HAL Id: hal-01970455
https://hal.archives-ouvertes.fr/hal-01970455
Submitted on 5 Jan 2019
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
Boundary Feedback Stabilization of the Boussinesq system with mixed boundary conditions
Mythily Ramaswamy, Jean-Pierre Raymond, Arnab Roy
To cite this version:
Mythily Ramaswamy, Jean-Pierre Raymond, Arnab Roy. Boundary Feedback Stabilization of the
Boussinesq system with mixed boundary conditions. Journal of Differential Equations, Elsevier, In
press. �hal-01970455�
Boundary Feedback Stabilization of the
Boussinesq system with mixed boundary conditions
Mythily Ramaswamy
∗, Jean-Pierre Raymond
†, Arnab Roy
∗Abstract
We study the feedback stabilization of the Boussinesq system in a two dimensional do- main, with mixed boundary conditions. After ascertaining the precise loss of regularity of the solution in such models, we prove first Green’s formulas for functions belonging to weighted Sobolev spaces and then correctly define the underlying control system. This provides a rigorous mathematical framework for models studied in the engineering literature. We prove the stabilizability by extending to the linearized Boussinesq system a local Carleman esti- mate already established for the Oseen system. Then we determine a feedback control law able to stabilize the linearized system around the stationary solution, with any prescribed exponential decay rate, and able to stabilize locally the nonlinear system.
Key words. Boussinesq system, mixed boundary conditions, boundary Feedback Stabiliza- tion, finite dimensional controllers, fractional Sobolev spaces.
AMS subject classifications. 93B52, 93C20, 93D15, 76D05, 76D55, 26B20.
1 Introduction
In the recent years, there has been a considerable interest in feedback stabilization of fluid flows around stationary solutions. However most of the works focus on either Dirichlet or Neumann boundary conditions on the entire boundary. Mixed boundary conditions do arise naturally in applications, for example, in energy efficient building structures.
In this paper, we would like to establish rigorous stabilization results for the Boussinesq system, around a stationary solution, by feedback controls of finite dimension, for geometrical domains and boundary conditions including models studied numerically in [1]. Stabilization problems for the Navier–Stokes system with mixed boundary conditions are already studied in [2]. Those results could be extended without major difficulties to the Boussinesq system. But
∗T.I.F.R Centre for Applicable Mathematics, Post Bag No. 6503, GKVK Post Office, Bangalore-560065, India, e-mails: mythily@math.tifrbng.res.in,arnabr@math.tifrbng.res.in
†Institut de Math´ematiques de Toulouse, Universit´e Paul Sabatier & CNRS, 31062 Toulouse Cedex, France, e-mail: jean-pierre.raymond@math.univ-toulouse.fr
in [2], only the case of a right angle junction between Dirichlet and Neumann boundary condi- tions is considered. Therefore the results obtained in [2] cannot be applied to the geometrical configurations studied in [1].
The goal of this paper is to fill this gap. An extension of the results of [2] is not obvious.
Indeed the analysis of such control systems relies on fine regularity results and on Green’s formulas to characterize adjoint operators. These Green’s formulas, which are known for velocity fields and temperatures with H
3/2+εspatial regularity with ε > 0, may be wrong for H
3/2−εspatial regularity when ε > 0 (see [3]). In addition to that, the H
3/2−εspatial regularity with ε > 0 that we have for the models considered here, is not sufficient to deal with the nonlinear term of the control Navier–Stokes equations. This is why a careful analysis of these difficulties is carried out in the present paper. The main idea to overcome the drawbacks linked to the loss of regularity is to work with weighted Sobolev spaces. The analysis of the stationary case can be handled with results proved in [4] for the Laplace and Stokes equations separately, but the approach proposed in the present paper is totally new both for the analysis and for the control of such instationary systems, even for the case of the Navier–Stokes system alone.
The precise assumptions on Ω are stated in Section 2.1. Its boundary Γ = ∂Ω is split as follows Γ = Γ
w∪ Γ
c∪ Γ
n, where Γ
w, Γ
cand Γ
nare relatively open subsets in Γ, two by two disjoint. An example of domain satisfying the assumptions stated in Section 2.1 is given in Fig. 1. The control Boussinesq system satisfied by the temperature τ , the fluid velocity u and
w
Ω
Γ Γ
Γ
c
n
Γ w
Figure 1: Rectangular domain with inflow and outflow boundary condition.
pressure q is
∂τ
∂t − κ∆τ + u · ∇τ = f
sin Q, τ (0) = τ
0in Ω, τ = g
τ,s+ θ
con Σ
c, ∂τ
∂n = 0 on Σ \ Σ
c,
∂u
∂t + (u · ∇)u − div σ(u, q) = β τ , div u = 0 in Q, u = 0 on Σ
w, u = g
u,s+ v
con Σ
c, σ(u, q)n = 0 on Σ
n, Πu(0) = Πu
0,
(1.1)
where κ is the thermal diffusivity, ν is the fluid viscosity, σ(u, q) = ν(∇u + (∇u)
T) − qI is the
Cauchy stress tensor, f
s, g
τ,sand g
u,sare stationary data, Q = Ω × (0, ∞), Σ = Γ × (0, ∞), Σ
n= Γ
n× (0, ∞), Σ
w= Γ
w× (0, ∞), Σ
c= Γ
c× (0, ∞), θ
cand v
care control functions, β ∈ L
∞(Ω; R
2), and Π is the Leray projector introduced in Section 2.2.
Let (τ
s, u
s, q
s) be a solution to the stationary Boussinesq system
−κ∆τ
s+ u
s· ∇τ
s= f
sin Ω, τ
s= g
τ,son Γ
c, ∂τ
s∂n = 0 on Γ \ Γ
c,
(u
s· ∇)u
s− div(σ(u
s, q
s)) = β τ
s, div u
s= 0 in Ω, u
s= g
u,son Γ
c, u
s= 0 on Γ
w, σ(u
s, q
s)n = 0 on Γ
n.
(1.2)
We assume that (τ, u, q) is an unstable solution for the instationary Boussinesq system (1.1).
Our goal is to stabilize system (1.1) in a neighborhood of the stationary solution (τ
s, u
s, q
s) at a given decay rate ω, by finite dimensional boundary controls in the form
θ
c(x, t) =
N
X
i=1
f
i(t)g
θ,i(x) and v
c(x, t) =
N
X
i=1
f
i(t)g
v,i(x). (1.3) The functions g
i= (g
θ,i, g
v,i)
Tare localized in Γ
c, and they play the role of actuators. They have to be chosen in order to satisfy some stabilizability properties. The function f = (f
i)
1≤i≤N∈ L
2(0, ∞; R
N) is the control variable. In order to determine the control f in feedback form, we write below the system satisfied by (θ, v, p) = (τ − τ
s, u − u
s, q − q
s):
∂θ
∂t − κ∆θ + u
s· ∇θ + v · ∇τ
s+ v · ∇θ = 0 in Q, θ = θ
con Σ
c, ∂θ
∂n = 0 on Σ \ Σ
c, θ(0) = θ
0in Ω,
∂v
∂t − div σ(v, p) + (u
s· ∇)v + (v · ∇)u
s+ (v · ∇)v = β θ in Q, div v = 0 in Q, v = v
con Σ
c, v = 0 on Σ
w,
σ(v, p)n = 0 on Σ
n, Πv(0) = Πv
0in Ω,
(1.4)
where θ
0= τ
0− τ
sand v
0= u
0− u
s. The linearized system around the steady state solution (τ
s, u
s, q
s) is given by
∂θ
∂t − κ∆θ + u
s· ∇θ + v · ∇τ
s= 0 in Q, θ = θ
con Σ
c, ∂θ
∂n = 0 on Σ \ Σ
c, θ(0) = θ
0in Ω,
∂v
∂t − div σ(v, p) + (u
s· ∇)v + (v · ∇)u
s= β θ, div v = 0 in Q, v = v
con Σ
c, v = 0 on Σ
w, σ(v, p)n = 0 on Σ
n,
Πv(0) = Πv
0in Ω.
(1.5)
The main issues that we have to deal with are
– obtaining regularity results for the linearized system, for the closed-loop homogeneous and
nonhomogeneous linearized systems,
– writing the linearized system as a control system,
– studying the stabilizability of the control system, and determining stabilizing feedback control laws,
– studying the stability of the closed-loop nonlinear system.
Let us address successively these issues. In Section 3, by using the variational formulation of the stationary Boussinesq system, we define the Boussinesq operator (A, D(A)) in Z = L
2(Ω) ×V
n,Γ0d
(Ω), where V
0n,Γd
(Ω) = {v ∈ L
2(Ω; R
2) | div v = 0 in Ω, v ·n = 0 on Γ
d} and Γ
d= Γ
w∪ Γ
c(see Section 2.1). The control operator B is defined in Section 3.5, and in Section 4, we show that the linearized system (1.5) satisfied by z = (θ, Πv)
Tcan be written in the form
z
0= Az + Bf, z(0) = z
0,
and that (I − Π)v satisfies an algebraic equation. This representation is essential to apply the stabilizability Hautus test. We also need the precise characterization of the adjoint oper- ators A
∗, B
∗. This is obtained by using Green’s formulas and regularity results for functions belonging to D(A) and to D(A
∗). In particular, we prove in Section 3 that
D(A) ⊂ H
δ2(Ω) × H
δ2(Ω; R
2) and D(A
∗) ⊂ H
δ2(Ω) × H
δ2(Ω; R
2),
where H
δ2(Ω) × H
δ2(Ω; R
2) is a weighted Sobolev space introduced in Section 2.1. We emphasize that these regularity results are obtained by using the mixed (variational) formulation of the stationary Boussinesq system, the Green’s formulas are obtained by using the PDE formulation of this system and the variational formulation is used in writing the PDE in operator form. That is why we show the equivalence between these three formulations. Due to the loss of regularity induced by the Dirichlet–Neumann, Neumann–Neumann and Dirichlet–Dirichlet junctions, the analysis is more tricky.
Stabilizability results may be deduced from null controllability results. Local boundary con- trollability to trajectories of the Boussinesq system is established in [5], [6], [7]. See also [8], for additional results of local exact controllability to the trajectories of the Boussinesq systems with interior controls. But all these controllability results are for Dirichlet conditions for both velocity and temperature and when the boundary of the domain is regular. As far as we know, there are no similar controllability or stabilizability results in the case of mixed boundary condi- tions and the analysis of our problem is more delicate. Here, adapting to the Boussinesq system the approach used in [9] for the Oseen system, we prove in Appendix a local Carleman estimate for the adjoint stationary Boussinesq system. Using this result, we prove that the control sys- tem corresponding to (1.5) is stabilizable. Then the feedback control law is defined as in [10]
or [2]. The regularity of solutions to the nonhomogeneous linearized closed loop system is also established in Section 5.
In particular, we show in Section 5.3 that if the initial condition (θ
0, v
0) belongs to H
ε(Ω) × V
n,Γεd
(Ω), where V
εn,Γd
(Ω) = V
0n,Γd
(Ω) ∩ H
ε(Ω; R
2), and if the right hand side of the nonhomo- geneous linearized closed-loop system belongs to L
2(0, ∞; H
Γ−1+εc
(Ω)) × L
2(0, ∞; H
−1+εΓd
(Ω)), for ε ∈ (0, 1/2), then the solution to the closed-loop linearized system belongs to the Hilbert space
X =
L
2(0, ∞; D((λ
0I − A)
12+ε2)) ∩ H
12+ε2(0, ∞; Z)
+H
1(0, ∞; H
32−ε(Ω) × H
32−ε(Ω)).
In Section 6, we carefully estimate the nonlinear terms v · ∇θ and (v · ∇) v in order to show, via a fixed point method, that the closed-loop nonlinear system (6.1) admits a unique solution in some ball of X. More precisely we prove the following theorem in Section 6.2.
Theorem 1.1. Let ε belong to (0, 1/2). For a given ω > 0, there exist a family of actuators g
i= (g
θ,i, g
v,i) ∈ H
003/2(Γ
c) ×H
3/200(Γ
c) explicitly given in (5.5), a constant µ
0> 0 and a constant C
0> 0 such that if µ ∈ (0, µ
0) and if k(θ
0, v
0)k
Hε(Ω)×Vεn,Γd(Ω)
≤ C
0µ, then the system (6.1) admits a unique solution in the ball B
µ=
(θ, v) ∈ X : ke
ωt(θ, v)k
X≤ µ , and it satisfies k(θ(t), v(t))k
Hε(Ω)×Hε(Ω)≤ Ce
−ωt,
where C depends on kθ
0k
Hε(Ω)and kv
0k
Hε(Ω).
For the definition of the spaces H
003/2(Γ
c) and H
3/200(Γ
c) see Section 2.1.
Therefore, our paper provides a rigorous framework for models studied in [1]. Let us mention some additional references linked to our paper. The idea of using finite dimensional controllers for the stabilization of linear parabolic systems goes back to [11]. The stabilization of the Navier–Stokes system by means of finite dimensional feedback controllers is obtained in [12], in the case of an internal control. The case of a boundary control is studied in [13], [10], [14]. See also additional results in [15], under the more restrictive assumption that the unstable spectrum of the linearized operator is semi-simple.
2 Preliminaries
2.1 Functional framework and assumptions
We set L
2(Ω) = H
0(Ω; R
2), H
s(Ω) = H
s(Ω; R
2), ∀ s > 0 and the same notation conventions will be used for trace spaces. We denote by H
1/200(Γ
c) (respectively H
3/200(Γ
c)) the subset of functions belonging to H
1/2(Γ
c) (respectively H
3/2(Γ
c)) whose extension by 0 to the whole boundary Γ belongs to H
1/2(Γ) (respectively H
3/2(Γ)). A similar definition is valid for H
001/2(Γ
d). Now we introduce the spaces
V
1Γd
(Ω) = { v ∈ H
1(Ω) | div v = 0 in Ω, v = 0 on Γ
d}, V
0n,Γd
(Ω) is the closure of V
1Γd
(Ω) in L
2(Ω), H
Γ1n(Ω) =
p ∈ H
1(Ω) | p = 0 on Γ
n, Z = L
2(Ω) × V
0n,Γd
(Ω).
From the above definition of V
n,Γ0d
(Ω), it follows that, for all v ∈ V
0n,Γd
(Ω), v · n|
Γd= 0 as element in (H
001/2(Γ
d))
0.
For ε ∈ [0, 1/2), we define the spaces H
Γ−1+εc
(Ω) as the dual space of H
Γ1−εc
(Ω) with respect to L
2(Ω) and H
−1+εΓd
(Ω) as the dual space of H
1−εΓd
(Ω) with respect to L
2(Ω).
Throughout the paper, the following assumptions are assumed to be satisfied.
(H
1) For the two dimensional domain Ω, the boundary Γ is split in the form Γ =
Ns
S
i=1
Γ
i, where each Γ
iis a submanifold of class C
2and the boundary condition on Γ
ifor the fluid velocity (or temperature) is only of one type, either Dirichlet or Neumann.
(H
2) The set of junction points between these submanifolds is J = {J
k| j = 1, · · · , N
J}. For each vertex J
k∈ J , there exists r
k> 0 such that {x ∈ R
2| dist(x, J
k) ≤ r
k} ∩ Γ is the union of two segments.
(H
3) If Γ
i∩ Γ
j= {J
k}, for some 1 ≤ i, j ≤ N
sand some 1 ≤ k ≤ N
J, the angle at J
kinterior to Ω is at most π if it corresponds to a Dirichlet–Neumann junction, and it is strictly less than 2π otherwise.
(H
4) The control zone Γ
ccorresponds to one of the submanifolds of the family (Γ
i)
1≤i≤Ns. We assume that Γ
cis of class C
3.
(H
5) The solution (τ
s, u
s, q
s) of the stationary problem (1.2) belongs to H
1+ε0(Ω) × H
1+ε0(Ω) × H
ε0(Ω) and (τ
s, u
s)|
Ωc,ε∈ H
2+ε0(Ω
c,ε) × H
2+ε0(Ω
c,ε) for all ε > 0 and some ε
0> 0, where Ω
c,ε= {x ∈ Ω | dist(x, Γ \ Γ
c) > ε}.
In our work, we need weighted Sobolev spaces for further analysis. For all −1 < δ < 1, s ∈ N, we introduce H
δs(Ω; R
2), the closure of C
∞(Ω) in the norms
kvk
2Hsδ(Ω;R2)
= X
|α|≤s
Z
Ω
r
2δ|∂
αv|
2,
where r stands for the distance to J, α = (α
1, α
2) ∈ N
2denotes a two-index, |α| = α
1+ α
2and ∂
αdenotes the corresponding partial differential operator. We denote the weighted Sobolev spaces as
L
2δ(Ω) = H
δ0(Ω; R
2) and H
sδ(Ω) = H
δs(Ω; R
2), ∀ s > 0.
Note that, following [16], we can also define the space H
δs(Ω) when s > 0 is not an integer. We also recall that [16, Theorem 3]
H
δs(Ω) , → H
s−δ(Ω) for all s ≥ δ ≥ 0,
and the trace operator γ
0is continuous from H
δ2(Ω) into H
3/2−δ(Γ) for all 0 < δ < 3/2. We also introduce the following time-dependent weighted Sobolev space
H
2,1δ(Ω) = L
2(0, T ; H
2δ(Ω)) ∩ H
1(0, T ; L
2(Ω)).
2.2 Some useful results
The orthogonal projection in L
2(Ω) onto V
0n,Γd
(Ω) is denoted by Π. The following result can be proved as in [2, Lemma 2.2]:
Lemma 2.1. We have the following orthogonal decomposition L
2(Ω) = V
0n,Γd(Ω) ⊕ ∇H
Γ1n
(Ω).
The orthogonal projection Π from L
2(Ω) to V
0n,Γd
(Ω) is defined by Πu = u − ∇q
1− ∇q
2, where
q
1∈ H
01(Ω), ∆q
1= div u in Ω, q
2∈ H
Γ1n(Ω), ∆q
2= 0 in Ω, ∂q
2∂n = (u − ∇q
1) · n on Γ
d.
We have introduced weighted Sobolev spaces in order to study the regularity of solutions to the following elliptic equations
λ
0θ − κ∆θ = f in Ω, θ = 0 on Γ
c, ∂θ
∂n = 0 on Γ \ Γ
c, (2.1) and
λ
0v − div σ(v, p) = h in Ω,
div v = 0 in Ω, v = 0 on Γ
d, σ(v, p)n = 0 on Γ
n, (2.2) where λ
0≥ 0, f ∈ L
2(Ω), and h ∈ L
2(Ω). It is well known that equation (2.1) admits a unique solution θ in H
1(Ω), and that the system (2.2) admits a unique solution (v, p) ∈ H
1(Ω) ×L
2(Ω).
Under the assumptions made on Γ in Section 2.1, we can prove the following proposition.
Proposition 2.2. The solution θ ∈ H
1(Ω) to equation (2.1) belongs to H
δ2(Ω) and satisfies kθk
H2δ(Ω)
≤ C
δkf k
L2(Ω),
for all δ ∈ (1/2, 1). The solution (v, p) ∈ H
1(Ω) × L
2(Ω) to system (2.2) belongs to H
2δ(Ω) × H
δ1(Ω) for all δ ∈ (1/2, 1), and it satisfies
kvk
H2δ(Ω)
+ kpk
H1δ(Ω)
≤ C
δkhk
L2(Ω).
Proof. Since the angle of junctions between Dirichlet–Neumann (respectively Neumann–Neumann and Dirichlet–Dirichlet) is less than or equal to π (respectively less than 2π), the regularity for θ follows from [4, Theorem 6.4.6], and the regularity for (v, p) follows from [4, Theorem 9.4.5].
The existence result for stationary Navier–Stokes equation for polyhedral domain is given in [4, Theorem 11.2.1]. We can adapt this idea to establish the existence and regularity result for nonlinear stationary Boussinesq system (1.2).
Theorem 2.3. Let f
s∈ H
Γ−1c
, (g
τ,s, g
u,s) ∈ H
001/2(Γ
c) × H
1/200(Γ
c). There exist constants ε
1> 0 and C
1> 0 such that if
k(g
τ,s, g
u,s)k
H1/2(Γc)×H1/2(Γc)+ kf
sk
H−1Γc(Ω)
≤ C
1ε
1, (2.3) then the problem (1.2) admits a solution in the ball
B
ε1=
(τ, u, q) ∈ H
1(Ω) × H
1(Ω) × L
2(Ω) : k(τ, u, q)k
H1(Ω)×H1(Ω)×L2(Ω)≤ ε
1.
If moreover f
s∈ L
2(Ω), (g
τ,s, g
u,s) ∈ H
003/2(Γ
c) × H
3/200(Γ
c), then (τ, u, q) ∈ H
1+ε0(Ω) ×
H
1+ε0(Ω) × H
ε0(Ω), for some ε
0> 0.
3 Stationary linearized system
To define the different operators involved in the Boussinesq system, we first consider the following stationary equation
λ
0θ − κ∆θ + u
s· ∇θ + v · ∇τ
s= f in Ω, θ = θ
con Γ
c, ∂θ
∂n = 0 on Γ \ Γ
c,
λ
0v − div σ(v, p) + (u
s· ∇)v + (v · ∇)u
s− β θ = h in Ω, div v = 0 in Ω,
v = v
con Γ
c, v = 0 on Γ
w, σ(v, p)n = 0 on Γ
n,
(3.1)
where λ
0> 0 is chosen later on, v
c∈ H
3/200(Γ
c), θ
c∈ H
003/2(Γ
c), and (f, h) ∈ L
2(Ω) × L
2(Ω).
Definition 3.1. Let δ belong to (1/2, 1). A triplet (θ, v, p) ∈ H
δ2(Ω) × H
2δ(Ω) × H
δ1(Ω) is a solution to (3.1) iff the equations (3.1)
1, (3.1)
3, (3.1)
4are satisfied in the sense of distributions, and (3.1)
2and (3.1)
5are satisfied in the sense of traces.
To prove the existence and uniqueness of solution to system (3.1), via the mixed (variational) formulation, we introduce the continuous bilinear form on H
1(Ω) × H
1(Ω) defined by
a((θ, v), (ξ, φ)) = Z
Ω
(λ
0θξ + κ∇θ · ∇ξ + (u
s· ∇θ)ξ + (v · ∇τ
s)ξ) dx +
Z
Ω
(λ
0v · φ + 2νDv : Dφ + [(u
s· ∇)v + (v · ∇)u
s] · φ − θβ · φ) dx,
where Dv =
12(∇v + (∇v)
T), and the continuous linear form on H
1(Ω) × L
2(Ω) defined by b(φ, p) =
Z
Ω
p div φ dx.
The mixed formulation for system (3.1) is
Find (θ, v, p) ∈ H
1(Ω) × H
1(Ω) × L
2(Ω) such that θ = θ
con Γ
cand v = v
con Γ
c, v = 0 on Γ
w, a((θ, v), (ξ, φ)) − b(φ, p) =
Z
Ω
(f ξ + h · φ) for all (ξ, φ) ∈ H
Γ1c(Ω) × H
1Γd
(Ω), b(v, ψ) = 0 for all ψ ∈ L
2(Ω).
(3.2)
The variational formulation for system (3.1) is
Find (θ, v) ∈ H
1(Ω) × V
1(Ω) such that θ = θ
con Γ
cand v = v
con Γ
c, v = 0 on Γ
w, a((θ, v), (ξ, φ)) =
Z
Ω
(f ξ + h · φ) ∀ (ξ, φ) ∈ H
Γ1c(Ω) × V
1Γd(Ω).
(3.3)
The mixed formulation (3.2) will be used to analyze the regularity of solutions to equation (3.1), while the variational formulation (3.3) will be used to write equation (3.1) in an operator form.
Later we will prove the equivalence of these formulations and of Definition 3.1.
3.1 Homogeneous boundary conditions
Proposition 3.2. There exists λ
0> 0 depending only on κ, ν, u
s, τ
sand β such that, for all (θ, v) ∈ H
Γ1c(Ω) × V
Γ1d
(Ω), we have a((θ, v), (θ, v)) ≥ 1
2 min(κ, 2ν )k(θ, v)k
2H1Γc(Ω)×V1Γ
d(Ω)
. (3.4)
Proof. The proposition is a direct consequence of classical majorizations combined with Korn’s and Young’s inequalities.
Consider the linearized stationary Boussinesq system with homogeneous boundary conditions λ
0θ − κ∆θ + u
s· ∇θ + v · ∇τ
s= f in Ω,
θ = 0 on Γ
c, ∂θ
∂n = 0 on Γ \ Γ
c,
λ
0v − div σ(v, p) + (u
s· ∇)v + (v · ∇)u
s− β θ = h in Ω, div v = 0 in Ω, v = 0 on Γ
d, σ(v, p) n = 0 on Γ
n.
(3.5)
Proposition 3.3. Let λ
0be large so that (3.4) holds. Let (f, h) belong to L
2(Ω) × L
2(Ω) and assume that θ
c= 0, v
c= 0. The variational formulation (3.3) admits a unique solution (θ, v).
The mixed formulation (3.2) admits a unique solution (θ, v, p). This solution belongs to H
δ2(Ω) × H
2δ(Ω) × H
δ1(Ω) for all δ ∈ (1/2, 1), and we have
kθk
H2δ(Ω)
+ kvk
H2δ(Ω)
+ kpk
H1δ(Ω)
≤ C
δkf k
L2(Ω)+ khk
L2(Ω). (3.6)
Proof. The existence and uniqueness of (θ, v, p) belonging to H
Γ1c(Ω) × V
Γ1d
(Ω) × L
2(Ω) follows from the Lax–Milgram Lemma and from the surjectivity of the divergence map as in [4, Theorem 9.1.5]. To prove the regularity, we can write (3.5) in the form
λ
0v − div σ(v, p) = h + βθ − (u
s· ∇)v − (v · ∇)u
sin Ω, div v = 0 in Ω, v = 0 on Γ
d, σ(v, p)n = 0 on Γ
n,
λ
0θ − κ∆θ = f − u
s· ∇θ − v · ∇τ
sin Ω, θ = 0 on Γ
c,
∂θ
∂n = 0 on Γ \ Γ
c.
Since u
s∈ H
1+ε0(Ω), we have (h + βθ − (u
s· ∇)v − (v · ∇)u
s) ∈ L
2(Ω). From Proposi- tion 2.2, it follows that (v, p) ∈ H
2δ(Ω) × H
δ1(Ω). As (u
s, τ
s) ∈ H
1+ε0(Ω) × H
1+ε0(Ω), we have (f − u
s· ∇θ − v · ∇τ
s) ∈ L
2(Ω). From Proposition 2.2, it follows that θ ∈ H
δ2(Ω).
3.2 Green’s formulas
Now we state Green’s formulas for the temperature and velocity separately.
Theorem 3.4. Let δ belong to (1/2, 1), and ε be positive. Let us set Γ
c,ε= {x ∈ Γ | dist(x, Γ
c) <
ε}, and let us define
H
Γ1c,ε(Ω) =
θ ∈ H
Γ1c| θ|
Γc,ε= 0 . For all θ ∈ H
δ2(Ω) and for all ξ ∈ H
Γ1c,ε
(Ω), we have hξ, ∆θi
L2−δ(Ω),L2δ(Ω)
= − Z
Ω
∇θ · ∇ξ + Z
Γ\Γc,ε
ξ ∂θ
∂n . (3.7)
If in addition ∂θ
∂n ∈ L
2(Γ \ Γ
c), then hξ, ∆θi
L2−δ(Ω),L2δ(Ω)
= − Z
Ω
∇θ · ∇ξ + Z
Γ\Γc
ξ ∂θ
∂n , for all ξ ∈ H
Γ1c(Ω). (3.8) Proof. Let us first notice that with the help of Hardy’s inequality ([4, page 223])
Z
Ω
r
−2δ|u|
2dx ≤ C Z
Ω
r
2(1−δ)|∇u|
2dx,
we have H
Γ1c(Ω) ⊂ L
2−δ(Ω), for δ ∈ (1/2, 1). The Green’s formula (3.7) can be first proved for θ ∈ C
∞(Ω) and ξ ∈ C
∞(Ω) ∩ H
Γ1c,ε
(Ω). Next the theorem follows from the density of C
∞(Ω) in H
δ2(Ω) and that of C
∞(Ω) ∩ H
Γ1c,ε
(Ω) in H
Γ1c,ε
(Ω). The second Green’s formula (3.8) can be proved by using the density of H
Γ1c,ε
(Ω) into H
Γ1c
(Ω).
Theorem 3.5. Let δ belong to (1/2, 1), and ε be positive. Let us set Γ
d,ε= {x ∈ Γ| dist(x, Γ
d) < ε}, and let us define
H
1Γd,ε(Ω) =
v ∈ H
1Γd| v|
Γd,ε= 0 . For all (v, p) ∈ H
2δ(Ω) × H
δ1(Ω) and for all φ ∈ H
1Γd,ε
(Ω), we have hφ, div σ(v, p)i
L2−δ(Ω),L2δ(Ω)
= −2ν Z
Ω
Dv : Dφ + Z
Ω
p div φ + Z
Γ\Γd,ε
σ(v, p)n · φ. (3.9) If in addition σ(v, p)n ∈ L
2(Γ
n), then we have
hφ, div σ(v, p)i
L2−δ(Ω),L2δ(Ω)
= −2ν Z
Ω
Dv : Dφ + Z
Ω
p div φ + Z
Γn
σ(v, p)n · φ, (3.10) for all φ ∈ H
1Γd
(Ω).
Proof. The Green’s formula (3.10) can be first proved for (v, p) ∈ C
∞(Ω) × C
∞(Ω) and φ ∈ C
∞(Ω)∩H
1Γd,ε
(Ω). Next (3.10) follows from the density of C
∞(Ω) in H
2δ(Ω), of C
∞(Ω)∩H
1Γd,ε
(Ω) in H
1Γd,ε
(Ω), and of C
∞(Ω) in H
δ1(Ω). The second Green’s formula (3.9) follows from the density of H
Γ1d,ε
(Ω) into H
Γ1d
(Ω).
These Green’s formulas are used to establish the equivalence between Definition 3.1, mixed
formulation (3.2), and variational formulation (3.3).
Theorem 3.6. Let λ
0be large so that (3.4) holds. Let (f, h) belong to L
2(Ω) × L
2(Ω), and assume that θ
c= 0, v
c= 0.
(1) A triplet (θ, v, p) ∈ H
δ2(Ω) × H
2δ(Ω) × H
δ1(Ω) is a solution to (3.5) in the sense of Definition 3.1 if and only if (θ, v, p) is a solution to the mixed formulation (3.2).
(2) If (θ, v, p) is a solution to the mixed formulation (3.2), then (θ, v) is a solution to the variational formulation (3.3).
(3) If (θ, v) is a solution to the variational formulation (3.3), then there exists p ∈ L
2(Ω) such that (θ, v, p) is a solution to the mixed formulation (3.2).
Proof. (1) Let (θ, v, p) ∈ H
δ2(Ω) × H
2δ(Ω) × H
δ1(Ω) be solution of equation (3.5). Multiplying (3.5)
1by ξ ∈ H
Γ1c(Ω) and (3.5)
3by φ ∈ H
1Γd
(Ω) and using the Green’s formulas (3.8)–(3.10), we obtain the mixed variational formulation.
Conversely, assume that (θ, v, p) ∈ H
δ2(Ω) × H
2δ(Ω) × H
δ1(Ω) satisfies the mixed variational formulation (3.2) with θ
c= 0 and v
c= 0. By using the Green’s formulas (3.8)–(3.10) successively with test functions (ξ, 0) and (0, φ) for ξ ∈ C
c∞(Ω) and φ ∈ C
c∞(Ω; R
2) in (3.2), we can prove that (θ, v, p) satisfies (3.5)
1and (3.5)
3in the sense of distributions.
Now, by using equation (3.5)
1, (3.7) and the mixed formulation (3.2) with (ξ, 0), ξ ∈ H
Γ1c,ε
(Ω), we obtain
Z
Γ\Γc,ε
∂θ
∂n ξ = 0, ∀ ξ ∈ H
Γ1c,ε(Ω). (3.11) Thus ∂θ
∂n = 0 on Γ \ Γ
c,ε. Since this is true for all ε > 0, we conclude that ∂θ
∂n = 0 on Γ \ Γ
c. With similar arguments, we can also recover the boundary condition σ(v, p)n = 0 on Γ
n.
(2) It is very easy to see that if we choose (ξ, φ) ∈ H
Γ1c
(Ω) × V
1Γd
(Ω) in (3.2), then we obtain the variational formulation (3.3).
(3) This follows from the uniqueness of solution (θ, v) to the variational formulation (3.3) and the uniqueness of (θ, v, p) to the mixed formulation (3.2) proved in Proposition 3.3.
We define the unbounded operator (A, D(A)) in Z by D(A) = {(θ, v) ∈ H
Γ1c
(Ω) × V
1Γd
(Ω) | (ξ, φ) 7→ a((θ, v), (ξ, φ)) is continuous in Z}, and
h(λ
0I − A)(θ, v), (ξ, φ)i
Z= a((θ, v), (ξ, φ)) for all (θ, v) ∈ D(A), (ξ, φ) ∈ H
Γ1c(Ω) × V
1Γd
(Ω).
Theorem 3.7. The operator (A, D(A)) is the infinitesimal generator of a strongly continuous analytic semigroup on Z, and its resolvent is compact.
Proof. Using the coercivity of the bilinear form a((θ, v), (θ, v)) in H
Γ1c(Ω)×V
1Γd
(Ω), we conclude
from [17, Theorem 2.12, page 115] that (A, D(A)) generates an analytic semigroup on Z. From
Proposition 3.3, it follows that (λ
0I − A)
−1is bounded from Z into H
Γ1c(Ω) × V
1Γd
(Ω). Since, the imbedding from H
Γ1c(Ω) ×V
1Γd
(Ω) into Z is compact, the resolvent of A is compact in Z.
3.3 Adjoint problem
Let us introduce the adjoint equation
λ
0ξ − κ∆ξ − u
s· ∇ξ − β · φ = f in Ω, ξ = 0 on Γ
c, κ ∂ξ
∂n + (u
s· n)ξ = 0 on Γ \ Γ
c,
λ
0φ − div σ(φ, ψ) − (u
s· ∇)φ + (∇u
s)
Tφ + (∇τ
s)
Tξ = h in Ω, div φ = 0 in Ω, φ = 0 on Γ
d, σ(φ, ψ)n + (u
s· n)φ = 0 on Γ
n,
(3.12)
where (f, h) ∈ L
2(Ω) × L
2(Ω). We define the following bilinear form a
#and the linear form `, on H
Γ1c
(Ω) × V
1Γd
(Ω), by a
#((ξ, φ), (θ, v)) =
Z
Ω
(λ
0ξθ + κ∇ξ · ∇θ − (u
s· ∇ξ)θ − (β · φ)θ) dx +
Z
Ω
λ
0φ · v + 2νDφ : Dv − ((u
s· ∇)φ) · v + (∇u
s)
Tφ · v + (∇τ
s)
Tξ · v
dx +
Z
Γ\Γc
(u
s· n)ξθ dx + Z
Γn
v · (u
s· n)φ dx,
`(θ, v) = Z
Ω
(f θ + h · v) .
The mixed (variational) formulation for equation (3.12) is defined by Find (ξ, φ, ψ) ∈ H
Γ1c
(Ω) × H
1Γd
(Ω) × L
2(Ω) such that
a
#((ξ, φ), (θ, v)) − b(v, ψ) = `(θ, v) ∀ (θ, v) ∈ H
Γ1c(Ω) × H
1Γd(Ω), b(φ, p) = 0 for all p ∈ L
2(Ω).
(3.13)
We can prove that the mixed formulation (3.13) admits a unique solution (ξ, φ, ψ) ∈ H
δ2(Ω) × H
2δ(Ω) × H
δ1(Ω) for all δ ∈ (
12, 1), as in Proposition 3.3. Associated with a
#, we introduce the operator (A
#, D(A
#)) defined by
D(A
#) = {(ξ, φ) ∈ H
Γ1c
(Ω) × V
1Γd
(Ω) | (θ, v) 7→ a
#((ξ, φ), (θ, v)) is continuous in Z},
and
(λ
0I − A
#)(ξ, φ), (θ, v)
Z
= a
#((ξ, φ), (θ, v)) for all (ξ, φ) ∈ D(A
#) and (θ, v) ∈ H
Γ1c(Ω)×
V
Γ1d
(Ω). Let (A
∗, D(A
∗)) be the adjoint of (A, D(A)). Since a
#((ξ, φ), (θ, v)) = a((θ, v), (ξ, φ)), for all (ξ, φ) and all (θ, v) belonging to H
Γ1c(Ω) × V
1Γd
(Ω), we have D(A
∗) = D(A
#) and A
∗= A
#.
Let λ
0be fixed so that (3.4) holds. Let us analyze the fractional powers of the domain of the operator A and its adjoint A
∗used later on.
Theorem 3.8. We have
D((λ
0I − A)
12) = H
Γ1c(Ω) × V
1Γd(Ω), D((λ
0I − A
∗)
12) = H
Γ1c(Ω) × V
1Γd(Ω), (D((λ
0I − A
∗)
12))
0= H
Γ−1c
(Ω) × V
−1Γd
(Ω). (3.14)
Furthermore, for ε ∈ (0,
12),
(D((λ
0I − A
∗)
1/2−ε/2))
0= [L
2(Ω), H
Γ−1c
(Ω)]
1−ε× [V
0n,Γd
(Ω), V
Γ−1d
(Ω)]
1−ε= H
Γ−1+εc(Ω) × H
−1+εΓd
(Ω) ∩ V
−1Γd
(Ω)
, (3.15)
D((λ
0I − A)
1/2+ε/2) ⊂
H
Γ1c(Ω) ∩ H
1+η(ε)(Ω)
×
V
1Γd(Ω) ∩ H
1+η(ε)(Ω)
, (3.16) where η(ε) =
ε2− ε
2.
Proof. Step 1. Proof of (3.14). The first and second identities in (3.14) follow by adapting the proof from [2, Theorem 2.13] to our case. For that it is sufficient to establish
[H
Γ1c(Ω) × V
Γ1d(Ω), H
Γ−1c(Ω) × V
−1Γd
(Ω)]
1/2= Z, (3.17)
and to show that (λ
0I − A) is a closed maximal monotone operator in H
Γ−1c
(Ω) × V
−1Γd
(Ω). The last identity of (3.14) follows by duality arguments.
Step 2. Proof of (3.15). Now using (3.14), for ε ∈ (0, 1/2), we have
D((λ
0I − A
∗)
1/2−ε/2) = [H
Γ1c(Ω) × V
1Γd(Ω), L
2(Ω) × V
0n,Γd(Ω)]
ε.
By duality, we get the first identity in (3.15). To get the second identity in (3.15), as in [2, Lemma 2.12], we first note that
V
n,Γ0 d(Ω) = L
2(Ω) ∩ V
Γ−1d
(Ω).
Then, for ε ∈ (0, 1/2), we can write [V
n,Γ0d
(Ω), V
−1Γd
(Ω)]
1−ε= [L
2(Ω) ∩ V
−1Γd
(Ω), H
−1Γd
(Ω) ∩ V
−1Γd
(Ω)]
1−ε= [L
2(Ω), H
−1Γd
(Ω)]
1−ε∩ V
−1Γd
(Ω) = H
−1+εΓd
(Ω) ∩ V
−1Γd
(Ω).
(3.18) We also have
[L
2(Ω), H
Γ−1c(Ω)]
1−ε= H
Γ−1+εc(Ω).
Combining both the results yields the second identity in (3.15).
Step 3. Proof of (3.16). Let us denote by (θ, v) the solution to the equation
(λ
0I − A)(θ, v) = (f, h). (3.19) Let ε ∈ (0, 1/2) be fixed. We know that λ
0I − A is an isomorphism from (D((λ
0I − A
∗)
1/2−ε/2))
0into D((λ
0I − A)
1/2+ε/2). Thus we have the following estimate
k(f, h)k
(D((λ0I−A∗)1/2−ε/2))0
≤ Ck(θ, v)k
D((λ0I−A)1/2+ε/2)
. (3.20)
Notice further that the solution mapping from (f, h) to (θ, v) is continuous from H
Γ−1c
(Ω) × V
Γ−1d
(Ω) into H
Γ1c(Ω) × V
Γ1d
(Ω) and also from Z into (H
3/2−ε(Ω) ∩ H
Γ1c(Ω)) × (H
3/2−ε(Ω) ∩ V
Γ1d
(Ω)). Thus by interpolation, it is also continuous from (D((λ
0I − A
∗))
1/2−ε/2)
0= [L
2(Ω) × V
n,Γ0d
(Ω), H
Γ−1c(Ω) × V
Γ−1d
(Ω)]
1−εinto
[ll][(H
3/2−ε(Ω) ∩ H
Γ1c(Ω)) × (H
3/2−ε(Ω) ∩ V
Γ1d(Ω)), H
Γ1c× V
1Γd]
1−ε= (H
1+η(ε)(Ω) ∩ H
Γ1c(Ω)) × (H
1+η(ε)(Ω) ∩ V
1Γd
(Ω)), where η(ε) =
ε2− ε
2. Thus we have
k(θ, v)k
H1+η(ε)×H1+η(ε)≤ Ck(f, h)k
(D((λ0I−A∗))1/2−ε/2)0
. This continuity result, with estimate (3.20), gives the last inclusion in (3.16).
3.4 Nonhomogeneous boundary conditions
We now want to prove the existence of a solution (θ, v, p) of the mixed formulation for the system (3.1) with nonhomogeneous boundary condition.
Theorem 3.9. Assume that (f, h) belongs to L
2(Ω) × L
2(Ω) and (θ
c, v
c) belongs to H
1 2
00
(Γ
c) × H
1 2
00
(Γ
c). Then, the mixed formulation (3.2) admits a unique solution (θ, v, p).
Moreover, if (θ
c, v
c) belongs to H
3 2
00
(Γ
c) × H
3 2
00
(Γ
c), then this solution belongs to H
δ2(Ω) × H
2δ(Ω) × H
δ1(Ω) for all δ ∈ (
12, 1), and it satisfies
[ll]kθk
H2δ(Ω)
+ kvk
H2δ(Ω)
+ kpk
H1 δ(Ω)≤ C
δkf k
L2(Ω)+ khk
L2(Ω)+ kθ
ck
H32(Γc)
+ kv
ck
H32(Γc)
. (3.21)
Proof. The uniqueness follows from the uniqueness result stated in Proposition 3.3. To prove the existence of solution, we first consider the equation
λ
0θ b − κ∆b θ = f in Ω, θ b = θ
con Γ
c, ∂b θ
∂n = 0 on Γ \ Γ
c, λ
0b v − div σ( v, b p) b − β θ b = h in Ω,
div v b = 0 in Ω, v b = 0 on Γ
c, v b = 0 on Γ
w, σ( b v, p)n b = 0 on Γ
n.
(3.22)
The equation for θ b is decoupled from the equation for ( v, b p), thus we can apply results already b existing in the literature for each equation separately. Since θ
c∈ H
3 2
00
(Γ
c), from [4, Theorem 6.4.6], it follows that system (3.22)
1–(3.22)
2admits a unique solution θ b ∈ H
1(Ω), and that this solution belongs to H
δ2(Ω) for all δ ∈ (
12, 1). Since v
c∈ H
3 2
00
(Γ
c), from [4, Theorem 9.1.5,
Theorem 9.4.5], it follows that system (3.22)
3–(3.22)
4admits a unique solution ( b v, p) b ∈ V
1(Ω) ×
L
2(Ω), and that this solution belongs to H
2δ(Ω) × H
δ1(Ω) for all δ ∈ (1/2, 1).
We look for a solution (θ, v, p) to the mixed formulation (3.2) of the form (θ, v, p) = ( θ, b v, b p) + (e b θ, e v, p). Thus, (e e θ, e v, p) must satisfy e
λ
0θ e − κ∆ θ e + u
s.∇e θ + v.∇τ e
s= −u
s· ∇b θ − v b · ∇τ
sin Ω, θ e = 0 on Γ
c, ∂e θ
∂n = 0 on Γ \ Γ
c,
λ
0e v − div σ( v, e p) + (u e
s.∇) v e + ( v.∇)u e
s− β θ e
1=
−(u
s.∇) b v + ( v b · ∇)u
sin Ω,
div v e = 0 in Ω, e v = 0 on Γ
c, v e = 0 on Γ
w, σ( v, e p) e n = 0 on Γ
n.
(3.23)
Since the right hand side in (3.23)
1(respectively (3.23)
4) belongs to L
2(Ω) (respectively L
2(Ω)), from Proposition 3.3, it follows that this system has a unique solution belonging to H
δ2(Ω) × H
2δ(Ω) × H
δ1(Ω) for all δ ∈ (
12, 1).
Remark 3.10. We can prove the equivalence of the three formulations for the solution as before, even when (f, h) belong to L
2(Ω) × L
2(Ω) and (θ
c, v
c) belongs to H
1 2
00
(Γ
c) × H
1 2
00
(Γ
c) by using Theorem 3.9.
3.5 Rewriting equation (3.1) in the operator form
We define the Dirichlet operators G ∈ L(H
001/2(Γ
c) × H
1/200(Γ
c), H
1(Ω) × H
1(Ω)) and G
p∈ L(H
001/2(Γ
c) × H
1/200(Γ
c), L
2(Ω)) by
G(θ
c, v
c) = (G
θ(θ
c, v
c), G
v(θ
c, v
c))
T= (θ, v)
T, G
p(θ
c, v
c) = p, (3.24) where (θ, v, p) satisfies equation (3.1) with (f, h) = (0, 0).
Due to Theorem 3.9, the operators G and G
pare continuous linear operators from (H
003/2(Γ
c)×
H
3/200(Γ
c)) into (H
δ2(Ω) × H
2δ(Ω)) and from (H
003/2(Γ
c) × H
3/200(Γ
c)) into H
δ1(Ω) respectively.
We can extend the operator A to an unbounded operator A e with domain D( A) = e L
2(Ω) × V
n0(Ω) in (D(A
∗))
0, in order that ( A, D( e A)) is the infinitesimal generator of an analytic semi- e group on (D(A
∗))
0, with compact resolvent. Later on, we shall denote this extension by A.
Theorem 3.11. Let (θ
c, v
c) ∈ H
003/2(Γ
c) × H
3/200(Γ
c). Let us set
B (θ
c, v
c) = (λ
0I − A) ΠG(θ e
c, v
c), (3.25) where G is defined in 3.24 and Π = e
Id 0
0 Π
. The operator B belongs to L(H
003/2(Γ
c)) × H
3/200(Γ
c)), (D(A
∗))
0).
If a triplet (θ, v, p) ∈ H
δ2(Ω) × H
2δ(Ω) × H
δ1(Ω) is a solution to (3.1), then (θ, v) satisfies (λ
0I − A)(θ, Πv)
T− B(θ
c, v
c)
T= (f, Πh)
Tin (D(A)
∗)
0,
(I − Π)v = P
Ni=1
f
i(I − Π)G
v(θ
c, v
c). (3.26)
Conversely, if (θ, v) ∈ H
δ2(Ω) × H
2δ(Ω) satisfies (3.26), then there exists a unique p ∈ H
δ1(Ω)
such that (θ, v, p) ∈ H
δ2(Ω) × H
2δ(Ω) × H
δ1(Ω) is a solution to (3.1).
Proof. Since G is continuous from (H
003/2(Γ
c) × H
3/200(Γ
c)) into (H
δ2(Ω) × H
2δ(Ω)), it is obvious that B ∈ L(H
003/2(Γ
c)) × H
3/200(Γ
c)), (D(A
∗))
0).
If (θ, v, p) satisfies equation (3.1), then we can write (θ, v, p) = ( θ, e e v, p) + ( e θ, b b v, p), b
where ( e θ, e v, p) is the solution to (3.5) and ( e θ, b b v, p) is the solution to (3.1) with (f, b h) = (0, 0).
Then we obtain
(λ
0I − A)( θ, e Π e v)
T= (f, Πh)
T, (3.27) and
( b θ, v) b
T= G(θ
c, v
c)
Timplies (λ
0I − A)( θ, b Π v) b
T= B (θ
c, v
c)
T(3.28) From (3.27) and (3.28), we obtain (3.26).
The proof of the converse statement is based on the decomposition (θ, v) = (e θ, e v) + (b θ, b v) and on Theorem 3.6-(3). It is left to the reader.
When (θ
c, v
c) is in the form (1.3), it is convenient to introduce the operator B ∈ L( R
N, (D(A
∗))
0) defined by
Bf =
N
X
i=1
f
iB(g
θ,i, g
v,i). (3.29)
Thus we can rewrite (3.26)
1as
(λ
0I − A)(θ, Πv)
T− Bf = (f, Πh) in (D(A)
∗)
0. (3.30) In order to compute the adjoint of the control operator, we now prove a Green’s formula between the solution (G(θ
c, v
c), G
p(θ
c, v
c)) to the nonhomogeneous boundary value problem (3.1) with (θ
c, v
c) as boundary condition and solution (ξ, φ, ψ) to equation (3.12). Before proving that we need one technical lemma about the existence of normal derivative in suitable space.
Lemma 3.12. Let Ω be a bounded open subset of R
2satisfying assumptions (H
1)–(H
4). Let us define
E
1(Ω) = n
ξ ∈ H
2−δ(Ω) | ∆ξ ∈ L
2(Ω) o
, E
2(Ω) = n
[v] ∈ H
1−δ(Ω; R
2×2) | div[v] ∈ L
2(Ω) o , where [v] is a (2 × 2) matrix and the divergence is taken line by line. Then
(1) The linear mapping γ
1: ξ 7→ ∂ξ
∂n is continuous from E
1(Ω) into H
1/2−δ(Γ).
(2) The linear mapping γ
n: [v] 7→ [v]n is continuous from E
2(Ω) into H
1/2−δ(Γ).
Proof. Let us set E(Ω) =
ξ ∈ H
1(Ω) | ∆ξ ∈ L
2(Ω) . We know that γ
1∈ L(E(Ω), H
−1/2(Γ)) [18, Chapter VII, Lemma 1, page 381] and γ
1∈ L(H
2(Ω), H
1/2(Γ)) [19, Theorem III.2.23, page 158]. Thus by interpolation, it follows that γ
1is also bounded from E
1(Ω) into H
1/2−δ(Γ), since [E(Ω), H
2(Ω)]
1−δ= E
1(Ω) and [H
−1/2(Γ), H
1/2(Γ)]
1−δ= H
1/2−δ(Γ).
The proof for γ
nis similar.
Theorem 3.13. Let Ω be a bounded open subset of R
2satisfying assumptions (H
1) − (H
4).
Let 1/2 < δ < 1. Assume that (θ
c, v
c) belongs to (H
003/2(Γ
c) × H
3/200(Γ
c)). Let ( b g, b g) belong to L
2(Γ \ Γ
c) × L
2(Γ
n).
For all θ ∈ H
δ2(Ω) satisfying θ = θ
con Γ
c, ∂θ
∂n = 0 on Γ \ Γ
c, and for all ξ ∈ H
δ2(Ω) ∩ H
Γ1c
(Ω) satisfying
∂n∂ξ= b g on Γ \ Γ
c, we have
hξ, ∆θi
L2−δ(Ω),L2δ(Ω)
− hθ, ∆ξi
L2−δ(Ω),L2δ(Ω)
= − ∂ξ
∂n , θ
cH1/2−δ(Γc),Hδ−1/2(Γc)
− Z
Γ\Γc
gθ. b (3.31)
For all (v, p) ∈ H
2δ(Ω) × H
δ1(Ω) satisfying div v = 0 in Ω, v = v
con Γ
c, v = 0 on Γ
w, σ(v, p)n = 0 on Γ
nand for all (φ, ψ) ∈
H
2δ(Ω) ∩ V
1Γd
(Ω)
× H
δ1(Ω) satisfying σ(φ, ψ)n = b g on Γ
n, we have
hφ, div σ(v, p)i
L2−δ(Ω),L2δ(Ω)
− hv div σ(φ, ψ)i
L2−δ(Ω),L2δ(Ω)
= − hσ(φ, ψ)n, v
ci
H1/2−δ(Γc),Hδ−1/2(Γc)− Z
Γn
b g · v. (3.32)
Proof. Step 1. Proof of (3.31). Since θ belongs to H
δ2(Ω) and ∂θ
∂n = 0 on Γ \ Γ
cand ξ ∈ H
δ2(Ω) ∩ H
Γ1c
(Ω), from Theorem 3.4, it follows that hξ, ∆θi
L2−δ(Ω),L2δ(Ω)
= − Z
Ω
∇θ · ∇ξ. (3.33)
For (θ
k)
k⊂ C
∞(Ω) and (ξ
k)
k⊂ C
∞(Ω), we have Z
Ω
θ
k∆ξ
k= − Z
Ω
∇θ
k· ∇ξ
k+ Z
Γ
θ
k∂ξ
k∂n . (3.34)
By density of C
∞(Ω) in H
δ2(Ω), passing to the limit when k tends to infinity and using Lemma 3.12, we obtain
hθ, ∆ξi
L2−δ(Ω),L2δ(Ω)
= − Z
Ω
∇θ · ∇ξ + ∂ξ
∂n , θ
H1/2−δ(Γ),Hδ−1/2(Γ)
. (3.35)
Since
∂ξ
∂n , θ
H1/2−δ(Γ),Hδ−1/2(Γ)
= ∂ξ
∂n , θ
H1/2−δ(Γc),Hδ−1/2(Γc)
+ ∂ξ
∂n , θ
H1/2−δ(Γ\Γc),Hδ−1/2(Γ\Γc)
= ∂ξ
∂n , θ
H1/2−δ(Γ),Hδ−1/2(Γ)
+ Z
Γ\Γc
b gθ, the proof of (3.31) is complete.
Step 2. Proof of (3.32). As (v, p) ∈ H
2δ(Ω) × H
δ1(Ω) satisfies σ(v, p)n = 0 on Γ
nand φ ∈ H
2δ(Ω) ∩ V
1Γd