math0410131/BKM.tex
1: 
2: \documentclass[12pt]{article}     \usepackage{amsmath}
3: \usepackage{amsthm}               \usepackage{amsopn}
4: \usepackage{psfig}                \usepackage{graphics}
5: \usepackage{latexsym}             \usepackage{amsfonts}
6: \usepackage{amssymb}
7: \setlength{\topmargin}{0.0in}     \setlength{\textwidth}{5.5in}
8: \setlength{\oddsidemargin}{0.5in} \setlength{\textheight}{8.5in}
9: % \documentclass{article}       \pagestyle{empty}
10: \swapnumbers
11: \theoremstyle{plain}
12: \newtheorem{theorem}{Theorem}[section]
13: \newtheorem{proposition}[theorem]{Proposition}
14: \newtheorem{lemma}[theorem]{Lemma}
15: \newtheorem{corollary}[theorem]{Corollary}
16: 
17: \theoremstyle{definition}
18: \newtheorem{definition}[theorem]{Definition}
19: \newtheorem{remark}[theorem]{Remark}
20: 
21: \newenvironment{pf}{{\noindent\bf Proof.}}
22: 
23: \newcommand{\nc}[2]{ \newcommand{#1}{#2} }
24: \newcommand{\pnu}[1]{ \ensuremath{\frac{\partial #1}{\partial\nu}} }
25: % partial derivative in the normal direction
26: \nc{\avint}{ {- \hspace{-3.5mm} \int} }  % average integral
27: \newcommand{\myavint}[1]{ \int_{#1} \mkern-27mu
28:                           \rule[.033 in]{.12 in}{.01 in} \ \ \  }
29: \nc{\R}{{\rm {I \! R}}}  % Real Numbers
30: \newcommand{\closure}[1]{ \stackrel{\rule{.1 in}{.01 in}}{#1} }
31: \newcommand{\dclosure}[1]{ \stackrel{\rule{.2 in}{.01 in}}{#1} }
32: \newcommand{\tclosure}[1]{ \stackrel{\rule{.3 in}{.01 in}}{#1} }
33: \newcommand{\qclosure}[1]{ \stackrel{\rule{.4 in}{.01 in}}{#1} }
34: \newcommand{\norm}[4]{ || #1 ||_{_{{ #2 }^{ #3 }( #4 )}} }
35: \newcommand{\chisub}[1]{ {\mathbf{\chi}}_{_{#1}} }
36: 
37: \newcommand{\newsec}[2]{ \section{#1} \label{sec-#2}  % starts new section,
38:                          \setcounter{equation}{0}     % resets counters,
39:                          \setcounter{theorem}{0} }    % makes label
40: \newcommand{\newsub}[2]{ \subsection{#1} \label{sub-#2} }
41: % starts new subsection, makes label
42: \newcommand{\refeqn}[1]{ (\!\!~\ref{eq:#1}) } % gives references to
43: \newcommand{\refthm}[1]{ (\!\!~\ref{#1}) }    % equations or theorems
44: 
45: \nc{\Holder}{H\"{o}lder\ }
46: \newcommand{\vhat}[1]{ \hat{\mathbf{#1}} }
47: \newcommand{\vnorm}[1]{ ||\vec{#1}|| }
48: \newcommand{\parder}[2]{ \frac{\partial #1}{\partial #2} }
49: \newcommand{\dotvecs}[2]{ \vec{#1} \bullet \vec{#2} }
50: \newcommand{\crvecs}[2]{ \vec{#1} \times \vec{#2} }
51: \nc{\ith}{ \ensuremath{\text{i}^{\text{th}}} }
52: \nc{\jth}{ \ensuremath{\text{j}^{\text{th}}} }
53: \nc{\kth}{ \ensuremath{\text{k}^{\text{th}}} }
54: \nc{\curl}{ \nabla \times }
55: \nc{\Div}{ \nabla \cdot }
56: 
57: \newcommand{\BVP}[4]{  % 1 = equation, 2 = bdry vals, 3 = label, 4 = set.
58:   \begin{equation}
59:         \begin{array}{rl}
60:            #1 & \ \text{in}
61:                \ \ #4 \vspace{.05in} \\
62:            #2 & \ \text{on} \ \ \partial #4 \;.
63:         \end{array}
64:   \label{eq:#3}
65:   \end{equation}    }
66: \newcommand{\BVPom}[3]{   \BVP{#1}{#2}{#3}{ \Omega }  }
67: \newcommand{\BVPb}[3]{   \BVP{#1}{#2}{#3}{ B_{1} }  }
68: 
69: \begin{document}
70: 
71: \numberwithin{equation}{section}
72: 
73: \title{ The Hele-Shaw problem as a ``Mesa'' limit\\
74:         of Stefan problems: Existence, uniqueness,\\
75:         and regularity of the free boundary }
76: \author{Ivan A. Blank\\
77:         Marianne K. Korten\\
78:         Charles N. Moore\\
79: \normalsize}
80: %\date{}
81: 
82: \maketitle
83: 
84: \begin{abstract}
85: We study a Hele-Shaw problem with a mushy region obtained as a mesa
86: type limit of one phase Stefan problems in exterior domains.  We
87: study the convergence, determine some of the qualitative properties
88: and regularity of the unique limiting solution, and prove regularity
89: of the free boundary of this limit under very general conditions on
90: the initial data.  Indeed, our results handle changes in topology
91: and multiple injection slots.
92: 
93: \
94: 
95: \noindent
96: \textit{Key Words:} Mesa problem; Hele-Shaw problem; Stefan problem; free boundary;
97: mushy region; singular limit;
98: 
99: \
100: 
101: \noindent
102: \textit{2000 Mathematics Subject Classification:} 76D27, 35K65, 49J40
103: \end{abstract}
104: 
105: \newsec{Introduction}{Intro}
106: 
107: Given a bounded domain $D \in \R^{n}$ with smooth boundary $\partial D,$ a 
108: finite set of closed curves $\{ s^{j}_{0}(x) = 0 \}$ such that $D$ is contained
109: in the union of their interiors, and a continuous function $p(x,t)$ defined on 
110: $\partial D \times (0, \infty),$ the classical Hele-Shaw problem with Dirichlet 
111: data is frequently formulated as follows:  Find a function $V(x,t)$ and a family 
112: of domains $S(t)$ (which each contain $D$) with $\partial S(t) = \{ t = s(x) \}$
113: and such that
114: \begin{equation}
115: \begin{array}{rll}
116:    \Delta_x V &= 0 \ \ &(x, \; t) \in S(t) \setminus D \\
117:    V(x, \; t) &= 0 \ \ &(x, \; t) \in \partial S(t) \\
118:    V(x, \; t) &= p(x) \ \ &(x, \; t) \in \partial D \times (0, \; \infty) \\
119:    \nabla_x V \cdot \nabla_x s &= \frac{\partial s}{\partial t} \ \ 
120:                                &(x, \; t) \in \partial S(t) \\
121:    \partial S(0) &= \cup_j \{ s^{j}_{0}(x) = 0 \} \;. \ \ & \ 
122: \end{array}
123: \label{eq:classicalHS}
124: \end{equation}
125: 
126: This problem models the advance of the slick formed by injecting oil between
127: two nearby plates, and has further been used in injection molding (used in
128: turn in the packaging industry, and more generally for the production of
129: plastic components, for example interior pieces of cars and aircraft), in
130: electrochemical machining (see \cite{MR}), and even to predict tumor growth
131: (see \cite{BF}).  Sometimes the normal derivative of $V$ at the ``slot,'' 
132: $\partial D,$ is prescribed, or curvature dependent terms are included in the
133: free boundary condition.  Among the most pressing open questions concerning
134: the Hele-Shaw problem are finding a weak formulation, studying the
135: regularity of $V(x,t)$ in $t,$ and determining the regularity of the free
136: boundary $t = s(x).$  Another question which has long attracted interest
137: is whether a Hele-Shaw problem could have a ``mushy'' region.
138: 
139: In turn, the ``Mesa'' problem describes the limit pattern 
140: $\lim_{m \rightarrow \infty} u_m$ of solutions $u_m$ of, say, the porous
141: medium equation, when the initial data are held fixed.  This problem
142: first appeared in connection with the modeling of problems related to
143: transistors (see \cite{EHKO}).  Caffarelli and Friedman studied the 
144: initial value problem in $\R^{n} \times [0,T]$ in \cite{CF}.  
145: They proved that the limit exists, that it is independent of the chosen 
146: subsequence, that it is independent of time, that it is equal to the 
147: characteristic function of one set plus the initial data times the 
148: characteristic function of the complement of that set, and finally
149: that that set can be characterized as the noncoincidence set of a 
150: variational inequality.  Further developments showed that the same 
151: conclusions hold for the limit when $u^m$ is replaced by a fairly 
152: general monotone constitutive function $\phi(u)$ with $\phi(0) = 0$ 
153: (see \cite{FH} for example), moreover, this behavior is a property of 
154: fairly general semigroups (see \cite{I}, \cite{BEG}).  A Mesa problem 
155: for an equation which gives rise to a mushy region was studied in 
156: \cite{BKM}.  For the Mesa problem we study in this paper we show all of 
157: the properties shown by \cite{CF} mentioned above, except that our 
158: limits will not be independent of time.  (This evolution in time is 
159: natural since we work in an outer domain where the inner boundary data 
160: serves as a source.)
161: 
162: In this paper the authors exploit a Mesa limit setting in an outer
163: domain $D^{c}$ to obtain naturally a weak formulation of the Hele-Shaw
164: problem (with Dirichlet condition as in Equation\refeqn{classicalHS}on 
165: the slot as above).  The use of one-phase Stefan problems with with 
166: ``mushy'' regions and with increasing diffusivities naturally produces a mushy
167: region when we permit initial data $u_{I}$ for the approximating problems
168: to take values in the interval $[0, \; 1].$  These $u_{I}$ can be thought
169: of as generalized characteristic functions.  Another aspect of this approach
170: which is extremely attractive is the fact that changes in the topology of the 
171: ``wet'' region do not interfere with the construction.  In short, whereas other
172: authors have taken a priori assumptions which ensure that their free boundary
173: stays smooth, we have been able to show existence of weak solutions for all
174: time, regardless of the possible changes in topology.  (Note that in \cite{CF} and
175: \cite{FH} they assume that their data is starlike with respect to the origin, and
176: in \cite{DL} log concavity of initial data is assumed to guarantee existence and
177: smoothness of the solutions.)  Indeed there are some very natural problems arising
178: in the applications where the topology should change.  Consider for example the 
179: problem of what happens with Hele-Shaw flow around an obstacle.  In this case, 
180: there is automatically a change in topology when the flow meets itself on the 
181: other side.  An interesting question is whether or not an air bubble will be left 
182: behind in the wake of the obstacle.  Another obvious problem from applications
183: where there will be changes in topology is if there are multiple injection slots.
184: In fact, this paper already deals with the second situation, since we never 
185: assume that $D$ is connected.
186: 
187: The classical version of the m-approximating problem which we use is given as follows:
188: \begin{definition}[m-approximating problem]   \label{mWholeMHS}
189: We let $u^{(m)}$ denote the solution of the following partial differential equation,
190: \begin{equation}
191: u^{(m)}_t = m \Delta (u^{(m)}(x,t) - 1)_{+}, \ \ \ \
192:             (x,t) \in D^{c} \times (0, \; + \infty) \cap \{ u^{(m)} > 1 \},
193: \label{eq:MesaHS}
194: \end{equation}
195: with boundary data given by
196: \begin{equation}
197:    \left.
198:    \begin{array}{rll}
199:       u^{(m)}(x,0) \! &= u_{I}(x), \; \ & x \in D^{c}, \\
200: %      m(u^{(m)}(x,t) - 1)_{+} \! &= p(x,t), \; \
201:       m(u^{(m)}(x,t) - 1)_{+} \! &= p(x), \; \
202:             & (x,t) \in \partial D \times (0, \; + \infty),
203:    \end{array}
204:    \ \ \ \ \right\}
205: \label{eq:bdryHS}
206: \end{equation}
207: and with the free boundary condition
208: \begin{equation}
209:     (-\nabla m(u^{(m)} - 1)_{+}, (1 - u_{I}) \;) \cdot \nu = 0 \;, \ \ 
210:     (x,t) \in \partial \{ u^{(m)} > 1 \} \;.
211: \label{eq:fbmHS}
212: \end{equation}
213: Here $\nu$ is the outer (n + 1) dimensional normal to the set $\{ u^{(m)} > 1 \},$
214: and this free boundary condition will be satisfied when the free boundary is smooth.
215: The weak formulation we give at the beginning of the next section will not require
216: any regularity assumptions on the free boundary or on the initial data.
217: For a fixed $m > 0,$ we call the free boundary problem determined by the equations
218: above the \textit{m-approximating problem}.  We will assume
219: \begin{equation}
220:    \begin{array}{l}
221:      0 \leq u_{I} \leq 1 \ \text{is compactly supported, } \\
222:      0 < p(x) \in C^{2, \alpha}(\partial D) \;, \\
223: %     p(x,t) \ \text{is nondecreasing for each fixed} \ t \;, \\
224:      D \ \text{is a bounded set with} \ \partial D \in C^{2, \alpha}. \\
225:    \end{array}
226: \label{eq:makeass}
227: \end{equation}
228: \end{definition}
229: \noindent
230: \begin{remark}[Simplifying Assumptions]  \label{rmk:SA}
231: There are two assumptions which we make which serve to simplify the exposition but 
232: which are absolutely \textit{not} necessary for the derivation of our results:
233: \begin{enumerate}
234:    \item
235: The assumption that $u_{I}$ has compact support is only technical and serves to
236: simplify the construction of certain comparison functions in the fifth section.  We 
237: leave to the reader the verification that it suffices to assume only that 
238: $\{u_{I} = 1\}$ is compact for all of the results before the sixth section.
239:    \item
240: The assumption that the Dirichlet data, $p(x),$ on the boundary of the slot is a 
241: function of $x$ alone can be replaced with the assumption that the Dirichlet data
242: $p(x,t)$ is a nondecreasing function of $t$ for each fixed $x.$  On the other hand,
243: if $p(x,t)$ is allowed to decrease in time, then Lemma\refthm{Monoumint}does not 
244: hold in general, and this appears to be essential for the methods presented in this
245: paper.
246: \end{enumerate}
247: \end{remark}
248: For convenience, we will define
249: \begin{equation}
250:    M := ||p||_{L^{\infty}(\partial D)} \;.
251: \label{eq:Mdef}
252: \end{equation}
253: %We will refer to the set $D$ as the \textit{slot.}
254: Now in terms of the positivity
255: assumption on $p(x),$ we note that $p(x) \equiv 0$ leads to a trivial case:
256: There will be no evolution at all.  To see this fact extend each $u^{(m)}$ by
257: $1$ across all of $D.$ In short, the positivity of $p(x)$ is driving the
258: evolution.
259: 
260: In this formulation $u^{(m)}$ is energy (or enthalpy), and $m(u^{(m)}(x,t) - 1)_{+}$
261: is temperature.  Equation\refeqn{MesaHS}comes from conservation of energy.
262: Basically, as $m$ increases, the diffusion happens faster.  Competing with this
263: increase in diffusion is the fact that the boundary data for $u^{(m)}$ on 
264: $\partial D$ is decreasing down to $1.$  As $m \rightarrow \infty$ we have
265: convergence of our operators to the following picture which is typical for Mesa
266: problems:
267:  
268: \
269: 
270: \psfig{file=mpic.eps}
271: 
272: \
273: 
274: We show that as $m \rightarrow \infty,$ the $u^{(m)}$ converge pointwise
275: to a limit $0 \leq u^{(\infty)} \leq 1,$ and $m(u^{(m)} - 1)_{+}$ tend 
276: pointwise to a bounded function $V(x, \; t).$  Furthermore, the function $V$ is
277: identically zero in $\{u^{(\infty)} < 1 \}$ and positive and harmonic for 
278: fixed $t$ in the component of the set $\{u^{(\infty)} = 1\}$ which contains $D.$ 
279: (On sets where $\{u^{(\infty)} = 1\}$ which are isolated from $D$ we will have
280: $V \equiv 0,$ see Remark\refthm{Vnc}\!\!.)
281: Finally, the pair $(u^{(\infty)}, \; V)$ solves our weak Hele-Shaw problem with 
282: a mushy region which we give formally in Definition\refthm{HSmr}\!\!.  In this way
283: we obtain the following results.
284: \begin{enumerate}
285:   \item We get a natural generalization of the weak formulation of the Hele-Shaw 
286:         problem of DiBenedetto and Friedman (see \cite{DF}).  Our formulation
287:         allows for a ``mushy'' region.  Moreover, after invoking a result
288:         of Bouillet, we will be able to say that solutions of our formulation
289:         are unique, and hence the Mesa limiting procedure we use gives an
290:         effective method of constructing the solution.
291:   \item We determine the regularity of the spatial slices of the free boundary 
292:         by invoking the results of Blank, Caffarelli, and Kinderlehrer and 
293:         Nirenberg for the regularity of the free boundary in the obstacle
294:         problem (see \cite{Bl}, \cite{C}, \cite{KN}).  Indeed, if we define
295:      \begin{equation}
296:         U(x,t) := \int_{0}^{t} u^{(\infty)}_{t}(x, \; s) \; ds
297:                 = (1 - u_{I}(x))\chisub{A(t)}(x)
298:      \label{eq:Unowdef}
299:      \end{equation}
300:         where $A(t)$ gives the ``puddle'' at time $t,$ and
301:      \begin{equation}
302:         W(x,t) := \int_{0}^{t} V(x, \; s) \; ds \;,
303:      \label{eq:Wnowdef}
304:      \end{equation}
305:         and we formally apply the Baiocchi transformation to the equation
306:      \begin{equation}
307:         u^{(\infty)}_{t}(x, \; t) = \Delta_x V(x, \; t) \;,
308:      \label{eq:prelin}
309:      \end{equation}
310:         then for every fixed time $t_0,$ $W$ satisfies the obstacle problem:
311:      \begin{equation}
312:         0 \leq W(x,t_0) \;, \ \ \ \ \Delta_x W(x,t_0) = U(x,t_0) \;.
313:      \label{eq:UWsatOP}
314:      \end{equation}
315: \end{enumerate}
316: In a subsequent paper the authors will address the rectifiability and further space-time
317: regularity of the free boundary.
318: 
319: The paper is arranged as follows:  In section 2 we introduce our notion of weak
320: solutions to the approximating problem, show some qualitative properties of these
321: solutions, derive the existence of the limits (for now in weak-$* \ L^{\infty}$),
322: and give some trivial bounds on these limits.  Most of the results in this section
323: draw from the maximum principle and from the papers \cite{AK}, \cite{K1}, \cite{K2}, 
324: and \cite{K3}.  Section 3 gives a simple counter-example which motivates the definition 
325: of the free boundary and of the diffusive region.  In section 4 we derive some 
326: monotonicity properties of our sequences and limits, and as consequences we improve our 
327: weak-$* \ L^{\infty}$ convergence to pointwise convergence and give an explicit 
328: representation of $u^{(\infty)}.$  In section 5 we show that the limiting problem has a 
329: free boundary for all finite time (as opposed to if it ``escaped to infinity'' in zero 
330: time or by a fixed time), and we use explicit subsolutions to show that if an open ball 
331: within the diffusive region has a boundary point on the free boundary, then the free 
332: boundary will have nonzero velocity at that point.  In section 6 we apply the Baiocchi 
333: transformation to Equation\refeqn{prelin}and thereby derive a family of 
334: obstacle problems that this procedure yields.  At that point, under suitable assumptions
335: on the initial data, we can invoke the regularity theory for the obstacle problem (see 
336: \cite{Bl} and \cite{C}) to immediately derive regularity in space for the free boundary 
337: of our Hele-Shaw problem for almost every time.  In section 7, we extend the results of 
338: section 6 to include every time.  In the process of extending to every time, we establish
339: the continuity of the measure of the diffusive region with respect to time when the
340: initial data is strictly less than one.  Section 8 is an appendix which gives some of the 
341: barrier functions and some of their properties that we need in some of the earlier 
342: sections.
343: 
344: I. Blank wishes to thank Kansas State University for their hospitality while
345: the bulk of this paper was written.  M. Korten is indebted to Juan Luis 
346: V\'{a}zquez for introducing her to the Hele-Shaw problem.  M. Korten was partially 
347: supported by NSF EPSCoR Grant \#530517 under agreement NSF32169/KAN32170, 
348: and I. Blank's visits to Kansas State University were also supported by this 
349: grant.  All three authors wish to thank Barbara Korten for her patience and 
350: support while this paper was written.
351: 
352: %-----------------------------------------------------------------------
353: 
354: \newsec{The weak formulations}{TWF}
355: 
356: For our m-approximating problem, we need an appropriate weak formulation which we
357: give here:
358: 
359: \begin{definition}[Weak solutions of the m-approximating problem] \label{WSmAP}
360: The nonnegative function $u^{(m)}(x, \; t) \in L^{1}_{loc}$ is a weak solution of
361: the m-approximating problem if for any
362: $\varphi \in C^{\infty}\left(\R^n \times [0, \infty)\right),$ such that
363: $\varphi \equiv 0$ on $\partial D \times [0, \infty),$ and $\varphi(x,t) \rightarrow 0$
364: as either $t \rightarrow \infty$ or $|x| \rightarrow \infty$ we have
365: \begin{equation}
366: \begin{array}{l}
367:  \displaystyle{\int_{D^c} \int_{0}^{\infty} \varphi_t(x,t) u^{(m)}(x,t) \; dt \; dx +
368:                  \int_{D^c} \int_{0}^{\infty} 
369:    \Delta_x \varphi(x,t) m \left[ u^{(m)}(x,t) - 1 \right]_{+} \; dt \; dx} \\
370:    \ \\
371:    = \displaystyle{\int_{\partial D} \int_{0}^{\infty}
372:         \frac{\partial \varphi}{\partial \nu}(x,t) p(x) \; dt \; d\mathcal{H}^{n - 1} x
373:                  -  \int_{D^c} \varphi(x,0) u_{I}(x) \; dx} \;.
374: \end{array}
375: \label{eq:wmap}
376: \end{equation}
377: \end{definition}  \noindent
378: We observe that the traces of $u^{(m)}$ and $m(u^{(m)} - 1)_{+}$ on the boundaries of
379: our domain will be well-defined by the work of Korten even if we only assume that we
380: are dealing with local solutions and that the initial trace is between zero and one.  
381: (See \cite{K1} which adapts the work of Dahlberg and Kenig, \cite{DK}, and see Lemma 
382: 3.2 of \cite{K2}.)  Furthermore, we observe that the free boundary condition for the 
383: classical formulation (i.e. Equation\refeqn{fbmHS}\!) will be satisfied, whenever the 
384: functions and sets are sufficiently smooth.
385: 
386: For strictly local situations, it may be
387: helpful to note that if $\varphi \in C_{0}^{\infty}(\Omega)$ where
388: $\Omega \subset \subset D^c \times (0, + \infty),$ then we have
389: \begin{alignat*}{1}
390:    &\int_{D^c} \int_{0}^{\infty} \varphi_t(x,t) u^{(m)}(x,t) \; dt \; dx \\
391:    = - &\int_{D^c} \int_{0}^{\infty} m \Delta_x \varphi(x,t) \left[ u^{(m)}(x,t) - 1 \right]_{+}
392:                                \; dt \; dx \\
393:    = + &\int_{D^c} \int_{0}^{\infty} m \nabla_x \varphi(x,t) \cdot
394:                                      \nabla_x \left[ u^{(m)}(x,t) - 1 \right]_{+}
395:                                \; dt \; dx \;.
396: \end{alignat*}
397: The last integration by parts requires that we invoke the known regularity theory
398: and energy estimates for solutions of Equation\refeqn{MesaHS}\!\!.  See Lemma 1.2 of
399: \cite{AK}.
400: 
401: Since we want to discuss regularity of functions in Sobolev spaces, we need to fix ideas
402: about which representative we will use.  Although the ``Lebesgue point'' representative
403: where points are filled in by taking limits of averages over balls with radii going to
404: zero is the most common procedure, we will use a slightly different approach which exploits
405: the fact that the functions $u^{(m)}(x,t)$ are nondecreasing in $t$ for almost every $x$
406: by Lemma 4.2 of \cite{K1} and by the fact that the boundary data $p(x,t)$ is nondecreasing
407: in time for each fixed $x.$
408: \begin{definition}[Representative]   \label{fctrep}
409: For $t > 0,$ we set $$\tilde{u}^{(m)}(x, \; t) := \lim_{s \uparrow t} u^{(m)}(x, \; s) 
410: \ \ \text{for} \ a.e. \ x.$$
411: \end{definition}  \noindent
412: For economy of notation we will not bother to relabel any of our $u^{(m)}$'s, but always
413: understand that we are using $\tilde{u}^{(m)}$ as our representative for $u^{(m)}$ in all
414: pointwise matters.
415: 
416: We cannot expect $u^{(m)}$ to be continuous across the free boundary, and in
417: fact, the following theorem summarizes the results of Lemma 2.4 and Lemma 2.5 of \cite{K3}:
418: \begin{theorem}[$u^{(m)}$ must jump]  \label{ummj}
419: The set 
420: $$E := \{ x \in \R^n \; : 
421:        \; \exists \ t \ s.t. \; u_I(x) < u^{(m)}(x, \; t) < 1 \} $$
422: has Lebesgue $n$-dimensional measure equal to zero.
423: \end{theorem}
424: \noindent
425: On the other hand, $m(u^{(m)} - 1)_{+}$ will be continuous across the free boundary.  
426: (See \cite{K1} which verifies the required assumptions of \cite{DB} for continuity.)
427: For a fixed $m$ we now have the following picture of $u^{(m)}$ viewed from the side:
428: 
429: \
430: 
431: \psfig{file=mprofile.eps}
432: 
433: \
434: 
435: 
436: By using the boundedness of the data along with the maximum principle that
437: the equation above enjoys, we can conclude that the solutions are bounded and
438: therefore by elementary functional analysis we can conclude that the following 
439: limit
440: \begin{equation}
441:          u^{(\infty)} := \lim_{m \rightarrow \infty} u^{(m)}(x,t), 
442: \label{eq:MesaLims}
443: \end{equation}
444: exists weak-$* \ L^{\infty}$ along a subsequence of $m$ in the entire domain.  
445: Again by using the maximum principle it is immediate that
446: \begin{equation}
447:          0 \leq u^{(\infty)} \leq 1 \;.
448: \label{eq:OneBdd}
449: \end{equation}
450: In fact, by using the maximum principle again, we can assert that
451: \begin{equation}
452:     0 \leq u^{(m)} \leq 1 + \frac{M}{m} \ \ \ \ \ \text{or} \ \ \ \ \ 
453:     0 \leq m(u^{(m)} - 1)_{+} \leq M \;,
454: \label{eq:mMfun}
455: \end{equation}
456: and we stress that $M$ is independent of $m$ and $t.$  Because of this fact, we
457: can take a further subsequence to ensure that $m(u^{(m)} - 1)_{+}$ has a 
458: limit $V$ in the weak-$* \ L^{\infty}$ topology of the entire domain.  
459: 
460: \begin{remark} \label{WooHooI} By combining Equation\refeqn{mMfun}with 
461: Theorem\refthm{ummj}we can conclude that the essential range of 
462: $u^{(m)}(x, \; \cdot)$ is a subset of $\{ u_I(x) \} \cup [1, \; 1 + M/m] \;.$
463: \end{remark}
464: 
465: Next, for $\varphi \in H^{1}_{0}(\Omega)$ we have:
466: 
467: \begin{alignat*}{1}
468:     \int_{0}^{+\infty} \int_{D^{c}} 
469:  \nabla_x \left[ m(u^{(m)} - 1)_{+} \right] \cdot \nabla_x \varphi \; dx \; dt
470:   & = \int_{0}^{+\infty} \int_{D^{c}} u^{(m)} \varphi_t \; dx \; dt \\
471:   & \rightarrow \int_{0}^{+\infty} \int_{D^{c}} u^{(\infty)} \varphi_t \; dx \; dt \;.
472: \end{alignat*}
473: Since $u^{(\infty)} \in L^{\infty}(\Omega) \subset L^{2}(\Omega),$ we have
474: \begin{equation}
475:    u^{(\infty)}_{t} \in H^{-1}(\Omega)
476: \label{eq:hmo}
477: \end{equation}
478: We summarize with the following lemma whose proof is now obtained trivially by
479: using the weak-$* \ L^{\infty}$ compactness we have and taking the limits as 
480: $m \rightarrow \infty$ in the weak formulation of the m-approximating problem. 
481: \begin{lemma}[Limits solve the limiting problem]  \label{limlimp}
482: Under our assumptions as above, we have 
483: \begin{equation}
484:    0 \leq V \leq M \;, 
485: \label{eq:VlimProb}
486: \end{equation}
487: and the pair $(u^{(\infty)}, \; V)$ satisfies
488: \begin{equation}
489: \!\!
490: \begin{array}{rl}
491:     &\displaystyle{\int_{D^c} \int_{0}^{\infty} \varphi_t(x,t) u^{(\infty)}(x,t) \; dt \; dx +
492:    \int_{D^c} \int_{0}^{\infty} \Delta_x \varphi(x,t) V(x,t) \; dt \; dx} \\
493:    \ \\
494:    = &\displaystyle{\int_{\partial D} \int_{0}^{\infty}
495:         \frac{\partial \varphi}{\partial \nu}(x,t) p(x) \; dt \; d\mathcal{H}^{n - 1} x 
496:     - \int_{D^c} \varphi(x,0) u_{I}(x) \; dx} \;.
497: \end{array}
498: \label{eq:whsprob}
499: \end{equation}
500: for any
501: $\varphi \in C^{\infty}\left(\R^n \times [0, \infty)\right),$ such that
502: $\varphi \equiv 0$ on $\partial D \times [0, \infty),$ and $\varphi(x,t) \rightarrow 0$
503: as either $t \rightarrow \infty$ or $|x| \rightarrow \infty.$
504: \end{lemma}
505: \begin{definition}[Hele-Shaw with mushy region]  \label{HSmr}
506: Any pair $(u^{(\infty)}, \; V)$ which satisfies Equation\refeqn{whsprob}for $\varphi$ as
507: in the lemma above will be called a weak solution of the Hele-Shaw problem with boundary
508: data $p(x)$ and initial data $u_{I}(x).$
509: \end{definition}
510: \begin{remark}[Inclusion of earlier models]  \label{Cwom}
511: The weak formulation we have above generalizes the Hele-Shaw formulation of DiBenedetto and
512: Friedman to include the situation where there is a mushy region (see \cite{DF}).  In
513: addition, our a priori regularity assumptions on the solutions are weaker.  (We only assume
514: $L^{1}_{loc}.$)
515: \end{remark}
516: %\begin{remark}[Traces are well-behaved]  \label{tawb}
517: %Because local solutions of
518: %\begin{equation}
519: %   \int_{D^c} \int_{0}^{\infty} \varphi_t(x,t) u^{(\infty)}(x,t) \; dt \; dx 
520: %   = - \int_{D^c} \int_{0}^{\infty} \Delta_x \varphi(x,t) V(x,t) \; dt \; dx 
521: %\label{eq:lstwb}
522: %\end{equation}
523: %have well-defined traces (see \cite{K2}), we can conclude that $V(x,t) = p(x)$ on
524: %$\partial D,$ in a suitable sense.
525: %\end{remark}
526: \begin{theorem}[Uniqueness among all solutions]   \label{yitoo}
527: The solution of the limiting problem of Lemma\refthm{limlimp}is unique, so all
528: solutions of the Hele-Shaw problem as given in Definition\refthm{HSmr}are recoverable 
529: via the Mesa limit process we have introduced.
530: \end{theorem}
531: \begin{pf}
532: This theorem is an immediate application of the uniqueness theorem in section 3 
533: of \cite{Bo}.
534: \newline Q.E.D. \newline
535: \end{pf}
536: 
537: %-----------------------------------------------------------------------
538: 
539: \newsec{Counter-example to regularity in time}{Cert}
540: Based on the assumptions we have so far, the function $V(x,t)$ will \textit{not} be
541: continuous in time in general.  To show this, we will assume for the sake of this
542: example that $u_{I}$ is a continuous function.  We will also assume in this section 
543: that the set $D$ is connected.  The trouble appears in certain cases where the set
544: \begin{equation}
545:    W := D \cup \{ x \in \R^n : u_{I}(x) = 1 \}
546: \label{eq:SetTroub}
547: \end{equation}
548: is disconnected.  If $K$ is a component of the set $\{ u_{I} = 1 \}$ and
549: $K$ is a positive distance away from $D,$ then the $u^{(m)}$ and therefore the
550: $u^{(\infty)}$ should not evolve on this set of $x$ until the free boundary
551: comes into contact with it.  Essentially, the ``patch'' $K$ will not ``see'' 
552: the input of the slot until the component of the set $\{ u^{(m)} \geq 1 \}$ 
553: which surrounds $D$ connects to it.
554: \begin{remark}[$V$ is not always continuous]  \label{Vnc}
555: To produce a situation where $V$ must be discontinuous in time,
556: simply consider the radially symmetric situation where $D = B_1$ and $u_{I}(r)$
557: is taken such that it is identically one on the set $3 \leq r \leq 5,$ but smaller
558: outside of it.  When what we want to call the free boundary reaches $r = 3,$ then 
559: it will instantaneously jump to $r = 5,$ and this leads to an immediate jump in 
560: the height of $V.$
561: \end{remark}
562:  
563: \
564:  
565: \psfig{file=Vjump.eps}
566:  
567: \
568:  
569: \noindent
570: With these pictures in mind we make the following definitions:
571: \begin{definition}[Defining the free boundary]  \label{DFB}
572: We set
573: \begin{equation}
574: \begin{array}{l}
575:  t_{m}(x) := \inf \{ t \; : \; m(u^{(m)}(x,t) - 1)_{+} > 0 \} \\
576:  t_{\infty}(x) := \inf \{ t \; : \; V(x,t) > 0 \}
577: \end{array}
578: \label{eq:tdefs}
579: \end{equation}
580: and
581: \begin{equation}
582: \begin{array}{l}
583:  \tau_{m}(x) := \inf \{ t \; : \; u^{(m)}(x,t) \geq 1 \} \\
584:  \tau_{\infty}(x) := \inf \{ t \; : \; u^{(\infty)}(x,t) = 1 \} \;.
585: \end{array}
586: \label{eq:taudefs}
587: \end{equation}
588: We can also now define the diffusive region at time $t$ for the m-approximating
589: problem and the limiting Hele-Shaw problem to be respectively,
590: \begin{equation}
591: \begin{array}{rl}
592:   A^{(m)}(t) \!\! &:= \{ x \in D^c : \; m(u^{(m)}(x, \; t) - 1)_{+} > 0 \} \\
593:   A(t) \!\! &:= \{ x \in D^c \; : \; V(x, \; t) > 0 \} \;.
594: \end{array}
595: \label{eq:DiffReg}
596: \end{equation}
597: Now we simply define the free boundary at time $t$ to be 
598: $FB^{(m)}(t) := \partial A^{(m)}(t) \setminus \partial D,$ or 
599: $FB(t) := \partial A(t) \setminus \partial D,$ according to which problem we 
600: are considering.  (Here ``$\partial K$'' denotes the topological boundary of $K.$)
601: \end{definition}
602: \noindent
603: Observe that because of the radially symmetric example given above where $V$
604: is discontinuous in time, it is also clear that the free boundary cannot be
605: expected to vary continuously in time.
606: \begin{remark}[``Diffusive'' versus ``Wet'']    \label{DVW}
607: The reader may observe that ``Wet'' may be a more appropriate name for what we
608: are calling the diffusive region in the Hele-Shaw problem.  On the other hand,
609: ``wet'' might more properly describe any region where $u^{(\infty)}(x,t) > 0,$
610: so we will stick to what is already appropriate in the approximating problems.
611: \end{remark}
612: 
613: By considering the figure above, it is clear that in general
614: $t_{m}(x)$ does not have to equal $\tau_{m}(x),$ and similarly with
615: $t_{\infty}(x)$ and $\tau_{\infty}(x).$  In the figure, they would
616: differ on the set $K.$  For $x \in K$ we have 
617: $\tau_{m}(x) = \tau_{\infty}(x) = 0,$ while $t_{m}(x)$ would be the
618: positive time when the moving part of the boundary of 
619: $\{u^{(m)} \geq 1 \}$ crosses $K,$ and $t_{\infty}(x)$ would be the
620: positive time when the moving part of the boundary of
621: $\{u^{(\infty)} = 1 \}$ crosses $K.$
622: 
623: %-----------------------------------------------------------------------
624: 
625: \newsec{Monotonicity, inclusions, and consequences}{mic}
626: We start with a lemma which is a simple consequence of Lemma 4.2 in \cite{K1}.
627: \begin{lemma}[Monotonicity of $u^{(m)}$]   \label{Monoumint}
628: For each $m > 0$ and any $x \in D^c$ we have $u^{(m)}(x,t)$ is an increasing
629: function of time.
630: \end{lemma}
631: \begin{pf}
632: If $u^{(m)}(x,t) < 1,$ then this follows immediately from Lemma 4.2 of
633: \cite{K1}.  As a consequence of this fact, we can already say that the 
634: $A^{(m)}(t)$ are nested nondecreasing sets.  Now we consider the set 
635: $S := \{ (x,t) \; : \; u^{(m)}(x,t) > 1 \}.$  By observing that $u^{(m)}$
636: satisfies the heat equation within $S$ and invoking standard parabolic 
637: regularity theory, we can conclude that $u^{(m)}(x,t)$ will assume the
638: boundary values $1 + \frac{p(x)}{m}$ on $\partial D$ continuously, except 
639: possibly at the initial corner.  By Corollary 1.3  of \cite{AK} the function 
640: $(u^{(m)} - 1)_{+}$ is continuous within $D^c,$ so we can extend $u^{(m)}$ 
641: to be $1$ on the rest of the parabolic boundary of $S$ in a continuous 
642: fashion, except again at $t = 0$ where this set meets $\partial D.$  Now 
643: we consider the function 
644: \begin{equation}
645:     U(x,t) := u^{(m)}(x, \; t + \epsilon) - u^{(m)}(x, \; t) \;.
646: \label{eq:BUdef}
647: \end{equation}
648: Within $S,$ this function satisfies the heat equation.  On 
649: $\partial S \cap \partial D,$ we have $U(x,t) \equiv 0.$  (If we allow $p(x)$
650: to also depend on time, but we require it to be nondecreasing in time, then
651: we have $U(x,t) \geq 0$ here which is also fine.  It is not at all clear, 
652: however, how to deal with the most general case where $p(x,t)$ is allowed to
653: decrease in time.)  On the rest of the parabolic boundary of $S$ we have 
654: $U(x,t) = u^{(m)}(x, \; t + \epsilon) - 1 \geq 0.$  In the corner where
655: $U(x,t)$ is discontinuous, it stays bounded between zero and $1 + \frac{M}{m},$
656: and hence by the weak maximum principle, $U(x,t)$ is nonnegative in all of $D.$
657: Hence
658: \begin{equation}
659:     \frac{u^{(m)}(x, \; t + \epsilon) - u^{(m)}(x, \; t)}{\epsilon}
660: \label{eq:diffquotint}
661: \end{equation}
662: is nonnegative, and the result follows by taking $\epsilon \downarrow 0.$
663: \newline Q.E.D. \newline
664: \end{pf}
665: 
666: \begin{lemma}[Monotonicity of $u^{(\infty)}$]  \label{MonoUInf}
667: $u^{(\infty)}$ is monotone increasing in $t$ for $\mathcal{L}^n$ a.e. $x.$
668: \end{lemma}
669: \begin{pf}
670: By the last lemma the $u^{(m)}$ are monotone increasing in $t.$  Now to come to 
671: a contradiction, suppose $h > 0$ is a number and $\varphi(x,t)$ is a bounded 
672: nonnegative function which is compactly supported in our domain and which 
673: satisfies
674:   $$\int \int \varphi(x, \; t) u^{(\infty)}(x, \; t) \; dx \; dt >
675:     \int \int \varphi(x, \; t - h) u^{(\infty)}(x, \; t) \; dx \; dt \;.$$
676: Such a construction is possible if the conclusions of our theorem are not satisfied.
677: Then $\psi(x,t) := \varphi(x, \; t) - \varphi(x, \; t - h)$ is an admissible test
678: function which satisfies
679:   $$\int \int \psi(x, \; t) u^{(\infty)}(x, \; t) \; dx \; dt > 0 \;, \ \ \ \text{and} \ \ \
680:     \int \int \psi(x, \; t) u^{(m)}(x, \; t) \; dx \; dt \leq 0$$
681: which contradicts the weak-$* \ L^{\infty}$ convergence of the 
682: $u^{(m)}$ to $u^{(\infty)}.$
683: \newline Q.E.D. \newline
684: \end{pf}
685: Now without data on the slot which ``competes'' with the increasing diffusivity (or if
686: the functions solve the equation on all of $\R^n$), it is immediate by rescaling that
687: the diffusive regions (the $A^{(m)}(t)$) must be nondecreasing with $m.$  In our case, however, 
688: we need to produce an appropriate barrier.
689: 
690: \begin{theorem}[Temperature increases with $m.$]  \label{drim}
691: If $m < k,$ then
692: \begin{equation}
693: m[u^{(m)} - 1]_{+} \leq k[u^{(k)} - 1]_{+} \;.
694: \label{eq:tempinc}
695: \end{equation}
696: As a trivial consequence we can say that the $A^{(m)}(t)$ are nested:  
697: $m < k$ implies $A^{(m)}(t) \subset A^{(k)}(t) \;.$
698: \end{theorem}
699: \begin{pf}
700: We make the following definition:
701: \begin{equation}
702:    v^{(k)}(x, \; t) \; := \; \left\{
703:         \begin{array}{ll}
704:            u^{(m)}(x, \; t)  \ \ \ &\text{if} \ u^{(m)}(x, \; t) < 1 \\
705:            \ \\
706:            1 + \frac{m}{k}(u^{(m)}(x, \; t) - 1) \ \ \ &\text{if} \ u^{(m)}(x, \; t) \geq 1 \\
707:         \end{array}          \right.
708: \label{eq:sdvkdef}
709: \end{equation}
710: and claim that $v^{(k)}(x, \; t)$ is a subsolution of the k-approximating problem.
711: Indeed, this follows very quickly from the following two easily verifiable statements:
712: \begin{enumerate}
713:    \item The set where $v^{(k)} > 1$ is identical to the set where $u^{(m)} > 1.$
714:    \item $k(v^{(k)} - 1)_{+} \equiv m(u^{(m)} - 1)_{+}.$
715: \end{enumerate}
716: (Only the definition of $v^{(k)}$ is needed to verify these statements!)
717: By the first observation the free boundaries of $v^{(k)}$ and $u^{(m)}$ are identical 
718: in time and space, so the speed of these boundaries at every point is also identical.  
719: On the other hand, the second observation quickly leads to the conclusion that the function
720: $v^{(k)}$ satisfies the free boundary condition and the boundary value condition on the 
721: slot for the k-approximating problem exactly.  Finally, we simply compute:
722: \begin{alignat*}{1}
723:    v^{(k)}_{t}(x, \; t) & \leq u^{(m)}_{t}(x, \; t) \\
724:                         & = m \Delta (u^{(m)}(x, \; t) - 1)_{+} \\
725:                         & = k \Delta (v^{(k)}(x, \; t) - 1)_{+} 
726: \end{alignat*}
727: which shows that $v^{(k)}$ is a local subsolution to the k-approximating problem, and
728: therefore $v^{(k)}(x, \; t) \leq u^{(k)}(x, \; t).$  (In the first inequality above we 
729: have used Lemma\refthm{Monoumint}\!.)  We now have the simple consequence
730: \begin{equation}
731:    k[u^{(k)} - 1]_{+} \geq k[v^{(k)} - 1]_{+} \equiv m[u^{(m)} - 1]_{+} \;,
732: \label{eq:HGtm}
733: \end{equation}
734: which is the monotonicity we require.
735: \newline Q.E.D. \newline
736: \end{pf}
737: \begin{corollary}[Pointwise convergence of the temperature]  \label{pctemp}
738: The sequence of functions $\{m[u^{(m)} - 1]_{+}\}$ converges pointwise almost everywhere
739: to $V.$  In particular, the limiting function $V$ is unique.  (No subsequence is ever
740: needed.)
741: \end{corollary}
742: \begin{pf}
743: By the last theorem combined with the estimates in place already, at each point
744: we have a bounded increasing sequence.
745: \newline Q.E.D. \newline
746: \end{pf}
747: \begin{corollary}[Representation of $u^{(\infty)}$]  \label{uirep}
748: There is an increasing set-valued function of $t$ which we call $Q(t)$ such that
749: $u^{(\infty)}(x, \; t)$ admits the representation for almost every $(x, \; t)$:
750: \begin{equation}
751:    u^{(\infty)}(x,t) = \chisub{Q(t)}(x) + u_I(x) \chisub{Q(t)^{c}}(x) \;.
752: \label{eq:crurep}
753: \end{equation}
754: Furthermore, $u^{(\infty)}(x, \; t)$ is the pointwise limit of the functions
755: $u^{(m)}(x, \; t)$ almost everywhere, and $Q(t)$ can be chosen to be equal to 
756: the set $\{ x \in \R^{n} \; : \; \tau_{\infty}(x) < t \}$  ($Q(t)$ is increasing, means in 
757: the sense of set inclusion.)
758: \end{corollary}
759: \begin{pf}
760: By Lemma\refthm{MonoUInf}\!\!, we know that $u^{(\infty)}(x, \; t)$ is an increasing 
761: function of $t$ and so by using Lemma\refthm{MonoUInf}along with Equation\refeqn{OneBdd}we
762: can conclude that $u_{I}(x) \leq u^{(\infty)}(x, \; t) \leq 1.$  So, to prove 
763: Equation\refeqn{crurep}it suffices to show that $u^{(\infty)}(x, \; \cdot)$ does 
764: not attain values strictly between $u_{I}(x)$ and $1$ for a.e. $x.$  Indeed, by 
765: combining the previous theorem with Remark\refthm{WooHooI}we see that the limit
766: \begin{equation}
767:    \lim_{m \rightarrow \infty} u^{(m)}(x_0, \; t_0)
768: \label{eq:ptwiselim}
769: \end{equation}
770: exists for almost every $(x_0, t_0) \in \R^{n + 1}$ and is either equal to $1$ or to
771: $u_{I}(x_0).$  At that point it is a simple exercise to show that this pointwise limit
772: coincides with the weak-$* \ L^{\infty}$ limit almost everywhere.
773: 
774: In terms of the ``choice'' of $Q(t),$ it is clear that the only $x$ where there can 
775: be a choice is on the (possibly empty) set $\{ x \in \R^{n} \; : \; u_{I}(x) = 1 \}.$  
776: Examination of the definition of $Q(t)$ combined with the monotonicity of
777: $u^{(\infty)}(x, \; t)$ in $t$ makes it clear that 
778: $Q(t) := \{ x \in \R^{n} \; : \; \tau_{\infty}(x) < t \}$ will suffice.  (Note that
779: an equally good choice for $Q(t)$ would be to take 
780: $Q(t) := \{ x \in \R^{n} \; : \; t_{\infty}(x) < t \}.$)
781: \newline Q.E.D. \newline
782: \end{pf}
783: %\begin{remark}[Choice of $S(t)$]  \label{CST}
784: %In choosing 
785: %\end{remark}
786: \begin{remark}[Uniqueness of limits of subsequences]  \label{Unique}
787: Although we had originally needed subsequences to be sure that our $u^{(m)}$ would
788: converge in weak-$* \ L^{\infty},$ the last result makes it clear that $u^{(m)}$ will
789: converge both pointwise and weak-$* \ L^{\infty}$ to $u^{(\infty)}$ without the need
790: to extract a subsequence.  
791: %Next by using this observation along with the fact that
792: %$(u^{(\infty)}, \; V)$ satisfies Equation\refeqn{whsprob}(see Lemma\refthm{limlimp}\!\!) 
793: %we can conclude that subsequences are also unnecessary when discussing the 
794: %weak-$* \ L^{\infty}$ convergence of the $m(u^{(m)} - 1)_{+}$ to $V.$
795: \end{remark}
796: \begin{corollary}[$V$ is harmonic within the diffusive region]  \label{DVEZ}
797: The spatial Laplacian of $V$ is zero for $x$ in the interior of $A(t).$  As a
798: consequence, if $\partial D \in C^{k, \alpha}$ where $k \geq 2,$ then $V$ attaches 
799: to the slot $D$ in a $C^{k, \alpha}$ fashion.
800: \end{corollary}
801: \noindent
802: Although the function $V$ is discontinuous in general, we do get regularity in space.
803: \begin{theorem}[Spatial continuity of $V.$]
804: $V(\; \cdot \;, \; t)$ is continuous for almost every time, $t.$
805: \end{theorem}
806: \begin{pf}
807: We will prove that $V(\; \cdot \;, \; t)$ is continuous by proving that it is both upper
808: and lower semicontinuous.  Lower semicontinuity follows immediately from the fact that
809: the functions $m[u^{(m)}(\; \cdot \;, \; t) - 1]_{+}$ are continuous and are increasing as 
810: functions of $m.$  To get upper semicontinuity we have a little bit more work.
811: 
812: By Lemma\refthm{MonoUInf}\!\!, $d\nu = u^{(\infty)}_t$ is a nonnegative measure.  Our Radon
813: measure can be ``sliced'' into Radon measures of one dimension less for a.e. t. (See
814: \cite{M} p. 139 - 142, and Equation (10.3) on p. 140 in particular.)  We will call
815: these slices $d\nu_t.$  Since for a.e. t we now have $\Delta_{x} V(x,t) = d\nu_{t},$
816: we know $V( \cdot, t)$ is subharmonic, and therefore upper semicontinuous for those
817: values of $t.$
818: \newline Q.E.D. \newline
819: \end{pf}
820: 
821: %-----------------------------------------------------------------------
822: 
823: \newsec{Nontriviality and nondegeneracy}{NN}
824: Due to the competition between the decreasing data on $\partial D$ and the increasing
825: diffusivity of our m-approximating problems, we need now to rule out two trivial cases.
826: The possibilities that we need to exclude are:
827: \begin{enumerate}
828:    \item The possibility that the free boundary moves to infinity as soon as $t > 0.$
829:    \item The possibility that the free boundary never moves.
830: \end{enumerate}
831: Conceptually, our elimination of these cases is trivial.  We simply produce a family 
832: of supersolutions (and then subsolutions) to our m-approximating problems 
833: (Equation\refeqn{MesaHS}\!) whose free boundaries move in a suitable manner, 
834: independent of $m,$ and then we make use of the comparison principles that our
835: equations enjoy (see \cite{K1}).
836: 
837: \begin{theorem}[Boundedness of the free boundary]   \label{BFB}
838: If $u_{I}$ is compactly supported, then so is $u^{(\infty)}(x,t_0)$ for any $t_0 > 0.$
839: In fact, no matter what the initial data, there is always a supersolution whose free 
840: boundary has constant speed.
841: \end{theorem}
842: \noindent
843: Note that we need the compact support of the initial data for this result to hold.
844: 
845: \noindent
846: \begin{pf}
847: By using the boundedness of $D,$ and by translating and rescaling it will suffice to 
848: produce a supersolution to the problem where 
849: $D^c = B_1,$ \; $u_I = \chisub{ \{ B_2 \setminus B_1 \} },$ and $p(x) \equiv k.$
850: We claim that the function
851: \begin{equation}
852:    v^{(m)}(r,t) := \left( \frac{k(2 - r + \ell t)}{m(1 + \ell t)} + 1 \right)
853:                    \chisub{\displaystyle{(1, \; 2 + \ell t]}}(r)
854: \label{eq:vm}
855: \end{equation}
856: will suffice if $\ell$ is sufficiently large.  Note that this function is linear in the
857: radial variable, and that the free boundary moves with speed $\ell$ for all time.  The
858: following computations are elementary, and they prove our claim that $v^{(m)}$ is a
859: local supersolution:
860: \begin{equation}
861:   m \Delta (v^{(m)} - 1)_{+} = \frac{k(1 - n)}{r(1 + \ell t)} \leq 0 
862:                           \leq \frac{k \ell (r - 1)}{m(1 + \ell t)^2} = v^{(m)}_{t} \;.
863: \label{eq:locsuper}
864: \end{equation}
865: If we are on the free boundary, so that $r = 2 + \ell t,$ then
866: \begin{equation}
867:   \left| \frac{\partial}{\partial r} \left[ m(v^{(m)} - 1)_{+} \right] \right| = 
868:   \frac{k}{(1 + \ell t)} \leq \ell = \text{speed of the free boundary.}
869: \label{eq:fbsuper}
870: \end{equation}
871: \newline Q.E.D. \newline
872: \end{pf}
873: \begin{theorem}[Nontrivial motion of the free boundary]  \label{NMFB}
874: The boundary of the diffusive region does not remain stationary, and for any 
875: $R > 0,$ there exists a time $t_0$ such that for all $t > t_0,$ we have
876: $B_R \subset A(t_0).$
877: \end{theorem}
878: \begin{remark}   \label{EasWay}
879: If we only want to show that the free boundary does not remain stationary, then we can
880: use the same subsolutions as in the proof of Theorem\refthm{drim}\!.  On the other hand,
881: the subsolutions we use here are also needed in the proof of the next theorem.
882: \end{remark}
883: \begin{pf}
884: Using the fact that $D$ contains an open set, and by translating and rescaling it will 
885: suffice to produce a subsolution to the problem where (for any $\alpha \geq 0,$)
886: $D^c = B_{1 + \alpha},$ \; $u_I = \chisub{ \{ B_{2 + \alpha} \setminus B_{1 + \alpha} \} },$ 
887: and $p(x) \equiv k,$ and such that the free boundary of this subsolution moves a fixed
888: distance (independent of $\alpha$) in a finite time $T(\alpha).$
889: 
890: We will create our subsolution in a couple of steps.
891: First, let $w^{(m)}(r,t) = w^{(m)}_{1}(r,t) + w^{(m)}_{2}(r,t)$ where
892: $w^{(m)}_{1}(r,t)$ solves for each fixed $t$
893: \begin{equation}
894:   \begin{array}{rll}
895:     \Delta_x w^{(m)}_{1}(r,t) \! 
896:     &= 0 \ \ \ 
897:     & \text{in} \ B_{2 + \alpha + \ell t} \setminus B_{1 + \alpha} \\
898:     \ \\
899:     w^{(m)}_{1}(r,t) \! &= \displaystyle{1 + \frac{k}{m}} \ \ \ 
900:     & \text{on} \ \partial B_{1 + \alpha} \\
901:     \ \\
902:     w^{(m)}_{1}(r,t) \! &= 1 \ \ \ & \text{on} \ \partial B_{2 + \alpha + \ell t} \\
903:   \end{array}
904: \label{eq:wm1}
905: \end{equation}
906: and
907: $w^{(m)}_{2}(r,t)$ solves (again for each fixed $t$)
908: \begin{equation}
909:   \begin{array}{rll}
910:     \Delta_x w^{(m)}_{2}(r,t) \! 
911:     &= \displaystyle{\frac{\epsilon}{m}} \ \ \ 
912:     & \text{in} \ B_{2 + \alpha + \ell t} \setminus B_{1 + \alpha} \\
913:     \ \\
914:     w^{(m)}_{2}(r,t) \! 
915:     &= 0 \ \ \ 
916:     & \text{on} \ \partial \{ B_{2 + \alpha + \ell t} \setminus B_{1 + \alpha} \} \;.
917:   \end{array}
918: \label{eq:wm2}
919: \end{equation}
920: Since the free boundary will be given by $r = 2 + \alpha + \ell t,$ and we only need it
921: to move a fixed distance, we will also assume that $\ell t \leq 1.$  The functions 
922: $w^{(m)}_{1}$ and $w^{(m)}_{2}$ can be given explicitly, and their relevant properties
923: are given in the appendix.
924: 
925: By invoking Corollary\refthm{BOB}from the appendix, we have
926: \begin{equation}
927:   -\infty < -\tilde{C}_{1}(n)k \leq
928:   \frac{\partial}{\partial r} m w^{(m)}_{1}(2 + \alpha + \ell t, \; t) \leq
929:   -\tilde{C}_{2}(n)k < 0
930: \label{eq:Dmw1}
931: \end{equation}
932: where we stress that the constants are independent of $\alpha$ and $\ell t.$  
933: (See the appendix for a few more details, and remember that $0 \leq \ell t \leq 1.$)
934: By the same corollary we can conclude that
935: \begin{equation}
936:   0 < \tilde{C}_{3}(n) \epsilon \leq
937:   \frac{\partial}{\partial r} m w^{(m)}_{2}(2 + \alpha + \ell t, \; t) \leq
938:   \tilde{C}_{4}(n) \epsilon < \infty 
939: \label{eq:Dmw2}
940: \end{equation}
941: again with constants independent of $\alpha$ and $\ell t.$
942: Now we take $\epsilon > 0$ sufficiently small to ensure that
943: \begin{equation}
944:   -\infty < -C_1 k \leq
945:   \frac{\partial}{\partial r} \left[ m w^{(m)}(2 + \alpha + \ell t, \; t) \right] \leq
946:   -C_2 k < 0 \;.
947: \label{eq:Dmw}
948: \end{equation}
949: Now by taking $\ell \leq C_2 k/2$ we can be sure that our function is a subsolution
950: along the free boundary.  We note that 
951: $\Delta m(w^{(m)}(r, \; t) - 1)_{+} = \epsilon > 0$ in the region where
952: $w^{(m)}(r, \; t) > 1.$  Since
953: \begin{equation}
954:    \lim_{m \rightarrow \infty} w^{(m)}_t(r, \; t) = 0 \;,
955: \label{eq:wtmLIM}
956: \end{equation}
957: Once $m$ is sufficiently large, we automatically have
958: \begin{equation}
959:    \Delta m (w^{(m)}(r, \; t) - 1)_{+} \geq w^{(m)}_t(r, \; t) \;.
960: \label{eq:wmlocsub}
961: \end{equation}
962: Since the free boundary of our subsolution moves with speed $\ell > 0,$ we are done.
963: \newline Q.E.D. \newline
964: \end{pf}
965: 
966: The following theorem shows the instantaneous detachment of the free boundary
967: from the slot, $\partial D,$ even if $u_{I}(x) \leq \lambda < 1$ in all of $D^{c}.$
968: \begin{theorem}[Instantaneous formation of the diffusive region]  \label{IFDR}
969: If $t_0 > 0,$ The set $A(t_0)$ contains an open neighborhood of $\partial D.$
970: \end{theorem}
971: \begin{pf}
972: Because we have assumed that $\partial D \in C^{2, \alpha} ,$ every point on 
973: $\partial D$ can be touched from within with a tangent ball, and then we can use
974: the same subsolutions of the previous theorem to force instantaneous movement of
975: the free boundary.
976: \newline Q.E.D. \newline
977: \end{pf}
978: 
979: %-----------------------------------------------------------------------
980: 
981: \newsec{Spatial regularity results}{SRFB}
982: In this section we will derive spatial regularity for both the function
983: $V(x,t)$ and for the free boundary.  We apply the Baiocchi 
984: transformation to $V(x,t)$ and define:
985: \begin{equation}
986:      W(x,t) := \int_{0}^{t} V(x, \; s) \; ds \;.
987: \label{eq:WBaio}
988: \end{equation}
989: Observe that by Lemma\refthm{MonoUInf}(which shows that the diffusive regions
990: are increasing in time) and by the positivity of $V(x,t)$ in the diffusive
991: region, it is clear that the set $\{ W > 0 \}$ is identical to the set
992: $\{ V > 0 \}.$  Now to find the regularity of 
993: $\partial \{ x \in \R^{n} \; : \; W(x, T) > 0 \} $ we will show that 
994: $W(\cdot,t)$ belongs to $H^1_{loc}(D^{c})$ for almost every $t,$ and 
995: then that $W(x, t)$ is a weak solution of the following obstacle 
996: problem in almost every time slice $t = T$
997: \begin{equation}
998:    0 \leq W(x, T), \ \ \ \ \ 
999:    \Delta_x W(x, T) = \chisub{ \{ W(x, T) > 0 \} }(x)(1 - u_{I}(x)) \;.
1000: \label{eq:WobProb}
1001: \end{equation}
1002: After that we will be able to invoke regularity results for the obstacle problem 
1003: due to Caffarelli, Kinderlehrer, Nirenberg, and Blank.
1004: 
1005: For simplicity, we let $\alpha_{m}(s) := m(s - 1)_{+}.$  We start by stating some
1006: simple trace results.  Basically, we need to adapt Equation\refeqn{wmap}to some
1007: situations with slightly different test functions.
1008: 
1009: \begin{lemma}[First Trace result] \label{TrRes}
1010: If $\psi(x) \in C^{\infty}$ is supported in the interior of $D^{c},$ then the following formula
1011: holds for a.e. $T$:
1012: \begin{equation}
1013: \begin{array}{l}
1014:  \displaystyle{\int_{D^c} \int_{0}^{T}
1015:    [\Delta_x \psi(x)] \; \alpha_{m}(u^{(m)}(x,t)) \; dt \; dx} \\
1016:    \ \\
1017:    = \displaystyle{ \int_{D^c} \psi(x) [u^{(m)}(x,T) - u_{I}(x)] \; dx} \;.
1018: \end{array}
1019: \label{eq:wmap2}
1020: \end{equation}
1021: \end{lemma}
1022: \begin{pf}
1023: We make the following definition
1024: \begin{equation}
1025:    \varphi(x,t) := \left\{
1026:        \begin{array}{rl}
1027:           \psi(x) \ \ \ & t \leq T \\
1028:           \ \\
1029:           \Theta(x,t) \ \ \ & t > T \\
1030:        \end{array}
1031:    \right.
1032: \label{eq:nicetest}
1033: \end{equation}
1034: where $\Theta(x,t)$ is chosen to ensure that $\varphi(x,t)$ is a permissible
1035: test function for our m-approximating problem.  In particular, we need
1036: $\varphi(x,t) \in C^{\infty}$ and we need it to converge to zero as
1037: $t \rightarrow \infty.$  Neither requirement poses any difficulty.
1038: 
1039: By using the trace result of \cite{AK} (see Theorem 1.1 of \cite{AK}) our
1040: functions $u^{(m)}(x,t)$ solve our m-approximating problem starting at time $T$
1041: with initial data $u^{(m)}(x,T)$ for almost every $T,$ and so (for those $T$)
1042: we can use $\Theta(x,t)$ as the test function in Equation\refeqn{wmap}to obtain:
1043: \begin{equation}
1044: \begin{array}{rl}
1045:  \ & \displaystyle{\int_{D^c} \int_{T}^{\infty} \Theta_t(x,t) u^{(m)}(x,t) \; dt \; dx} \\
1046:    \ \\
1047:  + & \displaystyle{\int_{D^c} \int_{T}^{\infty}
1048:    \Delta_x \Theta(x,t) \alpha_{m}(u^{(m)}(x,t)) \; dt \; dx} \\
1049:    \ \\
1050:  = - & \displaystyle{\int_{D^c} \psi(x) u^{(m)}(x,T) \; dx} \;.
1051: \end{array}
1052: \label{eq:wmapT}
1053: \end{equation}
1054: (Note that $\Theta(x,T) = \psi(x)$ since we have required that $\varphi(x,t)$ be
1055: smooth.)  Now by subtracting this equation from what we have when we plug in the
1056: function $\varphi(x,t)$ defined in Equation\refeqn{nicetest}into
1057: Equation\refeqn{wmap}we get Equation\refeqn{wmap2}immediately.
1058: \newline Q.E.D. \newline
1059: \end{pf}
1060: 
1061: The proof of the following result is almost identical to the proof of the trace result
1062: above, so we omit it.
1063: \begin{lemma}[Second Trace result] \label{TrRes2}
1064: For a.e. $t_0, t_1$ such that $0 \leq t_0 < t_1 < \infty,$ and for 
1065: $\varphi \in C^{\infty}\left(\R^n \times [t_0,t_1]\right),$ which satisfies
1066: $\varphi \equiv 0$ on an open set containing $D \times [t_0,t_1],$ we have
1067: \begin{equation}
1068: \begin{array}{l}
1069:  \displaystyle{\int_{D^c} \int_{t_0}^{t_1} \varphi_t(x,t) u^{(m)}(x,t) \; dt \; dx +
1070:                  \int_{D^c} \int_{t_0}^{t_1}
1071:    \Delta_x \varphi(x,t) \alpha_{m}(u^{(m)}(x,t)) \; dt \; dx} \\
1072:    \ \\
1073:    = \displaystyle{\int_{D^c} \left( \left. \rule[-.02 in]{.0 in}{.22 in}
1074:                \left[  \varphi(x,s) u^{(m)}(x,s) \right] \right|_{s = t_0}^{s = t_1} \right) \; dx} \;.
1075: \end{array}
1076: \label{eq:wmap3}
1077: \end{equation}
1078: \end{lemma}
1079: 
1080: Now we state the standard energy estimate for our situation.
1081: \begin{lemma}[Energy estimates for the $u^{(m)}$]   \label{umenergy}
1082: Let $0 < r < R$ and $0 \leq t_0 < t_1.$  Then there is a constant of
1083: the form
1084: \begin{equation}
1085:    C = \frac{C(n)}{(R - r)^2}
1086: \label{eq:rtdep}
1087: \end{equation}
1088: such that the following energy estimate holds:
1089: \begin{equation}
1090:     \int_{t_0}^{t_1} \int_{B_{r}(x_0)} 
1091:     | \nabla \alpha_{m}(u^{(m)}) |^2 \; dx \; dt \leq
1092:     C\int_{t_0}^{t_1} \int_{B_{R}(x_0)} \alpha_{m}(u^{(m)})^2 \; dx \; dt \;.
1093: \label{eq:umspace}
1094: \end{equation}
1095: % and
1096: %\begin{equation}
1097: %    \int_{t_1}^{t_2} \int_{B_{r}(x_0)} 
1098:     %\left| \frac{\partial}{\partial t} \alpha_{m}(u^{(m)}) \right|^2 \; dx \; dt \leq
1099:     %C m \int_{t_0}^{t_2} \int_{B_{R}(x_0)} \alpha_{m}(u^{(m)})^2 \; dx \; dt \;.
1100: %\label{eq:umtime}
1101: %\end{equation}
1102: \end{lemma}
1103: \begin{remark}[Independence of Time]   \label{rmk:IoT}
1104: Notice that the constant $C$ is independent of time and notice that the time
1105: intervals in the integrals in each side of the inequality are identical.
1106: \end{remark}
1107: \begin{pf}
1108: %We choose $\psi(t) \in C^{\infty}([t_0,t_1]),$ and
1109: %\begin{enumerate}
1110:    %\item $\psi \equiv 1$ on $[t_1, +\infty),$ \ $\psi \equiv 0$ on $(-\infty, t_0],$
1111:    %\item $0 \leq \psi \leq 1,$ and
1112:    %\item $0 \leq \psi^{\prime} \leq 4(t_1 - t_0)^{-1} \;.$
1113: %\end{enumerate}
1114: We choose $\eta(x) \in C^{\infty}_{0}(B_R)$ such that
1115: \begin{enumerate}
1116:    \item $\eta \equiv 1$ on $\closure{B_{r}},$
1117:    \item $0 \leq \eta \leq 1,$ and
1118:    \item $|\nabla \eta| \leq 4(R - r)^{-1} \;.$
1119: \end{enumerate}
1120: Now let $\varphi(x,t) := \alpha_{m}(u^{(m)}) \eta(x)^2,$ and apply the last lemma
1121: and Green's identity to obtain:
1122: \begin{equation}
1123: \begin{array}{l}
1124:  \displaystyle{\int_{B_R} \int_{t_0}^{t_1} \varphi_t(x,t) u^{(m)}(x,t) \; dt \; dx -
1125:                \int_{B_R} \left( \left. \rule[-.02 in]{.0 in}{.22 in} 
1126:                \left[  \varphi(x,s) u^{(m)}(x,s) \right] \right|_{s = t_0}^{s = t_1} \right) \; dx} \\
1127:    \ \\
1128:    = \displaystyle{\int_{B_R} \int_{t_0}^{t_1} \nabla_{x} \varphi(x,t) \nabla_{x} \alpha_{m}(u^{(m)}) \; dt \; dx} \;.
1129: \end{array}
1130: \label{eq:prehappy}
1131: \end{equation}
1132: The fact that we can apply Green's identity above can be justified using Lemma 1.2
1133: of \cite{AK}.  By Lemma\refthm{Monoumint} we know that for a.e. $x \in D^c, \ \mu_x(t):=u_t(x,t)$
1134: is a non-negative Radon measure, whence it suffices to have $\phi(t) \in C_0(0, +\infty)$ for the
1135: distributional pairing $(\mu_x(t), \phi(t))$ to be defined, and non-negative if $\phi \geq 0.$
1136: Using Equation\refeqn{wmap}it is easy to see that the function $g(x) := (\mu_x(t), \chisub{[t_0,t_1]}(t))$
1137: is locally integrable in $x.$  In other words, we can integrate the left hand side of
1138: Equation\refeqn{prehappy}by parts in time to give us the following inequality:
1139: \begin{alignat*}{1}
1140:   \ & \ \ \ \int_{B_R} \int_{t_0}^{t_1} \nabla_{x} \left( \alpha_{m}(u^{(m)}) \eta^2(x) \right) 
1141:                                     \nabla_{x} \alpha_{m}(u^{(m)}) \; dt \; dx \\
1142:   \ &= \; - \int_{B_R} \left[ \int_{t_0}^{t_1} \left( \alpha_{m}(u^{(m)}) \eta^2(x) \right) 
1143:                                            d\mu_x(t) \right] \; dx \\
1144:   \ &\leq 0.
1145: \end{alignat*}
1146: This inequality implies
1147: $$\begin{array}{c}
1148:     \displaystyle{
1149: \int_{B_R} \int_{t_0}^{t_1} \eta^2 \left| \nabla_{x} \alpha_{m}(u^{(m)}) \right|^2 \; dt \; dx \leq} \\
1150:     \ \\
1151:     \displaystyle{
1152:   2\int_{B_R} \int_{t_0}^{t_1} \left( \eta \left| \nabla_{x} \alpha_{m}(u^{(m)}) \right| \right)
1153:                               \left( \alpha_{m}(u^{(m)}) \left| \nabla_{x} \eta \right| \right)
1154:                                     \; dt \; dx}
1155:   \end{array}$$
1156: and so Cauchy-Schwarz gives
1157: \begin{equation}
1158: \begin{array}{l}
1159:  \displaystyle{\int_{t_0}^{t_1} \left[ 
1160:     \int_{B_R} \eta(x)^2 \left| \nabla_{x} \alpha_{m}(u^{(m)}) \right|^2 dx \; \right] dt } \\
1161:    \ \\
1162:    \leq \displaystyle{4\int_{t_0}^{t_1} \left[
1163:     \int_{B_R} \left| \nabla_{x} \eta(x) \right|^2 \alpha_{m}(u^{(m)})^2 dx \; \right] dt } \;.
1164: \end{array}
1165: \label{eq:extrahappy}
1166: \end{equation}
1167: \newline Q.E.D. \newline
1168: \end{pf}
1169: The aforementioned independence of time of the constant $C$ leads immediately to the following
1170: theorem:
1171: \begin{theorem}[Time-slice energy estimates] \label{T-seeum}
1172: For almost every time $t$ we have
1173: \begin{equation}
1174:     \int_{B_{r}(x_0)} | \nabla \alpha_{m}(u^{(m)}(x,t)) |^2 \; dx \leq
1175:     C\int_{B_{R}(x_0)} \alpha_{m}(u^{(m)}(x,t))^2 \; dx \;,
1176: \label{eq:umspace2}
1177: \end{equation}
1178: and also
1179: \begin{equation}
1180:     \int_{B_{r}(x_0)} | \nabla V(x,t) |^2 \; dx \leq
1181:     C\int_{B_{R}(x_0)} V(x,t)^2 \; dx \;,
1182: \label{eq:VspaceCacc}
1183: \end{equation}
1184: where once again, the constant $C$ has the form given in Equation\refeqn{rtdep}\!.
1185: \end{theorem}
1186: \begin{pf}
1187: Equation\refeqn{umspace2}is already known from the previous lemma, and Equation\refeqn{VspaceCacc}follows
1188: from Equation\refeqn{umspace2}by using the $L^2$ convergence of $\alpha_{m}(u^{(m)}(x,t))$ to
1189: $V(x,t)$ and by using the lower semicontinuity of the Dirichlet integral.
1190: \newline Q.E.D. \newline
1191: \end{pf}
1192: \begin{corollary}[Energy estimates for $V(\cdot,t)$]  \label{VLip}
1193: For almost every $t, \ V(\cdot,t) \in H^{1}_{loc}(D^c).$
1194: \end{corollary}
1195: %The proof of this lemma follows along the lines of the proofs for the case
1196: %$m = 1$ in \cite{AK} and \cite{KM}.
1197: % Marianne's new stuff
1198: \begin{theorem}[Energy estimates for $W(\cdot,t)$]  \label{WH1inSpace}
1199: For all $t > 0, \ W(\cdot,t) \in H^{1}_{loc}(D^c).$
1200: \end{theorem}
1201: \begin{pf}
1202: Let $K$ be a compact subset of $D^{c}.$  Since $W(\cdot,t) \leq Mt,$ we have
1203: $W(\cdot,t) \in L^{\infty}(K) \subset L^{2}(K).$  It remains to show that
1204: $\frac{\partial W}{\partial x_i}(\cdot,t) \in L^{2}(K).$
1205: 
1206: Let $\varphi \in C_{0}^{1}(K)$  
1207: %Let $\mu_{\delta}(s) \in C_{0}^{\infty}(0,t)$
1208: %satisfy $0 \leq \mu_{\delta}(s) \leq 1,$ and $\mu_{\delta}(s) \equiv 1$ on
1209: %$[\delta, t - \delta].$
1210: and estimate.
1211:   \begin{alignat*}{1}
1212:   \left( \frac{\partial W}{\partial x_i}, \; \varphi \right) \;
1213:   &= \; - \int_{K} \; W(x,t) \frac{\partial \varphi}{\partial x_i}(x) \; dx \\
1214:   &= \; - \int_{K} \int_{0}^{t} V(x, \; s) \frac{\partial \varphi}{\partial x_i}(x) \; ds \; dx \\
1215:   &= \; - \int_{K} \int_{0}^{t} \lim_{m \rightarrow +\infty} \alpha_{m}(u^{(m)}(x, \; s)) \;
1216:           \frac{\partial \varphi}{\partial x_i}(x) \; ds \; dx \\
1217:   &= \; - \lim_{m \rightarrow +\infty} \int_{K} \int_{0}^{t} \alpha_{m}(u^{(m)}(x, \; s)) \;
1218:           \frac{\partial \varphi}{\partial x_i}(x) \; ds \; dx \\
1219: %  &= \; - \lim_{m \rightarrow +\infty} \lim_{\delta \downarrow 0} \int_{K} \int_{0}^{t}
1220: %          \alpha_{m}(u^{(m)}(x, \; s)) \mu_{\delta}(s) \frac{\partial \varphi}{\partial x_i}(x)
1221: %          \; ds \; dx \\
1222: %  &  \text{Ivan's Question:  Do we really need} \ \mu_{\delta}(s) \ \text{?} \\
1223: %  & \ \\
1224: %  &= \; \lim_{m \rightarrow +\infty} \lim_{\delta \downarrow 0} \int_{K} \int_{0}^{t}
1225: %        \frac{\partial}{\partial x_i} \left( \alpha_{m}(u^{(m)}(x, \; s)) \right) \cdot
1226: %        \mu_{\delta}(s) \varphi(x) \; ds \; dx \\
1227:   &= \; \lim_{m \rightarrow +\infty} \int_{K} \int_{0}^{t} 
1228:         \frac{\partial}{\partial x_i} \left( \alpha_{m}(u^{(m)}(x, \; s)) \right) \varphi(x) \; ds \; dx \\
1229:   &= \; \lim_{m \rightarrow +\infty} \int_{K} \varphi(x) \left[ \int_{0}^{t}
1230:         \frac{\partial}{\partial x_i} \left( \alpha_{m}(u^{(m)}(x, \; s)) \right) \; ds \right] \; dx \\
1231:   \end{alignat*}
1232: where we have used Lebesgue's Dominated Convergence Theorem repeatedly in the computation above.
1233: Now, by using Minkowski's integral inequality and the Cauchy-Schwarz Inequality we have
1234:   \begin{alignat*}{1}
1235:   \ & \left| \int_{K} \varphi(x) \left[ \int_{0}^{t} \frac{\partial}{\partial x_i} 
1236:       \left( \alpha_{m}(u^{(m)}(x, \; s)) \right) \; ds \right] \; dx \right| \; \\
1237:   &\leq \left| \int_{K} \varphi(x)^2 \; dx \right|^{1/2} \left| \int_{K} \left(
1238:         \int_{0}^{t} \frac{\partial}{\partial x_i} 
1239:         \left( \alpha_{m}(u^{(m)}(x, \; s)) \right) \; ds \right)^{2} dx \right|^{1/2} \\
1240:   &\leq ||\varphi||_{L^2(K)} \int_{0}^{t} \left| \int_{K} \left( \frac{\partial}{\partial x_i}
1241:         \left( \alpha_{m}(u^{(m)}(x, \; s)) \right) \right)^2 dx \right|^{1/2} ds \\
1242:   &\leq ||\varphi||_{L^2(K)} \left| \int_{0}^{t} \int_{K} \left( \frac{\partial}{\partial x_i}
1243:         \left( \alpha_{m}(u^{(m)}(x, \; s)) \right) \right)^2 dx \; ds \right|^{1/2} \cdot
1244:         \left[ \int_{0}^{t} ds \right]^{1/2} \\
1245:   &\leq t^{1/2} ||\varphi||_{L^2(K)} ||\nabla_{x} \alpha_{m}(u^{(m)}(x, \; s)) ||_{L^2(K \times (0,t))} \\
1246:   &\leq C t^{1/2} ||\varphi||_{L^2(K)}
1247:   \end{alignat*}
1248: where the last constant is independent of $m$ by Lemma\refthm{umenergy}\!.
1249: \newline Q.E.D. \newline
1250: \end{pf}
1251: % Older stuff
1252: %\begin{theorem}[Energy estimates for $V(\cdot,t)$]  \label{VLip}
1253: %For almost every $t, \ V(\cdot,t) \in H^{1}_{loc}(D^c).$
1254: %\end{theorem}
1255: %\begin{pf}
1256: %Fix $T > 0.$  By Theorem\refthm{BFB}there exists a $\gamma = \gamma(T)$ such
1257: %that $V(x,t) \equiv 0$ outside of $B_{\gamma}$ for all $t \leq T.$  For 
1258: %$x_0 \in D^{c}$ we consider the ball $B_R(x_0) \subset D^{c}$ and assume
1259: %that $0 < t < T.$  Take $t_0$ and $t_1$ with $0 < t_0 < t_1 < T$ and 
1260: %$t = (t_0 + t_1)/2.$ 
1261: %By using the estimate $\alpha_m(u^{(m)}) \leq M$ which follows from
1262: %Equation\refeqn{mMfun}we obtain
1263: %\begin{equation}
1264: %  \int_{t_0}^{t_1} \int_{B_{R/2}(x_0)}
1265: %  | \nabla \alpha_{m}(u^{(m)}) |^2 \; dx \; dt \leq
1266: %  \frac{C(t_1 - t_0) |B_R| M^2}{R^2} \;.
1267: %\label{eq:NotLipEst1}
1268: %\end{equation}
1269: %By taking limits we conclude that
1270: %\begin{equation}
1271: %  | \nabla \alpha_{m}(u^{(m)}) | \leq CM \ \ \text{for} \ a.e. \ (x,t) \;.
1272: %\label{eq:LipEst2}
1273: %\end{equation}
1274: %???So, for almost every $t,$ this equation must hold for almost every $x,$??? 
1275: %Now for $\epsilon > 0,$ let $T_{\epsilon}$ denote the set of $t \in [t_0, t_1]$ such that
1276: %\begin{equation}
1277: %  \int_{B_r(x_0)}
1278: %  | \nabla \alpha_{m}(u^{(m)}) |^2 \; dx >
1279: %  C \epsilon^{-1} \;.
1280: %  \frac{C(t_1 - t_0) |B_R| M^2}{(\min \{ r, t_1 - t_0 \})^2} \;.
1281: %\label{eq:NotLipEst2}
1282: %\end{equation}
1283: %It follows from Equation\refeqn{NotLipEst1}that the measure of $T_{\epsilon}$ is
1284: %bounded by a constant times epsilon.  So, for $t$ in the complement of $T_{\epsilon}$
1285: %if $\varphi \in H^{1}_{0}(K)$ with $K \subset \subset D^{c}$ then we have
1286: %\begin{alignat*}{1}
1287: %  \left| \int_{K} \alpha_m(u^{(m)}) \; \nabla \varphi \; dx \right|
1288: %  &= \left| \int_{K \cap B_{\gamma}} \nabla \alpha_m(u^{(m)}) \; \varphi \; dx \right| \\
1289: %  &\leq C \epsilon^{-1} |B_{\gamma}|^{1/2} ||\varphi||_{L^2(K)} \;.
1290: %\end{alignat*}
1291: %Now we can take $m \rightarrow \infty$ to conclude that $V(\cdot, t) \in H^{1}_{loc}(D^c).$
1292: %and then, by simply invoking Equation\refeqn{LipEst2}again, we conclude that 
1293: %$V(\cdot, t) \in C^{0,1}(D^{c}).$
1294: %\newline Q.E.D. \newline
1295: %\end{pf}
1296: %\begin{corollary}  \label{WH1}
1297: %$W(\cdot,t) \in H^{1}_{loc}(D^c)$ for every $t.$
1298: %\end{corollary}
1299: \begin{remark}   \label{WC1al}
1300: In fact, once we show that $W(\cdot,t)$ is a solution of the obstacle problem, we
1301: will be able to infer from elliptic regularity theory, that $W(\cdot,t)$ is 
1302: $C^{1,\alpha}$ in space for all $\alpha < 1.$
1303: \end{remark}
1304: %\begin{pf}
1305: %Assume $\varphi \in H^{1}_{0}(D^{c})$ and compute
1306: %\begin{alignat*}{1}
1307: %   \left| \int_{D^{c}} W(x,t) \; \nabla \varphi(x) \; dx \right|
1308: %   &= \left| \int_{D^{c}} \left( \int_{0}^{t} V(x,s) \; ds \right) \; \nabla \varphi(x) \; dx \right| \\
1309: %   &= \left| \int_{0}^{t} \int_{D^c} V(x,s) \; \nabla \varphi(x) \; dx \; ds \right| \\
1310: %   &= \left| \int_{0}^{t} \int_{D^c} \nabla V(x,s) \; \varphi(x) \; dx \; ds \right| \\
1311: %   &= \left| \int_{0}^{t} ||\nabla V(x,s)||_{L^2(D^c)} ||\varphi||_{L^2(D^c)} \right| \\
1312: %   &\leq CMt |B_{\gamma}|^{1/2} ||\varphi||_{L^2(D^c)}
1313: %\end{alignat*}
1314: %Q.E.D. \newline
1315: %\end{pf}
1316: In order to derive Equation\refeqn{WobProb}we will need to commute the 
1317: Laplacian with the integral in time.  To accomplish this commutation we turn 
1318: back to the approximating problem where this switch is simpler. 
1319: 
1320: %Corollary 1.3 of \cite{AK} implies that the temperature functions 
1321: %$m[u^{(m)}(x,t) - 1]_{+}$ are continuous.  This fact implies that the functions
1322: %$u^{(m)}$ are ...
1323: %$\varphi(x,t) := \psi(x) \chisub{[0,T]}(t),$ where $\psi$ is supported in the interior
1324: %of $D^{c},$ then $\varphi_t \equiv 0$ and the following formula holds:
1325: By using weak-$* \ L^{\infty}$ convergence of the temperature functions we have
1326: \begin{alignat*}{1}
1327:    \lim_{m \rightarrow \infty} &\int_{D^c} \int_{0}^{T}
1328:     [\Delta_x \psi(x)] \; m \left[ u^{(m)}(x,t) - 1 \right]_{+} \; dt \; dx \\
1329:    = &\int_{D^c} \int_{0}^{T} [\Delta_x \psi(x)] \; V(x, t) \; dt \; dx \\
1330:    = &\int_{D^c} [\Delta_x \psi(x)] W(x,T) \; dx \;.
1331: \end{alignat*}
1332: On the other hand, by using the dominated convergence theorem we conclude that
1333: for almost every $T$
1334: \begin{alignat*}{1}
1335:    \lim_{m \rightarrow \infty} &\int_{D^c} \psi(x) [u^{(m)}(x,T) - u_{I}(x)] \; dx \\
1336:    = &\int_{D^c} \psi(x) [u^{(\infty)}(x,T) - u_{I}(x)] \; dx \\
1337:    = &\int_{D^c} \psi(x) \chisub{ \{ W(x, T) > 0 \} }(x)(1 - u_{I}(x))
1338: \end{alignat*}
1339: 
1340: Since $W \geq 0,$ we can combine the last two computations with the previous
1341: lemma to conclude that $W(\cdot, \; t)$ solves the obstacle problem for a.e. $t > 0.$
1342: Equation\refeqn{WobProb}and this fact allows us to use the technology of 
1343: \cite{Bl} and \cite{C} to infer regularity of the free boundary here as long 
1344: as we satisfy the nondegeneracy condition:
1345: \begin{equation}
1346:    u_I(x) \leq \lambda < 1 \;.
1347: \label{eq:ndsit}
1348: \end{equation}
1349: ($u_I$ satisfying Equation\refeqn{ndsit}will be referred to as \textit{nondegenerate
1350: initial data}.)  In this case we will have the next theorem which we state after one 
1351: simple definition.
1352: \begin{definition}[Minimum diameter]  \label{mindiam}
1353: The minimum diameter of a set $S \subset \R^n$ (denoted ``$m.d.(S)$'') is the
1354: infimum of the distances between parallel hyperplanes enclosing $S.$
1355: \end{definition}
1356: \begin{theorem}[Regularity of the free boundary in space]  \label{RFBS}
1357: Assume $u_{I}$ is continuous, nondegenerate initial data.  Then there is a modulus 
1358: of continuity $\sigma$ which depends on $\lambda, n,$ and the modulus of continuity 
1359: of $u_{I}$ such that at almost any time $t,$ (indeed every time $t_0$ where 
1360: Equation\refeqn{WobProb}is valid) and for any free boundary point $(x_0,t_0)$ 
1361: contained in the interior of $D^{c}$ we have either (with $B_r(x)$ denoting the 
1362: (spatial) ball centered at $x$ with radius $r$)
1363: \begin{equation}
1364:    m.d.(B_r(x_0) \cap A(t)^{c}) \leq r \sigma(r) \ \ \ \ \text{for all} \ r \leq 1 \;,
1365: \label{eq:singularpt}
1366: \end{equation}
1367: (so that the nondiffusive region is ``cusp-like'') or
1368: \begin{equation}
1369:    \lim_{r \rightarrow 0} \frac{|B_r(x_0) \cap A(t)|}{|B_r|} = \frac{1}{2} 
1370: \label{eq:regpt}
1371: \end{equation}
1372: in which case we will say that $(x_0, t_0)$ is a regular point of the free boundary.
1373: \end{theorem}
1374: \begin{pf}
1375: Simply use Equation\refeqn{WobProb}together with the results from \cite{Bl}.
1376: \newline Q.E.D. \newline
1377: \end{pf}
1378: \begin{theorem}[Better regularity] \label{betreg}
1379: Assume the hypotheses of the previous theorem and assume that $(x_0, t_0)$ is a
1380: regular point of the free boundary.  Then the free boundary intersected with
1381: $\{t = t_0\}$ will be Reifenberg vanishing near $x_0.$  (For a definition of
1382: Reifenberg vanishing, see \cite{Bl}.)  If $u_I$ is Dini continuous, then the
1383: free boundary will be $C^1$ (in space) near $x_0,$ and if $u_I \in C^{k,\alpha}$ 
1384: then the free boundary will be $C^{k+1,\alpha}$ (in space) near $x_0.$
1385: \end{theorem}
1386: \begin{pf}
1387: Again, just combine Equation\refeqn{WobProb}with the results from \cite{Bl}.
1388: \newline Q.E.D. \newline
1389: \end{pf}
1390: \begin{remark}[Clarification]  \label{clear}
1391: When we say ``near'' $x_0$ in the theorem above we mean near in space only.
1392: In other words, we are specifically talking about the free boundary restricted
1393: to the time slice $t = t_0.$
1394: \end{remark}
1395: At this point we need to call attention to our assumptions that $u_I$ is continuous
1396: and nondegenerate.  Indeed it is well known that there are examples of ``persistent 
1397: corners'' in Hele-Shaw problems.  (See \cite{KLV} for example.)  Our theorem does 
1398: not contradict this fact.  Our spatial regularity theorem says nothing if $u_I$ is 
1399: discontinuous, or if the set $\{ u_I = 1 \} \cap D^{c}$ is nonempty.  On the other 
1400: hand, our assumptions do not rule out ``focusing'' or changes in topology of the 
1401: diffusive region, so it is certainly nontrivial that corners do not arise in this 
1402: setting.  Consider a case where $u_I$ is very close to zero except for an $\epsilon$ 
1403: neighborhood of an annulus which contains the slot.  In this annulus assume that 
1404: $u_I$ is extremely close to one.  (The epsilon neighborhood is needed to make $u_I$
1405: continuous.)  In this case, the diffusive region will make its way around the annulus
1406: in each direction very quickly, but expand slowly into the region inside and outside
1407: of the annulus.  Thus, it will meet itself on the other side of the annulus long before
1408: it fills in the interior.
1409: 
1410: \
1411: 
1412: \psfig{file=Fingers.eps}
1413: 
1414: \
1415: 
1416: %-----------------------------------------------------------------------
1417: 
1418: \newsec{Continuity and dealing with all times}{CDWAT}
1419: We wish now to improve the results from the previous section by extending
1420: them from almost every time to every time.  Indeed, this prevents corners
1421: from arising as ``transitions'' from the situations which are permissible
1422: for sets of time with positive measure.  (Consider for example the sets
1423: $S(t) := \{ (x,y) \; : \; xy < t - 1$ for the interval $t \in [0, 2].$  By
1424: the results of the last section, it would still be possible for $S(t)$ to
1425: be the diffusive region for a Hele-Shaw flow, since the corner only occurs
1426: when $t$ belongs to the zero measure set $\{ 1 \}.$)  Of course in this 
1427: entire section we make the standing assumption that $u_I$ is nondegenerate 
1428: initial data (see Equation\refeqn{ndsit}\!).  We start with a simple lemma
1429: summarizing some of the regularity we have for $W.$
1430: 
1431: \begin{lemma}[Regularity for $W$]   \label{RegW}
1432: $W(x,t)$ is continuous in space and continuous and convex and nondecreasing 
1433: in time.  For almost every time, $T,$ $W(x, T)$ satisfies 
1434: Equation\refeqn{WobProb}with boundary data
1435: \begin{equation}
1436:     W(x, T) = p(x)T \ \ \ \ \ x \in \partial D \;.
1437: \label{eq:Wbdrydata}
1438: \end{equation}
1439: \end{lemma}
1440: \begin{pf}
1441: By using the maximum principle together with the fact that the diffusive
1442: region increases with time, we see that $V(x, t)$ must be an increasing
1443: function of time.  Now convexity of $W$ follows from this fact and from
1444: the definition of $W.$  The rest of the lemma follows immediately from 
1445: the definition of $W$ and the spatial continuity of $V(x,t)$ along with
1446: its boundary data on $\partial D.$
1447: \newline Q.E.D. \newline
1448: \end{pf}
1449: \begin{theorem}[Measure of the diffusive region]   \label{medire}
1450: There exists a constant $C = C(n, \alpha, \partial D)$ such that if 
1451: $0 \leq t - s \leq 1,$ then
1452: \begin{equation}
1453:    |A(t) \setminus A(s)| \leq C ||p||_{C^{2,\alpha}(\partial D)} 
1454:        \left( \frac{t - s}{1 - \lambda} \right) \;.
1455: \label{eq:mcwrtt}
1456: \end{equation}
1457: (Recall that $A(t)$ is the diffusive region at time $t,$ and for
1458: $S \subset \R^n,$ we let $|S|$ denote the Lebesgue n-dimensional measure
1459: of $S.$)
1460: \end{theorem}
1461: \begin{pf}
1462: We adapt the proof of Theorem 4.1 of \cite{Bl} to the current setting.
1463: Fix $t$ and $s.$  Because the diffusive regions are nested, and because 
1464: Equation\refeqn{WobProb}holds for almost every time, without loss of generality
1465: we can assume that it holds for both $t$ and $s.$  Let 
1466: $L := A(t) \setminus A(s),$ let $\Psi(x) := W(x,t) - W(x,s)$ and observe that
1467: $\Delta \Psi(x) = \chisub{L} (1 - u_{I}) \geq 0,$ and 
1468: \begin{equation}
1469: \begin{array}{ll}
1470:     \Psi(x) = 0 \ & x \in FB(t), \\
1471:     \Psi(x) = (t - s)p(x) \ & x \in \partial D \;.
1472: \end{array}
1473: \label{eq:bvd4wsmwt}
1474: \end{equation}
1475: From this fact and by the weak maximum principle, it follows that
1476: \begin{equation}
1477: 0 \leq W(x,t) - W(x,s) \leq (t - s)||p||_{L^{\infty}(\partial D)} 
1478: \label{eq:beW}
1479: \end{equation}
1480: for all $x \in A(t).$
1481: (Nonnegativity is actually a consequence of the previous lemma.)
1482: 
1483: Observe that $W(x,t) - W(x,s)$ is harmonic within $A(s)$ so that
1484: \begin{alignat*}{1}
1485: 0 &= \int_{A(s)} \Delta(W(x,t) - W(x,s)) dx \\
1486:   &= \int_{\partial D} \frac{\partial}{\partial \nu} (W(x,t) - W(x,s)) d\mathcal{H}^{n-1}
1487:    - \int_{FB(s)} \frac{\partial}{\partial \nu} (W(x,t) - W(x,s)) d\mathcal{H}^{n-1} \\
1488:   &= \int_{\partial D} \frac{\partial}{\partial \nu} (W(x,t) - W(x,s)) d\mathcal{H}^{n-1}
1489:    - \int_{FB(s)} \frac{\partial}{\partial \nu} W(x,t) d\mathcal{H}^{n-1} 
1490: \end{alignat*}
1491: Now by using boundary regularity for harmonic functions combined with 
1492: Equation\refeqn{beW}we can conclude
1493: \begin{equation}
1494:    \left| \int_{\partial D} \frac{\partial}{\partial \nu} 
1495:           (W(x,t) - W(x,s)) d\mathcal{H}^{n-1} \right| \leq 
1496:    C(n, \alpha, \partial D) (t - s)||p||_{C^{2,\alpha}(\partial D)}
1497: \label{eq:ellregest}
1498: \end{equation}
1499: By combining this fact with the last computation, we conclude that
1500: \begin{equation}
1501:    \left| \int_{FB(s)} \frac{\partial}{\partial \nu} W(x,t) d\mathcal{H}^{n-1} \right|
1502:    \leq C(n, \alpha, \partial D) (t - s) ||p||_{C^{2,\alpha}(\partial D)} \;.
1503: \label{eq:bdryestI}
1504: \end{equation}
1505: On the other hand we have
1506: \begin{alignat*}{1}
1507:    (1 - \lambda)|L| 
1508:       &\leq \int_{L} (1 - u_{I}) \; dx \\
1509:       &= \int_{L} \Delta W(x,t) \; dx \\
1510:       &= \int_{\partial L} \frac{\partial}{\partial \nu} W(x,t) d\mathcal{H}^{n-1} \\
1511:       &= \int_{FB(s)} \frac{\partial}{\partial \nu} W(x,t) d\mathcal{H}^{n-1}
1512: \end{alignat*}
1513: which we can combine with Equation\refeqn{bdryestI}to give us what we need.
1514: \newline Q.E.D. \newline
1515: \end{pf}
1516: \begin{corollary}[Continuity in $L^p$]  \label{CLp}
1517: Under the assumptions made at the beginning of this section, the map from $t$ 
1518: to the function
1519: $$\chisub{\{ W(x,t) > 0\}} (1 - u_{I})$$
1520: is a continuous function from $\R$ into $L^{p}(D^{c})$ for $1 \leq p < \infty.$
1521: \end{corollary}
1522: \begin{corollary}[Spatial regularity for every time]   \label{HOP}
1523: All of the results of Theorems\refthm{RFBS}and\refthm{betreg}hold for every time.
1524: \end{corollary}
1525: \begin{pf}
1526: It suffices to show that Equation\refeqn{WobProb}holds for all time.  Fix
1527: $\tilde{t} > 0$ and let $t_n \rightarrow \tilde{t}$ with $t_n$ chosen so that
1528: Equation\refeqn{WobProb}holds at each $t_n.$  Now take a ball, $B_R$ which is large
1529: enough to contain $A(\tilde{t})$ in its interior, and let 
1530: $\Omega := B_R \setminus D.$  For each $n$ we let $w_n(x)$ solve the boundary value
1531: problem
1532: \begin{equation}
1533:    \begin{array}{ll}
1534:             \Delta w_n(x) = \chisub{A(t_n)}(1 - u_{I}) & \ \text{in}
1535:                \ \ \Omega \vspace{.05in} \\
1536:             w_n(x) = t_n p(x) & \ \text{on} \ \ \partial D \\
1537:             w_n(x) = 0 & \ \text{on} \ \ \partial B_R \;.
1538:    \end{array}
1539: \label{eq:bvpop}
1540: \end{equation}
1541: By standard uniqueness results, $w_n(x) \equiv W(x, t_n),$ as they satisfy the same
1542: boundary value problem.  By standard elliptic regularity theory, since the boundary
1543: data on $\partial D$ will converge to $\tilde{t} p(x)$ and by the last corollary the 
1544: right hand side of the equation will converge in $L^{p}(\Omega)$ to 
1545: $\chisub{A(\tilde{t})}(1 - u_{I}),$ we can conclude that $w_n$ will converge to a 
1546: function $\tilde{w}$ which satisfies
1547: \begin{equation}
1548:    \begin{array}{ll}
1549:             \Delta \tilde{w}(x) = \chisub{A(\tilde{t})}(1 - u_{I}) & \ \text{in}
1550:                \ \ \Omega \vspace{.05in} \\
1551:             \tilde{w}(x) = \tilde{t} p(x) & \ \text{on} \ \ \partial D \\
1552:             \tilde{w}(x) = 0 & \ \text{on} \ \ \partial B_R \;.
1553:    \end{array}
1554: \label{eq:bvpoplim}
1555: \end{equation}
1556: On the other hand, $w_n(x) = W(x, t_n)$ converges to $W(x, \tilde{t})$ by the continuity
1557: of $W$ in time.
1558: \newline Q.E.D. \newline
1559: \end{pf}
1560: 
1561: %-----------------------------------------------------------------------
1562: 
1563: \newsec{Appendix}{App}
1564: Here we collect some facts we need to construct our subsolutions in the proof of
1565: Theorem\refthm{NMFB}\!.  We let $u(r; \alpha, \beta)$ denote the solution to:
1566: \begin{equation}
1567:   \begin{array}{rll}
1568:     \Delta_x u(r; \alpha, \beta) \!
1569:     &= 0 \ \ \
1570:     & \text{in} \ B_{1 + \alpha + \beta} \setminus B_{\alpha} \\
1571:     \ \\
1572:     u(r; \alpha, \beta) \! &= 1 \ \ \ & \text{on} \ \partial B_{\alpha} \\
1573:     \ \\
1574:     u(r; \alpha, \beta) \! &= 0 \ \ \ & \text{on} \ \partial B_{1 + \alpha + \beta} \\
1575:   \end{array}
1576: \label{eq:TestU}
1577: \end{equation}
1578: and we let $v(r; \alpha, \beta)$ denote the solution to:
1579: \begin{equation}
1580:   \begin{array}{rll}
1581:     \Delta v(r; \alpha, \beta) \!
1582:     &= 2n \ \ \
1583:     & \text{in} \ B_{1 + \alpha + \beta} \setminus B_{\alpha} \\
1584:     \ \\
1585:     v(r; \alpha, \beta) \!
1586:     &= 0 \ \ \
1587:     & \text{on} \ \partial \{ B_{1 + \alpha + \beta} \setminus B_{\alpha} \} \;.
1588:   \end{array}
1589: \label{eq:TestV}
1590: \end{equation}
1591: We will always assume that $\alpha \geq 1$ and $0 \leq \beta \leq 1.$
1592: 
1593: \begin{lemma}[Explicit Forms of Our Comparison Functions]  \label{RHFAR}
1594: If $n > 2,$ then $u$ has the explicit form
1595: \begin{equation}
1596:    u(r; \alpha, \beta) = 
1597:       \frac{r^{2 - n} - (1 + \alpha + \beta)^{2 - n}}
1598:       {\alpha^{2 - n} - (1 + \alpha + \beta)^{2 - n}} \;,
1599: \label{eq:Uform}
1600: \end{equation}
1601: and $v$ has the explicit form
1602: \begin{equation}
1603:    v(r; \alpha, \beta) = (r^2 - \alpha^2) + 
1604:       \frac{(\alpha^2 - (1 + \alpha + \beta)^2)(r^{2 - n} - \alpha^{2 - n})}
1605:            {(1 + \alpha + \beta)^{2 - n} - \alpha^{2 - n}} \;.
1606: \label{eq:Vform}
1607: \end{equation}
1608: If $n = 2,$ then $u$ has the explicit form
1609: \begin{equation}
1610:    u(r; \alpha, \beta) = \left[ \log \left( \frac{r}{1 + \alpha + \beta} \right) \right] /
1611:                     \left[ \log \left( \frac{\alpha}{1 + \alpha + \beta} \right) \right] \;,
1612: \label{eq:Uform2}
1613: \end{equation}
1614: and $v$ has the explicit form
1615: \begin{equation}
1616:    v(r; \alpha, \beta) = (r^2 - \alpha^2) + 
1617:    \frac{ \log \left( \displaystyle{ \frac{r}{\alpha} } \right) 
1618:    (\alpha^2 - (1 + \alpha + \beta)^2)}{\log \left( \displaystyle{
1619:       \frac{1 + \alpha + \beta}{\alpha} } \right) } \;.
1620: \label{eq:Vform2}
1621: \end{equation}
1622: \end{lemma}
1623: 
1624: \begin{lemma}[Derivatives on the Outer Boundaries]  \label{DOB}
1625: If $n > 2,$ then
1626: \begin{equation}
1627:    u_r(1 + \alpha + \beta; \alpha, \beta) = \frac{(2 - n)(1 + \alpha + \beta)^{1 - n}}
1628:                      {\alpha^{2 - n} - (1 + \alpha + \beta)^{2 - n}} < 0 \;,
1629: \label{eq:urfb}
1630: \end{equation}
1631: and
1632: \begin{equation}
1633:      v_r(1 + \alpha + \beta; \alpha, \beta) = 2(1 + \alpha + \beta) +
1634:  (n - 2)(1 + \alpha + \beta) \frac{(1 + \alpha + \beta)^2 - \alpha^2}
1635:                                   {(1 + \alpha + \beta)^{2 - n} - \alpha^{2 - n}} > 0 \;.
1636: \label{eq:vrfb}
1637: \end{equation}
1638: If $n = 2,$ then
1639: \begin{equation}
1640:    u_r(1 + \alpha + \beta; \alpha, \beta) = \frac{1}
1641:   {(1 + \alpha + \beta) \log \left( \displaystyle{\frac{\alpha}{1 + \alpha + \beta}} \right)} < 0 \;,
1642: \label{eq:u2rfb}
1643: \end{equation}
1644: and
1645: \begin{equation}
1646:      v_r(1 + \alpha + \beta; \alpha, \beta) = 2(1 + \alpha + \beta) +
1647:         \frac{\alpha^2 - (1 + \alpha + \beta)^2}{(1 + \alpha + \beta) \log \left(
1648:         \displaystyle{\frac{1 + \alpha + \beta}{\alpha}} \right)}
1649:                                    > 0 \;.
1650: \label{eq:v2rfb}
1651: \end{equation}
1652: Also, for any $n \geq 2,$ we have
1653: \begin{equation}
1654:    \lim_{\alpha \rightarrow \infty} u_r(1 + \alpha + \beta; \alpha, \beta) =
1655:    \frac{-1}{1 + \beta} < 0 \;,
1656: \label{eq:urfbLIM}
1657: \end{equation}
1658: and
1659: \begin{equation}
1660:    \lim_{\alpha \rightarrow \infty} v_r(1 + \alpha + \beta; \alpha, \beta) =
1661:    n(1 + \beta) > 0 \;.
1662: \label{eq:vrfbLIM}
1663: \end{equation}
1664: \end{lemma}
1665: \begin{corollary}[Bounds on the Outer Boundaries]  \label{BOB}
1666: For $0 \leq \beta \leq 1$ and $\alpha \geq 1,$ there exist constants $\gamma_i$
1667: which are all independent of $\alpha$ and $\beta$ such that
1668: \begin{equation}
1669:    -\gamma_1 \leq u_r(1 + \alpha + \beta; \alpha, \beta) \leq -\gamma_2 < 0 \;,
1670: \label{eq:uHG}
1671: \end{equation}
1672: and
1673: \begin{equation}
1674:    0 < \gamma_3 \leq v_r(1 + \alpha + \beta; \alpha, \beta) \leq \gamma_4 < \infty \;.
1675: \label{eq:vHG}
1676: \end{equation}
1677: \end{corollary}
1678: \begin{pf}
1679: The proof for $u_r$ is essentially the same as the proof for $v_r,$ so we will only
1680: deal with $u_r.$  Because the limit as $\alpha \rightarrow \infty$ is strictly negative
1681: by the previous lemma there is a large $\alpha_0$ such that for $\alpha \geq \alpha_0$
1682: we have the desired lower and upper bounds.  Next we simply use compactness for the 
1683: rest of the strip.
1684: \newline Q.E.D. \newline
1685: \end{pf}
1686: 
1687: \begin{thebibliography}{99}
1688: 
1689: \parskip=0pt
1690: \itemsep=0pt
1691: 
1692: \bibitem[AK]{AK}  D. Andreucci and M. Korten, Initial traces
1693: of solutions to a one-phase Stefan problem in an infinite
1694: strip, { \em Rev. Mat. Iberoamericana,} 9(1993), no. 2,
1695: 315--332.
1696: 
1697: \bibitem[BF]{BF}  B. Bazaliy and A. Friedman, Global existence
1698: and asymptotic stability for an elliptic-parabolic free boundary problem:
1699: An application to a model of tumor growth, { \em Indiana Univ. Math. J.,, }
1700: 52(2003), no. 5, 1265--1304.
1701: 
1702: \bibitem[BBH]{BBH}  Ph. B\'{e}nilan, L. Boccardo, and M.A. Herrero,
1703: On the limit of solutions of $u_t = \Delta u^m$ as $m \rightarrow
1704: \infty,$ { \em Rend. Mat. Univ. Pol. Torino, } Fascicolo Speciale (1989),
1705: Nonl. PDE's, 1--13.
1706: 
1707: \bibitem[BEG]{BEG}  Ph. B\'{e}nilan, L.C. Evans, R.F. Gariepy,
1708: On some singular limits of homogeneous semigroups, { \em J. Evol. Eq., }
1709: 3(2003), no. 2, 203--214.
1710: 
1711: \bibitem[Bl]{Bl}  I. Blank, Sharp results for the regularity
1712: and stability of the free boundary in the obstacle problem,
1713: { \em Indiana Univ. Math. J.} 50(2001), no. 3, 1077--1112.
1714: 
1715: \bibitem[Bo]{Bo}  J.E. Bouillet, Nonuniqueness in $L^{\infty}$: An
1716: example, { \em Lecture notes in pure and applied mathematics,}
1717: Vol. 148, Marcel Dekker, Inc., (1993).
1718: 
1719: \bibitem[BKM]{BKM}  J.E. Bouillet, M.K. Korten, and V. M\'{a}rquez,
1720: Singular limits and the ``Mesa'' problem, { \em Rev. Un. Mat.
1721: Argentina, } 41(1998), no. 1, 27--40.
1722: 
1723: \bibitem[C]{C}  L.A. Caffarelli,
1724: The obstacle problem revisited,
1725: { \em J. Fourier Anal. Appl., } 4(1998),  no. 4-5,
1726: 383--402.
1727: 
1728: \bibitem[CF]{CF}  L.A. Caffarelli and A. Friedman, Asymptotic
1729: behavior of solutions of $u_t = \Delta u^{m}$ as
1730: $m \rightarrow \infty,$ { \em Indiana Univ. Math. J.,} 36(1987),
1731: no. 4, 203--224.
1732: 
1733: \bibitem[DK]{DK}  B.E.J. Dahlberg and C.E. Kenig, Weak solutions of
1734: the porous medium equation, { \em Trans. Am. Math. Soc., } 336(1993),
1735: no. 2, 711--725.
1736: 
1737: \bibitem[DL]{DL}  P. Daskalopoulos and K. Lee, All time smooth
1738: solutions of the one-phase Stefan problem and the Hele-Shaw flow,
1739: { \em Comm. PDE, } 29(2004), no. 1-2, 71--88.
1740: 
1741: \bibitem[DB]{DB}  E. DiBenedetto, Continuity of weak solutions to
1742: certain singular parabolic equations, { \em Ann. Mat. Pura Appl., }
1743: (4), CXXX(1982), 131--176.
1744: 
1745: \bibitem[DF]{DF}  E. DiBenedetto and A. Friedman, The ill-posed
1746: Hele-Shaw model and the Stefan problem for supercooled water,
1747: { \em Trans. Amer. Math. Soc., } 282(1984), no. 1, 183--204.
1748: 
1749: \bibitem[EHKO]{EHKO}  C.M. Elliott, M.A. Herrero, J.R. King, and
1750: J.R. Ockendon, The mesa problem: Diffusion patterns for
1751: $u_t = \nabla \cdot (u^m \nabla u)$ as $m \rightarrow +\infty,$
1752: { \em IMA J. Appl. Math., } 37(1986), no. 2, 147--154.
1753: 
1754: %\bibitem[EJ]{EJ}  C.M. Elliott and V. Janovsk\'{y}, A variational inequality
1755: %approach to the Hele-Shaw flow with a moving boundary, { \em Proc. Royal
1756: %Soc. Edinburgh, } 88A, 93--107.   ????????????????????
1757: 
1758: \bibitem[FH]{FH}  A. Friedman and S. Huang, Asymptotic
1759: behavior of solutions of $u_t = \Delta \phi_m(u)$ as
1760: $m \rightarrow \infty$ with inconsistent initial values,
1761: { \em Analyse math\'{e}matique et applications, }
1762: Gauthier-Villars, Paris, (1988), 165--180.
1763: 
1764: \bibitem[GQ]{GQ}  O. Gil and F. Quir\'{o}s, Convergence of the
1765: porous medium equation to Hele-Shaw, { \em Nonlin. Anal., }
1766: 44(2001), no. 8, Ser. A: Theory Methods, 1111--1131.
1767: 
1768: \bibitem[I]{I}  N. Igbida, The mesa-limit of the porous-medium
1769: equation and the Hele-Shaw problem, { \em Diff. Int. Eqns., }
1770: 15(2002), no. 2, 129--146.
1771: 
1772: \bibitem[KT]{KT}  C.E. Kenig and T. Toro, Free boundary regularity
1773: for harmonic measures and Poisson Kernels, { \em Ann. of Math., }
1774: (2)150(1999), no. 2, 369--454.
1775: 
1776: \bibitem[KN]{KN}  D. Kinderlehrer and L. Nirenberg,
1777: Regularity in free boundary problems, { \em Ann.
1778: Scuola Norm. Sup. Pisa Cl. Sci., } (4)4(1977), no. 2,
1779: 373--391.
1780: 
1781: \bibitem[KLV]{KLV}  J.R. King, A.A. Lacey, and J.L. Vazquez,
1782: Persistence of corners in Hele-Shaw flow, { \em Euro. J. Appl.
1783: Math., } 6(1995), no. 5, 455--490.
1784: 
1785: \bibitem[K1]{K1}  M.K. Korten, Non-negative solutions of
1786: $u_t = \Delta(u - 1)_{+}:$ Regularity and uniqueness for
1787: the Cauchy problem, { \em Nonl. Anal., Th., Meth., and
1788: Appl.,} 27(1996), no. 5, 589--603.
1789: 
1790: \bibitem[K2]{K2}  M.K. Korten, A Fatou theorem for the
1791: equation $u_t = \Delta(u - 1)_{+},$ { \em Proc. Am.
1792: Math. Soc.,} 128(1999), no. 2, 439--444.
1793: 
1794: \bibitem[K3]{K3}  M.K. Korten, On the rectifiability of the free
1795: boundary of the one phase Stefan problem, to appear in { \em Comm.
1796: Anal. and Geom. }
1797: 
1798: \bibitem[KM]{KM}  M.K. Korten and C.N. Moore, The two phase Stefan
1799: problem:  Regularity of solutions to the Cauchy problem,
1800: submitted.
1801: 
1802: \bibitem[M]{M}  P. Mattila, Geometry of Sets and Measures
1803: in Euclidean Spaces, { \em Cambridges studies in advanced
1804: mathematics,} Vol. 44, Cambridge University Press, (1995).
1805: 
1806: \bibitem[MR]{MR}  J.A. McGeough and H. Rasmussen, On the
1807: derivation of the quasi-steady model in electrochemical
1808: machining, { \em J. Inst. Math. Appl., } 13(1974), 13--21.
1809: 
1810: \vspace{.3in} \noindent
1811: Department of Mathematics \\
1812: Worcester Polytechnic Institute \\
1813: Worcester, MA 01609 \\
1814: E-mail address: blanki@wpi.edu \\
1815: 
1816: \vspace{.2in} \noindent
1817: Department of Mathematics \\
1818: Kansas State University \\
1819: Manhattan, KS 66506 \\
1820: E-mail address: marianne@math.ksu.edu, cnmoore@math.ksu.edu \\
1821: 
1822: \end{thebibliography}
1823: 
1824: \end{document}
1825: 
1826: