Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
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:
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
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.
ISBN 978-3-0348-9632-0
987654321 www.birkhauser-science.com
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
Partial Non-stationary
Perturbation Determinants
Vadim Adamyan and Heinz Langer
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)
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
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))
Partial Non-stationary Perturbation Determinants 3
where the integral on the right-hand side exists with respect to the norm of Sl and
belongs to Sl, and
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)
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),
Partial Non-stationary Perturbation Determinants 5
and, finally,
d~d:'s) = -itr[V(t))]~(t,s).
Observing that ~(s, s) = 1 the relation (2.2) follows.
D
Further,
det [UO(Z)-IU(Z)] = det [(Ho - z)(Ho - z)-I(H - z)(H - Z)-I]
H = ( A;* V ~) (3.1)
(3.5)
is invertible,
(1 + X* X)-1
T- l = ( -(1 +XX*)-IX (3.6)
Partial Non-stationary Perturbation Determinants 7
and it diagonalizes H:
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) .
and hence
(3.11)
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
Observing that
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
= JCXJ
E;,H/Ho(>\) d)'"
)...-z '
8'z -I- 0,
-CXJ
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 '
such that
H := (I + T*)Ho(1 + T) -I- Ho
10 v. Adamyan and H. Langer
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
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.
Partial Non-stationary Perturbation Determinants 11
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],
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
H=(:* ~) (3.22)
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.
12 v. Adamyan and H. Langer
II(I + X* X)I/ 2x l
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
i J
00
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
where the functions gl"v(z) are the entries of the n x n-matrix function
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
J
00
the representation
J
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
and hence
eitAP1e-itHIHl = 1 - eitAR(t)e-itA, (4.10)
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
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
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
x - R(t)x = f,
that is
(4.15)
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
It follows that
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
18 V. Adamyan and H. Langer
defined by (4.14) then jar the partial perturbation determinant ~l(t) it holds
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:
Advances and Applications, Vol. 154, 19-33
2004 Birkhiiuser Verlag Basel/Switzerland
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.
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)
1 Notice that if V(A) = 'It, then the inclusion (1) becomes an equality.
Noncommuting Domination 21
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).
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
3 For instance, [; = DOC (H) or [; = the set of all bounded vectors of H, d. [5].
Noncommuting Domination 25
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 '
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.
(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
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<}'
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)*,
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
=
J=O
(22)
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 Cf. [5],
Noncommuting Domination 31
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.
Noncommuting Domination 33
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
U niwersytet J agiellonski
ul. Reymonta 4
PL-30059 Krakow
e-mail: stochel<!lim.uj.edu.pl
Franciszek Hugon Szafraniec
Instytut Matematyki
U niwersytet J agiellonski
ul. Reymonta 4
PL-30059 Krakow
e-mail: fhszafra<!lim.uj.edu.pl
Operator Theory:
Advances and Applications, Vol. 154, 35-50
2004 Birkhiiuser Verlag Basel/Switzerland
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.
other applications. We also note the discussion undertaken in [4] (see also [7]) on
the connection with related results from the works [1], [20].
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'.
Finiteness of Point Spectrum 39
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
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).
Finiteness of Point Spectrum 43
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).
The next theorem stands out as an alternate version of Corollary 3.6 for the
considered discrete case of operators.
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.
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,
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
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.
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
)
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.
Finiteness of Point Spectrum 49
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
2004 Birkhiiuser Verlag Basel/Switzerland
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
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
54 v. Derkach, S. Hassi and H. de Snoo
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
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
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
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
FE SQ.
In this case the Weyl function of SQ associated with the boundary triplet Irq
coincides with the matrix polynomial Q, cf. [9J.
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
58 V. Derkach, S. Hassi and H. de Snoo
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).
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
Singular Perturbations and Range Rerturbations 59
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)).
(3.12)
(3.13)
60 v. Derkach, S. Hassi and H. de Snoo
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).
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
62 V. Derkach, S. Hassi and H. de Snoo
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.
Singular Perturbations and Range Rerturbations 63
(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.
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
Singular Perturbations and Range Rerturbations 65
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
pr810 = (I - p)r~lo = 0.
(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,
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).
Singular Perturbations and Range Rerturbations 67
[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).
68 V. Derkach, S. Hassi and H. de Snoo
[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:
Advances and Applications, Vol. 154, 69-90
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
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
Proof of Theorem 2.1. Consider the linear relations Ao and Uo in (Q) defined by
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
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 .
Minimal Realizations of Generalized Nevanlinna Functions 75
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
76 A. Dijksma, H. Langer, A. Luger and Yu. Shondin
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
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)
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
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*:
Bo ({f,g}) = c, B1 ({f,g}) = d.
Hence the relation AT =: AA,B is described by the formula
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
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),
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 .
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
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
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
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)
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.
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)
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.
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) :
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):
84 A. Dijksma, H. Langer, A. Luger and Yu. Shondin
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
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*
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
Minimal Realizations of Generalized Nevanlinna Functions 85
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,
~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.
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
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.
Minimal Realizations of Generalized Nevanlinna Functions 87
(f)
Proof. Note that the span on the right-hand side of (4.1) is contained in Ho.
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*
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}.
88 A. Dijksma, H. Langer, A. Luger and Yu. Shondin
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
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.
Minimal Realizations of Generalized Nevanlinna Functions 89
[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
Department of Mathematics
University of Groningen
P.O. Box 800
NL-9700 AV Groningen, The Netherlands
~mail: a.dijksma@math.rug.nl
Heinz Langer
Institute for Analysis and Scientific Computing
Vienna University of Technology
Wiedner Hauptstrasse 8-10
A-1040 Vienna, Austria
~mail: hlanger@mail.zserv.tuwien.ac.at
Annemarie Luger
Institute for Analysis and Scientific Computing
Vienna University of Technology
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
2004 Birkhauser Verlag Basel/Switzerland
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
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
L L
00 00
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).
L jn-I) l'l/Jnl 2+ L
00 00
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
~ (C 2'I/J,'I/J) = ~ (C'I/J,C'I/J)
L
00
k=l
+L
00
+ (x - y)2
k=l
L 2(x -
00
(x - y?'I/J~k-l
k=l
L (x -
00
~ y)2'I/J~k_1 .
k=l
D
Eigenvalues and Spectral Gaps 95
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
CXJ
k=1
L
CXJ
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
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.
L a~ l'n+1 -
00
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
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.
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
"21 (2
C'l/J,'l/J ) ="21 (C'l/J,C'l/J)
L
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=l
D
Eigenvalues and Spectral Gaps 99
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
= =
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
Department of Mathematics
Wright State University
Dayton, OH 45435, USA
e-mail: joanne. dombrowski<owright. edu
Operator Theory:
Advances and Applications, Vol. 154, 101-138
2004 Birkhiiuser Verlag Basel/Switzerland
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
2. Motivation
a. Differential equations - the Langer transformation
Consider the differential equation,
2
_p3/2(t)
3
= it
to
k(u)du (2.2)
(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
104 J.S. Geronimo, O. Bruno and W. Van Assche
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.
Here to is such that ~;;-(~~~';~ = 1. We now define the Langer transformation for
the above difference equation as
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,
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].
106 J.S. Geronimo, O. Bruno and W. Van Assche
Also
t:i k2(t, y, E) E CO([tin' tf;] X [0, EO]) i = 0, ... ,3,
1.
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)
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.
WKB and Turning Point Theory 107
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
and
Xi,k(p,p) = ~1P sXi,k_l(p,s)sinh..jP(p-s)ds. (3.5)
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)
= lp'
p S
( - sinh yps + - cosh yps) ds,
4yIP
s2
, -
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.
and
W = clos(W).
WKB and Turning Point Theory 109
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
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)
- 2cosh(pl/2P')Ai(C 2/ 3p(t))
Here,
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
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
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
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)
WKB and Turning Point Theory 113
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,
Since u(j) satisfy the Airy equation it is not difficult to compute their
Wronskian
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,
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
114 J.S. Geronimo, O. Bruno and W. Van Assche
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
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
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)
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)
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
WKB and Turning Point Theory 115
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
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
+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.
116 J.S. Geronimo, O. Bruno and W. Van Assche
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
Proof. Set
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
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
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
118 J.S. Geronimo, O. Bruno and W. Van Assche
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
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.
WKB and Turning Point Theory 119
and
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
120 J.S. Geronimo, O. Bruno and W. Van Assche
(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],
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
Since the above function and Bi all satisfy the same differential equation there is
a relation among them given by ([0, p. 414)),
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
122 J.S. Geronimo, 0. Bruno and W. Van Assche
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,
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
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
WKB and Turning Point Theory 123
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
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,
+
(i:~(p( i-1)3/2) >0)
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.
124 J.S. Geronimo, O. Bruno and W. Van Assche
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.
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) ,
Set
j
vti,tj(g) = lt Ig(t,E)ldt, (5.8)
+ vt,HE((4a'a())2(t + E - .))),
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
J~,Sl (t) =
1
2E
I t
t
s~'(u)(t E - u)2du
+ I t
t
s~(u)(t E - u)du,
and
128 J.S. Geronimo, O. Bruno and W. Van Assche
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
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).
WKB and Turning Point Theory 129
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
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)
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
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
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
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)).
132 J.S. Geronimo, O. Bruno and W. Van Assche
Then there exists an EK such that for (n., E) E [0, tjl x (0, EKl
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)
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) ,
WKB and Thrning Point Theory 133
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
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.
134 J.S. Geronimo, 0. Bruno and W. Van Assche
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)
WKB and Thrning Point Theory 135
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)
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
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
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.
138 J.S. Geronimo, O. Bruno and W. Van Assche
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:
Advances and Applications, Vol. 154, 139-149
2004 Birkhauser Verlag Basel/Switzerland
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
(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)
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)
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.
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
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
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
144 D.J. Gilbert, B.J. Harris and S.M. Riehl
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].
11 00
e2iv'Xtq(t)dtl :::; a(x)r]()..),
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
Bounds for the Points of Spectral Concentration 145
(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
"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].
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 ---+
, 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,
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,
Bounds for the Points of Spectral Concentration 147
(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,
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.
148 D.J. Gilbert, B.J. Harris and S.M. Riehl
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,
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.
Bounds for the Points of Spectral Concentration 149
[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
Department of Mathematics
University of Northern Iowa, USA
Operator Theory:
Advances and Applications, Vol. 154, 151-154
2004 Birkhauser Verlag Basel/Switzerland
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.
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
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
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)
(3-:- = _ TI)=1
N ( -:- - A)
J-L
n-l
2 TI l#i (-
J-Li - J-LI-)TI 1=1- ( J-Li-
N n
- J-LI+)' i=1
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.
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.
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,
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
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
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 Asymptotics of Solutions of Difference Equations 159
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
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
On the Asymptotics of Solutions of Difference Equations 161
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
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
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:
Advances and Applications, Vol. 154, 163-177
2004 Birkhauser Verlag Basel/Switzerland
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
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
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.
11
initial data:
u ="2 -00
00
[
I \l2
uol + q(x)luol 2+ lUll ] dx.
The evolution operator of Cauchy data
(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
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}
-,
then
(:I+h, :I-g) = .;f; 1 dx Jdk Jdk [_e- ikx + S(k)e ikx ] h+(k)
00
+ 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
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
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
Spectral Meaning of Functional Model 169
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
.cit = ( Au + r 2+ [~:~x~)
.~
+ u+(O)] ) .
Z dx
2This condition can be essentially relaxed, see the remark after Theorem 3.2.
170 B. Pavlov
(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 <I>(A) in upper
half-plane <;SA > -21'
P K <I>()PK =
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
Spectral Meaning of Functional Model 171
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)
(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
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
Spectral Meaning of Functional Model 173
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
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
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 )
= 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-
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].
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
( fin) (FW_f(O))
fout - FW+f(O)
where F is the standard Fourier transform in L 2 :
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)
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.
Spectral Meaning of Functional Model 177
[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:
Advances and Applications, Vol. 154, 179-184
2004 Birkhauser Verlag Basel/Switzerland
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).
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.
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
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
2
sup IImzlll(B - Z)-l wl1 < 00 (1)
zEIC
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
1: 1:
in k from -N to N,
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
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),
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:
Advances and Applications, Vol. 154, 185-201
2004 Birkhiiuser Verlag Basel/Switzerland
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)
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)
(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}.
(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
n~O n
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)
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)
111(-. A) liP ~
P
(1 + I~A)P-l L n~O
e- lmAn . 2PIlfllfp(L 1)
(2.6)
x-oo
lim ip (x) = O.
The function ip will be specified in the proof
Consider hand h separately. Integrating (2.1) by parts and taking into account
that h E C l yield
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:
(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
where j(A) is the usual Fourier transform and (2.15) implies the Riemann-Le-
besgue Lemma.
(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.
-1=
mations
q1 (x, A) = e2iAS q (8 + x) d8 = -q(x, 2A). (3.1)
WKB asymptotics 191
y (x, >.) = 1+ 1 00
Kn (x, s, >.) y (s, >.) ds, (3.4)
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
(3.8)
192 A. Rybkin
An (x, A) := iAX + rt
10 m=l
qm (s, A) ds.
18m +, (x, s, -X) I' :5 exp { -2lm.ls (1- 1~.1 t, IIq, (-, .I) "= ) } (3.12)
:; 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
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,
: ; 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
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
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
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.
m
(A2) = \[1' (0, A)
\[1 (0, A) ,A
2
E C+. (4.2)
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
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
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
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)
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.
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
JIgl = ~.
b
Set now
a
12 = fXIR\[a,b] + g.
By (5.1) and (5.2), we have
J
b
[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:
Advances and Applications, Vol. 154, 203-218
2004 Birkhiiuser Verlag Basel/Switzerland
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
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
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
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
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.
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
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
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
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.
Discrete Uniform Levinson Type Theorems 209
It remains to prove property 3. We must show that if <P E Xk(No) then (5)
holds for {(Tk<P)n(A)}n>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
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
AEJ I ITi=No vi (oX) I m=No
But
8-1
sup { n_l 1 (k) L
IIp(1)lJIn(oX)IJI;;-'~1 (oX)Rm(oX)CPm(oX)ll}
AEJ I ITi=No vi (oX) I m=No
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
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
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
(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.
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
Discrete Uniform Levinson Type Theorems 213
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 ,
difficult to show that we can find an m2 ::::: m1, such that there exist G1 , G2 , G3 > 0
for which the following estimates hold
This last estimate, together with (22) and conditions iv and v, yields the bound-
edness of
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
n-1
supllx~k)(A)/ II VY)(A) - Pkll -+ 0, as n -+ 00,
.\EJ i=m
)
to the canonical basis in l2(N) is a Jacobi matrix of the form
...
... , (23)
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
(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
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
Discrete Uniform Levinson Type Theorems 217
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:
Advances and Applications, Vol. 154, 219-231
2004 Birkhauser Verlag Basel/Switzerland
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
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].
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
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:
~o = 1E
( Mo No
et(O)*
)
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;.
Free Functional Model for Domains 223
1) ( 7ri ) (7r, T)
Tt
1 0
= ( 0 PclasRan6 );
2) ( 7ri )
T1
(7r~, T~) = (
8~
~
~~
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
This projection plays a central role in the paper and have the following properties:
226 A. Tikhonov
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
Free Functional Model for Domains 227
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].
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 <I> : '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, <I>(z) = T; 2) 11<1>11 < 00;
3) II'ljJnll < K, lim 'ljJn(z) = 0 a.e. z E C ===} s- lim <I>('ljJn) = o.
n~~ n~~
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
c,
J (-rrI+X+(UI)(I - PIe)J)(ZI) dZ I = o.
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.
Free Functional Model for Domains 231
[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:
Advances and Applications, Vol. 154, 233-238
2004 Birkhauser Verlag Basel/Switzerland
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
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
((U,;;,BIU~,B))WA,B = J U,;;,B(x)dWA,B(x)U~,B(X)* = 6n ,m I .
IR
Mn = ~JJ4-X2(At+Btdt
27r
-2
Moreover
SUppWA,B = U u(At + B) .
tE[-2,2]
Proof. We have
Jacobi Block Matrices with Constant Matrix Terms 235
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
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
1 f
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
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
Proof. Let
At + B = U(t)A(t)U(t)* . (3)
Putting (3) into (1) gives
J J J
2 2
where
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'