1: %0. Target journal= ARMA.
2: %TODO:
3: %1. The definition of PW version of D is backwards, in fact
4: %defines $\bar D$ by our (std) complex inner product convention..
5: %FIX?
6: %
7: %2. Add ref. to JLW along with CJLW?
8: %
9: %3. Update refs (in prep \to preprint, preprint \to publication)
10: %as possible.
11: %
12: %
13: %OLD NOTES: Add ref. to Schlichting (thanks, Mark Williams!):
14: %I'm pretty sure you are talking about the book
15: %
16: %NOTES:
17: %1. ref (from Mark W.): Boundary Layer Theory
18: %
19: %by Hermann Schlichting, who was (I think) a student of Prandtl. It's a really impressive book; you'll find all kinds of things in there, including characteristic layers, turbulent layers,....Chapter 14 on boundary layer control talks about boundary conditions that involve suction or injection.
20: %DONE.
21: %
22: %2. IDEA: More general approach? (for full gas, say???)
23: %about convergence to shock case.
24: % To verify that $\beta(\lambda)\ne 0$, try
25: %``Projection lemma'' in spirit of \cite{GZ}, but
26: %for real visc. case. I think this may work, since
27: %it is a $2\times 2$ system. That was the main ingredient
28: %of the argument in \cite{GZ}.
29: %Recall: Projection lemma states that, in phase variables
30: %$(u,u')^T$ for strictly parabolic case, the projection onto
31: %the first (or second) variable is one-to-one on either the
32: %stable or unstable subspace of the limiting coefficient matrix
33: %of the eigenvalue equations written as a first-order system.
34: %Here, we want projection onto third coordinate nonvanishing
35: %for unstable manifold of $A_+$.
36: %See also \cite{Z.3} for generalization to real visc. case.
37: %(Yes, I think this may do it!-KZ)
38: \documentclass[11pt]{amsart}
39: \usepackage{amsmath,amsfonts,amsthm,amscd,amssymb,graphicx}
40:
41: \newtheorem{theorem}{Theorem}[section]
42: \newtheorem{lemma}[theorem]{Lemma}
43: \newtheorem{proposition}[theorem]{Proposition}
44: \newtheorem{corollary}[theorem]{Corollary}
45: \newtheorem{assumption}[theorem]{Assumption}
46: \newtheorem{example}[theorem]{Example}
47: \newtheorem{remark}[theorem]{Remark}
48:
49:
50:
51: \newcommand{\ipt}[2]{\langle #1,#2 \rangle}
52: \newcommand{\ip}{\int_{\mathbb{R}^+}}
53: \newcommand{\tr}{\mbox{tr}}
54: \newcommand{\ad}[2]{\mbox{Ad}_{#1} #2}
55: \newcommand{\adj}[1]{\mbox{Ad}_{#1}}
56: \renewcommand{\ker}[1]{\mathcal{N}(#1)}
57: \newcommand{\ran}[1]{\mathcal{R}(#1)}
58: \newcommand{\C}{\mathbb{C}}
59: \newcommand{\bbF}{\mathbb{F}}
60: \newcommand{\M}{M_{n}}
61:
62:
63: \renewcommand{\labelenumi}{(\roman{enumi}).}
64:
65:
66: \def\bb1{{1\!\!1}}
67: %
68: \def\CalA{\mathcal{A}}
69: \def\CalL{\mathcal{L}}
70: \def\CalO{\mathcal{O}}
71: \def\CalV{\mathcal{V}}
72: \def\CalU{\mathcal{U}}
73: %
74: \def\bu{\hat{u}}
75: \def\bv{\hat{v}}
76: \def\cU{{\check{U}}}
77: \def\bU{{\hat{U}}}
78: \def\bV{{\hat{v}}}
79: %
80: \def\cu{\check{u}}
81: \def\cv{\check{v}}
82: \def\hu{\hat{u}}
83: \def\hv{\hat{v}}
84: \def\hU{\widehat{U}}
85: %
86: \def \tu{\tilde{u}}
87: \def \tv{\tilde{v}}
88: \def\tU{{\widetilde{U}}}
89: \def\tV{{\widetilde{V}}}
90: %
91: \def\R{\Re e}
92: \def\I{\Im m}
93:
94: \def\I{\Im m}
95: \def\sgn{\rm sgn}
96:
97: \begin{document}
98: \title[Stability of boundary layers]
99: {Spectral stability of noncharacteristic isentropic
100: Navier--Stokes boundary layers}
101: \author[Costanzino, Humpherys, Nguyen, and Zumbrun]{Nicola Costanzino, Jeffrey Humpherys, \\ Toan Nguyen, and Kevin Zumbrun}
102:
103: \date{Last Updated: June 22, 2007}
104:
105: \thanks{ This work was supported in part by the National Science Foundation award numbers DMS-0607721 and DMS-0300487.}
106:
107: \address{Department of Mathematics, Pennsylvania State
108: University, University Park, PA, 16802}
109: \email{costanzi@math.psu.edu}
110: \address{Department of Mathematics, Brigham Young University, Provo, UT 84602}
111: \email{jeffh@math.byu.edu}
112: \address{Department of Mathematics, Indiana University, Bloomington, IN 47402}
113: \email{nguyentt@indiana.edu}
114: \address{Department of Mathematics, Indiana University, Bloomington, IN 47402}
115: \email{kzumbrun@indiana.edu}
116:
117:
118:
119: \begin{abstract}
120: Building on work of Barker, Humpherys, Lafitte, Rudd, and Zumbrun
121: in the shock wave case, we study stability of
122: compressive, or ``shock-like'',
123: boundary layers of the isentropic compressible
124: Navier--Stokes equations with $\gamma$-law pressure
125: by a combination of asymptotic ODE estimates and numerical Evans function
126: computations.
127: Our results indicate stability for $\gamma\in [1, 3]$ for all
128: compressive boundary-layers, independent of amplitude, save for
129: inflow layers in the characteristic limit (not treated).
130: Expansive inflow boundary-layers have been shown to be
131: stable for all amplitudes by Matsumura and Nishihara using energy estimates.
132: Besides the parameter of amplitude appearing in the shock case,
133: the boundary-layer case features an additional parameter
134: measuring displacement of the background profile, which greatly
135: complicates the resulting case structure.
136: Moreover, inflow boundary layers turn out to have quite delicate
137: stability in both large-displacement and large-amplitude limits,
138: necessitating the additional use of a mod-two stability index
139: studied earlier by Serre and Zumbrun in order to decide stability.
140: \end{abstract}
141:
142: \maketitle
143:
144: %\clearpage
145: \tableofcontents
146: %\clearpage
147: %%%%%%%%%%%%%%%%%
148: \bigbreak
149: \section{Introduction} \label{int}
150:
151: Consider the isentropic compressible Navier-Stokes equations
152: \begin{equation}
153: \begin{split}
154: \label{eulerian}
155: \rho_{t}+ (\rho u)_{x} &=0, \\
156: (\rho u)_{t}+ (\rho u^2)_x + p(\rho)_{x} &= u_{xx}
157: \end{split}
158: \end{equation}
159: on the quarter-plane $x,t \ge 0$,
160: where $\rho>0$, $u$, $p$ denote density, velocity, and pressure
161: at spatial location $x$ and time $t$,
162: with $\gamma$-law pressure function
163: \begin{equation}\label{gaslaw}
164: p(\rho) = a_0 \rho^{\gamma},
165: \qquad
166: a_0>0, \, \gamma \geq 1,
167: \end{equation}
168: and noncharacteristic constant ``inflow'' or ``outflow''
169: boundary conditions
170: \begin{equation}\label{inflow}
171: (\rho, u)(0,t) \equiv (\rho_0, u_0), \qquad u_0>0
172: \end{equation}
173: or
174: \begin{equation}\label{outflow}
175: u(0,t) \equiv u_0 \qquad u_0<0
176: \end{equation}
177: as discussed in \cite{SZ,GMWZ.5,GMWZ.6}.
178: The sign of the velocity at $x=0$ determines whether
179: characteristics of the hyperbolic transport equation
180: $\rho_t+ u\rho_x= f$ enter the domain
181: (considering $f:=\rho u_x$ as a lower-order forcing term),
182: and thus whether $\rho(0,t)$ should be prescribed.
183: The variable-coefficient parabolic equation
184: $\rho u_t - u_{xx}= g$ requires prescription of $u(0,t)$ in
185: %CHANGED: note, I've subtracted out first eqn. to get
186: %terms that are truly lower-order... $\rho_t$ no good..-K
187: either case, with $g:= -\rho (u^2/2)_x -p(\rho)_x $.
188: %ENDCHANGED
189:
190: %NOTE: nontrivial BL exists for weak inflow (one hyp. char enters)
191: %or strong outflow (no hyp. char. enters domain).
192: By comparison, the purely hyperbolic isentropic Euler equations
193: \begin{equation}
194: \begin{split}
195: \label{euler}
196: \rho_{t}+ (\rho u)_{x} &=0, \\
197: (\rho u)_{t}+ (\rho u^2)_x + p(\rho)_{x} &= 0
198: \end{split}
199: \end{equation}
200: have characteristic speeds $a= u\pm \sqrt{p'(\rho)}$, hence,
201: depending on the values of $(\rho, u)(0,t)$, may have one, two,
202: or no characteristics entering the domain, hence require one,
203: two, or no prescribed boundary values.
204: In particular, there is a discrepancy between the number of
205: prescribed boundary values for \eqref{eulerian} and \eqref{euler}
206: in the case of mild inflow $u_0>0$ small (two for \eqref{eulerian}, one
207: for \eqref{euler}) or
208: strong outflow $u_0<0$ large (one for \eqref{eulerian}, none
209: for \eqref{euler}), indicating the possibility of {\it boundary layers},
210: or asymptotically-constant stationary solutions of \eqref{eulerian}:
211: \begin{equation}\label{BL}
212: (\rho, u)(x,t)\equiv (\hat \rho, \hat u)(x),
213: \qquad
214: \lim_{z\to +\infty} (\hat \rho, \hat u)(z)= (\rho_+, u_+).
215: \end{equation}
216: Indeed, existence of such solutions is straightforward to verify
217: by direct computations on the (scalar) stationary-wave ODE;
218: see \cite{MeZ,SZ,MN.2,KNZ,GMWZ.5,GMWZ.6} or Section \ref{profsec}.
219: These may be either of ``expansive'' type, resembling rarefaction
220: wave solutions on the whole line, or ``compressive'' type,
221: resembling viscous shock solutions.
222:
223: A fundamental question is whether or not such boundary layer solutions
224: are {\it stable} in the sense of PDE.
225: For the expansive inflow case, it has been shown
226: in \cite{MN.2} that {\it all} boundary layers are stable,
227: independent of amplitude, by energy estimates similar
228: to those used to prove the corresponding result for rarefactions
229: on the whole line.
230: Here, we concentrate on the complementary, {\it compressive case}
231: (though see discussion, Section \ref{discussion}).
232:
233: Linearized and nonlinear stability of general
234: (expansive or compressive) {\it small-amplitude} noncharacteristic
235: boundary layers of \eqref{eulerian} have been established
236: in \cite{MN.2,R,KNZ,GMWZ.5}.
237: More generally, it has been shown in \cite{GMWZ.5,YZ} that linearized
238: and nonlinear stability are equivalent to {spectral stability},
239: or nonexistence of nonstable (nonnegative real part) eigenvalues of
240: the linearized operator about the layer, for boundary layers of arbitrary amplitude.
241: However, up to now the spectral stability of {\it large-amplitude
242: compressive} boundary layers has remained largely undetermined.\footnote{
243: See, however, the
244: investigations of \cite{SZ} on stability index, or parity of the number
245: of nonstable eigenvalues of the linearized operator about the layer.}
246:
247: We resolve this question in the present paper, carrying out a systematic,
248: global study classifying the stability of all possible compressive
249: boundary-layer solutions of \eqref{eulerian}.
250: Our method of analysis is by a combination of asymptotic ODE techniques
251: and numerical Evans function computations,
252: following a basic approach introduced recently in \cite{HLZ,BHRZ} for
253: the study of the closely related shock wave case.
254: Here, there are interesting complications associated with the
255: richer class of boundary-layer solutions as compared to possible
256: shock solutions, the delicate stability properties of the
257: inflow case, and, in the outflow case, the nonstandard eigenvalue problem
258: arising from reduction to Lagrangian coordinates.
259: %TODO: rephrase the last sentence? NO.
260:
261: Our conclusions are, for both inflow and outflow conditions,
262: that compressive boundary layers that are uniformly noncharacteristic
263: in a sense to be made precise later (specifically, $v_+$ bounded
264: away from $1$, in the terminology of Section \ref{profsec})
265: are {\it unconditionally stable}, independent of amplitude,
266: on the range $\gamma \in [1,3]$ considered in our numerical
267: computations.
268: We show by energy estimates that {\it outflow boundary layers are stable
269: also in the characteristic limit}.
270: The omitted characteristic limit in the inflow case,
271: analogous to the small-amplitude limit
272: for the shock case should be treatable by the singular perturbation
273: methods used in \cite{PZ,FS} to treat the small-amplitude shock case;
274: however, we do not consider this case here.
275:
276: In the inflow case, our results, together with
277: those of \cite{MN.2}, completely resolve the question of stability
278: of isentropic (expansive or compressive) uniformly noncharacteristic
279: boundary layers for $\gamma\in [1,3]$, yielding
280: {\it unconditional stability independent of amplitude or type.}
281: In the outflow case, we show stability of all {compressive} boundary
282: layers without the assumption of uniform noncharacteristicity.
283:
284: \subsection{Discussion and open problems}\label{discussion}
285: The small-amplitude results obtained in \cite{MN.2,KNZ,R,GMWZ.5}
286: are of ``general type'', making little use of the specific
287: structure of the equations.
288: Essentially, they all require that the difference between the
289: boundary layer solution and its constant limit at $|x|=\infty$
290: be small in $L^1$.\footnote{Alternatively, as in
291: \cite{MN.2,R}, the essentially equivalent condition
292: that $x\hat v'(x)$ be small in $L^1$.
293: (For monotone profiles,
294: $\int_0^{+\infty} |\hat v-v_+|dx=
295: \pm\int_0^{+\infty} (\hat v-v_+) dx=
296: \mp\int_0^{+\infty} x\hat v'dx$.)}
297: %Rousset requires derivative have small first moment in L^1, same
298: %thing essentially... Interesting question: What do GG require???
299: As pointed out in \cite{GMWZ.5}, this is
300: the ``gap lemma'' regime in which standard asymptotic ODE
301: estimates show that behavior is essentially governed by
302: the limiting constant-coefficient equations at infinity,
303: and thus stability may be concluded immediately from stability
304: (computable by exact solution) of the constant layer
305: identically equal to the limiting state.
306: These methods do not suffice to treat either the (small-amplitude)
307: characteristic limit or the large-amplitude case, which require
308: more refined analyses.
309: %
310: In particular, up to now, {\it there was no analysis considering boundary
311: layers approaching a full viscous shock profile},
312: {\it not even a profile of vanishingly small amplitude}.
313: Our analysis of this limit indicates why: the appearance of a small
314: eigenvalue near zero prevents uniform estimates such as would be obtained
315: by usual types of energy estimates.
316:
317: By contrast, the large-amplitude results obtained here and (for expansive
318: layers) in \cite{MN.2}
319: make use of the specific form of the equations.
320: In particular, both analyses make use of the advantageous structure
321: in Lagrangian coordinates.
322: The possibility to work in Lagrangian coordinates was first pointed out
323: by Matsumura--Nishihara \cite{MN.2} in the inflow case,
324: for which the stationary boundary transforms to a moving boundary
325: with constant speed.
326: Here we show how to convert the outflow problem also to Lagrangian
327: coordinates, by converting the resulting variable-speed
328: boundary problem to a constant-speed one with modified boundary condition.
329: This trick seems of general use.
330: %CHANGED
331: In particular,
332: %we expect that the
333: %energy methods of \cite{MN.2} applied in this framework should yield
334: it might be possible that the
335: energy methods of \cite{MN.2} applied in this framework would yield
336: unconditional stability of expansive boundary-layers,
337: completing the analysis of the outflow case.
338: %We see no reason why the methods of this paper
339: %should not apply in the expansive case as well, to yield
340: %stability by a different approach.
341: Alternatively, this case could be attacked by the methods of the
342: present paper.
343: %ENDCHANGED
344: These are two further interesting direction for future investigation.
345:
346: In the outflow case, a further transformation to the
347: ``balanced flux form'' introduced in \cite{PZ}, in which the equations
348: take the form of the integrated shock equations,
349: allows us to establish stability in the characteristic limit by
350: energy estimates like those of \cite{MN} in the shock case.
351: The treatment of the characteristic inflow limit by the methods of \cite{PZ,FS}
352: seems to be another extremely interesting direction for future study.
353:
354: Finally, we point to the extension of the present methods to
355: full (nonisentropic) gas dynamics and multidimensions as
356: the two outstanding open problems in this area.
357:
358: New features of the present analysis as compared to the shock
359: case considered in \cite{BHRZ,HLZ} are the presence of two parameters,
360: strength and displacement, indexing possible boundary layers,
361: vs. the single parameter of strength in the shock case, and
362: the fact that the limiting equations in several asymptotic regimes
363: possess zero eigenvalues, making the limiting stability analysis much
364: more delicate than in the shock case.
365: The latter is seen, for example, in the limit as a compressive
366: boundary layer approaches a full stationary shock solution,
367: which we show to be spectrally equivalent to the situation
368: of unintegrated shock equations on the whole line.
369: As the equations on the line
370: possess always a translational eigenvalue at $\lambda=0$, we
371: may conclude existence of a zero at $\lambda=0$ for the limiting equations
372: and thus a zero {\it near} $\lambda=0$ as we approach this limit,
373: which could be stable or unstable.
374: Similarly, the Evans function in the inflow case is shown to
375: converge in the large-strength limit to a function with a
376: zero at $\lambda=0$, with the same conclusions;
377: see Section \ref{description} for further details.
378:
379: %CHANGED: better wording-KZ
380: %To deal with this latter possibility, we find it necessary
381: To deal with this latter circumstance, we find it necessary
382: %ENDCHANGED
383: to make use also of topological information provided by the
384: stability index of \cite{PW,GZ,SZ}, a mod-two index counting
385: the parity of the number of unstable eigenvalues.
386: Together with the information that there is at most one
387: unstable zero, the parity information provided by the stability
388: index is sufficient to determine whether an unstable zero
389: does or does not occur.
390: Remarkably, in the isentropic case we are able to compute {explicitly}
391: the stability index for all parameter values, recovering
392: results obtained by indirect argument in \cite{SZ},
393: and thereby completing the stability analysis in the presence
394: of a single possibly unstable zero.
395: %NOTE: discussion of vanishing at infinity that was given in
396: %SZ is wrong! nonvanishing does NOT follow from garding ineq.,
397: %or rather Garding doesn't immediatley follow in this setting..
398: %something to do here! Nonetheless, very complete discussion...
399: %in particular, stab. index already done...!
400:
401: %CHANGED: deleted- weak ending..-KZ
402: %In summary, we here illustrate the robustness of the methods introduced
403: %in \cite{HLZ} for the shock case
404: %by the application to the considerably more complicated case of boundary
405: %layers, in the process introducing a number of refinements that may be of
406: %use in more general situations.
407:
408: \section{Preliminaries}\label{prelim}
409: We begin by carrying out a number of preliminary steps similar to
410: those carried out in \cite{BHRZ,HLZ} for the shock case, but
411: complicated somewhat by the need to treat the boundary and
412: its different conditions in the inflow and outflow case.
413:
414: \subsection{Lagrangian formulation.}\label{lagrangiansec}
415:
416: The analyses of \cite{HLZ,BHRZ} in the shock wave case were carried
417: out in Lagrangian coordinates, which proved to be particularly
418: convenient.
419: Our first step, therefore, is to convert the Eulerian formulation
420: \eqref{eulerian} into Lagrangian coordinates similar to those of the shock case.
421: However, standard Lagrangian coordinates in which the spatial variable
422: $\tilde x$ is constant on particle paths are not appropriate for the
423: boundary-value problem with inflow/outflow.
424: We therefore introduce instead ``psuedo-Lagrangian'' coordinates
425: \begin{equation}\label{psuedo}
426: \tilde x:=\int_0^x \rho(y,t)\, dy, \quad \tilde t:=t,
427: \end{equation}
428: in which the physical boundary $x=0$ remains fixed at $\tilde x=0$.
429:
430: Straightforward calculation reveals that in these coordinates
431: \eqref{eulerian} becomes
432:
433: \begin{equation}\label{NS}
434: \begin{aligned}
435: v_t - s v_{\tilde x} - u_{\tilde x} &= \sigma(t)v_{\tilde x} \\
436: u_t - s u_{\tilde x} + p(v)_{\tilde x} -\left(\frac{u_{\tilde
437: x}}{v}\right)_{\tilde x} &= \sigma(t) u_{\tilde x}
438: \end{aligned}
439: %\hspace{0.5cm} \right\} \;\; \mbox{in} \; x>0
440: \end{equation}
441: on $x>0$, where
442: \begin{equation}\label{s(t)}
443: s = -\frac{u_0}{v_0}, \; \sigma(t) = m(t) - s, \; m(t) := -
444: \rho(0,t) u(0,t)=-u(0,t)/v(0,t),
445: \end{equation} so that $m(t)$ is the negative of the momentum at the
446: boundary $x= \tilde x = 0$. From now on, we drop the tilde, denoting
447: $\tilde x$ simply as $x$.
448:
449:
450:
451: \subsubsection{Inflow case}\label{lagin}
452: For the inflow case, $u_0>0$ so we may prescribe {\em two} boundary
453: conditions on \eqref{NS}, namely
454: \begin{equation}
455: v|_{x=0} = v_0 > 0, \hspace{0.5cm} u|_{x=0} = u_0 > 0
456: \end{equation} where both $u_0, v_0$ are constant.
457:
458:
459: \subsubsection{Outflow case}\label{lagout}
460: For the outflow case, $u_0<0$ so we may prescribe {\em only one}
461: boundary condition on \eqref{NS}, namely
462: \begin{equation}
463: u|_{x=0} = u_0 < 0.
464: \end{equation} Thus $v(0,t)$ is an unknown in the problem, which
465: %CHANGED: grammar-KZ
466: %makes the analysis of the outflow case more subtle.
467: makes the analysis of the outflow case more subtle than that of
468: the inflow case.
469: %ENDCHANGED
470:
471: \subsection{Rescaled coordinates}\label{renormalization}
472: Our next step is to rescale the equations in such a way
473: that coefficients remain bounded in the strong boundary-layer
474: limit.
475: Consider the change of variables
476: \begin{equation}\label{scaling}
477: (x,t,v,u) \rightarrow (-\varepsilon s x, \varepsilon s^2 t,
478: v/\varepsilon, -u/(\varepsilon s)),
479: \end{equation}
480: where $\varepsilon$ is chosen so that
481: \begin{equation}
482: 0 < v_+ < v_- = 1,
483: \end{equation}
484: where $v_+$ is the limit as $x\to +\infty$ of the boundary layer
485: (stationary solution) $(\hat v, \hat u)$ under consideration
486: and $v_-$ is the limit as $x\to -\infty$ of its continuation
487: into $x<0$ as a solution of the standing-wave ODE (discussed in
488: more detail just below).
489: Under the rescaling \eqref{scaling}, \eqref{NS} becomes
490: \begin{equation} \label{rescaled}
491: \begin{aligned}
492: v_t + v_x - u_x &= \sigma(t)v_x,\\
493: u_t + u_x + (a v^{-\gamma})_x &= \sigma(t) u_x +
494: \left(\frac{u_x}{v}\right)_x
495: \end{aligned}
496: %\hspace{0.5cm} \right\} \;\; \mbox{on} \; x\gtrless 0
497: \end{equation}
498: where $a = a_0 \varepsilon^{-\gamma-1} s^{-2}$, $\sigma=-u(0,t)/v(0,t)+1$,
499: on respective domains
500: $$
501: x>0 \, \hbox{\rm (inflow case)}
502: \qquad x<0 \, \hbox{\rm (outflow case)}.
503: $$
504:
505: \subsection{Stationary boundary layers}\label{profsec}
506: Stationary boundary layers
507: %CHANGED: typesetting
508: %$ (v,u)(x,t) = (\bv,\bu)(x)$
509: $$
510: (v,u)(x,t) = (\bv,\bu)(x)
511: $$
512: %ENDCHANGED
513: of \eqref{rescaled} satisfy
514: \begin{equation}\label{stationarybl}
515: \begin{split}
516: (a) \hspace{0.5cm} & \bv' - \bu' = 0 \\
517: (b) \hspace{0.5cm} & \bu' + (a\bv^{-\gamma}) = \left(\frac{\bu'}{\bv}\right)' \\
518: (c) \hspace{0.5cm} & (\bv,\bu)|_{x=0} = (v_0,u_0)\\
519: (d) \hspace{0.5cm} & \lim_{x\rightarrow \pm\infty}(\bv,\bu) =
520: (v,u)_\pm,
521: \end{split}
522: \end{equation}
523: where (d) is imposed at $+\infty$ in the inflow case, $-\infty$
524: in the outflow case and (imposing $\sigma=0$) $u_0=v_0$.
525: %with $u_0 > 0$ in the inflow case and $u_0 < 0$ in the outflow case.
526: Using \eqref{stationarybl}(a) we can reduce
527: this to the study of the scalar ODE,
528: \begin{equation}\label{scalarode}
529: \bv' + (a\bv^{-\gamma})' = \left(\frac{\bv'}{\bv}\right)' \\
530: \end{equation} with the same boundary conditions at $x=0$ and
531: %CHANGED: both infinities
532: %$x=+\infty$ as above.
533: $x=\pm\infty$ as above.
534: %ENDCHANGED
535: Taking the antiderivative of this equation
536: yields
537: \begin{equation}\label{profeqC}
538: \bv' = \mathcal{H}_C(\bv) = \bv(\bv + a\bv^{-\gamma} +C ),
539: \end{equation}
540: where $C$ is a constant of integration.
541:
542: Noting that $\mathcal{H}_C$ is convex, we find that there
543: are precisely two rest points of \eqref{profeqC} whenever boundary-layer
544: profiles exist, except at the single parameter value on the boundary
545: between existence and nonexistence of solutions, for which
546: there is a degenerate rest point (double root of $\mathcal{H}_C$).
547: Ignoring this degenerate case, we see that boundary layers terminating
548: at rest point $v_+$ as $x\to +\infty$ must either continue
549: backward into $x<0$ to terminate at a second rest point $v_-$ as
550: $x\to -\infty$, or else blow up to infinity as $x\to -\infty$.
551: The first case we shall call {\it compressive}, the second
552: {\it expansive}.
553:
554: In the first case, the extended solution on the whole line
555: may be recognized as a standing viscous shock wave;
556: that is, {\it for isentropic gas
557: dynamics, compressive boundary layers are just restrictions to the half-line
558: $x\ge 0$ [resp. $x\le 0$] of standing shock waves.}
559: In the second case, as discussed in \cite{MN.2}, the boundary layers
560: are somewhat analogous to rarefaction waves on the whole line.
561: {From here on, we concentrate exclusively on the compressive case.}
562:
563: With the choice $v_-=1$, we may carry out the integration of
564: \eqref{scalarode} once more, this time as a definite
565: integral from $-\infty$ to $x$, to obtain
566: \begin{equation}\label{profeq}
567: \bv' = H(\bv) = \bv(\bv-1 + a(\bv^{-\gamma} - 1)),
568: \end{equation}
569: where $a$ is found by letting $x \rightarrow +\infty$, yielding
570: \begin{equation}
571: \label{RH} a = -\frac{v_+ - 1}{v_+^{-\gamma} - 1} = v_+^\gamma
572: \frac{1-v_+}{1-v_+^\gamma} \, ;
573: \end{equation}
574: %
575: in particular,
576: $ a\sim v_+^\gamma $ in the large boundary layer limit $v_+\to 0$.
577: %NOTE: $a\sim 1/\gamma$ in char. limit $v_+\to 1$.
578: This is exactly the equation for viscous shock profiles
579: considered in \cite{HLZ}.
580:
581: \subsection{Eigenvalue equations}\label{eigsec}
582: Linearizing \eqref{rescaled} about $(\bv,\bu)$, we obtain
583:
584: \begin{equation} \label{linearized}
585: \begin{split}
586: & \tv_t + \tv_x - \tu_x = \frac{\tv(0,t)}{ v_0}\bv ' \\
587: & \tu_t + \tu_x - \left(\frac{h(\bv)}{\bv^{\gamma + 1}} \tilde v \right)_x
588: - \left( \frac{\tu_x}{\bv} \right)_x=
589: \frac{\tv(0,t)}{v_0}\bu ' \\
590: & (\tv,\tu)|_{x=0} = (\tv_0(t),0) \\
591: & \lim_{x\rightarrow +\infty}(\tv,\tu) = (0,0)
592: \end{split}
593: \end{equation}
594: where $v_0=\hat v(0)$,
595: \begin{equation}\label{f}
596: h(\bv) = -\bv^{\gamma + 1} + a(\gamma - 1) + (a+1)\bv^\gamma
597: \end{equation}
598: and $\tv,\tu$ denote perturbations of $\bv, \bu$.
599:
600: \subsubsection{Inflow case}\label{ineigen}
601: In the inflow case, $\tilde u(0,t)=\tilde v(0,t)\equiv 0$, yielding
602: \begin{equation}\label{eigen1}
603: \begin{aligned}
604: \lambda v + v_x - u_x &= 0\\
605: \lambda u + u_x - \left(\frac{h(\bv)}{\bv^{\gamma + 1}} v \right)_x
606: &= \left( \frac{u_x}{\bv} \right)_x\\
607: \end{aligned}
608: \end{equation}
609: on $x>0$, with full Dirichlet conditions $(v,u)|_{x=0}=(0,0)$.
610:
611: \subsubsection{Outflow case}\label{outeigen}
612: Letting $\tU := (\tv,\tu)^{T}$, $\hat U:=(\hat v, \hat u)^T$,
613: and denoting by $\CalL$ the operator
614: associated to the linearization about boundary-layer $(\hat v,\hat u)$,
615: \begin{equation}\label{linoperator}
616: \CalL := \partial_x A(x) - \partial_x B(x)\partial_x,
617: \end{equation} where
618: \begin{equation}
619: A(x) = \left( \begin{array}{cc} 1 & -1 \\
620: -h(\bv)/\bv^{\gamma + 1} & 1 \\
621: \end{array} \right), \hspace{0.5cm} B(x) = \left( \begin{array}{cc} 0 & 0 \\
622: 0 & \bv^{-1} \end{array} \right),
623: \end{equation}
624: we have
625: $\tilde U_t - \CalL \tilde U = \frac{\tv_0(t)}{v_0}\hat U'(x)$,
626: with associated eigenvalue equation
627: \begin{equation} \label{vectorevalue}
628: %CHANGED
629: \lambda \tilde U - \CalL \tilde U = \frac{\tv(0,\lambda)}{v_0}\hat U'(x),
630: %ENDCHANGED
631: \end{equation}
632: where $\bU' = (\bv',\bu')$.
633:
634:
635: To eliminate the nonstandard inhomogeneous term on the righthand side of
636: \eqref{vectorevalue}, we introduce
637: a ``good unknown" (c.f. \cite{Al,CJLW,GMWZ.3,HuZ.3})
638: \begin{equation}
639: U := \tU - \lambda^{-1} \frac{\tv(0,\lambda)}{v_0} \hat U'(x).
640: \end{equation}
641: Since $\CalL \bU ' = 0$ by differentiation of the boundary-layer equation,
642: the system expressed in the good unknown becomes simply
643: \begin{equation} \label{lin-stability-prob}
644: U_t - \CalL U = 0 \hspace{0.5cm} \mbox{in} \; x<0 ,
645: \end{equation}
646: or, equivalently, \eqref{eigen1}
647: with boundary conditions
648: \begin{equation}
649: \begin{aligned}
650: & U|_{x=0} = \frac{\tv(0,\lambda)}{v_0}(1 - \lambda^{-1}\bv '(0) ,
651: \, -\lambda^{-1} \bu '(0))^{T} \\
652: & \lim_{x \rightarrow +\infty} U = 0.
653: \end{aligned}
654: \end{equation}
655: Solving for $u|_{x=0}$ in terms of $v|_{x=0}$ and
656: recalling that $\hat v'=\hat u'$ by \eqref{profeq},
657: we obtain finally
658: \begin{equation}\label{outflowBC}
659: u|_{x=0} =\alpha(\lambda) v|_{x=0},
660: \qquad
661: \alpha(\lambda):= \frac{-\bv'(0)}{\lambda - \bv '(0)}.
662: \end{equation}
663:
664: \begin{remark}\label{outrmk}
665: Problems \eqref{vectorevalue} and \eqref{lin-stability-prob}--\eqref{eigen1}
666: are evidently equivalent for all $\lambda\ne 0$,
667: but are not equivalent for $\lambda=0$ (for which the change of coordinates
668: to good unknown becomes singular). For, $U=\hat U'$ by inspection
669: is a solution of \eqref{lin-stability-prob}, but is not
670: a solution of \eqref{vectorevalue}.
671: That is, we have introduced by this transformation a spurious
672: eigenvalue at $\lambda=0$, which we shall have to account for later.
673: \end{remark}
674:
675: \subsection{Preliminary estimates}\label{prelimests}
676:
677: \begin{proposition} [\cite{BHRZ}] \label{profdecay}
678: For each $\gamma\ge 1$, $0<v_+\le 1/12<v_0 < 1$,
679: \eqref{profeq} has a unique
680: (up to translation) monotone
681: decreasing solution $\hat v$ decaying to endstates $v_\pm$
682: with a uniform exponential rate for $v_+$ uniformly bounded away from $v_-=1$.
683: In particular, for $0<v_+\le 1/12$,
684: \begin{subequations}
685: \label{decaybd}
686: \begin{align}
687: |\bv(x)-v_+|&\le C e^{-\frac{3(x-\delta)} {4}} \quad x\ge \delta,
688: \label{decaybd_1}\\
689: |\bv(x)-v_-|&\le
690: Ce^{\frac{(x-\delta)}{2}} \quad x\le \delta
691: \label{decaybd_2}
692: \end{align}
693: \end{subequations}
694: where $\delta$ is defined by $\hat v(\delta)=(v_-+v_+)/2$.
695: \end{proposition}
696:
697: \begin{proof}
698: Existence and monotonicity follow trivially
699: by the fact that \eqref{profeq} is a scalar
700: first-order ODE with convex righthand side.
701: Exponential convergence as $x\to +\infty$ follows by
702: $H(v, v_+) =
703: (v-v_+)\Big(v - \Big(\frac{1-v_+}{1-v_+^{\gamma}}\Big)
704: \Big(\frac{1 - \big(\frac{v_+}{v}\big)^{\gamma}}{1 -
705: \big(\frac{v_+}{v}\big)}\Big)\Big),
706: $
707: whence
708: $v- \gamma \le \frac{H(v,v_+)}{v-v_+}\le v-(1-v_+)$
709: by $1\le \frac{1-x^\gamma}{1-x}\le \gamma$
710: for $0\le x\le 1$. Exponential convergence as $x\to -\infty$
711: follows by a similar, but more straightforward calculation,
712: where, in the ``centered'' coordinate $\tilde x:=x-\delta$,
713: the constants $C>0$ are uniform with respect to $v_+, v_0$.
714: See \cite{BHRZ} for details.
715: \end{proof}
716:
717: The following estimates are established in Appendices \ref{basicproof}
718: and \ref{outbasicproof}.
719:
720: \begin{proposition} \label{hf}
721: Nonstable eigenvalues $\lambda$ of \eqref{eigen1}, i.e., eigenvalues
722: with nonnegative real part, are confined for any $0<v_+\le 1$ to the
723: region
724: \begin{equation}
725: \label{hfbounds1} \Lambda:= \{\lambda:\, \R(\lambda) + |\I(\lambda)|
726: \leq \frac 12\Big(2\sqrt{\gamma}+1\Big)^2\}.
727: \end{equation}
728: for the inflow case, and to the region
729: \begin{equation}
730: \label{hfbounds2} \Lambda:= \{\lambda:\, \R(\lambda) + |\I
731: (\lambda)| \le \max\{\frac{3\sqrt2}{2},3\gamma+ \frac{3}{8}\}
732: \end{equation}
733: for the outflow case.
734: \end{proposition}
735: %NOTE: around $10$, so use this as numerical radius.
736:
737: \subsection{Evans function formulation}\label{evanssec}
738: Setting $w:=\frac{u'}{\hat v} +\frac{h(\hat v)}{\hat v^{\gamma+1}}v - u$,
739: we may express \eqref{eigen1} as a first-order system
740: \begin{equation}\label{firstorder}
741: W' = A(x,\lambda) W,
742: \end{equation}
743: where
744: \begin{equation}
745: \label{evans_ode}
746: A(x,\lambda) = \begin{pmatrix}
747: 0 & \lambda & \lambda\\
748: 0 & 0 & \lambda\\
749: \hat v & \hat v & f(\hat v)- \lambda\\ \end{pmatrix},
750: \quad W = \begin{pmatrix}
751: w\\u-v\\v\end{pmatrix},\quad \prime = \frac{d}{dx},
752: \end{equation}
753: where
754: \begin{equation}\label{feq}
755: f(\bv) = \bv- \bv^{-\gamma} h(\bv)
756: = 2\bv - a(\gamma-1)\bv^{-\gamma} - (a+1),
757: \end{equation}
758: with $h$ as in \eqref{f} and $a$ as in \eqref{RH},
759: or, equivalently,
760: \begin{equation}\label{feq2}
761: f(\bv) = 2\bv - (\gamma-1)
762: \Big(\frac{1-v_+}{1-v_+^\gamma}\Big) \Big(\frac{v_+}{\bv}\Big)^{\gamma}
763: - \Big(\frac{1-v_+}{1-v_+^\gamma}\Big) v_+^{\gamma} -1.
764: \end{equation}
765:
766: \begin{remark}\label{shockrel}
767: The coefficient matrix $A$ may be recognized as a rescaled version
768: %CHANGED: more clear-KZ
769: %of the one $\mathcal{A}$ appearing in the shock case \cite{BHRZ,HLZ},
770: of the coefficient matrix
771: $\mathcal{A}$ appearing in the shock case \cite{BHRZ,HLZ},
772: %ENDCHANGED
773: with
774: $$
775: \mathcal{A}=
776: \begin{pmatrix}1&0&0\\0&1&0\\0&0&\lambda\end{pmatrix}
777: A\begin{pmatrix}1&0&0\\0&1&0\\0&0&1/\lambda\end{pmatrix}.
778: $$
779: The choice of variables $(w,u-v,v)^T$ may be recognized as the modified flux
780: form of \cite{PZ}, adapted to the hyperbolic--parabolic case.
781: \end{remark}
782:
783: Eigenvalues of \eqref{eigen1} correspond to nontrivial solutions $W$
784: for which the boundary conditions $W(\pm\infty)=0$ are satisfied.
785: Because $A(x,\lambda)$ as a function of $\bv$ is asymptotically constant
786: in $x$, the behavior near $x=\pm \infty$ of solutions of
787: \eqref{evans_ode} is governed by the limiting constant-coefficient
788: systems
789: \begin{equation}
790: \label{apm}
791: W' = A_\pm(\lambda) W, \qquad
792: A_\pm(\lambda):=A(\pm \infty,\lambda),
793: \end{equation}
794: from which we readily find on the (nonstable) domain $\Re \lambda \ge 0$,
795: $\lambda\ne 0$ of
796: interest that there is a one-dimensional
797: unstable manifold $W_1^-(x)$ of solutions decaying at $x=-\infty$ and
798: a two-dimensional stable manifold $W_2^+(x) \wedge W_3^+(x)$ of
799: solutions decaying at $x=+\infty$, analytic in $\lambda$, with
800: asymptotic behavior
801: \begin{equation}\label{asymptotics}
802: W_j^\pm(x,\lambda) \sim e^{\mu_\pm(\lambda) x} V_j^\pm(\lambda)
803: \end{equation}
804: as $x\to \pm \infty$, where $\mu_\pm(\lambda)$ and $V_j^\pm(\lambda)$
805: are eigenvalues and associated analytically chosen
806: eigenvectors of the limiting coefficient matrices $A_\pm (\lambda)$.
807: %
808: A standard choice of eigenvectors $V_j^\pm$ \cite{GZ,BrZ,BDG,HSZ},
809: uniquely specifying $W_j^\pm$ (up to constant factor) is obtained by
810: Kato's ODE \cite{Kato}, a linear, analytic ODE whose solution
811: can be alternatively characterized by the property
812: that there exist corresponding left eigenvectors $\tilde V_j^\pm$ such that
813: \begin{equation}\label{katoprop}
814: (\tilde V_j\cdot V_j)^\pm \equiv {\rm constant}, \quad
815: (\tilde V_j \cdot \dot V_j)^\pm \equiv 0,
816: \end{equation}
817: where ``$\, \, \dot{ }\, \,$'' denotes $d/d\lambda$;
818: for further discussion, see \cite{Kato,GZ,HSZ}.
819:
820: \subsubsection{Inflow case}\label{inev}
821: In the inflow case, $0\le x\le +\infty$, we define
822: the {\it Evans function} $D$
823: %associated with operator $L$
824: as the analytic function
825: \begin{equation}\label{evansdef}
826: D_{\rm in}(\lambda): = \det(W_1^0, W_2^+, W_3^+)_{\mid x=0},
827: \end{equation}
828: where $W^+_j$ are as defined above, and $W^0_1$ is a solution
829: satisfying the boundary conditions $(v,u)=(0,0)$ at $x=0$,
830: specifically,
831: \begin{equation}\label{w01}
832: W^0_1|_{x=0}=(1,0,0)^T.
833: \end{equation}
834: With this definition, eigenvalues of $\mathcal{L}$
835: correspond to zeroes of $D$ both
836: in location and multiplicity; moreover, the Evans function
837: extends analytically to $\lambda=0$, i.e., to all of $\Re \lambda \ge 0$.
838: See \cite{AGJ,GZ,MZ.3,Z.3} for further details.
839:
840: Equivalently,
841: following \cite{PW,BHRZ}, we may express the Evans function as
842: \begin{equation}\label{adjevans}
843: D_{\rm in}(\lambda)=
844: \big( W_1^0 \cdot \widetilde{W}_1^+ \big)_{\mid x=0},
845: \end{equation}
846: where $\widetilde{W}_1^+(x)$ spans the one-dimensional
847: unstable manifold of solutions decaying at $x=+\infty$
848: (necessarily orthogonal to the span of
849: $W_2^+(x)$ and $W_3^+(x)$)
850: of the adjoint eigenvalue ODE
851: \begin{equation}\label{adjode}
852: \widetilde{W}' =
853: - A(x,\lambda)^* \widetilde{W}.
854: \end{equation}
855: The simpler representation \eqref{adjevans}
856: is the one that we shall use here.
857:
858: \subsubsection{Outflow case}\label{outev}
859: In the outflow case, $-\infty \le x\le 0$, we define
860: the {\it Evans function} as
861: \begin{equation}\label{evansdef2}
862: D_{\rm out}(\lambda): = \det(W_1^-, W_2^0, W_3^0)_{\mid x=0},
863: \end{equation}
864: where $W^-_1$ is as defined above, and $W^0_j$ are a basis
865: of solutions of \eqref{firstorder}
866: satisfying the boundary conditions \eqref{outflowBC},
867: specifically,
868: \begin{equation}\label{w023}
869: W^0_2|_{x=0}=(1,0,0)^T,
870: \qquad
871: W^0_3|_{x=0}=\Big(0, -\frac{\lambda}{\lambda - \bv '(0)}, 1\Big)^T,
872: \end{equation}
873: or, equivalently, as
874: \begin{equation}\label{adjevans2}
875: D_{\rm out}(\lambda)=\big(W_1^-\cdot \widetilde{W}_1^0 \big)_{\mid x=0},
876: \end{equation}
877: where
878: \begin{equation}\label{tildew1}
879: \widetilde{W}^0_1= \Big(0, -1,
880: -\frac{\bar \lambda}{\bar \lambda - \bv '(0)}\Big)^T
881: \end{equation}
882: is the solution of the adjoint eigenvalue ODE dual to $W^0_2$ and $W^0_3$.
883:
884: \begin{remark}\label{outevansrmk}
885: As discussed in Remark \ref{outrmk},
886: $D_{\rm out}$ has a spurious zero at $\lambda=0$
887: introduced by the coordinate change to ``good unknown''.
888: \end{remark}
889:
890: \section{Main results}\label{description}
891:
892: We can now state precisely our main results.
893:
894: \subsection{The strong layer limit}\label{strong}
895: Taking a formal limit as $v_+\to 0$ of the rescaled equations
896: \eqref{rescaled} and recalling that $a\sim v_+^\gamma$, we obtain
897: a {limiting evolution equation}
898: \begin{equation}
899: \begin{split}
900: \label{pressureless}
901: v_t + v_x - u_x &= 0,\\
902: u_t + u_x &= \left(\frac{u_x}{v}\right)_x
903: \end{split}
904: \end{equation}
905: corresponding to a {\it pressureless gas}, or $\gamma=0$.
906:
907:
908: The associated {limiting profile equation} $v' =v(v-1)$
909: has explicit solution
910: \begin{equation}\label{v^0}
911: \hat v^0(x)= \frac{1-\tanh \big(\frac{x-\delta}{2}\big)}{2},
912: \end{equation}
913: $\hat v^0(0)= \frac{1-\tanh (-\delta/2)}{2}= v_0$;
914: the limiting eigenvalue system is
915: $
916: W' = A^0(x,\lambda) W,
917: $
918: \begin{equation}
919: \label{limevans_ode}
920: A^0(x,\lambda) = \begin{pmatrix}0 & \lambda & \lambda\\0 & 0 & \lambda\\
921: \bv^0& \bv^0 &f^0(\bv^0)-\lambda \end{pmatrix},
922: \end{equation}
923: where
924: $f^0(\bv^0) = 2\bv^0 - 1 = -\tanh\big(\frac{x+\delta}{2}\big).$
925:
926: Convergence of the profile and eigenvalue equations is {\it uniform} on any
927: interval $\hat v^0\ge \epsilon >0$, or, equivalently,
928: $x-\delta \le L$, for $L$ any positive constant, where
929: the sequence of coefficient matrices is
930: therefore a {\it regular perturbation} of its limit.
931: Following \cite{HLZ}, we call $x\le L+\delta$ the ``regular region''.
932: For $\hat v_0\to 0$ on the other hand, or $x\to \infty$, the limit
933: is less well-behaved, as may be seen by the fact that $\partial f/\partial \hat v\sim \hat v^{-1}$ as $\hat v\to v_+$, a consequence of the
934: appearance of $\big(\frac{v_+}{\hat v}\big)$ in
935: the expression \eqref{feq2} for $f$.
936: Similarly, $A(x,\lambda)$ does not converge
937: to $A_+(\lambda)$ as $x\to +\infty$ with uniform exponential rate
938: independent of $v_+$, $\gamma$, but rather as $C\hat v^{-1}e^{-x/2}$.
939: As in the shock case, this makes problematic the treatment of
940: $x\ge L+\delta$.
941: Following \cite{HLZ} we call
942: $x\ge L+\delta$ the ``singular region''.
943:
944: To put things in another way, the effects of pressure are not lost
945: as $v_+\to 0$, but rather pushed to $x=+\infty$, where they must
946: be studied by a careful boundary-layer analysis.
947: (Note: this is not a boundary-layer in the same sense
948: as the background solution, nor is it a
949: singular perturbation in the usual sense,
950: at least as we have framed the problem here.)
951:
952: \begin{remark}\label{newparam}
953: A significant difference from the shock case of
954: \cite{HLZ} is the appearance of the second parameter $v_0$
955: that survives in the $v_+\to 0$ limit.
956: \end{remark}
957:
958: \subsubsection{Inflow case}
959: Observe that the limiting coefficient matrix
960: \begin{equation}\label{A0+}
961: \begin{aligned}
962: A^0_+(\lambda) &:=A^0(+\infty, \lambda)
963: =
964: \begin{pmatrix}0 & \lambda & \lambda\\0 & 0 & \lambda\\
965: 0& 0 &-1 -\lambda \end{pmatrix},
966: \end{aligned}
967: \end{equation}
968: is nonhyperbolic (in ODE sense) for all $\lambda$,
969: having eigenvalues $0,0,-1-\lambda$; in particular,
970: the stable manifold drops to dimension one in the limit $v_+\to 0$,
971: and so the prescription of an associated Evans function is
972: {\it underdetermined}.
973:
974: This difficulty is resolved by a careful boundary-layer analysis
975: in \cite{HLZ}, determining a special ``slow stable'' mode
976: $$
977: V_2^+ \pm(1,0,0)^T
978: $$
979: augmenting the ``fast stable'' mode
980: $$
981: V_3:=(\lambda/\mu)(\lambda/\mu+1), \lambda/\mu, 1)^T
982: $$
983: associated with the single stable eigenvalue
984: $\mu=-1-\lambda$ of $A^0_+$.
985: This determines a {\it limiting Evans function} $D^0_{\rm in}(\lambda)$
986: by the prescription \eqref{evansdef}, \eqref{asymptotics} of
987: Section \ref{evanssec},
988: or alternatively via \eqref{adjevans} as
989: \begin{equation}\label{duallimD}
990: D^0_{\rm in}(\lambda)=
991: \big(W_1^{00}\cdot \widetilde{W}_1^{0+} \big)_{\mid x=0},
992: \end{equation}
993: with $\widetilde{W}_1^{0+}$ defined analogously as a solution of
994: the adjoint limiting system
995: lying asymptotically at $x=+\infty$ in direction
996: \begin{equation}\label{tildeV}
997: \widetilde V_1:=
998: (0, -1, \bar \lambda/\bar \mu)^T
999: \end{equation}
1000: orthogonal to the span of $V_2$ and $V_3$,
1001: where `` $\bar{ }$ '' denotes complex conjugate,
1002: and $W^{00}_1$ defined as the solution of the limiting eigenvalue
1003: equations satisfying boundary condition \eqref{w01},
1004: i.e., $(W_1^{00})_{\mid x=0}= (1,0,0)^T$.
1005:
1006: \subsubsection{Outflow case}
1007: We have no such difficulties in the outflow case, since
1008: $A^0_-=A^0(-\infty)$ remains uniformly hyperbolic,
1009: and we may define a limiting Evans function $D^0_{\rm out}$ directly by
1010: \eqref{evansdef2}, \eqref{asymptotics}, \eqref{tildew1},
1011: at least so long as $v_0$ remains bounded from zero.
1012: (As perhaps already hinted by Remark \ref{newparam}, there are complications
1013: associated with the double limit $(v_0,v_+)\to (0,0)$.)
1014:
1015: \subsection{Analytical results}\label{analytical}
1016: With the above definitions, we have the following main theorems
1017: characterizing the strong-layer limit $v_+\to 0$ as well as the
1018: limits $v_0\to 0, \, 1$.
1019:
1020:
1021: \begin{theorem}\label{mainthm}
1022: For $v_0\ge \eta>0$ and $\lambda$ in any compact subset
1023: of $\Re \lambda \ge 0$,
1024: $D_{\rm in}(\lambda)$ and $D_{\rm out}(\lambda)$ converge
1025: uniformly to $D^0_{\rm in}(\lambda)$ and $D^0_{\rm out}(\lambda)$
1026: as $v_+\to 0$.
1027: \end{theorem}
1028:
1029: %TODO: Put in linking result about stable if limit is? NO.
1030:
1031: \begin{theorem}\label{main3}
1032: For $\lambda$ in any compact subset of $\Re \lambda \ge 0$
1033: and $v_+$ bounded from $1$,
1034: $D_{\rm in}(\lambda)$, appropriately renormalized by a nonvanishing
1035: analytic factor,
1036: converges uniformly as $v_0\to 1$ to the Evans function for the (unintegrated)
1037: eigenvalue equations of the associated viscous shock wave connecting
1038: $v_-=1$ to $v_+$;
1039: likewise, $D^0_{\rm out}(\lambda)$, appropriately renormalized,
1040: converges uniformly as $v_0\to 0$ to the same limit for $\lambda$ uniformly
1041: bounded away from zero.
1042: \end{theorem}
1043:
1044: By similar computations, we obtain also the following direct result.
1045:
1046: \begin{theorem}\label{main2}
1047: Inflow boundary layers are stable for $v_0$ sufficiently small.
1048: %NOTE: independent of $v_+$.
1049: \end{theorem}
1050:
1051: We have also the following parity information, obtained by
1052: stability-index computations as in \cite{SZ}.\footnote{Indeed,
1053: these may be deduced from the results of \cite{SZ}, taking account
1054: of the difference between Eulerian and Lagrangian coordinates.}
1055:
1056: \begin{lemma}[Stability index]\label{index}
1057: For any $\gamma\ge 1$, $v_0$, and $v_+$, $D_{\rm in}(0)\ne 0$,
1058: hence the number of unstable roots of $D_{\rm in}$ is even;
1059: on the other hand $D^0_{\rm in}(0)=0$
1060: and $\lim_{v_0\to 0}D^0_{\rm in}(\lambda)\equiv 0$.
1061: Likewise, $(D^0_{\rm in})'(0)$, $D_{\rm out}'(0)\ne
1062: 0$, $(D^0_{\rm out})'(0)\ne 0$, hence the number of nonzero unstable
1063: roots of $D^0_{\rm in}$, $D_{\rm out}$, $D^0_{\rm out}$ is even.
1064: \end{lemma}
1065:
1066: Finally, we have the following auxiliary results established
1067: by energy estimates in Appendices \ref{stronglimit}, \ref{outnonv},
1068: \ref{char}, and \ref{nonvanish-expansive-inflow}.
1069:
1070: \begin{proposition}\label{redenergy}
1071: The limiting Evans function $D^0_{\rm in}$ is nonzero for
1072: $\lambda \ne 0$ on $\R \lambda \ge 0$, for all $1>v_0>0$.
1073: The limiting Evans function $D^0_{\rm out}$ is nonzero for
1074: $\lambda \ne 0$ on $\R \lambda \ge 0$, for $1>v_0> v_*$,
1075: where $v_* \approx 0.0899$ is determined by the functional equation
1076: $v_*= e^{-2/(1-v_*)^2}$.
1077: \end{proposition}
1078:
1079: \begin{proposition} \label{charsmallamp}
1080: Compressive outflow boundary layers are stable for $v_+$
1081: sufficiently close to $1$.
1082: \end{proposition}
1083:
1084: \begin{proposition}[\cite{MN.2}] \label{expansive}
1085: Expansive inflow boundary layers are stable for all
1086: %$\gamma \ge 1$, $ v_+ \ge 1$, $1>v_0>v_+$.
1087: parameter values.
1088: \end{proposition}
1089:
1090: Collecting information, we have the following analytical stability results.
1091:
1092: \begin{corollary}\label{v01}
1093: For $v_0$ or $v_+$ sufficiently small, compressive
1094: inflow boundary layers are stable.
1095: For $v_0$ sufficiently small,
1096: $v_+$ sufficiently close to $1$,
1097: or $v_0> v_*\approx .0899$ and
1098: $v_+$ sufficiently small, compressive
1099: outflow layers are stable.
1100: Expansive inflow boundary layers are stable
1101: for all parameter values.
1102: \end{corollary}
1103:
1104:
1105: Stability of inflow boundary layers
1106: in the characteristic limit $v_+\to 1$ is not treated here,
1107: but should be treatable analytically by the asymptotic ODE methods
1108: used in \cite{PZ,FS} to study the small-amplitude (characteristic)
1109: shock limit.
1110: This would be an interesting direction for future investigation.
1111: The characteristic limit is not accessible numerically, since the
1112: exponential decay rate of the background profile
1113: decays to zero as $|1-v_+|$, so that the numerical
1114: domain of integration needed to resolve the eigenvalue
1115: ODE becomes infinitely large as $v_+\to 1$.
1116:
1117: \begin{remark}\label{weakreg}
1118: Stability in the noncharacteristic weak layer limit $v_0\to v_+$ [resp. $1$]
1119: in the inflow [outflow] case, for $v_+$ bounded away from the
1120: strong and characteristic limits $0$ and $1$ has already
1121: been established in \cite{GMWZ.5,R}.
1122: Indeed, it is shown in \cite{GMWZ.5} that the Evans function
1123: converges to that for a constant solution, and this is a {\it regular}
1124: perturbation.
1125: %(TODO: compute the limiting function with Kato basis- anything nice?)
1126: \end{remark}
1127:
1128: \begin{remark}\label{extra}
1129: Stability of $D^0_{\rm in}$, $D^0_{\rm out}$ may also be
1130: determined numerically, in particular in the
1131: region $v_0\le v_*$ not covered by Proposition \ref{redenergy}.
1132: %CHANGED: deleted, unclear-KZ
1133: %(Recall that the limiting shock case $v_0\to 0$ is treated elsewhere;
1134: %see also Remark \ref{smallcase}.)
1135: %ENDCHANGED
1136: \end{remark}
1137:
1138:
1139: \subsection{Numerical results}\label{numresults}
1140: The asymptotic results of Section \ref{analytical} reduce
1141: the problem of
1142: %TODO: remove disclaimer? NO, OK.
1143: (uniformly noncharacteristic, $v_+$ bounded away from $v_-=1$)
1144: %
1145: boundary layer stability to a bounded parameter
1146: range on which the Evans function may be efficiently computed
1147: numerically in a way that is uniformly well-conditioned; see \cite{BrZ}.
1148: Specifically, we may map a semicircle
1149: $$
1150: \partial\{\Re \lambda \ge 0\}\cap \{|\lambda|\le 10\}
1151: $$
1152: enclosing $\Lambda$ for $\gamma\in [1,3]$ by
1153: $D^0_{\rm in}$, $D^0_{\rm out}$, $D_{\rm in}$, $D_{\rm out}$ and
1154: compute the winding number of its image about the origin
1155: to determine the number of zeroes of the various Evans functions
1156: within the semicircle, and thus within $\Lambda$.
1157: For details of the numerical algorithm, see \cite{BHRZ,BrZ}.
1158:
1159: In all cases, we obtain results consistent with {stability};
1160: that is, a winding number of zero or one, depending on the situation.
1161: In the case of a single nonzero root, we know from our limiting
1162: analysis that this root may be quite near $\lambda=0$, making
1163: delicate the direct determination of its stability; however,
1164: in this case we do not attempt to determine the stability numerically,
1165: but rely on the analytically computed stability index to conclude
1166: stability.
1167: See Section \ref{computations} for further details.
1168:
1169: %TODO: or, put some pictures already here? NO, NOT NEEDED.
1170:
1171: \subsection{Conclusions}\label{conclusions}
1172: As in the shock case \cite{BHRZ,HLZ}, our results indicate
1173: {\it unconditional stability} of uniformly noncharacteristic
1174: boundary-layers for isentropic
1175: Navier--Stokes equations (and, for outflow layer, in
1176: the characteristic limit as well),
1177: despite the additional complexity of the boundary-layer case.
1178: However, two additional comments are in order, perhaps related.
1179: First, we point out that the apparent symmetry of Theorem
1180: \ref{main3} in the $v_0\to 0$ outflow and $v_0\to 1$ inflow limits
1181: is somewhat misleading.
1182: For, the limiting, shock Evans function possesses a single zero
1183: at $\lambda=0$, indicating that stability of inflow boundary
1184: layers is somewhat delicate as $v_0\to 1$: specifically, they
1185: have an eigenvalue near zero, which, though stable, is (since
1186: vanishingly small in the shock limit) not ``very'' stable.
1187: Likewise, the limiting Evans function $D^0_{\rm in}$ as $v_+\to 0$
1188: possesses a zero at $\lambda=0$, with the same conclusions.
1189:
1190: By contrast, the Evans functions of outflow boundary layers
1191: possess a spurious zero at $\lambda=0$, so that convergence to
1192: the shock or strong-layer limit in this case implies the {\it absence}
1193: of any eigenvalues near zero, or ``uniform'' stability as $v_+\to 0$.
1194: In this sense, strong outflow boundary layers appear to be more stable
1195: than inflow boundary layers.
1196: One may make interesting comparisons to physical attempts
1197: to stabilize laminar flow along an air- or hydro-foil by suction (outflow)
1198: along the boundary.
1199: See, for example, the interesting treatise \cite{S}.
1200:
1201: Second, we point out the result of {\it instability} obtained
1202: in \cite{SZ} for inflow boundary-layers of the full (nonisentropic)
1203: ideal-gas equations for appropriate ratio of the coefficients
1204: of viscosity and heat conduction.
1205: This suggests that the small eigenvalues of the strong inflow-layer
1206: limit may in some cases perturb to the unstable side.
1207: It would be very interesting to make these connections more precise,
1208: as we hope to do in future work.
1209:
1210: \section{Boundary-layer analysis}\label{singular}
1211:
1212: Since the structure of \eqref{evans_ode} is essentially the same
1213: as that of the shock case, we may follow exactly the treatment
1214: in \cite{HLZ} analyzing the flow of \eqref{evans_ode}
1215: in the singular region $x\to +\infty$.
1216: As we shall need the details for further computations
1217: %CHANGED: added this- KZ
1218: (specifically, the proof of Theorem \ref{main2}),
1219: %ENDCHANGED
1220: we repeat the analysis here in full.
1221:
1222: Our starting point is the observation that
1223: \begin{equation}
1224: \label{a+}
1225: A(x,\lambda) = \begin{pmatrix}0 & \lambda & \lambda\\0 & 0 & \lambda\\
1226: \hat v& \hat v &f(\hat v)-\lambda \end{pmatrix}
1227: \end{equation}
1228: is approximately block upper-triangular for $\hat v$ sufficiently small,
1229: with diagonal blocks
1230: $\begin{pmatrix}
1231: 0 & \lambda \\
1232: 0 & 0\\
1233: \end{pmatrix}$
1234: and
1235: $\begin{pmatrix}
1236: f(\hat v)-\lambda
1237: \end{pmatrix}$
1238: that are uniformly spectrally separated on $\R \lambda \ge 0$,
1239: as follows by
1240: \begin{equation}\label{fneg}
1241: f(\hat v)\le \hat v-1 \le -3/4.
1242: \end{equation}
1243: We exploit this structure by a judicious coordinate change
1244: converting \eqref{evans_ode}
1245: to a system in exact upper triangular form, for which the
1246: decoupled ``slow'' upper lefthand $2\times 2$ block undergoes
1247: a {\it regular perturbation} that can be analyzed by standard
1248: tools introduced in \cite{PZ}.
1249: Meanwhile, the fast, lower righthand $1\times 1$ block, since
1250: scalar, may be solved exactly.
1251:
1252: \subsection{Preliminary transformation}\label{pretrans}
1253: We first block upper-triangularize by a static (constant) coordinate
1254: transformation the limiting matrix
1255: \begin{equation}
1256: \label{lima+}
1257: A_+=A(+\infty,\lambda) = \begin{pmatrix}0 & \lambda & \lambda\\0 & 0 & \lambda\\
1258: v_+& v_+ &f(v_+)-\lambda \end{pmatrix}
1259: \end{equation}
1260: at $x=+\infty$ using special block lower-triangular transformations
1261: \begin{equation}\label{statictrans}
1262: R_+:=\begin{pmatrix}
1263: I& 0\\
1264: v_+ \theta_+ & 1\\
1265: \end{pmatrix},
1266: \qquad
1267: L_+:=R_+^{-1}=\begin{pmatrix}
1268: I& 0\\
1269: -v_+\theta_+ & 1\\
1270: \end{pmatrix},
1271: \end{equation}
1272: where $I$ denotes the $2\times 2$ identity matrix and
1273: $\theta_+\in \C^{1\times 2}$ is a $1\times 2$ row vector.
1274:
1275: \begin{lemma}\label{pretranslem}
1276: On any compact subset of $\R \lambda \ge 0$, for each $v_+>0$
1277: sufficiently small,
1278: there exists a unique $\theta_+=\theta_+(v_+,\lambda)$ such that
1279: $\hat A_+:=L_+A_+R_+$ is upper block-triangular,
1280: \begin{equation}\label{hata+}
1281: \begin{aligned}
1282: \hat A_+&=
1283: \begin{pmatrix}
1284: \lambda (J+ v_+\bb1 \theta_+) & \lambda \bb1 \\
1285: 0 & f(v_+)-\lambda -\lambda v_+ \theta_+\bb1 \\
1286: \end{pmatrix},
1287: \end{aligned}
1288: \end{equation}
1289: where $J=\begin{pmatrix} 0 & 1 \\ 0 & 0 \end{pmatrix}$
1290: and
1291: $\bb1=\begin{pmatrix} 1 \\ 1 \\ \end{pmatrix} $,
1292: satisfying a uniform bound
1293: \begin{equation}\label{theta+bd}
1294: |\theta_+|\le C.
1295: \end{equation}
1296: \end{lemma}
1297:
1298: \begin{proof}
1299: Setting the $2-1$ block of $\hat A_+$ to zero, we obtain the
1300: matrix equation
1301: $$
1302: \theta_+ (aI-\lambda J)
1303: = -\bb1^T + \lambda v_+ \theta_+ \bb1 \theta_+,
1304: $$
1305: where $a=f(v_+)-\lambda$, or, equivalently, the fixed-point equation
1306: \begin{equation}\label{fix1}
1307: \theta_+ =
1308: (aI-\lambda J)^{-1}
1309: \Big( -\bb1^T + \lambda v_+\theta_+ \bb1 \theta_+\Big).
1310: \end{equation}
1311: By $\det (aI-\lambda J)= a^2\ne 0$,
1312: $(aI-\lambda J)^{-1}$ is uniformly bounded
1313: on compact subsets of $\R \lambda \ge 0$
1314: (indeed, it is uniformly bounded on all of $\R \lambda \ge 0$),
1315: whence, for $|\lambda|$ bounded and $v_+$ sufficiently small,
1316: there exists a unique
1317: solution by the Contraction Mapping Theorem,
1318: which, moreover, satisfies \eqref{theta+bd}.
1319: \end{proof}
1320:
1321: \subsection{Dynamic triangularization}\label{dynamic}
1322: Defining now $Y:=L_+W$ and
1323: $$
1324: \begin{aligned}
1325: &\hat A(x,\lambda)= L_+A(x, \lambda) R_+(x,\lambda)=\\
1326: &\quad \begin{pmatrix}\\
1327: \lambda (J+v_+\bb1 \theta_+) &\quad & \lambda \bb1 \\
1328: (\hat v-v_+) \bb1^T
1329: -v_+(f(\hat v)-f(v_+))\theta_+
1330: & \quad& f(\hat v)-\lambda -\lambda v_+ \theta_+\bb1 \\
1331: \end{pmatrix},
1332: \end{aligned}
1333: $$
1334: we have converted \eqref{evans_ode} to an asymptotically block
1335: upper-triangular system
1336: \begin{equation}\label{tri1}
1337: Y'=\hat A(x,\lambda) Y,
1338: \end{equation}
1339: with $\hat A_+=\hat A(+\infty, \lambda)$ as in \eqref{hata+}.
1340: %
1341: Our next step is to choose a {\it dynamic} transformation of
1342: the same form
1343: \begin{equation}\label{dyntrans}
1344: \tilde R:=\begin{pmatrix}
1345: I& 0\\
1346: \tilde \Theta & 1\\
1347: \end{pmatrix},
1348: \qquad
1349: \tilde L:=\tilde R^{-1}=\begin{pmatrix}
1350: I& 0\\
1351: -\tilde \Theta & 1\\
1352: \end{pmatrix},
1353: \end{equation}
1354: converting \eqref{tri1} to an exactly block upper-triangular
1355: system, with $\tilde \Theta$ uniformly exponentially decaying
1356: at $x=+\infty$: that is, a {\it regular perturbation} of the identity.
1357:
1358: \begin{lemma}\label{dyntranslem}
1359: On any compact subset of $\R \lambda \ge 0$,
1360: for $L$ sufficiently large and each $v_+>0$ sufficiently small,
1361: there exists a unique $\Theta=\Theta_+(x,\lambda, v_+)$ such that
1362: $\tilde A:=\tilde L \hat A(x,\lambda)\tilde R
1363: + \tilde L'\tilde R$ is upper block-triangular,
1364: \begin{equation}\label{ahat}
1365: \begin{aligned}
1366: \tilde A&=
1367: \begin{pmatrix}
1368: \lambda (J+v_+\bb1 \theta_+ + \bb1 \tilde \Theta) & \lambda \bb1 \\
1369: 0 & f(\hat v)-\lambda -\lambda \theta_+\bb1 -\lambda \tilde \Theta \bb1 \\
1370: \end{pmatrix},
1371: \end{aligned}
1372: \end{equation}
1373: and
1374: $\tilde \Theta(L)=0$, satisfying a uniform bound
1375: \begin{equation}\label{Thetabd}
1376: |\tilde \Theta(x,\lambda, v_+)|\le Ce^{-\eta x},
1377: \qquad \eta>0, \, x\ge L,
1378: \end{equation}
1379: independent of the choice of $L$, $v_+$.
1380: \end{lemma}
1381:
1382: \begin{proof}
1383: Setting the $2-1$ block of $\tilde A$ to zero and computing
1384: $$
1385: \tilde L'\tilde R=
1386: \begin{pmatrix}
1387: 0 & 0\\
1388: -\tilde \Theta' & 0
1389: \end{pmatrix}
1390: \begin{pmatrix}
1391: I & 0\\
1392: \tilde \Theta & I
1393: \end{pmatrix}
1394: =
1395: \begin{pmatrix}
1396: 0 & 0\\
1397: -\tilde \Theta' & 0,
1398: \end{pmatrix}
1399: $$
1400: we obtain the matrix equation
1401: \begin{equation}\label{mat}
1402: \tilde \Theta' - \tilde \Theta \big(aI-\lambda (J +v_+\bb1\theta_+)\big)
1403: = \zeta+ \lambda\tilde \Theta \bb1 \tilde \Theta,
1404: \end{equation}
1405: where $a(x):= f(\hat v)-\lambda -\lambda v_+ \theta_+\bb1 $
1406: and the forcing term
1407: $$
1408: \zeta:=
1409: -(\hat v-v_+) \bb1^T
1410: +v_+(f(\hat v)-f(v_+))\theta_+
1411: $$
1412: by derivative estimate $df/d\hat v\le C\hat v^{-1}$
1413: together with the Mean Value Theorem
1414: is uniformly exponentially decaying:
1415: \begin{equation}\label{zetabd}
1416: \begin{aligned}
1417: |\zeta|\le C |\hat v-v_+|\le
1418: C_2 e^{-\eta x},
1419: \qquad
1420: \eta>0.
1421: \end{aligned}
1422: \end{equation}
1423:
1424: Initializing $\tilde \Theta(L)=0$, we obtain by Duhamel's Principle/Variation
1425: of Constants the representation (supressing the argument $\lambda$)
1426: \begin{equation}\label{duhamel}
1427: \tilde \Theta(x)=
1428: \int_L^{x}
1429: S^{y\to x}
1430: (\zeta+ \lambda\tilde \Theta \bb1 \tilde \Theta)(y)
1431: \, dy,
1432: \end{equation}
1433: where $S^{y\to x}$ is the solution operator for the homogeneous
1434: equation
1435: $$
1436: \tilde \Theta' - \tilde \Theta \big(aI-\lambda (J +v_+\bb1\theta_+)\big)=0,
1437: $$
1438: or, explicitly,
1439: $$
1440: S^{y\to x}=
1441: e^{\int_y^x a(y)dy}
1442: e^{ -\lambda (J +v_+\bb1\theta_+)(x-y)}.
1443: $$
1444:
1445: For $|\lambda|$ bounded and $v_+$ sufficiently small,
1446: we have by matrix perturbation theory
1447: that the eigenvalues of $ -\lambda (J +v_+\bb1\theta_+)$
1448: are small and the entries are bounded, hence
1449: $$
1450: |e^{ -\lambda (J +v_+\bb1\theta_+)z}|\le Ce^{\epsilon z}
1451: $$
1452: for $z\ge 0$. Recalling the uniform spectral gap
1453: $\R a =f(\hat v)-\R \lambda \le -1/2$ for $\R \lambda \ge 0$,
1454: we thus have
1455: \begin{equation}\label{Sbd}
1456: |S^{y\to x}|\le Ce^{\eta (x-y)}
1457: \end{equation}
1458: for some $C$, $\eta>0$.
1459: Combining \eqref{zetabd} and \eqref{Sbd}, we obtain
1460: \begin{equation}
1461: \begin{aligned}
1462: \Big|\int_L^{x} S^{y\to x} \zeta(y)\, dy\Big|&\le
1463: \int_L^x
1464: C_2 e^{-\eta(x-y)}e^{-(\eta/2) y} dy \\
1465: & =C_3 e^{-(\eta/2)x}.
1466: \end{aligned}
1467: \end{equation}
1468:
1469: Defining $\tilde \Theta(x) =:\tilde \theta(x) e^{-(\eta/2)x}$
1470: and recalling \eqref{duhamel} we thus have
1471: \begin{equation}\label{duhamel2}
1472: \tilde \theta(x)=
1473: f + e^{(\eta/2)x}\int_L^{x}
1474: S^{y\to x}
1475: e^{-\eta y}\lambda\tilde\theta \bb1 \tilde \theta(y)
1476: \, dy,
1477: \end{equation}
1478: where $f:= e^{(\eta/2)x}\int_L^{x} S^{y\to x} \zeta(y) \, dy$
1479: is uniformly bounded, $|f|\le C_3$, and
1480: $
1481: e^{(\eta/2)x}\int_L^{x} S^{y\to x}
1482: e^{-\eta y}\lambda\tilde \theta \bb1 \tilde \theta(y)
1483: \, dy
1484: $
1485: is contractive with arbitrarily small contraction constant $\epsilon>0$
1486: in $L^\infty[L,+\infty)$ for $|\tilde \theta|\le 2C_3$ for $L$ sufficiently
1487: large, by the calculation
1488: $$
1489: \begin{aligned}
1490: &\Big| e^{(\eta/2)x}\int_L^{x} S^{y\to x}
1491: e^{-\eta y}\lambda\tilde \theta_1 \bb1 \tilde \theta_1(y)
1492: -
1493: e^{(\eta/2)x}\int_L^{x} S^{y\to x}
1494: e^{-\eta y}\lambda\tilde \theta_2 \bb1 \tilde \theta_2(y)
1495: \Big|
1496: \\
1497: & \qquad \qquad \le
1498: \Big|e^{(\eta/2)x}\int_L^{x} Ce^{-\eta(x-y)} e^{-\eta y} \, dy \Big|
1499: |\lambda|
1500: \|\tilde \theta_1- \tilde \theta_2\|_\infty
1501: \max_j \|\tilde \theta_j\|_\infty
1502: \\
1503: & \qquad \qquad \le
1504: e^{-(\eta/2)L}\Big|\int_L^{x} Ce^{-(\eta/2)(x-y)} \, dy \Big|
1505: |\lambda|
1506: \|\tilde \theta_1- \tilde \theta_2\|_\infty
1507: \max_j \|\tilde \theta_j\|_\infty \\
1508: & \qquad \qquad =
1509: C_3e^{-(\eta/2)L}
1510: |\lambda|
1511: \|\tilde \theta_1- \tilde \theta_2\|_\infty
1512: \max_j \|\tilde \theta_j\|_\infty.
1513: \end{aligned}
1514: $$
1515: It follows by the Contraction Mapping Principle that there exists
1516: a unique solution $\tilde \theta$ of fixed point equation
1517: \eqref{duhamel2} with $|\tilde \theta(x)|\le 2C_3$
1518: for $x\ge L$,
1519: or, equivalently (redefining the unspecified constant $\eta$), \eqref{Thetabd}.
1520: \end{proof}
1521:
1522: \subsection{Fast/Slow dynamics}\label{slow}
1523: Making now the further change of coordinates
1524: $$
1525: Z=\tilde LY
1526: $$
1527: and computing
1528: $$
1529: \begin{aligned}
1530: (\tilde LY)'=\tilde L Y' + \tilde L' Y
1531: &=(\tilde LA_++\tilde L')Y,\\
1532: &=(\tilde LA_+\tilde R+\tilde L'\tilde R)Z,\\
1533: \end{aligned}
1534: $$
1535: we find that we have converted \eqref{tri1} to a block-triangular system
1536: \begin{equation} \label{tri2}
1537: Z'=\tilde AZ=
1538: \begin{pmatrix}
1539: \lambda (J+v_+\bb1 \theta_+ + \bb1 \tilde \Theta) & \lambda\bb1 \\
1540: 0 & f(\hat v)-\lambda -\lambda v_+ \theta_+\bb1 -\lambda\tilde \Theta \bb1 \\
1541: \end{pmatrix}Z,
1542: \end{equation}
1543: related to the original eigenvalue system \eqref{evans_ode} by
1544: \begin{equation}\label{WZ}
1545: W=LZ,\quad
1546: R:=R_+R=
1547: \begin{pmatrix}
1548: I & 0\\
1549: \Theta & 1
1550: \end{pmatrix},
1551: \quad
1552: L:=R^{-1}=
1553: \begin{pmatrix}
1554: I & 0\\
1555: -\Theta & 1
1556: \end{pmatrix},
1557: \end{equation}
1558: where
1559: \begin{equation}\label{Theta}
1560: \Theta= \tilde \Theta + v_+ \theta_+.
1561: \end{equation}
1562:
1563: Since it is triangular, \eqref{tri2} may be solved completely
1564: if we can solve the component systems associated with its diagonal
1565: blocks. The {\it fast system}
1566: $$
1567: z'=
1568: \Big(f(\hat v)-\lambda -\lambda v_+ \theta_+\bb1 -
1569: \lambda \tilde \Theta \bb1 \Big)z
1570: $$
1571: associated to the lower righthand block features rapidly-varying
1572: coefficients. However, because it is scalar,
1573: it can be solved explicitly by exponentiation.
1574:
1575: The {\it slow system }
1576: \begin{equation} \label{slowsys}
1577: z'= \lambda(J+v_+\bb1 \theta_+ + \bb1 \tilde \Theta ) z
1578: \end{equation}
1579: associated to the upper lefthand block, on the other hand,
1580: by \eqref{Thetabd}, is an exponentially decaying
1581: perturbation of a constant-coefficient system
1582: \begin{equation}\label{cc}
1583: z'= \lambda (J+v_+\bb1 \theta_+)z
1584: \end{equation}
1585: that can be explicitly solved by exponentiation, and thus
1586: can be well-estimated by comparison with \eqref{cc}.
1587: A rigorous version of this statement is given by the
1588: {\it conjugation lemma} of \cite{MeZ}:
1589:
1590: \begin{proposition}[\cite{MeZ}] \label{conjugation}
1591: Let $M(x,\lambda)=M_+(\lambda)+ \Theta(x,\lambda)$,
1592: with $M_+$ continuous in $\lambda$ and $|\Theta(x,\lambda)|\le Ce^{-\eta x}$,
1593: for $\lambda$ in some compact set $\Lambda$.
1594: Then, there exists a globally invertible matrix
1595: $P(x,\lambda)=I + Q(x,\lambda)$ such that the
1596: coordinate change $z=Pv$ converts the variable-coefficient ODE
1597: $
1598: z'=M(x,\lambda)z
1599: $
1600: to a constant-coefficient equation
1601: $$
1602: v'=M_+(\lambda)v,
1603: $$
1604: satisfying for any $L$, $0<\hat \eta < \eta$ a uniform bound
1605: \begin{equation}\label{qdecay}
1606: |Q(x,\lambda)|\le
1607: C(L,\hat \eta, \eta, \max |(M_+)_{ij}|, \dim M_+)e^{-\hat \eta x}
1608: \quad \hbox{for $x\ge L$}.
1609: \end{equation}
1610: \end{proposition}
1611:
1612: \begin{proof}
1613: See \cite{MeZ,Z.3}, or Appendix C, \cite{HLZ}.
1614: \end{proof}
1615:
1616: By Proposition \ref{conjugation}, the solution operator for \eqref{slowsys}
1617: is given by
1618: \begin{equation}\label{slowsoln}
1619: P(y,\lambda) e^{\lambda (J+v_+\bb1 \theta_+(\lambda, v_+))(x-y)}
1620: P(x,\lambda)^{-1},
1621: \end{equation}
1622: where $P$ is a uniformly small perturbation of the identity
1623: for $x\ge L$ and $L>0$ sufficiently large.
1624:
1625:
1626: \section{Proof of the main theorems}\label{proofsec}
1627:
1628: With these preparations, we turn now to the proofs of the main theorems.
1629:
1630: \subsection{Boundary estimate}\label{estW1}
1631: We begin by recalling the following estimates established in
1632: \cite{HLZ} on $\widetilde W_1^+(L+\delta)$, that is, the value of
1633: the dual mode $\widetilde W_1^+$ appearing in \eqref{adjevans}
1634: at the boundary $x=L+\delta$ between regular and singular regions.
1635: For completeness, and because we shall need the details in further
1636: computations, we repeat the proof in full.
1637:
1638: \begin{lemma}[\cite{HLZ}]\label{matching}
1639: For $\lambda$ on any compact subset of $\R \lambda \ge 0$,
1640: and $L>0$ sufficiently large,
1641: with $\widetilde W_1^+$ normalized as in \cite{GZ,PZ,BHRZ},
1642: \begin{equation}\label{wcon}
1643: |\widetilde W_1^+(L+\delta)-\widetilde V_1| \le Ce^{-\eta L}
1644: \end{equation}
1645: as $v_+\to 0$, uniformly in $\lambda$, where $C$, $\eta>0$ are
1646: independent of $L$
1647: %NOTE: clearly not uniform in $L$..
1648: and
1649: $$
1650: \widetilde V_1:= (0, -1, \lambda/\mu)^T
1651: $$
1652: is the limiting direction vector \eqref{tildeV}
1653: appearing in the definition of $D^0_{\rm in}$.
1654: \end{lemma}
1655:
1656: \begin{corollary}[\cite{HLZ}]\label{matching2}
1657: Under the hypotheses of Lemma \ref{matching},
1658: \begin{equation}\label{limcon}
1659: |\tilde W_1^{0+}(L+\delta)-\widetilde V_1| \le Ce^{-\eta L}
1660: \end{equation}
1661: and
1662: \begin{equation}\label{wcon2}
1663: |\widetilde W^{+}_1(L+\delta) -\widetilde W^{0+}_1(L+\delta)|\le Ce^{-\eta L}
1664: \end{equation}
1665: as $v_+\to 0$, uniformly in $\lambda$, where $C$, $\eta>0$ are
1666: independent of $L$ and $\widetilde W_1^{0+}$ is the solution
1667: of the limiting adjoint eigenvalue system
1668: appearing in definition \eqref{duallimD} of $D^0$.
1669: \end{corollary}
1670:
1671: \begin{proof}[Proof of Lemma \ref{matching}]
1672: First, make the independent coordinate change $x\to x-\delta$
1673: normalizing the background wave to match the shock-wave case.
1674: Making the dependent coordinate-change
1675: \begin{equation}\label{dualL}
1676: \tilde Z:= R^* \tilde W,
1677: \end{equation}
1678: $R$ as in \eqref{WZ}, reduces the adjoint equation
1679: $\tilde W'=-A^*\tilde W$ to block lower-triangular form,
1680: %
1681: \begin{equation} \label{dualtri2}
1682: \begin{aligned}
1683: &\tilde Z'=-\tilde A^* \tilde Z=\\
1684: &\,
1685: \begin{pmatrix}
1686: -\bar \lambda( J^T +
1687: v_+\bb1 \theta_+ + \bb1 \tilde \Theta)^* & 0 \\
1688: -\bar\lambda \bb1^T &
1689: -f(\hat v)+\bar \lambda +\bar \lambda v_+ (\theta_+\bb1
1690: + \tilde \Theta \bb1)^* \\
1691: \end{pmatrix}Z,
1692: \end{aligned}
1693: \end{equation}
1694: with `` $\bar{ }$ '' denoting complex conjugate.
1695:
1696: Denoting by $\tilde V^+_1$ a suitably normalized
1697: element of the one-dimensional (slow) stable subspace
1698: of $-\tilde A^*$, we find readily
1699: (see \cite{HLZ} for further discussion)
1700: that, without loss of generality,
1701: \begin{equation}\label{dualVlim}
1702: \tilde V^+_1 \to (0, 1, \bar \lambda (\gamma+\bar \lambda)^{-1})^T
1703: \end{equation}
1704: %TODO: should really be negative of this to match earlier section,
1705: %but no harm in the analysis.. (FIX? Or, just explain that makes
1706: %no difference? -KZ)
1707: as $v_+\to 0$, while the associated eigenvalue $\tilde \mu_1^+\to 0,$
1708: uniformly for $\lambda$ on an compact subset of $\R \lambda\ge 0$.
1709: The dual mode $\tilde Z_1^+=R^* \tilde W_1^+$ is uniquely determined
1710: by the property that it is asymptotic as $x\to +\infty$
1711: to the corresponding constant-coefficient solution
1712: $e^{\tilde \mu_1^+}\tilde V_1^+$
1713: (the standard normalization of \cite{GZ,PZ,BHRZ}).
1714:
1715: By lower block-triangular form \eqref{dualtri2}, the equations
1716: for the slow variable $\tilde z^T:=(\tilde Z_1, \tilde Z_2)$ decouples
1717: as a slow system
1718: \begin{equation} \label{dualslowsys}
1719: \tilde z'= -\Big(\lambda (J+v_+\bb1 \theta_+ + \bb1 \tilde \Theta) \Big)^*
1720: \tilde z
1721: \end{equation}
1722: dual to \eqref{slowsys}, with solution operator
1723: \begin{equation}\label{dualslowsoln}
1724: P^*(x,\lambda)^{-1} e^{-\bar \lambda (J+v_+\bb1 \theta_+)^*)(x-y)}
1725: P(y,\lambda)^{*}
1726: \end{equation}
1727: dual to \eqref{slowsoln}, i.e. (fixing $y=L$, say), solutions of general form
1728: \begin{equation}\label{genform}
1729: \tilde z(\lambda,x)=
1730: P^*(x,\lambda)^{-1} e^{-\bar \lambda (J+v_+\bb1 \theta_+)^*)(x-y)}
1731: \tilde v,
1732: \end{equation}
1733: $\tilde v \in \C^2$ arbitrary.
1734:
1735: Denoting by
1736: $$
1737: \tilde Z_1^+(L):=R^*\tilde W_1^+(L),
1738: $$
1739: therefore, the unique (up to constant factor) decaying solution
1740: at $+\infty$, and
1741: $\tilde v_1^+:=((\tilde V_1^+)_1 , (\tilde V_1^+)_2)^T$,
1742: we thus have evidently
1743: %\begin{equation}\label{zform}
1744: $$
1745: \tilde z_1^+(x,\lambda)=
1746: P^*(x,\lambda)^{-1} e^{-\bar \lambda (J+v_+\bb1 \theta_+)^*)x}
1747: \tilde v_1^+,
1748: $$
1749: %\end{equation}
1750: which, as $v_+\to 0$, is uniformly bounded by
1751: \begin{equation}\label{weakexp}
1752: |\tilde z_1^+(x,\lambda)|\le Ce^{\epsilon x}
1753: \end{equation}
1754: for arbitrarily small $\epsilon>0$
1755: and, by \eqref{dualVlim}, converges for $x$ less than or equal to
1756: $X-\delta$ for any fixed $X$ simply to
1757: \begin{equation}\label{simplelim}
1758: \lim_{v_+\to 0}\tilde z_1^+(x,\lambda)=
1759: P^*(x,\lambda)^{-1} (0,1)^T.
1760: \end{equation}
1761:
1762:
1763: Defining by $\tilde q:=(\tilde Z_1^+)_3$
1764: the fast coordinate of $\tilde Z_1^+$, we have, by \eqref{dualtri2},
1765: $$
1766: \tilde q'
1767: +\Big(f(\hat v)-\bar \lambda -(\lambda v_+ \theta_+\bb1 + \lambda
1768: \tilde \Theta \bb1)^* \Big)
1769: \tilde q=
1770: \bar \lambda \bb1^T \tilde z_1^+,
1771: $$
1772: whence, by Duhamel's principle, any decaying solution is given by
1773: $$
1774: \tilde q(x,\lambda)=
1775: \int_x^{+\infty} e^{\int_y^x a(z,\lambda, v_+)dz}\bar \lambda \bb1^T z_1^+(y) \, dy,
1776: $$
1777: where
1778: $$
1779: a(y,\lambda,v_+):=
1780: -\Big(f(\hat v)-\bar \lambda -(\lambda v_+ \theta_+\bb1 + \lambda
1781: \tilde \Theta \bb1)^* \Big).
1782: $$
1783: %TODO: a() is not the best variable, since a is used elsewhere for
1784: %small constant... (CHANGE? NO, I THINK OK, though not optimal...-KZ)
1785: Recalling, for $\R \lambda \ge 0$, that $\R a \ge 1/2$, combining
1786: \eqref{weakexp} and \eqref{simplelim},
1787: and noting that $a$ converges uniformly on $y\le Y$ as $v_+\to 0$ for
1788: any $Y>0$ to
1789: $$
1790: \begin{aligned}
1791: a_0(y, \lambda)&:=
1792: -f_0(\hat v)+\bar \lambda
1793: +(\lambda\tilde \Theta_0 \bb1)^* \\
1794: &= (1+\bar \lambda)
1795: +O(e^{-\eta y})
1796: \end{aligned}
1797: $$
1798: we obtain by the Lebesgue
1799: Dominated Convergence Theorem that
1800: $$
1801: \begin{aligned}
1802: \tilde q(L,\lambda)&\to
1803: \int_L^{+\infty} e^{\int_y^L a_0(z,\lambda)dz}\bar \lambda \bb1^T (0,1)^T \, dy\\
1804: &=
1805: \bar \lambda\int_L^{+\infty}
1806: e^{(1+\bar \lambda)(L-y)+ \int_y^L O(e^{-\eta z})dz }
1807: \, dy\\
1808: &=
1809: \bar \lambda
1810: (1+\bar \lambda)^{-1}(1 +O(e^{-\eta L})).
1811: \end{aligned}
1812: $$
1813: %
1814: Recalling, finally, \eqref{simplelim}, and the fact that
1815: $$
1816: |P-Id|(L,\lambda), \, |R-Id|(L,\lambda) \le Ce^{-\eta L}
1817: $$
1818: for $v_+$ sufficiently small, we obtain \eqref{wcon} as claimed.
1819: \end{proof}
1820:
1821: \begin{proof}[Proof of Corollary \ref{matching2}]
1822: Again, make the coordinate change $x\to x-\delta$ normalizing
1823: the background wave to match the shock-wave case.
1824: Applying Proposition \ref{conjugation} to the limiting adjoint system
1825: $$
1826: \tilde W'=-(A^0)^* \tilde W=
1827: \begin{pmatrix}0 & 0 & 0\\-\bar \lambda & 0 & 0\\
1828: -1& -1 & 1+\bar \lambda \end{pmatrix}
1829: \tilde W + O(e^{-\eta x})\tilde W,
1830: $$
1831: we find that, up to an $Id +O(e^{-\eta x})$ coordinate change,
1832: $\tilde W_1^{0+}(x)$ is given by the exact solution
1833: $\tilde W\equiv \tilde V_1$ of the limiting, constant-coefficient
1834: system
1835: $$
1836: \tilde W'=-(A^0)^* \tilde W=
1837: \begin{pmatrix}0 & 0 & 0\\-\bar \lambda & 0 & 0\\
1838: -1& -1 & 1+\bar \lambda \end{pmatrix}
1839: \tilde W.
1840: $$
1841: This yields immediately \eqref{limcon},
1842: which, together with \eqref{wcon}, yields \eqref{wcon2}.
1843: \end{proof}
1844:
1845: \subsection{Convergence to $D^0$}\label{convergence}
1846:
1847: The rest of our analysis is standard.
1848:
1849: \begin{lemma}\label{regconj}
1850: On $x\le L-\delta$ for any fixed $L>0$, there exists a coordinate-change
1851: $W=TZ$ conjugating \eqref{evans_ode} to the limiting equations
1852: \eqref{limevans_ode}, $T=T(x,\lambda, v_+)$, satisfying a uniform bound
1853: \begin{equation}\label{Tbd}
1854: |T-Id|\le C(L)v_+
1855: \end{equation}
1856: for all $v_+> 0$ sufficiently small.
1857: \end{lemma}
1858:
1859: \begin{proof}
1860: Make the coordinate change $x\to x-\delta$ normalizing the background profile.
1861: For $x\in (-\infty, 0]$, this is a consequence of the
1862: {\it Convergence Lemma} of \cite{PZ}, a variation on
1863: Proposition \ref{conjugation}, together with uniform
1864: convergence of the profile and eigenvalue equations.
1865: For $x\in [0,L]$, it is essentially continuous dependence;
1866: more precisely, observing that
1867: $ |A-A^0|\le C_1(L)v_+$ for $x\in [0,L]$,
1868: setting $S:=T-Id$, and writing the
1869: homological equation expressing conjugacy of \eqref{evans_ode}
1870: and \eqref{limevans_ode}, we obtain
1871: $$
1872: S'- (AS-SA^0)= (A-A^0),
1873: $$
1874: which, considered as an inhomogeneous linear matrix-valued equation, yields
1875: an exponential growth bound
1876: \[
1877: S(x)\le e^{Cx}(S(0)+ C^{-1}C_1(L)v_+)\]
1878: for some $C>0$, giving the result.
1879: \end{proof}
1880:
1881: \begin{proof}[Proof of Theorem \ref{mainthm}: inflow case]
1882: Make the coordinate change $x\to x-\delta$ normalizing the background profile.
1883: Lemma \ref{regconj}, together with convergence as $v_+\to 0$
1884: of the unstable subspace of $A_-$ to the unstable subspace
1885: of $A^0_-$ at the same rate $O(v_+)$ (as follows by spectral separation
1886: of the unstable eigenvalue of $A^0$ and standard matrix
1887: perturbation theory) yields
1888: \begin{equation}\label{Wbd}
1889: |W_1^0(0,\lambda)- W_1^{00}(0,\lambda)|\le C(L)v_+.
1890: \end{equation}
1891:
1892: Likewise, Lemma \ref{regconj} gives
1893: \begin{equation}\label{tildeWbd}
1894: \begin{aligned}
1895: |\tilde W_1^+(0,\lambda)- \tilde W_1^{0+}(0,\lambda)|&\le
1896: C(L)v_+
1897: |\tilde W_1^+(0,\lambda)|\\
1898: &\quad +
1899: |S_0^{L \to 0}|
1900: |\tilde W_1^+(L,\lambda)- \tilde W_1^{0+}(L,\lambda)|,
1901: \end{aligned}
1902: \end{equation}
1903: where $S_0^{y\to x}$ denotes the solution operator of
1904: the limiting adjoint eigenvalue equation $\tilde W'=-(A^0)^*\tilde W$.
1905: Applying Proposition \ref{conjugation} to the limiting system, we obtain
1906: $$
1907: |S_0^{L\to 0}|\le C_2e^{-A^0_+ L}\le C_2L|\lambda|
1908: $$
1909: by direct computation of $e^{-A^0_+ L}$, where $C_2$ is independent of $L>0$.
1910: Together with \eqref{wcon2} and \eqref{tildeWbd}, this gives
1911: $$
1912: |\tilde W_1^+(0,\lambda)- \tilde W_1^{0+}(0,\lambda)|\le
1913: C(L)v_+
1914: |\tilde W_1^+(0,\lambda)| + L|\lambda|C_2Ce^{-\eta L},
1915: $$
1916: hence, for $|\lambda|$ bounded,
1917: \begin{equation}\label{lastbd}
1918: \begin{aligned}
1919: |\tilde W_1^+(0,\lambda)- \tilde W_1^{0+}(0,\lambda)|&\le
1920: C_3(L)v_+ |\tilde W_1^{0+}(0,\lambda)| + LC_4e^{-\eta L}\\
1921: &\le
1922: C_5(L)v_+ + LC_4e^{-\eta L}.\\
1923: \end{aligned}
1924: \end{equation}
1925: Taking first $L\to \infty$ and then $v_+\to 0$, we obtain
1926: therefore convergence of $W^+_1(0,\lambda)$ and $\tilde W_1^+(0,\lambda)$ to
1927: $W^{0+}_1(0,\lambda)$ and $\tilde W_1^{0+}(0,\lambda)$, yielding the result
1928: by definitions \eqref{adjevans} and \eqref{duallimD}.
1929: \end{proof}
1930:
1931: \begin{proof}[Proof of Theorem \ref{mainthm}: outflow case]
1932: Straightforward, following the previous argument in the regular
1933: region only.
1934: \end{proof}
1935:
1936: \subsection{Convergence to the shock case}\label{shocklim}
1937:
1938:
1939: \begin{proof}[Proof of Theorem \ref{main2}: inflow case]
1940: First make the coordinate change $x\to x-\delta$ normalizing
1941: the background profile location to that of the shock wave case,
1942: where $\delta \to +\infty$ as $v_0\to 1$.
1943: By standard duality properties,
1944: $$
1945: D_{\rm in}=
1946: W^0_1 \cdot \tilde W_1^+|_{x=x_0}
1947: $$
1948: is independent of $x_0$, so we may evaluate at $x=0$ as in the
1949: shock case.
1950: Denote by ${\mathcal{W}}^-_1$, $\tilde {\mathcal{W}}_1^+$
1951: the corresponding modes in the shock case, and
1952: $$
1953: \mathcal{D}=
1954: \mathcal{W}^-_1\cdot \tilde {\mathcal{W}}_1^+|_{x=0}
1955: $$
1956: the resulting Evans function.
1957:
1958: Noting that $\tilde {\mathcal{W}}^1_+$
1959: and $\tilde W^1_+$ are asymptotic to the unique stable mode
1960: at $+\infty$ of the (same) adjoint eigenvalue equation,
1961: but with translated decay rates, we see immediately that
1962: $ \tilde {\mathcal{W}}^+_1=\tilde W^1_+ e^{-\delta \tilde \mu_1^+}. $
1963: %
1964: $W^0_1 $ on the other hand is initialized at $x=-\delta$
1965: (in the new coordinates $\tilde x=x-\delta$)
1966: as
1967: $$
1968: W^0_1(-\delta)=(1,0,0)^T,
1969: $$
1970: whereas ${\mathcal{W}}^-_1$ is the unique unstable mode at $-\infty$
1971: decaying as $e^{\mu_1^-x}V_1^-$, where $ V_1^- $ is the unstable
1972: right eigenvector of
1973: $$
1974: A_-=
1975: \begin{pmatrix}
1976: 0 & \lambda & \lambda\\
1977: 0 & 0 & \lambda\\
1978: 1& 1 & f(1)-\lambda\\
1979: \end{pmatrix}.
1980: $$
1981:
1982: Denote by $\tilde V^-_1$ the associated dual unstable left eigenvector and
1983: $$
1984: \Pi^-_1:=V^-_1(\tilde V^-_1)^T
1985: $$
1986: the eigenprojection onto the stable vector $V^-_1$.
1987: By direct computation,
1988: $$
1989: \tilde V^-_1=c(\lambda)(1, 1+\lambda/\mu^-_1,\mu^-_1)^T,
1990: \quad c(\lambda)\ne 0,
1991: $$
1992: yielding
1993: \begin{equation}\label{goodproj}
1994: \Pi^-_1 W^0_1=:\beta(\lambda)=c(\lambda)\ne 0
1995: \end{equation}
1996: for $\Re \lambda\ge 0$, on which $\Re \mu^-_1>0$.
1997:
1998: Once we know \eqref{goodproj}, we may finish by a standard argument,
1999: concluding by exponential attraction
2000: in the positive $x$-direction of the unstable mode that other modes
2001: decay exponentially as $x\to 0$, leaving the
2002: contribution from $\beta(\lambda)V_1^-$ plus a negligible $O(e^{-\eta \delta})$
2003: error, $\eta>0$, from which we may conclude that
2004: ${\mathcal{W}}^-_1|_{x=0}\sim \beta^{-1}e^{-\delta \mu_1^-} W^0_1|_{x=0}.$
2005: Collecting information, we find that
2006: $$
2007: \mathcal{D}(\lambda)=
2008: \beta(\lambda)^{-1}
2009: e^{-\delta
2010: (\bar \mu_1^-+ \tilde \mu_1^+)
2011: (\lambda)} D_{\rm in}(\lambda) + O(e^{-\eta \delta}),
2012: $$
2013: $\eta>0$, yielding the claimed convergence as $v_0\to 1$, $\delta\to +\infty$.
2014: \end{proof}
2015:
2016: \begin{proof}[Proof of Theorem \ref{main2}: outflow case]
2017: For $\lambda$ uniformly bounded from zero,
2018: $\tilde W^0_1=(0,-1, -\bar \lambda /(\bar \lambda -\hat v'(0)))^T$
2019: converges uniformly as $v_0\to 0$ to
2020: $$
2021: (0,-1, -1)^T,
2022: $$
2023: whereas the shock Evans function $\mathcal{D}$ is initiated by
2024: $\tilde{\mathcal{ W}}^+_1$ proportional to
2025: $$
2026: \tilde{\mathcal{V}}^+_1= (0,-1, -1-\lambda)^T
2027: $$
2028: agreeing in the first two coordinates with $\tilde W^0_1$.
2029: By the boundary-layer analysis of Section \ref{estW1}, the
2030: backward (i.e., decreasing $x$) evolution of the adjoint
2031: eigenvalue ODE reduces in the asymptotic
2032: limit $v_+\to 0$ (forced by $v_0\to 0$) to a decoupled slow flow
2033: $$
2034: \tilde w'=\begin{pmatrix}0 & \bar \lambda\\ 0 & 0\end{pmatrix}w,
2035: \qquad w\in \C^2
2036: $$
2037: in the first two coordinates, driving an exponentially slaved
2038: fast flow in the third coordinate.
2039: From this, we may conclude that solutions agreeing in the first
2040: two coordinates converge exponentially as $x$ decreases.
2041: Performing an appropriate normalization, as in the inflow case
2042: just treated, we thus obtain the result.
2043: We omit the details, which follow what has already been done
2044: in previous cases.
2045: \end{proof}
2046:
2047: \subsection{The stability index}\label{stabsection}
2048: Following \cite{SZ,GMWZ.5}, we note that $D_{\rm in}(\lambda)$ is
2049: real for real $\lambda$, and nonvanishing for real $\lambda$
2050: sufficiently large, hence $\sgn D_{\rm in}(+\infty)$ is well-defined
2051: and constant on the entire (connected) parameter range. The number
2052: of roots of $D_{\rm in}$ on $\Re \lambda \ge 0$ is therefore even or
2053: odd depending on the {\it stability index}
2054: $$
2055: \sgn [D_{\rm in}(0)D_{\rm in}(+\infty)].
2056: $$
2057: Similarly, recalling that $D_{\rm out}(0)\equiv 0$, we find that the
2058: number of roots of $D_{\rm out}$ on $\Re \lambda \ge 0$ is even or
2059: odd depending on
2060: $$
2061: \sgn [D_{\rm out}'(0)D_{\rm out}(+\infty)].
2062: $$
2063:
2064: \begin{proof}[Proof of Lemma \ref{index}: inflow case]
2065: Examining the adjoint equation at $\lambda=0$,
2066: $$
2067: \tilde W'=-A^*\tilde W,
2068: \qquad
2069: -A^*(x,0)=
2070: \begin{pmatrix}
2071: 0 & 0 & -\hat v\\
2072: 0 & 0 & -\hat v\\
2073: 0 & 0 & -f(\hat v)\\
2074: \end{pmatrix},
2075: $$
2076: $-f(v_+)>0$, we find by explicit computation that
2077: the only solutions that are bounded as $x\to +\infty$ are
2078: the {\it constant solutions} $\tilde W\equiv (a,b,0)^T$.
2079: Taking the limit $V^+_1(0)$ as $\lambda \to 0^+$ along the real axis
2080: of the unique stable eigenvector of $-A^*_+(\lambda)$,
2081: we find (see, e.g., \cite{Z.3}) that it lies in the direction
2082: $(1, 2+a_j^+,0)^T$, where $a_j^+>0$ is the positive characteristic
2083: speed of the hyperbolic convection matrix
2084: $\begin{pmatrix}
2085: 1 & - 1\\
2086: -h(v_+)/v_+^{\gamma+1} & 1\\
2087: \end{pmatrix}$, i.e.,
2088: $V_1^-= c(v_0,v_+)(1, 2+a_j^+,0)^T$, $c(v_0,v_+)\ne 0$. Thus,
2089: $D_{\rm in}(0)= V_1^-\cdot (1,0,0)^T=\bar c(v_0,v_+) \ne 0$ as
2090: claimed. On the other hand, the same computation carried out for
2091: $D^0_{\rm in}(0)$ yields $D^0_{\rm in}(0)\equiv 0$. (Note: $a_j\sim
2092: v_+^{-1/2}\to +\infty$ as $v_+\to 0$.)
2093: %
2094: Similarly, as $v_0\to 0$,
2095: $$
2096: D^0_{\rm in} (\lambda)\to (1,0,0)^T\cdot
2097: (0,1,*)^T\equiv 0.
2098: $$ Finally, note $D_{\rm in}(0)\ne 0$ implies that the stability
2099: index, since continuously varying so long as it doesn't vanish and
2100: taking discrete values $\pm 1$, must be constant on the connected
2101: set of parameter values. Since inflow boundary layers are known to
2102: be stable on some part of the parameter regime by energy estimates
2103: (Theorem \ref{main2}), we may conclude that the stability index is
2104: identically one and therefore there are an even number of unstable
2105: roots for all $1>v_0\ge v_+>0$.
2106:
2107: To establish that $(D^0_{\rm in})'(0)\ne 0$, we compute
2108: $$
2109: D^0_{\rm in}\; '(0) = (\partial_\lambda W_1^{00})\cdot
2110: \widetilde{W}_1^{0+} + W_1^{00}\cdot (\partial_\lambda
2111: \widetilde{W}_1^{0+}).
2112: $$ Since $W_1^{00} \equiv (1,0,0)$ is independent of $\lambda$, we
2113: need only show that the first component of $\partial_\lambda
2114: \widetilde{W}_1^{0+}$ is nonzero. Note that $\partial_\lambda
2115: W_1^{0+}$ solves the limiting adjoint variational equations
2116:
2117:
2118: $$
2119: (\partial_\lambda \widetilde{W}_1^{0+})'(0) + (A^0)^*(x,0)
2120: \partial_\lambda \widetilde{W}_1^{0+} = b(x)
2121: $$ with
2122: $$
2123: (A^0)^*(x,0)\tilde =
2124: \begin{pmatrix}
2125: 0 & 0 & \hat v^0\\
2126: 0 & 0 & \hat v^0\\
2127: 0 & 0 & f^0(\hat v^0)\\
2128: \end{pmatrix}, \qquad b(x) = \begin{pmatrix}
2129: 0 \\
2130: \hat v^0 + \hat u^0 - \frac{\hat v^0\, '}{\hat v^0} - 1 \\
2131: 3\hat v^0 - \frac{\hat v^0\, '}{\hat v^0} - 1 \end{pmatrix}.
2132: $$ By \eqref{tildeV}, and the fact that
2133: $\partial_\lambda \tilde \mu_1^{0+}\equiv 0$,
2134: $\partial_\lambda \widetilde{W}_1^{0+}(x)$ is chosen so
2135: that asymptotically at $x=+\infty$ it lies in the direction of
2136: $\partial_\lambda \tilde V_1 = (0,0,-1)$. Set $\partial_\lambda
2137: \widetilde{W}_1^{0+} = (\partial_\lambda \widetilde{W}_{1,\;1}^{0+},
2138: \partial_\lambda \widetilde{W}_{1,\;2}^{0+}, \partial_\lambda
2139: \widetilde{W}_{1,\;3}^{0+})^T$. Then the third component solves
2140: $$
2141: (\partial_\lambda \widetilde{W}_{1,\;3}^{0+})' + \hat v ^0
2142: \partial_\lambda \widetilde{W}_{1,\; 3}^{0+} =b_3 := 3\hat v^0 -
2143: \frac{\hat v^0\, '}{\hat v^0} - 1
2144: $$ which has solution
2145: $$
2146: \partial_\lambda \widetilde{W}_{1,\;3}^{0+}(x) = \partial_\lambda \widetilde{W}_{1,\;3}^{0+}(+\infty)\varphi(x) -
2147: \varphi(x) \int_x^\infty \varphi^{-1}(y) b_3(y)dy
2148: $$ where
2149: $$
2150: \varphi(x) = e^{\int_x^\infty \hat v^0(y) dy}.
2151: $$ Integrating the equation for the first component of $\partial_\lambda \widetilde{W}_1^{0+}$ yields
2152: \begin{align*}
2153: \begin{split}
2154: \partial_\lambda \widetilde{W}_{1,\;1}^{0+}(x) & = \partial_\lambda
2155: \widetilde{W}_{1,\;1}^{0+}(+\infty) + \int_x^\infty \partial_\lambda
2156: \widetilde{W}_{1,\;3}^{0+}(y) dy \\
2157: &= \partial_\lambda \widetilde{W}_{1,\;1}^{0+}(+\infty) +
2158: \partial_\lambda \widetilde{W}_{1,\;3}^{0+}(+\infty) \int_x^\infty
2159: \hat v^0(y) \varphi(y) dy \\
2160: & - \int_x^\infty \left( \varphi(y) \int_y^\infty \varphi^{-1}(z)
2161: b_3(z) dz \right) dy.
2162: \end{split}
2163: \end{align*} Using the condition $\partial_\lambda \widetilde{W}_{1}^{0+}(+\infty) =
2164: (0,0,-1)^T$ we have $\partial_\lambda
2165: \widetilde{W}_{1,\;1}^{0+}(+\infty) = 0, \partial_\lambda
2166: \widetilde{W}_{1,\;3}^{0+}(+\infty) = -1$ so that
2167:
2168: $$
2169: \partial_\lambda \widetilde{W}_{1,\;1}^{0+}|_{x=0} = - \int_0^\infty \hat v^0(y) \varphi(y) dy - \int_x^\infty \left(
2170: \varphi(y) \int_y^\infty \varphi^{-1}(z) b_3 dz \right) dy.
2171: $$ Finally, note that by using \eqref{v^0} we have $b_3=1-\tanh(\frac{x-\delta}{2})$ so that for all $x\geq 0$, $\varphi(x), b_3(x) \geq
2172: 0$ which implies
2173: $$
2174: D^0_{\rm in}\, '(0) = \partial_\lambda
2175: \widetilde{W}_{1,\;1}^{0+}|_{x=0} \neq 0.
2176: $$
2177: \end{proof}
2178:
2179: \begin{remark}\label{apparent}
2180: The result $D_{\rm in}(0)\ne 0$ at first sight appears to contradict that
2181: of Theorem \ref{main3}, since $\mathcal{D}(0)=0$ for
2182: the shock wave case. This apparent contradiction is explained
2183: by the fact that the normalizing factor
2184: $e^{-\delta (\bar \mu_1^- + \tilde \mu_1^+) }$
2185: is exponentially decaying in $\delta$ for $\lambda=0$, since
2186: $\tilde \mu_1^+(0)=0$, while $\Re\mu^-_1>0$.
2187: Recalling that $\delta\to +\infty$ as $v_0\to 1$, we recover the result
2188: of Theorem \ref{main3}.
2189: \end{remark}
2190:
2191: \begin{proof}[Proof of Lemma \ref{index}: outflow case]
2192: Similarly, we compute
2193: $$
2194: D'_{\rm out}(0)=
2195: \partial \lambda W^-_1 \cdot \tilde W^0_1,
2196: $$
2197: where $ \partial \lambda W^-_1|_{\lambda=0}$ satisfies the variational
2198: equation $L\partial_\lambda U^-_1(0)=U^-_1=\hat U'$, or, written
2199: as a first-order system,
2200: $$
2201: (\partial \lambda W^-_1)'- A(x,0) \partial \lambda W^-_1=
2202: \begin{pmatrix}\hat u_x\\ \hat v_x\\ -\hat v_x \end{pmatrix},
2203: \qquad
2204: A(x,0)=
2205: \begin{pmatrix} 0 & 0 & 0\\
2206: 0 & 0 & 0\\
2207: \hat v & \hat v & f(\hat v)\\
2208: \end{pmatrix},
2209: $$
2210: which may be solved exactly for the unique solution
2211: decaying at $-\infty$ of
2212: $$
2213: W^-_1(0)=
2214: \begin{pmatrix} 0 \\ 0\\ \hat v' \end{pmatrix},
2215: \qquad
2216: (\partial \lambda W^-_1)(0)=
2217: \begin{pmatrix}\hat u- u_-\\ \hat v- v_-\\ * \end{pmatrix}.
2218: $$
2219: Recalling from \eqref{tildew1} that
2220: $\widetilde{W}^0_1(\lambda)= (0, -1, -\lambda/(\lambda -\hat v'(0)))^T$,
2221: hence
2222: $$
2223: \widetilde{W}^0_1(0)= (0, -1, 0)^T,
2224: \qquad
2225: \partial_\lambda \widetilde{W}^0_1(0)= (0, 0, 1/\hat v'(0))^T,
2226: $$
2227: we thus find that
2228: $$
2229: \begin{aligned}
2230: D_{\rm out}'(0)&=\partial_\lambda W^-_1(0)\cdot \widetilde{W}^0_1(0)
2231: +W^-_1(0)\cdot \partial_\lambda\widetilde{W}^0_1(0)
2232: \\
2233: &= -(\hat v(0)-1)+1 =2-v_0\ne 0
2234: \end{aligned}
2235: $$
2236: as claimed. The proof that $(D^0_{\rm out})'(0)\ne 0$ goes
2237: similarly.
2238:
2239: Finally, as in the proof of the inflow case, we note that nonvanishing
2240: implies that the stability index is constant across the entire
2241: (connected) parameter range, hence we may conclude that it is identically
2242: one by existence of a stable case (Corollary \ref{v01}), and
2243: therefore that the number of nonzero unstable roots is even, as claimed.
2244: \end{proof}
2245:
2246: \subsection{Stability in the shock limit}\label{smallv0}
2247:
2248: \begin{proof}[Proof of Corollary \ref{v01}: inflow case]
2249: By Proposition \ref{redenergy}
2250: %By convergence to the shock case, we find that $D_{\rm in}$
2251: we find that $D_{\rm in}$
2252: has at most a single zero in $\Re \lambda \ge 0$. However,
2253: by our stability index results, Theorem \ref{index}, the
2254: number of eigenvalues in $\Re \lambda \ge 0$ is even.
2255: Thus, it must be zero, giving the result.
2256: \end{proof}
2257:
2258: \begin{proof}[Proof of Corollary \ref{v01}: outflow case]
2259: By Theorem \ref{main3}, $D_{\rm out}$, suitably renormalized,
2260: converges as $v_0\to 0$ to the Evans function for the (unintegrated)
2261: shock wave case.
2262: But, the shock Evans function by the results of \cite{BHRZ,HLZ}
2263: has just a single zero at $\lambda=0$ on $\Re \lambda\ge 0$,
2264: already accounted for in $D_{\rm out}$
2265: by the spurious root at $\lambda=0$ introduced by recoordinatization
2266: to ``good unknown''.
2267: %(NOTE: as mentioned just above, we don't in fact need connection
2268: %to shock case, unless we insist to do this analytically. Numerical
2269: %verification is easy, since more stable.)
2270: \end{proof}
2271:
2272: \subsection{Stability for small $v_0$}\label{corner}
2273:
2274: Finally, we treat the remaining,
2275: ``corner case'' as $v_+$, $v_0$ simultaneously approach zero.
2276: The fact (Lemma \ref{index}) that
2277: $$
2278: \lim_{v_0\to 0}\lim_{v_+\to 0}D_{\rm in}(\lambda)\equiv 0
2279: $$
2280: shows that this limit is quite delicate; indeed, this
2281: is the most delicate part of our analysis.
2282:
2283: \begin{proof}[Proof of Theorem \ref{main2}: inflow case]
2284: Consider again the adjoint system
2285: $$
2286: \tilde W'=-A^*(x,\lambda)\tilde W,
2287: \qquad
2288: A^*(x,\lambda)\tilde =
2289: \begin{pmatrix}
2290: 0 & 0 & \hat v\\
2291: \bar \lambda & 0 & \hat v\\
2292: \bar \lambda & \bar \lambda & f(\hat v) - \bar \lambda\\
2293: \end{pmatrix}.
2294: $$
2295: By the boundary analysis of Section \ref{estW1},
2296: $$
2297: \tilde W=
2298: \Big(\alpha, 1, \frac{\alpha \tilde \mu-\bar \lambda(\alpha +1)}
2299: {-f(\hat v)+\bar \lambda}\Big)^T
2300: + O(e^{-\eta |x-\delta|}),
2301: $$
2302: where $\alpha:=\frac{\tilde \mu_+}{\tilde \mu_++ \bar \lambda}$,
2303: and $\tilde \mu$ is the unique stable eigenvalue of $A^*_+$,
2304: satisfying (by matrix perturbation calculation)
2305: $$
2306: \tilde \mu= \bar \lambda(v_+^{1/2} + O(v_+))
2307: $$
2308: and thus $\alpha =v_+^{1/2}+O(v_+)$
2309: as $v_0\to 0$ (hence $v_+\to 0$) on bounded subsets of $\Re \lambda \ge 0$.
2310: Combining these expansions, we have
2311: $$
2312: \tilde W_1(+\infty)=v_+^{1/2}(1+o(1)),
2313: \qquad
2314: \tilde W_3=
2315: \frac{-\bar \lambda}
2316: {-f(\hat v)+\bar \lambda}
2317: (1 + o(1))
2318: $$
2319: for $v_0$ sufficiently small.
2320:
2321: From the $\tilde W_1$ equation $\tilde W'= \hat v \tilde W_3$,
2322: we thus obtain
2323: $$
2324: \begin{aligned}
2325: \tilde W_1(0)&=
2326: \tilde W_1(+\infty)- \int_0^{+\infty}\hat v \tilde W_3(y)\, dy\\
2327: &= (1+o(1))\times
2328: \Big(
2329: v_+^{1/2} + \int_0^{+\infty}
2330: \frac{\bar \lambda \hat v}{-f(\hat v)+\bar \lambda}(y)\, dy \Big).
2331: \end{aligned}
2332: $$
2333: Observing, finally, that, for $\Re \lambda \ge 0$, the
2334: ratio of real to imaginary parts of
2335: $\frac{\bar \lambda \hat v}{-f(\hat v)+\bar \lambda}(y)$ is
2336: uniformly positive, we find that $\Re \tilde W_1(0)\ne 0$
2337: for $v_0$ sufficiently small, which yields nonvanishing of
2338: $D_{\rm in}(\lambda)$ on $\Re \lambda \ge 0$ as claimed.
2339: \end{proof}
2340:
2341: \section{Numerical computations}\label{computations}
2342:
2343: In this section, we show, through a systematic numerical Evans function
2344: study, that there are no unstable eigenvalues for
2345: \[
2346: (\gamma,v_+) \in[1,3]\times(0,1],
2347: \]
2348: in either inflow or outflow cases. As defined in Section \ref{evanssec},
2349: the Evans function is analytic in the right-half plane and reports a value
2350: of zero precisely at the eigenvalues of the linearized operator
2351: \eqref{linearized}. Hence we can use the argument principle to determine
2352: if there are any unstable eigenvalues for this system. Our approach
2353: closely follows that of \cite{BHRZ,HLZ} for the shock case with only two
2354: major differences. First, our shooting algorithm is only one sided as we
2355: have the boundary conditions \eqref{w01} and \eqref{tildew1} for the
2356: inflow and outflow cases, respectfully. Second, we ``correct'' for the
2357: displacement in the boundary layer when $v_0\approx 1$ in the inflow case
2358: and $v_0\approx 0$ in the outflow case so that the Evans function
2359: converges to the shock case as studied in \cite{BHRZ,HLZ} (see discussion
2360: in Section \ref{sectionshocklimit}).
2361:
2362: \begin{figure}[t]
2363: \begin{center}
2364: $\begin{array}{cc}
2365: \includegraphics[width=6.4cm,height=4.5cm]{inflow1} &
2366: \includegraphics[width=6.4cm,height=4.5cm]{inflow2} \\
2367: \mbox{\bf (a)} & \mbox{\bf (b)}\\
2368: \includegraphics[width=5.75cm]{inflow3} &
2369: \includegraphics[width=5.6cm]{inflow4} \\
2370: \mbox{\bf (c)} & \mbox{\bf (d)}
2371: \end{array}$
2372: \end{center}
2373: \caption{Typical examples of the inflow case, showing convergence to the
2374: limiting Evans function as $v_+\to 0$ for a monatomic gas, $\gamma=5/3$,
2375: with $(a)$ $v_0=0.1$, $(b)$ $v_0=0.2$, $(c)$ $v_0=0.4$, and $(d)$
2376: $v_0=0.7$. The contours depicted, going from inner to outer, are images
2377: of the semicircle $\phi$ under $D$ for
2378: $v_+=1e\!-\!2,1e\!-\!3,1e\!-\!4,1e\!-\!5,1e\!-\!6$, with the outer-most
2379: contour given by the image of $\phi$ under $D^0$, that is, when $v_+= 0$.
2380: Each contour consists of $60$ points in $\lambda$.}
2381: \label{first}
2382: \end{figure}
2383:
2384: The profiles were generated using Matlab's {\tt bvp4c} routine, which is
2385: an adaptive Lobatto quadrature scheme. The shooting portion of the Evans
2386: function computation was performed using Matlab's {\tt ode45} package,
2387: which is the standard 4th order adaptive Runge-Kutta-Fehlberg method
2388: (RKF45). The error tolerances for both the profiles and the shooting were
2389: set to {\tt AbsTol=1e-6} and {\tt RelTol=1e-8}. We remark that Kato's ODE
2390: (see Section \ref{evanssec} and \cite{Kato,HSZ} for details) is used to
2391: analytically choose the initial eigenbasis for the stable/unstable
2392: manifolds at the numerical values of infinity at $L=\pm 18$. Finally in Section
2393: \ref{errorstudy}, we carry out a numerical convergence study similar to that
2394: in \cite{BHRZ}.
2395:
2396: \begin{figure}[t]
2397: \begin{center}
2398: $\begin{array}{cc}
2399: \includegraphics[width=5.75cm]{outflow1} &
2400: \includegraphics[width=5.75cm]{outflow2} \\
2401: \mbox{\bf (a)} & \mbox{\bf (b)}\\
2402: \includegraphics[width=5.75cm]{outflow3} &
2403: \includegraphics[width=5.3cm]{outflow4} \\
2404: \mbox{\bf (c)} & \mbox{\bf (d)}
2405: \end{array}$
2406: \end{center}
2407: \caption{Typical examples of the outflow case, showing convergence to the
2408: limiting Evans function as $v_+\to 0$ for a monatomic gas, $\gamma=5/3$,
2409: with $(a)$ $v_0=0.2$, $(b)$ $v_0=0.4$, $(c)$ $v_0=0.6$, and $(d)$
2410: $v_0=0.8$. The contours depicted are images of the semicircle $\phi$
2411: under $D$ for $v_+=1e\!-\!2,1e\!-\!3,1e\!-\!4,1e\!-\!5,1e\!-\!6$, and the
2412: limiting case $v_+= 0$. Interestingly the contours are essentially
2413: (visually) indistinguishable in this parameter range. Each contour
2414: consists of $60$ points in $\lambda$}
2415: \label{second1}
2416: \end{figure}
2417:
2418: \subsection{Winding number computations}
2419:
2420: The high-frequency estimates in Proposition \ref{hf} restrict the set of
2421: admissible unstable eigenvalues to a fixed compact triangle $\Lambda$ in
2422: the right-half plane (see \eqref{hfbounds1} and \eqref{hfbounds2} for the
2423: inflow and outflow cases, respectively). We reiterate the remarkable
2424: property that $\Lambda$ does not depend on the choice of $v_+$ or $v_0$.
2425: Hence, to demonstrate stability for a given $\gamma$, $v_+$ and $v_0$, it
2426: suffices to show that the winding number of the Evans function along a
2427: contour containing $\Lambda$ is zero. Note that in our region of
2428: interest, $\gamma\in[1,3]$, the semi-circular contour given by
2429: \[
2430: \phi:=\partial(\{\lambda\mid \R\geq 0\}\cup\{\lambda\mid |\lambda|\leq 10\}),
2431: \]
2432: contains $\Lambda$ in both the inflow and outflow cases. Hence, for
2433: consistency we use this same semicircle for all of our winding number
2434: computations.
2435:
2436: A remarkable feature of the Evans function for this system, and one that
2437: is shared with the shock case in \cite{BHRZ,HLZ}, is that the Evans function
2438: has limiting behavior as the amplitude increases, Section \ref{analytical}. For
2439: the inflow case, we see in Figure \ref{first}, the mapping of the contour $\phi$
2440: for the monatomic case ($\gamma=5/3$), for several different choices of $v_0$,
2441: as $v_+\rightarrow 0$. We remark that the winding numbers for $0\leq v_+\leq 1$
2442: are all zero, and the limiting contour touches zero due to the emergence of a
2443: zero root in the limit. Note that the limiting case contains the contours of all other
2444: amplitudes. Hence, we have spectral stability for all amplitudes.
2445:
2446: The outflow case likewise has a limiting behavior, however, all contours
2447: cross through zero due to the eigenvalue at the origin. Nonetheless,
2448: since the contours only wind around once, we can likewise conclude that
2449: these profiles are spectrally stable. We remark that the outflow case
2450: converges to the limiting case faster than the inflow case as is clear
2451: from Figure \ref{second1}.
2452: Indeed, $v_+=1e\!-\!2$ and the limiting case $v_+=0$, as well as all of
2453: the values of $v_+$ in between, are virtually indistinguishable.
2454:
2455: \begin{figure}[t]
2456: \begin{center}
2457: $\begin{array}{cc}
2458: \includegraphics[width=5.75cm]{index1} &
2459: \includegraphics[width=5.75cm]{index2} \\
2460: \mbox{\bf (a)} & \mbox{\bf (b)}
2461: \end{array}$
2462: \end{center}
2463: \caption{Typical examples of the Evans function evaluated along the
2464: positive real axis. The $(a)$ inflow case is computed for $v_0 = 0.7$ and
2465: $v_0=0$ and $(b)$ the outflow case is computed for $v_0=0.3$ and
2466: $v_+=0.001$. Not the transversality at the origin in both cases. Both
2467: graphs consist of $50$ points in $\lambda$.}
2468: \label{second2}
2469: \end{figure}
2470:
2471: In our study, we systematically varied $v_0$ in the interval $[.01,.99]$
2472: and took the $v_+\rightarrow 0$ limit at each step, starting from a
2473: $v_+=.9$ (or some other appropriate value, for example when $v_0<.9$) on
2474: the small-amplitude end and decreased $v_+$ steadily to $10^{-k}$ for
2475: $k=1,2,3,\ldots,6$, followed by evaluation at $v_+=0$. For both inflow
2476: and outflow cases, over $2000$ contours were computed. We remark that in
2477: the $v_+\rightarrow 0$ limit, the system becomes pressureless, and thus
2478: all of the contours in the large-amplitude limit look the same regardless
2479: of the value of $\gamma$ chosen.
2480:
2481: \subsection{Nonexistence of unstable real eigenvalues}
2482:
2483: As an additional verification of stability, we computed the Evans function
2484: along the unstable real axis on the interval $[0,15]$ for varying
2485: parameters to show that there are no real unstable eigenvalues. Since the
2486: Evans function has a root at the origin in the limiting system for the inflow case,
2487: and for all values of $v_+$ in the outflow case, we can perform in these cases a
2488: sort of {\em numerical stability index analysis} to verify that the Evans function
2489: cuts transversely through the origin and is otherwise nonzero, indicating
2490: that there are no unstable real eigenvalues as expected. In Figure
2491: \ref{second2}, we see a typical example of $(a)$ the inflow and $(b)$
2492: outflow cases. Note that in both images, the Evans function cuts
2493: transversally through the origin and is otherwise nonzero as $\lambda$
2494: increases.
2495:
2496: \begin{figure}[t]
2497: \begin{center}
2498: $\begin{array}{cc}
2499: \includegraphics[width=5.4cm]{limit1} &
2500: \includegraphics[width=5.9cm]{limit2} \\
2501: \mbox{\bf (a)} & \mbox{\bf (b)}
2502: \end{array}$
2503: \end{center}
2504: \caption{Shock limit for $(a)$ inflow and $(b)$ outflow cases, both for
2505: $\gamma=5/3$. Note that the images look very similar to those of
2506: \cite{BHRZ,HLZ}. }
2507: \label{second3}
2508: \end{figure}
2509:
2510: \subsection{The shock limit}
2511: \label{sectionshocklimit}
2512: When $v_0$ is far from the midpoint $(1-v_+)/2$ of the end states, the the
2513: Evans function of the boundary layer is similar to the Evans function of
2514: the shock case evaluated at the displacement point $x_0$. Hence, when we
2515: compute the boundary layer Evans function near the shock limits,
2516: $v_0\approx 1$ for the inflow case and $v_0\approx 0$ for the outflow
2517: case, we multiply for the correction factor $c(\lambda)$ so that our
2518: output looks close to that of the shock case studied in \cite{BHRZ,HLZ}.
2519: The correction factors are
2520: \[
2521: c(\lambda) = e^{(-\mu^+ - \bar{\mu}^-)x_0}
2522: \]
2523: for the inflow case and
2524: \[
2525: c(\lambda) = e^{(-\bar{\mu}^+ - \mu^-)x_0},
2526: \]
2527: for the outflow case, where $\mu^-$ is the growth mode of $A_-(\lambda)$
2528: and $\mu^+$ is the decay mode of $A_+(\lambda)$. In Figure \ref{second3},
2529: we see that these highly displaced profiles appear to be very similar to
2530: the shock cases with one notable difference. These images have a small
2531: dimple near $\lambda=0$ to account for the eigenvalue there,
2532: whereas those in the shock case \cite{BHRZ,HLZ} were computed
2533: in integrated coordinates and thus have no root at the origin.
2534:
2535: \begin{table}[t]
2536: \begin{center}
2537: \begin{tabular}{ccccccc}
2538: \hline
2539: \multicolumn{7}{c}{Inflow Case}\\
2540: \hline
2541: $L$ & $\gamma=1.2$ & $\gamma=1.4$ & $\gamma=1.666$ & $\gamma=2.0$ & $\gamma=2.5$ & $\gamma=3.0$ \\
2542: \hline
2543: 8 & 7.8(-1) & 8.4(-1) & 9.2(-1) & 1.0(0) & 1.2(0) & 1.3(0) \\
2544: 10 & 1.4(-1) & 1.2(-1) & 9.2(-2) & 6.8(-2) & 4.4(-2) & 2.8(-2) \\
2545: 12 & 1.4(-2) & 7.9(-3) & 3.6(-3) & 1.3(-3) & 3.1(-4) & 7.3(-5) \\
2546: 14& 1.3(-3) & 4.9(-4) & 1.3(-4) & 2.4(-5) & 8.7(-6) & 8.2(-6)\\
2547: 16 & 1.2(-4) & 3.0(-5) & 4.7(-6) & 2.8(-6) & 2.7(-6) & 2.6(-6)\\
2548: 18& 1.1(-5) & 5.8(-6) & 8.0(-6) & 8.1(-6) & 8.0(-6) & 8.0(-6) \\
2549: \hline
2550: %\end{tabular}
2551: %\begin{tabular}{ccccc}
2552: \multicolumn{7}{c}{Outflow Case}\\
2553: \hline
2554: $L$ & $\gamma=1.2$ & $\gamma=1.4$ & $\gamma=1.666$ & $\gamma=2.0$ & $\gamma=2.5$ & $\gamma=3.0$ \\
2555: \hline
2556: 8 & 5.4(-3) & 5.4(-3) & 5.4(-3) & 5.4(-3) & 5.4(-3) & 5.4(-3)\\
2557: 10 & 9.2(-4) & 9.1(-4) & 9.1(-4) & 9.1(-4) & 9.1(-4) & 9.1(-4) \\
2558: 12 & 1.5(-4) & 1.5(-4) & 1.5(-4) & 1.5(-4) & 1.5(-4) & 1.5(-4) \\
2559: 14& 2.5(-5) & 2.7(-5) & 2.0(-5) & 2.0(-5) & 2.0(-5) & 2.0(-5) \\
2560: 16 & 2.3(-6) & 2.6(-6) & 2.6(-6) & 2.5(-6) & 2.5(-6) & 2.5(-6) \\
2561: 18& 6.6(-6) & 3.6(-6) & 8.7(-6) & 8.7(-6) & 8.7(-6) & 8.7(-6) \\
2562: \hline
2563: \end{tabular}
2564: \caption{Relative errors in $D(\lambda)$ for the inflow and outflow cases are computed by taking the maximum relative error for 60 contour points evaluated along the semicircle $\phi$. Samples were taken for varying $L$ and $\gamma$, leaving $v_+$ fixed at $v_+=10^{-4}$ and $v_0=0.6$. We used $L=8,10,12,14,16,18,20$ and $\gamma=1.2,1.4,1.666,2.0$. Relative errors were computed using the next value of $L$ as the baseline.}
2565: \end{center}
2566: \end{table}
2567:
2568: \begin{table}[h]
2569: \begin{center}
2570: \begin{tabular}{ccccccc}
2571: \hline
2572: \multicolumn{7}{c}{Inflow Case}\\
2573: \hline
2574: \hline
2575: Abs/Rel & $\gamma=1.2$ & $\gamma=1.4$ & $\gamma=1.666$ & $\gamma=2.0$ & $\gamma=2.5$ & $\gamma=3.0$ \\
2576: \hline
2577: $10^{-3}/10^{-5}$ & 5.4(-4) & 4.1(-4) & 4.0(-4) & 5.0(-4) & 3.4(-4) & 8.6(-4)\\
2578: $10^{-4}/10^{-6}$ & 3.1(-5) & 4.6(-5) & 3.4(-5) & 3.3(-5) & 3.3(-5) & 3.2(-5)\\
2579: $10^{-5}/10^{-7}$ & 2.9(-6) & 3.6(-6) & 3.9(-6) & 6.8(-6) & 2.7(-6) & 2.5(-6)\\
2580: $10^{-6}/10^{-8}$ & 4.6(-7) & 9.9(-7) & 1.1(-6) & 6.0(-7) & 2.9(-7) & 3.2(-7)\\
2581: \hline
2582: %\end{tabular}
2583: %\begin{tabular}{ccccc}
2584: \multicolumn{7}{c}{Outflow Case}\\
2585: \hline
2586: Abs/Rel & $\gamma=1.2$ & $\gamma=1.4$ & $\gamma=1.666$ & $\gamma=2.0$ & $\gamma=2.5$ & $\gamma=3.0$\\
2587: \hline
2588: $10^{-3}/10^{-5}$ & 9.2(-4) & 9.2(-4) & 9.1(-4) & 9.1(-4) & 9.1(-4) & 9.2(-4) \\
2589: $10^{-4}/10^{-6}$ & 5.3(-5) & 4.9(-5) & 5.3(-5) & 5.3(-5) & 5.3(-5) & 5.3(-5)\\
2590: $10^{-5}/10^{-7}$ & 6.7(-5) & 6.7(-5) & 6.7(-5) & 6.7(-5) & 6.7(-5) & 6.7(-5)\\
2591: $10^{-6}/10^{-8}$ & 2.9(-6) & 2.9(-6) & 2.9(-6) & 2.9(-6) & 2.9(-6) & 2.9(-6)\\
2592: \hline
2593: \end{tabular}
2594: \caption{Relative errors in $D(\lambda)$ for the inflow and outflow cases are computed by taking the maximum relative error for 60 contour points evaluated along the semicircle $\phi$. Samples were taken for varying the absolute and relative error tolerances and $\gamma$ in the ODE solver, leaving $L=18$ and $\gamma=1.666$, $v_+=10^{-4}$, and $v_0=0.6$ fixed. Relative errors were computed using the next run as the baseline.}
2595: \end{center}
2596: \end{table}
2597:
2598: \subsection{Numerical convergence study}
2599: \label{errorstudy}
2600: As in \cite{BHRZ}, we carry out a numerical convergence study to show that our results are accurate. We varied the absolute and relative error tolerances, as well as the length of the numerical domain $[-L,L]$. In Tables 1--2, we demonstrate that our choices of $L=18$, {\tt AbsTol=1e-6} and {\tt RelTol=1e-8} provide accurate results.
2601:
2602: %
2603: %
2604: %
2605: %
2606: \appendix
2607:
2608: \section{Proof of preliminary estimate: inflow case}\label{basicproof}
2609: Our starting point is Remark \ref{shockrel}, in which we observed
2610: that the first-order eigensystem \eqref{evans_ode}
2611: in variable $W=(w,u-v,v)^T$ may be converted by the
2612: rescaling $W\to \tilde W:= (w,u-v, \lambda v)^T$ to a system identical
2613: to that of the integrated equations in the shock case; see \cite{PZ}.
2614: Artificially defining $(\tilde u, \tilde v, \tilde v')^T:= \tilde W$, we
2615: obtain a system
2616: \begin{subequations}\label{ep}
2617: \begin{align}
2618: &\lambda \tilde v + \tilde v' - \tilde u' =0, \label{ep:1}\\
2619: &\lambda \tilde u + \tilde u' - \frac{h(\bV)}{\bV^{\gamma+1}} \tilde v'
2620: = \frac{\tilde u''}{\bV}.\label{ep:2}
2621: \end{align}
2622: \end{subequations}
2623: identical to that in the integrated shock case \cite{BHRZ}, but with
2624: boundary conditions
2625: \begin {equation}\label{newbc}
2626: \tilde v(0)=\tilde v'(0)=\tilde u'(0)=0
2627: \end{equation}
2628: imposed at $x=0$.
2629: This new eigenvalue problem differs spectrally from \eqref{eigen1} only at $\lambda=0$, hence spectral stability of \eqref{eigen1} is implied by spectral stability of \eqref{ep}.
2630: Hereafter, we drop the tildes, and refer simply to $u$, $v$.
2631:
2632: With these coordinates, we may establish \eqref{hf} by exactly
2633: the same argument used in the shock case in \cite{BHRZ,HLZ},
2634: for completeness reproduced here.
2635:
2636: \begin{lemma}
2637: The following identity holds for $\R \lambda \geq 0$:
2638: \begin{align}
2639: (\R(\lambda) + |\I (\lambda)|) & \ip \bV |u|^2 + \ip |u'|^2\notag\\
2640: &\leq \sqrt{2} \ip \frac{h(\bV)}{\bV^\gamma} |v'| |u| + \sqrt2\ip \bV |u'||u|\label{id1}.
2641: \end{align}
2642: \end{lemma}
2643:
2644: \begin{proof}
2645: We multiply \eqref{ep:2} by $\bV {\bar u}$ and integrate along $x$. This yields
2646: \[
2647: \lambda \ip \bV |u|^2 + \ip \bV u'\bar{u} + \ip |u'|^2 = \ip \frac{h(\bV)}{\bV^\gamma} v'\bar{u} .
2648: \]
2649: We get \eqref{id1} by taking the real and imaginary parts and adding them together, and noting that $|\R(z)| + |\I(z)| \leq \sqrt{2}|z|$.
2650: \end{proof}
2651:
2652: \begin{lemma}
2653: \label{kawashima}
2654: The following identity holds for $\R \lambda \geq 0$:
2655: %
2656: % TODO (JH) Doesn't fit
2657: %
2658: \begin{equation}
2659: \label{id3}
2660: \ip |u'|^2 = 2\R(\lambda)^2\ip|v|^2 + \R(\lambda)\ip \frac{|v'|^2}{\bV} + \frac{1}{2} \ip \left[\frac{h(\bV)}{\bV^{\gamma+1}} + \frac{a\gamma}{\bV^{\gamma+1}} \right] |v'|^2
2661: \end{equation}
2662: \end{lemma}
2663:
2664: \begin{proof}
2665: We multiply \eqref{ep:2} by ${\bar v'}$ and integrate along $x$. This yields
2666: \[
2667: \lambda \ip u\bar{v}' + \ip u'\bar{v}' - \ip \frac{h(\bV)}{\bV^{\gamma+1}}|v'|^2 = \ip \frac{1}{\bV}u''\bar{v}' = \ip \frac{1}{\bV}(\lambda v' + v''){\bar v'}.
2668: \]
2669: Using \eqref{ep:1} on the right-hand side, integrating by parts, and taking the real part gives
2670: \[
2671: \R \left[ \lambda \ip u\bar{v}' + \ip u'\bar{v}'\right] = \ip \left[\frac{h(\bV)}{\bV^{\gamma+1}} + \frac{\bV_x}{2 \bV^2} \right] |v'|^2 + \R(\lambda)\ip \frac{|v'|^2}{\bV}.
2672: \]
2673: The right hand side can be rewritten as
2674: %
2675: % TODO (JH) Doesn't fit
2676: %
2677: \begin{equation}
2678: \label{id3_1}
2679: \R \left[ \lambda \ip u\bar{v}' + \ip u'\bar{v}'\right] = \frac{1}{2} \ip \left[\frac{h(\bV)}{\bV^{\gamma+1}} + \frac{a\gamma}{\bV^{\gamma+1}} \right] |v'|^2 + \R(\lambda)\ip \frac{|v'|^2}{\bV}.
2680: \end{equation}
2681: Now we manipulate the left-hand side. Note that
2682: \begin{align*}
2683: \lambda \ip u\bar{v}' + \ip u'\bar{v}' &= (\lambda+\bar{\lambda}) \ip u\bar{v}' - \ip u(\bar{\lambda}\bar{v}' + \bar{v}'')\\
2684: &= -2\R(\lambda) \ip u' \bar{v} - \ip u \bar{u}''\\
2685: &= -2\R(\lambda) \ip (\lambda v + v') \bar{v} + \ip |u'|^2.
2686: \end{align*}
2687: Hence, by taking the real part we get
2688: \[
2689: \R \left[ \lambda \ip u\bar{v}' + \ip u'\bar{v}'\right] = \ip |u'|^2 - 2\R(\lambda)^2 \ip |v|^2.
2690: \]
2691: This combines with \eqref{id3_1} to give \eqref{id3}.
2692: \end{proof}
2693:
2694: \begin{lemma}[\cite{BHRZ}]\label{Hlem}
2695: For $h(\bV)$ as in \eqref{f}, we have
2696: \begin{equation}
2697: \label{id2}
2698: \sup_{\bV} \left| \frac{h(\bV)}{\bV^\gamma}\right| = \gamma
2699: \frac{1-v_+}{1-v_+^\gamma}
2700: \leq \gamma,
2701: \end{equation}
2702: where $\bV$ is the profile solution to \eqref{profeq}.
2703: \end{lemma}
2704:
2705: \begin{proof}
2706: Defining
2707: \begin{equation}\label{H}
2708: g(\bV):=h(\bV)\bV^{-\gamma} = -\bV + a(\gamma-1)\bV^{-\gamma} +
2709: (a+1),
2710: \end{equation}
2711: we have $g'(\bV)= -1 -a\gamma(\gamma-1)\bV^{-\gamma-1}<0$ for
2712: $0<v_+\le \bV\le v_-= 1$, hence the maximum of $g$ on $\bV\in
2713: [v_+,v_-]$ is achieved at $\bV=v_+$. Substituting \eqref{RH} into
2714: \eqref{H} and simplifying yields \eqref{id2}.
2715: \end{proof}
2716:
2717: \begin{proof}[Proof of Proposition \ref{hf}]
2718: Using Young's inequality twice on right-hand side of \eqref{id1} together with \eqref{id2}, we get
2719: \begin{align*}
2720: (\R&(\lambda) + |\I (\lambda)|) \ip \bV |u|^2 + \ip |u'|^2 \\
2721: &\leq \sqrt{2} \ip \frac{h(\bV)}{\bV^\gamma} |v'| |u| + \sqrt2\ip \bV |u'||u|\\
2722: &\leq \theta \ip \frac{h(\bV)}{\bV^{\gamma+1}} |v'|^2 + \frac{(\sqrt{2})^2}{4\theta} \ip \frac{h(\bV)}{\bV^\gamma} \bV |u|^2 + \epsilon \ip \bV |u'|^2 + \frac{1}{4 \epsilon} \ip \bV |u|^2\\
2723: &< \theta \ip \frac{h(\bV)}{\bV^{\gamma+1}} |v'|^2 + \epsilon \ip
2724: |u'|^2 + \left[\frac{\gamma}{2\theta} + \frac{1}{2 \epsilon}\right]
2725: \ip \bV |u|^2.
2726: \end{align*}
2727: Assuming that $0<\epsilon<1$ and $\theta = (1-\epsilon)/2$, this simplifies to
2728: \begin{align*}
2729: (\R(\lambda) + |\I (\lambda)|) & \ip \bV |u|^2 + (1-\epsilon) \ip |u'|^2 \\
2730: &<\frac{1-\epsilon}{2} \ip \frac{h(\bV)}{\bV^{\gamma+1}} |v'|^2 +
2731: \left[\frac{\gamma}{2\theta} + \frac{1}{2 \epsilon}\right] \ip \bV
2732: |u|^2.
2733: \end{align*}
2734: Applying \eqref{id3} yields
2735: \[
2736: (\R(\lambda) + |\I (\lambda)|) \ip \bV |u|^2 <
2737: \left[\frac{\gamma}{1-\epsilon} + \frac{1}{2 \epsilon}\right] \ip
2738: \bV |u|^2,
2739: \]
2740: or equivalently,
2741: \[
2742: (\R(\lambda) + |\I (\lambda)|) < \frac{(2 \gamma-1)\epsilon +
2743: 1}{2\epsilon(1-\epsilon)}.
2744: \]
2745: Setting $\epsilon = 1/(2\sqrt{\gamma}+1)$ gives \eqref{hfbounds1}.
2746: \end{proof}
2747:
2748:
2749: \section{Proof of preliminary estimate: outflow case}\label{outbasicproof}
2750:
2751: %\newcommand{\balin}{\begin{align*}}
2752: %\newcommand{\ealin}{\end{align*}}
2753: %\newcommand{\bali}{\begin{align}}
2754: %\newcommand{\eali}{\end{align}}
2755: \newcommand{\ipo}{\int_{\mathbb{R}^-}}
2756: \newcommand{\ipi}[1]{\int_0^\infty{#1}dx}
2757:
2758:
2759:
2760: %CHANGED: grammar-KZ
2761: %Similarly to the inflow case, we can convert the eigenvalue
2762: Similarly as in the inflow case, we can convert the eigenvalue
2763: %ENDCHANGED
2764: equations into the integrated equations as in the shock case; see
2765: \cite{PZ}. Artificially defining $(\tilde u, \tilde v, \tilde
2766: v')^T:= \tilde W$, we obtain a system
2767: \begin{subequations}\label{outep}
2768: \begin{align}
2769: &\lambda \tilde v + \tilde v' - \tilde u' =0, \label{outep:1}\\
2770: &\lambda \tilde u + \tilde u' - \frac{h(\bV)}{\bV^{\gamma+1}}
2771: \tilde v' = \frac{\tilde u''}{\bV}.\label{outep:2}
2772: \end{align}
2773: \end{subequations}
2774: identical to that in the integrated shock case \cite{BHRZ}, but with
2775: boundary conditions
2776: \begin {equation}\label{outnewbc}
2777: %\tilde u(0) \ne 0,\quad
2778: \tilde v'(0) = \frac{\lambda}{\alpha-1}\tilde v(0),\quad \tilde
2779: u'(0) = \alpha\tilde v'(0)
2780: \end{equation}
2781: imposed at $x=0$.
2782: %NOTE: Thus, $\tilde u'(0)=\lambda \alpha/(\alpha -1)\tilde v(0)$, where
2783: %$\alpha \sim 1/\lambda$ for $\lambda $ large,
2784: %or $\tilde u'(0)\sim \tilde v(0)$.
2785: %This shows how $\tilde u'$ boundary terms can be traded for $\tilde v$
2786: %terms for which we have already good control!.
2787: We shall write $w_0$ for $w(0)$, for any function $w$. This new
2788: eigenvalue problem differs spectrally from \eqref{eigen1} only at
2789: $\lambda=0$, hence spectral stability of \eqref{eigen1} is implied
2790: by spectral stability of \eqref{outep}. Hereafter, we drop the
2791: tildes, and refer simply to $u$, $v$.
2792:
2793: %With these coordinates, we may establish \eqref{hf} by exactly the
2794: %same argument used in the shock case in \cite{BHRZ,HLZ}, for
2795: %completeness reproduced here.
2796:
2797: \begin{lemma}
2798: The following identity holds for $\R \lambda \geq 0$:
2799: \begin{align}
2800: (\R(\lambda) + |\I (\lambda)|) & \ipo \bV |u|^2 - \frac{1}{2}\ipo \bV_x |u|^2 + \ipo |u'|^2+\frac12\bv_0|u_0|^2\notag\\
2801: &\leq \sqrt{2} \ipo \frac{h(\bV)}{\bV^\gamma} |v'| |u| + \ipo \bV |u'||u| + \sqrt2|\alpha||v'_0||u_0|\label{outid1}.
2802: \end{align}
2803: \end{lemma}
2804:
2805: \begin{proof}
2806: We multiply \eqref{outep:2} by $\bV {\bar u}$ and integrate along
2807: $x$. This yields
2808: \[
2809: \lambda \ipo \bV |u|^2 + \ipo \bV u'\bar{u} + \ipo |u'|^2 = \ipo
2810: \frac{h(\bV)}{\bV^\gamma} v'\bar{u} +u'_0\bar u_0.
2811: \]
2812: We get \eqref{outid1} by taking the real and imaginary parts and
2813: adding them together, and noting that $|\R(z)| + |\I(z)| \leq
2814: \sqrt{2}|z|$.
2815: \end{proof}
2816:
2817: \begin{lemma}
2818: \label{outkawashima} The following inequality holds for $\R \lambda
2819: \geq 0$:
2820: %
2821: % TODO (JH) Doesn't fit
2822: %
2823: %Changed
2824: %\begin{equation}
2825: %\label{outid3} \frac{1}{2} \ipo \left[\frac{h(\bV)}{\bV^{\gamma+1}} +
2826: %\frac{a\gamma}{\bV^{\gamma+1}} \right] |v'|^2 + \frac
2827: %{|v'_0|^2}{4\bv_0}\le \ipo |u'|^2+\R\lambda
2828: %\ipo \bv|u|^2+{\bv_0|u_0|^2}.
2829: %\end{equation}
2830: \begin{align}\label{outid3} \frac{1}{2} \ipo
2831: \left[\frac{h(\bV)}{\bV^{\gamma+1}} + \frac{a\gamma}{\bV^{\gamma+1}}
2832: \right] |v'|^2 &+ \R(\lambda)\ipo \frac{|v'|^2}{\bV}+ \frac
2833: {|v'_0|^2}{4\bv_0} + 2\R\lambda ^2\ipo |v|^2\notag \\&\le \ipo
2834: |u'|^2 +\bv_0|u_0|^2.\end{align}
2835: %ENDCHANGED
2836:
2837: \end{lemma}
2838:
2839: \begin{proof}
2840: We multiply \eqref{outep:2} by ${\bar v'}$ and integrate along $x$.
2841: This yields
2842: \[
2843: \lambda \ipo u\bar{v}' + \ipo u'\bar{v}' - \ipo
2844: \frac{h(\bV)}{\bV^{\gamma+1}}|v'|^2 = \ipo \frac{1}{\bV}u''\bar{v}'
2845: = \ipo \frac{1}{\bV}(\lambda v' + v''){\bar v'}.
2846: \]
2847: Using \eqref{outep:1} on the right-hand side, integrating by parts,
2848: and taking the real part gives
2849: \[
2850: \R \left[ \lambda \ipo u\bar{v}' + \ipo u'\bar{v}'\right] = \ipo
2851: \left[\frac{h(\bV)}{\bV^{\gamma+1}} + \frac{\bV_x}{2 \bV^2} \right]
2852: |v'|^2 + \R(\lambda)\ipo \frac{|v'|^2}{\bV} + \frac
2853: {|v'_0|^2}{2\bv_0}.
2854: \]
2855: The right hand side can be rewritten as
2856: %
2857: % TODO (JH) Doesn't fit
2858: %
2859: \begin{align}
2860: \label{outid3_1} \R &\left[ \lambda \ipo u\bar{v}' + \ipo
2861: u'\bar{v}'\right] \notag\\&= \frac{1}{2} \ipo
2862: \left[\frac{h(\bV)}{\bV^{\gamma+1}} + \frac{a\gamma}{\bV^{\gamma+1}}
2863: \right] |v'|^2 + \R(\lambda)\ipo \frac{|v'|^2}{\bV}+ \frac
2864: {|v'_0|^2}{2\bv_0}.
2865: \end{align}
2866: Now we manipulate the left-hand side. Note that
2867: %CHANGED
2868: %\begin{align*}
2869: %\lambda \ipo u\bar{v}' + \ipo u'\bar{v}' &= (\lambda+\bar{\lambda}) \ipo u\bar{v}' + \ipo (u'\bar{v}' - \bar \lambda u \bar v')\\
2870: %&= 2\R(\lambda) \ipo u \bar{v}'+\ipo u'(\bar v' + \bar \lambda \bar v)
2871: %-\bar \lambda u_0 \bar
2872: %v_0\\
2873: %&= 2\R(\lambda) \ipo u \bar{v}' +\ipo |u'|^2 -\bar \lambda u_0 \bar
2874: %v_0.
2875: %\end{align*}
2876: %Hence, by taking the real part we get
2877: %\[
2878: %\R \left[ \lambda \ipo u\bar{v}' + \ipo u'\bar{v}'\right] = \ipo |u'|^2
2879: %+ 2\R(\lambda)\R\ipo u \bar{v}'-\R((\bar \alpha-1) \bar v'_0u_0).
2880: %\]
2881: %This combines with \eqref{outid3_1} to give
2882: %\begin{align*} \frac{1}{2} \ipo
2883: %\left[\frac{h(\bV)}{\bV^{\gamma+1}} + \frac{a\gamma}{\bV^{\gamma+1}}
2884: %\right] |v'|^2 &+ \R(\lambda)\ipo \frac{|v'|^2}{\bV}+ \frac
2885: %{|v'_0|^2}{2\bv_0}\\& \le \ipo |u'|^2 +2\R(\lambda)\ipo |u
2886: %\bar{v}'|+|\alpha-1| | v'_0||u_0|.\end{align*}
2887: \begin{align*}
2888: \lambda \ipo u\bar{v}' &+ \ipo u'\bar{v}' = (\lambda+\bar{\lambda}) \ipo u\bar{v}' + \ipo (u'\bar{v}' - \bar \lambda u \bar v')\\
2889: &= -2\R(\lambda) \ipo u' \bar{v} + 2\R\lambda u_0\bar v_0+\ipo
2890: u'(\bar v' + \bar \lambda \bar v) -\bar \lambda u_0 \bar
2891: v_0\\
2892: &= -2\R(\lambda) \ipo (\lambda v + v') \bar{v} +\ipo |u'|^2
2893: +2\R\lambda u_0\bar v_0-\bar \lambda u_0 \bar v_0.
2894: \end{align*}
2895: Hence, by taking the real part and noting that $$\R(2\R\lambda
2896: u_0\bar v_0-\bar \lambda u_0 \bar v_0) =\R\lambda \R(u_0\bar
2897: v_0)-\I\lambda \I(u_0 \bar v_0) = \R(\lambda u_0\bar v_0)$$ we get
2898: \[
2899: \R \left[ \lambda \ipo u\bar{v}' + \ipo u'\bar{v}'\right] = \ipo
2900: |u'|^2 - 2\R\lambda ^2\ipo |v|^2 - \R\lambda |v_0|^2 + \R(\lambda
2901: u_0\bar v_0).
2902: \]
2903:
2904: This combines with \eqref{outid3_1} to give
2905: \begin{align*} \frac{1}{2} \ipo
2906: \left[\frac{h(\bV)}{\bV^{\gamma+1}} + \frac{a\gamma}{\bV^{\gamma+1}}
2907: \right] |v'|^2 &+ \R(\lambda)\ipo \frac{|v'|^2}{\bV}+ \frac
2908: {|v'_0|^2}{2\bv_0} + 2\R\lambda ^2\ipo |v|^2\\& + \R\lambda
2909: |v_0|^2 = \ipo |u'|^2 +\R(\lambda u_0\bar v_0).\end{align*}
2910: %ENDCHANGED
2911:
2912: We get \eqref{outid3} by observing that \eqref{outnewbc} and Young's
2913: inequality yield
2914: $$|\R(\lambda u_0\bar v_0)|\le |\alpha-1||v'_0v_0| \le |v'_0v_0|\le \frac{|v'_0|^2}{4\bv_0} +
2915: \bv_0|u_0|^2.$$ Here we used $|\alpha-1| =
2916: \frac{|\lambda|}{|\lambda-\bv'_0|}\le 1$. Note that $\R\lambda \ge0$
2917: and $\bv'_0\le0$.\end{proof}
2918:
2919: \begin{proof}[Proof of Proposition \ref{hf}]
2920: Using Young's inequality twice on right-hand side of \eqref{outid1}
2921: together with \eqref{id2}, and denoting the boundary term on the
2922: right by $I_b$, we get
2923: \begin{align*}
2924: (\R&(\lambda) + |\I (\lambda)|) \ipo \bV |u|^2 - \frac{1}{2}\ipo \bV_x |u|^2 + \ipo |u'|^2+\frac12\bv_0|u_0|^2 \\
2925: &\leq \sqrt{2} \ipo \frac{h(\bV)}{\bV^\gamma} |v'| |u| + \ipo \bV |u'||u|+I_b\\
2926: &\leq \theta \ipo \frac{h(\bV)}{\bV^{\gamma+1}} |v'|^2 + \frac{1}{2\theta} \ipo \frac{h(\bV)}{\bV^\gamma} \bV |u|^2 + \epsilon \ipo \bV |u'|^2 + \frac{1}{4 \epsilon} \ipo \bV |u|^2+I_b\\
2927: &< \theta \ipo \frac{h(\bV)}{\bV^{\gamma+1}} |v'|^2 + \epsilon
2928: \ipo |u'|^2 + \left[\frac{\gamma}{2\theta} + \frac{1}{4
2929: \epsilon}\right] \ipo \bV |u|^2+I_b.
2930: \end{align*}
2931: Here we treat the boundary term by
2932: \begin{align*}I_b&\le\sqrt2|\alpha||v'_0||u_0|\le \frac \theta 2 \frac{|v'_0|^2}{\bv_0}+\frac 1\theta |\alpha|^2 \bv_0|u_0|^2.\end{align*}
2933:
2934: Therefore using \eqref{outid3}, we simply obtain from the above
2935: estimates
2936: %CHANGED
2937: %\begin{align*}
2938: %(\R(\lambda) &+ |\I (\lambda)|) \ipo \bV |u|^2 + (1-\epsilon )\ipo |u'|^2 +\frac12\bv_0|u_0|^2\\
2939: %&<\theta \ipo \frac{h(\bV)}{\bV^{\gamma+1}} |v'|^2 +\frac \theta 2
2940: %\frac{|v'_0|^2}{\bv_0} + \left[\frac{\gamma}{2\theta} + \frac{1}{4
2941: %\epsilon}\right] \ipo \bV |u|^2+\frac 1\theta |\alpha|^2
2942: %\bv_0|u_0|^2\\&< 2\theta \ipo |u'|^2 +\left[\frac{\gamma}{2\theta} +
2943: %\frac{1}{4 \epsilon}+2\theta\R\lambda\right] \ipo \bV |u|^2+J_b
2944: %\end{align*} where $J_b:=(\frac
2945: %1\theta |\alpha|^2 +2\theta )\bv_0|u_0|^2.$ Assuming that $\epsilon
2946: %+ 2\theta \le 1$, this simplifies to
2947: %\begin{align*}
2948: %(\R(\lambda) + |\I (\lambda)|) & \ipo \bV |u|^2
2949: %+\frac12\bv_0|u_0|^2<\left[\frac{\gamma}{2\theta} + \frac{1}{4
2950: %\epsilon}+2\theta\R\lambda\right] \ipo \bV |u|^2+J_b.
2951: %\end{align*}
2952: \begin{align*}
2953: (\R(\lambda) &+ |\I (\lambda)|) \ipo \bV |u|^2 + (1-\epsilon )\ipo |u'|^2 +\frac12\bv_0|u_0|^2\\
2954: &<\theta \ipo \frac{h(\bV)}{\bV^{\gamma+1}} |v'|^2 +\frac \theta 2
2955: \frac{|v'_0|^2}{\bv_0} + \left[\frac{\gamma}{2\theta} + \frac{1}{4
2956: \epsilon}\right] \ipo \bV |u|^2+\frac 1\theta |\alpha|^2
2957: \bv_0|u_0|^2\\&< 2\theta \ipo |u'|^2 +\left[\frac{\gamma}{2\theta}
2958: + \frac{1}{4 \epsilon}\right] \ipo \bV |u|^2+J_b
2959: \end{align*} where $J_b:=(\frac
2960: 1\theta |\alpha|^2 +2\theta )\bv_0|u_0|^2.$ Assuming that $\epsilon
2961: + 2\theta \le 1$, this simplifies to
2962: \begin{align*}
2963: (\R(\lambda) + |\I (\lambda)|) & \ipo \bV |u|^2
2964: +\frac12\bv_0|u_0|^2<\left[\frac{\gamma}{2\theta} + \frac{1}{4
2965: \epsilon}\right] \ipo \bV |u|^2+J_b.
2966: \end{align*}
2967:
2968:
2969: Note that $|\alpha|\le \frac{-\bv'_0}{|\lambda|}\le \frac
2970: 1{4|\lambda|}$. Therefore for $|\lambda|\ge \frac1{4\theta}$, we get
2971: $|\alpha|\le \theta$ and $J_b\le 3\theta\bv_0|u_0|^2.$ For sake of
2972: simplicity, choose $\theta =1/6$ and $\epsilon = 2/3$. This shows
2973: that $J_b$ can be absorbed into the left by the term $\frac
2974: 12\bv_0|u_0|^2$ and thus we get
2975: \begin{align*}
2976: (\R(\lambda) + |\I (\lambda)|) & \ipo \bV |u|^2 <\left[
2977: \frac{\gamma}{2\theta}+ \frac{1}{4 \epsilon}\right] \ipo \bV |u|^2
2978: = \left[ 3\gamma+ \frac{3}{8}\right] \ipo \bV |u|^2,
2979: \end{align*} provided that $|\lambda|\ge 1/(4\theta) =3/2$.
2980:
2981: This shows
2982: \[
2983: (\R(\lambda) + |\I (\lambda)|) < \max\{\frac{3\sqrt2}{2},3\gamma+
2984: \frac{3}{8}\}.
2985: \]
2986: \end{proof}
2987:
2988: \section{Nonvanishing of $D^0_{\rm in}$}\label{stronglimit}
2989:
2990: Working in $(\tilde v, \tilde u)$ variables as in \eqref{ep},
2991: the limiting eigenvalue system and boundary
2992: conditions take the form
2993: \begin{subequations}\label{limitep}
2994: \begin{align}
2995: &\lambda \tilde v + \tilde v' - \tilde u' =0, \label{limitep:1}\\
2996: &\lambda \tilde u + \tilde u' - \frac{1-\bV}{\bV} \tilde v' =
2997: \frac{\tilde u''}{\bV}.\label{limitep:2}
2998: \end{align}
2999: \end{subequations}
3000: corresponding to a pressureless gas, $\gamma=0$,
3001: with
3002: \begin{equation}\label{PLBC}
3003: (\tilde u,\tilde u',\tilde v,\tilde v')(0)=(d,0,0,0),\:\:
3004: (\tilde u,\tilde u',\tilde v,\tilde v')(+\infty)=(c,0,0,0).
3005: \end{equation}
3006: Hereafter, we drop the tildes.
3007:
3008: \begin{proof}[Proof of Proposition \ref{redenergy}]
3009: Multiplying \eqref{limitep:2} by $\bV \bar{u}/(1-\bV)$ and integrating on
3010: $[0,b]\subset\mathbb{R}^+$, we obtain
3011: \[
3012: \lambda \int^b_0 \frac{\bV}{1-\bV}|u|^2 dx + \int^b_0 \frac{\bV}{1-\bV}
3013: u'\bar{u} dx - \int^b_0 v'\bar{u} dx = \int^b_0\frac{u'' \bar{u}}{1-\bV}
3014: dx.
3015: \]
3016: Integrating the third and fourth terms by parts yields
3017: \begin{align*}
3018: \lambda \int^b_0 \frac{\bV}{1-\bV}|u|^2 dx &+ \int^b_0 \left[
3019: \frac{\bV}{1-\bV} + \left( \frac{1}{1-\bV} \right)'\right] u'\bar{u} dx \\
3020: &\quad+
3021: \int^b_0 \frac{|u'|^2}{1-\bV} dx
3022: + \int^b_0 v (\overline{\lambda v + v'}) dx \\
3023: &= \left[ v \bar{u} + \frac{u'\bar{u}}{1-\bV} \right] \Big|^b_0.\\
3024: \end{align*}
3025: Taking the real part, we have
3026:
3027: \begin{align}
3028: &\R(\lambda) \int^b_0 \left( \frac{\bV}{1-\bV}|u|^2 + |v|^2\right) dx +
3029: \int^b_0 g(\bV) |u|^2 dx + \int^b_0 \frac{|u'|^2}{1-\bV} dx\notag\\
3030: &\quad = \R \left[ v \bar{u} +
3031: \frac{u'\bar{u}}{1-\bV} - \frac{1}{2}\left[\frac{\bV}{1-\bV} + \left(
3032: \frac{1}{1-\bV} \right)'\right] |u|^2 - \frac{|v|^2 }{2} \right] \Big|^b_0,\label{MNen}
3033: \end{align}
3034: where
3035: \[
3036: g(\bV) = -\frac{1}{2}\left[ \left(\frac{\bV}{1-\bV}\right)' + \left(
3037: \frac{1}{1-\bV} \right)''\right].
3038: \]
3039: Note that
3040: \[
3041: \frac{d}{dx}\left(\frac{1}{1-\bV}\right) = - \frac{(1-\bV)'}{(1-\bV)^2} =
3042: \frac{\bV_x}{(1-\bV)^2} = \frac{\bV(\bV-1)}{(1-\bV)^2} =
3043: -\frac{\bV}{1-\bV}.
3044: \]
3045: Thus, $g(\bV)\equiv 0$ and the third term on the right-hand side vanishes,
3046: leaving
3047: \begin{align*}
3048: \R(\lambda) \int^b_0 \left( \frac{\bV}{1-\bV}|u|^2 + |v|^2\right) dx &+
3049: \int^b_0 \frac{|u'|^2}{1-\bV} dx\\
3050: &\quad = \left[ \R(v \bar{u}) + \frac{\R(u'\bar{u})}{1-\bV} - \frac{|v|^2
3051: }{2} \right] \Big|^b_0\\
3052: &\quad = \left[ \R(v \bar{u}) + \frac{\R(u'\bar{u})}{1-\bV} - \frac{|v|^2
3053: }{2} \right](b).\\
3054: \end{align*}
3055:
3056: We show finally that the right-hand side goes to zero in the limit as
3057: $b\rightarrow\infty$.
3058: By Proposition \ref{conjugation}, the behavior of $u$, $v$ near
3059: $\pm \infty$ is governed by the limiting constant--coefficient
3060: systems $W'=A^0_\pm(\lambda)W$, where $W=(u,v,v')^T$
3061: and $A^0_\pm =A^0(\pm \infty, \lambda)$.
3062: In particular, solutions $W$ asymptotic to $(1,0,0)$ at
3063: $x=+\infty$ decay exponentially in $(u',v,v')$ and are bounded
3064: in coordinate $u$ as $x\to +\infty$.
3065: Observing that $1-\hat v\to 1$ as $x\to +\infty$, we thus see immediately
3066: that the boundary contribution at $b$ vanishes as $b\to +\infty$.
3067:
3068: Thus, in the limit as $b\to +\infty$,
3069: \begin{equation}
3070: \R(\lambda) \int^{+\infty}_{0} \left( \frac{\bV}{1-\bV}|u|^2 + |v|^2\right) dx
3071: + \int^{+\infty}_{0} \frac{|u'|^2}{1-\bV} dx=0.
3072: \end{equation}
3073: But, for $\R \lambda\ge 0$, this implies $u'\equiv 0$,
3074: or $u\equiv \hbox{\rm constant}$, which, by $u(0)=1$,
3075: implies $u\equiv 1$.
3076: This reduces \eqref{limitep:1} to $v'=\lambda v$, yielding the
3077: explicit solution $v=Ce^{\lambda x}$. By $v(0)=0$,
3078: therefore, $v\equiv 0$ for $\R \lambda\ge 0$.
3079: Substituting into \eqref{limitep:2}, we obtain $\lambda =0$.
3080: It follows that there are no nontrivial solutions of \eqref{limitep},
3081: \eqref{PLBC} for $\R \lambda\ge 0$ except at $\lambda=0$.
3082: %
3083: %On the other hand, $(u,v,v')\equiv (1,0,0)$ is evidently a solution
3084: %for $\lambda=0$, which translates back to a solution $(1,0,0)$
3085: %in the original coordinates as well.
3086: \end{proof}
3087:
3088: \begin{remark}\label{symmetrizers}
3089: The above energy estimate is essentially identical to that used
3090: in \cite{HLZ} to treat the limiting shock case.
3091: \end{remark}
3092:
3093: \section{Nonvanishing of $D^0_{\rm out}$}\label{outnonv}
3094:
3095: Working in $(\tilde v, \tilde u)$ variables as in \eqref{ep},
3096: the limiting eigenvalue system and boundary
3097: conditions take the form
3098: \begin{subequations}\label{outlimitep}
3099: \begin{align}
3100: &\lambda \tilde v + \tilde v' - \tilde u' =0, \label{outlimitep:1}\\
3101: &\lambda \tilde u + \tilde u' - \frac{1-\bV}{\bV} \tilde v' =
3102: \frac{\tilde u''}{\bV}.\label{outlimitep:2}
3103: \end{align}
3104: \end{subequations}
3105: corresponding to a pressureless gas, $\gamma=0$,
3106: with
3107: \begin{equation}\label{outPLBC}
3108: (\tilde u,\tilde u',\tilde v,\tilde v')(-\infty)=(0,0,0,0),
3109: \end{equation}
3110: \begin {equation}
3111: \tilde v'(0) = \frac{\lambda}{\alpha-1}\tilde v(0),\quad \tilde
3112: u'(0) = \alpha\tilde v'(0).
3113: \end{equation}
3114: In particular,
3115: \begin{equation}\label{useful}
3116: \tilde u'(0)=\frac{\lambda \alpha}{\alpha -1}\tilde v(0)=
3117: \hat v'(0)\tilde v(0)
3118: =
3119: (v_0-1)\hat v_0 \tilde v(0).
3120: \end{equation}
3121: Hereafter, we drop the tildes.
3122:
3123: \begin{proof}[Proof of Proposition \ref{redenergy}]
3124: Multiplying \eqref{outlimitep:2} by $\bV \bar{u}/(1-\bV)$ and integrating on
3125: $[a,0]\subset\mathbb{R}^-$, we obtain
3126: \[
3127: \lambda \int^0_a \frac{\bV}{1-\bV}|u|^2 dx + \int^b_a \frac{\bV}{1-\bV}
3128: u'\bar{u} dx - \int^0_a v'\bar{u} dx = \int^0_a\frac{u'' \bar{u}}{1-\bV}
3129: dx.
3130: \]
3131: Integrating the third and fourth terms by parts yields
3132: \begin{align*}
3133: \lambda \int^0_a \frac{\bV}{1-\bV}|u|^2 dx &+ \int^0_a \left[
3134: \frac{\bV}{1-\bV} + \left( \frac{1}{1-\bV} \right)'\right] u'\bar{u} dx \\
3135: &\quad+
3136: \int^0_a \frac{|u'|^2}{1-\bV} dx
3137: + \int^0_a v (\overline{\lambda v + v'}) dx \\
3138: &= \left[ v \bar{u} + \frac{u'\bar{u}}{1-\bV} \right] \Big|^0_a.\\
3139: \end{align*}
3140: Taking the real part, we have
3141:
3142: \begin{align}
3143: &\R(\lambda) \int^0_a \left( \frac{\bV}{1-\bV}|u|^2 + |v|^2\right) dx +
3144: \int^0_a g(\bV) |u|^2 dx + \int^0_a \frac{|u'|^2}{1-\bV} dx\notag\\
3145: &\quad = \R \left[ v \bar{u} +
3146: \frac{u'\bar{u}}{1-\bV} - \frac{1}{2}\left[\frac{\bV}{1-\bV} + \left(
3147: \frac{1}{1-\bV} \right)'\right] |u|^2 - \frac{|v|^2 }{2} \right] \Big|^0_a,\label{MNen2}
3148: \end{align}
3149: where
3150: \[
3151: g(\bV) = -\frac{1}{2}\left[ \left(\frac{\bV}{1-\bV}\right)' + \left(
3152: \frac{1}{1-\bV} \right)''\right]\equiv 0
3153: \]
3154: and the third term on the right-hand side vanishes,
3155: as shown in Section \ref{stronglimit}, leaving
3156: \begin{align*}
3157: \R(\lambda) \int^0_a \left( \frac{\bV}{1-\bV}|u|^2 + |v|^2\right) dx &+
3158: \int^0_a \frac{|u'|^2}{1-\bV} dx\\
3159: &\quad = \left[ \R(v \bar{u}) + \frac{\R(u'\bar{u})}{1-\bV} - \frac{|v|^2
3160: }{2} \right] \Big|^0_a.\\
3161: \end{align*}
3162:
3163: A boundary analysis similar to that of Section \ref{stronglimit}
3164: shows that the contribution at $a$ on the righthand side
3165: vanishes as $a\to -\infty$; see \cite{HLZ} for details.
3166: Thus, in the limit as $a\to -\infty$ we obtain
3167: \begin{align*}
3168: \R(\lambda) \int^0_{-\infty} \left( \frac{\bV}{1-\bV}|u|^2 + |v|^2\right) dx &+
3169: \int^0_{-\infty} \frac{|u'|^2}{1-\bV} dx\\
3170: &\quad = \left[ \R(v \bar{u}) + \frac{\R(u'\bar{u})}{1-\bV} - \frac{|v|^2
3171: }{2} \right] (0)\\
3172: &\quad = \left[(1-v_0)\R(v \bar{u}) - \frac{|v|^2
3173: }{2} \right] (0),\\
3174: &\quad \le
3175: \left[(1- v_0) |v||u| - \frac{|v|^2 }{2}\right](0)\\
3176: &\quad \le (1- v_0 )^2\frac{|u(0)|^2}{2},
3177: \end{align*}
3178: where the second equality follows by \eqref{useful} and the
3179: final line by Young's inequality.
3180:
3181: Next, observe the Sobolev-type bound
3182: $$
3183: \begin{aligned}
3184: |u(0)|^2 &\le \Big(
3185: \int_{-\infty}^0 |u'(x)| dx
3186: \Big)^2
3187: \le
3188: \int_{-\infty}^0 \frac{|u'|^2}{1-\hat v}(x) dx
3189: \int_{-\infty}^0 (1-\hat v)(x) dx,
3190: \end{aligned}
3191: $$
3192: together with
3193: $$
3194: \begin{aligned}
3195: \int_{-\infty}^0 (1-\hat v)(x) dx&=
3196: \int_{-\infty}^0 -\frac{\hat v'}{\hat v}(x)dx
3197: =\int_{-\infty}^0 (\log \hat v^{-1})'(x) dx
3198: = \log v_0^{-1},
3199: \end{aligned}
3200: $$
3201: hence
3202: $\int_{-\infty}^0 (1-\hat v)(x) dx< \frac{2}{(1-v_0)^2}$
3203: for $v_0>v_*$, where $v_*<e^{-2}$ is the unique solution of
3204: \begin{equation}\label{fnl}
3205: v_*=e^{-2/(1-v_*)^2}.
3206: \end{equation}
3207:
3208:
3209: Thus, for $v_0>v_*$,
3210: \begin{equation}
3211: \R(\lambda) \int^{0}_{-\infty} \left( \frac{\bV}{1-\bV}|u|^2 + |v|^2\right) dx
3212: + \epsilon \int^{0}_{-\infty} \frac{|u'|^2}{1-\bV} dx\le 0,
3213: \end{equation}
3214: for $\epsilon:= \frac{(1-v_0)^2}{2}-
3215: \frac{1}{\int_{-\infty}^0 (1-\hat v)(x) dx}>0$.
3216: For $\R \lambda\ge 0$, this implies $u'\equiv 0$,
3217: or $u\equiv \hbox{\rm constant}$, which, by $u(-\infty)=0$,
3218: implies $u\equiv 0$.
3219: This reduces \eqref{outlimitep:1} to $v'=\lambda v$, yielding the
3220: explicit solution $v=Ce^{\lambda x}$. By $v(0)=0$,
3221: therefore, $v\equiv 0$ for $\R \lambda\ge 0$.
3222: It follows that there are no nontrivial solutions of \eqref{outlimitep},
3223: \eqref{outPLBC} for $\R \lambda\ge 0$ except at $\lambda=0$.
3224:
3225: By iteration, starting with $v_*\approx 0$, we obtain
3226: first $v_*<e^{-2}\approx 0.14$ then $v_*> e^{2/(1-.14)^2}\approx .067$,
3227: then $v_*< e^{2/(1-.067)^2}\approx .10$, then
3228: $v_*> e^{2/(1-.10)^2}\approx .085$, then
3229: $v_*< e^{2/(1-.085)}\approx .091$ and
3230: $v_*> e^{2/(1-.091)}\approx .0889$, terminating with $v_*\approx .0899$.
3231: %TODO: improve the above?
3232: \end{proof}
3233:
3234: \begin{remark}\label{smallcase}
3235: Our Evans function results show that
3236: the case $v_0$ small not treated corresponds to the shock
3237: limit for which stability is already known by \cite{HLZ}.
3238: This suggests that a more sophisticated energy estimate
3239: combining the above with a boundary-layer analysis from
3240: $x=0$ back to $x=L+\delta$ might yield nonvanishing for all $1>v_0>0$.
3241: \end{remark}
3242:
3243: \section{The characteristic limit: outflow case}\label{char}
3244:
3245: We now show stability of compressive outflow boundary layers in the
3246: characteristic limit $v_+\to 1$, by essentially the same energy
3247: estimate used in \cite{MN} to show stability of small-amplitude
3248: shock waves.
3249:
3250: As in the above section on the outflow case, we obtain a system
3251: \begin{subequations}\label{charep}
3252: \begin{align}
3253: &\lambda \tilde v + \tilde v' - \tilde u' =0, \label{charep:1}\\
3254: &\lambda \tilde u + \tilde u' - \frac{h(\bV)}{\bV^{\gamma+1}}
3255: \tilde v' = \frac{\tilde u''}{\bV}.\label{charep:2}
3256: \end{align}
3257: \end{subequations}
3258: identical to that in the integrated shock case \cite{BHRZ}, but with
3259: boundary conditions
3260: \begin {equation}
3261: \tilde v'(0) = \frac{\lambda}{\alpha-1}\tilde v(0),\quad \tilde
3262: u'(0) = \alpha\tilde v'(0).
3263: \end{equation}
3264: In particular,
3265: \begin{equation}
3266: \tilde u'(0)=\frac{\lambda \alpha}{\alpha -1}\tilde v(0)= \hat
3267: v'(0)\tilde v(0).
3268: \end{equation}
3269: This new eigenvalue problem differs spectrally from \eqref{eigen1}
3270: only at $\lambda=0$, hence spectral stability of \eqref{eigen1} is
3271: implied by spectral stability of \eqref{charep}. Hereafter, we drop
3272: the tildes, and refer simply to $u$, $v$.
3273:
3274: \begin{proof} [Proof of Proposition \ref{charsmallamp}]
3275:
3276: \newcommand{\ipco}[1]{\int_{-\infty}^0 {#1} dx}
3277:
3278: We note that $h(\bV) > 0$. By multiplying \eqref{charep:2} by both
3279: the conjugate $\bar{u}$ and $\bV^{\gamma+1}/h(\bV)$ and integrating
3280: along $x$ from $-\infty$ to $0$, we have
3281: \[
3282: \ipco{ \frac{\lambda u \bar{u}\bV^{\gamma+1}}{h(\bV)} }+ \ipco{
3283: \frac{u' \bar{u}\bV^{\gamma+1}}{h(\bV)} }- \ipco{ v' \bar{u}} =
3284: \ipco{ \frac{u''\bar{u}\bV^\gamma}{h(\bV)}}.
3285: \]
3286: Integrating the last three terms by parts and appropriately using
3287: \eqref{charep:1} to substitute for $u'$ in the third term gives us
3288: \begin{align*}
3289: \ipco{ \frac{\lambda |u|^2 \bV^{\gamma+1}}{h(\bV)} }&+ \ipco{
3290: \frac{u' \bar{u}\bV^{\gamma+1}}{h(\bV)}} + \ipco{ v
3291: (\overline{\lambda v + v'})} + \ipco{
3292: \frac{\bV^\gamma|u'|^2}{h(\bV)}}
3293: \\&= -\ipco{ \left(\frac{\bV^\gamma}{h(\bV)}\right)' u'\bar{u}} +
3294: \left[v\bar u+\frac{v^\gamma u'\bar u}{h(\bv)}\right]\Big|_{x=0}.
3295: \end{align*}
3296: We take the real part and appropriately integrate by parts to get
3297: \begin{align}\label{outid2}
3298: \R(\lambda)\ipco{ \left[ \frac{\bV^{\gamma+1}}{h(\bV)}|u|^2+|v|^2
3299: \right]} + \ipco{ g(\bV) |u|^2} + \ipco{
3300: \frac{\bV^\gamma}{h(\bV)}|u'|^2}= G(0),
3301: \end{align}
3302: where
3303: \[
3304: g(\bV) = - \frac{1}{2}
3305: \left[\left(\frac{\bV^{\gamma+1}}{h(\bV)}\right)' +
3306: \left(\frac{\bV^\gamma}{h(\bV)}\right)'' \right]
3307: \]
3308: and
3309: \[
3310: G(0) = - \frac{1}{2} \left[\frac{\bV^{\gamma+1}}{h(\bV)} +
3311: \left(\frac{\bV^\gamma}{h(\bV)}\right)' \right]|u|^2 + \R\left[v\bar
3312: u+\frac{v^\gamma u'\bar u}{h(\bv)}\right] - \frac{|v|^2}{2}
3313: \]
3314: evaluated at $x=0$. Here, the boundary term appearing on the
3315: righthand side is the only difference from the corresponding
3316: estimate appearing in the treatment of the shock case in \cite{MN,
3317: BHRZ}. We shall show that as $\bv_+\to1$, the boundary term $G(0)$
3318: is nonpositive. Observe that boundary conditions yield
3319: $$\left[v\bar u+\frac{v^\gamma u'\bar u}{h(\bv)}\right]\Big|_{x=0} =
3320: \R(v(0)\bar u(0))\left[1+\frac{\bv^\gamma
3321: \bv'}{h(\bv)}\right]\Big|_{x=0}.$$
3322:
3323:
3324: We first note, as established in \cite{MN,BHRZ}, that $g(\bV) \geq
3325: 0$ on $[v_+,1]$, under certain conditions including the case
3326: $\bv_+\to1$. Straightforward computation gives identities:
3327: \begin{align}
3328: \gamma h(\bV) - \bV h'(\bV) &= a\gamma(\gamma-1) + \bV^{\gamma+1}\quad\mbox{and}\label{charell1}\\
3329: \bV^{\gamma-1}\bV_x &= a\gamma - h(\bV)\label{charI2}.
3330: \end{align}
3331: Using \eqref{charell1} and \eqref{charI2}, we abbreviate a few
3332: intermediate steps below:
3333: \begin{align}
3334: g(\bV) &= -\frac{\bV_x}{2}\left[ \frac{(\gamma+1)\bV^\gamma h(\bV) - \bV^{\gamma+1}h'(\bV)}{h(\bV)^2} + \frac{d}{d\bV}\left[ \frac{\gamma \bV^{\gamma-1}h(\bV)-\bV^\gamma h'(\bV)}{h(\bV)^2} \bV_x \right]\right]\notag\\
3335: &= -\frac{\bV_x}{2}\left[ \frac{\bV^\gamma\left((\gamma+1)h(\bV) - \bV h'(\bV)\right)}{h(\bV)^2} + \frac{d}{d\bV}\left[ \frac{\gamma h(\bV)-\bV h'(\bV)}{h(\bV)^2} (a\gamma-h(\bV)) \right]\right]\notag\\
3336: %
3337: &=-\frac{a\bV_x\bV^{\gamma-1}}{2 h(\bV)^3} \times\notag\\
3338: & \qquad\left[ \gamma^2(\gamma+1)\bV^{\gamma+2} - 2 (a+1)\gamma(\gamma^2-1)\bV^{\gamma+1}+(a+1)^2\gamma^2(\gamma-1)\bV^\gamma\right.\notag\\
3339: &\qquad\qquad +\left. a\gamma(\gamma+2)(\gamma^2-1)\bV-a (a+1) \gamma^2 (\gamma^2-1) \right]\notag\\
3340: &=-\frac{a\bV_x\bV^{\gamma-1}}{2 h(\bV)^3}[(\gamma+1)\bV^{\gamma+2}+\bV^\gamma(\gamma-1)\left((\gamma+1)\bV-(a+1)\gamma\right)^2 \label{charpreMN}\\
3341: &\qquad + a\gamma(\gamma^2-1)(\gamma+2)\bV-a (a+1)\gamma^2(\gamma^2-1)]\notag\\
3342: &\geq -\frac{a\bV_x\bV^{\gamma-1}}{2 h(\bV)^3}[(\gamma+1)\bV^{\gamma+2}+ a\gamma(\gamma^2-1)(\gamma+2)\bV-a (a+1)\gamma^2(\gamma^2-1)]\notag\\
3343: &\geq-\frac{\gamma^2 a^3 \bV_x (\gamma+1)}{2 h(\bV)^3
3344: v_+}\left[\left(\frac{v_+^{\gamma+1}}{a\gamma}\right)^2+2(\gamma-1)\left(\frac{v_+^{\gamma+1}}{a\gamma}\right)-(\gamma-1)\right].\label{charMN}
3345: \end{align}
3346: This verifies $g(\bv) \ge 0$ as $\bv_+\to1$.
3347:
3348: Second, examine
3349: \[ G(0) = - \frac{1}{2} \left[\frac{\bV^{\gamma+1}}{h(\bV)} +
3350: \left(\frac{\bV^\gamma}{h(\bV)}\right)'
3351: \right]|u(0)|^2+\left[1+\frac{\bv^\gamma
3352: \bv'}{h(\bv)}\right]\R(v(0)\bar u(0))- \frac{|v(0)|^2}{2}.
3353: \]
3354:
3355: Applying Young's inequality to the middle term, we easily get $$G(0)
3356: \le - \frac{1}{2} \left[\frac{\bV^{\gamma+1}}{h(\bV)} +
3357: \left(\frac{\bV^\gamma}{h(\bV)}\right)' -\left(1+\frac{\bv^\gamma
3358: \bv'}{h(\bv)}\right)^2\right]|u(0)|^2 =: -\frac 12 I|u(0)|^2.$$
3359:
3360: Now observe that $I$ can be written as
3361: \begin{align*} I=\frac{\bV^{\gamma+1}}{h(\bV)} &-1+
3362: \left[\frac{\gamma\bv^{\gamma-1}}{h(\bv)}-\frac{2\bv^\gamma}{h(\bv)}
3363: -\frac{\bv^{2\gamma}\bv'}{h^2(\bv)}\right]\bv' - \frac{\bv^\gamma
3364: h'(\bv)}{h^2(\bv)}.\end{align*}
3365:
3366: Using \eqref{charell1} and \eqref{charI2}, we get
3367: $$\frac{\bV^{\gamma+1}}{h(\bV)} -1 = -\frac{(\gamma-1)\bv^{\gamma-1}\bv' + \bv
3368: h'(\bv)}{h(\bv)}$$ and thus
3369: \begin{align*} I= -\frac{(\gamma-1)\bv^{\gamma-1}\bv' + \bv
3370: h'(\bv)}{h(\bv)}+
3371: \left[\frac{\gamma\bv^{\gamma-1}}{h(\bv)}-2\frac{\bv^\gamma}{h(\bv)}
3372: -\frac{\bv^{2\gamma}\bv'}{h^2(\bv)}\right]\bv' - \frac{\bv^\gamma
3373: h'(\bv)}{h^2(\bv)}.\end{align*}
3374:
3375:
3376: Now since $h'(\bv) =
3377: -(\gamma+1)\bv^\gamma\bv'+(a+1)\gamma\bv^{\gamma-1}\bv'$, as
3378: $\bv_+\to 1$, $I\sim -\bv' \ge0$. Therefore, as $\bv_+$ is close to
3379: $1$, $G(0)\le \frac 14\bv'(0)|u(0)|^2\le0$. This, $g(\bv)\ge0$, and
3380: \eqref{outid2} give, as $\bv_+$ is close enough to $1$,
3381: \begin{align}\label{outid4}
3382: \R(\lambda)\ipco{ \left[ \frac{\bV^{\gamma+1}}{h(\bV)}|u|^2+|v|^2
3383: \right]} &+\ipco {\frac{\bV^\gamma}{h(\bV)}|u'|^2}\le 0,
3384: \end{align} which evidently gives stability as claimed.
3385: \end{proof}
3386:
3387:
3388: \section{Nonvanishing of $D_{\rm in}$: expansive inflow case}\label{nonvanish-expansive-inflow}
3389:
3390: For completeness, we recall the argument of \cite{MN.2}
3391: %CHANGED: wording-KZ
3392: %for the expansive inflow case.
3393: in the expansive inflow case.
3394: %ENDCHANGED
3395:
3396: {\bf Profile equation.} Note that, in the expansive inflow case, we
3397: assume $v_0 < v_+$. Therefore we can still follow the scaling
3398: \eqref{scaling} to get
3399: $$0<v_0< v_+ =1.$$
3400:
3401: Then the stationary boundary layer $(\bv,\bu)$ satisfies
3402: \eqref{stationarybl} with $v_0 <v_+=1$. Now by integrating
3403: \eqref{scalarode} from $x$ to $+\infty$ with noting that
3404: $\bv(+\infty)=1$ and $\bv'(+\infty)=0$, we get the profile equation
3405: $$\bv'= \bv(\bv-1 + a(\bv^{-\gamma} - 1)).$$
3406:
3407: Note that $\bv'>0$. We now follow
3408: the same method for compressive inflow case to get the following
3409: eigenvalue system
3410: \begin{subequations}\label{expinep}
3411: \begin{align}
3412: &\lambda v + v' - u' =0, \label{expinep:1}\\
3413: &\lambda u + u' - (f v)' = \left(\frac{
3414: u'}{\bV}\right)'.\label{expinep:2}
3415: \end{align}
3416: \end{subequations}with boundary
3417: conditions
3418: \begin{equation} u(0) = v(0) =0,
3419: \end{equation} where $f(\bv) = \frac{h(\bv)}{\bv^{\gamma+1}}$.
3420:
3421: \begin{proof}[Proof of Proposition \ref{expansive}] Multiply the equation \eqref{expinep:2} by $\bar u$
3422: and integrate along $x$. By integration by parts, we get $$\lambda
3423: \ipi{|u|^2} + \ipi{u'\bar u + fv\bar u' + \frac{|u'|^2}{\bv}} =0.$$
3424:
3425: Using \eqref{expinep:1} and taking the real part of the above yield
3426: \begin{align}\label{exeq}\R\lambda \ipi{|u|^2+f|v|^2} -\frac 12 \ipi{f'|v|^2} +
3427: \ipi{\frac{|u'|^2}{\bv}} =0.\end{align}
3428:
3429: Note that $$f' = \left(1+a+\frac{a(\gamma^2-1)}{\bv^\gamma}\right)
3430: \frac{-\bv'}{\bv^2} \le 0$$ which together with \eqref{exeq} gives
3431: $\R\lambda <0$, the proposition is proved. \end{proof}
3432:
3433: \def\cprime{$'$}
3434: \begin{thebibliography}{10}
3435:
3436: \bibitem{AGJ}
3437: J.~Alexander, R.~Gardner, and C.~Jones.
3438: \newblock A topological invariant arising in the stability analysis of
3439: travelling waves.
3440: \newblock {\em J. Reine Angew. Math.}, 410:167--212, 1990.
3441:
3442: \bibitem{Al}
3443: S.~Alinhac.
3444: \newblock Existence d'ondes de rar\'efaction pour des syst\`emes
3445: quasi-lin\'eaires hyperboliques multidimensionnels.
3446: \newblock {\em Comm. Partial Differential Equations}, 14(2):173--230, 1989.
3447:
3448: \bibitem{BHRZ}
3449: B.~Barker, J.~Humpherys, K.~Rudd, and K.~Zumbrun.
3450: \newblock Stability of viscous shocks in isentropic gas dynamics.
3451: \newblock Preprint, 2007.
3452:
3453: \bibitem{BDG}
3454: T.~J. Bridges, G.~Derks, and G.~Gottwald.
3455: \newblock Stability and instability of solitary waves of the fifth-order
3456: {K}d{V} equation: a numerical framework.
3457: \newblock {\em Phys. D}, 172(1-4):190--216, 2002.
3458:
3459: \bibitem{BrZ}
3460: L.~Q. Brin and K.~Zumbrun.
3461: \newblock Analytically varying eigenvectors and the stability of viscous shock
3462: waves.
3463: \newblock {\em Mat. Contemp.}, 22:19--32, 2002.
3464: \newblock Seventh Workshop on Partial Differential Equations, Part I (Rio de
3465: Janeiro, 2001).
3466:
3467: \bibitem{CJLW}
3468: N.~Costanzino, K.~J. Helge, G.~Lyng, and M.~Williams.
3469: \newblock Existence and stability of curved multidimensional detonation fronts.
3470: \newblock {\em Indiana Univ. Math. J.}, 56, 2007.
3471:
3472: \bibitem{FS}
3473: H.~Freist{\"u}hler and P.~Szmolyan.
3474: \newblock Spectral stability of small shock waves.
3475: \newblock {\em Arch. Ration. Mech. Anal.}, 164(4):287--309, 2002.
3476:
3477: \bibitem{GZ}
3478: R.~A. Gardner and K.~Zumbrun.
3479: \newblock The gap lemma and geometric criteria for instability of viscous shock
3480: profiles.
3481: \newblock {\em Comm. Pure Appl. Math.}, 51(7):797--855, 1998.
3482:
3483: \bibitem{GMWZ.6}
3484: C.~M. I.~O. Gu{\`e}s, G.~M{\'e}tivier, M.~Williams, and K.~Zumbrun.
3485: \newblock Viscous variable-coefficient systems.
3486: \newblock Preprint, 2006.
3487:
3488: \bibitem{GMWZ.5}
3489: C.~M. I.~O. Gu{\`e}s, G.~M{\'e}tivier, M.~Williams, and K.~Zumbrun.
3490: \newblock Multidimensional stability of small-amplitude noncharacteristic
3491: boundary layers.
3492: \newblock Preprint, 2007.
3493:
3494: \bibitem{GMWZ.3}
3495: O.~Gu{\`e}s, G.~M{\'e}tivier, M.~Williams, and K.~Zumbrun.
3496: \newblock Existence and stability of multidimensional shock fronts in the
3497: vanishing viscosity limit.
3498: \newblock {\em Arch. Ration. Mech. Anal.}, 175(2):151--244, 2005.
3499:
3500: \bibitem{HLZ}
3501: J.~Humpherys, O.~Lafitte, and K.~Zumbrun.
3502: \newblock Stability of isentropic viscous shock profiles in the high-mach
3503: number limit.
3504: \newblock Preprint, 2007.
3505:
3506: \bibitem{HSZ}
3507: J.~Humpherys, B.~Sandstede, and K.~Zumbrun.
3508: \newblock Efficient computation of analytic bases in {E}vans function analysis
3509: of large systems.
3510: \newblock {\em Numer. Math.}, 103(4):631--642, 2006.
3511:
3512: \bibitem{HuZ.3}
3513: J.~Humpherys and K.~Zumbrun.
3514: \newblock A fast algorithm for numerical stability analysis of detonation waves
3515: in {ZND}.
3516: \newblock In preparation, 2007.
3517:
3518: \bibitem{Kato}
3519: T.~Kato.
3520: \newblock {\em Perturbation theory for linear operators}.
3521: \newblock Classics in Mathematics. Springer-Verlag, Berlin, 1995.
3522: \newblock Reprint of the 1980 edition.
3523:
3524: \bibitem{KNZ}
3525: S.~Kawashima, S.~Nishibata, and P.~Zhu.
3526: \newblock Asymptotic stability of the stationary solution to the compressible
3527: navier-stokes equations in the half space.
3528: \newblock {\em Comm. Math. Phys.}, 240(3):483--500, 2003.
3529:
3530: \bibitem{MZ.3}
3531: C.~Mascia and K.~Zumbrun.
3532: \newblock Pointwise {G}reen function bounds for shock profiles of systems with
3533: real viscosity.
3534: \newblock {\em Arch. Ration. Mech. Anal.}, 169(3):177--263, 2003.
3535:
3536: \bibitem{MN}
3537: A.~Matsumura and K.~Nishihara.
3538: \newblock On the stability of travelling wave solutions of a one-dimensional
3539: model system for compressible viscous gas.
3540: \newblock {\em Japan J. Appl. Math.}, 2(1):17--25, 1985.
3541:
3542: \bibitem{MN.2}
3543: A.~Matsumura and K.~Nishihara.
3544: \newblock Large-time behaviors of solutions to an inflow problem in the half
3545: space for a one-dimensional system of compressible viscous gas.
3546: \newblock {\em Comm. Math. Phys.}, 222(3):449--474, 2001.
3547:
3548: \bibitem{MeZ}
3549: G.~M{\'e}tivier and K.~Zumbrun.
3550: \newblock Large viscous boundary layers for noncharacteristic nonlinear
3551: hyperbolic problems.
3552: \newblock {\em Mem. Amer. Math. Soc.}, 175(826):vi+107, 2005.
3553:
3554: \bibitem{PW}
3555: R.~L. Pego and M.~I. Weinstein.
3556: \newblock Eigenvalues, and instabilities of solitary waves.
3557: \newblock {\em Philos. Trans. Roy. Soc. London Ser. A}, 340(1656):47--94, 1992.
3558:
3559: \bibitem{PZ}
3560: R.~Plaza and K.~Zumbrun.
3561: \newblock An {E}vans function approach to spectral stability of small-amplitude
3562: shock profiles.
3563: \newblock {\em Discrete Contin. Dyn. Syst.}, 10(4):885--924, 2004.
3564: \newblock Preprint, 2002.
3565:
3566: \bibitem{R}
3567: F.~Rousset.
3568: \newblock Viscous approximation of strong shocks of systems of conservation
3569: laws.
3570: \newblock {\em SIAM J. Math. Anal.}, 35(2):492--519 (electronic), 2003.
3571:
3572: \bibitem{S}
3573: H.~Schlichting.
3574: \newblock {\em Boundary layer theory}.
3575: \newblock Translated by J. Kestin. 4th ed. McGraw-Hill Series in Mechanical
3576: Engineering. McGraw-Hill Book Co., Inc., New York, 1960.
3577:
3578: \bibitem{SZ}
3579: D.~Serre and K.~Zumbrun.
3580: \newblock Boundary layer stability in real vanishing viscosity limit.
3581: \newblock {\em Comm. Math. Phys.}, 221(2):267--292, 2001.
3582:
3583: \bibitem{YZ}
3584: S.~Yarahmadian and K.~Zumbrun.
3585: \newblock Pointwise green function bounds and long-time stability of strong
3586: noncharacteristic boundary layers.
3587: \newblock In preparation, 2007.
3588:
3589: \bibitem{Z.3}
3590: K.~Zumbrun.
3591: \newblock Stability of large-amplitude shock waves of compressible
3592: {N}avier-{S}tokes equations.
3593: \newblock In {\em Handbook of mathematical fluid dynamics. Vol. III}, pages
3594: 311--533. North-Holland, Amsterdam, 2004.
3595: \newblock With an appendix by Helge Kristian Jenssen and Gregory Lyng.
3596:
3597: \end{thebibliography}
3598:
3599:
3600: \end{document}
3601: