You are on page 1of 29

J.

Differential Equations 249 (2010) 30523080


Contents lists available at ScienceDirect
Journal of Differential Equations
www.elsevier.com/locate/jde
New approach to the incompressible MaxwellBoussinesq
approximation: Existence, uniqueness and shape sensitivity
L. Consiglieri
a,1
, . Ne casov
b,2
, J. Sokolowski
c,d,,2
a
Mathematics Dep/FCUL and CMAF, University of Lisbon, 1749-016 Lisboa, Portugal
b
Mathematical Institute of the Academy of Sciences of the Czech Republic, itn 25, 115 67 Praha 1, Czech Republic
c
Institut lie Cartan, UMR 7502 Nancy-Universit-CNRS-INRIA, Laboratoire de Mathmatiques, Universit Henri Poincar Nancy 1,
B.P. 239, 54506 Vandoeuvre Ls Nancy Cedex, France
d
Systems Research Institute of the Polish Academy of Sciences, ul. Newelska 6, 01-447 Warszawa, Poland
a r t i c l e i n f o a b s t r a c t
Article history:
Received 19 November 2008
Revised 18 September 2010
Available online 8 October 2010
Keywords:
Magnetohydrodynamic ows
Existence
Uniqueness
Shape sensitivity
The Boussinesq approximation to the FourierNavierStokes
(FNS) ows under the electromagnetic eld is considered. Such
a model is the so-called MaxwellBoussinesq approximation. We pro-
pose a new approach to the problem. We prove the existence and
uniqueness of weak solutions to the variational formulation of the
model. Some further regularity in W
1,2+
, > 0, is obtained for
the weak solutions. The shape sensitivity analysis by the bound-
ary variations technique is performed for the weak solutions. As
a result, the existence of the strong material derivatives for the
weak solutions of the problem is shown. The result can be used
to establish the shape differentiability for a broad class of shape
functionals for the models of FourierNavierStokes ows under
the electromagnetic eld.
2010 Elsevier Inc. All rights reserved.
1. Introduction
The magnetohydrodynamic ows have been studied in [1,18,19] for differential operators with
constant coecients. We would like also to mention the work of Duvaut and Lions [8], Sermange and
*
Corresponding author at: Institut lie Cartan, UMR 7502 Nancy-Universit-CNRS-INRIA, Laboratoire de Mathmatiques,
Universit Henri Poincar Nancy 1, B.P. 239, 54506 Vandoeuvre Ls Nancy Cedex, France.
E-mail address: Jan.Sokolowski@iecn.u-nancy.fr (J. Sokolowski).
1
The work of L.C. was supported by the Ne cas Center for Mathematical Modeling LC06052 nanced by MSMT.
2
The works of .N. and J.S. were partially supported by the project between Czech Academy and CNRS. The work of .N.
was supported by the Grant Agency of the Academy of Sciences No. IAA100190804, further her research was supported by the
Academy of Sciences of the Czech Republic, Institutional Research Plan No. AV0Z10190503.
0022-0396/$ see front matter 2010 Elsevier Inc. All rights reserved.
doi:10.1016/j.jde.2010.09.029
L. Consiglieri et al. / J. Differential Equations 249 (2010) 30523080 3053
Temam [25], Gerbeau and Le Bris [14,15] on the existence of solutions. However, the temperature
dependent parameters seem to be more realistic in diverse situations. In what follows we deal with
the existence and uniqueness of weak solutions to the thermoelectromagnetic ow (FourierMaxwell
NavierStokes) model with viscosity, as well as with electric and thermal conductivities and magnetic
permeability dependent not only on the temperature but also on the space variables. The present
model is provided for all magnetic elds H H
1
() such that H = 0 in and H n = 0 on .
A different approach of the complete FNS system coupled with Maxwell equations can be found
in [4], where the shape sensitivity is analyzed under the assumption that the velocity eld of the
perturbation of the domain is divergence free.
The shape derivatives of shape functionals governed by NavierStokes ow are investigated by
Boisgrault and Zolsio [2] in the framework of the speed method. The expression for the Navier
Stokes equation transported to the xed domain is given provided that the domain deformation
for the purposes of shape sensitivity analysis is induced by a smooth vector eld. Under regular-
ity assumptions on the initial data and the forcing terms, and under a uniqueness assumption for
the solution of a linearized version of the NavierStokes equation, the implicit function theorem is
used to prove a regularity result for the solutions the NavierStokes equation transported to the xed
domain. The latter result is extended to the case of nonhomogeneous boundary conditions and shape-
dependent forces in [3]. The shape sensitivity analysis of the NavierStokes system is investigated by
Dziri, Moubachir and Zolsio [11] with respect to the dynamical deformations of the uid domain
boundary. The shape differentiability for the heat-conduction equation is considered in a work of Dziri
and Zolsio [9]. The existence of shapes for the NavierStokes ow with heat convection is studied by
Dziri and Zolsio [10]. The shape differentiability of solutions for compressible perturbations of Stokes
problem is performed in [24] in the framework of local theory. The shape gradients are recovered by
means of singular limits of the expressions obtained for the material derivatives, cf. [23].
In the present paper, the coecients are assumed to be bounded and continuous, and there are
no restrictions imposed on the derivatives of coecients. The presented here model can also include
the parameter dependent magnetic eld if some boundedness and continuity properties of the model
coecients are still available. However, for the simplicity of the presentation we avoid such depen-
dence.
The uniqueness of solutions is established under the smallness assumption on the data, related to
the coecients characterizing the viscous stress (the kinematic viscosity), the induced electric current
(the electric conductivity) and the heat ux (the thermal conductivity). In the present setting some
assumptions in the form of Lipschitz-type conditions on the parameters are also required. The shape
analysis concerns the system with nonconstant coecients as functions only on the temperature then
the nal elliptic system characterizing the strong material derivatives is provided if the additional
C
1
-assumption on the coecients is taken into account.
The outline of this work is as follows. In Section 2 the thermoelectromagnetoow is formulated as
a MaxwellBoussinesq approximation. In Section 3 we state the functional framework and the main
existence, regularity and uniqueness results. Section 4 is devoted to some auxiliary existence results
that will be needed in the proofs to the main theorems (Sections 5 and 6). In Section 7 we state
some useful results in order to analyze the shape sensitivity. We deal with the shape sensitivity of
the problem in Section 8. Applying the speed method introduced in the monograph by Sokolowski
and Zolsio the problem of nonconservation of solenoidality appears. To overcome this diculty we
can use the test functions, which are not divergence free and to handle the estimate for pressure we
apply the Bogovskii operator. We have to show that Bogovskii operator is independent of the shape
perturbations parameter which governs small variations of the boundary . Another possibility is
to use the domain transformations built by the divergence free vector elds V of the speed method,
or to apply the Piola transform to recover the solenoidal vector elds.
2. Thermoelectromagnetoow model
Let be an open bounded subset of R
3
with the boundary of class C
1,1
which is split into
two parts =
D

N
, where
D
is an open nonempty subset of and
N
= \
D
. The
electromagnetic eld is described by the Maxwell equations:
3054 L. Consiglieri et al. / J. Differential Equations 249 (2010) 30523080
B
t
+ E = 0; (1)
B = 0; (2)
D
t
+J = H; (3)
D =, (4)
where E and H are the electric and magnetic intensity elds, D and B are the electric and magnetic
induction elds and J is the current density. The constitutive laws are
D =E, B =H,
where is the permittivity and is the magnetic permeability. The Ohm law reads
J =J
0
+(E +u B),
where J
0
denotes a given applied current, is the electric conductivity and u is the uid velocity
vector.
It is known that:
by (2) there is A such that B = A;
introducing A in (1), we obtain (
A
t
+E) = 0;
hence, there exists the potential , such that
A
t
+E = .
For the steady-state case, the latter equation is reduced to the well-known relation E = .
The following coupled system of equations derived from the motion and energy equations models
the steady-state motion of FNS uids in :

_
(T )Du
_
+(u )u +p =(T ) rot HH+f G(T )T ; (5)
divu =
3

i=1
u
i
x
i
= 0; (6)

_
k(T )T
_
+u T = f ; (7)
where T is the temperature, Du = (D
i j
) = (
i
u
j
+
j
u
i
)/2 (i, j = 1, 2, 3) is the symmetrized gradient
of the velocity, is the viscosity, p denotes the pressure, k is the thermal conductivity, f and f denote
the external forces and heat sources, respectively. The buoyancy force as in the Boussinesq approxi-
mation is described by G(T ) = (T )(0, 0, g)

, where denotes the coecient of thermal dilatation


and g is the constant of gravity. The mass density is assumed to be constant, we set = 1. The ex-
istence of two body forces in the uid, the Lorentz force J B = ( H) (H) and the buoyancy
force, results from the presence of the magnetic eld.
The thermoelectromagnetoow problem under study has the following boundary conditions
u =g on , (8)
T = 0 on
D
, k(T )
T
n
+T =h on
N
, (9)
H n = 0, E n =0 on , (10)
L. Consiglieri et al. / J. Differential Equations 249 (2010) 30523080 3055
where represents the convective heat transfer coecient. To simplify the presentation it is assumed
that g =0. Indeed, under a convenient choice of a lift the general case is recovered.
For the steady-state, taking the divergence of Eq. (3) leads to

_
(T )E
_
=
_
J
0
+(T )(T )u H
_
.
Let us consider the electric eld E = and the mixed boundary value problem for the potential
of the form

_
(T )
_
=
_
J
0
+(T )(T )u H
_
in , (11)
= 0 on
D
and

n
= 0 on
N
. (12)
3. Assumptions and main existence results
We need some assumptions on the model, which are listed below.
Let us assume that
(H1) , , , k : RR are Caratheodory functions such that

#
,
#
>0 :
#
(, )
#
, a.e. in , R; (13)

#
,
#
>0 :
#
(, )
#
, a.e. in , R; (14)

#
,
#
>0 :
#
(, )
#
, a.e. in , R; (15)
k
#
, k
#
>0 : k
#
k(, ) k
#
, a.e. in , R; (16)
(H2) G = (0, 0,
T
g) where
T
is a real, continuous, and bounded function and g is a constant, we
denote G
#
= g
#
, where
#
denotes the upper bound for the function
T
;
(H3) L
2
+
(
N
) = { L
2
(
N
): 0};
(H4) And
f L
2
(), J
0
L
2
(), f L
2
() and h L
2
(
N
). (17)
(H5) In the variable domain setting (see Section 7), the function
T
is given by the restriction to

of a given H
1
-function dened in R
3
.
(H6) In the variable domain setting, the elements
f

L
2
(

), J

0
L
2
(

), f

L
2
(

) and h

L
2
_

N
_
,
stand for the data in boundary value problems in

, are simply given by restrictions to

of
some functions
f H
1
_
R
3
_
, J
0
H
1
_
R
3
_
, f H
1
_
R
3
_
and h H
1
_
R
3
_
, (18)
dened in all space. In this way the shape derivatives of all the data vanish, except for h, and
the material derivatives are just given by the scalar products of the gradients of the data with
respect to spatial variables with the velocity vector eld, e.g.,

f = f V , provided that all data
are given in the Sobolev spaces H
1
(R
3
).
3056 L. Consiglieri et al. / J. Differential Equations 249 (2010) 30523080
In the framework of function spaces of the Lebesgue and Sobolev type, the norms are denoted
by the symbols ,
1
,

N
in spaces L
2
(), H
1
(), L
2
(
N
), respectively, and there scalar and
vector function spaces are not distinguished in our notations. Providing that the meaning remains
clear, the canonical norm in L
p
() for p = 1, 2 is denoted by
p
. We introduce the Hilbert spaces
V =
_
v H
1
0
(): divv = 0 in
_
,
Z =
_
H
1
(): = 0 on
D
_
,
equipped with their standard scalar products. We recall that the norm
Z
is equivalent to the usual
seminorm and also to the norm
1
on space H
1
().
We state the main results of the paper considering a free magnetic eld.
Theorem3.1. Under the above assumptions (H1)(H4), the problem (5)(9), (11)(12) has a weak solution in
the following sense:
For every magnetic eld H H
1
() such that H = 0 in and H n = 0 on , there is the triplet
(u, , T ) V Z
2
which satises the following integral identities
_

(T )Du : Dvdx +
_

(v u) : udx
=
_

_
(T )( H) H+f G(T )T
_
vdx, v V; (19)
_

(T ) dx =
_

_
J
0
(T )(T )Hu
_
dx, Z; (20)
_

k(T )T dx +
_

u T dx +
_

N
T ds
=
_

f dx +
_

N
hds, Z. (21)
Theorem 3.2. Assume that f L
2+
1
() and H W
1,2+
1
(), where
1
> 2/5, then the weak solution
obtained by Theorem 3.1 enjoys the additional regularity, actually (u, , T ) W
1,2+
() W
1,2+
()
W
1,2+
() for some , , > 0. Furthermore, under the following Lipschitz-type continuity assumptions on
the temperature dependent function parameters of the model
>0 :

_
T
2
_

_
T
1
_

T
2
T
1

3/(2+)
, (22)
>0 :

_
T
2
_

_
T
1
_

T
2
T
1

, (23)
>0 :

_
T
2
_

_
T
1
_

T
2
T
1

, G = g, (24)
>0 :

_
T
2
_

_
T
1
_

T
2
T
1

3/(2+)
, (25)
k >0 :

k
_
T
2
_
k
_
T
1
_

T
2
T
1

3/(2+)
, T
2
, T
1
R, (26)
the weak solution is unique for small data.
L. Consiglieri et al. / J. Differential Equations 249 (2010) 30523080 3057
The existence of the pressure p in the space of distributions follows from the well-known results
by using the divergence-free test functions v C

0
() in (19). Moreover, the pressure is unique up to
a constant.
Finally, we recall the Tychonoff extension to weak topologies of the Schauder xed point theo-
rem [7, pp. 453456 and 470].
Theorem 3.3. Let K be a nonempty weakly sequentially compact convex subset of a locally convex linear
topological vector space V . Let L : K K be a weakly sequentially continuous operator. Then L has at least
one xed point.
4. Auxiliary results
In this section it is assumed that there are given elements w, with the following properties. The
vector function w L
4
() is the solenoidal function, i.e., divw = 0 in the sense of the distributions,
and w n = 0 on , furthermore L
1
().
In order to apply the xed point argument (cf. Theorem 3.3), let us introduce the nonlinear map-
ping
L : (w, ) (u, T ), (27)
where T and u will be solutions to the variational problems (28) and (30), respectively. Note that
the potential can be determined provided that the pair (u, T ) is given. The existence of solutions
for the following elliptic boundary value problems is the straightforward application of the classical
existence theory, hence their proofs are omitted here.
Proposition 4.1. Assume that conditions (16), (H3) and (17) are fullled. Then there exists a unique T Z
such that
_

k()T dx +
_

w T dx +
_

N
T ds
=
_

f dx +
_

N
hds, Z. (28)
Moreover, the energy estimate holds
k
#
T
1
f +h

N
. (29)
Proposition 4.2. Assume that conditions (13), (H2) and (17) are fullled. Then there exists a unique u V
such that
_

()Du : Dvdx +
_

(v w) : udx
=
_

_
()( H) H+f G(T )T
_
vdx, v V. (30)
Moreover, the energy estimate holds

#
u
1

#
HH
L
3 +f + G
#
T
L
6/5 . (31)
3058 L. Consiglieri et al. / J. Differential Equations 249 (2010) 30523080
Proposition 4.3. Assume that conditions (14)(15) and (17) are fullled. Then there exists a unique Z
satisfying (20). Moreover, the energy estimate holds

1
J
0
+
#

#
Hu. (32)
We point out that in order to prove Theorem 3.1 by the xed point argument, it suces to estab-
lish the continuity result for the mapping L.
Lemma 4.4 (Weak continuity of L). Let {(w
m
,
m
)} be a sequence such that w
m
w in L
4
() and
m

in L
1
(). Let {(u
m
, T
m
)} be the corresponding sequence of weak solutions given by Propositions 4.1 and 4.2.
Then there exists the weak solution (u, T ) for the limit functions (w, ) such that
u
m
u in V; T
m
T in Z.
Proof. Take w
m
w in L
4
() and
m
in L
1
() and denote by (u
m
, T
m
) the corresponding
solutions to the integral identities (30) and (28), for m N. From the estimates (31) and (29), the
sequence (u
m
, T
m
) is bounded in V Z, which implies the existence of a limit, (u, T ) V Z such
that the weak convergence holds, possibly for a subsequence, still denoted by (u
m
, T
m
). In order to
show that the limit (u, T ) is a solution of the required problem (30) and (28), we pass to the limit
as m + in the integral identities (30) and (28), simply replacing w, , u and T by the sequences
w
m
,
m
, u
m
and T
m
, respectively. Indeed, the passage to the limit can be justied due to the continuity
properties of the Nemytskii operators in the coecients combined with the standard arguments. 2
5. Proof of Theorem 3.1
Let L be the mapping dened by (27). In view of Propositions 4.1 and 4.2, the associated operator
L is well dened. Lemma 4.4 yields its weak continuity, since we have the compact embeddings
V
_
w L
4
(): w= 0 in , w n = 0 on
_
,
Z L
1
().
Finally, L maps the ball K = {(w, ) V Z: w
1
+
1
R} into itself, and taking into account
(29) and (31) we set
R :=
1

#
_

#
HH
L
3 +f
_
+
_
G
#

#
+1
_
1
k
#
_
f +h

N
_
.
Therefore, Theorem 3.3 can be applied to the mapping L and in this way the existence of a weak
solution (u, T ) is established. Then, the complete solution is obtained in view of Proposition 4.3.
6. Proof of Theorem 3.2
For weak solutions, the property (u, , T ) W
1,2+
() W
1,2+
() W
1,2+
(), for some , ,
>0, is a consequence of the following regularity results.
Proposition 6.1. Assume that f L
2+
1
() and H W
1,2+
1
(), for some
1
> 2/5, then there exists a
constant >0 such that the weak solution u V of (19) belongs to W
1,2+
(), i.e.
u
2+
K
1
,
with a constant K
1
>0 only dependent on the data.
L. Consiglieri et al. / J. Differential Equations 249 (2010) 30523080 3059
Proof. For every x
0
, 0 < r < R small enough, (x
0
, R) := B(x
0
, R), ]0, 1[ and some
positive constants B
1
, B
2
, independent of u and T , we have the following reverse estimate (cf. [5,
Lemma 3.2])
_ _
(x
0
,r)
|u|
2
dx
_
1/2

_ _
(x
0
,R)
|u|
2
dx
_
1/2
+
B
1
R r
_ _
(x
0
,R)
|u|
6/5
dx
_
5/6
+
B
2
R r
_ _
(x
0
,R)
_
|u u|
2
+|F|
2
+1
_
dx
_
1/2
.
By the Sobolev embedding W
1,2+
1
() L
3(2+
1
)/(1
1
)
() it follows that rot H H
L
3(2+
1
)/(4
1
)
(). In view of 3(2 +
1
)/(4
1
) = 2 + , where > 0 since
1
> 2/5, we get
F = f + (T ) rot H H G(T )T L
2+
(). Since u u L
3
() then the Gehring inequality [13]
guarantees the higher integrability u W
1,2+
() for some 0 < < <1. 2
For the regularity of the potential and the temperature T it will be sucient to state the follow-
ing simplied version of the general result on the higher regularity for weak solutions to the mixed
boundary value problems (cf. [16, Theorem 1]).
Theorem 6.2. Let be a bounded domain of class C
1
and u Z be a solution to the second-order elliptic
differential equation Au = F Z

, and satisfy natural boundary conditions on \


D
. Set A : Z Z

the
operator such that Au = (a(, u)u) where a : R R is a Caratheodory function such that there
exist constants a
#
>a
#
>0
a
#
a(, ) a
#
, a.e. in , R. (33)
Then A maps Z
p
= {v W
1,p
(): v = 0 on
D
} onto Z

p
for some p >2.
Thus we can prove the following two results.
Proposition 6.3. If J
0
L
2
() then there exists a constant > 0 such that the weak solution Z of (20)
belongs to W
1,2+
(), i.e.

2+
K
2
,
with a constant K
2
>0 only dependent on the data.
Proof. In order to apply Theorem 6.2, let u = be a weak solution verifying (20) and denote the
operator A by
A, =
_

(T ) dx.
The boundedness property (33) is fullled, considering that the assumption on (15) holds under
T L
1
(). Since
3060 L. Consiglieri et al. / J. Differential Equations 249 (2010) 30523080
F =
_
J
0
(T )(T )Hu
_
Z

p
, p 2,
then Theorem 6.2 guarantees that W
1,2+
() for some >0. 2
Proposition 6.4. If f L
2
() and h L
2
(
N
) then there exists a constant >0 such that the weak solution
T Z of (21) belongs to W
1,2+
(), i.e.
T
2+
K
3
,
with a constant K
3
>0 only dependent on the data.
Proof. In order to apply Theorem 6.2, let u = T be a weak solution verifying (21) and denote the
operator A by
AT , =
_

k(T )T dx.
The boundedness property (33) is fullled, considering that the assumption on k (16) holds. Since
u V, T H
1
() and (H3) holds, we have f u T , h T Z

p
, for all p 2, then Theo-
rem 6.2 guarantees that T W
1,2+
() for some >0. 2
To show the uniqueness let us assume that (u
1
,
1
, T
1
) and (u
2
,
2
, T
2
) are two weak solutions
to problem (19)(21). Thus, the differences u =u
1
u
2
, =
1

2
and T = T
1
T
2
satisfy
_

_
T
1
_
|Du|
2
dx =
_

_
T
2
_

_
T
1
__
Du
2
: Dudx
_

(u u) : u
2
dx
+
_

_
T
1
_

_
T
2
__
( H) H udx
_

_
G
_
T
1
_
T
1
G
_
T
2
_
T
2
_
udx;
_

_
T
1
_
||
2
dx =
_

_
T
2
_

_
T
1
__

2
dx
+
_

_
T
2
_

_
T
2
_
Hu
2

_
T
1
_

_
T
1
_
Hu
1
_
dx;
_

k
_
T
1
_
|T |
2
dx +
_

N
|T |
2
ds =
_

_
k
_
T
2
_
k
_
T
1
__
T
2
T dx

u T
2
T dx.
Using the assumptions (H3), (13)(16), and applying Hlder and Young inequalities leads to

#
2
Du
2

#
_
_
_

_
T
2
_

_
T
1
__
Du
2
_
_
2
+
_
_
|u|
2
_
_
_
_
u
2
_
_
+
C
1

#
__
_
_

_
T
1
_

_
T
2
__
( H) H
_
_
6/5
+
_
_
G
_
T
1
_
T
1
G
_
T
2
_
T
2
_
_
6/5
_
2
,
L. Consiglieri et al. / J. Differential Equations 249 (2010) 30523080 3061

#
2

2

#
_
_
_

_
T
2
_

_
T
1
__

2
_
_
2
+
C
1

#
_
_

_
T
2
_

_
T
2
_
Hu
2

_
T
1
_

_
T
1
_
Hu
1
_
_
2
,
k
#
2
T
2

1
k
#
_
_
_
k
_
T
2
_
k
_
T
1
__
T
2
_
_
2
+
C
1
k
#
_
_
u T
2
_
_
2
6/5
,
where C
1
is the Sobolev constant of the embedding H
1
() L
6
(). Furthermore, using this time
the Lipschitz continuity assumptions (22)(26), and applying Hlder and Young inequalities result in

#
2
Du
2

#
T
6/(2+)
6
_
_
Du
2
_
_
2
2+
+ C
2
2
Du
2
_
_
u
2
_
_
+
C
1

#
_
T
6
_
_
( H) H
_
_
3/2
+ G
#
T
6/5
+ GT
6
_
_
T
2
_
_
3/2
_
2
; (34)

#
2

2

#
T
6/(2+)
6
_
_

2
_
_
2
2+
+
C
1

#
_

#
H
4
u
4
+
#
T
3/(2+)
6
H
2(2+)/(4)
_
_
u
2
_
_
6
+
#
T
6
H
6
_
_
u
2
_
_
6
_
2
; (35)
k
#
2
T
2

k
k
#
T
6/(2+)
6
_
_
T
2
_
_
2
2+
+
C
1
k
#
u
2
6
_
_
T
2
_
_
2
3/2
, (36)
where C
2
is the Sobolev constant of the embedding H
1
() L
4
(). In the sequel C stands for
different Sobolev constants. Let K
1
, K
2
and K
3
be the upper bounds derived in Propositions 6.1, 6.3
and 6.4, respectively, and K
4
and K
5
be the upper bounds in estimates (29) and (31), namely,
K
5
=
1

#
_

#
HH
1
+f + G
#
K
4
_
, K
4
=
1
k
#
_
f +h

N
_
.
Now, summing up (34)(36) we get
_

#
2
C
2
K
5

C
#

#
H
2
1

C
k
#
K
2
4
_
Du
2
+

#
2

2
+
_
k
#
2

C

#
K
2
1

C

#
_

_
_
( H) H
_
_
3/2
+ G
#
+ GK
4
_
2

#
_
K
2
2
+
_

#
+
#
_
H
2
1
K
2
5
_

Ck
k
#
K
2
3
_
T
2
0,
and the uniqueness holds for small data. Note that the uniqueness of the potential Z is obtained
without any smallness assumption on the data.
7. Shape sensitivity preliminaries
We apply the framework established in [26] to the shape sensitivity analysis. A slightly different
approach to the stationary compressible ows is proposed in [22,24] (see also [6,20,21]). The main
outcome from the approach is the direct evaluation of the shape gradients by means of the appro-
priate singular limits of volume integrals. It seems that the approach is the only possibility to derive
the shape derivatives of solutions and of the functionals for the nonlinear boundary value problems
in uid and gas ows.
3062 L. Consiglieri et al. / J. Differential Equations 249 (2010) 30523080
To investigate the sensitivity of solutions to the Boussinesq approximation (5)(9) under the free
magnetic eld problem (11)(12) with respect to perturbations of the shape we use the speed method
developed, among others, by Sokolowski and Zolsio, all details can be found in [26]. We derive
the strong material derivatives for weak solutions of the model under study. Such shape sensitivity
results can be used, in particular, to obtain the form of the shape gradients for a broad class of shape
functionals governed by the MaxwellNavierStokes model investigated in the paper.
The speed method is briey described with our applications in mind. First, a family of mappings
T

: R
3
R
3
associated with a given velocity eld V (, x) is constructed. We dene the family of
perturbations of a given initial conguration by

= T

(), each specic family parametrized


by is dened in the direction of a given vector eld V . The evolution of geometrical domains, if the
vector eld V is chosen, is governed by the real parameter , and the so-called variable domain

=
T

() depends on two parameters, a vector eld V and the real variable , therefore, the variable
has the meaning of the time in our setting. The eld V is compactly supported with respect to the
spatial variable x, i.e.,
V C
_

1
,
1
; D
2
_
; R
3
__
, supp V ,
for some positive constant
1
. The mapping is given by the system of differential equations
d
d
x() = V
_
, x()
_
, x(0) = X, (37)
with the solution denoted by x() = x(, X), (
1
,
1
), X R
3
. Then the variable domains are
dened by the images of the mapping, and denoted by

= {x R
3
| x = x(, X), X }.
In order to measure how the weak solutions depends on the geometrical domain, it is convenient
to dene our model in the variable domain setting. However, the direct shape sensitivity analysis in
the variable domain setting, which results in the so-called shape derivatives, is usually dicult to
justify and requires some additional regularity of the weak solutions to the boundary value prob-
lem under study. Therefore, the shape differentiability is shown in the xed domain setting, the
domain perturbations are transformed into the variable coecients of the integral variational formu-
lations of the equations in question. In our setting all equations dened in variable domain

can
be transported to the reference domain which is also called the xed domain , using the inverse
transformation T
1

.
7.1. Properties of the mapping T

We consider the general case of the domain transformations T

, for R. Let D be a domain


in R
N
with the boundary D piecewise C
k
for a given integer k 0. Let T

be a one-to-one mapping
from D onto T

(D) such that


(A1) T

and T
1

belong to C
k
(D) = C
k
(D; R
N
) and C
k
(T

(D)), respectively;
(A2) T

(X), T
1

(x) C(
1
,
1
), X D, x T

(D).
Thus (, X) T

(X) C(
1
,
1
; C
k
(D)). For any X D, the point x() =T

(X) moves along the


trajectory x() with the velocity
d
d
x() =

(X). (38)
It is obvious that V (, x) takes the form
V (, x) =
_

_
T
1

(x). (39)
L. Consiglieri et al. / J. Differential Equations 249 (2010) 30523080 3063
Theorem 7.1 below fully characterizes the families of domains transformations and by the Nagumo
Theorem the transformations leave invariant, by construction, the hold-all-domain D. Such a construc-
tion serves for the investigations of shape differentiability of solutions to boundary value problems
as well as of the shape functionals. In particular, by the constructions, all admissible domains for a
specic problem of shape optimization are included in the suciently large hold-all-domain D.
By (38) and (39), the vector eld V () dened as V ()(x) = V (, x) satises the relation
V C
_

1
,
1
; C
k
_
D; R
N
__
. (40)
If V is a vector eld such that (40) holds, then the transformation T

, depending on V , and such that


conditions (38), (39) are satised, is dened by (37).
Theorem 7.1. Let D be a bounded domain in R
N
with the piecewise smooth boundary D, and V
C(
1
,
1
; C
k
(D; R
N
)) be a given vector eld which satises
V (, x) n(x) = 0 for a.e. x D, (41)
and we set
V (, x) = 0 if n =n(x) is not dened as a singular point x D. (42)
Then there exists an interval I (
1
,
1
), 0 I , and the one-to-one transformation T

(V ) : R
N
R
N
such
that T

(V ) maps D onto T

(V )(D). Furthermore T

(V ) satises the conditions (38), (39). In particular the


vector eld V can be written in the form
V =

(V ) T

(V )
1
. (43)
On the other hand, if T

= T

(V ) is a transformation of D, T

satises (38), (39) and



V is dened by the
formula

V =

T
1

, (44)
then (41) and (42) hold for

V . Furthermore,

V C(
1
,
1
; C
k
(D; R
N
)) and the transformation T

(X) =
x(, X) is dened as the local solution to the system of ordinary differential equations (37), that is, T

=
T

(V ).
Remark 7.2. There exists an interval I = (, ), 0 <
1
and a one-to-one transformation T

(V )
for each I which satises all properties of Theorem 7.1.
For the strong differentiability of composed mapping, we refer the reader to [26]. We recall here
some results required in our framework of boundary variations technique for the shape sensitivity
analysis. If D is the hold-all-domain, it means that by the Nagumo Theorem the boundary D is
invariant under the ow of admissible vector elds which satisfy conditions (41), (42). Given function
dened on D, or on R
3
, we can obtain the explicit form of the material and shape derivatives of the
function in the direction of an admissible vector eld V .
Proposition 7.3. Let f H
1
(D), V C(
1
,
1
; D
k
(R
N
+
; R
N
)), k 1 is a given vector eld such that condi-
tions (41), (42) are satised, then the mapping f T

is strongly differentiable in the space L


2
(D), and
its derivative is f V .
3064 L. Consiglieri et al. / J. Differential Equations 249 (2010) 30523080
For the proof see [26]. We recall here, that the above result is in fact equivalent to the shape
differentiability of the integral shape functional J (

) =
_

f (x) dx, where

D for (, ).
It is clear, that the results established in our paper lead to the shape differentiability for a class of
shape functionals, however this subject is not developed here, we restrict ourselves to the strong
shape differentiability of solutions to the boundary value problems.
Very important is also the property of a weak differentiability of T

with respect to . We address


the question, if for a given domain D R
N
and a given function f in L
2
the composed mapping
f T

is strongly differentiable in H
1
(D). The answer is negative, for a counterexample see [26],
however the differentiability in the weak topology of the space H
1
(D) can be shown.
Proposition 7.4. Let f L
2
(D), V C(
1
,
1
; D
k
(R
N
+
; R
N
)) be given, k 1, then the mapping f T

is weakly differentiable in the space H


1
(D).
For the proof see [26].
Proposition 7.5. Let f L
2
(R
N
), V C(
1
,
1
; D
k
(R
N
; R
N
)) be given, then the mapping f T

is
strongly differentiable in the space H
2
(R
N
).
For the proof see [26].
7.2. Bogovskii operator
We need some elementary properties of the operator which denes, in an appropriate way, the in-
verse of the div differential operator, the so-called Bogovskii operator. Consider an auxiliary problem:
Given
g L
q
(

),
_

g dx

= 0, 1 <q <, (45)


nd a vector eld v =B

[g] such that


v W
1,q
0
(

), divv = g a.a. in

and |v|
1,q
cg
q
, (46)
where c = c(N, q, ).
We recall the following results and denition, for more details see Chapters II and III.3 in
Galdi [12].
Denition 7.6. The domain is called star-shaped domain if there exist a point x and a continu-
ous positive function h on the unit sphere such that
=
_
x R
N
: |x x| <h
_
x x
|x x|
__
.
Proposition 7.7. Let be locally Lipschitzian. Then there exist m locally Lipschitzian domains G
1
, . . . , G
m
such that
(i)

m
i=1
G
i
;
(ii) the domains
i
= G
i
, i = 1, . . . , m are (locally Lipschitzian and) star-shaped with respect to every
point of a ball B
i
with B
i

i
.
L. Consiglieri et al. / J. Differential Equations 249 (2010) 30523080 3065
Proposition 7.8. Let R
N
, N 2, be such that
=
M
_
k=1

k
, M 1, (47)
where each
k
is star-shaped domain with respect to some open ball B
k
with B
k

k
, and let g L
q
()
satisfy (45). Then, there exist M functions g
k
such that for all k = 1, . . . , M:
(i) g
k
L
q
();
(ii) supp(g
k
)
k
;
(iii)
_

k
g
k
= 0;
(iv) g
k

q
Cg
q
, with
C =
_
1 +
|
k
|
|
k
D
k
|
_
k1

i=1
_
1 +|F
i
|
1/q1
|D
i

i
|
11/q
_
, (48)
and where D
i
=

M
s=i+1

s
and F
i
= |
i
D
i
|, i = 1, . . . , M 1.
Theorem 7.9. Let be a bounded domain of R
N
, N 2, such that (47) holds where each
k
is star-shaped
with respect to some open ball B
k
with B
k

k
. For instance satises the cone condition. Then, given
f L
q
() verifying
_

f = 0, there exists at least one solution v to (46). Furthermore, the constant c entering
in inequality (46) admits the following estimate:
c c
0
C
_
()
R
0
_
N
_
1 +
()
R
0
_
, (49)
where R
0
is the smallest radius of the balls B
k
, c
0
= c
0
(N, q), C is given as (48), and () is the diameter
of ,
() = sup
x,y
|x y|.
Finally, if f is of compact support in so is v.
Proposition 7.10. For each > 0 small enough there is a solution operator B

associated to problem (45)


(46) such that
_
_
B

[g]
_
_
W
1,q
0
(

)
c(N, q, )g
L
q
(

)
. (50)
Moreover, the norm of B

is independent of .
Proof. It is known that a domain R
N
is C
k,1
if the boundary can be expressed as a graph of
a C
k,1
function
a : S R
N1
{0} R
with an appropriate open set S R
N
and k N. We assume that =

M
i=1

i
and we know that
on every
i
Bogovskii operator is valid. Now we transform from to

and we know that this


transformation satises that (, X) T

(X) C(
1
,
1
; C
k
(D)). It is known that if we introduce
3066 L. Consiglieri et al. / J. Differential Equations 249 (2010) 30523080
the transformation of R
N
into itself then constant c in inequality (46) does not change in case of
homothetic transformation or rotation (see Galdi [12, Chapter III, Lemma 3.3.]). In our case it can be
seen that in our more general situation we can choose supremum over in (

) to get independent
constant c on in the estimate for Bogovskii operator. 2
8. Shape sensitivity analysis
We consider the model (5)(9), (11)(12), under the assumption that all coecients , , k and
are continuous functions only on the temperature. We perform the shape sensitivity analysis of the
considered model (cf. Theorem 3.1).
8.1. Perturbed problem in variable domain
Denition 8.1. The following system of equations dened in

is called a perturbed problem to the


model (5)(9), (11)(12),

_

_
T

_
Du

_
+
_
u


_
u

+p

=
_
T

_
rot H

+f

G
_
T

_
T

; divu

= 0; (51)

_

_
T

_
=
_
J

0
+
_
T

_
T

_
u

_
; (52)

_
k
_
T

_
T

_
+u

= f

, (53)
along with the boundary conditions:
u

=0 on

; (54)
T

= 0 on

D
; (55)
k
_
T

_
T

=h

and

= 0 on

N
. (56)
We introduce the Hilbert space
Z

=
_
H
1
(

): = 0 on

D
_
equipped with its standard inner product.
Theorem 8.2. Under the assumptions (H1)(H6), the problem (51)(56) has a weak solution in the following
sense:
For every magnetic eld H

H
1
(

) such that H

= 0 in

and H

= 0 on

, there exists
(u

, T

, p

) H
1
0
(

) (Z

)
2
L
2
(

) satisfying, for all v

H
1
0
(

) and

,
_

_
T

_
Du

: Dv

dx

+
_

_
v

_
: u

dx

=
_

dx

+
_

_
T

__
H

_
H

+f

G
_
T

_
T

_
v

dx

, (57)
_

_
T

dx

=
_

_
J

0

_
T

_
T

_
H

dx

, (58)
L. Consiglieri et al. / J. Differential Equations 249 (2010) 30523080 3067
_

k
_
T

_
T

dx

+
_

dx

+
_

ds

=
_

dx

+
_

N
h

ds

. (59)
Proof. Only difference is in term with pressure. Indeed we can argue as in the proof of Theorem 3.1
and then apply the De Rham Theorem to obtain the existence of the pressure. To estimate the pressure
we use as a test function
v

=B

_
p

1
|

|
_

dx

_
and we get
_
_
p

_
_
2
c. 2
Theorem 8.3. Assume f

L
2+
1
(

) and H

W
1,2+
1
(

), for some
1
> 2/5, then the solution in ac-
cordance to Theorem 8.2 is for some , , > 0 such that (u

, T

) W
1,2+
(

) W
1,2+
(

)
W
1,2+
(

) and it is unique under small data on (24).


Proof. See the proof of Theorem 3.2. 2
8.2. Transported problem
The transported solution to the xed domain is denoted by u

= u

, H

= H

, T

= T

with data f

= f

, G

= G

, J
0
= J

0
T

, f

= f

T

and h

=h

.
Remark 8.4. The shape analysis for the Maxwell equations coupled with heat equation and an evo-
lution equation for the volume fraction of the high temperature phase in steel was investigated by
Hmberg and Sokolowski [17]. Here we deal with the more general situation: incompressible heat
conductive uid with Maxwell equations.
We recall the following important results.
Proposition 8.5. Denote by J T

the Jacobian of T

and for any matrix B the transposed matrix is denoted


by

B. Then we have
(i) (grad w) T

=(

J T
1

)(w T

) for all w H
1
();
(ii) (divw) T

=()
1
(() J T
1

) (w T

) for all w H
1
();
(iii) (curl w) T

=(

J T
1

) (w T

) for all w H
1
().
Remark 8.6. From Proposition 8.5, it follows that functions which are divergence free on

gen-
erally loose this property when they are transported to the xed domain. For more details see [17,
Remark 6.2].
3068 L. Consiglieri et al. / J. Differential Equations 249 (2010) 30523080
Proposition 8.7. (See [26, Proposition 2.47].) For any f L
1
(

),
_

f ds

=
_

f T

_
_
M(T

) n
_
_
R
3
ds,
where M(T

) = det( J T

J T
1

is the cofactor matrix of the Jacobian matrix J T

.
We introduce the following notations
() = det( J T

),
() =

J T
1

,
A() =()

()(),
B() =()(),
() =
_
_
M( J T

) n
_
_
R
3
,
and we dene the following forms
(F1)
0
(, u, T , v) =
_

(T )()(()Du) : (()Dv) dx =
_

(T )A() : (DuDv) dx,


(F2)
1
(, u, v) =
_

()(()u) : (v u) dx =
_

B()u : (v u) dx,
(F3)
2
(, H, T , v) =
_

(T )()(((()) H) H) vdx,
(F4)
3
(, f, T , v) =
_

()(f G(T )T ) vdx,


(F5)
4
(, p, v) =
_

()p(()) vdx =
_

pB() : vdx,
(F6)
1
(, T , , ) =
_

(T )(()) (()) dx =
_

(T )A() : ( ) dx,
(F7)
2
(, J
0
, T , H, u, ) =
_

B() : ((J
0
(T )(T )Hu) ) dx,
(F8)
1
(, T , ) =
_

k(T )A() : (T ) dx,


(F9)
2
(, u, T , ) =
_

()u (())T dx =
_

B() : (u T )dx,
(F10)
3
(, , T , ) =
_

N
T () ds,
(F11)
4
(, f , ) =
_

f () dx,
(F12)
5
(, h, ) =
_

N
h() ds.
In view of Propositions 8.5 and 8.7, the problem (57)(59) in variable domains can be transported
to the xed domain, and rewritten in the weak form as the following system of integral identities.
Denition 8.8. The following system of integral identities is called the transported problem to the
xed domain

0
(, u

, T

, v) +
1
(, u

, v) =
2
(, H

, T

, v) +
3
(, f

, T

, v) +
4
(, p

, v),

1
(, T

, ) =
2
(, J
0

, T

, H

, u

, ),

1
(, T

, ) +
2
(, u

, T

, ) +
3
(,

, T

, ) =
4
(, f

, ) +
5
(, h

, ),
for all v H
1
0
(), , Z. Here
L. Consiglieri et al. / J. Differential Equations 249 (2010) 30523080 3069
H

= 0,
_
divH

_
T

=
1
()
div
_
() J T
1

_
,
and

= T

= 0 on
D
and u

=0 on .
Remark 8.9. First, we prove the formula for the transport of the divergence operator (to be sure). We
have in the variable domain
_

divvdx

=
_

v dx

then, the transport to the xed domain leads to, denote = T

,
_

_
(divv) T

_
() dx =
_

[v T

]
_
() T

_
() dx
=
_

[v T

]
_

J T
1


_
() dx
which conrms the above formula.
For a given divergence free vector eld v

dened in the variable domain

it follows that the


transported divergence is given by the expression
_
divv

_
T

=
1
()
div
_
() J T
1

_
which shows the following equivalence for v

=v

divv

= 0 in if and only if div


_
() J T
1

_
= 0 in

.
Therefore, if we introduce the new function v

= () J T
1

dened in the xed domain, then it


follows that
divv

= div
_
() J T
1

_
=()
_
divv

_
T

= 0
for the divergence free eld v

. In this way we have v

=
1
() J T

and v

=
1
() J T

.
Therefore, to keep divergence free the unknown velocity eld we should replace in the transported
problem u

by u

=
1
() J T

, and we have divu

= 0 since divu

= 0.
Lemma 8.10. Under our assumptions on the domain transformations the functions (), A(), B() are
differentiable at = 0. Furthermore, for ||
1
and
1
small enough, the following Taylor expansions are
obtained
(i) () = 1 +

(0) +o(),
(ii) () = I +

(0) + O(),
(iii) A() = I + A

(0) + O(),
(iv) B() = I + B

(0) + O(),
3070 L. Consiglieri et al. / J. Differential Equations 249 (2010) 30523080
where the derivatives at = 0 are given by
(v)

(0) = div V (0),


(vi)

(0) =

J V (0),
(vii) A

(0) = div V (0)I 2D(V (0)),


(viii) B

(0) = div V (0)I

J V (0).
As in Section 2, D(V (0)) denotes the symmetrized part of J V (0), i.e. D(V (0)) =
1
2
( J V (0) +

J V (0)).
For the proof see Sokolowski and Zolsio [26, Section 2.13].
Theorem 8.11. Suppose that the assumptions (H1)(H6) are fullled. For every magnetic eld H

H
1
()
such that H

= 0 in and H

= 0 on , there exists a weak solution (u,

T ,p) H
1
0
() Z
2

L
2
() to the transported problem in accordance to Denition 8.8.
Proof. The existence of a solution is consequence of Theorem 3.1. Only difference is in term with
pressure. To estimate the pressure we use as a test function
v

=B
_
p


1
||
_

dx
_
and we get
p

2
c. 2
Theorem 8.12. Assume f

L
2+
1
() and H

W
1,2+
1
(), for some
1
> 2/5, then the solution in ac-
cordance to Theorem 8.11 belongs to W
1,2+
() W
1,2+
() W
1,2+
() for some , , > 0 and the
solution is unique under small data on (24), that is,
(u,

T ,p) =(u

, T

, p

).
Proof. See the proof of Theorem 3.2. 2
Consequence of Lemma 8.10 is the following corollary.
Corollary 8.13. Let ||
1
and
1
be small enough, then there exist realvalued functions g
i
satisfying g
i
() =
o(), i = 0, . . . , 11 and forms
i
(, . . .), i = 0, 1, 2, 3, 4 and

(, . . .), i = 1, 2, and (, . . .), i = 1, . . . , 5,
such that the following statements are valid.
(B1) For all u, v H
1
0
() and T L
1
(),

0
(, u, T , v) =
0
_
0, u, T , v
_
+
0,
_
0, u, T , v
_
+
0
_
, u, T , v
_
,

0,
_
0, u, T , v
_
=
_

(T )A

(0) : (DuDv) dx,



0
_
, u, T , v
_
g
0
()
#
u
1
v
1
.
L. Consiglieri et al. / J. Differential Equations 249 (2010) 30523080 3071
(B2) For all u, v H
1
0
(),

1
_
, u, v
_
=
1
_
0, u, v
_
+
1,
_
0, u, v
_
+
1
_
, u, v
_
,

1,
_
0, u, v
_
=
_

(0)u : (v u) dx,

1
_
, u, v
_
g
1
()u
2
1
v
1
.
(B3) For all H H
1
(), T L
1
() and v H
1
0
(),

2
_
, H, T , v
_
=
2
_
0, H, T , v
_
+
2,
_
0, H, T , v
_
+
2
_
, H, T , v
_
,

2,
_
0, H, T , v
_
=
_

(T )
_

(0)( H) H
+
__

(0)
_
H
_
H
_
vdx,

2
_
, H, T , v
_
g
2
()
#
HH
1
v
1
.
(B4) For all f L
2
(), T Z and v H
1
0
(),

3
(, f, T , v) =
3
(0, f, T , v) +
3,
(0, f, T , v) +
3
(, f, T , v),

3,
(0, f, T , v) =
_

(0)
_
f G(T )T
_
vdx,

3
(, f, T , v) g
3
()
_
f + G
#
T
_
v.
(B5) For all p L
2
() and v H
1
0
(),

4
(, p, v) =
4
(0, p, v) +
4,
(0, p, v) +
4
(, p, v),

4,
(0, p, v) =
_

pB

(0) : vdx,

4
(, p, v) g
4
()pv
1
.
(B6) For all T L
1
() and , Z,

1
(, T , , ) =
1
(0, T , , ) +
1,
(0, T , , ) +

1
(, T , , ),

1,
(0, T , , ) =
_

(T )A

(0) : ( ) dx,

1
(, T , , ) g
5
()
#

1
.
3072 L. Consiglieri et al. / J. Differential Equations 249 (2010) 30523080
(B7) For all J
0
L
2
(), T L
1
(), H H
1
(), u H
1
0
() and Z,

2
(, J
0
, T , H, u, ) =
2
(0, J
0
, T , H, u, ) +
2,
(0, J
0
, T , H, u, )
+

2
(, J
0
, T , H, u, ),

2,
(0, J
0
, T , H, u, ) =
_

(0) :
__
J
0
(T )(T )Hu
_

_
dx,

2
(, J
0
, T , H, u, ) g
6
()
_
J
0
+
#

#
H
1
u
1
_

1
.
(B8) For all T , Z,

1
(, T , ) =
1
(0, T , ) +
1,
(0, T , ) +
1
(, T , ),

1,
(0, T , ) =
_

k(T )A

(0) : (T ) dx,

1
(, T , ) g
7
()k
#
T
1

1
.
(B9) For all u H
1
0
() and T , Z,

2
(, u, T , ) =
2
(0, u, T , ) +
2,
(0, u, T , ) +
2
(, u, T , ),

2,
(0, u, T , ) =
_

(0) : (u T )dx,

2
(, u, T , ) g
8
()u
1
T
1

1
.
(B10) For all L
2
(
N
) and T , Z,

3
(, , T , ) =
3
(0, , T , ) +
3,
(0, , T , ) +
3
(, , T , ),

3,
(0, , T , ) =
_

N
T

(0) ds,

3
(, , T , ) g
9
()

N
T
1

1
.
(B11) For all f L
2
() and Z,

4
(, f , ) =
4
(0, f , ) +
4,
(0, f , ) +
4
(, f , ),

4,
(0, f , ) =
_

(0) dx,

4
(, f , ) g
10
() f .
(B12) For all h L
2
(
N
) and Z,

5
(, h, ) =
5
(0, h, ) +
5,
(0, h, ) +
5
(, h, ),

5,
(0, h, ) =
_

N
h

(0) ds,

5
(, h, ) g
11
()h

1
.
L. Consiglieri et al. / J. Differential Equations 249 (2010) 30523080 3073
Applying Taylor polynomials of degree one (cf. Lemma 8.10) we can prove the stability result.
Proposition 8.14. Under the assumptions of Theorem 8.12, if (u

, T

) is a transported solution which


converges to (u, , T ) with 0, and in addition the following assumptions on the continuity of the data are
fullled:
(M1) H

H
1
C||,
(M2) f

f C||,
(M3) G(T

)T

G(T )T
6/5
C||,
(M4) J
0
J
0
C||,
(M5)

N
C||,
(M6) h

N
C||,
(M7) f

f C||,
then we have
u

u
1
C||; (60)


1
C||; (61)
T

T
1
C||; (62)
p

p C||, (63)
where C stands for a generic constant.
Proof. For small enough and
i
[0, ], i = 0, . . . , 4, we can write
() = 1 +

(
0
),
() = I +

(
1
),
A() = I + A

(
2
),
B() = I + B

(
3
),
() = 1 +

(
4
),
where () c

1
>0 for ||
1
and A, B, are positive denite for ||
1
.
Observing that
4
is linear with respect to the second argument, we write

4
(, p

, v)
4
(0, p, v) =
4
(0, p

p, v) +
_

(
3
) : vdx. (64)
Observe that
i
(i = 0, 1, 2, 3) is no more linear with respect to its arguments, thus we write

0
(, u

, T

, v)
0
(0, u, T , v)
=
0
(0, u

, T

, v)
0
(0, u, T , v) +
_

(T

)A

(
2
) : (Du

Dv) dx; (65)


3074 L. Consiglieri et al. / J. Differential Equations 249 (2010) 30523080

1
(, u

, v)
1
(0, u, v)
=
1
(0, u

, v)
1
(0, u, v) +
_

(
3
)u

: (v u

) dx; (66)

2
(, H

, T

, v)
2
(0, H, T , v)
=
2
(0, H

, T

, v)
2
(0, H, T , v)
+
_

(T

)
_

(
0
)( H

) H

+
__

(
1
)
_
H

_
H

_
vdx; (67)

3
(, f

, T

, v)
3
(0, f, T , v)
=
3
(0, f

, T

, v)
3
(0, f, T , v) +
_

(
0
)
_
f

G(T

)T

_
vdx. (68)
Considering that u is the particular case ( = 0) to the perturbed u

it follows that
RHS of (65) +RHS of (66) = RHS of (67) +RHS of (68) +RHS of (64).
Next, we write
(N1)
1
(, T

, )
1
(0, T , , ) =
1
(0, T

, )
1
(0, T , , ) +
_

(T

)A

(
2
) : (


) dx;
(N2)
2
(, J
0
, T

, H

, u

, )
2
(0, J
0
, T , H, u, ) =
2
(0, J
0
, T

, H

, u

, )
2
(0, J
0
, T , H, u, ) +

(
3
) : ((J
0
(T

)(T

)H

) ) dx;
(N3)
1
(, T

, )
1
(0, T , ) =
1
(0, T

, )
1
(0, T , ) +
_

k(T

)A

(
2
) : (T

) dx;
(N4)
2
(, u

, T

, )
2
(0, u, T , ) =
2
(0, u

, T

, )
2
(0, u, T , ) +
_

(
3
) : (u

)dx;
(N5)
3
(,

, T

, )
3
(0, , T , ) =
3
(0,

, T

, )
3
(0, , T , ) +
_

(
4
) ds;
(N6)
4
(, f

, )
4
(0, f , ) =
4
(0, f

f , ) +
_

(
0
) dx;
(N7)
5
(, h

, )
5
(0, h, ) =
5
(0, h

h, ) +
_

N
h

(
4
) ds.
Considering that and T are the particular case ( = 0) to the perturbed

, and T

, respectively,
we can argue as in the proof of Theorem 3.2, setting v = u

u, =

and = T

T to get
the following estimate
_

#
C
2
K
5

C
#

#
H
2
1

C
k
#
K
2
4
_
u

u
2
1
+
#


2
1
+
_
k
#

C

#
K
2
1

C

#
_

_
_
( H) H
_
_
3/2
+ G
#
+ GK
4
_
2

#
_
K
2
2
+
_

#
+
#
_
H
2
1
K
2
5
_

Ck
k
#
K
2
3
_
T

T
2
1

#
_

#
_
_
( H

) H

( H) H
_
_
6/5
+f

f +p

p +||
_
2
+
C

#
_
J
0

J
0
+
#

#
H

H
1
+||
_
2
+
C
k
#
_

N
+ f

f +h

N
+||
_
2
.
L. Consiglieri et al. / J. Differential Equations 249 (2010) 30523080 3075
Under the smallness condition on the data, we argue as in Theorem 8.11 and from assumptions (M1)
(M7), and we get (60)(63). 2
8.3. Material derivative
Denition 8.15. The following limit in the function space norm H

f = lim
0
f () f (0)

is called the strong material derivative



f of f in H.
Denition 8.16. The shape derivative u

of u() in the direction of the vector eld V is dened by


the formula
u

= u u V
provided that there exists the material derivative u.
We recall that A(0) = B(0) =(0) = I , (0) =(0) = 1, =

(0),

A = A

(0),

B = B

(0),

=

(0)
and =

(0), and we state the following result on the existence of material derivatives.
Theorem 8.17. Under the assumptions

f L
2
(),

J L
2
(),

f L
2
(), ,

h L
2
(
N
), for every magnetic
eld

H H
1
() such that

H = 0 in ,

H n = 0 on and , , , k C
1
() the triple ( u,

,

T ) VZ
2
and it satises:
Momentum equations:
_

(T )(

ADu + D u) : Dvdx +
_

(T )

T Du : Dvdx
+
_

Bu + u) : (v u) dx +
_

u : (v u) dx
+
_

_
(

p + p) v + p( ) v
_
dx
=
_

(T )

_
( H) H
_
vdx +
_

(T )
__
( ) H
_
H
_
vdx
+
_

(T )
_
(

H) H+( H)

H
_
vdx
+
_

(T )

T ( H) H
_
vdx
+
_

f

G(T )T G(T )

T +
_
f G(T )T
_

_
vdx, v H
1
0
(); (69)
3076 L. Consiglieri et al. / J. Differential Equations 249 (2010) 30523080
Equation for the electric eld:
_

(T )(

A +

) dx +
_

(T )

T dx
=
_

B :
_
J
0
(T )(T )Hu
_
dx
+
_

J
0
(T )(T )

Hu (T )(T )H u
_
dx
+
_

__

(T )

T (T ) +(T )

(T )

T
_
Hu
_
dx, Z; (70)
Energy equation:
_

k(T )(

AT +

T ) dx +
_

(T )

T T dx
+
_

B : u T + u T +u

T )dx
+
_

N
(

T + T )ds +
_

N
T ds
=
_

f + f

)dx +
_

N
(

h +h )ds, Z; (71)
and the following estimate

T
1
+ u
1
+

1
C
__
1 +u
1
+

N
+

N
_
T
1
+
_
1 +u
1
_
u
1
+
1
+

f + f +

N
+h

N
+
_

H
1
+H
1
_
H
1
+

f +f +
_

G
#
+ G
#
_
T
1
+

Hu +H u +Hu +

J
0
+J
0

_
, (72)
where C denotes a positive constant depending on the upper and lower bounds of , , , k and its derivatives.
Proof. The estimate (72) we get as before, namely from (69), (70) and (71) using the uniqueness
argument of Theorem 3.2. Now, it remains to prove that they are material derivative.
The transported solution given by Theorem 8.12 veries, for all v H
1
0
() and , Z,
_

(T

)A() : (Du

Dv) +
_

B()u

: (v u

)
=
_

B() : Dv +()
_
(T

)
__
()
_
H

_
H

+f

G(T

)T

_
vdx; (73)
L. Consiglieri et al. / J. Differential Equations 249 (2010) 30523080 3077
_

(T

)A() : (

) dx
=
_

B() :
__
J
0

(T

)(T

)H

_
dx; (74)
_

k(T

)A() : (T

) dx +
_

B() : (u

)dx +
_

() ds
=
_

()dx +
_

N
h

() ds. (75)
The solution given by Theorem 3.2 veries (20)(21), and applying the De Rham Theorem, also
_

(T )Du : Dvdx +
_

Du : (u v) dx
=
_

p vdx +
_

_
(T )( H) H+f G(T )T
_
vdx, v H
1
0
(). (76)
Let
z

=
1

(u

u);

=
1

(T

T );

=
1

);

=
1

(p

p).
From Proposition 8.14 we nd the following results (as 0):
(i) z

u weakly in H
1
0
(), u

u strongly in H
1
0
();
(ii)



T weakly in Z, T

T strongly in Z;
(iii)



weakly in Z,

strongly in Z;
(iv)

p weakly in L
2
(), p

p strongly in L
2
().
Applying the SobolevKondrachov compact embedding to (i)(iii), the sequences z

and

are
strongly convergent in L
q
() for any 1 q < 6 and, in particular, it follows u

u, T

T and

strongly in L
6
(), L
6
() and L
6
(), respectively. Thus, considering the relation
1

_
(T

) (T )
_
=

(T

, for some T

between T

and T ,
applying (22) and the continuity of

we obtain
(v)
(T

)(T )

(T )

T weakly in L
2(2+)/
() and strongly in L
q(2+)/(3)
().
Recalling that Du L
2+
() it results from (v) that Du((T

) (T ))/

(T )

T Du weakly
in L
2
().
Analogously we have
(vi)
(T

)(T )

(T )

T weakly in L
6
() and strongly in L
q
();
(vii)
(T

)(T )

(T )

T weakly in L
2(2+)/
() and strongly in L
q(2+)/(3)
();
(viii)
k(T

)k(T )

(T )

T weakly in L
2(2+)/
() and strongly in L
q(2+)/(3)
().
3078 L. Consiglieri et al. / J. Differential Equations 249 (2010) 30523080
Subtracting (73), (74) and (75) by (76), (20) and (21), respectively, dividing by = 0 we get:
Momentum equations:
_

(T

)
_
1

_
A() I
_
: DuDv + A() : Dz

Dv
_
dx
+
_

_
(T

) (T )
_
Du : Dvdx
+
_

B()z

: v u

dx +
_

B()Du : v z

dx
+
_

_
B() I
_
u : (v u) dx
+
_

_
p

B() + p
_
1

_
B() I
_
__
: Dvdx
=
_

(T

)()
_
_
()
_

_
H

vdx
+
_

(T

)()
__
()
_
H

(H

H) vdx
+
_

(T

)()
___
1

_
() I
_

_
H

_
H

_
vdx
+
_

_
(T

) (T )
_
()
___
()
_
H
_
H
_
vdx
+
_

_
1

_
() I
_
_
_
(T )
__
()
_
H
_
H+f G(T )T
_
vdx
+
_

()
_
1

(f

f)
_
1

_
G(T

) G(T )
_
_
T

+
1

(T

T )G(T )
_
vdx,
v H
1
0
();
Electric equation:
_

(T

)A() : (

) dx +
_

(T

)
_
1

_
A() I
_
_
: ( ) dx
+
_

_
(T

) (T )
_
dx
=
_

B() :
_
1

(J
0

J
0
) (T

)(T

)
_
H

u +H

__
dx
L. Consiglieri et al. / J. Differential Equations 249 (2010) 30523080 3079
+
_

B() :
_
1

__
(T

) (T )
_
(T ) +(T )
_
(T

) (T )
__
Hu
_
dx
+
_

_
1

_
B() I
_
_
:
__
J
0
(T )(T )H

_
dx, Z;
Energy equation:
_

k(T

)A() : (

)dx +
_

k(T

)
_
1

_
A() I
_
_
: (T ) dx
+
_

_
k(T

) k(T )
_
T dx
+
_

__
1

_
B() I
_
_
u T + B()z

+ B()u

_
dx
+
_

() ds +
_

T
_
1

_
() I
_
_
ds +
_

N
_
1

)
_
T ds
=
_

_
1

( f

f )
_
() dx +
_

_
1

_
() I
_
_
dx
+
_

N
_
1

(h

h)
_
() ds +
_

N
h

_
1

_
() I
_
_
ds, Z.
Therefore passing to the limit in the momentum, electric and energy equations as tends to
zero, applying the convergences (i)(viii), Lemma 8.10 and Corollary 8.13 we conclude (69), (70)
and (71). 2
References
[1] G.V. Alekseev, Steady model of magnetohydrodynamics viscous uid with heat transfer, Uspekhi Mech. (2006) 66116
(in Russian).
[2] S. Boisgrault, J.P. Zolsio, Shape derivative of sharp functionals governed by NavierStokes ow, in: Partial Differential
Equations, Praha, 1998, in: Chapman & Hall/CRC Res. Notes Math., vol. 406, 2000, pp. 4963.
[3] S. Boisgrault, J.P. Zolsio, Boundary variations in the NavierStokes equations and Lagrangian functionals, in: Shape Op-
timization and Optimal Design, Cambridge, 1999, in: Lect. Notes Pure Appl. Math., vol. 216, Dekker, New York, 2001,
pp. 726.
[4] L. Consiglieri, rka Ne casov, Jan Sokolowski, Incompressible MaxwellBoussinesq approximation: Existence, uniqueness
and shape sensitivity, Control Cybernet. 38 (4A) (2009) 11931215.
[5] L. Consiglieri, T. Shilkin, Regularity to stationary weak solutions for generalized Newtonian uids with energy transfer, Zap.
Nauchn. Sem. POMI 271 (2000) 122150.
[6] M. Delfour, J.P. Zolsio, Shapes and Geometries: Analysis, Differential Calculus, and Optimisation, Adv. Des. Control, SIAM,
Philadelphia, 2001.
[7] N. Dunford, J.T. Schwartz, Linear Operators, Part I, Interscience Publ., 1958.
[8] G. Duvaut, J.L. Lions, Inquations en thermolasticit et magntohydro-dynamique, Arch. Ration. Mech. Anal. 46 (1972)
241279.
[9] R. Dziri, J.P. Zolsio, Shape derivative for the heat-conduction equation with Lipschitz continuous coecients, Boll. Unione
Mat. Ital. Sez. B 10 (3) (1996) 569594.
[10] R. Dziri, J.P. Zolsio, Shape existence in NavierStokes ow with heat convection, Ann. Sc. Norm. Super. Pisa 24 (1) (1997)
165192.
[11] R. Dziri, M. Moubachir, J.P. Zolsio, Dynamical shape gradient for the NavierStokes system, C. R. Math. 338 (2) (2004)
183186.
3080 L. Consiglieri et al. / J. Differential Equations 249 (2010) 30523080
[12] G.P. Galdi, An Introduction to the Mathematical Theory of the NavierStokes Equations, I, Springer-Verlag, New York, 1994.
[13] F.W. Gehring, The L
p
-integrability of the partial derivatives of a quasi conformal mappings, Acta Math. 130 (1973) 265277.
[14] J.F. Gerbeau, C. Le Bris, A coupled system arising in magnetohydrodynamics, Appl. Math. Lett. 12 (3) (1999) 5357.
[15] J.F. Gerbeau, C. Le Bris, Existence of solution for a density-dependent magnetohydrodynamic equation, Adv. Differential
Equations 2 (3) (1997) 427452.
[16] K. Groger, A W
1,p
-estimate for solutions to mixed boundary value problems for second order elliptic differential equations,
Math. Ann. 283 (1989) 679687.
[17] D. Hmberg, J. Sokolowski, Optimal shape design of inductor coils for surface hardening, SIAM J. Control Optim. 42 (3)
(2003) 10871117.
[18] A.J. Meir, Thermally coupled magnetohydrodynamics ow, Appl. Math. Comput. 65 (1994) 7994.
[19] A.J. Meir, P.G. Schmidt, On electromagnetically and thermally driven liquid-metal ows, Nonlinear Anal. 47 (2001) 3281
3294.
[20] F. Murat, J. Simon, Sur la Controle par un Domaine Geometrique, Publications du Laboratoire dAnalyse Numerique, Univer-
site de Paris VI, 1976.
[21] O. Pironneau, Optimal Shape Design for Elliptic Systems, Springer Ser. Comput. Phys., Springer-Verlag, New York, 1984.
[22] P.I. Plotnikov, J. Sokolowski, Domain dependence of solutions to compressible NavierStokes equations, SIAM J. Control
Optim. 45 (4) (2006) 11471539.
[23] P.I. Plotnikov, J. Sokolowski, Compressible NavierStokes equations, Discrete Contin. Dyn. Syst Suppl. (2009) 502511.
[24] P.I. Plotnikov, E.V. Ruban, J. Sokolowski, Inhomogeneous boundary value problems for compressible NavierStokes equa-
tions, well-posedness and sensitivity analysis, SIAM J. Math. Anal. 40 (3) (2008) 11521200.
[25] M. Sermange, R. Temam, Some mathematical questions related to MHD equations, Comm. Pure Appl. Math. XXXVI (1983)
635664.
[26] J. Sokolowski, J.P. Zolsio, Introduction to Shape Optimization, Springer-Verlag, 1992.