MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
COMMUNICATIONS IN PARTIAL DIFFERENTIAL EQUATIONS
Vol. 28, Nos. 7 & 8, pp. 1391–1436, 2003
A Nonhomogeneous Boundary-Value Problem for the
Korteweg–de Vries Equation Posed on a Finite Domain
Jerry L. Bona,1,* Shu Ming Sun,2 and Bing-Yu Zhang3
1
Department of Mathematics, University of Illinois at Chicago,
Chicago, Illinois, USA
2
Department of Mathematics, Virginia Polytechnic Institute and
State University, Blacksburg, Virginia, USA
3
Department of Mathematical Sciences, University of Cincinnati,
Cincinnati, Ohio, USA
ABSTRACT
Studied here is an initial- and boundary-value problem for the Korteweg–de
Vries equation posed on a bounded interval with nonhomogeneous boundary
conditions. This particular problem arises naturally in certain circumstances
when the equation is used as a model for waves and a numerical scheme is
needed. It is shown here that this initial-boundary-value problem is globally
well-posed in the L2-based Sobolev space H s(0, 1) for any s 0. In addition,
the mapping that associates to appropriate initial- and boundary-data the
corresponding solution is shown to be analytic as a function between appropriate
Banach spaces.
*Correspondence: Jerry L. Bona, Department of Mathematics, University of Illinois at
Chicago, Chicago, IL 60607, USA; E-mail: bona@math.uic.edu.
1391
DOI: 10.1081/PDE-120024373
Copyright & 2003 by Marcel Dekker, Inc.
0360-5302 (Print); 1532-4133 (Online)
www.dekker.com
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
1392
Bona, Sun, and Zhang
1. INTRODUCTION
This article is concerned with the Korteweg–de Vries equation (KdV-equation
henceforth)
ut þ ux þ uux þ uxxx ¼ 0
ð1:1Þ
posed as an initial- and boundary-value problem. In the conception pursued here,
one asks for a solution of (1.1) for ðx, tÞ 2 Rþ where is an interval in R,
subject to an initial condition
uðx, 0Þ ¼ ðxÞ,
for x 2 ,
ð1:2Þ
and appropriate boundary conditions at the ends of the interval. In applications to
physical problems, the independent variable x is often a coordinate representing
position in the medium of propagation, t is proportional to elapsed time, and
u(x,t) is a velocity or an amplitude at the point x at time t. Here and below, if
f ¼ f ðx, tÞ is a function of x and t, then fx is shorthand for @x f and similarly
ft ¼ @t f . When ¼ R, the entire real line, this is the classical problem whose
study was initiated by Gardner et al. (1974) and Lax (1968) in the middle 1960’s
by way of the inverse scattering theory and by Sjöberg (1970) and Temam (1969) in
the late 1960’s using the then new methods for the analysis of nonlinear partial
differential equations, and by many others since. As will be described presently,
this pure initial-value problem continues to attract attention and its mathematical
theory has proved to be subtle.
Another configuration that arises naturally in making predictions of waves is to
take ¼ Rþ ¼ fx j x > 0g and specify uð0, tÞ for t > 0 and uðx, 0Þ ¼ 0, say, for x > 0.
This corresponds to a known wavetrain generated at one end and propagating into
a quiescent region of the medium of propagation. If u(0, t) is of small amplitude
(small compared to one in this scaling) and has primarily low frequency content,
then the waves generated by the boundary disturbance will satisfy the assumptions
underlying the derivation of the KdV-equation. The semi-infinite aspect of the
domain mirrors the fact that the KdV-equation written in the form (1.1) is an
approximation only for waves moving in the direction of increasing values of x.
Once the incoming waves encounter a boundary, reflection will come into play,
and the KdV-equation is no longer expected provide an accurate rendition of reality.
The problem of imposition of boundary data at the right-hand end of the domain
does not arise when the KdV-equation is posed on R+ with zero initial data, say, and
input from the left-hand boundary. Indeed, the zero boundary conditions at x ¼ þ1
implicit in the formulation may be imposed by function-class restrictions (e.g.,
uð , tÞ 2 L2 ðRþ Þ for all relevant values of t). This initial-boundary-value problem
fits well with laboratory studies wherein waves are generated by a wavemaker at
the left-hand end and these are monitored as they propagate down the channel, with
the experiment ceasing as soon as the waves reach the other end of the channel and
reflected components intrude (see Bona et al., 1981; Hammack, 1973; Hammack and
Segur, 1974; Zabusky and Galvin, 1971). Similarly, when modeling surface waves
arriving from deep water into near-shore zones or large-scale internal
waves propagating from the deep ocean onto the continental shelf, reflection may
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
Korteweg–de Vries Equation on Finite Domain
1393
sometimes be safely ignored and one encounters a variable-coefficient version of
Eq. (1.1) posed on Rþ ð0, T Þ with a time-dependent Dirichlet boundary condition
at x ¼ 0 (see Boczar-Karakiewicz et al., 1991; Boczar-Karakiewicz et al., (submitted)
for example). The quarter-plane problem just outlined has been considered recently
by the present authors (see Bona et al., 2001) and by several others (see the references
in the last-quoted article) and there is a satisfactory theory of well-posedness for
this problem.
However, if one is interested in implementing a numerical scheme to approximate solutions of the quarter-plane problem, there arises the issue of cutting off the
spatial domain. Once this is done, two more boundary conditions are needed to
specify the solution completely. Because the model cannot countenance waves
moving to the left, it is usual, as suggested above, to apply the model only on a
time scale T short enough that significant wave motion has not reached the righthand boundary. If the right-hand boundary is located at x ¼ r, say, then it is therefore natural in regard to the physical problem to impose uðr, tÞ ¼ ux ðr, tÞ ¼ 0 for
0 t T to obtain a complete set of boundary conditions. Of course, one might
also imagine imposing ux(0, t) rather than ux(r, t), but in practical situations, one does
not normally have information that warrants the imposition of a second boundary
condition at the left-hand, or wavemaker end of the medium of propagation. As far
as mathematical analysis is concerned, it makes relatively little difference whether or
not the boundary conditions are homogeneous. In consequence, consideration is
given here to (1.1)–(1.2) completed by the general nonhomogeneous boundary conditions
uð0, tÞ ¼ h1 ðtÞ,
uðr, tÞ ¼ h2 ðtÞ,
ux ðr, tÞ ¼ h3 ðtÞ,
for t 0,
ð1:3Þ
where the initial value and the boundary data hj, j ¼ 1, 2, 3 are given functions. The
principal concern of the present essay is the well-posedness of the initial-boundaryvalue problem (IBVP henceforth) (1.1)–(1.3). That is, we aim to establish
existence, uniqueness, and persistence properties of solutions corresponding to
reasonable auxiliary data, together with continuous dependence of the solution
upon the auxiliary data. A brief review of the mathematical theory currently
available is now presented. The pure initial-value problem (IVP) for (1.1) and its
relatives where the initial datum is specified on the entire real axis R has received a
lot of attention in the last three decades, both in case lies in an L2(R)-based
Sobolev space and in case is periodic (see Bona and Scott, 1976; Bona and
Smith, 1978; Bourgain, 1993a; Bourgain, 1993b; Constantin and Saut, 1988;
Hammack, 1973; Hammack and Segur, 1974; Kato, 1975; Kato, 1979; Kato,
1983; Kenig et al., 1991a; Kenig et al., 1991b; Kenig et al., 1993a; Kenig et al.,
1993b; Kenig et al., 1996; Lax, 1968; Miura, 1976; Russell and Zhang, 1993;
Russell and Zhang, 1995; Russell and Zhang, 1996; Saut and Temam, 1976; Sun,
1996; Temam, 1969; Zhang, 1995a, 1995b, 1995c). In particular, various smoothing
properties have been discovered for solutions of the (1.1) when posed on the whole
line R or on a periodic domain S (e.g., the unit circle in the plane). It is those
smoothing properties that enable one to prove that the IVP (1.1)–(1.2) is wellposed in the space H s(R) for s > 3/4 when posed on R and is well-posed in the
space H s(S) for s > 1/2 when posed on the periodic domain S (Bourgain, 1993a;
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
1394
Bona, Sun, and Zhang
Bourgain, 1993b; Kenig et al., 1993b; Kenig et al., 1996). By contrast, the study of
the KdV-equation posed on the half line R+ or on a finite interval has received much
less attention and the results available thus far appear to be not as sharp as those for
the IVP on R. For the initial-boundary-value problem (IBVP henceforth) for the
KdV-equation posed on the half line R+,
ut þ ux þ uux þ uxxx ¼ 0,
uð0, tÞ ¼ hðtÞ,
)
uðx, 0Þ ¼ ðxÞ,
ð1:4Þ
for x, t 2 Rþ , we have provided a review in our recent article Bona et al., (2001) (see
the earlier work of Bona and Dougalis, 1980; Bona and Scott, 1974; Bona and
Winther, 1989; the related article of Benjamin et al., 1972, on the BBM equation,
and the recent article by Colliander and Kenig, 2002). In Bona et al. (2001), we
pointed out that for the linear problem obtained from (1.4) by omitting the quadratic
term, there are smoothing properties similar to those established by Kenig et al.
(1991b) for (1.1)–(1.2) posed on all of R. Consequently, we were able to show the
IBVP (1.4) to be well-posed in the space Cð ½0, T; H s ðRþ ÞÞ for any s > 3/4 provided
the data ð, hÞ is drawn from H s ðRþ Þ H ðsþ1Þ=3 ð0, T Þ, by applying the contractionmapping principle. The corresponding solution map was shown to be analytic. In
their recent work, Colliander and Kenig (2002) showed that (1.4) is well-posed for
s 0.
For the KdV-equation posed on a finite interval, Bubnov (1979, 1980) studied
the general two-point boundary-value problem
8
ut þ uux þ uxxx ¼ f ðx, tÞ,
uðx, 0Þ ¼ 0,
>
>
>
>
>
< 1 uxx ð0, tÞ þ 2 ux ð0, tÞ þ 3 uð0, tÞ ¼ 0,
ð1:5Þ
>
1 uxx ð1, tÞ þ 2 ux ð1, tÞ þ 3 uð1, tÞ ¼ 0,
>
>
>
>
:
1 ux ð1, tÞ þ 2 uð1, tÞ ¼ 0,
posed on the interval ð0, 1Þ (see also the related work Bona and Dougalis (1980) on
the BBM-equation). Here, i , i , j 2 R, i ¼ 1, 2, 3, j ¼ 1, 2 are real constants and
assumptions are imposed so that the L2 norm of the solutions of the linear version
of (1.5) (obtained by dropping the nonlinear term uux ) is decreasing. It was shown in
Bubnov (1979) that for given T > 0 and f 2 H 1 ð ½0, T; L2 ð0, 1ÞÞ, there exists a T > 0
depending on k f kH 1 ð ½0, T;L2 ð0, 1ÞÞ such that (1.5) admits a unique solution
u 2 L2 ð ½0, T ; H 3 ð0, 1ÞÞ,
ut 2 L1 ð ½0, T ; L2 ð0, 1ÞÞ \ L2 ð ½0, T ; H 1 ð0, 1ÞÞ:
In Zhang (1994), Zhang considered boundary control of the KdV-equation posed on
a finite interval ð0, 1Þ with Dirichlet boundary conditions. A feedback control law
was introduced to stabilize the system, leading to the initial-boundary-value problem
ut þ uux þ uxxx ¼ 0,
uð0, tÞ ¼ 0,
uð1, tÞ ¼ 0,
uðx, 0Þ ¼ ðxÞ,
)
x 2 ð0, 1Þ, t 0,
ux ð1, tÞ ¼ ux ð0, tÞ,
t 0,
ð1:6Þ
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
Korteweg–de Vries Equation on Finite Domain
1395
with 0 j j < 1. Note that when ¼ 0, the system (1.6) is (1.1)–(1.3) with homogeneous boundary conditions. It was shown in Zhang (1994) that Eq. (1.6) is globally
well-posed in the space H 3kþ1 ð0, 1Þ for k ¼ 0, 1, . . .. In a recent article Colin and
Ghidaglia (2001), the authors considered the following initial-boundary value
problem
)
ut þ uux þ uxxx ¼ 0,
uðx, 0Þ ¼ ðxÞ, x 2 ð0, 1Þ, t 0,
uð0, tÞ ¼ h1 ðtÞ,
ux ð1, tÞ ¼ h2 ðtÞ,
uxx ð1, tÞ ¼ h3 ðtÞ,
t 0,
and showed it to be locally well-posed in the space H1(0,1) with the initial data
drawn from H 1 ð0, 1Þ and the boundary data ðh1 , h2 , h3 Þ taken from the product space
C 1 ½0, T Þ C 1 ½0, T Þ C 1 ½0, T Þ. In addition, Rosier (1997) studied the control
problem for the system
)
ut þ uux þ uxxx ¼ 0,
uðx, 0Þ ¼ ðxÞ, x 2 ð0, 1Þ, t 0,
uð0, tÞ ¼ 0,
uð1, tÞ ¼ 0
ux ð1, tÞ ¼ hðtÞ,
t 0,
where the boundary function h is considered as a control input. Rosier showed the
system is (locally) exactly controllable in the space L2(0,1). A similar problem was
also considered by Zhang (1999) for the system (1.1)–(1.3) where the boundary
value functions hj ðtÞ, j ¼ 1, 2, 3 are all taken to be control inputs. This system is
shown to be exactly controllable in the space H s(0, 1) for any s 0 in a neighborhood
of any smooth solution of the KdV-equation. (Exact controllability means, roughly,
that for a given time T > 0 and a given pair of functions and
in the space
H s(0, 1), there exist appropriate controls such that the corresponding system
possesses a solution u which exactly equals at t ¼ 0 and equals
at t ¼ T. Put
colloquially, given two states and , there is a control h that will drive the system
from to
in time T. Of course, there are obvious approximate controllability
analogs of this concept. Readers who are interested in control issues are referred
to the excellent review article of Russell (1978) for commentary on controllability
and stabilizability of linear partial differential equations and to Russell and
Zhang (1993, 1995, 1996) for theory of controllability and stabilizability of the
KdV-equation.
In this article, the nonhomogeneous boundary-value problem (1.1)–(1.3) is
considered. The aim is to establish the well-posedness of (1.1)–(1.3) in the space
H s(0, r) when the initial data is drawn from H s(0, r) and the boundary data
(h1, h2, h3, ) lies in the product space H s1 (0,T ) H s2(0,T ) H s3(0,T ) for some
appropriate indices s1, s2, and s3 that depend on s. As we will see later, the natural
choices of s1, s2 and s3 are s1 ¼ s2 ¼ (s+1)/3 and s3 ¼ s/3. For convenience of writing,
we take the underlying spatial domain (0, r) to be (0, 1) throughout. This is a restriction of no consequence as far as the theory is concerned. The well-posedness result
for the IBVP (1.1)–(1.3) we establish in this article appears to require some compatibility conditions relating the initial datum (x) and the boundary data hj (t),
j ¼ 1, 2, 3. A simple computation shows that if u is a C1-smooth solution of the
IBVP (1.1)–(1.3), then its initial data uðx, 0Þ ¼ ðxÞ and its boundary values hj (t),
j ¼ 1, 2, 3 must satisfy the following compatibility conditions:
k ð0Þ ¼ h1ðkÞ ð0Þ,
k ð1Þ ¼ hðkÞ
2 ð0Þ,
0k ð1Þ ¼ hðkÞ
3 ð0Þ
ð1:7Þ
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
1396
Bona, Sun, and Zhang
for k ¼ 0, 1, . . . , where hðkÞ
j ðtÞ is the kth order derivative of hj and
0 ðxÞ ¼ ðxÞ
0
Pk1
0
k ðxÞ ¼ 000
k1 ðxÞ þ k1 ðxÞ þ
j¼0 j ðxÞkj1 ðxÞ
ð1:8Þ
for k ¼ 1, 2, . . . : When the well-posedness of (1.1)–(1.3) is considered in the space
H s(0, 1) for some finite value s 0, the following s-compatibility conditions thus
arise naturally.
Definition 1.1. (s-compatibility) Let T > 0 and s 0 be given. A four-tuple
ð, h~Þ ¼ ð, h1 , h2 , h3 Þ 2 H s ð0, 1Þ H ðsþ1Þ=3 ð0, T Þ H ðsþ1Þ=3 ð0, T Þ H s=3 ð0, T Þ
is
said to be s-compatible if
k ð0Þ ¼ hðkÞ
1 ð0Þ,
k ð1Þ ¼ h2ðkÞ ð0Þ
ð1:9Þ
holds for k ¼ 0, 1, . . . , ½s=3 1 when s 3½s=3
k ¼ 0, 1, . . . , ½s=3 when 1=2 < s 3½s=3 3=2 and
k ð0Þ ¼ hðkÞ
1 ð0Þ,
1=2,
or (1.9) holds for
0k ð1Þ ¼ hðkÞ
3 ð0Þ
k ð1Þ ¼ h2ðkÞ ð0Þ,
holds for k ¼ 0, 1, . . . , ½s=3 when s 3½s=3 > 3=2. We adopt the convention that
Eq. (1.9) is vacuous if ½s=3 1 < 0.
With this compatibility notation, we may state the following two theorems,
which comprise the main results of this article.
Theorem 1.2. (Local well-posedness) Let T > 0 and s 0 be given. Suppose that
ð, h~Þ 2 H s ð0, 1Þ H ðsþ1Þ=3 ð0, T Þ H ðsþ1Þ=3 ð0, T Þ H s=3 ð0, T Þ
is s-compatible. Then there exists a T 2 ð0, T depending only on the norm of ð, h~Þ
in the space H s ð0, 1Þ H ðsþ1Þ=3 ð0, T Þ H ðsþ1Þ=3 ð0, T Þ H s=3 ð0, T Þ such that Eqs.
(1.1)–(1.3) admits a unique solution
u 2 Cð ½0, T ; H s ð0, 1ÞÞ \ L2 ð ½0, T ; H sþ1 ð0, 1ÞÞ:
Moreover, the solution depends continuously in this latter space on variations of the
auxiliary data in their respective function classes.
Theorem 1.3. (Global well-posedness) Let T > 0 be arbitrary and s 0. For any
s-compatible
ð, h~Þ 2 H s ð0, 1Þ H
where
1 ðsÞ
¼
2 ðsÞ
¼
1 ðsÞ
ð0, T Þ H
þ ð5s þ 9Þ=18
ðs þ 1Þ=3
þ ð5s þ 3Þ=18
s=3
if 0
1 ðsÞ
ð0, T Þ H
s < 3,
if s 3;
if 0 s < 3,
if s 3
2 ðsÞ
ð0, T Þ,
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
Korteweg–de Vries Equation on Finite Domain
1397
and is any positive constant, the IBVP (1.1)–(1.3) admits a unique solution
u 2 Cð ½0, T; H s ð0, 1ÞÞ \ L2 ð ½0, T; H sþ1 ð0, 1ÞÞ:
Moreover, the solution depends continuously on variations of the auxiliary data in their
respective function classes.
Remark 1.4. The global well-posedness result presented in Theorem 1.3 requires
slightly stronger regularity of the boundary values hj , j ¼ 1, 2, 3 in case 0 s < 3
when compared with the local well-posedness result in Theorem 1.2. The same
situation appears in the global well-posedness theory for the KdV-equation posed
in a quarter plane in Bona et al. (2001).
The proof of our well-posedness result for (1.1)–(1.3) relies on the smoothing
properties of the associated linear problem
)
ut þ ux þ uxxx ¼ f ,
uðx, 0Þ ¼ ðxÞ,
ð1:10Þ
uð0, tÞ ¼ h1 ðtÞ,
uð1, tÞ ¼ h2 ðtÞ, ux ð1, tÞ ¼ h3 ðtÞ :
There are three types of smoothing associated with solving (1.10); these are the
smoothing effects of the solution u with respect to the forcing f, the initial value
and the boundary data hj ¼ 0, j ¼ 1, 2, 3, respectively. It will be demonstrated that
(i)
(ii)
(iii)
For 2 L2 ð0, 1Þ with f ¼ 0, hj ¼ 0, j ¼ 1, 2, 3, the solution u of (1.10)
belongs to the space CðRþ ; L2 ð0, 1ÞÞ \ L2 ðRþ ; H 1 ð0, 1ÞÞ and ux 2 Cð ½0, 1,
L2, t ðRþ ÞÞ;
For f 2 L1 ðRþ ; L2 ð0, 1ÞÞ with ¼ 0, hj ¼ 0, j ¼ 1, 2, 3, the solution u of
Eq. (1.10) belongs to the space CðRþ ; L2 ð0, 1ÞÞ \ L2 ðRþ ; H 1 ð0, 1ÞÞ
and ux 2 Cð ½0, 1, L2, t ðRþ ÞÞ;
1=3
ðRþ Þ, h3 2 L2, loc ðRþ Þ with f ¼ 0 and ¼ 0, the solution u
For h1 , h2 2 Hloc
of (1.10) belongs to the space CðRþ ; L2 ð0, 1ÞÞ \ L2, loc ðRþ ; H 1 ð0, 1ÞÞ and
ux 2 Cð ½0, 1, L2, t ðRþ ÞÞ.
Various other related linear estimates will also be derived. Once these linear
estimates are in hand, a local well-posedness result for (1.1)–(1.3) may be established
using the contraction-mapping principle. The long-time results are obtained by finding
global a priori estimates for smooth solutions of (1.1)–(1.3). It is interesting to note
that while an L2-estimate of solutions is relatively straightforward to establish, the
global H1- and H2-bounds on solutions seem difficult to obtain by the usual energytype methods. The approach used here is to obtain an L2-estimate of the time derivative ut of solutions, which, in turn, provides a global H3-estimate. Nonlinear interpolation theory (Bona and Scott, 1976; Tartar, 1972) is then used to obtain the global
H s-estimates for 0 < s < 3. Global a priori H s-estimates for s > 3 are established by
obtaining a priori bounds on @ kt u for k ¼ 1, 2, . . . , ½s=3.
Because of its well-posedness, the IBVP (1.1)–(1.3) defines a continuous nonlinear map Ks, T from the space
Xs, T ¼ H s ð0, 1Þ H ðsþ1Þ=3 ð0, T Þ H ðsþ1Þ=3 ð0, T Þ H s=3 ð0, T Þ
ð1:11Þ
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
1398
Bona, Sun, and Zhang
to the space Cð ½0, T; H s ð0, 1ÞÞ \ L2 ð ½0, T; H sþ1 ð0, 1ÞÞ for given T > 0 and s 0.
It follows readily from the proof presented here that Ks, T is locally Lipschitz continuous. In fact, the map Ks, T is much smoother than just Lipschitz. According to
the local existence theory, for a given ð, h~Þ 2 Xs, T which is s-compatible,
h~ ¼ ðh1 , h2 , h3 Þ, there is a unique local solution u of (1.1)–(1.3). Of course, the existence time T for this solution need not be T. Let DðKs, T Þ connote those elements of
Xs, T for which the solution exists on ½0, T Þ. As will appear from our detailed theory,
DðKs, T Þ is an open neighborhood of the zero element in Xs, T if 0 s 7=2. In this
case, the mapping Ks, T is analytic from DðKs, T Þ to the space
Cð ½0, T; H s ð0, 1ÞÞ \ L2 ð ½0, T; H sþ1 ð0, 1ÞÞ. That is to say, for given ð, h~Þ in
DðKs, T Þ, there exists a > 0 such that for any ð1 , h~1 Þ 2 Xs, T with
ð, h~Þ þ ð1 , h~1 Þ 2 DðKs, T Þ and kð1 , h~1 ÞkXs, T , then Ks, T ð þ 1 , h~ þ h~1 Þ has a
Taylor series expansion which is uniformly convergent in the space
Cð ½0, T; H s ð0, 1ÞÞ \ L2 ð ½0, T; H sþ1 ð0, 1ÞÞ. Each term in the Taylor series is determined by the solution of a forced linear KdV-equation. Thus one obtains the attractive result that solutions of the nonlinear problem (1.1)–(1.3) can be obtained by
solving an infinite sequence of linear problems. When s > 7=2, because of the compatibility conditions, DðKs, T Þ is no longer a neighborhood of zero in Xs, T . However,
we can view solutions of the IBVP (1.1)–(1.3) as a special class of solutions of IBVP’s
for a system of nonlinear equations. Viewed this way, it may be shown that the IBVP
for this nonlinear system is well-posed and the corresponding nonlinear map is again
analytic.
The approach developed in this article can also be used to obtain similar results
for the following general nonhomogeneous boundary-value problem for the KdVequation:
8
ut þ ux þ uux þ uxxx ¼ f ðx, tÞ,
uðx, 0Þ ¼ ðxÞ, the
>
>
>
< u ð0, tÞ þ u ð0, tÞ þ uð0, tÞ ¼ h ðtÞ,
1 xx
2 x
3
1
> 1 uxx ð1, tÞ þ 2 ux ð1, tÞ þ 3 uð1, tÞ ¼ h2 ðtÞ,
>
>
:
1 ux ð1, tÞ þ 2 uð1, tÞ ¼ h3 ðtÞ,
ð1:12Þ
with x 2 ð0, 1Þ and t 0 (cf. Bubnov, 1979, 1980). Roughly speaking, if the parameters j, j, and j, j ¼ 1, 2, 3, are chosen such that the solution u of the associated
homogeneous linear problem (obtained by dropping uux and setting f ¼ 0 and hj ¼ 0
for j ¼ 1, 2, 3 in (1.12)) satisfies
Z
d 1
juðx, tÞj2 dx 0
ð1:13Þ
dt 0
for any t > 0, the detailed techniques developed in the remainder of the article apply
and one may establish that the IBVP (1.12) is well-posed in the space H s(0, 1) for any
s 0. In case (1.13) is not valid, the issue of local well-posedness may be more
challenging. As far as global existence is concerned, we can give conditions on
(1.12) for this to hold. Indeed, because of the strong smoothing resulting
from the boundedness of the domain, all that is required is to keep the L2(0, 1)—
norm bounded on bounded time intervals. We will not enter into the details of this
development here.
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
Korteweg–de Vries Equation on Finite Domain
1399
With global well-posedness results in hand, a natural further question arises
about the solutions of the IBVP (1.1)–(1.3), namely their long time, asymptotic
behavior. Because the imposition of boundary conditions may exert a weak
dissipative mechanism, it is expected that the solutions of the nonlinear system
(1.1)–(1.3) will decay as t ! þ1, at least in case the initial value is small and
the boundary data hj ðtÞ, j ¼ 1, 2, 3 decay to zero as t ! 1. A special situation occurs
when the boundary data are all periodic of some period , say. Experiments Bona et
al., (1981) suggest that in this case the solution will eventually become time periodic
of period . This has been rigorously established in Bona et al., (2003) for the
quarter-plane problem (1.4) with a damping term included. It would be interesting
and useful to have similar results for the finite domain problem considered here. A
related question is whether the (1.1) possesses a strictly time-periodic solution if its
boundary forcing h1, h2, and h3 are time-periodic functions defined on all of R
(cf. Bona et al., 2003; Wayne, 1990, 1997). Those issues will be addressed in our
subsequent articles.
The article is organized as follows. In Sec. 2, several estimates pertaining to
solutions of the linear problem (1.10) are established which display the smoothing
properties mentioned earlier. In Sec. 3, the linear estimates are used to prove that
(1.1)–(1.3) is locally well-posed. The global well-posedness of (1.1)–(1.3) is established
in Sec. 4. Analyticity of the nonlinear map Ks, T defined by the IBVP (1.1)–(1.3) is
discussed in Sec. 5.
2. LINEAR ESTIMATES AND SMOOTHING PROPERTIES
In this section, various smoothing properties that accrue to the linear system
Eq. (1.10) will be discussed. As (1.10) is linear, it is convenient to break up the
analysis. Considered first is the problem
)
ut þ ux þ uxxx ¼ 0, uðx, 0Þ ¼ ðxÞ,
ð2:1Þ
uð0, tÞ ¼ 0, uð1, tÞ ¼ 0, ux ð1, tÞ ¼ 0
with homogeneous boundary conditions and no forcing. Then we will consider
problem (1.10) with non-trivial forcing f but with all three boundary conditions
set to zero. The outcome of the analysis of these problems are recorded in
Propositions 2.1 and 2.4. Next, problem (1.10) with zero forcing, but non-trivial
boundary conditions is taken up. We use the Laplace transform in t to obtain a
solution formula. Whilst a little complicated, the representation formulas (2.14) and
following are completely explicit. Consequently, their analysis may be carried out in
detail. The outcome is recorded in a sequence of propositions that conclude the
section.
Let A be the linear operator defined by
Af ¼ f 000 f 0 :
Consider A as an unbounded operator on L2(0, 1) with the domain
DðAÞ ¼ f f 2 H 3 ð0, 1Þ, f ð0Þ ¼ f ð1Þ ¼ f 0 ð1Þ ¼ 0g:
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
1400
Bona, Sun, and Zhang
The IBVP (2.1) can be written as the initial-value problem of an abstract
evolution equation in the space L2(0,1), viz
du
¼ Au,
uð0Þ ¼ ,
ð2:2Þ
dt
where the spatial variable is suppressed. It is easily verified that both A and its
adjoint A* are dissipative, which is to say
hAf , f iL2 ð0, 1Þ
hA g, giL2 ð0, 1Þ
0,
0
for any f 2 DðAÞ and g 2 DðA Þ, where A g ¼ g000 þ g0 and
DðA Þ ¼ f f 2 H 3 ð0, 1Þ; f ð0Þ ¼ f 0 ð0Þ ¼ f ð1Þ ¼ 0g:
Thus the operator A is the infinitesimal generator of a C0-semigroup W0(t) in the
space L2(0, 1) (see Pazy, 1983). By standard semigroup theory applied in the overlying
space L2(0, 1), for any 2 L2 ð0, 1Þ,
uðtÞ ¼ W0 ðtÞ
belongs to the space Cb ðRþ ; L2 ð0, 1ÞÞ. The function u thus defined is called a mild
solution of (2.1). Such solutions certainly solve (2.1) in the sense of distributions (cf.
Bona and Winther (1983), Sec. 2). If 2 DðAÞ, then uðtÞ ¼ W0 ðtÞ belongs to the
much smaller space Cð0, 1; H 3 ð0, 1ÞÞ \ C1 ð0, 1; L2 ð0, 1ÞÞ and uðtÞ 2 DðAÞ for all
t 0. Moreover, the equation is satisfied in the sense of Cð0, 1; L2 ð0, 1ÞÞ, and in
particular, pointwise almost everywhere. Such solutions are called strong solutions.
For strong solutions, the boundary values are taken on pointwise. In what follows, a
solution of (2.1) is either a mild solution or strong solution in the semigroup context.
Proposition 2.1. For any 2 L2 ð0, 1Þ, uðtÞ ¼ W0 ðtÞ satisfies
Zt
u2x ð0, Þ d ¼ kk2L2 ð0, 1Þ
kuð , tÞk2L2 ð0, 1Þ þ
0
ð2:3Þ
and
Z
1
0
xu2 ðx, tÞ dx þ 3
Z tZ
0
1
0
u2x ðx, Þ dx d
ð1 þ tÞ
Z
0
1
2 ðxÞ dx
ð2:4Þ
for any t 0.
Remark 2.2. The relation (2.3) provides a trace result at x ¼ 0 which reveals a
boundary smoothing effect of the system (2.1).
Remark 2.3. Combining inequalities (2.3) and (2.4) gives
kukL2 ð0, t;H 1 ð0, 1ÞÞ
Cð1 þ tÞ1=2 kkL2 ð0, 1Þ ,
ð2:5Þ
which is a Kato-type smoothing effect. Note that the original Kato smoothing effect
for solutions of (1.1)–(1.2) posed on the whole real line R is local, which is to say,
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
Korteweg–de Vries Equation on Finite Domain
1401
1
ðRÞÞ. In contrast, the smoothing effect (2.5) is
2 L2 ðRÞ implies that u 2 L2 ð0, T; Hloc
global. As will be seen later, this global Kato-smoothing effect alone is enough to
establish the well-posedness of (1.1)–(1.3) in the space H s(0, 1) for s 0. This is in
sharp contrast to the problem (1.1)–(1.2) posed on the unbounded domain R or the
IBVP (1.4) posed on the unbounded domain R+, where both the Kato smoothing
and the Strichartz smoothing or the Bourgain smoothing are used to establish their
well-posedness in weak spaces.
Proof. Assume first that 2 DðAÞ. Then uðtÞ 2 DðAÞ for any t 0 and
u 2 C 1 ð0, 1; L2 ð0, 1ÞÞ. To obtain (2.3), multiply both sides of the differential
equation in (2.1) by 2u and integrate over ð0, 1Þ with respect to x and over
(0, t) with respect to t. Integration by parts then leads to (2.3). For inequality
(2.4), multiply both sides of the equation in (2.1) by 2xu, integrate the result over
½0, 1 ½0, 1, and integrate by parts to reach the relation
Z
1
0
xu2 ðx, tÞ dx þ 3
Z tZ
0
1
0
u2x ðx, Þ dx d ¼
Z
1
0
x2 ðxÞ dx þ
Z tZ
0
1
0
u2 ðx, Þ dx d
from which (2.4) follows on account of (2.3). If, instead, 2 L2 ð0, 1Þ, choose a
sequence fn g from DðAÞ such that n converges to in L2(0, 1) as n ! 1. Define
un to be
un ¼ W0 ðtÞn ,
n ¼ 1, 2, . . . :
As we have just shown, both (2.3) and (2.1) hold with u replaced by un and replaced
by n. Let T > 0 be fixed. Then the sequence fun g is bounded in the spaces
Cð ½0, T; L2 ð0, 1ÞÞ, L2 ð0, T; H 1 ð0, 1ÞÞ and Cð ½0, T; L2 ð0, 1; xdxÞÞ. Here L2 ð0, 1; xdxÞ
is the weighted L2-space with the weight x. Moreover un, x ð0, tÞ @x un ð0, tÞ is
a bounded sequence in the space L2 ð0, T Þ. Thus there exists a subsequence funk g
of fun g and a
u 2 L2 ð0, T; H 1 ð0, 1ÞÞ \ L1 ð0, T; L2 ð0, 1ÞÞ \ L1 ð0, T; L2 ð0, 1; xdxÞÞ
with ux ð0, tÞ 2 L2 ð0, T Þ such that funk g is convergent to u weakly in L2 ð0, T; H 1 ð0, 1ÞÞ
and weak-star in both spaces L1 ð0, T; L2 ð0, 1ÞÞ and L1 ð0, T; L2 ð0, 1; xdxÞÞ.
Furthermore, f@x unk ð0, tÞg is weakly convergent to ux ð0, tÞ in the space L2 ð0, T Þ. On
account of the lower semi-continuity of the various norms with regard to weak
convergence, it is adduced that (2.3) and (2.4) hold for u . On the other hand un
converges strongly to u ¼ W0 ðtÞ in L1 ð0, T; L2 ð0, 1ÞÞ. We conclude that
u 2 L2 ð0, T; H 1 ð0, 1ÞÞ \ L1 ð0, T; L2 ð0, 1ÞÞ \ L1 ð0, T; L2 ð0, 1; xdxÞÞ,
ux ð0, tÞ 2 L2 ð0, T Þ and (2.3) and (2.4) hold for u.
Next, attention is turned to the inhomogeneous linear problem
)
ut þ ux þ uxxx ¼ f ðx, tÞ, uðx, 0Þ ¼ 0,
uð0, tÞ ¼ 0,
uð1, tÞ ¼ 0,
ux ð1, tÞ ¼ 0:
œ
ð2:6Þ
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
1402
Bona, Sun, and Zhang
In terms of the operator A defined above, one may write (2.6) as an initial-value
problem for an abstract nonhomogeneous evolution equation, viz.
du
¼ Au þ f ,
dt
uð0Þ ¼ 0:
ð2:7Þ
By standard semigroup theory (see again Pazy, 1983), for any f 2 L1, loc ðRþ ; L2 ð0, 1ÞÞ,
Z
uðtÞ ¼
t
0
W0 ðt Þ f ðÞ d
ð2:8Þ
belongs to the space CðRþ ; L2 ð0, 1ÞÞ and is called a mild solution of (2.7). It is a weak
solution of (2.6) in the sense of distribution. In addition, if f ðtÞ 2 DðAÞ for t > 0 and
Af 2 L1, loc ðRþ ; L2 ð0, 1ÞÞ, then u(t) given by (2.8) solves (2.7) a.e. on ½0, T Þ and is
called a strong solution of (2.7).
Proposition 2.4. There exists a constant C such that for any f 2 L1, loc ðRþ ; L2 ð0, 1ÞÞ,
the solution u of Eq. (2.6) satisfies
kuð , tÞkL2 ð0, 1ÞÞ þ kux ð0, ÞkL2 ð0, tÞ
Ck f kL1 ð0, t;L2 ð0, 1ÞÞ
ð2:9Þ
and
Z
1
2
xu ðx, tÞ dx þ
0
Z tZ
1
0
0
2ð1 þ tÞk f k2L1 ð0, t;L2 ð0, 1ÞÞ
u2x ðx, Þ dx d
ð2:10Þ
for any t 0.
Proof. Without loss of generality, we assume that u is a strong solution. The general
case follows using a limiting procedure similar to that appearing in the proof of
Proposition 2.1. Multiply the equation in (2.6) by 2u and integrate over ð0, 1Þ with
respect to x. Integration by parts leads to
d
dt
Z
1
0
u2 ðx, tÞ dx þ u2x ð0, tÞ
2k f ð , tÞkL2 ð0, 1Þ kukL2 ð0, 1Þ
from which (2.9) follows. To prove (2.10), multiply both sides of the equation in (2.6)
by 2xu and integrate over the rectangle ð0, 1Þ ð0, tÞ in space-time. After integrations
by parts, it is seen that
Z
0
1
xu2 ðx, tÞ dx þ 3
Z tZ
¼2
0
Z
0
t
1
0
Z tZ
0
1
0
u2x ðx, Þ dx d
xf ðx, Þuðx, Þ dx d þ
Z tZ
0
1
0
u2 ðx, Þ dx d
kx1=2 f ð , ÞkL2 ð0, 1Þ kx1=2 uð , ÞkL2 ð0, 1Þ d þ
Z tZ
0
0
1
u2 ðx, Þ dx d
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
Korteweg–de Vries Equation on Finite Domain
Z
t
kx
1=2
1
1
sup kx1=2 uð , Þk2L2 ð0, 1Þ þ
20 t
2
Z tZ 1
u2 ðx, Þ dx d,
þ
Z
1=2
sup kx
0 t
uð , ÞkL2 ð0, 1Þ
0
1403
f ð , ÞkL2 ð0, 1Þ d þ
t
0
Z tZ
0
0
1
u2 ðx, Þ dx d
2
kx1=2 f ð , ÞkL2 ð0, 1Þ d
0
0
œ
which yields the inequality (2.10).
Next, consider the non-homogeneous boundary-value problem
)
ut þ ux þ uxxx ¼ 0,
uðx, 0Þ ¼ 0,
uð0, tÞ ¼ h1 ðtÞ,
uð1, tÞ ¼ h2 ðtÞ,
ux ð1, tÞ ¼ h3 ðtÞ:
ð2:11Þ
A common approach to (2.11) is to render its boundary conditions homogeneous as
follows. The solution u of (2.11) can be written as
uðx, tÞ ¼ wðx, tÞ þ vðx, tÞ
with
vðx, tÞ ¼ ð1 xÞh1 ðtÞ þ xh2 ðtÞ þ xð1 xÞðh3 ðtÞ h2 ðtÞ þ h1 ðtÞÞ
and w satisfying
wt þ wx þ wxxx ¼ vt vx ,
wð0, tÞ ¼ 0,
wð1, tÞ ¼ 0,
)
wðx, 0Þ ¼ 0,
wx ð1, tÞ ¼ 0:
ð2:12Þ
Thus to solve (2.11), one only need solve (2.12), which can be done by applying
Proposition 2.2. Here we assume hj ð0Þ ¼ 0 for j ¼ 1, 2, 3. However there is a serious
drawback to this approach; it is required that h0j 2 L1 ðRÞ for j ¼ 1, 2, 3 to obtain even
a mild solution u of (2.12) in the space Cb ð0, T; L2 ð0, 1ÞÞ \ L2 ð0, T; H 1 ð0, 1ÞÞ.
Furthermore, for such a mild solution u, although both uð0, tÞ and uð1, tÞ are defined
thanks to the Kato smoothing, it seems that the trace of ux ðx, tÞ at x ¼ 1 does not
make sense since ux is only known to be in the space L2 ð0, T; L2 ð0, 1ÞÞ. This suggests
that a stronger boundary smoothing property of (2.11) is needed if one wants to
solve (2.11) in the space Cb ð0, T; L2 ð0, 1ÞÞ \ L2 ð0, T; H 1 ð0, 1ÞÞ.
Our approach to solve (2.11) is to seek an explicit solution formula in terms of its
boundary values via the Laplace transform as we did in Bona et al., (2001) for the
KdV-equation in a quarter plane.
Applying the Laplace transform with respect to t, (2.11) is converted to
)
su^ ðx, sÞ þ u^ x ðx, sÞ þ u^ xxx ðx, sÞ ¼ 0,
ð2:13Þ
u^ ð0, sÞ ¼ h^1 ðsÞ, u^ ð1, sÞ ¼ h^2 ðsÞ, u^ x ð1, sÞ ¼ h^3 ðsÞ,
where
u^ ðx, sÞ ¼
and
Z
þ1
0
est uðx, tÞ dt
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
1404
Bona, Sun, and Zhang
h^j ðsÞ ¼
Z
þ1
0
est hj ðtÞ dt,
j ¼ 1, 2, 3:
The solution u^ ðx, sÞ of (2.13) can be written in the form
u^ ðx, sÞ ¼
3
X
j¼1
cj ðsÞej ðsÞx ,
where j ðsÞ, j ¼ 1, 2, 3, are the three solutions of the characteristic equation
s þ þ 3 ¼ 0
and cj ¼ cj ðsÞ, j ¼ 1, 2, 3, solve the linear system
8
^
>
>
< c1 þ c2 þ c3 ¼ h1 ðsÞ,
c1 e1 ðsÞ þ c2 e2 ðsÞ þ c3 e3 ðsÞ ¼ h^2 ðsÞ,
>
>
:
c1 1 ðsÞe1 ðsÞ þ c2 2 ðsÞe2 ðsÞ þ c3 3 ðsÞe3 ðsÞ ¼ h^3 ðsÞ :
Let (s) be the determinant of the coefficient matrix and i (s) the determinants of
the matrices that are obtained by replacing the ith-column of (s) by the column
vector ðh^1 ðsÞ, h^2 ðsÞ, h^3 ðsÞÞT , i ¼ 1, 2, 3. Cramer’s rule implies that
cj ¼
j ðsÞ
,
ðsÞ
j ¼ 1, 2, 3:
Taking the inverse Laplace transform of u^ yields
uðx, tÞ ¼
1
2i
Z
rþi1
ri1
est u^ ðx, sÞ ds ¼
Z
3
X
1 rþi1 st j ðsÞ j ðsÞx
e
e
ds
2i ri1
ðsÞ
j¼1
for any r > 0. The solution u of (2.11) may also be written in the form
uðx, tÞ ¼ u1 ðx, tÞ þ u2 ðx, tÞ þ u3 ðx, tÞ
where um ðx, tÞ solves (2.11) with hj 0 when j 6¼ m, m, j ¼ 1, 2, 3; thus um has the
representation
Z
3
X
1 rþi1 st j, m ðsÞ j ðsÞx ^
e
e
hm ðsÞ ds Wm ðtÞhm
um ðx, tÞ ¼
ðsÞ
2i ri1
j¼1
ð2:14Þ
for m ¼ 1, 2, 3. Here j, m ðsÞ is obtained from j ðsÞ by letting h^m ðtÞ ¼ 1 and hk ðtÞ 0
for k 6¼ m, k, m ¼ 1, 2, 3. It is straightforward to determine that in the last two
formulas, the right-hand sides are continuous with respect to r for r 0. As the
left-hand sides do not depend on r, it follows that we may take r ¼ 0 in these
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
Korteweg–de Vries Equation on Finite Domain
1405
formulas and in those appearing below. Write um in the form
um ðx, tÞ ¼
Z
3
X
1 þi1 st j, m ðsÞ j ðsÞx ^
e
hm ðsÞ ds
e
2i 0
ðsÞ
j¼1
Z
3
X
1 0 st j, m ðsÞ j ðsÞx ^
þ
hm ðsÞ ds
e
e
2i i1
ðsÞ
j¼1
Im ðx, tÞ þ IIm ðx, tÞ,
for m ¼ 1, 2, 3. Letting s ¼ ið 3 Þ with 1
< þ1 in the characteristic equation
s þ þ 3 ¼ 0,
the three roots are given in terms of by
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
32 4 i
þ
þ
,
1 ð Þ ¼ i, 2 ð Þ ¼
2
ð2:15Þ
þ
3 ð Þ
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
32 4 i
¼
2
and thus Im ðx, tÞ and IIm ðx, tÞ may be written in the form
Im ¼
Z
þ
3
X
1 þ1 ið 3 Þt þj ð Þx j, m ð Þ
2
^þ
e
e
þ ð Þ ð3 1Þhm ð Þ d
2
1
j¼1
and
IIm ¼
Z
3
X
1 þ1 ið 3 Þt j ð Þx
j, m ð Þ
2
^
e
e
ð Þ ð3 1Þhm ð Þ d
2
1
j¼1
3
þ
þ
^
where h^þ
m ð Þ ¼ hm ðið ÞÞ, ð Þ and j, m ð Þ are obtained from ðsÞ and j, m ðsÞ,
3
respectively, by replacing s with ið Þ and j ðsÞ with þ
j ð Þ, for j ¼ 1, 2, 3. Notice
that, with an obvious notation, ð Þ ¼ þ ð Þ and
ð
Þ
¼ þ
j, m
j, m ð Þ for j ¼ 1, 2, 3,
þ
^
^
and hm ð Þ ¼ hm ð Þ.
The next result is a technical lemma that will find frequent use in this section.
It plays the same role in dealing with the finite-interval problem as does Plancherel’s
theorem for the pure initial-value problem posed on the line.
Lemma 2.5. For any f 2 L2 ð0 þ 1Þ, let Kf be the function defined by
Z þ1
e ð Þx f ð Þ d
Kf ðxÞ ¼
0
where ð Þ is a continuous complex-valued function defined on ð0, 1Þ satisfying the
following two conditions:
(i)
There exist > 0 and b > 0 such that
Re ð Þ
sup
b;
0< <
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
1406
(ii)
Bona, Sun, and Zhang
There exists a complex number þ i such that
ð Þ
lim
!þ1
¼ þ i:
Then there exists a constant C such that for all f 2 L2 ð0, 1Þ,
kKf kL2 ð0, 1Þ
CðkeRe
ðÞ
f ð ÞkL2 ðRþ Þ þ k f ð ÞkL2 ðRþ Þ Þ:
Proof. Observe that
kKf k2L2 ð0, 1Þ
Notice also that
ð yÞ
þ1
0
Z þ1
0
0
Z þ1
Re ð ðsÞþ ðyÞÞ
0
¼
keRe
Z 1Z
Z
e
Re ð ðsÞxÞ
0
þ1 Z þ1
Z
1
j f ðsÞj ds
eRe ð
þ1
Z
eRe ð
ð yÞxÞ
0
ðsÞþ ð yÞÞx
j f ð yÞj dy dx
dx j f ðsÞjj f ð yÞj ds dy
j f ðsÞ f ð yÞj
ds dy
þ1
jReð ðsÞ þ ð yÞÞj
0
0Z
þ1 eRe ðsÞ j f ðsÞj ds
keRe ð Þ f ð ÞkL2 ðRþ Þ
0 jReð ðsÞ þ ð yÞÞj
L2 ðRþ Þ
Z þ1
j f ðsÞj ds
k f kL2 ðRþ Þ :
þ
jReð ðsÞ þ ð yÞÞjL2 ðRþ Þ
0
e
1=2
f ð yÞkL2 ðRþ Þ
keRe
ðÞ
f ð ÞkL2 ðRþ Þ
and for any y 2 ð0, þ 1Þ,
y
jReð ð yÞ þ ð yÞÞj
C
:
þ1
Using the integral version of Minkowski’s inequality yields
Z
þ1 eRe ðsÞ j f ðsÞj ds
0 jReð ðsÞ þ ð yÞÞj
L2 ðRþ Þ
Z
þ1 eRe ð yÞ j f ð yÞjy d
¼
0
jReð ð yÞ þ ð yÞÞj
L2 ðRþ Þ
Z þ1 Re ð yÞ
f ð yÞy
e
d
Reð ð yÞ þ ð yÞÞ
0
þ
L2 ðR Þ
C
Z
þ1
0
CkeRe
1
d keRe
pffiffiffiffi
ð1 þ Þ
ðÞ
f kL2 ðRþ Þ
ðÞ
f ð ÞkL2 ðRþ Þ
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
Korteweg–de Vries Equation on Finite Domain
1407
for some absolute constant C. The same argument also gives
Z þ1
j f ðsÞj ds
Ck f kL2 ðRþ Þ :
jReð ðsÞ þ ð yÞÞjL2 ðRþ Þ
0
The proof is complete.
œ
Lemma 2.6. Let a > 0 be given. For any f 2 L2 ð0, aÞ, let Gf be the function defined by
Za
Gf ðxÞ ¼
eið Þx f ð Þ d
0
where is a continuous real-valued function defined on the interval ½0, a which is C1 on
the open interval ð0, aÞ and such that there is a constant C1 for which ð1=j0 ð ÞjÞ C1
for 0 < < a. Then there exists a constant C such that for all f 2 L2 ð0, aÞ,
Ck f kL2 ð0, aÞ :
kGf kL2 ð0, 1Þ
Proof. Let ! ¼ ð Þ. Since 0 ð Þ 6¼ 0 for 2 ð0, aÞ, is strictly monotone and so is
invertible. Let ¼ 1 ð!Þ denote its inverse. Note that d! ¼ 0 ð Þd and so by a
change of variables, we may write Gf in terms of ! thusly:
Z ðaÞ
1
Gf ðxÞ ¼
ei!x f ð1 ð!ÞÞ 0 1
d!:
ð ð!ÞÞ
ð0Þ
It follows from Plancherel’s theorem that
Z
1 ðaÞ 1 2
1
f ð ð!Þ
0
1
2 ð0Þ
ð ð!ÞÞ
Z
1 a
1
¼
d
j f ð Þj2 0
2 0
j ð Þj
Z
C1 a
j f ð Þj2 d
2 0
kGf k2L2 ð0, 1Þ ¼
2
d!
œ
which is the advertised inequality.
The following three propositions provide estimates for u1 , u2 , and u3 , respectively.
They show clearly various smoothing properties that accrue through implementation
of the boundary conditions for the linear system (2.7) (cf. Remark 2.3).
Proposition 2.7. There exists a constant C such that
ku1 kL2 ðRþ ;H 1 ð0, 1ÞÞ þ sup ku1 ð , tÞkL2 ð0, 1Þ
0 t<þ1
and @x u1 2 Cb ð ½0, 1; L2 ðRþ ÞÞ with
sup k@x u1 ðx, ÞkL2 ðRþ Þ Ckh1 kH 1=3 ðRþ Þ
x2ð0, 1Þ
for all h1 2 H 1=3 ðRþ Þ:
Ckh1 kH 1=3 ðRþ Þ
ð2:16Þ
ð2:17Þ
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
1408
Bona, Sun, and Zhang
Proof. Since
1 ðsÞ þ 2 ðsÞ þ 3 ðsÞ 0,
it is readily seen that
1, 1 ðsÞ ¼ ð3 ðsÞ 2 ðsÞÞe1 ðsÞ ,
2, 1 ðsÞ ¼ ð1 ðsÞ 3 ðsÞÞe2 ðsÞ ,
3, 1 ðsÞ ¼ ð2 ðsÞ 1 ðsÞÞe3 ðsÞ
and thus
ðsÞ ¼ ð3 ðsÞ 2 ðsÞÞe1 ðsÞ þ ð1 ðsÞ 3 ðsÞÞe2 ðsÞ þ ð2 ðsÞ 1 ðsÞÞe3 ðsÞ :
In consequence, it follows readily that as a function of the variable introduced
above and defined by the relation s ¼ ið 3 Þ,
pffi
þ
3
1, 1 ð Þ
2 ,
e
þ ð Þ
pffiffi
þ
2, 1 ð Þ
3
e
,
þ ð Þ
þ
3, 1 ð Þ
1
þ ð Þ
as ! þ1. An application of Lemma 2.5 produces a constant C such that
2
3 Z þ1 þ
2
X
2
j, 1 ð Þ Re þj ð Þ
2
2
kI1 ð , tÞkL2 ð0, 1Þ
þ 1 h^þ
e
þ
1 ð Þð3 1Þ d
ð
Þ
1
j¼1
Z þ1
2
2
2
C
jh^þ
1 ð Þj ð3 1Þ d
1
C
Z
þ1
0
Z
ð1 þ Þ2=3
0
þ1
2
ei h1 ðÞ d d
Ckh1 k2H 1=3 ðRþ Þ :
The same argument applied to II1 ðx, tÞ gives
kII1 ð , tÞkL2 ð0, 1Þ
Ckh1 kH 1=3 ðRþ Þ :
Thus (2.16) holds. To prove (2.17), observe that
Z
þ
3
X
1 þ1 ið 3 Þt þ
þ
ð Þx 1, j ð Þ
2
^þ
j
@x I1 ðx, tÞ ¼
e
j ð Þe
þ ð Þ ð3 1Þh1 ð Þ d
2
1
j¼1
Z
3
X 1 þ1 i t þ
þ
þ
j, 1 ðð ÞÞ ^
h ði Þ d
¼
e j ðð ÞÞej ðð ÞÞx þ
ðð ÞÞ 1
2 0
j¼1
where ð Þ is the real solution of
¼ 3 for 1. Using the Plancherel
Theorem (with respect to t) yields that for any x 2 ð0, 1Þ,
2
Z
3
X
þ
ðð ÞÞ ^
1 þ1 þ
j,
1
ðð
ÞÞx
þ
2
2
k@x I1 ðx, ÞkL2 ðRþ Þ
j ðð ÞÞe j
þ ðð ÞÞ jh1 ði Þj d :
2
0
j¼1
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
Korteweg–de Vries Equation on Finite Domain
1409
Thus, one finds there is a constant C such that
Z
0
1
k@x I1 ðx, Þk2L2 ðRþ Þ dx
sup k@x I1 ðx, Þk2L2 ðRþ Þ
x2ð0, 1Þ
C
3 Z
X
j¼1
C
3 Z
X
j¼1
0
þ1
þ1
0
þ
j ðð
þ
ÞÞx 2 j, 1 ðð
j þ
ðð
2
þ
ÞÞ sup jej ðð
x2ð0, 1Þ
ð1 þ Þ2=3 jh^1 ði Þj2 d
2
ÞÞ ^
jh ði Þj2 d
ÞÞ 1
Ckh1 k2H 1=3 ðRþ Þ :
The following estimates were used in obtaining the last inequality:
pffiffi
sup je1 ð Þx j2 C,
sup je2 ð Þx j2 C e 3 þ 1 ,
x2ð0, 1Þ
x2ð0, 1Þ
sup je3 ð Þx j2
x2ð0, 1Þ
pffiffi
C1 e 3 þ 1 :
To see @x I1 is continuous from ½0, 1 to the space L2 ðRþ Þ, choose any x0 2 ½0, 1 and
x 2 ð0, 1Þ and observe that
@x I1 ðx, tÞ @x I1 ðx0 , tÞ
Z
3
X
þ
1 þ1 i t þ
e j ðð ÞÞ ej ðð
¼
2 0
j¼1
ÞÞx
þ
ej ðð
ÞÞx0
þ
j, 1 ðð ÞÞ ^
h ði Þ d :
þ ðð ÞÞ 1
Using the Plancherel theorem with respect to t as above yields
k@x I1 ðx, Þ @x I1 ðx0 , Þk2L2 ðRþ Þ
Z
3
X
þ
1 þ1 þ
j ðð ÞÞðej ðð
2 0
j¼1
C
3 Z
X
j¼1
þ1
0
ÞÞx
e
þ
j ðð ÞÞx0
ð1 þ Þ2=3 jh^1 ði Þj2 d :
2
þ
j, 1 ðð ÞÞ ^
Þ þ
jh ði Þj2 d
ðð ÞÞ 1
An application of Fatou’s lemma gives
lim k@x I1 ðx, Þ @x I1 ðx0 , Þk2L2 ðRþ Þ
x!x0
Z
3
X
þ
1 þ1 þ
j ðð ÞÞ lim ej ðð
x!x0
2 0
j¼1
¼0
ÞÞx
e
þ
j ðð ÞÞx0
2
þ
j, 1 ðð ÞÞ ^
jh ði Þj2 d
þ
ðð ÞÞ 1
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
1410
Bona, Sun, and Zhang
Similar considerations establish that
Z1
k@x I2 ðx, Þk2L2 ðRþ Þ dx
sup k@x I2 ðx, Þk2L2 ðRþ Þ
x2ð0, 1Þ
0
Ckh1 k2H 1=3 ðRþ Þ
and I2 ðx, Þ 2 Cb ð ½0, 1; L2 ðRþ ÞÞ. The proof is complete.
œ
Proposition 2.8. There exists a constant C such that
ku2 kL2 ðRþ ;H 1 ð0, 1ÞÞ þ sup ku2 ð , tÞkL2 ð0, 1Þ
0 t<þ1
Ckh2 kH 1=3 ðRþ Þ
ð2:18Þ
and @x u2 2 Cb ð ½0, 1; L2, t ðRþ ÞÞ with
sup k@x u2 ðx, ÞkL2 ðRþ Þ
x2ð0, 1Þ
Ckh2 kH 1=3 ðRþ Þ
ð2:19Þ
for all h2 2 H 1=3 ðRþ Þ.
Proof. Let u2 ðx, tÞ ¼ I2 ðx, tÞ þ II2 ðx, tÞ with
Z
3
X
1 þi1 st j, 2 ðsÞ j ðsÞx ^
I2 ðx, tÞ ¼
h2 ðsÞ ds
e
e
2i 0
ðsÞ
j¼1
and
II2 ðx, tÞ ¼
Z
3
X
1 0 st j, 2 ðsÞ j ðsÞx ^
e
h2 ðsÞ ds:
e
ðsÞ
2i i1
j¼1
As in the proof of Proposition 2.7, one has
I2 ðx, tÞ ¼
Z
þ
3
X
1 þ1 ið 3 Þt þj ð Þx j, 2 ð Þ
ð32 1Þh^þ
e
e
2 ð Þ d
þ
2
ð
Þ
1
j¼1
3
þ
3
^
where h^þ
2 ð Þ ¼ h2 ðið ÞÞ and j, 2 ð Þ ¼ j, 2 ðið ÞÞ for j ¼ 1, 2, 3. Note that
1, 2 ðsÞ ¼ 2 e2 3 e3 ,
3, 2 ðsÞ ¼ 1 e1 3 e2 :
2, 2 ðsÞ ¼ 3 e3 1 e1 ,
One readily obtains that, as ! þ1,
þ
1, 2 ð Þ
1,
þ ð Þ
pffi
þ
3
2, 2 ð Þ
e 2 ,
þ
ð Þ
þ
3, 2 ð Þ
1:
þ ð Þ
Using Lemma 2.5, it is adduced that there is a constant C for which
2
3 Z þ1 þ
X
2
j, 2 ð Þ Re þj ð Þ
2
2
kI2 ð , tÞkL2 ð0, 1Þ
þ 1bigÞ2 h^þ
e
þ
2 ð Þð3 1Þ d
ð Þ
j¼1 1
Z þ1
þ
h^2 ð Þð32 1Þ2 d Ckh2 k2 1=3 þ :
C
H ðR Þ
1
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
Korteweg–de Vries Equation on Finite Domain
1411
The same estimate holds for II2 ðx, tÞ. To prove (2.19), note that
Z
3
X
þ
þ
1 þ1 ið 3 Þt þ
j, 2 ð Þ
ð32 1Þh^þ
e
j ð Þej ð Þx þ
2 ð Þ d
2
ð
Þ
1
j¼1
Z
3
X
þ
þ
1 þ1 i t þ
j, 2 ðð ÞÞ ^
h ði Þ d :
¼
e j ðð ÞÞej ðð ÞÞx þ
2 0
ðð ÞÞ 2
j¼1
@x I2 ðx, tÞ ¼
Using the Plancherel theorem (with respect to t),
2
Z þ1
þ
3
X
ðð
ÞÞ
þ
1
^þ
j,
2
j ðð ÞÞx
2
k@x I2 ðx, Þk2L2 ðRþ Þ
þ
j ðð ÞÞe
þ ðð ÞÞ jh2 ði Þj d ,
2
0
j¼1
from which follows
k@x I1 ðx, tÞk2L2 ð0, 1;L2 ðRþ ÞÞ
sup k@x I1 ðx, Þk2L2 ðRþ Þ
x2ð0, 1Þ
C
3 Z
X
j¼1
C
3 Z
X
j¼1
0
þ1
þ1
0
þ
2
þ
sup ej ðð
j ðð ÞÞ
x2ð0, 1Þ
2
ð1 þ Þ2=3 jh^þ
2 ði Þj d
þ
ÞÞx 2 j, 2 ðð
þ
ðð
2
ÞÞ ^
jh ði Þj2 d
ÞÞ 2
Ckh2 k2H 1=3 ðRþ Þ :
The same estimate holds for II2 ðx, tÞ. Moreover, a similar argument as that used
in the proof of Proposition 2.7 shows that both I2 and II2 are continuous from ½0, 1
to the space L2 ðRþ Þ. The proof is complete.
œ
Proposition 2.9. There exists a constant C such that
ku3 kL2 ðRþ ;H 1 ð0, 1ÞÞ þ sup ku3 ð , tÞkL2 ð0, 1Þ Ckh3 kL2 ðRþ Þ
0 t<þ1
ð2:20Þ
and @x u3 2 Cb ð ½0, 1; L2 ðRþ ÞÞ with
sup k@x u3 ðx, ÞkL2 ðRþ Þ
x2ð0, 1Þ
Ckh3 kL2 ðRþ Þ
ð2:21Þ
for all h3 2 L2 ðRþ Þ.
Proof. The function u3 ðx, tÞ can be written in the form u3 ðx, tÞ ¼ I3 ðx, tÞ þ II3 ðx, tÞ with
Z
3
X
1 þi1 st j, 3 ðsÞ j ðsÞx ^
e
e
h3 ðsÞ ds
I3 ðx, tÞ ¼
2i 0
ðsÞ
j¼1
and
II3 ðx, tÞ ¼
Z
3
X
1 0 st j, 3 ðsÞ j ðsÞx ^
e
h3 ðsÞ ds:
e
2i i1
ðsÞ
j¼1
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
1412
Bona, Sun, and Zhang
As in the proof of Proposition 2.7, one has
Z
þ
3
X
1 þ1 ið 3 Þt þj ð Þx j, 3 ð Þ
ð32 1Þh^þ
e
e
I3 ðx, tÞ ¼
2 ð Þ d
þ
2
ð
Þ
1
j¼1
3
þ
3
^
where h^þ
3 ð Þ ¼ h3 ðið ÞÞ and j, 3 ð Þ ¼ j, 3 ðið ÞÞ, j ¼ 1, 2, 3. Since
1, 3 ðsÞ ¼ e3 ðsÞ e2 ðsÞ ,
2, 3 ðsÞ ¼ e1 ðsÞ e3 ðsÞ
and
3, 3 ðsÞ ¼ e2 ðsÞ e1 ðsÞ ,
it follows that
þ
1, 3 ð Þ
1 ,
þ ð Þ
pffiffi
þ
2, 3 ð Þ
1 eð 3=2Þ ,
þ
ð Þ
þ
3, 2 ð Þ
1
þ ð Þ
as ! þ1. The remainder of the proof follows the lines developed above.
œ
Let h~ðtÞ ¼ ðh1 ðtÞ, h2 ðtÞ, h2 ðtÞÞ and write the solution u of (2.11) as
uðtÞ ¼ Wb ðtÞh~ ¼
3
X
j¼1
Wj ðtÞhj
ð2:22Þ
where the spatial variable x is suppressed and the Wj are as defined in (2.14). For
s 0 and T > 0, let
Hs, T ¼ H ðsþ1Þ=3 ð0, T Þ H ðsþ1Þ=3 ð0, T Þ H s=3 ð0, T Þ:
For any h~ 2 Hs, T ,
1=2
kh~kHs, T kh1 k2H ðsþ1Þ=3 ð0, T Þ þ kh2 k2H ðsþ1Þ=3 ð0, T Þ þ kh3 k2H s=3 ð0, T Þ
:
If T ¼ 1, denote Hs, T by Hs . Combining Propositions 2.7–2.9 yields the following
theorem about the linear IBVP (2.11).
Theorem 2.10. For any h~ 2 H0 , the IBVP Eq. (2.11) admits a unique solution
uðx, tÞ ¼ ½Wb ðtÞh~ðtÞðxÞ
which belongs to the space Cb ðRþ ; L2 ð0, 1ÞÞ \ L2 ðRþ ; H 1 ð0, 1ÞÞ with ux 2 Cb ð ½0, 1;
L2 ðRþ ÞÞ. Moreover there exists a constant C such that
kukL2 ðRþ ;H 1 ð0, 1ÞÞ þ sup kuð , tÞkL2 ð0, 1Þ
0 t<þ1
Ckh~kH0
ð2:23Þ
and
sup kux ðx, ÞkL2 ðRþ Þ
x2ð0, 1Þ
for all h~ 2 H0 .
Ckh~kH0
ð2:24Þ
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
Korteweg–de Vries Equation on Finite Domain
1413
Remark 2.11. The estimate (2.23) reveals a Kato-type smoothing effect of the system
(2.11) while (2.24) shows that the system (2.11) possesses a stronger smoothing effect,
the so-called sharp Kato smoothing (see Kato (1983), Kenig et al. (1991a); Vega
(1988)). It is this smoothing property that provides the rationale for being able to
impose in a strong sense the boundary condition ux ð1, tÞ ¼ h3 ðtÞ in (2.11).
Remark 2.12. The condition imposed on h~ in Theorem 2.10 appears to be sharp in
the context of our approach to the analysis. From the explicit solution formula for
(2.11), using arguments similar to those appearing in the proof of Propositions 2.7–
2.9, one shows that u 2 Cb ð ½0, 1; Ht1=3 ðRþ ÞÞ and
sup kuðx, ÞkH 1=3 ðRþ Þ
x2ð0, 1Þ
Ckh~kH0 :
Of course, this does not mean that results with s < 0 are not possible, mearly that the
present argument would not be adequate to the task.
Finally we return to the homogeneous IBVP (2.1) to show that it possesses the
sharp Kato-smoothing property. Let a function be defined on the interval ð0, 1Þ
and let be its extension by zero to the whole line R. Assume that v ¼ vðx, tÞ is the
solution of
vt þ vx þ vxxx ¼ 0,
vðx, 0Þ ¼ ðxÞ
for x 2 R, t 0. If
g1 ðtÞ ¼ vð0, tÞ,
g2 ðtÞ ¼ vð1, tÞ,
g3 ðtÞ ¼ vx ð1, tÞ,
then in terms of Wb ðtÞ defined in (2.22),
vg~ ¼ vg~ðx, tÞ Wb ðtÞ~
g
is the corresponding solution of the nonhomogeneous boundary-value problem
Eq. (2.1) with boundary conditions hj ðtÞ ¼ gj ðtÞ, j ¼ 1, 2, 3, for t 0. It is clear that
for x 2 ð0, 1Þ, the function vðx, tÞ vg~ðx, tÞ solves the IBVP (2.1), and this in turn
leads to a representation of the semigroup W0 ðtÞ in terms of Wb ðtÞ and WR ðtÞ,
where WR ðtÞ is the C0 -semigroup in the space L2 ðRÞ generated by the operator AR
defined by
AR f ¼ f 0 f 000
with domain DðAR Þ ¼ H 3 ðRÞ and vðx, tÞ ¼ WR ðtÞ ðxÞ.
Proposition 2.13. For any 2 L2 ð0, 1Þ, if is its zero-extension to R, then W0 ðtÞ may
be written in the form
W0 ðtÞ ¼ WR ðtÞ Wb ðtÞ~
g
for any x 2 ð0, 1Þ, t > 0, where g~ ¼ ðg1 , g2 , g3 Þ,
g1 ðtÞ ¼ vð0, tÞ,
with v ¼ WR ðtÞ .
g2 ðtÞ ¼ vð1, tÞ,
g3 ðtÞ ¼ vx ð1, tÞ,
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
1414
Bona, Sun, and Zhang
Remark 2.14. Of course, g~ depends upon the particular extension of chosen
here. However, the intrusion of is simply as an intermediary for obtaining the
trace estimate in Proposition 2.16 below. It plays no other role in the theory.
To have appropriate estimates of W0 ðtÞ, the following trace result related to the
semi-group WR ðtÞ is needed.
Lemma 2.15. There exists a constant C such that for any
WR ðtÞ ðxÞ satisfies
sup kvðx, ÞkH 1=3 ðRÞ
x2R
2 L2 ðRÞ, vðx, tÞ ¼
Ck kL2 ðRÞ :
Moreover, vx 2 Cb ðR; L2 ðRÞÞ and
sup kvx ðx, ÞkL2 ðRÞ
x2R
Ck kL2 ðRÞ :
Proof. This lemma follows as a special case of Lemma 2.1 in Kenig et al., (1991b)
except the continuity of vx ðx, Þ from R to the space L2 ðRÞ, which can be verified using
Fatou’s lemma and the argument that appears in the proof of Proposition 2.7. œ
The following estimate for W0 ðtÞ follows from Proposition 2.13, Lemma 2.15,
and the estimates of Wb ðtÞ established earlier in Theorem 2.10.
Proposition 2.16. For any 2 L2 ð0, 1Þ, u ¼ W0 ðtÞ has the property
ux 2 Cb ð ½0, 1; L2, t ðRþ ÞÞ
and there exists a constant C such that
sup kux ðx, ÞkL2 ðRþ Þ CkkL2 ð0, 1Þ :
x2ð0, 1Þ
We conclude this section with the following proposition which, like the foregoing
results, will be needed later (see Proposition 3.2).
Proposition 2.17. Let T > 0 be given and
Zt
uðx, tÞ ¼
W0 ðt Þ f ð , Þ d:
0
Then
sup kux ðx, ÞkL2 ð0, T Þ
x2ð0, 1Þ
C
Z
0
T
k f ð , ÞkL2 ð0, 1Þ d:
Proof. Observe that
Z
Zt
ux ðx, tÞ ¼
@x ðW0 ðt Þ f ð , ÞÞ d ¼
0
T
0
ð0, tÞ ðÞ@x ðW0 ðt Þ f ð , ÞÞ d
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
Korteweg–de Vries Equation on Finite Domain
1415
where
1 if 2 ð0, tÞ;
0 if > t:
Using the Minkowski’s integral inequality gives us
ð0, tÞ ðÞ ¼
n
Z
kux ðx, ÞkL2 ð0, T Þ
¼
T
0
Z
T
0
Z
0
Z
1=2
T
ð0, tÞ ðÞ@x ðW0 ðt Þ f ð , ÞÞ2 dt
1=2
T
@x ðW0 ðt Þ f ð , ÞÞ2 dt
d
d:
Thus, invoking Proposition 2.16 (with the initial time Þ gives us
sup kux ðx, ÞkL2 ð0, T Þ
x2ð0, 1Þ
Z
T
sup
x2ð0, 1Þ
0
C
Z
0
T
Z
T
1=2
@x ðW0 ðt Þ f ð , ÞÞ2 dt
d
k f ð , ÞkL2 ð0, 1Þ d:
œ
The proof is complete.
Remark 2.18. It is worth highlighting the crucial role played by the formulas (2.14)
and following, which provided an explicit representation of solutions directly in
terms of the boundary data. Our theory devolves in large part on the efficacy of
these formulas.
3. LOCAL WELL-POSEDNESS
In this section, attention will be given to the full nonlinear IBVP
ut þ ux þ uux þ uxxx ¼ 0,
uðx, 0Þ ¼ ðxÞ,
uð0, tÞ ¼ h1 ðtÞ,
uð1, tÞ ¼ h2 ðtÞ,
ux ð1, tÞ ¼ h3 ðtÞ
ð3:1Þ
introduced at the outset of our discussion.
For any T > 0 and s 0, let Xs, T be as defined in (1.11) with its usual product
topology and let Ys, T be the collection of
v 2 Cð ½0, T; H s ð0, 1ÞÞ \ L2 ð ½0, T; H sþ1 ð0, 1ÞÞ
with vx 2 Cð ½0, 1; L2 ð0, T ÞÞ. A norm k kYs, T on the space Ys, T is defined by
1=2
kvkYs, T :¼ kvk2Cð ½0, T;H s ð0, 1ÞÞ þ kvk2L2 ð ½0, T;H sþ1 ð0, 1ÞÞ þ kvx k2Cð ½0, 1;L2 ð0, T ÞÞ
for v 2 Ys, T .a The space Ys, T possesses the following helpful property.
a
The reader may notice that the space Ys;T need not include the finiteness of kvx kCð½0;1;L2 ð0;TÞÞ
for the arguments that follow to be valid, and hence for proving well-posedness of the IBVP
Eq. (3.1) in the space H s ð0; 1Þ. However, by keeping this term, we are able to determine at a
stroke that solutions of Eq. (3.1) possess the sharp Kato smoothing effect.
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
1416
Bona, Sun, and Zhang
Lemma 3.1. Let s 0 be given. There exists a constant C such that for any T > 0 and
u, v 2 Ys, T ,
ZT
kðuð , tÞvð , tÞÞx kH s ð0, 1Þ dt CðT 1=2 þ T 1=3 ÞkukYs, T kvkYs, T :
ð3:2Þ
0
Proof. The proof is given for 0 s 1. The proof for other values of s is similar.
Notice first that
ZT
ZT
ZT
kuð , tÞvx ð , tÞkL2 ð0,1Þ dt:
kux ð ,tÞvð ,tÞkL2 ð0,1Þ dt þ
kðuð , tÞvð , tÞÞx kL2 ð0, 1Þ dt
0
0
0
Using the Poincare inequality, there obtains
kuð , tÞvx ð ,tÞkL2 ð0,1Þ
kuð ,tÞkL1 ð0, 1Þ kvx ð ,tÞkL2 ð0,1Þ
1=2
Cðkuð , tÞkL2 ð0,1Þ þ kuð , tÞk1=2
L2 ð0,1Þ kux ð , tÞkL2 ð0,1Þ Þkvx ð ,tÞkL2 ð0,1Þ :
These two terms, when integrated with respect to t, are bounded thusly:
ZT
ZT
kvx ð ,tÞkL2 ð0,1Þ dt
kuð ,tÞkL2 ð0,1Þ kvx ð ,tÞkL2 ð0,1Þ dt sup kuð ,tÞkL2 ð0,1Þ
0 t T
0
0
T 1=2 sup kuð ,tÞkL2 ð0,1Þ
0 t T
CT
1=2
Z
T
0
1=2
kvx ð ,tÞk2L2 ð0,1Þ dt
kukY0,T kvkY0,T
and
Z
T
0
1=2
kuð , tÞk1=2
L2 ð0, 1Þ kux ð , tÞkL2 ð0, 1Þ kvx ð , tÞkL2 ð0, 1Þ dt
Z
sup kuð , tÞk1=2
L2 ð0, 1Þ
0 t T
CT
1=3
kukY0, T kvkY0, T :
T
0
1=4
kux ð , tÞk2L2 ð0, 1Þ dt
Z
0
T
3=4
kvx ð , tÞkL4=3
dt
2 ð0, 1Þ
The last three inequalities combine to establish that
ZT
kuð , tÞvx ð , tÞkL2 ð0, 1Þ dt CðT 1=2 þ T 1=3 ÞkukY0, T kvkY0, T :
0
Similarly, one sees that
ZT
kux ð , tÞvð , tÞkL2 ð0, 1Þ dt
0
CðT 1=2 þ T 1=3 ÞkukY0, T kvkY0, T :
In consequence, estimate (3.2) holds with s ¼ 0. To see that (3.2) is true for s ¼ 1,
argue as follows. Observe that
kðuð , tÞvð , tÞÞx ÞkH 1 ð0, 1Þ
kðuð , tÞvð , tÞÞx kL2 ð0, 1Þ þ kðuð , tÞvð , tÞÞxx kL2 ð0, 1Þ
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
Korteweg–de Vries Equation on Finite Domain
1417
and that
kðuð , tÞvð , tÞÞxx kL2 ð0, 1Þ
kðux ð , tÞvð , tÞÞx kL2 ð0, 1Þ þ kðuð , tÞvx ð , tÞÞx kL2 ð0, 1Þ :
The inequality (3.2) with s ¼ 0, just established, gives
Z
T
CðT 1=2 þ T 1=3 Þðkux kY0, T kvkY0, T þ kukY0, T kvx kY0, T Þ
kðuð , tÞvð , tÞÞxx kL2 ð0, 1Þ dt
0
CðT 1=2 þ T 1=3 ÞkukY1, T kvkY1, T
which together with (3.2) (again with s ¼ 0) yields
Z
T
0
kuð , tÞvð , tÞkH 1 ð0, 1Þ dt
CðT 1=2 þ T 1=3 ÞkukY1, T kvkY1, T :
The estimate (3.2) with 0 < s < 1 now follows from the nonlinear interpolation
theory developed in Bona and Scott (1976). The proof is complete.
œ
The next step is to show that the IBVP (3.1) is locally well-posed in the space X0, T .
Proposition 3.2. Let T > 0 be given. For any ð, h~Þ 2 X0, T with h~ ¼ ðh1 , h2 , h3 Þ, there is
a T 2 ð0, T depending on kð, h~ÞkX0, T such that the IBVP Eq. (3.1) admits a unique
solution u 2 Y0, T . Moreover, for any T 0 < T , there is a neighborhood U of ð, h~Þ
such that the IBVP Eqs. (1.1)–(1.3) admits a unique solution in the space Y0, T 0 for
any ð , h~1 Þ 2 U and the corresponding solution map from U to Y0, T 0 is Lipschitz
continuous.
Proof. Write the IBVP (3.1) in its integral equation form
Zt
uðtÞ ¼ W0 ðtÞ þ Wb ðtÞh~
W0 ðt Þðuux ÞðÞ d
0
ð3:3Þ
where the operator Wb ðtÞ is as defined in formulas (2.14) and (2.22) in Sec. 2 and
the spatial variable is suppressed throughout. For given ð, h~Þ 2 X0, T , let r > 0 and
> 0 be constants to be determined. Let
S, r ¼ fv 2 Y0, , kvkY0,
rg:
The set S, r is a closed, convex, and bounded subset of the space Y0, and
therefore is a complete metric space in the topology induced from Y0, . Define a
map on S, r by
Zt
ðvÞ ¼ W0 ðtÞ þ Wb ðtÞh~
W0 ðt Þðvvx ÞðÞ d
0
for v 2 S, r . The crux of the matter is the following inequality. For any v 2 S, r ,
kðvÞkY0,
C0 kð, h~ÞkX0, T þ C1
Z
kvvx ð , ÞkL2 ð0, 1Þ d
0
1=2
C0 kð, h~ÞkX0, T þ C1 ð
þ 1=3 Þkvk2Y0,
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
1418
Bona, Sun, and Zhang
where C0 and C1 are constants. As the norm on Y0, has three parts, this amounts
to three inequalities, all of which follow immmediately from the linear estimates in
Sec. 2 and in Lemma 3.1. Choosing r > 0 and > 0 so that
(
r ¼ 2C0 kð, h~ÞkX0, T ,
C1 ð1=2 þ 1=3 Þ r
ð3:4Þ
1
2,
then
kðvÞkY0,
r
for any v 2 S, r . Thus, with such a choice of r and , maps S, r into S, r . The same
inequalities allow one to deduce that for r and chosen as in (3.4),
kðv1 Þ ðv2 ÞkY0,
1
kv v2 kY0,
2 1
for any v1 , v2 2 S, r . In other words, the map is a contraction mapping of Sr, . Its
fixed point u ¼ ðuÞ is the unique solution of the IBVP (1.1)–(1.3) in S, r .
œ
Consider the forced linear problem
ut þ ux þ uxxx ¼ f ,
uð0, tÞ ¼ h1 ðtÞ,
uðx, 0Þ ¼ ðxÞ,
uð1, tÞ ¼ h2 ðtÞ,
ux ð1, tÞ ¼ h3 ðtÞ:
)
ð3:5Þ
Applying the linear estimates derived in Sec. 2, for ð, h~Þ 2 X0, T and f 2 L1 ð0, T;
L2 ð0, 1ÞÞ, the corresponding solution u of (3.5) belongs to the space Y0, T and satisfies
kukY0, T
Cðkð, h~ÞkX0, T þ k f kL1 ð0, T;L2 ð0, 1ÞÞ Þ
ð3:6Þ
for some constant C independent of , hj , j ¼ 1, 2, 3 and f. The next lemma gives an
estimate for solutions of (3.5) in the space Ys, T with s in the range of 0 s 3.
Lemma 3.3. For given T > 0 and s in the range ½0, 3, let there be given
f 2 W s=3, 1 ð ½0, T; L2 ð0, 1ÞÞ and ð, h~Þ 2 Xs, T satisfying the compatibility conditions
ð0Þ ¼ h1 ð0Þ,
ð0Þ ¼ h1 ð0Þ,
ð1Þ ¼ h2 ð0Þ
ð1Þ ¼ h2 ð0Þ,
0 ð1Þ ¼ h3 ð0Þ
if 1/2 < s
if 3/2 < s
3/2 , or
3.
ð3:7Þ
Then Eq. (3.5) admits a unique solution u 2 Ys, T and
kukYs, T
Cðkð, h~ÞkXs, T þ k f kW s=3, 1 ð0, T;L2 ð0, 1ÞÞ Þ
ð3:8Þ
for some constant C > 0 independent of , h~, and f. Moreover, if s ¼ 3, ut 2 Y0, T and
kut kY0, T
Cðkð, h~ÞkX3, T þ k f kW s=3, 1 ð0, T;L2 ð0, 1ÞÞ Þ:
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
Korteweg–de Vries Equation on Finite Domain
1419
Proof. The proof is provided for s ¼ 3 since the result for other values of s can be
established by interpolation and (3.6). For the solution u of (3.5), let v ¼ ut . Then the
function v is a solution of
)
vt þ vx þ vxxx ¼ ft , vðx, 0Þ ¼ f ðx, 0Þ 000 ðxÞ 0 ðxÞ,
ð3:9Þ
vð0, tÞ ¼ h01 ðtÞ, vð1, tÞ ¼ h02 ðtÞ, vx ð1, tÞ ¼ h03 ðtÞ:
Applying (3.6) to v in (3.9) yields that
Cðk ft kL1 ð0, T;ð0, 1ÞÞ þ kð f ð , 0Þ 000 ð Þ 0 ð Þ, h~0 ÞkX0, T Þ:
kvkY0, T
Define
uðx, tÞ ¼
Z
t
0
vðx, Þ d þ ðxÞ:
Then uðx, 0Þ ¼ ðxÞ and
Zt
uð0, tÞ ¼
vð0, Þ d þ ð0Þ
0
Zt
h01 ðÞ d þ ð0Þ
¼
0
¼ h1 ðtÞ h1 ð0Þ þ ð0Þ ¼ h1 ðtÞ:
Similarly, uð1, tÞ ¼ h2 ðtÞ and ux ð1, tÞ ¼ h3 ðtÞ. Furthermore, it is easily verified that
ut ðx, tÞ þ ux ðx, tÞ þ uxxx ðx, tÞ
Zt
¼ vðx, tÞ þ ðvx ðx, Þ þ vxxx ðx, ÞÞ d þ 0 ðxÞ þ 000 ðxÞ
0
Zt
¼ vðx, 0Þ þ ð ft ðx, Þ vx ðx, Þ vxxx ðx, ÞÞ d
0
Zt
þ ðvx ðx, Þ þ vxxx ðx, ÞÞ d þ 0 ðxÞ þ 000 ðxÞ ¼ 0:
0
Thus u solves the IBVP (3.5). Since
uxxx ¼ f ut ux ¼ f v ux ,
it follows that u 2 Y3, T and satisfies (3.8) with s ¼ 3. The proof is complete.
œ
Here is the promised local well-posedness result for the IBVP (3.1) in Xs, T .
Theorem 3.4. Let T > 0 and s 0 be given. Suppose that ð, h~Þ 2 Xs, T satisfies the
s-compatibility conditions. Then there exists a T 2 ð0, T depending only on
kð, h~ÞkXs, T such that Eq. (3.1) admits a unique solution u 2 Ys, T with
@ jt u 2 Ys3j, T
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
1420
Bona, Sun, and Zhang
for j ¼ 0, 1, 2, . . . , ½s=3 1, ½s=3. Moreover, for any T 0 < T , there is a neighborhood
U of ð, h~Þ such that the IBVP Eq. (3.1) admits a unique solution in the space Ys, T 0 for
any ð , h~1 Þ 2 U and the corresponding solution map is Lipschitz continuous.
Proof. For given s-compatible ð, h~Þ 2 Xs, T , let r > 0 and > 0 be given and S, r
be the collection of functions v in the space Cð ½0, ; L2 ð0, 1ÞÞ \ L2 ð0, ; H 1 ð0, 1ÞÞ
satisfying
@ tj v 2 Y3, , for j ¼ 0, 1, 2, . . . , ½s=3 1 and @ t½s=3 v 2 Ys3½s=3, ,
and
k@½s=3
vkYs3½s=3 þ
t
½s=31
X
j¼0
k@ tj vkY3,
r:
Let
Y s, ¼ Ys3½s=3,
½s=31
Y
Y3,
j¼0
with the usual product topology. Then the set S, r may be viewed as a closed subset
of Y s, via the mapping v ! ðv, @t v, . . . , @ x½s=3 vÞ v~, and therefore is a complete
metric space. For any v 2 S, r , consider the system of equations
9
P
k
ðkÞ
ðkÞ
ðkÞ
ð j Þ ðkjÞ
1
k!
=
uðkÞ
¼
þ
u
þ
u
,
u
ðx,
0Þ
¼
ðxÞ,
@
v
v
k
xxx
x
t
j¼0 j!ðkjÞ!
2 x
ð3:10Þ
;
ðkÞ
ðkÞ
ðtÞ,
uð1,
tÞ
¼
h
ðtÞ,
u
ð1,
tÞ
¼
h
ðtÞ,
uð0, tÞ ¼ hðkÞ
x
1
2
3
ðkÞ
for k ¼ 0, 1, 2, . . . , ½s=3, where uðkÞ @ kt u, vðkÞ @ kt v and k , h1ðkÞ , hðkÞ
2 , h3 are defined
in (1.7) and (1.8). By Lemma 3.3, the IBVP (3.10) defines a map from S, r to the
space Y s, . Moreover,
kð~
vÞkY s, Ckð, h~ÞkXs, T þ C 1=2 þ 1=3 k~
vkY s,
for some constant C independent of h~, , and . Thus, the argument presented in the
proof of Proposition 3.2 shows that is a contraction map from S, r to S, r if r and
are appropriately chosen. As a result, its fixed point u~ 2 S, r is the unique solution of
(3.5). Thus the proof is complete when s 3. In case s > 3, the result just established
shows that
uð j Þ 2 Cð ½0, ; H 3 ð0, 1ÞÞ \ L2 ð ½0, ; H 4 ð0, 1ÞÞ
for j ¼ 0, 1, . . . , ½s=3 1 and
u½s=3 2 Ys3½s=3, ¼ Cð ½0, ; H s3½s=3 ð0, 1ÞÞ \ L2 ð ½0, ; H sþ13½s=3 ð0, 1ÞÞ:
In case k ¼ ½s=3 1, (3.10) implies that
ð ½s=31Þ
uxxx
¼
utð ½s=31Þ
uxð ½s=31Þ
1
@x
2
ð ½s=31Þ
X
j¼0
C j½s=31 uð j Þ uð ½s=31jÞ
!
:
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
Korteweg–de Vries Equation on Finite Domain
1421
We thereby arrive at the conclusion
uð ½s=31Þ 2 Cð ½0, ; H s3½s=3þ2 ð0, 1ÞÞ:
It is further implied that the left-hand side of the last equation belongs to
Cð ½0, ; H s3ð ½s=3 ð0, 1ÞÞ \ L2 ð ½0, ; H sþ13½s=3 ð0, 1ÞÞ:
Consequently, it must be the case that
uð ½s=31Þ 2 Cð ½0, ; H sþ33ð ½s=3 ð0, 1ÞÞ \ L2 ð ½0, ; H sþ43½s=3 ð0, 1ÞÞ:
Repeating this argument
L2 ð ½0, ; H sþ1 ð0, 1ÞÞ with
if
necessary
yields
that
u 2 Cð ½0, ; H s ð0, 1ÞÞ \
@ jt u 2 Cð ½0, ; H s3j ð0, 1ÞÞ \ L2 ð ½0, ; H sþ13j ð0, 1ÞÞ
for j ¼ 1, 2, . . . , ½s=3 1 and
@ t½s=3 u 2 Cð ½0, ; H s3½s=3 ð0, 1ÞÞ \ L2 ð ½0, ; H sþ13½s=3 ð0, 1ÞÞ:
œ
The proof is complete.
4. GLOBAL WELL-POSEDNESS
The results presented in Theorem 3.4 are local in the sense that the time interval
ð0, T Þ on which the solution exists depends on kð, h~ÞkXs, T . In general, the larger
kð, h~ÞkXs, T , the smaller will be T . However, if T ¼ T no matter what the size of
kð, h~ÞkXs, T , the IBVP (3.1) is said to be globally well-posed. In this section we study
global well-posedness of the problem (3.1). First we introduce a helpful Banach
space. For given s 0 and T > 0, let
Zs, T H s ð0, 1Þ H þð5sþ9Þ=18 ð0, T Þ H þð5sþ9Þ=18 ð0, T Þ H þð5sþ3Þ=18 ð0, T Þ
if 0
s
Zs, T
3 and
Xs, T
if s > 3, where is any positive constant. Of course, for s 3, Zs, T depends on , but
this dependence is suppressed. The Sobolev indices when s lies in ½0, 3 may look a
little odd. We feel it likely that they are an artifact of our proof. The strange indices
derive from slightly inadequate smoothing results and are the best we can do with
what is in hand. Note this inelegance ceases as soon as s 3, and hence for the case
of classical solutions when s > 7=2. The same issue arose in Bona et al., (2001) for
the quarter-plane problem, so the issue does not necessarily devolve upon the third
boundary condition ux ð1, tÞ ¼ h3 ðtÞ.
Theorem 4.1. Let T > 0 and s 0. For any s-compatible ð, h~Þ 2 Zs, T , the IBVP
Eq. (3.1) admits a unique solution u 2 Ys, T with @ jt u 2 Ys3j, T for j ¼ 0, 1, 2, . . . , ½s=3.
Moreover, the corresponding solution map of the IBVP Eq. (3.1) is Lipschitz continuous.
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
1422
Bona, Sun, and Zhang
Proof of Theorem 4.1. In the context of an established local well-posedness result,
it suffices to prove the following global a priori H s -estimate for smooth solutions of
the IBVP (3.1).
Proposition 4.2. For given T > 0 and s 0, there exists a continuous and nondecreasing function s : Rþ ! Rþ such that for any smooth solution u of Eq. (3.1),
sup kuð , tÞkH s ð0, 1Þ
0 t T
~
s ðkð, hÞkZs, T Þ:
ð4:1Þ
The proof of Proposition 4.3 consists of four parts. In part (i), estimate (4.1) is
shown to be true for s ¼ 0. In part (ii), estimate (4.1) is shown to be true for s ¼ 3.
Then, in part (iii), Tartar’s nonlinear interpolation theory is used to show that (4.1)
holds for 0 < s < 3. The validity of (4.1) for other values of s is established in part
(iv).
Part (i). For a smooth solution u of the IBVP (3.1), write u ¼ w þ v, where v solves
)
vt þ vx þ vxxx ¼ 0,
vðx, 0Þ ¼ ðxÞ,
ð4:2Þ
vð0, tÞ ¼ h1 ðtÞ, vð1, tÞ ¼ h2 ðtÞ, vx ð1, tÞ ¼ h3 ðtÞ,
with
ðxÞ ¼ ð1 xÞh1 ð0Þ þ xh2 ð0Þ þ xð1 xÞðh3 ð0Þ h2 ð0Þ þ h1 ð0ÞÞ
and w solves
wt þ wx þ wwx þ wxxx ¼ ðwvÞx vvx ,
wð0, tÞ ¼ 0,
wð1, tÞ ¼ 0,
wx ð1, tÞ ¼ 0:
)
wðx, 0Þ ¼ ðxÞ ðxÞ,
ð4:3Þ
By Lemma 3.3
Ckð , h~ÞkXs, T
kvkYs, T
for 0
s
3. In particular, for 3=2 < s
kvkYs, T
ð4:4Þ
3,
Cðkh1 kH s=3 ð0, T Þ þ kh2 kH s=3 ð0, T Þ þ kh3 kH ðs1Þ=3 ð0, T Þ Þ
ð4:5Þ
since
k kH s ð0, 1Þ
Cðkh1 kH s=3 ð0, T Þ þ kh2 kH s=3 ð0, T Þ þ kh3 kH ðs1Þ=3 ð0, T Þ Þ:
Multiply both sides of the equation in (4.3) by w and integrate over ð0, 1Þ with respect
to x. Integration by parts leads to
Z1
Z1
d
2
2
kwð , tÞkL2 ð0, 1Þ C
jvx ð , tÞw ð , tÞj dx þ C
jvx ð , tÞvð , tÞwð , tÞj dx:
dt
0
0
Observe that
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
Korteweg–de Vries Equation on Finite Domain
Z
1
0
jvx ð , tÞw2 ð , tÞj dx
1423
sup vx ðx, tÞkwð , tÞk2L2 ð0, 1Þ
x2ð0, 1Þ
C kvð , tÞkH 3=2þ ð0, 1Þ kwð , tÞk2L2 ð0, 1Þ
and
Z
1
0
jvx ð , tÞvð , tÞwð , tÞj dx
sup vx ðx, tÞkvð , tÞkL2 ð0, 1Þ kwð , tÞkL2 ð0, 1Þ
x2ð0, 1Þ
C kvð , tÞk2H 3=2þ ð0, 1Þ kwð , tÞkL2 ð0, 1Þ
where is any fixed positive constant.
In consequence, one has that
d
kwð , tÞkL2 ð0, 1Þ
dt
C kvð , tÞkH 3=2þ ð0, 1Þ kwð , tÞkL2 ð0, 1Þ þ C kvð , tÞk2H 3=2þ ð0, 1Þ
for any t 0 The estimate (4.1) with s ¼ 0 then follows by using Gronwall’s inequality and (4.5).
Part (ii). For a smooth solution u, v ¼ ut solves
vt þ vx þ ðuvÞx þ vxxx ¼ 0,
vð0, tÞ ¼ h01 ðtÞ,
vðx, 0Þ ¼ ðxÞ
vð1, tÞ ¼ h02 ðtÞ,
vx ð1, tÞ ¼ h03 ðtÞ
)
where ðxÞ ¼ 0 ðxÞ 0 ðxÞðxÞ 000 ðxÞ. By Lemma 3.3, there exists a constant
C > 0 such that for any T 0 T,
kvkY0, T 0
Ckð , h~0 ÞkX0, T þ CðT 01=2 þ T 01=3 ÞkukY0, T kvkY0, T :
Choose T 0 T such that CðT 01=2 þ T 01=3 ÞkukY0, T ¼ 1=2; with such a choice,
kvkY0, T 0 2Ckð , h~0 ÞkX0, T : Note that T 0 only depends on kukY0, T , and therefore
depends only on kð, h~ÞkZ0, T by the estimate proved in Part (i). By a standard density
argument,
kvkY0, T
C1 kð, h~ÞkZ3, T
where C1 depends only on T and kð, h~ÞkZ0, T . The estimate (4.1) with s ¼ 3 then
follows from
v ¼ ðuxxx þ ux þ uux Þ
by a now familiar argument.
Part (iii). Here is a précis of the (real) interpolation theory as it will be used below.
Let B0 and B1 be two Banach spaces such that B1 B0 with the inclusion map
continuous. Let f 2 B0 and, for t 0, define
Kð f , tÞ ¼ inf fk f gkB0 þ tkgkB1 g:
g2B1
For 0 < < 1 and 1
p
þ1, define
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
1424
Bona, Sun, and Zhang
½B0 , B1 , p ¼ B, p ¼
(
f 2 B0 : k f k, p ¼
Z
þ1
1=p
p p1
Kð f , tÞ t
0
dt
< þ1
)
with the usual modification for the case p ¼ þ1. Then B, p is a Banach space with
norm kk, p . Given two pairs of indices ð1 , p1 Þ and ð2 , p2 Þ as above, then
ð1 , p1 Þ ! ð2 , p2 Þ means
1 < 2 ,
1 ¼ 2
or
and p1 > p2 :
If ð1 , p1 Þ ! ð2 , p2 Þ then B2 , p2
B1 , p1 with the inclusion map continuous.
Theorem 4.3. (Bona and Scott, 1976) Let B0j and B1j be Banach spaces such that
B j1 B0j with continuous inclusion mappings, j ¼ 1, 2. Let and q lie in the ranges
0 < < 1 and 1 q þ1. Suppose A is a mapping such that
(i )
A : B1, q ! B20 and for f , g 2 B1, q ,
kAf AgkB2
0
C0 ðk f kB1, q þ kgkB1, q Þk f gkB1
0
and
(ii)
A : B11 ! B21 and for h 2 B11
kAhkB2
1
C1 ðkhkB1, q ÞkhkB1 ,
1
where Cj : Rþ ! Rþ are continuous nondecreasing functions, j ¼ 0, 1.
Then if ð, pÞ ð, qÞ, A maps B1, p into B2, p and for f 2 B1, p
kAf kB2, p
Cðk f kB1, q Þk f kB1, p ,
where for r > 0, CðrÞ ¼ 4C0 ð4rÞ1 C1 ð3rÞ :
Remark 4.4. This theorem is identical with Theorem 2 of Tartar (1972) except that
Tartar makes the more restrictive assumption that the constants C0 and C1 depend
only on the B10 norms of the functions in question. Theorem 4.3 was used by
Bona and Scott to give the first proof of global well-posedness of the pure initialvalue problem for the KdV-equation on the whole line in fractional order Sobolev
spaces H s(R).
To prove that estimate (4.1) holds for T > 0 and 0 s 3, let
Z s, T ¼ fð, h~Þ 2 Zs, T satisfying s-compatibility conditiong
with the inherited norm from the space Zs, T . Choose
B10 ¼ Z 0, T ,
B11 ¼ Z 3, T ,
B20 ¼ Cð ½0, T; L2 ð0, 1ÞÞ,
B21 ¼ Cð ½0, T; H 3 ð0, 1ÞÞ:
Let A be the solution map of the IBVP (3.1): u ¼ Að, h~Þ. For given s with 0 < s < 3,
choose p ¼ 2 and ¼ s=3. Then
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
Korteweg–de Vries Equation on Finite Domain
B2, p ¼ Cð ½0, T; H s ð0, 1ÞÞ,
1425
B1, p ¼ Z s, T :
In this case, assumption (ii) of Theorem 4.3 is (4.1) with s ¼ 3, which we have already
proved. It remains to verify assumption (i) of Theorem 4.3.
To this end, let u1 ¼ Að1 , h~1 Þ, u2 ¼ Að2 , h~2 Þ, and w ¼ u1 u2 . It is seen that w
solves the variable coefficient problem
)
wt þ wx þ ðzwÞx þ wxxx ¼ 0, vðx, 0Þ ¼ 1 ðxÞ 2 ðxÞ
wð0, tÞ ¼ h1, 1 ðtÞ h2, 1 ðtÞ,
wx ð1, tÞ ¼ h1, 2 ðtÞ h2, 2 ðtÞ, wð1, tÞ ¼ h1, 3 ðtÞ h2, 3 ðtÞ
with z ¼ ð1=2Þðu1 u2 Þ. Applying Lemma 3.3 with s ¼ 0 yields that, for any
0 T 0 T,
kwkY0, T 0
Cðkð1 , h~1 Þ ð2 , h~2 ÞkX0, T þ kzwkL1 ð0, T 0 ;L2 ð0, 1ÞÞ Þ
Ckð1 , h~1 Þ ð2 , h~2 ÞkX0, T þ C T 01=2 þ T 01=3 kzkY0, T kwkY0, T 0 :
Because of Part (ii), the estimate
kzkY0, T
ðkð1 , h~1 kZ0, T þ kð2 , h~2 ÞkZ0, T Þ
is obtained for z. If T 0 is chosen such that
C T 01=2 þ T 01=3 kzkY0, T ¼ 1=2
then it follows that
kwkY0, T 0
2Ckð1 , h~1 Þ ð2 , h~2 ÞkX0, T :
kwkY0, T
ðkð1 , h~1 ÞkZ0, T þ kð2 , h~2 ÞkZ0, T Þkð1 , h~1 Þ ð2 , h~2 ÞkX0, T :
ð4:6Þ
Since T only depends on kzkY0, T which in turn only depends on kð1 , h~1 ÞkZ0, T þ
kð2 , h~2 ÞkZ0, T , by a standard extension argument, one arrives at
0
Thus assumption (i) of Theorem 4.3 is satisfied. Estimate (4.1) is established for
0 < s < 3 by invoking Theorem 4.3.
Part (iv). We prove that (4.1) holds for 3 < s < 6. The same argument can be
invoked for s 6. For a smooth solution u of the IBVP (3.1), v ¼ ut solves
)
vt þ vx þ ðuvÞx þ vxxx ¼ 0, vðx, 0Þ ¼ ðxÞ
vð0, tÞ ¼ h01 ðtÞ,
vð1, tÞ ¼ h02 ðtÞ,
Applying Lemma 3.3 for any 0 < T 0
kvkYs3, T 0
vx ð1, tÞ ¼ h03 ðtÞ :
T gives the inequality
Ckð, h~ÞkZs, T þ C T 01=2 þ T 01=3 kvkYs3, T 0 kukYs3, T
for some constant C > 0 independent of T 0 and ð, h~Þ. Thus, if one chooses T 0
such that
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
1426
Bona, Sun, and Zhang
C T 01=2 þ T 01=3 kukYs3, T ¼ 1=2,
then
2Ckð, h~ÞkZs, T :
kvkYs3, T 0
Because T 0 only depends on kukYs3, T , which by the estimate (4.1) proved in Part (iii)
only depends on kð, h~ÞkZs3, T , one obtains
~
~
~
~
s3 ðkð, hÞkZs3, T Þkð, hÞkZs, T :
kvkYs3, T
Consequently,
kukYs, T
C
s3 ðkð, hÞkZs3, T Þkð, hÞkZs, T
œ
and the proof is complete.
5. ANALYTICITY
For given T > 0 and s 0, let X s, T be the collection of all s-compatible functions
ð, h~Þ 2 Xs, T . By the definition of s-compatibility, X s, T is a linear subspace of the
Banach space Xs, T if and only if 0 s 7=2. When 0 s 7=2, we consider X s, T as
a Banach space with its norm inherited from Xs, T . By the results established in Secs. 3
and 4, the IBVP (3.1) defines a nonlinear map KI from X s, T to the space Ys, T for any
s 0. From the proofs of the results given in Sec. 3, the map KI is known to be
locally Lipschitz continuous from DðKI Þ, the domain of KI , to Ys, T . In this section
it is shown that this nonlinear map KI is analytic. More precisely, when 0 s 7=2,
for any g 2 DðKI Þ, there exists an > 0 such that for any w 2 X s, T with kwkX s, T ,
we have g þ w 2 X s, T and KI ðg þ wÞ has the following Taylor series expansion:
KI ðg þ wÞ ¼ KI ðgÞ þ
1
X
K ðnÞ ðgÞ½wn
I
n¼1
n!
where KIðnÞ ðgÞ is the nth order Fréchet derivative of KI evaluated at g and the series
converges in the space YTs . In case s > 7=2, the Taylor series expansion does not hold
as just written since the space X s, T is no longer a linear vector space. In this case, we
consider the initial-boundary value problem for a general m-nonlinear system, which
includes the IBVP (3.1) as a special case, and show that the corresponding nonlinear
solution map KI is analytic in this context.
In pursuit of this program, we present a well-posedness result for the linearized
KdV-equation with variable coefficients, viz.
)
ut þ ux þ ðauÞx þ uxxx ¼ f ðx, tÞ,
uðx, 0Þ ¼ ðxÞ,
ð5:1Þ
uð0, tÞ ¼ h1 ðtÞ, uð1, tÞ ¼ h2 ðtÞ, ux ð1, tÞ ¼ h3 ðtÞ:
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
Korteweg–de Vries Equation on Finite Domain
1427
Proposition 5.1. Let 0 s 3 and T > 0 be given. Assume that a 2 Ys, T . Then for
any f 2 H s=3 ð ½0, T; H ð3sÞ=3 ð0, 1ÞÞ and ð, h~Þ 2 X s, T , Eq. (5.1) admits a unique
solution u 2 Ys, T satisfying
kukYs, T
Cðkð, h~ÞkXs, T þ k f kH s=3 ð ½0, T;H s=3 ð0, 1ÞÞ Þ
where C > 0 only depends on kakYs, T .
Proof. The proof is similar to that of Theorem 3.4, and so we only provide a sketch.
For given 0 < T and r > 0, let
S, r ¼ fw 2 Ys, : kwkYs,
rg:
For specified a 2 Ys, T , f 2 H s=3 ð ½0, T; H s=3 ð0, 1ÞÞ and ð, h~, f Þ 2 Xs, T H s=3 ð ½0, T;
H s=3 ð0, 1ÞÞ with ð, h~Þ 2 X s, T , consider a map : S, r ! Ys, defined by
u ¼ ðvÞ
where u is the unique solution of
ut þ ux þ uxxx ¼ f ðx, tÞ ðavÞx ,
uð0, tÞ ¼ h1 ðtÞ,
uð1, tÞ ¼ h2 ðtÞ,
)
uðx, 0Þ ¼ ðxÞ,
ux ð1, tÞ ¼ h3 ðtÞ,
for v 2 S, r . Applying Lemmas 3.3 and 3.1 yields
kðvÞÞkYs,
Choose 0 <
Ckð, h~, f ÞkXs, T H s=3 ð0, T;H s=3 ð0, 1ÞÞ þ Cð1=3 þ 1=2 ÞkakYs, T kvkYs, :
T and r such that
r ¼ 2Ckð, h~, f ÞkXs, T H s=3 ð0, T;H s=3 ð0, 1ÞÞ
ð5:2Þ
Cð1=3 þ 1=2 ÞkakYs, T r
ð5:3Þ
and
1=2:
It follows that
kðvÞkYs,
r
for any v 2 S, r and that for any v1 , v2 2 S, r ,
kððv1 Þ ðv2 ÞÞkYs,
1
kv v2 kYs, :
2 1
Thus is a contraction from S, r to S, r . Its unique fixed point is the desired solution
of (5.1) for 0 t . However, since is chosen according to (5.2) and (5.3) which
only depends on kakYs, T , this local argument can be iterated to extend the solution to
the entire temporal interval 0 t T. The proof is complete.
œ
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
1428
Bona, Sun, and Zhang
Formally, if KI is an analytic mapping from X s, T to Ys, T , then, for
n ¼ 0, 1, 2, . . . , its n-th order Fréchet derivative KIðnÞ ðgÞ at g 2 X s, T exists and is
the symmetric, n-linear map from X s, T to Ys, T given by
(
!)
n
X
@n
ðnÞ
KI ðgÞ½w1 , . . . :, wn ¼
k wk
K gþ
@1
@n I
k¼1
0,..., 0
for any w1 , w2 , . . . , wn 2 X s, T . The homogeneous polynomial KIðnÞ ðgÞ½wn of degree n
induced by KIðnÞ ðgÞ, where wn ¼ ðw, w, . . . , wÞ (n components), is
dn
K
ðg
þ
wÞ
KIðnÞ ðgÞ½wn ¼
dn I
¼0
for w ¼ ðw , wh Þ 2 X s, T . If we define yn by
yn ¼ KIðnÞ ðgÞ½wn ,
then it is formally ascertained that for 0 < t < T, ð y1 , y2 , . . . , yn Þ solves the system of
the equations
)
@t y1 þ @x y1 þ @x ðuy1 Þ þ @ 3x y1 ¼ 0,
y1 ðx, 0Þ ¼ w ðxÞ,
ð5:4Þ
y1 ð0, tÞ ¼ wh1 ðtÞ, y1 ð1, tÞ ¼ wh2 ðtÞ, @x y1 ð1, tÞ ¼ wh3 ðtÞ
and
@t yk þ @x yk þ @x ðuyk Þ þ
yk ðx, 0Þ ¼ 0,
@ 3x yk
yk ð0, tÞ ¼ 0,
!
9
k1 k
1X
>
@x ð yj ykj Þ, =
¼
2 j¼0 j
>
;
yk ð1, tÞ ¼ 0, @yk ð1, tÞ ¼ 0
ð5:5Þ
for 2 k n, where u ¼ KI ðgÞ and w ¼ ðw , wh1 , wh2 , wh3 Þ 2 X s, T .
On the other hand, for any g ¼ ð, h~Þ 2 DðKI Þ, let u ¼ KI ðgÞ and consider solving
the linear systems (5.4)–(5.5). It follows from Proposition 5.1 that (5.4)–(5.5) define a
homogeneous polynomial of degree n from X s, T to Ys, T as described by the following proposition.
Proposition 5.2. Let T > 0, 0 s 3, and g 2 X s, T be given and let u ¼ KI ðgÞ. Then
Eqs. (5.4)–(5.5) define a homogeneous polynomial KIðnÞ ðgÞ½wn of degree n from X s, T to
Ys, T . Moreover, there exists a constant c3 such that
kyn kYs, T
cn3 n!kwknX s, T
ð5:6Þ
for any n 2, where c3 ¼ c3 ðT, kukYs, T Þ, and it may be that c3 ! þ1 as T ! þ1 or
kukYs, T ! þ1, but in any case c3 ! 0 if T ! 0.
Proof. The proof is a straightforward consequence of the linear estimates in Sec. 2
and Proposition 5.1 (cf. Zhang, 1995b), Proposition 3.3 for a detailed argument in
related circumstances).
œ
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
Korteweg–de Vries Equation on Finite Domain
Define a Taylor polynomial Pn ðwÞ of degree n for w 2 X s, T by
n
n
X
X
KIðkÞ ðgÞ½wk
yk
¼ KI ðgÞ þ
Pn ðwÞ ¼
k!
k!
k¼1
k¼0
1429
ð5:7Þ
and a Taylor series by
PðwÞ ¼
1
X
K ðkÞ ðgÞ½wk
I
k¼0
k!
:
ð5:8Þ
Proposition 5.3. Let T > 0 and 0 s 3 be given. For any g ¼ ð, h~Þ 2 DðKI Þ, there
exists an > 0 depending only on kKI ðgÞkYs, T such that the formal Taylor series (5.8)
is uniformly convergent in the space Ys, T with respect to w 2 X s, T with kwkX s, T .
Moreover, if v ¼ PðwÞ, then v 2 Ys, T solves the problem
)
vt þ vx þ vvx þ vxxx ¼ 0,
vðx, 0Þ ¼ ðxÞ þ w ðxÞ
ð5:9Þ
vð0, tÞ ¼ h1 þ wh1 , vð1, tÞ ¼ h2 þ wh2 , vx ð1, tÞ ¼ h3 þ wh3
for 0
t
T.
Proof. It is readily seen that the sequence fPn ðwÞg1
n¼0 of Taylor polynomials is
Cauchy in Ys, T uniformly for w in the ball of radius in X s, T for suitable .
Indeed, because of Proposition 5.2, for m n 0,
m
m
m ky k
X
X
X
yk
k Ys, T
kPn ðwÞ Pm ðwÞkYs, T ¼
ck3 khkkX s, T :
k¼n k!
k!
k¼n
k¼n
Ys, T
If is chosen so that
1=ð2c3 Þ,
then for w 2 X s, T with kwkX s, T ,
m
X
1
kPn ðwÞ Pm ðwÞkYs, T
k
2
k¼n
ð5:10Þ
which goes to zero uniformly as n, m ! 1.
Since fPn ðwÞg1
n¼0 is a Cauchy sequence in the space Ys, T , it makes sense to define
v ¼ PðwÞ 2 Ys, T as its limit as n ! 1. It is then readily verified that v solves the
IBVP (3.1). The proof is complete.
œ
The following theorem is now adduced.
Theorem 5.4. (Analyticity) For any T > 0 and 0 s 3, the IBVP Eq. (3.1) establishes an analytic map KI from the space X s, T to the space Ys, T in the sense that for any
g 2 DðKI Þ there exists an > 0 such that for any w 2 X s, T with kwkX s, T , the
Taylor series expansion
1
X
KIðnÞ ðgÞ½wn
KI ðg þ wÞ ¼
n!
n¼0
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
1430
Bona, Sun, and Zhang
converges in the space Ys, T . Moreover, the convergence is uniform with regard to w in
the aforementioned ball in X s, T .
Remark 5.5. The above theorem holds also for 3 < s 7=2. Since the reasoning in
this case is similar to that put forward for the system discussed below, we include
analysis for this range of s in our next theorem.
Now, consider the case wherein s > 3. This situation is a little more involved
than the previous case because the compatibility conditions are no longer linear
restrictions. One could attempt to deal directly with the geometric situation implied
by the nonlinear compatibility conditions, but another approach presents itself
which is more transparent. That is to link the single equation faithfully with a
class of systems to be discussed presently.
As in the Sec. 4, for any s > 3, write s ¼ 3m þ s0 where m > 0 is an integer and
0 < s0 3. For T > 0, define the space Y sT to be
Y sT ¼ Y3, T Y3, T
Y3, T Ys0 , T
and the space X sT as
X sT ¼ X3, T X3, T
X3, T Xs0 , T :
Consider the system
u~ðx, 0Þ ¼ ~ðxÞ
u~t þ u~x þ ðFð~
uÞ~
uÞx þ u~xxx ¼ 0,
u~ð0, tÞ ¼ h~1 ,
u~ð1, tÞ ¼ h~2 ,
u~x ð1, tÞ ¼ h~3
)
ð5:11Þ
where
u~ ¼ ðu0 , u1 , . . . , um ÞT ,
~ ¼ ð0 , 1 , . . . , m ÞT ,
h~j ¼ ðhj, 0 , hj, 1 , . . . , hj, m ÞT
for j ¼ 1, 2, 3 and
Fð~
uÞ ¼ ð1=2Þ
u20 ,
m
X
m
u u
2u0 u1 , . . . ,
k k mk
k¼0
!T
:
By Theorem 4.1, for any s-compatible ð, h1 , h2 , h3 Þ 2 Xs, T , the IBVP (3.1) has a
unique solution u 2 Ys, T . If one defines 0 by 0 ¼ and let k be obtained from
k
by (3.4) with hj, k ¼ hðkÞ
j , uk ¼ @ t u for j ¼ 1, 2, 3 and k ¼ 0, 1, . . . , m, then
s
ð~, h~1 , h~2 , h~3 Þ 2 X T and u~ is a solution of (5.11). In this sense, the IBVP (3.1) is a
specialization of the system (5.11).
Theorem 5.6. Let T > 0 and s > 3 be given with s ¼ 3m þ s0 and 0 s0 < 3. Then for
any ð~, h~Þ 2 X sT , the system (5.11) admits a unique solution u~ 2 Y sT .
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
Korteweg–de Vries Equation on Finite Domain
1431
Proof. Observe that the nonlinear system (5.11) consists of initial-boundary-value
problems for m þ 1 scalar equations. Among them, the first one is the IBVP (3.1)
which only involves u0 . The second one involves only u0 and u1 . If u0 is known, then
the second IBVP is a linear problem. Similar remarks apply for the rest of the
equations. Thus we can solve the nonlinear system by solving for u0 from the first
equation, plugging u0 into the second equation and solving the corresponding linearized problem to obtain u1 , etc. Using Theorem 4.1 and Proposition 5.1, it is
deduced inductively that uk 2 Y3, T for k ¼ 0, 1, . . . , m 1. Now the equation related
to um has the form
)
um ðx, 0Þ ¼ m ðxÞ
@t um þ @x um þ @x ðaum Þ þ @ 3x um ¼ f ,
ð5:12Þ
um ð0, tÞ ¼ h1, m ðtÞ, um ð1, tÞ ¼ h2, m ðtÞ, @x um ð1, tÞ ¼ h3, m ðtÞ
0
where f 2 Cð ½0, T; H s ð0, 1ÞÞ and a 2 Y 3m
T are known. Using Lemmas 4.1–4.6, the
contraction principle and arguments similar to those appearing in the proof of
Theorem 4.1, it can be shown that for any ðm , h1, m , h2, m , h3, m Þ 2 X s0 , T , (5.12)
admits a unique solution um 2 Ys0 , T . The proof is complete.
œ
The last result implies the nonlinear system (5.11) defines a nonlinear map KI
from the space X sT to Y sT for given T > 0 and s ¼ 3m þ s0 with 0 s0 < 3. We claim
this map KI is analytic from X sT to Y sT . For the purpose of establishing this contention, consider the linearized system corresponding to the nonlinear system (5.11),
namely
)
@t w~ þ @x w~ þ @x ðJð~
aÞw~ Þ þ @ 3x w~ ¼ f~,
w~ ðx, 0Þ ¼ ~ðxÞ,
ð5:13Þ
w~ ð0, tÞ ¼ h~1 ðtÞ, w~ ð1, tÞ ¼ h~2 ðtÞ, @x w~ ð1, tÞ ¼ h~3 ðtÞ
where J is the Jacobian matrix of F at u~ ¼ a~, viz.
@Fð~
uÞ
Jð~
aÞ ¼
@~
u u~¼~a
!
k
X
k
¼
ði, jÞakj þ aj ði, k jÞ
j
j¼0
0 k, i m
and
ði, jÞ ¼
1
0
if i ¼ j,
if i ¼
6 j:
Proposition 5.7. Let T > 0 and s > 3 be given. Suppose a~ 2 Y sT and
f~ 2 FTs ¼ L1 ð0, T; H 3 ð0, 1ÞÞ
0
L1 ð0, T; H 3 ð0, 1ÞÞ L1 ð0, T; H s ð0, 1ÞÞ:
Then for any ð~, h~1 , h~2 , h~3 Þ 2 X sT , Eq. (5.13) admits a unique solution w~ 2 Y sT .
Moreover,
kw~ kY Ts
where
ðk~
akY sT Þðkð~, h~1 , h~2 , h~3 ÞkX sT þ k f~kFTs Þ
: Rþ ! Rþ is a continuous nondecreasing function.
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
1432
Bona, Sun, and Zhang
Proof. The proof is similar to that of Proposition 5.1 and therefore omitted.
For given u~ ¼ KI ðð~, h~ÞÞ with ð~, h~Þ 2 X sT , consider the linear systems
9
~ ~,=
y~1 ðx, 0Þ ¼ w
uÞ~
y1 Þ þ @ 3x y~1 ¼ 0,
@t y~1 þ @x y~1 þ @x ðJð~
y~1 ð0, tÞ ¼ w~ h~ ðtÞ,
y~1 ð1, tÞ ¼ w~ h~ ðtÞ,
1
2
and
@x y~1 ð1, tÞ ¼ w~ h~ ðtÞ;
for 2
n
ð5:14Þ
3
y1 , . . . , y~n1 Þ,
uÞ~
yn Þ þ @ 3x y~n ¼ Fn ð~
@t y~n þ @x y~n þ @x ðJð~
y~n ð0, tÞ ¼ 0,
œ
y~n ð1, tÞ ¼ 0,
@x y~n ð1, tÞ ¼ 0
n
y y
i i, j ni, kj
!
y~n ðx, 0Þ ¼ 0,
)
ð5:15Þ
N, where
Fn ¼ ð fn, 0 , fn, 1 , . . . , fn, m ÞT
with
fn, k
k X
n1
X
1
k
¼ @x
j
2
j¼0 i¼1
for k ¼ 0, 1, . . . , m.
Proposition 5.8. Given T > 0, s > 3, and g~ ¼ ð~, h~Þ 2 Y sT , let u~ ¼ KI ðð~, h~ÞÞ. Then the
systems (5.14)–(5.15) defines a homogeneous polynomial KðnÞ
gÞ½w~ n of degree n from
I ð~
s
s
X T to Y T . Moreover, there exists a constant C > 0 such that
Cn n!kw~ knX sT
k~
yn kY sT
for any n 2, where C ¼ CðT, k~
ukY sT Þ. Here C may go to þ1 when T ! 1 or
k~
ukY sT ! 1, but must go to 0 if T ! 0,
Proof. This follows from Proposition 5.4 by direct computation.
œ
~ Þ of degree n for w~ 2 X sT by
Define a Taylor polynomial Pn ðw
Pn ðw~ Þ ¼
N
X
KðkÞ ðgÞ½w~ k
I
k!
k¼0
¼ KI ð~
gÞ þ
N
X
y~k
k¼1
k!
,
ð5:16Þ
and a Taylor series by
Pðw~ Þ ¼
1
X
~k
KðkÞ ð~
gÞ½w
I
k¼0
k!
:
ð5:17Þ
A proof similar to that given for Proposition 5.3 yields the following
proposition.
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
Korteweg–de Vries Equation on Finite Domain
1433
Proposition 5.9. Let T > 0 and s > 0 be given. For any g~ ¼ ð~, h~1 , h~2 , h~3 ÞÞ 2 X sT , there
exists an > 0 depending only on kKI ð~
gÞkY sT such that the formal Taylor series (5.17) is
uniformly convergent in the space Y sT with respect to w~ 2 X sT with kw~ kX sT .
Moreover, if v~ ¼ Pðw~ Þ, then v~ 2 Y sT solves the problem
@t v~ þ @x v~ þ @x ðFð~vÞx v~Þ þ @ 3x v~ ¼ 0,
~ h~ ðtÞ,
v~ð0, tÞ ¼ h~1 ðtÞ þ w
1
v~ðx, 0Þ ¼ ~ðxÞ þ w~ ~ðxÞ
v~ð1, tÞ ¼ h~2 ðtÞ þ w~ h~ ðtÞ,
2
9
=
@x v~ð1, tÞ ¼ h~3 ðtÞ þ w~ h~ ðtÞ ;
3
ð5:18Þ
for 0
t
T.
Consequently, we have the following theorem.
Theorem 5.10. (Analyticity) For any T > 0 and s > 3, the nonlinear problem (5.11)
establishes a map KI from the space X sT to the space Y sT . The map KI is analytic from
X sT to Y sT in the sense that for any g~ 2 X sT , there exists an > 0 such that for any
w~ 2 X sT with kw~ kX sT , the Taylor series expansion
~Þ ¼
KI ð~
gþw
1
X
KðnÞ
gÞ½w~ n
I ð~
n!
n0
converges in the space Y sT . Moreover, the convergence is uniform with regard to h~ in the
aforementioned ball in X sT .
ACKNOWLEDGMENTS
JLB was partially supported by the National Science Foundation and by the
W. M. Keck Foundation. SMS was partially supported by the National Science
Foundation. BYZ was partially supported by the Charles H. Taft Memorial
Fund. The manuscript benefitted very considerably from detailed commentary by
an anonymous referee. The authors are grateful for this advice.
REFERENCES
Benjamin, T. B., Bona, J. L., Mahony, J. J. (1972). Model equations for long waves
in nonlinear dispersive systems. Philos. Trans. Royal Soc. London Ser. A
272:47–78.
Boczar-Karakiewicz, B., Bona, J. L., Pelchat, B. (1991). Interaction of internal waves
with the sea bed on continental shelves. Continental Shelf Res. 11:1181–1197.
Boczar-Karakiewicz, B., Bona, J. L., Romńczyk, W., Thornton, E. B. Seasonal and
interseasonal variability of sand bars at Duck, NC, USA. Observations and
model predictions, submitted.
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
1434
Bona, Sun, and Zhang
Bona, J. L., Bryant, P. J. (1973). A mathematical model for long waves generated by
wavemakers in nonlinear dispersive systems. Proc. Cambridge Philos. Soc.
73:391–405.
Bona, J. L., Dougalis, V. A. (1980). An initial- and boundary-value problem for a
model equation for propagation of long waves. J. Math. Anal. Appl. 75:503–522.
Bona, J. L., Luo, L. (1995). Initial-boundary value problems for model equations for
the propagation of long waves. In: Gerreyra, G., Goldstein, G., Neubrander, F.,
ed. Evolution equations. Lecture Notes in Pure and Appl. Math. New York:
Marcel Dekker, Inc., 168:65–94.
Bona, J. L., Pritchard, W. G., Scott, L. R., (1981). An evaluation of a model equation
for water waves. Philos. Trans. Royal Soc. London Ser. A 302:457–510.
Bona, J. L., Scott, L. R. (1976). Solutions of the Korteweg-de Vries equation in
fractional order Sobolev spaces. Duke Math. J. 43:87–99.
Bona, J. L., Smith, R. (1978). The initial-value problem for the Korteweg-de Vries
equation. Philos. Trans. Royal Soc. London Ser. A 278:555–601.
Bona, J. L., Sun, S. M., Zhang, B.-Y. (2001). The initial-boundary value problem for
the KdV-equation on a quarter plane. Trans. American Math. Soc. 354:427–490.
Bona, J. L., Sun, S. M., Zhang, B.-Y. (2003). Forced oscillations of a damped KdVequation in a quarter plane, to appear in Commun. Contemporary Math.
Bona, J. L., Winther, R. (1983). The Korteweg-de Vries equation, posed in a quarter
plane. SIAM J. Math. Anal. 14:1056–1106.
Bona, J. L., Winther, R. (1989). The Korteweg-de Vries equation in a quarter-plane,
continuous dependence results. Diff. and Integral Eq. 2:228–250.
Bourgain, J. (1993a). Fourier transform restriction phenomena for certain lattice
subsets and applications to nonlinear evolution equations, part I: Schrödinger
equations. Geometric & Funct. Anal. 3:107–156.
Bourgain, J. (1993b). Fourier transform restriction phenomena for certain lattice
subsets and applications to nonlinear evolution equations, part II: the KdVequation. Geometric & Funct. Anal. 3:209–262.
Bubnov, B. A. (1979). Generalized boundary value problems for the Korteweg-de
Vries equation in bounded domain. Diff. Eq. 15:17–21.
Bubnov, B. A. (1980). Solvability in the large of nonlinear boundary-value problems
for the Korteweg-de Vries equations. Diff. Eq. 16:24–30.
Colin, T., Ghidaglia, J.-M. (2001). An initial-boundary-value problem for the
Korteweg-de Vries equation posed on a finite interval. Advances Diff. Eq.
6:1463–1492.
Colliander, J. E., Kenig, C. E., (2002). The generalized KdV-equation on the
half-line. Commun. Partial Diff. Eq. 27:2187-2266.
Constantin, P., Saut, J.-C. (1988). Local smoothing properties of dispersive
equations. J. American Math. Soc. 1:413–446.
Craig, W., Kappeler, T., Strauss, W. A. (1992). Gain of regularity for equations of
the Korteweg-de Vries type. Ann. Inst. Henri Poincare´ 9:147–186.
Gardner, C. S., Greene, J. M., Kruskal, M. D., Miura, R. M. (1974). Korteweg-de
Vries equation and generalizations. VI. Methods for exact solution. Commun.
Pure Appl. Math. 27:97–133.
Hammack, J. L. (1973). A note on tsunamis: their generation and propagation in an
ocean of uniform depth. J. Fluid Mech. 60:769–799.
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
Korteweg–de Vries Equation on Finite Domain
1435
Hammack, J. L., Segur, H. (1974). The Korteweg-de Vries equation and water
waves, Part 2. Comparison with experiments. J. Fluid Mech. 65:237–246.
Kato, T. (1975). Quasilinear equations of evolution, with applications to partial
differential equations. Lecture Notes in Math. New York-Berlin-HeidelbergTokyo: Springer-Verlag 448:27–50.
Kato, T. (1979). On the Korteweg-de Vries equation. Manuscripta Math. 29:89–99.
Kato, T. (1983). On the Cauchy problem for the (generalized) Korteweg-de Vries
equations. Advances in Mathematics Supplementary Studies, Stud. Appl. Math.
8:93–128.
Kenig, C. E., Ponce, G., Vega, L. (1991a). Oscillatory integrals and regularity of
dispersive equations. Indiana Univ. Math. J. 40:33–69.
Kenig, C. E., Ponce, G., Vega, L. (1991b). Well-posedness of the initial value problem
for the Korteweg-de Vries equation. J. American Math. Soc. 4:323–347.
Kenig, C. E., Ponce, G.. Vega, L. (1993a). Well-posedness and scattering results for
the generalized Korteweg-de Vries equation via the contraction principle.
Commun. Pure Appl. Math. 46:527–620.
Kenig, C. E., Ponce, G., Vega, L. (1993b). The Cauchy problem for the Korteweg-de
Vries equation in Sobolev spaces of negative indices. Duke Math. J. 71:1–21.
Kenig, C. E., Ponce, G., Vega, L. (1996). A bilinear estimate with applications to the
KdV-equation. J. American Math. Soc. 9:573–603.
Korteweg, D. J., de Vries, G. (1895). On the change of form of long waves advancing
in a rectangular canal, and on a new type of long stationary waves. Philos. Mag.
39:422–443.
Lax, P. D. (1968). Integrals of nonlinear equations of evolution and solitary waves.
Commun. Pure Appl. Math. 21:467–490.
Miura, R. M. (1976). The Korteweg-de Vries equation: a survey of results. SIAM
Review 18:412–459.
Pazy, A. (1983). Semigroups of Linear Operators and Applications to Partial
Differential Equations, Applied Mathematical Sciences 44. New York-BerlinHeidelberg-Tokyo: Springer-Verlag.
Rosier, L. (1997). Exact boundary controllability for the Korteweg-de Vries
equation on a bounded domain. ESIAM: Control, Optim. & Calc. Variations
2:33–55.
Rudin, W. (1966). Real and Complex Analysis. New York: McGraw-Hill.
Russell, D. L. (1978). Controllability and stabilizability theory for linear partial
differential equations: recent progress and open questions. SIAM Review
20:639–739.
Russell, D. L., Zhang, B.-Y. (1993). Controllability and stabilizability of the
third-order linear dispersion equation on a periodic domain. SIAM J.
Control & Optim. 31:659–676.
Russell, D. L., Zhang, B.-Y. (1995). Smoothing properties of solutions of
the Korteweg-de Vries equation on a periodic domain with point dissipation.
J. Math. Anal. Appl. 190:449–488.
Russell, D. L., Zhang, B.-Y. (1996). Exact controllability and stabilizibility of the
Korteweg-de Vries equation. Trans. American Math. Soc. 348:3643–3672.
Saut, J.-C., Temam, R. (1976). Remarks on the Korteweg-de Vries equation. Israel
J. Math. 24:78–87.
MARCEL DEKKER, INC. • 270 MADISON AVENUE • NEW YORK, NY 10016
©2003 Marcel Dekker, Inc. All rights reserved. This material may not be used or reproduced in any form without the express written permission of Marcel Dekker, Inc.
1436
Bona, Sun, and Zhang
Sjöberg, A. (1970). On the Korteweg-de Vries equation: existence and uniqueness.
J. Math. Anal. Appl. 29:569-579.
Sun, S. M., (1996). The Korteweg-de Vries equation on a periodic domain with
singular-point dissipation. SIAM J. Control & Optim. 34:892–912.
Tartar, L. (1972). Interpolation non linéaire et régularité. J. Funct. Anal. 9:469–489.
Temam, R. (1969). Sur un probléme non linéaire. J. Math. Pures Appl. 48:159–172.
Vega, L. (1988). Schrödinger equations: pointwise convergence to the initial data.
Proc. American Math. Soc. 102:874–878.
Wayne, C. E. (1990). Periodic and quasi-periodic solutions of nonlinear wave
equations via KAM theory. Commun. Math. Phys. 127:479–528.
Wayne, C. E. (1997). Periodic solutions of nonlinear partial differential equations.
Notices American Math. Soc. 44:895–902.
Zabusky, N. J., Galvin, C. J. (1971). Shallow-water waves, the Korteweg-de Vries
equation and solitons. J. Fluid Mech. 47:811–824.
Zhang. B.-Y. (1994). Boundary stabilization of the Korteweg-de Vries equations.
Proc. of International Conference on Control and Estimation of Distributed
Parameter Systems: Nonlinear Phenomena. Desch, W., Kappel, F., Kunisch, K.,
eds. International Series of Numerical Mathematics 118, Birkhauser, Basel,
371–389.
Zhang. B.-Y. (1995a). A remark on the Cauchy problem for the Korteweg-de Vries
equation on a periodic domain. Diff. & Integral Eq. 8:1191–1204.
Zhang. B.-Y. (1995b). Analyticity of solutions for the generalized Korteweg-de
Vries equation with respect to their initial datum. SIAM J. Math. Anal.
26:1488–1513.
Zhang. B.-Y. (1995c). Taylor series expansion for solutions of the KdV-equation
with respect to their initial values. J. Funct. Anal. 129:293–324.
Zhang. B.-Y. (1999). Exact boundary controllability of the Korteweg-de Vries
equation on a bounded domain. SIAM J. Control & Optim. 37:543 –565.
Received May 2002
Revised March 2003