,q+2 0
= h + I_q+1(0, ... ,0,Pn ) - I_ 2q+1(bn) , bn = J.Ln-q

~,Pn = Z
An'
Matrix B(n) is a so-called transfer matrix corresponding to (1) with the restriction
that all coefficients an,n-i, (i = -q + 1, ... , q - 1) for n > q are equal to zero. If
q = 1,
B(n) = (-~n p~),

On the Asymptotics of Solutions of Difference Equations 157
One can easily check that for n - 2q + 1 >q

B(n, n - 2q + 1) == B(n)B(n - 1) ... B(n - 2q + 1)
= B(n, n - q + l)B(n - q, n - 2q + 1)
q times
~
= [Iq + diag(O, ... , 0, Pn-q+1, Pn-q+2,, Pn) - L q(bn- q+1, ... , bn )]
q times
~
X + diag(O, ... , 0 Pn-2q+1, Pn-2q+2,. Pn-q) - Lq(bn-2q+1, ... ,bn- q)]
[Iq
= diag( -bn- 2q+1, ... ,-bn- q, -bn- q+1, ... ,-bn) + Iq(Pn-2q+I, ... , Pn-q)
- I- q(Pn-q+1 bn-2q+1, Pn-q+2 bn-2q+2, ... ,Pnbn-q),
where bi = bi - PiPi-q. Since {O, {6} E [2, we get
J.1n-q qa 1 En6n
bn = -\- = (1 - - + O( 2" ))(1- En + --,) = 1 -!3n + rn ,
An n n 1 + Un
where En = 6n - ~n-q,!3n = qa/n + En, {rn} E [1. We also have
{Pnbn-q - Pn}, {b n - bn } E [\
because a > 1/2.
In view of the above, we can write
B(n, n - 2q + 1) = B(n, n - 2q + 1) + Rn , where
B(n, n - 2q + 1) = diag( -bn- 2q +l, ... ,-bn )
+Iq(Pn-2q+1, ... , Pn-q) - I- q(Pn-q+1, ... , Pn),
and {II Rn II} E [1. Again, using the condition a > 1/2 we obtain that
{Pn - Pn-d E [1 for i = 1, ... , 2q - 1 and {Pn} E [2, so the formula for
B(n, n - 2q + 1) can be written as
- exp( -PnP)[I - diag(!3n-2q+l, ... , !3n)][I + Sn],
where P = Iq - L q , {IISnll} E [1, and I is the identity matrix.
Now assume that for a fixed index i = 1, ... , 2q - I, {En - En-d E [1
for n = q(2k + I), kENo Then {!3n - !3n-d E [1, and we get
B(n, n - 2q + 1) = -(1 - !3n) exp( -PnP)[I + Bn], where {IIBnll} E [1.
Thus, for n = q(2k + 1) we are coming to the formula

n-q
""""'2q
U n+1 = (-1) n 2-;,q II (1 - !3q(2j+1)) exp( -PnP)(I + Bn) (5)

j=l
We have
exp( -PnP) = Cos(Pn)I - sin(Pn)Iq + sin(Pn)L q .
Hence, for real Pn this is a unitary matrix.
158 A.S.Osipov
Theorem 1. Suppose that for the system (1-3) the following conditions are fulfilled
(i) o:(n) = 'Y(n) = 0: for all n > q.
(ii) The series Ek Ek is convergent.
(iii) There exists an integer m, 0:::; m :::; 2q - 1 such that {En - En-i} E ll,
n = 2qk+m, k E N,i = 1, ... ,2q-1.
(iv) For all k > q, Pk are real.
If u = {Un}~=q+l is a non-zero solution of (1-3), then for Un defined by (4)
Un =
c< 1 n
pnn-'(exp(-- L
PkP)(J + (0(1))), n -+ 00 (6)
2q k=q+l
for a certain vector f E C 2q and a vector 0(1) of norm 0(1).
The method of proof is the same as for the Theorem 3.2 from [5J (see also the
example in the next section). If An = J.Ln, n = q + 1, ... ,00 (as in [5J for q = 1),
then
k=q+l k=N-q+l k=1

so the condition (ii) is automatically fulfilled (since {~d E l2). Note that for-
mula (5) used in the proof has been derived from the additional assumption that
an,n-i = 0 for all i = -q + 1, ... ,q -1, n > q. But, if in case of nonzero an,n-i the
condition (3) is fulfilled then the corresponding additional terms in the transfer
matrix B(n) give a summable error, and the formula (5) is also valid for this case.
Theorem 2. Under the assumptions of the Theorem 1 for any non-zero solution u
of (1) one obtains the asymptotic formula as N -+ 00
~ I 12 {O(N 1 - a ), 0: =f. 1,
~ Un = O(ln(N)), otherwise.
Proof. Let the conditions of the previous theorem be fulfilled for m = q. Then, for
!'!..::::.!J.
N = q(2k + 1), E:=llun I2 :::; Ep~~ Iluq(2p+l)+111 2. Applying (5) and using, that
II exp( -Pn P ) I = 1 we get
E~~~+I) lun l2:::; E7=1 I1~=111- ,Bq(2i+l) 12(1 + IISq(2i+l) 112)lluq+l 112
:::; C1E7=1 I1~=111- ,Bq(2i+1)1 2I1 U q+111 2
:::; C 2E7=1 exp( -2 E~=1 l,Bq(2i+l) I) Iluq+111 2
:::; C 3 E7=1 exp( -0: E~=1 J) I uq+111 2 :::; C 4 E7=1 l-a,
for some suitable positive constants Cm, m = 1, ... ,4. Here we used that
{,B}, {E} E l2 and the condition (i) of the Theorem 1.
On the other hand, there exists No so large, that IISq(2J+1) II < 1 and
LBq(2J+I) I < 1 for j ~ No So we obtain
q(2k+l) k I
L lunl2 ~ C 5 L II 1(1- j1q(2J+I))12(1-IISq(2J+I)1I2)lluq+1112 ...
n=1 I=No j=No
k
>
_ C 6 '"' l-O: ,
~
I=No
for some C 5 and C 6 . Thus the claim of the theorem is proved for N = q(2k + 1)
and for the other N it follows from the fact, that IIB(n)111 ---+ 1 as n ---+ 00. 0
This theorem can be applied to study the subordinated solutions of (1) for
q = 1. A solution u =f. 0 is called a subordinate solution of (1)
if for every solution
v linearly independent with u we have limn-->oo I:~=I IUkl 2/ I:~=I IVkl2 = O. We
mention here that one of the main tools in the spectral analysis of infinite Jacobi
matrices is the study of subordinated solutions of the systems (1) related to such
matrices [7], [8].
Let u = {Un}~=1 and v = {Vn}~=1 be two linearly independent solutions of
(1) for q = 1. One can easily check that for n ~ 2
UnVn+1 - VnUn+1 Mn-I( Un-IVn -

=~ Vn-IUn )
_ Mn-I ... MI ( _ ) _ 8n- 1Wo

- \ \ UI V2 VI U2 - \ '
An . A2 An
where 8n = t::t and Wo = UIV2 - VIU2 Assuming that infn 18nl ~ a > 0 we get
N I I N+I N+I
~ 7;n~ ~ 2(~ lunI2)~(~ IVnI2)~.
Applying the above theorem and using the same arguments as in [5] page 222, we
finally obtain
",N+I I 12
L.m=1 Un > >0
",N+I I 12 - Co .
L ..m=1 Vn
Thus the following theorem is proved.
Theorem 3. If the system (1-3), with q = 1, satisfies the assumptions of the
Theorem 1 and the additional assumption infnl8nl > 0, then it has no subordinated
solutions.
2. Application
The above asymptotic formulas were established under the assumption that Pk are
real, so exp( -PkP) are unitary. To establish (6) for the complex Pk we have to
additionally assume, that exp(I:%"=1 PkP) is convergent, which is fulfilled, e.g., for
160 A.S.Osipov
a> 1. For such a, as it follows from (6), all non-zero solutions of (1) are decreas-
ing not slower than n- a / 2 Nevertheless, for some cases when exp(2::%':l PkP) is
divergent, it is also possible to establish the analog of (6).
As an example, we consider the application of the studied asymptotic method
to the analysis of invariance of lP property for (1), with respect to z. Namely, the
following generalization of the Hellinger-Wall theorem [9] takes place.
Theorem 4. Suppose that the coefficients of the system
/-tn-qun- q + an,n-q+lUn-q+l + ... + an,n+r-lUn+r-l + AnUn+r = ZUn , (7)
n > q, q, r ~ 1, /-tn-q i= 0, An i= 0, Z E C,
satisfy the condition
inf IOi I > 0 > 0, (8)

where
s: /-tl/-t2 . /-ti
s:
Ul = /-tl, Ui = , i ~ 2.
Aq+lAq+2 ... Aq+i-l
Suppose also that for some p, 1 ::; p ::; 2, and Z = Zo E C, the series
2:::1 lUi (zo)IP is convergent for every solution U = u(z) = {Ui(Z)}~l of (7),and,
therefore u(zo) E lP. Then, for any M > 0, the series 2:::llui(Z)IP is uniformly
convergent for Iz - zol < M.
The proof is contained in [10]. Thus, the property of all solutions of (7) to
belong to lP (the lP property) is invariant with respect to Z E C for 1 ::; p ::; 2.
To check whether or not the statement of this theorem is true for lP, p > 2
we consider the system:
(9)
where
/-tn = {n(n + 1), = 2k, _ {n + 1, = 2k,
n \
2
n k
E
1M
/-tl = 2.
= 2k + 1, = 2k + 1,
An - l'l,
n+1, n n , n
Or, in other words,
nUn-l + (n + l)Un+l = ZUn for even n,
(n - l)nu n -l + n 2u n = ZU n otherwise.
It satisfies the condition (8) of Theorem 4. For even n a(n) = 'Y(n - 1) = 1,
on = lin, en-l = 1/(n-1); for odd n a(n) = 'Y(n-1) = 2, on = 0, en-l = 1/(n-1),
and for all n /-tn-dAn = I-lin. Our purpose is to obtain an asymptotic formula
for the solutions of (9). Since a(n) i= 'Y(n) we cannot apply here Theorem 1
(the conditions (ii)-(iii) are also not fulfilled), but we can use similar techniques.
Namely, consider the case of odd n. Then
+1
B(n)B(n -1) ( -1 o n -1'+~)
1 +
n
( 0
... _...
n3 n2
1 --
-(1 - -) exp( -PnP)(I + Sn),
n
where Pn = n'=-l'
22 z2
S
n
= (- n2(n-1)2
z
+ n(n-1)2
z
n(n-1) - n2(n-1)
Hence for odd n we have:
= B(n)B(n - 1) ... B(2)U2 = (-1)-2 (1 -
n-l n~l
Un+1 I1k~l 2k~1)

X exp( -Pn F )(I + Sn) exp( -Pn-2F)(1 + Sn-2)'" exp( -P3F)(I + S3)U2
= (-1) n;-" I1:E~ (1 - 2k~1) exp( - 2:::E~ PkF) I1:E~ (I + R2k+l)U2,
where R3 = S;{,
and for k > 1
k-1 k-1
R2k+1 = exp(LP2i+1F)S2k+1 exp(L -P2i+1F)
i=l i=l
= (1 2::7:112i~1) ((2k)2(~k+1) ~ (2kl"f:k+1)2 (2k)2~~~+1)2) (1

o 1 2k(2k+1) (2k)(2k+1)2 (2k)(2k+1)2 0
Thus the elements of R2k+1 are of order Ocn~~k)) and therefore {IIR 2k + 1 11} Ell.
Then, using that
n-l
-2- 1
II(I- 2k +l):::::exP(-L2k+l)'
n-l
-2- 1
p2=_1, (whereP=
(0
-1
1)
0);
k=l k=l
n-l n-l
-2- 1 l l -2- 1 I 1
n
L 2 k+l ="2 L k-"2 L (2k-2k+l)'

k=l k=2 k=l
for odd n we obtain

n-l n-l n-l
Un+1 = pn-1 exp( - 2::k~1 2k~1) exp( - 2::k~1 P2k+1 F ) I1k~l (I + R2k+1)U2
= p n - 1 exp( -~ 2::~~i i) exp( -~ 2::~~i iJkF)(e + 0(1))
= pn+1(n + 1)-1/2 exp( -~ 2::~~i PkF)(J + 0(1)), n -7 00,
where e and f are some vectors in C 2 and 0(1) is a vector of norm 0(1). Since
B(n) -7 P as n -7 00, the asymptotics for even n is the same.
Hence for all n
162 A.S.Osipov
where 9 is a certain vector in (:2. From this, one can immediately see that for z = 0
or z = i all solutions of the equation are of order O(n-l/2) and therefore belong
to the space Z2+0 for any f > O. However, for z = -1 there is a one-dimensional
subspace of solutions which are bounded but do not tend to zero, as n ---7 00. Thus
the following proposition is proved:
Proposition. The statement of the Theorem 4 is not true for p > 2.
Acknowledgment
The author is thankful to Bernhard Beckermann for important suggestions and
remarks and to the anonymous referee for valuable comments.
References
[1] A.I. Aptekarev V. Kaliaguine and W. Van Assche, Criterion for the resolvent set
of nonsymmetric tridiagonal matrix, Proc. Amer. Math Soc., vol. 123 no 8, (1995),
2423-2430.
[2] V.A. Kaliaguine, Hermite-Pade approximants and spectral analysis of nonsymmetric
operators, Mat. Sb., vol. 185 (1994) 79-100 (In Russian). English transl. in Russian
Acad. Sci. Sb. Math., vol. 82 (1995) 199-216.
[3] Yu.V. Sidorov, M.V. Fedoryuk, M.A. Shabunin, Lectures on the theory of functions
of a complex variable. Textbook, Third ed.(Russian) Moscow, Nauka, 478 p. (1989).
[4] Z. Benzaid and D.A. Lutz, Asymptotic representation of solutions of perturbed sys-
tems of linear difference equations, Studies in Applied Mathematics, Vol. 77, (1987)
195-221.
[5] J. Janas, S. Naboko, Jacobi matrices with power-like weights - grouping in blocks
approach, Journal of Functional Analysis, vol. 166., (1999), 218-243.
[6] E.A. Coddington, N. Levinson, Theory of ordinary differential equations, McGraw-
Hill Book Company, NY, 1955.
[7] S. Khan and D.E. Pearson, Subordinacy and spectral theory for infinite matrices,
Helv. Phys. Acta, vol. 65, no. 4, (1992), 505-527.
[8] G. Stolz, Spectral theory for slowly oscillating potentials. I. Jacobi matrices,
Manuscripta Math., vol. 84, no. 3-4, (1994), 245-260.
[9] H. Wall,Analytic Theory of Continued Fractions, Chelsea, Bronx, NY, 1973.
[10] A.S. Osipov, On the Hellinger theorem and lP properties of solutions of difference
equations, Journal of Difference Equations and Applications, vol. 9, no. 9, (2003),
841-851.
Andrey S. Osipov
Scientific Research Institute for System Studies
Russian Academy of Sciences
N akhimovskii pr. 36-1
Moscow, 117218, Russia
e-mail: andrei@sgo.fi
Operator Theory:
A Remark on Spectral Meaning of the

Symmetric Functional Model
Boris Pavlov
Abstract. The imaginary part of a dissipative operator L is weak if it is pre-

2
sented by a positive operator T such that the square T of it is a product
of an operator with a finite trace and an operator from Macaev class. For a
dissipative operator with a weak imaginary part the families of incoming and
outgoing scattered waves form a non-orthogonal and often even over-complete
system {Win, Wout} of eigenfunctions of the corresponding self-adjoint dila-
tion , . The rescription of L in the spectral representation associated with
{Win, Wout} gives the Symmetric Functional Model of L, and the characteris-
tic function S of L coincides with the transmission coefficient of the outgoing
waves. A general construction based on the self-adjoint delation and an ex-
ample of the Lax-Phillips Semigroup for the I-D wave equation on the infinite
string with a bounded non-negative potential supported by semi-axis are con-
sidered.
Mathematics Subject Classification (2000). 47A45.
Keywords. Symmetric functional model.
1. Introduction
The classical Nagy-Foias Functional Model of a dissipative operator is obtained
as an image of the dissipative operator in incoming or outgoing spectral repre-
sentation of it's self-adjoint dilation, see [10]. The symmetric rescription of the
Nagy-Foias Functional Model (we will call it further just a Symmetric Model) was
suggested in [12, 13] and developed in [6, 8, 15, 11, 17, 18], see complete bibli-
ography in [15, 11, 18]. The symmetric rescription of the Nagy-Foias functional
model gives most simple algebraic formulae for spectral objects of the model (the
eigenfunctions, the spectral density and others) and permits to calculate directly
important spectral characteristics, see [15, 11, 18]. The Symmetric Model can be
The author recognizes support from the grant of the Russian Academy of Sciences, RFBR 03-
01-00090.
164 B. Pavlov
obtained formally from the original Nagy-Foias model just by a non-degenerate

transformation of the space, see [12], which transforms the spectral density matrix
of the dilation into the spectral density matrix of the symmetric model:
s+
(~ 1- ~+ S ) ---+ ( 1 ~ ) (~ 1- ~+ S ) (~ ~)
=(~ S1+ ) := p(),), (1.1)
where S = S(k + iO) is the limit value of the characteristic function of the original
dissipative operator on the real axis from the upper half-plane and S+ are the limit
values of S+(k - iO) on the real axis from the lower half-plane. The above trans-
formation of the spectral data connects actually different kinds of eigenfunctions
which may be used for construction of the spectral representation. The standard
Nagy-Foias model was connected with the orthogonal system of eigenfunctions of
incoming and complementary ("radiating") components of the dilation, see [13, 14].
We denote them by {'lj;_, 'lj;<} respectively. An equivalent model may be connected
with outgoing {'lj;+} and "absorbing" eigenfunction {'lj;>} of the dilation. Then the
symmetric model is formally connected by the above formula (1.1) to the orthogo-
nal system {'lj;<,'lj;_ - S+'lj;<} or the dual system N>,'lj;+ - S'lj;>}. Both systems
look rather ugly, and hence the recipe of the construction of the functional model
as an image of the original dissipative operator in the spectral representation
associated with this system seems neither elegant nor persuading. Besides the sug-
gested transformation of the classical Nagy-Foias model to the symmetric one is
not unique. Hence there exists an extended set of systems of eigenfunctions of the
dilation which can be used for construction of the spectral representation relevant
to the symmetric model.
We will reveal the spectral meaning of the symmetric model via finding the
most natural spectral representation of the dilation, such that the rescription of the
original dissipative operator in terms of it coincides with the symmetric functional
model, similarly to the classical case. We show that the symmetric functional
model is associated with a certain set of eigenfunctions of the absolutely-continuous
spectrum of the model: the incoming and outgoing scattered waves {'lj;_, 'lj;+}.
Generally, this set of eigenfunctions is not orthogonal. Hence the corresponding
matrix p of the spectral density is non-diagonal and may degenerate. Nevertheless
the spectral representation is properly defined, up to non-essential addenda, as
stated in [13, 11].
The plan of actual paper is the following: in the second section we revisit
the construction of the functional model in [12] for the scalar wave-equation in
R 1 , paying a special attention to the calculation of the symmetric non-orthogonal
spectral representation. Then in the third section we study an abstract dissipa-
tive operator B with finite-dimensional defect dim (B - B+) < 00 1 and square
characteristic matrix-function. We obtain the corresponding eigenfunctions of the
lSee the remark after Theorem 3.2.

Spectral Meaning of Functional Model 165
self-adjoint dilation as limits of the resolvent in the properly rigged space. Though
the spectral components are, generally, non-uniquelly defined, we calculate them
with the help of the wave-operators. The characteristic function is interpreted
both as a stationary transmission coefficient and as a (non-stationary) limit, in
full agreement with the classical Adamyan-Arov results.
2. Symmetric representation for Lax-Phillips semigroup

Assuming that q is a piece-wise continuous non-negative locally bounded function
supported by the left semi-axis R_ = (-00,0), consider the wave-equation on
R = (-00,00)
Utt - Uxx + q(x)u = 0, -00 < x < 00, (2.1)
with proper initial data with finite energy:
u(x,O) = Uo, Ut(x,O) = UI.
There exists a unique generalized solution of the above equation from the do-
main of the quadratic form of the corresponding Schrodinger operator L, Lu =
-U xx + q(x)u, defined as a Friedrichs extension of the corresponding symmetric
operator defined on smooth compactly-supported functions. The energy of the so-
lution u(x, t) is conserved in course of evolution and is equal to the energy of the
11
initial data:
u ="2 -00
00
[
I \l2
uol + q(x)luol 2+ lUll ] dx.
The evolution operator of Cauchy data
( Uo ) ----+ ( u(x, t) ) = u(t)

ul Ut(x, t)
is unitary in the space of all Cauchy data u supplied with the energy dot-product:
(u, v) ="211 00
-00 [(\lU, \lV) + q(x)uv + UtVt] dx.
The generator of the evolution of Cauchy data is a self-adjoint operator C in
which appears in the right side of the wave-equation represented in Schrodinger
1 (0 -1)
form:
at = Z.
i au L O u := Cu.
The incoming and outgoing subspaces Din,out E consist of all Cauchy data of
incoming and outgoing waves u(x t) supported by the right semi-axis and are
mutually orthogonal in , see [7].
The spectrum of the operator L is absolutely continuous on the interval
R+ = [0, (0), possibly with varying multiplicity. We assume that the spectral
analysis of the Schrodinger operator L is done and a complete orthogonal set of
166 B. Pavlov
eigenfunction of the absolutely-continuous spectrum - the scattered waves 'l/J - is

constructed as
e- ikx + Se ikx if x>O
'l/J+(x, k) = {
a+x+(x, k) if x<O
and
eikx + Se- ikx if x>O
'l/J-(x, k) = { a_x_(x, k) if x<O
where a are corresponding transmission coefficients and X(x, k) = cp(x, k) +
m(k)(}(x, k) are the limit values (from the upper and lower half-planes) of Weyl-
solutions of the homogeneous equation Lu = k 2 u which are square integrable
on (-00,0] for k in upper or lower half-plane '2sk > 0, '2sk < 0 respectively. Here
cp(x, k), (}(x, k) are the standard solutions ofthe homogeneous equation with initial
data (0,1), (1,0) and m(k) are the corresponding Weyl-Titchmarsh functions,
see [19], selected with the condition of analyticity in upper or lower half-planes
respectively. The role of the spectral parameter is played by k 2 with the branch
of the square root Jk2 = k defined for S by the condition '2sk > 0 and for S by
'2sk < O.
Theorem 2.1. The reflection coefficients on the real axis '2sk = 0 are defined by the
limit values of the corresponding Weyl- Titchmarsh functions on real axis k from
the upper (lower) half-planes respectively:
S(k) = _ m+(k) + ik S(k) = _ m_(k) - ik

m+(k) - ik' m_(k) + ik
and admit an analytic continuation into the upper and lower half-planes respec-
tively. The incoming and outgoing solutions of the wave-equation on the positive
semi-axis 0 < x < 00 are parametrized by elements of the Hardy classes h E Hi
as
Uin,out(X, t) -
00
_1 -00
1
e ikt 'l/J(x, k)--:-kh(k)dk.
z
The incoming and outgoing subspaces Din, D out of the energy-normed space [. are
images of the corresponding Hardy-classes in spectral representations associated
with proper eigenfunctions
I]! (x k) = ( 'l/J(x, k) )
, 'l/J(x, k)
of the generator :
D.
In
= {jOO eikt (
-00
'l/J-(x, k) ) h (k)dk h E
'l/J-(x,k) - ,-
H2}
-,
Dout = {I: e ikt ( ~~(i~k~) ) h+(k)dp, h+ E H~ } (2.2)

and the correspondence between the functional parameters h and the relevant
Cauchy data is isometric in energy-normed space.
The invariant subspaces [ of the generator of the evolution developed from

V in , out are isometrical images of L2 (R) defined by the spectral map
:I : h ----> ~ J iJ!(x,p)h(p)dp = :Ih, , hE L2
[ = {J iJ!(x, k)h(k)dk, hE L2(R)} , 1:Ihl., = IhI L2
In particular, if f, g are elements of the energy-normed space presented as
f = :1+1+ + :1-1- := :I ( j~ ), g = :I+g+ + :I-g- :=:1 ( ~~ ) ,
then
Each of systems of eigenfunctions iJ! of the generator is non-complete, if the

modulo lSI of the Scattering matrix is not equal to one identically:
[ 1= [, [t = [> 1= 0, [~ = [< 1= 0,
but the joining of them {iJ!} is complete [+ + [_ = [ and the corresponding
Parseval identity (2.3) is true with proper spectral density matrix.
Proof. The proof of the statement is obtained by straightforward verification of the

corresponding algebra, see, for instance [13]. The only analytical question appears
when deriving the above Parseval identity (2.3). It is actually equivalent to the
following equations:
(:I+h, :1+ g) = (h, g) L2, (:I+h, :I-g) = (Sh, g)L 2
(2.4)
(:I-h, :I+g) = (Sh, g)L 2, (:I-h,:I-g) = (h, gh2'
We will verify only one of them assuming that h = h+ E H~ and 9 = eiktg_, g_ E
H'!... , t > O. Then the integration in the energy dot-product is extended on the
positive semi-axis 0 < x < 00 where the eigenfunctions 7/J are presented just as
linear combinations of exponentials. Then assuming that the elements h+, g_ are
smooth and rapidly decreasing we may accomplish the integration over the space
variable x, acquiring proper delta-functions. Then the integral over wave number
k is reduced, for any positive t, to:
(:I+h, :I-g) = .;f; 1 dx Jdk Jdk [_e- ikx + S(k)e ikx ] h+(k)
00
[e ikx _ S(k)C ikX ] e-iktL(k)
+ 4~ 1 JJ
00
dx dk dk [e- ikx + S(k)e ikx ] h+(k) [e ikX - S(k)e- ikX ] e-iktL(k).
168 B. Pavlov
The ultimate expression may be presented as an integral over the semi-axis 0 <
x < 00 and, after cancellation of few terms which do not contain S, results in:
2~1 JJ 00
dx dk dkei(k-k)xS(k)h+(k)lL(k)e- ikt
+2~ 1 J J 00
dke-i(k-k)xS(k)h+(k)JL(k)e- ikt
I: I:
dx dk
e- ikt [Sh+9-] dk = [Sh+e- ikt 9_] dk.
Note that the integral with negative t is equal to zero because of orthogonality of
incoming and outgoing subspaces.
The obtained result may be extended to the dense set of elements via con-
sidering the products h = eikt1h+, 9 = e ikt2 g_ for any finite t l , t2. Taking a limit
h, t2 -+ oo we obtain the statement
(.hh, J-g)E; = (Sh, g)
for any h, 9 E L 2(R), that is for any elements J+h E E+, J-g E E_ from the in-
variant subspaces of the generator , produced by the development of the incoming
and outgoing subspaces. Other announced statements may be proved in a similar
~ 0
Note that the column of "symmetric" coordinates ( ;~ ) of an element f =

J+f+ + J-f- in the symmetric spectral representation is not defined uniquely.
Nevertheless, one may suggest a natural procedure of recovering the symmetric
coordinates of an element (u, Ut) from the energy-normed space of Cauchy data
of the wave equation based on Adamyan-Arov wave-operators.
Really, consider the non-perturbed wave equation
Utt - u xx = 0
on the whole axis (-00,00). The corresponding unitary evolution group eiCot in
proper "non-perturbed" energy-normed space Eo has a common pair V out , in of
incoming and outgoing subspaces with the evolution group generated by the equa-
tion (2.1) in E. Denote by ..A.n, out the orthogonal projections in the energy-normed
space E onto incoming and outgoing subspaces Vin, out.
Theorem 2.2. The Adamyan-Arov wave operators, see [1]
W+ = s - lim e- iCot Poute-iCt, W_ = s - lim e- iCot ..A.ne-iCt
t---+CX) t---+-oo
exist as strong limits and are isometrical from the invariant subspaces Ein, out ob-
tained via development of Vin, out with perturbed evolution - into the correspond-
ing invariant subspaces V~n out of the unitary evolution group e iCot in the "non-
perturbed" energy-normed space Eo spanned by Cauchy data of incoming ''from the
right" and outgoing "to the right" waves
h( x t ) --
V
1
rn=
2rr
1 00
-00
i h in, out (k) dk 'ki h in,out:= F W 'f h( 0,
eik(tx) k ) (2.5)
where
h(O) = ( f(x,O) )
!t(x, 0)
I:
is the column of Cauchy data of the original perturbed problem and F
f(x) -7 ~ eikX dx : (Ff) (k)

is a standard Fourier transform in L 2
Corollary. The column of coordinates ( ~: ) of an element f = ..1-f- + ..1+f+
in the symmetric spectral representation may be chosen as ( ~:t ), where the
components fin,out := -ikFW'fF(O) are defined by the initial data F(O) = ( ~~ )
of the perturbed wave equation.
3. Self-adjoint dilation and symmetric model

Consider a dissipative operator in a Hilbert space K
i
L= A+ -r+r
2
with a real part A = A+ and a finite-dimensional 2 positive imaginary part.
Assuming that r K = E contains a generating subspace of the operator A, consider
the extended space . = L2 (R_ , E) EB K EB L2 (R+, E) of vector functions
and the operator.c in . defined on vector functions it, u E D(A), u E Wi(R, E)

submitted to the condition u_(O) - u+(O) = iru:
.cit = ( Au + r 2+ [~:~x~)
.~
+ u+(O)] ) .
Z dx
Without loss of generality we can assume that the operator r is self-adjoint r = r+

and positive, but still we will use both r, r+ in further formulae, if not specified
otherwise. The elements from the domain of .c, which vanish on either of semi-axes
2This condition can be essentially relaxed, see the remark after Theorem 3.2.
170 B. Pavlov
R, satisfy the corresponding homogeneous boundary conditions u_(O) = ifu,

or -u+(O) = ifu. These conditions are fulfilled for the absorbing and radiating
eigenvectors of the dilation, see (3.4) below.
Theorem 3.1. The operator is a self-adjoint operator in . Moreover, the com-
pression of the resolvent R>., <;SA < 0 of the operator onto the subspace K coin-
cides with the resolvent of the operator L:
(3.1)
Proof of the corresponding statement for the dissipative Schrodinger operator with
complex potential is given in [14]. Proof of the announced abstract result follows
the same pattern: to obtain the first statement we have to verify the symmetry
of and the symmetry of the adjoint operator; to prove the second statement we
may use a simple algebra and the basic fact of existence of limits of R-function on
the real axis from the upper half-plane. Then using Riesz-integral of the resolvent
one may derive from (3.1) that for any bounded analytic function (A) in upper
half-plane <;SA > -21'
P K ()PK =
-~ JCXJ-iE PK (A) [ - Ur 1 PKdA =

21fZ -CXJ-iE
-~ JCXJ-iE (A)[L - Alr1dA = (L)

21fZ -CXJ-iE
which means, in particular, that for t > 0: e iLt = P K e iCt P K . 0
The unitary group e iCt is a unitary dilation [10] of the contracting semigroup
e iLt and the operator is the self-adjoint dilation of the dissipative operator L.
The constructed dilation is minimal - i.e., it does not have proper self-adjoint
parts - if the subspace E is a generating subspace of the operator A.
We can construct the symmetric spectral representation for the original dis-
sipative operator following the pattern of the previous section. We begin with
description of eigenfunctions of the dilation.
The space of the dilation may be decomposed into orthogonal sum of
invariant subspaces generated by incoming and outgoing waves and corresponding
complementary ("radiating" and "absorbing" ) components = _ EB < = + EB
>. The spectrum of in each of components is absolutely continuous with
the constant multiplicity dim f on the whole real axis R. The spectrum of the
"absorbing" and "radiating" components >, < in subspaces >, < consists
of intervals of real axis where nonzero generalized solutions of the homogeneous
equation 'ljJ - A'ljJ = 0 exist which vanish on L2 (R+) (for 'ljJ or vanish on L2 (R_)
(for 'ljJ<). The corresponding eigen-functions of the dilation in each component
can be found, according to philosophy developed in [5, 4, 3] as elements of some
rigged space constructed with a help of some Hilbert-Schmidt operator T which
has a dense range, i.e., with all non-zero eigenvalues. This general statement can
be specified in our case by selection of a special class of eigenfunctions which

play a role of Scattered waves. This result can be obtained via selection of a
special rigging (i.e., the operator 7) correlated with the imaginary part of the
considered dissipative operator. Without loss of generality we can assume that the
operator r = r+ is a part of the positive Hilbert-Schmidt operator 7 acting in
K, (Ku,u) > O. Moreover we can assume that the operator 7 2 > 0 is presented
as a product of an operator of the trace class and an operator from Matsaev class,
so that its eigenvalues are sn(7 2 ) = O(a n f3n) with En lanl < 00, En f3n/n < 00
and f3n tend to zero monotonically. Consider the Gelfand triple [5] associated with
1 -1
the operator 7 as 7K = K1 eKe K = 7 K. Then the following statement
is true:
Theorem 3.2. The incoming and outgoing eigen-functions of the ditation , can be
presented as generalized solutions of the corresponding homogeneous equation with
exponential behavior in L 2 (R, E):
{ ,-<k", eE E, x E R_,
K1 ,
'I/J-(e) = u_(e)' in
e-tkxS+e, eE E, xER+,
{ ,-""8, eE E, x E R_,
K1 ,
'I/J+(e) = u+(e) , in (3.2)
e-tkxe, eE E, xER+.
These eigenfunctions are labelled by the "direction vectors" 3 e E E . The mid-
components u'f are generalized solutions of the non-homogeneous equation in com-
plex plane and are uniquely defined by the direction vectors e E E, see (3.4) below,
as 7- 1 images of strong limits of properly framed resolvent of the self-adjoint opera-
tor A or the resolvent of L, L+ on the real axis from the lower (upper) half-planes.
The transmission coefficients S, S+ are also uniquely defined from the homoge-
neous equation. In particular, S, S+ are analytic matrix-function in upper and
lower half-planes ':Sk > 0, ':Sk < 0
I 1- ir-I-r+
S+(k - iO) =I - i lim r---r+ = lim 2 A-AI
A-+k-iO L - AI A-+k-iO 1+ .!.r-I-r+'
2 A-AI
S (k + iO) = I + i A-+k+tO
lim. r L + I
-
AI r+ . (3.3)
_ 1 1
( ) --"2A-(k-iO) ( +( .)) _ I +
u_e I+S k-zO e--L-(k-iO/ e,eEE.
_
u+(e ) - -"21 A _ I
(k + iO) ( (
1+ S k
.)) _
+ zO e- -
I
L+ _ (k + iO/ + e, e E E. (3.4)
3The term is borrowed from [2J.

172 B. Pavlov
The eigenfunctions 'Ij.;>, 'Ij.;< of components of the dilation in complementary sub-

spaces e - = < and e + = > have a form:
(3.5)
when choosing vectors e>, e< as eigenvectors of operators Do> = I - S+S, Do < =
I - SS+ with non-zero eigenvalues 0>, 0< respectively, we obtain:
1
u>(e = 0> [u_(e - u+(S+e] ,
1
ue<) = 0< [u+(e<) - u_(Se<)] .
Proof. It is easy to verify the above formulae for the eigenfunctions on a formal
level, see for instance [14]. Note that analysis of the absolutely-continuous spectrum
of the symmetric model in the rigged space is presented in [17, 18, 15]. We suggest
below only the sketch of the proof of existence of the scattered wave of the dilation
(actually, the proof of existence of their mid-components) based on the Theorem 7
from [9]. It is proved in that theorem, in particular, that the non-tangential limits
exist on the real axis for the operator-valued R-function presented by the properly
framed resolvent of a self-adjoint operator. The remark attached to the theorem
shows that the statement remains true for the limit of the R-function
I
7 P K .c _ ).JPK 7 =
I
7 L+ -).J 7
from the upper half-plane>. --+ k + iO and for the adjoint function 7 L!.)..J 7 from
the lower half-plane, >. --+ k - iO. Now it is easy to verify the existence of the
mid-component u+ of the outgoing scattered wave. Really, the limit
lim
A-+k-iO
exists in the trace class, hence the mid-component can be presented as

-1 I 1
u+ = 7 lim 7 L+ _ >.I 7e E K
A--+k-iO
Similarly the mid-component of the scattered wave 'Ij.;_ can be obtained.The mid-
components u<, u> of the eigenfunctions in the complementary subspaces can
be obtained as linear combinations of them with proper coefficients and properly
chosen direction vectors. 0
Remark. One can see that the above calculation can be applied to the situation
when the imaginary part of the dissipative operator is a positive operator pre-
sented as a part in E of the positive operator 7 such that the square 7 2 of it is
product of an operator with a finite trace and an operator from Macaev class. This
is actually the natural class of dissipative operators for which the symmetric func-
tional model may be obtained by the procedure described above. This class can
be extended via considering the corresponding relative classes with the imaginary
part subordinated to the real part. It will be done elsewhere.
Based on the explicit formulae for the eigenfunctions one can prove that the
characteristic function obtained above as a stationary transmission coefficient can
be also interpreted in non-stationary terms.
Considering the non-perturbed shift generator in the space Din EB DOllt
L 2 (R-, E) EB L 2 (R+, E):
.d
La = Z-.
dx
Then the characteristic function of the original dissipative operator as Adamjan-
Arov scattering matrix, see [1] for the pair L, La is
s - lim Joe-iCotP+e2iI.:.tP_e-iCot.Jc;.
t-H)Q
Theorem 3.3. The Adamjan-Arov scattering matrix for the pair L, La coincides
with the transmission coefficient4 :
I
S+(k - iO) = 1- i lim r--r+ =
A--+k-iO L - ),,1
. I - 2:ir A-AI
I r+
hm . I ' (3.6)
A--+k-iO I + ~r A_AIr+
Proofis obtained by the straightforward calculation using the fact that the spectral
representation JO for the non-perturbed operator is defined by Fourier transform.
Hence the scattering matrix coincides with the transmission coefficient S in front
of the exponential e- ikx in the formula for the scattered wave 1/J- in the outgoing
subspace. 0
We construct now the symmetric functional model for the original operator
L based on eigenfunctions 1/J of its self-adjoint dilation, see (3.2).
Theorem 3.4. Consider the maps J of the spaces L2(E) into E:
J+h+ = rn=
1
V 21f
1 00
1/J+ (h+(p)) dp, h+ E L 2(E),
1
-00
J-h- = rn=
1
CXJ
1/J- (h-(p)) dp, h- E L 2(E),
v21f -CXJ
and the map J of the column ( ~~ ) := h into E:
Jh = J+h+ + J-h-.
4The numerator and denominator of the announced representation for the Scattering matrix are
commuting, so the order of them is not important.
174 B. Pavlov
Then the followin9 Parseval identity is true:

(.:Jf, .:Jg) = (1+,9+) + (S+ f-, 9+) + (SI+, 9-) + (1-,9-) =
J(( ~ SI+) f, g)E$E dk (3.7)
Proof. Note that for L E H:' (E)
with h_(x) = f:O e-ikxh_(k)dk non equal to zero identically if x E R_. Similarly
for h+ E H~(E)
:r ( h~ ~ ) ( hJX) )
with h+(x) = f~oo e-ikxh+(k)dk =I- 0 if x E R+. Hence incoming and outgo-
ing subspacesVin,out = L 2 (R) are mutually orthogonal. The invariant subspaces
tv 1
E of the dilation developed from the incoming and outgoing subspaces,
= 00
eikt'ljJ (h) dk, h E Hi,
t=-oo -00
are represented as
+ = .:J ( ~2 ) , L = .:J ( ~2 )
Then for 1+ E H~, 9_ E H:' we obtain:
(.:J ( ~ ) ,.:J ( eik~9_ ))
= 1-00
0
(Sf+,e
ikt
9-)L2 dx
for any finite t. Following the pattern of the previous section one may derive from
it that for any 1+,9- E L2
= JJ
2~ [0 00
dx dk dke-ikxeiicxS(k)1+9_
+ 2~ 1 J J 00
dx dk dke-ikxeikx f+s+(k)9-
+(u+(I+), U-(9-))K = (SI+, 9-)L2'

since (u+(!+),u-(g-))K = 0, and remaining integrals over semi-axes should be

combined to the delta-function 8(k - k). This way the announced statement is
verified for special elements
The proof is accomplished based on similar arguments for various choice of special
elements and linearity of the map 3. D
Consider the non-perturbed operator 0 = 0 EB 0 in L 2 (R, E) EB L 2 (R, E).
The corresponding evolution group Ut : u(x) -+ u(x - t) has unilateral invariant
subspaces L 2 (R_,E) := Din, L 2 (R+,E) := D out . We denote by P;.n,out the or-
thogonal projections onto Din ,out respectively. Similarly to the above reasoning in
section 2 we calculate the symmetric spectral representation via Arov-Adamyan
wave operators, see [1].
Theorem 3.S. The wave-operators

W_ =s- lim e-iCotp;'neiCt
t---t-OO
W+ = s - lim e- iCot PouteiCt

t--+CXJ
exist as strong limits and are isometric operators from the invariant subspaces
in, out C obtained by development of the incoming and outgoing subspaces
L 2 (R_, E) and L 2 (R+, E) with evolution generated by . The column
( fin) := f
fout
defines the symmetric spectral map as
3f = 3-fin + 3+fout
which is calculated from the column of Cauchy data as ( j~ ) = f(O) as
( fin) (FW_f(O))
fout - FW+f(O)
where F is the standard Fourier transform in L 2 :
f(x) -+ vk: J eikx f(x)dx = f(k).
Note that we suggested a unique recipe of construction of coordinates of

the symmetric spectral representation of the dilation, but, once constructed, the
column of coordinates may be a subject to change within proper limits caused
by possible presence of intervals where the scattering matrix is unitary, see the
discussion in [14].
176 B. Pavlov
Remark. Note that the eigenfunctions of the complementary component are found
uniquely, up to the parametrization with the direction vectors. Their mid-compo-
nents u<, u> <, > may serve a canonic system of eigenfunctions of the abso-
lutely continuous spectrum of the original dissipative operator and adjoint opera-
tor, respectively. The corresponding spectral expansion
u =
1
21f
1
au
IS(k)1 2 - 1 < >
S+(k) u (k)(u, u (k))dk, (3.8)
is converging for elements u represented as orthogonal projections of elements of

the complementary subspace < onto K. This set is dense in the absolutely con-
tinuous subspace of the operator L, see [10] and the detailed discussion of the
eigenfunction expansion of the dissipative Schrodinger operator with complex po-
tential in [14, 18]. Thus the incoming-outgoing eigenfunctions of the dilation and
eigenfunctions in the complementary subspaces <, > play essentially different
roles in spectral problem for the dissipative operator. The above formula (3.8)
shows that the problem of proper choice of the canonic system of eigenfunction
of the absolutely-continuous spectrum for dissipative operators is naturally re-
solved. Note that similar question about a canonic system of eigenfunctions of
absolutely continuous spectrum of a self-adjoint operator remains obscure. The
only bridge between the General Spectral Theorem for self-adjoint operators and
the expansion theorem is formed by classical results of I. Gelfand-A. Kostyuchenko
[3] on differentiation of the spectral measure of a self-adjoint operator in properly
rigged spaces. We hope to discuss the important question of the construction of
the canonic system of eigenvectors of the abstract self-adjoint operator somewhere.
For discussion of choice of the canonic system of eigenfunctions of the absolutely
continuous spectrum in case of spectral multiplicity one for a unitary operator and
a canonic system of eigenfunction of its contracting perturbation see [16].
Acknowledgement
The author is grateful to the referee for the suggestion to rewrite one of statements
in more formal manner (Theorem 3.2).
References
[1] V. Adamjan, V. Arov On unitary couplings of semi-unitary perators, Matematich-
eskie Isslenovanija, 1, 2, 1966, 3-64.
[2] N.I. Akhiezer, I.M. Glazman, Theory of linear operators in Hilbert space. Translated
from the Russian and with a preface by Merlynd Nestell. Reprint of the 1961 and
1963 translations. Bd. 1, 2 Dover Publications, Inc., New York, 1993.
[3] I.M. Gelfand, A.G. Kostyuchenko Expansion in eigenfunctions of differential opera-
tors In: Dokl. AN SSSR 103 (1955), 349-352.
[4] I.M. Gelfand Some questions of analysis and differential equations In: Uspekhi Mat.
nauk (in Russian) 14,3 (87), 1959,3-19.
[5] I.M. Gelfand, G.E. Shilov generalized functions Volume 3: Theory of differential
equations, 'franslated from Russian by Menhard E. Mayer, Academic Press [Harcourt
Brace Iovanovich Publisher] New-York-London, 1967 [1977]
[6] S. Khruschev, N. Nikolskij A functional model and some problems of spectral theory
of functions, 'frudy Mat. Inst. Steklov, 176, 1987. 97-210.
[7] P. Lax, R. Phillips Scattering theory Academic press, New York, 1967.
[8] S. N aboko A functional model of perturbation theory and its application to scattering
theory, Proceedings of Steklov Institute, 2, 1981, 85-116.
[9] S.N. Naboko Non-tangential boundary values of operator-valued R-functions in a
half-plane, Leningrad Math. Journal, Vol. 1,5, 1990, 1255-1277.
[10] B. Sz.-Nagy, C. Foias Analyse harmonique des operateurs de l'espace de Hilbert,
Academiai Kiado, Budapest, 1970.
[11] N. Nikolski, Operators, Functions and Systems: An Easy Reading, AMS, 2002, Vol
2.
[12] B. Pavlov, The continuous spectrum of resonances on a non-physical sheet Dokl. AN
USSR, 206, 1972, 1301-1304.
[13] B. Pavlov, On conditions of separability of the spectral componentrs of a dissipative
operators, Izv. AN USSR, 39, 1975, 123-148.
[14] B. Pavlov, Self-adjoint dilation of the dissipative Schroedinger operator and its res-
olution in terms of eigenfunctions, Math. USSR Sbornik, v31, 4, 457-478 (1977).
[15] B. Pavlov, Spectral Analusis of a Dissipative Schrodinger operator in terms of a
functional model, in the book: "Partial Differential Equations" , ed. by M. Shubin in
the series Encycl. Math. Sciences, 65, Springer 1995, 87-153.
[16] B.Pavlov, M.Faddeev, Spectral analysis of unitary perturbations of contractions,
Proc. LOMI, v115,1982. (English translation in: J. of Sov. Math. vol. 28, 768-776
(1985) )
[17] V.A. Ryzhov Construction of the functional model of non-self-adjoint operator with
nonempty regular set, Dep. VINITI 21.12.98, 389-B90, St. Petersburg (1990), 92
pages.
[18] V.A. Ryzhov Absolutely-continuous subspace of the hon-self-adjoint operator and
Scattering theory, PhD Thesis, supervised by S. Naboko, St. Petersburg 1994, 147
pages.
[19] E.c. Titchmarsh Eigenfunctions expansions associated with second-order differential
equations, Partl, Clarendon Press, Oxford(1946) 247 pages.
[20] P.Lax, R. Phillips Scattering Theory Academic press, New-York - London,1967.
Boris Pavlov
Department of Mathematics University of Auckland
Private Bag 92019
Auckland, New Zealand
e-mail: pavlov@math.auckland.ac.nz
Operator Theory:
A Remark on Equivalence of Weak and Strong

Definitions of the Absolutely Continuous
Subspace for N onself-adjoint Operators
Roman Romanov
Abstract. We prove the equivalence of weak and strong definitions of the

absolutely continuous subspace for nonself-adjoint dissipative operators.
Mathematics Subject Classification (2000). 47B44.
Keywords. absolutely continuous spectrum, dissipative operators.
In this note we analyze two known natural definitions of the absolutely continuous
(a.c.) subspace for an abstract nonself-adjoint operator. So far, their equivalence
has been proved [1] under certain strong assumptions about boundary behavior of
the resolvent. Here we establish the equivalence without any additional assump-
tions in the case of dissipative operators.
Throughout the paper, L is a closed operator in a Hilbert space H such
that (J"ess(L) c JR. In particular, the resolvent of L is defined for all z . JR ex-
cept for at most count ably many points as a bounded operator in H. The vector
Hardy classes H~ are the collections of analytic functions f : C --t H satisfying
suP,,>o JRIlf(k ic)112 dk < 00, respectively.
Definition 1. The subspace
de! -
H:;;)L) = clos H~(L),
-
H~(L) ~
d! {UE H: (i) (L - Z)-l u is analytic in C \ JR }
(ii) ((L-z)-lU,v)lc EHiforallvEH '
is called the weak a.c. subspace of the operator L. Elements of the linear set H~(L)
are called weak smooth vectors of L.
This work was supported in part by EPSRC Grant GR/ R20885 and by RFBR Grant 03-01-
00090.
180 R. Romanov
The weak a.c. subspace has first been introduced and studied by A. Tikhonov
[2]. In the case of a self-adjoint L, it is well known, see, e.g., [3], that the definition
above agrees with the standard one.
Another, historically first, definition of the a.c. subspace suggested by
L. Sakhnovich [4] initially referred to the case when L is dissipative. A conve-
nient equivalent formulation and an extension of it to general nonself-adjoint case
was given in [5]. For clarity, we first restrict our consideration to the situation of
the perturbation theory and discuss the general case afterwards. Namely, except
for two remarks following Theorem 2 it is assumed throughout that
The operator L is a completely nonself-adjoint operator 1 of the form L =
A+iV, A = A*, V = V*, V(L): = V(A) c V(V), and V is A-bounded with
a relative bound less than 1, that is, IIVul12 :'S a IIAul1 2 + b Ilu11 2, a < 1, for
all u E V(A).
Definition 2. The subspace

de! -
Hac(L) = clos Hac(L),
uEH : (i) (L - z)-1 u is analytic in <C \ lR}
Hac(L) d;J {
(ii) 1V11/2 (L - z)-1 ul c
E H1 '
is called the (strong) a.c. subspace of the operator L. Elements of the linear set
Hac(L) are called (strong) smooth vectors of L.
In the dissipative case the subspace Hac (L) coincides with the invariant sub-
space corresponding to the canonical factorization of the characteristic function of
the operator L in the sense of the Szokefalvi-Nagy-Foias functional model [5, 6].
It is easy to see that (L - A)-1 H::c c H::", (L - A)-1 Hac C Hac for all
A E p(L). An important property of strong smooth vectors is expressed by the
following
Proposition [5]. There exists a Hilbert space N, an a.c. self-adjoint operator Ao
in N, and a bounded operator P: N ---+ H, RanP = Hac, such that for all g EN
and z ~ lR, Z E p( L)
(L - Z)-1 Pg = P(A o - z)-1 g.
Corollary 1. H::c ( L) :J Hac ( L ) .
Proof. Let g E N satisfy driq E LOO(lR) where the measure dJ1g is the matrix
element of the spectral measure of Ao on the vector g (an LOO-vector of Ao), and
let u = Pg. The spectral theorem for operator Ao gives
\(L-Z)-1 U ,V)=\(Ao -z)-1 g ,p*v)= Jk~zp(k)dk,

IThat is, L has no reducing subspaces on which it induces a self-adjoint operator.
On Equivalence of Definitions 181
where p E L2, by the choice of g. This shows that the restrictions of the left-hand
side are in H~ in the respective half-planes for all v, that is, U E H;:C. The result
follows since the set of LCXJ-vectors is dense in N by absolute continuity of Ao. 0
Theorem 2. If L is dissipative (V :::: 0), then H::;J L) = Hac (L ) .

Using a functional model for nonself-adjoint operators due to Naboko [5],
Ryzhov has proved [1] the equality H;:C = Hac assuming that the characteristic
function, 8(z), of L has weak non-tangential boundary values, 8(k iO), a.e. on
the real axis. In the dissipative case, this condition is equivalent to the requirement
that the inverse characteristic function has weak boundary values from above. It
appears to have not been noticed so far that in the dissipative case the inclusion
H;:C(L) c Hac(L) holds unconditionally. Our proof of it is elementary and is based
on the following simple property of weak smooth vectors.
Lemma 3. Let B be a closed operator such that uess(B) c lR. Then
2
sup IImzlll(B - Z)-l wl1 < 00 (1)
zEIC
for any w E H;:C(B).
Proof. From the Riesz integral representation and Schwartz inequality we have
(signs refer to z E C, respectively)
I((B-z) w,v)1 =
-1 If ~J-L(k)dk::;
1
z
I
11J-LII2(lR) Jr1/2
IImzl
1/2'
where J-L E L2(JR) satisfy 1IJ-LIIL2 ::; C w Ilvll in view of the uniform bounded ness
principle. By arbitrariness of v, (1) follows. 0
Proof of the theorem. We have to show that H:c(L) c Hac(L). Define

U = (L - zO)-l W, wE H;:C, Zo E C_, then U E H;:C.
Let us integrate the identity (z = k + ic)
1: 1:
in k from -N to N,
IIV1/2 (L - uI12 dk = c II(L - Z)-l uI12 dk

1:
Z)-l (2)
-1m ((L-z)-l u ,u)dk.

182 R. Romanov
Consider first the case when c > O. The first term in the r.h.s., denoted by (1), is
estimated as follows
(1) = c jN II(L - z)-l (L - ZO)-l wl1 2 dk :::; jN _1 2 c II(L _ z)-l wl1 2dk
-N -N Iz zol
+cll(L_Zo)- l W2jN
I1
- N
dk
1z - Zo 1
2 :::;CW1
c-
17mZo :::;C
c 1 w
Here we have taken (1) into account. We conclude that (I) is bounded uniformly
I: \ I:
in Nand c. In a similar way, we have
(L - Z)-l u, u) dk = z ~ Zo (((L - Z)-l W, u) - (u, u)) dk
_jN _l_/(L Z)-l W, u) dk -In -N

N + i~ - Zo Ilu11 2.
-N z- Zo \ + Zc - Zo
When N -+ 00, the first term in the right-hand side vanishes, the second tends to
-i7rlluI1 2, hence the left-hand side has a finite limit, and this limit is independent of
00 of the left-hand side in (2) exists
c. Combining these, we find that the limit N -+
and is bounded in c. This and a similar consideration for the case 2 c < 0 yield
that u E Hac. It remains to notice that VzoEL (L - zo)- H;:C = H::!c because
-- 1--
iT (L + iT)-l ~ I, T -+ +00. (3)

The latter easily follows from dissipativity of L [7]. 0
Remark. Let L be an arbitrary maximal dissipative completely nonself-adjoint op-
erator. The definition of the strong a.c. subspace of L is obtained [1] by substituting
the condition (ii) in definition 2 with the following one (see (2))
~~~~ (cll(L_k_ic)-luI12 -1m \(L-k-ic)-l U,U))dk<OO.

With this definition, Theorem 2 holds in the general case, with the same proof.
Remark. The proof of Theorem 2 shows that the linear sets of smooth and weak
smooth vectors of a maximal dissipative operator L satisfy (L - zo) -1 H;:C (L) c
Hac(L) for all non-real Zo E p(L).
In the situation of the perturbation theory, the following simple sufficient
condition of triviality of the subspace H::!c (L) is useful.
Proposition 4. H:;c(L) = {O} if for a. e. k E IR we have (z = k + ic)
D(z) == y'c (L* - Z)-l 1V11/2 ~ 0 (4)
as c 1 o.
2Because of the dissipativity of L, it actually suffices to consider the case c > 0 only, for the
function V 1 / 2 (L - z)-l u restricted to IC_ is in H=- for all u E H [5]. In order to keep the
elementary nature of the proof, we prefer not to use this fact, since the proof of it in [5] is based
on the existence of a self-adjoint dilation of L.
On Equivalence of Definitions 183
In the case when L is dissipative the condition (4) is also necessary, see [8]. It
was used in [9] to establish triviality of the a.c. subspace for dissipative Schrodinger
operators with slowly decaying imaginary part of the potential.
Proof. Given awE H:i:c and v E D(V), define the functions F(z) to be the
restrictions of the function ((L -
Z)-1 W, 1V11/2 v)
to C, respectively. We shall
show that if (4) is satisfied, then F vanish identically for all v E H. Let us
first derive the result from this assertion. We have, (w, (L* - z)-11V1 1/ 2 v) =
((L - z)-1 w, 1V11/2 v) = 0 for all non-real z E p(L) and v E D(V). On the other
hand, the closed linear hull of the linear sets (L* - Z)-1 Ran V, z ~ JR, z E p(L),
coincides with H, since L is a completely nonself-adjoint operator [5]. Hence,
w = 0, as required.
Since F E Hl by the choice of w, it suffices to show that the boundary
values F+(k) = F_(k) for a.e. k E JR. Indeed, we have (z = k + iE, E > 0),
F+(z) - F_(z) = 2iE ((L - z)-1 (L - z)-1 w, 1V11/2 v)

= 2i (....rc (L - z) -1 w, D(z )v) ----40
010
for a.e. k E R D
References
[1] V.A. Ryzhov, On a singular and an absolutely continuous subspace of a non-self-
adjoint operator whose characteristic function has boundary values on the real axis.
Funktsional. Anal. i Prilozhen. 32 (1998), no. 3, 83-87 (Russian); English transl. in:
Funct. Anal. Appl. 32 (1999), No.3, 208-212.
[2] A.S. Tikhonov, Absolutely continuous spectrum of a linear operator and problems of
scattering theory and factorization of operator-functions. Ph. D. Thesis, Simferopol
State University, Simferopol (1989) (in Russian).
[3] M. Reed and B. Simon, "Methods of Modern Mathematical Physics" , v. 4, Academic
Press, New York, 1978.
[4] L.A. Sahnovic, Dissipative operators with absolutely continuous spectrum, Trans.
Moscow Math. Soc. 19 (1968).
[5] S.N. Naboko, A functional model of perturbation theory and its applications to
scattering theory, Trudy MIAN 147 (1980), 86-114 (Russian); English transl. in:
Proc. Steklov Inst. Math. (1981), No.2, 85-116.
[6] B. Sz.-Nagy and C. Foias, "Analyse Harmonique des Operateurs de I'Espase de
Hilbert", Masson et C ie / Academiai Kiado, 1967.
[7] E. Hille and R. Phillips, "Functional analysis and semi-groups", AMS Colloquium
Publications, Vol. XXXI, AMS, Providence, R.I., 1974.
184 R. Romanov
[8] B.S. Pavlov, Self-adjoint dilation of the dissipative Schrodinger operator and its
resolution in terms of eigenfunctions, Math. USSR Sbornik 31 (1977), No.4, 457-
478.
[9] R. Romanov, On instability of the absolutely continuous spectrum of dissipative
Schrodinger operators and Jacobi matrices, submitted.
Roman Romanov
School of Computer Science
Cardiff University, Cardiff
Queen's Buildings, PO Box 916
Newport Road, Cardiff CF24 3XF
UK
and
Laboratory of Quantum Networks
Institute for Physics
Saint Petersburg State University
198904 St. Petersburg
Russia
e-mail: roma@rvr.stud.pu.ru.R.V.Romanov@cs.cf.ac . uk
Operator Theory:
On a Transformation of the Sturm-Liouville

Equation with Slowly Decaying Potentials
and the Titchmarsh-Weyl m-function
Alexei Rybkin
Dedicated to the memory of F. Atkinson
Abstract. We put forward a new transformation of the half-line Sturm-Liou-

ville equation with non-smooth potentials from Lp with p 2: 2. This trans-
formation yields existence of the Weyl solution with higher order WKB-type
asymptotic behavior (spatial and spectral parameter). We apply our approach
to the study of high-energy asymptotics for the Titchmarsh-Weyl m-function,
improving on some relevant results of others.
Mathematics Subject Classification (2000). 34EIO, 34E20, 34B20, 34L40.

Keywords. Sturm-Liouville problem, Schrodinger operator, long-range poten-
tials, WKB-type asymptotics, Titchmarsh-Weyl m-function.
1. Introduction
We will be concerned with the Sturm-Liouville equation
-u" + q (x) u = A2 u for x E lR+ := (0, (0) (1.1 )
with real potentials q essentially bounded on lR+. With applications to the spectral
analysis in mind we will be especially interested in the behavior of the solutions
to (1.1) at infinity. By variation of parameters (1.1) formally transforms to the
integral equation
y(X,A) = 1 + 1 00
K(x,s,A)y(s,A)ds, (1.2)
where y := e-i)\X u and

e- 2iA (s-x) - 1
K(x,s,A):= 2iA q(s). (1.3)
186 A. Rybkin
If q E L1 (lR.+) then (1.2) can be solved by iteration and hence the original equation
(1.1) has the unique solution u (x,'\) such that
u (x , ,\) '" e i>.x , X ---+ 00 , (1.4)
for all ,\ E R Such a solution is referred to as Jost and its existence is the main
feature of the short-range scattering. Unfortunately if we go beyond the condition
q E L1 (lR.+) the transformation from (1.1) to (1.2) becomes singular and in general
one has to look for some other transformations. Eq. (1.1) with non-smooth poten-
tials from Lp (lR.+) with some p < 2 was the main object of the recent research
due to Christ-Kiselev (see, e.g., [7]). They found a sequence of transformations for
(1.1) that, combined with some subtle results on almost everywhere convergence
of certain integral, yields a series solution u (x,'\) to (1.1), which is absolutely
convergent for a.e. real ,\ and has the WKB-type asymptotic behavior
u (x,'\) '" exp { i'\x + 2~'\ fox q (s) dS} ,x ---+ 00, (1.5)
for a.e. ,\ E R By the Gilbert-Pearson subordinacy theory [12] the existence of

asymptotics (1.5) implies that the absolutely continuous spectrum of the Schrodin-
ger operators defined by (1.1) is lR.+. Note that if q is merely L2 then it is still an
open question if (1.5) holds for a.e. ,\ E lR.. In the general case of q E Lp (lR.+) ,p > 2,
we can only claim that lR.+ is the essential spectrum and hence solutions with
behavior (1.5) for real ,\ need not exist.
Of course, if we assume some smoothness of q then the situation may con-
siderably improve. For instance, the Green-Liouville transformation of (1.1) (see,
e.g., [25])
(1.6)
where y = {,\2 - q (x)} 1/4 u, may come into play. Thansformation (1.1) ::::} (1.6) is
a crucial ingredient in the WKB-analysis and serves a very wide range of potentials
(even growing at infinity) but requires the existence of q' and q" which appears to
be too much in the setting of slowly decaying non-smooth and random potentials.
Even for smooth potentials like q (x) = x-a sinx,i3, 0< a ~ (3 ~ 1, the transforma-
tion is not of much help since q' and q" unboundedly oscillate at 00. Note, that,
as it was shown by Buslaev-Matveev [5], the Green-Liouville transformation (1.6)
works for slowly decaying potentials subject to
Iq(n) (x)1 ~ Cx- a - n , a> 0, n = 0,1,2. (1.7)
In the present paper we are going to deal with the case of potentials q E
Lp (lR.+) , p 2: 2, without any smoothness assumptions. We put forward a transfor-
mation of the original equation (1.1) that yields solutions with higher-order WKB-
type asymptotic for every complex'\. We then use it to study the Titchmarsh-Weyl
m-function. We leave the background information and a discussion of the intensive
literature for Section 4.
WKB asymptotics 187
Notation. L p ,1 ::::: p ::::: 00, as usual, stands for the Lebesgue class of functions f
with the finite norm 1
Ilfll~:= l l f (xW dx,p < 00; IlfllCXl := ess- sup {If (x)1 ,x E IR}.
Cn,n E N, is a set of functions on IR such that Ilf(j)IICXl < oo,j = 0,1, ... ,no We
write f E Xo if f E X and lim f (x) = O. The class lp (L 1 ) ,1 ::::: p ::::: 00, with
X----+C'XJ
the norm
Ilfllf (L
p 1) := n~CXl (i n
+1If (x)1 dX) P, P< 00;
IlfI1 100 (L , ) := sup {i n
+1 lf (x)1 dx, n E IR}.
We write f E Co (Ld if f E lCXl (Ld and lim .r,n+1

n-----+CX) n
If (x)1 dx = o. It is clear that
h (Ld = L 1, Lp + L1 c lp (L 1) but Lp + L1 i- lp (Ld. With compactness of our
exposition in mind and whenever it leads to no confusion, we write
2. A Fourier-type transform and a Riemann-Lebesgue-type lemma

Let f be a measurable function on IR+ which class will be specified later. Fixed
>. E C+ consider the following Fourier-type transform
(2.1)
For every>. E C+, the decaying exponential under the integral sign in (2.1) makes
the "transformed function" [(x, >.) well defined for a broad class of functions.
Lemma 1. If f E lp (L1) ,1 ::::: p ::::: 00. Then
II[e>.)t::::: 2(1 + I~>') IlfI11p(L,) (2.2)
Proof. Assume p E [1,00). By Jensen's inequality
I[(x,>.) I Linrn+1 eiASf(s+x)ds

p
=
p
n~O n
1 If a function f is initially defined only on a subset of IR we continue it to the whole IR as zero.

188
r- ~
A. Rybkin
< (~e-ImAn I
e- ImAn ([+1 If (s +x)lds),
= (1 _e~ ImA ) p-1 ~ e- ImAn ( i n+1 If (s + x)1 dS) P
Integrating (2.3) with respect to x, one has
111(, A)II: ~ (1 + I~A) p-1 ~ e- lmAn l (in+1If (s + x)1 dS) P dx. (2.4)
Observe now that
r(I
JIR
n 1
n + If(s+x)lds )P dx= ~ l m
m+1 (l
x
x
)P dx
+1 If I
~ ~ ( lm +
m 2
If I
)P ~ 2P- 1 ~ { ( lmm+1 If I)P + (lm+2
m+1 If I
)P}
(2.5)
Plugging (2.5) into (2.4), one has
111(-. A) liP ~
P
(1 + I~A)P-l L n~O
e- lmAn . 2PIlfllfp(L 1)
~ P(1 + I~A)P Ilfllfp(LI)

2
that yields (2.2). Similarly one proves (2.2) for p = 00. D
Given a function ip (x) ~ 0, x E 1R+, set A (ip) := {>. : 1m A ~ ip (lAin. The

following statement will play an important role in our consideration.
Proposition 1. Let f E C 1 + Co (L1) then
(2.6)
where ip is monotonically decreasing function (depending on f) such that
x-oo
lim ip (x) = O.
The function ip will be specified in the proof
Proof. By Lemma 2 every function f E C 1 + Co (Ld can be represented as
f = h + 12, where h E C 1 and IlhI11oo(h) < c. (2.7)

WKB asymptotics 189
Consider hand h separately. Integrating (2.1) by parts and taking into account
that h E C l yield
h (x, >.) = - i~ h (x) - i~ LX) ei)..s f{ (8 + x) d8 (2.8)
and hence
Ilf~1(., >.)11 00 -< Ilhlloo + Ilf{ll oo
1>'1 1>'1
1 0
00
e
-Im)..sd
8
= Ilh 1100
1>'1
Ilf{ 1100 <
+ 1>'1 1m>. -
(~
1>'1
1)
+ 1>'1 1m>.
II
h
II
C"
(2.9)
For 12 we have:
LiT)r+ Le-1m)..n inr+

l l
= e-1m)..s 112 (s + x)1 ds ~ 112 (s + x)1 ds (2.10)
n~O n n~O n
~ 1- e~Im).. Ilhll l =(L , ) < (1 + l~>') E.
Combining (2.9) and (2.10) , one has
11 1 (-' >.)1100 ~ C~I + 1>'1 :m>.) Ilhlb + ( 1 + l~>') E. (2.11)

Setting Al := {>.: 1m>. 2: I} and letting in (2.11) 1>'1----+ 00,>' E AI, we obtain that
for any 10 > 0
lim 111(, >.)1100 ~ 210.
That is
lim 111 (-, >.) 1100 = 0, 1>'1 ----+ 00, >. E AI. (2.12)
Consider now the case when 1m>. ----+ O. Assuming 1m>. < 1, (2.11) implies
111(, >.) 1100 ~ l~ >. {lml~~ 1 Ilhlb + (1 + 1m >.) 10 } (2.13)
~ l~>' C~lllhlb +10)

Define now a function 9 (t) , t > 0, as follows
g(t):= infinf(C l llhllc1
10>0
+10),
where the first inf is taken over all representations (2.7) with fixed 10 > O. It is a
monotonically decreasing function and 9 (t) ----+ 0, t ----+ 00. Take now any continuous
non-negative function if such that if (t) ----+ 0, t ----+ 00, and
g(t) = O(if(t)),t ----+ 00,
190 A. Rybkin
(e.g., cP = y'g) and set A2 (cp) := p.: cP (IAI) :::; ImA < I}. It follows then from
(2.13) that
and hence
(2.14)
Combining (2.12) and (2.14) proves the proposition with A (cp) = A1 U A2 (cp). D
Remark 1. The standard Riemann-Lebesgue Lemma follows from the proof of
Proposition 1. Indeed, every f E 1 admits representation (2.7) with h E CO'
and 12 such that 1112111 < c. Estimate (2.11) for real A can be improved to read
Ij(x, A) I :::; IAI- 1(Ih (x)1 + Ilh 111) + c, VA, 1m A = O. (2.15)

It follows from (2.1) that (1m A = 0)
where j(A) is the usual Fourier transform and (2.15) implies the Riemann-Le-
besgue Lemma.
Remark 2. If we suppose that f E lp (1) ,p > 1, then the conclusion of Proposi-

tion 1 can be improved to read
(2.16)
with some monotonic function CPP (x) ----t O,x ----t 00. Moreover, if P1 < P2 then
CPPI < -
_ CPP2 an d 11m CPPI
-( (x)
-) = O . 1ar, CPP <
,zn part zcu < h . .
_ CPoo _ cP were cP zs as zn
x-+oo CPP2 X
Proposition 1.
Remark 3. If f E C n then similarly to (2.9) one has
IIF (., A)lloo :::; (lAin ImA)-lllfll c n
and (2.6) holds with cP (t) = 0 (1/t n - 1) ,t ----t 00.
3. A transformation of the original equation

Given potential q (x) and fixed A E C+ consider the following chain of transfor-
-1=
mations
q1 (x, A) = e2iAS q (8 + x) d8 = -q(x, 2A). (3.1)
WKB asymptotics 191
qn+l (x, >.) = 1 e;

00
(x, s, >.) q; (s + x, >.) ds, n E N (3.2)
en (x, s, >.) := exp {i>'S + I 1;

s x
+ qm (t, >.) dt} ,
en (s, >.) := en (0, s, >.) .

This way we get the sequence of functions {qn (., >.)} n> 1 which can be viewed as
"momentum dependent" transformations of the original potential q. Note that for
n ?: 2 these transformations are highly nonlinear and were previously considered
by many authors (see, e.g., [15], [17], [18]). The main feature of the transformation
qn (., >.) -> qn+l (., >.) is that in a way it improves the rate of decay at infinity.
The following theorem is our main result.
Theorem 1. Let q E lp (Ld ,p = 2n with some n E N, then for any>. from
A:= {>. E C+ : 111 00

e2iAS q (s + x) dslloo < ~ 1m>.}
the equation
-u" + q (x) u = >.2U, x E IR+, (3.3)
can be transformed into the Volterra type integral equation:
y (x, >.) = 1+ 1 00
Kn (x, s, >.) y (s, >.) ds, (3.4)
u (x, >.) = en (x, >.) y (x, >.) ,

with the kernel
Kn (x, s, >.) := (qnen)2 (s, >.) I e;;-2

S
(t, >.) dt. (3.5)
The kernel Kn (x, s, >.) satisfies the bound (>. E A)
1 00
IKn (x, s, >')1 ds ~ (I~>') p-l (2 + I~>') p IlqXxllfp(LIl' (3.6)
where Xx is the characteristic function of (x, (0), and therefore equation (3.4) is
solvable by iteration.
Proof. Observe that for any n E N
en (x, >.) = exp {i>'X + fox fl qm (s, >.) dS}

is a solution to
(3.7)
for any x?: 0 and>' E C+. Indeed, assume that (3.7) holds for n -1,n?: 2. Then
(3.8)
192 A. Rybkin
Note that qn is differentiable and

(3.9)
where
An (x, A) := iAX + rt
10 m=l
qm (s, A) ds.
Differentiating (3.8) twice and using (3.9) one has
e~ = {e~-l + 2e~_lqn + (q~ + q~) en-I} exp fox qn

= (q~-l + q - A2 + 2An- 1qn + q~ - q~-l - 2An- 1qn) en
=(q+q~-A2)en,
which is (3.7) for n 2: 2. By a straightforward computation one verifies (3.7) for

n = 1 and by induction (3.7) holds for any n E N.
If U1 (x, A) = en (x, A) is a solution to (3.7) then as the other solution we
choose
U2 (x, A) = en (x, A) lax e;;-2 (s, A) ds.
Rewriting the original equation (3.3) as
-u" + (q (x) + q~ (x, A)) u - A2 U = q~ (x, A) U (3.10)
and by variation of parameters one easily verifies (formally) that (3.10) implies
(3.4). To justify this formal computation it is enough to show that Kn (x, s, A) E
L1 (lR+, ds) for every x E lR+, A E A and, by choosing x large enough, we can
achieve 11Kn (x, ., A) 111 < 1. Let us prove first that for any p 2: 1 and natural m
qm E L 2p ===} qm+1 E Lp. (3.11)
Since lp (Ld c Co (L1) , by Proposition 1 we can achieve
15:= I~A IIq1 (,A)ll oo = I~A 11(f(,2A)11 00 < ~

and by Lemma 3 of Appendix
and hence it follows from (3.2) that
18m +, (x, s, -X) I' :5 exp { -2lm.ls (1- 1~.1 t, IIq, (-, .I) "= ) } (3.12)
::; exp { - 2 (1 - 215) 1m AS} < exp ( - 1m AS) .

WKB asymptotics 193
By Jensen's inequality then

(3.13)
:; 1 00
18 m+1 (x, s, A)1 2 1qm (s + x, A)1 2p ds (1 00
18 m+ 1 (x, s, A)12 dS) p-1
::; (1m A)-P+1 1 00

exp (- 1m AS) Iqm (s + x, A) 12P ds.
Integrating (3.13) with respect to x, one has
Ilqm+1 (', A) II~ ::; C~ A y-1 1 00

exp (- 1m AS) (1 00
Iqm (x, A) 12p dX) ds
::; (I~ A) p Ilqm (-, A) II;~

and finally
1 2
Ilqm+1 (', A)llp ::; ImA Ilqm (', A)112p ,p 2: 1, n E 1'1, (3.14)
that proves (3.11). By induction, (3.14) yields
2
Ilqn(-,A)11 2 ::;
(1ImA
)P-2
Ilqd,A)II~, p=2 n ,
that is
(3.15)
By (3.15), q;, (-,A) ELI and it is only left to demonstrate that 8;' (s,AH: 8;;2 (t,A) dt
is bounded. Indeed,
i8~ (s, A) is 8~2 (t, A) dti = lis exp2 {iA (s - t) + 1~

s
qm (', A) } dtl
: ; 11 S
-
X
exp2 { - 1m At (1 - I~ A t 1 Ilqm (-, A) 1100 ) }
::; 1o
00
exp (- 1m At) dt = -.
1
1m A
Estimate now L oo IKn (x, s, A)I ds. It immediately follows from (3.1) and (3.2) that
q1 (x, A) = -qxx (x, 2A), (qXx)n (x, A) = qn (x, A),
194 A. Rybkin
1= 1= 1= I(qXs)~
and hence
x IKn (x, s, A) Ids ::; 1m1 A x Iq; (s, A) Ids = 1m1 A x (s, A) Ids
: ; C~ y-l ( + I~ Y Ilfp(L') .
A 2 A IlqXx
that proves (3.6). The theorem is proven. o

Remark 4. The domain A c <C+ but it follows from Proposition 1 that
lim dist (A, k) = 0,
k--->=
that is A approaches the real line.
Remark 5. The conditions of Theorem 1 are clearly not optimal. Using arguments
of Proposition 1 one can achieve the contractive property of the integral operator
in (3.4) for any 1m A> 0 and x > 0 by choosing IAI large enough.
Remark 6. We do not assume potentials q in Theorem 1 real.
Corollary 1. Under conditions of Theorem 1, equation (3.3) has the Weyl solution
IlT (that is IlT (x, A) E L2 (lR+) , A E <C+) satisfying the following asymptotics
IlT (x, A) '" exp {iAX +l

x
tl qm (s, A) dS} ,x 00, ---+ (3.16)
for every A E <C+. This immediately follows from Theorem 1 if A E A. However

with some additional effort (see Remark 5) (3.16) can be proven as stated.
Remark 7. The structure of {qn} is very messy. One can prove though that if
A ---+00
in some sector in the upper half-plane then
ql (x, A) '" (2iA)-1 q (x), qn (x, A) rv - (2iA)-p+l qP/2 (x) ,p = 2n.
If we call exp {iAX + 2~A J; q (s) dS} the WKB-type asymptotics then (3.16) can
be referred as to higher-order WKB-type asymptotics.
Remark 8. It is natural to ask when (3.16) holds for real A. As we have already
mentioned in Introduction, in general there is no answer to this question even
for q E L 2 . The best-known result here belongs to Christ-Kiselev [7] (see also the
extensive literature cited therein) saying that if q E lp (Ld ,p < 2, then equation
(3.3) has a bounded solution u (x, A) for almost all A E lR with the WKB-type
asymptotics
u (x, A) '" exp {iAX + 2~Alx q (s) dS} ,x ---+ 00. (3.17)
On the other hand it is proven that under the only condition q E Lp,p > 2, the
spectrum of H = -d 2/dx 2 + q(x) ,u(O) = 0,2 may be purely singular. In such
20r any other self-adjoint boundary condition

WKB asymptotics 195
situations (3.16) holds for real A only on a set of Lebesgue measure zero. As it
was recently proven by the author [22] that q E l2 (LI) implies that the absolutely
continuous spectrum is lR+. However, the existence of the asymptotics (3.17) is
still an open problem related to some difficult issues of harmonic analysis (see,
e.g., [21])
In the conclusion of this section we note that in the setting of Schrodinger
operators with long-range potentials solutions of type (3.16) playa role of the
Jost solution in the short-range scattering. It can be proven (see also [22] in this
context) that if q E l2 (LI) then the perturbation 2-determinant ~2 of the pair
(H, H o ), (where Ho = H with q = 0) defined by
~2 (z) := det2 {I + (Ho - z)-1/2 q (Ho - z)-1/2}
exists and admits the representation:

~2 (A2) = \[1 (O,A) ,A E C+, (3.18)
where \[1 is the Weyl solution to (3.3) subject to
\[1 (x, A) rv exp {iAX + lax ql (8, A) d8} ,X -+ 00. (3.19)
This will be discussed in detail elsewhere. Note that (3.18) was first proven by
Koplienko [20] under conditions (1.7).
Theorem 1 type assertions are important tools in the study of perturbation
p-determinants ~p (A2) with p > 2 for which similar to (3.18) relations hold.
Theorem 1 for p = 2 was a crucial argument in [22] (see Remark 8). Note however,
that (3.18) does not hold for ~p (A2) with p > 2 - one should consider some
different (but quite similar) from (3.1 )-(3.2) transformations resulting in a different
expression for the phase in (3.19). We hope to return to these important issues
elsewhere.
4. Applications to the Weyl-Titchmarsh m-function

Let us consider
(4.1)
on the half-line lR+ with q E L 1 ,loc (lR+). Assume that q is real and in the limit
point case at 00 (see, e.g., [25]). Then the Weyl solution \[1 (x, A) of (4.1) is unique
up to a multiplicative constant and one can define the (Dirichlet) Titchmarsh-Weyl
m-function
m
(A2) = \[1' (0, A)
\[1 (0, A) ,A
2
E C+. (4.2)
It is well known that the Titchmarsh-Weyl m-function is analytic from C+ to C+

and plays a crucial role in spectral analysis of the one-dimensional Schrodinger op-
erator. Since there is no explicit formula for m in terms of q, it becomes important
196 A. Rybkin
to study m for large A. The research in this direction was originated by Everitt
[11] in 1972 who proved that
m (A2) = iA+o(1) ,A ---+ OO,A E SE:= {A: c:::; argA:::; 7f/2 -c,c > O}. (4.3)
In 1981 Atkinson [1] improved (4.3) to read (a is any positive)
m (A2) = iA + l a 2iAX
e q (x) dx +0 (IAI- 1 ) , (4.4)
as A ---+ 00 in SE: (or some parabolic domains allowing ImA ---+ 0). Atkinson's repre-
sentation (4.4) received much of attention and has been improved and generalized
in many directions by Bennewitz [4], Brown-Knowles-Weikard [6], Clark-Gesztesy
[8], [9], Gesztesy-Simon [13], Harris [14], [15], [16], Hinton-Klaus-Shaw [17], Kaper-
Kwong [18], [19], Rybkin [23], [24] and many others (see the extensive literature
cited therein). Note that one of the last papers by Atkinson [2] was also devoted
to generalization of (4.4) for potentials with strong singularities at the origin.
The result below is not new and in different forms has been discussed by
many authors but our statement appears to be optimal and its proof comes as a
simple corollary of Theorem 1 and the following proposition.
Proposition 2. Assume q to be real and from L 1 ,loc (IR+) , Ii = qX[O,aj, a > O. Let
m, iii be the Titchmarsh- Weyl m-functions corresponding to q, Ii respectively. If
ImA2 >0 andImA>max{4Joa lql,a- 1 In6}, then
1m (A2) - iii (A2) I ::; 8~4 IAI exp (-2a 1m A) . (4.5)
In an implicit form, Proposition 2 is due to Atkinson [1]. In the present form,

it was recently found by Simon and studied in great detail in Gesztesy-Simon [13].
Note that in [13] Gesztesy-Simon established that the converse for Proposition 2
is also true generalizing the Borg-Marchenko uniqueness result. (See also [3] for a
simpler proof.)
Theorem 2. Assume q E L 1 ,loc (lR+), real and in the limit point case at 00. Let
(qa)n(A) (qa)n(O,A) be as before definedforqa = qX[O,aj,a > O. Then for
A2 E C+
m (A2) = iA + L (qa)n (A) + r (A) , (4.6)
n2:1
where the remainder r (A) admits the estimate
11' (A)I :::; 8:41AI e-2almA
holding uniformly in A = {A2 E C+ : 1m A::::: max {4 Joa Iql ,a- 1 In6}} .
Proof. Let ma be the Titchmarsh-Weyl m-function corresponding to qa' Since

qa ELl, Theorem 1 applies with any n E N and equation (4.1) can be solved by
WKB asymptotics 197
iteration. By Proposition 1, for any fixed 1m A > 0 by choosing IAI large enough
one can make (1m A) -1 Ilia (-, A) 1100 < 1/2 and by Lemma 3 then
' " ( ) (A) < ' " II( ) (. A)II <

L...t qa n
n2:l
- L...t
n2:l
qa n ' 00 - 1 _ (1 mIlia (', A)lloo
A)-l II ~ (. A)II .
qa, 00
Corollary 1 applies with any n and one has
iii (A2) = iA + L (qa)n (A) , (4.7)
n2:l
where the series on the right converges at least in the domain Ilia (', A) (-, A) 1100 <
1/2 1m A. Applying Proposition 2 completes the proof. 0
Remark 9. Our representation (4.6) is not asymptotic: the series in (4.6) is abso-
lutely convergent in C+. In the previous literature cited above the series in (4.6) is
usually considered as asymptotic in sectorial domains
SE = {.A.: s ~ argA ~ 7r/2-s,s > O}.
The next assertion says when the error term in (4.6) can be dropped and
appears to be new.
Theorem 3. If q is real and from c l + Co (L l ) then
m (A2) = iA + L qn (A), (4.8)
for any A subject to the condition
1~AI11O e2iAsq(s+X)dSII00 < l. (4.9)
Proof. By Proposition 1, condition (4.9) is satisfied for some domain A asymp-

totically approaching the real line. Consider (4.7). By Lemma 2, Ln>l qn (0, A)
converges absolutely for any A E A and it is clear from the proof of the same
lemma that (qa) n ----+ qn, a ----+ 00, and the right-hand side of (4.7) converges to that
of (4.8). Since by our condition on q guarantees the limit point case at 00, one
concludes [25] that ma (A2) ----+ m (A2) ,a ----+ 00, on any compact set in C+ and a
passage to the limit in (4.7) yields (4.8). 0
Remark 10. The question when representation of type (4.8) holds was a focus of
[15] where representation (4.8) was proven for A from sectorial domains Se; under
the condition
11 00
e 2iAS q (s + x) dsl ~ a (x) b (A), (4.10)
with a (x) E LdlR+) and b (A) ----+ O,IAI ----) 00, A ESE' Condition (4.10), as ap-
posed to ours, assumes a strong decay of q at infinity (for instance, (fxoo IqjP) lip E
Ll (lR+) ,p > 1) . Note in this connection, that the conclusion of Theorem 3 holds
even for periodic potentials q integrable on the periods.
198 A. Rybkin
Remark 11. Since the spectrum of the operator -d2 /dx 2 + q (x), u (0) = 0, coin-
cides with the complement of the set {t E ~ : 1m m (t + iO) = O}, representations
of type (4.8) can be used in spectral analysis of Schrodinger operators on the line.
However, to make substantial assertions about the spectrum one needs to improve
condition (4.9) allowing ).2 in (4.8) to approach the real line (or at least be paral-
lel to it). In view of Remark 3, one of the trivial ways to achieve it is to require
uniform boundedness of q,q', and q". We have a strong reason to believe that a
more suibtle analysis of qn and a proper refinement of Proposition 1 should yield
in Theorem 3 a much stronger result on the domain of convergence of the series
in (4.8). We hope to address this issue elsewhere.
5. Appendix. Some auxiliary facts

We give here two known facts which we have used in the previous sections. For
the reader convenience we provide their simple proofs.
Lemma 2. Co (L 1 ) is a Banach space with respect to the norm
Proof. We are going to show that for any c > 0 a function f E Co (L 1 ) can be
represented as
f = II + 12, II E C8", 111211 < c.
Let X[a,b] be the characteristic function of a compact interval [a, b]. We have
f = fX[a,b] + fXIR\[a,b]. (5.1)
Choose now a, b E ~ so that
IlfXIR\[a,b] I = sup {Jr+1

x
} c
If I : x + 1 < a, x> b ="2. (5.2)
Since clearly fX[a,b] E L1 (a - 1, b + 1) and fX[a,b] = 0, x tt [a, b] , the function

fX[a,b] can be represented as
fX[a,b] = II + g,
where II E Co [a - 1, b + 1] and II = 0 on [a - 1, b + 1] \ (a, b), and 9 is such that
JIgl = ~.
b
Set now
a
12 = fXIR\[a,b] + g.
By (5.1) and (5.2), we have
J
b
111211::; IlfXIR\[a,b] II + Ilgll ::; ; + Igl = c

a
and the lemma is proven. o
WKB asymptotics 199
Lemma 3. Let qn be defined by (3.2) and

1
On:= 1m>."qn (,>')11= ,n E N. (5.3)
If 01 = 0 < 1/2 then
(5.4)
Proof. We conduct our proof by induction. It follows from (3.2) with n = 1 that
(a := 1m>.)
~a IIq2 (., >')11= ::; ~a sup +

ixr qI(., >.)}
= 21qn (s, >')1 2 ds
02
xElRix{= lexp {i>. (s - x) 1
::; -1 IlqI(, >.) II!, . 1= exp2 {( -a + IlqI(, >')11=) (s - x)} ds

a x
2 {= O? 2
(5.5)
= a01 . io exp {-2a (1 - 01) s} ds = 2 (1 _ 01) < 0 .
Assume that (5.4) holds for some n, then

n
~1 Om < l; om < ~1 (l)m
n
2"
n
= 1-
(l)n
2"
and similarly to (5.5) one has
1
On+1 = -llqn+1 (., >')11=
a
::; ~ IIqn (., >')II!' . 1= exp { -2as (1 - ~ t111qm (., >')11= ) } ds
~ ~ liOn (, A) II;;' {2a (1- ~ ~ 110m (-, A)II= ) }

r 2;
-J
~ ~,~ (1- t, 'm < ,'n ~ 2n- J Gf d J

,;n+J n+ J D
References
[1] F.V. Atkinson, On the location of the Weyl circles, Proc. Roy. Soc. Edinburgh Sect.
A 88 (1981), no. 3-4, 345-356.
[2] F.V. Atkinson and C. T. Fulton, Asymptotics of the Titchmarsh- Weyl m-coefficient
for non-integrable potentials, Proc. Roy. Soc. Edinburgh Sect. A 129 (1999), no. 4,
663-683
[3] C. Bennewitz, A proof of the local Borg-Marchenko theorem, Comm. Math. Phys.
218 (2001), no. 1, 131-132.
[4] C. Bennewitz, Spectral asymptotics for Sturm-Liouville equations, Proc. London
Math. Soc. (3) 59 (1989), no. 2, 294-338.
200 A. Rybkin
[5] V.S. Buslaev and V.B. Matveev, Wave operators for the Schrodinger equation with
slowly decreasing potential, Theoret. and Math. Phys. 2 (1970), no. 3, 266-274.
[6] B.M. Brown, I. Knowles, and R. Weikard, On the inverse resonace problem, J. Lon-
don Math. Soc. (2) 68 (2003), no. 2, 383-40l.
[7] M. Christ and A. Kiselev, Scattering and wave operators for one-dimensional
Schrodinger operators with slowly decaying nonsmooth potentials. Geom. Funet.
Anal. 12 (2002), no. 6, 1174-1234.
[8] S. Clark and F. Gesztesy, Weyl- Titchmarsh M -function asymptotics, local unique-
ness results, trace formulas, and Borg-type theorems for Dirac operators, Trans.
Amer. Math. Soc. 354 (2002), no. 9, 3475-3534
[9] S. Clark and F. Gesztesy, Weyl-Titchmarsh M -function asymptotics for matrix-
valued Schrodinger operators, Proc. London Math. Soc. (3) 82 (2001), no. 3, 701-724.
[10] P. Deift and R. Killip, On the absolutely continuous spectrum of one-dimensional
Schrodinger operators with square summable potentials, Commun. Math. Phys. 203
(1999), 341-347.
[11] W.N. Everitt, On a property of the m -coefficient of a second-order linear differential
equation, J. London Math. Soc. (2) 4 (1971/72), 443-457.
[12] D.J. Gilbert and D.B. Pearson, On subordinacy and analysis of the spectrum of one-
dimensional Schrodinger operators, J. Math. Anal. Appl. 128 (1987), 30-56.
[13] F. Gesztesy and B. Simon, A new approach to inverse spectral theory. II. General
real potentials and the connection to the spectral measure, Ann. of Math. (2) 152
(2000), no. 2, 593-643.
[14] B.J. Harris, The asymptotic form of the Titchmarsh- Weyl m-function associated with
a non-definite, linear, second order differential equation, Mathematika 43 (1996), no.
1, 209-222.
[15] B.J. Harris, An exact method for the calculation of certain Titchmarsh- Weyl m-
functions. Proc. Roy. Soc. Edinburgh Sect. A 106 (1987), no. 1-2, 137-142.
[16] B.J. Harris, The asymptotic form of the Titchmarsh- Weyl m-function associated with
a second-order differential equation with locally integrable coefficient. Proc. Roy. Soc.
Edinburgh Sect. A 102 (1986), no. 3-4, 243-25l.
[17] D.B. Hinton, M. Klaus, and J.K. Shaw, Series representation and asymptotics for
Titchmarsh- Weyl m-functions., Differential Integral Equations 2 (1989), no. 4, 419-
429.
[18] H.G. Kaper and M.K. Kwong, Asymptotics of the Titchmarsh- Weyl m-coefficient
for integrable potentials. II. Differential equations and mathematical physics (Birm-
ingham, Ala., 1986), 222-229, Lecture Notes in Math., 1285, Springer, Berlin, 1987.
[19] H.G. Kaper and M.K. Kwong, Asymptotics of the Titchmarsh- Weyl m-coefficient
for integrable potentials, Proc. Roy. Soc. Edinburgh Sect. A 103 (1986), no. 3-4,
347-358.
[20] L.S. Koplienko, Regularized function of spectral shift for a one-dimensional Schro-
dinger operator with slowly decreasing potential, Siberian Math. J. 26 (1985), no. 3,
365-369.
[21] C. Muscalu, T. Tao and C. Thiele, A counterexample to a multilinear endpoint ques-
tion of Christ and Kiselev. Math. Res. Lett. 10 (2003), no. 2-3, 237-246.
WKB asymptotics 201
[22] A. Rybkin, On the absolutely continuous spectrum and negative discrete spectra of
Schrodinger operators on the line with locally integrable globally square summable
potentials, to appear in JMP.
[23] A. Rybkin, Some new and old asymptotic representations of the Jost solution and
the Weyl m-function for Schrodinger operators on the line. Bull. London Math. Soc.
34 (2002), no. 1, 61-72.
[24] A. Rybkin, On a complete analysis of high-energy scattering matrix asymptotics for
one-dimensional Schrodinger operators with integrable potentials, Proc. Amer. Math.
Soc. 130 (2002), no. 1, 59-67.
[25] E. C. Titchmarsh, Eigenfunction expansions associated with second-order differential
equations. Part I, Second Edition, Clarendon Press, Oxford 1962, 203 pp.
Alexei Rybkin
Department of Mathematical Sciences
University of Alaska Fairbanks
PO Box 756660
Fairbanks, AK 99775, USA
e-mail: ffavr(!)uaf. edu
Operator Theory:
Uniform Levinson Type Theorems for

Discrete Linear Systems
Luis O. Silva
Abstract. For the solutions of parametric discrete linear systems we obtain

an asymptotic formula whose remainder is estimated uniformly with respect
to the parameter. This result is an extension of the discrete Levinson theorem
and allows us, in particular, to study the parametric discrete linear systems
that arise when dealing with the generalized eigenvectors of Jacobi matrices.
Mathematics Subject Classification (2000). Primary 39Al1;Secondary 47B36.
Keywords. Asymptotic behavior, Levinson theorem, Jacobi Matrices.
1. Introduction
The Levinson theorem, in its discrete version, describes the asymptotic behavior
as n ---+ 00 of the sequence of d-dimensional vectors {xn}n~no being a solution of
the linear system
(1)
where the sequence of d x d matrices {An}n>no satisfies certain conditions be-
ginning at some nl ~ no. If, for instance, (1) is a diagonal system, that is,
An = diag{v~k)H=1 for all n ~ no, then a fundamental system of solutions of (1)
. gIven
IS . Xn n;:>:no (k =,
b y the sequences {_(k)} o::tCk) = Il i=no
1 ... , d) , suchth a t Xn n - 1 (k)_
Vi ek,
where {en}~=1 is the canonical basis in Cd ( ek is the vector with all components
zero except the k-th, which is one). Therefore, in this case, a basis in the space of
solutions of (1) behaves asymptotically, as n ---+ 00, as Il~:;o vY)ek. The discrete
Levinson theorem asserts that a similar asymptotic behavior takes place when
{An}n>no is perturbed by a sequence of matrices whose norms form a summable
series, and even when An is not diagonal, but diagonalizable in a specific sense (see
below). This assertion is the discrete analogue of the well-known Levinson theorem
on the asymptotics of solutions of a system of ordinary differential equations [2].
204 L.O. Silva
The asymptotic analysis of solutions of discrete linear systems has already

a long history. Important results on the asymptotics of the solutions of (I), in-
cluding the discrete Levinson theorem, are in [1] and [3]. A modern approach to
the asymptotic analysis of solutions of difference equations is in [7], where various
discrete Levinson type theorems are proved.
In 1992 the theory of subordinacy, developed some years earlier as a tool in
the spectral analysis of ordinary differential operators by Gilbert and Pearson [5]
[4], was successfully carried over into the discrete domain [11]. Since then, various
asymptotic methods have been used to study the spectral properties of difference
operators, in particular, Jacobi matrices. One of these techniques has been the
discrete Levinson theorem (cf. [8], [7] and [13]).
It is well known that the three term recurrence equation for determining the
generalized eigenvectors of a Jacobi matrix is equivalent to a discrete linear system
similar to (I), which depends on the parameter A E JR, i.e.,
for n 2': no, (2)
where, 'VA E JR, {A n (A)}n2 no is a sequence of 2 x 2 matrices and {xn (A)}n2 no
is a sequence of two-dimensional vectors (see [9], [10] and [7]). Since we are most
interested in applications to general difference operators, we shall consider systems
of the form (2), but in a broader setting, namely, we regard {A n (A)}n2 no to be a
sequence of d x d complex matrices for every A in some interval J c R Besides,
in order to avoid trivial cases, we always assume that
Notice that, under this assumption, the solutions of the system (2) form a d-
dimensional space and any solution {x n (A)}n2 no is uniquely determined by
n-l
Xn(A) = II Ai (A)Xno (A) , (3)
i=no
Here, and later on in this paper, the product of matrices is considered in "chrono-
logical" order, i.e., I1~==-;o Ai(A) = An-I ... Ano The simplicity of (3) is delusive,
since it generally does not give direct information about the asymptotic behavior
of the solutions.
Traditionally, equation (2) has been studied pointwise with respect to A.
However, in many cases, there is a uniform behavior of the system with respect to
this parameter ([7] and [13]). Making use of this uniformity to obtain a uniform
estimate of the remainder in the asymptotic expansion of the solutions as n -7 00
turns out to be important in the spectral analysis of Jacobi matrices.
Recent developments [13] have shown that, when studying Jacobi matrices, a
complete spectral analysis requires an extension of the discrete Levinson theorem
to deal with the parametric recurrence systems that arise from the equation for
determining the generalized eigenvectors of operators associated with Jacobi ma-
trices. The point is that the spectral analysis given by the theory of subordinacy
is not complete in the sense that it left unanswered the question of whether pure
Discrete Uniform Levinson Type Theorems 205
point parts of the spectrum have points of accumulation on any finite interval.
The method used to answer this question requires, among other things, a uniform
estimate of the asymptotic remainder of the solutions of (2) with respect to A.
Peculiarities of this method are its wide applicability and that it is based on the
asymptotic behavior of solutions, i.e., it is in the spirit of subordinacy theory. In
this paper we develop the first "ingredient" of this method, the discrete uniform
Levinson theorem. The other parts are planned to be the material of other papers.
It was decided to present the uniform Levinson theorem separately because the
author thinks that it may have other applications besides the one described in this
introduction.
2. Preliminaries
For the definitions below, we consider a sequence V = {Vn(A)}n~no' such that,
for every A in some interval J c ~, Vn(A) is a complex d x d matrix with the
eigenvalues {v~k) (A)}~=l.
Definition 2.1. We shall say that the sequence V = {Vn(A)}n~no satisfies the
Levinson condition for k (L.c.(k)) when there exist an N 2: no and a constant
number M > 1 such that, k being fixed, each j (1:::; j :::; d) falls into one and only
one of the two classes h or h, where
(a) j E h if 'VA E J
ITI~-=-~ V;k) (A)I

-t 00 asn-too, and
ITI~:~ v?) (A) 1
In:~ll Vi(k\>.) I
ITI ~2-1 v(j)(A)1
z=nl 'l,
(b) j E 12 if 'VA E J
ITIn 2 - 1 v(k) (A)I
<=nl < <M 'Vn2 ,nl such that n2 > nl 2: N.
n 2 - 1 v(j)(A)1 '
ITI'l.=nl Z
Remark 1. It is easy to show that when the sequence V = {Vn(A)}n~no is such

that
where V00 is a matrix whose eigenvalues are different from zero and have pair-
wise distinct absolute values, the sequence V satisfies the (L.c.(k) for every
k = 1, ... ,d.
Definition 2.2. Assuming that v~k) (A) I=- 0 'V k, n, A, we define for each k the normed
space X k (no) that consists of all sequences cP = {CPn (A) }~=no of functions defined
206 L.O. Silva
on some interval J c ffi. and with range in Cd, such that
sup suP{IItP'n('~)lllCd 1 1 (k) } < 00

n>no
- AEJ nn-
1,=no v 1- (A)
1 1
and where the norm is defined by
11tP'llxdno) = sup sup{lltP'n(A)llcd 1nn-1 1 (k)( )I}' (4)

n?:no AEJ i=no vi A
Clearly, this space is complete.
In the space Xdno) we shall also consider the subspace 1 XZ(no) which con-
tains all functions of Xk(nO) such that
suP{IItP'n(A)lk~d 1 1 (k) } ----* 0 as n ----* 00 (5)

AEJ 1 nn-
'Z=no
v 'I,
(A) 1
Consider now the linear space of sequences V = {Vn (A)} ~=no' In this space
we define the operator Do by
(6)
Definition 2.3. The sequence V = {Vn}~=no is said to be in the class 151 iff
{sup II(DoV)nll}~=no E [1.
AEJ
This class is just the uniform analogue of one of the classes defined in [14]
In addition to these notations and concepts, we also shall need the following
projectors acting in the linear space Cd. Consider again the canonical orthonormal
basis in Cd, {ed%=l' Let Pk be the projector to the one-dimensional space gener-
ated by ek, i.e., Pk is the d x d diagonal matrix whose diagonal has the coordinates
of ek as its elements:
k = 1, ... ,d. (7)
Using these projectors and the classes h, h from the definition of (L.c.(k)), we
define
p(i) = LPj i = 1,2. (8)
JEI i
3. Auxiliary results
In the following lemma we introduce, with the help of two sequences of matri-
ces satisfying certain conditions, an operator in Xk(no) and establish some of its
properties.
IThat is a closed subset with norm induced from Xk(no).

Lemma 3.1. Consider the sequences {An (>.)} ;::'=no and {Rn (>.)} ;::'=no' such that,
for every >. in some interval 'J c JR, An(>') and Rn(>') are d x d complex matrices
and An(>') has the eigenvalues {v~k) (>')}~=l' such that v~k) (>.) =I- 0 Vk, n, >.. If the
following conditions hold
1. the sequence {An(>')};::'=no satisfies (L.c.(k)) for k = 1, ... , d.
ii. there exists an Nl ~ no such that for all n > N 1 , and all >. E 'J, the matrices
An(>') are diagonal matrices.
iii. there exists a C > 0 such that SUPAEJ f:
II~~?)II < C
n=no IVn (A)I
iv. for any f > 0 there exists an N, (which depends only on f) such that V>' E 'J
we have
~ IIRn(>.)11 < f
~ (k) ,
n=N, IVn (>')1
then there exists a natural number No ~ Nl such that the operator Tk defined for
every <1 = {<1n(>')};::'=No in Xk(No ) (k = 1, ... ,d) by the following expression
n-l
(Tk<1)n(>') = L p(1hl1n(>')W;:;'~l (>.)Rm (>')<1m (>.)
m=No (9)
00
m=n
where Wn (>.) = I1~:~o Ai(>'), has the following properties:

1. Tk is a correctly defined operator on Xk(No), that is, the series in (9) con-
verges, provided that <1 is in Xk(No).
2. IITkl1 < 1
3. TkXk(No) c XZ(No)
Proof. We begin by choosing a natural number No ~ max{Nl' N}, where Nl is

from condition ii and N is from i (see definition of L.c.(k)) and such that
M sup
AEJ
f IIRn(>') II
n=No Iv~k) (>.) 1
< l. (10)
Firstly we show that nis correctly defined on the space Xk(No ). Let us begin by
estimating the series in (9).
where
l E h,
else.
208 L.O. Silva
Therefore, taking into account that m 2': n,

n- 1 (I) (.\) I I
I_I TI i=No vi
(k) (k) (k)
Ih(l) 1_ 1 1 Vn Vn +1 .Vm
n.m - TI:NoVi(I)(.\) -lv~i)v~l~l ... v~1 Iv~k)v~~l ... v~)1 Iv~l)v~IL ... v~1
1 TI n - 1 v(k) (.\)
<M -MI 2=No 2 I
Iv~k) V~~l .. . V~) I - TI:No vik) (.\) .
Here, we have used (b). Thus, from (4) and condition iii, it is thus clear that for
any sequence r.j5 = {r.j5n(.\)}~=No in Xk(No) we have convergence of the series in (9).
Let us now study the first term of the equation
p(l)ll1n('\)Il1;;-'~l (.\) = diag{f~l,~(.\)}t=l'
where
l E h,
else.
Hence, taking into account that in this case m ::::: n and using (a), we have
(l) I - ITI;:~o vii) (.\) I _ I (I)

Ij n,m (I) (l) I
- m (l) - v m + 1 v m + 2 Vn - 1
TIi=No Vi (.\)
I
I IVm +1 Vm +2 Vn -
(l) (I) (I)
_I (k) (k) (k) 1
- VmH Vm +2 Vn - 1 (k) (k) (k)
Ivm+1vm+2 vn_11
n-1 (k) (.\)
TI i=No
< MI (k) (k) (k) I _ MI vi I.
- Vm + 1 Vm +2 Vn - 1 - m (k)
TIi=No vi (.\)
We now show that operator Tk is a contraction.
IITkr.j5l1x.(No) = sup sup{ n-l 1 (k)

n?No AEJ ITIi=No Vi (.\)1
n-l 00
m=No m=n
<
- sup
{M ~ IIRm(.\)11
L... (k)
II r.j5m (.\)II
m-l } <- I rp~II XdNo)
AEJ m=No IVm (.\)1 I TI V?) (.\)1
i=No
Note that in the second line we have written Il1 n instead of Il1 n (.\) to simplify the
writing of the formula.
It remains to prove property 3. We must show that if No' Using the definition of Tk (see (9)) we obtain
Since <P is in Xk(No) it follows at once from (b) that the second term on the
right-hand side of the last expression can be done as little as we want if we take n
sufficiently big.
Consider now the first term on the right-hand side of (11).
Provided that s > No this inequality holds for every n ;::: No no matter how big it
is. Now, since
and taking into account that {<Pn(A)}n>No

-
E Xk(No) implies {<Pn(A)}n>s
-
E Xk(s),
there exists an s > No, such that for every 1 > 0
210 L.O. Silva
Thus, we have shown that the first term on the right-hand side of (11) is estimated
as follows
n-l
sup { n_l 1 (k) II L
p(I)lJIn(.~)IJI;;-'~1 (oX)Rm(oX)CPm(oX)ll}
AEJ I ITi=No vi (oX) I m=No
8-1
:S sup{ n_l 1 (k) L IIp(I)lJIn(oX)IJI;;-'~1 (oX)Rm (oX)CPm (oX) II } + El
But
8-1
sup { n_l 1 (k) L
IIp(1)lJIn(oX)IJI;;-'~1 (oX)Rm(oX)CPm(oX)ll}
C {"p(1)lJIn(oX)ll}
:S ~~~ I IT~:~o v?) (oX)1 '
where C is a positive constant. Now, as a consequence of (a) we know that
{ 11P(1) IJI n (oX) II } { I IT~:~ vl 1) (oX) I }

asn-+oo.
~~~ IIT~:~ovlk)(oX)1 =~~~ IIT~:~:v}k)(oX)1 -+0
This completes the proof of the lemma. D
Now we show that, using the operator introduced in the previous lemma,
we can write an equation which turns out to be equivalent to a recurrence linear
system of the form (2) defined by the sequences {An(oX)}n~no and {Rn(oX)}n~no.
Lemma 3.2. Let the sequences of function matrices {An(oX)}n~no and {Rn(oX)}n~no
satisfy the conditions of the previous lemma. Let No ~ no be such that Tk given
by (9) is correctly defined:. The sequence of function vectors cP = {~k) (oX)}n~No
in Xk(No) is a solution of the recurrence equation
CPn+l(oX) = (An (oX) + Rn (oX))CPn (oX) n~ No (12)
iff cP satisfies equation
cP = -(k) + TkCP (13)
where the sequence -(k) = {-~k) (oX)}n~No is defined by
n-l
-~k) = IJIn(oX)ek = II vl k)(oX)ek
i=No
Proof. The proof is straightforward. Using the definition of Tk in (13) and substi-
tuting this into (12) one obtains an identity (it has to be taken into account that
An(oX)lJIn(oX) = IJI n+1(oX) and that p(l) + p(2) = I, where I is the d x d-identity
matrix). D
2The existence of No is guaranteed by Lemma 3.1.

Remark 2. A straightforward consequence of Tk's property 3, stated in Lemma

3.1, is that if r.j5 E Xk(No) satisfies equation (13) then
r.j5 - ;j;(k) E XZ(No ) . (14)
It is of practical importance to deal with systems such as (12), but when
the sequence {An(>')}n>no is not necessarily diagonal. The following lemma gives
sufficient conditions for a sequence of matrices to be diagonalizable in a specific
sense.
Lemma 3.3. Let {Vn(>\)}n2 no be a sequence of function matrices defined on every
>.. in some interval J c ffi., such that
1. {Vn(>")}n>no is in 15 1
.\EJ
11.Vn (>..) :::::t V00 as n -+ 00, where V00 is a d x d matrix with pairwise distinct
eigenvalues {v(k)}~=1
Then we can find an m :::: no such that there exists a sequence of diagonal matrices
{An(>")}n2m and a sequence of invertible matrices {G n(>")}n2m such that
1. {An(>")}n>m E 15 1
2. {Gn(>")}n~m E 15 1
3. Vn (>..) = Gn(>")An(>")G;;:I(>..) for n:::: m for all >.. E J
.\EJ
4. An(>") :::::t A as n -+ 00, where A = diag{v(k)}~=1
.\EJ
5. Gn(>") :::::t G as n -+ 00, where G is invertible and Voo = GAG- 1
Proof. The proof is very similar to the one in [6] for the non-parametric analogous
assertion.
As we have done before, let us denote by {vAk)(>")}~=1 the set of eigenvalues
of Vn (.\). Since the matrix Voo has pairwise distinct eigenvalues, by choosing E > 0
small enough, the operators
p(k) = f (voo - zI)- l dz k = 1, ... , d (15)

Iz-v(k)I=E
are projectors onto the eigenspaces of Voo corresponding to v(k) (k = 1, ... , d).
Now, taking E even smaller if necessary, it is clear that there exists some N :::: no,
such that the operators, defined for all >.. E J and each n :::: N by
PAkl(>..) = f (Vn(>") - zI)- l dz k = 1, ... , d (16)

Iz_v~k) (.\)I=E
are projectors onto the eigenspaces of Vn(>..) corresponding to the eigenvalue vAk) (>..)
.\EJ
(k = 1, ... , d). From the conditions of the lemma it is clear that vAk) (>..) :::::t v(k)
and, therefore, k i=- j implies
vAk) (>..) i=- v~) (>..) \j n :::: N, and \j >.. E J. (17)
212 L.O. Silva
We define the matrices

An(A) = diag{v~kl(A)}%=1
From (17) we have property 4. Furthermore, using (17) and ii, from (15), (16) we
have
AEJ
p~kl (A) ~ p(kl and {p~kl (A) }n~N E 151
Now, for k = 1, ... ,d we choose an arbitrary non-zero vector Vk in Ran(Pk ) and
construct the function matrix Gn(A) with columns
P~1)(A)Vl"'" p~dl(A)Vd'
Therefore
Vn(A)Gn(A) = Gn(A)An(A) for n ~ N, VA E J.
AEJ
Now, since p~kl (A)Vk ~ p(kl vk = Vk I=- 0 as n _ 00, we can find an m ~ N
such that Gn(A) are invertible for n ~ m. Thus property 3 holds. Moreover,
{p~kl (A)}n~m E 151 implies 2. Now, let us consider the matrix G with columns
given by the vectors VI, ... , Vd' Clearly
(18)
and G is invertible. Thus 5 holds. So this, together with 2, implies
{Gn(A)-I}n~m E 151 .
Finally property 1 takes place since 15 1 is an algebra. o
Remark 3. It is not difficult to understand that if the sequence {Vn(A)}n~m satis-
fies the Levinson condition (L.c.(k)) for a certain k, then the diagonalized sequence
{An(A)}n~m also satisfies this condition for the same k.
4. The asymptotic behavior of solutions

In this section we obtain Levinson type assertions that give a uniform estimate of
the remainder in the asymptotic expansion.
Theorem 4.1. Let the sequences {An(A)}n~no and {Rn(A)}n~no satisfy the condi-
tions of Lemma 3.1. Then we can find an No E N such that there exists a basis
{~kl(A)}%=1 in the space of solutions of (12) satisfying
~kl(A) _
~~~II rr~:';' V?l(A) - ekll- 0, as n - 00, for k = 1, ... ,d. (19)
Proof The statement of the theorem easily follows from Lemmas 3.1 and 3.2.
Note that property 2 implies that (13) yields linearly independent solutions for
k = 1, ... ,d. Equation (19) follows directly from (14). 0
Now we use Lemma 3.3 to replace the restricting condition of diagonality

that in Theorem 4.1 is imposed on {An(A)}n>no by a weaker assumption, viz.,
that a sequence is diagonalizable in the sense specified in Lemma 3.3.
Theorem 4.2. Let Voo be a dxd matrix whose non-zero eigenvalues v Ck ) are pairwise
distinct. Consider the sequences {Vn (A)} ~=no and {Rn (A)} ~=no' such that, for
every A in some interval J c JR, Vn(A) and Rn(A) are d x d complex matrices
and Vn(A) has the eigenvalues {v~k\A)}%=I' such that v~k) (A) tf 0 V k, n, A. If the
following conditions hold
>'EJ
i. Vn(A) ~ Voo as n ---> 00.
11. the sequence {Vn(A)}~=no satisfies (L.c.(k)) for k = 1, ... , d.
iii. {Vn(A)}~no E 151

iv. there exists a C > 0 such that sUP>'EJ I::=no IIRn(A)11 < C.
v. for any f > 0 there exists a N (which depends only on f) such that VA E J
we have
00
L IIRn(A)11 < f,
then we can find an mEN such that there exists a basis {x~k) (A) }%=1 in the space
of solutions of the recurrence relation
(20)
such that
x}.k) (A) ~
supll
>'EJ
nn-l Ck)(\) - Pkll
i=m vi /\
---> 0, as n ---> 00 ,
where Pk is the eigenvector of Voo corresponding to v(k).

Proof Since {Vn(A)}~=no satisfies here the conditions of Lemma 3.3 we can find
an ml ~ no such that VA E J
Vn(A) = G~I(A)An(A)Gn(A) for n ~ mI.
Therefore, letting rp},k) (A) = Gn(A)x}.k) (A) in (20), considered for n ~ ml, we have
~k21(A) = Gn+l(A) (Vn(A) + Rn(A))G~I(A)~k)(A) for n ~ ml
This recurrence system can be rewritten as follows
~~1 = (An(A) + Rn(A)) ~k), (21)
where Rn(A) = (~G)n(A)Vn(A)G~I(A) + Gn+l(A)Rn(A)G~I(A) (recall the oper-

ator ~ introduced in Sec. 2). Clearly the sequence {An(A)}~=ml' by Lemma 3.3,
satisfies the conditions of Lemma 3.1, in particular (L.c.(k)) for k = 1, ... , d.
Next we show that the sequence {Rn(A)}n~ml satisfies the conditions iii and ivof
Lemma 3.1. Indeed, considering condition i and property 5 of Lemma 3.3, it is not
214 L.O. Silva
difficult to show that we can find an m2 ::::: m1, such that there exist G1 , G2 , G3 > 0
for which the following estimates hold
IIRn(A)11 :::; GIiI(~G)n(A)11 + G21IRn(A)11 VA E J, Vn::::: m2

(22)
Iv~k)(A)I:::::G3 VAEJ, Vn:::::m2 (k=l, ... ,d).
Now, since {Gn(A)}~=ml is in 151, there exist a G > 0 such that

00
This last estimate, together with (22) and conditions iv and v, yields the bound-
edness of
and the uniform tail's estimate
Hence the conditions of Lemmas 3.1 and 3.2 are satisfied. This means that we can
apply Theorem 4.1 to the system (21). Therefore for an m ::::: m2 there exists a
basis {<p k)(A)}%=l in the space of solutions of (21) such that
n-1
supll~k)(A)/ II V;k) (A) - ekll -+ 0, as n -+ 00,
.\EJ i=m
Thus, (20) has the linearly independent solutions x~k) (A)

k = 1, ... ,d. It is also not difficult to show that
n-1
supllx~k)(A)/ II VY)(A) - Pkll -+ 0, as n -+ 00,
.\EJ i=m
where Pk = G-1ek is the eigenvector of Voo corresponding to v(k). o
Remark 4. If in Theorem 4.2 we consider the sequences {Vn(A)}n>no and Voo

defined as in Remark 1, then Voo satisfies the conditions of the theorem and i and
ii are fulfilled. In this case we have the "classical" discrete Levinson theorem [7],
but in its uniform variant.
5. Asymptotics of the generalized eigenvectors of

a Jacobi matrix with rapidly increasing weights
Let us consider an operator J in l2 (N) whose matrix representation with respect
)
to the canonical basis in l2(N) is a Jacobi matrix of the form
...
... , (23)
where b = {bn}nEN is a sequence of positive numbers. By means of the subordinacy

theory, the spectral characterization of J can be done by analyzing the asymptotic
behavior, as n ---+ 00, of the solutions of the following equation
n>l, >'ElR. (24)
The solutions u = {un(>')}~=l of (24) are called the generalized eigenvectors of J
corresponding to >.. Equation (24) can be written as follows
Un+l = Bn(>')un , n> 1, >. E lR. (25)
where un(>') = (u::~;~)) and Bn(>') ( _ b~:' b~) is the so-called transfer
matrix. It turns out that it is possible to group the transfer matrices Bn(>') in
such a way that the resulting system satisfies the conditions of Theorem 4.2 or, at
least Theorem 4.1. Let us illustrate this for the case when the sequence b = {bn}~=l
is defined by
nEN, (26)
where 1 < a < 2 and {Cn}nEN is a positive periodic sequence of even period,
i.e., Cn+T = Cn for all n E N and a fixed even natural number T = 2L. This
class of Jacobi matrices with rapidly growing weights is based on an example
suggested by A.G. Kostyuchenko and K.A. Mirzoev [12]3. The asymptotics of
the generalized eigenvectors for this class, has been studied in [7] and [13] with
different approaches. We use here mainly the same reasoning as in [7], but applying
our uniform version of the Levinson theorem.
First we define how we should group the transfer matrices in order to apply
the main result of this paper. Consider the following recurrence equation for the
sequences of two-dimensional vectors x(>') = {xn(>')}~=l
nEN, >'ElR. (27)
where An(>') = TI~':~ BnT+k(>'). It is clear that for any solution x(>') of (27) there
exists a unique sequence u being a solution of (24) whose elements are determined
by the second component of the vector UnT(>') = xn(.A). Moreover, there is an
3 Actually, this class is not restricted to the case a < 2. This restriction is not important for this
example, but simplifies some formulas.
216 L.O. Silva
isomorphism x f--7 u of the space of solutions of (27) to the space of solutions of

(24) and we have for every n E N and the integers s = 1, ... , T
(28)
where e2 = (?) (el = (6)) and (., .)c2 is the inner product in C 2, which is used to
single out the second component of the vector.
Let us calculate An(>") for our particular matrix. Simple algebraic computa-
tions give us
(29)
where
o>. ) ,
TQ(1+0(~J)
and Qk(n) is a matrix which does not depend on >.. and tends to a constant matrix
when n -+ 00. Thus,
(30)
where I is the 2 x 2-identity matrix and where the sequence of matrices {Rn(>") }nEl\I
tends, uniformly with respect to >.. in any finite interval 'J of JR, to a constant
matrix. Indeed, from (29) it straightforwardly follows that each element of Rn(>")
is a product of a polynomial of >.. and a function of n which does not depend on >..
and tends to a constant as n -+ 00. As regards the second term of the right-hand
side of (30) we have
Now, assuming that
and
it is easy to show that the sequences {Vn(>")}nEl\I and {Rn(>")}nEl\I satisfy the
conditions of Theorem 4.1 for the system (27). Thus, there exists an mEN such
that, as n ----+ 00,
where k = 1,2.
Now, reasoning as in [7], one can obtain a basis u(k) = {U~k)}n~m' k = 1,2,
in the space of solutions of (24) such that, for s = 1, ... ,T,
(k)
u nT +s
(,X) _ (_l)nL
-
nrr-I (1 _Z-I (::: +
2
(_l)k+I 2:~:I T(-l)j Cj)) w nT
(k)
+s
(,X)
'
l=m
where
supllw~k) (,X) - (nek' e2)c211 ----+ 0, as n ----+ 00, k = 1,2.
AEJ
Hence we have obtained uniform estimates of the asymptotic remainder term for
the generalized eigenvectors of J.
References
[1] Z. Benzaid and D. A. Lutz. Asymptotic representation of solutions of perturbed
systems of linear difference equations. Studies Appl. Math., 77(1987) 195-22l.
[2] E. Coddington and N. Levinson. Theory of Ordinary Differential Equations. Mc-
Graw-Hill, New York, 1955.
[3] S. N. Elaydi. An Introduction to Difference Equations. Springer-Verlag, New York,
1999.
[4] D. Gilbert. On subordinacy and analysis of the spectrum of Schrodinger operators
with two singular end points. Pmc. Roy. Soc. Edinb., 112A(1989) 213-229.
[5] D. Gilbert and D. Pearson. On subordinacy and analysis of the spectrum of one-
dimensional Schrodinger operators. J. Math. Anal. Appl., 128(1987) 30--56.
[6] J. Janas and M. Moszynski. Alternative approaches to the absolute continuity of
Jacobi matrices with monotonic weights. Integr. Equ. Oper. Theory, 43(2002) 397-
416.
[7] J. Janas and M. Moszynski. Spectral properties of Jacobi matrices by asymptotic
analysis. J. Approx. Theory, 120(2003) 309-336.
[8] J. Janas and S. Naboko. Multithreshold spectral phase transitions for a class of
Jacobi matrices. In Recent advances in operator theory (Gmningen, 1998), volume
124 of Oper. Theory Adv. Appl. Birkhauser, Basel, 2001, 267-285.
[9] J. Janas and S.N. Naboko. Jacobi matrices with absolutely continuous spectrum.
Pmc. Amer. Math. Soc., 127(1999) 791-800.
[10] J. Janas and S.N. Naboko. Jacobi matrices with power-like weights, grouping in
blocks approach. J. Functional Analysis, 166(1999) 218-243.
[11] S. Khan and D. Pearson. Subordinacy and spectral theory for infinite matrices. Helv.
Phys. Acta, 65(1992) 505-527.
218 L.O. Silva
[12J A.G. Kostyuchenko and K.A. Mirzoev. Generalized Jacobi matrices and deficiency
indices of differential operators with polynomial coefficients. Funct. Anal. Appl.,
33(1999) 30-45.
[13J L.O. Silva. Spectral properties of Jacobi matrices with rapidly growing power-like
weights. In Operator methods in ordinary and partial differential equations (Stock-
holm, 2000), volume 132 of Oper. Theory Adv. Appl. Birkhiiuser, Basel, 2002, 387-
394.
[14J G. Stolz. Spectral theory of slowly oscillating potentials I. Jacobi matrices. Manu-
scripta Math., 84(1994) 245-260.
Luis O. Silva
Departamento de Metodos Matematicos y N umericos
IIMAS-UNAM
Apdo. Postal 20-726
Mexico, D.F. 01000
e-mail: silva@leibniz.iimas.unam.mx
Operator Theory:
Free Functional Model Related

to Simply-connected Domains
Alexey Tikhonov
Abstract. The aim of this paper is to extend the S.-Nagy-Foill functional

model to the case of functions for contractive operators. This model meets
general requirements and can be used for studying trace class perturbations
of normal operators with spectrum on a curve.
Mathematics Subject Classification (2000). Primary 47A45; Secondary 47 A48,

47A55.
Keywords. functional model, operator, Hilbert space.
Introduction
The functional model of S.-Nagy-Foia [1] of contractive (dissipative) operators is
well known and has many important and useful applications. It was generalized
many times and in various directions. Here we single out three of them: 1) models
of unitary nodes [2]; 2) models of operators those are analogous of contractive
operators [3, 4]; 3) models of operators those are close to unitary (self-adjoint)
operators but are not contractive (dissipative) [5,6].
On the other hand, trace class perturbations of normal operators with spec-
trum on a curve is an important object to study. In [7J the author revealed that
such operators can be represented in the form <p(To) + NoxMo, where <p is a
conformal map of the unit disk, To, No, Mo are entries of an unitary operator-
matrix 2l0 , and x is a bounded operator. This representation enables us to
extend scheme from [5, 6J to the trace class perturbations of normal operators
provided that we have a suitable functional model for the triple of operators
(<p(To), Mo, No). Problems, which are studied in [7], impose some requirements
on this model. For instance, the definition of spectral components yields that the
Hardy-Smirnov spaces E2(G) [8J and the corresponding (non-orthogonal) projec-
tions P (RanP = E2(G), Ker P = E2(G~)) should be important ingredients
220 A. Tikhonov
of the model (here G + = int C , G _ = ext C are the interior and exterior do-
mains for a closed curve C, respectively). Besides, it is desirable that we be able to
work with a "dilation" space and with a projection of it onto the model subspace.
Both of them should be consistent with Hardy-Smirnov spaces. Note that models
from [3, 4] do not meet all our requirements on a model.
In the present paper we construct linearly similar functional model for triples
of operators
(T,M,N) = ('P(To), MoX+(To), X-(To)No) (*)
where 'P E CM(ID, G+), the operators To, No, Mo are entries of an unitary node
210 = (~o ~~) E [H EB 1)1], 21~210 = I, 21021~ = I,

and x+ = y({l/('l/J+ 0 'P), x- = y(jl ('l/J- 0 'P). We assume that the curve C
is simple closed C1+E:- smooth, the unitary node 210 is simple (that is, the con-
traction To is completely nonunitary), and 'l/J, 1/'l/J E HOO(G+). (Note that
C1+E:- smoothness of the curve C implies 'P', 1/'P' E HOO(G+) [9]). In the paper
the class of all conformal maps of G 1 onto G 2 is denoted by CM(G 1 , G 2 ), the set
of all bounded operators acting in a Hilbert space H is denoted by [H]. A lin-
early similar model of triple (T, M, N) is understood to be a triple of operators
(T, M, N) such that there exists bounded invertible operator Wand
TW=WT, MW=M, N=WN.
Basic requirements for our variant of a functional model are: 1) possibility
to work with the "dilation" space and the projection onto the model subspace;
2) possibility to work with the dual model; 3) possibility to treat the absolutely
continuous subspace; 4) existence of simple explicit transformations between a
model of an operator and a model of a function of it; 5) possibility to construct
a model for a basic operator T and channel operators M, N simultaneously. Note
that, although these requirements arose in the process of solving of the duality
problem for spectral components [7], they are not specific for this problem only.
At the same time they are quite general and natural. Possibility to construct such
model and, of course, the construction itself is a matter of this paper.
The paper is organized as follows. In Section 1 we review the known construc-
tion of the S.-Nagy-Foia functional model [1] for which we use the coordinate-free
approach from [10]. In so doing, we consider features of the model that are essen-
tial for our generalization. In Section 2 the construction of the model is extended
to the case of a closed smooth curve. We employ certain "change of variable" from
the model for a curve to the model for the unit circle. Next, we consider the model
subspace Ke and the projection Fe onto this subspace and study their relation-
ships to the dual model and the "change of variable". In Section 3 we introduce
model operators T, M, N and find the transformations of them when we do the
"change of variable". Using these transformations we establish the main result of
the paper: triple (T, M, N) is a linearly similar model for certain triple (T, M, N)
Free Functional Model for Domains 221
of the form (*) and, conversely, for any triple (T, M, N) of the form (*) there ex-
ists a linearly similar model (T, M, it). The most part of presenting results was
announced without proofs in [7].
1. Model for the unit circle

Let lJ1, 'H be separable Hilbert spaces and let 1fo E [L2(1I', lJ1), 'H] be mappings
satisfying the conditions:
(i) 0 1fo 1fo = I ;
(11.. )01 (1fo_1fo+)z = z(1fo_ 1fo+);
(iii)o Ran 1fo+ V Ran 1fo- = 'H ,
where P+ and P_ are orthogonal projections on the Hardy spaces H2(lJ1) and
H:'(lJ1), respectively. We set 8~ = 1fO'f1fO and ~~ = (I - 8~8~f)1/2. It is
easy to show that 8~, ~~ E LOO(1I', [lJ1]). (Usually, we employ one and the same
symbol for an operator-valued function and for the operator of multiplication by
it.) Moreover, operator-valued functions 8~ admit analytic continuation to JD)
and have contractive values there. This follows from (ii)o and the self-adjointness
of projections P. Isometries TO : clos Ran ~~ f--t 1i are uniquely determined
by the relations TO ~~ = (I - 1fo 1fo )1fO'f . We extend them to (Ran ~~).l by
O. For them, we have TO = ((~~)-l(1fO'f - 8~1fo))* E [L2(1I', lJ1), 1i] and the
following identities hold:
1) (1fZ) (1fo, TO)

TO
= (~ pO)
L'l.g'
clos Ran
;
2) 8P
~ri ).,
- 0 clos Ran L'l..f
3) 1fo1fo + TOTO = I.
We define also the dual mappings 1f*0 = -i1fO'fCl/z, where (C1/zJU)(z) =
(i/z)J(l/z). It is easy to check that the maps 1f*0 satisfy conditions (i)o, (ii)o,
(iii)o and we have also 8;0 = (8~)~, ~;o = (~ri)~, where A~(z) = A(z)*.
The following duality relations hold
(f, g)H =< 1foJ, 1f:O'f g > 1[' + < ToJ, T:O'f g > 1[', J, g E 1i ,
1 )
where
< u, v > 1['= -1. (u(z, V(Z))91 dz, U E L 2 (1I', lJ1), vEL 2 (1I', lJ1) .
21ft 1['
Further, from the identity 111f0+ ZU+ + 1fo- ZU_ II = II1fo+ u+ + 1fo- u_ II and the
condition (iii)o it follows that there exists a unique unitary operator Uo E [1i]
with absolutely continuous spectrum such that Uo1fo = 1foZ . It is easy to check
the following identities UOTo = TOZ , 1foUo = Z1fo , TOUO = ZTo .
222 A. Tikhonov
Define the orthogonal projections Qo = 7fOP7fo and subspaces Do =

Ran Qo. It is easy to check that QoQO'f = 0 and that the subspaces Do
are invariant under operators UtI. Then we arrive at the Lax-Phillips-Adamyan-
Arov scheme [11]:
Do+1- D o-, nUtnDo = {O},

nEfiI
UUoDo+ V UUoDo- = 'H.
nEZ nEZ
It is clear that, conversely, one can easily construct operators 7fo that satisfy
conditions (i)o, (ii)o, and (iii)o whenever the corresponding outgoing and incoming
subspaces Do are given.
As is well known [11], the generalized Lax-Phillips scheme is equivalent to
the functional model of S.-Nagy-Foi~ [1]. For this model, the original Nikolski-
Vasyunin [10] properties of operators 7fo are:
7fo 7fo = I ; 7foZ = Uo7fo ;

7foH2(]J))) = Do ; Ran 7fo+ V Ran 7fo- = 'H ,
where Uo E ['H] is a minimal unitary dilation of some completely nonunitary
contraction To E [H], Do+ = (Vn~oUoH) e H, Do- = 'H e (Vn~oUOH).
Therefore our set of axioms (i)o, (ii)o, (iii)o and Nikolski-Vasyunin's axioms
are equivalent. But in our settings we use neither a unitary dilation nor orthogonal
complements. This reformulation enables us to extend the functional model to the
case of a closed curve. We present this generalization in the next section.
Now we define model operators ib E [Ko] , Mo E [Ko, m], No E [m, Ko] :
Tol = Uol - 7fo+Mol, Mol = (7f~+Uof)(oo), Non = (I - 7fo+P+7f~+)7fo_n,
where IE Ko, n E m, Ko = RanPo , Po = 1- Qo+ - Qo-. Note that operators

-go and No are (up to unitary factor) the defect operators [1, 2] of the contraction
To, the operator Po is an orthogonal projection and
~o = 1E
( Mo No
et(O)*
)
is a simple unitary node.

The inverse is also true. If ~o E [H EB m] is a simple unitary node, then there
exists a pair ITo = (7fo+, 7fo-) satisfying conditions (i)o, (ii)o, (iii)o, and a unitary
operator Wo E [H, Ko] such that
ToWo = WoTo, -
MoWo = Mo, No = WoNo, et(o)* = Lo
Note also that for dual objects we have the following relations
P*o = Po, K*o = Ko, T*o = T~, M*o = N;, N*o = M;.
2. Model for a curve (geometry)

Since the curve C is C1+c-smooth, there exist (non-orthogonal) projections P E
[L 2(C,5)1)J such that RanP = E2(G, 5)1), Ker P = E2(G~, 5)1). Let mappings
1l" E [L2(C, 5)1), ltJ satisfy conditions:
(i)1 V'Ij; E Loo(C, [5)1]) (1l"1l")'Ij; = 'Ij;(1l"1l"); (i)2 (1l"1l")-1 E [L 2(C,5)1)J;
(ii)1 (1l"~1l"+)z = z(1l"~1l"+); (ii)2 P-(1l"~1l"+)P+ = 0;
(iii) Ran 1l"+ V Ran 1l"_ = 1t.
Here 1l" ~ is the Moore-Penrose inverse operator for 1l" _: 1l"~ f = (1l" _I (Ker 1l" _ ) J. ) -1 f,
f E Ran1l"_; 1l"~f = 0, f.lRan1l"_.
Conditions (i) , (ii), and (iii) are generalization of (i) 0, (ii) 0, and (iii) to
the case of a curve. Indeed, we have 1l"L = 1l"o_ . On the other hand, there ex-
ists a transformation ("change of variable") between 11 = (1l" +, 1l" _) and 110 =
(1l"0+, 1l"0- ), which we are going to describe.
First, note that from conditions (i)1 and (i)2 it follows that 1l"1l" = bI,
where b,l/b E Loo(C). Then there exist [1, 12J outer analytic scalar func-
tions 'Ij; E Hoo(G+) such that for their boundary values 1'Ij;12 = b. Clearly,
1/'Ij; E Hoo(G+). Note also that, since b do not vanish on '][', we have 'Ij; E
C1+c(closG+) {:} b E C1+c(C) (see, e.g., [13]). Define the unitary operator
C<.p21 E [L2(C2, 5)1), L2(C1, 5)1)J by formula
(C<.p2J( ))(zd = J'P~1(Z1)f('P21(Z1))' Z1 E C1 , f E L2(C2,5)1) ,

where 'P21 E CM(G1+, G2+). Let 'P E CM(JD), G+). Put 1l"0 = 1l"1/'Ij;C<.p-1 .
We will check that 110 = (1l"0+,1l"0-) satisfy conditions (i)o, (ii)o, and (iii)o.
It is easy to show that operators 1l"1/'Ij; are isometries. Hence, 1l"0 sat-
isfy the condition (i)o. Obviously, 1l" = 1l"OC<.p'I/J. Since Ker1l" = Ker1l"O =
{O}, we have 1l"1 = 1/'Ij;C:p11l"6 = 1/'Ij;C<.p-11l"o E [1t, L2(C, 5)1)J. Whence,
1l"o = c<.p'lj;1l"l and 1l"o_1l"o+ = C<.p'lj;_1l"~1l"+1/'Ij;+C<.p-1. Taking into account the
conditions (ii)1 and (ii)2 we can regard the operator 1l"~ 1l"+ as an operator
of multiplication by boundary values of bounded analytic operator-valued func-
tion e+(z), z E G+. Since 'I/J, 1/'I/J E Hoo(G+) , the operator-valued function
et = ('Ij;_/'Ij;+e) 0 'P is bounded and analytic in the unit disk. Then the con-
ditions (ii)~ and (ii)~ are fulfilled for the operator 1l"o_1l"o+ of multiplication by
boundary values of et(zo), Zo E JD). The condition (iii)o follows from the identity
Ran 1l"0 = Ran 1l".
Note that one can consider slightly more general "change of variable" 1l"2 =
1l"lC<.p21"l, where 'P21 E CM(G1+,G2+), "l, 1/"l E Hoo(G2+). Then the pair
112 = (1l"2+,1l"2-) satisfy conditions (i), (ii), (iii) {:} the pair 111 = (1l"1+,1l"1-)
satisfy the same conditions.
We set e = 1l"~1l", b, = (I-ee~)1/2. It is easy to show that e, b, E
Loo(C, [5)1]). As we already know, e+ E Hoo(G+, [5)1]). However, now we cannot
224 A. Tikhonov
assert that the operator-valued function 8- = 7r ~ 7r _ admits analytic continuation

to G_.
Further, put T = ((~)-1(7r~ - 87ri))t . It is easy to check that T =
ToC'P'lj;~, T1 = 1/'lj;~C'P-1TJ. Combining this with the corresponding relations
for 7ro, TO from Section 1, we obtain
Proposition 1. One has
1) ( 7ri ) (7r, T)
Tt
1 0
= ( 0 PclasRan6 );
2) ( 7ri )
T1
(7r~, T~) = (
8~
~
~~
-8 Pclas Ran 6'1' }

3) 7r7ri + TT1 = 1.
We set 7rd = 7r*oC'P~l/'lj;':f. It is clear that II* = (7r*+,7r*_) satisfy conditions
(i), (ii), (iii) and 8; = (8)~, ~; = (~~)~. Here again A~(z) = A(z)*.
Proposition 2. One has
(j, g)H = (7rif, 7r!~g)c + (T1f, Tl~g)c, f, g E H,
where
(u, v)c = 2~i J
c
(u(z), v(z)ht dz, u E L2(C, 1)1), v E L2(0, 1)1).
Proof Using obvious properties of the pairing, we obtain

(7rof, 7r:o~g) 11' = (ccp'lj;7rif, Ccp~ 1N7r!~g) 11'
= ('lj;7rif, l/'lj;7r!~g)c = ((l/'lj;)'lj;7rif,7r!~g)c = (7rif, 7r!~g)c.
Similarly, we get the corresponding identity for T , T*~ . It remains to make use
of the relations of duality for the unit circle. D
Proposition 3. Conditions U 7r = 7r z uniquely determine the normal operator U.
The spectrum of U is absolutely continuous and lies on the curve C.
Proof Since 7rZ = 7roCcp'lj;z = 7ro<p(zo)Ccp'lj; = <P(Uo)7roCcp'lj; = <p(Uo)7r,
we have U = <p(Uo). D
In the same way, we get UT = TZ, 7riU = z7ri, T1u = zT1. From the duality
relations we obtain U* = U* .
We pass to the model subspace and the projection onto it. To this end we con-
sider auxiliary projections q = 7rp7ri. These operators have the following
properties.
Lemma 4. 1) qi = q; 2) Kerq = Ker P7ri; 3) Ranq = 7rE2(G) =

Ker p~7ri nKerT1; 4) qd = q~; 5) q2-ql+ = q2-(I - q1-) = (1 - q2+)ql+ = 0,
where 7r2 = 7r1CCP21 'f], <P21 E CM(Gl+' G2+), 'f], 1/rt E HOO(G2+) .
Proof. This technical lemma bases mainly on the condition (ii)2. We have
1) d=q~~q~~=q~~~=q~~=~.
2) This follows from identities q = 7r(P7rl), P7rl = 7rlq.
3) Let 1 Ranq. Then 1 = qg = 7rU, U = P7rlg. Conversely, let
E
P~u = o. Then 7rU = 7r(P + P~)u = 7rPU = (7rP7rl)7ru.
Whence, Ranq = 7rE2(G).
Next, since p~7rl7r = P~, Tl7r = 0, we have 7rE2(G) C Ker P~7rlnKerTl.
Conversely, let 1 E Ker p~7rl n KefTl. Then 1 = (7r7rl + TTl)1 = 7r7rl1 =
7r(P + p~)7r11 = 7rU, U = P7r11 E E2(G). Whence, 7rE2(G) =
Ker p~7r1 n KefTl.
4) Let 1 E Ranq, g E Kerq*~. Then, by Proposition 2, we have (j,g) =<
7r11, 7rl~g >c, where 7r11 E E2(G), 7rl~g E E2(G d ). Taking into ac-
count that E2(G)<~> = E 2(Gd) , we get (j,g) = o.
Conversely, let g..lRanq. If we take 1 = 7rU, U E E 2(G) , then 0 =
(j,g) =< 7r11,7rl~g >c=< U,7r!~g >c . Whence, 7rl~g E E 2(Gd) , i.e.,
g E Kerq*~. Therefore, (Ranq)~ = Kerq*~. Similarly, (Kerq)~ = Ranq*~.
On the other hand, for any linear operator (Kerq)~ = Ranq and (Ranq)~ =
Ker q . Since q*~ and q are projections with the same ranges and kernels, we
obtain q*~ = q .
5) Let A = 1/TJC:;;21, . Then we have 7rl = 7r2A and P2_AP1+ = o.
Using the latter observation and the obvious identity ql-q1+ = 0, we get
q2-(I - ql-) = 7r2-P2-7rL(I - 7rl-Pl-7rL)

= 7r2-P2-A_7rL(I - 7rl-Pl-7rL)
= 7r2-P2-A-(I - P 1 -)7rL = 7r2-(P2-A_P1+)7rL = 0,
(I - q2+)q1+ = (J - 7r2+P2+7r~+)7r1+P1+7r!+
= (I - 7r1+A+l P2+A+7r!+)7r1+ P1+7r!+
= 7r1+(I - A+l P2+A+)P1+7ri+
= 7r1+A+l(P2_A+P1+)7r!+ = 0,
q2-q1+ = q2-ql-q1+ + q2-(I - ql-)q1+ = 0 + 0 = o. o
We set Q+ = q+(I - q_), Q- = q_ and D = RanQ. Since Q+q+ = q+, we
have RanQ = Ranq. It is easy to check that Q1 = Q, QQ~ = o. Now we
can define the projection Pe onto the model subspace Ke = Ran Pe :
This projection plays a central role in the paper and have the following properties:
226 A. Tikhonov
Theorem 5. 1) P~ = Pe; 2) P*e = P e ; 3) P3eP2eP1e = P3eP1e, where

7r3 = 7r2CCP32 'f/3, 7r2 = 7r1CCP21 'f/2 .
Proof. 1) This follows from the properties of Q .
2) Using Lemma 4(4), we get
P*e = (1- q*+)(l- q*_) = (1 - q:')(l- q~) = ((1 - q+)(l- q-))* = Pe.
3) Using Lemma 4(5), we get
(1- q2-)(1 - q1+)(l - q1-) = (1- q2- - q1+)(l- q1-)
= ((1 - q1+) - q2-)(1- q1-) = (1- q1+)(l - ql-),
(1- q3+)(1 - q3-)(1 - q2+) = (1- q3+)(l - q3- - q2+)
= (1- q3+)((1- q3-) - q2+) = (1- q3+)(1 - q3-). 0
Remark. We can rewrite statement 3) in the form Z31 = Z32Z21 , whereZij =
PielKje E [Kje,K ie ]. In particular, we have Z2i1 = Z12. Note that the
model
subspace Ke varies depending on "change of variable". In general, K 1e =F K 2e .
e
Remark. In contrast to the case of the unit circle, we have P =F Pe, Q + =F q+ ,
and Pe = (1 - q+)(l- q_) =F (1 - q_)(l- q+) = (1 - q+ - q-).
Remark. Besides, we have '</'IjJ E HOO(G) : (I - Q)'IjJ(U))Q = o. That
means, the subspaces D are invariant under operators 'IjJ(U) . But, on the other
hand, we have only '</'IjJ E HOO(G+) : Q_'ljJ(U)(I - Q_) = o. Thus we have some
asymmetry here as well as in the previous remark.
Remark. Nevertheless, the model subspace Ke admits the following symmetric
description:
Ke = {f E 1t : 7rtf E E2(G_, sn), 7r~f E E2(G+, sn)} = Ker P+7rt n Ker P_7r~.
Indeed, let 7rtf E E2(G_, sn), 7r~f E E2(G+, sn). Then we get Pef = (I -
7r+P+7rt)(I -7r-P-7r~)f = (I -7r+P+7rt)f = f and f EKe Conversely, we have
P+7rt Pe = P+7rt(I - 7r+P+7rt)(I - 7r_P_7r~) = P+P_7rt(I - 7r_P_7r~) = 0
and P-7r~Pe = P_7r~(I - 7r+P+7rt)(I - 7r_P_7r~) = P_7r~(I - 7r_P_7r~) =
P_P+7r~ = o.
Example. Let 8 E HOO(G+, [sn]), sup 118(z)-111 < 00. Then, for the space
zEC
1t = L2(C, sn) and the mappings 7r+ = I, 7r_ = 8- 1 , we have
Ke = {f E L2(C, sn) : f E E2(G_, sn), 8f E E2(G+, sn)}
{f E E2(G_, sn) : 8f E E2(G+, sn)} = Ker P+ n Ker P_8.
Thus we arrive at Yakubovich's model [4] for the class of Coo operators. In [4]
Yakubovich considers a more general case where the domain G+ is multiply con-
nected and the curve C is less smooth. Note that the most part of presenting
above results can be extended to this case. But, in contrast to [4], we do not
restrict ourself to the case of Coo operators and allow them to have absolutely
continuous spectrum. Another difference is the presence of the projection Pe in
our model. This projection enables us to lift various objects (e.g., spectral compo-
nents, commutant, etc.) from the model space Ke up to the level of the dilation
space H.
Our triples of operators correspond to Yakubovich's 3-systems. In [4] Yaku-
bovich studies the problem: how to construct a model for a given 3-system. He has
solved this problem, but under the assumption that a characteristic function is
known (more precisely, it must belong to some family of operator-valued functions
corresponding to the transfer function of a 3-system). In the next section we also
study this problem for the case of simple connected domains. We do not assume
that the characteristic function e+ is given. But we suppose instead that the
functions X are known. Such a statement of the problem justifies in applications
of this result to the study trace class perturbations of normal operators [7].
3. Model for a curve (operators)

First we introduce model operators:
Tf = Uf - 7r+Mf, Mf = ~
27rz
J(7r~f)(Z) dz = (7r~Uf)(oo),
C
Nn = Pe7r-n = (I - 7r+P+7r~)7r_n, where f E K c,,>, n E l)1.
Theorem 6. The mapping : 'ljJ f---+ Pe'ljJ(U)IKe, 'ljJ E HOO(G+) is a homomor-
phism from the algebra HOO(G+) into the algebra [Ke]. This homomorphism has
the following properties: 1) <1>(1) = I, (z) = T; 2) 11<1>11 < 00;
3) II'ljJnll < K, lim 'ljJn(z) = 0 a.e. z E C ===} s- lim ('ljJn) = o.
n~~ n~~
Proof. Using notation 'ljJ(T) = <1>( 'ljJ), we check multiplicativity:

'ljJl (T)'ljJ2(T)f = Pe'ljJl (U)Pe 'ljJ2(U)f = Pe'ljJl (U)'ljJ2(U)f
- Pe 'ljJl(U)(I - Pe )'ljJ2(U)f = Pe'ljJl(U)'ljJ2(U)f
- (I - Q_)(I - Q+)'ljJl(U)(Q+ + Q-)'ljJ2(U) (I - Q_)(I - Q+)J
= Pe('ljJl'ljJ2)(U)f = ('ljJl'ljJ2)(T)f, f EKe.
Above we have exploited identities (I-Q+)'ljJl(U)Q+ = 0, Q_'ljJ2(U)(I-Q_) = O.
Now we pass to calculation of (z):
(z)f PeUf = (I - 7r+P+7r~)(I - LP-7r~)uf
(I -7r+P+7r~)(Uf - LP_Z7r~f) = (I -7r+P+7r~)uf
U f - 7r+(Z7r~f)(oo) = Tf, f EKe
Properties 2) and 3) follow from the corresponding properties of the functional
calculus for normal operators. 0
228 A. Tikhonov
The following assertion answers how the model operators depend on "change of
variable" .
Proposition 7. Let 7r2 = 7rlCp"7, <p E CM(G H , G2+), "7, 1/"7 E HOO(G 2+).
Then
where Z = P 2s1K 1s, X+ = #/("7+ 0 <p), X- = # ("7- 0 <p).

Proof. First we note that Z-l = P 1s1K 2s. We shall also use the identities
P2sQH = 0, Q2-PIS = 0, which follow easily from Lemma 4(5).
1) By Theorem 6 and Proposition 3, we have
T 2ZP1S - Z<p(TdP1s = P2S(U2P2S - P1s<p(ud)P1S =
P2s(<p(Ud(I - Q2+ - Q2-) - (I - QH - Ql-)<p(U1))P1S =
P2S ((QH + Ql-)<p(U1) - <p(U1)(Q2+ + Q2-))P1S.
Since (I - Q2+)U2Q2+ = 0, Ql-<p(Ud(I - Ql-) = 0, we get
T2ZP1S - Z<p(T1)P1S = P2S (QH<P(Ul) - <P(U1)Q2-)P1S .
Finally, using above mentioned identities we obtain T2Z - Z<p(T1) = o.
2) Let! E K 2S . Then
M2! 1 . J(7r~+!)(Z2) dZ2 = -2

= -2 1 . J ~() (7ri+J) (<p-l (Z2)) J<P 1(Z2)' dZ2
7rZ 7rZ "7+ Z2
C2 C2
- 1 J J<pl(zd t - 1 J t
- -2. ( ( )) (7r H J) (Zl) dZ1 - -2 x+(zd(7r H J)(Zl) dZ1
7rZ "7+ <p Zl 7rZ
C1 C1
= 2~i J (7ri+X+(Ul)J)(Zl) dZ1 = 2~i J (7ri+P1sx+(Ul)P1S J)(Zl) dZ1

C1 C1
+ 2~i J (7ri+((I - P1s)x+(udP1s! + X+(Ul)(I - P1S)J))(Zl) dz1.

C1
Taking into account that QI-X+(U1)(I -Ql-) =0 and PIS = (I -Ql-)(I -QH)'
we get
(I - P1S)X+(Ul)P1S = (QH + Ql-)x+(Ul)P1S = QHX+(Ul)P1S.
Whence, by the Cauchy theorem,
J (7ri+(I - P1s)x+(udP1SJ)(zd dZ1 = o.

C1
Next, since Q2-PIS =0 and (I - QH)x+(Ul)QH = 0, we get

X+(Ul)(I - PIS)! X+(Ul)(QH + Ql-)P2S!
= X+(Ul)Q H P2S! = QHX+(Ul)Q H P2S!.
Then, again by the Cauchy theorem, we have
c,
J (-rrI+X+(UI)(I - PIe)J)(ZI) dZ I = o.
3) Let n E 1)1. Then

N 2n = P2e 7r2- n = P2e7rl-x-(zl)n = P2e X-(UI) 7rI-n
= P2e(PIeX-(UdPIe + PIex-(Ud(I - PIe) + (I - PIe )x-(Ud)7rI-n.
Taking into account that (I - Q1+)x-(UdQ1+ = 0, we get
PIex-(Ud(I - PIe) = (I - Ql_)(I - Q1+)x-(Ul)(Q1+ + Ql-)
= P1eX-(UI)Ql-.
Since QI-7rl-n = 0, we get P 1eX-(U1)(I - P1e)7rI-n = o.
Further, since P 2e Q1+ = 0 and QI-x-(Ud(I - Ql-) = 0, we obtain
P2e (I - P1e)X-(UI) = P2e(Q1+ + Ql-)X-(Ul)
= P2eQI-X-(Ud = P 2 eQI-X-(UdQI- .
Again, since QI-7rl-n = 0, we get P 2e (I - P1e)X-(UI)7rI-n = o. Therefore,
o
Applying this proposition for "change of variable" 7r = 7roCcp1/J , we get that
the triple (T, M, N) is a linearly similar model for a triple (T, M, N) of the
form (*). So, we have proved the following
Theorem 8. For any pair II = (7r+,7r_) satisfying conditions (i), (ii), (iii) there
exist an operator W E [H, Ke] and a triple (T, M, N) of the form (*) such that
W- 1 E [Ke,H] and TW=WT, MW=M, N=WN.
Inverse is also true.
Theorem 9. Let (T, M, N) be a triple of operators of the form (*). Then there
exist pair II = (7r+, 7r_) satisfying conditions (i), (ii), (iii) and operators W, W* E
[H, H] such that WW: = Pe, W:W = I and TW = WT, MW = M, N = WN;
T*W* = W*T*, M*W* = N*, N* = W*M*.
Remark. Here the triple (T*, M*, N*) corresponds to the dual pair II* = (7r*+, 7r*_).
Note that the following identities hold
(Tf,g) = (j,T*g) , (Mf,n) = (j,N*n) , (Nn,g) = (n,M"g) ,
where f E Ke, g E K*e, n E 1)1 .
230 A. Tikhonov
Proof. Since (T, M, N) is a triple of the form (*), there exists a simple unitary
node 210 such that (T, M, N) = (cp(To), MoX+(To), X-(To)No). For 210 , there
exists (see Section 1) a pair ITo = (1fo+,1f0-) such that the node 2to is unitarly
equivalent to the node 21 0 . Let Wo be a unitary operator that realizes this equiv-
alence. We take "change of variable" 1f = 1foCcp and 1fd = 1f*0Ccp~1/';f.
Applying Proposition 7, we have
Cr, M, N) (Zcp(To)Z-\ MoX+(To)Z-\ ZX-(To)No),
(T*, M.., N*) (Z*cp~(T*O)Z;l, M*oX~(T..O)Z;l, Z*X,+(T*o)N*o),
where Z = PelKo, Z* = PelKo, and Ko is the model subspace for ITo. Using
relations from Section 1, we get
(T*, M*, N*) (Z*cp(To)* Z;\ NO'X_(To)* Z;\ Z*X+(To)*MO').
Let W = Pe VWo, W* = P VWo, e where~ V E [Ko, HLis embeddi~g: V f
f, f Then it is easy to check that TW = WT, MW = M, N = WN;
E Ko
T*W* = W*T*, M*W* = N*, N* = W*M*. Next, we have W: = WO'V*Pe.
Whence, using Theorem 5(3), we get
W:W= WO'V*PoPePoVWo = WO'V*PoVWo = WO'V*VWo = I,
WW: = Pe VWo WO'V* Pe = Pe VV* Pe = PePoPe = Pe. 0
Remark. The relation W: W = I can be rewritten in the form (W* f, W g)
(f,g). Thus here we have the same type of the duality as in [4, 14].
Remark. One can also consider the transfer function Y(z) = M(T - Z)-l N. It
is interesting to note that the transfer function Y(z) uniquely determines the pure
part of the characteristic function e+(z) provided the functions X are known.
References
[1] Szokefalvi-Nagy B., Foia C., Harmonic analysis of operators on Hilbert space. North-
Holland, Amsterdam-London, 1970.
[2] Brodskiy M.S., Unitary operator nodes and their characteristic functions, Uspehi
mat. nauk 33 (1978), no. 4, 141-168.
[3] Abrahamse M.B., Douglas R.G., Operators on multiply-connected domains. Proc.
Royal Irish Acad., 74 A (1974), 135-14l.
[4] Yakubovich D.V. Linearly similar model of Sz.-Nagy - Foias type in a domain.
Algebra i Analiz, 15 (2003), no.2, 180-227.
[5] Naboko S.N., Functional model for perturbation theory and its applications to scat-
tering theory. Trudy Mat. Inst. Steklov 147 (1980), 86-114.
[6] Makarov N.G., Vasyunin V.I., A model for noncontraction and stability of the con-
tinuous spectrum. Lect. Notes in Math., 864 (1981), 365-412.
[7] Tikhonov A.S., Functional model and duality of spectral components for operators
with continuous spectrum on a curve. Algebra i Analiz, 14 (2002), no.4, 158-195.
[8] Duren P.L., Theory of HP spaces, Pure Appl. Math., vol. 38, Academic Press, New
York-London, 1970.
[9] Pommerenke C., Univalent functions. Vandenhoek and Ruprecht, Gottingen, 1975.
[10] Nikolski N.K., Vasyunin V.I., Elements of spectral theory in terms of the free func-
tional model. Part I: Basic constructions, Holomorphic spaces (eds. Sh. Axler, J. Mc-
Carthy, D. Sarason), MSRI Publications 33 (1998), 211-302.
[11] Adamyan V.M., Arov D.Z., Unitary couplings of semi-unitary operators. Mat. Issled.,
1 (1966), no.2, 3-64.
[12] Garnett J.B., Bounded analytic functions, Pure Appl. Math., vol. 96, Academic
Press, New York-London, 1981.
[13] Shirokov N.A., Properties of modulus for analytic functions smooth up to the bound-
ary. Dokl. Akad. Nauk SSSR, 269 (1966), no.6, 1320--1323.
[14] Yakubovich D. V., Dual piecewise analytic bundle shift models of linear operators.
J. Func. Anal., 136 (1996), no.2, 294-330.
Alexey Tikhonov
Mathematical analysis division
Taurida National University
Yaltinskaya str., 4
Simferopol 95007 Crimea
Ukraine
e-mail: tikhonov@club.cris.net
Operator Theory:
Jacobi Block Matrices with

Constant Matrix Terms
Marcin J. Zygmunt
Abstract. We investigate a solution of the difference equation

tU~,B(t) = AU;;+~(t) + BU~,B(t) + AU;;-'-~(t)
with the boundary conditions U()-,B = I, U:: 1B = 0, where A,
B are hermit-
ian matrices. U;;,B are usually called matrix Chebyshev polynomials of the
second kind. The above equation cannot be easily simplified as in scalar case
because A and B do not need to commute. However we are able to compute
spectrum of the corresponding orthogonality measure which is very important
to investigate discrete Schrodinger operator related to U;;,B.
Mathematics Subject Classification (2000). Primary: 47B36 secondary: 39A70,
39B42.
Keywords. Chebyshev polynomials, discrete Schrodinger operator, Jacobi
block-matrices, matrix orthogonal polynomials.
Let C NxN denote a space of all quadratic N x N matrices. We will write A 2: 0

(or A > 0 respectively) if A is a positive definite (or strictly positive respectively)
Hermitian matrix. In the following the inequality A 2: B (or A > B respectively)
will be equivalent to A - B 2: 0 (or A - B > 0 respectively) for A, B E C NxN .
Denote by .e2 (C NXN ) a space of sequences X = (XO,X1 , ... ) of matrices from
00
C NXN , for which the series L Xn * Xn converges.
n=O
We introduce an "NxN-matrix" product on .e2 (C NxN ):
00
((X!Y))i2 = LYn * Xn E C NXN .

n=O
We have ((XIX))c 2 2: 0 in the sense stating before. Then a system
En = (0, ... ,0,1,0, .. . ),
The author was supported by KBN's grant 2P03A00723 and RTN network: HPRN-CT-2002-
00279.
234 M.J. Zygmunt
where I denotes the identity of C NXN , appears only in the nth position, forms an
"orthonormal" basis of fi2(C NXN ).
In a similar way we can define an L2-space of square integrable matrix-valued
functions. Let I: be a positive matrix-valued Borel measure, i.e., I:(~) is positive
definite (I:(~) ~ 0) for all Borel subsets ~ c R For matrix-valued functions F(x)
and G(x) we define an "NxN-matrix" product
((FIG))2; = J F(x)dI:(x)G(x)* E C NxN .
IR
Now the space L2(I:) consists of all matrix-valued functions F(x) for which
((F!F))2; is convergent. More details can be found in [AN], [B], [Dl-3], [DV]
and [Z2].
Let A, B E 2(C NXN ) be Hermitian matrices and let J A,B be an operator on
2(C NXN ) acting as follows:
(JA,BX)O = BXo + AXI , (JA,BX)n = AXn- 1 + BXn + AXn+ I .
Hence JA,B = BY + A(S + S*), where S denotes the "shift" on 2(C NXN ) and Y
- the identity operator. With J A,B there are associated matrix-valued Chebyshev
polynomials of the second kind U ,;;,B
(x), i.e., polynomials satisfying the recurrence
formula
XU,;;,B(X) = AU:+~(x) + BU,;;,B(x) + AU:~~(x).
Denote by Mn the n-th moment of WA,B
Mn = J xndWA,B(X) = ((xn III))wA,B

IR
Theorem 1. Let W A,B be an matrix-valued measure which orthogonalizes polyno-

mials U,;;,B, i. e.,
((U,;;,BIU~,B))WA,B = J U,;;,B(x)dWA,B(x)U~,B(X)* = 6n ,m I .
IR
Then the moments Mn of the measure WA,B are equal to

2
Mn = ~JJ4-X2(At+Btdt
27r
-2
Moreover
SUppWA,B = U u(At + B) .
tE[-2,2]
Proof. We have
Let
n
(B + Ax)n = L Ck,n Xn .
k=O
Then
(((JA,B)nEoIEo))2 = (((BI + A(S + S*)t EoIEo))2
n
= I: Ck,n(((S+S*)kEoIEo))2'
k=O
On the other hand SEn = E n+ 1 , hence the behavior of S is the same as of the
shift operator 8 on 2(N). Hence ((En lEo)) 2 = r5n,oI = (enleo)I, where {en} nEJII
so
(((S + s*)n EoIEo))2 = ((8 + 8*)neoleo)I.
The operator 8 + 8* is a well-known discrete Schrodinger operator (related to
the classical Chebyshev polynomials of the second kind), which spectrum is equal
to [-2,2] and its spectral measure w is equal to dW(x) = 2~ V4 - x 2 dx. Hence
f
2
((8 + 8*) ne oleo) = xnw(x)dx.

-2
Thus
2 n
Mn = J w(t) I: Ck,n tn dt
-2 k=O
2
= J w(t)(At + B)ndt
-2
which proves the first part of the theorem.
To prove the second part note that we have the equality
1 f
2
p(x)dWA,B (x) = w(t)p(At + B)dt (1)

-2
for every polynomial p E <C[x]. Let now ~ = U cr(At + B). ~ is a compact

tE[-2,2]
subset of the real line R By polynomial approximation we have
1 f
2
f(x)dWA,B(x) = w(t)f(At + B)dt (2)

-2
for every continuous function f E C(~). Now it is not to hard to see that the
support of W A,B is equal to ~. 0
Corollary 1.1. The support of the measure WA,B is equal to
supp WA,B = U cr(At + B).

tE[-2,2]
236 M.J. Zygmunt
Moreover if the matrix A has non-zero determinant (i.e., it is invertible), the mea-
sure WA,B is absolutely continuous with respect to the Lebesgue measure multiplied
by the identity matrix I.
Corollary 1.2. The spectrum of J A,B is equal to a (JA,B) = supp w A,B and con-
sists of at most N non-degenerate intervals of the real line ~.
Proof. By Theorem 2.4, Ch. VII, [Ber], the spectrum of the operator J A,B is equal
to the support of the measure W A,B. The second statement of the corollary holds
because of the continuity of spectrum. D
Theorem 2. Let U(t)A(t)U(t)*, where U(t) is unitary and A(t) diagonal matrix,
be the spectral decomposition of At + B. Then
J
2
WA,B(E) = w(t)U(t)XE (A(t)) U(t)*dt

-2
for every Borel subset E c ~.
Proof. Let
At + B = U(t)A(t)U(t)* . (3)
Putting (3) into (1) gives
J J J
2 2
p(x)dWA,B(x) = w(t)p(U(t)A(t)U(t)*)dt = w(t)U(t)p(A(t))U(t)*dt

IR -2 -2
for every polynomial p E C[xJ. Now by approximation of the characteristic func-

tions of given Borel subset E c ~ we get the thesis. D
Example 3. This example shows that the assumption on hermitianity of A cannot
be omitted if we want to save the absolutely continuity of the orthogonalizing
measure.
Let us consider the polynomials Pn(x) which satisfy the recurrence formula:
tPk(t) = APk+l (t) + A* Pk-l (t),
where
A=(O 0b) '
a
and a > b > O. Let
v = V* = (~ ~)
Define now A(t) = Vk Pk(t). Polynomials A are still orthonormal with respect to
the same matrix-valued measure as polynomials Pk, moreover Po = id, P- 1 = O. It
can be easily verified that polynomials Pk satisfy the following recurrence formula:
tP2k (t) = A 1 P2k+1 (t) + A 2P2k- 1 (t)
tP2k + 1 (t) = A 2P2k+2(t) + A 1 P2k (t)
where
Ai = V2k AV2k+1 = (~ ~), A2 = V2k AV 2k - i = (~ ~).

Hence
A = (p; ~)
where polynomials Pk, qk satisfy
tP2k(t) = bP2k+i(t) + ap2k-i(t), tP2k+i(t) = ap2k+2(t) + bp2k(t),
tq2k(t) = aq2k+i(t) + bq2k-i(t), tq2k+i(t) = bq2k+2(t) + aq2k(t).
Thus, the orthogonality measure for polynomials Fk (and so for P k ) is diago-
nal, where entries on the diagonal are scalar measures /-lp and /-lq orthogonaliz-
ing polynomials Pk and qk respectively. It was shown in [Ch] that in the case
a > b the support of the measures /-lp and /-lq consists of the union of the intervals
[-(a + b), -(a - b)] U [a - b; a + b], where both measures are absolutely continu-
ous. Moreover, /-lp has a non-zero atom at 0 with the weight 1- b2 ja 2 . This shows
that the matrix-valued measure orthogonalizing polynomials Pk has a non-zero
atom at 0, with the weight equal to
References
[AN] A.I. Aptekarev and E.M. Nikishin, The scattering problem for a discrete Sturm-
Liouville operator, Mat. Sb., 121(163):327-358; Mat. USSR Sd., 49:325-355, 1965.
[B] Y.M. Berezanski, Expansions in eigenfunctions of self-adjoint operators. AMS,
Providence, 1968.
[Ch] T. Chihara, An Introduction to Orthogonal Polynomials, Gordon & Breach Sc. Pub.,
New York, 1979.
[D1] A.J. Duran, A generalization of Favard's theorem for polynomials satisfying a re-
currence relation, J. Approx. Theory, 74, 83-109, 1993.
[D2] A.J. Duran, On orthogonal polynomials with respect to a positive definite matrix
of measures, Ganad. J. Math., 47(1):88-112, 1995.
[D3] A.J. Duran, Ratio asymptotics for orthogonal matrix polynomials, J. Approx. The-
ory, 100(2):304-344, 1999.
[DV] A.J. Duran and W. Van Assche, Orthogonal matrix polynomials and higher-order
recurrences, Linear Algebra Appl., 219:261-270, 1995.
[Sz] G. Szego, Orthogonal Polynomials, AMS ColI. Pub., vo1.23, AMS, Providence, 1975
(4th edition).
[Zl] M.J. Zygmunt, General Chebyshev polynomials and discrete Schrodinger operators,
J. Phys. A: Math. Gen., (34) 48, 2001, 10613-10616.
[Z2] M.J. Zygmunt, Matrix Chebyshev Polynomials and Continued Fractions, Linear AL-
gebra Appl.,340(1-3):155-168. 2002.
238 M.J. Zygmunt
Marcin J. Zygmunt
University of Wroclaw
Institute of Mathematics
pI. Grunwaldzki 2/4
50-384 Wroclaw, Poland
http://www.math.uni.wroc.pl/~zygmunt
and
Institute of Mathematics
Wroclaw University of Technology
ul. Wybrzeze Wyspianskiego 27
50-370 Wroclaw, Poland
e-mail: zygmunt@math.uni.wroc.pl
List of Lecture Titles
V. Adamyan
Non-stationary Perturbation Determinants
M. Baro
Dissipative Schrodinger-Poisson Systems
J. Brasche
Discussion of Singular Perturbation of -b. with the Aid of (-b.)~
D. Cichon
Three Term Relations Modulo an Ideal and Ortogonality of Polynomials
P. Cojuhari
Spectral Analysis of Jacobi Matrices
M. Demuth
On the Equilibrium Potential in Scattering Theory
J. Derezinski
Simple Models of Infra-Read Problem in Quantum Field Theory
J. Dombrowski
Jacobi Matrices: Perturbation of Double-weights
P. Exner
Two Strongly Singular Point-Interactions Problems
S-M. Fei
PT-Symmetric Point Interactions
R. Fruboes
Singular Perturbations of Operators - Simple Renormalization Models
J.S. Geronimo
Turning Point Theory for Difference Equations with Applications
A. Gheondea
Invariant Hermitian Kernels: A Unifying Approach
E. Giere
The Wegner Estimate in One Dimension for Non - Overlapping Single Site
Potentials
240 List of Lecture Titles
D. Gilbert
Bounds for the Points of Spectral Concentration for the One-dimensional
Schrodinger Operators
P. Hislop
Continuity of the Integral Density of States for Some Random Operators
on ffi,d
R. Hryniv
Inverse Spectral Problems for Sturm-Liouville Operators with Singular Po-
tentials
Z. Jablonski
Hyperexpansive Composition Operators
W. Karwowski
Unpenetrable Barriers and Canonical Quantization
R. Killip
Trace Formulae and Spectral Theory
A. V. Kisielev
Some Spectral Properties of the Non-Selfadjoint Friedrichs Model
S. Kondej
Spectrum of the Laplace Operator in L 2 (ffi,3) with a Singular Perturbation
Supported by a Not Flat Surface in ffi,3
V. K oshmanenko
The Aronszajn-Donoghue Theory for Rank One Perturbations of the 'H-2-
class
P. Kurasov
Rank one Perturbations in Mathematical Physics
M. Mantoiu
Anisotropic Differential Operators: the Essential Spectrum and Localization
Properties
R. Minlos
The Lower Branches of the Spectrum of the System like Polaron
M. M oszynski
On the Point Spectrum of Some Jacobi Matrices in lP Spaces
B. Pavlov
Dirichlet-to-Neumann Map and Design of a quantum switch
A. Osipov
On some Issues Related to the Study of Difference Operators Generated by
Infinite Band Matrices
List of Lecture Titles 241
A. Posilicano
Nonlinear Singular Perturbations
R. Purice
Eigenfunction Decay for Embedded Eigenvalues of Perturbed Periodic Schro-
dinger Hamiltonians
R. del Rio
Coexistence of Spectra in Rank One Perturbation Problems
A. Rybkin
On a New Transformation of the One Dimensional Schrodinger Equation and
High Order WKB-type Asymptotics
A. Shkalikov
Limit Spectral Curves for the Orr-Sommerfeld Problem and Quasi-classical
Eigenvalue Distribution Along the Curves
P.D. Siafarikas
Linear and Non-linear Partial Difference Equations: Some Existence and
Uniqueness Results with Applications
S. Sontz
Euler's Constant in Mathematical Physics
J. Stochel
A pecularity of the Creation Operator
G. Stolz
Fractional Moment Methods for Continuum Anderson-type Models
F.H. SzaJraniec
Subnormality - the Missed Cell in the Quantization Procedure
G. Teschl
Oscillation Theory for Jacobi Operators
1. M. Tkachenko
Sum Rules and Exact Relations in Quantal Coulomb Systems
W. Van Assche
Jacobi Matrices, Orthogonal Polynomials and the Continuum Limit of the
Toda Lattice
J. Yngvason
Bose-Einstein Condensation in Dilute Gases
L. Zielinski
On The Asymptotic Distribution of Eigenvalues for a Class of Schrodinger
Operators
M.J. Zygmunt
Generalized Chebyshev Polynomials and Discrete Schrodinger Operators
Your Specialized
Publisher in
Mathematics ~
Birkhiiuser IlIJI!
F'o' orders ~ kim .. .,..,. Ihr-'d
tlQIII~~
8ifthioMr ~ NO
,
do ~"'!I'" GmbH & co
""""~
...,...........
Edited by [>.69126 Heodelberg
F"I: <-49 J 6221 J lAS ( 229
Gohberg. I., School of Mathematical Sciences, Tel Avtv University, HniII:~ ....dt
Ramat Avtv, Israel
for o:dtn II'lOJMnI1ln Ihr
This series is devoted to the publication of current research in ope- ~MlfrIU:
rator theory, with particular emphasis on applications to classical
analysis and the theory of integral equations, as well as to numeri- ."""'"
mMNdowlrd~
cal analysis. mathematical physics and mathematical methods in ""."""",
NJ 0109l-U91
electrical engineering. F.,*1201}C8(S05
.. lI\IItorder~
OT , 5]: Gaspar. O. I GoIIberg. 1. 1TImotin. D. I OT 141: Gohberg.l .1 dos Santos, A.F.I Speck. F.-D.I
VasilHCU, F.H.I Zsido, L Recent Advances in Operator Teilreira, F.5. 1WendlMd. W. (Eds.) Operator Theofetical
Theory, Operator Algebfas, and their Applications. Methods and Applications to M.lthematical Physics. The
Proceedings of the XIXth Intemational Conference on Emard Meister Memorial Volume (2004)
Operator Theory, nmisoara (Rumania), 2002 (2004). ISBN 37643-6634-6
ISBN 3-7643-71277
OT 146: Kopachevsky. N.D. I Krein. 5.G. Operator
OT 152: Eid~ 5.0. I lvalOys/Ien. 5.0.1 Kochubei. AppIoach to Unear Problems of H','drodynamics. VokJrne 2:
"-N. Analytic Methods in the Theory of Differemial and Nor!seIfadjoint Problems for VISCOUS Auids (2003)
f'seudo-differential Equations of Parabolic Type (2004) ISBN ].7643-2190-3
ISBN 3-7643-7115-3
OT 14S: A1beverio, 5. 1 Demuth, M.I Schrohe. E.I
OT 151 : Gil. J.8. I Kralnet T. I WItt, I. (Eds.) Aspects of Schulze. 8.-W. (Eds.) Nonlinear Hyperbolic Equations.
Boundary Problems in Analysis and Geometry (2004) Spectral Theory, and Wavelet Transformations (2003)
ISBN ].7643-70696 IsaN ].7643-21687
OT 144: BeHtsldi, G. I Ttachenko, V. One-dimensional
OT 150: Rabinovich, V. I Roch. S. 1Silbermann. 8. functional Equations (2003)
limit Operators and their ~Ikations in Operator Theory ISBN 3-764].(1084-1
(2004)
ISBN ).7643-7081-5 OT 14]: A1pay, D.(Ed.) Rep!odoong Kernel Spaces and
Applications (2003)
OT 149: Ball, J.A- I Helton, J.W.I KIHs, M.I ISBN 3-764]-{)068X
Rodman, L(Eds.) Current TrMd5 in Operator Theory and
its ~Iications (2004) OT '42: Bottcher, A.I dos Santos, A.F.I Kaashoek.
ISBN 3-7643-706J.X M.A. I Britts Lebre. A.I Speck, F.O. (Eds.) Singular
Integral OperatDfs, factDfization and Applications.
OT 148: Asnyralyev, A- I Sobolnrsll:ii, P.E.New Internatiooal Workshop on Operator Theory and
Difference SdIemes for Partial Diffefential Equations (2004) Applications, IWOTA 2000 - Portugai (2003)
ISBN 3-764].7054-8 ISBN 3-7643-6947-7
Your Specialized
Publisher in
Mathematics ~
Birkhiiuser lIIJ2l
or 141 : dos nICK, A.f. I Goh~g. I. I .... nojlovic, It (Eds.). Of 129: Bori<hev, A.A. I Nikolski. N.K.(Eds). SVSlt'm$,
fKlD'ilMlOl'l .-t Integrable s"stems.PrIXtfdl1gs 0I1ht Sumrnrr AWrOli'nation. SingINr Integr" ()ptrator\, and ReIatrd Topics
SdmI. f..o. PorWgal. 2000 (2003) (2001)
ISBN 3-1&43-69388 ISBN 3-7643-66451
or 140: Ellis, R.I GoIlberg.1. 0rth0g0NI s"stemund Of 121: KopacMvsky. N.D. I Krein. 5.G. (lp!ralOr AWroach 10
CorMJIut;on Optrators (2002) I.ftar I'roblftnsof HydIodynarnics. VoIt.me 1: 5tII.~ I'r~
ISBN 3 7643692H for iIIlldeal Fluid (2001)
ISBN37643~2
01 139: Muller, V.Spffi,~ Theory of linN' Optratofl and
Spffiral Systems in Baoach AIgeOr~ (2003) Of l Z1: K~rdry.l.1 Folas, C.1. 1 Gohberg.1.1 Lln~r. H.
IsaN 3764369124 (ds). Recent Aot.ran<es In ()ptratOf Theory and Related Topics
(1001)
Of 138: AJbeVHIQ, S. I Demuth, M.I Schroht. E.I Sdlulze, lSIIN 3-764366079
" W. (Eds.). Parabolidty. Vollena Cak1.M, and Conical
OT 126: Demuth,. M.I 5chc1~ 8.W. ((ds). Partial Differenbal
SinglNrilits (2002)
E~1ions iII'Id 5ptctral Thtory (1001)
ISBN 3-760-6906-X
ISBH 3-7643-6119-7
Of 1)1: Dybin. V.I Grud$ky. S.M. ~o6.oaion IOIht Theory 01 OT 125: Gil LB. I Gries. D. 11.nch. M. (lds.). AWroadlrtS 10
Toeplitz (lp!raun 'Mill Infinr~ IrocB (2002) Sir"gWf AnaIy\Is (2001)
ISBN 3-1613-6906-X ISBN 3-7643-6518-t1
Of 136: Wong. M.W. W.welet Trill'lSforms and lOOlliution OT 12': Oijksmf. A. / Kaas/Ioek., M.A. I Ran. A.C.M. (Eds.).
OptralOfS (2002) Rttenl~ on OptratOf Thtory(2001)
ISBN37643-678H ISBN 37643-657HI
OT 135: l!ittchtr. A.I Goh~g. 1.1 Jungt..nns. P. (Eds.). OT 123: Alpay. 0./ Vlnnlkov. V. (Edi). OprratOl Theory. System
TotpIOU MWOct$, COO'I'IOlJtion ClpfWOI$, iIIld Integral E~tiom. Theory and Related Topics (2001)
TIlt ~nd SilbrnnaM ~ VoIufM {2002) ISBN 37643-65234
IsaN )-7643-68172
OT 122: Bart. H. I Gohberg. I. I Ran. A.C.M. (Eds~ ()ptratof
Of 134: AJpay. 0. 1 Gohberg. I. I Vinnikov, V. (Eds.). Theory and Analysis (2001)
ln~bOn Theory. s"stems Theory and Related Topics. Tht Kirry ISBN 3-7643-6499-B
Dym Atri.oer5aoy Vob'nr (2002)
ISBN 3-7613-67628 OT 121; EIsdwIer.. J. I Gohberg.1. 1 5il~ .. (Eds.).
Problems ld Method!. on Ma1!lernatQI Physocs (2001)
Of 1)); Itnll. A.M. HiRItn Sp.te, Iku:Idy Valut ProbItms and ISBH 3-7643-64777
0nh0g0naI ~(2002)
OT 120: Beltita, 0.1 Sabac, M. lie AJo}eIIras of Boooded
ISBN 37643-6701-6 OperalOl'l (2001)
or 132: Albeverio,. 5. 1 ElandH. N.I E"ftfitl W.N.I KurlSOv. IS8N 3-7643-64041
P. (Eds.). ()ptratOl Methods in Or~ and PiII~al Oilferen~al OT 119: litvinov. W.G. Optlmization In E~iptOc Problems 'Ni!h
Equations. S. Kowlevsky Sym~um. Univer50ty 0/ Stoc\J-.oIm. June AWliultioos to MhanIr::s oIDf101mabie ~ and Fluid
2000 (2002) MecNri:s (2000)
IsaN ) 76436190-3 IS8N 37643-6199-9
OT 131: Bottcher, A. I Karlovid!, Y.I. I SpitklMky,l.M. OT lIB: Kaashoet. M.A. l langer. H .I ~. G. (Eds).
COfMI1uuon Oplfaun and FoKtOrilation 01 ~ Periodic MatrIX Optrator Theory and R$tfti Topic:s (2000)
FI.O'ICbOI'IS(2CI02) ISBN H643-6lB8-X
ISBN 3-7613-6672-9
OT 111: Mamyan. V.M.I GolIbtrg. I. I Gotbachuk. M. I
Of 130: Gohberg. 1.1 ~ H. (Eds). ~ ()peroJlOfS and Gotbachuk. V. tlds.). OrffeIenNI OperalOfS and ReIa!!d TopICS
Matrices (2001) (lWl>
ISBN 3-7&43-66559 ISBN 3-7643-6287'

Spectral Methods for Operators of Mathematical Physics-Birkhäuser B - (Operator Theory_ Advances and Applications 154) Vadim Adamyan, Heinz Langer (Auth.), Jan Janas, Pavel Kurasov, Sergei Naboko (Eds.)

Caricato da

Informazioni sul documento

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Spectral Methods for Operators of Mathematical Physics-Birkhäuser B - (Operator Theory_ Advances and Applications 154) Vadim Adamyan, Heinz Langer (Auth.), Jan Janas, Pavel Kurasov, Sergei Naboko (Eds.)

Caricato da

Copyright:

Formati disponibili

Operator Theory: Advances and

Jan Janas Pavel Kurasov

A CIP catalogue record for this book is available from the

Bibliographic information published by Die Deutsche Bibliothek

ISBN 978-3-0348-9632-0 ISBN 978-3-0348-7947-7 (eBook)

2004 Springer Basel AG

Introduction ............................................................. vii

Vadim Adamyan and Heinz Langer

Dariusz Cichon, Jan Stochel and Franciszek Hugon SzaJraniec

List of Lecture Titles ..................................................... 239

This volume contains proceedings of the International Conference: Operator The-

for generous financial support. It is impossible to imagine the conference without

Krakow-Lund-St. Petersburg, March 2004

Abstract. A partial non-stationary perturbation determinant ~I(t) is defined

where A is a self-adjoint operator in some Hilbert space 1i l , H is a self-adjoint

onto Ji l , and H(H - A)IJ-ll is a trace class operator in Ji l . We are interested in

with non-negative numbers a and b. Moreover, under an additional assumption

2. Non-stationary perturbation determinants

Wo(t, s)*W(t, s) = I - i it WO(T, S)*V(T)W(T, s) dT, (2.1)

i.l(t,s) := det (Wo(t,s)*W(t,s)) = e

= -i (V(t)W(t, s)x, Wo(t, s)y).

If, in Theorem 2.1, Ho(t) == Ho and H(t) == H are constant self-adjoint

A discrete analog is as follows. Let Ho and H be as in Corollary 2.2. For

Proof. The properties of determinants (cf. [7]) imply

= det [(H - z)-I(Ho - z)(Ho - z)-I(H - z)]

3. Asymptotic behavior of ~l(t)

which is supposed to be self-adjoint. We assume that V, B E SI and, additionally,

Moreover, X is a trace class operator: X E SI, ranX C domD, ranX* C domA,

H=T( A+VO+BX o (3.7)

It follows that also the operator e- itH admits the representation

If P l denotes again the projection onto Hl in H, then

Proof. Using the identity

det (e itA P e-itHI ) = det (eitAe-it(A+V+BX)) .6.(t)

and (3.10) is proved. o

Proof. First we observe that

(U(t)'l/Jv, 'l/JI") = ei'\t fl"v(>")d>", -00 <t< +00,

where fl"v E 1L 1 (-00, 00). The Riemann-Lebesgue lemma yields

we get from (3.13)

Ho := (~ ~), H := (:* ~) (3.14)

with B E SI the relation

Proof. If B = 0 then, evidently, E;,H/Ho(>\) = 0 a.e.

where In denotes the continuous branch of the logarithm with In 1 = O. Therefore

For the self-adjoint operator Ho and Q E 51 we have

From (3.15), (3.16), (3.18) we obtain

- ~tr ((Ho - 0(:3 )

= -~tr ((Ho - iy)-lQ(Ho - iy)-lQ) + 0(:3) , y i 00,

and, using (3.17), tr Q2 = 0, which because of the self-adjointness of Q implies

Note that in general for a pair of self-adjoint operators H, Ho with H - Ho E

Then for z with 'Sz -I- 0

Li(t) = det (I + eit(A+V+BX) X*e-it(D-B* x*) X) = 1 + 0(1), t -t 00. (3.20)

The self-adjointness of the operator A + V + BX in (HI, ((I + X* X), . )) means

is a group of unitary operators in HI with generator A which has absolutely

a = tr (V + BX) = ~H/iIo (A) dA + tr V. (3.21)

which, because of ~H/Ho(A) ~ 0 a.e. on (a,oo), yields ~H/Ho(A) = 0 a.e. Finally,

In the following for the self-adjoint operator

in 1i = 1i1 EB 1i2 the invertibility of the operator

plays some role.

Proof. We consider the representation (3.3) of the spectral subspaces L_ and L+

Applying the operator (J + X* X)I/2, using the identities

= II(I +X*X)I/2e it (A+BX)(J +X*X)-1/2(1 +X*X)I/2X*e- it (D-B*X*) Xx ll

< 11(1 +XX*)1/2e- it (D-B*X*)(J +XX*)-1/2(J +XX*)1/2Xxll

= IIX(J + X* X)I/2XII < 11(1 + X* X)I/ 2XII,

4. Partial perturbation determinants and Fredholm resolvents

~I(t) = det (eitAPIe-itHI1tJ.

Wo(t, s)W(t, s) = I - i it WO(T, S)V(T)W(T, s) dT, (2.1)

Li(t) = det (I + eit(A+V+BX) Xe-it(D-B x*) X) = 1 + 0(1), t -t 00. (3.20)

= II(I +XX)I/2e it (A+BX)(J +XX)-1/2(1 +XX)I/2Xe- it (D-BX) Xx ll

< 11(1 +XX)1/2e- it (D-BX*)(J +XX)-1/2(J +XX)1/2Xxll