Removal of the resolvent-like energy dependence from interactions and invariant subspaces of a total Hamiltonian
Extracted source text ยท arXiv funct-an/9606002
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\documentstyle[12pt,aps,floats]{revtex}
\renewcommand{\baselinestretch}{1.00}
\newcommand{\be}{\begin{equation}}
\newcommand{\ee}{\end{equation}}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\newtheorem{theorem}{Theorem}
\newtheorem{lemma}{Lemma}
\newtheorem{corollary}{Corollary}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{document}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
\title
{Removal of the resolvent-like energy dependence from
interactions and invariant subspaces of a total Hamiltonian}
%
\author{A.K.Motovilov}
\address{Laboratory of Theoretical Physics, JINR,
141980 Dubna, Moscow region, Russia\\
E--address: motovilv@thsun1.jinr.dubna.su}
\maketitle
\bigskip
\bigskip
\begin{abstract}
%
The spectral problem $(A + V(z))\psi=z\psi$ is considered where the main
Hamiltonian $A$ is a self-adjoint operator of sufficiently arbitrary
nature. The perturbation $V(z)=-B(A'-z)^{-1}B^{*}$ depends
on the energy $z$ as resolvent of another self-adjoint operator $A'$.
The latter is usually interpreted as Hamiltonian describing an
internal structure of physical system. The operator $B$ is assumed to
have a finite Hilbert-Schmidt norm. The conditions are formulated
when one can replace the perturbation $V(z)$ with an
energy-independent ``potential'' $W$ such that the Hamiltonian $H=A
+W$ has the same spectrum (more exactly a part of spectrum) and the
same eigenfunctions as the initial spectral problem. The Hamiltonian
$H$ is constructed as a solution of the non-linear operator equation
$H=A+V(H)$. It is established that this equation is closely connected
with the problem of searching for invariant subspaces of the
Hamiltonian
$
{\bf H}=\left[ \begin{array}{lr} A & B \\
B^{*} & A' \end{array}\right].
$
The orthogonality and expansion theorems are proved for eigenfunction
systems of the Hamiltonian $ H=A + W $. Scattering theory is
developed for this Hamiltonian in the case where the operator $A$ has
continuous spectrum.
\end{abstract}
%
\bigskip
\bigskip
AMS subject classification (1980) 47B99, 81G05.
%
\pacs{02.30.Tb, 03.65.Nk, 12.40.--y}
\bigskip
\noindent Published in
J. Math. Phys., 1995, vol. 36 (12), pp. 6647-6664.
LANL E-print {\tt funct-an/9606002}.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{ INTRODUCTION}\label{Intro}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Perturbations, depending on the spectral
parameter (usually energy of system) arise
in a lot of physical
problems (see papers~\cite{Dashen}--\cite{IMU} and Refs. therein).
In particular, such are the interaction
potentials between clusters formed by quantum
particles~\cite{Dashen}--\cite{NarodetskyPalanga}.
The perturbations of this type appear
typically~\cite{Dashen}--\cite{Kerbikov}, \cite{Pavlov84}--\cite{IMU}
as a result of
dividing the Hilbert space ${\cal H}$ of physical system in two subspaces,
${\cal H}={\cal H}_1\oplus{\cal H}_2$.
The first one, say ${\cal H}_1$, is interpreted as a space of
``external'' (for example, hadronic) degrees of freedom. The second one,
${\cal H}_2$, is associated with an ``internal''
(for example, quark) structure
of the system. The Hamiltonian ${{\bf H}}$ of the system looks as a matrix,
\be
\label{twochannel}
%
{{\bf H}}=\left[
%
\begin{array}{cc}
%
A_1 & B_{12} \\
%
B_{21} & A_2
%
\end{array}
%
\right]
%
\ee
with $A_1$, $A_2$, the channel Hamiltonians (self-adjoint
operators) and
$B_{12}$, $B_{21}=B_{12}^{*}$, the coupling operators.
Reducing the spectral problem
${{\bf H}}U=zU$ , $U=\{ u_1 ,u_2 \}$ to the
channel ${\alpha}$ only one gets the spectral problem
\be
%
\label{ini}
[A_{\alpha} +V_{\alpha} (z)]u_{\alpha} =zu_{\alpha} ,
%
\quad {\alpha} =1,2,
%
\ee
where the perturbation
\be
\label{epot}
%
V_{\alpha} (z)=-B_{\alpha\beta} (A_{\beta} -z)^{-1}B_{\beta\alpha} ,
%
\quad {\beta} \neq {\alpha},
%
\ee
depends on the spectral parameter $z$
as the resolvent $(A_{\beta} - z)^{-1}$ of
the Hamiltonian $A_{\beta}$.
In more complicated cases $V_{\alpha}(z)$
can include also linear terms in respect
with $z$. Other types of dependency of the potentials
$V_{\alpha}(z)$ on the spectral parameter z give, in a general way,
the spectral problems~(\ref{ini}) with a complex spectrum.
The present paper is a continuation of the author's
works~\cite{MotJMPh91}--\cite{SPbWorkshop} devoted to a study of the
possibility to ``remove'' the energy dependence from perturbations
of the type~(\ref{epot}). Namely, in~\cite{MotJMPh91}--\cite{SPbWorkshop}
we construct such new potential $W_{\alpha}$
that spectrum of the Hamiltonian
$H_{\alpha}=A_{\alpha}+W_{\alpha}$
is a part of the spectrum of the problem~(\ref{ini}).
At the same time, the respective eigenvectors of $H_{\alpha}$ become
also those for~(\ref{ini}).
Hamiltonians $H_{\alpha}$ are found as solutions of the
non-linear operator equations
\be
\label{basic}
%
H_{\alpha}=A_{\alpha} + V_{\alpha} (H_{\alpha})
%
\ee
first appeared in the paper~\cite{Braun}
by M.A.Braun in connection with consideration
of the quasipotential equation. The operator-value function
$V_{\alpha}(Y)$ of the operator variable
$Y$, $Y:\, {\cal H}_{\alpha}\rightarrow{\cal H}_{\alpha},$ is defined
by us in such a way (see Sec.~\ref{BasicEquation})
that eigenvectors $\psi$ of $Y$,
$Y\psi =z\psi$, become automatically those for $V_{\alpha}(Y)$ and
$V_{\alpha}(Y)\psi=V_{\alpha}(z)\psi.$
Note that the problem of removal of the dependence on a spectral
parameter from perturbation is interesting not only in itself.
In particular, there is a rather conceptual question
(see for instance Refs.~\onlinecite{McKellarMcKay},
\onlinecite{Schmid} and~\onlinecite{MotJMPh91})
concerning a use of the two--body energy--dependent potentials
in few--body nonrelativistic scattering problems. Since the energies
of pair subsystems are not fixed in the N--body (N$\geq 3$) problem,
a direct embedding of such potentials into the few--body Hamiltonian
is impossible. Thus, the replacements of the type (\ref{epot})
energy--dependent potentials with the respective new
potentials $W_\alpha$ could be considered as a way to overcome
this difficulty.
In Ref.~\onlinecite{MotJMPh91}, the problem of the removal of the
energy dependence from the type (\ref{epot}) perturbations was
considered in details when one of the operators $A_{\alpha}$ is the
Schr\"{o}dinger operator in $L_2({\bf R}^n)$ and another one has a
discrete spectrum only. The reports~\cite{Kazimierz},
\cite{SPbWorkshop} announce the results concerning the
equations~(\ref{basic}) and properties of their solutions $H_{\alpha}$ in a
rather more general situation where the Hamiltonian ${{\bf H}}$ may be
rewritten in terms of a two-channel variant of the Friedrichs model
investigated by O.A.Ladyzhenskaya and L.D.Faddeev in
Ref.~\onlinecite{LadyzhFaddeev} and L.D.Faddeev in
Ref.~\onlinecite{Faddeev64}. In the paper~\cite{IMU} the
method~\cite{MotJMPh91}--\cite{SPbWorkshop} is applied for
construction of an effective cluster Hamiltonian for atoms adsorbed
by the metal surface.
In the present paper, we specify the assertions
from~\cite{Kazimierz}, \cite{SPbWorkshop} and give proofs for them.
Also, we pay attention to an important circumstance disclosing a
nature of solutions of the basic equations (\ref{basic}). Thing is
that the potentials $W_{\alpha}=V_{\alpha}(H_{\alpha})$
may be presented in the form
$W_{\alpha}=B_{\alpha\beta} Q_{\beta\alpha}$
where the operators $Q_{\beta\alpha}$ satisfy the stationary
Riccati equations~(\ref{QbasicSym}) (see Sec.~\ref{BasicEquation}).
Exactly the same equations arise in the method of construction of
invariant subspaces developed by V.A.Malyshev and
R.A.Minlos~\cite{MalyshevMinlos1},~\cite{MalyshevMinlos2} for a class
of self-adjoint operators in statistical physics. It follows from
the results of~\cite{MalyshevMinlos1}, \cite{MalyshevMinlos2} that
operators $H_{\alpha}$, ${\alpha}=1,2,$ determine in fact, parts of the
two-channel Hamiltonian ${{\bf H}}$ acting in corresponding invariant
subspaces (see Theorem~\ref{ThInvariant} and comments to it).
Recently, the author came to know about the
work~\cite{LangerAdamyan}
by V.M.Adamjan and H.Langer studying
the operator-value functions written in our notation as
$F_{\alpha}(z)=z-A_{\alpha}\pm
B_{\alpha\beta}(A_{\beta}-z)^{-1}B_{\beta\alpha}$.
In particular Adamjan and Langer show in this work
that a subset of eigenvectors of $F_{\alpha}$ can be chosen
to form a Riesz basis in
${\cal H}_{\alpha}$. There is a certain intersection
of their results and ours from
Refs.~\onlinecite{MotJMPh91}--\onlinecite{SPbWorkshop}.
However the methods are different.
The paper is organized as follows.
In Sec.~\ref{Initial} we describe the Hamiltonian ${{\bf H}}$ as a
two-channel variant of the Friedrichs
model~\cite{LadyzhFaddeev}, \cite{Faddeev64}.
We suppose that both operators $A_{\alpha}$, ${\alpha}=1,2,$
may have continuous
spectrum. When properties of objects connected with this spectrum
(wave operators and scattering matrices) are considered in
following sections, the coupling operators
$B_{\alpha\beta}$ in~(\ref{twochannel})
are assumed to be integral ones with kernels
$B_{\alpha\beta}(\lambda,\mu)$, the
H\"{o}lder functions in both variables $\lambda,\mu$.
In Sec.~\ref{BasicEquation} the equations~(\ref{basic}) are studied. As in
Refs.~\onlinecite{MalyshevMinlos1},~\onlinecite{MalyshevMinlos2}
we suppose that
spectra $\sigma(A_1)$ and $\sigma(A_2)$ of the operators $A_1$ and
$A_2$ are separated, ${\rm dist}\{ \sigma(A_1), \sigma(A_2) \}>0$.
Existence of solutions of Eqs.~(\ref{basic}) is established
only in the case where the Hilbert-Schmidt norms $\| B_{\alpha\beta}\|_2$
of the coupling operators $B_{\alpha\beta}$ satisfy
the condition
$\| B_{\alpha\beta}\|_2 < \frac{1}{2}{\rm dist}
\{ \sigma(A_1),\sigma(A_2) \} $.
In Sec.~\ref{Eigenfunctions} the eigenfunctions systems
of the operators $H_{\alpha}$ are studied and theorems of their
orthogonality and completeness are proved. We show here
in particular that spectrum of the Hamiltonian ${{\bf H}}$ is distributed
between the solutions $H_1=A_1 + B_{12} Q_{21}$ and
$H_2=A_2 + B_{21} Q_{12}$, $Q_{21}=-Q_{12}^{*}, $
of the basic equations~(\ref{basic})
in such a way that $H_1$ and $H_2$ have not ``common'' eigenfunctions
$U=\{u_1, u_2\} $ of ${{\bf H}}$: simultaneously, the component $u_1$ can
not be eigenfunction for $H_1$, and the component $u_2$, for $H_2$.
In Sec.~\ref{InnerProduct} we introduce new inner products
in the Hilbert spaces ${\cal H}_{\alpha}$, ${\alpha}=1,2,$
making the Hamiltonians $H_{\alpha}$ self-adjoint.
In Sec.~\ref{ScatteringProblem} we give a non-stationary formulation
of the scattering problem for a system described by the Hamiltonian
$H_{\alpha}$ constructed in the case where $A_\alpha$ has continuous
spectrum. We show that this formulation is correct and scattering
operator is exactly the same as in initial spectral problem.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{ INITIAL SPECTRAL PROBLEM AND TWO-CHANNEL HAMILTONIAN}
\label{Initial}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Let $A_{1}$ and $A_{2}$ be self-adjoint operators acting,
respectively, in ``external'', ${{\cal H}} _{1}$, and ``internal'',
${{\cal H}} _{2}$, Hilbert spaces.
We study the spectral problem~(\ref{ini})
with perturbation $V_{\alpha}(z)$ given by~(\ref{epot}).
We suppose that
$B_{\alpha\beta} \in {{\bf B}} ({{\cal H}} _{\alpha} ,
{{\cal H}} _{\beta} )$ where
${{\bf B}} ({{\cal H}} _{\alpha} ,{{\cal H}} _{\beta} )$ is the
Banach space of bounded linear
operators acting from ${{\cal H}} _{\alpha} $ to ${{\cal H}} _{\beta} $.
Note that the method developed in the present paper works also in the
case of more general perturbations$^{\onlinecite{RfunfNaboko}}$
$V_{\alpha}(z)=-{\cal R}_{\alpha}(z)$ containing linear terms,
%
\be
\label{Rfunction}
%
{\cal R}_{\alpha}(z)= N_{\alpha} z +
B_{\alpha\beta}(A_{\beta}-N_{\beta} z -z)^{-1}B_{\beta\alpha}
%
\ee
with $N_{\alpha}$, self-adjoint bounded operator
in ${\cal H}_{\alpha}$ such that
$N_{\alpha}\geq ({\delta} -1)I_{\alpha}$ where
$\,\,{\delta} > 0 $ and $I_{\alpha}$
is the identity operator in ${\cal H}_{\alpha}$.
Thing is that the equation~(\ref{ini}) with
$V_{\alpha}(z)=-{\cal R}_{\alpha}(z)$
can be easily rewritten in the form
(\ref{ini}),(\ref{epot}). To do this,
one has only to make the replacements
$u_{\alpha}\rightarrow u_{\alpha}'=
(I_{\alpha}+N_{\alpha})^{1/2}u_{\alpha},$
$
\,\, A_{\alpha}\rightarrow A_{\alpha}'=
(I_{\alpha}+N_{\alpha})^{-1/2} A_{\alpha} (I_{\alpha}+N_{\alpha})^{-1/2}
$
and
$
B_{\alpha\beta}\rightarrow B_{\alpha\beta}'=
%
(I_{\alpha}+N_{\alpha})^{-1/2}B_{\alpha\beta}
(I_{\beta}+N_{\beta})^{-1/2}
$.
Therefore we shall consider further only the initial spectral problem
(\ref{ini}),(\ref{epot}).
We shall assume that operators $A_{\alpha}$, ${\alpha}=1,2,$
may have continuous
spectra $\sigma_{\alpha}^c$. To deal with
these spectra we accept below some
presuppositions in respect with $A_{\alpha}$
restricting us to the case of
a two-channel variant of the Friedrichs
model~\cite{LadyzhFaddeev}, \cite{Faddeev64}.
Note that these presuppositions are not necessary
for a part of statements (Lemma~\ref{LQdefinition},
Theorems~\ref{ThSolvability}--\ref{ThDiscSpectrum}
and~\ref{ThInnerProduct}) which stay
correct also in general case.
The presuppositions are following.
At first, we assume that
Hamiltonian $H$ is defined in that representation where
operators $A_{\alpha} , {\alpha} =1,2,$ are diagonal. We suppose
that continuous spectra $\sigma ^{c}_{\alpha} $
of the operators $A_{\alpha} ,
{\alpha} =1,2,$ are absolutely continuous and consist of a finite
number of finite (and may be one or two infinite) intervals
$(a_{\alpha}^{(j)} ,b_{\alpha}^{(j)} )$,
$ -\infty\leq a_{\alpha}^{(j)} < b_{\alpha}^{(j)} \leq +\infty$,
$j=1,2,\ldots,n_{\alpha},$ $n_{\alpha} < \infty$.
At second, we suppose that discrete spectra $\sigma_{\alpha}^d$ of
the operators $A_{\alpha}$, ${\alpha}=1,2,$
do not intersect with $\sigma_{\alpha}^c$,
$\sigma_{\alpha}^d \bigcap\sigma_{\alpha}^c =\emptyset$,
and consist of a finite number of points with finite multiplicity.
In this case the
space ${{\cal H}} _{\alpha} $ may be present as the direct
integral~\cite{BirmanSolomiak}
\be
%
\label{Neumann}
%
{{\cal H}} _{\alpha}
%
=\sum\mbox{\hskip-2em}\int\limits_{\lambda\in\sigma_{\alpha} }
%
\oplus {{\cal G}} _{\alpha} (\lambda )d\lambda
%
\equiv \mathop{\oplus}\limits_{\lambda\in\sigma^{d}_{\alpha} }
%
{{\cal G}} _{\alpha} (\lambda )\oplus
%
\int\limits_{\lambda\in\sigma_{\alpha}^c}
%
\oplus{{\cal G}} _{\alpha} (\lambda )d\lambda , \quad
%
\sigma_{\alpha} =\sigma^{c}_{\alpha}\bigcup
\sigma^{d}_{\alpha}\subset{\bf R} .
%
\ee
The space ${{\cal H}} _{\alpha} $
consists of the
measurable functions $f_{\alpha} $
which are defined on $\sigma _{\alpha} $
and
have the values
$f_{\alpha} (\lambda )$
from corresponding Hilbert
spaces ${{\cal G}} _{\alpha} (\lambda )$.
By $\langle\,\cdot\, ,\,\cdot\,\rangle$
we denote the inner product in
%
${{\cal H}} _{\alpha} ,\qquad $
%
$$
%
\langle f_{\alpha} ,g_{\alpha} \rangle=
%
\mathop{\mbox{$\sum$\hskip-1em{\LARGE $\int$}\hskip0.1em}}
\limits_{\lambda\in\sigma_{\alpha} }
%
(f_{\alpha} (\lambda ),g_{\alpha} (\lambda ))
%
\equiv \sum\limits_{\lambda\in\sigma^{d}_{\alpha} }
%
(f_{\alpha} (\lambda ),g_{\alpha} (\lambda ))
%
+\int\limits_{\lambda\in\sigma_{\alpha}^c}
%
d\lambda (f_{\alpha} (\lambda ),g_{\alpha} (\lambda )),
%
$$
where $(\,\cdot\, ,\,\cdot\,)$ stands for inner product in
${{\cal G}} _{\alpha} (\lambda )$. By $|\cdot| $ we denote
norm of vectors and
operators in ${{\cal G}} _{\alpha} (\lambda )$ and by
$\parallel\cdot\parallel $,
the norm in ${{\cal H}} _{\alpha} $.
Operator $A_{\alpha} $
acts in ${{\cal H}} _{\alpha} $
as the independent variable multiplication operator,
\be
%
\label{multi}
%
(A_{\alpha} f_{\alpha} )(\lambda )=
%
\lambda \cdot f_{\alpha} (\lambda ), \quad {\alpha} =1,2.
%
\ee
It's domain
%
${\cal D}(A_{\alpha} )$
%
consists of those functions $f_{\alpha} \in {{\cal H}} _{\alpha} $
which satisfy the condition
\newline
$\mathop{\mbox{$\sum$\hskip-1em{\LARGE $\int$}\hskip0.1em}}
\limits_{\lambda\in\sigma _{\alpha} }\lambda ^{2}$
$|f_{\alpha} (\lambda )|^{2}<\infty $. For the
sake of simplicity we assume that
${{\cal G}} _{\alpha}(\lambda )$ does not depend on
$\lambda\in\sigma_{\alpha}^c$, i.e.
$
%
{{\cal G}} _{\alpha}(\lambda )\equiv {{\cal G}} ^{c}_{\alpha}
$
for each
$
\lambda \in \sigma^{c}_{\alpha} .
$
Hence,
$
\int\limits_{\sigma ^{c}_{\alpha} }
%
\oplus {{\cal G}} _{\alpha} (\lambda )d\lambda =
%
L_{2}(\sigma ^{c}_{\alpha} ,{{\cal G}} ^{c}_{\alpha} )
%
\equiv {{\cal H}} ^{c}_{\alpha}
%
$.
By $E_{\alpha} (d\lambda )$ we denote a
spectral measure~\cite{BirmanSolomiak} of
the operator $A_{\alpha} ,$
$
A_{\alpha} =\int\limits_{\sigma_{\alpha}}\lambda
E_{\alpha} (d\lambda )
$.
In the diagonal representation considered,
the spectral projector $E_{\alpha} $ acts on
$f\in {{\cal H}} _{\alpha} $ as
\be
%
\label{mera}
%
(E_{\alpha} (\Delta )f)(\lambda )=\chi_{\Delta }(\lambda )f(\lambda )
%
\ee
for any Borelian set
$
\Delta \subset \sigma _{\alpha}
$.
Here, $\chi _{\Delta }$
is a characteristic function of
$\Delta ,$
$ \chi _{\Delta }(\lambda )=1$
if
$\lambda \in \Delta $,
and $\chi _{\Delta }(\lambda )=0$
if
$\lambda \not\in \Delta $.
Let ${{\cal B}} ^{{\alpha}{\beta}}_{\theta\gamma}$
be a class of functions $F$ defined on
$\sigma _{\alpha} \times \sigma _{\beta} ,$ ${\alpha},{\beta}=1,2$,
for each $\lambda \in \sigma _{\alpha} , \mu \in \sigma _{\beta} $
as operator
$F(\lambda ,\mu ):$
${{\cal G}} _{\beta} (\mu )\rightarrow {{\cal G}} _{\alpha} (\lambda )$,
with $\parallel F\parallel_{{{\cal B}} } < \infty $, where
$$
\parallel F\parallel _{{{\cal B}} } =
%
\mathop{\rm sup}\limits_{
%
\mbox{
\scriptsize
$\begin{array}{c}
%
\mu\in\sigma_{\beta}\\
\lambda\in\sigma_{\alpha}
%
\end{array}$
}
}(1+|\lambda |)^{\theta }
%
(1+|\mu |)^{\theta }| F(\lambda ,\mu ) | +
$$
$$+\mathop{\rm sup}\limits_{
%
\mbox{
\scriptsize
$\begin{array}{c}
%
\lambda,\lambda'\in\sigma_{\alpha}^c\\
\mu\in\sigma_{\beta}
%
\end{array}$
}
}
\left\{
%
(1+|\mu |)^{\theta}
%
\frac{|F(\lambda ,\mu )-F(\lambda ',\mu )|}
{|\lambda -\lambda '|^{\gamma }}
%
\right\}+
$$
$$
%
+\mathop{\rm sup}\limits_{
%
\mbox{
\scriptsize
$\begin{array}{c}
%
\lambda\in\sigma_{\alpha}\\
\mu,\mu'\in\sigma_{\beta}^c
%
\end{array}$
}
}
\left\{ (1+|\lambda|)^{\theta }
%
\frac{|F(\lambda ,\mu )-F(\lambda ,\mu ')|}{|\mu-\mu'|^{\gamma }}
%
\right\}+
%
$$
$$
%
+\mathop{\rm sup}\limits_{
%
\mbox{
\scriptsize
$\begin{array}{c}
%
\lambda,\lambda'\in\sigma_{\alpha}^c\\
\mu,\mu'\in\sigma_{\beta}^c
%
\end{array}$
} }
%
\left\{ \frac{|F(\lambda,\mu)-F(\lambda',\mu)-
%
F(\lambda,\mu')+F(\lambda',\mu')|}
%
{|\lambda-\lambda'|^{\gamma}|\mu-\mu'|^{\gamma }}
%
\right\} .
$$
With the norm $\parallel\cdot\parallel_{{{\cal B}} }$
this class will constitute a Banach space.
We introduce also the Banach space
${\cal M}_{\theta \gamma }(\sigma _{\alpha} )$ of
functions $f$ defined on $\sigma _{\alpha} $ with the norm
$$
\parallel f\parallel_{{\cal M}}
%
=\mathop{\rm sup}\limits_{\lambda\in\sigma_{\alpha}}
(1+|\lambda |)^{\theta }|f(\lambda)|
%
+\mathop{\rm sup}\limits_{\lambda,\lambda'\in\sigma_{\alpha}^c}
%
\frac{|f(\lambda ) - f(\lambda ')|}
%
{|\lambda -\lambda '|^{\gamma }}<\infty .
$$
The value $f(\lambda )$
of the function $f\in {\cal M}_{\theta \gamma }(\sigma _{\alpha} )$
is an operator in ${{\cal G}} _{\alpha} (\lambda )$.
Let $B_{\alpha\beta} $ be an integral operator with a
kernel $B_{\alpha\beta} (\lambda ,\mu )$
from the space ${{\cal B}} ^{{\alpha} {\beta} }_{\theta \gamma },$
$\theta >{1\over 2}, {1\over 2}<\gamma <1.$ We assume that
$B_{\alpha\beta} (\lambda ,\mu )$ is
a compact operator, $B_{\alpha\beta} (\lambda ,\mu ):$
${{\cal G}} _{\beta} (\mu )\rightarrow {{\cal G}} _{\alpha} (\lambda )$,
for each $\lambda \in \sigma _{\alpha} , \mu \in \sigma _{\beta} $ and
$B_{\alpha\beta} (\lambda ,\mu )=0$
if $\lambda $ belongs to the boundary of $\sigma ^{c}_{\alpha} $
or $\mu $ belongs to the
boundary of $\sigma ^{c}_{\beta} $.
With this presuppositions the Hamiltonian ${{\bf H}}$
may be considered as a two-channel variant of the Friedrichs
model~\cite{LadyzhFaddeev}, \cite{Faddeev64}. Investigation of ${{\bf H}}$
repeats almost literally the analysis from
Ref.~\onlinecite{Faddeev64}. Therefore we describe
here only final results which are quite analogous to
\cite{LadyzhFaddeev}, \cite{Faddeev64}. These results are
following.
The operator ${{\bf H}}$ is self-adjoint on the set
${\cal D}({{\bf H}})={\cal D}(A_{1})\oplus {\cal D}(A_{2})$.
Continuous spectrum of ${{\bf H}}$ is
situated on the set $\sigma _{c}({{\bf H}})=
\sigma ^{c}_{1}\cup \sigma ^{c}_{2}$.
Let ${{\bf H}}^{c}$ be the part of ${{\bf H}}$
acting in the invariant subspace corresponding to
continuous spectrum. The operator ${{\bf H}}^{c}$
is unitary equivalent to the operator
${{\bf H}}_{0}=A^{(0)}_{1}\oplus A^{(0)}_{2}$ with
$A^{(0)}_{\alpha},$ ${\alpha}=1,2,$ the restriction of the operator
$A_{\alpha}$ on ${\cal H}^{c}_{\alpha}$.
Namely, there exist wave operators $U^{(+)}$ and $U^{(-)}, $
$
%
U^{(\pm )}=
%
\left(
%
\begin{array}{lr}
%
u^{(\pm )}_{11} & u^{(\pm )}_{12} \\
%
u^{(\pm )}_{21} & u^{(\pm )}_{22}
%
\end{array}
%
\right)=
%
$
%
%
$
%
s\!-\!\mathop{\rm lim}\limits_{t\rightarrow\mp\infty}
{\rm e}^{i{{\bf H}}t}{\rm e}^{-i{{\bf H}}_0 t},
%
$
with the following properties:
%
${{\bf H}}U^{(\pm )}=U^{(\pm )}{{\bf H}}_{0},$
$U^{(\pm )*}U^{(\pm )}=I,$
%
$U^{(\pm )}U^{(\pm )*}= I-P.$
%
Here, $P$ is an
orthogonal projector on subspace
corresponding to the discrete spectrum $\sigma _{d}({{\bf H}})$
of the operator ${{\bf H}}$.
The kernel $u^{(\pm )}_{{\alpha} {\alpha} }(\lambda ,\lambda ')$
of the operator $u^{(\pm )}_{{\alpha} {\alpha} },{\alpha} =1,2,$
represents an eigenfunction of the continuous spectrum
of the problem~(\ref{ini}) for
$
%
z=\lambda '\pm i0,
%
$
$
%
\lambda '\in \sigma ^{c}_{\alpha}
%
$, and satisfies the integral equation
\be
\label{wavefunction}
%
u^{(\pm )}_{{\alpha} {\alpha} }(\lambda ,\lambda ')=
%
I_{\alpha}^c {\delta} (\lambda -\lambda ')-
%
[(A_{\alpha} -\lambda '\mp i0)^{-1}
%
V_{\alpha} (\lambda '\pm i0)u^{(\pm )}_{{\alpha} {\alpha} }]
%
(\lambda ,\lambda '),
%
\ee
where $I_{\alpha}^c $
is identity operator in
$
{{\cal G}} ^{c}_{\alpha} , \lambda \in \sigma _{\alpha}
$.
For each concrete sign (plus or minus) and for each
$
\lambda '\in \sigma ^{c}_{\alpha},
\lambda '\not\in \sigma _{d}({{\bf H}})
$
the function
$
u^{(\pm )}_{{\alpha} {\alpha} }(\lambda ,\lambda ')
$
is an unique solution of eq.(\ref{wavefunction})
in the class of the distributions
\be
%
\label{amplitude}
%
f^{(\pm )}_{\alpha} (\lambda )=
%
I_{\alpha} {\delta} (\lambda -\lambda ')+
%
{f(\lambda )\over \lambda -\lambda '\mp i0},
%
\quad f\in {\cal M}_{\theta '\gamma '},
%
\ee
where
$
{1\over 2}<\theta '<\theta ,
$
$
{1\over 2}<\gamma '<\gamma
$.
At the same time
$$
%
u^{(\pm )}_{\alpha\beta} (\lambda ,\lambda ')=
%
-[(A_{\alpha} -\lambda '\mp i0)^{-1}
%
B_{\alpha\beta} u^{(\pm )}_{{\beta} {\beta} }](\lambda ,\lambda '),
%
\quad {\beta} \neq {\alpha} ,
%
$$
is the problem ~(\ref{ini}) eigenfunction corresponding to
$\lambda '\in \sigma_{\beta}^c$.
The functions $u_{{\beta}{\alpha}}^{(\pm)}$,
${\alpha},{\beta}=1,2,$ can be explicitly
expressed in terms of kernels of the operator
$$
%
T(z)=B-B({{\bf H}}-z)^{-1}B,\quad
%
B=\left[\begin{array}{lr} 0 & B_{12} \\
B_{21} & 0 \end{array}\right].
%
$$
Corresponding formulae read as
$$
%
u_{{\beta}{\alpha}}^{(\pm)}(\lambda,\lambda')=
%
{\delta} _{{\beta}{\alpha}}I_{\alpha}^c{\delta} (\lambda-\lambda')-
%
\frac{T_{{\beta}{\alpha}}(\mu,\lambda',\lambda'\pm i0)}
%
{\mu-\lambda'\mp i0}, \quad
%
\mu\in\sigma_{\beta},\,\, \lambda'\in\sigma_{\alpha}^c,
%
$$
with $t$-matrices
$$
%
T_{{\alpha}{\alpha}}=B_{\alpha\beta}
\left[z-A_{\beta}+B_{{\beta}{\alpha}}
(A_{\alpha}-z)^{-1}B_{\alpha\beta}\right]^{-1}
%
B_{{\beta}{\alpha}}
%
$$
and
$$
%
T_{{\beta}{\alpha}}=B_{{\beta}{\alpha}}
\left[z-A_{\alpha}+B_{{\alpha}{\beta}}(A_{\beta}-z)^{-1}
%
B_{{\beta}{\alpha}}\right]^{-1}(z-A_{\alpha})=
%
$$
%
$$
%
=(z-A_{\beta})\left[z-A_{\beta}+
B_{{\beta}{\alpha}}(A_{\alpha}-z)^{-1}B_{\alpha\beta}\right]^{-1}
%
B_{{\beta}{\alpha}},\quad {\beta}\neq{\alpha}.
%
$$
Considering the equation for $T(z)$,
$T(z)=B-B(A-z)^{-1}T(z)$, $A=A_1\oplus A_2$, one shows
in the same way as in~\cite{LadyzhFaddeev}, \cite{Faddeev64}
that for all $z\in{\bf C}\setminus\sigma({{\bf H}})$, each kernel
$ T_{{\beta}{\alpha}}(\mu,\lambda,z)$, ${\alpha},{\beta}=1,2,$
belongs to the class
${{\cal B}} _{\theta'\gamma'}^{{\beta}{\alpha}}$
with arbitrary $\theta',\gamma'$
such that $\frac{1}{2}<\theta'<\theta$,
$\frac{1}{2}<\gamma'<\gamma$.
In respect with variable $z$,
the kernel of $T_{{\beta}{\alpha}}(z)$
is continuous in the
${{\cal B}} _{\theta'\gamma'}^{{\beta}{\alpha}}$--norm
right up to the upper and lower borders of the set
$\sigma_c({{\bf H}})\setminus\sigma_d({{\bf H}})$.
Scattering operator $S=U^{(-)*}U^{(+)}$ for a system described by the
Hamiltonian ${{\bf H}}$ is unitary
in ${\cal H}_{\alpha}^c$. It's kernels
$s_{{\beta}{\alpha}}(\mu,\lambda)$,
${\alpha},{\beta}=1,2$, are given by expressions
\be
\label{scattering}
%
s_{{\beta}{\alpha}}(\mu,\lambda)={\delta} (\mu-\lambda)
%
\left[ {\delta} _{{\beta}{\alpha}}I_{\alpha}^c-
%
2\pi i\, T_{{\beta}{\alpha}}(\mu,\lambda,\lambda+i0) \right].
%
\ee
By $U_{j},$ $j=1,2,\ldots$,
we denote eigenvectors,
$U_{j}=\{u^{(j)} _{1},u^{(j)} _{2}\},$
$ U_{j}\in {\cal D}({{\bf H}})$,
$\parallel U_{j} \parallel =1,$ and by
$z_{j},$ $ z_{j}\in {\bf R}$, the
respective eigenvalues of the operator ${{\bf H}}$ discrete
spectrum $\sigma_d({{\bf H}})$. The component
$u^{(j)} _{\alpha} ,$ ${\alpha} =1,2,$ of the vector $U_{j}$
is a solution of Eq.~(\ref{ini}) at $z=z_{j}$.
If $z_{j}\!\in\!\sigma ^{c}_{\beta} $ then
$(B_{\beta\alpha} u^{(j)} _{\alpha} )(z_{j})=0.$
%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{ CONSTRUCTION OF THE OPERATORS $H_\alpha$}
\label{BasicEquation}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
The paper is devoted to construction of such
operator $H_{\alpha} $
that it's each eigenfunction
$u_{\alpha} , H_{\alpha} u_{\alpha} =zu_{\alpha} $,
together with eigenvalue $z$, satisfies Eq.~(\ref{ini}). This
operator will be found as a solution of the non-linear
operator equation~(\ref{basic}). To obtain this equation we need
the following operator-value function $V_{\alpha} (Y)$
of the operator
variable $Y:\quad$
%%%%%
$$
%
V_{\alpha} (Y)=B_{\alpha\beta}\int\limits_{\sigma _{\beta} }
%
E_{\beta} (d\mu )B_{\beta\alpha} (Y-\mu )^{-1},
%
$$
%%%%%
$Y:$ ${{\cal H}} _{\alpha} \rightarrow {{\cal H}} _{\alpha} $ .
We suppose here that
%%%%%
$
%
(Y-\mu I)^{-1}\in L_{\infty }
%
(\sigma _{\beta} ,{{\bf B}}({{\cal H}}_{\alpha},{{\cal H}}_{\alpha}))
%
$
if $\mu \in \sigma _{\beta} $.
This means that $\sigma_{\beta}$ has not
to be included into the spectrum of
the operator $Y$. The integral
$
%
Q(T)=\int\limits_{\sigma _{\beta} }
%
E_{\beta} (d\mu )B_{\beta\alpha} T(\mu )
%
$
may be constructed in the same way as integrals of scalar functions
over spectral measure~\cite{BirmanSolomiak}.
Namely, we consider $Q(T)$ for
$
%
T\in L_{\infty }(\sigma _{\beta} ,{{\bf B}} ({{\cal H}} _{\alpha} ,
%
{{\cal H}} _{\alpha} )),
%
$
$
%
\|{T}\|_{\infty }=
$
$
E_{\beta}\!-\!\mathop{\rm sup}\limits_{\mu\in\sigma_{\beta}}
\|T(\mu)\|<\infty
$
(hereafter $E_{\beta}\!-\!\mathop{\rm sup}\limits_{}$
means the supremum with respect to $E_\beta$),
as a limit value, with respect to the operator
norm in ${{\bf B}}({{\cal H}}_{\alpha},{{\cal H}}_{\alpha})$,
of finite integral sums for piecewise--constant
operator--value functions
approximating $T$ in
$
%
L_{\infty }(\sigma _{\beta} ,{{\bf B}}
({{\cal H}} _{\alpha} ,{{\cal H}} _{\alpha} ))
%
$.
We show the existence of this integral at least in the case where
the Hilbert-Schmidt norm $\|{B}_{{\alpha}{\beta}}\|_{2}$ is finite.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{lemma}\label{LQdefinition}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Let
$
%
T\in L_{\infty }(\sigma _{\beta} ,
%
{{\bf B}} ({{\cal H}} _{\alpha} ,{{\cal H}} _{\alpha} ))
%
$
and
%
$
%
\parallel B_{\alpha\beta} \parallel_{2}<\infty $.
Then the
integral $Q(T)$ exists as a bounded operator in ${\cal H}_\alpha$
with norm satisfying the estimate
$
\|Q(T)\|\le \| T\|_{\infty }\!\cdot\!\|B_{\beta\alpha}\|_{2}
$.
\end{lemma}
\noindent{\sl Proof}. We prove the Lemma in the
diagonal representation~(\ref{Neumann}),~(\ref{multi}).
By~(\ref{mera}) we have
$$
%
(Qf)(\mu )=
%
\mathop{\mbox{$\sum$\hskip-1em{\LARGE $\int$}\hskip0.1em}}
\limits_{\sigma_{\alpha}} B_{\beta\alpha} (\mu ,\lambda )
%
(T(\mu )f)(\lambda) d\lambda
%
$$
for
any $f\in {{\cal H}} _{\alpha} $.
It means that
$$
%
| (Qf)(\mu )|^{2}\!\le\!\sum
\mbox{\hskip-2em}\int\limits_{\lambda\in\sigma_{\alpha}}
%
d\lambda \cdot
%
| B_{\beta\alpha} (\mu,\lambda )|^{2}
%
\sum\mbox{\hskip-2em}\int\limits_{\lambda\in\sigma_{\alpha}}d\lambda
|({T(\mu)}f)(\lambda)|^{2}=
%
$$
%
$$
%
=\sum\mbox{\hskip-2em}
\int\limits_{\lambda\in\sigma_{\alpha}}
d\lambda|B_{{\beta}{\alpha}}(\mu,\lambda)|^{2}\cdot
%
\|T(\mu)f\|^2
%
\leq \sum\mbox{\hskip-2em}
\int\limits_{\lambda\in\sigma_{\alpha}}
d\lambda|B_{{\beta}{\alpha}}(\mu,\lambda)|^{2}\cdot
%
\|T(\mu)\|^{2}\cdot\|f\|^2.
%
$$
Hence,
integrating over
%
$\mu \in \sigma_{\beta} $
%
we come to the relation
%%
$$
%
\parallel{Qf}\parallel^{2}\le
\parallel{B_{{\beta}{\alpha}}}\parallel^2_2\!\cdot\!
%
\parallel{T}\parallel^{2}_{\infty }\!\cdot\!
%
\parallel{f}\parallel^{2}
%
$$
%%
which completes the proof.
\medskip
Let us suppose that
%%
$
%
(H_{\alpha} -\mu I)^{-1}\in L_{\infty }
%
(\sigma _{\beta} ,{{\bf B}}
({{\cal H}} _{\alpha} ,{{\cal H}} _{\alpha} ))
%
$.
%%
We note that if
%
$H_{\alpha} \psi_{\alpha} =z\psi_{\alpha} $,
%
then automatically
%
$$
%
V_{\alpha} (H_{\alpha} )\psi_{\alpha} =
%
B_{\alpha\beta}\int\limits_{\sigma _{\beta} }
%
E_{\beta} (d\mu )
%
B_{\beta\alpha} (z-\mu )^{-1}\psi_{\alpha} =
$$
\be
\label{Vaction}
%
=B_{\alpha\beta} (z-A_{\beta} )^{-1}B_{\beta\alpha} \psi_{\alpha}
%
=V_{\alpha} (z) \psi_{\alpha} .
%
\ee
It follows from~(\ref{Vaction})
that $H_{\alpha} $ satisfies the relation
$
H_{\alpha} \psi_{\alpha}
%
=(A_{\alpha} +V_{\alpha} (H_{\alpha} )) \psi_{\alpha}
$
and we can spread this relation
over all the linear combinations of $H_{\alpha} $ eigenfunctions.
Supposing that the eigenfunctions system of $H_{\alpha} $ is
dense in ${\cal H}_{\alpha}$ we spread this
equation over ${\cal D}(A_{\alpha} )$. As a result
we come to the desired {\it basic equation}
(\ref{basic})
for $H_{\alpha} $ (see also Refs.~\onlinecite{Braun},
\onlinecite{MotJMPh91}--\onlinecite{SPbWorkshop}).
Eq.~(\ref{basic}) means that the construction
of the operator $H_{\alpha} $
comes to the searching for the operator
\be
\label{Qbasic}
%
Q_{\beta\alpha}
%
=\int\limits_{\sigma_{\beta}} E_{\beta} (d\mu )
%
B_{\beta\alpha} (H_{\alpha} -\mu )^{-1}.
%
\ee
Since
$
H_{\alpha} =A_{\alpha} +B_{\alpha\beta} Q_{\beta\alpha}
$,
we have
\be
\label{Qbasic1}
%
Q_{\beta\alpha} =
%
\int\limits_{\sigma _{\beta} }E_{\beta} (d\mu )
%
B_{\beta\alpha} (A_{\alpha} +
%
B_{\alpha\beta} Q_{\beta\alpha} -\mu )^{-1},
%
\quad {\beta} \neq {\alpha}.
%
\ee
%
In this paper we restrict ourselves to the study
of Eq.~(\ref{Qbasic1})
solvability only in the case where
spectra $\sigma_{1}$ and $\sigma_{2}$
are separated,
\be
\label{dist}
%
d_{0}={\rm dist}(\sigma _{1},\sigma _{2})>0.
%
\ee
Using the Lemma~\ref{LQdefinition} and
the contracting mapping Theorem,
we prove the following:
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{theorem}\label{ThSolvability}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Let $M_{\beta\alpha} ({\delta} )$
%
be a set of bounded operators
$X,$ $X:$
${{\cal H}} _{\alpha} \rightarrow {{\cal H}} _{\beta} $,
satisfying the inequality
$ \|{X}\| $ $ \le{\delta} $ with ${\delta} > 0$.
If this ${\delta} $ and the norm
$\|{B_{\alpha\beta} }\|_{2}$
%
satisfy the condition
$
%
\| B_{\alpha\beta}\|_{2}< d_{0}
%
\mathop{\rm min}\limits_{}
\{ \frac{1}{1+{\delta} },
\frac{{\delta} }{1+{\delta} ^2} \}$,
then Eq.~(\ref{Qbasic1})
is uniquely
solvable in
$M_{\beta\alpha} ({\delta} )$.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\end{theorem}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\noindent{\sl Proof.} Let
\be
\label{funF}
%
F(X)=\int\limits_{\sigma_{\beta}}
E_{\beta}(d\mu)B_{{\beta}{\alpha}}
(A_{\alpha}+B_{\alpha\beta} X-\mu)^{-1}
%
\ee
with $X$, the operator from
${{\bf B}} ({\cal H}_{\alpha},{\cal H}_{\beta})$.
Firstly, consider conditions where the function $F$ maps the set
$M_{\alpha}({\delta} )$ into itself. We suppose here that
$B_{\alpha\beta}$ and $X$ are such
that
\be
\label{ne1}
%
\|B_{\alpha\beta}\|_2 \|X\|
\leq {\delta} \|B_{\alpha\beta}\|_2 < d_0
%
\ee
and consequently, $\|B_{\alpha\beta} X\|\leq d_0 . $
This means that spectrum of the operator
$A_{\alpha} +B_{\alpha\beta} X$
does not intersect with the set $\sigma_{{\beta}}$.
Hence, the resolvent
$(A_{\alpha} +B_{\alpha\beta} X-\mu)^{-1}$
exists and is bounded for any $\mu\in\sigma_{\beta}$.
Thus, by Lemma~\ref{LQdefinition} we have
$$
%
\|F(X)\| \leq \|B_{\alpha\beta}\|_2\cdot
E_{\beta}\!-\!\mathop{\rm sup}\limits_{\mu\in\sigma_{\beta}}
%
\|(A_{\alpha} +B_{\alpha\beta} X-\mu)^{-1}\|.
%
$$
Due to identity
$$
%
(A_{\alpha} +B_{\alpha\beta} X-\mu)^{-1}=
\left( I+(A_{\alpha}-\mu)^{-1}B_{\alpha\beta} X\right)^{-1}
%
(A_{\alpha}-\mu)^{-1}
%
$$
and inequality $ \|B_{\alpha\beta}\| \leq \|B_{\alpha\beta}\|_2 $
we make estimation
$$
%
\| (A_{\alpha} +B_{\alpha\beta} X-\mu)^{-1} \| \leq
%
\frac{1}{ 1 - \| (A_{\alpha}-\mu)^{-1}\| \|B_{\alpha\beta}\|_2 \|X\| }
%
\|(A_{\alpha}-\mu)^{-1}\| \leq
%
$$
\be
\label{estires}
%
\leq \frac{1}{ 1-\frac{1}{d_0} \|B_{\alpha\beta}\|_2 {\delta} }\cdot
%
\frac{1}{d_0}=
%
\frac{1}{d_0 - \|B_{\alpha\beta}\|_2 {\delta} }.
%
\ee
Therefore, the set $M_{\alpha}({\delta} )$
will be mapped by $F$ into itself if
$\|B_{\alpha\beta}\|_2$ and ${\delta} $ are such that
\be
\label{ne2}
%
\|B_{\alpha\beta}\|_2 \cdot
\frac{1}{d_0 - \|B_{\alpha\beta}\|_2{\delta} } \leq {\delta} .
%
\ee
Secondly, study conditions for the
function $F$ to be a contracting mapping.
Now, we consider the difference
$$
%
F(X)-F(Y)=\int\limits_{\sigma_{\beta}}
E_{\beta}(d\mu) B_{\beta\alpha}
%
\left[ (A_{\alpha}+B_{\alpha\beta} X -\mu)^{-1}
- (A_{\alpha}+B_{\alpha\beta} Y -\mu)^{-1} \right]=
%
$$
%
$$
%
=\int\limits_{\sigma_{\beta}} E_{\beta}(d\mu) B_{\beta\alpha}
%
(A_{\alpha}+B_{\alpha\beta} X -\mu)^{-1}
B_{\alpha\beta} (Y-X) (A_{\alpha}+B_{\alpha\beta} Y -\mu)^{-1}.
%
$$
Again, by Lemma~\ref{LQdefinition}, we have
$$
\|F(X)-F(Y)\| \leq
$$
%
$$
%
\leq \|B_{\alpha\beta}\|_2^2\cdot
\mathop{\rm sup}\limits_{\mu\in\sigma_{\beta}}
%
\| (A_{\alpha}+B_{\alpha\beta} X -\mu)^{-1}\| \cdot
%
\mathop{\rm sup}\limits_{\mu\in\sigma_{\beta}}
\| (A_{\alpha}+B_{\alpha\beta} Y -\mu)^{-1}\| \cdot \|(Y-X)\| .
%
$$
With~(\ref{estires}) we come to the estimate
$$
%
\|F(X)-F(Y)\| \leq \|B_{\alpha\beta}\|_2^2 \cdot
%
\frac{1}{(d_0 -\|B_{\alpha\beta}\|_2{\delta} )^2}\cdot \|Y-X\|.
$$
The function $F$ becomes a contracting mapping if
\be
\label{ne3}
%
\frac{\|B_{\alpha\beta}\|_2^2}
{(d_0 -\|B_{\alpha\beta}\|_2{\delta} )^2} < 1.
%
\ee
Solving system of the inequalities~(\ref{ne1}),~(\ref{ne2}) and
(\ref{ne3}) we find
$$
%
\|B_{\alpha\beta}\|_2 < d_0 \mathop{\rm min}\limits_{} \left\{
%
\frac{{\delta} }{1+{\delta} ^2}, \frac{1}{1+{\delta} } \right\}
%
$$
and this completes the proof of Theorem~\ref{ThSolvability}.
\begin{corollary}\label{CBestimate}
%
Equation (\ref{Qbasic1})
is uniquely solvable in the unit
ball $M_{\beta\alpha}(1)\subset$
${{\bf B}} ({\cal H}_{\alpha},{\cal H}_{\beta})$
for any $B_{\alpha\beta}$ such that
\be
\label{ineqB}
%
\|B_{\alpha\beta}\|_2 < \frac{1}{2} d_0.
%
\ee
\end{corollary}
To prove the inequality~(\ref{ineqB}), note that
$
%
\mathop{\rm max}\limits_{{\delta} \geq 0}\,
\mathop{\rm min}\limits_{} \left\{
%
\frac{{\delta} }{1+{\delta} ^2}, \frac{1}{1+{\delta} } \right\}
%
=\frac{1}{2}
%
$
(at ${\delta} =1$). Hence, if~(\ref{ineqB})
takes place then the function
(\ref{funF}) is a contracting mapping of
the unit ball $M_{\beta\alpha}(1)$ into
itself.
\medskip
\noindent{\bf Remark.} In the proofs of Lemma~\ref{LQdefinition}
and Theorem~\ref{ThSolvability}
we did not use the assumption
about finiteness of the numbers $n_{\alpha}$
of intervals included in continuous
spectra $\sigma_{\alpha}^c$ of the operators
$A_{\alpha}$, ${\alpha}=1,2$. Really,
these assertions take place in the case of arbitrary spectrum
$\sigma_{\alpha} $.
Finiteness at least of one of the numbers $n_1$ and $n_2$ will be used
at the moment.
If $n_1$ and/or $n_2$ are finite and
\be
\label{BQd}
%
\|B_{\alpha\beta} Q_{\beta\alpha}\| < d_0=
{\rm dist}\{\sigma_1,\sigma_2\},
%
\quad {\alpha}=1,2,\, {\beta}\neq{\alpha},
%
\ee
we can state that
\be
\label{resolvest}
%
\|( A_{\alpha} + B_{\alpha\beta} Q_{\beta\alpha} -
\mu )^{-1}\| \leq \frac{C_{\alpha\beta}}{1+|\mu|}, \,\,
%
{\alpha}=1,2, \quad
\mbox{at any $\mu\in\sigma_{\beta}$, ${\beta}\neq{\alpha}$,}
%
\ee
with some $C_{\alpha\beta} > 0$,
$C_{\alpha\beta}\sim 1/(d_0 - \|B_{\alpha\beta} Q_{\beta\alpha}\| )$.
Of course this
estimate is essential only in the case
where $\sigma_{\beta}$ is unbounded.
It follows immediately from Eq.~(\ref{Qbasic1}) that
if $n_1$ and/or $n_2$ are finite then
$Q_{\beta\alpha} f_{\alpha}\in
{\cal D}(H_{\beta})={\cal D}(A_{\beta})$
for any $f_{\alpha}\in {\cal H}_{\alpha}$.
In this case we can rewrite Eq.~(\ref{Qbasic1}) in symmetric
form as a stationary Riccati equation
(see the book~\cite{Reid} and Refs. therein),
\be
\label{QbasicSym}
%
Q_{\beta\alpha} A_{\alpha} -
%
A_{\beta} Q_{\beta\alpha} +
%
Q_{\beta\alpha} B_{\alpha\beta} Q_{\beta\alpha} =B_{\beta\alpha} .
%
\ee
%
To make this, it is sufficient to calculate the
expression
$
%
Q_{\beta\alpha} H_{\alpha} -A_{\beta} Q_{\beta\alpha}
%
$
for both parts of eq.(\ref{Qbasic1}) having in mind that we apply it
to $f_{\alpha}\in {\cal D}(H_{\alpha})$. Did, we have
$$
%
Q_{\beta\alpha} H_{\alpha} -A_{\beta} Q_{\beta\alpha} =
%
Q_{\beta\alpha}(A_{\alpha}+
B_{\alpha\beta} Q_{\beta\alpha})-A_{\beta} Q_{\beta\alpha}=
Q_{\beta\alpha} A_{\alpha} -A_{\beta} Q_{\beta\alpha} +
Q_{\beta\alpha} B_{\alpha\beta} Q_{\beta\alpha}.
%
$$
On the other hand,
$$
%
Q_{\beta\alpha} H_{\alpha} -A_{\beta} Q_{\beta\alpha} =
%
\int\limits_{\sigma_{\beta}}[E_{\beta}(d\mu)
B_{\beta\alpha}(H_{\alpha}-\mu)^{-1}H_{\alpha}-
%
\mu E_{\beta}(d\mu)B_{\beta\alpha}(H_{\alpha}-\mu)^{-1}]=
B_{\beta\alpha}.
%
$$
One finds immediately from Eqs.~(\ref{QbasicSym}),
${\alpha}=1,2$, that
if $Q_{\beta\alpha}$ gives solution
$H_{\alpha}=A_{\alpha}+B_{\alpha\beta} Q_{\beta\alpha}$
of the problem
(\ref{basic}) in the channel ${\alpha}$ then
\be
\label{Qadj}
%
Q_{\alpha\beta}=-Q_{\beta\alpha}^{*}=
-\int\limits_{\sigma_{\alpha} }( H^{*}_{\alpha}
%
-\mu)^{-1}B_{\alpha\beta} E_{\beta}(d\mu )
%
\ee
gives analogous solution
$H_{\beta}=A_{\beta}+B_{\beta\alpha} Q_{\alpha\beta}$
in the channel ${\beta}$.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{theorem}\label{ThInvariant}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Let $Q_{\beta\alpha}$,
$Q_{\beta\alpha}\in{{\bf B}}({\cal H}_{\alpha},{\cal H}_{\beta})$,
be a solution of Eq.~(\ref{QbasicSym}) satisfying together with
$Q_{\alpha\beta}=-Q_{\beta\alpha}^{*}$ the conditions
(\ref{resolvest}). Then the transform
${{\bf H}}'={\cal Q}^{-1}{{\bf H}}{\cal Q}$ with
$
%
{\cal Q}=\left[\begin{array}{lr} I_1 & Q_{12} \\
Q_{21} & I_2
\end{array}\right]
%
$
reduces the operator ${{\bf H}}$ to the block--diagonal form,
${{\bf H}}'=\mathop{\rm diag}\{ H_1, H_2 \}$
where $H_{\alpha}=A_{\alpha} + B_{\alpha\beta} Q_{\beta\alpha},$
${\alpha},{\beta}=1,2,$ ${\beta}\neq{\alpha}.$
At the same time, the operators
$
%
{\cal O}_{\alpha}=\left[\begin{array}{lr} I_{\alpha} & 0 \\
Q_{\beta\alpha} & I_{\beta}
\end{array}\right]
%
$
reduce the Hamiltonian ${{\bf H}}$,
$
%
{{\bf H}}=\left[\begin{array}{lr}
A_{\alpha} & B_{\alpha\beta} \\
B_{\beta\alpha} & A_{\beta}
\end{array}\right],
%
$
to triangular form,
$
%
{{\bf H}}^{(\alpha)}\equiv
{\cal O}_{\alpha}^{-1} {{\bf H}}{\cal O}_{\alpha}=
%
\left[\begin{array}{lr} H_{\alpha} & B_{\alpha\beta} \\
0 & H_{\beta}^{*}
\end{array}\right].
%
$
\end{theorem}
\noindent{\sl Proofs} of both statements are done by direct substituting
${\cal Q}$ and ${\cal O}_{\alpha}$ into the definitions
of ${{\bf H}}'$ and ${{\bf H}}^{(\alpha)}$ and
using the equations~(\ref{QbasicSym}) .
We have to note only that operator ${\cal Q}$
is invertible since, according to
~(\ref{Qadj}),
\be
\label{X}
%
X_{\alpha}=I_{\alpha}-Q_{\alpha\beta} Q_{\beta\alpha} =
I_{\alpha} +Q_{\alpha\beta} Q_{\alpha\beta}^{*} \geq I_{\alpha},
\quad {\alpha}=1,2,
%
\ee
and
\be
\label{Qrevers}
%
{\cal Q}^{-1}=\left[ \begin{array}{lr} X_1^{-1} & 0 \\
0 & X_2^{-1}
\end{array}\right] \cdot
%
\left[ \begin{array}{lr} I_1 & -Q_{12} \\
-Q_{21} & I_2
\end{array}\right] .
%
\ee
\begin{corollary}\label{CReducing}
Subspaces
${\cal H}^{(\alpha)} = {\cal O} _{\alpha}
({\cal H}_{\alpha}\oplus\{ {\bf 0} \} ) =$
%
$
\{ f:\,\, f=\{ f_{\alpha}, f_{\beta} \}\in{\cal H},\,\,
%
f_{\alpha}\in{\cal H}_{\alpha},\,\,
f_{\beta}=Q_{\beta\alpha} f_{\alpha} \}
$
are orthogonal, $ {\cal H}^{(1)} \perp {\cal H}^{(2)}, $
and reducing for ${{\bf H}}$,
${{\bf H}}\left( {\cal D}({{\bf H}})\bigcap
{\cal H}^{(\alpha)} \right)\subseteq {\cal H}^{(\alpha)}. $
\end{corollary}
Really, if $f\in{\cal H}^{(\alpha)}$, $g\in{\cal H}^{(\beta)}$ and
$f=\{ f_{\alpha}, Q_{\beta\alpha} f_{\alpha} \},$
$g=\{ Q_{\alpha\beta} g_{\beta}, g_{\beta} \},$ then
$\langle f,g \rangle = $
$\langle f_{\alpha}, Q_{\alpha\beta} g_{\beta} \rangle +
\langle Q_{\beta\alpha} f_{\alpha}, g_{\beta} \rangle = 0 $
since $Q_{\beta\alpha} = -Q_{\alpha\beta}^{*}$.
The invariance of ${\cal H}^{(\alpha)}$, ${\alpha}=1,2,$
in respect with ${{\bf H}}$ follows
from the equality $ {{\bf H}}{\cal Q} ={\cal Q} {{\bf H}}'.$
Assertions quite analogous to the Theorem~\ref{ThInvariant}
and Corollary~\ref{CReducing} one
can find in Refs.~\onlinecite{MalyshevMinlos1},
\onlinecite{MalyshevMinlos2}.
Solvability (for sufficiently small $\|B_{\alpha\beta}\|$)
of the equation~(\ref{QbasicSym}) was proved
in~\onlinecite{MalyshevMinlos1},
\onlinecite{MalyshevMinlos2} by rather different method
also in the supposition~(\ref{dist}).
\medskip
\noindent{\bf Remark}. It follows from
Theorem~\ref{ThInvariant} that operator
$ \tilde{{\cal Q}}={\cal Q} X^{-1/2} $
with $X=\mathop{\rm diag}\{X_1,X_2\}$
is unitary. Consequently, the operator
$ {{\bf H}}''= \tilde{{\cal Q}}^{*} {{\bf H}}
\tilde{{\cal Q}}=$ $X^{1/2} {{\bf H}}' X^{-1/2}$
becomes self-adjoint in ${\cal H} $.
Since $ {{\bf H}}''=\mathop{\rm diag}\{ H''_1, H''_1 \}$
with $ H''_{\alpha}=X_{\alpha}^{1/2} H_{\alpha} X_{\alpha}^{-1/2}$,
the operators $H''_{\alpha}$, $\,\,\,{\alpha}=1,2,$
are self-adjoint on ${\cal D}(A_{\alpha})$
in ${\cal H}_{\alpha} $. Moreover the operators
$
%
{{\bf H}}^{(\alpha)}=\tilde{{\cal Q}}\cdot
\mathop{\rm diag}\{ H''_{\alpha}, 0\}\cdot \tilde{{\cal Q}}^{*} =
%
$
%
${\cal Q} \cdot\mathop{\rm diag}\{ H_{\alpha}, 0 \}\cdot
{\cal Q}^{-1} $ represent
parts of the Hamiltonian ${{\bf H}}$ in the
corresponding invariant subspaces
${\cal H}^{(1)}$ and ${\cal H}^{(2)}$
(see also Refs.~\onlinecite{MalyshevMinlos1},
\onlinecite{MalyshevMinlos2}).
Unfortunately, eigenvectors ${{\psi}''_{\alpha}}$
of the operators $H''_{\alpha}$ differ from those
for the initial spectral problem~(\ref{ini}):
${{\psi}''}_{\alpha}=X_{\alpha}^{1/2}\psi_{\alpha}$.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{lemma}\label{LHolder}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Let the kernel $B_{\beta\alpha}(\mu,\lambda)$,
${\beta}\neq{\alpha}$,
of the operator $B_{\beta\alpha}$ belong
to the class ${\cal B}^{{\beta}{\alpha}}_{\theta\gamma}$
with $\theta > \frac{1}{2}$
and $Q_{\beta\alpha}$ be a solution of
Eq.~(\ref{Qbasic1}) satisfying together
with $Q_{\alpha\beta}=-Q_{\beta\alpha}^{*}$ the conditions
(\ref{resolvest}). Then
(a) the operator $Q_{\beta\alpha}$ is an integral operator,
$Q_{\beta\alpha}:$
${\cal H}_{\alpha}\rightarrow{\cal H}_{\beta},$
with a kernel $Q_{\beta\alpha}(\mu,\lambda)$
belonging to ${\cal B}^{{\beta}{\alpha}}_{\theta\gamma}$;
(b) the potential
$W_{\alpha}\equiv B_{\alpha\beta} Q_{\beta\alpha}$ is
an integral operator, $W_{\alpha}:$
${\cal H}_{\alpha}\rightarrow{\cal H}_{\alpha},$
with a kernel $W_{\alpha}(\lambda,\lambda')$
belonging to ${\cal B}^{{\alpha}{\alpha}}_{\theta\gamma}$.
%
\end{lemma}
\noindent{\sl Proof.} At the beginning we prove the assertion (b).
According to~(\ref{Qadj}),
\be
\label{w}
%
W_{\alpha}=-B_{\alpha\beta}\int\limits_{\sigma_{\alpha}}
(H_{\beta}^{*}-\lambda)^{-1}B_{\beta\alpha} E_{\alpha}(d\lambda)
%
\ee
with $H_{\beta\alpha}^{*}=A_{\beta}+W_{\beta}^{*}=
A_{\beta}+Q_{\alpha\beta}^{*}B_{\alpha\beta}$. Since the inequalities
(\ref{resolvest}) take place we write
$$
%
\| (H_{\beta}^{*}-\lambda)^{-1} \| = \| (H_{\beta}-\lambda)^{-1} \|
%
\leq C_{\beta\alpha}
%
$$
for any $\lambda\in\sigma_{\alpha}$. In the diagonal representation
(\ref{Neumann}),(\ref{multi}), the equation~(\ref{w}) turns in
$$
%
W_{\alpha}(\lambda,\lambda')=
-B_{\alpha\beta}(\lambda,\,\cdot\,)(H_{\beta}^{*}-\lambda)^{-1}
%
B_{\beta\alpha}(\,\cdot\, ,\lambda').
%
$$
It means that
$$
%
|W_{\alpha}(\lambda,\lambda')| \leq
%
\|B_{\alpha\beta}(\lambda,\,\cdot\, )\|_{{\cal H}_{\beta}}\cdot
%
\|(H_{\beta}^{*}-\lambda)^{-1}\|\cdot
%
\| B_{\beta\alpha}(\,\cdot\, ,\lambda')\|_{{\cal H}_{\beta}}\leq
%
$$
%
\be
\label{West}
%
\leq C_{\beta\alpha}
\|B_{\alpha\beta}(\lambda,\,\cdot\, )\|_{{\cal H}_{\beta}}\cdot
%
\| B_{\beta\alpha}(\,\cdot\, ,\lambda')\|_{{\cal H}_{\beta}}.
%
\ee
Here,
$
%
\|B_{\alpha\beta}(\lambda,\,\cdot\, )\|_{{\cal H}_{\beta}}=
%
\left[ \mathop{\mbox{$\sum$\hskip-1em{\LARGE $\int$}\hskip0.1em}}
\limits_{\sigma_{\beta}}
|B_{\alpha\beta}(\lambda,\mu)|^{2}d\mu \right]^{1/2}.
%
$
Since $\theta >\frac{1}{2},$ we have
$
%
\|B_{\alpha\beta}(\lambda,\,\cdot\, )\|_{{\cal H}_{\beta}}
%
$ $\leq$ $ \frac{c(\theta)}{(1+|\lambda|)^\theta}\cdot \|B\|_{{\cal B}}
%
$
with some $c(\theta)$, $\, c(\theta) > 0$, depending only on $\theta$.
Analogously,
$$
%
\|B_{\beta\alpha}(\,\cdot\, ,\lambda')\|_{{\cal H}_{\beta}}=
%
\|\overline{B_{\alpha\beta}}(\lambda',\,\cdot\,)\|_{{\cal H}_{\beta}}
%
\leq \frac{c(\theta)}{(1+|\lambda|)^\theta}\cdot \|B\|_{{\cal B}},
%
$$
where the operator
$\overline{B_{\alpha\beta}}(\lambda,\mu),$
$\overline{B_{\alpha\beta}}(\lambda,\mu):$
${\cal G}_{\beta}(\mu)\rightarrow{\cal G}_{\alpha}(\lambda)$,
is adjoint to $B_{\beta\alpha}(\mu,\lambda)$.
Estimations similar to~(\ref{West}) may be done also for
$|W_{\alpha}(\lambda'',\lambda')-W_{\alpha}(\lambda,\lambda')|,$
%
$\, \lambda,\lambda''\in\sigma_{\alpha}^c,$
%
$\, \lambda\in\sigma_{\alpha}$,
%
$|W_{\alpha}(\lambda,\lambda''')-W_{\alpha}(\lambda,\lambda')|,$
%
$\, \lambda\in\sigma_{\alpha}$,
%
$\, \lambda''',\lambda'\in\sigma_{\alpha}^c,$
and
%
$ |W_{\alpha}(\lambda,\lambda')-W_{\alpha}(\lambda'',\lambda') -$
%
$W_{\alpha}(\lambda,\lambda''')+W_{\alpha}(\lambda'',\lambda''')|,\,$
%
$\, \lambda,\lambda',\lambda'',\lambda'''\in\sigma_{\alpha}^c$,
%
in terms of the norms
$
%
\|B_{\alpha\beta}(\lambda,\,\cdot\,) -
B_{\alpha\beta}(\lambda'',\,\cdot\,)\|_{{\cal H}_{\beta}}
%
$
and
$
%
\|B_{\beta\alpha}(\,\cdot\, ,\lambda''') -
B_{\beta\alpha}(\,\cdot\, ,\lambda')\|_{{\cal H}_{\beta}}.
%
$
Estimating the latter through $ \| B_{\alpha\beta}\|_{\cal B} $
we come to the inequality
$$
%
\| W_{\alpha}\|_{{\cal B}^{{\alpha}{\alpha}}_{\theta\gamma}}
%
\leq c(\theta)C_{\beta\alpha}\cdot
%
\| B_{\beta\alpha}\|^2_{{\cal B}^{{\beta}{\alpha}}_{\theta\gamma}}
$$
with $0< c(\theta) < \infty. $
Therefore, we have proved the assertion (b).
To prove the statement (a) we note that according to~(\ref{Qbasic1}),
$$
%
Q_{\beta\alpha}=
\int\limits_{\sigma_{\beta}} E_{\beta}(d\mu) B_{\beta\alpha}
\left[
%
(A_{\alpha} -\mu)^{-1}-(H_{\alpha}-\mu)^{-1}
W_{\alpha}(A_{\alpha}-\mu)^{-1} \right]
%
$$
or, in the diagonal representation~(\ref{Neumann}),(\ref{multi}),
$$
%
Q_{\beta\alpha}(\mu,\lambda)=
\frac{B_{\beta\alpha}(\mu,\lambda)}{\lambda-\mu}
%
- \frac{B_{\beta\alpha}(\mu,\,\cdot\,)
(H_{\alpha}-\mu)^{-1}W_{\alpha}(\,\cdot\, ,\lambda)}
%
{\lambda-\mu}.
%
$$
Repeating literally the last part of the proof of the assertion (b)
we come to the inequality
$$
%
\| Q_{\beta\alpha}\|_{{\cal B}^{{\beta}{\alpha}}_{\theta\gamma}} \leq
%
\mathop{\rm sup}\limits_{
%
\mbox{
\scriptsize
$\begin{array}{c}
%
\mu\in\sigma_{\beta}\\
\lambda\in\sigma_{\alpha}
%
\end{array}$
}
}
%
\frac{1}{ |\lambda-\mu| } \cdot
%
\left\{
%
\| B_{\beta\alpha}
\|_{ {\cal B}^{{\beta}{\alpha}}_{\theta\gamma} } +
%
\mbox{\phantom{
$\mathop{\rm sup}\limits_{\mu\in\sigma_{\beta}}$ }}\right.
%
$$
%
$$
%
\left. \mbox{\phantom{{\Large I}}}
%
+ c(\theta)\cdot
\| B_{\beta\alpha}\|_{{\cal B}^{{\beta}{\alpha}}_{\theta\gamma}}
\cdot
%
\mathop{\rm sup}\limits_{\mu\in\sigma_{\beta}}
\| (H_{\alpha}-\mu)^{-1} \|\cdot
%
\| W_{\alpha}\|_{{\cal B}^{{\alpha}{\alpha}}_{\theta\gamma}}
%
\right\},
%
\quad 0 < c(\theta) < +\infty.
$$
Consequently
$Q_{\beta\alpha}\in{\cal B}^{{\beta}{\alpha}}_{\theta\gamma} $
and
$$
%
\|Q_{\beta\alpha}\|_{{\cal B}} \leq \frac{1}{d_0}\cdot
%
\left\{
%
\|B_{\beta\alpha}\|_{{\cal B}} +
%
c(\theta)C_{\alpha\beta} C_{\beta\alpha}
%
\cdot \|B_{\beta\alpha}\|_{{\cal B}}^3
\right\},
%
\quad 0 <c(\theta) < +\infty.
%
$$
This completes the proof of Lemma~\ref{LHolder}.
\noindent\begin{corollary}\label{CHolderSmoothness}
{\it If $B_{\beta\alpha}\in{\cal B}^{{\beta}{\alpha}}_{\theta\gamma}$,
$\,\theta > \frac{1}{2},$
then the solution of Eq.~(\ref{Qbasic1}) described by
Theorem~\ref{ThSolvability}
belongs to the class
${\cal B}^{{\beta}{\alpha}}_{\theta\gamma}$, too. }
\end{corollary}
This statement is based on the fact that the mentioned solution satisfies
automatically the conditions~(\ref{BQd}) and, hence, the conditions
(\ref{resolvest}).
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{EIGENFUNCTIONS AND THE EXPANSION THEOREM}
\label{Eigenfunctions}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
In the preceding section, we have proved the existence
(in the unit ball
$M_{\beta\alpha}(1)\subset
{\bf B}({\cal H}_{\alpha},{\cal H}_{\beta})$)
of a solution
$Q_{\beta\alpha}$ of the basic
equation~(\ref{Qbasic1}) only in the case where spectra
$\sigma_1,$ $\sigma_2$ of the operators $A_1$, $A_2$ are separated,
${\rm dist}\{ \sigma_1,\sigma_2\}$ = $d_0$ $> 0$, and
$\|B_{12}\|_2 = \| B_{21} \|_2 < \frac{d_0}{2}$. May be, however,
Eqs.~(\ref{basic}) and~(\ref{Qbasic1}) have solutions also in other
cases. That is why we study the spectral properties of the operator
$H_{\alpha}=A_{\alpha}+B_{\alpha\beta} Q_{\beta\alpha}$
not supposing that
$\|B_{\alpha\beta}\|_2 < \frac{d_0}{2}$
and using more general requirements~(\ref{resolvest}) only, with
$C_{\alpha\beta}$, some positive numbers,
${\alpha},{\beta}=1,2$, ${\beta}\neq{\alpha}$. Of course, we
assume again that the condition~(\ref{dist}) takes place.
Remember that the requirements~(\ref{resolvest}) are sufficient for
existence of the operators
$V_{\alpha}(H_{\alpha})$. As well, the equations
(\ref{QbasicSym}) and~(\ref{Qadj}) take place and the assertions of
Theorem~\ref{ThInvariant} and Lemma~\ref{LHolder} are valid.
So, let us suppose that
$Q_{\beta\alpha}$ and $Q_{\alpha\beta}=-Q_{\beta\alpha}^{*}$
are solutions
of Eqs.~(\ref{Qbasic1}) and~(\ref{QbasicSym}) satisfying the
conditions~(\ref{resolvest}).
It follows from Lemma~\ref{LQdefinition} that
$Q_{\beta\alpha}\in{\bf B}_{\beta\alpha}
({\cal H}_{\alpha},{\cal H}_{\beta})$
as well as
$Q_{\alpha\beta}\in{\bf B}_{\alpha\beta}
({\cal H}_{\beta},{\cal H}_{\alpha})$.
If $B_{\beta\alpha}\in{\cal B}_{\theta\gamma}^{{\beta}{\alpha}}$,
$\theta>\frac{1}{2}$, then,
according to Lemma~\ref{LHolder},
$Q_{\beta\alpha}\in{\cal B}_{\theta\gamma}^{{\beta}{\alpha}}$
and $Q_{\alpha\beta}\in{\cal B}_{\theta\gamma}^{{\alpha}{\beta}}$.
By Theorem~\ref{ThInvariant}, the operator
${{\bf H}}'=\mathop{\rm diag}\{ H_1, H_2 \}$ is connected with
the (self-adjoint) operator ${{\bf H}}$
by a similarity transform. Thus, the spectra
$\sigma(H_1)$ and $\sigma(H_2)$ of
the operators $H_{\alpha}$, ${\alpha}=1,2$,
are real and
$\sigma(H_1)\bigcup\sigma(H_2)=\sigma({{\bf H}})$.
Continuous spectrum
$\sigma_c (H_{\alpha})$
of the each operator $H_{\alpha}$ coincides with that of the
operator $A_{\alpha}$, $\sigma_c (H_{\alpha})=\sigma_{\alpha}^c$,
since due to $\|B_{\alpha\beta}\|_2 < +\infty$,
the potential $W_{\alpha}=B_{\alpha\beta} Q_{\beta\alpha} $ is
a compact operator. Since $\sigma_1^c \bigcap\sigma_2^c =\emptyset $
we have $\sigma_c (H_1)\bigcap\sigma_c (H_2) =\emptyset$.
We show now that the
discrete spectra $\sigma_d (H_{\alpha}),$ ${\alpha}=1,2$,
satisfy a similar condition.
Let us suppose that $\sigma _{d}(H_{\alpha} )\neq\emptyset $,
$z\in \sigma _{d}(H_{\alpha} )$ and $\psi _{\alpha} $ is
the corresponding eigenfunction of $H_{\alpha} ,$
$ H_{\alpha} \psi _{\alpha} =z\psi _{\alpha} ,$
$\psi _{\alpha} \in {\cal D}(H_{\alpha} )={\cal D}(A_{\alpha} )$.
Then, according to construction of $H_{\alpha}$,
we have
%
$H_{\alpha} \psi _{\alpha} =$
%
$(A_{\alpha} +V_{\alpha} (H_{\alpha} ))\psi _{\alpha} =$
%
$(A_{\alpha} +V_{\alpha} (z))\psi _{\alpha} =z\psi _{\alpha} $.
%
Thus if $z\in \sigma _{d}(H_{\alpha} )$ then $z$ becomes
automatically a point of the discrete
spectrum of the initial spectral problem ~(\ref{ini}).
At the same time $\psi_{\alpha} $ becomes it's eigenfunction.
Let us further denote the eigenfunctions of the
operator $H_{\alpha} $
discrete spectrum by $\psi ^{(j)} _{\alpha} , $
$\psi ^{(j)} _{\alpha} =u^{(j)} _{\alpha} $,
keeping for them the same numeration as for
eigenvectors of $U_j$, $U_j=\{ u_{\alpha}^{(j)}, u_{\beta}^{(j)} \}, $
of the Hamiltonian ${{\bf H}}$,
${{\bf H}}U_j =z_j U_j$, $z_j\in\sigma_d ({{\bf H}}).$
We assume that in the case
of multiple discrete eigenvalues, certain $z_j$ may be repeated
in this numeration.
By ${\cal U} ^{d}$
we denote the set ${\cal U} ^{d}=\{U_{j}, j=1,2,\ldots\}$
of all the eigenvectors $U_j$.
Let ${\cal U} ^{d}_{\alpha} $
be such a subset of ${\cal U} ^{d}$
that it's elements have
the operator $H_{\alpha}$ eigenvectors $\psi^{(j)}_{\alpha}$
in the capacity of the channel ${\alpha}$
components:
%
${\cal U} _{\alpha}^d =$ $\{U_{j}:$
%
$ U_{j}=\{u^{(j)} _{1},u^{(j)} _{2}\},$
%
$ u^{(j)} _{\alpha} =\psi ^{(j)} _{\alpha} \}$.
%
By Theorem~\ref{ThInvariant}, we have
${\cal U} ^{d}_1 \bigcup {\cal U} ^{d}_2 = {\cal U} ^{d}.$
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{theorem}\label{ThDiscSpectrum}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Let $H_{\beta} =A_{\beta} + B_{\beta\alpha} Q_{\alpha\beta} ,$
correspond (for
$\| B_{\beta\alpha} \|_{2} < +\infty $)
to the same solution $Q_{\alpha\beta} =- Q_{\beta\alpha}^{*} $
of Eqs.~(\ref{Qbasic1}) and~(\ref{QbasicSym}) as
$H_{\alpha} =A_{\alpha} +B_{\alpha\beta} Q_{\beta\alpha} $,
and the conditions~(\ref{resolvest}) are valid.
Let $z_{j}\in \sigma _{d}(H_{\alpha} )$ and
$H_{\alpha} u^{(j)} _{\alpha} =
z_{j}u^{(j)} _{\alpha} $ with $u_{\alpha}^{(j)} $,
the channel ${\alpha}$ component of the eigenvector
$U_j =\{ u_{\alpha}^{(j)},u_{\beta}^{(j)} \}$
of the operator ${{\bf H}}$, ${{\bf H}}U_j =z_j U_j$.
Then either
$z_{j}\not\in \sigma _{d}(H_{\beta})$, ${\beta}\neq{\alpha}$,
or (if $z_{j}\in\sigma_{d}(H_{\beta} ))$
the vector $u_{\beta}^{(j)} $ is not
an eigenvector of $H_{\beta}$.
\end{theorem}
\begin{corollary}\label{CDiscSpectrum}
$
%
{\cal U} ^{d}_1
%
\bigcap{\cal U} ^{d}_2 =\emptyset
%
$.
\end{corollary}
Statement of Theorem~\ref{ThDiscSpectrum} means that discrete
spectrum $\sigma_d ({{\bf H}}) $ is distributed between
discrete spectra $\sigma_d (H_1)$ and
$\sigma_d (H_2)$ in such a way that
operators $H_1$ and $H_2$ have not ``common'' eigenvectors
$U_j =\{ u_1^{(j)},u_2^{(j)} \}$:
simultaneously, component $u_1^{(j)}$
can not be eigenvector
for $H_1$, and $u_2^{(j)}$ with the same $j$, for $H_2$.
\medskip
\noindent{\sl Proof} of the Theorem will be given by contradiction.
Let us suppose that $\psi_{\alpha}^{(j)} = u_{\alpha}^{(j)}$
is an eigenvector of $H_{\alpha}$
corresponding to $z_j$ i.e.
\be
\label{Aeigen}
%
(A_{\alpha}+B_{\alpha\beta} Q_{\beta\alpha} -z_j)
\psi_{\alpha}^{(j)}=0.
%
\ee
If $z_j\in\sigma_{\alpha}=\sigma(A_{\alpha})$ then automatically
$z_j\not\in\sigma_d(H_{\beta})$
since due to conditions~(\ref{resolvest})
we have $\sigma(H_{\beta})\bigcap\sigma(A_{\alpha})=\emptyset$.
Thus in the case
where $z_j\in\sigma_{\alpha}$ the assertion of Theorem is valid.
Let $z_j\not\in\sigma(A_{\alpha})$. In this case we can rewrite
Eq.~(\ref{Aeigen}) in the form
\be
\label{LSch}
%
\psi_{\alpha}^{(j)} =
-(A_{\alpha} -z_j)^{-1}B_{\alpha\beta}
Q_{\beta\alpha}\psi_{\alpha}^{(j)}.
%
\ee
Let $y_{\beta}^{(j)}=Q_{\beta\alpha}\psi_{\alpha}^{(j)}$.
It follows from~(\ref{LSch}) that
\be
\label{y}
%
y_{\beta}^{(j)} +
Q_{\beta\alpha}(A_{\alpha} - z_j)^{-1} y_{\beta}^{(j)} =0.
%
\ee
We will show that the vector $ y_{\beta}^{(j)} $
is a solution of the initial
spectral problem~(\ref{ini}) in the channel ${\beta}$ at $z=z_j$ and
$\tilde{U}_j=\{ \psi_{\alpha}^{(j)}, y_{\beta}^{(j)} \}$
is an eigenvector of ${{\bf H}}$,
${\bf H}\tilde{U}_j=z_j \tilde{U}_j$.
To do this, we act on both parts of
Eq.~(\ref{y}) by $ H^{*}_{\beta}-z_j$
remembering that, according to
(\ref{Qadj}),
$Q_{\beta\alpha}=-Q_{\alpha\beta}^{*}$
%
$=-\int\limits_{\sigma_{\alpha}}(H_{\beta}^{*}-\lambda)^{-1}
B_{\beta\alpha} E_{\alpha}(d\lambda)$.
%
We obtain
$$
%
(H_{\beta}^{*}-z_j)y_{\beta}^{(j)}+
\int\limits_{\sigma_{\alpha}} (H_{\beta}^{*}-z_j)
%
(H_{\beta}^{*}-\lambda)^{-1}(z_j -\lambda)^{-1}
B_{\beta\alpha} E_{\alpha}(d\lambda)
%
B_{\alpha\beta} y_{\beta}^{(j)} =0.
%
$$
Using the identity
%
$ (H-z)(H-\lambda)^{-1}(z-\lambda)^{-1}=
%
(z-\lambda)^{-1}-(H-\lambda)^{-1} $
%
we find
$$
%
(H_{\beta}^{*}-z_j)y_{\beta}^{(j)}+\int\limits_{\sigma_{\alpha}}
%
[(z_j -\lambda)^{-1} - (H_{\beta}^{*}-z_j) ]
%
B_{\beta\alpha} E_{\alpha}(d\lambda)
B_{\alpha\beta} y_{\beta}^{(j)} =0
%
$$
or, and it is the same,
\be
\label{yy}
%
(H_{\beta}^{*}-z_j)y_{\beta}^{(j)} -
B_{\alpha\beta}(A_{\alpha} -z_j)^{-1} B_{\alpha\beta} y_{\beta}^{(j)}
%
+ Q_{\beta\alpha} B_{\alpha\beta} y_{\beta}^{(j)} =0.
%
\ee
However
$ H_{\beta}^{*}=A_{\beta} -
Q_{\beta\alpha} B_{\alpha\beta} $.
Hence the relation~(\ref{yy})
turns in equation~(\ref{ini}) for the channel ${\beta}$,
$$
%
[A_{\beta} - B_{\beta\alpha}
(A_{\alpha} -z_j)^{-1} B_{\alpha\beta} - z_j ] y_{\beta}^{(j)} = 0.
%
$$
So, we have proved that $y_{\beta}^{(j)}$
is a solution of the initial problem
in the channel ${\beta}$ and we did deal with an eigenvector
$ U_j=\{ u_{\alpha}^{(j)}, u_{\beta}^{(j)} \} $
of the operator ${{\bf H}}$ having the components
$u_{\alpha}^{(j)}=\psi_{\alpha}^{(j)}$ and
$u_{\beta}^{(j)} =y_{\alpha}^{(j)} $.
Let us show that $y_{\beta}^{(j)}$ can not be
an eigenvector of $H_{\beta}$
corresponding to the eigenvalue $z_j$.
Actually, due to~(\ref{y}) we have
$$
%
{\rm a}\equiv \langle y_{\beta}^{(j)} +
Q_{\beta\alpha} (A_{\alpha} -z_j)^{-1}
B_{\alpha\beta} y_{\beta}^{(j)},
%
y_{\beta}^{(j)} \rangle =0.
%
$$
On the other hand
$$
%
{\rm a}= \| y_{\beta}^{(j)}\|^2 +
\langle (A_{\alpha}-z_j)^{-1} B_{\alpha\beta} y_{\beta}^{(j)},
%
Q_{\beta\alpha}^{*}y_{\beta}^{(j)} \rangle.
%
$$
If $y_{\beta}^{(j)}$ is an eigenvector of
$H_{\beta}$, $H_{\beta} y_{\beta}^{(j)} =z_j y_{\beta}^{(j)}$,
then
$$
%
Q_{\beta\alpha}^{*} y_{\beta}^{(j)} =
-Q_{\alpha\beta} y_{\beta}^{(j)} =
-\int\limits_{\sigma_{\alpha}}
%
E_{\alpha}(d\lambda) B_{\alpha\beta}
(H_{\beta} -\lambda)^{-1} y_{\beta}^{(j)} =
%
(A_{\alpha} -z_j)^{-1} B_{\alpha\beta} y_{\beta}^{(j)}.
%
$$
It means that
$$
%
{\rm a} = \|y_{\beta}^{(j)}\|^2 +
\| (A_{\alpha}-z_j)^{-1} B_{\alpha\beta} y_{\beta}^{(j)}\|^2
%
\geq \|y_{\beta}^{(j)}\|^2 .
%
$$
Since ${\rm a} =0 $ we get $y_{\beta}^{(j)} =0 $
and, due to~(\ref{LSch}), $\psi_{\alpha}^{(j)} =0.$
However, by supposition,
$ \psi_{\alpha}^{(j)} \neq 0.$
Thus, we come to a contradiction and
$y_{\beta}^{(j)}$ can not be
an eigenvector of $H_{\beta}$.
And so, if $z_j \in \sigma_d (H_{\alpha})$ and
$H_{\alpha} u_{\alpha}^{(j)} =z_j u_{\alpha}^{(j)}$
then $u_{\beta}^{(j)} $ is not an eigenvector of $H_{\beta}$.
The proof of Theorem~\ref{ThDiscSpectrum} is completed.
\medskip
Let us pay attention to the continuous spectrum of $H_{\alpha}$
assuming here that
$B_{\alpha\beta}\in{\cal B}^{{\beta}{\alpha}}_{\theta\gamma},$
$\theta > \frac{1}{2},$ $\gamma > \frac{1}{2}$, and consequently,
$Q_{\alpha\beta}\in{\cal B}^{{\beta}{\alpha}}_{\theta\gamma},$
${\alpha},{\beta}=1,2$, ${\beta}\neq{\alpha}.$
Consider at $\lambda'\in\sigma_{\alpha}^c$
the integral equations
\be
\label{LSchpsi}
%
\psi ^{(\pm )}_{\alpha} (\lambda ,\lambda ')=
%
I_{\alpha}^c {\delta} (\lambda -\lambda ')-
%
[(A_{\alpha} -\lambda '\mp i0)^{-1}
%
W_{\alpha}
%
\psi ^{(\pm )}_{\alpha}](\lambda ,\lambda '), \,\, {\alpha}=1,2,
%
\ee
where as usually
$W_{\alpha}=B_{\alpha\beta} Q_{\beta\alpha}.$
Since
$W_{\alpha}\in{\cal B}^{{\alpha}{\alpha}}_{\theta\gamma},$
the integral operator with the kernel
$
%
\frac{W_{\alpha}(\lambda,\lambda')}{\lambda -\lambda '\mp i0}
%
$
is compact in
$
{\cal M}_{\theta '\gamma '},
$
%
$
\frac{1}{2}<\theta '<\theta ,
$
%
$
0<\gamma '<\gamma
$
%
(cf. Refs.~\onlinecite{LadyzhFaddeev}, \onlinecite{Faddeev64}).
If $\lambda '\not\in \sigma _{d}(H_{\alpha} )$
then Eq.~(\ref{LSchpsi}) for
$\psi^{(+)}_{{\alpha}}$
as well as for $\psi^{(-)}_{{\alpha}}$ is
uniquely solvable (see Ref.~\onlinecite{Faddeev64}) in
the class of the form ~(\ref{amplitude})
distributions.
Denote by
$
\Psi ^{(\pm )}_{\alpha} ,
$
$\Psi ^{(\pm )}_{\alpha} :$
$
{{\cal H}} ^{c}_{\alpha} \rightarrow {{\cal H}} _{\alpha}
$,
the in\-te\-gral ope\-ra\-tor
with the ker\-nel
$
\psi ^{(\pm )}_{\alpha} (\lambda ,\lambda ')
$.
The operator $\Psi ^{(\pm )}_{\alpha} $ is
bounded and
$
\Psi ^{(\pm )}_{\alpha} {\cal D}(A^{(0)}_{\alpha} )
%
\subseteq {\cal D}(H_{\alpha} )$%
~\cite{LadyzhFaddeev}, \cite{Faddeev64}.
It follows from ~(\ref{LSchpsi})
that $\Psi ^{(\pm )}_{\alpha} $
has the property
$
H_{\alpha} \Psi ^{(\pm )}_{\alpha} =
%
\Psi ^{(\pm )}_{\alpha} A^{(0)}_{\alpha}
$.
Thus,
$
Q_{\beta\alpha} \Psi ^{(\pm )}_{\alpha} (\cdot ,\lambda ')=
$
%
$
%
(\lambda '- A_{\beta} )^{-1}
%
$
$
%
B_{\beta\alpha} \Psi ^{(\pm )}_{\alpha} (\cdot ,\lambda ')
$.
Substitution of this
expression in~(\ref{LSchpsi}) shows that
$
\psi ^{(\pm )}_{\alpha}
$
satisfies~(\ref{wavefunction}).
Due to the uniqueness
of Eq.~(\ref{wavefunction}) solution at
$\lambda'\not\in\sigma_d({{\bf H}})$
we have
$
%
\psi ^{(\pm )}_{\alpha} (\lambda ,\lambda ')=
%
u^{(\pm )}_{{\alpha}{\alpha}}(\lambda,\lambda')
$.
This means that each
eigenfunction
$
u^{(\pm )}_{{\alpha} {\alpha} }(\lambda ,\lambda '),
$
%
%
$
\lambda ^\prime \in \sigma ^{c}_{\alpha} ,
$
%
%
$
\lambda '\not\in \sigma _{d}({{\bf H}})
$
of the initial spectral problem
problem~(\ref{ini}) is also an eigenfunction of $H_{\alpha} $.
Consider the functions
$
%
\tilde{\psi}^{(j)} _{\alpha} =
%
\psi ^{(j)} _{\alpha} -Q_{\alpha\beta} u^{(j)} _{\beta}
%
$
and \newline
$
\tilde{\psi}^{(\pm )}_{\alpha} (\cdot ,\lambda ')=
$
$
%
\psi ^{(\pm )}_{\alpha} (\,\cdot\, ,\lambda ')-
$
$
%
Q_{\alpha\beta} u^{(\pm )}_{\beta\alpha}
%
(\,\cdot\, ,\lambda '),
$
%
%
$
\lambda '\in \sigma ^{c}_{\alpha}
$.
Let
$
\tilde{\Psi }^{(\pm )}_{\alpha} ,
$
%
%
$
\tilde{\Psi }^{(\pm )}_{\alpha} :
$
%
%
$
{{\cal H}} ^{c}_{\alpha} \rightarrow {{\cal H}} _{\alpha}
$,
be the integral operator with the
kernel
$
\tilde{\psi}^{(\pm )}_{\alpha} (\lambda ,\lambda ')
$.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{theorem}\label{ThExpansion}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
The functions
$
\tilde{\psi}^{(j)} _{\alpha}
$
(with $j$ such that
$
U_{j}\in {\cal U} ^{d}_{\alpha} )
$
are eigenfunctions of adjoint operator
$
H^{*}_{\alpha} , H^{*}_{\alpha} =
%
A_{\alpha} +Q^{*}_{\beta\alpha} B_{\beta\alpha} ,
$
discrete spectrum,
$
H^{*}_{\alpha} \tilde{\psi}^{(j)} _{\alpha} =
$
%
%
$
z_{j}\tilde{\psi}^{(j)} _{\alpha}
$.
Operators
$
\tilde{\Psi }^{(\pm )}_{\alpha}
$
have the property
$
H^{*}_{\alpha} \tilde{\Psi }^{(\pm )}_{\alpha}=
$
%
%
$
\tilde{\Psi }^{(\pm )}_{\alpha} A^{(0)}_{\alpha}
$.
At the same time the orthogonality relations take
place:
$
\langle\psi^{(j)} _{\alpha} ,\tilde{\psi}^{(k)}_{\alpha} \rangle=
%
{\delta} _{jk},
$
%
%
$
\Psi ^{(\pm )*}_{\alpha} \tilde{\Psi }^{(\pm )}_{\alpha} =
\left.{I_{\alpha}}\right|_{{\cal H}_{\alpha}^c},
$
%
%
$
\tilde{\Psi }^{(\pm )*}_{\alpha} \psi ^{(j)} _{\alpha} =0
$
and
$
\Psi_{\alpha}^{(\pm)*}\tilde{\psi}^{(j)} _{\alpha} =0.
$
Also, the following completeness relations are valid,
\be
\label{completeness}
%
\sum\limits_{j:\, U_{j}\in{\cal U} _{\alpha}^d}
%
{\psi}^{(j)} _{\alpha} \langle\cdot,
\tilde{\psi}^{(j)} _{\alpha} \rangle+
%
\Psi^{(\pm )}_{\alpha} \tilde{\Psi }^{(\pm )*}_{\alpha} =
%
I_{\alpha} , \quad {\alpha} =1,2,
%
\ee
\end{theorem}
\noindent{\sl Proof.} Show for example that
\be
\label{Hadjeigen}
%
H_{\alpha}^{*}\tilde{\psi}_{\alpha}^{(j)}=
%
z_j\tilde{\psi}_{\alpha}^{(j)}
%
\ee
(remember that $z_j\in{\bf R}$). We have
$$
%
H_{\alpha}^{*}\tilde{\psi}_{\alpha}^{(j)}=
(A_{\alpha}-Q_{\alpha\beta} B_{\beta\alpha})(\tilde{\psi}_{\alpha}^{(j)}
-Q_{\alpha\beta} u_{\beta}^{(j)})=
%
$$
\be
\label{Heq1}
%
=(A_{\alpha}-Q_{\alpha\beta}
B_{\beta\alpha})\psi_{\alpha}^{(j)} -
(A_{\alpha} Q_{\alpha\beta}-
Q_{\alpha\beta} B_{\beta\alpha} Q_{\alpha\beta})u_{\beta}^{(j)}.
%
\ee
Note that $A_{\alpha}=H_{\alpha}-
B_{\alpha\beta} Q_{\beta\alpha}$ and, hence,
\be
\label{Heq2}
%
(A_{\alpha}-Q_{\alpha\beta}
B_{\beta\alpha})\psi_{\alpha}^{(j)} =
z_j\psi_{\alpha}^{(j)} -
%
(B_{\alpha\beta} Q_{\beta\alpha}+
Q_{\alpha\beta} B_{\beta\alpha})\psi_{\alpha}^{(j)}.
%
\ee
The second term in the right part of~(\ref{Heq2}) may be easily
expressed by $u_{\beta}^{(j)}$. Actually,
%
$
u_{\beta}^{(j)}=-(A_{\beta}-z_j)^{-1}B_{\beta\alpha}
\psi_{\alpha}^{(j)} $
%
[here, we use again the
property $\sigma(H_{\alpha})\bigcap\sigma_{\beta}=\emptyset$
following
from~(\ref{resolvest})]. Since Eqs.~(\ref{Qbasic}) and
$H_{\alpha}\psi_{\alpha}^{(j)}=z_j\psi_{\alpha}^{(j)}$
take place, we find
$Q_{\beta\alpha}\psi_{\alpha}^{(j)}=
B_{\alpha\beta} u_{\beta}^{(j)}.$ Thus,
$$
(B_{\alpha\beta} Q_{\beta\alpha} +
Q_{\alpha\beta} B_{\beta\alpha})\psi_{\alpha}^{(j)}=
%
B_{\beta\alpha} Q_{\beta\alpha}\psi_{\alpha}^{(j)}+
Q_{\alpha\beta}(A_{\beta}-z_j)
(A_{\beta}-z_j)^{-1}B_{\beta\alpha}\psi_{\alpha}^{(j)}=
%
$$
%
$$
=B_{\beta\alpha} u_{\beta}^{(j)} -Q_{\alpha\beta}
(A_{\beta} - z_j)u_{\beta}^{(j)}.
$$
Substituting the expressions obtained into~(\ref{Heq2}) and then
into~(\ref{Heq1}), we get
$$
%
H_{\alpha}^{*}\tilde{\psi}_{\alpha}^{(j)}=
z_j(\psi_{\alpha}^{(j)}-Q_{\alpha\beta} u_{\beta}^{(j)})
%
+ [-B_{\alpha\beta}+Q_{\alpha\beta} A_{\beta}-
A_{\alpha} Q_{\alpha\beta} +
Q_{\alpha\beta} B_{\beta\alpha} Q_{\alpha\beta}]u_{\beta}^{(j)}.
%
$$
According to the equations~(\ref{QbasicSym}), the expression in the square
brackets is equal to zero and we come to~(\ref{Hadjeigen}).
The equalities
$
H_{\alpha}^{*}\tilde{\psi}^{(\pm)}_{\alpha}(\,\cdot\, ,\lambda')=
$
%
$
\lambda'\tilde{\psi}^{(\pm)}_{\alpha}(\,\cdot\, ,\lambda'),
$
%
$\lambda'\in\sigma_{\alpha}^c ,$ are proved quite analogously.
The orthogonality relations
$
\langle\psi^{(j)} _{\alpha} ,\tilde{\psi}^{(k)}_{\alpha} \rangle=
%
{\delta}_{jk},
$
%
%
$
\tilde{\Psi }^{(\pm )*}_{\alpha} \psi ^{(j)} _{\alpha} =0
$
and
$
\Psi_{\alpha}^{(\pm)*}\tilde{\psi}^{(j)} _{\alpha} =0
$
are trivial. Proofs of the relation
%
$
\Psi ^{(\pm )*}_{\alpha} \tilde{\Psi }^{(\pm )}_{\alpha} =
\left.{I_{\alpha}}\right|_{{\cal H}_{\alpha}^c},
$
%
and the equality~(\ref{completeness}) are very similar.
Both these proofs are
based on use of properties of the wave
operators $U^{(\pm)}$. As a sample,
we give a proof of the completeness relation~(\ref{completeness}).
Consider the operator
$$
%
{\cal A}=\sum\limits_{j:\, U_j\in{\cal U}_{\alpha}^d}
%
\psi_{\alpha}^{(j)} \langle\,\cdot\, ,
\tilde{\psi}_{\alpha}^{(j)}\rangle +
%
\Psi_{\alpha}\tilde{\Psi}_{\alpha}^{*} =
%
$$
\be
\label{Aeq0}
%
=\sum\limits_{j:\, U_j\in{\cal U}_{\alpha}^d}
%
\psi_{\alpha}^{(j)} \langle\,\cdot\, ,
\psi_{\alpha}^{(j)} -Q_{\alpha\beta} u_{\beta}^{(j)}\rangle +
%
\Psi_{\alpha}[\Psi_{\alpha}^{*} -(Q_{\alpha\beta} u_{\beta\alpha})^{*}].
%
\ee
For convenience, we omit signs ``$\pm$'' in notations of
${\Psi}^{(\pm)}_{\alpha} \equiv u^{(\pm)}_{\alpha\alpha}$,
$ u^{(\pm)}_{\beta\alpha}$
and
$\tilde{\Psi}^{(\pm)}_{\alpha}$
taking in mind for example the case of sign ``$+$''. We have from
(\ref{Aeq0}):
$$
%
{\cal A}=\sum\limits_{j:\, U_j\in{\cal U}_{\alpha}^d}
%
\psi_{\alpha}^{(j)} \langle\,\cdot\, ,\psi_{\alpha}^{(j)} \rangle +
%
\Psi_{\alpha} \Psi_{\alpha}^{*} -
%
\sum\limits_{j:\, U_j\in{\cal U}_{\alpha}^d}
%
\psi_{\alpha}^{(j)} \langle\,\cdot\, ,
Q_{\alpha\beta} u_{\beta}^{(j)} \rangle
%
-\Psi_{\alpha} u_{\beta\alpha}^{*} Q_{\alpha\beta}^{*}.
%
$$
It follows from the completeness relations $U^{(\pm)*}U^{(\pm)}=I-P$
for wave operators $U^{(\pm)}$ that
\be
\label{Aeq1}
%
\Psi_{\alpha} u_{\beta\alpha}^{*}\equiv
u_{\alpha\alpha} u_{\beta\alpha}^{*}=
%
-u_{\alpha\beta} u_{\beta\beta}^{*} -
\sum\limits_{z_j\in\sigma({{\bf H}})}
%
u_{\alpha}^{(j)} \langle \,\cdot\, ,u_{\beta}^{(j)}\rangle.
%
\ee
Since $ u_{\beta\beta}^{*} Q_{\alpha\beta}^{*} =
(Q_{\alpha\beta} u_{\beta\beta})^{*} = u_{\alpha\beta}^{*},$
we can write
with a help of~(\ref{Aeq1}) that
$$
%
{\cal A}=u_{\alpha\alpha} u_{\alpha\alpha}^{*} +
u_{\alpha\beta} u_{\alpha\beta}^{*} +
%
\sum\limits_{j:\, U_j\in{\cal U}_{\alpha}^d}
%
\psi_{\alpha}^{(j)} \langle \,\cdot\, ,
\psi_{\alpha}^{(j)} \rangle +
%
\sum\limits_{ \mbox{
\scriptsize
$\begin{array}{c}
%
z_j\in\sigma_d (\mbox{{\footnotesize${\bf H}$}})\\
%
U_j \not\in {\cal U}_{\alpha}^d
%
\end{array}$
}
}
%
u_{\alpha}^{(j)} \langle \,\cdot\, ,
Q_{\alpha\beta} u_{\beta}^{(j)} \rangle .
%
$$
In the last sum, the conditions
$ z_j\in\sigma_d ({{\bf H}})$ and
$ U_j \not\in {\cal U}_{\alpha}^d $
mean really that we deal with any $j$ such that
$U_j\in{\cal U}_{\beta}^d$. This follows from the equalities
$ {\cal U}_1^d \bigcup {\cal U}_2^d = {\cal U}^d $ and
$ {\cal U}_1^d \bigcap {\cal U}_2^d = \emptyset $ (see
Theorem~\ref{ThDiscSpectrum} and Corollary~\ref{CDiscSpectrum}).
For $ U_j\in{\cal U}_{\beta}^d$, the vector
$u_{\beta}^{(j)}$ is eigenfunction of $H_{\beta}$,
$u_{\beta}^{(j)}=\psi_{\beta}^{(j)}$, and
$Q_{\alpha\beta} u_{\beta}^{(j)} =$
$Q_{\alpha\beta} \psi_{\beta}^{(j)} = u_{\alpha}^{(j)} .$
Thus, ${\cal A}$ turns in
$$
%
{\cal A}=u_{\alpha\alpha} u_{\alpha\alpha}^{*} +
u_{\alpha\beta} u_{\beta\alpha}^{*}
%
+ \sum\limits_{z_j \in \sigma_d ({{\bf H}})}
%
u_{\alpha}^{(j)} \langle \,\cdot\, ,
u_{\alpha}^{(j)} \rangle
%
=(U^{(\pm)}U^{(\pm)*} + P)_{\alpha\alpha}.
%
$$
Since $U^{(\pm)}U^{(\pm)*}+P=I$ we find
${\cal A}=I_{\alpha}$ and this completes
the proof of Theorem~\ref{ThExpansion}.
\medskip
Theorem~\ref{ThExpansion} means in particular that {\it
part $H^{c}_{\alpha} $ of
operator $H_{\alpha} $ acting in the invariant subspace
corresponding to it's continuous
spectrum $\sigma_{\alpha}^c$, is similar
to the operator
$
%
A^{(0)}_{\alpha},
%
$
%
$
%
H^{c}_{\alpha} =
%
\Psi ^{(\pm )}_{\alpha} A^{(0)}_{\alpha}
%
\tilde{\Psi }^{(\pm )*}_{\alpha}
%
$,
and spectrum $\sigma_{\alpha}^c$ is absolutely continuous.}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{INNER PRODUCT MAKING NEW HAMILTONIANS SELF-ADJOINT}
\label{InnerProduct}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
We introduce now a new inner product $[\, .\, ,\, .\,]_{\alpha}$ in
$
{{\cal H}} _{\alpha} ,
$
%%%%
$
[f_{\alpha},g_{\alpha}]_{\alpha}=
\langle X_{\alpha} f_{\alpha},g_{\alpha} \rangle,
$
%
$f_{\alpha},g_{\alpha}\in{\cal H}_{\alpha},$
with $X_{\alpha}$ defined as in Theorem~\ref{ThInvariant},
$X_{\alpha}=I_{\alpha} +Q_{\alpha\beta} Q_{\alpha\beta}^{*}$,
${\alpha}=1,2.$
The operator $X_{\alpha} $ is positive definite,
$X_{\alpha}\geq I_{\alpha}$.
This means that $[\, .\, ,\, .\,]_{\alpha}$
satisfies all the axioms of
inner product.
\begin{theorem}\label{ThInnerProduct}
%
The operator $H_{\alpha}, $ ${\alpha}=1,2,$
is self-adjoint on ${\cal D}(A_{\alpha})$
in respect with the inner product $[\, .\, ,\, .\,]_{\alpha}$.
%
\end{theorem}
\noindent{\sl Proof.} It follows from Theorem~\ref{ThInvariant}
that operator ${{\bf H}}'$ is
self-adjoint in ${\cal H}={\cal H}_1 \oplus{\cal H}_2 $
in respect with the inner product
$[\, .\, ,\, .\,]$, $[f,g]=[Xf,g]$
with $X=\mathop{\rm diag}\{ X_1, X_2 \}$. Did, since
${\cal Q}^{-1}={\cal Q}^{*}X^{-1}=X^{-1}{\cal Q}^{*}$,
we have for
$f,g\in{\cal D}({{\bf H}}')={\cal D}({{\bf H}})=
{\cal D}(A_1)\oplus{\cal D}(A_2)$:
$$
%
[{{\bf H}}'f,g]=\langle
X{\cal Q}^{-1} {{\bf H}} {\cal Q} f,g \rangle =
%
\langle X\cdot X^{-1} {\cal Q}^{*}{{\bf H}}{\cal Q} f, g \rangle =
$$
%
$$
= \langle f, {\cal Q}^{*} {{\bf H}}{\cal Q} g \rangle =
%
\langle f, X\cdot X^{-1} {\cal Q}^{*} {{\bf H}} {\cal Q} g \rangle =
%
[f,{{\bf H}}'g].
$$
Here, we used the fact that in the case of~(\ref{resolvest}),
${\cal Q} f \in {\cal D}(A_1)\oplus{\cal D}(A_2)$
if
$ f \in {\cal D}(A_1)\oplus{\cal D}(A_2).$
Taking elements $f,$ $g$ in the equality $[{{\bf H}}'f,g]=[f,{{\bf H}}'g]$
in the form $f=\{ f_1, 0 \}$, $g=\{ g_1, 0 \}$
or $f=\{ 0, f_2 \}$, $g=\{ 0, g_2 \}$
with one of the components equal to zero
and $f_{\alpha},g_{\alpha}\in{\cal D}(A_{\alpha})$,
${\alpha}=1,2$, one comes to the statement of Theorem.
\medskip
\noindent{\bf Remark.} This Theorem may be proved also in another way
making use of the equality
\be
\label{QQ}
%
I_{\alpha} + Q_{\alpha\beta} Q_{\alpha\beta}^{*} =
%
\sum\limits_{j:\,\, U_j\in{\cal U} _{\alpha}^d}
%
\tilde{\psi}_{\alpha}^{(j)} \langle\,\cdot\, ,
\tilde{\psi}_{\alpha}^{(j)} \rangle +
%
\tilde{\Psi}^{(\pm)}_{\alpha} \tilde{\Psi}^{(\pm)} {}^{*}_{\alpha}
%
\ee
which is valid for both signs ``$+$'' and ``$-$''.
In this case, a self-adjointness of $H_{\alpha}$ in respect with
$[\,\cdot\, , \,\cdot\, ]_{\alpha}$ follows
from the fact that it's spectrum is real and also from
relations
$
H^{*}_{\alpha} \tilde{\Psi }^{(\pm )}_{\alpha}
%
\tilde{\Psi }^{(\pm )*}_{\alpha} =
$
%
%
$
\tilde{\Psi }^{(\pm )}_{\alpha} A^{(0)}_{\alpha}
%
\tilde{\Psi }^{(\pm )*}_{\alpha} =
$
%
%
$
\tilde{\Psi }^{(\pm )}_{\alpha}
%
\tilde{\Psi }^{(\pm )*}_{\alpha} H_{\alpha}
$.
The equality~(\ref{QQ}) itself is proved by calculating it's right part
in the same way as it was done where the completeness relations
(\ref{completeness}) were established
(see proof of Theorem~\ref{ThExpansion}).
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{SCATTERING PROBLEM}\label{ScatteringProblem}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
We establish now that operators
$
\Psi_{\alpha}^{(+)}
$
and
$
\Psi_{\alpha}^{(-)}
$
play the same important role describing a time asymptotics of solutions
of the Schr\"{o}dinger equation
\be
\label{Schrod}
%
i\frac{d}{dt}f_{\alpha}(t)=H_{\alpha} f_{\alpha}(t)
%
\ee
as in the usual self-adjoint case~\cite{LadyzhFaddeev},
\cite{Faddeev64}.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{theorem}\label{ThWave}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Operator $U_{\alpha}(t)=\exp(iH_{\alpha} t)\exp(-iA_{\alpha}^{(0)} t)$
converges strongly if $t\rightarrow\mp\infty$,
in respect with the norm
$\parallel\cdot\parallel_{\alpha}^X$
corresponding to the inner product
$[\,\cdot\, ,\,\cdot \,]_{\alpha}$
in ${{\cal H}} _{\alpha}$. The limit is equal to
$\,\,\, s\!-\!\mathop{\rm lim}\limits_{t\rightarrow\mp\infty}
U_{\alpha}(t)=\Psi_{\alpha}^{(\pm)}.$
%
\end{theorem}
Since the norms $\| \,\cdot\, \|^{X}_{\alpha} $ and
$\| \,\cdot\,\| $ in ${\cal H}_{\alpha}$
are equivalent,
$\|f\|\leq\|f\|^{X}_{\alpha} \leq
(1+\|Q_{\alpha\beta}\|\cdot\|Q_{\beta\alpha}\|)^{1/2}\|f\|$,
the same statement takes place also in respect
with the initial norm
$\| \,\cdot\,\| $.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{theorem}\label{ThScattering}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
For any element $f_{\alpha}^{(-)}\in{\cal H}_{\alpha}^c$
one can find such unique element $f_{\alpha}^{(0)}$
that solution
$f_{\alpha}(t)=\exp(-iH_{\alpha} t)f_{\alpha}^{(0)}$
of Eq.~(\ref{Schrod})
satisfies the asymptotic condition
$$
%
\mathop{\rm lim}\limits_{t\rightarrow-\infty}
\parallel{f_{\alpha}(t)-\exp(-iA_{\alpha}^{(0)}t)
%
f_{\alpha}^{(-)}}\parallel_{\alpha}^X=0.
%
$$
There exists the unique element
$f_{\alpha}^{(+)}\in{\cal H}_{\alpha}^c$ such that
$$
%
\mathop{\rm lim}
\limits_{t\rightarrow+\infty}\parallel{f_{\alpha}(t)-
\exp(-iA_{\alpha}^{(0)}t)
%
f_{\alpha}^{(+)}}\parallel_{\alpha}^X=0.
%
$$
Elements $f_{\alpha}^{(-)}$ and $f_{\alpha}^{(+)}$ are
connected by the relation
$f_{\alpha}^{(+)}=S^{({\alpha})}f_{\alpha}^{(-)}$
with
\newline
$
%
S^{({\alpha})}=
$
%
$
\Psi_{\alpha}^{(-)-1}\Psi_{\alpha}^{(+)}=
$
%
$
\tilde{\Psi}_{\alpha}^{(-)*}\Psi_{\alpha}^{(+)}=
$
%
$
\Psi_{\alpha}^{(-)*} X_{\alpha} \Psi_{\alpha}^{(+)}.
%
$
\end{theorem}
We do not give here proofs of the Theorems~\ref{ThWave}
and~\ref{ThScattering} because they are exactly
the same as in the case of one-particle Schr\"{o}dinger operator in
Ref.~\onlinecite{Faddeev63}
Theorem~\ref{ThScattering} gives the
non-stationary formulation of the scattering problem
for a system described by Hamiltonian $H_{\alpha}$.
Moreover $S^{({\alpha})}$ is a scattering operator for this system.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{theorem}\label{ThScChannel}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Scattering operator $S^{({\alpha})}$ coincides with the component
$s_{\alpha\alpha}$ of the scattering operator $S$, $S=U^{(-)*}U^{(+)}$,
for a system described by the two--channel Hamiltonian ${{\bf H}}$.
%
\end{theorem}
\noindent{\sl Proof.} Let us show that operator $S^{({\alpha})}$
has the kernel $s_{\alpha\alpha}(\lambda,\lambda')$
given by Eq.~(\ref{scattering}).
To do this, remember that
$\tilde{\Psi}_{\alpha}^{(-)}=\Psi_{\alpha}^{(-)}-
Q_{\alpha\beta} u_{\beta\alpha}^{(-)}$
(see Theorem~\ref{ThExpansion}). Therefore,
$$
%
S^{({\alpha})}=(\Psi_{\alpha}^{(-)*}-
u_{\beta\alpha}^{(-)*} Q_{\alpha\beta}^{*})\Psi_{\alpha}^{(+)}=
%
\Psi_{\alpha}^{(-)*} \Psi_{\alpha}^{(+)} +
u_{\beta\alpha}^{(-)*} Q_{\beta\alpha} \Psi_{\alpha}^{(+)} =
%
u_{\alpha\alpha}^{(-)*} u_{\alpha\alpha}^{(+)} +
u_{\beta\alpha}^{(-)*} u_{\beta\alpha}^{(+)} .
%
$$
Here, we have used the properties
${\Psi}^{(\pm)}_{\alpha} = u_{\alpha\alpha}^{(\pm)} $,
$Q_{\alpha\beta}^{*}=-Q_{\beta\alpha}$ and
$Q_{\beta\alpha} \Psi_{\alpha}^{(+)} =u_{\beta\alpha}^{(+)}$
established above.
Since
$$
%
u_{\alpha\alpha}^{(-)*} u_{\alpha\alpha}^{(+)} +
u_{\beta\alpha}^{(-)*} u_{\beta\alpha}^{(+)} =
%
\left( U^{(-)*}U^{(+)} \right) _{\alpha\alpha} = s_{\alpha\alpha},
%
$$
we come to the statement of Theorem. The proof of
Theorem~\ref{ThScattering} is completed.
\medskip
A kernel of the scattering operator $S^{({\alpha})}$ may be presented
also in a usual way~(\ref{scattering}) in terms of the $t$-matrix
$t_{\alpha}(z)=W_{\alpha}-W_{\alpha}(H_{\alpha}-z)^{-1}W_{\alpha}$,
taken on the energy--shell.
Note that $t_{\alpha}(z)$ differs from $T_{\alpha\alpha}(z)$
introduced in Sec.~\ref{Initial}.
Did, easy calculations show that
\be
\label{Tmatrix}
%
t_{\alpha}(z)=B_{\alpha\beta}
[I_{\beta} +Q_{\beta\alpha} (A_{\alpha} -z)^{-1} B_{\alpha\beta}]^{-1}
Q_{\beta\alpha} .
%
\ee
Using the basic equation~(\ref{QbasicSym}) one can rewrite~(\ref{Tmatrix})
in the form
$$
%
t_{\alpha}(z)=T_{\alpha\alpha}(z) + \tilde{t}_{\alpha}(z)
%
$$
where
$$
%
\tilde{t}_{\alpha}(z)= B_{\alpha\beta} [A_{\beta} -
%
B_{\beta\alpha}(A_{\alpha} -z)^{-1}
B_{\alpha\beta}]^{-1}Q_{\alpha\beta} (A_{\alpha} -z)\not\equiv 0.
%
$$
However the additional term $\tilde{t}_{\alpha}(z)$
is evidently disappearing on the energy-shell due to presence of the
difference $A_{\alpha} -z$ as an end factor. Actually, in the
diagonal representation~(\ref{Neumann}),(\ref{multi}),
$A_{\alpha} -z$ acts as the factor $\lambda-z$
vanishing at $z=\lambda+i0$.
Therefore, kernels of $t$-matrices
$t_{\alpha}$ and $T_{\alpha\alpha}$ coincide
on the energy surface.
Note also that in our case
$\sigma_1^c\bigcap\sigma_2^c=\emptyset$. Hence we have
$s_{{\beta}{\alpha}}=0$ and
$S^{({\alpha})}=s_{{\alpha}{\alpha}}$ is unitary.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\medskip
\acknowledgements
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Author is thankful to Prof. B.S.Pavlov and Dr.~K.A.Makarov for
support and interest to this work. The author is much indebted to
Prof.~R.A.Minlos and participants of his seminar in Moscow
University for discussion of the presented results. Particularly,
the author is grateful to Dr.~S.A.Stepin for the literary
indications.
{The work is supported in part by the International Science Foundation
(Grants~\#RFB000 and~\#RFB300) and the RAS Academy of Natural Sciences.}
%
\newpage
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{references}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\bibitem{Dashen} R.F.Dashen and J.B.Healy, I.J.Muzinich,
{ Ann. Phys.} {\bf 102}, 1 (1976).%
\bibitem{JaffeLow} R.L.Jaffe and F.E.Low, { Phys. Rev.}
{\bf D 19}, 2105 (1979).%
\bibitem{Simonov} Yu.A.Simonov, { Yad.Fiz.} {\bf 36}, 722 (1982).%
\bibitem{Kerbikov} B.O.Kerbikov, { Yad. Fiz.} {\bf 65}, 379 (1985).%
\bibitem{NakaishiSchmid} S.Nakaishi--Maeda and E.W.Schmid,
{ Phys. Rev.} {\bf C 35}, 799 (1987).%
\bibitem{NarodetskyPalanga} I.M.Narodetsky, in
{ III All-Union School on Few Body and Quark--Hadronic
Problems} (Vilnous, 1986, in Russian), Part~I, p.~116.%
\bibitem{McKellarMcKay}
B.H.J.McKellar and C.M.McKay, { Aust.J.Phys.}{\bf 36},
607 (1983).%
\bibitem{OrlowskiKim} M.Orlowski and Y.E.Kim, { Phys. Rev. }
{\bf D 32}, 1376 (1985).%
\bibitem{Braun}
M.A.Braun, { Teor.Mat.Fiz.}{\bf 72}, 394 (1987).%
\bibitem{Schmid}
E.W.Schmid, { Helvetica Phys.Acta} {\bf 60}, 394 (1987).%
\bibitem{Pavlov84} B.S.Pavlov, { Teor. Mat. Fiz.}
{\bf 59}, 345 (1984).%
\bibitem{YaF88}
Yu.A.Kuperin, K.A.Makarov, S.P.Merkuriev and
A.K.Motovilov, { Yad.Fiz.} {\bf 48}, 358 (1988).%
\bibitem{KMMMP} Yu.A.Kuperin, K.A.Makarov, S.P.Merkuriev,
A.K.Motovilov and B.S.Pavlov, { J. Math. Phys.} {\bf 31},
1681 (1990).%
\bibitem{MinlosTMF} R.A.Minlos, { Teor. Mat. Fiz.}
{\bf 92}, 255 (1992) [English translation in
Theor. Math. Phys.].%
\bibitem{MyTMF93} A.K.Motovilov, { Teor. Mat. Fiz.}
{\bf 97}, 163 (1993) [English translation in
Theor. Math. Phys.].%
\bibitem{IMU} K.N.Ilinsky, A.K.Motovilov and V.M.Uzdin,
{ Helv. Phys. Acta} {\bf 66}, 337 (1993);
{ Poverkhnost (Russian J.~Surface) }
{\bf 2}, 12 (1994).%
\bibitem{MotJMPh91}
A.K.Motovilov, { J.Math.Phys.} {\bf 32}/12, 3509 (1991).%
\bibitem{Kazimierz} A.K.Motovilov, in
{Proceedings of the
XIII Warsaw Sympos. on Elementary
Particle Physics} /Eds.: Z.Ajduk {\it et al.}
(World Scientific, 1990), pp.~494--499.%
\bibitem{SPbWorkshop} A.K.Motovilov, in {
Invited Talks of the Intern.Workshop ``Mathematical
Aspects of the Scattering Theory and Applications''}
(St.Petersburg, 1991), pp.~101--108.%
\bibitem{LadyzhFaddeev}
O.A.Ladyzhenskaya and L.D.Faddeev, { Doklady AN SSSR}
{\bf 120}, 1187 (1958).%
\bibitem{Faddeev64}
L.D.Faddeev, { Trudy Mat. In-ta AN SSSR} {\bf 73}, 292
(1964).%
\bibitem{MalyshevMinlos1} V.A.Malyshev and R.A.Minlos,
{ J. Stat. Phys.} {\bf 21}, 231 (1979).%
\bibitem{MalyshevMinlos2} V.A.Malyshev and R.A.Minlos,
{ Comm. Math. Phys.} {\bf 82}/2, 211 (1981).%
\bibitem{LangerAdamyan} V.M.Adamjan and H.Langer,
{``Spectral properties of a class
of rational operator-value functions''},
accepted for publication
in {J.Operator Theory}.%
\bibitem{RfunfNaboko} Remember that if $N_{\alpha}\geq 0$
then Eq.~(\ref{Rfunction}) gives a
general form of $R$-function on
${\cal H}_{\alpha}$, i.e. an analytic at
${\rm Im}\, z\neq 0$
${{\bf B}} ({\cal H}_{\alpha},
{\cal H}_{\alpha})$--value function with
positive imaginary part for $z:\, {\rm Im}\, z > 0$
(see paper~\cite{Naboko} and Refs. therein).%
\bibitem{Naboko} S.N.Naboko, { Funktsionalnyi analiz i ego
prilozhenia} {\bf 25}, 1 (1991).%
\bibitem{BirmanSolomiak}
M.S.Birman and M.Z.Solomiak, {``Spectral Theory of
Self-Adjoint Operators in Hilbert Space''} (Leningrad
U.P., Leningrad, 1980, in Russian).%
\bibitem{Reid}
W.T.Reid, {``Riccati differential equations}
(Academic Press'', N.Y.---London, 1972).%
\bibitem{Faddeev63}
L.D.Faddeev, { Trudy Mat. In-ta AN SSSR} {\bf 69}, 1-125
(1963) [English translation: {``Mathematical
aspects of the three--body problem in quantum mechanics''.}
Israel Program for Scientific Translations,
Jerusalem, 1965].%
\vspace*{-0.24cm}
\end{references}
\end{document}