\documentclass[twoside]{article}
\usepackage{amsfonts} % used for R in Real numbers

\pagestyle{myheadings} \markboth{ Existence of a principal
eigenvalue } { Daniela Lupo \& Kevin R. Payne}
\begin{document}
\setcounter{page}{173}
\title{\vspace{-1in}\parbox{\linewidth}{\footnotesize\noindent
Nonlinear Differential Equations, \newline
Electron. J. Diff. Eqns., Conf. 05, 2000, pp. 173--180\newline
http://ejde.math.swt.edu or http://ejde.math.unt.edu
\newline ftp  ejde.math.swt.edu or ejde.math.unt.edu (login: ftp)}
 \vspace{\bigskipamount} \\
%
 Existence of a principal eigenvalue for the Tricomi problem
%
\thanks{ {\em Mathematics Subject Classifications:}  35M10, 35P05, 35B10
 \hfil\break\indent
{\em Key words:} Tricomi problem, mixed type equations, spectral theory,
principal eigenvalue, maximum principle
 \hfil\break\indent
\copyright 2000 Southwest Texas State University.
\hfil\break\indent Published October 25, 2000.  \hfil\break\indent
Both authors were supported by MURST, Project
``Metodi Variazionali ed Equazioni  \hfil\break\indent
Differenziali Non Lineari''} }

\date{}
\author{ Daniela Lupo \& Kevin R. Payne \\[12pt]
{\em Dedicated to Alan Lazer} \\
{\em on  his 60th birthday }}
\maketitle

\begin{abstract}
The existence of a principal eigenvalue is established for the Tricomi problem
in normal domains; that is, the existence of a positive
eigenvalue of minimum modulus with an associated positive eigenfunction.
The argument here uses prior results of the authors on the generalized
solvability in weighted Sobolev spaces and associated
maximum/minimum principles \cite{[LP2]} coupled with known results
of Krein-Rutman type.
\end{abstract}

\renewcommand{\theequation}{\thesection.\arabic{equation}}
\catcode`@=11 \@addtoreset{equation}{section}
\catcode`@=12


\section{Introduction}

In this note, we are interested in
establishing the existence of a principal eigenvalue associated to generalized
solutions of the classical linear Tricomi problem. That is; we seek to
find an eigenvalue-eigenfunction pair $(\lambda, u)$ with $\lambda >
0$ and $u$ a positive generalized solution to the problem
$$\displaylines{
\hfill Tu  =\lambda u \quad \mbox{in } \Omega \hfill\llap{(LTE)}\cr
\quad u=0 \quad\mbox{on } AC\cup\sigma.
} $$
The results will apply as well to the conjugate problem (LTE)* in which the boundary
conditions $u = 0$ are placed on $BC\cup\sigma$ instead.
Here $T\equiv -y\partial_x^2-\partial_y^2$ is the
Tricomi operator on ${\mathbb R}^2$
and $\Omega$ is a bounded region in ${\mathbb R}^2$ with piecewise smooth boundary $\partial \Omega$
of the classical Tricomi form. That is, $\partial \Omega$
consists of a smooth arc $\sigma$ in the elliptic region $y>0$, with
endpoints on the $x$-axis at $A=(-x_0,0)$ and $B=(x_0,0)$,
and two characteristic arcs $AC$ and $BC$ for the Tricomi operator
in the hyperbolic region $y<0$ issuing from $A$ and $B$ and
meeting at the point $C$ on the $y$-axis (we assume without loss of
generality that $A$ and $B$ are symmetric with respect to the $y$-axis).
One knows that
$$      AC:\ (x + x_0) -{2\over 3}(-y)^{3/2}=0 \quad\mbox{and}\quad
 BC:\ (x - x_0) +{2\over 3}(-y)^{3/2}= 0\,.
$$
We will call such a domain a {\em Tricomi domain}.
Moreover, we will assume that the domain is {\em normal} in the
sense that $\sigma$ is perpendicular to the x-axis in the points
$A$ and $B$.

    The underlying Tricomi problem
$$ \displaylines{
\hfill Tu  = f \quad\mbox{in }\Omega \hfill\llap{(LT)}\cr
 u=0 \quad\mbox{on }AC\cup\sigma
}$$
has a notable physical importance. It describes, in the
hodograph plane, the problem of transonic flow through a nozzle; a
connection first established by Frankl' \cite{[F]}. The placement
of the boundary condition on only a portion $AC\cup\sigma$ of the boundary
can yield a well posed problem for classical solutions as first
established by Tricomi himself \cite{[T]} in special cases, whereas placement
of data on larger portions of the
boundary will overdetermine the problem for classical solutions
due to the presence of hyperbolicity; energy integral methods
based on the work of Friedrichs \cite{[Fr]} yield suitable
uniqueness theorems. In addition, there are a wealth of results on
existence and uniqueness of strong solutions in Hilbert spaces
well adapted to the boundary condition. However, despite its physical importance
and despite some 70 years of
study, results on the linear Tricomi problems (LT) and (LT)*
are not complete. In particular, there is an almost complete
absence of spectral theory which is glaring in its own right and
impedes substantially progress on associated nonlinear problems.
Recent works such as \cite{[LP1]} and \cite{[LMP]} have attempted
to make progress in this direction. One major difficulty lies in the
fact that the problem (LT) is {\bf not} self-adjoint.

    In part as preparation for the current study, we have
established results on the generalized solvability of (LT)
in weighted Sobolev spaces and associated maximum/minimum principles for
normal domains \cite{[LP2]} (which will be recalled in section 2).
It is well known that the presence of extremum principles if
interpretable as the invariance of a positive cone in a suitable
Banach space can be a first step in developing a spectral theory
even in cases where the operator is not self adjoint. Such ideas
were pioneered by Krein and Rutman \cite{[KR]} and have played an
important role in nonlinear analysis (cf.\ Krasnoselskii
\cite{[K]}). As noted in \cite{[LP2]}, the solution operators
$\widetilde T_{AC}^{-1}$ and $\widetilde T_{BC}^{-1}$ to the Tricomi problems (LT)
and (LT)* are compact as operators on $L^2(\Omega)$ and thanks to the
maximum/minimum principles preserve the positive cone in $L^2(\Omega)$. Combining
these facts with an argument of Krein-Rutman type using the Hopf maximum principle
in the elliptic part of the domain, we establish the following
result on the existence of a principal eigenvalue. The relevant
definitions and notations will be recalled in section 2.

\vspace{2ex}
\noindent
{\bf Theorem 1.1} {\em Let $\Omega$ be a normal admissible Tricomi
domain. Then there exists a positive eigenvalue $\lambda_0$
of minimum modulus in the sense that $|\lambda_0| \leq |\lambda|$ for every
$\lambda \in \sigma(\widetilde T_{AC})$. Moreover, associated to $\lambda_0$ there are
corresponding eigenfunctions $u_0\in \widetilde{W}^1_{AC \cup \sigma}\subset L^2(\Omega)$
and  $v_0\in \widetilde{W}^1_{BC \cup \sigma}\subset L^2(\Omega)$ for the problems
$(LTE)$ and $(LTE)^*$ which are positive; that is, $u_0, v_0 \geq 0$ a.e. in $\Omega$.}

\vspace{2ex}

    We point out that $\lambda_0$ is called a principal eigenvalue
due to the positivity of the associated eigenfunction and its being of
minimum modulus, as done by Lazer in \cite{[Lz], [CL]}. However, we cannot say
at present that the associated eigenspace is simple, nor that other
eigenspaces do not contain eigenfunctions that are non-negative
almost everywhere, as happens in the purely elliptic case (cf.\
Manes - Micheletti \cite{[MM]} as well as Hess - Kato \cite{[HK]}).


\section{Generalized solvability and maximum/minimum principles}

In this section, we will recall the main definitions, notation,
and results on generalized solutions that will be used in the
proof of the main theorem. We begin by recalling the spaces of
functions in which we will work. In all that follows, $\Gamma$
will be a connected subset of $\partial \Omega$ which is assumed
to be piecewise $C^1$ (in order to apply the divergence theorem).
We consider the following spaces of smooth functions
\begin{equation}
    C^{\infty}_{0, \Gamma}(\overline{\Omega}) = \left\{ \psi \in C^{\infty}(\overline{\Omega}) : \psi \equiv 0
\ {\rm on} \  N_{\epsilon} \Gamma \ \ {\rm for\ some}\ \epsilon > 0 \right\},
\end{equation}
where $N_{\epsilon}\Gamma$ is an $\epsilon$ neighborhood of
$\Gamma$. We denote by $\widetilde{W}^1_{\Gamma} (\Omega)$ the Sobolev space
obtained as closure of the spaces in $(2.1)$ with respect to the
norm
$$
    || \psi ||^2_{\widetilde{W}^1_{\Gamma} (\Omega)}  = || \psi ||^2_{\widetilde{W}^{1,2}
    (\Omega)} = \int_{\Omega} \left( |y| \psi_x^2 + \psi_y^2 + \psi^2 \right) \, dx\,dy ,
    $$
and denote by $\widetilde{W}^{-1}_{\Gamma}$ the dual space to $\widetilde{W}^1_{\Gamma}$
equipped with its negative norm in the sense of Lax (cf. \cite{[L]}). We note that using the
definition of the $\widetilde{W}^{-1}_{\Gamma} (\Omega)$ norm, one has the following
estimates: for each $\Omega$ with $\partial \Omega$ piecewise $C^1$, there
exist constants $C_1, C_2 > 0$ such that
\begin{equation}
    ||Tu||_{\widetilde{W}^{-1}_{BC \cup \sigma} (\Omega)} \leq C_1
    ||u||_{\widetilde{W}^1_{AC \cup \sigma} (\Omega)},\ \ u \in
    C^{\infty}_{0, AC \cup \sigma}(\overline{\Omega})
    \end{equation}
and
\begin{equation}
 ||Tv||_{\widetilde{W}^{-1}_{AC \cup \sigma} (\Omega)} \leq C_2
 ||v||_{\widetilde{W}^1_{BC \cup \sigma} (\Omega)},\ \ v \in
    C^{\infty}_{0, BC \cup \sigma}(\overline{\Omega})
    \end{equation}
which are just continuity estimates. They give rise to continuous
extensions of the Tricomi operator $T$ (defined on dense subspaces of smooth functions)
such as
\begin{equation}
 \widetilde T_{AC} : \widetilde{W}^1_{AC \cup \sigma} (\Omega) \to \widetilde{W}^{-1}_{BC \cup \sigma} (\Omega) \ \ {\rm and}
    \ \ \widetilde T_{BC}: \widetilde{W}^1_{BC \cup \sigma} (\Omega) \to \widetilde{W}^{-1}_{AC \cup \sigma} (\Omega) .
\end{equation}
We recall that the placement of the boundary conditions on only a
portion of the boundary implies that the problems (LT) and
(LT)* are not self adjoint. In fact, one has that the operators
in $(2.4)$ satisfy $\widetilde T_{BC} = \widetilde T_{AC}^*$.

    As shown by Didenko, a necessary and sufficient condition to have
the generalized solvability for the problem (LT) and (LT)*
for each $f \in L^2(\Omega)$ is to have the continuity estimates
$(2.2)$ and $(2.3)$ as well as the {\em a priori}\ estimates of {\em
admissibility} as encoded in the following definition.

\vspace{2ex}

\noindent
{\bf Definition 2.1.} {\em A Tricomi domain $\Omega$ will be said to be admissible
if there exist positive constants $C_3$ and $C_4$ such that}
\begin{equation}
    \Vert u \Vert_{L^2(\Omega)} \leq C_3 \Vert Tu \Vert_{\widetilde{W}^{-1}_{BC \cup \sigma}},
            \ \ \ \ u \in C^{\infty}_{0, AC \cup \sigma}(\overline{\Omega})
\end{equation}
{\em and}
\begin{equation}
        \Vert v \Vert_{L^2 (\Omega)} \leq C_4 \Vert Tv \Vert_{\widetilde{W}^{-1}_{AC \cup \sigma} (\Omega)},
            \ \ \ \ v \in C^{\infty}_{0, BC \cup \sigma}(\overline{\Omega}) .
\end{equation}

\vspace{2ex}

\noindent
The class of admissible normal domains includes convex domains as well as those
that contain a convex subdomain; we record the
following result of \cite{[LP2]} as an example.

\vspace{2ex}
\noindent
{\bf Example  2.2.} {\em Let $\Omega$ be a normal Tricomi domain,
with boundary $AC\cup BC\cup\sigma$ such that}
\begin{description}
    \item[{i)}] {\em $\Omega$ contains $\Omega_0$ as a subdomain where $\Omega_0$
     has boundary
    $AC \cup BC \cup \sigma_0$ such that the elliptic boundary arc $\sigma_0$ is given as a graph
$\{ (x,y): y = g(x),\ -x_0 \leq x \leq x_0 \}$ which
satisfies the following hypotheses: $ g \in C^2((-x_0, x_0))$,
$ g(\pm x_0) = 0 $, $ g'(\mp x_0^\pm) = \pm \infty$ and, for every  $x \in
(-x_0,x_0)$,
$ g(x) > 0$
and $ g''(x) \leq - k < 0$.
}

    \item[{ii)}] {\em  There exists an  $\epsilon>0$ such that the elliptic boundaries
    $\sigma$ and $\sigma_0$ of
    $\Omega$ and $\Omega_0$ coincide in a  strip $\{(x,y)\ :\ 0\leq y\leq \epsilon\}$.}
        \end{description}

\noindent
{\em Then $\Omega$ is admissible in the sense of Definition 2.1.}

\vspace{2ex}

    As noted above, the {\em a priori} estimates of admissibility together
with some functional analysis yield the following kind of solvability result
(cf.\ section 2 of \cite{[LP2]}).

\vspace{2ex}
\noindent
{\bf Theorem 2.3.} {\em Let $\Omega$ be normal admissible
Tricomi domain in the sense of Definition 2.1.  Then
for every $f \in L^2(\Omega)$ there exists a unique
generalized solution $u \in \widetilde{W}^1_{AC \cup \sigma} (\Omega)$ to the Tricomi
problem (LT) in the sense that
there exists a sequence $\{u_j \} \subset C^{\infty}_{0, AC \cup \sigma}(\overline{\Omega}) $ such that}$$
    \lim_{j \to \infty} \Vert u_j - u \Vert_{\widetilde{W}^1_{AC \cup \sigma}
    (\Omega)} = 0 \quad \mbox{and}\quad
        \lim_{j \to \infty} \Vert Tu_j - f
\Vert_{\widetilde{W}^{-1}_{BC \cup \sigma} (\Omega)}=0.
              $$

\vspace{2ex}

    As a final preliminary, we recall the following
maximum/minimum principle for generalized solutions which is proven by
regularizing the problem, applying a variant of the maximum principle
of Agmon-Nirenberg-Protter for regular solutions \cite{[ANP]}, and
using the continuity of the solution operator $(\widetilde T_{AC})^{-1}$
(cf.\ Theorem 3.1 of \cite{[LP2]}).

\vspace{2ex}
\noindent
{\bf Theorem 2.4.} {\em Let $\Omega$ be an admissible normal Tricomi
domain,
$f\in L^2(\Omega)$ and $u \in \widetilde{W}^1_{AC \cup \sigma} (\Omega)$ the unique generalized solution to the problem (LT).
Then $f \geq 0 (\leq 0)$ a.e.\ in $\Omega$ implies $u \geq 0 (\leq 0)$
a.e.\ in $\Omega$. A similar statement holds for the conjugate problem}
$(LT)^*$


\section{Proof of Theorem 1.1}

    Having recalled the machinery of generalized solvability and
extremum principles, we will show how they give rise to the existence of
a principal eigenvalue. The proof proceeds in three steps. The first makes
use of the solvability theory.

\vspace{2ex}

\noindent
{\em Step 1. (Passage to the inverse operator)}

\vspace{1ex}

    An immediate consequence of the solvabilty result Theorem 2.3 is the existence
of a continuous right inverse $K$ defined on all of $L^2(\Omega)$ whose image
$\widetilde W_A$ is a dense proper subspace of $\widetilde{W}^1_{AC \cup \sigma} (\Omega)$. The operator $K$ gives rise to
a compact operator
$
K=\widetilde T_{AC}^{-1}:L^2(\Omega)\to L^2(\Omega).
$
This follows from standard functional analysis (cf.\ section 2 of
\cite{[LP1]}). It is therefore clear by a classical result (cf. \cite{[Br]} or
\cite{[CH]}) that this injective, non surjective compact operator $K$ has spectrum $\sigma
(K)= \{\mu_j\}_{j\in {\mathbb Z}}$ consisting of eigenvalues of finite
multiplicity with $0$ as the only possible accumulation point.
Therefore it suffices to find an eigenvalue
$\mu_0\in{\mathbb R}^+$ and an associated eigenfunction $v_0\in L^2(\Omega)$ for $K$
satisfying $v_0\geq 0 $ a.e. in $\Omega$, as then
\begin{equation}
\lambda_0=1/\mu_0 \quad {\rm is\ an\ eigenvalue\ for\ }\widetilde T_{AC}
\end{equation}
and
\begin{equation}
u_0=K v_0 \quad {\rm is\ an\ eigenfunction\ for\ }\widetilde T_{AC}.
\end{equation}
Indeed, if $Kv_0=\mu_0v_0$ with $v_0\in L^2(\Omega)$ then one has
$u_0=\mu_0(\widetilde T_{AC} u_0)$ from which it follows that $\widetilde T_{AC} u_0=\lambda_0u_0 $. Moreover,
since $u_0=Kv_0$ is the unique generalized solution to $\widetilde T_{AC} u_0=v_0$,
Theorem 2.4 shows that this solution must obey $u_0\geq 0$ a.e. in $\Omega$
if $v \geq 0$ a.e.\ in $\Omega$.

\vspace{2ex}
\noindent
{\em Step 2. (A result of Krein and Rutman)}

\vspace{1ex}

    By Step 1, it suffices to find a positive eigenvalue $\mu_0$ for
$K=\widetilde T_{AC}^{-1}$ with an associated positive eigenfunction $v_0\in
L^2(\Omega)$ satisfying $v_0 \geq 0$. If we denote by $L(\Omega)^+ = \{ f \in L^2(\Omega) :
f \geq 0 \ \ {\rm a.e.\ in}\ \Omega \}$, the positive cone in $L^2(\Omega)$, Theorem 2.4 shows
that the compact operator $K$ leaves this cone invariant in the sense that
$K \left( L^2(\Omega)^+ \right) \subset L^2(\Omega)^+$. We recall the following special case
of the result of Krein and Rutman (cf.\ Theorem 6.2 of \cite{[KR]}) which suits our needs.

\vspace{2ex}
\noindent
{\bf Lemma 3.1.} {\em Let $K$ be a compact linear operator on
$L^2(\Omega)$ that leaves the positive cone $L^2(\Omega)^+$ invariant. If there exist
an $\alpha>0$ and an $f \in L^2(\Omega)^+$ with $||f||_{L^2(\Omega)} = 1$
such that}
\begin{equation}
    Kf\geq \alpha f \ \ {\rm a.e.\ in}\ \ \Omega,
        \end{equation}
{\em then}
\begin{description}
\item[{i)}] {\em $K$ has nonzero eigenvalues}
\item[{ii)}] {\em Among the eigenvalues of maximum modulus there is
a positive one $\mu_0 \geq \alpha$ to which there correspond eigenvectors
$v_0, w_0 \in L^2(\Omega)^+$ of the operators $K$ and $K^*$ respectively.}
\end{description}

\vspace{2ex}
    The work that remains is to construct a suitable function $f \in L^2(\Omega)^+$
and parameter $\alpha$ so that $(3.3)$ holds. We claim that it is enough to select
any nontrivial $f$ such that
\begin{eqnarray}
    &f \in C^0(\overline{\Omega})\,,& \\
    &f \geq 0 \quad\mbox{in }\Omega\,,& \\
 &\mathop{\rm supp}f \subset \overline{B_r} \subset \Omega^{+}
 = \Omega \cap \{ y  > 0 \}\,,&
        \end{eqnarray}
where $B_r$ is any ball with center in the elliptic
region $\Omega^{+}$ and radius $r$ small enough so that $\overline{B_r} \subset \Omega^{+}$ holds
and to pick
    \begin{equation}
        \alpha = 1/n
        \end{equation}
for $n$ sufficiently large. Indeed, selecting $f \not \equiv 0$ satisfying
$(3.4) - (3.6)$ and then normalizing $f$ to satisfy $||f||_{L^2(\Omega)} = 1$, it remains
to verify condition $(3.3)$ by choosing $n$ large enough in $(3.7)$.

\vspace{2ex}

\noindent
{\em Step 3. (Verification of condition (3.3))}

\vspace{1ex}

    Since $f \in C^0(\overline{\Omega}) \subset L^2(\Omega)$ with $\Omega$ normal and admissible,
one can apply both Theorem 2.3 as well as the classical solvability result of
Agmon \cite{[A]} to show that there exists a unique generalized
solution $u \in \widetilde{W}^1_{AC \cup \sigma} (\Omega) \bigcap C^0(\overline{\Omega})$ to the Tricomi
problem (LT).

    If one restricts the solution $u$ to the set
$\Omega_{\delta}^{+} = \Omega^{+} \cap \{ y > \delta \}$, where $\delta > 0$
is chosen so that $\overline{B_r} \subset \Omega_{\delta}^{+}$, one
has that this restriction belongs to $C^2(\Omega_{\delta}^{+}) \bigcap
C^0(\overline{\Omega_{\delta}^{+}})$ and satisfies
$$\displaylines{
   Tu = f \quad\mbox{in } \Omega_{\delta}^{+} \cr
   u _{|\sigma \cap \{ y \geq \delta \}} = 0 \,.
}$$
Moreover, the restriction of $u$ to $\overline{A'B'}$ is an element
$ g(x) \in C^0({\overline{A'B'}})$
with $u(A') = u(B') = 0$ where $A'$ and $B'$ are the unique points
of intersection of $\sigma$ with the line $y = \delta$. In light
of the needed admissibility of $\Omega$, one might as well assume
that $\Omega \cap \{ y < \delta\}$ is a strip for $\delta$ small
enough (cf.\ the hypotheses in Example 2.2).

    One now may apply the Hopf maximum principle (cf.\ Theorem 3.5
of \cite{[GT]}) to the restriction of $u$ to
$\overline{\Omega_{\delta}^{+}}$ where we note that the Tricomi
operator is uniformly elliptic on the subdomain
$\Omega_{\delta}^{+}$. This yields
$$  u > 0 \quad\mbox{on } \Omega_{\delta}^{+}\,,
$$
where we note that $u$ is not constant since $u$ vanishes on the upper boundary
which is a subset of $\sigma$ but $u \not \equiv 0$ since $f \not \equiv 0$
in $\Omega_{\delta}^{+}$.

    We are now ready to show that
\begin{equation}
   u(x,y) \geq \alpha f(x,y) \ \ {\rm for\ each}\ \ (x,y) \in \Omega
\end{equation}
by choosing $\alpha$ small enough. We have $u \geq 0$ using $(3.5)$ and the
maximum principle of Theorem 2.4 and hence outside the support of
$f = \overline{B_r}$ one has $(3.8)$ for each $\alpha > 0$. Within the support of
$f$, since $u, f \in C^0(\overline{B_r})$, there exist minimum and
maximum values $m, M$ such that
\begin{equation}
    0 < m = \min_{\overline{B_r}} u \quad\mbox{and}\quad
    M = \max_{\overline{B_r}} f \,.
\end{equation}
Using the bounds (3.9) it suffices to pick $\alpha = 1/n$
with $n$ large enough to give $n \geq M/m$. This completes the
proof.


\begin{thebibliography}{99} {\frenchspacing

\bibitem{[A]} S. Agmon, Boundary value problems for equations of
mixed type,{\em Convegno Internazionale sulle Equazioni Lineari alle Derivate
Parziali, Trieste}, (1954), pp. 54--68. Edizioni Cremonese, Roma, 1955.

\bibitem{[ANP]} S. Agmon, L. Nirenberg, and M. H. Protter, A maximum principle
for a class of hyperbolic equations and applications to equations of
mixed elliptic-hyperbolic type, {\em Comm. Pure Appl. Math.}\ {\bf 6}\
(1953), 455-470.

\bibitem{[Br]} H. Brezis, ``Analyse Functionelle'', Masson, Paris, 1983.

\bibitem{[CL]} A. Castro and A. C. Lazer, Results on periodic
solutions of parabolic equations suggested by elliptic theory,
{\em Boll. Un. Mat. Ital. B (6)}\ {\bf 1}\ (1982), 1089-1104.

\bibitem{[CH]} R. Courant and D. Hilbert, `` Methods of Mathematical
Physics, Vol. 1'', Interscience, New York, 1953.

\bibitem{[F]} F. I. Frankl', On the problems of Chaplygin for mixed
sub- and supersonic flows, {\em Isv. Akad. Nauk. USSR Ser. Mat.}
{\bf 9}\ (1945), 121-143.

\bibitem{[Fr]} K. O. Friedrichs, Symmetric positive linear differential
equations, {\em Comm. Pure Appl. Math.} {\bf 11}\ (1958), 338-418.

\bibitem{[G]} N. N. Gadai, Existence of a spectrum for Tricomi's
operator, {\em Differential Equations}, {\bf 17}\ (1981), 20-25.

\bibitem{[GT]} D. Gilbarg and N. S. Trudinger, ``Elliptic
Partial Differential Equations of Second Order'', Springer-Verlag,
New York, 1977.

\bibitem{[HK]} P. Hess and T. Kato, On some linear and nonlinear
eigenvalue problems with an indefinite weight function, {\em
Comm. Partial Differential Equations}\ {\bf 5}\ (1980), 999-1030.

\bibitem{[K]} M. A. Krasnoselskii, ``Positive solutions of
operators equations'', Noordhoof Ltd., Groningen, 1964.

\bibitem{[KR]} M. G. Krein and M. A. Rutman, Linear operators
leaving invariant a cone in a Banach space, {\em Uspehi Matem.\ Nauk.}\ {\bf 3} N. 1(23)
(1948), 3-95; translation in {\em Amer.\ Math.\ Soc.\ Translation}\ {\bf
1950}\ n.\ 26\ (1950).


\bibitem{[L]} P. Lax, On Cauchy's problem for hyperbolic equations
and the differentiability of solutions of elliptic equations, {\em Comm. Pure
Appl. Math.} {\bf 8}\ (1955), 615-633.

\bibitem{[Lz]} A. C. Lazer, Some remarks on periodic solutions
of parabolic differential equations, in ``Dynamical Systems II
(Gainesville, Fla. 1981)'', Academic Press, New-York-London,
1982, 227-246.

\bibitem{[LMP]} D. Lupo,\ A. M. Micheletti and K. R. Payne,\
Existence of eigenvalues for reflected Tricomi operators and applications
to multiplicity of sultions for sublinear and asymptotically linear nonlocal Tricomi problems,
{\em Advances in Diff. Equations}\ {\bf 4}\ (1999), 391-412.

\bibitem{[LP1]} D. Lupo and K. R. Payne, A dual variational approach to
a class of nonlocal semilinear Tricomi problems,
{\em NoDEA Nonlinear Differential Equations Appl.}\ {\bf 6}\ (1999), 247-266 .

\bibitem{[LP2]} D. Lupo and K. R. Payne,
On the maximum principle for generalized solutions to the Tricomi problem,
{\em Commun.\ Contemp.\ Math.}, to appear.

\bibitem{[MM]} A. Manes and  A. M. Micheletti, Un'estensione della teoria
variazionale classica degli autovalori per operatori ellittici del secondo
ordine, {\em Boll. Un. Mat. Ital. (4)}, {\bf 7}(1973), 285-301.

\bibitem{[T]} F. G. Tricomi, Sulle equazioni lineari alle derivate
parziali di secondo ordine, di tipo misto, {\em Atti Acad. Naz. Lincei Mem.\
Cl. Fis. Mat. Nat.}\  (5)\ {\bf 14}\ (1923), 134-247.

}\end{thebibliography}
\medskip

\noindent
{\sc Daniela Lupo} (e-mail: danlup@mate.polimi.it) \\
{\sc Kevin R. Payne} (e-mail: paynek@mate.polimi.it) \\
 Dipartimento di Matematica, Politecnico di Milano\\
 Piazza Leonardo da Vinci, 32\\
 20133 Milano, Italy


\end{document}