1: \NeedsTeXFormat{LaTeX2e}[1994/06/01]
2: % LaTeX 2.09 can't be used (nor non-LaTeX)
3: %
4: \documentclass[11pt, a4paper, leqno]{article}
5: \usepackage[intlimits]{amsmath}
6: \usepackage{amsthm,amssymb}
7: \usepackage{amsfonts}
8: \usepackage{fancyhdr}
9: \usepackage{verbatim}
10: %\usepackage[notref, notcite,color]{showkeys}
11: %\usepackage{color}
12: %\definecolor{labelkey}{rgb}{1,0.5,0.5}
13: \usepackage{graphicx}
14: %usepackage[first,bottomafter,light]{draftcopy}
15: %
16: %\draftcopyName{DRAFT M.Levitin@ma.hw.ac.uk}{60}
17: %
18: %\input /home/levitin/latex2emathdef
19: %\input latex2emathdef
20: %
21: %\setlength{\headheight}{15pt}
22: %
23: \pagestyle{fancy}
24: \setlength{\headheight}{14pt}
25: \lhead{\bfseries Johnson/Levitin/Parnovski --- September 2004}
26: \chead{}
27: \rhead{Page \thepage}
28: \lfoot{\bfseries Existence of eigenvalues of an operator pencil in a curved waveguide}
29: \cfoot{}
30: \rfoot{}
31: \renewcommand{\footrulewidth}{\headrulewidth}
32: %
33: \renewcommand{\familydefault}{cmss}
34: %
35: %
36: \DeclareMathOperator{\im}{Im}
37: \DeclareMathOperator{\re}{Re}
38: \DeclareMathOperator{\Tr}{Tr}
39: \DeclareMathOperator{\dist}{dist}
40: \DeclareMathOperator{\grad}{grad}
41: \DeclareMathOperator{\diver}{div}
42: \DeclareMathOperator{\supp}{supp}
43: \DeclareMathOperator{\Dom}{Dom}
44: \newcommand{\bgrad}{\text{\bf grad}\,}
45: %
46: \newcommand{\alp}{\alpha}
47: \newcommand{\bet}{\beta}
48: \newcommand{\gam}{\gamma}
49: \newcommand{\del}{\delta}
50: \newcommand{\eps}{\varepsilon}
51: \newcommand{\zet}{\zeta}
52: \newcommand{\tet}{\theta}
53: \newcommand{\kap}{\kappa}
54: \newcommand{\lam}{\lambda}
55: \newcommand{\sig}{\sigma}
56: \newcommand{\varsig}{\varsigma}
57: \newcommand{\vphi}{\varphi}
58: \newcommand{\Del}{\Delta}
59: \newcommand{\Lam}{\Lambda}
60: %
61: \newcommand{\ubold}{\mathbf{u}}
62: \newcommand{\xbold}{\mathbf{x}}
63: \newcommand{\Acal}{\mathcal{A}}
64: \newcommand{\Bcal}{\mathcal{B}}
65: \newcommand{\Lcal}{\mathcal{L}}
66: \newcommand{\Mcal}{\mathcal{M}}
67: \newcommand{\Rbb}[1][]{\mathbb{R}^{#1}}
68: \newcommand{\Cbb}[1][]{\mathbb{C}^{#1}}
69: \newcommand{\Nbb}{\mathbb{N}}
70: \newcommand{\dpar}{\partial}
71: \newcommand{\drm}{\mathrm{d}}
72: \newcommand{\erm}{\mathrm{e}}
73: \newcommand{\muu}{\underline{\mu}{}}
74: %
75: \newcommand{\dsp}{\displaystyle}
76: %\newcommand{\mlremark}[1]{\marginpar{\color{red}\raisebox{0.5cm}{\fbox{\parbox{\marginparwidth}{ML: #1}}}}}
77: %\newcommand{\lpremark}[1]{\marginpar{\color{blue}\raisebox{0.5cm}{\fbox{\parbox{\marginparwidth}{LP: #1}}}}}
78: %\newcommand{\tjremark}[1]{\marginpar{\color{green}\raisebox{0.5cm}{\fbox{\parbox{\marginparwidth}{TJ: #1}}}}}
79: %
80: \allowdisplaybreaks[4]
81: %
82: % additional \newcommand should be put here
83: \newcommand{\scal}[2]{\ensuremath{\langle #1,#2\rangle}}
84: \DeclareMathOperator{\spec}{spec}
85: \DeclareMathOperator{\spp}{\spec_{\text{pt}}}
86: \DeclareMathOperator{\spe}{\spec_{\text{ess}}}
87: \DeclareMathOperator{\spd}{\spec_{\text{dis}}}
88: %
89: \numberwithin{equation}{section}
90: %
91: \theoremstyle{plain}% default
92: \newtheorem{thm}{Theorem}[section]
93: \newtheorem{tm}{Theorem}[section]
94: \newtheorem{lem}[thm]{Lemma}
95: \newtheorem{lm}[thm]{Lemma}
96: \newtheorem{prop}[thm]{Proposition}
97: \newtheorem{cor}[thm]{Corollary}
98: %\newtheorem*{VT}{Virial Lemma} %new entry, moved from below
99: %
100: \theoremstyle{definition}
101: \newtheorem{defn}[thm]{Definition}
102: \newtheorem{conj}[thm]{Conjecture}
103: \newtheorem{exmp}[thm]{Example}
104: %
105: \theoremstyle{remark}
106: \newtheorem{rem}[thm]{Remark}
107: \newtheorem{note}[thm]{Note}
108: %
109: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
110: \newcommand{\pd}[2]{\frac{\partial{#1}}{\partial{#2}}}
111: \newcommand{\bu}{\mathbf{u}}
112: \newcommand{\bdiv}{\text{\bf div}\,}
113: \newcommand{\nhat}{\hat{\mathbf{n}}}
114: \newcommand{\ii}{\rm i}
115: \newcommand{\const}{\text{const}}
116: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
117: \begin{document}
118: %
119: \title{Existence of eigenvalues of a linear operator pencil in a curved waveguide --- localized shelf
120: waves on a curved coast}
121: \author{
122: E R Johnson\\
123: \normalsize\small Department of Mathematics, University College London\\
124: \normalsize\small Gower Street, London WC1E 6BT, U.~K.\\
125: \normalsize\small email {\sffamily erj@math.ucl.ac.uk}
126: \and
127: Michael Levitin\\
128: \normalsize\small Department of Mathematics, Heriot-Watt University\\
129: \normalsize\small Riccarton, Edinburgh EH14 4AS, U.~K.\\
130: \normalsize\small email {\sffamily M.Levitin@ma.hw.ac.uk}
131: \and
132: Leonid Parnovski\\
133: \normalsize\small Department of Mathematics, University College London\\
134: \normalsize\small Gower Street, London WC1E 6BT, U.~K.\\
135: \normalsize\small email {\sffamily Leonid@math.ucl.ac.uk}}
136: %
137: %
138: %\subjclass{}
139: %
140: \date{%
141: %\small Version 7
142: September 2004%\today
143: \thanks{The research of M.L. and L.P. was
144: partially supported by the EPSRC Spectral Theory Network}}
145: %
146: \maketitle
147: %
148: \begin{abstract} \noindent The study of the possibility of existence of the non-propagating, trapped
149: continental shelf waves (CSWs)along curved coasts reduces mathematically to a spectral problem for a self-adjoint operator
150: pencil in a curved strip. Using the methods developed in the setting
151: of the waveguide trapped mode problem, we show that such CSWs exist for a wide class of coast curvature and depth profiles.
152: \end{abstract}
153: %
154: {\small \textbf{Keywords:} continental shelf waves; curved coasts; trapped modes; essential spectrum; operator pencil}
155:
156: \
157:
158: \noindent{\small \textbf{2000 Mathematics Subject Classification: }
159: 35P05,35P15,86A05, 76U05}
160: \newpage
161: %
162: \section{Introduction}
163:
164: Measurements of velocity fields along the coasts of oceans throughout the
165: world show that much of the fluid energy is contained in motions with periods
166: of a few days or longer. The comparison of measurements at different places
167: along the same coast show that in general these low-frequency disturbances
168: propagate along coasts with shallow water to the right in the northern
169: hemisphere and to the left in the southern hemisphere. These waves have come
170: to be known as continental shelf waves (CSWs). The purpose of the present
171: paper is to demonstrate, using the most straightforward model possible, the
172: possibility of the non-propagating, trapped CSWs along curved coasts.
173: The existence of such non-propagating modes would be significant as they
174: would tend to be forced by atmospheric weather systems, which have similar
175: periods of a few days, similar horizontal extent, and a reasonably broad
176: spectrum in space and time. Areas where such modes were trapped would thus
177: appear to be likely to show higher than normal energy in the low frequency
178: horizontal velocity field.
179:
180: %Below in this Introduction we give a brief summary of the derivation of the governing equations
181: %and boundary conduction and an overview of previous results.
182:
183:
184: The simplest models for CSWs take the coastal oceans to be inviscid and of
185: constant density. Both these assumptions might be expected to fail in various regions such as when strong
186: currents pass sharp capes or when the coastal flow is strongly stratified.
187: However for small amplitude CSWs in quiescent flow along smooth coasts
188: viscous separation is negligible.
189: Similarly most disturbance energy is concentrated in the modes with the least vertical
190: structure, which are well described by the constant density model \cite{LeBloM78}. The
191: governing equations are then simply the rotating incompressible Euler
192: equations. Further, coastal flows are shallow in the sense that the ratio
193: of depth to typical horizontal scale is small. Expanding the rotating incompressible Euler equations in
194: powers of this ratio and retaining only the leading order terms gives the rotating
195: shallow water equations \cite{Pedlo86}.
196: \begin{align}
197: \label{SWE1}
198: \pd{\bu}{t} + \bu\cdot\bgrad\bu - 2\Omega\mathbf{k}\times\bu &= -g\: \bgrad \widetilde{H}, \\
199: \pd{\widetilde{H}}{t} + \bdiv [(\widetilde{H}+H)\bu] &= 0.
200: \label{SWE2}
201: \end{align}
202: Here $x$ and $y$ are horizontal coordinates in a frame fixed to the rotating Earth, $\mathbf{k}$
203: is a vertical unit vector, $\bu(x,y,t)$ is the horizontal velocity, $\Omega$ is the (locally constant) vertical
204: component of the
205: Earth's rotation, $g$ is gravitational acceleration, $\widetilde{H}(x,y,t)$ is the vertical
206: displacement of the free surface and $H(x,y)$ is the local undisturbed fluid depth.
207:
208:
209: System \eqref{SWE1},\eqref{SWE2} admits waves of two types, denoted Class 1 and Class 2 by \cite{Lamb32}.
210: Class 1 are fast high-frequency waves, the rotation modified form of
211: the usual free surface water waves, although here present only as long,
212: non-dispersive waves %(because $\delta\ll1$)
213: with speeds of order $\sqrt{gH}$.
214: Class 2 waves are slower, low-frequency waves that vanish in the absence of
215: depth change or in the absence of rotation. It is the Class 2 waves that give
216: CSWs. They have little signature in the vertical height field $\widetilde{H}(x,y,t)$ and
217: are observed through their associated horizontal velocity fields \cite{Hamon66}.
218: The Class 1 waves can be removed from \eqref{SWE1},\eqref{SWE2} by considering the
219: `rigid-lid' limit, where the external Rossby radius $\sqrt{gH}/2\Omega$ (which gives the relaxation
220: distance of the free surface) is large compared to the horizontal
221: scale of the motion. This is perhaps the most accurate of the approximations
222: noted here, causing the time-dependent term to vanish from \eqref{SWE2} and the
223: right sides of \eqref{SWE1} to become simple pressure gradients.
224:
225: For small amplitude waves the nonlinear terms in \eqref{SWE1},\eqref{SWE2} are negligible and cross-differentiating
226: gives
227: \begin{align}
228: \label{lin1}
229: \pd{\zeta}{t} + 2\Omega\bdiv\bu &= 0, \\
230: \bdiv (H\bu) &= 0.
231: \label{lin2}
232: \end{align}
233: where $\zeta=\pd{v}{x}-\pd{u}{y}$ is the vertical component of relative vorticity. Equation \eqref{lin2}
234: is satisfied by introducing the volume flux streamfunction defined though
235: \begin{equation}
236: Hu = -\pd{\psi}{y}, \qquad Hv = \pd{\psi}{x},
237: \end{equation}
238: allowing \eqref{lin1} to be written as the single equation
239: \begin{equation}
240: \bdiv \left( \frac{1}{H}\bgrad\pd{\psi}{t}\right) + 2\Omega\mathbf{k}\cdot\bgrad\psi\times\bgrad(1/H) = 0.
241: \label{twe}
242: \end{equation}
243: Equation \eqref{twe} is generally described as the
244: topographic Rossby-wave equation or the equation for barotropic CSWs. Many
245: solutions have been presented for straight coasts, where the coast lies along
246: $y=0$ (say) and the depth $H$ is a function of $y$ alone (described as rectilinear
247: topography here) \cite{LeBloM78}. These have shown
248: excellent agreement with observations of CSWs, as in \cite{Hamon66}. There has
249: been far less discussion of non-rectilinear geometries, where either the coast or the
250: depth profile or both are not functions of a single coordinate. Yet interesting
251: results appear. \cite{StockH86,StockH87} present extensive numerical integrations of a low-order spectral model
252: of a rectangular lake with idealized topography. For their chosen depth profiles normal modes can be divided into
253: two types: {\em basin-wide modes} which extend throughout the lake and localized {\em bay modes}. These bay modes
254: correspond to the high-frequency modes found in a finite-element model of the Lake of Lugarno by \cite{Trosch84} and
255: observed by \cite{StockHSTZ87}. \cite{Johns89a,StockJ89,StockJ91} give a variational formulation and describe simplified
256: quasi-analytical models that admit localized trapped bay modes. However the geometry changes
257: in these models are large, with the sloping lower boundary terminating abruptly
258: where it strikes a coastal wall. Further \cite{Johns87a} notes that \eqref{twe} is
259: invariant under conformal mappings and so any geometry that can be mapped
260: conformally to a rectilinear shelf cannot support trapped modes. The question thus
261: arises as to whether it is only for the most extreme topographic changes that
262: shelf waves can be trapped or whether trapping can occur on smoothly varying
263: shelves. The purpose of the present paper is to show that this is not so: trapped modes can exist on smoothly curving coasts.
264:
265: The geometry considered here is that of a shelf of finite width lying along an impermeable coast. Thus sufficiently
266: far from the coast the undisturbed fluid depth becomes the constant depth of the open ocean. It is shown in
267: \cite{Johns91b} that at the shelf-ocean boundary of finite-width rectilinear shelves the tangential velocity component
268: $u$ vanishes for waves sufficiently long compared to the shelf width. The wavelength of long propagating
269: disturbances is proportional to their frequency which is in turn proportional to the slope of the shelf.
270: Thus it appears that for sufficiently weakly sloping shelves the tangential velocity component,
271: i.e. the normal derivative of the streamfunction, at the shelf-ocean boundary can be made arbitrarily small.
272: Here this will be taken as also giving a close approximation to the boundary condition at the shelf-ocean boundary
273: when this boundary is no longer straight. The unapproximated boundary condition is that the streamfunction and its
274: normal derivative are continuous across the boundary where they match to the decaying solution of Laplace's equation
275: (to which \eqref{twe} reduces in regions of constant depth). This gives a linear integral condition along the boundary.
276: The unapproximated problem will not be pursued further here. The boundary condition at the coast is simply one of impermeability
277: and thus on both rectilinear and curving coasts is simply that the streamfunction vanishes. Now consider flows of the form
278: \begin{equation}
279: \psi(x,y,t)=\re \{ \Phi(x,y) \exp (-2\ii\omega\Omega t\},
280: \end{equation}
281: so $\Phi(x,y)$ gives the spatial structure of the flow and $\omega$ its non-dimensional frequency. Then $\Phi$
282: satisfies
283: \begin{align}
284: \label{prob1}
285: \frac{1}{H}\,\Delta\,\Phi + \bgrad\left(\frac{1}{H}\right)\cdot\bgrad\Phi+
286: \frac{i}{\omega}\mathbf{k}\cdot\left(\bgrad\Phi\times\bgrad\left(\frac{1}{H}\right)\right)=0,& \\
287: \label{bc1}
288: \Phi = 0 \text{ at the coast},& \\
289: \label{bc2}
290: \nhat\cdot\bgrad\Phi =0 \text{ at the shelf-ocean boundary},
291: \end{align}
292: where $\nhat$ is normal to the shelf-ocean boundary.
293:
294: Mathematically, we are going to study the existence of trapped modes (i.e., the eigenvalues either
295: embedded into the essential spectrum or lying in the gap of the essential spectrum) for the problem \eqref{prob1}--\eqref{bc2}
296: in a curved strip. Similar problems for the Laplace operator have been extensively
297: studied in the literature --- either in a curved strip, or in a straight strip with an
298: obstacle, or in a strip with compactly perturbed boundary.
299: In the case of Laplacians with Dirichlet boundary
300: conditions these problems are usually called `quantum waveguides'; the Neumann case
301: is usually referred to as `acoustic waveguides'. The important result concerning
302: quantum waveguides was established in \cite{ES}, \cite{DE}: in the curved waveguides
303: there always exist a trapped mode. Later this result was extended to more general settings;
304: in particular, in \cite{KK} it was shown that in the case of mixed boundary
305: conditions (i.e., Dirichlet conditions on one side of the strip and Neumann conditions
306: on the other side) trapped modes exist if the strip is curved `in the direction of the
307: Dirichlet boundary'.
308:
309: The case of quantum waveguides is more complicated because any eventual eigenvalues are
310: embedded into the essential spectrum and are, therefore, highly unstable. Therefore,
311: it is believed that in general the existence of trapped modes in this case
312: is due to some sort of the
313: symmetry of the problem (see \cite{ELV}, \cite{DP}, \cite{APV}).
314:
315: In the present paper we use the approach similar to the one used in \cite{DE}
316: and \cite{ELV}; however, we have to modify this approach substantially due to the
317: fact that we are working with a spectral problem for an operator pencil rather than that for
318: an ordinary operator.
319:
320: The rest of the paper is organized in the following way. In Section 2, we discuss the rigorous mathematical statement
321: of the problem; in Section 3, we study the essential spectrum, and in Section 4, state and prove the main result
322: on the existence of a discrete spectrum (Theorem 4.1). In particular we show that a trapped mode always exists of
323: all of the following conditions are satisfied: (a) the depth profile $H$ does not depend upon the longitudinal coordinate;
324: (b) the channel is curved in the direction of the Dirichlet boundary; (c) the curvature is sufficiently small.
325:
326: Similar results can be obtained in a straight strip if the depth profile $H$ depends nontrivially upon the longitudinal
327: coordinate; we however do not discuss this problem here.
328:
329: \section{Mathematical statement of the problem}
330:
331: \subsection{Geometry}
332:
333: The original geometry is a straight planar strip of width $\delta$:
334: %
335: $$
336: G_0 = \{(x,y):x\in\Rbb,\ y\in(0,\delta)\}\,.
337: $$
338:
339: Deformed geometry $G$ is assumed to be a curved planar strip of constant
340: width $\delta$. To describe it precisely, we introduce the
341: curve $\Gamma=\{(x=X(\xi),\ y=Y(\xi))\},\,\xi\in\Rbb$, where $\xi$ is a natural arc-length parameter,
342: i.e., $X'(\xi)^2+Y'(\xi)^2\equiv 1$ . By $\gamma(\xi) = X''(\xi)Y'(\xi) - X'(\xi)Y''(\xi)$ we denote a
343: (signed) curvature of $\Gamma$ (see Figure~\ref{fig:one}). Note that
344: $|\gamma(\xi)|^2 = X''(\xi)^2+Y''(\xi)^2$.
345:
346: \begin{figure}[thb!]
347: \begin{center}
348: \fbox{\resizebox{0.9\textwidth}{!}{\includegraphics*{pict}}}
349: \caption{Domain $G$ and curvilinear coordinates $\xi$, $\eta$. The solid line denotes the boundary
350: $\partial_1 G$ with the Dirichlet boundary condition, and the dotted line --- the boundary
351: $\partial_2 G$ with the Neumann boundary condition.\label{fig:one}}
352: \end{center}
353: \end{figure}
354:
355: We additionally assume
356: %
357: \begin{equation}\label{eq:gam_compact}
358: \supp\gamma\Subset[-R,R]\qquad\text{for some }R>0\,,
359: \end{equation}
360: %
361: and set
362: %
363: \begin{equation}\label{eq:gam_ext}
364: \kappa^+=\sup_{\xi\in [-R,R]}\gamma(\xi)\,,\qquad \kappa^-=-\inf_{\xi\in [-R,R]}\gamma(\xi)\,.
365: \end{equation}
366: %
367:
368: We shall assume throughout the paper the smoothness condition
369: \begin{equation}\label{eq:gam_smooth}
370: \gamma\in C^\infty(\Rbb)\,,
371: \end{equation}
372: which can be obviously softened.
373:
374:
375: Now we can introduce, in a neighbourhood of $\Gamma$,
376: the curvilinear coordinates $(\xi,\eta)$ as
377: %
378: \begin{equation}\label{eq:curv_coords}
379: x=X(\xi)-\eta Y'(\xi)\,,\qquad y=Y(\xi)+\eta X'(\xi)\,,
380: \end{equation}
381: %
382: (where $\eta$ is a distance from a point $(x,y)$ to $\Gamma$)
383: and describe the \emph{deformed strip} $G$ in these coordinates as
384: %
385: \begin{equation}\label{eq:Omega}
386: G=G_\gamma=\{(\xi,\eta):\ \xi\in\Rbb,\ \eta\in(0,\delta)\}\,.
387: \end{equation}
388: %
389:
390: \begin{rem}
391: As sets of points, $G_\gamma \equiv G_0$ for any $\gamma$, but the metrics are different, see below.
392: We shall often omit index $\gamma$ if the metric is obvious from the context.
393: \end{rem}
394:
395: To avoid self-intersections, we must restrict the width of the strip by natural conditions
396: %
397: \begin{equation}\label{eq:kappa_rest}
398: \kappa^\pm\le A\delta^{-1}\,,\qquad A = \const \in [0,1)\,.
399: \end{equation}
400: %
401:
402: Finally, it is an easy computation to show that the Euclidean metric in
403: the curvilinear coordinates has a form $dx^2+dy^2=gd\xi^2+d\eta^2$, where
404: %
405: $$
406: g(\xi,\eta) = (1+\eta\gamma(\xi))^2\,.
407: $$
408: %
409: Later on, we shall widely use the notation
410: \begin{equation}\label{eq:p}
411: p(\xi,\eta)=(g(\xi,\eta))^{1/2} = 1+\eta\gamma(\xi)\,.
412: \end{equation}
413: Note that in all the volume integrals,
414: $$
415: \drm G_\gamma=p(\xi,\eta)\,\drm\xi\,\drm\eta=(1+\eta\gamma(\xi))\,\drm\xi\,\drm\eta=p(\xi,\eta)\,\drm G_0\,.
416: $$
417:
418:
419: \subsection{Governing equations}
420:
421:
422: For a given positive continuously differentiable function $H(\xi,\eta)$ (describing a depth profile), we are looking for a
423: function $\Phi(\xi,\eta)$ satisfying \eqref{prob1} with spectral parameter $\omega$.
424:
425:
426: By substituting
427: %
428: \begin{equation}\label{eq:beta}
429: \beta(\xi,\eta):=\ln H(\xi,\eta)\,,
430: \end{equation}
431: %
432: and using explicit expressions for differential operators in curvilinear coordinates, we can re-write
433: \eqref{prob1} as
434: %
435: \begin{equation}\label{eq:govern_coords}
436: \begin{split}
437: &\omega\left(
438: -\frac{1}{p^2}\,\frac{\dpar^2\Phi}{\dpar\xi^2}
439: -\frac{\dpar^2\Phi}{\dpar\eta^2}
440: +
441: \left(\frac{1}{p^3}\,\frac{\dpar p}{\dpar\xi}+\frac{1}{p^2}\,\frac{\dpar \beta}{\dpar\xi}\right)\,\frac{\dpar \Phi}{\dpar\xi}
442: +\left(\frac{\dpar \beta}{\dpar\eta}-\frac{1}{p}\,\frac{\dpar p}{\dpar\eta}\right)\,\frac{\dpar \Phi}{\dpar\eta}
443: \right)
444: \\
445: &\qquad =\frac{i}{p}\left(
446: \frac{\dpar \beta}{\dpar\xi}\,\frac{\dpar \Phi}{\dpar\eta}
447: -\frac{\dpar \beta}{\dpar\eta}\,\frac{\dpar \Phi}{\dpar\xi}
448: \right)\,.
449: \end{split}
450: \end{equation}
451: %
452:
453: \begin{rem}\label{rem:h}
454: When deducing equation \eqref{eq:govern_coords}, we have cancelled, on both sides, a common positive
455: factor $\dsp h(\xi,\eta):=\frac{1}{H(\xi,\eta)}=\erm^{-\beta(\xi,\eta)}$. However, we have to
456: use this factor when considering corresponding variational equations, in order to keep the
457: resulting forms symmetric. This leads to a special choice of weighted Hilbert spaces below.
458: \end{rem}
459: Further on, we only consider the case of a longitudinally uniform monotone depth profile
460: %
461: \begin{equation}\label{eq:beta_eta}
462: \beta(\xi,\eta)\equiv \beta(\eta)\,,\qquad\beta'(\eta)>0\,,
463: \end{equation}
464: %
465: in which case the equation \eqref{eq:govern_coords} simplifies to
466: %
467: \begin{equation}\label{eq:govern_simply}
468: \begin{split}
469: &\omega\left(-\frac{1}{p^2}\,\frac{\dpar^2\Phi}{\dpar\xi^2}-\frac{\dpar^2\Phi}{\dpar\eta^2}
470: +\frac{1}{p^3}\,\frac{\dpar p}{\dpar\xi}\,\frac{\dpar \Phi}{\dpar\xi}
471: +\left(\beta'-\frac{1}{p}\,\frac{\dpar p}{\dpar\eta}\right)\,\frac{\dpar \Phi}{\dpar\eta}\right) \\
472: &\qquad =-\frac{i}{p}\,\beta'\,\frac{\dpar \Phi}{\dpar\xi}\,,
473: \end{split}
474: \end{equation}
475: %
476: with $\dsp \beta'=\frac{\drm\beta}{\drm\eta}$.
477:
478:
479: \subsection{Boundary conditions}
480:
481: Let $\partial_1 G=\{(\xi,0):\xi\in\Rbb\}$ and $\partial_2
482: G=\{(\xi,\delta):\xi\in\Rbb\}$ denote the lower and the upper
483: boundary of the strip $G$, respectively.
484: Boundary conditions \eqref{bc1}, \eqref{bc2} then become
485: \begin{equation}\label{eq:bc}
486: \left.\Phi\right|_{\dpar_1 G}=\left.\frac{\dpar \Phi}{\dpar\eta}\right|_{\dpar_2 G}=0\,.
487: \end{equation}
488: \begin{rem}
489: If the flow is confined to a channel then the Dirichlet boundary condition \eqref{bc1}
490: applies on both channel walls. This leads to a mathematically different problem which we do not consider in this
491: paper.
492: \end{rem}
493:
494:
495:
496: \subsection{Function spaces and rigorous operator statement}
497:
498:
499: We want to discuss the function spaces in which everything acts. Let us denote by
500: $L_2(G; h)$ the Hilbert space of functions $\phi:G\to\Cbb$ which are square-integrable on $G$
501: with the weight $\dsp h(\eta)\equiv \frac{1}{H}=\exp(-\beta(\eta))$:
502: $$
503: \|\phi\|_{L_2(G;h)}^2=\int_G |\phi(\xi,\eta)|^2\,h(\eta)\,\drm G=
504: \int_{\Rbb}\int_0^\delta |\phi(\xi,\eta)|^2\,h(\eta)p(\xi,\eta)\,\drm\eta\drm\xi\,<\,\infty\,.
505: $$
506: The corresponding inner product will be denoted $\dsp {\scal{\cdot}{\cdot}}_{L_2(G;h)}$.
507: Similarly we can define the space $L_2(F; h)$ for an arbitrary open subset $F$ of $G$.
508:
509:
510: Let us formally introduce the operators
511: $$
512: \Lcal_\gamma: \Phi\mapsto -\frac{1}{p^2}\,\frac{\dpar^2\Phi}{\dpar\xi^2}-\frac{\dpar^2\Phi}{\dpar\eta^2}
513: +\frac{1}{p^3}\,\frac{\dpar p}{\dpar\xi}\,\frac{\dpar \Phi}{\dpar\xi}
514: +\left(\beta'-\frac{1}{p}\,\frac{\dpar p}{\dpar\eta}\right)\,\frac{\dpar \Phi}{\dpar\eta}
515: $$
516: and
517: $$
518: \Mcal_\gamma: \Phi\mapsto -\frac{i}{p}\,\beta'\,\frac{\dpar \Phi}{\dpar\xi}\,.
519: $$
520: (The dependence on $\gamma$ is of course via $p$, see \eqref{eq:p}.)
521: Then \eqref{eq:govern_simply} can be formally re-written as
522: \begin{equation}\label{eq:LM}
523: \omega\Lcal_\gamma \Phi=\Mcal_\gamma\Phi\,,
524: \end{equation}
525: or via an {\it operator pencil\/}
526: \begin{equation}
527: \Acal_\gamma \equiv \Acal_\gamma(\omega) = \omega\Lcal_\gamma- \Mcal_\gamma
528: \end{equation}
529: as
530: \begin{equation}\label{eq:A}
531: \Acal_\gamma(\omega) \Phi = 0
532: \end{equation}
533: The \emph{domain} of the pencil $\Acal_\gamma$ in the $L_2$-sense is
534: naturally defined as
535: %
536: \begin{equation}\label{eq:DomA}
537: \Dom(\Acal_\gamma)=\{\Phi\in H^2(G)\,,\ \Phi\text{ satisfies (\ref{eq:bc})}\}\,,
538: \end{equation}
539: where $H^2$ denotes a standard Sobolev space.
540:
541: On the domain \eqref{eq:DomA}, $\Mcal_\gamma$ is symmetric, and $\Lcal_\gamma$ is symmetric and positive in the sense of
542: the scalar product $\dsp {\scal{\cdot}{\cdot}}_{L_2(G;h)}$, with
543: $$
544: {\scal{\Lcal_\gamma\Phi}{\Phi}}_{L_2(G;h)}=\int_{\Rbb}\int_0^\delta
545: \left(\frac{1}{p}\left|\frac{\dpar\Phi}{\dpar\xi}\right|^2+
546: p\left|\frac{\dpar\Phi}{\dpar\eta}\right|^2\right)\erm^{-\beta}\,\drm\eta\drm\xi
547: $$
548:
549: Later on, we shall use a \emph{weak} (or \emph{variational}) form of \eqref{eq:A}, and shall require
550: some other
551: function spaces described below.
552: Let $F\subseteq G$, and suppose its boundary
553: is decomposed into two disjoint parts: $\partial F = \partial_1 F \sqcup \partial_2 F$.
554: We introduce the space
555: \begin{multline*}
556: \widetilde{C}_0^\infty(F; \partial_1 F) = \{\phi\in C^\infty(F): \overline{\supp\phi}\cap\partial_1 F =
557: \emptyset,\\
558: \text{ and there exists }r>0\text{ such that }
559: \phi(\xi, \eta)=0\text{ for }(\xi,\eta)\in F,\ |\xi|\ge r\}\,,
560: \end{multline*}
561: consisting of smooth functions with compact support vanishing near $\partial_1 F$.
562: By $\widetilde{H}_0^1(F;\partial_1 F; h)$ we denote its closure with respect to the scalar product
563: $$
564: {\scal{\phi}{\psi}}_{\widetilde{H}_0^1(F,\partial_1 F;h)} = {\scal{\phi}{\psi}}_{L_2(F;h)} +
565: {\scal{\bgrad\phi}{\bgrad\psi}}_{L_2(F;h)}\,.
566: $$
567:
568: In what follows we shall study the operators $\Lcal_\gamma$, $\Mcal_\gamma$, and the pencil
569: $\Acal_\gamma$ from a variational point of view. The details are given in the next Section; here we note only that
570: from now we understand the expression $\scal{\Lcal_\gamma\Psi}{\Psi}_{L_2(G_\gamma,h)}$ as the quadratic form for the
571: operator $\Lcal_\gamma$, with the quadratic form domain $\widetilde{H}_0^1(G;\partial_1 G; h)$.
572:
573: The main purpose of this paper is to study the \emph{spectral
574: properties} of the operator pencil $\Acal_\gamma$. We recall the
575: following definitions.
576:
577: A number $\omega\in\Cbb$ is said to belong to the \emph{spectrum} of $\Acal_\gamma$
578: (denoted $\spec(\Acal_\gamma)$) if
579: $\Acal_\gamma(\omega)$ is not invertible.
580:
581: It is easily seen that in our case the spectrum of $\Acal_\gamma$ is real.
582:
583: We say that $\omega\in\Rbb$ belongs to the \emph{essential spectrum} of the operator pencil
584: $\Acal_\gamma$ (denoted $\omega\in\spe(\Acal_\gamma)$) if for this $\omega$ the operator $\Acal_\gamma(\omega)$
585: is non-Fredholm.
586:
587: We say that $\omega\in\Cbb$ belongs to the {\it point spectrum\/} of the operator pencil
588: $\Acal_\gamma$ (denoted $\omega\in\spp(\Acal_\gamma)$),
589: or, in other words, say that $\omega$ is an {\it eigenvalue\/}, if for this $\omega$ there exists a
590: non-trivial solution $\Psi\in \Dom(\Acal_\gamma)$ of the problem $\Acal_\gamma(\omega)\Psi=0$.
591:
592: It is known that the essential spectrum is a closed subset of $\Rbb$, and that any point of the spectrum
593: outside the essential spectrum is an isolated eigenvalue of finite multiplicity. The set of all such
594: points is called the \emph{discrete spectrum}, and will be denoted $\spd(\Acal_\gamma)$
595: There may, however, exist the points of the spectrum which belong
596: both to the essential spectrum and the point spectrum.
597:
598: Our main result (Theorem \ref{thm:main} below) establishes some conditions on the curvature $\gamma$ of the waveguide which
599: guarantee the existence of eigenvalues of $\Acal_\gamma$.
600:
601: It is more convenient to deal with problems of this type variationally, and we start the next Section with
602: an abstract variational scheme suitable for self-adjoint pencils with non-empty essential spectrum.
603:
604:
605: \section{Essential spectrum}
606:
607: \subsection{Variational principle for the essential spectrum}
608:
609: \begin{defn} We set, for $j\in\Nbb$,
610: \begin{equation}\label{eq:mu_j}
611: \mu_{\gamma,j} = \sup_{\substack{U\subset \widetilde{H}_0^1(G;\partial_1 G; h)\\\dim U = j}}
612: \,\inf_{\substack{\Psi\in U,\,\Psi\ne 0}}
613: \frac{\dsp\scal{\Mcal_\gamma\Psi}{\Psi}_{L_2(G_\gamma,h)}}{\dsp\scal{\Lcal_\gamma\Psi}{\Psi}_{L_2(G_\gamma,h)}}
614: \end{equation}
615: \end{defn}
616:
617: As $\dsp\scal{\Lcal_\gamma\Psi}{\Psi}_{L_2(G_\gamma,h)}$ is positive, the right-hand sides of \eqref{eq:mu_j} are well defined,
618: though the numbers $\mu_{\gamma,j}$ may \emph{a priori} be finite or infinite.
619:
620: Obviously, for any fixed curvature profile $\gamma$ the numbers $\mu_{\gamma,j}$ form a
621: non-increasing sequence:
622: $$
623: \mu_{\gamma,1}\ge\mu_{\gamma,2}\ge\dots\ge\mu_{\gamma,j}\ge\mu_{\gamma,j+1}\ge\dots\,.
624: $$
625:
626: \begin{defn} Denote
627: \begin{equation}\label{eq:muu}
628: \muu_\gamma = \lim_{j\to\infty}\mu_{\gamma,j}\,.
629: \end{equation}
630: \end{defn}
631:
632: For general self-adjoint operator pencils the analog of \eqref{eq:muu} may be finite or equal to $\pm\infty$; as we shall
633: see below,in our case $\muu_\gamma$ is finite
634:
635:
636: The following result is a modification, to the case of an abstract self-adjoint linear pencil, of the
637: general variational principle for a self-adjoint operator with an essential spectrum, see
638: \cite[Prop. 4.5.2]{EBD}.
639:
640: \begin{prop}\label{prop:EBD}
641: Either
642: \begin{enumerate}
643: \item[(i)] $\muu_\gamma>-\infty$, and then
644: $\sup\spe(\Acal_\gamma)=\muu_\gamma$,
645: \end{enumerate}
646: or
647: \begin{enumerate}
648: \item[(ii)] $\muu_\gamma=-\infty$,
649: and then $\spe(\Acal_\gamma)=\emptyset$.
650: \end{enumerate}
651: Moreover, if $\mu_{\gamma,j}>\muu_\gamma$, then $\mu_{\gamma,j}\in\spd(\Acal_\gamma)$.
652: \end{prop}
653:
654: Proposition \ref{prop:EBD} ensures that we can use the variational principle \eqref{eq:mu_j} in order
655: to find the eigenvalues of the pencil $\Acal_\gamma$ lying \emph{above} the supremum $\muu_\gamma$ of the
656: essential spectrum.
657:
658: \subsection{Essential spectrum for the straight strip}
659:
660: The spectral analysis in the case of a straight strip ($\gamma\equiv 0$) is rather straightforward
661: as the problem admits in this case the separation of variables.
662:
663: Let us seek the solutions of \eqref{eq:govern_simply},
664: \eqref{eq:bc} in the case of a straight strip ($\gamma\equiv 0$, and so $p\equiv 1$)
665: in the form
666: \begin{equation}\label{eq:phi_def}
667: \Phi(\xi,\eta)=\phi(\eta)\exp(i\alpha\xi)\,;
668: \end{equation}
669: it is sufficient to consider only real values of $\alpha$.
670:
671:
672: After separation of variables, \eqref{eq:govern_simply}, \eqref{eq:bc} are
673: written, for each $\alpha$, as a one-dimensional transversal spectral problem
674: \begin{equation}\label{eq:LM_1d}
675: \omega(-\phi''+\beta'\phi'+\alpha^2\phi) =\alpha\beta'\phi\,,
676: \qquad \phi(0)=\phi'(\delta)=0\,,
677: \end{equation}
678: Alternatively, introduce operators
679: $$
680: \mathfrak{l}_\alpha : \phi\mapsto -\phi''+\beta'\phi'+\alpha^2\phi\,,\qquad
681: \mathfrak{m}_\alpha : \phi\mapsto\alpha\beta'\phi\,,
682: $$
683: and a pencil
684: $$
685: \mathfrak{a}_\alpha(\omega) = \omega\mathfrak{l}_\alpha-\mathfrak{m}_\alpha\,,
686: $$
687: (again understood in an $L_2((0,\delta);h)$ sense with the domain defined similarly to \eqref{eq:DomA}),
688: and consider a one-dimensional operator pencil spectral problem $\mathfrak{a}_\alpha(\omega)\phi = 0$.
689:
690: For a fixed value of $\alpha$, the one-dimensional linear operator
691: pencil \eqref{eq:LM_1d} has the essential spectrum $\{0\}$ and a discrete spectrum $\spec(\mathfrak{a}_\alpha)$; note
692: that $\spec(\mathfrak{a}_{-\alpha})=-\spec(\mathfrak{a}_\alpha)$.
693: Denote, for $\alpha>0$, the top of the spectrum of this transversal problem
694: by $\omega_\alpha = \sup\spec(\mathfrak{a}_\alpha)$.
695:
696: \begin{lem}\label{lem:1d_sp} Let $\alpha>0$. Then, under condition \eqref{eq:beta_eta},
697: \begin{enumerate}
698: \item[(i)] $\spec(\mathfrak{a}_\alpha)\subset[0, +\infty)$;
699: \item[(ii)] $0<\omega_\alpha<+\infty$;
700: \item[(iii)] $\omega_\alpha\to +0$ as $\alpha\to\infty$.
701: \end{enumerate}
702: \end{lem}
703:
704: \begin{proof} By the variational principle analogous to Proposition~\ref{prop:EBD}(i),
705: $$
706: \omega_\alpha = \sup_{\substack{\phi\in \widetilde H^1_0((0,\delta),0,h)\\\phi\ne0}} J_\alpha(\phi)\,,
707: $$
708: where we set
709: \begin{equation}\label{eq:vp_1d}
710: J_\alpha(\phi) =
711: \frac{\dsp\scal{\mathfrak{m}_\alpha\phi}{\phi}_{L_2((0,\delta),h)}}
712: {\dsp\scal{\mathfrak{l}_\alpha\phi}{\phi}_{L_2((0,\delta),h)}}
713: = \frac
714: {\dsp \int_0^\delta \alpha \beta'(\eta)\phi(\eta)^2
715: h(\eta)\,\drm\eta}{\dsp \int_0^\delta
716: (-\phi''(\eta)+\beta'(\eta)\phi'(\eta)+\alpha^2\phi(\eta))\phi(\eta)h(\eta)\,\drm\eta}
717: \end{equation}
718:
719: After integrating by parts using $\dsp h(\eta)=\erm^{-\beta(\eta)}$, and inverting the quotient,
720: we get
721: \begin{equation}\label{eq:Jalp}
722: J_\alpha(\phi) = \left(\alpha J^{(1)}(\phi)+\frac{1}{\alpha} J^{(2)}(\phi)\right)^{-1}\,,
723: \end{equation}
724: where we denote
725: $$
726: J^{(1)}(\phi) = \frac{\dsp\int_0^\delta \erm^{-\beta(\eta)}|\phi(\eta)|^2\,\drm\eta}{\dsp\int_0^\delta
727: \beta'(\eta)\erm^{-\beta(\eta)}|\phi(\eta)|^2\,\drm\eta}
728: $$
729: and
730: $$
731: J^{(2)}(\phi) = \frac{\dsp\int_0^\delta\erm^{-\beta(\eta)}|\phi'(\eta)|^2\,\drm\eta}{\dsp\int_0^\delta
732: \beta'(\eta)\erm^{-\beta(\eta)}|\phi(\eta)|^2\,\drm\eta}
733: $$
734: The statements (ii) and (iii) of the Lemma now follow immediately from the estimates
735: $$
736: J^{(1)}(\phi)\ge
737: \frac{\dsp \inf_{\eta\in(0,\delta)}\erm^{-\beta(\eta)}}{\dsp \sup_{\eta\in(0,\delta)}(\beta'(\eta)\erm^{-\beta(\eta)})}
738: $$
739: and
740: $$
741: J^{(2)}(\phi)\ge\frac{\pi^2}{4\delta^2}\,
742: \frac{\dsp \inf_{\eta\in(0,\delta)}\erm^{-\beta(\eta)}}{\dsp
743: \sup_{\eta\in(0,\delta)}(\beta'(\eta)\erm^{-\beta(\eta)})}\,,
744: $$
745: where the latter inequality uses the variational principle and the fact that the principle eigenvalue of the
746: mixed Dirichlet--Neumann problem spectral problem for the operator $-\dsp\frac{\drm^2}{\drm \eta^2}$
747: on the interval $(0,\delta)$ is equal to $\dsp \frac{\pi^2}{4\delta^2}$. The statement (i) follows
748: from the positivity of the right-hand side of \eqref{eq:Jalp}.
749: \end{proof}
750:
751:
752: We are now able to find the essential spectrum of the problem in a
753: straight strip.
754:
755:
756: \begin{lem}\label{lem:sp_str}
757: Assume that conditions \eqref{eq:beta_eta} hold. Then
758: \begin{equation}\label{eq:speB}
759: \spe(\Acal_0)=[-\Omega_*,\Omega_*]\,,
760: \end{equation}
761: where
762: \begin{equation}\label{eq:Omstar}
763: \Omega_* = \sup_{\phi\in \widetilde H^1_0((0,\delta),0,h)}
764: \frac{\dsp\frac{1}{2}\int_0^\delta
765: \beta'(\eta)\erm^{-\beta(\eta)}|\phi(\eta)|^2\,\drm\eta}
766: {\sqrt{\dsp\int_0^\delta\erm^{-\beta(\eta)}|\phi'(\eta)|^2\,\drm\eta\cdot
767: \dsp\int_0^\delta \erm^{-\beta(\eta)}|\phi(\eta)|^2\,\drm\eta}}>0\,.
768: \end{equation}
769: \end{lem}
770:
771: \begin{proof} It is standard that
772: $$
773: \spe(\Acal_0)=\overline{\bigcup_{\alpha\in\Rbb} \spec(\mathfrak{a}_\alpha)}\,.
774: $$
775: Thus, by Lemma \ref{lem:1d_sp}, and with account of the
776: anti-symmetry of the spectrum of $\mathfrak{a}_\alpha$ with
777: respect to $\alpha$ and its positivity for $\alpha>0$, we have
778: $$
779: \sup\spe(\Acal_0)=\sup_{\alpha>0}\omega_\alpha=
780: \sup_{\alpha>0}\sup_{\phi\in \widetilde H^1_0((0,\delta),0,h)} J_\alpha(\phi)\,.
781: $$
782:
783: By maximizing first with respect to $\alpha$, we obtain, from \eqref{eq:Jalp},
784: $$
785: J_\alpha(\phi)\le J_{\alpha_*(\phi)}(\phi)
786: $$
787: with the maximizer
788: $$
789: \alpha_*(\phi) = \sqrt{\frac{J^{(2)}(\phi)}{J^{(1)}(\phi)}}\,.
790: $$
791:
792:
793: Maximizing now with respect to $\phi$ gives
794: $\sup\spe(\Acal_0)=\Omega_*$, with $\Omega_*$ given by \eqref{eq:Omstar}.
795: Finally, the statement follows from the fact that $\omega_\alpha$ depends continuously on $\alpha>0$ and
796: thus takes all the values in $(0, \Omega_*]$ by Lemma~\ref{lem:1d_sp}.
797: \end{proof}
798:
799:
800: \subsection{Essential spectrum for a curved strip}
801:
802:
803: It is now a standard procedure to show that under our conditions the essential spectrum of the problem in a
804: curved strip coincides with the essential spectrum of the problem in a straight strip given by
805: Lemma~\ref{lem:sp_str}. Namely, we have the following
806:
807: \begin{lem}\label{lem:sp_cur}
808: Let us assume the conditions \eqref{eq:gam_compact}, \eqref{eq:gam_smooth} and \eqref{eq:beta_eta} hold.
809: Then
810: $$
811: \spe(\Acal_\gamma)=\spe(\Acal_0)=[-\Omega_*,\Omega_*]\,
812: $$
813: with $\Omega_*$ given by \eqref{eq:Omstar}.
814: \end{lem}
815:
816: The proof is based on the fact that any solution of the problem \eqref{eq:govern_simply}, \eqref{eq:bc} with $\gamma\not\equiv 0$
817: (and thus $p\not\equiv 1$) should coincide in
818: $G\cap\{|\xi|>R>\max(|\inf\supp\gamma|,|\sup\supp\gamma|)\}$ with a
819: solution of the
820: same problem for $\gamma\equiv0$.
821: An analogous result has been proved in a number of similar situations elsewhere, see e.g. \cite{ES, ELV, DP}, so
822: we omit the details of the proof.
823: We briefly note that the inclusion $\spe(\Acal_\gamma)\subseteq\spe(\Acal_0)$ is proved using the separation of variables
824: as above and a construction of appropriate Weyl's sequences, and in order to prove the inclusion
825: $\spe(\Acal_\gamma)\supseteq\spe(\Acal_0)$ one can use the Dirichlet-Neumann bracketing and the discreteness of the spectrum of the problem
826: \eqref{eq:govern_simply}, \eqref{eq:bc} considered in $G\cap\{|\xi|<R\}$ with additional Dirichlet or Neumann boundary conditions imposed on the ``cuts'' $\{\xi=\pm R\}$.
827:
828:
829: \section{Main result}
830:
831: Our main result consists in stating some sufficient conditions on the depth profile $\beta(\eta)$ and
832: the curvature profile $\gamma(\xi)$ which guarantee the existence of an eigenvalue of the pencil $\Acal_\gamma$ lying
833: outside the essential spectrum.
834:
835: \begin{thm}\label{thm:main}
836: Assume, as before, that the condition \eqref{eq:beta_eta} holds.
837: Assume additionally that
838: \begin{equation}\label{eq:beta_2diff}
839: \beta''(\eta)<0\qquad\text{for }\eta\in(0,\delta)\,.
840: \end{equation}
841: Then there exists a constant $C_\beta>0$, which depends only on the depth profile $\beta$, such that
842: $\spd(\Acal_\gamma)\ne\emptyset$ whenever $\gamma$ satisfies conditions
843: \eqref{eq:gam_compact}, \eqref{eq:gam_smooth} and
844: %%
845: \begin{equation}\label{eq:main_cond}
846: \int \gamma(\xi)\drm\xi > C_\beta \int \gamma(\xi)^2\drm\xi
847: \end{equation}
848: \end{thm}
849:
850: We give an explicit expression for $C_\beta$ below, see \eqref{eq:C_beta}.
851:
852: An integral sufficient condition \eqref{eq:main_cond} may be
853: replaced by a pointwise, although more restrictive, condition:
854:
855: \begin{cor}\label{cor:main}
856: Assume that the conditions \eqref{eq:beta_eta} and \eqref{eq:beta_2diff}
857: hold. Then there exists a constant $\dsp c_{\beta, R}=\frac{C_\beta}{2R}$ which depends only on the
858: depth profile $\beta$ and a given $R>0$
859: such that $\spd(\Acal_\gamma)\ne\emptyset$ whenever $\gamma\not\equiv 0$ satisfies conditions
860: \eqref{eq:gam_compact}, \eqref{eq:gam_smooth} and
861: %%
862: \begin{equation}\label{eq:cor_cond}
863: 0\le\gamma(\xi)<c_{\beta, R}\qquad\text{ for }|\xi|\le R\,.
864: \end{equation}
865: \end{cor}
866:
867: We prove Theorem~\ref{thm:main} using a number of simple lemmas, the central of which
868: is the following
869: \begin{lem}\label{lem:test_fn}
870: Suppose there exists a function $\widetilde\Psi\in \widetilde H^1_0(G_\gamma,h)$ such that
871: \begin{equation}\label{eq:main_ineq}
872: \frac{\dsp\scal{\Mcal_\gamma\widetilde\Psi}{\widetilde\Psi}_{L_2(G_\gamma,h)}}
873: {\dsp\scal{\Lcal_\gamma\widetilde\Psi}{\widetilde\Psi}_{L_2(G_\gamma,h)}} > \Omega_*\,.
874: \end{equation}
875: Then there exists $\omega>\Omega_*$ which belongs to $\spd(\Acal_\gamma)$.
876: \end{lem}
877:
878: Lemma \ref{lem:test_fn} is just a re-statement of the variational principle of Proposition~\ref{prop:EBD}.
879: The main difficulty in its application is of course the choice of an appropriate test function $\widetilde\Psi$.
880: However such choice becomes much easier if we use the following modification of this Lemma which allows us
881: to consider test functions which are not necessarily square integrable on $G_\gamma$.
882:
883: Denote, for brevity, $G^r_\gamma=G_\gamma\cap\{|\xi|<r\}$.
884:
885: \begin{lem}\label{lem:test_fn_mod}
886: Suppose there exists a function $\Psi$ and a constant $D$ such that, for any $r>R$, we have
887: $\Psi \in \widetilde H^1_0(G^r_\gamma,h)$ and
888: \begin{equation}\label{eq:main_ineq_mod}
889: \scal{\Mcal_\gamma\Psi}{\Psi}_{L_2(G^r_\gamma,h)}- \Omega_*\scal{\Lcal_\gamma\Psi}{\Psi}_{L_2(G^r_\gamma,h)} \ge D > 0\,.
890: \end{equation}
891: Then there exists $\omega>\Omega_*$ which belongs to $\spd(\Acal_\gamma)$.
892: \end{lem}
893:
894: The proof of Lemma~\ref{lem:test_fn_mod} uses the construction of an appropriate cut-off function $\chi(\xi)$
895: such that $\widetilde\Psi=\chi\Psi$ satisfies the conditions of Lemma~\ref{lem:test_fn}, cf. \cite[Prop.~1]{DP}.
896:
897: We now proceed as follows.
898:
899: Let $\phi_*(\eta)$ be a minimizer in \eqref{eq:Omstar}, and set
900: $$
901: \Psi(\xi,\eta) = \phi_*(\eta) \erm^{i\alpha_{\bullet} \xi}\,,
902: $$
903: where
904: \begin{equation}\label{eq:alpha_crit}
905: \alpha_{\bullet}=\alpha_*(\phi_*) = \sqrt{\frac{J^{(2)}(\phi_*)}{J^{(1)}(\phi_*)}}\,.
906: \end{equation}
907:
908: It is important to note that $\Psi$ is in fact an ``eigenfunction'' of the essential spectrum
909: of $\Acal_\gamma$ corresponding to its highest positive point $\Omega_*$ and that $\phi_*$ is an
910: eigenfunction of \eqref{eq:LM_1d} with $\alpha=\alpha_{\bullet}$
911: (i.e. of the pencil $\mathfrak{a}_{\alpha_{\bullet}}$) again corresponding to the eigenvalue $\Omega_*$, and so
912: \begin{equation}\label{eq:phi_star_eq}
913: \phi''_* = \beta'\phi'_*+(\alpha_{\bullet}^2-\Lambda_*\alpha_{\bullet}\beta')\phi_*\,,\qquad
914: \phi_*(0)=\phi'_*(\delta)=0
915: \end{equation}
916: with $\dsp \Lambda_*:=\frac{1}{\Omega_*}$
917: (cf. \eqref{eq:LM_1d}).
918:
919: For future use, we summarize the relations obtained so far:
920: \begin{align*}
921: \Lcal_0\Psi &= (-\phi_*''(\eta)+\beta'(\eta)\phi_*'(\eta)+\alpha_{\bullet}\phi_*(\eta))\,\erm^{i\alpha_{\bullet} \xi} =
922: (\mathfrak{l}_{\alpha_{\bullet}}\phi_*)\,\erm^{i\alpha_{\bullet} \xi}\,,
923: \\
924: \Mcal_0\Psi &= \alpha_{\bullet}\beta'(\eta)\phi_*(\eta)\,\erm^{i\alpha_{\bullet} \xi}
925: = (\mathfrak{m}_{\alpha_{\bullet}}\phi_*)\,\erm^{i\alpha_{\bullet} \xi}\,,
926: \\
927: \Lcal_\gamma\Psi &= \left(-\phi_*''(\eta)
928: +\left(\beta'(\eta)-\frac{1}{p(\xi,\eta)}\,\frac{\dpar p(\xi,\eta)}{\dpar\eta}\right)\phi_*'(\eta)
929: \right.
930: \\
931: &\qquad\qquad\left.+\left(\frac{i\alpha_{\bullet}}{p(\xi,\eta)^3}\,\frac{\dpar p(\xi,\eta)}{\dpar\xi}+
932: \frac{\alpha_{\bullet}^2}{p(\xi,\eta)^2}\right)\,\phi_*(\eta)
933: \right)\,\erm^{i\alpha_{\bullet} \xi} \,,
934: \\
935: \Mcal_\gamma\Psi &= \frac{\alpha_{\bullet}}{p(\xi,\eta)}\beta'(\eta)\phi_*(\eta)\,\erm^{i\alpha_{\bullet} \xi} \,,
936: \\
937: p(\xi,\eta)&=1+\gamma(\xi)\eta\,,
938: \end{align*}
939: (with ${}'$ denoting differentiation with respect to $\eta$).
940:
941: It is important to note that for any $r>0$
942: $$
943: \frac{\dsp\scal{\Mcal_0\Psi}{\Psi}_{L_2(G^r_0,h)}}{\dsp\scal{\Lcal_0\Psi}{\Psi}_{L_2(G^r_0,h)}}
944: =\frac{\dsp\scal{\mathfrak{m}_{\alpha_{\bullet}}\phi_*}{\phi_*}_{L_2((0,\delta),h)}}
945: {\dsp\scal{\mathfrak{l}_{\alpha_{\bullet}}\phi_*}{\phi_*}_{L_2((0,\delta),h)}}%
946: =\Omega_*>0
947: $$
948: and, as explicit formulae above show,
949: \begin{equation}\label{eq:m_equal}
950: \begin{split}
951: \scal{\Mcal_\gamma\Psi}{\Psi}_{L_2(G^r_\gamma,h)}&=\scal{\Mcal_0\Psi}{\Psi}_{L_2(G^r_0,h)}
952: =\alpha_{\bullet}\int_{-r}^r\int_0^\delta \beta'(\eta) \erm^{-\beta(\eta)} |\phi_*(\eta)|^2\,\drm\eta\drm\xi\\
953: &= 2r\alpha_{\bullet}\int_0^\delta \beta'(\eta) \erm^{-\beta(\eta)} |\phi_*(\eta)|^2\,\drm\eta>0\,.
954: \end{split}
955: \end{equation}
956:
957: We want to show that under conditions of Theorem~\ref{thm:main}
958: and with the choice of $\Psi$ as above, the inequality
959: \eqref{eq:main_ineq_mod} holds for any $r>R$.
960:
961: In view of \eqref{eq:m_equal}, it is enough to show that
962: $$
963: D_\gamma:=\scal{\Lcal_\gamma\Psi}{\Psi}_{L_2(G^r_\gamma,h)}-\scal{\Lcal_0\Psi}{\Psi}_{L_2(G^r_0,h)}
964: $$
965: is \emph{negative} for $r>R$.
966:
967: Explicit substitution gives, after taking into account the formula
968: $$
969: \int \frac{1}{p(\xi,\eta)^2}\,\frac{\dpar p(\xi,\eta)}{\dpar\xi}\,\drm\xi = 0
970: $$
971: (due to \eqref{eq:gam_compact}, with account of \eqref{eq:p}), the
972: following expression:
973: $$
974: D_\gamma = \int_{-r}^r\int_0^\delta \gamma(\xi)\eta\erm^{-\beta(\eta)}|\phi'_*(\eta)|^2\,\drm\eta\drm\xi
975: -\int_{-r}^r\int_0^\delta \alpha_{\bullet}^2 \frac{\eta\gamma(\xi)}{1+\eta\gamma(\xi)}
976: \erm^{-\beta(\eta)}|\phi_*(\eta)|^2\,\drm\eta\drm\xi\,.
977: $$
978: This, in turn, can be re-written, using the obvious identity
979: $$
980: \frac{\eta\gamma(\xi)}{1+\eta\gamma(\xi)} = \eta\gamma(\xi) - \frac{\eta^2\gamma(\xi)^2}{1+\eta\gamma(\xi)}
981: $$
982: as
983: \begin{equation}\label{eq:D_gamma}
984: \begin{split}
985: D_\gamma &= \int_{-r}^r\gamma(\xi)\int_0^\delta
986: \eta\erm^{-\beta(\eta)}
987: \left(|\phi'_*(\eta)|^2-\alpha_{\bullet}^2|\phi_*(\eta)|^2\right)\,\drm\eta\drm\xi
988: \\
989: &+\alpha_{\bullet}^2\int_{-r}^r\int_0^\delta \frac{\eta^2\gamma(\xi)^2}{1+\eta\gamma(\xi)}
990: \erm^{-\beta(\eta)}|\phi_*(\eta)|^2\,\drm\eta\drm\xi\,.
991: \end{split}
992: \end{equation}
993:
994: We shall deal with the two terms in \eqref{eq:D_gamma} separately.
995:
996: The first one is more difficult. As \eqref{eq:alpha_crit} yields explicitly
997: $$
998: \alpha_{\bullet}^2=\frac{\dsp \int_0^\delta \erm^{-\beta(\eta)}|\phi'_*(\eta)|^2\,\drm\eta}%
999: {\dsp \int_0^\delta \erm^{-\beta(\eta)}|\phi_*(\eta)|^2\,\drm\eta}\,,
1000: $$
1001: we get
1002: \begin{equation}\label{eq:I1}
1003: \begin{split}
1004: I_1&:=\int_0^\delta\eta\erm^{-\beta(\eta)}
1005: \left(|\phi'_*(\eta)|^2-\alpha_{\bullet}^2|\phi_*(\eta)|^2\right)\,\drm\eta
1006: \\
1007: &=\frac{1}{\dsp \int_0^\delta \erm^{-\beta(\eta)}|\phi_*(\eta)|^2\,\drm\eta}
1008: \times\left(\int_0^\delta\eta\erm^{-\beta(\eta)}|\phi'_*(\eta)|^2\,\drm\eta
1009: \cdot
1010: \int_0^\delta\erm^{-\beta(\eta)}|\phi_*(\eta)|^2\,\drm\eta
1011: \right.\\
1012: &\qquad\qquad\qquad\left.-
1013: \int_0^\delta\erm^{-\beta(\eta)}|\phi'_*(\eta)|^2\,\drm\eta
1014: \cdot
1015: \int_0^\delta\eta\erm^{-\beta(\eta)}|\phi_*(\eta)|^2\,\drm\eta\right)
1016: \,.
1017: \end{split}
1018: \end{equation}
1019:
1020: We want to show that the term in brackets is negative under some reasonable assumptions:
1021:
1022: \begin{lem}\label{lem:I1}
1023: Assume that the conditions of theorem~\ref{thm:main} hold. Then
1024: $I_1<0$.
1025: \end{lem}
1026:
1027: The proof of Lemma~\ref{lem:I1} uses the following
1028: simple fact:\footnote{we are grateful to Daniel Elton for a useful suggestion helping to prove this Lemma.}
1029:
1030: \begin{lem}\label{lem:Daniel}
1031: Let $(a,b)\subset(0,+\infty)$ be a finite interval, and let a
1032: function $g:(a,b)\to\Rbb$ be non-negative and non-increasing.
1033: Then
1034: $$
1035: \left(\int_a^b x g(x)f(x)\,\drm x\right)\cdot\left(\int_a^b f(x)\,\drm x\right) -
1036: \left(\int_a^b g(x)f(x)\,\drm x\right)\cdot\left(\int_a^b xf(x)\,\drm x\right)\le 0
1037: $$
1038: for any non-negative function $f:(a,b)\to\Rbb$.
1039: \end{lem}
1040:
1041: \begin{proof}[Proof of Lemma~\ref{lem:Daniel}]
1042: We have
1043: \begin{align*}
1044: &\left(\int_a^b x g(x)f(x)\,\drm x\right)\cdot\left(\int_a^b f(x)\,\drm x\right) -
1045: \left(\int_a^b g(x)f(x)\,\drm x\right)\cdot\left(\int_a^b xf(x)\,\drm x\right)\\
1046: &\qquad=
1047: \int_a^b\int_a^b x g(x)f(x) f(y)\,\drm x\drm y - \int_a^b\int_a^b g(x)f(x)yf(y)\,\drm x\drm y\\
1048: &\qquad=
1049: \int_a^b\int_a^y (x-y) f(x) f(y) g(x)\,\drm x\drm y + \int_a^b\int_y^b (x-y) f(x) f(y) g(x)\,\drm x\drm y
1050: \end{align*}
1051: Interchanging the variables $x$ and $y$ in the last integral, we obtain that the whole expression is equal to
1052: $$
1053: \int_a^b\int_a^y \underbrace{(x-y)}_{\text{non-positive}} \underbrace{f(x) f(y)(g(x)-g(y))}_{\text{non-negative}}\,\drm x\drm y\,,
1054: $$
1055: and is therefore non-positive.
1056: \end{proof}
1057:
1058: We can now proceed with evaluating $I_1$.
1059:
1060: \begin{proof}[Proof of Lemma~\ref{lem:I1}] We act by doing a lot of integrations by part.
1061: We shall also use \eqref{eq:phi_star_eq}.
1062:
1063:
1064: We have (all integrals are over $[0,\delta]$ and with respect to $\eta$)
1065: \begin{equation*}
1066: \begin{split}
1067: \int \eta\erm^{-\beta}|\phi'_*|^2 &= - \int \phi_*\cdot(\eta \erm^{-\beta} \phi'_*)'\\
1068: &= -\int \phi_*\cdot(\erm^{-\beta} \phi'_*-\beta'\eta\erm^{-\beta}\phi'_*+\eta\erm^{-\beta} \phi''_*)\\
1069: &=-\int \phi_*\cdot(\erm^{-\beta} \phi'_*+(\alpha_{\bullet}^2-\Lambda_*\alpha_{\bullet}\beta')\erm^{-\beta}\eta\phi_*)\,.
1070: \end{split}
1071: \end{equation*}
1072: Further,
1073: \begin{equation*}
1074: -\int (\phi_*\erm^{-\beta})\phi'_*=\left(\int
1075: \left(\phi'_*\erm^{-\beta}-\beta'\phi_*\erm^{-\beta}\right)\phi_*\right)-\erm^{-\beta(\delta)}\phi_*^2(\delta)\,,
1076: \end{equation*}
1077: thus producing
1078: \begin{equation}\label{eq:int1}
1079: \begin{split}
1080: \int \eta\erm^{-\beta}|\phi'_*|^2 &= -\frac{1}{2} \int \beta'\phi_*^2\erm^{-\beta}
1081: \underbrace{-\frac{1}{2}\erm^{-\beta(\delta)}\phi_*^2(\delta)}_\text{negative constant}\\
1082: & -\alpha_{\bullet}^2\int\eta\erm^{-\beta}|\phi_*|^2
1083: +\Lambda_*\alpha_{\bullet}\int\eta\beta'\erm^{-\beta}|\phi_*|^2
1084: \end{split}
1085: \end{equation}
1086:
1087: Also,
1088: \begin{equation}\label{eq:int2}
1089: \begin{split}
1090: \int \erm^{-\beta}|\phi'_*|^2 &= -\int \erm^{-\beta}\phi_*(-\beta'\phi'_*+\phi''_*)\\
1091: &= -\int
1092: \erm^{-\beta}\phi_*\left(-\beta'\phi_*'+\beta'\phi_*'+\alpha_{\bullet}^2\phi_*-\Lambda_*\alpha_{\bullet}\beta'\phi_*\right)\\
1093: &=
1094: -\int \erm^{-\beta}\phi_*^2\,(\alpha_{\bullet}^2-\Lambda_*\alpha_{\bullet}\beta')\,.
1095: \end{split}
1096: \end{equation}
1097:
1098: Substituting \eqref{eq:int1} and \eqref{eq:int2} into \eqref{eq:I1}, and simplifying, we get
1099: \begin{equation}\label{eq:I1_1}
1100: \begin{split}
1101: &I_1 \cdot \underbrace{\dsp\erm^{-\beta(\eta)}|\phi_*(\eta)|^2}_\text{positive integral}
1102: =
1103: \int \eta\erm^{-\beta}|\phi'_*|^2 \cdot \int \erm^{-\beta}|\phi_*|^2
1104: - \int \erm^{-\beta}|\phi'_*|^2 \cdot \int \eta\erm^{-\beta}|\phi_*|^2\\
1105: &\qquad=
1106: \left(\underbrace{\left(-\frac{1}{2} \int \beta'\erm^{-\beta}|\phi_*|^2\right)}_{\text{negative as }\beta'>0}+
1107: \underbrace{\left(-\frac{1}{2}\erm^{-\beta(\delta)}\phi_*^2(\delta)\right)}_{\text{negative constant}}\right)
1108: \int \erm^{-\beta}|\phi_*|^2\\
1109: &\qquad+\underbrace{\left(\Lambda_*\alpha_{\bullet}\right)}_{\text{$+$ve constant}}
1110: \times
1111: \underbrace{\left(\int \eta\beta'\erm^{-\beta}|\phi_*|^2\cdot \int \erm^{-\beta}|\phi_*|^2-
1112: \int \beta'\erm^{-\beta}|\phi_*|^2\cdot \int \eta\erm^{-\beta}|\phi_*|^2\right)}_{\text{non-positive by Lemma~\ref{lem:Daniel} with }
1113: g=\beta',\ f=\erm^{-\beta}|\phi_*|^2\text{ as }g'=\beta''\le 0}
1114: \end{split}
1115: \end{equation}
1116: Thus $I_1<0$.
1117: \end{proof}
1118:
1119:
1120: Let us now return to \eqref{eq:D_gamma} and deal with the second term in the right-hand side. We have,
1121: with account of \eqref{eq:gam_ext} and \eqref{eq:kappa_rest},
1122: $$
1123: \frac{\eta^2\gamma(\xi)^2}{1+\eta\gamma(\xi)}\le
1124: \begin{cases}
1125: \eta^2\gamma(\xi)^2\,,\qquad&\text{if }\gamma(\xi)\ge 0\\
1126: \dsp\frac{1}{1-A}\eta^2\gamma(\xi)^2\,,\qquad&\text{if }\gamma(\xi)< 0
1127: \end{cases}\qquad\le\max\{1,\dsp\frac{1}{1-A}\}\eta^2\gamma(\xi)^2\,,
1128: $$
1129: and so
1130: $$
1131: \alpha_{\bullet}^2\int_{-r}^r\int_0^\delta \frac{\eta^2\gamma(\xi)^2}{1+\eta\gamma(\xi)}
1132: \erm^{-\beta(\eta)}|\phi_*(\eta)|^2\,\drm\eta\drm\xi\le
1133: I_2\int_{-r}^r \gamma(\xi)^2\drm\xi\,,
1134: $$
1135: where
1136: \begin{equation}\label{eq:I2}
1137: I_2:=\max\{1,\dsp\frac{1}{1-A}\}\alpha_{\bullet}^2\int_0^\delta \eta^2\erm^{-\beta(\eta)}|\phi_*(\eta)|^2\,\drm\eta\ge 0\,.
1138: \end{equation}
1139:
1140:
1141: Thus, as $\gamma(\xi)$ vanishes for $|\xi|>R$, we have
1142: $$
1143: D_\gamma=I_1\int \gamma(\xi)\drm\xi+I_2\int \gamma(\xi)^2\drm\xi=
1144: (-I_1)\left(C_\beta\int \gamma(\xi)^2\drm\xi-\int \gamma(\xi)\drm\xi\right)\,,
1145: $$
1146: where
1147: \begin{equation}\label{eq:C_beta}
1148: C_\beta=\frac{I_2}{-I_1}=\frac{\dsp\max\{1,\dsp\frac{1}{1-A}\}\alpha_{\bullet}^2\int_0^\delta \eta^2\erm^{-\beta(\eta)}|\phi_*(\eta)|^2\,\drm\eta}
1149: {\dsp\int_0^\delta\eta\erm^{-\beta(\eta)}
1150: \left(|\phi'_*(\eta)|^2-\alpha_{\bullet}^2|\phi_*(\eta)|^2\right)\,\drm\eta}
1151: \end{equation}
1152: is a positive constant.
1153:
1154: As soon as \eqref{eq:main_cond} holds, $D_\gamma$ is negative, and so \eqref{eq:main_ineq_mod} holds. This proves Theorem~\ref{thm:main}.
1155:
1156: Finally, it is sufficient to note that \eqref{eq:cor_cond} implies $\dsp\int \gamma(\xi)^2\drm\xi<2Rc_{\beta,R}\int \gamma(\xi)\drm\xi$,
1157: which proves Corollary~\ref{cor:main}.
1158:
1159: \section{Conclusions}
1160: It has been shown that a trapped mode is possible in the model presented here.
1161: To increase confidence that such modes exist on real coasts further work is clearly
1162: required to demonstrate that this mode is not an artifact of the modelling assumptions.
1163: However these assumptions are the usual ones for the simple theory of CSWs and extensions
1164: to include stratification and more realistic boundary conditions have not in general
1165: contradicted them \cite{LeBloM78}.
1166: The result here suggests that it would be of interest to compare low frequency
1167: velocity records in the neighborhood of capes with those on nearby straight
1168: coasts to determine whether there is indeed enhanced energy at the cape. Both
1169: the above endeavors are being pursued.
1170:
1171: \bibliographystyle{alpha}
1172: %\bibliography{/flip/erj/Bibliographies/ERJv5}
1173: \begin{thebibliography}{StHuSaTrZa}
1174:
1175: \bibitem[AsPaVa]{APV}
1176: A.~Aslanyan, L.~Parnovski, and D.~Vassiliev, \emph{Complex resonances in
1177: acoustic waveguides}, Q.~Jl~Mech. Appl. Math. \textbf{53} (2000), 429--447.
1178:
1179: \bibitem[Da]{EBD}
1180: E.~B. Davies, \emph{Spectral theory and differential operators}, Camb. Univ
1181: Press, Cambridge, 1995.
1182:
1183: \bibitem[DaPa]{DP}
1184: E.~B. Davies and L.~Parnovski, \emph{Trapped modes in acoustic waveguides},
1185: Q.~Jl~Mech. Appl. Math. \textbf{51} (1998), 477--492.
1186:
1187: \bibitem[DuEx]{DE}
1188: P.~Duclos and P.~Exner, \emph{{C}urvature-induced bound states in quantum
1189: waveguides in two and three dimensions}, Rev. Math. Phys. \textbf{7} (1995),
1190: 73--102.
1191:
1192: \bibitem[EvLeVa]{ELV}
1193: D.~V. Evans, M.~Levitin, and D.~Vassiliev, \emph{Existence theorems for trapped
1194: modes}, J.~Fluid Mech. \textbf{261} (1994), 21--31.
1195:
1196: \bibitem[ExSe]{ES}
1197: P.~Exner and P.~{\v S}eba, \emph{Bound states in curved quantum waveguides},
1198: J.~Math.~Phys. \textbf{30} (1989), 2574--2580.
1199:
1200:
1201: \bibitem[Ha]{Hamon66}
1202: B.~V. Hamon,
1203: \emph{Continental shelf waves and the effect of atmospheric pressure and
1204: wind stress on sea level}, J. Geophys. Res \textbf{71} (1966), 2883--2893.
1205:
1206: \bibitem[Jo1]{Johns87a}
1207: E.~R. Johnson, \emph{A conformal-mapping technique for topographic-wave problems ---
1208: semi-infinite channels and elongated basins},
1209: J.~Fluid Mech., \textbf{177} (1987), 395--405.
1210:
1211: \bibitem[Jo2]{Johns89a}
1212: E.~R. Johnson,
1213: \emph{Topographic waves in open domains. I. Boundary conditions and
1214: frequency estimates},
1215: J.~Fluid Mech. \textbf{200} (1989), 69--76.
1216:
1217: \bibitem[Jo3]{Johns91b}
1218: E.~R. Johnson,
1219: \emph{Low-frequency barotropic scattering on a shelf bordering
1220: an ocean}
1221: J. Phys. Oceanog. \textbf{21} (1991), 720--727.
1222:
1223: \bibitem[KrKr]{KK}
1224: D.~Krej\v{c}i\v{r}\'{\i}k and J.~K\v{r}\'{\i}\v{z}, \emph{On the Spectrum of Curved Quantum Waveguides},
1225: preprint \texttt{math-ph/0306008} (2003).
1226:
1227: \bibitem[La]{Lamb32}
1228: H.~Lamb,
1229: \emph{Hydrodynamics}.
1230: Cambridge University Press, 6th edition, 1932.
1231:
1232: \bibitem[LBM]{LeBloM78}
1233: P.~H. LeBlond and L.~A. Mysak,
1234: \emph{Waves in the {O}cean}.
1235: Elsevier, 1978.
1236:
1237: \bibitem[Pe]{Pedlo86}
1238: J.~Pedlosky,
1239: \emph{Geophysical {F}luid {D}ynamics}.
1240: Springer, 1986.
1241:
1242: \bibitem[StHu1]{StockH86}
1243: T.~F. Stocker and K.~Hutter,
1244: \emph{One-dimensional models for topographic rossby waves in elongated
1245: basins on the f-plane},
1246: J. Fluid Mech. \textbf{170} (1986), 435--459.
1247:
1248: \bibitem[StHu2]{StockH87}
1249: T.~F. Stocker and K.~Hutter,
1250: \emph{Topographic waves in rectangular basins},
1251: J. Fluid Mech. \textbf{185} (1987), 107--120.
1252:
1253: \bibitem[StHuSaTrZa]{StockHSTZ87}
1254: T.~F. Stocker, K.~Hutter, G.~Salvade, J.~Tr\"osch, and F.~Zamboni,
1255: \emph{Observations and analysis of internal seiches in the southern basin
1256: of {L}ake of {L}ugano},
1257: Ann. Geophys. B --- Terr. Planet. Phys. \textbf{5} (1987), 553--568.
1258:
1259: \bibitem[StJo1]{StockJ89}
1260: T.~F. Stocker and E.~R. Johnson,
1261: \emph{Topographic waves in open domains. II. Bay modes and resonances},
1262: J. Fluid Mech. \textbf{200} (1989), 77--93.
1263:
1264: \bibitem[StJo2]{StockJ91}
1265: T.~F. Stocker and E.~R. Johnson,
1266: \emph{The trapping and scattering of topographic waves by estuaries and
1267: headlands},
1268: J. Fluid Mech. \textbf{222} (1991) 501--524.
1269:
1270: \bibitem[Tr]{Trosch84}
1271: J.~Tr\"osch,
1272: \emph{Finite element calculation of topographic waves in lakes},
1273: in {Han Min Hsia}, {You Li Chou}, {Shu Yi Wang}, and {Sheng Jii
1274: Hsieh}, editors, {\em Proc. 4th {I}ntl {C}onf. {A}ppl. {N}umerical
1275: {M}odeling. {T}ainan, {T}aiwan} (1984), 307--311.
1276:
1277:
1278: \end{thebibliography}
1279:
1280: \end{document}
1281: