CAMASSAHOLM EQUATION
HELGE HOLDEN AND XAVIER RAYNAUD
Abstract. We show that the periodic CamassaHolm equation ut uxxt +
3uux 2uxuxx uuxxx = 0 possesses a global continuous semigroup of weak
conservative solutions for initial data u
t=0
in H
1
per
. The result is obtained
by introducing a coordinate transformation into Lagrangian coordinates. To
characterize conservative solutions it is necessary to include the energy density
given by the positive Radon measure with ac = (u
2
+ u
2
x
) dx. The total
energy is preserved by the solution.
1. Introduction
The CamassaHolm equation
u
t
u
xxt
+ 3uu
x
2u
x
u
xx
uu
xxx
= 0 (1.1)
was rst studied extensively in 1993 [6, 7]. It can be derived as a model for shallow
water waves. Furthermore, the equation can be derived in the context of geodesic
ows of a certain invariant metric on the BottVirasoro group [25, 3].
The equation possesses many fascinating properties that has made it a popular
equation. In particular, it is biHamiltonian, completely integrable, has innitely
many conserved quantities, and has solitary waves, called (multi)peakons, that in
teract like KdVsolitons. Another interesting aspect is that it enjoys wavebreaking
in nite time in the sense that the spatial derivative u
x
of the solution blows up
while the solution u itself as well as its energy, the H
1
norm, both remain nite.
Continuation of the solution beyond wave breaking has been a challenge. Several
entropy conditions that single out the proper continuation have been analyzed.
The Cauchy problem for (1.1) has been studied in two dierent settings; on the
full line R and the periodic case on [0, 1]. We here address the latter case, and
for reasons of space we restrict the general references mainly to the periodic case.
Constantin and Moulinet have proved [14, p. 60] that for initial data u[
t=0
= u
H
1
([0, 1]) such that m = u u
xx
is a nonnegative Radonmeasure, the equation
(1.1) possesses a unique solution u C
1
((0, ), L
2
([0, 1])) C((0, ), H
1
([0, 1])).
Furthermore, the quantities
_
[0,1]
u dx,
_
[0,1]
(u
2
+ u
2
x
) dx, and
_
[0,1]
(u
3
+ uu
2
x
) dx
are all conserved quantities. Regarding blowup, Constantin and Escher [13] have
derived the following result. Let u
0
H
3
([0, 1]). Then there exists a maximal T > 0
such that (1.1) has a unique solution u C([0, T), H
3
([0, 1]))C
1
([0, T), H
2
([0, 1])).
If u is nonzero and
_
[0,1]
( u
3
+ u u
2
x
) dx = 0, then T is nite. See also [10, 26].
The question about how to continue the solution beyond wavebreaking can be
nicely studied in the case of multipeakons (we here give the description on the full
Date: October 12, 2006.
2000 Mathematics Subject Classication. Primary: 65M06, 65M12; Secondary: 35B10, 35Q53.
Key words and phrases. CamassaHolm equation, periodic solution.
This research was supported in part by the Research Council of Norway.
1
2 HOLDEN AND RAYNAUD
line). Multipeakons are given by (see, e.g., [21] and references therein)
u(t, x) =
n
i=1
p
i
(t)e
xqi(t)
, (1.2)
where the (p
i
(t), q
i
(t)) satisfy the explicit system of ordinary dierential equations
q
i
=
n
j=1
p
j
e
qiqj
, p
i
=
n
j=1
p
i
p
j
sgn(q
i
q
j
)e
qiqj 
.
Observe that the solution (1.2) is not smooth even with continuous functions
(p
i
(t), q
i
(t)); one possible way to interpret (1.2) as a weak solution of (1.1) is to
rewrite the equation (1.1) as
u
t
+
_
1
2
u
2
+ (1
2
x
)
1
(u
2
+
1
2
u
2
x
)
_
x
= 0.
Wave breaking may appear when at least two of the q
i
s coincide. If all the p
i
(0)
have the same sign, the peakons move in the same direction, the solution experiences
no wave breaking, and one has a global solution. Higher peakons move faster
than the smaller ones, and when a higher peakon overtakes a smaller, there is an
exchange of mass, but no wave breaking takes place. Furthermore, the q
i
(t) remain
distinct. However, if some of p
i
(0) have opposite sign, wave breaking may incur.
For simplicity, consider the case with n = 2 and one peakon p
1
(0) > 0 (moving to
the right) and one antipeakon p
2
(0) < 0 (moving to the left). In the symmetric case
(p
1
(0) = p
2
(0) and q
1
(0) = q
2
(0) < 0) the solution will vanish pointwise at the
collision time t
when q
1
(t
) = q
2
(t
,
and the other possibility is to let the peakon and antipeakon pass through each
other in a way that is consistent with the CamassaHolm equation. In the rst
case the energy
_
(u
2
+u
2
x
) dx decreases to zero at t
(u
2
+u
2
x
) dx (1.4)
where U corresponds to the Lagrangian velocity while H could be interpreted as
the Lagrangian cumulative energy distribution. In the periodic case one denes
Q =
1
2(e 1)
_
1
0
sinh(y() y())(U
2
y
+H
)() d (1.5)
1
4
_
1
0
sgn( ) exp
_
sgn( )(y() y())
_
(U
2
y
+H
) d,
P =
1
2(e 1)
_
1
0
cosh(y() y())(U
2
y
+H
)() d (1.6)
+
1
4
_
1
0
exp
_
sgn( )(y() y())
_
(U
2
y
+H
) d.
Then one can show that
_
_
y
t
= U,
U
t
= Q,
H
t
= U
3
2PU,
(1.7)
is equivalent to the CamassaHolm equation. Global existence of solutions of (1.7)
is obtained starting from a contraction argument, see Theorem 2.7. The uniqueness
issue is resolved by considering the set T (see Denition 3.1) which consists of pairs
(u, ) such that (u, ) T if u H
1
per
and is a positive Radon measure with period
one, and whose absolutely continuous part satises
ac
= (u
2
+u
2
x
) dx. With three
Lagrangian variables (y, U, H) versus two Eulerian variables (u, ), it is clear that
there can be no bijection between the two coordinates systems. However, we dene
a group of transformations which acts on the Lagrangian variables and lets the
system of equations (1.7) invariant. We are able to establish a bijection between the
space of Eulerian variables and the space of Lagrangian variables when we identify
variables that are invariant under the action of the group. This bijection allows us to
transform the results obtained in the Lagrangian framework (in which the equation
is wellposed) into the Eulerian framework (in which the situation is much more
subtle). In particular, and this constitutes the main result of this paper, we obtain a
metric d
D
on T and a continuous semigroup of solutions on (T, d
D
). The distance
d
D
gives T the structure of a complete metric space. This metric is compared with
some more standard topologies, and we obtain that convergence in H
1
per
implies
convergence in (T, d
D
) which itself implies convergence in L
, see Propositions
5.1 and 5.2. The properties of the spaces as well as the various mappings between
them are described in great detail, see Section 3. Our main result, Theorem 4.2,
states that there exists a continuous semigroup T : T R T such that, for any
( u, ) T, if we denote (u(t), (t)) = T
t
( u, ), then u(t) is a weak solution of the
CamassaHolm equation. The topology on T is of course given by the metric d
D
4 HOLDEN AND RAYNAUD
and, by continuity of the semigroup, we mean that if ( u
n
,
n
) ( u, ) in T, then
(u
n
(t),
n
(t) (u(t), (t)) in T, i.e., we use the same topology on the set of initial
data as on the set of solutions, which shows that the complete metric d
D
is the
appropriate metric for conservative solutions of the CamassaHolm equation. The
associated measure (t) has constant total mass, i.e., (t)([0, 1)) = (0)([0, 1)) for
all t, which corresponds to the total energy of the system. This is the reason why
our solutions are called conservative.
Many of the ideas used in this article originate from [21] where the case of the
full line is treated and some of the proofs are indeed adaptations to the periodic
case. There is, however, one signicant and important dierence. It concerns the
group of transformations here denoted
G acting on the Lagrangian variables. By
introducing Lagrangian variables, one introduces a degree of arbitrariness which is
captured by the group of transformation acting on the new variables and which
is removed when one takes the quotient space. The determination of the correct
group is crucial as it enables us to return to the Eulerian coordinates via the
quotient space and to construct the continuous semigroup of solutions in Eulerian
coordinates. On the full line, this group consists of the group of dieomorphism
with some regularity condition denoted G, which is also a natural choice taking into
account the geometric interpretation of the equation, see [3, 15]. In the periodic
case, the same group G is needed but we also have to take into account that we
introduce an additional degree of freedom by considering the cumulative energy.
Indeed, the energy is like a potential and dened up to a constant. In the case of
the full line, we normalize this constant to zero at . We cannot do that in the
periodic case and instead we expand the group G to
G = GR. The action of the
group (R, +) corresponds to the degree of freedom resulting from the fact that the
energy is dened up to a constant.
The method described here can be studied in detail for multipeakons, see [21]
for details on the full line. The results can be extended, as in [24], to show global
existence of conservative periodic solutions for the generalized hyperelasticrod wave
equation
_
_
_
u
t
+f(u)
x
+P
x
= 0,
P P
xx
= g(u) +
1
2
f
(u)u
2
x
,
(1.8)
where f, g C
0
. (2.4)
From (2.3), the periodicity of u and the fact that y V
1
, we can check that, for all
R,
H(t, + 1) H(t, ) = H(t, 1) H(t, 0).
Moreover, from (2.4), we can verify that H(t, 1) H(t, 0) is constant in time so
that H(t, 1) H(t, 0) = H(0, 1) H(0, 0). For all t, H belongs to the vector space
V dened as follows
V = f H
1
loc
(R) [ there exists R
such that f( + 1) = f() +, for all R.
We equip V with the norm f
V
= f
H
1
([0,1])
. Later we will prove that V is a
Banach space. To simplify the notation, we will denote H
1
([0, 1]) by H
1
and follow
the same convention for the other norms we will consider.
We now derive formally a system equivalent to (1.3). From the denition of the
characteristics, it follows that
U
t
(t, ) = u
t
(t, y) +y
t
(t, )u
x
(t, y) = P
x
y (t, ). (2.5)
This last term can be expressed uniquely in term of U, y, and H. From (1.3b), we
obtain the following explicit expression for P,
P(t, x) =
1
2
_
R
e
xz
(u
2
(t, z) +
1
2
u
2
x
(t, z)) dz. (2.6)
Thus we have
P
x
y (t, ) =
1
2
_
R
sgn(y(t, ) z)e
y(t,)z
(u
2
(t, z) +
1
2
u
2
x
(t, z)) dz
and, after the change of variables z = y(t, ),
P
x
y(t, ) =
1
2
_
R
_
sgn(y(t, ) y(t, ))e
y(t,)y(t,)
_
u
2
(t, y(t, )) +
1
2
u
2
x
(t, y(t, ))
_
y
(t, )
_
d. (2.7)
We have
H
= (u
2
+u
2
x
)y y
. (2.8)
Therefore, (2.7) can be rewritten as
P
x
y () =
1
4
_
R
sgn(y() y()) exp([y() y()[)
_
U
2
y
+H
_
() d (2.9)
where the t variable has been dropped to simplify the notation. Later we will prove
that y is an increasing function for any xed time t. If, for the moment, we take
this for granted, then P
x
y is equivalent to Q where
Q(t, ) =
1
4
_
R
sgn() exp
_
sgn()(y()y())
__
U
2
y
+H
_
() d, (2.10)
6 HOLDEN AND RAYNAUD
and, slightly abusing the notation, we write
P(t, ) =
1
4
_
R
exp
_
sgn( )(y() y())
__
U
2
y
+H
_
() d. (2.11)
The derivatives of Q and P are given by
Q
=
1
2
H
_
1
2
U
2
P
_
y
and P
= Qy
. (2.12)
For [0, 1], using the fact that y( +1) = y() +1 and the periodicity of H
and
U, these expressions can be rewritten as
Q =
1
2(e 1)
_
1
0
sinh(y() y())(U
2
y
+H
)() d
1
4
_
1
0
sgn( ) exp
_
sgn( )(y() y())
_
(U
2
y
+H
) d (2.13)
and
P =
1
2(e 1)
_
1
0
cosh(y() y())(U
2
y
+H
)() d
+
1
4
_
1
0
exp
_
sgn( )(y() y())
_
(U
2
y
+H
) d. (2.14)
Thus P
x
y and P y can be replaced by equivalent expressions given by (2.10)
and (2.11) which only depend on our new variables U, H, and y. We now derive
a new system of equations, formally equivalent to the CamassaHolm equation.
Equations (2.5), (2.4) and (2.1) give us
_
_
y
t
= U,
U
t
= Q,
H
t
=
_
U
3
2PU
0
.
(2.15)
Dierentiating (2.15) yields
_
_
y
t
= U
,
U
t
=
1
2
H
+
_
1
2
U
2
P
_
y
,
H
t
= 2QUy
+
_
3U
2
2P
_
U
.
(2.16)
The system (2.16) is semilinear with respect to the variables y
, U
, and H
.
2.2. Existence and uniqueness of solutions of the equivalent system. In
this section, we focus our attention on the system of equations (2.15) and prove,
by a contraction argument, that it admits a unique solution. Let Id denote the
identity, i.e., Id() = . We claim that the linear map : (, h) f = + hId is
an homeomorphism from H
1
per
R to V where H
1
per
denotes the Banach space
H
1
per
= H
1
loc
(R) [ ( + 1) = () for all R
with the norm 
H
1
per
= 
H
1
. It is clear that is invertible and, for any f V ,
its inverse (, h) =
1
f is given by h = f(1) f(0) and = f hId. Let
f = (, h), we have
f
H
1 
H
1 +[h[ Id
H
1 = 
H
1 +
_
2
3
[h[
and therefore is continuous. Conversely,
[h[ = [f(1) f(0)[ 2 f
L
2C f
H
1 ,
PERIODIC CONSERVATIVE SOLUTIONS OF THE CAMASSAHOLM EQUATION 7
and

H
1 f
H
1 + 2 f
L
Id
H
1 (1 + 2
_
2
3
C) f
H
1
where the constant C denotes the constant of the Sobolev embedding H
1
L
.
Hence,
1
is continuous. Since H
1
per
R is a Banach space, V is also a Banach
space. We introduce = y Id and (, h) =
1
(H), i.e., h = H(t, 1) H(t, 0)
and = H hId. The system (2.15) is then equivalent to
_
t
= U,
U
t
= Q,
t
=
_
U
3
2PU
0
,
h
t
= 0.
(2.17)
We will prove that the system (2.17) is a wellposed system of ordinary dierential
equations in the Banach space E where
E = H
1
per
H
1
per
H
1
per
R.
There is a bijection (, U, , h) (y, U, H) between E and V
1
H
1
per
V given by
y = + Id, H = + hId and U is unchanged, so that in the remaining we will
use both set of variables. However, for the contraction argument it is important to
have a Banach space and we use E and the variables (, U, , h) (note that V
1
and
a fortiori V
1
H
1
per
V are not Banach spaces). The following lemma gives the
Lipschitz bounds we need on Q and P.
Lemma 2.1. For any X = (, U, , h) in E, we dene the maps Q and T as
Q(X) = Q and T(X) = P where Q and P are given by (2.10) and (2.11), re
spectively. Then, T and Q are Lipschitz maps on bounded sets from E to H
1
per
.
Moreover, we have
Q
=
1
2
(
+h)
_
1
2
U
2
P
_
(1 +
), (2.18)
P
= Q(1 +
). (2.19)
Proof. Let B
M
= X = (, U, , h) E [ X
E
M. Let us rst prove that
T and Q are Lipschitz maps from B
M
to L
per
. Let X = (, U, , h) and
X =
(
,
U, ,
h) be two elements of B
M
. We have
y
L
= Id +
L
1 +C 
H
1 (1 +CM).
and  y
L
(1 + CM). Since the map x coshx is locally Lipschitz, it is
Lipschitz on x R [ [x[ 2(1 + CM). Hence, if we denote by C a generic
constant that only depends on M, we have
[cosh(y() y()) cosh( y() y())[ C [y() y() y() + y()[
C
_
_
_
_
_
_
L
for all , in [0, 1]. It follows that, for all [0, 1],
_
_
_cosh(y() y( ))U
2
y
( ) cosh( y() y( ))
U
2
y
( )
_
_
_
L
2
C
_
_
_
_
_
_
_
L
+
_
_
_U
U
_
_
_
L
+
_
_
_
_
_
_
L
2
_
and the map X = (, U, , h)
1
2(e1)
_
1
0
cosh(y() y())(U
2
y
)() d which
corresponds to the rst term in (2.14) is Lipschitz from B
M
to L
per
. We handle the
other terms in (2.14) in the same way and we prove that T is Lipschitz from B
M
to
L
per
. Similarly, one proves that Q: B
M
L
per
is Lipschitz. Direct dierentiation
8 HOLDEN AND RAYNAUD
gives us the expressions (2.12) for the derivatives P
and Q
of P and Q. Since, as
we have just proved, T and Q are Lipschitz from B
M
to L
per
, we have
_
_
Q(X)
Q(
X)
_
_
L
2
=
_
_
_
_
y
T(X) y
T(
X)
1
2
(U
2
y
U
2
y
+h
h)
_
_
_
_
L
2
C
_
_
_
_T(X) T(
X)
_
_
_
L
+
_
_
_U
U
_
_
_
L
+
_
_
_
_
_
_
L
2
+

L
2
+
_
C
_
_
_X
X
_
_
_
E
where we have used the fact that U and
U are bounded in B
M
so that
_
_
_U
2
U
2
_
_
_
L
C
_
_
_U
U
_
_
_
L
,
U, , h) in E, there exists a time T depending only
on
_
_
X
_
_
E
such that the system (2.15) admits a unique solution in C
1
([0, T], E) with
initial data
X.
Proof. Solutions of (2.15) can be rewritten as
X(t) =
X +
_
t
0
F(X()) d (2.20)
where F : E E is given by F(X) = (U, Q(X), [U
3
2T(X)U]
0
, 0) where X =
(, U, , h). The integrals are dened as Riemann integrals of continuous functions
on the Banach space E. To prove that X [U
3
2T(U)U]
0
is Lipschitz from
bounded set of E to H
1
per
, we proceed as in the proof of Lemma 2.1. Hence, F
is Lipschitz on bounded set, and the theorem follows from the standard theory of
ordinary dierential equations, see, for example, [1].
We now turn to the proof of existence of global solutions of (2.15). We are
interested in a particular class of initial data that we are going to make precise later,
see Denition 2.5. In particular, we will only consider initial data that belong to
_
W
1,
per
3
R where W
1,
per
= f W
1,
loc
(R) [ f( +1) = f() for all R, which
is a Banach space for the norm f
W
1,
per
= f
W
1,
. Of course,
_
W
1,
per
3
R is
a subset of E. Given
X = (
,
U, ,
h) [W
1,
]
3
R, we consider the shorttime
solution X = (, U, , h) C([0, T], E) of (2.15) given by Theorem 2.2. Using the
fact that Q and T are locally Lipschitz (Lemma 2.1) and, since X C([0, T], E),
we can prove that P and Q belongs to C([0, T], H
1
per
). We now consider U, P, and
Q as given functions in C([0, T], H
1
per
). Then, for any xed R, we can solve the
system of ordinary dierential equations in R
3
given by
_
_
d
dt
(t, ) = (t, ),
d
dt
(t, ) =
1
2
((t, ) +
h) +
_
(
1
2
U
2
P)(t, )
_
(1 +(t, )),
d
dt
(t, ) = [2(QU)(t, )] (1 +(t, )) +
_
(3U
2
2P)(t, )
(t, ),
(2.21)
and which is obtained by substituting
, U
and
()
_
_
U
_
_
L
, [
()[ 

L
()
_
_
_
_
L
.
Since we assumed
X [W
1,
]
3
R, we have that / has full measure, that is,
meas(/
c
) = 0. For / we dene ((0, ), (0, ), (0, )) = (
(),
U
(),
()).
However, for /
c
we take ((0, ), (0, ), (0, )) = (0, 0, 0).
Lemma 2.3. Given initial condition
X = (
,
U, ,
h) [W
1,
]
3
R, we consider
the solution X = (, U, , h) C
1
([0, T], E) of (2.21) given by Theorem 2.2. Then,
X C
1
([0, T], [W
1,
]
3
R). The functions (t, ), (t, ) and (t, ) which are
obtained by solving (2.21) for any xed given with the initial condition specied
above, coincide for almost every and for all time t with
, U
and
, respectively,
that is, for all t [0, T], we have
((t, ), (t, ), (t, )) = (
(t, ), U
(t, ),
(t, )) (2.22)
for almost every R.
Thus, this lemma allows us to pick up a special representative for (
, U
)
given by (, , ), which is dened for all R and which, for any given , satises
the ordinary dierential equation (2.21) in R
3
. In the remaining we will of course
identify the two and set (
, U
per
whose norm is naturally given by f
B
per
= sup
[0,1]
[f()[.
Obviously, the periodic continuous functions belong to B
per
. We dene (, , ) as
the solution of (2.21) in
_
B
per
3
with initial data as given above. Thus, strictly
speaking, this is a dierent denition than the one given in the lemma but we will
see that they are in fact equivalent. We note that the system (2.21) is ane (it con
sists of a sum of a linear transformation and a constant) and therefore it is not hard
to prove, by using a contraction argument in
_
B
per
3
, the shorttime existence of
solutions. Moreover, due to the ane structure, a direct application of Gronwalls
lemma shows that the solution exists on [0, T], the interval on which (, U, , h) is
dened. For any given , the map f f() from B
per
to R is linear and continuous
(the space B
per
was precisely introduced in order to make this map continuous).
Hence, after applying this map to each term in (2.21) written in integral form and
using Proposition 2.4, we recover the original denition of , and as solutions,
for any given R, of the system (2.21) of ordinary dierential equations in R
3
.
The derivation map
d
d
is continuous from H
1
per
into L
2
per
. We can apply it to each
term in (2.15) written in integral from and, by Proposition 2.4, this map commutes
with the integral. We end up with, after using (2.18) and (2.19),
_
(t) =
+
_
t
0
U
() d,
U
(t) =
U
+
_
t
0
_
1
2
(
+
h) + (
1
2
U
2
P)(1 +
)
_
() d,
(t) =
+
_
t
0
_
2QU(1 +
) + (3U
2
2P)U
_
() d.
(2.23)
The map from B
per
to L
2
per
is also continuous, we can apply it to (2.21) writ
ten in integral form, and again use Proposition 2.4. Then, we subtract each
10 HOLDEN AND RAYNAUD
equation in (2.23) from the corresponding one in (2.21), take the norm and add
them. After introducing Z(t) = (t, )
(t, )
L
2
+ (t, ) U
(t, )
L
2
+
(t, )
(t, )
L
2
, we end up with the following equation
Z(t) Z(0) +C
_
t
0
Z() d
where C is a constant which, again, only depends on the C([0, T], H
1
)norms, of
U, P, and Q. By assumption on the initial conditions, we have Z(0) = 0 because
(0) =
, (0) =
U
, (0) =
= u
2
+ u
2
x
and
H
hId,
h) belongs
to ( where ( is dened as follows.
Denition 2.5. The set ( is composed of all (, U, , h) E such that
(, U, , h)
_
W
1,
3
R, (2.25a)
y
0, H
0, y
+H
= y
2
U
2
+U
2
0, H is an increasing function
and h = H(1) H(0) 0. Note that if all functions are smooth and y
> 0, we
have u
x
y =
U
,
U, ,
,
U, ,
h). Then,
(i) X(t) belongs to ( for all t [0, T],
(ii) for almost every t [0, T], y
(),
(), and
U
() all
are smaller than
_
_
X
_
_
[W
1,
]
3
R
and where the inequalities in (2.25b) and (2.25c)
are satised for y
,
U
and
H
,
U
) equal to zero on /
c
. Thus, as allowed by Lemma 2.3, we choose
PERIODIC CONSERVATIVE SOLUTIONS OF THE CAMASSAHOLM EQUATION 11
a special representative for ((t, ), U(t, ), (t, )) whose derivative satises (2.16)
as an ordinary dierential equation, for every R. The proof of this lemma is
almost the same as in [22]. We repeat it here for completeness.
Proof. (i) We already proved in Lemma 2.3 that the space [W
1,
]
3
R is preserved
and X(t) satises (2.25a) for all t [0, T]. Let us prove that (2.25c) and the
inequalities in (2.25b) hold for any / and therefore almost everywhere. We
consider a xed in / and drop it in the notation when there is no ambiguity.
From (2.16), we have, on the one hand,
(y
)
t
= y
t
H
+H
t
y
= U
+ (3U
2
U
2y
QU 2PU
)y
,
and, on the other hand,
(y
2
U
2
+U
2
)
t
= 2y
t
y
U
2
+ 2y
2
U
t
U + 2U
t
U
= 3U
U
2
y
2PU
+H
2y
2
QU.
Thus, (y
y
2
U
2
U
2
)
t
= 0, and since y
(0) = (y
2
U
2
+ U
2
)(0), we have
y
(t) = (y
2
U
2
+ U
2
given by
t
= supt [0, T] [ y
(t
) 0 for all t
[0, t].
Here we recall that we consider a xed / and dropped it in the notation.
Assume that t
< T. Since y
(t
) = 0. (2.26)
Hence, from (2.25c) that we just proved, U
(t
) = 0 and, by (2.16),
y
t
(t
) = U
(t
) = 0. (2.27)
From (2.16), since y
(t
) = U
(t
) = 0, we get
y
tt
(t
) = U
t
(t
) =
1
2
H
(t
). (2.28)
If H
(t
) = 0, then (y
, U
, H
)(t
, U
, H
)(t) = 0 for all t [0, T]. This contradicts the fact that y
(0) and H
(0)
cannot vanish at the same time ( y
+
H
(t
) < 0, then
y
tt
(t
.
Hence, H
(t
(t
) = y
t
(t
) = 0 and y
tt
(t
) > 0,
there exists a neighborhood of t
(t) 0 for all t [0, T]. This follows from (2.25c) when y
(t) = 0, then U
(t
+H
)(t) = 0. Since y
and H
(t) = H
(0) = U
(0) = H
+H
)(0) > 0.
(ii) We dene the set
^ = (t, ) [0, T] R [ y
(t, ) = 0.
12 HOLDEN AND RAYNAUD
Fubinis theorem gives us
meas(^) =
_
R
meas(^
) d =
_
[0,T]
meas(^
t
) dt (2.29)
where ^
and ^
t
are the section and tsection of ^, respectively, that is,
^
= t [0, T] [ y
(t, ) = 0 and ^
t
= R [ y
(t, ) = 0.
Let us prove that, for all /, meas(^
dened
as
^
n
= t [0, T] [ y
(t, ) = 0 and y
(t
[t 1/n, t + 1/n] t,
then
^
=
_
nN
^
n
. (2.30)
Indeed, for all t ^
, we have y
(t, ) = 0, y
t
(t, ) = 0 from (2.25c) and (2.16)
and y
tt
(t, ) =
1
2
H
and H
cannot vanish at
the same time for /). This implies that, on a small punctured neighborhood
of t, y
) = 0.
It follows from (2.29) and since meas(/
c
) = 0 that
meas(^
t
) = 0 for almost every t [0, T]. (2.31)
We denote by / the set of times such that meas(^
t
) > 0, i.e.,
/ = t R
+
[ meas(^
t
) > 0. (2.32)
By (2.31), meas(/) = 0. For all t /
c
, y
,
U, ,
2h and sup
t[0,T)
(t, )
L
(R)
PERIODIC CONSERVATIVE SOLUTIONS OF THE CAMASSAHOLM EQUATION 13
is bounded by 2
() d +
_
1
0
e
y()y()
H
() d.
Hence, [Q[ C(H(1)H(0)) = Ch = C
(R)
is nite. Similarly, one prove that sup
t[0,T)
P(t, )
L
(R)
< . Since U
t
= Q, it
follows that sup
t[0,T)
U(t, )
L
(R)
< and, since
t
= U, sup
t[0,T)
(t, )
L
(R)
is also nite. We have proved that
C
1
= sup
t[0,T)
U(t, )
L
+P(t, )
L
+Q(t, )
L
(t, )
L
2
+ U
(t, )
L
2
+ H
(t, )
L
2
. Using the semi
linearity of (2.16), we obtain
Z(t) Z(0) +C
_
t
0
Z() d
where C is a constant depending only on C
1
. It follows from Gronwalls lemma
that sup
t[0,T)
Z(t) is nite and, as
= y
Id and = H
ac
, satises
ac
= (u
2
+u
2
x
) dx. (3.1)
A Radon measure is said to be 1periodic if (1 +B) = (B) for all Borel sets
B. The equivalent system (2.15) was derived by using the characteristics. Since y
satises (2.1), y, for a given , can also be seen as the position of a particle evolving
in the velocity eld u, where u is the solution of the CamassaHolm equation. We
are then working in Lagrangian coordinates. In [15], the CamassaHolm equation
is derived as a geodesic equation on the group of dieomorphism equipped with
a rightinvariant metric. In the present paper, the geodesic curves correspond to
y(t, ). Note that y does not remain a dieomorphism since it can become non
invertible, which agrees with the fact that the solutions of the geodesic equation
may break down, see [11]. The rightinvariance of the metric can be interpreted
as an invariance with respect to relabeling as noted in [3]. This is a property that
we also observe in our setting. We denote by G the subgroup of the group of
homeomorphisms on the unit circle dened as follows: f G if f is invertible,
f W
1,
loc
(R), f( + 1) = f() + 1 for all R, and (3.2)
f Id and f
1
Id both belong to W
1,
per
. (3.3)
The set G can be interpreted as the set of relabeling functions. For any > 1, we
introduce the subsets G
of G dened by
G
= f G [ f Id
W
1, +
_
_
f
1
Id
_
_
W
1,
.
14 HOLDEN AND RAYNAUD
The subsets G
.
Lemma 3.2. Let 0. If f belongs to G
, then 1/(1 + ) f
1 +
almost everywhere. Conversely, if f satises (3.2) and there exists c 1 such that
1/c f
f
1
(f(
)) f
1
(f())
f(
) f()
= (f
1
)
(f())
because f is continuous and f
1
is dierentiable at f(). On the other hand, we
have
f
1
(f(
)) f
1
(f())
f(
) f()
=
f(
) f()
.
Hence, f is dierentiable for any f
1
(B) and
f
()
1
(f
1
)

L
1
1 +
. (3.4)
The estimate (3.4) holds only on f
1
(B) but, since meas(B
c
) = 0 and f
1
is
Lipschitz and onetoone, meas(f
1
(B
c
)) = 0 (see, e.g., [2, Remark 2.72]), and
therefore (3.4) holds almost everywhere. We have f
1 +f
1
L
1 +.
Let us now consider a function f that satises (3.2) and such that 1/c f
c
almost everywhere for some c 1. Since f
1
,
2
in R such that
1
<
2
,
f
1
(
2
) f
1
(
1
) =
_
[f
1
(1),f
1
(2)]
f
d c(
2
1
).
Hence, f
1
is Lipschitz and (f
1
)
c. We have f
1
(
= f() for
. Finally, we get
f Id
W
1, +
_
_
f
1
Id
_
_
W
1,
2 f Id
L
+ 2
+f

L
+
_
_
(f
1
)
_
_
L
2 f Id
L
+ 2 + 2c.
Since f Id
L
_
1
0
([f
and T of ( as follows
T
= X = (y, U, H) ( [
1
1 +h
(y +H) G
,
and
T = X = (y, U, H) ( [
1
1 +h
(y +H) G.
We recall that h = H( + 1) H() = H(1) H(0). For = 0, G
0
= Id.
As we will see, the space T
0
will play a special role. These sets are relevant only
because they are in some sense preserved by the governing equation (2.15) as the
next lemma shows.
Lemma 3.3. The space T is preserved by the governing equation (2.15). More
precisely, given , T 0 and
X T
, we have
S
t
(
X) T
h
( y()+
c almost
everywhere, for some constant c > 1 depending only . We consider a xed and
drop it in the notation. Applying Gronwalls inequality backward in time to (2.16),
we obtain
[y
(0)[ +[H
(0)[ +[U
(0)[ e
CT
([y
(t)[ +[H
(t)[ +[U
(t)[) (3.5)
for some constant C which depends on X(t)
C([0,T],E)
, which itself depends only
on
_
_
X
_
_
E
and T. From (2.25c), we have
[U
(t)[
_
y
(t)H
(t)
1
2
(y
(t) +H
(t)).
Hence, since y
and H
+
H
3
2
e
CT
(y
(t) +H
(t)),
and v
(t) =
1
1+h
(y
(t) + H
(t))
2
3c
e
CT
. Similarly, by applying Gronwalls
lemma forward in time, we obtain v
=
1
1+h
(y
(t) + H
(t))
3
2
ce
CT
. We have
1
1+h
(y +H)(t, +1) =
1
1+h
(y +H)(t, ) +1. Hence, applying Lemma 3.2, we obtain
that
1
1+h
(y(t, ) +H(t, )) G
for some
depending
only on , T and
_
_
X
_
_
E
.
For the sake of simplicity, for any X = (y, U, H) T and any function f G,
we denote (y f, U f, H f) by X f. We denote by
G the product group GR.
The group operation on
G is given by (f
1
,
1
) (f
2
,
2
) = (f
2
f
1
,
1
+
2
) where
(f
1
,
1
) and (f
2
,
2
) are two elements of
G. We dene the map :
G T T as
follows
_
_
y = y f,
H = H f +,
U = U f,
where ( y,
U,
H) = (f, ), (y, U, H). We denote ( y,
U,
H) = (f, ) (y, U, H).
Proposition 3.4. The map denes a group action of
G on T.
Proof. It is clear that satises the fundamental property of a group action, that
is, (f
2
,
2
) ((f
1
,
1
) X) = (f
1
f
2
,
1
+
2
) X for all X T and all (f
1
,
1
) and
(f
2
,
2
) in
G. It remains to prove that (f, ) X indeed belongs to T. It is not hard
to check that (0, ) X belongs to T. Thus, by the group action property, we only
have to show that (f, 0)X = Xf belongs to T. We denote
X = ( y,
U,
H) = Xf.
As compositions of two Lipschitz maps, y,
U and
H are Lipschitz. It is not hard to
check that y( +1) = y()+1,
U( +1) =
U() and
H( +1) =
H()+H(1)H(0),
for all R. Let us prove that
y
= y
f f
,
U
= U
f f
and
H
= H
f f
(3.6)
almost everywhere. Let B
1
be the set where y is dierentiable and B
2
the set where
y and f are dierentiable. Using Radamachers theorem, we get that meas(B
c
1
) =
meas(B
c
2
) = 0. For B
3
= B
2
f
1
(B
1
), we consider a sequence
i
converging
to (
i
,= ). We have
y(f(
i
)) y(f())
f(
i
) f()
f(
i
) f()
=
y(
i
) y()
. (3.7)
16 HOLDEN AND RAYNAUD
Since f is continuous, f(
i
) converges to f() and, as y is dierentiable at f(), the
lefthand side of (3.7) tends to y
f() f
(f())f
() = y
() (3.8)
for all B
3
. Since f
1
is Lipschitz, onetoone and meas(B
c
1
) = 0, we have
meas(f
1
(B
1
)
c
) = 0 and therefore (3.8) holds everywhere. One proves the two
other identities in (3.6) similarly. From Lemma 3.2, we have that f
> 0 almost
everywhere. Then, using (3.6) we easily check that (2.25b) and (2.25c) are fullled.
Thus, we have proved that ( y,
U,
H) fullls (2.25). We have
h =
H(1)
H(0) =
H(1) H(0) = h. Hence,
1
1+
h
( y +
H) =
1
1+h
(y + H) f which implies, since
1
1+h
(y +H) and f belongs to G and G is a group, that
1
1+
h
( y +
H) G. Therefore
1
(X) = X f
1
with f =
1
1+h
(y +H) T and X = (y, U, H), and
_
_
y = y( a)
H = H( a) + (1 +h)a
U = U( a)
with a =
_
1
0
y() d and ( y,
U,
H) =
2
(y, U, H). In fact,
1
(X) = (f
1
, 0) X and
2
(X) = (
a
, (1 + h)a) X where
a
denotes the translation by a. After noting
that the group action let invariant the quantity h = H(1) H(0), it is not hard to
check that
1
(X) indeed belongs to T
0
, that is,
1
1+
h
( y +
H) = Id. Let us prove that
( y,
U,
H) =
2
(y, U, H) belongs to H for any (y, U, H) T
0
. On the one hand, we
have
1
1 +
h
( y +
H)() =
1
1 +h
[(y +H) ( a) + (1 +h)a] =
because
h = h and
1
1+h
(y +H) = Id as (y, U, H) T
0
. On the other hand,
_
1
0
y() d =
_
1a
a
y() d =
_
1
0
y() d +
_
0
a
y() d +
_
1a
1
y() d
and, since y( + 1) = y() + 1, we obtain
_
1
0
y() d =
_
1
0
y() d +
_
0
a
y() d +
_
a
0
y() d a =
_
1
0
y() dx a = 0.
Thus
2
(X) H. We denote the composition map
2
1
from T to H by . We
have
(X) = (
a
, (1 +h)a) ((f
1
, 0) X) = (f
1
a
, (1 +h)a) X
where f and a has been dened above. Thus, (X) belongs to the same equivalence
class as X, and we can dene the map
: T/
2
let invariant H so that
H
= Id
H
. Hence,
H
= Id
H
and it follows that
is a bijection from T/
G to H.
Any topology dened on H is naturally transported into T/
G by the bijection
.
We equip H with the metric induced by the Enorm, i.e., d
H
(X, X
) = X X

E
for all X, X
G as
d
F/
G
([X], [X
]) =
_
_
_
([X])
([X
])
_
_
_
E
,
for any [X], [X
] T/
G. Then, T/
to H.
Proof. We prove rst that
1
is continuous from T
to T
0
and then, that
2
is
continuous from T
0
to H. We equip T
. We denote
X
n
= ( y
n
,
U
n
,
H
n
) =
1
(X
n
) and
X = ( y,
U,
H) =
1
(X). By
denition of T
0
, we have
H
n
=
n
+
h
n
(recall that
n
= y
n
Id). Let us prove
rst that
H
n
tends to
H in L
per
. We denote f
n
=
1
1+hn
(y
n
+H
n
), f =
1
1+h
(y +H),
and we have f
n
, f G
. Thus
H
n
H = (H
n
H) f
n
1
+
H f f
n
1
H and
we have
_
_
H
n
H
_
_
L
H
n
H
L
+
_
_
H f
H f
n
_
_
L
. (3.9)
From the denition of T
0
, we know that
H is Lipschitz with Lipschitz constant
smaller than 1 +
h
n
. Hence,
_
_
H f
H f
n
_
_
L
(1 +
h
n
) f
n
f
L
. (3.10)
Since H
n
and f
n
converges to H and f, respectively, in L
per
and
h
n
= h
n
converges
to h, from (3.9) and (3.10), we get that
H
n
converges to
H in L
per
. Let us prove
now that
H
n,
tend to
H
in L
2
per
. We have
H
n,
H
=
H
n,
f
n,
f
n
1
f
1
which can be decomposed into
H
n,
H
=
_
H
n,
H
f
n,
_
f
n
1
+
H
f
n,
f
n
1
f
1
. (3.11)
Since f
n
G
f
n,
_
f
n
1
_
_
_
_
2
L
2
=
_
1
0
(H
n,
H
)
2
1
f
n,
d c H
n,
H

2
L
2
, (3.12)
where we have made the change of variables
= f
n
1
(). Hence, the lefthand
side of (3.12) converges to zero. If we can prove that
H
f
n,
f
n
1
f
1
in
L
2
per
, then, using (3.11), we get that
H
n,
H
in L
2
per
, which is the desired result.
We recall that, since the space V and H
1
per
R are homeomorphic, H
n
H in V
is equivalent to (
n
, h
n
) (, h) in H
1
per
R. We have
H
f
n,
f
n
1
=
(
H
f)f
f
n,
f
n
1
= (
H
g
n
)g
n,
where g
n
= f f
n
1
. Let us prove that lim
n
g
n,
1
L
2
= 0. We have, after
using a change of variables,
g
n,
1
2
L
2
=
_
1
0
_
f
f
n,
f
n
1
1
_
2
d = c f
f
n,

2
L
2
. (3.13)
18 HOLDEN AND RAYNAUD
Hence, since f
n,
f
in L
2
per
, lim
n
g
n,
1
L
2
= 0. We have
_
_
H
g
n
g
n,
H
_
_
L
2
_
_
H
g
n
_
_
L
g
n,
1
L
2
+
_
_
H
g
n
_
_
L
2
. (3.14)
We have
_
_
H
g
n
_
_
L
1 + h
n
since, as we already noted,
H is Lipschitz with
Lipschitz constant smaller than 1 +
h
n
= 1 +h
n
. Hence, the rst term in the sum
in (3.14) converges to zero. As far as the second term is concerned, one can always
approximate
H
in L
2
per
by a periodic continuous function v. After observing that
1/c
2
g
n,
c
2
almost everywhere, we can prove, as we have done several times
now, that H
g
n
v g
n

2
L
2
c
2
H
v
2
L
2
and vg
n
can be chosen arbitrarily
close to H
g
n
in L
2
independently of n, that is, for all > 0, there exists v such
that
H
g
n
v g
n

L
2
3
and H
v
L
2
3
(3.15)
for all n. By Lebesgues dominated convergence theorem, we have vg
n
v in L
2
per
.
Hence, for n large enough, we have v g
n
v
L
2
3
which, together with (3.15),
implies
_
_
H
g
n
_
_
L
2
, and
H
g
n
H
in L
2
per
. It remains to prove that
2
is continuous from T
0
to H. We consider a sequence X
n
= (y
n
, U
n
, H
n
) in T
0
which converges to X = (y, U, H) and denote a
n
=
_
1
0
y
n
() d and a =
_
1
0
y() d.
We set
X
n
=
2
(X
n
) and
X =
2
(X). Since y
n
y in L
, a
n
a. We have
y
n
= y
n
an
,
H = H
n
an
+(1+h
n
)a
n
,
U
n
= U
an
where h
n
= H
n
(1) H
n
(0).
Since,
an
a
in H
1
and
an,
= 1 so that the
an,
are clearly uniformly bounded
away from zero and innity, we can repeat the proof of continuity of
1
and prove
that y
n
an
y
a
, H
n
an
H
a
and U
n
an
U
a
in H
1
. Then, as
a
n
a, it follows that
X
n
X and the continuity of
2
is proved.
Remark 3.6. The map
1
is not continuous from T to T
0
and therefore neither
is the map from T to H. The spaces T
G to T/
G given by
S
t
([X]) = [S
t
X]
is welldened. It generates a continuous semigroup.
Proof. For any X
0
= (y
0
, U
0
, H
0
) T and (f, )
G, we denote
X
0
= ( y
0
,
U
0
,
H
0
) =
(f, ) X
0
, X(t) = S
t
(X
0
) and
X(t) = S
t
(
X
0
). We claim that (f, ) X(t) sat
ises (2.15) and therefore, since (f, ) X(t) and
X(t) satisfy the same system
of dierential equation with the same initial data, they are equal. We denote
X(t) = ( y(t),
U(t),
H(t)) = (f, ) X(t). We have
U
t
=
1
4
_
R
sgn( ) exp
_
sgn( )( y() y())
_ _
U()
2
y
() +H
()
d.
(3.17)
We have y
() = y
(f())f
() and
H
() = H
(f())f
U
t
=
1
4
_
R
sgn( ) exp
_
sgn( )( y() y())
_
_
U()
2
y
() +
H
()
_
d.
PERIODIC CONSERVATIVE SOLUTIONS OF THE CAMASSAHOLM EQUATION 19
We treat similarly the other terms in (2.15), and it follows that ( y,
U,
H) is a
solution of (2.15). Since ( y,
U,
H) and ( y,
U,
H) satisfy the same system of ordinary
dierential equations with the same initial data, they are equal, i.e.,
X(t) = (f, )
X(t) and (3.16) is proved. We have the following diagram:
H
//
T/
G
T
OO
H
St
OO
//
T/
St
OO
(3.18)
on a bounded domain of H whose diameter together with t determines the constant
, see Lemma 3.3. By the denition of the metric on T/
G, the map is an isometry
from H to T/
G
is continuous if and only if S
t
: H H is continuous. Let us prove that
S
t
: H H is sequentially continuous. We consider a sequence X
n
H that
converges to X H in H, that is, lim
n
X
n
X
E
= 0. From Theorem 2.7,
we get that lim
n
S
t
(X
n
) S
t
(X)
E
= 0. Since X
n
X in E, there exists a
constant C 0 such that X
n
 C for all n. Lemma 3.3 gives us that S
t
(X
n
) T
. Then, by
Lemma 3.5, we obtain that S
t
(X
n
) S
t
(X) in H.
3.2. Maps between the two coordinate systems. Our next task is to derive
the correspondence between Eulerian coordinates (functions in T) and Lagrangian
coordinates (functions in T/
G). Earlier we considered initial data in T with a
special structure: The energy density was given by (u
2
+ u
2
x
) dx and therefore
did not have any singular part. The set T however allows the energy density to
have a singular part and a positive amount of energy can concentrate on a set of
Lebesgue measure zero. We constructed corresponding initial data in T
0
by the
means of (2.24). This construction can be generalized to the set T. To any positive
periodic Radon measure , we associate the function F
dened as
F
(x) =
_
_
([0, x)) if x > 0,
0 if x = 0,
([x, 0)) if x < 0.
The function F
(b) F
(z + 1) = F
(z)+z < (1+h), and it follows that y( +1) = y()+1 and H( +1) = H()+h
for all R. The function F
(z) + z
and therefore y are also increasing. Let us prove that y is Lipschitz with Lipschitz
constant at most 1+h. We consider ,
) (the
case y() = y(
i
, and a decreasing one, x
i
such that lim
i
x
i
= y(),
lim
i
x
i
= y(
) with F
(x
i
) + x
i
< (1 + h)
and F
(x
i
) + x
i
(1 + h).
Subtracting the second inequality from the rst, we obtain
F
(x
i
) F
(x
i
) +x
i
x
i
< (1 +h)(
). (3.22)
For i large enough, since by assumption y() < y(
), we have x
i
< x
i
and therefore
F
(x
i
) F
(x
i
) = ([x
i
, x
i
)) 0. Hence, x
i
x
i
< (1 +h)(
). Letting i tend
to innity, we get y(
) y() (1+h)(
(x
i
) + x
i
< (1 + h). Since F
is lower semi
continuous, lim
i
F
(x
i
) = F
and U
i
in A such that
1
2
(y(
i
)+y(
i
)) =
y() and y(
i
) y(
i
) <
1
i
. We have, by (3.19) and (3.23), since y(
i
) and y(
i
)
belong to A,
([y(
i
), y(
i
))) +y(
i
) y(
i
) = (1 +h)(
i
). (3.25)
Since y(
i
) / supp(
s
), (y(
i
)) = 0 and ([y(
i
), y(
i
))) = ((y(
i
), y(
i
))).
Dividing (3.25) by
i
i
and letting i tend to , we obtain
y
()(u
2
+u
2
x
)(y()) +y
() = 1 +h (3.26)
PERIODIC CONSERVATIVE SOLUTIONS OF THE CAMASSAHOLM EQUATION 21
where y is dierentiable in y
1
(B
1
), that is, almost everywhere in y
1
(B
1
). We
now derive a short lemma which will be useful several times in this proof.
Lemma 3.9. Given an increasing Lipschitz function f : R R, for any set B of
measure zero, we have f
= 0 almost everywhere in f
1
(B).
Proof of Lemma 3.9. The lemma follows directly from the area formula:
_
f
1
(B)
f
() d =
_
R
H
0
_
f
1
(B) f
1
(x)
_
dx (3.27)
where H
0
is the multiplicity function, see [2] for the formula and the precise deni
tion of H
0
. The function H
0
_
f
1
(B) f
1
(x)
_
is Lebesgue measurable (see [2])
and it vanishes on B
c
. Hence,
_
f
1
(B)
f
0, f
= 0
almost everywhere in f
1
(B).
We apply Lemma 3.9 to B
c
1
and get, since meas(B
c
1
) = 0, that y
= 0 almost
everywhere on y
1
(B
c
1
). On y
1
(B
1
), we proved that y
= 1 +h y
, that
H
0. In the same way as we proved that y was Lipschitz with Lipschitz constant
at most 1 + h, we can prove that the function
_
y()
u
2
x
dx is also Lipschitz.
Indeed, from (3.22), for i large enough, we have
_
x
i
xi
u
2
x
dx
ac
([x
i
, x
i
)) ([x
i
, x
i
)) = F
(x
i
) F
(x
i
) < (1 +h)(
).
Since lim
i
x
i
= y(
) and lim
i
x
i
= y(), letting i tend to innity, we obtain
_
y(
)
y()
u
2
x
dx < (1 + h)(
) R
2
, we have, after using the
CauchySchwarz inequality,
[U(
) U()[ =
_
y(
)
y()
u
x
dx
_
y(
) y()
_
y(
)
y()
u
2
x
dx (3.28)
(1 +h) [
[
because y and
_
y()
0
u
2
x
dx are Lipschitz with Lipschitz constant at most 1 +h.
Hence, U is also Lipschitz and therefore dierentiable almost everywhere. We
denote by B
2
the set of Lebesgue points of u
x
in B
1
, i.e.,
B
2
= x B
1
[ lim
0
1
_
x+
x
u
x
(t) dt = u
x
(x).
We have meas(B
c
2
) = 0. We choose a sequence
i
and
i
such that
1
2
(y(
i
)+y(
i
)) =
x and y(
i
) y(
i
)
1
i
. Thus
U(
i
) U(
i
)
i
=
_
y(
i
)
y(i)
u
x
(t) dt
y(
i
) y(
i
)
y(
i
) y(
i
)
i
.
Hence, letting i tend to innity, we get that for every in y
1
(B
2
) where U and y
are dierentiable, that is, almost everywhere on y
1
(B
2
),
U
() = y
()u
x
(y()). (3.29)
22 HOLDEN AND RAYNAUD
From (3.28) and using the fact that
_
y()
0
u
2
x
dx is Lipschitz with Lipschitz
constant at most 1 +h, we get
U(
) U()
1 + h
y(
) y()
.
Hence,
[U
()[
_
y
(). (3.30)
Since meas(B
c
2
) = 0, we have by Lemma 3.9, that y
= 0 almost everywhere on
y
1
(B
c
2
). Hence, U
= 0 almost everywhere on y
1
(B
c
2
). Thus, we have computed
U
= y
(1 + h y
) = y
2
(u
2
+ u
2
x
) y and, nally, y
=
y
2
U
2
+ U
2
almost everywhere on y
1
(B
2
). On y
1
(B
c
2
), we have y
= U
= 0
almost everywhere. Therefore (2.25c) is satised almost everywhere. By denition,
we have
1
1+h
(y +H) = Id, which concludes the proof of the theorem.
Remark 3.10. If is absolutely continuous, then = (u
2
+ u
2
x
)dx and, from
(3.23), we get
_
y()
0
(u
2
+u
2
x
) dx +y() = (1 +h)
for all R.
At the very beginning, H(t, ) was introduced as the energy contained in a strip
between y(0, ) and y(t, ), see (2.3). This interpretation still holds. We obtain
, the energy density in Eulerian coordinates, by pushing forward by y the energy
density in Lagrangian coordinates, H
d) (3.31b)
belongs to T and is independent of the representative X = (y, U, H) T we choose
for [X]. We denote by M: T/
G T the map which to any [X] in T/
G associates
(u, ) as given by (3.31).
Proof. First we have to prove that the denition of u makes sense. Since y is
surjective, there exists , which may not be unique, such that x = y(). It remains
to prove that, given
1
and
2
such that x = y(
1
) = y(
2
), we have
U(
1
) = U(
2
). (3.32)
Since y() is an increasing function in , we must have y() = x for all [
1
,
2
]
and therefore y
() = 0 in [
1
,
2
]. From (2.25c), we get that U
() = 0 for all
[
1
,
2
] and (3.32) follows. It is not hard to check that u(x + 1) = u(x).
Since y is proper and H
such that x = y(
) and u(x) =
U(
). Let = f
1
(
).
As x = y(
), we
nally obtain u(x) = u(x). For any continuous function with compact support ,
we have
_
R
d =
_
R
y(
)
H
) d
,
see [2]. Hence, after making the change of variables
= f(), we obtain
_
R
d =
_
R
y f()
H
f() f
() d
and, since H
=
H
ff
almost everywhere,
_
R
d =
_
R
y()H
() d =
_
R
d.
Since was arbitrary in C
c
(R), we get = . This proves that X and
X give raise
to the same pair (u, ), which therefore does not depend on the representative of
[X] we choose.
Let us prove that u H
1
. We have u L
() d =
_
y
1
(U)
U
()( y)() d
(3.33)
after integrating by parts. Let B
1
= y
1
() [ y
() > 0. Because of
(2.25c), and since y
= 0 almost everywhere on
B
c
1
. Hence, we can restrict the integration domain in (3.33) to B
1
. We divide and
multiply by
_
U
u
x
dx
_
B1
U
( y)
_
B1
U
2
_
B1
( y)
2
y
d.
By (2.25c), we have
U
2
_
U
u
x
dx
_
y
1
(U)
H
d 
L
2
(U)
. (3.34)
Since lim
y() = , y
1
() is bounded and (3.34) implies that u
x
L
2
().
As  was an arbitrary bounded open set, it follows that u
x
L
2
loc
.
Let us prove that the absolutely continuous part of is equal to (u
2
+ u
2
x
) dx.
We introduce the sets Z and B dened as follows
Z =
_
R [ y is dierentiable at and y
() = 0
or y or U are not dierentiable at
_
and
B = x y(Z)
c
[ u is dierentiable at x .
Since u belongs to H
1
, it is dierentiable almost everywhere. We have, since y is
Lipschitz and by the denition of Z, that meas(y(Z)) =
_
Z
y
() d = 0. Hence,
meas(B
c
) = 0. For any y
1
(B), we denote x = y(). By necessity, we have
Z
c
. Let
i
be a sequence converging to such that
i
,= for all i. We write
24 HOLDEN AND RAYNAUD
x
i
= y(
i
). Since y
=
u(x
i
) u(x)
x
i
x
x
i
x
.
Since u is dierentiable at x and belongs to Z
c
, we obtain, after letting i tend to
innity, that
U
() = u
x
(y())y
(). (3.35)
For all subsets B
of B, we have
(B
) =
_
y
1
(B
)
H
d =
_
y
1
(B
)
_
U
2
+
U
2
y
2
_
y
d.
We can divide by y
) =
_
B
(u
2
+u
2
x
) dx. (3.36)
Since (3.36) holds for any set B
B and meas(B
c
) = 0, we have
ac
= (u
2
+
u
2
x
) dx.
The next theorem shows that the transformation from Eulerian to Lagrangian
coordinates is a bijection.
Theorem 3.12. The maps M and L are invertible. We have
L M = Id
F/
G
and M L = Id
D
.
Proof. Given [X] in T/
G, we choose X = (y, U, H) =
([X]) as a representative
of [X] and consider (u, ) given by (3.31) for this particular X. Note that, from
the denition of
, we have X H. Let
X = ( y,
U,
H) be the representative of
L(u, ) in T
0
given by the formulas (3.20). We claim that ( y,
U,
H) and (y, U, H)
are equivalent and therefore L M = Id
F/
G
. Let
g(x) = sup R [ y() < x. (3.37)
It is not hard to prove, using the fact that y is increasing and continuous, that
y(g(x)) = x (3.38)
and y
1
([a, b)) = [g(a), g(b)) for any a < b in R. Hence, by (3.31b), we have
([a, b)) =
_
y
1
([a,b))
H
d =
_
g(b)
g(a)
H
, it follows
then that
F
G
.
Given (u, ) in T, we denote by (y, U, H) the representative of L(u, ) in T
0
given by (3.20). Then, let ( u, ) = M L(u, ). We claim that ( u, ) = (u, ). Let
g be the function dened as before by (3.37). The same computation that leads to
(3.40) now gives
F
(x) +x = (1 +h)(g(x) g(0)). (3.45)
Given R, we consider an increasing sequence x
i
which converges to y() and
such that F
(x
i
) + x
i
< (1 + h). The existence of such sequence is guaranteed
by (3.20b). Passing to the limit and since F
(x) + x
(1 + h)
i
. Letting i tend to innity, we obtain F
= F
+(1+h)g(0). Hence = . It is
clear from the denitions that u = u. Hence, ( u, ) = (u, ) and M L = Id
D
.
4. Continuous semigroup of solutions on T
The fact that we have been able to establish a bijection between the two coor
dinate systems, T/
G and T, enables us now to transport the topology dened in
T/
G
(L(u, ), L( u, )).
Since T/
G equipped with d
F/
G
is a complete metric space, we have the following
theorem.
Theorem 4.1. T equipped with the metric d
D
is a complete metric space.
For each t R, we dene the map T
t
from T to T by
T
t
= M
S
t
L.
We have the following commutative diagram:
T T/
G
M
oo
T
Tt
OO
L
//
T/
St
OO
(4.1)
Our main theorem reads as follows.
26 HOLDEN AND RAYNAUD
Theorem 4.2. T : T R
+
T (where T is dened by Denition 3.1) denes a
continuous semigroup of solutions of the CamassaHolm equation, that is, given
( u, ) T, if we denote t (u(t), (t)) = T
t
( u, ) the corresponding trajectory,
then u is a weak solution of the CamassaHolm equation (1.3). Moreover is a
weak solution of the following transport equation for the energy density
t
+ (u)
x
= (u
3
2Pu)
x
. (4.2)
Furthermore, we have that
(t)([0, 1)) = (0)([0, 1)) for all t, (4.3)
and
(t)([0, 1)) =
ac
(t)([0, 1)) = u(t)
2
H
1
per
= (0)([0, 1)) for almost all t. (4.4)
Remark 4.3. We denote the unique solution described in the theorem as a con
servative weak solution of the CamassaHolm equation.
Proof. The proof is similar to the non periodic case. We want to prove that, for all
C
(R
+
R) with compact support,
_
R+R
[u(t, x)
t
(t, x) +u(t, x)u
x
(t, x)(t, x)] dxdt =
_
R+R
P
x
(t, x)(t, x) dxdt
(4.5)
where P is given by (2.11) or equivalently (2.6). Let (y(t), U(t), H(t)) be a rep
resentative of L(u(t), (t)) which is solution of (2.15). Since y is Lipschitz in
and invertible for t /
c
(see (2.32) for the denition of /, in particular, we have
meas(/) = 0), we can use the change of variables x = y(t, ) and, using (3.29), we
get
_
R+R
[u(t, x)
t
(t, x) +u(t, x)u
x
(t, x)(t, x)] dxdt
=
_
R+R
[U(t, )y
(t, )
t
(t, y(t, )) +U(t, )U
y)
t
(U
2
)
= Uy
t
y UU
y +U
t
y
y. (4.7)
After integrating (4.7) over R
+
R, the lefthand side of (4.7) vanishes and we
obtain
_
R+R
[Uy
t
y +UU
y ] ddt
=
1
4
_
R+R
2
_
sgn()e
{sgn()(y()y()}
_
U
2
y
+H
_
()y
()y()
_
dddt
(4.8)
by (2.15). Again, to simplify the notation, we deliberately omitted the t variable.
On the other hand, by using the change of variables x = y(t, ) and z = y(t, )
when t /
c
, we have
_
R+R
P
x
(t, x)(t, x) dxdt =
1
2
_
R+R
2
_
sgn(y() y())e
y()y()
_
u
2
(t, y()) +
1
2
u
2
x
(t, y())
_
(t, y())y
()y
()
_
dddt.
PERIODIC CONSERVATIVE SOLUTIONS OF THE CAMASSAHOLM EQUATION 27
Since, from Lemma 2.6, y
(t, )/y
_
R+R
P
x
(t, x)(t, x) dxdt =
1
4
_
R+R
2
_
sgn() exp
_
sgn()(y()y()
_
_
U
2
y
+H
_
()y
()(t, y())
_
dddt. (4.9)
Thus, comparing (4.8) and (4.9), we get
_
R+R
[Uy
t
(t, y) +UU
] ddt =
_
R+R
P
x
(t, x)(t, x) dxdt
and (4.5) follows from (4.6). Similarly, one proves that (t) is solution of (4.2). We
have y
1
([0, 1)) = [g(0), g(1)) where g is given by (3.37). From (3.37) and the fact
that y( +1) = y() +1 for all , we infer that g(x+1) = g(x) +1. Hence, it follows
from (3.31a), since H
is periodic, that
(t)([0, 1)) =
_
[g(0),g(0)+1)
H
d =
_
[0,1)
H
d = H(t, 1) H(t, 0)
which is constant in time, from the governing equation (2.15). Hence, (4.3) is
proved. We know from Lemma 2.6 (ii) that, for t /
c
, y
d =
_
y
1
(B)
_
U
2
+
U
2
y
2
_
y
d (4.10)
from (2.25c) and because y
= U
per
.
Proposition 5.1. The map
u (u, (u
2
+u
2
x
)dx)
is continuous from H
1
per
into T. In other words, given a sequence u
n
H
1
per
converging to u in H
1
per
, then (u
n
, (u
2
n
+ u
2
nx
)dx) converges to (u, (u
2
+ u
2
x
)dx) in
T.
Proof. Let X
n
= (y
n
, U
n
, H
n
) and X = (y, U, H) be the representatives in T
0
given
by (3.20) of L(u
n
, (u
2
n
+u
2
nx
)dx) and L(u, (u
2
+u
2
x
)dx), respectively. By denition
of the topology of T and T/
G, we have to prove that (X
n
) (X) in H. Since
is continuous from T
0
into H, see Lemma 3.5, it is enough to prove that X
n
X
28 HOLDEN AND RAYNAUD
in E. We write g
n
= u
2
n
+ u
2
n,x
and g = u
2
+ u
2
x
; g
n
and g are periodic. Following
Remark 3.10, we have
_
y()
0
g(x) dx +y() = (1 +h) ,
_
yn()
0
g
n
(x) dx +y
n
() = (1 +h
n
) (5.1)
and, after taking the dierence between the two equations, we obtain
_
y()
0
(g g
n
)(x) dx +
_
y()
yn()
g
n
(x) dx +y() y
n
() = (h
n
h). (5.2)
Since g
n
is positive,
y y
n
+
_
y
yn
g
n
(x) d)
= [y y
n
[ +
_
y
yn
g
n
(x) d)
and (5.2)
implies
[y() y
n
()[
_
y()
0
[g g
n
[ dx +[h
n
h[ [[ g g
n

L
1 +[h
n
h[
because y(0) = 0 and therefore [0, y()] [0, 1] for [0, 1]. Since u
n
u in
H
1
, g
n
g in L
1
and it follows that
n
and H
n
H in L
. We recall that
() = y() and H = h (as X, X
n
T
0
). We have
U
n
U = u
n
y
n
u y
n
+u y
n
u y. (5.3)
Since u
n
u in L
, u
n
y
n
u y
n
in L
, u y
n
u y in L
. The measures (u
2
+ u
2
x
)dx and (u
2
n
+ u
2
n,x
)dx have, by
denition, no singular part and in that case (3.26) holds almost everywhere, that
is,
y
=
1
g y + 1
and y
n,
=
1
g
n
y
n
+ 1
(5.4)
almost everywhere. Hence,
n,
= (g y g
n
y
n
)y
n,
y
= (g y g y
n
)y
n,
y
+ (g y
n
g
n
y
n
)y
n,
y
. (5.5)
Since 0 y
1 +h, we have
_
[0,1]
[g y
n
g
n
y
n
[ y
n,
y
d (1 + h)
_
[0,1]
[g y
n
g
n
y
n
[ y
n,
d (5.6)
= (1 + h) g g
n

L
1
.
Let C = sup
n
(1 + h
n
). For any > 0, there exists a continuous function v with
compact support such that g v
L
1 /3C. We can decompose the rst term in
the righthand side of (5.5) into
(g y g y
n
)y
n,
y
= (g y v y)y
n,
y
+ (v y v y
n
)y
n,
y
+ (v y
n
g y
n
)y
n,
y
. (5.7)
Then, we have
_
[0,1]
[g y v y[ y
n,
y
d (1 +h
n
)
_
[g y v y[ y
d C g v
L
1 /3
and, similarly, we obtain
_
[0,1]
[g y
n
v y
n
[ y
n,
y
d /3. Since y
n
y in L
d C
2
v y v y
n

L
1 /3.
PERIODIC CONSERVATIVE SOLUTIONS OF THE CAMASSAHOLM EQUATION 29
Hence, from (5.7), we get that
_
[0,1]
[g y g y
n
[ y
n,
y
d so that
lim
n
_
[0,1]
[g y g y
n
[ y
n,
y
d = 0,
and, from (5.5) and (5.6), it follows that
n,
in L
1
. Since X
n
T
0
,
n,
is
bounded in L
in L
2
and, by (3.20c), H
n,
H
in L
2
. It remains to prove that U
n,
U
in L
2
. Since y
n,
, H
n,
and U tend to
y
, H
and U in L
2
, respectively and U
n

L
, y
n,

L

L
2
. (5.8)
Once we have proved that U
n,
converges weakly to U
strongly in L
2
, see, for example, [28, section V.1]. For any continuous
function with compact support, we have
_
R
U
n,
d =
_
R
u
n,x
y
n
y
n,
d =
_
R
u
n,x
y
n
1
d. (5.9)
By assumption, we have u
n,x
u
x
in L
2
. Since y
n
y in L
, the support of
y
n
1
is contained in some compact that can be chosen to be independent of
n. Thus, after using Lebesgues dominated convergence theorem, we obtain that
y
n
1
y
1
in L
2
and therefore
lim
n
_
R
U
n,
d =
_
R
u
x
y
1
d =
_
R
U
d. (5.10)
From (5.8), we have that U
n,
is bounded and therefore, by a density argument,
(5.10) holds for any function in L
2
and U
n,
U
weakly in L
2
.
Proposition 5.2. Let (u
n
,
n
) be a sequence in T that converges to (u, ) in T.
Then
u
n
u in L
per
and
n
.
Proof. We denote by X
n
= (y
n
, U
n
, H
n
) and X = (y, U, H) the representative in
H of L(u
n
,
n
) and L(u, ). Let C = sup
n
(1 + h
n
). For any x R, there exists
n
and , which may not be unique, such that x = y
n
(
n
) and x = y(). We set
x
n
= y
n
(). We have
u
n
(x) u(x) = u
n
(x) u
n
(x
n
) +U
n
() U(), (5.11)
and
[u
n
(x) u
n
(x
n
)[ =
_
n
U
n,
() d
_
[
n
[
_
_
n
U
2
n,
d
_
1/2
(CauchySchwarz)
_
[
n
[
_
_
n
y
n,
H
n,
d
_
1/2
(from (2.25c))
C
_
[
n
[
_
[y
n
(
n
) y
n
()[ (since H
n,
C)
= C
_
[
n
[
_
[y() y
n
()[
C
_
[
n
[ y y
n

1/2
L
. (5.12)
We have
[y
n
(
n
) y
n
()[ = [y() y
n
()[ y
n
y
L
. (5.13)
30 HOLDEN AND RAYNAUD
Without loss of generality, we can assume that y
n
y
L
< 1 so that (5.13)
implies [
n
[ < 1 as y
n
is increasing and y
n
(
+ 1) = y
n
(
) + 1 for all
. Hence,
(5.12) implies
[u
n
(x) u
n
(x
n
)[ C y y
n

1/2
L
. (5.14)
Since y
n
y and U
n
U in L
d (5.16)
see [2, Denition 1.70]. Since y
n
y in L
, the support of y
n
is contained
in some compact which can be chosen independently of n and, from Lebesgues
dominated convergence theorem, we have that y
n
y in L
2
. Hence, since
H
n,
H
in L
2
,
lim
n
_
R
y
n
H
n,
d =
_
R
yH
d,
and (5.15) follows from (5.16).
Гораздо больше, чем просто документы.
Откройте для себя все, что может предложить Scribd, включая книги и аудиокниги от крупных издательств.
Отменить можно в любой момент.