1: \documentclass[12pt]{article}
2: \usepackage{amsmath,amssymb,subfigure}
3: \usepackage[dvips]{epsfig}
4: \renewcommand\baselinestretch{2.0} %1.667%
5: %\topmargin=-.4688in
6: %\textheight=8.48in
7: %\textwidth=5.8in
8: %\oddsidemargin=.568in
9: \topmargin=-.5in
10: \usepackage{amsmath,amssymb,subfigure}
11: \usepackage[dvips]{epsfig}
12: \renewcommand\baselinestretch{1.5} %1.667%
13: %\topmargin=-.4688in
14: %\textheight=8.48in
15: %\textwidth=5.8in
16: %\oddsidemargin=.568in
17: \topmargin=-.5in
18: \textheight=8.5in
19: \textwidth=6in
20: \oddsidemargin=.5in
21: \makeatletter \@addtoreset{equation}{section} \makeatother
22: \renewcommand{\theequation}{\arabic{section}.\arabic{equation}}
23: \begin{document}
24: \pagestyle{empty}
25: \newtheorem{theo}{Theorem}[section]
26: \newtheorem{lem}[theo]{Lemma}
27: \newtheorem{cor}[theo]{Corollary}
28: \newtheorem{pro}[theo]{Proposition}
29: \newtheorem{prop}[theo]{Property}
30: \newcommand{\no}{\nonumber}
31: \newcommand{\p}{\partial}
32: \newcommand{\lb}{\lambda}
33: \newcommand{\vu}{\vec{u}}
34: \newcommand{\tvu}{\vec{\tilde{u}}}
35: \newcommand{\e}{\epsilon}
36: \newcommand{\ra}{\rightarrow}
37: \newcommand{\proof}{{\bf Proof}:\quad}
38:
39: \begin{center}
40: {\large\bf Large Time Behavior of the Zero Dispersion Limit of the Fifth Order KdV Equation}
41:
42: \vspace*{.3in}
43:
44: Virgil Pierce and Fei-Ran Tian
45:
46: \vspace{.1in}
47:
48: {\em Department of Mathematics, Ohio State University, Columbus, OH 43210 \\
49: vpierce@math.ohio-state.edu \ , ~ tian@math.ohio-state.edu}
50:
51:
52: \vspace*{.1in}
53:
54: \end{center}
55:
56: \vspace*{.5in}
57:
58: \noindent
59: {\bf Abstract:} \quad
60: We study the zero dispersion limit of the fifth order KdV equations when time is sufficiently large.
61: In general, the weak limit may be described by an arbitrary odd number of
62: hyperbolic equations. Unlike the KdV case, these are non-strictly hyperbolic equations.
63: However,
64: we show that the weak limit is governed by three hyperbolic equations
65: in a domain in the space-time for all times bigger than a large time.
66: Outside this domain, the weak limit satisfies a single hyperbolic equation.
67:
68: \thispagestyle{empty}
69:
70: \newpage
71:
72:
73: \setcounter{page}{1}
74: \pagestyle{plain}
75: %\setcounter{section}{0}
76: %\section{Introduction}
77: \refstepcounter{section}
78: \begin{center}
79: {\bf $\S$ 1 \quad Introduction}
80: \end{center}
81:
82: It is well known that the solution of the KdV equation
83: \begin{equation}
84: u_t + 6 u u_x + \epsilon^2 u_{xxx} = 0 \label{KdV}
85: \end{equation}
86: has a weak limit as $\epsilon \rightarrow 0$ while the initial values
87: \begin{equation}
88: u(x, 0; \epsilon) = u_0(x) \label{ini}
89: \end{equation}
90: are fixed \cite{lax, ven}.
91: This weak limit satisfies the Burgers equation
92: \begin{equation}
93: \label{Burgers}
94: u_t + (3 u^2)_x = 0
95: \end{equation}
96: until its solution develops shocks. Immediately after, the weak limit is governed
97: by the Whitham equations \cite{fla, lax, ven, whi}
98: \begin{equation}
99: \label{KdVW}
100: u_{it} + \lambda_i(u_1, u_2, u_3) u_{ix} = 0 \ , \quad i=1, 2, 3,
101: \end{equation}
102: where the $\lambda_i$'s are given by formulae (\ref{lambda}).
103: After the breaking of the solution of (\ref{KdVW}), the weak limit is described by systems of
104: at least five hyperbolic equations similar to (\ref{KdVW}).
105:
106: The KdV equation (\ref{KdV}) is just the first member of an infinite sequence
107: of equations, the second of which is the so-called fifth order KdV equation
108: \begin{equation}
109: \label{5KdV}
110: u_t + 30 u^2 u_x + 20 \epsilon^2 u_x u_{xx} + 10 \epsilon^2 u u_{xxx} + \epsilon^4 u_{xxxxx} = 0 \ .
111: \end{equation}
112: The solution of the fifth order KdV equation (\ref{5KdV}) also has a weak limit as $\epsilon \to 0$ \cite{lax2}.
113: As in the KdV case, this weak limit satisfies the Burgers type
114: equation
115: \begin{equation}
116: \label{5Burgers}
117: u_t + (10 u^3)_x = 0
118: \end{equation}
119: until the solution of (\ref{5Burgers}) forms a shock. Later, the limit is governed by equations similar
120: to (\ref{KdVW}), namely,
121: \begin{equation}
122: \label{5KdVW}
123: u_{it} + \mu_i(u_1, u_2, u_3) u_{ix} = 0 \ , \quad i=1, 2, 3,
124: \end{equation}
125: where $\mu_i$'s are given in (\ref{mu}). They will be also be called the Whitham equations \cite{pie}.
126:
127: In this paper, we are interested in the large time behavior of the weak limit of
128: the fifth order KdV equation (\ref{5KdV}).
129:
130: For simplicity,
131: we consider $u_0(x)$ of (\ref{ini}) which is
132: a decreasing function and is bounded at $x= \pm \infty$:
133: \begin{equation}
134: \label{cu}
135: \lim_{x \rightarrow - \infty }u_0(x) = 1 \ , ~~~~
136: \lim_{x \rightarrow + \infty }u_0(x) = 0 \ .
137: \end{equation}
138:
139: The large time behavior of the weak limit of the KdV equation (\ref{KdV})
140: has been studied \cite{Tian3}.
141: For generic initial data of (\ref{cu}), the weak limit is governed by
142: the Whitham equations (\ref{KdVW}) in a domain in the space-time when $t$ is sufficiently
143: large. Outside this domain, the weak limit satisfies the Burgers equation (\ref{Burgers}).
144: We note that the weak limit may be described by an arbitrary odd number of hyperbolic equations
145: in the intermedia times.
146:
147: The strong hyperbolicity of the Whitham equations (\ref{KdVW}) for the KdV
148: plays an important role in the paper of \cite{Tian3}.
149: It is well known that equations (\ref{KdVW}) are strictly hyperbolic:
150: $$ \lambda_i(u_1, u_2, u_3) \neq \lambda_j(u_1, u_2, u_3) \ , \quad i, j = 1, 2, 3; i \neq j,$$
151: and genuinely nonlinear:
152: $$ {\p \over \p u_i} \lambda_i(u_1, u_2, u_3) \neq 0 \ , \quad i = 1, 2, 3, $$
153: for $u_1 > u_2 > u_3$ \cite{lev}.
154:
155: However, in the case of fifth order KdV equation (\ref{5KdV}), the Whitham equations (\ref{5KdVW})
156: are neither strictly hyperbolic nor genuinely nonlinear \cite{pie}.
157:
158: In this paper, we use the method developed in \cite{gra1} and \cite{gra2} to study the weak limit of
159: the fifth order KdV (\ref{5KdV}) for sufficiently large time $t$. Our approach does not require
160: the Whitham equations to be strongly hyperbolic. Our result is quite similar to the
161: KdV result. Namely, For generic initial data of (\ref{cu}), the weak limit is governed by
162: the Whitham equations (\ref{5KdVW}) in a domain $x^-(t) < x < x^+(t)$ in the space time
163: when $t$ is sufficiently
164: large (see Figure 1.). Outside this domain, the weak limit satisfies the Burgers type
165: equation (\ref{5Burgers}).
166:
167: We also find some difference between the KdV case and the fifth order KdV case. In the latter case,
168: the trailing edge is ``lifted'', i.e.,
169: $$u_1 \approx 1 \ , \quad u_2 = u_3 \approx {1 \over 4} \quad \mbox{for large $t>0$}$$
170: at the trailing edge $x=x^-(t)$ (see Figure 1.). In the KdV case, instead we have
171: $$u_1 \approx 1 \ , \quad u_2 = u_3 \approx 0 \quad \mbox{for large $t>0$}$$
172: at the trailing edge \cite{gur, lax}.
173:
174: This phenomenon of ``lifted'' trailing edge has also been observed in the weak limit of
175: the fifth order KdV equation
176: when the initial values $u_0(x)$ of (\ref{ini}) are given by a step-like function
177: $$u_0(x) = \left\{ \begin{matrix} 1 & x < 0 \\
178: 0 & x > 0 \end{matrix} \right. \ . $$
179: In this case, we have
180: $$u_1 = 1 \ , \quad u_2 = u_3 = {1 \over 4}$$
181: at the trailing edge $x=-15t$ \cite{pie}.
182:
183: The organization of this paper is as follows. In Section 2, we will summarize the method
184: of \cite{gra1}. In Section 3, we will describe our main theorem. In Section 4, we
185: will study the trailing and leading edges, which separate the Burgers type solutions and the
186: Whitham solutions,
187: when $t$ is large. In Section 5, we will study the
188: Whitham solutions that live between the trailing and leading edges.
189:
190:
191: \refstepcounter{section}
192: \begin{center}
193: {\bf $\S$ 2 \quad A Minimization Problem}
194: \end{center}
195:
196: As in the KdV case, the weak limit of the fifth order KdV equation (\ref{5KdV})
197: is determined by
198: a minimization problem with constraints \cite{lax2}.
199:
200: The minimization problem is
201: \begin{equation}
202: \label{mini}
203: \underset{\{\psi \geq 0, \ \psi \in L^1 \}}
204: {\rm
205: Minimize}
206: \{ - \frac{1}{2 \pi} \int_0^1 \int_0^1 \log \Big|\frac{\eta - \mu}
207: {\eta + \mu
208: }\Big|
209: \psi(\eta) \psi(\mu) d \eta d \mu + \int_0^1 a(\eta, x, t)
210: \psi(\eta) d \eta \} \ .
211: \end{equation}
212: Function $a(\eta, x, t)$ is given by
213: \begin{eqnarray}
214: a(\eta, x, t) &=& \eta x - 16 \eta^5 t - \theta(\eta) \ , \label{a} \\
215: \theta(\eta) &=& \eta f(\eta^2) + \int_{f(\eta^2)}^{+ \infty}
216: [\eta - \sqrt{\eta^2 - u_0(x)}] d x \ , \no
217: \end{eqnarray}
218: where $f(u)$ is the inverse function of the initial data $u_0(x)$.
219: We note that in the KdV case, we have $a(\eta, x, t) = \eta x - 4 \eta^3 t - \theta(\eta)$
220: instead of (\ref{a}) \cite{gra1, lax, ven}.
221:
222: We now give a brief summary of our approach to the minimization problem (\ref{mini}).
223:
224: We first follow Lax $\&$ Levermore \cite{lax} and Venakides \cite{ven, ven2} to make the ansatz
225: that the support of $\psi$ consists
226: of a finite union of disjoint intervals
227: \begin{equation}
228: \label{support}
229: S = [0, \sqrt{u_{2g+1}}] \cup [\sqrt{u_{2g}}, \sqrt{u_{2g-1}}] \cup \cdots
230: \cup [\sqrt{u_2}, \sqrt{u_1}] \ ,
231: \end{equation}
232: where $0 < u_{2g+1} < \cdots < u_2 < u_1 < 1$. Hence, $g$ is the number of gaps in the support
233: $S$.
234:
235: We then introduce
236: \begin{equation}
237: \label{P}
238: P(\xi,\vec{u}) = 2 R^2(\xi, \vu) \Phi_g(\xi,\vec{u}) + Q(\xi,\vec{u}) \ ,
239: \end{equation}
240: where $\vu$ denotes $(u_1, u_2, \cdots, u_{2g+1})$ and
241: $R(\xi, \vec{u}) = \sqrt{(\xi - u_1)\cdots (\xi - u_{2g+1})}$ with the sign
242: given by $\sqrt{1} = 1$ \cite{gra1}.
243:
244: The function $\Phi_g(\xi,\vec{u})$
245: is the unique solution of the boundary value problem for
246: the Euler-Poisson-Darboux equations
247: \begin{eqnarray}
248: 2(u_i - u_j) {\p^2 \Phi_g \over \p u_i \p u_j} &=& {\p \Phi_g \over \p u_i} -
249: {\p \Phi_g \over \p u_j} \ , \quad i,j = 1, 2, \cdots, 2g+1, \label{ph1} \\
250: 2(\xi - u_i) {\p^2 \Phi_g \over \p \xi \p u_i} &=& {\p \Phi_g \over \p \xi} -
251: 2{\p \Phi_g \over \p u_i} \ , \quad i = 1, 2, \cdots, 2g+1, \label{ph2} \\
252: \Phi_g(u, u, \cdots, u) &=& {2^g \over (2g+1)!!} {d^{g+1} \over d u^{g+1}}
253: [ x - 30 t u^2 - f(u)]
254: \ . \label{ph3}
255: \end{eqnarray}
256:
257: The function $Q(\xi,\vec{u})$ is a polynomial of degree $2g$ in $\xi$:
258:
259: \begin{enumerate}
260:
261: \item for $g=0$,
262: $Q(\xi, u_1) = x - 30 u_1^2 t - f(u_1)$;
263:
264: \item for $g > 0$,
265: \begin{eqnarray*}
266: Q(\xi, \vec{u}) &=& 2 \sum_{i = 1}^{2g+1}
267: [\prod_{l=1, l \neq i}^{2g+1} (\xi-u_l)]
268: {\p q_{g,g}(\vec{u}) \over \p u_i} \no \\
269: & & + \sum_{k=1}^{g} [(2k-1) \sum_{l=0}^{g-k} \Gamma_l
270: (\vec{u})q_{g,k+l}(\vec{u})]
271: P_{g, k-1}(\xi, \vec{u}) \ .
272: \end{eqnarray*}
273: \end{enumerate}
274: Here $\Gamma_{l}(\vec{u})$ are the coefficients of the
275: expansion at $\xi = \infty$
276: \begin{displaymath}
277: R(\xi, \vu) = \xi^{g+{1 \over 2}} [\Gamma_{0}(\vu) +
278: {\Gamma_{1}(\vu) \over
279: \xi} + {\Gamma_{2}(\vu) \over \xi^2} + \cdots ]
280: \end{displaymath}
281: and $q_{g,k}(\vec{u})$ is the solution of the boundary value problem for
282: another version of the Euler-Poisson-Darboux equations
283: \begin{eqnarray}
284: 2(u_i - u_j) {\p^2 q_{g,k} \over \p u_i \p u_j} = {\p q_{g,k} \over \p u_i} -
285: {\p q_{g,k} \over \p u_j} \ , \quad \mbox{$i, j= 1, \cdots, 2g+1$} \ , \label{q1} \\
286: q_{g,k}(u, \cdots, u) = \frac{ 2^{g-1}}{(2g-1)!!}
287: u^{-k + \frac{1}
288: {2}} \frac{d^{g-k} }{d u^{g-k}} \{ u^{g - \frac{1}{2}} [ \frac{
289: d^{k-1} }{d u^{k-1}} \left( x - 30 t u^2 - f(u) \right) ] \}. \label{q2}
290: \end{eqnarray}
291: The polynomial
292: \begin{equation}
293: \label{pgn}
294: P_{g,n}(\xi,\vec{u}) = \xi^{g+n} + a_{g,1} \xi^{g+n-1} + \cdots + a_{g, g+n}
295: \end{equation}
296: is defined such that
297: \begin{equation}
298: \label{Pgn}
299: \frac{P_{g,n}(\xi,\vec{u})}{R(\xi, \vu)} =
300: \xi^{n-\frac{1}{2}} +
301: O(\xi^{-\frac{3}{2}}) ~~~~~~~
302: \mbox{for large $|\xi|$} \ ,
303: \end{equation}
304: and
305: \begin{equation}
306: \label{lc}
307: \int_{u_{2k+1}}^{u_{2k}} \frac{P_{g,n}(\xi,\vec{u})}{R(\xi, \vu)} d \xi
308: = 0 \ , ~~~~~~~~
309: k = 1, 2, \cdots, g \ ,
310: \end{equation}
311: where $\vu=(u_1, u_2, \cdots, u_{2g+1})$.
312:
313: Solutions of the boundary value problem for the Euler-Poisson-Darboux equations
314: can be constructed using those of the following simpler problem as building
315: blocks \cite{Tian1}
316: \begin{eqnarray}
317: \label{EPD}
318: 2(x_{1} - x_{2})\frac{\partial^{2} v}{\partial x_{1} \partial
319: x_{2}} & = & \frac{\partial v}{\partial x_{1}} - \rho \frac{\partial v}
320: {\partial x_{2}} \ , ~~~~~ \rho > 0 ~ is ~ a ~ constant \ , \\
321: v(x_1, x_1) & = & g(x_1) \ . \label{EPDd}
322: \end{eqnarray}
323: The boundary value problem (\ref{EPD}-\ref{EPDd}) has one and
324: only one solution for smooth boundary
325: data $g(x)$ \cite{Tian1}. A simple calculation shows that the solution is
326: given by the formula
327: \begin{displaymath}
328: v(x_1, x_2) = C \int_{-1}^{1} \frac{g(\frac{1+\mu}{2}x_1 + \frac{1-\mu}{2}x_2)}{
329: \sqrt{1 - \mu^{2}}} (1 + \mu)^{\frac{\rho - 1}{2}} d \mu \ ,
330: \end{displaymath}
331: where
332: \begin{displaymath}
333: C = \frac{1}{\int_{-1}^{1} \frac{(1+\mu)^{\frac{\rho - 1}{2}}}{\sqrt{1-\mu^{2}}}
334: d \mu} \ .
335: \end{displaymath}
336: A change of integration variable gives another
337: expression for the solution
338: \begin{equation}
339: \label{EPQsS}
340: v(x_1,x_2) = C \left[{2 \over x_1-x_2}\right]^{\rho - 1 \over 2} \int_{x_2}
341: ^{x_1} g(x) {(x - x_2)^{\rho - 2 \over 2} \over (x_1 - x)^{1 \over 2}} d x \ ,
342: \end{equation}
343: where the square root is set to be positive for $x$ between $x_1$ and $x_2$.
344: In particular, when $g=0$, the solution of the boundary value problem
345: (\ref{ph1}-\ref{ph3}) is
346: \begin{equation}
347: \label{phi0}
348: \Phi_0(\xi, u_1) = {1 \over 2 \sqrt{2}} \int_{-1}^{1} {H'({1 + \mu \over 2}
349: \xi + {1- \mu \over 2} u_1) \over \sqrt{1 - \mu} } d \mu \ ,
350: \end{equation}
351: where $H(u) = x - 30 t u^2 - f(u)$.
352: More generally, the solution of equations (\ref{ph1}-\ref{ph3}) can be solved explicitly
353: \begin{eqnarray}
354: \lefteqn{\Phi_g(\xi, \vec{u}) = } \no \\
355: &M& \int_{-1}^{1} \cdots \int_{-1}^{1}
356: { H^{(g+1)}({1 + \mu_{2g+1} \over 2} (\cdots ({1 + \mu_{2} \over 2}
357: ({1 + \mu_{1} \over 2} \xi + {1 - \mu_{1} \over 2}u_1) + \cdots ) +
358: {1 - \mu_{2g+1} \over 2} u_{2g+1} ) \over \sqrt{(1-\mu_1)(1-\mu_2) \cdots
359: (1 - \mu_{2g+1})}} \no \\
360: & & \times (1+\mu_2)^{1 \over 2} (1+\mu_3)^{2 \over 2} \cdots
361: (1+\mu_{2g+1})^{2g \over 2} d \mu_1 d \mu_2 \cdots d \mu_{2g+1} \ , \label{phi}
362: \end{eqnarray}
363: where the constant $M$ is chosen so that
364: the boundary condition (\ref{ph3}) is
365: satisfied \cite{gra1}.
366:
367: The solution $q_{g,k}(\vec{u})$ of another
368: boundary value problem
369: (\ref{q1}), (\ref{q2}) is also given by a multiple integral formula similar to (\ref{phi}).
370:
371: We now list some identities of $\Phi_g$ and $q_{g,k}$. They will be useful
372: in the subsequent calculations.
373:
374: \begin{lem} \cite{gra1}
375: \label{gidentity}
376: \begin{eqnarray}
377: 2 (\xi - u_i) {\p \over \p u_i} \Phi_g(\xi, \vec{u})
378: &=& \Phi_g(\xi, \vec{u}) - \Phi_g(u_i, \vec{u}) \ , \label{Phiu} \\
379: \Phi_g(\xi, \vu)|_{u_l=u_{l+1} = u^*} &=&
380: \frac{\Phi_{g-1}(\xi,\tvu)-\Phi_{g-1}(u^*,\tvu)}{\xi-u^*} \ , \quad
381: \label{lim1} \\
382: \Phi_g(u^*,\vec{u})|_{u_l=u_{l+1} = u^*} &=& {\p \Phi_{g-1}(\xi, \tvu)
383: \over \p \xi}|_{\xi
384: = u^*} \ , \label{lim2} \\
385: \Phi_{g-1}(\xi,\tvu) &=& \left(\sum_{i=1}^{2g+1}
386: {\p \over \p u_i}q_{g,g}(\vu)\right)
387: |_{u_{l}=u_{l+1}=\xi} \ , \label{lim3}
388: \end{eqnarray}
389: where $\vu=(u_1,\dots, ,u_{2g+1})$,
390: $\tvu=(u_1,\dots, u_{l-1},u_{l+2},\dots,u_{2g+1})$ and $1\leq l\leq 2g$.
391: \end{lem}
392:
393: Our method of solution to the minimization problem is summarized in the
394: following
395: theorem.
396: \begin{theo} \cite{gra1}
397: \label{main}
398: If $x$, $t$ and $\vu$ are connected by the equations
399: \begin{equation}
400: \label{P2}
401: P(u_i, \vu) = 0 ~~~~~~ \mbox{for $i=1, 2, \cdots, 2g+1$},
402: \end{equation}
403: where $P$ is defined by (\ref{P}), and if the inequalities for
404: $k=0, 1, \cdots, g$, $u_0 = 1$ and $u_{2g+2} = 0$
405: \begin{eqnarray}
406: Re\{\sqrt{-1}R(\xi,\vu)\}\Phi_g(\xi,
407: \vu) & < & 0 \quad
408: \mbox{for $u_{2k+2} < \xi < u_{2k+1}$} \ , \label{constraint} \\
409: \int_{u_{2k+1}}^{\xi} R(\mu, \vu) \Phi_g(\mu,\vu)
410: d \mu & > & 0 \quad \mbox{for $u_{2k+1}<\xi<u_{2k}$} \ ,
411: \label{constraint2}
412: \end{eqnarray}
413: are satisfied, the function
414: \begin{equation}
415: \label{psi}
416: \psi(\eta) = - 2 \eta Re\{\sqrt{-1}R(\eta^2,\vu)\}\Phi_g(\eta^2,\vu)
417: \end{equation}
418: is the minimizer of (\ref{mini}). Its support is given by
419: (\ref{support}).
420: \end{theo}
421:
422: We now analyze equations (\ref{P2}). The boundary conditions
423: (\ref{ph3}) and (\ref{q2}) are linear in $x$ and $t$, as is the function
424: $P(\xi, \vec{u})$. We then use formulae (\ref{P}-\ref{q2}) to write
425: \begin{equation}
426: \label{1}
427: P(\xi, \vec{u}) = x P_{g,0}(\xi, \vec{u}) - 80 t P_{g,2}(\xi, \vec{u}) - P_f(\xi, \vec{u}) \ ,
428: \end{equation}
429: where $P_{g,0}$ and $P_{g,2}$ are defined by (\ref{pgn}-\ref{lc}), and
430: $P_f$ is also determined by equations (\ref{P}-\ref{q2}) with boundary data (\ref{ph3})
431: and (\ref{q2}) depending only on $f$ and its derivatives.
432:
433: Equations (\ref{P2}) can then be written as
434: \begin{equation}
435: \label{P3}
436: x = \lambda_{g, i}(\vec{u})t + w_{g,i}(\vec{u}) \ , \quad i = 1, 2, \cdots , 2g+1 \ ,
437: \end{equation}
438: where $$ \lambda_{g, i} = 80 {P_{g,2}(u_i, \vec{u}) \over P_{g,0}(u_i, \vec{u})} \ ,
439: \quad w_{g,i}(\vec{u}) = {P_f(u_i, \vec{u}) \over P_{g,0}(u_i, \vec{u})} \ .$$
440:
441: It is well known that the solutions $u_1$, $u_2$, $\cdots$, $u_{2g+1}$ of (\ref{P3})
442: or equivalently (\ref{P2}), as functions of $x$ and $t$, satisfy an odd number of
443: hyperbolic partial differential
444: equations \cite{dub, gra1, kri, Tian2, tsa}
445: \begin{equation}
446: \label{P4}
447: u_{it} + \lambda_{g, i}(u_1, u_2, \cdots, u_{2g+1}) u_{ix} = 0 \ , \quad i = 1,2, \cdots, 2g+1 \ .
448: \end{equation}
449: In other words, the square roots of the end points of the support (\ref{support}) of the minimizer of
450: the variational problem (\ref{mini}) satisfy the PDE (\ref{P4}).
451:
452: The coefficients $\lambda_{g,i}$ of (\ref{P4}) involve complete hyperelliptic integrals
453: of genus $g$ because $P_{g,n}$'s do so in view of (\ref{pgn}-\ref{lc}). Equations (\ref{P4}) are then
454: called the $g$-phase Whitham equations and its solution a $g$-phase solution.
455:
456: In particular, when $g=0$, we have $\lambda_{0, 1}(u_1) = 30 u_1^2$ and equation (\ref{P4}) becomes
457: the Burgers type equation (\ref{5Burgers}).
458:
459: When $g=1$, equations (\ref{P4}) turn out to be the Whitham equations (\ref{5KdVW}), where
460: \begin{equation}
461: \label{mu}
462: \mu_i(u_1, u_2, u_3) = \lambda_{1, i}(u_1,u_2,u_3) = 80 {P_{1,2}(u_i, u_1, u_2, u_3)
463: \over P_{1,0}(u_i, u_1, u_2, u_3)} \ , \quad i=1,2,3 \ .
464: \end{equation}
465:
466: We note that, for the KdV equation, $\lambda_i$'s of (\ref{KdVW}) are given by
467: \begin{equation}
468: \label{lambda}
469: \lambda_i(u_1, u_2, u_3) = 12 {P_{1,1}(u_i, u_1, u_2, u_3) \over P_{1,0}(u_i, u_1, u_2, u_3)} \ ,
470: \quad i = 1, 2, 3 \ .
471: \end{equation}
472:
473: To solve the system of algebraic equations (\ref{P2}), we rely on the
474: Implicit Function Theorem. Although the system is complicated, its Jacobian
475: matrix is diagonal and its determinant is easily calculated.
476: \begin{theo} \cite{gra1}
477: \label{jaco}
478: On the solution $\vec{u}$ of (\ref{P2}),
479: \begin{enumerate}
480: \item the Jacobian matrix of the hodograph transform (\ref{P2}) is diagonal, i.e.,
481: $${\p P(u_i, \vec{u}) \over \p u_j} = 0 \ , \quad i, j = 1, 2, \cdots, 2g+1 \ ; i \neq j \ ,$$
482:
483: \item the Jacobian is
484: \begin{displaymath}
485: det \left({\p P(u_i, \vec{u}) \over \p u_j} \right)
486: = (-1)^g \prod_{i > j} (u_i - u_j)^2 \prod_{i=1}^{2g +1}
487: \Phi_g(u_i, \vu) \ .
488: \end{displaymath}
489: \end{enumerate}
490: \end{theo}
491:
492: Violation of inequalities (\ref{constraint}-\ref{constraint2}) gives rise to
493: phase changes. Indeed,
494: when (\ref{constraint}) first fails, function $\Phi_g(\xi,\vu)$
495: must have a double $\xi$-zero for $\sqrt{\xi}$ in the interior of $S$ of (\ref{support}),
496: say, the zero $u^*$ where $\sqrt{u^*} \in S$.
497: The resulting equations are
498: \begin{eqnarray}
499: \Phi_g(u^*, \vu)
500: &=& 0 \ , \label{teg1} \\
501: {\p \over \p \xi} \Phi_g(\xi, \vu)|_{
502: \xi = u^*} &=& 0 \ , \label{teg2} \\
503: P(u_i, \vu) &=& 0 \ , \quad i = 1, 2, \cdots, 2g+1 \ .
504: \label{teg3}
505: \end{eqnarray}
506: They govern the
507: so-called trailing edge separating
508: $g$ and $g+1$ phases.
509: Similarly, when (\ref{constraint2}) is first violated, the
510: indefinite integral must also
511: have a double $\xi$-zero for $\sqrt{\xi}$ in the complement of $S$,
512: say, the zero $u^* \in (u_{2k+1}, u_{2k})$.
513: The resulting equations are
514: \begin{eqnarray}
515: \Phi_g(u^*, \tvu)
516: &=& 0 \ , \label{leg1} \\
517: \int_{u_{2k+1}}^{u^*} \Phi_g(\xi, \vu)
518: R(\xi, \vu) d \xi &=& 0 \ , \label{leg2} \\
519: P(u_i, \vu) &=& 0 \ , \quad i = 1, 2, \cdots, 2g+1 \ ,
520: \label{leg3}
521: \end{eqnarray}
522: which govern
523: the so-called leading edge separating $g$ and $g+1$ phases.
524:
525: We close this section with a remark on the other higher order KdV equations.
526: All the zero dispersion limits of the equations in the KdV hierarchy are determined by
527: the minimization problem (\ref{mini}) with $a(\eta, x, t)$ of (\ref{a}) replaced by \cite{lax2}
528: \begin{equation}
529: \label{am}
530: a(\eta, x, t) = \eta x - 4^m \eta^{2m+1} t - \theta(\eta)
531: \end{equation}
532: for $m=1 \ , 2 \ , \cdots \ .$ The function $x-30tu^2 - f(u)$ in the
533: boundary conditions (\ref{ph3}) and (\ref{q2})
534: must then be replaced by $x - C_m t u^m - f(u)$, where \cite{gra1}
535: $$C_m = {(2m+1)2^{2m+1} \over \pi} \int_0^1 {t^{2m} \over \sqrt{1 - t^2}} d t \ .$$
536: In particular, $m=1$ corresponds to the KdV case and $m=2$ to the fifth order KdV case.
537: All the analysis of Section 3 can be carried over from the fifth order KdV ($m=2$) case to all
538: the other equations ($m \neq 2$) in the KdV hierarchy.
539:
540: In the subsequent sections, we will only study the fifth order KdV ($m=2$) case. The KdV
541: ($m=1$) case has been well understood \cite{Tian3}. We will indicate why the
542: $m > 2$ case is technically more difficult than the $m=2$ case when the opportunity presents itself.
543:
544:
545:
546: \bigskip
547:
548: \refstepcounter{section}
549: \begin{center}
550: {\bf $\S$ 3 \quad Main Theorem}
551: \end{center}
552:
553: We first make assumption on the initial data $u_0(x)$ or its inverse
554: $f(u)$. Since $u_0(x)$ has the limits (\ref{cu}) as $x$ goes to the infinity,
555: its inverse $f(u)$ behaves as
556: \begin{equation}
557: \label{cf1}
558: \lim_{u \rightarrow 0} f(u) = + \infty \ , ~~~~~~~~~~~~~~~~~
559: \lim_{u \rightarrow 1} f(u) = - \infty \ .
560: \end{equation}
561: We further assume that
562: \begin{equation}
563: \label{cf2}
564: f'''(u) < 0 ~~~~~~~~ \mbox{in the neighborhood of $u=0$ and $u=1$} \ .
565: \end{equation}
566: Conditions (\ref{cf1}-\ref{cf2}) immediately imply \cite{Tian3}
567: \begin{equation}
568: \label{cf3}
569: \lim_{u \rightarrow 0} f'(u) = \lim_{u \rightarrow 1} f'(u) = - \infty \ , ~~
570: \lim_{u \rightarrow 0} f''(u) = + \infty \ , ~
571: \lim_{u \rightarrow 1} f''(u) = - \infty \ .
572: \end{equation}
573:
574: \begin{theo} (see Figure 1.)
575: \label{Main}
576: Under the conditions (\ref{cf1}) and (\ref{cf2}), there exists a $T>0$ so that when
577: $t > T$, the weak
578: limit of the fifth order KdV equation is governed by the Whitham equations
579: (\ref{5KdVW}) for $x^-(t) < x < x^+(t)$, where $x^-(t)$ and $x^+(t)$ are some functions of $t$.
580: The weak limit satisfies the Burgers
581: type equation (\ref{5Burgers}) for $x < x^-(t)$ and $x > x^+(t)$.
582: \end{theo}
583:
584: This theorem is a consequence of Theorems \ref{Burgers2} and \ref{Whitham}, which will be
585: proved in the subsequent sections.
586:
587: \begin{figure}[h] \label{fig1}
588: \begin{center}
589: \resizebox{15cm}{5cm}{\includegraphics{gen4.eps}}
590: \caption{Large time behavior of the weak limit of the fifth order KdV
591: for generic initial data.
592: The limit is governed by
593: the Whitham equations (\ref{5KdVW}) for $x^-(t) < x < x^+(t)$ and by
594: the Burgers type equation (\ref{5Burgers}) otherwise. The trailing edge $x=x^-(t)$ is
595: characterized by $u_1 \approx 1 \ , u_2=u_3 \approx 1/4$.}
596: \end{center}\end{figure}
597:
598: \bigskip
599:
600: \refstepcounter{section}
601: \begin{center}
602: {\bf $\S$ 4 \quad Trailing and Leading Edges}
603: \end{center}
604:
605: In this section, we will study the trailing and leading edges of a single phase
606: solution when $t$ is sufficiently large.
607:
608: \medskip
609:
610: \noindent
611: {\bf $\S$ 4.1 \quad The trailing edge}
612:
613: We first analyze the tail of a single phase solution at which $u_2=u_3$ (see Figure 1.).
614: It is a phase transition boundary between zero and single phases; it corresponds to
615: $g=0$ in equations (\ref{teg1}-\ref{teg3}):
616: \begin{eqnarray}
617: W_1(u_1, u_3, t) := \Phi_0(u_3, u_1)
618: &=& 0 \ , \label{teg10} \\
619: W_2(u_1, u_3, t) := {\p \over \p \xi} \Phi_0(\xi, u_1)|_{
620: \xi = u_3} &=& 0 \ , \label{teg20} \\
621: x - 30 t u_1^2 - f(u_1) &=& 0 \ . \label{teg30}
622: \end{eqnarray}
623:
624: We split $\Phi_0(\xi, u_1)$ into
625: \begin{equation}
626: \label{split}
627: \Phi_0(\xi, u_1) = - 30 t U_0(\xi, u_1) - F_0(\xi, u_1) \ ,
628: \end{equation}
629: where $U_0$ and $F_0$ are the solutions of the Euler-Poisson-Darboux equations
630: (\ref{ph1}-\ref{ph3}) with boundary data $2 u$ and $f'(u)$, respectively.
631: In view of formula (\ref{phi0}), we have
632: \begin{eqnarray}
633: F_0(\xi, u_1) &=& {1 \over 2 \sqrt{2}} \int_{-1}^{1} {f'({1 + \mu \over 2}
634: \xi + {1- \mu \over 2} u_1) \over \sqrt{1 - \mu} } d \mu \ , \label{fF} \\
635: U_0(\xi, u_1) &=& {1 \over 2 \sqrt{2}} \int_{-1}^{1} { 2({1 + \mu \over 2}
636: \xi + {1- \mu \over 2} u_1) \over \sqrt{1 - \mu} } d \mu =
637: {4 \over 3} \xi + {2 \over 3} u_1 \ . \label{fU}
638: \end{eqnarray}
639:
640: Equations (\ref{teg10}) and (\ref{teg20}) are then equivalent to
641: \begin{eqnarray}
642: {F_0 \over U_0} + 30 t &=& 0
643: \ , \label{teg10'} \\
644: {\p \over \p \xi} [{F_0 \over U_0} + 30 t]_{\xi = u_3}
645: &=& 0 \ . \label{teg20'}
646: \end{eqnarray}
647:
648: Equation (\ref{teg20'}) suggests that the point $\xi = u_3$ is a critical
649: point of the function $[F_0/U_0 + 30 t]$.
650: We therefore consider an auxiliary minimum problem
651: \begin{equation}
652: \label{min}
653: Min_{0 < \xi < 1} \left[ - {F_0(\xi, u_1) \over U_0(\xi, u_1)} \right]
654: \end{equation}
655: for each $u_1$ close to $1$.
656:
657: To study the minimization problem (\ref{min}), we need the following lemma.
658: \begin{lem}
659: \label{trailing}
660:
661: \begin{enumerate}
662:
663: \item
664: \begin{equation}
665: \label{F}
666: \lim_{\xi \rightarrow 0} F_0(\xi, u_1) = \lim_{\xi \rightarrow 1} F_0(\xi, u_1)
667: = - \infty
668: \end{equation}
669: for each $u_1 \in (0, 1)$.
670:
671: \item There exists a $\delta > 0$ such that
672: \begin{equation}
673: \label{F_0''}
674: {\p^2 \over \p \xi^2} F_0(\xi, u_1) < 0 ~~~~~ \mbox{for $0 < \xi < 1$ and
675: $1 - \delta < u_1 < 1$} \ .
676: \end{equation}
677:
678: \item
679: \begin{equation}
680: \label{F'}
681: \lim_{u_1 \rightarrow 1} { {\p F_0(\xi, u_1) \over \p \xi} \over F_0(\xi, u_1) } = {1 \over 1 - \xi}
682: \end{equation}
683: uniformly for $\xi$ on every compact subset of the interval $(0,1)$.
684:
685: \end{enumerate}
686: \end{lem}
687: Proof: To prove the first limit of (\ref{F}), we use (\ref{fF}) to
688: rewrite $F_0(\xi, u_1)$ as
689: $$F_0(\xi, u_1) = {1 \over 2 \sqrt{\xi - u_1}} \int_{u_1}^{\xi} {f'(u)
690: \over \sqrt{\xi - u} } d u ~~~~~ \mbox{when $\xi > u_1$} \ . $$
691: Since $f'(u) < 0$, we estimate the integral when $\xi$ is close to $1$:
692: \begin{equation*}
693: F_0(\xi, u_1) \leq {1 \over 2(\xi - u_1)} \int_{u_1}^{\xi} f'(u) d u
694: = {1 \over 2(\xi - u_1) } [ f(\xi) - f(u_1) ] \ .
695: \end{equation*}
696: This together with (\ref{cf1}) proves the second limit of (\ref{F}).
697: The first limit can be shown in the same way.
698:
699: To prove inequality (\ref{F_0''}), in view of condition (\ref{cf2}) we
700: assume $f'''(u) < 0$ for $u$ outside the interval $(\delta_1, 1 - \delta_1)$
701: where $\delta_1$ is a small positive number.
702: Since $f'''(u) < 0$ for $u > 1 - \delta_1$, it follows from formula
703: (\ref{fF}) that it suffices to prove (\ref{F_0''}) for $\xi \leq 1 - \delta_1$.
704:
705: We use formula (\ref{fF}) to write for $u_1 > 1 - {\delta_1 \over 2} \geq \xi$,
706: \begin{eqnarray}
707: {\p^2 \over \p \xi^2} F_0(\xi, u_1) &=& {1 \over 2 (u_1 - \xi)^{5 \over 2}}
708: \int_{\xi}^{u_1} f'''(u) {(u_1 - u)^2 \over \sqrt{u - \xi} } d u \no \\
709: &=& {1 \over 2 (u_1 - \xi)^{5 \over 2}} [ \int_{1 - {\delta_1 \over 2}}^{u_1}
710: + \int_{\xi}^{1 - {\delta_1 \over 2}} ] \ . \label{F0''}
711: \end{eqnarray}
712: Since $f'''(u) < 0$ for $u < \delta_1/2$,
713: the second integral in the parenthesis of
714: (\ref{F0''}) is smaller than
715: $$\int_{\delta_1}^{1 - {\delta_1 \over 2}} {|f'''(u)| \over \sqrt{u - \xi}}
716: d u \ ,$$
717: which, since $f'''(u)$ is bounded on the closed interval $[\delta_1, 1 - \delta_1/2]$,
718: is uniformly bounded for $\xi \leq 1 - \delta_1/2$.
719:
720: Since $f'''(u) < 0$ for $u > 1 - \delta_1$, the first integral of (\ref{F0''})
721: is less than
722: \begin{eqnarray*}
723: \lefteqn{\int_{1 - {\delta_1 \over 2}}^{u_1}
724: f'''(u) (u_1 - u)^2 d u} \\
725: &=& 2 f(u_1) -
726: f''(1 - {\delta_1 \over 2}) (u_1 - 1 + {\delta_1 \over 2})^2 -
727: 2 f'(1 - {\delta_1 \over 2}) (u_1 - 1 + {\delta_1 \over 2}) -
728: 2f(1 - {\delta_1 \over 2}) \ ,
729: \end{eqnarray*}
730: which, in view of (\ref{cf1}), goes to $- \infty$ as $u_1 \rightarrow
731: 1$. This proves inequality (\ref{F_0''}).
732:
733: To prove (\ref{F'}), we
734: use formula (\ref{fF}) to write for $u_1 > \xi$,
735: \begin{eqnarray}
736: F_0(\xi, u_1) &=& {1 \over 2 \sqrt{u_1 - \xi}} \int_{\xi}^{u_1} {f'(u)
737: \over \sqrt{u - \xi } } d u \ , \label{fF2} \\
738: {\p \over \p \xi} F_0(\xi, u_1) &=& {1 \over 2 (u_1 - \xi)^{3 \over 2}}
739: \int_{\xi}^{u_1} f''(u) {u_1 - u \over \sqrt{u - \xi} } d u \label{fF2'} \ .
740: \end{eqnarray}
741:
742: We first observe that
743: \begin{equation}
744: \label{f'}
745: \lim_{u_1 \rightarrow 1} \int_{\xi}^{u_1} {f'(u)
746: \over \sqrt{u - \xi } } d u = - \infty
747: \end{equation}
748: uniformly for $\xi$ on every compact subset of the interval $(0,1)$.
749: Since $f'(u) < 0$, we estimate for $u_1 > \xi$,
750: $$\int_{\xi}^{u_1} {f'(u) \over \sqrt{u - \xi } } d u \leq \int_{\xi}^{u_1} f'(u) d u
751: = f(u_1) - f(\xi) \ .$$
752: As $u_1 \rightarrow 1$, the right hand side goes to $- \infty$ uniformly for $\xi$
753: on every compact subset of the interval $(0,1)$ because $f(u_1) \rightarrow - \infty$
754: in view of (\ref{cf1}) and $f(u)$ is bounded on every compact subset.
755: This proves the limit (\ref{f'}).
756:
757: To calculate the limit (\ref{F'}), we turn to the following lemma on the uniform
758: convergence version of L'Hospital's rule. The proof of Lemma \ref{L'Hospital}
759: is simply a modification of the standard proof of the classical L'Hospital's rule.
760:
761: \begin{lem}
762: \label{L'Hospital}
763: Let $g_1(x,y)$ and $g_2(x,y)$ be defined on $(0, 1) \times Y$, where $Y$ is a subset of
764: $\Re$. If $g_1(x,y)$ and $g_2(x,y)$ satisfy the conditions
765: \begin{enumerate}
766:
767: \item for each $x \in (0,1)$, $g_1(x,y)$ and $g_2(x,y)$ are uniformly bounded for $y$ on $Y$,
768:
769: \item the partial derivatives $g_{1x}(x,y)$ and $g_{2x}(x,y)$ exist for each $(x,y) \in (0, 1)
770: \times Y$,
771:
772: \item $\lim_{x \rightarrow 1} g_2(x,y) = \infty$ uniformly for $y$ on $Y$,
773:
774: \item $\lim_{x \rightarrow 1} [g_{1x}(x,y) / g_{2x}(x,y)] = L(y)$ uniformly for $y$ on $Y$,
775:
776: \item $L(y)$ is uniformly bounded on $Y$.
777:
778: \end{enumerate}
779: then
780: $\lim_{x \rightarrow 1} [g_1(x,y) / g_2(x,y)] = L(y)$
781: uniformly for $y$ on $Y$.
782: \end{lem}
783:
784: Using formulae (\ref{fF2}) and (\ref{fF2'}), we apply Lemma \ref{L'Hospital} twice to calculate the limit
785: \begin{eqnarray*}
786: \lim_{u_1 \rightarrow 1} { {\p F_0(\xi, u_1) \over \p \xi} \over F_0(\xi, u_1) }
787: &=& \lim_{u_1 \rightarrow 1} {1 \over u_1 - \xi} { \int_{\xi}^{u_1}
788: { f''(u) \over \sqrt{u - \xi} } (u_1 - u) d u \over \int_{\xi}^{u_1} {f'(u)
789: \over \sqrt{u - \xi } } d u } \\
790: &=& {1 \over 1 - \xi} \lim_{u_1 \rightarrow 1} {\int_{\xi}^{u_1}
791: {f''(u) \over \sqrt{u - \xi} } d u \over {f'(u_1) \over \sqrt{u_1 - \xi} }} \\
792: &=& {1 \over \sqrt{1 - \xi} } \lim_{u_1 \rightarrow 1} { {f''(u_1) \over \sqrt{u_1 - \xi}}
793: \over f''(u_1)} \\
794: &=& {1 \over 1 - \xi} \ ,
795: \end{eqnarray*}
796: uniformly for $\xi$ on every compact subset of $(0,1)$.
797: Here, the limit (\ref{f'}) and the second limit of (\ref{cf3}) guarantee that Lemma \ref{L'Hospital}
798: applies.
799: This proves the limit (\ref{F'}).
800:
801: The proof of Lemma \ref{trailing} is now complete.
802:
803: We now use Lemma \ref{trailing} to study the minimum problem (\ref{min}). Because
804: of (\ref{F}), $[- F_0(\xi, u_1) / U_0(\xi, u_1)]$ goes to $+ \infty$ as
805: $\xi \rightarrow 0, 1$. Hence, the minimum is reached somewhere in
806: the interval $(0, 1)$.
807:
808: We next show that the minimum problem (\ref{min}) has only one critical point when $u_1$ is close
809: to $1$.
810: To see this, we calculate the second derivatives of $[- F_0 / U_0]$
811: at any of its critical points
812: \begin{equation}
813: \label{''} {\p^2 \over \p \xi^2} [- {F_0(\xi, u_1) \over U_0(\xi, u_1)}] =
814: - {F_{0 \xi \xi} U_0 - F_0 U_{0 \xi \xi} \over U_0^2} \ .
815: \end{equation}
816: Since $U_{0 \xi \xi} \equiv 0$ in view of (\ref{fU}), it follows from inequality
817: (\ref{F_0''}) that the right hand side is positive if $u_1$ is close to $1$.
818: Hence, all the critical points of $[- F_0 / U_0]$ must be local
819: minimizing points.
820: For any two minimizing points, some point between them
821: must be a local maximizing point; an impossibility. This proves
822: the uniqueness of the critical point and hence of the minimizing point
823: in the minimum problem (\ref{min}) when $u_1$ is close to $1$.
824:
825: \noindent
826: {\bf Remark:} \ {\em This is the first place where $m=2$ in (\ref{am}) is important.
827: It is because of $m=2$ that we have $U_{0 \xi \xi} \equiv 0$ which makes
828: it possible to conclude from formula (\ref{''}) the uniqueness of the critical point
829: in the the minimum problem (\ref{min}).}
830:
831: We also want to determine the asymptotics of the critical point when $t$ is large.
832: To do this, we write
833: \begin{equation}
834: \label{F0/U0}
835: {\p \over \p \xi} [- {F_0(\xi, u_1) \over U_0(\xi, u_1)}] =
836: - {F_0 \over U_0}[ {{\p F_0 \over \p \xi} \over F_0} -
837: {{\p U_0 \over \p \xi} \over U_0} ] \ .
838: \end{equation}
839: Since $F_0$ and $U_0$ are non-zero when $0 < \xi < 1$ and $0 < u_1 < 1$,
840: $[- F_0/U_0 ]$ has a unique critical point within the interval $(0,1)$
841: if and only if the quantity in the parenthesis on the right hand side does so.
842: We now consider a
843: compact set $[\epsilon_0, 1 - \epsilon_0]$, where $\epsilon_0$ is a small positive
844: number. According to (\ref{F'}),
845: the quantity in the parenthesis of (\ref{F0/U0}) has the uniform limit
846: \begin{equation}
847: \label{U0}
848: {1 \over 1 - \xi} - {{\p U_0(\xi, 1) \over \p \xi} \over U_0(\xi, 1)}
849: \end{equation}
850: on this compact set as $u_1 \rightarrow 1$. In view of formula (\ref{fU}) for $U_0$, we deduce that
851: the function (\ref{U0}) has only one zero $\xi = 1/4$ in the interval $0 < \xi < 1$,
852: counting multiplicities.
853: Hence, the quantity
854: in the parenthesis of
855: (\ref{F0/U0}) will have a zero near $\xi = 1/4$ when $u_1$ is close to $1$. This zero is exactly the unique
856: critical point of the minimum problem (\ref{min}).
857:
858: We summarize the above result as a lemma.
859: \begin{lem}
860: \label{trailing2}
861: The minimum problem
862: (\ref{min}) has one and only one critical point
863: when $u_1$ is close to $1$. Furthermore, the critical point approaches $1/4$ as
864: $u_1$ goes to $1$ .
865: \end{lem}
866:
867: We are now ready to solve equations (\ref{teg10}-\ref{teg30}). By Lemma \ref{trailing2},
868: equation (\ref{teg20'}) has
869: a unique solution $\xi = u_3$ for each $u_1$ close to $1$.
870: We then use (\ref{teg10'}) to determine $t$. Hence, equations (\ref{teg10'}-\ref{teg20'})
871: can be solved to give $u_3$ and $t$ as functions of $u_1$.
872:
873: We then calculate the Jacobian of equations (\ref{teg10}-\ref{teg20}) with respect to
874: $u_1$ and $u_3$ on the solution of (\ref{teg10'}-\ref{teg20'}):
875: \begin{eqnarray}
876: {\p W_1 \over \p u_1} &=& {\Phi_0(u_1, u_1) \over 2(u_1 - u_3)} > 0 \ , \label{W11} \\
877: {\p W_1 \over \p u_3} &=& 0 \ , \label{W13} \\
878: {\p W_2 \over \p u_1} &=& {\Phi_0(u_1, u_1) \over 2(u_1 - u_3)^2} > 0 \ , \label{W21} \\
879: {\p W_2 \over \p u_3} &=& - {\p^2 \over \p \xi^2} F_0(\xi, u_1)|_{\xi = u_3} > 0 \label{W23} \ .
880: \end{eqnarray}
881:
882: To prove the equality of (\ref{W11}), we use the special case of identity (\ref{Phiu})
883: $$2(\xi - u_1) {\p \over \p u_1} \Phi_0(\xi, u_1) = \Phi_0(\xi, u_1) - \Phi_0(u_1, u_1) \ .$$
884: This together with equation (\ref{teg10}) proves the equality of (\ref{W11}).
885:
886: Equation (\ref{W13}) is the result of (\ref{teg20}).
887:
888: The equality of (\ref{W21}) follows from (\ref{ph2}), (\ref{teg20}) and (\ref{W11}).
889:
890: The equality of (\ref{W23}) is the result of
891: the split (\ref{split}) and $U_{0 \xi \xi} \equiv 0$.
892:
893: The sign in (\ref{W23}) comes from (\ref{F_0''}).
894:
895: To determine the signs in (\ref{W11}) and (\ref{W21}), we split $\Phi_0(u_1, u_1)$
896: according to (\ref{split}),
897: \begin{eqnarray*}
898: \Phi_0(u_1,u_1) &=& - 30 t U_0(u_1,u_1) - F_0(u_1, u_1) \\
899: &=& U_0(u_1, u_1) [ {F_0(u_3, u_1) \over U_0(u_3, u_1)} -
900: {F_0(u_1, u_1) \over U_0(u_1, u_1)} ] > 0 \ .
901: \end{eqnarray*}
902: Here, we have used (\ref{teg10'}) in the second equality. The inequality follows
903: from the fact that $u_3$ is the only minimizing point of the minimum problem
904: (\ref{min}).
905:
906: It now follows from the Implicit Function Theorem that equations (\ref{teg10}-\ref{teg20})
907: can be solved to give $u_1$ and $u_3$ as functions of $t$,
908: respectively, for $t$ larger than a finite time. We have therefore established
909: the following theorem.
910:
911: \begin{theo}
912: \label{trailing3}
913: Under the conditions (\ref{cf1}) and (\ref{cf2}), there exists a constant $T^- > 0$ such that
914: for $t > T^-$, equations (\ref{teg10}-\ref{teg30})
915: have a unique solution $u_1^-(t)$, $u_3^-(t)$ and $x^-(t)$.
916: %with $u_3^-(t)$ being the
917: %unique minimizer of (\ref{min}) for each $u_1=u_1^-(t)$.
918: Function $u_1^-(t)$ is an increasing function of $t$.
919: Furthermore,
920: $$\lim_{t \rightarrow + \infty} u_1^-(t) = 1 \ , ~~~~~
921: \lim_{t \rightarrow + \infty} u_3^-(t) = {1 \over 4} \ . $$
922: \end{theo}
923:
924: \medskip
925:
926: \noindent
927: {\bf $\S$ 4.2 \quad The leading edge}
928:
929: We now turn to the leading edge at which $u_1 = u_2$. It is the head of a single phase solution (see Figure 1.).
930: It is governed by equations (\ref{leg1}-\ref{leg3})
931: for $g=0$:
932: \begin{eqnarray}
933: W_3(u_1, u_3, t) =: \Phi_0(u_1, u_3)
934: &=& 0 \ , \label{leg10} \\
935: W_4(u_1, u_3, t) =: \int_{u_3}^{u_1} \Phi_0(\xi, u_3) \sqrt{\xi - u_3} d \xi &=& 0 \ ,
936: \label{leg20} \\
937: x - 30 t u_3^2 - f(u_3) &=& 0 \ . \label{leg30}
938: \end{eqnarray}
939:
940: We again use the split $\Phi_0(\xi, u_3) = - 30 t U_0(\xi, u_3) - F_0(\xi, u_3)$,
941: which is similar to (\ref{split}),
942: to rewrite (\ref{leg10}-\ref{leg20}) as
943: \begin{eqnarray}
944: {\int_{u_3}^{u_1} F_0(u, u_3) \sqrt{u - u_3} d u \over
945: \int_{u_3}^{u_1} U_0(u, u_3) \sqrt{u - u_3} d u} + 30 t &=& 0 \ , \label{leg10'} \\
946: {\p \over \p u_1} \left[ {\int_{u_3}^{u_1} F_0(u, u_3) \sqrt{u - u_3} d u \over
947: \int_{u_3}^{u_1} U_0(u, u_3) \sqrt{u - u_3} d u} + 30 t \right ] &=& 0 \ . \label{leg20'}
948: \end{eqnarray}
949:
950: We first simplify equations (\ref{leg10'}-\ref{leg20'}).
951:
952: \begin{lem}
953: \label{FG}
954: $${1 \over 2 (u_1 - u_3)^{3 \over 2}} \int_{u_3}^{u_1} F_0(u, u_3) \sqrt{u - u_3} d u
955: = {1 \over 4 \sqrt{2}} \int_{-1}^1 f'({1 - \mu \over 2} u_3 + {1 + \mu \over 2} u_1)
956: \sqrt{1 - \mu} d \mu \ .$$
957: \end{lem}
958: Proof:
959: \begin{eqnarray*}
960: {\p \over \p u_1} \int_{u_3}^{u_1} F_0(u, u_3) \sqrt{u - u_3} d u
961: &=& F_0(u_1, u_3) \sqrt{u_1 - u_3} \\
962: &=& {1 \over 2} \int_{u_3}^{u_1} {f'(u) \over \sqrt{u_1 - u}} d u \\
963: &=& {\p \over \p u_1} \int_{u_3}^{u_1} f'(u) \sqrt{u_1 - u} d u \\
964: &=& {\p \over \p u_1} [{(u_1 - u_3)^{3 \over 2} \over 2 \sqrt{2}}
965: \int_{-1}^1 f'({1 - \mu \over 2} u_3 + {1 + \mu \over 2} u_1)
966: \sqrt{1 - \mu} d \mu] \ ,
967: \end{eqnarray*}
968: which proves the lemma.
969:
970: Let
971: \begin{eqnarray}
972: G_0(\xi, u_3) &=& {1 \over 4 \sqrt{2}} \int_{-1}^1 f'({1 - \mu \over 2} u_3 + {1 + \mu \over 2} \xi)
973: \sqrt{1 - \mu} d \mu \ , \label{fG} \\
974: V_0(\xi, u_3) &=& {1 \over 4 \sqrt{2}} \int_{-1}^1 2 ({1 - \mu \over 2} u_3 + {1 + \mu \over 2} \xi)
975: \sqrt{1 - \mu} d \mu = {4 \over 15 } \xi + {2 \over 5} u_3 \ . \label{fV}
976: \end{eqnarray}
977:
978: We use Lemma \ref{FG} to rewrite equations (\ref{leg10'}-\ref{leg20'}) as
979: \begin{eqnarray}
980: {G_0(\xi, u_3) \over V_0(\xi, u_3)} + 30 t &=& 0
981: \ , \label{leg10''} \\
982: {\p \over \p \xi} [{G_0(\xi, u_3) \over V_0(\xi, u_3)} + 30 t]_{\xi = u_1}
983: &=& 0 \ . \label{leg20''}
984: \end{eqnarray}
985: As before, equation (\ref{leg20''}) motivates us to consider another auxiliary minimum problem
986: \begin{equation}
987: \label{min2}
988: Min_{0 < \xi < 1} \left[ - {G_0(\xi, u_3) \over V_0(\xi, u_3)} \right]
989: \end{equation}
990: for each $u_3$ close to $0$.
991:
992: \begin{lem}
993: \label{leading}
994:
995: \begin{enumerate}
996:
997: \item For each $u_3 \in (0, 1)$,
998: \begin{equation}
999: \label{G}
1000: lim_{\xi \rightarrow 1} {\p \over \p \xi}
1001: \left[ - {G_0(\xi, u_3) \over V_0(\xi, u_3)} \right] = + \infty \ .
1002: \end{equation}
1003:
1004: \item For any small $\epsilon_1 > 0$, there exists a $\delta^* > 0$ such that
1005: \begin{equation}
1006: \label{G'}
1007: {\p \over \p \xi}
1008: \left[ - {G_0(\xi, u_3) \over V_0(\xi, u_3)} \right] < 0
1009: \end{equation}
1010: for $0 < \xi < 1 - \epsilon_1$ and $0 < u_3 < \delta^*$.
1011:
1012: \item
1013: \begin{equation}
1014: \label{G'2}
1015: {\p^2 \over \p \xi^2} G_0(\xi, u_3) < 0 \quad uniformly \ for \ 0 < \xi < 1
1016: \end{equation}
1017: when $u_3$ is close to $0$.
1018:
1019: \end{enumerate}
1020:
1021: \end{lem}
1022:
1023: Proof: We omit the proof of (\ref{G'2}) since it is similar to the proof of
1024: (\ref{F_0''}).
1025:
1026: To prove (\ref{G}), we write
1027: \begin{equation}
1028: \label{G/V}
1029: {\p \over \p \xi} [- {G_0(\xi, u_3) \over V_0(\xi, u_3)}] =
1030: - {{\p G_0 \over \p \xi} V_0 - G_0
1031: {\p V_0 \over \p \xi} \over V_0^2} \ .
1032: \end{equation}
1033:
1034: We use formula (\ref{fG}) to calculate $\p_{\xi} G_0$ and integrate by parts to obtain
1035: \begin{equation}
1036: \label{GF0}
1037: {\p G_0 \over \p \xi} = {F_0(\xi, u_3) - 3 G_0(\xi, u_3) \over 2(\xi - u_3)} \ ,
1038: \end{equation}
1039: where $F_0$ is previously defined by (\ref{fF}).
1040:
1041: Using (\ref{fG}) again, we write
1042: \begin{equation*}
1043: G_0(\xi, u_3) = {1 \over 2(\xi - u_3)^{3 \over 2}} \int_{u_3}^{\xi} f'(u) \sqrt{\xi -u}
1044: \ d u \ .
1045: \end{equation*}
1046: Since the integral is a decreasing function of $\xi$ because $f'(u) < 0$, $G_0(\xi, u_3)$
1047: goes to either a finite negative number or $- \infty$ as $\xi \rightarrow 1$.
1048:
1049: For those initial functions $f(u)$ such that $G_0(\xi, u_3)$ converges as $\xi \rightarrow 1$,
1050: the right hand side
1051: of (\ref{GF0}) goes to $- \infty$ because $F_0(\xi, u_3)$ does so by Lemma \ref{trailing};
1052: so $G_{0 \xi}(\xi, u_3) \rightarrow - \infty$. This together with (\ref{G/V}) gives the limit
1053: (\ref{G}).
1054:
1055: For those initial functions $f(u)$ such that $G_0(\xi, u_3) \rightarrow - \infty$ as $\xi \rightarrow 1$, we estimate
1056: $${F_0(\xi, u_3) \over G_0(\xi, u_3)} = { \int_{u_3}^{\xi} {f'(u) \over \sqrt{\xi - u}} d u \over (\xi - u_3)
1057: \int_{u_3}^{\xi} f'(u) \sqrt{\xi - u} d u} \geq { \int_{u_3}^{\xi} {f'(u) \over \sqrt{1 - u}} d u \over
1058: \int_{u_3}^{\xi} f'(u) \sqrt{1 - u} d u} \ .$$
1059: Using L'Hospital's rule, it is easy to see that the right hand side goes to $+ \infty$ as
1060: $\xi \rightarrow 1$. This limit, when combined with (\ref{GF0}), gives
1061: $$\lim_{\xi \rightarrow 1} {{\p G_0(\xi, u_3) \over \p \xi} \over G_0(\xi, u_3)}
1062: = + \infty \ , $$
1063: which, in view of (\ref{G/V}), proves (\ref{G}).
1064:
1065: The proof of the first part of Lemma \ref{leading} is completed.
1066:
1067: To prove (\ref{G'}), we use (\ref{G/V}) again.
1068: Since $V_0$ and $\p_{\xi} V_0$ are positive functions and $G_0$ is negative,
1069: it suffices to show that
1070: \begin{equation}
1071: \label{G0}
1072: {\p G_0(\xi, u_3) \over \p \xi} > 0
1073: \end{equation}
1074: for $0 < \xi < 1 - \epsilon_1$ and $0 < u_3 < \delta$.
1075:
1076: In view of the third limit of (\ref{cf3}), there exists a $\delta_2 > 0$ such that
1077: \begin{equation}
1078: \label{f''}
1079: f''(u) > 0 \quad \mbox{for $0 < u < \delta_2$} \ .
1080: \end{equation}
1081: It then follows from formula (\ref{fG}) for $G_0$ that
1082: inequality (\ref{G0}) is true for $0 < \xi < \delta_2$ and $0 < u_3 < \delta_2$.
1083:
1084: For $\delta_2 \leq \xi < 1 - \epsilon_1$ and $0 < u_3 < \delta_2 / 2$, we use formula (\ref{fG}) again to write
1085: \begin{eqnarray*}
1086: {\p G_0(\xi, u_3) \over \p \xi} &=& {1 \over 2(\xi - u_3)^{5 \over 2}} \int_{u_3}^{\xi} f''(u) \sqrt{\xi -u}
1087: (u - u_3) d u \no \\
1088: &=& {1 \over 2(\xi - u_3)^{5 \over 2}} \int_{u_3}^{\delta_2 \over 2} f''(u) \sqrt{\xi -u}
1089: (u - u_3) d u \\
1090: && + {1 \over 2(\xi - u_3)^{5 \over 2}} \int_{\delta_2 \over 2}^{\xi} f''(u) \sqrt{\xi -u} (u - u_3) d u \ .
1091: \end{eqnarray*}
1092: The second term is bounded from below by
1093: $$ - \ {2^{3 \over 2} \over \delta_2^{5 \over 2}} \int_{\delta_2 \over 2}^{1 - \epsilon_1} |f''(u)|
1094: d u \ ,$$
1095: which is a constant.
1096: Since $f'' > 0$ for $0 < u < \delta_2$, the first term is bigger than
1097: $$ {\sqrt{\delta_2} \over 2 \sqrt{2}} \int_{u_3}^{\delta_2 \over 2} f''(u) (u - u_3) d u
1098: = {\sqrt{\delta_2} \over 2 \sqrt{2}} [({\delta_2 \over 2} - u_3)f'({\delta_2 \over 2}) - f({\delta_2 \over 2}) +
1099: f(u_3)] \ ,$$
1100: which goes to $+ \infty$ as $u_3 \rightarrow 0$ because of (\ref{cf1}).
1101: We have therefore proved (\ref{G0}) and hence (\ref{G'}).
1102:
1103: The proof of Lemma \ref{leading} is complete.
1104:
1105: We now use Lemma \ref{leading} to study the minimization problem (\ref{min2}).
1106: For each small $u_3$, the function $[- G_0(\xi, u_3) / V_0(\xi, u_3)]$ is decreasing when $\xi < 1 - \epsilon_1$
1107: in view of (\ref{G'}) and tending to $+ \infty$ as $\xi \rightarrow 1$ in view of (\ref{G}).
1108: Hence, the minimum is reached somewhere in the
1109: interval $(0, 1)$.
1110:
1111: We next show that the minimum problem (\ref{min2}) has only one critical point when $u_3$ is close
1112: to $0$.
1113: To see this, we calculate the second derivatives of $[- G_0 / V_0]$
1114: at any of its critical points
1115: \begin{equation}
1116: \label{GV2}
1117: {\p^2 \over \p \xi^2} [- {G_0(\xi, u_3) \over V_0(\xi, u_3)}] =
1118: - {G_{0 \xi \xi} V_0 - G_0 V_{0 \xi \xi} \over V_0^2} \ .
1119: \end{equation}
1120: Since $V_{0 \xi \xi} \equiv 0$ from (\ref{fV}), the right hand side of
1121: (\ref{GV2}) is positive
1122: if $u_3$ is close to $0$, in view
1123: of (\ref{G'2}). Hence, all the critical points of $[- G_0 / V_0]$ must be local
1124: minimizing points.
1125: For any two minimizing points, some point between them
1126: must be a local maximizing point; an impossibility. This proves
1127: the uniqueness of the critical point and hence of the minimizing point
1128: in the minimum problem (\ref{min2}) when $u_3$ is close to $0$.
1129:
1130: \noindent
1131: {\bf Remark:} \ {\em This is the second place where $m=2$ in (\ref{am}) plays an important role.
1132: Since $m=2$, $V$ of (\ref{fV}) is a linear function. It then becomes possible to conclude from
1133: formula (\ref{GV2}) that the minimum problem (\ref{min2}) has a unique critical point.}
1134:
1135: The critical point approaches $1$ as $u_3$ goes to $0$. This follows from
1136: (\ref{G'}).
1137:
1138: We summarize the above result as a lemma.
1139: \begin{lem}
1140: \label{leading2}
1141: The minimum problem
1142: (\ref{min2}) has one and only one critical point and hence minimizing point
1143: when $u_3$ is close to $0$. Furthermore, the critical point tends to $1$ as
1144: $u_3$ goes to $0$.
1145: \end{lem}
1146:
1147: We are now ready to solve equations (\ref{leg10}-\ref{leg30}). By Lemma \ref{leading2},
1148: equation (\ref{leg20''}) has
1149: a unique solution $\xi = u_1$ for each $u_3$ close to $0$.
1150: We then use (\ref{leg10''}) to determine $t$. Hence, equations (\ref{leg10''}-\ref{leg20''})
1151: or equivalently equations (\ref{leg10}-\ref{leg30}) can be solved to give $u_3$ and $t$ as
1152: functions of $u_1$.
1153:
1154: We then calculate the Jacobian of equations (\ref{leg10}-\ref{leg20}) with respect to
1155: $u_1$ and $u_3$ on the solution of (\ref{leg10''}-\ref{leg20''}):
1156: \begin{eqnarray}
1157: {\p W_3 \over \p u_1} &=& {\p \Phi_0(\xi, u_3) \over \p \xi}|_{\xi = u_1} > 0 \ , \label{W31} \\
1158: {\p W_3 \over \p u_3} &=& - {\Phi_0(u_3, u_3) \over 2(u_1 - u_3)} < 0 \ , \label{W33} \\
1159: {\p W_4 \over \p u_1} &=& 0 \ , \label{W41} \\
1160: {\p W_4 \over \p u_3} &=& - \Phi_0(u_3, u_3) \sqrt{u_1 - u_3} < 0 \ . \label{W43}
1161: \end{eqnarray}
1162:
1163: The equalities in (\ref{W33}) and (\ref{W43}) follow from the related identity (\ref{Phiu})
1164: for $\Phi_0(\xi, u_3)$.
1165:
1166: Equation (\ref{W41}) is the result of (\ref{teg20}).
1167:
1168: The signs in (\ref{W33}) and (\ref{W43}) are determined as follows.
1169: Since the minimum problem (\ref{min2}) has only one minimizing point at
1170: $\xi = u_1$, we must have
1171: $$ - {G_0(u_3, u_3) \over V_0(u_3, u_3)} > - {G_0(u_1, u_3) \over V_0(u_1, u_3)} \ .$$
1172: The right hand side is equal to $30 t$ in view of (\ref{leg10''}). Hence,
1173: $- 30 t V_0(u_3, u_3) - G_0(u_3, u_3) > 0 \ ,$ which by formulae (\ref{fG}-\ref{fV}), reduces to
1174: $$- 60 t u_3 - f'(u_3) > 0 \ .$$
1175: In view of the boundary condition (\ref{ph3}) for $\Phi_0$, this is exactly $\Phi_0(u_3, u_3) > 0$,
1176: which proves the inequalities in (\ref{W33}) and (\ref{W43}).
1177:
1178: To determine the sign in (\ref{W31}), since $\xi = u_1$ is the minimizing point in the minimum
1179: problem (\ref{min2}), we have
1180: $$ {\p^2 \over \p \xi^2} [ - {G_0(\xi, u_3) \over V_0(\xi, u_3)} ] |_{\xi = u_1} > 0 \ ,$$
1181: which, in view of Lemma \ref{FG}, is equivalent to
1182: $${\p^2 \over \p \xi^2} \left[ - {\int_{u_3}^{\xi} F_0(u, u_3) \sqrt{u - u_3} d u \over
1183: \int_{u_3}^{\xi} U_0(u, u_3) \sqrt{u - u_3} d u} \right ]|_{\xi = u_1} > 0 \ . $$
1184: This together with (\ref{leg10'}-\ref{leg20'})) gives
1185: $${\p^2 \over \p \xi^2} \int_{u_3}^{\xi} [- 30 t U_0(u, u_3) - F_0(u, u_3)] \sqrt{u - u_3} d u
1186: > 0 $$ at $\xi = u_1$. Since $\Phi_0 = - 30 t U_0 - F_0$, the last inequality together
1187: with (\ref{leg10}) implies that
1188: $${\p \Phi_0(\xi, u_3) \over \p \xi}|_{\xi=u_1} > 0 \ ,$$
1189: which proves the inequality in (\ref{W31}).
1190:
1191: We have therefore established the following theorem.
1192:
1193: \begin{theo}
1194: \label{leading3}
1195: Under the conditions (\ref{cf1}) and (\ref{cf2}), there exists a constant $T^+ > 0$ such that
1196: equations (\ref{leg10}-\ref{leg30})
1197: have a unique solution $u_1^+(t)$, $u_3^+(t)$ and $x^+(t)$ for $t > T^+$.
1198: Functions $u_1^+(t)$ and $u_3^+(t)$ are increasing and decreasing functions of $t$, respectively.
1199: They have the limits:
1200: $$\lim_{t \rightarrow + \infty} u_1^+(t) = 1 \ , ~~~~~
1201: \lim_{t \rightarrow + \infty} u_3^+(t) = 0 \ . $$
1202: Furthermore,
1203: $$ 1 - \delta < u_1^+(t) < 1 \ , \quad 0 < u_3^+(t) < \delta_2 \quad \mbox{for $t > T^+$} \ ,$$
1204: where $\delta$ and $\delta_2$ are defined in (\ref{F_0''}) and (\ref{f''}), respectively.
1205: \end{theo}
1206:
1207: The second part of Theorem \ref{leading3} is only for technical purpose. It will be needed
1208: in the next section.
1209:
1210:
1211: \bigskip
1212:
1213: \refstepcounter{section}
1214: \begin{center}
1215: {\bf $\S$ 5 \quad Analysis of Zero and Single Phase Solutions}
1216: \end{center}
1217:
1218: In this section, we shall show that the hodograph transform (\ref{P2})
1219: $g=0, 1$, can be inverted to give solutions $u_1(x,t)$, $\cdots$, $u_{2g+1}(x,t)$
1220: for $- \infty < x < \infty$ and $t > max \{ T^-, T^+ \}$, where $T^-$ and $T^+$
1221: are given in Theorems \ref{trailing3} and \ref{leading3}, respectively. We will also verify
1222: inequalities (\ref{constraint}-\ref{constraint2}). By theorem \ref{main}, the
1223: function $\psi$ of (\ref{psi}) will be the minimizer of (\ref{mini}).
1224:
1225: For each $t > max \{ T^-, T^+ \}$, we split the space into three pieces $(- \infty, x^-(t)]$,
1226: $[x^-(t), x^+(t)]$ and $[x^+(t), + \infty)$. Here, $x^-(t)$ and $x^+(t)$ are given
1227: in Theorems \ref{trailing3} and \ref{leading3}, respectively.
1228:
1229: \noindent
1230: {\bf $\S$ 5.1 \quad Zero phases: $x \leq x^-(t)$ and $x \geq x^+(t)$}
1231:
1232: We shall focus on the zero phase over the interval $x \geq x^+(t)$.
1233: The other zero phase over $x \leq x^-(t)$ can be handled in the same way.
1234:
1235: For $x \geq x^+(t)$, we shall use the auxiliary minimum problem (\ref{min2}) to
1236: show that the minimizer of the minimization problem (\ref{mini})
1237: has a compact support $[0, \sqrt{u_3(x,t)}]$. The function $u_3(x,t)$ is governed by
1238: equation (\ref{P2}), which, when $g=0$, becomes
1239: \begin{equation}
1240: \label{5Burgers'}
1241: x - 30 t u_3^2 - f(u_3) = 0 \ .
1242: \end{equation}
1243:
1244: For each $0 < u_3 < u_3^+(t)$ where $u_3^+(t)$ is given in Theorem \ref{leading3}, since
1245: $u_3^+(t)$ is a continuous function of $t$ and $u_3^+(t) \rightarrow 0$ as $t \rightarrow
1246: + \infty$,
1247: there is a $t^* > t$ such that $u_3 = u_3^+(t^*)$. In view of the minimum problem
1248: (\ref{min2}), we have for $0 < \xi < 1$,
1249: $$ - {G_0(\xi, u_3) \over V_0(\xi, u_3)} \geq - {G_0(u_1^+(t^*), u_3) \over V_0(u_1^+(t^*), u_3)}
1250: = 30 t^* > 30 t \ ,$$
1251: where the equality follows from equation (\ref{leg10''}). Hence,
1252: \begin{equation}
1253: \label{B}
1254: - 30 t V_0(\xi, u_3) - G_0(\xi, u_3) > 0
1255: \end{equation}
1256: for all $0 < \xi < 1$. In particular, when $\xi = u_3$, inequality (\ref{B})
1257: reduces to $- 60 t u_3 - f'(u_3) > 0$
1258: in view of formulae (\ref{fG}) and (\ref{fV}). This shows that equation (\ref{5Burgers'})
1259: can be inverted to give $u_3$ as a decreasing function of $x$ for $x > x^+(t)$.
1260:
1261: We next verify inequalities (\ref{constraint}-\ref{constraint2}), which now take the form
1262: \begin{eqnarray}
1263: Re\{\sqrt{-1} \sqrt{\xi - u_3}\}\Phi_0(\xi,
1264: u_3) & < & 0 \quad
1265: \mbox{for $0 < \xi < u_3$} \ , \label{constraint'} \\
1266: \int_{u_3}^{\xi}\sqrt{u - u_3} \Phi_0(u, u_3)
1267: d u & > & 0 \quad \mbox{for $\xi > u_3$} \ .
1268: \label{constraint2'}
1269: \end{eqnarray}
1270:
1271: Inequality (\ref{constraint2'}) follows easily from (\ref{B}) and Lemma \ref{FG}.
1272:
1273: To prove (\ref{constraint'}), we use $u_3 < u_3^+(t)$ again. Since $u_3^+(t) < \delta_2$ according
1274: to Theorem \ref{leading3}, we have $u_3 < \delta_2$. We then use (\ref{f''}) in formula
1275: (\ref{fF}) for $F_0$ to obtain that
1276: $${\p F_0(\xi, u_3) \over \p \xi} > 0$$
1277: for $\xi \leq u_3 < \delta_2$. Hence,
1278: $$ {\p \Phi_0(\xi, u_3) \over \p \xi} = - 30 t {\p U_0(\xi, u_3) \over \p \xi}
1279: - {\p F_0(\xi, u_3) \over \p \xi} < 0 $$
1280: for all $\xi \leq u_3$. This together with $\Phi_0(u_3, u_3) =
1281: - 60 t u_3 - f'(u_3) > 0$ proves $\Phi_0(\xi, u_3) > 0$ for
1282: $\xi \leq u_3$ and hence inequality (\ref{constraint'}).
1283:
1284: \begin{theo}
1285: \label{Burgers2}
1286: Under the conditions (\ref{cf1}) and (\ref{cf2}), for each $t > max \{ T^-, T^+ \}$, the
1287: minimizer of the minimization problem (\ref{mini}) is supported on a single interval
1288: when $x \leq x^-(t)$ and $x \geq x^+(t)$.
1289: \end{theo}
1290:
1291: \noindent
1292: {\bf $\S$ 5.2 \quad Single phase: $x^-(t) < x < x^+(t)$}
1293:
1294: For $t > max \{ T^-, T^+ \}$ and $x^-(t) < x < x^+(t)$, we shall
1295: show that the minimizer of (\ref{mini})
1296: is supported on two disjoint intervals $[0, \sqrt{u_3(x,t)}]$ and $[\sqrt{u_2(x,t)},
1297: \sqrt{u_1(x,t)}]$, where $u_1(x,t)$, $u_2(x,t)$ and $u_3(x,t)$ are governed by
1298: equations (\ref{P2}) when $g=1$
1299: \begin{equation}
1300: \label{P2'}
1301: P(u_1, u_1, u_2, u_3) = 0 \ , ~~ P(u_2, u_1, u_2, u_3) = 0 \ , ~~ P(u_3, u_1, u_2, u_3) = 0 \ .
1302: \end{equation}
1303:
1304: We will first solve equations (\ref{P2'}) near the leading edge $x=x^+(t)$.
1305: We will then extend the solution over the whole interval $x^-(t) < x < x^+(t)$.
1306:
1307: The first two equations of (\ref{P2'}) are degenerate near the leading edge
1308: at which $u_1 = u_2$. We replace them by
1309: \begin{eqnarray}
1310: W_5(u_1, u_2, u_3, t) =: {\p q_{1,1} \over \p u_1} + {\p q_{1,1} \over \p u_2}
1311: + {\p q_{1,1} \over \p u_3} &=& 0 , \label{l1} \\
1312: W_6(u_1, u_2, u_3, t) =: \int_{u_3}^{u_2} \sqrt{(u_1 - \xi)(u_2 - \xi)(\xi - u_3)}
1313: \Phi_1(\xi, u_1, u_2, u_3) d \xi &=& 0 , \label{l2}
1314: \end{eqnarray}
1315: where $\Phi_1$ and $q_{1,1}$ are given by (\ref{ph1}-\ref{ph3}) and (\ref{q1}-\ref{q2}), respectively.
1316: It can be shown that equations (\ref{l1}) and (\ref{l2}) together with $P(u_3, \vu) = 0$
1317: are equivalent
1318: to equations (\ref{P2'}) when $u_1 > u_2 > u_3$ \cite{gra1}.
1319:
1320: It is also known that equations (\ref{l1}) and (\ref{l2}) transform into
1321: equations (\ref{leg10}-\ref{leg20}) when $u_1=u_2$ and
1322: that equation $P(u_3, \vu) = 0$ into (\ref{leg30}) \cite{gra1}.
1323:
1324: For each fixed $t$, we now solve equations (\ref{l1}) and (\ref{l2}) for $u_1$ and $u_3$ in terms of
1325: $u_2$ near the leading edge. We use the identities of Lemma \ref{gidentity} to calculate the derivatives
1326: of $W_5$ and $W_6$ at the point $(u_1^+(t), u_1^+(t), u_3^+(t))$:
1327: \begin{eqnarray}
1328: {\p W_5 \over \p u_1} &=& {\p W_5 \over \p u_2} = {1 \over 2} {\p \Phi_0(\xi, u_3) \over \p \xi}|_{\xi = u_1} > 0 \ , \label{W51} \\
1329: {\p W_5 \over \p u_3} &=& {\Phi_0(u_1, u_3) - \Phi_0(u_3, u_3) \over 2(u_1 - u_3)}
1330: = - {\Phi_0(u_3, u_3) \over 2(u_1 - u_3)} < 0 \ , \label{W53} \\
1331: {\p W_6 \over \p u_1} &=& {\p W_6 \over \p u_2} = {1 \over 2} \int_{u_3}^{u_2} \sqrt{\xi - u_3}
1332: [2(u_1 - \xi) {\p \Phi_1(\xi, \vu) \over \p u_1} + \Phi_1(\xi, \vu) ] d \xi \no \\
1333: &=& {(u_1 - u_3)^{3 \over 2} \over 3} \Phi_1(u_1, \vu) = {(u_1 - u_3)^{3 \over 2} \over 3} {\p \Phi_0(\xi, u_3)
1334: \over \p \xi}|_{\xi = u_1} > 0 \ , \label{W61} \\
1335: {\p W_6 \over \p u_3} &=& - \int_{u_3}^{u_2} {u_1 - \xi \over 2 \sqrt{\xi - u_3} }
1336: [ 2 (u_3 - \xi) {\p \Phi_1(\xi, \vu) \over \p u_3} + \Phi_1(\xi, \vu) ] d \xi \no \\
1337: &=& - \int_{u_3}^{u_2} {u_1 - \xi \over 2 \sqrt{\xi - u_3} } \Phi_1(u_3, \vu) d \xi \no \\
1338: &=& - {2 (u_1 - u_3)^{3 \over 2} \over 3} [ {\Phi_0(u_3, u_3) - \Phi_0(u_1, u_3) \over u_3 - u_1} ] \no \\
1339: &=& {2 \sqrt{u_1 - u_3} \over 3 } \Phi_0(u_3, u_3) > 0 \ , \label{W63}
1340: \end{eqnarray}
1341: where $\vec{u} = (u_1, u_2, u_3)$.
1342:
1343: The second equality of (\ref{W51}) follows from identity (\ref{lim3}).
1344:
1345: The first equality of (\ref{W53}) is a consequence of (\ref{Phiu}) and (\ref{lim3}). The second
1346: equality follows from equation (\ref{leg10}).
1347:
1348: The third and last equalities of (\ref{W61}) follows from identities (\ref{Phiu}) and (\ref{lim2}),
1349: respectively.
1350:
1351: The second, third and last equalities of (\ref{W63}) follows from (\ref{Phiu}), (\ref{lim1}) and
1352: (\ref{leg10}), respectively.
1353:
1354: All the signs are easily determined according to (\ref{W31}-\ref{W43}).
1355:
1356: Therefore, equations (\ref{l1}) and (\ref{l2}) can be inverted to give $u_1$ and $u_3$ as
1357: functions of $u_2$ when $u_2$ is close to and less than $u_1^+(t)$. Moreover,
1358: $u_1$ decreases as $u_2$ increases. We then use the last equation of (\ref{P2'}) to give
1359: $x$ as a function of $u_2$. We summarize these result in the lemma.
1360:
1361: \begin{lem}
1362: \label{leadingn}
1363: For each $t > T$, equations (\ref{P2'}) can be inverted to give $x$, $u_1$ and $u_3$
1364: as functions of $u_2$ when $u_2$ is close to and less than $u_1^+(t)$.
1365: \end{lem}
1366:
1367: Having solved equations (\ref{P2'}) for $x$, $u_1$ and $u_3$ as functions of $u_2$
1368: near the leading edge $u_2 = u_2^+(t)$, we now extend the solution by decreasing $u_2$.
1369: To calculate the Jacobian of (\ref{P2'}), we turn to Lemma \ref{jaco}, which says
1370: that the zeros of the Jacobian are determined by the $\xi$-zeros of $\Phi_1(\xi, u_1,u_2,u_3)$.
1371: We again split
1372: \begin{equation}
1373: \label{split3}
1374: \Phi_1(\xi, u_1,u_2,u_3) = - 30 t U_1(\xi, u_1,u_2,u_3) - F_1(\xi, u_1,u_2,u_3) \ ,
1375: \end{equation}
1376: where $U_1$ and $F_1$ satisfy the Euler-Poisson-Darboux equations (\ref{ph1}-\ref{ph3})
1377: with the boundary values $4/3$ and $2 f''(u)/3$, respectively. Hence, $U_1$ is a constant;
1378: indeed, $U_1 = 4/3$.
1379:
1380: \begin{lem}
1381: \label{Ph1}
1382: For each $t$, $u_1>u_2>u_3$,
1383: $$lim_{\xi \rightarrow 0} \Phi_1(\xi, u_1,u_2,u_3) = - \infty \ , ~~~~~
1384: lim_{\xi \rightarrow 1} \Phi_1(\xi, u_1,u_2,u_3) = + \infty \ .$$
1385: \end{lem}
1386: Proof: Since $U_1$ is a constant, we see from (\ref{split3}) that it suffices to prove that
1387: \begin{equation}
1388: \label{F1L}
1389: lim_{\xi \rightarrow 0} F_1(\xi, u_1,u_2,u_3) = + \infty \ , ~~~~~
1390: lim_{\xi \rightarrow 1} F_1(\xi, u_1,u_2,u_3) = - \infty \ .
1391: \end{equation}
1392:
1393: We first
1394: write $F_1(\xi, u_1, u_2, u_3)$ in terms of
1395: $F_0(\xi, u_1)$. We see from
1396: (\ref{ph1}) that
1397: $F_1$ satisfies
1398: \begin{displaymath}
1399: 2(u_2 - u_3) {\p^2 F_1 \over \p u_2 \p u_3} = {\p F_1 \over \p u_2}
1400: - {\p F_1 \over \p u_3}
1401: \end{displaymath}
1402: and that it has the boundary values $F_1(\xi, u_1, u, u) = {F_0(\xi, u_1)
1403: - F_0(u, u_1) \over \xi - u}$ on account of (\ref{lim1}).
1404: It then follows from formula (\ref{EPQsS}) at $\rho=1$ that
1405: \begin{equation*}
1406: F_1(\xi, u_1,u_2,u_3) = {1 \over \pi} \int_{u_3}^{u_2}
1407: {F_0(\xi, u_1) - F_0(u, u_1) \over (\xi -u) \sqrt{(u_2 -u)(u - u_3)}}
1408: d u \ ,
1409: \end{equation*}
1410: which together with (\ref{F}) gives (\ref{F1L}). This proves Lemma \ref{Ph1}.
1411:
1412: \begin{lem}
1413: \label{m=2}
1414: Under the conditions (\ref{cf1}) and (\ref{cf2}),
1415: $\Phi_1(\xi, u_1,u_2,u_3)$ has at most one $\xi$-zero for all $t$, $u_1$, $u_2$ and
1416: $u_3$ whenever $1 - \delta < u_1 < 1$, where $\delta$ is given by Theorem \ref{leading3}.
1417: \end{lem}
1418: Proof: We first observe that
1419: \begin{equation}
1420: \label{unique}
1421: {\p \over \p \xi} F_1(\xi, u_1,u_2,u_3) < 0
1422: \end{equation}
1423: for all $\xi$, $u_1$, $u_2$ and $u_3$ whenever $1 - \delta < u_1 < 1$.
1424: To see this, we will derive a new formula for $F_1$ in terms of $F_0$. For each fixed $u_1$,
1425: we view $F_1(\xi, u_1, u_2, u_3)$
1426: as a function of $\xi$, $u_2$ and $u_3$ only. Hence, equations (\ref{ph1}) and (\ref{ph2}) for $F_1$
1427: reduce to
1428: \begin{equation}
1429: \label{new1}
1430: 2(u_2 - u_3) {\p^2 F_1 \over \p u_2 \p u_3} = {\p F_1 \over \p u_2} - {\p F_1 \over \p u_3} \ ,
1431: \ 2(\xi - u_i) {\p^2 F_1 \over \p \xi \p u_i} = {\p F_1 \over \p \xi} - 2 {\p F_1 \over \p u_i} \ , \ i = 2 ,3 \ .
1432: \end{equation}
1433: The new boundary condition is
1434: \begin{equation}
1435: \label{new2}
1436: F_1(u, u_1, u, u) = {\p F_0(\xi, u_1) \over \p \xi}|_{\xi = u}
1437: \end{equation}
1438: on account of (\ref{lim2}).
1439: One can then use the method of Section 2 to derive a double integral formula, similar to (\ref{phi}),
1440: for the solution of
1441: equations (\ref{new1}-\ref{new2}). Indeed, we have
1442: \begin{eqnarray*}
1443: \lefteqn{F_1(\xi, u_1,u_2,u_3)} \no \\
1444: &=& {1 \over 2 \sqrt{2} \pi} \int_{-1}^1 \int_{-1}^1
1445: {F_{0 \xi}({1 + \mu \over 2}{1 + \nu \over 2} \xi + {1 + \mu \over 2}{1 - \nu \over 2} u_2
1446: +{1 - \mu \over 2} u_3, u_1) \over \sqrt{(1 - \mu)(1 - \nu)}} \sqrt{1 + \mu} \ d \mu d \nu \ ,
1447: \end{eqnarray*}
1448: which, when combined with (\ref{F_0''}), gives (\ref{unique}).
1449:
1450: Since $U_1$ is a constant, we deduce from
1451: the split (\ref{split3}) that
1452: \begin{equation}
1453: \label{u2}
1454: {\p \over \p \xi} \Phi_1(\xi, u_1,u_2,u_3) = - {\p \over \p \xi} F_1(\xi, u_1,u_2,u_3) > 0
1455: \quad \mbox{for $1 - \delta < u_1 < 1$} \ ,
1456: \end{equation}
1457: where the inequality follows from (\ref{unique}).
1458: This proves the
1459: uniqueness of the $\xi$-zero of $\Phi_1(\xi, u_1,u_2,u_3)$ and hence Lemma \ref{m=2}.
1460:
1461: \noindent
1462: {\bf Remark:} \ {\em This is the third and also the last place where $m=2$ is important.
1463: If $m > 2$, (\ref{u2}) is not true.}
1464:
1465: We now continue to extend the solution of equations (\ref{P2'}) by decreasing $u_2$.
1466:
1467: By Lemma \ref{Ph1}, the function $\Phi_1(\xi, u_1,u_2,u_3)$ must have at least a $\xi$-zero.
1468: Indeed, it
1469: has a zero, denoted by $u^*$, between $u_3$ and $u_2$ in view of equation
1470: (\ref{l2}). According to Lemma \ref{m=2}, $u^*$ is the only zero of $\Phi_1$.
1471: Hence,
1472: \begin{equation}
1473: \label{Ph1><}
1474: \Phi_1(\xi, u_1, u_2, u_3) > 0 \quad \mbox{for $\xi > u^*$} \ and
1475: \ \Phi_1(\xi, u_1, u_2, u_3) < 0 \quad \mbox{for $\xi < u^*$} \ .
1476: \end{equation}
1477: Since $u_3 < u_2 < u_1$ and $u_3 < u^* < u_2$, we have
1478: \begin{equation}
1479: \label{2}
1480: \Phi_1(u_1, u_1,u_2,u_3) > 0 \ , ~ \Phi_1(u_2, u_1,u_2,u_3) > 0 \ , ~ \Phi_1(u_3, u_1,u_2,u_3) < 0 \ .
1481: \end{equation}
1482:
1483: For $P_{1,0}$ of (\ref{pgn}-\ref{lc}), we also have
1484: \begin{equation}
1485: \label{3}
1486: P_{1,0}(u_1, u_1,u_2,u_3) > 0 \ , ~ P_{1,0}(u_2, u_1,u_2,u_3) > 0 \ , ~ P_{1,0}(u_3, u_1,u_2,u_3) < 0 \ .
1487: \end{equation}
1488: To see this, we observe from (\ref{pgn}) that $P_{1,0}$ is a linear function of $\xi$ and that
1489: it thus has only one $\xi$-zero. This zero is between $u_3$ and $u_2$ on account of equation
1490: (\ref{lc}). Inequalities (\ref{3}) are then justified.
1491:
1492: By Lemma \ref{jaco}, we deduce from inequalities (\ref{2}) that the Jacobian of (\ref{P2'}) is always non-zero.
1493: It then follows from the
1494: decomposition (\ref{1}) for $P$ and inequalities (\ref{3}) that equations (\ref{P2'})
1495: can be inverted to give $x$, $u_1$, and $u_3$ as increasing, decreasing, and decreasing functions of $u_2$,
1496: respectively, provided that $u_3 < u_2 < u_1$ and $1 - \delta < u_1 < 1$.
1497: We now extend the solution as far as possible by decreasing
1498: $u_2$; so $u_1$ always stays in the interval $(1 - \delta, 1)$ because $u_1$ increases as $u_2$ decreases.
1499: Since both $u_1$ and $u_3$ increase as $u_2$ decreases, the solution
1500: will stop at $u_2$ = $u_3$. At this point, $x$, $u_1$, $u_2 = u_3$ satisfy the trailing edge equations
1501: (\ref{teg10}-\ref{teg30}). According to Lemma \ref{trailing}, the equations have a unique
1502: solution for $t > T$. We thus have $x=x^-(t)$, $u_1= u_1^-(t)$, $u_2=u_3=u_3^-(t)$.
1503:
1504: We have therefore proved that equations (\ref{P2'}) can be inverted to give $u_1$, $u_2$ and
1505: $u_3$ as functions of $x$ for $x^-(t) < x < x^+(t)$.
1506:
1507: Inequalities (\ref{constraint}-\ref{constraint2}) are immediate consequences of (\ref{l2}) and (\ref{Ph1><}).
1508: We have therefore established the following theorem.
1509: \begin{theo}
1510: \label{Whitham}
1511: Under the conditions (\ref{cf1}) and (\ref{cf2}), for each $t > max \{ T^-, T^+ \}$, the
1512: minimizer of the minimization problem (\ref{mini}) is supported on two disjoint intervals
1513: when $x^-(t) < x < x^+(t)$.
1514: \end{theo}
1515:
1516: The main theorem of this paper, Theorem \ref{Main}, follows from Theorems \ref{Burgers2} and
1517: \ref{Whitham}.
1518:
1519:
1520:
1521: \bigskip
1522:
1523: {\bf Acknowledgments.}
1524: We thank Tamara Grava for valuable discussions.
1525: V.P. was supported in part by NSF Grant DMS-0135308. F.-R. T. was supported in part by
1526: NSF Grant DMS-0404931 and by a John Simon Guggenheim Fellowship.
1527:
1528:
1529:
1530: \begin{thebibliography}{50}
1531:
1532: \bibitem{dub} B.A. Dubrovin and S.P. Novikov, ``Hydrodynamics of Weakly
1533: Deformed Soliton Lattices. Differential Geometry and Hamiltonian Theory'',
1534: Russian Math. Surveys 44:6 (1989), 35-124.
1535:
1536: %\bibitem{dub3} B.A. Dubrovin, ``Functionals of Peierls-Frolich type and
1537: %Variational Principal of Whitham Equations'', Amer. Math. Soc. Transl.
1538: %179 (1997), 35-44.
1539:
1540: \bibitem{fla} H. Flaschka, M.G. Forest and D.W. McLaughlin,
1541: ``Multiphase Averaging and the Inverse Spectral Solution of the
1542: Korteweg-de Vries Equation'', Comm. Pure Appl. Math. 33 (1980), 739-784.
1543:
1544: \bibitem{gra1} T. Grava and F.R. Tian, ``The Generation, Propagation and
1545: Extinction of Multiphases in the KdV Zero Dispersion Limit'', Comm. Pure Appl. Math.
1546: 55(2002), 1569-1639.
1547:
1548: \bibitem{gra2} T. Grava and F.R. Tian, ``Large Parameter Behavior of Equilibrium Measures'',
1549: Comm. Math. Sci. 4(2006), 551-573.
1550:
1551: \bibitem{gur} A.V. Gurevich and L.P. Pitaevskii, ``Non-stationary Structure
1552: of a Collisionless Shock Wave'', Soviet Phys. JETP 38 (1974), 291-297.
1553:
1554: \bibitem{kri} I.M. Krichever, ``The Method of Averaging for
1555: Two-dimensional `Integrable' Equations'', Functional Anal. App.
1556: 22 (1988), 200-213.
1557:
1558: \bibitem{lax} P.D. Lax and C.D. Levermore, ``The Small Dispersion
1559: Limit for the Korteweg-de Vries Equation I, II, and III'', Comm. Pure Appl.
1560: Math.
1561: 36 (1983), 253-290, 571-593, 809-830.
1562:
1563: \bibitem{lax2} P.D. Lax, C.D. Levermore and S. Venakides, ``The Generation and
1564: Propagation of Oscillations in Dispersive IVPs and Their Limiting Behavior''
1565: in Important Developments in Soliton Theory 1980-1990, T. Fokas and V.E.
1566: Zakharov eds., Springer-Verlag, Berlin (1992).
1567:
1568: \bibitem{lev} C.D. Levermore, ``The Hyperbolic Nature of the Zero Dispersion
1569: KdV Limit'', Comm. P.D.E. 13 (1988), 495-514.
1570:
1571: \bibitem{pie} V.U. Pierce and F.R. Tian, ``Self-similar Solutions of the Non-Strictly Hyperbolic
1572: Whitham Equations'', preprint, arxiv.org, nlin.SI/0607061.
1573:
1574: \bibitem{Tian1} F.R. Tian, ``Oscillations of the Zero Dispersion Limit of the
1575: Korteweg-de Vries Equation'', Comm. Pure Appl. Math. 46 (1993), 1093-1129.
1576:
1577: \bibitem{Tian2} F.R. Tian, ``The Whitham Type Equations and Linear
1578: Overdetermined
1579: Systems of Euler-Poisson-Darboux Type'', Duke Math. Jour. 74 (1994), 203-221.
1580:
1581: \bibitem{Tian3} F.R. Tian, ``The Initial Value problem for the Whitham Averaged
1582: System'', Comm. Math. Phys. 166 (1994), 79-115.
1583:
1584: \bibitem{tsa} S.P. Tsarev, ``Poisson Brackets and One-dimensional
1585: Hamiltonian Systems of Hydrodynamic Type'', Soviet Math. Dokl.
1586: 31 (1985), 488-491.
1587:
1588: \bibitem{ven} S. Venakides, ``The Zero Dispersion Limit of the KdV Equation
1589: with Nontrivial Reflection Coefficient'', Comm. Pure Appl. Math.
1590: 38 (1985), 125-155.
1591:
1592: \bibitem{ven2} S. Venakides, ``Higher order Lax-Levermore Theory'',
1593: Comm. Pure Appl. Math. 43 (1990), 335-362.
1594:
1595: \bibitem{whi} G.B. Whitham, ``Non-linear Dispersive Waves'', Proc. Royal Soc.
1596: London Ser. A 139 (1965), 283-291.
1597:
1598: \end{thebibliography}
1599:
1600: \end{document}
1601:
1602: