Regularity of The Interfaces With Sign Changes of Solutions of The One-Dimensional Porous Medium Equation

Journal of Differential Equations 178, 1–59 (2002)
doi:10.1006/jdeq.2000.4002, available online at http://www.idealibrary.com on
Regularity of the Interfaces with Sign Changes

of Solutions of the One-Dimensional Porous
Medium Equation 1
Shigeru Sakaguchi
Department of Mathematical Sciences, Faculty of Science, Ehime University,

2-5 Bunkyo-cho, Matsuyama-shi, Ehime 790-8577, Japan
E-mail: sakaguch@dpc.ehime-u.ac.jp
Received October 28, 1998
In previous papers we considered the Cauchy problem for the one-dimensional

evolution p-Laplacian equation for nonzero, bounded, and nonnegative initial data
having compact support, and showed that after a finite time the set of spatial
critical points of the nonnegative solution u=u(x, t) in {u > 0} consists of one
point, the spatial maximum point of u, and the curve of the spatial maximum points
is continuous with respect to the time variable. Since the spatial derivative “x u
satisfies the porous medium equation with sign changes, the curve of the spatial
maximum points is regarded as an interface with sign changes of “x u. On the other
hand, in a paper by M. Bertsch and D. Hilhorst (1991, Appl. Anal. 41, 111–130) the
interfaces where the solutions change their sign were studied in detail for the initial-
boundary value problems of the generalized porous medium equation over
two-dimensional cylinders. But the monotonicity of the initial data is assumed
there. As is noted in Section 4 of our earlier work (1996, J. Math. Anal. Appl. 203,
78–103), the monotonicity of “x u( · , t) in some neighborhood of the spatial
maximum point of u( · , t) cannot be assumed, and therefore, if this monotonicity
for some large t > 0 is proved, then by the method of Bertsch and Hilhorst (cited
above) one may get more precise regularity properties of the curve of the spatial
maximum points. The purpose of the present paper is twofold. One is to remove
some monotonicity assumption for initial data in Bertsch and Hilhorst’s theorem
concerning the regularity of the interfaces with sign changes of solutions of the one-
dimensional generalized porous medium equation. By comparing the solution with
appropriate symmetric nonnegative solutions we shall get the monotonicity of the
solution near the interface after a finite time. The other is as a by-product of the
method to get C 1 regularity of the curves of the spatial maximum points of non-
negative solutions of the Cauchy problem for the evolution p-Laplacian equation
for sufficiently large t. © 2002 Elsevier Science
Key Words: degenerate diffusion equation; singular diffusion equation; initial-
boundary value problem; spatial maximum point; interface with sign changes;
porous medium equation; p-Laplacian; asymptotic behaviour.
1
Partially supported by a Grant-in-Aid for Scientific Research (10640175), Ministry of
Education, Science, Sports, and Culture, Japan.
1
⁄ 0022-0396/02 $35.00
© 2002 Elsevier Science
All rights reserved.
2 SHIGERU SAKAGUCHI
Contents.
1. Introduction.
2. Preliminaries.
3. The interface is a continuous curve.
4. Continuous differentiability of the interface.
5. The curve of the spatial maximum points.
6. Appendix: Proof of Lemma 2.1.
1. INTRODUCTION
In the previous papers [Sak 1, Sak 2] we considered the Cauchy problem

for one-dimensional evolution p-Laplacian equation for nonzero, bounded,
and nonnegative initial data having compact support, and showed that
after a finite time the set of spatial critical points of the nonnegative solu-
tion, in the set where the solution is positive, consists of one point, the
spatial maximum point, and the curve of the spatial maximum points is
continuous with respect to the time variable. Precisely, let u=u(x, t) be the
unique bounded weak solution of the Cauchy problem for the one-
dimensional evolution p-Laplacian equation with p > 1 and p ] 2
“t u=“x (|“x u| p − 2 “x u) in R × R+ and u(x, 0)=u0 (x) in R, (1.1)
where u0 is a nonzero, nonnegative, and bounded function in R having

compact support, and where R+=(0, .). Let C(t) be the set of spatial
critical points of u where u is positive:
C(t)={x ¥ R; “x u(x, t)=0 and u(x, t) > 0}. (1.2)
Then it was shown in [Sak 1, Sak 2] that there exist a time T > 0 and a
continuous function x=x(t) on [T, .) such that C(t)={x(t)} for each
t F T. Namely, after a finite time for each time the set C(t) consists of the
unique spatial maximum point of u( · , t). If we put v=“x u, then v satisfies
the porous medium equation with sign changes
“t v=“ 2x (|v| p − 2 v) in R × R+. (1.3)
Therefore, the curve x=x(t), the curve of the spatial maximum points of
u, is regarded as an interface with sign changes of the solution v of the
porous medium equation.
On the other hand, in Bertsch and Hilhorst [BH] the interfaces where
the weak solutions change their sign were studied in detail for the initial-
boundary value problems of the generalized porous medium equation over
THE POROUS MEDIUM EQUATION 3
two-dimensional cylinders. But the monotonicity of the initial data is

assumed there. As is noted in [Sak 1, Sect. 4, p. 100], the monotonicity of
v( · , t)=“x u( · , t) in some neighborhood of the spatial maximum point of
u( · , t) cannot be assumed, and therefore, if this monotonicity for some
large t > 0 is proved, then by the method of [BH] one may get more
precise regularity properties of the curve of the spatial maximum points of u.
The purpose of the present paper is twofold. One is to remove some
monotonicity assumption for initial data in Bertsch and Hilhorst’s theorem
[BH] concerning the regularity of the interfaces with sign changes of weak
solutions of the one-dimensional generalized porous medium equation. By
comparing the solution with appropriate symmetric nonnegative weak
solutions we shall get the monotonicity of the solution near the interface
after a finite time. The other is as a by-product of the method to get C 1
regularity of the curve x=x(t), the curve of the spatial maximum points of
the nonnegative weak solution of the Cauchy problem for the evolution
p-Laplacian equation, for sufficiently large t.
Let us state our main result concerning the regularity of the interfaces
with sign changes of solutions of the one-dimensional generalized porous
medium equation precisely. Let f=f(s) be a real-valued function on R
belonging to C(R) 5 C 3(R 0 {0}) and f satisfies
f(0)=0 and f −(s) > 0 for s ] 0. (1.4)
A typical example of f is given by
f(s)=˛ kk |s||s|
1
2
m1 − 1
m2 − 1
s
s
if s F 0,
if s < 0,
where k1 , k2 , m1 , and m2 are positive constants. Consider the following

initial-boundary value problem
˛ “t v=“ 2x (f(v))
f(v)(−L, t)=a, f(v)(L, t)=−b
v(x, 0)=v0 (x)
in Q=(−L, L) × R+,
for t > 0,
for x ¥ (−L, L),
(1.5)
where a, b, and L are positive constants, and where v0 ¥ C([ − L, L])

satisfies
˛ f(v0 )(−L)=a and f(v0 )(L)=−b.
There exists a point x0 ¥ (−L, L) satisfying v0 (x) ˛ >< 00 if − L E x < x0 ,

if x0 < x E L.
(1.6)
4 SHIGERU SAKAGUCHI
In order to define a weak solution generally, by letting “p Q be the

parabolic boundary of Q given by
“p Q=((−L, L) × {0}) 2 ({ − L, L} × [0, .)),
we consider the problem,
˛ “v=g
v=“ (f(v))
t
2
x in Q=(−L, L) × R+,
on “p Q,
(1.7)
where g ¥ C(“p Q) is a given function. Note that in problem (1.5)
˛ v0
g= f −1(a)
−1
f (−b)
on
on
on
(−L, L) × {0},
{ − L} × [0, .),
{L} × [0, .).
For T > 0 , let
Q0, T =(−L, L) × (0, T] and

“p Q0, T =((−L, L) × {0}) 2 ({ − L, L} × [0, T]).
We also consider the problem,
˛ v=gt
2
“ v=“ (f(v))
x in Q0, T ,
on “p Q0, T ,
(1.8)
where g ¥ C(“p Q0, T ) is a given function. A function v=v(x, t) is said to be

a weak solution of problem (1.8), if v ¥ C(Q0, T ) 5 L .(Q0, T ) and v satisfies
the integral identity
FF {f(v) “ 2x t+v “t t} dx dt
Q0, T
T
=F {f(g)(L, t) “x t(L, t) − f(g)(−L, t) “x t(−L, t)} dt
0
L
+F {v(x, T) t(x, T) − g(x, 0) t(x, 0)} dx (1.9)
−L
for any t ¥ C 2(Q0, T ) satisfying t(L, t)=t(−L, t)=0 for t ¥ [0, T]. Also, a
function v=v(x, t) is said to be a weak solution of problem (1.7), if for any
T > 0 v is a weak solution of problem (1.8). Our main result is an improve-
ment of the result of Bertsch and Hilhorst [BH, Theorem 1.3, p. 113]:
Theorem 1.1. Let v ¥ C(Q̄) be the unique weak solution of problem

(1.5). Then there exist a continuous function x=x(t) on [0, .) with
x(0)=x0 and a time TÄ F 0 satisfying:
(1) For any t F 0, x(t) ¥ (−L, L) and
v(x, t)˛ >< 00 if x < x(t),

if x > x(t).
(2) x( · ) ¥ C 1(TÄ , .).

(3) When > d0 f −(s)/s ds=−> 0−d f −(s)/s ds=. for some d > 0, TÄ =0.
(4) When > d0 f −(s)/s ds < . or − > 0−d f −(s)/s ds < . for some d > 0,
x(t)=x0 for any t ¥ [0, TÄ ].
(5) When > d0 f −(s)/s ds < . and − > 0−d f −(s)/s ds < . for some d > 0,
x(t) is uniformly Hölder continuous with exponent 12 on [0, .).
Remark 1.2. (i) Existence and uniqueness of the bounded weak
solution of problem (1.5) and the comparison principle were proved in
[Hua, DK] with the help of the regularity result of Sacks [Sac]. Further-
more, compared with some continuous solutions we shall see that this
solution is continuous on Q̄ (see Lemma 2.4 in the present paper). If the
initial and boundary values are nonnegative, then in [Gil 3] it was shown
that there exists a unique bounded weak solution which is continuous on Q̄.
(ii) It is known that > d0 f −(s)/s ds < . for some d > 0 if and only if
the equation “t v=“ 2x (f(v)) has the property of finite speed of propagation
for nonnegative solutions (see [Gil 2, Pe], for example).
(iii) When f(s)=|s| m − 1 s with 0 < m < 1, the condition that
> 0 f (s)/s ds=−> 0−d f −(s)/s ds=. for some d > 0 is satisfied. Therefore (3)
d −
of Theorem 1.1 implies that TÄ =0. Namely, in the fast diffusion case for
the porous medium equation the interface x=x(t) becomes C 1 instantly.
(Note that m=p − 1 for Eq. (1.3).)
(iv) In [BH], the locally Lipschitz continuity of f on R is assumed,
and concerning assumption (1.6) the monotonicity of the initial data v0 on
some neighborhood of x0 is assumed. This monotonicity is essential in
[BH] in order to use the transformation (x, t) Q (f(v), t) (that is, they
consider the value of f(v) as a new independent variable near the inter-
face). In [BH] under this monotonicity assumption for initial data the
authors proved that when > d0 f −(s)/s ds=. or − > 0−d f −(s)/s ds=. for
some d > 0, TÄ =0 (see [BH, (iv) of Theorem 1.3, p. 113, or p. 123]). In the
present paper, without this monotonicity assumption for initial data, we
prove (3) and (4) of Theorem 1.1. Namely, when > d0 f −(s)/s ds < . and
− > 0−d f −(s)/s ds=., or when > d0 f −(s)/s ds=. and − > 0−d f −(s)/s ds < .
6 SHIGERU SAKAGUCHI
for some d > 0, we can not prove that TÄ =0 without this monotonicity
assumption for initial data. The reason is that technically we need a time
in order to gain the monotonicity of v with respect to the space variable
x on some neighborhood of the interface, when > d0 f −(s)/s ds < . or
− > 0−d f −(s)/s ds < . (see the proof of Lemma 3.4 in the present paper).
(v) In view of the proof of Theorem 1.1 in the present paper (in
particular see (4.12) for the determination of TÄ ), by the same proof as in
[BH, pp. 126–127] we can get the same formula as in [BH, (1.6), p. 113 or
p. 127]:
“x (f(v))(x, t) − “x (f(v))(x(t), t)
x −(t)=− lim for any t > TÄ .
x Q x(t) v(x, t)
The remainder of the present paper is organized as follows. In Section 2

we introduce a class of symmetric nonnegative weak solutions of the
Cauchy problem (see Lemma 2.1 and Remark 2.2), which is used in the
following sections instead of Barenblatt solutions of the porous medium
equation. Also, we give the comparison principle (see Lemma 2.3) and
approximate problems for (1.5) (see problem (2.12)), and we obtain the
existence and regularity of solutions of problem (1.5) (see Lemma 2.4).
Finally in Section 2, we give a useful lemma (see Lemma 2.5), which shows
that the number of spatial sign changes of the difference of two continuous
weak solutions of (1.8) always equals 1, provided the difference has no
zeroes on the lateral boundaries and provided the number equals 1 initially.
This comes from the principle that the number of spatial sign changes or
spatial zeroes of the difference of two classical solutions of a one-dimen-
sional uniformly parabolic equation never increases with time, provided the
difference has no zeroes on the lateral boundaries (see [An 1, Hen, Ma 1,
Sat]). This principle has many applications (see for example [An 2, GV,
SGKM]). Section 3 is devoted to show that the interface of the solution of
problem (1.5) is a continuous curve of the form x=x(t). In particular,
Lemma 3.4 together with Lemma 3.7 is the key to the applicability of the
arguments of [BH], that is, these lemmas show that after a finite time the
solution becomes a monotone function with respect to the space variable x
near the interface. The proof of this key lemma consists of comparing the
solution with appropriate symmetric nonnegative weak solutions given by
Lemma 2.1 and Remark 2.2 of Section 2. This comparison is done by
Lemma 2.5 of Section 2. In Section 4 we show that this curve x=x(t) is
C 1, which completes the proof of Theorem 1.1. In Section 5 we get C 1
regularity of the curves of the spatial maximum points of nonnegative
solutions of the Cauchy problem for the evolution p-Laplacian equation
(see Theorem 5.1). The last Section 6 (Appendix) is devoted to the proof of
Lemma 2.1.
2. PRELIMINARIES
First of all, let us consider the following one-dimensional Cauchy

problem,
“t u=“ 2x (f(u)) in R × R+ and u(x, 0)=u0 (x) in R, (2.1)
where u0 is a bounded, continuous, and nonnegative function on R. A

function u=u(x, t) is said to be a nonnegative weak solution of (2.1), if u is
bounded, nonnegative, and continuous on R × [0, .) with u(x, 0)=u0 (x)
for x ¥ R, and u satisfies “t u=“ 2x (f(u)) in the sense of distributions on
R × R+, precisely,
FF {f(u) “ 2x t+u “t t} dx dt=0 for any t ¥ C. +

0 (R × R ). (2.2)
R × R+
Note that for any nonnegative weak solution u of (2.1), since u is continu-
ous on R × [0, .), u is also a weak solution of problem (1.7) with initial
and boundary values g=u|“p Q , as is noted in [DK, Proof of Corollary 2.11,
p. 273] (see [AC, Proof of Proposition 1.2, pp. 353–355] for its proof).
It is known that there exists a unique nonnegative weak solution u of
(2.1) (see [Gil 3, Theorems 1 and 4, p. 182 and p. 199]). Also, comparison
principles for nonnegative weak solutions of initial-boundary value
problems are proved in [Gil 3]. We shall use the following class of non-
negative weak solutions of (2.1) instead of Barenblatt solutions of the
porous medium equation. We give a proof of the following lemma in
Section 6 (Appendix) for convenience.
Lemma 2.1. Denote by G the class of nonzero and nonnegative functions

gŒs in C .(R) having compact support, say supp g, and satisfying:
(i) g(x)=g(−x) for any x ¥ R.
(ii) If g(x) > 0 and x > 0, then g −(x) < 0.
(iii) maxx ¥ R g(x)(=g(0)) < lims Q . f(s)(E .).
Take two numbers M0 and M1 such that 0 < M0 < lims Q . f(s) and M1 > 0.
For each g ¥ G with maxR g E M0 and maxR |g −| E M1 , let f=f −1(g) (f −1
denotes the inverse function of f) and let u be the unique nonnegative weak
solution of (2.1) for u0 =f. Then u satisfies the following:
(1) 0 E u E maxx ¥ R f(x) in R × [0, .).
(2) For all (x1 , t1 ), (x2 , t2 ) ¥ R × [0, .)
1 1
f −1(max [0, f(u)(x1 , t1 ) − 12 M1 |x1 − x2 | − M1 |t1 − t2 | 2]) − M1 |t1 − t2 | 2
1
E f −1(min [M0 , f(u)(x2 , t2 )+12 M1 |x1 − x2 |+M1 |t1 − t2 | 2]).
8 SHIGERU SAKAGUCHI
(3) For D={(x, t) ¥ R × R+; u(x, t) > 0}, u, “x (f(u)), and “t (f(u))
belong to C 2, 1(D), where C 2, 1(D) is the set of functions in D twice and once
continuously differentiable with respect to x and t, respectively, and u satisfies
the equation of (2.1) in the classical sense in D.
(4) u(x, t)=u(−x, t) for any (x, t) ¥ R × R+.
(5) >R u(x, t) dx=>R f(x) dx for any t > 0.
(6) For each t > 0, u( · , t) is monotone nonincreasing on [0, .),
further, “x u(x, t) < 0, provided u(x, t) > 0 and x > 0. In particular, u(0, t)=
maxy ¥ R u(y, t) > 0 and u(0, t) is monotone nonincreasing in t.
(7) u(x, t) Q 0 as t Q . uniformly in R.
(8) Under the condition that > d0 f −(s)/s ds < . for some d > 0, if we
put g(t)=sup{x ¥ R; u(x, t) > 0} then g is a real-valued, continuous, and
monotone nondecreasing function on [0, .), satisfying limt Q . g(t)=.
and {x ¥ R; u(x, t) > 0}=(−g(t), g(t)) for any t F 0, and putting C=
−1 1
2(> 0f (M0 ) f −(s)/s ds) 2 yields
1
g(t2 ) E g(t1 )+C(t2 − t1 ) 2 for any t2 > t1 F 0.
(Note that C a 0 as M0 a 0.) Furthermore, for each y > 0, g is uniformly
Lipschitz continuous on [y, .).
(9) Under the condition that > d0 f −(s)/s ds=. for some d > 0, u > 0 in
R × R+.
Remark 2.2. Instead of Barenblatt solutions of the porous medium
equation, we shall use symmetric nonnegative weak solutions vŒs of
“t v=“ 2x (f(v)) given by
v(x, t)=u(x − y, t − s) for (x, t) ¥ R × [s, .),
where u is a nonnegative weak solution of (2.1) with u0 (x)=f −1(g)(x) for
some g ¥ G and where y ¥ R and s F 0. For convenience let us write
v(x, t)=vsym (x, t)=v(x, t; y, s, g). (2.3)
In particular we put v(x, t; y, s, 0) — 0.
In order to prove Theorem 1.1 let us quote a comparison principle from
Dahlberg and Kenig [DK, Lemma 2.3, p. 271], which implies the unique-
ness of the weak solution of (1.8) as well as of (1.7).
Lemma 2.3 (Comparison Principle). Let g1 , g2 ¥ C(“p Q0, T ) and let v1 , v2

be weak solutions of (1.8) with g=g1 , g2 respectively. If g1 E g2 on “p Q0, T ,
then v1 E v2 in Q0, T . In particular, if g1 , g2 ¥ C(“p Q) satisfy g1 E g2 on “p Q
and if v1 , v2 are weak solutions of (1.7) with g=g1 , g2 respectively, then
v1 E v2 in Q.
Proof. Although in [DK] only nonnegative weak solutions belonging

to the function space C([0, T], L 1((−L, L))) 5 L .(Q0, T ) are considered,
the same proof works for our weak solutions since our weak solutions
belong to C(Q0, T ) 5 L .(Q0, T ). L
Let us introduce approximate problems for (1.5), and for convenience let
us prove existence of the solution v. For this purpose we shall first intro-
duce a sequence of functions approximating the inverse function f −1 of f.
Since for the solution v of (1.5) we have from Lemma 2.3
min v0 (x) E v E max v0 (x) in Q̄, (2.4)

x ¥ [ − L, L] x ¥ [ − L, L]
so by extending f appropriately we may assume additionally that
lim f(s)=−. and lim f(s)=..

s Q −. sQ.
Put
b(s)=f −1(s) for s ¥ R. (2.5)
Let us use a smooth approximation result for b as in [Hua, Theorem 2.1,

p. 964]. Precisely, let us define functions M, m: R+ × R+ Q R+ by
M(r, R)= ˛ max b (s) r E |s| E R
max{b (−R), b (R)} −

−
−
if 0 < r E R,
if r F R,
m(r, R)=˛
−
min b (s) r E |s| E R if 0 < r E R,
− −
min{b (−R), b (R)} if r F R.
Then we see that M( · , R) is monotone nonincreasing, m( · , R) is monotone

nondecreasing, and
m(|s|, R) E b −(s) E M(|s|, R) for 0 < |s| E R. (2.6)
For e ¥ (0, 12 ), define
e
b (s)= 0 ˛ b(s+e) − b(e)
b(s − e) − b(−e)
if s > 0,
if s=0,
if s < 0.
10 SHIGERU SAKAGUCHI
Let r ¥ C .
0 (R) be an even function satisfying
.
F r(s) ds=1, r F 0 in R, and r(s)=0 if |s| F 1.
−.
Namely, r is a standard mollifier. Set re (s)=e −1r(s/e) and put for any
integer n F 3
1
bn (s)=(re f b e)(s) − (re f b e)(0) with e= .
n
Then, as is shown in [Hua, Theorem 2.1, pp. 964–965], we see that for
each n F 3 bn ¥ C .(R), b −n > 0 in R, bn (0)=0, bn (R)=R, and
m(|s|, R+1) E b −n (s) E M(|s|, R+1) for 0 < |s| E R. (2.7)
Furthermore, we have
bn Q b as n Q . uniformly on any bounded interval of R. (2.8)

{bn ; n F 3} is equicontinuous on any bounded interval of R. (2.9)
In order to preserve the nodal property of the initial data f(v0 ) for later use
(see Lemma 2.5 in this section), let us regularize f(v0 ) by using the heat
equation. Precisely, let H ¥ C .([ − L, L] × (0, 1]) 5 C(Q0, 1 ) be the unique
classical solution of the problem:
˛ “t H=“ 2x H
H( ± L, t)=f(v0 )( ± L)
H(x, 0)=f(v0 )(x)
in Q0, 1 =(−L, L) × (0, 1],
for t ¥ (0, 1] respectively,
for x ¥ (−L, L).
(2.10)
Then, in particular, putting Hn (x)=H(x, 1n ) for each n F 3 yields that
Hn ¥ C .([ − L, L]), and Hn Q f(v0 ) as n Q . uniformly on [ − L, L].

(2.11)
Furthermore, by using the theory of uniformly parabolic equations [LSU]

we see that for each n F 3 there exists a unique classical solution Vn ¥
C .(Q̄) of the problem:
˛ “t (bn (Vn ))=“ 2x Vn

Vn (−L, t)=a, Vn (L, t)=−b
Vn (x, 0)=Hn (x)
in Q=(−L, L) × R+,
for t > 0,
for x ¥ (−L, L).
(2.12)
Then, by applying the maximum principle to H and Vn , and using (2.8), we

have
˛ min
x ¥ [ − L, L]
f(v0 )(x) E Vn E max
sup |bn (Vn )| E C,

Q
x ¥ [ − L, L]
f(v0 )(x) in Q̄,
(2.13)
for some constant C > 0 independent of n F 3. Therefore, in view of (2.7)

and (2.13) by the result of [Sac, Theorem 1.1, p. 389] we see that the
family {Vn } is equicontinuous on every compact set in Q. Hence it follows
from (2.8), (2.11), and the uniqueness of the weak solution of (1.5) that
there exists a unique weak solution v ¥ C(Q) 5 L .(Q) such that
Vn Q f(v) and bn (Vn ) Q v as n Q . uniformly on any compact set in Q.

(2.14)
Furthermore we have:
Lemma 2.4. The solution v is continuous in Q̄, if the value of v on “p Q is

defined by the initial and boundary conditions. Furthermore, if we put
D={(x, t) ¥ Q; v(x, t) ] 0}, then v, “x (f(v)), and “t (f(v)) belong to
C 2, 1(D), v satisfies the equation of (1.5) in the classical sense in D, and for
each compact set E in D
Vn Q f(v), bn (Vn ) Q v, “x Vn Q “x (f(v)), and

“t Vn Q “t (f(v)) as n Q . in C 2, 1(E).
Proof. With the help of (2.14) and the fact that v ¥ C(Q), the latter half
follows from the standard theory of uniformly parabolic equations [Fr].
(Besides the regularity for f(v) follows from the fact that “t (f(v))=
f −(v) “ 2x (f(v)) in D in the classical sense.) Therefore it suffices to prove the
former half. Let us first prove that v is continuous at any point on
{ − L, L} × (0, .). Consider the boundary { − L} × (0, .). Take any e > 0.
Since the initial data v0 ¥ C([ − L, L]) satisfies (1.6), there exist two con-
stants a+ and a− such that
V− < f(v0 ) < V+ on [ − L, L], (2.15)
where V± (x)=a ± (x+L)+a ± e. If we put v ± =b(V± ), then both v+ and

v− satisfy the partial differential equation of (1.5). Hence it follows from
(1.6), (2.15), and Lemma 2.3 that
v− E v E v+ in Q.
This implies that for any point (−L, t0 ) ¥ { − L} × (0, .)
b(a − e) E lim inf v(x, t) E lim sup v(x, t) E b(a+e). (2.16)

(x, t) Q (−L, t0 ) (x, t) Q (−L, t0 )
Since e > 0 is chosen arbitrarily, so v is continuous at (−L, t0 ) if we put

f(v)=a there in view of the boundary condition. The continuity at any
point on {L} × (0, .) is proved similarly.
It remains to prove that v is continuous at any point on [ − L, L] × {0} if
we put v(x, 0)=v0 (x) for any x ¥ [ − L, L]. Take any point y ¥ [ − L, L]
and take any e > 0. Let us show that
lim inf v(x, t) F v0 (y) − e. (2.17)

(x, t) Q (y, 0)
Since the initial data v0 ¥ C([ − L, L]) satisfies (1.6), then there exists a
function h ¥ C .([ − L, L]) such that
supp h − … (−L, L), h(y)=f(v0 (y) − e), and

h < f(v0 ) in [ − L, L]. (2.18)
For n F 3 let Ṽn ¥ C .(Q̄) be the unique classical solution of the problem:
˛ “t (bn (Ṽn ))=“ 2x Ṽn

Ṽn ( ± L, t)=h( ± L)
Ṽn (x, 0)=h(x)
in Q,
for t > 0 respectively,
for x ¥ (−L, L).
(2.19)
Then, by the maximum principle and (2.8) we have
˛ min
x ¥ [ − L, L]
h(x) E Ṽn E max
sup |bn (Ṽn )| E C,

Q
x ¥ [ − L, L]
h(x) in Q̄,
(2.20)
for some constant C > 0 independent of n F 3. Furthermore, as in [DP,

Proof of Lemma 3, pp. 178–179] by putting K=maxx ¥ [ − L, L] |h −(x)| we can
see that
|“x Ṽn | E K in Q̄ for any n F 3. (2.21)
Indeed, by putting W± (x)= ± K(x+L)+h(−L) we have
W− E h E W+ on [ − L, L]. (2.22)
Then both W+ and W− satisfy the partial differential equation of (2.19) for
any n F 3. Hence it follows from (2.22) and the comparison principle that
W− E Ṽn E W+ in Q̄. (2.23)
Since W− (−L)=Ṽn (−L, t)=W+(−L) for any t F 0, this implies that for
any n F 3
− K E “x Ṽn (−L, t) E K for any t F 0.
By estimating “x Ṽn (L, t) similarly, we conclude that for any n F 3
max{|“x Ṽn (−L, t)|, |“x Ṽn (L, t)|} E K for any t F 0. (2.24)
Therefore, by applying the maximum principle to “x Ṽn we get (2.21), since

putting Z=“x Ṽn yields
1 b ' (Ṽ ) “ Ṽ
“t Z= − “ 2x Z − n − n x 2 n “x Z in Q.
b n (Ṽn ) (b n (Ṽn ))
The same argument as in proving (2.14) implies that there exists a unique
weak solution ṽ ¥ C(Q) 5 L .(Q) of the problem,
˛ “t ṽ=“ 2x (f(ṽ))
f(ṽ)( ± L, t)=h( ± L)
ṽ(x, 0)=b(h)(x)
in Q,
for t > 0 respectively,
for x ¥ (−L, L),
(2.25)
such that
Ṽn Q f(ṽ) and bn (Ṽn ) Q ṽ as n Q . uniformly on any compact set in Q.

(2.26)
Then it follows from (2.18) and Lemma 2.3 that
ṽ E v in Q. (2.27)
As in the beginning of the proof of Lemma 2.4, by putting f(ṽ)( ± L, t)

=h( ± L) for any t > 0 we see that ṽ belongs to C([ − L, L] × (0, .)). By
setting ṽ=b(h) on [ − L, L] × {0}, we regard ṽ as a function defined in Q̄.
The definition of K, (2.21), and (2.26) imply that
|f(ṽ)(x1 , t) − f(ṽ)(x2 , t)| E K |x1 − x2 | for any points (x1 , t) and (x2 , t) ¥ Q̄.
(2.28)
Let us consider the continuity of ṽ with respect to t F 0. Let t2 > t1 F 0 and

(t2 − t1 ) 1/2 < 2L. Take any xg ¥ [ − L, L]. Then we can take a rectangle
R=[xg1 , xg2 ] × [t1 , t2 ] … Q̄ such that xg ¥ [xg1 , xg2 ] and xg2 − xg1 =
(t2 − t1 ) 1/2. Integrating the partial differential equation of (2.19) over R
yields that for any n F 3
xg2 t2
F {bn (Ṽn )(x, t2 ) − bn (Ṽn )(x, t1 )} dx=F {“x Ṽn (xg2 , t) − “x Ṽn (xg1 , t)} dt.
xg1 t1
Hence by (2.21)
:F xg 2
xg1
:
{bn (Ṽn )(x, t2 ) − bn (Ṽn )(x, t1 )} dx E 2K |t2 − t1 |.
In view of (2.26) and (2.8), by letting n Q . we get
:Fxg1
xg2
:
{ṽ(x, t2 ) − ṽ(x, t1 )} dx E 2K |t2 − t1 |. (2.29)
Here, since xg ¥ [xg 1 , xg2 ] and xg2 − xg1 =(t2 − t1 ) 1/2, from (2.28) we have
for j=1, 2 and for any x ¥ [xg1 , xg2 ]
1 1
b(f(ṽ)(xg , tj ) − K(t2 − t1 ) 2) E ṽ(x, tj ) E b(f(ṽ)(xg , tj )+K(t2 − t1 ) 2).
Then, by combining these inequalities with (2.29) we conclude that for any
1
xg ¥ [ − L, L] and for any t2 > t1 F 0 with (t2 − t1 ) 2 < 2L
˛ b(f(ṽ)(x g
g
2 2
2
1
1
1
, t ) − K(t − t ) ) E b(f(ṽ)(x , t )+K(t − t ) )+2K(t − t ) ,
2
1
g 1
b(f(ṽ)(x , t ) − K(t − t ) ) E b(f(ṽ)(x , t )+K(t − t ) )+2K(t − t ) .

1
2
g 2
2
2
1
1
1
2
1
2
2
2
1
1
1
2
1
2
(2.30)
These inequalities together with (2.28) imply that ṽ ¥ C(Q̄). Therefore, since
ṽ(x, 0)=b(h(x)) for x ¥ [ − L, L], from (2.18) and (2.27) we get (2.17).
Similarly we get
lim sup v(x, t) E v0 (y)+e. (2.31)

(x, t) Q (y, 0)
Since e > 0 and y ¥ [ − L, L] are chosen arbitrarily, it follows from (2.17)

and (2.31) that v is continuous at any point on [ − L, L] × {0} if we
put v(x, 0)=v0 (x) for any x ¥ [ − L, L]. This completes the proof of
Lemma 2.4. L
Finally in this section let us give a lemma concerning the number of

spatial sign changes of the difference of two weak solutions of problem
(1.8). For this purpose, for a function u ¥ C([ − L, L]) satisfying
u(−L) u(L) ] 0, define the number of sign changes of u on [ − L, L] to be
the supremum over all k such that there exists a finite sequence
− L E x1 < · · · < xk+1 E L with u(xj ) u(xj+1 ) < 0 for any 1 E j E k.
Lemma 2.5. Let v1 , v2 ¥ C(Q0, T ) be weak solutions of problem (1.8) with

g=v1 | “p Q0, T , v2 | “p Q0, T respectively. Suppose that the number of sign changes of
v1 ( · , 0) − v2 ( · , 0) on [ − L, L] equals 1 and moreover
v1 − v2 ˛ >< 00 on
on
{ − L} × [0, T],
{L} × [0, T].
(2.32)
Then the number of sign changes of v1 ( · , t) − v2 ( · , t) on [ − L, L] equals 1

for any t ¥ [0, T]. Furthermore, if we assume that f ¥ C 3(R) and f −(s) > 0
for any s ¥ R additionally, then for each t ¥ (0, T) the number of zeros of
v1 ( · , t) − v2 ( · , t) on [ − L, L] equals 1 and
“x (v1 − v2 )(x, t) < 0 for a unique x ¥ (−L, L) satisfying (v1 − v2 )(x, t)=0.
Proof. Let us first regularize the initial data f(vj )( · , 0)(j=1, 2) as in

(2.10) and (2.11). Precisely, for each j=1, 2, by replacing v0 ( · ) by vj ( · , 0)
in problem (2.10), we let H j ¥ C .([ − L, L] × (0, 1]) 5 C(Q0, 1 ) be the
corresponding solution of problem (2.10). Let H g=H 1 − H 2. Then H g
satisfies the heat equation. In view of the hypotheses, we observe that for
each t ¥ (0, 1] the number of sign changes of H g on the parabolic bound-
ary “p Q0, t equals 1. Here, we recall the result of Sattinger [Sat, Corollary 5,
p. 84] or of Matano [Ma 1, Lemma 2.6, pp. 408–409] for one-dimensional
uniformly parabolic equations with bounded coefficients, which says in our
situation that the number of sign changes of H g( · , t) on [ − L, L] is not
greater than the number of sign changes of H g on “p Q0, t . Therefore, in
view of the boundary condition of (2.10) and hypothesis (2.32), we see that
the number of sign changes of H g( · , t) on [ − L, L]

equals 1 for any t ¥ (0, 1]. (2.33)
On the other hand, putting H nj (x)=H j(x, 1n ) for each n F 3 yields that for
each j=1, 2
H nj ¥ C .([ − L, L]), and H nj ( · ) Q f(vj )( · , 0)

as n Q . uniformly on [ − L, L].
In view of (2.32) and the continuity of v1 and v2 , we can take four sequences
{G nj+ }, {G nj − } (j=1, 2) in C .([0, T]) such that
˛ G nj ± ( · ) Q f(vj )( ± L, · )
G j±
n =f(vj )( ± L, 0)
G 1n − − G 2n − > 0 and G 1+
as n Q . uniformly on [0, T] respectively,
in a neighbourhood of t=0, respectively, (2.34)
2+
n − Gn < 0 on [0, T].
By using the theory of uniformly parabolic equations [LSU] we see that

for each j=1, 2 and for n F 3 there exists a unique classical solution
V nj ¥ C .(Q0, T ) of the problem:
˛ “t (bn (V nj ))=“ 2x V nj
V nj ( ± L, t)=G nj ± (t)
j
V (x, 0)=H (x)
n
j
n
in Q0, T =(−L, L) × (0, T],
for t ¥ [0, T], respectively,
for x ¥ (−L, L).
(2.35)
Then by the same argument as in proving (2.14) we get for each j=1, 2
V jn Q f(vj ) and bn (V jn ) Q vj as n Q . (2.36)
uniformly on any compact set in Q0, T . Let Nn (t) be the number of sign
changes of V 1n ( · , t) − V 2n ( · , t), or of bn (V 1n )( · , t) − bn (V 2n )( · , t), on
[ − L, L] for each t ¥ [0, T]. Observe that putting Un =V 1n − V 2n yields
1 “ c(x, t)
“t Un = “ 2x Un − t U in Q0, T , (2.37)
c(x, t) c(x, t) n
where c(x, t)=> 10 b −n (V 2n +hUn ) dh. Also, by using (2.33) and the third line
of (2.34), we observe that for each t ¥ (0, T] the number of sign changes of
Un on the parabolic boundary “p Q0, t equals 1. Then we can use the result
of Sattinger [Sat, Corollary 5, p. 84] or of Matano [Ma 1, Lemma 2.6,
pp. 408–409] again, and we see that
Nn (t)=1 for any t ¥ [0, T]. (2.38)
Hence, by letting n Q ., from (2.32) and (2.36) we see that the number of
sign changes of v1 ( · , t) − v2 ( · , t) on [ − L, L] equals 1 for any t ¥ [0, T].
It remains to consider the case where f ¥ C 3(R) and f −(s) > 0 for any
s ¥ R. In this case it follows from the theory of uniformly parabolic equa-
tions [LSU] that both v1 and v2 belong to C 2, 1(Q0, T ), and these satisfy the
differential equation in the classical sense. Let us put U=f(v1 ) − f(v2 ) and
let N(t) be the number of sign changes of U( · , t) on [ − L, L] for each

t ¥ (0, T]. Then we already know that
N(t)=1 for any t ¥ [0, T]. (2.39)
Observe that U is regarded as a classical solution of a uniformly parabolic

equation similar to (2.37) for Un . With the help of (2.39) and this fact, we
can show
if U(x, t)=0 for some (x, t) ¥ (−L, L) × (0, T), then “x U(x, t) < 0. (2.40)
Indeed, let U(x̃, t̃)=0 for some (x̃, t̃) ¥ (−L, L) × (0, T). Suppose that
“x U(x̃, t̃) F 0. When “x U(x̃, t̃) > 0, in view of (2.32) we immediately see
that N(t̃) F 3, which is a contradiction. When “x U(x̃, t̃)=0, we have
U(x̃, t̃)=“x U(x̃, t̃)=0. (2.41)
Therefore, it follows from (2.41) and the theory of Angenent concerning

the zero sets of solutions of one-dimensional uniformly parabolic equations
(see [An 1, Theorem B with Theorem D, pp. 80–82]) that there exist
y > 0, e > 0, and at least two points, say xj ¥ (x̃ − e, x̃+e)(j=1, 2), such
that
U(xj , t̃ − y)=0 and “x U(xj , t̃ − y) ] 0 for j=1, 2. (2.42)
(Notice that the theory of Angenent is applicable because of the hypotheses

that f ¥ C 3(R) and f −(s) > 0 for any s ¥ R.) Relation (2.42) together with
(2.32) implies that N(t̃ − y) F 3, which is a contradiction. Consequently, we
see that (2.40) holds.
Condition (2.40) together with (2.39) implies that for each t ¥ (0, T) the
function U( · , t) has only one zero, say x ¥ (−L, L), and “x U(x, t) < 0. This
completes the proof. L
Remark 2.6. In view of differential equation (2.37), we needed to
regularize the initial data in order to use the result of [Sat, Ma 1].
Furthermore, in order to preserve the nodal property of the initial data we
used the heat equation as above.
3. THE INTERFACE IS A CONTINUOUS CURVE
Now, let us consider the interface of v. We begin with:
Lemma 3.1. For any t > 0 there exist x− (t) and x+(t) with − L < x− (t)
E x+(t) < L such that
˛
v(x, t) =0
>0
<0
if x < x− (t),
if x− (t) E x E x+(t),
if x > x+(t).
Proof. As in [Sak 1, Lemma 3.6, pp. 95–96], by employing an argu-

ment in [Hen, Proof of Theorem 6, pp. 180–181] we can prove this lemma.
(See also [Sat, Proof of Theorem 4, p. 84, Ma 1, Proofs of Lemmas 2.5 and
2.6, pp. 408–409].) For convenience let us give a proof. It suffices to show
that for any t > 0 the set {x ¥ [ − L, L]; v(x, t) > 0} equals [ − L, x− (t)) for
some x− (t) ¥ (−L, L), since we can prove the other one similarly. For
simplicity put
Q(t)={x ¥ [ − L, L]; v(x, t) > 0}. (3.1)
Let tg > 0. Since v(−L, tg )=b(a) > 0, we let wg be the component of Q(tg )
containing − L. Note that wg is a half-open interval, since v(L, tg )=
b(−b) < 0. Suppose that there exists another component of Q(tg ), say w.
Then w 5 wg =”. Let Sw and Sw g be the components of {(x, t) ¥
[ − L, L] × [0, tg ]; v(x, t) > 0} containing w and wg , respectively. Hence,
since v(−L, t)=b(a) > 0 and v(L, t)=b(−b) < 0 for any t F 0, it follows
from (1.6) that Sw g contains the whole of Q(0) × {0}(=[ − L, x0 ) × {0}).
First let us show that Sw =Sw g . Indeed, suppose that Sw 5 Sw g =”. Then
Sw 5 ([ − L, L] × {0}) is empty. Observe that
v > 0 in Sw , v=0 on “Sw 0 w, and Sw is compact. (3.2)
Therefore the positive maximum of v over Sw is attained in the interior of

Sw or in w. However this contradicts the strong maximum principle (see
[Fr, RR]), since by Lemma 2.4, v is regarded as a classical solution of a
linear uniformly parabolic equation without zero-order term in each
compact set contained in D={(x, t) ¥ Q; v(x, t) ] 0}. Hence we have
Sw =Sw g .
Note that any interval in {(x, tg ); x ¥ (−L, L)} which meets w and wg
must have a point (x, tg ) with v(x, tg )=0. Since Sw 5 ((−L, L) × (0, tg )) is
open and connected (hence path connected), there exists a simple path in
Sw 5 ((−L, L) × (0, tg )) with one end point in w × {tg } and the other in
wg × {tg }. Adding a line segment in {(x, tg ); x ¥ (−L, L)} joining these
endpoints, we obtain a simple closed curve bounding a bounded region, say
W. Let c be the simple curve consisting of “W 5 ((−L, L) × (0, tg )) and the
end points of “W 5 ((−L, L) × {tg }). Then, by the continuity of v we have
vFg>0 on c (3.3)
for some positive constant g > 0. By (2.14) there exists n0 F 3 such that for
any n F n0
bn (Vn ) F 12 g on c. (3.4)
Since each bn (Vn ) is regarded as a solution of a linear uniformly parabolic

equation without zero-order term, the maximum principle implies that for
any n F n0
bn (Vn ) F 12 g in W̄. (3.5)
Then, letting n Q . we get from (2.14)
v F 12 g in W̄.
This contradicts the fact that v has a zero on “W 5 ((−L, L) × {tg }).
Therefore the proof of Lemma 3.1 is completed. L
Concerning (1) of Theorem 1.1 we have:
Lemma 3.2. Suppose that > d0 f −(s)/s ds < . or − > 0−d f −(s)/s ds < .
for some d > 0. Then x− (t)=x+(t) for any t > 0. Furthermore, if one
puts x(t)=x− (t) for any t > 0 and x(0)=x0 , then x(t) is continuous on
[0, .) and (1) of Theorem 1.1 holds. When > d0 f −(s)/s ds < . and
− > 0−d f −(s)/s ds < . for some d > 0, x(t) is uniformly Hölder continuous
with exponent 12 on [0, .), that is, (5) of Theorem 1.1 holds.
Remark 3.3. When > d0 f −(s)/s ds=−> 0−d f −(s)/s ds=. for some d > 0,
we shall prove (1) of Theorem 1.1 in Lemma 3.10. If we assume moreover
that f(s)=−f(−s) for any s ¥ R, we can prove this by using an antisym-
metric solution as in [Sak 2, Proof of Lemma 2.3, pp. 1055–1057].
Proof of Lemma 3.2. Once it is proved that x− (t)=x+(t) for any t > 0,
the continuity of x(t) together with x(0)=x0 follows from the continuity
of v and (1.6). So let us show that x− (t)=x+(t) for any t > 0. For d > 0 we
distinguish the following three cases:
(i) > d0 f −(s)/s ds < . and − > 0−d f −(s)/s ds=..

(ii) > d0 f −(s)/s ds=. and − > 0−d f −(s)/s ds < ..
(iii) > d0 f −(s)/s ds < . and − > 0−d f −(s)/s ds < ..
Since we can deal with cases (i) and (ii) similarly, let us consider case (i)
only. In this case the equation has the property of finite speed of propaga-
tion for nonnegative solutions, and it has the property of infinite speed of
propagation for nonpositive solutions. Suppose that there exists t0 > 0 such
that x− (t0 ) < x+(t0 ). Then we have
lim inf x− (t) F x+(t0 ). (3.6)

t ‘ t0
Indeed, by putting xg =lim inft ‘ t0 x− (t), we suppose that xg < x+(t0 ). Then
there exists a sequence {tj } with tj ‘ t0 as j ‘ . such that
xg =lim x− (tj ) < x+(t0 ). (3.7)

jQ.
Let vj be the unique nonnegative weak solution of the problem:
˛ “t vj =“ 2x (f(vj ))
f(vj )(−L, t)=a
vj (x, tj )=max{v(x, tj ), 0}
in (−L, .) × (tj , .),
for t > tj ,
for x ¥ (−L, .),
(3.8)
where existence and uniqueness of vj are proved in [Gil 3, Theorems 2 and 5,

pp. 186 and 199]. Then, since > d0 f −(s)/s ds < ., by using the result of
Gilding [Gil 2, Theorem 5, p. 260] concerning the interface of nonnegative
solutions, we see that there exists a constant C > 0 independent of j
satisfying
1
zj (t) E x− (tj )+C(t − tj ) 2 for any t > tj , (3.9)
where zj (t)=sup{x ¥ (−L, .); vj (x, t) > 0}. Indeed, the constant C is
1
given by C=2(> M −
0 f (s)/s ds)
2 with M=max
x ¥ [ − L, L] v0 (x) in [Gil 2].
Therefore, in view of (3.7) we can choose j sufficiently large to get
zj (t0 ) < x+(t0 ). On the other hand, by Lemma 2.3 we get
v E vj in (−L, L) × (tj , .).
Therefore, we have
vE0 in [zj (t0 ), L] × [tj , t0 ].
Then, since − > 0−d f −(s)/s ds=., so the property of infinite speed of
propagation implies that
v<0 in (zj (t0 ), L] × [tj , t0 ].
This contradicts the fact that v(x+(t0 ), t0 )=0 because of zj (t0 ) < x+(t0 ).
Consequently we get (3.6).
In particular, we have from (3.6)
lim inf x− (t) > x− (t0 ),

t ‘ t0
which implies that there exists e > 0 satisfying
v>0 in [ − L, x− (t0 )+e] × [t0 − e, t0 ).
Therefore, since > d0 f −(s)/s ds < ., by comparing v with a symmetric non-

negative weak solution v( · , · ; x− (t0 ), t0 − e, g) given in Remark 2.2, we see
that v(x− (t0 ), t0 ) > 0, which is a contradiction and the proof for case (i) is
completed. Here we chose maxR g and supp g sufficiently small in (2.3).
It remains to consider case (iii). Remark that in this case the equation
has the property of finite speed of propagation both for nonnegative
solutions and for nonpositive solutions (see [Gil 2]). We can deal with this
case as in [Sak 1, Lemmas 3.7 and 3.8, pp. 96–100]. Suppose that there
exists t0 > 0 such that x− (t0 ) < x+(t0 ). Let v+ and v− be the unique non-
negative weak solutions of the problems,
˛ “t v+=“ 2x (f(v+))
f(v+)(−L, t)=a
v+(x, t0 )=max{v(x, t0 ), 0}
in (−L, .) × (t0 , .),
for t > t0 ,
for x ¥ (−L, .),
(3.10)
and
˛ “t v− =“ 2x ( − f(−v− ))
− f(−v− )(L, t)=b
v− (x, t0 )=max{ − v(x, t0 ), 0}
in (−., L) × (t0 , .),
for t > t0 ,
for x ¥ (−., L),
(3.11)
respectively. Since x− (t0 ) < x+(t0 ), by the property of finite speed of

propagation and by the uniqueness of the weak solution of the problem:
˛ “t v=“ 2x (f(v))
f(v)(−L, t)=a, f(v)(L, t)=−b
v(x, t0 )=v(x, t0 )
in Qt0 , . =(−L, L) × (t0 , .),
for t > t0 ,
for x ¥ (−L, L),
(3.12)
we see that v=v+ − v− in [ − L, L] × [t0 , T], where T is determined by
T=inf{t > t0 ; supp v+( · , t) 5 supp v− ( · , t) ] ”}. (3.13)

Namely, max{v, 0}=v+ and max{ − v, 0}=v− in [ − L, L] × [t0 , T].

Therefore it follows from Lemma 3.1 that the curves x=x− (t) and
x=x+(t) for t ¥ [t0 , T] equal the right-hand interface of v+ and the
left-hand interface of v− , respectively. Hence both x− (t) and − x+(t) are
monotone nondecreasing and continuous functions in [t0 , T] and
x− (T)=x+(T) (see [Gil 2]). Next, let us show that both x− (t) and x+(t)
are continuous at t=t0 . It suffices to show the continuity from below of
x− (t) and x+(t) at t=t0 . Let us show that limt ‘ t0 x− (t)=x− (t0 ). The
continuity of v and Lemma 3.1 imply that
lim inf x− (t) F x− (t0 ). (3.14)

t ‘ t0
Hence let us show that lim supt ‘ t0 x− (t) E x− (t0 ). By putting x g=

lim supt ‘ t0 x− (t), we suppose that x g > x− (t0 ). Then there exists a sequence
{tj } with tj ‘ t0 as j ‘ . such that
x g=lim x− (tj ) > x− (t0 ). (3.15)

jQ.
As in the proof of (3.6), let v j be the unique nonnegative weak solution of

the problem:
˛ “t v j=“ 2x ( − f(−v j))

j
− f(−v )(L, t)=b
v j(x, tj )=max{ − v(x, tj ), 0}
in (−., L) × (tj , .),
for t > tj ,
for x ¥ (−., L).
(3.16)
Then it follows from Lemma 2.3 that
v F − vj in [ − L, L] × [tj , .). (3.17)
Let x=z j(t) be the left-hand interface of v j defined by
z j(t)=inf{x ¥ (−., L); v j(x, t) > 0} for any t F tj .
Then it follows from the result of [Gil 2, Theorem 5, p. 260] that z j(t) is a
monotone nonincreasing and continuous function in [tj , .) with
z j(tj )=x+(tj ) such that
1
z j(t) F x+(tj ) − C(t − tj ) 2 for any t > tj (3.18)
for some constant C > 0 independent of j. Since x− (tj ) Q x g and tj ‘ t0 as

j Q ., from (3.18) we can choose j sufficiently large to get z j(t0 ) > x− (t0 ).
Therefore (3.17) implies that z j(t0 ) E x+(t0 ) and
v=0 in [x− (t0 ), z j(t0 )] × [tj , t0 ]. (3.19)
Indeed, since v F 0 in [x− (t0 ), z j(t0 )] × [tj , t0 ], if v(x̃, t̃) > 0 for some (x̃, t̃)
in the interior of this rectangle, then by comparing v with a symmetric
nonnegative weak solution v( · , · ; x̃, t̃, g) given in Remark 2.2 we see that
v(x̃, t0 ) > 0, which contradicts the fact that v=0 on [x− (t0 ), z j(t0 )] × {t0 }.
Here we chose maxR g and supp g sufficiently small in (2.3).
On the other hand, (3.19) and (3.15) contradict Lemma 3.1. Namely
we get
lim sup x− (t) E x− (t0 ).

t ‘ t0
Combining this with (3.14) yields
lim x− (t)=x− (t0 ).

t ‘ t0
Similarly we get limt ‘ t0 x+(t)=x+(t0 ) and conclude that both x− (t) and
x+(t) are continuous at t=t0 .
By repeating these arguments for a little smaller time instead of t0 , we see
that both x− (t) and − x+(t) are monotone nondecreasing and continuous
functions on (0, t0 ]. Hence
lim x− (t) E x− (t0 ) < x+(t0 ) E lim x+(t). (3.20)

ta0 ta0
Therefore it follows from Lemma 3.1 and the continuity of v that

v(x, 0)=v0 (x)=0 for any x ¥ [x− (t0 ), x+(t0 )]. This contradicts (1.6).
It remains to show the Hölder continuity of x(t). Take t0 F 0 arbitrarily.
Consider v+ and v− solving (3.10) and (3.11), respectively. Then by Lemma 2.3
− v− E v E v+ in Qt0 , . . (3.21)
In view of the property of finite speed of propagation, let x=z+(t) and

x=z− (t) be the right-hand interface of v+ and the left-hand interface of v−
for t F t0 , respectively. Then it follows from the result of [Gil 2, Theorem 5,
p. 260] that z+(t) and − z− (t) are monotone nondecreasing and continuous
functions in [t0 , .) with z+(t0 )=z− (t0 )=x(t0 ) such that
1 1
z+(t) E x(t0 )+C(t − t0 ) 2 and z− (t) F x(t0 ) − C(t − t0 ) 2 for any t F t0 ,
where C is a positive constant depending only on maxx ¥ [ − L, L] |v0 (x)| and f.

Combining these inequalities with (3.21) yields
1
|x(t) − x(t0 )| E C |t − t0 | 2 for t > t0 F 0. (3.22)
Thus the proof for case (iii) is completed. L
The following key lemma ensures that after a finite time v becomes a
monotone function with respect to x near the interface.
Lemma 3.4. There exists T1 > 0 such that one can find three numbers
e1 > 0, c1 , and c2 with − L < c1 < x− (T1 ) E x+(T1 ) < c2 < L satisfying:
˛ v(x, T1 ) ˛ >< e− e
“x v(x, T1 ) < 0
1
1
if x ¥ [ − L, c1 ),
if x ¥ (c2 , L],
if x ¥ [c1 , x− (T1 )) 2 (x+(T1 ), c2 ].
Furthermore the following hold:
(1) When > d0 f −(s)/s ds=−> 0−d f −(s)/s ds=. for some d > 0, one can
take any number T1 > 0, in particular, one can choose T1 > 0 as small as one
wants.
(2) When > d0 f −(s)/s ds < . or − > 0−d f −(s)/s ds < . for some d > 0, if
x(t1 ) ] x(0)(=x0 ) for some t1 > 0, then one can take T1 =t1 . Here x(t) is
given in Lemma 3.2.
Proof. In view of (1.6), take g ¥ G in Lemma 2.1 such that
˛ max g=g(0) < a,

R
supp g( · +L) … (−2L − x0 , x0 ),

g(x+L) < f(v0 )(x) for any x ¥ [ − L, x0 ).
(3.23)
Consider the symmetric nonnegative weak solution vsym (x, t)=v(x, t;

− L, 0, g) given in Remark 2.2. Then we have
v(x, 0) − vsym (x, 0) ˛ >< 00 if x ¥ [ − L, x0 ),

if x ¥ (x0 , L].
(3.24)
Let N(t) be the number of sign changes of v( · , t) − vsym ( · , t) on [ − L, L]

for each t F 0. (3.24) means that N(0)=1. Since by (1) of Lemma 2.1 we
have 0 E vsym E maxR b(g) ( < b(a)), so
v − vsym ˛ <> 00 on { − L} × [0, .),

on {L} × [0, .).
(3.25)
Therefore it follows from Lemma 2.5 that
N(t)=1 for any t > 0. (3.26)
Furthermore, by using (3.26) we can show
if v(x, t)=vsym (x, t) > 0 for some (x, t) ¥ Q, then “x v(x, t) E “x vsym (x, t) < 0.
(3.27)
Indeed, let v(x̃, t̃)=vsym (x̃, t̃) > 0 for some (x̃, t̃) ¥ Q. Then it follows from
Lemma 2.4 and (3) of Lemma 2.1 that both v and vsym are classical solu-
tions in some neighborhood of (x̃, t̃). Since by (6) of Lemma 2.1 we have
“x vsym (x̃, t̃) < 0, it suffices to show that “x v(x̃, t̃) − “x vsym (x̃, t̃) E 0. So
suppose that “x v(x̃, t̃ ) − “x vsym (x̃, t̃) > 0. Then this together with (3.25)
immediately implies that N(t̃) F 3. This contradicts (3.26) and shows that
(3.27) holds.
For d > 0 let us distinguish the following two cases:
(i) > d0 f −(s)/s ds < ..
(ii) > d0 f −(s)/s ds=..
In case (i), by (8) of Lemma 2.1 there exists T > 0 such that
vsym > 0 in [ − L, L] × [T, .). (3.28)
Take T1 F T arbitrarily and put
z=sup{y ¥ [ − L, L]; v(y, T1 ) − vsym (y, T1 ) F 0}.
Then it follows from (3.25) and (3.28) that
− L < z < x− (T1 ) and v(z, T1 )=vsym (z, T1 ) > 0. (3.29)
Moreover we have
0 < v(x, T1 ) < vsym (x, T1 ) for any x ¥ (z, x− (T1 )). (3.30)
From (3.27) and (3.29) we see that
“x v(z, T1 ) < 0. (3.31)
For each l ¥ [0, 1] consider the symmetric nonnegative weak solution in

Remark 2.2, say v lsym , given by
v lsym (x, t)=v(x, t; − L, 0, lg). (3.32)
Note that v 1sym — vsym and v 0sym — 0. By the construction of symmetric non-
negative weak solutions in Lemma 2.1, we have
0 E v lsym
1
E v lsym
2
if 0 E l1 < l2 E 1. (3.33)
Furthermore, by (1) and (2) of Lemma 2.1, in particular the mapping:

l W v lsym ( · , T1 ) from [0, 1] to C([ − L, L]) is continuous, and v lsym a 0
(— v 0sym ) as l a 0. Note that (3.27) holds even if one replaces vsym by v lsym . In
view of (3.30) we observe that
v 0sym (x, T1 ) — 0 < v(x, T1 ) < vsym (x, T1 )

— v 1sym (x, T1 ) for each x ¥ (z, x− (T1 )). (3.34)
Hence it follows from the continuity of v lsym with respect to l that for each
x ¥ (z, x− (T1 )) there exists l ¥ (0, 1) satisfying
v(x, T1 )=v lsym (x, T1 ) > 0. (3.35)
Then (3.27) implies that
“x v(x, T1 ) < 0 for any x ¥ (z, x− (T1 )). (3.36)
Consequently, by (3.31) and (3.36) we conclude that for each T1 F T there

exists z ¥ (−L, x(T1 )) satisfying
“x v(x, T1 ) < 0 for any x ¥ [z, x(T1 )). (3.37)
In case (ii), instead of (3.28) we have by (9) of Lemma 2.1
vsym > 0 in [ − L, L] × (0, .). (3.38)
Therefore we can take T1 > 0 arbitrarily, and by the same argument as

above we conclude that for each T1 > 0 there exists z ¥ (−L, x− (T1 ))
satisfying
“x v(x, T1 ) < 0 for any x ¥ [z, x− (T1 )). (3.39)

By using symmetric nonnegative weak solutions vsym =v(x, t; L, 0, g) for

the equation “t vsym =“ 2x ( − f(−vsym )) and considering v − (−vsym ), similarly
we can get the following:
(i) − When − > 0−d f −(s)/s ds < . for some d > 0, there exists T − > 0
such that for each T1 F T − we can find z ¥ (x(T1 ), L) satisfying
“x v(x, T1 ) < 0 for any x ¥ (x(T1 ), z]. (3.40)
(ii) − When − > 0−d f −(s)/s ds=. for some d > 0, for each T1 > 0 we
can find z ¥ (x+(T1 ), L) satisfying
“x v(x, T1 ) < 0 for any x ¥ (x+(T1 ), z]. (3.41)
Consequently, in view of (3.37), (3.39), (3.40), and (3.41), we see that

Lemma 3.4 holds except (2).
It remains to prove (2). In this case by Lemma 3.2 we note that
x− (t)=x+(t)=x(t) for any t > 0. Remark that in the above proof we can
take any T1 > 0 in each case of (ii) and (ii) −. Therefore, in order to prove (2)
it suffices to consider cases (i) and (i) −. Since these two cases can be dealt
with similarly as in the above proof, it suffices to consider case (i) and
prove that there exists z ¥ (−L, x(t1 )) satisfying
“x v(x, t1 ) < 0 for any x ¥ [z, x(t1 )). (3.42)
In view of x(t1 ) ] x(0), we distinguish the following two cases:

(A) x(0) < x(t1 ).
(B) x(0) > x(t1 ).
Consider case (A) first. Take a small number r > 0 satisfying
x(0) < x(t1 ) − 3r < x(t1 ) < x(t1 )+r < L. (3.43)
Put x1 =x(t1 ) − r and take another g ¥ G such that
˛ supp g( · − x1 )=[x1 − 2r, x1 +2r],

max g=g(0) < min{a, f(v)(x1 , t1 )}.
(3.44)
R

x1 , 0, g) given in Remark 2.2. Then from (3.44) we have (3.24). Let
N(t) be the number of sign changes of v( · , t) − vsym ( · , t) on [ − L, L] for
each t F 0. (3.24) means that N(0)=1. Since by (1) of Lemma 2.1 we have
0 E vsym E maxR b(g)( < b(a)), so we have (3.25). Therefore by Lemma 2.5
we have (3.26). In view of (3.43), (3.44), and (1), (8) of Lemma 2.1, we
observe that
0 < max vsym (y, t1 )=vsym (x1 , t1 ) < v(x1 , t1 ) and

y¥R
vsym (x(t1 ), t1 ) > 0=v(x(t1 ), t1 ). (3.45)
Hence, following the argument from (3.27) to (3.31) yields that there exists
z ¥ (x1 , x(t1 )) with v(z, t1 )=vsym (z, t1 ) > 0, and moreover
˛ “0 <v(z,v(x,t )t E) <“ vv
x 1
1
x
sym (x, t1 )
sym (z, t1 ) < 0.
for any x ¥ (z, x(t1 )),
(3.46)
Introducing the family of the symmetric nonnegative weak solutions

{v lsym ; l ¥ [0, 1]} given by
v lsym (x, t)=v(x, t; x1 , 0, lg),
and following the argument from (3.32) to (3.37), yield (3.42). This
completes the proof in case (A).
Let us proceed to case (B). Take a small number r > 0 satisfying
− L < x(t1 ) − r < x(t1 )+r < x(0). (3.47)
Put x1 =x(t1 ) − r and take another g ¥ G such that
˛ supp g( · − x1 )=[x1 − 2r, x1 +2r],

max g=g(0) < min{a, f(v)(x1 , t1 )},
R
g(x − x1 ) < f(v)(x, 0) for any x ¥ [ − L, x(0)).

(3.48)
Then, by considering the symmetric nonnegative weak solution vsym (x, t)=
v(x, t; x1 , 0, g) given in Remark 2.2 and proceeding as in case (A), we get
(3.24), (3.25), and (3.26). Furthermore, in view of (3.47), (3.48), and (1), (8)
of Lemma 2.1, we observe that (3.45) also holds. Therefore, by the same
proof as in case (A) we see that (3.42) holds also in case (B). This completes
the proof. L
In view of Lemma 3.4, instead of Vn (see (2.12)), for each n F 3 let us
introduce a unique classical solution Wn ¥ C .(QT1 , . ) of the problem:
˛ “t (bn (Wn ))=“ 2x Wn

Wn (−L, t)=a, Wn (L, t)=−b
W (x, T )=H̃ 1 x, 2
n 1
1
n
in QT1 , . =(−L, L) × (T1 , .),
for t > T1 ,
for x ¥ (−L, L).

(3.49)
Here H̃ ¥ C .([ − L, L] × (0, 1]) 5 C(Q0, 1 ) is the unique classical solution

of problem (2.10) where v0 ( · ) is replaced by v( · , T1 ). Then we notice:
Remark 3.5. If we replace Vn , f(v0 )(x), and Q by Wn , f(v)(x, T1 ), and
QT1 , . , respectively, then by the same proof we get (2.13), (2.14), and
Lemma 2.4.
Furthermore, by Lemma 3.4 and Lemma 2.5 we have:
Lemma 3.6. For each (h, t) ¥ [ − e1 , e1 ] × [T1 , .), there exists a unique
x ¥ (−L, L) satisfying Wn (x, t)=f(h). Furthermore “x Wn (x, t) < 0 for such
x. Here T1 > 0 and e1 > 0 are the numbers given in Lemma 3.4.
Proof. Let h ¥ [ − e1 , e1 ]. Then, it follows from Lemma 3.4 that the
number of sign changes of f(v)( · , T1 ) − f(h) equals 1. Observe that the
constant f(h) satisfies the heat equation, and in problem (2.10), where
v0 ( · ) is replaced by v( · , T1 ),
H̃ − f(h) ˛ >< 00 on { − L} × [0, 1],

on {L} × [0, 1].
(3.50)
Here we can apply Lemma 2.5 to the heat equation by considering f(s) — s
and see that for each t ¥ (0, 1) the number of zeros of H̃( · , t) − f(h) on
[ − L, L] equals 1 and
“x (H̃ − f(h))(x, t) < 0 for a unique x ¥ (−L, L)

satisfying (H̃ − f(h))(x, t)=0. (3.51)
Hence, in problem (3.49), the number of sign changes of bn (Wn )( · , T1 ) −

bn (f(h)) on [ − L, L] equals 1. Further, observe that
bn (Wn ) − bn (f(h)) ˛ >< 00 on { − L} × [T1 , .),

on {L} × [T1 , .).
(3.52)
Here, once more, we can apply Lemma 2.5 to the equation “t w=

“ 2x (b n−1 (w)) for w=bn (Wn ) by considering f(s) — b n−1 (s) and see that for
each t ¥ (T1 , .) the number of zeroes of bn (Wn )( · , t) − bn (f(h)) on
[ − L, L] equals 1 and
˛ “ (bsatisfying
x (W ) − b (f(h)))(x, t) < 0
n n n for a unique x ¥ (−L, L)
(b (W ) − b (f(h)))(x, t)=0.
n n n (3.53)
Consequently, since “x (bn (Wn ) − bn (f(h)))=b −n (Wn ) “x Wn , so (3.53) together

with (3.51) completes the proof. L
Furthermore we have:
Lemma 3.7. For each (h, t) ¥ ([ − e1 , 0) 2 (0, e1 ]) × [T1 , .), there exists
a unique x ¥ (−L, L) satisfying v(x, t)=h. Furthermore “x v(x, t) < 0 for
such x. Here T1 > 0 and e1 > 0 are the numbers given in Lemma 3.4.
Proof. Lemma 3.4 guarantees that for t=T1 the statement is true. First
of all, we take h ¥ (0, e1 ]. Let N(t) be the number of sign changes of
v( · , t) − h on [ − L, L] for each t ¥ [T1 , .). Observe that the constant h is a
trivial solution of the equation of (1.5), N(T1 )=1, and
v−h ˛ >< 00 on { − L} × [T1 , .),

on {L} × [T1 , .).
(3.54)
Therefore it follows from Lemma 2.5 that
N(t)=1 for any t ¥ [T1 , .). (3.55)
On the other hand, as in the proof of Lemma 2.5, putting Û=f(v) − f(h)
yields that N(t) equals the number of sign changes of Û( · , t) on [ − L, L]
for each t F T1 . By Lemma 2.4 and the fact that h > 0, we see that Û is
regarded as a classical solution of a uniformly parabolic equation similar to
(2.37) for Un in each compact set contained in the set {v > 0}. With the help
of (3.55) we can show
if Û(x, t)=0 for some (x, t) ¥ QT1 , . , then “x Û(x, t) < 0. (3.56)
Indeed, let Û(x̃, t̃)=0 for some (x̃, t̃) ¥ QT1 , . . Then we notice that v(x̃, t̃)=
h > 0. Suppose that “x Û(x̃, t̃) F 0. When “x Û(x̃, t̃) > 0, in view of (3.54) we
immediately get N(t̃) F 3. This contradicts (3.55). When “x Û(x̃, t̃)=0, we
have
Û(x̃, t̃)=“x Û(x̃, t̃)=0. (3.57)
Since Û(x− (t̃), t̃)=0 − f(h) < 0, so it follows from (3.54) and the conti-
nuity of Û that there exist a small m > 0 and a rectangle R=(−L, x− (t̃)
− m) × [t̃ − m, t̃+m] satisfying:
˛ Û(−L,
(x̃, t̃) ¥ R,
t) > 0
b … {v > 0} 5 ([ − L, L) × (T1 , .)),
R
and Û(x− (t̃) − m, t) < 0 for any t ¥ [t̃ − m, t̃+m].
Note that Û is regarded as a classical solution on R and is continuous on

b . Hence, by using the theory concerning the zero sets of solutions of
R
one-dimensional uniformly parabolic equations [An 1, Theorem B with
Theorem D, pp. 80–82] again as in the proof of Lemma 2.5, from (3.57) we
see that there exist y > 0, e > 0, and at least two points, say xj ¥
(x̃ − e, x̃+e)(j=1, 2), such that
Û(xj , t̃ − y)=0 and “x Û(xj , t̃ − y) ] 0 for j=1, 2.
This together with (3.54) implies that N(t̃ − y) F 3, which contradicts (3.55).
Consequently we see that (3.56) holds.
Condition (3.56) together with (3.55) implies that for each t > T1 the
function Û( · , t) has only one zero, say x ¥ (−L, L), and “x Û(x, t) < 0.
Namely, we conclude that for each (h, t) ¥ (0, e1 ] × [T1 , .), there exists a
unique x ¥ (−L, L) satisfying v(x, t)=h, and furthermore “x v(x, t) < 0 for
such x. Since the similar argument is applicable to (h, t) ¥ [ − e1 , 0)
× [T1 , .), we come to the conclusion of Lemma 3.7. L
In view of Lemma 3.6, as in [BH] for each n F 3 let us introduce the

new independent variable y and the function Xn =Xn (y, t) by
x=Xn (y, t) and Wn (Xn (y, t), t)=y for (y, t) ¥ J × [T1 , .), (3.58)
where we put
J=[f(−e1 ), f(e1 )]. (3.59)
Then we have:
Lemma 3.8. For each n F 3, Xn ¥ C .(J × [T1 , .)). Furthermore, for

each t3 > t2 > T1 there exists a constant C > 0 depending only on t2 and t3
such that for any n F 3
|“t Xn | E C in J × [t2 , t3 ].
Proof. The smoothness of Xn follows from the implicit function

theorem. By (3.58) we get the following as in [BH, Proof of Lemma 2.3,
p. 118]: in J × (T1 , .)
˛ “y Xn =
1
“x Wn
“W
< 0,
“2 W
“t Xn = − t n = − − x n
“x Wn b n (Wn ) “x Wn
=− −
1
b n (y)
“y
1
“y Xn
. 1 2
(3.60)
Let t3 > t2 > T1 . As in [BH] put
1
q= t −
T1 +t2
2
2 “ X − 12 X .
t n n (3.61)
Then q satisfies
“ 2y q “t Xn “y q
“t q= − 2−2 in J × (T1 , .). (3.62)
b n (y)(“y Xn ) “y Xn
Let us apply the maximum principle to q in J × [(T1 +t2 )/2, t3 ]. In view of

Remark 3.5, it follows from Lemma 2.4, Lemma 3.6, and Lemma 3.7 that
there exists a constant C0 > 0 independent of n F 3 satisfying
|q| E C0 on {f(−e1 ), f(e1 )} × 5T +t2 , t 6.

1 2
3 (3.63)
Also, since q=− 12 Xn on J × {(T1 +t2 )/2}, so |q(y, (T1 +t2 )/2)| E 12 L for
any y ¥ J. Therefore, it follows from this inequality, (3.62), (3.63), and the
maximum principle that
|q| E max{C0 , 12 L} in J × 5T +t2 , t 6 .

1 2
3
This completes the proof, since |Xn | E L. L

In view of Lemma 3.7 let us introduce the function X=X(y, t) for
y ¥ J 0 {0} by
x=X(y, t) and f(v)(X(y, t), t)=y for (y, t) ¥ (J 0 {0}) × [T1 , .).
(3.64)
Then we have:
Lemma 3.9. The following hold:

(1) Xn Q X as n Q . in C 2, 1(E) for any compact set E in (J 0 {0})
× (T1 , .).
(2) For each t3 > t2 > T1 there exists a constant C > 0 depending only
on t2 and t3 such that
|“t X| E C in (J 0 {0}) × [t2 , t3 ].
(3) X(y, t) ‘ x− (t) as y a 0, and X(y, t) a x+(t) as y ‘ 0 uniformly on

any compact interval in (T1 , .).
(4) Both x=x− (t) and x=x+(t) are locally Lipschitz continuous on
(T1 , .).
Proof. Take any compact set E in (J 0 {0}) × (T1 , .). Then it follows
from Lemma 2.4 with Remark 3.5, Lemma 3.6, and Lemma 3.7 that
{|(“x Wn )(Xn (y, t), t)|; n F 3} is uniformly bounded and uniformly bounded
away from zero on E, and (1) holds. (2) follows from (1) and Lemma 3.8.
Finally, (3) and (4) follows from (2) and Lemma 3.7. L
In view of Lemma 3.2 and Remark 3.3, in order to prove (1) of Theorem 1.1
it remains to show that x− (t)=x+(t) for any t > 0 when > d0 f −(s)/s ds=
− > 0−d f −(s)/s ds=. for some d > 0. Now, with the help of (4) of Lemma
3.9, in this case we prove:
Lemma 3.10. When > d0 f −(s)/s ds= − > 0−d f −(s)/s ds=. for some d > 0,
one has
x− (t)=x+(t) for any t > 0.
Furthermore, if one puts x(t)=x− (t) for any t > 0 and x(0)=x0 , then x(t) is
continuous on [0, .) and (1) of Theorem 1.1 holds.
Proof. When > d0 f −(s)/s ds= − > 0−d f −(s)/s ds=. for some d > 0, by (1)
of Lemma 3.4 we can choose T1 > 0 as small as we want. Hence combining
this fact with (4) of Lemma 3.9 yields that both x− (t) and x+(t) are locally
Lipschitz continuous on (0, .). Suppose that there exists t0 > 0 satisfying
x− (t0 ) < x+(t0 ). Then it follows from the continuity of x− (t) and x+(t) that
there exist a nonempty interval [y1 , y2 ] … (0, .) with t0 ¥ (y1 , y2 ) and a
nonempty interval (x1 , x2 ) such that
x− (t) < x1 < x2 < x+(t) for any t ¥ [y1 , y2 ]. (3.65)
Hence, by putting w=max{v, 0} and using (3.65), we see that w ¥

C([ − L, L] × [y1 , y2 ]) is a weak solution of the problem:
˛ “t w=“ 2x (f(w))
f(w)(−L, t)=a, f(w)(L, t)=0
w(x, 0)=max{v(x, y1 ), 0}
in Qy1 , y2 =(−L, L) × (y1 , y2 ],
for t ¥ (y1 , y2 ],
for x ¥ (−L, L).
On the other hand, since > d0 f −(s)/s ds=. and w F 0, the property of
infinite speed of propagation implies that w > 0 in Qy1 , y2 . This is a
contradiction. L
By Lemma 3.2 and Lemma 3.10, hereafter let us put
x(t)=x− (t)=x+(t) for any t > 0 and x(0)=x0 . (3.66)
Then in addition to (3.64) X(0, t) is defined by
X(0, t)=x(t) for any t F T1 , and X ¥ C(J × [T1 , .)). (3.67)
4. CONTINUOUS DIFFERENTIABILITY OF THE INTERFACE
By the same argument as in [BH, Sect. 3] we get the following three

lemmas:
Lemma 4.1. For each t3 > t2 > T1 there exists a constant C > 0 depending
only on t2 and t3 such that for each n F 3 and for each t ¥ [t2 , t3 ]
|“x Wn (x1 , t) − “x Wn (x2 , t)| E C |bn (Wn )(x1 , t) − bn (Wn )(x2 , t)| (4.1)
for any x1 , x2 ¥ Xn (J, t), where Xn (J, t)=[Xn (f(e1 ), t), Xn (f(−e1 ), t)].
In particular, {|(“x Wn )(Xn (y, t), t)|; n F 3} is uniformly bounded on any
compact set in J × (T1 , .). Here, Wn is the solution of problem (3.49), and Xn
and J are given by (3.58) and (3.59), respectively.
Proof. Observe that
“t Xn “x (bn (Wn ))=−“t (bn (Wn ))=−“ 2x Wn in 0 (Xn (J, t) × {t}).

t > T1
Hence it follows from Lemma 3.8 that for each t3 > t2 > T1 there exists a
constant C > 0 depending only on t2 and t3 such that for each n F 3
C “x (bn (Wn )) E “ 2x Wn E − C “x (bn (Wn )) in 0 (Xn (J, t) × {t}).

t ¥ [t2 , t3 ]
Integrating this with respect to x gives the desired inequality (4.1). Ob-
serve that the right-hand side of (4.1) is bounded because of (2.13) with
Remark 3.5. Therefore, (4.1) implies that {|(“x Wn )(Xn (y, t), t)|; n F 3} is
uniformly bounded on any compact set in J × (T1 , .), since this set is uni-
formly bounded on any compact set in {f(−e1 )} × (T1 , .) as is mentioned
in the proof of Lemma 3.9. The proof is completed. L
With the help of Lemma 4.1 we have:
Lemma 4.2. “x (f(v)) ¥ C(QT1 , . ).
Proof. In view of (3.66) and Lemma 2.4 with Remark 3.5, it suffices to
show the partial differentiability of f(v) and the continuity of “x (f(v)) at
any point (x(t), t) on the interface x=x(t) for t > T1 . Take t0 > T1
arbitrarily. It follows from (1) of Lemma 3.9 that there exist e2 > 0 and
y > 0 with t0 − y > T1 such that for any t ¥ [t0 − y, t0 +y]
x(t) ¥ (x(t0 ) − e2 , x(t0 )+e2 ) … 3 Xn (J, t). (4.2)

nF3
Observe that in the right-hand side of inequality (4.1) the family

{bn (Wn ); n F 3} is equicontinuous on any compact set in QT1 , . , since by the
result of [Sac, Theorem 1.1, p. 389] the uniformly bounded family {Wn } is
equicontinuous on any compact set in QT1 , . and the family {bn } is also
equicontinuous on any bounded interval of R (see (2.9)). Therefore, by
putting [t2 , t3 ]=[t0 − y, t0 +y] in Lemma 4.1 we see that for each
t ¥ [t0 − y, t0 +y] the family {“x Wn ( · , t); n F 3} is equicontinuous and
uniformly bounded on [x(t0 ) − e2 , x(t0 )+e2 ]. Hence it follows from (2.14)
with Remark 3.5, (4.1), and (4.2) that for each t ¥ [t0 − y, t0 +y], f(v)( · , t)
is differentiable with respect to x on [x(t0 ) − e2 , x(t0 )+e2 ], “x Wn ( · , t) Q
“x (f(v))( · , t) as n Q . uniformly on [x(t0 ) − e2 , x(t0 )+e2 ], and for any
x1 , x2 ¥ [x(t0 ) − e2 , x(t0 )+e2 ]
|“x (f(v))(x1 , t) − “x (f(v))(x2 , t)| E C |v(x1 , t) − v(x2 , t)|.
This inequality implies that the family {“x (f(v))( · , t); t ¥ [t0 − y, t0 +y]} is
equicontinuous on [x(t0 ) − e2 , x(t0 )+e2 ]. Combining this equicontinuity
with both the continuity of “x (f(v)) in {(x, t) ¥ [x(t0 ) − e2 , x(t0 )+e2 ] ×
[t0 − y, t0 +y]; x ] x(t)} (see Lemma 2.4) and the continuity of the inter-
face x=x(t) yields the continuity of “x (f(v)) on [x(t0 ) − e2 , x(t0 )+e2 ]
× [t0 − y, t0 +y]. This completes the proof. L
By using Lemma 4.2 we have:
Lemma 4.3. For T1 > 0 given in Lemma 3.4, there exists a unique Tg =
Tg (T1 ) F T1 such that
“x (f(v))(x(t), t) ˛=0
<0
if t ¥ (T1 , Tg ],
if t ¥ (Tg , .).

(1) When > d0 f −(s)/s ds=. or − > 0−d f −(s)/s ds=. for some d > 0, it
holds that Tg =T1 .
(2) When > d0 f −(s)/s ds < . and − > 0−d f −(s)/s ds < . for some d > 0,
x(t)=x(T1 ) for any t ¥ [T1 , Tg ].
Proof. First, let us show that there exists t > T1 such that “x (f(v))
(x(t), t) < 0. Suppose that for some T > T1
“x (f(v))(x(t), t)=0 for any t ¥ (T1 , T]. (4.3)
Then, in view of Lemma 4.2 and the continuity of the interface x=x(t), by
putting v ± =max{ ± v, 0}, we see that v ± ¥ C(QT1 , T ) are weak solutions of
the problems:
˛ “t v+=“ 2x (f(v+))
f(v+)(−L, t)=a, f(v+)(L, t)=0
v+(x, 0)=max{v(x, T1 ), 0}
in QT1 , T =(−L, L) × (T1 , T],
for t ¥ (T1 , T],
for x ¥ (−L, L),
(4.4)
and
˛ “t v− =“ 2x (−f(−v− ))
− f(−v− )(−L, t)=0, −f(−v− )(L, t)=b
v− (x, 0)=max{ − v(x, T1 ), 0}
in QT1 , T ,
for t ¥ (T1 , T], (4.5)
for x ¥ (−L, L),
respectively. For d > 0 let us distinguish the following two cases:

(i) > d0 f −(s)/s ds=. or − > 0−d f −(s)/s ds=..
(ii) > d0 f −(s)/s ds < . and − > 0−d f −(s)/s ds < ..
In case (i) the property of infinite speed of propagation immediately implies
that v+ > 0 or v− > 0 in QT1 , T . Therefore, for any T > T1 (4.3) yields a
contradiction. In case (ii) the property of finite speed of propagation
implies that T must be finite in (4.3). Indeed, if (4.3) holds for any T > T1 ,
then there exists T̃ > T1 such that v+ > 0 in (−L, L) × [T̃, .), which is a
contradiction. Note that in case (ii) if (4.3) holds for T > T1 , then by the
monotonicity of the interfaces of both v+ and v− we have
x(t)=x(T1 ) for any t ¥ (T1 , T]. (4.6)
Thus we have proved that there exists t > T1 such that “x (f(v))(x(t), t) < 0,
since “x (f(v))(x(t), t) E 0.
Next, let us show that “x (f(v))(x(t), t) < 0 for any t > t0 provided
“x (f(v))(x(t0 ), t0 ) < 0 for some t0 > T1 . For this purpose, take any t0 > T1
with “x (f(v))(x(t0 ), t0 ) < 0 and take any T > t0 . Then it follows from
Lemma 3.7 and Lemma 4.2 that there exists e > 0 satisfying
“x (f(v)) E − e < 0 on (X(J, t0 ) × {t0 }) 2 0 (“X(J, t) × {t}),

t ¥ [t0 , T]
(4.7)
where X(J, t)=[X(f(e1 ), t), X(f(−e1 ), t)] and “X(J, t)={X(f(e1 ), t),
X(f(−e1 ), t)}. Since in the proof of Lemma 4.2 we observed that for each
t > T1
“x Wn ( · , t) Q “x (f(v))( · , t) as n Q .
uniformly on some neighborhood of x(t), (4.8)
by Lemma 2.4 with Remark 3.5 and (1) of Lemma 3.9 we see that for
sufficiently large n
“x Wn E − 12 e < 0 on (X(J, t0 ) × {t0 }) 2 0 (“X(J, t) × {t}).

t ¥ [t0 , T]
(4.9)
Since putting Z=“x Wn yields
1 b ' (W ) “ W
“t Z= − “ 2x Z − n − n x 2 n “x Z in QT1 , . ,
b n (Wn ) (b n (Wn ))
it follows from (4.9) and the maximum principle that for sufficiently large n
“x Wn E − 12 e < 0 in 0 (X(J, t) × {t}). (4.10)

t ¥ [t0 , T]
Therefore, by using (4.8) and Lemma 2.4 with Remark 3.5 again and
letting n Q ., we get
“x (f(v)) E − 12 e < 0 in 0 (X(J, t) × {t}).

t ¥ [t0 , T]
Since T > t0 > T1 with “x (f(v))(x(t0 ), t0 ) < 0 are chosen arbitrarily, we

conclude that if “x (f(v))(x(t0 ), t0 ) < 0 for some t0 > T1 , then “x (f(v))
(x(t), t) < 0 for any t > t0 . Therefore Tg =Tg (T1 )( F T1 ) is determined by
Tg =inf{t0 > T1 ; “x (f(v))(x(t), t) < 0 for any t F t0 }. (4.11)
This completes the proof. L

Now, in order to complete the proof of Theorem 1.1, by using the
estimates of Bertsch and Hilhorst [BH, Sect. 4, pp. 123–127] we shall
prove:
Lemma 4.4. The following hold:
x=x(t) ¥ C 1(Tg , .) and “t (f(v)) ¥ C(QT1 , . ),
where T1 and Tg are given in Lemma 3.4 and Lemma 4.3, respectively.
The continuity of “t (f(v)) is not necessary to prove Theorem 1.1, but we
state it, since as a by-product it is easily obtained as in [BH].
Completion of the Proof of Theorem 1.1. Parts (1) and (5) of Theorem
1.1 were already proved by Lemma 3.2 and Lemma 3.10. Let us complete
the proof of Theorem 1.1 by using Lemma 4.4. In view of Lemma 4.4 we
can determine a time TÄ F 0 by
TÄ =inf{Tg (T1 ); T1 ( > 0) has the property P(T1 )}, (4.12)
where Tg =Tg (T1 ) is determined by (4.11) in the proof of Lemma 4.3, and
where P(T1 ) is
˛ there exist two numbers z1 , z2 satisfying:

(i) − L < z1 < x(T1 ) < z2 < L.
(ii) “x v(x, T1 ) < 0 for any x ¥ [z1 , x(T1 )) 2 (x(T1 ), z2 ].
P(T1 )
Since we already know that x− (t)=x+(t)=x(t) for any t > 0, the property
P(T1 ) is equivalent to that T1 has in the former half of Lemma 3.4. (2) of
Theorem 1.1 follows from our determination of TÄ and Lemma 4.4.
Combining Lemma 4.4 with (1) of Lemma 3.4 and (1) of Lemma 4.3 yields
(3) of Theorem 1.1. It remains to show (4) of Theorem 1.1. We consider the
case where > d0 f −(s)/s ds < . or − > 0−d f −(s)/s ds < . for some d > 0. It
suffices to consider the case where TÄ > 0. Suppose that there exists
t1 ¥ (0, TÄ ) such that x(t1 ) ] x(0). Then by (2) of Lemma 3.4 we can take
T1 =t1 . Hence, when > d0 f −(s)/s ds=. or − > 0−d f −(s)/s ds=. for some
d > 0, by (1) of Lemma 4.3 for T1 =t1 we have Tg (t1 )=t1 . This contradicts
the fact that Tg (t1 ) F TÄ > t1 . When > d0 f −(s)/s ds < . and − > 0−d f −(s)/s ds
< ., (2) of Lemma 4.3 and the fact that Tg F TÄ imply that x(t1 )=x(TÄ ).
On the other hand by the continuity of x=x(t) there exists another
t2 ¥ (0, t1 ) such that x(t2 ) ] x(0) and x(t2 ) ] x(t1 ). By the same argument
as above we see that x(t2 )=x(TÄ ). This contradicts the fact that x(t1 )=
x(TÄ ). Consequently, we conclude that x(t)=x(0) for any t ¥ [0, TÄ ],
which completes the proof of (4) of Theorem 1.1. L
To prove Lemma 4.4 let us use notations similar to those in [BH]. Take
y > 0 and T > Tg +2y arbitrarily. In order to prove that x( · ) ¥ C 1(Tg , .), it
suffices to show that x( · ) ¥ C 1(Tg +2y, T). In view of Lemma 3.6, Lemma
4.1, Lemma 2.4 with Remark 3.5, and (4.10) with Lemma 4.3, we conclude
that the family {|“x Wn (Xn (y, t), t)|; n F 3} is uniformly bounded and
uniformly bounded away from zero in J × [Tg +y, T]. Therefore the first
equality of (3.60) implies that for some d2 > d1 > 0 independent of n F 3
1
d1 < |“y Xn |= < d2 in J × [Tg +y, T]. (4.13)
|“x Wn |
As in [BH, Lemma 4.2, pp. 123–124] we have:
Lemma 4.5. There exists a constant C > 0 independent of n F 3 satisfying:
1
2 F b −n (y)(“t Xn ) 2 (y, T) dy+d 2−2 F (“y “t Xn ) 2 dy dt E C. (4.14)
J J × [Tg +y, T]
Proof. Put p=“t Xn . Then by the second equality of (3.60)
“t p= −
1
b n (y)
“y 1 “y p
(“y Xn ) 2
2 in J × (T1 , .). (4.15)
Multiplying (4.15) by b −n (y) p and integrating by parts over J × [Tg +y, T]

yield
1
2 F b −n (y) p 2(y, T) dy+F (“y p) 2 (“y Xn ) −2 dy dt
J J × [Tg +y, T]
T
=12 F b −n (y) p 2(y, Tg +y) dy+F [p “y p(“y Xn ) −2] y=f(e1)
y=f(−e1 ) dt.
J Tg +y
(4.16)
By using (4.13) for the second term in the left-hand side of (4.16), we see
that the left-hand side of (4.14) is bounded by that of (4.16). By Lemma 3.8
the first term in the right-hand side of (4.16) is bounded by
1
2 C 2 F b −n (y) dy=12 C 2{bn (f(e1 )) − bn (f(−e1 ))}.
J
Since by using (3.60) we have
“y p=−(“x Wn ) −2 {p “ 2x Wn +“x “t Wn }, (4.17)
therefore it follows from Lemma 3.6, Lemma 3.7, and Lemma 2.4 with
Remark 3.5 that |p “y p(“y Xn ) −2| is bounded by a constant independent of
n F 3 on “J × [Tg +y, T]. Consequently, the right-hand side of (4.16) is
bounded above by a constant independent of n F 3, and this completes the
proof. L
By using this lemma, by the same proof as in [BH] we get:
Lemma 4.6. There exist two constants e > 0 and C > 0 both of which are
independent of n F 3 such that [ − e, e] … J and for any t ¥ [Tg +2y, T]
e 1 t e 1
F (“y “t Xn ) 2 (y, t) dy+ 2 F F (“ 2 “ X ) 2 dy dt E C. (4.18)
−e 2d 2 Tg +2y −e b −n (y) y t n
Proof. The proof is the same as in [BH, Proof of Lemma 4.3,

pp. 124–126]. In the proof, for q=“y p=“t “y Xn , the authors used the fact
that {|q|; n F 3} and {|“y q|; n F 3} are uniformly bounded on any compact
set in (J 0 {0}) × (T1 , .). This fact is shown as follows: Since {|q|; n F 3}
was dealt with in the proof of Lemma 4.5 (see (4.17)), it suffices to consider
{|“y q|; n F 3}. By differentiating (4.17) with respect to y we get
“y q=2(“x Wn ) −4 “ 2x Wn {p “ 2x Wn +“x “t Wn }
− (“x Wn ) −2 {“y p “ 2x Wn +p “ 3x Wn (“x Wn ) −1+“ 2x “t Wn (“x Wn ) −1}.
Therefore it follows from Lemma 3.6, Lemma 3.7, and Lemma 2.4 with
Remark 3.5 that {|“y q|; n F 3} is uniformly bounded on any compact set in
(J 0 {0}) × (T1 , .). L
Proof of Lemma 4.4. In [BH] the regularity result of Gilding [Gil 1] is
used to show that “t X is Hölder continuous with respect to t in [ − e, e]
× [Tg +2y, T]. The locally Lipschitz continuity of f in R is necessary to
apply the result of [Gil 1]. So in our situation we cannot use it.
By Lemma 4.6 for each n F 3 and for each t ¥ [Tg +2y, T] we have
|“t Xn (y2 , t) − “t Xn (y1 , t)| E F : y2
y1
“y “t Xn (y, t) dy :
E:F |“ “ X (y, t)| dy : |y − y |
1
y2 2 1
2 2
y t n 2 1
y1
1 1
E C 2 |y2 − y1 | 2 (4.19)
for any y1 , y2 ¥ [ − e, e]. Hence, in view of (1) of Lemma 3.9, by letting

n Q . we get for each t ¥ [Tg +2y, T]
1 1
|“t X(y2 , t) − “t X(y1 , t)| E C 2 |y2 − y1 | 2 for any y1 , y2 ¥ [ − e, 0) 2 (0, e].
(4.20)
Take a sequence {yj } in [ − e, 0) 2 (0, e] such that yj Q 0 as j Q .. Then

inequality (4.20) implies that {“t X(yj , · )} is a Cauchy sequence in
C([Tg +2y, T]). Since by (3) of Lemma 3.9 X(yj , · ) Q x( · )( — X(0, · )) as
j Q . in C([Tg +2y, T]), so we conclude that x( · ) ¥ C 1([Tg +2y, T]) and
“t X(yj , · ) Q “t x( · ) as j Q . in C([Tg +2y, T]). Since y > 0 and T > Tg +2y
are chosen arbitrarily, we see that x( · ) ¥ C 1(Tg , .) and “t X ¥ C(J ×
(Tg , .)).
It remains to show that “t (f(v)) ¥ C(QT1 , . ). Since we already know
that X ¥ C 1(J × (Tg , .)) and “y X < 0, so in view of the relation
x=X(f(v)(x, t), t) by the implicit function theorem together with Lemma
2.4, f(v) ¥ C 1(QTg , . ). Therefore, in view of Lemma 4.3, it suffices to
consider the case where Tg > T1 , > d0 f −(s)/s ds < . and − > 0−d f −(s)/s ds < .
for some d > 0, and x(t)=x(T1 ) for any t ¥ [T1 , Tg ]. Observe that
“t (f(v))(x, t)=−(“t X)(f(v)(x, t), t) “x (f(v))(x, t) in A, (4.21)
where we put A=(1t > T1 (X(J, t) × {t})) 0 (1t ¥ (T1 , Tg ] {(x(T1 ), t)}). Then it
follows from (2) of Lemma 3.9 and Lemma 4.2 with Lemma 4.3 that for
any t0 ¥ (T1 , Tg ]
lim “t (f(v))(x, t)=0. (4.22)

(x, t) Q (x(t0 ), t0 ) with (x, t) ¥ A
On the other hand, since x(t)=x(T1 ) and f(v)(x(T1 ), t)=0 for any
t ¥ [T1 , Tg ], so for any t ¥ (T1 , Tg ) “t (f(v))=0 at (x(t), t) and the left deriv-
ative of f(v) with respect to t at (x(T1 ), Tg ) equals zero. Also, (4.22)
implies that the right derivative with respect to t at (x(T1 ), Tg ) equals zero.
Consequently we get “t (f(v)) ¥ C(QT1 , . ), which completes the proof of
Lemma 4.4. L
5. THE CURVE OF THE SPATIAL MAXIMUM POINTS
As in Section 1 (Introduction), let u=u(x, t) be the unique bounded

weak solution of the Cauchy problem for the one-dimensional evolution
p-Laplacian equation with p > 1 and p ] 2
“t u=“x (|“x u| p − 2 “x u) in R × R+ and u(x, 0)=u0 (x) in R, (1.1)
where u0 is a nonzero, nonnegative, and bounded function in R having

compact support, supp u0 . Then u and “x u are Hölder continuous in R × R+
(see the book of DiBenedetto [dB]). Let C(t) be the set of spatial critical
points of u where u is positive:
C(t)={x ¥ R; “x u(x, t)=0 and u(x, t) > 0}. (1.2)
Then it was shown in [Sak 1, Sak 2] that C(t) is contained in the closed
convex hull of supp u0 for any t > 0, and there exist a time T > 0 and a
continuous function x=x(t) on [T, .) such that C(t)={x(t)} for each
t F T. Namely, after a finite time for each time the set C(t) consists of the
unique spatial maximum point of u( · , t). (References [Sak 1, Theorem 2,
p. 80; Sak 2, Theorem 1, p. 1051] dealt with the cases where p > 2 and
1 2, because of the
property of finite speed of propagation, we have for any t F T
{x ¥ R; u(x, t) > 0}=(z1 (t), z2 (t)) (5.1)
for some uniformly Lipschitz continuous functions z1 (t) and z2 (t) on

[T, .). Moreover − z1 (t) and z2 (t) are monotone nondecreasing on [T, .)
(see [EV, Sect. 4] for the regularity of z1 (t) and z2 (t)). The purpose of this
section is to prove:
Theorem 5.1. There exists a time TÄ F T satisfying:
(1) x( · ) ¥ C 1(TÄ , .).

(2) When 1 2, x(t)=x(T) for any t ¥ [T, TÄ ].
Let f(s)=|s| p − 2 s for s ¥ R, and put v=“x u. Then v ¥ C(R × R+) satisfies
“t v=“ 2x (f(v)) in R × R+. (5.2)

Note that v ¥ C .({v ] 0}) by the regularity theory of uniformly parabolic

equations ([LSU, Fr]). Furthermore it follows from (5.1) that when p > 2,
for any t F T
˛ {x ¥ R; v(x, t) > 0}=(z1 (t), x(t)),

{x ¥ R; v(x, t) < 0}=(x(t), z2 (t)).
(5.3)
On the other hand, because of the property of infinite speed of propa-

gation, when 1 0}=(−., x(t)),

{x ¥ R; v(x, t) < 0}=(x(t), .).
(5.4)
Instead of Lemma 3.4 we have:
Lemma 5.2. There exist two numbers Ta , Tr ¥ (T, .) such that one can
find two points za , zr with za < x(Ta ) and zr > x(Tr ) satisfying
“x v(x, Ta ) < 0 for any x ¥ [za , x(Ta )), and

“x v(x, Tr ) < 0 for any x ¥ (x(Tr ), zr ].

(1) When 1 2, if x(t1 ) ] x(T) for some t1 > T, then one can take
Ta =Tr ¥ (T, t1 ].
Proof. Let us show only that “x v(x, Ta ) < 0 for any x ¥ [za , x(Ta )), since
the other one can be shown similarly. We distinguish the following two
cases:
(i) p > 2.
(ii) 1 < p < 2.
Let us consider case (i) first. In view of (5.3), by taking a small positive
number s with 0 < s E 14 min{x(T) − z1 (T), z2 (T) − x(T)}, we set
x1 =z1 (T)+s, x2 =x(T) − s, and x3 =z2 (T) − s. (5.5)
Note that
z1 (T) < x1 < x(T) − 2s < x2 < x(T) < x(T)+2s < x3 < z2 (T).
Take g ¥ G in Lemma 2.1 such that
˛ supp g( · − x ) … (x − s, x(T)),
2
g(x − x ) < f(v)(x, T)

2
2
for any x ¥ (z1 (T), x(T)).

(5.6)

x2 , T, g) given in Remark 2.2. Then we have
v(x, T) − vsym (x, T) ˛ <> 00 if x ¥ [x1 , x(T)),

if x ¥ (x(T), x3 ].
(5.7)
Here it follows from (8) of Lemma 2.1 that there exists a real-valued,
continuous, and monotone nondecreasing function g(t) on [T, .)
satisfying limt Q . g(t)=. and
{x ¥ R; vsym (x, t) > 0}=(x2 − g(t), x2 +g(t)) for any t F T. (5.8)
Therefore we can define a finite time t̂ > T by
t̂=sup{t ¥ (T, .); x2 +g(t) E x(T)}. (5.9)
Then, by the monotonicity and continuity of g(t), we have
x2 +g(t) ˛ >E x(T)

x(T) if t ¥ [T, t̂],
if t ¥ (t̂, .).
(5.10)
In view of the continuity of x(t), let us distinguish the following three

cases:
(a) x(t)=x(T) for any t ¥ [T, t̂+1].

(b) There exists t+ ¥ (T, t̂+1] such that x(T) < x(t+) and x(t) ¥
[x(T) − s, x(T)+s] for any t ¥ [T, t+].
(c) There exists t− ¥ (T, t̂+1] such that x(T) > x(t− ) and x(t) ¥
[x(T) − s, x(T)+s] for any t ¥ [T, t− ].
Consider case (a) first. In case (a), we have from (5.7), (5.8), and (5.10)
˛ zv (t)(x,< T)x << xv(x,− g(t)

1
sym
1 2
T)
< x +g(t) E x(T)=x(t)
2
for any x ¥ [x , x(T)).

1
for any t ¥ [T, t̂],
(5.11)
Then, applying Lemma 2.3 to vsym and v in the rectangle (x1 , x(T)) × (T, t̂]
yields
vsym E v in [x1 , x(T)] × [T, t̂]. (5.12)
Since f(v) − f(vsym ) is regarded as a classical solution of a uniformly

parabolic equation similar to (2.37) for Un in each compact set contained in
the set {vsym > 0 and v > 0}, so it follows from (5.12) and the strong
maximum principle (see [Fr, RR]) that in particular
vsym (x, t̂) < v(x, t̂) for any x ¥ [x1 , x(T)). (5.13)
Therefore, in view of (5.10), (5.11), and (5.13), by the continuity we see that
there exists Ta ¥ (t̂, t̂+1) satisfying
˛ v(x2 , Ta ) > vsym (x2 , Ta )=max vsym (y, Ta ) > 0,

y¥R
z1 (Ta ) < x1 < x2 − g(Ta ) < x2 < x(Ta )=x(T) < x2 +g(Ta ) < x3 < z2 (Ta ).
(5.14)
Let N(t) be the number of sign changes of v( · , t) − vsym ( · , t) on [x1 , x3 ] for

each t ¥ [T, Ta ]. (5.7) means that N(T)=1. In view of the first line of
inequalities of (5.11), the second line of inequalities of (5.14), and the
monotonicity of three functions z1 (t), z2 (t), and g(t), we observe that
v − vsym ˛ >< 00 on {x1 } × [T, Ta ],

on {x3 } × [T, Ta ].
(5.15)
Therefore, with the help of (5.7) and (5.15), by using Lemma 2.5 on the
rectangle (x1 , x3 ) × (T, Ta ], we have
N(t)=1 for any t ¥ [T, Ta ]. (5.16)
Hence, in view of (5.14), by following the proof of Lemma 3.4 (from

(3.27) to (3.31)), we see that there exists za ¥ (x2 , x(Ta )) with v(za , Ta )=
vsym (za , Ta ) > 0, and moreover
˛ “0 <v(zv(x,, TT) )E<“vv

x a a
a
x
(x, Ta )
sym for any x ¥ (za , x(Ta )),
sym (za , Ta ) < 0.
(5.17)
Introducing the family of the symmetric nonnegative weak solutions

{v lsym ; l ¥ [0, 1]} given by
v lsym (x, t)=v(x, t; x2 , T, lg),
and following the proof of Lemma 3.4 (from (3.32) to (3.37)), yield
“x v(x, Ta ) < 0 for any x ¥ [za , x(Ta )). (5.18)
This completes the proof in case (a).

Although cases (b) and (c) are similar to cases (A) and (B) in the proof of
Lemma 3.4, respectively, in view of the behaviour of v and vsym on the
lateral boundaries, we need more careful consideration. Consider case (b).
In this case we first put
Ta =t+. (5.19)
Remark that x(t) ¥ (x1 , x3 ) for any t ¥ [T, Ta ]. By letting r=14 (x(Ta ) −
x(T)) (¥ (0, 14 s]), we set
x4 =x(T)+3r(=x(Ta ) − r). (5.20)
Note that 12 (x(T)+x(Ta )) < x4 < x(Ta ). Take another g ¥ G in Lemma 2.1
such that
˛ supp g( · − x4 )=[x4 − 2r, x4 +2r]( … (x(T), x3 )),

max g=g(0) < f(v)(x4 , Ta ).
R
(5.21)

x4 , T, g) given in Remark 2.2. Then we have (5.7) here. In view of (8) of
Lemma 2.1, by choosing maxR g smaller if necessary, we get
{x ¥ R; vsym (x, t) > 0}=(x4 − g(t), x4 +g(t)) … (x1 , x3 ) for any t ¥ [T, Ta ]
(5.22)
for some real-valued, continuous, and monotone nondecreasing function

g(t) on [T, .). Then we have
v − vsym ˛ >< 00 on {x1 } × [T, Ta ],

on {x3 } × [T, Ta ].
(5.23)

each t ¥ [T, Ta ]. (5.7) means that N(T)=1. In view of (5.23), by using
Lemma 2.5 as in case (a), we see that (5.16) also holds. Then, in view of
(5.21), (5.20), and (1), (8) of Lemma 2.1, we observe that
0 < max vsym (y, Ta )=vsym (x4 , Ta ) < v(x4 , Ta ) and

y¥R
vsym (x(Ta ), Ta ) > 0=v(x(Ta ), Ta ). (5.24)
Therefore, as in case (a), from (5.16) we see that there exists za ¥ (x4 , x(Ta ))
with v(za , Ta )=vsym (za , Ta ) > 0, and moreover (5.17) also holds. Hence,
introducing the family {v lsym ; l ¥ [0, 1]} given by v lsym (x, t)=v(x, t;
x4 , T, lg), and following the proof of Lemma 3.4, yield the conclusion
(5.18). This completes the proof in case (b).
Let us proceed to case (c). In this case we first put
Ta =t− . (5.25)
Remark that x(t) ¥ (x1 , x3 ) for any t ¥ [T, Ta ]. By letting r=14 (x(T) −
x(Ta ))( ¥ (0, 14 s]), we set in this case
x4 =x(Ta ) − r. (5.26)
Note that x4 < x(Ta ) < x4 +2r < x(T). Take another g ¥ G in Lemma 2.1
such that
˛ supp g( · − x4 )=[x4 − 2r, x4 +2r]( … (x1 , x(T))),

max g=g(0) < f(v)(x4 , Ta ),
R
g(x − x4 ) < f(v)(x, T) for any x ¥ [x1 , x(T)).

(5.27)

x4 , T, g) given in Remark 2.2. Then we have (5.7) here. In view of (8) of
Lemma 2.1, by choosing maxR g smaller if necessary, we get (5.22) and
(5.23) here. Following the proof in case (b) yields that (5.16) also holds.
Then, in view of (5.27), (5.26), and (1), (8) of Lemma 2.1, we observe that
(5.24) also holds. Therefore, as in case (a), from (5.16) we see that there
exists za ¥ (x4 , x(Ta )) with v(za , Ta )=vsym (za , Ta ) > 0, and moreover (5.17)
holds. Introducing the family {v lsym ; l ¥ [0, 1]} and following the proof of
Lemma 3.4, yield the conclusion (5.18). This completes the proof in case
(c). Thus we complete the proof of this lemma in case (i) (that is, the case
where p > 2) except (2). Let us show (2). Suppose that x(t1 ) ] x(T) for
some t1 > T. Then, in view of the continuity of x(t), we can find either t+
or t− in (T, min{t̂+1, t1 }] such that either case (b) or case (c) occurs.
Therefore from (5.19) and (5.25) we see that (2) holds.
It remains to consider case (ii), that is, the case where 1 T arbitrarily. In view of the continuity of x(t), take two points x1 , x3
such that
x1 < x(t) < x3 for any t ¥ [T, Ta ]. (5.28)
Then, as in the proof of Lemma 3.4, in view of (5.4) we take g ¥ G in

Lemma 2.1 such that
˛ max g=g(0) < min{f(v)(x1 , t); t ¥ [T, Ta ]},

R
supp g( · − x1 ) … (2x1 − x(T), x(T)),

g(x − x1 ) < f(v)(x, T) for any x ¥ [x1 , x(T)).
(5.29)

x1 , T, g) given in Remark 2.2. Then we have
v(x, T) − vsym (x, T) ˛ >< 00 if x ¥ [x1 , x(T)),

if x ¥ (x(T), x3 ].
(5.30)

each t ¥ [T, Ta ]. (5.30) means that N(T)=1. Since by (1) of Lemma 2.1 we
have 0 E vsym E f −1(g(0)), so by (5.29), (5.28), and (5.4)
v − vsym ˛ >< 00 on {x1 } × [T, Ta ],

on {x3 } × [T, Ta ].
(5.31)
Therefore, by using Lemma 2.5 as in case (a) of (i), from (5.31) and (5.30)
we see that (5.16) also holds. In view of (9) of Lemma 2.1, we have that
vsym > 0 in R × (T, .). Hence, in particular
0 < max vsym (y, Ta )=vsym (x1 , Ta ) < v(x1 , Ta ) and

y¥R
vsym (x(Ta ), Ta ) > 0=v(x(Ta ), Ta ). (5.32)
Consequently, with the help of (5.16) and (5.32), by the same argument as
in case (a) of (i), we see that there exists za ¥ (x1 , x(Ta )) and the conclusion
(5.18) also holds. Since Ta > T is taken arbitrarily, so (1) holds. This
completes the proof of the lemma. L
In view of Lemma 5.2, let us put
T1 =max{Ta , Tr }. (5.33)
By using Lemma 5.2, instead of Lemma 3.7 we have:
Lemma 5.3. For each T − > T1 one can find two numbers e1 > 0 and d1 > 0
satisfying:
(1) For any t ¥ [T, T −], v(x(t) − d1 , t) > e1 and v(x(t)+d1 , t) < −e1 .
(2) For each (h, t) ¥ ([ − e1 , 0) 2 (0, e1 ]) × [T1 , T −) there exists a
unique x ¥ (x(t) − d1 , x(t)+d1 ) satisfying v(x, t)=h. Moreover “x v(x, t) < 0
for such x.
Proof. Note that T1 > T, since Ta > T and Tr > T. Take T − > T1
arbitrarily. In view of (5.3) and (5.4), we can find d1 > 0 such that
˛ v(x(t) − d , t) > 0
1
v(x(t)+d , t) < 0
1
for any t ¥ [T, T −],
for any t ¥ [T, T −].
(5.34)
Therefore, it follows from Lemma 5.2 and the continuity of v and x(t) that
there exist e1 > 0, ca , and cr with x(Ta ) − d1 < ca < x(Ta ) and x(Tr ) < cr <
x(Tr )+d1 such that
˛ v(x(t) − d1 , t) > e1
v(x(t)+d1 , t) < − e1
v(x, Ta ) > e1
“x v(x, Ta ) < 0
v(x, Tr ) < − e1
“x v(x, Tr ) < 0
if t ¥ [T, T −],
if t ¥ [T, T −],
if x ¥ [x(Ta ) − d1 , ca ),
if x ¥ [ca , x(Ta )),
if x ¥ (cr , x(Tr )+d1 ],
if x ¥ (x(Tr ), cr ].
(5.35)
The first and second inequalities of (5.35) imply (1) of the lemma. Let
h ¥ (0, e1 ]. Since the constant h is a trivial solution of (5.2), by putting
U=f(v) − f(h) we let N(t) be the number of zeroes of U( · , t) on
[x(t) − d1 , x(t)] for each t ¥ (T, T −). Then it follows from the third and
fourth inequalities of (5.35) that N(Ta )=1. Moreover, from the first
inequality of (5.35) and the definition of x(t), we have
U(x(t) − d1 , t) > 0 and U(x(t), t)=−f(h) < 0 for any t ¥ [T, T −].
(5.36)
Put
S= 0 ((x(t) − d1 , x(t)) × {t}).

t ¥ (T, T −)
Then, since U is continuous on S̄ and it is regarded as a classical solution

of a uniformly parabolic equation similar to (2.37) for Un in each compact
set contained in S, so by using the theory concerning the zero sets of
solutions of one-dimensional uniformly parabolic equations of Angenent
[An 1, Theorem D, p. 82], from (5.36) we see that N(t) is finite for each
t ¥ (T, T −) and that if U(x, t)=“x U(x, t)=0 for some (x, t) ¥ S, then
N(t1 ) > N(t2 ) for any t1 ¥ (T, t) and for any t2 ¥ (t, T −). In particular, N(t)
is monotone nonincreasing in t on (T, T −). (Precisely, finitely many times
we used Theorem D of [An 1] on closed rectangles contained in S such
that U has no zeros on their lateral boundaries.) Thus we get
N(t)=N(Ta )=1 for any t ¥ [Ta , T −), (5.37)
and moreover for each t ¥ (Ta , T −)
“x U(x, t) < 0 for a unique x ¥ (x(t) − d1 , x(t)) satisfying U(x, t)=0.

(5.38)
This together with Lemma 5.2 means that for each (h, t) ¥ (0, e1 ] × [Ta , T −)
there exists a unique x ¥ (x(t) − d1 , x(t)) satisfying v(x, t)=h, and more-
over “x v(x, t) < 0 for such x. Similarly we can see that for each (h, t) ¥
[ − e1 , 0) × [Tr , T −) there exists a unique x ¥ (x(t), x(t)+d1 ) satisfying
v(x, t)=h, and moreover “x v(x, t) < 0 for such x. From (5.33) this implies
(2) and completes the proof. L
Proceeding as in the proof of Theorem 1.1 yields:
Lemma 5.4. “x (f(v)) ¥ C((R × (T1 , .)) 5 {u > 0}).

Proof. It suffices to show the partial differentiability of f(v) and the
continuity of “x (f(v)) at any point (x(t), t) on the interface x=x(t) for
t > T1 . Take t0 > T1 arbitrarily and take T − > t0 . Then by Lemma 5.3 there
exist two numbers e1 > 0 and d1 > 0 satisfying (1) and (2) in Lemma 5.3. By
the continuity of v and x(t), we find a rectangle R=(x(t0 ) − d2 , x(t0 )
+d2 ) × (t0 − s, t0 +s] for some d2 ¥ (0, d1 ) and s > 0 with [t0 − s, t0 +s] …
(T1 , T −), and we have for any t ¥ [t0 − s, t0 +s]
˛ x(t) − d < x(t ) − d < x(t) < x(t )+d < x(t)+d ,
1
v(x(t ) − d , t) > e
0 2
0
and 1
2 0
v(x(t )+d , t) < − e .

0
2
2
1
1
(5.39)
Let us consider approximate solutions for f(v) on R. For this purpose,

take two sequences {G+ − .
n }, {G n } in C ([t0 − s, t0 +s]) such that
˛ G n± ( · ) Q f(v)(x(t0 ) ± d2 , · )
[t0 − s, t0 +s] respectively,
G n± =f(v)(x(t0 ) ± d2 , t0 − s)
t=t0 − s respectively,
G n− > f(e1 ) and G+
as n Q . uniformly on
in a neighbourhood of
n < f(−e1 ) on [t0 − s, t0 +s].

(5.40)
Let H ¥ C .([x(t0 ) − d2 , x(t0 )+d2 ] × (0, 1]) 5 C([x(t0 ) − d2 , x(t0 )+d2 ] ×

[0, 1]) be the unique classical solution of the problem:
˛ “t H=“ 2x H in (x(t0 ) − d2 , x(t0 )+d2 ) × (0, 1],

H(x(t0 ) ± d2 , t)=f(v)(x(t0 ) ± d2 , t0 − s) for t ¥ (0, 1] respectively,
H(x, 0)=f(v)(x, t0 − s) for x ¥ (x(t0 ) − d2 , x(t0 )+d2 ).
(5.41)
Then, in particular, putting Hn (x)=H(x, 1n ) for each n F 3 yields that
˛ HH (¥· )CQ([x(t
n
n
.
) − d , x(t )+d ]),
0
f(v)( · , t − s)
0
2 0 2
as n Q . uniformly on [x(t ) − d , x(t )+d ].

0 2 0 2
(5.42)
By using the theory of uniformly parabolic equations [LSU] we see that

b ) of the
for n F 3 there exists a unique classical solution Wn ¥ C .(R
problem:
˛ “t (bn (Wn ))=“ 2x Wn

±
in R,
Wn (x(t0 ) ± d2 , t)=G (t) for t ¥ [t0 − s, t0 +s] respectively,
Wn (x, t0 − s)=Hn (x)
n
for x ¥ (x(t0 ) − d2 , x(t0 )+d2 ).

(5.43)
Then by the same argument as in proving (2.14) we get
Wn Q f(v) and bn (Wn ) Q v as n Q . (5.44)
uniformly on any compact set in R. By the same proof as that of Lemma 3.6,
we see that
Fact. For each (h, t) ¥ [ − e1 , e2 ] × [t0 − s, t0 − s), there exists a unique

x ¥ (x(t0 ) − d2 , x(t0 )+d2 ) satisfying Wn (x, t)=f(h). Furthermore “x Wn (x, t)
< 0 for such x.
Hence, for each n F 3, we can introduce the new independent variable y
and the function Xn =Xn (y, t) by
x=Xn (y, t) and Wn (Xn (y, t), t)=y for (y, t) ¥ J × [t0 − s, t0 +s),
(5.45)
where J=[f(−e1 ), f(e1 )]. Also, from Lemma 5.3 we can introduce the
function X=X(y, t) by
x=X(y, t) and f(v)(X(y, t), t)=y for (y, t) ¥ J × [t0 − s, t0 +s).

(5.46)
By proceeding as in the proof of Theorem 1.1, we can obtain the corre-

sponding results to Lemmas 3.8, 3.9, 4.1, and 4.2. Thus we see that “x (f(v)) ¥
C((x(t0 ) − d2 , x(t0 )+d2 ) × (t0 − s, t0 +s)). This completes the proof, since
t0 > T1 is taken arbitrarily. L
By using Lemma 5.4 we have:
Lemma 5.5. For T1 ¥ (T, .) given by (5.33), there exists a unique Tg =

Tg (T1 ) F T1 such that
“x (f(v))(x(t), t) ˛=0
<0
if t ¥ (T1 , Tg ],
if t ¥ (Tg , .).

(1) When 1 2, x(t)=x(T1 ) for any t ¥ [T1 , Tg ].
Proof. By modifying the proof of Lemma 4.3, we can prove this lemma.
Indeed, in order to show that there exists t > T1 such that “x (f(v))
(x(t), t) < 0, it suffice to consider two nonnegative functions v ± =
max{ ± v, 0} and use the well-known properties of the interfaces of non-
negative solutions as in the former half of the proof of Lemma 4.3. Also,
for each t > t0 ( > T1 ), by introducing a chain of a finite number of rec-
tangles R, each of which is given in the proof of Lemma 5.4, such that the
first rectangle contains the point (x(t0 ), t0 ) and the last rectangle contains
(x(t), t), and by using the approximate solutions {Wn } defined on each
rectangle and given in the proof of Lemma 5.4 as in the latter half
of the proof of Lemma 4.3, we can show that if “x (f(v))(x(t0 ), t0 ) < 0 for
some t0 > T1 , then “x (f(v))(x(t), t) < 0 for any t > t0 . Then Tg is

determined by (4.11). L
Finally, since the differentiability is a local character, by introducing a

rectangle R together with the approximate solutions {Wn } as in the proof
of Lemma 5.4, by the same argument as in the proof of Lemma 4.4 we
obtain with the help of Lemma 5.5:
Lemma 5.6. The following hold,
x=x(t) ¥ C 1(Tg , .) and “t (f(v)) ¥ C((R × (T1 , .)) 5 {u > 0}),
where T1 and Tg are given in (5.33) and Lemma 5.5, respectively.
Consequently, combining Lemma 5.6 with Lemma 5.2, Lemma 5.3,

(5.33), and Lemma 5.5, yields Theorem 5.1 as in the proof of Theorem 1.1.
As in the proof of Theorem 1.1 (in particular, see (4.12)), TÄ F T is
determined by
TÄ =inf{Tg (T1 ); T1 ( > T) has the property P̃(T1 )}, (5.47)
where Tg =Tg (T1 ) is given in Lemma 5.5 and where P̃(T1 ) is:
˛ there exist two numbers z1 , z2 satisfying:

(i) z1 < x(T1 ) < z2 .
(ii) “x v(x, T1 ) < 0 for any x ¥ [z1 , x(T1 )) 2 (x(T1 ), z2 ].
P̃(T1 )
In view of Lemma 5.2, (5.33), and Lemma 5.3, we notice that the property
P̃(T1 ) is equivalent to that T1 has in Lemma 5.3.
APPENDIX: PROOF OF LEMMA 2.1
Let us prove Lemma 2.1. Take g ¥ G with
max g E M0 and max |g −| E M1 .

R R
By introducing the approximate solutions, we prove this lemma. Take

R0 > 0 such that the support of g is contained in (−R0 , R0 ) and take n0 ¥ N
such that maxR g+1/n0 < lims Q . f(s). Then, for any R F R0 and n F n0 ,
by the theory of uniformly parabolic equations [LSU], there exists a
unique bounded classical solution un, R ¥ C 2, 1([ − R, R] × [0, .)) of the

initial-boundary value problem:
˛ “t un, R =“ 2x (f(un, R ))
un, R =f −1 1 1n 2
un, R (x, 0)=f −1 g(x)+
1
n
1 2
in (−R, R) × R+,
on { − R, R} × R+,
in (−R, R),
(6.1)
which satisfies
f −1 1 1n 2 E u n, R
1
E max f −1 g(y)+
y¥R
1
n
2 in [ − R, R] × [0, .). (6.2)
Furthermore, if we put w=f(un, R ), then w satisfies
˛ “t w=f −(un, R ) “ 2x w
w=
1
n
w(x, 0)=g(x)+
1
n
in (−R, R) × R+,
on { − R, R} × R+,
in (−R, R).
(6.3)
Then, because of (6.2) by the standard regularity theory of uniformly

parabolic equations (see [Fr, Theorem 12, p. 75] for example), we see that
both w and “x w belong to C 2, 1([ − R, R] × [0, .)). Since g is continuous,
there exists a constant c > 0 such that
1 1
E g(x)+ E min{w+(x), w −(x)} in [ − R, R] for any R F R0 , (6.4)
n n
where w ± (x)=Rc (R ± x)+1n . Since both w+ and w − satisfy the partial

differential equation of (6.3), in view of (6.4) from the comparison principle
we see that
1
E w E min{w+, w −} in [ − R, R] × [0, .) for any R F R0 . (6.5)
n
Then it follows that
c
|“x (f(un, R ))|=|“x w| E on { − R, R} × [0, .) for any R F R0 . (6.6)
R
Therefore, by applying the maximum principle to the function “x w we get
|“x (f(un, R ))| E max max |g −|, 3 R

c
R
4 in [ − R, R] × [0, .) for any R F R0 .
(6.7)
We use the Aleksandrov’s reflection principle (see [GNN, Ma 2, AC] for

example). Let l ¥ [0, R) and define the function u ln, R (x, t)=un, R (2l − x, t).
We observe that
˛ “t u ln, R =“ 2x (f(u ln, R ))
u ln, R =f −1
u
u
l
n, R
l
n, R
=un, R
1 1n 2 E un, R
(x, 0) E un, R (x, 0)

in (2l − R, l) × R+,
on {2l − R} × R+,
on {l} × R , +
in (2l − R, l).
(6.8)
Here (ii) was used to show the last inequality. Therefore, it follows from
the comparison principle that
u ln, R E un, R in [2l − R, l] × [0, .). (6.9)
In particular, if we take l=0, since both the second and the last inequali-
ties of (6.8) become equalities, we get
un, R (x, t)=un, R (−x, t) for any (x, t) ¥ [ − R, R] × [0, .). (6.10)
Also, by the comparison principle and (6.2) we notice that
un1 , R1 E un2 , R2 on [ − R1 , R1 ] × [0, .) provided that n1 F n2 and R1 E R2 .

(6.11)
In view of (6.2), from the standard estimates of [LSU] and the

compactness arguments with the monotonicity (6.11) we get for each n F n0
un, R Q un as R Q . in C 2, 1([ − L, L] × [0, T]) for any L > 0 and T > 0,

(6.12)
where un is the unique bounded classical solution of the Cauchy problem,
“t un =“ 2x (f(un )) in R × R+ and 1 1 2
un (x, 0)=f −1 g(x)+ in R,
n
(6.13)
which satisfies
f −1 1 1n 2 E u E f 1 max g(y)+1n 2 E f 1 M +n1 2

n
−1
y¥R
−1
0
0
in R × [0, .).
(6.14)
Moreover, it follows from (6.7), (6.9), (6.10), and (6.11) that for each n F n0 ,
|“x (f(un ))| E M1 in R × [0, .), (6.15)

l
u E un
n in (−., l] × [0, .) for each l F 0, (6.16)
un (x, t)=un (−x, t) for any (x, t) ¥ R × [0, .), (6.17)
un+1 E un in R × [0, .), (6.18)
where u ln is defined by u ln (x, t)=un (2l − x, t). In view of (6.14) and (6.15),
we can apply an estimate of Gilding [Gil 3, Lemma 4 and its proof,
pp. 180–181] to un for n F n0 and we see that
1 5 1 62 − M |t − t |
1
f −1 max 0, f(un )(x1 , t1 ) − M1 |x1 − x2 | − M1 |t1 − t2 | 2
2 1 1 2
1
2
E f 1 min 5 M + , f(u )(x , t )+ M |x − x |+M |t − t | 62

−1 1 1 1
2
0 n 2 2 1 1 2 1 1 2
n 0 2
(6.19)
for all (x1 , t1 ), (x2 , t2 ) ¥ R × [0, .). Notice that if one puts t1 =t2 , then this
inequality is reduced to (6.15).
Therefore it follows from (6.14) and (6.19) that the sequence {un } . n=n0 is
uniformly bounded and equicontinuous on R × [0, .). By Ascoli-Arzelà
theorem and the monotonicity (6.18) we see that
un Q u as n Q . uniformly on any compact subset of R × [0, .),

(6.20)
where u is the unique nonnegative weak solution of (2.1) for u0 =f. Hence,
(6.14) and (6.19) imply (1) and (2). With the help of (6.20) and the conti-
nuity of u, (3) follows from the standard theory of uniformly parabolic
equations [Fr]. (In particular, in order to obtain the regularity for f(u), we
use the equation “t (f(un ))=f −(un ) “ 2x (f(un )) to get necessary estimates for
f(un ) on each compact set contained in D.) (4) follows from (6.17). (5) is
proved in [Gil 2, Proposition 1, p. 248]. Let us prove (6). It follows from
(6.16) that u( · , t) is monotone nonincreasing on [0, .) for each t > 0.

Furthermore, from (3) we have
“x u E 0 in D 5 (R+) 2. (6.21)
Suppose that there exists (x0 , t0 ) ¥ D 5 (R+) 2 such that “x u(x0 , t0 )=0.
Then it follows from the strong maximum principle (see [Fr, RR]), the
monotonicity of u( · , t), and (3) that “x u(x, t0 )=0 for any x F x0 − e0 with
some e0 > 0 and u(x, t0 ) — u(x0 , t0 ) > 0 for any x F x0 − e0 . This contradicts
(5). The last part of (6) follows from the fact that “t u(0, t)=
“ 2x (f(u))(0, t) E 0.
Let us proceed to the proof of (7). By (4), (5), and (6), we get
1
u(x, t) E F f(x) dx for any (x, t) ¥ R × [0, .). (6.22)
2 |x| R
1
Take any n F n0 . Choose R F R0 sufficiently large such that 2R >R f(x) dx
−1 1 −1 1
< f ( n ). Since (6.22) implies that u < f ( n ) on { − R, R} × R+, by the
comparison principle (see [Gil 3, Theorem 14, p. 204] or Lemma 2.3 in the
present paper) we have
u E un, R in [ − R, R] × [0, .). (6.23)
On the other hand, as is proved in [BP, Proof of Lemma 3, pp. 37–38]

with the help of a Lyapunov functional and the result of [Da], we have
un, R (x, t) Q f −1 1 1n 2 as t Q . uniformly in [ − R, R].
Therefore it follows from this and (6.23) that
lim sup u(0, t) E f −1

tQ.
1 1n 2 . (6.24)
Since n F n0 is chosen arbitrarily and f −1(0)=0, (6.24) and (6) imply (7).
Parts (8) and (9) are proved in [Gil 2] except limt Q . g(t)=.. On the
other hand this limit is obtained by (5) and (7).
REFERENCES
[An 1] S. Angenent, The zero set of a solution of a parabolic equation, J. Reine Angew.
Math. 390 (1988), 79–96.
[An 2] S. Angenent, Nodal properties of solutions of parabolic equations, Rocky
Mountain J. Math. 21 (1991), 585–592.
[AC] D. G. Aronson and L. A. Caffarelli, The initial trace of a solution of the porous
medium equation, Trans. Amer. Math. Soc. 280 (1983), 351–366.
[BH] M. Bertsch and D. Hilhorst, The interface between regions where u < 0 and u > 0
in the porous medium equation, Appl. Anal. 41 (1991), 111–130.
[BP] M. Bertsch and L. A. Peletier, A positivity property of solutions of nonlinear
diffusion equations, J. Differential Equations 53 (1984), 30–47.
[Da] C. M. Dafermos, Asymptotic behavior of solutions of evolution equations, in
‘‘Nonlinear Evolution Equations’’ (M. G. Crandall, Ed.), pp. 103–123, Academic
Press, New York, 1978.
[DK] B. E. J. Dahlberg and C. E. Kenig, Non-negative solutions of generalized porous
medium equations, Rev. Mat. Iberoamericana 2 (1986), 267–305.
[dB] E. DiBenedetto, ‘‘Degenerate Parabolic Equations,’’ Springer-Verlag, New York/
Berlin, 1993.
[DP] C. J. van Duyn and L. A. Peletier, Nonstationary filtration in partially saturated
porous media, Arch. Rational Mech. Anal. 78 (1982), 173–198.
[EV] J. R. Esteban and J. L. Vazquez, Homogeneous diffusion in R with power-like
nonlinear diffusivity, Arch. Rational Mech. Anal. 103 (1988), 39–80.
[Fr] A. Friedman, ‘‘Partial Differential Equations of Parabolic Type,’’ Prentice–Hall,
Englewood Cliffs, NJ, 1964.
[GV] V. A. Galaktionov and J. L. Vazquez, Geometrical properties of the solutions of
one-dimensional nonlinear parabolic equations, Math. Ann. 303 (1995), 741–769.
[GNN] B. Gidas, W. M. Ni, and L. Nirenberg, Symmetry and related properties via the
maximum principle, Comm. Math. Phys. 68 (1979), 209–243.
[Gil 1] B. H. Gilding, Hölder continuity of solutions of parabolic equations, J. London
Math. Soc. 13 (1976), 103–106.
[Gil 2] B. H. Gilding, The occurrence of interfaces in nonlinear diffusion-advection
processes, Arch. Rational Mech. Anal. 100 (1988), 243–263.
[Gil 3] B. H. Gilding, Improved theory for a nonlinear degenerate parabolic equation,
Ann. Scuola Norm. Sup. Pisa Ser. IV 16 (1989), 165–224; Errata-Corrige, 17 (1990),
479.
[Hen] D. B. Henry, Some infinite-dimensional Morse–Smale systems defined by parabolic
partial differential equations, J. Differential Equations 59 (1985), 165–205.
[Hua] Zhongdan Huan, Some basic facts for a class of degenerate parabolic equations,
Differential Integral Equations 5 (1992), 961–972.
[LSU] O. A. Ladyženskaja, V. A. Solonnikov, and N. N. Ural’ceva, ‘‘Linear and
Quasilinear Equations of Parabolic Type,’’ Transl. Math. Monogr., Vol. 23, Amer.
Math. Soc., Providence, 1968.
[Ma 1] H. Matano, Nonincrease of the lap-number of a solution for a one-dimen-
sional semilinear parabolic equation, J. Fac. Sci. Univ. Tokyo Sect. IA 29 (1982),
401–441.
[Ma 2] H. Matano, Asymptotic behaviour of the free boundaries arising in one phase
Stefan problems in multi-dimensional spaces, in ‘‘Lecture Notes,’’ Numer. Appl.
Anal., Vol. 5, pp. 133–151, Kinokuniya, Tokyo, 1982.
[Pe] L. A. Peletier, A necessary and sufficient condition for the existence of an interface
in flows through porous media, Arch. Rational Mech. Anal. 56 (1974), 183–190.
[RR] M. Renardy and R. C. Rogers, ‘‘An Introduction to Partial Differential
Equations,’’ Springer-Verlag, New York/Berlin/Heidelberg, 1992.
[Sac] P. Sacks, Continuity of solutions of a singular parabolic equation, Nonlinear Anal.
7 (1983), 387–409.
[Sak 1] S. Sakaguchi, The number of peaks of nonnegative solutions to some nonlinear
degenerate parabolic equations, J. Math. Anal. Appl. 203 (1996), 78–103.
[Sak 2]S. Sakaguchi, Spatial critical points of nonnegative solutions of the evolution
p-Laplacian equation: The fast diffusion case, Differential Integral Equations 10
(1997), 1049–1063.
[SGKM] A. A. Samarskii, V. A. Galaktionov, S. P. Kurdyumov, and A. P. Mikhailov,
‘‘Blow-up in Quasilinear Parabolic Equations,’’ Nauka, Moscow, 1987 [In
Russian]; English translation, de Gruyter, Berlin, 1995.
[Sat] D. H. Sattinger, On the total variation of solutions of parabolic equations, Math.
Ann. 183 (1969), 78–92.

Regularity of The Interfaces With Sign Changes of Solutions of The One-Dimensional Porous Medium Equation

Uploaded by

Copyright:

Available Formats

Regularity of The Interfaces With Sign Changes of Solutions of The One-Dimensional Porous Medium Equation

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Regularity of The Interfaces With Sign Changes of Solutions of The One-Dimensional Porous Medium Equation

Uploaded by

Copyright:

Available Formats

Journal of Differential Equations 178, 1–59 (2002)

doi:10.1006/jdeq.2000.4002, available online at http://www.idealibrary.com on

Regularity of the Interfaces with Sign Changes

Department of Mathematical Sciences, Faculty of Science, Ehime University,

Received October 28, 1998

In previous papers we considered the Cauchy problem for the one-dimensional

In the previous papers [Sak 1, Sak 2] we considered the Cauchy problem

“t u=“x (|“x u| p − 2 “x u) in R × R+ and u(x, 0)=u0 (x) in R, (1.1)

where u0 is a nonzero, nonnegative, and bounded function in R having

C(t)={x ¥ R; “x u(x, t)=0 and u(x, t) > 0}. (1.2)

“t v=“ 2x (|v| p − 2 v) in R × R+. (1.3)

two-dimensional cylinders. But the monotonicity of the initial data is

f(0)=0 and f −(s) > 0 for s ] 0. (1.4)

A typical example of f is given by

where k1 , k2 , m1 , and m2 are positive constants. Consider the following

where a, b, and L are positive constants, and where v0 ¥ C([ − L, L])

˛ f(v0 )(−L)=a and f(v0 )(L)=−b.

There exists a point x0 ¥ (−L, L) satisfying v0 (x) ˛ >< 00 if − L E x < x0 ,

In order to define a weak solution generally, by letting “p Q be the

“p Q=((−L, L) × {0}) 2 ({ − L, L} × [0, .)),

we consider the problem,

where g ¥ C(“p Q) is a given function. Note that in problem (1.5)

For T > 0 , let

Q0, T =(−L, L) × (0, T] and

We also consider the problem,

where g ¥ C(“p Q0, T ) is a given function. A function v=v(x, t) is said to be

Theorem 1.1. Let v ¥ C(Q̄) be the unique weak solution of problem

v(x, t)˛ >< 00 if x < x(t),

(2) x( · ) ¥ C 1(TÄ , .).

The remainder of the present paper is organized as follows. In Section 2

First of all, let us consider the following one-dimensional Cauchy

“t u=“ 2x (f(u)) in R × R+ and u(x, 0)=u0 (x) in R, (2.1)

where u0 is a bounded, continuous, and nonnegative function on R. A

FF {f(u) “ 2x t+u “t t} dx dt=0 for any t ¥ C. +

Lemma 2.1. Denote by G the class of nonzero and nonnegative functions

Lemma 2.3 (Comparison Principle). Let g1 , g2 ¥ C(“p Q0, T ) and let v1 , v2

Proof. Although in [DK] only nonnegative weak solutions belonging

min v0 (x) E v E max v0 (x) in Q̄, (2.4)

so by extending f appropriately we may assume additionally that

lim f(s)=−. and lim f(s)=..

b(s)=f −1(s) for s ¥ R. (2.5)

Let us use a smooth approximation result for b as in [Hua, Theorem 2.1,

M(r, R)= ˛ max b (s) r E |s| E R

max{b (−R), b (R)} −

Then we see that M( · , R) is monotone nonincreasing, m( · , R) is monotone

m(|s|, R) E b −(s) E M(|s|, R) for 0 < |s| E R. (2.6)

For e ¥ (0, 12 ), define

m(|s|, R+1) E b −n (s) E M(|s|, R+1) for 0 < |s| E R. (2.7)

bn Q b as n Q . uniformly on any bounded interval of R. (2.8)

Then, in particular, putting Hn (x)=H(x, 1n ) for each n F 3 yields that

Hn ¥ C .([ − L, L]), and Hn Q f(v0 ) as n Q . uniformly on [ − L, L].

Furthermore, by using the theory of uniformly parabolic equations [LSU]

˛ “t (bn (Vn ))=“ 2x Vn

Then, by applying the maximum principle to H and Vn , and using (2.8), we

sup |bn (Vn )| E C,

for some constant C > 0 independent of n F 3. Therefore, in view of (2.7)

Vn Q f(v) and bn (Vn ) Q v as n Q . uniformly on any compact set in Q.

Lemma 2.4. The solution v is continuous in Q̄, if the value of v on “p Q is

Vn Q f(v), bn (Vn ) Q v, “x Vn Q “x (f(v)), and